blob: 45fe924f82bce2ff76e3e74b45ec1833729433ea [file] [log] [blame]
Amir Goldsteindae1e522011-06-27 19:40:50 -04001/*
2 * linux/fs/ext4/indirect.c
3 *
4 * from
5 *
6 * linux/fs/ext4/inode.c
7 *
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
12 *
13 * from
14 *
15 * linux/fs/minix/inode.c
16 *
17 * Copyright (C) 1991, 1992 Linus Torvalds
18 *
19 * Goal-directed block allocation by Stephen Tweedie
20 * (sct@redhat.com), 1993, 1998
21 */
22
Kent Overstreeta27bb332013-05-07 16:19:08 -070023#include <linux/aio.h>
Amir Goldsteindae1e522011-06-27 19:40:50 -040024#include "ext4_jbd2.h"
25#include "truncate.h"
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;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500149 int ret = -EIO;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400150
151 *err = 0;
152 /* i_data is not going away, no lock needed */
153 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
154 if (!p->key)
155 goto no_block;
156 while (--depth) {
157 bh = sb_getblk(sb, le32_to_cpu(p->key));
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500158 if (unlikely(!bh)) {
159 ret = -ENOMEM;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400160 goto failure;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500161 }
Amir Goldsteindae1e522011-06-27 19:40:50 -0400162
163 if (!bh_uptodate_or_lock(bh)) {
164 if (bh_submit_read(bh) < 0) {
165 put_bh(bh);
166 goto failure;
167 }
168 /* validate block references */
169 if (ext4_check_indirect_blockref(inode, bh)) {
170 put_bh(bh);
171 goto failure;
172 }
173 }
174
175 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
176 /* Reader: end */
177 if (!p->key)
178 goto no_block;
179 }
180 return NULL;
181
182failure:
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500183 *err = ret;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400184no_block:
185 return p;
186}
187
188/**
189 * ext4_find_near - find a place for allocation with sufficient locality
190 * @inode: owner
191 * @ind: descriptor of indirect block.
192 *
193 * This function returns the preferred place for block allocation.
194 * It is used when heuristic for sequential allocation fails.
195 * Rules are:
196 * + if there is a block to the left of our position - allocate near it.
197 * + if pointer will live in indirect block - allocate near that block.
198 * + if pointer will live in inode - allocate in the same
199 * cylinder group.
200 *
201 * In the latter case we colour the starting block by the callers PID to
202 * prevent it from clashing with concurrent allocations for a different inode
203 * in the same block group. The PID is used here so that functionally related
204 * files will be close-by on-disk.
205 *
206 * Caller must make sure that @ind is valid and will stay that way.
207 */
208static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
209{
210 struct ext4_inode_info *ei = EXT4_I(inode);
211 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
212 __le32 *p;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400213
214 /* Try to find previous block */
215 for (p = ind->p - 1; p >= start; p--) {
216 if (*p)
217 return le32_to_cpu(*p);
218 }
219
220 /* No such thing, so let's try location of indirect block */
221 if (ind->bh)
222 return ind->bh->b_blocknr;
223
224 /*
225 * It is going to be referred to from the inode itself? OK, just put it
226 * into the same cylinder group then.
227 */
Eric Sandeenf86186b2011-06-28 10:01:31 -0400228 return ext4_inode_to_goal_block(inode);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400229}
230
231/**
232 * ext4_find_goal - find a preferred place for allocation.
233 * @inode: owner
234 * @block: block we want
235 * @partial: pointer to the last triple within a chain
236 *
237 * Normally this function find the preferred place for block allocation,
238 * returns it.
239 * Because this is only used for non-extent files, we limit the block nr
240 * to 32 bits.
241 */
242static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
243 Indirect *partial)
244{
245 ext4_fsblk_t goal;
246
247 /*
248 * XXX need to get goal block from mballoc's data structures
249 */
250
251 goal = ext4_find_near(inode, partial);
252 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
253 return goal;
254}
255
256/**
257 * ext4_blks_to_allocate - Look up the block map and count the number
258 * of direct blocks need to be allocated for the given branch.
259 *
260 * @branch: chain of indirect blocks
261 * @k: number of blocks need for indirect blocks
262 * @blks: number of data blocks to be mapped.
263 * @blocks_to_boundary: the offset in the indirect block
264 *
265 * return the total number of blocks to be allocate, including the
266 * direct and indirect blocks.
