1 /*
2 * linux/fs/ext4/inode.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
17 *
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
19 */
20
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40
41 #include "ext4_jbd2.h"
42 #include "xattr.h"
43 #include "acl.h"
44 #include "truncate.h"
45
46 #include <trace/events/ext4.h>
47
48 #define MPAGE_DA_EXTENT_TAIL 0x01
49
50 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
52 {
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
54 __u16 csum_lo;
55 __u16 csum_hi = 0;
56 __u32 csum;
57
58 csum_lo = raw->i_checksum_lo;
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = raw->i_checksum_hi;
63 raw->i_checksum_hi = 0;
64 }
65
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
68
69 raw->i_checksum_lo = csum_lo;
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = csum_hi;
73
74 return csum;
75 }
76
77 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
79 {
80 __u32 provided, calculated;
81
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
86 return 1;
87
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
93 else
94 calculated &= 0xFFFF;
95
96 return provided == calculated;
97 }
98
99 static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
101 {
102 __u32 csum;
103
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
108 return;
109
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
115 }
116
117 static inline int ext4_begin_ordered_truncate(struct inode *inode,
118 loff_t new_size)
119 {
120 trace_ext4_begin_ordered_truncate(inode, new_size);
121 /*
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
126 */
127 if (!EXT4_I(inode)->jinode)
128 return 0;
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
131 new_size);
132 }
133
134 static void ext4_invalidatepage(struct page *page, unsigned long offset);
135 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
136 struct buffer_head *bh_result, int create);
137 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
138 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
139 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
140 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
141 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
142 struct inode *inode, struct page *page, loff_t from,
143 loff_t length, int flags);
144
145 /*
146 * Test whether an inode is a fast symlink.
147 */
148 static int ext4_inode_is_fast_symlink(struct inode *inode)
149 {
150 int ea_blocks = EXT4_I(inode)->i_file_acl ?
151 (inode->i_sb->s_blocksize >> 9) : 0;
152
153 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
154 }
155
156 /*
157 * Restart the transaction associated with *handle. This does a commit,
158 * so before we call here everything must be consistently dirtied against
159 * this transaction.
160 */
161 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
162 int nblocks)
163 {
164 int ret;
165
166 /*
167 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
168 * moment, get_block can be called only for blocks inside i_size since
169 * page cache has been already dropped and writes are blocked by
170 * i_mutex. So we can safely drop the i_data_sem here.
171 */
172 BUG_ON(EXT4_JOURNAL(inode) == NULL);
173 jbd_debug(2, "restarting handle %p\n", handle);
174 up_write(&EXT4_I(inode)->i_data_sem);
175 ret = ext4_journal_restart(handle, nblocks);
176 down_write(&EXT4_I(inode)->i_data_sem);
177 ext4_discard_preallocations(inode);
178
179 return ret;
180 }
181
182 /*
183 * Called at the last iput() if i_nlink is zero.
184 */
185 void ext4_evict_inode(struct inode *inode)
186 {
187 handle_t *handle;
188 int err;
189
190 trace_ext4_evict_inode(inode);
191
192 ext4_ioend_wait(inode);
193
194 if (inode->i_nlink) {
195 /*
196 * When journalling data dirty buffers are tracked only in the
197 * journal. So although mm thinks everything is clean and
198 * ready for reaping the inode might still have some pages to
199 * write in the running transaction or waiting to be
200 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 * (via truncate_inode_pages()) to discard these buffers can
202 * cause data loss. Also even if we did not discard these
203 * buffers, we would have no way to find them after the inode
204 * is reaped and thus user could see stale data if he tries to
205 * read them before the transaction is checkpointed. So be
206 * careful and force everything to disk here... We use
207 * ei->i_datasync_tid to store the newest transaction
208 * containing inode's data.
209 *
210 * Note that directories do not have this problem because they
211 * don't use page cache.
212 */
213 if (ext4_should_journal_data(inode) &&
214 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
216 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
217
218 jbd2_log_start_commit(journal, commit_tid);
219 jbd2_log_wait_commit(journal, commit_tid);
220 filemap_write_and_wait(&inode->i_data);
221 }
222 truncate_inode_pages(&inode->i_data, 0);
223 goto no_delete;
224 }
225
226 if (!is_bad_inode(inode))
227 dquot_initialize(inode);
228
229 if (ext4_should_order_data(inode))
230 ext4_begin_ordered_truncate(inode, 0);
231 truncate_inode_pages(&inode->i_data, 0);
232
233 if (is_bad_inode(inode))
234 goto no_delete;
235
236 /*
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
239 */
240 sb_start_intwrite(inode->i_sb);
241 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
242 if (IS_ERR(handle)) {
243 ext4_std_error(inode->i_sb, PTR_ERR(handle));
244 /*
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
247 * cleaned up.
248 */
249 ext4_orphan_del(NULL, inode);
250 sb_end_intwrite(inode->i_sb);
251 goto no_delete;
252 }
253
254 if (IS_SYNC(inode))
255 ext4_handle_sync(handle);
256 inode->i_size = 0;
257 err = ext4_mark_inode_dirty(handle, inode);
258 if (err) {
259 ext4_warning(inode->i_sb,
260 "couldn't mark inode dirty (err %d)", err);
261 goto stop_handle;
262 }
263 if (inode->i_blocks)
264 ext4_truncate(inode);
265
266 /*
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
271 */
272 if (!ext4_handle_has_enough_credits(handle, 3)) {
273 err = ext4_journal_extend(handle, 3);
274 if (err > 0)
275 err = ext4_journal_restart(handle, 3);
276 if (err != 0) {
277 ext4_warning(inode->i_sb,
278 "couldn't extend journal (err %d)", err);
279 stop_handle:
280 ext4_journal_stop(handle);
281 ext4_orphan_del(NULL, inode);
282 sb_end_intwrite(inode->i_sb);
283 goto no_delete;
284 }
285 }
286
287 /*
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
294 */
295 ext4_orphan_del(handle, inode);
296 EXT4_I(inode)->i_dtime = get_seconds();
297
298 /*
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
303 * fails.
304 */
305 if (ext4_mark_inode_dirty(handle, inode))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode);
308 else
309 ext4_free_inode(handle, inode);
310 ext4_journal_stop(handle);
311 sb_end_intwrite(inode->i_sb);
312 return;
313 no_delete:
314 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
315 }
316
317 #ifdef CONFIG_QUOTA
318 qsize_t *ext4_get_reserved_space(struct inode *inode)
319 {
320 return &EXT4_I(inode)->i_reserved_quota;
321 }
322 #endif
323
324 /*
325 * Calculate the number of metadata blocks need to reserve
326 * to allocate a block located at @lblock
327 */
328 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
329 {
330 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
331 return ext4_ext_calc_metadata_amount(inode, lblock);
332
333 return ext4_ind_calc_metadata_amount(inode, lblock);
334 }
335
336 /*
337 * Called with i_data_sem down, which is important since we can call
338 * ext4_discard_preallocations() from here.
339 */
340 void ext4_da_update_reserve_space(struct inode *inode,
341 int used, int quota_claim)
342 {
343 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
344 struct ext4_inode_info *ei = EXT4_I(inode);
345
346 spin_lock(&ei->i_block_reservation_lock);
347 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
348 if (unlikely(used > ei->i_reserved_data_blocks)) {
349 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
350 "with only %d reserved data blocks",
351 __func__, inode->i_ino, used,
352 ei->i_reserved_data_blocks);
353 WARN_ON(1);
354 used = ei->i_reserved_data_blocks;
355 }
356
357 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
358 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d "
359 "with only %d reserved metadata blocks\n", __func__,
360 inode->i_ino, ei->i_allocated_meta_blocks,
361 ei->i_reserved_meta_blocks);
362 WARN_ON(1);
363 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
364 }
365
366 /* Update per-inode reservations */
367 ei->i_reserved_data_blocks -= used;
368 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
369 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
370 used + ei->i_allocated_meta_blocks);
371 ei->i_allocated_meta_blocks = 0;
372
373 if (ei->i_reserved_data_blocks == 0) {
374 /*
375 * We can release all of the reserved metadata blocks
376 * only when we have written all of the delayed
377 * allocation blocks.
378 */
379 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
380 ei->i_reserved_meta_blocks);
381 ei->i_reserved_meta_blocks = 0;
382 ei->i_da_metadata_calc_len = 0;
383 }
384 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
385
386 /* Update quota subsystem for data blocks */
387 if (quota_claim)
388 dquot_claim_block(inode, EXT4_C2B(sbi, used));
389 else {
390 /*
391 * We did fallocate with an offset that is already delayed
392 * allocated. So on delayed allocated writeback we should
393 * not re-claim the quota for fallocated blocks.
394 */
395 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
396 }
397
398 /*
399 * If we have done all the pending block allocations and if
400 * there aren't any writers on the inode, we can discard the
401 * inode's preallocations.
402 */
403 if ((ei->i_reserved_data_blocks == 0) &&
404 (atomic_read(&inode->i_writecount) == 0))
405 ext4_discard_preallocations(inode);
406 }
407
408 static int __check_block_validity(struct inode *inode, const char *func,
409 unsigned int line,
410 struct ext4_map_blocks *map)
411 {
412 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
413 map->m_len)) {
414 ext4_error_inode(inode, func, line, map->m_pblk,
415 "lblock %lu mapped to illegal pblock "
416 "(length %d)", (unsigned long) map->m_lblk,
417 map->m_len);
418 return -EIO;
419 }
420 return 0;
421 }
422
423 #define check_block_validity(inode, map) \
424 __check_block_validity((inode), __func__, __LINE__, (map))
425
426 /*
427 * Return the number of contiguous dirty pages in a given inode
428 * starting at page frame idx.
429 */
430 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
431 unsigned int max_pages)
432 {
433 struct address_space *mapping = inode->i_mapping;
434 pgoff_t index;
435 struct pagevec pvec;
436 pgoff_t num = 0;
437 int i, nr_pages, done = 0;
438
439 if (max_pages == 0)
440 return 0;
441 pagevec_init(&pvec, 0);
442 while (!done) {
443 index = idx;
444 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
445 PAGECACHE_TAG_DIRTY,
446 (pgoff_t)PAGEVEC_SIZE);
447 if (nr_pages == 0)
448 break;
449 for (i = 0; i < nr_pages; i++) {
450 struct page *page = pvec.pages[i];
451 struct buffer_head *bh, *head;
452
453 lock_page(page);
454 if (unlikely(page->mapping != mapping) ||
455 !PageDirty(page) ||
456 PageWriteback(page) ||
457 page->index != idx) {
458 done = 1;
459 unlock_page(page);
460 break;
461 }
462 if (page_has_buffers(page)) {
463 bh = head = page_buffers(page);
464 do {
465 if (!buffer_delay(bh) &&
466 !buffer_unwritten(bh))
467 done = 1;
468 bh = bh->b_this_page;
469 } while (!done && (bh != head));
470 }
471 unlock_page(page);
472 if (done)
473 break;
474 idx++;
475 num++;
476 if (num >= max_pages) {
477 done = 1;
478 break;
479 }
480 }
481 pagevec_release(&pvec);
482 }
483 return num;
484 }
485
486 /*
487 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
488 */
489 static void set_buffers_da_mapped(struct inode *inode,
490 struct ext4_map_blocks *map)
491 {
492 struct address_space *mapping = inode->i_mapping;
493 struct pagevec pvec;
494 int i, nr_pages;
495 pgoff_t index, end;
496
497 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
498 end = (map->m_lblk + map->m_len - 1) >>
499 (PAGE_CACHE_SHIFT - inode->i_blkbits);
500
501 pagevec_init(&pvec, 0);
502 while (index <= end) {
503 nr_pages = pagevec_lookup(&pvec, mapping, index,
504 min(end - index + 1,
505 (pgoff_t)PAGEVEC_SIZE));
506 if (nr_pages == 0)
507 break;
508 for (i = 0; i < nr_pages; i++) {
509 struct page *page = pvec.pages[i];
510 struct buffer_head *bh, *head;
511
512 if (unlikely(page->mapping != mapping) ||
513 !PageDirty(page))
514 break;
515
516 if (page_has_buffers(page)) {
517 bh = head = page_buffers(page);
518 do {
519 set_buffer_da_mapped(bh);
520 bh = bh->b_this_page;
521 } while (bh != head);
522 }
523 index++;
524 }
525 pagevec_release(&pvec);
526 }
527 }
528
529 /*
530 * The ext4_map_blocks() function tries to look up the requested blocks,
531 * and returns if the blocks are already mapped.
532 *
533 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
534 * and store the allocated blocks in the result buffer head and mark it
535 * mapped.
536 *
537 * If file type is extents based, it will call ext4_ext_map_blocks(),
538 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
539 * based files
540 *
541 * On success, it returns the number of blocks being mapped or allocate.
542 * if create==0 and the blocks are pre-allocated and uninitialized block,
543 * the result buffer head is unmapped. If the create ==1, it will make sure
544 * the buffer head is mapped.
545 *
546 * It returns 0 if plain look up failed (blocks have not been allocated), in
547 * that case, buffer head is unmapped
548 *
549 * It returns the error in case of allocation failure.
550 */
551 int ext4_map_blocks(handle_t *handle, struct inode *inode,
552 struct ext4_map_blocks *map, int flags)
553 {
554 int retval;
555
556 map->m_flags = 0;
557 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
558 "logical block %lu\n", inode->i_ino, flags, map->m_len,
559 (unsigned long) map->m_lblk);
560 /*
561 * Try to see if we can get the block without requesting a new
562 * file system block.
563 */
564 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
565 down_read((&EXT4_I(inode)->i_data_sem));
566 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
567 retval = ext4_ext_map_blocks(handle, inode, map, flags &
568 EXT4_GET_BLOCKS_KEEP_SIZE);
569 } else {
570 retval = ext4_ind_map_blocks(handle, inode, map, flags &
571 EXT4_GET_BLOCKS_KEEP_SIZE);
572 }
573 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
574 up_read((&EXT4_I(inode)->i_data_sem));
575
576 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
577 int ret = check_block_validity(inode, map);
578 if (ret != 0)
579 return ret;
580 }
581
582 /* If it is only a block(s) look up */
583 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
584 return retval;
585
586 /*
587 * Returns if the blocks have already allocated
588 *
589 * Note that if blocks have been preallocated
590 * ext4_ext_get_block() returns the create = 0
591 * with buffer head unmapped.
592 */
593 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
594 return retval;
595
596 /*
597 * When we call get_blocks without the create flag, the
598 * BH_Unwritten flag could have gotten set if the blocks
599 * requested were part of a uninitialized extent. We need to
600 * clear this flag now that we are committed to convert all or
601 * part of the uninitialized extent to be an initialized
602 * extent. This is because we need to avoid the combination
603 * of BH_Unwritten and BH_Mapped flags being simultaneously
604 * set on the buffer_head.
605 */
606 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
607
608 /*
609 * New blocks allocate and/or writing to uninitialized extent
610 * will possibly result in updating i_data, so we take
611 * the write lock of i_data_sem, and call get_blocks()
612 * with create == 1 flag.
613 */
614 down_write((&EXT4_I(inode)->i_data_sem));
615
616 /*
617 * if the caller is from delayed allocation writeout path
618 * we have already reserved fs blocks for allocation
619 * let the underlying get_block() function know to
620 * avoid double accounting
621 */
622 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
623 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
624 /*
625 * We need to check for EXT4 here because migrate
626 * could have changed the inode type in between
627 */
628 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
629 retval = ext4_ext_map_blocks(handle, inode, map, flags);
630 } else {
631 retval = ext4_ind_map_blocks(handle, inode, map, flags);
632
633 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
634 /*
635 * We allocated new blocks which will result in
636 * i_data's format changing. Force the migrate
637 * to fail by clearing migrate flags
638 */
639 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
640 }
641
642 /*
643 * Update reserved blocks/metadata blocks after successful
644 * block allocation which had been deferred till now. We don't
645 * support fallocate for non extent files. So we can update
646 * reserve space here.
647 */
648 if ((retval > 0) &&
649 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
650 ext4_da_update_reserve_space(inode, retval, 1);
651 }
652 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
653 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
654
655 /* If we have successfully mapped the delayed allocated blocks,
656 * set the BH_Da_Mapped bit on them. Its important to do this
657 * under the protection of i_data_sem.
658 */
659 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
660 set_buffers_da_mapped(inode, map);
661 }
662
663 up_write((&EXT4_I(inode)->i_data_sem));
664 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
665 int ret = check_block_validity(inode, map);
666 if (ret != 0)
667 return ret;
668 }
669 return retval;
670 }
671
672 /* Maximum number of blocks we map for direct IO at once. */
673 #define DIO_MAX_BLOCKS 4096
674
675 static int _ext4_get_block(struct inode *inode, sector_t iblock,
676 struct buffer_head *bh, int flags)
677 {
678 handle_t *handle = ext4_journal_current_handle();
679 struct ext4_map_blocks map;
680 int ret = 0, started = 0;
681 int dio_credits;
682
683 map.m_lblk = iblock;
684 map.m_len = bh->b_size >> inode->i_blkbits;
685
686 if (flags && !handle) {
687 /* Direct IO write... */
688 if (map.m_len > DIO_MAX_BLOCKS)
689 map.m_len = DIO_MAX_BLOCKS;
690 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
691 handle = ext4_journal_start(inode, dio_credits);
692 if (IS_ERR(handle)) {
693 ret = PTR_ERR(handle);
694 return ret;
695 }
696 started = 1;
697 }
698
699 ret = ext4_map_blocks(handle, inode, &map, flags);
700 if (ret > 0) {
701 map_bh(bh, inode->i_sb, map.m_pblk);
702 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
703 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
704 ret = 0;
705 }
706 if (started)
707 ext4_journal_stop(handle);
708 return ret;
709 }
710
711 int ext4_get_block(struct inode *inode, sector_t iblock,
712 struct buffer_head *bh, int create)
713 {
714 return _ext4_get_block(inode, iblock, bh,
715 create ? EXT4_GET_BLOCKS_CREATE : 0);
716 }
717
718 /*
719 * `handle' can be NULL if create is zero
720 */
721 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
722 ext4_lblk_t block, int create, int *errp)
723 {
724 struct ext4_map_blocks map;
725 struct buffer_head *bh;
726 int fatal = 0, err;
727
728 J_ASSERT(handle != NULL || create == 0);
729
730 map.m_lblk = block;
731 map.m_len = 1;
732 err = ext4_map_blocks(handle, inode, &map,
733 create ? EXT4_GET_BLOCKS_CREATE : 0);
734
735 /* ensure we send some value back into *errp */
736 *errp = 0;
737
738 if (err < 0)
739 *errp = err;
740 if (err <= 0)
741 return NULL;
742
743 bh = sb_getblk(inode->i_sb, map.m_pblk);
744 if (!bh) {
745 *errp = -EIO;
746 return NULL;
747 }
748 if (map.m_flags & EXT4_MAP_NEW) {
749 J_ASSERT(create != 0);
750 J_ASSERT(handle != NULL);
751
752 /*
753 * Now that we do not always journal data, we should
754 * keep in mind whether this should always journal the
755 * new buffer as metadata. For now, regular file
756 * writes use ext4_get_block instead, so it's not a
757 * problem.
