Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * linux/fs/ext4/indirect.c
4 : *
5 : * from
6 : *
7 : * linux/fs/ext4/inode.c
8 : *
9 : * Copyright (C) 1992, 1993, 1994, 1995
10 : * Remy Card (card@masi.ibp.fr)
11 : * Laboratoire MASI - Institut Blaise Pascal
12 : * Universite Pierre et Marie Curie (Paris VI)
13 : *
14 : * from
15 : *
16 : * linux/fs/minix/inode.c
17 : *
18 : * Copyright (C) 1991, 1992 Linus Torvalds
19 : *
20 : * Goal-directed block allocation by Stephen Tweedie
21 : * (sct@redhat.com), 1993, 1998
22 : */
23 :
24 : #include "ext4_jbd2.h"
25 : #include "truncate.h"
26 : #include <linux/dax.h>
27 : #include <linux/uio.h>
28 :
29 : #include <trace/events/ext4.h>
30 :
31 : typedef struct {
32 : __le32 *p;
33 : __le32 key;
34 : struct buffer_head *bh;
35 : } Indirect;
36 :
37 : static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
38 : {
39 230842 : p->key = *(p->p = v);
40 230842 : p->bh = bh;
41 : }
42 :
43 : /**
44 : * ext4_block_to_path - parse the block number into array of offsets
45 : * @inode: inode in question (we are only interested in its superblock)
46 : * @i_block: block number to be parsed
47 : * @offsets: array to store the offsets in
48 : * @boundary: set this non-zero if the referred-to block is likely to be
49 : * followed (on disk) by an indirect block.
50 : *
51 : * To store the locations of file's data ext4 uses a data structure common
52 : * for UNIX filesystems - tree of pointers anchored in the inode, with
53 : * data blocks at leaves and indirect blocks in intermediate nodes.
54 : * This function translates the block number into path in that tree -
55 : * return value is the path length and @offsets[n] is the offset of
56 : * pointer to (n+1)th node in the nth one. If @block is out of range
57 : * (negative or too large) warning is printed and zero returned.
58 : *
59 : * Note: function doesn't find node addresses, so no IO is needed. All
60 : * we need to know is the capacity of indirect blocks (taken from the
61 : * inode->i_sb).
62 : */
63 :
64 : /*
65 : * Portability note: the last comparison (check that we fit into triple
66 : * indirect block) is spelled differently, because otherwise on an
67 : * architecture with 32-bit longs and 8Kb pages we might get into trouble
68 : * if our filesystem had 8Kb blocks. We might use long long, but that would
69 : * kill us on x86. Oh, well, at least the sign propagation does not matter -
70 : * i_block would have to be negative in the very beginning, so we would not
71 : * get there at all.
72 : */
73 :
74 144286 : static int ext4_block_to_path(struct inode *inode,
75 : ext4_lblk_t i_block,
76 : ext4_lblk_t offsets[4], int *boundary)
77 : {
78 144286 : int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
79 144286 : int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
80 144286 : const long direct_blocks = EXT4_NDIR_BLOCKS,
81 144286 : indirect_blocks = ptrs,
82 144286 : double_blocks = (1 << (ptrs_bits * 2));
83 144286 : int n = 0;
84 144286 : int final = 0;
85 :
86 144286 : if (i_block < direct_blocks) {
87 971 : offsets[n++] = i_block;
88 971 : final = direct_blocks;
89 143315 : } else if ((i_block -= direct_blocks) < indirect_blocks) {
90 15977 : offsets[n++] = EXT4_IND_BLOCK;
91 15977 : offsets[n++] = i_block;
92 15977 : final = ptrs;
93 127338 : } else if ((i_block -= indirect_blocks) < double_blocks) {
94 127335 : offsets[n++] = EXT4_DIND_BLOCK;
95 127335 : offsets[n++] = i_block >> ptrs_bits;
96 127335 : offsets[n++] = i_block & (ptrs - 1);
97 127335 : final = ptrs;
98 3 : } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
99 3 : offsets[n++] = EXT4_TIND_BLOCK;
100 3 : offsets[n++] = i_block >> (ptrs_bits * 2);
101 3 : offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
102 3 : offsets[n++] = i_block & (ptrs - 1);
103 3 : final = ptrs;
104 : } else {
105 0 : ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
106 : i_block + direct_blocks +
107 : indirect_blocks + double_blocks, inode->i_ino);
108 : }
109 144286 : if (boundary)
110 144239 : *boundary = final - 1 - (i_block & (ptrs - 1));
111 144286 : return n;
112 : }
113 :
114 : /**
115 : * ext4_get_branch - read the chain of indirect blocks leading to data
116 : * @inode: inode in question
117 : * @depth: depth of the chain (1 - direct pointer, etc.)
118 : * @offsets: offsets of pointers in inode/indirect blocks
119 : * @chain: place to store the result
120 : * @err: here we store the error value
121 : *
122 : * Function fills the array of triples <key, p, bh> and returns %NULL
123 : * if everything went OK or the pointer to the last filled triple
124 : * (incomplete one) otherwise. Upon the return chain[i].key contains
125 : * the number of (i+1)-th block in the chain (as it is stored in memory,
126 : * i.e. little-endian 32-bit), chain[i].p contains the address of that
127 : * number (it points into struct inode for i==0 and into the bh->b_data
128 : * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
129 : * block for i>0 and NULL for i==0. In other words, it holds the block
130 : * numbers of the chain, addresses they were taken from (and where we can
131 : * verify that chain did not change) and buffer_heads hosting these
132 : * numbers.
133 : *
134 : * Function stops when it stumbles upon zero pointer (absent block)
135 : * (pointer to last triple returned, *@err == 0)
136 : * or when it gets an IO error reading an indirect block
137 : * (ditto, *@err == -EIO)
138 : * or when it reads all @depth-1 indirect blocks successfully and finds
139 : * the whole chain, all way to the data (returns %NULL, *err == 0).
140 : *
141 : * Need to be called with
142 : * down_read(&EXT4_I(inode)->i_data_sem)
143 : */
144 144261 : static Indirect *ext4_get_branch(struct inode *inode, int depth,
145 : ext4_lblk_t *offsets,
146 : Indirect chain[4], int *err)
147 : {
148 144261 : struct super_block *sb = inode->i_sb;
149 144261 : Indirect *p = chain;
150 144261 : struct buffer_head *bh;
151 144261 : unsigned int key;
152 144261 : int ret = -EIO;
153 :
154 144261 : *err = 0;
155 : /* i_data is not going away, no lock needed */
156 144261 : add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
157 144261 : if (!p->key)
158 63897 : goto no_block;
159 88852 : while (--depth) {
160 86567 : key = le32_to_cpu(p->key);
161 86567 : if (key > ext4_blocks_count(EXT4_SB(sb)->s_es)) {
162 : /* the block was out of range */
163 0 : ret = -EFSCORRUPTED;
164 0 : goto failure;
165 : }
166 86567 : bh = sb_getblk(sb, key);
167 86580 : if (unlikely(!bh)) {
168 0 : ret = -ENOMEM;
169 0 : goto failure;
170 : }
171 :
172 86580 : if (!bh_uptodate_or_lock(bh)) {
173 1997 : if (ext4_read_bh(bh, 0, NULL) < 0) {
174 0 : put_bh(bh);
175 0 : goto failure;
176 : }
177 : /* validate block references */
178 1997 : if (ext4_check_indirect_blockref(inode, bh)) {
179 0 : put_bh(bh);
180 0 : goto failure;
181 : }
182 : }
183 :
184 86581 : add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
185 : /* Reader: end */
186 86581 : if (!p->key)
187 78093 : goto no_block;
188 : }
189 : return NULL;
190 :
191 0 : failure:
192 0 : *err = ret;
193 : no_block:
194 : return p;
195 : }
196 :
197 : /**
198 : * ext4_find_near - find a place for allocation with sufficient locality
199 : * @inode: owner
200 : * @ind: descriptor of indirect block.
