]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/ext4/indirect.c
Remove rw from {,__,do_}blockdev_direct_IO()
[mirror_ubuntu-zesty-kernel.git] / fs / ext4 / indirect.c
CommitLineData
dae1e52c
AG
1/*
2 * linux/fs/ext4/indirect.c
3 *
4 * from
5 *
6 * linux/fs/ext4/inode.c
7 *
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
12 *
13 * from
14 *
15 * linux/fs/minix/inode.c
16 *
17 * Copyright (C) 1991, 1992 Linus Torvalds
18 *
19 * Goal-directed block allocation by Stephen Tweedie
20 * (sct@redhat.com), 1993, 1998
21 */
22
dae1e52c
AG
23#include "ext4_jbd2.h"
24#include "truncate.h"
e2e40f2c 25#include <linux/uio.h>
dae1e52c
AG
26
27#include <trace/events/ext4.h>
28
29typedef struct {
30 __le32 *p;
31 __le32 key;
32 struct buffer_head *bh;
33} Indirect;
34
35static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
36{
37 p->key = *(p->p = v);
38 p->bh = bh;
39}
40
41/**
42 * ext4_block_to_path - parse the block number into array of offsets
43 * @inode: inode in question (we are only interested in its superblock)
44 * @i_block: block number to be parsed
45 * @offsets: array to store the offsets in
46 * @boundary: set this non-zero if the referred-to block is likely to be
47 * followed (on disk) by an indirect block.
48 *
49 * To store the locations of file's data ext4 uses a data structure common
50 * for UNIX filesystems - tree of pointers anchored in the inode, with
51 * data blocks at leaves and indirect blocks in intermediate nodes.
52 * This function translates the block number into path in that tree -
53 * return value is the path length and @offsets[n] is the offset of
54 * pointer to (n+1)th node in the nth one. If @block is out of range
55 * (negative or too large) warning is printed and zero returned.
56 *
57 * Note: function doesn't find node addresses, so no IO is needed. All
58 * we need to know is the capacity of indirect blocks (taken from the
59 * inode->i_sb).
60 */
61
62/*
63 * Portability note: the last comparison (check that we fit into triple
64 * indirect block) is spelled differently, because otherwise on an
65 * architecture with 32-bit longs and 8Kb pages we might get into trouble
66 * if our filesystem had 8Kb blocks. We might use long long, but that would
67 * kill us on x86. Oh, well, at least the sign propagation does not matter -
68 * i_block would have to be negative in the very beginning, so we would not
69 * get there at all.
70 */
71
72static int ext4_block_to_path(struct inode *inode,
73 ext4_lblk_t i_block,
74 ext4_lblk_t offsets[4], int *boundary)
75{
76 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
77 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
78 const long direct_blocks = EXT4_NDIR_BLOCKS,
79 indirect_blocks = ptrs,
80 double_blocks = (1 << (ptrs_bits * 2));
81 int n = 0;
82 int final = 0;
83
84 if (i_block < direct_blocks) {
85 offsets[n++] = i_block;
86 final = direct_blocks;
87 } else if ((i_block -= direct_blocks) < indirect_blocks) {
88 offsets[n++] = EXT4_IND_BLOCK;
89 offsets[n++] = i_block;
90 final = ptrs;
91 } else if ((i_block -= indirect_blocks) < double_blocks) {
92 offsets[n++] = EXT4_DIND_BLOCK;
93 offsets[n++] = i_block >> ptrs_bits;
94 offsets[n++] = i_block & (ptrs - 1);
95 final = ptrs;
96 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
97 offsets[n++] = EXT4_TIND_BLOCK;
98 offsets[n++] = i_block >> (ptrs_bits * 2);
99 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
100 offsets[n++] = i_block & (ptrs - 1);
101 final = ptrs;
102 } else {
103 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
104 i_block + direct_blocks +
105 indirect_blocks + double_blocks, inode->i_ino);
106 }
107 if (boundary)
108 *boundary = final - 1 - (i_block & (ptrs - 1));
109 return n;
110}
111
112/**
113 * ext4_get_branch - read the chain of indirect blocks leading to data
114 * @inode: inode in question
115 * @depth: depth of the chain (1 - direct pointer, etc.)
116 * @offsets: offsets of pointers in inode/indirect blocks
117 * @chain: place to store the result
118 * @err: here we store the error value
119 *
120 * Function fills the array of triples <key, p, bh> and returns %NULL
121 * if everything went OK or the pointer to the last filled triple
122 * (incomplete one) otherwise. Upon the return chain[i].key contains
123 * the number of (i+1)-th block in the chain (as it is stored in memory,
124 * i.e. little-endian 32-bit), chain[i].p contains the address of that
125 * number (it points into struct inode for i==0 and into the bh->b_data
126 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
127 * block for i>0 and NULL for i==0. In other words, it holds the block
128 * numbers of the chain, addresses they were taken from (and where we can
129 * verify that chain did not change) and buffer_heads hosting these
130 * numbers.
131 *
132 * Function stops when it stumbles upon zero pointer (absent block)
133 * (pointer to last triple returned, *@err == 0)
134 * or when it gets an IO error reading an indirect block
135 * (ditto, *@err == -EIO)
136 * or when it reads all @depth-1 indirect blocks successfully and finds
137 * the whole chain, all way to the data (returns %NULL, *err == 0).
138 *
139 * Need to be called with
140 * down_read(&EXT4_I(inode)->i_data_sem)
141 */
142static Indirect *ext4_get_branch(struct inode *inode, int depth,
143 ext4_lblk_t *offsets,
144 Indirect chain[4], int *err)
145{
146 struct super_block *sb = inode->i_sb;
147 Indirect *p = chain;
148 struct buffer_head *bh;
860d21e2 149 int ret = -EIO;
dae1e52c
AG
150
151 *err = 0;
152 /* i_data is not going away, no lock needed */
153 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
154 if (!p->key)
155 goto no_block;
156 while (--depth) {
157 bh = sb_getblk(sb, le32_to_cpu(p->key));
860d21e2
TT
158 if (unlikely(!bh)) {
159 ret = -ENOMEM;
dae1e52c 160 goto failure;
860d21e2 161 }
dae1e52c
AG
162
163 if (!bh_uptodate_or_lock(bh)) {
164 if (bh_submit_read(bh) < 0) {
165 put_bh(bh);
166 goto failure;
167 }
168 /* validate block references */
169 if (ext4_check_indirect_blockref(inode, bh)) {
170 put_bh(bh);
171 goto failure;
172 }
173 }
174
175 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
176 /* Reader: end */
177 if (!p->key)
178 goto no_block;
179 }
180 return NULL;
181
182failure:
860d21e2 183 *err = ret;
dae1e52c
AG
184no_block:
185 return p;
186}
187
188/**
189 * ext4_find_near - find a place for allocation with sufficient locality
190 * @inode: owner
191 * @ind: descriptor of indirect block.
