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 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
21 *
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 */
24
25 #include <linux/module.h>
26 #include <linux/fs.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include <linux/workqueue.h>
41
42 #include "ext4_jbd2.h"
43 #include "xattr.h"
44 #include "acl.h"
45 #include "ext4_extents.h"
46
47 #include <trace/events/ext4.h>
48
49 #define MPAGE_DA_EXTENT_TAIL 0x01
50
51 static inline int ext4_begin_ordered_truncate(struct inode *inode,
52 loff_t new_size)
53 {
54 return jbd2_journal_begin_ordered_truncate(
55 EXT4_SB(inode->i_sb)->s_journal,
56 &EXT4_I(inode)->jinode,
57 new_size);
58 }
59
60 static void ext4_invalidatepage(struct page *page, unsigned long offset);
61
62 /*
63 * Test whether an inode is a fast symlink.
64 */
65 static int ext4_inode_is_fast_symlink(struct inode *inode)
66 {
67 int ea_blocks = EXT4_I(inode)->i_file_acl ?
68 (inode->i_sb->s_blocksize >> 9) : 0;
69
70 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
71 }
72
73 /*
74 * The ext4 forget function must perform a revoke if we are freeing data
75 * which has been journaled. Metadata (eg. indirect blocks) must be
76 * revoked in all cases.
77 *
78 * "bh" may be NULL: a metadata block may have been freed from memory
79 * but there may still be a record of it in the journal, and that record
80 * still needs to be revoked.
81 *
82 * If the handle isn't valid we're not journaling, but we still need to
83 * call into ext4_journal_revoke() to put the buffer head.
84 */
85 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
86 struct buffer_head *bh, ext4_fsblk_t blocknr)
87 {
88 int err;
89
90 might_sleep();
91
92 BUFFER_TRACE(bh, "enter");
93
94 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
95 "data mode %x\n",
96 bh, is_metadata, inode->i_mode,
97 test_opt(inode->i_sb, DATA_FLAGS));
98
99 /* Never use the revoke function if we are doing full data
100 * journaling: there is no need to, and a V1 superblock won't
101 * support it. Otherwise, only skip the revoke on un-journaled
102 * data blocks. */
103
104 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
105 (!is_metadata && !ext4_should_journal_data(inode))) {
106 if (bh) {
107 BUFFER_TRACE(bh, "call jbd2_journal_forget");
108 return ext4_journal_forget(handle, bh);
109 }
110 return 0;
111 }
112
113 /*
114 * data!=journal && (is_metadata || should_journal_data(inode))
115 */
116 BUFFER_TRACE(bh, "call ext4_journal_revoke");
117 err = ext4_journal_revoke(handle, blocknr, bh);
118 if (err)
119 ext4_abort(inode->i_sb, __func__,
120 "error %d when attempting revoke", err);
121 BUFFER_TRACE(bh, "exit");
122 return err;
123 }
124
125 /*
126 * Work out how many blocks we need to proceed with the next chunk of a
127 * truncate transaction.
128 */
129 static unsigned long blocks_for_truncate(struct inode *inode)
130 {
131 ext4_lblk_t needed;
132
133 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
134
135 /* Give ourselves just enough room to cope with inodes in which
136 * i_blocks is corrupt: we've seen disk corruptions in the past
137 * which resulted in random data in an inode which looked enough
138 * like a regular file for ext4 to try to delete it. Things
139 * will go a bit crazy if that happens, but at least we should
140 * try not to panic the whole kernel. */
141 if (needed < 2)
142 needed = 2;
143
144 /* But we need to bound the transaction so we don't overflow the
145 * journal. */
146 if (needed > EXT4_MAX_TRANS_DATA)
147 needed = EXT4_MAX_TRANS_DATA;
148
149 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
150 }
151
152 /*
153 * Truncate transactions can be complex and absolutely huge. So we need to
154 * be able to restart the transaction at a conventient checkpoint to make
155 * sure we don't overflow the journal.
156 *
157 * start_transaction gets us a new handle for a truncate transaction,
158 * and extend_transaction tries to extend the existing one a bit. If
159 * extend fails, we need to propagate the failure up and restart the
160 * transaction in the top-level truncate loop. --sct
161 */
162 static handle_t *start_transaction(struct inode *inode)
163 {
164 handle_t *result;
165
166 result = ext4_journal_start(inode, blocks_for_truncate(inode));
167 if (!IS_ERR(result))
168 return result;
169
170 ext4_std_error(inode->i_sb, PTR_ERR(result));
171 return result;
172 }
173
174 /*
175 * Try to extend this transaction for the purposes of truncation.
176 *
177 * Returns 0 if we managed to create more room. If we can't create more
178 * room, and the transaction must be restarted we return 1.
179 */
180 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
181 {
182 if (!ext4_handle_valid(handle))
183 return 0;
184 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
185 return 0;
186 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
187 return 0;
188 return 1;
189 }
190
191 /*
192 * Restart the transaction associated with *handle. This does a commit,
193 * so before we call here everything must be consistently dirtied against
194 * this transaction.
195 */
196 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
197 int nblocks)
198 {
199 int ret;
200
201 /*
202 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
203 * moment, get_block can be called only for blocks inside i_size since
204 * page cache has been already dropped and writes are blocked by
205 * i_mutex. So we can safely drop the i_data_sem here.
206 */
207 BUG_ON(EXT4_JOURNAL(inode) == NULL);
208 jbd_debug(2, "restarting handle %p\n", handle);
209 up_write(&EXT4_I(inode)->i_data_sem);
210 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
211 down_write(&EXT4_I(inode)->i_data_sem);
212 ext4_discard_preallocations(inode);
213
214 return ret;
215 }
216
217 /*
218 * Called at the last iput() if i_nlink is zero.
219 */
220 void ext4_delete_inode(struct inode *inode)
221 {
222 handle_t *handle;
223 int err;
224
225 if (ext4_should_order_data(inode))
226 ext4_begin_ordered_truncate(inode, 0);
227 truncate_inode_pages(&inode->i_data, 0);
228
229 if (is_bad_inode(inode))
230 goto no_delete;
231
232 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
233 if (IS_ERR(handle)) {
234 ext4_std_error(inode->i_sb, PTR_ERR(handle));
235 /*
236 * If we're going to skip the normal cleanup, we still need to
237 * make sure that the in-core orphan linked list is properly
238 * cleaned up.
239 */
240 ext4_orphan_del(NULL, inode);
241 goto no_delete;
242 }
243
244 if (IS_SYNC(inode))
245 ext4_handle_sync(handle);
246 inode->i_size = 0;
247 err = ext4_mark_inode_dirty(handle, inode);
248 if (err) {
249 ext4_warning(inode->i_sb, __func__,
250 "couldn't mark inode dirty (err %d)", err);
251 goto stop_handle;
252 }
253 if (inode->i_blocks)
254 ext4_truncate(inode);
255
256 /*
257 * ext4_ext_truncate() doesn't reserve any slop when it
258 * restarts journal transactions; therefore there may not be
259 * enough credits left in the handle to remove the inode from
260 * the orphan list and set the dtime field.
261 */
262 if (!ext4_handle_has_enough_credits(handle, 3)) {
263 err = ext4_journal_extend(handle, 3);
264 if (err > 0)
265 err = ext4_journal_restart(handle, 3);
266 if (err != 0) {
267 ext4_warning(inode->i_sb, __func__,
268 "couldn't extend journal (err %d)", err);
269 stop_handle:
270 ext4_journal_stop(handle);
271 goto no_delete;
272 }
273 }
274
275 /*
276 * Kill off the orphan record which ext4_truncate created.
277 * AKPM: I think this can be inside the above `if'.
278 * Note that ext4_orphan_del() has to be able to cope with the
279 * deletion of a non-existent orphan - this is because we don't
280 * know if ext4_truncate() actually created an orphan record.
281 * (Well, we could do this if we need to, but heck - it works)
282 */
283 ext4_orphan_del(handle, inode);
284 EXT4_I(inode)->i_dtime = get_seconds();
285
286 /*
287 * One subtle ordering requirement: if anything has gone wrong
288 * (transaction abort, IO errors, whatever), then we can still
289 * do these next steps (the fs will already have been marked as
290 * having errors), but we can't free the inode if the mark_dirty
291 * fails.
292 */
293 if (ext4_mark_inode_dirty(handle, inode))
294 /* If that failed, just do the required in-core inode clear. */
295 clear_inode(inode);
296 else
297 ext4_free_inode(handle, inode);
298 ext4_journal_stop(handle);
299 return;
300 no_delete:
301 clear_inode(inode); /* We must guarantee clearing of inode... */
302 }
303
304 typedef struct {
305 __le32 *p;
306 __le32 key;
307 struct buffer_head *bh;
308 } Indirect;
309
310 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
311 {
312 p->key = *(p->p = v);
313 p->bh = bh;
314 }
315
316 /**
317 * ext4_block_to_path - parse the block number into array of offsets
318 * @inode: inode in question (we are only interested in its superblock)
319 * @i_block: block number to be parsed
320 * @offsets: array to store the offsets in
321 * @boundary: set this non-zero if the referred-to block is likely to be
322 * followed (on disk) by an indirect block.
323 *
324 * To store the locations of file's data ext4 uses a data structure common
325 * for UNIX filesystems - tree of pointers anchored in the inode, with
326 * data blocks at leaves and indirect blocks in intermediate nodes.
327 * This function translates the block number into path in that tree -
328 * return value is the path length and @offsets[n] is the offset of
329 * pointer to (n+1)th node in the nth one. If @block is out of range
330 * (negative or too large) warning is printed and zero returned.
331 *
332 * Note: function doesn't find node addresses, so no IO is needed. All
333 * we need to know is the capacity of indirect blocks (taken from the
334 * inode->i_sb).
335 */
336
337 /*
338 * Portability note: the last comparison (check that we fit into triple
339 * indirect block) is spelled differently, because otherwise on an
340 * architecture with 32-bit longs and 8Kb pages we might get into trouble
341 * if our filesystem had 8Kb blocks. We might use long long, but that would
342 * kill us on x86. Oh, well, at least the sign propagation does not matter -
343 * i_block would have to be negative in the very beginning, so we would not
344 * get there at all.
345 */
346
347 static int ext4_block_to_path(struct inode *inode,
348 ext4_lblk_t i_block,
349 ext4_lblk_t offsets[4], int *boundary)
350 {
351 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
352 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
353 const long direct_blocks = EXT4_NDIR_BLOCKS,
354 indirect_blocks = ptrs,
355 double_blocks = (1 << (ptrs_bits * 2));
356 int n = 0;
357 int final = 0;
358
359 if (i_block < 0) {
360 ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
361 } else if (i_block < direct_blocks) {
362 offsets[n++] = i_block;
363 final = direct_blocks;
364 } else if ((i_block -= direct_blocks) < indirect_blocks) {
365 offsets[n++] = EXT4_IND_BLOCK;
366 offsets[n++] = i_block;
367 final = ptrs;
368 } else if ((i_block -= indirect_blocks) < double_blocks) {
369 offsets[n++] = EXT4_DIND_BLOCK;
370 offsets[n++] = i_block >> ptrs_bits;
371 offsets[n++] = i_block & (ptrs - 1);
372 final = ptrs;
373 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
374 offsets[n++] = EXT4_TIND_BLOCK;
375 offsets[n++] = i_block >> (ptrs_bits * 2);
376 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
377 offsets[n++] = i_block & (ptrs - 1);
378 final = ptrs;
379 } else {
380 ext4_warning(inode->i_sb, "ext4_block_to_path",
381 "block %lu > max in inode %lu",
382 i_block + direct_blocks +
383 indirect_blocks + double_blocks, inode->i_ino);
384 }
385 if (boundary)
386 *boundary = final - 1 - (i_block & (ptrs - 1));
387 return n;
388 }
389
390 static int __ext4_check_blockref(const char *function, struct inode *inode,
391 __le32 *p, unsigned int max)
392 {
393 __le32 *bref = p;
394 unsigned int blk;
395
396 while (bref < p+max) {
397 blk = le32_to_cpu(*bref++);
398 if (blk &&
399 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
400 blk, 1))) {
401 ext4_error(inode->i_sb, function,
402 "invalid block reference %u "
403 "in inode #%lu", blk, inode->i_ino);
404 return -EIO;
405 }
406 }
407 return 0;
408 }
409
410
411 #define ext4_check_indirect_blockref(inode, bh) \
412 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
413 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
414
415 #define ext4_check_inode_blockref(inode) \
416 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
417 EXT4_NDIR_BLOCKS)
418
419 /**
420 * ext4_get_branch - read the chain of indirect blocks leading to data
421 * @inode: inode in question
422 * @depth: depth of the chain (1 - direct pointer, etc.)
423 * @offsets: offsets of pointers in inode/indirect blocks
424 * @chain: place to store the result
425 * @err: here we store the error value
426 *
427 * Function fills the array of triples <key, p, bh> and returns %NULL
428 * if everything went OK or the pointer to the last filled triple
429 * (incomplete one) otherwise. Upon the return chain[i].key contains
430 * the number of (i+1)-th block in the chain (as it is stored in memory,
431 * i.e. little-endian 32-bit), chain[i].p contains the address of that
432 * number (it points into struct inode for i==0 and into the bh->b_data
433 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
434 * block for i>0 and NULL for i==0. In other words, it holds the block
435 * numbers of the chain, addresses they were taken from (and where we can
436 * verify that chain did not change) and buffer_heads hosting these
437 * numbers.
438 *
439 * Function stops when it stumbles upon zero pointer (absent block)
440 * (pointer to last triple returned, *@err == 0)
441 * or when it gets an IO error reading an indirect block
442 * (ditto, *@err == -EIO)
443 * or when it reads all @depth-1 indirect blocks successfully and finds
444 * the whole chain, all way to the data (returns %NULL, *err == 0).
445 *
446 * Need to be called with
447 * down_read(&EXT4_I(inode)->i_data_sem)
448 */
449 static Indirect *ext4_get_branch(struct inode *inode, int depth,
450 ext4_lblk_t *offsets,
451 Indirect chain[4], int *err)
452 {
453 struct super_block *sb = inode->i_sb;
454 Indirect *p = chain;
455 struct buffer_head *bh;
456
457 *err = 0;
458 /* i_data is not going away, no lock needed */
459 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
460 if (!p->key)
461 goto no_block;
462 while (--depth) {
463 bh = sb_getblk(sb, le32_to_cpu(p->key));
464 if (unlikely(!bh))
465 goto failure;
466
467 if (!bh_uptodate_or_lock(bh)) {
468 if (bh_submit_read(bh) < 0) {
469 put_bh(bh);
470 goto failure;
471 }
472 /* validate block references */
473 if (ext4_check_indirect_blockref(inode, bh)) {
474 put_bh(bh);
475 goto failure;
476 }
477 }
478
479 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
480 /* Reader: end */
481 if (!p->key)
482 goto no_block;
483 }
484 return NULL;
485
486 failure:
487 *err = -EIO;
488 no_block:
489 return p;
490 }
491
492 /**
493 * ext4_find_near - find a place for allocation with sufficient locality
494 * @inode: owner
495 * @ind: descriptor of indirect block.
496 *
497 * This function returns the preferred place for block allocation.
498 * It is used when heuristic for sequential allocation fails.
499 * Rules are:
500 * + if there is a block to the left of our position - allocate near it.
501 * + if pointer will live in indirect block - allocate near that block.
502 * + if pointer will live in inode - allocate in the same
503 * cylinder group.
504 *
505 * In the latter case we colour the starting block by the callers PID to
506 * prevent it from clashing with concurrent allocations for a different inode
507 * in the same block group. The PID is used here so that functionally related
508 * files will be close-by on-disk.
509 *
510 * Caller must make sure that @ind is valid and will stay that way.
511 */
512 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
513 {
514 struct ext4_inode_info *ei = EXT4_I(inode);
515 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
516 __le32 *p;
517 ext4_fsblk_t bg_start;
518 ext4_fsblk_t last_block;
519 ext4_grpblk_t colour;
520 ext4_group_t block_group;
521 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
522
523 /* Try to find previous block */
524 for (p = ind->p - 1; p >= start; p--) {
525 if (*p)
526 return le32_to_cpu(*p);
527 }
528
529 /* No such thing, so let's try location of indirect block */
530 if (ind->bh)
531 return ind->bh->b_blocknr;
532
533 /*
534 * It is going to be referred to from the inode itself? OK, just put it
535 * into the same cylinder group then.
536 */
537 block_group = ei->i_block_group;
538 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
539 block_group &= ~(flex_size-1);
540 if (S_ISREG(inode->i_mode))
541 block_group++;
542 }
543 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
544 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
545
546 /*
547 * If we are doing delayed allocation, we don't need take
548 * colour into account.
549 */
550 if (test_opt(inode->i_sb, DELALLOC))
551 return bg_start;
552
553 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
554 colour = (current->pid % 16) *
555 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
556 else
557 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
558 return bg_start + colour;
559 }
560
561 /**
562 * ext4_find_goal - find a preferred place for allocation.
563 * @inode: owner
564 * @block: block we want
565 * @partial: pointer to the last triple within a chain
566 *
567 * Normally this function find the preferred place for block allocation,
568 * returns it.
569 * Because this is only used for non-extent files, we limit the block nr
570 * to 32 bits.
571 */
572 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
573 Indirect *partial)
574 {
575 ext4_fsblk_t goal;
576
577 /*
578 * XXX need to get goal block from mballoc's data structures
579 */
580
581 goal = ext4_find_near(inode, partial);
582 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
583 return goal;
584 }
585
586 /**
587 * ext4_blks_to_allocate: Look up the block map and count the number
588 * of direct blocks need to be allocated for the given branch.
589 *
590 * @branch: chain of indirect blocks
591 * @k: number of blocks need for indirect blocks
592 * @blks: number of data blocks to be mapped.
593 * @blocks_to_boundary: the offset in the indirect block
594 *
595 * return the total number of blocks to be allocate, including the
596 * direct and indirect blocks.
597 */
598 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
599 int blocks_to_boundary)
600 {
601 unsigned int count = 0;
602
603 /*
604 * Simple case, [t,d]Indirect block(s) has not allocated yet
605 * then it's clear blocks on that path have not allocated
606 */
607 if (k > 0) {
608 /* right now we don't handle cross boundary allocation */
609 if (blks < blocks_to_boundary + 1)
610 count += blks;
611 else
612 count += blocks_to_boundary + 1;
613 return count;
614 }
615
616 count++;
617 while (count < blks && count <= blocks_to_boundary &&
618 le32_to_cpu(*(branch[0].p + count)) == 0) {
619 count++;
620 }
621 return count;
622 }
623
624 /**
625 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
626 * @indirect_blks: the number of blocks need to allocate for indirect
627 * blocks
628 *
629 * @new_blocks: on return it will store the new block numbers for
630 * the indirect blocks(if needed) and the first direct block,
631 * @blks: on return it will store the total number of allocated
632 * direct blocks
633 */
634 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
635 ext4_lblk_t iblock, ext4_fsblk_t goal,
636 int indirect_blks, int blks,
637 ext4_fsblk_t new_blocks[4], int *err)
638 {
639 struct ext4_allocation_request ar;
640 int target, i;
641 unsigned long count = 0, blk_allocated = 0;
642 int index = 0;
643 ext4_fsblk_t current_block = 0;
644 int ret = 0;
645
646 /*
647 * Here we try to allocate the requested multiple blocks at once,
648 * on a best-effort basis.
649 * To build a branch, we should allocate blocks for
650 * the indirect blocks(if not allocated yet), and at least
651 * the first direct block of this branch. That's the
652 * minimum number of blocks need to allocate(required)
653 */
654 /* first we try to allocate the indirect blocks */
655 target = indirect_blks;
656 while (target > 0) {
657 count = target;
658 /* allocating blocks for indirect blocks and direct blocks */
659 current_block = ext4_new_meta_blocks(handle, inode,
660 goal, &count, err);
661 if (*err)
662 goto failed_out;
663
664 BUG_ON(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS);
665
666 target -= count;
667 /* allocate blocks for indirect blocks */
668 while (index < indirect_blks && count) {
669 new_blocks[index++] = current_block++;
670 count--;
671 }
672 if (count > 0) {
673 /*
674 * save the new block number
675 * for the first direct block
676 */
677 new_blocks[index] = current_block;
678 printk(KERN_INFO "%s returned more blocks than "
679 "requested\n", __func__);
680 WARN_ON(1);
681 break;
682 }
683 }
684
685 target = blks - count ;
686 blk_allocated = count;
687 if (!target)
688 goto allocated;
689 /* Now allocate data blocks */
690 memset(&ar, 0, sizeof(ar));
691 ar.inode = inode;
692 ar.goal = goal;
693 ar.len = target;
694 ar.logical = iblock;
695 if (S_ISREG(inode->i_mode))
696 /* enable in-core preallocation only for regular files */
697 ar.flags = EXT4_MB_HINT_DATA;
698
699 current_block = ext4_mb_new_blocks(handle, &ar, err);
700 BUG_ON(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS);
701
702 if (*err && (target == blks)) {
703 /*
704 * if the allocation failed and we didn't allocate
705 * any blocks before
706 */
707 goto failed_out;
708 }
709 if (!*err) {
710 if (target == blks) {
711 /*
712 * save the new block number
713 * for the first direct block
714 */
715 new_blocks[index] = current_block;
716 }
717 blk_allocated += ar.len;
718 }
719 allocated:
720 /* total number of blocks allocated for direct blocks */
721 ret = blk_allocated;
722 *err = 0;
723 return ret;
724 failed_out:
725 for (i = 0; i < index; i++)
726 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
727 return ret;
728 }
729
730 /**
731 * ext4_alloc_branch - allocate and set up a chain of blocks.
732 * @inode: owner
733 * @indirect_blks: number of allocated indirect blocks
734 * @blks: number of allocated direct blocks
735 * @offsets: offsets (in the blocks) to store the pointers to next.
736 * @branch: place to store the chain in.
737 *
738 * This function allocates blocks, zeroes out all but the last one,
739 * links them into chain and (if we are synchronous) writes them to disk.
740 * In other words, it prepares a branch that can be spliced onto the
741 * inode. It stores the information about that chain in the branch[], in
742 * the same format as ext4_get_branch() would do. We are calling it after
743 * we had read the existing part of chain and partial points to the last
744 * triple of that (one with zero ->key). Upon the exit we have the same
745 * picture as after the successful ext4_get_block(), except that in one
746 * place chain is disconnected - *branch->p is still zero (we did not
747 * set the last link), but branch->key contains the number that should
748 * be placed into *branch->p to fill that gap.
749 *
750 * If allocation fails we free all blocks we've allocated (and forget
751 * their buffer_heads) and return the error value the from failed
752 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
753 * as described above and return 0.
754 */
755 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
756 ext4_lblk_t iblock, int indirect_blks,
757 int *blks, ext4_fsblk_t goal,
758 ext4_lblk_t *offsets, Indirect *branch)
759 {
760 int blocksize = inode->i_sb->s_blocksize;
761 int i, n = 0;
762 int err = 0;
763 struct buffer_head *bh;
764 int num;
765 ext4_fsblk_t new_blocks[4];
766 ext4_fsblk_t current_block;
767
768 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
769 *blks, new_blocks, &err);
770 if (err)
771 return err;
772
773 branch[0].key = cpu_to_le32(new_blocks[0]);
774 /*
775 * metadata blocks and data blocks are allocated.
