Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/ext3/inode.c | |
3 | * | |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
15 | * Goal-directed block allocation by Stephen Tweedie | |
e9ad5620 | 16 | * (sct@redhat.com), 1993, 1998 |
1da177e4 LT |
17 | * Big-endian to little-endian byte-swapping/bitmaps by |
18 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
19 | * 64-bit file support on 64-bit platforms by Jakub Jelinek | |
e9ad5620 | 20 | * (jj@sunsite.ms.mff.cuni.cz) |
1da177e4 LT |
21 | * |
22 | * Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000 | |
23 | */ | |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/time.h> | |
28 | #include <linux/ext3_jbd.h> | |
29 | #include <linux/jbd.h> | |
1da177e4 LT |
30 | #include <linux/highuid.h> |
31 | #include <linux/pagemap.h> | |
32 | #include <linux/quotaops.h> | |
33 | #include <linux/string.h> | |
34 | #include <linux/buffer_head.h> | |
35 | #include <linux/writeback.h> | |
36 | #include <linux/mpage.h> | |
37 | #include <linux/uio.h> | |
caa38fb0 | 38 | #include <linux/bio.h> |
68c9d702 | 39 | #include <linux/fiemap.h> |
b5ed3112 | 40 | #include <linux/namei.h> |
1da177e4 LT |
41 | #include "xattr.h" |
42 | #include "acl.h" | |
43 | ||
44 | static int ext3_writepage_trans_blocks(struct inode *inode); | |
45 | ||
46 | /* | |
47 | * Test whether an inode is a fast symlink. | |
48 | */ | |
d6859bfc | 49 | static int ext3_inode_is_fast_symlink(struct inode *inode) |
1da177e4 LT |
50 | { |
51 | int ea_blocks = EXT3_I(inode)->i_file_acl ? | |
52 | (inode->i_sb->s_blocksize >> 9) : 0; | |
53 | ||
d6859bfc | 54 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
1da177e4 LT |
55 | } |
56 | ||
d6859bfc AM |
57 | /* |
58 | * The ext3 forget function must perform a revoke if we are freeing data | |
1da177e4 | 59 | * which has been journaled. Metadata (eg. indirect blocks) must be |
ae6ddcc5 | 60 | * revoked in all cases. |
1da177e4 LT |
61 | * |
62 | * "bh" may be NULL: a metadata block may have been freed from memory | |
63 | * but there may still be a record of it in the journal, and that record | |
64 | * still needs to be revoked. | |
65 | */ | |
d6859bfc | 66 | int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode, |
1c2bf374 | 67 | struct buffer_head *bh, ext3_fsblk_t blocknr) |
1da177e4 LT |
68 | { |
69 | int err; | |
70 | ||
71 | might_sleep(); | |
72 | ||
73 | BUFFER_TRACE(bh, "enter"); | |
74 | ||
75 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " | |
76 | "data mode %lx\n", | |
77 | bh, is_metadata, inode->i_mode, | |
78 | test_opt(inode->i_sb, DATA_FLAGS)); | |
79 | ||
80 | /* Never use the revoke function if we are doing full data | |
81 | * journaling: there is no need to, and a V1 superblock won't | |
82 | * support it. Otherwise, only skip the revoke on un-journaled | |
83 | * data blocks. */ | |
84 | ||
85 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA || | |
86 | (!is_metadata && !ext3_should_journal_data(inode))) { | |
87 | if (bh) { | |
88 | BUFFER_TRACE(bh, "call journal_forget"); | |
89 | return ext3_journal_forget(handle, bh); | |
90 | } | |
91 | return 0; | |
92 | } | |
93 | ||
94 | /* | |
95 | * data!=journal && (is_metadata || should_journal_data(inode)) | |
96 | */ | |
97 | BUFFER_TRACE(bh, "call ext3_journal_revoke"); | |
98 | err = ext3_journal_revoke(handle, blocknr, bh); | |
99 | if (err) | |
e05b6b52 | 100 | ext3_abort(inode->i_sb, __func__, |
1da177e4 LT |
101 | "error %d when attempting revoke", err); |
102 | BUFFER_TRACE(bh, "exit"); | |
103 | return err; | |
104 | } | |
105 | ||
106 | /* | |
d6859bfc | 107 | * Work out how many blocks we need to proceed with the next chunk of a |
1da177e4 LT |
108 | * truncate transaction. |
109 | */ | |
ae6ddcc5 | 110 | static unsigned long blocks_for_truncate(struct inode *inode) |
1da177e4 LT |
111 | { |
112 | unsigned long needed; | |
113 | ||
114 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
115 | ||
116 | /* Give ourselves just enough room to cope with inodes in which | |
117 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
118 | * which resulted in random data in an inode which looked enough | |
119 | * like a regular file for ext3 to try to delete it. Things | |
120 | * will go a bit crazy if that happens, but at least we should | |
121 | * try not to panic the whole kernel. */ | |
122 | if (needed < 2) | |
123 | needed = 2; | |
124 | ||
125 | /* But we need to bound the transaction so we don't overflow the | |
126 | * journal. */ | |
ae6ddcc5 | 127 | if (needed > EXT3_MAX_TRANS_DATA) |
1da177e4 LT |
128 | needed = EXT3_MAX_TRANS_DATA; |
129 | ||
1f54587b | 130 | return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
1da177e4 LT |
131 | } |
132 | ||
ae6ddcc5 | 133 | /* |
1da177e4 LT |
134 | * Truncate transactions can be complex and absolutely huge. So we need to |
135 | * be able to restart the transaction at a conventient checkpoint to make | |
136 | * sure we don't overflow the journal. | |
137 | * | |
138 | * start_transaction gets us a new handle for a truncate transaction, | |
139 | * and extend_transaction tries to extend the existing one a bit. If | |
140 | * extend fails, we need to propagate the failure up and restart the | |
ae6ddcc5 | 141 | * transaction in the top-level truncate loop. --sct |
1da177e4 | 142 | */ |
ae6ddcc5 | 143 | static handle_t *start_transaction(struct inode *inode) |
1da177e4 LT |
144 | { |
145 | handle_t *result; | |
146 | ||
147 | result = ext3_journal_start(inode, blocks_for_truncate(inode)); | |
148 | if (!IS_ERR(result)) | |
149 | return result; | |
150 | ||
151 | ext3_std_error(inode->i_sb, PTR_ERR(result)); | |
152 | return result; | |
153 | } | |
154 | ||
155 | /* | |
156 | * Try to extend this transaction for the purposes of truncation. | |
157 | * | |
158 | * Returns 0 if we managed to create more room. If we can't create more | |
159 | * room, and the transaction must be restarted we return 1. | |
160 | */ | |
161 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
162 | { | |
163 | if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS) | |
164 | return 0; | |
165 | if (!ext3_journal_extend(handle, blocks_for_truncate(inode))) | |
166 | return 0; | |
167 | return 1; | |
168 | } | |
169 | ||
170 | /* | |
171 | * Restart the transaction associated with *handle. This does a commit, | |
172 | * so before we call here everything must be consistently dirtied against | |
173 | * this transaction. | |
174 | */ | |
00171d3c | 175 | static int truncate_restart_transaction(handle_t *handle, struct inode *inode) |
1da177e4 | 176 | { |
00171d3c JK |
177 | int ret; |
178 | ||
1da177e4 | 179 | jbd_debug(2, "restarting handle %p\n", handle); |
00171d3c JK |
180 | /* |
181 | * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle | |
182 | * At this moment, get_block can be called only for blocks inside | |
183 | * i_size since page cache has been already dropped and writes are | |
184 | * blocked by i_mutex. So we can safely drop the truncate_mutex. | |
185 | */ | |
186 | mutex_unlock(&EXT3_I(inode)->truncate_mutex); | |
187 | ret = ext3_journal_restart(handle, blocks_for_truncate(inode)); | |
188 | mutex_lock(&EXT3_I(inode)->truncate_mutex); | |
189 | return ret; | |
1da177e4 LT |
190 | } |
191 | ||
192 | /* | |
193 | * Called at the last iput() if i_nlink is zero. | |
194 | */ | |
195 | void ext3_delete_inode (struct inode * inode) | |
196 | { | |
197 | handle_t *handle; | |
198 | ||
fef26658 MF |
199 | truncate_inode_pages(&inode->i_data, 0); |
200 | ||
1da177e4 LT |
201 | if (is_bad_inode(inode)) |
202 | goto no_delete; | |
203 | ||
204 | handle = start_transaction(inode); | |
205 | if (IS_ERR(handle)) { | |
d6859bfc AM |
206 | /* |
207 | * If we're going to skip the normal cleanup, we still need to | |
208 | * make sure that the in-core orphan linked list is properly | |
209 | * cleaned up. | |
210 | */ | |
1da177e4 LT |
211 | ext3_orphan_del(NULL, inode); |
212 | goto no_delete; | |
213 | } | |
214 | ||
215 | if (IS_SYNC(inode)) | |
216 | handle->h_sync = 1; | |
217 | inode->i_size = 0; | |
218 | if (inode->i_blocks) | |
219 | ext3_truncate(inode); | |
220 | /* | |
221 | * Kill off the orphan record which ext3_truncate created. | |
222 | * AKPM: I think this can be inside the above `if'. | |
223 | * Note that ext3_orphan_del() has to be able to cope with the | |
224 | * deletion of a non-existent orphan - this is because we don't | |
225 | * know if ext3_truncate() actually created an orphan record. | |
226 | * (Well, we could do this if we need to, but heck - it works) | |
227 | */ | |
228 | ext3_orphan_del(handle, inode); | |
229 | EXT3_I(inode)->i_dtime = get_seconds(); | |
230 | ||
ae6ddcc5 | 231 | /* |
1da177e4 LT |
232 | * One subtle ordering requirement: if anything has gone wrong |
233 | * (transaction abort, IO errors, whatever), then we can still | |
234 | * do these next steps (the fs will already have been marked as | |
235 | * having errors), but we can't free the inode if the mark_dirty | |
ae6ddcc5 | 236 | * fails. |
1da177e4 LT |
237 | */ |
238 | if (ext3_mark_inode_dirty(handle, inode)) | |
239 | /* If that failed, just do the required in-core inode clear. */ | |
240 | clear_inode(inode); | |
241 | else | |
242 | ext3_free_inode(handle, inode); | |
243 | ext3_journal_stop(handle); | |
244 | return; | |
245 | no_delete: | |
246 | clear_inode(inode); /* We must guarantee clearing of inode... */ | |
247 | } | |
248 | ||
1da177e4 LT |
249 | typedef struct { |
250 | __le32 *p; | |
251 | __le32 key; | |
252 | struct buffer_head *bh; | |
253 | } Indirect; | |
254 | ||
255 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
256 | { | |
257 | p->key = *(p->p = v); | |
258 | p->bh = bh; | |
259 | } | |
260 | ||
d6859bfc | 261 | static int verify_chain(Indirect *from, Indirect *to) |
1da177e4 LT |
262 | { |
263 | while (from <= to && from->key == *from->p) | |
264 | from++; | |
265 | return (from > to); | |
266 | } | |
267 | ||
268 | /** | |
269 | * ext3_block_to_path - parse the block number into array of offsets | |
270 | * @inode: inode in question (we are only interested in its superblock) | |
271 | * @i_block: block number to be parsed | |
272 | * @offsets: array to store the offsets in | |
273 | * @boundary: set this non-zero if the referred-to block is likely to be | |
274 | * followed (on disk) by an indirect block. | |
275 | * | |
276 | * To store the locations of file's data ext3 uses a data structure common | |
277 | * for UNIX filesystems - tree of pointers anchored in the inode, with | |
278 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
279 | * This function translates the block number into path in that tree - | |
280 | * return value is the path length and @offsets[n] is the offset of | |
281 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
282 | * (negative or too large) warning is printed and zero returned. | |
283 | * | |
284 | * Note: function doesn't find node addresses, so no IO is needed. All | |
285 | * we need to know is the capacity of indirect blocks (taken from the | |
286 | * inode->i_sb). | |
287 | */ | |
288 | ||
289 | /* | |
290 | * Portability note: the last comparison (check that we fit into triple | |
291 | * indirect block) is spelled differently, because otherwise on an | |
292 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
293 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
294 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
295 | * i_block would have to be negative in the very beginning, so we would not | |
296 | * get there at all. | |
297 | */ | |
298 | ||
299 | static int ext3_block_to_path(struct inode *inode, | |
300 | long i_block, int offsets[4], int *boundary) | |
301 | { | |
302 | int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb); | |
303 | int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
304 | const long direct_blocks = EXT3_NDIR_BLOCKS, | |
305 | indirect_blocks = ptrs, | |
306 | double_blocks = (1 << (ptrs_bits * 2)); | |
307 | int n = 0; | |
308 | int final = 0; | |
309 | ||
310 | if (i_block < 0) { | |
311 | ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0"); | |
312 | } else if (i_block < direct_blocks) { | |
313 | offsets[n++] = i_block; | |
314 | final = direct_blocks; | |
315 | } else if ( (i_block -= direct_blocks) < indirect_blocks) { | |
316 | offsets[n++] = EXT3_IND_BLOCK; | |
317 | offsets[n++] = i_block; | |
318 | final = ptrs; | |
319 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
320 | offsets[n++] = EXT3_DIND_BLOCK; | |
321 | offsets[n++] = i_block >> ptrs_bits; | |
322 | offsets[n++] = i_block & (ptrs - 1); | |
323 | final = ptrs; | |
324 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
325 | offsets[n++] = EXT3_TIND_BLOCK; | |
326 | offsets[n++] = i_block >> (ptrs_bits * 2); | |
327 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
328 | offsets[n++] = i_block & (ptrs - 1); | |
329 | final = ptrs; | |
330 | } else { | |
d6859bfc | 331 | ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big"); |
1da177e4 LT |
332 | } |
333 | if (boundary) | |
89747d36 | 334 | *boundary = final - 1 - (i_block & (ptrs - 1)); |
1da177e4 LT |
335 | return n; |
336 | } | |
337 | ||
338 | /** | |
339 | * ext3_get_branch - read the chain of indirect blocks leading to data | |
340 | * @inode: inode in question | |
341 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
342 | * @offsets: offsets of pointers in inode/indirect blocks | |
343 | * @chain: place to store the result | |
344 | * @err: here we store the error value | |
345 | * | |
346 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
347 | * if everything went OK or the pointer to the last filled triple | |
348 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
349 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
350 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
351 | * number (it points into struct inode for i==0 and into the bh->b_data | |
352 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
353 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
354 | * numbers of the chain, addresses they were taken from (and where we can | |
355 | * verify that chain did not change) and buffer_heads hosting these | |
356 | * numbers. | |
357 | * | |
358 | * Function stops when it stumbles upon zero pointer (absent block) | |
359 | * (pointer to last triple returned, *@err == 0) | |
360 | * or when it gets an IO error reading an indirect block | |
361 | * (ditto, *@err == -EIO) | |
362 | * or when it notices that chain had been changed while it was reading | |
363 | * (ditto, *@err == -EAGAIN) | |
364 | * or when it reads all @depth-1 indirect blocks successfully and finds | |
365 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
366 | */ | |
367 | static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets, | |
368 | Indirect chain[4], int *err) | |
369 | { | |
370 | struct super_block *sb = inode->i_sb; | |
371 | Indirect *p = chain; | |
372 | struct buffer_head *bh; | |
373 | ||
374 | *err = 0; | |
375 | /* i_data is not going away, no lock needed */ | |
376 | add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets); | |
377 | if (!p->key) | |
378 | goto no_block; | |
379 | while (--depth) { | |
380 | bh = sb_bread(sb, le32_to_cpu(p->key)); | |
381 | if (!bh) | |
382 | goto failure; | |
383 | /* Reader: pointers */ | |
384 | if (!verify_chain(chain, p)) | |
385 | goto changed; | |
386 | add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); | |
387 | /* Reader: end */ | |
388 | if (!p->key) | |
389 | goto no_block; | |
390 | } | |
391 | return NULL; | |
392 | ||
393 | changed: | |
394 | brelse(bh); | |
395 | *err = -EAGAIN; | |
396 | goto no_block; | |
397 | failure: | |
398 | *err = -EIO; | |
399 | no_block: | |
400 | return p; | |
401 | } | |
402 | ||
403 | /** | |
404 | * ext3_find_near - find a place for allocation with sufficient locality | |
405 | * @inode: owner | |
406 | * @ind: descriptor of indirect block. | |
407 | * | |
1cc8dcf5 | 408 | * This function returns the preferred place for block allocation. |
1da177e4 LT |
409 | * It is used when heuristic for sequential allocation fails. |
410 | * Rules are: | |
411 | * + if there is a block to the left of our position - allocate near it. | |
412 | * + if pointer will live in indirect block - allocate near that block. | |
413 | * + if pointer will live in inode - allocate in the same | |
ae6ddcc5 | 414 | * cylinder group. |
1da177e4 LT |
415 | * |
416 | * In the latter case we colour the starting block by the callers PID to | |
417 | * prevent it from clashing with concurrent allocations for a different inode | |
418 | * in the same block group. The PID is used here so that functionally related | |
419 | * files will be close-by on-disk. | |
420 | * | |
421 | * Caller must make sure that @ind is valid and will stay that way. | |
422 | */ | |
43d23f90 | 423 | static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind) |
1da177e4 LT |
424 | { |
425 | struct ext3_inode_info *ei = EXT3_I(inode); | |
426 | __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data; | |
427 | __le32 *p; | |
43d23f90 MC |
428 | ext3_fsblk_t bg_start; |
429 | ext3_grpblk_t colour; | |
1da177e4 LT |
430 | |
431 | /* Try to find previous block */ | |
