blob: b505a145a5937a426184d5c289b6825ee62d0e17 [file] [log] [blame]
Amir Goldsteindae1e522011-06-27 19:40:50 -04001/*
2 * linux/fs/ext4/indirect.c
3 *
4 * from
5 *
6 * linux/fs/ext4/inode.c
7 *
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
12 *
13 * from
14 *
15 * linux/fs/minix/inode.c
16 *
17 * Copyright (C) 1991, 1992 Linus Torvalds
18 *
19 * Goal-directed block allocation by Stephen Tweedie
20 * (sct@redhat.com), 1993, 1998
21 */
22
Amir Goldsteindae1e522011-06-27 19:40:50 -040023#include "ext4_jbd2.h"
24#include "truncate.h"
Theodore Ts'o4a092d72012-11-28 13:03:30 -050025#include "ext4_extents.h" /* Needed for EXT_MAX_BLOCKS */
Amir Goldsteindae1e522011-06-27 19:40:50 -040026
27#include <trace/events/ext4.h>
28
29typedef struct {
30 __le32 *p;
31 __le32 key;
32 struct buffer_head *bh;
33} Indirect;
34
35static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
36{
37 p->key = *(p->p = v);
38 p->bh = bh;
39}
40
41/**
42 * ext4_block_to_path - parse the block number into array of offsets
43 * @inode: inode in question (we are only interested in its superblock)
44 * @i_block: block number to be parsed
45 * @offsets: array to store the offsets in
46 * @boundary: set this non-zero if the referred-to block is likely to be
47 * followed (on disk) by an indirect block.
48 *
49 * To store the locations of file's data ext4 uses a data structure common
50 * for UNIX filesystems - tree of pointers anchored in the inode, with
51 * data blocks at leaves and indirect blocks in intermediate nodes.
52 * This function translates the block number into path in that tree -
53 * return value is the path length and @offsets[n] is the offset of
54 * pointer to (n+1)th node in the nth one. If @block is out of range
55 * (negative or too large) warning is printed and zero returned.
56 *
57 * Note: function doesn't find node addresses, so no IO is needed. All
58 * we need to know is the capacity of indirect blocks (taken from the
59 * inode->i_sb).
60 */
61
62/*
63 * Portability note: the last comparison (check that we fit into triple
64 * indirect block) is spelled differently, because otherwise on an
65 * architecture with 32-bit longs and 8Kb pages we might get into trouble
66 * if our filesystem had 8Kb blocks. We might use long long, but that would
67 * kill us on x86. Oh, well, at least the sign propagation does not matter -
68 * i_block would have to be negative in the very beginning, so we would not
69 * get there at all.
70 */
71
72static int ext4_block_to_path(struct inode *inode,
73 ext4_lblk_t i_block,
74 ext4_lblk_t offsets[4], int *boundary)
75{
76 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
77 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
78 const long direct_blocks = EXT4_NDIR_BLOCKS,
79 indirect_blocks = ptrs,
80 double_blocks = (1 << (ptrs_bits * 2));
81 int n = 0;
82 int final = 0;
83
84 if (i_block < direct_blocks) {
85 offsets[n++] = i_block;
86 final = direct_blocks;
87 } else if ((i_block -= direct_blocks) < indirect_blocks) {
88 offsets[n++] = EXT4_IND_BLOCK;
89 offsets[n++] = i_block;
90 final = ptrs;
91 } else if ((i_block -= indirect_blocks) < double_blocks) {
92 offsets[n++] = EXT4_DIND_BLOCK;
93 offsets[n++] = i_block >> ptrs_bits;
94 offsets[n++] = i_block & (ptrs - 1);
95 final = ptrs;
96 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
97 offsets[n++] = EXT4_TIND_BLOCK;
98 offsets[n++] = i_block >> (ptrs_bits * 2);
99 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
100 offsets[n++] = i_block & (ptrs - 1);
101 final = ptrs;
102 } else {
103 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
104 i_block + direct_blocks +
105 indirect_blocks + double_blocks, inode->i_ino);
106 }
107 if (boundary)
108 *boundary = final - 1 - (i_block & (ptrs - 1));
109 return n;
110}
111
112/**
113 * ext4_get_branch - read the chain of indirect blocks leading to data
114 * @inode: inode in question
115 * @depth: depth of the chain (1 - direct pointer, etc.)
116 * @offsets: offsets of pointers in inode/indirect blocks
117 * @chain: place to store the result
118 * @err: here we store the error value
119 *
120 * Function fills the array of triples <key, p, bh> and returns %NULL
121 * if everything went OK or the pointer to the last filled triple
122 * (incomplete one) otherwise. Upon the return chain[i].key contains
123 * the number of (i+1)-th block in the chain (as it is stored in memory,
124 * i.e. little-endian 32-bit), chain[i].p contains the address of that
125 * number (it points into struct inode for i==0 and into the bh->b_data
126 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
127 * block for i>0 and NULL for i==0. In other words, it holds the block
128 * numbers of the chain, addresses they were taken from (and where we can
129 * verify that chain did not change) and buffer_heads hosting these
130 * numbers.
131 *
132 * Function stops when it stumbles upon zero pointer (absent block)
133 * (pointer to last triple returned, *@err == 0)
134 * or when it gets an IO error reading an indirect block
135 * (ditto, *@err == -EIO)
136 * or when it reads all @depth-1 indirect blocks successfully and finds
137 * the whole chain, all way to the data (returns %NULL, *err == 0).
138 *
139 * Need to be called with
140 * down_read(&EXT4_I(inode)->i_data_sem)
141 */
142static Indirect *ext4_get_branch(struct inode *inode, int depth,
143 ext4_lblk_t *offsets,
144 Indirect chain[4], int *err)
145{
146 struct super_block *sb = inode->i_sb;
147 Indirect *p = chain;
148 struct buffer_head *bh;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500149 int ret = -EIO;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400150
151 *err = 0;
152 /* i_data is not going away, no lock needed */
153 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
154 if (!p->key)
155 goto no_block;
156 while (--depth) {
157 bh = sb_getblk(sb, le32_to_cpu(p->key));
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500158 if (unlikely(!bh)) {
159 ret = -ENOMEM;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400160 goto failure;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500161 }
Amir Goldsteindae1e522011-06-27 19:40:50 -0400162
163 if (!bh_uptodate_or_lock(bh)) {
164 if (bh_submit_read(bh) < 0) {
165 put_bh(bh);
166 goto failure;
167 }
168 /* validate block references */
169 if (ext4_check_indirect_blockref(inode, bh)) {
170 put_bh(bh);
171 goto failure;
172 }
173 }
174
175 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
176 /* Reader: end */
177 if (!p->key)
178 goto no_block;
179 }
180 return NULL;
181
182failure:
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500183 *err = ret;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400184no_block:
185 return p;
186}
187
188/**
189 * ext4_find_near - find a place for allocation with sufficient locality
190 * @inode: owner
191 * @ind: descriptor of indirect block.
192 *
193 * This function returns the preferred place for block allocation.
194 * It is used when heuristic for sequential allocation fails.
195 * Rules are:
196 * + if there is a block to the left of our position - allocate near it.
197 * + if pointer will live in indirect block - allocate near that block.
198 * + if pointer will live in inode - allocate in the same
199 * cylinder group.
200 *
201 * In the latter case we colour the starting block by the callers PID to
202 * prevent it from clashing with concurrent allocations for a different inode
203 * in the same block group. The PID is used here so that functionally related
204 * files will be close-by on-disk.
