~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ext4/inode.c

Version: ~ [ linux-4.20-rc6 ] ~ [ linux-4.19.8 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.87 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.144 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.166 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.128 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.61 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.31.14 ] ~ [ linux-2.6.30.10 ] ~ [ linux-2.6.29.6 ] ~ [ linux-2.6.28.10 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  *  linux/fs/ext4/inode.c
  4  *
  5  * Copyright (C) 1992, 1993, 1994, 1995
  6  * Remy Card (card@masi.ibp.fr)
  7  * Laboratoire MASI - Institut Blaise Pascal
  8  * Universite Pierre et Marie Curie (Paris VI)
  9  *
 10  *  from
 11  *
 12  *  linux/fs/minix/inode.c
 13  *
 14  *  Copyright (C) 1991, 1992  Linus Torvalds
 15  *
 16  *  64-bit file support on 64-bit platforms by Jakub Jelinek
 17  *      (jj@sunsite.ms.mff.cuni.cz)
 18  *
 19  *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
 20  */
 21 
 22 #include <linux/fs.h>
 23 #include <linux/time.h>
 24 #include <linux/highuid.h>
 25 #include <linux/pagemap.h>
 26 #include <linux/dax.h>
 27 #include <linux/quotaops.h>
 28 #include <linux/string.h>
 29 #include <linux/buffer_head.h>
 30 #include <linux/writeback.h>
 31 #include <linux/pagevec.h>
 32 #include <linux/mpage.h>
 33 #include <linux/namei.h>
 34 #include <linux/uio.h>
 35 #include <linux/bio.h>
 36 #include <linux/workqueue.h>
 37 #include <linux/kernel.h>
 38 #include <linux/printk.h>
 39 #include <linux/slab.h>
 40 #include <linux/bitops.h>
 41 #include <linux/iomap.h>
 42 #include <linux/iversion.h>
 43 
 44 #include "ext4_jbd2.h"
 45 #include "xattr.h"
 46 #include "acl.h"
 47 #include "truncate.h"
 48 
 49 #include <trace/events/ext4.h>
 50 
 51 #define MPAGE_DA_EXTENT_TAIL 0x01
 52 
 53 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
 54                               struct ext4_inode_info *ei)
 55 {
 56         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 57         __u32 csum;
 58         __u16 dummy_csum = 0;
 59         int offset = offsetof(struct ext4_inode, i_checksum_lo);
 60         unsigned int csum_size = sizeof(dummy_csum);
 61 
 62         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
 63         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
 64         offset += csum_size;
 65         csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
 66                            EXT4_GOOD_OLD_INODE_SIZE - offset);
 67 
 68         if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
 69                 offset = offsetof(struct ext4_inode, i_checksum_hi);
 70                 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
 71                                    EXT4_GOOD_OLD_INODE_SIZE,
 72                                    offset - EXT4_GOOD_OLD_INODE_SIZE);
 73                 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
 74                         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
 75                                            csum_size);
 76                         offset += csum_size;
 77                 }
 78                 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
 79                                    EXT4_INODE_SIZE(inode->i_sb) - offset);
 80         }
 81 
 82         return csum;
 83 }
 84 
 85 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
 86                                   struct ext4_inode_info *ei)
 87 {
 88         __u32 provided, calculated;
 89 
 90         if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
 91             cpu_to_le32(EXT4_OS_LINUX) ||
 92             !ext4_has_metadata_csum(inode->i_sb))
 93                 return 1;
 94 
 95         provided = le16_to_cpu(raw->i_checksum_lo);
 96         calculated = ext4_inode_csum(inode, raw, ei);
 97         if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
 98             EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
 99                 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
100         else
101                 calculated &= 0xFFFF;
102 
103         return provided == calculated;
104 }
105 
106 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
107                                 struct ext4_inode_info *ei)
108 {
109         __u32 csum;
110 
111         if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
112             cpu_to_le32(EXT4_OS_LINUX) ||
113             !ext4_has_metadata_csum(inode->i_sb))
114                 return;
115 
116         csum = ext4_inode_csum(inode, raw, ei);
117         raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
118         if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
119             EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
120                 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
121 }
122 
123 static inline int ext4_begin_ordered_truncate(struct inode *inode,
124                                               loff_t new_size)
125 {
126         trace_ext4_begin_ordered_truncate(inode, new_size);
127         /*
128          * If jinode is zero, then we never opened the file for
129          * writing, so there's no need to call
130          * jbd2_journal_begin_ordered_truncate() since there's no
131          * outstanding writes we need to flush.
132          */
133         if (!EXT4_I(inode)->jinode)
134                 return 0;
135         return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
136                                                    EXT4_I(inode)->jinode,
137                                                    new_size);
138 }
139 
140 static void ext4_invalidatepage(struct page *page, unsigned int offset,
141                                 unsigned int length);
142 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
143 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
144 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
145                                   int pextents);
146 
147 /*
148  * Test whether an inode is a fast symlink.
149  * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150  */
151 int ext4_inode_is_fast_symlink(struct inode *inode)
152 {
153         if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
154                 int ea_blocks = EXT4_I(inode)->i_file_acl ?
155                                 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
156 
157                 if (ext4_has_inline_data(inode))
158                         return 0;
159 
160                 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161         }
162         return S_ISLNK(inode->i_mode) && inode->i_size &&
163                (inode->i_size < EXT4_N_BLOCKS * 4);
164 }
165 
166 /*
167  * Restart the transaction associated with *handle.  This does a commit,
168  * so before we call here everything must be consistently dirtied against
169  * this transaction.
170  */
171 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
172                                  int nblocks)
173 {
174         int ret;
175 
176         /*
177          * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
178          * moment, get_block can be called only for blocks inside i_size since
179          * page cache has been already dropped and writes are blocked by
180          * i_mutex. So we can safely drop the i_data_sem here.
181          */
182         BUG_ON(EXT4_JOURNAL(inode) == NULL);
183         jbd_debug(2, "restarting handle %p\n", handle);
184         up_write(&EXT4_I(inode)->i_data_sem);
185         ret = ext4_journal_restart(handle, nblocks);
186         down_write(&EXT4_I(inode)->i_data_sem);
187         ext4_discard_preallocations(inode);
188 
189         return ret;
190 }
191 
192 /*
193  * Called at the last iput() if i_nlink is zero.
194  */
195 void ext4_evict_inode(struct inode *inode)
196 {
197         handle_t *handle;
198         int err;
199         int extra_credits = 3;
200         struct ext4_xattr_inode_array *ea_inode_array = NULL;
201 
202         trace_ext4_evict_inode(inode);
203 
204         if (inode->i_nlink) {
205                 /*
206                  * When journalling data dirty buffers are tracked only in the
207                  * journal. So although mm thinks everything is clean and
208                  * ready for reaping the inode might still have some pages to
209                  * write in the running transaction or waiting to be
210                  * checkpointed. Thus calling jbd2_journal_invalidatepage()
211                  * (via truncate_inode_pages()) to discard these buffers can
212                  * cause data loss. Also even if we did not discard these
213                  * buffers, we would have no way to find them after the inode
214                  * is reaped and thus user could see stale data if he tries to
215                  * read them before the transaction is checkpointed. So be
216                  * careful and force everything to disk here... We use
217                  * ei->i_datasync_tid to store the newest transaction
218                  * containing inode's data.
219                  *
220                  * Note that directories do not have this problem because they
221                  * don't use page cache.
222                  */
223                 if (inode->i_ino != EXT4_JOURNAL_INO &&
224                     ext4_should_journal_data(inode) &&
225                     (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
226                     inode->i_data.nrpages) {
227                         journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
228                         tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
229 
230                         jbd2_complete_transaction(journal, commit_tid);
231                         filemap_write_and_wait(&inode->i_data);
232                 }
233                 truncate_inode_pages_final(&inode->i_data);
234 
235                 goto no_delete;
236         }
237 
238         if (is_bad_inode(inode))
239                 goto no_delete;
240         dquot_initialize(inode);
241 
242         if (ext4_should_order_data(inode))
243                 ext4_begin_ordered_truncate(inode, 0);
244         truncate_inode_pages_final(&inode->i_data);
245 
246         /*
247          * Protect us against freezing - iput() caller didn't have to have any
248          * protection against it
249          */
250         sb_start_intwrite(inode->i_sb);
251 
252         if (!IS_NOQUOTA(inode))
253                 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
254 
255         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
256                                  ext4_blocks_for_truncate(inode)+extra_credits);
257         if (IS_ERR(handle)) {
258                 ext4_std_error(inode->i_sb, PTR_ERR(handle));
259                 /*
260                  * If we're going to skip the normal cleanup, we still need to
261                  * make sure that the in-core orphan linked list is properly
262                  * cleaned up.
263                  */
264                 ext4_orphan_del(NULL, inode);
265                 sb_end_intwrite(inode->i_sb);
266                 goto no_delete;
267         }
268 
269         if (IS_SYNC(inode))
270                 ext4_handle_sync(handle);
271 
272         /*
273          * Set inode->i_size to 0 before calling ext4_truncate(). We need
274          * special handling of symlinks here because i_size is used to
275          * determine whether ext4_inode_info->i_data contains symlink data or
276          * block mappings. Setting i_size to 0 will remove its fast symlink
277          * status. Erase i_data so that it becomes a valid empty block map.
278          */
279         if (ext4_inode_is_fast_symlink(inode))
280                 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
281         inode->i_size = 0;
282         err = ext4_mark_inode_dirty(handle, inode);
283         if (err) {
284                 ext4_warning(inode->i_sb,
285                              "couldn't mark inode dirty (err %d)", err);
286                 goto stop_handle;
287         }
288         if (inode->i_blocks) {
289                 err = ext4_truncate(inode);
290                 if (err) {
291                         ext4_error(inode->i_sb,
292                                    "couldn't truncate inode %lu (err %d)",
293                                    inode->i_ino, err);
294                         goto stop_handle;
295                 }
296         }
297 
298         /* Remove xattr references. */
299         err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
300                                       extra_credits);
301         if (err) {
302                 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
303 stop_handle:
304                 ext4_journal_stop(handle);
305                 ext4_orphan_del(NULL, inode);
306                 sb_end_intwrite(inode->i_sb);
307                 ext4_xattr_inode_array_free(ea_inode_array);
308                 goto no_delete;
309         }
310 
311         /*
312          * Kill off the orphan record which ext4_truncate created.
313          * AKPM: I think this can be inside the above `if'.
314          * Note that ext4_orphan_del() has to be able to cope with the
315          * deletion of a non-existent orphan - this is because we don't
316          * know if ext4_truncate() actually created an orphan record.
317          * (Well, we could do this if we need to, but heck - it works)
318          */
319         ext4_orphan_del(handle, inode);
320         EXT4_I(inode)->i_dtime  = (__u32)ktime_get_real_seconds();
321 
322         /*
323          * One subtle ordering requirement: if anything has gone wrong
324          * (transaction abort, IO errors, whatever), then we can still
325          * do these next steps (the fs will already have been marked as
326          * having errors), but we can't free the inode if the mark_dirty
327          * fails.
328          */
329         if (ext4_mark_inode_dirty(handle, inode))
330                 /* If that failed, just do the required in-core inode clear. */
331                 ext4_clear_inode(inode);
332         else
333                 ext4_free_inode(handle, inode);
334         ext4_journal_stop(handle);
335         sb_end_intwrite(inode->i_sb);
336         ext4_xattr_inode_array_free(ea_inode_array);
337         return;
338 no_delete:
339         ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
340 }
341 
342 #ifdef CONFIG_QUOTA
343 qsize_t *ext4_get_reserved_space(struct inode *inode)
344 {
345         return &EXT4_I(inode)->i_reserved_quota;
346 }
347 #endif
348 
349 /*
350  * Called with i_data_sem down, which is important since we can call
351  * ext4_discard_preallocations() from here.
352  */
353 void ext4_da_update_reserve_space(struct inode *inode,
354                                         int used, int quota_claim)
355 {
356         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
357         struct ext4_inode_info *ei = EXT4_I(inode);
358 
359         spin_lock(&ei->i_block_reservation_lock);
360         trace_ext4_da_update_reserve_space(inode, used, quota_claim);
361         if (unlikely(used > ei->i_reserved_data_blocks)) {
362                 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
363                          "with only %d reserved data blocks",
364                          __func__, inode->i_ino, used,
365                          ei->i_reserved_data_blocks);
366                 WARN_ON(1);
367                 used = ei->i_reserved_data_blocks;
368         }
369 
370         /* Update per-inode reservations */
371         ei->i_reserved_data_blocks -= used;
372         percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
373 
374         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
375 
376         /* Update quota subsystem for data blocks */
377         if (quota_claim)
378                 dquot_claim_block(inode, EXT4_C2B(sbi, used));
379         else {
380                 /*
381                  * We did fallocate with an offset that is already delayed
382                  * allocated. So on delayed allocated writeback we should
383                  * not re-claim the quota for fallocated blocks.
384                  */
385                 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
386         }
387 
388         /*
389          * If we have done all the pending block allocations and if
390          * there aren't any writers on the inode, we can discard the
391          * inode's preallocations.
392          */
393         if ((ei->i_reserved_data_blocks == 0) &&
394             (atomic_read(&inode->i_writecount) == 0))
395                 ext4_discard_preallocations(inode);
396 }
397 
398 static int __check_block_validity(struct inode *inode, const char *func,
399                                 unsigned int line,
400                                 struct ext4_map_blocks *map)
401 {
402         if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
403                                    map->m_len)) {
404                 ext4_error_inode(inode, func, line, map->m_pblk,
405                                  "lblock %lu mapped to illegal pblock %llu "
406                                  "(length %d)", (unsigned long) map->m_lblk,
407                                  map->m_pblk, map->m_len);
408                 return -EFSCORRUPTED;
409         }
410         return 0;
411 }
412 
413 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
414                        ext4_lblk_t len)
415 {
416         int ret;
417 
418         if (ext4_encrypted_inode(inode))
419                 return fscrypt_zeroout_range(inode, lblk, pblk, len);
420 
421         ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
422         if (ret > 0)
423                 ret = 0;
424 
425         return ret;
426 }
427 
428 #define check_block_validity(inode, map)        \
429         __check_block_validity((inode), __func__, __LINE__, (map))
430 
431 #ifdef ES_AGGRESSIVE_TEST
432 static void ext4_map_blocks_es_recheck(handle_t *handle,
433                                        struct inode *inode,
434                                        struct ext4_map_blocks *es_map,
435                                        struct ext4_map_blocks *map,
436                                        int flags)
437 {
438         int retval;
439 
440         map->m_flags = 0;
441         /*
442          * There is a race window that the result is not the same.
443          * e.g. xfstests #223 when dioread_nolock enables.  The reason
444          * is that we lookup a block mapping in extent status tree with
445          * out taking i_data_sem.  So at the time the unwritten extent
446          * could be converted.
447          */
448         down_read(&EXT4_I(inode)->i_data_sem);
449         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
450                 retval = ext4_ext_map_blocks(handle, inode, map, flags &
451                                              EXT4_GET_BLOCKS_KEEP_SIZE);
452         } else {
453                 retval = ext4_ind_map_blocks(handle, inode, map, flags &
454                                              EXT4_GET_BLOCKS_KEEP_SIZE);
455         }
456         up_read((&EXT4_I(inode)->i_data_sem));
457 
458         /*
459          * We don't check m_len because extent will be collpased in status
460          * tree.  So the m_len might not equal.
461          */
462         if (es_map->m_lblk != map->m_lblk ||
463             es_map->m_flags != map->m_flags ||
464             es_map->m_pblk != map->m_pblk) {
465                 printk("ES cache assertion failed for inode: %lu "
466                        "es_cached ex [%d/%d/%llu/%x] != "
467                        "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
468                        inode->i_ino, es_map->m_lblk, es_map->m_len,
469                        es_map->m_pblk, es_map->m_flags, map->m_lblk,
470                        map->m_len, map->m_pblk, map->m_flags,
471                        retval, flags);
472         }
473 }
474 #endif /* ES_AGGRESSIVE_TEST */
475 
476 /*
477  * The ext4_map_blocks() function tries to look up the requested blocks,
478  * and returns if the blocks are already mapped.
479  *
480  * Otherwise it takes the write lock of the i_data_sem and allocate blocks
481  * and store the allocated blocks in the result buffer head and mark it
482  * mapped.
483  *
484  * If file type is extents based, it will call ext4_ext_map_blocks(),
485  * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
486  * based files
487  *
488  * On success, it returns the number of blocks being mapped or allocated.  if
489  * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
490  * is marked as unwritten. If the create == 1, it will mark @map as mapped.
491  *
492  * It returns 0 if plain look up failed (blocks have not been allocated), in
493  * that case, @map is returned as unmapped but we still do fill map->m_len to
494  * indicate the length of a hole starting at map->m_lblk.
495  *
496  * It returns the error in case of allocation failure.
497  */
498 int ext4_map_blocks(handle_t *handle, struct inode *inode,
499                     struct ext4_map_blocks *map, int flags)
500 {
501         struct extent_status es;
502         int retval;
503         int ret = 0;
504 #ifdef ES_AGGRESSIVE_TEST
505         struct ext4_map_blocks orig_map;
506 
507         memcpy(&orig_map, map, sizeof(*map));
508 #endif
509 
510         map->m_flags = 0;
511         ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
512                   "logical block %lu\n", inode->i_ino, flags, map->m_len,
513                   (unsigned long) map->m_lblk);
514 
515         /*
516          * ext4_map_blocks returns an int, and m_len is an unsigned int
517          */
518         if (unlikely(map->m_len > INT_MAX))
519                 map->m_len = INT_MAX;
520 
521         /* We can handle the block number less than EXT_MAX_BLOCKS */
522         if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
523                 return -EFSCORRUPTED;
524 
525         /* Lookup extent status tree firstly */
526         if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
527                 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
528                         map->m_pblk = ext4_es_pblock(&es) +
529                                         map->m_lblk - es.es_lblk;
530                         map->m_flags |= ext4_es_is_written(&es) ?
531                                         EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
532                         retval = es.es_len - (map->m_lblk - es.es_lblk);
533                         if (retval > map->m_len)
534                                 retval = map->m_len;
535                         map->m_len = retval;
536                 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
537                         map->m_pblk = 0;
538                         retval = es.es_len - (map->m_lblk - es.es_lblk);
539                         if (retval > map->m_len)
540                                 retval = map->m_len;
541                         map->m_len = retval;
542                         retval = 0;
543                 } else {
544                         BUG_ON(1);
545                 }
546 #ifdef ES_AGGRESSIVE_TEST
547                 ext4_map_blocks_es_recheck(handle, inode, map,
548                                            &orig_map, flags);
549 #endif
550                 goto found;
551         }
552 
553         /*
554          * Try to see if we can get the block without requesting a new
555          * file system block.
556          */
557         down_read(&EXT4_I(inode)->i_data_sem);
558         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
559                 retval = ext4_ext_map_blocks(handle, inode, map, flags &
560                                              EXT4_GET_BLOCKS_KEEP_SIZE);
561         } else {
562                 retval = ext4_ind_map_blocks(handle, inode, map, flags &
563                                              EXT4_GET_BLOCKS_KEEP_SIZE);
564         }
565         if (retval > 0) {
566                 unsigned int status;
567 
568                 if (unlikely(retval != map->m_len)) {
569                         ext4_warning(inode->i_sb,
570                                      "ES len assertion failed for inode "
571                                      "%lu: retval %d != map->m_len %d",
572                                      inode->i_ino, retval, map->m_len);
573                         WARN_ON(1);
574                 }
575 
576                 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
577                                 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
578                 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
579                     !(status & EXTENT_STATUS_WRITTEN) &&
580                     ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
581                                        map->m_lblk + map->m_len - 1))
582                         status |= EXTENT_STATUS_DELAYED;
583                 ret = ext4_es_insert_extent(inode, map->m_lblk,
584                                             map->m_len, map->m_pblk, status);
585                 if (ret < 0)
586                         retval = ret;
587         }
588         up_read((&EXT4_I(inode)->i_data_sem));
589 
590 found:
591         if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
592                 ret = check_block_validity(inode, map);
593                 if (ret != 0)
594                         return ret;
595         }
596 
597         /* If it is only a block(s) look up */
598         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
599                 return retval;
600 
601         /*
602          * Returns if the blocks have already allocated
603          *
604          * Note that if blocks have been preallocated
605          * ext4_ext_get_block() returns the create = 0
606          * with buffer head unmapped.
607          */
608         if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
609                 /*
610                  * If we need to convert extent to unwritten
611                  * we continue and do the actual work in
612                  * ext4_ext_map_blocks()
613                  */
614                 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
615                         return retval;
616 
617         /*
618          * Here we clear m_flags because after allocating an new extent,
619          * it will be set again.
620          */
621         map->m_flags &= ~EXT4_MAP_FLAGS;
622 
623         /*
624          * New blocks allocate and/or writing to unwritten extent
625          * will possibly result in updating i_data, so we take
626          * the write lock of i_data_sem, and call get_block()
627          * with create == 1 flag.
628          */
629         down_write(&EXT4_I(inode)->i_data_sem);
630 
631         /*
632          * We need to check for EXT4 here because migrate
633          * could have changed the inode type in between
634          */
635         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
636                 retval = ext4_ext_map_blocks(handle, inode, map, flags);
637         } else {
638                 retval = ext4_ind_map_blocks(handle, inode, map, flags);
639 
640                 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
641                         /*
642                          * We allocated new blocks which will result in
643                          * i_data's format changing.  Force the migrate
644                          * to fail by clearing migrate flags
645                          */
646                         ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
647                 }
648 
649                 /*
650                  * Update reserved blocks/metadata blocks after successful
651                  * block allocation which had been deferred till now. We don't
652                  * support fallocate for non extent files. So we can update
653                  * reserve space here.
654                  */
655                 if ((retval > 0) &&
656                         (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
657                         ext4_da_update_reserve_space(inode, retval, 1);
658         }
659 
660         if (retval > 0) {
661                 unsigned int status;
662 
663                 if (unlikely(retval != map->m_len)) {
664                         ext4_warning(inode->i_sb,
665                                      "ES len assertion failed for inode "
666                                      "%lu: retval %d != map->m_len %d",
667                                      inode->i_ino, retval, map->m_len);
668                         WARN_ON(1);
669                 }
670 
671                 /*
672                  * We have to zeroout blocks before inserting them into extent
673                  * status tree. Otherwise someone could look them up there and
674                  * use them before they are really zeroed. We also have to
675                  * unmap metadata before zeroing as otherwise writeback can
676                  * overwrite zeros with stale data from block device.
677                  */
678                 if (flags & EXT4_GET_BLOCKS_ZERO &&
679                     map->m_flags & EXT4_MAP_MAPPED &&
680                     map->m_flags & EXT4_MAP_NEW) {
681                         clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
682                                            map->m_len);
683                         ret = ext4_issue_zeroout(inode, map->m_lblk,
684                                                  map->m_pblk, map->m_len);
685                         if (ret) {
686                                 retval = ret;
687                                 goto out_sem;
688                         }
689                 }
690 
691                 /*
692                  * If the extent has been zeroed out, we don't need to update
693                  * extent status tree.
694                  */
695                 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
696                     ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
697                         if (ext4_es_is_written(&es))
698                                 goto out_sem;
699                 }
700                 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
701                                 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
702                 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
703                     !(status & EXTENT_STATUS_WRITTEN) &&
704                     ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
705                                        map->m_lblk + map->m_len - 1))
706                         status |= EXTENT_STATUS_DELAYED;
707                 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
708                                             map->m_pblk, status);
709                 if (ret < 0) {
710                         retval = ret;
711                         goto out_sem;
712                 }
713         }
714 
715 out_sem:
716         up_write((&EXT4_I(inode)->i_data_sem));
717         if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
718                 ret = check_block_validity(inode, map);
719                 if (ret != 0)
720                         return ret;
721 
722                 /*
723                  * Inodes with freshly allocated blocks where contents will be
724                  * visible after transaction commit must be on transaction's
725                  * ordered data list.
726                  */
727                 if (map->m_flags & EXT4_MAP_NEW &&
728                     !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
729                     !(flags & EXT4_GET_BLOCKS_ZERO) &&
730                     !ext4_is_quota_file(inode) &&
731                     ext4_should_order_data(inode)) {
732                         if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
733                                 ret = ext4_jbd2_inode_add_wait(handle, inode);
734                         else
735                                 ret = ext4_jbd2_inode_add_write(handle, inode);
736                         if (ret)
737                                 return ret;
738                 }
739         }
740         return retval;
741 }
742 
743 /*
744  * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
745  * we have to be careful as someone else may be manipulating b_state as well.
746  */
747 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
748 {
749         unsigned long old_state;
750         unsigned long new_state;
751 
752         flags &= EXT4_MAP_FLAGS;
753 
754         /* Dummy buffer_head? Set non-atomically. */
755         if (!bh->b_page) {
756                 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
757                 return;
758         }
759         /*
760          * Someone else may be modifying b_state. Be careful! This is ugly but
761          * once we get rid of using bh as a container for mapping information
762          * to pass to / from get_block functions, this can go away.
763          */
764         do {
765                 old_state = READ_ONCE(bh->b_state);
766                 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
767         } while (unlikely(
768                  cmpxchg(&bh->b_state, old_state, new_state) != old_state));
769 }
770 
771 static int _ext4_get_block(struct inode *inode, sector_t iblock,
772                            struct buffer_head *bh, int flags)
773 {
774         struct ext4_map_blocks map;
775         int ret = 0;
776 
777         if (ext4_has_inline_data(inode))
778                 return -ERANGE;
779 
780         map.m_lblk = iblock;
781         map.m_len = bh->b_size >> inode->i_blkbits;
782 
783         ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
784                               flags);
785         if (ret > 0) {
786                 map_bh(bh, inode->i_sb, map.m_pblk);
787                 ext4_update_bh_state(bh, map.m_flags);
788                 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
789                 ret = 0;
790         } else if (ret == 0) {
791                 /* hole case, need to fill in bh->b_size */
792                 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
793         }
794         return ret;
795 }
796 
797 int ext4_get_block(struct inode *inode, sector_t iblock,
798                    struct buffer_head *bh, int create)
799 {
800         return _ext4_get_block(inode, iblock, bh,
801                                create ? EXT4_GET_BLOCKS_CREATE : 0);
802 }
803 
804 /*
805  * Get block function used when preparing for buffered write if we require
806  * creating an unwritten extent if blocks haven't been allocated.  The extent
807  * will be converted to written after the IO is complete.