192 *
193 * This function returns the preferred place for block allocation.
194 * It is used when heuristic for sequential allocation fails.
195 * Rules are:
196 * + if there is a block to the left of our position - allocate near it.
197 * + if pointer will live in indirect block - allocate near that block.
198 * + if pointer will live in inode - allocate in the same
199 * cylinder group.
200 *
201 * In the latter case we colour the starting block by the callers PID to
202 * prevent it from clashing with concurrent allocations for a different inode
203 * in the same block group. The PID is used here so that functionally related
204 * files will be close-by on-disk.
205 *
206 * Caller must make sure that @ind is valid and will stay that way.
207 */
208static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
209{
210 struct ext4_inode_info *ei = EXT4_I(inode);
211 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
212 __le32 *p;
dae1e52c
AG
213
214 /* Try to find previous block */
215 for (p = ind->p - 1; p >= start; p--) {
216 if (*p)
217 return le32_to_cpu(*p);
218 }
219
220 /* No such thing, so let's try location of indirect block */
221 if (ind->bh)
222 return ind->bh->b_blocknr;
223
224 /*
225 * It is going to be referred to from the inode itself? OK, just put it
226 * into the same cylinder group then.
227 */
f86186b4 228 return ext4_inode_to_goal_block(inode);
dae1e52c
AG
229}
230
231/**
232 * ext4_find_goal - find a preferred place for allocation.
233 * @inode: owner
234 * @block: block we want
235 * @partial: pointer to the last triple within a chain
236 *
237 * Normally this function find the preferred place for block allocation,
238 * returns it.
239 * Because this is only used for non-extent files, we limit the block nr
240 * to 32 bits.
241 */
242static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
243 Indirect *partial)
244{
245 ext4_fsblk_t goal;
246
247 /*
248 * XXX need to get goal block from mballoc's data structures
249 */
250
251 goal = ext4_find_near(inode, partial);
252 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
253 return goal;
254}
255
256/**
257 * ext4_blks_to_allocate - Look up the block map and count the number
258 * of direct blocks need to be allocated for the given branch.
259 *
260 * @branch: chain of indirect blocks
261 * @k: number of blocks need for indirect blocks
262 * @blks: number of data blocks to be mapped.
263 * @blocks_to_boundary: the offset in the indirect block
264 *
265 * return the total number of blocks to be allocate, including the
266 * direct and indirect blocks.
267 */
268static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
269 int blocks_to_boundary)
270{
271 unsigned int count = 0;
272
273 /*
274 * Simple case, [t,d]Indirect block(s) has not allocated yet
275 * then it's clear blocks on that path have not allocated
276 */
277 if (k > 0) {
278 /* right now we don't handle cross boundary allocation */
279 if (blks < blocks_to_boundary + 1)
280 count += blks;
281 else
282 count += blocks_to_boundary + 1;
283 return count;
284 }
285
286 count++;
287 while (count < blks && count <= blocks_to_boundary &&
288 le32_to_cpu(*(branch[0].p + count)) == 0) {
289 count++;
290 }
291 return count;
292}
293
dae1e52c
AG
294/**
295 * ext4_alloc_branch - allocate and set up a chain of blocks.
296 * @handle: handle for this transaction
297 * @inode: owner
298 * @indirect_blks: number of allocated indirect blocks
299 * @blks: number of allocated direct blocks
300 * @goal: preferred place for allocation
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 */
a5211002
TT
321static int ext4_alloc_branch(handle_t *handle,
322 struct ext4_allocation_request *ar,
323 int indirect_blks, ext4_lblk_t *offsets,
324 Indirect *branch)
dae1e52c 325{
781f143e
TT
326 struct buffer_head * bh;
327 ext4_fsblk_t b, new_blocks[4];
328 __le32 *p;
329 int i, j, err, len = 1;
dae1e52c 330
781f143e
TT
331 for (i = 0; i <= indirect_blks; i++) {
332 if (i == indirect_blks) {
a5211002 333 new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
781f143e 334 } else
a5211002 335 ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
e3cf5d5d
TT
336 ar->inode, ar->goal,
337 ar->flags & EXT4_MB_DELALLOC_RESERVED,
338 NULL, &err);
781f143e
TT
339 if (err) {
340 i--;
341 goto failed;
342 }
343 branch[i].key = cpu_to_le32(new_blocks[i]);
344 if (i == 0)
345 continue;
346
a5211002 347 bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
dae1e52c 348 if (unlikely(!bh)) {
860d21e2 349 err = -ENOMEM;
dae1e52c
AG
350 goto failed;
351 }
dae1e52c
AG
352 lock_buffer(bh);
353 BUFFER_TRACE(bh, "call get_create_access");
354 err = ext4_journal_get_create_access(handle, bh);
355 if (err) {
dae1e52c
AG
356 unlock_buffer(bh);
357 goto failed;
358 }
359
781f143e
TT
360 memset(bh->b_data, 0, bh->b_size);
361 p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
362 b = new_blocks[i];
363
364 if (i == indirect_blks)
a5211002 365 len = ar->len;
781f143e
TT
366 for (j = 0; j < len; j++)
367 *p++ = cpu_to_le32(b++);
368
dae1e52c
AG
369 BUFFER_TRACE(bh, "marking uptodate");
370 set_buffer_uptodate(bh);
371 unlock_buffer(bh);
372
373 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
a5211002 374 err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
dae1e52c
AG
375 if (err)
376 goto failed;
377 }
781f143e 378 return 0;
dae1e52c 379failed:
781f143e 380 for (; i >= 0; i--) {
c5c7b8dd
JK
381 /*
382 * We want to ext4_forget() only freshly allocated indirect
383 * blocks. Buffer for new_blocks[i-1] is at branch[i].bh and
384 * buffer at branch[0].bh is indirect block / inode already
385 * existing before ext4_alloc_branch() was called.
386 */
387 if (i > 0 && i != indirect_blks && branch[i].bh)
a5211002 388 ext4_forget(handle, 1, ar->inode, branch[i].bh,
781f143e 389 branch[i].bh->b_blocknr);
a5211002
TT
390 ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
391 (i == indirect_blks) ? ar->len : 1, 0);
dae1e52c 392 }
dae1e52c
AG
393 return err;
394}
395
396/**
397 * ext4_splice_branch - splice the allocated branch onto inode.
