1 /*
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
23 #include "ext4_jbd2.h"
24 #include "truncate.h"
25
26 #include <trace/events/ext4.h>
27
28 typedef struct {
29 __le32 *p;
30 __le32 key;
31 struct buffer_head *bh;
32 } Indirect;
33
add_chain(Indirect * p,struct buffer_head * bh,__le32 * v)34 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
35 {
36 p->key = *(p->p = v);
37 p->bh = bh;
38 }
39
40 /**
41 * ext4_block_to_path - parse the block number into array of offsets
42 * @inode: inode in question (we are only interested in its superblock)
43 * @i_block: block number to be parsed
44 * @offsets: array to store the offsets in
45 * @boundary: set this non-zero if the referred-to block is likely to be
46 * followed (on disk) by an indirect block.
47 *
48 * To store the locations of file's data ext4 uses a data structure common
49 * for UNIX filesystems - tree of pointers anchored in the inode, with
50 * data blocks at leaves and indirect blocks in intermediate nodes.
51 * This function translates the block number into path in that tree -
52 * return value is the path length and @offsets[n] is the offset of
53 * pointer to (n+1)th node in the nth one. If @block is out of range
54 * (negative or too large) warning is printed and zero returned.
55 *
56 * Note: function doesn't find node addresses, so no IO is needed. All
57 * we need to know is the capacity of indirect blocks (taken from the
58 * inode->i_sb).
59 */
60
61 /*
62 * Portability note: the last comparison (check that we fit into triple
63 * indirect block) is spelled differently, because otherwise on an
64 * architecture with 32-bit longs and 8Kb pages we might get into trouble
65 * if our filesystem had 8Kb blocks. We might use long long, but that would
66 * kill us on x86. Oh, well, at least the sign propagation does not matter -
67 * i_block would have to be negative in the very beginning, so we would not
68 * get there at all.
69 */
70
ext4_block_to_path(struct inode * inode,ext4_lblk_t i_block,ext4_lblk_t offsets[4],int * boundary)71 static int ext4_block_to_path(struct inode *inode,
72 ext4_lblk_t i_block,
73 ext4_lblk_t offsets[4], int *boundary)
74 {
75 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
76 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
77 const long direct_blocks = EXT4_NDIR_BLOCKS,
78 indirect_blocks = ptrs,
79 double_blocks = (1 << (ptrs_bits * 2));
80 int n = 0;
81 int final = 0;
82
83 if (i_block < direct_blocks) {
84 offsets[n++] = i_block;
85 final = direct_blocks;
86 } else if ((i_block -= direct_blocks) < indirect_blocks) {
87 offsets[n++] = EXT4_IND_BLOCK;
88 offsets[n++] = i_block;
89 final = ptrs;
90 } else if ((i_block -= indirect_blocks) < double_blocks) {
91 offsets[n++] = EXT4_DIND_BLOCK;
92 offsets[n++] = i_block >> ptrs_bits;
93 offsets[n++] = i_block & (ptrs - 1);
94 final = ptrs;
95 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
96 offsets[n++] = EXT4_TIND_BLOCK;
97 offsets[n++] = i_block >> (ptrs_bits * 2);
98 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
99 offsets[n++] = i_block & (ptrs - 1);
100 final = ptrs;
101 } else {
102 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
103 i_block + direct_blocks +
104 indirect_blocks + double_blocks, inode->i_ino);
105 }
106 if (boundary)
107 *boundary = final - 1 - (i_block & (ptrs - 1));
108 return n;
109 }
110
111 /**
112 * ext4_get_branch - read the chain of indirect blocks leading to data
113 * @inode: inode in question
114 * @depth: depth of the chain (1 - direct pointer, etc.)
115 * @offsets: offsets of pointers in inode/indirect blocks
116 * @chain: place to store the result
117 * @err: here we store the error value
118 *
119 * Function fills the array of triples <key, p, bh> and returns %NULL
120 * if everything went OK or the pointer to the last filled triple
121 * (incomplete one) otherwise. Upon the return chain[i].key contains
122 * the number of (i+1)-th block in the chain (as it is stored in memory,
123 * i.e. little-endian 32-bit), chain[i].p contains the address of that
124 * number (it points into struct inode for i==0 and into the bh->b_data
125 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
126 * block for i>0 and NULL for i==0. In other words, it holds the block
127 * numbers of the chain, addresses they were taken from (and where we can
128 * verify that chain did not change) and buffer_heads hosting these
129 * numbers.
130 *
131 * Function stops when it stumbles upon zero pointer (absent block)
132 * (pointer to last triple returned, *@err == 0)
133 * or when it gets an IO error reading an indirect block
134 * (ditto, *@err == -EIO)
135 * or when it reads all @depth-1 indirect blocks successfully and finds
136 * the whole chain, all way to the data (returns %NULL, *err == 0).
