]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/ext4/inode.c
Merge branch 'pstore-efi' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
[mirror_ubuntu-artful-kernel.git] / fs / ext4 / inode.c
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 #include <linux/kernel.h>
42 #include <linux/printk.h>
43 #include <linux/slab.h>
44 #include <linux/ratelimit.h>
45
46 #include "ext4_jbd2.h"
47 #include "xattr.h"
48 #include "acl.h"
49 #include "ext4_extents.h"
50
51 #include <trace/events/ext4.h>
52
53 #define MPAGE_DA_EXTENT_TAIL 0x01
54
55 static inline int ext4_begin_ordered_truncate(struct inode *inode,
56 loff_t new_size)
57 {
58 trace_ext4_begin_ordered_truncate(inode, new_size);
59 /*
60 * If jinode is zero, then we never opened the file for
61 * writing, so there's no need to call
62 * jbd2_journal_begin_ordered_truncate() since there's no
63 * outstanding writes we need to flush.
64 */
65 if (!EXT4_I(inode)->jinode)
66 return 0;
67 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
68 EXT4_I(inode)->jinode,
69 new_size);
70 }
71
72 static void ext4_invalidatepage(struct page *page, unsigned long offset);
73 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
74 struct buffer_head *bh_result, int create);
75 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
76 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
77 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
78 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
79
80 /*
81 * Test whether an inode is a fast symlink.
82 */
83 static int ext4_inode_is_fast_symlink(struct inode *inode)
84 {
85 int ea_blocks = EXT4_I(inode)->i_file_acl ?
86 (inode->i_sb->s_blocksize >> 9) : 0;
87
88 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
89 }
90
91 /*
92 * Work out how many blocks we need to proceed with the next chunk of a
93 * truncate transaction.
94 */
95 static unsigned long blocks_for_truncate(struct inode *inode)
96 {
97 ext4_lblk_t needed;
98
99 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
100
101 /* Give ourselves just enough room to cope with inodes in which
102 * i_blocks is corrupt: we've seen disk corruptions in the past
103 * which resulted in random data in an inode which looked enough
104 * like a regular file for ext4 to try to delete it. Things
105 * will go a bit crazy if that happens, but at least we should
106 * try not to panic the whole kernel. */
107 if (needed < 2)
108 needed = 2;
109
110 /* But we need to bound the transaction so we don't overflow the
111 * journal. */
112 if (needed > EXT4_MAX_TRANS_DATA)
113 needed = EXT4_MAX_TRANS_DATA;
114
115 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
116 }
117
118 /*
119 * Truncate transactions can be complex and absolutely huge. So we need to
120 * be able to restart the transaction at a conventient checkpoint to make
121 * sure we don't overflow the journal.
122 *
123 * start_transaction gets us a new handle for a truncate transaction,
124 * and extend_transaction tries to extend the existing one a bit. If
125 * extend fails, we need to propagate the failure up and restart the
126 * transaction in the top-level truncate loop. --sct
127 */
128 static handle_t *start_transaction(struct inode *inode)
129 {
130 handle_t *result;
131
132 result = ext4_journal_start(inode, blocks_for_truncate(inode));
133 if (!IS_ERR(result))
134 return result;
135
136 ext4_std_error(inode->i_sb, PTR_ERR(result));
137 return result;
138 }
139
140 /*
141 * Try to extend this transaction for the purposes of truncation.
142 *
143 * Returns 0 if we managed to create more room. If we can't create more
144 * room, and the transaction must be restarted we return 1.
145 */
146 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
147 {
148 if (!ext4_handle_valid(handle))
149 return 0;
150 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
151 return 0;
152 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
153 return 0;
154 return 1;
155 }
156
157 /*
158 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against
160 * this transaction.
161 */
162 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
163 int nblocks)
164 {
165 int ret;
166
167 /*
168 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
169 * moment, get_block can be called only for blocks inside i_size since
170 * page cache has been already dropped and writes are blocked by
171 * i_mutex. So we can safely drop the i_data_sem here.
172 */
173 BUG_ON(EXT4_JOURNAL(inode) == NULL);
174 jbd_debug(2, "restarting handle %p\n", handle);
175 up_write(&EXT4_I(inode)->i_data_sem);
176 ret = ext4_journal_restart(handle, nblocks);
177 down_write(&EXT4_I(inode)->i_data_sem);
178 ext4_discard_preallocations(inode);
179
180 return ret;
181 }
182
183 /*
184 * Called at the last iput() if i_nlink is zero.
185 */
186 void ext4_evict_inode(struct inode *inode)
187 {
188 handle_t *handle;
189 int err;
190
191 trace_ext4_evict_inode(inode);
192 if (inode->i_nlink) {
193 truncate_inode_pages(&inode->i_data, 0);
194 goto no_delete;
195 }
196
197 if (!is_bad_inode(inode))
198 dquot_initialize(inode);
199
200 if (ext4_should_order_data(inode))
201 ext4_begin_ordered_truncate(inode, 0);
202 truncate_inode_pages(&inode->i_data, 0);
203
204 if (is_bad_inode(inode))
205 goto no_delete;
206
207 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
208 if (IS_ERR(handle)) {
209 ext4_std_error(inode->i_sb, PTR_ERR(handle));
210 /*
211 * If we're going to skip the normal cleanup, we still need to
212 * make sure that the in-core orphan linked list is properly
213 * cleaned up.
214 */
215 ext4_orphan_del(NULL, inode);
216 goto no_delete;
217 }
218
219 if (IS_SYNC(inode))
220 ext4_handle_sync(handle);
221 inode->i_size = 0;
222 err = ext4_mark_inode_dirty(handle, inode);
223 if (err) {
224 ext4_warning(inode->i_sb,
225 "couldn't mark inode dirty (err %d)", err);
226 goto stop_handle;
227 }
228 if (inode->i_blocks)
229 ext4_truncate(inode);
230
231 /*
232 * ext4_ext_truncate() doesn't reserve any slop when it
233 * restarts journal transactions; therefore there may not be
234 * enough credits left in the handle to remove the inode from
235 * the orphan list and set the dtime field.
236 */
237 if (!ext4_handle_has_enough_credits(handle, 3)) {
238 err = ext4_journal_extend(handle, 3);
239 if (err > 0)
240 err = ext4_journal_restart(handle, 3);
241 if (err != 0) {
242 ext4_warning(inode->i_sb,
243 "couldn't extend journal (err %d)", err);
244 stop_handle:
245 ext4_journal_stop(handle);
246 ext4_orphan_del(NULL, inode);
247 goto no_delete;
248 }
249 }
250
251 /*
252 * Kill off the orphan record which ext4_truncate created.
253 * AKPM: I think this can be inside the above `if'.
254 * Note that ext4_orphan_del() has to be able to cope with the
255 * deletion of a non-existent orphan - this is because we don't
256 * know if ext4_truncate() actually created an orphan record.
257 * (Well, we could do this if we need to, but heck - it works)
258 */
259 ext4_orphan_del(handle, inode);
260 EXT4_I(inode)->i_dtime = get_seconds();
261
262 /*
263 * One subtle ordering requirement: if anything has gone wrong
264 * (transaction abort, IO errors, whatever), then we can still
265 * do these next steps (the fs will already have been marked as
266 * having errors), but we can't free the inode if the mark_dirty
267 * fails.
268 */
269 if (ext4_mark_inode_dirty(handle, inode))
270 /* If that failed, just do the required in-core inode clear. */
271 ext4_clear_inode(inode);
272 else
273 ext4_free_inode(handle, inode);
274 ext4_journal_stop(handle);
275 return;
276 no_delete:
277 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
278 }
279
280 typedef struct {
281 __le32 *p;
282 __le32 key;
283 struct buffer_head *bh;
284 } Indirect;
285
286 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
287 {
288 p->key = *(p->p = v);
289 p->bh = bh;
290 }
291
292 /**
293 * ext4_block_to_path - parse the block number into array of offsets
294 * @inode: inode in question (we are only interested in its superblock)
295 * @i_block: block number to be parsed
296 * @offsets: array to store the offsets in
297 * @boundary: set this non-zero if the referred-to block is likely to be
298 * followed (on disk) by an indirect block.
299 *
300 * To store the locations of file's data ext4 uses a data structure common
301 * for UNIX filesystems - tree of pointers anchored in the inode, with
302 * data blocks at leaves and indirect blocks in intermediate nodes.
303 * This function translates the block number into path in that tree -
304 * return value is the path length and @offsets[n] is the offset of
305 * pointer to (n+1)th node in the nth one. If @block is out of range
306 * (negative or too large) warning is printed and zero returned.
307 *
308 * Note: function doesn't find node addresses, so no IO is needed. All
309 * we need to know is the capacity of indirect blocks (taken from the
310 * inode->i_sb).
311 */
312
313 /*
314 * Portability note: the last comparison (check that we fit into triple
315 * indirect block) is spelled differently, because otherwise on an
316 * architecture with 32-bit longs and 8Kb pages we might get into trouble
317 * if our filesystem had 8Kb blocks. We might use long long, but that would
318 * kill us on x86. Oh, well, at least the sign propagation does not matter -
319 * i_block would have to be negative in the very beginning, so we would not
320 * get there at all.
321 */
322
323 static int ext4_block_to_path(struct inode *inode,
324 ext4_lblk_t i_block,
325 ext4_lblk_t offsets[4], int *boundary)
326 {
327 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
328 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
329 const long direct_blocks = EXT4_NDIR_BLOCKS,
330 indirect_blocks = ptrs,
331 double_blocks = (1 << (ptrs_bits * 2));
332 int n = 0;
333 int final = 0;
334
335 if (i_block < direct_blocks) {
336 offsets[n++] = i_block;
337 final = direct_blocks;
338 } else if ((i_block -= direct_blocks) < indirect_blocks) {
339 offsets[n++] = EXT4_IND_BLOCK;
340 offsets[n++] = i_block;
341 final = ptrs;
342 } else if ((i_block -= indirect_blocks) < double_blocks) {
343 offsets[n++] = EXT4_DIND_BLOCK;
344 offsets[n++] = i_block >> ptrs_bits;
345 offsets[n++] = i_block & (ptrs - 1);
346 final = ptrs;
347 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
348 offsets[n++] = EXT4_TIND_BLOCK;
349 offsets[n++] = i_block >> (ptrs_bits * 2);
350 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
351 offsets[n++] = i_block & (ptrs - 1);
352 final = ptrs;
353 } else {
354 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
355 i_block + direct_blocks +
356 indirect_blocks + double_blocks, inode->i_ino);
357 }
358 if (boundary)
359 *boundary = final - 1 - (i_block & (ptrs - 1));
360 return n;
361 }
362
363 static int __ext4_check_blockref(const char *function, unsigned int line,
364 struct inode *inode,
365 __le32 *p, unsigned int max)
366 {
367 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
368 __le32 *bref = p;
369 unsigned int blk;
370
371 while (bref < p+max) {
372 blk = le32_to_cpu(*bref++);
373 if (blk &&
374 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
375 blk, 1))) {
376 es->s_last_error_block = cpu_to_le64(blk);
377 ext4_error_inode(inode, function, line, blk,
378 "invalid block");
379 return -EIO;
380 }
381 }
382 return 0;
383 }
384
385
386 #define ext4_check_indirect_blockref(inode, bh) \
387 __ext4_check_blockref(__func__, __LINE__, inode, \
388 (__le32 *)(bh)->b_data, \
389 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
390
391 #define ext4_check_inode_blockref(inode) \
392 __ext4_check_blockref(__func__, __LINE__, inode, \
393 EXT4_I(inode)->i_data, \
394 EXT4_NDIR_BLOCKS)
395
396 /**
397 * ext4_get_branch - read the chain of indirect blocks leading to data
398 * @inode: inode in question
399 * @depth: depth of the chain (1 - direct pointer, etc.)
400 * @offsets: offsets of pointers in inode/indirect blocks
401 * @chain: place to store the result
402 * @err: here we store the error value
403 *
404 * Function fills the array of triples <key, p, bh> and returns %NULL
405 * if everything went OK or the pointer to the last filled triple
406 * (incomplete one) otherwise. Upon the return chain[i].key contains
407 * the number of (i+1)-th block in the chain (as it is stored in memory,
408 * i.e. little-endian 32-bit), chain[i].p contains the address of that
409 * number (it points into struct inode for i==0 and into the bh->b_data
410 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
411 * block for i>0 and NULL for i==0. In other words, it holds the block
412 * numbers of the chain, addresses they were taken from (and where we can
413 * verify that chain did not change) and buffer_heads hosting these
414 * numbers.
415 *
416 * Function stops when it stumbles upon zero pointer (absent block)
417 * (pointer to last triple returned, *@err == 0)
418 * or when it gets an IO error reading an indirect block
419 * (ditto, *@err == -EIO)
420 * or when it reads all @depth-1 indirect blocks successfully and finds
421 * the whole chain, all way to the data (returns %NULL, *err == 0).
422 *
423 * Need to be called with
424 * down_read(&EXT4_I(inode)->i_data_sem)
425 */
426 static Indirect *ext4_get_branch(struct inode *inode, int depth,
427 ext4_lblk_t *offsets,
428 Indirect chain[4], int *err)
429 {
430 struct super_block *sb = inode->i_sb;
431 Indirect *p = chain;
432 struct buffer_head *bh;
433
434 *err = 0;
435 /* i_data is not going away, no lock needed */
436 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
437 if (!p->key)
438 goto no_block;
439 while (--depth) {
440 bh = sb_getblk(sb, le32_to_cpu(p->key));
441 if (unlikely(!bh))
442 goto failure;
443
444 if (!bh_uptodate_or_lock(bh)) {
445 if (bh_submit_read(bh) < 0) {
446 put_bh(bh);
447 goto failure;
448 }
449 /* validate block references */
450 if (ext4_check_indirect_blockref(inode, bh)) {
451 put_bh(bh);
452 goto failure;
453 }
454 }
455
456 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
457 /* Reader: end */
458 if (!p->key)
459 goto no_block;
460 }
461 return NULL;
462
463 failure:
464 *err = -EIO;
465 no_block:
466 return p;
467 }
468
469 /**
470 * ext4_find_near - find a place for allocation with sufficient locality
471 * @inode: owner
472 * @ind: descriptor of indirect block.
473 *
474 * This function returns the preferred place for block allocation.
475 * It is used when heuristic for sequential allocation fails.
476 * Rules are:
477 * + if there is a block to the left of our position - allocate near it.
478 * + if pointer will live in indirect block - allocate near that block.
479 * + if pointer will live in inode - allocate in the same
480 * cylinder group.
481 *
482 * In the latter case we colour the starting block by the callers PID to
483 * prevent it from clashing with concurrent allocations for a different inode
484 * in the same block group. The PID is used here so that functionally related
485 * files will be close-by on-disk.
486 *
487 * Caller must make sure that @ind is valid and will stay that way.
488 */
489 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
490 {
491 struct ext4_inode_info *ei = EXT4_I(inode);
492 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
493 __le32 *p;
494 ext4_fsblk_t bg_start;
495 ext4_fsblk_t last_block;
496 ext4_grpblk_t colour;
497 ext4_group_t block_group;
498 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
499
500 /* Try to find previous block */
501 for (p = ind->p - 1; p >= start; p--) {
502 if (*p)
503 return le32_to_cpu(*p);
504 }
505
506 /* No such thing, so let's try location of indirect block */
507 if (ind->bh)
508 return ind->bh->b_blocknr;
509
510 /*
511 * It is going to be referred to from the inode itself? OK, just put it
512 * into the same cylinder group then.
513 */
514 block_group = ei->i_block_group;
515 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
516 block_group &= ~(flex_size-1);
517 if (S_ISREG(inode->i_mode))
518 block_group++;
519 }
520 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
521 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
522
523 /*
524 * If we are doing delayed allocation, we don't need take
525 * colour into account.
526 */
527 if (test_opt(inode->i_sb, DELALLOC))
528 return bg_start;
529
530 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
531 colour = (current->pid % 16) *
532 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
533 else
534 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
535 return bg_start + colour;
536 }
537
538 /**
539 * ext4_find_goal - find a preferred place for allocation.
540 * @inode: owner
541 * @block: block we want
542 * @partial: pointer to the last triple within a chain
543 *
544 * Normally this function find the preferred place for block allocation,
545 * returns it.
546 * Because this is only used for non-extent files, we limit the block nr
547 * to 32 bits.
548 */
549 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
550 Indirect *partial)
551 {
552 ext4_fsblk_t goal;
553
554 /*
555 * XXX need to get goal block from mballoc's data structures
556 */
557
558 goal = ext4_find_near(inode, partial);
559 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
560 return goal;
561 }
562
563 /**
564 * ext4_blks_to_allocate - Look up the block map and count the number
565 * of direct blocks need to be allocated for the given branch.
566 *
567 * @branch: chain of indirect blocks
568 * @k: number of blocks need for indirect blocks
569 * @blks: number of data blocks to be mapped.
570 * @blocks_to_boundary: the offset in the indirect block
571 *
572 * return the total number of blocks to be allocate, including the
573 * direct and indirect blocks.
574 */
575 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
576 int blocks_to_boundary)
577 {
578 unsigned int count = 0;
579
580 /*
581 * Simple case, [t,d]Indirect block(s) has not allocated yet
582 * then it's clear blocks on that path have not allocated
583 */
584 if (k > 0) {
585 /* right now we don't handle cross boundary allocation */
586 if (blks < blocks_to_boundary + 1)
587 count += blks;
588 else
589 count += blocks_to_boundary + 1;
590 return count;
591 }
592
593 count++;
594 while (count < blks && count <= blocks_to_boundary &&
595 le32_to_cpu(*(branch[0].p + count)) == 0) {
596 count++;
597 }
598 return count;
599 }
600
601 /**
602 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
603 * @handle: handle for this transaction
604 * @inode: inode which needs allocated blocks
605 * @iblock: the logical block to start allocated at
606 * @goal: preferred physical block of allocation
607 * @indirect_blks: the number of blocks need to allocate for indirect
608 * blocks
609 * @blks: number of desired blocks
610 * @new_blocks: on return it will store the new block numbers for
611 * the indirect blocks(if needed) and the first direct block,
612 * @err: on return it will store the error code
613 *
614 * This function will return the number of blocks allocated as
615 * requested by the passed-in parameters.
616 */
617 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
618 ext4_lblk_t iblock, ext4_fsblk_t goal,
619 int indirect_blks, int blks,
620 ext4_fsblk_t new_blocks[4], int *err)
621 {
622 struct ext4_allocation_request ar;
623 int target, i;
624 unsigned long count = 0, blk_allocated = 0;
625 int index = 0;
626 ext4_fsblk_t current_block = 0;
627 int ret = 0;
628
629 /*
630 * Here we try to allocate the requested multiple blocks at once,
631 * on a best-effort basis.
632 * To build a branch, we should allocate blocks for
633 * the indirect blocks(if not allocated yet), and at least
634 * the first direct block of this branch. That's the
635 * minimum number of blocks need to allocate(required)
636 */
637 /* first we try to allocate the indirect blocks */
638 target = indirect_blks;
639 while (target > 0) {
640 count = target;
641 /* allocating blocks for indirect blocks and direct blocks */
642 current_block = ext4_new_meta_blocks(handle, inode, goal,
643 0, &count, err);
644 if (*err)
645 goto failed_out;
646
647 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
648 EXT4_ERROR_INODE(inode,
649 "current_block %llu + count %lu > %d!",
650 current_block, count,
651 EXT4_MAX_BLOCK_FILE_PHYS);
652 *err = -EIO;
653 goto failed_out;
654 }
655
656 target -= count;
657 /* allocate blocks for indirect blocks */
658 while (index < indirect_blks && count) {
659 new_blocks[index++] = current_block++;
660 count--;
661 }
662 if (count > 0) {
663 /*
664 * save the new block number
665 * for the first direct block
666 */
667 new_blocks[index] = current_block;
668 printk(KERN_INFO "%s returned more blocks than "
669 "requested\n", __func__);
670 WARN_ON(1);
671 break;
672 }
673 }
674
675 target = blks - count ;
676 blk_allocated = count;
677 if (!target)
678 goto allocated;
679 /* Now allocate data blocks */
680 memset(&ar, 0, sizeof(ar));
681 ar.inode = inode;
682 ar.goal = goal;
683 ar.len = target;
684 ar.logical = iblock;
685 if (S_ISREG(inode->i_mode))
686 /* enable in-core preallocation only for regular files */
687 ar.flags = EXT4_MB_HINT_DATA;
688
689 current_block = ext4_mb_new_blocks(handle, &ar, err);
690 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
691 EXT4_ERROR_INODE(inode,
692 "current_block %llu + ar.len %d > %d!",
693 current_block, ar.len,
694 EXT4_MAX_BLOCK_FILE_PHYS);
695 *err = -EIO;
696 goto failed_out;
697 }
698
699 if (*err && (target == blks)) {
700 /*
701 * if the allocation failed and we didn't allocate
702 * any blocks before
703 */
704 goto failed_out;
705 }
706 if (!*err) {
707 if (target == blks) {
708 /*
709 * save the new block number
710 * for the first direct block
711 */
712 new_blocks[index] = current_block;
713 }
714 blk_allocated += ar.len;
715 }
716 allocated:
717 /* total number of blocks allocated for direct blocks */
718 ret = blk_allocated;
719 *err = 0;
720 return ret;
721 failed_out:
722 for (i = 0; i < index; i++)
723 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
724 return ret;
725 }
726
727 /**
728 * ext4_alloc_branch - allocate and set up a chain of blocks.
729 * @handle: handle for this transaction
730 * @inode: owner
731 * @indirect_blks: number of allocated indirect blocks
732 * @blks: number of allocated direct blocks
733 * @goal: preferred place for allocation
734 * @offsets: offsets (in the blocks) to store the pointers to next.
735 * @branch: place to store the chain in.
736 *
737 * This function allocates blocks, zeroes out all but the last one,
738 * links them into chain and (if we are synchronous) writes them to disk.
739 * In other words, it prepares a branch that can be spliced onto the
740 * inode. It stores the information about that chain in the branch[], in
741 * the same format as ext4_get_branch() would do. We are calling it after
742 * we had read the existing part of chain and partial points to the last
743 * triple of that (one with zero ->key). Upon the exit we have the same
744 * picture as after the successful ext4_get_block(), except that in one
745 * place chain is disconnected - *branch->p is still zero (we did not
746 * set the last link), but branch->key contains the number that should
747 * be placed into *branch->p to fill that gap.
748 *
749 * If allocation fails we free all blocks we've allocated (and forget
750 * their buffer_heads) and return the error value the from failed
751 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
752 * as described above and return 0.
753 */
754 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
755 ext4_lblk_t iblock, int indirect_blks,
756 int *blks, ext4_fsblk_t goal,
757 ext4_lblk_t *offsets, Indirect *branch)
758 {
759 int blocksize = inode->i_sb->s_blocksize;
760 int i, n = 0;
761 int err = 0;
762 struct buffer_head *bh;
763 int num;
764 ext4_fsblk_t new_blocks[4];
765 ext4_fsblk_t current_block;
766
767 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
768 *blks, new_blocks, &err);
769 if (err)
770 return err;
771
772 branch[0].key = cpu_to_le32(new_blocks[0]);
773 /*
774 * metadata blocks and data blocks are allocated.
775 */
776 for (n = 1; n <= indirect_blks; n++) {
777 /*
778 * Get buffer_head for parent block, zero it out
779 * and set the pointer to new one, then send
780 * parent to disk.