398 * @handle: handle for this transaction
399 * @inode: owner
400 * @block: (logical) number of block we are adding
401 * @chain: chain of indirect blocks (with a missing link - see
402 * ext4_alloc_branch)
403 * @where: location of missing link
404 * @num: number of indirect blocks we are adding
405 * @blks: number of direct blocks we are adding
406 *
407 * This function fills the missing link and does all housekeeping needed in
408 * inode (->i_blocks, etc.). In case of success we end up with the full
409 * chain to new block and return 0.
410 */
a5211002
TT
411static int ext4_splice_branch(handle_t *handle,
412 struct ext4_allocation_request *ar,
413 Indirect *where, int num)
dae1e52c
AG
414{
415 int i;
416 int err = 0;
417 ext4_fsblk_t current_block;
418
419 /*
420 * If we're splicing into a [td]indirect block (as opposed to the
421 * inode) then we need to get write access to the [td]indirect block
422 * before the splice.
423 */
424 if (where->bh) {
425 BUFFER_TRACE(where->bh, "get_write_access");
426 err = ext4_journal_get_write_access(handle, where->bh);
427 if (err)
428 goto err_out;
429 }
430 /* That's it */
431
432 *where->p = where->key;
433
434 /*
435 * Update the host buffer_head or inode to point to more just allocated
436 * direct blocks blocks
437 */
a5211002 438 if (num == 0 && ar->len > 1) {
dae1e52c 439 current_block = le32_to_cpu(where->key) + 1;
a5211002 440 for (i = 1; i < ar->len; i++)
dae1e52c
AG
441 *(where->p + i) = cpu_to_le32(current_block++);
442 }
443
444 /* We are done with atomic stuff, now do the rest of housekeeping */
445 /* had we spliced it onto indirect block? */
446 if (where->bh) {
447 /*
448 * If we spliced it onto an indirect block, we haven't
449 * altered the inode. Note however that if it is being spliced
450 * onto an indirect block at the very end of the file (the
451 * file is growing) then we *will* alter the inode to reflect
452 * the new i_size. But that is not done here - it is done in
453 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
454 */
455 jbd_debug(5, "splicing indirect only\n");
456 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
a5211002 457 err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
dae1e52c
AG
458 if (err)
459 goto err_out;
460 } else {
461 /*
462 * OK, we spliced it into the inode itself on a direct block.
463 */
a5211002 464 ext4_mark_inode_dirty(handle, ar->inode);
dae1e52c
AG
465 jbd_debug(5, "splicing direct\n");
466 }
467 return err;
468
469err_out:
470 for (i = 1; i <= num; i++) {
471 /*
472 * branch[i].bh is newly allocated, so there is no
473 * need to revoke the block, which is why we don't
474 * need to set EXT4_FREE_BLOCKS_METADATA.
475 */
a5211002 476 ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
dae1e52c
AG
477 EXT4_FREE_BLOCKS_FORGET);
478 }
a5211002
TT
479 ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
480 ar->len, 0);
dae1e52c
AG
481
482 return err;
483}
484
485/*
486 * The ext4_ind_map_blocks() function handles non-extents inodes
487 * (i.e., using the traditional indirect/double-indirect i_blocks
488 * scheme) for ext4_map_blocks().
489 *
490 * Allocation strategy is simple: if we have to allocate something, we will
491 * have to go the whole way to leaf. So let's do it before attaching anything
492 * to tree, set linkage between the newborn blocks, write them if sync is
493 * required, recheck the path, free and repeat if check fails, otherwise
494 * set the last missing link (that will protect us from any truncate-generated
495 * removals - all blocks on the path are immune now) and possibly force the
496 * write on the parent block.
497 * That has a nice additional property: no special recovery from the failed
498 * allocations is needed - we simply release blocks and do not touch anything
499 * reachable from inode.
500 *
501 * `handle' can be NULL if create == 0.
502 *
503 * return > 0, # of blocks mapped or allocated.
504 * return = 0, if plain lookup failed.
505 * return < 0, error case.
506 *
507 * The ext4_ind_get_blocks() function should be called with
508 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
509 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
510 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
511 * blocks.
512 */
513int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
514 struct ext4_map_blocks *map,
515 int flags)
516{
a5211002 517 struct ext4_allocation_request ar;
dae1e52c
AG
518 int err = -EIO;
519 ext4_lblk_t offsets[4];
520 Indirect chain[4];
521 Indirect *partial;
dae1e52c
AG
522 int indirect_blks;
523 int blocks_to_boundary = 0;
524 int depth;
525 int count = 0;
526 ext4_fsblk_t first_block = 0;
527
528 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
529 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
530 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
531 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
532 &blocks_to_boundary);
533
534 if (depth == 0)
535 goto out;
536
537 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
538
539 /* Simplest case - block found, no allocation needed */
540 if (!partial) {
541 first_block = le32_to_cpu(chain[depth - 1].key);
542 count++;
543 /*map more blocks*/
544 while (count < map->m_len && count <= blocks_to_boundary) {
545 ext4_fsblk_t blk;
546
547 blk = le32_to_cpu(*(chain[depth-1].p + count));
548
549 if (blk == first_block + count)
550 count++;
551 else
552 break;
553 }
554 goto got_it;
555 }
556
557 /* Next simple case - plain lookup or failed read of indirect block */
558 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
559 goto cleanup;
560
561 /*
562 * Okay, we need to do block allocation.
563 */
bab08ab9
TT
564 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
565 EXT4_FEATURE_RO_COMPAT_BIGALLOC)) {
566 EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
567 "non-extent mapped inodes with bigalloc");
568 return -ENOSPC;
569 }
570
a5211002
TT
571 /* Set up for the direct block allocation */
572 memset(&ar, 0, sizeof(ar));
573 ar.inode = inode;
574 ar.logical = map->m_lblk;
575 if (S_ISREG(inode->i_mode))
576 ar.flags = EXT4_MB_HINT_DATA;
e3cf5d5d
TT
577 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
578 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
a5211002
TT
579
580 ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
dae1e52c
AG
581
582 /* the number of blocks need to allocate for [d,t]indirect blocks */
583 indirect_blks = (chain + depth) - partial - 1;
584
585 /*
586 * Next look up the indirect map to count the totoal number of
587 * direct blocks to allocate for this branch.