267 */
268static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
269 int blocks_to_boundary)
270{
271 unsigned int count = 0;
272
273 /*
274 * Simple case, [t,d]Indirect block(s) has not allocated yet
275 * then it's clear blocks on that path have not allocated
276 */
277 if (k > 0) {
278 /* right now we don't handle cross boundary allocation */
279 if (blks < blocks_to_boundary + 1)
280 count += blks;
281 else
282 count += blocks_to_boundary + 1;
283 return count;
284 }
285
286 count++;
287 while (count < blks && count <= blocks_to_boundary &&
288 le32_to_cpu(*(branch[0].p + count)) == 0) {
289 count++;
290 }
291 return count;
292}
293
294/**
Amir Goldsteindae1e522011-06-27 19:40:50 -0400295 * 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 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400321static int ext4_alloc_branch(handle_t *handle,
322 struct ext4_allocation_request *ar,
323 int indirect_blks, ext4_lblk_t *offsets,
324 Indirect *branch)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400325{
Theodore Ts'o781f1432013-04-03 12:43:17 -0400326 struct buffer_head * bh;
327 ext4_fsblk_t b, new_blocks[4];
328 __le32 *p;
329 int i, j, err, len = 1;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400330
Theodore Ts'o781f1432013-04-03 12:43:17 -0400331 for (i = 0; i <= indirect_blks; i++) {
332 if (i == indirect_blks) {
Theodore Ts'oa5211002014-09-04 18:06:25 -0400333 new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
Theodore Ts'o781f1432013-04-03 12:43:17 -0400334 } else
Theodore Ts'oa5211002014-09-04 18:06:25 -0400335 ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -0400336 ar->inode, ar->goal,
337 ar->flags & EXT4_MB_DELALLOC_RESERVED,
338 NULL, &err);
Theodore Ts'o781f1432013-04-03 12:43:17 -0400339 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
Theodore Ts'oa5211002014-09-04 18:06:25 -0400347 bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400348 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500349 err = -ENOMEM;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400350 goto failed;
351 }
Amir Goldsteindae1e522011-06-27 19:40:50 -0400352 lock_buffer(bh);
353 BUFFER_TRACE(bh, "call get_create_access");
354 err = ext4_journal_get_create_access(handle, bh);
355 if (err) {
Amir Goldsteindae1e522011-06-27 19:40:50 -0400356 unlock_buffer(bh);
357 goto failed;
358 }
359
Theodore Ts'o781f1432013-04-03 12:43:17 -0400360 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)
Theodore Ts'oa5211002014-09-04 18:06:25 -0400365 len = ar->len;
Theodore Ts'o781f1432013-04-03 12:43:17 -0400366 for (j = 0; j < len; j++)
367 *p++ = cpu_to_le32(b++);
368
Amir Goldsteindae1e522011-06-27 19:40:50 -0400369 BUFFER_TRACE(bh, "marking uptodate");
370 set_buffer_uptodate(bh);
371 unlock_buffer(bh);
372
373 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Theodore Ts'oa5211002014-09-04 18:06:25 -0400374 err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400375 if (err)
376 goto failed;
377 }
Theodore Ts'o781f1432013-04-03 12:43:17 -0400378 return 0;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400379failed:
Theodore Ts'o781f1432013-04-03 12:43:17 -0400380 for (; i >= 0; i--) {
Jan Karac5c7b8d2014-06-15 23:46:28 -0400381 /*
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)
Theodore Ts'oa5211002014-09-04 18:06:25 -0400388 ext4_forget(handle, 1, ar->inode, branch[i].bh,
Theodore Ts'o781f1432013-04-03 12:43:17 -0400389 branch[i].bh->b_blocknr);
Theodore Ts'oa5211002014-09-04 18:06:25 -0400390 ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
391 (i == indirect_blks) ? ar->len : 1, 0);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400392 }
Amir Goldsteindae1e522011-06-27 19:40:50 -0400393 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 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400411static int ext4_splice_branch(handle_t *handle,
412 struct ext4_allocation_request *ar,
413 Indirect *where, int num)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400414{
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 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400438 if (num == 0 && ar->len > 1) {
Amir Goldsteindae1e522011-06-27 19:40:50 -0400439 current_block = le32_to_cpu(where->key) + 1;
Theodore Ts'oa5211002014-09-04 18:06:25 -0400440 for (i = 1; i < ar->len; i++)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400441 *(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");
Theodore Ts'oa5211002014-09-04 18:06:25 -0400457 err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400458 if (err)
459 goto err_out;
460 } else {
461 /*
462 * OK, we spliced it into the inode itself on a direct block.
