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

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

Version: ~ [ linux-5.15-rc5 ] ~ [ linux-5.14.11 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.72 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.152 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.210 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.250 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.286 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.288 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ 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             !inode_is_open_for_write(inode))
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_has_feature_journal(inode->i_sb) &&
403             (inode->i_ino ==
404              le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
405                 return 0;
406         if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
407                                    map->m_len)) {
408                 ext4_error_inode(inode, func, line, map->m_pblk,
409                                  "lblock %lu mapped to illegal pblock %llu "
410                                  "(length %d)", (unsigned long) map->m_lblk,
411                                  map->m_pblk, map->m_len);
412                 return -EFSCORRUPTED;
413         }
414         return 0;
415 }
416 
417 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
418                        ext4_lblk_t len)
419 {
420         int ret;
421 
422         if (IS_ENCRYPTED(inode))
423                 return fscrypt_zeroout_range(inode, lblk, pblk, len);
424 
425         ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
426         if (ret > 0)
427                 ret = 0;
428 
429         return ret;
430 }
431 
432 #define check_block_validity(inode, map)        \
433         __check_block_validity((inode), __func__, __LINE__, (map))
434 
435 #ifdef ES_AGGRESSIVE_TEST
436 static void ext4_map_blocks_es_recheck(handle_t *handle,
437                                        struct inode *inode,
438                                        struct ext4_map_blocks *es_map,
439                                        struct ext4_map_blocks *map,
440                                        int flags)
441 {
442         int retval;
443 
444         map->m_flags = 0;
445         /*
446          * There is a race window that the result is not the same.
447          * e.g. xfstests #223 when dioread_nolock enables.  The reason
448          * is that we lookup a block mapping in extent status tree with
449          * out taking i_data_sem.  So at the time the unwritten extent
450          * could be converted.
451          */
452         down_read(&EXT4_I(inode)->i_data_sem);
453         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
454                 retval = ext4_ext_map_blocks(handle, inode, map, flags &
455                                              EXT4_GET_BLOCKS_KEEP_SIZE);
456         } else {
457                 retval = ext4_ind_map_blocks(handle, inode, map, flags &
458                                              EXT4_GET_BLOCKS_KEEP_SIZE);
459         }
460         up_read((&EXT4_I(inode)->i_data_sem));
461 
462         /*
463          * We don't check m_len because extent will be collpased in status
464          * tree.  So the m_len might not equal.
465          */
466         if (es_map->m_lblk != map->m_lblk ||
467             es_map->m_flags != map->m_flags ||
468             es_map->m_pblk != map->m_pblk) {
469                 printk("ES cache assertion failed for inode: %lu "
470                        "es_cached ex [%d/%d/%llu/%x] != "
471                        "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
472                        inode->i_ino, es_map->m_lblk, es_map->m_len,
473                        es_map->m_pblk, es_map->m_flags, map->m_lblk,
474                        map->m_len, map->m_pblk, map->m_flags,
475                        retval, flags);
476         }
477 }
478 #endif /* ES_AGGRESSIVE_TEST */
479 
480 /*
481  * The ext4_map_blocks() function tries to look up the requested blocks,
482  * and returns if the blocks are already mapped.
483  *
484  * Otherwise it takes the write lock of the i_data_sem and allocate blocks
485  * and store the allocated blocks in the result buffer head and mark it
486  * mapped.
487  *
488  * If file type is extents based, it will call ext4_ext_map_blocks(),
489  * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
490  * based files
491  *
492  * On success, it returns the number of blocks being mapped or allocated.  if
493  * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
494  * is marked as unwritten. If the create == 1, it will mark @map as mapped.
495  *
496  * It returns 0 if plain look up failed (blocks have not been allocated), in
497  * that case, @map is returned as unmapped but we still do fill map->m_len to
498  * indicate the length of a hole starting at map->m_lblk.
499  *
500  * It returns the error in case of allocation failure.
501  */
502 int ext4_map_blocks(handle_t *handle, struct inode *inode,
503                     struct ext4_map_blocks *map, int flags)
504 {
505         struct extent_status es;
506         int retval;
507         int ret = 0;
508 #ifdef ES_AGGRESSIVE_TEST
509         struct ext4_map_blocks orig_map;
510 
511         memcpy(&orig_map, map, sizeof(*map));
512 #endif
513 
514         map->m_flags = 0;
515         ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
516                   "logical block %lu\n", inode->i_ino, flags, map->m_len,
517                   (unsigned long) map->m_lblk);
518 
519         /*
520          * ext4_map_blocks returns an int, and m_len is an unsigned int
521          */
522         if (unlikely(map->m_len > INT_MAX))
523                 map->m_len = INT_MAX;
524 
525         /* We can handle the block number less than EXT_MAX_BLOCKS */
526         if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
527                 return -EFSCORRUPTED;
528 
529         /* Lookup extent status tree firstly */
530         if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
531                 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
532                         map->m_pblk = ext4_es_pblock(&es) +
533                                         map->m_lblk - es.es_lblk;
534                         map->m_flags |= ext4_es_is_written(&es) ?
535                                         EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
536                         retval = es.es_len - (map->m_lblk - es.es_lblk);
537                         if (retval > map->m_len)
538                                 retval = map->m_len;
539                         map->m_len = retval;
540                 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
541                         map->m_pblk = 0;
542                         retval = es.es_len - (map->m_lblk - es.es_lblk);
543                         if (retval > map->m_len)
544                                 retval = map->m_len;
545                         map->m_len = retval;
546                         retval = 0;
547                 } else {
548                         BUG();
549                 }
550 #ifdef ES_AGGRESSIVE_TEST
551                 ext4_map_blocks_es_recheck(handle, inode, map,
552                                            &orig_map, flags);
553 #endif
554                 goto found;
555         }
556 
557         /*
558          * Try to see if we can get the block without requesting a new
559          * file system block.
560          */
561         down_read(&EXT4_I(inode)->i_data_sem);
562         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
563                 retval = ext4_ext_map_blocks(handle, inode, map, flags &
564                                              EXT4_GET_BLOCKS_KEEP_SIZE);
565         } else {
566                 retval = ext4_ind_map_blocks(handle, inode, map, flags &
567                                              EXT4_GET_BLOCKS_KEEP_SIZE);
568         }
569         if (retval > 0) {
570                 unsigned int status;
571 
572                 if (unlikely(retval != map->m_len)) {
573                         ext4_warning(inode->i_sb,
574                                      "ES len assertion failed for inode "
575                                      "%lu: retval %d != map->m_len %d",
576                                      inode->i_ino, retval, map->m_len);
577                         WARN_ON(1);
578                 }
579 
580                 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
581                                 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
582                 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
583                     !(status & EXTENT_STATUS_WRITTEN) &&
584                     ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
585                                        map->m_lblk + map->m_len - 1))
586                         status |= EXTENT_STATUS_DELAYED;
587                 ret = ext4_es_insert_extent(inode, map->m_lblk,
588                                             map->m_len, map->m_pblk, status);
589                 if (ret < 0)
590                         retval = ret;
591         }
592         up_read((&EXT4_I(inode)->i_data_sem));
593 
594 found:
595         if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
596                 ret = check_block_validity(inode, map);
597                 if (ret != 0)
598                         return ret;
599         }
600 
601         /* If it is only a block(s) look up */
602         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
603                 return retval;
604 
605         /*
606          * Returns if the blocks have already allocated
607          *
608          * Note that if blocks have been preallocated
609          * ext4_ext_get_block() returns the create = 0
610          * with buffer head unmapped.
611          */
612         if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
613                 /*
614                  * If we need to convert extent to unwritten
615                  * we continue and do the actual work in
616                  * ext4_ext_map_blocks()
617                  */
618                 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
619                         return retval;
620 
621         /*
622          * Here we clear m_flags because after allocating an new extent,
623          * it will be set again.
624          */
625         map->m_flags &= ~EXT4_MAP_FLAGS;
626 
627         /*
628          * New blocks allocate and/or writing to unwritten extent
629          * will possibly result in updating i_data, so we take
630          * the write lock of i_data_sem, and call get_block()
631          * with create == 1 flag.
632          */
633         down_write(&EXT4_I(inode)->i_data_sem);
634 
635         /*
636          * We need to check for EXT4 here because migrate
637          * could have changed the inode type in between
638          */
639         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
640                 retval = ext4_ext_map_blocks(handle, inode, map, flags);
641         } else {
642                 retval = ext4_ind_map_blocks(handle, inode, map, flags);
643 
644                 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
645                         /*
646                          * We allocated new blocks which will result in
647                          * i_data's format changing.  Force the migrate
648                          * to fail by clearing migrate flags
649                          */
650                         ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
651                 }
652 
653                 /*
654                  * Update reserved blocks/metadata blocks after successful
655                  * block allocation which had been deferred till now. We don't
656                  * support fallocate for non extent files. So we can update
657                  * reserve space here.
658                  */
659                 if ((retval > 0) &&
660                         (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
661                         ext4_da_update_reserve_space(inode, retval, 1);
662         }
663 
664         if (retval > 0) {
665                 unsigned int status;
666 
667                 if (unlikely(retval != map->m_len)) {
668                         ext4_warning(inode->i_sb,
669                                      "ES len assertion failed for inode "
670                                      "%lu: retval %d != map->m_len %d",
671                                      inode->i_ino, retval, map->m_len);
672                         WARN_ON(1);
673                 }
674 
675                 /*
676                  * We have to zeroout blocks before inserting them into extent
677                  * status tree. Otherwise someone could look them up there and
678                  * use them before they are really zeroed. We also have to
679                  * unmap metadata before zeroing as otherwise writeback can
680                  * overwrite zeros with stale data from block device.
681                  */
682                 if (flags & EXT4_GET_BLOCKS_ZERO &&
683                     map->m_flags & EXT4_MAP_MAPPED &&
684                     map->m_flags & EXT4_MAP_NEW) {
685                         ret = ext4_issue_zeroout(inode, map->m_lblk,
686                                                  map->m_pblk, map->m_len);
687                         if (ret) {
688                                 retval = ret;
689                                 goto out_sem;
690                         }
691                 }
692 
693                 /*
694                  * If the extent has been zeroed out, we don't need to update
695                  * extent status tree.
696                  */
697                 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
698                     ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
699                         if (ext4_es_is_written(&es))
700                                 goto out_sem;
701                 }
702                 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
703                                 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
704                 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
705                     !(status & EXTENT_STATUS_WRITTEN) &&
706                     ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
707                                        map->m_lblk + map->m_len - 1))
708                         status |= EXTENT_STATUS_DELAYED;
709                 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
710                                             map->m_pblk, status);
711                 if (ret < 0) {
712                         retval = ret;
713                         goto out_sem;
714                 }
715         }
716 
717 out_sem:
718         up_write((&EXT4_I(inode)->i_data_sem));
719         if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
720                 ret = check_block_validity(inode, map);
721                 if (ret != 0)
722                         return ret;
723 
724                 /*
725                  * Inodes with freshly allocated blocks where contents will be
726                  * visible after transaction commit must be on transaction's
727                  * ordered data list.
728                  */
729                 if (map->m_flags & EXT4_MAP_NEW &&
730                     !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
731                     !(flags & EXT4_GET_BLOCKS_ZERO) &&
732                     !ext4_is_quota_file(inode) &&
733                     ext4_should_order_data(inode)) {
734                         loff_t start_byte =
735                                 (loff_t)map->m_lblk << inode->i_blkbits;
736                         loff_t length = (loff_t)map->m_len << inode->i_blkbits;
737 
738                         if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
739                                 ret = ext4_jbd2_inode_add_wait(handle, inode,
740                                                 start_byte, length);
741                         else
742                                 ret = ext4_jbd2_inode_add_write(handle, inode,
743                                                 start_byte, length);
744                         if (ret)
745                                 return ret;
746                 }
747         }
748         return retval;
749 }
750 
751 /*
752  * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
753  * we have to be careful as someone else may be manipulating b_state as well.
754  */
755 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
756 {
757         unsigned long old_state;
758         unsigned long new_state;
759 
760         flags &= EXT4_MAP_FLAGS;
761 
762         /* Dummy buffer_head? Set non-atomically. */
763         if (!bh->b_page) {
764                 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
765                 return;
766         }
767         /*
768          * Someone else may be modifying b_state. Be careful! This is ugly but
769          * once we get rid of using bh as a container for mapping information
770          * to pass to / from get_block functions, this can go away.
771          */
772         do {
773                 old_state = READ_ONCE(bh->b_state);
774                 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
775         } while (unlikely(
776                  cmpxchg(&bh->b_state, old_state, new_state) != old_state));
777 }
778 
779 static int _ext4_get_block(struct inode *inode, sector_t iblock,
780                            struct buffer_head *bh, int flags)
781 {
782         struct ext4_map_blocks map;
783         int ret = 0;
784 
785         if (ext4_has_inline_data(inode))
786                 return -ERANGE;
787 
788         map.m_lblk = iblock;
789         map.m_len = bh->b_size >> inode->i_blkbits;
790 
791         ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
792                               flags);
793         if (ret > 0) {
794                 map_bh(bh, inode->i_sb, map.m_pblk);
795                 ext4_update_bh_state(bh, map.m_flags);
796                 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
797                 ret = 0;
798         } else if (ret == 0) {
799                 /* hole case, need to fill in bh->b_size */
800                 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
801         }
802         return ret;
803 }
804 
805 int ext4_get_block(struct inode *inode, sector_t iblock,
806                    struct buffer_head *bh, int create)
807 {
808         return _ext4_get_block(inode, iblock, bh,
809                                create ? EXT4_GET_BLOCKS_CREATE : 0);
810 }
811 
812 /*
813  * Get block function used when preparing for buffered write if we require
814  * creating an unwritten extent if blocks haven't been allocated.  The extent
815  * will be converted to written after the IO is complete.
816  */
817 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
818                              struct buffer_head *bh_result, int create)
819 {
820         ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
821                    inode->i_ino, create);
822         return _ext4_get_block(inode, iblock, bh_result,
823                                EXT4_GET_BLOCKS_IO_CREATE_EXT);
824 }
825 
826 /* Maximum number of blocks we map for direct IO at once. */
827 #define DIO_MAX_BLOCKS 4096
828 
829 /*
830  * Get blocks function for the cases that need to start a transaction -
831  * generally difference cases of direct IO and DAX IO. It also handles retries
832  * in case of ENOSPC.
833  */
834 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
835                                 struct buffer_head *bh_result, int flags)
836 {
837         int dio_credits;
838         handle_t *handle;
839         int retries = 0;
840         int ret;
841 
842         /* Trim mapping request to maximum we can map at once for DIO */
843         if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
844                 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
845         dio_credits = ext4_chunk_trans_blocks(inode,
846                                       bh_result->b_size >> inode->i_blkbits);
847 retry:
848         handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
849         if (IS_ERR(handle))
850                 return PTR_ERR(handle);
851 
852         ret = _ext4_get_block(inode, iblock, bh_result, flags);
853         ext4_journal_stop(handle);
854 
855         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
856                 goto retry;
857         return ret;
858 }
859 
860 /* Get block function for DIO reads and writes to inodes without extents */
861 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
862                        struct buffer_head *bh, int create)
863 {
864         /* We don't expect handle for direct IO */
865         WARN_ON_ONCE(ext4_journal_current_handle());
866 
867         if (!create)
868                 return _ext4_get_block(inode, iblock, bh, 0);
869         return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
870 }
871 
872 /*
873  * Get block function for AIO DIO writes when we create unwritten extent if
874  * blocks are not allocated yet. The extent will be converted to written
875  * after IO is complete.
876  */
877 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
878                 sector_t iblock, struct buffer_head *bh_result, int create)
879 {
880         int ret;
881 
882         /* We don't expect handle for direct IO */
883         WARN_ON_ONCE(ext4_journal_current_handle());
884 
885         ret = ext4_get_block_trans(inode, iblock, bh_result,
886                                    EXT4_GET_BLOCKS_IO_CREATE_EXT);
887 
888         /*
889          * When doing DIO using unwritten extents, we need io_end to convert
890          * unwritten extents to written on IO completion. We allocate io_end
891          * once we spot unwritten extent and store it in b_private. Generic
892          * DIO code keeps b_private set and furthermore passes the value to
893          * our completion callback in 'private' argument.
894          */
895         if (!ret && buffer_unwritten(bh_result)) {
896                 if (!bh_result->b_private) {
897                         ext4_io_end_t *io_end;
898 
899                         io_end = ext4_init_io_end(inode, GFP_KERNEL);
900                         if (!io_end)
901                                 return -ENOMEM;
902                         bh_result->b_private = io_end;
903                         ext4_set_io_unwritten_flag(inode, io_end);
904                 }
905                 set_buffer_defer_completion(bh_result);
906         }
907 
908         return ret;
909 }
910 
911 /*
912  * Get block function for non-AIO DIO writes when we create unwritten extent if
913  * blocks are not allocated yet. The extent will be converted to written
914  * after IO is complete by ext4_direct_IO_write().
915  */
916 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
917                 sector_t iblock, struct buffer_head *bh_result, int create)
918 {
919         int ret;
920 
921         /* We don't expect handle for direct IO */
922         WARN_ON_ONCE(ext4_journal_current_handle());
923 
924         ret = ext4_get_block_trans(inode, iblock, bh_result,
925                                    EXT4_GET_BLOCKS_IO_CREATE_EXT);
926 
927         /*
928          * Mark inode as having pending DIO writes to unwritten extents.
929          * ext4_direct_IO_write() checks this flag and converts extents to
930          * written.
931          */
932         if (!ret && buffer_unwritten(bh_result))
933                 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
934 
935         return ret;
936 }
937 
938 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
939                    struct buffer_head *bh_result, int create)
940 {
941         int ret;
942 
943         ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
944                    inode->i_ino, create);
945         /* We don't expect handle for direct IO */
946         WARN_ON_ONCE(ext4_journal_current_handle());
947 
948         ret = _ext4_get_block(inode, iblock, bh_result, 0);
949         /*
950          * Blocks should have been preallocated! ext4_file_write_iter() checks
951          * that.
952          */
953         WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
954 
955         return ret;
956 }
957 
958 
959 /*
960  * `handle' can be NULL if create is zero
961  */
962 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
963                                 ext4_lblk_t block, int map_flags)
964 {
965         struct ext4_map_blocks map;
966         struct buffer_head *bh;
967         int create = map_flags & EXT4_GET_BLOCKS_CREATE;
968         int err;
969 
970         J_ASSERT(handle != NULL || create == 0);
971 
972         map.m_lblk = block;
973         map.m_len = 1;
974         err = ext4_map_blocks(handle, inode, &map, map_flags);
975 
976         if (err == 0)
977                 return create ? ERR_PTR(-ENOSPC) : NULL;
978         if (err < 0)
979                 return ERR_PTR(err);
980 
981         bh = sb_getblk(inode->i_sb, map.m_pblk);
982         if (unlikely(!bh))
983                 return ERR_PTR(-ENOMEM);
984         if (map.m_flags & EXT4_MAP_NEW) {
985                 J_ASSERT(create != 0);
986                 J_ASSERT(handle != NULL);
987 
988                 /*
989                  * Now that we do not always journal data, we should
990                  * keep in mind whether this should always journal the
991                  * new buffer as metadata.  For now, regular file
992                  * writes use ext4_get_block instead, so it's not a
993                  * problem.
994                  */
995                 lock_buffer(bh);
996                 BUFFER_TRACE(bh, "call get_create_access");
997                 err = ext4_journal_get_create_access(handle, bh);
998                 if (unlikely(err)) {
999                         unlock_buffer(bh);
1000                         goto errout;
1001                 }
1002                 if (!buffer_uptodate(bh)) {
1003                         memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1004                         set_buffer_uptodate(bh);
1005                 }
1006                 unlock_buffer(bh);
1007                 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1008                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1009                 if (unlikely(err))
1010                         goto errout;
1011         } else
1012                 BUFFER_TRACE(bh, "not a new buffer");
1013         return bh;
1014 errout:
1015         brelse(bh);
1016         return ERR_PTR(err);
1017 }
1018 
1019 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1020                                ext4_lblk_t block, int map_flags)
1021 {
1022         struct buffer_head *bh;
1023 
1024         bh = ext4_getblk(handle, inode, block, map_flags);
1025         if (IS_ERR(bh))
1026                 return bh;
1027         if (!bh || buffer_uptodate(bh))
1028                 return bh;
1029         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1030         wait_on_buffer(bh);
1031         if (buffer_uptodate(bh))
1032                 return bh;
1033         put_bh(bh);
1034         return ERR_PTR(-EIO);
1035 }
1036 
1037 /* Read a contiguous batch of blocks. */
1038 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1039                      bool wait, struct buffer_head **bhs)
1040 {
1041         int i, err;
1042 
1043         for (i = 0; i < bh_count; i++) {
1044                 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1045                 if (IS_ERR(bhs[i])) {
1046                         err = PTR_ERR(bhs[i]);
1047                         bh_count = i;
1048                         goto out_brelse;
1049                 }
1050         }
1051 
1052         for (i = 0; i < bh_count; i++)
1053                 /* Note that NULL bhs[i] is valid because of holes. */
1054                 if (bhs[i] && !buffer_uptodate(bhs[i]))
1055                         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1056                                     &bhs[i]);
1057 
1058         if (!wait)
1059                 return 0;
1060 
1061         for (i = 0; i < bh_count; i++)
1062                 if (bhs[i])
1063                         wait_on_buffer(bhs[i]);
1064 
1065         for (i = 0; i < bh_count; i++) {
1066                 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1067                         err = -EIO;
1068                         goto out_brelse;
1069                 }
1070         }
1071         return 0;
1072 
1073 out_brelse:
1074         for (i = 0; i < bh_count; i++) {
1075                 brelse(bhs[i]);
1076                 bhs[i] = NULL;
1077         }
1078         return err;
1079 }
1080 
1081 int ext4_walk_page_buffers(handle_t *handle,
1082                            struct buffer_head *head,
1083                            unsigned from,
1084                            unsigned to,
1085                            int *partial,
1086                            int (*fn)(handle_t *handle,
1087                                      struct buffer_head *bh))
1088 {
1089         struct buffer_head *bh;
1090         unsigned block_start, block_end;
1091         unsigned blocksize = head->b_size;
1092         int err, ret = 0;
1093         struct buffer_head *next;
1094 
1095         for (bh = head, block_start = 0;
1096              ret == 0 && (bh != head || !block_start);
1097              block_start = block_end, bh = next) {
1098                 next = bh->b_this_page;
1099                 block_end = block_start + blocksize;
1100                 if (block_end <= from || block_start >= to) {
1101                         if (partial && !buffer_uptodate(bh))
1102                                 *partial = 1;
1103                         continue;
1104                 }
1105                 err = (*fn)(handle, bh);
1106                 if (!ret)
1107                         ret = err;
1108         }
1109         return ret;
1110 }
1111 
1112 /*
1113  * To preserve ordering, it is essential that the hole instantiation and
1114  * the data write be encapsulated in a single transaction.  We cannot
1115  * close off a transaction and start a new one between the ext4_get_block()
1116  * and the commit_write().  So doing the jbd2_journal_start at the start of
1117  * prepare_write() is the right place.