205 *
206 * Caller must make sure that @ind is valid and will stay that way.
207 */
208static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
209{
210 struct ext4_inode_info *ei = EXT4_I(inode);
211 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
212 __le32 *p;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400213
214 /* Try to find previous block */
215 for (p = ind->p - 1; p >= start; p--) {
216 if (*p)
217 return le32_to_cpu(*p);
218 }
219
220 /* No such thing, so let's try location of indirect block */
221 if (ind->bh)
222 return ind->bh->b_blocknr;
223
224 /*
225 * It is going to be referred to from the inode itself? OK, just put it
226 * into the same cylinder group then.
227 */
Eric Sandeenf86186b2011-06-28 10:01:31 -0400228 return ext4_inode_to_goal_block(inode);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400229}
230
231/**
232 * ext4_find_goal - find a preferred place for allocation.
233 * @inode: owner
234 * @block: block we want
235 * @partial: pointer to the last triple within a chain
236 *
237 * Normally this function find the preferred place for block allocation,
238 * returns it.
239 * Because this is only used for non-extent files, we limit the block nr
240 * to 32 bits.
241 */
242static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
243 Indirect *partial)
244{
245 ext4_fsblk_t goal;
246
247 /*
248 * XXX need to get goal block from mballoc's data structures
249 */
250
251 goal = ext4_find_near(inode, partial);
252 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
253 return goal;
254}
255
256/**
257 * ext4_blks_to_allocate - Look up the block map and count the number
258 * of direct blocks need to be allocated for the given branch.
259 *
260 * @branch: chain of indirect blocks
261 * @k: number of blocks need for indirect blocks
262 * @blks: number of data blocks to be mapped.
263 * @blocks_to_boundary: the offset in the indirect block
264 *
265 * return the total number of blocks to be allocate, including the
266 * direct and indirect blocks.
267 */
268static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
269 int blocks_to_boundary)
270{
271 unsigned int count = 0;
272
273 /*
274 * Simple case, [t,d]Indirect block(s) has not allocated yet
275 * then it's clear blocks on that path have not allocated
276 */
277 if (k > 0) {
278 /* right now we don't handle cross boundary allocation */
279 if (blks < blocks_to_boundary + 1)
280 count += blks;
281 else
282 count += blocks_to_boundary + 1;
283 return count;
284 }
285
286 count++;
287 while (count < blks && count <= blocks_to_boundary &&
288 le32_to_cpu(*(branch[0].p + count)) == 0) {
289 count++;
290 }
291 return count;
292}
293
294/**
295 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
296 * @handle: handle for this transaction
297 * @inode: inode which needs allocated blocks
298 * @iblock: the logical block to start allocated at
299 * @goal: preferred physical block of allocation
300 * @indirect_blks: the number of blocks need to allocate for indirect
301 * blocks
302 * @blks: number of desired blocks
303 * @new_blocks: on return it will store the new block numbers for
304 * the indirect blocks(if needed) and the first direct block,
305 * @err: on return it will store the error code
306 *
307 * This function will return the number of blocks allocated as
308 * requested by the passed-in parameters.
309 */
310static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
311 ext4_lblk_t iblock, ext4_fsblk_t goal,
312 int indirect_blks, int blks,
313 ext4_fsblk_t new_blocks[4], int *err)
314{
315 struct ext4_allocation_request ar;
316 int target, i;
317 unsigned long count = 0, blk_allocated = 0;
318 int index = 0;
319 ext4_fsblk_t current_block = 0;
320 int ret = 0;
321
322 /*
323 * Here we try to allocate the requested multiple blocks at once,
324 * on a best-effort basis.
325 * To build a branch, we should allocate blocks for
326 * the indirect blocks(if not allocated yet), and at least
327 * the first direct block of this branch. That's the
328 * minimum number of blocks need to allocate(required)
329 */
330 /* first we try to allocate the indirect blocks */
331 target = indirect_blks;
332 while (target > 0) {
333 count = target;
334 /* allocating blocks for indirect blocks and direct blocks */
335 current_block = ext4_new_meta_blocks(handle, inode, goal,
336 0, &count, err);
337 if (*err)
338 goto failed_out;
339
340 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
341 EXT4_ERROR_INODE(inode,
342 "current_block %llu + count %lu > %d!",
343 current_block, count,
344 EXT4_MAX_BLOCK_FILE_PHYS);
345 *err = -EIO;
346 goto failed_out;
347 }
348
349 target -= count;
350 /* allocate blocks for indirect blocks */
351 while (index < indirect_blks && count) {
352 new_blocks[index++] = current_block++;
353 count--;
354 }
355 if (count > 0) {
356 /*
357 * save the new block number
358 * for the first direct block
359 */
360 new_blocks[index] = current_block;
Julia Lawall524c19e2013-02-01 20:07:21 -0500361 WARN(1, KERN_INFO "%s returned more blocks than "
Amir Goldsteindae1e522011-06-27 19:40:50 -0400362 "requested\n", __func__);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400363 break;
364 }
365 }
366
367 target = blks - count ;
368 blk_allocated = count;
369 if (!target)
370 goto allocated;
371 /* Now allocate data blocks */
372 memset(&ar, 0, sizeof(ar));
373 ar.inode = inode;
374 ar.goal = goal;
375 ar.len = target;
376 ar.logical = iblock;
377 if (S_ISREG(inode->i_mode))
378 /* enable in-core preallocation only for regular files */
379 ar.flags = EXT4_MB_HINT_DATA;
380
381 current_block = ext4_mb_new_blocks(handle, &ar, err);
382 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
383 EXT4_ERROR_INODE(inode,
384 "current_block %llu + ar.len %d > %d!",
385 current_block, ar.len,
386 EXT4_MAX_BLOCK_FILE_PHYS);
387 *err = -EIO;
388 goto failed_out;
389 }
390
391 if (*err && (target == blks)) {
392 /*
393 * if the allocation failed and we didn't allocate
394 * any blocks before
395 */
396 goto failed_out;
397 }
398 if (!*err) {
399 if (target == blks) {
400 /*
401 * save the new block number
402 * for the first direct block
403 */
404 new_blocks[index] = current_block;
405 }
406 blk_allocated += ar.len;
407 }
408allocated:
409 /* total number of blocks allocated for direct blocks */
410 ret = blk_allocated;
411 *err = 0;
412 return ret;
413failed_out:
414 for (i = 0; i < index; i++)
415 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
416 return ret;
417}
418
419/**
420 * ext4_alloc_branch - allocate and set up a chain of blocks.
421 * @handle: handle for this transaction
422 * @inode: owner
423 * @indirect_blks: number of allocated indirect blocks
424 * @blks: number of allocated direct blocks
425 * @goal: preferred place for allocation
426 * @offsets: offsets (in the blocks) to store the pointers to next.
427 * @branch: place to store the chain in.
428 *
429 * This function allocates blocks, zeroes out all but the last one,
430 * links them into chain and (if we are synchronous) writes them to disk.
431 * In other words, it prepares a branch that can be spliced onto the
432 * inode. It stores the information about that chain in the branch[], in
433 * the same format as ext4_get_branch() would do. We are calling it after
434 * we had read the existing part of chain and partial points to the last
435 * triple of that (one with zero ->key). Upon the exit we have the same
436 * picture as after the successful ext4_get_block(), except that in one
437 * place chain is disconnected - *branch->p is still zero (we did not
438 * set the last link), but branch->key contains the number that should
439 * be placed into *branch->p to fill that gap.