201 : *
202 : * This function returns the preferred place for block allocation.
203 : * It is used when heuristic for sequential allocation fails.
204 : * Rules are:
205 : * + if there is a block to the left of our position - allocate near it.
206 : * + if pointer will live in indirect block - allocate near that block.
207 : * + if pointer will live in inode - allocate in the same
208 : * cylinder group.
209 : *
210 : * In the latter case we colour the starting block by the callers PID to
211 : * prevent it from clashing with concurrent allocations for a different inode
212 : * in the same block group. The PID is used here so that functionally related
213 : * files will be close-by on-disk.
214 : *
215 : * Caller must make sure that @ind is valid and will stay that way.
216 : */
217 4422 : static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
218 : {
219 4422 : struct ext4_inode_info *ei = EXT4_I(inode);
220 4422 : __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
221 4422 : __le32 *p;
222 :
223 : /* Try to find previous block */
224 30078 : for (p = ind->p - 1; p >= start; p--) {
225 29872 : if (*p)
226 4216 : return le32_to_cpu(*p);
227 : }
228 :
229 : /* No such thing, so let's try location of indirect block */
230 206 : if (ind->bh)
231 31 : return ind->bh->b_blocknr;
232 :
233 : /*
234 : * It is going to be referred to from the inode itself? OK, just put it
235 : * into the same cylinder group then.
236 : */
237 175 : return ext4_inode_to_goal_block(inode);
238 : }
239 :
240 : /**
241 : * ext4_find_goal - find a preferred place for allocation.
242 : * @inode: owner
243 : * @block: block we want
244 : * @partial: pointer to the last triple within a chain
245 : *
246 : * Normally this function find the preferred place for block allocation,
247 : * returns it.
248 : * Because this is only used for non-extent files, we limit the block nr
249 : * to 32 bits.
250 : */
251 : static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
252 : Indirect *partial)
253 : {
254 4423 : ext4_fsblk_t goal;
255 :
256 : /*
257 : * XXX need to get goal block from mballoc's data structures
258 : */
259 :
260 4423 : goal = ext4_find_near(inode, partial);
261 4421 : goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
262 4421 : return goal;
263 : }
264 :
265 : /**
266 : * ext4_blks_to_allocate - Look up the block map and count the number
267 : * of direct blocks need to be allocated for the given branch.
268 : *
269 : * @branch: chain of indirect blocks
270 : * @k: number of blocks need for indirect blocks
271 : * @blks: number of data blocks to be mapped.
272 : * @blocks_to_boundary: the offset in the indirect block
273 : *
274 : * return the total number of blocks to be allocate, including the
275 : * direct and indirect blocks.
276 : */
277 4421 : static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
278 : int blocks_to_boundary)
279 : {
280 4421 : unsigned int count = 0;
281 :
282 : /*
283 : * Simple case, [t,d]Indirect block(s) has not allocated yet
284 : * then it's clear blocks on that path have not allocated
285 : */
286 4421 : if (k > 0) {
287 : /* right now we don't handle cross boundary allocation */
288 202 : if (blks < blocks_to_boundary + 1)
289 : count += blks;
290 : else
291 115 : count += blocks_to_boundary + 1;
292 202 : return count;
293 : }
294 :
295 : count++;
296 10385 : while (count < blks && count <= blocks_to_boundary &&
297 6166 : le32_to_cpu(*(branch[0].p + count)) == 0) {
298 6166 : count++;
299 : }
300 4219 : return count;
301 : }
302 :
303 : /**
304 : * ext4_alloc_branch() - allocate and set up a chain of blocks
305 : * @handle: handle for this transaction
306 : * @ar: structure describing the allocation request
307 : * @indirect_blks: number of allocated indirect blocks
308 : * @offsets: offsets (in the blocks) to store the pointers to next.
309 : * @branch: place to store the chain in.
310 : *
311 : * This function allocates blocks, zeroes out all but the last one,
312 : * links them into chain and (if we are synchronous) writes them to disk.
313 : * In other words, it prepares a branch that can be spliced onto the
314 : * inode. It stores the information about that chain in the branch[], in
315 : * the same format as ext4_get_branch() would do. We are calling it after
316 : * we had read the existing part of chain and partial points to the last
317 : * triple of that (one with zero ->key). Upon the exit we have the same
318 : * picture as after the successful ext4_get_block(), except that in one
319 : * place chain is disconnected - *branch->p is still zero (we did not
320 : * set the last link), but branch->key contains the number that should
321 : * be placed into *branch->p to fill that gap.
322 : *
323 : * If allocation fails we free all blocks we've allocated (and forget
324 : * their buffer_heads) and return the error value the from failed
325 : * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
326 : * as described above and return 0.