1118  *
1119  * Also, this function can nest inside ext4_writepage().  In that case, we
1120  * *know* that ext4_writepage() has generated enough buffer credits to do the
1121  * whole page.  So we won't block on the journal in that case, which is good,
1122  * because the caller may be PF_MEMALLOC.
1123  *
1124  * By accident, ext4 can be reentered when a transaction is open via
1125  * quota file writes.  If we were to commit the transaction while thus
1126  * reentered, there can be a deadlock - we would be holding a quota
1127  * lock, and the commit would never complete if another thread had a
1128  * transaction open and was blocking on the quota lock - a ranking
1129  * violation.
1130  *
1131  * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1132  * will _not_ run commit under these circumstances because handle->h_ref
1133  * is elevated.  We'll still have enough credits for the tiny quotafile
1134  * write.
1135  */
1136 int do_journal_get_write_access(handle_t *handle,
1137                                 struct buffer_head *bh)
1138 {
1139         int dirty = buffer_dirty(bh);
1140         int ret;
1141 
1142         if (!buffer_mapped(bh) || buffer_freed(bh))
1143                 return 0;
1144         /*
1145          * __block_write_begin() could have dirtied some buffers. Clean
1146          * the dirty bit as jbd2_journal_get_write_access() could complain
1147          * otherwise about fs integrity issues. Setting of the dirty bit
1148          * by __block_write_begin() isn't a real problem here as we clear
1149          * the bit before releasing a page lock and thus writeback cannot
1150          * ever write the buffer.
1151          */
1152         if (dirty)
1153                 clear_buffer_dirty(bh);
1154         BUFFER_TRACE(bh, "get write access");
1155         ret = ext4_journal_get_write_access(handle, bh);
1156         if (!ret && dirty)
1157                 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1158         return ret;
1159 }
1160 
1161 #ifdef CONFIG_FS_ENCRYPTION
1162 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1163                                   get_block_t *get_block)
1164 {
1165         unsigned from = pos & (PAGE_SIZE - 1);
1166         unsigned to = from + len;
1167         struct inode *inode = page->mapping->host;
1168         unsigned block_start, block_end;
1169         sector_t block;
1170         int err = 0;
1171         unsigned blocksize = inode->i_sb->s_blocksize;
1172         unsigned bbits;
1173         struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1174         bool decrypt = false;
1175 
1176         BUG_ON(!PageLocked(page));
1177         BUG_ON(from > PAGE_SIZE);
1178         BUG_ON(to > PAGE_SIZE);
1179         BUG_ON(from > to);
1180 
1181         if (!page_has_buffers(page))
1182                 create_empty_buffers(page, blocksize, 0);
1183         head = page_buffers(page);
1184         bbits = ilog2(blocksize);
1185         block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1186 
1187         for (bh = head, block_start = 0; bh != head || !block_start;
1188             block++, block_start = block_end, bh = bh->b_this_page) {
1189                 block_end = block_start + blocksize;
1190                 if (block_end <= from || block_start >= to) {
1191                         if (PageUptodate(page)) {
1192                                 if (!buffer_uptodate(bh))
1193                                         set_buffer_uptodate(bh);
1194                         }
1195                         continue;
1196                 }
1197                 if (buffer_new(bh))
1198                         clear_buffer_new(bh);
1199                 if (!buffer_mapped(bh)) {
1200                         WARN_ON(bh->b_size != blocksize);
1201                         err = get_block(inode, block, bh, 1);
1202                         if (err)
1203                                 break;
1204                         if (buffer_new(bh)) {
1205                                 if (PageUptodate(page)) {
1206                                         clear_buffer_new(bh);
1207                                         set_buffer_uptodate(bh);
1208                                         mark_buffer_dirty(bh);
1209                                         continue;
1210                                 }
1211                                 if (block_end > to || block_start < from)
1212                                         zero_user_segments(page, to, block_end,
1213                                                            block_start, from);
1214                                 continue;
1215                         }
1216                 }
1217                 if (PageUptodate(page)) {
1218                         if (!buffer_uptodate(bh))
1219                                 set_buffer_uptodate(bh);
1220                         continue;
1221                 }
1222                 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1223                     !buffer_unwritten(bh) &&
1224                     (block_start < from || block_end > to)) {
1225                         ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1226                         *wait_bh++ = bh;
1227                         decrypt = IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode);
1228                 }
1229         }
1230         /*
1231          * If we issued read requests, let them complete.
1232          */
1233         while (wait_bh > wait) {
1234                 wait_on_buffer(*--wait_bh);
1235                 if (!buffer_uptodate(*wait_bh))
1236                         err = -EIO;
1237         }
1238         if (unlikely(err))
1239                 page_zero_new_buffers(page, from, to);
1240         else if (decrypt)
1241                 err = fscrypt_decrypt_page(page->mapping->host, page,
1242                                 PAGE_SIZE, 0, page->index);
1243         return err;
1244 }
1245 #endif
1246 
1247 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1248                             loff_t pos, unsigned len, unsigned flags,
1249                             struct page **pagep, void **fsdata)
1250 {
1251         struct inode *inode = mapping->host;
1252         int ret, needed_blocks;
1253         handle_t *handle;
1254         int retries = 0;
1255         struct page *page;
1256         pgoff_t index;
1257         unsigned from, to;
1258 
1259         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1260                 return -EIO;
1261 
1262         trace_ext4_write_begin(inode, pos, len, flags);
1263         /*
1264          * Reserve one block more for addition to orphan list in case
1265          * we allocate blocks but write fails for some reason
1266          */
1267         needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1268         index = pos >> PAGE_SHIFT;
1269         from = pos & (PAGE_SIZE - 1);
1270         to = from + len;
1271 
1272         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1273                 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1274                                                     flags, pagep);
1275                 if (ret < 0)
1276                         return ret;
1277                 if (ret == 1)
1278                         return 0;
1279         }
1280 
1281         /*
1282          * grab_cache_page_write_begin() can take a long time if the
1283          * system is thrashing due to memory pressure, or if the page
1284          * is being written back.  So grab it first before we start
1285          * the transaction handle.  This also allows us to allocate
1286          * the page (if needed) without using GFP_NOFS.
1287          */
1288 retry_grab:
1289         page = grab_cache_page_write_begin(mapping, index, flags);
1290         if (!page)
1291                 return -ENOMEM;
1292         unlock_page(page);
1293 
1294 retry_journal:
1295         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1296         if (IS_ERR(handle)) {
1297                 put_page(page);
1298                 return PTR_ERR(handle);
1299         }
1300 
1301         lock_page(page);
1302         if (page->mapping != mapping) {
1303                 /* The page got truncated from under us */
1304                 unlock_page(page);
1305                 put_page(page);
1306                 ext4_journal_stop(handle);
1307                 goto retry_grab;
1308         }
1309         /* In case writeback began while the page was unlocked */
1310         wait_for_stable_page(page);
1311 
1312 #ifdef CONFIG_FS_ENCRYPTION
1313         if (ext4_should_dioread_nolock(inode))
1314                 ret = ext4_block_write_begin(page, pos, len,
1315                                              ext4_get_block_unwritten);
1316         else
1317                 ret = ext4_block_write_begin(page, pos, len,
1318                                              ext4_get_block);
1319 #else
1320         if (ext4_should_dioread_nolock(inode))
1321                 ret = __block_write_begin(page, pos, len,
1322                                           ext4_get_block_unwritten);
1323         else
1324                 ret = __block_write_begin(page, pos, len, ext4_get_block);
1325 #endif
1326         if (!ret && ext4_should_journal_data(inode)) {
1327                 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1328                                              from, to, NULL,
1329                                              do_journal_get_write_access);
1330         }
1331 
1332         if (ret) {
1333                 unlock_page(page);
1334                 /*
1335                  * __block_write_begin may have instantiated a few blocks
1336                  * outside i_size.  Trim these off again. Don't need
1337                  * i_size_read because we hold i_mutex.
1338                  *
1339                  * Add inode to orphan list in case we crash before
1340                  * truncate finishes
1341                  */
1342                 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1343                         ext4_orphan_add(handle, inode);
1344 
1345                 ext4_journal_stop(handle);
1346                 if (pos + len > inode->i_size) {
1347                         ext4_truncate_failed_write(inode);
1348                         /*
1349                          * If truncate failed early the inode might
1350                          * still be on the orphan list; we need to
1351                          * make sure the inode is removed from the
1352                          * orphan list in that case.
1353                          */
1354                         if (inode->i_nlink)
1355                                 ext4_orphan_del(NULL, inode);
1356                 }
1357 
1358                 if (ret == -ENOSPC &&
1359                     ext4_should_retry_alloc(inode->i_sb, &retries))
1360                         goto retry_journal;
1361                 put_page(page);
1362                 return ret;
1363         }
1364         *pagep = page;
1365         return ret;
1366 }
1367 
1368 /* For write_end() in data=journal mode */
1369 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1370 {
1371         int ret;
1372         if (!buffer_mapped(bh) || buffer_freed(bh))
1373                 return 0;
1374         set_buffer_uptodate(bh);
1375         ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1376         clear_buffer_meta(bh);
1377         clear_buffer_prio(bh);
1378         return ret;
1379 }
1380 
1381 /*
1382  * We need to pick up the new inode size which generic_commit_write gave us
1383  * `file' can be NULL - eg, when called from page_symlink().
1384  *
1385  * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1386  * buffers are managed internally.
1387  */
1388 static int ext4_write_end(struct file *file,
1389                           struct address_space *mapping,
1390                           loff_t pos, unsigned len, unsigned copied,
1391                           struct page *page, void *fsdata)
1392 {
1393         handle_t *handle = ext4_journal_current_handle();
1394         struct inode *inode = mapping->host;
1395         loff_t old_size = inode->i_size;
1396         int ret = 0, ret2;
1397         int i_size_changed = 0;
1398         int inline_data = ext4_has_inline_data(inode);
1399 
1400         trace_ext4_write_end(inode, pos, len, copied);
1401         if (inline_data) {
1402                 ret = ext4_write_inline_data_end(inode, pos, len,
1403                                                  copied, page);
1404                 if (ret < 0) {
1405                         unlock_page(page);
1406                         put_page(page);
1407                         goto errout;
1408                 }
1409                 copied = ret;
1410         } else
1411                 copied = block_write_end(file, mapping, pos,
1412                                          len, copied, page, fsdata);
1413         /*
1414          * it's important to update i_size while still holding page lock:
1415          * page writeout could otherwise come in and zero beyond i_size.
1416          */
1417         i_size_changed = ext4_update_inode_size(inode, pos + copied);
1418         unlock_page(page);
1419         put_page(page);
1420 
1421         if (old_size < pos)
1422                 pagecache_isize_extended(inode, old_size, pos);
1423         /*
1424          * Don't mark the inode dirty under page lock. First, it unnecessarily
1425          * makes the holding time of page lock longer. Second, it forces lock
1426          * ordering of page lock and transaction start for journaling
1427          * filesystems.
1428          */
1429         if (i_size_changed || inline_data)
1430                 ext4_mark_inode_dirty(handle, inode);
1431 
1432         if (pos + len > inode->i_size && ext4_can_truncate(inode))
1433                 /* if we have allocated more blocks and copied
1434                  * less. We will have blocks allocated outside
1435                  * inode->i_size. So truncate them
1436                  */
1437                 ext4_orphan_add(handle, inode);
1438 errout:
1439         ret2 = ext4_journal_stop(handle);
1440         if (!ret)
1441                 ret = ret2;
1442 
1443         if (pos + len > inode->i_size) {
1444                 ext4_truncate_failed_write(inode);
1445                 /*
1446                  * If truncate failed early the inode might still be
1447                  * on the orphan list; we need to make sure the inode
1448                  * is removed from the orphan list in that case.
1449                  */
1450                 if (inode->i_nlink)
1451                         ext4_orphan_del(NULL, inode);
1452         }
1453 
1454         return ret ? ret : copied;
1455 }
1456 
1457 /*
1458  * This is a private version of page_zero_new_buffers() which doesn't
1459  * set the buffer to be dirty, since in data=journalled mode we need
1460  * to call ext4_handle_dirty_metadata() instead.
1461  */
1462 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1463                                             struct page *page,
1464                                             unsigned from, unsigned to)
1465 {
1466         unsigned int block_start = 0, block_end;
1467         struct buffer_head *head, *bh;
1468 
1469         bh = head = page_buffers(page);
1470         do {
1471                 block_end = block_start + bh->b_size;
1472                 if (buffer_new(bh)) {
1473                         if (block_end > from && block_start < to) {
1474                                 if (!PageUptodate(page)) {
1475                                         unsigned start, size;
1476 
1477                                         start = max(from, block_start);
1478                                         size = min(to, block_end) - start;
1479 
1480                                         zero_user(page, start, size);
1481                                         write_end_fn(handle, bh);
1482                                 }
1483                                 clear_buffer_new(bh);
1484                         }
1485                 }
1486                 block_start = block_end;
1487                 bh = bh->b_this_page;
1488         } while (bh != head);
1489 }
1490 
1491 static int ext4_journalled_write_end(struct file *file,
1492                                      struct address_space *mapping,
1493                                      loff_t pos, unsigned len, unsigned copied,
1494                                      struct page *page, void *fsdata)
1495 {
1496         handle_t *handle = ext4_journal_current_handle();
1497         struct inode *inode = mapping->host;
1498         loff_t old_size = inode->i_size;
1499         int ret = 0, ret2;
1500         int partial = 0;
1501         unsigned from, to;
1502         int size_changed = 0;
1503         int inline_data = ext4_has_inline_data(inode);
1504 
1505         trace_ext4_journalled_write_end(inode, pos, len, copied);
1506         from = pos & (PAGE_SIZE - 1);
1507         to = from + len;
1508 
1509         BUG_ON(!ext4_handle_valid(handle));
1510 
1511         if (inline_data) {
1512                 ret = ext4_write_inline_data_end(inode, pos, len,
1513                                                  copied, page);
1514                 if (ret < 0) {
1515                         unlock_page(page);
1516                         put_page(page);
1517                         goto errout;
1518                 }
1519                 copied = ret;
1520         } else if (unlikely(copied < len) && !PageUptodate(page)) {
1521                 copied = 0;
1522                 ext4_journalled_zero_new_buffers(handle, page, from, to);
1523         } else {
1524                 if (unlikely(copied < len))
1525                         ext4_journalled_zero_new_buffers(handle, page,
1526                                                          from + copied, to);
1527                 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1528                                              from + copied, &partial,
1529                                              write_end_fn);
1530                 if (!partial)
1531                         SetPageUptodate(page);
1532         }
1533         size_changed = ext4_update_inode_size(inode, pos + copied);
1534         ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1535         EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1536         unlock_page(page);
1537         put_page(page);
1538 
1539         if (old_size < pos)
1540                 pagecache_isize_extended(inode, old_size, pos);
1541 
1542         if (size_changed || inline_data) {
1543                 ret2 = ext4_mark_inode_dirty(handle, inode);
1544                 if (!ret)
1545                         ret = ret2;
1546         }
1547 
1548         if (pos + len > inode->i_size && ext4_can_truncate(inode))
1549                 /* if we have allocated more blocks and copied
1550                  * less. We will have blocks allocated outside
1551                  * inode->i_size. So truncate them
1552                  */
1553                 ext4_orphan_add(handle, inode);
1554 
1555 errout:
1556         ret2 = ext4_journal_stop(handle);
1557         if (!ret)
1558                 ret = ret2;
1559         if (pos + len > inode->i_size) {
1560                 ext4_truncate_failed_write(inode);
1561                 /*
1562                  * If truncate failed early the inode might still be
1563                  * on the orphan list; we need to make sure the inode
1564                  * is removed from the orphan list in that case.
1565                  */
1566                 if (inode->i_nlink)
1567                         ext4_orphan_del(NULL, inode);
1568         }
1569 
1570         return ret ? ret : copied;
1571 }
1572 
1573 /*
1574  * Reserve space for a single cluster
1575  */
1576 static int ext4_da_reserve_space(struct inode *inode)
1577 {
1578         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1579         struct ext4_inode_info *ei = EXT4_I(inode);
1580         int ret;
1581 
1582         /*
1583          * We will charge metadata quota at writeout time; this saves
1584          * us from metadata over-estimation, though we may go over by
1585          * a small amount in the end.  Here we just reserve for data.
1586          */
1587         ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1588         if (ret)
1589                 return ret;
1590 
1591         spin_lock(&ei->i_block_reservation_lock);
1592         if (ext4_claim_free_clusters(sbi, 1, 0)) {
1593                 spin_unlock(&ei->i_block_reservation_lock);
1594                 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1595                 return -ENOSPC;
1596         }
1597         ei->i_reserved_data_blocks++;
1598         trace_ext4_da_reserve_space(inode);
1599         spin_unlock(&ei->i_block_reservation_lock);
1600 
1601         return 0;       /* success */
1602 }
1603 
1604 void ext4_da_release_space(struct inode *inode, int to_free)
1605 {
1606         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1607         struct ext4_inode_info *ei = EXT4_I(inode);
1608 
1609         if (!to_free)
1610                 return;         /* Nothing to release, exit */
1611 
1612         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1613 
1614         trace_ext4_da_release_space(inode, to_free);
1615         if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1616                 /*
1617                  * if there aren't enough reserved blocks, then the
1618                  * counter is messed up somewhere.  Since this
1619                  * function is called from invalidate page, it's
1620                  * harmless to return without any action.
1621                  */
1622                 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1623                          "ino %lu, to_free %d with only %d reserved "
1624                          "data blocks", inode->i_ino, to_free,
1625                          ei->i_reserved_data_blocks);
1626                 WARN_ON(1);
1627                 to_free = ei->i_reserved_data_blocks;
1628         }
1629         ei->i_reserved_data_blocks -= to_free;
1630 
1631         /* update fs dirty data blocks counter */
1632         percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1633 
1634         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1635 
1636         dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1637 }
1638 
1639 static void ext4_da_page_release_reservation(struct page *page,
1640                                              unsigned int offset,
1641                                              unsigned int length)
1642 {
1643         int contiguous_blks = 0;
1644         struct buffer_head *head, *bh;
1645         unsigned int curr_off = 0;
1646         struct inode *inode = page->mapping->host;
1647         unsigned int stop = offset + length;
1648         ext4_fsblk_t lblk;
1649 
1650         BUG_ON(stop > PAGE_SIZE || stop < length);
1651 
1652         head = page_buffers(page);
1653         bh = head;
1654         do {
1655                 unsigned int next_off = curr_off + bh->b_size;
1656 
1657                 if (next_off > stop)
1658                         break;
1659 
1660                 if ((offset <= curr_off) && (buffer_delay(bh))) {
1661                         contiguous_blks++;
1662                         clear_buffer_delay(bh);
1663                 } else if (contiguous_blks) {
1664                         lblk = page->index <<
1665                                (PAGE_SHIFT - inode->i_blkbits);
1666                         lblk += (curr_off >> inode->i_blkbits) -
1667                                 contiguous_blks;
1668                         ext4_es_remove_blks(inode, lblk, contiguous_blks);
1669                         contiguous_blks = 0;
1670                 }
1671                 curr_off = next_off;
1672         } while ((bh = bh->b_this_page) != head);
1673 
1674         if (contiguous_blks) {
1675                 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1676                 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1677                 ext4_es_remove_blks(inode, lblk, contiguous_blks);
1678         }
1679 
1680 }
1681 
1682 /*
1683  * Delayed allocation stuff
1684  */
1685 
1686 struct mpage_da_data {
1687         struct inode *inode;
1688         struct writeback_control *wbc;
1689 
1690         pgoff_t first_page;     /* The first page to write */
1691         pgoff_t next_page;      /* Current page to examine */
1692         pgoff_t last_page;      /* Last page to examine */
1693         /*
1694          * Extent to map - this can be after first_page because that can be
1695          * fully mapped. We somewhat abuse m_flags to store whether the extent
1696          * is delalloc or unwritten.
1697          */
1698         struct ext4_map_blocks map;
1699         struct ext4_io_submit io_submit;        /* IO submission data */
1700         unsigned int do_map:1;
1701 };
1702 
1703 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1704                                        bool invalidate)
1705 {
1706         int nr_pages, i;
1707         pgoff_t index, end;
1708         struct pagevec pvec;
1709         struct inode *inode = mpd->inode;
1710         struct address_space *mapping = inode->i_mapping;
1711 
1712         /* This is necessary when next_page == 0. */
1713         if (mpd->first_page >= mpd->next_page)
1714                 return;
1715 
1716         index = mpd->first_page;
1717         end   = mpd->next_page - 1;
1718         if (invalidate) {
1719                 ext4_lblk_t start, last;
1720                 start = index << (PAGE_SHIFT - inode->i_blkbits);
1721                 last = end << (PAGE_SHIFT - inode->i_blkbits);
1722                 ext4_es_remove_extent(inode, start, last - start + 1);
1723         }
1724 
1725         pagevec_init(&pvec);
1726         while (index <= end) {
1727                 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1728                 if (nr_pages == 0)
1729                         break;
1730                 for (i = 0; i < nr_pages; i++) {
1731                         struct page *page = pvec.pages[i];
1732 
1733                         BUG_ON(!PageLocked(page));
1734                         BUG_ON(PageWriteback(page));
1735                         if (invalidate) {
1736                                 if (page_mapped(page))
1737                                         clear_page_dirty_for_io(page);
1738                                 block_invalidatepage(page, 0, PAGE_SIZE);
1739                                 ClearPageUptodate(page);
1740                         }
1741                         unlock_page(page);
1742                 }
1743                 pagevec_release(&pvec);
1744         }
1745 }
1746 
1747 static void ext4_print_free_blocks(struct inode *inode)
1748 {
1749         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1750         struct super_block *sb = inode->i_sb;
1751         struct ext4_inode_info *ei = EXT4_I(inode);
1752 
1753         ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1754                EXT4_C2B(EXT4_SB(inode->i_sb),
1755                         ext4_count_free_clusters(sb)));
1756         ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1757         ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1758                (long long) EXT4_C2B(EXT4_SB(sb),
1759                 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1760         ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1761                (long long) EXT4_C2B(EXT4_SB(sb),
1762                 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1763         ext4_msg(sb, KERN_CRIT, "Block reservation details");
1764         ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1765                  ei->i_reserved_data_blocks);
1766         return;
1767 }
1768 
1769 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1770 {
1771         return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1772 }
1773 
1774 /*
1775  * ext4_insert_delayed_block - adds a delayed block to the extents status
1776  *                             tree, incrementing the reserved cluster/block
1777  *                             count or making a pending reservation
1778  *                             where needed
1779  *
1780  * @inode - file containing the newly added block
1781  * @lblk - logical block to be added
1782  *
1783  * Returns 0 on success, negative error code on failure.