776 */
777 for (n = 1; n <= indirect_blks; n++) {
778 /*
779 * Get buffer_head for parent block, zero it out
780 * and set the pointer to new one, then send
781 * parent to disk.
782 */
783 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
784 branch[n].bh = bh;
785 lock_buffer(bh);
786 BUFFER_TRACE(bh, "call get_create_access");
787 err = ext4_journal_get_create_access(handle, bh);
788 if (err) {
789 unlock_buffer(bh);
790 brelse(bh);
791 goto failed;
792 }
793
794 memset(bh->b_data, 0, blocksize);
795 branch[n].p = (__le32 *) bh->b_data + offsets[n];
796 branch[n].key = cpu_to_le32(new_blocks[n]);
797 *branch[n].p = branch[n].key;
798 if (n == indirect_blks) {
799 current_block = new_blocks[n];
800 /*
801 * End of chain, update the last new metablock of
802 * the chain to point to the new allocated
803 * data blocks numbers
804 */
805 for (i = 1; i < num; i++)
806 *(branch[n].p + i) = cpu_to_le32(++current_block);
807 }
808 BUFFER_TRACE(bh, "marking uptodate");
809 set_buffer_uptodate(bh);
810 unlock_buffer(bh);
811
812 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
813 err = ext4_handle_dirty_metadata(handle, inode, bh);
814 if (err)
815 goto failed;
816 }
817 *blks = num;
818 return err;
819 failed:
820 /* Allocation failed, free what we already allocated */
821 for (i = 1; i <= n ; i++) {
822 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
823 ext4_journal_forget(handle, branch[i].bh);
824 }
825 for (i = 0; i < indirect_blks; i++)
826 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
827
828 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
829
830 return err;
831 }
832
833 /**
834 * ext4_splice_branch - splice the allocated branch onto inode.
835 * @inode: owner
836 * @block: (logical) number of block we are adding
837 * @chain: chain of indirect blocks (with a missing link - see
838 * ext4_alloc_branch)
839 * @where: location of missing link
840 * @num: number of indirect blocks we are adding
841 * @blks: number of direct blocks we are adding
842 *
843 * This function fills the missing link and does all housekeeping needed in
844 * inode (->i_blocks, etc.). In case of success we end up with the full
845 * chain to new block and return 0.
846 */
847 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
848 ext4_lblk_t block, Indirect *where, int num,
849 int blks)
850 {
851 int i;
852 int err = 0;
853 ext4_fsblk_t current_block;
854
855 /*
856 * If we're splicing into a [td]indirect block (as opposed to the
857 * inode) then we need to get write access to the [td]indirect block
858 * before the splice.
859 */
860 if (where->bh) {
861 BUFFER_TRACE(where->bh, "get_write_access");
862 err = ext4_journal_get_write_access(handle, where->bh);
863 if (err)
864 goto err_out;
865 }
866 /* That's it */
867
868 *where->p = where->key;
869
870 /*
871 * Update the host buffer_head or inode to point to more just allocated
872 * direct blocks blocks
873 */
874 if (num == 0 && blks > 1) {
875 current_block = le32_to_cpu(where->key) + 1;
876 for (i = 1; i < blks; i++)
877 *(where->p + i) = cpu_to_le32(current_block++);
878 }
879
880 /* We are done with atomic stuff, now do the rest of housekeeping */
881 /* had we spliced it onto indirect block? */
882 if (where->bh) {
883 /*
884 * If we spliced it onto an indirect block, we haven't
885 * altered the inode. Note however that if it is being spliced
886 * onto an indirect block at the very end of the file (the
887 * file is growing) then we *will* alter the inode to reflect
888 * the new i_size. But that is not done here - it is done in
889 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
890 */
891 jbd_debug(5, "splicing indirect only\n");
892 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
893 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
894 if (err)
895 goto err_out;
896 } else {
897 /*
898 * OK, we spliced it into the inode itself on a direct block.
899 */
900 ext4_mark_inode_dirty(handle, inode);
901 jbd_debug(5, "splicing direct\n");
902 }
903 return err;
904
905 err_out:
906 for (i = 1; i <= num; i++) {
907 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
908 ext4_journal_forget(handle, where[i].bh);
909 ext4_free_blocks(handle, inode,
910 le32_to_cpu(where[i-1].key), 1, 0);
911 }
912 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
913
914 return err;
915 }
916
917 /*
918 * The ext4_ind_get_blocks() function handles non-extents inodes
919 * (i.e., using the traditional indirect/double-indirect i_blocks
920 * scheme) for ext4_get_blocks().
921 *
922 * Allocation strategy is simple: if we have to allocate something, we will
923 * have to go the whole way to leaf. So let's do it before attaching anything
924 * to tree, set linkage between the newborn blocks, write them if sync is
925 * required, recheck the path, free and repeat if check fails, otherwise
926 * set the last missing link (that will protect us from any truncate-generated
927 * removals - all blocks on the path are immune now) and possibly force the
928 * write on the parent block.
929 * That has a nice additional property: no special recovery from the failed
930 * allocations is needed - we simply release blocks and do not touch anything
931 * reachable from inode.
932 *
933 * `handle' can be NULL if create == 0.
934 *
935 * return > 0, # of blocks mapped or allocated.
936 * return = 0, if plain lookup failed.
937 * return < 0, error case.
938 *
939 * The ext4_ind_get_blocks() function should be called with
940 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
941 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
942 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
943 * blocks.
944 */
945 static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
946 ext4_lblk_t iblock, unsigned int maxblocks,
947 struct buffer_head *bh_result,
948 int flags)
949 {
950 int err = -EIO;
951 ext4_lblk_t offsets[4];
952 Indirect chain[4];
953 Indirect *partial;
954 ext4_fsblk_t goal;
955 int indirect_blks;
956 int blocks_to_boundary = 0;
957 int depth;
958 int count = 0;
959 ext4_fsblk_t first_block = 0;
960
961 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
962 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
963 depth = ext4_block_to_path(inode, iblock, offsets,
964 &blocks_to_boundary);
965
966 if (depth == 0)
967 goto out;
968
969 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
970
971 /* Simplest case - block found, no allocation needed */
972 if (!partial) {
973 first_block = le32_to_cpu(chain[depth - 1].key);
974 clear_buffer_new(bh_result);
975 count++;
976 /*map more blocks*/
977 while (count < maxblocks && count <= blocks_to_boundary) {
978 ext4_fsblk_t blk;
979
980 blk = le32_to_cpu(*(chain[depth-1].p + count));
981
982 if (blk == first_block + count)
983 count++;
984 else
985 break;
986 }
987 goto got_it;
988 }
989
990 /* Next simple case - plain lookup or failed read of indirect block */
991 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
992 goto cleanup;
993
994 /*
995 * Okay, we need to do block allocation.
996 */
997 goal = ext4_find_goal(inode, iblock, partial);
998
999 /* the number of blocks need to allocate for [d,t]indirect blocks */
1000 indirect_blks = (chain + depth) - partial - 1;
1001
1002 /*
1003 * Next look up the indirect map to count the totoal number of
1004 * direct blocks to allocate for this branch.
1005 */
1006 count = ext4_blks_to_allocate(partial, indirect_blks,
1007 maxblocks, blocks_to_boundary);
1008 /*
1009 * Block out ext4_truncate while we alter the tree
1010 */
1011 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
1012 &count, goal,
1013 offsets + (partial - chain), partial);
1014
1015 /*
1016 * The ext4_splice_branch call will free and forget any buffers
1017 * on the new chain if there is a failure, but that risks using
1018 * up transaction credits, especially for bitmaps where the
1019 * credits cannot be returned. Can we handle this somehow? We
1020 * may need to return -EAGAIN upwards in the worst case. --sct
1021 */
1022 if (!err)
1023 err = ext4_splice_branch(handle, inode, iblock,
1024 partial, indirect_blks, count);
1025 if (err)
1026 goto cleanup;
1027
1028 set_buffer_new(bh_result);
1029
1030 ext4_update_inode_fsync_trans(handle, inode, 1);
1031 got_it:
1032 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1033 if (count > blocks_to_boundary)
1034 set_buffer_boundary(bh_result);
1035 err = count;
1036 /* Clean up and exit */
1037 partial = chain + depth - 1; /* the whole chain */
1038 cleanup:
1039 while (partial > chain) {
1040 BUFFER_TRACE(partial->bh, "call brelse");
1041 brelse(partial->bh);
1042 partial--;
1043 }
1044 BUFFER_TRACE(bh_result, "returned");
1045 out:
1046 return err;
1047 }
1048
1049 #ifdef CONFIG_QUOTA
1050 qsize_t *ext4_get_reserved_space(struct inode *inode)
1051 {
1052 return &EXT4_I(inode)->i_reserved_quota;
1053 }
1054 #endif
1055 /*
1056 * Calculate the number of metadata blocks need to reserve
1057 * to allocate @blocks for non extent file based file
1058 */
1059 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1060 {
1061 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1062 int ind_blks, dind_blks, tind_blks;
1063
1064 /* number of new indirect blocks needed */
1065 ind_blks = (blocks + icap - 1) / icap;
1066
1067 dind_blks = (ind_blks + icap - 1) / icap;
1068
1069 tind_blks = 1;
1070
1071 return ind_blks + dind_blks + tind_blks;
1072 }
1073
1074 /*
1075 * Calculate the number of metadata blocks need to reserve
1076 * to allocate given number of blocks
1077 */
1078 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1079 {
1080 if (!blocks)
1081 return 0;
1082
1083 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1084 return ext4_ext_calc_metadata_amount(inode, blocks);
1085
1086 return ext4_indirect_calc_metadata_amount(inode, blocks);
1087 }
1088
1089 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1090 {
1091 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1092 int total, mdb, mdb_free;
1093
1094 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1095 /* recalculate the number of metablocks still need to be reserved */
1096 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1097 mdb = ext4_calc_metadata_amount(inode, total);
1098
1099 /* figure out how many metablocks to release */
1100 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1101 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1102
1103 if (mdb_free) {
1104 /* Account for allocated meta_blocks */
1105 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1106
1107 /* update fs dirty blocks counter */
1108 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1109 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1110 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1111 }
1112
1113 /* update per-inode reservations */
1114 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1115 EXT4_I(inode)->i_reserved_data_blocks -= used;
1116 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1117
1118 /*
1119 * free those over-booking quota for metadata blocks
1120 */
1121 if (mdb_free)
1122 vfs_dq_release_reservation_block(inode, mdb_free);
1123
1124 /*
1125 * If we have done all the pending block allocations and if
1126 * there aren't any writers on the inode, we can discard the
1127 * inode's preallocations.
1128 */
1129 if (!total && (atomic_read(&inode->i_writecount) == 0))
1130 ext4_discard_preallocations(inode);
1131 }
1132
1133 static int check_block_validity(struct inode *inode, const char *msg,
1134 sector_t logical, sector_t phys, int len)
1135 {
1136 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
1137 ext4_error(inode->i_sb, msg,
1138 "inode #%lu logical block %llu mapped to %llu "
1139 "(size %d)", inode->i_ino,
1140 (unsigned long long) logical,
1141 (unsigned long long) phys, len);
1142 return -EIO;
1143 }
1144 return 0;
1145 }
1146
1147 /*
1148 * Return the number of contiguous dirty pages in a given inode
1149 * starting at page frame idx.
1150 */
1151 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1152 unsigned int max_pages)
1153 {
1154 struct address_space *mapping = inode->i_mapping;
1155 pgoff_t index;
1156 struct pagevec pvec;
1157 pgoff_t num = 0;
1158 int i, nr_pages, done = 0;
1159
1160 if (max_pages == 0)
1161 return 0;
1162 pagevec_init(&pvec, 0);
1163 while (!done) {
1164 index = idx;
1165 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1166 PAGECACHE_TAG_DIRTY,
1167 (pgoff_t)PAGEVEC_SIZE);
1168 if (nr_pages == 0)
1169 break;
1170 for (i = 0; i < nr_pages; i++) {
1171 struct page *page = pvec.pages[i];
1172 struct buffer_head *bh, *head;
1173
1174 lock_page(page);
1175 if (unlikely(page->mapping != mapping) ||
1176 !PageDirty(page) ||
1177 PageWriteback(page) ||
1178 page->index != idx) {
1179 done = 1;
1180 unlock_page(page);
1181 break;
1182 }
1183 if (page_has_buffers(page)) {
1184 bh = head = page_buffers(page);
1185 do {
1186 if (!buffer_delay(bh) &&
1187 !buffer_unwritten(bh))
1188 done = 1;
1189 bh = bh->b_this_page;
1190 } while (!done && (bh != head));
1191 }
1192 unlock_page(page);
1193 if (done)
1194 break;
1195 idx++;
1196 num++;
1197 if (num >= max_pages)
1198 break;
1199 }
1200 pagevec_release(&pvec);
1201 }
1202 return num;
1203 }
1204
1205 /*
1206 * The ext4_get_blocks() function tries to look up the requested blocks,
1207 * and returns if the blocks are already mapped.
1208 *
1209 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1210 * and store the allocated blocks in the result buffer head and mark it
1211 * mapped.
1212 *
1213 * If file type is extents based, it will call ext4_ext_get_blocks(),
1214 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1215 * based files
1216 *
1217 * On success, it returns the number of blocks being mapped or allocate.
1218 * if create==0 and the blocks are pre-allocated and uninitialized block,
1219 * the result buffer head is unmapped. If the create ==1, it will make sure
1220 * the buffer head is mapped.
1221 *
1222 * It returns 0 if plain look up failed (blocks have not been allocated), in
1223 * that casem, buffer head is unmapped
1224 *
1225 * It returns the error in case of allocation failure.
1226 */
1227 int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1228 unsigned int max_blocks, struct buffer_head *bh,
1229 int flags)
1230 {
1231 int retval;
1232
1233 clear_buffer_mapped(bh);
1234 clear_buffer_unwritten(bh);
1235
1236 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1237 "logical block %lu\n", inode->i_ino, flags, max_blocks,
1238 (unsigned long)block);
1239 /*
1240 * Try to see if we can get the block without requesting a new
1241 * file system block.
1242 */
1243 down_read((&EXT4_I(inode)->i_data_sem));
1244 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1245 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1246 bh, 0);
1247 } else {
1248 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
1249 bh, 0);
1250 }
1251 up_read((&EXT4_I(inode)->i_data_sem));
1252
1253 if (retval > 0 && buffer_mapped(bh)) {
1254 int ret = check_block_validity(inode, "file system corruption",
1255 block, bh->b_blocknr, retval);
1256 if (ret != 0)
1257 return ret;
1258 }
1259
1260 /* If it is only a block(s) look up */
1261 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1262 return retval;
1263
1264 /*
1265 * Returns if the blocks have already allocated
1266 *
1267 * Note that if blocks have been preallocated
1268 * ext4_ext_get_block() returns th create = 0
1269 * with buffer head unmapped.
1270 */
1271 if (retval > 0 && buffer_mapped(bh))
1272 return retval;
1273
1274 /*
1275 * When we call get_blocks without the create flag, the
1276 * BH_Unwritten flag could have gotten set if the blocks
1277 * requested were part of a uninitialized extent. We need to
1278 * clear this flag now that we are committed to convert all or
1279 * part of the uninitialized extent to be an initialized
1280 * extent. This is because we need to avoid the combination
1281 * of BH_Unwritten and BH_Mapped flags being simultaneously
1282 * set on the buffer_head.
1283 */
1284 clear_buffer_unwritten(bh);
1285
1286 /*
1287 * New blocks allocate and/or writing to uninitialized extent
1288 * will possibly result in updating i_data, so we take
1289 * the write lock of i_data_sem, and call get_blocks()
1290 * with create == 1 flag.
1291 */
1292 down_write((&EXT4_I(inode)->i_data_sem));
1293
1294 /*
1295 * if the caller is from delayed allocation writeout path
1296 * we have already reserved fs blocks for allocation
1297 * let the underlying get_block() function know to
1298 * avoid double accounting
1299 */
1300 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1301 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1302 /*
1303 * We need to check for EXT4 here because migrate
1304 * could have changed the inode type in between
1305 */
1306 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1307 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1308 bh, flags);
1309 } else {
1310 retval = ext4_ind_get_blocks(handle, inode, block,
1311 max_blocks, bh, flags);
1312
1313 if (retval > 0 && buffer_new(bh)) {
1314 /*
1315 * We allocated new blocks which will result in
1316 * i_data's format changing. Force the migrate
1317 * to fail by clearing migrate flags
1318 */
1319 EXT4_I(inode)->i_state &= ~EXT4_STATE_EXT_MIGRATE;
1320 }
1321 }
1322
1323 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1324 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1325
1326 /*
1327 * Update reserved blocks/metadata blocks after successful
1328 * block allocation which had been deferred till now.
1329 */
1330 if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE))
1331 ext4_da_update_reserve_space(inode, retval);
1332
1333 up_write((&EXT4_I(inode)->i_data_sem));
1334 if (retval > 0 && buffer_mapped(bh)) {
1335 int ret = check_block_validity(inode, "file system "
1336 "corruption after allocation",
1337 block, bh->b_blocknr, retval);
1338 if (ret != 0)
1339 return ret;
1340 }
1341 return retval;
1342 }
1343
1344 /* Maximum number of blocks we map for direct IO at once. */
1345 #define DIO_MAX_BLOCKS 4096
1346
1347 int ext4_get_block(struct inode *inode, sector_t iblock,
1348 struct buffer_head *bh_result, int create)
1349 {
1350 handle_t *handle = ext4_journal_current_handle();
1351 int ret = 0, started = 0;
1352 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1353 int dio_credits;
1354
1355 if (create && !handle) {
1356 /* Direct IO write... */
1357 if (max_blocks > DIO_MAX_BLOCKS)
1358 max_blocks = DIO_MAX_BLOCKS;
1359 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1360 handle = ext4_journal_start(inode, dio_credits);
1361 if (IS_ERR(handle)) {
1362 ret = PTR_ERR(handle);
1363 goto out;
1364 }
1365 started = 1;
1366 }
1367
1368 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
1369 create ? EXT4_GET_BLOCKS_CREATE : 0);
1370 if (ret > 0) {
1371 bh_result->b_size = (ret << inode->i_blkbits);
1372 ret = 0;
1373 }
1374 if (started)
1375 ext4_journal_stop(handle);
1376 out:
1377 return ret;
1378 }
1379
1380 /*
1381 * `handle' can be NULL if create is zero
1382 */
1383 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1384 ext4_lblk_t block, int create, int *errp)
1385 {
1386 struct buffer_head dummy;
1387 int fatal = 0, err;
1388 int flags = 0;
1389
1390 J_ASSERT(handle != NULL || create == 0);
1391
1392 dummy.b_state = 0;
1393 dummy.b_blocknr = -1000;
1394 buffer_trace_init(&dummy.b_history);
1395 if (create)
1396 flags |= EXT4_GET_BLOCKS_CREATE;
1397 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
1398 /*
1399 * ext4_get_blocks() returns number of blocks mapped. 0 in
1400 * case of a HOLE.
1401 */
1402 if (err > 0) {
1403 if (err > 1)
1404 WARN_ON(1);
1405 err = 0;
1406 }
1407 *errp = err;
1408 if (!err && buffer_mapped(&dummy)) {
1409 struct buffer_head *bh;
1410 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1411 if (!bh) {
1412 *errp = -EIO;
1413 goto err;
1414 }
1415 if (buffer_new(&dummy)) {
1416 J_ASSERT(create != 0);
1417 J_ASSERT(handle != NULL);
1418
1419 /*
1420 * Now that we do not always journal data, we should
1421 * keep in mind whether this should always journal the
1422 * new buffer as metadata. For now, regular file
1423 * writes use ext4_get_block instead, so it's not a
1424 * problem.
1425 */
1426 lock_buffer(bh);
1427 BUFFER_TRACE(bh, "call get_create_access");
1428 fatal = ext4_journal_get_create_access(handle, bh);
1429 if (!fatal && !buffer_uptodate(bh)) {
1430 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1431 set_buffer_uptodate(bh);
1432 }
1433 unlock_buffer(bh);
1434 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1435 err = ext4_handle_dirty_metadata(handle, inode, bh);
1436 if (!fatal)
1437 fatal = err;
1438 } else {
1439 BUFFER_TRACE(bh, "not a new buffer");
1440 }
1441 if (fatal) {
1442 *errp = fatal;
1443 brelse(bh);
1444 bh = NULL;
1445 }
1446 return bh;
1447 }
1448 err:
1449 return NULL;
1450 }
1451
1452 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1453 ext4_lblk_t block, int create, int *err)
1454 {
1455 struct buffer_head *bh;
1456
1457 bh = ext4_getblk(handle, inode, block, create, err);
1458 if (!bh)
1459 return bh;
1460 if (buffer_uptodate(bh))
1461 return bh;
1462 ll_rw_block(READ_META, 1, &bh);
1463 wait_on_buffer(bh);
1464 if (buffer_uptodate(bh))
1465 return bh;
1466 put_bh(bh);
1467 *err = -EIO;
1468 return NULL;
1469 }
1470
1471 static int walk_page_buffers(handle_t *handle,
1472 struct buffer_head *head,
1473 unsigned from,
1474 unsigned to,
1475 int *partial,
1476 int (*fn)(handle_t *handle,
1477 struct buffer_head *bh))
1478 {
1479 struct buffer_head *bh;
1480 unsigned block_start, block_end;
1481 unsigned blocksize = head->b_size;
1482 int err, ret = 0;
1483 struct buffer_head *next;
1484
1485 for (bh = head, block_start = 0;
1486 ret == 0 && (bh != head || !block_start);
1487 block_start = block_end, bh = next) {
1488 next = bh->b_this_page;
1489 block_end = block_start + blocksize;
1490 if (block_end <= from || block_start >= to) {
1491 if (partial && !buffer_uptodate(bh))
1492 *partial = 1;
1493 continue;
1494 }
1495 err = (*fn)(handle, bh);
1496 if (!ret)
1497 ret = err;
1498 }
1499 return ret;
1500 }
1501
1502 /*
1503 * To preserve ordering, it is essential that the hole instantiation and
1504 * the data write be encapsulated in a single transaction. We cannot
1505 * close off a transaction and start a new one between the ext4_get_block()
1506 * and the commit_write(). So doing the jbd2_journal_start at the start of
1507 * prepare_write() is the right place.
1508 *
1509 * Also, this function can nest inside ext4_writepage() ->
1510 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1511 * has generated enough buffer credits to do the whole page. So we won't
1512 * block on the journal in that case, which is good, because the caller may
1513 * be PF_MEMALLOC.
1514 *
1515 * By accident, ext4 can be reentered when a transaction is open via
1516 * quota file writes. If we were to commit the transaction while thus
1517 * reentered, there can be a deadlock - we would be holding a quota
1518 * lock, and the commit would never complete if another thread had a
1519 * transaction open and was blocking on the quota lock - a ranking
1520 * violation.
1521 *
1522 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1523 * will _not_ run commit under these circumstances because handle->h_ref
1524 * is elevated. We'll still have enough credits for the tiny quotafile
1525 * write.