781 */
782 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
783 if (unlikely(!bh)) {
784 err = -EIO;
785 goto failed;
786 }
787
788 branch[n].bh = bh;
789 lock_buffer(bh);
790 BUFFER_TRACE(bh, "call get_create_access");
791 err = ext4_journal_get_create_access(handle, bh);
792 if (err) {
793 /* Don't brelse(bh) here; it's done in
794 * ext4_journal_forget() below */
795 unlock_buffer(bh);
796 goto failed;
797 }
798
799 memset(bh->b_data, 0, blocksize);
800 branch[n].p = (__le32 *) bh->b_data + offsets[n];
801 branch[n].key = cpu_to_le32(new_blocks[n]);
802 *branch[n].p = branch[n].key;
803 if (n == indirect_blks) {
804 current_block = new_blocks[n];
805 /*
806 * End of chain, update the last new metablock of
807 * the chain to point to the new allocated
808 * data blocks numbers
809 */
810 for (i = 1; i < num; i++)
811 *(branch[n].p + i) = cpu_to_le32(++current_block);
812 }
813 BUFFER_TRACE(bh, "marking uptodate");
814 set_buffer_uptodate(bh);
815 unlock_buffer(bh);
816
817 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
818 err = ext4_handle_dirty_metadata(handle, inode, bh);
819 if (err)
820 goto failed;
821 }
822 *blks = num;
823 return err;
824 failed:
825 /* Allocation failed, free what we already allocated */
826 ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
827 for (i = 1; i <= n ; i++) {
828 /*
829 * branch[i].bh is newly allocated, so there is no
830 * need to revoke the block, which is why we don't
831 * need to set EXT4_FREE_BLOCKS_METADATA.
832 */
833 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
834 EXT4_FREE_BLOCKS_FORGET);
835 }
836 for (i = n+1; i < indirect_blks; i++)
837 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
838
839 ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
840
841 return err;
842 }
843
844 /**
845 * ext4_splice_branch - splice the allocated branch onto inode.
846 * @handle: handle for this transaction
847 * @inode: owner
848 * @block: (logical) number of block we are adding
849 * @chain: chain of indirect blocks (with a missing link - see
850 * ext4_alloc_branch)
851 * @where: location of missing link
852 * @num: number of indirect blocks we are adding
853 * @blks: number of direct blocks we are adding
854 *
855 * This function fills the missing link and does all housekeeping needed in
856 * inode (->i_blocks, etc.). In case of success we end up with the full
857 * chain to new block and return 0.
858 */
859 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
860 ext4_lblk_t block, Indirect *where, int num,
861 int blks)
862 {
863 int i;
864 int err = 0;
865 ext4_fsblk_t current_block;
866
867 /*
868 * If we're splicing into a [td]indirect block (as opposed to the
869 * inode) then we need to get write access to the [td]indirect block
870 * before the splice.
871 */
872 if (where->bh) {
873 BUFFER_TRACE(where->bh, "get_write_access");
874 err = ext4_journal_get_write_access(handle, where->bh);
875 if (err)
876 goto err_out;
877 }
878 /* That's it */
879
880 *where->p = where->key;
881
882 /*
883 * Update the host buffer_head or inode to point to more just allocated
884 * direct blocks blocks
885 */
886 if (num == 0 && blks > 1) {
887 current_block = le32_to_cpu(where->key) + 1;
888 for (i = 1; i < blks; i++)
889 *(where->p + i) = cpu_to_le32(current_block++);
890 }
891
892 /* We are done with atomic stuff, now do the rest of housekeeping */
893 /* had we spliced it onto indirect block? */
894 if (where->bh) {
895 /*
896 * If we spliced it onto an indirect block, we haven't
897 * altered the inode. Note however that if it is being spliced
898 * onto an indirect block at the very end of the file (the
899 * file is growing) then we *will* alter the inode to reflect
900 * the new i_size. But that is not done here - it is done in
901 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
902 */
903 jbd_debug(5, "splicing indirect only\n");
904 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
905 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
906 if (err)
907 goto err_out;
908 } else {
909 /*
910 * OK, we spliced it into the inode itself on a direct block.
911 */
912 ext4_mark_inode_dirty(handle, inode);
913 jbd_debug(5, "splicing direct\n");
914 }
915 return err;
916
917 err_out:
918 for (i = 1; i <= num; i++) {
919 /*
920 * branch[i].bh is newly allocated, so there is no
921 * need to revoke the block, which is why we don't
922 * need to set EXT4_FREE_BLOCKS_METADATA.
923 */
924 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
925 EXT4_FREE_BLOCKS_FORGET);
926 }
927 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
928 blks, 0);
929
930 return err;
931 }
932
933 /*
934 * The ext4_ind_map_blocks() function handles non-extents inodes
935 * (i.e., using the traditional indirect/double-indirect i_blocks
936 * scheme) for ext4_map_blocks().
937 *
938 * Allocation strategy is simple: if we have to allocate something, we will
939 * have to go the whole way to leaf. So let's do it before attaching anything
940 * to tree, set linkage between the newborn blocks, write them if sync is
941 * required, recheck the path, free and repeat if check fails, otherwise
942 * set the last missing link (that will protect us from any truncate-generated
943 * removals - all blocks on the path are immune now) and possibly force the
944 * write on the parent block.
945 * That has a nice additional property: no special recovery from the failed
946 * allocations is needed - we simply release blocks and do not touch anything
947 * reachable from inode.
948 *
949 * `handle' can be NULL if create == 0.
950 *
951 * return > 0, # of blocks mapped or allocated.
952 * return = 0, if plain lookup failed.
953 * return < 0, error case.
954 *
955 * The ext4_ind_get_blocks() function should be called with
956 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
957 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
958 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
959 * blocks.
960 */
961 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
962 struct ext4_map_blocks *map,
963 int flags)
964 {
965 int err = -EIO;
966 ext4_lblk_t offsets[4];
967 Indirect chain[4];
968 Indirect *partial;
969 ext4_fsblk_t goal;
970 int indirect_blks;
971 int blocks_to_boundary = 0;
972 int depth;
973 int count = 0;
974 ext4_fsblk_t first_block = 0;
975
976 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
977 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
978 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
979 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
980 &blocks_to_boundary);
981
982 if (depth == 0)
983 goto out;
984
985 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
986
987 /* Simplest case - block found, no allocation needed */
988 if (!partial) {
989 first_block = le32_to_cpu(chain[depth - 1].key);
990 count++;
991 /*map more blocks*/
992 while (count < map->m_len && count <= blocks_to_boundary) {
993 ext4_fsblk_t blk;
994
995 blk = le32_to_cpu(*(chain[depth-1].p + count));
996
997 if (blk == first_block + count)
998 count++;
999 else
1000 break;
1001 }
1002 goto got_it;
1003 }
1004
1005 /* Next simple case - plain lookup or failed read of indirect block */
1006 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
1007 goto cleanup;
1008
1009 /*
1010 * Okay, we need to do block allocation.
1011 */
1012 goal = ext4_find_goal(inode, map->m_lblk, partial);
1013
1014 /* the number of blocks need to allocate for [d,t]indirect blocks */
1015 indirect_blks = (chain + depth) - partial - 1;
1016
1017 /*
1018 * Next look up the indirect map to count the totoal number of
1019 * direct blocks to allocate for this branch.
1020 */
1021 count = ext4_blks_to_allocate(partial, indirect_blks,
1022 map->m_len, blocks_to_boundary);
1023 /*
1024 * Block out ext4_truncate while we alter the tree
1025 */
1026 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
1027 &count, goal,
1028 offsets + (partial - chain), partial);
1029
1030 /*
1031 * The ext4_splice_branch call will free and forget any buffers
1032 * on the new chain if there is a failure, but that risks using
1033 * up transaction credits, especially for bitmaps where the
1034 * credits cannot be returned. Can we handle this somehow? We
1035 * may need to return -EAGAIN upwards in the worst case. --sct
1036 */
1037 if (!err)
1038 err = ext4_splice_branch(handle, inode, map->m_lblk,
1039 partial, indirect_blks, count);
1040 if (err)
1041 goto cleanup;
1042
1043 map->m_flags |= EXT4_MAP_NEW;
1044
1045 ext4_update_inode_fsync_trans(handle, inode, 1);
1046 got_it:
1047 map->m_flags |= EXT4_MAP_MAPPED;
1048 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1049 map->m_len = count;
1050 if (count > blocks_to_boundary)
1051 map->m_flags |= EXT4_MAP_BOUNDARY;
1052 err = count;
1053 /* Clean up and exit */
1054 partial = chain + depth - 1; /* the whole chain */
1055 cleanup:
1056 while (partial > chain) {
1057 BUFFER_TRACE(partial->bh, "call brelse");
1058 brelse(partial->bh);
1059 partial--;
1060 }
1061 out:
1062 trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
1063 map->m_pblk, map->m_len, err);
1064 return err;
1065 }
1066
1067 #ifdef CONFIG_QUOTA
1068 qsize_t *ext4_get_reserved_space(struct inode *inode)
1069 {
1070 return &EXT4_I(inode)->i_reserved_quota;
1071 }
1072 #endif
1073
1074 /*
1075 * Calculate the number of metadata blocks need to reserve
1076 * to allocate a new block at @lblocks for non extent file based file
1077 */
1078 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1079 sector_t lblock)
1080 {
1081 struct ext4_inode_info *ei = EXT4_I(inode);
1082 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1083 int blk_bits;
1084
1085 if (lblock < EXT4_NDIR_BLOCKS)
1086 return 0;
1087
1088 lblock -= EXT4_NDIR_BLOCKS;
1089
1090 if (ei->i_da_metadata_calc_len &&
1091 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1092 ei->i_da_metadata_calc_len++;
1093 return 0;
1094 }
1095 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1096 ei->i_da_metadata_calc_len = 1;
1097 blk_bits = order_base_2(lblock);
1098 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1099 }
1100
1101 /*
1102 * Calculate the number of metadata blocks need to reserve
1103 * to allocate a block located at @lblock
1104 */
1105 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
1106 {
1107 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1108 return ext4_ext_calc_metadata_amount(inode, lblock);
1109
1110 return ext4_indirect_calc_metadata_amount(inode, lblock);
1111 }
1112
1113 /*
1114 * Called with i_data_sem down, which is important since we can call
1115 * ext4_discard_preallocations() from here.
1116 */
1117 void ext4_da_update_reserve_space(struct inode *inode,
1118 int used, int quota_claim)
1119 {
1120 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1121 struct ext4_inode_info *ei = EXT4_I(inode);
1122
1123 spin_lock(&ei->i_block_reservation_lock);
1124 trace_ext4_da_update_reserve_space(inode, used);
1125 if (unlikely(used > ei->i_reserved_data_blocks)) {
1126 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1127 "with only %d reserved data blocks\n",
1128 __func__, inode->i_ino, used,
1129 ei->i_reserved_data_blocks);
1130 WARN_ON(1);
1131 used = ei->i_reserved_data_blocks;
1132 }
1133
1134 /* Update per-inode reservations */
1135 ei->i_reserved_data_blocks -= used;
1136 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1137 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1138 used + ei->i_allocated_meta_blocks);
1139 ei->i_allocated_meta_blocks = 0;
1140
1141 if (ei->i_reserved_data_blocks == 0) {
1142 /*
1143 * We can release all of the reserved metadata blocks
1144 * only when we have written all of the delayed
1145 * allocation blocks.
1146 */
1147 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1148 ei->i_reserved_meta_blocks);
1149 ei->i_reserved_meta_blocks = 0;
1150 ei->i_da_metadata_calc_len = 0;
1151 }
1152 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1153
1154 /* Update quota subsystem for data blocks */
1155 if (quota_claim)
1156 dquot_claim_block(inode, used);
1157 else {
1158 /*
1159 * We did fallocate with an offset that is already delayed
1160 * allocated. So on delayed allocated writeback we should
1161 * not re-claim the quota for fallocated blocks.
1162 */
1163 dquot_release_reservation_block(inode, used);
1164 }
1165
1166 /*
1167 * If we have done all the pending block allocations and if
1168 * there aren't any writers on the inode, we can discard the
1169 * inode's preallocations.
1170 */
1171 if ((ei->i_reserved_data_blocks == 0) &&
1172 (atomic_read(&inode->i_writecount) == 0))
1173 ext4_discard_preallocations(inode);
1174 }
1175
1176 static int __check_block_validity(struct inode *inode, const char *func,
1177 unsigned int line,
1178 struct ext4_map_blocks *map)
1179 {
1180 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1181 map->m_len)) {
1182 ext4_error_inode(inode, func, line, map->m_pblk,
1183 "lblock %lu mapped to illegal pblock "
1184 "(length %d)", (unsigned long) map->m_lblk,
1185 map->m_len);
1186 return -EIO;
1187 }
1188 return 0;
1189 }
1190
1191 #define check_block_validity(inode, map) \
1192 __check_block_validity((inode), __func__, __LINE__, (map))
1193
1194 /*
1195 * Return the number of contiguous dirty pages in a given inode
1196 * starting at page frame idx.
1197 */
1198 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1199 unsigned int max_pages)
1200 {
1201 struct address_space *mapping = inode->i_mapping;
1202 pgoff_t index;
1203 struct pagevec pvec;
1204 pgoff_t num = 0;
1205 int i, nr_pages, done = 0;
1206
1207 if (max_pages == 0)
1208 return 0;
1209 pagevec_init(&pvec, 0);
1210 while (!done) {
1211 index = idx;
1212 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1213 PAGECACHE_TAG_DIRTY,
1214 (pgoff_t)PAGEVEC_SIZE);
1215 if (nr_pages == 0)
1216 break;
1217 for (i = 0; i < nr_pages; i++) {
1218 struct page *page = pvec.pages[i];
1219 struct buffer_head *bh, *head;
1220
1221 lock_page(page);
1222 if (unlikely(page->mapping != mapping) ||
1223 !PageDirty(page) ||
1224 PageWriteback(page) ||
1225 page->index != idx) {
1226 done = 1;
1227 unlock_page(page);
1228 break;
1229 }
1230 if (page_has_buffers(page)) {
1231 bh = head = page_buffers(page);
1232 do {
1233 if (!buffer_delay(bh) &&
1234 !buffer_unwritten(bh))
1235 done = 1;
1236 bh = bh->b_this_page;
1237 } while (!done && (bh != head));
1238 }
1239 unlock_page(page);
1240 if (done)
1241 break;
1242 idx++;
1243 num++;
1244 if (num >= max_pages) {
1245 done = 1;
1246 break;
1247 }
1248 }
1249 pagevec_release(&pvec);
1250 }
1251 return num;
1252 }
1253
1254 /*
1255 * The ext4_map_blocks() function tries to look up the requested blocks,
1256 * and returns if the blocks are already mapped.
1257 *
1258 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1259 * and store the allocated blocks in the result buffer head and mark it
1260 * mapped.
1261 *
1262 * If file type is extents based, it will call ext4_ext_map_blocks(),
1263 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1264 * based files
1265 *
1266 * On success, it returns the number of blocks being mapped or allocate.
1267 * if create==0 and the blocks are pre-allocated and uninitialized block,
1268 * the result buffer head is unmapped. If the create ==1, it will make sure
1269 * the buffer head is mapped.
1270 *
1271 * It returns 0 if plain look up failed (blocks have not been allocated), in
1272 * that casem, buffer head is unmapped
1273 *
1274 * It returns the error in case of allocation failure.
1275 */
1276 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1277 struct ext4_map_blocks *map, int flags)
1278 {
1279 int retval;
1280
1281 map->m_flags = 0;
1282 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1283 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1284 (unsigned long) map->m_lblk);
1285 /*
1286 * Try to see if we can get the block without requesting a new
1287 * file system block.
1288 */
1289 down_read((&EXT4_I(inode)->i_data_sem));
1290 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1291 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1292 } else {
1293 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1294 }
1295 up_read((&EXT4_I(inode)->i_data_sem));
1296
1297 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1298 int ret = check_block_validity(inode, map);
1299 if (ret != 0)
1300 return ret;
1301 }
1302
1303 /* If it is only a block(s) look up */
1304 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1305 return retval;
1306
1307 /*
1308 * Returns if the blocks have already allocated
1309 *
1310 * Note that if blocks have been preallocated
1311 * ext4_ext_get_block() returns th create = 0
1312 * with buffer head unmapped.
1313 */
1314 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1315 return retval;
1316
1317 /*
1318 * When we call get_blocks without the create flag, the
1319 * BH_Unwritten flag could have gotten set if the blocks
1320 * requested were part of a uninitialized extent. We need to
1321 * clear this flag now that we are committed to convert all or
1322 * part of the uninitialized extent to be an initialized
1323 * extent. This is because we need to avoid the combination
1324 * of BH_Unwritten and BH_Mapped flags being simultaneously
1325 * set on the buffer_head.
1326 */
1327 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1328
1329 /*
1330 * New blocks allocate and/or writing to uninitialized extent
1331 * will possibly result in updating i_data, so we take
1332 * the write lock of i_data_sem, and call get_blocks()
1333 * with create == 1 flag.
1334 */
1335 down_write((&EXT4_I(inode)->i_data_sem));
1336
1337 /*
1338 * if the caller is from delayed allocation writeout path
1339 * we have already reserved fs blocks for allocation
1340 * let the underlying get_block() function know to
1341 * avoid double accounting
1342 */
1343 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1344 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
1345 /*
1346 * We need to check for EXT4 here because migrate
1347 * could have changed the inode type in between
1348 */
1349 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1350 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1351 } else {
1352 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1353
1354 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1355 /*
1356 * We allocated new blocks which will result in
1357 * i_data's format changing. Force the migrate
1358 * to fail by clearing migrate flags
1359 */
1360 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1361 }
1362
1363 /*
1364 * Update reserved blocks/metadata blocks after successful
1365 * block allocation which had been deferred till now. We don't
1366 * support fallocate for non extent files. So we can update
1367 * reserve space here.
1368 */
1369 if ((retval > 0) &&
1370 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1371 ext4_da_update_reserve_space(inode, retval, 1);
1372 }
1373 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1374 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
1375
1376 up_write((&EXT4_I(inode)->i_data_sem));
1377 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1378 int ret = check_block_validity(inode, map);
1379 if (ret != 0)
1380 return ret;
1381 }
1382 return retval;
1383 }
1384
1385 /* Maximum number of blocks we map for direct IO at once. */
1386 #define DIO_MAX_BLOCKS 4096
1387
1388 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1389 struct buffer_head *bh, int flags)
1390 {
1391 handle_t *handle = ext4_journal_current_handle();
1392 struct ext4_map_blocks map;
1393 int ret = 0, started = 0;
1394 int dio_credits;
1395
1396 map.m_lblk = iblock;
1397 map.m_len = bh->b_size >> inode->i_blkbits;
1398
1399 if (flags && !handle) {
1400 /* Direct IO write... */
1401 if (map.m_len > DIO_MAX_BLOCKS)
1402 map.m_len = DIO_MAX_BLOCKS;
1403 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1404 handle = ext4_journal_start(inode, dio_credits);
1405 if (IS_ERR(handle)) {
1406 ret = PTR_ERR(handle);
1407 return ret;
1408 }
1409 started = 1;
1410 }
1411
1412 ret = ext4_map_blocks(handle, inode, &map, flags);
1413 if (ret > 0) {
1414 map_bh(bh, inode->i_sb, map.m_pblk);
1415 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1416 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1417 ret = 0;
1418 }
1419 if (started)
1420 ext4_journal_stop(handle);
1421 return ret;
1422 }
1423
1424 int ext4_get_block(struct inode *inode, sector_t iblock,
1425 struct buffer_head *bh, int create)
1426 {
1427 return _ext4_get_block(inode, iblock, bh,
1428 create ? EXT4_GET_BLOCKS_CREATE : 0);
1429 }
1430
1431 /*
1432 * `handle' can be NULL if create is zero
1433 */
1434 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1435 ext4_lblk_t block, int create, int *errp)
1436 {
1437 struct ext4_map_blocks map;
1438 struct buffer_head *bh;
1439 int fatal = 0, err;
1440
1441 J_ASSERT(handle != NULL || create == 0);
1442
1443 map.m_lblk = block;
1444 map.m_len = 1;
1445 err = ext4_map_blocks(handle, inode, &map,
1446 create ? EXT4_GET_BLOCKS_CREATE : 0);
1447
1448 if (err < 0)
1449 *errp = err;
1450 if (err <= 0)
1451 return NULL;
1452 *errp = 0;
1453
1454 bh = sb_getblk(inode->i_sb, map.m_pblk);
1455 if (!bh) {
1456 *errp = -EIO;
1457 return NULL;
1458 }
1459 if (map.m_flags & EXT4_MAP_NEW) {
1460 J_ASSERT(create != 0);
1461 J_ASSERT(handle != NULL);
1462
1463 /*
1464 * Now that we do not always journal data, we should
1465 * keep in mind whether this should always journal the
1466 * new buffer as metadata. For now, regular file
1467 * writes use ext4_get_block instead, so it's not a
1468 * problem.
1469 */
1470 lock_buffer(bh);
1471 BUFFER_TRACE(bh, "call get_create_access");
1472 fatal = ext4_journal_get_create_access(handle, bh);
1473 if (!fatal && !buffer_uptodate(bh)) {
1474 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1475 set_buffer_uptodate(bh);
1476 }
1477 unlock_buffer(bh);
1478 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1479 err = ext4_handle_dirty_metadata(handle, inode, bh);
1480 if (!fatal)
1481 fatal = err;
1482 } else {
1483 BUFFER_TRACE(bh, "not a new buffer");
1484 }
1485 if (fatal) {
1486 *errp = fatal;
1487 brelse(bh);
1488 bh = NULL;
1489 }
1490 return bh;
1491 }
1492
1493 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1494 ext4_lblk_t block, int create, int *err)
1495 {
1496 struct buffer_head *bh;
1497
1498 bh = ext4_getblk(handle, inode, block, create, err);
1499 if (!bh)
1500 return bh;
1501 if (buffer_uptodate(bh))
1502 return bh;
1503 ll_rw_block(READ_META, 1, &bh);
1504 wait_on_buffer(bh);
1505 if (buffer_uptodate(bh))
1506 return bh;
1507 put_bh(bh);
1508 *err = -EIO;
1509 return NULL;
1510 }
1511
1512 static int walk_page_buffers(handle_t *handle,
1513 struct buffer_head *head,
1514 unsigned from,
1515 unsigned to,
1516 int *partial,
1517 int (*fn)(handle_t *handle,
1518 struct buffer_head *bh))
1519 {
1520 struct buffer_head *bh;
1521 unsigned block_start, block_end;
1522 unsigned blocksize = head->b_size;
1523 int err, ret = 0;
1524 struct buffer_head *next;
1525
1526 for (bh = head, block_start = 0;
1527 ret == 0 && (bh != head || !block_start);
1528 block_start = block_end, bh = next) {
1529 next = bh->b_this_page;
1530 block_end = block_start + blocksize;
1531 if (block_end <= from || block_start >= to) {
1532 if (partial && !buffer_uptodate(bh))
1533 *partial = 1;
1534 continue;
1535 }
1536 err = (*fn)(handle, bh);
1537 if (!ret)
1538 ret = err;
1539 }
1540 return ret;
1541 }
1542
1543 /*
1544 * To preserve ordering, it is essential that the hole instantiation and
1545 * the data write be encapsulated in a single transaction. We cannot
1546 * close off a transaction and start a new one between the ext4_get_block()
1547 * and the commit_write(). So doing the jbd2_journal_start at the start of
1548 * prepare_write() is the right place.
1549 *
1550 * Also, this function can nest inside ext4_writepage() ->
1551 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1552 * has generated enough buffer credits to do the whole page. So we won't
1553 * block on the journal in that case, which is good, because the caller may
1554 * be PF_MEMALLOC.