463 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400464 ext4_mark_inode_dirty(handle, ar->inode);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400465 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 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400476 ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
Amir Goldsteindae1e522011-06-27 19:40:50 -0400477 EXT4_FREE_BLOCKS_FORGET);
478 }
Theodore Ts'oa5211002014-09-04 18:06:25 -0400479 ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
480 ar->len, 0);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400481
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{
Theodore Ts'oa5211002014-09-04 18:06:25 -0400517 struct ext4_allocation_request ar;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400518 int err = -EIO;
519 ext4_lblk_t offsets[4];
520 Indirect chain[4];
521 Indirect *partial;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400522 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 */
Theodore Ts'obab08ab2011-09-09 18:36:51 -0400564 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
Theodore Ts'oa5211002014-09-04 18:06:25 -0400571 /* 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;
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -0400577 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
578 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
Theodore Ts'oa5211002014-09-04 18:06:25 -0400579
580 ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400581
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 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400589 ar.len = ext4_blks_to_allocate(partial, indirect_blks,
590 map->m_len, blocks_to_boundary);
591
Amir Goldsteindae1e522011-06-27 19:40:50 -0400592 /*
593 * Block out ext4_truncate while we alter the tree
594 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400595 err = ext4_alloc_branch(handle, &ar, indirect_blks,
Amir Goldsteindae1e522011-06-27 19:40:50 -0400596 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)
Theodore Ts'oa5211002014-09-04 18:06:25 -0400606 err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400607 if (err)
608 goto cleanup;
609
610 map->m_flags |= EXT4_MAP_NEW;
611
612 ext4_update_inode_fsync_trans(handle, inode, 1);
Theodore Ts'oa5211002014-09-04 18:06:25 -0400613 count = ar.len;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400614got_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:
Theodore Ts'o21ddd562013-07-01 08:12:40 -0400630 trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400631 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,
Al Viro16b1f052014-03-04 22:14:00 -0500646 struct iov_iter *iter, loff_t offset)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400647{
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;
Al Viroa6cbcd42014-03-04 22:38:00 -0500654 size_t count = iov_iter_count(iter);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400655 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 */
Theodore Ts'o9924a922013-02-08 21:59:22 -0500662 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400663 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:
Jiaying Zhangdccaf332011-08-19 19:13:32 -0400679 if (rw == READ && ext4_should_dioread_nolock(inode)) {
Dmitry Monakhov17335dc2012-09-29 00:41:21 -0400680 /*
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 }
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800692 if (IS_DAX(inode))
693 ret = dax_do_io(rw, iocb, inode, iter, offset,
694 ext4_get_block, NULL, 0);
695 else
696 ret = __blockdev_direct_IO(rw, iocb, inode,
697 inode->i_sb->s_bdev, iter, offset,
698 ext4_get_block, NULL, NULL, 0);
Dmitry Monakhov17335dc2012-09-29 00:41:21 -0400699 inode_dio_done(inode);
Jiaying Zhangdccaf332011-08-19 19:13:32 -0400700 } else {
Dmitry Monakhov17335dc2012-09-29 00:41:21 -0400701locked:
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800702 if (IS_DAX(inode))
703 ret = dax_do_io(rw, iocb, inode, iter, offset,
704 ext4_get_block, NULL, DIO_LOCKING);
705 else
706 ret = blockdev_direct_IO(rw, iocb, inode, iter,
707 offset, ext4_get_block);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400708
709 if (unlikely((rw & WRITE) && ret < 0)) {
710 loff_t isize = i_size_read(inode);
Al Viro16b1f052014-03-04 22:14:00 -0500711 loff_t end = offset + count;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400712
713 if (end > isize)
714 ext4_truncate_failed_write(inode);
715 }
716 }
717 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
718 goto retry;
719
720 if (orphan) {
721 int err;
722
723 /* Credits for sb + inode write */
Theodore Ts'o9924a922013-02-08 21:59:22 -0500724 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400725 if (IS_ERR(handle)) {
726 /* This is really bad luck. We've written the data
727 * but cannot extend i_size. Bail out and pretend
728 * the write failed... */
729 ret = PTR_ERR(handle);
730 if (inode->i_nlink)
731 ext4_orphan_del(NULL, inode);
732
733 goto out;
734 }
735 if (inode->i_nlink)
736 ext4_orphan_del(handle, inode);
737 if (ret > 0) {
738 loff_t end = offset + ret;
739 if (end > inode->i_size) {
740 ei->i_disksize = end;
741 i_size_write(inode, end);
742 /*
743 * We're going to return a positive `ret'
744 * here due to non-zero-length I/O, so there's
745 * no way of reporting error returns from
746 * ext4_mark_inode_dirty() to userspace. So
747 * ignore it.
748 */
749 ext4_mark_inode_dirty(handle, inode);
750 }
751 }
752 err = ext4_journal_stop(handle);
753 if (ret == 0)
754 ret = err;
755 }
756out:
757 return ret;
758}
759
760/*
761 * Calculate the number of metadata blocks need to reserve
762 * to allocate a new block at @lblocks for non extent file based file
763 */
764int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
765{
766 struct ext4_inode_info *ei = EXT4_I(inode);
767 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
768 int blk_bits;
769
770 if (lblock < EXT4_NDIR_BLOCKS)
771 return 0;
772
773 lblock -= EXT4_NDIR_BLOCKS;
774
775 if (ei->i_da_metadata_calc_len &&
776 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
777 ei->i_da_metadata_calc_len++;
778 return 0;
779 }
780 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
781 ei->i_da_metadata_calc_len = 1;
782 blk_bits = order_base_2(lblock);
783 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
784}
785
Jan Karafa55a0e2013-06-04 12:56:55 -0400786/*
787 * Calculate number of indirect blocks touched by mapping @nrblocks logically
788 * contiguous blocks
789 */
790int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400791{
Amir Goldsteindae1e522011-06-27 19:40:50 -0400792 /*
Jan Karafa55a0e2013-06-04 12:56:55 -0400793 * With N contiguous data blocks, we need at most
794 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
795 * 2 dindirect blocks, and 1 tindirect block
Amir Goldsteindae1e522011-06-27 19:40:50 -0400796 */
Jan Karafa55a0e2013-06-04 12:56:55 -0400797 return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400798}
799
800/*
801 * Truncate transactions can be complex and absolutely huge. So we need to
802 * be able to restart the transaction at a conventient checkpoint to make
803 * sure we don't overflow the journal.
