Loading...
1/*
2 * linux/fs/ext3/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 ext3_get_block() by Al Viro, 2000
23 */
24
25#include <linux/highuid.h>
26#include <linux/quotaops.h>
27#include <linux/writeback.h>
28#include <linux/mpage.h>
29#include <linux/namei.h>
30#include "ext3.h"
31#include "xattr.h"
32#include "acl.h"
33
34static int ext3_writepage_trans_blocks(struct inode *inode);
35static int ext3_block_truncate_page(struct inode *inode, loff_t from);
36
37/*
38 * Test whether an inode is a fast symlink.
39 */
40static int ext3_inode_is_fast_symlink(struct inode *inode)
41{
42 int ea_blocks = EXT3_I(inode)->i_file_acl ?
43 (inode->i_sb->s_blocksize >> 9) : 0;
44
45 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
46}
47
48/*
49 * The ext3 forget function must perform a revoke if we are freeing data
50 * which has been journaled. Metadata (eg. indirect blocks) must be
51 * revoked in all cases.
52 *
53 * "bh" may be NULL: a metadata block may have been freed from memory
54 * but there may still be a record of it in the journal, and that record
55 * still needs to be revoked.
56 */
57int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
58 struct buffer_head *bh, ext3_fsblk_t blocknr)
59{
60 int err;
61
62 might_sleep();
63
64 trace_ext3_forget(inode, is_metadata, blocknr);
65 BUFFER_TRACE(bh, "enter");
66
67 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
68 "data mode %lx\n",
69 bh, is_metadata, inode->i_mode,
70 test_opt(inode->i_sb, DATA_FLAGS));
71
72 /* Never use the revoke function if we are doing full data
73 * journaling: there is no need to, and a V1 superblock won't
74 * support it. Otherwise, only skip the revoke on un-journaled
75 * data blocks. */
76
77 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
78 (!is_metadata && !ext3_should_journal_data(inode))) {
79 if (bh) {
80 BUFFER_TRACE(bh, "call journal_forget");
81 return ext3_journal_forget(handle, bh);
82 }
83 return 0;
84 }
85
86 /*
87 * data!=journal && (is_metadata || should_journal_data(inode))
88 */
89 BUFFER_TRACE(bh, "call ext3_journal_revoke");
90 err = ext3_journal_revoke(handle, blocknr, bh);
91 if (err)
92 ext3_abort(inode->i_sb, __func__,
93 "error %d when attempting revoke", err);
94 BUFFER_TRACE(bh, "exit");
95 return err;
96}
97
98/*
99 * Work out how many blocks we need to proceed with the next chunk of a
100 * truncate transaction.
101 */
102static unsigned long blocks_for_truncate(struct inode *inode)
103{
104 unsigned long needed;
105
106 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
107
108 /* Give ourselves just enough room to cope with inodes in which
109 * i_blocks is corrupt: we've seen disk corruptions in the past
110 * which resulted in random data in an inode which looked enough
111 * like a regular file for ext3 to try to delete it. Things
112 * will go a bit crazy if that happens, but at least we should
113 * try not to panic the whole kernel. */
114 if (needed < 2)
115 needed = 2;
116
117 /* But we need to bound the transaction so we don't overflow the
118 * journal. */
119 if (needed > EXT3_MAX_TRANS_DATA)
120 needed = EXT3_MAX_TRANS_DATA;
121
122 return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
123}
124
125/*
126 * Truncate transactions can be complex and absolutely huge. So we need to
127 * be able to restart the transaction at a conventient checkpoint to make
128 * sure we don't overflow the journal.
129 *
130 * start_transaction gets us a new handle for a truncate transaction,
131 * and extend_transaction tries to extend the existing one a bit. If
132 * extend fails, we need to propagate the failure up and restart the
133 * transaction in the top-level truncate loop. --sct
134 */
135static handle_t *start_transaction(struct inode *inode)
136{
137 handle_t *result;
138
139 result = ext3_journal_start(inode, blocks_for_truncate(inode));
140 if (!IS_ERR(result))
141 return result;
142
143 ext3_std_error(inode->i_sb, PTR_ERR(result));
144 return result;
145}
146
147/*
148 * Try to extend this transaction for the purposes of truncation.
149 *
150 * Returns 0 if we managed to create more room. If we can't create more
151 * room, and the transaction must be restarted we return 1.
152 */
153static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
154{
155 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
156 return 0;
157 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
158 return 0;
159 return 1;
160}
161
162/*
163 * Restart the transaction associated with *handle. This does a commit,
164 * so before we call here everything must be consistently dirtied against
165 * this transaction.
166 */
167static int truncate_restart_transaction(handle_t *handle, struct inode *inode)
168{
169 int ret;
170
171 jbd_debug(2, "restarting handle %p\n", handle);
172 /*
173 * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
174 * At this moment, get_block can be called only for blocks inside
175 * i_size since page cache has been already dropped and writes are
176 * blocked by i_mutex. So we can safely drop the truncate_mutex.
177 */
178 mutex_unlock(&EXT3_I(inode)->truncate_mutex);
179 ret = ext3_journal_restart(handle, blocks_for_truncate(inode));
180 mutex_lock(&EXT3_I(inode)->truncate_mutex);
181 return ret;
182}
183
184/*
185 * Called at inode eviction from icache
186 */
187void ext3_evict_inode (struct inode *inode)
188{
189 struct ext3_inode_info *ei = EXT3_I(inode);
190 struct ext3_block_alloc_info *rsv;
191 handle_t *handle;
192 int want_delete = 0;
193
194 trace_ext3_evict_inode(inode);
195 if (!inode->i_nlink && !is_bad_inode(inode)) {
196 dquot_initialize(inode);
197 want_delete = 1;
198 }
199
200 /*
201 * When journalling data dirty buffers are tracked only in the journal.
202 * So although mm thinks everything is clean and ready for reaping the
203 * inode might still have some pages to write in the running
204 * transaction or waiting to be checkpointed. Thus calling
205 * journal_invalidatepage() (via truncate_inode_pages()) to discard
206 * these buffers can cause data loss. Also even if we did not discard
207 * these buffers, we would have no way to find them after the inode
208 * is reaped and thus user could see stale data if he tries to read
209 * them before the transaction is checkpointed. So be careful and
210 * force everything to disk here... We use ei->i_datasync_tid to
211 * store the newest transaction containing inode's data.
212 *
213 * Note that directories do not have this problem because they don't
214 * use page cache.
215 *
216 * The s_journal check handles the case when ext3_get_journal() fails
217 * and puts the journal inode.
218 */
219 if (inode->i_nlink && ext3_should_journal_data(inode) &&
220 EXT3_SB(inode->i_sb)->s_journal &&
221 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
222 tid_t commit_tid = atomic_read(&ei->i_datasync_tid);
223 journal_t *journal = EXT3_SB(inode->i_sb)->s_journal;
224
225 log_start_commit(journal, commit_tid);
226 log_wait_commit(journal, commit_tid);
227 filemap_write_and_wait(&inode->i_data);
228 }
229 truncate_inode_pages(&inode->i_data, 0);
230
231 ext3_discard_reservation(inode);
232 rsv = ei->i_block_alloc_info;
233 ei->i_block_alloc_info = NULL;
234 if (unlikely(rsv))
235 kfree(rsv);
236
237 if (!want_delete)
238 goto no_delete;
239
240 handle = start_transaction(inode);
241 if (IS_ERR(handle)) {
242 /*
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
245 * cleaned up.
246 */
247 ext3_orphan_del(NULL, inode);
248 goto no_delete;
249 }
250
251 if (IS_SYNC(inode))
252 handle->h_sync = 1;
253 inode->i_size = 0;
254 if (inode->i_blocks)
255 ext3_truncate(inode);
256 /*
257 * Kill off the orphan record created when the inode lost the last
258 * link. Note that ext3_orphan_del() has to be able to cope with the
259 * deletion of a non-existent orphan - ext3_truncate() could
260 * have removed the record.
261 */
262 ext3_orphan_del(handle, inode);
263 ei->i_dtime = get_seconds();
264
265 /*
266 * One subtle ordering requirement: if anything has gone wrong
267 * (transaction abort, IO errors, whatever), then we can still
268 * do these next steps (the fs will already have been marked as
269 * having errors), but we can't free the inode if the mark_dirty
270 * fails.
271 */
272 if (ext3_mark_inode_dirty(handle, inode)) {
273 /* If that failed, just dquot_drop() and be done with that */
274 dquot_drop(inode);
275 clear_inode(inode);
276 } else {
277 ext3_xattr_delete_inode(handle, inode);
278 dquot_free_inode(inode);
279 dquot_drop(inode);
280 clear_inode(inode);
281 ext3_free_inode(handle, inode);
282 }
283 ext3_journal_stop(handle);
284 return;
285no_delete:
286 clear_inode(inode);
287 dquot_drop(inode);
288}
289
290typedef struct {
291 __le32 *p;
292 __le32 key;
293 struct buffer_head *bh;
294} Indirect;
295
296static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
297{
298 p->key = *(p->p = v);
299 p->bh = bh;
300}
301
302static int verify_chain(Indirect *from, Indirect *to)
303{
304 while (from <= to && from->key == *from->p)
305 from++;
306 return (from > to);
307}
308
309/**
310 * ext3_block_to_path - parse the block number into array of offsets
311 * @inode: inode in question (we are only interested in its superblock)
312 * @i_block: block number to be parsed
313 * @offsets: array to store the offsets in
314 * @boundary: set this non-zero if the referred-to block is likely to be
315 * followed (on disk) by an indirect block.
316 *
317 * To store the locations of file's data ext3 uses a data structure common
318 * for UNIX filesystems - tree of pointers anchored in the inode, with
319 * data blocks at leaves and indirect blocks in intermediate nodes.
320 * This function translates the block number into path in that tree -
321 * return value is the path length and @offsets[n] is the offset of
322 * pointer to (n+1)th node in the nth one. If @block is out of range
323 * (negative or too large) warning is printed and zero returned.
324 *
325 * Note: function doesn't find node addresses, so no IO is needed. All
326 * we need to know is the capacity of indirect blocks (taken from the
327 * inode->i_sb).
328 */
329
330/*
331 * Portability note: the last comparison (check that we fit into triple
332 * indirect block) is spelled differently, because otherwise on an
333 * architecture with 32-bit longs and 8Kb pages we might get into trouble
334 * if our filesystem had 8Kb blocks. We might use long long, but that would
335 * kill us on x86. Oh, well, at least the sign propagation does not matter -
336 * i_block would have to be negative in the very beginning, so we would not
337 * get there at all.
338 */
339
340static int ext3_block_to_path(struct inode *inode,
341 long i_block, int offsets[4], int *boundary)
342{
343 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
344 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
345 const long direct_blocks = EXT3_NDIR_BLOCKS,
346 indirect_blocks = ptrs,
347 double_blocks = (1 << (ptrs_bits * 2));
348 int n = 0;
349 int final = 0;
350
351 if (i_block < 0) {
352 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
353 } else if (i_block < direct_blocks) {
354 offsets[n++] = i_block;
355 final = direct_blocks;
356 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
357 offsets[n++] = EXT3_IND_BLOCK;
358 offsets[n++] = i_block;
359 final = ptrs;
360 } else if ((i_block -= indirect_blocks) < double_blocks) {
361 offsets[n++] = EXT3_DIND_BLOCK;
362 offsets[n++] = i_block >> ptrs_bits;
363 offsets[n++] = i_block & (ptrs - 1);
364 final = ptrs;
365 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
366 offsets[n++] = EXT3_TIND_BLOCK;
367 offsets[n++] = i_block >> (ptrs_bits * 2);
368 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
369 offsets[n++] = i_block & (ptrs - 1);
370 final = ptrs;
371 } else {
372 ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
373 }
374 if (boundary)
375 *boundary = final - 1 - (i_block & (ptrs - 1));
376 return n;
377}
378
379/**
380 * ext3_get_branch - read the chain of indirect blocks leading to data
381 * @inode: inode in question
382 * @depth: depth of the chain (1 - direct pointer, etc.)
383 * @offsets: offsets of pointers in inode/indirect blocks
384 * @chain: place to store the result
385 * @err: here we store the error value
386 *
387 * Function fills the array of triples <key, p, bh> and returns %NULL
388 * if everything went OK or the pointer to the last filled triple
389 * (incomplete one) otherwise. Upon the return chain[i].key contains
390 * the number of (i+1)-th block in the chain (as it is stored in memory,
391 * i.e. little-endian 32-bit), chain[i].p contains the address of that
392 * number (it points into struct inode for i==0 and into the bh->b_data
393 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
394 * block for i>0 and NULL for i==0. In other words, it holds the block
395 * numbers of the chain, addresses they were taken from (and where we can
396 * verify that chain did not change) and buffer_heads hosting these
397 * numbers.
398 *
399 * Function stops when it stumbles upon zero pointer (absent block)
400 * (pointer to last triple returned, *@err == 0)
401 * or when it gets an IO error reading an indirect block
402 * (ditto, *@err == -EIO)
403 * or when it notices that chain had been changed while it was reading
404 * (ditto, *@err == -EAGAIN)
405 * or when it reads all @depth-1 indirect blocks successfully and finds
406 * the whole chain, all way to the data (returns %NULL, *err == 0).
407 */
408static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
409 Indirect chain[4], int *err)
410{
411 struct super_block *sb = inode->i_sb;
412 Indirect *p = chain;
413 struct buffer_head *bh;
414
415 *err = 0;
416 /* i_data is not going away, no lock needed */
417 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
418 if (!p->key)
419 goto no_block;
420 while (--depth) {
421 bh = sb_bread(sb, le32_to_cpu(p->key));
422 if (!bh)
423 goto failure;
424 /* Reader: pointers */
425 if (!verify_chain(chain, p))
426 goto changed;
427 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
428 /* Reader: end */
429 if (!p->key)
430 goto no_block;
431 }
432 return NULL;
433
434changed:
435 brelse(bh);
436 *err = -EAGAIN;
437 goto no_block;
438failure:
439 *err = -EIO;
440no_block:
441 return p;
442}
443
444/**
445 * ext3_find_near - find a place for allocation with sufficient locality
446 * @inode: owner
447 * @ind: descriptor of indirect block.
448 *
449 * This function returns the preferred place for block allocation.
450 * It is used when heuristic for sequential allocation fails.
451 * Rules are:
452 * + if there is a block to the left of our position - allocate near it.
453 * + if pointer will live in indirect block - allocate near that block.
454 * + if pointer will live in inode - allocate in the same
455 * cylinder group.
456 *
457 * In the latter case we colour the starting block by the callers PID to
458 * prevent it from clashing with concurrent allocations for a different inode
459 * in the same block group. The PID is used here so that functionally related
460 * files will be close-by on-disk.
461 *
462 * Caller must make sure that @ind is valid and will stay that way.
