Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
dae1e52c AG |
2 | /* |
3 | * linux/fs/ext4/indirect.c | |
4 | * | |
5 | * from | |
6 | * | |
7 | * linux/fs/ext4/inode.c | |
8 | * | |
9 | * Copyright (C) 1992, 1993, 1994, 1995 | |
10 | * Remy Card (card@masi.ibp.fr) | |
11 | * Laboratoire MASI - Institut Blaise Pascal | |
12 | * Universite Pierre et Marie Curie (Paris VI) | |
13 | * | |
14 | * from | |
15 | * | |
16 | * linux/fs/minix/inode.c | |
17 | * | |
18 | * Copyright (C) 1991, 1992 Linus Torvalds | |
19 | * | |
20 | * Goal-directed block allocation by Stephen Tweedie | |
21 | * (sct@redhat.com), 1993, 1998 | |
22 | */ | |
23 | ||
dae1e52c AG |
24 | #include "ext4_jbd2.h" |
25 | #include "truncate.h" | |
c94c2acf | 26 | #include <linux/dax.h> |
e2e40f2c | 27 | #include <linux/uio.h> |
dae1e52c AG |
28 | |
29 | #include <trace/events/ext4.h> | |
30 | ||
31 | typedef struct { | |
32 | __le32 *p; | |
33 | __le32 key; | |
34 | struct buffer_head *bh; | |
35 | } Indirect; | |
36 | ||
37 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
38 | { | |
39 | p->key = *(p->p = v); | |
40 | p->bh = bh; | |
41 | } | |
42 | ||
43 | /** | |
44 | * ext4_block_to_path - parse the block number into array of offsets | |
45 | * @inode: inode in question (we are only interested in its superblock) | |
46 | * @i_block: block number to be parsed | |
47 | * @offsets: array to store the offsets in | |
48 | * @boundary: set this non-zero if the referred-to block is likely to be | |
49 | * followed (on disk) by an indirect block. | |
50 | * | |
51 | * To store the locations of file's data ext4 uses a data structure common | |
52 | * for UNIX filesystems - tree of pointers anchored in the inode, with | |
53 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
54 | * This function translates the block number into path in that tree - | |
55 | * return value is the path length and @offsets[n] is the offset of | |
56 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
57 | * (negative or too large) warning is printed and zero returned. | |
58 | * | |
59 | * Note: function doesn't find node addresses, so no IO is needed. All | |
60 | * we need to know is the capacity of indirect blocks (taken from the | |
61 | * inode->i_sb). | |
62 | */ | |
63 | ||
64 | /* | |
65 | * Portability note: the last comparison (check that we fit into triple | |
66 | * indirect block) is spelled differently, because otherwise on an | |
67 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
68 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
69 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
70 | * i_block would have to be negative in the very beginning, so we would not | |
71 | * get there at all. | |
72 | */ | |
73 | ||
74 | static int ext4_block_to_path(struct inode *inode, | |
75 | ext4_lblk_t i_block, | |
76 | ext4_lblk_t offsets[4], int *boundary) | |
77 | { | |
78 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
79 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
80 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
81 | indirect_blocks = ptrs, | |
82 | double_blocks = (1 << (ptrs_bits * 2)); | |
83 | int n = 0; | |
84 | int final = 0; | |
85 | ||
86 | if (i_block < direct_blocks) { | |
87 | offsets[n++] = i_block; | |
88 | final = direct_blocks; | |
89 | } else if ((i_block -= direct_blocks) < indirect_blocks) { | |
90 | offsets[n++] = EXT4_IND_BLOCK; | |
91 | offsets[n++] = i_block; | |
92 | final = ptrs; | |
93 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
94 | offsets[n++] = EXT4_DIND_BLOCK; | |
95 | offsets[n++] = i_block >> ptrs_bits; | |
96 | offsets[n++] = i_block & (ptrs - 1); | |
97 | final = ptrs; | |
98 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
99 | offsets[n++] = EXT4_TIND_BLOCK; | |
100 | offsets[n++] = i_block >> (ptrs_bits * 2); | |
101 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
102 | offsets[n++] = i_block & (ptrs - 1); | |
103 | final = ptrs; | |
104 | } else { | |
105 | ext4_warning(inode->i_sb, "block %lu > max in inode %lu", | |
106 | i_block + direct_blocks + | |
107 | indirect_blocks + double_blocks, inode->i_ino); | |
108 | } | |
109 | if (boundary) | |
110 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
111 | return n; | |
112 | } | |
113 | ||
114 | /** | |
115 | * ext4_get_branch - read the chain of indirect blocks leading to data | |
116 | * @inode: inode in question | |
117 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
118 | * @offsets: offsets of pointers in inode/indirect blocks | |
119 | * @chain: place to store the result | |
120 | * @err: here we store the error value | |
121 | * | |
122 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
123 | * if everything went OK or the pointer to the last filled triple | |
124 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
125 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
126 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
127 | * number (it points into struct inode for i==0 and into the bh->b_data | |
128 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
129 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
130 | * numbers of the chain, addresses they were taken from (and where we can | |
131 | * verify that chain did not change) and buffer_heads hosting these | |
132 | * numbers. | |
133 | * | |
134 | * Function stops when it stumbles upon zero pointer (absent block) | |
135 | * (pointer to last triple returned, *@err == 0) | |
136 | * or when it gets an IO error reading an indirect block | |
137 | * (ditto, *@err == -EIO) | |
138 | * or when it reads all @depth-1 indirect blocks successfully and finds | |
139 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
140 | * | |
141 | * Need to be called with | |
142 | * down_read(&EXT4_I(inode)->i_data_sem) | |
143 | */ | |
144 | static Indirect *ext4_get_branch(struct inode *inode, int depth, | |
145 | ext4_lblk_t *offsets, | |
146 | Indirect chain[4], int *err) | |
147 | { | |
148 | struct super_block *sb = inode->i_sb; | |
149 | Indirect *p = chain; | |
150 | struct buffer_head *bh; | |
860d21e2 | 151 | int ret = -EIO; |
dae1e52c AG |
152 | |
153 | *err = 0; | |
154 | /* i_data is not going away, no lock needed */ | |
155 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); | |
156 | if (!p->key) | |
157 | goto no_block; | |
158 | while (--depth) { | |
159 | bh = sb_getblk(sb, le32_to_cpu(p->key)); | |
860d21e2 TT |
160 | if (unlikely(!bh)) { |
161 | ret = -ENOMEM; | |
dae1e52c | 162 | goto failure; |
860d21e2 | 163 | } |
dae1e52c AG |
164 | |
165 | if (!bh_uptodate_or_lock(bh)) { | |
166 | if (bh_submit_read(bh) < 0) { | |
167 | put_bh(bh); | |
168 | goto failure; | |
169 | } | |
170 | /* validate block references */ | |
171 | if (ext4_check_indirect_blockref(inode, bh)) { | |
172 | put_bh(bh); | |
173 | goto failure; | |
174 | } | |
175 | } | |
176 | ||
177 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); | |