808  */
809 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
810                              struct buffer_head *bh_result, int create)
811 {
812         ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
813                    inode->i_ino, create);
814         return _ext4_get_block(inode, iblock, bh_result,
815                                EXT4_GET_BLOCKS_IO_CREATE_EXT);
816 }
817 
818 /* Maximum number of blocks we map for direct IO at once. */
819 #define DIO_MAX_BLOCKS 4096
820 
821 /*
822  * Get blocks function for the cases that need to start a transaction -
823  * generally difference cases of direct IO and DAX IO. It also handles retries
824  * in case of ENOSPC.
825  */
826 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
827                                 struct buffer_head *bh_result, int flags)
828 {
829         int dio_credits;
830         handle_t *handle;
831         int retries = 0;
832         int ret;
833 
834         /* Trim mapping request to maximum we can map at once for DIO */
835         if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
836                 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
837         dio_credits = ext4_chunk_trans_blocks(inode,
838                                       bh_result->b_size >> inode->i_blkbits);
839 retry:
840         handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
841         if (IS_ERR(handle))
842                 return PTR_ERR(handle);
843 
844         ret = _ext4_get_block(inode, iblock, bh_result, flags);
845         ext4_journal_stop(handle);
846 
847         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
848                 goto retry;
849         return ret;
850 }
851 
852 /* Get block function for DIO reads and writes to inodes without extents */
853 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
854                        struct buffer_head *bh, int create)
855 {
856         /* We don't expect handle for direct IO */
857         WARN_ON_ONCE(ext4_journal_current_handle());
858 
859         if (!create)
860                 return _ext4_get_block(inode, iblock, bh, 0);
861         return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
862 }
863 
864 /*
865  * Get block function for AIO DIO writes when we create unwritten extent if
866  * blocks are not allocated yet. The extent will be converted to written
867  * after IO is complete.
868  */
869 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
870                 sector_t iblock, struct buffer_head *bh_result, int create)
871 {
872         int ret;
873 
874         /* We don't expect handle for direct IO */
875         WARN_ON_ONCE(ext4_journal_current_handle());
876 
877         ret = ext4_get_block_trans(inode, iblock, bh_result,
878                                    EXT4_GET_BLOCKS_IO_CREATE_EXT);
879 
880         /*
881          * When doing DIO using unwritten extents, we need io_end to convert
882          * unwritten extents to written on IO completion. We allocate io_end
883          * once we spot unwritten extent and store it in b_private. Generic
884          * DIO code keeps b_private set and furthermore passes the value to
885          * our completion callback in 'private' argument.
886          */
887         if (!ret && buffer_unwritten(bh_result)) {
888                 if (!bh_result->b_private) {
889                         ext4_io_end_t *io_end;
890 
891                         io_end = ext4_init_io_end(inode, GFP_KERNEL);
892                         if (!io_end)
893                                 return -ENOMEM;
894                         bh_result->b_private = io_end;
895                         ext4_set_io_unwritten_flag(inode, io_end);
896                 }
897                 set_buffer_defer_completion(bh_result);
898         }
899 
900         return ret;
901 }
902 
903 /*
904  * Get block function for non-AIO DIO writes when we create unwritten extent if
905  * blocks are not allocated yet. The extent will be converted to written
906  * after IO is complete by ext4_direct_IO_write().
907  */
908 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
909                 sector_t iblock, struct buffer_head *bh_result, int create)
910 {
911         int ret;
912 
913         /* We don't expect handle for direct IO */
914         WARN_ON_ONCE(ext4_journal_current_handle());
915 
916         ret = ext4_get_block_trans(inode, iblock, bh_result,
917                                    EXT4_GET_BLOCKS_IO_CREATE_EXT);
918 
919         /*
920          * Mark inode as having pending DIO writes to unwritten extents.
921          * ext4_direct_IO_write() checks this flag and converts extents to
922          * written.
923          */
924         if (!ret && buffer_unwritten(bh_result))
925                 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
926 
927         return ret;
928 }
929 
930 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
931                    struct buffer_head *bh_result, int create)
932 {
933         int ret;
934 
935         ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
936                    inode->i_ino, create);
937         /* We don't expect handle for direct IO */
938         WARN_ON_ONCE(ext4_journal_current_handle());
939 
940         ret = _ext4_get_block(inode, iblock, bh_result, 0);
941         /*
942          * Blocks should have been preallocated! ext4_file_write_iter() checks
943          * that.
944          */
945         WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
946 
947         return ret;
948 }
949 
950 
951 /*
952  * `handle' can be NULL if create is zero
953  */
954 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
955                                 ext4_lblk_t block, int map_flags)
956 {
957         struct ext4_map_blocks map;
958         struct buffer_head *bh;
959         int create = map_flags & EXT4_GET_BLOCKS_CREATE;
960         int err;
961 
962         J_ASSERT(handle != NULL || create == 0);
963 
964         map.m_lblk = block;
965         map.m_len = 1;
966         err = ext4_map_blocks(handle, inode, &map, map_flags);
967 
968         if (err == 0)
969                 return create ? ERR_PTR(-ENOSPC) : NULL;
970         if (err < 0)
971                 return ERR_PTR(err);
972 
973         bh = sb_getblk(inode->i_sb, map.m_pblk);
974         if (unlikely(!bh))
975                 return ERR_PTR(-ENOMEM);
976         if (map.m_flags & EXT4_MAP_NEW) {
977                 J_ASSERT(create != 0);
978                 J_ASSERT(handle != NULL);
979 
980                 /*
981                  * Now that we do not always journal data, we should
982                  * keep in mind whether this should always journal the
983                  * new buffer as metadata.  For now, regular file
984                  * writes use ext4_get_block instead, so it's not a
985                  * problem.
986                  */
987                 lock_buffer(bh);
988                 BUFFER_TRACE(bh, "call get_create_access");
989                 err = ext4_journal_get_create_access(handle, bh);
990                 if (unlikely(err)) {
991                         unlock_buffer(bh);
992                         goto errout;
993                 }
994                 if (!buffer_uptodate(bh)) {
995                         memset(bh->b_data, 0, inode->i_sb->s_blocksize);
996                         set_buffer_uptodate(bh);
997                 }
998                 unlock_buffer(bh);
999                 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1000                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1001                 if (unlikely(err))
1002                         goto errout;
1003         } else
1004                 BUFFER_TRACE(bh, "not a new buffer");
1005         return bh;
1006 errout:
1007         brelse(bh);
1008         return ERR_PTR(err);
1009 }
1010 
1011 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1012                                ext4_lblk_t block, int map_flags)
1013 {
1014         struct buffer_head *bh;
1015 
1016         bh = ext4_getblk(handle, inode, block, map_flags);
1017         if (IS_ERR(bh))
1018                 return bh;
1019         if (!bh || buffer_uptodate(bh))
1020                 return bh;
1021         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1022         wait_on_buffer(bh);
1023         if (buffer_uptodate(bh))
1024                 return bh;
1025         put_bh(bh);
1026         return ERR_PTR(-EIO);
1027 }
1028 
1029 /* Read a contiguous batch of blocks. */
1030 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1031                      bool wait, struct buffer_head **bhs)
1032 {
1033         int i, err;
1034 
1035         for (i = 0; i < bh_count; i++) {
1036                 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1037                 if (IS_ERR(bhs[i])) {
1038                         err = PTR_ERR(bhs[i]);
1039                         bh_count = i;
1040                         goto out_brelse;
1041                 }
1042         }
1043 
1044         for (i = 0; i < bh_count; i++)
1045                 /* Note that NULL bhs[i] is valid because of holes. */
1046                 if (bhs[i] && !buffer_uptodate(bhs[i]))
1047                         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1048                                     &bhs[i]);
1049 
1050         if (!wait)
1051                 return 0;
1052 
1053         for (i = 0; i < bh_count; i++)
1054                 if (bhs[i])
1055                         wait_on_buffer(bhs[i]);
1056 
1057         for (i = 0; i < bh_count; i++) {
1058                 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1059                         err = -EIO;
1060                         goto out_brelse;
1061                 }
1062         }
1063         return 0;
1064 
1065 out_brelse:
1066         for (i = 0; i < bh_count; i++) {
1067                 brelse(bhs[i]);
1068                 bhs[i] = NULL;
1069         }
1070         return err;
1071 }
1072 
1073 int ext4_walk_page_buffers(handle_t *handle,
1074                            struct buffer_head *head,
1075                            unsigned from,
1076                            unsigned to,
1077                            int *partial,
1078                            int (*fn)(handle_t *handle,
1079                                      struct buffer_head *bh))
1080 {
1081         struct buffer_head *bh;
1082         unsigned block_start, block_end;
1083         unsigned blocksize = head->b_size;
1084         int err, ret = 0;
1085         struct buffer_head *next;
1086 
1087         for (bh = head, block_start = 0;
1088              ret == 0 && (bh != head || !block_start);
1089              block_start = block_end, bh = next) {
1090                 next = bh->b_this_page;
1091                 block_end = block_start + blocksize;
1092                 if (block_end <= from || block_start >= to) {
1093                         if (partial && !buffer_uptodate(bh))
1094                                 *partial = 1;
1095                         continue;
1096                 }
1097                 err = (*fn)(handle, bh);
1098                 if (!ret)
1099                         ret = err;
1100         }
1101         return ret;
1102 }
1103 
1104 /*
1105  * To preserve ordering, it is essential that the hole instantiation and
1106  * the data write be encapsulated in a single transaction.  We cannot
1107  * close off a transaction and start a new one between the ext4_get_block()
1108  * and the commit_write().  So doing the jbd2_journal_start at the start of
1109  * prepare_write() is the right place.
1110  *
1111  * Also, this function can nest inside ext4_writepage().  In that case, we
1112  * *know* that ext4_writepage() has generated enough buffer credits to do the
1113  * whole page.  So we won't block on the journal in that case, which is good,
1114  * because the caller may be PF_MEMALLOC.
1115  *
1116  * By accident, ext4 can be reentered when a transaction is open via
1117  * quota file writes.  If we were to commit the transaction while thus
1118  * reentered, there can be a deadlock - we would be holding a quota
1119  * lock, and the commit would never complete if another thread had a
1120  * transaction open and was blocking on the quota lock - a ranking
1121  * violation.
1122  *
1123  * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1124  * will _not_ run commit under these circumstances because handle->h_ref
1125  * is elevated.  We'll still have enough credits for the tiny quotafile
1126  * write.
1127  */
1128 int do_journal_get_write_access(handle_t *handle,
1129                                 struct buffer_head *bh)
1130 {
1131         int dirty = buffer_dirty(bh);
1132         int ret;
1133 
1134         if (!buffer_mapped(bh) || buffer_freed(bh))
1135                 return 0;
1136         /*
1137          * __block_write_begin() could have dirtied some buffers. Clean
1138          * the dirty bit as jbd2_journal_get_write_access() could complain
1139          * otherwise about fs integrity issues. Setting of the dirty bit
1140          * by __block_write_begin() isn't a real problem here as we clear
1141          * the bit before releasing a page lock and thus writeback cannot
1142          * ever write the buffer.
1143          */
1144         if (dirty)
1145                 clear_buffer_dirty(bh);
1146         BUFFER_TRACE(bh, "get write access");
1147         ret = ext4_journal_get_write_access(handle, bh);
1148         if (!ret && dirty)
1149                 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1150         return ret;
1151 }
1152 
1153 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1154 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1155                                   get_block_t *get_block)
1156 {
1157         unsigned from = pos & (PAGE_SIZE - 1);
1158         unsigned to = from + len;
1159         struct inode *inode = page->mapping->host;
1160         unsigned block_start, block_end;
1161         sector_t block;
1162         int err = 0;
1163         unsigned blocksize = inode->i_sb->s_blocksize;
1164         unsigned bbits;
1165         struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1166         bool decrypt = false;
1167 
1168         BUG_ON(!PageLocked(page));
1169         BUG_ON(from > PAGE_SIZE);
1170         BUG_ON(to > PAGE_SIZE);
1171         BUG_ON(from > to);
1172 
1173         if (!page_has_buffers(page))
1174                 create_empty_buffers(page, blocksize, 0);
1175         head = page_buffers(page);
1176         bbits = ilog2(blocksize);
1177         block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1178 
1179         for (bh = head, block_start = 0; bh != head || !block_start;
1180             block++, block_start = block_end, bh = bh->b_this_page) {
1181                 block_end = block_start + blocksize;
1182                 if (block_end <= from || block_start >= to) {
1183                         if (PageUptodate(page)) {
1184                                 if (!buffer_uptodate(bh))
1185                                         set_buffer_uptodate(bh);
1186                         }
1187                         continue;
1188                 }
1189                 if (buffer_new(bh))
1190                         clear_buffer_new(bh);
1191                 if (!buffer_mapped(bh)) {
1192                         WARN_ON(bh->b_size != blocksize);
1193                         err = get_block(inode, block, bh, 1);
1194                         if (err)
1195                                 break;
1196                         if (buffer_new(bh)) {
1197                                 clean_bdev_bh_alias(bh);
1198                                 if (PageUptodate(page)) {
1199                                         clear_buffer_new(bh);
1200                                         set_buffer_uptodate(bh);
1201                                         mark_buffer_dirty(bh);
1202                                         continue;
1203                                 }
1204                                 if (block_end > to || block_start < from)
1205                                         zero_user_segments(page, to, block_end,
1206                                                            block_start, from);
1207                                 continue;
1208                         }
1209                 }
1210                 if (PageUptodate(page)) {
1211                         if (!buffer_uptodate(bh))
1212                                 set_buffer_uptodate(bh);
1213                         continue;
1214                 }
1215                 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1216                     !buffer_unwritten(bh) &&
1217                     (block_start < from || block_end > to)) {
1218                         ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1219                         *wait_bh++ = bh;
1220                         decrypt = ext4_encrypted_inode(inode) &&
1221                                 S_ISREG(inode->i_mode);
1222                 }
1223         }
1224         /*
1225          * If we issued read requests, let them complete.
1226          */
1227         while (wait_bh > wait) {
1228                 wait_on_buffer(*--wait_bh);
1229                 if (!buffer_uptodate(*wait_bh))
1230                         err = -EIO;
1231         }
1232         if (unlikely(err))
1233                 page_zero_new_buffers(page, from, to);
1234         else if (decrypt)
1235                 err = fscrypt_decrypt_page(page->mapping->host, page,
1236                                 PAGE_SIZE, 0, page->index);
1237         return err;
1238 }
1239 #endif
1240 
1241 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1242                             loff_t pos, unsigned len, unsigned flags,
1243                             struct page **pagep, void **fsdata)
1244 {
1245         struct inode *inode = mapping->host;
1246         int ret, needed_blocks;
1247         handle_t *handle;
1248         int retries = 0;
1249         struct page *page;
1250         pgoff_t index;
1251         unsigned from, to;
1252 
1253         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1254                 return -EIO;
1255 
1256         trace_ext4_write_begin(inode, pos, len, flags);
1257         /*
1258          * Reserve one block more for addition to orphan list in case
1259          * we allocate blocks but write fails for some reason
1260          */
1261         needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1262         index = pos >> PAGE_SHIFT;
1263         from = pos & (PAGE_SIZE - 1);
1264         to = from + len;
1265 
1266         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1267                 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1268                                                     flags, pagep);
1269                 if (ret < 0)
1270                         return ret;
1271                 if (ret == 1)
1272                         return 0;
1273         }
1274 
1275         /*
1276          * grab_cache_page_write_begin() can take a long time if the
1277          * system is thrashing due to memory pressure, or if the page
1278          * is being written back.  So grab it first before we start
1279          * the transaction handle.  This also allows us to allocate
1280          * the page (if needed) without using GFP_NOFS.
1281          */
1282 retry_grab:
1283         page = grab_cache_page_write_begin(mapping, index, flags);
1284         if (!page)
1285                 return -ENOMEM;
1286         unlock_page(page);
1287 
1288 retry_journal:
1289         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1290         if (IS_ERR(handle)) {
1291                 put_page(page);
1292                 return PTR_ERR(handle);
1293         }
1294 
1295         lock_page(page);
1296         if (page->mapping != mapping) {
1297                 /* The page got truncated from under us */
1298                 unlock_page(page);
1299                 put_page(page);
1300                 ext4_journal_stop(handle);
1301                 goto retry_grab;
1302         }
1303         /* In case writeback began while the page was unlocked */
1304         wait_for_stable_page(page);
1305 
1306 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1307         if (ext4_should_dioread_nolock(inode))
1308                 ret = ext4_block_write_begin(page, pos, len,
1309                                              ext4_get_block_unwritten);
1310         else
1311                 ret = ext4_block_write_begin(page, pos, len,
1312                                              ext4_get_block);
1313 #else
1314         if (ext4_should_dioread_nolock(inode))
1315                 ret = __block_write_begin(page, pos, len,
1316                                           ext4_get_block_unwritten);
1317         else
1318                 ret = __block_write_begin(page, pos, len, ext4_get_block);
1319 #endif
1320         if (!ret && ext4_should_journal_data(inode)) {
1321                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1322                                              from, to, NULL,
1323                                              do_journal_get_write_access);
1324         }
1325 
1326         if (ret) {
1327                 unlock_page(page);
1328                 /*
1329                  * __block_write_begin may have instantiated a few blocks
1330                  * outside i_size.  Trim these off again. Don't need
1331                  * i_size_read because we hold i_mutex.
1332                  *
1333                  * Add inode to orphan list in case we crash before
1334                  * truncate finishes
1335                  */
1336                 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1337                         ext4_orphan_add(handle, inode);
1338 
1339                 ext4_journal_stop(handle);
1340                 if (pos + len > inode->i_size) {
1341                         ext4_truncate_failed_write(inode);
1342                         /*
1343                          * If truncate failed early the inode might
1344                          * still be on the orphan list; we need to
1345                          * make sure the inode is removed from the
1346                          * orphan list in that case.
1347                          */
1348                         if (inode->i_nlink)
1349                                 ext4_orphan_del(NULL, inode);
1350                 }
1351 
1352                 if (ret == -ENOSPC &&
1353                     ext4_should_retry_alloc(inode->i_sb, &retries))
1354                         goto retry_journal;
1355                 put_page(page);
1356                 return ret;
1357         }
1358         *pagep = page;
1359         return ret;
1360 }
1361 
1362 /* For write_end() in data=journal mode */
1363 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1364 {
1365         int ret;
1366         if (!buffer_mapped(bh) || buffer_freed(bh))
1367                 return 0;
1368         set_buffer_uptodate(bh);
1369         ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1370         clear_buffer_meta(bh);
1371         clear_buffer_prio(bh);
1372         return ret;
1373 }
1374 
1375 /*
1376  * We need to pick up the new inode size which generic_commit_write gave us
1377  * `file' can be NULL - eg, when called from page_symlink().
1378  *
1379  * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1380  * buffers are managed internally.
1381  */
1382 static int ext4_write_end(struct file *file,
1383                           struct address_space *mapping,
1384                           loff_t pos, unsigned len, unsigned copied,
1385                           struct page *page, void *fsdata)
1386 {
1387         handle_t *handle = ext4_journal_current_handle();
1388         struct inode *inode = mapping->host;
1389         loff_t old_size = inode->i_size;
1390         int ret = 0, ret2;
1391         int i_size_changed = 0;
1392         int inline_data = ext4_has_inline_data(inode);
1393 
1394         trace_ext4_write_end(inode, pos, len, copied);
1395         if (inline_data) {
1396                 ret = ext4_write_inline_data_end(inode, pos, len,
1397                                                  copied, page);
1398                 if (ret < 0) {
1399                         unlock_page(page);
1400                         put_page(page);
1401                         goto errout;
1402                 }
1403                 copied = ret;
1404         } else
1405                 copied = block_write_end(file, mapping, pos,
1406                                          len, copied, page, fsdata);
1407         /*
1408          * it's important to update i_size while still holding page lock:
1409          * page writeout could otherwise come in and zero beyond i_size.
1410          */
1411         i_size_changed = ext4_update_inode_size(inode, pos + copied);
1412         unlock_page(page);
1413         put_page(page);
1414 
1415         if (old_size < pos)
1416                 pagecache_isize_extended(inode, old_size, pos);
1417         /*
1418          * Don't mark the inode dirty under page lock. First, it unnecessarily
1419          * makes the holding time of page lock longer. Second, it forces lock
1420          * ordering of page lock and transaction start for journaling
1421          * filesystems.
1422          */
1423         if (i_size_changed || inline_data)
1424                 ext4_mark_inode_dirty(handle, inode);
1425 
1426         if (pos + len > inode->i_size && ext4_can_truncate(inode))
1427                 /* if we have allocated more blocks and copied
1428                  * less. We will have blocks allocated outside
1429                  * inode->i_size. So truncate them
1430                  */
1431                 ext4_orphan_add(handle, inode);
1432 errout:
1433         ret2 = ext4_journal_stop(handle);
1434         if (!ret)
1435                 ret = ret2;
1436 
1437         if (pos + len > inode->i_size) {
1438                 ext4_truncate_failed_write(inode);
1439                 /*
1440                  * If truncate failed early the inode might still be
1441                  * on the orphan list; we need to make sure the inode
1442                  * is removed from the orphan list in that case.
1443                  */
1444                 if (inode->i_nlink)
1445                         ext4_orphan_del(NULL, inode);
1446         }
1447 
1448         return ret ? ret : copied;
1449 }
1450 
1451 /*
1452  * This is a private version of page_zero_new_buffers() which doesn't
1453  * set the buffer to be dirty, since in data=journalled mode we need
1454  * to call ext4_handle_dirty_metadata() instead.
1455  */
1456 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1457                                             struct page *page,
1458                                             unsigned from, unsigned to)
1459 {
1460         unsigned int block_start = 0, block_end;
1461         struct buffer_head *head, *bh;
1462 
1463         bh = head = page_buffers(page);
1464         do {
1465                 block_end = block_start + bh->b_size;
1466                 if (buffer_new(bh)) {
1467                         if (block_end > from && block_start < to) {
1468                                 if (!PageUptodate(page)) {
1469                                         unsigned start, size;
1470 
1471                                         start = max(from, block_start);
1472                                         size = min(to, block_end) - start;
1473 
1474                                         zero_user(page, start, size);
1475                                         write_end_fn(handle, bh);
1476                                 }
1477                                 clear_buffer_new(bh);
1478                         }
1479                 }
1480                 block_start = block_end;
1481                 bh = bh->b_this_page;
1482         } while (bh != head);
1483 }
1484 
1485 static int ext4_journalled_write_end(struct file *file,
1486                                      struct address_space *mapping,
1487                                      loff_t pos, unsigned len, unsigned copied,
1488                                      struct page *page, void *fsdata)
1489 {
1490         handle_t *handle = ext4_journal_current_handle();
1491         struct inode *inode = mapping->host;
1492         loff_t old_size = inode->i_size;
1493         int ret = 0, ret2;
1494         int partial = 0;
1495         unsigned from, to;
1496         int size_changed = 0;
1497         int inline_data = ext4_has_inline_data(inode);
1498 
1499         trace_ext4_journalled_write_end(inode, pos, len, copied);
1500         from = pos & (PAGE_SIZE - 1);
1501         to = from + len;
1502 
1503         BUG_ON(!ext4_handle_valid(handle));
1504 
1505         if (inline_data) {
1506                 ret = ext4_write_inline_data_end(inode, pos, len,
1507                                                  copied, page);
1508                 if (ret < 0) {
1509                         unlock_page(page);
1510                         put_page(page);
1511                         goto errout;
1512                 }
1513                 copied = ret;
1514         } else if (unlikely(copied < len) && !PageUptodate(page)) {
1515                 copied = 0;
1516                 ext4_journalled_zero_new_buffers(handle, page, from, to);
1517         } else {
1518                 if (unlikely(copied < len))
1519                         ext4_journalled_zero_new_buffers(handle, page,
1520                                                          from + copied, to);
1521                 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1522                                              from + copied, &partial,
1523                                              write_end_fn);
1524                 if (!partial)
1525                         SetPageUptodate(page);
1526         }
1527         size_changed = ext4_update_inode_size(inode, pos + copied);
1528         ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1529         EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1530         unlock_page(page);
1531         put_page(page);
1532 
1533         if (old_size < pos)
1534                 pagecache_isize_extended(inode, old_size, pos);
1535 
1536         if (size_changed || inline_data) {
1537                 ret2 = ext4_mark_inode_dirty(handle, inode);
1538                 if (!ret)
1539                         ret = ret2;
1540         }
1541 
1542         if (pos + len > inode->i_size && ext4_can_truncate(inode))
1543                 /* if we have allocated more blocks and copied
1544                  * less. We will have blocks allocated outside
1545                  * inode->i_size. So truncate them
1546                  */
1547                 ext4_orphan_add(handle, inode);
1548 
1549 errout:
1550         ret2 = ext4_journal_stop(handle);
1551         if (!ret)
1552                 ret = ret2;
1553         if (pos + len > inode->i_size) {
1554                 ext4_truncate_failed_write(inode);
1555                 /*
1556                  * If truncate failed early the inode might still be
1557                  * on the orphan list; we need to make sure the inode
1558                  * is removed from the orphan list in that case.
1559                  */
1560                 if (inode->i_nlink)
1561                         ext4_orphan_del(NULL, inode);
1562         }
1563 
1564         return ret ? ret : copied;
1565 }
1566 
1567 /*
1568  * Reserve space for a single cluster
1569  */
1570 static int ext4_da_reserve_space(struct inode *inode)
1571 {
1572         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1573         struct ext4_inode_info *ei = EXT4_I(inode);
1574         int ret;
1575 
1576         /*
1577          * We will charge metadata quota at writeout time; this saves
1578          * us from metadata over-estimation, though we may go over by
1579          * a small amount in the end.  Here we just reserve for data.
1580          */
1581         ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1582         if (ret)
1583                 return ret;
1584 
1585         spin_lock(&ei->i_block_reservation_lock);
1586         if (ext4_claim_free_clusters(sbi, 1, 0)) {
1587                 spin_unlock(&ei->i_block_reservation_lock);
1588                 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1589                 return -ENOSPC;
1590         }
1591         ei->i_reserved_data_blocks++;
1592         trace_ext4_da_reserve_space(inode);
1593         spin_unlock(&ei->i_block_reservation_lock);
1594 
1595         return 0;       /* success */
1596 }
1597 
1598 void ext4_da_release_space(struct inode *inode, int to_free)
1599 {
1600         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1601         struct ext4_inode_info *ei = EXT4_I(inode);
1602 
1603         if (!to_free)
1604                 return;         /* Nothing to release, exit */
1605 
1606         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1607 
1608         trace_ext4_da_release_space(inode, to_free);
1609         if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1610                 /*
1611                  * if there aren't enough reserved blocks, then the
1612                  * counter is messed up somewhere.  Since this
1613                  * function is called from invalidate page, it's
1614                  * harmless to return without any action.