327 : */
328 4422 : static int ext4_alloc_branch(handle_t *handle,
329 : struct ext4_allocation_request *ar,
330 : int indirect_blks, ext4_lblk_t *offsets,
331 : Indirect *branch)
332 : {
333 4422 : struct buffer_head * bh;
334 4422 : ext4_fsblk_t b, new_blocks[4];
335 4422 : __le32 *p;
336 4422 : int i, j, err, len = 1;
337 :
338 9052 : for (i = 0; i <= indirect_blks; i++) {
339 4634 : if (i == indirect_blks) {
340 4422 : new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
341 : } else {
342 214 : ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
343 : ar->inode, ar->goal,
344 212 : ar->flags & EXT4_MB_DELALLOC_RESERVED,
345 : NULL, &err);
346 : /* Simplify error cleanup... */
347 214 : branch[i+1].bh = NULL;
348 : }
349 4634 : if (err) {
350 1 : i--;
351 1 : goto failed;
352 : }
353 4633 : branch[i].key = cpu_to_le32(new_blocks[i]);
354 4630 : if (i == 0)
355 4416 : continue;
356 :
357 214 : bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
358 214 : if (unlikely(!bh)) {
359 0 : err = -ENOMEM;
360 0 : goto failed;
361 : }
362 214 : lock_buffer(bh);
363 214 : BUFFER_TRACE(bh, "call get_create_access");
364 214 : err = ext4_journal_get_create_access(handle, ar->inode->i_sb,
365 : bh, EXT4_JTR_NONE);
366 212 : if (err) {
367 0 : unlock_buffer(bh);
368 0 : goto failed;
369 : }
370 :
371 212 : memset(bh->b_data, 0, bh->b_size);
372 212 : p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
373 212 : b = new_blocks[i];
374 :
375 214 : if (i == indirect_blks)
376 202 : len = ar->len;
377 120576 : for (j = 0; j < len; j++)
378 120362 : *p++ = cpu_to_le32(b++);
379 :
380 214 : BUFFER_TRACE(bh, "marking uptodate");
381 214 : set_buffer_uptodate(bh);
382 214 : unlock_buffer(bh);
383 :
384 214 : BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
385 214 : err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
386 214 : if (err)
387 0 : goto failed;
388 : }
389 : return 0;
390 1 : failed:
391 1 : if (i == indirect_blks) {
392 : /* Free data blocks */
393 0 : ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
394 0 : ar->len, 0);
395 0 : i--;
396 : }
397 1 : for (; i >= 0; i--) {
398 : /*
399 : * We want to ext4_forget() only freshly allocated indirect
400 : * blocks. Buffer for new_blocks[i] is at branch[i+1].bh
401 : * (buffer at branch[0].bh is indirect block / inode already
402 : * existing before ext4_alloc_branch() was called). Also
403 : * because blocks are freshly allocated, we don't need to
404 : * revoke them which is why we don't set
405 : * EXT4_FREE_BLOCKS_METADATA.
406 : */
407 0 : ext4_free_blocks(handle, ar->inode, branch[i+1].bh,
408 0 : new_blocks[i], 1,
409 0 : branch[i+1].bh ? EXT4_FREE_BLOCKS_FORGET : 0);
410 : }
411 1 : return err;
412 : }
413 :
414 : /**
415 : * ext4_splice_branch() - splice the allocated branch onto inode.
416 : * @handle: handle for this transaction
417 : * @ar: structure describing the allocation request
418 : * @where: location of missing link
419 : * @num: number of indirect blocks we are adding
420 : *
421 : * This function fills the missing link and does all housekeeping needed in
422 : * inode (->i_blocks, etc.). In case of success we end up with the full
423 : * chain to new block and return 0.
424 : */
425 4418 : static int ext4_splice_branch(handle_t *handle,
426 : struct ext4_allocation_request *ar,
427 : Indirect *where, int num)
428 : {
429 4418 : int i;
430 4418 : int err = 0;
431 4418 : ext4_fsblk_t current_block;
432 :
433 : /*
434 : * If we're splicing into a [td]indirect block (as opposed to the
435 : * inode) then we need to get write access to the [td]indirect block
436 : * before the splice.
437 : */
438 4418 : if (where->bh) {
439 4208 : BUFFER_TRACE(where->bh, "get_write_access");
440 4208 : err = ext4_journal_get_write_access(handle, ar->inode->i_sb,
441 : where->bh, EXT4_JTR_NONE);
442 4211 : if (err)
443 0 : goto err_out;
444 : }
445 : /* That's it */
446 :
447 4421 : *where->p = where->key;
448 :
449 : /*
450 : * Update the host buffer_head or inode to point to more just allocated
451 : * direct blocks blocks
452 : */
453 4421 : if (num == 0 && ar->len > 1) {
454 80 : current_block = le32_to_cpu(where->key) + 1;
455 5304 : for (i = 1; i < ar->len; i++)
456 5224 : *(where->p + i) = cpu_to_le32(current_block++);
457 : }
458 :
459 : /* We are done with atomic stuff, now do the rest of housekeeping */
460 : /* had we spliced it onto indirect block? */
461 4421 : if (where->bh) {
462 : /*
463 : * If we spliced it onto an indirect block, we haven't
464 : * altered the inode. Note however that if it is being spliced
465 : * onto an indirect block at the very end of the file (the
466 : * file is growing) then we *will* alter the inode to reflect
467 : * the new i_size. But that is not done here - it is done in
468 : * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
469 : */
470 4211 : ext4_debug("splicing indirect only\n");
471 4211 : BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
472 4211 : err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
473 4211 : if (err)
474 0 : goto err_out;
475 : } else {
476 : /*
477 : * OK, we spliced it into the inode itself on a direct block.
478 : */
479 210 : err = ext4_mark_inode_dirty(handle, ar->inode);
480 210 : if (unlikely(err))
481 0 : goto err_out;
482 : ext4_debug("splicing direct\n");
483 : }
484 : return err;
485 :
486 0 : err_out:
487 0 : for (i = 1; i <= num; i++) {
488 : /*
489 : * branch[i].bh is newly allocated, so there is no
490 : * need to revoke the block, which is why we don't
491 : * need to set EXT4_FREE_BLOCKS_METADATA.
492 : */
493 0 : ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
494 : EXT4_FREE_BLOCKS_FORGET);
495 : }
496 0 : ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
497 0 : ar->len, 0);
498 :
499 0 : return err;
500 : }
501 :
502 : /*
503 : * The ext4_ind_map_blocks() function handles non-extents inodes
504 : * (i.e., using the traditional indirect/double-indirect i_blocks
505 : * scheme) for ext4_map_blocks().
506 : *
507 : * Allocation strategy is simple: if we have to allocate something, we will
508 : * have to go the whole way to leaf. So let's do it before attaching anything
509 : * to tree, set linkage between the newborn blocks, write them if sync is
510 : * required, recheck the path, free and repeat if check fails, otherwise
511 : * set the last missing link (that will protect us from any truncate-generated
512 : * removals - all blocks on the path are immune now) and possibly force the
513 : * write on the parent block.
514 : * That has a nice additional property: no special recovery from the failed
515 : * allocations is needed - we simply release blocks and do not touch anything
516 : * reachable from inode.
517 : *
518 : * `handle' can be NULL if create == 0.
519 : *
520 : * return > 0, # of blocks mapped or allocated.
521 : * return = 0, if plain lookup failed.
522 : * return < 0, error case.
523 : *
524 : * The ext4_ind_get_blocks() function should be called with
525 : * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
526 : * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
527 : * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
528 : * blocks.