1526 */
1527 static int do_journal_get_write_access(handle_t *handle,
1528 struct buffer_head *bh)
1529 {
1530 if (!buffer_mapped(bh) || buffer_freed(bh))
1531 return 0;
1532 return ext4_journal_get_write_access(handle, bh);
1533 }
1534
1535 /*
1536 * Truncate blocks that were not used by write. We have to truncate the
1537 * pagecache as well so that corresponding buffers get properly unmapped.
1538 */
1539 static void ext4_truncate_failed_write(struct inode *inode)
1540 {
1541 truncate_inode_pages(inode->i_mapping, inode->i_size);
1542 ext4_truncate(inode);
1543 }
1544
1545 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1546 loff_t pos, unsigned len, unsigned flags,
1547 struct page **pagep, void **fsdata)
1548 {
1549 struct inode *inode = mapping->host;
1550 int ret, needed_blocks;
1551 handle_t *handle;
1552 int retries = 0;
1553 struct page *page;
1554 pgoff_t index;
1555 unsigned from, to;
1556
1557 trace_ext4_write_begin(inode, pos, len, flags);
1558 /*
1559 * Reserve one block more for addition to orphan list in case
1560 * we allocate blocks but write fails for some reason
1561 */
1562 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1563 index = pos >> PAGE_CACHE_SHIFT;
1564 from = pos & (PAGE_CACHE_SIZE - 1);
1565 to = from + len;
1566
1567 retry:
1568 handle = ext4_journal_start(inode, needed_blocks);
1569 if (IS_ERR(handle)) {
1570 ret = PTR_ERR(handle);
1571 goto out;
1572 }
1573
1574 /* We cannot recurse into the filesystem as the transaction is already
1575 * started */
1576 flags |= AOP_FLAG_NOFS;
1577
1578 page = grab_cache_page_write_begin(mapping, index, flags);
1579 if (!page) {
1580 ext4_journal_stop(handle);
1581 ret = -ENOMEM;
1582 goto out;
1583 }
1584 *pagep = page;
1585
1586 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1587 ext4_get_block);
1588
1589 if (!ret && ext4_should_journal_data(inode)) {
1590 ret = walk_page_buffers(handle, page_buffers(page),
1591 from, to, NULL, do_journal_get_write_access);
1592 }
1593
1594 if (ret) {
1595 unlock_page(page);
1596 page_cache_release(page);
1597 /*
1598 * block_write_begin may have instantiated a few blocks
1599 * outside i_size. Trim these off again. Don't need
1600 * i_size_read because we hold i_mutex.
1601 *
1602 * Add inode to orphan list in case we crash before
1603 * truncate finishes
1604 */
1605 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1606 ext4_orphan_add(handle, inode);
1607
1608 ext4_journal_stop(handle);
1609 if (pos + len > inode->i_size) {
1610 ext4_truncate_failed_write(inode);
1611 /*
1612 * If truncate failed early the inode might
1613 * still be on the orphan list; we need to
1614 * make sure the inode is removed from the
1615 * orphan list in that case.
1616 */
1617 if (inode->i_nlink)
1618 ext4_orphan_del(NULL, inode);
1619 }
1620 }
1621
1622 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1623 goto retry;
1624 out:
1625 return ret;
1626 }
1627
1628 /* For write_end() in data=journal mode */
1629 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1630 {
1631 if (!buffer_mapped(bh) || buffer_freed(bh))
1632 return 0;
1633 set_buffer_uptodate(bh);
1634 return ext4_handle_dirty_metadata(handle, NULL, bh);
1635 }
1636
1637 static int ext4_generic_write_end(struct file *file,
1638 struct address_space *mapping,
1639 loff_t pos, unsigned len, unsigned copied,
1640 struct page *page, void *fsdata)
1641 {
1642 int i_size_changed = 0;
1643 struct inode *inode = mapping->host;
1644 handle_t *handle = ext4_journal_current_handle();
1645
1646 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1647
1648 /*
1649 * No need to use i_size_read() here, the i_size
1650 * cannot change under us because we hold i_mutex.
1651 *
1652 * But it's important to update i_size while still holding page lock:
1653 * page writeout could otherwise come in and zero beyond i_size.
1654 */
1655 if (pos + copied > inode->i_size) {
1656 i_size_write(inode, pos + copied);
1657 i_size_changed = 1;
1658 }
1659
1660 if (pos + copied > EXT4_I(inode)->i_disksize) {
1661 /* We need to mark inode dirty even if
1662 * new_i_size is less that inode->i_size
1663 * bu greater than i_disksize.(hint delalloc)
1664 */
1665 ext4_update_i_disksize(inode, (pos + copied));
1666 i_size_changed = 1;
1667 }
1668 unlock_page(page);
1669 page_cache_release(page);
1670
1671 /*
1672 * Don't mark the inode dirty under page lock. First, it unnecessarily
1673 * makes the holding time of page lock longer. Second, it forces lock
1674 * ordering of page lock and transaction start for journaling
1675 * filesystems.
1676 */
1677 if (i_size_changed)
1678 ext4_mark_inode_dirty(handle, inode);
1679
1680 return copied;
1681 }
1682
1683 /*
1684 * We need to pick up the new inode size which generic_commit_write gave us
1685 * `file' can be NULL - eg, when called from page_symlink().
1686 *
1687 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1688 * buffers are managed internally.
1689 */
1690 static int ext4_ordered_write_end(struct file *file,
1691 struct address_space *mapping,
1692 loff_t pos, unsigned len, unsigned copied,
1693 struct page *page, void *fsdata)
1694 {
1695 handle_t *handle = ext4_journal_current_handle();
1696 struct inode *inode = mapping->host;
1697 int ret = 0, ret2;
1698
1699 trace_ext4_ordered_write_end(inode, pos, len, copied);
1700 ret = ext4_jbd2_file_inode(handle, inode);
1701
1702 if (ret == 0) {
1703 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1704 page, fsdata);
1705 copied = ret2;
1706 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1707 /* if we have allocated more blocks and copied
1708 * less. We will have blocks allocated outside
1709 * inode->i_size. So truncate them
1710 */
1711 ext4_orphan_add(handle, inode);
1712 if (ret2 < 0)
1713 ret = ret2;
1714 }
1715 ret2 = ext4_journal_stop(handle);
1716 if (!ret)
1717 ret = ret2;
1718
1719 if (pos + len > inode->i_size) {
1720 ext4_truncate_failed_write(inode);
1721 /*
1722 * If truncate failed early the inode might still be
1723 * on the orphan list; we need to make sure the inode
1724 * is removed from the orphan list in that case.
1725 */
1726 if (inode->i_nlink)
1727 ext4_orphan_del(NULL, inode);
1728 }
1729
1730
1731 return ret ? ret : copied;
1732 }
1733
1734 static int ext4_writeback_write_end(struct file *file,
1735 struct address_space *mapping,
1736 loff_t pos, unsigned len, unsigned copied,
1737 struct page *page, void *fsdata)
1738 {
1739 handle_t *handle = ext4_journal_current_handle();
1740 struct inode *inode = mapping->host;
1741 int ret = 0, ret2;
1742
1743 trace_ext4_writeback_write_end(inode, pos, len, copied);
1744 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1745 page, fsdata);
1746 copied = ret2;
1747 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1748 /* if we have allocated more blocks and copied
1749 * less. We will have blocks allocated outside
1750 * inode->i_size. So truncate them
1751 */
1752 ext4_orphan_add(handle, inode);
1753
1754 if (ret2 < 0)
1755 ret = ret2;
1756
1757 ret2 = ext4_journal_stop(handle);
1758 if (!ret)
1759 ret = ret2;
1760
1761 if (pos + len > inode->i_size) {
1762 ext4_truncate_failed_write(inode);
1763 /*
1764 * If truncate failed early the inode might still be
1765 * on the orphan list; we need to make sure the inode
1766 * is removed from the orphan list in that case.
1767 */
1768 if (inode->i_nlink)
1769 ext4_orphan_del(NULL, inode);
1770 }
1771
1772 return ret ? ret : copied;
1773 }
1774
1775 static int ext4_journalled_write_end(struct file *file,
1776 struct address_space *mapping,
1777 loff_t pos, unsigned len, unsigned copied,
1778 struct page *page, void *fsdata)
1779 {
1780 handle_t *handle = ext4_journal_current_handle();
1781 struct inode *inode = mapping->host;
1782 int ret = 0, ret2;
1783 int partial = 0;
1784 unsigned from, to;
1785 loff_t new_i_size;
1786
1787 trace_ext4_journalled_write_end(inode, pos, len, copied);
1788 from = pos & (PAGE_CACHE_SIZE - 1);
1789 to = from + len;
1790
1791 if (copied < len) {
1792 if (!PageUptodate(page))
1793 copied = 0;
1794 page_zero_new_buffers(page, from+copied, to);
1795 }
1796
1797 ret = walk_page_buffers(handle, page_buffers(page), from,
1798 to, &partial, write_end_fn);
1799 if (!partial)
1800 SetPageUptodate(page);
1801 new_i_size = pos + copied;
1802 if (new_i_size > inode->i_size)
1803 i_size_write(inode, pos+copied);
1804 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1805 if (new_i_size > EXT4_I(inode)->i_disksize) {
1806 ext4_update_i_disksize(inode, new_i_size);
1807 ret2 = ext4_mark_inode_dirty(handle, inode);
1808 if (!ret)
1809 ret = ret2;
1810 }
1811
1812 unlock_page(page);
1813 page_cache_release(page);
1814 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1815 /* if we have allocated more blocks and copied
1816 * less. We will have blocks allocated outside
1817 * inode->i_size. So truncate them
1818 */
1819 ext4_orphan_add(handle, inode);
1820
1821 ret2 = ext4_journal_stop(handle);
1822 if (!ret)
1823 ret = ret2;
1824 if (pos + len > inode->i_size) {
1825 ext4_truncate_failed_write(inode);
1826 /*
1827 * If truncate failed early the inode might still be
1828 * on the orphan list; we need to make sure the inode
1829 * is removed from the orphan list in that case.
1830 */
1831 if (inode->i_nlink)
1832 ext4_orphan_del(NULL, inode);
1833 }
1834
1835 return ret ? ret : copied;
1836 }
1837
1838 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1839 {
1840 int retries = 0;
1841 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1842 unsigned long md_needed, mdblocks, total = 0;
1843
1844 /*
1845 * recalculate the amount of metadata blocks to reserve
1846 * in order to allocate nrblocks
1847 * worse case is one extent per block
1848 */
1849 repeat:
1850 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1851 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1852 mdblocks = ext4_calc_metadata_amount(inode, total);
1853 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1854
1855 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1856 total = md_needed + nrblocks;
1857 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1858
1859 /*
1860 * Make quota reservation here to prevent quota overflow
1861 * later. Real quota accounting is done at pages writeout
1862 * time.
1863 */
1864 if (vfs_dq_reserve_block(inode, total))
1865 return -EDQUOT;
1866
1867 if (ext4_claim_free_blocks(sbi, total)) {
1868 vfs_dq_release_reservation_block(inode, total);
1869 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1870 yield();
1871 goto repeat;
1872 }
1873 return -ENOSPC;
1874 }
1875 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1876 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1877 EXT4_I(inode)->i_reserved_meta_blocks += md_needed;
1878 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1879
1880 return 0; /* success */
1881 }
1882
1883 static void ext4_da_release_space(struct inode *inode, int to_free)
1884 {
1885 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1886 int total, mdb, mdb_free, release;
1887
1888 if (!to_free)
1889 return; /* Nothing to release, exit */
1890
1891 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1892
1893 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1894 /*
1895 * if there is no reserved blocks, but we try to free some
1896 * then the counter is messed up somewhere.
1897 * but since this function is called from invalidate
1898 * page, it's harmless to return without any action
1899 */
1900 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1901 "blocks for inode %lu, but there is no reserved "
1902 "data blocks\n", to_free, inode->i_ino);
1903 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1904 return;
1905 }
1906
1907 /* recalculate the number of metablocks still need to be reserved */
1908 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1909 mdb = ext4_calc_metadata_amount(inode, total);
1910
1911 /* figure out how many metablocks to release */
1912 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1913 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1914
1915 release = to_free + mdb_free;
1916
1917 /* update fs dirty blocks counter for truncate case */
1918 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1919
1920 /* update per-inode reservations */
1921 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1922 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1923
1924 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1925 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1926 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1927
1928 vfs_dq_release_reservation_block(inode, release);
1929 }
1930
1931 static void ext4_da_page_release_reservation(struct page *page,
1932 unsigned long offset)
1933 {
1934 int to_release = 0;
1935 struct buffer_head *head, *bh;
1936 unsigned int curr_off = 0;
1937
1938 head = page_buffers(page);
1939 bh = head;
1940 do {
1941 unsigned int next_off = curr_off + bh->b_size;
1942
1943 if ((offset <= curr_off) && (buffer_delay(bh))) {
1944 to_release++;
1945 clear_buffer_delay(bh);
1946 }
1947 curr_off = next_off;
1948 } while ((bh = bh->b_this_page) != head);
1949 ext4_da_release_space(page->mapping->host, to_release);
1950 }
1951
1952 /*
1953 * mpage_da_submit_io - walks through extent of pages and try to write
1954 * them with writepage() call back
1955 *
1956 * @mpd->inode: inode
1957 * @mpd->first_page: first page of the extent
1958 * @mpd->next_page: page after the last page of the extent
1959 *
1960 * By the time mpage_da_submit_io() is called we expect all blocks
1961 * to be allocated. this may be wrong if allocation failed.
1962 *
1963 * As pages are already locked by write_cache_pages(), we can't use it
1964 */
1965 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1966 {
1967 long pages_skipped;
1968 struct pagevec pvec;
1969 unsigned long index, end;
1970 int ret = 0, err, nr_pages, i;
1971 struct inode *inode = mpd->inode;
1972 struct address_space *mapping = inode->i_mapping;
1973
1974 BUG_ON(mpd->next_page <= mpd->first_page);
1975 /*
1976 * We need to start from the first_page to the next_page - 1
1977 * to make sure we also write the mapped dirty buffer_heads.
1978 * If we look at mpd->b_blocknr we would only be looking
1979 * at the currently mapped buffer_heads.
1980 */
1981 index = mpd->first_page;
1982 end = mpd->next_page - 1;
1983
1984 pagevec_init(&pvec, 0);
1985 while (index <= end) {
1986 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1987 if (nr_pages == 0)
1988 break;
1989 for (i = 0; i < nr_pages; i++) {
1990 struct page *page = pvec.pages[i];
1991
1992 index = page->index;
1993 if (index > end)
1994 break;
1995 index++;
1996
1997 BUG_ON(!PageLocked(page));
1998 BUG_ON(PageWriteback(page));
1999
2000 pages_skipped = mpd->wbc->pages_skipped;
2001 err = mapping->a_ops->writepage(page, mpd->wbc);
2002 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2003 /*
2004 * have successfully written the page
2005 * without skipping the same
2006 */
2007 mpd->pages_written++;
2008 /*
2009 * In error case, we have to continue because
2010 * remaining pages are still locked
2011 * XXX: unlock and re-dirty them?
2012 */
2013 if (ret == 0)
2014 ret = err;
2015 }
2016 pagevec_release(&pvec);
2017 }
2018 return ret;
2019 }
2020
2021 /*
2022 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2023 *
2024 * @mpd->inode - inode to walk through
2025 * @exbh->b_blocknr - first block on a disk
2026 * @exbh->b_size - amount of space in bytes
2027 * @logical - first logical block to start assignment with
2028 *
2029 * the function goes through all passed space and put actual disk
2030 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2031 */
2032 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
2033 struct buffer_head *exbh)
2034 {
2035 struct inode *inode = mpd->inode;
2036 struct address_space *mapping = inode->i_mapping;
2037 int blocks = exbh->b_size >> inode->i_blkbits;
2038 sector_t pblock = exbh->b_blocknr, cur_logical;
2039 struct buffer_head *head, *bh;
2040 pgoff_t index, end;
2041 struct pagevec pvec;
2042 int nr_pages, i;
2043
2044 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2045 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2046 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2047
2048 pagevec_init(&pvec, 0);
2049
2050 while (index <= end) {
2051 /* XXX: optimize tail */
2052 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2053 if (nr_pages == 0)
2054 break;
2055 for (i = 0; i < nr_pages; i++) {
2056 struct page *page = pvec.pages[i];
2057
2058 index = page->index;
2059 if (index > end)
2060 break;
2061 index++;
2062
2063 BUG_ON(!PageLocked(page));
2064 BUG_ON(PageWriteback(page));
2065 BUG_ON(!page_has_buffers(page));
2066
2067 bh = page_buffers(page);
2068 head = bh;
2069
2070 /* skip blocks out of the range */
2071 do {
2072 if (cur_logical >= logical)
2073 break;
2074 cur_logical++;
2075 } while ((bh = bh->b_this_page) != head);
2076
2077 do {
2078 if (cur_logical >= logical + blocks)
2079 break;
2080
2081 if (buffer_delay(bh) ||
2082 buffer_unwritten(bh)) {
2083
2084 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2085
2086 if (buffer_delay(bh)) {
2087 clear_buffer_delay(bh);
2088 bh->b_blocknr = pblock;
2089 } else {
2090 /*
2091 * unwritten already should have
2092 * blocknr assigned. Verify that
2093 */
2094 clear_buffer_unwritten(bh);
2095 BUG_ON(bh->b_blocknr != pblock);
2096 }
2097
2098 } else if (buffer_mapped(bh))
2099 BUG_ON(bh->b_blocknr != pblock);
2100
2101 cur_logical++;
2102 pblock++;
2103 } while ((bh = bh->b_this_page) != head);
2104 }
2105 pagevec_release(&pvec);
2106 }
2107 }
2108
2109
2110 /*
2111 * __unmap_underlying_blocks - just a helper function to unmap
2112 * set of blocks described by @bh
2113 */
2114 static inline void __unmap_underlying_blocks(struct inode *inode,
2115 struct buffer_head *bh)
2116 {
2117 struct block_device *bdev = inode->i_sb->s_bdev;
2118 int blocks, i;
2119
2120 blocks = bh->b_size >> inode->i_blkbits;
2121 for (i = 0; i < blocks; i++)
2122 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
2123 }
2124
2125 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2126 sector_t logical, long blk_cnt)
2127 {
2128 int nr_pages, i;
2129 pgoff_t index, end;
2130 struct pagevec pvec;
2131 struct inode *inode = mpd->inode;
2132 struct address_space *mapping = inode->i_mapping;
2133
2134 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2135 end = (logical + blk_cnt - 1) >>
2136 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2137 while (index <= end) {
2138 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2139 if (nr_pages == 0)
2140 break;
2141 for (i = 0; i < nr_pages; i++) {
2142 struct page *page = pvec.pages[i];
2143 index = page->index;
2144 if (index > end)
2145 break;
2146 index++;
2147
2148 BUG_ON(!PageLocked(page));
2149 BUG_ON(PageWriteback(page));
2150 block_invalidatepage(page, 0);
2151 ClearPageUptodate(page);
2152 unlock_page(page);
2153 }
2154 }
2155 return;
2156 }
2157
2158 static void ext4_print_free_blocks(struct inode *inode)
2159 {
2160 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2161 printk(KERN_EMERG "Total free blocks count %lld\n",
2162 ext4_count_free_blocks(inode->i_sb));
2163 printk(KERN_EMERG "Free/Dirty block details\n");
2164 printk(KERN_EMERG "free_blocks=%lld\n",
2165 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
2166 printk(KERN_EMERG "dirty_blocks=%lld\n",
2167 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2168 printk(KERN_EMERG "Block reservation details\n");
2169 printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
2170 EXT4_I(inode)->i_reserved_data_blocks);
2171 printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
2172 EXT4_I(inode)->i_reserved_meta_blocks);
2173 return;
2174 }
2175
2176 /*
2177 * mpage_da_map_blocks - go through given space
2178 *
2179 * @mpd - bh describing space
2180 *
2181 * The function skips space we know is already mapped to disk blocks.
2182 *
2183 */
2184 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2185 {
2186 int err, blks, get_blocks_flags;
2187 struct buffer_head new;
2188 sector_t next = mpd->b_blocknr;
2189 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2190 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2191 handle_t *handle = NULL;
2192
2193 /*
2194 * We consider only non-mapped and non-allocated blocks
2195 */
2196 if ((mpd->b_state & (1 << BH_Mapped)) &&
2197 !(mpd->b_state & (1 << BH_Delay)) &&
2198 !(mpd->b_state & (1 << BH_Unwritten)))
2199 return 0;
2200
2201 /*
2202 * If we didn't accumulate anything to write simply return
2203 */
2204 if (!mpd->b_size)
2205 return 0;
2206
2207 handle = ext4_journal_current_handle();
2208 BUG_ON(!handle);
2209
2210 /*
2211 * Call ext4_get_blocks() to allocate any delayed allocation
2212 * blocks, or to convert an uninitialized extent to be
2213 * initialized (in the case where we have written into
2214 * one or more preallocated blocks).
2215 *
2216 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2217 * indicate that we are on the delayed allocation path. This
2218 * affects functions in many different parts of the allocation
2219 * call path. This flag exists primarily because we don't
2220 * want to change *many* call functions, so ext4_get_blocks()
2221 * will set the magic i_delalloc_reserved_flag once the
2222 * inode's allocation semaphore is taken.
2223 *
2224 * If the blocks in questions were delalloc blocks, set
2225 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2226 * variables are updated after the blocks have been allocated.
2227 */
2228 new.b_state = 0;
2229 get_blocks_flags = (EXT4_GET_BLOCKS_CREATE |
2230 EXT4_GET_BLOCKS_DELALLOC_RESERVE);
2231 if (mpd->b_state & (1 << BH_Delay))
2232 get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE;
2233 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
2234 &new, get_blocks_flags);
2235 if (blks < 0) {
2236 err = blks;
2237 /*
2238 * If get block returns with error we simply
2239 * return. Later writepage will redirty the page and
2240 * writepages will find the dirty page again
2241 */
2242 if (err == -EAGAIN)
2243 return 0;
2244
2245 if (err == -ENOSPC &&
2246 ext4_count_free_blocks(mpd->inode->i_sb)) {
2247 mpd->retval = err;
2248 return 0;
2249 }
2250
2251 /*
2252 * get block failure will cause us to loop in
2253 * writepages, because a_ops->writepage won't be able
2254 * to make progress. The page will be redirtied by
2255 * writepage and writepages will again try to write
2256 * the same.