1784  */
1785 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1786 {
1787         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1788         int ret;
1789         bool allocated = false;
1790 
1791         /*
1792          * If the cluster containing lblk is shared with a delayed,
1793          * written, or unwritten extent in a bigalloc file system, it's
1794          * already been accounted for and does not need to be reserved.
1795          * A pending reservation must be made for the cluster if it's
1796          * shared with a written or unwritten extent and doesn't already
1797          * have one.  Written and unwritten extents can be purged from the
1798          * extents status tree if the system is under memory pressure, so
1799          * it's necessary to examine the extent tree if a search of the
1800          * extents status tree doesn't get a match.
1801          */
1802         if (sbi->s_cluster_ratio == 1) {
1803                 ret = ext4_da_reserve_space(inode);
1804                 if (ret != 0)   /* ENOSPC */
1805                         goto errout;
1806         } else {   /* bigalloc */
1807                 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1808                         if (!ext4_es_scan_clu(inode,
1809                                               &ext4_es_is_mapped, lblk)) {
1810                                 ret = ext4_clu_mapped(inode,
1811                                                       EXT4_B2C(sbi, lblk));
1812                                 if (ret < 0)
1813                                         goto errout;
1814                                 if (ret == 0) {
1815                                         ret = ext4_da_reserve_space(inode);
1816                                         if (ret != 0)   /* ENOSPC */
1817                                                 goto errout;
1818                                 } else {
1819                                         allocated = true;
1820                                 }
1821                         } else {
1822                                 allocated = true;
1823                         }
1824                 }
1825         }
1826 
1827         ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1828 
1829 errout:
1830         return ret;
1831 }
1832 
1833 /*
1834  * This function is grabs code from the very beginning of
1835  * ext4_map_blocks, but assumes that the caller is from delayed write
1836  * time. This function looks up the requested blocks and sets the
1837  * buffer delay bit under the protection of i_data_sem.
1838  */
1839 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1840                               struct ext4_map_blocks *map,
1841                               struct buffer_head *bh)
1842 {
1843         struct extent_status es;
1844         int retval;
1845         sector_t invalid_block = ~((sector_t) 0xffff);
1846 #ifdef ES_AGGRESSIVE_TEST
1847         struct ext4_map_blocks orig_map;
1848 
1849         memcpy(&orig_map, map, sizeof(*map));
1850 #endif
1851 
1852         if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1853                 invalid_block = ~0;
1854 
1855         map->m_flags = 0;
1856         ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1857                   "logical block %lu\n", inode->i_ino, map->m_len,
1858                   (unsigned long) map->m_lblk);
1859 
1860         /* Lookup extent status tree firstly */
1861         if (ext4_es_lookup_extent(inode, iblock, &es)) {
1862                 if (ext4_es_is_hole(&es)) {
1863                         retval = 0;
1864                         down_read(&EXT4_I(inode)->i_data_sem);
1865                         goto add_delayed;
1866                 }
1867 
1868                 /*
1869                  * Delayed extent could be allocated by fallocate.
1870                  * So we need to check it.
1871                  */
1872                 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1873                         map_bh(bh, inode->i_sb, invalid_block);
1874                         set_buffer_new(bh);
1875                         set_buffer_delay(bh);
1876                         return 0;
1877                 }
1878 
1879                 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1880                 retval = es.es_len - (iblock - es.es_lblk);
1881                 if (retval > map->m_len)
1882                         retval = map->m_len;
1883                 map->m_len = retval;
1884                 if (ext4_es_is_written(&es))
1885                         map->m_flags |= EXT4_MAP_MAPPED;
1886                 else if (ext4_es_is_unwritten(&es))
1887                         map->m_flags |= EXT4_MAP_UNWRITTEN;
1888                 else
1889                         BUG();
1890 
1891 #ifdef ES_AGGRESSIVE_TEST
1892                 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1893 #endif
1894                 return retval;
1895         }
1896 
1897         /*
1898          * Try to see if we can get the block without requesting a new
1899          * file system block.
1900          */
1901         down_read(&EXT4_I(inode)->i_data_sem);
1902         if (ext4_has_inline_data(inode))
1903                 retval = 0;
1904         else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1905                 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1906         else
1907                 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1908 
1909 add_delayed:
1910         if (retval == 0) {
1911                 int ret;
1912 
1913                 /*
1914                  * XXX: __block_prepare_write() unmaps passed block,
1915                  * is it OK?
1916                  */
1917 
1918                 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1919                 if (ret != 0) {
1920                         retval = ret;
1921                         goto out_unlock;
1922                 }
1923 
1924                 map_bh(bh, inode->i_sb, invalid_block);
1925                 set_buffer_new(bh);
1926                 set_buffer_delay(bh);
1927         } else if (retval > 0) {
1928                 int ret;
1929                 unsigned int status;
1930 
1931                 if (unlikely(retval != map->m_len)) {
1932                         ext4_warning(inode->i_sb,
1933                                      "ES len assertion failed for inode "
1934                                      "%lu: retval %d != map->m_len %d",
1935                                      inode->i_ino, retval, map->m_len);
1936                         WARN_ON(1);
1937                 }
1938 
1939                 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1940                                 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1941                 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1942                                             map->m_pblk, status);
1943                 if (ret != 0)
1944                         retval = ret;
1945         }
1946 
1947 out_unlock:
1948         up_read((&EXT4_I(inode)->i_data_sem));
1949 
1950         return retval;
1951 }
1952 
1953 /*
1954  * This is a special get_block_t callback which is used by
1955  * ext4_da_write_begin().  It will either return mapped block or
1956  * reserve space for a single block.
1957  *
1958  * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1959  * We also have b_blocknr = -1 and b_bdev initialized properly
1960  *
1961  * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1962  * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1963  * initialized properly.
1964  */
1965 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1966                            struct buffer_head *bh, int create)
1967 {
1968         struct ext4_map_blocks map;
1969         int ret = 0;
1970 
1971         BUG_ON(create == 0);
1972         BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1973 
1974         map.m_lblk = iblock;
1975         map.m_len = 1;
1976 
1977         /*
1978          * first, we need to know whether the block is allocated already
1979          * preallocated blocks are unmapped but should treated
1980          * the same as allocated blocks.
1981          */
1982         ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1983         if (ret <= 0)
1984                 return ret;
1985 
1986         map_bh(bh, inode->i_sb, map.m_pblk);
1987         ext4_update_bh_state(bh, map.m_flags);
1988 
1989         if (buffer_unwritten(bh)) {
1990                 /* A delayed write to unwritten bh should be marked
1991                  * new and mapped.  Mapped ensures that we don't do
1992                  * get_block multiple times when we write to the same
1993                  * offset and new ensures that we do proper zero out
1994                  * for partial write.
1995                  */
1996                 set_buffer_new(bh);
1997                 set_buffer_mapped(bh);
1998         }
1999         return 0;
2000 }
2001 
2002 static int bget_one(handle_t *handle, struct buffer_head *bh)
2003 {
2004         get_bh(bh);
2005         return 0;
2006 }
2007 
2008 static int bput_one(handle_t *handle, struct buffer_head *bh)
2009 {
2010         put_bh(bh);
2011         return 0;
2012 }
2013 
2014 static int __ext4_journalled_writepage(struct page *page,
2015                                        unsigned int len)
2016 {
2017         struct address_space *mapping = page->mapping;
2018         struct inode *inode = mapping->host;
2019         struct buffer_head *page_bufs = NULL;
2020         handle_t *handle = NULL;
2021         int ret = 0, err = 0;
2022         int inline_data = ext4_has_inline_data(inode);
2023         struct buffer_head *inode_bh = NULL;
2024 
2025         ClearPageChecked(page);
2026 
2027         if (inline_data) {
2028                 BUG_ON(page->index != 0);
2029                 BUG_ON(len > ext4_get_max_inline_size(inode));
2030                 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2031                 if (inode_bh == NULL)
2032                         goto out;
2033         } else {
2034                 page_bufs = page_buffers(page);
2035                 if (!page_bufs) {
2036                         BUG();
2037                         goto out;
2038                 }
2039                 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2040                                        NULL, bget_one);
2041         }
2042         /*
2043          * We need to release the page lock before we start the
2044          * journal, so grab a reference so the page won't disappear
2045          * out from under us.
2046          */
2047         get_page(page);
2048         unlock_page(page);
2049 
2050         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2051                                     ext4_writepage_trans_blocks(inode));
2052         if (IS_ERR(handle)) {
2053                 ret = PTR_ERR(handle);
2054                 put_page(page);
2055                 goto out_no_pagelock;
2056         }
2057         BUG_ON(!ext4_handle_valid(handle));
2058 
2059         lock_page(page);
2060         put_page(page);
2061         if (page->mapping != mapping) {
2062                 /* The page got truncated from under us */
2063                 ext4_journal_stop(handle);
2064                 ret = 0;
2065                 goto out;
2066         }
2067 
2068         if (inline_data) {
2069                 ret = ext4_mark_inode_dirty(handle, inode);
2070         } else {
2071                 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2072                                              do_journal_get_write_access);
2073 
2074                 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2075                                              write_end_fn);
2076         }
2077         if (ret == 0)
2078                 ret = err;
2079         EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2080         err = ext4_journal_stop(handle);
2081         if (!ret)
2082                 ret = err;
2083 
2084         if (!ext4_has_inline_data(inode))
2085                 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2086                                        NULL, bput_one);
2087         ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2088 out:
2089         unlock_page(page);
2090 out_no_pagelock:
2091         brelse(inode_bh);
2092         return ret;
2093 }
2094 
2095 /*
2096  * Note that we don't need to start a transaction unless we're journaling data
2097  * because we should have holes filled from ext4_page_mkwrite(). We even don't
2098  * need to file the inode to the transaction's list in ordered mode because if
2099  * we are writing back data added by write(), the inode is already there and if
2100  * we are writing back data modified via mmap(), no one guarantees in which
2101  * transaction the data will hit the disk. In case we are journaling data, we
2102  * cannot start transaction directly because transaction start ranks above page
2103  * lock so we have to do some magic.
2104  *
2105  * This function can get called via...
2106  *   - ext4_writepages after taking page lock (have journal handle)
2107  *   - journal_submit_inode_data_buffers (no journal handle)
2108  *   - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2109  *   - grab_page_cache when doing write_begin (have journal handle)
2110  *
2111  * We don't do any block allocation in this function. If we have page with
2112  * multiple blocks we need to write those buffer_heads that are mapped. This
2113  * is important for mmaped based write. So if we do with blocksize 1K
2114  * truncate(f, 1024);
2115  * a = mmap(f, 0, 4096);
2116  * a[0] = 'a';
2117  * truncate(f, 4096);
2118  * we have in the page first buffer_head mapped via page_mkwrite call back
2119  * but other buffer_heads would be unmapped but dirty (dirty done via the
2120  * do_wp_page). So writepage should write the first block. If we modify
2121  * the mmap area beyond 1024 we will again get a page_fault and the
2122  * page_mkwrite callback will do the block allocation and mark the
2123  * buffer_heads mapped.
2124  *
2125  * We redirty the page if we have any buffer_heads that is either delay or
2126  * unwritten in the page.
2127  *
2128  * We can get recursively called as show below.
2129  *
2130  *      ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2131  *              ext4_writepage()
2132  *
2133  * But since we don't do any block allocation we should not deadlock.
2134  * Page also have the dirty flag cleared so we don't get recurive page_lock.
2135  */
2136 static int ext4_writepage(struct page *page,
2137                           struct writeback_control *wbc)
2138 {
2139         int ret = 0;
2140         loff_t size;
2141         unsigned int len;
2142         struct buffer_head *page_bufs = NULL;
2143         struct inode *inode = page->mapping->host;
2144         struct ext4_io_submit io_submit;
2145         bool keep_towrite = false;
2146 
2147         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2148                 ext4_invalidatepage(page, 0, PAGE_SIZE);
2149                 unlock_page(page);
2150                 return -EIO;
2151         }
2152 
2153         trace_ext4_writepage(page);
2154         size = i_size_read(inode);
2155         if (page->index == size >> PAGE_SHIFT)
2156                 len = size & ~PAGE_MASK;
2157         else
2158                 len = PAGE_SIZE;
2159 
2160         page_bufs = page_buffers(page);
2161         /*
2162          * We cannot do block allocation or other extent handling in this
2163          * function. If there are buffers needing that, we have to redirty
2164          * the page. But we may reach here when we do a journal commit via
2165          * journal_submit_inode_data_buffers() and in that case we must write
2166          * allocated buffers to achieve data=ordered mode guarantees.
2167          *
2168          * Also, if there is only one buffer per page (the fs block
2169          * size == the page size), if one buffer needs block
2170          * allocation or needs to modify the extent tree to clear the
2171          * unwritten flag, we know that the page can't be written at
2172          * all, so we might as well refuse the write immediately.
2173          * Unfortunately if the block size != page size, we can't as
2174          * easily detect this case using ext4_walk_page_buffers(), but
2175          * for the extremely common case, this is an optimization that
2176          * skips a useless round trip through ext4_bio_write_page().
2177          */
2178         if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2179                                    ext4_bh_delay_or_unwritten)) {
2180                 redirty_page_for_writepage(wbc, page);
2181                 if ((current->flags & PF_MEMALLOC) ||
2182                     (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2183                         /*
2184                          * For memory cleaning there's no point in writing only
2185                          * some buffers. So just bail out. Warn if we came here
2186                          * from direct reclaim.
2187                          */
2188                         WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2189                                                         == PF_MEMALLOC);
2190                         unlock_page(page);
2191                         return 0;
2192                 }
2193                 keep_towrite = true;
2194         }
2195 
2196         if (PageChecked(page) && ext4_should_journal_data(inode))
2197                 /*
2198                  * It's mmapped pagecache.  Add buffers and journal it.  There
2199                  * doesn't seem much point in redirtying the page here.
2200                  */
2201                 return __ext4_journalled_writepage(page, len);
2202 
2203         ext4_io_submit_init(&io_submit, wbc);
2204         io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2205         if (!io_submit.io_end) {
2206                 redirty_page_for_writepage(wbc, page);
2207                 unlock_page(page);
2208                 return -ENOMEM;
2209         }
2210         ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2211         ext4_io_submit(&io_submit);
2212         /* Drop io_end reference we got from init */
2213         ext4_put_io_end_defer(io_submit.io_end);
2214         return ret;
2215 }
2216 
2217 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2218 {
2219         int len;
2220         loff_t size;
2221         int err;
2222 
2223         BUG_ON(page->index != mpd->first_page);
2224         clear_page_dirty_for_io(page);
2225         /*
2226          * We have to be very careful here!  Nothing protects writeback path
2227          * against i_size changes and the page can be writeably mapped into
2228          * page tables. So an application can be growing i_size and writing
2229          * data through mmap while writeback runs. clear_page_dirty_for_io()
2230          * write-protects our page in page tables and the page cannot get
2231          * written to again until we release page lock. So only after
2232          * clear_page_dirty_for_io() we are safe to sample i_size for
2233          * ext4_bio_write_page() to zero-out tail of the written page. We rely
2234          * on the barrier provided by TestClearPageDirty in
2235          * clear_page_dirty_for_io() to make sure i_size is really sampled only
2236          * after page tables are updated.
2237          */
2238         size = i_size_read(mpd->inode);
2239         if (page->index == size >> PAGE_SHIFT)
2240                 len = size & ~PAGE_MASK;
2241         else
2242                 len = PAGE_SIZE;
2243         err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2244         if (!err)
2245                 mpd->wbc->nr_to_write--;
2246         mpd->first_page++;
2247 
2248         return err;
2249 }
2250 
2251 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2252 
2253 /*
2254  * mballoc gives us at most this number of blocks...
2255  * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2256  * The rest of mballoc seems to handle chunks up to full group size.
2257  */
2258 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2259 
2260 /*
2261  * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2262  *
2263  * @mpd - extent of blocks
2264  * @lblk - logical number of the block in the file
2265  * @bh - buffer head we want to add to the extent
2266  *
2267  * The function is used to collect contig. blocks in the same state. If the
2268  * buffer doesn't require mapping for writeback and we haven't started the
2269  * extent of buffers to map yet, the function returns 'true' immediately - the
2270  * caller can write the buffer right away. Otherwise the function returns true
2271  * if the block has been added to the extent, false if the block couldn't be
2272  * added.
2273  */
2274 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2275                                    struct buffer_head *bh)
2276 {
2277         struct ext4_map_blocks *map = &mpd->map;
2278 
2279         /* Buffer that doesn't need mapping for writeback? */
2280         if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2281             (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2282                 /* So far no extent to map => we write the buffer right away */
2283                 if (map->m_len == 0)
2284                         return true;
2285                 return false;
2286         }
2287 
2288         /* First block in the extent? */
2289         if (map->m_len == 0) {
2290                 /* We cannot map unless handle is started... */
2291                 if (!mpd->do_map)
2292                         return false;
2293                 map->m_lblk = lblk;
2294                 map->m_len = 1;
2295                 map->m_flags = bh->b_state & BH_FLAGS;
2296                 return true;
2297         }
2298 
2299         /* Don't go larger than mballoc is willing to allocate */
2300         if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2301                 return false;
2302 
2303         /* Can we merge the block to our big extent? */
2304         if (lblk == map->m_lblk + map->m_len &&
2305             (bh->b_state & BH_FLAGS) == map->m_flags) {
2306                 map->m_len++;
2307                 return true;
2308         }
2309         return false;
2310 }
2311 
2312 /*
2313  * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2314  *
2315  * @mpd - extent of blocks for mapping
2316  * @head - the first buffer in the page
2317  * @bh - buffer we should start processing from
2318  * @lblk - logical number of the block in the file corresponding to @bh
2319  *
2320  * Walk through page buffers from @bh upto @head (exclusive) and either submit
2321  * the page for IO if all buffers in this page were mapped and there's no
2322  * accumulated extent of buffers to map or add buffers in the page to the
2323  * extent of buffers to map. The function returns 1 if the caller can continue
2324  * by processing the next page, 0 if it should stop adding buffers to the
2325  * extent to map because we cannot extend it anymore. It can also return value
2326  * < 0 in case of error during IO submission.
2327  */
2328 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2329                                    struct buffer_head *head,
2330                                    struct buffer_head *bh,
2331                                    ext4_lblk_t lblk)
2332 {
2333         struct inode *inode = mpd->inode;
2334         int err;
2335         ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2336                                                         >> inode->i_blkbits;
2337 
2338         do {
2339                 BUG_ON(buffer_locked(bh));
2340 
2341                 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2342                         /* Found extent to map? */
2343                         if (mpd->map.m_len)
2344                                 return 0;
2345                         /* Buffer needs mapping and handle is not started? */
2346                         if (!mpd->do_map)
2347                                 return 0;
2348                         /* Everything mapped so far and we hit EOF */
2349                         break;
2350                 }
2351         } while (lblk++, (bh = bh->b_this_page) != head);
2352         /* So far everything mapped? Submit the page for IO. */
2353         if (mpd->map.m_len == 0) {
2354                 err = mpage_submit_page(mpd, head->b_page);
2355                 if (err < 0)
2356                         return err;
2357         }
2358         return lblk < blocks;
2359 }
2360 
2361 /*
2362  * mpage_map_buffers - update buffers corresponding to changed extent and
2363  *                     submit fully mapped pages for IO
2364  *
2365  * @mpd - description of extent to map, on return next extent to map
2366  *
2367  * Scan buffers corresponding to changed extent (we expect corresponding pages
2368  * to be already locked) and update buffer state according to new extent state.
2369  * We map delalloc buffers to their physical location, clear unwritten bits,
2370  * and mark buffers as uninit when we perform writes to unwritten extents
2371  * and do extent conversion after IO is finished. If the last page is not fully
2372  * mapped, we update @map to the next extent in the last page that needs
2373  * mapping. Otherwise we submit the page for IO.
2374  */
2375 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2376 {
2377         struct pagevec pvec;
2378         int nr_pages, i;
2379         struct inode *inode = mpd->inode;
2380         struct buffer_head *head, *bh;
2381         int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2382         pgoff_t start, end;
2383         ext4_lblk_t lblk;
2384         sector_t pblock;
2385         int err;
2386 
2387         start = mpd->map.m_lblk >> bpp_bits;
2388         end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2389         lblk = start << bpp_bits;
2390         pblock = mpd->map.m_pblk;
2391 
2392         pagevec_init(&pvec);
2393         while (start <= end) {
2394                 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2395                                                 &start, end);
2396                 if (nr_pages == 0)
2397                         break;
2398                 for (i = 0; i < nr_pages; i++) {
2399                         struct page *page = pvec.pages[i];
2400 
2401                         bh = head = page_buffers(page);
2402                         do {
2403                                 if (lblk < mpd->map.m_lblk)
2404                                         continue;
2405                                 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2406                                         /*
2407                                          * Buffer after end of mapped extent.
2408                                          * Find next buffer in the page to map.
2409                                          */
2410                                         mpd->map.m_len = 0;
2411                                         mpd->map.m_flags = 0;
2412                                         /*
2413                                          * FIXME: If dioread_nolock supports
2414                                          * blocksize < pagesize, we need to make
2415                                          * sure we add size mapped so far to
2416                                          * io_end->size as the following call
2417                                          * can submit the page for IO.