178 | /* Reader: end */ | |
179 | if (!p->key) | |
180 | goto no_block; | |
181 | } | |
182 | return NULL; | |
183 | ||
184 | failure: | |
860d21e2 | 185 | *err = ret; |
dae1e52c AG |
186 | no_block: |
187 | return p; | |
188 | } | |
189 | ||
190 | /** | |
191 | * ext4_find_near - find a place for allocation with sufficient locality | |
192 | * @inode: owner | |
193 | * @ind: descriptor of indirect block. | |
194 | * | |
195 | * This function returns the preferred place for block allocation. | |
196 | * It is used when heuristic for sequential allocation fails. | |
197 | * Rules are: | |
198 | * + if there is a block to the left of our position - allocate near it. | |
199 | * + if pointer will live in indirect block - allocate near that block. | |
200 | * + if pointer will live in inode - allocate in the same | |
201 | * cylinder group. | |
202 | * | |
203 | * In the latter case we colour the starting block by the callers PID to | |
204 | * prevent it from clashing with concurrent allocations for a different inode | |
205 | * in the same block group. The PID is used here so that functionally related | |
206 | * files will be close-by on-disk. | |
207 | * | |
208 | * Caller must make sure that @ind is valid and will stay that way. | |
209 | */ | |
210 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) | |
211 | { | |
212 | struct ext4_inode_info *ei = EXT4_I(inode); | |
213 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; | |
214 | __le32 *p; | |
dae1e52c AG |
215 | |
216 | /* Try to find previous block */ | |
217 | for (p = ind->p - 1; p >= start; p--) { | |
218 | if (*p) | |
219 | return le32_to_cpu(*p); | |
220 | } | |
221 | ||
222 | /* No such thing, so let's try location of indirect block */ | |
223 | if (ind->bh) | |
224 | return ind->bh->b_blocknr; | |
225 | ||
226 | /* | |
227 | * It is going to be referred to from the inode itself? OK, just put it | |
228 | * into the same cylinder group then. | |
229 | */ | |
f86186b4 | 230 | return ext4_inode_to_goal_block(inode); |
dae1e52c AG |
231 | } |
232 | ||
233 | /** | |
234 | * ext4_find_goal - find a preferred place for allocation. | |
235 | * @inode: owner | |
236 | * @block: block we want | |
237 | * @partial: pointer to the last triple within a chain | |
238 | * | |
239 | * Normally this function find the preferred place for block allocation, | |
240 | * returns it. | |
241 | * Because this is only used for non-extent files, we limit the block nr | |
242 | * to 32 bits. | |
243 | */ | |
244 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, | |
245 | Indirect *partial) | |
246 | { | |
247 | ext4_fsblk_t goal; | |
248 | ||
249 | /* | |
250 | * XXX need to get goal block from mballoc's data structures | |
251 | */ | |
252 | ||
253 | goal = ext4_find_near(inode, partial); | |
254 | goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; | |
255 | return goal; | |
256 | } | |
257 | ||
258 | /** | |
259 | * ext4_blks_to_allocate - Look up the block map and count the number | |
260 | * of direct blocks need to be allocated for the given branch. | |
261 | * | |
262 | * @branch: chain of indirect blocks | |
263 | * @k: number of blocks need for indirect blocks | |
264 | * @blks: number of data blocks to be mapped. | |
265 | * @blocks_to_boundary: the offset in the indirect block | |
266 | * | |
267 | * return the total number of blocks to be allocate, including the | |
268 | * direct and indirect blocks. | |
269 | */ | |
270 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, | |
271 | int blocks_to_boundary) | |
272 | { | |
273 | unsigned int count = 0; | |
274 | ||
275 | /* | |
276 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
277 | * then it's clear blocks on that path have not allocated | |
278 | */ | |
279 | if (k > 0) { | |
280 | /* right now we don't handle cross boundary allocation */ | |
281 | if (blks < blocks_to_boundary + 1) | |
282 | count += blks; | |
283 | else | |
284 | count += blocks_to_boundary + 1; | |
285 | return count; | |
286 | } | |
287 | ||
288 | count++; | |
289 | while (count < blks && count <= blocks_to_boundary && | |
290 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
291 | count++; | |
292 | } | |
293 | return count; | |
294 | } | |
295 | ||
dae1e52c | 296 | /** |
c60990b3 TT |
297 | * ext4_alloc_branch() - allocate and set up a chain of blocks |
298 | * @handle: handle for this transaction | |
299 | * @ar: structure describing the allocation request | |
300 | * @indirect_blks: number of allocated indirect blocks | |
301 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
302 | * @branch: place to store the chain in. | |
dae1e52c AG |
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 |
321 | static int ext4_alloc_branch(handle_t *handle, |
322 | struct ext4_allocation_request *ar, | |
323 | int indirect_blks, ext4_lblk_t *offsets, | |
324 | Indirect *branch) | |
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); |
f2890730 | 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); | |
f2890730 JK |
339 | /* Simplify error cleanup... */ |
340 | branch[i+1].bh = NULL; | |
341 | } | |
781f143e TT |
342 | if (err) { |
343 | i--; | |
344 | goto failed; | |
345 | } | |
346 | branch[i].key = cpu_to_le32(new_blocks[i]); | |
347 | if (i == 0) | |
348 | continue; | |
349 | ||
a5211002 | 350 | bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]); |
dae1e52c | 351 | if (unlikely(!bh)) { |
860d21e2 | 352 | err = -ENOMEM; |
dae1e52c AG |
353 | goto failed; |
354 | } | |
dae1e52c AG |
355 | lock_buffer(bh); |
356 | BUFFER_TRACE(bh, "call get_create_access"); | |
357 | err = ext4_journal_get_create_access(handle, bh); | |
358 | if (err) { | |
dae1e52c AG |
359 | unlock_buffer(bh); |
360 | goto failed; | |
361 | } | |
362 | ||
781f143e TT |
363 | memset(bh->b_data, 0, bh->b_size); |
364 | p = branch[i].p = (__le32 *) bh->b_data + offsets[i]; | |
365 | b = new_blocks[i]; | |
366 | ||
367 | if (i == indirect_blks) | |
a5211002 | 368 | len = ar->len; |
781f143e TT |
369 | for (j = 0; j < len; j++) |
370 | *p++ = cpu_to_le32(b++); | |
371 | ||
dae1e52c AG |
372 | BUFFER_TRACE(bh, "marking uptodate"); |
373 | set_buffer_uptodate(bh); | |
374 | unlock_buffer(bh); | |
375 | ||
376 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
a5211002 | 377 | err = ext4_handle_dirty_metadata(handle, ar->inode, bh); |
dae1e52c AG |
378 | if (err) |
379 | goto failed; | |
380 | } | |
781f143e | 381 | return 0; |
dae1e52c | 382 | failed: |
f2890730 JK |
383 | if (i == indirect_blks) { |
384 | /* Free data blocks */ | |
385 | ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i], | |
386 | ar->len, 0); | |
387 | i--; | |
388 | } | |
781f143e | 389 | for (; i >= 0; i--) { |
c5c7b8dd JK |
390 | /* |
391 | * We want to ext4_forget() only freshly allocated indirect | |
f2890730 JK |
392 | * blocks. Buffer for new_blocks[i] is at branch[i+1].bh |
393 | * (buffer at branch[0].bh is indirect block / inode already | |