1555 *
1556 * By accident, ext4 can be reentered when a transaction is open via
1557 * quota file writes. If we were to commit the transaction while thus
1558 * reentered, there can be a deadlock - we would be holding a quota
1559 * lock, and the commit would never complete if another thread had a
1560 * transaction open and was blocking on the quota lock - a ranking
1561 * violation.
1562 *
1563 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1564 * will _not_ run commit under these circumstances because handle->h_ref
1565 * is elevated. We'll still have enough credits for the tiny quotafile
1566 * write.
1567 */
1568 static int do_journal_get_write_access(handle_t *handle,
1569 struct buffer_head *bh)
1570 {
1571 int dirty = buffer_dirty(bh);
1572 int ret;
1573
1574 if (!buffer_mapped(bh) || buffer_freed(bh))
1575 return 0;
1576 /*
1577 * __block_write_begin() could have dirtied some buffers. Clean
1578 * the dirty bit as jbd2_journal_get_write_access() could complain
1579 * otherwise about fs integrity issues. Setting of the dirty bit
1580 * by __block_write_begin() isn't a real problem here as we clear
1581 * the bit before releasing a page lock and thus writeback cannot
1582 * ever write the buffer.
1583 */
1584 if (dirty)
1585 clear_buffer_dirty(bh);
1586 ret = ext4_journal_get_write_access(handle, bh);
1587 if (!ret && dirty)
1588 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1589 return ret;
1590 }
1591
1592 /*
1593 * Truncate blocks that were not used by write. We have to truncate the
1594 * pagecache as well so that corresponding buffers get properly unmapped.
1595 */
1596 static void ext4_truncate_failed_write(struct inode *inode)
1597 {
1598 truncate_inode_pages(inode->i_mapping, inode->i_size);
1599 ext4_truncate(inode);
1600 }
1601
1602 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1603 struct buffer_head *bh_result, int create);
1604 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1605 loff_t pos, unsigned len, unsigned flags,
1606 struct page **pagep, void **fsdata)
1607 {
1608 struct inode *inode = mapping->host;
1609 int ret, needed_blocks;
1610 handle_t *handle;
1611 int retries = 0;
1612 struct page *page;
1613 pgoff_t index;
1614 unsigned from, to;
1615
1616 trace_ext4_write_begin(inode, pos, len, flags);
1617 /*
1618 * Reserve one block more for addition to orphan list in case
1619 * we allocate blocks but write fails for some reason
1620 */
1621 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1622 index = pos >> PAGE_CACHE_SHIFT;
1623 from = pos & (PAGE_CACHE_SIZE - 1);
1624 to = from + len;
1625
1626 retry:
1627 handle = ext4_journal_start(inode, needed_blocks);
1628 if (IS_ERR(handle)) {
1629 ret = PTR_ERR(handle);
1630 goto out;
1631 }
1632
1633 /* We cannot recurse into the filesystem as the transaction is already
1634 * started */
1635 flags |= AOP_FLAG_NOFS;
1636
1637 page = grab_cache_page_write_begin(mapping, index, flags);
1638 if (!page) {
1639 ext4_journal_stop(handle);
1640 ret = -ENOMEM;
1641 goto out;
1642 }
1643 *pagep = page;
1644
1645 if (ext4_should_dioread_nolock(inode))
1646 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1647 else
1648 ret = __block_write_begin(page, pos, len, ext4_get_block);
1649
1650 if (!ret && ext4_should_journal_data(inode)) {
1651 ret = walk_page_buffers(handle, page_buffers(page),
1652 from, to, NULL, do_journal_get_write_access);
1653 }
1654
1655 if (ret) {
1656 unlock_page(page);
1657 page_cache_release(page);
1658 /*
1659 * __block_write_begin may have instantiated a few blocks
1660 * outside i_size. Trim these off again. Don't need
1661 * i_size_read because we hold i_mutex.
1662 *
1663 * Add inode to orphan list in case we crash before
1664 * truncate finishes
1665 */
1666 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1667 ext4_orphan_add(handle, inode);
1668
1669 ext4_journal_stop(handle);
1670 if (pos + len > inode->i_size) {
1671 ext4_truncate_failed_write(inode);
1672 /*
1673 * If truncate failed early the inode might
1674 * still be on the orphan list; we need to
1675 * make sure the inode is removed from the
1676 * orphan list in that case.
1677 */
1678 if (inode->i_nlink)
1679 ext4_orphan_del(NULL, inode);
1680 }
1681 }
1682
1683 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1684 goto retry;
1685 out:
1686 return ret;
1687 }
1688
1689 /* For write_end() in data=journal mode */
1690 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1691 {
1692 if (!buffer_mapped(bh) || buffer_freed(bh))
1693 return 0;
1694 set_buffer_uptodate(bh);
1695 return ext4_handle_dirty_metadata(handle, NULL, bh);
1696 }
1697
1698 static int ext4_generic_write_end(struct file *file,
1699 struct address_space *mapping,
1700 loff_t pos, unsigned len, unsigned copied,
1701 struct page *page, void *fsdata)
1702 {
1703 int i_size_changed = 0;
1704 struct inode *inode = mapping->host;
1705 handle_t *handle = ext4_journal_current_handle();
1706
1707 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1708
1709 /*
1710 * No need to use i_size_read() here, the i_size
1711 * cannot change under us because we hold i_mutex.
1712 *
1713 * But it's important to update i_size while still holding page lock:
1714 * page writeout could otherwise come in and zero beyond i_size.
1715 */
1716 if (pos + copied > inode->i_size) {
1717 i_size_write(inode, pos + copied);
1718 i_size_changed = 1;
1719 }
1720
1721 if (pos + copied > EXT4_I(inode)->i_disksize) {
1722 /* We need to mark inode dirty even if
1723 * new_i_size is less that inode->i_size
1724 * bu greater than i_disksize.(hint delalloc)
1725 */
1726 ext4_update_i_disksize(inode, (pos + copied));
1727 i_size_changed = 1;
1728 }
1729 unlock_page(page);
1730 page_cache_release(page);
1731
1732 /*
1733 * Don't mark the inode dirty under page lock. First, it unnecessarily
1734 * makes the holding time of page lock longer. Second, it forces lock
1735 * ordering of page lock and transaction start for journaling
1736 * filesystems.
1737 */
1738 if (i_size_changed)
1739 ext4_mark_inode_dirty(handle, inode);
1740
1741 return copied;
1742 }
1743
1744 /*
1745 * We need to pick up the new inode size which generic_commit_write gave us
1746 * `file' can be NULL - eg, when called from page_symlink().
1747 *
1748 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1749 * buffers are managed internally.
1750 */
1751 static int ext4_ordered_write_end(struct file *file,
1752 struct address_space *mapping,
1753 loff_t pos, unsigned len, unsigned copied,
1754 struct page *page, void *fsdata)
1755 {
1756 handle_t *handle = ext4_journal_current_handle();
1757 struct inode *inode = mapping->host;
1758 int ret = 0, ret2;
1759
1760 trace_ext4_ordered_write_end(inode, pos, len, copied);
1761 ret = ext4_jbd2_file_inode(handle, inode);
1762
1763 if (ret == 0) {
1764 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1765 page, fsdata);
1766 copied = ret2;
1767 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1768 /* if we have allocated more blocks and copied
1769 * less. We will have blocks allocated outside
1770 * inode->i_size. So truncate them
1771 */
1772 ext4_orphan_add(handle, inode);
1773 if (ret2 < 0)
1774 ret = ret2;
1775 }
1776 ret2 = ext4_journal_stop(handle);
1777 if (!ret)
1778 ret = ret2;
1779
1780 if (pos + len > inode->i_size) {
1781 ext4_truncate_failed_write(inode);
1782 /*
1783 * If truncate failed early the inode might still be
1784 * on the orphan list; we need to make sure the inode
1785 * is removed from the orphan list in that case.
1786 */
1787 if (inode->i_nlink)
1788 ext4_orphan_del(NULL, inode);
1789 }
1790
1791
1792 return ret ? ret : copied;
1793 }
1794
1795 static int ext4_writeback_write_end(struct file *file,
1796 struct address_space *mapping,
1797 loff_t pos, unsigned len, unsigned copied,
1798 struct page *page, void *fsdata)
1799 {
1800 handle_t *handle = ext4_journal_current_handle();
1801 struct inode *inode = mapping->host;
1802 int ret = 0, ret2;
1803
1804 trace_ext4_writeback_write_end(inode, pos, len, copied);
1805 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1806 page, fsdata);
1807 copied = ret2;
1808 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1809 /* if we have allocated more blocks and copied
1810 * less. We will have blocks allocated outside
1811 * inode->i_size. So truncate them
1812 */
1813 ext4_orphan_add(handle, inode);
1814
1815 if (ret2 < 0)
1816 ret = ret2;
1817
1818 ret2 = ext4_journal_stop(handle);
1819 if (!ret)
1820 ret = ret2;
1821
1822 if (pos + len > inode->i_size) {
1823 ext4_truncate_failed_write(inode);
1824 /*
1825 * If truncate failed early the inode might still be
1826 * on the orphan list; we need to make sure the inode
1827 * is removed from the orphan list in that case.
1828 */
1829 if (inode->i_nlink)
1830 ext4_orphan_del(NULL, inode);
1831 }
1832
1833 return ret ? ret : copied;
1834 }
1835
1836 static int ext4_journalled_write_end(struct file *file,
1837 struct address_space *mapping,
1838 loff_t pos, unsigned len, unsigned copied,
1839 struct page *page, void *fsdata)
1840 {
1841 handle_t *handle = ext4_journal_current_handle();
1842 struct inode *inode = mapping->host;
1843 int ret = 0, ret2;
1844 int partial = 0;
1845 unsigned from, to;
1846 loff_t new_i_size;
1847
1848 trace_ext4_journalled_write_end(inode, pos, len, copied);
1849 from = pos & (PAGE_CACHE_SIZE - 1);
1850 to = from + len;
1851
1852 if (copied < len) {
1853 if (!PageUptodate(page))
1854 copied = 0;
1855 page_zero_new_buffers(page, from+copied, to);
1856 }
1857
1858 ret = walk_page_buffers(handle, page_buffers(page), from,
1859 to, &partial, write_end_fn);
1860 if (!partial)
1861 SetPageUptodate(page);
1862 new_i_size = pos + copied;
1863 if (new_i_size > inode->i_size)
1864 i_size_write(inode, pos+copied);
1865 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1866 if (new_i_size > EXT4_I(inode)->i_disksize) {
1867 ext4_update_i_disksize(inode, new_i_size);
1868 ret2 = ext4_mark_inode_dirty(handle, inode);
1869 if (!ret)
1870 ret = ret2;
1871 }
1872
1873 unlock_page(page);
1874 page_cache_release(page);
1875 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1876 /* if we have allocated more blocks and copied
1877 * less. We will have blocks allocated outside
1878 * inode->i_size. So truncate them
1879 */
1880 ext4_orphan_add(handle, inode);
1881
1882 ret2 = ext4_journal_stop(handle);
1883 if (!ret)
1884 ret = ret2;
1885 if (pos + len > inode->i_size) {
1886 ext4_truncate_failed_write(inode);
1887 /*
1888 * If truncate failed early the inode might still be
1889 * on the orphan list; we need to make sure the inode
1890 * is removed from the orphan list in that case.
1891 */
1892 if (inode->i_nlink)
1893 ext4_orphan_del(NULL, inode);
1894 }
1895
1896 return ret ? ret : copied;
1897 }
1898
1899 /*
1900 * Reserve a single block located at lblock
1901 */
1902 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1903 {
1904 int retries = 0;
1905 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1906 struct ext4_inode_info *ei = EXT4_I(inode);
1907 unsigned long md_needed;
1908 int ret;
1909
1910 /*
1911 * recalculate the amount of metadata blocks to reserve
1912 * in order to allocate nrblocks
1913 * worse case is one extent per block
1914 */
1915 repeat:
1916 spin_lock(&ei->i_block_reservation_lock);
1917 md_needed = ext4_calc_metadata_amount(inode, lblock);
1918 trace_ext4_da_reserve_space(inode, md_needed);
1919 spin_unlock(&ei->i_block_reservation_lock);
1920
1921 /*
1922 * We will charge metadata quota at writeout time; this saves
1923 * us from metadata over-estimation, though we may go over by
1924 * a small amount in the end. Here we just reserve for data.
1925 */
1926 ret = dquot_reserve_block(inode, 1);
1927 if (ret)
1928 return ret;
1929 /*
1930 * We do still charge estimated metadata to the sb though;
1931 * we cannot afford to run out of free blocks.
1932 */
1933 if (ext4_claim_free_blocks(sbi, md_needed + 1, 0)) {
1934 dquot_release_reservation_block(inode, 1);
1935 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1936 yield();
1937 goto repeat;
1938 }
1939 return -ENOSPC;
1940 }
1941 spin_lock(&ei->i_block_reservation_lock);
1942 ei->i_reserved_data_blocks++;
1943 ei->i_reserved_meta_blocks += md_needed;
1944 spin_unlock(&ei->i_block_reservation_lock);
1945
1946 return 0; /* success */
1947 }
1948
1949 static void ext4_da_release_space(struct inode *inode, int to_free)
1950 {
1951 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1952 struct ext4_inode_info *ei = EXT4_I(inode);
1953
1954 if (!to_free)
1955 return; /* Nothing to release, exit */
1956
1957 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1958
1959 trace_ext4_da_release_space(inode, to_free);
1960 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1961 /*
1962 * if there aren't enough reserved blocks, then the
1963 * counter is messed up somewhere. Since this
1964 * function is called from invalidate page, it's
1965 * harmless to return without any action.
1966 */
1967 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1968 "ino %lu, to_free %d with only %d reserved "
1969 "data blocks\n", inode->i_ino, to_free,
1970 ei->i_reserved_data_blocks);
1971 WARN_ON(1);
1972 to_free = ei->i_reserved_data_blocks;
1973 }
1974 ei->i_reserved_data_blocks -= to_free;
1975
1976 if (ei->i_reserved_data_blocks == 0) {
1977 /*
1978 * We can release all of the reserved metadata blocks
1979 * only when we have written all of the delayed
1980 * allocation blocks.
1981 */
1982 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1983 ei->i_reserved_meta_blocks);
1984 ei->i_reserved_meta_blocks = 0;
1985 ei->i_da_metadata_calc_len = 0;
1986 }
1987
1988 /* update fs dirty data blocks counter */
1989 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1990
1991 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1992
1993 dquot_release_reservation_block(inode, to_free);
1994 }
1995
1996 static void ext4_da_page_release_reservation(struct page *page,
1997 unsigned long offset)
1998 {
1999 int to_release = 0;
2000 struct buffer_head *head, *bh;
2001 unsigned int curr_off = 0;
2002
2003 head = page_buffers(page);
2004 bh = head;
2005 do {
2006 unsigned int next_off = curr_off + bh->b_size;
2007
2008 if ((offset <= curr_off) && (buffer_delay(bh))) {
2009 to_release++;
2010 clear_buffer_delay(bh);
2011 }
2012 curr_off = next_off;
2013 } while ((bh = bh->b_this_page) != head);
2014 ext4_da_release_space(page->mapping->host, to_release);
2015 }
2016
2017 /*
2018 * Delayed allocation stuff
2019 */
2020
2021 /*
2022 * mpage_da_submit_io - walks through extent of pages and try to write
2023 * them with writepage() call back
2024 *
2025 * @mpd->inode: inode
2026 * @mpd->first_page: first page of the extent
2027 * @mpd->next_page: page after the last page of the extent
2028 *
2029 * By the time mpage_da_submit_io() is called we expect all blocks
2030 * to be allocated. this may be wrong if allocation failed.
2031 *
2032 * As pages are already locked by write_cache_pages(), we can't use it
2033 */
2034 static int mpage_da_submit_io(struct mpage_da_data *mpd,
2035 struct ext4_map_blocks *map)
2036 {
2037 struct pagevec pvec;
2038 unsigned long index, end;
2039 int ret = 0, err, nr_pages, i;
2040 struct inode *inode = mpd->inode;
2041 struct address_space *mapping = inode->i_mapping;
2042 loff_t size = i_size_read(inode);
2043 unsigned int len, block_start;
2044 struct buffer_head *bh, *page_bufs = NULL;
2045 int journal_data = ext4_should_journal_data(inode);
2046 sector_t pblock = 0, cur_logical = 0;
2047 struct ext4_io_submit io_submit;
2048
2049 BUG_ON(mpd->next_page <= mpd->first_page);
2050 memset(&io_submit, 0, sizeof(io_submit));
2051 /*
2052 * We need to start from the first_page to the next_page - 1
2053 * to make sure we also write the mapped dirty buffer_heads.
2054 * If we look at mpd->b_blocknr we would only be looking
2055 * at the currently mapped buffer_heads.
2056 */
2057 index = mpd->first_page;
2058 end = mpd->next_page - 1;
2059
2060 pagevec_init(&pvec, 0);
2061 while (index <= end) {
2062 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2063 if (nr_pages == 0)
2064 break;
2065 for (i = 0; i < nr_pages; i++) {
2066 int commit_write = 0, skip_page = 0;
2067 struct page *page = pvec.pages[i];
2068
2069 index = page->index;
2070 if (index > end)
2071 break;
2072
2073 if (index == size >> PAGE_CACHE_SHIFT)
2074 len = size & ~PAGE_CACHE_MASK;
2075 else
2076 len = PAGE_CACHE_SIZE;
2077 if (map) {
2078 cur_logical = index << (PAGE_CACHE_SHIFT -
2079 inode->i_blkbits);
2080 pblock = map->m_pblk + (cur_logical -
2081 map->m_lblk);
2082 }
2083 index++;
2084
2085 BUG_ON(!PageLocked(page));
2086 BUG_ON(PageWriteback(page));
2087
2088 /*
2089 * If the page does not have buffers (for
2090 * whatever reason), try to create them using
2091 * __block_write_begin. If this fails,
2092 * skip the page and move on.
2093 */
2094 if (!page_has_buffers(page)) {
2095 if (__block_write_begin(page, 0, len,
2096 noalloc_get_block_write)) {
2097 skip_page:
2098 unlock_page(page);
2099 continue;
2100 }
2101 commit_write = 1;
2102 }
2103
2104 bh = page_bufs = page_buffers(page);
2105 block_start = 0;
2106 do {
2107 if (!bh)
2108 goto skip_page;
2109 if (map && (cur_logical >= map->m_lblk) &&
2110 (cur_logical <= (map->m_lblk +
2111 (map->m_len - 1)))) {
2112 if (buffer_delay(bh)) {
2113 clear_buffer_delay(bh);
2114 bh->b_blocknr = pblock;
2115 }
2116 if (buffer_unwritten(bh) ||
2117 buffer_mapped(bh))
2118 BUG_ON(bh->b_blocknr != pblock);
2119 if (map->m_flags & EXT4_MAP_UNINIT)
2120 set_buffer_uninit(bh);
2121 clear_buffer_unwritten(bh);
2122 }
2123
2124 /* skip page if block allocation undone */
2125 if (buffer_delay(bh) || buffer_unwritten(bh))
2126 skip_page = 1;
2127 bh = bh->b_this_page;
2128 block_start += bh->b_size;
2129 cur_logical++;
2130 pblock++;
2131 } while (bh != page_bufs);
2132
2133 if (skip_page)
2134 goto skip_page;
2135
2136 if (commit_write)
2137 /* mark the buffer_heads as dirty & uptodate */
2138 block_commit_write(page, 0, len);
2139
2140 clear_page_dirty_for_io(page);
2141 /*
2142 * Delalloc doesn't support data journalling,
2143 * but eventually maybe we'll lift this
2144 * restriction.
2145 */
2146 if (unlikely(journal_data && PageChecked(page)))
2147 err = __ext4_journalled_writepage(page, len);
2148 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
2149 err = ext4_bio_write_page(&io_submit, page,
2150 len, mpd->wbc);
2151 else
2152 err = block_write_full_page(page,
2153 noalloc_get_block_write, mpd->wbc);
2154
2155 if (!err)
2156 mpd->pages_written++;
2157 /*
2158 * In error case, we have to continue because
2159 * remaining pages are still locked
2160 */
2161 if (ret == 0)
2162 ret = err;
2163 }
2164 pagevec_release(&pvec);
2165 }
2166 ext4_io_submit(&io_submit);
2167 return ret;
2168 }
2169
2170 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
2171 {
2172 int nr_pages, i;
2173 pgoff_t index, end;
2174 struct pagevec pvec;
2175 struct inode *inode = mpd->inode;
2176 struct address_space *mapping = inode->i_mapping;
2177
2178 index = mpd->first_page;
2179 end = mpd->next_page - 1;
2180 while (index <= end) {
2181 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2182 if (nr_pages == 0)
2183 break;
2184 for (i = 0; i < nr_pages; i++) {
2185 struct page *page = pvec.pages[i];
2186 if (page->index > end)
2187 break;
2188 BUG_ON(!PageLocked(page));
2189 BUG_ON(PageWriteback(page));
2190 block_invalidatepage(page, 0);
2191 ClearPageUptodate(page);
2192 unlock_page(page);
2193 }
2194 index = pvec.pages[nr_pages - 1]->index + 1;
2195 pagevec_release(&pvec);
2196 }
2197 return;
2198 }
2199
2200 static void ext4_print_free_blocks(struct inode *inode)
2201 {
2202 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2203 printk(KERN_CRIT "Total free blocks count %lld\n",
2204 ext4_count_free_blocks(inode->i_sb));
2205 printk(KERN_CRIT "Free/Dirty block details\n");
2206 printk(KERN_CRIT "free_blocks=%lld\n",
2207 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2208 printk(KERN_CRIT "dirty_blocks=%lld\n",
2209 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2210 printk(KERN_CRIT "Block reservation details\n");
2211 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2212 EXT4_I(inode)->i_reserved_data_blocks);
2213 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2214 EXT4_I(inode)->i_reserved_meta_blocks);
2215 return;
2216 }
2217
2218 /*
2219 * mpage_da_map_and_submit - go through given space, map them
2220 * if necessary, and then submit them for I/O
2221 *
2222 * @mpd - bh describing space
2223 *
2224 * The function skips space we know is already mapped to disk blocks.
2225 *
2226 */
2227 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
2228 {
2229 int err, blks, get_blocks_flags;
2230 struct ext4_map_blocks map, *mapp = NULL;
2231 sector_t next = mpd->b_blocknr;
2232 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2233 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2234 handle_t *handle = NULL;
2235
2236 /*
2237 * If the blocks are mapped already, or we couldn't accumulate
2238 * any blocks, then proceed immediately to the submission stage.
2239 */
2240 if ((mpd->b_size == 0) ||
2241 ((mpd->b_state & (1 << BH_Mapped)) &&
2242 !(mpd->b_state & (1 << BH_Delay)) &&
2243 !(mpd->b_state & (1 << BH_Unwritten))))
2244 goto submit_io;
2245
2246 handle = ext4_journal_current_handle();
2247 BUG_ON(!handle);
2248
2249 /*
2250 * Call ext4_map_blocks() to allocate any delayed allocation
2251 * blocks, or to convert an uninitialized extent to be
2252 * initialized (in the case where we have written into
2253 * one or more preallocated blocks).
2254 *
2255 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2256 * indicate that we are on the delayed allocation path. This
2257 * affects functions in many different parts of the allocation
2258 * call path. This flag exists primarily because we don't
2259 * want to change *many* call functions, so ext4_map_blocks()
2260 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
2261 * inode's allocation semaphore is taken.
2262 *
2263 * If the blocks in questions were delalloc blocks, set
2264 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2265 * variables are updated after the blocks have been allocated.
