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; | |
149 | ||
150 | *err = 0; | |
151 | /* i_data is not going away, no lock needed */ | |
152 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); | |
153 | if (!p->key) | |
154 | goto no_block; | |
155 | while (--depth) { | |
156 | bh = sb_getblk(sb, le32_to_cpu(p->key)); | |
157 | if (unlikely(!bh)) | |
158 | goto failure; | |
159 | ||
160 | if (!bh_uptodate_or_lock(bh)) { | |
161 | if (bh_submit_read(bh) < 0) { | |
162 | put_bh(bh); | |
163 | goto failure; | |
164 | } | |
165 | /* validate block references */ | |
166 | if (ext4_check_indirect_blockref(inode, bh)) { | |
167 | put_bh(bh); | |
168 | goto failure; | |
169 | } | |
170 | } | |
171 | ||
172 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); | |
173 | /* Reader: end */ | |
174 | if (!p->key) | |
175 | goto no_block; | |
176 | } | |
177 | return NULL; | |
178 | ||
179 | failure: | |
180 | *err = -EIO; | |
181 | no_block: | |
182 | return p; | |
183 | } | |
184 | ||
185 | /** | |
186 | * ext4_find_near - find a place for allocation with sufficient locality | |
187 | * @inode: owner | |
188 | * @ind: descriptor of indirect block. | |
189 | * | |
190 | * This function returns the preferred place for block allocation. | |
191 | * It is used when heuristic for sequential allocation fails. | |
192 | * Rules are: | |
193 | * + if there is a block to the left of our position - allocate near it. | |
194 | * + if pointer will live in indirect block - allocate near that block. | |
195 | * + if pointer will live in inode - allocate in the same | |
196 | * cylinder group. | |
197 | * | |
198 | * In the latter case we colour the starting block by the callers PID to | |
199 | * prevent it from clashing with concurrent allocations for a different inode | |
200 | * in the same block group. The PID is used here so that functionally related | |
201 | * files will be close-by on-disk. | |
202 | * | |
203 | * Caller must make sure that @ind is valid and will stay that way. | |
204 | */ | |
205 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) | |
206 | { | |
207 | struct ext4_inode_info *ei = EXT4_I(inode); | |
208 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; | |
209 | __le32 *p; | |
dae1e52c AG |
210 | |
211 | /* Try to find previous block */ | |
212 | for (p = ind->p - 1; p >= start; p--) { | |
213 | if (*p) | |
214 | return le32_to_cpu(*p); | |
215 | } | |
216 | ||
217 | /* No such thing, so let's try location of indirect block */ | |
218 | if (ind->bh) | |
219 | return ind->bh->b_blocknr; | |
220 | ||
221 | /* | |
222 | * It is going to be referred to from the inode itself? OK, just put it | |
223 | * into the same cylinder group then. | |
224 | */ | |
f86186b4 | 225 | return ext4_inode_to_goal_block(inode); |
dae1e52c AG |
226 | } |
227 | ||
228 | /** | |
229 | * ext4_find_goal - find a preferred place for allocation. | |
230 | * @inode: owner | |
231 | * @block: block we want | |
232 | * @partial: pointer to the last triple within a chain | |
233 | * | |
234 | * Normally this function find the preferred place for block allocation, | |
235 | * returns it. | |
236 | * Because this is only used for non-extent files, we limit the block nr | |
237 | * to 32 bits. | |
238 | */ | |
239 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, | |
240 | Indirect *partial) | |
241 | { | |
242 | ext4_fsblk_t goal; | |
243 | ||
244 | /* | |
245 | * XXX need to get goal block from mballoc's data structures | |
246 | */ | |
247 | ||
248 | goal = ext4_find_near(inode, partial); | |
249 | goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; | |
250 | return goal; | |
251 | } | |
252 | ||
253 | /** | |
254 | * ext4_blks_to_allocate - Look up the block map and count the number | |
255 | * of direct blocks need to be allocated for the given branch. | |
256 | * | |
257 | * @branch: chain of indirect blocks | |
258 | * @k: number of blocks need for indirect blocks | |
259 | * @blks: number of data blocks to be mapped. | |
260 | * @blocks_to_boundary: the offset in the indirect block | |
261 | * | |
262 | * return the total number of blocks to be allocate, including the | |
263 | * direct and indirect blocks. | |
264 | */ | |
265 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, | |
266 | int blocks_to_boundary) | |
267 | { | |
268 | unsigned int count = 0; | |
269 | ||
270 | /* | |
271 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
272 | * then it's clear blocks on that path have not allocated | |
273 | */ | |
274 | if (k > 0) { | |
275 | /* right now we don't handle cross boundary allocation */ | |
276 | if (blks < blocks_to_boundary + 1) | |
277 | count += blks; | |
278 | else | |
279 | count += blocks_to_boundary + 1; | |
280 | return count; | |
281 | } | |
282 | ||
283 | count++; | |
284 | while (count < blks && count <= blocks_to_boundary && | |
285 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
286 | count++; | |
287 | } | |
288 | return count; | |
289 | } | |
290 | ||
291 | /** | |
292 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch | |
293 | * @handle: handle for this transaction | |
294 | * @inode: inode which needs allocated blocks | |
295 | * @iblock: the logical block to start allocated at | |
296 | * @goal: preferred physical block of allocation | |
297 | * @indirect_blks: the number of blocks need to allocate for indirect | |
298 | * blocks | |
299 | * @blks: number of desired blocks | |
300 | * @new_blocks: on return it will store the new block numbers for | |
301 | * the indirect blocks(if needed) and the first direct block, | |
302 | * @err: on return it will store the error code | |
303 | * | |
304 | * This function will return the number of blocks allocated as | |
305 | * requested by the passed-in parameters. | |
306 | */ | |
307 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, | |
308 | ext4_lblk_t iblock, ext4_fsblk_t goal, | |
309 | int indirect_blks, int blks, | |
310 | ext4_fsblk_t new_blocks[4], int *err) | |
311 | { | |
312 | struct ext4_allocation_request ar; | |
313 | int target, i; | |
314 | unsigned long count = 0, blk_allocated = 0; | |
315 | int index = 0; | |
316 | ext4_fsblk_t current_block = 0; | |
317 | int ret = 0; | |
318 | ||
319 | /* | |
320 | * Here we try to allocate the requested multiple blocks at once, | |
321 | * on a best-effort basis. | |
322 | * To build a branch, we should allocate blocks for | |