588 */
a5211002
TT
589 ar.len = ext4_blks_to_allocate(partial, indirect_blks,
590 map->m_len, blocks_to_boundary);
591
dae1e52c
AG
592 /*
593 * Block out ext4_truncate while we alter the tree
594 */
a5211002 595 err = ext4_alloc_branch(handle, &ar, indirect_blks,
dae1e52c
AG
596 offsets + (partial - chain), partial);
597
598 /*
599 * The ext4_splice_branch call will free and forget any buffers
600 * on the new chain if there is a failure, but that risks using
601 * up transaction credits, especially for bitmaps where the
602 * credits cannot be returned. Can we handle this somehow? We
603 * may need to return -EAGAIN upwards in the worst case. --sct
604 */
605 if (!err)
a5211002 606 err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
dae1e52c
AG
607 if (err)
608 goto cleanup;
609
610 map->m_flags |= EXT4_MAP_NEW;
611
612 ext4_update_inode_fsync_trans(handle, inode, 1);
a5211002 613 count = ar.len;
dae1e52c
AG
614got_it:
615 map->m_flags |= EXT4_MAP_MAPPED;
616 map->m_pblk = le32_to_cpu(chain[depth-1].key);
617 map->m_len = count;
618 if (count > blocks_to_boundary)
619 map->m_flags |= EXT4_MAP_BOUNDARY;
620 err = count;
621 /* Clean up and exit */
622 partial = chain + depth - 1; /* the whole chain */
623cleanup:
624 while (partial > chain) {
625 BUFFER_TRACE(partial->bh, "call brelse");
626 brelse(partial->bh);
627 partial--;
628 }
629out:
21ddd568 630 trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
dae1e52c
AG
631 return err;
632}
633
634/*
635 * O_DIRECT for ext3 (or indirect map) based files
636 *
637 * If the O_DIRECT write will extend the file then add this inode to the
638 * orphan list. So recovery will truncate it back to the original size
639 * if the machine crashes during the write.
640 *
641 * If the O_DIRECT write is intantiating holes inside i_size and the machine
642 * crashes then stale disk data _may_ be exposed inside the file. But current
643 * VFS code falls back into buffered path in that case so we are safe.
644 */
645ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
16b1f05d 646 struct iov_iter *iter, loff_t offset)
dae1e52c
AG
647{
648 struct file *file = iocb->ki_filp;
649 struct inode *inode = file->f_mapping->host;
650 struct ext4_inode_info *ei = EXT4_I(inode);
651 handle_t *handle;
652 ssize_t ret;
653 int orphan = 0;
a6cbcd4a 654 size_t count = iov_iter_count(iter);
dae1e52c
AG
655 int retries = 0;
656
657 if (rw == WRITE) {
658 loff_t final_size = offset + count;
659
660 if (final_size > inode->i_size) {
661 /* Credits for sb + inode write */
9924a92a 662 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
dae1e52c
AG
663 if (IS_ERR(handle)) {
664 ret = PTR_ERR(handle);
665 goto out;
666 }
667 ret = ext4_orphan_add(handle, inode);
668 if (ret) {
669 ext4_journal_stop(handle);
670 goto out;
671 }
672 orphan = 1;
673 ei->i_disksize = inode->i_size;
674 ext4_journal_stop(handle);
675 }
676 }
677
678retry:
dccaf33f 679 if (rw == READ && ext4_should_dioread_nolock(inode)) {
17335dcc
DM
680 /*
681 * Nolock dioread optimization may be dynamically disabled
682 * via ext4_inode_block_unlocked_dio(). Check inode's state
683 * while holding extra i_dio_count ref.
684 */
685 atomic_inc(&inode->i_dio_count);
686 smp_mb();
687 if (unlikely(ext4_test_inode_state(inode,
688 EXT4_STATE_DIOREAD_LOCK))) {
689 inode_dio_done(inode);
690 goto locked;
691 }
923ae0ff
RZ
692 if (IS_DAX(inode))
693 ret = dax_do_io(rw, iocb, inode, iter, offset,
694 ext4_get_block, NULL, 0);
695 else
17f8c842
OS
696 ret = __blockdev_direct_IO(iocb, inode,
697 inode->i_sb->s_bdev, iter,
698 offset, ext4_get_block, NULL,
699 NULL, 0);
17335dcc 700 inode_dio_done(inode);
dccaf33f 701 } else {
17335dcc 702locked:
923ae0ff
RZ
703 if (IS_DAX(inode))
704 ret = dax_do_io(rw, iocb, inode, iter, offset,
705 ext4_get_block, NULL, DIO_LOCKING);
706 else
17f8c842
OS
707 ret = blockdev_direct_IO(iocb, inode, iter, offset,
708 ext4_get_block);
dae1e52c
AG
709
710 if (unlikely((rw & WRITE) && ret < 0)) {
711 loff_t isize = i_size_read(inode);
16b1f05d 712 loff_t end = offset + count;
dae1e52c
AG
713
714 if (end > isize)
715 ext4_truncate_failed_write(inode);
716 }
717 }
718 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
719 goto retry;
720
721 if (orphan) {
722 int err;
723
724 /* Credits for sb + inode write */
9924a92a 725 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
dae1e52c
AG
726 if (IS_ERR(handle)) {
727 /* This is really bad luck. We've written the data
728 * but cannot extend i_size. Bail out and pretend
729 * the write failed... */
730 ret = PTR_ERR(handle);
731 if (inode->i_nlink)
732 ext4_orphan_del(NULL, inode);
733
734 goto out;
735 }
736 if (inode->i_nlink)
737 ext4_orphan_del(handle, inode);
738 if (ret > 0) {
739 loff_t end = offset + ret;
740 if (end > inode->i_size) {
741 ei->i_disksize = end;
742 i_size_write(inode, end);
743 /*
744 * We're going to return a positive `ret'
745 * here due to non-zero-length I/O, so there's
746 * no way of reporting error returns from
747 * ext4_mark_inode_dirty() to userspace. So
748 * ignore it.
749 */
750 ext4_mark_inode_dirty(handle, inode);
751 }
752 }
753 err = ext4_journal_stop(handle);
754 if (ret == 0)
755 ret = err;
756 }
757out:
758 return ret;
759}
760
761/*
762 * Calculate the number of metadata blocks need to reserve
763 * to allocate a new block at @lblocks for non extent file based file
764 */
765int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
766{
767 struct ext4_inode_info *ei = EXT4_I(inode);
768 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
769 int blk_bits;
770
771 if (lblock < EXT4_NDIR_BLOCKS)
772 return 0;
773
774 lblock -= EXT4_NDIR_BLOCKS;
775
776 if (ei->i_da_metadata_calc_len &&
777 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
778 ei->i_da_metadata_calc_len++;
779 return 0;
780 }
781 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
782 ei->i_da_metadata_calc_len = 1;
783 blk_bits = order_base_2(lblock);
784 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
785}
786
fa55a0ed
JK
787/*
788 * Calculate number of indirect blocks touched by mapping @nrblocks logically
789 * contiguous blocks
790 */
791int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
dae1e52c 792{
dae1e52c 793 /*
fa55a0ed
JK
794 * With N contiguous data blocks, we need at most
795 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
796 * 2 dindirect blocks, and 1 tindirect block
dae1e52c 797 */
fa55a0ed 798 return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
dae1e52c
AG
799}
800
801/*
802 * Truncate transactions can be complex and absolutely huge. So we need to
803 * be able to restart the transaction at a conventient checkpoint to make
804 * sure we don't overflow the journal.