758 */
759 lock_buffer(bh);
760 BUFFER_TRACE(bh, "call get_create_access");
761 fatal = ext4_journal_get_create_access(handle, bh);
762 if (!fatal && !buffer_uptodate(bh)) {
763 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
764 set_buffer_uptodate(bh);
765 }
766 unlock_buffer(bh);
767 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
768 err = ext4_handle_dirty_metadata(handle, inode, bh);
769 if (!fatal)
770 fatal = err;
771 } else {
772 BUFFER_TRACE(bh, "not a new buffer");
773 }
774 if (fatal) {
775 *errp = fatal;
776 brelse(bh);
777 bh = NULL;
778 }
779 return bh;
780 }
781
782 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
783 ext4_lblk_t block, int create, int *err)
784 {
785 struct buffer_head *bh;
786
787 bh = ext4_getblk(handle, inode, block, create, err);
788 if (!bh)
789 return bh;
790 if (buffer_uptodate(bh))
791 return bh;
792 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
793 wait_on_buffer(bh);
794 if (buffer_uptodate(bh))
795 return bh;
796 put_bh(bh);
797 *err = -EIO;
798 return NULL;
799 }
800
801 static int walk_page_buffers(handle_t *handle,
802 struct buffer_head *head,
803 unsigned from,
804 unsigned to,
805 int *partial,
806 int (*fn)(handle_t *handle,
807 struct buffer_head *bh))
808 {
809 struct buffer_head *bh;
810 unsigned block_start, block_end;
811 unsigned blocksize = head->b_size;
812 int err, ret = 0;
813 struct buffer_head *next;
814
815 for (bh = head, block_start = 0;
816 ret == 0 && (bh != head || !block_start);
817 block_start = block_end, bh = next) {
818 next = bh->b_this_page;
819 block_end = block_start + blocksize;
820 if (block_end <= from || block_start >= to) {
821 if (partial && !buffer_uptodate(bh))
822 *partial = 1;
823 continue;
824 }
825 err = (*fn)(handle, bh);
826 if (!ret)
827 ret = err;
828 }
829 return ret;
830 }
831
832 /*
833 * To preserve ordering, it is essential that the hole instantiation and
834 * the data write be encapsulated in a single transaction. We cannot
835 * close off a transaction and start a new one between the ext4_get_block()
836 * and the commit_write(). So doing the jbd2_journal_start at the start of
837 * prepare_write() is the right place.
838 *
839 * Also, this function can nest inside ext4_writepage() ->
840 * block_write_full_page(). In that case, we *know* that ext4_writepage()
841 * has generated enough buffer credits to do the whole page. So we won't
842 * block on the journal in that case, which is good, because the caller may
843 * be PF_MEMALLOC.
844 *
845 * By accident, ext4 can be reentered when a transaction is open via
846 * quota file writes. If we were to commit the transaction while thus
847 * reentered, there can be a deadlock - we would be holding a quota
848 * lock, and the commit would never complete if another thread had a
849 * transaction open and was blocking on the quota lock - a ranking
850 * violation.
851 *
852 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
853 * will _not_ run commit under these circumstances because handle->h_ref
854 * is elevated. We'll still have enough credits for the tiny quotafile
855 * write.
856 */
857 static int do_journal_get_write_access(handle_t *handle,
858 struct buffer_head *bh)
859 {
860 int dirty = buffer_dirty(bh);
861 int ret;
862
863 if (!buffer_mapped(bh) || buffer_freed(bh))
864 return 0;
865 /*
866 * __block_write_begin() could have dirtied some buffers. Clean
867 * the dirty bit as jbd2_journal_get_write_access() could complain
868 * otherwise about fs integrity issues. Setting of the dirty bit
869 * by __block_write_begin() isn't a real problem here as we clear
870 * the bit before releasing a page lock and thus writeback cannot
871 * ever write the buffer.
872 */
873 if (dirty)
874 clear_buffer_dirty(bh);
875 ret = ext4_journal_get_write_access(handle, bh);
876 if (!ret && dirty)
877 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
878 return ret;
879 }
880
881 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
882 struct buffer_head *bh_result, int create);
883 static int ext4_write_begin(struct file *file, struct address_space *mapping,
884 loff_t pos, unsigned len, unsigned flags,
885 struct page **pagep, void **fsdata)
886 {
887 struct inode *inode = mapping->host;
888 int ret, needed_blocks;
889 handle_t *handle;
890 int retries = 0;
891 struct page *page;
892 pgoff_t index;
893 unsigned from, to;
894
895 trace_ext4_write_begin(inode, pos, len, flags);
896 /*
897 * Reserve one block more for addition to orphan list in case
898 * we allocate blocks but write fails for some reason
899 */
900 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
901 index = pos >> PAGE_CACHE_SHIFT;
902 from = pos & (PAGE_CACHE_SIZE - 1);
903 to = from + len;
904
905 retry:
906 handle = ext4_journal_start(inode, needed_blocks);
907 if (IS_ERR(handle)) {
908 ret = PTR_ERR(handle);
909 goto out;
910 }
911
912 /* We cannot recurse into the filesystem as the transaction is already
913 * started */
914 flags |= AOP_FLAG_NOFS;
915
916 page = grab_cache_page_write_begin(mapping, index, flags);
917 if (!page) {
918 ext4_journal_stop(handle);
919 ret = -ENOMEM;
920 goto out;
921 }
922 *pagep = page;
923
924 if (ext4_should_dioread_nolock(inode))
925 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
926 else
927 ret = __block_write_begin(page, pos, len, ext4_get_block);
928
929 if (!ret && ext4_should_journal_data(inode)) {
930 ret = walk_page_buffers(handle, page_buffers(page),
931 from, to, NULL, do_journal_get_write_access);
932 }
933
934 if (ret) {
935 unlock_page(page);
936 page_cache_release(page);
937 /*
938 * __block_write_begin may have instantiated a few blocks
939 * outside i_size. Trim these off again. Don't need
940 * i_size_read because we hold i_mutex.
941 *
942 * Add inode to orphan list in case we crash before
943 * truncate finishes
944 */
945 if (pos + len > inode->i_size && ext4_can_truncate(inode))
946 ext4_orphan_add(handle, inode);
947
948 ext4_journal_stop(handle);
949 if (pos + len > inode->i_size) {
950 ext4_truncate_failed_write(inode);
951 /*
952 * If truncate failed early the inode might
953 * still be on the orphan list; we need to
954 * make sure the inode is removed from the
955 * orphan list in that case.
956 */
957 if (inode->i_nlink)
958 ext4_orphan_del(NULL, inode);
959 }
960 }
961
962 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
963 goto retry;
964 out:
965 return ret;
966 }
967
968 /* For write_end() in data=journal mode */
969 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
970 {
971 if (!buffer_mapped(bh) || buffer_freed(bh))
972 return 0;
973 set_buffer_uptodate(bh);
974 return ext4_handle_dirty_metadata(handle, NULL, bh);
975 }
976
977 static int ext4_generic_write_end(struct file *file,
978 struct address_space *mapping,
979 loff_t pos, unsigned len, unsigned copied,
980 struct page *page, void *fsdata)
981 {
982 int i_size_changed = 0;
983 struct inode *inode = mapping->host;
984 handle_t *handle = ext4_journal_current_handle();
985
986 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
987
988 /*
989 * No need to use i_size_read() here, the i_size
990 * cannot change under us because we hold i_mutex.
991 *
992 * But it's important to update i_size while still holding page lock:
993 * page writeout could otherwise come in and zero beyond i_size.
994 */
995 if (pos + copied > inode->i_size) {
996 i_size_write(inode, pos + copied);
997 i_size_changed = 1;
998 }
999
1000 if (pos + copied > EXT4_I(inode)->i_disksize) {
1001 /* We need to mark inode dirty even if
1002 * new_i_size is less that inode->i_size
1003 * bu greater than i_disksize.(hint delalloc)
1004 */
1005 ext4_update_i_disksize(inode, (pos + copied));
1006 i_size_changed = 1;
1007 }
1008 unlock_page(page);
1009 page_cache_release(page);
1010
1011 /*
1012 * Don't mark the inode dirty under page lock. First, it unnecessarily
1013 * makes the holding time of page lock longer. Second, it forces lock
1014 * ordering of page lock and transaction start for journaling
1015 * filesystems.
1016 */
1017 if (i_size_changed)
1018 ext4_mark_inode_dirty(handle, inode);
1019
1020 return copied;
1021 }
1022
1023 /*
1024 * We need to pick up the new inode size which generic_commit_write gave us
1025 * `file' can be NULL - eg, when called from page_symlink().
1026 *
1027 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1028 * buffers are managed internally.
1029 */
1030 static int ext4_ordered_write_end(struct file *file,
1031 struct address_space *mapping,
1032 loff_t pos, unsigned len, unsigned copied,
1033 struct page *page, void *fsdata)
1034 {
1035 handle_t *handle = ext4_journal_current_handle();
1036 struct inode *inode = mapping->host;
1037 int ret = 0, ret2;
1038
1039 trace_ext4_ordered_write_end(inode, pos, len, copied);
1040 ret = ext4_jbd2_file_inode(handle, inode);
1041
1042 if (ret == 0) {
1043 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1044 page, fsdata);
1045 copied = ret2;
1046 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1047 /* if we have allocated more blocks and copied
1048 * less. We will have blocks allocated outside
1049 * inode->i_size. So truncate them
1050 */
1051 ext4_orphan_add(handle, inode);
1052 if (ret2 < 0)
1053 ret = ret2;
1054 } else {
1055 unlock_page(page);
1056 page_cache_release(page);
1057 }
1058
1059 ret2 = ext4_journal_stop(handle);
1060 if (!ret)
1061 ret = ret2;
1062
1063 if (pos + len > inode->i_size) {
1064 ext4_truncate_failed_write(inode);
1065 /*
1066 * If truncate failed early the inode might still be
1067 * on the orphan list; we need to make sure the inode
1068 * is removed from the orphan list in that case.
1069 */
1070 if (inode->i_nlink)
1071 ext4_orphan_del(NULL, inode);
1072 }
1073
1074
1075 return ret ? ret : copied;
1076 }
1077
1078 static int ext4_writeback_write_end(struct file *file,
1079 struct address_space *mapping,
1080 loff_t pos, unsigned len, unsigned copied,
1081 struct page *page, void *fsdata)
1082 {
1083 handle_t *handle = ext4_journal_current_handle();
1084 struct inode *inode = mapping->host;
1085 int ret = 0, ret2;
1086
1087 trace_ext4_writeback_write_end(inode, pos, len, copied);
1088 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1089 page, fsdata);
1090 copied = ret2;
1091 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1092 /* if we have allocated more blocks and copied
1093 * less. We will have blocks allocated outside
1094 * inode->i_size. So truncate them
1095 */
1096 ext4_orphan_add(handle, inode);
1097
1098 if (ret2 < 0)
1099 ret = ret2;
1100
1101 ret2 = ext4_journal_stop(handle);
1102 if (!ret)
1103 ret = ret2;
1104
1105 if (pos + len > inode->i_size) {
1106 ext4_truncate_failed_write(inode);
1107 /*
1108 * If truncate failed early the inode might still be
1109 * on the orphan list; we need to make sure the inode
1110 * is removed from the orphan list in that case.
1111 */
1112 if (inode->i_nlink)
1113 ext4_orphan_del(NULL, inode);
1114 }
1115
1116 return ret ? ret : copied;
1117 }
1118
1119 static int ext4_journalled_write_end(struct file *file,
1120 struct address_space *mapping,
1121 loff_t pos, unsigned len, unsigned copied,
1122 struct page *page, void *fsdata)
1123 {
1124 handle_t *handle = ext4_journal_current_handle();
1125 struct inode *inode = mapping->host;
1126 int ret = 0, ret2;
1127 int partial = 0;
1128 unsigned from, to;
1129 loff_t new_i_size;
1130
1131 trace_ext4_journalled_write_end(inode, pos, len, copied);
1132 from = pos & (PAGE_CACHE_SIZE - 1);
1133 to = from + len;
1134
1135 BUG_ON(!ext4_handle_valid(handle));
1136
1137 if (copied < len) {
1138 if (!PageUptodate(page))
1139 copied = 0;
1140 page_zero_new_buffers(page, from+copied, to);
1141 }
1142
1143 ret = walk_page_buffers(handle, page_buffers(page), from,
1144 to, &partial, write_end_fn);
1145 if (!partial)
1146 SetPageUptodate(page);
1147 new_i_size = pos + copied;
1148 if (new_i_size > inode->i_size)
1149 i_size_write(inode, pos+copied);
1150 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1151 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1152 if (new_i_size > EXT4_I(inode)->i_disksize) {
1153 ext4_update_i_disksize(inode, new_i_size);
1154 ret2 = ext4_mark_inode_dirty(handle, inode);
1155 if (!ret)
1156 ret = ret2;
1157 }
1158
1159 unlock_page(page);
1160 page_cache_release(page);
1161 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1162 /* if we have allocated more blocks and copied
1163 * less. We will have blocks allocated outside
1164 * inode->i_size. So truncate them
1165 */
1166 ext4_orphan_add(handle, inode);
1167
1168 ret2 = ext4_journal_stop(handle);
1169 if (!ret)
1170 ret = ret2;
1171 if (pos + len > inode->i_size) {
1172 ext4_truncate_failed_write(inode);
1173 /*
1174 * If truncate failed early the inode might still be
1175 * on the orphan list; we need to make sure the inode
1176 * is removed from the orphan list in that case.
1177 */
1178 if (inode->i_nlink)
1179 ext4_orphan_del(NULL, inode);
1180 }
1181
1182 return ret ? ret : copied;
1183 }
1184
1185 /*
1186 * Reserve a single cluster located at lblock
1187 */
1188 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1189 {
1190 int retries = 0;
1191 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1192 struct ext4_inode_info *ei = EXT4_I(inode);
1193 unsigned int md_needed;
1194 int ret;
1195 ext4_lblk_t save_last_lblock;
1196 int save_len;
1197
1198 /*
1199 * We will charge metadata quota at writeout time; this saves
1200 * us from metadata over-estimation, though we may go over by
1201 * a small amount in the end. Here we just reserve for data.
1202 */
1203 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1204 if (ret)
1205 return ret;
1206
1207 /*
1208 * recalculate the amount of metadata blocks to reserve
1209 * in order to allocate nrblocks
1210 * worse case is one extent per block
1211 */
1212 repeat:
1213 spin_lock(&ei->i_block_reservation_lock);
1214 /*
1215 * ext4_calc_metadata_amount() has side effects, which we have
1216 * to be prepared undo if we fail to claim space.
1217 */
1218 save_len = ei->i_da_metadata_calc_len;
1219 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1220 md_needed = EXT4_NUM_B2C(sbi,
1221 ext4_calc_metadata_amount(inode, lblock));
1222 trace_ext4_da_reserve_space(inode, md_needed);
1223
1224 /*
1225 * We do still charge estimated metadata to the sb though;
1226 * we cannot afford to run out of free blocks.
1227 */
1228 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
1229 ei->i_da_metadata_calc_len = save_len;
1230 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1231 spin_unlock(&ei->i_block_reservation_lock);
1232 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1233 yield();
1234 goto repeat;
1235 }
1236 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1237 return -ENOSPC;
1238 }
1239 ei->i_reserved_data_blocks++;
1240 ei->i_reserved_meta_blocks += md_needed;
1241 spin_unlock(&ei->i_block_reservation_lock);
1242
1243 return 0; /* success */
1244 }
1245
1246 static void ext4_da_release_space(struct inode *inode, int to_free)
1247 {
1248 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1249 struct ext4_inode_info *ei = EXT4_I(inode);
1250
1251 if (!to_free)
1252 return; /* Nothing to release, exit */
1253
1254 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1255
1256 trace_ext4_da_release_space(inode, to_free);
1257 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1258 /*
1259 * if there aren't enough reserved blocks, then the
1260 * counter is messed up somewhere. Since this
1261 * function is called from invalidate page, it's
1262 * harmless to return without any action.
1263 */
1264 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1265 "ino %lu, to_free %d with only %d reserved "
1266 "data blocks", inode->i_ino, to_free,
1267 ei->i_reserved_data_blocks);
1268 WARN_ON(1);
1269 to_free = ei->i_reserved_data_blocks;
1270 }
1271 ei->i_reserved_data_blocks -= to_free;
1272
1273 if (ei->i_reserved_data_blocks == 0) {
1274 /*
1275 * We can release all of the reserved metadata blocks
1276 * only when we have written all of the delayed
1277 * allocation blocks.
1278 * Note that in case of bigalloc, i_reserved_meta_blocks,
1279 * i_reserved_data_blocks, etc. refer to number of clusters.
1280 */
1281 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
1282 ei->i_reserved_meta_blocks);
1283 ei->i_reserved_meta_blocks = 0;
1284 ei->i_da_metadata_calc_len = 0;
1285 }
1286
1287 /* update fs dirty data blocks counter */
1288 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1289
1290 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1291
1292 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1293 }
1294
1295 static void ext4_da_page_release_reservation(struct page *page,
1296 unsigned long offset)
1297 {
1298 int to_release = 0;
1299 struct buffer_head *head, *bh;
1300 unsigned int curr_off = 0;
1301 struct inode *inode = page->mapping->host;
1302 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1303 int num_clusters;
1304
1305 head = page_buffers(page);
1306 bh = head;
1307 do {
1308 unsigned int next_off = curr_off + bh->b_size;
1309
1310 if ((offset <= curr_off) && (buffer_delay(bh))) {
1311 to_release++;
1312 clear_buffer_delay(bh);
1313 clear_buffer_da_mapped(bh);
1314 }
1315 curr_off = next_off;
1316 } while ((bh = bh->b_this_page) != head);
1317
1318 /* If we have released all the blocks belonging to a cluster, then we
1319 * need to release the reserved space for that cluster. */
1320 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1321 while (num_clusters > 0) {
1322 ext4_fsblk_t lblk;
1323 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1324 ((num_clusters - 1) << sbi->s_cluster_bits);
1325 if (sbi->s_cluster_ratio == 1 ||
1326 !ext4_find_delalloc_cluster(inode, lblk, 1))
1327 ext4_da_release_space(inode, 1);
1328
1329 num_clusters--;
1330 }
1331 }
1332
1333 /*
1334 * Delayed allocation stuff
1335 */
1336
1337 /*
1338 * mpage_da_submit_io - walks through extent of pages and try to write
1339 * them with writepage() call back
1340 *
1341 * @mpd->inode: inode
1342 * @mpd->first_page: first page of the extent
1343 * @mpd->next_page: page after the last page of the extent
1344 *
1345 * By the time mpage_da_submit_io() is called we expect all blocks
1346 * to be allocated. this may be wrong if allocation failed.