804 *
Theodore Ts'o819c4922013-04-03 12:47:17 -0400805 * Try to extend this transaction for the purposes of truncation. If
Amir Goldsteindae1e522011-06-27 19:40:50 -0400806 * extend fails, we need to propagate the failure up and restart the
807 * transaction in the top-level truncate loop. --sct
Amir Goldsteindae1e522011-06-27 19:40:50 -0400808 *
809 * Returns 0 if we managed to create more room. If we can't create more
810 * room, and the transaction must be restarted we return 1.
811 */
812static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
813{
814 if (!ext4_handle_valid(handle))
815 return 0;
816 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
817 return 0;
818 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
819 return 0;
820 return 1;
821}
822
823/*
824 * Probably it should be a library function... search for first non-zero word
825 * or memcmp with zero_page, whatever is better for particular architecture.
826 * Linus?
827 */
828static inline int all_zeroes(__le32 *p, __le32 *q)
829{
830 while (p < q)
831 if (*p++)
832 return 0;
833 return 1;
834}
835
836/**
837 * ext4_find_shared - find the indirect blocks for partial truncation.
838 * @inode: inode in question
839 * @depth: depth of the affected branch
840 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
841 * @chain: place to store the pointers to partial indirect blocks
842 * @top: place to the (detached) top of branch
843 *
844 * This is a helper function used by ext4_truncate().
845 *
846 * When we do truncate() we may have to clean the ends of several
847 * indirect blocks but leave the blocks themselves alive. Block is
848 * partially truncated if some data below the new i_size is referred
849 * from it (and it is on the path to the first completely truncated
850 * data block, indeed). We have to free the top of that path along
851 * with everything to the right of the path. Since no allocation
852 * past the truncation point is possible until ext4_truncate()
853 * finishes, we may safely do the latter, but top of branch may
854 * require special attention - pageout below the truncation point
855 * might try to populate it.
856 *
857 * We atomically detach the top of branch from the tree, store the
858 * block number of its root in *@top, pointers to buffer_heads of
859 * partially truncated blocks - in @chain[].bh and pointers to
860 * their last elements that should not be removed - in
861 * @chain[].p. Return value is the pointer to last filled element
862 * of @chain.
863 *
864 * The work left to caller to do the actual freeing of subtrees:
865 * a) free the subtree starting from *@top
866 * b) free the subtrees whose roots are stored in
867 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
868 * c) free the subtrees growing from the inode past the @chain[0].
869 * (no partially truncated stuff there). */
870
871static Indirect *ext4_find_shared(struct inode *inode, int depth,
872 ext4_lblk_t offsets[4], Indirect chain[4],
873 __le32 *top)
874{
875 Indirect *partial, *p;
876 int k, err;
877
878 *top = 0;
879 /* Make k index the deepest non-null offset + 1 */
880 for (k = depth; k > 1 && !offsets[k-1]; k--)
881 ;
882 partial = ext4_get_branch(inode, k, offsets, chain, &err);
883 /* Writer: pointers */
884 if (!partial)
885 partial = chain + k-1;
886 /*
887 * If the branch acquired continuation since we've looked at it -
888 * fine, it should all survive and (new) top doesn't belong to us.
889 */
890 if (!partial->key && *partial->p)
891 /* Writer: end */
892 goto no_top;
893 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
894 ;
895 /*
896 * OK, we've found the last block that must survive. The rest of our
897 * branch should be detached before unlocking. However, if that rest
898 * of branch is all ours and does not grow immediately from the inode
899 * it's easier to cheat and just decrement partial->p.
900 */
901 if (p == chain + k - 1 && p > chain) {
902 p->p--;
903 } else {
904 *top = *p->p;
905 /* Nope, don't do this in ext4. Must leave the tree intact */
906#if 0
907 *p->p = 0;
908#endif
909 }
910 /* Writer: end */
911
912 while (partial > p) {
913 brelse(partial->bh);
914 partial--;
915 }
916no_top:
917 return partial;
918}
919
920/*
921 * Zero a number of block pointers in either an inode or an indirect block.
922 * If we restart the transaction we must again get write access to the
923 * indirect block for further modification.
924 *
925 * We release `count' blocks on disk, but (last - first) may be greater
926 * than `count' because there can be holes in there.
927 *
928 * Return 0 on success, 1 on invalid block range
929 * and < 0 on fatal error.