1615                  */
1616                 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1617                          "ino %lu, to_free %d with only %d reserved "
1618                          "data blocks", inode->i_ino, to_free,
1619                          ei->i_reserved_data_blocks);
1620                 WARN_ON(1);
1621                 to_free = ei->i_reserved_data_blocks;
1622         }
1623         ei->i_reserved_data_blocks -= to_free;
1624 
1625         /* update fs dirty data blocks counter */
1626         percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1627 
1628         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1629 
1630         dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1631 }
1632 
1633 static void ext4_da_page_release_reservation(struct page *page,
1634                                              unsigned int offset,
1635                                              unsigned int length)
1636 {
1637         int contiguous_blks = 0;
1638         struct buffer_head *head, *bh;
1639         unsigned int curr_off = 0;
1640         struct inode *inode = page->mapping->host;
1641         unsigned int stop = offset + length;
1642         ext4_fsblk_t lblk;
1643 
1644         BUG_ON(stop > PAGE_SIZE || stop < length);
1645 
1646         head = page_buffers(page);
1647         bh = head;
1648         do {
1649                 unsigned int next_off = curr_off + bh->b_size;
1650 
1651                 if (next_off > stop)
1652                         break;
1653 
1654                 if ((offset <= curr_off) && (buffer_delay(bh))) {
1655                         contiguous_blks++;
1656                         clear_buffer_delay(bh);
1657                 } else if (contiguous_blks) {
1658                         lblk = page->index <<
1659                                (PAGE_SHIFT - inode->i_blkbits);
1660                         lblk += (curr_off >> inode->i_blkbits) -
1661                                 contiguous_blks;
1662                         ext4_es_remove_blks(inode, lblk, contiguous_blks);
1663                         contiguous_blks = 0;
1664                 }
1665                 curr_off = next_off;
1666         } while ((bh = bh->b_this_page) != head);
1667 
1668         if (contiguous_blks) {
1669                 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1670                 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1671                 ext4_es_remove_blks(inode, lblk, contiguous_blks);
1672         }
1673 
1674 }
1675 
1676 /*
1677  * Delayed allocation stuff
1678  */
1679 
1680 struct mpage_da_data {
1681         struct inode *inode;
1682         struct writeback_control *wbc;
1683 
1684         pgoff_t first_page;     /* The first page to write */
1685         pgoff_t next_page;      /* Current page to examine */
1686         pgoff_t last_page;      /* Last page to examine */
1687         /*
1688          * Extent to map - this can be after first_page because that can be
1689          * fully mapped. We somewhat abuse m_flags to store whether the extent
1690          * is delalloc or unwritten.
1691          */
1692         struct ext4_map_blocks map;
1693         struct ext4_io_submit io_submit;        /* IO submission data */
1694         unsigned int do_map:1;
1695 };
1696 
1697 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1698                                        bool invalidate)
1699 {
1700         int nr_pages, i;
1701         pgoff_t index, end;
1702         struct pagevec pvec;
1703         struct inode *inode = mpd->inode;
1704         struct address_space *mapping = inode->i_mapping;
1705 
1706         /* This is necessary when next_page == 0. */
1707         if (mpd->first_page >= mpd->next_page)
1708                 return;
1709 
1710         index = mpd->first_page;
1711         end   = mpd->next_page - 1;
1712         if (invalidate) {
1713                 ext4_lblk_t start, last;
1714                 start = index << (PAGE_SHIFT - inode->i_blkbits);
1715                 last = end << (PAGE_SHIFT - inode->i_blkbits);
1716                 ext4_es_remove_extent(inode, start, last - start + 1);
1717         }
1718 
1719         pagevec_init(&pvec);
1720         while (index <= end) {
1721                 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1722                 if (nr_pages == 0)
1723                         break;
1724                 for (i = 0; i < nr_pages; i++) {
1725                         struct page *page = pvec.pages[i];
1726 
1727                         BUG_ON(!PageLocked(page));
1728                         BUG_ON(PageWriteback(page));
1729                         if (invalidate) {
1730                                 if (page_mapped(page))
1731                                         clear_page_dirty_for_io(page);
1732                                 block_invalidatepage(page, 0, PAGE_SIZE);
1733                                 ClearPageUptodate(page);
1734                         }
1735                         unlock_page(page);
1736                 }
1737                 pagevec_release(&pvec);
1738         }
1739 }
1740 
1741 static void ext4_print_free_blocks(struct inode *inode)
1742 {
1743         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1744         struct super_block *sb = inode->i_sb;
1745         struct ext4_inode_info *ei = EXT4_I(inode);
1746 
1747         ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1748                EXT4_C2B(EXT4_SB(inode->i_sb),
1749                         ext4_count_free_clusters(sb)));
1750         ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1751         ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1752                (long long) EXT4_C2B(EXT4_SB(sb),
1753                 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1754         ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1755                (long long) EXT4_C2B(EXT4_SB(sb),
1756                 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1757         ext4_msg(sb, KERN_CRIT, "Block reservation details");
1758         ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1759                  ei->i_reserved_data_blocks);
1760         return;
1761 }
1762 
1763 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1764 {
1765         return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1766 }
1767 
1768 /*
1769  * ext4_insert_delayed_block - adds a delayed block to the extents status
1770  *                             tree, incrementing the reserved cluster/block
1771  *                             count or making a pending reservation
1772  *                             where needed
1773  *
1774  * @inode - file containing the newly added block
1775  * @lblk - logical block to be added
1776  *
1777  * Returns 0 on success, negative error code on failure.
1778  */
1779 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1780 {
1781         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1782         int ret;
1783         bool allocated = false;
1784 
1785         /*
1786          * If the cluster containing lblk is shared with a delayed,
1787          * written, or unwritten extent in a bigalloc file system, it's
1788          * already been accounted for and does not need to be reserved.
1789          * A pending reservation must be made for the cluster if it's
1790          * shared with a written or unwritten extent and doesn't already
1791          * have one.  Written and unwritten extents can be purged from the
1792          * extents status tree if the system is under memory pressure, so
1793          * it's necessary to examine the extent tree if a search of the
1794          * extents status tree doesn't get a match.
1795          */
1796         if (sbi->s_cluster_ratio == 1) {
1797                 ret = ext4_da_reserve_space(inode);
1798                 if (ret != 0)   /* ENOSPC */
1799                         goto errout;
1800         } else {   /* bigalloc */
1801                 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1802                         if (!ext4_es_scan_clu(inode,
1803                                               &ext4_es_is_mapped, lblk)) {
1804                                 ret = ext4_clu_mapped(inode,
1805                                                       EXT4_B2C(sbi, lblk));
1806                                 if (ret < 0)
1807                                         goto errout;
1808                                 if (ret == 0) {
1809                                         ret = ext4_da_reserve_space(inode);
1810                                         if (ret != 0)   /* ENOSPC */
1811                                                 goto errout;
1812                                 } else {
1813                                         allocated = true;
1814                                 }
1815                         } else {
1816                                 allocated = true;
1817                         }
1818                 }
1819         }
1820 
1821         ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1822 
1823 errout:
1824         return ret;
1825 }
1826 
1827 /*
1828  * This function is grabs code from the very beginning of
1829  * ext4_map_blocks, but assumes that the caller is from delayed write
1830  * time. This function looks up the requested blocks and sets the
1831  * buffer delay bit under the protection of i_data_sem.
1832  */
1833 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1834                               struct ext4_map_blocks *map,
1835                               struct buffer_head *bh)
1836 {
1837         struct extent_status es;
1838         int retval;
1839         sector_t invalid_block = ~((sector_t) 0xffff);
1840 #ifdef ES_AGGRESSIVE_TEST
1841         struct ext4_map_blocks orig_map;
1842 
1843         memcpy(&orig_map, map, sizeof(*map));
1844 #endif
1845 
1846         if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1847                 invalid_block = ~0;
1848 
1849         map->m_flags = 0;
1850         ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1851                   "logical block %lu\n", inode->i_ino, map->m_len,
1852                   (unsigned long) map->m_lblk);
1853 
1854         /* Lookup extent status tree firstly */
1855         if (ext4_es_lookup_extent(inode, iblock, &es)) {
1856                 if (ext4_es_is_hole(&es)) {
1857                         retval = 0;
1858                         down_read(&EXT4_I(inode)->i_data_sem);
1859                         goto add_delayed;
1860                 }
1861 
1862                 /*
1863                  * Delayed extent could be allocated by fallocate.
1864                  * So we need to check it.
1865                  */
1866                 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1867                         map_bh(bh, inode->i_sb, invalid_block);
1868                         set_buffer_new(bh);
1869                         set_buffer_delay(bh);
1870                         return 0;
1871                 }
1872 
1873                 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1874                 retval = es.es_len - (iblock - es.es_lblk);
1875                 if (retval > map->m_len)
1876                         retval = map->m_len;
1877                 map->m_len = retval;
1878                 if (ext4_es_is_written(&es))
1879                         map->m_flags |= EXT4_MAP_MAPPED;
1880                 else if (ext4_es_is_unwritten(&es))
1881                         map->m_flags |= EXT4_MAP_UNWRITTEN;
1882                 else
1883                         BUG_ON(1);
1884 
1885 #ifdef ES_AGGRESSIVE_TEST
1886                 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1887 #endif
1888                 return retval;
1889         }
1890 
1891         /*
1892          * Try to see if we can get the block without requesting a new
1893          * file system block.
1894          */
1895         down_read(&EXT4_I(inode)->i_data_sem);
1896         if (ext4_has_inline_data(inode))
1897                 retval = 0;
1898         else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1899                 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1900         else
1901                 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1902 
1903 add_delayed:
1904         if (retval == 0) {
1905                 int ret;
1906 
1907                 /*
1908                  * XXX: __block_prepare_write() unmaps passed block,
1909                  * is it OK?
1910                  */
1911 
1912                 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1913                 if (ret != 0) {
1914                         retval = ret;
1915                         goto out_unlock;
1916                 }
1917 
1918                 map_bh(bh, inode->i_sb, invalid_block);
1919                 set_buffer_new(bh);
1920                 set_buffer_delay(bh);
1921         } else if (retval > 0) {
1922                 int ret;
1923                 unsigned int status;
1924 
1925                 if (unlikely(retval != map->m_len)) {
1926                         ext4_warning(inode->i_sb,
1927                                      "ES len assertion failed for inode "
1928                                      "%lu: retval %d != map->m_len %d",
1929                                      inode->i_ino, retval, map->m_len);
1930                         WARN_ON(1);
1931                 }
1932 
1933                 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1934                                 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1935                 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1936                                             map->m_pblk, status);
1937                 if (ret != 0)
1938                         retval = ret;
1939         }
1940 
1941 out_unlock:
1942         up_read((&EXT4_I(inode)->i_data_sem));
1943 
1944         return retval;
1945 }
1946 
1947 /*
1948  * This is a special get_block_t callback which is used by
1949  * ext4_da_write_begin().  It will either return mapped block or
1950  * reserve space for a single block.
1951  *
1952  * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1953  * We also have b_blocknr = -1 and b_bdev initialized properly
1954  *
1955  * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1956  * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1957  * initialized properly.
1958  */
1959 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1960                            struct buffer_head *bh, int create)
1961 {
1962         struct ext4_map_blocks map;
1963         int ret = 0;
1964 
1965         BUG_ON(create == 0);
1966         BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1967 
1968         map.m_lblk = iblock;
1969         map.m_len = 1;
1970 
1971         /*
1972          * first, we need to know whether the block is allocated already
1973          * preallocated blocks are unmapped but should treated
1974          * the same as allocated blocks.
1975          */
1976         ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1977         if (ret <= 0)
1978                 return ret;
1979 
1980         map_bh(bh, inode->i_sb, map.m_pblk);
1981         ext4_update_bh_state(bh, map.m_flags);
1982 
1983         if (buffer_unwritten(bh)) {
1984                 /* A delayed write to unwritten bh should be marked
1985                  * new and mapped.  Mapped ensures that we don't do
1986                  * get_block multiple times when we write to the same
1987                  * offset and new ensures that we do proper zero out
1988                  * for partial write.
1989                  */
1990                 set_buffer_new(bh);
1991                 set_buffer_mapped(bh);
1992         }
1993         return 0;
1994 }
1995 
1996 static int bget_one(handle_t *handle, struct buffer_head *bh)
1997 {
1998         get_bh(bh);
1999         return 0;
2000 }
2001 
2002 static int bput_one(handle_t *handle, struct buffer_head *bh)
2003 {
2004         put_bh(bh);
2005         return 0;
2006 }
2007 
2008 static int __ext4_journalled_writepage(struct page *page,
2009                                        unsigned int len)
2010 {
2011         struct address_space *mapping = page->mapping;
2012         struct inode *inode = mapping->host;
2013         struct buffer_head *page_bufs = NULL;
2014         handle_t *handle = NULL;
2015         int ret = 0, err = 0;
2016         int inline_data = ext4_has_inline_data(inode);
2017         struct buffer_head *inode_bh = NULL;
2018 
2019         ClearPageChecked(page);
2020 
2021         if (inline_data) {
2022                 BUG_ON(page->index != 0);
2023                 BUG_ON(len > ext4_get_max_inline_size(inode));
2024                 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2025                 if (inode_bh == NULL)
2026                         goto out;
2027         } else {
2028                 page_bufs = page_buffers(page);
2029                 if (!page_bufs) {
2030                         BUG();
2031                         goto out;
2032                 }
2033                 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2034                                        NULL, bget_one);
2035         }
2036         /*
2037          * We need to release the page lock before we start the
2038          * journal, so grab a reference so the page won't disappear
2039          * out from under us.
2040          */
2041         get_page(page);
2042         unlock_page(page);
2043 
2044         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2045                                     ext4_writepage_trans_blocks(inode));
2046         if (IS_ERR(handle)) {
2047                 ret = PTR_ERR(handle);
2048                 put_page(page);
2049                 goto out_no_pagelock;
2050         }
2051         BUG_ON(!ext4_handle_valid(handle));
2052 
2053         lock_page(page);
2054         put_page(page);
2055         if (page->mapping != mapping) {
2056                 /* The page got truncated from under us */
2057                 ext4_journal_stop(handle);
2058                 ret = 0;
2059                 goto out;
2060         }
2061 
2062         if (inline_data) {
2063                 ret = ext4_mark_inode_dirty(handle, inode);
2064         } else {
2065                 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2066                                              do_journal_get_write_access);
2067 
2068                 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2069                                              write_end_fn);
2070         }
2071         if (ret == 0)
2072                 ret = err;
2073         EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2074         err = ext4_journal_stop(handle);
2075         if (!ret)
2076                 ret = err;
2077 
2078         if (!ext4_has_inline_data(inode))
2079                 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2080                                        NULL, bput_one);
2081         ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2082 out:
2083         unlock_page(page);
2084 out_no_pagelock:
2085         brelse(inode_bh);
2086         return ret;
2087 }
2088 
2089 /*
2090  * Note that we don't need to start a transaction unless we're journaling data
2091  * because we should have holes filled from ext4_page_mkwrite(). We even don't
2092  * need to file the inode to the transaction's list in ordered mode because if
2093  * we are writing back data added by write(), the inode is already there and if
2094  * we are writing back data modified via mmap(), no one guarantees in which
2095  * transaction the data will hit the disk. In case we are journaling data, we
2096  * cannot start transaction directly because transaction start ranks above page
2097  * lock so we have to do some magic.
2098  *
2099  * This function can get called via...
2100  *   - ext4_writepages after taking page lock (have journal handle)
2101  *   - journal_submit_inode_data_buffers (no journal handle)
2102  *   - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2103  *   - grab_page_cache when doing write_begin (have journal handle)
2104  *
2105  * We don't do any block allocation in this function. If we have page with
2106  * multiple blocks we need to write those buffer_heads that are mapped. This
2107  * is important for mmaped based write. So if we do with blocksize 1K
2108  * truncate(f, 1024);
2109  * a = mmap(f, 0, 4096);
2110  * a[0] = 'a';
2111  * truncate(f, 4096);
2112  * we have in the page first buffer_head mapped via page_mkwrite call back
2113  * but other buffer_heads would be unmapped but dirty (dirty done via the
2114  * do_wp_page). So writepage should write the first block. If we modify
2115  * the mmap area beyond 1024 we will again get a page_fault and the
2116  * page_mkwrite callback will do the block allocation and mark the
2117  * buffer_heads mapped.
2118  *
2119  * We redirty the page if we have any buffer_heads that is either delay or
2120  * unwritten in the page.
2121  *
2122  * We can get recursively called as show below.
2123  *
2124  *      ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2125  *              ext4_writepage()
2126  *
2127  * But since we don't do any block allocation we should not deadlock.
2128  * Page also have the dirty flag cleared so we don't get recurive page_lock.
2129  */
2130 static int ext4_writepage(struct page *page,
2131                           struct writeback_control *wbc)
2132 {
2133         int ret = 0;
2134         loff_t size;
2135         unsigned int len;
2136         struct buffer_head *page_bufs = NULL;
2137         struct inode *inode = page->mapping->host;
2138         struct ext4_io_submit io_submit;
2139         bool keep_towrite = false;
2140 
2141         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2142                 ext4_invalidatepage(page, 0, PAGE_SIZE);
2143                 unlock_page(page);
2144                 return -EIO;
2145         }
2146 
2147         trace_ext4_writepage(page);
2148         size = i_size_read(inode);
2149         if (page->index == size >> PAGE_SHIFT)
2150                 len = size & ~PAGE_MASK;
2151         else
2152                 len = PAGE_SIZE;
2153 
2154         page_bufs = page_buffers(page);
2155         /*
2156          * We cannot do block allocation or other extent handling in this
2157          * function. If there are buffers needing that, we have to redirty
2158          * the page. But we may reach here when we do a journal commit via
2159          * journal_submit_inode_data_buffers() and in that case we must write
2160          * allocated buffers to achieve data=ordered mode guarantees.
2161          *
2162          * Also, if there is only one buffer per page (the fs block
2163          * size == the page size), if one buffer needs block
2164          * allocation or needs to modify the extent tree to clear the
2165          * unwritten flag, we know that the page can't be written at
2166          * all, so we might as well refuse the write immediately.
2167          * Unfortunately if the block size != page size, we can't as
2168          * easily detect this case using ext4_walk_page_buffers(), but
2169          * for the extremely common case, this is an optimization that
2170          * skips a useless round trip through ext4_bio_write_page().
2171          */
2172         if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2173                                    ext4_bh_delay_or_unwritten)) {
2174                 redirty_page_for_writepage(wbc, page);
2175                 if ((current->flags & PF_MEMALLOC) ||
2176                     (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2177                         /*
2178                          * For memory cleaning there's no point in writing only
2179                          * some buffers. So just bail out. Warn if we came here
2180                          * from direct reclaim.
2181                          */
2182                         WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2183                                                         == PF_MEMALLOC);
2184                         unlock_page(page);
2185                         return 0;
2186                 }
2187                 keep_towrite = true;
2188         }
2189 
2190         if (PageChecked(page) && ext4_should_journal_data(inode))
2191                 /*
2192                  * It's mmapped pagecache.  Add buffers and journal it.  There
2193                  * doesn't seem much point in redirtying the page here.
2194                  */
2195                 return __ext4_journalled_writepage(page, len);
2196 
2197         ext4_io_submit_init(&io_submit, wbc);
2198         io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2199         if (!io_submit.io_end) {
2200                 redirty_page_for_writepage(wbc, page);
2201                 unlock_page(page);
2202                 return -ENOMEM;
2203         }
2204         ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2205         ext4_io_submit(&io_submit);
2206         /* Drop io_end reference we got from init */
2207         ext4_put_io_end_defer(io_submit.io_end);
2208         return ret;
2209 }
2210 
2211 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2212 {
2213         int len;
2214         loff_t size;
2215         int err;
2216 
2217         BUG_ON(page->index != mpd->first_page);
2218         clear_page_dirty_for_io(page);
2219         /*
2220          * We have to be very careful here!  Nothing protects writeback path
2221          * against i_size changes and the page can be writeably mapped into
2222          * page tables. So an application can be growing i_size and writing
2223          * data through mmap while writeback runs. clear_page_dirty_for_io()
2224          * write-protects our page in page tables and the page cannot get
2225          * written to again until we release page lock. So only after
2226          * clear_page_dirty_for_io() we are safe to sample i_size for
2227          * ext4_bio_write_page() to zero-out tail of the written page. We rely
2228          * on the barrier provided by TestClearPageDirty in
2229          * clear_page_dirty_for_io() to make sure i_size is really sampled only
2230          * after page tables are updated.
2231          */
2232         size = i_size_read(mpd->inode);
2233         if (page->index == size >> PAGE_SHIFT)
2234                 len = size & ~PAGE_MASK;
2235         else
2236                 len = PAGE_SIZE;
2237         err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2238         if (!err)
2239                 mpd->wbc->nr_to_write--;
2240         mpd->first_page++;
2241 
2242         return err;
2243 }
2244 
2245 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2246 
2247 /*
2248  * mballoc gives us at most this number of blocks...
2249  * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2250  * The rest of mballoc seems to handle chunks up to full group size.
2251  */
2252 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2253 
2254 /*
2255  * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2256  *
2257  * @mpd - extent of blocks
2258  * @lblk - logical number of the block in the file
2259  * @bh - buffer head we want to add to the extent
2260  *
2261  * The function is used to collect contig. blocks in the same state. If the
2262  * buffer doesn't require mapping for writeback and we haven't started the
2263  * extent of buffers to map yet, the function returns 'true' immediately - the
2264  * caller can write the buffer right away. Otherwise the function returns true
2265  * if the block has been added to the extent, false if the block couldn't be
2266  * added.
2267  */
2268 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2269                                    struct buffer_head *bh)
2270 {
2271         struct ext4_map_blocks *map = &mpd->map;
2272 
2273         /* Buffer that doesn't need mapping for writeback? */
2274         if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2275             (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2276                 /* So far no extent to map => we write the buffer right away */
2277                 if (map->m_len == 0)
2278                         return true;
2279                 return false;
2280         }
2281 
2282         /* First block in the extent? */
2283         if (map->m_len == 0) {
2284                 /* We cannot map unless handle is started... */
2285                 if (!mpd->do_map)
2286                         return false;
2287                 map->m_lblk = lblk;
2288                 map->m_len = 1;
2289                 map->m_flags = bh->b_state & BH_FLAGS;
2290                 return true;
2291         }
2292 
2293         /* Don't go larger than mballoc is willing to allocate */
2294         if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2295                 return false;
2296 
2297         /* Can we merge the block to our big extent? */
2298         if (lblk == map->m_lblk + map->m_len &&
2299             (bh->b_state & BH_FLAGS) == map->m_flags) {
2300                 map->m_len++;
2301                 return true;
2302         }
2303         return false;
2304 }
2305 
2306 /*
2307  * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2308  *
2309  * @mpd - extent of blocks for mapping
2310  * @head - the first buffer in the page
2311  * @bh - buffer we should start processing from
2312  * @lblk - logical number of the block in the file corresponding to @bh
2313  *
2314  * Walk through page buffers from @bh upto @head (exclusive) and either submit
2315  * the page for IO if all buffers in this page were mapped and there's no
2316  * accumulated extent of buffers to map or add buffers in the page to the
2317  * extent of buffers to map. The function returns 1 if the caller can continue
2318  * by processing the next page, 0 if it should stop adding buffers to the
2319  * extent to map because we cannot extend it anymore. It can also return value
2320  * < 0 in case of error during IO submission.
2321  */
2322 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2323                                    struct buffer_head *head,
2324                                    struct buffer_head *bh,
2325                                    ext4_lblk_t lblk)
2326 {
2327         struct inode *inode = mpd->inode;
2328         int err;
2329         ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2330                                                         >> inode->i_blkbits;
2331 
2332         do {
2333                 BUG_ON(buffer_locked(bh));
2334 
2335                 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2336                         /* Found extent to map? */
2337                         if (mpd->map.m_len)
2338                                 return 0;
2339                         /* Buffer needs mapping and handle is not started? */
2340                         if (!mpd->do_map)
2341                                 return 0;
2342                         /* Everything mapped so far and we hit EOF */
2343                         break;
2344                 }
2345         } while (lblk++, (bh = bh->b_this_page) != head);
2346         /* So far everything mapped? Submit the page for IO. */
2347         if (mpd->map.m_len == 0) {
2348                 err = mpage_submit_page(mpd, head->b_page);
2349                 if (err < 0)
2350                         return err;
2351         }
2352         return lblk < blocks;
2353 }
2354 
2355 /*
2356  * mpage_map_buffers - update buffers corresponding to changed extent and
2357  *                     submit fully mapped pages for IO
2358  *
2359  * @mpd - description of extent to map, on return next extent to map
2360  *
2361  * Scan buffers corresponding to changed extent (we expect corresponding pages
2362  * to be already locked) and update buffer state according to new extent state.
2363  * We map delalloc buffers to their physical location, clear unwritten bits,
2364  * and mark buffers as uninit when we perform writes to unwritten extents
2365  * and do extent conversion after IO is finished. If the last page is not fully
2366  * mapped, we update @map to the next extent in the last page that needs
2367  * mapping. Otherwise we submit the page for IO.
2368  */
2369 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2370 {
2371         struct pagevec pvec;
2372         int nr_pages, i;
2373         struct inode *inode = mpd->inode;
2374         struct buffer_head *head, *bh;
2375         int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2376         pgoff_t start, end;
2377         ext4_lblk_t lblk;
2378         sector_t pblock;
2379         int err;
2380 
2381         start = mpd->map.m_lblk >> bpp_bits;
2382         end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2383         lblk = start << bpp_bits;
2384         pblock = mpd->map.m_pblk;
2385 
2386         pagevec_init(&pvec);
2387         while (start <= end) {
2388                 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2389                                                 &start, end);
2390                 if (nr_pages == 0)
2391                         break;
2392                 for (i = 0; i < nr_pages; i++) {
2393                         struct page *page = pvec.pages[i];
2394 
2395                         bh = head = page_buffers(page);
2396                         do {
2397                                 if (lblk < mpd->map.m_lblk)
2398                                         continue;
2399                                 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2400                                         /*
2401                                          * Buffer after end of mapped extent.