137 *
138 * Need to be called with
139 * down_read(&EXT4_I(inode)->i_data_sem)
140 */
ext4_get_branch(struct inode * inode,int depth,ext4_lblk_t * offsets,Indirect chain[4],int * err)141 static Indirect *ext4_get_branch(struct inode *inode, int depth,
142 ext4_lblk_t *offsets,
143 Indirect chain[4], int *err)
144 {
145 struct super_block *sb = inode->i_sb;
146 Indirect *p = chain;
147 struct buffer_head *bh;
148 int ret = -EIO;
149
150 *err = 0;
151 /* i_data is not going away, no lock needed */
152 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
153 if (!p->key)
154 goto no_block;
155 while (--depth) {
156 bh = sb_getblk(sb, le32_to_cpu(p->key));
157 if (unlikely(!bh)) {
158 ret = -ENOMEM;
159 goto failure;
160 }
161
162 if (!bh_uptodate_or_lock(bh)) {
163 if (bh_submit_read(bh) < 0) {
164 put_bh(bh);
165 goto failure;
166 }
167 /* validate block references */
168 if (ext4_check_indirect_blockref(inode, bh)) {
169 put_bh(bh);
170 goto failure;
171 }
172 }
173
174 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
175 /* Reader: end */
176 if (!p->key)
177 goto no_block;
178 }
179 return NULL;
180
181 failure:
182 *err = ret;
183 no_block:
184 return p;
185 }
186
187 /**
188 * ext4_find_near - find a place for allocation with sufficient locality
189 * @inode: owner
190 * @ind: descriptor of indirect block.
191 *
192 * This function returns the preferred place for block allocation.
193 * It is used when heuristic for sequential allocation fails.
194 * Rules are:
195 * + if there is a block to the left of our position - allocate near it.
196 * + if pointer will live in indirect block - allocate near that block.
197 * + if pointer will live in inode - allocate in the same
198 * cylinder group.
199 *
200 * In the latter case we colour the starting block by the callers PID to
201 * prevent it from clashing with concurrent allocations for a different inode
202 * in the same block group. The PID is used here so that functionally related
203 * files will be close-by on-disk.
204 *
205 * Caller must make sure that @ind is valid and will stay that way.
206 */
ext4_find_near(struct inode * inode,Indirect * ind)207 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
208 {
209 struct ext4_inode_info *ei = EXT4_I(inode);
210 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
211 __le32 *p;
212
213 /* Try to find previous block */
214 for (p = ind->p - 1; p >= start; p--) {
215 if (*p)
216 return le32_to_cpu(*p);
217 }
218
219 /* No such thing, so let's try location of indirect block */
220 if (ind->bh)
221 return ind->bh->b_blocknr;
222
223 /*
224 * It is going to be referred to from the inode itself? OK, just put it
225 * into the same cylinder group then.
226 */
227 return ext4_inode_to_goal_block(inode);
228 }
229
230 /**
231 * ext4_find_goal - find a preferred place for allocation.
232 * @inode: owner
233 * @block: block we want
234 * @partial: pointer to the last triple within a chain
235 *
236 * Normally this function find the preferred place for block allocation,
237 * returns it.
238 * Because this is only used for non-extent files, we limit the block nr
239 * to 32 bits.
240 */
ext4_find_goal(struct inode * inode,ext4_lblk_t block,Indirect * partial)241 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
242 Indirect *partial)
243 {
244 ext4_fsblk_t goal;
245
246 /*
247 * XXX need to get goal block from mballoc's data structures
248 */
249
250 goal = ext4_find_near(inode, partial);
251 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
252 return goal;
253 }
254
255 /**
256 * ext4_blks_to_allocate - Look up the block map and count the number
257 * of direct blocks need to be allocated for the given branch.
258 *
259 * @branch: chain of indirect blocks
260 * @k: number of blocks need for indirect blocks
261 * @blks: number of data blocks to be mapped.
262 * @blocks_to_boundary: the offset in the indirect block
263 *
264 * return the total number of blocks to be allocate, including the
265 * direct and indirect blocks.
266 */
ext4_blks_to_allocate(Indirect * branch,int k,unsigned int blks,int blocks_to_boundary)267 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
268 int blocks_to_boundary)
269 {
270 unsigned int count = 0;
271
272 /*
273 * Simple case, [t,d]Indirect block(s) has not allocated yet
274 * then it's clear blocks on that path have not allocated
275 */
276 if (k > 0) {
277 /* right now we don't handle cross boundary allocation */
278 if (blks < blocks_to_boundary + 1)
279 count += blks;
280 else
281 count += blocks_to_boundary + 1;
282 return count;
283 }
284
285 count++;
286 while (count < blks && count <= blocks_to_boundary &&
287 le32_to_cpu(*(branch[0].p + count)) == 0) {
288 count++;
289 }
290 return count;
291 }
292
293 /**
294 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
295 * @handle: handle for this transaction
296 * @inode: inode which needs allocated blocks
297 * @iblock: the logical block to start allocated at
298 * @goal: preferred physical block of allocation
299 * @indirect_blks: the number of blocks need to allocate for indirect
300 * blocks
301 * @blks: number of desired blocks
302 * @new_blocks: on return it will store the new block numbers for
303 * the indirect blocks(if needed) and the first direct block,
304 * @err: on return it will store the error code
305 *
306 * This function will return the number of blocks allocated as
307 * requested by the passed-in parameters.