2266 */
2267 map.m_lblk = next;
2268 map.m_len = max_blocks;
2269 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2270 if (ext4_should_dioread_nolock(mpd->inode))
2271 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2272 if (mpd->b_state & (1 << BH_Delay))
2273 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2274
2275 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2276 if (blks < 0) {
2277 struct super_block *sb = mpd->inode->i_sb;
2278
2279 err = blks;
2280 /*
2281 * If get block returns EAGAIN or ENOSPC and there
2282 * appears to be free blocks we will just let
2283 * mpage_da_submit_io() unlock all of the pages.
2284 */
2285 if (err == -EAGAIN)
2286 goto submit_io;
2287
2288 if (err == -ENOSPC &&
2289 ext4_count_free_blocks(sb)) {
2290 mpd->retval = err;
2291 goto submit_io;
2292 }
2293
2294 /*
2295 * get block failure will cause us to loop in
2296 * writepages, because a_ops->writepage won't be able
2297 * to make progress. The page will be redirtied by
2298 * writepage and writepages will again try to write
2299 * the same.
2300 */
2301 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2302 ext4_msg(sb, KERN_CRIT,
2303 "delayed block allocation failed for inode %lu "
2304 "at logical offset %llu with max blocks %zd "
2305 "with error %d", mpd->inode->i_ino,
2306 (unsigned long long) next,
2307 mpd->b_size >> mpd->inode->i_blkbits, err);
2308 ext4_msg(sb, KERN_CRIT,
2309 "This should not happen!! Data will be lost\n");
2310 if (err == -ENOSPC)
2311 ext4_print_free_blocks(mpd->inode);
2312 }
2313 /* invalidate all the pages */
2314 ext4_da_block_invalidatepages(mpd);
2315
2316 /* Mark this page range as having been completed */
2317 mpd->io_done = 1;
2318 return;
2319 }
2320 BUG_ON(blks == 0);
2321
2322 mapp = &map;
2323 if (map.m_flags & EXT4_MAP_NEW) {
2324 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2325 int i;
2326
2327 for (i = 0; i < map.m_len; i++)
2328 unmap_underlying_metadata(bdev, map.m_pblk + i);
2329 }
2330
2331 if (ext4_should_order_data(mpd->inode)) {
2332 err = ext4_jbd2_file_inode(handle, mpd->inode);
2333 if (err)
2334 /* This only happens if the journal is aborted */
2335 return;
2336 }
2337
2338 /*
2339 * Update on-disk size along with block allocation.
2340 */
2341 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2342 if (disksize > i_size_read(mpd->inode))
2343 disksize = i_size_read(mpd->inode);
2344 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2345 ext4_update_i_disksize(mpd->inode, disksize);
2346 err = ext4_mark_inode_dirty(handle, mpd->inode);
2347 if (err)
2348 ext4_error(mpd->inode->i_sb,
2349 "Failed to mark inode %lu dirty",
2350 mpd->inode->i_ino);
2351 }
2352
2353 submit_io:
2354 mpage_da_submit_io(mpd, mapp);
2355 mpd->io_done = 1;
2356 }
2357
2358 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2359 (1 << BH_Delay) | (1 << BH_Unwritten))
2360
2361 /*
2362 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2363 *
2364 * @mpd->lbh - extent of blocks
2365 * @logical - logical number of the block in the file
2366 * @bh - bh of the block (used to access block's state)
2367 *
2368 * the function is used to collect contig. blocks in same state
2369 */
2370 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2371 sector_t logical, size_t b_size,
2372 unsigned long b_state)
2373 {
2374 sector_t next;
2375 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2376
2377 /*
2378 * XXX Don't go larger than mballoc is willing to allocate
2379 * This is a stopgap solution. We eventually need to fold
2380 * mpage_da_submit_io() into this function and then call
2381 * ext4_map_blocks() multiple times in a loop
2382 */
2383 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2384 goto flush_it;
2385
2386 /* check if thereserved journal credits might overflow */
2387 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2388 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2389 /*
2390 * With non-extent format we are limited by the journal
2391 * credit available. Total credit needed to insert
2392 * nrblocks contiguous blocks is dependent on the
2393 * nrblocks. So limit nrblocks.
2394 */
2395 goto flush_it;
2396 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2397 EXT4_MAX_TRANS_DATA) {
2398 /*
2399 * Adding the new buffer_head would make it cross the
2400 * allowed limit for which we have journal credit
2401 * reserved. So limit the new bh->b_size
2402 */
2403 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2404 mpd->inode->i_blkbits;
2405 /* we will do mpage_da_submit_io in the next loop */
2406 }
2407 }
2408 /*
2409 * First block in the extent
2410 */
2411 if (mpd->b_size == 0) {
2412 mpd->b_blocknr = logical;
2413 mpd->b_size = b_size;
2414 mpd->b_state = b_state & BH_FLAGS;
2415 return;
2416 }
2417
2418 next = mpd->b_blocknr + nrblocks;
2419 /*
2420 * Can we merge the block to our big extent?
2421 */
2422 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2423 mpd->b_size += b_size;
2424 return;
2425 }
2426
2427 flush_it:
2428 /*
2429 * We couldn't merge the block to our extent, so we
2430 * need to flush current extent and start new one
2431 */
2432 mpage_da_map_and_submit(mpd);
2433 return;
2434 }
2435
2436 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2437 {
2438 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2439 }
2440
2441 /*
2442 * This is a special get_blocks_t callback which is used by
2443 * ext4_da_write_begin(). It will either return mapped block or
2444 * reserve space for a single block.
2445 *
2446 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2447 * We also have b_blocknr = -1 and b_bdev initialized properly
2448 *
2449 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2450 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2451 * initialized properly.
2452 */
2453 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2454 struct buffer_head *bh, int create)
2455 {
2456 struct ext4_map_blocks map;
2457 int ret = 0;
2458 sector_t invalid_block = ~((sector_t) 0xffff);
2459
2460 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2461 invalid_block = ~0;
2462
2463 BUG_ON(create == 0);
2464 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2465
2466 map.m_lblk = iblock;
2467 map.m_len = 1;
2468
2469 /*
2470 * first, we need to know whether the block is allocated already
2471 * preallocated blocks are unmapped but should treated
2472 * the same as allocated blocks.
2473 */
2474 ret = ext4_map_blocks(NULL, inode, &map, 0);
2475 if (ret < 0)
2476 return ret;
2477 if (ret == 0) {
2478 if (buffer_delay(bh))
2479 return 0; /* Not sure this could or should happen */
2480 /*
2481 * XXX: __block_write_begin() unmaps passed block, is it OK?
2482 */
2483 ret = ext4_da_reserve_space(inode, iblock);
2484 if (ret)
2485 /* not enough space to reserve */
2486 return ret;
2487
2488 map_bh(bh, inode->i_sb, invalid_block);
2489 set_buffer_new(bh);
2490 set_buffer_delay(bh);
2491 return 0;
2492 }
2493
2494 map_bh(bh, inode->i_sb, map.m_pblk);
2495 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2496
2497 if (buffer_unwritten(bh)) {
2498 /* A delayed write to unwritten bh should be marked
2499 * new and mapped. Mapped ensures that we don't do
2500 * get_block multiple times when we write to the same
2501 * offset and new ensures that we do proper zero out
2502 * for partial write.
2503 */
2504 set_buffer_new(bh);
2505 set_buffer_mapped(bh);
2506 }
2507 return 0;
2508 }
2509
2510 /*
2511 * This function is used as a standard get_block_t calback function
2512 * when there is no desire to allocate any blocks. It is used as a
2513 * callback function for block_write_begin() and block_write_full_page().
2514 * These functions should only try to map a single block at a time.
2515 *
2516 * Since this function doesn't do block allocations even if the caller
2517 * requests it by passing in create=1, it is critically important that
2518 * any caller checks to make sure that any buffer heads are returned
2519 * by this function are either all already mapped or marked for
2520 * delayed allocation before calling block_write_full_page(). Otherwise,
2521 * b_blocknr could be left unitialized, and the page write functions will
2522 * be taken by surprise.
2523 */
2524 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2525 struct buffer_head *bh_result, int create)
2526 {
2527 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2528 return _ext4_get_block(inode, iblock, bh_result, 0);
2529 }
2530
2531 static int bget_one(handle_t *handle, struct buffer_head *bh)
2532 {
2533 get_bh(bh);
2534 return 0;
2535 }
2536
2537 static int bput_one(handle_t *handle, struct buffer_head *bh)
2538 {
2539 put_bh(bh);
2540 return 0;
2541 }
2542
2543 static int __ext4_journalled_writepage(struct page *page,
2544 unsigned int len)
2545 {
2546 struct address_space *mapping = page->mapping;
2547 struct inode *inode = mapping->host;
2548 struct buffer_head *page_bufs;
2549 handle_t *handle = NULL;
2550 int ret = 0;
2551 int err;
2552
2553 ClearPageChecked(page);
2554 page_bufs = page_buffers(page);
2555 BUG_ON(!page_bufs);
2556 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2557 /* As soon as we unlock the page, it can go away, but we have
2558 * references to buffers so we are safe */
2559 unlock_page(page);
2560
2561 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2562 if (IS_ERR(handle)) {
2563 ret = PTR_ERR(handle);
2564 goto out;
2565 }
2566
2567 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2568 do_journal_get_write_access);
2569
2570 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2571 write_end_fn);
2572 if (ret == 0)
2573 ret = err;
2574 err = ext4_journal_stop(handle);
2575 if (!ret)
2576 ret = err;
2577
2578 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2579 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2580 out:
2581 return ret;
2582 }
2583
2584 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2585 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2586
2587 /*
2588 * Note that we don't need to start a transaction unless we're journaling data
2589 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2590 * need to file the inode to the transaction's list in ordered mode because if
2591 * we are writing back data added by write(), the inode is already there and if
2592 * we are writing back data modified via mmap(), no one guarantees in which
2593 * transaction the data will hit the disk. In case we are journaling data, we
2594 * cannot start transaction directly because transaction start ranks above page
2595 * lock so we have to do some magic.
2596 *
2597 * This function can get called via...
2598 * - ext4_da_writepages after taking page lock (have journal handle)
2599 * - journal_submit_inode_data_buffers (no journal handle)
2600 * - shrink_page_list via pdflush (no journal handle)
2601 * - grab_page_cache when doing write_begin (have journal handle)
2602 *
2603 * We don't do any block allocation in this function. If we have page with
2604 * multiple blocks we need to write those buffer_heads that are mapped. This
2605 * is important for mmaped based write. So if we do with blocksize 1K
2606 * truncate(f, 1024);
2607 * a = mmap(f, 0, 4096);
2608 * a[0] = 'a';
2609 * truncate(f, 4096);
2610 * we have in the page first buffer_head mapped via page_mkwrite call back
2611 * but other bufer_heads would be unmapped but dirty(dirty done via the
2612 * do_wp_page). So writepage should write the first block. If we modify
2613 * the mmap area beyond 1024 we will again get a page_fault and the
2614 * page_mkwrite callback will do the block allocation and mark the
2615 * buffer_heads mapped.
2616 *
2617 * We redirty the page if we have any buffer_heads that is either delay or
2618 * unwritten in the page.
2619 *
2620 * We can get recursively called as show below.
2621 *
2622 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2623 * ext4_writepage()
2624 *
2625 * But since we don't do any block allocation we should not deadlock.
2626 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2627 */
2628 static int ext4_writepage(struct page *page,
2629 struct writeback_control *wbc)
2630 {
2631 int ret = 0, commit_write = 0;
2632 loff_t size;
2633 unsigned int len;
2634 struct buffer_head *page_bufs = NULL;
2635 struct inode *inode = page->mapping->host;
2636
2637 trace_ext4_writepage(page);
2638 size = i_size_read(inode);
2639 if (page->index == size >> PAGE_CACHE_SHIFT)
2640 len = size & ~PAGE_CACHE_MASK;
2641 else
2642 len = PAGE_CACHE_SIZE;
2643
2644 /*
2645 * If the page does not have buffers (for whatever reason),
2646 * try to create them using __block_write_begin. If this
2647 * fails, redirty the page and move on.
2648 */
2649 if (!page_has_buffers(page)) {
2650 if (__block_write_begin(page, 0, len,
2651 noalloc_get_block_write)) {
2652 redirty_page:
2653 redirty_page_for_writepage(wbc, page);
2654 unlock_page(page);
2655 return 0;
2656 }
2657 commit_write = 1;
2658 }
2659 page_bufs = page_buffers(page);
2660 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2661 ext4_bh_delay_or_unwritten)) {
2662 /*
2663 * We don't want to do block allocation, so redirty
2664 * the page and return. We may reach here when we do
2665 * a journal commit via journal_submit_inode_data_buffers.
2666 * We can also reach here via shrink_page_list
2667 */
2668 goto redirty_page;
2669 }
2670 if (commit_write)
2671 /* now mark the buffer_heads as dirty and uptodate */
2672 block_commit_write(page, 0, len);
2673
2674 if (PageChecked(page) && ext4_should_journal_data(inode))
2675 /*
2676 * It's mmapped pagecache. Add buffers and journal it. There
2677 * doesn't seem much point in redirtying the page here.
2678 */
2679 return __ext4_journalled_writepage(page, len);
2680
2681 if (buffer_uninit(page_bufs)) {
2682 ext4_set_bh_endio(page_bufs, inode);
2683 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2684 wbc, ext4_end_io_buffer_write);
2685 } else
2686 ret = block_write_full_page(page, noalloc_get_block_write,
2687 wbc);
2688
2689 return ret;
2690 }
2691
2692 /*
2693 * This is called via ext4_da_writepages() to
2694 * calculate the total number of credits to reserve to fit
2695 * a single extent allocation into a single transaction,
2696 * ext4_da_writpeages() will loop calling this before
2697 * the block allocation.
2698 */
2699
2700 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2701 {
2702 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2703
2704 /*
2705 * With non-extent format the journal credit needed to
2706 * insert nrblocks contiguous block is dependent on
2707 * number of contiguous block. So we will limit
2708 * number of contiguous block to a sane value
2709 */
2710 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2711 (max_blocks > EXT4_MAX_TRANS_DATA))
2712 max_blocks = EXT4_MAX_TRANS_DATA;
2713
2714 return ext4_chunk_trans_blocks(inode, max_blocks);
2715 }
2716
2717 /*
2718 * write_cache_pages_da - walk the list of dirty pages of the given
2719 * address space and accumulate pages that need writing, and call
2720 * mpage_da_map_and_submit to map a single contiguous memory region
2721 * and then write them.
2722 */
2723 static int write_cache_pages_da(struct address_space *mapping,
2724 struct writeback_control *wbc,
2725 struct mpage_da_data *mpd,
2726 pgoff_t *done_index)
2727 {
2728 struct buffer_head *bh, *head;
2729 struct inode *inode = mapping->host;
2730 struct pagevec pvec;
2731 unsigned int nr_pages;
2732 sector_t logical;
2733 pgoff_t index, end;
2734 long nr_to_write = wbc->nr_to_write;
2735 int i, tag, ret = 0;
2736
2737 memset(mpd, 0, sizeof(struct mpage_da_data));
2738 mpd->wbc = wbc;
2739 mpd->inode = inode;
2740 pagevec_init(&pvec, 0);
2741 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2742 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2743
2744 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2745 tag = PAGECACHE_TAG_TOWRITE;
2746 else
2747 tag = PAGECACHE_TAG_DIRTY;
2748
2749 *done_index = index;
2750 while (index <= end) {
2751 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2752 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2753 if (nr_pages == 0)
2754 return 0;
2755
2756 for (i = 0; i < nr_pages; i++) {
2757 struct page *page = pvec.pages[i];
2758
2759 /*
2760 * At this point, the page may be truncated or
2761 * invalidated (changing page->mapping to NULL), or
2762 * even swizzled back from swapper_space to tmpfs file
2763 * mapping. However, page->index will not change
2764 * because we have a reference on the page.
2765 */
2766 if (page->index > end)
2767 goto out;
2768
2769 *done_index = page->index + 1;
2770
2771 /*
2772 * If we can't merge this page, and we have
2773 * accumulated an contiguous region, write it
2774 */
2775 if ((mpd->next_page != page->index) &&
2776 (mpd->next_page != mpd->first_page)) {
2777 mpage_da_map_and_submit(mpd);
2778 goto ret_extent_tail;
2779 }
2780
2781 lock_page(page);
2782
2783 /*
2784 * If the page is no longer dirty, or its
2785 * mapping no longer corresponds to inode we
2786 * are writing (which means it has been
2787 * truncated or invalidated), or the page is
2788 * already under writeback and we are not
2789 * doing a data integrity writeback, skip the page
2790 */
2791 if (!PageDirty(page) ||
2792 (PageWriteback(page) &&
2793 (wbc->sync_mode == WB_SYNC_NONE)) ||
2794 unlikely(page->mapping != mapping)) {
2795 unlock_page(page);
2796 continue;
2797 }
2798
2799 wait_on_page_writeback(page);
2800 BUG_ON(PageWriteback(page));
2801
2802 if (mpd->next_page != page->index)
2803 mpd->first_page = page->index;
2804 mpd->next_page = page->index + 1;
2805 logical = (sector_t) page->index <<
2806 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2807
2808 if (!page_has_buffers(page)) {
2809 mpage_add_bh_to_extent(mpd, logical,
2810 PAGE_CACHE_SIZE,
2811 (1 << BH_Dirty) | (1 << BH_Uptodate));
2812 if (mpd->io_done)
2813 goto ret_extent_tail;
2814 } else {
2815 /*
2816 * Page with regular buffer heads,
2817 * just add all dirty ones
2818 */
2819 head = page_buffers(page);
2820 bh = head;
2821 do {
2822 BUG_ON(buffer_locked(bh));
2823 /*
2824 * We need to try to allocate
2825 * unmapped blocks in the same page.
2826 * Otherwise we won't make progress
2827 * with the page in ext4_writepage
2828 */
2829 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2830 mpage_add_bh_to_extent(mpd, logical,
2831 bh->b_size,
2832 bh->b_state);
2833 if (mpd->io_done)
2834 goto ret_extent_tail;
2835 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2836 /*
2837 * mapped dirty buffer. We need
2838 * to update the b_state
2839 * because we look at b_state
2840 * in mpage_da_map_blocks. We
2841 * don't update b_size because
2842 * if we find an unmapped
2843 * buffer_head later we need to
2844 * use the b_state flag of that
2845 * buffer_head.
2846 */
2847 if (mpd->b_size == 0)
2848 mpd->b_state = bh->b_state & BH_FLAGS;
2849 }
2850 logical++;
2851 } while ((bh = bh->b_this_page) != head);
2852 }
2853
2854 if (nr_to_write > 0) {
2855 nr_to_write--;
2856 if (nr_to_write == 0 &&
2857 wbc->sync_mode == WB_SYNC_NONE)
2858 /*
2859 * We stop writing back only if we are
2860 * not doing integrity sync. In case of
2861 * integrity sync we have to keep going
2862 * because someone may be concurrently
2863 * dirtying pages, and we might have
2864 * synced a lot of newly appeared dirty
2865 * pages, but have not synced all of the
2866 * old dirty pages.
2867 */
2868 goto out;
2869 }
2870 }
2871 pagevec_release(&pvec);
2872 cond_resched();
2873 }
2874 return 0;
2875 ret_extent_tail:
2876 ret = MPAGE_DA_EXTENT_TAIL;
2877 out:
2878 pagevec_release(&pvec);
2879 cond_resched();
2880 return ret;
2881 }
2882
2883
2884 static int ext4_da_writepages(struct address_space *mapping,
2885 struct writeback_control *wbc)
2886 {
2887 pgoff_t index;
2888 int range_whole = 0;
2889 handle_t *handle = NULL;
2890 struct mpage_da_data mpd;
2891 struct inode *inode = mapping->host;
2892 int pages_written = 0;
2893 unsigned int max_pages;
2894 int range_cyclic, cycled = 1, io_done = 0;
2895 int needed_blocks, ret = 0;
2896 long desired_nr_to_write, nr_to_writebump = 0;
2897 loff_t range_start = wbc->range_start;
2898 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2899 pgoff_t done_index = 0;
2900 pgoff_t end;
2901
2902 trace_ext4_da_writepages(inode, wbc);
2903
2904 /*
2905 * No pages to write? This is mainly a kludge to avoid starting
2906 * a transaction for special inodes like journal inode on last iput()
2907 * because that could violate lock ordering on umount
2908 */
2909 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2910 return 0;
2911
2912 /*
2913 * If the filesystem has aborted, it is read-only, so return
2914 * right away instead of dumping stack traces later on that
2915 * will obscure the real source of the problem. We test
2916 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2917 * the latter could be true if the filesystem is mounted
2918 * read-only, and in that case, ext4_da_writepages should
2919 * *never* be called, so if that ever happens, we would want
2920 * the stack trace.
2921 */
2922 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2923 return -EROFS;
2924
2925 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2926 range_whole = 1;
2927
2928 range_cyclic = wbc->range_cyclic;
2929 if (wbc->range_cyclic) {
2930 index = mapping->writeback_index;
2931 if (index)
2932 cycled = 0;
2933 wbc->range_start = index << PAGE_CACHE_SHIFT;
2934 wbc->range_end = LLONG_MAX;
2935 wbc->range_cyclic = 0;
2936 end = -1;
2937 } else {
2938 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2939 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2940 }
2941
2942 /*
2943 * This works around two forms of stupidity. The first is in
2944 * the writeback code, which caps the maximum number of pages
2945 * written to be 1024 pages. This is wrong on multiple
2946 * levels; different architectues have a different page size,
2947 * which changes the maximum amount of data which gets
2948 * written. Secondly, 4 megabytes is way too small. XFS
2949 * forces this value to be 16 megabytes by multiplying
2950 * nr_to_write parameter by four, and then relies on its
2951 * allocator to allocate larger extents to make them
2952 * contiguous. Unfortunately this brings us to the second
2953 * stupidity, which is that ext4's mballoc code only allocates
2954 * at most 2048 blocks. So we force contiguous writes up to
2955 * the number of dirty blocks in the inode, or
2956 * sbi->max_writeback_mb_bump whichever is smaller.
2957 */
2958 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2959 if (!range_cyclic && range_whole) {
2960 if (wbc->nr_to_write == LONG_MAX)
2961 desired_nr_to_write = wbc->nr_to_write;
2962 else
2963 desired_nr_to_write = wbc->nr_to_write * 8;
2964 } else
2965 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2966 max_pages);
2967 if (desired_nr_to_write > max_pages)
2968 desired_nr_to_write = max_pages;
2969
2970 if (wbc->nr_to_write < desired_nr_to_write) {
2971 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2972 wbc->nr_to_write = desired_nr_to_write;
2973 }
2974
2975 retry:
2976 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2977 tag_pages_for_writeback(mapping, index, end);
2978
2979 while (!ret && wbc->nr_to_write > 0) {
2980
2981 /*
2982 * we insert one extent at a time. So we need
2983 * credit needed for single extent allocation.
2984 * journalled mode is currently not supported
2985 * by delalloc
2986 */
2987 BUG_ON(ext4_should_journal_data(inode));
2988 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2989
2990 /* start a new transaction*/
2991 handle = ext4_journal_start(inode, needed_blocks);
2992 if (IS_ERR(handle)) {
2993 ret = PTR_ERR(handle);
2994 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2995 "%ld pages, ino %lu; err %d", __func__,
2996 wbc->nr_to_write, inode->i_ino, ret);
2997 goto out_writepages;
2998 }
2999
3000 /*
3001 * Now call write_cache_pages_da() to find the next
3002 * contiguous region of logical blocks that need
3003 * blocks to be allocated by ext4 and submit them.