2418                                          */
2419                                         err = mpage_process_page_bufs(mpd, head,
2420                                                                       bh, lblk);
2421                                         pagevec_release(&pvec);
2422                                         if (err > 0)
2423                                                 err = 0;
2424                                         return err;
2425                                 }
2426                                 if (buffer_delay(bh)) {
2427                                         clear_buffer_delay(bh);
2428                                         bh->b_blocknr = pblock++;
2429                                 }
2430                                 clear_buffer_unwritten(bh);
2431                         } while (lblk++, (bh = bh->b_this_page) != head);
2432 
2433                         /*
2434                          * FIXME: This is going to break if dioread_nolock
2435                          * supports blocksize < pagesize as we will try to
2436                          * convert potentially unmapped parts of inode.
2437                          */
2438                         mpd->io_submit.io_end->size += PAGE_SIZE;
2439                         /* Page fully mapped - let IO run! */
2440                         err = mpage_submit_page(mpd, page);
2441                         if (err < 0) {
2442                                 pagevec_release(&pvec);
2443                                 return err;
2444                         }
2445                 }
2446                 pagevec_release(&pvec);
2447         }
2448         /* Extent fully mapped and matches with page boundary. We are done. */
2449         mpd->map.m_len = 0;
2450         mpd->map.m_flags = 0;
2451         return 0;
2452 }
2453 
2454 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2455 {
2456         struct inode *inode = mpd->inode;
2457         struct ext4_map_blocks *map = &mpd->map;
2458         int get_blocks_flags;
2459         int err, dioread_nolock;
2460 
2461         trace_ext4_da_write_pages_extent(inode, map);
2462         /*
2463          * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2464          * to convert an unwritten extent to be initialized (in the case
2465          * where we have written into one or more preallocated blocks).  It is
2466          * possible that we're going to need more metadata blocks than
2467          * previously reserved. However we must not fail because we're in
2468          * writeback and there is nothing we can do about it so it might result
2469          * in data loss.  So use reserved blocks to allocate metadata if
2470          * possible.
2471          *
2472          * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2473          * the blocks in question are delalloc blocks.  This indicates
2474          * that the blocks and quotas has already been checked when
2475          * the data was copied into the page cache.
2476          */
2477         get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2478                            EXT4_GET_BLOCKS_METADATA_NOFAIL |
2479                            EXT4_GET_BLOCKS_IO_SUBMIT;
2480         dioread_nolock = ext4_should_dioread_nolock(inode);
2481         if (dioread_nolock)
2482                 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2483         if (map->m_flags & (1 << BH_Delay))
2484                 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2485 
2486         err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2487         if (err < 0)
2488                 return err;
2489         if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2490                 if (!mpd->io_submit.io_end->handle &&
2491                     ext4_handle_valid(handle)) {
2492                         mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2493                         handle->h_rsv_handle = NULL;
2494                 }
2495                 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2496         }
2497 
2498         BUG_ON(map->m_len == 0);
2499         return 0;
2500 }
2501 
2502 /*
2503  * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2504  *                               mpd->len and submit pages underlying it for IO
2505  *
2506  * @handle - handle for journal operations
2507  * @mpd - extent to map
2508  * @give_up_on_write - we set this to true iff there is a fatal error and there
2509  *                     is no hope of writing the data. The caller should discard
2510  *                     dirty pages to avoid infinite loops.
2511  *
2512  * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2513  * delayed, blocks are allocated, if it is unwritten, we may need to convert
2514  * them to initialized or split the described range from larger unwritten
2515  * extent. Note that we need not map all the described range since allocation
2516  * can return less blocks or the range is covered by more unwritten extents. We
2517  * cannot map more because we are limited by reserved transaction credits. On
2518  * the other hand we always make sure that the last touched page is fully
2519  * mapped so that it can be written out (and thus forward progress is
2520  * guaranteed). After mapping we submit all mapped pages for IO.
2521  */
2522 static int mpage_map_and_submit_extent(handle_t *handle,
2523                                        struct mpage_da_data *mpd,
2524                                        bool *give_up_on_write)
2525 {
2526         struct inode *inode = mpd->inode;
2527         struct ext4_map_blocks *map = &mpd->map;
2528         int err;
2529         loff_t disksize;
2530         int progress = 0;
2531 
2532         mpd->io_submit.io_end->offset =
2533                                 ((loff_t)map->m_lblk) << inode->i_blkbits;
2534         do {
2535                 err = mpage_map_one_extent(handle, mpd);
2536                 if (err < 0) {
2537                         struct super_block *sb = inode->i_sb;
2538 
2539                         if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2540                             EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2541                                 goto invalidate_dirty_pages;
2542                         /*
2543                          * Let the uper layers retry transient errors.
2544                          * In the case of ENOSPC, if ext4_count_free_blocks()
2545                          * is non-zero, a commit should free up blocks.
2546                          */
2547                         if ((err == -ENOMEM) ||
2548                             (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2549                                 if (progress)
2550                                         goto update_disksize;
2551                                 return err;
2552                         }
2553                         ext4_msg(sb, KERN_CRIT,
2554                                  "Delayed block allocation failed for "
2555                                  "inode %lu at logical offset %llu with"
2556                                  " max blocks %u with error %d",
2557                                  inode->i_ino,
2558                                  (unsigned long long)map->m_lblk,
2559                                  (unsigned)map->m_len, -err);
2560                         ext4_msg(sb, KERN_CRIT,
2561                                  "This should not happen!! Data will "
2562                                  "be lost\n");
2563                         if (err == -ENOSPC)
2564                                 ext4_print_free_blocks(inode);
2565                 invalidate_dirty_pages:
2566                         *give_up_on_write = true;
2567                         return err;
2568                 }
2569                 progress = 1;
2570                 /*
2571                  * Update buffer state, submit mapped pages, and get us new
2572                  * extent to map
2573                  */
2574                 err = mpage_map_and_submit_buffers(mpd);
2575                 if (err < 0)
2576                         goto update_disksize;
2577         } while (map->m_len);
2578 
2579 update_disksize:
2580         /*
2581          * Update on-disk size after IO is submitted.  Races with
2582          * truncate are avoided by checking i_size under i_data_sem.
2583          */
2584         disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2585         if (disksize > EXT4_I(inode)->i_disksize) {
2586                 int err2;
2587                 loff_t i_size;
2588 
2589                 down_write(&EXT4_I(inode)->i_data_sem);
2590                 i_size = i_size_read(inode);
2591                 if (disksize > i_size)
2592                         disksize = i_size;
2593                 if (disksize > EXT4_I(inode)->i_disksize)
2594                         EXT4_I(inode)->i_disksize = disksize;
2595                 up_write(&EXT4_I(inode)->i_data_sem);
2596                 err2 = ext4_mark_inode_dirty(handle, inode);
2597                 if (err2)
2598                         ext4_error(inode->i_sb,
2599                                    "Failed to mark inode %lu dirty",
2600                                    inode->i_ino);
2601                 if (!err)
2602                         err = err2;
2603         }
2604         return err;
2605 }
2606 
2607 /*
2608  * Calculate the total number of credits to reserve for one writepages
2609  * iteration. This is called from ext4_writepages(). We map an extent of
2610  * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2611  * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2612  * bpp - 1 blocks in bpp different extents.
2613  */
2614 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2615 {
2616         int bpp = ext4_journal_blocks_per_page(inode);
2617 
2618         return ext4_meta_trans_blocks(inode,
2619                                 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2620 }
2621 
2622 /*
2623  * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2624  *                               and underlying extent to map
2625  *
2626  * @mpd - where to look for pages
2627  *
2628  * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2629  * IO immediately. When we find a page which isn't mapped we start accumulating
2630  * extent of buffers underlying these pages that needs mapping (formed by
2631  * either delayed or unwritten buffers). We also lock the pages containing
2632  * these buffers. The extent found is returned in @mpd structure (starting at
2633  * mpd->lblk with length mpd->len blocks).
2634  *
2635  * Note that this function can attach bios to one io_end structure which are
2636  * neither logically nor physically contiguous. Although it may seem as an
2637  * unnecessary complication, it is actually inevitable in blocksize < pagesize
2638  * case as we need to track IO to all buffers underlying a page in one io_end.
2639  */
2640 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2641 {
2642         struct address_space *mapping = mpd->inode->i_mapping;
2643         struct pagevec pvec;
2644         unsigned int nr_pages;
2645         long left = mpd->wbc->nr_to_write;
2646         pgoff_t index = mpd->first_page;
2647         pgoff_t end = mpd->last_page;
2648         xa_mark_t tag;
2649         int i, err = 0;
2650         int blkbits = mpd->inode->i_blkbits;
2651         ext4_lblk_t lblk;
2652         struct buffer_head *head;
2653 
2654         if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2655                 tag = PAGECACHE_TAG_TOWRITE;
2656         else
2657                 tag = PAGECACHE_TAG_DIRTY;
2658 
2659         pagevec_init(&pvec);
2660         mpd->map.m_len = 0;
2661         mpd->next_page = index;
2662         while (index <= end) {
2663                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2664                                 tag);
2665                 if (nr_pages == 0)
2666                         goto out;
2667 
2668                 for (i = 0; i < nr_pages; i++) {
2669                         struct page *page = pvec.pages[i];
2670 
2671                         /*
2672                          * Accumulated enough dirty pages? This doesn't apply
2673                          * to WB_SYNC_ALL mode. For integrity sync we have to
2674                          * keep going because someone may be concurrently
2675                          * dirtying pages, and we might have synced a lot of
2676                          * newly appeared dirty pages, but have not synced all
2677                          * of the old dirty pages.
2678                          */
2679                         if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2680                                 goto out;
2681 
2682                         /* If we can't merge this page, we are done. */
2683                         if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2684                                 goto out;
2685 
2686                         lock_page(page);
2687                         /*
2688                          * If the page is no longer dirty, or its mapping no
2689                          * longer corresponds to inode we are writing (which
2690                          * means it has been truncated or invalidated), or the
2691                          * page is already under writeback and we are not doing
2692                          * a data integrity writeback, skip the page
2693                          */
2694                         if (!PageDirty(page) ||
2695                             (PageWriteback(page) &&
2696                              (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2697                             unlikely(page->mapping != mapping)) {
2698                                 unlock_page(page);
2699                                 continue;
2700                         }
2701 
2702                         wait_on_page_writeback(page);
2703                         BUG_ON(PageWriteback(page));
2704 
2705                         if (mpd->map.m_len == 0)
2706                                 mpd->first_page = page->index;
2707                         mpd->next_page = page->index + 1;
2708                         /* Add all dirty buffers to mpd */
2709                         lblk = ((ext4_lblk_t)page->index) <<
2710                                 (PAGE_SHIFT - blkbits);
2711                         head = page_buffers(page);
2712                         err = mpage_process_page_bufs(mpd, head, head, lblk);
2713                         if (err <= 0)
2714                                 goto out;
2715                         err = 0;
2716                         left--;
2717                 }
2718                 pagevec_release(&pvec);
2719                 cond_resched();
2720         }
2721         return 0;
2722 out:
2723         pagevec_release(&pvec);
2724         return err;
2725 }
2726 
2727 static int ext4_writepages(struct address_space *mapping,
2728                            struct writeback_control *wbc)
2729 {
2730         pgoff_t writeback_index = 0;
2731         long nr_to_write = wbc->nr_to_write;
2732         int range_whole = 0;
2733         int cycled = 1;
2734         handle_t *handle = NULL;
2735         struct mpage_da_data mpd;
2736         struct inode *inode = mapping->host;
2737         int needed_blocks, rsv_blocks = 0, ret = 0;
2738         struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2739         bool done;
2740         struct blk_plug plug;
2741         bool give_up_on_write = false;
2742 
2743         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2744                 return -EIO;
2745 
2746         percpu_down_read(&sbi->s_journal_flag_rwsem);
2747         trace_ext4_writepages(inode, wbc);
2748 
2749         /*
2750          * No pages to write? This is mainly a kludge to avoid starting
2751          * a transaction for special inodes like journal inode on last iput()
2752          * because that could violate lock ordering on umount
2753          */
2754         if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2755                 goto out_writepages;
2756 
2757         if (ext4_should_journal_data(inode)) {
2758                 ret = generic_writepages(mapping, wbc);
2759                 goto out_writepages;
2760         }
2761 
2762         /*
2763          * If the filesystem has aborted, it is read-only, so return
2764          * right away instead of dumping stack traces later on that
2765          * will obscure the real source of the problem.  We test
2766          * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2767          * the latter could be true if the filesystem is mounted
2768          * read-only, and in that case, ext4_writepages should
2769          * *never* be called, so if that ever happens, we would want
2770          * the stack trace.
2771          */
2772         if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2773                      sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2774                 ret = -EROFS;
2775                 goto out_writepages;
2776         }
2777 
2778         if (ext4_should_dioread_nolock(inode)) {
2779                 /*
2780                  * We may need to convert up to one extent per block in
2781                  * the page and we may dirty the inode.
2782                  */
2783                 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2784                                                 PAGE_SIZE >> inode->i_blkbits);
2785         }
2786 
2787         /*
2788          * If we have inline data and arrive here, it means that
2789          * we will soon create the block for the 1st page, so
2790          * we'd better clear the inline data here.
2791          */
2792         if (ext4_has_inline_data(inode)) {
2793                 /* Just inode will be modified... */
2794                 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2795                 if (IS_ERR(handle)) {
2796                         ret = PTR_ERR(handle);
2797                         goto out_writepages;
2798                 }
2799                 BUG_ON(ext4_test_inode_state(inode,
2800                                 EXT4_STATE_MAY_INLINE_DATA));
2801                 ext4_destroy_inline_data(handle, inode);
2802                 ext4_journal_stop(handle);
2803         }
2804 
2805         if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2806                 range_whole = 1;
2807 
2808         if (wbc->range_cyclic) {
2809                 writeback_index = mapping->writeback_index;
2810                 if (writeback_index)
2811                         cycled = 0;
2812                 mpd.first_page = writeback_index;
2813                 mpd.last_page = -1;
2814         } else {
2815                 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2816                 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2817         }
2818 
2819         mpd.inode = inode;
2820         mpd.wbc = wbc;
2821         ext4_io_submit_init(&mpd.io_submit, wbc);
2822 retry:
2823         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2824                 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2825         done = false;
2826         blk_start_plug(&plug);
2827 
2828         /*
2829          * First writeback pages that don't need mapping - we can avoid
2830          * starting a transaction unnecessarily and also avoid being blocked
2831          * in the block layer on device congestion while having transaction
2832          * started.
2833          */
2834         mpd.do_map = 0;
2835         mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2836         if (!mpd.io_submit.io_end) {
2837                 ret = -ENOMEM;
2838                 goto unplug;
2839         }
2840         ret = mpage_prepare_extent_to_map(&mpd);
2841         /* Unlock pages we didn't use */
2842         mpage_release_unused_pages(&mpd, false);
2843         /* Submit prepared bio */
2844         ext4_io_submit(&mpd.io_submit);
2845         ext4_put_io_end_defer(mpd.io_submit.io_end);
2846         mpd.io_submit.io_end = NULL;
2847         if (ret < 0)
2848                 goto unplug;
2849 
2850         while (!done && mpd.first_page <= mpd.last_page) {
2851                 /* For each extent of pages we use new io_end */
2852                 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2853                 if (!mpd.io_submit.io_end) {
2854                         ret = -ENOMEM;
2855                         break;
2856                 }
2857 
2858                 /*
2859                  * We have two constraints: We find one extent to map and we
2860                  * must always write out whole page (makes a difference when
2861                  * blocksize < pagesize) so that we don't block on IO when we
2862                  * try to write out the rest of the page. Journalled mode is
2863                  * not supported by delalloc.
2864                  */
2865                 BUG_ON(ext4_should_journal_data(inode));
2866                 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2867 
2868                 /* start a new transaction */
2869                 handle = ext4_journal_start_with_reserve(inode,
2870                                 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2871                 if (IS_ERR(handle)) {
2872                         ret = PTR_ERR(handle);
2873                         ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2874                                "%ld pages, ino %lu; err %d", __func__,
2875                                 wbc->nr_to_write, inode->i_ino, ret);
2876                         /* Release allocated io_end */
2877                         ext4_put_io_end(mpd.io_submit.io_end);
2878                         mpd.io_submit.io_end = NULL;
2879                         break;
2880                 }
2881                 mpd.do_map = 1;
2882 
2883                 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2884                 ret = mpage_prepare_extent_to_map(&mpd);
2885                 if (!ret) {
2886                         if (mpd.map.m_len)
2887                                 ret = mpage_map_and_submit_extent(handle, &mpd,
2888                                         &give_up_on_write);
2889                         else {
2890                                 /*
2891                                  * We scanned the whole range (or exhausted
2892                                  * nr_to_write), submitted what was mapped and
2893                                  * didn't find anything needing mapping. We are
2894                                  * done.
2895                                  */
2896                                 done = true;
2897                         }
2898                 }
2899                 /*
2900                  * Caution: If the handle is synchronous,
2901                  * ext4_journal_stop() can wait for transaction commit
2902                  * to finish which may depend on writeback of pages to
2903                  * complete or on page lock to be released.  In that
2904                  * case, we have to wait until after after we have
2905                  * submitted all the IO, released page locks we hold,
2906                  * and dropped io_end reference (for extent conversion
2907                  * to be able to complete) before stopping the handle.
2908                  */
2909                 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2910                         ext4_journal_stop(handle);
2911                         handle = NULL;
2912                         mpd.do_map = 0;
2913                 }
2914                 /* Unlock pages we didn't use */
2915                 mpage_release_unused_pages(&mpd, give_up_on_write);
2916                 /* Submit prepared bio */
2917                 ext4_io_submit(&mpd.io_submit);
2918 
2919                 /*
2920                  * Drop our io_end reference we got from init. We have
2921                  * to be careful and use deferred io_end finishing if
2922                  * we are still holding the transaction as we can
2923                  * release the last reference to io_end which may end
2924                  * up doing unwritten extent conversion.
2925                  */
2926                 if (handle) {
2927                         ext4_put_io_end_defer(mpd.io_submit.io_end);
2928                         ext4_journal_stop(handle);
2929                 } else
2930                         ext4_put_io_end(mpd.io_submit.io_end);
2931                 mpd.io_submit.io_end = NULL;
2932 
2933                 if (ret == -ENOSPC && sbi->s_journal) {
2934                         /*
2935                          * Commit the transaction which would
2936                          * free blocks released in the transaction
2937                          * and try again
2938                          */
2939                         jbd2_journal_force_commit_nested(sbi->s_journal);
2940                         ret = 0;
2941                         continue;
2942                 }
2943                 /* Fatal error - ENOMEM, EIO... */
2944                 if (ret)
2945                         break;
2946         }
2947 unplug:
2948         blk_finish_plug(&plug);
2949         if (!ret && !cycled && wbc->nr_to_write > 0) {
2950                 cycled = 1;
2951                 mpd.last_page = writeback_index - 1;
2952                 mpd.first_page = 0;
2953                 goto retry;
2954         }
2955 
2956         /* Update index */
2957         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2958                 /*
2959                  * Set the writeback_index so that range_cyclic
2960                  * mode will write it back later
2961                  */
2962                 mapping->writeback_index = mpd.first_page;
2963 
2964 out_writepages:
2965         trace_ext4_writepages_result(inode, wbc, ret,
2966                                      nr_to_write - wbc->nr_to_write);
2967         percpu_up_read(&sbi->s_journal_flag_rwsem);
2968         return ret;
2969 }
2970 
2971 static int ext4_dax_writepages(struct address_space *mapping,
2972                                struct writeback_control *wbc)
2973 {
2974         int ret;
2975         long nr_to_write = wbc->nr_to_write;
2976         struct inode *inode = mapping->host;
2977         struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2978 
2979         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2980                 return -EIO;
2981 
2982         percpu_down_read(&sbi->s_journal_flag_rwsem);
2983         trace_ext4_writepages(inode, wbc);
2984 
2985         ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
2986         trace_ext4_writepages_result(inode, wbc, ret,
2987                                      nr_to_write - wbc->nr_to_write);
2988         percpu_up_read(&sbi->s_journal_flag_rwsem);
2989         return ret;
2990 }
2991 
2992 static int ext4_nonda_switch(struct super_block *sb)
2993 {
2994         s64 free_clusters, dirty_clusters;
2995         struct ext4_sb_info *sbi = EXT4_SB(sb);
2996 
2997         /*
2998          * switch to non delalloc mode if we are running low
2999          * on free block. The free block accounting via percpu
3000          * counters can get slightly wrong with percpu_counter_batch getting
3001          * accumulated on each CPU without updating global counters
3002          * Delalloc need an accurate free block accounting. So switch
3003          * to non delalloc when we are near to error range.
3004          */
3005         free_clusters =
3006                 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3007         dirty_clusters =
3008                 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3009         /*
3010          * Start pushing delalloc when 1/2 of free blocks are dirty.
3011          */
3012         if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3013                 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3014 
3015         if (2 * free_clusters < 3 * dirty_clusters ||
3016             free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3017                 /*
3018                  * free block count is less than 150% of dirty blocks
3019                  * or free blocks is less than watermark
3020                  */
3021                 return 1;
3022         }
3023         return 0;
3024 }
3025 
3026 /* We always reserve for an inode update; the superblock could be there too */
3027 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3028 {
3029         if (likely(ext4_has_feature_large_file(inode->i_sb)))
3030                 return 1;
3031 
3032         if (pos + len <= 0x7fffffffULL)
3033                 return 1;
3034 
3035         /* We might need to update the superblock to set LARGE_FILE */
3036         return 2;
3037 }
3038 
3039 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3040                                loff_t pos, unsigned len, unsigned flags,
3041                                struct page **pagep, void **fsdata)
3042 {
3043         int ret, retries = 0;
3044         struct page *page;
3045         pgoff_t index;
3046         struct inode *inode = mapping->host;
3047         handle_t *handle;
3048 
3049         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3050                 return -EIO;
3051 
3052         index = pos >> PAGE_SHIFT;
3053 
3054         if (ext4_nonda_switch(inode->i_sb) ||
3055             S_ISLNK(inode->i_mode)) {
3056                 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3057                 return ext4_write_begin(file, mapping, pos,
3058                                         len, flags, pagep, fsdata);
3059         }
3060         *fsdata = (void *)0;
3061         trace_ext4_da_write_begin(inode, pos, len, flags);
3062 
3063         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3064                 ret = ext4_da_write_inline_data_begin(mapping, inode,
3065                                                       pos, len, flags,
3066                                                       pagep, fsdata);
3067                 if (ret < 0)
3068                         return ret;
3069                 if (ret == 1)
3070                         return 0;
3071         }
3072 
3073         /*
3074          * grab_cache_page_write_begin() can take a long time if the
3075          * system is thrashing due to memory pressure, or if the page
3076          * is being written back.  So grab it first before we start
3077          * the transaction handle.  This also allows us to allocate
3078          * the page (if needed) without using GFP_NOFS.