2257 */
2258 printk(KERN_EMERG "%s block allocation failed for inode %lu "
2259 "at logical offset %llu with max blocks "
2260 "%zd with error %d\n",
2261 __func__, mpd->inode->i_ino,
2262 (unsigned long long)next,
2263 mpd->b_size >> mpd->inode->i_blkbits, err);
2264 printk(KERN_EMERG "This should not happen.!! "
2265 "Data will be lost\n");
2266 if (err == -ENOSPC) {
2267 ext4_print_free_blocks(mpd->inode);
2268 }
2269 /* invalidate all the pages */
2270 ext4_da_block_invalidatepages(mpd, next,
2271 mpd->b_size >> mpd->inode->i_blkbits);
2272 return err;
2273 }
2274 BUG_ON(blks == 0);
2275
2276 new.b_size = (blks << mpd->inode->i_blkbits);
2277
2278 if (buffer_new(&new))
2279 __unmap_underlying_blocks(mpd->inode, &new);
2280
2281 /*
2282 * If blocks are delayed marked, we need to
2283 * put actual blocknr and drop delayed bit
2284 */
2285 if ((mpd->b_state & (1 << BH_Delay)) ||
2286 (mpd->b_state & (1 << BH_Unwritten)))
2287 mpage_put_bnr_to_bhs(mpd, next, &new);
2288
2289 if (ext4_should_order_data(mpd->inode)) {
2290 err = ext4_jbd2_file_inode(handle, mpd->inode);
2291 if (err)
2292 return err;
2293 }
2294
2295 /*
2296 * Update on-disk size along with block allocation.
2297 */
2298 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2299 if (disksize > i_size_read(mpd->inode))
2300 disksize = i_size_read(mpd->inode);
2301 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2302 ext4_update_i_disksize(mpd->inode, disksize);
2303 return ext4_mark_inode_dirty(handle, mpd->inode);
2304 }
2305
2306 return 0;
2307 }
2308
2309 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2310 (1 << BH_Delay) | (1 << BH_Unwritten))
2311
2312 /*
2313 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2314 *
2315 * @mpd->lbh - extent of blocks
2316 * @logical - logical number of the block in the file
2317 * @bh - bh of the block (used to access block's state)
2318 *
2319 * the function is used to collect contig. blocks in same state
2320 */
2321 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2322 sector_t logical, size_t b_size,
2323 unsigned long b_state)
2324 {
2325 sector_t next;
2326 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2327
2328 /* check if thereserved journal credits might overflow */
2329 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2330 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2331 /*
2332 * With non-extent format we are limited by the journal
2333 * credit available. Total credit needed to insert
2334 * nrblocks contiguous blocks is dependent on the
2335 * nrblocks. So limit nrblocks.
2336 */
2337 goto flush_it;
2338 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2339 EXT4_MAX_TRANS_DATA) {
2340 /*
2341 * Adding the new buffer_head would make it cross the
2342 * allowed limit for which we have journal credit
2343 * reserved. So limit the new bh->b_size
2344 */
2345 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2346 mpd->inode->i_blkbits;
2347 /* we will do mpage_da_submit_io in the next loop */
2348 }
2349 }
2350 /*
2351 * First block in the extent
2352 */
2353 if (mpd->b_size == 0) {
2354 mpd->b_blocknr = logical;
2355 mpd->b_size = b_size;
2356 mpd->b_state = b_state & BH_FLAGS;
2357 return;
2358 }
2359
2360 next = mpd->b_blocknr + nrblocks;
2361 /*
2362 * Can we merge the block to our big extent?
2363 */
2364 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2365 mpd->b_size += b_size;
2366 return;
2367 }
2368
2369 flush_it:
2370 /*
2371 * We couldn't merge the block to our extent, so we
2372 * need to flush current extent and start new one
2373 */
2374 if (mpage_da_map_blocks(mpd) == 0)
2375 mpage_da_submit_io(mpd);
2376 mpd->io_done = 1;
2377 return;
2378 }
2379
2380 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2381 {
2382 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2383 }
2384
2385 /*
2386 * __mpage_da_writepage - finds extent of pages and blocks
2387 *
2388 * @page: page to consider
2389 * @wbc: not used, we just follow rules
2390 * @data: context
2391 *
2392 * The function finds extents of pages and scan them for all blocks.
2393 */
2394 static int __mpage_da_writepage(struct page *page,
2395 struct writeback_control *wbc, void *data)
2396 {
2397 struct mpage_da_data *mpd = data;
2398 struct inode *inode = mpd->inode;
2399 struct buffer_head *bh, *head;
2400 sector_t logical;
2401
2402 if (mpd->io_done) {
2403 /*
2404 * Rest of the page in the page_vec
2405 * redirty then and skip then. We will
2406 * try to to write them again after
2407 * starting a new transaction
2408 */
2409 redirty_page_for_writepage(wbc, page);
2410 unlock_page(page);
2411 return MPAGE_DA_EXTENT_TAIL;
2412 }
2413 /*
2414 * Can we merge this page to current extent?
2415 */
2416 if (mpd->next_page != page->index) {
2417 /*
2418 * Nope, we can't. So, we map non-allocated blocks
2419 * and start IO on them using writepage()
2420 */
2421 if (mpd->next_page != mpd->first_page) {
2422 if (mpage_da_map_blocks(mpd) == 0)
2423 mpage_da_submit_io(mpd);
2424 /*
2425 * skip rest of the page in the page_vec
2426 */
2427 mpd->io_done = 1;
2428 redirty_page_for_writepage(wbc, page);
2429 unlock_page(page);
2430 return MPAGE_DA_EXTENT_TAIL;
2431 }
2432
2433 /*
2434 * Start next extent of pages ...
2435 */
2436 mpd->first_page = page->index;
2437
2438 /*
2439 * ... and blocks
2440 */
2441 mpd->b_size = 0;
2442 mpd->b_state = 0;
2443 mpd->b_blocknr = 0;
2444 }
2445
2446 mpd->next_page = page->index + 1;
2447 logical = (sector_t) page->index <<
2448 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2449
2450 if (!page_has_buffers(page)) {
2451 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2452 (1 << BH_Dirty) | (1 << BH_Uptodate));
2453 if (mpd->io_done)
2454 return MPAGE_DA_EXTENT_TAIL;
2455 } else {
2456 /*
2457 * Page with regular buffer heads, just add all dirty ones
2458 */
2459 head = page_buffers(page);
2460 bh = head;
2461 do {
2462 BUG_ON(buffer_locked(bh));
2463 /*
2464 * We need to try to allocate
2465 * unmapped blocks in the same page.
2466 * Otherwise we won't make progress
2467 * with the page in ext4_writepage
2468 */
2469 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2470 mpage_add_bh_to_extent(mpd, logical,
2471 bh->b_size,
2472 bh->b_state);
2473 if (mpd->io_done)
2474 return MPAGE_DA_EXTENT_TAIL;
2475 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2476 /*
2477 * mapped dirty buffer. We need to update
2478 * the b_state because we look at
2479 * b_state in mpage_da_map_blocks. We don't
2480 * update b_size because if we find an
2481 * unmapped buffer_head later we need to
2482 * use the b_state flag of that buffer_head.
2483 */
2484 if (mpd->b_size == 0)
2485 mpd->b_state = bh->b_state & BH_FLAGS;
2486 }
2487 logical++;
2488 } while ((bh = bh->b_this_page) != head);
2489 }
2490
2491 return 0;
2492 }
2493
2494 /*
2495 * This is a special get_blocks_t callback which is used by
2496 * ext4_da_write_begin(). It will either return mapped block or
2497 * reserve space for a single block.
2498 *
2499 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2500 * We also have b_blocknr = -1 and b_bdev initialized properly
2501 *
2502 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2503 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2504 * initialized properly.
2505 */
2506 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2507 struct buffer_head *bh_result, int create)
2508 {
2509 int ret = 0;
2510 sector_t invalid_block = ~((sector_t) 0xffff);
2511
2512 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2513 invalid_block = ~0;
2514
2515 BUG_ON(create == 0);
2516 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2517
2518 /*
2519 * first, we need to know whether the block is allocated already
2520 * preallocated blocks are unmapped but should treated
2521 * the same as allocated blocks.
2522 */
2523 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
2524 if ((ret == 0) && !buffer_delay(bh_result)) {
2525 /* the block isn't (pre)allocated yet, let's reserve space */
2526 /*
2527 * XXX: __block_prepare_write() unmaps passed block,
2528 * is it OK?
2529 */
2530 ret = ext4_da_reserve_space(inode, 1);
2531 if (ret)
2532 /* not enough space to reserve */
2533 return ret;
2534
2535 map_bh(bh_result, inode->i_sb, invalid_block);
2536 set_buffer_new(bh_result);
2537 set_buffer_delay(bh_result);
2538 } else if (ret > 0) {
2539 bh_result->b_size = (ret << inode->i_blkbits);
2540 if (buffer_unwritten(bh_result)) {
2541 /* A delayed write to unwritten bh should
2542 * be marked new and mapped. Mapped ensures
2543 * that we don't do get_block multiple times
2544 * when we write to the same offset and new
2545 * ensures that we do proper zero out for
2546 * partial write.
2547 */
2548 set_buffer_new(bh_result);
2549 set_buffer_mapped(bh_result);
2550 }
2551 ret = 0;
2552 }
2553
2554 return ret;
2555 }
2556
2557 /*
2558 * This function is used as a standard get_block_t calback function
2559 * when there is no desire to allocate any blocks. It is used as a
2560 * callback function for block_prepare_write(), nobh_writepage(), and
2561 * block_write_full_page(). These functions should only try to map a
2562 * single block at a time.
2563 *
2564 * Since this function doesn't do block allocations even if the caller
2565 * requests it by passing in create=1, it is critically important that
2566 * any caller checks to make sure that any buffer heads are returned
2567 * by this function are either all already mapped or marked for
2568 * delayed allocation before calling nobh_writepage() or
2569 * block_write_full_page(). Otherwise, b_blocknr could be left
2570 * unitialized, and the page write functions will be taken by
2571 * surprise.
2572 */
2573 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2574 struct buffer_head *bh_result, int create)
2575 {
2576 int ret = 0;
2577 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2578
2579 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2580
2581 /*
2582 * we don't want to do block allocation in writepage
2583 * so call get_block_wrap with create = 0
2584 */
2585 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2586 if (ret > 0) {
2587 bh_result->b_size = (ret << inode->i_blkbits);
2588 ret = 0;
2589 }
2590 return ret;
2591 }
2592
2593 static int bget_one(handle_t *handle, struct buffer_head *bh)
2594 {
2595 get_bh(bh);
2596 return 0;
2597 }
2598
2599 static int bput_one(handle_t *handle, struct buffer_head *bh)
2600 {
2601 put_bh(bh);
2602 return 0;
2603 }
2604
2605 static int __ext4_journalled_writepage(struct page *page,
2606 struct writeback_control *wbc,
2607 unsigned int len)
2608 {
2609 struct address_space *mapping = page->mapping;
2610 struct inode *inode = mapping->host;
2611 struct buffer_head *page_bufs;
2612 handle_t *handle = NULL;
2613 int ret = 0;
2614 int err;
2615
2616 page_bufs = page_buffers(page);
2617 BUG_ON(!page_bufs);
2618 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2619 /* As soon as we unlock the page, it can go away, but we have
2620 * references to buffers so we are safe */
2621 unlock_page(page);
2622
2623 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2624 if (IS_ERR(handle)) {
2625 ret = PTR_ERR(handle);
2626 goto out;
2627 }
2628
2629 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2630 do_journal_get_write_access);
2631
2632 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2633 write_end_fn);
2634 if (ret == 0)
2635 ret = err;
2636 err = ext4_journal_stop(handle);
2637 if (!ret)
2638 ret = err;
2639
2640 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2641 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2642 out:
2643 return ret;
2644 }
2645
2646 /*
2647 * Note that we don't need to start a transaction unless we're journaling data
2648 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2649 * need to file the inode to the transaction's list in ordered mode because if
2650 * we are writing back data added by write(), the inode is already there and if
2651 * we are writing back data modified via mmap(), noone guarantees in which
2652 * transaction the data will hit the disk. In case we are journaling data, we
2653 * cannot start transaction directly because transaction start ranks above page
2654 * lock so we have to do some magic.
2655 *
2656 * This function can get called via...
2657 * - ext4_da_writepages after taking page lock (have journal handle)
2658 * - journal_submit_inode_data_buffers (no journal handle)
2659 * - shrink_page_list via pdflush (no journal handle)
2660 * - grab_page_cache when doing write_begin (have journal handle)
2661 *
2662 * We don't do any block allocation in this function. If we have page with
2663 * multiple blocks we need to write those buffer_heads that are mapped. This
2664 * is important for mmaped based write. So if we do with blocksize 1K
2665 * truncate(f, 1024);
2666 * a = mmap(f, 0, 4096);
2667 * a[0] = 'a';
2668 * truncate(f, 4096);
2669 * we have in the page first buffer_head mapped via page_mkwrite call back
2670 * but other bufer_heads would be unmapped but dirty(dirty done via the
2671 * do_wp_page). So writepage should write the first block. If we modify
2672 * the mmap area beyond 1024 we will again get a page_fault and the
2673 * page_mkwrite callback will do the block allocation and mark the
2674 * buffer_heads mapped.
2675 *
2676 * We redirty the page if we have any buffer_heads that is either delay or
2677 * unwritten in the page.
2678 *
2679 * We can get recursively called as show below.
2680 *
2681 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2682 * ext4_writepage()
2683 *
2684 * But since we don't do any block allocation we should not deadlock.
2685 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2686 */
2687 static int ext4_writepage(struct page *page,
2688 struct writeback_control *wbc)
2689 {
2690 int ret = 0;
2691 loff_t size;
2692 unsigned int len;
2693 struct buffer_head *page_bufs;
2694 struct inode *inode = page->mapping->host;
2695
2696 trace_ext4_writepage(inode, page);
2697 size = i_size_read(inode);
2698 if (page->index == size >> PAGE_CACHE_SHIFT)
2699 len = size & ~PAGE_CACHE_MASK;
2700 else
2701 len = PAGE_CACHE_SIZE;
2702
2703 if (page_has_buffers(page)) {
2704 page_bufs = page_buffers(page);
2705 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2706 ext4_bh_delay_or_unwritten)) {
2707 /*
2708 * We don't want to do block allocation
2709 * So redirty the page and return
2710 * We may reach here when we do a journal commit
2711 * via journal_submit_inode_data_buffers.
2712 * If we don't have mapping block we just ignore
2713 * them. We can also reach here via shrink_page_list
2714 */
2715 redirty_page_for_writepage(wbc, page);
2716 unlock_page(page);
2717 return 0;
2718 }
2719 } else {
2720 /*
2721 * The test for page_has_buffers() is subtle:
2722 * We know the page is dirty but it lost buffers. That means
2723 * that at some moment in time after write_begin()/write_end()
2724 * has been called all buffers have been clean and thus they
2725 * must have been written at least once. So they are all
2726 * mapped and we can happily proceed with mapping them
2727 * and writing the page.
2728 *
2729 * Try to initialize the buffer_heads and check whether
2730 * all are mapped and non delay. We don't want to
2731 * do block allocation here.
2732 */
2733 ret = block_prepare_write(page, 0, len,
2734 noalloc_get_block_write);
2735 if (!ret) {
2736 page_bufs = page_buffers(page);
2737 /* check whether all are mapped and non delay */
2738 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2739 ext4_bh_delay_or_unwritten)) {
2740 redirty_page_for_writepage(wbc, page);
2741 unlock_page(page);
2742 return 0;
2743 }
2744 } else {
2745 /*
2746 * We can't do block allocation here
2747 * so just redity the page and unlock
2748 * and return
2749 */
2750 redirty_page_for_writepage(wbc, page);
2751 unlock_page(page);
2752 return 0;
2753 }
2754 /* now mark the buffer_heads as dirty and uptodate */
2755 block_commit_write(page, 0, len);
2756 }
2757
2758 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2759 /*
2760 * It's mmapped pagecache. Add buffers and journal it. There
2761 * doesn't seem much point in redirtying the page here.
2762 */
2763 ClearPageChecked(page);
2764 return __ext4_journalled_writepage(page, wbc, len);
2765 }
2766
2767 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2768 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
2769 else
2770 ret = block_write_full_page(page, noalloc_get_block_write,
2771 wbc);
2772
2773 return ret;
2774 }
2775
2776 /*
2777 * This is called via ext4_da_writepages() to
2778 * calulate the total number of credits to reserve to fit
2779 * a single extent allocation into a single transaction,
2780 * ext4_da_writpeages() will loop calling this before
2781 * the block allocation.
2782 */
2783
2784 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2785 {
2786 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2787
2788 /*
2789 * With non-extent format the journal credit needed to
2790 * insert nrblocks contiguous block is dependent on
2791 * number of contiguous block. So we will limit
2792 * number of contiguous block to a sane value
2793 */
2794 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) &&
2795 (max_blocks > EXT4_MAX_TRANS_DATA))
2796 max_blocks = EXT4_MAX_TRANS_DATA;
2797
2798 return ext4_chunk_trans_blocks(inode, max_blocks);
2799 }
2800
2801 static int ext4_da_writepages(struct address_space *mapping,
2802 struct writeback_control *wbc)
2803 {
2804 pgoff_t index;
2805 int range_whole = 0;
2806 handle_t *handle = NULL;
2807 struct mpage_da_data mpd;
2808 struct inode *inode = mapping->host;
2809 int no_nrwrite_index_update;
2810 int pages_written = 0;
2811 long pages_skipped;
2812 unsigned int max_pages;
2813 int range_cyclic, cycled = 1, io_done = 0;
2814 int needed_blocks, ret = 0;
2815 long desired_nr_to_write, nr_to_writebump = 0;
2816 loff_t range_start = wbc->range_start;
2817 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2818
2819 trace_ext4_da_writepages(inode, wbc);
2820
2821 /*
2822 * No pages to write? This is mainly a kludge to avoid starting
2823 * a transaction for special inodes like journal inode on last iput()
2824 * because that could violate lock ordering on umount
2825 */
2826 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2827 return 0;
2828
2829 /*
2830 * If the filesystem has aborted, it is read-only, so return
2831 * right away instead of dumping stack traces later on that
2832 * will obscure the real source of the problem. We test
2833 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2834 * the latter could be true if the filesystem is mounted
2835 * read-only, and in that case, ext4_da_writepages should
2836 * *never* be called, so if that ever happens, we would want
2837 * the stack trace.
2838 */
2839 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2840 return -EROFS;
2841
2842 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2843 range_whole = 1;
2844
2845 range_cyclic = wbc->range_cyclic;
2846 if (wbc->range_cyclic) {
2847 index = mapping->writeback_index;
2848 if (index)
2849 cycled = 0;
2850 wbc->range_start = index << PAGE_CACHE_SHIFT;
2851 wbc->range_end = LLONG_MAX;
2852 wbc->range_cyclic = 0;
2853 } else
2854 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2855
2856 /*
2857 * This works around two forms of stupidity. The first is in
2858 * the writeback code, which caps the maximum number of pages
2859 * written to be 1024 pages. This is wrong on multiple
2860 * levels; different architectues have a different page size,
2861 * which changes the maximum amount of data which gets
2862 * written. Secondly, 4 megabytes is way too small. XFS
2863 * forces this value to be 16 megabytes by multiplying
2864 * nr_to_write parameter by four, and then relies on its
2865 * allocator to allocate larger extents to make them
2866 * contiguous. Unfortunately this brings us to the second
2867 * stupidity, which is that ext4's mballoc code only allocates
2868 * at most 2048 blocks. So we force contiguous writes up to
2869 * the number of dirty blocks in the inode, or
2870 * sbi->max_writeback_mb_bump whichever is smaller.
2871 */
2872 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2873 if (!range_cyclic && range_whole)
2874 desired_nr_to_write = wbc->nr_to_write * 8;
2875 else
2876 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2877 max_pages);
2878 if (desired_nr_to_write > max_pages)
2879 desired_nr_to_write = max_pages;
2880
2881 if (wbc->nr_to_write < desired_nr_to_write) {
2882 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2883 wbc->nr_to_write = desired_nr_to_write;
2884 }
2885
2886 mpd.wbc = wbc;
2887 mpd.inode = mapping->host;
2888
2889 /*
2890 * we don't want write_cache_pages to update
2891 * nr_to_write and writeback_index
2892 */
2893 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2894 wbc->no_nrwrite_index_update = 1;
2895 pages_skipped = wbc->pages_skipped;
2896
2897 retry:
2898 while (!ret && wbc->nr_to_write > 0) {
2899
2900 /*
2901 * we insert one extent at a time. So we need
2902 * credit needed for single extent allocation.
2903 * journalled mode is currently not supported
2904 * by delalloc
2905 */
2906 BUG_ON(ext4_should_journal_data(inode));
2907 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2908
2909 /* start a new transaction*/
2910 handle = ext4_journal_start(inode, needed_blocks);
2911 if (IS_ERR(handle)) {
2912 ret = PTR_ERR(handle);
2913 printk(KERN_CRIT "%s: jbd2_start: "
2914 "%ld pages, ino %lu; err %d\n", __func__,
2915 wbc->nr_to_write, inode->i_ino, ret);
2916 dump_stack();
2917 goto out_writepages;
2918 }
2919
2920 /*
2921 * Now call __mpage_da_writepage to find the next
2922 * contiguous region of logical blocks that need
2923 * blocks to be allocated by ext4. We don't actually
2924 * submit the blocks for I/O here, even though
2925 * write_cache_pages thinks it will, and will set the
2926 * pages as clean for write before calling
2927 * __mpage_da_writepage().
2928 */
2929 mpd.b_size = 0;
2930 mpd.b_state = 0;
2931 mpd.b_blocknr = 0;
2932 mpd.first_page = 0;
2933 mpd.next_page = 0;
2934 mpd.io_done = 0;
2935 mpd.pages_written = 0;
2936 mpd.retval = 0;
2937 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2938 &mpd);
2939 /*
2940 * If we have a contigous extent of pages and we
2941 * haven't done the I/O yet, map the blocks and submit
2942 * them for I/O.
2943 */
2944 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2945 if (mpage_da_map_blocks(&mpd) == 0)
2946 mpage_da_submit_io(&mpd);
2947 mpd.io_done = 1;
2948 ret = MPAGE_DA_EXTENT_TAIL;
2949 }
2950 wbc->nr_to_write -= mpd.pages_written;
2951
2952 ext4_journal_stop(handle);
2953
2954 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2955 /* commit the transaction which would
2956 * free blocks released in the transaction
2957 * and try again
2958 */
2959 jbd2_journal_force_commit_nested(sbi->s_journal);
2960 wbc->pages_skipped = pages_skipped;
2961 ret = 0;
2962 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2963 /*
2964 * got one extent now try with
2965 * rest of the pages
2966 */
2967 pages_written += mpd.pages_written;
2968 wbc->pages_skipped = pages_skipped;
2969 ret = 0;
2970 io_done = 1;
2971 } else if (wbc->nr_to_write)
2972 /*
2973 * There is no more writeout needed
2974 * or we requested for a noblocking writeout
2975 * and we found the device congested
2976 */
2977 break;
2978 }
2979 if (!io_done && !cycled) {
2980 cycled = 1;
2981 index = 0;
2982 wbc->range_start = index << PAGE_CACHE_SHIFT;
2983 wbc->range_end = mapping->writeback_index - 1;
2984 goto retry;
2985 }
2986 if (pages_skipped != wbc->pages_skipped)
2987 printk(KERN_EMERG "This should not happen leaving %s "
2988 "with nr_to_write = %ld ret = %d\n",
2989 __func__, wbc->nr_to_write, ret);
2990
2991 /* Update index */
2992 index += pages_written;
2993 wbc->range_cyclic = range_cyclic;
2994 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2995 /*
2996 * set the writeback_index so that range_cyclic
2997 * mode will write it back later
2998 */
2999 mapping->writeback_index = index;
3000
3001 out_writepages:
3002 if (!no_nrwrite_index_update)
3003 wbc->no_nrwrite_index_update = 0;
3004 if (wbc->nr_to_write > nr_to_writebump)
3005 wbc->nr_to_write -= nr_to_writebump;
3006 wbc->range_start = range_start;
3007 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3008 return ret;
3009 }
3010
3011 #define FALL_BACK_TO_NONDELALLOC 1
3012 static int ext4_nonda_switch(struct super_block *sb)
3013 {
3014 s64 free_blocks, dirty_blocks;
3015 struct ext4_sb_info *sbi = EXT4_SB(sb);
3016
3017 /*
3018 * switch to non delalloc mode if we are running low
3019 * on free block. The free block accounting via percpu
3020 * counters can get slightly wrong with percpu_counter_batch getting
3021 * accumulated on each CPU without updating global counters
3022 * Delalloc need an accurate free block accounting. So switch
3023 * to non delalloc when we are near to error range.