529 : */
530 144239 : int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
531 : struct ext4_map_blocks *map,
532 : int flags)
533 : {
534 144239 : struct ext4_allocation_request ar;
535 144239 : int err = -EIO;
536 144239 : ext4_lblk_t offsets[4];
537 144239 : Indirect chain[4];
538 144239 : Indirect *partial;
539 144239 : int indirect_blks;
540 144239 : int blocks_to_boundary = 0;
541 144239 : int depth;
542 144239 : int count = 0;
543 144239 : ext4_fsblk_t first_block = 0;
544 :
545 144239 : trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
546 144233 : ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
547 144233 : ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
548 144233 : depth = ext4_block_to_path(inode, map->m_lblk, offsets,
549 : &blocks_to_boundary);
550 :
551 144234 : if (depth == 0)
552 0 : goto out;
553 :
554 144234 : partial = ext4_get_branch(inode, depth, offsets, chain, &err);
555 :
556 : /* Simplest case - block found, no allocation needed */
557 144249 : if (!partial) {
558 2283 : first_block = le32_to_cpu(chain[depth - 1].key);
559 2283 : count++;
560 : /*map more blocks*/
561 1855580 : while (count < map->m_len && count <= blocks_to_boundary) {
562 1853309 : ext4_fsblk_t blk;
563 :
564 1853309 : blk = le32_to_cpu(*(chain[depth-1].p + count));
565 :
566 1853309 : if (blk == first_block + count)
567 1853297 : count++;
568 : else
569 : break;
570 : }
571 2283 : goto got_it;
572 : }
573 :
574 : /* Next simple case - plain lookup failed */
575 141966 : if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
576 137543 : unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
577 137543 : int i;
578 :
579 : /*
580 : * Count number blocks in a subtree under 'partial'. At each
581 : * level we count number of complete empty subtrees beyond
582 : * current offset and then descend into the subtree only
583 : * partially beyond current offset.
584 : */
585 137543 : count = 0;
586 321399 : for (i = partial - chain + 1; i < depth; i++)
587 183855 : count = count * epb + (epb - offsets[i] - 1);
588 137544 : count++;
589 : /* Fill in size of a hole we found */
590 137544 : map->m_pblk = 0;
591 137544 : map->m_len = min_t(unsigned int, map->m_len, count);
592 137544 : goto cleanup;
593 : }
594 :
595 : /* Failed read of indirect block */
596 4423 : if (err == -EIO)
597 0 : goto cleanup;
598 :
599 : /*
600 : * Okay, we need to do block allocation.
601 : */
602 4423 : if (ext4_has_feature_bigalloc(inode->i_sb)) {
603 0 : EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
604 : "non-extent mapped inodes with bigalloc");
605 0 : err = -EFSCORRUPTED;
606 0 : goto out;
607 : }
608 :
609 : /* Set up for the direct block allocation */
610 4423 : memset(&ar, 0, sizeof(ar));
611 4423 : ar.inode = inode;
612 4423 : ar.logical = map->m_lblk;
613 4423 : if (S_ISREG(inode->i_mode))
614 4322 : ar.flags = EXT4_MB_HINT_DATA;
615 4423 : if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
616 205 : ar.flags |= EXT4_MB_DELALLOC_RESERVED;
617 4423 : if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
618 205 : ar.flags |= EXT4_MB_USE_RESERVED;
619 :
620 4423 : ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
621 :
622 : /* the number of blocks need to allocate for [d,t]indirect blocks */
623 4421 : indirect_blks = (chain + depth) - partial - 1;
624 :
625 : /*
626 : * Next look up the indirect map to count the totoal number of
627 : * direct blocks to allocate for this branch.
628 : */
629 4421 : ar.len = ext4_blks_to_allocate(partial, indirect_blks,
630 : map->m_len, blocks_to_boundary);
631 :
632 : /*
633 : * Block out ext4_truncate while we alter the tree
634 : */
635 8840 : err = ext4_alloc_branch(handle, &ar, indirect_blks,
636 4422 : offsets + (partial - chain), partial);
637 :
638 : /*
639 : * The ext4_splice_branch call will free and forget any buffers
640 : * on the new chain if there is a failure, but that risks using
641 : * up transaction credits, especially for bitmaps where the
642 : * credits cannot be returned. Can we handle this somehow? We
643 : * may need to return -EAGAIN upwards in the worst case. --sct
644 : */
645 4418 : if (!err)
646 4418 : err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
647 4421 : if (err)
648 1 : goto cleanup;
649 :
650 4420 : map->m_flags |= EXT4_MAP_NEW;
651 :
652 4420 : ext4_update_inode_fsync_trans(handle, inode, 1);
653 4421 : count = ar.len;
654 :
655 : /*
656 : * Update reserved blocks/metadata blocks after successful block
657 : * allocation which had been deferred till now.
658 : */
659 4421 : if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
660 205 : ext4_da_update_reserve_space(inode, count, 1);
661 :
662 4216 : got_it:
663 6704 : map->m_flags |= EXT4_MAP_MAPPED;
664 6704 : map->m_pblk = le32_to_cpu(chain[depth-1].key);
665 6705 : map->m_len = count;
666 6705 : if (count > blocks_to_boundary)
667 1941 : map->m_flags |= EXT4_MAP_BOUNDARY;
668 6705 : err = count;
669 : /* Clean up and exit */
670 6705 : partial = chain + depth - 1; /* the whole chain */
671 : cleanup:
672 231015 : while (partial > chain) {
673 86769 : BUFFER_TRACE(partial->bh, "call brelse");
674 86769 : brelse(partial->bh);
675 86765 : partial--;
676 : }
677 144246 : out:
678 144246 : trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
679 144245 : return err;
680 : }
681 :
682 : /*
683 : * Calculate number of indirect blocks touched by mapping @nrblocks logically
684 : * contiguous blocks
685 : */
686 4566 : int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
687 : {
688 : /*
689 : * With N contiguous data blocks, we need at most
690 : * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
691 : * 2 dindirect blocks, and 1 tindirect block
692 : */
693 4566 : return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
694 : }
695 :
696 0 : static int ext4_ind_trunc_restart_fn(handle_t *handle, struct inode *inode,
697 : struct buffer_head *bh, int *dropped)
698 : {
699 0 : int err;
700 :
701 0 : if (bh) {
702 0 : BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
703 0 : err = ext4_handle_dirty_metadata(handle, inode, bh);
704 0 : if (unlikely(err))
705 : return err;
706 : }
707 0 : err = ext4_mark_inode_dirty(handle, inode);
708 0 : if (unlikely(err))
709 : return err;
710 : /*
711 : * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
712 : * moment, get_block can be called only for blocks inside i_size since
713 : * page cache has been already dropped and writes are blocked by
714 : * i_rwsem. So we can safely drop the i_data_sem here.
715 : */
716 0 : BUG_ON(EXT4_JOURNAL(inode) == NULL);
717 0 : ext4_discard_preallocations(inode, 0);
718 0 : up_write(&EXT4_I(inode)->i_data_sem);
719 0 : *dropped = 1;
720 0 : return 0;
721 : }
722 :
723 : /*
724 : * Truncate transactions can be complex and absolutely huge. So we need to
725 : * be able to restart the transaction at a convenient checkpoint to make
726 : * sure we don't overflow the journal.
727 : *
728 : * Try to extend this transaction for the purposes of truncation. If
729 : * extend fails, we restart transaction.