394 | * existing before ext4_alloc_branch() was called). Also | |
395 | * because blocks are freshly allocated, we don't need to | |
396 | * revoke them which is why we don't set | |
397 | * EXT4_FREE_BLOCKS_METADATA. | |
c5c7b8dd | 398 | */ |
f2890730 JK |
399 | ext4_free_blocks(handle, ar->inode, branch[i+1].bh, |
400 | new_blocks[i], 1, | |
401 | branch[i+1].bh ? EXT4_FREE_BLOCKS_FORGET : 0); | |
dae1e52c | 402 | } |
dae1e52c AG |
403 | return err; |
404 | } | |
405 | ||
406 | /** | |
c60990b3 | 407 | * ext4_splice_branch() - splice the allocated branch onto inode. |
dae1e52c | 408 | * @handle: handle for this transaction |
c60990b3 | 409 | * @ar: structure describing the allocation request |
dae1e52c AG |
410 | * @where: location of missing link |
411 | * @num: number of indirect blocks we are adding | |
dae1e52c AG |
412 | * |
413 | * This function fills the missing link and does all housekeeping needed in | |
414 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
415 | * chain to new block and return 0. | |
416 | */ | |
a5211002 TT |
417 | static int ext4_splice_branch(handle_t *handle, |
418 | struct ext4_allocation_request *ar, | |
419 | Indirect *where, int num) | |
dae1e52c AG |
420 | { |
421 | int i; | |
422 | int err = 0; | |
423 | ext4_fsblk_t current_block; | |
424 | ||
425 | /* | |
426 | * If we're splicing into a [td]indirect block (as opposed to the | |
427 | * inode) then we need to get write access to the [td]indirect block | |
428 | * before the splice. | |
429 | */ | |
430 | if (where->bh) { | |
431 | BUFFER_TRACE(where->bh, "get_write_access"); | |
432 | err = ext4_journal_get_write_access(handle, where->bh); | |
433 | if (err) | |
434 | goto err_out; | |
435 | } | |
436 | /* That's it */ | |
437 | ||
438 | *where->p = where->key; | |
439 | ||
440 | /* | |
441 | * Update the host buffer_head or inode to point to more just allocated | |
442 | * direct blocks blocks | |
443 | */ | |
a5211002 | 444 | if (num == 0 && ar->len > 1) { |
dae1e52c | 445 | current_block = le32_to_cpu(where->key) + 1; |
a5211002 | 446 | for (i = 1; i < ar->len; i++) |
dae1e52c AG |
447 | *(where->p + i) = cpu_to_le32(current_block++); |
448 | } | |
449 | ||
450 | /* We are done with atomic stuff, now do the rest of housekeeping */ | |
451 | /* had we spliced it onto indirect block? */ | |
452 | if (where->bh) { | |
453 | /* | |
454 | * If we spliced it onto an indirect block, we haven't | |
455 | * altered the inode. Note however that if it is being spliced | |
456 | * onto an indirect block at the very end of the file (the | |
457 | * file is growing) then we *will* alter the inode to reflect | |
458 | * the new i_size. But that is not done here - it is done in | |
459 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. | |
460 | */ | |
461 | jbd_debug(5, "splicing indirect only\n"); | |
462 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); | |
a5211002 | 463 | err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh); |
dae1e52c AG |
464 | if (err) |
465 | goto err_out; | |
466 | } else { | |
467 | /* | |
468 | * OK, we spliced it into the inode itself on a direct block. | |
469 | */ | |
4209ae12 HS |
470 | err = ext4_mark_inode_dirty(handle, ar->inode); |
471 | if (unlikely(err)) | |
472 | goto err_out; | |
dae1e52c AG |
473 | jbd_debug(5, "splicing direct\n"); |
474 | } | |
475 | return err; | |
476 | ||
477 | err_out: | |
478 | for (i = 1; i <= num; i++) { | |
479 | /* | |
480 | * branch[i].bh is newly allocated, so there is no | |
481 | * need to revoke the block, which is why we don't | |
482 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
483 | */ | |
a5211002 | 484 | ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1, |
dae1e52c AG |
485 | EXT4_FREE_BLOCKS_FORGET); |
486 | } | |
a5211002 TT |
487 | ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key), |
488 | ar->len, 0); | |
dae1e52c AG |
489 | |
490 | return err; | |
491 | } | |
492 | ||
493 | /* | |
494 | * The ext4_ind_map_blocks() function handles non-extents inodes | |
495 | * (i.e., using the traditional indirect/double-indirect i_blocks | |
496 | * scheme) for ext4_map_blocks(). | |
497 | * | |
498 | * Allocation strategy is simple: if we have to allocate something, we will | |
499 | * have to go the whole way to leaf. So let's do it before attaching anything | |
500 | * to tree, set linkage between the newborn blocks, write them if sync is | |
501 | * required, recheck the path, free and repeat if check fails, otherwise | |
502 | * set the last missing link (that will protect us from any truncate-generated | |
503 | * removals - all blocks on the path are immune now) and possibly force the | |
504 | * write on the parent block. | |
505 | * That has a nice additional property: no special recovery from the failed | |
506 | * allocations is needed - we simply release blocks and do not touch anything | |
507 | * reachable from inode. | |
508 | * | |
509 | * `handle' can be NULL if create == 0. | |
510 | * | |
511 | * return > 0, # of blocks mapped or allocated. | |
512 | * return = 0, if plain lookup failed. | |
513 | * return < 0, error case. | |
514 | * | |
515 | * The ext4_ind_get_blocks() function should be called with | |
516 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem | |
517 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or | |
518 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system | |
519 | * blocks. | |
520 | */ | |
521 | int ext4_ind_map_blocks(handle_t *handle, struct inode *inode, | |
522 | struct ext4_map_blocks *map, | |
523 | int flags) | |
524 | { | |
a5211002 | 525 | struct ext4_allocation_request ar; |
dae1e52c AG |
526 | int err = -EIO; |
527 | ext4_lblk_t offsets[4]; | |
528 | Indirect chain[4]; | |
529 | Indirect *partial; | |
dae1e52c AG |
530 | int indirect_blks; |
531 | int blocks_to_boundary = 0; | |
532 | int depth; | |
533 | int count = 0; | |
534 | ext4_fsblk_t first_block = 0; | |
535 | ||
536 | trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); | |
537 | J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); | |
538 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); | |
539 | depth = ext4_block_to_path(inode, map->m_lblk, offsets, | |
540 | &blocks_to_boundary); | |
541 | ||
542 | if (depth == 0) | |
543 | goto out; | |
544 | ||
545 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); | |
546 | ||
547 | /* Simplest case - block found, no allocation needed */ | |
548 | if (!partial) { | |
549 | first_block = le32_to_cpu(chain[depth - 1].key); | |
550 | count++; | |
551 | /*map more blocks*/ | |
552 | while (count < map->m_len && count <= blocks_to_boundary) { | |
553 | ext4_fsblk_t blk; | |
554 | ||
555 | blk = le32_to_cpu(*(chain[depth-1].p + count)); | |
556 | ||
557 | if (blk == first_block + count) | |
558 | count++; | |
559 | else | |
560 | break; | |
561 | } | |
562 | goto got_it; | |
563 | } | |
564 | ||
facab4d9 JK |
565 | /* Next simple case - plain lookup failed */ |
566 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
567 | unsigned epb = inode->i_sb->s_blocksize / sizeof(u32); | |
568 | int i; | |
569 | ||
2ee3ee06 JK |
570 | /* |
571 | * Count number blocks in a subtree under 'partial'. At each | |
572 | * level we count number of complete empty subtrees beyond | |
573 | * current offset and then descend into the subtree only | |
574 | * partially beyond current offset. | |
575 | */ | |
576 | count = 0; | |
577 | for (i = partial - chain + 1; i < depth; i++) | |
578 | count = count * epb + (epb - offsets[i] - 1); | |
579 | count++; | |
facab4d9 JK |
580 | /* Fill in size of a hole we found */ |
581 | map->m_pblk = 0; | |
582 | map->m_len = min_t(unsigned int, map->m_len, count); | |
583 | goto cleanup; | |
584 | } | |
585 | ||
586 | /* Failed read of indirect block */ | |
587 | if (err == -EIO) | |
dae1e52c AG |
588 | goto cleanup; |
589 | ||
590 | /* | |
591 | * Okay, we need to do block allocation. | |
592 | */ | |
e2b911c5 | 593 | if (ext4_has_feature_bigalloc(inode->i_sb)) { |
bab08ab9 TT |
594 | EXT4_ERROR_INODE(inode, "Can't allocate blocks for " |
595 | "non-extent mapped inodes with bigalloc"); | |
6a797d27 | 596 | return -EFSCORRUPTED; |
bab08ab9 TT |
597 | } |
598 | ||
a5211002 TT |
599 | /* Set up for the direct block allocation */ |
600 | memset(&ar, 0, sizeof(ar)); | |
601 | ar.inode = inode; | |
602 | ar.logical = map->m_lblk; | |
603 | if (S_ISREG(inode->i_mode)) | |
604 | ar.flags = EXT4_MB_HINT_DATA; | |
e3cf5d5d TT |
605 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
606 | ar.flags |= EXT4_MB_DELALLOC_RESERVED; | |
c5e298ae TT |
607 | if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
608 | ar.flags |= EXT4_MB_USE_RESERVED; | |
a5211002 TT |
609 | |
610 | ar.goal = ext4_find_goal(inode, map->m_lblk, partial); | |
dae1e52c AG |
611 | |
612 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
613 | indirect_blks = (chain + depth) - partial - 1; | |
614 | ||
615 | /* | |
616 | * Next look up the indirect map to count the totoal number of | |
617 | * direct blocks to allocate for this branch. | |
618 | */ | |
a5211002 TT |
619 | ar.len = ext4_blks_to_allocate(partial, indirect_blks, |
620 | map->m_len, blocks_to_boundary); | |
621 | ||
dae1e52c AG |
622 | /* |
623 | * Block out ext4_truncate while we alter the tree | |
624 | */ | |
a5211002 | 625 | err = ext4_alloc_branch(handle, &ar, indirect_blks, |
dae1e52c AG |
626 | offsets + (partial - chain), partial); |
627 | ||
628 | /* | |
629 | * The ext4_splice_branch call will free and forget any buffers | |
630 | * on the new chain if there is a failure, but that risks using | |
631 | * up transaction credits, especially for bitmaps where the | |
632 | * credits cannot be returned. Can we handle this somehow? We | |
633 | * may need to return -EAGAIN upwards in the worst case. --sct | |
634 | */ | |
635 | if (!err) | |
a5211002 | 636 | err = ext4_splice_branch(handle, &ar, partial, indirect_blks); |
dae1e52c AG |
637 | if (err) |
638 | goto cleanup; | |
639 | ||
640 | map->m_flags |= EXT4_MAP_NEW; | |
641 | ||
642 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
a5211002 | 643 | count = ar.len; |
dae1e52c AG |
644 | got_it: |
645 | map->m_flags |= EXT4_MAP_MAPPED; | |
646 | map->m_pblk = le32_to_cpu(chain[depth-1].key); | |
647 | map->m_len = count; | |
648 | if (count > blocks_to_boundary) | |
649 | map->m_flags |= EXT4_MAP_BOUNDARY; | |
650 | err = count; | |
651 | /* Clean up and exit */ | |
652 | partial = chain + depth - 1; /* the whole chain */ | |
653 | cleanup: | |
654 | while (partial > chain) { | |
655 | BUFFER_TRACE(partial->bh, "call brelse"); | |
656 | brelse(partial->bh); | |
657 | partial--; | |
658 | } | |
659 | out: | |
21ddd568 | 660 | trace_ext4_ind_map_blocks_exit(inode, flags, map, err); |
dae1e52c AG |
661 | return err; |
662 | } | |
663 | ||
fa55a0ed JK |
664 | /* |
665 | * Calculate number of indirect blocks touched by mapping @nrblocks logically | |
666 | * contiguous blocks | |
667 | */ | |
668 | int ext4_ind_trans_blocks(struct inode *inode, int nrblocks) | |
dae1e52c | 669 | { |
dae1e52c | 670 | /* |
fa55a0ed JK |
671 | * With N contiguous data blocks, we need at most |
672 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks, | |
673 | * 2 dindirect blocks, and 1 tindirect block | |
dae1e52c | 674 | */ |
fa55a0ed | 675 | return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4; |
dae1e52c AG |
676 | } |
677 | ||
a4130367 JK |
678 | static int ext4_ind_trunc_restart_fn(handle_t *handle, struct inode *inode, |
679 | struct buffer_head *bh, int *dropped) | |
680 | { | |
681 | int err; | |
682 | ||
683 | if (bh) { | |
684 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
685 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
686 | if (unlikely(err)) | |
687 | return err; | |
688 | } | |
689 | err = ext4_mark_inode_dirty(handle, inode); | |
690 | if (unlikely(err)) | |
691 | return err; | |
692 | /* | |
693 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this | |
694 | * moment, get_block can be called only for blocks inside i_size since | |
695 | * page cache has been already dropped and writes are blocked by | |
696 | * i_mutex. So we can safely drop the i_data_sem here. | |
697 | */ | |
698 | BUG_ON(EXT4_JOURNAL(inode) == NULL); | |
699 | ext4_discard_preallocations(inode); | |
700 | up_write(&EXT4_I(inode)->i_data_sem); | |
701 | *dropped = 1; | |
702 | return 0; | |
703 | } | |
704 | ||
dae1e52c AG |
705 | /* |
706 | * Truncate transactions can be complex and absolutely huge. So we need to | |
707 | * be able to restart the transaction at a conventient checkpoint to make | |
708 | * sure we don't overflow the journal. | |
709 | * | |
819c4920 | 710 | * Try to extend this transaction for the purposes of truncation. If |
a4130367 | 711 | * extend fails, we restart transaction. |
dae1e52c | 712 | */ |
a4130367 JK |
713 | static int ext4_ind_truncate_ensure_credits(handle_t *handle, |
714 | struct inode *inode, | |
83448bdf JK |
715 | struct buffer_head *bh, |
716 | int revoke_creds) | |
dae1e52c | 717 | { |
a4130367 JK |
718 | int ret; |
719 | int dropped = 0; | |
720 | ||
721 | ret = ext4_journal_ensure_credits_fn(handle, EXT4_RESERVE_TRANS_BLOCKS, | |
83448bdf | 722 | ext4_blocks_for_truncate(inode), revoke_creds, |
a4130367 JK |
723 | ext4_ind_trunc_restart_fn(handle, inode, bh, &dropped)); |
724 | if (dropped) | |
725 | down_write(&EXT4_I(inode)->i_data_sem); | |
726 | if (ret <= 0) | |
727 | return ret; | |
728 | if (bh) { | |
729 | BUFFER_TRACE(bh, "retaking write access"); | |
730 | ret = ext4_journal_get_write_access(handle, bh); | |
731 | if (unlikely(ret)) | |
732 | return ret; | |
733 | } | |
734 | return 0; | |
dae1e52c AG |
735 | } |
736 | ||
737 | /* | |
738 | * Probably it should be a library function... search for first non-zero word | |
739 | * or memcmp with zero_page, whatever is better for particular architecture. | |
740 | * Linus? | |
741 | */ | |
742 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
743 | { | |
744 | while (p < q) | |
745 | if (*p++) | |
746 | return 0; | |
747 | return 1; | |
748 | } | |
749 | ||
750 | /** | |
751 | * ext4_find_shared - find the indirect blocks for partial truncation. | |
752 | * @inode: inode in question | |
753 | * @depth: depth of the affected branch | |
754 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) | |
755 | * @chain: place to store the pointers to partial indirect blocks | |
756 | * @top: place to the (detached) top of branch | |
757 | * | |
758 | * This is a helper function used by ext4_truncate(). | |
759 | * | |
760 | * When we do truncate() we may have to clean the ends of several | |
761 | * indirect blocks but leave the blocks themselves alive. Block is | |
762 | * partially truncated if some data below the new i_size is referred | |
763 | * from it (and it is on the path to the first completely truncated | |
764 | * data block, indeed). We have to free the top of that path along | |
765 | * with everything to the right of the path. Since no allocation | |
766 | * past the truncation point is possible until ext4_truncate() | |
767 | * finishes, we may safely do the latter, but top of branch may | |
768 | * require special attention - pageout below the truncation point | |
769 | * might try to populate it. | |
770 | * | |
771 | * We atomically detach the top of branch from the tree, store the | |
772 | * block number of its root in *@top, pointers to buffer_heads of | |
773 | * partially truncated blocks - in @chain[].bh and pointers to | |
774 | * their last elements that should not be removed - in | |
775 | * @chain[].p. Return value is the pointer to last filled element | |
776 | * of @chain. | |
777 | * | |
778 | * The work left to caller to do the actual freeing of subtrees: | |
779 | * a) free the subtree starting from *@top | |
780 | * b) free the subtrees whose roots are stored in | |
781 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
782 | * c) free the subtrees growing from the inode past the @chain[0]. | |
783 | * (no partially truncated stuff there). */ | |
784 | ||
785 | static Indirect *ext4_find_shared(struct inode *inode, int depth, | |
786 | ext4_lblk_t offsets[4], Indirect chain[4], | |
787 | __le32 *top) | |
788 | { | |
789 | Indirect *partial, *p; | |
790 | int k, err; | |
791 | ||
792 | *top = 0; | |
793 | /* Make k index the deepest non-null offset + 1 */ | |
794 | for (k = depth; k > 1 && !offsets[k-1]; k--) | |
795 | ; | |
796 | partial = ext4_get_branch(inode, k, offsets, chain, &err); | |
797 | /* Writer: pointers */ | |
798 | if (!partial) | |
799 | partial = chain + k-1; | |
800 | /* | |
801 | * If the branch acquired continuation since we've looked at it - | |
802 | * fine, it should all survive and (new) top doesn't belong to us. | |
803 | */ | |
804 | if (!partial->key && *partial->p) | |
805 | /* Writer: end */ | |
806 | goto no_top; | |
807 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) | |
808 | ; | |
809 | /* | |
810 | * OK, we've found the last block that must survive. The rest of our | |
811 | * branch should be detached before unlocking. However, if that rest | |
812 | * of branch is all ours and does not grow immediately from the inode | |
813 | * it's easier to cheat and just decrement partial->p. | |
814 | */ | |
815 | if (p == chain + k - 1 && p > chain) { | |
816 | p->p--; | |
817 | } else { | |
818 | *top = *p->p; | |
819 | /* Nope, don't do this in ext4. Must leave the tree intact */ | |
820 | #if 0 | |
821 | *p->p = 0; | |
822 | #endif | |
823 | } | |
824 | /* Writer: end */ | |
825 | ||
826 | while (partial > p) { | |
827 | brelse(partial->bh); | |
828 | partial--; | |
829 | } | |
830 | no_top: | |
831 | return partial; | |
832 | } | |
833 | ||
834 | /* | |
835 | * Zero a number of block pointers in either an inode or an indirect block. | |
836 | * If we restart the transaction we must again get write access to the | |
837 | * indirect block for further modification. | |
838 | * | |
839 | * We release `count' blocks on disk, but (last - first) may be greater | |
840 | * than `count' because there can be holes in there. | |
841 | * | |
842 | * Return 0 on success, 1 on invalid block range | |
843 | * and < 0 on fatal error. | |
844 | */ | |
845 | static int ext4_clear_blocks(handle_t *handle, struct inode *inode, | |
846 | struct buffer_head *bh, | |
847 | ext4_fsblk_t block_to_free, | |
848 | unsigned long count, __le32 *first, | |
849 | __le32 *last) | |
850 | { | |
851 | __le32 *p; | |
981250ca | 852 | int flags = EXT4_FREE_BLOCKS_VALIDATED; |
dae1e52c AG |
853 | int err; |
854 | ||
ddfa17e4 TE |
855 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) || |
856 | ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE)) | |
981250ca TT |
857 | flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA; |
858 | else if (ext4_should_journal_data(inode)) | |
859 | flags |= EXT4_FREE_BLOCKS_FORGET; | |
dae1e52c AG |
860 | |
861 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free, | |
862 | count)) { | |
863 | EXT4_ERROR_INODE(inode, "attempt to clear invalid " | |
864 | "blocks %llu len %lu", | |
865 | (unsigned long long) block_to_free, count); | |
866 | return 1; | |
867 | } | |
868 | ||
83448bdf JK |
869 | err = ext4_ind_truncate_ensure_credits(handle, inode, bh, |
870 | ext4_free_data_revoke_credits(inode, count)); | |
a4130367 JK |
871 | if (err < 0) |
872 | goto out_err; | |
dae1e52c AG |
873 | |
874 | for (p = first; p < last; p++) | |
875 | *p = 0; | |
876 | ||
877 | ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags); | |
878 | return 0; | |
879 | out_err: | |
880 | ext4_std_error(inode->i_sb, err); | |
881 | return err; | |
882 | } | |
883 | ||
884 | /** | |
885 | * ext4_free_data - free a list of data blocks | |
886 | * @handle: handle for this transaction | |
887 | * @inode: inode we are dealing with | |
888 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
889 | * @first: array of block numbers | |
890 | * @last: points immediately past the end of array | |
891 | * | |
892 | * We are freeing all blocks referred from that array (numbers are stored as | |
893 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
894 | * | |
895 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
896 | * blocks are contiguous then releasing them at one time will only affect one | |
897 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
898 | * actually use a lot of journal space. | |
899 | * | |
900 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
901 | * block pointers. | |
902 | */ | |
903 | static void ext4_free_data(handle_t *handle, struct inode *inode, | |
904 | struct buffer_head *this_bh, | |
905 | __le32 *first, __le32 *last) | |
906 | { | |
907 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ | |
908 | unsigned long count = 0; /* Number of blocks in the run */ | |
909 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
910 | corresponding to | |
911 | block_to_free */ | |
912 | ext4_fsblk_t nr; /* Current block # */ | |
913 | __le32 *p; /* Pointer into inode/ind | |
914 | for current block */ | |
915 | int err = 0; | |
916 | ||
917 | if (this_bh) { /* For indirect block */ | |
918 | BUFFER_TRACE(this_bh, "get_write_access"); | |
919 | err = ext4_journal_get_write_access(handle, this_bh); | |
920 | /* Important: if we can't update the indirect pointers | |
921 | * to the blocks, we can't free them. */ | |
922 | if (err) | |
923 | return; | |
924 | } | |
925 | ||
926 | for (p = first; p < last; p++) { | |
927 | nr = le32_to_cpu(*p); | |
928 | if (nr) { | |
929 | /* accumulate blocks to free if they're contiguous */ | |
930 | if (count == 0) { | |
931 | block_to_free = nr; | |
932 | block_to_free_p = p; | |
933 | count = 1; | |
934 | } else if (nr == block_to_free + count) { | |
935 | count++; | |
936 | } else { | |
937 | err = ext4_clear_blocks(handle, inode, this_bh, | |
938 | block_to_free, count, | |
939 | block_to_free_p, p); | |
940 | if (err) | |
941 | break; | |
942 | block_to_free = nr; | |
943 | block_to_free_p = p; | |
944 | count = 1; | |
945 | } | |
946 | } | |
947 | } | |
948 | ||
949 | if (!err && count > 0) | |
950 | err = ext4_clear_blocks(handle, inode, this_bh, block_to_free, | |
951 | count, block_to_free_p, p); | |
952 | if (err < 0) | |
953 | /* fatal error */ | |
954 | return; | |
955 | ||
956 | if (this_bh) { | |
957 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); | |
958 | ||
959 | /* | |
960 | * The buffer head should have an attached journal head at this | |
961 | * point. However, if the data is corrupted and an indirect | |
962 | * block pointed to itself, it would have been detached when | |
963 | * the block was cleared. Check for this instead of OOPSing. | |
964 | */ | |
965 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) | |
966 | ext4_handle_dirty_metadata(handle, inode, this_bh); | |
967 | else | |
968 | EXT4_ERROR_INODE(inode, | |
969 | "circular indirect block detected at " | |
970 | "block %llu", | |
971 | (unsigned long long) this_bh->b_blocknr); | |
972 | } | |
973 | } | |
974 | ||
975 | /** | |
976 | * ext4_free_branches - free an array of branches | |
977 | * @handle: JBD handle for this transaction | |
978 | * @inode: inode we are dealing with | |
979 | * @parent_bh: the buffer_head which contains *@first and *@last | |
980 | * @first: array of block numbers | |
981 | * @last: pointer immediately past the end of array | |
982 | * @depth: depth of the branches to free | |
983 | * | |
984 | * We are freeing all blocks referred from these branches (numbers are | |
985 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
986 | * appropriately. | |
987 | */ | |
988 | static void ext4_free_branches(handle_t *handle, struct inode *inode, | |
989 | struct buffer_head *parent_bh, | |
990 | __le32 *first, __le32 *last, int depth) | |
991 | { | |
992 | ext4_fsblk_t nr; | |
993 | __le32 *p; | |
994 | ||
995 | if (ext4_handle_is_aborted(handle)) | |
996 | return; | |
997 | ||
998 | if (depth--) { | |
999 | struct buffer_head *bh; | |
1000 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1001 | p = last; | |
1002 | while (--p >= first) { | |
1003 | nr = le32_to_cpu(*p); | |
1004 | if (!nr) | |
1005 | continue; /* A hole */ | |
1006 | ||
1007 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), | |
1008 | nr, 1)) { | |
1009 | EXT4_ERROR_INODE(inode, | |
1010 | "invalid indirect mapped " | |
1011 | "block %lu (level %d)", | |
1012 | (unsigned long) nr, depth); | |
1013 | break; | |
1014 | } | |
1015 | ||
1016 | /* Go read the buffer for the next level down */ | |
1017 | bh = sb_bread(inode->i_sb, nr); | |
1018 | ||
1019 | /* | |
1020 | * A read failure? Report error and clear slot | |
1021 | * (should be rare). | |
1022 | */ | |
1023 | if (!bh) { | |
54d3adbc | 1024 | ext4_error_inode_block(inode, nr, EIO, |
dae1e52c AG |
1025 | "Read failure"); |
1026 | continue; | |
1027 | } | |
1028 | ||
1029 | /* This zaps the entire block. Bottom up. */ | |
1030 | BUFFER_TRACE(bh, "free child branches"); | |
1031 | ext4_free_branches(handle, inode, bh, | |
1032 | (__le32 *) bh->b_data, | |
1033 | (__le32 *) bh->b_data + addr_per_block, | |
1034 | depth); | |
1035 | brelse(bh); | |
1036 | ||
1037 | /* | |
1038 | * Everything below this this pointer has been | |
1039 | * released. Now let this top-of-subtree go. | |
1040 | * | |
1041 | * We want the freeing of this indirect block to be | |
1042 | * atomic in the journal with the updating of the | |
1043 | * bitmap block which owns it. So make some room in | |
1044 | * the journal. | |
1045 | * | |
1046 | * We zero the parent pointer *after* freeing its | |
1047 | * pointee in the bitmaps, so if extend_transaction() | |
1048 | * for some reason fails to put the bitmap changes and | |
1049 | * the release into the same transaction, recovery | |
1050 | * will merely complain about releasing a free block, | |
1051 | * rather than leaking blocks. | |
1052 | */ | |
1053 | if (ext4_handle_is_aborted(handle)) | |
1054 | return; | |
a4130367 | 1055 | if (ext4_ind_truncate_ensure_credits(handle, inode, |
83448bdf JK |
1056 | NULL, |
1057 | ext4_free_metadata_revoke_credits( | |
1058 | inode->i_sb, 1)) < 0) | |
a4130367 | 1059 | return; |
dae1e52c AG |
1060 | |
1061 | /* | |
1062 | * The forget flag here is critical because if | |
1063 | * we are journaling (and not doing data | |
1064 | * journaling), we have to make sure a revoke | |
1065 | * record is written to prevent the journal | |
1066 | * replay from overwriting the (former) | |
1067 | * indirect block if it gets reallocated as a | |
1068 | * data block. This must happen in the same | |
1069 | * transaction where the data blocks are | |
1070 | * actually freed. | |
1071 | */ | |
1072 | ext4_free_blocks(handle, inode, NULL, nr, 1, | |
1073 | EXT4_FREE_BLOCKS_METADATA| | |
1074 | EXT4_FREE_BLOCKS_FORGET); | |
1075 | ||
1076 | if (parent_bh) { | |
1077 | /* | |
1078 | * The block which we have just freed is | |
1079 | * pointed to by an indirect block: journal it | |
1080 | */ | |
1081 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
1082 | if (!ext4_journal_get_write_access(handle, | |
1083 | parent_bh)){ | |
1084 | *p = 0; | |
1085 | BUFFER_TRACE(parent_bh, | |
1086 | "call ext4_handle_dirty_metadata"); | |
1087 | ext4_handle_dirty_metadata(handle, | |
1088 | inode, | |
1089 | parent_bh); | |
1090 | } | |
1091 | } | |
1092 | } | |
1093 | } else { | |
1094 | /* We have reached the bottom of the tree. */ | |
1095 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
1096 | ext4_free_data(handle, inode, parent_bh, first, last); | |
1097 | } | |
1098 | } | |
1099 | ||
819c4920 | 1100 | void ext4_ind_truncate(handle_t *handle, struct inode *inode) |
dae1e52c | 1101 | { |
dae1e52c AG |
1102 | struct ext4_inode_info *ei = EXT4_I(inode); |
1103 | __le32 *i_data = ei->i_data; | |
1104 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
dae1e52c AG |
1105 | ext4_lblk_t offsets[4]; |
1106 | Indirect chain[4]; | |
1107 | Indirect *partial; | |
1108 | __le32 nr = 0; | |
1109 | int n = 0; | |
1110 | ext4_lblk_t last_block, max_block; | |
1111 | unsigned blocksize = inode->i_sb->s_blocksize; | |
dae1e52c AG |
1112 | |
1113 | last_block = (inode->i_size + blocksize-1) | |
1114 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); | |
1115 | max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1) | |
1116 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); | |
1117 | ||
dae1e52c AG |
1118 | if (last_block != max_block) { |
1119 | n = ext4_block_to_path(inode, last_block, offsets, NULL); | |
1120 | if (n == 0) | |
819c4920 | 1121 | return; |
dae1e52c AG |
1122 | } |
1123 | ||
51865fda | 1124 | ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block); |
dae1e52c AG |
1125 | |
1126 | /* | |
1127 | * The orphan list entry will now protect us from any crash which | |
1128 | * occurs before the truncate completes, so it is now safe to propagate | |
1129 | * the new, shorter inode size (held for now in i_size) into the | |
1130 | * on-disk inode. We do this via i_disksize, which is the value which | |
1131 | * ext4 *really* writes onto the disk inode. | |
1132 | */ | |
1133 | ei->i_disksize = inode->i_size; | |
1134 | ||
1135 | if (last_block == max_block) { | |
1136 | /* | |
1137 | * It is unnecessary to free any data blocks if last_block is | |
1138 | * equal to the indirect block limit. | |
1139 | */ | |
819c4920 | 1140 | return; |
dae1e52c AG |
1141 | } else if (n == 1) { /* direct blocks */ |
1142 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], | |
1143 | i_data + EXT4_NDIR_BLOCKS); | |
1144 | goto do_indirects; | |
1145 | } | |
1146 | ||
1147 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); | |
1148 | /* Kill the top of shared branch (not detached) */ | |
1149 | if (nr) { | |
1150 | if (partial == chain) { | |
1151 | /* Shared branch grows from the inode */ | |
1152 | ext4_free_branches(handle, inode, NULL, | |
1153 | &nr, &nr+1, (chain+n-1) - partial); | |
1154 | *partial->p = 0; | |
1155 | /* | |
1156 | * We mark the inode dirty prior to restart, | |
1157 | * and prior to stop. No need for it here. | |
1158 | */ | |
1159 | } else { | |
1160 | /* Shared branch grows from an indirect block */ | |
1161 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
1162 | ext4_free_branches(handle, inode, partial->bh, | |
1163 | partial->p, | |
1164 | partial->p+1, (chain+n-1) - partial); | |
1165 | } | |
1166 | } | |
1167 | /* Clear the ends of indirect blocks on the shared branch */ | |
1168 | while (partial > chain) { | |
1169 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, | |
1170 | (__le32*)partial->bh->b_data+addr_per_block, | |
1171 | (chain+n-1) - partial); | |
1172 | BUFFER_TRACE(partial->bh, "call brelse"); | |
1173 | brelse(partial->bh); | |
1174 | partial--; | |
1175 | } | |
1176 | do_indirects: | |
1177 | /* Kill the remaining (whole) subtrees */ | |
1178 | switch (offsets[0]) { | |
1179 | default: | |
1180 | nr = i_data[EXT4_IND_BLOCK]; | |
1181 | if (nr) { | |
1182 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); | |
1183 | i_data[EXT4_IND_BLOCK] = 0; | |
1184 | } | |
793bc518 | 1185 | /* fall through */ |
dae1e52c AG |
1186 | case EXT4_IND_BLOCK: |
1187 | nr = i_data[EXT4_DIND_BLOCK]; | |
1188 | if (nr) { | |
1189 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); | |
1190 | i_data[EXT4_DIND_BLOCK] = 0; | |
1191 | } | |
793bc518 | 1192 | /* fall through */ |
dae1e52c AG |
1193 | case EXT4_DIND_BLOCK: |
1194 | nr = i_data[EXT4_TIND_BLOCK]; | |
1195 | if (nr) { | |
1196 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); | |
1197 | i_data[EXT4_TIND_BLOCK] = 0; | |
1198 | } | |
793bc518 | 1199 | /* fall through */ |
dae1e52c AG |
1200 | case EXT4_TIND_BLOCK: |
1201 | ; | |
1202 | } | |
dae1e52c AG |
1203 | } |
1204 | ||
4f579ae7 LC |
1205 | /** |
1206 | * ext4_ind_remove_space - remove space from the range | |
1207 | * @handle: JBD handle for this transaction | |
1208 | * @inode: inode we are dealing with | |
1209 | * @start: First block to remove | |
1210 | * @end: One block after the last block to remove (exclusive) | |
1211 | * | |
1212 | * Free the blocks in the defined range (end is exclusive endpoint of | |
1213 | * range). This is used by ext4_punch_hole(). | |
1214 | */ | |
1215 | int ext4_ind_remove_space(handle_t *handle, struct inode *inode, | |
1216 | ext4_lblk_t start, ext4_lblk_t end) | |
8bad6fc8 | 1217 | { |
4f579ae7 LC |
1218 | struct ext4_inode_info *ei = EXT4_I(inode); |
1219 | __le32 *i_data = ei->i_data; | |
8bad6fc8 | 1220 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
4f579ae7 LC |
1221 | ext4_lblk_t offsets[4], offsets2[4]; |
1222 | Indirect chain[4], chain2[4]; | |
1223 | Indirect *partial, *partial2; | |
5e86bdda | 1224 | Indirect *p = NULL, *p2 = NULL; |
4f579ae7 LC |
1225 | ext4_lblk_t max_block; |
1226 | __le32 nr = 0, nr2 = 0; | |
1227 | int n = 0, n2 = 0; | |
1228 | unsigned blocksize = inode->i_sb->s_blocksize; | |
a93cd4cf | 1229 | |
4f579ae7 LC |
1230 | max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1) |
1231 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); | |
1232 | if (end >= max_block) | |
1233 | end = max_block; | |
1234 | if ((start >= end) || (start > max_block)) | |
1235 | return 0; | |
1236 | ||
1237 | n = ext4_block_to_path(inode, start, offsets, NULL); | |
1238 | n2 = ext4_block_to_path(inode, end, offsets2, NULL); | |
1239 | ||
1240 | BUG_ON(n > n2); | |
1241 | ||
1242 | if ((n == 1) && (n == n2)) { | |
1243 | /* We're punching only within direct block range */ | |
1244 | ext4_free_data(handle, inode, NULL, i_data + offsets[0], | |
1245 | i_data + offsets2[0]); | |
1246 | return 0; | |
1247 | } else if (n2 > n) { | |
1248 | /* | |
1249 | * Start and end are on a different levels so we're going to | |
1250 | * free partial block at start, and partial block at end of | |
1251 | * the range. If there are some levels in between then | |
1252 | * do_indirects label will take care of that. | |
1253 | */ | |
1254 | ||
1255 | if (n == 1) { | |
1256 | /* | |
1257 | * Start is at the direct block level, free | |
1258 | * everything to the end of the level. | |
1259 | */ | |
1260 | ext4_free_data(handle, inode, NULL, i_data + offsets[0], | |
1261 | i_data + EXT4_NDIR_BLOCKS); | |
1262 | goto end_range; | |
1263 | } | |
1264 | ||
1265 | ||
5e86bdda | 1266 | partial = p = ext4_find_shared(inode, n, offsets, chain, &nr); |
4f579ae7 LC |
1267 | if (nr) { |
1268 | if (partial == chain) { | |
1269 | /* Shared branch grows from the inode */ | |
1270 | ext4_free_branches(handle, inode, NULL, | |
1271 | &nr, &nr+1, (chain+n-1) - partial); | |
1272 | *partial->p = 0; | |
a93cd4cf | 1273 | } else { |
4f579ae7 LC |
1274 | /* Shared branch grows from an indirect block */ |
1275 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
1276 | ext4_free_branches(handle, inode, partial->bh, | |
1277 | partial->p, | |
1278 | partial->p+1, (chain+n-1) - partial); | |
a93cd4cf | 1279 | } |
4f579ae7 LC |
1280 | } |
1281 | ||
1282 | /* | |
1283 | * Clear the ends of indirect blocks on the shared branch | |
1284 | * at the start of the range | |
1285 | */ | |
1286 | while (partial > chain) { | |
1287 | ext4_free_branches(handle, inode, partial->bh, | |
1288 | partial->p + 1, | |
1289 | (__le32 *)partial->bh->b_data+addr_per_block, | |
1290 | (chain+n-1) - partial); | |
4f579ae7 LC |
1291 | partial--; |
1292 | } | |
1293 | ||
1294 | end_range: | |
5e86bdda | 1295 | partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2); |
4f579ae7 LC |
1296 | if (nr2) { |
1297 | if (partial2 == chain2) { | |
1298 | /* | |
1299 | * Remember, end is exclusive so here we're at | |
1300 | * the start of the next level we're not going | |
1301 | * to free. Everything was covered by the start | |
1302 | * of the range. | |
1303 | */ | |
6f30b7e3 | 1304 | goto do_indirects; |
8bad6fc8 | 1305 | } |
4f579ae7 LC |
1306 | } else { |
1307 | /* | |
1308 | * ext4_find_shared returns Indirect structure which | |
1309 | * points to the last element which should not be | |
1310 | * removed by truncate. But this is end of the range | |
1311 | * in punch_hole so we need to point to the next element | |
1312 | */ | |
1313 | partial2->p++; | |
8bad6fc8 | 1314 | } |
4f579ae7 LC |
1315 | |
1316 | /* | |
1317 | * Clear the ends of indirect blocks on the shared branch | |
1318 | * at the end of the range | |
1319 | */ | |
1320 | while (partial2 > chain2) { | |
1321 | ext4_free_branches(handle, inode, partial2->bh, | |
1322 | (__le32 *)partial2->bh->b_data, | |
1323 | partial2->p, | |
1324 | (chain2+n2-1) - partial2); | |
4f579ae7 | 1325 | partial2--; |
8bad6fc8 | 1326 | } |
4f579ae7 | 1327 | goto do_indirects; |
8bad6fc8 ZL |
1328 | } |
1329 | ||
4f579ae7 | 1330 | /* Punch happened within the same level (n == n2) */ |
5e86bdda | 1331 | partial = p = ext4_find_shared(inode, n, offsets, chain, &nr); |
1332 | partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2); | |
6f30b7e3 OS |
1333 | |
1334 | /* Free top, but only if partial2 isn't its subtree. */ | |
1335 | if (nr) { | |
1336 | int level = min(partial - chain, partial2 - chain2); | |
1337 | int i; | |
1338 | int subtree = 1; | |
1339 | ||
1340 | for (i = 0; i <= level; i++) { | |
1341 | if (offsets[i] != offsets2[i]) { | |
1342 | subtree = 0; | |
1343 | break; | |
1344 | } | |
1345 | } | |
1346 | ||
1347 | if (!subtree) { | |
1348 | if (partial == chain) { | |
1349 | /* Shared branch grows from the inode */ | |
1350 | ext4_free_branches(handle, inode, NULL, | |
1351 | &nr, &nr+1, | |
1352 | (chain+n-1) - partial); | |
1353 | *partial->p = 0; | |
1354 | } else { | |
1355 | /* Shared branch grows from an indirect block */ | |
1356 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
4f579ae7 | 1357 | ext4_free_branches(handle, inode, partial->bh, |
6f30b7e3 OS |
1358 | partial->p, |
1359 | partial->p+1, | |
4f579ae7 | 1360 | (chain+n-1) - partial); |
4f579ae7 | 1361 | } |
8bad6fc8 | 1362 | } |
6f30b7e3 OS |
1363 | } |
1364 | ||
1365 | if (!nr2) { | |
4f579ae7 | 1366 | /* |
6f30b7e3 OS |
1367 | * ext4_find_shared returns Indirect structure which |
1368 | * points to the last element which should not be | |
1369 | * removed by truncate. But this is end of the range | |
1370 | * in punch_hole so we need to point to the next element | |
4f579ae7 | 1371 | */ |
6f30b7e3 OS |
1372 | partial2->p++; |
1373 | } | |
1374 | ||
1375 | while (partial > chain || partial2 > chain2) { | |
1376 | int depth = (chain+n-1) - partial; | |
1377 | int depth2 = (chain2+n2-1) - partial2; | |
1378 | ||
1379 | if (partial > chain && partial2 > chain2 && | |
1380 | partial->bh->b_blocknr == partial2->bh->b_blocknr) { | |
1381 | /* | |
1382 | * We've converged on the same block. Clear the range, | |
1383 | * then we're done. | |
1384 | */ | |
4f579ae7 | 1385 | ext4_free_branches(handle, inode, partial->bh, |
6f30b7e3 OS |
1386 | partial->p + 1, |
1387 | partial2->p, | |
1388 | (chain+n-1) - partial); | |
5e86bdda | 1389 | goto cleanup; |
4f579ae7 | 1390 | } |
6f30b7e3 | 1391 | |
4f579ae7 | 1392 | /* |
6f30b7e3 OS |
1393 | * The start and end partial branches may not be at the same |
1394 | * level even though the punch happened within one level. So, we | |
1395 | * give them a chance to arrive at the same level, then walk | |
1396 | * them in step with each other until we converge on the same | |
1397 | * block. | |
4f579ae7 | 1398 | */ |
6f30b7e3 OS |
1399 | if (partial > chain && depth <= depth2) { |
1400 | ext4_free_branches(handle, inode, partial->bh, | |
1401 | partial->p + 1, | |
1402 | (__le32 *)partial->bh->b_data+addr_per_block, | |
1403 | (chain+n-1) - partial); | |
6f30b7e3 OS |
1404 | partial--; |
1405 | } | |
1406 | if (partial2 > chain2 && depth2 <= depth) { | |
4f579ae7 LC |
1407 | ext4_free_branches(handle, inode, partial2->bh, |
1408 | (__le32 *)partial2->bh->b_data, | |
1409 | partial2->p, | |
6f30b7e3 | 1410 | (chain2+n2-1) - partial2); |
4f579ae7 | 1411 | partial2--; |
8bad6fc8 ZL |
1412 | } |
1413 | } | |
5e86bdda | 1414 | |
1415 | cleanup: | |
1416 | while (p && p > chain) { | |
1417 | BUFFER_TRACE(p->bh, "call brelse"); | |
1418 | brelse(p->bh); | |
1419 | p--; | |
1420 | } | |
1421 | while (p2 && p2 > chain2) { | |
1422 | BUFFER_TRACE(p2->bh, "call brelse"); | |
1423 | brelse(p2->bh); | |
1424 | p2--; | |
1425 | } | |
6f30b7e3 | 1426 | return 0; |
8bad6fc8 | 1427 | |
4f579ae7 LC |
1428 | do_indirects: |
1429 | /* Kill the remaining (whole) subtrees */ | |
1430 | switch (offsets[0]) { | |
1431 | default: | |
1432 | if (++n >= n2) | |
5e86bdda | 1433 | break; |
4f579ae7 LC |
1434 | nr = i_data[EXT4_IND_BLOCK]; |
1435 | if (nr) { | |
1436 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); | |
1437 | i_data[EXT4_IND_BLOCK] = 0; | |
1438 | } | |
793bc518 | 1439 | /* fall through */ |
4f579ae7 LC |
1440 | case EXT4_IND_BLOCK: |
1441 | if (++n >= n2) | |
5e86bdda | 1442 | break; |
4f579ae7 LC |
1443 | nr = i_data[EXT4_DIND_BLOCK]; |
1444 | if (nr) { | |
1445 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); | |
1446 | i_data[EXT4_DIND_BLOCK] = 0; | |
1447 | } | |
793bc518 | 1448 | /* fall through */ |
4f579ae7 LC |
1449 | case EXT4_DIND_BLOCK: |
1450 | if (++n >= n2) | |
5e86bdda | 1451 | break; |
4f579ae7 LC |
1452 | nr = i_data[EXT4_TIND_BLOCK]; |
1453 | if (nr) { | |
1454 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); | |
1455 | i_data[EXT4_TIND_BLOCK] = 0; | |
1456 | } | |
793bc518 | 1457 | /* fall through */ |
4f579ae7 LC |
1458 | case EXT4_TIND_BLOCK: |
1459 | ; | |
1460 | } | |
5e86bdda | 1461 | goto cleanup; |
8bad6fc8 | 1462 | } |