323 | * the indirect blocks(if not allocated yet), and at least | |
324 | * the first direct block of this branch. That's the | |
325 | * minimum number of blocks need to allocate(required) | |
326 | */ | |
327 | /* first we try to allocate the indirect blocks */ | |
328 | target = indirect_blks; | |
329 | while (target > 0) { | |
330 | count = target; | |
331 | /* allocating blocks for indirect blocks and direct blocks */ | |
332 | current_block = ext4_new_meta_blocks(handle, inode, goal, | |
333 | 0, &count, err); | |
334 | if (*err) | |
335 | goto failed_out; | |
336 | ||
337 | if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) { | |
338 | EXT4_ERROR_INODE(inode, | |
339 | "current_block %llu + count %lu > %d!", | |
340 | current_block, count, | |
341 | EXT4_MAX_BLOCK_FILE_PHYS); | |
342 | *err = -EIO; | |
343 | goto failed_out; | |
344 | } | |
345 | ||
346 | target -= count; | |
347 | /* allocate blocks for indirect blocks */ | |
348 | while (index < indirect_blks && count) { | |
349 | new_blocks[index++] = current_block++; | |
350 | count--; | |
351 | } | |
352 | if (count > 0) { | |
353 | /* | |
354 | * save the new block number | |
355 | * for the first direct block | |
356 | */ | |
357 | new_blocks[index] = current_block; | |
358 | printk(KERN_INFO "%s returned more blocks than " | |
359 | "requested\n", __func__); | |
360 | WARN_ON(1); | |
361 | break; | |
362 | } | |
363 | } | |
364 | ||
365 | target = blks - count ; | |
366 | blk_allocated = count; | |
367 | if (!target) | |
368 | goto allocated; | |
369 | /* Now allocate data blocks */ | |
370 | memset(&ar, 0, sizeof(ar)); | |
371 | ar.inode = inode; | |
372 | ar.goal = goal; | |
373 | ar.len = target; | |
374 | ar.logical = iblock; | |
375 | if (S_ISREG(inode->i_mode)) | |
376 | /* enable in-core preallocation only for regular files */ | |
377 | ar.flags = EXT4_MB_HINT_DATA; | |
378 | ||
379 | current_block = ext4_mb_new_blocks(handle, &ar, err); | |
380 | if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) { | |
381 | EXT4_ERROR_INODE(inode, | |
382 | "current_block %llu + ar.len %d > %d!", | |
383 | current_block, ar.len, | |
384 | EXT4_MAX_BLOCK_FILE_PHYS); | |
385 | *err = -EIO; | |
386 | goto failed_out; | |
387 | } | |
388 | ||
389 | if (*err && (target == blks)) { | |
390 | /* | |
391 | * if the allocation failed and we didn't allocate | |
392 | * any blocks before | |
393 | */ | |
394 | goto failed_out; | |
395 | } | |
396 | if (!*err) { | |
397 | if (target == blks) { | |
398 | /* | |
399 | * save the new block number | |
400 | * for the first direct block | |
401 | */ | |
402 | new_blocks[index] = current_block; | |
403 | } | |
404 | blk_allocated += ar.len; | |
405 | } | |
406 | allocated: | |
407 | /* total number of blocks allocated for direct blocks */ | |
408 | ret = blk_allocated; | |
409 | *err = 0; | |
410 | return ret; | |
411 | failed_out: | |
412 | for (i = 0; i < index; i++) | |
413 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0); | |
414 | return ret; | |
415 | } | |
416 | ||
417 | /** | |
418 | * ext4_alloc_branch - allocate and set up a chain of blocks. | |
419 | * @handle: handle for this transaction | |
420 | * @inode: owner | |
421 | * @indirect_blks: number of allocated indirect blocks | |
422 | * @blks: number of allocated direct blocks | |
423 | * @goal: preferred place for allocation | |
424 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
425 | * @branch: place to store the chain in. | |
426 | * | |
427 | * This function allocates blocks, zeroes out all but the last one, | |
428 | * links them into chain and (if we are synchronous) writes them to disk. | |
429 | * In other words, it prepares a branch that can be spliced onto the | |
430 | * inode. It stores the information about that chain in the branch[], in | |
431 | * the same format as ext4_get_branch() would do. We are calling it after | |
432 | * we had read the existing part of chain and partial points to the last | |
433 | * triple of that (one with zero ->key). Upon the exit we have the same | |
434 | * picture as after the successful ext4_get_block(), except that in one | |
435 | * place chain is disconnected - *branch->p is still zero (we did not | |
436 | * set the last link), but branch->key contains the number that should | |
437 | * be placed into *branch->p to fill that gap. | |
438 | * | |
439 | * If allocation fails we free all blocks we've allocated (and forget | |
440 | * their buffer_heads) and return the error value the from failed | |
441 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain | |
442 | * as described above and return 0. | |
443 | */ | |
444 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, | |
445 | ext4_lblk_t iblock, int indirect_blks, | |
446 | int *blks, ext4_fsblk_t goal, | |
447 | ext4_lblk_t *offsets, Indirect *branch) | |
448 | { | |
449 | int blocksize = inode->i_sb->s_blocksize; | |
450 | int i, n = 0; | |
451 | int err = 0; | |
452 | struct buffer_head *bh; | |
453 | int num; | |
454 | ext4_fsblk_t new_blocks[4]; | |
455 | ext4_fsblk_t current_block; | |
456 | ||
457 | num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks, | |
458 | *blks, new_blocks, &err); | |
459 | if (err) | |
460 | return err; | |
461 | ||
462 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
463 | /* | |
464 | * metadata blocks and data blocks are allocated. | |
465 | */ | |
466 | for (n = 1; n <= indirect_blks; n++) { | |
467 | /* | |
468 | * Get buffer_head for parent block, zero it out | |
469 | * and set the pointer to new one, then send | |
470 | * parent to disk. | |
471 | */ | |
472 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
473 | if (unlikely(!bh)) { | |
474 | err = -EIO; | |
475 | goto failed; | |
476 | } | |
477 | ||
478 | branch[n].bh = bh; | |
479 | lock_buffer(bh); | |
480 | BUFFER_TRACE(bh, "call get_create_access"); | |
481 | err = ext4_journal_get_create_access(handle, bh); | |
482 | if (err) { | |
483 | /* Don't brelse(bh) here; it's done in | |
484 | * ext4_journal_forget() below */ | |
485 | unlock_buffer(bh); | |
486 | goto failed; | |
487 | } | |
488 | ||
489 | memset(bh->b_data, 0, blocksize); | |
490 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
491 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
492 | *branch[n].p = branch[n].key; | |
493 | if (n == indirect_blks) { | |
494 | current_block = new_blocks[n]; | |
495 | /* | |
496 | * End of chain, update the last new metablock of | |