d6859bfc | 432 | for (p = ind->p - 1; p >= start; p--) { |
1da177e4 LT |
433 | if (*p) |
434 | return le32_to_cpu(*p); | |
d6859bfc | 435 | } |
1da177e4 LT |
436 | |
437 | /* No such thing, so let's try location of indirect block */ | |
438 | if (ind->bh) | |
439 | return ind->bh->b_blocknr; | |
440 | ||
441 | /* | |
d6859bfc AM |
442 | * It is going to be referred to from the inode itself? OK, just put it |
443 | * into the same cylinder group then. | |
1da177e4 | 444 | */ |
43d23f90 | 445 | bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group); |
1da177e4 LT |
446 | colour = (current->pid % 16) * |
447 | (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16); | |
448 | return bg_start + colour; | |
449 | } | |
450 | ||
451 | /** | |
1cc8dcf5 | 452 | * ext3_find_goal - find a preferred place for allocation. |
1da177e4 LT |
453 | * @inode: owner |
454 | * @block: block we want | |
1da177e4 | 455 | * @partial: pointer to the last triple within a chain |
1da177e4 | 456 | * |
1cc8dcf5 | 457 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 458 | * returns it. |
1da177e4 LT |
459 | */ |
460 | ||
43d23f90 | 461 | static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block, |
fb01bfda | 462 | Indirect *partial) |
1da177e4 | 463 | { |
d6859bfc AM |
464 | struct ext3_block_alloc_info *block_i; |
465 | ||
466 | block_i = EXT3_I(inode)->i_block_alloc_info; | |
1da177e4 LT |
467 | |
468 | /* | |
469 | * try the heuristic for sequential allocation, | |
470 | * failing that at least try to get decent locality. | |
471 | */ | |
472 | if (block_i && (block == block_i->last_alloc_logical_block + 1) | |
473 | && (block_i->last_alloc_physical_block != 0)) { | |
fe55c452 | 474 | return block_i->last_alloc_physical_block + 1; |
1da177e4 LT |
475 | } |
476 | ||
fe55c452 | 477 | return ext3_find_near(inode, partial); |
1da177e4 | 478 | } |
d6859bfc | 479 | |
b47b2478 MC |
480 | /** |
481 | * ext3_blks_to_allocate: Look up the block map and count the number | |
482 | * of direct blocks need to be allocated for the given branch. | |
483 | * | |
e9ad5620 | 484 | * @branch: chain of indirect blocks |
b47b2478 MC |
485 | * @k: number of blocks need for indirect blocks |
486 | * @blks: number of data blocks to be mapped. | |
487 | * @blocks_to_boundary: the offset in the indirect block | |
488 | * | |
489 | * return the total number of blocks to be allocate, including the | |
490 | * direct and indirect blocks. | |
491 | */ | |
d6859bfc | 492 | static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks, |
b47b2478 MC |
493 | int blocks_to_boundary) |
494 | { | |
495 | unsigned long count = 0; | |
496 | ||
497 | /* | |
498 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
499 | * then it's clear blocks on that path have not allocated | |
500 | */ | |
501 | if (k > 0) { | |
d6859bfc | 502 | /* right now we don't handle cross boundary allocation */ |
b47b2478 MC |
503 | if (blks < blocks_to_boundary + 1) |
504 | count += blks; | |
505 | else | |
506 | count += blocks_to_boundary + 1; | |
507 | return count; | |
508 | } | |
509 | ||
510 | count++; | |
511 | while (count < blks && count <= blocks_to_boundary && | |
512 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
513 | count++; | |
514 | } | |
515 | return count; | |
516 | } | |
517 | ||
518 | /** | |
519 | * ext3_alloc_blocks: multiple allocate blocks needed for a branch | |
520 | * @indirect_blks: the number of blocks need to allocate for indirect | |
521 | * blocks | |
522 | * | |
523 | * @new_blocks: on return it will store the new block numbers for | |
524 | * the indirect blocks(if needed) and the first direct block, | |
525 | * @blks: on return it will store the total number of allocated | |
526 | * direct blocks | |
527 | */ | |
528 | static int ext3_alloc_blocks(handle_t *handle, struct inode *inode, | |
43d23f90 MC |
529 | ext3_fsblk_t goal, int indirect_blks, int blks, |
530 | ext3_fsblk_t new_blocks[4], int *err) | |
b47b2478 MC |
531 | { |
532 | int target, i; | |
533 | unsigned long count = 0; | |
534 | int index = 0; | |
43d23f90 | 535 | ext3_fsblk_t current_block = 0; |
b47b2478 MC |
536 | int ret = 0; |
537 | ||
538 | /* | |
539 | * Here we try to allocate the requested multiple blocks at once, | |
540 | * on a best-effort basis. | |
541 | * To build a branch, we should allocate blocks for | |
542 | * the indirect blocks(if not allocated yet), and at least | |
543 | * the first direct block of this branch. That's the | |
544 | * minimum number of blocks need to allocate(required) | |
545 | */ | |
546 | target = blks + indirect_blks; | |
547 | ||
548 | while (1) { | |
549 | count = target; | |
550 | /* allocating blocks for indirect blocks and direct blocks */ | |
d6859bfc | 551 | current_block = ext3_new_blocks(handle,inode,goal,&count,err); |
b47b2478 MC |
552 | if (*err) |
553 | goto failed_out; | |
554 | ||
555 | target -= count; | |
556 | /* allocate blocks for indirect blocks */ | |
557 | while (index < indirect_blks && count) { | |
558 | new_blocks[index++] = current_block++; | |
559 | count--; | |
560 | } | |
561 | ||
562 | if (count > 0) | |
563 | break; | |
564 | } | |
565 | ||
566 | /* save the new block number for the first direct block */ | |
567 | new_blocks[index] = current_block; | |
568 | ||
569 | /* total number of blocks allocated for direct blocks */ | |
570 | ret = count; | |
571 | *err = 0; | |
572 | return ret; | |
573 | failed_out: | |
574 | for (i = 0; i <index; i++) | |
575 | ext3_free_blocks(handle, inode, new_blocks[i], 1); | |
576 | return ret; | |
577 | } | |
1da177e4 LT |
578 | |
579 | /** | |
580 | * ext3_alloc_branch - allocate and set up a chain of blocks. | |
581 | * @inode: owner | |
b47b2478 MC |
582 | * @indirect_blks: number of allocated indirect blocks |
583 | * @blks: number of allocated direct blocks | |
1da177e4 LT |
584 | * @offsets: offsets (in the blocks) to store the pointers to next. |
585 | * @branch: place to store the chain in. | |
586 | * | |
b47b2478 | 587 | * This function allocates blocks, zeroes out all but the last one, |
1da177e4 LT |
588 | * links them into chain and (if we are synchronous) writes them to disk. |
589 | * In other words, it prepares a branch that can be spliced onto the | |
590 | * inode. It stores the information about that chain in the branch[], in | |
591 | * the same format as ext3_get_branch() would do. We are calling it after | |
592 | * we had read the existing part of chain and partial points to the last | |
593 | * triple of that (one with zero ->key). Upon the exit we have the same | |
5b116879 | 594 | * picture as after the successful ext3_get_block(), except that in one |
1da177e4 LT |
595 | * place chain is disconnected - *branch->p is still zero (we did not |
596 | * set the last link), but branch->key contains the number that should | |
597 | * be placed into *branch->p to fill that gap. | |
598 | * | |
599 | * If allocation fails we free all blocks we've allocated (and forget | |
600 | * their buffer_heads) and return the error value the from failed | |
601 | * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain | |
602 | * as described above and return 0. | |
603 | */ | |
1da177e4 | 604 | static int ext3_alloc_branch(handle_t *handle, struct inode *inode, |
43d23f90 | 605 | int indirect_blks, int *blks, ext3_fsblk_t goal, |
b47b2478 | 606 | int *offsets, Indirect *branch) |
1da177e4 LT |
607 | { |
608 | int blocksize = inode->i_sb->s_blocksize; | |
b47b2478 | 609 | int i, n = 0; |
1da177e4 | 610 | int err = 0; |
b47b2478 MC |
611 | struct buffer_head *bh; |
612 | int num; | |
43d23f90 MC |
613 | ext3_fsblk_t new_blocks[4]; |
614 | ext3_fsblk_t current_block; | |
1da177e4 | 615 | |
b47b2478 MC |
616 | num = ext3_alloc_blocks(handle, inode, goal, indirect_blks, |
617 | *blks, new_blocks, &err); | |
618 | if (err) | |
619 | return err; | |
1da177e4 | 620 | |
b47b2478 MC |
621 | branch[0].key = cpu_to_le32(new_blocks[0]); |
622 | /* | |
623 | * metadata blocks and data blocks are allocated. | |
624 | */ | |
625 | for (n = 1; n <= indirect_blks; n++) { | |
626 | /* | |
627 | * Get buffer_head for parent block, zero it out | |
628 | * and set the pointer to new one, then send | |
629 | * parent to disk. | |
630 | */ | |
631 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
632 | branch[n].bh = bh; | |
633 | lock_buffer(bh); | |
634 | BUFFER_TRACE(bh, "call get_create_access"); | |
635 | err = ext3_journal_get_create_access(handle, bh); | |
636 | if (err) { | |
1da177e4 | 637 | unlock_buffer(bh); |
b47b2478 MC |
638 | brelse(bh); |
639 | goto failed; | |
640 | } | |
1da177e4 | 641 | |
b47b2478 MC |
642 | memset(bh->b_data, 0, blocksize); |
643 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
644 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
645 | *branch[n].p = branch[n].key; | |
646 | if ( n == indirect_blks) { | |
647 | current_block = new_blocks[n]; | |
648 | /* | |
649 | * End of chain, update the last new metablock of | |
650 | * the chain to point to the new allocated | |
651 | * data blocks numbers | |
652 | */ | |
653 | for (i=1; i < num; i++) | |
654 | *(branch[n].p + i) = cpu_to_le32(++current_block); | |
1da177e4 | 655 | } |
b47b2478 MC |
656 | BUFFER_TRACE(bh, "marking uptodate"); |
657 | set_buffer_uptodate(bh); | |
658 | unlock_buffer(bh); | |
1da177e4 | 659 | |
b47b2478 MC |
660 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
661 | err = ext3_journal_dirty_metadata(handle, bh); | |
662 | if (err) | |
663 | goto failed; | |
664 | } | |
665 | *blks = num; | |
666 | return err; | |
667 | failed: | |
1da177e4 | 668 | /* Allocation failed, free what we already allocated */ |
b47b2478 | 669 | for (i = 1; i <= n ; i++) { |
1da177e4 LT |
670 | BUFFER_TRACE(branch[i].bh, "call journal_forget"); |
671 | ext3_journal_forget(handle, branch[i].bh); | |
672 | } | |
b47b2478 MC |
673 | for (i = 0; i <indirect_blks; i++) |
674 | ext3_free_blocks(handle, inode, new_blocks[i], 1); | |
675 | ||
676 | ext3_free_blocks(handle, inode, new_blocks[i], num); | |
677 | ||
1da177e4 LT |
678 | return err; |
679 | } | |
680 | ||
681 | /** | |
d6859bfc AM |
682 | * ext3_splice_branch - splice the allocated branch onto inode. |
683 | * @inode: owner | |
684 | * @block: (logical) number of block we are adding | |
685 | * @chain: chain of indirect blocks (with a missing link - see | |
686 | * ext3_alloc_branch) | |
687 | * @where: location of missing link | |
688 | * @num: number of indirect blocks we are adding | |
689 | * @blks: number of direct blocks we are adding | |
690 | * | |
691 | * This function fills the missing link and does all housekeeping needed in | |
692 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
693 | * chain to new block and return 0. | |
1da177e4 | 694 | */ |
d6859bfc AM |
695 | static int ext3_splice_branch(handle_t *handle, struct inode *inode, |
696 | long block, Indirect *where, int num, int blks) | |
1da177e4 LT |
697 | { |
698 | int i; | |
699 | int err = 0; | |
d6859bfc | 700 | struct ext3_block_alloc_info *block_i; |
43d23f90 | 701 | ext3_fsblk_t current_block; |
fe8bc91c | 702 | struct ext3_inode_info *ei = EXT3_I(inode); |
d6859bfc | 703 | |
fe8bc91c | 704 | block_i = ei->i_block_alloc_info; |
1da177e4 LT |
705 | /* |
706 | * If we're splicing into a [td]indirect block (as opposed to the | |
707 | * inode) then we need to get write access to the [td]indirect block | |
708 | * before the splice. | |
709 | */ | |
710 | if (where->bh) { | |
711 | BUFFER_TRACE(where->bh, "get_write_access"); | |
712 | err = ext3_journal_get_write_access(handle, where->bh); | |
713 | if (err) | |
714 | goto err_out; | |
715 | } | |
1da177e4 LT |
716 | /* That's it */ |
717 | ||
718 | *where->p = where->key; | |
d6859bfc AM |
719 | |
720 | /* | |
721 | * Update the host buffer_head or inode to point to more just allocated | |
722 | * direct blocks blocks | |
723 | */ | |
b47b2478 | 724 | if (num == 0 && blks > 1) { |
5dea5176 | 725 | current_block = le32_to_cpu(where->key) + 1; |
b47b2478 MC |
726 | for (i = 1; i < blks; i++) |
727 | *(where->p + i ) = cpu_to_le32(current_block++); | |
728 | } | |
1da177e4 LT |
729 | |
730 | /* | |
731 | * update the most recently allocated logical & physical block | |
732 | * in i_block_alloc_info, to assist find the proper goal block for next | |
733 | * allocation | |
734 | */ | |
735 | if (block_i) { | |
b47b2478 | 736 | block_i->last_alloc_logical_block = block + blks - 1; |
d6859bfc | 737 | block_i->last_alloc_physical_block = |
5dea5176 | 738 | le32_to_cpu(where[num].key) + blks - 1; |
1da177e4 LT |
739 | } |
740 | ||
741 | /* We are done with atomic stuff, now do the rest of housekeeping */ | |
742 | ||
743 | inode->i_ctime = CURRENT_TIME_SEC; | |
744 | ext3_mark_inode_dirty(handle, inode); | |
fe8bc91c JK |
745 | /* ext3_mark_inode_dirty already updated i_sync_tid */ |
746 | atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid); | |
1da177e4 LT |
747 | |
748 | /* had we spliced it onto indirect block? */ | |
749 | if (where->bh) { | |
750 | /* | |
d6859bfc | 751 | * If we spliced it onto an indirect block, we haven't |
1da177e4 LT |
752 | * altered the inode. Note however that if it is being spliced |
753 | * onto an indirect block at the very end of the file (the | |
754 | * file is growing) then we *will* alter the inode to reflect | |
755 | * the new i_size. But that is not done here - it is done in | |
756 | * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode. | |
757 | */ | |
758 | jbd_debug(5, "splicing indirect only\n"); | |
759 | BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata"); | |
760 | err = ext3_journal_dirty_metadata(handle, where->bh); | |
ae6ddcc5 | 761 | if (err) |
1da177e4 LT |
762 | goto err_out; |
763 | } else { | |
764 | /* | |
765 | * OK, we spliced it into the inode itself on a direct block. | |
766 | * Inode was dirtied above. | |
767 | */ | |
768 | jbd_debug(5, "splicing direct\n"); | |
769 | } | |
770 | return err; | |
771 | ||
1da177e4 | 772 | err_out: |
b47b2478 | 773 | for (i = 1; i <= num; i++) { |
1da177e4 LT |
774 | BUFFER_TRACE(where[i].bh, "call journal_forget"); |
775 | ext3_journal_forget(handle, where[i].bh); | |
d6859bfc | 776 | ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1); |
1da177e4 | 777 | } |
b47b2478 MC |
778 | ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks); |
779 | ||
1da177e4 LT |
780 | return err; |
781 | } | |
782 | ||
783 | /* | |
784 | * Allocation strategy is simple: if we have to allocate something, we will | |
785 | * have to go the whole way to leaf. So let's do it before attaching anything | |
786 | * to tree, set linkage between the newborn blocks, write them if sync is | |
787 | * required, recheck the path, free and repeat if check fails, otherwise | |
788 | * set the last missing link (that will protect us from any truncate-generated | |
789 | * removals - all blocks on the path are immune now) and possibly force the | |
790 | * write on the parent block. | |
791 | * That has a nice additional property: no special recovery from the failed | |
792 | * allocations is needed - we simply release blocks and do not touch anything | |
793 | * reachable from inode. | |
794 | * | |
d6859bfc | 795 | * `handle' can be NULL if create == 0. |
1da177e4 LT |
796 | * |
797 | * The BKL may not be held on entry here. Be sure to take it early. | |
89747d36 MC |
798 | * return > 0, # of blocks mapped or allocated. |
799 | * return = 0, if plain lookup failed. | |
800 | * return < 0, error case. | |
1da177e4 | 801 | */ |
d6859bfc AM |
802 | int ext3_get_blocks_handle(handle_t *handle, struct inode *inode, |
803 | sector_t iblock, unsigned long maxblocks, | |
804 | struct buffer_head *bh_result, | |
43237b54 | 805 | int create) |
1da177e4 LT |
806 | { |
807 | int err = -EIO; | |
808 | int offsets[4]; | |
809 | Indirect chain[4]; | |
810 | Indirect *partial; | |
43d23f90 | 811 | ext3_fsblk_t goal; |
b47b2478 | 812 | int indirect_blks; |
89747d36 MC |
813 | int blocks_to_boundary = 0; |
814 | int depth; | |
1da177e4 | 815 | struct ext3_inode_info *ei = EXT3_I(inode); |
89747d36 | 816 | int count = 0; |
43d23f90 | 817 | ext3_fsblk_t first_block = 0; |
89747d36 | 818 | |
1da177e4 LT |
819 | |
820 | J_ASSERT(handle != NULL || create == 0); | |
d6859bfc | 821 | depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary); |
1da177e4 LT |
822 | |
823 | if (depth == 0) | |
824 | goto out; | |
825 | ||
1da177e4 LT |
826 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
827 | ||
828 | /* Simplest case - block found, no allocation needed */ | |
829 | if (!partial) { | |
5dea5176 | 830 | first_block = le32_to_cpu(chain[depth - 1].key); |
1da177e4 | 831 | clear_buffer_new(bh_result); |
89747d36 MC |
832 | count++; |
833 | /*map more blocks*/ | |
834 | while (count < maxblocks && count <= blocks_to_boundary) { | |
43d23f90 | 835 | ext3_fsblk_t blk; |
5dea5176 | 836 | |
e8ef7aae | 837 | if (!verify_chain(chain, chain + depth - 1)) { |
89747d36 MC |
838 | /* |
839 | * Indirect block might be removed by | |
840 | * truncate while we were reading it. | |
841 | * Handling of that case: forget what we've | |
842 | * got now. Flag the err as EAGAIN, so it | |
843 | * will reread. | |
844 | */ | |