805 *
819c4920 806 * Try to extend this transaction for the purposes of truncation. If
dae1e52c
AG
807 * extend fails, we need to propagate the failure up and restart the
808 * transaction in the top-level truncate loop. --sct
dae1e52c
AG
809 *
810 * Returns 0 if we managed to create more room. If we can't create more
811 * room, and the transaction must be restarted we return 1.
812 */
813static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
814{
815 if (!ext4_handle_valid(handle))
816 return 0;
817 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
818 return 0;
819 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
820 return 0;
821 return 1;
822}
823
824/*
825 * Probably it should be a library function... search for first non-zero word
826 * or memcmp with zero_page, whatever is better for particular architecture.
827 * Linus?
828 */
829static inline int all_zeroes(__le32 *p, __le32 *q)
830{
831 while (p < q)
832 if (*p++)
833 return 0;
834 return 1;
835}
836
837/**
838 * ext4_find_shared - find the indirect blocks for partial truncation.
839 * @inode: inode in question
840 * @depth: depth of the affected branch
841 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
842 * @chain: place to store the pointers to partial indirect blocks
843 * @top: place to the (detached) top of branch
844 *
845 * This is a helper function used by ext4_truncate().
846 *
847 * When we do truncate() we may have to clean the ends of several
848 * indirect blocks but leave the blocks themselves alive. Block is
849 * partially truncated if some data below the new i_size is referred
850 * from it (and it is on the path to the first completely truncated
851 * data block, indeed). We have to free the top of that path along
852 * with everything to the right of the path. Since no allocation
853 * past the truncation point is possible until ext4_truncate()
854 * finishes, we may safely do the latter, but top of branch may
855 * require special attention - pageout below the truncation point
856 * might try to populate it.
857 *
858 * We atomically detach the top of branch from the tree, store the
859 * block number of its root in *@top, pointers to buffer_heads of
860 * partially truncated blocks - in @chain[].bh and pointers to
861 * their last elements that should not be removed - in
862 * @chain[].p. Return value is the pointer to last filled element
863 * of @chain.
864 *
865 * The work left to caller to do the actual freeing of subtrees:
866 * a) free the subtree starting from *@top
867 * b) free the subtrees whose roots are stored in
868 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
869 * c) free the subtrees growing from the inode past the @chain[0].
870 * (no partially truncated stuff there). */
871
872static Indirect *ext4_find_shared(struct inode *inode, int depth,
873 ext4_lblk_t offsets[4], Indirect chain[4],
874 __le32 *top)
875{
876 Indirect *partial, *p;
877 int k, err;
878
879 *top = 0;
880 /* Make k index the deepest non-null offset + 1 */
881 for (k = depth; k > 1 && !offsets[k-1]; k--)
882 ;
883 partial = ext4_get_branch(inode, k, offsets, chain, &err);
884 /* Writer: pointers */
885 if (!partial)
886 partial = chain + k-1;
887 /*
888 * If the branch acquired continuation since we've looked at it -
889 * fine, it should all survive and (new) top doesn't belong to us.
890 */
891 if (!partial->key && *partial->p)
892 /* Writer: end */
893 goto no_top;
894 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
895 ;
896 /*
897 * OK, we've found the last block that must survive. The rest of our
898 * branch should be detached before unlocking. However, if that rest
899 * of branch is all ours and does not grow immediately from the inode
900 * it's easier to cheat and just decrement partial->p.
901 */
902 if (p == chain + k - 1 && p > chain) {
903 p->p--;
904 } else {
905 *top = *p->p;
906 /* Nope, don't do this in ext4. Must leave the tree intact */
907#if 0
908 *p->p = 0;
909#endif
910 }
911 /* Writer: end */
912
913 while (partial > p) {
914 brelse(partial->bh);
915 partial--;
916 }
917no_top:
918 return partial;
919}
920
921/*
922 * Zero a number of block pointers in either an inode or an indirect block.
923 * If we restart the transaction we must again get write access to the
924 * indirect block for further modification.
925 *
926 * We release `count' blocks on disk, but (last - first) may be greater
927 * than `count' because there can be holes in there.
928 *
929 * Return 0 on success, 1 on invalid block range
930 * and < 0 on fatal error.
931 */
932static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
933 struct buffer_head *bh,
934 ext4_fsblk_t block_to_free,
935 unsigned long count, __le32 *first,
936 __le32 *last)
937{
938 __le32 *p;
981250ca 939 int flags = EXT4_FREE_BLOCKS_VALIDATED;
dae1e52c
AG
940 int err;
941
942 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
981250ca
TT
943 flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
944 else if (ext4_should_journal_data(inode))
945 flags |= EXT4_FREE_BLOCKS_FORGET;
dae1e52c
AG
946
947 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
948 count)) {
949 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
950 "blocks %llu len %lu",
951 (unsigned long long) block_to_free, count);
952 return 1;
953 }
954
955 if (try_to_extend_transaction(handle, inode)) {
956 if (bh) {
957 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
958 err = ext4_handle_dirty_metadata(handle, inode, bh);
959 if (unlikely(err))
960 goto out_err;
961 }
962 err = ext4_mark_inode_dirty(handle, inode);
963 if (unlikely(err))
964 goto out_err;
965 err = ext4_truncate_restart_trans(handle, inode,
966 ext4_blocks_for_truncate(inode));
967 if (unlikely(err))
968 goto out_err;
969 if (bh) {
970 BUFFER_TRACE(bh, "retaking write access");
971 err = ext4_journal_get_write_access(handle, bh);
972 if (unlikely(err))
973 goto out_err;
974 }
975 }
976
977 for (p = first; p < last; p++)
978 *p = 0;
979
980 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
981 return 0;
982out_err:
983 ext4_std_error(inode->i_sb, err);
984 return err;
985}
986
987/**
988 * ext4_free_data - free a list of data blocks
989 * @handle: handle for this transaction
990 * @inode: inode we are dealing with
991 * @this_bh: indirect buffer_head which contains *@first and *@last
992 * @first: array of block numbers
993 * @last: points immediately past the end of array
994 *
995 * We are freeing all blocks referred from that array (numbers are stored as
996 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
997 *
998 * We accumulate contiguous runs of blocks to free. Conveniently, if these
999 * blocks are contiguous then releasing them at one time will only affect one
1000 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1001 * actually use a lot of journal space.