308 */
ext4_alloc_blocks(handle_t * handle,struct inode * inode,ext4_lblk_t iblock,ext4_fsblk_t goal,int indirect_blks,int blks,ext4_fsblk_t new_blocks[4],int * err)309 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
310 ext4_lblk_t iblock, ext4_fsblk_t goal,
311 int indirect_blks, int blks,
312 ext4_fsblk_t new_blocks[4], int *err)
313 {
314 struct ext4_allocation_request ar;
315 int target, i;
316 unsigned long count = 0, blk_allocated = 0;
317 int index = 0;
318 ext4_fsblk_t current_block = 0;
319 int ret = 0;
320
321 /*
322 * Here we try to allocate the requested multiple blocks at once,
323 * on a best-effort basis.
324 * To build a branch, we should allocate blocks for
325 * the indirect blocks(if not allocated yet), and at least
326 * the first direct block of this branch. That's the
327 * minimum number of blocks need to allocate(required)
328 */
329 /* first we try to allocate the indirect blocks */
330 target = indirect_blks;
331 while (target > 0) {
332 count = target;
333 /* allocating blocks for indirect blocks and direct blocks */
334 current_block = ext4_new_meta_blocks(handle, inode, goal,
335 0, &count, err);
336 if (*err)
337 goto failed_out;
338
339 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
340 EXT4_ERROR_INODE(inode,
341 "current_block %llu + count %lu > %d!",
342 current_block, count,
343 EXT4_MAX_BLOCK_FILE_PHYS);
344 *err = -EIO;
345 goto failed_out;
346 }
347
348 target -= count;
349 /* allocate blocks for indirect blocks */
350 while (index < indirect_blks && count) {
351 new_blocks[index++] = current_block++;
352 count--;
353 }
354 if (count > 0) {
355 /*
356 * save the new block number
357 * for the first direct block
358 */
359 new_blocks[index] = current_block;
360 printk(KERN_INFO "%s returned more blocks than "
361 "requested\n", __func__);
362 WARN_ON(1);
363 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 }
408 allocated:
409 /* total number of blocks allocated for direct blocks */
410 ret = blk_allocated;
411 *err = 0;
412 return ret;
413 failed_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 */
ext4_alloc_branch(handle_t * handle,struct inode * inode,ext4_lblk_t iblock,int indirect_blks,int * blks,ext4_fsblk_t goal,ext4_lblk_t * offsets,Indirect * branch)446 static 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)) {
476 err = -ENOMEM;
477 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;
516 failed:
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 */
ext4_splice_branch(handle_t * handle,struct inode * inode,ext4_lblk_t block,Indirect * where,int num,int blks)551 static 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
609 err_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 */
ext4_ind_map_blocks(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,int flags)653 int 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 */
704 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
711 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);
745 got_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 */
754 cleanup:
755 while (partial > chain) {
756 BUFFER_TRACE(partial->bh, "call brelse");
757 brelse(partial->bh);
758 partial--;
759 }
760 out:
761 trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
762 map->m_pblk, map->m_len, err);
763 return err;
764 }
765
766 /*
767 * O_DIRECT for ext3 (or indirect map) based files
768 *
769 * If the O_DIRECT write will extend the file then add this inode to the
770 * orphan list. So recovery will truncate it back to the original size
771 * if the machine crashes during the write.
772 *
773 * If the O_DIRECT write is intantiating holes inside i_size and the machine
774 * crashes then stale disk data _may_ be exposed inside the file. But current
775 * VFS code falls back into buffered path in that case so we are safe.