2402                                          * Find next buffer in the page to map.
2403                                          */
2404                                         mpd->map.m_len = 0;
2405                                         mpd->map.m_flags = 0;
2406                                         /*
2407                                          * FIXME: If dioread_nolock supports
2408                                          * blocksize < pagesize, we need to make
2409                                          * sure we add size mapped so far to
2410                                          * io_end->size as the following call
2411                                          * can submit the page for IO.
2412                                          */
2413                                         err = mpage_process_page_bufs(mpd, head,
2414                                                                       bh, lblk);
2415                                         pagevec_release(&pvec);
2416                                         if (err > 0)
2417                                                 err = 0;
2418                                         return err;
2419                                 }
2420                                 if (buffer_delay(bh)) {
2421                                         clear_buffer_delay(bh);
2422                                         bh->b_blocknr = pblock++;
2423                                 }
2424                                 clear_buffer_unwritten(bh);
2425                         } while (lblk++, (bh = bh->b_this_page) != head);
2426 
2427                         /*
2428                          * FIXME: This is going to break if dioread_nolock
2429                          * supports blocksize < pagesize as we will try to
2430                          * convert potentially unmapped parts of inode.
2431                          */
2432                         mpd->io_submit.io_end->size += PAGE_SIZE;
2433                         /* Page fully mapped - let IO run! */
2434                         err = mpage_submit_page(mpd, page);
2435                         if (err < 0) {
2436                                 pagevec_release(&pvec);
2437                                 return err;
2438                         }
2439                 }
2440                 pagevec_release(&pvec);
2441         }
2442         /* Extent fully mapped and matches with page boundary. We are done. */
2443         mpd->map.m_len = 0;
2444         mpd->map.m_flags = 0;
2445         return 0;
2446 }
2447 
2448 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2449 {
2450         struct inode *inode = mpd->inode;
2451         struct ext4_map_blocks *map = &mpd->map;
2452         int get_blocks_flags;
2453         int err, dioread_nolock;
2454 
2455         trace_ext4_da_write_pages_extent(inode, map);
2456         /*
2457          * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2458          * to convert an unwritten extent to be initialized (in the case
2459          * where we have written into one or more preallocated blocks).  It is
2460          * possible that we're going to need more metadata blocks than
2461          * previously reserved. However we must not fail because we're in
2462          * writeback and there is nothing we can do about it so it might result
2463          * in data loss.  So use reserved blocks to allocate metadata if
2464          * possible.
2465          *
2466          * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2467          * the blocks in question are delalloc blocks.  This indicates
2468          * that the blocks and quotas has already been checked when
2469          * the data was copied into the page cache.
2470          */
2471         get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2472                            EXT4_GET_BLOCKS_METADATA_NOFAIL |
2473                            EXT4_GET_BLOCKS_IO_SUBMIT;
2474         dioread_nolock = ext4_should_dioread_nolock(inode);
2475         if (dioread_nolock)
2476                 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2477         if (map->m_flags & (1 << BH_Delay))
2478                 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2479 
2480         err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2481         if (err < 0)
2482                 return err;
2483         if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2484                 if (!mpd->io_submit.io_end->handle &&
2485                     ext4_handle_valid(handle)) {
2486                         mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2487                         handle->h_rsv_handle = NULL;
2488                 }
2489                 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2490         }
2491 
2492         BUG_ON(map->m_len == 0);
2493         if (map->m_flags & EXT4_MAP_NEW) {
2494                 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2495                                    map->m_len);
2496         }
2497         return 0;
2498 }
2499 
2500 /*
2501  * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2502  *                               mpd->len and submit pages underlying it for IO
2503  *
2504  * @handle - handle for journal operations
2505  * @mpd - extent to map
2506  * @give_up_on_write - we set this to true iff there is a fatal error and there
2507  *                     is no hope of writing the data. The caller should discard
2508  *                     dirty pages to avoid infinite loops.
2509  *
2510  * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2511  * delayed, blocks are allocated, if it is unwritten, we may need to convert
2512  * them to initialized or split the described range from larger unwritten
2513  * extent. Note that we need not map all the described range since allocation
2514  * can return less blocks or the range is covered by more unwritten extents. We
2515  * cannot map more because we are limited by reserved transaction credits. On
2516  * the other hand we always make sure that the last touched page is fully
2517  * mapped so that it can be written out (and thus forward progress is
2518  * guaranteed). After mapping we submit all mapped pages for IO.
2519  */
2520 static int mpage_map_and_submit_extent(handle_t *handle,
2521                                        struct mpage_da_data *mpd,
2522                                        bool *give_up_on_write)
2523 {
2524         struct inode *inode = mpd->inode;
2525         struct ext4_map_blocks *map = &mpd->map;
2526         int err;
2527         loff_t disksize;
2528         int progress = 0;
2529 
2530         mpd->io_submit.io_end->offset =
2531                                 ((loff_t)map->m_lblk) << inode->i_blkbits;
2532         do {
2533                 err = mpage_map_one_extent(handle, mpd);
2534                 if (err < 0) {
2535                         struct super_block *sb = inode->i_sb;
2536 
2537                         if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2538                             EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2539                                 goto invalidate_dirty_pages;
2540                         /*
2541                          * Let the uper layers retry transient errors.
2542                          * In the case of ENOSPC, if ext4_count_free_blocks()
2543                          * is non-zero, a commit should free up blocks.
2544                          */
2545                         if ((err == -ENOMEM) ||
2546                             (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2547                                 if (progress)
2548                                         goto update_disksize;
2549                                 return err;
2550                         }
2551                         ext4_msg(sb, KERN_CRIT,
2552                                  "Delayed block allocation failed for "
2553                                  "inode %lu at logical offset %llu with"
2554                                  " max blocks %u with error %d",
2555                                  inode->i_ino,
2556                                  (unsigned long long)map->m_lblk,
2557                                  (unsigned)map->m_len, -err);
2558                         ext4_msg(sb, KERN_CRIT,
2559                                  "This should not happen!! Data will "
2560                                  "be lost\n");
2561                         if (err == -ENOSPC)
2562                                 ext4_print_free_blocks(inode);
2563                 invalidate_dirty_pages:
2564                         *give_up_on_write = true;
2565                         return err;
2566                 }
2567                 progress = 1;
2568                 /*
2569                  * Update buffer state, submit mapped pages, and get us new
2570                  * extent to map
2571                  */
2572                 err = mpage_map_and_submit_buffers(mpd);
2573                 if (err < 0)
2574                         goto update_disksize;
2575         } while (map->m_len);
2576 
2577 update_disksize:
2578         /*
2579          * Update on-disk size after IO is submitted.  Races with
2580          * truncate are avoided by checking i_size under i_data_sem.
2581          */
2582         disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2583         if (disksize > EXT4_I(inode)->i_disksize) {
2584                 int err2;
2585                 loff_t i_size;
2586 
2587                 down_write(&EXT4_I(inode)->i_data_sem);
2588                 i_size = i_size_read(inode);
2589                 if (disksize > i_size)
2590                         disksize = i_size;
2591                 if (disksize > EXT4_I(inode)->i_disksize)
2592                         EXT4_I(inode)->i_disksize = disksize;
2593                 up_write(&EXT4_I(inode)->i_data_sem);
2594                 err2 = ext4_mark_inode_dirty(handle, inode);
2595                 if (err2)
2596                         ext4_error(inode->i_sb,
2597                                    "Failed to mark inode %lu dirty",
2598                                    inode->i_ino);
2599                 if (!err)
2600                         err = err2;
2601         }
2602         return err;
2603 }
2604 
2605 /*
2606  * Calculate the total number of credits to reserve for one writepages
2607  * iteration. This is called from ext4_writepages(). We map an extent of
2608  * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2609  * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2610  * bpp - 1 blocks in bpp different extents.
2611  */
2612 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2613 {
2614         int bpp = ext4_journal_blocks_per_page(inode);
2615 
2616         return ext4_meta_trans_blocks(inode,
2617                                 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2618 }
2619 
2620 /*
2621  * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2622  *                               and underlying extent to map
2623  *
2624  * @mpd - where to look for pages
2625  *
2626  * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2627  * IO immediately. When we find a page which isn't mapped we start accumulating
2628  * extent of buffers underlying these pages that needs mapping (formed by
2629  * either delayed or unwritten buffers). We also lock the pages containing
2630  * these buffers. The extent found is returned in @mpd structure (starting at
2631  * mpd->lblk with length mpd->len blocks).
2632  *
2633  * Note that this function can attach bios to one io_end structure which are
2634  * neither logically nor physically contiguous. Although it may seem as an
2635  * unnecessary complication, it is actually inevitable in blocksize < pagesize
2636  * case as we need to track IO to all buffers underlying a page in one io_end.
2637  */
2638 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2639 {
2640         struct address_space *mapping = mpd->inode->i_mapping;
2641         struct pagevec pvec;
2642         unsigned int nr_pages;
2643         long left = mpd->wbc->nr_to_write;
2644         pgoff_t index = mpd->first_page;
2645         pgoff_t end = mpd->last_page;
2646         xa_mark_t tag;
2647         int i, err = 0;
2648         int blkbits = mpd->inode->i_blkbits;
2649         ext4_lblk_t lblk;
2650         struct buffer_head *head;
2651 
2652         if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2653                 tag = PAGECACHE_TAG_TOWRITE;
2654         else
2655                 tag = PAGECACHE_TAG_DIRTY;
2656 
2657         pagevec_init(&pvec);
2658         mpd->map.m_len = 0;
2659         mpd->next_page = index;
2660         while (index <= end) {
2661                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2662                                 tag);
2663                 if (nr_pages == 0)
2664                         goto out;
2665 
2666                 for (i = 0; i < nr_pages; i++) {
2667                         struct page *page = pvec.pages[i];
2668 
2669                         /*
2670                          * Accumulated enough dirty pages? This doesn't apply
2671                          * to WB_SYNC_ALL mode. For integrity sync we have to
2672                          * keep going because someone may be concurrently
2673                          * dirtying pages, and we might have synced a lot of
2674                          * newly appeared dirty pages, but have not synced all
2675                          * of the old dirty pages.
2676                          */
2677                         if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2678                                 goto out;
2679 
2680                         /* If we can't merge this page, we are done. */
2681                         if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2682                                 goto out;
2683 
2684                         lock_page(page);
2685                         /*
2686                          * If the page is no longer dirty, or its mapping no
2687                          * longer corresponds to inode we are writing (which
2688                          * means it has been truncated or invalidated), or the
2689                          * page is already under writeback and we are not doing
2690                          * a data integrity writeback, skip the page
2691                          */
2692                         if (!PageDirty(page) ||
2693                             (PageWriteback(page) &&
2694                              (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2695                             unlikely(page->mapping != mapping)) {
2696                                 unlock_page(page);
2697                                 continue;
2698                         }
2699 
2700                         wait_on_page_writeback(page);
2701                         BUG_ON(PageWriteback(page));
2702 
2703                         if (mpd->map.m_len == 0)
2704                                 mpd->first_page = page->index;
2705                         mpd->next_page = page->index + 1;
2706                         /* Add all dirty buffers to mpd */
2707                         lblk = ((ext4_lblk_t)page->index) <<
2708                                 (PAGE_SHIFT - blkbits);
2709                         head = page_buffers(page);
2710                         err = mpage_process_page_bufs(mpd, head, head, lblk);
2711                         if (err <= 0)
2712                                 goto out;
2713                         err = 0;
2714                         left--;
2715                 }
2716                 pagevec_release(&pvec);
2717                 cond_resched();
2718         }
2719         return 0;
2720 out:
2721         pagevec_release(&pvec);
2722         return err;
2723 }
2724 
2725 static int ext4_writepages(struct address_space *mapping,
2726                            struct writeback_control *wbc)
2727 {
2728         pgoff_t writeback_index = 0;
2729         long nr_to_write = wbc->nr_to_write;
2730         int range_whole = 0;
2731         int cycled = 1;
2732         handle_t *handle = NULL;
2733         struct mpage_da_data mpd;
2734         struct inode *inode = mapping->host;
2735         int needed_blocks, rsv_blocks = 0, ret = 0;
2736         struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2737         bool done;
2738         struct blk_plug plug;
2739         bool give_up_on_write = false;
2740 
2741         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2742                 return -EIO;
2743 
2744         percpu_down_read(&sbi->s_journal_flag_rwsem);
2745         trace_ext4_writepages(inode, wbc);
2746 
2747         /*
2748          * No pages to write? This is mainly a kludge to avoid starting
2749          * a transaction for special inodes like journal inode on last iput()
2750          * because that could violate lock ordering on umount
2751          */
2752         if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2753                 goto out_writepages;
2754 
2755         if (ext4_should_journal_data(inode)) {
2756                 ret = generic_writepages(mapping, wbc);
2757                 goto out_writepages;
2758         }
2759 
2760         /*
2761          * If the filesystem has aborted, it is read-only, so return
2762          * right away instead of dumping stack traces later on that
2763          * will obscure the real source of the problem.  We test
2764          * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2765          * the latter could be true if the filesystem is mounted
2766          * read-only, and in that case, ext4_writepages should
2767          * *never* be called, so if that ever happens, we would want
2768          * the stack trace.
2769          */
2770         if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2771                      sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2772                 ret = -EROFS;
2773                 goto out_writepages;
2774         }
2775 
2776         if (ext4_should_dioread_nolock(inode)) {
2777                 /*
2778                  * We may need to convert up to one extent per block in
2779                  * the page and we may dirty the inode.
2780                  */
2781                 rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2782         }
2783 
2784         /*
2785          * If we have inline data and arrive here, it means that
2786          * we will soon create the block for the 1st page, so
2787          * we'd better clear the inline data here.
2788          */
2789         if (ext4_has_inline_data(inode)) {
2790                 /* Just inode will be modified... */
2791                 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2792                 if (IS_ERR(handle)) {
2793                         ret = PTR_ERR(handle);
2794                         goto out_writepages;
2795                 }
2796                 BUG_ON(ext4_test_inode_state(inode,
2797                                 EXT4_STATE_MAY_INLINE_DATA));
2798                 ext4_destroy_inline_data(handle, inode);
2799                 ext4_journal_stop(handle);
2800         }
2801 
2802         if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2803                 range_whole = 1;
2804 
2805         if (wbc->range_cyclic) {
2806                 writeback_index = mapping->writeback_index;
2807                 if (writeback_index)
2808                         cycled = 0;
2809                 mpd.first_page = writeback_index;
2810                 mpd.last_page = -1;
2811         } else {
2812                 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2813                 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2814         }
2815 
2816         mpd.inode = inode;
2817         mpd.wbc = wbc;
2818         ext4_io_submit_init(&mpd.io_submit, wbc);
2819 retry:
2820         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2821                 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2822         done = false;
2823         blk_start_plug(&plug);
2824 
2825         /*
2826          * First writeback pages that don't need mapping - we can avoid
2827          * starting a transaction unnecessarily and also avoid being blocked
2828          * in the block layer on device congestion while having transaction
2829          * started.
2830          */
2831         mpd.do_map = 0;
2832         mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2833         if (!mpd.io_submit.io_end) {
2834                 ret = -ENOMEM;
2835                 goto unplug;
2836         }
2837         ret = mpage_prepare_extent_to_map(&mpd);
2838         /* Submit prepared bio */
2839         ext4_io_submit(&mpd.io_submit);
2840         ext4_put_io_end_defer(mpd.io_submit.io_end);
2841         mpd.io_submit.io_end = NULL;
2842         /* Unlock pages we didn't use */
2843         mpage_release_unused_pages(&mpd, false);
2844         if (ret < 0)
2845                 goto unplug;
2846 
2847         while (!done && mpd.first_page <= mpd.last_page) {
2848                 /* For each extent of pages we use new io_end */
2849                 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2850                 if (!mpd.io_submit.io_end) {
2851                         ret = -ENOMEM;
2852                         break;
2853                 }
2854 
2855                 /*
2856                  * We have two constraints: We find one extent to map and we
2857                  * must always write out whole page (makes a difference when
2858                  * blocksize < pagesize) so that we don't block on IO when we
2859                  * try to write out the rest of the page. Journalled mode is
2860                  * not supported by delalloc.
2861                  */
2862                 BUG_ON(ext4_should_journal_data(inode));
2863                 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2864 
2865                 /* start a new transaction */
2866                 handle = ext4_journal_start_with_reserve(inode,
2867                                 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2868                 if (IS_ERR(handle)) {
2869                         ret = PTR_ERR(handle);
2870                         ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2871                                "%ld pages, ino %lu; err %d", __func__,
2872                                 wbc->nr_to_write, inode->i_ino, ret);
2873                         /* Release allocated io_end */
2874                         ext4_put_io_end(mpd.io_submit.io_end);
2875                         mpd.io_submit.io_end = NULL;
2876                         break;
2877                 }
2878                 mpd.do_map = 1;
2879 
2880                 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2881                 ret = mpage_prepare_extent_to_map(&mpd);
2882                 if (!ret) {
2883                         if (mpd.map.m_len)
2884                                 ret = mpage_map_and_submit_extent(handle, &mpd,
2885                                         &give_up_on_write);
2886                         else {
2887                                 /*
2888                                  * We scanned the whole range (or exhausted
2889                                  * nr_to_write), submitted what was mapped and
2890                                  * didn't find anything needing mapping. We are
2891                                  * done.
2892                                  */
2893                                 done = true;
2894                         }
2895                 }
2896                 /*
2897                  * Caution: If the handle is synchronous,
2898                  * ext4_journal_stop() can wait for transaction commit
2899                  * to finish which may depend on writeback of pages to
2900                  * complete or on page lock to be released.  In that
2901                  * case, we have to wait until after after we have
2902                  * submitted all the IO, released page locks we hold,
2903                  * and dropped io_end reference (for extent conversion
2904                  * to be able to complete) before stopping the handle.
2905                  */
2906                 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2907                         ext4_journal_stop(handle);
2908                         handle = NULL;
2909                         mpd.do_map = 0;
2910                 }
2911                 /* Submit prepared bio */
2912                 ext4_io_submit(&mpd.io_submit);
2913                 /* Unlock pages we didn't use */
2914                 mpage_release_unused_pages(&mpd, give_up_on_write);
2915                 /*
2916                  * Drop our io_end reference we got from init. We have
2917                  * to be careful and use deferred io_end finishing if
2918                  * we are still holding the transaction as we can
2919                  * release the last reference to io_end which may end
2920                  * up doing unwritten extent conversion.
2921                  */
2922                 if (handle) {
2923                         ext4_put_io_end_defer(mpd.io_submit.io_end);
2924                         ext4_journal_stop(handle);
2925                 } else
2926                         ext4_put_io_end(mpd.io_submit.io_end);
2927                 mpd.io_submit.io_end = NULL;
2928 
2929                 if (ret == -ENOSPC && sbi->s_journal) {
2930                         /*
2931                          * Commit the transaction which would
2932                          * free blocks released in the transaction
2933                          * and try again
2934                          */
2935                         jbd2_journal_force_commit_nested(sbi->s_journal);
2936                         ret = 0;
2937                         continue;
2938                 }
2939                 /* Fatal error - ENOMEM, EIO... */
2940                 if (ret)
2941                         break;
2942         }
2943 unplug:
2944         blk_finish_plug(&plug);
2945         if (!ret && !cycled && wbc->nr_to_write > 0) {
2946                 cycled = 1;
2947                 mpd.last_page = writeback_index - 1;
2948                 mpd.first_page = 0;
2949                 goto retry;
2950         }
2951 
2952         /* Update index */
2953         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2954                 /*
2955                  * Set the writeback_index so that range_cyclic
2956                  * mode will write it back later
2957                  */
2958                 mapping->writeback_index = mpd.first_page;
2959 
2960 out_writepages:
2961         trace_ext4_writepages_result(inode, wbc, ret,
2962                                      nr_to_write - wbc->nr_to_write);
2963         percpu_up_read(&sbi->s_journal_flag_rwsem);
2964         return ret;
2965 }
2966 
2967 static int ext4_dax_writepages(struct address_space *mapping,
2968                                struct writeback_control *wbc)
2969 {
2970         int ret;
2971         long nr_to_write = wbc->nr_to_write;
2972         struct inode *inode = mapping->host;
2973         struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2974 
2975         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2976                 return -EIO;
2977 
2978         percpu_down_read(&sbi->s_journal_flag_rwsem);
2979         trace_ext4_writepages(inode, wbc);
2980 
2981         ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
2982         trace_ext4_writepages_result(inode, wbc, ret,
2983                                      nr_to_write - wbc->nr_to_write);
2984         percpu_up_read(&sbi->s_journal_flag_rwsem);
2985         return ret;
2986 }
2987 
2988 static int ext4_nonda_switch(struct super_block *sb)
2989 {
2990         s64 free_clusters, dirty_clusters;
2991         struct ext4_sb_info *sbi = EXT4_SB(sb);
2992 
2993         /*
2994          * switch to non delalloc mode if we are running low
2995          * on free block. The free block accounting via percpu
2996          * counters can get slightly wrong with percpu_counter_batch getting
2997          * accumulated on each CPU without updating global counters
2998          * Delalloc need an accurate free block accounting. So switch
2999          * to non delalloc when we are near to error range.
3000          */
3001         free_clusters =
3002                 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3003         dirty_clusters =
3004                 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3005         /*
3006          * Start pushing delalloc when 1/2 of free blocks are dirty.
3007          */
3008         if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3009                 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3010 
3011         if (2 * free_clusters < 3 * dirty_clusters ||
3012             free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3013                 /*
3014                  * free block count is less than 150% of dirty blocks
3015                  * or free blocks is less than watermark
3016                  */
3017                 return 1;
3018         }
3019         return 0;
3020 }
3021 
3022 /* We always reserve for an inode update; the superblock could be there too */
3023 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3024 {
3025         if (likely(ext4_has_feature_large_file(inode->i_sb)))
3026                 return 1;
3027 
3028         if (pos + len <= 0x7fffffffULL)
3029                 return 1;
3030 
3031         /* We might need to update the superblock to set LARGE_FILE */
3032         return 2;
3033 }
3034 
3035 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3036                                loff_t pos, unsigned len, unsigned flags,
3037                                struct page **pagep, void **fsdata)
3038 {
3039         int ret, retries = 0;
3040         struct page *page;
3041         pgoff_t index;
3042         struct inode *inode = mapping->host;
3043         handle_t *handle;
3044 
3045         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3046                 return -EIO;
3047 
3048         index = pos >> PAGE_SHIFT;
3049 
3050         if (ext4_nonda_switch(inode->i_sb) ||
3051             S_ISLNK(inode->i_mode)) {
3052                 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3053                 return ext4_write_begin(file, mapping, pos,
3054                                         len, flags, pagep, fsdata);
3055         }
3056         *fsdata = (void *)0;
3057         trace_ext4_da_write_begin(inode, pos, len, flags);
3058 
3059         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3060                 ret = ext4_da_write_inline_data_begin(mapping, inode,
3061                                                       pos, len, flags,
3062                                                       pagep, fsdata);
3063                 if (ret < 0)
3064                         return ret;
3065                 if (ret == 1)
3066                         return 0;
3067         }
3068 
3069         /*
3070          * grab_cache_page_write_begin() can take a long time if the
3071          * system is thrashing due to memory pressure, or if the page
3072          * is being written back.  So grab it first before we start
3073          * the transaction handle.  This also allows us to allocate
3074          * the page (if needed) without using GFP_NOFS.
3075          */
3076 retry_grab:
3077         page = grab_cache_page_write_begin(mapping, index, flags);
3078         if (!page)
3079                 return -ENOMEM;
3080         unlock_page(page);
3081 
3082         /*
3083          * With delayed allocation, we don't log the i_disksize update
3084          * if there is delayed block allocation. But we still need
3085          * to journalling the i_disksize update if writes to the end
3086          * of file which has an already mapped buffer.
3087          */
3088 retry_journal:
3089         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3090                                 ext4_da_write_credits(inode, pos, len));
3091         if (IS_ERR(handle)) {
3092                 put_page(page);
3093                 return PTR_ERR(handle);
3094         }
3095 
3096         lock_page(page);
3097         if (page->mapping != mapping) {
3098                 /* The page got truncated from under us */
3099                 unlock_page(page);
3100                 put_page(page);
3101                 ext4_journal_stop(handle);
3102                 goto retry_grab;
3103         }
3104         /* In case writeback began while the page was unlocked */
3105         wait_for_stable_page(page);
3106 
3107 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3108         ret = ext4_block_write_begin(page, pos, len,
3109                                      ext4_da_get_block_prep);
3110 #else
3111         ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3112 #endif
3113         if (ret < 0) {
3114                 unlock_page(page);
3115                 ext4_journal_stop(handle);
3116                 /*
3117                  * block_write_begin may have instantiated a few blocks
3118                  * outside i_size.  Trim these off again. Don't need
3119                  * i_size_read because we hold i_mutex.
3120                  */
3121                 if (pos + len > inode->i_size)
3122                         ext4_truncate_failed_write(inode);
3123 
3124                 if (ret == -ENOSPC &&
3125                     ext4_should_retry_alloc(inode->i_sb, &retries))
3126                         goto retry_journal;
3127 
3128                 put_page(page);
3129                 return ret;
3130         }
3131 
3132         *pagep = page;
3133         return ret;
3134 }
3135 
3136 /*
3137  * Check if we should update i_disksize
3138  * when write to the end of file but not require block allocation
3139  */
3140 static int ext4_da_should_update_i_disksize(struct page *page,
3141                                             unsigned long offset)
3142 {
3143         struct buffer_head *bh;
3144         struct inode *inode = page->mapping->host;
3145         unsigned int idx;
3146         int i;
3147 
3148         bh = page_buffers(page);
3149         idx = offset >> inode->i_blkbits;
3150 
3151         for (i = 0; i < idx; i++)
3152                 bh = bh->b_this_page;
3153 
3154         if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3155                 return 0;
3156         return 1;
3157 }
3158 
3159 static int ext4_da_write_end(struct file *file,
3160                              struct address_space *mapping,
3161                              loff_t pos, unsigned len, unsigned copied,
3162                              struct page *page, void *fsdata)
3163 {
3164         struct inode *inode = mapping->host;
3165         int ret = 0, ret2;
3166         handle_t *handle = ext4_journal_current_handle();
3167         loff_t new_i_size;
3168         unsigned long start, end;
3169         int write_mode = (int)(unsigned long)fsdata;
3170 
3171         if (write_mode == FALL_BACK_TO_NONDELALLOC)
3172                 return ext4_write_end(file, mapping, pos,
3173                                       len, copied, page, fsdata);
3174 
3175         trace_ext4_da_write_end(inode, pos, len, copied);
3176         start = pos & (PAGE_SIZE - 1);
3177         end = start + copied - 1;
3178 
3179         /*
3180          * generic_write_end() will run mark_inode_dirty() if i_size
3181          * changes.  So let's piggyback the i_disksize mark_inode_dirty
3182          * into that.