497 | * the chain to point to the new allocated | |
498 | * data blocks numbers | |
499 | */ | |
500 | for (i = 1; i < num; i++) | |
501 | *(branch[n].p + i) = cpu_to_le32(++current_block); | |
502 | } | |
503 | BUFFER_TRACE(bh, "marking uptodate"); | |
504 | set_buffer_uptodate(bh); | |
505 | unlock_buffer(bh); | |
506 | ||
507 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
508 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
509 | if (err) | |
510 | goto failed; | |
511 | } | |
512 | *blks = num; | |
513 | return err; | |
514 | failed: | |
515 | /* Allocation failed, free what we already allocated */ | |
516 | ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0); | |
517 | for (i = 1; i <= n ; i++) { | |
518 | /* | |
519 | * branch[i].bh is newly allocated, so there is no | |
520 | * need to revoke the block, which is why we don't | |
521 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
522 | */ | |
523 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, | |
524 | EXT4_FREE_BLOCKS_FORGET); | |
525 | } | |
526 | for (i = n+1; i < indirect_blks; i++) | |
527 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0); | |
528 | ||
529 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0); | |
530 | ||
531 | return err; | |
532 | } | |
533 | ||
534 | /** | |
535 | * ext4_splice_branch - splice the allocated branch onto inode. | |
536 | * @handle: handle for this transaction | |
537 | * @inode: owner | |
538 | * @block: (logical) number of block we are adding | |
539 | * @chain: chain of indirect blocks (with a missing link - see | |
540 | * ext4_alloc_branch) | |
541 | * @where: location of missing link | |
542 | * @num: number of indirect blocks we are adding | |
543 | * @blks: number of direct blocks we are adding | |
544 | * | |
545 | * This function fills the missing link and does all housekeeping needed in | |
546 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
547 | * chain to new block and return 0. | |
548 | */ | |
549 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, | |
550 | ext4_lblk_t block, Indirect *where, int num, | |
551 | int blks) | |
552 | { | |
553 | int i; | |
554 | int err = 0; | |
555 | ext4_fsblk_t current_block; | |
556 | ||
557 | /* | |
558 | * If we're splicing into a [td]indirect block (as opposed to the | |
559 | * inode) then we need to get write access to the [td]indirect block | |
560 | * before the splice. | |
561 | */ | |
562 | if (where->bh) { | |
563 | BUFFER_TRACE(where->bh, "get_write_access"); | |
564 | err = ext4_journal_get_write_access(handle, where->bh); | |
565 | if (err) | |
566 | goto err_out; | |
567 | } | |
568 | /* That's it */ | |
569 | ||
570 | *where->p = where->key; | |
571 | ||
572 | /* | |
573 | * Update the host buffer_head or inode to point to more just allocated | |
574 | * direct blocks blocks | |
575 | */ | |
576 | if (num == 0 && blks > 1) { | |
577 | current_block = le32_to_cpu(where->key) + 1; | |
578 | for (i = 1; i < blks; i++) | |
579 | *(where->p + i) = cpu_to_le32(current_block++); | |
580 | } | |
581 | ||
582 | /* We are done with atomic stuff, now do the rest of housekeeping */ | |
583 | /* had we spliced it onto indirect block? */ | |
584 | if (where->bh) { | |
585 | /* | |
586 | * If we spliced it onto an indirect block, we haven't | |
587 | * altered the inode. Note however that if it is being spliced | |
588 | * onto an indirect block at the very end of the file (the | |
589 | * file is growing) then we *will* alter the inode to reflect | |
590 | * the new i_size. But that is not done here - it is done in | |
591 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. | |
592 | */ | |
593 | jbd_debug(5, "splicing indirect only\n"); | |
594 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); | |
595 | err = ext4_handle_dirty_metadata(handle, inode, where->bh); | |
596 | if (err) | |
597 | goto err_out; | |
598 | } else { | |
599 | /* | |
600 | * OK, we spliced it into the inode itself on a direct block. | |
601 | */ | |
602 | ext4_mark_inode_dirty(handle, inode); | |
603 | jbd_debug(5, "splicing direct\n"); | |
604 | } | |
605 | return err; | |
606 | ||
607 | err_out: | |
608 | for (i = 1; i <= num; i++) { | |
609 | /* | |
610 | * branch[i].bh is newly allocated, so there is no | |
611 | * need to revoke the block, which is why we don't | |
612 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
613 | */ | |
614 | ext4_free_blocks(handle, inode, where[i].bh, 0, 1, | |
615 | EXT4_FREE_BLOCKS_FORGET); | |
616 | } | |
617 | ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key), | |
618 | blks, 0); | |
619 | ||
620 | return err; | |
621 | } | |
622 | ||
623 | /* | |
624 | * The ext4_ind_map_blocks() function handles non-extents inodes | |
625 | * (i.e., using the traditional indirect/double-indirect i_blocks | |
626 | * scheme) for ext4_map_blocks(). | |
627 | * | |
628 | * Allocation strategy is simple: if we have to allocate something, we will | |
629 | * have to go the whole way to leaf. So let's do it before attaching anything | |
630 | * to tree, set linkage between the newborn blocks, write them if sync is | |
631 | * required, recheck the path, free and repeat if check fails, otherwise | |
632 | * set the last missing link (that will protect us from any truncate-generated | |
633 | * removals - all blocks on the path are immune now) and possibly force the | |
634 | * write on the parent block. | |
635 | * That has a nice additional property: no special recovery from the failed | |
636 | * allocations is needed - we simply release blocks and do not touch anything | |
637 | * reachable from inode. | |
638 | * | |
639 | * `handle' can be NULL if create == 0. | |
640 | * | |
641 | * return > 0, # of blocks mapped or allocated. | |
642 | * return = 0, if plain lookup failed. | |
643 | * return < 0, error case. | |
644 | * | |
645 | * The ext4_ind_get_blocks() function should be called with | |
646 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem | |
647 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or | |
648 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system | |
649 | * blocks. | |
650 | */ | |
651 | int ext4_ind_map_blocks(handle_t *handle, struct inode *inode, | |
652 | struct ext4_map_blocks *map, | |
653 | int flags) | |
654 | { | |
655 | int err = -EIO; | |
656 | ext4_lblk_t offsets[4]; | |
657 | Indirect chain[4]; | |
658 | Indirect *partial; | |