845 | err = -EAGAIN; | |
846 | count = 0; | |
847 | break; | |
848 | } | |
5dea5176 MC |
849 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
850 | ||
851 | if (blk == first_block + count) | |
89747d36 MC |
852 | count++; |
853 | else | |
854 | break; | |
855 | } | |
856 | if (err != -EAGAIN) | |
857 | goto got_it; | |
1da177e4 LT |
858 | } |
859 | ||
860 | /* Next simple case - plain lookup or failed read of indirect block */ | |
fe55c452 MC |
861 | if (!create || err == -EIO) |
862 | goto cleanup; | |
863 | ||
97461518 | 864 | mutex_lock(&ei->truncate_mutex); |
fe55c452 MC |
865 | |
866 | /* | |
867 | * If the indirect block is missing while we are reading | |
868 | * the chain(ext3_get_branch() returns -EAGAIN err), or | |
869 | * if the chain has been changed after we grab the semaphore, | |
870 | * (either because another process truncated this branch, or | |
871 | * another get_block allocated this branch) re-grab the chain to see if | |
872 | * the request block has been allocated or not. | |
873 | * | |
874 | * Since we already block the truncate/other get_block | |
875 | * at this point, we will have the current copy of the chain when we | |
876 | * splice the branch into the tree. | |
877 | */ | |
878 | if (err == -EAGAIN || !verify_chain(chain, partial)) { | |
1da177e4 | 879 | while (partial > chain) { |
1da177e4 LT |
880 | brelse(partial->bh); |
881 | partial--; | |
882 | } | |
fe55c452 MC |
883 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
884 | if (!partial) { | |
89747d36 | 885 | count++; |
97461518 | 886 | mutex_unlock(&ei->truncate_mutex); |
fe55c452 MC |
887 | if (err) |
888 | goto cleanup; | |
889 | clear_buffer_new(bh_result); | |
890 | goto got_it; | |
891 | } | |
1da177e4 LT |
892 | } |
893 | ||
894 | /* | |
fe55c452 MC |
895 | * Okay, we need to do block allocation. Lazily initialize the block |
896 | * allocation info here if necessary | |
897 | */ | |
898 | if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) | |
1da177e4 | 899 | ext3_init_block_alloc_info(inode); |
1da177e4 | 900 | |
fb01bfda | 901 | goal = ext3_find_goal(inode, iblock, partial); |
1da177e4 | 902 | |
b47b2478 MC |
903 | /* the number of blocks need to allocate for [d,t]indirect blocks */ |
904 | indirect_blks = (chain + depth) - partial - 1; | |
1da177e4 | 905 | |
b47b2478 MC |
906 | /* |
907 | * Next look up the indirect map to count the totoal number of | |
908 | * direct blocks to allocate for this branch. | |
909 | */ | |
910 | count = ext3_blks_to_allocate(partial, indirect_blks, | |
911 | maxblocks, blocks_to_boundary); | |
1da177e4 LT |
912 | /* |
913 | * Block out ext3_truncate while we alter the tree | |
914 | */ | |
b47b2478 | 915 | err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal, |
fe55c452 | 916 | offsets + (partial - chain), partial); |
1da177e4 | 917 | |
fe55c452 MC |
918 | /* |
919 | * The ext3_splice_branch call will free and forget any buffers | |
1da177e4 LT |
920 | * on the new chain if there is a failure, but that risks using |
921 | * up transaction credits, especially for bitmaps where the | |
922 | * credits cannot be returned. Can we handle this somehow? We | |
fe55c452 MC |
923 | * may need to return -EAGAIN upwards in the worst case. --sct |
924 | */ | |
1da177e4 | 925 | if (!err) |
b47b2478 MC |
926 | err = ext3_splice_branch(handle, inode, iblock, |
927 | partial, indirect_blks, count); | |
97461518 | 928 | mutex_unlock(&ei->truncate_mutex); |
1da177e4 LT |
929 | if (err) |
930 | goto cleanup; | |
931 | ||
932 | set_buffer_new(bh_result); | |
fe55c452 MC |
933 | got_it: |
934 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); | |
20acaa18 | 935 | if (count > blocks_to_boundary) |
fe55c452 | 936 | set_buffer_boundary(bh_result); |
89747d36 | 937 | err = count; |
fe55c452 MC |
938 | /* Clean up and exit */ |
939 | partial = chain + depth - 1; /* the whole chain */ | |
940 | cleanup: | |
1da177e4 | 941 | while (partial > chain) { |
fe55c452 | 942 | BUFFER_TRACE(partial->bh, "call brelse"); |
1da177e4 LT |
943 | brelse(partial->bh); |
944 | partial--; | |
945 | } | |
fe55c452 MC |
946 | BUFFER_TRACE(bh_result, "returned"); |
947 | out: | |
948 | return err; | |
1da177e4 LT |
949 | } |
950 | ||
bd1939de JK |
951 | /* Maximum number of blocks we map for direct IO at once. */ |
952 | #define DIO_MAX_BLOCKS 4096 | |
953 | /* | |
954 | * Number of credits we need for writing DIO_MAX_BLOCKS: | |
955 | * We need sb + group descriptor + bitmap + inode -> 4 | |
956 | * For B blocks with A block pointers per block we need: | |
957 | * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect). | |
958 | * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25. | |
959 | */ | |
960 | #define DIO_CREDITS 25 | |
1da177e4 | 961 | |
f91a2ad2 BP |
962 | static int ext3_get_block(struct inode *inode, sector_t iblock, |
963 | struct buffer_head *bh_result, int create) | |
1da177e4 | 964 | { |
3e4fdaf8 | 965 | handle_t *handle = ext3_journal_current_handle(); |
bd1939de | 966 | int ret = 0, started = 0; |
1d8fa7a2 | 967 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
1da177e4 | 968 | |
bd1939de JK |
969 | if (create && !handle) { /* Direct IO write... */ |
970 | if (max_blocks > DIO_MAX_BLOCKS) | |
971 | max_blocks = DIO_MAX_BLOCKS; | |
972 | handle = ext3_journal_start(inode, DIO_CREDITS + | |
973 | 2 * EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb)); | |
974 | if (IS_ERR(handle)) { | |
1da177e4 | 975 | ret = PTR_ERR(handle); |
bd1939de | 976 | goto out; |
1da177e4 | 977 | } |
bd1939de | 978 | started = 1; |
1da177e4 LT |
979 | } |
980 | ||
bd1939de | 981 | ret = ext3_get_blocks_handle(handle, inode, iblock, |
43237b54 | 982 | max_blocks, bh_result, create); |
bd1939de JK |
983 | if (ret > 0) { |
984 | bh_result->b_size = (ret << inode->i_blkbits); | |
985 | ret = 0; | |
89747d36 | 986 | } |
bd1939de JK |
987 | if (started) |
988 | ext3_journal_stop(handle); | |
989 | out: | |
1da177e4 LT |
990 | return ret; |
991 | } | |
992 | ||
68c9d702 JB |
993 | int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
994 | u64 start, u64 len) | |
995 | { | |
996 | return generic_block_fiemap(inode, fieinfo, start, len, | |
997 | ext3_get_block); | |
998 | } | |
999 | ||
1da177e4 LT |
1000 | /* |
1001 | * `handle' can be NULL if create is zero | |
1002 | */ | |
d6859bfc AM |
1003 | struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode, |
1004 | long block, int create, int *errp) | |
1da177e4 LT |
1005 | { |
1006 | struct buffer_head dummy; | |
1007 | int fatal = 0, err; | |
1008 | ||
1009 | J_ASSERT(handle != NULL || create == 0); | |
1010 | ||
1011 | dummy.b_state = 0; | |
1012 | dummy.b_blocknr = -1000; | |
1013 | buffer_trace_init(&dummy.b_history); | |
89747d36 | 1014 | err = ext3_get_blocks_handle(handle, inode, block, 1, |
43237b54 | 1015 | &dummy, create); |
3665d0e5 BP |
1016 | /* |
1017 | * ext3_get_blocks_handle() returns number of blocks | |
1018 | * mapped. 0 in case of a HOLE. | |
1019 | */ | |
1020 | if (err > 0) { | |
1021 | if (err > 1) | |
1022 | WARN_ON(1); | |
89747d36 | 1023 | err = 0; |
89747d36 MC |
1024 | } |
1025 | *errp = err; | |
1026 | if (!err && buffer_mapped(&dummy)) { | |
1da177e4 LT |
1027 | struct buffer_head *bh; |
1028 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); | |
2973dfdb GOC |
1029 | if (!bh) { |
1030 | *errp = -EIO; | |
1031 | goto err; | |
1032 | } | |
1da177e4 LT |
1033 | if (buffer_new(&dummy)) { |
1034 | J_ASSERT(create != 0); | |
c80544dc | 1035 | J_ASSERT(handle != NULL); |
1da177e4 | 1036 | |
d6859bfc AM |
1037 | /* |
1038 | * Now that we do not always journal data, we should | |
1039 | * keep in mind whether this should always journal the | |
1040 | * new buffer as metadata. For now, regular file | |
1041 | * writes use ext3_get_block instead, so it's not a | |
1042 | * problem. | |
1043 | */ | |
1da177e4 LT |
1044 | lock_buffer(bh); |
1045 | BUFFER_TRACE(bh, "call get_create_access"); | |
1046 | fatal = ext3_journal_get_create_access(handle, bh); | |
1047 | if (!fatal && !buffer_uptodate(bh)) { | |
d6859bfc | 1048 | memset(bh->b_data,0,inode->i_sb->s_blocksize); |
1da177e4 LT |
1049 | set_buffer_uptodate(bh); |
1050 | } | |
1051 | unlock_buffer(bh); | |
1052 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
1053 | err = ext3_journal_dirty_metadata(handle, bh); | |
1054 | if (!fatal) | |
1055 | fatal = err; | |
1056 | } else { | |
1057 | BUFFER_TRACE(bh, "not a new buffer"); | |
1058 | } | |
1059 | if (fatal) { | |
1060 | *errp = fatal; | |
1061 | brelse(bh); | |
1062 | bh = NULL; | |
1063 | } | |
1064 | return bh; | |
1065 | } | |
2973dfdb | 1066 | err: |
1da177e4 LT |
1067 | return NULL; |
1068 | } | |
1069 | ||
d6859bfc | 1070 | struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode, |
1da177e4 LT |
1071 | int block, int create, int *err) |
1072 | { | |
1073 | struct buffer_head * bh; | |
1074 | ||
1075 | bh = ext3_getblk(handle, inode, block, create, err); | |
1076 | if (!bh) | |
1077 | return bh; | |
1078 | if (buffer_uptodate(bh)) | |
1079 | return bh; | |
caa38fb0 | 1080 | ll_rw_block(READ_META, 1, &bh); |
1da177e4 LT |
1081 | wait_on_buffer(bh); |
1082 | if (buffer_uptodate(bh)) | |
1083 | return bh; | |
1084 | put_bh(bh); | |
1085 | *err = -EIO; | |
1086 | return NULL; | |
1087 | } | |
1088 | ||
1089 | static int walk_page_buffers( handle_t *handle, | |
1090 | struct buffer_head *head, | |
1091 | unsigned from, | |
1092 | unsigned to, | |
1093 | int *partial, | |
1094 | int (*fn)( handle_t *handle, | |
1095 | struct buffer_head *bh)) | |
1096 | { | |
1097 | struct buffer_head *bh; | |
1098 | unsigned block_start, block_end; | |
1099 | unsigned blocksize = head->b_size; | |
1100 | int err, ret = 0; | |
1101 | struct buffer_head *next; | |
1102 | ||
1103 | for ( bh = head, block_start = 0; | |
1104 | ret == 0 && (bh != head || !block_start); | |
e9ad5620 | 1105 | block_start = block_end, bh = next) |
1da177e4 LT |
1106 | { |
1107 | next = bh->b_this_page; | |
1108 | block_end = block_start + blocksize; | |
1109 | if (block_end <= from || block_start >= to) { | |
1110 | if (partial && !buffer_uptodate(bh)) | |
1111 | *partial = 1; | |
1112 | continue; | |
1113 | } | |
1114 | err = (*fn)(handle, bh); | |
1115 | if (!ret) | |
1116 | ret = err; | |
1117 | } | |
1118 | return ret; | |
1119 | } | |
1120 | ||
1121 | /* | |
1122 | * To preserve ordering, it is essential that the hole instantiation and | |
1123 | * the data write be encapsulated in a single transaction. We cannot | |
1124 | * close off a transaction and start a new one between the ext3_get_block() | |
1125 | * and the commit_write(). So doing the journal_start at the start of | |
1126 | * prepare_write() is the right place. | |
1127 | * | |
1128 | * Also, this function can nest inside ext3_writepage() -> | |
1129 | * block_write_full_page(). In that case, we *know* that ext3_writepage() | |
1130 | * has generated enough buffer credits to do the whole page. So we won't | |
1131 | * block on the journal in that case, which is good, because the caller may | |
1132 | * be PF_MEMALLOC. | |
1133 | * | |
1134 | * By accident, ext3 can be reentered when a transaction is open via | |
1135 | * quota file writes. If we were to commit the transaction while thus | |
1136 | * reentered, there can be a deadlock - we would be holding a quota | |
1137 | * lock, and the commit would never complete if another thread had a | |
1138 | * transaction open and was blocking on the quota lock - a ranking | |
1139 | * violation. | |
1140 | * | |
1141 | * So what we do is to rely on the fact that journal_stop/journal_start | |
1142 | * will _not_ run commit under these circumstances because handle->h_ref | |
1143 | * is elevated. We'll still have enough credits for the tiny quotafile | |
ae6ddcc5 | 1144 | * write. |
1da177e4 | 1145 | */ |
d6859bfc AM |
1146 | static int do_journal_get_write_access(handle_t *handle, |
1147 | struct buffer_head *bh) | |
1da177e4 LT |
1148 | { |
1149 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1150 | return 0; | |
1151 | return ext3_journal_get_write_access(handle, bh); | |
1152 | } | |
1153 | ||
f4fc66a8 NP |
1154 | static int ext3_write_begin(struct file *file, struct address_space *mapping, |
1155 | loff_t pos, unsigned len, unsigned flags, | |
1156 | struct page **pagep, void **fsdata) | |
1da177e4 | 1157 | { |
f4fc66a8 | 1158 | struct inode *inode = mapping->host; |
695f6ae0 | 1159 | int ret; |
1da177e4 LT |
1160 | handle_t *handle; |
1161 | int retries = 0; | |
f4fc66a8 NP |
1162 | struct page *page; |
1163 | pgoff_t index; | |
1164 | unsigned from, to; | |
695f6ae0 JK |
1165 | /* Reserve one block more for addition to orphan list in case |
1166 | * we allocate blocks but write fails for some reason */ | |
1167 | int needed_blocks = ext3_writepage_trans_blocks(inode) + 1; | |
f4fc66a8 NP |
1168 | |
1169 | index = pos >> PAGE_CACHE_SHIFT; | |
1170 | from = pos & (PAGE_CACHE_SIZE - 1); | |
1171 | to = from + len; | |
1da177e4 LT |
1172 | |
1173 | retry: | |
54566b2c | 1174 | page = grab_cache_page_write_begin(mapping, index, flags); |
f4fc66a8 NP |
1175 | if (!page) |
1176 | return -ENOMEM; | |
1177 | *pagep = page; | |
1178 | ||
1da177e4 | 1179 | handle = ext3_journal_start(inode, needed_blocks); |
1aa9b4b9 | 1180 | if (IS_ERR(handle)) { |
f4fc66a8 NP |
1181 | unlock_page(page); |
1182 | page_cache_release(page); | |
1aa9b4b9 AM |
1183 | ret = PTR_ERR(handle); |
1184 | goto out; | |
1185 | } | |
f4fc66a8 NP |
1186 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
1187 | ext3_get_block); | |
1da177e4 | 1188 | if (ret) |
f4fc66a8 | 1189 | goto write_begin_failed; |
1da177e4 LT |
1190 | |
1191 | if (ext3_should_journal_data(inode)) { | |
1192 | ret = walk_page_buffers(handle, page_buffers(page), | |
1193 | from, to, NULL, do_journal_get_write_access); | |
1194 | } | |
f4fc66a8 NP |
1195 | write_begin_failed: |
1196 | if (ret) { | |
5ec8b75e AK |
1197 | /* |
1198 | * block_write_begin may have instantiated a few blocks | |
1199 | * outside i_size. Trim these off again. Don't need | |
1200 | * i_size_read because we hold i_mutex. | |
695f6ae0 JK |
1201 | * |
1202 | * Add inode to orphan list in case we crash before truncate | |
9eaaa2d5 JK |
1203 | * finishes. Do this only if ext3_can_truncate() agrees so |
1204 | * that orphan processing code is happy. | |
5ec8b75e | 1205 | */ |
9eaaa2d5 | 1206 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 JK |
1207 | ext3_orphan_add(handle, inode); |
1208 | ext3_journal_stop(handle); | |
1209 | unlock_page(page); | |
1210 | page_cache_release(page); | |
5ec8b75e | 1211 | if (pos + len > inode->i_size) |
9eaaa2d5 | 1212 | ext3_truncate(inode); |
f4fc66a8 | 1213 | } |
1da177e4 LT |
1214 | if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries)) |
1215 | goto retry; | |
1aa9b4b9 | 1216 | out: |
1da177e4 LT |
1217 | return ret; |
1218 | } | |
1219 | ||
f4fc66a8 | 1220 | |
d6859bfc | 1221 | int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh) |
1da177e4 LT |
1222 | { |
1223 | int err = journal_dirty_data(handle, bh); | |
1224 | if (err) | |
e05b6b52 | 1225 | ext3_journal_abort_handle(__func__, __func__, |
f4fc66a8 | 1226 | bh, handle, err); |
1da177e4 LT |
1227 | return err; |
1228 | } | |
1229 | ||
695f6ae0 JK |
1230 | /* For ordered writepage and write_end functions */ |
1231 | static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh) | |
1232 | { | |
1233 | /* | |
1234 | * Write could have mapped the buffer but it didn't copy the data in | |
1235 | * yet. So avoid filing such buffer into a transaction. | |
1236 | */ | |
1237 | if (buffer_mapped(bh) && buffer_uptodate(bh)) | |
1238 | return ext3_journal_dirty_data(handle, bh); | |
1239 | return 0; | |
1240 | } | |
1241 | ||
f4fc66a8 NP |
1242 | /* For write_end() in data=journal mode */ |
1243 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
1da177e4 LT |
1244 | { |
1245 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1246 | return 0; | |
1247 | set_buffer_uptodate(bh); | |
1248 | return ext3_journal_dirty_metadata(handle, bh); | |
1249 | } | |
1250 | ||
f4fc66a8 | 1251 | /* |
695f6ae0 JK |
1252 | * This is nasty and subtle: ext3_write_begin() could have allocated blocks |
1253 | * for the whole page but later we failed to copy the data in. Update inode | |
1254 | * size according to what we managed to copy. The rest is going to be | |
1255 | * truncated in write_end function. | |
f4fc66a8 | 1256 | */ |
695f6ae0 | 1257 | static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied) |
f4fc66a8 | 1258 | { |
695f6ae0 JK |
1259 | /* What matters to us is i_disksize. We don't write i_size anywhere */ |
1260 | if (pos + copied > inode->i_size) | |
1261 | i_size_write(inode, pos + copied); | |
1262 | if (pos + copied > EXT3_I(inode)->i_disksize) { | |
1263 | EXT3_I(inode)->i_disksize = pos + copied; | |
f4fc66a8 NP |
1264 | mark_inode_dirty(inode); |
1265 | } | |
f4fc66a8 NP |
1266 | } |
1267 | ||
1da177e4 LT |
1268 | /* |
1269 | * We need to pick up the new inode size which generic_commit_write gave us | |
1270 | * `file' can be NULL - eg, when called from page_symlink(). | |
1271 | * | |
1272 | * ext3 never places buffers on inode->i_mapping->private_list. metadata | |
1273 | * buffers are managed internally. | |