930 */
931static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
932 struct buffer_head *bh,
933 ext4_fsblk_t block_to_free,
934 unsigned long count, __le32 *first,
935 __le32 *last)
936{
937 __le32 *p;
Theodore Ts'o981250c2013-06-12 11:48:29 -0400938 int flags = EXT4_FREE_BLOCKS_VALIDATED;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400939 int err;
940
941 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
Theodore Ts'o981250c2013-06-12 11:48:29 -0400942 flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
943 else if (ext4_should_journal_data(inode))
944 flags |= EXT4_FREE_BLOCKS_FORGET;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400945
946 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
947 count)) {
948 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
949 "blocks %llu len %lu",
950 (unsigned long long) block_to_free, count);
951 return 1;
952 }
953
954 if (try_to_extend_transaction(handle, inode)) {
955 if (bh) {
956 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
957 err = ext4_handle_dirty_metadata(handle, inode, bh);
958 if (unlikely(err))
959 goto out_err;
960 }
961 err = ext4_mark_inode_dirty(handle, inode);
962 if (unlikely(err))
963 goto out_err;
964 err = ext4_truncate_restart_trans(handle, inode,
965 ext4_blocks_for_truncate(inode));
966 if (unlikely(err))
967 goto out_err;
968 if (bh) {
969 BUFFER_TRACE(bh, "retaking write access");
970 err = ext4_journal_get_write_access(handle, bh);
971 if (unlikely(err))
972 goto out_err;
973 }
974 }
975
976 for (p = first; p < last; p++)
977 *p = 0;
978
979 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
980 return 0;
981out_err:
982 ext4_std_error(inode->i_sb, err);
983 return err;
984}
985
986/**
987 * ext4_free_data - free a list of data blocks
988 * @handle: handle for this transaction
989 * @inode: inode we are dealing with
990 * @this_bh: indirect buffer_head which contains *@first and *@last
991 * @first: array of block numbers
992 * @last: points immediately past the end of array
993 *
994 * We are freeing all blocks referred from that array (numbers are stored as
995 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
996 *
997 * We accumulate contiguous runs of blocks to free. Conveniently, if these
998 * blocks are contiguous then releasing them at one time will only affect one
999 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1000 * actually use a lot of journal space.
1001 *
1002 * @this_bh will be %NULL if @first and @last point into the inode's direct
1003 * block pointers.
1004 */
1005static void ext4_free_data(handle_t *handle, struct inode *inode,
1006 struct buffer_head *this_bh,
1007 __le32 *first, __le32 *last)
1008{
1009 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
1010 unsigned long count = 0; /* Number of blocks in the run */
1011 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
1012 corresponding to
1013 block_to_free */
1014 ext4_fsblk_t nr; /* Current block # */
1015 __le32 *p; /* Pointer into inode/ind
1016 for current block */
1017 int err = 0;
1018
1019 if (this_bh) { /* For indirect block */
1020 BUFFER_TRACE(this_bh, "get_write_access");
1021 err = ext4_journal_get_write_access(handle, this_bh);
1022 /* Important: if we can't update the indirect pointers
1023 * to the blocks, we can't free them. */
1024 if (err)
1025 return;
1026 }
1027
1028 for (p = first; p < last; p++) {
1029 nr = le32_to_cpu(*p);
1030 if (nr) {
1031 /* accumulate blocks to free if they're contiguous */
1032 if (count == 0) {
1033 block_to_free = nr;
1034 block_to_free_p = p;
1035 count = 1;
1036 } else if (nr == block_to_free + count) {
1037 count++;
1038 } else {
1039 err = ext4_clear_blocks(handle, inode, this_bh,
1040 block_to_free, count,
1041 block_to_free_p, p);
1042 if (err)
1043 break;
1044 block_to_free = nr;
1045 block_to_free_p = p;
1046 count = 1;
1047 }
1048 }
1049 }
1050
1051 if (!err && count > 0)
1052 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1053 count, block_to_free_p, p);
1054 if (err < 0)
1055 /* fatal error */
1056 return;
1057
1058 if (this_bh) {
1059 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1060
1061 /*
1062 * The buffer head should have an attached journal head at this
1063 * point. However, if the data is corrupted and an indirect
1064 * block pointed to itself, it would have been detached when
1065 * the block was cleared. Check for this instead of OOPSing.
1066 */
1067 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1068 ext4_handle_dirty_metadata(handle, inode, this_bh);
1069 else
1070 EXT4_ERROR_INODE(inode,
1071 "circular indirect block detected at "
1072 "block %llu",
1073 (unsigned long long) this_bh->b_blocknr);
1074 }
1075}
1076
1077/**
1078 * ext4_free_branches - free an array of branches
1079 * @handle: JBD handle for this transaction
1080 * @inode: inode we are dealing with
1081 * @parent_bh: the buffer_head which contains *@first and *@last
1082 * @first: array of block numbers
1083 * @last: pointer immediately past the end of array
1084 * @depth: depth of the branches to free
1085 *
1086 * We are freeing all blocks referred from these branches (numbers are
1087 * stored as little-endian 32-bit) and updating @inode->i_blocks
1088 * appropriately.
1089 */
1090static void ext4_free_branches(handle_t *handle, struct inode *inode,
1091 struct buffer_head *parent_bh,
1092 __le32 *first, __le32 *last, int depth)
1093{
1094 ext4_fsblk_t nr;
1095 __le32 *p;
1096
1097 if (ext4_handle_is_aborted(handle))
1098 return;
1099
1100 if (depth--) {
1101 struct buffer_head *bh;
1102 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1103 p = last;
1104 while (--p >= first) {
1105 nr = le32_to_cpu(*p);
1106 if (!nr)
1107 continue; /* A hole */
1108
1109 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1110 nr, 1)) {
1111 EXT4_ERROR_INODE(inode,
1112 "invalid indirect mapped "
1113 "block %lu (level %d)",
1114 (unsigned long) nr, depth);
1115 break;
1116 }
1117
1118 /* Go read the buffer for the next level down */
1119 bh = sb_bread(inode->i_sb, nr);
1120
1121 /*
1122 * A read failure? Report error and clear slot
1123 * (should be rare).