1347 *
1348 * As pages are already locked by write_cache_pages(), we can't use it
1349 */
1350 static int mpage_da_submit_io(struct mpage_da_data *mpd,
1351 struct ext4_map_blocks *map)
1352 {
1353 struct pagevec pvec;
1354 unsigned long index, end;
1355 int ret = 0, err, nr_pages, i;
1356 struct inode *inode = mpd->inode;
1357 struct address_space *mapping = inode->i_mapping;
1358 loff_t size = i_size_read(inode);
1359 unsigned int len, block_start;
1360 struct buffer_head *bh, *page_bufs = NULL;
1361 int journal_data = ext4_should_journal_data(inode);
1362 sector_t pblock = 0, cur_logical = 0;
1363 struct ext4_io_submit io_submit;
1364
1365 BUG_ON(mpd->next_page <= mpd->first_page);
1366 memset(&io_submit, 0, sizeof(io_submit));
1367 /*
1368 * We need to start from the first_page to the next_page - 1
1369 * to make sure we also write the mapped dirty buffer_heads.
1370 * If we look at mpd->b_blocknr we would only be looking
1371 * at the currently mapped buffer_heads.
1372 */
1373 index = mpd->first_page;
1374 end = mpd->next_page - 1;
1375
1376 pagevec_init(&pvec, 0);
1377 while (index <= end) {
1378 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1379 if (nr_pages == 0)
1380 break;
1381 for (i = 0; i < nr_pages; i++) {
1382 int commit_write = 0, skip_page = 0;
1383 struct page *page = pvec.pages[i];
1384
1385 index = page->index;
1386 if (index > end)
1387 break;
1388
1389 if (index == size >> PAGE_CACHE_SHIFT)
1390 len = size & ~PAGE_CACHE_MASK;
1391 else
1392 len = PAGE_CACHE_SIZE;
1393 if (map) {
1394 cur_logical = index << (PAGE_CACHE_SHIFT -
1395 inode->i_blkbits);
1396 pblock = map->m_pblk + (cur_logical -
1397 map->m_lblk);
1398 }
1399 index++;
1400
1401 BUG_ON(!PageLocked(page));
1402 BUG_ON(PageWriteback(page));
1403
1404 /*
1405 * If the page does not have buffers (for
1406 * whatever reason), try to create them using
1407 * __block_write_begin. If this fails,
1408 * skip the page and move on.
1409 */
1410 if (!page_has_buffers(page)) {
1411 if (__block_write_begin(page, 0, len,
1412 noalloc_get_block_write)) {
1413 skip_page:
1414 unlock_page(page);
1415 continue;
1416 }
1417 commit_write = 1;
1418 }
1419
1420 bh = page_bufs = page_buffers(page);
1421 block_start = 0;
1422 do {
1423 if (!bh)
1424 goto skip_page;
1425 if (map && (cur_logical >= map->m_lblk) &&
1426 (cur_logical <= (map->m_lblk +
1427 (map->m_len - 1)))) {
1428 if (buffer_delay(bh)) {
1429 clear_buffer_delay(bh);
1430 bh->b_blocknr = pblock;
1431 }
1432 if (buffer_da_mapped(bh))
1433 clear_buffer_da_mapped(bh);
1434 if (buffer_unwritten(bh) ||
1435 buffer_mapped(bh))
1436 BUG_ON(bh->b_blocknr != pblock);
1437 if (map->m_flags & EXT4_MAP_UNINIT)
1438 set_buffer_uninit(bh);
1439 clear_buffer_unwritten(bh);
1440 }
1441
1442 /*
1443 * skip page if block allocation undone and
1444 * block is dirty
1445 */
1446 if (ext4_bh_delay_or_unwritten(NULL, bh))
1447 skip_page = 1;
1448 bh = bh->b_this_page;
1449 block_start += bh->b_size;
1450 cur_logical++;
1451 pblock++;
1452 } while (bh != page_bufs);
1453
1454 if (skip_page)
1455 goto skip_page;
1456
1457 if (commit_write)
1458 /* mark the buffer_heads as dirty & uptodate */
1459 block_commit_write(page, 0, len);
1460
1461 clear_page_dirty_for_io(page);
1462 /*
1463 * Delalloc doesn't support data journalling,
1464 * but eventually maybe we'll lift this
1465 * restriction.
1466 */
1467 if (unlikely(journal_data && PageChecked(page)))
1468 err = __ext4_journalled_writepage(page, len);
1469 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1470 err = ext4_bio_write_page(&io_submit, page,
1471 len, mpd->wbc);
1472 else if (buffer_uninit(page_bufs)) {
1473 ext4_set_bh_endio(page_bufs, inode);
1474 err = block_write_full_page_endio(page,
1475 noalloc_get_block_write,
1476 mpd->wbc, ext4_end_io_buffer_write);
1477 } else
1478 err = block_write_full_page(page,
1479 noalloc_get_block_write, mpd->wbc);
1480
1481 if (!err)
1482 mpd->pages_written++;
1483 /*
1484 * In error case, we have to continue because
1485 * remaining pages are still locked
1486 */
1487 if (ret == 0)
1488 ret = err;
1489 }
1490 pagevec_release(&pvec);
1491 }
1492 ext4_io_submit(&io_submit);
1493 return ret;
1494 }
1495
1496 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
1497 {
1498 int nr_pages, i;
1499 pgoff_t index, end;
1500 struct pagevec pvec;
1501 struct inode *inode = mpd->inode;
1502 struct address_space *mapping = inode->i_mapping;
1503
1504 index = mpd->first_page;
1505 end = mpd->next_page - 1;
1506
1507 pagevec_init(&pvec, 0);
1508 while (index <= end) {
1509 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1510 if (nr_pages == 0)
1511 break;
1512 for (i = 0; i < nr_pages; i++) {
1513 struct page *page = pvec.pages[i];
1514 if (page->index > end)
1515 break;
1516 BUG_ON(!PageLocked(page));
1517 BUG_ON(PageWriteback(page));
1518 block_invalidatepage(page, 0);
1519 ClearPageUptodate(page);
1520 unlock_page(page);
1521 }
1522 index = pvec.pages[nr_pages - 1]->index + 1;
1523 pagevec_release(&pvec);
1524 }
1525 return;
1526 }
1527
1528 static void ext4_print_free_blocks(struct inode *inode)
1529 {
1530 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1531 struct super_block *sb = inode->i_sb;
1532
1533 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1534 EXT4_C2B(EXT4_SB(inode->i_sb),
1535 ext4_count_free_clusters(inode->i_sb)));
1536 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1537 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1538 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1539 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1540 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1541 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1542 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1543 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1544 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1545 EXT4_I(inode)->i_reserved_data_blocks);
1546 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
1547 EXT4_I(inode)->i_reserved_meta_blocks);
1548 return;
1549 }
1550
1551 /*
1552 * mpage_da_map_and_submit - go through given space, map them
1553 * if necessary, and then submit them for I/O
1554 *
1555 * @mpd - bh describing space
1556 *
1557 * The function skips space we know is already mapped to disk blocks.
1558 *
1559 */
1560 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1561 {
1562 int err, blks, get_blocks_flags;
1563 struct ext4_map_blocks map, *mapp = NULL;
1564 sector_t next = mpd->b_blocknr;
1565 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1566 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1567 handle_t *handle = NULL;
1568
1569 /*
1570 * If the blocks are mapped already, or we couldn't accumulate
1571 * any blocks, then proceed immediately to the submission stage.
1572 */
1573 if ((mpd->b_size == 0) ||
1574 ((mpd->b_state & (1 << BH_Mapped)) &&
1575 !(mpd->b_state & (1 << BH_Delay)) &&
1576 !(mpd->b_state & (1 << BH_Unwritten))))
1577 goto submit_io;
1578
1579 handle = ext4_journal_current_handle();
1580 BUG_ON(!handle);
1581
1582 /*
1583 * Call ext4_map_blocks() to allocate any delayed allocation
1584 * blocks, or to convert an uninitialized extent to be
1585 * initialized (in the case where we have written into
1586 * one or more preallocated blocks).
1587 *
1588 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1589 * indicate that we are on the delayed allocation path. This
1590 * affects functions in many different parts of the allocation
1591 * call path. This flag exists primarily because we don't
1592 * want to change *many* call functions, so ext4_map_blocks()
1593 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1594 * inode's allocation semaphore is taken.
1595 *
1596 * If the blocks in questions were delalloc blocks, set
1597 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1598 * variables are updated after the blocks have been allocated.
1599 */
1600 map.m_lblk = next;
1601 map.m_len = max_blocks;
1602 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1603 if (ext4_should_dioread_nolock(mpd->inode))
1604 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1605 if (mpd->b_state & (1 << BH_Delay))
1606 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1607
1608 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1609 if (blks < 0) {
1610 struct super_block *sb = mpd->inode->i_sb;
1611
1612 err = blks;
1613 /*
1614 * If get block returns EAGAIN or ENOSPC and there
1615 * appears to be free blocks we will just let
1616 * mpage_da_submit_io() unlock all of the pages.
1617 */
1618 if (err == -EAGAIN)
1619 goto submit_io;
1620
1621 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
1622 mpd->retval = err;
1623 goto submit_io;
1624 }
1625
1626 /*
1627 * get block failure will cause us to loop in
1628 * writepages, because a_ops->writepage won't be able
1629 * to make progress. The page will be redirtied by
1630 * writepage and writepages will again try to write
1631 * the same.
1632 */
1633 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1634 ext4_msg(sb, KERN_CRIT,
1635 "delayed block allocation failed for inode %lu "
1636 "at logical offset %llu with max blocks %zd "
1637 "with error %d", mpd->inode->i_ino,
1638 (unsigned long long) next,
1639 mpd->b_size >> mpd->inode->i_blkbits, err);
1640 ext4_msg(sb, KERN_CRIT,
1641 "This should not happen!! Data will be lost\n");
1642 if (err == -ENOSPC)
1643 ext4_print_free_blocks(mpd->inode);
1644 }
1645 /* invalidate all the pages */
1646 ext4_da_block_invalidatepages(mpd);
1647
1648 /* Mark this page range as having been completed */
1649 mpd->io_done = 1;
1650 return;
1651 }
1652 BUG_ON(blks == 0);
1653
1654 mapp = ↦
1655 if (map.m_flags & EXT4_MAP_NEW) {
1656 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1657 int i;
1658
1659 for (i = 0; i < map.m_len; i++)
1660 unmap_underlying_metadata(bdev, map.m_pblk + i);
1661
1662 if (ext4_should_order_data(mpd->inode)) {
1663 err = ext4_jbd2_file_inode(handle, mpd->inode);
1664 if (err) {
1665 /* Only if the journal is aborted */
1666 mpd->retval = err;
1667 goto submit_io;
1668 }
1669 }
1670 }
1671
1672 /*
1673 * Update on-disk size along with block allocation.
1674 */
1675 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1676 if (disksize > i_size_read(mpd->inode))
1677 disksize = i_size_read(mpd->inode);
1678 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1679 ext4_update_i_disksize(mpd->inode, disksize);
1680 err = ext4_mark_inode_dirty(handle, mpd->inode);
1681 if (err)
1682 ext4_error(mpd->inode->i_sb,
1683 "Failed to mark inode %lu dirty",
1684 mpd->inode->i_ino);
1685 }
1686
1687 submit_io:
1688 mpage_da_submit_io(mpd, mapp);
1689 mpd->io_done = 1;
1690 }
1691
1692 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1693 (1 << BH_Delay) | (1 << BH_Unwritten))
1694
1695 /*
1696 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1697 *
1698 * @mpd->lbh - extent of blocks
1699 * @logical - logical number of the block in the file
1700 * @bh - bh of the block (used to access block's state)
1701 *
1702 * the function is used to collect contig. blocks in same state
1703 */
1704 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1705 sector_t logical, size_t b_size,
1706 unsigned long b_state)
1707 {
1708 sector_t next;
1709 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1710
1711 /*
1712 * XXX Don't go larger than mballoc is willing to allocate
1713 * This is a stopgap solution. We eventually need to fold
1714 * mpage_da_submit_io() into this function and then call
1715 * ext4_map_blocks() multiple times in a loop
1716 */
1717 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1718 goto flush_it;
1719
1720 /* check if thereserved journal credits might overflow */
1721 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1722 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1723 /*
1724 * With non-extent format we are limited by the journal
1725 * credit available. Total credit needed to insert
1726 * nrblocks contiguous blocks is dependent on the
1727 * nrblocks. So limit nrblocks.
1728 */
1729 goto flush_it;
1730 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1731 EXT4_MAX_TRANS_DATA) {
1732 /*
1733 * Adding the new buffer_head would make it cross the
1734 * allowed limit for which we have journal credit
1735 * reserved. So limit the new bh->b_size
1736 */
1737 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1738 mpd->inode->i_blkbits;
1739 /* we will do mpage_da_submit_io in the next loop */
1740 }
1741 }
1742 /*
1743 * First block in the extent
1744 */
1745 if (mpd->b_size == 0) {
1746 mpd->b_blocknr = logical;
1747 mpd->b_size = b_size;
1748 mpd->b_state = b_state & BH_FLAGS;
1749 return;
1750 }
1751
1752 next = mpd->b_blocknr + nrblocks;
1753 /*
1754 * Can we merge the block to our big extent?
1755 */
1756 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1757 mpd->b_size += b_size;
1758 return;
1759 }
1760
1761 flush_it:
1762 /*
1763 * We couldn't merge the block to our extent, so we
1764 * need to flush current extent and start new one
1765 */
1766 mpage_da_map_and_submit(mpd);
1767 return;
1768 }
1769
1770 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1771 {
1772 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1773 }
1774
1775 /*
1776 * This function is grabs code from the very beginning of
1777 * ext4_map_blocks, but assumes that the caller is from delayed write
1778 * time. This function looks up the requested blocks and sets the
1779 * buffer delay bit under the protection of i_data_sem.
1780 */
1781 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1782 struct ext4_map_blocks *map,
1783 struct buffer_head *bh)
1784 {
1785 int retval;
1786 sector_t invalid_block = ~((sector_t) 0xffff);
1787
1788 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1789 invalid_block = ~0;
1790
1791 map->m_flags = 0;
1792 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1793 "logical block %lu\n", inode->i_ino, map->m_len,
1794 (unsigned long) map->m_lblk);
1795 /*
1796 * Try to see if we can get the block without requesting a new
1797 * file system block.
1798 */
1799 down_read((&EXT4_I(inode)->i_data_sem));
1800 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1801 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1802 else
1803 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1804
1805 if (retval == 0) {
1806 /*
1807 * XXX: __block_prepare_write() unmaps passed block,
1808 * is it OK?
1809 */
1810 /* If the block was allocated from previously allocated cluster,
1811 * then we dont need to reserve it again. */
1812 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1813 retval = ext4_da_reserve_space(inode, iblock);
1814 if (retval)
1815 /* not enough space to reserve */
1816 goto out_unlock;
1817 }
1818
1819 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1820 * and it should not appear on the bh->b_state.
1821 */
1822 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1823
1824 map_bh(bh, inode->i_sb, invalid_block);
1825 set_buffer_new(bh);
1826 set_buffer_delay(bh);
1827 }
1828
1829 out_unlock:
1830 up_read((&EXT4_I(inode)->i_data_sem));
1831
1832 return retval;
1833 }
1834
1835 /*
1836 * This is a special get_blocks_t callback which is used by
1837 * ext4_da_write_begin(). It will either return mapped block or
1838 * reserve space for a single block.
1839 *
1840 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1841 * We also have b_blocknr = -1 and b_bdev initialized properly
1842 *
1843 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1844 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1845 * initialized properly.
1846 */
1847 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1848 struct buffer_head *bh, int create)
1849 {
1850 struct ext4_map_blocks map;
1851 int ret = 0;
1852
1853 BUG_ON(create == 0);
1854 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1855
1856 map.m_lblk = iblock;
1857 map.m_len = 1;
1858
1859 /*
1860 * first, we need to know whether the block is allocated already
1861 * preallocated blocks are unmapped but should treated
1862 * the same as allocated blocks.
1863 */
1864 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1865 if (ret <= 0)
1866 return ret;
1867
1868 map_bh(bh, inode->i_sb, map.m_pblk);
1869 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1870
1871 if (buffer_unwritten(bh)) {
1872 /* A delayed write to unwritten bh should be marked
1873 * new and mapped. Mapped ensures that we don't do
1874 * get_block multiple times when we write to the same
1875 * offset and new ensures that we do proper zero out
1876 * for partial write.
1877 */
1878 set_buffer_new(bh);
1879 set_buffer_mapped(bh);
1880 }
1881 return 0;
1882 }
1883
1884 /*
1885 * This function is used as a standard get_block_t calback function
1886 * when there is no desire to allocate any blocks. It is used as a
1887 * callback function for block_write_begin() and block_write_full_page().
1888 * These functions should only try to map a single block at a time.
1889 *
1890 * Since this function doesn't do block allocations even if the caller
1891 * requests it by passing in create=1, it is critically important that
1892 * any caller checks to make sure that any buffer heads are returned
1893 * by this function are either all already mapped or marked for
1894 * delayed allocation before calling block_write_full_page(). Otherwise,
1895 * b_blocknr could be left unitialized, and the page write functions will
1896 * be taken by surprise.
1897 */
1898 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1899 struct buffer_head *bh_result, int create)
1900 {
1901 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1902 return _ext4_get_block(inode, iblock, bh_result, 0);
1903 }
1904
1905 static int bget_one(handle_t *handle, struct buffer_head *bh)
1906 {
1907 get_bh(bh);
1908 return 0;
1909 }
1910
1911 static int bput_one(handle_t *handle, struct buffer_head *bh)
1912 {
1913 put_bh(bh);
1914 return 0;
1915 }
1916
1917 static int __ext4_journalled_writepage(struct page *page,
1918 unsigned int len)
1919 {
1920 struct address_space *mapping = page->mapping;
1921 struct inode *inode = mapping->host;
1922 struct buffer_head *page_bufs;
1923 handle_t *handle = NULL;
1924 int ret = 0;
1925 int err;
1926
1927 ClearPageChecked(page);
1928 page_bufs = page_buffers(page);
1929 BUG_ON(!page_bufs);
1930 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1931 /* As soon as we unlock the page, it can go away, but we have
1932 * references to buffers so we are safe */
1933 unlock_page(page);
1934
1935 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1936 if (IS_ERR(handle)) {
1937 ret = PTR_ERR(handle);
1938 goto out;
1939 }
1940
1941 BUG_ON(!ext4_handle_valid(handle));
1942
1943 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1944 do_journal_get_write_access);
1945
1946 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1947 write_end_fn);
1948 if (ret == 0)
1949 ret = err;
1950 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1951 err = ext4_journal_stop(handle);
1952 if (!ret)
1953 ret = err;
1954
1955 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
1956 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1957 out:
1958 return ret;
1959 }
1960
1961 /*
1962 * Note that we don't need to start a transaction unless we're journaling data
1963 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1964 * need to file the inode to the transaction's list in ordered mode because if
1965 * we are writing back data added by write(), the inode is already there and if
1966 * we are writing back data modified via mmap(), no one guarantees in which
1967 * transaction the data will hit the disk. In case we are journaling data, we
1968 * cannot start transaction directly because transaction start ranks above page
1969 * lock so we have to do some magic.