1274 | */ | |
f4fc66a8 NP |
1275 | static int ext3_ordered_write_end(struct file *file, |
1276 | struct address_space *mapping, | |
1277 | loff_t pos, unsigned len, unsigned copied, | |
1278 | struct page *page, void *fsdata) | |
1da177e4 LT |
1279 | { |
1280 | handle_t *handle = ext3_journal_current_handle(); | |
f4fc66a8 NP |
1281 | struct inode *inode = file->f_mapping->host; |
1282 | unsigned from, to; | |
1da177e4 LT |
1283 | int ret = 0, ret2; |
1284 | ||
695f6ae0 | 1285 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
f4fc66a8 | 1286 | |
695f6ae0 JK |
1287 | from = pos & (PAGE_CACHE_SIZE - 1); |
1288 | to = from + copied; | |
1da177e4 | 1289 | ret = walk_page_buffers(handle, page_buffers(page), |
695f6ae0 | 1290 | from, to, NULL, journal_dirty_data_fn); |
1da177e4 | 1291 | |
695f6ae0 JK |
1292 | if (ret == 0) |
1293 | update_file_sizes(inode, pos, copied); | |
1294 | /* | |
1295 | * There may be allocated blocks outside of i_size because | |
1296 | * we failed to copy some data. Prepare for truncate. | |
1297 | */ | |
9eaaa2d5 | 1298 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 | 1299 | ext3_orphan_add(handle, inode); |
1da177e4 LT |
1300 | ret2 = ext3_journal_stop(handle); |
1301 | if (!ret) | |
1302 | ret = ret2; | |
f4fc66a8 NP |
1303 | unlock_page(page); |
1304 | page_cache_release(page); | |
1305 | ||
695f6ae0 | 1306 | if (pos + len > inode->i_size) |
9eaaa2d5 | 1307 | ext3_truncate(inode); |
f4fc66a8 | 1308 | return ret ? ret : copied; |
1da177e4 LT |
1309 | } |
1310 | ||
f4fc66a8 NP |
1311 | static int ext3_writeback_write_end(struct file *file, |
1312 | struct address_space *mapping, | |
1313 | loff_t pos, unsigned len, unsigned copied, | |
1314 | struct page *page, void *fsdata) | |
1da177e4 LT |
1315 | { |
1316 | handle_t *handle = ext3_journal_current_handle(); | |
f4fc66a8 | 1317 | struct inode *inode = file->f_mapping->host; |
695f6ae0 | 1318 | int ret; |
1da177e4 | 1319 | |
695f6ae0 JK |
1320 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
1321 | update_file_sizes(inode, pos, copied); | |
1322 | /* | |
1323 | * There may be allocated blocks outside of i_size because | |
1324 | * we failed to copy some data. Prepare for truncate. | |
1325 | */ | |
9eaaa2d5 | 1326 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 JK |
1327 | ext3_orphan_add(handle, inode); |
1328 | ret = ext3_journal_stop(handle); | |
f4fc66a8 NP |
1329 | unlock_page(page); |
1330 | page_cache_release(page); | |
1331 | ||
695f6ae0 | 1332 | if (pos + len > inode->i_size) |
9eaaa2d5 | 1333 | ext3_truncate(inode); |
f4fc66a8 | 1334 | return ret ? ret : copied; |
1da177e4 LT |
1335 | } |
1336 | ||
f4fc66a8 NP |
1337 | static int ext3_journalled_write_end(struct file *file, |
1338 | struct address_space *mapping, | |
1339 | loff_t pos, unsigned len, unsigned copied, | |
1340 | struct page *page, void *fsdata) | |
1da177e4 LT |
1341 | { |
1342 | handle_t *handle = ext3_journal_current_handle(); | |
f4fc66a8 | 1343 | struct inode *inode = mapping->host; |
1da177e4 LT |
1344 | int ret = 0, ret2; |
1345 | int partial = 0; | |
f4fc66a8 | 1346 | unsigned from, to; |
1da177e4 | 1347 | |
f4fc66a8 NP |
1348 | from = pos & (PAGE_CACHE_SIZE - 1); |
1349 | to = from + len; | |
1350 | ||
1351 | if (copied < len) { | |
1352 | if (!PageUptodate(page)) | |
1353 | copied = 0; | |
695f6ae0 JK |
1354 | page_zero_new_buffers(page, from + copied, to); |
1355 | to = from + copied; | |
f4fc66a8 | 1356 | } |
1da177e4 LT |
1357 | |
1358 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
f4fc66a8 | 1359 | to, &partial, write_end_fn); |
1da177e4 LT |
1360 | if (!partial) |
1361 | SetPageUptodate(page); | |
695f6ae0 JK |
1362 | |
1363 | if (pos + copied > inode->i_size) | |
1364 | i_size_write(inode, pos + copied); | |
1365 | /* | |
1366 | * There may be allocated blocks outside of i_size because | |
1367 | * we failed to copy some data. Prepare for truncate. | |
1368 | */ | |
9eaaa2d5 | 1369 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 | 1370 | ext3_orphan_add(handle, inode); |
1da177e4 LT |
1371 | EXT3_I(inode)->i_state |= EXT3_STATE_JDATA; |
1372 | if (inode->i_size > EXT3_I(inode)->i_disksize) { | |
1373 | EXT3_I(inode)->i_disksize = inode->i_size; | |
1374 | ret2 = ext3_mark_inode_dirty(handle, inode); | |
ae6ddcc5 | 1375 | if (!ret) |
1da177e4 LT |
1376 | ret = ret2; |
1377 | } | |
f4fc66a8 | 1378 | |
1da177e4 LT |
1379 | ret2 = ext3_journal_stop(handle); |
1380 | if (!ret) | |
1381 | ret = ret2; | |
f4fc66a8 NP |
1382 | unlock_page(page); |
1383 | page_cache_release(page); | |
1384 | ||
695f6ae0 | 1385 | if (pos + len > inode->i_size) |
9eaaa2d5 | 1386 | ext3_truncate(inode); |
f4fc66a8 | 1387 | return ret ? ret : copied; |
1da177e4 LT |
1388 | } |
1389 | ||
ae6ddcc5 | 1390 | /* |
1da177e4 LT |
1391 | * bmap() is special. It gets used by applications such as lilo and by |
1392 | * the swapper to find the on-disk block of a specific piece of data. | |
1393 | * | |
1394 | * Naturally, this is dangerous if the block concerned is still in the | |
1395 | * journal. If somebody makes a swapfile on an ext3 data-journaling | |
1396 | * filesystem and enables swap, then they may get a nasty shock when the | |
1397 | * data getting swapped to that swapfile suddenly gets overwritten by | |
1398 | * the original zero's written out previously to the journal and | |
ae6ddcc5 | 1399 | * awaiting writeback in the kernel's buffer cache. |
1da177e4 LT |
1400 | * |
1401 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
ae6ddcc5 | 1402 | * take extra steps to flush any blocks which might be in the cache. |
1da177e4 LT |
1403 | */ |
1404 | static sector_t ext3_bmap(struct address_space *mapping, sector_t block) | |
1405 | { | |
1406 | struct inode *inode = mapping->host; | |
1407 | journal_t *journal; | |
1408 | int err; | |
1409 | ||
1410 | if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) { | |
ae6ddcc5 | 1411 | /* |
1da177e4 LT |
1412 | * This is a REALLY heavyweight approach, but the use of |
1413 | * bmap on dirty files is expected to be extremely rare: | |
1414 | * only if we run lilo or swapon on a freshly made file | |
ae6ddcc5 | 1415 | * do we expect this to happen. |
1da177e4 LT |
1416 | * |
1417 | * (bmap requires CAP_SYS_RAWIO so this does not | |
1418 | * represent an unprivileged user DOS attack --- we'd be | |
1419 | * in trouble if mortal users could trigger this path at | |
ae6ddcc5 | 1420 | * will.) |
1da177e4 LT |
1421 | * |
1422 | * NB. EXT3_STATE_JDATA is not set on files other than | |
1423 | * regular files. If somebody wants to bmap a directory | |
1424 | * or symlink and gets confused because the buffer | |
1425 | * hasn't yet been flushed to disk, they deserve | |
1426 | * everything they get. | |
1427 | */ | |
1428 | ||
1429 | EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA; | |
1430 | journal = EXT3_JOURNAL(inode); | |
1431 | journal_lock_updates(journal); | |
1432 | err = journal_flush(journal); | |
1433 | journal_unlock_updates(journal); | |
1434 | ||
1435 | if (err) | |
1436 | return 0; | |
1437 | } | |
1438 | ||
1439 | return generic_block_bmap(mapping,block,ext3_get_block); | |
1440 | } | |
1441 | ||
1442 | static int bget_one(handle_t *handle, struct buffer_head *bh) | |
1443 | { | |
1444 | get_bh(bh); | |
1445 | return 0; | |
1446 | } | |
1447 | ||
1448 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1449 | { | |
1450 | put_bh(bh); | |
1451 | return 0; | |
1452 | } | |
1453 | ||
9e80d407 JK |
1454 | static int buffer_unmapped(handle_t *handle, struct buffer_head *bh) |
1455 | { | |
1456 | return !buffer_mapped(bh); | |
1457 | } | |
695f6ae0 | 1458 | |
1da177e4 LT |
1459 | /* |
1460 | * Note that we always start a transaction even if we're not journalling | |
1461 | * data. This is to preserve ordering: any hole instantiation within | |
1462 | * __block_write_full_page -> ext3_get_block() should be journalled | |
1463 | * along with the data so we don't crash and then get metadata which | |
1464 | * refers to old data. | |
1465 | * | |
1466 | * In all journalling modes block_write_full_page() will start the I/O. | |
1467 | * | |
1468 | * Problem: | |
1469 | * | |
1470 | * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1471 | * ext3_writepage() | |
1472 | * | |
1473 | * Similar for: | |
1474 | * | |
1475 | * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ... | |
1476 | * | |
1477 | * Same applies to ext3_get_block(). We will deadlock on various things like | |
97461518 | 1478 | * lock_journal and i_truncate_mutex. |
1da177e4 LT |
1479 | * |
1480 | * Setting PF_MEMALLOC here doesn't work - too many internal memory | |
1481 | * allocations fail. | |
1482 | * | |
1483 | * 16May01: If we're reentered then journal_current_handle() will be | |
1484 | * non-zero. We simply *return*. | |
1485 | * | |
1486 | * 1 July 2001: @@@ FIXME: | |
1487 | * In journalled data mode, a data buffer may be metadata against the | |
1488 | * current transaction. But the same file is part of a shared mapping | |
1489 | * and someone does a writepage() on it. | |
1490 | * | |
1491 | * We will move the buffer onto the async_data list, but *after* it has | |
1492 | * been dirtied. So there's a small window where we have dirty data on | |
1493 | * BJ_Metadata. | |
1494 | * | |
1495 | * Note that this only applies to the last partial page in the file. The | |
1496 | * bit which block_write_full_page() uses prepare/commit for. (That's | |
1497 | * broken code anyway: it's wrong for msync()). | |
1498 | * | |
1499 | * It's a rare case: affects the final partial page, for journalled data | |
1500 | * where the file is subject to bith write() and writepage() in the same | |
1501 | * transction. To fix it we'll need a custom block_write_full_page(). | |
1502 | * We'll probably need that anyway for journalling writepage() output. | |
1503 | * | |
1504 | * We don't honour synchronous mounts for writepage(). That would be | |
1505 | * disastrous. Any write() or metadata operation will sync the fs for | |
1506 | * us. | |
1507 | * | |
1508 | * AKPM2: if all the page's buffers are mapped to disk and !data=journal, | |
1509 | * we don't need to open a transaction here. | |
1510 | */ | |
1511 | static int ext3_ordered_writepage(struct page *page, | |
d6859bfc | 1512 | struct writeback_control *wbc) |
1da177e4 LT |
1513 | { |
1514 | struct inode *inode = page->mapping->host; | |
1515 | struct buffer_head *page_bufs; | |
1516 | handle_t *handle = NULL; | |
1517 | int ret = 0; | |
1518 | int err; | |
1519 | ||
1520 | J_ASSERT(PageLocked(page)); | |
1521 | ||
1522 | /* | |
1523 | * We give up here if we're reentered, because it might be for a | |
1524 | * different filesystem. | |
1525 | */ | |
1526 | if (ext3_journal_current_handle()) | |
1527 | goto out_fail; | |
1528 | ||
9e80d407 JK |
1529 | if (!page_has_buffers(page)) { |
1530 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
1531 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
430db323 JK |
1532 | page_bufs = page_buffers(page); |
1533 | } else { | |
1534 | page_bufs = page_buffers(page); | |
1535 | if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE, | |
1536 | NULL, buffer_unmapped)) { | |
1537 | /* Provide NULL get_block() to catch bugs if buffers | |
1538 | * weren't really mapped */ | |
1539 | return block_write_full_page(page, NULL, wbc); | |
1540 | } | |
9e80d407 | 1541 | } |
1da177e4 LT |
1542 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
1543 | ||
1544 | if (IS_ERR(handle)) { | |
1545 | ret = PTR_ERR(handle); | |
1546 | goto out_fail; | |
1547 | } | |
1548 | ||
1da177e4 LT |
1549 | walk_page_buffers(handle, page_bufs, 0, |
1550 | PAGE_CACHE_SIZE, NULL, bget_one); | |
1551 | ||
1552 | ret = block_write_full_page(page, ext3_get_block, wbc); | |
1553 | ||
1554 | /* | |
1555 | * The page can become unlocked at any point now, and | |
1556 | * truncate can then come in and change things. So we | |
1557 | * can't touch *page from now on. But *page_bufs is | |
1558 | * safe due to elevated refcount. | |
1559 | */ | |
1560 | ||
1561 | /* | |
ae6ddcc5 | 1562 | * And attach them to the current transaction. But only if |
1da177e4 LT |
1563 | * block_write_full_page() succeeded. Otherwise they are unmapped, |
1564 | * and generally junk. | |
1565 | */ | |
1566 | if (ret == 0) { | |
1567 | err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, | |
1568 | NULL, journal_dirty_data_fn); | |
1569 | if (!ret) | |
1570 | ret = err; | |
1571 | } | |
1572 | walk_page_buffers(handle, page_bufs, 0, | |
1573 | PAGE_CACHE_SIZE, NULL, bput_one); | |
1574 | err = ext3_journal_stop(handle); | |
1575 | if (!ret) | |
1576 | ret = err; | |
1577 | return ret; | |
1578 | ||
1579 | out_fail: | |
1580 | redirty_page_for_writepage(wbc, page); | |
1581 | unlock_page(page); | |
1582 | return ret; | |
1583 | } | |
1584 | ||
1da177e4 LT |
1585 | static int ext3_writeback_writepage(struct page *page, |
1586 | struct writeback_control *wbc) | |
1587 | { | |
1588 | struct inode *inode = page->mapping->host; | |
1589 | handle_t *handle = NULL; | |
1590 | int ret = 0; | |
1591 | int err; | |
1592 | ||
1593 | if (ext3_journal_current_handle()) | |
1594 | goto out_fail; | |
1595 | ||
430db323 JK |
1596 | if (page_has_buffers(page)) { |
1597 | if (!walk_page_buffers(NULL, page_buffers(page), 0, | |
1598 | PAGE_CACHE_SIZE, NULL, buffer_unmapped)) { | |
1599 | /* Provide NULL get_block() to catch bugs if buffers | |
1600 | * weren't really mapped */ | |
1601 | return block_write_full_page(page, NULL, wbc); | |
1602 | } | |
1603 | } | |
1604 | ||
1da177e4 LT |
1605 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
1606 | if (IS_ERR(handle)) { | |
1607 | ret = PTR_ERR(handle); | |
1608 | goto out_fail; | |
1609 | } | |
1610 | ||
0e31f51d | 1611 | if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode)) |
1da177e4 LT |
1612 | ret = nobh_writepage(page, ext3_get_block, wbc); |
1613 | else | |
1614 | ret = block_write_full_page(page, ext3_get_block, wbc); | |
1615 | ||
1616 | err = ext3_journal_stop(handle); | |
1617 | if (!ret) | |
1618 | ret = err; | |
1619 | return ret; | |
1620 | ||
1621 | out_fail: | |
1622 | redirty_page_for_writepage(wbc, page); | |
1623 | unlock_page(page); | |
1624 | return ret; | |
1625 | } | |
1626 | ||
1627 | static int ext3_journalled_writepage(struct page *page, | |
1628 | struct writeback_control *wbc) | |
1629 | { | |
1630 | struct inode *inode = page->mapping->host; | |
1631 | handle_t *handle = NULL; | |
1632 | int ret = 0; | |
1633 | int err; | |
1634 | ||
1635 | if (ext3_journal_current_handle()) | |
1636 | goto no_write; | |
1637 | ||
1638 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); | |
1639 | if (IS_ERR(handle)) { | |
1640 | ret = PTR_ERR(handle); | |
1641 | goto no_write; | |
1642 | } | |
1643 | ||
1644 | if (!page_has_buffers(page) || PageChecked(page)) { | |
1645 | /* | |
1646 | * It's mmapped pagecache. Add buffers and journal it. There | |
1647 | * doesn't seem much point in redirtying the page here. | |
1648 | */ | |
1649 | ClearPageChecked(page); | |
1650 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, | |
1651 | ext3_get_block); | |
ab4eb43c DL |
1652 | if (ret != 0) { |
1653 | ext3_journal_stop(handle); | |
1da177e4 | 1654 | goto out_unlock; |
ab4eb43c | 1655 | } |
1da177e4 LT |
1656 | ret = walk_page_buffers(handle, page_buffers(page), 0, |
1657 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); | |
1658 | ||
1659 | err = walk_page_buffers(handle, page_buffers(page), 0, | |
f4fc66a8 | 1660 | PAGE_CACHE_SIZE, NULL, write_end_fn); |
1da177e4 LT |
1661 | if (ret == 0) |
1662 | ret = err; | |
1663 | EXT3_I(inode)->i_state |= EXT3_STATE_JDATA; | |
1664 | unlock_page(page); | |
1665 | } else { | |
1666 | /* | |
1667 | * It may be a page full of checkpoint-mode buffers. We don't | |
1668 | * really know unless we go poke around in the buffer_heads. | |
1669 | * But block_write_full_page will do the right thing. | |
1670 | */ | |
1671 | ret = block_write_full_page(page, ext3_get_block, wbc); | |
1672 | } | |
1673 | err = ext3_journal_stop(handle); | |
1674 | if (!ret) | |
1675 | ret = err; | |
1676 | out: | |
1677 | return ret; | |
1678 | ||
1679 | no_write: | |
1680 | redirty_page_for_writepage(wbc, page); | |
1681 | out_unlock: | |
1682 | unlock_page(page); | |
1683 | goto out; | |
1684 | } | |
1685 | ||
1686 | static int ext3_readpage(struct file *file, struct page *page) | |
1687 | { | |
1688 | return mpage_readpage(page, ext3_get_block); | |
1689 | } | |
1690 | ||
1691 | static int | |
1692 | ext3_readpages(struct file *file, struct address_space *mapping, | |
1693 | struct list_head *pages, unsigned nr_pages) | |
1694 | { | |
1695 | return mpage_readpages(mapping, pages, nr_pages, ext3_get_block); | |
1696 | } | |
1697 | ||
2ff28e22 | 1698 | static void ext3_invalidatepage(struct page *page, unsigned long offset) |
1da177e4 LT |
1699 | { |
1700 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); | |
1701 | ||
1702 | /* | |
1703 | * If it's a full truncate we just forget about the pending dirtying | |
1704 | */ | |
1705 | if (offset == 0) | |
1706 | ClearPageChecked(page); | |
1707 | ||
2ff28e22 | 1708 | journal_invalidatepage(journal, page, offset); |
1da177e4 LT |
1709 | } |
1710 | ||