3004 */
3005 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
3006 /*
3007 * If we have a contiguous extent of pages and we
3008 * haven't done the I/O yet, map the blocks and submit
3009 * them for I/O.
3010 */
3011 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3012 mpage_da_map_and_submit(&mpd);
3013 ret = MPAGE_DA_EXTENT_TAIL;
3014 }
3015 trace_ext4_da_write_pages(inode, &mpd);
3016 wbc->nr_to_write -= mpd.pages_written;
3017
3018 ext4_journal_stop(handle);
3019
3020 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3021 /* commit the transaction which would
3022 * free blocks released in the transaction
3023 * and try again
3024 */
3025 jbd2_journal_force_commit_nested(sbi->s_journal);
3026 ret = 0;
3027 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3028 /*
3029 * got one extent now try with
3030 * rest of the pages
3031 */
3032 pages_written += mpd.pages_written;
3033 ret = 0;
3034 io_done = 1;
3035 } else if (wbc->nr_to_write)
3036 /*
3037 * There is no more writeout needed
3038 * or we requested for a noblocking writeout
3039 * and we found the device congested
3040 */
3041 break;
3042 }
3043 if (!io_done && !cycled) {
3044 cycled = 1;
3045 index = 0;
3046 wbc->range_start = index << PAGE_CACHE_SHIFT;
3047 wbc->range_end = mapping->writeback_index - 1;
3048 goto retry;
3049 }
3050
3051 /* Update index */
3052 wbc->range_cyclic = range_cyclic;
3053 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3054 /*
3055 * set the writeback_index so that range_cyclic
3056 * mode will write it back later
3057 */
3058 mapping->writeback_index = done_index;
3059
3060 out_writepages:
3061 wbc->nr_to_write -= nr_to_writebump;
3062 wbc->range_start = range_start;
3063 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3064 return ret;
3065 }
3066
3067 #define FALL_BACK_TO_NONDELALLOC 1
3068 static int ext4_nonda_switch(struct super_block *sb)
3069 {
3070 s64 free_blocks, dirty_blocks;
3071 struct ext4_sb_info *sbi = EXT4_SB(sb);
3072
3073 /*
3074 * switch to non delalloc mode if we are running low
3075 * on free block. The free block accounting via percpu
3076 * counters can get slightly wrong with percpu_counter_batch getting
3077 * accumulated on each CPU without updating global counters
3078 * Delalloc need an accurate free block accounting. So switch
3079 * to non delalloc when we are near to error range.
3080 */
3081 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3082 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3083 if (2 * free_blocks < 3 * dirty_blocks ||
3084 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3085 /*
3086 * free block count is less than 150% of dirty blocks
3087 * or free blocks is less than watermark
3088 */
3089 return 1;
3090 }
3091 /*
3092 * Even if we don't switch but are nearing capacity,
3093 * start pushing delalloc when 1/2 of free blocks are dirty.
3094 */
3095 if (free_blocks < 2 * dirty_blocks)
3096 writeback_inodes_sb_if_idle(sb);
3097
3098 return 0;
3099 }
3100
3101 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3102 loff_t pos, unsigned len, unsigned flags,
3103 struct page **pagep, void **fsdata)
3104 {
3105 int ret, retries = 0;
3106 struct page *page;
3107 pgoff_t index;
3108 struct inode *inode = mapping->host;
3109 handle_t *handle;
3110
3111 index = pos >> PAGE_CACHE_SHIFT;
3112
3113 if (ext4_nonda_switch(inode->i_sb)) {
3114 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3115 return ext4_write_begin(file, mapping, pos,
3116 len, flags, pagep, fsdata);
3117 }
3118 *fsdata = (void *)0;
3119 trace_ext4_da_write_begin(inode, pos, len, flags);
3120 retry:
3121 /*
3122 * With delayed allocation, we don't log the i_disksize update
3123 * if there is delayed block allocation. But we still need
3124 * to journalling the i_disksize update if writes to the end
3125 * of file which has an already mapped buffer.
3126 */
3127 handle = ext4_journal_start(inode, 1);
3128 if (IS_ERR(handle)) {
3129 ret = PTR_ERR(handle);
3130 goto out;
3131 }
3132 /* We cannot recurse into the filesystem as the transaction is already
3133 * started */
3134 flags |= AOP_FLAG_NOFS;
3135
3136 page = grab_cache_page_write_begin(mapping, index, flags);
3137 if (!page) {
3138 ext4_journal_stop(handle);
3139 ret = -ENOMEM;
3140 goto out;
3141 }
3142 *pagep = page;
3143
3144 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3145 if (ret < 0) {
3146 unlock_page(page);
3147 ext4_journal_stop(handle);
3148 page_cache_release(page);
3149 /*
3150 * block_write_begin may have instantiated a few blocks
3151 * outside i_size. Trim these off again. Don't need
3152 * i_size_read because we hold i_mutex.
3153 */
3154 if (pos + len > inode->i_size)
3155 ext4_truncate_failed_write(inode);
3156 }
3157
3158 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3159 goto retry;
3160 out:
3161 return ret;
3162 }
3163
3164 /*
3165 * Check if we should update i_disksize
3166 * when write to the end of file but not require block allocation
3167 */
3168 static int ext4_da_should_update_i_disksize(struct page *page,
3169 unsigned long offset)
3170 {
3171 struct buffer_head *bh;
3172 struct inode *inode = page->mapping->host;
3173 unsigned int idx;
3174 int i;
3175
3176 bh = page_buffers(page);
3177 idx = offset >> inode->i_blkbits;
3178
3179 for (i = 0; i < idx; i++)
3180 bh = bh->b_this_page;
3181
3182 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3183 return 0;
3184 return 1;
3185 }
3186
3187 static int ext4_da_write_end(struct file *file,
3188 struct address_space *mapping,
3189 loff_t pos, unsigned len, unsigned copied,
3190 struct page *page, void *fsdata)
3191 {
3192 struct inode *inode = mapping->host;
3193 int ret = 0, ret2;
3194 handle_t *handle = ext4_journal_current_handle();
3195 loff_t new_i_size;
3196 unsigned long start, end;
3197 int write_mode = (int)(unsigned long)fsdata;
3198
3199 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3200 if (ext4_should_order_data(inode)) {
3201 return ext4_ordered_write_end(file, mapping, pos,
3202 len, copied, page, fsdata);
3203 } else if (ext4_should_writeback_data(inode)) {
3204 return ext4_writeback_write_end(file, mapping, pos,
3205 len, copied, page, fsdata);
3206 } else {
3207 BUG();
3208 }
3209 }
3210
3211 trace_ext4_da_write_end(inode, pos, len, copied);
3212 start = pos & (PAGE_CACHE_SIZE - 1);
3213 end = start + copied - 1;
3214
3215 /*
3216 * generic_write_end() will run mark_inode_dirty() if i_size
3217 * changes. So let's piggyback the i_disksize mark_inode_dirty
3218 * into that.
3219 */
3220
3221 new_i_size = pos + copied;
3222 if (new_i_size > EXT4_I(inode)->i_disksize) {
3223 if (ext4_da_should_update_i_disksize(page, end)) {
3224 down_write(&EXT4_I(inode)->i_data_sem);
3225 if (new_i_size > EXT4_I(inode)->i_disksize) {
3226 /*
3227 * Updating i_disksize when extending file
3228 * without needing block allocation
3229 */
3230 if (ext4_should_order_data(inode))
3231 ret = ext4_jbd2_file_inode(handle,
3232 inode);
3233
3234 EXT4_I(inode)->i_disksize = new_i_size;
3235 }
3236 up_write(&EXT4_I(inode)->i_data_sem);
3237 /* We need to mark inode dirty even if
3238 * new_i_size is less that inode->i_size
3239 * bu greater than i_disksize.(hint delalloc)
3240 */
3241 ext4_mark_inode_dirty(handle, inode);
3242 }
3243 }
3244 ret2 = generic_write_end(file, mapping, pos, len, copied,
3245 page, fsdata);
3246 copied = ret2;
3247 if (ret2 < 0)
3248 ret = ret2;
3249 ret2 = ext4_journal_stop(handle);
3250 if (!ret)
3251 ret = ret2;
3252
3253 return ret ? ret : copied;
3254 }
3255
3256 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3257 {
3258 /*
3259 * Drop reserved blocks
3260 */
3261 BUG_ON(!PageLocked(page));
3262 if (!page_has_buffers(page))
3263 goto out;
3264
3265 ext4_da_page_release_reservation(page, offset);
3266
3267 out:
3268 ext4_invalidatepage(page, offset);
3269
3270 return;
3271 }
3272
3273 /*
3274 * Force all delayed allocation blocks to be allocated for a given inode.
3275 */
3276 int ext4_alloc_da_blocks(struct inode *inode)
3277 {
3278 trace_ext4_alloc_da_blocks(inode);
3279
3280 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3281 !EXT4_I(inode)->i_reserved_meta_blocks)
3282 return 0;
3283
3284 /*
3285 * We do something simple for now. The filemap_flush() will
3286 * also start triggering a write of the data blocks, which is
3287 * not strictly speaking necessary (and for users of
3288 * laptop_mode, not even desirable). However, to do otherwise
3289 * would require replicating code paths in:
3290 *
3291 * ext4_da_writepages() ->
3292 * write_cache_pages() ---> (via passed in callback function)
3293 * __mpage_da_writepage() -->
3294 * mpage_add_bh_to_extent()
3295 * mpage_da_map_blocks()
3296 *
3297 * The problem is that write_cache_pages(), located in
3298 * mm/page-writeback.c, marks pages clean in preparation for
3299 * doing I/O, which is not desirable if we're not planning on
3300 * doing I/O at all.
3301 *
3302 * We could call write_cache_pages(), and then redirty all of
3303 * the pages by calling redirty_page_for_writepage() but that
3304 * would be ugly in the extreme. So instead we would need to
3305 * replicate parts of the code in the above functions,
3306 * simplifying them because we wouldn't actually intend to
3307 * write out the pages, but rather only collect contiguous
3308 * logical block extents, call the multi-block allocator, and
3309 * then update the buffer heads with the block allocations.
3310 *
3311 * For now, though, we'll cheat by calling filemap_flush(),
3312 * which will map the blocks, and start the I/O, but not
3313 * actually wait for the I/O to complete.
3314 */
3315 return filemap_flush(inode->i_mapping);
3316 }
3317
3318 /*
3319 * bmap() is special. It gets used by applications such as lilo and by
3320 * the swapper to find the on-disk block of a specific piece of data.
3321 *
3322 * Naturally, this is dangerous if the block concerned is still in the
3323 * journal. If somebody makes a swapfile on an ext4 data-journaling
3324 * filesystem and enables swap, then they may get a nasty shock when the
3325 * data getting swapped to that swapfile suddenly gets overwritten by
3326 * the original zero's written out previously to the journal and
3327 * awaiting writeback in the kernel's buffer cache.
3328 *
3329 * So, if we see any bmap calls here on a modified, data-journaled file,
3330 * take extra steps to flush any blocks which might be in the cache.
3331 */
3332 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3333 {
3334 struct inode *inode = mapping->host;
3335 journal_t *journal;
3336 int err;
3337
3338 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3339 test_opt(inode->i_sb, DELALLOC)) {
3340 /*
3341 * With delalloc we want to sync the file
3342 * so that we can make sure we allocate
3343 * blocks for file
3344 */
3345 filemap_write_and_wait(mapping);
3346 }
3347
3348 if (EXT4_JOURNAL(inode) &&
3349 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3350 /*
3351 * This is a REALLY heavyweight approach, but the use of
3352 * bmap on dirty files is expected to be extremely rare:
3353 * only if we run lilo or swapon on a freshly made file
3354 * do we expect this to happen.
3355 *
3356 * (bmap requires CAP_SYS_RAWIO so this does not
3357 * represent an unprivileged user DOS attack --- we'd be
3358 * in trouble if mortal users could trigger this path at
3359 * will.)
3360 *
3361 * NB. EXT4_STATE_JDATA is not set on files other than
3362 * regular files. If somebody wants to bmap a directory
3363 * or symlink and gets confused because the buffer
3364 * hasn't yet been flushed to disk, they deserve
3365 * everything they get.
3366 */
3367
3368 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3369 journal = EXT4_JOURNAL(inode);
3370 jbd2_journal_lock_updates(journal);
3371 err = jbd2_journal_flush(journal);
3372 jbd2_journal_unlock_updates(journal);
3373
3374 if (err)
3375 return 0;
3376 }
3377
3378 return generic_block_bmap(mapping, block, ext4_get_block);
3379 }
3380
3381 static int ext4_readpage(struct file *file, struct page *page)
3382 {
3383 trace_ext4_readpage(page);
3384 return mpage_readpage(page, ext4_get_block);
3385 }
3386
3387 static int
3388 ext4_readpages(struct file *file, struct address_space *mapping,
3389 struct list_head *pages, unsigned nr_pages)
3390 {
3391 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3392 }
3393
3394 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3395 {
3396 struct buffer_head *head, *bh;
3397 unsigned int curr_off = 0;
3398
3399 if (!page_has_buffers(page))
3400 return;
3401 head = bh = page_buffers(page);
3402 do {
3403 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3404 && bh->b_private) {
3405 ext4_free_io_end(bh->b_private);
3406 bh->b_private = NULL;
3407 bh->b_end_io = NULL;
3408 }
3409 curr_off = curr_off + bh->b_size;
3410 bh = bh->b_this_page;
3411 } while (bh != head);
3412 }
3413
3414 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3415 {
3416 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3417
3418 trace_ext4_invalidatepage(page, offset);
3419
3420 /*
3421 * free any io_end structure allocated for buffers to be discarded
3422 */
3423 if (ext4_should_dioread_nolock(page->mapping->host))
3424 ext4_invalidatepage_free_endio(page, offset);
3425 /*
3426 * If it's a full truncate we just forget about the pending dirtying
3427 */
3428 if (offset == 0)
3429 ClearPageChecked(page);
3430
3431 if (journal)
3432 jbd2_journal_invalidatepage(journal, page, offset);
3433 else
3434 block_invalidatepage(page, offset);
3435 }
3436
3437 static int ext4_releasepage(struct page *page, gfp_t wait)
3438 {
3439 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3440
3441 trace_ext4_releasepage(page);
3442
3443 WARN_ON(PageChecked(page));
3444 if (!page_has_buffers(page))
3445 return 0;
3446 if (journal)
3447 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3448 else
3449 return try_to_free_buffers(page);
3450 }
3451
3452 /*
3453 * O_DIRECT for ext3 (or indirect map) based files
3454 *
3455 * If the O_DIRECT write will extend the file then add this inode to the
3456 * orphan list. So recovery will truncate it back to the original size
3457 * if the machine crashes during the write.
3458 *
3459 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3460 * crashes then stale disk data _may_ be exposed inside the file. But current
3461 * VFS code falls back into buffered path in that case so we are safe.
3462 */
3463 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3464 const struct iovec *iov, loff_t offset,
3465 unsigned long nr_segs)
3466 {
3467 struct file *file = iocb->ki_filp;
3468 struct inode *inode = file->f_mapping->host;
3469 struct ext4_inode_info *ei = EXT4_I(inode);
3470 handle_t *handle;
3471 ssize_t ret;
3472 int orphan = 0;
3473 size_t count = iov_length(iov, nr_segs);
3474 int retries = 0;
3475
3476 if (rw == WRITE) {
3477 loff_t final_size = offset + count;
3478
3479 if (final_size > inode->i_size) {
3480 /* Credits for sb + inode write */
3481 handle = ext4_journal_start(inode, 2);
3482 if (IS_ERR(handle)) {
3483 ret = PTR_ERR(handle);
3484 goto out;
3485 }
3486 ret = ext4_orphan_add(handle, inode);
3487 if (ret) {
3488 ext4_journal_stop(handle);
3489 goto out;
3490 }
3491 orphan = 1;
3492 ei->i_disksize = inode->i_size;
3493 ext4_journal_stop(handle);
3494 }
3495 }
3496
3497 retry:
3498 if (rw == READ && ext4_should_dioread_nolock(inode))
3499 ret = __blockdev_direct_IO(rw, iocb, inode,
3500 inode->i_sb->s_bdev, iov,
3501 offset, nr_segs,
3502 ext4_get_block, NULL, NULL, 0);
3503 else {
3504 ret = blockdev_direct_IO(rw, iocb, inode, iov,
3505 offset, nr_segs, ext4_get_block);
3506
3507 if (unlikely((rw & WRITE) && ret < 0)) {
3508 loff_t isize = i_size_read(inode);
3509 loff_t end = offset + iov_length(iov, nr_segs);
3510
3511 if (end > isize)
3512 ext4_truncate_failed_write(inode);
3513 }
3514 }
3515 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3516 goto retry;
3517
3518 if (orphan) {
3519 int err;
3520
3521 /* Credits for sb + inode write */
3522 handle = ext4_journal_start(inode, 2);
3523 if (IS_ERR(handle)) {
3524 /* This is really bad luck. We've written the data
3525 * but cannot extend i_size. Bail out and pretend
3526 * the write failed... */
3527 ret = PTR_ERR(handle);
3528 if (inode->i_nlink)
3529 ext4_orphan_del(NULL, inode);
3530
3531 goto out;
3532 }
3533 if (inode->i_nlink)
3534 ext4_orphan_del(handle, inode);
3535 if (ret > 0) {
3536 loff_t end = offset + ret;
3537 if (end > inode->i_size) {
3538 ei->i_disksize = end;
3539 i_size_write(inode, end);
3540 /*
3541 * We're going to return a positive `ret'
3542 * here due to non-zero-length I/O, so there's
3543 * no way of reporting error returns from
3544 * ext4_mark_inode_dirty() to userspace. So
3545 * ignore it.
3546 */
3547 ext4_mark_inode_dirty(handle, inode);
3548 }
3549 }
3550 err = ext4_journal_stop(handle);
3551 if (ret == 0)
3552 ret = err;
3553 }
3554 out:
3555 return ret;
3556 }
3557
3558 /*
3559 * ext4_get_block used when preparing for a DIO write or buffer write.
3560 * We allocate an uinitialized extent if blocks haven't been allocated.
3561 * The extent will be converted to initialized after the IO is complete.
3562 */
3563 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3564 struct buffer_head *bh_result, int create)
3565 {
3566 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3567 inode->i_ino, create);
3568 return _ext4_get_block(inode, iblock, bh_result,
3569 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3570 }
3571
3572 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3573 ssize_t size, void *private, int ret,
3574 bool is_async)
3575 {
3576 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
3577 ext4_io_end_t *io_end = iocb->private;
3578 struct workqueue_struct *wq;
3579 unsigned long flags;
3580 struct ext4_inode_info *ei;
3581
3582 /* if not async direct IO or dio with 0 bytes write, just return */
3583 if (!io_end || !size)
3584 goto out;
3585
3586 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3587 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3588 iocb->private, io_end->inode->i_ino, iocb, offset,
3589 size);
3590
3591 /* if not aio dio with unwritten extents, just free io and return */
3592 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3593 ext4_free_io_end(io_end);
3594 iocb->private = NULL;
3595 out:
3596 if (is_async)
3597 aio_complete(iocb, ret, 0);
3598 inode_dio_done(inode);
3599 return;
3600 }
3601
3602 io_end->offset = offset;
3603 io_end->size = size;
3604 if (is_async) {
3605 io_end->iocb = iocb;
3606 io_end->result = ret;
3607 }
3608 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3609
3610 /* Add the io_end to per-inode completed aio dio list*/
3611 ei = EXT4_I(io_end->inode);
3612 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3613 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3614 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3615
3616 /* queue the work to convert unwritten extents to written */
3617 queue_work(wq, &io_end->work);
3618 iocb->private = NULL;
3619
3620 /* XXX: probably should move into the real I/O completion handler */
3621 inode_dio_done(inode);
3622 }
3623
3624 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3625 {
3626 ext4_io_end_t *io_end = bh->b_private;
3627 struct workqueue_struct *wq;
3628 struct inode *inode;
3629 unsigned long flags;
3630
3631 if (!test_clear_buffer_uninit(bh) || !io_end)
3632 goto out;
3633
3634 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3635 printk("sb umounted, discard end_io request for inode %lu\n",
3636 io_end->inode->i_ino);
3637 ext4_free_io_end(io_end);
3638 goto out;
3639 }
3640
3641 io_end->flag = EXT4_IO_END_UNWRITTEN;
3642 inode = io_end->inode;
3643
3644 /* Add the io_end to per-inode completed io list*/
3645 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3646 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3647 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3648
3649 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3650 /* queue the work to convert unwritten extents to written */
3651 queue_work(wq, &io_end->work);
3652 out:
3653 bh->b_private = NULL;
3654 bh->b_end_io = NULL;
3655 clear_buffer_uninit(bh);
3656 end_buffer_async_write(bh, uptodate);
3657 }
3658
3659 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3660 {
3661 ext4_io_end_t *io_end;
3662 struct page *page = bh->b_page;
3663 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3664 size_t size = bh->b_size;
3665
3666 retry:
3667 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3668 if (!io_end) {
3669 pr_warn_ratelimited("%s: allocation fail\n", __func__);
3670 schedule();
3671 goto retry;
3672 }
3673 io_end->offset = offset;
3674 io_end->size = size;
3675 /*
3676 * We need to hold a reference to the page to make sure it
3677 * doesn't get evicted before ext4_end_io_work() has a chance
3678 * to convert the extent from written to unwritten.
3679 */
3680 io_end->page = page;
3681 get_page(io_end->page);
3682
3683 bh->b_private = io_end;
3684 bh->b_end_io = ext4_end_io_buffer_write;
3685 return 0;
3686 }
3687
3688 /*
3689 * For ext4 extent files, ext4 will do direct-io write to holes,
3690 * preallocated extents, and those write extend the file, no need to
3691 * fall back to buffered IO.
3692 *
3693 * For holes, we fallocate those blocks, mark them as uninitialized
3694 * If those blocks were preallocated, we mark sure they are splited, but
3695 * still keep the range to write as uninitialized.
3696 *
3697 * The unwrritten extents will be converted to written when DIO is completed.
3698 * For async direct IO, since the IO may still pending when return, we
3699 * set up an end_io call back function, which will do the conversion
3700 * when async direct IO completed.
3701 *
3702 * If the O_DIRECT write will extend the file then add this inode to the
3703 * orphan list. So recovery will truncate it back to the original size
3704 * if the machine crashes during the write.
3705 *
3706 */
3707 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3708 const struct iovec *iov, loff_t offset,
3709 unsigned long nr_segs)
3710 {
3711 struct file *file = iocb->ki_filp;
3712 struct inode *inode = file->f_mapping->host;
3713 ssize_t ret;
3714 size_t count = iov_length(iov, nr_segs);
3715
3716 loff_t final_size = offset + count;
3717 if (rw == WRITE && final_size <= inode->i_size) {
3718 /*
3719 * We could direct write to holes and fallocate.
3720 *
3721 * Allocated blocks to fill the hole are marked as uninitialized
3722 * to prevent parallel buffered read to expose the stale data
3723 * before DIO complete the data IO.
3724 *
3725 * As to previously fallocated extents, ext4 get_block
3726 * will just simply mark the buffer mapped but still
3727 * keep the extents uninitialized.
3728 *
3729 * for non AIO case, we will convert those unwritten extents
3730 * to written after return back from blockdev_direct_IO.
3731 *
3732 * for async DIO, the conversion needs to be defered when
3733 * the IO is completed. The ext4 end_io callback function
3734 * will be called to take care of the conversion work.
3735 * Here for async case, we allocate an io_end structure to
3736 * hook to the iocb.