463 */
464static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
465{
466 struct ext3_inode_info *ei = EXT3_I(inode);
467 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
468 __le32 *p;
469 ext3_fsblk_t bg_start;
470 ext3_grpblk_t colour;
471
472 /* Try to find previous block */
473 for (p = ind->p - 1; p >= start; p--) {
474 if (*p)
475 return le32_to_cpu(*p);
476 }
477
478 /* No such thing, so let's try location of indirect block */
479 if (ind->bh)
480 return ind->bh->b_blocknr;
481
482 /*
483 * It is going to be referred to from the inode itself? OK, just put it
484 * into the same cylinder group then.
485 */
486 bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
487 colour = (current->pid % 16) *
488 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
489 return bg_start + colour;
490}
491
492/**
493 * ext3_find_goal - find a preferred place for allocation.
494 * @inode: owner
495 * @block: block we want
496 * @partial: pointer to the last triple within a chain
497 *
498 * Normally this function find the preferred place for block allocation,
499 * returns it.
500 */
501
502static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
503 Indirect *partial)
504{
505 struct ext3_block_alloc_info *block_i;
506
507 block_i = EXT3_I(inode)->i_block_alloc_info;
508
509 /*
510 * try the heuristic for sequential allocation,
511 * failing that at least try to get decent locality.
512 */
513 if (block_i && (block == block_i->last_alloc_logical_block + 1)
514 && (block_i->last_alloc_physical_block != 0)) {
515 return block_i->last_alloc_physical_block + 1;
516 }
517
518 return ext3_find_near(inode, partial);
519}
520
521/**
522 * ext3_blks_to_allocate - Look up the block map and count the number
523 * of direct blocks need to be allocated for the given branch.
524 *
525 * @branch: chain of indirect blocks
526 * @k: number of blocks need for indirect blocks
527 * @blks: number of data blocks to be mapped.
528 * @blocks_to_boundary: the offset in the indirect block
529 *
530 * return the total number of blocks to be allocate, including the
531 * direct and indirect blocks.
532 */
533static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
534 int blocks_to_boundary)
535{
536 unsigned long count = 0;
537
538 /*
539 * Simple case, [t,d]Indirect block(s) has not allocated yet
540 * then it's clear blocks on that path have not allocated
541 */
542 if (k > 0) {
543 /* right now we don't handle cross boundary allocation */
544 if (blks < blocks_to_boundary + 1)
545 count += blks;
546 else
547 count += blocks_to_boundary + 1;
548 return count;
549 }
550
551 count++;
552 while (count < blks && count <= blocks_to_boundary &&
553 le32_to_cpu(*(branch[0].p + count)) == 0) {
554 count++;
555 }
556 return count;
557}
558
559/**
560 * ext3_alloc_blocks - multiple allocate blocks needed for a branch
561 * @handle: handle for this transaction
562 * @inode: owner
563 * @goal: preferred place for allocation
564 * @indirect_blks: the number of blocks need to allocate for indirect
565 * blocks
566 * @blks: number of blocks need to allocated for direct blocks
567 * @new_blocks: on return it will store the new block numbers for
568 * the indirect blocks(if needed) and the first direct block,
569 * @err: here we store the error value
570 *
571 * return the number of direct blocks allocated
572 */
573static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
574 ext3_fsblk_t goal, int indirect_blks, int blks,
575 ext3_fsblk_t new_blocks[4], int *err)
576{
577 int target, i;
578 unsigned long count = 0;
579 int index = 0;
580 ext3_fsblk_t current_block = 0;
581 int ret = 0;
582
583 /*
584 * Here we try to allocate the requested multiple blocks at once,
585 * on a best-effort basis.
586 * To build a branch, we should allocate blocks for
587 * the indirect blocks(if not allocated yet), and at least
588 * the first direct block of this branch. That's the
589 * minimum number of blocks need to allocate(required)
590 */
591 target = blks + indirect_blks;
592
593 while (1) {
594 count = target;
595 /* allocating blocks for indirect blocks and direct blocks */
596 current_block = ext3_new_blocks(handle,inode,goal,&count,err);
597 if (*err)
598 goto failed_out;
599
600 target -= count;
601 /* allocate blocks for indirect blocks */
602 while (index < indirect_blks && count) {
603 new_blocks[index++] = current_block++;
604 count--;
605 }
606
607 if (count > 0)
608 break;
609 }
610
611 /* save the new block number for the first direct block */
612 new_blocks[index] = current_block;
613
614 /* total number of blocks allocated for direct blocks */
615 ret = count;
616 *err = 0;
617 return ret;
618failed_out:
619 for (i = 0; i <index; i++)
620 ext3_free_blocks(handle, inode, new_blocks[i], 1);
621 return ret;
622}
623
624/**
625 * ext3_alloc_branch - allocate and set up a chain of blocks.
626 * @handle: handle for this transaction
627 * @inode: owner
628 * @indirect_blks: number of allocated indirect blocks
629 * @blks: number of allocated direct blocks
630 * @goal: preferred place for allocation
631 * @offsets: offsets (in the blocks) to store the pointers to next.
632 * @branch: place to store the chain in.
633 *
634 * This function allocates blocks, zeroes out all but the last one,
635 * links them into chain and (if we are synchronous) writes them to disk.
636 * In other words, it prepares a branch that can be spliced onto the
637 * inode. It stores the information about that chain in the branch[], in
638 * the same format as ext3_get_branch() would do. We are calling it after
639 * we had read the existing part of chain and partial points to the last
640 * triple of that (one with zero ->key). Upon the exit we have the same
641 * picture as after the successful ext3_get_block(), except that in one
642 * place chain is disconnected - *branch->p is still zero (we did not
643 * set the last link), but branch->key contains the number that should
644 * be placed into *branch->p to fill that gap.
645 *
646 * If allocation fails we free all blocks we've allocated (and forget
647 * their buffer_heads) and return the error value the from failed
648 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
649 * as described above and return 0.
650 */
651static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
652 int indirect_blks, int *blks, ext3_fsblk_t goal,
653 int *offsets, Indirect *branch)
654{
655 int blocksize = inode->i_sb->s_blocksize;
656 int i, n = 0;
657 int err = 0;
658 struct buffer_head *bh;
659 int num;
660 ext3_fsblk_t new_blocks[4];
661 ext3_fsblk_t current_block;
662
663 num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
664 *blks, new_blocks, &err);
665 if (err)
666 return err;
667
668 branch[0].key = cpu_to_le32(new_blocks[0]);
669 /*
670 * metadata blocks and data blocks are allocated.
671 */
672 for (n = 1; n <= indirect_blks; n++) {
673 /*
674 * Get buffer_head for parent block, zero it out
675 * and set the pointer to new one, then send
676 * parent to disk.
677 */
678 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
679 branch[n].bh = bh;
680 lock_buffer(bh);
681 BUFFER_TRACE(bh, "call get_create_access");
682 err = ext3_journal_get_create_access(handle, bh);
683 if (err) {
684 unlock_buffer(bh);
685 brelse(bh);
686 goto failed;
687 }
688
689 memset(bh->b_data, 0, blocksize);
690 branch[n].p = (__le32 *) bh->b_data + offsets[n];
691 branch[n].key = cpu_to_le32(new_blocks[n]);
692 *branch[n].p = branch[n].key;
693 if ( n == indirect_blks) {
694 current_block = new_blocks[n];
695 /*
696 * End of chain, update the last new metablock of
697 * the chain to point to the new allocated
698 * data blocks numbers
699 */
700 for (i=1; i < num; i++)
701 *(branch[n].p + i) = cpu_to_le32(++current_block);
702 }
703 BUFFER_TRACE(bh, "marking uptodate");
704 set_buffer_uptodate(bh);
705 unlock_buffer(bh);
706
707 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
708 err = ext3_journal_dirty_metadata(handle, bh);
709 if (err)
710 goto failed;
711 }
712 *blks = num;
713 return err;
714failed:
715 /* Allocation failed, free what we already allocated */
716 for (i = 1; i <= n ; i++) {
717 BUFFER_TRACE(branch[i].bh, "call journal_forget");
718 ext3_journal_forget(handle, branch[i].bh);
719 }
720 for (i = 0; i <indirect_blks; i++)
721 ext3_free_blocks(handle, inode, new_blocks[i], 1);
722
723 ext3_free_blocks(handle, inode, new_blocks[i], num);
724
725 return err;
726}
727
728/**
729 * ext3_splice_branch - splice the allocated branch onto inode.
730 * @handle: handle for this transaction
731 * @inode: owner
732 * @block: (logical) number of block we are adding
733 * @where: location of missing link
734 * @num: number of indirect blocks we are adding
735 * @blks: number of direct blocks we are adding
736 *
737 * This function fills the missing link and does all housekeeping needed in
738 * inode (->i_blocks, etc.). In case of success we end up with the full
739 * chain to new block and return 0.
740 */
741static int ext3_splice_branch(handle_t *handle, struct inode *inode,
742 long block, Indirect *where, int num, int blks)
743{
744 int i;
745 int err = 0;
746 struct ext3_block_alloc_info *block_i;
747 ext3_fsblk_t current_block;
748 struct ext3_inode_info *ei = EXT3_I(inode);
749 struct timespec now;
750
751 block_i = ei->i_block_alloc_info;
752 /*
753 * If we're splicing into a [td]indirect block (as opposed to the
754 * inode) then we need to get write access to the [td]indirect block
755 * before the splice.
756 */
757 if (where->bh) {
758 BUFFER_TRACE(where->bh, "get_write_access");
759 err = ext3_journal_get_write_access(handle, where->bh);
760 if (err)
761 goto err_out;
762 }
763 /* That's it */
764
765 *where->p = where->key;
766
767 /*
768 * Update the host buffer_head or inode to point to more just allocated
769 * direct blocks blocks
770 */
771 if (num == 0 && blks > 1) {
772 current_block = le32_to_cpu(where->key) + 1;
773 for (i = 1; i < blks; i++)
774 *(where->p + i ) = cpu_to_le32(current_block++);
775 }
776
777 /*
778 * update the most recently allocated logical & physical block
779 * in i_block_alloc_info, to assist find the proper goal block for next
780 * allocation
781 */
782 if (block_i) {
783 block_i->last_alloc_logical_block = block + blks - 1;
784 block_i->last_alloc_physical_block =
785 le32_to_cpu(where[num].key) + blks - 1;
786 }
787
788 /* We are done with atomic stuff, now do the rest of housekeeping */
789 now = CURRENT_TIME_SEC;
790 if (!timespec_equal(&inode->i_ctime, &now) || !where->bh) {
791 inode->i_ctime = now;
792 ext3_mark_inode_dirty(handle, inode);
793 }
794 /* ext3_mark_inode_dirty already updated i_sync_tid */
795 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
796
797 /* had we spliced it onto indirect block? */
798 if (where->bh) {
799 /*
800 * If we spliced it onto an indirect block, we haven't
801 * altered the inode. Note however that if it is being spliced
802 * onto an indirect block at the very end of the file (the
803 * file is growing) then we *will* alter the inode to reflect
804 * the new i_size. But that is not done here - it is done in
805 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
806 */
807 jbd_debug(5, "splicing indirect only\n");
808 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
809 err = ext3_journal_dirty_metadata(handle, where->bh);
810 if (err)
811 goto err_out;
812 } else {
813 /*
814 * OK, we spliced it into the inode itself on a direct block.
815 * Inode was dirtied above.
816 */
817 jbd_debug(5, "splicing direct\n");
818 }
819 return err;
820
821err_out:
822 for (i = 1; i <= num; i++) {
823 BUFFER_TRACE(where[i].bh, "call journal_forget");
824 ext3_journal_forget(handle, where[i].bh);
825 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
826 }
827 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
828
829 return err;
830}
831
832/*
833 * Allocation strategy is simple: if we have to allocate something, we will
834 * have to go the whole way to leaf. So let's do it before attaching anything
835 * to tree, set linkage between the newborn blocks, write them if sync is
836 * required, recheck the path, free and repeat if check fails, otherwise
837 * set the last missing link (that will protect us from any truncate-generated
838 * removals - all blocks on the path are immune now) and possibly force the
839 * write on the parent block.
840 * That has a nice additional property: no special recovery from the failed
841 * allocations is needed - we simply release blocks and do not touch anything
842 * reachable from inode.
843 *
844 * `handle' can be NULL if create == 0.
845 *
846 * The BKL may not be held on entry here. Be sure to take it early.
847 * return > 0, # of blocks mapped or allocated.
848 * return = 0, if plain lookup failed.
849 * return < 0, error case.
850 */
851int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
852 sector_t iblock, unsigned long maxblocks,
853 struct buffer_head *bh_result,
854 int create)
855{
856 int err = -EIO;
857 int offsets[4];
858 Indirect chain[4];
859 Indirect *partial;
860 ext3_fsblk_t goal;
861 int indirect_blks;
862 int blocks_to_boundary = 0;
863 int depth;
864 struct ext3_inode_info *ei = EXT3_I(inode);
865 int count = 0;
866 ext3_fsblk_t first_block = 0;
867
868
869 trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create);
870 J_ASSERT(handle != NULL || create == 0);
871 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
872
873 if (depth == 0)
874 goto out;
875
876 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
877
878 /* Simplest case - block found, no allocation needed */
879 if (!partial) {
880 first_block = le32_to_cpu(chain[depth - 1].key);
881 clear_buffer_new(bh_result);
882 count++;
883 /*map more blocks*/
884 while (count < maxblocks && count <= blocks_to_boundary) {
885 ext3_fsblk_t blk;
886
887 if (!verify_chain(chain, chain + depth - 1)) {
888 /*
889 * Indirect block might be removed by
890 * truncate while we were reading it.
891 * Handling of that case: forget what we've
892 * got now. Flag the err as EAGAIN, so it
893 * will reread.
894 */
895 err = -EAGAIN;
896 count = 0;
897 break;
898 }
899 blk = le32_to_cpu(*(chain[depth-1].p + count));
900
901 if (blk == first_block + count)
902 count++;
903 else
904 break;
905 }
906 if (err != -EAGAIN)
907 goto got_it;
908 }
909
910 /* Next simple case - plain lookup or failed read of indirect block */
911 if (!create || err == -EIO)
912 goto cleanup;
913
914 /*
915 * Block out ext3_truncate while we alter the tree
916 */
917 mutex_lock(&ei->truncate_mutex);
918
919 /*
920 * If the indirect block is missing while we are reading
921 * the chain(ext3_get_branch() returns -EAGAIN err), or
922 * if the chain has been changed after we grab the semaphore,
923 * (either because another process truncated this branch, or
924 * another get_block allocated this branch) re-grab the chain to see if
925 * the request block has been allocated or not.
926 *
927 * Since we already block the truncate/other get_block
928 * at this point, we will have the current copy of the chain when we
929 * splice the branch into the tree.
930 */
931 if (err == -EAGAIN || !verify_chain(chain, partial)) {
932 while (partial > chain) {
933 brelse(partial->bh);
934 partial--;
935 }
936 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
937 if (!partial) {
938 count++;
939 mutex_unlock(&ei->truncate_mutex);
940 if (err)
941 goto cleanup;
942 clear_buffer_new(bh_result);
943 goto got_it;
944 }
945 }
946
947 /*
948 * Okay, we need to do block allocation. Lazily initialize the block
949 * allocation info here if necessary
950 */
951 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
952 ext3_init_block_alloc_info(inode);
953
954 goal = ext3_find_goal(inode, iblock, partial);
955
956 /* the number of blocks need to allocate for [d,t]indirect blocks */
957 indirect_blks = (chain + depth) - partial - 1;
958
959 /*
960 * Next look up the indirect map to count the totoal number of
961 * direct blocks to allocate for this branch.
