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> |
785c4bcc | 41 | #include <trace/events/ext3.h> |
1da177e4 LT |
42 | #include "xattr.h" |
43 | #include "acl.h" | |
44 | ||
45 | static int ext3_writepage_trans_blocks(struct inode *inode); | |
ee3e77f1 | 46 | static int ext3_block_truncate_page(struct inode *inode, loff_t from); |
1da177e4 LT |
47 | |
48 | /* | |
49 | * Test whether an inode is a fast symlink. | |
50 | */ | |
d6859bfc | 51 | static int ext3_inode_is_fast_symlink(struct inode *inode) |
1da177e4 LT |
52 | { |
53 | int ea_blocks = EXT3_I(inode)->i_file_acl ? | |
54 | (inode->i_sb->s_blocksize >> 9) : 0; | |
55 | ||
d6859bfc | 56 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
1da177e4 LT |
57 | } |
58 | ||
d6859bfc AM |
59 | /* |
60 | * The ext3 forget function must perform a revoke if we are freeing data | |
1da177e4 | 61 | * which has been journaled. Metadata (eg. indirect blocks) must be |
ae6ddcc5 | 62 | * revoked in all cases. |
1da177e4 LT |
63 | * |
64 | * "bh" may be NULL: a metadata block may have been freed from memory | |
65 | * but there may still be a record of it in the journal, and that record | |
66 | * still needs to be revoked. | |
67 | */ | |
d6859bfc | 68 | int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode, |
1c2bf374 | 69 | struct buffer_head *bh, ext3_fsblk_t blocknr) |
1da177e4 LT |
70 | { |
71 | int err; | |
72 | ||
73 | might_sleep(); | |
74 | ||
785c4bcc | 75 | trace_ext3_forget(inode, is_metadata, blocknr); |
1da177e4 LT |
76 | BUFFER_TRACE(bh, "enter"); |
77 | ||
78 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " | |
79 | "data mode %lx\n", | |
80 | bh, is_metadata, inode->i_mode, | |
81 | test_opt(inode->i_sb, DATA_FLAGS)); | |
82 | ||
83 | /* Never use the revoke function if we are doing full data | |
84 | * journaling: there is no need to, and a V1 superblock won't | |
85 | * support it. Otherwise, only skip the revoke on un-journaled | |
86 | * data blocks. */ | |
87 | ||
88 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA || | |
89 | (!is_metadata && !ext3_should_journal_data(inode))) { | |
90 | if (bh) { | |
91 | BUFFER_TRACE(bh, "call journal_forget"); | |
92 | return ext3_journal_forget(handle, bh); | |
93 | } | |
94 | return 0; | |
95 | } | |
96 | ||
97 | /* | |
98 | * data!=journal && (is_metadata || should_journal_data(inode)) | |
99 | */ | |
100 | BUFFER_TRACE(bh, "call ext3_journal_revoke"); | |
101 | err = ext3_journal_revoke(handle, blocknr, bh); | |
102 | if (err) | |
e05b6b52 | 103 | ext3_abort(inode->i_sb, __func__, |
1da177e4 LT |
104 | "error %d when attempting revoke", err); |
105 | BUFFER_TRACE(bh, "exit"); | |
106 | return err; | |
107 | } | |
108 | ||
109 | /* | |
d6859bfc | 110 | * Work out how many blocks we need to proceed with the next chunk of a |
1da177e4 LT |
111 | * truncate transaction. |
112 | */ | |
ae6ddcc5 | 113 | static unsigned long blocks_for_truncate(struct inode *inode) |
1da177e4 LT |
114 | { |
115 | unsigned long needed; | |
116 | ||
117 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
118 | ||
119 | /* Give ourselves just enough room to cope with inodes in which | |
120 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
121 | * which resulted in random data in an inode which looked enough | |
122 | * like a regular file for ext3 to try to delete it. Things | |
123 | * will go a bit crazy if that happens, but at least we should | |
124 | * try not to panic the whole kernel. */ | |
125 | if (needed < 2) | |
126 | needed = 2; | |
127 | ||
128 | /* But we need to bound the transaction so we don't overflow the | |
129 | * journal. */ | |
ae6ddcc5 | 130 | if (needed > EXT3_MAX_TRANS_DATA) |
1da177e4 LT |
131 | needed = EXT3_MAX_TRANS_DATA; |
132 | ||
1f54587b | 133 | return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
1da177e4 LT |
134 | } |
135 | ||
ae6ddcc5 | 136 | /* |
1da177e4 LT |
137 | * Truncate transactions can be complex and absolutely huge. So we need to |
138 | * be able to restart the transaction at a conventient checkpoint to make | |
139 | * sure we don't overflow the journal. | |
140 | * | |
141 | * start_transaction gets us a new handle for a truncate transaction, | |
142 | * and extend_transaction tries to extend the existing one a bit. If | |
143 | * extend fails, we need to propagate the failure up and restart the | |