776 */
ext4_ind_direct_IO(int rw,struct kiocb * iocb,const struct iovec * iov,loff_t offset,unsigned long nr_segs)777 ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
778 const struct iovec *iov, loff_t offset,
779 unsigned long nr_segs)
780 {
781 struct file *file = iocb->ki_filp;
782 struct inode *inode = file->f_mapping->host;
783 struct ext4_inode_info *ei = EXT4_I(inode);
784 handle_t *handle;
785 ssize_t ret;
786 int orphan = 0;
787 size_t count = iov_length(iov, nr_segs);
788 int retries = 0;
789
790 if (rw == WRITE) {
791 loff_t final_size = offset + count;
792
793 if (final_size > inode->i_size) {
794 /* Credits for sb + inode write */
795 handle = ext4_journal_start(inode, 2);
796 if (IS_ERR(handle)) {
797 ret = PTR_ERR(handle);
798 goto out;
799 }
800 ret = ext4_orphan_add(handle, inode);
801 if (ret) {
802 ext4_journal_stop(handle);
803 goto out;
804 }
805 orphan = 1;
806 ei->i_disksize = inode->i_size;
807 ext4_journal_stop(handle);
808 }
809 }
810
811 retry:
812 if (rw == READ && ext4_should_dioread_nolock(inode)) {
813 if (unlikely(!list_empty(&ei->i_completed_io_list))) {
814 mutex_lock(&inode->i_mutex);
815 ext4_flush_completed_IO(inode);
816 mutex_unlock(&inode->i_mutex);
817 }
818 ret = __blockdev_direct_IO(rw, iocb, inode,
819 inode->i_sb->s_bdev, iov,
820 offset, nr_segs,
821 ext4_get_block, NULL, NULL, 0);
822 } else {
823 ret = blockdev_direct_IO(rw, iocb, inode, iov,
824 offset, nr_segs, ext4_get_block);
825
826 if (unlikely((rw & WRITE) && ret < 0)) {
827 loff_t isize = i_size_read(inode);
828 loff_t end = offset + iov_length(iov, nr_segs);
829
830 if (end > isize)
831 ext4_truncate_failed_write(inode);
832 }
833 }
834 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
835 goto retry;
836
837 if (orphan) {
838 int err;
839
840 /* Credits for sb + inode write */
841 handle = ext4_journal_start(inode, 2);
842 if (IS_ERR(handle)) {
843 /* This is really bad luck. We've written the data
844 * but cannot extend i_size. Bail out and pretend
845 * the write failed... */
846 ret = PTR_ERR(handle);
847 if (inode->i_nlink)
848 ext4_orphan_del(NULL, inode);
849
850 goto out;
851 }
852 if (inode->i_nlink)
853 ext4_orphan_del(handle, inode);
854 if (ret > 0) {
855 loff_t end = offset + ret;
856 if (end > inode->i_size) {
857 ei->i_disksize = end;
858 i_size_write(inode, end);
859 /*
860 * We're going to return a positive `ret'
861 * here due to non-zero-length I/O, so there's
862 * no way of reporting error returns from
863 * ext4_mark_inode_dirty() to userspace. So
864 * ignore it.
865 */
866 ext4_mark_inode_dirty(handle, inode);
867 }
868 }
869 err = ext4_journal_stop(handle);
870 if (ret == 0)
871 ret = err;
872 }
873 out:
874 return ret;
875 }
876
877 /*
878 * Calculate the number of metadata blocks need to reserve
879 * to allocate a new block at @lblocks for non extent file based file
880 */
ext4_ind_calc_metadata_amount(struct inode * inode,sector_t lblock)881 int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
882 {
883 struct ext4_inode_info *ei = EXT4_I(inode);
884 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
885 int blk_bits;
886
887 if (lblock < EXT4_NDIR_BLOCKS)
888 return 0;
889
890 lblock -= EXT4_NDIR_BLOCKS;
891
892 if (ei->i_da_metadata_calc_len &&
893 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
894 ei->i_da_metadata_calc_len++;
895 return 0;
896 }
897 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
898 ei->i_da_metadata_calc_len = 1;
899 blk_bits = order_base_2(lblock);
900 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
901 }
902
ext4_ind_trans_blocks(struct inode * inode,int nrblocks,int chunk)903 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk)
904 {
905 int indirects;
906
907 /* if nrblocks are contiguous */
908 if (chunk) {
909 /*
910 * With N contiguous data blocks, we need at most
911 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
912 * 2 dindirect blocks, and 1 tindirect block
913 */
914 return DIV_ROUND_UP(nrblocks,
915 EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
916 }
917 /*
918 * if nrblocks are not contiguous, worse case, each block touch
919 * a indirect block, and each indirect block touch a double indirect
920 * block, plus a triple indirect block
921 */
922 indirects = nrblocks * 2 + 1;
923 return indirects;
924 }
925
926 /*
927 * Truncate transactions can be complex and absolutely huge. So we need to
928 * be able to restart the transaction at a conventient checkpoint to make
929 * sure we don't overflow the journal.
930 *
931 * start_transaction gets us a new handle for a truncate transaction,
932 * and extend_transaction tries to extend the existing one a bit. If
933 * extend fails, we need to propagate the failure up and restart the
934 * transaction in the top-level truncate loop. --sct
935 */
start_transaction(struct inode * inode)936 static handle_t *start_transaction(struct inode *inode)
937 {
938 handle_t *result;
939
940 result = ext4_journal_start(inode, ext4_blocks_for_truncate(inode));
941 if (!IS_ERR(result))
942 return result;
943
944 ext4_std_error(inode->i_sb, PTR_ERR(result));
945 return result;
946 }
947
948 /*
949 * Try to extend this transaction for the purposes of truncation.