27496a8c | 1711 | static int ext3_releasepage(struct page *page, gfp_t wait) |
1da177e4 LT |
1712 | { |
1713 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); | |
1714 | ||
1715 | WARN_ON(PageChecked(page)); | |
1716 | if (!page_has_buffers(page)) | |
1717 | return 0; | |
1718 | return journal_try_to_free_buffers(journal, page, wait); | |
1719 | } | |
1720 | ||
1721 | /* | |
1722 | * If the O_DIRECT write will extend the file then add this inode to the | |
1723 | * orphan list. So recovery will truncate it back to the original size | |
1724 | * if the machine crashes during the write. | |
1725 | * | |
1726 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
bd1939de JK |
1727 | * crashes then stale disk data _may_ be exposed inside the file. But current |
1728 | * VFS code falls back into buffered path in that case so we are safe. | |
1da177e4 LT |
1729 | */ |
1730 | static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb, | |
1731 | const struct iovec *iov, loff_t offset, | |
1732 | unsigned long nr_segs) | |
1733 | { | |
1734 | struct file *file = iocb->ki_filp; | |
1735 | struct inode *inode = file->f_mapping->host; | |
1736 | struct ext3_inode_info *ei = EXT3_I(inode); | |
bd1939de | 1737 | handle_t *handle; |
1da177e4 LT |
1738 | ssize_t ret; |
1739 | int orphan = 0; | |
1740 | size_t count = iov_length(iov, nr_segs); | |
ea0174a7 | 1741 | int retries = 0; |
1da177e4 LT |
1742 | |
1743 | if (rw == WRITE) { | |
1744 | loff_t final_size = offset + count; | |
1745 | ||
1da177e4 | 1746 | if (final_size > inode->i_size) { |
bd1939de JK |
1747 | /* Credits for sb + inode write */ |
1748 | handle = ext3_journal_start(inode, 2); | |
1749 | if (IS_ERR(handle)) { | |
1750 | ret = PTR_ERR(handle); | |
1751 | goto out; | |
1752 | } | |
1da177e4 | 1753 | ret = ext3_orphan_add(handle, inode); |
bd1939de JK |
1754 | if (ret) { |
1755 | ext3_journal_stop(handle); | |
1756 | goto out; | |
1757 | } | |
1da177e4 LT |
1758 | orphan = 1; |
1759 | ei->i_disksize = inode->i_size; | |
bd1939de | 1760 | ext3_journal_stop(handle); |
1da177e4 LT |
1761 | } |
1762 | } | |
1763 | ||
ea0174a7 | 1764 | retry: |
ae6ddcc5 | 1765 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, |
1da177e4 | 1766 | offset, nr_segs, |
f91a2ad2 | 1767 | ext3_get_block, NULL); |
ea0174a7 ES |
1768 | if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries)) |
1769 | goto retry; | |
1da177e4 | 1770 | |
bd1939de | 1771 | if (orphan) { |
1da177e4 LT |
1772 | int err; |
1773 | ||
bd1939de JK |
1774 | /* Credits for sb + inode write */ |
1775 | handle = ext3_journal_start(inode, 2); | |
1776 | if (IS_ERR(handle)) { | |
1777 | /* This is really bad luck. We've written the data | |
1778 | * but cannot extend i_size. Bail out and pretend | |
1779 | * the write failed... */ | |
1780 | ret = PTR_ERR(handle); | |
1781 | goto out; | |
1782 | } | |
1783 | if (inode->i_nlink) | |
1da177e4 | 1784 | ext3_orphan_del(handle, inode); |
bd1939de | 1785 | if (ret > 0) { |
1da177e4 LT |
1786 | loff_t end = offset + ret; |
1787 | if (end > inode->i_size) { | |
1788 | ei->i_disksize = end; | |
1789 | i_size_write(inode, end); | |
1790 | /* | |
1791 | * We're going to return a positive `ret' | |
1792 | * here due to non-zero-length I/O, so there's | |
1793 | * no way of reporting error returns from | |
1794 | * ext3_mark_inode_dirty() to userspace. So | |
1795 | * ignore it. | |
1796 | */ | |
1797 | ext3_mark_inode_dirty(handle, inode); | |
1798 | } | |
1799 | } | |
1800 | err = ext3_journal_stop(handle); | |
1801 | if (ret == 0) | |
1802 | ret = err; | |
1803 | } | |
1804 | out: | |
1805 | return ret; | |
1806 | } | |
1807 | ||
1808 | /* | |
1809 | * Pages can be marked dirty completely asynchronously from ext3's journalling | |
1810 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do | |
1811 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
1812 | * not necessarily locked. | |
1813 | * | |
1814 | * We cannot just dirty the page and leave attached buffers clean, because the | |
1815 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
1816 | * or jbddirty because all the journalling code will explode. | |
1817 | * | |
1818 | * So what we do is to mark the page "pending dirty" and next time writepage | |
1819 | * is called, propagate that into the buffers appropriately. | |
1820 | */ | |
1821 | static int ext3_journalled_set_page_dirty(struct page *page) | |
1822 | { | |
1823 | SetPageChecked(page); | |
1824 | return __set_page_dirty_nobuffers(page); | |
1825 | } | |
1826 | ||
f5e54d6e | 1827 | static const struct address_space_operations ext3_ordered_aops = { |
8ab22b9a HH |
1828 | .readpage = ext3_readpage, |
1829 | .readpages = ext3_readpages, | |
1830 | .writepage = ext3_ordered_writepage, | |
1831 | .sync_page = block_sync_page, | |
1832 | .write_begin = ext3_write_begin, | |
1833 | .write_end = ext3_ordered_write_end, | |
1834 | .bmap = ext3_bmap, | |
1835 | .invalidatepage = ext3_invalidatepage, | |
1836 | .releasepage = ext3_releasepage, | |
1837 | .direct_IO = ext3_direct_IO, | |
1838 | .migratepage = buffer_migrate_page, | |
1839 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 1840 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
1841 | }; |
1842 | ||
f5e54d6e | 1843 | static const struct address_space_operations ext3_writeback_aops = { |
8ab22b9a HH |
1844 | .readpage = ext3_readpage, |
1845 | .readpages = ext3_readpages, | |
1846 | .writepage = ext3_writeback_writepage, | |
1847 | .sync_page = block_sync_page, | |
1848 | .write_begin = ext3_write_begin, | |
1849 | .write_end = ext3_writeback_write_end, | |
1850 | .bmap = ext3_bmap, | |
1851 | .invalidatepage = ext3_invalidatepage, | |
1852 | .releasepage = ext3_releasepage, | |
1853 | .direct_IO = ext3_direct_IO, | |
1854 | .migratepage = buffer_migrate_page, | |
1855 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 1856 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
1857 | }; |
1858 | ||
f5e54d6e | 1859 | static const struct address_space_operations ext3_journalled_aops = { |
8ab22b9a HH |
1860 | .readpage = ext3_readpage, |
1861 | .readpages = ext3_readpages, | |
1862 | .writepage = ext3_journalled_writepage, | |
1863 | .sync_page = block_sync_page, | |
1864 | .write_begin = ext3_write_begin, | |
1865 | .write_end = ext3_journalled_write_end, | |
1866 | .set_page_dirty = ext3_journalled_set_page_dirty, | |
1867 | .bmap = ext3_bmap, | |
1868 | .invalidatepage = ext3_invalidatepage, | |
1869 | .releasepage = ext3_releasepage, | |
1870 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 1871 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
1872 | }; |
1873 | ||
1874 | void ext3_set_aops(struct inode *inode) | |
1875 | { | |
1876 | if (ext3_should_order_data(inode)) | |
1877 | inode->i_mapping->a_ops = &ext3_ordered_aops; | |
1878 | else if (ext3_should_writeback_data(inode)) | |
1879 | inode->i_mapping->a_ops = &ext3_writeback_aops; | |
1880 | else | |
1881 | inode->i_mapping->a_ops = &ext3_journalled_aops; | |
1882 | } | |
1883 | ||
1884 | /* | |
1885 | * ext3_block_truncate_page() zeroes out a mapping from file offset `from' | |
1886 | * up to the end of the block which corresponds to `from'. | |
1887 | * This required during truncate. We need to physically zero the tail end | |
1888 | * of that block so it doesn't yield old data if the file is later grown. | |
1889 | */ | |
1890 | static int ext3_block_truncate_page(handle_t *handle, struct page *page, | |
1891 | struct address_space *mapping, loff_t from) | |
1892 | { | |
43d23f90 | 1893 | ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
1894 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
1895 | unsigned blocksize, iblock, length, pos; | |
1896 | struct inode *inode = mapping->host; | |
1897 | struct buffer_head *bh; | |
1898 | int err = 0; | |
1da177e4 LT |
1899 | |
1900 | blocksize = inode->i_sb->s_blocksize; | |
1901 | length = blocksize - (offset & (blocksize - 1)); | |
1902 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
1903 | ||
1904 | /* | |
1905 | * For "nobh" option, we can only work if we don't need to | |
1906 | * read-in the page - otherwise we create buffers to do the IO. | |
1907 | */ | |
cd6ef84e BP |
1908 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && |
1909 | ext3_should_writeback_data(inode) && PageUptodate(page)) { | |
eebd2aa3 | 1910 | zero_user(page, offset, length); |
cd6ef84e BP |
1911 | set_page_dirty(page); |
1912 | goto unlock; | |
1da177e4 LT |
1913 | } |
1914 | ||
1915 | if (!page_has_buffers(page)) | |
1916 | create_empty_buffers(page, blocksize, 0); | |
1917 | ||
1918 | /* Find the buffer that contains "offset" */ | |
1919 | bh = page_buffers(page); | |
1920 | pos = blocksize; | |
1921 | while (offset >= pos) { | |
1922 | bh = bh->b_this_page; | |
1923 | iblock++; | |
1924 | pos += blocksize; | |
1925 | } | |
1926 | ||
1927 | err = 0; | |
1928 | if (buffer_freed(bh)) { | |
1929 | BUFFER_TRACE(bh, "freed: skip"); | |
1930 | goto unlock; | |
1931 | } | |
1932 | ||
1933 | if (!buffer_mapped(bh)) { | |
1934 | BUFFER_TRACE(bh, "unmapped"); | |
1935 | ext3_get_block(inode, iblock, bh, 0); | |
1936 | /* unmapped? It's a hole - nothing to do */ | |
1937 | if (!buffer_mapped(bh)) { | |
1938 | BUFFER_TRACE(bh, "still unmapped"); | |
1939 | goto unlock; | |
1940 | } | |
1941 | } | |
1942 | ||
1943 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
1944 | if (PageUptodate(page)) | |
1945 | set_buffer_uptodate(bh); | |
1946 | ||
1947 | if (!buffer_uptodate(bh)) { | |
1948 | err = -EIO; | |
1949 | ll_rw_block(READ, 1, &bh); | |
1950 | wait_on_buffer(bh); | |
1951 | /* Uhhuh. Read error. Complain and punt. */ | |
1952 | if (!buffer_uptodate(bh)) | |
1953 | goto unlock; | |
1954 | } | |
1955 | ||
1956 | if (ext3_should_journal_data(inode)) { | |
1957 | BUFFER_TRACE(bh, "get write access"); | |
1958 | err = ext3_journal_get_write_access(handle, bh); | |
1959 | if (err) | |
1960 | goto unlock; | |
1961 | } | |
1962 | ||
eebd2aa3 | 1963 | zero_user(page, offset, length); |
1da177e4 LT |
1964 | BUFFER_TRACE(bh, "zeroed end of block"); |
1965 | ||
1966 | err = 0; | |
1967 | if (ext3_should_journal_data(inode)) { | |
1968 | err = ext3_journal_dirty_metadata(handle, bh); | |
1969 | } else { | |
1970 | if (ext3_should_order_data(inode)) | |
1971 | err = ext3_journal_dirty_data(handle, bh); | |
1972 | mark_buffer_dirty(bh); | |
1973 | } | |
1974 | ||
1975 | unlock: | |
1976 | unlock_page(page); | |
1977 | page_cache_release(page); | |
1978 | return err; | |
1979 | } | |
1980 | ||
1981 | /* | |
1982 | * Probably it should be a library function... search for first non-zero word | |
1983 | * or memcmp with zero_page, whatever is better for particular architecture. | |
1984 | * Linus? | |
1985 | */ | |
1986 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
1987 | { | |
1988 | while (p < q) | |
1989 | if (*p++) | |
1990 | return 0; | |
1991 | return 1; | |
1992 | } | |
1993 | ||
1994 | /** | |
1995 | * ext3_find_shared - find the indirect blocks for partial truncation. | |
1996 | * @inode: inode in question | |
1997 | * @depth: depth of the affected branch | |
1998 | * @offsets: offsets of pointers in that branch (see ext3_block_to_path) | |
1999 | * @chain: place to store the pointers to partial indirect blocks | |
2000 | * @top: place to the (detached) top of branch | |
2001 | * | |
2002 | * This is a helper function used by ext3_truncate(). | |
2003 | * | |
2004 | * When we do truncate() we may have to clean the ends of several | |
2005 | * indirect blocks but leave the blocks themselves alive. Block is | |
2006 | * partially truncated if some data below the new i_size is refered | |
2007 | * from it (and it is on the path to the first completely truncated | |
2008 | * data block, indeed). We have to free the top of that path along | |
2009 | * with everything to the right of the path. Since no allocation | |
2010 | * past the truncation point is possible until ext3_truncate() | |
2011 | * finishes, we may safely do the latter, but top of branch may | |
2012 | * require special attention - pageout below the truncation point | |
2013 | * might try to populate it. | |
2014 | * | |
2015 | * We atomically detach the top of branch from the tree, store the | |
2016 | * block number of its root in *@top, pointers to buffer_heads of | |
2017 | * partially truncated blocks - in @chain[].bh and pointers to | |
2018 | * their last elements that should not be removed - in | |
2019 | * @chain[].p. Return value is the pointer to last filled element | |
2020 | * of @chain. | |
2021 | * | |
2022 | * The work left to caller to do the actual freeing of subtrees: | |
2023 | * a) free the subtree starting from *@top | |
2024 | * b) free the subtrees whose roots are stored in | |
2025 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
2026 | * c) free the subtrees growing from the inode past the @chain[0]. | |
2027 | * (no partially truncated stuff there). */ | |
2028 | ||
d6859bfc AM |
2029 | static Indirect *ext3_find_shared(struct inode *inode, int depth, |
2030 | int offsets[4], Indirect chain[4], __le32 *top) | |
1da177e4 LT |
2031 | { |
2032 | Indirect *partial, *p; | |
2033 | int k, err; | |
2034 | ||
2035 | *top = 0; | |
bf48aabb | 2036 | /* Make k index the deepest non-null offset + 1 */ |
1da177e4 LT |
2037 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
2038 | ; | |
2039 | partial = ext3_get_branch(inode, k, offsets, chain, &err); | |
2040 | /* Writer: pointers */ | |
2041 | if (!partial) | |
2042 | partial = chain + k-1; | |
2043 | /* | |
2044 | * If the branch acquired continuation since we've looked at it - | |
2045 | * fine, it should all survive and (new) top doesn't belong to us. | |
2046 | */ | |
2047 | if (!partial->key && *partial->p) | |
2048 | /* Writer: end */ | |
2049 | goto no_top; | |
2050 | for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) | |
2051 | ; | |
2052 | /* | |
2053 | * OK, we've found the last block that must survive. The rest of our | |
2054 | * branch should be detached before unlocking. However, if that rest | |
2055 | * of branch is all ours and does not grow immediately from the inode | |
2056 | * it's easier to cheat and just decrement partial->p. | |
2057 | */ | |
2058 | if (p == chain + k - 1 && p > chain) { | |
2059 | p->p--; | |
2060 | } else { | |
2061 | *top = *p->p; | |
2062 | /* Nope, don't do this in ext3. Must leave the tree intact */ | |
2063 | #if 0 | |
2064 | *p->p = 0; | |
2065 | #endif | |
2066 | } | |
2067 | /* Writer: end */ | |
2068 | ||
d6859bfc | 2069 | while(partial > p) { |
1da177e4 LT |
2070 | brelse(partial->bh); |
2071 | partial--; | |
2072 | } | |
2073 | no_top: | |
2074 | return partial; | |
2075 | } | |
2076 | ||
2077 | /* | |
2078 | * Zero a number of block pointers in either an inode or an indirect block. | |
2079 | * If we restart the transaction we must again get write access to the | |
2080 | * indirect block for further modification. | |
2081 | * | |
2082 | * We release `count' blocks on disk, but (last - first) may be greater | |
2083 | * than `count' because there can be holes in there. | |
2084 | */ | |
d6859bfc | 2085 | static void ext3_clear_blocks(handle_t *handle, struct inode *inode, |
43d23f90 | 2086 | struct buffer_head *bh, ext3_fsblk_t block_to_free, |
d6859bfc | 2087 | unsigned long count, __le32 *first, __le32 *last) |
1da177e4 LT |
2088 | { |
2089 | __le32 *p; | |
2090 | if (try_to_extend_transaction(handle, inode)) { | |
2091 | if (bh) { | |
2092 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
2093 | ext3_journal_dirty_metadata(handle, bh); | |
2094 | } | |
2095 | ext3_mark_inode_dirty(handle, inode); | |
00171d3c | 2096 | truncate_restart_transaction(handle, inode); |
1da177e4 LT |
2097 | if (bh) { |
2098 | BUFFER_TRACE(bh, "retaking write access"); | |
2099 | ext3_journal_get_write_access(handle, bh); | |
2100 | } | |
2101 | } | |
2102 | ||
2103 | /* | |
2104 | * Any buffers which are on the journal will be in memory. We find | |
2105 | * them on the hash table so journal_revoke() will run journal_forget() | |
2106 | * on them. We've already detached each block from the file, so | |
2107 | * bforget() in journal_forget() should be safe. | |
2108 | * | |
2109 | * AKPM: turn on bforget in journal_forget()!!! | |
2110 | */ | |
2111 | for (p = first; p < last; p++) { | |
2112 | u32 nr = le32_to_cpu(*p); | |
2113 | if (nr) { | |
2114 | struct buffer_head *bh; | |
2115 | ||
2116 | *p = 0; | |
2117 | bh = sb_find_get_block(inode->i_sb, nr); | |
2118 | ext3_forget(handle, 0, inode, bh, nr); | |
2119 | } | |
2120 | } | |
2121 | ||
2122 | ext3_free_blocks(handle, inode, block_to_free, count); | |
2123 | } | |
2124 | ||
2125 | /** | |
2126 | * ext3_free_data - free a list of data blocks | |
2127 | * @handle: handle for this transaction | |
2128 | * @inode: inode we are dealing with | |
2129 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
2130 | * @first: array of block numbers | |
2131 | * @last: points immediately past the end of array | |
2132 | * | |
2133 | * We are freeing all blocks refered from that array (numbers are stored as | |
2134 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
2135 | * | |
2136 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