962 */
963 count = ext3_blks_to_allocate(partial, indirect_blks,
964 maxblocks, blocks_to_boundary);
965 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
966 offsets + (partial - chain), partial);
967
968 /*
969 * The ext3_splice_branch call will free and forget any buffers
970 * on the new chain if there is a failure, but that risks using
971 * up transaction credits, especially for bitmaps where the
972 * credits cannot be returned. Can we handle this somehow? We
973 * may need to return -EAGAIN upwards in the worst case. --sct
974 */
975 if (!err)
976 err = ext3_splice_branch(handle, inode, iblock,
977 partial, indirect_blks, count);
978 mutex_unlock(&ei->truncate_mutex);
979 if (err)
980 goto cleanup;
981
982 set_buffer_new(bh_result);
983got_it:
984 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
985 if (count > blocks_to_boundary)
986 set_buffer_boundary(bh_result);
987 err = count;
988 /* Clean up and exit */
989 partial = chain + depth - 1; /* the whole chain */
990cleanup:
991 while (partial > chain) {
992 BUFFER_TRACE(partial->bh, "call brelse");
993 brelse(partial->bh);
994 partial--;
995 }
996 BUFFER_TRACE(bh_result, "returned");
997out:
998 trace_ext3_get_blocks_exit(inode, iblock,
999 depth ? le32_to_cpu(chain[depth-1].key) : 0,
1000 count, err);
1001 return err;
1002}
1003
1004/* Maximum number of blocks we map for direct IO at once. */
1005#define DIO_MAX_BLOCKS 4096
1006/*
1007 * Number of credits we need for writing DIO_MAX_BLOCKS:
1008 * We need sb + group descriptor + bitmap + inode -> 4
1009 * For B blocks with A block pointers per block we need:
1010 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
1011 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
1012 */
1013#define DIO_CREDITS 25
1014
1015static int ext3_get_block(struct inode *inode, sector_t iblock,
1016 struct buffer_head *bh_result, int create)
1017{
1018 handle_t *handle = ext3_journal_current_handle();
1019 int ret = 0, started = 0;
1020 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1021
1022 if (create && !handle) { /* Direct IO write... */
1023 if (max_blocks > DIO_MAX_BLOCKS)
1024 max_blocks = DIO_MAX_BLOCKS;
1025 handle = ext3_journal_start(inode, DIO_CREDITS +
1026 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
1027 if (IS_ERR(handle)) {
1028 ret = PTR_ERR(handle);
1029 goto out;
1030 }
1031 started = 1;
1032 }
1033
1034 ret = ext3_get_blocks_handle(handle, inode, iblock,
1035 max_blocks, bh_result, create);
1036 if (ret > 0) {
1037 bh_result->b_size = (ret << inode->i_blkbits);
1038 ret = 0;
1039 }
1040 if (started)
1041 ext3_journal_stop(handle);
1042out:
1043 return ret;
1044}
1045
1046int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1047 u64 start, u64 len)
1048{
1049 return generic_block_fiemap(inode, fieinfo, start, len,
1050 ext3_get_block);
1051}
1052
1053/*
1054 * `handle' can be NULL if create is zero
1055 */
1056struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1057 long block, int create, int *errp)
1058{
1059 struct buffer_head dummy;
1060 int fatal = 0, err;
1061
1062 J_ASSERT(handle != NULL || create == 0);
1063
1064 dummy.b_state = 0;
1065 dummy.b_blocknr = -1000;
1066 buffer_trace_init(&dummy.b_history);
1067 err = ext3_get_blocks_handle(handle, inode, block, 1,
1068 &dummy, create);
1069 /*
1070 * ext3_get_blocks_handle() returns number of blocks
1071 * mapped. 0 in case of a HOLE.
1072 */
1073 if (err > 0) {
1074 if (err > 1)
1075 WARN_ON(1);
1076 err = 0;
1077 }
1078 *errp = err;
1079 if (!err && buffer_mapped(&dummy)) {
1080 struct buffer_head *bh;
1081 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1082 if (!bh) {
1083 *errp = -EIO;
1084 goto err;
1085 }
1086 if (buffer_new(&dummy)) {
1087 J_ASSERT(create != 0);
1088 J_ASSERT(handle != NULL);
1089
1090 /*
1091 * Now that we do not always journal data, we should
1092 * keep in mind whether this should always journal the
1093 * new buffer as metadata. For now, regular file
1094 * writes use ext3_get_block instead, so it's not a
1095 * problem.
1096 */
1097 lock_buffer(bh);
1098 BUFFER_TRACE(bh, "call get_create_access");
1099 fatal = ext3_journal_get_create_access(handle, bh);
1100 if (!fatal && !buffer_uptodate(bh)) {
1101 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1102 set_buffer_uptodate(bh);
1103 }
1104 unlock_buffer(bh);
1105 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1106 err = ext3_journal_dirty_metadata(handle, bh);
1107 if (!fatal)
1108 fatal = err;
1109 } else {
1110 BUFFER_TRACE(bh, "not a new buffer");
1111 }
1112 if (fatal) {
1113 *errp = fatal;
1114 brelse(bh);
1115 bh = NULL;
1116 }
1117 return bh;
1118 }
1119err:
1120 return NULL;
1121}
1122
1123struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
1124 int block, int create, int *err)
1125{
1126 struct buffer_head * bh;
1127
1128 bh = ext3_getblk(handle, inode, block, create, err);
1129 if (!bh)
1130 return bh;
1131 if (bh_uptodate_or_lock(bh))
1132 return bh;
1133 get_bh(bh);
1134 bh->b_end_io = end_buffer_read_sync;
1135 submit_bh(READ | REQ_META | REQ_PRIO, bh);
1136 wait_on_buffer(bh);
1137 if (buffer_uptodate(bh))
1138 return bh;
1139 put_bh(bh);
1140 *err = -EIO;
1141 return NULL;
1142}
1143
1144static int walk_page_buffers( handle_t *handle,
1145 struct buffer_head *head,
1146 unsigned from,
1147 unsigned to,
1148 int *partial,
1149 int (*fn)( handle_t *handle,
1150 struct buffer_head *bh))
1151{
1152 struct buffer_head *bh;
1153 unsigned block_start, block_end;
1154 unsigned blocksize = head->b_size;
1155 int err, ret = 0;
1156 struct buffer_head *next;
1157
1158 for ( bh = head, block_start = 0;
1159 ret == 0 && (bh != head || !block_start);
1160 block_start = block_end, bh = next)
1161 {
1162 next = bh->b_this_page;
1163 block_end = block_start + blocksize;
1164 if (block_end <= from || block_start >= to) {
1165 if (partial && !buffer_uptodate(bh))
1166 *partial = 1;
1167 continue;
1168 }
1169 err = (*fn)(handle, bh);
1170 if (!ret)
1171 ret = err;
1172 }
1173 return ret;
1174}
1175
1176/*
1177 * To preserve ordering, it is essential that the hole instantiation and
1178 * the data write be encapsulated in a single transaction. We cannot
1179 * close off a transaction and start a new one between the ext3_get_block()
1180 * and the commit_write(). So doing the journal_start at the start of
1181 * prepare_write() is the right place.
1182 *
1183 * Also, this function can nest inside ext3_writepage() ->
1184 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1185 * has generated enough buffer credits to do the whole page. So we won't
1186 * block on the journal in that case, which is good, because the caller may
1187 * be PF_MEMALLOC.
1188 *
1189 * By accident, ext3 can be reentered when a transaction is open via
1190 * quota file writes. If we were to commit the transaction while thus
1191 * reentered, there can be a deadlock - we would be holding a quota
1192 * lock, and the commit would never complete if another thread had a
1193 * transaction open and was blocking on the quota lock - a ranking
1194 * violation.
1195 *
1196 * So what we do is to rely on the fact that journal_stop/journal_start
1197 * will _not_ run commit under these circumstances because handle->h_ref
1198 * is elevated. We'll still have enough credits for the tiny quotafile
1199 * write.
1200 */
1201static int do_journal_get_write_access(handle_t *handle,
1202 struct buffer_head *bh)
1203{
1204 int dirty = buffer_dirty(bh);
1205 int ret;
1206
1207 if (!buffer_mapped(bh) || buffer_freed(bh))
1208 return 0;
1209 /*
1210 * __block_prepare_write() could have dirtied some buffers. Clean
1211 * the dirty bit as jbd2_journal_get_write_access() could complain
1212 * otherwise about fs integrity issues. Setting of the dirty bit
1213 * by __block_prepare_write() isn't a real problem here as we clear
1214 * the bit before releasing a page lock and thus writeback cannot
1215 * ever write the buffer.
1216 */
1217 if (dirty)
1218 clear_buffer_dirty(bh);
1219 ret = ext3_journal_get_write_access(handle, bh);
1220 if (!ret && dirty)
1221 ret = ext3_journal_dirty_metadata(handle, bh);
1222 return ret;
1223}
1224
1225/*
1226 * Truncate blocks that were not used by write. We have to truncate the
1227 * pagecache as well so that corresponding buffers get properly unmapped.
1228 */
1229static void ext3_truncate_failed_write(struct inode *inode)
1230{
1231 truncate_inode_pages(inode->i_mapping, inode->i_size);
1232 ext3_truncate(inode);
1233}
1234
1235/*
1236 * Truncate blocks that were not used by direct IO write. We have to zero out
1237 * the last file block as well because direct IO might have written to it.
1238 */
1239static void ext3_truncate_failed_direct_write(struct inode *inode)
1240{
1241 ext3_block_truncate_page(inode, inode->i_size);
1242 ext3_truncate(inode);
1243}
1244
1245static int ext3_write_begin(struct file *file, struct address_space *mapping,
1246 loff_t pos, unsigned len, unsigned flags,
1247 struct page **pagep, void **fsdata)
1248{
1249 struct inode *inode = mapping->host;
1250 int ret;
1251 handle_t *handle;
1252 int retries = 0;
1253 struct page *page;
1254 pgoff_t index;
1255 unsigned from, to;
1256 /* Reserve one block more for addition to orphan list in case
1257 * we allocate blocks but write fails for some reason */
1258 int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
1259
1260 trace_ext3_write_begin(inode, pos, len, flags);
1261
1262 index = pos >> PAGE_CACHE_SHIFT;
1263 from = pos & (PAGE_CACHE_SIZE - 1);
1264 to = from + len;
1265
1266retry:
1267 page = grab_cache_page_write_begin(mapping, index, flags);
1268 if (!page)
1269 return -ENOMEM;
1270 *pagep = page;
1271
1272 handle = ext3_journal_start(inode, needed_blocks);
1273 if (IS_ERR(handle)) {
1274 unlock_page(page);
1275 page_cache_release(page);
1276 ret = PTR_ERR(handle);
1277 goto out;
1278 }
1279 ret = __block_write_begin(page, pos, len, ext3_get_block);
1280 if (ret)
1281 goto write_begin_failed;
1282
1283 if (ext3_should_journal_data(inode)) {
1284 ret = walk_page_buffers(handle, page_buffers(page),
1285 from, to, NULL, do_journal_get_write_access);
1286 }
1287write_begin_failed:
1288 if (ret) {
1289 /*
1290 * block_write_begin may have instantiated a few blocks
1291 * outside i_size. Trim these off again. Don't need
1292 * i_size_read because we hold i_mutex.
1293 *
1294 * Add inode to orphan list in case we crash before truncate
1295 * finishes. Do this only if ext3_can_truncate() agrees so
1296 * that orphan processing code is happy.
1297 */
1298 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1299 ext3_orphan_add(handle, inode);
1300 ext3_journal_stop(handle);
1301 unlock_page(page);
1302 page_cache_release(page);
1303 if (pos + len > inode->i_size)
1304 ext3_truncate_failed_write(inode);
1305 }
1306 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1307 goto retry;
1308out:
1309 return ret;
1310}
1311
1312
1313int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1314{
1315 int err = journal_dirty_data(handle, bh);
1316 if (err)
1317 ext3_journal_abort_handle(__func__, __func__,
1318 bh, handle, err);
1319 return err;
1320}
1321
1322/* For ordered writepage and write_end functions */
1323static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1324{
1325 /*
1326 * Write could have mapped the buffer but it didn't copy the data in
1327 * yet. So avoid filing such buffer into a transaction.
1328 */
1329 if (buffer_mapped(bh) && buffer_uptodate(bh))
1330 return ext3_journal_dirty_data(handle, bh);
1331 return 0;
1332}
1333
1334/* For write_end() in data=journal mode */
1335static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1336{
1337 if (!buffer_mapped(bh) || buffer_freed(bh))
1338 return 0;
1339 set_buffer_uptodate(bh);
1340 return ext3_journal_dirty_metadata(handle, bh);
1341}
1342
1343/*
1344 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1345 * for the whole page but later we failed to copy the data in. Update inode
1346 * size according to what we managed to copy. The rest is going to be
1347 * truncated in write_end function.
1348 */
1349static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
1350{
1351 /* What matters to us is i_disksize. We don't write i_size anywhere */
1352 if (pos + copied > inode->i_size)
1353 i_size_write(inode, pos + copied);
1354 if (pos + copied > EXT3_I(inode)->i_disksize) {
1355 EXT3_I(inode)->i_disksize = pos + copied;
1356 mark_inode_dirty(inode);
1357 }
1358}
1359
1360/*
1361 * We need to pick up the new inode size which generic_commit_write gave us
1362 * `file' can be NULL - eg, when called from page_symlink().
1363 *
1364 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1365 * buffers are managed internally.
1366 */
1367static int ext3_ordered_write_end(struct file *file,
1368 struct address_space *mapping,
1369 loff_t pos, unsigned len, unsigned copied,
1370 struct page *page, void *fsdata)
1371{
1372 handle_t *handle = ext3_journal_current_handle();
1373 struct inode *inode = file->f_mapping->host;
1374 unsigned from, to;
1375 int ret = 0, ret2;
1376
1377 trace_ext3_ordered_write_end(inode, pos, len, copied);
1378 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1379
1380 from = pos & (PAGE_CACHE_SIZE - 1);
1381 to = from + copied;
1382 ret = walk_page_buffers(handle, page_buffers(page),
1383 from, to, NULL, journal_dirty_data_fn);
1384
1385 if (ret == 0)
1386 update_file_sizes(inode, pos, copied);
1387 /*
1388 * There may be allocated blocks outside of i_size because
1389 * we failed to copy some data. Prepare for truncate.