950 *
951 * Returns 0 if we managed to create more room. If we can't create more
952 * room, and the transaction must be restarted we return 1.
953 */
try_to_extend_transaction(handle_t * handle,struct inode * inode)954 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
955 {
956 if (!ext4_handle_valid(handle))
957 return 0;
958 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
959 return 0;
960 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
961 return 0;
962 return 1;
963 }
964
965 /*
966 * Probably it should be a library function... search for first non-zero word
967 * or memcmp with zero_page, whatever is better for particular architecture.
968 * Linus?
969 */
all_zeroes(__le32 * p,__le32 * q)970 static inline int all_zeroes(__le32 *p, __le32 *q)
971 {
972 while (p < q)
973 if (*p++)
974 return 0;
975 return 1;
976 }
977
978 /**
979 * ext4_find_shared - find the indirect blocks for partial truncation.
980 * @inode: inode in question
981 * @depth: depth of the affected branch
982 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
983 * @chain: place to store the pointers to partial indirect blocks
984 * @top: place to the (detached) top of branch
985 *
986 * This is a helper function used by ext4_truncate().
987 *
988 * When we do truncate() we may have to clean the ends of several
989 * indirect blocks but leave the blocks themselves alive. Block is
990 * partially truncated if some data below the new i_size is referred
991 * from it (and it is on the path to the first completely truncated
992 * data block, indeed). We have to free the top of that path along
993 * with everything to the right of the path. Since no allocation
994 * past the truncation point is possible until ext4_truncate()
995 * finishes, we may safely do the latter, but top of branch may
996 * require special attention - pageout below the truncation point
997 * might try to populate it.
998 *
999 * We atomically detach the top of branch from the tree, store the
1000 * block number of its root in *@top, pointers to buffer_heads of
1001 * partially truncated blocks - in @chain[].bh and pointers to
1002 * their last elements that should not be removed - in
1003 * @chain[].p. Return value is the pointer to last filled element
1004 * of @chain.
1005 *
1006 * The work left to caller to do the actual freeing of subtrees:
1007 * a) free the subtree starting from *@top
1008 * b) free the subtrees whose roots are stored in
1009 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1010 * c) free the subtrees growing from the inode past the @chain[0].
1011 * (no partially truncated stuff there). */
1012
ext4_find_shared(struct inode * inode,int depth,ext4_lblk_t offsets[4],Indirect chain[4],__le32 * top)1013 static Indirect *ext4_find_shared(struct inode *inode, int depth,
1014 ext4_lblk_t offsets[4], Indirect chain[4],
1015 __le32 *top)
1016 {
1017 Indirect *partial, *p;
1018 int k, err;
1019
1020 *top = 0;
1021 /* Make k index the deepest non-null offset + 1 */
1022 for (k = depth; k > 1 && !offsets[k-1]; k--)
1023 ;
1024 partial = ext4_get_branch(inode, k, offsets, chain, &err);
1025 /* Writer: pointers */
1026 if (!partial)
1027 partial = chain + k-1;
1028 /*
1029 * If the branch acquired continuation since we've looked at it -
1030 * fine, it should all survive and (new) top doesn't belong to us.
1031 */
1032 if (!partial->key && *partial->p)
1033 /* Writer: end */
1034 goto no_top;
1035 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
1036 ;
1037 /*
1038 * OK, we've found the last block that must survive. The rest of our
1039 * branch should be detached before unlocking. However, if that rest
1040 * of branch is all ours and does not grow immediately from the inode
1041 * it's easier to cheat and just decrement partial->p.
1042 */
1043 if (p == chain + k - 1 && p > chain) {
1044 p->p--;
1045 } else {
1046 *top = *p->p;
1047 /* Nope, don't do this in ext4. Must leave the tree intact */
1048 #if 0
1049 *p->p = 0;
1050 #endif
1051 }
1052 /* Writer: end */
1053
1054 while (partial > p) {
1055 brelse(partial->bh);
1056 partial--;
1057 }
1058 no_top:
1059 return partial;
1060 }
1061
1062 /*
1063 * Zero a number of block pointers in either an inode or an indirect block.
1064 * If we restart the transaction we must again get write access to the
1065 * indirect block for further modification.
1066 *
1067 * We release `count' blocks on disk, but (last - first) may be greater
1068 * than `count' because there can be holes in there.
1069 *
1070 * Return 0 on success, 1 on invalid block range
1071 * and < 0 on fatal error.