3079          */
3080 retry_grab:
3081         page = grab_cache_page_write_begin(mapping, index, flags);
3082         if (!page)
3083                 return -ENOMEM;
3084         unlock_page(page);
3085 
3086         /*
3087          * With delayed allocation, we don't log the i_disksize update
3088          * if there is delayed block allocation. But we still need
3089          * to journalling the i_disksize update if writes to the end
3090          * of file which has an already mapped buffer.
3091          */
3092 retry_journal:
3093         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3094                                 ext4_da_write_credits(inode, pos, len));
3095         if (IS_ERR(handle)) {
3096                 put_page(page);
3097                 return PTR_ERR(handle);
3098         }
3099 
3100         lock_page(page);
3101         if (page->mapping != mapping) {
3102                 /* The page got truncated from under us */
3103                 unlock_page(page);
3104                 put_page(page);
3105                 ext4_journal_stop(handle);
3106                 goto retry_grab;
3107         }
3108         /* In case writeback began while the page was unlocked */
3109         wait_for_stable_page(page);
3110 
3111 #ifdef CONFIG_FS_ENCRYPTION
3112         ret = ext4_block_write_begin(page, pos, len,
3113                                      ext4_da_get_block_prep);
3114 #else
3115         ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3116 #endif
3117         if (ret < 0) {
3118                 unlock_page(page);
3119                 ext4_journal_stop(handle);
3120                 /*
3121                  * block_write_begin may have instantiated a few blocks
3122                  * outside i_size.  Trim these off again. Don't need
3123                  * i_size_read because we hold i_mutex.
3124                  */
3125                 if (pos + len > inode->i_size)
3126                         ext4_truncate_failed_write(inode);
3127 
3128                 if (ret == -ENOSPC &&
3129                     ext4_should_retry_alloc(inode->i_sb, &retries))
3130                         goto retry_journal;
3131 
3132                 put_page(page);
3133                 return ret;
3134         }
3135 
3136         *pagep = page;
3137         return ret;
3138 }
3139 
3140 /*
3141  * Check if we should update i_disksize
3142  * when write to the end of file but not require block allocation
3143  */
3144 static int ext4_da_should_update_i_disksize(struct page *page,
3145                                             unsigned long offset)
3146 {
3147         struct buffer_head *bh;
3148         struct inode *inode = page->mapping->host;
3149         unsigned int idx;
3150         int i;
3151 
3152         bh = page_buffers(page);
3153         idx = offset >> inode->i_blkbits;
3154 
3155         for (i = 0; i < idx; i++)
3156                 bh = bh->b_this_page;
3157 
3158         if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3159                 return 0;
3160         return 1;
3161 }
3162 
3163 static int ext4_da_write_end(struct file *file,
3164                              struct address_space *mapping,
3165                              loff_t pos, unsigned len, unsigned copied,
3166                              struct page *page, void *fsdata)
3167 {
3168         struct inode *inode = mapping->host;
3169         int ret = 0, ret2;
3170         handle_t *handle = ext4_journal_current_handle();
3171         loff_t new_i_size;
3172         unsigned long start, end;
3173         int write_mode = (int)(unsigned long)fsdata;
3174 
3175         if (write_mode == FALL_BACK_TO_NONDELALLOC)
3176                 return ext4_write_end(file, mapping, pos,
3177                                       len, copied, page, fsdata);
3178 
3179         trace_ext4_da_write_end(inode, pos, len, copied);
3180         start = pos & (PAGE_SIZE - 1);
3181         end = start + copied - 1;
3182 
3183         /*
3184          * generic_write_end() will run mark_inode_dirty() if i_size
3185          * changes.  So let's piggyback the i_disksize mark_inode_dirty
3186          * into that.
3187          */
3188         new_i_size = pos + copied;
3189         if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3190                 if (ext4_has_inline_data(inode) ||
3191                     ext4_da_should_update_i_disksize(page, end)) {
3192                         ext4_update_i_disksize(inode, new_i_size);
3193                         /* We need to mark inode dirty even if
3194                          * new_i_size is less that inode->i_size
3195                          * bu greater than i_disksize.(hint delalloc)
3196                          */
3197                         ext4_mark_inode_dirty(handle, inode);
3198                 }
3199         }
3200 
3201         if (write_mode != CONVERT_INLINE_DATA &&
3202             ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3203             ext4_has_inline_data(inode))
3204                 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3205                                                      page);
3206         else
3207                 ret2 = generic_write_end(file, mapping, pos, len, copied,
3208                                                         page, fsdata);
3209 
3210         copied = ret2;
3211         if (ret2 < 0)
3212                 ret = ret2;
3213         ret2 = ext4_journal_stop(handle);
3214         if (!ret)
3215                 ret = ret2;
3216 
3217         return ret ? ret : copied;
3218 }
3219 
3220 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3221                                    unsigned int length)
3222 {
3223         /*
3224          * Drop reserved blocks
3225          */
3226         BUG_ON(!PageLocked(page));
3227         if (!page_has_buffers(page))
3228                 goto out;
3229 
3230         ext4_da_page_release_reservation(page, offset, length);
3231 
3232 out:
3233         ext4_invalidatepage(page, offset, length);
3234 
3235         return;
3236 }
3237 
3238 /*
3239  * Force all delayed allocation blocks to be allocated for a given inode.
3240  */
3241 int ext4_alloc_da_blocks(struct inode *inode)
3242 {
3243         trace_ext4_alloc_da_blocks(inode);
3244 
3245         if (!EXT4_I(inode)->i_reserved_data_blocks)
3246                 return 0;
3247 
3248         /*
3249          * We do something simple for now.  The filemap_flush() will
3250          * also start triggering a write of the data blocks, which is
3251          * not strictly speaking necessary (and for users of
3252          * laptop_mode, not even desirable).  However, to do otherwise
3253          * would require replicating code paths in:
3254          *
3255          * ext4_writepages() ->
3256          *    write_cache_pages() ---> (via passed in callback function)
3257          *        __mpage_da_writepage() -->
3258          *           mpage_add_bh_to_extent()
3259          *           mpage_da_map_blocks()
3260          *
3261          * The problem is that write_cache_pages(), located in
3262          * mm/page-writeback.c, marks pages clean in preparation for
3263          * doing I/O, which is not desirable if we're not planning on
3264          * doing I/O at all.
3265          *
3266          * We could call write_cache_pages(), and then redirty all of
3267          * the pages by calling redirty_page_for_writepage() but that
3268          * would be ugly in the extreme.  So instead we would need to
3269          * replicate parts of the code in the above functions,
3270          * simplifying them because we wouldn't actually intend to
3271          * write out the pages, but rather only collect contiguous
3272          * logical block extents, call the multi-block allocator, and
3273          * then update the buffer heads with the block allocations.
3274          *
3275          * For now, though, we'll cheat by calling filemap_flush(),
3276          * which will map the blocks, and start the I/O, but not
3277          * actually wait for the I/O to complete.
3278          */
3279         return filemap_flush(inode->i_mapping);
3280 }
3281 
3282 /*
3283  * bmap() is special.  It gets used by applications such as lilo and by
3284  * the swapper to find the on-disk block of a specific piece of data.
3285  *
3286  * Naturally, this is dangerous if the block concerned is still in the
3287  * journal.  If somebody makes a swapfile on an ext4 data-journaling
3288  * filesystem and enables swap, then they may get a nasty shock when the
3289  * data getting swapped to that swapfile suddenly gets overwritten by
3290  * the original zero's written out previously to the journal and
3291  * awaiting writeback in the kernel's buffer cache.
3292  *
3293  * So, if we see any bmap calls here on a modified, data-journaled file,
3294  * take extra steps to flush any blocks which might be in the cache.
3295  */
3296 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3297 {
3298         struct inode *inode = mapping->host;
3299         journal_t *journal;
3300         int err;
3301 
3302         /*
3303          * We can get here for an inline file via the FIBMAP ioctl
3304          */
3305         if (ext4_has_inline_data(inode))
3306                 return 0;
3307 
3308         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3309                         test_opt(inode->i_sb, DELALLOC)) {
3310                 /*
3311                  * With delalloc we want to sync the file
3312                  * so that we can make sure we allocate
3313                  * blocks for file
3314                  */
3315                 filemap_write_and_wait(mapping);
3316         }
3317 
3318         if (EXT4_JOURNAL(inode) &&
3319             ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3320                 /*
3321                  * This is a REALLY heavyweight approach, but the use of
3322                  * bmap on dirty files is expected to be extremely rare:
3323                  * only if we run lilo or swapon on a freshly made file
3324                  * do we expect this to happen.
3325                  *
3326                  * (bmap requires CAP_SYS_RAWIO so this does not
3327                  * represent an unprivileged user DOS attack --- we'd be
3328                  * in trouble if mortal users could trigger this path at
3329                  * will.)
3330                  *
3331                  * NB. EXT4_STATE_JDATA is not set on files other than
3332                  * regular files.  If somebody wants to bmap a directory
3333                  * or symlink and gets confused because the buffer
3334                  * hasn't yet been flushed to disk, they deserve
3335                  * everything they get.
3336                  */
3337 
3338                 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3339                 journal = EXT4_JOURNAL(inode);
3340                 jbd2_journal_lock_updates(journal);
3341                 err = jbd2_journal_flush(journal);
3342                 jbd2_journal_unlock_updates(journal);
3343 
3344                 if (err)
3345                         return 0;
3346         }
3347 
3348         return generic_block_bmap(mapping, block, ext4_get_block);
3349 }
3350 
3351 static int ext4_readpage(struct file *file, struct page *page)
3352 {
3353         int ret = -EAGAIN;
3354         struct inode *inode = page->mapping->host;
3355 
3356         trace_ext4_readpage(page);
3357 
3358         if (ext4_has_inline_data(inode))
3359                 ret = ext4_readpage_inline(inode, page);
3360 
3361         if (ret == -EAGAIN)
3362                 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3363                                                 false);
3364 
3365         return ret;
3366 }
3367 
3368 static int
3369 ext4_readpages(struct file *file, struct address_space *mapping,
3370                 struct list_head *pages, unsigned nr_pages)
3371 {
3372         struct inode *inode = mapping->host;
3373 
3374         /* If the file has inline data, no need to do readpages. */
3375         if (ext4_has_inline_data(inode))
3376                 return 0;
3377 
3378         return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3379 }
3380 
3381 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3382                                 unsigned int length)
3383 {
3384         trace_ext4_invalidatepage(page, offset, length);
3385 
3386         /* No journalling happens on data buffers when this function is used */
3387         WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3388 
3389         block_invalidatepage(page, offset, length);
3390 }
3391 
3392 static int __ext4_journalled_invalidatepage(struct page *page,
3393                                             unsigned int offset,
3394                                             unsigned int length)
3395 {
3396         journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3397 
3398         trace_ext4_journalled_invalidatepage(page, offset, length);
3399 
3400         /*
3401          * If it's a full truncate we just forget about the pending dirtying
3402          */
3403         if (offset == 0 && length == PAGE_SIZE)
3404                 ClearPageChecked(page);
3405 
3406         return jbd2_journal_invalidatepage(journal, page, offset, length);
3407 }
3408 
3409 /* Wrapper for aops... */
3410 static void ext4_journalled_invalidatepage(struct page *page,
3411                                            unsigned int offset,
3412                                            unsigned int length)
3413 {
3414         WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3415 }
3416 
3417 static int ext4_releasepage(struct page *page, gfp_t wait)
3418 {
3419         journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3420 
3421         trace_ext4_releasepage(page);
3422 
3423         /* Page has dirty journalled data -> cannot release */
3424         if (PageChecked(page))
3425                 return 0;
3426         if (journal)
3427                 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3428         else
3429                 return try_to_free_buffers(page);
3430 }
3431 
3432 static bool ext4_inode_datasync_dirty(struct inode *inode)
3433 {
3434         journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3435 
3436         if (journal)
3437                 return !jbd2_transaction_committed(journal,
3438                                         EXT4_I(inode)->i_datasync_tid);
3439         /* Any metadata buffers to write? */
3440         if (!list_empty(&inode->i_mapping->private_list))
3441                 return true;
3442         return inode->i_state & I_DIRTY_DATASYNC;
3443 }
3444 
3445 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3446                             unsigned flags, struct iomap *iomap)
3447 {
3448         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3449         unsigned int blkbits = inode->i_blkbits;
3450         unsigned long first_block, last_block;
3451         struct ext4_map_blocks map;
3452         bool delalloc = false;
3453         int ret;
3454 
3455         if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3456                 return -EINVAL;
3457         first_block = offset >> blkbits;
3458         last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3459                            EXT4_MAX_LOGICAL_BLOCK);
3460 
3461         if (flags & IOMAP_REPORT) {
3462                 if (ext4_has_inline_data(inode)) {
3463                         ret = ext4_inline_data_iomap(inode, iomap);
3464                         if (ret != -EAGAIN) {
3465                                 if (ret == 0 && offset >= iomap->length)
3466                                         ret = -ENOENT;
3467                                 return ret;
3468                         }
3469                 }
3470         } else {
3471                 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3472                         return -ERANGE;
3473         }
3474 
3475         map.m_lblk = first_block;
3476         map.m_len = last_block - first_block + 1;
3477 
3478         if (flags & IOMAP_REPORT) {
3479                 ret = ext4_map_blocks(NULL, inode, &map, 0);
3480                 if (ret < 0)
3481                         return ret;
3482 
3483                 if (ret == 0) {
3484                         ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3485                         struct extent_status es;
3486 
3487                         ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3488                                                   map.m_lblk, end, &es);
3489 
3490                         if (!es.es_len || es.es_lblk > end) {
3491                                 /* entire range is a hole */
3492                         } else if (es.es_lblk > map.m_lblk) {
3493                                 /* range starts with a hole */
3494                                 map.m_len = es.es_lblk - map.m_lblk;
3495                         } else {
3496                                 ext4_lblk_t offs = 0;
3497 
3498                                 if (es.es_lblk < map.m_lblk)
3499                                         offs = map.m_lblk - es.es_lblk;
3500                                 map.m_lblk = es.es_lblk + offs;
3501                                 map.m_len = es.es_len - offs;
3502                                 delalloc = true;
3503                         }
3504                 }
3505         } else if (flags & IOMAP_WRITE) {
3506                 int dio_credits;
3507                 handle_t *handle;
3508                 int retries = 0;
3509 
3510                 /* Trim mapping request to maximum we can map at once for DIO */
3511                 if (map.m_len > DIO_MAX_BLOCKS)
3512                         map.m_len = DIO_MAX_BLOCKS;
3513                 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3514 retry:
3515                 /*
3516                  * Either we allocate blocks and then we don't get unwritten
3517                  * extent so we have reserved enough credits, or the blocks
3518                  * are already allocated and unwritten and in that case
3519                  * extent conversion fits in the credits as well.
3520                  */
3521                 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3522                                             dio_credits);
3523                 if (IS_ERR(handle))
3524                         return PTR_ERR(handle);
3525 
3526                 ret = ext4_map_blocks(handle, inode, &map,
3527                                       EXT4_GET_BLOCKS_CREATE_ZERO);
3528                 if (ret < 0) {
3529                         ext4_journal_stop(handle);
3530                         if (ret == -ENOSPC &&
3531                             ext4_should_retry_alloc(inode->i_sb, &retries))
3532                                 goto retry;
3533                         return ret;
3534                 }
3535 
3536                 /*
3537                  * If we added blocks beyond i_size, we need to make sure they
3538                  * will get truncated if we crash before updating i_size in
3539                  * ext4_iomap_end(). For faults we don't need to do that (and
3540                  * even cannot because for orphan list operations inode_lock is
3541                  * required) - if we happen to instantiate block beyond i_size,
3542                  * it is because we race with truncate which has already added
3543                  * the inode to the orphan list.
3544                  */
3545                 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3546                     (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3547                         int err;
3548 
3549                         err = ext4_orphan_add(handle, inode);
3550                         if (err < 0) {
3551                                 ext4_journal_stop(handle);
3552                                 return err;
3553                         }
3554                 }
3555                 ext4_journal_stop(handle);
3556         } else {
3557                 ret = ext4_map_blocks(NULL, inode, &map, 0);
3558                 if (ret < 0)
3559                         return ret;
3560         }
3561 
3562         iomap->flags = 0;
3563         if (ext4_inode_datasync_dirty(inode))
3564                 iomap->flags |= IOMAP_F_DIRTY;
3565         iomap->bdev = inode->i_sb->s_bdev;
3566         iomap->dax_dev = sbi->s_daxdev;
3567         iomap->offset = (u64)first_block << blkbits;
3568         iomap->length = (u64)map.m_len << blkbits;
3569 
3570         if (ret == 0) {
3571                 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3572                 iomap->addr = IOMAP_NULL_ADDR;
3573         } else {
3574                 if (map.m_flags & EXT4_MAP_MAPPED) {
3575                         iomap->type = IOMAP_MAPPED;
3576                 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3577                         iomap->type = IOMAP_UNWRITTEN;
3578                 } else {
3579                         WARN_ON_ONCE(1);
3580                         return -EIO;
3581                 }
3582                 iomap->addr = (u64)map.m_pblk << blkbits;
3583         }
3584 
3585         if (map.m_flags & EXT4_MAP_NEW)
3586                 iomap->flags |= IOMAP_F_NEW;
3587 
3588         return 0;
3589 }
3590 
3591 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3592                           ssize_t written, unsigned flags, struct iomap *iomap)
3593 {
3594         int ret = 0;
3595         handle_t *handle;
3596         int blkbits = inode->i_blkbits;
3597         bool truncate = false;
3598 
3599         if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3600                 return 0;
3601 
3602         handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3603         if (IS_ERR(handle)) {
3604                 ret = PTR_ERR(handle);
3605                 goto orphan_del;
3606         }
3607         if (ext4_update_inode_size(inode, offset + written))
3608                 ext4_mark_inode_dirty(handle, inode);
3609         /*
3610          * We may need to truncate allocated but not written blocks beyond EOF.
3611          */
3612         if (iomap->offset + iomap->length > 
3613             ALIGN(inode->i_size, 1 << blkbits)) {
3614                 ext4_lblk_t written_blk, end_blk;
3615 
3616                 written_blk = (offset + written) >> blkbits;
3617                 end_blk = (offset + length) >> blkbits;
3618                 if (written_blk < end_blk && ext4_can_truncate(inode))
3619                         truncate = true;
3620         }
3621         /*
3622          * Remove inode from orphan list if we were extending a inode and
3623          * everything went fine.
3624          */
3625         if (!truncate && inode->i_nlink &&
3626             !list_empty(&EXT4_I(inode)->i_orphan))
3627                 ext4_orphan_del(handle, inode);
3628         ext4_journal_stop(handle);
3629         if (truncate) {
3630                 ext4_truncate_failed_write(inode);
3631 orphan_del:
3632                 /*
3633                  * If truncate failed early the inode might still be on the
3634                  * orphan list; we need to make sure the inode is removed from
3635                  * the orphan list in that case.
3636                  */
3637                 if (inode->i_nlink)
3638                         ext4_orphan_del(NULL, inode);
3639         }
3640         return ret;
3641 }
3642 
3643 const struct iomap_ops ext4_iomap_ops = {
3644         .iomap_begin            = ext4_iomap_begin,
3645         .iomap_end              = ext4_iomap_end,
3646 };
3647 
3648 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3649                             ssize_t size, void *private)
3650 {
3651         ext4_io_end_t *io_end = private;
3652 
3653         /* if not async direct IO just return */
3654         if (!io_end)
3655                 return 0;
3656 
3657         ext_debug("ext4_end_io_dio(): io_end 0x%p "
3658                   "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3659                   io_end, io_end->inode->i_ino, iocb, offset, size);
3660 
3661         /*
3662          * Error during AIO DIO. We cannot convert unwritten extents as the
3663          * data was not written. Just clear the unwritten flag and drop io_end.
3664          */
3665         if (size <= 0) {
3666                 ext4_clear_io_unwritten_flag(io_end);
3667                 size = 0;
3668         }
3669         io_end->offset = offset;
3670         io_end->size = size;
3671         ext4_put_io_end(io_end);
3672 
3673         return 0;
3674 }
3675 
3676 /*
3677  * Handling of direct IO writes.
3678  *
3679  * For ext4 extent files, ext4 will do direct-io write even to holes,
3680  * preallocated extents, and those write extend the file, no need to
3681  * fall back to buffered IO.
3682  *
3683  * For holes, we fallocate those blocks, mark them as unwritten
3684  * If those blocks were preallocated, we mark sure they are split, but
3685  * still keep the range to write as unwritten.
3686  *
3687  * The unwritten extents will be converted to written when DIO is completed.
3688  * For async direct IO, since the IO may still pending when return, we
3689  * set up an end_io call back function, which will do the conversion
3690  * when async direct IO completed.
3691  *
3692  * If the O_DIRECT write will extend the file then add this inode to the
3693  * orphan list.  So recovery will truncate it back to the original size
3694  * if the machine crashes during the write.
3695  *
3696  */
3697 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3698 {
3699         struct file *file = iocb->ki_filp;
3700         struct inode *inode = file->f_mapping->host;
3701         struct ext4_inode_info *ei = EXT4_I(inode);
3702         ssize_t ret;
3703         loff_t offset = iocb->ki_pos;
3704         size_t count = iov_iter_count(iter);
3705         int overwrite = 0;
3706         get_block_t *get_block_func = NULL;
3707         int dio_flags = 0;
3708         loff_t final_size = offset + count;
3709         int orphan = 0;
3710         handle_t *handle;
3711 
3712         if (final_size > inode->i_size || final_size > ei->i_disksize) {
3713                 /* Credits for sb + inode write */
3714                 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3715                 if (IS_ERR(handle)) {
3716                         ret = PTR_ERR(handle);
3717                         goto out;
3718                 }
3719                 ret = ext4_orphan_add(handle, inode);
3720                 if (ret) {
3721                         ext4_journal_stop(handle);
3722                         goto out;
3723                 }
3724                 orphan = 1;
3725                 ext4_update_i_disksize(inode, inode->i_size);
3726                 ext4_journal_stop(handle);
3727         }
3728 
3729         BUG_ON(iocb->private == NULL);
3730 
3731         /*
3732          * Make all waiters for direct IO properly wait also for extent
3733          * conversion. This also disallows race between truncate() and
3734          * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3735          */
3736         inode_dio_begin(inode);
3737 
3738         /* If we do a overwrite dio, i_mutex locking can be released */
3739         overwrite = *((int *)iocb->private);
3740 
3741         if (overwrite)
3742                 inode_unlock(inode);
3743 
3744         /*
3745          * For extent mapped files we could direct write to holes and fallocate.