1390 */
1391 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1392 ext3_orphan_add(handle, inode);
1393 ret2 = ext3_journal_stop(handle);
1394 if (!ret)
1395 ret = ret2;
1396 unlock_page(page);
1397 page_cache_release(page);
1398
1399 if (pos + len > inode->i_size)
1400 ext3_truncate_failed_write(inode);
1401 return ret ? ret : copied;
1402}
1403
1404static int ext3_writeback_write_end(struct file *file,
1405 struct address_space *mapping,
1406 loff_t pos, unsigned len, unsigned copied,
1407 struct page *page, void *fsdata)
1408{
1409 handle_t *handle = ext3_journal_current_handle();
1410 struct inode *inode = file->f_mapping->host;
1411 int ret;
1412
1413 trace_ext3_writeback_write_end(inode, pos, len, copied);
1414 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1415 update_file_sizes(inode, pos, copied);
1416 /*
1417 * There may be allocated blocks outside of i_size because
1418 * we failed to copy some data. Prepare for truncate.
1419 */
1420 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1421 ext3_orphan_add(handle, inode);
1422 ret = ext3_journal_stop(handle);
1423 unlock_page(page);
1424 page_cache_release(page);
1425
1426 if (pos + len > inode->i_size)
1427 ext3_truncate_failed_write(inode);
1428 return ret ? ret : copied;
1429}
1430
1431static int ext3_journalled_write_end(struct file *file,
1432 struct address_space *mapping,
1433 loff_t pos, unsigned len, unsigned copied,
1434 struct page *page, void *fsdata)
1435{
1436 handle_t *handle = ext3_journal_current_handle();
1437 struct inode *inode = mapping->host;
1438 struct ext3_inode_info *ei = EXT3_I(inode);
1439 int ret = 0, ret2;
1440 int partial = 0;
1441 unsigned from, to;
1442
1443 trace_ext3_journalled_write_end(inode, pos, len, copied);
1444 from = pos & (PAGE_CACHE_SIZE - 1);
1445 to = from + len;
1446
1447 if (copied < len) {
1448 if (!PageUptodate(page))
1449 copied = 0;
1450 page_zero_new_buffers(page, from + copied, to);
1451 to = from + copied;
1452 }
1453
1454 ret = walk_page_buffers(handle, page_buffers(page), from,
1455 to, &partial, write_end_fn);
1456 if (!partial)
1457 SetPageUptodate(page);
1458
1459 if (pos + copied > inode->i_size)
1460 i_size_write(inode, pos + copied);
1461 /*
1462 * There may be allocated blocks outside of i_size because
1463 * we failed to copy some data. Prepare for truncate.
1464 */
1465 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1466 ext3_orphan_add(handle, inode);
1467 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1468 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
1469 if (inode->i_size > ei->i_disksize) {
1470 ei->i_disksize = inode->i_size;
1471 ret2 = ext3_mark_inode_dirty(handle, inode);
1472 if (!ret)
1473 ret = ret2;
1474 }
1475
1476 ret2 = ext3_journal_stop(handle);
1477 if (!ret)
1478 ret = ret2;
1479 unlock_page(page);
1480 page_cache_release(page);
1481
1482 if (pos + len > inode->i_size)
1483 ext3_truncate_failed_write(inode);
1484 return ret ? ret : copied;
1485}
1486
1487/*
1488 * bmap() is special. It gets used by applications such as lilo and by
1489 * the swapper to find the on-disk block of a specific piece of data.
1490 *
1491 * Naturally, this is dangerous if the block concerned is still in the
1492 * journal. If somebody makes a swapfile on an ext3 data-journaling
1493 * filesystem and enables swap, then they may get a nasty shock when the
1494 * data getting swapped to that swapfile suddenly gets overwritten by
1495 * the original zero's written out previously to the journal and
1496 * awaiting writeback in the kernel's buffer cache.
1497 *
1498 * So, if we see any bmap calls here on a modified, data-journaled file,
1499 * take extra steps to flush any blocks which might be in the cache.
1500 */
1501static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1502{
1503 struct inode *inode = mapping->host;
1504 journal_t *journal;
1505 int err;
1506
1507 if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) {
1508 /*
1509 * This is a REALLY heavyweight approach, but the use of
1510 * bmap on dirty files is expected to be extremely rare:
1511 * only if we run lilo or swapon on a freshly made file
1512 * do we expect this to happen.
1513 *
1514 * (bmap requires CAP_SYS_RAWIO so this does not
1515 * represent an unprivileged user DOS attack --- we'd be
1516 * in trouble if mortal users could trigger this path at
1517 * will.)
1518 *
1519 * NB. EXT3_STATE_JDATA is not set on files other than
1520 * regular files. If somebody wants to bmap a directory
1521 * or symlink and gets confused because the buffer
1522 * hasn't yet been flushed to disk, they deserve
1523 * everything they get.
1524 */
1525
1526 ext3_clear_inode_state(inode, EXT3_STATE_JDATA);
1527 journal = EXT3_JOURNAL(inode);
1528 journal_lock_updates(journal);
1529 err = journal_flush(journal);
1530 journal_unlock_updates(journal);
1531
1532 if (err)
1533 return 0;
1534 }
1535
1536 return generic_block_bmap(mapping,block,ext3_get_block);
1537}
1538
1539static int bget_one(handle_t *handle, struct buffer_head *bh)
1540{
1541 get_bh(bh);
1542 return 0;
1543}
1544
1545static int bput_one(handle_t *handle, struct buffer_head *bh)
1546{
1547 put_bh(bh);
1548 return 0;
1549}
1550
1551static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1552{
1553 return !buffer_mapped(bh);
1554}
1555
1556/*
1557 * Note that we always start a transaction even if we're not journalling
1558 * data. This is to preserve ordering: any hole instantiation within
1559 * __block_write_full_page -> ext3_get_block() should be journalled
1560 * along with the data so we don't crash and then get metadata which
1561 * refers to old data.
1562 *
1563 * In all journalling modes block_write_full_page() will start the I/O.
1564 *
1565 * Problem:
1566 *
1567 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1568 * ext3_writepage()
1569 *
1570 * Similar for:
1571 *
1572 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1573 *
1574 * Same applies to ext3_get_block(). We will deadlock on various things like
1575 * lock_journal and i_truncate_mutex.
1576 *
1577 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1578 * allocations fail.
1579 *
1580 * 16May01: If we're reentered then journal_current_handle() will be
1581 * non-zero. We simply *return*.
1582 *
1583 * 1 July 2001: @@@ FIXME:
1584 * In journalled data mode, a data buffer may be metadata against the
1585 * current transaction. But the same file is part of a shared mapping
1586 * and someone does a writepage() on it.
1587 *
1588 * We will move the buffer onto the async_data list, but *after* it has
1589 * been dirtied. So there's a small window where we have dirty data on
1590 * BJ_Metadata.
1591 *
1592 * Note that this only applies to the last partial page in the file. The
1593 * bit which block_write_full_page() uses prepare/commit for. (That's
1594 * broken code anyway: it's wrong for msync()).
1595 *
1596 * It's a rare case: affects the final partial page, for journalled data
1597 * where the file is subject to bith write() and writepage() in the same
1598 * transction. To fix it we'll need a custom block_write_full_page().
1599 * We'll probably need that anyway for journalling writepage() output.
1600 *
1601 * We don't honour synchronous mounts for writepage(). That would be
1602 * disastrous. Any write() or metadata operation will sync the fs for
1603 * us.
1604 *
1605 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1606 * we don't need to open a transaction here.
1607 */
1608static int ext3_ordered_writepage(struct page *page,
1609 struct writeback_control *wbc)
1610{
1611 struct inode *inode = page->mapping->host;
1612 struct buffer_head *page_bufs;
1613 handle_t *handle = NULL;
1614 int ret = 0;
1615 int err;
1616
1617 J_ASSERT(PageLocked(page));
1618 /*
1619 * We don't want to warn for emergency remount. The condition is
1620 * ordered to avoid dereferencing inode->i_sb in non-error case to
1621 * avoid slow-downs.
1622 */
1623 WARN_ON_ONCE(IS_RDONLY(inode) &&
1624 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
1625
1626 /*
1627 * We give up here if we're reentered, because it might be for a
1628 * different filesystem.
1629 */
1630 if (ext3_journal_current_handle())
1631 goto out_fail;
1632
1633 trace_ext3_ordered_writepage(page);
1634 if (!page_has_buffers(page)) {
1635 create_empty_buffers(page, inode->i_sb->s_blocksize,
1636 (1 << BH_Dirty)|(1 << BH_Uptodate));
1637 page_bufs = page_buffers(page);
1638 } else {
1639 page_bufs = page_buffers(page);
1640 if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1641 NULL, buffer_unmapped)) {
1642 /* Provide NULL get_block() to catch bugs if buffers
1643 * weren't really mapped */
1644 return block_write_full_page(page, NULL, wbc);
1645 }
1646 }
1647 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1648
1649 if (IS_ERR(handle)) {
1650 ret = PTR_ERR(handle);
1651 goto out_fail;
1652 }
1653
1654 walk_page_buffers(handle, page_bufs, 0,
1655 PAGE_CACHE_SIZE, NULL, bget_one);
1656
1657 ret = block_write_full_page(page, ext3_get_block, wbc);
1658
1659 /*
1660 * The page can become unlocked at any point now, and
1661 * truncate can then come in and change things. So we
1662 * can't touch *page from now on. But *page_bufs is
1663 * safe due to elevated refcount.
1664 */
1665
1666 /*
1667 * And attach them to the current transaction. But only if
1668 * block_write_full_page() succeeded. Otherwise they are unmapped,
1669 * and generally junk.
1670 */
1671 if (ret == 0) {
1672 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1673 NULL, journal_dirty_data_fn);
1674 if (!ret)
1675 ret = err;
1676 }
1677 walk_page_buffers(handle, page_bufs, 0,
1678 PAGE_CACHE_SIZE, NULL, bput_one);
1679 err = ext3_journal_stop(handle);
1680 if (!ret)
1681 ret = err;
1682 return ret;
1683
1684out_fail:
1685 redirty_page_for_writepage(wbc, page);
1686 unlock_page(page);
1687 return ret;
1688}
1689
1690static int ext3_writeback_writepage(struct page *page,
1691 struct writeback_control *wbc)
1692{
1693 struct inode *inode = page->mapping->host;
1694 handle_t *handle = NULL;
1695 int ret = 0;
1696 int err;
1697
1698 J_ASSERT(PageLocked(page));
1699 /*
1700 * We don't want to warn for emergency remount. The condition is
1701 * ordered to avoid dereferencing inode->i_sb in non-error case to
1702 * avoid slow-downs.
1703 */
1704 WARN_ON_ONCE(IS_RDONLY(inode) &&
1705 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
1706
1707 if (ext3_journal_current_handle())
1708 goto out_fail;
1709
1710 trace_ext3_writeback_writepage(page);
1711 if (page_has_buffers(page)) {
1712 if (!walk_page_buffers(NULL, page_buffers(page), 0,
1713 PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1714 /* Provide NULL get_block() to catch bugs if buffers
1715 * weren't really mapped */
1716 return block_write_full_page(page, NULL, wbc);
1717 }
1718 }
1719
1720 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1721 if (IS_ERR(handle)) {
1722 ret = PTR_ERR(handle);
1723 goto out_fail;
1724 }
1725
1726 ret = block_write_full_page(page, ext3_get_block, wbc);
1727
1728 err = ext3_journal_stop(handle);
1729 if (!ret)
1730 ret = err;
1731 return ret;
1732
1733out_fail:
1734 redirty_page_for_writepage(wbc, page);
1735 unlock_page(page);
1736 return ret;
1737}
1738
1739static int ext3_journalled_writepage(struct page *page,
1740 struct writeback_control *wbc)
1741{
1742 struct inode *inode = page->mapping->host;
1743 handle_t *handle = NULL;
1744 int ret = 0;
1745 int err;
1746
1747 J_ASSERT(PageLocked(page));
1748 /*
1749 * We don't want to warn for emergency remount. The condition is
1750 * ordered to avoid dereferencing inode->i_sb in non-error case to
1751 * avoid slow-downs.
1752 */
1753 WARN_ON_ONCE(IS_RDONLY(inode) &&
1754 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
1755
1756 if (ext3_journal_current_handle())
1757 goto no_write;
1758
1759 trace_ext3_journalled_writepage(page);
1760 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1761 if (IS_ERR(handle)) {
1762 ret = PTR_ERR(handle);
1763 goto no_write;
1764 }
1765
1766 if (!page_has_buffers(page) || PageChecked(page)) {
1767 /*
1768 * It's mmapped pagecache. Add buffers and journal it. There
1769 * doesn't seem much point in redirtying the page here.
1770 */
1771 ClearPageChecked(page);
1772 ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE,
1773 ext3_get_block);
1774 if (ret != 0) {
1775 ext3_journal_stop(handle);
1776 goto out_unlock;
1777 }
1778 ret = walk_page_buffers(handle, page_buffers(page), 0,
1779 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1780
1781 err = walk_page_buffers(handle, page_buffers(page), 0,
1782 PAGE_CACHE_SIZE, NULL, write_end_fn);
1783 if (ret == 0)
1784 ret = err;
1785 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1786 atomic_set(&EXT3_I(inode)->i_datasync_tid,
1787 handle->h_transaction->t_tid);
1788 unlock_page(page);
1789 } else {
1790 /*
1791 * It may be a page full of checkpoint-mode buffers. We don't
1792 * really know unless we go poke around in the buffer_heads.
1793 * But block_write_full_page will do the right thing.
1794 */
1795 ret = block_write_full_page(page, ext3_get_block, wbc);
1796 }
1797 err = ext3_journal_stop(handle);
1798 if (!ret)
1799 ret = err;
1800out:
1801 return ret;
1802
1803no_write:
1804 redirty_page_for_writepage(wbc, page);
1805out_unlock:
1806 unlock_page(page);
1807 goto out;
1808}
1809
1810static int ext3_readpage(struct file *file, struct page *page)
1811{
1812 trace_ext3_readpage(page);
1813 return mpage_readpage(page, ext3_get_block);
1814}
1815
1816static int
1817ext3_readpages(struct file *file, struct address_space *mapping,
1818 struct list_head *pages, unsigned nr_pages)
1819{
1820 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1821}
1822
1823static void ext3_invalidatepage(struct page *page, unsigned long offset)
1824{
1825 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1826
1827 trace_ext3_invalidatepage(page, offset);
1828
1829 /*
1830 * If it's a full truncate we just forget about the pending dirtying
1831 */
1832 if (offset == 0)
1833 ClearPageChecked(page);
1834
1835 journal_invalidatepage(journal, page, offset);
1836}
1837
1838static int ext3_releasepage(struct page *page, gfp_t wait)
1839{
1840 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1841
1842 trace_ext3_releasepage(page);
1843 WARN_ON(PageChecked(page));
1844 if (!page_has_buffers(page))
1845 return 0;
1846 return journal_try_to_free_buffers(journal, page, wait);
1847}
1848
1849/*
1850 * If the O_DIRECT write will extend the file then add this inode to the
1851 * orphan list. So recovery will truncate it back to the original size
1852 * if the machine crashes during the write.
1853 *
1854 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1855 * crashes then stale disk data _may_ be exposed inside the file. But current
1856 * VFS code falls back into buffered path in that case so we are safe.