3024 */
3025 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3026 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3027 if (2 * free_blocks < 3 * dirty_blocks ||
3028 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3029 /*
3030 * free block count is less that 150% of dirty blocks
3031 * or free blocks is less that watermark
3032 */
3033 return 1;
3034 }
3035 return 0;
3036 }
3037
3038 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3039 loff_t pos, unsigned len, unsigned flags,
3040 struct page **pagep, void **fsdata)
3041 {
3042 int ret, retries = 0;
3043 struct page *page;
3044 pgoff_t index;
3045 unsigned from, to;
3046 struct inode *inode = mapping->host;
3047 handle_t *handle;
3048
3049 index = pos >> PAGE_CACHE_SHIFT;
3050 from = pos & (PAGE_CACHE_SIZE - 1);
3051 to = from + len;
3052
3053 if (ext4_nonda_switch(inode->i_sb)) {
3054 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3055 return ext4_write_begin(file, mapping, pos,
3056 len, flags, pagep, fsdata);
3057 }
3058 *fsdata = (void *)0;
3059 trace_ext4_da_write_begin(inode, pos, len, flags);
3060 retry:
3061 /*
3062 * With delayed allocation, we don't log the i_disksize update
3063 * if there is delayed block allocation. But we still need
3064 * to journalling the i_disksize update if writes to the end
3065 * of file which has an already mapped buffer.
3066 */
3067 handle = ext4_journal_start(inode, 1);
3068 if (IS_ERR(handle)) {
3069 ret = PTR_ERR(handle);
3070 goto out;
3071 }
3072 /* We cannot recurse into the filesystem as the transaction is already
3073 * started */
3074 flags |= AOP_FLAG_NOFS;
3075
3076 page = grab_cache_page_write_begin(mapping, index, flags);
3077 if (!page) {
3078 ext4_journal_stop(handle);
3079 ret = -ENOMEM;
3080 goto out;
3081 }
3082 *pagep = page;
3083
3084 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
3085 ext4_da_get_block_prep);
3086 if (ret < 0) {
3087 unlock_page(page);
3088 ext4_journal_stop(handle);
3089 page_cache_release(page);
3090 /*
3091 * block_write_begin may have instantiated a few blocks
3092 * outside i_size. Trim these off again. Don't need
3093 * i_size_read because we hold i_mutex.
3094 */
3095 if (pos + len > inode->i_size)
3096 ext4_truncate_failed_write(inode);
3097 }
3098
3099 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3100 goto retry;
3101 out:
3102 return ret;
3103 }
3104
3105 /*
3106 * Check if we should update i_disksize
3107 * when write to the end of file but not require block allocation
3108 */
3109 static int ext4_da_should_update_i_disksize(struct page *page,
3110 unsigned long offset)
3111 {
3112 struct buffer_head *bh;
3113 struct inode *inode = page->mapping->host;
3114 unsigned int idx;
3115 int i;
3116
3117 bh = page_buffers(page);
3118 idx = offset >> inode->i_blkbits;
3119
3120 for (i = 0; i < idx; i++)
3121 bh = bh->b_this_page;
3122
3123 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3124 return 0;
3125 return 1;
3126 }
3127
3128 static int ext4_da_write_end(struct file *file,
3129 struct address_space *mapping,
3130 loff_t pos, unsigned len, unsigned copied,
3131 struct page *page, void *fsdata)
3132 {
3133 struct inode *inode = mapping->host;
3134 int ret = 0, ret2;
3135 handle_t *handle = ext4_journal_current_handle();
3136 loff_t new_i_size;
3137 unsigned long start, end;
3138 int write_mode = (int)(unsigned long)fsdata;
3139
3140 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3141 if (ext4_should_order_data(inode)) {
3142 return ext4_ordered_write_end(file, mapping, pos,
3143 len, copied, page, fsdata);
3144 } else if (ext4_should_writeback_data(inode)) {
3145 return ext4_writeback_write_end(file, mapping, pos,
3146 len, copied, page, fsdata);
3147 } else {
3148 BUG();
3149 }
3150 }
3151
3152 trace_ext4_da_write_end(inode, pos, len, copied);
3153 start = pos & (PAGE_CACHE_SIZE - 1);
3154 end = start + copied - 1;
3155
3156 /*
3157 * generic_write_end() will run mark_inode_dirty() if i_size
3158 * changes. So let's piggyback the i_disksize mark_inode_dirty
3159 * into that.
3160 */
3161
3162 new_i_size = pos + copied;
3163 if (new_i_size > EXT4_I(inode)->i_disksize) {
3164 if (ext4_da_should_update_i_disksize(page, end)) {
3165 down_write(&EXT4_I(inode)->i_data_sem);
3166 if (new_i_size > EXT4_I(inode)->i_disksize) {
3167 /*
3168 * Updating i_disksize when extending file
3169 * without needing block allocation
3170 */
3171 if (ext4_should_order_data(inode))
3172 ret = ext4_jbd2_file_inode(handle,
3173 inode);
3174
3175 EXT4_I(inode)->i_disksize = new_i_size;
3176 }
3177 up_write(&EXT4_I(inode)->i_data_sem);
3178 /* We need to mark inode dirty even if
3179 * new_i_size is less that inode->i_size
3180 * bu greater than i_disksize.(hint delalloc)
3181 */
3182 ext4_mark_inode_dirty(handle, inode);
3183 }
3184 }
3185 ret2 = generic_write_end(file, mapping, pos, len, copied,
3186 page, fsdata);
3187 copied = ret2;
3188 if (ret2 < 0)
3189 ret = ret2;
3190 ret2 = ext4_journal_stop(handle);
3191 if (!ret)
3192 ret = ret2;
3193
3194 return ret ? ret : copied;
3195 }
3196
3197 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3198 {
3199 /*
3200 * Drop reserved blocks
3201 */
3202 BUG_ON(!PageLocked(page));
3203 if (!page_has_buffers(page))
3204 goto out;
3205
3206 ext4_da_page_release_reservation(page, offset);
3207
3208 out:
3209 ext4_invalidatepage(page, offset);
3210
3211 return;
3212 }
3213
3214 /*
3215 * Force all delayed allocation blocks to be allocated for a given inode.
3216 */
3217 int ext4_alloc_da_blocks(struct inode *inode)
3218 {
3219 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3220 !EXT4_I(inode)->i_reserved_meta_blocks)
3221 return 0;
3222
3223 /*
3224 * We do something simple for now. The filemap_flush() will
3225 * also start triggering a write of the data blocks, which is
3226 * not strictly speaking necessary (and for users of
3227 * laptop_mode, not even desirable). However, to do otherwise
3228 * would require replicating code paths in:
3229 *
3230 * ext4_da_writepages() ->
3231 * write_cache_pages() ---> (via passed in callback function)
3232 * __mpage_da_writepage() -->
3233 * mpage_add_bh_to_extent()
3234 * mpage_da_map_blocks()
3235 *
3236 * The problem is that write_cache_pages(), located in
3237 * mm/page-writeback.c, marks pages clean in preparation for
3238 * doing I/O, which is not desirable if we're not planning on
3239 * doing I/O at all.
3240 *
3241 * We could call write_cache_pages(), and then redirty all of
3242 * the pages by calling redirty_page_for_writeback() but that
3243 * would be ugly in the extreme. So instead we would need to
3244 * replicate parts of the code in the above functions,
3245 * simplifying them becuase we wouldn't actually intend to
3246 * write out the pages, but rather only collect contiguous
3247 * logical block extents, call the multi-block allocator, and
3248 * then update the buffer heads with the block allocations.
3249 *
3250 * For now, though, we'll cheat by calling filemap_flush(),
3251 * which will map the blocks, and start the I/O, but not
3252 * actually wait for the I/O to complete.
3253 */
3254 return filemap_flush(inode->i_mapping);
3255 }
3256
3257 /*
3258 * bmap() is special. It gets used by applications such as lilo and by
3259 * the swapper to find the on-disk block of a specific piece of data.
3260 *
3261 * Naturally, this is dangerous if the block concerned is still in the
3262 * journal. If somebody makes a swapfile on an ext4 data-journaling
3263 * filesystem and enables swap, then they may get a nasty shock when the
3264 * data getting swapped to that swapfile suddenly gets overwritten by
3265 * the original zero's written out previously to the journal and
3266 * awaiting writeback in the kernel's buffer cache.
3267 *
3268 * So, if we see any bmap calls here on a modified, data-journaled file,
3269 * take extra steps to flush any blocks which might be in the cache.
3270 */
3271 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3272 {
3273 struct inode *inode = mapping->host;
3274 journal_t *journal;
3275 int err;
3276
3277 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3278 test_opt(inode->i_sb, DELALLOC)) {
3279 /*
3280 * With delalloc we want to sync the file
3281 * so that we can make sure we allocate
3282 * blocks for file
3283 */
3284 filemap_write_and_wait(mapping);
3285 }
3286
3287 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
3288 /*
3289 * This is a REALLY heavyweight approach, but the use of
3290 * bmap on dirty files is expected to be extremely rare:
3291 * only if we run lilo or swapon on a freshly made file
3292 * do we expect this to happen.
3293 *
3294 * (bmap requires CAP_SYS_RAWIO so this does not
3295 * represent an unprivileged user DOS attack --- we'd be
3296 * in trouble if mortal users could trigger this path at
3297 * will.)
3298 *
3299 * NB. EXT4_STATE_JDATA is not set on files other than
3300 * regular files. If somebody wants to bmap a directory
3301 * or symlink and gets confused because the buffer
3302 * hasn't yet been flushed to disk, they deserve
3303 * everything they get.
3304 */
3305
3306 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
3307 journal = EXT4_JOURNAL(inode);
3308 jbd2_journal_lock_updates(journal);
3309 err = jbd2_journal_flush(journal);
3310 jbd2_journal_unlock_updates(journal);
3311
3312 if (err)
3313 return 0;
3314 }
3315
3316 return generic_block_bmap(mapping, block, ext4_get_block);
3317 }
3318
3319 static int ext4_readpage(struct file *file, struct page *page)
3320 {
3321 return mpage_readpage(page, ext4_get_block);
3322 }
3323
3324 static int
3325 ext4_readpages(struct file *file, struct address_space *mapping,
3326 struct list_head *pages, unsigned nr_pages)
3327 {
3328 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3329 }
3330
3331 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3332 {
3333 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3334
3335 /*
3336 * If it's a full truncate we just forget about the pending dirtying
3337 */
3338 if (offset == 0)
3339 ClearPageChecked(page);
3340
3341 if (journal)
3342 jbd2_journal_invalidatepage(journal, page, offset);
3343 else
3344 block_invalidatepage(page, offset);
3345 }
3346
3347 static int ext4_releasepage(struct page *page, gfp_t wait)
3348 {
3349 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3350
3351 WARN_ON(PageChecked(page));
3352 if (!page_has_buffers(page))
3353 return 0;
3354 if (journal)
3355 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3356 else
3357 return try_to_free_buffers(page);
3358 }
3359
3360 /*
3361 * O_DIRECT for ext3 (or indirect map) based files
3362 *
3363 * If the O_DIRECT write will extend the file then add this inode to the
3364 * orphan list. So recovery will truncate it back to the original size
3365 * if the machine crashes during the write.
3366 *
3367 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3368 * crashes then stale disk data _may_ be exposed inside the file. But current
3369 * VFS code falls back into buffered path in that case so we are safe.
3370 */
3371 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3372 const struct iovec *iov, loff_t offset,
3373 unsigned long nr_segs)
3374 {
3375 struct file *file = iocb->ki_filp;
3376 struct inode *inode = file->f_mapping->host;
3377 struct ext4_inode_info *ei = EXT4_I(inode);
3378 handle_t *handle;
3379 ssize_t ret;
3380 int orphan = 0;
3381 size_t count = iov_length(iov, nr_segs);
3382 int retries = 0;
3383
3384 if (rw == WRITE) {
3385 loff_t final_size = offset + count;
3386
3387 if (final_size > inode->i_size) {
3388 /* Credits for sb + inode write */
3389 handle = ext4_journal_start(inode, 2);
3390 if (IS_ERR(handle)) {
3391 ret = PTR_ERR(handle);
3392 goto out;
3393 }
3394 ret = ext4_orphan_add(handle, inode);
3395 if (ret) {
3396 ext4_journal_stop(handle);
3397 goto out;
3398 }
3399 orphan = 1;
3400 ei->i_disksize = inode->i_size;
3401 ext4_journal_stop(handle);
3402 }
3403 }
3404
3405 retry:
3406 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3407 offset, nr_segs,
3408 ext4_get_block, NULL);
3409 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3410 goto retry;
3411
3412 if (orphan) {
3413 int err;
3414
3415 /* Credits for sb + inode write */
3416 handle = ext4_journal_start(inode, 2);
3417 if (IS_ERR(handle)) {
3418 /* This is really bad luck. We've written the data
3419 * but cannot extend i_size. Bail out and pretend
3420 * the write failed... */
3421 ret = PTR_ERR(handle);
3422 goto out;
3423 }
3424 if (inode->i_nlink)
3425 ext4_orphan_del(handle, inode);
3426 if (ret > 0) {
3427 loff_t end = offset + ret;
3428 if (end > inode->i_size) {
3429 ei->i_disksize = end;
3430 i_size_write(inode, end);
3431 /*
3432 * We're going to return a positive `ret'
3433 * here due to non-zero-length I/O, so there's
3434 * no way of reporting error returns from
3435 * ext4_mark_inode_dirty() to userspace. So
3436 * ignore it.
3437 */
3438 ext4_mark_inode_dirty(handle, inode);
3439 }
3440 }
3441 err = ext4_journal_stop(handle);
3442 if (ret == 0)
3443 ret = err;
3444 }
3445 out:
3446 return ret;
3447 }
3448
3449 static int ext4_get_block_dio_write(struct inode *inode, sector_t iblock,
3450 struct buffer_head *bh_result, int create)
3451 {
3452 handle_t *handle = NULL;
3453 int ret = 0;
3454 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
3455 int dio_credits;
3456
3457 ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
3458 inode->i_ino, create);
3459 /*
3460 * DIO VFS code passes create = 0 flag for write to
3461 * the middle of file. It does this to avoid block
3462 * allocation for holes, to prevent expose stale data
3463 * out when there is parallel buffered read (which does
3464 * not hold the i_mutex lock) while direct IO write has
3465 * not completed. DIO request on holes finally falls back
3466 * to buffered IO for this reason.
3467 *
3468 * For ext4 extent based file, since we support fallocate,
3469 * new allocated extent as uninitialized, for holes, we
3470 * could fallocate blocks for holes, thus parallel
3471 * buffered IO read will zero out the page when read on
3472 * a hole while parallel DIO write to the hole has not completed.
3473 *
3474 * when we come here, we know it's a direct IO write to
3475 * to the middle of file (<i_size)
3476 * so it's safe to override the create flag from VFS.
3477 */
3478 create = EXT4_GET_BLOCKS_DIO_CREATE_EXT;
3479
3480 if (max_blocks > DIO_MAX_BLOCKS)
3481 max_blocks = DIO_MAX_BLOCKS;
3482 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
3483 handle = ext4_journal_start(inode, dio_credits);
3484 if (IS_ERR(handle)) {
3485 ret = PTR_ERR(handle);
3486 goto out;
3487 }
3488 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
3489 create);
3490 if (ret > 0) {
3491 bh_result->b_size = (ret << inode->i_blkbits);
3492 ret = 0;
3493 }
3494 ext4_journal_stop(handle);
3495 out:
3496 return ret;
3497 }
3498
3499 static void ext4_free_io_end(ext4_io_end_t *io)
3500 {
3501 BUG_ON(!io);
3502 iput(io->inode);
3503 kfree(io);
3504 }
3505 static void dump_aio_dio_list(struct inode * inode)
3506 {
3507 #ifdef EXT4_DEBUG
3508 struct list_head *cur, *before, *after;
3509 ext4_io_end_t *io, *io0, *io1;
3510
3511 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3512 ext4_debug("inode %lu aio dio list is empty\n", inode->i_ino);
3513 return;
3514 }
3515
3516 ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode->i_ino);
3517 list_for_each_entry(io, &EXT4_I(inode)->i_aio_dio_complete_list, list){
3518 cur = &io->list;
3519 before = cur->prev;
3520 io0 = container_of(before, ext4_io_end_t, list);
3521 after = cur->next;
3522 io1 = container_of(after, ext4_io_end_t, list);
3523
3524 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3525 io, inode->i_ino, io0, io1);
3526 }
3527 #endif
3528 }
3529
3530 /*
3531 * check a range of space and convert unwritten extents to written.
3532 */
3533 static int ext4_end_aio_dio_nolock(ext4_io_end_t *io)
3534 {
3535 struct inode *inode = io->inode;
3536 loff_t offset = io->offset;
3537 size_t size = io->size;
3538 int ret = 0;
3539
3540 ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
3541 "list->prev 0x%p\n",
3542 io, inode->i_ino, io->list.next, io->list.prev);
3543
3544 if (list_empty(&io->list))
3545 return ret;
3546
3547 if (io->flag != DIO_AIO_UNWRITTEN)
3548 return ret;
3549
3550 if (offset + size <= i_size_read(inode))
3551 ret = ext4_convert_unwritten_extents(inode, offset, size);
3552
3553 if (ret < 0) {
3554 printk(KERN_EMERG "%s: failed to convert unwritten"
3555 "extents to written extents, error is %d"
3556 " io is still on inode %lu aio dio list\n",
3557 __func__, ret, inode->i_ino);
3558 return ret;
3559 }
3560
3561 /* clear the DIO AIO unwritten flag */
3562 io->flag = 0;
3563 return ret;
3564 }
3565 /*
3566 * work on completed aio dio IO, to convert unwritten extents to extents
3567 */
3568 static void ext4_end_aio_dio_work(struct work_struct *work)
3569 {
3570 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3571 struct inode *inode = io->inode;
3572 int ret = 0;
3573
3574 mutex_lock(&inode->i_mutex);
3575 ret = ext4_end_aio_dio_nolock(io);
3576 if (ret >= 0) {
3577 if (!list_empty(&io->list))
3578 list_del_init(&io->list);
3579 ext4_free_io_end(io);
3580 }
3581 mutex_unlock(&inode->i_mutex);
3582 }
3583 /*
3584 * This function is called from ext4_sync_file().
3585 *
3586 * When AIO DIO IO is completed, the work to convert unwritten
3587 * extents to written is queued on workqueue but may not get immediately
3588 * scheduled. When fsync is called, we need to ensure the
3589 * conversion is complete before fsync returns.
3590 * The inode keeps track of a list of completed AIO from DIO path
3591 * that might needs to do the conversion. This function walks through
3592 * the list and convert the related unwritten extents to written.
3593 */
3594 int flush_aio_dio_completed_IO(struct inode *inode)
3595 {
3596 ext4_io_end_t *io;
3597 int ret = 0;
3598 int ret2 = 0;
3599
3600 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list))
3601 return ret;
3602
3603 dump_aio_dio_list(inode);
3604 while (!list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3605 io = list_entry(EXT4_I(inode)->i_aio_dio_complete_list.next,
3606 ext4_io_end_t, list);
3607 /*
3608 * Calling ext4_end_aio_dio_nolock() to convert completed
3609 * IO to written.
3610 *
3611 * When ext4_sync_file() is called, run_queue() may already
3612 * about to flush the work corresponding to this io structure.
3613 * It will be upset if it founds the io structure related
3614 * to the work-to-be schedule is freed.
3615 *
3616 * Thus we need to keep the io structure still valid here after
3617 * convertion finished. The io structure has a flag to
3618 * avoid double converting from both fsync and background work
3619 * queue work.
3620 */
3621 ret = ext4_end_aio_dio_nolock(io);
3622 if (ret < 0)
3623 ret2 = ret;
3624 else
3625 list_del_init(&io->list);
3626 }
3627 return (ret2 < 0) ? ret2 : 0;
3628 }
3629
3630 static ext4_io_end_t *ext4_init_io_end (struct inode *inode)
3631 {
3632 ext4_io_end_t *io = NULL;
3633
3634 io = kmalloc(sizeof(*io), GFP_NOFS);
3635
3636 if (io) {
3637 igrab(inode);
3638 io->inode = inode;
3639 io->flag = 0;
3640 io->offset = 0;
3641 io->size = 0;
3642 io->error = 0;
3643 INIT_WORK(&io->work, ext4_end_aio_dio_work);
3644 INIT_LIST_HEAD(&io->list);
3645 }
3646
3647 return io;
3648 }
3649
3650 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3651 ssize_t size, void *private)
3652 {
3653 ext4_io_end_t *io_end = iocb->private;
3654 struct workqueue_struct *wq;
3655
3656 /* if not async direct IO or dio with 0 bytes write, just return */
3657 if (!io_end || !size)
3658 return;
3659
3660 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3661 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3662 iocb->private, io_end->inode->i_ino, iocb, offset,
3663 size);
3664
3665 /* if not aio dio with unwritten extents, just free io and return */
3666 if (io_end->flag != DIO_AIO_UNWRITTEN){
3667 ext4_free_io_end(io_end);
3668 iocb->private = NULL;
3669 return;
3670 }
3671
3672 io_end->offset = offset;
3673 io_end->size = size;
3674 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3675
3676 /* queue the work to convert unwritten extents to written */
3677 queue_work(wq, &io_end->work);
3678
3679 /* Add the io_end to per-inode completed aio dio list*/
3680 list_add_tail(&io_end->list,
3681 &EXT4_I(io_end->inode)->i_aio_dio_complete_list);
3682 iocb->private = NULL;
3683 }
3684 /*
3685 * For ext4 extent files, ext4 will do direct-io write to holes,
3686 * preallocated extents, and those write extend the file, no need to
3687 * fall back to buffered IO.
3688 *
3689 * For holes, we fallocate those blocks, mark them as unintialized
3690 * If those blocks were preallocated, we mark sure they are splited, but
3691 * still keep the range to write as unintialized.
3692 *
3693 * The unwrritten extents will be converted to written when DIO is completed.
3694 * For async direct IO, since the IO may still pending when return, we
3695 * set up an end_io call back function, which will do the convertion
3696 * when async direct IO completed.