2137 | * blocks are contiguous then releasing them at one time will only affect one | |
2138 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
2139 | * actually use a lot of journal space. | |
2140 | * | |
2141 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
2142 | * block pointers. | |
2143 | */ | |
2144 | static void ext3_free_data(handle_t *handle, struct inode *inode, | |
2145 | struct buffer_head *this_bh, | |
2146 | __le32 *first, __le32 *last) | |
2147 | { | |
43d23f90 | 2148 | ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ae6ddcc5 | 2149 | unsigned long count = 0; /* Number of blocks in the run */ |
1da177e4 LT |
2150 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind |
2151 | corresponding to | |
2152 | block_to_free */ | |
43d23f90 | 2153 | ext3_fsblk_t nr; /* Current block # */ |
1da177e4 LT |
2154 | __le32 *p; /* Pointer into inode/ind |
2155 | for current block */ | |
2156 | int err; | |
2157 | ||
2158 | if (this_bh) { /* For indirect block */ | |
2159 | BUFFER_TRACE(this_bh, "get_write_access"); | |
2160 | err = ext3_journal_get_write_access(handle, this_bh); | |
2161 | /* Important: if we can't update the indirect pointers | |
2162 | * to the blocks, we can't free them. */ | |
2163 | if (err) | |
2164 | return; | |
2165 | } | |
2166 | ||
2167 | for (p = first; p < last; p++) { | |
2168 | nr = le32_to_cpu(*p); | |
2169 | if (nr) { | |
2170 | /* accumulate blocks to free if they're contiguous */ | |
2171 | if (count == 0) { | |
2172 | block_to_free = nr; | |
2173 | block_to_free_p = p; | |
2174 | count = 1; | |
2175 | } else if (nr == block_to_free + count) { | |
2176 | count++; | |
2177 | } else { | |
ae6ddcc5 | 2178 | ext3_clear_blocks(handle, inode, this_bh, |
1da177e4 LT |
2179 | block_to_free, |
2180 | count, block_to_free_p, p); | |
2181 | block_to_free = nr; | |
2182 | block_to_free_p = p; | |
2183 | count = 1; | |
2184 | } | |
2185 | } | |
2186 | } | |
2187 | ||
2188 | if (count > 0) | |
2189 | ext3_clear_blocks(handle, inode, this_bh, block_to_free, | |
2190 | count, block_to_free_p, p); | |
2191 | ||
2192 | if (this_bh) { | |
2193 | BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata"); | |
3ccc3167 DG |
2194 | |
2195 | /* | |
2196 | * The buffer head should have an attached journal head at this | |
2197 | * point. However, if the data is corrupted and an indirect | |
2198 | * block pointed to itself, it would have been detached when | |
2199 | * the block was cleared. Check for this instead of OOPSing. | |
2200 | */ | |
2201 | if (bh2jh(this_bh)) | |
2202 | ext3_journal_dirty_metadata(handle, this_bh); | |
2203 | else | |
2204 | ext3_error(inode->i_sb, "ext3_free_data", | |
2205 | "circular indirect block detected, " | |
2206 | "inode=%lu, block=%llu", | |
2207 | inode->i_ino, | |
2208 | (unsigned long long)this_bh->b_blocknr); | |
1da177e4 LT |
2209 | } |
2210 | } | |
2211 | ||
2212 | /** | |
2213 | * ext3_free_branches - free an array of branches | |
2214 | * @handle: JBD handle for this transaction | |
2215 | * @inode: inode we are dealing with | |
2216 | * @parent_bh: the buffer_head which contains *@first and *@last | |
2217 | * @first: array of block numbers | |
2218 | * @last: pointer immediately past the end of array | |
2219 | * @depth: depth of the branches to free | |
2220 | * | |
2221 | * We are freeing all blocks refered from these branches (numbers are | |
2222 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
2223 | * appropriately. | |
2224 | */ | |
2225 | static void ext3_free_branches(handle_t *handle, struct inode *inode, | |
2226 | struct buffer_head *parent_bh, | |
2227 | __le32 *first, __le32 *last, int depth) | |
2228 | { | |
43d23f90 | 2229 | ext3_fsblk_t nr; |
1da177e4 LT |
2230 | __le32 *p; |
2231 | ||
2232 | if (is_handle_aborted(handle)) | |
2233 | return; | |
2234 | ||
2235 | if (depth--) { | |
2236 | struct buffer_head *bh; | |
2237 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); | |
2238 | p = last; | |
2239 | while (--p >= first) { | |
2240 | nr = le32_to_cpu(*p); | |
2241 | if (!nr) | |
2242 | continue; /* A hole */ | |
2243 | ||
2244 | /* Go read the buffer for the next level down */ | |
2245 | bh = sb_bread(inode->i_sb, nr); | |
2246 | ||
2247 | /* | |
2248 | * A read failure? Report error and clear slot | |
2249 | * (should be rare). | |
2250 | */ | |
2251 | if (!bh) { | |
2252 | ext3_error(inode->i_sb, "ext3_free_branches", | |
eee194e7 | 2253 | "Read failure, inode=%lu, block="E3FSBLK, |
1da177e4 LT |
2254 | inode->i_ino, nr); |
2255 | continue; | |
2256 | } | |
2257 | ||
2258 | /* This zaps the entire block. Bottom up. */ | |
2259 | BUFFER_TRACE(bh, "free child branches"); | |
2260 | ext3_free_branches(handle, inode, bh, | |
2261 | (__le32*)bh->b_data, | |
2262 | (__le32*)bh->b_data + addr_per_block, | |
2263 | depth); | |
2264 | ||
2265 | /* | |
2266 | * We've probably journalled the indirect block several | |
2267 | * times during the truncate. But it's no longer | |
2268 | * needed and we now drop it from the transaction via | |
2269 | * journal_revoke(). | |
2270 | * | |
2271 | * That's easy if it's exclusively part of this | |
2272 | * transaction. But if it's part of the committing | |
2273 | * transaction then journal_forget() will simply | |
2274 | * brelse() it. That means that if the underlying | |
2275 | * block is reallocated in ext3_get_block(), | |
2276 | * unmap_underlying_metadata() will find this block | |
2277 | * and will try to get rid of it. damn, damn. | |
2278 | * | |
2279 | * If this block has already been committed to the | |
2280 | * journal, a revoke record will be written. And | |
2281 | * revoke records must be emitted *before* clearing | |
2282 | * this block's bit in the bitmaps. | |
2283 | */ | |
2284 | ext3_forget(handle, 1, inode, bh, bh->b_blocknr); | |
2285 | ||
2286 | /* | |
2287 | * Everything below this this pointer has been | |
2288 | * released. Now let this top-of-subtree go. | |
2289 | * | |
2290 | * We want the freeing of this indirect block to be | |
2291 | * atomic in the journal with the updating of the | |
2292 | * bitmap block which owns it. So make some room in | |
2293 | * the journal. | |
2294 | * | |
2295 | * We zero the parent pointer *after* freeing its | |
2296 | * pointee in the bitmaps, so if extend_transaction() | |
2297 | * for some reason fails to put the bitmap changes and | |
2298 | * the release into the same transaction, recovery | |
2299 | * will merely complain about releasing a free block, | |
2300 | * rather than leaking blocks. | |
2301 | */ | |
2302 | if (is_handle_aborted(handle)) | |
2303 | return; | |
2304 | if (try_to_extend_transaction(handle, inode)) { | |
2305 | ext3_mark_inode_dirty(handle, inode); | |
00171d3c | 2306 | truncate_restart_transaction(handle, inode); |
1da177e4 LT |
2307 | } |
2308 | ||
2309 | ext3_free_blocks(handle, inode, nr, 1); | |
2310 | ||
2311 | if (parent_bh) { | |
2312 | /* | |
2313 | * The block which we have just freed is | |
2314 | * pointed to by an indirect block: journal it | |
2315 | */ | |
2316 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
2317 | if (!ext3_journal_get_write_access(handle, | |
2318 | parent_bh)){ | |
2319 | *p = 0; | |
2320 | BUFFER_TRACE(parent_bh, | |
2321 | "call ext3_journal_dirty_metadata"); | |
ae6ddcc5 | 2322 | ext3_journal_dirty_metadata(handle, |
1da177e4 LT |
2323 | parent_bh); |
2324 | } | |
2325 | } | |
2326 | } | |
2327 | } else { | |
2328 | /* We have reached the bottom of the tree. */ | |
2329 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
2330 | ext3_free_data(handle, inode, parent_bh, first, last); | |
2331 | } | |
2332 | } | |
2333 | ||
ae76dd9a DG |
2334 | int ext3_can_truncate(struct inode *inode) |
2335 | { | |
2336 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
2337 | return 0; | |
2338 | if (S_ISREG(inode->i_mode)) | |
2339 | return 1; | |
2340 | if (S_ISDIR(inode->i_mode)) | |
2341 | return 1; | |
2342 | if (S_ISLNK(inode->i_mode)) | |
2343 | return !ext3_inode_is_fast_symlink(inode); | |
2344 | return 0; | |
2345 | } | |
2346 | ||
1da177e4 LT |
2347 | /* |
2348 | * ext3_truncate() | |
2349 | * | |
2350 | * We block out ext3_get_block() block instantiations across the entire | |
2351 | * transaction, and VFS/VM ensures that ext3_truncate() cannot run | |
2352 | * simultaneously on behalf of the same inode. | |
2353 | * | |
2354 | * As we work through the truncate and commmit bits of it to the journal there | |
2355 | * is one core, guiding principle: the file's tree must always be consistent on | |
2356 | * disk. We must be able to restart the truncate after a crash. | |
2357 | * | |
2358 | * The file's tree may be transiently inconsistent in memory (although it | |
2359 | * probably isn't), but whenever we close off and commit a journal transaction, | |
2360 | * the contents of (the filesystem + the journal) must be consistent and | |
2361 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
2362 | * left-to-right works OK too). | |
2363 | * | |
2364 | * Note that at recovery time, journal replay occurs *before* the restart of | |
2365 | * truncate against the orphan inode list. | |
2366 | * | |
2367 | * The committed inode has the new, desired i_size (which is the same as | |
2368 | * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see | |
2369 | * that this inode's truncate did not complete and it will again call | |
2370 | * ext3_truncate() to have another go. So there will be instantiated blocks | |
2371 | * to the right of the truncation point in a crashed ext3 filesystem. But | |
2372 | * that's fine - as long as they are linked from the inode, the post-crash | |
2373 | * ext3_truncate() run will find them and release them. | |
2374 | */ | |
d6859bfc | 2375 | void ext3_truncate(struct inode *inode) |
1da177e4 LT |
2376 | { |
2377 | handle_t *handle; | |
2378 | struct ext3_inode_info *ei = EXT3_I(inode); | |
2379 | __le32 *i_data = ei->i_data; | |
2380 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); | |
2381 | struct address_space *mapping = inode->i_mapping; | |
2382 | int offsets[4]; | |
2383 | Indirect chain[4]; | |
2384 | Indirect *partial; | |
2385 | __le32 nr = 0; | |
2386 | int n; | |
2387 | long last_block; | |
2388 | unsigned blocksize = inode->i_sb->s_blocksize; | |
2389 | struct page *page; | |
2390 | ||
ae76dd9a | 2391 | if (!ext3_can_truncate(inode)) |
ef43618a | 2392 | goto out_notrans; |
1da177e4 | 2393 | |
f7ab34ea TT |
2394 | if (inode->i_size == 0 && ext3_should_writeback_data(inode)) |
2395 | ei->i_state |= EXT3_STATE_FLUSH_ON_CLOSE; | |
2396 | ||
1da177e4 LT |
2397 | /* |
2398 | * We have to lock the EOF page here, because lock_page() nests | |
2399 | * outside journal_start(). | |
2400 | */ | |
2401 | if ((inode->i_size & (blocksize - 1)) == 0) { | |
2402 | /* Block boundary? Nothing to do */ | |
2403 | page = NULL; | |
2404 | } else { | |
2405 | page = grab_cache_page(mapping, | |
2406 | inode->i_size >> PAGE_CACHE_SHIFT); | |
2407 | if (!page) | |
ef43618a | 2408 | goto out_notrans; |
1da177e4 LT |
2409 | } |
2410 | ||
2411 | handle = start_transaction(inode); | |
2412 | if (IS_ERR(handle)) { | |
2413 | if (page) { | |
2414 | clear_highpage(page); | |
2415 | flush_dcache_page(page); | |
2416 | unlock_page(page); | |
2417 | page_cache_release(page); | |
2418 | } | |
ef43618a | 2419 | goto out_notrans; |
1da177e4 LT |
2420 | } |
2421 | ||
2422 | last_block = (inode->i_size + blocksize-1) | |
2423 | >> EXT3_BLOCK_SIZE_BITS(inode->i_sb); | |
2424 | ||
2425 | if (page) | |
2426 | ext3_block_truncate_page(handle, page, mapping, inode->i_size); | |
2427 | ||
2428 | n = ext3_block_to_path(inode, last_block, offsets, NULL); | |
2429 | if (n == 0) | |
2430 | goto out_stop; /* error */ | |
2431 | ||
2432 | /* | |
2433 | * OK. This truncate is going to happen. We add the inode to the | |
2434 | * orphan list, so that if this truncate spans multiple transactions, | |
2435 | * and we crash, we will resume the truncate when the filesystem | |
2436 | * recovers. It also marks the inode dirty, to catch the new size. | |
2437 | * | |
2438 | * Implication: the file must always be in a sane, consistent | |
2439 | * truncatable state while each transaction commits. | |
2440 | */ | |
2441 | if (ext3_orphan_add(handle, inode)) | |
2442 | goto out_stop; | |
2443 | ||
2444 | /* | |
2445 | * The orphan list entry will now protect us from any crash which | |
2446 | * occurs before the truncate completes, so it is now safe to propagate | |
2447 | * the new, shorter inode size (held for now in i_size) into the | |
2448 | * on-disk inode. We do this via i_disksize, which is the value which | |
2449 | * ext3 *really* writes onto the disk inode. | |
2450 | */ | |
2451 | ei->i_disksize = inode->i_size; | |
2452 | ||
2453 | /* | |
2454 | * From here we block out all ext3_get_block() callers who want to | |
2455 | * modify the block allocation tree. | |
2456 | */ | |
97461518 | 2457 | mutex_lock(&ei->truncate_mutex); |
1da177e4 LT |
2458 | |
2459 | if (n == 1) { /* direct blocks */ | |
2460 | ext3_free_data(handle, inode, NULL, i_data+offsets[0], | |
2461 | i_data + EXT3_NDIR_BLOCKS); | |
2462 | goto do_indirects; | |
2463 | } | |
2464 | ||
2465 | partial = ext3_find_shared(inode, n, offsets, chain, &nr); | |
2466 | /* Kill the top of shared branch (not detached) */ | |
2467 | if (nr) { | |
2468 | if (partial == chain) { | |
2469 | /* Shared branch grows from the inode */ | |
2470 | ext3_free_branches(handle, inode, NULL, | |
2471 | &nr, &nr+1, (chain+n-1) - partial); | |
2472 | *partial->p = 0; | |
2473 | /* | |
2474 | * We mark the inode dirty prior to restart, | |
2475 | * and prior to stop. No need for it here. | |
2476 | */ | |
2477 | } else { | |
2478 | /* Shared branch grows from an indirect block */ | |
2479 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
2480 | ext3_free_branches(handle, inode, partial->bh, | |
2481 | partial->p, | |
2482 | partial->p+1, (chain+n-1) - partial); | |
2483 | } | |
2484 | } | |
2485 | /* Clear the ends of indirect blocks on the shared branch */ | |
2486 | while (partial > chain) { | |
2487 | ext3_free_branches(handle, inode, partial->bh, partial->p + 1, | |
2488 | (__le32*)partial->bh->b_data+addr_per_block, | |
2489 | (chain+n-1) - partial); | |
2490 | BUFFER_TRACE(partial->bh, "call brelse"); | |
2491 | brelse (partial->bh); | |
2492 | partial--; | |
2493 | } | |
2494 | do_indirects: | |
2495 | /* Kill the remaining (whole) subtrees */ | |
2496 | switch (offsets[0]) { | |
d6859bfc AM |
2497 | default: |
2498 | nr = i_data[EXT3_IND_BLOCK]; | |
2499 | if (nr) { | |
2500 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1); | |
2501 | i_data[EXT3_IND_BLOCK] = 0; | |
2502 | } | |
2503 | case EXT3_IND_BLOCK: | |
2504 | nr = i_data[EXT3_DIND_BLOCK]; | |
2505 | if (nr) { | |
2506 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2); | |
2507 | i_data[EXT3_DIND_BLOCK] = 0; | |
2508 | } | |
2509 | case EXT3_DIND_BLOCK: | |
2510 | nr = i_data[EXT3_TIND_BLOCK]; | |
2511 | if (nr) { | |
2512 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3); | |
2513 | i_data[EXT3_TIND_BLOCK] = 0; | |
2514 | } | |
2515 | case EXT3_TIND_BLOCK: | |
2516 | ; | |
1da177e4 LT |
2517 | } |
2518 | ||
2519 | ext3_discard_reservation(inode); | |
2520 | ||
97461518 | 2521 | mutex_unlock(&ei->truncate_mutex); |
1da177e4 LT |
2522 | inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC; |
2523 | ext3_mark_inode_dirty(handle, inode); | |
2524 | ||
d6859bfc AM |
2525 | /* |
2526 | * In a multi-transaction truncate, we only make the final transaction | |
2527 | * synchronous | |
2528 | */ | |
1da177e4 LT |
2529 | if (IS_SYNC(inode)) |
2530 | handle->h_sync = 1; | |
2531 | out_stop: | |
2532 | /* | |
2533 | * If this was a simple ftruncate(), and the file will remain alive | |
2534 | * then we need to clear up the orphan record which we created above. | |
2535 | * However, if this was a real unlink then we were called by | |
2536 | * ext3_delete_inode(), and we allow that function to clean up the | |
2537 | * orphan info for us. | |
2538 | */ | |
2539 | if (inode->i_nlink) | |
2540 | ext3_orphan_del(handle, inode); | |
2541 | ||
2542 | ext3_journal_stop(handle); | |
ef43618a JK |
2543 | return; |
2544 | out_notrans: | |
2545 | /* | |
2546 | * Delete the inode from orphan list so that it doesn't stay there | |
2547 | * forever and trigger assertion on umount. | |
2548 | */ | |
2549 | if (inode->i_nlink) | |
2550 | ext3_orphan_del(NULL, inode); | |
1da177e4 LT |
2551 | } |
2552 | ||
43d23f90 | 2553 | static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb, |
1da177e4 LT |
2554 | unsigned long ino, struct ext3_iloc *iloc) |
2555 | { | |
e0e369a7 | 2556 | unsigned long block_group; |
43d23f90 MC |
2557 | unsigned long offset; |
2558 | ext3_fsblk_t block; | |
e0e369a7 | 2559 | struct ext3_group_desc *gdp; |
1da177e4 | 2560 | |
2ccb48eb NB |
2561 | if (!ext3_valid_inum(sb, ino)) { |
2562 | /* | |
2563 | * This error is already checked for in namei.c unless we are | |
2564 | * looking at an NFS filehandle, in which case no error | |
2565 | * report is needed | |
2566 | */ | |
1da177e4 LT |
2567 | return 0; |
2568 | } | |
2ccb48eb | 2569 | |
1da177e4 | 2570 | block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); |