1857 */
1858static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1859 const struct iovec *iov, loff_t offset,
1860 unsigned long nr_segs)
1861{
1862 struct file *file = iocb->ki_filp;
1863 struct inode *inode = file->f_mapping->host;
1864 struct ext3_inode_info *ei = EXT3_I(inode);
1865 handle_t *handle;
1866 ssize_t ret;
1867 int orphan = 0;
1868 size_t count = iov_length(iov, nr_segs);
1869 int retries = 0;
1870
1871 trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
1872
1873 if (rw == WRITE) {
1874 loff_t final_size = offset + count;
1875
1876 if (final_size > inode->i_size) {
1877 /* Credits for sb + inode write */
1878 handle = ext3_journal_start(inode, 2);
1879 if (IS_ERR(handle)) {
1880 ret = PTR_ERR(handle);
1881 goto out;
1882 }
1883 ret = ext3_orphan_add(handle, inode);
1884 if (ret) {
1885 ext3_journal_stop(handle);
1886 goto out;
1887 }
1888 orphan = 1;
1889 ei->i_disksize = inode->i_size;
1890 ext3_journal_stop(handle);
1891 }
1892 }
1893
1894retry:
1895 ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1896 ext3_get_block);
1897 /*
1898 * In case of error extending write may have instantiated a few
1899 * blocks outside i_size. Trim these off again.
1900 */
1901 if (unlikely((rw & WRITE) && ret < 0)) {
1902 loff_t isize = i_size_read(inode);
1903 loff_t end = offset + iov_length(iov, nr_segs);
1904
1905 if (end > isize)
1906 ext3_truncate_failed_direct_write(inode);
1907 }
1908 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1909 goto retry;
1910
1911 if (orphan) {
1912 int err;
1913
1914 /* Credits for sb + inode write */
1915 handle = ext3_journal_start(inode, 2);
1916 if (IS_ERR(handle)) {
1917 /* This is really bad luck. We've written the data
1918 * but cannot extend i_size. Truncate allocated blocks
1919 * and pretend the write failed... */
1920 ext3_truncate_failed_direct_write(inode);
1921 ret = PTR_ERR(handle);
1922 goto out;
1923 }
1924 if (inode->i_nlink)
1925 ext3_orphan_del(handle, inode);
1926 if (ret > 0) {
1927 loff_t end = offset + ret;
1928 if (end > inode->i_size) {
1929 ei->i_disksize = end;
1930 i_size_write(inode, end);
1931 /*
1932 * We're going to return a positive `ret'
1933 * here due to non-zero-length I/O, so there's
1934 * no way of reporting error returns from
1935 * ext3_mark_inode_dirty() to userspace. So
1936 * ignore it.
1937 */
1938 ext3_mark_inode_dirty(handle, inode);
1939 }
1940 }
1941 err = ext3_journal_stop(handle);
1942 if (ret == 0)
1943 ret = err;
1944 }
1945out:
1946 trace_ext3_direct_IO_exit(inode, offset,
1947 iov_length(iov, nr_segs), rw, ret);
1948 return ret;
1949}
1950
1951/*
1952 * Pages can be marked dirty completely asynchronously from ext3's journalling
1953 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1954 * much here because ->set_page_dirty is called under VFS locks. The page is
1955 * not necessarily locked.
1956 *
1957 * We cannot just dirty the page and leave attached buffers clean, because the
1958 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1959 * or jbddirty because all the journalling code will explode.
1960 *
1961 * So what we do is to mark the page "pending dirty" and next time writepage
1962 * is called, propagate that into the buffers appropriately.
1963 */
1964static int ext3_journalled_set_page_dirty(struct page *page)
1965{
1966 SetPageChecked(page);
1967 return __set_page_dirty_nobuffers(page);
1968}
1969
1970static const struct address_space_operations ext3_ordered_aops = {
1971 .readpage = ext3_readpage,
1972 .readpages = ext3_readpages,
1973 .writepage = ext3_ordered_writepage,
1974 .write_begin = ext3_write_begin,
1975 .write_end = ext3_ordered_write_end,
1976 .bmap = ext3_bmap,
1977 .invalidatepage = ext3_invalidatepage,
1978 .releasepage = ext3_releasepage,
1979 .direct_IO = ext3_direct_IO,
1980 .migratepage = buffer_migrate_page,
1981 .is_partially_uptodate = block_is_partially_uptodate,
1982 .error_remove_page = generic_error_remove_page,
1983};
1984
1985static const struct address_space_operations ext3_writeback_aops = {
1986 .readpage = ext3_readpage,
1987 .readpages = ext3_readpages,
1988 .writepage = ext3_writeback_writepage,
1989 .write_begin = ext3_write_begin,
1990 .write_end = ext3_writeback_write_end,
1991 .bmap = ext3_bmap,
1992 .invalidatepage = ext3_invalidatepage,
1993 .releasepage = ext3_releasepage,
1994 .direct_IO = ext3_direct_IO,
1995 .migratepage = buffer_migrate_page,
1996 .is_partially_uptodate = block_is_partially_uptodate,
1997 .error_remove_page = generic_error_remove_page,
1998};
1999
2000static const struct address_space_operations ext3_journalled_aops = {
2001 .readpage = ext3_readpage,
2002 .readpages = ext3_readpages,
2003 .writepage = ext3_journalled_writepage,
2004 .write_begin = ext3_write_begin,
2005 .write_end = ext3_journalled_write_end,
2006 .set_page_dirty = ext3_journalled_set_page_dirty,
2007 .bmap = ext3_bmap,
2008 .invalidatepage = ext3_invalidatepage,
2009 .releasepage = ext3_releasepage,
2010 .is_partially_uptodate = block_is_partially_uptodate,
2011 .error_remove_page = generic_error_remove_page,
2012};
2013
2014void ext3_set_aops(struct inode *inode)
2015{
2016 if (ext3_should_order_data(inode))
2017 inode->i_mapping->a_ops = &ext3_ordered_aops;
2018 else if (ext3_should_writeback_data(inode))
2019 inode->i_mapping->a_ops = &ext3_writeback_aops;
2020 else
2021 inode->i_mapping->a_ops = &ext3_journalled_aops;
2022}
2023
2024/*
2025 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
2026 * up to the end of the block which corresponds to `from'.
2027 * This required during truncate. We need to physically zero the tail end
2028 * of that block so it doesn't yield old data if the file is later grown.
2029 */
2030static int ext3_block_truncate_page(struct inode *inode, loff_t from)
2031{
2032 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2033 unsigned offset = from & (PAGE_CACHE_SIZE - 1);
2034 unsigned blocksize, iblock, length, pos;
2035 struct page *page;
2036 handle_t *handle = NULL;
2037 struct buffer_head *bh;
2038 int err = 0;
2039
2040 /* Truncated on block boundary - nothing to do */
2041 blocksize = inode->i_sb->s_blocksize;
2042 if ((from & (blocksize - 1)) == 0)
2043 return 0;
2044
2045 page = grab_cache_page(inode->i_mapping, index);
2046 if (!page)
2047 return -ENOMEM;
2048 length = blocksize - (offset & (blocksize - 1));
2049 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2050
2051 if (!page_has_buffers(page))
2052 create_empty_buffers(page, blocksize, 0);
2053
2054 /* Find the buffer that contains "offset" */
2055 bh = page_buffers(page);
2056 pos = blocksize;
2057 while (offset >= pos) {
2058 bh = bh->b_this_page;
2059 iblock++;
2060 pos += blocksize;
2061 }
2062
2063 err = 0;
2064 if (buffer_freed(bh)) {
2065 BUFFER_TRACE(bh, "freed: skip");
2066 goto unlock;
2067 }
2068
2069 if (!buffer_mapped(bh)) {
2070 BUFFER_TRACE(bh, "unmapped");
2071 ext3_get_block(inode, iblock, bh, 0);
2072 /* unmapped? It's a hole - nothing to do */
2073 if (!buffer_mapped(bh)) {
2074 BUFFER_TRACE(bh, "still unmapped");
2075 goto unlock;
2076 }
2077 }
2078
2079 /* Ok, it's mapped. Make sure it's up-to-date */
2080 if (PageUptodate(page))
2081 set_buffer_uptodate(bh);
2082
2083 if (!bh_uptodate_or_lock(bh)) {
2084 err = bh_submit_read(bh);
2085 /* Uhhuh. Read error. Complain and punt. */
2086 if (err)
2087 goto unlock;
2088 }
2089
2090 /* data=writeback mode doesn't need transaction to zero-out data */
2091 if (!ext3_should_writeback_data(inode)) {
2092 /* We journal at most one block */
2093 handle = ext3_journal_start(inode, 1);
2094 if (IS_ERR(handle)) {
2095 clear_highpage(page);
2096 flush_dcache_page(page);
2097 err = PTR_ERR(handle);
2098 goto unlock;
2099 }
2100 }
2101
2102 if (ext3_should_journal_data(inode)) {
2103 BUFFER_TRACE(bh, "get write access");
2104 err = ext3_journal_get_write_access(handle, bh);
2105 if (err)
2106 goto stop;
2107 }
2108
2109 zero_user(page, offset, length);
2110 BUFFER_TRACE(bh, "zeroed end of block");
2111
2112 err = 0;
2113 if (ext3_should_journal_data(inode)) {
2114 err = ext3_journal_dirty_metadata(handle, bh);
2115 } else {
2116 if (ext3_should_order_data(inode))
2117 err = ext3_journal_dirty_data(handle, bh);
2118 mark_buffer_dirty(bh);
2119 }
2120stop:
2121 if (handle)
2122 ext3_journal_stop(handle);
2123
2124unlock:
2125 unlock_page(page);
2126 page_cache_release(page);
2127 return err;
2128}
2129
2130/*
2131 * Probably it should be a library function... search for first non-zero word
2132 * or memcmp with zero_page, whatever is better for particular architecture.
2133 * Linus?
2134 */
2135static inline int all_zeroes(__le32 *p, __le32 *q)
2136{
2137 while (p < q)
2138 if (*p++)
2139 return 0;
2140 return 1;
2141}
2142
2143/**
2144 * ext3_find_shared - find the indirect blocks for partial truncation.
2145 * @inode: inode in question
2146 * @depth: depth of the affected branch
2147 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2148 * @chain: place to store the pointers to partial indirect blocks
2149 * @top: place to the (detached) top of branch
2150 *
2151 * This is a helper function used by ext3_truncate().
2152 *
2153 * When we do truncate() we may have to clean the ends of several
2154 * indirect blocks but leave the blocks themselves alive. Block is
2155 * partially truncated if some data below the new i_size is referred
2156 * from it (and it is on the path to the first completely truncated
2157 * data block, indeed). We have to free the top of that path along
2158 * with everything to the right of the path. Since no allocation
2159 * past the truncation point is possible until ext3_truncate()
2160 * finishes, we may safely do the latter, but top of branch may
2161 * require special attention - pageout below the truncation point
2162 * might try to populate it.
2163 *
2164 * We atomically detach the top of branch from the tree, store the
2165 * block number of its root in *@top, pointers to buffer_heads of
2166 * partially truncated blocks - in @chain[].bh and pointers to
2167 * their last elements that should not be removed - in
2168 * @chain[].p. Return value is the pointer to last filled element
2169 * of @chain.
2170 *
2171 * The work left to caller to do the actual freeing of subtrees:
2172 * a) free the subtree starting from *@top
2173 * b) free the subtrees whose roots are stored in
2174 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2175 * c) free the subtrees growing from the inode past the @chain[0].
2176 * (no partially truncated stuff there). */
2177
2178static Indirect *ext3_find_shared(struct inode *inode, int depth,
2179 int offsets[4], Indirect chain[4], __le32 *top)
2180{
2181 Indirect *partial, *p;
2182 int k, err;
2183
2184 *top = 0;
2185 /* Make k index the deepest non-null offset + 1 */
2186 for (k = depth; k > 1 && !offsets[k-1]; k--)
2187 ;
2188 partial = ext3_get_branch(inode, k, offsets, chain, &err);
2189 /* Writer: pointers */
2190 if (!partial)
2191 partial = chain + k-1;
2192 /*
2193 * If the branch acquired continuation since we've looked at it -
2194 * fine, it should all survive and (new) top doesn't belong to us.
2195 */
2196 if (!partial->key && *partial->p)
2197 /* Writer: end */
2198 goto no_top;
2199 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2200 ;
2201 /*
2202 * OK, we've found the last block that must survive. The rest of our
2203 * branch should be detached before unlocking. However, if that rest
2204 * of branch is all ours and does not grow immediately from the inode
2205 * it's easier to cheat and just decrement partial->p.
2206 */
2207 if (p == chain + k - 1 && p > chain) {
2208 p->p--;
2209 } else {
2210 *top = *p->p;
2211 /* Nope, don't do this in ext3. Must leave the tree intact */
2212#if 0
2213 *p->p = 0;
2214#endif
2215 }
2216 /* Writer: end */
2217
2218 while(partial > p) {
2219 brelse(partial->bh);
2220 partial--;
2221 }
2222no_top:
2223 return partial;
2224}
2225
2226/*
2227 * Zero a number of block pointers in either an inode or an indirect block.
2228 * If we restart the transaction we must again get write access to the
2229 * indirect block for further modification.
2230 *
2231 * We release `count' blocks on disk, but (last - first) may be greater
2232 * than `count' because there can be holes in there.
2233 */
2234static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
2235 struct buffer_head *bh, ext3_fsblk_t block_to_free,
2236 unsigned long count, __le32 *first, __le32 *last)
2237{
2238 __le32 *p;
2239 if (try_to_extend_transaction(handle, inode)) {
2240 if (bh) {
2241 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2242 if (ext3_journal_dirty_metadata(handle, bh))
2243 return;
2244 }
2245 ext3_mark_inode_dirty(handle, inode);
2246 truncate_restart_transaction(handle, inode);
2247 if (bh) {
2248 BUFFER_TRACE(bh, "retaking write access");
2249 if (ext3_journal_get_write_access(handle, bh))
2250 return;
2251 }
2252 }
2253
2254 /*
2255 * Any buffers which are on the journal will be in memory. We find
2256 * them on the hash table so journal_revoke() will run journal_forget()
2257 * on them. We've already detached each block from the file, so
2258 * bforget() in journal_forget() should be safe.
2259 *
2260 * AKPM: turn on bforget in journal_forget()!!!
2261 */
2262 for (p = first; p < last; p++) {
2263 u32 nr = le32_to_cpu(*p);
2264 if (nr) {
2265 struct buffer_head *bh;
2266
2267 *p = 0;
2268 bh = sb_find_get_block(inode->i_sb, nr);
2269 ext3_forget(handle, 0, inode, bh, nr);
2270 }
2271 }
2272
2273 ext3_free_blocks(handle, inode, block_to_free, count);
2274}
2275
2276/**
2277 * ext3_free_data - free a list of data blocks
2278 * @handle: handle for this transaction
2279 * @inode: inode we are dealing with
2280 * @this_bh: indirect buffer_head which contains *@first and *@last
2281 * @first: array of block numbers
2282 * @last: points immediately past the end of array
2283 *
2284 * We are freeing all blocks referred from that array (numbers are stored as
2285 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2286 *
2287 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2288 * blocks are contiguous then releasing them at one time will only affect one
2289 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2290 * actually use a lot of journal space.