3697 *
3698 * If the O_DIRECT write will extend the file then add this inode to the
3699 * orphan list. So recovery will truncate it back to the original size
3700 * if the machine crashes during the write.
3701 *
3702 */
3703 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3704 const struct iovec *iov, loff_t offset,
3705 unsigned long nr_segs)
3706 {
3707 struct file *file = iocb->ki_filp;
3708 struct inode *inode = file->f_mapping->host;
3709 ssize_t ret;
3710 size_t count = iov_length(iov, nr_segs);
3711
3712 loff_t final_size = offset + count;
3713 if (rw == WRITE && final_size <= inode->i_size) {
3714 /*
3715 * We could direct write to holes and fallocate.
3716 *
3717 * Allocated blocks to fill the hole are marked as uninitialized
3718 * to prevent paralel buffered read to expose the stale data
3719 * before DIO complete the data IO.
3720 *
3721 * As to previously fallocated extents, ext4 get_block
3722 * will just simply mark the buffer mapped but still
3723 * keep the extents uninitialized.
3724 *
3725 * for non AIO case, we will convert those unwritten extents
3726 * to written after return back from blockdev_direct_IO.
3727 *
3728 * for async DIO, the conversion needs to be defered when
3729 * the IO is completed. The ext4 end_io callback function
3730 * will be called to take care of the conversion work.
3731 * Here for async case, we allocate an io_end structure to
3732 * hook to the iocb.
3733 */
3734 iocb->private = NULL;
3735 EXT4_I(inode)->cur_aio_dio = NULL;
3736 if (!is_sync_kiocb(iocb)) {
3737 iocb->private = ext4_init_io_end(inode);
3738 if (!iocb->private)
3739 return -ENOMEM;
3740 /*
3741 * we save the io structure for current async
3742 * direct IO, so that later ext4_get_blocks()
3743 * could flag the io structure whether there
3744 * is a unwritten extents needs to be converted
3745 * when IO is completed.
3746 */
3747 EXT4_I(inode)->cur_aio_dio = iocb->private;
3748 }
3749
3750 ret = blockdev_direct_IO(rw, iocb, inode,
3751 inode->i_sb->s_bdev, iov,
3752 offset, nr_segs,
3753 ext4_get_block_dio_write,
3754 ext4_end_io_dio);
3755 if (iocb->private)
3756 EXT4_I(inode)->cur_aio_dio = NULL;
3757 /*
3758 * The io_end structure takes a reference to the inode,
3759 * that structure needs to be destroyed and the
3760 * reference to the inode need to be dropped, when IO is
3761 * complete, even with 0 byte write, or failed.
3762 *
3763 * In the successful AIO DIO case, the io_end structure will be
3764 * desctroyed and the reference to the inode will be dropped
3765 * after the end_io call back function is called.
3766 *
3767 * In the case there is 0 byte write, or error case, since
3768 * VFS direct IO won't invoke the end_io call back function,
3769 * we need to free the end_io structure here.
3770 */
3771 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3772 ext4_free_io_end(iocb->private);
3773 iocb->private = NULL;
3774 } else if (ret > 0 && (EXT4_I(inode)->i_state &
3775 EXT4_STATE_DIO_UNWRITTEN)) {
3776 int err;
3777 /*
3778 * for non AIO case, since the IO is already
3779 * completed, we could do the convertion right here
3780 */
3781 err = ext4_convert_unwritten_extents(inode,
3782 offset, ret);
3783 if (err < 0)
3784 ret = err;
3785 EXT4_I(inode)->i_state &= ~EXT4_STATE_DIO_UNWRITTEN;
3786 }
3787 return ret;
3788 }
3789
3790 /* for write the the end of file case, we fall back to old way */
3791 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3792 }
3793
3794 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3795 const struct iovec *iov, loff_t offset,
3796 unsigned long nr_segs)
3797 {
3798 struct file *file = iocb->ki_filp;
3799 struct inode *inode = file->f_mapping->host;
3800
3801 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3802 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3803
3804 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3805 }
3806
3807 /*
3808 * Pages can be marked dirty completely asynchronously from ext4's journalling
3809 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3810 * much here because ->set_page_dirty is called under VFS locks. The page is
3811 * not necessarily locked.
3812 *
3813 * We cannot just dirty the page and leave attached buffers clean, because the
3814 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3815 * or jbddirty because all the journalling code will explode.
3816 *
3817 * So what we do is to mark the page "pending dirty" and next time writepage
3818 * is called, propagate that into the buffers appropriately.
3819 */
3820 static int ext4_journalled_set_page_dirty(struct page *page)
3821 {
3822 SetPageChecked(page);
3823 return __set_page_dirty_nobuffers(page);
3824 }
3825
3826 static const struct address_space_operations ext4_ordered_aops = {
3827 .readpage = ext4_readpage,
3828 .readpages = ext4_readpages,
3829 .writepage = ext4_writepage,
3830 .sync_page = block_sync_page,
3831 .write_begin = ext4_write_begin,
3832 .write_end = ext4_ordered_write_end,
3833 .bmap = ext4_bmap,
3834 .invalidatepage = ext4_invalidatepage,
3835 .releasepage = ext4_releasepage,
3836 .direct_IO = ext4_direct_IO,
3837 .migratepage = buffer_migrate_page,
3838 .is_partially_uptodate = block_is_partially_uptodate,
3839 };
3840
3841 static const struct address_space_operations ext4_writeback_aops = {
3842 .readpage = ext4_readpage,
3843 .readpages = ext4_readpages,
3844 .writepage = ext4_writepage,
3845 .sync_page = block_sync_page,
3846 .write_begin = ext4_write_begin,
3847 .write_end = ext4_writeback_write_end,
3848 .bmap = ext4_bmap,
3849 .invalidatepage = ext4_invalidatepage,
3850 .releasepage = ext4_releasepage,
3851 .direct_IO = ext4_direct_IO,
3852 .migratepage = buffer_migrate_page,
3853 .is_partially_uptodate = block_is_partially_uptodate,
3854 };
3855
3856 static const struct address_space_operations ext4_journalled_aops = {
3857 .readpage = ext4_readpage,
3858 .readpages = ext4_readpages,
3859 .writepage = ext4_writepage,
3860 .sync_page = block_sync_page,
3861 .write_begin = ext4_write_begin,
3862 .write_end = ext4_journalled_write_end,
3863 .set_page_dirty = ext4_journalled_set_page_dirty,
3864 .bmap = ext4_bmap,
3865 .invalidatepage = ext4_invalidatepage,
3866 .releasepage = ext4_releasepage,
3867 .is_partially_uptodate = block_is_partially_uptodate,
3868 };
3869
3870 static const struct address_space_operations ext4_da_aops = {
3871 .readpage = ext4_readpage,
3872 .readpages = ext4_readpages,
3873 .writepage = ext4_writepage,
3874 .writepages = ext4_da_writepages,
3875 .sync_page = block_sync_page,
3876 .write_begin = ext4_da_write_begin,
3877 .write_end = ext4_da_write_end,
3878 .bmap = ext4_bmap,
3879 .invalidatepage = ext4_da_invalidatepage,
3880 .releasepage = ext4_releasepage,
3881 .direct_IO = ext4_direct_IO,
3882 .migratepage = buffer_migrate_page,
3883 .is_partially_uptodate = block_is_partially_uptodate,
3884 };
3885
3886 void ext4_set_aops(struct inode *inode)
3887 {
3888 if (ext4_should_order_data(inode) &&
3889 test_opt(inode->i_sb, DELALLOC))
3890 inode->i_mapping->a_ops = &ext4_da_aops;
3891 else if (ext4_should_order_data(inode))
3892 inode->i_mapping->a_ops = &ext4_ordered_aops;
3893 else if (ext4_should_writeback_data(inode) &&
3894 test_opt(inode->i_sb, DELALLOC))
3895 inode->i_mapping->a_ops = &ext4_da_aops;
3896 else if (ext4_should_writeback_data(inode))
3897 inode->i_mapping->a_ops = &ext4_writeback_aops;
3898 else
3899 inode->i_mapping->a_ops = &ext4_journalled_aops;
3900 }
3901
3902 /*
3903 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3904 * up to the end of the block which corresponds to `from'.
3905 * This required during truncate. We need to physically zero the tail end
3906 * of that block so it doesn't yield old data if the file is later grown.
3907 */
3908 int ext4_block_truncate_page(handle_t *handle,
3909 struct address_space *mapping, loff_t from)
3910 {
3911 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3912 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3913 unsigned blocksize, length, pos;
3914 ext4_lblk_t iblock;
3915 struct inode *inode = mapping->host;
3916 struct buffer_head *bh;
3917 struct page *page;
3918 int err = 0;
3919
3920 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3921 mapping_gfp_mask(mapping) & ~__GFP_FS);
3922 if (!page)
3923 return -EINVAL;
3924
3925 blocksize = inode->i_sb->s_blocksize;
3926 length = blocksize - (offset & (blocksize - 1));
3927 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3928
3929 /*
3930 * For "nobh" option, we can only work if we don't need to
3931 * read-in the page - otherwise we create buffers to do the IO.
3932 */
3933 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3934 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3935 zero_user(page, offset, length);
3936 set_page_dirty(page);
3937 goto unlock;
3938 }
3939
3940 if (!page_has_buffers(page))
3941 create_empty_buffers(page, blocksize, 0);
3942
3943 /* Find the buffer that contains "offset" */
3944 bh = page_buffers(page);
3945 pos = blocksize;
3946 while (offset >= pos) {
3947 bh = bh->b_this_page;
3948 iblock++;
3949 pos += blocksize;
3950 }
3951
3952 err = 0;
3953 if (buffer_freed(bh)) {
3954 BUFFER_TRACE(bh, "freed: skip");
3955 goto unlock;
3956 }
3957
3958 if (!buffer_mapped(bh)) {
3959 BUFFER_TRACE(bh, "unmapped");
3960 ext4_get_block(inode, iblock, bh, 0);
3961 /* unmapped? It's a hole - nothing to do */
3962 if (!buffer_mapped(bh)) {
3963 BUFFER_TRACE(bh, "still unmapped");
3964 goto unlock;
3965 }
3966 }
3967
3968 /* Ok, it's mapped. Make sure it's up-to-date */
3969 if (PageUptodate(page))
3970 set_buffer_uptodate(bh);
3971
3972 if (!buffer_uptodate(bh)) {
3973 err = -EIO;
3974 ll_rw_block(READ, 1, &bh);
3975 wait_on_buffer(bh);
3976 /* Uhhuh. Read error. Complain and punt. */
3977 if (!buffer_uptodate(bh))
3978 goto unlock;
3979 }
3980
3981 if (ext4_should_journal_data(inode)) {
3982 BUFFER_TRACE(bh, "get write access");
3983 err = ext4_journal_get_write_access(handle, bh);
3984 if (err)
3985 goto unlock;
3986 }
3987
3988 zero_user(page, offset, length);
3989
3990 BUFFER_TRACE(bh, "zeroed end of block");
3991
3992 err = 0;
3993 if (ext4_should_journal_data(inode)) {
3994 err = ext4_handle_dirty_metadata(handle, inode, bh);
3995 } else {
3996 if (ext4_should_order_data(inode))
3997 err = ext4_jbd2_file_inode(handle, inode);
3998 mark_buffer_dirty(bh);
3999 }
4000
4001 unlock:
4002 unlock_page(page);
4003 page_cache_release(page);
4004 return err;
4005 }
4006
4007 /*
4008 * Probably it should be a library function... search for first non-zero word
4009 * or memcmp with zero_page, whatever is better for particular architecture.
4010 * Linus?
4011 */
4012 static inline int all_zeroes(__le32 *p, __le32 *q)
4013 {
4014 while (p < q)
4015 if (*p++)
4016 return 0;
4017 return 1;
4018 }
4019
4020 /**
4021 * ext4_find_shared - find the indirect blocks for partial truncation.
4022 * @inode: inode in question
4023 * @depth: depth of the affected branch
4024 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4025 * @chain: place to store the pointers to partial indirect blocks
4026 * @top: place to the (detached) top of branch
4027 *
4028 * This is a helper function used by ext4_truncate().
4029 *
4030 * When we do truncate() we may have to clean the ends of several
4031 * indirect blocks but leave the blocks themselves alive. Block is
4032 * partially truncated if some data below the new i_size is refered
4033 * from it (and it is on the path to the first completely truncated
4034 * data block, indeed). We have to free the top of that path along
4035 * with everything to the right of the path. Since no allocation
4036 * past the truncation point is possible until ext4_truncate()
4037 * finishes, we may safely do the latter, but top of branch may
4038 * require special attention - pageout below the truncation point
4039 * might try to populate it.
4040 *
4041 * We atomically detach the top of branch from the tree, store the
4042 * block number of its root in *@top, pointers to buffer_heads of
4043 * partially truncated blocks - in @chain[].bh and pointers to
4044 * their last elements that should not be removed - in
4045 * @chain[].p. Return value is the pointer to last filled element
4046 * of @chain.
4047 *
4048 * The work left to caller to do the actual freeing of subtrees:
4049 * a) free the subtree starting from *@top
4050 * b) free the subtrees whose roots are stored in
4051 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4052 * c) free the subtrees growing from the inode past the @chain[0].
4053 * (no partially truncated stuff there). */
4054
4055 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4056 ext4_lblk_t offsets[4], Indirect chain[4],
4057 __le32 *top)
4058 {
4059 Indirect *partial, *p;
4060 int k, err;
4061
4062 *top = 0;
4063 /* Make k index the deepest non-null offest + 1 */
4064 for (k = depth; k > 1 && !offsets[k-1]; k--)
4065 ;
4066 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4067 /* Writer: pointers */
4068 if (!partial)
4069 partial = chain + k-1;
4070 /*
4071 * If the branch acquired continuation since we've looked at it -
4072 * fine, it should all survive and (new) top doesn't belong to us.
4073 */
4074 if (!partial->key && *partial->p)
4075 /* Writer: end */
4076 goto no_top;
4077 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4078 ;
4079 /*
4080 * OK, we've found the last block that must survive. The rest of our
4081 * branch should be detached before unlocking. However, if that rest
4082 * of branch is all ours and does not grow immediately from the inode
4083 * it's easier to cheat and just decrement partial->p.
4084 */
4085 if (p == chain + k - 1 && p > chain) {
4086 p->p--;
4087 } else {
4088 *top = *p->p;
4089 /* Nope, don't do this in ext4. Must leave the tree intact */
4090 #if 0
4091 *p->p = 0;
4092 #endif
4093 }
4094 /* Writer: end */
4095
4096 while (partial > p) {
4097 brelse(partial->bh);
4098 partial--;
4099 }
4100 no_top:
4101 return partial;
4102 }
4103
4104 /*
4105 * Zero a number of block pointers in either an inode or an indirect block.
4106 * If we restart the transaction we must again get write access to the
4107 * indirect block for further modification.
4108 *
4109 * We release `count' blocks on disk, but (last - first) may be greater
4110 * than `count' because there can be holes in there.
4111 */
4112 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
4113 struct buffer_head *bh,
4114 ext4_fsblk_t block_to_free,
4115 unsigned long count, __le32 *first,
4116 __le32 *last)
4117 {
4118 __le32 *p;
4119 int is_metadata = S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode);
4120
4121 if (try_to_extend_transaction(handle, inode)) {
4122 if (bh) {
4123 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4124 ext4_handle_dirty_metadata(handle, inode, bh);
4125 }
4126 ext4_mark_inode_dirty(handle, inode);
4127 ext4_truncate_restart_trans(handle, inode,
4128 blocks_for_truncate(inode));
4129 if (bh) {
4130 BUFFER_TRACE(bh, "retaking write access");
4131 ext4_journal_get_write_access(handle, bh);
4132 }
4133 }
4134
4135 /*
4136 * Any buffers which are on the journal will be in memory. We
4137 * find them on the hash table so jbd2_journal_revoke() will
4138 * run jbd2_journal_forget() on them. We've already detached
4139 * each block from the file, so bforget() in
4140 * jbd2_journal_forget() should be safe.
4141 *
4142 * AKPM: turn on bforget in jbd2_journal_forget()!!!
4143 */
4144 for (p = first; p < last; p++) {
4145 u32 nr = le32_to_cpu(*p);
4146 if (nr) {
4147 struct buffer_head *tbh;
4148
4149 *p = 0;
4150 tbh = sb_find_get_block(inode->i_sb, nr);
4151 ext4_forget(handle, is_metadata, inode, tbh, nr);
4152 }
4153 }
4154
4155 ext4_free_blocks(handle, inode, block_to_free, count, is_metadata);
4156 }
4157
4158 /**
4159 * ext4_free_data - free a list of data blocks
4160 * @handle: handle for this transaction
4161 * @inode: inode we are dealing with
4162 * @this_bh: indirect buffer_head which contains *@first and *@last
4163 * @first: array of block numbers
4164 * @last: points immediately past the end of array
4165 *
4166 * We are freeing all blocks refered from that array (numbers are stored as
4167 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4168 *
4169 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4170 * blocks are contiguous then releasing them at one time will only affect one
4171 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4172 * actually use a lot of journal space.
4173 *
4174 * @this_bh will be %NULL if @first and @last point into the inode's direct
4175 * block pointers.
4176 */
4177 static void ext4_free_data(handle_t *handle, struct inode *inode,
4178 struct buffer_head *this_bh,
4179 __le32 *first, __le32 *last)
4180 {
4181 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4182 unsigned long count = 0; /* Number of blocks in the run */
4183 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4184 corresponding to
4185 block_to_free */
4186 ext4_fsblk_t nr; /* Current block # */
4187 __le32 *p; /* Pointer into inode/ind
4188 for current block */
4189 int err;
4190
4191 if (this_bh) { /* For indirect block */
4192 BUFFER_TRACE(this_bh, "get_write_access");
4193 err = ext4_journal_get_write_access(handle, this_bh);
4194 /* Important: if we can't update the indirect pointers
4195 * to the blocks, we can't free them. */
4196 if (err)
4197 return;
4198 }
4199
4200 for (p = first; p < last; p++) {
4201 nr = le32_to_cpu(*p);
4202 if (nr) {
4203 /* accumulate blocks to free if they're contiguous */
4204 if (count == 0) {
4205 block_to_free = nr;
4206 block_to_free_p = p;
4207 count = 1;
4208 } else if (nr == block_to_free + count) {
4209 count++;
4210 } else {
4211 ext4_clear_blocks(handle, inode, this_bh,
4212 block_to_free,
4213 count, block_to_free_p, p);
4214 block_to_free = nr;
4215 block_to_free_p = p;
4216 count = 1;
4217 }
4218 }
4219 }
4220
4221 if (count > 0)
4222 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4223 count, block_to_free_p, p);
4224
4225 if (this_bh) {
4226 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4227
4228 /*
4229 * The buffer head should have an attached journal head at this
4230 * point. However, if the data is corrupted and an indirect
4231 * block pointed to itself, it would have been detached when
4232 * the block was cleared. Check for this instead of OOPSing.
4233 */
4234 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4235 ext4_handle_dirty_metadata(handle, inode, this_bh);
4236 else
4237 ext4_error(inode->i_sb, __func__,
4238 "circular indirect block detected, "
4239 "inode=%lu, block=%llu",
4240 inode->i_ino,
4241 (unsigned long long) this_bh->b_blocknr);
4242 }
4243 }
4244
4245 /**
4246 * ext4_free_branches - free an array of branches
4247 * @handle: JBD handle for this transaction
4248 * @inode: inode we are dealing with
4249 * @parent_bh: the buffer_head which contains *@first and *@last
4250 * @first: array of block numbers
4251 * @last: pointer immediately past the end of array
4252 * @depth: depth of the branches to free
4253 *
4254 * We are freeing all blocks refered from these branches (numbers are
4255 * stored as little-endian 32-bit) and updating @inode->i_blocks
4256 * appropriately.
4257 */
4258 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4259 struct buffer_head *parent_bh,
4260 __le32 *first, __le32 *last, int depth)
4261 {
4262 ext4_fsblk_t nr;
4263 __le32 *p;
4264
4265 if (ext4_handle_is_aborted(handle))
4266 return;
4267
4268 if (depth--) {
4269 struct buffer_head *bh;
4270 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4271 p = last;
4272 while (--p >= first) {
4273 nr = le32_to_cpu(*p);
4274 if (!nr)
4275 continue; /* A hole */
4276
4277 /* Go read the buffer for the next level down */
4278 bh = sb_bread(inode->i_sb, nr);
4279
4280 /*
4281 * A read failure? Report error and clear slot
4282 * (should be rare).
4283 */
4284 if (!bh) {
4285 ext4_error(inode->i_sb, "ext4_free_branches",
4286 "Read failure, inode=%lu, block=%llu",
4287 inode->i_ino, nr);
4288 continue;
4289 }
4290
4291 /* This zaps the entire block. Bottom up. */
4292 BUFFER_TRACE(bh, "free child branches");
4293 ext4_free_branches(handle, inode, bh,
4294 (__le32 *) bh->b_data,
4295 (__le32 *) bh->b_data + addr_per_block,
4296 depth);
4297
4298 /*
4299 * We've probably journalled the indirect block several
4300 * times during the truncate. But it's no longer
4301 * needed and we now drop it from the transaction via
4302 * jbd2_journal_revoke().
4303 *
4304 * That's easy if it's exclusively part of this
4305 * transaction. But if it's part of the committing
4306 * transaction then jbd2_journal_forget() will simply
4307 * brelse() it. That means that if the underlying
4308 * block is reallocated in ext4_get_block(),
4309 * unmap_underlying_metadata() will find this block
4310 * and will try to get rid of it. damn, damn.
4311 *
4312 * If this block has already been committed to the
4313 * journal, a revoke record will be written. And
4314 * revoke records must be emitted *before* clearing
4315 * this block's bit in the bitmaps.
4316 */
4317 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
4318
4319 /*
4320 * Everything below this this pointer has been
4321 * released. Now let this top-of-subtree go.
4322 *
4323 * We want the freeing of this indirect block to be
4324 * atomic in the journal with the updating of the
4325 * bitmap block which owns it. So make some room in
4326 * the journal.
4327 *
4328 * We zero the parent pointer *after* freeing its
4329 * pointee in the bitmaps, so if extend_transaction()
4330 * for some reason fails to put the bitmap changes and
4331 * the release into the same transaction, recovery
4332 * will merely complain about releasing a free block,
4333 * rather than leaking blocks.