440 *
441 * If allocation fails we free all blocks we've allocated (and forget
442 * their buffer_heads) and return the error value the from failed
443 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
444 * as described above and return 0.
445 */
446static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
447 ext4_lblk_t iblock, int indirect_blks,
448 int *blks, ext4_fsblk_t goal,
449 ext4_lblk_t *offsets, Indirect *branch)
450{
451 int blocksize = inode->i_sb->s_blocksize;
452 int i, n = 0;
453 int err = 0;
454 struct buffer_head *bh;
455 int num;
456 ext4_fsblk_t new_blocks[4];
457 ext4_fsblk_t current_block;
458
459 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
460 *blks, new_blocks, &err);
461 if (err)
462 return err;
463
464 branch[0].key = cpu_to_le32(new_blocks[0]);
465 /*
466 * metadata blocks and data blocks are allocated.
467 */
468 for (n = 1; n <= indirect_blks; n++) {
469 /*
470 * Get buffer_head for parent block, zero it out
471 * and set the pointer to new one, then send
472 * parent to disk.
473 */
474 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
475 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500476 err = -ENOMEM;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400477 goto failed;
478 }
479
480 branch[n].bh = bh;
481 lock_buffer(bh);
482 BUFFER_TRACE(bh, "call get_create_access");
483 err = ext4_journal_get_create_access(handle, bh);
484 if (err) {
485 /* Don't brelse(bh) here; it's done in
486 * ext4_journal_forget() below */
487 unlock_buffer(bh);
488 goto failed;
489 }
490
491 memset(bh->b_data, 0, blocksize);
492 branch[n].p = (__le32 *) bh->b_data + offsets[n];
493 branch[n].key = cpu_to_le32(new_blocks[n]);
494 *branch[n].p = branch[n].key;
495 if (n == indirect_blks) {
496 current_block = new_blocks[n];
497 /*
498 * End of chain, update the last new metablock of
499 * the chain to point to the new allocated
500 * data blocks numbers
501 */
502 for (i = 1; i < num; i++)
503 *(branch[n].p + i) = cpu_to_le32(++current_block);
504 }
505 BUFFER_TRACE(bh, "marking uptodate");
506 set_buffer_uptodate(bh);
507 unlock_buffer(bh);
508
509 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
510 err = ext4_handle_dirty_metadata(handle, inode, bh);
511 if (err)
512 goto failed;
513 }
514 *blks = num;
515 return err;
516failed:
517 /* Allocation failed, free what we already allocated */
518 ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
519 for (i = 1; i <= n ; i++) {
520 /*
521 * branch[i].bh is newly allocated, so there is no
522 * need to revoke the block, which is why we don't
523 * need to set EXT4_FREE_BLOCKS_METADATA.
524 */
525 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
526 EXT4_FREE_BLOCKS_FORGET);
527 }
528 for (i = n+1; i < indirect_blks; i++)
529 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
530
531 ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
532
533 return err;
534}
535
536/**
537 * ext4_splice_branch - splice the allocated branch onto inode.
538 * @handle: handle for this transaction
539 * @inode: owner
540 * @block: (logical) number of block we are adding
541 * @chain: chain of indirect blocks (with a missing link - see
542 * ext4_alloc_branch)
543 * @where: location of missing link
544 * @num: number of indirect blocks we are adding
545 * @blks: number of direct blocks we are adding
546 *
547 * This function fills the missing link and does all housekeeping needed in
548 * inode (->i_blocks, etc.). In case of success we end up with the full
549 * chain to new block and return 0.
550 */
551static int ext4_splice_branch(handle_t *handle, struct inode *inode,
552 ext4_lblk_t block, Indirect *where, int num,
553 int blks)
554{
555 int i;
556 int err = 0;
557 ext4_fsblk_t current_block;
558
559 /*
560 * If we're splicing into a [td]indirect block (as opposed to the
561 * inode) then we need to get write access to the [td]indirect block
562 * before the splice.
563 */
564 if (where->bh) {
565 BUFFER_TRACE(where->bh, "get_write_access");
566 err = ext4_journal_get_write_access(handle, where->bh);
567 if (err)
568 goto err_out;
569 }
570 /* That's it */
571
572 *where->p = where->key;
573
574 /*
575 * Update the host buffer_head or inode to point to more just allocated
576 * direct blocks blocks
577 */
578 if (num == 0 && blks > 1) {
579 current_block = le32_to_cpu(where->key) + 1;
580 for (i = 1; i < blks; i++)
581 *(where->p + i) = cpu_to_le32(current_block++);
582 }
583
584 /* We are done with atomic stuff, now do the rest of housekeeping */
585 /* had we spliced it onto indirect block? */
586 if (where->bh) {
587 /*
588 * If we spliced it onto an indirect block, we haven't
589 * altered the inode. Note however that if it is being spliced
590 * onto an indirect block at the very end of the file (the
591 * file is growing) then we *will* alter the inode to reflect
592 * the new i_size. But that is not done here - it is done in
593 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
594 */
595 jbd_debug(5, "splicing indirect only\n");
596 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
597 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
598 if (err)
599 goto err_out;
600 } else {
601 /*
602 * OK, we spliced it into the inode itself on a direct block.
603 */
604 ext4_mark_inode_dirty(handle, inode);
605 jbd_debug(5, "splicing direct\n");
606 }
607 return err;
608
609err_out:
610 for (i = 1; i <= num; i++) {
611 /*
612 * branch[i].bh is newly allocated, so there is no
613 * need to revoke the block, which is why we don't
614 * need to set EXT4_FREE_BLOCKS_METADATA.
615 */
616 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
617 EXT4_FREE_BLOCKS_FORGET);
618 }
619 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
620 blks, 0);
621
622 return err;
623}
624
625/*
626 * The ext4_ind_map_blocks() function handles non-extents inodes
627 * (i.e., using the traditional indirect/double-indirect i_blocks
628 * scheme) for ext4_map_blocks().
629 *
630 * Allocation strategy is simple: if we have to allocate something, we will
631 * have to go the whole way to leaf. So let's do it before attaching anything
632 * to tree, set linkage between the newborn blocks, write them if sync is
633 * required, recheck the path, free and repeat if check fails, otherwise
634 * set the last missing link (that will protect us from any truncate-generated
635 * removals - all blocks on the path are immune now) and possibly force the
636 * write on the parent block.
637 * That has a nice additional property: no special recovery from the failed
638 * allocations is needed - we simply release blocks and do not touch anything
639 * reachable from inode.
640 *
641 * `handle' can be NULL if create == 0.
642 *
643 * return > 0, # of blocks mapped or allocated.
644 * return = 0, if plain lookup failed.
645 * return < 0, error case.
646 *
647 * The ext4_ind_get_blocks() function should be called with
648 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
649 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
650 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
651 * blocks.