2291 *
2292 * @this_bh will be %NULL if @first and @last point into the inode's direct
2293 * block pointers.
2294 */
2295static void ext3_free_data(handle_t *handle, struct inode *inode,
2296 struct buffer_head *this_bh,
2297 __le32 *first, __le32 *last)
2298{
2299 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
2300 unsigned long count = 0; /* Number of blocks in the run */
2301 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2302 corresponding to
2303 block_to_free */
2304 ext3_fsblk_t nr; /* Current block # */
2305 __le32 *p; /* Pointer into inode/ind
2306 for current block */
2307 int err;
2308
2309 if (this_bh) { /* For indirect block */
2310 BUFFER_TRACE(this_bh, "get_write_access");
2311 err = ext3_journal_get_write_access(handle, this_bh);
2312 /* Important: if we can't update the indirect pointers
2313 * to the blocks, we can't free them. */
2314 if (err)
2315 return;
2316 }
2317
2318 for (p = first; p < last; p++) {
2319 nr = le32_to_cpu(*p);
2320 if (nr) {
2321 /* accumulate blocks to free if they're contiguous */
2322 if (count == 0) {
2323 block_to_free = nr;
2324 block_to_free_p = p;
2325 count = 1;
2326 } else if (nr == block_to_free + count) {
2327 count++;
2328 } else {
2329 ext3_clear_blocks(handle, inode, this_bh,
2330 block_to_free,
2331 count, block_to_free_p, p);
2332 block_to_free = nr;
2333 block_to_free_p = p;
2334 count = 1;
2335 }
2336 }
2337 }
2338
2339 if (count > 0)
2340 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2341 count, block_to_free_p, p);
2342
2343 if (this_bh) {
2344 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
2345
2346 /*
2347 * The buffer head should have an attached journal head at this
2348 * point. However, if the data is corrupted and an indirect
2349 * block pointed to itself, it would have been detached when
2350 * the block was cleared. Check for this instead of OOPSing.
2351 */
2352 if (bh2jh(this_bh))
2353 ext3_journal_dirty_metadata(handle, this_bh);
2354 else
2355 ext3_error(inode->i_sb, "ext3_free_data",
2356 "circular indirect block detected, "
2357 "inode=%lu, block=%llu",
2358 inode->i_ino,
2359 (unsigned long long)this_bh->b_blocknr);
2360 }
2361}
2362
2363/**
2364 * ext3_free_branches - free an array of branches
2365 * @handle: JBD handle for this transaction
2366 * @inode: inode we are dealing with
2367 * @parent_bh: the buffer_head which contains *@first and *@last
2368 * @first: array of block numbers
2369 * @last: pointer immediately past the end of array
2370 * @depth: depth of the branches to free
2371 *
2372 * We are freeing all blocks referred from these branches (numbers are
2373 * stored as little-endian 32-bit) and updating @inode->i_blocks
2374 * appropriately.
2375 */
2376static void ext3_free_branches(handle_t *handle, struct inode *inode,
2377 struct buffer_head *parent_bh,
2378 __le32 *first, __le32 *last, int depth)
2379{
2380 ext3_fsblk_t nr;
2381 __le32 *p;
2382
2383 if (is_handle_aborted(handle))
2384 return;
2385
2386 if (depth--) {
2387 struct buffer_head *bh;
2388 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2389 p = last;
2390 while (--p >= first) {
2391 nr = le32_to_cpu(*p);
2392 if (!nr)
2393 continue; /* A hole */
2394
2395 /* Go read the buffer for the next level down */
2396 bh = sb_bread(inode->i_sb, nr);
2397
2398 /*
2399 * A read failure? Report error and clear slot
2400 * (should be rare).
2401 */
2402 if (!bh) {
2403 ext3_error(inode->i_sb, "ext3_free_branches",
2404 "Read failure, inode=%lu, block="E3FSBLK,
2405 inode->i_ino, nr);
2406 continue;
2407 }
2408
2409 /* This zaps the entire block. Bottom up. */
2410 BUFFER_TRACE(bh, "free child branches");
2411 ext3_free_branches(handle, inode, bh,
2412 (__le32*)bh->b_data,
2413 (__le32*)bh->b_data + addr_per_block,
2414 depth);
2415
2416 /*
2417 * Everything below this this pointer has been
2418 * released. Now let this top-of-subtree go.
2419 *
2420 * We want the freeing of this indirect block to be
2421 * atomic in the journal with the updating of the
2422 * bitmap block which owns it. So make some room in
2423 * the journal.
2424 *
2425 * We zero the parent pointer *after* freeing its
2426 * pointee in the bitmaps, so if extend_transaction()
2427 * for some reason fails to put the bitmap changes and
2428 * the release into the same transaction, recovery
2429 * will merely complain about releasing a free block,
2430 * rather than leaking blocks.
2431 */
2432 if (is_handle_aborted(handle))
2433 return;
2434 if (try_to_extend_transaction(handle, inode)) {
2435 ext3_mark_inode_dirty(handle, inode);
2436 truncate_restart_transaction(handle, inode);
2437 }
2438
2439 /*
2440 * We've probably journalled the indirect block several
2441 * times during the truncate. But it's no longer
2442 * needed and we now drop it from the transaction via
2443 * journal_revoke().
2444 *
2445 * That's easy if it's exclusively part of this
2446 * transaction. But if it's part of the committing
2447 * transaction then journal_forget() will simply
2448 * brelse() it. That means that if the underlying
2449 * block is reallocated in ext3_get_block(),
2450 * unmap_underlying_metadata() will find this block
2451 * and will try to get rid of it. damn, damn. Thus
2452 * we don't allow a block to be reallocated until
2453 * a transaction freeing it has fully committed.
2454 *
2455 * We also have to make sure journal replay after a
2456 * crash does not overwrite non-journaled data blocks
2457 * with old metadata when the block got reallocated for
2458 * data. Thus we have to store a revoke record for a
2459 * block in the same transaction in which we free the
2460 * block.
2461 */
2462 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2463
2464 ext3_free_blocks(handle, inode, nr, 1);
2465
2466 if (parent_bh) {
2467 /*
2468 * The block which we have just freed is
2469 * pointed to by an indirect block: journal it
2470 */
2471 BUFFER_TRACE(parent_bh, "get_write_access");
2472 if (!ext3_journal_get_write_access(handle,
2473 parent_bh)){
2474 *p = 0;
2475 BUFFER_TRACE(parent_bh,
2476 "call ext3_journal_dirty_metadata");
2477 ext3_journal_dirty_metadata(handle,
2478 parent_bh);
2479 }
2480 }
2481 }
2482 } else {
2483 /* We have reached the bottom of the tree. */
2484 BUFFER_TRACE(parent_bh, "free data blocks");
2485 ext3_free_data(handle, inode, parent_bh, first, last);
2486 }
2487}
2488
2489int ext3_can_truncate(struct inode *inode)
2490{
2491 if (S_ISREG(inode->i_mode))
2492 return 1;
2493 if (S_ISDIR(inode->i_mode))
2494 return 1;
2495 if (S_ISLNK(inode->i_mode))
2496 return !ext3_inode_is_fast_symlink(inode);
2497 return 0;
2498}
2499
2500/*
2501 * ext3_truncate()
2502 *
2503 * We block out ext3_get_block() block instantiations across the entire
2504 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2505 * simultaneously on behalf of the same inode.
2506 *
2507 * As we work through the truncate and commit bits of it to the journal there
2508 * is one core, guiding principle: the file's tree must always be consistent on
2509 * disk. We must be able to restart the truncate after a crash.
2510 *
2511 * The file's tree may be transiently inconsistent in memory (although it
2512 * probably isn't), but whenever we close off and commit a journal transaction,
2513 * the contents of (the filesystem + the journal) must be consistent and
2514 * restartable. It's pretty simple, really: bottom up, right to left (although
2515 * left-to-right works OK too).
2516 *
2517 * Note that at recovery time, journal replay occurs *before* the restart of
2518 * truncate against the orphan inode list.
2519 *
2520 * The committed inode has the new, desired i_size (which is the same as
2521 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2522 * that this inode's truncate did not complete and it will again call
2523 * ext3_truncate() to have another go. So there will be instantiated blocks
2524 * to the right of the truncation point in a crashed ext3 filesystem. But
2525 * that's fine - as long as they are linked from the inode, the post-crash
2526 * ext3_truncate() run will find them and release them.
2527 */
2528void ext3_truncate(struct inode *inode)
2529{
2530 handle_t *handle;
2531 struct ext3_inode_info *ei = EXT3_I(inode);
2532 __le32 *i_data = ei->i_data;
2533 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2534 int offsets[4];
2535 Indirect chain[4];
2536 Indirect *partial;
2537 __le32 nr = 0;
2538 int n;
2539 long last_block;
2540 unsigned blocksize = inode->i_sb->s_blocksize;
2541
2542 trace_ext3_truncate_enter(inode);
2543
2544 if (!ext3_can_truncate(inode))
2545 goto out_notrans;
2546
2547 if (inode->i_size == 0 && ext3_should_writeback_data(inode))
2548 ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE);
2549
2550 handle = start_transaction(inode);
2551 if (IS_ERR(handle))
2552 goto out_notrans;
2553
2554 last_block = (inode->i_size + blocksize-1)
2555 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2556 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2557 if (n == 0)
2558 goto out_stop; /* error */
2559
2560 /*
2561 * OK. This truncate is going to happen. We add the inode to the
2562 * orphan list, so that if this truncate spans multiple transactions,
2563 * and we crash, we will resume the truncate when the filesystem
2564 * recovers. It also marks the inode dirty, to catch the new size.
2565 *
2566 * Implication: the file must always be in a sane, consistent
2567 * truncatable state while each transaction commits.
2568 */
2569 if (ext3_orphan_add(handle, inode))
2570 goto out_stop;
2571
2572 /*
2573 * The orphan list entry will now protect us from any crash which
2574 * occurs before the truncate completes, so it is now safe to propagate
2575 * the new, shorter inode size (held for now in i_size) into the
2576 * on-disk inode. We do this via i_disksize, which is the value which
2577 * ext3 *really* writes onto the disk inode.
2578 */
2579 ei->i_disksize = inode->i_size;
2580
2581 /*
2582 * From here we block out all ext3_get_block() callers who want to
2583 * modify the block allocation tree.
2584 */
2585 mutex_lock(&ei->truncate_mutex);
2586
2587 if (n == 1) { /* direct blocks */
2588 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2589 i_data + EXT3_NDIR_BLOCKS);
2590 goto do_indirects;
2591 }
2592
2593 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2594 /* Kill the top of shared branch (not detached) */
2595 if (nr) {
2596 if (partial == chain) {
2597 /* Shared branch grows from the inode */
2598 ext3_free_branches(handle, inode, NULL,
2599 &nr, &nr+1, (chain+n-1) - partial);
2600 *partial->p = 0;
2601 /*
2602 * We mark the inode dirty prior to restart,
2603 * and prior to stop. No need for it here.
2604 */
2605 } else {
2606 /* Shared branch grows from an indirect block */
2607 ext3_free_branches(handle, inode, partial->bh,
2608 partial->p,
2609 partial->p+1, (chain+n-1) - partial);
2610 }
2611 }
2612 /* Clear the ends of indirect blocks on the shared branch */
2613 while (partial > chain) {
2614 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2615 (__le32*)partial->bh->b_data+addr_per_block,
2616 (chain+n-1) - partial);
2617 BUFFER_TRACE(partial->bh, "call brelse");
2618 brelse (partial->bh);
2619 partial--;
2620 }
2621do_indirects:
2622 /* Kill the remaining (whole) subtrees */
2623 switch (offsets[0]) {
2624 default:
2625 nr = i_data[EXT3_IND_BLOCK];
2626 if (nr) {
2627 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2628 i_data[EXT3_IND_BLOCK] = 0;
2629 }
2630 case EXT3_IND_BLOCK:
2631 nr = i_data[EXT3_DIND_BLOCK];
2632 if (nr) {
2633 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2634 i_data[EXT3_DIND_BLOCK] = 0;
2635 }
2636 case EXT3_DIND_BLOCK:
2637 nr = i_data[EXT3_TIND_BLOCK];
2638 if (nr) {
2639 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2640 i_data[EXT3_TIND_BLOCK] = 0;
2641 }
2642 case EXT3_TIND_BLOCK:
2643 ;
2644 }
2645
2646 ext3_discard_reservation(inode);
2647
2648 mutex_unlock(&ei->truncate_mutex);
2649 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2650 ext3_mark_inode_dirty(handle, inode);
2651
2652 /*
2653 * In a multi-transaction truncate, we only make the final transaction
2654 * synchronous
2655 */
2656 if (IS_SYNC(inode))
2657 handle->h_sync = 1;
2658out_stop:
2659 /*
2660 * If this was a simple ftruncate(), and the file will remain alive
2661 * then we need to clear up the orphan record which we created above.
2662 * However, if this was a real unlink then we were called by
2663 * ext3_evict_inode(), and we allow that function to clean up the
2664 * orphan info for us.
2665 */
2666 if (inode->i_nlink)
2667 ext3_orphan_del(handle, inode);
2668
2669 ext3_journal_stop(handle);
2670 trace_ext3_truncate_exit(inode);
2671 return;
2672out_notrans:
2673 /*
2674 * Delete the inode from orphan list so that it doesn't stay there
2675 * forever and trigger assertion on umount.
2676 */
2677 if (inode->i_nlink)
2678 ext3_orphan_del(NULL, inode);
2679 trace_ext3_truncate_exit(inode);
2680}
2681
2682static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
2683 unsigned long ino, struct ext3_iloc *iloc)
2684{
2685 unsigned long block_group;
2686 unsigned long offset;
2687 ext3_fsblk_t block;
2688 struct ext3_group_desc *gdp;
2689
2690 if (!ext3_valid_inum(sb, ino)) {
2691 /*
2692 * This error is already checked for in namei.c unless we are
2693 * looking at an NFS filehandle, in which case no error
2694 * report is needed
2695 */
2696 return 0;
2697 }
2698
2699 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2700 gdp = ext3_get_group_desc(sb, block_group, NULL);
2701 if (!gdp)
2702 return 0;
2703 /*
2704 * Figure out the offset within the block group inode table
2705 */
2706 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2707 EXT3_INODE_SIZE(sb);
2708 block = le32_to_cpu(gdp->bg_inode_table) +
2709 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2710
2711 iloc->block_group = block_group;
2712 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2713 return block;
2714}
2715
2716/*
2717 * ext3_get_inode_loc returns with an extra refcount against the inode's
2718 * underlying buffer_head on success. If 'in_mem' is true, we have all
2719 * data in memory that is needed to recreate the on-disk version of this
2720 * inode.