3737 */
3738 iocb->private = NULL;
3739 EXT4_I(inode)->cur_aio_dio = NULL;
3740 if (!is_sync_kiocb(iocb)) {
3741 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3742 if (!iocb->private)
3743 return -ENOMEM;
3744 /*
3745 * we save the io structure for current async
3746 * direct IO, so that later ext4_map_blocks()
3747 * could flag the io structure whether there
3748 * is a unwritten extents needs to be converted
3749 * when IO is completed.
3750 */
3751 EXT4_I(inode)->cur_aio_dio = iocb->private;
3752 }
3753
3754 ret = __blockdev_direct_IO(rw, iocb, inode,
3755 inode->i_sb->s_bdev, iov,
3756 offset, nr_segs,
3757 ext4_get_block_write,
3758 ext4_end_io_dio,
3759 NULL,
3760 DIO_LOCKING | DIO_SKIP_HOLES);
3761 if (iocb->private)
3762 EXT4_I(inode)->cur_aio_dio = NULL;
3763 /*
3764 * The io_end structure takes a reference to the inode,
3765 * that structure needs to be destroyed and the
3766 * reference to the inode need to be dropped, when IO is
3767 * complete, even with 0 byte write, or failed.
3768 *
3769 * In the successful AIO DIO case, the io_end structure will be
3770 * desctroyed and the reference to the inode will be dropped
3771 * after the end_io call back function is called.
3772 *
3773 * In the case there is 0 byte write, or error case, since
3774 * VFS direct IO won't invoke the end_io call back function,
3775 * we need to free the end_io structure here.
3776 */
3777 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3778 ext4_free_io_end(iocb->private);
3779 iocb->private = NULL;
3780 } else if (ret > 0 && ext4_test_inode_state(inode,
3781 EXT4_STATE_DIO_UNWRITTEN)) {
3782 int err;
3783 /*
3784 * for non AIO case, since the IO is already
3785 * completed, we could do the conversion right here
3786 */
3787 err = ext4_convert_unwritten_extents(inode,
3788 offset, ret);
3789 if (err < 0)
3790 ret = err;
3791 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3792 }
3793 return ret;
3794 }
3795
3796 /* for write the the end of file case, we fall back to old way */
3797 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3798 }
3799
3800 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3801 const struct iovec *iov, loff_t offset,
3802 unsigned long nr_segs)
3803 {
3804 struct file *file = iocb->ki_filp;
3805 struct inode *inode = file->f_mapping->host;
3806 ssize_t ret;
3807
3808 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3809 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3810 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3811 else
3812 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3813 trace_ext4_direct_IO_exit(inode, offset,
3814 iov_length(iov, nr_segs), rw, ret);
3815 return ret;
3816 }
3817
3818 /*
3819 * Pages can be marked dirty completely asynchronously from ext4's journalling
3820 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3821 * much here because ->set_page_dirty is called under VFS locks. The page is
3822 * not necessarily locked.
3823 *
3824 * We cannot just dirty the page and leave attached buffers clean, because the
3825 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3826 * or jbddirty because all the journalling code will explode.
3827 *
3828 * So what we do is to mark the page "pending dirty" and next time writepage
3829 * is called, propagate that into the buffers appropriately.
3830 */
3831 static int ext4_journalled_set_page_dirty(struct page *page)
3832 {
3833 SetPageChecked(page);
3834 return __set_page_dirty_nobuffers(page);
3835 }
3836
3837 static const struct address_space_operations ext4_ordered_aops = {
3838 .readpage = ext4_readpage,
3839 .readpages = ext4_readpages,
3840 .writepage = ext4_writepage,
3841 .write_begin = ext4_write_begin,
3842 .write_end = ext4_ordered_write_end,
3843 .bmap = ext4_bmap,
3844 .invalidatepage = ext4_invalidatepage,
3845 .releasepage = ext4_releasepage,
3846 .direct_IO = ext4_direct_IO,
3847 .migratepage = buffer_migrate_page,
3848 .is_partially_uptodate = block_is_partially_uptodate,
3849 .error_remove_page = generic_error_remove_page,
3850 };
3851
3852 static const struct address_space_operations ext4_writeback_aops = {
3853 .readpage = ext4_readpage,
3854 .readpages = ext4_readpages,
3855 .writepage = ext4_writepage,
3856 .write_begin = ext4_write_begin,
3857 .write_end = ext4_writeback_write_end,
3858 .bmap = ext4_bmap,
3859 .invalidatepage = ext4_invalidatepage,
3860 .releasepage = ext4_releasepage,
3861 .direct_IO = ext4_direct_IO,
3862 .migratepage = buffer_migrate_page,
3863 .is_partially_uptodate = block_is_partially_uptodate,
3864 .error_remove_page = generic_error_remove_page,
3865 };
3866
3867 static const struct address_space_operations ext4_journalled_aops = {
3868 .readpage = ext4_readpage,
3869 .readpages = ext4_readpages,
3870 .writepage = ext4_writepage,
3871 .write_begin = ext4_write_begin,
3872 .write_end = ext4_journalled_write_end,
3873 .set_page_dirty = ext4_journalled_set_page_dirty,
3874 .bmap = ext4_bmap,
3875 .invalidatepage = ext4_invalidatepage,
3876 .releasepage = ext4_releasepage,
3877 .is_partially_uptodate = block_is_partially_uptodate,
3878 .error_remove_page = generic_error_remove_page,
3879 };
3880
3881 static const struct address_space_operations ext4_da_aops = {
3882 .readpage = ext4_readpage,
3883 .readpages = ext4_readpages,
3884 .writepage = ext4_writepage,
3885 .writepages = ext4_da_writepages,
3886 .write_begin = ext4_da_write_begin,
3887 .write_end = ext4_da_write_end,
3888 .bmap = ext4_bmap,
3889 .invalidatepage = ext4_da_invalidatepage,
3890 .releasepage = ext4_releasepage,
3891 .direct_IO = ext4_direct_IO,
3892 .migratepage = buffer_migrate_page,
3893 .is_partially_uptodate = block_is_partially_uptodate,
3894 .error_remove_page = generic_error_remove_page,
3895 };
3896
3897 void ext4_set_aops(struct inode *inode)
3898 {
3899 if (ext4_should_order_data(inode) &&
3900 test_opt(inode->i_sb, DELALLOC))
3901 inode->i_mapping->a_ops = &ext4_da_aops;
3902 else if (ext4_should_order_data(inode))
3903 inode->i_mapping->a_ops = &ext4_ordered_aops;
3904 else if (ext4_should_writeback_data(inode) &&
3905 test_opt(inode->i_sb, DELALLOC))
3906 inode->i_mapping->a_ops = &ext4_da_aops;
3907 else if (ext4_should_writeback_data(inode))
3908 inode->i_mapping->a_ops = &ext4_writeback_aops;
3909 else
3910 inode->i_mapping->a_ops = &ext4_journalled_aops;
3911 }
3912
3913 /*
3914 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3915 * up to the end of the block which corresponds to `from'.
3916 * This required during truncate. We need to physically zero the tail end
3917 * of that block so it doesn't yield old data if the file is later grown.
3918 */
3919 int ext4_block_truncate_page(handle_t *handle,
3920 struct address_space *mapping, loff_t from)
3921 {
3922 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3923 unsigned length;
3924 unsigned blocksize;
3925 struct inode *inode = mapping->host;
3926
3927 blocksize = inode->i_sb->s_blocksize;
3928 length = blocksize - (offset & (blocksize - 1));
3929
3930 return ext4_block_zero_page_range(handle, mapping, from, length);
3931 }
3932
3933 /*
3934 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3935 * starting from file offset 'from'. The range to be zero'd must
3936 * be contained with in one block. If the specified range exceeds
3937 * the end of the block it will be shortened to end of the block
3938 * that cooresponds to 'from'
3939 */
3940 int ext4_block_zero_page_range(handle_t *handle,
3941 struct address_space *mapping, loff_t from, loff_t length)
3942 {
3943 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3944 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3945 unsigned blocksize, max, pos;
3946 ext4_lblk_t iblock;
3947 struct inode *inode = mapping->host;
3948 struct buffer_head *bh;
3949 struct page *page;
3950 int err = 0;
3951
3952 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3953 mapping_gfp_mask(mapping) & ~__GFP_FS);
3954 if (!page)
3955 return -EINVAL;
3956
3957 blocksize = inode->i_sb->s_blocksize;
3958 max = blocksize - (offset & (blocksize - 1));
3959
3960 /*
3961 * correct length if it does not fall between
3962 * 'from' and the end of the block
3963 */
3964 if (length > max || length < 0)
3965 length = max;
3966
3967 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3968
3969 if (!page_has_buffers(page))
3970 create_empty_buffers(page, blocksize, 0);
3971
3972 /* Find the buffer that contains "offset" */
3973 bh = page_buffers(page);
3974 pos = blocksize;
3975 while (offset >= pos) {
3976 bh = bh->b_this_page;
3977 iblock++;
3978 pos += blocksize;
3979 }
3980
3981 err = 0;
3982 if (buffer_freed(bh)) {
3983 BUFFER_TRACE(bh, "freed: skip");
3984 goto unlock;
3985 }
3986
3987 if (!buffer_mapped(bh)) {
3988 BUFFER_TRACE(bh, "unmapped");
3989 ext4_get_block(inode, iblock, bh, 0);
3990 /* unmapped? It's a hole - nothing to do */
3991 if (!buffer_mapped(bh)) {
3992 BUFFER_TRACE(bh, "still unmapped");
3993 goto unlock;
3994 }
3995 }
3996
3997 /* Ok, it's mapped. Make sure it's up-to-date */
3998 if (PageUptodate(page))
3999 set_buffer_uptodate(bh);
4000
4001 if (!buffer_uptodate(bh)) {
4002 err = -EIO;
4003 ll_rw_block(READ, 1, &bh);
4004 wait_on_buffer(bh);
4005 /* Uhhuh. Read error. Complain and punt. */
4006 if (!buffer_uptodate(bh))
4007 goto unlock;
4008 }
4009
4010 if (ext4_should_journal_data(inode)) {
4011 BUFFER_TRACE(bh, "get write access");
4012 err = ext4_journal_get_write_access(handle, bh);
4013 if (err)
4014 goto unlock;
4015 }
4016
4017 zero_user(page, offset, length);
4018
4019 BUFFER_TRACE(bh, "zeroed end of block");
4020
4021 err = 0;
4022 if (ext4_should_journal_data(inode)) {
4023 err = ext4_handle_dirty_metadata(handle, inode, bh);
4024 } else {
4025 if (ext4_should_order_data(inode) && EXT4_I(inode)->jinode)
4026 err = ext4_jbd2_file_inode(handle, inode);
4027 mark_buffer_dirty(bh);
4028 }
4029
4030 unlock:
4031 unlock_page(page);
4032 page_cache_release(page);
4033 return err;
4034 }
4035
4036 /*
4037 * Probably it should be a library function... search for first non-zero word
4038 * or memcmp with zero_page, whatever is better for particular architecture.
4039 * Linus?
4040 */
4041 static inline int all_zeroes(__le32 *p, __le32 *q)
4042 {
4043 while (p < q)
4044 if (*p++)
4045 return 0;
4046 return 1;
4047 }
4048
4049 /**
4050 * ext4_find_shared - find the indirect blocks for partial truncation.
4051 * @inode: inode in question
4052 * @depth: depth of the affected branch
4053 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4054 * @chain: place to store the pointers to partial indirect blocks
4055 * @top: place to the (detached) top of branch
4056 *
4057 * This is a helper function used by ext4_truncate().
4058 *
4059 * When we do truncate() we may have to clean the ends of several
4060 * indirect blocks but leave the blocks themselves alive. Block is
4061 * partially truncated if some data below the new i_size is referred
4062 * from it (and it is on the path to the first completely truncated
4063 * data block, indeed). We have to free the top of that path along
4064 * with everything to the right of the path. Since no allocation
4065 * past the truncation point is possible until ext4_truncate()
4066 * finishes, we may safely do the latter, but top of branch may
4067 * require special attention - pageout below the truncation point
4068 * might try to populate it.
4069 *
4070 * We atomically detach the top of branch from the tree, store the
4071 * block number of its root in *@top, pointers to buffer_heads of
4072 * partially truncated blocks - in @chain[].bh and pointers to
4073 * their last elements that should not be removed - in
4074 * @chain[].p. Return value is the pointer to last filled element
4075 * of @chain.
4076 *
4077 * The work left to caller to do the actual freeing of subtrees:
4078 * a) free the subtree starting from *@top
4079 * b) free the subtrees whose roots are stored in
4080 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4081 * c) free the subtrees growing from the inode past the @chain[0].
4082 * (no partially truncated stuff there). */
4083
4084 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4085 ext4_lblk_t offsets[4], Indirect chain[4],
4086 __le32 *top)
4087 {
4088 Indirect *partial, *p;
4089 int k, err;
4090
4091 *top = 0;
4092 /* Make k index the deepest non-null offset + 1 */
4093 for (k = depth; k > 1 && !offsets[k-1]; k--)
4094 ;
4095 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4096 /* Writer: pointers */
4097 if (!partial)
4098 partial = chain + k-1;
4099 /*
4100 * If the branch acquired continuation since we've looked at it -
4101 * fine, it should all survive and (new) top doesn't belong to us.
4102 */
4103 if (!partial->key && *partial->p)
4104 /* Writer: end */
4105 goto no_top;
4106 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4107 ;
4108 /*
4109 * OK, we've found the last block that must survive. The rest of our
4110 * branch should be detached before unlocking. However, if that rest
4111 * of branch is all ours and does not grow immediately from the inode
4112 * it's easier to cheat and just decrement partial->p.
4113 */
4114 if (p == chain + k - 1 && p > chain) {
4115 p->p--;
4116 } else {
4117 *top = *p->p;
4118 /* Nope, don't do this in ext4. Must leave the tree intact */
4119 #if 0
4120 *p->p = 0;
4121 #endif
4122 }
4123 /* Writer: end */
4124
4125 while (partial > p) {
4126 brelse(partial->bh);
4127 partial--;
4128 }
4129 no_top:
4130 return partial;
4131 }
4132
4133 /*
4134 * Zero a number of block pointers in either an inode or an indirect block.
4135 * If we restart the transaction we must again get write access to the
4136 * indirect block for further modification.
4137 *
4138 * We release `count' blocks on disk, but (last - first) may be greater
4139 * than `count' because there can be holes in there.
4140 *
4141 * Return 0 on success, 1 on invalid block range
4142 * and < 0 on fatal error.
4143 */
4144 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4145 struct buffer_head *bh,
4146 ext4_fsblk_t block_to_free,
4147 unsigned long count, __le32 *first,
4148 __le32 *last)
4149 {
4150 __le32 *p;
4151 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4152 int err;
4153
4154 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4155 flags |= EXT4_FREE_BLOCKS_METADATA;
4156
4157 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4158 count)) {
4159 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4160 "blocks %llu len %lu",
4161 (unsigned long long) block_to_free, count);
4162 return 1;
4163 }
4164
4165 if (try_to_extend_transaction(handle, inode)) {
4166 if (bh) {
4167 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4168 err = ext4_handle_dirty_metadata(handle, inode, bh);
4169 if (unlikely(err))
4170 goto out_err;
4171 }
4172 err = ext4_mark_inode_dirty(handle, inode);
4173 if (unlikely(err))
4174 goto out_err;
4175 err = ext4_truncate_restart_trans(handle, inode,
4176 blocks_for_truncate(inode));
4177 if (unlikely(err))
4178 goto out_err;
4179 if (bh) {
4180 BUFFER_TRACE(bh, "retaking write access");
4181 err = ext4_journal_get_write_access(handle, bh);
4182 if (unlikely(err))
4183 goto out_err;
4184 }
4185 }
4186
4187 for (p = first; p < last; p++)
4188 *p = 0;
4189
4190 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
4191 return 0;
4192 out_err:
4193 ext4_std_error(inode->i_sb, err);
4194 return err;
4195 }
4196
4197 /**
4198 * ext4_free_data - free a list of data blocks
4199 * @handle: handle for this transaction
4200 * @inode: inode we are dealing with
4201 * @this_bh: indirect buffer_head which contains *@first and *@last
4202 * @first: array of block numbers
4203 * @last: points immediately past the end of array
4204 *
4205 * We are freeing all blocks referred from that array (numbers are stored as
4206 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4207 *
4208 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4209 * blocks are contiguous then releasing them at one time will only affect one
4210 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4211 * actually use a lot of journal space.
4212 *
4213 * @this_bh will be %NULL if @first and @last point into the inode's direct
4214 * block pointers.
4215 */
4216 static void ext4_free_data(handle_t *handle, struct inode *inode,
4217 struct buffer_head *this_bh,
4218 __le32 *first, __le32 *last)
4219 {
4220 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4221 unsigned long count = 0; /* Number of blocks in the run */
4222 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4223 corresponding to
4224 block_to_free */
4225 ext4_fsblk_t nr; /* Current block # */
4226 __le32 *p; /* Pointer into inode/ind
4227 for current block */
4228 int err = 0;
4229
4230 if (this_bh) { /* For indirect block */
4231 BUFFER_TRACE(this_bh, "get_write_access");
4232 err = ext4_journal_get_write_access(handle, this_bh);
4233 /* Important: if we can't update the indirect pointers
4234 * to the blocks, we can't free them. */
4235 if (err)
4236 return;
4237 }
4238
4239 for (p = first; p < last; p++) {
4240 nr = le32_to_cpu(*p);
4241 if (nr) {
4242 /* accumulate blocks to free if they're contiguous */
4243 if (count == 0) {
4244 block_to_free = nr;
4245 block_to_free_p = p;
4246 count = 1;
4247 } else if (nr == block_to_free + count) {
4248 count++;
4249 } else {
4250 err = ext4_clear_blocks(handle, inode, this_bh,
4251 block_to_free, count,
4252 block_to_free_p, p);
4253 if (err)
4254 break;
4255 block_to_free = nr;
4256 block_to_free_p = p;
4257 count = 1;
4258 }
4259 }
4260 }
4261
4262 if (!err && count > 0)
4263 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4264 count, block_to_free_p, p);
4265 if (err < 0)
4266 /* fatal error */
4267 return;
4268
4269 if (this_bh) {
4270 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4271
4272 /*
4273 * The buffer head should have an attached journal head at this
4274 * point. However, if the data is corrupted and an indirect
4275 * block pointed to itself, it would have been detached when
4276 * the block was cleared. Check for this instead of OOPSing.
4277 */
4278 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4279 ext4_handle_dirty_metadata(handle, inode, this_bh);
4280 else
4281 EXT4_ERROR_INODE(inode,
4282 "circular indirect block detected at "
4283 "block %llu",
4284 (unsigned long long) this_bh->b_blocknr);
4285 }
4286 }
4287
4288 /**
4289 * ext4_free_branches - free an array of branches
4290 * @handle: JBD handle for this transaction
4291 * @inode: inode we are dealing with
4292 * @parent_bh: the buffer_head which contains *@first and *@last
4293 * @first: array of block numbers
4294 * @last: pointer immediately past the end of array
4295 * @depth: depth of the branches to free
4296 *
4297 * We are freeing all blocks referred from these branches (numbers are
4298 * stored as little-endian 32-bit) and updating @inode->i_blocks
4299 * appropriately.
4300 */
4301 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4302 struct buffer_head *parent_bh,
4303 __le32 *first, __le32 *last, int depth)
4304 {
4305 ext4_fsblk_t nr;
4306 __le32 *p;
4307
4308 if (ext4_handle_is_aborted(handle))
4309 return;
4310
4311 if (depth--) {
4312 struct buffer_head *bh;
4313 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4314 p = last;
4315 while (--p >= first) {
4316 nr = le32_to_cpu(*p);
4317 if (!nr)
4318 continue; /* A hole */
4319
4320 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4321 nr, 1)) {
4322 EXT4_ERROR_INODE(inode,
4323 "invalid indirect mapped "
4324 "block %lu (level %d)",
4325 (unsigned long) nr, depth);
4326 break;
4327 }
4328
4329 /* Go read the buffer for the next level down */
4330 bh = sb_bread(inode->i_sb, nr);
4331
4332 /*
4333 * A read failure? Report error and clear slot
4334 * (should be rare).
4335 */
4336 if (!bh) {
4337 EXT4_ERROR_INODE_BLOCK(inode, nr,
4338 "Read failure");
4339 continue;
4340 }
4341
4342 /* This zaps the entire block. Bottom up. */
4343 BUFFER_TRACE(bh, "free child branches");
4344 ext4_free_branches(handle, inode, bh,
4345 (__le32 *) bh->b_data,
4346 (__le32 *) bh->b_data + addr_per_block,
4347 depth);
4348 brelse(bh);
4349
4350 /*
4351 * Everything below this this pointer has been
4352 * released. Now let this top-of-subtree go.
4353 *
4354 * We want the freeing of this indirect block to be
4355 * atomic in the journal with the updating of the
4356 * bitmap block which owns it. So make some room in
4357 * the journal.
4358 *
4359 * We zero the parent pointer *after* freeing its
4360 * pointee in the bitmaps, so if extend_transaction()
4361 * for some reason fails to put the bitmap changes and
4362 * the release into the same transaction, recovery
4363 * will merely complain about releasing a free block,
4364 * rather than leaking blocks.
4365 */
4366 if (ext4_handle_is_aborted(handle))
4367 return;
4368 if (try_to_extend_transaction(handle, inode)) {
4369 ext4_mark_inode_dirty(handle, inode);
4370 ext4_truncate_restart_trans(handle, inode,
4371 blocks_for_truncate(inode));
4372 }
4373
4374 /*
4375 * The forget flag here is critical because if
4376 * we are journaling (and not doing data
4377 * journaling), we have to make sure a revoke
4378 * record is written to prevent the journal
4379 * replay from overwriting the (former)
4380 * indirect block if it gets reallocated as a
4381 * data block. This must happen in the same
4382 * transaction where the data blocks are
4383 * actually freed.
4384 */
4385 ext4_free_blocks(handle, inode, NULL, nr, 1,
4386 EXT4_FREE_BLOCKS_METADATA|
4387 EXT4_FREE_BLOCKS_FORGET);
4388
4389 if (parent_bh) {
4390 /*
4391 * The block which we have just freed is
4392 * pointed to by an indirect block: journal it
4393 */
4394 BUFFER_TRACE(parent_bh, "get_write_access");
4395 if (!ext4_journal_get_write_access(handle,
4396 parent_bh)){
4397 *p = 0;
4398 BUFFER_TRACE(parent_bh,
4399 "call ext4_handle_dirty_metadata");
4400 ext4_handle_dirty_metadata(handle,
4401 inode,
4402 parent_bh);
4403 }
4404 }
4405 }
4406 } else {
4407 /* We have reached the bottom of the tree. */
4408 BUFFER_TRACE(parent_bh, "free data blocks");
4409 ext4_free_data(handle, inode, parent_bh, first, last);
4410 }
4411 }
4412
4413 int ext4_can_truncate(struct inode *inode)
4414 {
4415 if (S_ISREG(inode->i_mode))
4416 return 1;
4417 if (S_ISDIR(inode->i_mode))
4418 return 1;
4419 if (S_ISLNK(inode->i_mode))
4420 return !ext4_inode_is_fast_symlink(inode);
4421 return 0;
4422 }
4423
4424 /*
4425 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
4426 * associated with the given offset and length
4427 *
4428 * @inode: File inode
4429 * @offset: The offset where the hole will begin
4430 * @len: The length of the hole
4431 *
4432 * Returns: 0 on sucess or negative on failure
4433 */
4434
4435 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4436 {
4437 struct inode *inode = file->f_path.dentry->d_inode;
4438 if (!S_ISREG(inode->i_mode))
4439 return -ENOTSUPP;
4440
4441 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4442 /* TODO: Add support for non extent hole punching */
4443 return -ENOTSUPP;
4444 }
4445
4446 return ext4_ext_punch_hole(file, offset, length);
4447 }
4448
4449 /*
4450 * ext4_truncate()
4451 *
4452 * We block out ext4_get_block() block instantiations across the entire
4453 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4454 * simultaneously on behalf of the same inode.