4334 */
4335 if (ext4_handle_is_aborted(handle))
4336 return;
4337 if (try_to_extend_transaction(handle, inode)) {
4338 ext4_mark_inode_dirty(handle, inode);
4339 ext4_truncate_restart_trans(handle, inode,
4340 blocks_for_truncate(inode));
4341 }
4342
4343 ext4_free_blocks(handle, inode, nr, 1, 1);
4344
4345 if (parent_bh) {
4346 /*
4347 * The block which we have just freed is
4348 * pointed to by an indirect block: journal it
4349 */
4350 BUFFER_TRACE(parent_bh, "get_write_access");
4351 if (!ext4_journal_get_write_access(handle,
4352 parent_bh)){
4353 *p = 0;
4354 BUFFER_TRACE(parent_bh,
4355 "call ext4_handle_dirty_metadata");
4356 ext4_handle_dirty_metadata(handle,
4357 inode,
4358 parent_bh);
4359 }
4360 }
4361 }
4362 } else {
4363 /* We have reached the bottom of the tree. */
4364 BUFFER_TRACE(parent_bh, "free data blocks");
4365 ext4_free_data(handle, inode, parent_bh, first, last);
4366 }
4367 }
4368
4369 int ext4_can_truncate(struct inode *inode)
4370 {
4371 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4372 return 0;
4373 if (S_ISREG(inode->i_mode))
4374 return 1;
4375 if (S_ISDIR(inode->i_mode))
4376 return 1;
4377 if (S_ISLNK(inode->i_mode))
4378 return !ext4_inode_is_fast_symlink(inode);
4379 return 0;
4380 }
4381
4382 /*
4383 * ext4_truncate()
4384 *
4385 * We block out ext4_get_block() block instantiations across the entire
4386 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4387 * simultaneously on behalf of the same inode.
4388 *
4389 * As we work through the truncate and commmit bits of it to the journal there
4390 * is one core, guiding principle: the file's tree must always be consistent on
4391 * disk. We must be able to restart the truncate after a crash.
4392 *
4393 * The file's tree may be transiently inconsistent in memory (although it
4394 * probably isn't), but whenever we close off and commit a journal transaction,
4395 * the contents of (the filesystem + the journal) must be consistent and
4396 * restartable. It's pretty simple, really: bottom up, right to left (although
4397 * left-to-right works OK too).
4398 *
4399 * Note that at recovery time, journal replay occurs *before* the restart of
4400 * truncate against the orphan inode list.
4401 *
4402 * The committed inode has the new, desired i_size (which is the same as
4403 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4404 * that this inode's truncate did not complete and it will again call
4405 * ext4_truncate() to have another go. So there will be instantiated blocks
4406 * to the right of the truncation point in a crashed ext4 filesystem. But
4407 * that's fine - as long as they are linked from the inode, the post-crash
4408 * ext4_truncate() run will find them and release them.
4409 */
4410 void ext4_truncate(struct inode *inode)
4411 {
4412 handle_t *handle;
4413 struct ext4_inode_info *ei = EXT4_I(inode);
4414 __le32 *i_data = ei->i_data;
4415 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4416 struct address_space *mapping = inode->i_mapping;
4417 ext4_lblk_t offsets[4];
4418 Indirect chain[4];
4419 Indirect *partial;
4420 __le32 nr = 0;
4421 int n;
4422 ext4_lblk_t last_block;
4423 unsigned blocksize = inode->i_sb->s_blocksize;
4424
4425 if (!ext4_can_truncate(inode))
4426 return;
4427
4428 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4429 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
4430
4431 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
4432 ext4_ext_truncate(inode);
4433 return;
4434 }
4435
4436 handle = start_transaction(inode);
4437 if (IS_ERR(handle))
4438 return; /* AKPM: return what? */
4439
4440 last_block = (inode->i_size + blocksize-1)
4441 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4442
4443 if (inode->i_size & (blocksize - 1))
4444 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4445 goto out_stop;
4446
4447 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4448 if (n == 0)
4449 goto out_stop; /* error */
4450
4451 /*
4452 * OK. This truncate is going to happen. We add the inode to the
4453 * orphan list, so that if this truncate spans multiple transactions,
4454 * and we crash, we will resume the truncate when the filesystem
4455 * recovers. It also marks the inode dirty, to catch the new size.
4456 *
4457 * Implication: the file must always be in a sane, consistent
4458 * truncatable state while each transaction commits.
4459 */
4460 if (ext4_orphan_add(handle, inode))
4461 goto out_stop;
4462
4463 /*
4464 * From here we block out all ext4_get_block() callers who want to
4465 * modify the block allocation tree.
4466 */
4467 down_write(&ei->i_data_sem);
4468
4469 ext4_discard_preallocations(inode);
4470
4471 /*
4472 * The orphan list entry will now protect us from any crash which
4473 * occurs before the truncate completes, so it is now safe to propagate
4474 * the new, shorter inode size (held for now in i_size) into the
4475 * on-disk inode. We do this via i_disksize, which is the value which
4476 * ext4 *really* writes onto the disk inode.
4477 */
4478 ei->i_disksize = inode->i_size;
4479
4480 if (n == 1) { /* direct blocks */
4481 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4482 i_data + EXT4_NDIR_BLOCKS);
4483 goto do_indirects;
4484 }
4485
4486 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4487 /* Kill the top of shared branch (not detached) */
4488 if (nr) {
4489 if (partial == chain) {
4490 /* Shared branch grows from the inode */
4491 ext4_free_branches(handle, inode, NULL,
4492 &nr, &nr+1, (chain+n-1) - partial);
4493 *partial->p = 0;
4494 /*
4495 * We mark the inode dirty prior to restart,
4496 * and prior to stop. No need for it here.
4497 */
4498 } else {
4499 /* Shared branch grows from an indirect block */
4500 BUFFER_TRACE(partial->bh, "get_write_access");
4501 ext4_free_branches(handle, inode, partial->bh,
4502 partial->p,
4503 partial->p+1, (chain+n-1) - partial);
4504 }
4505 }
4506 /* Clear the ends of indirect blocks on the shared branch */
4507 while (partial > chain) {
4508 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4509 (__le32*)partial->bh->b_data+addr_per_block,
4510 (chain+n-1) - partial);
4511 BUFFER_TRACE(partial->bh, "call brelse");
4512 brelse(partial->bh);
4513 partial--;
4514 }
4515 do_indirects:
4516 /* Kill the remaining (whole) subtrees */
4517 switch (offsets[0]) {
4518 default:
4519 nr = i_data[EXT4_IND_BLOCK];
4520 if (nr) {
4521 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4522 i_data[EXT4_IND_BLOCK] = 0;
4523 }
4524 case EXT4_IND_BLOCK:
4525 nr = i_data[EXT4_DIND_BLOCK];
4526 if (nr) {
4527 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4528 i_data[EXT4_DIND_BLOCK] = 0;
4529 }
4530 case EXT4_DIND_BLOCK:
4531 nr = i_data[EXT4_TIND_BLOCK];
4532 if (nr) {
4533 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4534 i_data[EXT4_TIND_BLOCK] = 0;
4535 }
4536 case EXT4_TIND_BLOCK:
4537 ;
4538 }
4539
4540 up_write(&ei->i_data_sem);
4541 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4542 ext4_mark_inode_dirty(handle, inode);
4543
4544 /*
4545 * In a multi-transaction truncate, we only make the final transaction
4546 * synchronous
4547 */
4548 if (IS_SYNC(inode))
4549 ext4_handle_sync(handle);
4550 out_stop:
4551 /*
4552 * If this was a simple ftruncate(), and the file will remain alive
4553 * then we need to clear up the orphan record which we created above.
4554 * However, if this was a real unlink then we were called by
4555 * ext4_delete_inode(), and we allow that function to clean up the
4556 * orphan info for us.
4557 */
4558 if (inode->i_nlink)
4559 ext4_orphan_del(handle, inode);
4560
4561 ext4_journal_stop(handle);
4562 }
4563
4564 /*
4565 * ext4_get_inode_loc returns with an extra refcount against the inode's
4566 * underlying buffer_head on success. If 'in_mem' is true, we have all
4567 * data in memory that is needed to recreate the on-disk version of this
4568 * inode.
4569 */
4570 static int __ext4_get_inode_loc(struct inode *inode,
4571 struct ext4_iloc *iloc, int in_mem)
4572 {
4573 struct ext4_group_desc *gdp;
4574 struct buffer_head *bh;
4575 struct super_block *sb = inode->i_sb;
4576 ext4_fsblk_t block;
4577 int inodes_per_block, inode_offset;
4578
4579 iloc->bh = NULL;
4580 if (!ext4_valid_inum(sb, inode->i_ino))
4581 return -EIO;
4582
4583 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4584 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4585 if (!gdp)
4586 return -EIO;
4587
4588 /*
4589 * Figure out the offset within the block group inode table
4590 */
4591 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4592 inode_offset = ((inode->i_ino - 1) %
4593 EXT4_INODES_PER_GROUP(sb));
4594 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4595 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4596
4597 bh = sb_getblk(sb, block);
4598 if (!bh) {
4599 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4600 "inode block - inode=%lu, block=%llu",
4601 inode->i_ino, block);
4602 return -EIO;
4603 }
4604 if (!buffer_uptodate(bh)) {
4605 lock_buffer(bh);
4606
4607 /*
4608 * If the buffer has the write error flag, we have failed
4609 * to write out another inode in the same block. In this
4610 * case, we don't have to read the block because we may
4611 * read the old inode data successfully.
4612 */
4613 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4614 set_buffer_uptodate(bh);
4615
4616 if (buffer_uptodate(bh)) {
4617 /* someone brought it uptodate while we waited */
4618 unlock_buffer(bh);
4619 goto has_buffer;
4620 }
4621
4622 /*
4623 * If we have all information of the inode in memory and this
4624 * is the only valid inode in the block, we need not read the
4625 * block.
4626 */
4627 if (in_mem) {
4628 struct buffer_head *bitmap_bh;
4629 int i, start;
4630
4631 start = inode_offset & ~(inodes_per_block - 1);
4632
4633 /* Is the inode bitmap in cache? */
4634 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4635 if (!bitmap_bh)
4636 goto make_io;
4637
4638 /*
4639 * If the inode bitmap isn't in cache then the
4640 * optimisation may end up performing two reads instead
4641 * of one, so skip it.
4642 */
4643 if (!buffer_uptodate(bitmap_bh)) {
4644 brelse(bitmap_bh);
4645 goto make_io;
4646 }
4647 for (i = start; i < start + inodes_per_block; i++) {
4648 if (i == inode_offset)
4649 continue;
4650 if (ext4_test_bit(i, bitmap_bh->b_data))
4651 break;
4652 }
4653 brelse(bitmap_bh);
4654 if (i == start + inodes_per_block) {
4655 /* all other inodes are free, so skip I/O */
4656 memset(bh->b_data, 0, bh->b_size);
4657 set_buffer_uptodate(bh);
4658 unlock_buffer(bh);
4659 goto has_buffer;
4660 }
4661 }
4662
4663 make_io:
4664 /*
4665 * If we need to do any I/O, try to pre-readahead extra
4666 * blocks from the inode table.
4667 */
4668 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4669 ext4_fsblk_t b, end, table;
4670 unsigned num;
4671
4672 table = ext4_inode_table(sb, gdp);
4673 /* s_inode_readahead_blks is always a power of 2 */
4674 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4675 if (table > b)
4676 b = table;
4677 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4678 num = EXT4_INODES_PER_GROUP(sb);
4679 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4680 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4681 num -= ext4_itable_unused_count(sb, gdp);
4682 table += num / inodes_per_block;
4683 if (end > table)
4684 end = table;
4685 while (b <= end)
4686 sb_breadahead(sb, b++);
4687 }
4688
4689 /*
4690 * There are other valid inodes in the buffer, this inode
4691 * has in-inode xattrs, or we don't have this inode in memory.
4692 * Read the block from disk.
4693 */
4694 get_bh(bh);
4695 bh->b_end_io = end_buffer_read_sync;
4696 submit_bh(READ_META, bh);
4697 wait_on_buffer(bh);
4698 if (!buffer_uptodate(bh)) {
4699 ext4_error(sb, __func__,
4700 "unable to read inode block - inode=%lu, "
4701 "block=%llu", inode->i_ino, block);
4702 brelse(bh);
4703 return -EIO;
4704 }
4705 }
4706 has_buffer:
4707 iloc->bh = bh;
4708 return 0;
4709 }
4710
4711 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4712 {
4713 /* We have all inode data except xattrs in memory here. */
4714 return __ext4_get_inode_loc(inode, iloc,
4715 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4716 }
4717
4718 void ext4_set_inode_flags(struct inode *inode)
4719 {
4720 unsigned int flags = EXT4_I(inode)->i_flags;
4721
4722 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4723 if (flags & EXT4_SYNC_FL)
4724 inode->i_flags |= S_SYNC;
4725 if (flags & EXT4_APPEND_FL)
4726 inode->i_flags |= S_APPEND;
4727 if (flags & EXT4_IMMUTABLE_FL)
4728 inode->i_flags |= S_IMMUTABLE;
4729 if (flags & EXT4_NOATIME_FL)
4730 inode->i_flags |= S_NOATIME;
4731 if (flags & EXT4_DIRSYNC_FL)
4732 inode->i_flags |= S_DIRSYNC;
4733 }
4734
4735 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4736 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4737 {
4738 unsigned int flags = ei->vfs_inode.i_flags;
4739
4740 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4741 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4742 if (flags & S_SYNC)
4743 ei->i_flags |= EXT4_SYNC_FL;
4744 if (flags & S_APPEND)
4745 ei->i_flags |= EXT4_APPEND_FL;
4746 if (flags & S_IMMUTABLE)
4747 ei->i_flags |= EXT4_IMMUTABLE_FL;
4748 if (flags & S_NOATIME)
4749 ei->i_flags |= EXT4_NOATIME_FL;
4750 if (flags & S_DIRSYNC)
4751 ei->i_flags |= EXT4_DIRSYNC_FL;
4752 }
4753
4754 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4755 struct ext4_inode_info *ei)
4756 {
4757 blkcnt_t i_blocks ;
4758 struct inode *inode = &(ei->vfs_inode);
4759 struct super_block *sb = inode->i_sb;
4760
4761 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4762 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4763 /* we are using combined 48 bit field */
4764 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4765 le32_to_cpu(raw_inode->i_blocks_lo);
4766 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4767 /* i_blocks represent file system block size */
4768 return i_blocks << (inode->i_blkbits - 9);
4769 } else {
4770 return i_blocks;
4771 }
4772 } else {
4773 return le32_to_cpu(raw_inode->i_blocks_lo);
4774 }
4775 }
4776
4777 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4778 {
4779 struct ext4_iloc iloc;
4780 struct ext4_inode *raw_inode;
4781 struct ext4_inode_info *ei;
4782 struct inode *inode;
4783 journal_t *journal = EXT4_SB(sb)->s_journal;
4784 long ret;
4785 int block;
4786
4787 inode = iget_locked(sb, ino);
4788 if (!inode)
4789 return ERR_PTR(-ENOMEM);
4790 if (!(inode->i_state & I_NEW))
4791 return inode;
4792
4793 ei = EXT4_I(inode);
4794 iloc.bh = 0;
4795
4796 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4797 if (ret < 0)
4798 goto bad_inode;
4799 raw_inode = ext4_raw_inode(&iloc);
4800 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4801 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4802 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4803 if (!(test_opt(inode->i_sb, NO_UID32))) {
4804 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4805 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4806 }
4807 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4808
4809 ei->i_state = 0;
4810 ei->i_dir_start_lookup = 0;
4811 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4812 /* We now have enough fields to check if the inode was active or not.
4813 * This is needed because nfsd might try to access dead inodes
4814 * the test is that same one that e2fsck uses
4815 * NeilBrown 1999oct15
4816 */
4817 if (inode->i_nlink == 0) {
4818 if (inode->i_mode == 0 ||
4819 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4820 /* this inode is deleted */
4821 ret = -ESTALE;
4822 goto bad_inode;
4823 }
4824 /* The only unlinked inodes we let through here have
4825 * valid i_mode and are being read by the orphan
4826 * recovery code: that's fine, we're about to complete
4827 * the process of deleting those. */
4828 }
4829 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4830 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4831 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4832 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4833 ei->i_file_acl |=
4834 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4835 inode->i_size = ext4_isize(raw_inode);
4836 ei->i_disksize = inode->i_size;
4837 #ifdef CONFIG_QUOTA
4838 ei->i_reserved_quota = 0;
4839 #endif
4840 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4841 ei->i_block_group = iloc.block_group;
4842 ei->i_last_alloc_group = ~0;
4843 /*
4844 * NOTE! The in-memory inode i_data array is in little-endian order
4845 * even on big-endian machines: we do NOT byteswap the block numbers!
4846 */
4847 for (block = 0; block < EXT4_N_BLOCKS; block++)
4848 ei->i_data[block] = raw_inode->i_block[block];
4849 INIT_LIST_HEAD(&ei->i_orphan);
4850
4851 /*
4852 * Set transaction id's of transactions that have to be committed
4853 * to finish f[data]sync. We set them to currently running transaction
4854 * as we cannot be sure that the inode or some of its metadata isn't
4855 * part of the transaction - the inode could have been reclaimed and
4856 * now it is reread from disk.
4857 */
4858 if (journal) {
4859 transaction_t *transaction;
4860 tid_t tid;
4861
4862 spin_lock(&journal->j_state_lock);
4863 if (journal->j_running_transaction)
4864 transaction = journal->j_running_transaction;
4865 else
4866 transaction = journal->j_committing_transaction;
4867 if (transaction)
4868 tid = transaction->t_tid;
4869 else
4870 tid = journal->j_commit_sequence;
4871 spin_unlock(&journal->j_state_lock);
4872 ei->i_sync_tid = tid;
4873 ei->i_datasync_tid = tid;
4874 }
4875
4876 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4877 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4878 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4879 EXT4_INODE_SIZE(inode->i_sb)) {
4880 ret = -EIO;
4881 goto bad_inode;
4882 }
4883 if (ei->i_extra_isize == 0) {
4884 /* The extra space is currently unused. Use it. */
4885 ei->i_extra_isize = sizeof(struct ext4_inode) -
4886 EXT4_GOOD_OLD_INODE_SIZE;
4887 } else {
4888 __le32 *magic = (void *)raw_inode +
4889 EXT4_GOOD_OLD_INODE_SIZE +
4890 ei->i_extra_isize;
4891 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4892 ei->i_state |= EXT4_STATE_XATTR;
4893 }
4894 } else
4895 ei->i_extra_isize = 0;
4896
4897 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4898 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4899 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4900 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4901
4902 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4903 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4904 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4905 inode->i_version |=
4906 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4907 }
4908
4909 ret = 0;
4910 if (ei->i_file_acl &&
4911 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4912 ext4_error(sb, __func__,
4913 "bad extended attribute block %llu in inode #%lu",
4914 ei->i_file_acl, inode->i_ino);
4915 ret = -EIO;
4916 goto bad_inode;
4917 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
4918 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4919 (S_ISLNK(inode->i_mode) &&
4920 !ext4_inode_is_fast_symlink(inode)))
4921 /* Validate extent which is part of inode */
4922 ret = ext4_ext_check_inode(inode);
4923 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4924 (S_ISLNK(inode->i_mode) &&
4925 !ext4_inode_is_fast_symlink(inode))) {
4926 /* Validate block references which are part of inode */
4927 ret = ext4_check_inode_blockref(inode);
4928 }
4929 if (ret)
4930 goto bad_inode;
4931
4932 if (S_ISREG(inode->i_mode)) {
4933 inode->i_op = &ext4_file_inode_operations;
4934 inode->i_fop = &ext4_file_operations;
4935 ext4_set_aops(inode);
4936 } else if (S_ISDIR(inode->i_mode)) {
4937 inode->i_op = &ext4_dir_inode_operations;
4938 inode->i_fop = &ext4_dir_operations;
4939 } else if (S_ISLNK(inode->i_mode)) {
4940 if (ext4_inode_is_fast_symlink(inode)) {
4941 inode->i_op = &ext4_fast_symlink_inode_operations;
4942 nd_terminate_link(ei->i_data, inode->i_size,
4943 sizeof(ei->i_data) - 1);
4944 } else {
4945 inode->i_op = &ext4_symlink_inode_operations;
4946 ext4_set_aops(inode);
4947 }
4948 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4949 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4950 inode->i_op = &ext4_special_inode_operations;
4951 if (raw_inode->i_block[0])
4952 init_special_inode(inode, inode->i_mode,
4953 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4954 else
4955 init_special_inode(inode, inode->i_mode,
4956 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4957 } else {
4958 ret = -EIO;
4959 ext4_error(inode->i_sb, __func__,
4960 "bogus i_mode (%o) for inode=%lu",
4961 inode->i_mode, inode->i_ino);
4962 goto bad_inode;
4963 }
4964 brelse(iloc.bh);
4965 ext4_set_inode_flags(inode);
4966 unlock_new_inode(inode);
4967 return inode;
4968
4969 bad_inode:
4970 brelse(iloc.bh);
4971 iget_failed(inode);
4972 return ERR_PTR(ret);
4973 }
4974
4975 static int ext4_inode_blocks_set(handle_t *handle,
4976 struct ext4_inode *raw_inode,
4977 struct ext4_inode_info *ei)
4978 {
4979 struct inode *inode = &(ei->vfs_inode);
4980 u64 i_blocks = inode->i_blocks;
4981 struct super_block *sb = inode->i_sb;
4982
4983 if (i_blocks <= ~0U) {
4984 /*
4985 * i_blocks can be represnted in a 32 bit variable
4986 * as multiple of 512 bytes
4987 */
4988 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4989 raw_inode->i_blocks_high = 0;
4990 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4991 return 0;
4992 }
4993 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4994 return -EFBIG;
4995
4996 if (i_blocks <= 0xffffffffffffULL) {
4997 /*
4998 * i_blocks can be represented in a 48 bit variable
4999 * as multiple of 512 bytes
5000 */
5001 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5002 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5003 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
5004 } else {
5005 ei->i_flags |= EXT4_HUGE_FILE_FL;
5006 /* i_block is stored in file system block size */
5007 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5008 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5009 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5010 }
5011 return 0;
5012 }
5013
5014 /*
5015 * Post the struct inode info into an on-disk inode location in the
5016 * buffer-cache. This gobbles the caller's reference to the
5017 * buffer_head in the inode location struct.
5018 *
5019 * The caller must have write access to iloc->bh.
5020 */
5021 static int ext4_do_update_inode(handle_t *handle,
5022 struct inode *inode,
5023 struct ext4_iloc *iloc)
5024 {
5025 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5026 struct ext4_inode_info *ei = EXT4_I(inode);
5027 struct buffer_head *bh = iloc->bh;
5028 int err = 0, rc, block;
5029
5030 /* For fields not not tracking in the in-memory inode,
5031 * initialise them to zero for new inodes. */
5032 if (ei->i_state & EXT4_STATE_NEW)
5033 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5034
5035 ext4_get_inode_flags(ei);
5036 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5037 if (!(test_opt(inode->i_sb, NO_UID32))) {
5038 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5039 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5040 /*
5041 * Fix up interoperability with old kernels. Otherwise, old inodes get
5042 * re-used with the upper 16 bits of the uid/gid intact
5043 */
5044 if (!ei->i_dtime) {
5045 raw_inode->i_uid_high =
5046 cpu_to_le16(high_16_bits(inode->i_uid));
5047 raw_inode->i_gid_high =
5048 cpu_to_le16(high_16_bits(inode->i_gid));
5049 } else {
5050 raw_inode->i_uid_high = 0;
5051 raw_inode->i_gid_high = 0;
5052 }
5053 } else {
5054 raw_inode->i_uid_low =
5055 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5056 raw_inode->i_gid_low =
5057 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5058 raw_inode->i_uid_high = 0;
5059 raw_inode->i_gid_high = 0;
5060 }
5061 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5062
5063 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5064 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5065 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5066 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5067
5068 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5069 goto out_brelse;
5070 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5071 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5072 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5073 cpu_to_le32(EXT4_OS_HURD))
5074 raw_inode->i_file_acl_high =
5075 cpu_to_le16(ei->i_file_acl >> 32);
5076 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5077 ext4_isize_set(raw_inode, ei->i_disksize);
5078 if (ei->i_disksize > 0x7fffffffULL) {
5079 struct super_block *sb = inode->i_sb;
5080 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5081 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5082 EXT4_SB(sb)->s_es->s_rev_level ==
5083 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5084 /* If this is the first large file
5085 * created, add a flag to the superblock.