652 */
653int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
654 struct ext4_map_blocks *map,
655 int flags)
656{
657 int err = -EIO;
658 ext4_lblk_t offsets[4];
659 Indirect chain[4];
660 Indirect *partial;
661 ext4_fsblk_t goal;
662 int indirect_blks;
663 int blocks_to_boundary = 0;
664 int depth;
665 int count = 0;
666 ext4_fsblk_t first_block = 0;
667
668 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
669 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
670 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
671 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
672 &blocks_to_boundary);
673
674 if (depth == 0)
675 goto out;
676
677 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
678
679 /* Simplest case - block found, no allocation needed */
680 if (!partial) {
681 first_block = le32_to_cpu(chain[depth - 1].key);
682 count++;
683 /*map more blocks*/
684 while (count < map->m_len && count <= blocks_to_boundary) {
685 ext4_fsblk_t blk;
686
687 blk = le32_to_cpu(*(chain[depth-1].p + count));
688
689 if (blk == first_block + count)
690 count++;
691 else
692 break;
693 }
694 goto got_it;
695 }
696
697 /* Next simple case - plain lookup or failed read of indirect block */
698 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
699 goto cleanup;
700
701 /*
702 * Okay, we need to do block allocation.
703 */
Theodore Ts'obab08ab2011-09-09 18:36:51 -0400704 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
705 EXT4_FEATURE_RO_COMPAT_BIGALLOC)) {
706 EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
707 "non-extent mapped inodes with bigalloc");
708 return -ENOSPC;
709 }
710
Amir Goldsteindae1e522011-06-27 19:40:50 -0400711 goal = ext4_find_goal(inode, map->m_lblk, partial);
712
713 /* the number of blocks need to allocate for [d,t]indirect blocks */
714 indirect_blks = (chain + depth) - partial - 1;
715
716 /*
717 * Next look up the indirect map to count the totoal number of
718 * direct blocks to allocate for this branch.
719 */
720 count = ext4_blks_to_allocate(partial, indirect_blks,
721 map->m_len, blocks_to_boundary);
722 /*
723 * Block out ext4_truncate while we alter the tree
724 */
725 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
726 &count, goal,
727 offsets + (partial - chain), partial);
728
729 /*
730 * The ext4_splice_branch call will free and forget any buffers
731 * on the new chain if there is a failure, but that risks using
732 * up transaction credits, especially for bitmaps where the
733 * credits cannot be returned. Can we handle this somehow? We
734 * may need to return -EAGAIN upwards in the worst case. --sct
735 */
736 if (!err)
737 err = ext4_splice_branch(handle, inode, map->m_lblk,
738 partial, indirect_blks, count);
739 if (err)
740 goto cleanup;
741
742 map->m_flags |= EXT4_MAP_NEW;
743
744 ext4_update_inode_fsync_trans(handle, inode, 1);
745got_it:
746 map->m_flags |= EXT4_MAP_MAPPED;
747 map->m_pblk = le32_to_cpu(chain[depth-1].key);
748 map->m_len = count;
749 if (count > blocks_to_boundary)
750 map->m_flags |= EXT4_MAP_BOUNDARY;
751 err = count;
752 /* Clean up and exit */
753 partial = chain + depth - 1; /* the whole chain */
754cleanup:
755 while (partial > chain) {
756 BUFFER_TRACE(partial->bh, "call brelse");
757 brelse(partial->bh);
758 partial--;
759 }
760out:
Zheng Liu19b303d2012-11-08 14:34:04 -0500761 trace_ext4_ind_map_blocks_exit(inode, map, err);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400762 return err;
763}
764
765/*
766 * O_DIRECT for ext3 (or indirect map) based files
767 *
768 * If the O_DIRECT write will extend the file then add this inode to the
769 * orphan list. So recovery will truncate it back to the original size
770 * if the machine crashes during the write.
771 *
772 * If the O_DIRECT write is intantiating holes inside i_size and the machine
773 * crashes then stale disk data _may_ be exposed inside the file. But current
774 * VFS code falls back into buffered path in that case so we are safe.
775 */
776ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
777 const struct iovec *iov, loff_t offset,
778 unsigned long nr_segs)
779{
780 struct file *file = iocb->ki_filp;
781 struct inode *inode = file->f_mapping->host;
782 struct ext4_inode_info *ei = EXT4_I(inode);
783 handle_t *handle;
784 ssize_t ret;
785 int orphan = 0;
786 size_t count = iov_length(iov, nr_segs);
787 int retries = 0;
788
789 if (rw == WRITE) {
790 loff_t final_size = offset + count;
791
792 if (final_size > inode->i_size) {
793 /* Credits for sb + inode write */
Theodore Ts'o9924a922013-02-08 21:59:22 -0500794 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400795 if (IS_ERR(handle)) {
796 ret = PTR_ERR(handle);
797 goto out;
798 }
799 ret = ext4_orphan_add(handle, inode);
800 if (ret) {
801 ext4_journal_stop(handle);
802 goto out;
803 }
804 orphan = 1;
805 ei->i_disksize = inode->i_size;
806 ext4_journal_stop(handle);
807 }
808 }
809
810retry:
Jiaying Zhangdccaf332011-08-19 19:13:32 -0400811 if (rw == READ && ext4_should_dioread_nolock(inode)) {
Dmitry Monakhovc2785312012-10-05 11:31:55 -0400812 if (unlikely(atomic_read(&EXT4_I(inode)->i_unwritten))) {
813 mutex_lock(&inode->i_mutex);
814 ext4_flush_unwritten_io(inode);
815 mutex_unlock(&inode->i_mutex);
816 }
Dmitry Monakhov17335dc2012-09-29 00:41:21 -0400817 /*
818 * Nolock dioread optimization may be dynamically disabled
819 * via ext4_inode_block_unlocked_dio(). Check inode's state
820 * while holding extra i_dio_count ref.
821 */
822 atomic_inc(&inode->i_dio_count);
823 smp_mb();
824 if (unlikely(ext4_test_inode_state(inode,
825 EXT4_STATE_DIOREAD_LOCK))) {
826 inode_dio_done(inode);
827 goto locked;
828 }
Amir Goldsteindae1e522011-06-27 19:40:50 -0400829 ret = __blockdev_direct_IO(rw, iocb, inode,
830 inode->i_sb->s_bdev, iov,
831 offset, nr_segs,
832 ext4_get_block, NULL, NULL, 0);
Dmitry Monakhov17335dc2012-09-29 00:41:21 -0400833 inode_dio_done(inode);
Jiaying Zhangdccaf332011-08-19 19:13:32 -0400834 } else {
Dmitry Monakhov17335dc2012-09-29 00:41:21 -0400835locked:
Linus Torvalds60ad4462011-08-01 13:56:03 -1000836 ret = blockdev_direct_IO(rw, iocb, inode, iov,
837 offset, nr_segs, ext4_get_block);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400838
839 if (unlikely((rw & WRITE) && ret < 0)) {
840 loff_t isize = i_size_read(inode);
841 loff_t end = offset + iov_length(iov, nr_segs);
842
843 if (end > isize)
844 ext4_truncate_failed_write(inode);
845 }
846 }
847 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
848 goto retry;
849
850 if (orphan) {
851 int err;
852
853 /* Credits for sb + inode write */
Theodore Ts'o9924a922013-02-08 21:59:22 -0500854 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400855 if (IS_ERR(handle)) {
856 /* This is really bad luck. We've written the data
857 * but cannot extend i_size. Bail out and pretend
858 * the write failed... */
859 ret = PTR_ERR(handle);
860 if (inode->i_nlink)
861 ext4_orphan_del(NULL, inode);
862
863 goto out;
864 }
865 if (inode->i_nlink)
866 ext4_orphan_del(handle, inode);
867 if (ret > 0) {
868 loff_t end = offset + ret;
869 if (end > inode->i_size) {
870 ei->i_disksize = end;
871 i_size_write(inode, end);
872 /*
873 * We're going to return a positive `ret'
874 * here due to non-zero-length I/O, so there's
875 * no way of reporting error returns from
876 * ext4_mark_inode_dirty() to userspace. So
877 * ignore it.