1002 *
1003 * @this_bh will be %NULL if @first and @last point into the inode's direct
1004 * block pointers.
1005 */
1006static void ext4_free_data(handle_t *handle, struct inode *inode,
1007 struct buffer_head *this_bh,
1008 __le32 *first, __le32 *last)
1009{
1010 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
1011 unsigned long count = 0; /* Number of blocks in the run */
1012 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
1013 corresponding to
1014 block_to_free */
1015 ext4_fsblk_t nr; /* Current block # */
1016 __le32 *p; /* Pointer into inode/ind
1017 for current block */
1018 int err = 0;
1019
1020 if (this_bh) { /* For indirect block */
1021 BUFFER_TRACE(this_bh, "get_write_access");
1022 err = ext4_journal_get_write_access(handle, this_bh);
1023 /* Important: if we can't update the indirect pointers
1024 * to the blocks, we can't free them. */
1025 if (err)
1026 return;
1027 }
1028
1029 for (p = first; p < last; p++) {
1030 nr = le32_to_cpu(*p);
1031 if (nr) {
1032 /* accumulate blocks to free if they're contiguous */
1033 if (count == 0) {
1034 block_to_free = nr;
1035 block_to_free_p = p;
1036 count = 1;
1037 } else if (nr == block_to_free + count) {
1038 count++;
1039 } else {
1040 err = ext4_clear_blocks(handle, inode, this_bh,
1041 block_to_free, count,
1042 block_to_free_p, p);
1043 if (err)
1044 break;
1045 block_to_free = nr;
1046 block_to_free_p = p;
1047 count = 1;
1048 }
1049 }
1050 }
1051
1052 if (!err && count > 0)
1053 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1054 count, block_to_free_p, p);
1055 if (err < 0)
1056 /* fatal error */
1057 return;
1058
1059 if (this_bh) {
1060 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1061
1062 /*
1063 * The buffer head should have an attached journal head at this
1064 * point. However, if the data is corrupted and an indirect
1065 * block pointed to itself, it would have been detached when
1066 * the block was cleared. Check for this instead of OOPSing.
1067 */
1068 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1069 ext4_handle_dirty_metadata(handle, inode, this_bh);
1070 else
1071 EXT4_ERROR_INODE(inode,
1072 "circular indirect block detected at "
1073 "block %llu",
1074 (unsigned long long) this_bh->b_blocknr);
1075 }
1076}
1077
1078/**
1079 * ext4_free_branches - free an array of branches
1080 * @handle: JBD handle for this transaction
1081 * @inode: inode we are dealing with
1082 * @parent_bh: the buffer_head which contains *@first and *@last
1083 * @first: array of block numbers
1084 * @last: pointer immediately past the end of array
1085 * @depth: depth of the branches to free
1086 *
1087 * We are freeing all blocks referred from these branches (numbers are
1088 * stored as little-endian 32-bit) and updating @inode->i_blocks
1089 * appropriately.
1090 */
1091static void ext4_free_branches(handle_t *handle, struct inode *inode,
1092 struct buffer_head *parent_bh,
1093 __le32 *first, __le32 *last, int depth)
1094{
1095 ext4_fsblk_t nr;
1096 __le32 *p;
1097
1098 if (ext4_handle_is_aborted(handle))
1099 return;
1100
1101 if (depth--) {
1102 struct buffer_head *bh;
1103 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1104 p = last;
1105 while (--p >= first) {
1106 nr = le32_to_cpu(*p);
1107 if (!nr)
1108 continue; /* A hole */
1109
1110 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1111 nr, 1)) {
1112 EXT4_ERROR_INODE(inode,
1113 "invalid indirect mapped "
1114 "block %lu (level %d)",
1115 (unsigned long) nr, depth);
1116 break;
1117 }
1118
1119 /* Go read the buffer for the next level down */
1120 bh = sb_bread(inode->i_sb, nr);
1121
1122 /*
1123 * A read failure? Report error and clear slot
1124 * (should be rare).
1125 */
1126 if (!bh) {
1127 EXT4_ERROR_INODE_BLOCK(inode, nr,
1128 "Read failure");
1129 continue;
1130 }
1131
1132 /* This zaps the entire block. Bottom up. */
1133 BUFFER_TRACE(bh, "free child branches");
1134 ext4_free_branches(handle, inode, bh,
1135 (__le32 *) bh->b_data,
1136 (__le32 *) bh->b_data + addr_per_block,
1137 depth);
1138 brelse(bh);
1139
1140 /*
1141 * Everything below this this pointer has been
1142 * released. Now let this top-of-subtree go.
1143 *
1144 * We want the freeing of this indirect block to be
1145 * atomic in the journal with the updating of the
1146 * bitmap block which owns it. So make some room in
1147 * the journal.
1148 *
1149 * We zero the parent pointer *after* freeing its
1150 * pointee in the bitmaps, so if extend_transaction()
1151 * for some reason fails to put the bitmap changes and
1152 * the release into the same transaction, recovery
1153 * will merely complain about releasing a free block,
1154 * rather than leaking blocks.
1155 */
1156 if (ext4_handle_is_aborted(handle))
1157 return;
1158 if (try_to_extend_transaction(handle, inode)) {
1159 ext4_mark_inode_dirty(handle, inode);
1160 ext4_truncate_restart_trans(handle, inode,
1161 ext4_blocks_for_truncate(inode));
1162 }
1163
1164 /*
1165 * The forget flag here is critical because if
1166 * we are journaling (and not doing data
1167 * journaling), we have to make sure a revoke
1168 * record is written to prevent the journal
1169 * replay from overwriting the (former)
1170 * indirect block if it gets reallocated as a
1171 * data block. This must happen in the same
1172 * transaction where the data blocks are
1173 * actually freed.