730 : */
731 83 : static int ext4_ind_truncate_ensure_credits(handle_t *handle,
732 : struct inode *inode,
733 : struct buffer_head *bh,
734 : int revoke_creds)
735 : {
736 83 : int ret;
737 83 : int dropped = 0;
738 :
739 83 : ret = ext4_journal_ensure_credits_fn(handle, EXT4_RESERVE_TRANS_BLOCKS,
740 : ext4_blocks_for_truncate(inode), revoke_creds,
741 : ext4_ind_trunc_restart_fn(handle, inode, bh, &dropped));
742 83 : if (dropped)
743 0 : down_write(&EXT4_I(inode)->i_data_sem);
744 83 : if (ret <= 0)
745 : return ret;
746 0 : if (bh) {
747 0 : BUFFER_TRACE(bh, "retaking write access");
748 0 : ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
749 : EXT4_JTR_NONE);
750 0 : if (unlikely(ret))
751 0 : return ret;
752 : }
753 : return 0;
754 : }
755 :
756 : /*
757 : * Probably it should be a library function... search for first non-zero word
758 : * or memcmp with zero_page, whatever is better for particular architecture.
759 : * Linus?
760 : */
761 : static inline int all_zeroes(__le32 *p, __le32 *q)
762 : {
763 2330 : while (p < q)
764 2323 : if (*p++)
765 : return 0;
766 : return 1;
767 : }
768 :
769 : /**
770 : * ext4_find_shared - find the indirect blocks for partial truncation.
771 : * @inode: inode in question
772 : * @depth: depth of the affected branch
773 : * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
774 : * @chain: place to store the pointers to partial indirect blocks
775 : * @top: place to the (detached) top of branch
776 : *
777 : * This is a helper function used by ext4_truncate().
778 : *
779 : * When we do truncate() we may have to clean the ends of several
780 : * indirect blocks but leave the blocks themselves alive. Block is
781 : * partially truncated if some data below the new i_size is referred
782 : * from it (and it is on the path to the first completely truncated
783 : * data block, indeed). We have to free the top of that path along
784 : * with everything to the right of the path. Since no allocation
785 : * past the truncation point is possible until ext4_truncate()
786 : * finishes, we may safely do the latter, but top of branch may
787 : * require special attention - pageout below the truncation point
788 : * might try to populate it.
789 : *
790 : * We atomically detach the top of branch from the tree, store the
791 : * block number of its root in *@top, pointers to buffer_heads of
792 : * partially truncated blocks - in @chain[].bh and pointers to
793 : * their last elements that should not be removed - in
794 : * @chain[].p. Return value is the pointer to last filled element
795 : * of @chain.
796 : *
797 : * The work left to caller to do the actual freeing of subtrees:
798 : * a) free the subtree starting from *@top
799 : * b) free the subtrees whose roots are stored in
800 : * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
801 : * c) free the subtrees growing from the inode past the @chain[0].
802 : * (no partially truncated stuff there). */
803 :
804 31 : static Indirect *ext4_find_shared(struct inode *inode, int depth,
805 : ext4_lblk_t offsets[4], Indirect chain[4],
806 : __le32 *top)
807 : {
808 31 : Indirect *partial, *p;
809 31 : int k, err;
810 :
811 31 : *top = 0;
812 : /* Make k index the deepest non-null offset + 1 */
813 31 : for (k = depth; k > 1 && !offsets[k-1]; k--)
814 : ;
815 31 : partial = ext4_get_branch(inode, k, offsets, chain, &err);
816 : /* Writer: pointers */
817 31 : if (!partial)
818 2 : partial = chain + k-1;
819 : /*
820 : * If the branch acquired continuation since we've looked at it -
821 : * fine, it should all survive and (new) top doesn't belong to us.
822 : */
823 31 : if (!partial->key && *partial->p)
824 : /* Writer: end */
825 0 : goto no_top;
826 62 : for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
827 : ;
828 : /*
829 : * OK, we've found the last block that must survive. The rest of our
830 : * branch should be detached before unlocking. However, if that rest
831 : * of branch is all ours and does not grow immediately from the inode
832 : * it's easier to cheat and just decrement partial->p.
833 : */
834 31 : if (p == chain + k - 1 && p > chain) {
835 24 : p->p--;
836 : } else {
837 7 : *top = *p->p;
838 : /* Nope, don't do this in ext4. Must leave the tree intact */
839 : #if 0
840 : *p->p = 0;
841 : #endif
842 : }
843 : /* Writer: end */
844 :
845 38 : while (partial > p) {
846 7 : brelse(partial->bh);
847 7 : partial--;
848 : }
849 31 : no_top:
850 31 : return partial;
851 : }
852 :
853 : /*
854 : * Zero a number of block pointers in either an inode or an indirect block.
855 : * If we restart the transaction we must again get write access to the
856 : * indirect block for further modification.
857 : *
858 : * We release `count' blocks on disk, but (last - first) may be greater
859 : * than `count' because there can be holes in there.
860 : *
861 : * Return 0 on success, 1 on invalid block range
862 : * and < 0 on fatal error.
863 : */
864 65 : static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
865 : struct buffer_head *bh,
866 : ext4_fsblk_t block_to_free,
867 : unsigned long count, __le32 *first,
868 : __le32 *last)
869 : {
870 65 : __le32 *p;
871 65 : int flags = EXT4_FREE_BLOCKS_VALIDATED;
872 65 : int err;
873 :
874 65 : if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
875 : ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
876 : flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
877 55 : else if (ext4_should_journal_data(inode))
878 0 : flags |= EXT4_FREE_BLOCKS_FORGET;
879 :
880 65 : if (!ext4_inode_block_valid(inode, block_to_free, count)) {
881 0 : EXT4_ERROR_INODE(inode, "attempt to clear invalid "
882 : "blocks %llu len %lu",
883 : (unsigned long long) block_to_free, count);
884 0 : return 1;
885 : }
886 :
887 65 : err = ext4_ind_truncate_ensure_credits(handle, inode, bh,
888 : ext4_free_data_revoke_credits(inode, count));
889 65 : if (err < 0)
890 0 : goto out_err;
891 :
892 17845 : for (p = first; p < last; p++)
893 17780 : *p = 0;
894 :
895 65 : ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
896 65 : return 0;
897 : out_err:
898 0 : ext4_std_error(inode->i_sb, err);
899 0 : return err;
900 : }
901 :
902 : /**
903 : * ext4_free_data - free a list of data blocks
904 : * @handle: handle for this transaction
905 : * @inode: inode we are dealing with
906 : * @this_bh: indirect buffer_head which contains *@first and *@last
907 : * @first: array of block numbers
908 : * @last: points immediately past the end of array
909 : *
910 : * We are freeing all blocks referred from that array (numbers are stored as
911 : * little-endian 32-bit) and updating @inode->i_blocks appropriately.
912 : *
913 : * We accumulate contiguous runs of blocks to free. Conveniently, if these
914 : * blocks are contiguous then releasing them at one time will only affect one
915 : * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
916 : * actually use a lot of journal space.
917 : *
918 : * @this_bh will be %NULL if @first and @last point into the inode's direct
919 : * block pointers.