878 */
879 ext4_mark_inode_dirty(handle, inode);
880 }
881 }
882 err = ext4_journal_stop(handle);
883 if (ret == 0)
884 ret = err;
885 }
886out:
887 return ret;
888}
889
890/*
891 * Calculate the number of metadata blocks need to reserve
892 * to allocate a new block at @lblocks for non extent file based file
893 */
894int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
895{
896 struct ext4_inode_info *ei = EXT4_I(inode);
897 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
898 int blk_bits;
899
900 if (lblock < EXT4_NDIR_BLOCKS)
901 return 0;
902
903 lblock -= EXT4_NDIR_BLOCKS;
904
905 if (ei->i_da_metadata_calc_len &&
906 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
907 ei->i_da_metadata_calc_len++;
908 return 0;
909 }
910 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
911 ei->i_da_metadata_calc_len = 1;
912 blk_bits = order_base_2(lblock);
913 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
914}
915
916int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk)
917{
918 int indirects;
919
920 /* if nrblocks are contiguous */
921 if (chunk) {
922 /*
923 * With N contiguous data blocks, we need at most
924 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
925 * 2 dindirect blocks, and 1 tindirect block
926 */
927 return DIV_ROUND_UP(nrblocks,
928 EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
929 }
930 /*
931 * if nrblocks are not contiguous, worse case, each block touch
932 * a indirect block, and each indirect block touch a double indirect
933 * block, plus a triple indirect block
934 */
935 indirects = nrblocks * 2 + 1;
936 return indirects;
937}
938
939/*
940 * Truncate transactions can be complex and absolutely huge. So we need to
941 * be able to restart the transaction at a conventient checkpoint to make
942 * sure we don't overflow the journal.
943 *
944 * start_transaction gets us a new handle for a truncate transaction,
945 * and extend_transaction tries to extend the existing one a bit. If
946 * extend fails, we need to propagate the failure up and restart the
947 * transaction in the top-level truncate loop. --sct
948 */
949static handle_t *start_transaction(struct inode *inode)
950{
951 handle_t *result;
952
Theodore Ts'o9924a922013-02-08 21:59:22 -0500953 result = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
954 ext4_blocks_for_truncate(inode));
Amir Goldsteindae1e522011-06-27 19:40:50 -0400955 if (!IS_ERR(result))
956 return result;
957
958 ext4_std_error(inode->i_sb, PTR_ERR(result));
959 return result;
960}
961
962/*
963 * Try to extend this transaction for the purposes of truncation.
964 *
965 * Returns 0 if we managed to create more room. If we can't create more
966 * room, and the transaction must be restarted we return 1.
967 */
968static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
969{
970 if (!ext4_handle_valid(handle))
971 return 0;
972 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
973 return 0;
974 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
975 return 0;
976 return 1;
977}
978
979/*
980 * Probably it should be a library function... search for first non-zero word
981 * or memcmp with zero_page, whatever is better for particular architecture.
982 * Linus?
983 */
984static inline int all_zeroes(__le32 *p, __le32 *q)
985{
986 while (p < q)
987 if (*p++)
988 return 0;
989 return 1;
990}
991
992/**
993 * ext4_find_shared - find the indirect blocks for partial truncation.
994 * @inode: inode in question
995 * @depth: depth of the affected branch
996 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
997 * @chain: place to store the pointers to partial indirect blocks
998 * @top: place to the (detached) top of branch
999 *
1000 * This is a helper function used by ext4_truncate().
1001 *
1002 * When we do truncate() we may have to clean the ends of several
1003 * indirect blocks but leave the blocks themselves alive. Block is
1004 * partially truncated if some data below the new i_size is referred
1005 * from it (and it is on the path to the first completely truncated
1006 * data block, indeed). We have to free the top of that path along
1007 * with everything to the right of the path. Since no allocation
1008 * past the truncation point is possible until ext4_truncate()
1009 * finishes, we may safely do the latter, but top of branch may
1010 * require special attention - pageout below the truncation point
1011 * might try to populate it.
1012 *
1013 * We atomically detach the top of branch from the tree, store the
1014 * block number of its root in *@top, pointers to buffer_heads of
1015 * partially truncated blocks - in @chain[].bh and pointers to
1016 * their last elements that should not be removed - in
1017 * @chain[].p. Return value is the pointer to last filled element
1018 * of @chain.
1019 *
1020 * The work left to caller to do the actual freeing of subtrees:
1021 * a) free the subtree starting from *@top
1022 * b) free the subtrees whose roots are stored in
1023 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1024 * c) free the subtrees growing from the inode past the @chain[0].
1025 * (no partially truncated stuff there). */
1026
1027static Indirect *ext4_find_shared(struct inode *inode, int depth,
1028 ext4_lblk_t offsets[4], Indirect chain[4],
1029 __le32 *top)
1030{
1031 Indirect *partial, *p;
1032 int k, err;
1033
1034 *top = 0;
1035 /* Make k index the deepest non-null offset + 1 */
1036 for (k = depth; k > 1 && !offsets[k-1]; k--)
1037 ;
1038 partial = ext4_get_branch(inode, k, offsets, chain, &err);
1039 /* Writer: pointers */
1040 if (!partial)
1041 partial = chain + k-1;
1042 /*
1043 * If the branch acquired continuation since we've looked at it -
1044 * fine, it should all survive and (new) top doesn't belong to us.
1045 */
1046 if (!partial->key && *partial->p)
1047 /* Writer: end */
1048 goto no_top;
1049 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
1050 ;
1051 /*
1052 * OK, we've found the last block that must survive. The rest of our
1053 * branch should be detached before unlocking. However, if that rest
1054 * of branch is all ours and does not grow immediately from the inode
1055 * it's easier to cheat and just decrement partial->p.
1056 */
1057 if (p == chain + k - 1 && p > chain) {
1058 p->p--;
1059 } else {
1060 *top = *p->p;
1061 /* Nope, don't do this in ext4. Must leave the tree intact */
1062#if 0
1063 *p->p = 0;
1064#endif
1065 }
1066 /* Writer: end */
1067
1068 while (partial > p) {
1069 brelse(partial->bh);
1070 partial--;
1071 }
1072no_top:
1073 return partial;
1074}
1075
1076/*
1077 * Zero a number of block pointers in either an inode or an indirect block.
1078 * If we restart the transaction we must again get write access to the
1079 * indirect block for further modification.
1080 *
1081 * We release `count' blocks on disk, but (last - first) may be greater
1082 * than `count' because there can be holes in there.
1083 *
1084 * Return 0 on success, 1 on invalid block range
1085 * and < 0 on fatal error.