5086 */
5087 err = ext4_journal_get_write_access(handle,
5088 EXT4_SB(sb)->s_sbh);
5089 if (err)
5090 goto out_brelse;
5091 ext4_update_dynamic_rev(sb);
5092 EXT4_SET_RO_COMPAT_FEATURE(sb,
5093 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5094 sb->s_dirt = 1;
5095 ext4_handle_sync(handle);
5096 err = ext4_handle_dirty_metadata(handle, inode,
5097 EXT4_SB(sb)->s_sbh);
5098 }
5099 }
5100 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5101 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5102 if (old_valid_dev(inode->i_rdev)) {
5103 raw_inode->i_block[0] =
5104 cpu_to_le32(old_encode_dev(inode->i_rdev));
5105 raw_inode->i_block[1] = 0;
5106 } else {
5107 raw_inode->i_block[0] = 0;
5108 raw_inode->i_block[1] =
5109 cpu_to_le32(new_encode_dev(inode->i_rdev));
5110 raw_inode->i_block[2] = 0;
5111 }
5112 } else
5113 for (block = 0; block < EXT4_N_BLOCKS; block++)
5114 raw_inode->i_block[block] = ei->i_data[block];
5115
5116 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5117 if (ei->i_extra_isize) {
5118 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5119 raw_inode->i_version_hi =
5120 cpu_to_le32(inode->i_version >> 32);
5121 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5122 }
5123
5124 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5125 rc = ext4_handle_dirty_metadata(handle, inode, bh);
5126 if (!err)
5127 err = rc;
5128 ei->i_state &= ~EXT4_STATE_NEW;
5129
5130 ext4_update_inode_fsync_trans(handle, inode, 0);
5131 out_brelse:
5132 brelse(bh);
5133 ext4_std_error(inode->i_sb, err);
5134 return err;
5135 }
5136
5137 /*
5138 * ext4_write_inode()
5139 *
5140 * We are called from a few places:
5141 *
5142 * - Within generic_file_write() for O_SYNC files.
5143 * Here, there will be no transaction running. We wait for any running
5144 * trasnaction to commit.
5145 *
5146 * - Within sys_sync(), kupdate and such.
5147 * We wait on commit, if tol to.
5148 *
5149 * - Within prune_icache() (PF_MEMALLOC == true)
5150 * Here we simply return. We can't afford to block kswapd on the
5151 * journal commit.
5152 *
5153 * In all cases it is actually safe for us to return without doing anything,
5154 * because the inode has been copied into a raw inode buffer in
5155 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5156 * knfsd.
5157 *
5158 * Note that we are absolutely dependent upon all inode dirtiers doing the
5159 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5160 * which we are interested.
5161 *
5162 * It would be a bug for them to not do this. The code:
5163 *
5164 * mark_inode_dirty(inode)
5165 * stuff();
5166 * inode->i_size = expr;
5167 *
5168 * is in error because a kswapd-driven write_inode() could occur while
5169 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5170 * will no longer be on the superblock's dirty inode list.
5171 */
5172 int ext4_write_inode(struct inode *inode, int wait)
5173 {
5174 int err;
5175
5176 if (current->flags & PF_MEMALLOC)
5177 return 0;
5178
5179 if (EXT4_SB(inode->i_sb)->s_journal) {
5180 if (ext4_journal_current_handle()) {
5181 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5182 dump_stack();
5183 return -EIO;
5184 }
5185
5186 if (!wait)
5187 return 0;
5188
5189 err = ext4_force_commit(inode->i_sb);
5190 } else {
5191 struct ext4_iloc iloc;
5192
5193 err = ext4_get_inode_loc(inode, &iloc);
5194 if (err)
5195 return err;
5196 if (wait)
5197 sync_dirty_buffer(iloc.bh);
5198 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5199 ext4_error(inode->i_sb, __func__,
5200 "IO error syncing inode, "
5201 "inode=%lu, block=%llu",
5202 inode->i_ino,
5203 (unsigned long long)iloc.bh->b_blocknr);
5204 err = -EIO;
5205 }
5206 }
5207 return err;
5208 }
5209
5210 /*
5211 * ext4_setattr()
5212 *
5213 * Called from notify_change.
5214 *
5215 * We want to trap VFS attempts to truncate the file as soon as
5216 * possible. In particular, we want to make sure that when the VFS
5217 * shrinks i_size, we put the inode on the orphan list and modify
5218 * i_disksize immediately, so that during the subsequent flushing of
5219 * dirty pages and freeing of disk blocks, we can guarantee that any
5220 * commit will leave the blocks being flushed in an unused state on
5221 * disk. (On recovery, the inode will get truncated and the blocks will
5222 * be freed, so we have a strong guarantee that no future commit will
5223 * leave these blocks visible to the user.)
5224 *
5225 * Another thing we have to assure is that if we are in ordered mode
5226 * and inode is still attached to the committing transaction, we must
5227 * we start writeout of all the dirty pages which are being truncated.
5228 * This way we are sure that all the data written in the previous
5229 * transaction are already on disk (truncate waits for pages under
5230 * writeback).
5231 *
5232 * Called with inode->i_mutex down.
5233 */
5234 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5235 {
5236 struct inode *inode = dentry->d_inode;
5237 int error, rc = 0;
5238 const unsigned int ia_valid = attr->ia_valid;
5239
5240 error = inode_change_ok(inode, attr);
5241 if (error)
5242 return error;
5243
5244 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5245 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5246 handle_t *handle;
5247
5248 /* (user+group)*(old+new) structure, inode write (sb,
5249 * inode block, ? - but truncate inode update has it) */
5250 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5251 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5252 if (IS_ERR(handle)) {
5253 error = PTR_ERR(handle);
5254 goto err_out;
5255 }
5256 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
5257 if (error) {
5258 ext4_journal_stop(handle);
5259 return error;
5260 }
5261 /* Update corresponding info in inode so that everything is in
5262 * one transaction */
5263 if (attr->ia_valid & ATTR_UID)
5264 inode->i_uid = attr->ia_uid;
5265 if (attr->ia_valid & ATTR_GID)
5266 inode->i_gid = attr->ia_gid;
5267 error = ext4_mark_inode_dirty(handle, inode);
5268 ext4_journal_stop(handle);
5269 }
5270
5271 if (attr->ia_valid & ATTR_SIZE) {
5272 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
5273 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5274
5275 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5276 error = -EFBIG;
5277 goto err_out;
5278 }
5279 }
5280 }
5281
5282 if (S_ISREG(inode->i_mode) &&
5283 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
5284 handle_t *handle;
5285
5286 handle = ext4_journal_start(inode, 3);
5287 if (IS_ERR(handle)) {
5288 error = PTR_ERR(handle);
5289 goto err_out;
5290 }
5291
5292 error = ext4_orphan_add(handle, inode);
5293 EXT4_I(inode)->i_disksize = attr->ia_size;
5294 rc = ext4_mark_inode_dirty(handle, inode);
5295 if (!error)
5296 error = rc;
5297 ext4_journal_stop(handle);
5298
5299 if (ext4_should_order_data(inode)) {
5300 error = ext4_begin_ordered_truncate(inode,
5301 attr->ia_size);
5302 if (error) {
5303 /* Do as much error cleanup as possible */
5304 handle = ext4_journal_start(inode, 3);
5305 if (IS_ERR(handle)) {
5306 ext4_orphan_del(NULL, inode);
5307 goto err_out;
5308 }
5309 ext4_orphan_del(handle, inode);
5310 ext4_journal_stop(handle);
5311 goto err_out;
5312 }
5313 }
5314 }
5315
5316 rc = inode_setattr(inode, attr);
5317
5318 /* If inode_setattr's call to ext4_truncate failed to get a
5319 * transaction handle at all, we need to clean up the in-core
5320 * orphan list manually. */
5321 if (inode->i_nlink)
5322 ext4_orphan_del(NULL, inode);
5323
5324 if (!rc && (ia_valid & ATTR_MODE))
5325 rc = ext4_acl_chmod(inode);
5326
5327 err_out:
5328 ext4_std_error(inode->i_sb, error);
5329 if (!error)
5330 error = rc;
5331 return error;
5332 }
5333
5334 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5335 struct kstat *stat)
5336 {
5337 struct inode *inode;
5338 unsigned long delalloc_blocks;
5339
5340 inode = dentry->d_inode;
5341 generic_fillattr(inode, stat);
5342
5343 /*
5344 * We can't update i_blocks if the block allocation is delayed
5345 * otherwise in the case of system crash before the real block
5346 * allocation is done, we will have i_blocks inconsistent with
5347 * on-disk file blocks.
5348 * We always keep i_blocks updated together with real
5349 * allocation. But to not confuse with user, stat
5350 * will return the blocks that include the delayed allocation
5351 * blocks for this file.
5352 */
5353 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5354 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5355 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5356
5357 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5358 return 0;
5359 }
5360
5361 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5362 int chunk)
5363 {
5364 int indirects;
5365
5366 /* if nrblocks are contiguous */
5367 if (chunk) {
5368 /*
5369 * With N contiguous data blocks, it need at most
5370 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5371 * 2 dindirect blocks
5372 * 1 tindirect block
5373 */
5374 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5375 return indirects + 3;
5376 }
5377 /*
5378 * if nrblocks are not contiguous, worse case, each block touch
5379 * a indirect block, and each indirect block touch a double indirect
5380 * block, plus a triple indirect block
5381 */
5382 indirects = nrblocks * 2 + 1;
5383 return indirects;
5384 }
5385
5386 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5387 {
5388 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
5389 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5390 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5391 }
5392
5393 /*
5394 * Account for index blocks, block groups bitmaps and block group
5395 * descriptor blocks if modify datablocks and index blocks
5396 * worse case, the indexs blocks spread over different block groups
5397 *
5398 * If datablocks are discontiguous, they are possible to spread over
5399 * different block groups too. If they are contiugous, with flexbg,
5400 * they could still across block group boundary.
5401 *
5402 * Also account for superblock, inode, quota and xattr blocks
5403 */
5404 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5405 {
5406 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5407 int gdpblocks;
5408 int idxblocks;
5409 int ret = 0;
5410
5411 /*
5412 * How many index blocks need to touch to modify nrblocks?
5413 * The "Chunk" flag indicating whether the nrblocks is
5414 * physically contiguous on disk
5415 *
5416 * For Direct IO and fallocate, they calls get_block to allocate
5417 * one single extent at a time, so they could set the "Chunk" flag
5418 */
5419 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5420
5421 ret = idxblocks;
5422
5423 /*
5424 * Now let's see how many group bitmaps and group descriptors need
5425 * to account
5426 */
5427 groups = idxblocks;
5428 if (chunk)
5429 groups += 1;
5430 else
5431 groups += nrblocks;
5432
5433 gdpblocks = groups;
5434 if (groups > ngroups)
5435 groups = ngroups;
5436 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5437 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5438
5439 /* bitmaps and block group descriptor blocks */
5440 ret += groups + gdpblocks;
5441
5442 /* Blocks for super block, inode, quota and xattr blocks */
5443 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5444
5445 return ret;
5446 }
5447
5448 /*
5449 * Calulate the total number of credits to reserve to fit
5450 * the modification of a single pages into a single transaction,
5451 * which may include multiple chunks of block allocations.
5452 *
5453 * This could be called via ext4_write_begin()
5454 *
5455 * We need to consider the worse case, when
5456 * one new block per extent.
5457 */
5458 int ext4_writepage_trans_blocks(struct inode *inode)
5459 {
5460 int bpp = ext4_journal_blocks_per_page(inode);
5461 int ret;
5462
5463 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5464
5465 /* Account for data blocks for journalled mode */
5466 if (ext4_should_journal_data(inode))
5467 ret += bpp;
5468 return ret;
5469 }
5470
5471 /*
5472 * Calculate the journal credits for a chunk of data modification.
5473 *
5474 * This is called from DIO, fallocate or whoever calling
5475 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5476 *
5477 * journal buffers for data blocks are not included here, as DIO
5478 * and fallocate do no need to journal data buffers.
5479 */
5480 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5481 {
5482 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5483 }
5484
5485 /*
5486 * The caller must have previously called ext4_reserve_inode_write().
5487 * Give this, we know that the caller already has write access to iloc->bh.
5488 */
5489 int ext4_mark_iloc_dirty(handle_t *handle,
5490 struct inode *inode, struct ext4_iloc *iloc)
5491 {
5492 int err = 0;
5493
5494 if (test_opt(inode->i_sb, I_VERSION))
5495 inode_inc_iversion(inode);
5496
5497 /* the do_update_inode consumes one bh->b_count */
5498 get_bh(iloc->bh);
5499
5500 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5501 err = ext4_do_update_inode(handle, inode, iloc);
5502 put_bh(iloc->bh);
5503 return err;
5504 }
5505
5506 /*
5507 * On success, We end up with an outstanding reference count against
5508 * iloc->bh. This _must_ be cleaned up later.
5509 */
5510
5511 int
5512 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5513 struct ext4_iloc *iloc)
5514 {
5515 int err;
5516
5517 err = ext4_get_inode_loc(inode, iloc);
5518 if (!err) {
5519 BUFFER_TRACE(iloc->bh, "get_write_access");
5520 err = ext4_journal_get_write_access(handle, iloc->bh);
5521 if (err) {
5522 brelse(iloc->bh);
5523 iloc->bh = NULL;
5524 }
5525 }
5526 ext4_std_error(inode->i_sb, err);
5527 return err;
5528 }
5529
5530 /*
5531 * Expand an inode by new_extra_isize bytes.
5532 * Returns 0 on success or negative error number on failure.
5533 */
5534 static int ext4_expand_extra_isize(struct inode *inode,
5535 unsigned int new_extra_isize,
5536 struct ext4_iloc iloc,
5537 handle_t *handle)
5538 {
5539 struct ext4_inode *raw_inode;
5540 struct ext4_xattr_ibody_header *header;
5541 struct ext4_xattr_entry *entry;
5542
5543 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5544 return 0;
5545
5546 raw_inode = ext4_raw_inode(&iloc);
5547
5548 header = IHDR(inode, raw_inode);
5549 entry = IFIRST(header);
5550
5551 /* No extended attributes present */
5552 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5553 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5554 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5555 new_extra_isize);
5556 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5557 return 0;
5558 }
5559
5560 /* try to expand with EAs present */
5561 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5562 raw_inode, handle);
5563 }
5564
5565 /*
5566 * What we do here is to mark the in-core inode as clean with respect to inode
5567 * dirtiness (it may still be data-dirty).
5568 * This means that the in-core inode may be reaped by prune_icache
5569 * without having to perform any I/O. This is a very good thing,
5570 * because *any* task may call prune_icache - even ones which
5571 * have a transaction open against a different journal.
5572 *
5573 * Is this cheating? Not really. Sure, we haven't written the
5574 * inode out, but prune_icache isn't a user-visible syncing function.
5575 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5576 * we start and wait on commits.
5577 *
5578 * Is this efficient/effective? Well, we're being nice to the system
5579 * by cleaning up our inodes proactively so they can be reaped
5580 * without I/O. But we are potentially leaving up to five seconds'
5581 * worth of inodes floating about which prune_icache wants us to
5582 * write out. One way to fix that would be to get prune_icache()
5583 * to do a write_super() to free up some memory. It has the desired
5584 * effect.
5585 */
5586 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5587 {
5588 struct ext4_iloc iloc;
5589 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5590 static unsigned int mnt_count;
5591 int err, ret;
5592
5593 might_sleep();
5594 err = ext4_reserve_inode_write(handle, inode, &iloc);
5595 if (ext4_handle_valid(handle) &&
5596 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5597 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5598 /*
5599 * We need extra buffer credits since we may write into EA block
5600 * with this same handle. If journal_extend fails, then it will
5601 * only result in a minor loss of functionality for that inode.
5602 * If this is felt to be critical, then e2fsck should be run to
5603 * force a large enough s_min_extra_isize.
5604 */
5605 if ((jbd2_journal_extend(handle,
5606 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5607 ret = ext4_expand_extra_isize(inode,
5608 sbi->s_want_extra_isize,
5609 iloc, handle);
5610 if (ret) {
5611 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5612 if (mnt_count !=
5613 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5614 ext4_warning(inode->i_sb, __func__,
5615 "Unable to expand inode %lu. Delete"
5616 " some EAs or run e2fsck.",
5617 inode->i_ino);
5618 mnt_count =
5619 le16_to_cpu(sbi->s_es->s_mnt_count);
5620 }
5621 }
5622 }
5623 }
5624 if (!err)
5625 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5626 return err;
5627 }
5628
5629 /*
5630 * ext4_dirty_inode() is called from __mark_inode_dirty()
5631 *
5632 * We're really interested in the case where a file is being extended.
5633 * i_size has been changed by generic_commit_write() and we thus need
5634 * to include the updated inode in the current transaction.
5635 *
5636 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5637 * are allocated to the file.
5638 *
5639 * If the inode is marked synchronous, we don't honour that here - doing
5640 * so would cause a commit on atime updates, which we don't bother doing.
5641 * We handle synchronous inodes at the highest possible level.
5642 */
5643 void ext4_dirty_inode(struct inode *inode)
5644 {
5645 handle_t *current_handle = ext4_journal_current_handle();
5646 handle_t *handle;
5647
5648 handle = ext4_journal_start(inode, 2);
5649 if (IS_ERR(handle))
5650 goto out;
5651
5652 jbd_debug(5, "marking dirty. outer handle=%p\n", current_handle);
5653 ext4_mark_inode_dirty(handle, inode);
5654
5655 ext4_journal_stop(handle);
5656 out:
5657 return;
5658 }
5659
5660 #if 0
5661 /*
5662 * Bind an inode's backing buffer_head into this transaction, to prevent
5663 * it from being flushed to disk early. Unlike
5664 * ext4_reserve_inode_write, this leaves behind no bh reference and
5665 * returns no iloc structure, so the caller needs to repeat the iloc
5666 * lookup to mark the inode dirty later.
5667 */
5668 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5669 {
5670 struct ext4_iloc iloc;
5671
5672 int err = 0;
5673 if (handle) {
5674 err = ext4_get_inode_loc(inode, &iloc);
5675 if (!err) {
5676 BUFFER_TRACE(iloc.bh, "get_write_access");
5677 err = jbd2_journal_get_write_access(handle, iloc.bh);
5678 if (!err)
5679 err = ext4_handle_dirty_metadata(handle,
5680 inode,
5681 iloc.bh);
5682 brelse(iloc.bh);
5683 }
5684 }
5685 ext4_std_error(inode->i_sb, err);
5686 return err;
5687 }
5688 #endif
5689
5690 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5691 {
5692 journal_t *journal;
5693 handle_t *handle;
5694 int err;
5695
5696 /*
5697 * We have to be very careful here: changing a data block's
5698 * journaling status dynamically is dangerous. If we write a
5699 * data block to the journal, change the status and then delete
5700 * that block, we risk forgetting to revoke the old log record
5701 * from the journal and so a subsequent replay can corrupt data.
5702 * So, first we make sure that the journal is empty and that
5703 * nobody is changing anything.
5704 */
5705
5706 journal = EXT4_JOURNAL(inode);
5707 if (!journal)
5708 return 0;
5709 if (is_journal_aborted(journal))
5710 return -EROFS;
5711
5712 jbd2_journal_lock_updates(journal);
5713 jbd2_journal_flush(journal);
5714
5715 /*
5716 * OK, there are no updates running now, and all cached data is
5717 * synced to disk. We are now in a completely consistent state
5718 * which doesn't have anything in the journal, and we know that
5719 * no filesystem updates are running, so it is safe to modify
5720 * the inode's in-core data-journaling state flag now.
5721 */
5722
5723 if (val)
5724 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5725 else
5726 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5727 ext4_set_aops(inode);
5728
5729 jbd2_journal_unlock_updates(journal);
5730
5731 /* Finally we can mark the inode as dirty. */
5732
5733 handle = ext4_journal_start(inode, 1);
5734 if (IS_ERR(handle))
5735 return PTR_ERR(handle);
5736
5737 err = ext4_mark_inode_dirty(handle, inode);
5738 ext4_handle_sync(handle);
5739 ext4_journal_stop(handle);
5740 ext4_std_error(inode->i_sb, err);
5741
5742 return err;
5743 }
5744
5745 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5746 {
5747 return !buffer_mapped(bh);
5748 }
5749
5750 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5751 {
5752 struct page *page = vmf->page;
5753 loff_t size;
5754 unsigned long len;
5755 int ret = -EINVAL;
5756 void *fsdata;
5757 struct file *file = vma->vm_file;
5758 struct inode *inode = file->f_path.dentry->d_inode;
5759 struct address_space *mapping = inode->i_mapping;
5760
5761 /*
5762 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5763 * get i_mutex because we are already holding mmap_sem.
5764 */
5765 down_read(&inode->i_alloc_sem);
5766 size = i_size_read(inode);
5767 if (page->mapping != mapping || size <= page_offset(page)
5768 || !PageUptodate(page)) {
5769 /* page got truncated from under us? */
5770 goto out_unlock;
5771 }
5772 ret = 0;
5773 if (PageMappedToDisk(page))
5774 goto out_unlock;
5775
5776 if (page->index == size >> PAGE_CACHE_SHIFT)
5777 len = size & ~PAGE_CACHE_MASK;
5778 else
5779 len = PAGE_CACHE_SIZE;
5780
5781 lock_page(page);
5782 /*
5783 * return if we have all the buffers mapped. This avoid
5784 * the need to call write_begin/write_end which does a
5785 * journal_start/journal_stop which can block and take
5786 * long time
5787 */
5788 if (page_has_buffers(page)) {
5789 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5790 ext4_bh_unmapped)) {
5791 unlock_page(page);
5792 goto out_unlock;
5793 }
5794 }
5795 unlock_page(page);
5796 /*
5797 * OK, we need to fill the hole... Do write_begin write_end
5798 * to do block allocation/reservation.We are not holding
5799 * inode.i__mutex here. That allow * parallel write_begin,
5800 * write_end call. lock_page prevent this from happening
5801 * on the same page though
5802 */
5803 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5804 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5805 if (ret < 0)
5806 goto out_unlock;
5807 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5808 len, len, page, fsdata);
5809 if (ret < 0)
5810 goto out_unlock;
5811 ret = 0;
5812 out_unlock:
5813 if (ret)
5814 ret = VM_FAULT_SIGBUS;
5815 up_read(&inode->i_alloc_sem);
5816 return ret;
5817 }
5818
|
This page was automatically generated by the
LXR engine.
|