ae6ddcc5 | 144 | * transaction in the top-level truncate loop. --sct |
1da177e4 | 145 | */ |
ae6ddcc5 | 146 | static handle_t *start_transaction(struct inode *inode) |
1da177e4 LT |
147 | { |
148 | handle_t *result; | |
149 | ||
150 | result = ext3_journal_start(inode, blocks_for_truncate(inode)); | |
151 | if (!IS_ERR(result)) | |
152 | return result; | |
153 | ||
154 | ext3_std_error(inode->i_sb, PTR_ERR(result)); | |
155 | return result; | |
156 | } | |
157 | ||
158 | /* | |
159 | * Try to extend this transaction for the purposes of truncation. | |
160 | * | |
161 | * Returns 0 if we managed to create more room. If we can't create more | |
162 | * room, and the transaction must be restarted we return 1. | |
163 | */ | |
164 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
165 | { | |
166 | if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS) | |
167 | return 0; | |
168 | if (!ext3_journal_extend(handle, blocks_for_truncate(inode))) | |
169 | return 0; | |
170 | return 1; | |
171 | } | |
172 | ||
173 | /* | |
174 | * Restart the transaction associated with *handle. This does a commit, | |
175 | * so before we call here everything must be consistently dirtied against | |
176 | * this transaction. | |
177 | */ | |
00171d3c | 178 | static int truncate_restart_transaction(handle_t *handle, struct inode *inode) |
1da177e4 | 179 | { |
00171d3c JK |
180 | int ret; |
181 | ||
1da177e4 | 182 | jbd_debug(2, "restarting handle %p\n", handle); |
00171d3c JK |
183 | /* |
184 | * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle | |
185 | * At this moment, get_block can be called only for blocks inside | |
186 | * i_size since page cache has been already dropped and writes are | |
187 | * blocked by i_mutex. So we can safely drop the truncate_mutex. | |
188 | */ | |
189 | mutex_unlock(&EXT3_I(inode)->truncate_mutex); | |
190 | ret = ext3_journal_restart(handle, blocks_for_truncate(inode)); | |
191 | mutex_lock(&EXT3_I(inode)->truncate_mutex); | |
192 | return ret; | |
1da177e4 LT |
193 | } |
194 | ||
195 | /* | |
ac14a95b | 196 | * Called at inode eviction from icache |
1da177e4 | 197 | */ |
ac14a95b | 198 | void ext3_evict_inode (struct inode *inode) |
1da177e4 | 199 | { |
b22570d9 | 200 | struct ext3_inode_info *ei = EXT3_I(inode); |
ac14a95b | 201 | struct ext3_block_alloc_info *rsv; |
1da177e4 | 202 | handle_t *handle; |
ac14a95b | 203 | int want_delete = 0; |
1da177e4 | 204 | |
785c4bcc | 205 | trace_ext3_evict_inode(inode); |
ac14a95b | 206 | if (!inode->i_nlink && !is_bad_inode(inode)) { |
871a2931 | 207 | dquot_initialize(inode); |
ac14a95b AV |
208 | want_delete = 1; |
209 | } | |
907f4554 | 210 | |
b22570d9 JK |
211 | /* |
212 | * When journalling data dirty buffers are tracked only in the journal. | |
213 | * So although mm thinks everything is clean and ready for reaping the | |
214 | * inode might still have some pages to write in the running | |
215 | * transaction or waiting to be checkpointed. Thus calling | |
216 | * journal_invalidatepage() (via truncate_inode_pages()) to discard | |
217 | * these buffers can cause data loss. Also even if we did not discard | |
218 | * these buffers, we would have no way to find them after the inode | |
219 | * is reaped and thus user could see stale data if he tries to read | |
220 | * them before the transaction is checkpointed. So be careful and | |
221 | * force everything to disk here... We use ei->i_datasync_tid to | |
222 | * store the newest transaction containing inode's data. | |
223 | * | |
224 | * Note that directories do not have this problem because they don't | |
225 | * use page cache. | |
226 | */ | |
227 | if (inode->i_nlink && ext3_should_journal_data(inode) && | |
228 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) { | |
229 | tid_t commit_tid = atomic_read(&ei->i_datasync_tid); | |
230 | journal_t *journal = EXT3_SB(inode->i_sb)->s_journal; | |
231 | ||
232 | log_start_commit(journal, commit_tid); | |
233 | log_wait_commit(journal, commit_tid); | |
234 | filemap_write_and_wait(&inode->i_data); | |
235 | } | |
fef26658 MF |
236 | truncate_inode_pages(&inode->i_data, 0); |
237 | ||
ac14a95b | 238 | ext3_discard_reservation(inode); |
b22570d9 JK |
239 | rsv = ei->i_block_alloc_info; |
240 | ei->i_block_alloc_info = NULL; | |
ac14a95b AV |
241 | if (unlikely(rsv)) |