1174 */
1175 ext4_free_blocks(handle, inode, NULL, nr, 1,
1176 EXT4_FREE_BLOCKS_METADATA|
1177 EXT4_FREE_BLOCKS_FORGET);
1178
1179 if (parent_bh) {
1180 /*
1181 * The block which we have just freed is
1182 * pointed to by an indirect block: journal it
1183 */
1184 BUFFER_TRACE(parent_bh, "get_write_access");
1185 if (!ext4_journal_get_write_access(handle,
1186 parent_bh)){
1187 *p = 0;
1188 BUFFER_TRACE(parent_bh,
1189 "call ext4_handle_dirty_metadata");
1190 ext4_handle_dirty_metadata(handle,
1191 inode,
1192 parent_bh);
1193 }
1194 }
1195 }
1196 } else {
1197 /* We have reached the bottom of the tree. */
1198 BUFFER_TRACE(parent_bh, "free data blocks");
1199 ext4_free_data(handle, inode, parent_bh, first, last);
1200 }
1201}
1202
819c4920 1203void ext4_ind_truncate(handle_t *handle, struct inode *inode)
dae1e52c 1204{
dae1e52c
AG
1205 struct ext4_inode_info *ei = EXT4_I(inode);
1206 __le32 *i_data = ei->i_data;
1207 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
dae1e52c
AG
1208 ext4_lblk_t offsets[4];
1209 Indirect chain[4];
1210 Indirect *partial;
1211 __le32 nr = 0;
1212 int n = 0;
1213 ext4_lblk_t last_block, max_block;
1214 unsigned blocksize = inode->i_sb->s_blocksize;
dae1e52c
AG
1215
1216 last_block = (inode->i_size + blocksize-1)
1217 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1218 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1219 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1220
dae1e52c
AG
1221 if (last_block != max_block) {
1222 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1223 if (n == 0)
819c4920 1224 return;
dae1e52c
AG
1225 }
1226
51865fda 1227 ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
dae1e52c
AG
1228
1229 /*
1230 * The orphan list entry will now protect us from any crash which
1231 * occurs before the truncate completes, so it is now safe to propagate
1232 * the new, shorter inode size (held for now in i_size) into the
1233 * on-disk inode. We do this via i_disksize, which is the value which
1234 * ext4 *really* writes onto the disk inode.
1235 */
1236 ei->i_disksize = inode->i_size;
1237
1238 if (last_block == max_block) {
1239 /*
1240 * It is unnecessary to free any data blocks if last_block is
1241 * equal to the indirect block limit.
1242 */
819c4920 1243 return;
dae1e52c
AG
1244 } else if (n == 1) { /* direct blocks */
1245 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1246 i_data + EXT4_NDIR_BLOCKS);
1247 goto do_indirects;
1248 }
1249
1250 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1251 /* Kill the top of shared branch (not detached) */
1252 if (nr) {
1253 if (partial == chain) {
1254 /* Shared branch grows from the inode */
1255 ext4_free_branches(handle, inode, NULL,
1256 &nr, &nr+1, (chain+n-1) - partial);
1257 *partial->p = 0;
1258 /*
1259 * We mark the inode dirty prior to restart,
1260 * and prior to stop. No need for it here.
1261 */
1262 } else {
1263 /* Shared branch grows from an indirect block */
1264 BUFFER_TRACE(partial->bh, "get_write_access");
1265 ext4_free_branches(handle, inode, partial->bh,
1266 partial->p,
1267 partial->p+1, (chain+n-1) - partial);
1268 }
1269 }
1270 /* Clear the ends of indirect blocks on the shared branch */
1271 while (partial > chain) {
1272 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1273 (__le32*)partial->bh->b_data+addr_per_block,
1274 (chain+n-1) - partial);
1275 BUFFER_TRACE(partial->bh, "call brelse");
1276 brelse(partial->bh);
1277 partial--;
1278 }
1279do_indirects:
1280 /* Kill the remaining (whole) subtrees */
1281 switch (offsets[0]) {
1282 default:
1283 nr = i_data[EXT4_IND_BLOCK];
1284 if (nr) {
1285 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1286 i_data[EXT4_IND_BLOCK] = 0;
1287 }
1288 case EXT4_IND_BLOCK:
1289 nr = i_data[EXT4_DIND_BLOCK];
1290 if (nr) {
1291 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1292 i_data[EXT4_DIND_BLOCK] = 0;
1293 }
1294 case EXT4_DIND_BLOCK:
1295 nr = i_data[EXT4_TIND_BLOCK];
1296 if (nr) {
1297 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1298 i_data[EXT4_TIND_BLOCK] = 0;
1299 }
1300 case EXT4_TIND_BLOCK:
1301 ;
1302 }
dae1e52c
AG
1303}
1304
4f579ae7
LC
1305/**
1306 * ext4_ind_remove_space - remove space from the range
1307 * @handle: JBD handle for this transaction
1308 * @inode: inode we are dealing with
1309 * @start: First block to remove
1310 * @end: One block after the last block to remove (exclusive)
1311 *
1312 * Free the blocks in the defined range (end is exclusive endpoint of
1313 * range). This is used by ext4_punch_hole().
1314 */
1315int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1316 ext4_lblk_t start, ext4_lblk_t end)
8bad6fc8 1317{
4f579ae7
LC
1318 struct ext4_inode_info *ei = EXT4_I(inode);
1319 __le32 *i_data = ei->i_data;
8bad6fc8 1320 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4f579ae7
LC
1321 ext4_lblk_t offsets[4], offsets2[4];
1322 Indirect chain[4], chain2[4];
1323 Indirect *partial, *partial2;
1324 ext4_lblk_t max_block;
1325 __le32 nr = 0, nr2 = 0;
1326 int n = 0, n2 = 0;
1327 unsigned blocksize = inode->i_sb->s_blocksize;
a93cd4cf 1328
4f579ae7
LC
1329 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1330 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1331 if (end >= max_block)
1332 end = max_block;
1333 if ((start >= end) || (start > max_block))
1334 return 0;
1335
1336 n = ext4_block_to_path(inode, start, offsets, NULL);
1337 n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1338
1339 BUG_ON(n > n2);
1340
1341 if ((n == 1) && (n == n2)) {
1342 /* We're punching only within direct block range */
1343 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1344 i_data + offsets2[0]);
1345 return 0;
1346 } else if (n2 > n) {
1347 /*
1348 * Start and end are on a different levels so we're going to
1349 * free partial block at start, and partial block at end of
1350 * the range. If there are some levels in between then
1351 * do_indirects label will take care of that.
1352 */
1353
1354 if (n == 1) {
1355 /*
1356 * Start is at the direct block level, free
1357 * everything to the end of the level.
1358 */
1359 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1360 i_data + EXT4_NDIR_BLOCKS);
1361 goto end_range;
1362 }
1363
1364
1365 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1366 if (nr) {
1367 if (partial == chain) {
1368 /* Shared branch grows from the inode */
1369 ext4_free_branches(handle, inode, NULL,
1370 &nr, &nr+1, (chain+n-1) - partial);
1371 *partial->p = 0;
a93cd4cf 1372 } else {
4f579ae7
LC
1373 /* Shared branch grows from an indirect block */
1374 BUFFER_TRACE(partial->bh, "get_write_access");
1375 ext4_free_branches(handle, inode, partial->bh,
1376 partial->p,
1377 partial->p+1, (chain+n-1) - partial);
a93cd4cf 1378 }
4f579ae7
LC
1379 }
1380
1381 /*
1382 * Clear the ends of indirect blocks on the shared branch
1383 * at the start of the range
1384 */
1385 while (partial > chain) {
1386 ext4_free_branches(handle, inode, partial->bh,
1387 partial->p + 1,
1388 (__le32 *)partial->bh->b_data+addr_per_block,
1389 (chain+n-1) - partial);
1390 BUFFER_TRACE(partial->bh, "call brelse");
1391 brelse(partial->bh);
1392 partial--;
1393 }
1394
1395end_range:
1396 partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1397 if (nr2) {
1398 if (partial2 == chain2) {
1399 /*
1400 * Remember, end is exclusive so here we're at
1401 * the start of the next level we're not going
1402 * to free. Everything was covered by the start
1403 * of the range.