920 : */
921 49 : static void ext4_free_data(handle_t *handle, struct inode *inode,
922 : struct buffer_head *this_bh,
923 : __le32 *first, __le32 *last)
924 : {
925 49 : ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
926 49 : unsigned long count = 0; /* Number of blocks in the run */
927 49 : __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
928 : corresponding to
929 : block_to_free */
930 49 : ext4_fsblk_t nr; /* Current block # */
931 49 : __le32 *p; /* Pointer into inode/ind
932 : for current block */
933 49 : int err = 0;
934 :
935 49 : if (this_bh) { /* For indirect block */
936 33 : BUFFER_TRACE(this_bh, "get_write_access");
937 33 : err = ext4_journal_get_write_access(handle, inode->i_sb,
938 : this_bh, EXT4_JTR_NONE);
939 : /* Important: if we can't update the indirect pointers
940 : * to the blocks, we can't free them. */
941 33 : if (err)
942 : return;
943 : }
944 :
945 22708 : for (p = first; p < last; p++) {
946 22659 : nr = le32_to_cpu(*p);
947 22659 : if (nr) {
948 : /* accumulate blocks to free if they're contiguous */
949 753 : if (count == 0) {
950 : block_to_free = nr;
951 : block_to_free_p = p;
952 : count = 1;
953 718 : } else if (nr == block_to_free + count) {
954 688 : count++;
955 : } else {
956 30 : err = ext4_clear_blocks(handle, inode, this_bh,
957 : block_to_free, count,
958 : block_to_free_p, p);
959 30 : if (err)
960 : break;
961 : block_to_free = nr;
962 : block_to_free_p = p;
963 : count = 1;
964 : }
965 : }
966 : }
967 :
968 49 : if (!err && count > 0)
969 35 : err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
970 : count, block_to_free_p, p);
971 49 : if (err < 0)
972 : /* fatal error */
973 : return;
974 :
975 49 : if (this_bh) {
976 33 : BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
977 :
978 : /*
979 : * The buffer head should have an attached journal head at this
980 : * point. However, if the data is corrupted and an indirect
981 : * block pointed to itself, it would have been detached when
982 : * the block was cleared. Check for this instead of OOPSing.
983 : */
984 33 : if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
985 33 : ext4_handle_dirty_metadata(handle, inode, this_bh);
986 : else
987 0 : EXT4_ERROR_INODE(inode,
988 : "circular indirect block detected at "
989 : "block %llu",
990 : (unsigned long long) this_bh->b_blocknr);
991 : }
992 : }
993 :
994 : /**
995 : * ext4_free_branches - free an array of branches
996 : * @handle: JBD handle for this transaction
997 : * @inode: inode we are dealing with
998 : * @parent_bh: the buffer_head which contains *@first and *@last
999 : * @first: array of block numbers
1000 : * @last: pointer immediately past the end of array
1001 : * @depth: depth of the branches to free
1002 : *
1003 : * We are freeing all blocks referred from these branches (numbers are
1004 : * stored as little-endian 32-bit) and updating @inode->i_blocks
1005 : * appropriately.
1006 : */
1007 54 : static void ext4_free_branches(handle_t *handle, struct inode *inode,
1008 : struct buffer_head *parent_bh,
1009 : __le32 *first, __le32 *last, int depth)
1010 : {
1011 54 : ext4_fsblk_t nr;
1012 54 : __le32 *p;
1013 :
1014 54 : if (ext4_handle_is_aborted(handle))
1015 : return;
1016 :
1017 54 : if (depth--) {
1018 21 : struct buffer_head *bh;
1019 21 : int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1020 21 : p = last;
1021 6057 : while (--p >= first) {
1022 6036 : nr = le32_to_cpu(*p);
1023 6036 : if (!nr)
1024 6018 : continue; /* A hole */
1025 :
1026 18 : if (!ext4_inode_block_valid(inode, nr, 1)) {
1027 0 : EXT4_ERROR_INODE(inode,
1028 : "invalid indirect mapped "
1029 : "block %lu (level %d)",
1030 : (unsigned long) nr, depth);
1031 0 : break;
1032 : }
1033 :
1034 : /* Go read the buffer for the next level down */
1035 18 : bh = ext4_sb_bread(inode->i_sb, nr, 0);
1036 :
1037 : /*
1038 : * A read failure? Report error and clear slot
1039 : * (should be rare).
1040 : */
1041 18 : if (IS_ERR(bh)) {
1042 0 : ext4_error_inode_block(inode, nr, -PTR_ERR(bh),
1043 : "Read failure");
1044 0 : continue;
1045 : }
1046 :
1047 : /* This zaps the entire block. Bottom up. */
1048 18 : BUFFER_TRACE(bh, "free child branches");
1049 18 : ext4_free_branches(handle, inode, bh,
1050 : (__le32 *) bh->b_data,
1051 18 : (__le32 *) bh->b_data + addr_per_block,
1052 : depth);
1053 18 : brelse(bh);
1054 :
1055 : /*
1056 : * Everything below this pointer has been
1057 : * released. Now let this top-of-subtree go.
1058 : *
1059 : * We want the freeing of this indirect block to be
1060 : * atomic in the journal with the updating of the
1061 : * bitmap block which owns it. So make some room in
1062 : * the journal.
1063 : *
1064 : * We zero the parent pointer *after* freeing its
1065 : * pointee in the bitmaps, so if extend_transaction()
1066 : * for some reason fails to put the bitmap changes and
1067 : * the release into the same transaction, recovery
1068 : * will merely complain about releasing a free block,
1069 : * rather than leaking blocks.
1070 : */
1071 18 : if (ext4_handle_is_aborted(handle))
1072 : return;
1073 18 : if (ext4_ind_truncate_ensure_credits(handle, inode,
1074 : NULL,
1075 : ext4_free_metadata_revoke_credits(
1076 : inode->i_sb, 1)) < 0)
1077 : return;
1078 :
1079 : /*
1080 : * The forget flag here is critical because if
1081 : * we are journaling (and not doing data
1082 : * journaling), we have to make sure a revoke
1083 : * record is written to prevent the journal
1084 : * replay from overwriting the (former)
1085 : * indirect block if it gets reallocated as a
1086 : * data block. This must happen in the same
1087 : * transaction where the data blocks are
1088 : * actually freed.