1970 *
1971 * This function can get called via...
1972 * - ext4_da_writepages after taking page lock (have journal handle)
1973 * - journal_submit_inode_data_buffers (no journal handle)
1974 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1975 * - grab_page_cache when doing write_begin (have journal handle)
1976 *
1977 * We don't do any block allocation in this function. If we have page with
1978 * multiple blocks we need to write those buffer_heads that are mapped. This
1979 * is important for mmaped based write. So if we do with blocksize 1K
1980 * truncate(f, 1024);
1981 * a = mmap(f, 0, 4096);
1982 * a[0] = 'a';
1983 * truncate(f, 4096);
1984 * we have in the page first buffer_head mapped via page_mkwrite call back
1985 * but other buffer_heads would be unmapped but dirty (dirty done via the
1986 * do_wp_page). So writepage should write the first block. If we modify
1987 * the mmap area beyond 1024 we will again get a page_fault and the
1988 * page_mkwrite callback will do the block allocation and mark the
1989 * buffer_heads mapped.
1990 *
1991 * We redirty the page if we have any buffer_heads that is either delay or
1992 * unwritten in the page.
1993 *
1994 * We can get recursively called as show below.
1995 *
1996 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1997 * ext4_writepage()
1998 *
1999 * But since we don't do any block allocation we should not deadlock.
2000 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2001 */
2002 static int ext4_writepage(struct page *page,
2003 struct writeback_control *wbc)
2004 {
2005 int ret = 0, commit_write = 0;
2006 loff_t size;
2007 unsigned int len;
2008 struct buffer_head *page_bufs = NULL;
2009 struct inode *inode = page->mapping->host;
2010
2011 trace_ext4_writepage(page);
2012 size = i_size_read(inode);
2013 if (page->index == size >> PAGE_CACHE_SHIFT)
2014 len = size & ~PAGE_CACHE_MASK;
2015 else
2016 len = PAGE_CACHE_SIZE;
2017
2018 /*
2019 * If the page does not have buffers (for whatever reason),
2020 * try to create them using __block_write_begin. If this
2021 * fails, redirty the page and move on.
2022 */
2023 if (!page_has_buffers(page)) {
2024 if (__block_write_begin(page, 0, len,
2025 noalloc_get_block_write)) {
2026 redirty_page:
2027 redirty_page_for_writepage(wbc, page);
2028 unlock_page(page);
2029 return 0;
2030 }
2031 commit_write = 1;
2032 }
2033 page_bufs = page_buffers(page);
2034 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2035 ext4_bh_delay_or_unwritten)) {
2036 /*
2037 * We don't want to do block allocation, so redirty
2038 * the page and return. We may reach here when we do
2039 * a journal commit via journal_submit_inode_data_buffers.
2040 * We can also reach here via shrink_page_list but it
2041 * should never be for direct reclaim so warn if that
2042 * happens
2043 */
2044 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2045 PF_MEMALLOC);
2046 goto redirty_page;
2047 }
2048 if (commit_write)
2049 /* now mark the buffer_heads as dirty and uptodate */
2050 block_commit_write(page, 0, len);
2051
2052 if (PageChecked(page) && ext4_should_journal_data(inode))
2053 /*
2054 * It's mmapped pagecache. Add buffers and journal it. There
2055 * doesn't seem much point in redirtying the page here.
2056 */
2057 return __ext4_journalled_writepage(page, len);
2058
2059 if (buffer_uninit(page_bufs)) {
2060 ext4_set_bh_endio(page_bufs, inode);
2061 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2062 wbc, ext4_end_io_buffer_write);
2063 } else
2064 ret = block_write_full_page(page, noalloc_get_block_write,
2065 wbc);
2066
2067 return ret;
2068 }
2069
2070 /*
2071 * This is called via ext4_da_writepages() to
2072 * calculate the total number of credits to reserve to fit
2073 * a single extent allocation into a single transaction,
2074 * ext4_da_writpeages() will loop calling this before
2075 * the block allocation.
2076 */
2077
2078 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2079 {
2080 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2081
2082 /*
2083 * With non-extent format the journal credit needed to
2084 * insert nrblocks contiguous block is dependent on
2085 * number of contiguous block. So we will limit
2086 * number of contiguous block to a sane value
2087 */
2088 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2089 (max_blocks > EXT4_MAX_TRANS_DATA))
2090 max_blocks = EXT4_MAX_TRANS_DATA;
2091
2092 return ext4_chunk_trans_blocks(inode, max_blocks);
2093 }
2094
2095 /*
2096 * write_cache_pages_da - walk the list of dirty pages of the given
2097 * address space and accumulate pages that need writing, and call
2098 * mpage_da_map_and_submit to map a single contiguous memory region
2099 * and then write them.
2100 */
2101 static int write_cache_pages_da(struct address_space *mapping,
2102 struct writeback_control *wbc,
2103 struct mpage_da_data *mpd,
2104 pgoff_t *done_index)
2105 {
2106 struct buffer_head *bh, *head;
2107 struct inode *inode = mapping->host;
2108 struct pagevec pvec;
2109 unsigned int nr_pages;
2110 sector_t logical;
2111 pgoff_t index, end;
2112 long nr_to_write = wbc->nr_to_write;
2113 int i, tag, ret = 0;
2114
2115 memset(mpd, 0, sizeof(struct mpage_da_data));
2116 mpd->wbc = wbc;
2117 mpd->inode = inode;
2118 pagevec_init(&pvec, 0);
2119 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2120 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2121
2122 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2123 tag = PAGECACHE_TAG_TOWRITE;
2124 else
2125 tag = PAGECACHE_TAG_DIRTY;
2126
2127 *done_index = index;
2128 while (index <= end) {
2129 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2130 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2131 if (nr_pages == 0)
2132 return 0;
2133
2134 for (i = 0; i < nr_pages; i++) {
2135 struct page *page = pvec.pages[i];
2136
2137 /*
2138 * At this point, the page may be truncated or
2139 * invalidated (changing page->mapping to NULL), or
2140 * even swizzled back from swapper_space to tmpfs file
2141 * mapping. However, page->index will not change
2142 * because we have a reference on the page.
2143 */
2144 if (page->index > end)
2145 goto out;
2146
2147 *done_index = page->index + 1;
2148
2149 /*
2150 * If we can't merge this page, and we have
2151 * accumulated an contiguous region, write it
2152 */
2153 if ((mpd->next_page != page->index) &&
2154 (mpd->next_page != mpd->first_page)) {
2155 mpage_da_map_and_submit(mpd);
2156 goto ret_extent_tail;
2157 }
2158
2159 lock_page(page);
2160
2161 /*
2162 * If the page is no longer dirty, or its
2163 * mapping no longer corresponds to inode we
2164 * are writing (which means it has been
2165 * truncated or invalidated), or the page is
2166 * already under writeback and we are not
2167 * doing a data integrity writeback, skip the page
2168 */
2169 if (!PageDirty(page) ||
2170 (PageWriteback(page) &&
2171 (wbc->sync_mode == WB_SYNC_NONE)) ||
2172 unlikely(page->mapping != mapping)) {
2173 unlock_page(page);
2174 continue;
2175 }
2176
2177 wait_on_page_writeback(page);
2178 BUG_ON(PageWriteback(page));
2179
2180 if (mpd->next_page != page->index)
2181 mpd->first_page = page->index;
2182 mpd->next_page = page->index + 1;
2183 logical = (sector_t) page->index <<
2184 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2185
2186 if (!page_has_buffers(page)) {
2187 mpage_add_bh_to_extent(mpd, logical,
2188 PAGE_CACHE_SIZE,
2189 (1 << BH_Dirty) | (1 << BH_Uptodate));
2190 if (mpd->io_done)
2191 goto ret_extent_tail;
2192 } else {
2193 /*
2194 * Page with regular buffer heads,
2195 * just add all dirty ones
2196 */
2197 head = page_buffers(page);
2198 bh = head;
2199 do {
2200 BUG_ON(buffer_locked(bh));
2201 /*
2202 * We need to try to allocate
2203 * unmapped blocks in the same page.
2204 * Otherwise we won't make progress
2205 * with the page in ext4_writepage
2206 */
2207 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2208 mpage_add_bh_to_extent(mpd, logical,
2209 bh->b_size,
2210 bh->b_state);
2211 if (mpd->io_done)
2212 goto ret_extent_tail;
2213 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2214 /*
2215 * mapped dirty buffer. We need
2216 * to update the b_state
2217 * because we look at b_state
2218 * in mpage_da_map_blocks. We
2219 * don't update b_size because
2220 * if we find an unmapped
2221 * buffer_head later we need to
2222 * use the b_state flag of that
2223 * buffer_head.
2224 */
2225 if (mpd->b_size == 0)
2226 mpd->b_state = bh->b_state & BH_FLAGS;
2227 }
2228 logical++;
2229 } while ((bh = bh->b_this_page) != head);
2230 }
2231
2232 if (nr_to_write > 0) {
2233 nr_to_write--;
2234 if (nr_to_write == 0 &&
2235 wbc->sync_mode == WB_SYNC_NONE)
2236 /*
2237 * We stop writing back only if we are
2238 * not doing integrity sync. In case of
2239 * integrity sync we have to keep going
2240 * because someone may be concurrently
2241 * dirtying pages, and we might have
2242 * synced a lot of newly appeared dirty
2243 * pages, but have not synced all of the
2244 * old dirty pages.
2245 */
2246 goto out;
2247 }
2248 }
2249 pagevec_release(&pvec);
2250 cond_resched();
2251 }
2252 return 0;
2253 ret_extent_tail:
2254 ret = MPAGE_DA_EXTENT_TAIL;
2255 out:
2256 pagevec_release(&pvec);
2257 cond_resched();
2258 return ret;
2259 }
2260
2261
2262 static int ext4_da_writepages(struct address_space *mapping,
2263 struct writeback_control *wbc)
2264 {
2265 pgoff_t index;
2266 int range_whole = 0;
2267 handle_t *handle = NULL;
2268 struct mpage_da_data mpd;
2269 struct inode *inode = mapping->host;
2270 int pages_written = 0;
2271 unsigned int max_pages;
2272 int range_cyclic, cycled = 1, io_done = 0;
2273 int needed_blocks, ret = 0;
2274 long desired_nr_to_write, nr_to_writebump = 0;
2275 loff_t range_start = wbc->range_start;
2276 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2277 pgoff_t done_index = 0;
2278 pgoff_t end;
2279 struct blk_plug plug;
2280
2281 trace_ext4_da_writepages(inode, wbc);
2282
2283 /*
2284 * No pages to write? This is mainly a kludge to avoid starting
2285 * a transaction for special inodes like journal inode on last iput()
2286 * because that could violate lock ordering on umount
2287 */
2288 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2289 return 0;
2290
2291 /*
2292 * If the filesystem has aborted, it is read-only, so return
2293 * right away instead of dumping stack traces later on that
2294 * will obscure the real source of the problem. We test
2295 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2296 * the latter could be true if the filesystem is mounted
2297 * read-only, and in that case, ext4_da_writepages should
2298 * *never* be called, so if that ever happens, we would want
2299 * the stack trace.
2300 */
2301 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2302 return -EROFS;
2303
2304 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2305 range_whole = 1;
2306
2307 range_cyclic = wbc->range_cyclic;
2308 if (wbc->range_cyclic) {
2309 index = mapping->writeback_index;
2310 if (index)
2311 cycled = 0;
2312 wbc->range_start = index << PAGE_CACHE_SHIFT;
2313 wbc->range_end = LLONG_MAX;
2314 wbc->range_cyclic = 0;
2315 end = -1;
2316 } else {
2317 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2318 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2319 }
2320
2321 /*
2322 * This works around two forms of stupidity. The first is in
2323 * the writeback code, which caps the maximum number of pages
2324 * written to be 1024 pages. This is wrong on multiple
2325 * levels; different architectues have a different page size,
2326 * which changes the maximum amount of data which gets
2327 * written. Secondly, 4 megabytes is way too small. XFS
2328 * forces this value to be 16 megabytes by multiplying
2329 * nr_to_write parameter by four, and then relies on its
2330 * allocator to allocate larger extents to make them
2331 * contiguous. Unfortunately this brings us to the second
2332 * stupidity, which is that ext4's mballoc code only allocates
2333 * at most 2048 blocks. So we force contiguous writes up to
2334 * the number of dirty blocks in the inode, or
2335 * sbi->max_writeback_mb_bump whichever is smaller.
2336 */
2337 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2338 if (!range_cyclic && range_whole) {
2339 if (wbc->nr_to_write == LONG_MAX)
2340 desired_nr_to_write = wbc->nr_to_write;
2341 else
2342 desired_nr_to_write = wbc->nr_to_write * 8;
2343 } else
2344 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2345 max_pages);
2346 if (desired_nr_to_write > max_pages)
2347 desired_nr_to_write = max_pages;
2348
2349 if (wbc->nr_to_write < desired_nr_to_write) {
2350 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2351 wbc->nr_to_write = desired_nr_to_write;
2352 }
2353
2354 retry:
2355 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2356 tag_pages_for_writeback(mapping, index, end);
2357
2358 blk_start_plug(&plug);
2359 while (!ret && wbc->nr_to_write > 0) {
2360
2361 /*
2362 * we insert one extent at a time. So we need
2363 * credit needed for single extent allocation.
2364 * journalled mode is currently not supported
2365 * by delalloc
2366 */
2367 BUG_ON(ext4_should_journal_data(inode));
2368 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2369
2370 /* start a new transaction*/
2371 handle = ext4_journal_start(inode, needed_blocks);
2372 if (IS_ERR(handle)) {
2373 ret = PTR_ERR(handle);
2374 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2375 "%ld pages, ino %lu; err %d", __func__,
2376 wbc->nr_to_write, inode->i_ino, ret);
2377 blk_finish_plug(&plug);
2378 goto out_writepages;
2379 }
2380
2381 /*
2382 * Now call write_cache_pages_da() to find the next
2383 * contiguous region of logical blocks that need
2384 * blocks to be allocated by ext4 and submit them.
2385 */
2386 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2387 /*
2388 * If we have a contiguous extent of pages and we
2389 * haven't done the I/O yet, map the blocks and submit
2390 * them for I/O.
2391 */
2392 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2393 mpage_da_map_and_submit(&mpd);
2394 ret = MPAGE_DA_EXTENT_TAIL;
2395 }
2396 trace_ext4_da_write_pages(inode, &mpd);
2397 wbc->nr_to_write -= mpd.pages_written;
2398
2399 ext4_journal_stop(handle);
2400
2401 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2402 /* commit the transaction which would
2403 * free blocks released in the transaction
2404 * and try again
2405 */
2406 jbd2_journal_force_commit_nested(sbi->s_journal);
2407 ret = 0;
2408 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2409 /*
2410 * Got one extent now try with rest of the pages.
2411 * If mpd.retval is set -EIO, journal is aborted.
2412 * So we don't need to write any more.
2413 */
2414 pages_written += mpd.pages_written;
2415 ret = mpd.retval;
2416 io_done = 1;
2417 } else if (wbc->nr_to_write)
2418 /*
2419 * There is no more writeout needed
2420 * or we requested for a noblocking writeout
2421 * and we found the device congested
2422 */
2423 break;
2424 }
2425 blk_finish_plug(&plug);
2426 if (!io_done && !cycled) {
2427 cycled = 1;
2428 index = 0;
2429 wbc->range_start = index << PAGE_CACHE_SHIFT;
2430 wbc->range_end = mapping->writeback_index - 1;
2431 goto retry;
2432 }
2433
2434 /* Update index */
2435 wbc->range_cyclic = range_cyclic;
2436 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2437 /*
2438 * set the writeback_index so that range_cyclic
2439 * mode will write it back later
2440 */
2441 mapping->writeback_index = done_index;
2442
2443 out_writepages:
2444 wbc->nr_to_write -= nr_to_writebump;
2445 wbc->range_start = range_start;
2446 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2447 return ret;
2448 }
2449
2450 #define FALL_BACK_TO_NONDELALLOC 1
2451 static int ext4_nonda_switch(struct super_block *sb)
2452 {
2453 s64 free_blocks, dirty_blocks;
2454 struct ext4_sb_info *sbi = EXT4_SB(sb);
2455
2456 /*
2457 * switch to non delalloc mode if we are running low
2458 * on free block. The free block accounting via percpu
2459 * counters can get slightly wrong with percpu_counter_batch getting
2460 * accumulated on each CPU without updating global counters
2461 * Delalloc need an accurate free block accounting. So switch
2462 * to non delalloc when we are near to error range.
2463 */
2464 free_blocks = EXT4_C2B(sbi,
2465 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2466 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2467 /*
2468 * Start pushing delalloc when 1/2 of free blocks are dirty.
2469 */
2470 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2471 !writeback_in_progress(sb->s_bdi) &&
2472 down_read_trylock(&sb->s_umount)) {
2473 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2474 up_read(&sb->s_umount);
2475 }
2476
2477 if (2 * free_blocks < 3 * dirty_blocks ||
2478 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
2479 /*
2480 * free block count is less than 150% of dirty blocks
2481 * or free blocks is less than watermark
2482 */
2483 return 1;
2484 }
2485 return 0;
2486 }
2487
2488 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2489 loff_t pos, unsigned len, unsigned flags,
2490 struct page **pagep, void **fsdata)
2491 {
2492 int ret, retries = 0;
2493 struct page *page;
2494 pgoff_t index;
2495 struct inode *inode = mapping->host;
2496 handle_t *handle;
2497
2498 index = pos >> PAGE_CACHE_SHIFT;
2499
2500 if (ext4_nonda_switch(inode->i_sb)) {
2501 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2502 return ext4_write_begin(file, mapping, pos,
2503 len, flags, pagep, fsdata);
2504 }
2505 *fsdata = (void *)0;
2506 trace_ext4_da_write_begin(inode, pos, len, flags);
2507 retry:
2508 /*
2509 * With delayed allocation, we don't log the i_disksize update
2510 * if there is delayed block allocation. But we still need
2511 * to journalling the i_disksize update if writes to the end
2512 * of file which has an already mapped buffer.