1086 */
1087static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
1088 struct buffer_head *bh,
1089 ext4_fsblk_t block_to_free,
1090 unsigned long count, __le32 *first,
1091 __le32 *last)
1092{
1093 __le32 *p;
1094 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
1095 int err;
1096
1097 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1098 flags |= EXT4_FREE_BLOCKS_METADATA;
1099
1100 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
1101 count)) {
1102 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
1103 "blocks %llu len %lu",
1104 (unsigned long long) block_to_free, count);
1105 return 1;
1106 }
1107
1108 if (try_to_extend_transaction(handle, inode)) {
1109 if (bh) {
1110 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1111 err = ext4_handle_dirty_metadata(handle, inode, bh);
1112 if (unlikely(err))
1113 goto out_err;
1114 }
1115 err = ext4_mark_inode_dirty(handle, inode);
1116 if (unlikely(err))
1117 goto out_err;
1118 err = ext4_truncate_restart_trans(handle, inode,
1119 ext4_blocks_for_truncate(inode));
1120 if (unlikely(err))
1121 goto out_err;
1122 if (bh) {
1123 BUFFER_TRACE(bh, "retaking write access");
1124 err = ext4_journal_get_write_access(handle, bh);
1125 if (unlikely(err))
1126 goto out_err;
1127 }
1128 }
1129
1130 for (p = first; p < last; p++)
1131 *p = 0;
1132
1133 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
1134 return 0;
1135out_err:
1136 ext4_std_error(inode->i_sb, err);
1137 return err;
1138}
1139
1140/**
1141 * ext4_free_data - free a list of data blocks
1142 * @handle: handle for this transaction
1143 * @inode: inode we are dealing with
1144 * @this_bh: indirect buffer_head which contains *@first and *@last
1145 * @first: array of block numbers
1146 * @last: points immediately past the end of array
1147 *
1148 * We are freeing all blocks referred from that array (numbers are stored as
1149 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1150 *
1151 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1152 * blocks are contiguous then releasing them at one time will only affect one
1153 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1154 * actually use a lot of journal space.
1155 *
1156 * @this_bh will be %NULL if @first and @last point into the inode's direct
1157 * block pointers.
1158 */
1159static void ext4_free_data(handle_t *handle, struct inode *inode,
1160 struct buffer_head *this_bh,
1161 __le32 *first, __le32 *last)
1162{
1163 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
1164 unsigned long count = 0; /* Number of blocks in the run */
1165 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
1166 corresponding to
1167 block_to_free */
1168 ext4_fsblk_t nr; /* Current block # */
1169 __le32 *p; /* Pointer into inode/ind
1170 for current block */
1171 int err = 0;
1172
1173 if (this_bh) { /* For indirect block */
1174 BUFFER_TRACE(this_bh, "get_write_access");
1175 err = ext4_journal_get_write_access(handle, this_bh);
1176 /* Important: if we can't update the indirect pointers
1177 * to the blocks, we can't free them. */
1178 if (err)
1179 return;
1180 }
1181
1182 for (p = first; p < last; p++) {
1183 nr = le32_to_cpu(*p);
1184 if (nr) {
1185 /* accumulate blocks to free if they're contiguous */
1186 if (count == 0) {
1187 block_to_free = nr;
1188 block_to_free_p = p;
1189 count = 1;
1190 } else if (nr == block_to_free + count) {
1191 count++;
1192 } else {
1193 err = ext4_clear_blocks(handle, inode, this_bh,
1194 block_to_free, count,
1195 block_to_free_p, p);
1196 if (err)
1197 break;
1198 block_to_free = nr;
1199 block_to_free_p = p;
1200 count = 1;
1201 }
1202 }
1203 }
1204
1205 if (!err && count > 0)
1206 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1207 count, block_to_free_p, p);
1208 if (err < 0)
1209 /* fatal error */
1210 return;
1211
1212 if (this_bh) {
1213 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1214
1215 /*
1216 * The buffer head should have an attached journal head at this
1217 * point. However, if the data is corrupted and an indirect
1218 * block pointed to itself, it would have been detached when
1219 * the block was cleared. Check for this instead of OOPSing.
1220 */
1221 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1222 ext4_handle_dirty_metadata(handle, inode, this_bh);
1223 else
1224 EXT4_ERROR_INODE(inode,
1225 "circular indirect block detected at "
1226 "block %llu",
1227 (unsigned long long) this_bh->b_blocknr);
1228 }
1229}
1230
1231/**
1232 * ext4_free_branches - free an array of branches
1233 * @handle: JBD handle for this transaction
1234 * @inode: inode we are dealing with
1235 * @parent_bh: the buffer_head which contains *@first and *@last
1236 * @first: array of block numbers
1237 * @last: pointer immediately past the end of array
1238 * @depth: depth of the branches to free
1239 *
1240 * We are freeing all blocks referred from these branches (numbers are
1241 * stored as little-endian 32-bit) and updating @inode->i_blocks
1242 * appropriately.
1243 */
1244static void ext4_free_branches(handle_t *handle, struct inode *inode,
1245 struct buffer_head *parent_bh,
1246 __le32 *first, __le32 *last, int depth)
1247{
1248 ext4_fsblk_t nr;
1249 __le32 *p;
1250
1251 if (ext4_handle_is_aborted(handle))
1252 return;
1253
1254 if (depth--) {
1255 struct buffer_head *bh;
1256 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1257 p = last;
1258 while (--p >= first) {
1259 nr = le32_to_cpu(*p);
1260 if (!nr)
1261 continue; /* A hole */
1262
1263 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1264 nr, 1)) {
1265 EXT4_ERROR_INODE(inode,
1266 "invalid indirect mapped "
1267 "block %lu (level %d)",
1268 (unsigned long) nr, depth);
1269 break;
1270 }
1271
1272 /* Go read the buffer for the next level down */
1273 bh = sb_bread(inode->i_sb, nr);
1274
1275 /*
1276 * A read failure? Report error and clear slot
1277 * (should be rare).
1278 */
1279 if (!bh) {
1280 EXT4_ERROR_INODE_BLOCK(inode, nr,
1281 "Read failure");
1282 continue;
1283 }
1284
1285 /* This zaps the entire block. Bottom up. */
1286 BUFFER_TRACE(bh, "free child branches");
1287 ext4_free_branches(handle, inode, bh,
1288 (__le32 *) bh->b_data,
1289 (__le32 *) bh->b_data + addr_per_block,
1290 depth);
1291 brelse(bh);
1292
1293 /*
1294 * Everything below this this pointer has been
1295 * released. Now let this top-of-subtree go.
1296 *
1297 * We want the freeing of this indirect block to be
1298 * atomic in the journal with the updating of the
1299 * bitmap block which owns it. So make some room in
1300 * the journal.
1301 *
1302 * We zero the parent pointer *after* freeing its
1303 * pointee in the bitmaps, so if extend_transaction()
1304 * for some reason fails to put the bitmap changes and
1305 * the release into the same transaction, recovery
1306 * will merely complain about releasing a free block,
1307 * rather than leaking blocks.
1308 */
1309 if (ext4_handle_is_aborted(handle))
1310 return;
1311 if (try_to_extend_transaction(handle, inode)) {
1312 ext4_mark_inode_dirty(handle, inode);
1313 ext4_truncate_restart_trans(handle, inode,
1314 ext4_blocks_for_truncate(inode));
1315 }
1316
1317 /*
1318 * The forget flag here is critical because if
1319 * we are journaling (and not doing data
1320 * journaling), we have to make sure a revoke
1321 * record is written to prevent the journal
1322 * replay from overwriting the (former)
1323 * indirect block if it gets reallocated as a
1324 * data block. This must happen in the same
1325 * transaction where the data blocks are
1326 * actually freed.