1124 */
1125 if (!bh) {
1126 EXT4_ERROR_INODE_BLOCK(inode, nr,
1127 "Read failure");
1128 continue;
1129 }
1130
1131 /* This zaps the entire block. Bottom up. */
1132 BUFFER_TRACE(bh, "free child branches");
1133 ext4_free_branches(handle, inode, bh,
1134 (__le32 *) bh->b_data,
1135 (__le32 *) bh->b_data + addr_per_block,
1136 depth);
1137 brelse(bh);
1138
1139 /*
1140 * Everything below this this pointer has been
1141 * released. Now let this top-of-subtree go.
1142 *
1143 * We want the freeing of this indirect block to be
1144 * atomic in the journal with the updating of the
1145 * bitmap block which owns it. So make some room in
1146 * the journal.
1147 *
1148 * We zero the parent pointer *after* freeing its
1149 * pointee in the bitmaps, so if extend_transaction()
1150 * for some reason fails to put the bitmap changes and
1151 * the release into the same transaction, recovery
1152 * will merely complain about releasing a free block,
1153 * rather than leaking blocks.
1154 */
1155 if (ext4_handle_is_aborted(handle))
1156 return;
1157 if (try_to_extend_transaction(handle, inode)) {
1158 ext4_mark_inode_dirty(handle, inode);
1159 ext4_truncate_restart_trans(handle, inode,
1160 ext4_blocks_for_truncate(inode));
1161 }
1162
1163 /*
1164 * The forget flag here is critical because if
1165 * we are journaling (and not doing data
1166 * journaling), we have to make sure a revoke
1167 * record is written to prevent the journal
1168 * replay from overwriting the (former)
1169 * indirect block if it gets reallocated as a
1170 * data block. This must happen in the same
1171 * transaction where the data blocks are
1172 * actually freed.
1173 */
1174 ext4_free_blocks(handle, inode, NULL, nr, 1,
1175 EXT4_FREE_BLOCKS_METADATA|
1176 EXT4_FREE_BLOCKS_FORGET);
1177
1178 if (parent_bh) {
1179 /*
1180 * The block which we have just freed is
1181 * pointed to by an indirect block: journal it
1182 */
1183 BUFFER_TRACE(parent_bh, "get_write_access");
1184 if (!ext4_journal_get_write_access(handle,
1185 parent_bh)){
1186 *p = 0;
1187 BUFFER_TRACE(parent_bh,
1188 "call ext4_handle_dirty_metadata");
1189 ext4_handle_dirty_metadata(handle,
1190 inode,
1191 parent_bh);
1192 }
1193 }
1194 }
1195 } else {
1196 /* We have reached the bottom of the tree. */
1197 BUFFER_TRACE(parent_bh, "free data blocks");
1198 ext4_free_data(handle, inode, parent_bh, first, last);
1199 }
1200}
1201
Theodore Ts'o819c4922013-04-03 12:47:17 -04001202void ext4_ind_truncate(handle_t *handle, struct inode *inode)
Amir Goldsteindae1e522011-06-27 19:40:50 -04001203{
Amir Goldsteindae1e522011-06-27 19:40:50 -04001204 struct ext4_inode_info *ei = EXT4_I(inode);
1205 __le32 *i_data = ei->i_data;
1206 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Amir Goldsteindae1e522011-06-27 19:40:50 -04001207 ext4_lblk_t offsets[4];
1208 Indirect chain[4];
1209 Indirect *partial;
1210 __le32 nr = 0;
1211 int n = 0;
1212 ext4_lblk_t last_block, max_block;
1213 unsigned blocksize = inode->i_sb->s_blocksize;
Amir Goldsteindae1e522011-06-27 19:40:50 -04001214
1215 last_block = (inode->i_size + blocksize-1)
1216 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1217 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1218 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1219
Amir Goldsteindae1e522011-06-27 19:40:50 -04001220 if (last_block != max_block) {
1221 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1222 if (n == 0)
Theodore Ts'o819c4922013-04-03 12:47:17 -04001223 return;
Amir Goldsteindae1e522011-06-27 19:40:50 -04001224 }
1225
Zheng Liu51865fd2012-11-08 21:57:32 -05001226 ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
Amir Goldsteindae1e522011-06-27 19:40:50 -04001227
1228 /*
1229 * The orphan list entry will now protect us from any crash which
1230 * occurs before the truncate completes, so it is now safe to propagate
1231 * the new, shorter inode size (held for now in i_size) into the
1232 * on-disk inode. We do this via i_disksize, which is the value which
1233 * ext4 *really* writes onto the disk inode.
1234 */
1235 ei->i_disksize = inode->i_size;
1236
1237 if (last_block == max_block) {
1238 /*
1239 * It is unnecessary to free any data blocks if last_block is
1240 * equal to the indirect block limit.
1241 */
Theodore Ts'o819c4922013-04-03 12:47:17 -04001242 return;
Amir Goldsteindae1e522011-06-27 19:40:50 -04001243 } else if (n == 1) { /* direct blocks */
1244 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1245 i_data + EXT4_NDIR_BLOCKS);
1246 goto do_indirects;
1247 }
1248
1249 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1250 /* Kill the top of shared branch (not detached) */
1251 if (nr) {
1252 if (partial == chain) {
1253 /* Shared branch grows from the inode */
1254 ext4_free_branches(handle, inode, NULL,
1255 &nr, &nr+1, (chain+n-1) - partial);
1256 *partial->p = 0;
1257 /*
1258 * We mark the inode dirty prior to restart,
1259 * and prior to stop. No need for it here.