2513 */
2514 handle = ext4_journal_start(inode, 1);
2515 if (IS_ERR(handle)) {
2516 ret = PTR_ERR(handle);
2517 goto out;
2518 }
2519 /* We cannot recurse into the filesystem as the transaction is already
2520 * started */
2521 flags |= AOP_FLAG_NOFS;
2522
2523 page = grab_cache_page_write_begin(mapping, index, flags);
2524 if (!page) {
2525 ext4_journal_stop(handle);
2526 ret = -ENOMEM;
2527 goto out;
2528 }
2529 *pagep = page;
2530
2531 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2532 if (ret < 0) {
2533 unlock_page(page);
2534 ext4_journal_stop(handle);
2535 page_cache_release(page);
2536 /*
2537 * block_write_begin may have instantiated a few blocks
2538 * outside i_size. Trim these off again. Don't need
2539 * i_size_read because we hold i_mutex.
2540 */
2541 if (pos + len > inode->i_size)
2542 ext4_truncate_failed_write(inode);
2543 }
2544
2545 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2546 goto retry;
2547 out:
2548 return ret;
2549 }
2550
2551 /*
2552 * Check if we should update i_disksize
2553 * when write to the end of file but not require block allocation
2554 */
2555 static int ext4_da_should_update_i_disksize(struct page *page,
2556 unsigned long offset)
2557 {
2558 struct buffer_head *bh;
2559 struct inode *inode = page->mapping->host;
2560 unsigned int idx;
2561 int i;
2562
2563 bh = page_buffers(page);
2564 idx = offset >> inode->i_blkbits;
2565
2566 for (i = 0; i < idx; i++)
2567 bh = bh->b_this_page;
2568
2569 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2570 return 0;
2571 return 1;
2572 }
2573
2574 static int ext4_da_write_end(struct file *file,
2575 struct address_space *mapping,
2576 loff_t pos, unsigned len, unsigned copied,
2577 struct page *page, void *fsdata)
2578 {
2579 struct inode *inode = mapping->host;
2580 int ret = 0, ret2;
2581 handle_t *handle = ext4_journal_current_handle();
2582 loff_t new_i_size;
2583 unsigned long start, end;
2584 int write_mode = (int)(unsigned long)fsdata;
2585
2586 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2587 switch (ext4_inode_journal_mode(inode)) {
2588 case EXT4_INODE_ORDERED_DATA_MODE:
2589 return ext4_ordered_write_end(file, mapping, pos,
2590 len, copied, page, fsdata);
2591 case EXT4_INODE_WRITEBACK_DATA_MODE:
2592 return ext4_writeback_write_end(file, mapping, pos,
2593 len, copied, page, fsdata);
2594 default:
2595 BUG();
2596 }
2597 }
2598
2599 trace_ext4_da_write_end(inode, pos, len, copied);
2600 start = pos & (PAGE_CACHE_SIZE - 1);
2601 end = start + copied - 1;
2602
2603 /*
2604 * generic_write_end() will run mark_inode_dirty() if i_size
2605 * changes. So let's piggyback the i_disksize mark_inode_dirty
2606 * into that.
2607 */
2608
2609 new_i_size = pos + copied;
2610 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
2611 if (ext4_da_should_update_i_disksize(page, end)) {
2612 down_write(&EXT4_I(inode)->i_data_sem);
2613 if (new_i_size > EXT4_I(inode)->i_disksize) {
2614 /*
2615 * Updating i_disksize when extending file
2616 * without needing block allocation
2617 */
2618 if (ext4_should_order_data(inode))
2619 ret = ext4_jbd2_file_inode(handle,
2620 inode);
2621
2622 EXT4_I(inode)->i_disksize = new_i_size;
2623 }
2624 up_write(&EXT4_I(inode)->i_data_sem);
2625 /* We need to mark inode dirty even if
2626 * new_i_size is less that inode->i_size
2627 * bu greater than i_disksize.(hint delalloc)
2628 */
2629 ext4_mark_inode_dirty(handle, inode);
2630 }
2631 }
2632 ret2 = generic_write_end(file, mapping, pos, len, copied,
2633 page, fsdata);
2634 copied = ret2;
2635 if (ret2 < 0)
2636 ret = ret2;
2637 ret2 = ext4_journal_stop(handle);
2638 if (!ret)
2639 ret = ret2;
2640
2641 return ret ? ret : copied;
2642 }
2643
2644 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2645 {
2646 /*
2647 * Drop reserved blocks
2648 */
2649 BUG_ON(!PageLocked(page));
2650 if (!page_has_buffers(page))
2651 goto out;
2652
2653 ext4_da_page_release_reservation(page, offset);
2654
2655 out:
2656 ext4_invalidatepage(page, offset);
2657
2658 return;
2659 }
2660
2661 /*
2662 * Force all delayed allocation blocks to be allocated for a given inode.
2663 */
2664 int ext4_alloc_da_blocks(struct inode *inode)
2665 {
2666 trace_ext4_alloc_da_blocks(inode);
2667
2668 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2669 !EXT4_I(inode)->i_reserved_meta_blocks)
2670 return 0;
2671
2672 /*
2673 * We do something simple for now. The filemap_flush() will
2674 * also start triggering a write of the data blocks, which is
2675 * not strictly speaking necessary (and for users of
2676 * laptop_mode, not even desirable). However, to do otherwise
2677 * would require replicating code paths in:
2678 *
2679 * ext4_da_writepages() ->
2680 * write_cache_pages() ---> (via passed in callback function)
2681 * __mpage_da_writepage() -->
2682 * mpage_add_bh_to_extent()
2683 * mpage_da_map_blocks()
2684 *
2685 * The problem is that write_cache_pages(), located in
2686 * mm/page-writeback.c, marks pages clean in preparation for
2687 * doing I/O, which is not desirable if we're not planning on
2688 * doing I/O at all.
2689 *
2690 * We could call write_cache_pages(), and then redirty all of
2691 * the pages by calling redirty_page_for_writepage() but that
2692 * would be ugly in the extreme. So instead we would need to
2693 * replicate parts of the code in the above functions,
2694 * simplifying them because we wouldn't actually intend to
2695 * write out the pages, but rather only collect contiguous
2696 * logical block extents, call the multi-block allocator, and
2697 * then update the buffer heads with the block allocations.
2698 *
2699 * For now, though, we'll cheat by calling filemap_flush(),
2700 * which will map the blocks, and start the I/O, but not
2701 * actually wait for the I/O to complete.
2702 */
2703 return filemap_flush(inode->i_mapping);
2704 }
2705
2706 /*
2707 * bmap() is special. It gets used by applications such as lilo and by
2708 * the swapper to find the on-disk block of a specific piece of data.
2709 *
2710 * Naturally, this is dangerous if the block concerned is still in the
2711 * journal. If somebody makes a swapfile on an ext4 data-journaling
2712 * filesystem and enables swap, then they may get a nasty shock when the
2713 * data getting swapped to that swapfile suddenly gets overwritten by
2714 * the original zero's written out previously to the journal and
2715 * awaiting writeback in the kernel's buffer cache.
2716 *
2717 * So, if we see any bmap calls here on a modified, data-journaled file,
2718 * take extra steps to flush any blocks which might be in the cache.
2719 */
2720 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2721 {
2722 struct inode *inode = mapping->host;
2723 journal_t *journal;
2724 int err;
2725
2726 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2727 test_opt(inode->i_sb, DELALLOC)) {
2728 /*
2729 * With delalloc we want to sync the file
2730 * so that we can make sure we allocate
2731 * blocks for file
2732 */
2733 filemap_write_and_wait(mapping);
2734 }
2735
2736 if (EXT4_JOURNAL(inode) &&
2737 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2738 /*
2739 * This is a REALLY heavyweight approach, but the use of
2740 * bmap on dirty files is expected to be extremely rare:
2741 * only if we run lilo or swapon on a freshly made file
2742 * do we expect this to happen.
2743 *
2744 * (bmap requires CAP_SYS_RAWIO so this does not
2745 * represent an unprivileged user DOS attack --- we'd be
2746 * in trouble if mortal users could trigger this path at
2747 * will.)
2748 *
2749 * NB. EXT4_STATE_JDATA is not set on files other than
2750 * regular files. If somebody wants to bmap a directory
2751 * or symlink and gets confused because the buffer
2752 * hasn't yet been flushed to disk, they deserve
2753 * everything they get.
2754 */
2755
2756 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2757 journal = EXT4_JOURNAL(inode);
2758 jbd2_journal_lock_updates(journal);
2759 err = jbd2_journal_flush(journal);
2760 jbd2_journal_unlock_updates(journal);
2761
2762 if (err)
2763 return 0;
2764 }
2765
2766 return generic_block_bmap(mapping, block, ext4_get_block);
2767 }
2768
2769 static int ext4_readpage(struct file *file, struct page *page)
2770 {
2771 trace_ext4_readpage(page);
2772 return mpage_readpage(page, ext4_get_block);
2773 }
2774
2775 static int
2776 ext4_readpages(struct file *file, struct address_space *mapping,
2777 struct list_head *pages, unsigned nr_pages)
2778 {
2779 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2780 }
2781
2782 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2783 {
2784 struct buffer_head *head, *bh;
2785 unsigned int curr_off = 0;
2786
2787 if (!page_has_buffers(page))
2788 return;
2789 head = bh = page_buffers(page);
2790 do {
2791 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2792 && bh->b_private) {
2793 ext4_free_io_end(bh->b_private);
2794 bh->b_private = NULL;
2795 bh->b_end_io = NULL;
2796 }
2797 curr_off = curr_off + bh->b_size;
2798 bh = bh->b_this_page;
2799 } while (bh != head);
2800 }
2801
2802 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2803 {
2804 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2805
2806 trace_ext4_invalidatepage(page, offset);
2807
2808 /*
2809 * free any io_end structure allocated for buffers to be discarded
2810 */
2811 if (ext4_should_dioread_nolock(page->mapping->host))
2812 ext4_invalidatepage_free_endio(page, offset);
2813 /*
2814 * If it's a full truncate we just forget about the pending dirtying
2815 */
2816 if (offset == 0)
2817 ClearPageChecked(page);
2818
2819 if (journal)
2820 jbd2_journal_invalidatepage(journal, page, offset);
2821 else
2822 block_invalidatepage(page, offset);
2823 }
2824
2825 static int ext4_releasepage(struct page *page, gfp_t wait)
2826 {
2827 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2828
2829 trace_ext4_releasepage(page);
2830
2831 WARN_ON(PageChecked(page));
2832 if (!page_has_buffers(page))
2833 return 0;
2834 if (journal)
2835 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2836 else
2837 return try_to_free_buffers(page);
2838 }
2839
2840 /*
2841 * ext4_get_block used when preparing for a DIO write or buffer write.
2842 * We allocate an uinitialized extent if blocks haven't been allocated.
2843 * The extent will be converted to initialized after the IO is complete.
2844 */
2845 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2846 struct buffer_head *bh_result, int create)
2847 {
2848 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2849 inode->i_ino, create);
2850 return _ext4_get_block(inode, iblock, bh_result,
2851 EXT4_GET_BLOCKS_IO_CREATE_EXT);
2852 }
2853
2854 static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
2855 struct buffer_head *bh_result, int flags)
2856 {
2857 handle_t *handle = ext4_journal_current_handle();
2858 struct ext4_map_blocks map;
2859 int ret = 0;
2860
2861 ext4_debug("ext4_get_block_write_nolock: inode %lu, flag %d\n",
2862 inode->i_ino, flags);
2863
2864 flags = EXT4_GET_BLOCKS_NO_LOCK;
2865
2866 map.m_lblk = iblock;
2867 map.m_len = bh_result->b_size >> inode->i_blkbits;
2868
2869 ret = ext4_map_blocks(handle, inode, &map, flags);
2870 if (ret > 0) {
2871 map_bh(bh_result, inode->i_sb, map.m_pblk);
2872 bh_result->b_state = (bh_result->b_state & ~EXT4_MAP_FLAGS) |
2873 map.m_flags;
2874 bh_result->b_size = inode->i_sb->s_blocksize * map.m_len;
2875 ret = 0;
2876 }
2877 return ret;
2878 }
2879
2880 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2881 ssize_t size, void *private, int ret,
2882 bool is_async)
2883 {
2884 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2885 ext4_io_end_t *io_end = iocb->private;
2886
2887 /* if not async direct IO or dio with 0 bytes write, just return */
2888 if (!io_end || !size)
2889 goto out;
2890
2891 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2892 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2893 iocb->private, io_end->inode->i_ino, iocb, offset,
2894 size);
2895
2896 iocb->private = NULL;
2897
2898 /* if not aio dio with unwritten extents, just free io and return */
2899 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2900 ext4_free_io_end(io_end);
2901 out:
2902 if (is_async)
2903 aio_complete(iocb, ret, 0);
2904 inode_dio_done(inode);
2905 return;
2906 }
2907
2908 io_end->offset = offset;
2909 io_end->size = size;
2910 if (is_async) {
2911 io_end->iocb = iocb;
2912 io_end->result = ret;
2913 }
2914
2915 ext4_add_complete_io(io_end);
2916 }
2917
2918 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2919 {
2920 ext4_io_end_t *io_end = bh->b_private;
2921 struct inode *inode;
2922
2923 if (!test_clear_buffer_uninit(bh) || !io_end)
2924 goto out;
2925
2926 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2927 ext4_msg(io_end->inode->i_sb, KERN_INFO,
2928 "sb umounted, discard end_io request for inode %lu",
2929 io_end->inode->i_ino);
2930 ext4_free_io_end(io_end);
2931 goto out;
2932 }
2933
2934 /*
2935 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2936 * but being more careful is always safe for the future change.
2937 */
2938 inode = io_end->inode;
2939 ext4_set_io_unwritten_flag(inode, io_end);
2940 ext4_add_complete_io(io_end);
2941 out:
2942 bh->b_private = NULL;
2943 bh->b_end_io = NULL;
2944 clear_buffer_uninit(bh);
2945 end_buffer_async_write(bh, uptodate);
2946 }
2947
2948 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2949 {
2950 ext4_io_end_t *io_end;
2951 struct page *page = bh->b_page;
2952 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2953 size_t size = bh->b_size;
2954
2955 retry:
2956 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2957 if (!io_end) {
2958 pr_warn_ratelimited("%s: allocation fail\n", __func__);
2959 schedule();
2960 goto retry;
2961 }
2962 io_end->offset = offset;
2963 io_end->size = size;
2964 /*
2965 * We need to hold a reference to the page to make sure it
2966 * doesn't get evicted before ext4_end_io_work() has a chance
2967 * to convert the extent from written to unwritten.
2968 */
2969 io_end->page = page;
2970 get_page(io_end->page);
2971
2972 bh->b_private = io_end;
2973 bh->b_end_io = ext4_end_io_buffer_write;
2974 return 0;
2975 }
2976
2977 /*
2978 * For ext4 extent files, ext4 will do direct-io write to holes,
2979 * preallocated extents, and those write extend the file, no need to
2980 * fall back to buffered IO.
2981 *
2982 * For holes, we fallocate those blocks, mark them as uninitialized
2983 * If those blocks were preallocated, we mark sure they are splited, but
2984 * still keep the range to write as uninitialized.
2985 *
2986 * The unwrritten extents will be converted to written when DIO is completed.
2987 * For async direct IO, since the IO may still pending when return, we
2988 * set up an end_io call back function, which will do the conversion
2989 * when async direct IO completed.
2990 *
2991 * If the O_DIRECT write will extend the file then add this inode to the
2992 * orphan list. So recovery will truncate it back to the original size
2993 * if the machine crashes during the write.
2994 *
2995 */
2996 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2997 const struct iovec *iov, loff_t offset,
2998 unsigned long nr_segs)
2999 {
3000 struct file *file = iocb->ki_filp;
3001 struct inode *inode = file->f_mapping->host;
3002 ssize_t ret;
3003 size_t count = iov_length(iov, nr_segs);
3004
3005 loff_t final_size = offset + count;
3006 if (rw == WRITE && final_size <= inode->i_size) {
3007 int overwrite = 0;
3008
3009 BUG_ON(iocb->private == NULL);
3010
3011 /* If we do a overwrite dio, i_mutex locking can be released */
3012 overwrite = *((int *)iocb->private);
3013
3014 if (overwrite) {
3015 atomic_inc(&inode->i_dio_count);
3016 down_read(&EXT4_I(inode)->i_data_sem);
3017 mutex_unlock(&inode->i_mutex);
3018 }
3019
3020 /*
3021 * We could direct write to holes and fallocate.
3022 *
3023 * Allocated blocks to fill the hole are marked as uninitialized
3024 * to prevent parallel buffered read to expose the stale data
3025 * before DIO complete the data IO.
3026 *
3027 * As to previously fallocated extents, ext4 get_block
3028 * will just simply mark the buffer mapped but still
3029 * keep the extents uninitialized.
3030 *
3031 * for non AIO case, we will convert those unwritten extents
3032 * to written after return back from blockdev_direct_IO.
3033 *
3034 * for async DIO, the conversion needs to be defered when
3035 * the IO is completed. The ext4 end_io callback function
3036 * will be called to take care of the conversion work.
3037 * Here for async case, we allocate an io_end structure to
3038 * hook to the iocb.
3039 */
3040 iocb->private = NULL;
3041 ext4_inode_aio_set(inode, NULL);
3042 if (!is_sync_kiocb(iocb)) {
3043 ext4_io_end_t *io_end =
3044 ext4_init_io_end(inode, GFP_NOFS);
3045 if (!io_end) {
3046 ret = -ENOMEM;
3047 goto retake_lock;
3048 }
3049 io_end->flag |= EXT4_IO_END_DIRECT;
3050 iocb->private = io_end;
3051 /*
3052 * we save the io structure for current async
3053 * direct IO, so that later ext4_map_blocks()
3054 * could flag the io structure whether there
3055 * is a unwritten extents needs to be converted
3056 * when IO is completed.
3057 */
3058 ext4_inode_aio_set(inode, io_end);
3059 }
3060
3061 if (overwrite)
3062 ret = __blockdev_direct_IO(rw, iocb, inode,
3063 inode->i_sb->s_bdev, iov,
3064 offset, nr_segs,
3065 ext4_get_block_write_nolock,
3066 ext4_end_io_dio,
3067 NULL,
3068 0);
3069 else
3070 ret = __blockdev_direct_IO(rw, iocb, inode,
3071 inode->i_sb->s_bdev, iov,
3072 offset, nr_segs,
3073 ext4_get_block_write,
3074 ext4_end_io_dio,
3075 NULL,
3076 DIO_LOCKING);
3077 if (iocb->private)
3078 ext4_inode_aio_set(inode, NULL);
3079 /*
3080 * The io_end structure takes a reference to the inode,
3081 * that structure needs to be destroyed and the
3082 * reference to the inode need to be dropped, when IO is
3083 * complete, even with 0 byte write, or failed.
3084 *
3085 * In the successful AIO DIO case, the io_end structure will be
3086 * desctroyed and the reference to the inode will be dropped
3087 * after the end_io call back function is called.