1327 */
1328 ext4_free_blocks(handle, inode, NULL, nr, 1,
1329 EXT4_FREE_BLOCKS_METADATA|
1330 EXT4_FREE_BLOCKS_FORGET);
1331
1332 if (parent_bh) {
1333 /*
1334 * The block which we have just freed is
1335 * pointed to by an indirect block: journal it
1336 */
1337 BUFFER_TRACE(parent_bh, "get_write_access");
1338 if (!ext4_journal_get_write_access(handle,
1339 parent_bh)){
1340 *p = 0;
1341 BUFFER_TRACE(parent_bh,
1342 "call ext4_handle_dirty_metadata");
1343 ext4_handle_dirty_metadata(handle,
1344 inode,
1345 parent_bh);
1346 }
1347 }
1348 }
1349 } else {
1350 /* We have reached the bottom of the tree. */
1351 BUFFER_TRACE(parent_bh, "free data blocks");
1352 ext4_free_data(handle, inode, parent_bh, first, last);
1353 }
1354}
1355
1356void ext4_ind_truncate(struct inode *inode)
1357{
1358 handle_t *handle;
1359 struct ext4_inode_info *ei = EXT4_I(inode);
1360 __le32 *i_data = ei->i_data;
1361 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1362 struct address_space *mapping = inode->i_mapping;
1363 ext4_lblk_t offsets[4];
1364 Indirect chain[4];
1365 Indirect *partial;
1366 __le32 nr = 0;
1367 int n = 0;
1368 ext4_lblk_t last_block, max_block;
Allison Henderson189e8682011-09-06 21:49:44 -04001369 loff_t page_len;
Amir Goldsteindae1e522011-06-27 19:40:50 -04001370 unsigned blocksize = inode->i_sb->s_blocksize;
Allison Henderson189e8682011-09-06 21:49:44 -04001371 int err;
Amir Goldsteindae1e522011-06-27 19:40:50 -04001372
1373 handle = start_transaction(inode);
1374 if (IS_ERR(handle))
1375 return; /* AKPM: return what? */
1376
1377 last_block = (inode->i_size + blocksize-1)
1378 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1379 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1380 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1381
Allison Henderson189e8682011-09-06 21:49:44 -04001382 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
1383 page_len = PAGE_CACHE_SIZE -
1384 (inode->i_size & (PAGE_CACHE_SIZE - 1));
1385
1386 err = ext4_discard_partial_page_buffers(handle,
1387 mapping, inode->i_size, page_len, 0);
1388
1389 if (err)
Amir Goldsteindae1e522011-06-27 19:40:50 -04001390 goto out_stop;
Allison Henderson189e8682011-09-06 21:49:44 -04001391 }
Amir Goldsteindae1e522011-06-27 19:40:50 -04001392
1393 if (last_block != max_block) {
1394 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1395 if (n == 0)
1396 goto out_stop; /* error */
1397 }
1398
1399 /*
1400 * OK. This truncate is going to happen. We add the inode to the
1401 * orphan list, so that if this truncate spans multiple transactions,
1402 * and we crash, we will resume the truncate when the filesystem
1403 * recovers. It also marks the inode dirty, to catch the new size.
1404 *
1405 * Implication: the file must always be in a sane, consistent
1406 * truncatable state while each transaction commits.
1407 */
1408 if (ext4_orphan_add(handle, inode))
1409 goto out_stop;
1410
1411 /*
1412 * From here we block out all ext4_get_block() callers who want to
1413 * modify the block allocation tree.
1414 */
1415 down_write(&ei->i_data_sem);
1416
1417 ext4_discard_preallocations(inode);
Zheng Liu51865fd2012-11-08 21:57:32 -05001418 ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
Amir Goldsteindae1e522011-06-27 19:40:50 -04001419
1420 /*
1421 * The orphan list entry will now protect us from any crash which
1422 * occurs before the truncate completes, so it is now safe to propagate
1423 * the new, shorter inode size (held for now in i_size) into the
1424 * on-disk inode. We do this via i_disksize, which is the value which
1425 * ext4 *really* writes onto the disk inode.
1426 */
1427 ei->i_disksize = inode->i_size;
1428
1429 if (last_block == max_block) {
1430 /*
1431 * It is unnecessary to free any data blocks if last_block is
1432 * equal to the indirect block limit.
1433 */
1434 goto out_unlock;
1435 } else if (n == 1) { /* direct blocks */
1436 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1437 i_data + EXT4_NDIR_BLOCKS);
1438 goto do_indirects;
1439 }
1440
1441 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1442 /* Kill the top of shared branch (not detached) */
1443 if (nr) {
1444 if (partial == chain) {
1445 /* Shared branch grows from the inode */
1446 ext4_free_branches(handle, inode, NULL,
1447 &nr, &nr+1, (chain+n-1) - partial);
1448 *partial->p = 0;
1449 /*
1450 * We mark the inode dirty prior to restart,
1451 * and prior to stop. No need for it here.
1452 */
1453 } else {
1454 /* Shared branch grows from an indirect block */
1455 BUFFER_TRACE(partial->bh, "get_write_access");
1456 ext4_free_branches(handle, inode, partial->bh,
1457 partial->p,
1458 partial->p+1, (chain+n-1) - partial);
1459 }
1460 }
1461 /* Clear the ends of indirect blocks on the shared branch */
1462 while (partial > chain) {
1463 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1464 (__le32*)partial->bh->b_data+addr_per_block,
1465 (chain+n-1) - partial);
1466 BUFFER_TRACE(partial->bh, "call brelse");
1467 brelse(partial->bh);
1468 partial--;
1469 }
1470do_indirects:
1471 /* Kill the remaining (whole) subtrees */
1472 switch (offsets[0]) {
1473 default:
1474 nr = i_data[EXT4_IND_BLOCK];
1475 if (nr) {
1476 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1477 i_data[EXT4_IND_BLOCK] = 0;
1478 }
1479 case EXT4_IND_BLOCK:
1480 nr = i_data[EXT4_DIND_BLOCK];
1481 if (nr) {
1482 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1483 i_data[EXT4_DIND_BLOCK] = 0;
1484 }
1485 case EXT4_DIND_BLOCK:
1486 nr = i_data[EXT4_TIND_BLOCK];
1487 if (nr) {
1488 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1489 i_data[EXT4_TIND_BLOCK] = 0;
1490 }
1491 case EXT4_TIND_BLOCK:
1492 ;
1493 }
1494
1495out_unlock:
1496 up_write(&ei->i_data_sem);
1497 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1498 ext4_mark_inode_dirty(handle, inode);
1499
1500 /*
1501 * In a multi-transaction truncate, we only make the final transaction
1502 * synchronous
1503 */
1504 if (IS_SYNC(inode))
1505 ext4_handle_sync(handle);
1506out_stop:
1507 /*
1508 * If this was a simple ftruncate(), and the file will remain alive
1509 * then we need to clear up the orphan record which we created above.
1510 * However, if this was a real unlink then we were called by
1511 * ext4_delete_inode(), and we allow that function to clean up the
1512 * orphan info for us.