1072 */
ext4_clear_blocks(handle_t * handle,struct inode * inode,struct buffer_head * bh,ext4_fsblk_t block_to_free,unsigned long count,__le32 * first,__le32 * last)1073 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
1074 struct buffer_head *bh,
1075 ext4_fsblk_t block_to_free,
1076 unsigned long count, __le32 *first,
1077 __le32 *last)
1078 {
1079 __le32 *p;
1080 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
1081 int err;
1082
1083 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1084 flags |= EXT4_FREE_BLOCKS_METADATA;
1085
1086 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
1087 count)) {
1088 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
1089 "blocks %llu len %lu",
1090 (unsigned long long) block_to_free, count);
1091 return 1;
1092 }
1093
1094 if (try_to_extend_transaction(handle, inode)) {
1095 if (bh) {
1096 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1097 err = ext4_handle_dirty_metadata(handle, inode, bh);
1098 if (unlikely(err))
1099 goto out_err;
1100 }
1101 err = ext4_mark_inode_dirty(handle, inode);
1102 if (unlikely(err))
1103 goto out_err;
1104 err = ext4_truncate_restart_trans(handle, inode,
1105 ext4_blocks_for_truncate(inode));
1106 if (unlikely(err))
1107 goto out_err;
1108 if (bh) {
1109 BUFFER_TRACE(bh, "retaking write access");
1110 err = ext4_journal_get_write_access(handle, bh);
1111 if (unlikely(err))
1112 goto out_err;
1113 }
1114 }
1115
1116 for (p = first; p < last; p++)
1117 *p = 0;
1118
1119 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
1120 return 0;
1121 out_err:
1122 ext4_std_error(inode->i_sb, err);
1123 return err;
1124 }
1125
1126 /**
1127 * ext4_free_data - free a list of data blocks
1128 * @handle: handle for this transaction
1129 * @inode: inode we are dealing with
1130 * @this_bh: indirect buffer_head which contains *@first and *@last
1131 * @first: array of block numbers
1132 * @last: points immediately past the end of array
1133 *
1134 * We are freeing all blocks referred from that array (numbers are stored as
1135 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1136 *
1137 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1138 * blocks are contiguous then releasing them at one time will only affect one
1139 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1140 * actually use a lot of journal space.
1141 *
1142 * @this_bh will be %NULL if @first and @last point into the inode's direct
1143 * block pointers.
1144 */
ext4_free_data(handle_t * handle,struct inode * inode,struct buffer_head * this_bh,__le32 * first,__le32 * last)1145 static void ext4_free_data(handle_t *handle, struct inode *inode,
1146 struct buffer_head *this_bh,
1147 __le32 *first, __le32 *last)
1148 {
1149 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
1150 unsigned long count = 0; /* Number of blocks in the run */
1151 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
1152 corresponding to
1153 block_to_free */
1154 ext4_fsblk_t nr; /* Current block # */
1155 __le32 *p; /* Pointer into inode/ind
1156 for current block */
1157 int err = 0;
1158
1159 if (this_bh) { /* For indirect block */
1160 BUFFER_TRACE(this_bh, "get_write_access");
1161 err = ext4_journal_get_write_access(handle, this_bh);
1162 /* Important: if we can't update the indirect pointers
1163 * to the blocks, we can't free them. */
1164 if (err)
1165 return;
1166 }
1167
1168 for (p = first; p < last; p++) {
1169 nr = le32_to_cpu(*p);
1170 if (nr) {
1171 /* accumulate blocks to free if they're contiguous */
1172 if (count == 0) {
1173 block_to_free = nr;
1174 block_to_free_p = p;
1175 count = 1;
1176 } else if (nr == block_to_free + count) {
1177 count++;
1178 } else {
1179 err = ext4_clear_blocks(handle, inode, this_bh,
1180 block_to_free, count,
1181 block_to_free_p, p);
1182 if (err)
1183 break;
1184 block_to_free = nr;
1185 block_to_free_p = p;
1186 count = 1;
1187 }
1188 }
1189 }
1190
1191 if (!err && count > 0)
1192 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1193 count, block_to_free_p, p);
1194 if (err < 0)
1195 /* fatal error */
1196 return;
1197
1198 if (this_bh) {
1199 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1200
1201 /*
1202 * The buffer head should have an attached journal head at this
1203 * point. However, if the data is corrupted and an indirect
1204 * block pointed to itself, it would have been detached when
1205 * the block was cleared. Check for this instead of OOPSing.
1206 */
1207 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1208 ext4_handle_dirty_metadata(handle, inode, this_bh);
1209 else
1210 EXT4_ERROR_INODE(inode,
1211 "circular indirect block detected at "
1212 "block %llu",
1213 (unsigned long long) this_bh->b_blocknr);
1214 }
1215 }
1216
1217 /**
1218 * ext4_free_branches - free an array of branches
1219 * @handle: JBD handle for this transaction
1220 * @inode: inode we are dealing with
1221 * @parent_bh: the buffer_head which contains *@first and *@last
1222 * @first: array of block numbers
1223 * @last: pointer immediately past the end of array
1224 * @depth: depth of the branches to free
1225 *
1226 * We are freeing all blocks referred from these branches (numbers are
1227 * stored as little-endian 32-bit) and updating @inode->i_blocks
1228 * appropriately.