659 | ext4_fsblk_t goal; | |
660 | int indirect_blks; | |
661 | int blocks_to_boundary = 0; | |
662 | int depth; | |
663 | int count = 0; | |
664 | ext4_fsblk_t first_block = 0; | |
665 | ||
666 | trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); | |
667 | J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); | |
668 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); | |
669 | depth = ext4_block_to_path(inode, map->m_lblk, offsets, | |
670 | &blocks_to_boundary); | |
671 | ||
672 | if (depth == 0) | |
673 | goto out; | |
674 | ||
675 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); | |
676 | ||
677 | /* Simplest case - block found, no allocation needed */ | |
678 | if (!partial) { | |
679 | first_block = le32_to_cpu(chain[depth - 1].key); | |
680 | count++; | |
681 | /*map more blocks*/ | |
682 | while (count < map->m_len && count <= blocks_to_boundary) { | |
683 | ext4_fsblk_t blk; | |
684 | ||
685 | blk = le32_to_cpu(*(chain[depth-1].p + count)); | |
686 | ||
687 | if (blk == first_block + count) | |
688 | count++; | |
689 | else | |
690 | break; | |
691 | } | |
692 | goto got_it; | |
693 | } | |
694 | ||
695 | /* Next simple case - plain lookup or failed read of indirect block */ | |
696 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO) | |
697 | goto cleanup; | |
698 | ||
699 | /* | |
700 | * Okay, we need to do block allocation. | |
701 | */ | |
bab08ab9 TT |
702 | if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, |
703 | EXT4_FEATURE_RO_COMPAT_BIGALLOC)) { | |
704 | EXT4_ERROR_INODE(inode, "Can't allocate blocks for " | |
705 | "non-extent mapped inodes with bigalloc"); | |
706 | return -ENOSPC; | |
707 | } | |
708 | ||
dae1e52c AG |
709 | goal = ext4_find_goal(inode, map->m_lblk, partial); |
710 | ||
711 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
712 | indirect_blks = (chain + depth) - partial - 1; | |
713 | ||
714 | /* | |
715 | * Next look up the indirect map to count the totoal number of | |
716 | * direct blocks to allocate for this branch. | |
717 | */ | |
718 | count = ext4_blks_to_allocate(partial, indirect_blks, | |
719 | map->m_len, blocks_to_boundary); | |
720 | /* | |
721 | * Block out ext4_truncate while we alter the tree | |
722 | */ | |
723 | err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks, | |
724 | &count, goal, | |
725 | offsets + (partial - chain), partial); | |
726 | ||
727 | /* | |
728 | * The ext4_splice_branch call will free and forget any buffers | |
729 | * on the new chain if there is a failure, but that risks using | |
730 | * up transaction credits, especially for bitmaps where the | |
731 | * credits cannot be returned. Can we handle this somehow? We | |
732 | * may need to return -EAGAIN upwards in the worst case. --sct | |
733 | */ | |
734 | if (!err) | |
735 | err = ext4_splice_branch(handle, inode, map->m_lblk, | |
736 | partial, indirect_blks, count); | |
737 | if (err) | |
738 | goto cleanup; | |
739 | ||
740 | map->m_flags |= EXT4_MAP_NEW; | |
741 | ||
742 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
743 | got_it: | |
744 | map->m_flags |= EXT4_MAP_MAPPED; | |
745 | map->m_pblk = le32_to_cpu(chain[depth-1].key); | |
746 | map->m_len = count; | |
747 | if (count > blocks_to_boundary) | |
748 | map->m_flags |= EXT4_MAP_BOUNDARY; | |
749 | err = count; | |
750 | /* Clean up and exit */ | |
751 | partial = chain + depth - 1; /* the whole chain */ | |
752 | cleanup: | |
753 | while (partial > chain) { | |
754 | BUFFER_TRACE(partial->bh, "call brelse"); | |
755 | brelse(partial->bh); | |
756 | partial--; | |
757 | } | |
758 | out: | |
19b303d8 | 759 | trace_ext4_ind_map_blocks_exit(inode, map, err); |
dae1e52c AG |
760 | return err; |
761 | } | |
762 | ||
763 | /* | |
764 | * O_DIRECT for ext3 (or indirect map) based files | |
765 | * | |
766 | * If the O_DIRECT write will extend the file then add this inode to the | |
767 | * orphan list. So recovery will truncate it back to the original size | |
768 | * if the machine crashes during the write. | |
769 | * | |
770 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
771 | * crashes then stale disk data _may_ be exposed inside the file. But current | |
772 | * VFS code falls back into buffered path in that case so we are safe. | |
773 | */ | |
774 | ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb, | |
775 | const struct iovec *iov, loff_t offset, | |
776 | unsigned long nr_segs) | |
777 | { | |
778 | struct file *file = iocb->ki_filp; | |
779 | struct inode *inode = file->f_mapping->host; | |
780 | struct ext4_inode_info *ei = EXT4_I(inode); | |
781 | handle_t *handle; | |
782 | ssize_t ret; | |
783 | int orphan = 0; | |
784 | size_t count = iov_length(iov, nr_segs); | |
785 | int retries = 0; | |
786 | ||
787 | if (rw == WRITE) { | |
788 | loff_t final_size = offset + count; | |
789 | ||
790 | if (final_size > inode->i_size) { | |
791 | /* Credits for sb + inode write */ | |
792 | handle = ext4_journal_start(inode, 2); | |
793 | if (IS_ERR(handle)) { | |
794 | ret = PTR_ERR(handle); | |
795 | goto out; | |
796 | } | |
797 | ret = ext4_orphan_add(handle, inode); | |
798 | if (ret) { | |
799 | ext4_journal_stop(handle); | |
800 | goto out; | |
801 | } | |
802 | orphan = 1; | |
803 | ei->i_disksize = inode->i_size; | |
804 | ext4_journal_stop(handle); | |
805 | } | |
806 | } | |
807 | ||
808 | retry: | |
dccaf33f | 809 | if (rw == READ && ext4_should_dioread_nolock(inode)) { |
c278531d DM |
810 | if (unlikely(atomic_read(&EXT4_I(inode)->i_unwritten))) { |
811 | mutex_lock(&inode->i_mutex); | |
812 | ext4_flush_unwritten_io(inode); | |
813 | mutex_unlock(&inode->i_mutex); | |
814 | } | |
17335dcc DM |
815 | /* |
816 | * Nolock dioread optimization may be dynamically disabled | |
817 | * via ext4_inode_block_unlocked_dio(). Check inode's state | |
818 | * while holding extra i_dio_count ref. | |
819 | */ | |
820 | atomic_inc(&inode->i_dio_count); | |
821 | smp_mb(); | |
822 | if (unlikely(ext4_test_inode_state(inode, | |
823 | EXT4_STATE_DIOREAD_LOCK))) { | |
824 | inode_dio_done(inode); | |
825 | goto locked; | |
826 | } | |
dae1e52c AG |
827 | ret = __blockdev_direct_IO(rw, iocb, inode, |
828 | inode->i_sb->s_bdev, iov, | |
829 | offset, nr_segs, | |
830 | ext4_get_block, NULL, NULL, 0); | |
17335dcc | 831 | inode_dio_done(inode); |
dccaf33f | 832 | } else { |
17335dcc | 833 | locked: |
60ad4466 LT |
834 | ret = blockdev_direct_IO(rw, iocb, inode, iov, |
835 | offset, nr_segs, ext4_get_block); | |
dae1e52c AG |
836 | |
837 | if (unlikely((rw & WRITE) && ret < 0)) { | |
838 | loff_t isize = i_size_read(inode); | |
839 | loff_t end = offset + iov_length(iov, nr_segs); | |
840 | ||
841 | if (end > isize) | |
842 | ext4_truncate_failed_write(inode); | |
843 | } | |
844 | } | |
845 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
846 | goto retry; | |
847 | ||
848 | if (orphan) { | |
849 | int err; | |
850 | ||
851 | /* Credits for sb + inode write */ | |
852 | handle = ext4_journal_start(inode, 2); | |
853 | if (IS_ERR(handle)) { | |
854 | /* This is really bad luck. We've written the data | |
855 | * but cannot extend i_size. Bail out and pretend | |
856 | * the write failed... */ | |
857 | ret = PTR_ERR(handle); | |
858 | if (inode->i_nlink) | |
859 | ext4_orphan_del(NULL, inode); | |
860 | ||
861 | goto out; | |
862 | } | |
863 | if (inode->i_nlink) | |
864 | ext4_orphan_del(handle, inode); | |
865 | if (ret > 0) { | |
866 | loff_t end = offset + ret; | |
867 | if (end > inode->i_size) { | |
868 | ei->i_disksize = end; | |
869 | i_size_write(inode, end); | |
870 | /* | |
871 | * We're going to return a positive `ret' | |
872 | * here due to non-zero-length I/O, so there's | |
873 | * no way of reporting error returns from | |
874 | * ext4_mark_inode_dirty() to userspace. So | |
875 | * ignore it. | |
876 | */ | |
877 | ext4_mark_inode_dirty(handle, inode); | |
878 | } | |
879 | } | |
880 | err = ext4_journal_stop(handle); | |
881 | if (ret == 0) | |
882 | ret = err; | |
883 | } | |
884 | out: | |
885 | return ret; | |
886 | } | |
887 | ||
888 | /* | |
889 | * Calculate the number of metadata blocks need to reserve | |
890 | * to allocate a new block at @lblocks for non extent file based file | |
891 | */ | |
892 | int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock) | |
893 | { | |
894 | struct ext4_inode_info *ei = EXT4_I(inode); | |
895 | sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1); | |
896 | int blk_bits; | |
897 | ||
898 | if (lblock < EXT4_NDIR_BLOCKS) | |
899 | return 0; | |
900 | ||
901 | lblock -= EXT4_NDIR_BLOCKS; | |
902 | ||
903 | if (ei->i_da_metadata_calc_len && | |
904 | (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) { | |
905 | ei->i_da_metadata_calc_len++; | |
906 | return 0; | |
907 | } | |
908 | ei->i_da_metadata_calc_last_lblock = lblock & dind_mask; | |
909 | ei->i_da_metadata_calc_len = 1; | |
910 | blk_bits = order_base_2(lblock); | |
911 | return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1; | |
912 | } | |
913 | ||
914 | int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
915 | { | |
916 | int indirects; | |
917 | ||
918 | /* if nrblocks are contiguous */ | |
919 | if (chunk) { | |
920 | /* | |
921 | * With N contiguous data blocks, we need at most | |
922 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks, | |
923 | * 2 dindirect blocks, and 1 tindirect block | |
924 | */ | |
925 | return DIV_ROUND_UP(nrblocks, | |
926 | EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4; | |
927 | } | |
928 | /* | |
929 | * if nrblocks are not contiguous, worse case, each block touch | |
930 | * a indirect block, and each indirect block touch a double indirect | |
931 | * block, plus a triple indirect block | |
932 | */ | |
933 | indirects = nrblocks * 2 + 1; | |
934 | return indirects; | |
935 | } | |
936 | ||
937 | /* | |
938 | * Truncate transactions can be complex and absolutely huge. So we need to | |
939 | * be able to restart the transaction at a conventient checkpoint to make | |
940 | * sure we don't overflow the journal. | |
941 | * | |
942 | * start_transaction gets us a new handle for a truncate transaction, | |
943 | * and extend_transaction tries to extend the existing one a bit. If | |
944 | * extend fails, we need to propagate the failure up and restart the | |
945 | * transaction in the top-level truncate loop. --sct | |
946 | */ | |
947 | static handle_t *start_transaction(struct inode *inode) | |
948 | { | |
949 | handle_t *result; | |
950 | ||
951 | result = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)); | |
952 | if (!IS_ERR(result)) | |
953 | return result; | |
954 | ||
955 | ext4_std_error(inode->i_sb, PTR_ERR(result)); | |
956 | return result; | |
957 | } | |
958 | ||
959 | /* | |
960 | * Try to extend this transaction for the purposes of truncation. | |
961 | * | |
962 | * Returns 0 if we managed to create more room. If we can't create more | |
963 | * room, and the transaction must be restarted we return 1. | |
964 | */ | |
965 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
966 | { | |
967 | if (!ext4_handle_valid(handle)) | |
968 | return 0; | |
969 | if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) | |
970 | return 0; | |
971 | if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode))) | |
972 | return 0; | |
973 | return 1; | |
974 | } | |
975 | ||
976 | /* | |
977 | * Probably it should be a library function... search for first non-zero word | |
978 | * or memcmp with zero_page, whatever is better for particular architecture. | |
979 | * Linus? | |
980 | */ | |
981 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
982 | { | |
983 | while (p < q) | |
984 | if (*p++) | |
985 | return 0; | |
986 | return 1; | |
987 | } | |
988 | ||
989 | /** | |
990 | * ext4_find_shared - find the indirect blocks for partial truncation. | |
991 | * @inode: inode in question | |
992 | * @depth: depth of the affected branch | |
993 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) | |
994 | * @chain: place to store the pointers to partial indirect blocks | |
995 | * @top: place to the (detached) top of branch | |
996 | * | |
997 | * This is a helper function used by ext4_truncate(). | |
998 | * | |
999 | * When we do truncate() we may have to clean the ends of several | |
1000 | * indirect blocks but leave the blocks themselves alive. Block is | |
1001 | * partially truncated if some data below the new i_size is referred | |
1002 | * from it (and it is on the path to the first completely truncated | |
1003 | * data block, indeed). We have to free the top of that path along | |
1004 | * with everything to the right of the path. Since no allocation | |
1005 | * past the truncation point is possible until ext4_truncate() | |
1006 | * finishes, we may safely do the latter, but top of branch may | |
1007 | * require special attention - pageout below the truncation point | |
1008 | * might try to populate it. | |
1009 | * | |
1010 | * We atomically detach the top of branch from the tree, store the | |
1011 | * block number of its root in *@top, pointers to buffer_heads of | |
1012 | * partially truncated blocks - in @chain[].bh and pointers to | |
1013 | * their last elements that should not be removed - in | |
1014 | * @chain[].p. Return value is the pointer to last filled element | |
1015 | * of @chain. | |
1016 | * | |
1017 | * The work left to caller to do the actual freeing of subtrees: | |
1018 | * a) free the subtree starting from *@top | |
1019 | * b) free the subtrees whose roots are stored in | |
1020 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
1021 | * c) free the subtrees growing from the inode past the @chain[0]. | |
1022 | * (no partially truncated stuff there). */ | |
1023 | ||
1024 | static Indirect *ext4_find_shared(struct inode *inode, int depth, | |
1025 | ext4_lblk_t offsets[4], Indirect chain[4], | |
1026 | __le32 *top) | |
1027 | { | |
1028 | Indirect *partial, *p; | |
1029 | int k, err; | |
1030 | ||
1031 | *top = 0; | |
1032 | /* Make k index the deepest non-null offset + 1 */ | |
1033 | for (k = depth; k > 1 && !offsets[k-1]; k--) | |
1034 | ; | |
1035 | partial = ext4_get_branch(inode, k, offsets, chain, &err); | |
1036 | /* Writer: pointers */ | |
1037 | if (!partial) | |
1038 | partial = chain + k-1; | |
1039 | /* | |
1040 | * If the branch acquired continuation since we've looked at it - | |
1041 | * fine, it should all survive and (new) top doesn't belong to us. | |
1042 | */ | |
1043 | if (!partial->key && *partial->p) | |
1044 | /* Writer: end */ | |
1045 | goto no_top; | |
1046 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) | |
1047 | ; | |
1048 | /* | |
1049 | * OK, we've found the last block that must survive. The rest of our | |
1050 | * branch should be detached before unlocking. However, if that rest | |
1051 | * of branch is all ours and does not grow immediately from the inode | |
1052 | * it's easier to cheat and just decrement partial->p. | |
1053 | */ | |
1054 | if (p == chain + k - 1 && p > chain) { | |
1055 | p->p--; | |
1056 | } else { | |
1057 | *top = *p->p; | |
1058 | /* Nope, don't do this in ext4. Must leave the tree intact */ | |
1059 | #if 0 | |
1060 | *p->p = 0; | |
1061 | #endif | |
1062 | } | |
1063 | /* Writer: end */ | |
1064 | ||
1065 | while (partial > p) { | |
1066 | brelse(partial->bh); | |
1067 | partial--; | |
1068 | } | |
1069 | no_top: | |
1070 | return partial; | |
1071 | } | |
1072 | ||
1073 | /* | |
1074 | * Zero a number of block pointers in either an inode or an indirect block. | |
1075 | * If we restart the transaction we must again get write access to the | |
1076 | * indirect block for further modification. | |
1077 | * | |
1078 | * We release `count' blocks on disk, but (last - first) may be greater | |
1079 | * than `count' because there can be holes in there. | |
1080 | * | |
1081 | * Return 0 on success, 1 on invalid block range | |
1082 | * and < 0 on fatal error. | |
1083 | */ | |
1084 | static int ext4_clear_blocks(handle_t *handle, struct inode *inode, | |
1085 | struct buffer_head *bh, | |
1086 | ext4_fsblk_t block_to_free, | |
1087 | unsigned long count, __le32 *first, | |
1088 | __le32 *last) | |
1089 | { | |
1090 | __le32 *p; | |
1091 | int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED; | |
1092 | int err; | |
1093 | ||
1094 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
1095 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
1096 | ||
1097 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free, | |
1098 | count)) { | |
1099 | EXT4_ERROR_INODE(inode, "attempt to clear invalid " | |
1100 | "blocks %llu len %lu", | |
1101 | (unsigned long long) block_to_free, count); | |
1102 | return 1; | |
1103 | } | |
1104 | ||
1105 | if (try_to_extend_transaction(handle, inode)) { | |
1106 | if (bh) { | |
1107 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
1108 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
1109 | if (unlikely(err)) | |
1110 | goto out_err; | |
1111 | } | |
1112 | err = ext4_mark_inode_dirty(handle, inode); | |
1113 | if (unlikely(err)) | |
1114 | goto out_err; | |
1115 | err = ext4_truncate_restart_trans(handle, inode, | |
1116 | ext4_blocks_for_truncate(inode)); | |
1117 | if (unlikely(err)) | |
1118 | goto out_err; | |
1119 | if (bh) { | |
1120 | BUFFER_TRACE(bh, "retaking write access"); | |
1121 | err = ext4_journal_get_write_access(handle, bh); | |
1122 | if (unlikely(err)) | |
1123 | goto out_err; | |
1124 | } | |
1125 | } | |
1126 | ||
1127 | for (p = first; p < last; p++) | |
1128 | *p = 0; | |
1129 | ||
1130 | ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags); | |
1131 | return 0; | |
1132 | out_err: | |
1133 | ext4_std_error(inode->i_sb, err); | |
1134 | return err; | |
1135 | } | |
1136 | ||
1137 | /** | |
1138 | * ext4_free_data - free a list of data blocks | |
1139 | * @handle: handle for this transaction | |
1140 | * @inode: inode we are dealing with | |
1141 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
1142 | * @first: array of block numbers | |
1143 | * @last: points immediately past the end of array | |
1144 | * | |
1145 | * We are freeing all blocks referred from that array (numbers are stored as | |
1146 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
1147 | * | |
1148 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
1149 | * blocks are contiguous then releasing them at one time will only affect one | |
1150 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
1151 | * actually use a lot of journal space. | |
1152 | * | |
1153 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
1154 | * block pointers. | |
1155 | */ | |
1156 | static void ext4_free_data(handle_t *handle, struct inode *inode, | |
1157 | struct buffer_head *this_bh, | |
1158 | __le32 *first, __le32 *last) | |
1159 | { | |
1160 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ | |
1161 | unsigned long count = 0; /* Number of blocks in the run */ | |
1162 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
1163 | corresponding to | |
1164 | block_to_free */ | |
1165 | ext4_fsblk_t nr; /* Current block # */ | |
1166 | __le32 *p; /* Pointer into inode/ind | |
1167 | for current block */ | |
1168 | int err = 0; | |
1169 | ||
1170 | if (this_bh) { /* For indirect block */ | |
1171 | BUFFER_TRACE(this_bh, "get_write_access"); | |
1172 | err = ext4_journal_get_write_access(handle, this_bh); | |
1173 | /* Important: if we can't update the indirect pointers | |
1174 | * to the blocks, we can't free them. */ | |
1175 | if (err) | |
1176 | return; | |
1177 | } | |
1178 | ||
1179 | for (p = first; p < last; p++) { | |
1180 | nr = le32_to_cpu(*p); | |
1181 | if (nr) { | |
1182 | /* accumulate blocks to free if they're contiguous */ | |
1183 | if (count == 0) { | |
1184 | block_to_free = nr; | |
1185 | block_to_free_p = p; | |
1186 | count = 1; | |
1187 | } else if (nr == block_to_free + count) { | |
1188 | count++; | |
1189 | } else { | |
1190 | err = ext4_clear_blocks(handle, inode, this_bh, | |
1191 | block_to_free, count, | |
1192 | block_to_free_p, p); | |
1193 | if (err) | |
1194 | break; | |
1195 | block_to_free = nr; | |
1196 | block_to_free_p = p; | |
1197 | count = 1; | |
1198 | } | |
1199 | } | |
1200 | } | |
1201 | ||
1202 | if (!err && count > 0) | |
1203 | err = ext4_clear_blocks(handle, inode, this_bh, block_to_free, | |
1204 | count, block_to_free_p, p); | |
1205 | if (err < 0) | |
1206 | /* fatal error */ | |
1207 | return; | |
1208 | ||
1209 | if (this_bh) { | |
1210 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); | |
1211 | ||
1212 | /* | |
1213 | * The buffer head should have an attached journal head at this | |
1214 | * point. However, if the data is corrupted and an indirect | |
1215 | * block pointed to itself, it would have been detached when | |
1216 | * the block was cleared. Check for this instead of OOPSing. | |
1217 | */ | |
1218 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) | |
1219 | ext4_handle_dirty_metadata(handle, inode, this_bh); | |
1220 | else | |
1221 | EXT4_ERROR_INODE(inode, | |
1222 | "circular indirect block detected at " | |
1223 | "block %llu", | |
1224 | (unsigned long long) this_bh->b_blocknr); | |
1225 | } | |
1226 | } | |
1227 | ||
1228 | /** | |
1229 | * ext4_free_branches - free an array of branches | |
1230 | * @handle: JBD handle for this transaction | |
1231 | * @inode: inode we are dealing with | |
1232 | * @parent_bh: the buffer_head which contains *@first and *@last | |
1233 | * @first: array of block numbers | |
1234 | * @last: pointer immediately past the end of array | |
1235 | * @depth: depth of the branches to free | |
1236 | * | |
1237 | * We are freeing all blocks referred from these branches (numbers are | |
1238 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
1239 | * appropriately. | |
1240 | */ | |
1241 | static void ext4_free_branches(handle_t *handle, struct inode *inode, | |
1242 | struct buffer_head *parent_bh, | |
1243 | __le32 *first, __le32 *last, int depth) | |
1244 | { | |
1245 | ext4_fsblk_t nr; | |
1246 | __le32 *p; | |
1247 | ||
1248 | if (ext4_handle_is_aborted(handle)) | |
1249 | return; | |
1250 | ||
1251 | if (depth--) { | |
1252 | struct buffer_head *bh; | |
1253 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1254 | p = last; | |
1255 | while (--p >= first) { | |
1256 | nr = le32_to_cpu(*p); | |
1257 | if (!nr) | |
1258 | continue; /* A hole */ | |
1259 | ||
1260 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), | |
1261 | nr, 1)) { | |
1262 | EXT4_ERROR_INODE(inode, | |
1263 | "invalid indirect mapped " | |
1264 | "block %lu (level %d)", | |
1265 | (unsigned long) nr, depth); | |
1266 | break; | |
1267 | } | |
1268 | ||
1269 | /* Go read the buffer for the next level down */ | |
1270 | bh = sb_bread(inode->i_sb, nr); | |
1271 | ||
1272 | /* | |
1273 | * A read failure? Report error and clear slot | |
1274 | * (should be rare). | |
1275 | */ | |
1276 | if (!bh) { | |
1277 | EXT4_ERROR_INODE_BLOCK(inode, nr, | |
1278 | "Read failure"); | |
1279 | continue; | |
1280 | } | |
1281 | ||
1282 | /* This zaps the entire block. Bottom up. */ | |
1283 | BUFFER_TRACE(bh, "free child branches"); | |
1284 | ext4_free_branches(handle, inode, bh, | |
1285 | (__le32 *) bh->b_data, | |
1286 | (__le32 *) bh->b_data + addr_per_block, | |
1287 | depth); | |
1288 | brelse(bh); | |
1289 | ||
1290 | /* | |
1291 | * Everything below this this pointer has been | |
1292 | * released. Now let this top-of-subtree go. | |
1293 | * | |
1294 | * We want the freeing of this indirect block to be | |
1295 | * atomic in the journal with the updating of the | |
1296 | * bitmap block which owns it. So make some room in | |
1297 | * the journal. | |
1298 | * | |
1299 | * We zero the parent pointer *after* freeing its | |
1300 | * pointee in the bitmaps, so if extend_transaction() | |
1301 | * for some reason fails to put the bitmap changes and | |
1302 | * the release into the same transaction, recovery | |
1303 | * will merely complain about releasing a free block, | |
1304 | * rather than leaking blocks. | |
1305 | */ | |
1306 | if (ext4_handle_is_aborted(handle)) | |
1307 | return; | |
1308 | if (try_to_extend_transaction(handle, inode)) { | |
1309 | ext4_mark_inode_dirty(handle, inode); | |
1310 | ext4_truncate_restart_trans(handle, inode, | |
1311 | ext4_blocks_for_truncate(inode)); | |
1312 | } | |
1313 | ||
1314 | /* | |
1315 | * The forget flag here is critical because if | |
1316 | * we are journaling (and not doing data | |
1317 | * journaling), we have to make sure a revoke | |
1318 | * record is written to prevent the journal | |
1319 | * replay from overwriting the (former) | |
1320 | * indirect block if it gets reallocated as a | |
1321 | * data block. This must happen in the same | |
1322 | * transaction where the data blocks are | |
1323 | * actually freed. | |
1324 | */ | |
1325 | ext4_free_blocks(handle, inode, NULL, nr, 1, | |
1326 | EXT4_FREE_BLOCKS_METADATA| | |
1327 | EXT4_FREE_BLOCKS_FORGET); | |
1328 | ||
1329 | if (parent_bh) { | |
1330 | /* | |
1331 | * The block which we have just freed is | |
1332 | * pointed to by an indirect block: journal it | |
1333 | */ | |
1334 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
1335 | if (!ext4_journal_get_write_access(handle, | |
1336 | parent_bh)){ | |
1337 | *p = 0; | |
1338 | BUFFER_TRACE(parent_bh, | |
1339 | "call ext4_handle_dirty_metadata"); | |
1340 | ext4_handle_dirty_metadata(handle, | |
1341 | inode, | |
1342 | parent_bh); | |
1343 | } | |
1344 | } | |
1345 | } | |
1346 | } else { | |
1347 | /* We have reached the bottom of the tree. */ | |
1348 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
1349 | ext4_free_data(handle, inode, parent_bh, first, last); | |
1350 | } | |
1351 | } | |
1352 | ||
1353 | void ext4_ind_truncate(struct inode *inode) | |
1354 | { | |
1355 | handle_t *handle; | |
1356 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1357 | __le32 *i_data = ei->i_data; | |
1358 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1359 | struct address_space *mapping = inode->i_mapping; | |
1360 | ext4_lblk_t offsets[4]; | |
1361 | Indirect chain[4]; | |
1362 | Indirect *partial; | |
1363 | __le32 nr = 0; | |
1364 | int n = 0; | |
1365 | ext4_lblk_t last_block, max_block; | |
189e868f | 1366 | loff_t page_len; |
dae1e52c | 1367 | unsigned blocksize = inode->i_sb->s_blocksize; |
189e868f | 1368 | int err; |
dae1e52c AG |
1369 | |
1370 | handle = start_transaction(inode); | |
1371 | if (IS_ERR(handle)) | |
1372 | return; /* AKPM: return what? */ | |
1373 | ||
1374 | last_block = (inode->i_size + blocksize-1) | |
1375 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); | |
1376 | max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1) | |
1377 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); | |
1378 | ||
189e868f AH |
1379 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
1380 | page_len = PAGE_CACHE_SIZE - | |
1381 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
1382 | ||
1383 | err = ext4_discard_partial_page_buffers(handle, | |
1384 | mapping, inode->i_size, page_len, 0); | |
1385 | ||
1386 | if (err) | |
dae1e52c | 1387 | goto out_stop; |
189e868f | 1388 | } |
dae1e52c AG |
1389 | |
1390 | if (last_block != max_block) { | |
1391 | n = ext4_block_to_path(inode, last_block, offsets, NULL); | |
1392 | if (n == 0) | |
1393 | goto out_stop; /* error */ | |
1394 | } | |
1395 | ||
1396 | /* | |
1397 | * OK. This truncate is going to happen. We add the inode to the | |
1398 | * orphan list, so that if this truncate spans multiple transactions, | |
1399 | * and we crash, we will resume the truncate when the filesystem | |
1400 | * recovers. It also marks the inode dirty, to catch the new size. | |
1401 | * | |
1402 | * Implication: the file must always be in a sane, consistent | |
1403 | * truncatable state while each transaction commits. | |
1404 | */ | |
1405 | if (ext4_orphan_add(handle, inode)) | |
1406 | goto out_stop; | |
1407 | ||
1408 | /* | |
1409 | * From here we block out all ext4_get_block() callers who want to | |
1410 | * modify the block allocation tree. | |
1411 | */ | |
1412 | down_write(&ei->i_data_sem); | |
1413 | ||
1414 | ext4_discard_preallocations(inode); | |
51865fda | 1415 | ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block); |
dae1e52c AG |
1416 | |
1417 | /* | |
1418 | * The orphan list entry will now protect us from any crash which | |
1419 | * occurs before the truncate completes, so it is now safe to propagate | |
1420 | * the new, shorter inode size (held for now in i_size) into the | |
1421 | * on-disk inode. We do this via i_disksize, which is the value which | |
1422 | * ext4 *really* writes onto the disk inode. | |
1423 | */ | |
1424 | ei->i_disksize = inode->i_size; | |
1425 | ||
1426 | if (last_block == max_block) { | |
1427 | /* | |
1428 | * It is unnecessary to free any data blocks if last_block is | |
1429 | * equal to the indirect block limit. | |
1430 | */ | |
1431 | goto out_unlock; | |
1432 | } else if (n == 1) { /* direct blocks */ | |
1433 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], | |
1434 | i_data + EXT4_NDIR_BLOCKS); | |
1435 | goto do_indirects; | |
1436 | } | |
1437 | ||
1438 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); | |
1439 | /* Kill the top of shared branch (not detached) */ | |
1440 | if (nr) { | |
1441 | if (partial == chain) { | |
1442 | /* Shared branch grows from the inode */ | |
1443 | ext4_free_branches(handle, inode, NULL, | |
1444 | &nr, &nr+1, (chain+n-1) - partial); | |
1445 | *partial->p = 0; | |
1446 | /* | |
1447 | * We mark the inode dirty prior to restart, | |
1448 | * and prior to stop. No need for it here. | |
1449 | */ | |
1450 | } else { | |
1451 | /* Shared branch grows from an indirect block */ | |
1452 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
1453 | ext4_free_branches(handle, inode, partial->bh, | |
1454 | partial->p, | |
1455 | partial->p+1, (chain+n-1) - partial); | |
1456 | } | |
1457 | } | |
1458 | /* Clear the ends of indirect blocks on the shared branch */ | |
1459 | while (partial > chain) { | |
1460 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, | |
1461 | (__le32*)partial->bh->b_data+addr_per_block, | |
1462 | (chain+n-1) - partial); | |
1463 | BUFFER_TRACE(partial->bh, "call brelse"); | |
1464 | brelse(partial->bh); | |
1465 | partial--; | |
1466 | } | |
1467 | do_indirects: | |
1468 | /* Kill the remaining (whole) subtrees */ | |
1469 | switch (offsets[0]) { | |
1470 | default: | |
1471 | nr = i_data[EXT4_IND_BLOCK]; | |
1472 | if (nr) { | |
1473 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); | |
1474 | i_data[EXT4_IND_BLOCK] = 0; | |
1475 | } | |
1476 | case EXT4_IND_BLOCK: | |
1477 | nr = i_data[EXT4_DIND_BLOCK]; | |
1478 | if (nr) { | |
1479 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); | |
1480 | i_data[EXT4_DIND_BLOCK] = 0; | |
1481 | } | |
1482 | case EXT4_DIND_BLOCK: | |
1483 | nr = i_data[EXT4_TIND_BLOCK]; | |
1484 | if (nr) { | |
1485 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); | |
1486 | i_data[EXT4_TIND_BLOCK] = 0; | |
1487 | } | |
1488 | case EXT4_TIND_BLOCK: | |
1489 | ; | |
1490 | } | |
1491 | ||
1492 | out_unlock: | |
1493 | up_write(&ei->i_data_sem); | |
1494 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
1495 | ext4_mark_inode_dirty(handle, inode); | |
1496 | ||
1497 | /* | |
1498 | * In a multi-transaction truncate, we only make the final transaction | |
1499 | * synchronous | |
1500 | */ | |
1501 | if (IS_SYNC(inode)) | |
1502 | ext4_handle_sync(handle); | |
1503 | out_stop: | |
1504 | /* | |
1505 | * If this was a simple ftruncate(), and the file will remain alive | |
1506 | * then we need to clear up the orphan record which we created above. | |
1507 | * However, if this was a real unlink then we were called by | |
1508 | * ext4_delete_inode(), and we allow that function to clean up the | |
1509 | * orphan info for us. | |
1510 | */ | |
1511 | if (inode->i_nlink) | |
1512 | ext4_orphan_del(handle, inode); | |
1513 | ||
1514 | ext4_journal_stop(handle); | |
1515 | trace_ext4_truncate_exit(inode); | |
1516 | } | |
1517 |