1513 */
1514 if (inode->i_nlink)
1515 ext4_orphan_del(handle, inode);
1516
1517 ext4_journal_stop(handle);
1518 trace_ext4_truncate_exit(inode);
1519}
1520
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001521static int free_hole_blocks(handle_t *handle, struct inode *inode,
1522 struct buffer_head *parent_bh, __le32 *i_data,
1523 int level, ext4_lblk_t first,
1524 ext4_lblk_t count, int max)
1525{
1526 struct buffer_head *bh = NULL;
1527 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1528 int ret = 0;
1529 int i, inc;
1530 ext4_lblk_t offset;
1531 __le32 blk;
1532
1533 inc = 1 << ((EXT4_BLOCK_SIZE_BITS(inode->i_sb) - 2) * level);
1534 for (i = 0, offset = 0; i < max; i++, i_data++, offset += inc) {
1535 if (offset >= count + first)
1536 break;
1537 if (*i_data == 0 || (offset + inc) <= first)
1538 continue;
1539 blk = *i_data;
1540 if (level > 0) {
1541 ext4_lblk_t first2;
1542 bh = sb_bread(inode->i_sb, blk);
1543 if (!bh) {
1544 EXT4_ERROR_INODE_BLOCK(inode, blk,
1545 "Read failure");
1546 return -EIO;
1547 }
1548 first2 = (first > offset) ? first - offset : 0;
1549 ret = free_hole_blocks(handle, inode, bh,
1550 (__le32 *)bh->b_data, level - 1,
1551 first2, count - offset,
1552 inode->i_sb->s_blocksize >> 2);
1553 if (ret) {
1554 brelse(bh);
1555 goto err;
1556 }
1557 }
1558 if (level == 0 ||
1559 (bh && all_zeroes((__le32 *)bh->b_data,
1560 (__le32 *)bh->b_data + addr_per_block))) {
1561 ext4_free_data(handle, inode, parent_bh, &blk, &blk+1);
1562 *i_data = 0;
1563 }
1564 brelse(bh);
1565 bh = NULL;
1566 }
1567
1568err:
1569 return ret;
1570}
1571
1572static int ext4_free_hole_blocks(handle_t *handle, struct inode *inode,
1573 ext4_lblk_t first, ext4_lblk_t stop)
1574{
1575 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1576 int level, ret = 0;
1577 int num = EXT4_NDIR_BLOCKS;
1578 ext4_lblk_t count, max = EXT4_NDIR_BLOCKS;
1579 __le32 *i_data = EXT4_I(inode)->i_data;
1580
1581 count = stop - first;
1582 for (level = 0; level < 4; level++, max *= addr_per_block) {
1583 if (first < max) {
1584 ret = free_hole_blocks(handle, inode, NULL, i_data,
1585 level, first, count, num);
1586 if (ret)
1587 goto err;
1588 if (count > max - first)
1589 count -= max - first;
1590 else
1591 break;
1592 first = 0;
1593 } else {
1594 first -= max;
1595 }
1596 i_data += num;
1597 if (level == 0) {
1598 num = 1;
1599 max = 1;
1600 }
1601 }
1602
1603err:
1604 return ret;
1605}
1606
1607int ext4_ind_punch_hole(struct file *file, loff_t offset, loff_t length)
1608{
Al Viro6131ffa2013-02-27 16:59:05 -05001609 struct inode *inode = file_inode(file);
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001610 struct super_block *sb = inode->i_sb;
1611 ext4_lblk_t first_block, stop_block;
1612 struct address_space *mapping = inode->i_mapping;
1613 handle_t *handle = NULL;
1614 loff_t first_page, last_page, page_len;
1615 loff_t first_page_offset, last_page_offset;
1616 int err = 0;
1617
1618 /*
1619 * Write out all dirty pages to avoid race conditions
1620 * Then release them.
1621 */
1622 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
1623 err = filemap_write_and_wait_range(mapping,
1624 offset, offset + length - 1);
1625 if (err)
1626 return err;
1627 }
1628
1629 mutex_lock(&inode->i_mutex);
1630 /* It's not possible punch hole on append only file */
1631 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
1632 err = -EPERM;
1633 goto out_mutex;
1634 }
1635 if (IS_SWAPFILE(inode)) {
1636 err = -ETXTBSY;
1637 goto out_mutex;
1638 }
1639
1640 /* No need to punch hole beyond i_size */
1641 if (offset >= inode->i_size)
1642 goto out_mutex;
1643
1644 /*
1645 * If the hole extents beyond i_size, set the hole
1646 * to end after the page that contains i_size
1647 */
1648 if (offset + length > inode->i_size) {
1649 length = inode->i_size +
1650 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
1651 offset;
1652 }
1653
1654 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1655 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
1656
1657 first_page_offset = first_page << PAGE_CACHE_SHIFT;
1658 last_page_offset = last_page << PAGE_CACHE_SHIFT;
1659
1660 /* Now release the pages */
1661 if (last_page_offset > first_page_offset) {
1662 truncate_pagecache_range(inode, first_page_offset,
1663 last_page_offset - 1);
1664 }
1665
1666 /* Wait all existing dio works, newcomers will block on i_mutex */
1667 inode_dio_wait(inode);
1668
1669 handle = start_transaction(inode);
1670 if (IS_ERR(handle))
1671 goto out_mutex;
1672
1673 /*
1674 * Now we need to zero out the non-page-aligned data in the
1675 * pages at the start and tail of the hole, and unmap the buffer
1676 * heads for the block aligned regions of the page that were
1677 * completely zerod.
1678 */
1679 if (first_page > last_page) {
1680 /*
1681 * If the file space being truncated is contained within a page
1682 * just zero out and unmap the middle of that page
1683 */
1684 err = ext4_discard_partial_page_buffers(handle,
1685 mapping, offset, length, 0);
1686 if (err)
1687 goto out;
1688 } else {
1689 /*
1690 * Zero out and unmap the paritial page that contains
1691 * the start of the hole
1692 */
1693 page_len = first_page_offset - offset;
1694 if (page_len > 0) {
1695 err = ext4_discard_partial_page_buffers(handle, mapping,
1696 offset, page_len, 0);
1697 if (err)
1698 goto out;
1699 }
1700
1701 /*
1702 * Zero out and unmap the partial page that contains
1703 * the end of the hole
1704 */
1705 page_len = offset + length - last_page_offset;
1706 if (page_len > 0) {
1707 err = ext4_discard_partial_page_buffers(handle, mapping,
1708 last_page_offset, page_len, 0);
1709 if (err)
1710 goto out;
1711 }
1712 }
1713
1714 /*
1715 * If i_size contained in the last page, we need to
1716 * unmap and zero the paritial page after i_size
1717 */
1718 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
1719 inode->i_size % PAGE_CACHE_SIZE != 0) {
1720 page_len = PAGE_CACHE_SIZE -
1721 (inode->i_size & (PAGE_CACHE_SIZE - 1));
1722 if (page_len > 0) {
1723 err = ext4_discard_partial_page_buffers(handle,
1724 mapping, inode->i_size, page_len, 0);
1725 if (err)
1726 goto out;
1727 }
1728 }
1729
1730 first_block = (offset + sb->s_blocksize - 1) >>
1731 EXT4_BLOCK_SIZE_BITS(sb);
1732 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
1733
1734 if (first_block >= stop_block)
1735 goto out;
1736
1737 down_write(&EXT4_I(inode)->i_data_sem);
1738 ext4_discard_preallocations(inode);
1739
1740 err = ext4_es_remove_extent(inode, first_block,
1741 stop_block - first_block);
1742 err = ext4_free_hole_blocks(handle, inode, first_block, stop_block);
1743
1744 ext4_discard_preallocations(inode);
1745
1746 if (IS_SYNC(inode))
1747 ext4_handle_sync(handle);
1748
1749 up_write(&EXT4_I(inode)->i_data_sem);
1750
1751out:
1752 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1753 ext4_mark_inode_dirty(handle, inode);
1754 ext4_journal_stop(handle);
1755
1756out_mutex:
1757 mutex_unlock(&inode->i_mutex);
1758
1759 return err;
1760}