4455 *
4456 * As we work through the truncate and commmit bits of it to the journal there
4457 * is one core, guiding principle: the file's tree must always be consistent on
4458 * disk. We must be able to restart the truncate after a crash.
4459 *
4460 * The file's tree may be transiently inconsistent in memory (although it
4461 * probably isn't), but whenever we close off and commit a journal transaction,
4462 * the contents of (the filesystem + the journal) must be consistent and
4463 * restartable. It's pretty simple, really: bottom up, right to left (although
4464 * left-to-right works OK too).
4465 *
4466 * Note that at recovery time, journal replay occurs *before* the restart of
4467 * truncate against the orphan inode list.
4468 *
4469 * The committed inode has the new, desired i_size (which is the same as
4470 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4471 * that this inode's truncate did not complete and it will again call
4472 * ext4_truncate() to have another go. So there will be instantiated blocks
4473 * to the right of the truncation point in a crashed ext4 filesystem. But
4474 * that's fine - as long as they are linked from the inode, the post-crash
4475 * ext4_truncate() run will find them and release them.
4476 */
4477 void ext4_truncate(struct inode *inode)
4478 {
4479 handle_t *handle;
4480 struct ext4_inode_info *ei = EXT4_I(inode);
4481 __le32 *i_data = ei->i_data;
4482 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4483 struct address_space *mapping = inode->i_mapping;
4484 ext4_lblk_t offsets[4];
4485 Indirect chain[4];
4486 Indirect *partial;
4487 __le32 nr = 0;
4488 int n = 0;
4489 ext4_lblk_t last_block, max_block;
4490 unsigned blocksize = inode->i_sb->s_blocksize;
4491
4492 trace_ext4_truncate_enter(inode);
4493
4494 if (!ext4_can_truncate(inode))
4495 return;
4496
4497 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4498
4499 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4500 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4501
4502 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4503 ext4_ext_truncate(inode);
4504 trace_ext4_truncate_exit(inode);
4505 return;
4506 }
4507
4508 handle = start_transaction(inode);
4509 if (IS_ERR(handle))
4510 return; /* AKPM: return what? */
4511
4512 last_block = (inode->i_size + blocksize-1)
4513 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4514 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
4515 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4516
4517 if (inode->i_size & (blocksize - 1))
4518 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4519 goto out_stop;
4520
4521 if (last_block != max_block) {
4522 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4523 if (n == 0)
4524 goto out_stop; /* error */
4525 }
4526
4527 /*
4528 * OK. This truncate is going to happen. We add the inode to the
4529 * orphan list, so that if this truncate spans multiple transactions,
4530 * and we crash, we will resume the truncate when the filesystem
4531 * recovers. It also marks the inode dirty, to catch the new size.
4532 *
4533 * Implication: the file must always be in a sane, consistent
4534 * truncatable state while each transaction commits.
4535 */
4536 if (ext4_orphan_add(handle, inode))
4537 goto out_stop;
4538
4539 /*
4540 * From here we block out all ext4_get_block() callers who want to
4541 * modify the block allocation tree.
4542 */
4543 down_write(&ei->i_data_sem);
4544
4545 ext4_discard_preallocations(inode);
4546
4547 /*
4548 * The orphan list entry will now protect us from any crash which
4549 * occurs before the truncate completes, so it is now safe to propagate
4550 * the new, shorter inode size (held for now in i_size) into the
4551 * on-disk inode. We do this via i_disksize, which is the value which
4552 * ext4 *really* writes onto the disk inode.
4553 */
4554 ei->i_disksize = inode->i_size;
4555
4556 if (last_block == max_block) {
4557 /*
4558 * It is unnecessary to free any data blocks if last_block is
4559 * equal to the indirect block limit.
4560 */
4561 goto out_unlock;
4562 } else if (n == 1) { /* direct blocks */
4563 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4564 i_data + EXT4_NDIR_BLOCKS);
4565 goto do_indirects;
4566 }
4567
4568 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4569 /* Kill the top of shared branch (not detached) */
4570 if (nr) {
4571 if (partial == chain) {
4572 /* Shared branch grows from the inode */
4573 ext4_free_branches(handle, inode, NULL,
4574 &nr, &nr+1, (chain+n-1) - partial);
4575 *partial->p = 0;
4576 /*
4577 * We mark the inode dirty prior to restart,
4578 * and prior to stop. No need for it here.
4579 */
4580 } else {
4581 /* Shared branch grows from an indirect block */
4582 BUFFER_TRACE(partial->bh, "get_write_access");
4583 ext4_free_branches(handle, inode, partial->bh,
4584 partial->p,
4585 partial->p+1, (chain+n-1) - partial);
4586 }
4587 }
4588 /* Clear the ends of indirect blocks on the shared branch */
4589 while (partial > chain) {
4590 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4591 (__le32*)partial->bh->b_data+addr_per_block,
4592 (chain+n-1) - partial);
4593 BUFFER_TRACE(partial->bh, "call brelse");
4594 brelse(partial->bh);
4595 partial--;
4596 }
4597 do_indirects:
4598 /* Kill the remaining (whole) subtrees */
4599 switch (offsets[0]) {
4600 default:
4601 nr = i_data[EXT4_IND_BLOCK];
4602 if (nr) {
4603 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4604 i_data[EXT4_IND_BLOCK] = 0;
4605 }
4606 case EXT4_IND_BLOCK:
4607 nr = i_data[EXT4_DIND_BLOCK];
4608 if (nr) {
4609 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4610 i_data[EXT4_DIND_BLOCK] = 0;
4611 }
4612 case EXT4_DIND_BLOCK:
4613 nr = i_data[EXT4_TIND_BLOCK];
4614 if (nr) {
4615 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4616 i_data[EXT4_TIND_BLOCK] = 0;
4617 }
4618 case EXT4_TIND_BLOCK:
4619 ;
4620 }
4621
4622 out_unlock:
4623 up_write(&ei->i_data_sem);
4624 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4625 ext4_mark_inode_dirty(handle, inode);
4626
4627 /*
4628 * In a multi-transaction truncate, we only make the final transaction
4629 * synchronous
4630 */
4631 if (IS_SYNC(inode))
4632 ext4_handle_sync(handle);
4633 out_stop:
4634 /*
4635 * If this was a simple ftruncate(), and the file will remain alive
4636 * then we need to clear up the orphan record which we created above.
4637 * However, if this was a real unlink then we were called by
4638 * ext4_delete_inode(), and we allow that function to clean up the
4639 * orphan info for us.
4640 */
4641 if (inode->i_nlink)
4642 ext4_orphan_del(handle, inode);
4643
4644 ext4_journal_stop(handle);
4645 trace_ext4_truncate_exit(inode);
4646 }
4647
4648 /*
4649 * ext4_get_inode_loc returns with an extra refcount against the inode's
4650 * underlying buffer_head on success. If 'in_mem' is true, we have all
4651 * data in memory that is needed to recreate the on-disk version of this
4652 * inode.
4653 */
4654 static int __ext4_get_inode_loc(struct inode *inode,
4655 struct ext4_iloc *iloc, int in_mem)
4656 {
4657 struct ext4_group_desc *gdp;
4658 struct buffer_head *bh;
4659 struct super_block *sb = inode->i_sb;
4660 ext4_fsblk_t block;
4661 int inodes_per_block, inode_offset;
4662
4663 iloc->bh = NULL;
4664 if (!ext4_valid_inum(sb, inode->i_ino))
4665 return -EIO;
4666
4667 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4668 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4669 if (!gdp)
4670 return -EIO;
4671
4672 /*
4673 * Figure out the offset within the block group inode table
4674 */
4675 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4676 inode_offset = ((inode->i_ino - 1) %
4677 EXT4_INODES_PER_GROUP(sb));
4678 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4679 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4680
4681 bh = sb_getblk(sb, block);
4682 if (!bh) {
4683 EXT4_ERROR_INODE_BLOCK(inode, block,
4684 "unable to read itable block");
4685 return -EIO;
4686 }
4687 if (!buffer_uptodate(bh)) {
4688 lock_buffer(bh);
4689
4690 /*
4691 * If the buffer has the write error flag, we have failed
4692 * to write out another inode in the same block. In this
4693 * case, we don't have to read the block because we may
4694 * read the old inode data successfully.
4695 */
4696 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4697 set_buffer_uptodate(bh);
4698
4699 if (buffer_uptodate(bh)) {
4700 /* someone brought it uptodate while we waited */
4701 unlock_buffer(bh);
4702 goto has_buffer;
4703 }
4704
4705 /*
4706 * If we have all information of the inode in memory and this
4707 * is the only valid inode in the block, we need not read the
4708 * block.
4709 */
4710 if (in_mem) {
4711 struct buffer_head *bitmap_bh;
4712 int i, start;
4713
4714 start = inode_offset & ~(inodes_per_block - 1);
4715
4716 /* Is the inode bitmap in cache? */
4717 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4718 if (!bitmap_bh)
4719 goto make_io;
4720
4721 /*
4722 * If the inode bitmap isn't in cache then the
4723 * optimisation may end up performing two reads instead
4724 * of one, so skip it.
4725 */
4726 if (!buffer_uptodate(bitmap_bh)) {
4727 brelse(bitmap_bh);
4728 goto make_io;
4729 }
4730 for (i = start; i < start + inodes_per_block; i++) {
4731 if (i == inode_offset)
4732 continue;
4733 if (ext4_test_bit(i, bitmap_bh->b_data))
4734 break;
4735 }
4736 brelse(bitmap_bh);
4737 if (i == start + inodes_per_block) {
4738 /* all other inodes are free, so skip I/O */
4739 memset(bh->b_data, 0, bh->b_size);
4740 set_buffer_uptodate(bh);
4741 unlock_buffer(bh);
4742 goto has_buffer;
4743 }
4744 }
4745
4746 make_io:
4747 /*
4748 * If we need to do any I/O, try to pre-readahead extra
4749 * blocks from the inode table.
4750 */
4751 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4752 ext4_fsblk_t b, end, table;
4753 unsigned num;
4754
4755 table = ext4_inode_table(sb, gdp);
4756 /* s_inode_readahead_blks is always a power of 2 */
4757 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4758 if (table > b)
4759 b = table;
4760 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4761 num = EXT4_INODES_PER_GROUP(sb);
4762 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4763 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4764 num -= ext4_itable_unused_count(sb, gdp);
4765 table += num / inodes_per_block;
4766 if (end > table)
4767 end = table;
4768 while (b <= end)
4769 sb_breadahead(sb, b++);
4770 }
4771
4772 /*
4773 * There are other valid inodes in the buffer, this inode
4774 * has in-inode xattrs, or we don't have this inode in memory.
4775 * Read the block from disk.
4776 */
4777 trace_ext4_load_inode(inode);
4778 get_bh(bh);
4779 bh->b_end_io = end_buffer_read_sync;
4780 submit_bh(READ_META, bh);
4781 wait_on_buffer(bh);
4782 if (!buffer_uptodate(bh)) {
4783 EXT4_ERROR_INODE_BLOCK(inode, block,
4784 "unable to read itable block");
4785 brelse(bh);
4786 return -EIO;
4787 }
4788 }
4789 has_buffer:
4790 iloc->bh = bh;
4791 return 0;
4792 }
4793
4794 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4795 {
4796 /* We have all inode data except xattrs in memory here. */
4797 return __ext4_get_inode_loc(inode, iloc,
4798 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4799 }
4800
4801 void ext4_set_inode_flags(struct inode *inode)
4802 {
4803 unsigned int flags = EXT4_I(inode)->i_flags;
4804
4805 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4806 if (flags & EXT4_SYNC_FL)
4807 inode->i_flags |= S_SYNC;
4808 if (flags & EXT4_APPEND_FL)
4809 inode->i_flags |= S_APPEND;
4810 if (flags & EXT4_IMMUTABLE_FL)
4811 inode->i_flags |= S_IMMUTABLE;
4812 if (flags & EXT4_NOATIME_FL)
4813 inode->i_flags |= S_NOATIME;
4814 if (flags & EXT4_DIRSYNC_FL)
4815 inode->i_flags |= S_DIRSYNC;
4816 }
4817
4818 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4819 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4820 {
4821 unsigned int vfs_fl;
4822 unsigned long old_fl, new_fl;
4823
4824 do {
4825 vfs_fl = ei->vfs_inode.i_flags;
4826 old_fl = ei->i_flags;
4827 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4828 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4829 EXT4_DIRSYNC_FL);
4830 if (vfs_fl & S_SYNC)
4831 new_fl |= EXT4_SYNC_FL;
4832 if (vfs_fl & S_APPEND)
4833 new_fl |= EXT4_APPEND_FL;
4834 if (vfs_fl & S_IMMUTABLE)
4835 new_fl |= EXT4_IMMUTABLE_FL;
4836 if (vfs_fl & S_NOATIME)
4837 new_fl |= EXT4_NOATIME_FL;
4838 if (vfs_fl & S_DIRSYNC)
4839 new_fl |= EXT4_DIRSYNC_FL;
4840 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4841 }
4842
4843 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4844 struct ext4_inode_info *ei)
4845 {
4846 blkcnt_t i_blocks ;
4847 struct inode *inode = &(ei->vfs_inode);
4848 struct super_block *sb = inode->i_sb;
4849
4850 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4851 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4852 /* we are using combined 48 bit field */
4853 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4854 le32_to_cpu(raw_inode->i_blocks_lo);
4855 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4856 /* i_blocks represent file system block size */
4857 return i_blocks << (inode->i_blkbits - 9);
4858 } else {
4859 return i_blocks;
4860 }
4861 } else {
4862 return le32_to_cpu(raw_inode->i_blocks_lo);
4863 }
4864 }
4865
4866 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4867 {
4868 struct ext4_iloc iloc;
4869 struct ext4_inode *raw_inode;
4870 struct ext4_inode_info *ei;
4871 struct inode *inode;
4872 journal_t *journal = EXT4_SB(sb)->s_journal;
4873 long ret;
4874 int block;
4875
4876 inode = iget_locked(sb, ino);
4877 if (!inode)
4878 return ERR_PTR(-ENOMEM);
4879 if (!(inode->i_state & I_NEW))
4880 return inode;
4881
4882 ei = EXT4_I(inode);
4883 iloc.bh = NULL;
4884
4885 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4886 if (ret < 0)
4887 goto bad_inode;
4888 raw_inode = ext4_raw_inode(&iloc);
4889 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4890 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4891 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4892 if (!(test_opt(inode->i_sb, NO_UID32))) {
4893 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4894 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4895 }
4896 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4897
4898 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4899 ei->i_dir_start_lookup = 0;
4900 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4901 /* We now have enough fields to check if the inode was active or not.
4902 * This is needed because nfsd might try to access dead inodes
4903 * the test is that same one that e2fsck uses
4904 * NeilBrown 1999oct15
4905 */
4906 if (inode->i_nlink == 0) {
4907 if (inode->i_mode == 0 ||
4908 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4909 /* this inode is deleted */
4910 ret = -ESTALE;
4911 goto bad_inode;
4912 }
4913 /* The only unlinked inodes we let through here have
4914 * valid i_mode and are being read by the orphan
4915 * recovery code: that's fine, we're about to complete
4916 * the process of deleting those. */
4917 }
4918 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4919 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4920 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4921 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4922 ei->i_file_acl |=
4923 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4924 inode->i_size = ext4_isize(raw_inode);
4925 ei->i_disksize = inode->i_size;
4926 #ifdef CONFIG_QUOTA
4927 ei->i_reserved_quota = 0;
4928 #endif
4929 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4930 ei->i_block_group = iloc.block_group;
4931 ei->i_last_alloc_group = ~0;
4932 /*
4933 * NOTE! The in-memory inode i_data array is in little-endian order
4934 * even on big-endian machines: we do NOT byteswap the block numbers!
4935 */
4936 for (block = 0; block < EXT4_N_BLOCKS; block++)
4937 ei->i_data[block] = raw_inode->i_block[block];
4938 INIT_LIST_HEAD(&ei->i_orphan);
4939
4940 /*
4941 * Set transaction id's of transactions that have to be committed
4942 * to finish f[data]sync. We set them to currently running transaction
4943 * as we cannot be sure that the inode or some of its metadata isn't
4944 * part of the transaction - the inode could have been reclaimed and
4945 * now it is reread from disk.
4946 */
4947 if (journal) {
4948 transaction_t *transaction;
4949 tid_t tid;
4950
4951 read_lock(&journal->j_state_lock);
4952 if (journal->j_running_transaction)
4953 transaction = journal->j_running_transaction;
4954 else
4955 transaction = journal->j_committing_transaction;
4956 if (transaction)
4957 tid = transaction->t_tid;
4958 else
4959 tid = journal->j_commit_sequence;
4960 read_unlock(&journal->j_state_lock);
4961 ei->i_sync_tid = tid;
4962 ei->i_datasync_tid = tid;
4963 }
4964
4965 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4966 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4967 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4968 EXT4_INODE_SIZE(inode->i_sb)) {
4969 ret = -EIO;
4970 goto bad_inode;
4971 }
4972 if (ei->i_extra_isize == 0) {
4973 /* The extra space is currently unused. Use it. */
4974 ei->i_extra_isize = sizeof(struct ext4_inode) -
4975 EXT4_GOOD_OLD_INODE_SIZE;
4976 } else {
4977 __le32 *magic = (void *)raw_inode +
4978 EXT4_GOOD_OLD_INODE_SIZE +
4979 ei->i_extra_isize;
4980 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4981 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4982 }
4983 } else
4984 ei->i_extra_isize = 0;
4985
4986 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4987 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4988 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4989 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4990
4991 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4992 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4993 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4994 inode->i_version |=
4995 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4996 }
4997
4998 ret = 0;
4999 if (ei->i_file_acl &&
5000 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5001 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5002 ei->i_file_acl);
5003 ret = -EIO;
5004 goto bad_inode;
5005 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5006 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5007 (S_ISLNK(inode->i_mode) &&
5008 !ext4_inode_is_fast_symlink(inode)))
5009 /* Validate extent which is part of inode */
5010 ret = ext4_ext_check_inode(inode);
5011 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5012 (S_ISLNK(inode->i_mode) &&
5013 !ext4_inode_is_fast_symlink(inode))) {
5014 /* Validate block references which are part of inode */
5015 ret = ext4_check_inode_blockref(inode);
5016 }
5017 if (ret)
5018 goto bad_inode;
5019
5020 if (S_ISREG(inode->i_mode)) {
5021 inode->i_op = &ext4_file_inode_operations;
5022 inode->i_fop = &ext4_file_operations;
5023 ext4_set_aops(inode);
5024 } else if (S_ISDIR(inode->i_mode)) {
5025 inode->i_op = &ext4_dir_inode_operations;
5026 inode->i_fop = &ext4_dir_operations;
5027 } else if (S_ISLNK(inode->i_mode)) {
5028 if (ext4_inode_is_fast_symlink(inode)) {
5029 inode->i_op = &ext4_fast_symlink_inode_operations;
5030 nd_terminate_link(ei->i_data, inode->i_size,
5031 sizeof(ei->i_data) - 1);
5032 } else {
5033 inode->i_op = &ext4_symlink_inode_operations;
5034 ext4_set_aops(inode);
5035 }
5036 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5037 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5038 inode->i_op = &ext4_special_inode_operations;
5039 if (raw_inode->i_block[0])
5040 init_special_inode(inode, inode->i_mode,
5041 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5042 else
5043 init_special_inode(inode, inode->i_mode,
5044 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5045 } else {
5046 ret = -EIO;
5047 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5048 goto bad_inode;
5049 }
5050 brelse(iloc.bh);
5051 ext4_set_inode_flags(inode);
5052 unlock_new_inode(inode);
5053 return inode;
5054
5055 bad_inode:
5056 brelse(iloc.bh);
5057 iget_failed(inode);
5058 return ERR_PTR(ret);
5059 }
5060
5061 static int ext4_inode_blocks_set(handle_t *handle,
5062 struct ext4_inode *raw_inode,
5063 struct ext4_inode_info *ei)
5064 {
5065 struct inode *inode = &(ei->vfs_inode);
5066 u64 i_blocks = inode->i_blocks;
5067 struct super_block *sb = inode->i_sb;
5068
5069 if (i_blocks <= ~0U) {
5070 /*
5071 * i_blocks can be represnted in a 32 bit variable
5072 * as multiple of 512 bytes
5073 */
5074 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5075 raw_inode->i_blocks_high = 0;
5076 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5077 return 0;
5078 }
5079 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5080 return -EFBIG;
5081
5082 if (i_blocks <= 0xffffffffffffULL) {
5083 /*
5084 * i_blocks can be represented in a 48 bit variable
5085 * as multiple of 512 bytes
5086 */
5087 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5088 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5089 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5090 } else {
5091 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5092 /* i_block is stored in file system block size */
5093 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5094 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5095 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5096 }
5097 return 0;
5098 }
5099
5100 /*
5101 * Post the struct inode info into an on-disk inode location in the
5102 * buffer-cache. This gobbles the caller's reference to the
5103 * buffer_head in the inode location struct.
5104 *
5105 * The caller must have write access to iloc->bh.
5106 */
5107 static int ext4_do_update_inode(handle_t *handle,
5108 struct inode *inode,
5109 struct ext4_iloc *iloc)
5110 {
5111 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5112 struct ext4_inode_info *ei = EXT4_I(inode);
5113 struct buffer_head *bh = iloc->bh;
5114 int err = 0, rc, block;
5115
5116 /* For fields not not tracking in the in-memory inode,
5117 * initialise them to zero for new inodes. */
5118 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5119 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5120
5121 ext4_get_inode_flags(ei);
5122 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5123 if (!(test_opt(inode->i_sb, NO_UID32))) {
5124 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5125 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5126 /*
5127 * Fix up interoperability with old kernels. Otherwise, old inodes get
5128 * re-used with the upper 16 bits of the uid/gid intact
5129 */
5130 if (!ei->i_dtime) {
5131 raw_inode->i_uid_high =
5132 cpu_to_le16(high_16_bits(inode->i_uid));
5133 raw_inode->i_gid_high =
5134 cpu_to_le16(high_16_bits(inode->i_gid));
5135 } else {
5136 raw_inode->i_uid_high = 0;
5137 raw_inode->i_gid_high = 0;
5138 }
5139 } else {
5140 raw_inode->i_uid_low =
5141 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5142 raw_inode->i_gid_low =
5143 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5144 raw_inode->i_uid_high = 0;
5145 raw_inode->i_gid_high = 0;
5146 }
5147 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5148
5149 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5150 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5151 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5152 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5153
5154 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5155 goto out_brelse;
5156 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5157 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5158 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5159 cpu_to_le32(EXT4_OS_HURD))
5160 raw_inode->i_file_acl_high =
5161 cpu_to_le16(ei->i_file_acl >> 32);
5162 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5163 ext4_isize_set(raw_inode, ei->i_disksize);
5164 if (ei->i_disksize > 0x7fffffffULL) {
5165 struct super_block *sb = inode->i_sb;
5166 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5167 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5168 EXT4_SB(sb)->s_es->s_rev_level ==
5169 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5170 /* If this is the first large file
5171 * created, add a flag to the superblock.
