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