3746          *
3747          * Allocated blocks to fill the hole are marked as unwritten to prevent
3748          * parallel buffered read to expose the stale data before DIO complete
3749          * the data IO.
3750          *
3751          * As to previously fallocated extents, ext4 get_block will just simply
3752          * mark the buffer mapped but still keep the extents unwritten.
3753          *
3754          * For non AIO case, we will convert those unwritten extents to written
3755          * after return back from blockdev_direct_IO. That way we save us from
3756          * allocating io_end structure and also the overhead of offloading
3757          * the extent convertion to a workqueue.
3758          *
3759          * For async DIO, the conversion needs to be deferred when the
3760          * IO is completed. The ext4 end_io callback function will be
3761          * called to take care of the conversion work.  Here for async
3762          * case, we allocate an io_end structure to hook to the iocb.
3763          */
3764         iocb->private = NULL;
3765         if (overwrite)
3766                 get_block_func = ext4_dio_get_block_overwrite;
3767         else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3768                    round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3769                 get_block_func = ext4_dio_get_block;
3770                 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3771         } else if (is_sync_kiocb(iocb)) {
3772                 get_block_func = ext4_dio_get_block_unwritten_sync;
3773                 dio_flags = DIO_LOCKING;
3774         } else {
3775                 get_block_func = ext4_dio_get_block_unwritten_async;
3776                 dio_flags = DIO_LOCKING;
3777         }
3778         ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3779                                    get_block_func, ext4_end_io_dio, NULL,
3780                                    dio_flags);
3781 
3782         if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3783                                                 EXT4_STATE_DIO_UNWRITTEN)) {
3784                 int err;
3785                 /*
3786                  * for non AIO case, since the IO is already
3787                  * completed, we could do the conversion right here
3788                  */
3789                 err = ext4_convert_unwritten_extents(NULL, inode,
3790                                                      offset, ret);
3791                 if (err < 0)
3792                         ret = err;
3793                 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3794         }
3795 
3796         inode_dio_end(inode);
3797         /* take i_mutex locking again if we do a ovewrite dio */
3798         if (overwrite)
3799                 inode_lock(inode);
3800 
3801         if (ret < 0 && final_size > inode->i_size)
3802                 ext4_truncate_failed_write(inode);
3803 
3804         /* Handle extending of i_size after direct IO write */
3805         if (orphan) {
3806                 int err;
3807 
3808                 /* Credits for sb + inode write */
3809                 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3810                 if (IS_ERR(handle)) {
3811                         /*
3812                          * We wrote the data but cannot extend
3813                          * i_size. Bail out. In async io case, we do
3814                          * not return error here because we have
3815                          * already submmitted the corresponding
3816                          * bio. Returning error here makes the caller
3817                          * think that this IO is done and failed
3818                          * resulting in race with bio's completion
3819                          * handler.
3820                          */
3821                         if (!ret)
3822                                 ret = PTR_ERR(handle);
3823                         if (inode->i_nlink)
3824                                 ext4_orphan_del(NULL, inode);
3825 
3826                         goto out;
3827                 }
3828                 if (inode->i_nlink)
3829                         ext4_orphan_del(handle, inode);
3830                 if (ret > 0) {
3831                         loff_t end = offset + ret;
3832                         if (end > inode->i_size || end > ei->i_disksize) {
3833                                 ext4_update_i_disksize(inode, end);
3834                                 if (end > inode->i_size)
3835                                         i_size_write(inode, end);
3836                                 /*
3837                                  * We're going to return a positive `ret'
3838                                  * here due to non-zero-length I/O, so there's
3839                                  * no way of reporting error returns from
3840                                  * ext4_mark_inode_dirty() to userspace.  So
3841                                  * ignore it.
3842                                  */
3843                                 ext4_mark_inode_dirty(handle, inode);
3844                         }
3845                 }
3846                 err = ext4_journal_stop(handle);
3847                 if (ret == 0)
3848                         ret = err;
3849         }
3850 out:
3851         return ret;
3852 }
3853 
3854 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3855 {
3856         struct address_space *mapping = iocb->ki_filp->f_mapping;
3857         struct inode *inode = mapping->host;
3858         size_t count = iov_iter_count(iter);
3859         ssize_t ret;
3860 
3861         /*
3862          * Shared inode_lock is enough for us - it protects against concurrent
3863          * writes & truncates and since we take care of writing back page cache,
3864          * we are protected against page writeback as well.
3865          */
3866         inode_lock_shared(inode);
3867         ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3868                                            iocb->ki_pos + count - 1);
3869         if (ret)
3870                 goto out_unlock;
3871         ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3872                                    iter, ext4_dio_get_block, NULL, NULL, 0);
3873 out_unlock:
3874         inode_unlock_shared(inode);
3875         return ret;
3876 }
3877 
3878 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3879 {
3880         struct file *file = iocb->ki_filp;
3881         struct inode *inode = file->f_mapping->host;
3882         size_t count = iov_iter_count(iter);
3883         loff_t offset = iocb->ki_pos;
3884         ssize_t ret;
3885 
3886 #ifdef CONFIG_FS_ENCRYPTION
3887         if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
3888                 return 0;
3889 #endif
3890 
3891         /*
3892          * If we are doing data journalling we don't support O_DIRECT
3893          */
3894         if (ext4_should_journal_data(inode))
3895                 return 0;
3896 
3897         /* Let buffer I/O handle the inline data case. */
3898         if (ext4_has_inline_data(inode))
3899                 return 0;
3900 
3901         trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3902         if (iov_iter_rw(iter) == READ)
3903                 ret = ext4_direct_IO_read(iocb, iter);
3904         else
3905                 ret = ext4_direct_IO_write(iocb, iter);
3906         trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3907         return ret;
3908 }
3909 
3910 /*
3911  * Pages can be marked dirty completely asynchronously from ext4's journalling
3912  * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
3913  * much here because ->set_page_dirty is called under VFS locks.  The page is
3914  * not necessarily locked.
3915  *
3916  * We cannot just dirty the page and leave attached buffers clean, because the
3917  * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3918  * or jbddirty because all the journalling code will explode.
3919  *
3920  * So what we do is to mark the page "pending dirty" and next time writepage
3921  * is called, propagate that into the buffers appropriately.
3922  */
3923 static int ext4_journalled_set_page_dirty(struct page *page)
3924 {
3925         SetPageChecked(page);
3926         return __set_page_dirty_nobuffers(page);
3927 }
3928 
3929 static int ext4_set_page_dirty(struct page *page)
3930 {
3931         WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3932         WARN_ON_ONCE(!page_has_buffers(page));
3933         return __set_page_dirty_buffers(page);
3934 }
3935 
3936 static const struct address_space_operations ext4_aops = {
3937         .readpage               = ext4_readpage,
3938         .readpages              = ext4_readpages,
3939         .writepage              = ext4_writepage,
3940         .writepages             = ext4_writepages,
3941         .write_begin            = ext4_write_begin,
3942         .write_end              = ext4_write_end,
3943         .set_page_dirty         = ext4_set_page_dirty,
3944         .bmap                   = ext4_bmap,
3945         .invalidatepage         = ext4_invalidatepage,
3946         .releasepage            = ext4_releasepage,
3947         .direct_IO              = ext4_direct_IO,
3948         .migratepage            = buffer_migrate_page,
3949         .is_partially_uptodate  = block_is_partially_uptodate,
3950         .error_remove_page      = generic_error_remove_page,
3951 };
3952 
3953 static const struct address_space_operations ext4_journalled_aops = {
3954         .readpage               = ext4_readpage,
3955         .readpages              = ext4_readpages,
3956         .writepage              = ext4_writepage,
3957         .writepages             = ext4_writepages,
3958         .write_begin            = ext4_write_begin,
3959         .write_end              = ext4_journalled_write_end,
3960         .set_page_dirty         = ext4_journalled_set_page_dirty,
3961         .bmap                   = ext4_bmap,
3962         .invalidatepage         = ext4_journalled_invalidatepage,
3963         .releasepage            = ext4_releasepage,
3964         .direct_IO              = ext4_direct_IO,
3965         .is_partially_uptodate  = block_is_partially_uptodate,
3966         .error_remove_page      = generic_error_remove_page,
3967 };
3968 
3969 static const struct address_space_operations ext4_da_aops = {
3970         .readpage               = ext4_readpage,
3971         .readpages              = ext4_readpages,
3972         .writepage              = ext4_writepage,
3973         .writepages             = ext4_writepages,
3974         .write_begin            = ext4_da_write_begin,
3975         .write_end              = ext4_da_write_end,
3976         .set_page_dirty         = ext4_set_page_dirty,
3977         .bmap                   = ext4_bmap,
3978         .invalidatepage         = ext4_da_invalidatepage,
3979         .releasepage            = ext4_releasepage,
3980         .direct_IO              = ext4_direct_IO,
3981         .migratepage            = buffer_migrate_page,
3982         .is_partially_uptodate  = block_is_partially_uptodate,
3983         .error_remove_page      = generic_error_remove_page,
3984 };
3985 
3986 static const struct address_space_operations ext4_dax_aops = {
3987         .writepages             = ext4_dax_writepages,
3988         .direct_IO              = noop_direct_IO,
3989         .set_page_dirty         = noop_set_page_dirty,
3990         .bmap                   = ext4_bmap,
3991         .invalidatepage         = noop_invalidatepage,
3992 };
3993 
3994 void ext4_set_aops(struct inode *inode)
3995 {
3996         switch (ext4_inode_journal_mode(inode)) {
3997         case EXT4_INODE_ORDERED_DATA_MODE:
3998         case EXT4_INODE_WRITEBACK_DATA_MODE:
3999                 break;
4000         case EXT4_INODE_JOURNAL_DATA_MODE:
4001                 inode->i_mapping->a_ops = &ext4_journalled_aops;
4002                 return;
4003         default:
4004                 BUG();
4005         }
4006         if (IS_DAX(inode))
4007                 inode->i_mapping->a_ops = &ext4_dax_aops;
4008         else if (test_opt(inode->i_sb, DELALLOC))
4009                 inode->i_mapping->a_ops = &ext4_da_aops;
4010         else
4011                 inode->i_mapping->a_ops = &ext4_aops;
4012 }
4013 
4014 static int __ext4_block_zero_page_range(handle_t *handle,
4015                 struct address_space *mapping, loff_t from, loff_t length)
4016 {
4017         ext4_fsblk_t index = from >> PAGE_SHIFT;
4018         unsigned offset = from & (PAGE_SIZE-1);
4019         unsigned blocksize, pos;
4020         ext4_lblk_t iblock;
4021         struct inode *inode = mapping->host;
4022         struct buffer_head *bh;
4023         struct page *page;
4024         int err = 0;
4025 
4026         page = find_or_create_page(mapping, from >> PAGE_SHIFT,
4027                                    mapping_gfp_constraint(mapping, ~__GFP_FS));
4028         if (!page)
4029                 return -ENOMEM;
4030 
4031         blocksize = inode->i_sb->s_blocksize;
4032 
4033         iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4034 
4035         if (!page_has_buffers(page))
4036                 create_empty_buffers(page, blocksize, 0);
4037 
4038         /* Find the buffer that contains "offset" */
4039         bh = page_buffers(page);
4040         pos = blocksize;
4041         while (offset >= pos) {
4042                 bh = bh->b_this_page;
4043                 iblock++;
4044                 pos += blocksize;
4045         }
4046         if (buffer_freed(bh)) {
4047                 BUFFER_TRACE(bh, "freed: skip");
4048                 goto unlock;
4049         }
4050         if (!buffer_mapped(bh)) {
4051                 BUFFER_TRACE(bh, "unmapped");
4052                 ext4_get_block(inode, iblock, bh, 0);
4053                 /* unmapped? It's a hole - nothing to do */
4054                 if (!buffer_mapped(bh)) {
4055                         BUFFER_TRACE(bh, "still unmapped");
4056                         goto unlock;
4057                 }
4058         }
4059 
4060         /* Ok, it's mapped. Make sure it's up-to-date */
4061         if (PageUptodate(page))
4062                 set_buffer_uptodate(bh);
4063 
4064         if (!buffer_uptodate(bh)) {
4065                 err = -EIO;
4066                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4067                 wait_on_buffer(bh);
4068                 /* Uhhuh. Read error. Complain and punt. */
4069                 if (!buffer_uptodate(bh))
4070                         goto unlock;
4071                 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
4072                         /* We expect the key to be set. */
4073                         BUG_ON(!fscrypt_has_encryption_key(inode));
4074                         BUG_ON(blocksize != PAGE_SIZE);
4075                         WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4076                                                 page, PAGE_SIZE, 0, page->index));
4077                 }
4078         }
4079         if (ext4_should_journal_data(inode)) {
4080                 BUFFER_TRACE(bh, "get write access");
4081                 err = ext4_journal_get_write_access(handle, bh);
4082                 if (err)
4083                         goto unlock;
4084         }
4085         zero_user(page, offset, length);
4086         BUFFER_TRACE(bh, "zeroed end of block");
4087 
4088         if (ext4_should_journal_data(inode)) {
4089                 err = ext4_handle_dirty_metadata(handle, inode, bh);
4090         } else {
4091                 err = 0;
4092                 mark_buffer_dirty(bh);
4093                 if (ext4_should_order_data(inode))
4094                         err = ext4_jbd2_inode_add_write(handle, inode, from,
4095                                         length);
4096         }
4097 
4098 unlock:
4099         unlock_page(page);
4100         put_page(page);
4101         return err;
4102 }
4103 
4104 /*
4105  * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4106  * starting from file offset 'from'.  The range to be zero'd must
4107  * be contained with in one block.  If the specified range exceeds
4108  * the end of the block it will be shortened to end of the block
4109  * that cooresponds to 'from'
4110  */
4111 static int ext4_block_zero_page_range(handle_t *handle,
4112                 struct address_space *mapping, loff_t from, loff_t length)
4113 {
4114         struct inode *inode = mapping->host;
4115         unsigned offset = from & (PAGE_SIZE-1);
4116         unsigned blocksize = inode->i_sb->s_blocksize;
4117         unsigned max = blocksize - (offset & (blocksize - 1));
4118 
4119         /*
4120          * correct length if it does not fall between
4121          * 'from' and the end of the block
4122          */
4123         if (length > max || length < 0)
4124                 length = max;
4125 
4126         if (IS_DAX(inode)) {
4127                 return iomap_zero_range(inode, from, length, NULL,
4128                                         &ext4_iomap_ops);
4129         }
4130         return __ext4_block_zero_page_range(handle, mapping, from, length);
4131 }
4132 
4133 /*
4134  * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4135  * up to the end of the block which corresponds to `from'.
4136  * This required during truncate. We need to physically zero the tail end
4137  * of that block so it doesn't yield old data if the file is later grown.
4138  */
4139 static int ext4_block_truncate_page(handle_t *handle,
4140                 struct address_space *mapping, loff_t from)
4141 {
4142         unsigned offset = from & (PAGE_SIZE-1);
4143         unsigned length;
4144         unsigned blocksize;
4145         struct inode *inode = mapping->host;
4146 
4147         /* If we are processing an encrypted inode during orphan list handling */
4148         if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
4149                 return 0;
4150 
4151         blocksize = inode->i_sb->s_blocksize;
4152         length = blocksize - (offset & (blocksize - 1));
4153 
4154         return ext4_block_zero_page_range(handle, mapping, from, length);
4155 }
4156 
4157 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4158                              loff_t lstart, loff_t length)
4159 {
4160         struct super_block *sb = inode->i_sb;
4161         struct address_space *mapping = inode->i_mapping;
4162         unsigned partial_start, partial_end;
4163         ext4_fsblk_t start, end;
4164         loff_t byte_end = (lstart + length - 1);
4165         int err = 0;
4166 
4167         partial_start = lstart & (sb->s_blocksize - 1);
4168         partial_end = byte_end & (sb->s_blocksize - 1);
4169 
4170         start = lstart >> sb->s_blocksize_bits;
4171         end = byte_end >> sb->s_blocksize_bits;
4172 
4173         /* Handle partial zero within the single block */
4174         if (start == end &&
4175             (partial_start || (partial_end != sb->s_blocksize - 1))) {
4176                 err = ext4_block_zero_page_range(handle, mapping,
4177                                                  lstart, length);
4178                 return err;
4179         }
4180         /* Handle partial zero out on the start of the range */
4181         if (partial_start) {
4182                 err = ext4_block_zero_page_range(handle, mapping,
4183                                                  lstart, sb->s_blocksize);
4184                 if (err)
4185                         return err;
4186         }
4187         /* Handle partial zero out on the end of the range */
4188         if (partial_end != sb->s_blocksize - 1)
4189                 err = ext4_block_zero_page_range(handle, mapping,
4190                                                  byte_end - partial_end,
4191                                                  partial_end + 1);
4192         return err;
4193 }
4194 
4195 int ext4_can_truncate(struct inode *inode)
4196 {
4197         if (S_ISREG(inode->i_mode))
4198                 return 1;
4199         if (S_ISDIR(inode->i_mode))
4200                 return 1;
4201         if (S_ISLNK(inode->i_mode))
4202                 return !ext4_inode_is_fast_symlink(inode);
4203         return 0;
4204 }
4205 
4206 /*
4207  * We have to make sure i_disksize gets properly updated before we truncate
4208  * page cache due to hole punching or zero range. Otherwise i_disksize update
4209  * can get lost as it may have been postponed to submission of writeback but
4210  * that will never happen after we truncate page cache.
4211  */
4212 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4213                                       loff_t len)
4214 {
4215         handle_t *handle;
4216         loff_t size = i_size_read(inode);
4217 
4218         WARN_ON(!inode_is_locked(inode));
4219         if (offset > size || offset + len < size)
4220                 return 0;
4221 
4222         if (EXT4_I(inode)->i_disksize >= size)
4223                 return 0;
4224 
4225         handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4226         if (IS_ERR(handle))
4227                 return PTR_ERR(handle);
4228         ext4_update_i_disksize(inode, size);
4229         ext4_mark_inode_dirty(handle, inode);
4230         ext4_journal_stop(handle);
4231 
4232         return 0;
4233 }
4234 
4235 static void ext4_wait_dax_page(struct ext4_inode_info *ei)
4236 {
4237         up_write(&ei->i_mmap_sem);
4238         schedule();
4239         down_write(&ei->i_mmap_sem);
4240 }
4241 
4242 int ext4_break_layouts(struct inode *inode)
4243 {
4244         struct ext4_inode_info *ei = EXT4_I(inode);
4245         struct page *page;
4246         int error;
4247 
4248         if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4249                 return -EINVAL;
4250 
4251         do {
4252                 page = dax_layout_busy_page(inode->i_mapping);
4253                 if (!page)
4254                         return 0;
4255 
4256                 error = ___wait_var_event(&page->_refcount,
4257                                 atomic_read(&page->_refcount) == 1,
4258                                 TASK_INTERRUPTIBLE, 0, 0,
4259                                 ext4_wait_dax_page(ei));
4260         } while (error == 0);
4261 
4262         return error;
4263 }
4264 
4265 /*
4266  * ext4_punch_hole: punches a hole in a file by releasing the blocks
4267  * associated with the given offset and length
4268  *
4269  * @inode:  File inode
4270  * @offset: The offset where the hole will begin
4271  * @len:    The length of the hole
4272  *
4273  * Returns: 0 on success or negative on failure
4274  */
4275 
4276 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4277 {
4278         struct super_block *sb = inode->i_sb;
4279         ext4_lblk_t first_block, stop_block;
4280         struct address_space *mapping = inode->i_mapping;
4281         loff_t first_block_offset, last_block_offset;
4282         handle_t *handle;
4283         unsigned int credits;
4284         int ret = 0;
4285 
4286         if (!S_ISREG(inode->i_mode))
4287                 return -EOPNOTSUPP;
4288 
4289         trace_ext4_punch_hole(inode, offset, length, 0);
4290 
4291         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4292         if (ext4_has_inline_data(inode)) {
4293                 down_write(&EXT4_I(inode)->i_mmap_sem);
4294                 ret = ext4_convert_inline_data(inode);
4295                 up_write(&EXT4_I(inode)->i_mmap_sem);
4296                 if (ret)
4297                         return ret;
4298         }
4299 
4300         /*
4301          * Write out all dirty pages to avoid race conditions
4302          * Then release them.
4303          */
4304         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4305                 ret = filemap_write_and_wait_range(mapping, offset,
4306                                                    offset + length - 1);
4307                 if (ret)
4308                         return ret;
4309         }
4310 
4311         inode_lock(inode);
4312 
4313         /* No need to punch hole beyond i_size */
4314         if (offset >= inode->i_size)
4315                 goto out_mutex;
4316 
4317         /*
4318          * If the hole extends beyond i_size, set the hole
4319          * to end after the page that contains i_size
4320          */
4321         if (offset + length > inode->i_size) {
4322                 length = inode->i_size +
4323                    PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4324                    offset;
4325         }
4326 
4327         if (offset & (sb->s_blocksize - 1) ||
4328             (offset + length) & (sb->s_blocksize - 1)) {
4329                 /*
4330                  * Attach jinode to inode for jbd2 if we do any zeroing of
4331                  * partial block
4332                  */
4333                 ret = ext4_inode_attach_jinode(inode);
4334                 if (ret < 0)
4335                         goto out_mutex;
4336 
4337         }
4338 
4339         /* Wait all existing dio workers, newcomers will block on i_mutex */
4340         inode_dio_wait(inode);
4341 
4342         /*
4343          * Prevent page faults from reinstantiating pages we have released from
4344          * page cache.