3088 *
3089 * In the case there is 0 byte write, or error case, since
3090 * VFS direct IO won't invoke the end_io call back function,
3091 * we need to free the end_io structure here.
3092 */
3093 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3094 ext4_free_io_end(iocb->private);
3095 iocb->private = NULL;
3096 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3097 EXT4_STATE_DIO_UNWRITTEN)) {
3098 int err;
3099 /*
3100 * for non AIO case, since the IO is already
3101 * completed, we could do the conversion right here
3102 */
3103 err = ext4_convert_unwritten_extents(inode,
3104 offset, ret);
3105 if (err < 0)
3106 ret = err;
3107 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3108 }
3109
3110 retake_lock:
3111 /* take i_mutex locking again if we do a ovewrite dio */
3112 if (overwrite) {
3113 inode_dio_done(inode);
3114 up_read(&EXT4_I(inode)->i_data_sem);
3115 mutex_lock(&inode->i_mutex);
3116 }
3117
3118 return ret;
3119 }
3120
3121 /* for write the the end of file case, we fall back to old way */
3122 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3123 }
3124
3125 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3126 const struct iovec *iov, loff_t offset,
3127 unsigned long nr_segs)
3128 {
3129 struct file *file = iocb->ki_filp;
3130 struct inode *inode = file->f_mapping->host;
3131 ssize_t ret;
3132
3133 /*
3134 * If we are doing data journalling we don't support O_DIRECT
3135 */
3136 if (ext4_should_journal_data(inode))
3137 return 0;
3138
3139 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3140 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3141 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3142 else
3143 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3144 trace_ext4_direct_IO_exit(inode, offset,
3145 iov_length(iov, nr_segs), rw, ret);
3146 return ret;
3147 }
3148
3149 /*
3150 * Pages can be marked dirty completely asynchronously from ext4's journalling
3151 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3152 * much here because ->set_page_dirty is called under VFS locks. The page is
3153 * not necessarily locked.
3154 *
3155 * We cannot just dirty the page and leave attached buffers clean, because the
3156 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3157 * or jbddirty because all the journalling code will explode.
3158 *
3159 * So what we do is to mark the page "pending dirty" and next time writepage
3160 * is called, propagate that into the buffers appropriately.
3161 */
3162 static int ext4_journalled_set_page_dirty(struct page *page)
3163 {
3164 SetPageChecked(page);
3165 return __set_page_dirty_nobuffers(page);
3166 }
3167
3168 static const struct address_space_operations ext4_ordered_aops = {
3169 .readpage = ext4_readpage,
3170 .readpages = ext4_readpages,
3171 .writepage = ext4_writepage,
3172 .write_begin = ext4_write_begin,
3173 .write_end = ext4_ordered_write_end,
3174 .bmap = ext4_bmap,
3175 .invalidatepage = ext4_invalidatepage,
3176 .releasepage = ext4_releasepage,
3177 .direct_IO = ext4_direct_IO,
3178 .migratepage = buffer_migrate_page,
3179 .is_partially_uptodate = block_is_partially_uptodate,
3180 .error_remove_page = generic_error_remove_page,
3181 };
3182
3183 static const struct address_space_operations ext4_writeback_aops = {
3184 .readpage = ext4_readpage,
3185 .readpages = ext4_readpages,
3186 .writepage = ext4_writepage,
3187 .write_begin = ext4_write_begin,
3188 .write_end = ext4_writeback_write_end,
3189 .bmap = ext4_bmap,
3190 .invalidatepage = ext4_invalidatepage,
3191 .releasepage = ext4_releasepage,
3192 .direct_IO = ext4_direct_IO,
3193 .migratepage = buffer_migrate_page,
3194 .is_partially_uptodate = block_is_partially_uptodate,
3195 .error_remove_page = generic_error_remove_page,
3196 };
3197
3198 static const struct address_space_operations ext4_journalled_aops = {
3199 .readpage = ext4_readpage,
3200 .readpages = ext4_readpages,
3201 .writepage = ext4_writepage,
3202 .write_begin = ext4_write_begin,
3203 .write_end = ext4_journalled_write_end,
3204 .set_page_dirty = ext4_journalled_set_page_dirty,
3205 .bmap = ext4_bmap,
3206 .invalidatepage = ext4_invalidatepage,
3207 .releasepage = ext4_releasepage,
3208 .direct_IO = ext4_direct_IO,
3209 .is_partially_uptodate = block_is_partially_uptodate,
3210 .error_remove_page = generic_error_remove_page,
3211 };
3212
3213 static const struct address_space_operations ext4_da_aops = {
3214 .readpage = ext4_readpage,
3215 .readpages = ext4_readpages,
3216 .writepage = ext4_writepage,
3217 .writepages = ext4_da_writepages,
3218 .write_begin = ext4_da_write_begin,
3219 .write_end = ext4_da_write_end,
3220 .bmap = ext4_bmap,
3221 .invalidatepage = ext4_da_invalidatepage,
3222 .releasepage = ext4_releasepage,
3223 .direct_IO = ext4_direct_IO,
3224 .migratepage = buffer_migrate_page,
3225 .is_partially_uptodate = block_is_partially_uptodate,
3226 .error_remove_page = generic_error_remove_page,
3227 };
3228
3229 void ext4_set_aops(struct inode *inode)
3230 {
3231 switch (ext4_inode_journal_mode(inode)) {
3232 case EXT4_INODE_ORDERED_DATA_MODE:
3233 if (test_opt(inode->i_sb, DELALLOC))
3234 inode->i_mapping->a_ops = &ext4_da_aops;
3235 else
3236 inode->i_mapping->a_ops = &ext4_ordered_aops;
3237 break;
3238 case EXT4_INODE_WRITEBACK_DATA_MODE:
3239 if (test_opt(inode->i_sb, DELALLOC))
3240 inode->i_mapping->a_ops = &ext4_da_aops;
3241 else
3242 inode->i_mapping->a_ops = &ext4_writeback_aops;
3243 break;
3244 case EXT4_INODE_JOURNAL_DATA_MODE:
3245 inode->i_mapping->a_ops = &ext4_journalled_aops;
3246 break;
3247 default:
3248 BUG();
3249 }
3250 }
3251
3252
3253 /*
3254 * ext4_discard_partial_page_buffers()
3255 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3256 * This function finds and locks the page containing the offset
3257 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3258 * Calling functions that already have the page locked should call
3259 * ext4_discard_partial_page_buffers_no_lock directly.
3260 */
3261 int ext4_discard_partial_page_buffers(handle_t *handle,
3262 struct address_space *mapping, loff_t from,
3263 loff_t length, int flags)
3264 {
3265 struct inode *inode = mapping->host;
3266 struct page *page;
3267 int err = 0;
3268
3269 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3270 mapping_gfp_mask(mapping) & ~__GFP_FS);
3271 if (!page)
3272 return -ENOMEM;
3273
3274 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3275 from, length, flags);
3276
3277 unlock_page(page);
3278 page_cache_release(page);
3279 return err;
3280 }
3281
3282 /*
3283 * ext4_discard_partial_page_buffers_no_lock()
3284 * Zeros a page range of length 'length' starting from offset 'from'.
3285 * Buffer heads that correspond to the block aligned regions of the
3286 * zeroed range will be unmapped. Unblock aligned regions
3287 * will have the corresponding buffer head mapped if needed so that
3288 * that region of the page can be updated with the partial zero out.
3289 *
3290 * This function assumes that the page has already been locked. The
3291 * The range to be discarded must be contained with in the given page.
3292 * If the specified range exceeds the end of the page it will be shortened
3293 * to the end of the page that corresponds to 'from'. This function is
3294 * appropriate for updating a page and it buffer heads to be unmapped and
3295 * zeroed for blocks that have been either released, or are going to be
3296 * released.
3297 *
3298 * handle: The journal handle
3299 * inode: The files inode
3300 * page: A locked page that contains the offset "from"
3301 * from: The starting byte offset (from the beginning of the file)
3302 * to begin discarding
3303 * len: The length of bytes to discard
3304 * flags: Optional flags that may be used:
3305 *
3306 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3307 * Only zero the regions of the page whose buffer heads
3308 * have already been unmapped. This flag is appropriate
3309 * for updating the contents of a page whose blocks may
3310 * have already been released, and we only want to zero
3311 * out the regions that correspond to those released blocks.
3312 *
3313 * Returns zero on success or negative on failure.
3314 */
3315 static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
3316 struct inode *inode, struct page *page, loff_t from,
3317 loff_t length, int flags)
3318 {
3319 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3320 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3321 unsigned int blocksize, max, pos;
3322 ext4_lblk_t iblock;
3323 struct buffer_head *bh;
3324 int err = 0;
3325
3326 blocksize = inode->i_sb->s_blocksize;
3327 max = PAGE_CACHE_SIZE - offset;
3328
3329 if (index != page->index)
3330 return -EINVAL;
3331
3332 /*
3333 * correct length if it does not fall between
3334 * 'from' and the end of the page
3335 */
3336 if (length > max || length < 0)
3337 length = max;
3338
3339 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3340
3341 if (!page_has_buffers(page))
3342 create_empty_buffers(page, blocksize, 0);
3343
3344 /* Find the buffer that contains "offset" */
3345 bh = page_buffers(page);
3346 pos = blocksize;
3347 while (offset >= pos) {
3348 bh = bh->b_this_page;
3349 iblock++;
3350 pos += blocksize;
3351 }
3352
3353 pos = offset;
3354 while (pos < offset + length) {
3355 unsigned int end_of_block, range_to_discard;
3356
3357 err = 0;
3358
3359 /* The length of space left to zero and unmap */
3360 range_to_discard = offset + length - pos;
3361
3362 /* The length of space until the end of the block */
3363 end_of_block = blocksize - (pos & (blocksize-1));
3364
3365 /*
3366 * Do not unmap or zero past end of block
3367 * for this buffer head
3368 */
3369 if (range_to_discard > end_of_block)
3370 range_to_discard = end_of_block;
3371
3372
3373 /*
3374 * Skip this buffer head if we are only zeroing unampped
3375 * regions of the page
3376 */
3377 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3378 buffer_mapped(bh))
3379 goto next;
3380
3381 /* If the range is block aligned, unmap */
3382 if (range_to_discard == blocksize) {
3383 clear_buffer_dirty(bh);
3384 bh->b_bdev = NULL;
3385 clear_buffer_mapped(bh);
3386 clear_buffer_req(bh);
3387 clear_buffer_new(bh);
3388 clear_buffer_delay(bh);
3389 clear_buffer_unwritten(bh);
3390 clear_buffer_uptodate(bh);
3391 zero_user(page, pos, range_to_discard);
3392 BUFFER_TRACE(bh, "Buffer discarded");
3393 goto next;
3394 }
3395
3396 /*
3397 * If this block is not completely contained in the range
3398 * to be discarded, then it is not going to be released. Because
3399 * we need to keep this block, we need to make sure this part
3400 * of the page is uptodate before we modify it by writeing
3401 * partial zeros on it.
3402 */
3403 if (!buffer_mapped(bh)) {
3404 /*
3405 * Buffer head must be mapped before we can read
3406 * from the block
3407 */
3408 BUFFER_TRACE(bh, "unmapped");
3409 ext4_get_block(inode, iblock, bh, 0);
3410 /* unmapped? It's a hole - nothing to do */
3411 if (!buffer_mapped(bh)) {
3412 BUFFER_TRACE(bh, "still unmapped");
3413 goto next;
3414 }
3415 }
3416
3417 /* Ok, it's mapped. Make sure it's up-to-date */
3418 if (PageUptodate(page))
3419 set_buffer_uptodate(bh);
3420
3421 if (!buffer_uptodate(bh)) {
3422 err = -EIO;
3423 ll_rw_block(READ, 1, &bh);
3424 wait_on_buffer(bh);
3425 /* Uhhuh. Read error. Complain and punt.*/
3426 if (!buffer_uptodate(bh))
3427 goto next;
3428 }
3429
3430 if (ext4_should_journal_data(inode)) {
3431 BUFFER_TRACE(bh, "get write access");
3432 err = ext4_journal_get_write_access(handle, bh);
3433 if (err)
3434 goto next;
3435 }
3436
3437 zero_user(page, pos, range_to_discard);
3438
3439 err = 0;
3440 if (ext4_should_journal_data(inode)) {
3441 err = ext4_handle_dirty_metadata(handle, inode, bh);
3442 } else
3443 mark_buffer_dirty(bh);
3444
3445 BUFFER_TRACE(bh, "Partial buffer zeroed");
3446 next:
3447 bh = bh->b_this_page;
3448 iblock++;
3449 pos += range_to_discard;
3450 }
3451
3452 return err;
3453 }
3454
3455 int ext4_can_truncate(struct inode *inode)
3456 {
3457 if (S_ISREG(inode->i_mode))
3458 return 1;
3459 if (S_ISDIR(inode->i_mode))
3460 return 1;
3461 if (S_ISLNK(inode->i_mode))
3462 return !ext4_inode_is_fast_symlink(inode);
3463 return 0;
3464 }
3465
3466 /*
3467 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3468 * associated with the given offset and length
3469 *
3470 * @inode: File inode
3471 * @offset: The offset where the hole will begin
3472 * @len: The length of the hole
3473 *
3474 * Returns: 0 on success or negative on failure
3475 */
3476
3477 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3478 {
3479 struct inode *inode = file->f_path.dentry->d_inode;
3480 if (!S_ISREG(inode->i_mode))
3481 return -EOPNOTSUPP;
3482
3483 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3484 /* TODO: Add support for non extent hole punching */
3485 return -EOPNOTSUPP;
3486 }
3487
3488 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3489 /* TODO: Add support for bigalloc file systems */
3490 return -EOPNOTSUPP;
3491 }
3492
3493 return ext4_ext_punch_hole(file, offset, length);
3494 }
3495
3496 /*
3497 * ext4_truncate()
3498 *
3499 * We block out ext4_get_block() block instantiations across the entire
3500 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3501 * simultaneously on behalf of the same inode.
3502 *
3503 * As we work through the truncate and commit bits of it to the journal there
3504 * is one core, guiding principle: the file's tree must always be consistent on
3505 * disk. We must be able to restart the truncate after a crash.
3506 *
3507 * The file's tree may be transiently inconsistent in memory (although it
3508 * probably isn't), but whenever we close off and commit a journal transaction,
3509 * the contents of (the filesystem + the journal) must be consistent and
3510 * restartable. It's pretty simple, really: bottom up, right to left (although
3511 * left-to-right works OK too).
3512 *
3513 * Note that at recovery time, journal replay occurs *before* the restart of
3514 * truncate against the orphan inode list.
3515 *
3516 * The committed inode has the new, desired i_size (which is the same as
3517 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3518 * that this inode's truncate did not complete and it will again call
3519 * ext4_truncate() to have another go. So there will be instantiated blocks
3520 * to the right of the truncation point in a crashed ext4 filesystem. But
3521 * that's fine - as long as they are linked from the inode, the post-crash
3522 * ext4_truncate() run will find them and release them.
3523 */
3524 void ext4_truncate(struct inode *inode)
3525 {
3526 trace_ext4_truncate_enter(inode);
3527
3528 if (!ext4_can_truncate(inode))
3529 return;
3530
3531 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3532
3533 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3534 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3535
3536 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3537 ext4_ext_truncate(inode);
3538 else
3539 ext4_ind_truncate(inode);
3540
3541 trace_ext4_truncate_exit(inode);
3542 }
3543
3544 /*
3545 * ext4_get_inode_loc returns with an extra refcount against the inode's
3546 * underlying buffer_head on success. If 'in_mem' is true, we have all
3547 * data in memory that is needed to recreate the on-disk version of this
3548 * inode.
3549 */
3550 static int __ext4_get_inode_loc(struct inode *inode,
3551 struct ext4_iloc *iloc, int in_mem)
3552 {
3553 struct ext4_group_desc *gdp;
3554 struct buffer_head *bh;
3555 struct super_block *sb = inode->i_sb;
3556 ext4_fsblk_t block;
3557 int inodes_per_block, inode_offset;
3558
3559 iloc->bh = NULL;
3560 if (!ext4_valid_inum(sb, inode->i_ino))
3561 return -EIO;
3562
3563 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3564 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3565 if (!gdp)
3566 return -EIO;
3567
3568 /*
3569 * Figure out the offset within the block group inode table
3570 */
3571 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3572 inode_offset = ((inode->i_ino - 1) %
3573 EXT4_INODES_PER_GROUP(sb));
3574 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3575 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3576
3577 bh = sb_getblk(sb, block);
3578 if (!bh) {
3579 EXT4_ERROR_INODE_BLOCK(inode, block,
3580 "unable to read itable block");
3581 return -EIO;
3582 }
3583 if (!buffer_uptodate(bh)) {
3584 lock_buffer(bh);
3585
3586 /*
3587 * If the buffer has the write error flag, we have failed
3588 * to write out another inode in the same block. In this
3589 * case, we don't have to read the block because we may
3590 * read the old inode data successfully.
3591 */
3592 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3593 set_buffer_uptodate(bh);
3594
3595 if (buffer_uptodate(bh)) {
3596 /* someone brought it uptodate while we waited */
3597 unlock_buffer(bh);
3598 goto has_buffer;
3599 }
3600
3601 /*
3602 * If we have all information of the inode in memory and this
3603 * is the only valid inode in the block, we need not read the
3604 * block.
3605 */
3606 if (in_mem) {
3607 struct buffer_head *bitmap_bh;
3608 int i, start;
3609
3610 start = inode_offset & ~(inodes_per_block - 1);
3611
3612 /* Is the inode bitmap in cache? */
3613 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3614 if (!bitmap_bh)
3615 goto make_io;
3616
3617 /*
3618 * If the inode bitmap isn't in cache then the
3619 * optimisation may end up performing two reads instead
3620 * of one, so skip it.
3621 */
3622 if (!buffer_uptodate(bitmap_bh)) {
3623 brelse(bitmap_bh);
3624 goto make_io;
3625 }
3626 for (i = start; i < start + inodes_per_block; i++) {
3627 if (i == inode_offset)
3628 continue;
3629 if (ext4_test_bit(i, bitmap_bh->b_data))
3630 break;
3631 }
3632 brelse(bitmap_bh);
3633 if (i == start + inodes_per_block) {
3634 /* all other inodes are free, so skip I/O */
3635 memset(bh->b_data, 0, bh->b_size);
3636 set_buffer_uptodate(bh);
3637 unlock_buffer(bh);
3638 goto has_buffer;
3639 }
3640 }
3641
3642 make_io:
3643 /*
3644 * If we need to do any I/O, try to pre-readahead extra
3645 * blocks from the inode table.