1260 */
1261 } else {
1262 /* Shared branch grows from an indirect block */
1263 BUFFER_TRACE(partial->bh, "get_write_access");
1264 ext4_free_branches(handle, inode, partial->bh,
1265 partial->p,
1266 partial->p+1, (chain+n-1) - partial);
1267 }
1268 }
1269 /* Clear the ends of indirect blocks on the shared branch */
1270 while (partial > chain) {
1271 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1272 (__le32*)partial->bh->b_data+addr_per_block,
1273 (chain+n-1) - partial);
1274 BUFFER_TRACE(partial->bh, "call brelse");
1275 brelse(partial->bh);
1276 partial--;
1277 }
1278do_indirects:
1279 /* Kill the remaining (whole) subtrees */
1280 switch (offsets[0]) {
1281 default:
1282 nr = i_data[EXT4_IND_BLOCK];
1283 if (nr) {
1284 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1285 i_data[EXT4_IND_BLOCK] = 0;
1286 }
1287 case EXT4_IND_BLOCK:
1288 nr = i_data[EXT4_DIND_BLOCK];
1289 if (nr) {
1290 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1291 i_data[EXT4_DIND_BLOCK] = 0;
1292 }
1293 case EXT4_DIND_BLOCK:
1294 nr = i_data[EXT4_TIND_BLOCK];
1295 if (nr) {
1296 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1297 i_data[EXT4_TIND_BLOCK] = 0;
1298 }
1299 case EXT4_TIND_BLOCK:
1300 ;
1301 }
Amir Goldsteindae1e522011-06-27 19:40:50 -04001302}
1303
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001304/**
1305 * ext4_ind_remove_space - remove space from the range
1306 * @handle: JBD handle for this transaction
1307 * @inode: inode we are dealing with
1308 * @start: First block to remove
1309 * @end: One block after the last block to remove (exclusive)
1310 *
1311 * Free the blocks in the defined range (end is exclusive endpoint of
1312 * range). This is used by ext4_punch_hole().
1313 */
1314int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1315 ext4_lblk_t start, ext4_lblk_t end)
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001316{
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001317 struct ext4_inode_info *ei = EXT4_I(inode);
1318 __le32 *i_data = ei->i_data;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001319 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001320 ext4_lblk_t offsets[4], offsets2[4];
1321 Indirect chain[4], chain2[4];
1322 Indirect *partial, *partial2;
1323 ext4_lblk_t max_block;
1324 __le32 nr = 0, nr2 = 0;
1325 int n = 0, n2 = 0;
1326 unsigned blocksize = inode->i_sb->s_blocksize;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001327
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001328 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1329 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1330 if (end >= max_block)
1331 end = max_block;
1332 if ((start >= end) || (start > max_block))
1333 return 0;
Jan Karaa93cd4c2014-06-26 12:30:54 -04001334
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001335 n = ext4_block_to_path(inode, start, offsets, NULL);
1336 n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1337
1338 BUG_ON(n > n2);
1339
1340 if ((n == 1) && (n == n2)) {
1341 /* We're punching only within direct block range */
1342 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1343 i_data + offsets2[0]);
1344 return 0;
1345 } else if (n2 > n) {
1346 /*
1347 * Start and end are on a different levels so we're going to
1348 * free partial block at start, and partial block at end of
1349 * the range. If there are some levels in between then
1350 * do_indirects label will take care of that.
1351 */
1352
1353 if (n == 1) {
1354 /*
1355 * Start is at the direct block level, free
1356 * everything to the end of the level.
1357 */
1358 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1359 i_data + EXT4_NDIR_BLOCKS);
1360 goto end_range;
1361 }
1362
1363
1364 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1365 if (nr) {
1366 if (partial == chain) {
1367 /* Shared branch grows from the inode */
1368 ext4_free_branches(handle, inode, NULL,
1369 &nr, &nr+1, (chain+n-1) - partial);
1370 *partial->p = 0;
Jan Karaa93cd4c2014-06-26 12:30:54 -04001371 } else {
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001372 /* Shared branch grows from an indirect block */
1373 BUFFER_TRACE(partial->bh, "get_write_access");
1374 ext4_free_branches(handle, inode, partial->bh,
1375 partial->p,
1376 partial->p+1, (chain+n-1) - partial);
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001377 }
1378 }
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001379
1380 /*
1381 * Clear the ends of indirect blocks on the shared branch
1382 * at the start of the range
1383 */
1384 while (partial > chain) {
1385 ext4_free_branches(handle, inode, partial->bh,
1386 partial->p + 1,
1387 (__le32 *)partial->bh->b_data+addr_per_block,
1388 (chain+n-1) - partial);
1389 BUFFER_TRACE(partial->bh, "call brelse");
1390 brelse(partial->bh);
1391 partial--;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001392 }
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001393
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001394end_range:
1395 partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1396 if (nr2) {
1397 if (partial2 == chain2) {
1398 /*
1399 * Remember, end is exclusive so here we're at
1400 * the start of the next level we're not going
1401 * to free. Everything was covered by the start
1402 * of the range.