4345          */
4346         down_write(&EXT4_I(inode)->i_mmap_sem);
4347 
4348         ret = ext4_break_layouts(inode);
4349         if (ret)
4350                 goto out_dio;
4351 
4352         first_block_offset = round_up(offset, sb->s_blocksize);
4353         last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4354 
4355         /* Now release the pages and zero block aligned part of pages*/
4356         if (last_block_offset > first_block_offset) {
4357                 ret = ext4_update_disksize_before_punch(inode, offset, length);
4358                 if (ret)
4359                         goto out_dio;
4360                 truncate_pagecache_range(inode, first_block_offset,
4361                                          last_block_offset);
4362         }
4363 
4364         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4365                 credits = ext4_writepage_trans_blocks(inode);
4366         else
4367                 credits = ext4_blocks_for_truncate(inode);
4368         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4369         if (IS_ERR(handle)) {
4370                 ret = PTR_ERR(handle);
4371                 ext4_std_error(sb, ret);
4372                 goto out_dio;
4373         }
4374 
4375         ret = ext4_zero_partial_blocks(handle, inode, offset,
4376                                        length);
4377         if (ret)
4378                 goto out_stop;
4379 
4380         first_block = (offset + sb->s_blocksize - 1) >>
4381                 EXT4_BLOCK_SIZE_BITS(sb);
4382         stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4383 
4384         /* If there are blocks to remove, do it */
4385         if (stop_block > first_block) {
4386 
4387                 down_write(&EXT4_I(inode)->i_data_sem);
4388                 ext4_discard_preallocations(inode);
4389 
4390                 ret = ext4_es_remove_extent(inode, first_block,
4391                                             stop_block - first_block);
4392                 if (ret) {
4393                         up_write(&EXT4_I(inode)->i_data_sem);
4394                         goto out_stop;
4395                 }
4396 
4397                 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4398                         ret = ext4_ext_remove_space(inode, first_block,
4399                                                     stop_block - 1);
4400                 else
4401                         ret = ext4_ind_remove_space(handle, inode, first_block,
4402                                                     stop_block);
4403 
4404                 up_write(&EXT4_I(inode)->i_data_sem);
4405         }
4406         if (IS_SYNC(inode))
4407                 ext4_handle_sync(handle);
4408 
4409         inode->i_mtime = inode->i_ctime = current_time(inode);
4410         ext4_mark_inode_dirty(handle, inode);
4411         if (ret >= 0)
4412                 ext4_update_inode_fsync_trans(handle, inode, 1);
4413 out_stop:
4414         ext4_journal_stop(handle);
4415 out_dio:
4416         up_write(&EXT4_I(inode)->i_mmap_sem);
4417 out_mutex:
4418         inode_unlock(inode);
4419         return ret;
4420 }
4421 
4422 int ext4_inode_attach_jinode(struct inode *inode)
4423 {
4424         struct ext4_inode_info *ei = EXT4_I(inode);
4425         struct jbd2_inode *jinode;
4426 
4427         if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4428                 return 0;
4429 
4430         jinode = jbd2_alloc_inode(GFP_KERNEL);
4431         spin_lock(&inode->i_lock);
4432         if (!ei->jinode) {
4433                 if (!jinode) {
4434                         spin_unlock(&inode->i_lock);
4435                         return -ENOMEM;
4436                 }
4437                 ei->jinode = jinode;
4438                 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4439                 jinode = NULL;
4440         }
4441         spin_unlock(&inode->i_lock);
4442         if (unlikely(jinode != NULL))
4443                 jbd2_free_inode(jinode);
4444         return 0;
4445 }
4446 
4447 /*
4448  * ext4_truncate()
4449  *
4450  * We block out ext4_get_block() block instantiations across the entire
4451  * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4452  * simultaneously on behalf of the same inode.
4453  *
4454  * As we work through the truncate and commit bits of it to the journal there
4455  * is one core, guiding principle: the file's tree must always be consistent on
4456  * disk.  We must be able to restart the truncate after a crash.
4457  *
4458  * The file's tree may be transiently inconsistent in memory (although it
4459  * probably isn't), but whenever we close off and commit a journal transaction,
4460  * the contents of (the filesystem + the journal) must be consistent and
4461  * restartable.  It's pretty simple, really: bottom up, right to left (although
4462  * left-to-right works OK too).
4463  *
4464  * Note that at recovery time, journal replay occurs *before* the restart of
4465  * truncate against the orphan inode list.
4466  *
4467  * The committed inode has the new, desired i_size (which is the same as
4468  * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
4469  * that this inode's truncate did not complete and it will again call
4470  * ext4_truncate() to have another go.  So there will be instantiated blocks
4471  * to the right of the truncation point in a crashed ext4 filesystem.  But
4472  * that's fine - as long as they are linked from the inode, the post-crash
4473  * ext4_truncate() run will find them and release them.
4474  */
4475 int ext4_truncate(struct inode *inode)
4476 {
4477         struct ext4_inode_info *ei = EXT4_I(inode);
4478         unsigned int credits;
4479         int err = 0;
4480         handle_t *handle;
4481         struct address_space *mapping = inode->i_mapping;
4482 
4483         /*
4484          * There is a possibility that we're either freeing the inode
4485          * or it's a completely new inode. In those cases we might not
4486          * have i_mutex locked because it's not necessary.
4487          */
4488         if (!(inode->i_state & (I_NEW|I_FREEING)))
4489                 WARN_ON(!inode_is_locked(inode));
4490         trace_ext4_truncate_enter(inode);
4491 
4492         if (!ext4_can_truncate(inode))
4493                 return 0;
4494 
4495         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4496 
4497         if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4498                 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4499 
4500         if (ext4_has_inline_data(inode)) {
4501                 int has_inline = 1;
4502 
4503                 err = ext4_inline_data_truncate(inode, &has_inline);
4504                 if (err)
4505                         return err;
4506                 if (has_inline)
4507                         return 0;
4508         }
4509 
4510         /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4511         if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4512                 if (ext4_inode_attach_jinode(inode) < 0)
4513                         return 0;
4514         }
4515 
4516         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4517                 credits = ext4_writepage_trans_blocks(inode);
4518         else
4519                 credits = ext4_blocks_for_truncate(inode);
4520 
4521         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4522         if (IS_ERR(handle))
4523                 return PTR_ERR(handle);
4524 
4525         if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4526                 ext4_block_truncate_page(handle, mapping, inode->i_size);
4527 
4528         /*
4529          * We add the inode to the orphan list, so that if this
4530          * truncate spans multiple transactions, and we crash, we will
4531          * resume the truncate when the filesystem recovers.  It also
4532          * marks the inode dirty, to catch the new size.
4533          *
4534          * Implication: the file must always be in a sane, consistent
4535          * truncatable state while each transaction commits.
4536          */
4537         err = ext4_orphan_add(handle, inode);
4538         if (err)
4539                 goto out_stop;
4540 
4541         down_write(&EXT4_I(inode)->i_data_sem);
4542 
4543         ext4_discard_preallocations(inode);
4544 
4545         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4546                 err = ext4_ext_truncate(handle, inode);
4547         else
4548                 ext4_ind_truncate(handle, inode);
4549 
4550         up_write(&ei->i_data_sem);
4551         if (err)
4552                 goto out_stop;
4553 
4554         if (IS_SYNC(inode))
4555                 ext4_handle_sync(handle);
4556 
4557 out_stop:
4558         /*
4559          * If this was a simple ftruncate() and the file will remain alive,
4560          * then we need to clear up the orphan record which we created above.
4561          * However, if this was a real unlink then we were called by
4562          * ext4_evict_inode(), and we allow that function to clean up the
4563          * orphan info for us.
4564          */
4565         if (inode->i_nlink)
4566                 ext4_orphan_del(handle, inode);
4567 
4568         inode->i_mtime = inode->i_ctime = current_time(inode);
4569         ext4_mark_inode_dirty(handle, inode);
4570         ext4_journal_stop(handle);
4571 
4572         trace_ext4_truncate_exit(inode);
4573         return err;
4574 }
4575 
4576 /*
4577  * ext4_get_inode_loc returns with an extra refcount against the inode's
4578  * underlying buffer_head on success. If 'in_mem' is true, we have all
4579  * data in memory that is needed to recreate the on-disk version of this
4580  * inode.
4581  */
4582 static int __ext4_get_inode_loc(struct inode *inode,
4583                                 struct ext4_iloc *iloc, int in_mem)
4584 {
4585         struct ext4_group_desc  *gdp;
4586         struct buffer_head      *bh;
4587         struct super_block      *sb = inode->i_sb;
4588         ext4_fsblk_t            block;
4589         int                     inodes_per_block, inode_offset;
4590 
4591         iloc->bh = NULL;
4592         if (inode->i_ino < EXT4_ROOT_INO ||
4593             inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4594                 return -EFSCORRUPTED;
4595 
4596         iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4597         gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4598         if (!gdp)
4599                 return -EIO;
4600 
4601         /*
4602          * Figure out the offset within the block group inode table
4603          */
4604         inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4605         inode_offset = ((inode->i_ino - 1) %
4606                         EXT4_INODES_PER_GROUP(sb));
4607         block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4608         iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4609 
4610         bh = sb_getblk(sb, block);
4611         if (unlikely(!bh))
4612                 return -ENOMEM;
4613         if (!buffer_uptodate(bh)) {
4614                 lock_buffer(bh);
4615 
4616                 /*
4617                  * If the buffer has the write error flag, we have failed
4618                  * to write out another inode in the same block.  In this
4619                  * case, we don't have to read the block because we may
4620                  * read the old inode data successfully.
4621                  */
4622                 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4623                         set_buffer_uptodate(bh);
4624 
4625                 if (buffer_uptodate(bh)) {
4626                         /* someone brought it uptodate while we waited */
4627                         unlock_buffer(bh);
4628                         goto has_buffer;
4629                 }
4630 
4631                 /*
4632                  * If we have all information of the inode in memory and this
4633                  * is the only valid inode in the block, we need not read the
4634                  * block.
4635                  */
4636                 if (in_mem) {
4637                         struct buffer_head *bitmap_bh;
4638                         int i, start;
4639 
4640                         start = inode_offset & ~(inodes_per_block - 1);
4641 
4642                         /* Is the inode bitmap in cache? */
4643                         bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4644                         if (unlikely(!bitmap_bh))
4645                                 goto make_io;
4646 
4647                         /*
4648                          * If the inode bitmap isn't in cache then the
4649                          * optimisation may end up performing two reads instead
4650                          * of one, so skip it.
4651                          */
4652                         if (!buffer_uptodate(bitmap_bh)) {
4653                                 brelse(bitmap_bh);
4654                                 goto make_io;
4655                         }
4656                         for (i = start; i < start + inodes_per_block; i++) {
4657                                 if (i == inode_offset)
4658                                         continue;
4659                                 if (ext4_test_bit(i, bitmap_bh->b_data))
4660                                         break;
4661                         }
4662                         brelse(bitmap_bh);
4663                         if (i == start + inodes_per_block) {
4664                                 /* all other inodes are free, so skip I/O */
4665                                 memset(bh->b_data, 0, bh->b_size);
4666                                 set_buffer_uptodate(bh);
4667                                 unlock_buffer(bh);
4668                                 goto has_buffer;
4669                         }
4670                 }
4671 
4672 make_io:
4673                 /*
4674                  * If we need to do any I/O, try to pre-readahead extra
4675                  * blocks from the inode table.
4676                  */
4677                 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4678                         ext4_fsblk_t b, end, table;
4679                         unsigned num;
4680                         __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4681 
4682                         table = ext4_inode_table(sb, gdp);
4683                         /* s_inode_readahead_blks is always a power of 2 */
4684                         b = block & ~((ext4_fsblk_t) ra_blks - 1);
4685                         if (table > b)
4686                                 b = table;
4687                         end = b + ra_blks;
4688                         num = EXT4_INODES_PER_GROUP(sb);
4689                         if (ext4_has_group_desc_csum(sb))
4690                                 num -= ext4_itable_unused_count(sb, gdp);
4691                         table += num / inodes_per_block;
4692                         if (end > table)
4693                                 end = table;
4694                         while (b <= end)
4695                                 sb_breadahead(sb, b++);
4696                 }
4697 
4698                 /*
4699                  * There are other valid inodes in the buffer, this inode
4700                  * has in-inode xattrs, or we don't have this inode in memory.
4701                  * Read the block from disk.
4702                  */
4703                 trace_ext4_load_inode(inode);
4704                 get_bh(bh);
4705                 bh->b_end_io = end_buffer_read_sync;
4706                 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4707                 wait_on_buffer(bh);
4708                 if (!buffer_uptodate(bh)) {
4709                         EXT4_ERROR_INODE_BLOCK(inode, block,
4710                                                "unable to read itable block");
4711                         brelse(bh);
4712                         return -EIO;
4713                 }
4714         }
4715 has_buffer:
4716         iloc->bh = bh;
4717         return 0;
4718 }
4719 
4720 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4721 {
4722         /* We have all inode data except xattrs in memory here. */
4723         return __ext4_get_inode_loc(inode, iloc,
4724                 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4725 }
4726 
4727 static bool ext4_should_use_dax(struct inode *inode)
4728 {
4729         if (!test_opt(inode->i_sb, DAX))
4730                 return false;
4731         if (!S_ISREG(inode->i_mode))
4732                 return false;
4733         if (ext4_should_journal_data(inode))
4734                 return false;
4735         if (ext4_has_inline_data(inode))
4736                 return false;
4737         if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4738                 return false;
4739         return true;
4740 }
4741 
4742 void ext4_set_inode_flags(struct inode *inode)
4743 {
4744         unsigned int flags = EXT4_I(inode)->i_flags;
4745         unsigned int new_fl = 0;
4746 
4747         if (flags & EXT4_SYNC_FL)
4748                 new_fl |= S_SYNC;
4749         if (flags & EXT4_APPEND_FL)
4750                 new_fl |= S_APPEND;
4751         if (flags & EXT4_IMMUTABLE_FL)
4752                 new_fl |= S_IMMUTABLE;
4753         if (flags & EXT4_NOATIME_FL)
4754                 new_fl |= S_NOATIME;
4755         if (flags & EXT4_DIRSYNC_FL)
4756                 new_fl |= S_DIRSYNC;
4757         if (ext4_should_use_dax(inode))
4758                 new_fl |= S_DAX;
4759         if (flags & EXT4_ENCRYPT_FL)
4760                 new_fl |= S_ENCRYPTED;
4761         if (flags & EXT4_CASEFOLD_FL)
4762                 new_fl |= S_CASEFOLD;
4763         inode_set_flags(inode, new_fl,
4764                         S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4765                         S_ENCRYPTED|S_CASEFOLD);
4766 }
4767 
4768 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4769                                   struct ext4_inode_info *ei)
4770 {
4771         blkcnt_t i_blocks ;
4772         struct inode *inode = &(ei->vfs_inode);
4773         struct super_block *sb = inode->i_sb;
4774 
4775         if (ext4_has_feature_huge_file(sb)) {
4776                 /* we are using combined 48 bit field */
4777                 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4778                                         le32_to_cpu(raw_inode->i_blocks_lo);
4779                 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4780                         /* i_blocks represent file system block size */
4781                         return i_blocks  << (inode->i_blkbits - 9);
4782                 } else {
4783                         return i_blocks;
4784                 }
4785         } else {
4786                 return le32_to_cpu(raw_inode->i_blocks_lo);
4787         }
4788 }
4789 
4790 static inline int ext4_iget_extra_inode(struct inode *inode,
4791                                          struct ext4_inode *raw_inode,
4792                                          struct ext4_inode_info *ei)
4793 {
4794         __le32 *magic = (void *)raw_inode +
4795                         EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4796 
4797         if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4798             EXT4_INODE_SIZE(inode->i_sb) &&
4799             *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4800                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4801                 return ext4_find_inline_data_nolock(inode);
4802         } else
4803                 EXT4_I(inode)->i_inline_off = 0;
4804         return 0;
4805 }
4806 
4807 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4808 {
4809         if (!ext4_has_feature_project(inode->i_sb))
4810                 return -EOPNOTSUPP;
4811         *projid = EXT4_I(inode)->i_projid;
4812         return 0;
4813 }
4814 
4815 /*
4816  * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4817  * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4818  * set.
4819  */
4820 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4821 {
4822         if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4823                 inode_set_iversion_raw(inode, val);
4824         else
4825                 inode_set_iversion_queried(inode, val);
4826 }
4827 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4828 {
4829         if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4830                 return inode_peek_iversion_raw(inode);
4831         else
4832                 return inode_peek_iversion(inode);
4833 }
4834 
4835 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4836                           ext4_iget_flags flags, const char *function,
4837                           unsigned int line)
4838 {
4839         struct ext4_iloc iloc;
4840         struct ext4_inode *raw_inode;
4841         struct ext4_inode_info *ei;
4842         struct inode *inode;
4843         journal_t *journal = EXT4_SB(sb)->s_journal;
4844         long ret;
4845         loff_t size;
4846         int block;
4847         uid_t i_uid;
4848         gid_t i_gid;
4849         projid_t i_projid;
4850 
4851         if ((!(flags & EXT4_IGET_SPECIAL) &&
4852              (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4853             (ino < EXT4_ROOT_INO) ||
4854             (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4855                 if (flags & EXT4_IGET_HANDLE)
4856                         return ERR_PTR(-ESTALE);
4857                 __ext4_error(sb, function, line,
4858                              "inode #%lu: comm %s: iget: illegal inode #",
4859                              ino, current->comm);
4860                 return ERR_PTR(-EFSCORRUPTED);
4861         }
4862 
4863         inode = iget_locked(sb, ino);
4864         if (!inode)
4865                 return ERR_PTR(-ENOMEM);
4866         if (!(inode->i_state & I_NEW))
4867                 return inode;
4868 
4869         ei = EXT4_I(inode);
4870         iloc.bh = NULL;
4871 
4872         ret = __ext4_get_inode_loc(inode, &iloc, 0);
4873         if (ret < 0)
4874                 goto bad_inode;
4875         raw_inode = ext4_raw_inode(&iloc);
4876 
4877         if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4878                 ext4_error_inode(inode, function, line, 0,
4879                                  "iget: root inode unallocated");
4880                 ret = -EFSCORRUPTED;
4881                 goto bad_inode;
4882         }
4883 
4884         if ((flags & EXT4_IGET_HANDLE) &&
4885             (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4886                 ret = -ESTALE;
4887                 goto bad_inode;
4888         }
4889 
4890         if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4891                 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4892                 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4893                         EXT4_INODE_SIZE(inode->i_sb) ||
4894                     (ei->i_extra_isize & 3)) {
4895                         ext4_error_inode(inode, function, line, 0,
4896                                          "iget: bad extra_isize %u "
4897                                          "(inode size %u)",
4898                                          ei->i_extra_isize,
4899                                          EXT4_INODE_SIZE(inode->i_sb));
4900                         ret = -EFSCORRUPTED;
4901                         goto bad_inode;
4902                 }
4903         } else
4904                 ei->i_extra_isize = 0;
4905 
4906         /* Precompute checksum seed for inode metadata */
4907         if (ext4_has_metadata_csum(sb)) {
4908                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4909                 __u32 csum;
4910                 __le32 inum = cpu_to_le32(inode->i_ino);
4911                 __le32 gen = raw_inode->i_generation;
4912                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4913                                    sizeof(inum));
4914                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4915                                               sizeof(gen));
4916         }
4917 
4918         if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4919                 ext4_error_inode(inode, function, line, 0,
4920                                  "iget: checksum invalid");
4921                 ret = -EFSBADCRC;
4922                 goto bad_inode;
4923         }
4924 
4925         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4926         i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4927         i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4928         if (ext4_has_feature_project(sb) &&
4929             EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4930             EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4931                 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4932         else
4933                 i_projid = EXT4_DEF_PROJID;
4934 
4935         if (!(test_opt(inode->i_sb, NO_UID32))) {
4936                 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4937                 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4938         }
4939         i_uid_write(inode, i_uid);
4940         i_gid_write(inode, i_gid);
4941         ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4942         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4943 
4944         ext4_clear_state_flags(ei);     /* Only relevant on 32-bit archs */
4945         ei->i_inline_off = 0;
4946         ei->i_dir_start_lookup = 0;
4947         ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4948         /* We now have enough fields to check if the inode was active or not.
4949          * This is needed because nfsd might try to access dead inodes
4950          * the test is that same one that e2fsck uses
4951          * NeilBrown 1999oct15
4952          */
4953         if (inode->i_nlink == 0) {
4954                 if ((inode->i_mode == 0 ||
4955                      !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4956                     ino != EXT4_BOOT_LOADER_INO) {
4957                         /* this inode is deleted */
4958                         ret = -ESTALE;
4959                         goto bad_inode;
4960                 }
4961                 /* The only unlinked inodes we let through here have
4962                  * valid i_mode and are being read by the orphan
4963                  * recovery code: that's fine, we're about to complete
4964                  * the process of deleting those.
4965                  * OR it is the EXT4_BOOT_LOADER_INO which is
4966                  * not initialized on a new filesystem. */
4967         }
4968         ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4969         ext4_set_inode_flags(inode);
4970         inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4971         ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4972         if (ext4_has_feature_64bit(sb))
4973                 ei->i_file_acl |=
4974                         ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4975         inode->i_size = ext4_isize(sb, raw_inode);
4976         if ((size = i_size_read(inode)) < 0) {
4977                 ext4_error_inode(inode, function, line, 0,
4978                                  "iget: bad i_size value: %lld", size);
4979                 ret = -EFSCORRUPTED;
4980                 goto bad_inode;
4981         }
4982         ei->i_disksize = inode->i_size;
4983 #ifdef CONFIG_QUOTA
4984         ei->i_reserved_quota = 0;
4985 #endif
4986         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4987         ei->i_block_group = iloc.block_group;
4988         ei->i_last_alloc_group = ~0;
4989         /*
4990          * NOTE! The in-memory inode i_data array is in little-endian order
4991          * even on big-endian machines: we do NOT byteswap the block numbers!
4992          */
4993         for (block = 0; block < EXT4_N_BLOCKS; block++)
4994                 ei->i_data[block] = raw_inode->i_block[block];
4995         INIT_LIST_HEAD(&ei->i_orphan);
4996 
4997         /*
4998          * Set transaction id's of transactions that have to be committed
4999          * to finish f[data]sync. We set them to currently running transaction
5000          * as we cannot be sure that the inode or some of its metadata isn't
5001          * part of the transaction - the inode could have been reclaimed and
5002          * now it is reread from disk.