3183          */
3184         new_i_size = pos + copied;
3185         if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3186                 if (ext4_has_inline_data(inode) ||
3187                     ext4_da_should_update_i_disksize(page, end)) {
3188                         ext4_update_i_disksize(inode, new_i_size);
3189                         /* We need to mark inode dirty even if
3190                          * new_i_size is less that inode->i_size
3191                          * bu greater than i_disksize.(hint delalloc)
3192                          */
3193                         ext4_mark_inode_dirty(handle, inode);
3194                 }
3195         }
3196 
3197         if (write_mode != CONVERT_INLINE_DATA &&
3198             ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3199             ext4_has_inline_data(inode))
3200                 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3201                                                      page);
3202         else
3203                 ret2 = generic_write_end(file, mapping, pos, len, copied,
3204                                                         page, fsdata);
3205 
3206         copied = ret2;
3207         if (ret2 < 0)
3208                 ret = ret2;
3209         ret2 = ext4_journal_stop(handle);
3210         if (!ret)
3211                 ret = ret2;
3212 
3213         return ret ? ret : copied;
3214 }
3215 
3216 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3217                                    unsigned int length)
3218 {
3219         /*
3220          * Drop reserved blocks
3221          */
3222         BUG_ON(!PageLocked(page));
3223         if (!page_has_buffers(page))
3224                 goto out;
3225 
3226         ext4_da_page_release_reservation(page, offset, length);
3227 
3228 out:
3229         ext4_invalidatepage(page, offset, length);
3230 
3231         return;
3232 }
3233 
3234 /*
3235  * Force all delayed allocation blocks to be allocated for a given inode.
3236  */
3237 int ext4_alloc_da_blocks(struct inode *inode)
3238 {
3239         trace_ext4_alloc_da_blocks(inode);
3240 
3241         if (!EXT4_I(inode)->i_reserved_data_blocks)
3242                 return 0;
3243 
3244         /*
3245          * We do something simple for now.  The filemap_flush() will
3246          * also start triggering a write of the data blocks, which is
3247          * not strictly speaking necessary (and for users of
3248          * laptop_mode, not even desirable).  However, to do otherwise
3249          * would require replicating code paths in:
3250          *
3251          * ext4_writepages() ->
3252          *    write_cache_pages() ---> (via passed in callback function)
3253          *        __mpage_da_writepage() -->
3254          *           mpage_add_bh_to_extent()
3255          *           mpage_da_map_blocks()
3256          *
3257          * The problem is that write_cache_pages(), located in
3258          * mm/page-writeback.c, marks pages clean in preparation for
3259          * doing I/O, which is not desirable if we're not planning on
3260          * doing I/O at all.
3261          *
3262          * We could call write_cache_pages(), and then redirty all of
3263          * the pages by calling redirty_page_for_writepage() but that
3264          * would be ugly in the extreme.  So instead we would need to
3265          * replicate parts of the code in the above functions,
3266          * simplifying them because we wouldn't actually intend to
3267          * write out the pages, but rather only collect contiguous
3268          * logical block extents, call the multi-block allocator, and
3269          * then update the buffer heads with the block allocations.
3270          *
3271          * For now, though, we'll cheat by calling filemap_flush(),
3272          * which will map the blocks, and start the I/O, but not
3273          * actually wait for the I/O to complete.
3274          */
3275         return filemap_flush(inode->i_mapping);
3276 }
3277 
3278 /*
3279  * bmap() is special.  It gets used by applications such as lilo and by
3280  * the swapper to find the on-disk block of a specific piece of data.
3281  *
3282  * Naturally, this is dangerous if the block concerned is still in the
3283  * journal.  If somebody makes a swapfile on an ext4 data-journaling
3284  * filesystem and enables swap, then they may get a nasty shock when the
3285  * data getting swapped to that swapfile suddenly gets overwritten by
3286  * the original zero's written out previously to the journal and
3287  * awaiting writeback in the kernel's buffer cache.
3288  *
3289  * So, if we see any bmap calls here on a modified, data-journaled file,
3290  * take extra steps to flush any blocks which might be in the cache.
3291  */
3292 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3293 {
3294         struct inode *inode = mapping->host;
3295         journal_t *journal;
3296         int err;
3297 
3298         /*
3299          * We can get here for an inline file via the FIBMAP ioctl
3300          */
3301         if (ext4_has_inline_data(inode))
3302                 return 0;
3303 
3304         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3305                         test_opt(inode->i_sb, DELALLOC)) {
3306                 /*
3307                  * With delalloc we want to sync the file
3308                  * so that we can make sure we allocate
3309                  * blocks for file
3310                  */
3311                 filemap_write_and_wait(mapping);
3312         }
3313 
3314         if (EXT4_JOURNAL(inode) &&
3315             ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3316                 /*
3317                  * This is a REALLY heavyweight approach, but the use of
3318                  * bmap on dirty files is expected to be extremely rare:
3319                  * only if we run lilo or swapon on a freshly made file
3320                  * do we expect this to happen.
3321                  *
3322                  * (bmap requires CAP_SYS_RAWIO so this does not
3323                  * represent an unprivileged user DOS attack --- we'd be
3324                  * in trouble if mortal users could trigger this path at
3325                  * will.)
3326                  *
3327                  * NB. EXT4_STATE_JDATA is not set on files other than
3328                  * regular files.  If somebody wants to bmap a directory
3329                  * or symlink and gets confused because the buffer
3330                  * hasn't yet been flushed to disk, they deserve
3331                  * everything they get.
3332                  */
3333 
3334                 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3335                 journal = EXT4_JOURNAL(inode);
3336                 jbd2_journal_lock_updates(journal);
3337                 err = jbd2_journal_flush(journal);
3338                 jbd2_journal_unlock_updates(journal);
3339 
3340                 if (err)
3341                         return 0;
3342         }
3343 
3344         return generic_block_bmap(mapping, block, ext4_get_block);
3345 }
3346 
3347 static int ext4_readpage(struct file *file, struct page *page)
3348 {
3349         int ret = -EAGAIN;
3350         struct inode *inode = page->mapping->host;
3351 
3352         trace_ext4_readpage(page);
3353 
3354         if (ext4_has_inline_data(inode))
3355                 ret = ext4_readpage_inline(inode, page);
3356 
3357         if (ret == -EAGAIN)
3358                 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3359                                                 false);
3360 
3361         return ret;
3362 }
3363 
3364 static int
3365 ext4_readpages(struct file *file, struct address_space *mapping,
3366                 struct list_head *pages, unsigned nr_pages)
3367 {
3368         struct inode *inode = mapping->host;
3369 
3370         /* If the file has inline data, no need to do readpages. */
3371         if (ext4_has_inline_data(inode))
3372                 return 0;
3373 
3374         return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3375 }
3376 
3377 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3378                                 unsigned int length)
3379 {
3380         trace_ext4_invalidatepage(page, offset, length);
3381 
3382         /* No journalling happens on data buffers when this function is used */
3383         WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3384 
3385         block_invalidatepage(page, offset, length);
3386 }
3387 
3388 static int __ext4_journalled_invalidatepage(struct page *page,
3389                                             unsigned int offset,
3390                                             unsigned int length)
3391 {
3392         journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3393 
3394         trace_ext4_journalled_invalidatepage(page, offset, length);
3395 
3396         /*
3397          * If it's a full truncate we just forget about the pending dirtying
3398          */
3399         if (offset == 0 && length == PAGE_SIZE)
3400                 ClearPageChecked(page);
3401 
3402         return jbd2_journal_invalidatepage(journal, page, offset, length);
3403 }
3404 
3405 /* Wrapper for aops... */
3406 static void ext4_journalled_invalidatepage(struct page *page,
3407                                            unsigned int offset,
3408                                            unsigned int length)
3409 {
3410         WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3411 }
3412 
3413 static int ext4_releasepage(struct page *page, gfp_t wait)
3414 {
3415         journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3416 
3417         trace_ext4_releasepage(page);
3418 
3419         /* Page has dirty journalled data -> cannot release */
3420         if (PageChecked(page))
3421                 return 0;
3422         if (journal)
3423                 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3424         else
3425                 return try_to_free_buffers(page);
3426 }
3427 
3428 static bool ext4_inode_datasync_dirty(struct inode *inode)
3429 {
3430         journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3431 
3432         if (journal)
3433                 return !jbd2_transaction_committed(journal,
3434                                         EXT4_I(inode)->i_datasync_tid);
3435         /* Any metadata buffers to write? */
3436         if (!list_empty(&inode->i_mapping->private_list))
3437                 return true;
3438         return inode->i_state & I_DIRTY_DATASYNC;
3439 }
3440 
3441 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3442                             unsigned flags, struct iomap *iomap)
3443 {
3444         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3445         unsigned int blkbits = inode->i_blkbits;
3446         unsigned long first_block, last_block;
3447         struct ext4_map_blocks map;
3448         bool delalloc = false;
3449         int ret;
3450 
3451         if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3452                 return -EINVAL;
3453         first_block = offset >> blkbits;
3454         last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3455                            EXT4_MAX_LOGICAL_BLOCK);
3456 
3457         if (flags & IOMAP_REPORT) {
3458                 if (ext4_has_inline_data(inode)) {
3459                         ret = ext4_inline_data_iomap(inode, iomap);
3460                         if (ret != -EAGAIN) {
3461                                 if (ret == 0 && offset >= iomap->length)
3462                                         ret = -ENOENT;
3463                                 return ret;
3464                         }
3465                 }
3466         } else {
3467                 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3468                         return -ERANGE;
3469         }
3470 
3471         map.m_lblk = first_block;
3472         map.m_len = last_block - first_block + 1;
3473 
3474         if (flags & IOMAP_REPORT) {
3475                 ret = ext4_map_blocks(NULL, inode, &map, 0);
3476                 if (ret < 0)
3477                         return ret;
3478 
3479                 if (ret == 0) {
3480                         ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3481                         struct extent_status es;
3482 
3483                         ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3484                                                   map.m_lblk, end, &es);
3485 
3486                         if (!es.es_len || es.es_lblk > end) {
3487                                 /* entire range is a hole */
3488                         } else if (es.es_lblk > map.m_lblk) {
3489                                 /* range starts with a hole */
3490                                 map.m_len = es.es_lblk - map.m_lblk;
3491                         } else {
3492                                 ext4_lblk_t offs = 0;
3493 
3494                                 if (es.es_lblk < map.m_lblk)
3495                                         offs = map.m_lblk - es.es_lblk;
3496                                 map.m_lblk = es.es_lblk + offs;
3497                                 map.m_len = es.es_len - offs;
3498                                 delalloc = true;
3499                         }
3500                 }
3501         } else if (flags & IOMAP_WRITE) {
3502                 int dio_credits;
3503                 handle_t *handle;
3504                 int retries = 0;
3505 
3506                 /* Trim mapping request to maximum we can map at once for DIO */
3507                 if (map.m_len > DIO_MAX_BLOCKS)
3508                         map.m_len = DIO_MAX_BLOCKS;
3509                 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3510 retry:
3511                 /*
3512                  * Either we allocate blocks and then we don't get unwritten
3513                  * extent so we have reserved enough credits, or the blocks
3514                  * are already allocated and unwritten and in that case
3515                  * extent conversion fits in the credits as well.
3516                  */
3517                 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3518                                             dio_credits);
3519                 if (IS_ERR(handle))
3520                         return PTR_ERR(handle);
3521 
3522                 ret = ext4_map_blocks(handle, inode, &map,
3523                                       EXT4_GET_BLOCKS_CREATE_ZERO);
3524                 if (ret < 0) {
3525                         ext4_journal_stop(handle);
3526                         if (ret == -ENOSPC &&
3527                             ext4_should_retry_alloc(inode->i_sb, &retries))
3528                                 goto retry;
3529                         return ret;
3530                 }
3531 
3532                 /*
3533                  * If we added blocks beyond i_size, we need to make sure they
3534                  * will get truncated if we crash before updating i_size in
3535                  * ext4_iomap_end(). For faults we don't need to do that (and
3536                  * even cannot because for orphan list operations inode_lock is
3537                  * required) - if we happen to instantiate block beyond i_size,
3538                  * it is because we race with truncate which has already added
3539                  * the inode to the orphan list.
3540                  */
3541                 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3542                     (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3543                         int err;
3544 
3545                         err = ext4_orphan_add(handle, inode);
3546                         if (err < 0) {
3547                                 ext4_journal_stop(handle);
3548                                 return err;
3549                         }
3550                 }
3551                 ext4_journal_stop(handle);
3552         } else {
3553                 ret = ext4_map_blocks(NULL, inode, &map, 0);
3554                 if (ret < 0)
3555                         return ret;
3556         }
3557 
3558         iomap->flags = 0;
3559         if (ext4_inode_datasync_dirty(inode))
3560                 iomap->flags |= IOMAP_F_DIRTY;
3561         iomap->bdev = inode->i_sb->s_bdev;
3562         iomap->dax_dev = sbi->s_daxdev;
3563         iomap->offset = (u64)first_block << blkbits;
3564         iomap->length = (u64)map.m_len << blkbits;
3565 
3566         if (ret == 0) {
3567                 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3568                 iomap->addr = IOMAP_NULL_ADDR;
3569         } else {
3570                 if (map.m_flags & EXT4_MAP_MAPPED) {
3571                         iomap->type = IOMAP_MAPPED;
3572                 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3573                         iomap->type = IOMAP_UNWRITTEN;
3574                 } else {
3575                         WARN_ON_ONCE(1);
3576                         return -EIO;
3577                 }
3578                 iomap->addr = (u64)map.m_pblk << blkbits;
3579         }
3580 
3581         if (map.m_flags & EXT4_MAP_NEW)
3582                 iomap->flags |= IOMAP_F_NEW;
3583 
3584         return 0;
3585 }
3586 
3587 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3588                           ssize_t written, unsigned flags, struct iomap *iomap)
3589 {
3590         int ret = 0;
3591         handle_t *handle;
3592         int blkbits = inode->i_blkbits;
3593         bool truncate = false;
3594 
3595         if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3596                 return 0;
3597 
3598         handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3599         if (IS_ERR(handle)) {
3600                 ret = PTR_ERR(handle);
3601                 goto orphan_del;
3602         }
3603         if (ext4_update_inode_size(inode, offset + written))
3604                 ext4_mark_inode_dirty(handle, inode);
3605         /*
3606          * We may need to truncate allocated but not written blocks beyond EOF.
3607          */
3608         if (iomap->offset + iomap->length > 
3609             ALIGN(inode->i_size, 1 << blkbits)) {
3610                 ext4_lblk_t written_blk, end_blk;
3611 
3612                 written_blk = (offset + written) >> blkbits;
3613                 end_blk = (offset + length) >> blkbits;
3614                 if (written_blk < end_blk && ext4_can_truncate(inode))
3615                         truncate = true;
3616         }
3617         /*
3618          * Remove inode from orphan list if we were extending a inode and
3619          * everything went fine.
3620          */
3621         if (!truncate && inode->i_nlink &&
3622             !list_empty(&EXT4_I(inode)->i_orphan))
3623                 ext4_orphan_del(handle, inode);
3624         ext4_journal_stop(handle);
3625         if (truncate) {
3626                 ext4_truncate_failed_write(inode);
3627 orphan_del:
3628                 /*
3629                  * If truncate failed early the inode might still be on the
3630                  * orphan list; we need to make sure the inode is removed from
3631                  * the orphan list in that case.
3632                  */
3633                 if (inode->i_nlink)
3634                         ext4_orphan_del(NULL, inode);
3635         }
3636         return ret;
3637 }
3638 
3639 const struct iomap_ops ext4_iomap_ops = {
3640         .iomap_begin            = ext4_iomap_begin,
3641         .iomap_end              = ext4_iomap_end,
3642 };
3643 
3644 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3645                             ssize_t size, void *private)
3646 {
3647         ext4_io_end_t *io_end = private;
3648 
3649         /* if not async direct IO just return */
3650         if (!io_end)
3651                 return 0;
3652 
3653         ext_debug("ext4_end_io_dio(): io_end 0x%p "
3654                   "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3655                   io_end, io_end->inode->i_ino, iocb, offset, size);
3656 
3657         /*
3658          * Error during AIO DIO. We cannot convert unwritten extents as the
3659          * data was not written. Just clear the unwritten flag and drop io_end.
3660          */
3661         if (size <= 0) {
3662                 ext4_clear_io_unwritten_flag(io_end);
3663                 size = 0;
3664         }
3665         io_end->offset = offset;
3666         io_end->size = size;
3667         ext4_put_io_end(io_end);
3668 
3669         return 0;
3670 }
3671 
3672 /*
3673  * Handling of direct IO writes.
3674  *
3675  * For ext4 extent files, ext4 will do direct-io write even to holes,
3676  * preallocated extents, and those write extend the file, no need to
3677  * fall back to buffered IO.
3678  *
3679  * For holes, we fallocate those blocks, mark them as unwritten
3680  * If those blocks were preallocated, we mark sure they are split, but
3681  * still keep the range to write as unwritten.
3682  *
3683  * The unwritten extents will be converted to written when DIO is completed.
3684  * For async direct IO, since the IO may still pending when return, we
3685  * set up an end_io call back function, which will do the conversion
3686  * when async direct IO completed.
3687  *
3688  * If the O_DIRECT write will extend the file then add this inode to the
3689  * orphan list.  So recovery will truncate it back to the original size
3690  * if the machine crashes during the write.
3691  *
3692  */
3693 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3694 {
3695         struct file *file = iocb->ki_filp;
3696         struct inode *inode = file->f_mapping->host;
3697         struct ext4_inode_info *ei = EXT4_I(inode);
3698         ssize_t ret;
3699         loff_t offset = iocb->ki_pos;
3700         size_t count = iov_iter_count(iter);
3701         int overwrite = 0;
3702         get_block_t *get_block_func = NULL;
3703         int dio_flags = 0;
3704         loff_t final_size = offset + count;
3705         int orphan = 0;
3706         handle_t *handle;
3707 
3708         if (final_size > inode->i_size || final_size > ei->i_disksize) {
3709                 /* Credits for sb + inode write */
3710                 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3711                 if (IS_ERR(handle)) {
3712                         ret = PTR_ERR(handle);
3713                         goto out;
3714                 }
3715                 ret = ext4_orphan_add(handle, inode);
3716                 if (ret) {
3717                         ext4_journal_stop(handle);
3718                         goto out;
3719                 }
3720                 orphan = 1;
3721                 ext4_update_i_disksize(inode, inode->i_size);
3722                 ext4_journal_stop(handle);
3723         }
3724 
3725         BUG_ON(iocb->private == NULL);
3726 
3727         /*
3728          * Make all waiters for direct IO properly wait also for extent
3729          * conversion. This also disallows race between truncate() and
3730          * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3731          */
3732         inode_dio_begin(inode);
3733 
3734         /* If we do a overwrite dio, i_mutex locking can be released */
3735         overwrite = *((int *)iocb->private);
3736 
3737         if (overwrite)
3738                 inode_unlock(inode);
3739 
3740         /*
3741          * For extent mapped files we could direct write to holes and fallocate.
3742          *
3743          * Allocated blocks to fill the hole are marked as unwritten to prevent
3744          * parallel buffered read to expose the stale data before DIO complete
3745          * the data IO.
3746          *
3747          * As to previously fallocated extents, ext4 get_block will just simply
3748          * mark the buffer mapped but still keep the extents unwritten.
3749          *
3750          * For non AIO case, we will convert those unwritten extents to written
3751          * after return back from blockdev_direct_IO. That way we save us from
3752          * allocating io_end structure and also the overhead of offloading
3753          * the extent convertion to a workqueue.
3754          *
3755          * For async DIO, the conversion needs to be deferred when the
3756          * IO is completed. The ext4 end_io callback function will be
3757          * called to take care of the conversion work.  Here for async
3758          * case, we allocate an io_end structure to hook to the iocb.
3759          */
3760         iocb->private = NULL;
3761         if (overwrite)
3762                 get_block_func = ext4_dio_get_block_overwrite;
3763         else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3764                    round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3765                 get_block_func = ext4_dio_get_block;
3766                 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3767         } else if (is_sync_kiocb(iocb)) {
3768                 get_block_func = ext4_dio_get_block_unwritten_sync;
3769                 dio_flags = DIO_LOCKING;
3770         } else {
3771                 get_block_func = ext4_dio_get_block_unwritten_async;
3772                 dio_flags = DIO_LOCKING;
3773         }
3774         ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3775                                    get_block_func, ext4_end_io_dio, NULL,
3776                                    dio_flags);
3777 
3778         if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3779                                                 EXT4_STATE_DIO_UNWRITTEN)) {
3780                 int err;
3781                 /*
3782                  * for non AIO case, since the IO is already
3783                  * completed, we could do the conversion right here
3784                  */
3785                 err = ext4_convert_unwritten_extents(NULL, inode,
3786                                                      offset, ret);
3787                 if (err < 0)
3788                         ret = err;
3789                 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3790         }
3791 
3792         inode_dio_end(inode);
3793         /* take i_mutex locking again if we do a ovewrite dio */
3794         if (overwrite)
3795                 inode_lock(inode);
3796 
3797         if (ret < 0 && final_size > inode->i_size)
3798                 ext4_truncate_failed_write(inode);
3799 
3800         /* Handle extending of i_size after direct IO write */
3801         if (orphan) {
3802                 int err;
3803 
3804                 /* Credits for sb + inode write */
3805                 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3806                 if (IS_ERR(handle)) {
3807                         /*
3808                          * We wrote the data but cannot extend
3809                          * i_size. Bail out. In async io case, we do
3810                          * not return error here because we have
3811                          * already submmitted the corresponding
3812                          * bio. Returning error here makes the caller
3813                          * think that this IO is done and failed
3814                          * resulting in race with bio's completion
3815                          * handler.
3816                          */
3817                         if (!ret)
3818                                 ret = PTR_ERR(handle);
3819                         if (inode->i_nlink)
3820                                 ext4_orphan_del(NULL, inode);
3821 
3822                         goto out;
3823                 }
3824                 if (inode->i_nlink)
3825                         ext4_orphan_del(handle, inode);
3826                 if (ret > 0) {
3827                         loff_t end = offset + ret;
3828                         if (end > inode->i_size || end > ei->i_disksize) {
3829                                 ext4_update_i_disksize(inode, end);
3830                                 if (end > inode->i_size)
3831                                         i_size_write(inode, end);
3832                                 /*
3833                                  * We're going to return a positive `ret'
3834                                  * here due to non-zero-length I/O, so there's
3835                                  * no way of reporting error returns from
3836                                  * ext4_mark_inode_dirty() to userspace.  So
3837                                  * ignore it.
3838                                  */
3839                                 ext4_mark_inode_dirty(handle, inode);
3840                         }
3841                 }
3842                 err = ext4_journal_stop(handle);
3843                 if (ret == 0)
3844                         ret = err;
3845         }
3846 out:
3847         return ret;
3848 }
3849 
3850 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3851 {
3852         struct address_space *mapping = iocb->ki_filp->f_mapping;
3853         struct inode *inode = mapping->host;
3854         size_t count = iov_iter_count(iter);
3855         ssize_t ret;
3856 
3857         /*
3858          * Shared inode_lock is enough for us - it protects against concurrent
3859          * writes & truncates and since we take care of writing back page cache,
3860          * we are protected against page writeback as well.
3861          */
3862         inode_lock_shared(inode);
3863         ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3864                                            iocb->ki_pos + count - 1);
3865         if (ret)
3866                 goto out_unlock;
3867         ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3868                                    iter, ext4_dio_get_block, NULL, NULL, 0);
3869 out_unlock:
3870         inode_unlock_shared(inode);
3871         return ret;
3872 }
3873 
3874 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3875 {
3876         struct file *file = iocb->ki_filp;
3877         struct inode *inode = file->f_mapping->host;
3878         size_t count = iov_iter_count(iter);
3879         loff_t offset = iocb->ki_pos;
3880         ssize_t ret;
3881 
3882 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3883         if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3884                 return 0;
3885 #endif
3886 
3887         /*
3888          * If we are doing data journalling we don't support O_DIRECT
3889          */
3890         if (ext4_should_journal_data(inode))
3891                 return 0;
3892 
3893         /* Let buffer I/O handle the inline data case. */
3894         if (ext4_has_inline_data(inode))
3895                 return 0;
3896 
3897         trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3898         if (iov_iter_rw(iter) == READ)
3899                 ret = ext4_direct_IO_read(iocb, iter);
3900         else
3901                 ret = ext4_direct_IO_write(iocb, iter);
3902         trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3903         return ret;
3904 }
3905 
3906 /*
3907  * Pages can be marked dirty completely asynchronously from ext4's journalling
3908  * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
3909  * much here because ->set_page_dirty is called under VFS locks.  The page is
3910  * not necessarily locked.
3911  *
3912  * We cannot just dirty the page and leave attached buffers clean, because the
3913  * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3914  * or jbddirty because all the journalling code will explode.
3915  *
3916  * So what we do is to mark the page "pending dirty" and next time writepage
3917  * is called, propagate that into the buffers appropriately.