2721 */
2722static int __ext3_get_inode_loc(struct inode *inode,
2723 struct ext3_iloc *iloc, int in_mem)
2724{
2725 ext3_fsblk_t block;
2726 struct buffer_head *bh;
2727
2728 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2729 if (!block)
2730 return -EIO;
2731
2732 bh = sb_getblk(inode->i_sb, block);
2733 if (!bh) {
2734 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2735 "unable to read inode block - "
2736 "inode=%lu, block="E3FSBLK,
2737 inode->i_ino, block);
2738 return -EIO;
2739 }
2740 if (!buffer_uptodate(bh)) {
2741 lock_buffer(bh);
2742
2743 /*
2744 * If the buffer has the write error flag, we have failed
2745 * to write out another inode in the same block. In this
2746 * case, we don't have to read the block because we may
2747 * read the old inode data successfully.
2748 */
2749 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2750 set_buffer_uptodate(bh);
2751
2752 if (buffer_uptodate(bh)) {
2753 /* someone brought it uptodate while we waited */
2754 unlock_buffer(bh);
2755 goto has_buffer;
2756 }
2757
2758 /*
2759 * If we have all information of the inode in memory and this
2760 * is the only valid inode in the block, we need not read the
2761 * block.
2762 */
2763 if (in_mem) {
2764 struct buffer_head *bitmap_bh;
2765 struct ext3_group_desc *desc;
2766 int inodes_per_buffer;
2767 int inode_offset, i;
2768 int block_group;
2769 int start;
2770
2771 block_group = (inode->i_ino - 1) /
2772 EXT3_INODES_PER_GROUP(inode->i_sb);
2773 inodes_per_buffer = bh->b_size /
2774 EXT3_INODE_SIZE(inode->i_sb);
2775 inode_offset = ((inode->i_ino - 1) %
2776 EXT3_INODES_PER_GROUP(inode->i_sb));
2777 start = inode_offset & ~(inodes_per_buffer - 1);
2778
2779 /* Is the inode bitmap in cache? */
2780 desc = ext3_get_group_desc(inode->i_sb,
2781 block_group, NULL);
2782 if (!desc)
2783 goto make_io;
2784
2785 bitmap_bh = sb_getblk(inode->i_sb,
2786 le32_to_cpu(desc->bg_inode_bitmap));
2787 if (!bitmap_bh)
2788 goto make_io;
2789
2790 /*
2791 * If the inode bitmap isn't in cache then the
2792 * optimisation may end up performing two reads instead
2793 * of one, so skip it.
2794 */
2795 if (!buffer_uptodate(bitmap_bh)) {
2796 brelse(bitmap_bh);
2797 goto make_io;
2798 }
2799 for (i = start; i < start + inodes_per_buffer; i++) {
2800 if (i == inode_offset)
2801 continue;
2802 if (ext3_test_bit(i, bitmap_bh->b_data))
2803 break;
2804 }
2805 brelse(bitmap_bh);
2806 if (i == start + inodes_per_buffer) {
2807 /* all other inodes are free, so skip I/O */
2808 memset(bh->b_data, 0, bh->b_size);
2809 set_buffer_uptodate(bh);
2810 unlock_buffer(bh);
2811 goto has_buffer;
2812 }
2813 }
2814
2815make_io:
2816 /*
2817 * There are other valid inodes in the buffer, this inode
2818 * has in-inode xattrs, or we don't have this inode in memory.
2819 * Read the block from disk.
2820 */
2821 trace_ext3_load_inode(inode);
2822 get_bh(bh);
2823 bh->b_end_io = end_buffer_read_sync;
2824 submit_bh(READ | REQ_META | REQ_PRIO, bh);
2825 wait_on_buffer(bh);
2826 if (!buffer_uptodate(bh)) {
2827 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2828 "unable to read inode block - "
2829 "inode=%lu, block="E3FSBLK,
2830 inode->i_ino, block);
2831 brelse(bh);
2832 return -EIO;
2833 }
2834 }
2835has_buffer:
2836 iloc->bh = bh;
2837 return 0;
2838}
2839
2840int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2841{
2842 /* We have all inode data except xattrs in memory here. */
2843 return __ext3_get_inode_loc(inode, iloc,
2844 !ext3_test_inode_state(inode, EXT3_STATE_XATTR));
2845}
2846
2847void ext3_set_inode_flags(struct inode *inode)
2848{
2849 unsigned int flags = EXT3_I(inode)->i_flags;
2850
2851 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2852 if (flags & EXT3_SYNC_FL)
2853 inode->i_flags |= S_SYNC;
2854 if (flags & EXT3_APPEND_FL)
2855 inode->i_flags |= S_APPEND;
2856 if (flags & EXT3_IMMUTABLE_FL)
2857 inode->i_flags |= S_IMMUTABLE;
2858 if (flags & EXT3_NOATIME_FL)
2859 inode->i_flags |= S_NOATIME;
2860 if (flags & EXT3_DIRSYNC_FL)
2861 inode->i_flags |= S_DIRSYNC;
2862}
2863
2864/* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2865void ext3_get_inode_flags(struct ext3_inode_info *ei)
2866{
2867 unsigned int flags = ei->vfs_inode.i_flags;
2868
2869 ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2870 EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2871 if (flags & S_SYNC)
2872 ei->i_flags |= EXT3_SYNC_FL;
2873 if (flags & S_APPEND)
2874 ei->i_flags |= EXT3_APPEND_FL;
2875 if (flags & S_IMMUTABLE)
2876 ei->i_flags |= EXT3_IMMUTABLE_FL;
2877 if (flags & S_NOATIME)
2878 ei->i_flags |= EXT3_NOATIME_FL;
2879 if (flags & S_DIRSYNC)
2880 ei->i_flags |= EXT3_DIRSYNC_FL;
2881}
2882
2883struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
2884{
2885 struct ext3_iloc iloc;
2886 struct ext3_inode *raw_inode;
2887 struct ext3_inode_info *ei;
2888 struct buffer_head *bh;
2889 struct inode *inode;
2890 journal_t *journal = EXT3_SB(sb)->s_journal;
2891 transaction_t *transaction;
2892 long ret;
2893 int block;
2894 uid_t i_uid;
2895 gid_t i_gid;
2896
2897 inode = iget_locked(sb, ino);
2898 if (!inode)
2899 return ERR_PTR(-ENOMEM);
2900 if (!(inode->i_state & I_NEW))
2901 return inode;
2902
2903 ei = EXT3_I(inode);
2904 ei->i_block_alloc_info = NULL;
2905
2906 ret = __ext3_get_inode_loc(inode, &iloc, 0);
2907 if (ret < 0)
2908 goto bad_inode;
2909 bh = iloc.bh;
2910 raw_inode = ext3_raw_inode(&iloc);
2911 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2912 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2913 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2914 if(!(test_opt (inode->i_sb, NO_UID32))) {
2915 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2916 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2917 }
2918 i_uid_write(inode, i_uid);
2919 i_gid_write(inode, i_gid);
2920 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
2921 inode->i_size = le32_to_cpu(raw_inode->i_size);
2922 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2923 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2924 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
2925 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2926
2927 ei->i_state_flags = 0;
2928 ei->i_dir_start_lookup = 0;
2929 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2930 /* We now have enough fields to check if the inode was active or not.
2931 * This is needed because nfsd might try to access dead inodes
2932 * the test is that same one that e2fsck uses
2933 * NeilBrown 1999oct15
2934 */
2935 if (inode->i_nlink == 0) {
2936 if (inode->i_mode == 0 ||
2937 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2938 /* this inode is deleted */
2939 brelse (bh);
2940 ret = -ESTALE;
2941 goto bad_inode;
2942 }
2943 /* The only unlinked inodes we let through here have
2944 * valid i_mode and are being read by the orphan
2945 * recovery code: that's fine, we're about to complete
2946 * the process of deleting those. */
2947 }
2948 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2949 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2950#ifdef EXT3_FRAGMENTS
2951 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2952 ei->i_frag_no = raw_inode->i_frag;
2953 ei->i_frag_size = raw_inode->i_fsize;
2954#endif
2955 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2956 if (!S_ISREG(inode->i_mode)) {
2957 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2958 } else {
2959 inode->i_size |=
2960 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2961 }
2962 ei->i_disksize = inode->i_size;
2963 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2964 ei->i_block_group = iloc.block_group;
2965 /*
2966 * NOTE! The in-memory inode i_data array is in little-endian order
2967 * even on big-endian machines: we do NOT byteswap the block numbers!
2968 */
2969 for (block = 0; block < EXT3_N_BLOCKS; block++)
2970 ei->i_data[block] = raw_inode->i_block[block];
2971 INIT_LIST_HEAD(&ei->i_orphan);
2972
2973 /*
2974 * Set transaction id's of transactions that have to be committed
2975 * to finish f[data]sync. We set them to currently running transaction
2976 * as we cannot be sure that the inode or some of its metadata isn't
2977 * part of the transaction - the inode could have been reclaimed and
2978 * now it is reread from disk.
2979 */
2980 if (journal) {
2981 tid_t tid;
2982
2983 spin_lock(&journal->j_state_lock);
2984 if (journal->j_running_transaction)
2985 transaction = journal->j_running_transaction;
2986 else
2987 transaction = journal->j_committing_transaction;
2988 if (transaction)
2989 tid = transaction->t_tid;
2990 else
2991 tid = journal->j_commit_sequence;
2992 spin_unlock(&journal->j_state_lock);
2993 atomic_set(&ei->i_sync_tid, tid);
2994 atomic_set(&ei->i_datasync_tid, tid);
2995 }
2996
2997 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2998 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2999 /*
3000 * When mke2fs creates big inodes it does not zero out
3001 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
3002 * so ignore those first few inodes.
3003 */
3004 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3005 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3006 EXT3_INODE_SIZE(inode->i_sb)) {
3007 brelse (bh);
3008 ret = -EIO;
3009 goto bad_inode;
3010 }
3011 if (ei->i_extra_isize == 0) {
3012 /* The extra space is currently unused. Use it. */
3013 ei->i_extra_isize = sizeof(struct ext3_inode) -
3014 EXT3_GOOD_OLD_INODE_SIZE;
3015 } else {
3016 __le32 *magic = (void *)raw_inode +
3017 EXT3_GOOD_OLD_INODE_SIZE +
3018 ei->i_extra_isize;
3019 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
3020 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
3021 }
3022 } else
3023 ei->i_extra_isize = 0;
3024
3025 if (S_ISREG(inode->i_mode)) {
3026 inode->i_op = &ext3_file_inode_operations;
3027 inode->i_fop = &ext3_file_operations;
3028 ext3_set_aops(inode);
3029 } else if (S_ISDIR(inode->i_mode)) {
3030 inode->i_op = &ext3_dir_inode_operations;
3031 inode->i_fop = &ext3_dir_operations;
3032 } else if (S_ISLNK(inode->i_mode)) {
3033 if (ext3_inode_is_fast_symlink(inode)) {
3034 inode->i_op = &ext3_fast_symlink_inode_operations;
3035 nd_terminate_link(ei->i_data, inode->i_size,
3036 sizeof(ei->i_data) - 1);
3037 } else {
3038 inode->i_op = &ext3_symlink_inode_operations;
3039 ext3_set_aops(inode);
3040 }
3041 } else {
3042 inode->i_op = &ext3_special_inode_operations;
3043 if (raw_inode->i_block[0])
3044 init_special_inode(inode, inode->i_mode,
3045 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3046 else
3047 init_special_inode(inode, inode->i_mode,
3048 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3049 }
3050 brelse (iloc.bh);
3051 ext3_set_inode_flags(inode);
3052 unlock_new_inode(inode);
3053 return inode;
3054
3055bad_inode:
3056 iget_failed(inode);
3057 return ERR_PTR(ret);
3058}
3059
3060/*
3061 * Post the struct inode info into an on-disk inode location in the
3062 * buffer-cache. This gobbles the caller's reference to the
3063 * buffer_head in the inode location struct.
3064 *
3065 * The caller must have write access to iloc->bh.
3066 */
3067static int ext3_do_update_inode(handle_t *handle,
3068 struct inode *inode,
3069 struct ext3_iloc *iloc)
3070{
3071 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
3072 struct ext3_inode_info *ei = EXT3_I(inode);
3073 struct buffer_head *bh = iloc->bh;
3074 int err = 0, rc, block;
3075 int need_datasync = 0;
3076 __le32 disksize;
3077 uid_t i_uid;
3078 gid_t i_gid;
3079
3080again:
3081 /* we can't allow multiple procs in here at once, its a bit racey */
3082 lock_buffer(bh);
3083
3084 /* For fields not not tracking in the in-memory inode,
3085 * initialise them to zero for new inodes. */
3086 if (ext3_test_inode_state(inode, EXT3_STATE_NEW))
3087 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
3088
3089 ext3_get_inode_flags(ei);
3090 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3091 i_uid = i_uid_read(inode);
3092 i_gid = i_gid_read(inode);
3093 if(!(test_opt(inode->i_sb, NO_UID32))) {
3094 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
3095 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
3096/*
3097 * Fix up interoperability with old kernels. Otherwise, old inodes get
3098 * re-used with the upper 16 bits of the uid/gid intact
3099 */
3100 if(!ei->i_dtime) {
3101 raw_inode->i_uid_high =
3102 cpu_to_le16(high_16_bits(i_uid));
3103 raw_inode->i_gid_high =
3104 cpu_to_le16(high_16_bits(i_gid));
3105 } else {
3106 raw_inode->i_uid_high = 0;
3107 raw_inode->i_gid_high = 0;
3108 }
3109 } else {
3110 raw_inode->i_uid_low =
3111 cpu_to_le16(fs_high2lowuid(i_uid));
3112 raw_inode->i_gid_low =
3113 cpu_to_le16(fs_high2lowgid(i_gid));
3114 raw_inode->i_uid_high = 0;
3115 raw_inode->i_gid_high = 0;
3116 }
3117 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3118 disksize = cpu_to_le32(ei->i_disksize);
3119 if (disksize != raw_inode->i_size) {
3120 need_datasync = 1;
3121 raw_inode->i_size = disksize;
3122 }
3123 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
3124 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
3125 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
3126 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
3127 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3128 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
3129#ifdef EXT3_FRAGMENTS
3130 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
3131 raw_inode->i_frag = ei->i_frag_no;
3132 raw_inode->i_fsize = ei->i_frag_size;
3133#endif
3134 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
3135 if (!S_ISREG(inode->i_mode)) {
3136 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
3137 } else {
3138 disksize = cpu_to_le32(ei->i_disksize >> 32);
3139 if (disksize != raw_inode->i_size_high) {
3140 raw_inode->i_size_high = disksize;
3141 need_datasync = 1;
3142 }
3143 if (ei->i_disksize > 0x7fffffffULL) {
3144 struct super_block *sb = inode->i_sb;
3145 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
3146 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
3147 EXT3_SB(sb)->s_es->s_rev_level ==
3148 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
3149 /* If this is the first large file
3150 * created, add a flag to the superblock.