1089 : */
1090 18 : ext4_free_blocks(handle, inode, NULL, nr, 1,
1091 : EXT4_FREE_BLOCKS_METADATA|
1092 : EXT4_FREE_BLOCKS_FORGET);
1093 :
1094 18 : if (parent_bh) {
1095 : /*
1096 : * The block which we have just freed is
1097 : * pointed to by an indirect block: journal it
1098 : */
1099 4 : BUFFER_TRACE(parent_bh, "get_write_access");
1100 4 : if (!ext4_journal_get_write_access(handle,
1101 : inode->i_sb, parent_bh,
1102 : EXT4_JTR_NONE)) {
1103 4 : *p = 0;
1104 4 : BUFFER_TRACE(parent_bh,
1105 : "call ext4_handle_dirty_metadata");
1106 4 : ext4_handle_dirty_metadata(handle,
1107 : inode,
1108 : parent_bh);
1109 : }
1110 : }
1111 : }
1112 : } else {
1113 : /* We have reached the bottom of the tree. */
1114 33 : BUFFER_TRACE(parent_bh, "free data blocks");
1115 33 : ext4_free_data(handle, inode, parent_bh, first, last);
1116 : }
1117 : }
1118 :
1119 31 : void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1120 : {
1121 31 : struct ext4_inode_info *ei = EXT4_I(inode);
1122 31 : __le32 *i_data = ei->i_data;
1123 31 : int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1124 31 : ext4_lblk_t offsets[4];
1125 31 : Indirect chain[4];
1126 31 : Indirect *partial;
1127 31 : __le32 nr = 0;
1128 31 : int n = 0;
1129 31 : ext4_lblk_t last_block, max_block;
1130 31 : unsigned blocksize = inode->i_sb->s_blocksize;
1131 :
1132 62 : last_block = (inode->i_size + blocksize-1)
1133 31 : >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1134 62 : max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1135 31 : >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1136 :
1137 31 : if (last_block != max_block) {
1138 31 : n = ext4_block_to_path(inode, last_block, offsets, NULL);
1139 31 : if (n == 0)
1140 0 : return;
1141 : }
1142 :
1143 31 : ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1144 :
1145 : /*
1146 : * The orphan list entry will now protect us from any crash which
1147 : * occurs before the truncate completes, so it is now safe to propagate
1148 : * the new, shorter inode size (held for now in i_size) into the
1149 : * on-disk inode. We do this via i_disksize, which is the value which
1150 : * ext4 *really* writes onto the disk inode.
1151 : */
1152 31 : ei->i_disksize = inode->i_size;
1153 :
1154 31 : if (last_block == max_block) {
1155 : /*
1156 : * It is unnecessary to free any data blocks if last_block is
1157 : * equal to the indirect block limit.
1158 : */
1159 : return;
1160 31 : } else if (n == 1) { /* direct blocks */
1161 16 : ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1162 : i_data + EXT4_NDIR_BLOCKS);
1163 16 : goto do_indirects;
1164 : }
1165 :
1166 15 : partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1167 : /* Kill the top of shared branch (not detached) */
1168 15 : if (nr) {
1169 4 : if (partial == chain) {
1170 : /* Shared branch grows from the inode */
1171 4 : ext4_free_branches(handle, inode, NULL,
1172 4 : &nr, &nr+1, (chain+n-1) - partial);
1173 4 : *partial->p = 0;
1174 : /*
1175 : * We mark the inode dirty prior to restart,
1176 : * and prior to stop. No need for it here.
1177 : */
1178 : } else {
1179 : /* Shared branch grows from an indirect block */
1180 0 : BUFFER_TRACE(partial->bh, "get_write_access");
1181 0 : ext4_free_branches(handle, inode, partial->bh,
1182 : partial->p,
1183 0 : partial->p+1, (chain+n-1) - partial);
1184 : }
1185 : }
1186 : /* Clear the ends of indirect blocks on the shared branch */
1187 28 : while (partial > chain) {
1188 13 : ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1189 13 : (__le32*)partial->bh->b_data+addr_per_block,
1190 13 : (chain+n-1) - partial);
1191 13 : BUFFER_TRACE(partial->bh, "call brelse");
1192 13 : brelse(partial->bh);
1193 13 : partial--;
1194 : }
1195 15 : do_indirects:
1196 : /* Kill the remaining (whole) subtrees */
1197 31 : switch (offsets[0]) {
1198 16 : default:
1199 16 : nr = i_data[EXT4_IND_BLOCK];
1200 16 : if (nr) {
1201 6 : ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1202 6 : i_data[EXT4_IND_BLOCK] = 0;
1203 : }
1204 29 : fallthrough;
1205 : case EXT4_IND_BLOCK:
1206 29 : nr = i_data[EXT4_DIND_BLOCK];
1207 29 : if (nr) {
1208 4 : ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1209 4 : i_data[EXT4_DIND_BLOCK] = 0;
1210 : }
1211 31 : fallthrough;
1212 : case EXT4_DIND_BLOCK:
1213 31 : nr = i_data[EXT4_TIND_BLOCK];
1214 31 : if (nr) {
1215 0 : ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1216 0 : i_data[EXT4_TIND_BLOCK] = 0;
1217 : }
1218 31 : fallthrough;
1219 : case EXT4_TIND_BLOCK:
1220 31 : ;
1221 : }
1222 : }
1223 :
1224 : /**
1225 : * ext4_ind_remove_space - remove space from the range
1226 : * @handle: JBD handle for this transaction
1227 : * @inode: inode we are dealing with
1228 : * @start: First block to remove
1229 : * @end: One block after the last block to remove (exclusive)
1230 : *
1231 : * Free the blocks in the defined range (end is exclusive endpoint of
1232 : * range). This is used by ext4_punch_hole().
1233 : */
1234 8 : int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1235 : ext4_lblk_t start, ext4_lblk_t end)
1236 : {
1237 8 : struct ext4_inode_info *ei = EXT4_I(inode);
1238 8 : __le32 *i_data = ei->i_data;
1239 8 : int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1240 8 : ext4_lblk_t offsets[4], offsets2[4];
1241 8 : Indirect chain[4], chain2[4];
1242 8 : Indirect *partial, *partial2;
1243 8 : Indirect *p = NULL, *p2 = NULL;
1244 8 : ext4_lblk_t max_block;
1245 8 : __le32 nr = 0, nr2 = 0;
1246 8 : int n = 0, n2 = 0;
1247 8 : unsigned blocksize = inode->i_sb->s_blocksize;
1248 :
1249 16 : max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1250 8 : >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1251 8 : if (end >= max_block)
1252 : end = max_block;
1253 8 : if ((start >= end) || (start > max_block))
1254 : return 0;
1255 :
1256 8 : n = ext4_block_to_path(inode, start, offsets, NULL);
1257 8 : n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1258 :
1259 8 : BUG_ON(n > n2);
1260 :
1261 8 : if ((n == 1) && (n == n2)) {
1262 : /* We're punching only within direct block range */
1263 0 : ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1264 0 : i_data + offsets2[0]);
1265 0 : return 0;
1266 8 : } else if (n2 > n) {
1267 : /*
1268 : * Start and end are on a different levels so we're going to
1269 : * free partial block at start, and partial block at end of
1270 : * the range. If there are some levels in between then
1271 : * do_indirects label will take care of that.
1272 : */
1273 :
1274 1 : if (n == 1) {
1275 : /*
1276 : * Start is at the direct block level, free
1277 : * everything to the end of the level.