1404 */
6f30b7e3 1405 goto do_indirects;
8bad6fc8 1406 }
4f579ae7
LC
1407 } else {
1408 /*
1409 * ext4_find_shared returns Indirect structure which
1410 * points to the last element which should not be
1411 * removed by truncate. But this is end of the range
1412 * in punch_hole so we need to point to the next element
1413 */
1414 partial2->p++;
8bad6fc8 1415 }
4f579ae7
LC
1416
1417 /*
1418 * Clear the ends of indirect blocks on the shared branch
1419 * at the end of the range
1420 */
1421 while (partial2 > chain2) {
1422 ext4_free_branches(handle, inode, partial2->bh,
1423 (__le32 *)partial2->bh->b_data,
1424 partial2->p,
1425 (chain2+n2-1) - partial2);
1426 BUFFER_TRACE(partial2->bh, "call brelse");
1427 brelse(partial2->bh);
1428 partial2--;
8bad6fc8 1429 }
4f579ae7 1430 goto do_indirects;
8bad6fc8
ZL
1431 }
1432
4f579ae7
LC
1433 /* Punch happened within the same level (n == n2) */
1434 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1435 partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
6f30b7e3
OS
1436
1437 /* Free top, but only if partial2 isn't its subtree. */
1438 if (nr) {
1439 int level = min(partial - chain, partial2 - chain2);
1440 int i;
1441 int subtree = 1;
1442
1443 for (i = 0; i <= level; i++) {
1444 if (offsets[i] != offsets2[i]) {
1445 subtree = 0;
1446 break;
1447 }
1448 }
1449
1450 if (!subtree) {
1451 if (partial == chain) {
1452 /* Shared branch grows from the inode */
1453 ext4_free_branches(handle, inode, NULL,
1454 &nr, &nr+1,
1455 (chain+n-1) - partial);
1456 *partial->p = 0;
1457 } else {
1458 /* Shared branch grows from an indirect block */
1459 BUFFER_TRACE(partial->bh, "get_write_access");
4f579ae7 1460 ext4_free_branches(handle, inode, partial->bh,
6f30b7e3
OS
1461 partial->p,
1462 partial->p+1,
4f579ae7 1463 (chain+n-1) - partial);
4f579ae7 1464 }
8bad6fc8 1465 }
6f30b7e3
OS
1466 }
1467
1468 if (!nr2) {
4f579ae7 1469 /*
6f30b7e3
OS
1470 * ext4_find_shared returns Indirect structure which
1471 * points to the last element which should not be
1472 * removed by truncate. But this is end of the range
1473 * in punch_hole so we need to point to the next element
4f579ae7 1474 */
6f30b7e3
OS
1475 partial2->p++;
1476 }
1477
1478 while (partial > chain || partial2 > chain2) {
1479 int depth = (chain+n-1) - partial;
1480 int depth2 = (chain2+n2-1) - partial2;
1481
1482 if (partial > chain && partial2 > chain2 &&
1483 partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1484 /*
1485 * We've converged on the same block. Clear the range,
1486 * then we're done.
1487 */
4f579ae7 1488 ext4_free_branches(handle, inode, partial->bh,
6f30b7e3
OS
1489 partial->p + 1,
1490 partial2->p,
1491 (chain+n-1) - partial);
4f579ae7
LC
1492 BUFFER_TRACE(partial->bh, "call brelse");
1493 brelse(partial->bh);
6f30b7e3
OS
1494 BUFFER_TRACE(partial2->bh, "call brelse");
1495 brelse(partial2->bh);
1496 return 0;
4f579ae7 1497 }
6f30b7e3 1498
4f579ae7 1499 /*
6f30b7e3
OS
1500 * The start and end partial branches may not be at the same
1501 * level even though the punch happened within one level. So, we
1502 * give them a chance to arrive at the same level, then walk
1503 * them in step with each other until we converge on the same
1504 * block.
4f579ae7 1505 */
6f30b7e3
OS
1506 if (partial > chain && depth <= depth2) {
1507 ext4_free_branches(handle, inode, partial->bh,
1508 partial->p + 1,
1509 (__le32 *)partial->bh->b_data+addr_per_block,
1510 (chain+n-1) - partial);
1511 BUFFER_TRACE(partial->bh, "call brelse");
1512 brelse(partial->bh);
1513 partial--;
1514 }
1515 if (partial2 > chain2 && depth2 <= depth) {
4f579ae7
LC
1516 ext4_free_branches(handle, inode, partial2->bh,
1517 (__le32 *)partial2->bh->b_data,
1518 partial2->p,
6f30b7e3 1519 (chain2+n2-1) - partial2);
4f579ae7
LC
1520 BUFFER_TRACE(partial2->bh, "call brelse");
1521 brelse(partial2->bh);
1522 partial2--;
8bad6fc8
ZL
1523 }
1524 }
6f30b7e3 1525 return 0;
8bad6fc8 1526
4f579ae7
LC
1527do_indirects:
1528 /* Kill the remaining (whole) subtrees */
1529 switch (offsets[0]) {
1530 default:
1531 if (++n >= n2)
1532 return 0;
1533 nr = i_data[EXT4_IND_BLOCK];
1534 if (nr) {
1535 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1536 i_data[EXT4_IND_BLOCK] = 0;
1537 }
1538 case EXT4_IND_BLOCK:
1539 if (++n >= n2)
1540 return 0;
1541 nr = i_data[EXT4_DIND_BLOCK];
1542 if (nr) {
1543 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1544 i_data[EXT4_DIND_BLOCK] = 0;
1545 }
1546 case EXT4_DIND_BLOCK:
1547 if (++n >= n2)
1548 return 0;
1549 nr = i_data[EXT4_TIND_BLOCK];
1550 if (nr) {
1551 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1552 i_data[EXT4_TIND_BLOCK] = 0;
1553 }
1554 case EXT4_TIND_BLOCK:
1555 ;
1556 }
1557 return 0;
8bad6fc8 1558}