3151 */
3152 unlock_buffer(bh);
3153 err = ext3_journal_get_write_access(handle,
3154 EXT3_SB(sb)->s_sbh);
3155 if (err)
3156 goto out_brelse;
3157
3158 ext3_update_dynamic_rev(sb);
3159 EXT3_SET_RO_COMPAT_FEATURE(sb,
3160 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
3161 handle->h_sync = 1;
3162 err = ext3_journal_dirty_metadata(handle,
3163 EXT3_SB(sb)->s_sbh);
3164 /* get our lock and start over */
3165 goto again;
3166 }
3167 }
3168 }
3169 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3170 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3171 if (old_valid_dev(inode->i_rdev)) {
3172 raw_inode->i_block[0] =
3173 cpu_to_le32(old_encode_dev(inode->i_rdev));
3174 raw_inode->i_block[1] = 0;
3175 } else {
3176 raw_inode->i_block[0] = 0;
3177 raw_inode->i_block[1] =
3178 cpu_to_le32(new_encode_dev(inode->i_rdev));
3179 raw_inode->i_block[2] = 0;
3180 }
3181 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
3182 raw_inode->i_block[block] = ei->i_data[block];
3183
3184 if (ei->i_extra_isize)
3185 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3186
3187 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
3188 unlock_buffer(bh);
3189 rc = ext3_journal_dirty_metadata(handle, bh);
3190 if (!err)
3191 err = rc;
3192 ext3_clear_inode_state(inode, EXT3_STATE_NEW);
3193
3194 atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
3195 if (need_datasync)
3196 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
3197out_brelse:
3198 brelse (bh);
3199 ext3_std_error(inode->i_sb, err);
3200 return err;
3201}
3202
3203/*
3204 * ext3_write_inode()
3205 *
3206 * We are called from a few places:
3207 *
3208 * - Within generic_file_write() for O_SYNC files.
3209 * Here, there will be no transaction running. We wait for any running
3210 * trasnaction to commit.
3211 *
3212 * - Within sys_sync(), kupdate and such.
3213 * We wait on commit, if tol to.
3214 *
3215 * - Within prune_icache() (PF_MEMALLOC == true)
3216 * Here we simply return. We can't afford to block kswapd on the
3217 * journal commit.
3218 *
3219 * In all cases it is actually safe for us to return without doing anything,
3220 * because the inode has been copied into a raw inode buffer in
3221 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3222 * knfsd.
3223 *
3224 * Note that we are absolutely dependent upon all inode dirtiers doing the
3225 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3226 * which we are interested.
3227 *
3228 * It would be a bug for them to not do this. The code:
3229 *
3230 * mark_inode_dirty(inode)
3231 * stuff();
3232 * inode->i_size = expr;
3233 *
3234 * is in error because a kswapd-driven write_inode() could occur while
3235 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3236 * will no longer be on the superblock's dirty inode list.
3237 */
3238int ext3_write_inode(struct inode *inode, struct writeback_control *wbc)
3239{
3240 if (current->flags & PF_MEMALLOC)
3241 return 0;
3242
3243 if (ext3_journal_current_handle()) {
3244 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3245 dump_stack();
3246 return -EIO;
3247 }
3248
3249 if (wbc->sync_mode != WB_SYNC_ALL)
3250 return 0;
3251
3252 return ext3_force_commit(inode->i_sb);
3253}
3254
3255/*
3256 * ext3_setattr()
3257 *
3258 * Called from notify_change.
3259 *
3260 * We want to trap VFS attempts to truncate the file as soon as
3261 * possible. In particular, we want to make sure that when the VFS
3262 * shrinks i_size, we put the inode on the orphan list and modify
3263 * i_disksize immediately, so that during the subsequent flushing of
3264 * dirty pages and freeing of disk blocks, we can guarantee that any
3265 * commit will leave the blocks being flushed in an unused state on
3266 * disk. (On recovery, the inode will get truncated and the blocks will
3267 * be freed, so we have a strong guarantee that no future commit will
3268 * leave these blocks visible to the user.)
3269 *
3270 * Called with inode->sem down.
3271 */
3272int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3273{
3274 struct inode *inode = dentry->d_inode;
3275 int error, rc = 0;
3276 const unsigned int ia_valid = attr->ia_valid;
3277
3278 error = inode_change_ok(inode, attr);
3279 if (error)
3280 return error;
3281
3282 if (is_quota_modification(inode, attr))
3283 dquot_initialize(inode);
3284 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3285 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
3286 handle_t *handle;
3287
3288 /* (user+group)*(old+new) structure, inode write (sb,
3289 * inode block, ? - but truncate inode update has it) */
3290 handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3291 EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3);
3292 if (IS_ERR(handle)) {
3293 error = PTR_ERR(handle);
3294 goto err_out;
3295 }
3296 error = dquot_transfer(inode, attr);
3297 if (error) {
3298 ext3_journal_stop(handle);
3299 return error;
3300 }
3301 /* Update corresponding info in inode so that everything is in
3302 * one transaction */
3303 if (attr->ia_valid & ATTR_UID)
3304 inode->i_uid = attr->ia_uid;
3305 if (attr->ia_valid & ATTR_GID)
3306 inode->i_gid = attr->ia_gid;
3307 error = ext3_mark_inode_dirty(handle, inode);
3308 ext3_journal_stop(handle);
3309 }
3310
3311 if (attr->ia_valid & ATTR_SIZE)
3312 inode_dio_wait(inode);
3313
3314 if (S_ISREG(inode->i_mode) &&
3315 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3316 handle_t *handle;
3317
3318 handle = ext3_journal_start(inode, 3);
3319 if (IS_ERR(handle)) {
3320 error = PTR_ERR(handle);
3321 goto err_out;
3322 }
3323
3324 error = ext3_orphan_add(handle, inode);
3325 if (error) {
3326 ext3_journal_stop(handle);
3327 goto err_out;
3328 }
3329 EXT3_I(inode)->i_disksize = attr->ia_size;
3330 error = ext3_mark_inode_dirty(handle, inode);
3331 ext3_journal_stop(handle);
3332 if (error) {
3333 /* Some hard fs error must have happened. Bail out. */
3334 ext3_orphan_del(NULL, inode);
3335 goto err_out;
3336 }
3337 rc = ext3_block_truncate_page(inode, attr->ia_size);
3338 if (rc) {
3339 /* Cleanup orphan list and exit */
3340 handle = ext3_journal_start(inode, 3);
3341 if (IS_ERR(handle)) {
3342 ext3_orphan_del(NULL, inode);
3343 goto err_out;
3344 }
3345 ext3_orphan_del(handle, inode);
3346 ext3_journal_stop(handle);
3347 goto err_out;
3348 }
3349 }
3350
3351 if ((attr->ia_valid & ATTR_SIZE) &&
3352 attr->ia_size != i_size_read(inode)) {
3353 truncate_setsize(inode, attr->ia_size);
3354 ext3_truncate(inode);
3355 }
3356
3357 setattr_copy(inode, attr);
3358 mark_inode_dirty(inode);
3359
3360 if (ia_valid & ATTR_MODE)
3361 rc = ext3_acl_chmod(inode);
3362
3363err_out:
3364 ext3_std_error(inode->i_sb, error);
3365 if (!error)
3366 error = rc;
3367 return error;
3368}
3369
3370
3371/*
3372 * How many blocks doth make a writepage()?
3373 *
3374 * With N blocks per page, it may be:
3375 * N data blocks
3376 * 2 indirect block
3377 * 2 dindirect
3378 * 1 tindirect
3379 * N+5 bitmap blocks (from the above)
3380 * N+5 group descriptor summary blocks
3381 * 1 inode block
3382 * 1 superblock.
3383 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3384 *
3385 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3386 *
3387 * With ordered or writeback data it's the same, less the N data blocks.
3388 *
3389 * If the inode's direct blocks can hold an integral number of pages then a
3390 * page cannot straddle two indirect blocks, and we can only touch one indirect
3391 * and dindirect block, and the "5" above becomes "3".
3392 *
3393 * This still overestimates under most circumstances. If we were to pass the
3394 * start and end offsets in here as well we could do block_to_path() on each
3395 * block and work out the exact number of indirects which are touched. Pah.
3396 */
3397
3398static int ext3_writepage_trans_blocks(struct inode *inode)
3399{
3400 int bpp = ext3_journal_blocks_per_page(inode);
3401 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3402 int ret;
3403
3404 if (ext3_should_journal_data(inode))
3405 ret = 3 * (bpp + indirects) + 2;
3406 else
3407 ret = 2 * (bpp + indirects) + indirects + 2;
3408
3409#ifdef CONFIG_QUOTA
3410 /* We know that structure was already allocated during dquot_initialize so
3411 * we will be updating only the data blocks + inodes */
3412 ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
3413#endif
3414
3415 return ret;
3416}
3417
3418/*
3419 * The caller must have previously called ext3_reserve_inode_write().
3420 * Give this, we know that the caller already has write access to iloc->bh.
3421 */
3422int ext3_mark_iloc_dirty(handle_t *handle,
3423 struct inode *inode, struct ext3_iloc *iloc)
3424{
3425 int err = 0;
3426
3427 /* the do_update_inode consumes one bh->b_count */
3428 get_bh(iloc->bh);
3429
3430 /* ext3_do_update_inode() does journal_dirty_metadata */
3431 err = ext3_do_update_inode(handle, inode, iloc);
3432 put_bh(iloc->bh);
3433 return err;
3434}
3435
3436/*
3437 * On success, We end up with an outstanding reference count against
3438 * iloc->bh. This _must_ be cleaned up later.
3439 */
3440
3441int
3442ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
3443 struct ext3_iloc *iloc)
3444{
3445 int err = 0;
3446 if (handle) {
3447 err = ext3_get_inode_loc(inode, iloc);
3448 if (!err) {
3449 BUFFER_TRACE(iloc->bh, "get_write_access");
3450 err = ext3_journal_get_write_access(handle, iloc->bh);
3451 if (err) {
3452 brelse(iloc->bh);
3453 iloc->bh = NULL;
3454 }
3455 }
3456 }
3457 ext3_std_error(inode->i_sb, err);
3458 return err;
3459}
3460
3461/*
3462 * What we do here is to mark the in-core inode as clean with respect to inode
3463 * dirtiness (it may still be data-dirty).
3464 * This means that the in-core inode may be reaped by prune_icache
3465 * without having to perform any I/O. This is a very good thing,
3466 * because *any* task may call prune_icache - even ones which
3467 * have a transaction open against a different journal.
3468 *
3469 * Is this cheating? Not really. Sure, we haven't written the
3470 * inode out, but prune_icache isn't a user-visible syncing function.
3471 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3472 * we start and wait on commits.
3473 *
3474 * Is this efficient/effective? Well, we're being nice to the system
3475 * by cleaning up our inodes proactively so they can be reaped
3476 * without I/O. But we are potentially leaving up to five seconds'
3477 * worth of inodes floating about which prune_icache wants us to
3478 * write out. One way to fix that would be to get prune_icache()
3479 * to do a write_super() to free up some memory. It has the desired
3480 * effect.
3481 */
3482int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3483{
3484 struct ext3_iloc iloc;
3485 int err;
3486
3487 might_sleep();
3488 trace_ext3_mark_inode_dirty(inode, _RET_IP_);
3489 err = ext3_reserve_inode_write(handle, inode, &iloc);
3490 if (!err)
3491 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3492 return err;
3493}
3494
3495/*
3496 * ext3_dirty_inode() is called from __mark_inode_dirty()
3497 *
3498 * We're really interested in the case where a file is being extended.
3499 * i_size has been changed by generic_commit_write() and we thus need
3500 * to include the updated inode in the current transaction.
3501 *
3502 * Also, dquot_alloc_space() will always dirty the inode when blocks
3503 * are allocated to the file.
3504 *
3505 * If the inode is marked synchronous, we don't honour that here - doing
3506 * so would cause a commit on atime updates, which we don't bother doing.
3507 * We handle synchronous inodes at the highest possible level.
3508 */
3509void ext3_dirty_inode(struct inode *inode, int flags)
3510{
3511 handle_t *current_handle = ext3_journal_current_handle();
3512 handle_t *handle;
3513
3514 handle = ext3_journal_start(inode, 2);
3515 if (IS_ERR(handle))
3516 goto out;
3517 if (current_handle &&
3518 current_handle->h_transaction != handle->h_transaction) {
3519 /* This task has a transaction open against a different fs */
3520 printk(KERN_EMERG "%s: transactions do not match!\n",
3521 __func__);
3522 } else {
3523 jbd_debug(5, "marking dirty. outer handle=%p\n",
3524 current_handle);
3525 ext3_mark_inode_dirty(handle, inode);
3526 }
3527 ext3_journal_stop(handle);
3528out:
3529 return;
3530}
3531
3532#if 0
3533/*
3534 * Bind an inode's backing buffer_head into this transaction, to prevent
3535 * it from being flushed to disk early. Unlike
3536 * ext3_reserve_inode_write, this leaves behind no bh reference and
3537 * returns no iloc structure, so the caller needs to repeat the iloc
3538 * lookup to mark the inode dirty later.
3539 */
3540static int ext3_pin_inode(handle_t *handle, struct inode *inode)
3541{
3542 struct ext3_iloc iloc;
3543
3544 int err = 0;
3545 if (handle) {
3546 err = ext3_get_inode_loc(inode, &iloc);
3547 if (!err) {
3548 BUFFER_TRACE(iloc.bh, "get_write_access");
3549 err = journal_get_write_access(handle, iloc.bh);
3550 if (!err)
3551 err = ext3_journal_dirty_metadata(handle,
3552 iloc.bh);
3553 brelse(iloc.bh);
3554 }
3555 }
3556 ext3_std_error(inode->i_sb, err);
3557 return err;
3558}
3559#endif
3560
3561int ext3_change_inode_journal_flag(struct inode *inode, int val)
3562{
3563 journal_t *journal;
3564 handle_t *handle;
3565 int err;
3566
3567 /*
3568 * We have to be very careful here: changing a data block's
3569 * journaling status dynamically is dangerous. If we write a
3570 * data block to the journal, change the status and then delete
3571 * that block, we risk forgetting to revoke the old log record
3572 * from the journal and so a subsequent replay can corrupt data.
3573 * So, first we make sure that the journal is empty and that
3574 * nobody is changing anything.
3575 */
3576
3577 journal = EXT3_JOURNAL(inode);
3578 if (is_journal_aborted(journal))
3579 return -EROFS;
3580
3581 journal_lock_updates(journal);
3582 journal_flush(journal);
3583
3584 /*
3585 * OK, there are no updates running now, and all cached data is
3586 * synced to disk. We are now in a completely consistent state
3587 * which doesn't have anything in the journal, and we know that
3588 * no filesystem updates are running, so it is safe to modify
3589 * the inode's in-core data-journaling state flag now.
3590 */
3591
3592 if (val)
3593 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3594 else
3595 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3596 ext3_set_aops(inode);
3597
3598 journal_unlock_updates(journal);
3599
3600 /* Finally we can mark the inode as dirty. */
3601
3602 handle = ext3_journal_start(inode, 1);
3603 if (IS_ERR(handle))
3604 return PTR_ERR(handle);
3605
3606 err = ext3_mark_inode_dirty(handle, inode);
3607 handle->h_sync = 1;
3608 ext3_journal_stop(handle);
3609 ext3_std_error(inode->i_sb, err);
3610
3611 return err;
3612}