5003          */
5004         if (journal) {
5005                 transaction_t *transaction;
5006                 tid_t tid;
5007 
5008                 read_lock(&journal->j_state_lock);
5009                 if (journal->j_running_transaction)
5010                         transaction = journal->j_running_transaction;
5011                 else
5012                         transaction = journal->j_committing_transaction;
5013                 if (transaction)
5014                         tid = transaction->t_tid;
5015                 else
5016                         tid = journal->j_commit_sequence;
5017                 read_unlock(&journal->j_state_lock);
5018                 ei->i_sync_tid = tid;
5019                 ei->i_datasync_tid = tid;
5020         }
5021 
5022         if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5023                 if (ei->i_extra_isize == 0) {
5024                         /* The extra space is currently unused. Use it. */
5025                         BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5026                         ei->i_extra_isize = sizeof(struct ext4_inode) -
5027                                             EXT4_GOOD_OLD_INODE_SIZE;
5028                 } else {
5029                         ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5030                         if (ret)
5031                                 goto bad_inode;
5032                 }
5033         }
5034 
5035         EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5036         EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5037         EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5038         EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5039 
5040         if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5041                 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5042 
5043                 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5044                         if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5045                                 ivers |=
5046                     (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5047                 }
5048                 ext4_inode_set_iversion_queried(inode, ivers);
5049         }
5050 
5051         ret = 0;
5052         if (ei->i_file_acl &&
5053             !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5054                 ext4_error_inode(inode, function, line, 0,
5055                                  "iget: bad extended attribute block %llu",
5056                                  ei->i_file_acl);
5057                 ret = -EFSCORRUPTED;
5058                 goto bad_inode;
5059         } else if (!ext4_has_inline_data(inode)) {
5060                 /* validate the block references in the inode */
5061                 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5062                    (S_ISLNK(inode->i_mode) &&
5063                     !ext4_inode_is_fast_symlink(inode))) {
5064                         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5065                                 ret = ext4_ext_check_inode(inode);
5066                         else
5067                                 ret = ext4_ind_check_inode(inode);
5068                 }
5069         }
5070         if (ret)
5071                 goto bad_inode;
5072 
5073         if (S_ISREG(inode->i_mode)) {
5074                 inode->i_op = &ext4_file_inode_operations;
5075                 inode->i_fop = &ext4_file_operations;
5076                 ext4_set_aops(inode);
5077         } else if (S_ISDIR(inode->i_mode)) {
5078                 inode->i_op = &ext4_dir_inode_operations;
5079                 inode->i_fop = &ext4_dir_operations;
5080         } else if (S_ISLNK(inode->i_mode)) {
5081                 /* VFS does not allow setting these so must be corruption */
5082                 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5083                         ext4_error_inode(inode, function, line, 0,
5084                                          "iget: immutable or append flags "
5085                                          "not allowed on symlinks");
5086                         ret = -EFSCORRUPTED;
5087                         goto bad_inode;
5088                 }
5089                 if (IS_ENCRYPTED(inode)) {
5090                         inode->i_op = &ext4_encrypted_symlink_inode_operations;
5091                         ext4_set_aops(inode);
5092                 } else if (ext4_inode_is_fast_symlink(inode)) {
5093                         inode->i_link = (char *)ei->i_data;
5094                         inode->i_op = &ext4_fast_symlink_inode_operations;
5095                         nd_terminate_link(ei->i_data, inode->i_size,
5096                                 sizeof(ei->i_data) - 1);
5097                 } else {
5098                         inode->i_op = &ext4_symlink_inode_operations;
5099                         ext4_set_aops(inode);
5100                 }
5101                 inode_nohighmem(inode);
5102         } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5103               S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5104                 inode->i_op = &ext4_special_inode_operations;
5105                 if (raw_inode->i_block[0])
5106                         init_special_inode(inode, inode->i_mode,
5107                            old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5108                 else
5109                         init_special_inode(inode, inode->i_mode,
5110                            new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5111         } else if (ino == EXT4_BOOT_LOADER_INO) {
5112                 make_bad_inode(inode);
5113         } else {
5114                 ret = -EFSCORRUPTED;
5115                 ext4_error_inode(inode, function, line, 0,
5116                                  "iget: bogus i_mode (%o)", inode->i_mode);
5117                 goto bad_inode;
5118         }
5119         brelse(iloc.bh);
5120 
5121         unlock_new_inode(inode);
5122         return inode;
5123 
5124 bad_inode:
5125         brelse(iloc.bh);
5126         iget_failed(inode);
5127         return ERR_PTR(ret);
5128 }
5129 
5130 static int ext4_inode_blocks_set(handle_t *handle,
5131                                 struct ext4_inode *raw_inode,
5132                                 struct ext4_inode_info *ei)
5133 {
5134         struct inode *inode = &(ei->vfs_inode);
5135         u64 i_blocks = inode->i_blocks;
5136         struct super_block *sb = inode->i_sb;
5137 
5138         if (i_blocks <= ~0U) {
5139                 /*
5140                  * i_blocks can be represented in a 32 bit variable
5141                  * as multiple of 512 bytes
5142                  */
5143                 raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
5144                 raw_inode->i_blocks_high = 0;
5145                 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5146                 return 0;
5147         }
5148         if (!ext4_has_feature_huge_file(sb))
5149                 return -EFBIG;
5150 
5151         if (i_blocks <= 0xffffffffffffULL) {
5152                 /*
5153                  * i_blocks can be represented in a 48 bit variable
5154                  * as multiple of 512 bytes
5155                  */
5156                 raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
5157                 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5158                 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5159         } else {
5160                 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5161                 /* i_block is stored in file system block size */
5162                 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5163                 raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
5164                 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5165         }
5166         return 0;
5167 }
5168 
5169 struct other_inode {
5170         unsigned long           orig_ino;
5171         struct ext4_inode       *raw_inode;
5172 };
5173 
5174 static int other_inode_match(struct inode * inode, unsigned long ino,
5175                              void *data)
5176 {
5177         struct other_inode *oi = (struct other_inode *) data;
5178 
5179         if ((inode->i_ino != ino) ||
5180             (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5181                                I_DIRTY_INODE)) ||
5182             ((inode->i_state & I_DIRTY_TIME) == 0))
5183                 return 0;
5184         spin_lock(&inode->i_lock);
5185         if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5186                                 I_DIRTY_INODE)) == 0) &&
5187             (inode->i_state & I_DIRTY_TIME)) {
5188                 struct ext4_inode_info  *ei = EXT4_I(inode);
5189 
5190                 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5191                 spin_unlock(&inode->i_lock);
5192 
5193                 spin_lock(&ei->i_raw_lock);
5194                 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5195                 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5196                 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5197                 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5198                 spin_unlock(&ei->i_raw_lock);
5199                 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5200                 return -1;
5201         }
5202         spin_unlock(&inode->i_lock);
5203         return -1;
5204 }
5205 
5206 /*
5207  * Opportunistically update the other time fields for other inodes in
5208  * the same inode table block.
5209  */
5210 static void ext4_update_other_inodes_time(struct super_block *sb,
5211                                           unsigned long orig_ino, char *buf)
5212 {
5213         struct other_inode oi;
5214         unsigned long ino;
5215         int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5216         int inode_size = EXT4_INODE_SIZE(sb);
5217 
5218         oi.orig_ino = orig_ino;
5219         /*
5220          * Calculate the first inode in the inode table block.  Inode
5221          * numbers are one-based.  That is, the first inode in a block
5222          * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5223          */
5224         ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5225         for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5226                 if (ino == orig_ino)
5227                         continue;
5228                 oi.raw_inode = (struct ext4_inode *) buf;
5229                 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5230         }
5231 }
5232 
5233 /*
5234  * Post the struct inode info into an on-disk inode location in the
5235  * buffer-cache.  This gobbles the caller's reference to the
5236  * buffer_head in the inode location struct.
5237  *
5238  * The caller must have write access to iloc->bh.
5239  */
5240 static int ext4_do_update_inode(handle_t *handle,
5241                                 struct inode *inode,
5242                                 struct ext4_iloc *iloc)
5243 {
5244         struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5245         struct ext4_inode_info *ei = EXT4_I(inode);
5246         struct buffer_head *bh = iloc->bh;
5247         struct super_block *sb = inode->i_sb;
5248         int err = 0, rc, block;
5249         int need_datasync = 0, set_large_file = 0;
5250         uid_t i_uid;
5251         gid_t i_gid;
5252         projid_t i_projid;
5253 
5254         spin_lock(&ei->i_raw_lock);
5255 
5256         /* For fields not tracked in the in-memory inode,
5257          * initialise them to zero for new inodes. */
5258         if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5259                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5260 
5261         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5262         i_uid = i_uid_read(inode);
5263         i_gid = i_gid_read(inode);
5264         i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5265         if (!(test_opt(inode->i_sb, NO_UID32))) {
5266                 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5267                 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5268 /*
5269  * Fix up interoperability with old kernels. Otherwise, old inodes get
5270  * re-used with the upper 16 bits of the uid/gid intact
5271  */
5272                 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5273                         raw_inode->i_uid_high = 0;
5274                         raw_inode->i_gid_high = 0;
5275                 } else {
5276                         raw_inode->i_uid_high =
5277                                 cpu_to_le16(high_16_bits(i_uid));
5278                         raw_inode->i_gid_high =
5279                                 cpu_to_le16(high_16_bits(i_gid));
5280                 }
5281         } else {
5282                 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5283                 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5284                 raw_inode->i_uid_high = 0;
5285                 raw_inode->i_gid_high = 0;
5286         }
5287         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5288 
5289         EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5290         EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5291         EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5292         EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5293 
5294         err = ext4_inode_blocks_set(handle, raw_inode, ei);
5295         if (err) {
5296                 spin_unlock(&ei->i_raw_lock);
5297                 goto out_brelse;
5298         }
5299         raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5300         raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5301         if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5302                 raw_inode->i_file_acl_high =
5303                         cpu_to_le16(ei->i_file_acl >> 32);
5304         raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5305         if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5306                 ext4_isize_set(raw_inode, ei->i_disksize);
5307                 need_datasync = 1;
5308         }
5309         if (ei->i_disksize > 0x7fffffffULL) {
5310                 if (!ext4_has_feature_large_file(sb) ||
5311                                 EXT4_SB(sb)->s_es->s_rev_level ==
5312                     cpu_to_le32(EXT4_GOOD_OLD_REV))
5313                         set_large_file = 1;
5314         }
5315         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5316         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5317                 if (old_valid_dev(inode->i_rdev)) {
5318                         raw_inode->i_block[0] =
5319                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
5320                         raw_inode->i_block[1] = 0;
5321                 } else {
5322                         raw_inode->i_block[0] = 0;
5323                         raw_inode->i_block[1] =
5324                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
5325                         raw_inode->i_block[2] = 0;
5326                 }
5327         } else if (!ext4_has_inline_data(inode)) {
5328                 for (block = 0; block < EXT4_N_BLOCKS; block++)
5329                         raw_inode->i_block[block] = ei->i_data[block];
5330         }
5331 
5332         if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5333                 u64 ivers = ext4_inode_peek_iversion(inode);
5334 
5335                 raw_inode->i_disk_version = cpu_to_le32(ivers);
5336                 if (ei->i_extra_isize) {
5337                         if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5338                                 raw_inode->i_version_hi =
5339                                         cpu_to_le32(ivers >> 32);
5340                         raw_inode->i_extra_isize =
5341                                 cpu_to_le16(ei->i_extra_isize);
5342                 }
5343         }
5344 
5345         BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5346                i_projid != EXT4_DEF_PROJID);
5347 
5348         if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5349             EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5350                 raw_inode->i_projid = cpu_to_le32(i_projid);
5351 
5352         ext4_inode_csum_set(inode, raw_inode, ei);
5353         spin_unlock(&ei->i_raw_lock);
5354         if (inode->i_sb->s_flags & SB_LAZYTIME)
5355                 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5356                                               bh->b_data);
5357 
5358         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5359         rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5360         if (!err)
5361                 err = rc;
5362         ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5363         if (set_large_file) {
5364                 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5365                 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5366                 if (err)
5367                         goto out_brelse;
5368                 ext4_set_feature_large_file(sb);
5369                 ext4_handle_sync(handle);
5370                 err = ext4_handle_dirty_super(handle, sb);
5371         }
5372         ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5373 out_brelse:
5374         brelse(bh);
5375         ext4_std_error(inode->i_sb, err);
5376         return err;
5377 }
5378 
5379 /*
5380  * ext4_write_inode()
5381  *
5382  * We are called from a few places:
5383  *
5384  * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5385  *   Here, there will be no transaction running. We wait for any running
5386  *   transaction to commit.
5387  *
5388  * - Within flush work (sys_sync(), kupdate and such).
5389  *   We wait on commit, if told to.
5390  *
5391  * - Within iput_final() -> write_inode_now()
5392  *   We wait on commit, if told to.
5393  *
5394  * In all cases it is actually safe for us to return without doing anything,
5395  * because the inode has been copied into a raw inode buffer in
5396  * ext4_mark_inode_dirty().  This is a correctness thing for WB_SYNC_ALL
5397  * writeback.
5398  *
5399  * Note that we are absolutely dependent upon all inode dirtiers doing the
5400  * right thing: they *must* call mark_inode_dirty() after dirtying info in
5401  * which we are interested.
5402  *
5403  * It would be a bug for them to not do this.  The code:
5404  *
5405  *      mark_inode_dirty(inode)
5406  *      stuff();
5407  *      inode->i_size = expr;
5408  *
5409  * is in error because write_inode() could occur while `stuff()' is running,
5410  * and the new i_size will be lost.  Plus the inode will no longer be on the
5411  * superblock's dirty inode list.
5412  */
5413 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5414 {
5415         int err;
5416 
5417         if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5418             sb_rdonly(inode->i_sb))
5419                 return 0;
5420 
5421         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5422                 return -EIO;
5423 
5424         if (EXT4_SB(inode->i_sb)->s_journal) {
5425                 if (ext4_journal_current_handle()) {
5426                         jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5427                         dump_stack();
5428                         return -EIO;
5429                 }
5430 
5431                 /*
5432                  * No need to force transaction in WB_SYNC_NONE mode. Also
5433                  * ext4_sync_fs() will force the commit after everything is
5434                  * written.
5435                  */
5436                 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5437                         return 0;
5438 
5439                 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5440                                                 EXT4_I(inode)->i_sync_tid);
5441         } else {
5442                 struct ext4_iloc iloc;
5443 
5444                 err = __ext4_get_inode_loc(inode, &iloc, 0);
5445                 if (err)
5446                         return err;
5447                 /*
5448                  * sync(2) will flush the whole buffer cache. No need to do
5449                  * it here separately for each inode.
5450                  */
5451                 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5452                         sync_dirty_buffer(iloc.bh);
5453                 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5454                         EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5455                                          "IO error syncing inode");
5456                         err = -EIO;
5457                 }
5458                 brelse(iloc.bh);
5459         }
5460         return err;
5461 }
5462 
5463 /*
5464  * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5465  * buffers that are attached to a page stradding i_size and are undergoing
5466  * commit. In that case we have to wait for commit to finish and try again.
5467  */
5468 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5469 {
5470         struct page *page;
5471         unsigned offset;
5472         journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5473         tid_t commit_tid = 0;
5474         int ret;
5475 
5476         offset = inode->i_size & (PAGE_SIZE - 1);
5477         /*
5478          * All buffers in the last page remain valid? Then there's nothing to
5479          * do. We do the check mainly to optimize the common PAGE_SIZE ==
5480          * blocksize case
5481          */
5482         if (offset > PAGE_SIZE - i_blocksize(inode))
5483                 return;
5484         while (1) {
5485                 page = find_lock_page(inode->i_mapping,
5486                                       inode->i_size >> PAGE_SHIFT);
5487                 if (!page)
5488                         return;
5489                 ret = __ext4_journalled_invalidatepage(page, offset,
5490                                                 PAGE_SIZE - offset);
5491                 unlock_page(page);
5492                 put_page(page);
5493                 if (ret != -EBUSY)
5494                         return;
5495                 commit_tid = 0;
5496                 read_lock(&journal->j_state_lock);
5497                 if (journal->j_committing_transaction)
5498                         commit_tid = journal->j_committing_transaction->t_tid;
5499                 read_unlock(&journal->j_state_lock);
5500                 if (commit_tid)
5501                         jbd2_log_wait_commit(journal, commit_tid);
5502         }
5503 }
5504 
5505 /*
5506  * ext4_setattr()
5507  *
5508  * Called from notify_change.
5509  *
5510  * We want to trap VFS attempts to truncate the file as soon as
5511  * possible.  In particular, we want to make sure that when the VFS
5512  * shrinks i_size, we put the inode on the orphan list and modify
5513  * i_disksize immediately, so that during the subsequent flushing of
5514  * dirty pages and freeing of disk blocks, we can guarantee that any
5515  * commit will leave the blocks being flushed in an unused state on
5516  * disk.  (On recovery, the inode will get truncated and the blocks will
5517  * be freed, so we have a strong guarantee that no future commit will
5518  * leave these blocks visible to the user.)
5519  *
5520  * Another thing we have to assure is that if we are in ordered mode
5521  * and inode is still attached to the committing transaction, we must
5522  * we start writeout of all the dirty pages which are being truncated.
5523  * This way we are sure that all the data written in the previous
5524  * transaction are already on disk (truncate waits for pages under
5525  * writeback).
5526  *
5527  * Called with inode->i_mutex down.
5528  */
5529 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5530 {
5531         struct inode *inode = d_inode(dentry);
5532         int error, rc = 0;
5533         int orphan = 0;
5534         const unsigned int ia_valid = attr->ia_valid;
5535 
5536         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5537                 return -EIO;
5538 
5539         if (unlikely(IS_IMMUTABLE(inode)))
5540                 return -EPERM;
5541 
5542         if (unlikely(IS_APPEND(inode) &&
5543                      (ia_valid & (ATTR_MODE | ATTR_UID |
5544                                   ATTR_GID | ATTR_TIMES_SET))))
5545                 return -EPERM;
5546 
5547         error = setattr_prepare(dentry, attr);
5548         if (error)
5549                 return error;
5550 
5551         error = fscrypt_prepare_setattr(dentry, attr);
5552         if (error)
5553                 return error;
5554 
5555         if (is_quota_modification(inode, attr)) {
5556                 error = dquot_initialize(inode);
5557                 if (error)
5558                         return error;
5559         }
5560         if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5561             (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5562                 handle_t *handle;
5563 
5564                 /* (user+group)*(old+new) structure, inode write (sb,
5565                  * inode block, ? - but truncate inode update has it) */
5566                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5567                         (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5568                          EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5569                 if (IS_ERR(handle)) {
5570                         error = PTR_ERR(handle);
5571                         goto err_out;
5572                 }
5573 
5574                 /* dquot_transfer() calls back ext4_get_inode_usage() which
5575                  * counts xattr inode references.
5576                  */
5577                 down_read(&EXT4_I(inode)->xattr_sem);
5578                 error = dquot_transfer(inode, attr);
5579                 up_read(&EXT4_I(inode)->xattr_sem);
5580 
5581                 if (error) {
5582                         ext4_journal_stop(handle);
5583                         return error;
5584                 }
5585                 /* Update corresponding info in inode so that everything is in
5586                  * one transaction */
5587                 if (attr->ia_valid & ATTR_UID)
5588                         inode->i_uid = attr->ia_uid;
5589                 if (attr->ia_valid & ATTR_GID)
5590                         inode->i_gid = attr->ia_gid;
5591                 error = ext4_mark_inode_dirty(handle, inode);
5592                 ext4_journal_stop(handle);
5593         }
5594 
5595         if (attr->ia_valid & ATTR_SIZE) {
5596                 handle_t *handle;
5597                 loff_t oldsize = inode->i_size;
5598                 int shrink = (attr->ia_size <= inode->i_size);
5599 
5600                 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5601                         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5602 
5603                         if (attr->ia_size > sbi->s_bitmap_maxbytes)
5604                                 return -EFBIG;
5605                 }
5606                 if (!S_ISREG(inode->i_mode))
5607                         return -EINVAL;
5608 
5609                 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5610                         inode_inc_iversion(inode);
5611 
5612                 if (ext4_should_order_data(inode) &&
5613                     (attr->ia_size < inode->i_size)) {
5614                         error = ext4_begin_ordered_truncate(inode,
5615                                                             attr->ia_size);
5616                         if (error)
5617                                 goto err_out;
5618                 }
5619                 if (attr->ia_size != inode->i_size) {
5620                         handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5621                         if (IS_ERR(handle)) {
5622                                 error = PTR_ERR(handle);
5623                                 goto err_out;
5624                         }
5625                         if (ext4_handle_valid(handle) && shrink) {
5626                                 error = ext4_orphan_add(handle, inode);
5627                                 orphan = 1;
5628                         }
5629                         /*
5630                          * Update c/mtime on truncate up, ext4_truncate() will
5631                          * update c/mtime in shrink case below
5632                          */
5633                         if (!shrink) {
5634                                 inode->i_mtime = current_time(inode);
5635                                 inode->i_ctime = inode->i_mtime;
5636                         }
5637                         down_write(&EXT4_I(inode)->i_data_sem);
5638                         EXT4_I(inode)->i_disksize = attr->ia_size;
5639                         rc = ext4_mark_inode_dirty(handle, inode);
5640                         if (!error)
5641                                 error = rc;
5642                         /*
5643                          * We have to update i_size under i_data_sem together
5644                          * with i_disksize to avoid races with writeback code
5645                          * running ext4_wb_update_i_disksize().
5646                          */
5647                         if (!error)
5648                                 i_size_write(inode, attr->ia_size);
5649                         up_write(&EXT4_I(inode)->i_data_sem);
5650                         ext4_journal_stop(handle);
5651                         if (error) {
5652                                 if (orphan && inode->i_nlink)
5653                                         ext4_orphan_del(NULL, inode);
5654                                 goto err_out;
5655                         }
5656                 }
5657                 if (!shrink) {
5658                         pagecache_isize_extended(inode, oldsize, inode->i_size);
5659                 } else {
5660                         /*
5661                          * Blocks are going to be removed from the inode. Wait
5662                          * for dio in flight.
5663                          */
5664                         inode_dio_wait(inode);
5665                 }
5666                 if (orphan &&