5172 */
5173 err = ext4_journal_get_write_access(handle,
5174 EXT4_SB(sb)->s_sbh);
5175 if (err)
5176 goto out_brelse;
5177 ext4_update_dynamic_rev(sb);
5178 EXT4_SET_RO_COMPAT_FEATURE(sb,
5179 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5180 sb->s_dirt = 1;
5181 ext4_handle_sync(handle);
5182 err = ext4_handle_dirty_metadata(handle, NULL,
5183 EXT4_SB(sb)->s_sbh);
5184 }
5185 }
5186 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5187 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5188 if (old_valid_dev(inode->i_rdev)) {
5189 raw_inode->i_block[0] =
5190 cpu_to_le32(old_encode_dev(inode->i_rdev));
5191 raw_inode->i_block[1] = 0;
5192 } else {
5193 raw_inode->i_block[0] = 0;
5194 raw_inode->i_block[1] =
5195 cpu_to_le32(new_encode_dev(inode->i_rdev));
5196 raw_inode->i_block[2] = 0;
5197 }
5198 } else
5199 for (block = 0; block < EXT4_N_BLOCKS; block++)
5200 raw_inode->i_block[block] = ei->i_data[block];
5201
5202 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5203 if (ei->i_extra_isize) {
5204 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5205 raw_inode->i_version_hi =
5206 cpu_to_le32(inode->i_version >> 32);
5207 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5208 }
5209
5210 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5211 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5212 if (!err)
5213 err = rc;
5214 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5215
5216 ext4_update_inode_fsync_trans(handle, inode, 0);
5217 out_brelse:
5218 brelse(bh);
5219 ext4_std_error(inode->i_sb, err);
5220 return err;
5221 }
5222
5223 /*
5224 * ext4_write_inode()
5225 *
5226 * We are called from a few places:
5227 *
5228 * - Within generic_file_write() for O_SYNC files.
5229 * Here, there will be no transaction running. We wait for any running
5230 * trasnaction to commit.
5231 *
5232 * - Within sys_sync(), kupdate and such.
5233 * We wait on commit, if tol to.
5234 *
5235 * - Within prune_icache() (PF_MEMALLOC == true)
5236 * Here we simply return. We can't afford to block kswapd on the
5237 * journal commit.
5238 *
5239 * In all cases it is actually safe for us to return without doing anything,
5240 * because the inode has been copied into a raw inode buffer in
5241 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5242 * knfsd.
5243 *
5244 * Note that we are absolutely dependent upon all inode dirtiers doing the
5245 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5246 * which we are interested.
5247 *
5248 * It would be a bug for them to not do this. The code:
5249 *
5250 * mark_inode_dirty(inode)
5251 * stuff();
5252 * inode->i_size = expr;
5253 *
5254 * is in error because a kswapd-driven write_inode() could occur while
5255 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5256 * will no longer be on the superblock's dirty inode list.
5257 */
5258 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5259 {
5260 int err;
5261
5262 if (current->flags & PF_MEMALLOC)
5263 return 0;
5264
5265 if (EXT4_SB(inode->i_sb)->s_journal) {
5266 if (ext4_journal_current_handle()) {
5267 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5268 dump_stack();
5269 return -EIO;
5270 }
5271
5272 if (wbc->sync_mode != WB_SYNC_ALL)
5273 return 0;
5274
5275 err = ext4_force_commit(inode->i_sb);
5276 } else {
5277 struct ext4_iloc iloc;
5278
5279 err = __ext4_get_inode_loc(inode, &iloc, 0);
5280 if (err)
5281 return err;
5282 if (wbc->sync_mode == WB_SYNC_ALL)
5283 sync_dirty_buffer(iloc.bh);
5284 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5285 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5286 "IO error syncing inode");
5287 err = -EIO;
5288 }
5289 brelse(iloc.bh);
5290 }
5291 return err;
5292 }
5293
5294 /*
5295 * ext4_setattr()
5296 *
5297 * Called from notify_change.
5298 *
5299 * We want to trap VFS attempts to truncate the file as soon as
5300 * possible. In particular, we want to make sure that when the VFS
5301 * shrinks i_size, we put the inode on the orphan list and modify
5302 * i_disksize immediately, so that during the subsequent flushing of
5303 * dirty pages and freeing of disk blocks, we can guarantee that any
5304 * commit will leave the blocks being flushed in an unused state on
5305 * disk. (On recovery, the inode will get truncated and the blocks will
5306 * be freed, so we have a strong guarantee that no future commit will
5307 * leave these blocks visible to the user.)
5308 *
5309 * Another thing we have to assure is that if we are in ordered mode
5310 * and inode is still attached to the committing transaction, we must
5311 * we start writeout of all the dirty pages which are being truncated.
5312 * This way we are sure that all the data written in the previous
5313 * transaction are already on disk (truncate waits for pages under
5314 * writeback).
5315 *
5316 * Called with inode->i_mutex down.
5317 */
5318 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5319 {
5320 struct inode *inode = dentry->d_inode;
5321 int error, rc = 0;
5322 int orphan = 0;
5323 const unsigned int ia_valid = attr->ia_valid;
5324
5325 error = inode_change_ok(inode, attr);
5326 if (error)
5327 return error;
5328
5329 if (is_quota_modification(inode, attr))
5330 dquot_initialize(inode);
5331 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5332 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5333 handle_t *handle;
5334
5335 /* (user+group)*(old+new) structure, inode write (sb,
5336 * inode block, ? - but truncate inode update has it) */
5337 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5338 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5339 if (IS_ERR(handle)) {
5340 error = PTR_ERR(handle);
5341 goto err_out;
5342 }
5343 error = dquot_transfer(inode, attr);
5344 if (error) {
5345 ext4_journal_stop(handle);
5346 return error;
5347 }
5348 /* Update corresponding info in inode so that everything is in
5349 * one transaction */
5350 if (attr->ia_valid & ATTR_UID)
5351 inode->i_uid = attr->ia_uid;
5352 if (attr->ia_valid & ATTR_GID)
5353 inode->i_gid = attr->ia_gid;
5354 error = ext4_mark_inode_dirty(handle, inode);
5355 ext4_journal_stop(handle);
5356 }
5357
5358 if (attr->ia_valid & ATTR_SIZE) {
5359 inode_dio_wait(inode);
5360
5361 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5362 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5363
5364 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5365 return -EFBIG;
5366 }
5367 }
5368
5369 if (S_ISREG(inode->i_mode) &&
5370 attr->ia_valid & ATTR_SIZE &&
5371 (attr->ia_size < inode->i_size)) {
5372 handle_t *handle;
5373
5374 handle = ext4_journal_start(inode, 3);
5375 if (IS_ERR(handle)) {
5376 error = PTR_ERR(handle);
5377 goto err_out;
5378 }
5379 if (ext4_handle_valid(handle)) {
5380 error = ext4_orphan_add(handle, inode);
5381 orphan = 1;
5382 }
5383 EXT4_I(inode)->i_disksize = attr->ia_size;
5384 rc = ext4_mark_inode_dirty(handle, inode);
5385 if (!error)
5386 error = rc;
5387 ext4_journal_stop(handle);
5388
5389 if (ext4_should_order_data(inode)) {
5390 error = ext4_begin_ordered_truncate(inode,
5391 attr->ia_size);
5392 if (error) {
5393 /* Do as much error cleanup as possible */
5394 handle = ext4_journal_start(inode, 3);
5395 if (IS_ERR(handle)) {
5396 ext4_orphan_del(NULL, inode);
5397 goto err_out;
5398 }
5399 ext4_orphan_del(handle, inode);
5400 orphan = 0;
5401 ext4_journal_stop(handle);
5402 goto err_out;
5403 }
5404 }
5405 }
5406
5407 if (attr->ia_valid & ATTR_SIZE) {
5408 if (attr->ia_size != i_size_read(inode)) {
5409 truncate_setsize(inode, attr->ia_size);
5410 ext4_truncate(inode);
5411 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
5412 ext4_truncate(inode);
5413 }
5414
5415 if (!rc) {
5416 setattr_copy(inode, attr);
5417 mark_inode_dirty(inode);
5418 }
5419
5420 /*
5421 * If the call to ext4_truncate failed to get a transaction handle at
5422 * all, we need to clean up the in-core orphan list manually.
5423 */
5424 if (orphan && inode->i_nlink)
5425 ext4_orphan_del(NULL, inode);
5426
5427 if (!rc && (ia_valid & ATTR_MODE))
5428 rc = ext4_acl_chmod(inode);
5429
5430 err_out:
5431 ext4_std_error(inode->i_sb, error);
5432 if (!error)
5433 error = rc;
5434 return error;
5435 }
5436
5437 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5438 struct kstat *stat)
5439 {
5440 struct inode *inode;
5441 unsigned long delalloc_blocks;
5442
5443 inode = dentry->d_inode;
5444 generic_fillattr(inode, stat);
5445
5446 /*
5447 * We can't update i_blocks if the block allocation is delayed
5448 * otherwise in the case of system crash before the real block
5449 * allocation is done, we will have i_blocks inconsistent with
5450 * on-disk file blocks.
5451 * We always keep i_blocks updated together with real
5452 * allocation. But to not confuse with user, stat
5453 * will return the blocks that include the delayed allocation
5454 * blocks for this file.
5455 */
5456 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5457
5458 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5459 return 0;
5460 }
5461
5462 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5463 int chunk)
5464 {
5465 int indirects;
5466
5467 /* if nrblocks are contiguous */
5468 if (chunk) {
5469 /*
5470 * With N contiguous data blocks, we need at most
5471 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
5472 * 2 dindirect blocks, and 1 tindirect block
5473 */
5474 return DIV_ROUND_UP(nrblocks,
5475 EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
5476 }
5477 /*
5478 * if nrblocks are not contiguous, worse case, each block touch
5479 * a indirect block, and each indirect block touch a double indirect
5480 * block, plus a triple indirect block
5481 */
5482 indirects = nrblocks * 2 + 1;
5483 return indirects;
5484 }
5485
5486 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5487 {
5488 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5489 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5490 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5491 }
5492
5493 /*
5494 * Account for index blocks, block groups bitmaps and block group
5495 * descriptor blocks if modify datablocks and index blocks
5496 * worse case, the indexs blocks spread over different block groups
5497 *
5498 * If datablocks are discontiguous, they are possible to spread over
5499 * different block groups too. If they are contiuguous, with flexbg,
5500 * they could still across block group boundary.
5501 *
5502 * Also account for superblock, inode, quota and xattr blocks
5503 */
5504 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5505 {
5506 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5507 int gdpblocks;
5508 int idxblocks;
5509 int ret = 0;
5510
5511 /*
5512 * How many index blocks need to touch to modify nrblocks?
5513 * The "Chunk" flag indicating whether the nrblocks is
5514 * physically contiguous on disk
5515 *
5516 * For Direct IO and fallocate, they calls get_block to allocate
5517 * one single extent at a time, so they could set the "Chunk" flag
5518 */
5519 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5520
5521 ret = idxblocks;
5522
5523 /*
5524 * Now let's see how many group bitmaps and group descriptors need
5525 * to account
5526 */
5527 groups = idxblocks;
5528 if (chunk)
5529 groups += 1;
5530 else
5531 groups += nrblocks;
5532
5533 gdpblocks = groups;
5534 if (groups > ngroups)
5535 groups = ngroups;
5536 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5537 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5538
5539 /* bitmaps and block group descriptor blocks */
5540 ret += groups + gdpblocks;
5541
5542 /* Blocks for super block, inode, quota and xattr blocks */
5543 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5544
5545 return ret;
5546 }
5547
5548 /*
5549 * Calculate the total number of credits to reserve to fit
5550 * the modification of a single pages into a single transaction,
5551 * which may include multiple chunks of block allocations.
5552 *
5553 * This could be called via ext4_write_begin()
5554 *
5555 * We need to consider the worse case, when
5556 * one new block per extent.
5557 */
5558 int ext4_writepage_trans_blocks(struct inode *inode)
5559 {
5560 int bpp = ext4_journal_blocks_per_page(inode);
5561 int ret;
5562
5563 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5564
5565 /* Account for data blocks for journalled mode */
5566 if (ext4_should_journal_data(inode))
5567 ret += bpp;
5568 return ret;
5569 }
5570
5571 /*
5572 * Calculate the journal credits for a chunk of data modification.
5573 *
5574 * This is called from DIO, fallocate or whoever calling
5575 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5576 *
5577 * journal buffers for data blocks are not included here, as DIO
5578 * and fallocate do no need to journal data buffers.
5579 */
5580 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5581 {
5582 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5583 }
5584
5585 /*
5586 * The caller must have previously called ext4_reserve_inode_write().
5587 * Give this, we know that the caller already has write access to iloc->bh.
5588 */
5589 int ext4_mark_iloc_dirty(handle_t *handle,
5590 struct inode *inode, struct ext4_iloc *iloc)
5591 {
5592 int err = 0;
5593
5594 if (test_opt(inode->i_sb, I_VERSION))
5595 inode_inc_iversion(inode);
5596
5597 /* the do_update_inode consumes one bh->b_count */
5598 get_bh(iloc->bh);
5599
5600 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5601 err = ext4_do_update_inode(handle, inode, iloc);
5602 put_bh(iloc->bh);
5603 return err;
5604 }
5605
5606 /*
5607 * On success, We end up with an outstanding reference count against
5608 * iloc->bh. This _must_ be cleaned up later.
5609 */
5610
5611 int
5612 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5613 struct ext4_iloc *iloc)
5614 {
5615 int err;
5616
5617 err = ext4_get_inode_loc(inode, iloc);
5618 if (!err) {
5619 BUFFER_TRACE(iloc->bh, "get_write_access");
5620 err = ext4_journal_get_write_access(handle, iloc->bh);
5621 if (err) {
5622 brelse(iloc->bh);
5623 iloc->bh = NULL;
5624 }
5625 }
5626 ext4_std_error(inode->i_sb, err);
5627 return err;
5628 }
5629
5630 /*
5631 * Expand an inode by new_extra_isize bytes.
5632 * Returns 0 on success or negative error number on failure.
5633 */
5634 static int ext4_expand_extra_isize(struct inode *inode,
5635 unsigned int new_extra_isize,
5636 struct ext4_iloc iloc,
5637 handle_t *handle)
5638 {
5639 struct ext4_inode *raw_inode;
5640 struct ext4_xattr_ibody_header *header;
5641
5642 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5643 return 0;
5644
5645 raw_inode = ext4_raw_inode(&iloc);
5646
5647 header = IHDR(inode, raw_inode);
5648
5649 /* No extended attributes present */
5650 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5651 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5652 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5653 new_extra_isize);
5654 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5655 return 0;
5656 }
5657
5658 /* try to expand with EAs present */
5659 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5660 raw_inode, handle);
5661 }
5662
5663 /*
5664 * What we do here is to mark the in-core inode as clean with respect to inode
5665 * dirtiness (it may still be data-dirty).
5666 * This means that the in-core inode may be reaped by prune_icache
5667 * without having to perform any I/O. This is a very good thing,
5668 * because *any* task may call prune_icache - even ones which
5669 * have a transaction open against a different journal.
5670 *
5671 * Is this cheating? Not really. Sure, we haven't written the
5672 * inode out, but prune_icache isn't a user-visible syncing function.
5673 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5674 * we start and wait on commits.
5675 *
5676 * Is this efficient/effective? Well, we're being nice to the system
5677 * by cleaning up our inodes proactively so they can be reaped
5678 * without I/O. But we are potentially leaving up to five seconds'
5679 * worth of inodes floating about which prune_icache wants us to
5680 * write out. One way to fix that would be to get prune_icache()
5681 * to do a write_super() to free up some memory. It has the desired
5682 * effect.
5683 */
5684 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5685 {
5686 struct ext4_iloc iloc;
5687 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5688 static unsigned int mnt_count;
5689 int err, ret;
5690
5691 might_sleep();
5692 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5693 err = ext4_reserve_inode_write(handle, inode, &iloc);
5694 if (ext4_handle_valid(handle) &&
5695 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5696 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5697 /*
5698 * We need extra buffer credits since we may write into EA block
5699 * with this same handle. If journal_extend fails, then it will
5700 * only result in a minor loss of functionality for that inode.
5701 * If this is felt to be critical, then e2fsck should be run to
5702 * force a large enough s_min_extra_isize.
5703 */
5704 if ((jbd2_journal_extend(handle,
5705 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5706 ret = ext4_expand_extra_isize(inode,
5707 sbi->s_want_extra_isize,
5708 iloc, handle);
5709 if (ret) {
5710 ext4_set_inode_state(inode,
5711 EXT4_STATE_NO_EXPAND);
5712 if (mnt_count !=
5713 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5714 ext4_warning(inode->i_sb,
5715 "Unable to expand inode %lu. Delete"
5716 " some EAs or run e2fsck.",
5717 inode->i_ino);
5718 mnt_count =
5719 le16_to_cpu(sbi->s_es->s_mnt_count);
5720 }
5721 }
5722 }
5723 }
5724 if (!err)
5725 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5726 return err;
5727 }
5728
5729 /*
5730 * ext4_dirty_inode() is called from __mark_inode_dirty()
5731 *
5732 * We're really interested in the case where a file is being extended.
5733 * i_size has been changed by generic_commit_write() and we thus need
5734 * to include the updated inode in the current transaction.
5735 *
5736 * Also, dquot_alloc_block() will always dirty the inode when blocks
5737 * are allocated to the file.
5738 *
5739 * If the inode is marked synchronous, we don't honour that here - doing
5740 * so would cause a commit on atime updates, which we don't bother doing.
5741 * We handle synchronous inodes at the highest possible level.
5742 */
5743 void ext4_dirty_inode(struct inode *inode, int flags)
5744 {
5745 handle_t *handle;
5746
5747 handle = ext4_journal_start(inode, 2);
5748 if (IS_ERR(handle))
5749 goto out;
5750
5751 ext4_mark_inode_dirty(handle, inode);
5752
5753 ext4_journal_stop(handle);
5754 out:
5755 return;
5756 }
5757
5758 #if 0
5759 /*
5760 * Bind an inode's backing buffer_head into this transaction, to prevent
5761 * it from being flushed to disk early. Unlike
5762 * ext4_reserve_inode_write, this leaves behind no bh reference and
5763 * returns no iloc structure, so the caller needs to repeat the iloc
5764 * lookup to mark the inode dirty later.
5765 */
5766 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5767 {
5768 struct ext4_iloc iloc;
5769
5770 int err = 0;
5771 if (handle) {
5772 err = ext4_get_inode_loc(inode, &iloc);
5773 if (!err) {
5774 BUFFER_TRACE(iloc.bh, "get_write_access");
5775 err = jbd2_journal_get_write_access(handle, iloc.bh);
5776 if (!err)
5777 err = ext4_handle_dirty_metadata(handle,
5778 NULL,
5779 iloc.bh);
5780 brelse(iloc.bh);
5781 }
5782 }
5783 ext4_std_error(inode->i_sb, err);
5784 return err;
5785 }
5786 #endif
5787
5788 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5789 {
5790 journal_t *journal;
5791 handle_t *handle;
5792 int err;
5793
5794 /*
5795 * We have to be very careful here: changing a data block's
5796 * journaling status dynamically is dangerous. If we write a
5797 * data block to the journal, change the status and then delete
5798 * that block, we risk forgetting to revoke the old log record
5799 * from the journal and so a subsequent replay can corrupt data.
5800 * So, first we make sure that the journal is empty and that
5801 * nobody is changing anything.
5802 */
5803
5804 journal = EXT4_JOURNAL(inode);
5805 if (!journal)
5806 return 0;
5807 if (is_journal_aborted(journal))
5808 return -EROFS;
5809
5810 jbd2_journal_lock_updates(journal);
5811 jbd2_journal_flush(journal);
5812
5813 /*
5814 * OK, there are no updates running now, and all cached data is
5815 * synced to disk. We are now in a completely consistent state
5816 * which doesn't have anything in the journal, and we know that
5817 * no filesystem updates are running, so it is safe to modify
5818 * the inode's in-core data-journaling state flag now.
5819 */
5820
5821 if (val)
5822 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5823 else
5824 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5825 ext4_set_aops(inode);
5826
5827 jbd2_journal_unlock_updates(journal);
5828
5829 /* Finally we can mark the inode as dirty. */
5830
5831 handle = ext4_journal_start(inode, 1);
5832 if (IS_ERR(handle))
5833 return PTR_ERR(handle);
5834
5835 err = ext4_mark_inode_dirty(handle, inode);
5836 ext4_handle_sync(handle);
5837 ext4_journal_stop(handle);
5838 ext4_std_error(inode->i_sb, err);
5839
5840 return err;
5841 }
5842
5843 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5844 {
5845 return !buffer_mapped(bh);
5846 }
5847
5848 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5849 {
5850 struct page *page = vmf->page;
5851 loff_t size;
5852 unsigned long len;
5853 int ret;
5854 struct file *file = vma->vm_file;
5855 struct inode *inode = file->f_path.dentry->d_inode;
5856 struct address_space *mapping = inode->i_mapping;
5857 handle_t *handle;
5858 get_block_t *get_block;
5859 int retries = 0;
5860
5861 /*
5862 * This check is racy but catches the common case. We rely on
5863 * __block_page_mkwrite() to do a reliable check.
5864 */
5865 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
5866 /* Delalloc case is easy... */
5867 if (test_opt(inode->i_sb, DELALLOC) &&
5868 !ext4_should_journal_data(inode) &&
5869 !ext4_nonda_switch(inode->i_sb)) {
5870 do {
5871 ret = __block_page_mkwrite(vma, vmf,
5872 ext4_da_get_block_prep);
5873 } while (ret == -ENOSPC &&
5874 ext4_should_retry_alloc(inode->i_sb, &retries));
5875 goto out_ret;
5876 }
5877
5878 lock_page(page);
5879 size = i_size_read(inode);
5880 /* Page got truncated from under us? */
5881 if (page->mapping != mapping || page_offset(page) > size) {
5882 unlock_page(page);
5883 ret = VM_FAULT_NOPAGE;
5884 goto out;
5885 }
5886
5887 if (page->index == size >> PAGE_CACHE_SHIFT)
5888 len = size & ~PAGE_CACHE_MASK;
5889 else
5890 len = PAGE_CACHE_SIZE;
5891 /*
5892 * Return if we have all the buffers mapped. This avoids the need to do
5893 * journal_start/journal_stop which can block and take a long time
5894 */
5895 if (page_has_buffers(page)) {
5896 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5897 ext4_bh_unmapped)) {
5898 /* Wait so that we don't change page under IO */
5899 wait_on_page_writeback(page);
5900 ret = VM_FAULT_LOCKED;
5901 goto out;
5902 }
5903 }
5904 unlock_page(page);
5905 /* OK, we need to fill the hole... */
5906 if (ext4_should_dioread_nolock(inode))
5907 get_block = ext4_get_block_write;
5908 else
5909 get_block = ext4_get_block;
5910 retry_alloc:
5911 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
5912 if (IS_ERR(handle)) {
5913 ret = VM_FAULT_SIGBUS;
5914 goto out;
5915 }
5916 ret = __block_page_mkwrite(vma, vmf, get_block);
5917 if (!ret && ext4_should_journal_data(inode)) {
5918 if (walk_page_buffers(handle, page_buffers(page), 0,
5919 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5920 unlock_page(page);
5921 ret = VM_FAULT_SIGBUS;
5922 goto out;
5923 }
5924 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5925 }
5926 ext4_journal_stop(handle);
5927 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5928 goto retry_alloc;
5929 out_ret:
5930 ret = block_page_mkwrite_return(ret);
5931 out:
5932 return ret;
5933 }