3918  */
3919 static int ext4_journalled_set_page_dirty(struct page *page)
3920 {
3921         SetPageChecked(page);
3922         return __set_page_dirty_nobuffers(page);
3923 }
3924 
3925 static int ext4_set_page_dirty(struct page *page)
3926 {
3927         WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3928         WARN_ON_ONCE(!page_has_buffers(page));
3929         return __set_page_dirty_buffers(page);
3930 }
3931 
3932 static const struct address_space_operations ext4_aops = {
3933         .readpage               = ext4_readpage,
3934         .readpages              = ext4_readpages,
3935         .writepage              = ext4_writepage,
3936         .writepages             = ext4_writepages,
3937         .write_begin            = ext4_write_begin,
3938         .write_end              = ext4_write_end,
3939         .set_page_dirty         = ext4_set_page_dirty,
3940         .bmap                   = ext4_bmap,
3941         .invalidatepage         = ext4_invalidatepage,
3942         .releasepage            = ext4_releasepage,
3943         .direct_IO              = ext4_direct_IO,
3944         .migratepage            = buffer_migrate_page,
3945         .is_partially_uptodate  = block_is_partially_uptodate,
3946         .error_remove_page      = generic_error_remove_page,
3947 };
3948 
3949 static const struct address_space_operations ext4_journalled_aops = {
3950         .readpage               = ext4_readpage,
3951         .readpages              = ext4_readpages,
3952         .writepage              = ext4_writepage,
3953         .writepages             = ext4_writepages,
3954         .write_begin            = ext4_write_begin,
3955         .write_end              = ext4_journalled_write_end,
3956         .set_page_dirty         = ext4_journalled_set_page_dirty,
3957         .bmap                   = ext4_bmap,
3958         .invalidatepage         = ext4_journalled_invalidatepage,
3959         .releasepage            = ext4_releasepage,
3960         .direct_IO              = ext4_direct_IO,
3961         .is_partially_uptodate  = block_is_partially_uptodate,
3962         .error_remove_page      = generic_error_remove_page,
3963 };
3964 
3965 static const struct address_space_operations ext4_da_aops = {
3966         .readpage               = ext4_readpage,
3967         .readpages              = ext4_readpages,
3968         .writepage              = ext4_writepage,
3969         .writepages             = ext4_writepages,
3970         .write_begin            = ext4_da_write_begin,
3971         .write_end              = ext4_da_write_end,
3972         .set_page_dirty         = ext4_set_page_dirty,
3973         .bmap                   = ext4_bmap,
3974         .invalidatepage         = ext4_da_invalidatepage,
3975         .releasepage            = ext4_releasepage,
3976         .direct_IO              = ext4_direct_IO,
3977         .migratepage            = buffer_migrate_page,
3978         .is_partially_uptodate  = block_is_partially_uptodate,
3979         .error_remove_page      = generic_error_remove_page,
3980 };
3981 
3982 static const struct address_space_operations ext4_dax_aops = {
3983         .writepages             = ext4_dax_writepages,
3984         .direct_IO              = noop_direct_IO,
3985         .set_page_dirty         = noop_set_page_dirty,
3986         .bmap                   = ext4_bmap,
3987         .invalidatepage         = noop_invalidatepage,
3988 };
3989 
3990 void ext4_set_aops(struct inode *inode)
3991 {
3992         switch (ext4_inode_journal_mode(inode)) {
3993         case EXT4_INODE_ORDERED_DATA_MODE:
3994         case EXT4_INODE_WRITEBACK_DATA_MODE:
3995                 break;
3996         case EXT4_INODE_JOURNAL_DATA_MODE:
3997                 inode->i_mapping->a_ops = &ext4_journalled_aops;
3998                 return;
3999         default:
4000                 BUG();
4001         }
4002         if (IS_DAX(inode))
4003                 inode->i_mapping->a_ops = &ext4_dax_aops;
4004         else if (test_opt(inode->i_sb, DELALLOC))
4005                 inode->i_mapping->a_ops = &ext4_da_aops;
4006         else
4007                 inode->i_mapping->a_ops = &ext4_aops;
4008 }
4009 
4010 static int __ext4_block_zero_page_range(handle_t *handle,
4011                 struct address_space *mapping, loff_t from, loff_t length)
4012 {
4013         ext4_fsblk_t index = from >> PAGE_SHIFT;
4014         unsigned offset = from & (PAGE_SIZE-1);
4015         unsigned blocksize, pos;
4016         ext4_lblk_t iblock;
4017         struct inode *inode = mapping->host;
4018         struct buffer_head *bh;
4019         struct page *page;
4020         int err = 0;
4021 
4022         page = find_or_create_page(mapping, from >> PAGE_SHIFT,
4023                                    mapping_gfp_constraint(mapping, ~__GFP_FS));
4024         if (!page)
4025                 return -ENOMEM;
4026 
4027         blocksize = inode->i_sb->s_blocksize;
4028 
4029         iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4030 
4031         if (!page_has_buffers(page))
4032                 create_empty_buffers(page, blocksize, 0);
4033 
4034         /* Find the buffer that contains "offset" */
4035         bh = page_buffers(page);
4036         pos = blocksize;
4037         while (offset >= pos) {
4038                 bh = bh->b_this_page;
4039                 iblock++;
4040                 pos += blocksize;
4041         }
4042         if (buffer_freed(bh)) {
4043                 BUFFER_TRACE(bh, "freed: skip");
4044                 goto unlock;
4045         }
4046         if (!buffer_mapped(bh)) {
4047                 BUFFER_TRACE(bh, "unmapped");
4048                 ext4_get_block(inode, iblock, bh, 0);
4049                 /* unmapped? It's a hole - nothing to do */
4050                 if (!buffer_mapped(bh)) {
4051                         BUFFER_TRACE(bh, "still unmapped");
4052                         goto unlock;
4053                 }
4054         }
4055 
4056         /* Ok, it's mapped. Make sure it's up-to-date */
4057         if (PageUptodate(page))
4058                 set_buffer_uptodate(bh);
4059 
4060         if (!buffer_uptodate(bh)) {
4061                 err = -EIO;
4062                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4063                 wait_on_buffer(bh);
4064                 /* Uhhuh. Read error. Complain and punt. */
4065                 if (!buffer_uptodate(bh))
4066                         goto unlock;
4067                 if (S_ISREG(inode->i_mode) &&
4068                     ext4_encrypted_inode(inode)) {
4069                         /* We expect the key to be set. */
4070                         BUG_ON(!fscrypt_has_encryption_key(inode));
4071                         BUG_ON(blocksize != PAGE_SIZE);
4072                         WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4073                                                 page, PAGE_SIZE, 0, page->index));
4074                 }
4075         }
4076         if (ext4_should_journal_data(inode)) {
4077                 BUFFER_TRACE(bh, "get write access");
4078                 err = ext4_journal_get_write_access(handle, bh);
4079                 if (err)
4080                         goto unlock;
4081         }
4082         zero_user(page, offset, length);
4083         BUFFER_TRACE(bh, "zeroed end of block");
4084 
4085         if (ext4_should_journal_data(inode)) {
4086                 err = ext4_handle_dirty_metadata(handle, inode, bh);
4087         } else {
4088                 err = 0;
4089                 mark_buffer_dirty(bh);
4090                 if (ext4_should_order_data(inode))
4091                         err = ext4_jbd2_inode_add_write(handle, inode);
4092         }
4093 
4094 unlock:
4095         unlock_page(page);
4096         put_page(page);
4097         return err;
4098 }
4099 
4100 /*
4101  * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4102  * starting from file offset 'from'.  The range to be zero'd must
4103  * be contained with in one block.  If the specified range exceeds
4104  * the end of the block it will be shortened to end of the block
4105  * that cooresponds to 'from'
4106  */
4107 static int ext4_block_zero_page_range(handle_t *handle,
4108                 struct address_space *mapping, loff_t from, loff_t length)
4109 {
4110         struct inode *inode = mapping->host;
4111         unsigned offset = from & (PAGE_SIZE-1);
4112         unsigned blocksize = inode->i_sb->s_blocksize;
4113         unsigned max = blocksize - (offset & (blocksize - 1));
4114 
4115         /*
4116          * correct length if it does not fall between
4117          * 'from' and the end of the block
4118          */
4119         if (length > max || length < 0)
4120                 length = max;
4121 
4122         if (IS_DAX(inode)) {
4123                 return iomap_zero_range(inode, from, length, NULL,
4124                                         &ext4_iomap_ops);
4125         }
4126         return __ext4_block_zero_page_range(handle, mapping, from, length);
4127 }
4128 
4129 /*
4130  * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4131  * up to the end of the block which corresponds to `from'.
4132  * This required during truncate. We need to physically zero the tail end
4133  * of that block so it doesn't yield old data if the file is later grown.
4134  */
4135 static int ext4_block_truncate_page(handle_t *handle,
4136                 struct address_space *mapping, loff_t from)
4137 {
4138         unsigned offset = from & (PAGE_SIZE-1);
4139         unsigned length;
4140         unsigned blocksize;
4141         struct inode *inode = mapping->host;
4142 
4143         /* If we are processing an encrypted inode during orphan list handling */
4144         if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4145                 return 0;
4146 
4147         blocksize = inode->i_sb->s_blocksize;
4148         length = blocksize - (offset & (blocksize - 1));
4149 
4150         return ext4_block_zero_page_range(handle, mapping, from, length);
4151 }
4152 
4153 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4154                              loff_t lstart, loff_t length)
4155 {
4156         struct super_block *sb = inode->i_sb;
4157         struct address_space *mapping = inode->i_mapping;
4158         unsigned partial_start, partial_end;
4159         ext4_fsblk_t start, end;
4160         loff_t byte_end = (lstart + length - 1);
4161         int err = 0;
4162 
4163         partial_start = lstart & (sb->s_blocksize - 1);
4164         partial_end = byte_end & (sb->s_blocksize - 1);
4165 
4166         start = lstart >> sb->s_blocksize_bits;
4167         end = byte_end >> sb->s_blocksize_bits;
4168 
4169         /* Handle partial zero within the single block */
4170         if (start == end &&
4171             (partial_start || (partial_end != sb->s_blocksize - 1))) {
4172                 err = ext4_block_zero_page_range(handle, mapping,
4173                                                  lstart, length);
4174                 return err;
4175         }
4176         /* Handle partial zero out on the start of the range */
4177         if (partial_start) {
4178                 err = ext4_block_zero_page_range(handle, mapping,
4179                                                  lstart, sb->s_blocksize);
4180                 if (err)
4181                         return err;
4182         }
4183         /* Handle partial zero out on the end of the range */
4184         if (partial_end != sb->s_blocksize - 1)
4185                 err = ext4_block_zero_page_range(handle, mapping,
4186                                                  byte_end - partial_end,
4187                                                  partial_end + 1);
4188         return err;
4189 }
4190 
4191 int ext4_can_truncate(struct inode *inode)
4192 {
4193         if (S_ISREG(inode->i_mode))
4194                 return 1;
4195         if (S_ISDIR(inode->i_mode))
4196                 return 1;
4197         if (S_ISLNK(inode->i_mode))
4198                 return !ext4_inode_is_fast_symlink(inode);
4199         return 0;
4200 }
4201 
4202 /*
4203  * We have to make sure i_disksize gets properly updated before we truncate
4204  * page cache due to hole punching or zero range. Otherwise i_disksize update
4205  * can get lost as it may have been postponed to submission of writeback but
4206  * that will never happen after we truncate page cache.
4207  */
4208 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4209                                       loff_t len)
4210 {
4211         handle_t *handle;
4212         loff_t size = i_size_read(inode);
4213 
4214         WARN_ON(!inode_is_locked(inode));
4215         if (offset > size || offset + len < size)
4216                 return 0;
4217 
4218         if (EXT4_I(inode)->i_disksize >= size)
4219                 return 0;
4220 
4221         handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4222         if (IS_ERR(handle))
4223                 return PTR_ERR(handle);
4224         ext4_update_i_disksize(inode, size);
4225         ext4_mark_inode_dirty(handle, inode);
4226         ext4_journal_stop(handle);
4227 
4228         return 0;
4229 }
4230 
4231 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
4232 {
4233         up_write(&ei->i_mmap_sem);
4234         schedule();
4235         down_write(&ei->i_mmap_sem);
4236 }
4237 
4238 int ext4_break_layouts(struct inode *inode)
4239 {
4240         struct ext4_inode_info *ei = EXT4_I(inode);
4241         struct page *page;
4242         int error;
4243 
4244         if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4245                 return -EINVAL;
4246 
4247         do {
4248                 page = dax_layout_busy_page(inode->i_mapping);
4249                 if (!page)
4250                         return 0;
4251 
4252                 error = ___wait_var_event(&page->_refcount,
4253                                 atomic_read(&page->_refcount) == 1,
4254                                 TASK_INTERRUPTIBLE, 0, 0,
4255                                 ext4_wait_dax_page(ei));
4256         } while (error == 0);
4257 
4258         return error;
4259 }
4260 
4261 /*
4262  * ext4_punch_hole: punches a hole in a file by releasing the blocks
4263  * associated with the given offset and length
4264  *
4265  * @inode:  File inode
4266  * @offset: The offset where the hole will begin
4267  * @len:    The length of the hole
4268  *
4269  * Returns: 0 on success or negative on failure
4270  */
4271 
4272 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4273 {
4274         struct super_block *sb = inode->i_sb;
4275         ext4_lblk_t first_block, stop_block;
4276         struct address_space *mapping = inode->i_mapping;
4277         loff_t first_block_offset, last_block_offset;
4278         handle_t *handle;
4279         unsigned int credits;
4280         int ret = 0;
4281 
4282         if (!S_ISREG(inode->i_mode))
4283                 return -EOPNOTSUPP;
4284 
4285         trace_ext4_punch_hole(inode, offset, length, 0);
4286 
4287         /*
4288          * Write out all dirty pages to avoid race conditions
4289          * Then release them.
4290          */
4291         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4292                 ret = filemap_write_and_wait_range(mapping, offset,
4293                                                    offset + length - 1);
4294                 if (ret)
4295                         return ret;
4296         }
4297 
4298         inode_lock(inode);
4299 
4300         /* No need to punch hole beyond i_size */
4301         if (offset >= inode->i_size)
4302                 goto out_mutex;
4303 
4304         /*
4305          * If the hole extends beyond i_size, set the hole
4306          * to end after the page that contains i_size
4307          */
4308         if (offset + length > inode->i_size) {
4309                 length = inode->i_size +
4310                    PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4311                    offset;
4312         }
4313 
4314         if (offset & (sb->s_blocksize - 1) ||
4315             (offset + length) & (sb->s_blocksize - 1)) {
4316                 /*
4317                  * Attach jinode to inode for jbd2 if we do any zeroing of
4318                  * partial block
4319                  */
4320                 ret = ext4_inode_attach_jinode(inode);
4321                 if (ret < 0)
4322                         goto out_mutex;
4323 
4324         }
4325 
4326         /* Wait all existing dio workers, newcomers will block on i_mutex */
4327         inode_dio_wait(inode);
4328 
4329         /*
4330          * Prevent page faults from reinstantiating pages we have released from
4331          * page cache.
4332          */
4333         down_write(&EXT4_I(inode)->i_mmap_sem);
4334 
4335         ret = ext4_break_layouts(inode);
4336         if (ret)
4337                 goto out_dio;
4338 
4339         first_block_offset = round_up(offset, sb->s_blocksize);
4340         last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4341 
4342         /* Now release the pages and zero block aligned part of pages*/
4343         if (last_block_offset > first_block_offset) {
4344                 ret = ext4_update_disksize_before_punch(inode, offset, length);
4345                 if (ret)
4346                         goto out_dio;
4347                 truncate_pagecache_range(inode, first_block_offset,
4348                                          last_block_offset);
4349         }
4350 
4351         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4352                 credits = ext4_writepage_trans_blocks(inode);
4353         else
4354                 credits = ext4_blocks_for_truncate(inode);
4355         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4356         if (IS_ERR(handle)) {
4357                 ret = PTR_ERR(handle);
4358                 ext4_std_error(sb, ret);
4359                 goto out_dio;
4360         }
4361 
4362         ret = ext4_zero_partial_blocks(handle, inode, offset,
4363                                        length);
4364         if (ret)
4365                 goto out_stop;
4366 
4367         first_block = (offset + sb->s_blocksize - 1) >>
4368                 EXT4_BLOCK_SIZE_BITS(sb);
4369         stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4370 
4371         /* If there are blocks to remove, do it */
4372         if (stop_block > first_block) {
4373 
4374                 down_write(&EXT4_I(inode)->i_data_sem);
4375                 ext4_discard_preallocations(inode);
4376 
4377                 ret = ext4_es_remove_extent(inode, first_block,
4378                                             stop_block - first_block);
4379                 if (ret) {
4380                         up_write(&EXT4_I(inode)->i_data_sem);
4381                         goto out_stop;
4382                 }
4383 
4384                 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4385                         ret = ext4_ext_remove_space(inode, first_block,
4386                                                     stop_block - 1);
4387                 else
4388                         ret = ext4_ind_remove_space(handle, inode, first_block,
4389                                                     stop_block);
4390 
4391                 up_write(&EXT4_I(inode)->i_data_sem);
4392         }
4393         if (IS_SYNC(inode))
4394                 ext4_handle_sync(handle);
4395 
4396         inode->i_mtime = inode->i_ctime = current_time(inode);
4397         ext4_mark_inode_dirty(handle, inode);
4398         if (ret >= 0)
4399                 ext4_update_inode_fsync_trans(handle, inode, 1);
4400 out_stop:
4401         ext4_journal_stop(handle);
4402 out_dio:
4403         up_write(&EXT4_I(inode)->i_mmap_sem);
4404 out_mutex:
4405         inode_unlock(inode);
4406         return ret;
4407 }
4408 
4409 int ext4_inode_attach_jinode(struct inode *inode)
4410 {
4411         struct ext4_inode_info *ei = EXT4_I(inode);
4412         struct jbd2_inode *jinode;
4413 
4414         if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4415                 return 0;
4416 
4417         jinode = jbd2_alloc_inode(GFP_KERNEL);
4418         spin_lock(&inode->i_lock);
4419         if (!ei->jinode) {
4420                 if (!jinode) {
4421                         spin_unlock(&inode->i_lock);
4422                         return -ENOMEM;
4423                 }
4424                 ei->jinode = jinode;
4425                 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4426                 jinode = NULL;
4427         }
4428         spin_unlock(&inode->i_lock);
4429         if (unlikely(jinode != NULL))
4430                 jbd2_free_inode(jinode);
4431         return 0;
4432 }
4433 
4434 /*
4435  * ext4_truncate()
4436  *
4437  * We block out ext4_get_block() block instantiations across the entire
4438  * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4439  * simultaneously on behalf of the same inode.
4440  *
4441  * As we work through the truncate and commit bits of it to the journal there
4442  * is one core, guiding principle: the file's tree must always be consistent on
4443  * disk.  We must be able to restart the truncate after a crash.
4444  *
4445  * The file's tree may be transiently inconsistent in memory (although it
4446  * probably isn't), but whenever we close off and commit a journal transaction,
4447  * the contents of (the filesystem + the journal) must be consistent and
4448  * restartable.  It's pretty simple, really: bottom up, right to left (although
4449  * left-to-right works OK too).
4450  *
4451  * Note that at recovery time, journal replay occurs *before* the restart of
4452  * truncate against the orphan inode list.
4453  *
4454  * The committed inode has the new, desired i_size (which is the same as
4455  * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
4456  * that this inode's truncate did not complete and it will again call
4457  * ext4_truncate() to have another go.  So there will be instantiated blocks
4458  * to the right of the truncation point in a crashed ext4 filesystem.  But
4459  * that's fine - as long as they are linked from the inode, the post-crash
4460  * ext4_truncate() run will find them and release them.
4461  */
4462 int ext4_truncate(struct inode *inode)
4463 {
4464         struct ext4_inode_info *ei = EXT4_I(inode);
4465         unsigned int credits;
4466         int err = 0;
4467         handle_t *handle;
4468         struct address_space *mapping = inode->i_mapping;
4469 
4470         /*
4471          * There is a possibility that we're either freeing the inode
4472          * or it's a completely new inode. In those cases we might not
4473          * have i_mutex locked because it's not necessary.
4474          */
4475         if (!(inode->i_state & (I_NEW|I_FREEING)))
4476                 WARN_ON(!inode_is_locked(inode));
4477         trace_ext4_truncate_enter(inode);
4478 
4479         if (!ext4_can_truncate(inode))
4480                 return 0;
4481 
4482         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4483 
4484         if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4485                 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4486 
4487         if (ext4_has_inline_data(inode)) {
4488                 int has_inline = 1;
4489 
4490                 err = ext4_inline_data_truncate(inode, &has_inline);
4491                 if (err)
4492                         return err;
4493                 if (has_inline)
4494                         return 0;
4495         }
4496 
4497         /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4498         if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4499                 if (ext4_inode_attach_jinode(inode) < 0)
4500                         return 0;
4501         }
4502 
4503         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4504                 credits = ext4_writepage_trans_blocks(inode);
4505         else
4506                 credits = ext4_blocks_for_truncate(inode);
4507 
4508         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4509         if (IS_ERR(handle))
4510                 return PTR_ERR(handle);
4511 
4512         if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4513                 ext4_block_truncate_page(handle, mapping, inode->i_size);
4514 
4515         /*
4516          * We add the inode to the orphan list, so that if this
4517          * truncate spans multiple transactions, and we crash, we will
4518          * resume the truncate when the filesystem recovers.  It also
4519          * marks the inode dirty, to catch the new size.
4520          *
4521          * Implication: the file must always be in a sane, consistent
4522          * truncatable state while each transaction commits.
4523          */
4524         err = ext4_orphan_add(handle, inode);
4525         if (err)
4526                 goto out_stop;
4527 
4528         down_write(&EXT4_I(inode)->i_data_sem);
4529 
4530         ext4_discard_preallocations(inode);
4531 
4532         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4533                 err = ext4_ext_truncate(handle, inode);
4534         else
4535                 ext4_ind_truncate(handle, inode);
4536 
4537         up_write(&ei->i_data_sem);
4538         if (err)
4539                 goto out_stop;
4540 
4541         if (IS_SYNC(inode))
4542                 ext4_handle_sync(handle);
4543 
4544 out_stop:
4545         /*
4546          * If this was a simple ftruncate() and the file will remain alive,
4547          * then we need to clear up the orphan record which we created above.
4548          * However, if this was a real unlink then we were called by
4549          * ext4_evict_inode(), and we allow that function to clean up the
4550          * orphan info for us.
4551          */
4552         if (inode->i_nlink)
4553                 ext4_orphan_del(handle, inode);
4554 
4555         inode->i_mtime = inode->i_ctime = current_time(inode);
4556         ext4_mark_inode_dirty(handle, inode);
4557         ext4_journal_stop(handle);
4558 
4559         trace_ext4_truncate_exit(inode);
4560         return err;
4561 }
4562 
4563 /*
4564  * ext4_get_inode_loc returns with an extra refcount against the inode's
4565  * underlying buffer_head on success. If 'in_mem' is true, we have all
4566  * data in memory that is needed to recreate the on-disk version of this
4567  * inode.
4568  */
4569 static int __ext4_get_inode_loc(struct inode *inode,
4570                                 struct ext4_iloc *iloc, int in_mem)
4571 {
4572         struct ext4_group_desc  *gdp;
4573         struct buffer_head      *bh;
4574         struct super_block      *sb = inode->i_sb;
4575         ext4_fsblk_t            block;
4576         int                     inodes_per_block, inode_offset;
4577 
4578         iloc->bh = NULL;
4579         if (inode->i_ino < EXT4_ROOT_INO ||
4580             inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4581                 return -EFSCORRUPTED;
4582 
4583         iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4584         gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4585         if (!gdp)
4586                 return -EIO;
4587 
4588         /*
4589          * Figure out the offset within the block group inode table
4590          */
4591         inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4592         inode_offset = ((inode->i_ino - 1) %
4593                         EXT4_INODES_PER_GROUP(sb));
4594         block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4595         iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4596 
4597         bh = sb_getblk(sb, block);
4598         if (unlikely(!bh))
4599                 return -ENOMEM;
4600         if (!buffer_uptodate(bh)) {
4601                 lock_buffer(bh);
4602 
4603                 /*
4604                  * If the buffer has the write error flag, we have failed
4605                  * to write out another inode in the same block.  In this
4606                  * case, we don't have to read the block because we may
4607                  * read the old inode data successfully.
4608                  */
4609                 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4610                         set_buffer_uptodate(bh);
4611 
4612                 if (buffer_uptodate(bh)) {
4613                         /* someone brought it uptodate while we waited */
4614                         unlock_buffer(bh);
4615                         goto has_buffer;
4616                 }
4617 
4618                 /*
4619                  * If we have all information of the inode in memory and this
4620                  * is the only valid inode in the block, we need not read the
4621                  * block.
4622                  */
4623                 if (in_mem) {
4624                         struct buffer_head *bitmap_bh;
4625                         int i, start;
4626 
4627                         start = inode_offset & ~(inodes_per_block - 1);
4628 
4629                         /* Is the inode bitmap in cache? */
4630                         bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4631                         if (unlikely(!bitmap_bh))
4632                                 goto make_io;
4633 
4634                         /*
4635                          * If the inode bitmap isn't in cache then the
4636                          * optimisation may end up performing two reads instead
4637                          * of one, so skip it.
4638                          */
4639                         if (!buffer_uptodate(bitmap_bh)) {
4640                                 brelse(bitmap_bh);
4641                                 goto make_io;
4642                         }
4643                         for (i = start; i < start + inodes_per_block; i++) {
4644                                 if (i == inode_offset)
4645                                         continue;
4646                                 if (ext4_test_bit(i, bitmap_bh->b_data))
4647                                         break;
4648                         }
4649                         brelse(bitmap_bh);
4650                         if (i == start + inodes_per_block) {
4651                                 /* all other inodes are free, so skip I/O */
4652                                 memset(bh->b_data, 0, bh->b_size);
4653                                 set_buffer_uptodate(bh);
4654                                 unlock_buffer(bh);
4655                                 goto has_buffer;
4656                         }
4657                 }
4658 
4659 make_io:
4660                 /*
4661                  * If we need to do any I/O, try to pre-readahead extra
4662                  * blocks from the inode table.
4663                  */
4664                 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4665                         ext4_fsblk_t b, end, table;
4666                         unsigned num;
4667                         __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4668 
4669                         table = ext4_inode_table(sb, gdp);
4670                         /* s_inode_readahead_blks is always a power of 2 */
4671                         b = block & ~((ext4_fsblk_t) ra_blks - 1);
4672                         if (table > b)
4673                                 b = table;
4674                         end = b + ra_blks;
4675                         num = EXT4_INODES_PER_GROUP(sb);
4676                         if (ext4_has_group_desc_csum(sb))
4677                                 num -= ext4_itable_unused_count(sb, gdp);
4678                         table += num / inodes_per_block;
4679                         if (end > table)
4680                                 end = table;
4681                         while (b <= end)
4682                                 sb_breadahead(sb, b++);
4683                 }
4684 
4685                 /*
4686                  * There are other valid inodes in the buffer, this inode
4687                  * has in-inode xattrs, or we don't have this inode in memory.
4688                  * Read the block from disk.