1403 */
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001404 goto do_indirects;
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001405 }
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001406 } else {
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001407 /*
1408 * ext4_find_shared returns Indirect structure which
1409 * points to the last element which should not be
1410 * removed by truncate. But this is end of the range
1411 * in punch_hole so we need to point to the next element
1412 */
1413 partial2->p++;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001414 }
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001415
1416 /*
1417 * Clear the ends of indirect blocks on the shared branch
1418 * at the end of the range
1419 */
1420 while (partial2 > chain2) {
1421 ext4_free_branches(handle, inode, partial2->bh,
1422 (__le32 *)partial2->bh->b_data,
1423 partial2->p,
1424 (chain2+n2-1) - partial2);
1425 BUFFER_TRACE(partial2->bh, "call brelse");
1426 brelse(partial2->bh);
1427 partial2--;
1428 }
1429 goto do_indirects;
1430 }
1431
1432 /* Punch happened within the same level (n == n2) */
1433 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1434 partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001435
1436 /* Free top, but only if partial2 isn't its subtree. */
1437 if (nr) {
1438 int level = min(partial - chain, partial2 - chain2);
1439 int i;
1440 int subtree = 1;
1441
1442 for (i = 0; i <= level; i++) {
1443 if (offsets[i] != offsets2[i]) {
1444 subtree = 0;
1445 break;
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001446 }
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001447 }
1448
1449 if (!subtree) {
1450 if (partial == chain) {
1451 /* Shared branch grows from the inode */
1452 ext4_free_branches(handle, inode, NULL,
1453 &nr, &nr+1,
1454 (chain+n-1) - partial);
1455 *partial->p = 0;
1456 } else {
1457 /* Shared branch grows from an indirect block */
1458 BUFFER_TRACE(partial->bh, "get_write_access");
1459 ext4_free_branches(handle, inode, partial->bh,
1460 partial->p,
1461 partial->p+1,
1462 (chain+n-1) - partial);
1463 }
1464 }
1465 }
1466
1467 if (!nr2) {
1468 /*
1469 * ext4_find_shared returns Indirect structure which
1470 * points to the last element which should not be
1471 * removed by truncate. But this is end of the range
1472 * in punch_hole so we need to point to the next element
1473 */
1474 partial2->p++;
1475 }
1476
1477 while (partial > chain || partial2 > chain2) {
1478 int depth = (chain+n-1) - partial;
1479 int depth2 = (chain2+n2-1) - partial2;
1480
1481 if (partial > chain && partial2 > chain2 &&
1482 partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1483 /*
1484 * We've converged on the same block. Clear the range,
1485 * then we're done.
1486 */
1487 ext4_free_branches(handle, inode, partial->bh,
1488 partial->p + 1,
1489 partial2->p,
1490 (chain+n-1) - partial);
1491 BUFFER_TRACE(partial->bh, "call brelse");
1492 brelse(partial->bh);
1493 BUFFER_TRACE(partial2->bh, "call brelse");
1494 brelse(partial2->bh);
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001495 return 0;
1496 }
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001497
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001498 /*
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001499 * The start and end partial branches may not be at the same
1500 * level even though the punch happened within one level. So, we
1501 * give them a chance to arrive at the same level, then walk
1502 * them in step with each other until we converge on the same
1503 * block.
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001504 */
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001505 if (partial > chain && depth <= depth2) {
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001506 ext4_free_branches(handle, inode, partial->bh,
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001507 partial->p + 1,
1508 (__le32 *)partial->bh->b_data+addr_per_block,
1509 (chain+n-1) - partial);
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001510 BUFFER_TRACE(partial->bh, "call brelse");
1511 brelse(partial->bh);
1512 partial--;
1513 }
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001514 if (partial2 > chain2 && depth2 <= depth) {
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001515 ext4_free_branches(handle, inode, partial2->bh,
1516 (__le32 *)partial2->bh->b_data,
1517 partial2->p,
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001518 (chain2+n2-1) - partial2);
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001519 BUFFER_TRACE(partial2->bh, "call brelse");
1520 brelse(partial2->bh);
1521 partial2--;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001522 }
1523 }
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001524 return 0;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001525
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001526do_indirects:
1527 /* Kill the remaining (whole) subtrees */
1528 switch (offsets[0]) {
1529 default:
1530 if (++n >= n2)
1531 return 0;
1532 nr = i_data[EXT4_IND_BLOCK];
1533 if (nr) {
1534 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1535 i_data[EXT4_IND_BLOCK] = 0;
1536 }
1537 case EXT4_IND_BLOCK:
1538 if (++n >= n2)
1539 return 0;
1540 nr = i_data[EXT4_DIND_BLOCK];
1541 if (nr) {
1542 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1543 i_data[EXT4_DIND_BLOCK] = 0;
1544 }
1545 case EXT4_DIND_BLOCK:
1546 if (++n >= n2)
1547 return 0;
1548 nr = i_data[EXT4_TIND_BLOCK];
1549 if (nr) {
1550 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1551 i_data[EXT4_TIND_BLOCK] = 0;
1552 }
1553 case EXT4_TIND_BLOCK:
1554 ;
1555 }
1556 return 0;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001557}