3646 */
3647 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3648 ext4_fsblk_t b, end, table;
3649 unsigned num;
3650
3651 table = ext4_inode_table(sb, gdp);
3652 /* s_inode_readahead_blks is always a power of 2 */
3653 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3654 if (table > b)
3655 b = table;
3656 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3657 num = EXT4_INODES_PER_GROUP(sb);
3658 if (ext4_has_group_desc_csum(sb))
3659 num -= ext4_itable_unused_count(sb, gdp);
3660 table += num / inodes_per_block;
3661 if (end > table)
3662 end = table;
3663 while (b <= end)
3664 sb_breadahead(sb, b++);
3665 }
3666
3667 /*
3668 * There are other valid inodes in the buffer, this inode
3669 * has in-inode xattrs, or we don't have this inode in memory.
3670 * Read the block from disk.
3671 */
3672 trace_ext4_load_inode(inode);
3673 get_bh(bh);
3674 bh->b_end_io = end_buffer_read_sync;
3675 submit_bh(READ | REQ_META | REQ_PRIO, bh);
3676 wait_on_buffer(bh);
3677 if (!buffer_uptodate(bh)) {
3678 EXT4_ERROR_INODE_BLOCK(inode, block,
3679 "unable to read itable block");
3680 brelse(bh);
3681 return -EIO;
3682 }
3683 }
3684 has_buffer:
3685 iloc->bh = bh;
3686 return 0;
3687 }
3688
3689 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3690 {
3691 /* We have all inode data except xattrs in memory here. */
3692 return __ext4_get_inode_loc(inode, iloc,
3693 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3694 }
3695
3696 void ext4_set_inode_flags(struct inode *inode)
3697 {
3698 unsigned int flags = EXT4_I(inode)->i_flags;
3699
3700 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3701 if (flags & EXT4_SYNC_FL)
3702 inode->i_flags |= S_SYNC;
3703 if (flags & EXT4_APPEND_FL)
3704 inode->i_flags |= S_APPEND;
3705 if (flags & EXT4_IMMUTABLE_FL)
3706 inode->i_flags |= S_IMMUTABLE;
3707 if (flags & EXT4_NOATIME_FL)
3708 inode->i_flags |= S_NOATIME;
3709 if (flags & EXT4_DIRSYNC_FL)
3710 inode->i_flags |= S_DIRSYNC;
3711 }
3712
3713 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3714 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3715 {
3716 unsigned int vfs_fl;
3717 unsigned long old_fl, new_fl;
3718
3719 do {
3720 vfs_fl = ei->vfs_inode.i_flags;
3721 old_fl = ei->i_flags;
3722 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3723 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3724 EXT4_DIRSYNC_FL);
3725 if (vfs_fl & S_SYNC)
3726 new_fl |= EXT4_SYNC_FL;
3727 if (vfs_fl & S_APPEND)
3728 new_fl |= EXT4_APPEND_FL;
3729 if (vfs_fl & S_IMMUTABLE)
3730 new_fl |= EXT4_IMMUTABLE_FL;
3731 if (vfs_fl & S_NOATIME)
3732 new_fl |= EXT4_NOATIME_FL;
3733 if (vfs_fl & S_DIRSYNC)
3734 new_fl |= EXT4_DIRSYNC_FL;
3735 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3736 }
3737
3738 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3739 struct ext4_inode_info *ei)
3740 {
3741 blkcnt_t i_blocks ;
3742 struct inode *inode = &(ei->vfs_inode);
3743 struct super_block *sb = inode->i_sb;
3744
3745 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3746 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3747 /* we are using combined 48 bit field */
3748 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3749 le32_to_cpu(raw_inode->i_blocks_lo);
3750 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3751 /* i_blocks represent file system block size */
3752 return i_blocks << (inode->i_blkbits - 9);
3753 } else {
3754 return i_blocks;
3755 }
3756 } else {
3757 return le32_to_cpu(raw_inode->i_blocks_lo);
3758 }
3759 }
3760
3761 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3762 {
3763 struct ext4_iloc iloc;
3764 struct ext4_inode *raw_inode;
3765 struct ext4_inode_info *ei;
3766 struct inode *inode;
3767 journal_t *journal = EXT4_SB(sb)->s_journal;
3768 long ret;
3769 int block;
3770 uid_t i_uid;
3771 gid_t i_gid;
3772
3773 inode = iget_locked(sb, ino);
3774 if (!inode)
3775 return ERR_PTR(-ENOMEM);
3776 if (!(inode->i_state & I_NEW))
3777 return inode;
3778
3779 ei = EXT4_I(inode);
3780 iloc.bh = NULL;
3781
3782 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3783 if (ret < 0)
3784 goto bad_inode;
3785 raw_inode = ext4_raw_inode(&iloc);
3786
3787 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3788 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3789 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3790 EXT4_INODE_SIZE(inode->i_sb)) {
3791 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3792 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3793 EXT4_INODE_SIZE(inode->i_sb));
3794 ret = -EIO;
3795 goto bad_inode;
3796 }
3797 } else
3798 ei->i_extra_isize = 0;
3799
3800 /* Precompute checksum seed for inode metadata */
3801 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3802 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3803 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3804 __u32 csum;
3805 __le32 inum = cpu_to_le32(inode->i_ino);
3806 __le32 gen = raw_inode->i_generation;
3807 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3808 sizeof(inum));
3809 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3810 sizeof(gen));
3811 }
3812
3813 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3814 EXT4_ERROR_INODE(inode, "checksum invalid");
3815 ret = -EIO;
3816 goto bad_inode;
3817 }
3818
3819 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3820 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3821 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3822 if (!(test_opt(inode->i_sb, NO_UID32))) {
3823 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3824 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3825 }
3826 i_uid_write(inode, i_uid);
3827 i_gid_write(inode, i_gid);
3828 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
3829
3830 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
3831 ei->i_dir_start_lookup = 0;
3832 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3833 /* We now have enough fields to check if the inode was active or not.
3834 * This is needed because nfsd might try to access dead inodes
3835 * the test is that same one that e2fsck uses
3836 * NeilBrown 1999oct15
3837 */
3838 if (inode->i_nlink == 0) {
3839 if (inode->i_mode == 0 ||
3840 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3841 /* this inode is deleted */
3842 ret = -ESTALE;
3843 goto bad_inode;
3844 }
3845 /* The only unlinked inodes we let through here have
3846 * valid i_mode and are being read by the orphan
3847 * recovery code: that's fine, we're about to complete
3848 * the process of deleting those. */
3849 }
3850 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3851 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3852 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3853 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3854 ei->i_file_acl |=
3855 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3856 inode->i_size = ext4_isize(raw_inode);
3857 ei->i_disksize = inode->i_size;
3858 #ifdef CONFIG_QUOTA
3859 ei->i_reserved_quota = 0;
3860 #endif
3861 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3862 ei->i_block_group = iloc.block_group;
3863 ei->i_last_alloc_group = ~0;
3864 /*
3865 * NOTE! The in-memory inode i_data array is in little-endian order
3866 * even on big-endian machines: we do NOT byteswap the block numbers!
3867 */
3868 for (block = 0; block < EXT4_N_BLOCKS; block++)
3869 ei->i_data[block] = raw_inode->i_block[block];
3870 INIT_LIST_HEAD(&ei->i_orphan);
3871
3872 /*
3873 * Set transaction id's of transactions that have to be committed
3874 * to finish f[data]sync. We set them to currently running transaction
3875 * as we cannot be sure that the inode or some of its metadata isn't
3876 * part of the transaction - the inode could have been reclaimed and
3877 * now it is reread from disk.
3878 */
3879 if (journal) {
3880 transaction_t *transaction;
3881 tid_t tid;
3882
3883 read_lock(&journal->j_state_lock);
3884 if (journal->j_running_transaction)
3885 transaction = journal->j_running_transaction;
3886 else
3887 transaction = journal->j_committing_transaction;
3888 if (transaction)
3889 tid = transaction->t_tid;
3890 else
3891 tid = journal->j_commit_sequence;
3892 read_unlock(&journal->j_state_lock);
3893 ei->i_sync_tid = tid;
3894 ei->i_datasync_tid = tid;
3895 }
3896
3897 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3898 if (ei->i_extra_isize == 0) {
3899 /* The extra space is currently unused. Use it. */
3900 ei->i_extra_isize = sizeof(struct ext4_inode) -
3901 EXT4_GOOD_OLD_INODE_SIZE;
3902 } else {
3903 __le32 *magic = (void *)raw_inode +
3904 EXT4_GOOD_OLD_INODE_SIZE +
3905 ei->i_extra_isize;
3906 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3907 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3908 }
3909 }
3910
3911 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3912 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3913 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3914 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3915
3916 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3917 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3918 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3919 inode->i_version |=
3920 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3921 }
3922
3923 ret = 0;
3924 if (ei->i_file_acl &&
3925 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3926 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3927 ei->i_file_acl);
3928 ret = -EIO;
3929 goto bad_inode;
3930 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3931 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3932 (S_ISLNK(inode->i_mode) &&
3933 !ext4_inode_is_fast_symlink(inode)))
3934 /* Validate extent which is part of inode */
3935 ret = ext4_ext_check_inode(inode);
3936 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3937 (S_ISLNK(inode->i_mode) &&
3938 !ext4_inode_is_fast_symlink(inode))) {
3939 /* Validate block references which are part of inode */
3940 ret = ext4_ind_check_inode(inode);
3941 }
3942 if (ret)
3943 goto bad_inode;
3944
3945 if (S_ISREG(inode->i_mode)) {
3946 inode->i_op = &ext4_file_inode_operations;
3947 inode->i_fop = &ext4_file_operations;
3948 ext4_set_aops(inode);
3949 } else if (S_ISDIR(inode->i_mode)) {
3950 inode->i_op = &ext4_dir_inode_operations;
3951 inode->i_fop = &ext4_dir_operations;
3952 } else if (S_ISLNK(inode->i_mode)) {
3953 if (ext4_inode_is_fast_symlink(inode)) {
3954 inode->i_op = &ext4_fast_symlink_inode_operations;
3955 nd_terminate_link(ei->i_data, inode->i_size,
3956 sizeof(ei->i_data) - 1);
3957 } else {
3958 inode->i_op = &ext4_symlink_inode_operations;
3959 ext4_set_aops(inode);
3960 }
3961 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3962 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3963 inode->i_op = &ext4_special_inode_operations;
3964 if (raw_inode->i_block[0])
3965 init_special_inode(inode, inode->i_mode,
3966 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3967 else
3968 init_special_inode(inode, inode->i_mode,
3969 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3970 } else {
3971 ret = -EIO;
3972 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3973 goto bad_inode;
3974 }
3975 brelse(iloc.bh);
3976 ext4_set_inode_flags(inode);
3977 unlock_new_inode(inode);
3978 return inode;
3979
3980 bad_inode:
3981 brelse(iloc.bh);
3982 iget_failed(inode);
3983 return ERR_PTR(ret);
3984 }
3985
3986 static int ext4_inode_blocks_set(handle_t *handle,
3987 struct ext4_inode *raw_inode,
3988 struct ext4_inode_info *ei)
3989 {
3990 struct inode *inode = &(ei->vfs_inode);
3991 u64 i_blocks = inode->i_blocks;
3992 struct super_block *sb = inode->i_sb;
3993
3994 if (i_blocks <= ~0U) {
3995 /*
3996 * i_blocks can be represented in a 32 bit variable
3997 * as multiple of 512 bytes
3998 */
3999 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4000 raw_inode->i_blocks_high = 0;
4001 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4002 return 0;
4003 }
4004 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4005 return -EFBIG;
4006
4007 if (i_blocks <= 0xffffffffffffULL) {
4008 /*
4009 * i_blocks can be represented in a 48 bit variable
4010 * as multiple of 512 bytes
4011 */
4012 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4013 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4014 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4015 } else {
4016 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4017 /* i_block is stored in file system block size */
4018 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4019 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4020 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4021 }
4022 return 0;
4023 }
4024
4025 /*
4026 * Post the struct inode info into an on-disk inode location in the
4027 * buffer-cache. This gobbles the caller's reference to the
4028 * buffer_head in the inode location struct.
4029 *
4030 * The caller must have write access to iloc->bh.
4031 */
4032 static int ext4_do_update_inode(handle_t *handle,
4033 struct inode *inode,
4034 struct ext4_iloc *iloc)
4035 {
4036 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4037 struct ext4_inode_info *ei = EXT4_I(inode);
4038 struct buffer_head *bh = iloc->bh;
4039 int err = 0, rc, block;
4040 int need_datasync = 0;
4041 uid_t i_uid;
4042 gid_t i_gid;
4043
4044 /* For fields not not tracking in the in-memory inode,
4045 * initialise them to zero for new inodes. */
4046 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4047 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4048
4049 ext4_get_inode_flags(ei);
4050 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4051 i_uid = i_uid_read(inode);
4052 i_gid = i_gid_read(inode);
4053 if (!(test_opt(inode->i_sb, NO_UID32))) {
4054 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4055 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4056 /*
4057 * Fix up interoperability with old kernels. Otherwise, old inodes get
4058 * re-used with the upper 16 bits of the uid/gid intact
4059 */
4060 if (!ei->i_dtime) {
4061 raw_inode->i_uid_high =
4062 cpu_to_le16(high_16_bits(i_uid));
4063 raw_inode->i_gid_high =
4064 cpu_to_le16(high_16_bits(i_gid));
4065 } else {
4066 raw_inode->i_uid_high = 0;
4067 raw_inode->i_gid_high = 0;
4068 }
4069 } else {
4070 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4071 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4072 raw_inode->i_uid_high = 0;
4073 raw_inode->i_gid_high = 0;
4074 }
4075 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4076
4077 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4078 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4079 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4080 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4081
4082 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4083 goto out_brelse;
4084 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4085 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4086 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4087 cpu_to_le32(EXT4_OS_HURD))
4088 raw_inode->i_file_acl_high =
4089 cpu_to_le16(ei->i_file_acl >> 32);
4090 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4091 if (ei->i_disksize != ext4_isize(raw_inode)) {
4092 ext4_isize_set(raw_inode, ei->i_disksize);
4093 need_datasync = 1;
4094 }
4095 if (ei->i_disksize > 0x7fffffffULL) {
4096 struct super_block *sb = inode->i_sb;
4097 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4098 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4099 EXT4_SB(sb)->s_es->s_rev_level ==
4100 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4101 /* If this is the first large file
4102 * created, add a flag to the superblock.
4103 */
4104 err = ext4_journal_get_write_access(handle,
4105 EXT4_SB(sb)->s_sbh);
4106 if (err)
4107 goto out_brelse;
4108 ext4_update_dynamic_rev(sb);
4109 EXT4_SET_RO_COMPAT_FEATURE(sb,
4110 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4111 ext4_handle_sync(handle);
4112 err = ext4_handle_dirty_super(handle, sb);
4113 }
4114 }
4115 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4116 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4117 if (old_valid_dev(inode->i_rdev)) {
4118 raw_inode->i_block[0] =
4119 cpu_to_le32(old_encode_dev(inode->i_rdev));
4120 raw_inode->i_block[1] = 0;
4121 } else {
4122 raw_inode->i_block[0] = 0;
4123 raw_inode->i_block[1] =
4124 cpu_to_le32(new_encode_dev(inode->i_rdev));
4125 raw_inode->i_block[2] = 0;
4126 }
4127 } else
4128 for (block = 0; block < EXT4_N_BLOCKS; block++)
4129 raw_inode->i_block[block] = ei->i_data[block];
4130
4131 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4132 if (ei->i_extra_isize) {
4133 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4134 raw_inode->i_version_hi =
4135 cpu_to_le32(inode->i_version >> 32);
4136 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4137 }
4138
4139 ext4_inode_csum_set(inode, raw_inode, ei);
4140
4141 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4142 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
4143 if (!err)
4144 err = rc;
4145 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
4146
4147 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
4148 out_brelse:
4149 brelse(bh);
4150 ext4_std_error(inode->i_sb, err);
4151 return err;
4152 }
4153
4154 /*
4155 * ext4_write_inode()
4156 *
4157 * We are called from a few places:
4158 *
4159 * - Within generic_file_write() for O_SYNC files.
4160 * Here, there will be no transaction running. We wait for any running
4161 * transaction to commit.
4162 *
4163 * - Within sys_sync(), kupdate and such.
4164 * We wait on commit, if tol to.
4165 *
4166 * - Within prune_icache() (PF_MEMALLOC == true)
4167 * Here we simply return. We can't afford to block kswapd on the
4168 * journal commit.
4169 *
4170 * In all cases it is actually safe for us to return without doing anything,
4171 * because the inode has been copied into a raw inode buffer in
4172 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4173 * knfsd.
4174 *
4175 * Note that we are absolutely dependent upon all inode dirtiers doing the
4176 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4177 * which we are interested.
4178 *
4179 * It would be a bug for them to not do this. The code:
4180 *
4181 * mark_inode_dirty(inode)
4182 * stuff();
4183 * inode->i_size = expr;
4184 *
4185 * is in error because a kswapd-driven write_inode() could occur while
4186 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4187 * will no longer be on the superblock's dirty inode list.
4188 */
4189 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
4190 {
4191 int err;
4192
4193 if (current->flags & PF_MEMALLOC)
4194 return 0;
4195
4196 if (EXT4_SB(inode->i_sb)->s_journal) {
4197 if (ext4_journal_current_handle()) {
4198 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4199 dump_stack();
4200 return -EIO;
4201 }
4202
4203 if (wbc->sync_mode != WB_SYNC_ALL)
4204 return 0;
4205
4206 err = ext4_force_commit(inode->i_sb);
4207 } else {
4208 struct ext4_iloc iloc;
4209
4210 err = __ext4_get_inode_loc(inode, &iloc, 0);
4211 if (err)
4212 return err;
4213 if (wbc->sync_mode == WB_SYNC_ALL)
4214 sync_dirty_buffer(iloc.bh);
4215 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
4216 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4217 "IO error syncing inode");
4218 err = -EIO;
4219 }
4220 brelse(iloc.bh);
4221 }
4222 return err;
4223 }
4224
4225 /*
4226 * ext4_setattr()
4227 *
4228 * Called from notify_change.
4229 *
4230 * We want to trap VFS attempts to truncate the file as soon as
4231 * possible. In particular, we want to make sure that when the VFS
4232 * shrinks i_size, we put the inode on the orphan list and modify
4233 * i_disksize immediately, so that during the subsequent flushing of
4234 * dirty pages and freeing of disk blocks, we can guarantee that any
4235 * commit will leave the blocks being flushed in an unused state on
4236 * disk. (On recovery, the inode will get truncated and the blocks will
4237 * be freed, so we have a strong guarantee that no future commit will
4238 * leave these blocks visible to the user.)
4239 *
4240 * Another thing we have to assure is that if we are in ordered mode
4241 * and inode is still attached to the committing transaction, we must
4242 * we start writeout of all the dirty pages which are being truncated.
4243 * This way we are sure that all the data written in the previous
4244 * transaction are already on disk (truncate waits for pages under
4245 * writeback).
4246 *
4247 * Called with inode->i_mutex down.