1278 : */
1279 0 : ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1280 : i_data + EXT4_NDIR_BLOCKS);
1281 0 : goto end_range;
1282 : }
1283 :
1284 :
1285 1 : partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1286 1 : if (nr) {
1287 0 : if (partial == chain) {
1288 : /* Shared branch grows from the inode */
1289 0 : ext4_free_branches(handle, inode, NULL,
1290 0 : &nr, &nr+1, (chain+n-1) - partial);
1291 0 : *partial->p = 0;
1292 : } else {
1293 : /* Shared branch grows from an indirect block */
1294 0 : BUFFER_TRACE(partial->bh, "get_write_access");
1295 0 : ext4_free_branches(handle, inode, partial->bh,
1296 : partial->p,
1297 0 : partial->p+1, (chain+n-1) - partial);
1298 : }
1299 : }
1300 :
1301 : /*
1302 : * Clear the ends of indirect blocks on the shared branch
1303 : * at the start of the range
1304 : */
1305 2 : while (partial > chain) {
1306 1 : ext4_free_branches(handle, inode, partial->bh,
1307 1 : partial->p + 1,
1308 1 : (__le32 *)partial->bh->b_data+addr_per_block,
1309 1 : (chain+n-1) - partial);
1310 1 : partial--;
1311 : }
1312 :
1313 1 : end_range:
1314 1 : partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1315 1 : if (nr2) {
1316 0 : if (partial2 == chain2) {
1317 : /*
1318 : * Remember, end is exclusive so here we're at
1319 : * the start of the next level we're not going
1320 : * to free. Everything was covered by the start
1321 : * of the range.
1322 : */
1323 0 : goto do_indirects;
1324 : }
1325 : } else {
1326 : /*
1327 : * ext4_find_shared returns Indirect structure which
1328 : * points to the last element which should not be
1329 : * removed by truncate. But this is end of the range
1330 : * in punch_hole so we need to point to the next element
1331 : */
1332 1 : partial2->p++;
1333 : }
1334 :
1335 : /*
1336 : * Clear the ends of indirect blocks on the shared branch
1337 : * at the end of the range
1338 : */
1339 3 : while (partial2 > chain2) {
1340 2 : ext4_free_branches(handle, inode, partial2->bh,
1341 2 : (__le32 *)partial2->bh->b_data,
1342 : partial2->p,
1343 2 : (chain2+n2-1) - partial2);
1344 2 : partial2--;
1345 : }
1346 1 : goto do_indirects;
1347 : }
1348 :
1349 : /* Punch happened within the same level (n == n2) */
1350 7 : partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1351 7 : partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1352 :
1353 : /* Free top, but only if partial2 isn't its subtree. */
1354 7 : if (nr) {
1355 2 : int level = min(partial - chain, partial2 - chain2);
1356 2 : int i;
1357 2 : int subtree = 1;
1358 :
1359 4 : for (i = 0; i <= level; i++) {
1360 2 : if (offsets[i] != offsets2[i]) {
1361 : subtree = 0;
1362 : break;
1363 : }
1364 : }
1365 :
1366 2 : if (!subtree) {
1367 0 : if (partial == chain) {
1368 : /* Shared branch grows from the inode */
1369 0 : ext4_free_branches(handle, inode, NULL,
1370 : &nr, &nr+1,
1371 0 : (chain+n-1) - partial);
1372 0 : *partial->p = 0;
1373 : } else {
1374 : /* Shared branch grows from an indirect block */
1375 0 : BUFFER_TRACE(partial->bh, "get_write_access");
1376 0 : ext4_free_branches(handle, inode, partial->bh,
1377 : partial->p,
1378 0 : partial->p+1,
1379 0 : (chain+n-1) - partial);
1380 : }
1381 : }
1382 : }
1383 :
1384 7 : if (!nr2) {
1385 : /*
1386 : * ext4_find_shared returns Indirect structure which
1387 : * points to the last element which should not be
1388 : * removed by truncate. But this is end of the range
1389 : * in punch_hole so we need to point to the next element
1390 : */
1391 6 : partial2->p++;
1392 : }
1393 :
1394 8 : while (partial > chain || partial2 > chain2) {
1395 6 : int depth = (chain+n-1) - partial;
1396 6 : int depth2 = (chain2+n2-1) - partial2;
1397 :
1398 6 : if (partial > chain && partial2 > chain2 &&
1399 5 : partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1400 : /*
1401 : * We've converged on the same block. Clear the range,
1402 : * then we're done.
1403 : */
1404 5 : ext4_free_branches(handle, inode, partial->bh,
1405 5 : partial->p + 1,
1406 : partial2->p,
1407 : (chain+n-1) - partial);
1408 5 : goto cleanup;
1409 : }
1410 :
1411 : /*
1412 : * The start and end partial branches may not be at the same
1413 : * level even though the punch happened within one level. So, we
1414 : * give them a chance to arrive at the same level, then walk
1415 : * them in step with each other until we converge on the same
1416 : * block.
1417 : */
1418 1 : if (partial > chain && depth <= depth2) {
1419 0 : ext4_free_branches(handle, inode, partial->bh,
1420 0 : partial->p + 1,
1421 0 : (__le32 *)partial->bh->b_data+addr_per_block,
1422 : (chain+n-1) - partial);
1423 0 : partial--;
1424 : }
1425 1 : if (partial2 > chain2 && depth2 <= depth) {
1426 1 : ext4_free_branches(handle, inode, partial2->bh,
1427 1 : (__le32 *)partial2->bh->b_data,
1428 : partial2->p,
1429 : (chain2+n2-1) - partial2);
1430 1 : partial2--;
1431 : }
1432 : }
1433 :
1434 8 : cleanup:
1435 15 : while (p && p > chain) {
1436 7 : BUFFER_TRACE(p->bh, "call brelse");
1437 7 : brelse(p->bh);
1438 7 : p--;
1439 : }
1440 17 : while (p2 && p2 > chain2) {
1441 9 : BUFFER_TRACE(p2->bh, "call brelse");
1442 9 : brelse(p2->bh);
1443 9 : p2--;
1444 : }
1445 : return 0;
1446 :
1447 1 : do_indirects:
1448 : /* Kill the remaining (whole) subtrees */
1449 1 : switch (offsets[0]) {
1450 0 : default:
1451 0 : if (++n >= n2)
1452 : break;
1453 0 : nr = i_data[EXT4_IND_BLOCK];
1454 0 : if (nr) {
1455 0 : ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1456 0 : i_data[EXT4_IND_BLOCK] = 0;
1457 : }
1458 1 : fallthrough;
1459 : case EXT4_IND_BLOCK:
1460 1 : if (++n >= n2)
1461 : break;
1462 0 : nr = i_data[EXT4_DIND_BLOCK];
1463 0 : if (nr) {
1464 0 : ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1465 0 : i_data[EXT4_DIND_BLOCK] = 0;
1466 : }
1467 0 : fallthrough;
1468 : case EXT4_DIND_BLOCK:
1469 0 : if (++n >= n2)
1470 : break;
1471 0 : nr = i_data[EXT4_TIND_BLOCK];
1472 0 : if (nr) {
1473 0 : ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1474 0 : i_data[EXT4_TIND_BLOCK] = 0;
1475 : }
1476 1 : fallthrough;
1477 : case EXT4_TIND_BLOCK:
1478 1 : ;
1479 : }
1480 1 : goto cleanup;
1481 : }
|