1229 */
ext4_free_branches(handle_t * handle,struct inode * inode,struct buffer_head * parent_bh,__le32 * first,__le32 * last,int depth)1230 static void ext4_free_branches(handle_t *handle, struct inode *inode,
1231 struct buffer_head *parent_bh,
1232 __le32 *first, __le32 *last, int depth)
1233 {
1234 ext4_fsblk_t nr;
1235 __le32 *p;
1236
1237 if (ext4_handle_is_aborted(handle))
1238 return;
1239
1240 if (depth--) {
1241 struct buffer_head *bh;
1242 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1243 p = last;
1244 while (--p >= first) {
1245 nr = le32_to_cpu(*p);
1246 if (!nr)
1247 continue; /* A hole */
1248
1249 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1250 nr, 1)) {
1251 EXT4_ERROR_INODE(inode,
1252 "invalid indirect mapped "
1253 "block %lu (level %d)",
1254 (unsigned long) nr, depth);
1255 break;
1256 }
1257
1258 /* Go read the buffer for the next level down */
1259 bh = sb_bread(inode->i_sb, nr);
1260
1261 /*
1262 * A read failure? Report error and clear slot
1263 * (should be rare).
1264 */
1265 if (!bh) {
1266 EXT4_ERROR_INODE_BLOCK(inode, nr,
1267 "Read failure");
1268 continue;
1269 }
1270
1271 /* This zaps the entire block. Bottom up. */
1272 BUFFER_TRACE(bh, "free child branches");
1273 ext4_free_branches(handle, inode, bh,
1274 (__le32 *) bh->b_data,
1275 (__le32 *) bh->b_data + addr_per_block,
1276 depth);
1277 brelse(bh);
1278
1279 /*
1280 * Everything below this this pointer has been
1281 * released. Now let this top-of-subtree go.
1282 *
1283 * We want the freeing of this indirect block to be
1284 * atomic in the journal with the updating of the
1285 * bitmap block which owns it. So make some room in
1286 * the journal.
1287 *
1288 * We zero the parent pointer *after* freeing its
1289 * pointee in the bitmaps, so if extend_transaction()
1290 * for some reason fails to put the bitmap changes and
1291 * the release into the same transaction, recovery
1292 * will merely complain about releasing a free block,
1293 * rather than leaking blocks.
1294 */
1295 if (ext4_handle_is_aborted(handle))
1296 return;
1297 if (try_to_extend_transaction(handle, inode)) {
1298 ext4_mark_inode_dirty(handle, inode);
1299 ext4_truncate_restart_trans(handle, inode,
1300 ext4_blocks_for_truncate(inode));
1301 }
1302
1303 /*
1304 * The forget flag here is critical because if
1305 * we are journaling (and not doing data
1306 * journaling), we have to make sure a revoke
1307 * record is written to prevent the journal
1308 * replay from overwriting the (former)
1309 * indirect block if it gets reallocated as a
1310 * data block. This must happen in the same
1311 * transaction where the data blocks are
1312 * actually freed.
1313 */
1314 ext4_free_blocks(handle, inode, NULL, nr, 1,
1315 EXT4_FREE_BLOCKS_METADATA|
1316 EXT4_FREE_BLOCKS_FORGET);
1317
1318 if (parent_bh) {
1319 /*
1320 * The block which we have just freed is
1321 * pointed to by an indirect block: journal it
1322 */
1323 BUFFER_TRACE(parent_bh, "get_write_access");
1324 if (!ext4_journal_get_write_access(handle,
1325 parent_bh)){
1326 *p = 0;
1327 BUFFER_TRACE(parent_bh,
1328 "call ext4_handle_dirty_metadata");
1329 ext4_handle_dirty_metadata(handle,
1330 inode,
1331 parent_bh);
1332 }
1333 }
1334 }
1335 } else {
1336 /* We have reached the bottom of the tree. */
1337 BUFFER_TRACE(parent_bh, "free data blocks");
1338 ext4_free_data(handle, inode, parent_bh, first, last);
1339 }
1340 }
1341
ext4_ind_truncate(struct inode * inode)1342 void ext4_ind_truncate(struct inode *inode)
1343 {
1344 handle_t *handle;
1345 struct ext4_inode_info *ei = EXT4_I(inode);
1346 __le32 *i_data = ei->i_data;
1347 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1348 struct address_space *mapping = inode->i_mapping;
1349 ext4_lblk_t offsets[4];
1350 Indirect chain[4];
1351 Indirect *partial;
1352 __le32 nr = 0;
1353 int n = 0;
1354 ext4_lblk_t last_block, max_block;
1355 loff_t page_len;
1356 unsigned blocksize = inode->i_sb->s_blocksize;
1357 int err;
1358
1359 handle = start_transaction(inode);
1360 if (IS_ERR(handle))
1361 return; /* AKPM: return what? */
1362
1363 last_block = (inode->i_size + blocksize-1)
1364 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1365 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1366 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1367
1368 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
1369 page_len = PAGE_CACHE_SIZE -
1370 (inode->i_size & (PAGE_CACHE_SIZE - 1));
1371
1372 err = ext4_discard_partial_page_buffers(handle,
1373 mapping, inode->i_size, page_len, 0);
1374
1375 if (err)
1376 goto out_stop;
1377 }
1378
1379 if (last_block != max_block) {
1380 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1381 if (n == 0)
1382 goto out_stop; /* error */
1383 }
1384
1385 /*
1386 * OK. This truncate is going to happen. We add the inode to the
1387 * orphan list, so that if this truncate spans multiple transactions,
1388 * and we crash, we will resume the truncate when the filesystem
1389 * recovers. It also marks the inode dirty, to catch the new size.