4248 */
4249 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4250 {
4251 struct inode *inode = dentry->d_inode;
4252 int error, rc = 0;
4253 int orphan = 0;
4254 const unsigned int ia_valid = attr->ia_valid;
4255
4256 error = inode_change_ok(inode, attr);
4257 if (error)
4258 return error;
4259
4260 if (is_quota_modification(inode, attr))
4261 dquot_initialize(inode);
4262 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4263 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
4264 handle_t *handle;
4265
4266 /* (user+group)*(old+new) structure, inode write (sb,
4267 * inode block, ? - but truncate inode update has it) */
4268 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
4269 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
4270 if (IS_ERR(handle)) {
4271 error = PTR_ERR(handle);
4272 goto err_out;
4273 }
4274 error = dquot_transfer(inode, attr);
4275 if (error) {
4276 ext4_journal_stop(handle);
4277 return error;
4278 }
4279 /* Update corresponding info in inode so that everything is in
4280 * one transaction */
4281 if (attr->ia_valid & ATTR_UID)
4282 inode->i_uid = attr->ia_uid;
4283 if (attr->ia_valid & ATTR_GID)
4284 inode->i_gid = attr->ia_gid;
4285 error = ext4_mark_inode_dirty(handle, inode);
4286 ext4_journal_stop(handle);
4287 }
4288
4289 if (attr->ia_valid & ATTR_SIZE) {
4290
4291 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4292 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4293
4294 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4295 return -EFBIG;
4296 }
4297 }
4298
4299 if (S_ISREG(inode->i_mode) &&
4300 attr->ia_valid & ATTR_SIZE &&
4301 (attr->ia_size < inode->i_size)) {
4302 handle_t *handle;
4303
4304 handle = ext4_journal_start(inode, 3);
4305 if (IS_ERR(handle)) {
4306 error = PTR_ERR(handle);
4307 goto err_out;
4308 }
4309 if (ext4_handle_valid(handle)) {
4310 error = ext4_orphan_add(handle, inode);
4311 orphan = 1;
4312 }
4313 EXT4_I(inode)->i_disksize = attr->ia_size;
4314 rc = ext4_mark_inode_dirty(handle, inode);
4315 if (!error)
4316 error = rc;
4317 ext4_journal_stop(handle);
4318
4319 if (ext4_should_order_data(inode)) {
4320 error = ext4_begin_ordered_truncate(inode,
4321 attr->ia_size);
4322 if (error) {
4323 /* Do as much error cleanup as possible */
4324 handle = ext4_journal_start(inode, 3);
4325 if (IS_ERR(handle)) {
4326 ext4_orphan_del(NULL, inode);
4327 goto err_out;
4328 }
4329 ext4_orphan_del(handle, inode);
4330 orphan = 0;
4331 ext4_journal_stop(handle);
4332 goto err_out;
4333 }
4334 }
4335 }
4336
4337 if (attr->ia_valid & ATTR_SIZE) {
4338 if (attr->ia_size != i_size_read(inode)) {
4339 truncate_setsize(inode, attr->ia_size);
4340 /* Inode size will be reduced, wait for dio in flight.
4341 * Temporarily disable dioread_nolock to prevent
4342 * livelock. */
4343 if (orphan) {
4344 ext4_inode_block_unlocked_dio(inode);
4345 inode_dio_wait(inode);
4346 ext4_inode_resume_unlocked_dio(inode);
4347 }
4348 }
4349 ext4_truncate(inode);
4350 }
4351
4352 if (!rc) {
4353 setattr_copy(inode, attr);
4354 mark_inode_dirty(inode);
4355 }
4356
4357 /*
4358 * If the call to ext4_truncate failed to get a transaction handle at
4359 * all, we need to clean up the in-core orphan list manually.
4360 */
4361 if (orphan && inode->i_nlink)
4362 ext4_orphan_del(NULL, inode);
4363
4364 if (!rc && (ia_valid & ATTR_MODE))
4365 rc = ext4_acl_chmod(inode);
4366
4367 err_out:
4368 ext4_std_error(inode->i_sb, error);
4369 if (!error)
4370 error = rc;
4371 return error;
4372 }
4373
4374 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4375 struct kstat *stat)
4376 {
4377 struct inode *inode;
4378 unsigned long delalloc_blocks;
4379
4380 inode = dentry->d_inode;
4381 generic_fillattr(inode, stat);
4382
4383 /*
4384 * We can't update i_blocks if the block allocation is delayed
4385 * otherwise in the case of system crash before the real block
4386 * allocation is done, we will have i_blocks inconsistent with
4387 * on-disk file blocks.
4388 * We always keep i_blocks updated together with real
4389 * allocation. But to not confuse with user, stat
4390 * will return the blocks that include the delayed allocation
4391 * blocks for this file.
4392 */
4393 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4394 EXT4_I(inode)->i_reserved_data_blocks);
4395
4396 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4397 return 0;
4398 }
4399
4400 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4401 {
4402 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4403 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
4404 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4405 }
4406
4407 /*
4408 * Account for index blocks, block groups bitmaps and block group
4409 * descriptor blocks if modify datablocks and index blocks
4410 * worse case, the indexs blocks spread over different block groups
4411 *
4412 * If datablocks are discontiguous, they are possible to spread over
4413 * different block groups too. If they are contiguous, with flexbg,
4414 * they could still across block group boundary.
4415 *
4416 * Also account for superblock, inode, quota and xattr blocks
4417 */
4418 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4419 {
4420 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4421 int gdpblocks;
4422 int idxblocks;
4423 int ret = 0;
4424
4425 /*
4426 * How many index blocks need to touch to modify nrblocks?
4427 * The "Chunk" flag indicating whether the nrblocks is
4428 * physically contiguous on disk
4429 *
4430 * For Direct IO and fallocate, they calls get_block to allocate
4431 * one single extent at a time, so they could set the "Chunk" flag
4432 */
4433 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4434
4435 ret = idxblocks;
4436
4437 /*
4438 * Now let's see how many group bitmaps and group descriptors need
4439 * to account
4440 */
4441 groups = idxblocks;
4442 if (chunk)
4443 groups += 1;
4444 else
4445 groups += nrblocks;
4446
4447 gdpblocks = groups;
4448 if (groups > ngroups)
4449 groups = ngroups;
4450 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4451 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4452
4453 /* bitmaps and block group descriptor blocks */
4454 ret += groups + gdpblocks;
4455
4456 /* Blocks for super block, inode, quota and xattr blocks */
4457 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4458
4459 return ret;
4460 }
4461
4462 /*
4463 * Calculate the total number of credits to reserve to fit
4464 * the modification of a single pages into a single transaction,
4465 * which may include multiple chunks of block allocations.
4466 *
4467 * This could be called via ext4_write_begin()
4468 *
4469 * We need to consider the worse case, when
4470 * one new block per extent.
4471 */
4472 int ext4_writepage_trans_blocks(struct inode *inode)
4473 {
4474 int bpp = ext4_journal_blocks_per_page(inode);
4475 int ret;
4476
4477 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4478
4479 /* Account for data blocks for journalled mode */
4480 if (ext4_should_journal_data(inode))
4481 ret += bpp;
4482 return ret;
4483 }
4484
4485 /*
4486 * Calculate the journal credits for a chunk of data modification.
4487 *
4488 * This is called from DIO, fallocate or whoever calling
4489 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4490 *
4491 * journal buffers for data blocks are not included here, as DIO
4492 * and fallocate do no need to journal data buffers.
4493 */
4494 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4495 {
4496 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4497 }
4498
4499 /*
4500 * The caller must have previously called ext4_reserve_inode_write().
4501 * Give this, we know that the caller already has write access to iloc->bh.
4502 */
4503 int ext4_mark_iloc_dirty(handle_t *handle,
4504 struct inode *inode, struct ext4_iloc *iloc)
4505 {
4506 int err = 0;
4507
4508 if (IS_I_VERSION(inode))
4509 inode_inc_iversion(inode);
4510
4511 /* the do_update_inode consumes one bh->b_count */
4512 get_bh(iloc->bh);
4513
4514 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4515 err = ext4_do_update_inode(handle, inode, iloc);
4516 put_bh(iloc->bh);
4517 return err;
4518 }
4519
4520 /*
4521 * On success, We end up with an outstanding reference count against
4522 * iloc->bh. This _must_ be cleaned up later.
4523 */
4524
4525 int
4526 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4527 struct ext4_iloc *iloc)
4528 {
4529 int err;
4530
4531 err = ext4_get_inode_loc(inode, iloc);
4532 if (!err) {
4533 BUFFER_TRACE(iloc->bh, "get_write_access");
4534 err = ext4_journal_get_write_access(handle, iloc->bh);
4535 if (err) {
4536 brelse(iloc->bh);
4537 iloc->bh = NULL;
4538 }
4539 }
4540 ext4_std_error(inode->i_sb, err);
4541 return err;
4542 }
4543
4544 /*
4545 * Expand an inode by new_extra_isize bytes.
4546 * Returns 0 on success or negative error number on failure.
4547 */
4548 static int ext4_expand_extra_isize(struct inode *inode,
4549 unsigned int new_extra_isize,
4550 struct ext4_iloc iloc,
4551 handle_t *handle)
4552 {
4553 struct ext4_inode *raw_inode;
4554 struct ext4_xattr_ibody_header *header;
4555
4556 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4557 return 0;
4558
4559 raw_inode = ext4_raw_inode(&iloc);
4560
4561 header = IHDR(inode, raw_inode);
4562
4563 /* No extended attributes present */
4564 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4565 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4566 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4567 new_extra_isize);
4568 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4569 return 0;
4570 }
4571
4572 /* try to expand with EAs present */
4573 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4574 raw_inode, handle);
4575 }
4576
4577 /*
4578 * What we do here is to mark the in-core inode as clean with respect to inode
4579 * dirtiness (it may still be data-dirty).
4580 * This means that the in-core inode may be reaped by prune_icache
4581 * without having to perform any I/O. This is a very good thing,
4582 * because *any* task may call prune_icache - even ones which
4583 * have a transaction open against a different journal.
4584 *
4585 * Is this cheating? Not really. Sure, we haven't written the
4586 * inode out, but prune_icache isn't a user-visible syncing function.
4587 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4588 * we start and wait on commits.
4589 */
4590 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4591 {
4592 struct ext4_iloc iloc;
4593 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4594 static unsigned int mnt_count;
4595 int err, ret;
4596
4597 might_sleep();
4598 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4599 err = ext4_reserve_inode_write(handle, inode, &iloc);
4600 if (ext4_handle_valid(handle) &&
4601 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4602 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4603 /*
4604 * We need extra buffer credits since we may write into EA block
4605 * with this same handle. If journal_extend fails, then it will
4606 * only result in a minor loss of functionality for that inode.
4607 * If this is felt to be critical, then e2fsck should be run to
4608 * force a large enough s_min_extra_isize.
4609 */
4610 if ((jbd2_journal_extend(handle,
4611 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4612 ret = ext4_expand_extra_isize(inode,
4613 sbi->s_want_extra_isize,
4614 iloc, handle);
4615 if (ret) {
4616 ext4_set_inode_state(inode,
4617 EXT4_STATE_NO_EXPAND);
4618 if (mnt_count !=
4619 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4620 ext4_warning(inode->i_sb,
4621 "Unable to expand inode %lu. Delete"
4622 " some EAs or run e2fsck.",
4623 inode->i_ino);
4624 mnt_count =
4625 le16_to_cpu(sbi->s_es->s_mnt_count);
4626 }
4627 }
4628 }
4629 }
4630 if (!err)
4631 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4632 return err;
4633 }
4634
4635 /*
4636 * ext4_dirty_inode() is called from __mark_inode_dirty()
4637 *
4638 * We're really interested in the case where a file is being extended.
4639 * i_size has been changed by generic_commit_write() and we thus need
4640 * to include the updated inode in the current transaction.
4641 *
4642 * Also, dquot_alloc_block() will always dirty the inode when blocks
4643 * are allocated to the file.
4644 *
4645 * If the inode is marked synchronous, we don't honour that here - doing
4646 * so would cause a commit on atime updates, which we don't bother doing.
4647 * We handle synchronous inodes at the highest possible level.
4648 */
4649 void ext4_dirty_inode(struct inode *inode, int flags)
4650 {
4651 handle_t *handle;
4652
4653 handle = ext4_journal_start(inode, 2);
4654 if (IS_ERR(handle))
4655 goto out;
4656
4657 ext4_mark_inode_dirty(handle, inode);
4658
4659 ext4_journal_stop(handle);
4660 out:
4661 return;
4662 }
4663
4664 #if 0
4665 /*
4666 * Bind an inode's backing buffer_head into this transaction, to prevent
4667 * it from being flushed to disk early. Unlike
4668 * ext4_reserve_inode_write, this leaves behind no bh reference and
4669 * returns no iloc structure, so the caller needs to repeat the iloc
4670 * lookup to mark the inode dirty later.
4671 */
4672 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4673 {
4674 struct ext4_iloc iloc;
4675
4676 int err = 0;
4677 if (handle) {
4678 err = ext4_get_inode_loc(inode, &iloc);
4679 if (!err) {
4680 BUFFER_TRACE(iloc.bh, "get_write_access");
4681 err = jbd2_journal_get_write_access(handle, iloc.bh);
4682 if (!err)
4683 err = ext4_handle_dirty_metadata(handle,
4684 NULL,
4685 iloc.bh);
4686 brelse(iloc.bh);
4687 }
4688 }
4689 ext4_std_error(inode->i_sb, err);
4690 return err;
4691 }
4692 #endif
4693
4694 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4695 {
4696 journal_t *journal;
4697 handle_t *handle;
4698 int err;
4699
4700 /*
4701 * We have to be very careful here: changing a data block's
4702 * journaling status dynamically is dangerous. If we write a
4703 * data block to the journal, change the status and then delete
4704 * that block, we risk forgetting to revoke the old log record
4705 * from the journal and so a subsequent replay can corrupt data.
4706 * So, first we make sure that the journal is empty and that
4707 * nobody is changing anything.
4708 */
4709
4710 journal = EXT4_JOURNAL(inode);
4711 if (!journal)
4712 return 0;
4713 if (is_journal_aborted(journal))
4714 return -EROFS;
4715 /* We have to allocate physical blocks for delalloc blocks
4716 * before flushing journal. otherwise delalloc blocks can not
4717 * be allocated any more. even more truncate on delalloc blocks
4718 * could trigger BUG by flushing delalloc blocks in journal.
4719 * There is no delalloc block in non-journal data mode.
4720 */
4721 if (val && test_opt(inode->i_sb, DELALLOC)) {
4722 err = ext4_alloc_da_blocks(inode);
4723 if (err < 0)
4724 return err;
4725 }
4726
4727 /* Wait for all existing dio workers */
4728 ext4_inode_block_unlocked_dio(inode);
4729 inode_dio_wait(inode);
4730
4731 jbd2_journal_lock_updates(journal);
4732
4733 /*
4734 * OK, there are no updates running now, and all cached data is
4735 * synced to disk. We are now in a completely consistent state
4736 * which doesn't have anything in the journal, and we know that
4737 * no filesystem updates are running, so it is safe to modify
4738 * the inode's in-core data-journaling state flag now.
4739 */
4740
4741 if (val)
4742 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4743 else {
4744 jbd2_journal_flush(journal);
4745 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4746 }
4747 ext4_set_aops(inode);
4748
4749 jbd2_journal_unlock_updates(journal);
4750 ext4_inode_resume_unlocked_dio(inode);
4751
4752 /* Finally we can mark the inode as dirty. */
4753
4754 handle = ext4_journal_start(inode, 1);
4755 if (IS_ERR(handle))
4756 return PTR_ERR(handle);
4757
4758 err = ext4_mark_inode_dirty(handle, inode);
4759 ext4_handle_sync(handle);
4760 ext4_journal_stop(handle);
4761 ext4_std_error(inode->i_sb, err);
4762
4763 return err;
4764 }
4765
4766 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4767 {
4768 return !buffer_mapped(bh);
4769 }
4770
4771 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4772 {
4773 struct page *page = vmf->page;
4774 loff_t size;
4775 unsigned long len;
4776 int ret;
4777 struct file *file = vma->vm_file;
4778 struct inode *inode = file->f_path.dentry->d_inode;
4779 struct address_space *mapping = inode->i_mapping;
4780 handle_t *handle;
4781 get_block_t *get_block;
4782 int retries = 0;
4783
4784 sb_start_pagefault(inode->i_sb);
4785 file_update_time(vma->vm_file);
4786 /* Delalloc case is easy... */
4787 if (test_opt(inode->i_sb, DELALLOC) &&
4788 !ext4_should_journal_data(inode) &&
4789 !ext4_nonda_switch(inode->i_sb)) {
4790 do {
4791 ret = __block_page_mkwrite(vma, vmf,
4792 ext4_da_get_block_prep);
4793 } while (ret == -ENOSPC &&
4794 ext4_should_retry_alloc(inode->i_sb, &retries));
4795 goto out_ret;
4796 }
4797
4798 lock_page(page);
4799 size = i_size_read(inode);
4800 /* Page got truncated from under us? */
4801 if (page->mapping != mapping || page_offset(page) > size) {
4802 unlock_page(page);
4803 ret = VM_FAULT_NOPAGE;
4804 goto out;
4805 }
4806
4807 if (page->index == size >> PAGE_CACHE_SHIFT)
4808 len = size & ~PAGE_CACHE_MASK;
4809 else
4810 len = PAGE_CACHE_SIZE;
4811 /*
4812 * Return if we have all the buffers mapped. This avoids the need to do
4813 * journal_start/journal_stop which can block and take a long time
4814 */
4815 if (page_has_buffers(page)) {
4816 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4817 ext4_bh_unmapped)) {
4818 /* Wait so that we don't change page under IO */
4819 wait_on_page_writeback(page);
4820 ret = VM_FAULT_LOCKED;
4821 goto out;
4822 }
4823 }
4824 unlock_page(page);
4825 /* OK, we need to fill the hole... */
4826 if (ext4_should_dioread_nolock(inode))
4827 get_block = ext4_get_block_write;
4828 else
4829 get_block = ext4_get_block;
4830 retry_alloc:
4831 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4832 if (IS_ERR(handle)) {
4833 ret = VM_FAULT_SIGBUS;
4834 goto out;
4835 }
4836 ret = __block_page_mkwrite(vma, vmf, get_block);
4837 if (!ret && ext4_should_journal_data(inode)) {
4838 if (walk_page_buffers(handle, page_buffers(page), 0,
4839 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4840 unlock_page(page);
4841 ret = VM_FAULT_SIGBUS;
4842 ext4_journal_stop(handle);
4843 goto out;
4844 }
4845 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4846 }
4847 ext4_journal_stop(handle);
4848 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4849 goto retry_alloc;
4850 out_ret:
4851 ret = block_page_mkwrite_return(ret);
4852 out:
4853 sb_end_pagefault(inode->i_sb);
4854 return ret;
4855 }
4856
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.