4689                  */
4690                 trace_ext4_load_inode(inode);
4691                 get_bh(bh);
4692                 bh->b_end_io = end_buffer_read_sync;
4693                 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4694                 wait_on_buffer(bh);
4695                 if (!buffer_uptodate(bh)) {
4696                         EXT4_ERROR_INODE_BLOCK(inode, block,
4697                                                "unable to read itable block");
4698                         brelse(bh);
4699                         return -EIO;
4700                 }
4701         }
4702 has_buffer:
4703         iloc->bh = bh;
4704         return 0;
4705 }
4706 
4707 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4708 {
4709         /* We have all inode data except xattrs in memory here. */
4710         return __ext4_get_inode_loc(inode, iloc,
4711                 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4712 }
4713 
4714 static bool ext4_should_use_dax(struct inode *inode)
4715 {
4716         if (!test_opt(inode->i_sb, DAX))
4717                 return false;
4718         if (!S_ISREG(inode->i_mode))
4719                 return false;
4720         if (ext4_should_journal_data(inode))
4721                 return false;
4722         if (ext4_has_inline_data(inode))
4723                 return false;
4724         if (ext4_encrypted_inode(inode))
4725                 return false;
4726         return true;
4727 }
4728 
4729 void ext4_set_inode_flags(struct inode *inode)
4730 {
4731         unsigned int flags = EXT4_I(inode)->i_flags;
4732         unsigned int new_fl = 0;
4733 
4734         if (flags & EXT4_SYNC_FL)
4735                 new_fl |= S_SYNC;
4736         if (flags & EXT4_APPEND_FL)
4737                 new_fl |= S_APPEND;
4738         if (flags & EXT4_IMMUTABLE_FL)
4739                 new_fl |= S_IMMUTABLE;
4740         if (flags & EXT4_NOATIME_FL)
4741                 new_fl |= S_NOATIME;
4742         if (flags & EXT4_DIRSYNC_FL)
4743                 new_fl |= S_DIRSYNC;
4744         if (ext4_should_use_dax(inode))
4745                 new_fl |= S_DAX;
4746         if (flags & EXT4_ENCRYPT_FL)
4747                 new_fl |= S_ENCRYPTED;
4748         inode_set_flags(inode, new_fl,
4749                         S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4750                         S_ENCRYPTED);
4751 }
4752 
4753 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4754                                   struct ext4_inode_info *ei)
4755 {
4756         blkcnt_t i_blocks ;
4757         struct inode *inode = &(ei->vfs_inode);
4758         struct super_block *sb = inode->i_sb;
4759 
4760         if (ext4_has_feature_huge_file(sb)) {
4761                 /* we are using combined 48 bit field */
4762                 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4763                                         le32_to_cpu(raw_inode->i_blocks_lo);
4764                 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4765                         /* i_blocks represent file system block size */
4766                         return i_blocks  << (inode->i_blkbits - 9);
4767                 } else {
4768                         return i_blocks;
4769                 }
4770         } else {
4771                 return le32_to_cpu(raw_inode->i_blocks_lo);
4772         }
4773 }
4774 
4775 static inline int ext4_iget_extra_inode(struct inode *inode,
4776                                          struct ext4_inode *raw_inode,
4777                                          struct ext4_inode_info *ei)
4778 {
4779         __le32 *magic = (void *)raw_inode +
4780                         EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4781 
4782         if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4783             EXT4_INODE_SIZE(inode->i_sb) &&
4784             *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4785                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4786                 return ext4_find_inline_data_nolock(inode);
4787         } else
4788                 EXT4_I(inode)->i_inline_off = 0;
4789         return 0;
4790 }
4791 
4792 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4793 {
4794         if (!ext4_has_feature_project(inode->i_sb))
4795                 return -EOPNOTSUPP;
4796         *projid = EXT4_I(inode)->i_projid;
4797         return 0;
4798 }
4799 
4800 /*
4801  * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4802  * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4803  * set.
4804  */
4805 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4806 {
4807         if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4808                 inode_set_iversion_raw(inode, val);
4809         else
4810                 inode_set_iversion_queried(inode, val);
4811 }
4812 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4813 {
4814         if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4815                 return inode_peek_iversion_raw(inode);
4816         else
4817                 return inode_peek_iversion(inode);
4818 }
4819 
4820 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4821 {
4822         struct ext4_iloc iloc;
4823         struct ext4_inode *raw_inode;
4824         struct ext4_inode_info *ei;
4825         struct inode *inode;
4826         journal_t *journal = EXT4_SB(sb)->s_journal;
4827         long ret;
4828         loff_t size;
4829         int block;
4830         uid_t i_uid;
4831         gid_t i_gid;
4832         projid_t i_projid;
4833 
4834         inode = iget_locked(sb, ino);
4835         if (!inode)
4836                 return ERR_PTR(-ENOMEM);
4837         if (!(inode->i_state & I_NEW))
4838                 return inode;
4839 
4840         ei = EXT4_I(inode);
4841         iloc.bh = NULL;
4842 
4843         ret = __ext4_get_inode_loc(inode, &iloc, 0);
4844         if (ret < 0)
4845                 goto bad_inode;
4846         raw_inode = ext4_raw_inode(&iloc);
4847 
4848         if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4849                 EXT4_ERROR_INODE(inode, "root inode unallocated");
4850                 ret = -EFSCORRUPTED;
4851                 goto bad_inode;
4852         }
4853 
4854         if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4855                 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4856                 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4857                         EXT4_INODE_SIZE(inode->i_sb) ||
4858                     (ei->i_extra_isize & 3)) {
4859                         EXT4_ERROR_INODE(inode,
4860                                          "bad extra_isize %u (inode size %u)",
4861                                          ei->i_extra_isize,
4862                                          EXT4_INODE_SIZE(inode->i_sb));
4863                         ret = -EFSCORRUPTED;
4864                         goto bad_inode;
4865                 }
4866         } else
4867                 ei->i_extra_isize = 0;
4868 
4869         /* Precompute checksum seed for inode metadata */
4870         if (ext4_has_metadata_csum(sb)) {
4871                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4872                 __u32 csum;
4873                 __le32 inum = cpu_to_le32(inode->i_ino);
4874                 __le32 gen = raw_inode->i_generation;
4875                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4876                                    sizeof(inum));
4877                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4878                                               sizeof(gen));
4879         }
4880 
4881         if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4882                 EXT4_ERROR_INODE(inode, "checksum invalid");
4883                 ret = -EFSBADCRC;
4884                 goto bad_inode;
4885         }
4886 
4887         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4888         i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4889         i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4890         if (ext4_has_feature_project(sb) &&
4891             EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4892             EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4893                 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4894         else
4895                 i_projid = EXT4_DEF_PROJID;
4896 
4897         if (!(test_opt(inode->i_sb, NO_UID32))) {
4898                 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4899                 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4900         }
4901         i_uid_write(inode, i_uid);
4902         i_gid_write(inode, i_gid);
4903         ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4904         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4905 
4906         ext4_clear_state_flags(ei);     /* Only relevant on 32-bit archs */
4907         ei->i_inline_off = 0;
4908         ei->i_dir_start_lookup = 0;
4909         ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4910         /* We now have enough fields to check if the inode was active or not.
4911          * This is needed because nfsd might try to access dead inodes
4912          * the test is that same one that e2fsck uses
4913          * NeilBrown 1999oct15
4914          */
4915         if (inode->i_nlink == 0) {
4916                 if ((inode->i_mode == 0 ||
4917                      !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4918                     ino != EXT4_BOOT_LOADER_INO) {
4919                         /* this inode is deleted */
4920                         ret = -ESTALE;
4921                         goto bad_inode;
4922                 }
4923                 /* The only unlinked inodes we let through here have
4924                  * valid i_mode and are being read by the orphan
4925                  * recovery code: that's fine, we're about to complete
4926                  * the process of deleting those.
4927                  * OR it is the EXT4_BOOT_LOADER_INO which is
4928                  * not initialized on a new filesystem. */
4929         }
4930         ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4931         ext4_set_inode_flags(inode);
4932         inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4933         ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4934         if (ext4_has_feature_64bit(sb))
4935                 ei->i_file_acl |=
4936                         ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4937         inode->i_size = ext4_isize(sb, raw_inode);
4938         if ((size = i_size_read(inode)) < 0) {
4939                 EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
4940                 ret = -EFSCORRUPTED;
4941                 goto bad_inode;
4942         }
4943         ei->i_disksize = inode->i_size;
4944 #ifdef CONFIG_QUOTA
4945         ei->i_reserved_quota = 0;
4946 #endif
4947         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4948         ei->i_block_group = iloc.block_group;
4949         ei->i_last_alloc_group = ~0;
4950         /*
4951          * NOTE! The in-memory inode i_data array is in little-endian order
4952          * even on big-endian machines: we do NOT byteswap the block numbers!
4953          */
4954         for (block = 0; block < EXT4_N_BLOCKS; block++)
4955                 ei->i_data[block] = raw_inode->i_block[block];
4956         INIT_LIST_HEAD(&ei->i_orphan);
4957 
4958         /*
4959          * Set transaction id's of transactions that have to be committed
4960          * to finish f[data]sync. We set them to currently running transaction
4961          * as we cannot be sure that the inode or some of its metadata isn't
4962          * part of the transaction - the inode could have been reclaimed and
4963          * now it is reread from disk.
4964          */
4965         if (journal) {
4966                 transaction_t *transaction;
4967                 tid_t tid;
4968 
4969                 read_lock(&journal->j_state_lock);
4970                 if (journal->j_running_transaction)
4971                         transaction = journal->j_running_transaction;
4972                 else
4973                         transaction = journal->j_committing_transaction;
4974                 if (transaction)
4975                         tid = transaction->t_tid;
4976                 else
4977                         tid = journal->j_commit_sequence;
4978                 read_unlock(&journal->j_state_lock);
4979                 ei->i_sync_tid = tid;
4980                 ei->i_datasync_tid = tid;
4981         }
4982 
4983         if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4984                 if (ei->i_extra_isize == 0) {
4985                         /* The extra space is currently unused. Use it. */
4986                         BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4987                         ei->i_extra_isize = sizeof(struct ext4_inode) -
4988                                             EXT4_GOOD_OLD_INODE_SIZE;
4989                 } else {
4990                         ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4991                         if (ret)
4992                                 goto bad_inode;
4993                 }
4994         }
4995 
4996         EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4997         EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4998         EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4999         EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5000 
5001         if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5002                 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5003 
5004                 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5005                         if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5006                                 ivers |=
5007                     (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5008                 }
5009                 ext4_inode_set_iversion_queried(inode, ivers);
5010         }
5011 
5012         ret = 0;
5013         if (ei->i_file_acl &&
5014             !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5015                 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5016                                  ei->i_file_acl);
5017                 ret = -EFSCORRUPTED;
5018                 goto bad_inode;
5019         } else if (!ext4_has_inline_data(inode)) {
5020                 /* validate the block references in the inode */
5021                 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5022                    (S_ISLNK(inode->i_mode) &&
5023                     !ext4_inode_is_fast_symlink(inode))) {
5024                         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5025                                 ret = ext4_ext_check_inode(inode);
5026                         else
5027                                 ret = ext4_ind_check_inode(inode);
5028                 }
5029         }
5030         if (ret)
5031                 goto bad_inode;
5032 
5033         if (S_ISREG(inode->i_mode)) {
5034                 inode->i_op = &ext4_file_inode_operations;
5035                 inode->i_fop = &ext4_file_operations;
5036                 ext4_set_aops(inode);
5037         } else if (S_ISDIR(inode->i_mode)) {
5038                 inode->i_op = &ext4_dir_inode_operations;
5039                 inode->i_fop = &ext4_dir_operations;
5040         } else if (S_ISLNK(inode->i_mode)) {
5041                 /* VFS does not allow setting these so must be corruption */
5042                 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5043                         EXT4_ERROR_INODE(inode,
5044                           "immutable or append flags not allowed on symlinks");
5045                         ret = -EFSCORRUPTED;
5046                         goto bad_inode;
5047                 }
5048                 if (ext4_encrypted_inode(inode)) {
5049                         inode->i_op = &ext4_encrypted_symlink_inode_operations;
5050                         ext4_set_aops(inode);
5051                 } else if (ext4_inode_is_fast_symlink(inode)) {
5052                         inode->i_link = (char *)ei->i_data;
5053                         inode->i_op = &ext4_fast_symlink_inode_operations;
5054                         nd_terminate_link(ei->i_data, inode->i_size,
5055                                 sizeof(ei->i_data) - 1);
5056                 } else {
5057                         inode->i_op = &ext4_symlink_inode_operations;
5058                         ext4_set_aops(inode);
5059                 }
5060                 inode_nohighmem(inode);
5061         } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5062               S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5063                 inode->i_op = &ext4_special_inode_operations;
5064                 if (raw_inode->i_block[0])
5065                         init_special_inode(inode, inode->i_mode,
5066                            old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5067                 else
5068                         init_special_inode(inode, inode->i_mode,
5069                            new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5070         } else if (ino == EXT4_BOOT_LOADER_INO) {
5071                 make_bad_inode(inode);
5072         } else {
5073                 ret = -EFSCORRUPTED;
5074                 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5075                 goto bad_inode;
5076         }
5077         brelse(iloc.bh);
5078 
5079         unlock_new_inode(inode);
5080         return inode;
5081 
5082 bad_inode:
5083         brelse(iloc.bh);
5084         iget_failed(inode);
5085         return ERR_PTR(ret);
5086 }
5087 
5088 struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
5089 {
5090         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
5091                 return ERR_PTR(-EFSCORRUPTED);
5092         return ext4_iget(sb, ino);
5093 }
5094 
5095 static int ext4_inode_blocks_set(handle_t *handle,
5096                                 struct ext4_inode *raw_inode,
5097                                 struct ext4_inode_info *ei)
5098 {
5099         struct inode *inode = &(ei->vfs_inode);
5100         u64 i_blocks = inode->i_blocks;
5101         struct super_block *sb = inode->i_sb;
5102 
5103         if (i_blocks <= ~0U) {
5104                 /*
5105                  * i_blocks can be represented in a 32 bit variable
5106                  * as multiple of 512 bytes
5107                  */
5108                 raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
5109                 raw_inode->i_blocks_high = 0;
5110                 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5111                 return 0;
5112         }
5113         if (!ext4_has_feature_huge_file(sb))
5114                 return -EFBIG;
5115 
5116         if (i_blocks <= 0xffffffffffffULL) {
5117                 /*
5118                  * i_blocks can be represented in a 48 bit variable
5119                  * as multiple of 512 bytes
5120                  */
5121                 raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
5122                 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5123                 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5124         } else {
5125                 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5126                 /* i_block is stored in file system block size */
5127                 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5128                 raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
5129                 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5130         }
5131         return 0;
5132 }
5133 
5134 struct other_inode {
5135         unsigned long           orig_ino;
5136         struct ext4_inode       *raw_inode;
5137 };
5138 
5139 static int other_inode_match(struct inode * inode, unsigned long ino,
5140                              void *data)
5141 {
5142         struct other_inode *oi = (struct other_inode *) data;
5143 
5144         if ((inode->i_ino != ino) ||
5145             (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5146                                I_DIRTY_INODE)) ||
5147             ((inode->i_state & I_DIRTY_TIME) == 0))
5148                 return 0;
5149         spin_lock(&inode->i_lock);
5150         if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5151                                 I_DIRTY_INODE)) == 0) &&
5152             (inode->i_state & I_DIRTY_TIME)) {
5153                 struct ext4_inode_info  *ei = EXT4_I(inode);
5154 
5155                 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5156                 spin_unlock(&inode->i_lock);
5157 
5158                 spin_lock(&ei->i_raw_lock);
5159                 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5160                 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5161                 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5162                 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5163                 spin_unlock(&ei->i_raw_lock);
5164                 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5165                 return -1;
5166         }
5167         spin_unlock(&inode->i_lock);
5168         return -1;
5169 }
5170 
5171 /*
5172  * Opportunistically update the other time fields for other inodes in
5173  * the same inode table block.
5174  */
5175 static void ext4_update_other_inodes_time(struct super_block *sb,
5176                                           unsigned long orig_ino, char *buf)
5177 {
5178         struct other_inode oi;
5179         unsigned long ino;
5180         int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5181         int inode_size = EXT4_INODE_SIZE(sb);
5182 
5183         oi.orig_ino = orig_ino;
5184         /*
5185          * Calculate the first inode in the inode table block.  Inode
5186          * numbers are one-based.  That is, the first inode in a block
5187          * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5188          */
5189         ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5190         for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5191                 if (ino == orig_ino)
5192                         continue;
5193                 oi.raw_inode = (struct ext4_inode *) buf;
5194                 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5195         }
5196 }
5197 
5198 /*
5199  * Post the struct inode info into an on-disk inode location in the
5200  * buffer-cache.  This gobbles the caller's reference to the
5201  * buffer_head in the inode location struct.
5202  *
5203  * The caller must have write access to iloc->bh.
5204  */
5205 static int ext4_do_update_inode(handle_t *handle,
5206                                 struct inode *inode,
5207                                 struct ext4_iloc *iloc)
5208 {
5209         struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5210         struct ext4_inode_info *ei = EXT4_I(inode);
5211         struct buffer_head *bh = iloc->bh;
5212         struct super_block *sb = inode->i_sb;
5213         int err = 0, rc, block;
5214         int need_datasync = 0, set_large_file = 0;
5215         uid_t i_uid;
5216         gid_t i_gid;
5217         projid_t i_projid;
5218 
5219         spin_lock(&ei->i_raw_lock);
5220 
5221         /* For fields not tracked in the in-memory inode,
5222          * initialise them to zero for new inodes. */
5223         if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5224                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5225 
5226         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5227         i_uid = i_uid_read(inode);
5228         i_gid = i_gid_read(inode);
5229         i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5230         if (!(test_opt(inode->i_sb, NO_UID32))) {
5231                 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5232                 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5233 /*
5234  * Fix up interoperability with old kernels. Otherwise, old inodes get
5235  * re-used with the upper 16 bits of the uid/gid intact
5236  */
5237                 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5238                         raw_inode->i_uid_high = 0;
5239                         raw_inode->i_gid_high = 0;
5240                 } else {
5241                         raw_inode->i_uid_high =
5242                                 cpu_to_le16(high_16_bits(i_uid));
5243                         raw_inode->i_gid_high =
5244                                 cpu_to_le16(high_16_bits(i_gid));
5245                 }
5246         } else {
5247                 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5248                 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5249                 raw_inode->i_uid_high = 0;
5250                 raw_inode->i_gid_high = 0;
5251         }
5252         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5253 
5254         EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5255         EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5256         EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5257         EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5258 
5259         err = ext4_inode_blocks_set(handle, raw_inode, ei);
5260         if (err) {
5261                 spin_unlock(&ei->i_raw_lock);
5262                 goto out_brelse;
5263         }
5264         raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5265         raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5266         if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5267                 raw_inode->i_file_acl_high =
5268                         cpu_to_le16(ei->i_file_acl >> 32);
5269         raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5270         if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5271                 ext4_isize_set(raw_inode, ei->i_disksize);
5272                 need_datasync = 1;
5273         }
5274         if (ei->i_disksize > 0x7fffffffULL) {
5275                 if (!ext4_has_feature_large_file(sb) ||
5276                                 EXT4_SB(sb)->s_es->s_rev_level ==
5277                     cpu_to_le32(EXT4_GOOD_OLD_REV))
5278                         set_large_file = 1;
5279         }
5280         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5281         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5282                 if (old_valid_dev(inode->i_rdev)) {
5283                         raw_inode->i_block[0] =
5284                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
5285                         raw_inode->i_block[1] = 0;
5286                 } else {
5287                         raw_inode->i_block[0] = 0;
5288                         raw_inode->i_block[1] =
5289                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
5290                         raw_inode->i_block[2] = 0;
5291                 }
5292         } else if (!ext4_has_inline_data(inode)) {
5293                 for (block = 0; block < EXT4_N_BLOCKS; block++)
5294                         raw_inode->i_block[block] = ei->i_data[block];
5295         }
5296 
5297         if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5298                 u64 ivers = ext4_inode_peek_iversion(inode);
5299 
5300                 raw_inode->i_disk_version = cpu_to_le32(ivers);
5301                 if (ei->i_extra_isize) {
5302                         if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5303                                 raw_inode->i_version_hi =
5304                                         cpu_to_le32(ivers >> 32);
5305                         raw_inode->i_extra_isize =
5306                                 cpu_to_le16(ei->i_extra_isize);
5307                 }
5308         }
5309 
5310         BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5311                i_projid != EXT4_DEF_PROJID);
5312 
5313         if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5314             EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5315                 raw_inode->i_projid = cpu_to_le32(i_projid);
5316 
5317         ext4_inode_csum_set(inode, raw_inode, ei);
5318         spin_unlock(&ei->i_raw_lock);
5319         if (inode->i_sb->s_flags & SB_LAZYTIME)
5320                 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5321                                               bh->b_data);
5322 
5323         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5324         rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5325         if (!err)
5326                 err = rc;
5327         ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5328         if (set_large_file) {
5329                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5330                 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5331                 if (err)
5332                         goto out_brelse;
5333                 ext4_update_dynamic_rev(sb);
5334                 ext4_set_feature_large_file(sb);
5335                 ext4_handle_sync(handle);
5336                 err = ext4_handle_dirty_super(handle, sb);
5337         }
5338         ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5339 out_brelse:
5340         brelse(bh);
5341         ext4_std_error(inode->i_sb, err);
5342         return err;
5343 }
5344 
5345 /*
5346  * ext4_write_inode()
5347  *
5348  * We are called from a few places:
5349  *
5350  * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5351  *   Here, there will be no transaction running. We wait for any running
5352  *   transaction to commit.
5353  *
5354  * - Within flush work (sys_sync(), kupdate and such).
5355  *   We wait on commit, if told to.
5356  *
5357  * - Within iput_final() -> write_inode_now()
5358  *   We wait on commit, if told to.
5359  *
5360  * In all cases it is actually safe for us to return without doing anything,
5361  * because the inode has been copied into a raw inode buffer in
5362  * ext4_mark_inode_dirty().  This is a correctness thing for WB_SYNC_ALL
5363  * writeback.
5364  *
5365  * Note that we are absolutely dependent upon all inode dirtiers doing the
5366  * right thing: they *must* call mark_inode_dirty() after dirtying info in
5367  * which we are interested.
5368  *
5369  * It would be a bug for them to not do this.  The code:
5370  *
5371  *      mark_inode_dirty(inode)
5372  *      stuff();
5373  *      inode->i_size = expr;
5374  *
5375  * is in error because write_inode() could occur while `stuff()' is running,
5376  * and the new i_size will be lost.  Plus the inode will no longer be on the
5377  * superblock's dirty inode list.
5378  */
5379 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5380 {
5381         int err;
5382 
5383         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5384                 return 0;
5385 
5386         if (EXT4_SB(inode->i_sb)->s_journal) {
5387                 if (ext4_journal_current_handle()) {
5388                         jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5389                         dump_stack();
5390                         return -EIO;
5391                 }
5392 
5393                 /*
5394                  * No need to force transaction in WB_SYNC_NONE mode. Also
5395                  * ext4_sync_fs() will force the commit after everything is
5396                  * written.
5397                  */
5398                 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5399                         return 0;
5400 
5401                 err = ext4_force_commit(inode->i_sb);
5402         } else {
5403                 struct ext4_iloc iloc;
5404 
5405                 err = __ext4_get_inode_loc(inode, &iloc, 0);
5406                 if (err)
5407                         return err;
5408                 /*
5409                  * sync(2) will flush the whole buffer cache. No need to do
5410                  * it here separately for each inode.
5411                  */
5412                 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5413                         sync_dirty_buffer(iloc.bh);
5414                 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5415                         EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5416                                          "IO error syncing inode");
5417                         err = -EIO;
5418                 }
5419                 brelse(iloc.bh);
5420         }
5421         return err;
5422 }
5423 
5424 /*
5425  * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5426  * buffers that are attached to a page stradding i_size and are undergoing
5427  * commit. In that case we have to wait for commit to finish and try again.
5428  */
5429 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5430 {
5431         struct page *page;
5432         unsigned offset;
5433         journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5434         tid_t commit_tid = 0;
5435         int ret;
5436 
5437         offset = inode->i_size & (PAGE_SIZE - 1);
5438         /*
5439          * All buffers in the last page remain valid? Then there's nothing to
5440          * do. We do the check mainly to optimize the common PAGE_SIZE ==
5441          * blocksize case
5442          */
5443         if (offset > PAGE_SIZE - i_blocksize(inode))
5444                 return;
5445         while (1) {
5446                 page = find_lock_page(inode->i_mapping,
5447                                       inode->i_size >> PAGE_SHIFT);
5448                 if (!page)
5449                         return;
5450                 ret = __ext4_journalled_invalidatepage(page, offset,
5451                                                 PAGE_SIZE - offset);
5452                 unlock_page(page);
5453                 put_page(page);
5454                 if (ret != -EBUSY)
5455                         return;
5456                 commit_tid = 0;
5457                 read_lock(&journal->j_state_lock);
5458                 if (journal->j_committing_transaction)
5459                         commit_tid = journal->j_committing_transaction->t_tid;
5460                 read_unlock(&journal->j_state_lock);
5461                 if (commit_tid)
5462                         jbd2_log_wait_commit(journal, commit_tid);
5463         }
5464 }
5465 
5466 /*
5467  * ext4_setattr()
5468  *
5469  * Called from notify_change.
5470  *
5471  * We want to trap VFS attempts to truncate the file as soon as
5472  * possible.  In particular, we want to make sure that when the VFS
5473  * shrinks i_size, we put the inode on the orphan list and modify
5474  * i_disksize immediately, so that during the subsequent flushing of
5475  * dirty pages and freeing of disk blocks, we can guarantee that any
5476  * commit will leave the blocks being flushed in an unused state on
5477  * disk.  (On recovery, the inode will get truncated and the blocks will
5478  * be freed, so we have a strong guarantee that no future commit will
5479  * leave these blocks visible to the user.)
5480  *
5481  * Another thing we have to assure is that if we are in ordered mode
5482  * and inode is still attached to the committing transaction, we must
5483  * we start writeout of all the dirty pages which are being truncated.
5484  * This way we are sure that all the data written in the previous
5485  * transaction are already on disk (truncate waits for pages under
5486  * writeback).
5487  *
5488  * Called with inode->i_mutex down.