1390 *
1391 * Implication: the file must always be in a sane, consistent
1392 * truncatable state while each transaction commits.
1393 */
1394 if (ext4_orphan_add(handle, inode))
1395 goto out_stop;
1396
1397 /*
1398 * From here we block out all ext4_get_block() callers who want to
1399 * modify the block allocation tree.
1400 */
1401 down_write(&ei->i_data_sem);
1402
1403 ext4_discard_preallocations(inode);
1404
1405 /*
1406 * The orphan list entry will now protect us from any crash which
1407 * occurs before the truncate completes, so it is now safe to propagate
1408 * the new, shorter inode size (held for now in i_size) into the
1409 * on-disk inode. We do this via i_disksize, which is the value which
1410 * ext4 *really* writes onto the disk inode.
1411 */
1412 ei->i_disksize = inode->i_size;
1413
1414 if (last_block == max_block) {
1415 /*
1416 * It is unnecessary to free any data blocks if last_block is
1417 * equal to the indirect block limit.
1418 */
1419 goto out_unlock;
1420 } else if (n == 1) { /* direct blocks */
1421 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1422 i_data + EXT4_NDIR_BLOCKS);
1423 goto do_indirects;
1424 }
1425
1426 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1427 /* Kill the top of shared branch (not detached) */
1428 if (nr) {
1429 if (partial == chain) {
1430 /* Shared branch grows from the inode */
1431 ext4_free_branches(handle, inode, NULL,
1432 &nr, &nr+1, (chain+n-1) - partial);
1433 *partial->p = 0;
1434 /*
1435 * We mark the inode dirty prior to restart,
1436 * and prior to stop. No need for it here.
1437 */
1438 } else {
1439 /* Shared branch grows from an indirect block */
1440 BUFFER_TRACE(partial->bh, "get_write_access");
1441 ext4_free_branches(handle, inode, partial->bh,
1442 partial->p,
1443 partial->p+1, (chain+n-1) - partial);
1444 }
1445 }
1446 /* Clear the ends of indirect blocks on the shared branch */
1447 while (partial > chain) {
1448 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1449 (__le32*)partial->bh->b_data+addr_per_block,
1450 (chain+n-1) - partial);
1451 BUFFER_TRACE(partial->bh, "call brelse");
1452 brelse(partial->bh);
1453 partial--;
1454 }
1455 do_indirects:
1456 /* Kill the remaining (whole) subtrees */
1457 switch (offsets[0]) {
1458 default:
1459 nr = i_data[EXT4_IND_BLOCK];
1460 if (nr) {
1461 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1462 i_data[EXT4_IND_BLOCK] = 0;
1463 }
1464 case EXT4_IND_BLOCK:
1465 nr = i_data[EXT4_DIND_BLOCK];
1466 if (nr) {
1467 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1468 i_data[EXT4_DIND_BLOCK] = 0;
1469 }
1470 case EXT4_DIND_BLOCK:
1471 nr = i_data[EXT4_TIND_BLOCK];
1472 if (nr) {
1473 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1474 i_data[EXT4_TIND_BLOCK] = 0;
1475 }
1476 case EXT4_TIND_BLOCK:
1477 ;
1478 }
1479
1480 out_unlock:
1481 up_write(&ei->i_data_sem);
1482 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1483 ext4_mark_inode_dirty(handle, inode);
1484
1485 /*
1486 * In a multi-transaction truncate, we only make the final transaction
1487 * synchronous
1488 */
1489 if (IS_SYNC(inode))
1490 ext4_handle_sync(handle);
1491 out_stop:
1492 /*
1493 * If this was a simple ftruncate(), and the file will remain alive
1494 * then we need to clear up the orphan record which we created above.
1495 * However, if this was a real unlink then we were called by
1496 * ext4_delete_inode(), and we allow that function to clean up the
1497 * orphan info for us.
1498 */
1499 if (inode->i_nlink)
1500 ext4_orphan_del(handle, inode);
1501
1502 ext4_journal_stop(handle);
1503 trace_ext4_truncate_exit(inode);
1504 }
1505
1506