e0e369a7 AM |
2571 | gdp = ext3_get_group_desc(sb, block_group, NULL); |
2572 | if (!gdp) | |
1da177e4 | 2573 | return 0; |
1da177e4 LT |
2574 | /* |
2575 | * Figure out the offset within the block group inode table | |
2576 | */ | |
2577 | offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) * | |
2578 | EXT3_INODE_SIZE(sb); | |
e0e369a7 | 2579 | block = le32_to_cpu(gdp->bg_inode_table) + |
1da177e4 LT |
2580 | (offset >> EXT3_BLOCK_SIZE_BITS(sb)); |
2581 | ||
2582 | iloc->block_group = block_group; | |
2583 | iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1); | |
2584 | return block; | |
2585 | } | |
2586 | ||
2587 | /* | |
2588 | * ext3_get_inode_loc returns with an extra refcount against the inode's | |
2589 | * underlying buffer_head on success. If 'in_mem' is true, we have all | |
2590 | * data in memory that is needed to recreate the on-disk version of this | |
2591 | * inode. | |
2592 | */ | |
2593 | static int __ext3_get_inode_loc(struct inode *inode, | |
2594 | struct ext3_iloc *iloc, int in_mem) | |
2595 | { | |
43d23f90 | 2596 | ext3_fsblk_t block; |
1da177e4 LT |
2597 | struct buffer_head *bh; |
2598 | ||
2599 | block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc); | |
2600 | if (!block) | |
2601 | return -EIO; | |
2602 | ||
2603 | bh = sb_getblk(inode->i_sb, block); | |
2604 | if (!bh) { | |
2605 | ext3_error (inode->i_sb, "ext3_get_inode_loc", | |
2606 | "unable to read inode block - " | |
43d23f90 MC |
2607 | "inode=%lu, block="E3FSBLK, |
2608 | inode->i_ino, block); | |
1da177e4 LT |
2609 | return -EIO; |
2610 | } | |
2611 | if (!buffer_uptodate(bh)) { | |
2612 | lock_buffer(bh); | |
95450f5a HK |
2613 | |
2614 | /* | |
2615 | * If the buffer has the write error flag, we have failed | |
2616 | * to write out another inode in the same block. In this | |
2617 | * case, we don't have to read the block because we may | |
2618 | * read the old inode data successfully. | |
2619 | */ | |
2620 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
2621 | set_buffer_uptodate(bh); | |
2622 | ||
1da177e4 LT |
2623 | if (buffer_uptodate(bh)) { |
2624 | /* someone brought it uptodate while we waited */ | |
2625 | unlock_buffer(bh); | |
2626 | goto has_buffer; | |
2627 | } | |
2628 | ||
2629 | /* | |
2630 | * If we have all information of the inode in memory and this | |
2631 | * is the only valid inode in the block, we need not read the | |
2632 | * block. | |
2633 | */ | |
2634 | if (in_mem) { | |
2635 | struct buffer_head *bitmap_bh; | |
2636 | struct ext3_group_desc *desc; | |
2637 | int inodes_per_buffer; | |
2638 | int inode_offset, i; | |
2639 | int block_group; | |
2640 | int start; | |
2641 | ||
2642 | block_group = (inode->i_ino - 1) / | |
2643 | EXT3_INODES_PER_GROUP(inode->i_sb); | |
2644 | inodes_per_buffer = bh->b_size / | |
2645 | EXT3_INODE_SIZE(inode->i_sb); | |
2646 | inode_offset = ((inode->i_ino - 1) % | |
2647 | EXT3_INODES_PER_GROUP(inode->i_sb)); | |
2648 | start = inode_offset & ~(inodes_per_buffer - 1); | |
2649 | ||
2650 | /* Is the inode bitmap in cache? */ | |
2651 | desc = ext3_get_group_desc(inode->i_sb, | |
2652 | block_group, NULL); | |
2653 | if (!desc) | |
2654 | goto make_io; | |
2655 | ||
2656 | bitmap_bh = sb_getblk(inode->i_sb, | |
2657 | le32_to_cpu(desc->bg_inode_bitmap)); | |
2658 | if (!bitmap_bh) | |
2659 | goto make_io; | |
2660 | ||
2661 | /* | |
2662 | * If the inode bitmap isn't in cache then the | |
2663 | * optimisation may end up performing two reads instead | |
2664 | * of one, so skip it. | |
2665 | */ | |
2666 | if (!buffer_uptodate(bitmap_bh)) { | |
2667 | brelse(bitmap_bh); | |
2668 | goto make_io; | |
2669 | } | |
2670 | for (i = start; i < start + inodes_per_buffer; i++) { | |
2671 | if (i == inode_offset) | |
2672 | continue; | |
2673 | if (ext3_test_bit(i, bitmap_bh->b_data)) | |
2674 | break; | |
2675 | } | |
2676 | brelse(bitmap_bh); | |
2677 | if (i == start + inodes_per_buffer) { | |
2678 | /* all other inodes are free, so skip I/O */ | |
2679 | memset(bh->b_data, 0, bh->b_size); | |
2680 | set_buffer_uptodate(bh); | |
2681 | unlock_buffer(bh); | |
2682 | goto has_buffer; | |
2683 | } | |
2684 | } | |
2685 | ||
2686 | make_io: | |
2687 | /* | |
2688 | * There are other valid inodes in the buffer, this inode | |
2689 | * has in-inode xattrs, or we don't have this inode in memory. | |
2690 | * Read the block from disk. | |
2691 | */ | |
2692 | get_bh(bh); | |
2693 | bh->b_end_io = end_buffer_read_sync; | |
caa38fb0 | 2694 | submit_bh(READ_META, bh); |
1da177e4 LT |
2695 | wait_on_buffer(bh); |
2696 | if (!buffer_uptodate(bh)) { | |
2697 | ext3_error(inode->i_sb, "ext3_get_inode_loc", | |
2698 | "unable to read inode block - " | |
43d23f90 | 2699 | "inode=%lu, block="E3FSBLK, |
1da177e4 LT |
2700 | inode->i_ino, block); |
2701 | brelse(bh); | |
2702 | return -EIO; | |
2703 | } | |
2704 | } | |
2705 | has_buffer: | |
2706 | iloc->bh = bh; | |
2707 | return 0; | |
2708 | } | |
2709 | ||
2710 | int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc) | |
2711 | { | |
2712 | /* We have all inode data except xattrs in memory here. */ | |
2713 | return __ext3_get_inode_loc(inode, iloc, | |
2714 | !(EXT3_I(inode)->i_state & EXT3_STATE_XATTR)); | |
2715 | } | |
2716 | ||
2717 | void ext3_set_inode_flags(struct inode *inode) | |
2718 | { | |
2719 | unsigned int flags = EXT3_I(inode)->i_flags; | |
2720 | ||
2721 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
2722 | if (flags & EXT3_SYNC_FL) | |
2723 | inode->i_flags |= S_SYNC; | |
2724 | if (flags & EXT3_APPEND_FL) | |
2725 | inode->i_flags |= S_APPEND; | |
2726 | if (flags & EXT3_IMMUTABLE_FL) | |
2727 | inode->i_flags |= S_IMMUTABLE; | |
2728 | if (flags & EXT3_NOATIME_FL) | |
2729 | inode->i_flags |= S_NOATIME; | |
2730 | if (flags & EXT3_DIRSYNC_FL) | |
2731 | inode->i_flags |= S_DIRSYNC; | |
2732 | } | |
2733 | ||
28be5abb JK |
2734 | /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */ |
2735 | void ext3_get_inode_flags(struct ext3_inode_info *ei) | |
2736 | { | |
2737 | unsigned int flags = ei->vfs_inode.i_flags; | |
2738 | ||
2739 | ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL| | |
2740 | EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL); | |
2741 | if (flags & S_SYNC) | |
2742 | ei->i_flags |= EXT3_SYNC_FL; | |
2743 | if (flags & S_APPEND) | |
2744 | ei->i_flags |= EXT3_APPEND_FL; | |
2745 | if (flags & S_IMMUTABLE) | |
2746 | ei->i_flags |= EXT3_IMMUTABLE_FL; | |
2747 | if (flags & S_NOATIME) | |
2748 | ei->i_flags |= EXT3_NOATIME_FL; | |
2749 | if (flags & S_DIRSYNC) | |
2750 | ei->i_flags |= EXT3_DIRSYNC_FL; | |
2751 | } | |
2752 | ||
473043dc | 2753 | struct inode *ext3_iget(struct super_block *sb, unsigned long ino) |
1da177e4 LT |
2754 | { |
2755 | struct ext3_iloc iloc; | |
2756 | struct ext3_inode *raw_inode; | |
473043dc | 2757 | struct ext3_inode_info *ei; |
1da177e4 | 2758 | struct buffer_head *bh; |
473043dc | 2759 | struct inode *inode; |
fe8bc91c JK |
2760 | journal_t *journal = EXT3_SB(sb)->s_journal; |
2761 | transaction_t *transaction; | |
473043dc | 2762 | long ret; |
1da177e4 LT |
2763 | int block; |
2764 | ||
473043dc DH |
2765 | inode = iget_locked(sb, ino); |
2766 | if (!inode) | |
2767 | return ERR_PTR(-ENOMEM); | |
2768 | if (!(inode->i_state & I_NEW)) | |
2769 | return inode; | |
2770 | ||
2771 | ei = EXT3_I(inode); | |
1da177e4 LT |
2772 | ei->i_block_alloc_info = NULL; |
2773 | ||
473043dc DH |
2774 | ret = __ext3_get_inode_loc(inode, &iloc, 0); |
2775 | if (ret < 0) | |
1da177e4 LT |
2776 | goto bad_inode; |
2777 | bh = iloc.bh; | |
2778 | raw_inode = ext3_raw_inode(&iloc); | |
2779 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); | |
2780 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
2781 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
2782 | if(!(test_opt (inode->i_sb, NO_UID32))) { | |
2783 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; | |
2784 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
2785 | } | |
2786 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
2787 | inode->i_size = le32_to_cpu(raw_inode->i_size); | |
4d7bf11d MR |
2788 | inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime); |
2789 | inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime); | |
2790 | inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime); | |
1da177e4 LT |
2791 | inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0; |
2792 | ||
2793 | ei->i_state = 0; | |
2794 | ei->i_dir_start_lookup = 0; | |
2795 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
2796 | /* We now have enough fields to check if the inode was active or not. | |
2797 | * This is needed because nfsd might try to access dead inodes | |
2798 | * the test is that same one that e2fsck uses | |
2799 | * NeilBrown 1999oct15 | |
2800 | */ | |
2801 | if (inode->i_nlink == 0) { | |
2802 | if (inode->i_mode == 0 || | |
2803 | !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) { | |
2804 | /* this inode is deleted */ | |
2805 | brelse (bh); | |
473043dc | 2806 | ret = -ESTALE; |
1da177e4 LT |
2807 | goto bad_inode; |
2808 | } | |
2809 | /* The only unlinked inodes we let through here have | |
2810 | * valid i_mode and are being read by the orphan | |
2811 | * recovery code: that's fine, we're about to complete | |
2812 | * the process of deleting those. */ | |
2813 | } | |
1da177e4 LT |
2814 | inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); |
2815 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); | |
2816 | #ifdef EXT3_FRAGMENTS | |
2817 | ei->i_faddr = le32_to_cpu(raw_inode->i_faddr); | |
2818 | ei->i_frag_no = raw_inode->i_frag; | |
2819 | ei->i_frag_size = raw_inode->i_fsize; | |
2820 | #endif | |
2821 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); | |
2822 | if (!S_ISREG(inode->i_mode)) { | |
2823 | ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); | |
2824 | } else { | |
2825 | inode->i_size |= | |
2826 | ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; | |
2827 | } | |
2828 | ei->i_disksize = inode->i_size; | |
2829 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); | |
2830 | ei->i_block_group = iloc.block_group; | |
2831 | /* | |
2832 | * NOTE! The in-memory inode i_data array is in little-endian order | |
2833 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
2834 | */ | |
2835 | for (block = 0; block < EXT3_N_BLOCKS; block++) | |
2836 | ei->i_data[block] = raw_inode->i_block[block]; | |
2837 | INIT_LIST_HEAD(&ei->i_orphan); | |
2838 | ||
fe8bc91c JK |
2839 | /* |
2840 | * Set transaction id's of transactions that have to be committed | |
2841 | * to finish f[data]sync. We set them to currently running transaction | |
2842 | * as we cannot be sure that the inode or some of its metadata isn't | |
2843 | * part of the transaction - the inode could have been reclaimed and | |
2844 | * now it is reread from disk. | |
2845 | */ | |
2846 | if (journal) { | |
2847 | tid_t tid; | |
2848 | ||
2849 | spin_lock(&journal->j_state_lock); | |
2850 | if (journal->j_running_transaction) | |
2851 | transaction = journal->j_running_transaction; | |
2852 | else | |
2853 | transaction = journal->j_committing_transaction; | |
2854 | if (transaction) | |
2855 | tid = transaction->t_tid; | |
2856 | else | |
2857 | tid = journal->j_commit_sequence; | |
2858 | spin_unlock(&journal->j_state_lock); | |
2859 | atomic_set(&ei->i_sync_tid, tid); | |
2860 | atomic_set(&ei->i_datasync_tid, tid); | |
2861 | } | |
2862 | ||
1da177e4 LT |
2863 | if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 && |
2864 | EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) { | |
2865 | /* | |
2866 | * When mke2fs creates big inodes it does not zero out | |
2867 | * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE, | |
2868 | * so ignore those first few inodes. | |
2869 | */ | |
2870 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
2871 | if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
e4a10a36 KK |
2872 | EXT3_INODE_SIZE(inode->i_sb)) { |
2873 | brelse (bh); | |
473043dc | 2874 | ret = -EIO; |
1da177e4 | 2875 | goto bad_inode; |
e4a10a36 | 2876 | } |
1da177e4 LT |
2877 | if (ei->i_extra_isize == 0) { |
2878 | /* The extra space is currently unused. Use it. */ | |
2879 | ei->i_extra_isize = sizeof(struct ext3_inode) - | |
2880 | EXT3_GOOD_OLD_INODE_SIZE; | |
2881 | } else { | |
2882 | __le32 *magic = (void *)raw_inode + | |
2883 | EXT3_GOOD_OLD_INODE_SIZE + | |
2884 | ei->i_extra_isize; | |
2885 | if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC)) | |
2886 | ei->i_state |= EXT3_STATE_XATTR; | |
2887 | } | |
2888 | } else | |
2889 | ei->i_extra_isize = 0; | |
2890 | ||
2891 | if (S_ISREG(inode->i_mode)) { | |
2892 | inode->i_op = &ext3_file_inode_operations; | |
2893 | inode->i_fop = &ext3_file_operations; | |
2894 | ext3_set_aops(inode); | |
2895 | } else if (S_ISDIR(inode->i_mode)) { | |
2896 | inode->i_op = &ext3_dir_inode_operations; | |
2897 | inode->i_fop = &ext3_dir_operations; | |
2898 | } else if (S_ISLNK(inode->i_mode)) { | |
b5ed3112 | 2899 | if (ext3_inode_is_fast_symlink(inode)) { |
1da177e4 | 2900 | inode->i_op = &ext3_fast_symlink_inode_operations; |
b5ed3112 DG |
2901 | nd_terminate_link(ei->i_data, inode->i_size, |
2902 | sizeof(ei->i_data) - 1); | |
2903 | } else { | |
1da177e4 LT |
2904 | inode->i_op = &ext3_symlink_inode_operations; |
2905 | ext3_set_aops(inode); | |
2906 | } | |
2907 | } else { | |
2908 | inode->i_op = &ext3_special_inode_operations; | |
2909 | if (raw_inode->i_block[0]) | |
2910 | init_special_inode(inode, inode->i_mode, | |
2911 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
ae6ddcc5 | 2912 | else |
1da177e4 LT |
2913 | init_special_inode(inode, inode->i_mode, |
2914 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
2915 | } | |
2916 | brelse (iloc.bh); | |
2917 | ext3_set_inode_flags(inode); | |
473043dc DH |
2918 | unlock_new_inode(inode); |
2919 | return inode; | |
1da177e4 LT |
2920 | |
2921 | bad_inode: | |
473043dc DH |
2922 | iget_failed(inode); |
2923 | return ERR_PTR(ret); | |
1da177e4 LT |
2924 | } |
2925 | ||
2926 | /* | |
2927 | * Post the struct inode info into an on-disk inode location in the | |
2928 | * buffer-cache. This gobbles the caller's reference to the | |
2929 | * buffer_head in the inode location struct. | |
2930 | * | |
2931 | * The caller must have write access to iloc->bh. | |
2932 | */ | |
ae6ddcc5 MC |
2933 | static int ext3_do_update_inode(handle_t *handle, |
2934 | struct inode *inode, | |
1da177e4 LT |
2935 | struct ext3_iloc *iloc) |
2936 | { | |
2937 | struct ext3_inode *raw_inode = ext3_raw_inode(iloc); | |
2938 | struct ext3_inode_info *ei = EXT3_I(inode); | |
2939 | struct buffer_head *bh = iloc->bh; | |
2940 | int err = 0, rc, block; | |
2941 | ||
4f003fd3 CM |
2942 | again: |
2943 | /* we can't allow multiple procs in here at once, its a bit racey */ | |
2944 | lock_buffer(bh); | |
2945 | ||
1da177e4 LT |
2946 | /* For fields not not tracking in the in-memory inode, |
2947 | * initialise them to zero for new inodes. */ | |
2948 | if (ei->i_state & EXT3_STATE_NEW) | |
2949 | memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size); | |
2950 | ||
28be5abb | 2951 | ext3_get_inode_flags(ei); |
1da177e4 LT |
2952 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
2953 | if(!(test_opt(inode->i_sb, NO_UID32))) { | |
2954 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); | |
2955 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
2956 | /* | |
2957 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
2958 | * re-used with the upper 16 bits of the uid/gid intact | |
2959 | */ | |
2960 | if(!ei->i_dtime) { | |
2961 | raw_inode->i_uid_high = | |
2962 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
2963 | raw_inode->i_gid_high = | |
2964 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
2965 | } else { | |
2966 | raw_inode->i_uid_high = 0; | |
2967 | raw_inode->i_gid_high = 0; | |
2968 | } | |
2969 | } else { | |
2970 | raw_inode->i_uid_low = | |
2971 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
2972 | raw_inode->i_gid_low = | |
2973 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
2974 | raw_inode->i_uid_high = 0; | |
2975 | raw_inode->i_gid_high = 0; | |
2976 | } | |
2977 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
2978 | raw_inode->i_size = cpu_to_le32(ei->i_disksize); | |
2979 | raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec); | |
2980 | raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec); | |
2981 | raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec); | |
2982 | raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); | |
2983 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); | |
2984 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); | |
2985 | #ifdef EXT3_FRAGMENTS | |
2986 | raw_inode->i_faddr = cpu_to_le32(ei->i_faddr); | |
2987 | raw_inode->i_frag = ei->i_frag_no; | |
2988 | raw_inode->i_fsize = ei->i_frag_size; | |
2989 | #endif | |
2990 | raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); | |
2991 | if (!S_ISREG(inode->i_mode)) { | |
2992 | raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); | |
2993 | } else { | |
2994 | raw_inode->i_size_high = | |