5489  */
5490 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5491 {
5492         struct inode *inode = d_inode(dentry);
5493         int error, rc = 0;
5494         int orphan = 0;
5495         const unsigned int ia_valid = attr->ia_valid;
5496 
5497         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5498                 return -EIO;
5499 
5500         error = setattr_prepare(dentry, attr);
5501         if (error)
5502                 return error;
5503 
5504         error = fscrypt_prepare_setattr(dentry, attr);
5505         if (error)
5506                 return error;
5507 
5508         if (is_quota_modification(inode, attr)) {
5509                 error = dquot_initialize(inode);
5510                 if (error)
5511                         return error;
5512         }
5513         if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5514             (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5515                 handle_t *handle;
5516 
5517                 /* (user+group)*(old+new) structure, inode write (sb,
5518                  * inode block, ? - but truncate inode update has it) */
5519                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5520                         (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5521                          EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5522                 if (IS_ERR(handle)) {
5523                         error = PTR_ERR(handle);
5524                         goto err_out;
5525                 }
5526 
5527                 /* dquot_transfer() calls back ext4_get_inode_usage() which
5528                  * counts xattr inode references.
5529                  */
5530                 down_read(&EXT4_I(inode)->xattr_sem);
5531                 error = dquot_transfer(inode, attr);
5532                 up_read(&EXT4_I(inode)->xattr_sem);
5533 
5534                 if (error) {
5535                         ext4_journal_stop(handle);
5536                         return error;
5537                 }
5538                 /* Update corresponding info in inode so that everything is in
5539                  * one transaction */
5540                 if (attr->ia_valid & ATTR_UID)
5541                         inode->i_uid = attr->ia_uid;
5542                 if (attr->ia_valid & ATTR_GID)
5543                         inode->i_gid = attr->ia_gid;
5544                 error = ext4_mark_inode_dirty(handle, inode);
5545                 ext4_journal_stop(handle);
5546         }
5547 
5548         if (attr->ia_valid & ATTR_SIZE) {
5549                 handle_t *handle;
5550                 loff_t oldsize = inode->i_size;
5551                 int shrink = (attr->ia_size <= inode->i_size);
5552 
5553                 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5554                         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5555 
5556                         if (attr->ia_size > sbi->s_bitmap_maxbytes)
5557                                 return -EFBIG;
5558                 }
5559                 if (!S_ISREG(inode->i_mode))
5560                         return -EINVAL;
5561 
5562                 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5563                         inode_inc_iversion(inode);
5564 
5565                 if (ext4_should_order_data(inode) &&
5566                     (attr->ia_size < inode->i_size)) {
5567                         error = ext4_begin_ordered_truncate(inode,
5568                                                             attr->ia_size);
5569                         if (error)
5570                                 goto err_out;
5571                 }
5572                 if (attr->ia_size != inode->i_size) {
5573                         handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5574                         if (IS_ERR(handle)) {
5575                                 error = PTR_ERR(handle);
5576                                 goto err_out;
5577                         }
5578                         if (ext4_handle_valid(handle) && shrink) {
5579                                 error = ext4_orphan_add(handle, inode);
5580                                 orphan = 1;
5581                         }
5582                         /*
5583                          * Update c/mtime on truncate up, ext4_truncate() will
5584                          * update c/mtime in shrink case below
5585                          */
5586                         if (!shrink) {
5587                                 inode->i_mtime = current_time(inode);
5588                                 inode->i_ctime = inode->i_mtime;
5589                         }
5590                         down_write(&EXT4_I(inode)->i_data_sem);
5591                         EXT4_I(inode)->i_disksize = attr->ia_size;
5592                         rc = ext4_mark_inode_dirty(handle, inode);
5593                         if (!error)
5594                                 error = rc;
5595                         /*
5596                          * We have to update i_size under i_data_sem together
5597                          * with i_disksize to avoid races with writeback code
5598                          * running ext4_wb_update_i_disksize().
5599                          */
5600                         if (!error)
5601                                 i_size_write(inode, attr->ia_size);
5602                         up_write(&EXT4_I(inode)->i_data_sem);
5603                         ext4_journal_stop(handle);
5604                         if (error) {
5605                                 if (orphan)
5606                                         ext4_orphan_del(NULL, inode);
5607                                 goto err_out;
5608                         }
5609                 }
5610                 if (!shrink)
5611                         pagecache_isize_extended(inode, oldsize, inode->i_size);
5612 
5613                 /*
5614                  * Blocks are going to be removed from the inode. Wait
5615                  * for dio in flight.  Temporarily disable
5616                  * dioread_nolock to prevent livelock.
5617                  */
5618                 if (orphan) {
5619                         if (!ext4_should_journal_data(inode)) {
5620                                 inode_dio_wait(inode);
5621                         } else
5622                                 ext4_wait_for_tail_page_commit(inode);
5623                 }
5624                 down_write(&EXT4_I(inode)->i_mmap_sem);
5625 
5626                 rc = ext4_break_layouts(inode);
5627                 if (rc) {
5628                         up_write(&EXT4_I(inode)->i_mmap_sem);
5629                         error = rc;
5630                         goto err_out;
5631                 }
5632 
5633                 /*
5634                  * Truncate pagecache after we've waited for commit
5635                  * in data=journal mode to make pages freeable.
5636                  */
5637                 truncate_pagecache(inode, inode->i_size);
5638                 if (shrink) {
5639                         rc = ext4_truncate(inode);
5640                         if (rc)
5641                                 error = rc;
5642                 }
5643                 up_write(&EXT4_I(inode)->i_mmap_sem);
5644         }
5645 
5646         if (!error) {
5647                 setattr_copy(inode, attr);
5648                 mark_inode_dirty(inode);
5649         }
5650 
5651         /*
5652          * If the call to ext4_truncate failed to get a transaction handle at
5653          * all, we need to clean up the in-core orphan list manually.
5654          */
5655         if (orphan && inode->i_nlink)
5656                 ext4_orphan_del(NULL, inode);
5657 
5658         if (!error && (ia_valid & ATTR_MODE))
5659                 rc = posix_acl_chmod(inode, inode->i_mode);
5660 
5661 err_out:
5662         ext4_std_error(inode->i_sb, error);
5663         if (!error)
5664                 error = rc;
5665         return error;
5666 }
5667 
5668 int ext4_getattr(const struct path *path, struct kstat *stat,
5669                  u32 request_mask, unsigned int query_flags)
5670 {
5671         struct inode *inode = d_inode(path->dentry);
5672         struct ext4_inode *raw_inode;
5673         struct ext4_inode_info *ei = EXT4_I(inode);
5674         unsigned int flags;
5675 
5676         if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5677                 stat->result_mask |= STATX_BTIME;
5678                 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5679                 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5680         }
5681 
5682         flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5683         if (flags & EXT4_APPEND_FL)
5684                 stat->attributes |= STATX_ATTR_APPEND;
5685         if (flags & EXT4_COMPR_FL)
5686                 stat->attributes |= STATX_ATTR_COMPRESSED;
5687         if (flags & EXT4_ENCRYPT_FL)
5688                 stat->attributes |= STATX_ATTR_ENCRYPTED;
5689         if (flags & EXT4_IMMUTABLE_FL)
5690                 stat->attributes |= STATX_ATTR_IMMUTABLE;
5691         if (flags & EXT4_NODUMP_FL)
5692                 stat->attributes |= STATX_ATTR_NODUMP;
5693 
5694         stat->attributes_mask |= (STATX_ATTR_APPEND |
5695                                   STATX_ATTR_COMPRESSED |
5696                                   STATX_ATTR_ENCRYPTED |
5697                                   STATX_ATTR_IMMUTABLE |
5698                                   STATX_ATTR_NODUMP);
5699 
5700         generic_fillattr(inode, stat);
5701         return 0;
5702 }
5703 
5704 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5705                       u32 request_mask, unsigned int query_flags)
5706 {
5707         struct inode *inode = d_inode(path->dentry);
5708         u64 delalloc_blocks;
5709 
5710         ext4_getattr(path, stat, request_mask, query_flags);
5711 
5712         /*
5713          * If there is inline data in the inode, the inode will normally not
5714          * have data blocks allocated (it may have an external xattr block).
5715          * Report at least one sector for such files, so tools like tar, rsync,
5716          * others don't incorrectly think the file is completely sparse.
5717          */
5718         if (unlikely(ext4_has_inline_data(inode)))
5719                 stat->blocks += (stat->size + 511) >> 9;
5720 
5721         /*
5722          * We can't update i_blocks if the block allocation is delayed
5723          * otherwise in the case of system crash before the real block
5724          * allocation is done, we will have i_blocks inconsistent with
5725          * on-disk file blocks.
5726          * We always keep i_blocks updated together with real
5727          * allocation. But to not confuse with user, stat
5728          * will return the blocks that include the delayed allocation
5729          * blocks for this file.
5730          */
5731         delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5732                                    EXT4_I(inode)->i_reserved_data_blocks);
5733         stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5734         return 0;
5735 }
5736 
5737 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5738                                    int pextents)
5739 {
5740         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5741                 return ext4_ind_trans_blocks(inode, lblocks);
5742         return ext4_ext_index_trans_blocks(inode, pextents);
5743 }
5744 
5745 /*
5746  * Account for index blocks, block groups bitmaps and block group
5747  * descriptor blocks if modify datablocks and index blocks
5748  * worse case, the indexs blocks spread over different block groups
5749  *
5750  * If datablocks are discontiguous, they are possible to spread over
5751  * different block groups too. If they are contiguous, with flexbg,
5752  * they could still across block group boundary.
5753  *
5754  * Also account for superblock, inode, quota and xattr blocks
5755  */
5756 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5757                                   int pextents)
5758 {
5759         ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5760         int gdpblocks;
5761         int idxblocks;
5762         int ret = 0;
5763 
5764         /*
5765          * How many index blocks need to touch to map @lblocks logical blocks
5766          * to @pextents physical extents?
5767          */
5768         idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5769 
5770         ret = idxblocks;
5771 
5772         /*
5773          * Now let's see how many group bitmaps and group descriptors need
5774          * to account
5775          */
5776         groups = idxblocks + pextents;
5777         gdpblocks = groups;
5778         if (groups > ngroups)
5779                 groups = ngroups;
5780         if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5781                 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5782 
5783         /* bitmaps and block group descriptor blocks */
5784         ret += groups + gdpblocks;
5785 
5786         /* Blocks for super block, inode, quota and xattr blocks */
5787         ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5788 
5789         return ret;
5790 }
5791 
5792 /*
5793  * Calculate the total number of credits to reserve to fit
5794  * the modification of a single pages into a single transaction,
5795  * which may include multiple chunks of block allocations.
5796  *
5797  * This could be called via ext4_write_begin()
5798  *
5799  * We need to consider the worse case, when
5800  * one new block per extent.
5801  */
5802 int ext4_writepage_trans_blocks(struct inode *inode)
5803 {
5804         int bpp = ext4_journal_blocks_per_page(inode);
5805         int ret;
5806 
5807         ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5808 
5809         /* Account for data blocks for journalled mode */
5810         if (ext4_should_journal_data(inode))
5811                 ret += bpp;
5812         return ret;
5813 }
5814 
5815 /*
5816  * Calculate the journal credits for a chunk of data modification.
5817  *
5818  * This is called from DIO, fallocate or whoever calling
5819  * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5820  *
5821  * journal buffers for data blocks are not included here, as DIO
5822  * and fallocate do no need to journal data buffers.
5823  */
5824 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5825 {
5826         return ext4_meta_trans_blocks(inode, nrblocks, 1);
5827 }
5828 
5829 /*
5830  * The caller must have previously called ext4_reserve_inode_write().
5831  * Give this, we know that the caller already has write access to iloc->bh.
5832  */
5833 int ext4_mark_iloc_dirty(handle_t *handle,
5834                          struct inode *inode, struct ext4_iloc *iloc)
5835 {
5836         int err = 0;
5837 
5838         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5839                 put_bh(iloc->bh);
5840                 return -EIO;
5841         }
5842         if (IS_I_VERSION(inode))
5843                 inode_inc_iversion(inode);
5844 
5845         /* the do_update_inode consumes one bh->b_count */
5846         get_bh(iloc->bh);
5847 
5848         /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5849         err = ext4_do_update_inode(handle, inode, iloc);
5850         put_bh(iloc->bh);
5851         return err;
5852 }
5853 
5854 /*
5855  * On success, We end up with an outstanding reference count against
5856  * iloc->bh.  This _must_ be cleaned up later.
5857  */
5858 
5859 int
5860 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5861                          struct ext4_iloc *iloc)
5862 {
5863         int err;
5864 
5865         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5866                 return -EIO;
5867 
5868         err = ext4_get_inode_loc(inode, iloc);
5869         if (!err) {
5870                 BUFFER_TRACE(iloc->bh, "get_write_access");
5871                 err = ext4_journal_get_write_access(handle, iloc->bh);
5872                 if (err) {
5873                         brelse(iloc->bh);
5874                         iloc->bh = NULL;
5875                 }
5876         }
5877         ext4_std_error(inode->i_sb, err);
5878         return err;
5879 }
5880 
5881 static int __ext4_expand_extra_isize(struct inode *inode,
5882                                      unsigned int new_extra_isize,
5883                                      struct ext4_iloc *iloc,
5884                                      handle_t *handle, int *no_expand)
5885 {
5886         struct ext4_inode *raw_inode;
5887         struct ext4_xattr_ibody_header *header;
5888         int error;
5889 
5890         raw_inode = ext4_raw_inode(iloc);
5891 
5892         header = IHDR(inode, raw_inode);
5893 
5894         /* No extended attributes present */
5895         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5896             header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5897                 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5898                        EXT4_I(inode)->i_extra_isize, 0,
5899                        new_extra_isize - EXT4_I(inode)->i_extra_isize);
5900                 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5901                 return 0;
5902         }
5903 
5904         /* try to expand with EAs present */
5905         error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5906                                            raw_inode, handle);
5907         if (error) {
5908                 /*
5909                  * Inode size expansion failed; don't try again
5910                  */
5911                 *no_expand = 1;
5912         }
5913 
5914         return error;
5915 }
5916 
5917 /*
5918  * Expand an inode by new_extra_isize bytes.
5919  * Returns 0 on success or negative error number on failure.
5920  */
5921 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5922                                           unsigned int new_extra_isize,
5923                                           struct ext4_iloc iloc,
5924                                           handle_t *handle)
5925 {
5926         int no_expand;
5927         int error;
5928 
5929         if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5930                 return -EOVERFLOW;
5931 
5932         /*
5933          * In nojournal mode, we can immediately attempt to expand
5934          * the inode.  When journaled, we first need to obtain extra
5935          * buffer credits since we may write into the EA block
5936          * with this same handle. If journal_extend fails, then it will
5937          * only result in a minor loss of functionality for that inode.
5938          * If this is felt to be critical, then e2fsck should be run to
5939          * force a large enough s_min_extra_isize.
5940          */
5941         if (ext4_handle_valid(handle) &&
5942             jbd2_journal_extend(handle,
5943                                 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5944                 return -ENOSPC;
5945 
5946         if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5947                 return -EBUSY;
5948 
5949         error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5950                                           handle, &no_expand);
5951         ext4_write_unlock_xattr(inode, &no_expand);
5952 
5953         return error;
5954 }
5955 
5956 int ext4_expand_extra_isize(struct inode *inode,
5957                             unsigned int new_extra_isize,
5958                             struct ext4_iloc *iloc)
5959 {
5960         handle_t *handle;
5961         int no_expand;
5962         int error, rc;
5963 
5964         if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5965                 brelse(iloc->bh);
5966                 return -EOVERFLOW;
5967         }
5968 
5969         handle = ext4_journal_start(inode, EXT4_HT_INODE,
5970                                     EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5971         if (IS_ERR(handle)) {
5972                 error = PTR_ERR(handle);
5973                 brelse(iloc->bh);
5974                 return error;
5975         }
5976 
5977         ext4_write_lock_xattr(inode, &no_expand);
5978 
5979         BUFFER_TRACE(iloc.bh, "get_write_access");
5980         error = ext4_journal_get_write_access(handle, iloc->bh);
5981         if (error) {
5982                 brelse(iloc->bh);
5983                 goto out_stop;
5984         }
5985 
5986         error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5987                                           handle, &no_expand);
5988 
5989         rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5990         if (!error)
5991                 error = rc;
5992 
5993         ext4_write_unlock_xattr(inode, &no_expand);
5994 out_stop:
5995         ext4_journal_stop(handle);
5996         return error;
5997 }
5998 
5999 /*
6000  * What we do here is to mark the in-core inode as clean with respect to inode
6001  * dirtiness (it may still be data-dirty).
6002  * This means that the in-core inode may be reaped by prune_icache
6003  * without having to perform any I/O.  This is a very good thing,
6004  * because *any* task may call prune_icache - even ones which
6005  * have a transaction open against a different journal.
6006  *
6007  * Is this cheating?  Not really.  Sure, we haven't written the
6008  * inode out, but prune_icache isn't a user-visible syncing function.
6009  * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6010  * we start and wait on commits.
6011  */
6012 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
6013 {
6014         struct ext4_iloc iloc;
6015         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6016         int err;
6017 
6018         might_sleep();
6019         trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6020         err = ext4_reserve_inode_write(handle, inode, &iloc);
6021         if (err)
6022                 return err;
6023 
6024         if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6025                 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6026                                                iloc, handle);
6027 
6028         return ext4_mark_iloc_dirty(handle, inode, &iloc);
6029 }
6030 
6031 /*
6032  * ext4_dirty_inode() is called from __mark_inode_dirty()
6033  *
6034  * We're really interested in the case where a file is being extended.
6035  * i_size has been changed by generic_commit_write() and we thus need
6036  * to include the updated inode in the current transaction.
6037  *
6038  * Also, dquot_alloc_block() will always dirty the inode when blocks
6039  * are allocated to the file.
6040  *
6041  * If the inode is marked synchronous, we don't honour that here - doing
6042  * so would cause a commit on atime updates, which we don't bother doing.
6043  * We handle synchronous inodes at the highest possible level.
6044  *
6045  * If only the I_DIRTY_TIME flag is set, we can skip everything.  If
6046  * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6047  * to copy into the on-disk inode structure are the timestamp files.
6048  */
6049 void ext4_dirty_inode(struct inode *inode, int flags)
6050 {
6051         handle_t *handle;
6052 
6053         if (flags == I_DIRTY_TIME)
6054                 return;
6055         handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6056         if (IS_ERR(handle))
6057                 goto out;
6058 
6059         ext4_mark_inode_dirty(handle, inode);
6060 
6061         ext4_journal_stop(handle);
6062 out:
6063         return;
6064 }
6065 
6066 #if 0
6067 /*
6068  * Bind an inode's backing buffer_head into this transaction, to prevent
6069  * it from being flushed to disk early.  Unlike
6070  * ext4_reserve_inode_write, this leaves behind no bh reference and
6071  * returns no iloc structure, so the caller needs to repeat the iloc
6072  * lookup to mark the inode dirty later.
6073  */
6074 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
6075 {
6076         struct ext4_iloc iloc;
6077 
6078         int err = 0;
6079         if (handle) {
6080                 err = ext4_get_inode_loc(inode, &iloc);
6081                 if (!err) {
6082                         BUFFER_TRACE(iloc.bh, "get_write_access");
6083                         err = jbd2_journal_get_write_access(handle, iloc.bh);
6084                         if (!err)
6085                                 err = ext4_handle_dirty_metadata(handle,
6086                                                                  NULL,
6087                                                                  iloc.bh);
6088                         brelse(iloc.bh);
6089                 }
6090         }
6091         ext4_std_error(inode->i_sb, err);
6092         return err;
6093 }
6094 #endif
6095 
6096 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6097 {
6098         journal_t *journal;
6099         handle_t *handle;
6100         int err;
6101         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6102 
6103         /*
6104          * We have to be very careful here: changing a data block's
6105          * journaling status dynamically is dangerous.  If we write a
6106          * data block to the journal, change the status and then delete
6107          * that block, we risk forgetting to revoke the old log record
6108          * from the journal and so a subsequent replay can corrupt data.
6109          * So, first we make sure that the journal is empty and that
6110          * nobody is changing anything.
6111          */
6112 
6113         journal = EXT4_JOURNAL(inode);
6114         if (!journal)
6115                 return 0;
6116         if (is_journal_aborted(journal))
6117                 return -EROFS;
6118 
6119         /* Wait for all existing dio workers */
6120         inode_dio_wait(inode);
6121 
6122         /*
6123          * Before flushing the journal and switching inode's aops, we have
6124          * to flush all dirty data the inode has. There can be outstanding
6125          * delayed allocations, there can be unwritten extents created by
6126          * fallocate or buffered writes in dioread_nolock mode covered by
6127          * dirty data which can be converted only after flushing the dirty
6128          * data (and journalled aops don't know how to handle these cases).
6129          */
6130         if (val) {
6131                 down_write(&EXT4_I(inode)->i_mmap_sem);
6132                 err = filemap_write_and_wait(inode->i_mapping);
6133                 if (err < 0) {
6134                         up_write(&EXT4_I(inode)->i_mmap_sem);
6135                         return err;
6136                 }
6137         }
6138 
6139         percpu_down_write(&sbi->s_journal_flag_rwsem);
6140         jbd2_journal_lock_updates(journal);
6141 
6142         /*
6143          * OK, there are no updates running now, and all cached data is
6144          * synced to disk.  We are now in a completely consistent state
6145          * which doesn't have anything in the journal, and we know that
6146          * no filesystem updates are running, so it is safe to modify
6147          * the inode's in-core data-journaling state flag now.
6148          */
6149 
6150         if (val)
6151                 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6152         else {
6153                 err = jbd2_journal_flush(journal);
6154                 if (err < 0) {
6155                         jbd2_journal_unlock_updates(journal);
6156                         percpu_up_write(&sbi->s_journal_flag_rwsem);
6157                         return err;
6158                 }
6159                 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6160         }
6161         ext4_set_aops(inode);
6162 
6163         jbd2_journal_unlock_updates(journal);
6164         percpu_up_write(&sbi->s_journal_flag_rwsem);
6165 
6166         if (val)
6167                 up_write(&EXT4_I(inode)->i_mmap_sem);
6168 
6169         /* Finally we can mark the inode as dirty. */
6170 
6171         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6172         if (IS_ERR(handle))
6173                 return PTR_ERR(handle);
6174 
6175         err = ext4_mark_inode_dirty(handle, inode);
6176         ext4_handle_sync(handle);
6177         ext4_journal_stop(handle);
6178         ext4_std_error(inode->i_sb, err);
6179 
6180         return err;
6181 }
6182 
6183 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6184 {
6185         return !buffer_mapped(bh);
6186 }
6187 
6188 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6189 {
6190         struct vm_area_struct *vma = vmf->vma;
6191         struct page *page = vmf->page;
6192         loff_t size;
6193         unsigned long len;
6194         int err;
6195         vm_fault_t ret;
6196         struct file *file = vma->vm_file;
6197         struct inode *inode = file_inode(file);
6198         struct address_space *mapping = inode->i_mapping;
6199         handle_t *handle;
6200         get_block_t *get_block;
6201         int retries = 0;
6202 
6203         sb_start_pagefault(inode->i_sb);
6204         file_update_time(vma->vm_file);
6205 
6206         down_read(&EXT4_I(inode)->i_mmap_sem);
6207 
6208         err = ext4_convert_inline_data(inode);
6209         if (err)
6210                 goto out_ret;
6211 
6212         /* Delalloc case is easy... */
6213         if (test_opt(inode->i_sb, DELALLOC) &&
6214             !ext4_should_journal_data(inode) &&
6215             !ext4_nonda_switch(inode->i_sb)) {
6216                 do {
6217                         err = block_page_mkwrite(vma, vmf,
6218                                                    ext4_da_get_block_prep);
6219                 } while (err == -ENOSPC &&
6220                        ext4_should_retry_alloc(inode->i_sb, &retries));
6221                 goto out_ret;
6222         }
6223 
6224         lock_page(page);
6225         size = i_size_read(inode);
6226         /* Page got truncated from under us? */
6227         if (page->mapping != mapping || page_offset(page) > size) {
6228                 unlock_page(page);
6229                 ret = VM_FAULT_NOPAGE;
6230                 goto out;
6231         }
6232 
6233         if (page->index == size >> PAGE_SHIFT)
6234                 len = size & ~PAGE_MASK;
6235         else
6236                 len = PAGE_SIZE;
6237         /*
6238          * Return if we have all the buffers mapped. This avoids the need to do
6239          * journal_start/journal_stop which can block and take a long time
6240          */
6241         if (page_has_buffers(page)) {
6242                 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6243                                             0, len, NULL,
6244                                             ext4_bh_unmapped)) {
6245                         /* Wait so that we don't change page under IO */
6246                         wait_for_stable_page(page);
6247                         ret = VM_FAULT_LOCKED;
6248                         goto out;
6249                 }
6250         }
6251         unlock_page(page);
6252         /* OK, we need to fill the hole... */
6253         if (ext4_should_dioread_nolock(inode))
6254                 get_block = ext4_get_block_unwritten;
6255         else
6256                 get_block = ext4_get_block;
6257 retry_alloc:
6258         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6259                                     ext4_writepage_trans_blocks(inode));
6260         if (IS_ERR(handle)) {
6261                 ret = VM_FAULT_SIGBUS;
6262                 goto out;
6263         }
6264         err = block_page_mkwrite(vma, vmf, get_block);
6265         if (!err && ext4_should_journal_data(inode)) {
6266                 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6267                           PAGE_SIZE, NULL, do_journal_get_write_access)) {
6268                         unlock_page(page);
6269                         ret = VM_FAULT_SIGBUS;
6270                         ext4_journal_stop(handle);
6271                         goto out;
6272                 }
6273                 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6274         }
6275         ext4_journal_stop(handle);
6276         if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6277                 goto retry_alloc;
6278 out_ret:
6279         ret = block_page_mkwrite_return(err);
6280 out:
6281         up_read(&EXT4_I(inode)->i_mmap_sem);
6282         sb_end_pagefault(inode->i_sb);
6283         return ret;
6284 }
6285 
6286 vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6287 {
6288         struct inode *inode = file_inode(vmf->vma->vm_file);
6289         vm_fault_t ret;
6290 
6291         down_read(&EXT4_I(inode)->i_mmap_sem);
6292         ret = filemap_fault(vmf);
6293         up_read(&EXT4_I(inode)->i_mmap_sem);
6294 
6295         return ret;
6296 }
6297 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp