Commit | Line | Data |
---|---|---|
dae1e52c AG |
1 | /* |
2 | * linux/fs/ext4/indirect.c | |
3 | * | |
4 | * from | |
5 | * | |
6 | * linux/fs/ext4/inode.c | |
7 | * | |
8 | * Copyright (C) 1992, 1993, 1994, 1995 | |
9 | * Remy Card (card@masi.ibp.fr) | |
10 | * Laboratoire MASI - Institut Blaise Pascal | |
11 | * Universite Pierre et Marie Curie (Paris VI) | |
12 | * | |
13 | * from | |
14 | * | |
15 | * linux/fs/minix/inode.c | |
16 | * | |
17 | * Copyright (C) 1991, 1992 Linus Torvalds | |
18 | * | |
19 | * Goal-directed block allocation by Stephen Tweedie | |
20 | * (sct@redhat.com), 1993, 1998 | |
21 | */ | |
22 | ||
dae1e52c AG |
23 | #include "ext4_jbd2.h" |
24 | #include "truncate.h" | |
4a092d73 | 25 | #include "ext4_extents.h" /* Needed for EXT_MAX_BLOCKS */ |
dae1e52c AG |
26 | |
27 | #include <trace/events/ext4.h> | |
28 | ||
29 | typedef struct { | |
30 | __le32 *p; | |
31 | __le32 key; | |
32 | struct buffer_head *bh; | |
33 | } Indirect; | |
34 | ||
35 | static 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 | ||
72 | static 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 | */ | |
142 | static Indirect *ext4_get_branch(struct inode *inode, int depth, | |
143 | ext4_lblk_t *offsets, | |
144 | Indirect chain[4], int *err) | |
145 | { | |
146 | struct super_block *sb = inode->i_sb; | |
147 | Indirect *p = chain; | |
148 | struct buffer_head *bh; | |
860d21e2 | 149 | int ret = -EIO; |
dae1e52c AG |
150 | |
151 | *err = 0; | |
152 | /* i_data is not going away, no lock needed */ | |
153 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); | |
154 | if (!p->key) | |
155 | goto no_block; | |
156 | while (--depth) { | |
157 | bh = sb_getblk(sb, le32_to_cpu(p->key)); | |
860d21e2 TT |
158 | if (unlikely(!bh)) { |
159 | ret = -ENOMEM; | |
dae1e52c | 160 | goto failure; |
860d21e2 | 161 | } |
dae1e52c AG |
162 | |
163 | if (!bh_uptodate_or_lock(bh)) { | |
164 | if (bh_submit_read(bh) < 0) { | |
165 | put_bh(bh); | |
166 | goto failure; | |
167 | } | |
168 | /* validate block references */ | |
169 | if (ext4_check_indirect_blockref(inode, bh)) { | |
170 | put_bh(bh); | |
171 | goto failure; | |
172 | } | |
173 | } | |
174 | ||
175 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); | |
176 | /* Reader: end */ | |
177 | if (!p->key) | |
178 | goto no_block; | |
179 | } | |
180 | return NULL; | |
181 | ||
182 | failure: | |
860d21e2 | 183 | *err = ret; |
dae1e52c AG |
184 | no_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 | */ | |
208 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) | |
209 | { | |
210 | struct ext4_inode_info *ei = EXT4_I(inode); | |
211 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; | |
212 | __le32 *p; | |
dae1e52c AG |
213 | |
214 | /* Try to find previous block */ | |
215 | for (p = ind->p - 1; p >= start; p--) { | |
216 | if (*p) | |
217 | return le32_to_cpu(*p); | |
218 | } | |
219 | ||
220 | /* No such thing, so let's try location of indirect block */ | |
221 | if (ind->bh) | |
222 | return ind->bh->b_blocknr; | |
223 | ||
224 | /* | |
225 | * It is going to be referred to from the inode itself? OK, just put it | |
226 | * into the same cylinder group then. | |
227 | */ | |
f86186b4 | 228 | return ext4_inode_to_goal_block(inode); |
dae1e52c AG |
229 | } |
230 | ||
231 | /** | |
232 | * ext4_find_goal - find a preferred place for allocation. | |
233 | * @inode: owner | |
234 | * @block: block we want | |
235 | * @partial: pointer to the last triple within a chain | |
236 | * | |
237 | * Normally this function find the preferred place for block allocation, | |
238 | * returns it. | |
239 | * Because this is only used for non-extent files, we limit the block nr | |
240 | * to 32 bits. | |
241 | */ | |
242 | static 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 | */ | |
268 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, | |
269 | int blocks_to_boundary) | |
270 | { | |
271 | unsigned int count = 0; | |
272 | ||
273 | /* | |
274 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
275 | * then it's clear blocks on that path have not allocated | |
276 | */ | |
277 | if (k > 0) { | |
278 | /* right now we don't handle cross boundary allocation */ | |
279 | if (blks < blocks_to_boundary + 1) | |
280 | count += blks; | |
281 | else | |
282 | count += blocks_to_boundary + 1; | |
283 | return count; | |
284 | } | |
285 | ||
286 | count++; | |
287 | while (count < blks && count <= blocks_to_boundary && | |
288 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
289 | count++; | |
290 | } | |
291 | return count; | |
292 | } | |
293 | ||
dae1e52c AG |
294 | /** |
295 | * ext4_alloc_branch - allocate and set up a chain of blocks. | |
296 | * @handle: handle for this transaction | |
297 | * @inode: owner | |
298 | * @indirect_blks: number of allocated indirect blocks | |
299 | * @blks: number of allocated direct blocks | |
300 | * @goal: preferred place for allocation | |
301 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
302 | * @branch: place to store the chain in. | |
303 | * | |
304 | * This function allocates blocks, zeroes out all but the last one, | |
305 | * links them into chain and (if we are synchronous) writes them to disk. | |
306 | * In other words, it prepares a branch that can be spliced onto the | |
307 | * inode. It stores the information about that chain in the branch[], in | |
308 | * the same format as ext4_get_branch() would do. We are calling it after | |
309 | * we had read the existing part of chain and partial points to the last | |
310 | * triple of that (one with zero ->key). Upon the exit we have the same | |
311 | * picture as after the successful ext4_get_block(), except that in one | |
312 | * place chain is disconnected - *branch->p is still zero (we did not | |
313 | * set the last link), but branch->key contains the number that should | |
314 | * be placed into *branch->p to fill that gap. | |
315 | * | |
316 | * If allocation fails we free all blocks we've allocated (and forget | |
317 | * their buffer_heads) and return the error value the from failed | |
318 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain | |
319 | * as described above and return 0. | |
320 | */ | |
321 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, | |
322 | ext4_lblk_t iblock, int indirect_blks, | |
323 | int *blks, ext4_fsblk_t goal, | |
324 | ext4_lblk_t *offsets, Indirect *branch) | |
325 | { | |
781f143e TT |
326 | struct ext4_allocation_request ar; |
327 | struct buffer_head * bh; | |
328 | ext4_fsblk_t b, new_blocks[4]; | |
329 | __le32 *p; | |
330 | int i, j, err, len = 1; | |
dae1e52c | 331 | |
dae1e52c | 332 | /* |
781f143e | 333 | * Set up for the direct block allocation |
dae1e52c | 334 | */ |
781f143e TT |
335 | memset(&ar, 0, sizeof(ar)); |
336 | ar.inode = inode; | |
337 | ar.len = *blks; | |
338 | ar.logical = iblock; | |
339 | if (S_ISREG(inode->i_mode)) | |
340 | ar.flags = EXT4_MB_HINT_DATA; | |
341 | ||
342 | for (i = 0; i <= indirect_blks; i++) { | |
343 | if (i == indirect_blks) { | |
344 | ar.goal = goal; | |
345 | new_blocks[i] = ext4_mb_new_blocks(handle, &ar, &err); | |
346 | } else | |
347 | goal = new_blocks[i] = ext4_new_meta_blocks(handle, inode, | |
348 | goal, 0, NULL, &err); | |
349 | if (err) { | |
350 | i--; | |
351 | goto failed; | |
352 | } | |
353 | branch[i].key = cpu_to_le32(new_blocks[i]); | |
354 | if (i == 0) | |
355 | continue; | |
356 | ||
357 | bh = branch[i].bh = sb_getblk(inode->i_sb, new_blocks[i-1]); | |
dae1e52c | 358 | if (unlikely(!bh)) { |
860d21e2 | 359 | err = -ENOMEM; |
dae1e52c AG |
360 | goto failed; |
361 | } | |
dae1e52c AG |
362 | lock_buffer(bh); |
363 | BUFFER_TRACE(bh, "call get_create_access"); | |
364 | err = ext4_journal_get_create_access(handle, bh); | |
365 | if (err) { | |
dae1e52c AG |
366 | unlock_buffer(bh); |
367 | goto failed; | |
368 | } | |
369 | ||
781f143e TT |
370 | memset(bh->b_data, 0, bh->b_size); |
371 | p = branch[i].p = (__le32 *) bh->b_data + offsets[i]; | |
372 | b = new_blocks[i]; | |
373 | ||
374 | if (i == indirect_blks) | |
375 | len = ar.len; | |
376 | for (j = 0; j < len; j++) | |
377 | *p++ = cpu_to_le32(b++); | |
378 | ||
dae1e52c AG |
379 | BUFFER_TRACE(bh, "marking uptodate"); |
380 | set_buffer_uptodate(bh); | |
381 | unlock_buffer(bh); | |
382 | ||
383 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
384 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
385 | if (err) | |
386 | goto failed; | |
387 | } | |
781f143e TT |
388 | *blks = ar.len; |
389 | return 0; | |
dae1e52c | 390 | failed: |
781f143e TT |
391 | for (; i >= 0; i--) { |
392 | if (i != indirect_blks && branch[i].bh) | |
393 | ext4_forget(handle, 1, inode, branch[i].bh, | |
394 | branch[i].bh->b_blocknr); | |
395 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], | |
396 | (i == indirect_blks) ? ar.len : 1, 0); | |
dae1e52c | 397 | } |
dae1e52c AG |
398 | return err; |
399 | } | |
400 | ||
401 | /** | |
402 | * ext4_splice_branch - splice the allocated branch onto inode. | |
403 | * @handle: handle for this transaction | |
404 | * @inode: owner | |
405 | * @block: (logical) number of block we are adding | |
406 | * @chain: chain of indirect blocks (with a missing link - see | |
407 | * ext4_alloc_branch) | |
408 | * @where: location of missing link | |
409 | * @num: number of indirect blocks we are adding | |
410 | * @blks: number of direct blocks we are adding | |
411 | * | |
412 | * This function fills the missing link and does all housekeeping needed in | |
413 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
414 | * chain to new block and return 0. | |
415 | */ | |
416 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, | |
417 | ext4_lblk_t block, Indirect *where, int num, | |
418 | int blks) | |
419 | { | |
420 | int i; | |
421 | int err = 0; | |
422 | ext4_fsblk_t current_block; | |
423 | ||
424 | /* | |
425 | * If we're splicing into a [td]indirect block (as opposed to the | |
426 | * inode) then we need to get write access to the [td]indirect block | |
427 | * before the splice. | |
428 | */ | |
429 | if (where->bh) { | |
430 | BUFFER_TRACE(where->bh, "get_write_access"); | |
431 | err = ext4_journal_get_write_access(handle, where->bh); | |
432 | if (err) | |
433 | goto err_out; | |
434 | } | |
435 | /* That's it */ | |
436 | ||
437 | *where->p = where->key; | |
438 | ||
439 | /* | |
440 | * Update the host buffer_head or inode to point to more just allocated | |
441 | * direct blocks blocks | |
442 | */ | |
443 | if (num == 0 && blks > 1) { | |
444 | current_block = le32_to_cpu(where->key) + 1; | |
445 | for (i = 1; i < blks; i++) | |
446 | *(where->p + i) = cpu_to_le32(current_block++); | |
447 | } | |
448 | ||
449 | /* We are done with atomic stuff, now do the rest of housekeeping */ | |
450 | /* had we spliced it onto indirect block? */ | |
451 | if (where->bh) { | |
452 | /* | |
453 | * If we spliced it onto an indirect block, we haven't | |
454 | * altered the inode. Note however that if it is being spliced | |
455 | * onto an indirect block at the very end of the file (the | |
456 | * file is growing) then we *will* alter the inode to reflect | |
457 | * the new i_size. But that is not done here - it is done in | |
458 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. | |
459 | */ | |
460 | jbd_debug(5, "splicing indirect only\n"); | |
461 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); | |
462 | err = ext4_handle_dirty_metadata(handle, inode, where->bh); | |
463 | if (err) | |
464 | goto err_out; | |
465 | } else { | |
466 | /* | |
467 | * OK, we spliced it into the inode itself on a direct block. | |
468 | */ | |
469 | ext4_mark_inode_dirty(handle, inode); | |
470 | jbd_debug(5, "splicing direct\n"); | |
471 | } | |
472 | return err; | |
473 | ||
474 | err_out: | |
475 | for (i = 1; i <= num; i++) { | |
476 | /* | |
477 | * branch[i].bh is newly allocated, so there is no | |
478 | * need to revoke the block, which is why we don't | |
479 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
480 | */ | |
481 | ext4_free_blocks(handle, inode, where[i].bh, 0, 1, | |
482 | EXT4_FREE_BLOCKS_FORGET); | |
483 | } | |
484 | ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key), | |
485 | blks, 0); | |
486 | ||
487 | return err; | |
488 | } | |
489 | ||
490 | /* | |
491 | * The ext4_ind_map_blocks() function handles non-extents inodes | |
492 | * (i.e., using the traditional indirect/double-indirect i_blocks | |
493 | * scheme) for ext4_map_blocks(). | |
494 | * | |
495 | * Allocation strategy is simple: if we have to allocate something, we will | |
496 | * have to go the whole way to leaf. So let's do it before attaching anything | |
497 | * to tree, set linkage between the newborn blocks, write them if sync is | |
498 | * required, recheck the path, free and repeat if check fails, otherwise | |
499 | * set the last missing link (that will protect us from any truncate-generated | |
500 | * removals - all blocks on the path are immune now) and possibly force the | |
501 | * write on the parent block. | |
502 | * That has a nice additional property: no special recovery from the failed | |
503 | * allocations is needed - we simply release blocks and do not touch anything | |
504 | * reachable from inode. | |
505 | * | |
506 | * `handle' can be NULL if create == 0. | |
507 | * | |
508 | * return > 0, # of blocks mapped or allocated. | |
509 | * return = 0, if plain lookup failed. | |
510 | * return < 0, error case. | |
511 | * | |
512 | * The ext4_ind_get_blocks() function should be called with | |
513 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem | |
514 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or | |
515 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system | |
516 | * blocks. | |
517 | */ | |
518 | int ext4_ind_map_blocks(handle_t *handle, struct inode *inode, | |
519 | struct ext4_map_blocks *map, | |
520 | int flags) | |
521 | { | |
522 | int err = -EIO; | |
523 | ext4_lblk_t offsets[4]; | |
524 | Indirect chain[4]; | |
525 | Indirect *partial; | |
526 | ext4_fsblk_t goal; | |
527 | int indirect_blks; | |
528 | int blocks_to_boundary = 0; | |
529 | int depth; | |
530 | int count = 0; | |
531 | ext4_fsblk_t first_block = 0; | |
532 | ||
533 | trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); | |
534 | J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); | |
535 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); | |
536 | depth = ext4_block_to_path(inode, map->m_lblk, offsets, | |
537 | &blocks_to_boundary); | |
538 | ||
539 | if (depth == 0) | |
540 | goto out; | |
541 | ||
542 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); | |
543 | ||
544 | /* Simplest case - block found, no allocation needed */ | |
545 | if (!partial) { | |
546 | first_block = le32_to_cpu(chain[depth - 1].key); | |
547 | count++; | |
548 | /*map more blocks*/ | |
549 | while (count < map->m_len && count <= blocks_to_boundary) { | |
550 | ext4_fsblk_t blk; | |
551 | ||
552 | blk = le32_to_cpu(*(chain[depth-1].p + count)); | |
553 | ||
554 | if (blk == first_block + count) | |
555 | count++; | |
556 | else | |
557 | break; | |
558 | } | |
559 | goto got_it; | |
560 | } | |
561 | ||
562 | /* Next simple case - plain lookup or failed read of indirect block */ | |
563 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO) | |
564 | goto cleanup; | |
565 | ||
566 | /* | |
567 | * Okay, we need to do block allocation. | |
568 | */ | |
bab08ab9 TT |
569 | if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, |
570 | EXT4_FEATURE_RO_COMPAT_BIGALLOC)) { | |
571 | EXT4_ERROR_INODE(inode, "Can't allocate blocks for " | |
572 | "non-extent mapped inodes with bigalloc"); | |
573 | return -ENOSPC; | |
574 | } | |
575 | ||
dae1e52c AG |
576 | goal = ext4_find_goal(inode, map->m_lblk, partial); |
577 | ||
578 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
579 | indirect_blks = (chain + depth) - partial - 1; | |
580 | ||
581 | /* | |
582 | * Next look up the indirect map to count the totoal number of | |
583 | * direct blocks to allocate for this branch. | |
584 | */ | |
585 | count = ext4_blks_to_allocate(partial, indirect_blks, | |
586 | map->m_len, blocks_to_boundary); | |
587 | /* | |
588 | * Block out ext4_truncate while we alter the tree | |
589 | */ | |
590 | err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks, | |
591 | &count, goal, | |
592 | offsets + (partial - chain), partial); | |
593 | ||
594 | /* | |
595 | * The ext4_splice_branch call will free and forget any buffers | |
596 | * on the new chain if there is a failure, but that risks using | |
597 | * up transaction credits, especially for bitmaps where the | |
598 | * credits cannot be returned. Can we handle this somehow? We | |
599 | * may need to return -EAGAIN upwards in the worst case. --sct | |
600 | */ | |
601 | if (!err) | |
602 | err = ext4_splice_branch(handle, inode, map->m_lblk, | |
603 | partial, indirect_blks, count); | |
604 | if (err) | |
605 | goto cleanup; | |
606 | ||
607 | map->m_flags |= EXT4_MAP_NEW; | |
608 | ||
609 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
610 | got_it: | |
611 | map->m_flags |= EXT4_MAP_MAPPED; | |
612 | map->m_pblk = le32_to_cpu(chain[depth-1].key); | |
613 | map->m_len = count; | |
614 | if (count > blocks_to_boundary) | |
615 | map->m_flags |= EXT4_MAP_BOUNDARY; | |
616 | err = count; | |
617 | /* Clean up and exit */ | |
618 | partial = chain + depth - 1; /* the whole chain */ | |
619 | cleanup: | |
620 | while (partial > chain) { | |
621 | BUFFER_TRACE(partial->bh, "call brelse"); | |
622 | brelse(partial->bh); | |
623 | partial--; | |
624 | } | |
625 | out: | |
19b303d8 | 626 | trace_ext4_ind_map_blocks_exit(inode, map, err); |
dae1e52c AG |
627 | return err; |
628 | } | |
629 | ||
630 | /* | |
631 | * O_DIRECT for ext3 (or indirect map) based files | |
632 | * | |
633 | * If the O_DIRECT write will extend the file then add this inode to the | |
634 | * orphan list. So recovery will truncate it back to the original size | |
635 | * if the machine crashes during the write. | |
636 | * | |
637 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
638 | * crashes then stale disk data _may_ be exposed inside the file. But current | |
639 | * VFS code falls back into buffered path in that case so we are safe. | |
640 | */ | |
641 | ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb, | |
642 | const struct iovec *iov, loff_t offset, | |
643 | unsigned long nr_segs) | |
644 | { | |
645 | struct file *file = iocb->ki_filp; | |
646 | struct inode *inode = file->f_mapping->host; | |
647 | struct ext4_inode_info *ei = EXT4_I(inode); | |
648 | handle_t *handle; | |
649 | ssize_t ret; | |
650 | int orphan = 0; | |
651 | size_t count = iov_length(iov, nr_segs); | |
652 | int retries = 0; | |
653 | ||
654 | if (rw == WRITE) { | |
655 | loff_t final_size = offset + count; | |
656 | ||
657 | if (final_size > inode->i_size) { | |
658 | /* Credits for sb + inode write */ | |
9924a92a | 659 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
dae1e52c AG |
660 | if (IS_ERR(handle)) { |
661 | ret = PTR_ERR(handle); | |
662 | goto out; | |
663 | } | |
664 | ret = ext4_orphan_add(handle, inode); | |
665 | if (ret) { | |
666 | ext4_journal_stop(handle); | |
667 | goto out; | |
668 | } | |
669 | orphan = 1; | |
670 | ei->i_disksize = inode->i_size; | |
671 | ext4_journal_stop(handle); | |
672 | } | |
673 | } | |
674 | ||
675 | retry: | |
dccaf33f | 676 | if (rw == READ && ext4_should_dioread_nolock(inode)) { |
c278531d DM |
677 | if (unlikely(atomic_read(&EXT4_I(inode)->i_unwritten))) { |
678 | mutex_lock(&inode->i_mutex); | |
679 | ext4_flush_unwritten_io(inode); | |
680 | mutex_unlock(&inode->i_mutex); | |
681 | } | |
17335dcc DM |
682 | /* |
683 | * Nolock dioread optimization may be dynamically disabled | |
684 | * via ext4_inode_block_unlocked_dio(). Check inode's state | |
685 | * while holding extra i_dio_count ref. | |
686 | */ | |
687 | atomic_inc(&inode->i_dio_count); | |
688 | smp_mb(); | |
689 | if (unlikely(ext4_test_inode_state(inode, | |
690 | EXT4_STATE_DIOREAD_LOCK))) { | |
691 | inode_dio_done(inode); | |
692 | goto locked; | |
693 | } | |
dae1e52c AG |
694 | ret = __blockdev_direct_IO(rw, iocb, inode, |
695 | inode->i_sb->s_bdev, iov, | |
696 | offset, nr_segs, | |
697 | ext4_get_block, NULL, NULL, 0); | |
17335dcc | 698 | inode_dio_done(inode); |
dccaf33f | 699 | } else { |
17335dcc | 700 | locked: |
60ad4466 LT |
701 | ret = blockdev_direct_IO(rw, iocb, inode, iov, |
702 | offset, nr_segs, ext4_get_block); | |
dae1e52c AG |
703 | |
704 | if (unlikely((rw & WRITE) && ret < 0)) { | |
705 | loff_t isize = i_size_read(inode); | |
706 | loff_t end = offset + iov_length(iov, nr_segs); | |
707 | ||
708 | if (end > isize) | |
709 | ext4_truncate_failed_write(inode); | |
710 | } | |
711 | } | |
712 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
713 | goto retry; | |
714 | ||
715 | if (orphan) { | |
716 | int err; | |
717 | ||
718 | /* Credits for sb + inode write */ | |
9924a92a | 719 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
dae1e52c AG |
720 | if (IS_ERR(handle)) { |
721 | /* This is really bad luck. We've written the data | |
722 | * but cannot extend i_size. Bail out and pretend | |
723 | * the write failed... */ | |
724 | ret = PTR_ERR(handle); | |
725 | if (inode->i_nlink) | |
726 | ext4_orphan_del(NULL, inode); | |
727 | ||
728 | goto out; | |
729 | } | |
730 | if (inode->i_nlink) | |
731 | ext4_orphan_del(handle, inode); | |
732 | if (ret > 0) { | |
733 | loff_t end = offset + ret; | |
734 | if (end > inode->i_size) { | |
735 | ei->i_disksize = end; | |
736 | i_size_write(inode, end); | |
737 | /* | |
738 | * We're going to return a positive `ret' | |
739 | * here due to non-zero-length I/O, so there's | |
740 | * no way of reporting error returns from | |
741 | * ext4_mark_inode_dirty() to userspace. So | |
742 | * ignore it. | |
743 | */ | |
744 | ext4_mark_inode_dirty(handle, inode); | |
745 | } | |
746 | } | |
747 | err = ext4_journal_stop(handle); | |
748 | if (ret == 0) | |
749 | ret = err; | |
750 | } | |
751 | out: | |
752 | return ret; | |
753 | } | |
754 | ||
755 | /* | |
756 | * Calculate the number of metadata blocks need to reserve | |
757 | * to allocate a new block at @lblocks for non extent file based file | |
758 | */ | |
759 | int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock) | |
760 | { | |
761 | struct ext4_inode_info *ei = EXT4_I(inode); | |
762 | sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1); | |
763 | int blk_bits; | |
764 | ||
765 | if (lblock < EXT4_NDIR_BLOCKS) | |
766 | return 0; | |
767 | ||
768 | lblock -= EXT4_NDIR_BLOCKS; | |
769 | ||
770 | if (ei->i_da_metadata_calc_len && | |
771 | (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) { | |
772 | ei->i_da_metadata_calc_len++; | |
773 | return 0; | |
774 | } | |
775 | ei->i_da_metadata_calc_last_lblock = lblock & dind_mask; | |
776 | ei->i_da_metadata_calc_len = 1; | |
777 | blk_bits = order_base_2(lblock); | |
778 | return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1; | |
779 | } | |
780 | ||
781 | int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
782 | { | |
783 | int indirects; | |
784 | ||
785 | /* if nrblocks are contiguous */ | |
786 | if (chunk) { | |
787 | /* | |
788 | * With N contiguous data blocks, we need at most | |
789 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks, | |
790 | * 2 dindirect blocks, and 1 tindirect block | |
791 | */ | |
792 | return DIV_ROUND_UP(nrblocks, | |
793 | EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4; | |
794 | } | |
795 | /* | |
796 | * if nrblocks are not contiguous, worse case, each block touch | |
797 | * a indirect block, and each indirect block touch a double indirect | |
798 | * block, plus a triple indirect block | |
799 | */ | |
800 | indirects = nrblocks * 2 + 1; | |
801 | return indirects; | |
802 | } | |
803 | ||
804 | /* | |
805 | * Truncate transactions can be complex and absolutely huge. So we need to | |
806 | * be able to restart the transaction at a conventient checkpoint to make | |
807 | * sure we don't overflow the journal. | |
808 | * | |
809 | * start_transaction gets us a new handle for a truncate transaction, | |
810 | * and extend_transaction tries to extend the existing one a bit. If | |
811 | * extend fails, we need to propagate the failure up and restart the | |
812 | * transaction in the top-level truncate loop. --sct | |
813 | */ | |
814 | static handle_t *start_transaction(struct inode *inode) | |
815 | { | |
816 | handle_t *result; | |
817 | ||
9924a92a TT |
818 | result = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
819 | ext4_blocks_for_truncate(inode)); | |
dae1e52c AG |
820 | if (!IS_ERR(result)) |
821 | return result; | |
822 | ||
823 | ext4_std_error(inode->i_sb, PTR_ERR(result)); | |
824 | return result; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Try to extend this transaction for the purposes of truncation. | |
829 | * | |
830 | * Returns 0 if we managed to create more room. If we can't create more | |
831 | * room, and the transaction must be restarted we return 1. | |
832 | */ | |
833 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
834 | { | |
835 | if (!ext4_handle_valid(handle)) | |
836 | return 0; | |
837 | if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) | |
838 | return 0; | |
839 | if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode))) | |
840 | return 0; | |
841 | return 1; | |
842 | } | |
843 | ||
844 | /* | |
845 | * Probably it should be a library function... search for first non-zero word | |
846 | * or memcmp with zero_page, whatever is better for particular architecture. | |
847 | * Linus? | |
848 | */ | |
849 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
850 | { | |
851 | while (p < q) | |
852 | if (*p++) | |
853 | return 0; | |
854 | return 1; | |
855 | } | |
856 | ||
857 | /** | |
858 | * ext4_find_shared - find the indirect blocks for partial truncation. | |
859 | * @inode: inode in question | |
860 | * @depth: depth of the affected branch | |
861 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) | |
862 | * @chain: place to store the pointers to partial indirect blocks | |
863 | * @top: place to the (detached) top of branch | |
864 | * | |
865 | * This is a helper function used by ext4_truncate(). | |
866 | * | |
867 | * When we do truncate() we may have to clean the ends of several | |
868 | * indirect blocks but leave the blocks themselves alive. Block is | |
869 | * partially truncated if some data below the new i_size is referred | |
870 | * from it (and it is on the path to the first completely truncated | |
871 | * data block, indeed). We have to free the top of that path along | |
872 | * with everything to the right of the path. Since no allocation | |
873 | * past the truncation point is possible until ext4_truncate() | |
874 | * finishes, we may safely do the latter, but top of branch may | |
875 | * require special attention - pageout below the truncation point | |
876 | * might try to populate it. | |
877 | * | |
878 | * We atomically detach the top of branch from the tree, store the | |
879 | * block number of its root in *@top, pointers to buffer_heads of | |
880 | * partially truncated blocks - in @chain[].bh and pointers to | |
881 | * their last elements that should not be removed - in | |
882 | * @chain[].p. Return value is the pointer to last filled element | |
883 | * of @chain. | |
884 | * | |
885 | * The work left to caller to do the actual freeing of subtrees: | |
886 | * a) free the subtree starting from *@top | |
887 | * b) free the subtrees whose roots are stored in | |
888 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
889 | * c) free the subtrees growing from the inode past the @chain[0]. | |
890 | * (no partially truncated stuff there). */ | |
891 | ||
892 | static Indirect *ext4_find_shared(struct inode *inode, int depth, | |
893 | ext4_lblk_t offsets[4], Indirect chain[4], | |
894 | __le32 *top) | |
895 | { | |
896 | Indirect *partial, *p; | |
897 | int k, err; | |
898 | ||
899 | *top = 0; | |
900 | /* Make k index the deepest non-null offset + 1 */ | |
901 | for (k = depth; k > 1 && !offsets[k-1]; k--) | |
902 | ; | |
903 | partial = ext4_get_branch(inode, k, offsets, chain, &err); | |
904 | /* Writer: pointers */ | |
905 | if (!partial) | |
906 | partial = chain + k-1; | |
907 | /* | |
908 | * If the branch acquired continuation since we've looked at it - | |
909 | * fine, it should all survive and (new) top doesn't belong to us. | |
910 | */ | |
911 | if (!partial->key && *partial->p) | |
912 | /* Writer: end */ | |
913 | goto no_top; | |
914 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) | |
915 | ; | |
916 | /* | |
917 | * OK, we've found the last block that must survive. The rest of our | |
918 | * branch should be detached before unlocking. However, if that rest | |
919 | * of branch is all ours and does not grow immediately from the inode | |
920 | * it's easier to cheat and just decrement partial->p. | |
921 | */ | |
922 | if (p == chain + k - 1 && p > chain) { | |
923 | p->p--; | |
924 | } else { | |
925 | *top = *p->p; | |
926 | /* Nope, don't do this in ext4. Must leave the tree intact */ | |
927 | #if 0 | |
928 | *p->p = 0; | |
929 | #endif | |
930 | } | |
931 | /* Writer: end */ | |
932 | ||
933 | while (partial > p) { | |
934 | brelse(partial->bh); | |
935 | partial--; | |
936 | } | |
937 | no_top: | |
938 | return partial; | |
939 | } | |
940 | ||
941 | /* | |
942 | * Zero a number of block pointers in either an inode or an indirect block. | |
943 | * If we restart the transaction we must again get write access to the | |
944 | * indirect block for further modification. | |
945 | * | |
946 | * We release `count' blocks on disk, but (last - first) may be greater | |
947 | * than `count' because there can be holes in there. | |
948 | * | |
949 | * Return 0 on success, 1 on invalid block range | |
950 | * and < 0 on fatal error. | |
951 | */ | |
952 | static int ext4_clear_blocks(handle_t *handle, struct inode *inode, | |
953 | struct buffer_head *bh, | |
954 | ext4_fsblk_t block_to_free, | |
955 | unsigned long count, __le32 *first, | |
956 | __le32 *last) | |
957 | { | |
958 | __le32 *p; | |
959 | int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED; | |
960 | int err; | |
961 | ||
962 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
963 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
964 | ||
965 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free, | |
966 | count)) { | |
967 | EXT4_ERROR_INODE(inode, "attempt to clear invalid " | |
968 | "blocks %llu len %lu", | |
969 | (unsigned long long) block_to_free, count); | |
970 | return 1; | |
971 | } | |
972 | ||
973 | if (try_to_extend_transaction(handle, inode)) { | |
974 | if (bh) { | |
975 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
976 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
977 | if (unlikely(err)) | |
978 | goto out_err; | |
979 | } | |
980 | err = ext4_mark_inode_dirty(handle, inode); | |
981 | if (unlikely(err)) | |
982 | goto out_err; | |
983 | err = ext4_truncate_restart_trans(handle, inode, | |
984 | ext4_blocks_for_truncate(inode)); | |
985 | if (unlikely(err)) | |
986 | goto out_err; | |
987 | if (bh) { | |
988 | BUFFER_TRACE(bh, "retaking write access"); | |
989 | err = ext4_journal_get_write_access(handle, bh); | |
990 | if (unlikely(err)) | |
991 | goto out_err; | |
992 | } | |
993 | } | |
994 | ||
995 | for (p = first; p < last; p++) | |
996 | *p = 0; | |
997 | ||
998 | ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags); | |
999 | return 0; | |
1000 | out_err: | |
1001 | ext4_std_error(inode->i_sb, err); | |
1002 | return err; | |
1003 | } | |
1004 | ||
1005 | /** | |
1006 | * ext4_free_data - free a list of data blocks | |
1007 | * @handle: handle for this transaction | |
1008 | * @inode: inode we are dealing with | |
1009 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
1010 | * @first: array of block numbers | |
1011 | * @last: points immediately past the end of array | |
1012 | * | |
1013 | * We are freeing all blocks referred from that array (numbers are stored as | |
1014 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
1015 | * | |
1016 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
1017 | * blocks are contiguous then releasing them at one time will only affect one | |
1018 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
1019 | * actually use a lot of journal space. | |
1020 | * | |
1021 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
1022 | * block pointers. | |
1023 | */ | |
1024 | static void ext4_free_data(handle_t *handle, struct inode *inode, | |
1025 | struct buffer_head *this_bh, | |
1026 | __le32 *first, __le32 *last) | |
1027 | { | |
1028 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ | |
1029 | unsigned long count = 0; /* Number of blocks in the run */ | |
1030 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
1031 | corresponding to | |
1032 | block_to_free */ | |
1033 | ext4_fsblk_t nr; /* Current block # */ | |
1034 | __le32 *p; /* Pointer into inode/ind | |
1035 | for current block */ | |
1036 | int err = 0; | |
1037 | ||
1038 | if (this_bh) { /* For indirect block */ | |
1039 | BUFFER_TRACE(this_bh, "get_write_access"); | |
1040 | err = ext4_journal_get_write_access(handle, this_bh); | |
1041 | /* Important: if we can't update the indirect pointers | |
1042 | * to the blocks, we can't free them. */ | |
1043 | if (err) | |
1044 | return; | |
1045 | } | |
1046 | ||
1047 | for (p = first; p < last; p++) { | |
1048 | nr = le32_to_cpu(*p); | |
1049 | if (nr) { | |
1050 | /* accumulate blocks to free if they're contiguous */ | |
1051 | if (count == 0) { | |
1052 | block_to_free = nr; | |
1053 | block_to_free_p = p; | |
1054 | count = 1; | |
1055 | } else if (nr == block_to_free + count) { | |
1056 | count++; | |
1057 | } else { | |
1058 | err = ext4_clear_blocks(handle, inode, this_bh, | |
1059 | block_to_free, count, | |
1060 | block_to_free_p, p); | |
1061 | if (err) | |
1062 | break; | |
1063 | block_to_free = nr; | |
1064 | block_to_free_p = p; | |
1065 | count = 1; | |
1066 | } | |
1067 | } | |
1068 | } | |
1069 | ||
1070 | if (!err && count > 0) | |
1071 | err = ext4_clear_blocks(handle, inode, this_bh, block_to_free, | |
1072 | count, block_to_free_p, p); | |
1073 | if (err < 0) | |
1074 | /* fatal error */ | |
1075 | return; | |
1076 | ||
1077 | if (this_bh) { | |
1078 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); | |
1079 | ||
1080 | /* | |
1081 | * The buffer head should have an attached journal head at this | |
1082 | * point. However, if the data is corrupted and an indirect | |
1083 | * block pointed to itself, it would have been detached when | |
1084 | * the block was cleared. Check for this instead of OOPSing. | |
1085 | */ | |
1086 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) | |
1087 | ext4_handle_dirty_metadata(handle, inode, this_bh); | |
1088 | else | |
1089 | EXT4_ERROR_INODE(inode, | |
1090 | "circular indirect block detected at " | |
1091 | "block %llu", | |
1092 | (unsigned long long) this_bh->b_blocknr); | |
1093 | } | |
1094 | } | |
1095 | ||
1096 | /** | |
1097 | * ext4_free_branches - free an array of branches | |
1098 | * @handle: JBD handle for this transaction | |
1099 | * @inode: inode we are dealing with | |
1100 | * @parent_bh: the buffer_head which contains *@first and *@last | |
1101 | * @first: array of block numbers | |
1102 | * @last: pointer immediately past the end of array | |
1103 | * @depth: depth of the branches to free | |
1104 | * | |
1105 | * We are freeing all blocks referred from these branches (numbers are | |
1106 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
1107 | * appropriately. | |
1108 | */ | |
1109 | static void ext4_free_branches(handle_t *handle, struct inode *inode, | |
1110 | struct buffer_head *parent_bh, | |
1111 | __le32 *first, __le32 *last, int depth) | |
1112 | { | |
1113 | ext4_fsblk_t nr; | |
1114 | __le32 *p; | |
1115 | ||
1116 | if (ext4_handle_is_aborted(handle)) | |
1117 | return; | |
1118 | ||
1119 | if (depth--) { | |
1120 | struct buffer_head *bh; | |
1121 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1122 | p = last; | |
1123 | while (--p >= first) { | |
1124 | nr = le32_to_cpu(*p); | |
1125 | if (!nr) | |
1126 | continue; /* A hole */ | |
1127 | ||
1128 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), | |
1129 | nr, 1)) { | |
1130 | EXT4_ERROR_INODE(inode, | |
1131 | "invalid indirect mapped " | |
1132 | "block %lu (level %d)", | |
1133 | (unsigned long) nr, depth); | |
1134 | break; | |
1135 | } | |
1136 | ||
1137 | /* Go read the buffer for the next level down */ | |
1138 | bh = sb_bread(inode->i_sb, nr); | |
1139 | ||
1140 | /* | |
1141 | * A read failure? Report error and clear slot | |
1142 | * (should be rare). | |
1143 | */ | |
1144 | if (!bh) { | |
1145 | EXT4_ERROR_INODE_BLOCK(inode, nr, | |
1146 | "Read failure"); | |
1147 | continue; | |
1148 | } | |
1149 | ||
1150 | /* This zaps the entire block. Bottom up. */ | |
1151 | BUFFER_TRACE(bh, "free child branches"); | |
1152 | ext4_free_branches(handle, inode, bh, | |
1153 | (__le32 *) bh->b_data, | |
1154 | (__le32 *) bh->b_data + addr_per_block, | |
1155 | depth); | |
1156 | brelse(bh); | |
1157 | ||
1158 | /* | |
1159 | * Everything below this this pointer has been | |
1160 | * released. Now let this top-of-subtree go. | |
1161 | * | |
1162 | * We want the freeing of this indirect block to be | |
1163 | * atomic in the journal with the updating of the | |
1164 | * bitmap block which owns it. So make some room in | |
1165 | * the journal. | |
1166 | * | |
1167 | * We zero the parent pointer *after* freeing its | |
1168 | * pointee in the bitmaps, so if extend_transaction() | |
1169 | * for some reason fails to put the bitmap changes and | |
1170 | * the release into the same transaction, recovery | |
1171 | * will merely complain about releasing a free block, | |
1172 | * rather than leaking blocks. | |
1173 | */ | |
1174 | if (ext4_handle_is_aborted(handle)) | |
1175 | return; | |
1176 | if (try_to_extend_transaction(handle, inode)) { | |
1177 | ext4_mark_inode_dirty(handle, inode); | |
1178 | ext4_truncate_restart_trans(handle, inode, | |
1179 | ext4_blocks_for_truncate(inode)); | |
1180 | } | |
1181 | ||
1182 | /* | |
1183 | * The forget flag here is critical because if | |
1184 | * we are journaling (and not doing data | |
1185 | * journaling), we have to make sure a revoke | |
1186 | * record is written to prevent the journal | |
1187 | * replay from overwriting the (former) | |
1188 | * indirect block if it gets reallocated as a | |
1189 | * data block. This must happen in the same | |
1190 | * transaction where the data blocks are | |
1191 | * actually freed. | |
1192 | */ | |
1193 | ext4_free_blocks(handle, inode, NULL, nr, 1, | |
1194 | EXT4_FREE_BLOCKS_METADATA| | |
1195 | EXT4_FREE_BLOCKS_FORGET); | |
1196 | ||
1197 | if (parent_bh) { | |
1198 | /* | |
1199 | * The block which we have just freed is | |
1200 | * pointed to by an indirect block: journal it | |
1201 | */ | |
1202 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
1203 | if (!ext4_journal_get_write_access(handle, | |
1204 | parent_bh)){ | |
1205 | *p = 0; | |
1206 | BUFFER_TRACE(parent_bh, | |
1207 | "call ext4_handle_dirty_metadata"); | |
1208 | ext4_handle_dirty_metadata(handle, | |
1209 | inode, | |
1210 | parent_bh); | |
1211 | } | |
1212 | } | |
1213 | } | |
1214 | } else { | |
1215 | /* We have reached the bottom of the tree. */ | |
1216 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
1217 | ext4_free_data(handle, inode, parent_bh, first, last); | |
1218 | } | |
1219 | } | |
1220 | ||
1221 | void ext4_ind_truncate(struct inode *inode) | |
1222 | { | |
1223 | handle_t *handle; | |
1224 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1225 | __le32 *i_data = ei->i_data; | |
1226 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1227 | struct address_space *mapping = inode->i_mapping; | |
1228 | ext4_lblk_t offsets[4]; | |
1229 | Indirect chain[4]; | |
1230 | Indirect *partial; | |
1231 | __le32 nr = 0; | |
1232 | int n = 0; | |
1233 | ext4_lblk_t last_block, max_block; | |
189e868f | 1234 | loff_t page_len; |
dae1e52c | 1235 | unsigned blocksize = inode->i_sb->s_blocksize; |
189e868f | 1236 | int err; |
dae1e52c AG |
1237 | |
1238 | handle = start_transaction(inode); | |
1239 | if (IS_ERR(handle)) | |
1240 | return; /* AKPM: return what? */ | |
1241 | ||
1242 | last_block = (inode->i_size + blocksize-1) | |
1243 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); | |
1244 | max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1) | |
1245 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); | |
1246 | ||
189e868f AH |
1247 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
1248 | page_len = PAGE_CACHE_SIZE - | |
1249 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
1250 | ||
1251 | err = ext4_discard_partial_page_buffers(handle, | |
1252 | mapping, inode->i_size, page_len, 0); | |
1253 | ||
1254 | if (err) | |
dae1e52c | 1255 | goto out_stop; |
189e868f | 1256 | } |
dae1e52c AG |
1257 | |
1258 | if (last_block != max_block) { | |
1259 | n = ext4_block_to_path(inode, last_block, offsets, NULL); | |
1260 | if (n == 0) | |
1261 | goto out_stop; /* error */ | |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * OK. This truncate is going to happen. We add the inode to the | |
1266 | * orphan list, so that if this truncate spans multiple transactions, | |
1267 | * and we crash, we will resume the truncate when the filesystem | |
1268 | * recovers. It also marks the inode dirty, to catch the new size. | |
1269 | * | |
1270 | * Implication: the file must always be in a sane, consistent | |
1271 | * truncatable state while each transaction commits. | |
1272 | */ | |
1273 | if (ext4_orphan_add(handle, inode)) | |
1274 | goto out_stop; | |
1275 | ||
1276 | /* | |
1277 | * From here we block out all ext4_get_block() callers who want to | |
1278 | * modify the block allocation tree. | |
1279 | */ | |
1280 | down_write(&ei->i_data_sem); | |
1281 | ||
1282 | ext4_discard_preallocations(inode); | |
51865fda | 1283 | ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block); |
dae1e52c AG |
1284 | |
1285 | /* | |
1286 | * The orphan list entry will now protect us from any crash which | |
1287 | * occurs before the truncate completes, so it is now safe to propagate | |
1288 | * the new, shorter inode size (held for now in i_size) into the | |
1289 | * on-disk inode. We do this via i_disksize, which is the value which | |
1290 | * ext4 *really* writes onto the disk inode. | |
1291 | */ | |
1292 | ei->i_disksize = inode->i_size; | |
1293 | ||
1294 | if (last_block == max_block) { | |
1295 | /* | |
1296 | * It is unnecessary to free any data blocks if last_block is | |
1297 | * equal to the indirect block limit. | |
1298 | */ | |
1299 | goto out_unlock; | |
1300 | } else if (n == 1) { /* direct blocks */ | |
1301 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], | |
1302 | i_data + EXT4_NDIR_BLOCKS); | |
1303 | goto do_indirects; | |
1304 | } | |
1305 | ||
1306 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); | |
1307 | /* Kill the top of shared branch (not detached) */ | |
1308 | if (nr) { | |
1309 | if (partial == chain) { | |
1310 | /* Shared branch grows from the inode */ | |
1311 | ext4_free_branches(handle, inode, NULL, | |
1312 | &nr, &nr+1, (chain+n-1) - partial); | |
1313 | *partial->p = 0; | |
1314 | /* | |
1315 | * We mark the inode dirty prior to restart, | |
1316 | * and prior to stop. No need for it here. | |
1317 | */ | |
1318 | } else { | |
1319 | /* Shared branch grows from an indirect block */ | |
1320 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
1321 | ext4_free_branches(handle, inode, partial->bh, | |
1322 | partial->p, | |
1323 | partial->p+1, (chain+n-1) - partial); | |
1324 | } | |
1325 | } | |
1326 | /* Clear the ends of indirect blocks on the shared branch */ | |
1327 | while (partial > chain) { | |
1328 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, | |
1329 | (__le32*)partial->bh->b_data+addr_per_block, | |
1330 | (chain+n-1) - partial); | |
1331 | BUFFER_TRACE(partial->bh, "call brelse"); | |
1332 | brelse(partial->bh); | |
1333 | partial--; | |
1334 | } | |
1335 | do_indirects: | |
1336 | /* Kill the remaining (whole) subtrees */ | |
1337 | switch (offsets[0]) { | |
1338 | default: | |
1339 | nr = i_data[EXT4_IND_BLOCK]; | |
1340 | if (nr) { | |
1341 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); | |
1342 | i_data[EXT4_IND_BLOCK] = 0; | |
1343 | } | |
1344 | case EXT4_IND_BLOCK: | |
1345 | nr = i_data[EXT4_DIND_BLOCK]; | |
1346 | if (nr) { | |
1347 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); | |
1348 | i_data[EXT4_DIND_BLOCK] = 0; | |
1349 | } | |
1350 | case EXT4_DIND_BLOCK: | |
1351 | nr = i_data[EXT4_TIND_BLOCK]; | |
1352 | if (nr) { | |
1353 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); | |
1354 | i_data[EXT4_TIND_BLOCK] = 0; | |
1355 | } | |
1356 | case EXT4_TIND_BLOCK: | |
1357 | ; | |
1358 | } | |
1359 | ||
1360 | out_unlock: | |
1361 | up_write(&ei->i_data_sem); | |
1362 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
1363 | ext4_mark_inode_dirty(handle, inode); | |
1364 | ||
1365 | /* | |
1366 | * In a multi-transaction truncate, we only make the final transaction | |
1367 | * synchronous | |
1368 | */ | |
1369 | if (IS_SYNC(inode)) | |
1370 | ext4_handle_sync(handle); | |
1371 | out_stop: | |
1372 | /* | |
1373 | * If this was a simple ftruncate(), and the file will remain alive | |
1374 | * then we need to clear up the orphan record which we created above. | |
1375 | * However, if this was a real unlink then we were called by | |
1376 | * ext4_delete_inode(), and we allow that function to clean up the | |
1377 | * orphan info for us. | |
1378 | */ | |
1379 | if (inode->i_nlink) | |
1380 | ext4_orphan_del(handle, inode); | |
1381 | ||
1382 | ext4_journal_stop(handle); | |
1383 | trace_ext4_truncate_exit(inode); | |
1384 | } | |
1385 | ||
8bad6fc8 ZL |
1386 | static int free_hole_blocks(handle_t *handle, struct inode *inode, |
1387 | struct buffer_head *parent_bh, __le32 *i_data, | |
1388 | int level, ext4_lblk_t first, | |
1389 | ext4_lblk_t count, int max) | |
1390 | { | |
1391 | struct buffer_head *bh = NULL; | |
1392 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1393 | int ret = 0; | |
1394 | int i, inc; | |
1395 | ext4_lblk_t offset; | |
1396 | __le32 blk; | |
1397 | ||
1398 | inc = 1 << ((EXT4_BLOCK_SIZE_BITS(inode->i_sb) - 2) * level); | |
1399 | for (i = 0, offset = 0; i < max; i++, i_data++, offset += inc) { | |
1400 | if (offset >= count + first) | |
1401 | break; | |
1402 | if (*i_data == 0 || (offset + inc) <= first) | |
1403 | continue; | |
1404 | blk = *i_data; | |
1405 | if (level > 0) { | |
1406 | ext4_lblk_t first2; | |
8cde7ad1 | 1407 | bh = sb_bread(inode->i_sb, le32_to_cpu(blk)); |
8bad6fc8 | 1408 | if (!bh) { |
8cde7ad1 | 1409 | EXT4_ERROR_INODE_BLOCK(inode, le32_to_cpu(blk), |
8bad6fc8 ZL |
1410 | "Read failure"); |
1411 | return -EIO; | |
1412 | } | |
1413 | first2 = (first > offset) ? first - offset : 0; | |
1414 | ret = free_hole_blocks(handle, inode, bh, | |
1415 | (__le32 *)bh->b_data, level - 1, | |
1416 | first2, count - offset, | |
1417 | inode->i_sb->s_blocksize >> 2); | |
1418 | if (ret) { | |
1419 | brelse(bh); | |
1420 | goto err; | |
1421 | } | |
1422 | } | |
1423 | if (level == 0 || | |
1424 | (bh && all_zeroes((__le32 *)bh->b_data, | |
1425 | (__le32 *)bh->b_data + addr_per_block))) { | |
1426 | ext4_free_data(handle, inode, parent_bh, &blk, &blk+1); | |
1427 | *i_data = 0; | |
1428 | } | |
1429 | brelse(bh); | |
1430 | bh = NULL; | |
1431 | } | |
1432 | ||
1433 | err: | |
1434 | return ret; | |
1435 | } | |
1436 | ||
26a4c0c6 TT |
1437 | int ext4_free_hole_blocks(handle_t *handle, struct inode *inode, |
1438 | ext4_lblk_t first, ext4_lblk_t stop) | |
8bad6fc8 ZL |
1439 | { |
1440 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1441 | int level, ret = 0; | |
1442 | int num = EXT4_NDIR_BLOCKS; | |
1443 | ext4_lblk_t count, max = EXT4_NDIR_BLOCKS; | |
1444 | __le32 *i_data = EXT4_I(inode)->i_data; | |
1445 | ||
1446 | count = stop - first; | |
1447 | for (level = 0; level < 4; level++, max *= addr_per_block) { | |
1448 | if (first < max) { | |
1449 | ret = free_hole_blocks(handle, inode, NULL, i_data, | |
1450 | level, first, count, num); | |
1451 | if (ret) | |
1452 | goto err; | |
1453 | if (count > max - first) | |
1454 | count -= max - first; | |
1455 | else | |
1456 | break; | |
1457 | first = 0; | |
1458 | } else { | |
1459 | first -= max; | |
1460 | } | |
1461 | i_data += num; | |
1462 | if (level == 0) { | |
1463 | num = 1; | |
1464 | max = 1; | |
1465 | } | |
1466 | } | |
1467 | ||
1468 | err: | |
1469 | return ret; | |
1470 | } | |
1471 |