2995 | cpu_to_le32(ei->i_disksize >> 32); | |
2996 | if (ei->i_disksize > 0x7fffffffULL) { | |
2997 | struct super_block *sb = inode->i_sb; | |
2998 | if (!EXT3_HAS_RO_COMPAT_FEATURE(sb, | |
2999 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE) || | |
3000 | EXT3_SB(sb)->s_es->s_rev_level == | |
3001 | cpu_to_le32(EXT3_GOOD_OLD_REV)) { | |
3002 | /* If this is the first large file | |
3003 | * created, add a flag to the superblock. | |
3004 | */ | |
4f003fd3 | 3005 | unlock_buffer(bh); |
1da177e4 LT |
3006 | err = ext3_journal_get_write_access(handle, |
3007 | EXT3_SB(sb)->s_sbh); | |
3008 | if (err) | |
3009 | goto out_brelse; | |
4f003fd3 | 3010 | |
1da177e4 LT |
3011 | ext3_update_dynamic_rev(sb); |
3012 | EXT3_SET_RO_COMPAT_FEATURE(sb, | |
3013 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE); | |
1da177e4 LT |
3014 | handle->h_sync = 1; |
3015 | err = ext3_journal_dirty_metadata(handle, | |
3016 | EXT3_SB(sb)->s_sbh); | |
4f003fd3 CM |
3017 | /* get our lock and start over */ |
3018 | goto again; | |
1da177e4 LT |
3019 | } |
3020 | } | |
3021 | } | |
3022 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
3023 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
3024 | if (old_valid_dev(inode->i_rdev)) { | |
3025 | raw_inode->i_block[0] = | |
3026 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
3027 | raw_inode->i_block[1] = 0; | |
3028 | } else { | |
3029 | raw_inode->i_block[0] = 0; | |
3030 | raw_inode->i_block[1] = | |
3031 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
3032 | raw_inode->i_block[2] = 0; | |
3033 | } | |
3034 | } else for (block = 0; block < EXT3_N_BLOCKS; block++) | |
3035 | raw_inode->i_block[block] = ei->i_data[block]; | |
3036 | ||
ff87b37d | 3037 | if (ei->i_extra_isize) |
1da177e4 LT |
3038 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
3039 | ||
3040 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
4f003fd3 | 3041 | unlock_buffer(bh); |
1da177e4 LT |
3042 | rc = ext3_journal_dirty_metadata(handle, bh); |
3043 | if (!err) | |
3044 | err = rc; | |
3045 | ei->i_state &= ~EXT3_STATE_NEW; | |
3046 | ||
fe8bc91c | 3047 | atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid); |
1da177e4 LT |
3048 | out_brelse: |
3049 | brelse (bh); | |
3050 | ext3_std_error(inode->i_sb, err); | |
3051 | return err; | |
3052 | } | |
3053 | ||
3054 | /* | |
3055 | * ext3_write_inode() | |
3056 | * | |
3057 | * We are called from a few places: | |
3058 | * | |
3059 | * - Within generic_file_write() for O_SYNC files. | |
3060 | * Here, there will be no transaction running. We wait for any running | |
3061 | * trasnaction to commit. | |
3062 | * | |
3063 | * - Within sys_sync(), kupdate and such. | |
3064 | * We wait on commit, if tol to. | |
3065 | * | |
3066 | * - Within prune_icache() (PF_MEMALLOC == true) | |
3067 | * Here we simply return. We can't afford to block kswapd on the | |
3068 | * journal commit. | |
3069 | * | |
3070 | * In all cases it is actually safe for us to return without doing anything, | |
3071 | * because the inode has been copied into a raw inode buffer in | |
3072 | * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for | |
3073 | * knfsd. | |
3074 | * | |
3075 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
3076 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
3077 | * which we are interested. | |
3078 | * | |
3079 | * It would be a bug for them to not do this. The code: | |
3080 | * | |
3081 | * mark_inode_dirty(inode) | |
3082 | * stuff(); | |
3083 | * inode->i_size = expr; | |
3084 | * | |
3085 | * is in error because a kswapd-driven write_inode() could occur while | |
3086 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
3087 | * will no longer be on the superblock's dirty inode list. | |
3088 | */ | |
3089 | int ext3_write_inode(struct inode *inode, int wait) | |
3090 | { | |
3091 | if (current->flags & PF_MEMALLOC) | |
3092 | return 0; | |
3093 | ||
3094 | if (ext3_journal_current_handle()) { | |
9ad163ae | 3095 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
1da177e4 LT |
3096 | dump_stack(); |
3097 | return -EIO; | |
3098 | } | |
3099 | ||
3100 | if (!wait) | |
3101 | return 0; | |
3102 | ||
3103 | return ext3_force_commit(inode->i_sb); | |
3104 | } | |
3105 | ||
3106 | /* | |
3107 | * ext3_setattr() | |
3108 | * | |
3109 | * Called from notify_change. | |
3110 | * | |
3111 | * We want to trap VFS attempts to truncate the file as soon as | |
3112 | * possible. In particular, we want to make sure that when the VFS | |
3113 | * shrinks i_size, we put the inode on the orphan list and modify | |
3114 | * i_disksize immediately, so that during the subsequent flushing of | |
3115 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
3116 | * commit will leave the blocks being flushed in an unused state on | |
3117 | * disk. (On recovery, the inode will get truncated and the blocks will | |
3118 | * be freed, so we have a strong guarantee that no future commit will | |
ae6ddcc5 | 3119 | * leave these blocks visible to the user.) |
1da177e4 LT |
3120 | * |
3121 | * Called with inode->sem down. | |
3122 | */ | |
3123 | int ext3_setattr(struct dentry *dentry, struct iattr *attr) | |
3124 | { | |
3125 | struct inode *inode = dentry->d_inode; | |
3126 | int error, rc = 0; | |
3127 | const unsigned int ia_valid = attr->ia_valid; | |
3128 | ||
3129 | error = inode_change_ok(inode, attr); | |
3130 | if (error) | |
3131 | return error; | |
3132 | ||
3133 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || | |
3134 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
3135 | handle_t *handle; | |
3136 | ||
3137 | /* (user+group)*(old+new) structure, inode write (sb, | |
3138 | * inode block, ? - but truncate inode update has it) */ | |
1f54587b JK |
3139 | handle = ext3_journal_start(inode, 2*(EXT3_QUOTA_INIT_BLOCKS(inode->i_sb)+ |
3140 | EXT3_QUOTA_DEL_BLOCKS(inode->i_sb))+3); | |
1da177e4 LT |
3141 | if (IS_ERR(handle)) { |
3142 | error = PTR_ERR(handle); | |
3143 | goto err_out; | |
3144 | } | |
81a05227 | 3145 | error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0; |
1da177e4 LT |
3146 | if (error) { |
3147 | ext3_journal_stop(handle); | |
3148 | return error; | |
3149 | } | |
3150 | /* Update corresponding info in inode so that everything is in | |
3151 | * one transaction */ | |
3152 | if (attr->ia_valid & ATTR_UID) | |
3153 | inode->i_uid = attr->ia_uid; | |
3154 | if (attr->ia_valid & ATTR_GID) | |
3155 | inode->i_gid = attr->ia_gid; | |
3156 | error = ext3_mark_inode_dirty(handle, inode); | |
3157 | ext3_journal_stop(handle); | |
3158 | } | |
3159 | ||
3160 | if (S_ISREG(inode->i_mode) && | |
3161 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { | |
3162 | handle_t *handle; | |
3163 | ||
3164 | handle = ext3_journal_start(inode, 3); | |
3165 | if (IS_ERR(handle)) { | |
3166 | error = PTR_ERR(handle); | |
3167 | goto err_out; | |
3168 | } | |
3169 | ||
3170 | error = ext3_orphan_add(handle, inode); | |
3171 | EXT3_I(inode)->i_disksize = attr->ia_size; | |
3172 | rc = ext3_mark_inode_dirty(handle, inode); | |
3173 | if (!error) | |
3174 | error = rc; | |
3175 | ext3_journal_stop(handle); | |
3176 | } | |
3177 | ||
3178 | rc = inode_setattr(inode, attr); | |
3179 | ||
1da177e4 LT |
3180 | if (!rc && (ia_valid & ATTR_MODE)) |
3181 | rc = ext3_acl_chmod(inode); | |
3182 | ||
3183 | err_out: | |
3184 | ext3_std_error(inode->i_sb, error); | |
3185 | if (!error) | |
3186 | error = rc; | |
3187 | return error; | |
3188 | } | |
3189 | ||
3190 | ||
3191 | /* | |
d6859bfc | 3192 | * How many blocks doth make a writepage()? |
1da177e4 LT |
3193 | * |
3194 | * With N blocks per page, it may be: | |
3195 | * N data blocks | |
3196 | * 2 indirect block | |
3197 | * 2 dindirect | |
3198 | * 1 tindirect | |
3199 | * N+5 bitmap blocks (from the above) | |
3200 | * N+5 group descriptor summary blocks | |
3201 | * 1 inode block | |
3202 | * 1 superblock. | |
3203 | * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files | |
3204 | * | |
3205 | * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS | |
3206 | * | |
3207 | * With ordered or writeback data it's the same, less the N data blocks. | |
3208 | * | |
3209 | * If the inode's direct blocks can hold an integral number of pages then a | |
3210 | * page cannot straddle two indirect blocks, and we can only touch one indirect | |
3211 | * and dindirect block, and the "5" above becomes "3". | |
3212 | * | |
3213 | * This still overestimates under most circumstances. If we were to pass the | |
3214 | * start and end offsets in here as well we could do block_to_path() on each | |
3215 | * block and work out the exact number of indirects which are touched. Pah. | |
3216 | */ | |
3217 | ||
3218 | static int ext3_writepage_trans_blocks(struct inode *inode) | |
3219 | { | |
3220 | int bpp = ext3_journal_blocks_per_page(inode); | |
3221 | int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3; | |
3222 | int ret; | |
3223 | ||
3224 | if (ext3_should_journal_data(inode)) | |
3225 | ret = 3 * (bpp + indirects) + 2; | |
3226 | else | |
3227 | ret = 2 * (bpp + indirects) + 2; | |
3228 | ||
3229 | #ifdef CONFIG_QUOTA | |
81a05227 | 3230 | /* We know that structure was already allocated during vfs_dq_init so |
1da177e4 | 3231 | * we will be updating only the data blocks + inodes */ |
1f54587b | 3232 | ret += 2*EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb); |
1da177e4 LT |
3233 | #endif |
3234 | ||
3235 | return ret; | |
3236 | } | |
3237 | ||
3238 | /* | |
3239 | * The caller must have previously called ext3_reserve_inode_write(). | |
3240 | * Give this, we know that the caller already has write access to iloc->bh. | |
3241 | */ | |
3242 | int ext3_mark_iloc_dirty(handle_t *handle, | |
3243 | struct inode *inode, struct ext3_iloc *iloc) | |
3244 | { | |
3245 | int err = 0; | |
3246 | ||
3247 | /* the do_update_inode consumes one bh->b_count */ | |
3248 | get_bh(iloc->bh); | |
3249 | ||
3250 | /* ext3_do_update_inode() does journal_dirty_metadata */ | |
3251 | err = ext3_do_update_inode(handle, inode, iloc); | |
3252 | put_bh(iloc->bh); | |
3253 | return err; | |
3254 | } | |
3255 | ||
ae6ddcc5 | 3256 | /* |
1da177e4 | 3257 | * On success, We end up with an outstanding reference count against |
ae6ddcc5 | 3258 | * iloc->bh. This _must_ be cleaned up later. |
1da177e4 LT |
3259 | */ |
3260 | ||
3261 | int | |
ae6ddcc5 | 3262 | ext3_reserve_inode_write(handle_t *handle, struct inode *inode, |
1da177e4 LT |
3263 | struct ext3_iloc *iloc) |
3264 | { | |
3265 | int err = 0; | |
3266 | if (handle) { | |
3267 | err = ext3_get_inode_loc(inode, iloc); | |
3268 | if (!err) { | |
3269 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
3270 | err = ext3_journal_get_write_access(handle, iloc->bh); | |
3271 | if (err) { | |
3272 | brelse(iloc->bh); | |
3273 | iloc->bh = NULL; | |
3274 | } | |
3275 | } | |
3276 | } | |
3277 | ext3_std_error(inode->i_sb, err); | |
3278 | return err; | |
3279 | } | |
3280 | ||
3281 | /* | |
d6859bfc AM |
3282 | * What we do here is to mark the in-core inode as clean with respect to inode |
3283 | * dirtiness (it may still be data-dirty). | |
1da177e4 LT |
3284 | * This means that the in-core inode may be reaped by prune_icache |
3285 | * without having to perform any I/O. This is a very good thing, | |
3286 | * because *any* task may call prune_icache - even ones which | |
3287 | * have a transaction open against a different journal. | |
3288 | * | |
3289 | * Is this cheating? Not really. Sure, we haven't written the | |
3290 | * inode out, but prune_icache isn't a user-visible syncing function. | |
3291 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
3292 | * we start and wait on commits. | |
3293 | * | |
3294 | * Is this efficient/effective? Well, we're being nice to the system | |
3295 | * by cleaning up our inodes proactively so they can be reaped | |
3296 | * without I/O. But we are potentially leaving up to five seconds' | |
3297 | * worth of inodes floating about which prune_icache wants us to | |
3298 | * write out. One way to fix that would be to get prune_icache() | |
3299 | * to do a write_super() to free up some memory. It has the desired | |
3300 | * effect. | |
3301 | */ | |
3302 | int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode) | |
3303 | { | |
3304 | struct ext3_iloc iloc; | |
3305 | int err; | |
3306 | ||
3307 | might_sleep(); | |
3308 | err = ext3_reserve_inode_write(handle, inode, &iloc); | |
3309 | if (!err) | |
3310 | err = ext3_mark_iloc_dirty(handle, inode, &iloc); | |
3311 | return err; | |
3312 | } | |
3313 | ||
3314 | /* | |
d6859bfc | 3315 | * ext3_dirty_inode() is called from __mark_inode_dirty() |
1da177e4 LT |
3316 | * |
3317 | * We're really interested in the case where a file is being extended. | |
3318 | * i_size has been changed by generic_commit_write() and we thus need | |
3319 | * to include the updated inode in the current transaction. | |
3320 | * | |
81a05227 | 3321 | * Also, vfs_dq_alloc_space() will always dirty the inode when blocks |
1da177e4 LT |
3322 | * are allocated to the file. |
3323 | * | |
3324 | * If the inode is marked synchronous, we don't honour that here - doing | |
3325 | * so would cause a commit on atime updates, which we don't bother doing. | |
3326 | * We handle synchronous inodes at the highest possible level. | |
3327 | */ | |
3328 | void ext3_dirty_inode(struct inode *inode) | |
3329 | { | |
3330 | handle_t *current_handle = ext3_journal_current_handle(); | |
3331 | handle_t *handle; | |
3332 | ||
3333 | handle = ext3_journal_start(inode, 2); | |
3334 | if (IS_ERR(handle)) | |
3335 | goto out; | |
3336 | if (current_handle && | |
3337 | current_handle->h_transaction != handle->h_transaction) { | |
3338 | /* This task has a transaction open against a different fs */ | |
3339 | printk(KERN_EMERG "%s: transactions do not match!\n", | |
e05b6b52 | 3340 | __func__); |
1da177e4 LT |
3341 | } else { |
3342 | jbd_debug(5, "marking dirty. outer handle=%p\n", | |
3343 | current_handle); | |
3344 | ext3_mark_inode_dirty(handle, inode); | |
3345 | } | |
3346 | ext3_journal_stop(handle); | |
3347 | out: | |
3348 | return; | |
3349 | } | |
3350 | ||
d6859bfc | 3351 | #if 0 |
ae6ddcc5 | 3352 | /* |
1da177e4 LT |
3353 | * Bind an inode's backing buffer_head into this transaction, to prevent |
3354 | * it from being flushed to disk early. Unlike | |
3355 | * ext3_reserve_inode_write, this leaves behind no bh reference and | |
3356 | * returns no iloc structure, so the caller needs to repeat the iloc | |
3357 | * lookup to mark the inode dirty later. | |
3358 | */ | |
d6859bfc | 3359 | static int ext3_pin_inode(handle_t *handle, struct inode *inode) |
1da177e4 LT |
3360 | { |
3361 | struct ext3_iloc iloc; | |
3362 | ||
3363 | int err = 0; | |
3364 | if (handle) { | |
3365 | err = ext3_get_inode_loc(inode, &iloc); | |
3366 | if (!err) { | |
3367 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
3368 | err = journal_get_write_access(handle, iloc.bh); | |
3369 | if (!err) | |
ae6ddcc5 | 3370 | err = ext3_journal_dirty_metadata(handle, |
1da177e4 LT |
3371 | iloc.bh); |
3372 | brelse(iloc.bh); | |
3373 | } | |
3374 | } | |
3375 | ext3_std_error(inode->i_sb, err); | |
3376 | return err; | |
3377 | } | |
3378 | #endif | |
3379 | ||
3380 | int ext3_change_inode_journal_flag(struct inode *inode, int val) | |
3381 | { | |
3382 | journal_t *journal; | |
3383 | handle_t *handle; | |
3384 | int err; | |
3385 | ||
3386 | /* | |
3387 | * We have to be very careful here: changing a data block's | |
3388 | * journaling status dynamically is dangerous. If we write a | |
3389 | * data block to the journal, change the status and then delete | |
3390 | * that block, we risk forgetting to revoke the old log record | |
3391 | * from the journal and so a subsequent replay can corrupt data. | |
3392 | * So, first we make sure that the journal is empty and that | |
3393 | * nobody is changing anything. | |
3394 | */ | |
3395 | ||
3396 | journal = EXT3_JOURNAL(inode); | |
e3a68e30 | 3397 | if (is_journal_aborted(journal)) |
1da177e4 LT |
3398 | return -EROFS; |
3399 | ||
3400 | journal_lock_updates(journal); | |
3401 | journal_flush(journal); | |
3402 | ||
3403 | /* | |
3404 | * OK, there are no updates running now, and all cached data is | |
3405 | * synced to disk. We are now in a completely consistent state | |
3406 | * which doesn't have anything in the journal, and we know that | |
3407 | * no filesystem updates are running, so it is safe to modify | |
3408 | * the inode's in-core data-journaling state flag now. | |
3409 | */ | |
3410 | ||
3411 | if (val) | |
3412 | EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL; | |
3413 | else | |
3414 | EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL; | |
3415 | ext3_set_aops(inode); | |
3416 | ||
3417 | journal_unlock_updates(journal); | |
3418 | ||
3419 | /* Finally we can mark the inode as dirty. */ | |
3420 | ||
3421 | handle = ext3_journal_start(inode, 1); | |
3422 | if (IS_ERR(handle)) | |
3423 | return PTR_ERR(handle); | |
3424 | ||
3425 | err = ext3_mark_inode_dirty(handle, inode); | |
3426 | handle->h_sync = 1; | |
3427 | ext3_journal_stop(handle); | |
3428 | ext3_std_error(inode->i_sb, err); | |
3429 | ||
3430 | return err; | |
3431 | } |