242 | kfree(rsv); | |
243 | ||
244 | if (!want_delete) | |
1da177e4 LT |
245 | goto no_delete; |
246 | ||
247 | handle = start_transaction(inode); | |
248 | if (IS_ERR(handle)) { | |
d6859bfc AM |
249 | /* |
250 | * If we're going to skip the normal cleanup, we still need to | |
251 | * make sure that the in-core orphan linked list is properly | |
252 | * cleaned up. | |
253 | */ | |
1da177e4 LT |
254 | ext3_orphan_del(NULL, inode); |
255 | goto no_delete; | |
256 | } | |
257 | ||
258 | if (IS_SYNC(inode)) | |
259 | handle->h_sync = 1; | |
260 | inode->i_size = 0; | |
261 | if (inode->i_blocks) | |
262 | ext3_truncate(inode); | |
263 | /* | |
40680f2f JK |
264 | * Kill off the orphan record created when the inode lost the last |
265 | * link. Note that ext3_orphan_del() has to be able to cope with the | |
266 | * deletion of a non-existent orphan - ext3_truncate() could | |
267 | * have removed the record. | |
1da177e4 LT |
268 | */ |
269 | ext3_orphan_del(handle, inode); | |
b22570d9 | 270 | ei->i_dtime = get_seconds(); |
1da177e4 | 271 | |
ae6ddcc5 | 272 | /* |
1da177e4 LT |
273 | * One subtle ordering requirement: if anything has gone wrong |
274 | * (transaction abort, IO errors, whatever), then we can still | |
275 | * do these next steps (the fs will already have been marked as | |
276 | * having errors), but we can't free the inode if the mark_dirty | |
ae6ddcc5 | 277 | * fails. |
1da177e4 | 278 | */ |
ac14a95b AV |
279 | if (ext3_mark_inode_dirty(handle, inode)) { |
280 | /* If that failed, just dquot_drop() and be done with that */ | |
281 | dquot_drop(inode); | |
282 | end_writeback(inode); | |
283 | } else { | |
284 | ext3_xattr_delete_inode(handle, inode); | |
285 | dquot_free_inode(inode); | |
286 | dquot_drop(inode); | |
287 | end_writeback(inode); | |
1da177e4 | 288 | ext3_free_inode(handle, inode); |
ac14a95b | 289 | } |
1da177e4 LT |
290 | ext3_journal_stop(handle); |
291 | return; | |
292 | no_delete: | |
ac14a95b AV |
293 | end_writeback(inode); |
294 | dquot_drop(inode); | |
1da177e4 LT |
295 | } |
296 | ||
1da177e4 LT |
297 | typedef struct { |
298 | __le32 *p; | |
299 | __le32 key; | |
300 | struct buffer_head *bh; | |
301 | } Indirect; | |
302 | ||
303 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
304 | { | |
305 | p->key = *(p->p = v); | |
306 | p->bh = bh; | |
307 | } | |
308 | ||
d6859bfc | 309 | static int verify_chain(Indirect *from, Indirect *to) |
1da177e4 LT |
310 | { |
311 | while (from <= to && from->key == *from->p) | |
312 | from++; | |
313 | return (from > to); | |
314 | } | |
315 | ||
316 | /** | |
317 | * ext3_block_to_path - parse the block number into array of offsets | |
318 | * @inode: inode in question (we are only interested in its superblock) | |
319 | * @i_block: block number to be parsed | |
320 | * @offsets: array to store the offsets in | |
321 | * @boundary: set this non-zero if the referred-to block is likely to be | |
322 | * followed (on disk) by an indirect block. | |
323 | * | |
324 | * To store the locations of file's data ext3 uses a data structure common | |
325 | * for UNIX filesystems - tree of pointers anchored in the inode, with | |
326 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
327 | * This function translates the block number into path in that tree - | |
328 | * return value is the path length and @offsets[n] is the offset of | |
329 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
330 | * (negative or too large) warning is printed and zero returned. | |
331 | * | |
332 | * Note: function doesn't find node addresses, so no IO is needed. All | |
333 | * we need to know is the capacity of indirect blocks (taken from the | |
334 | * inode->i_sb). | |
335 | */ | |
336 | ||
337 | /* | |
338 | * Portability note: the last comparison (check that we fit into triple | |
339 | * indirect block) is spelled differently, because otherwise on an | |
340 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
341 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
342 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
343 | * i_block would have to be negative in the very beginning, so we would not | |
344 | * get there at all. | |
345 | */ | |
346 | ||
347 | static int ext3_block_to_path(struct inode *inode, | |
348 | long i_block, int offsets[4], int *boundary) | |
349 | { | |
350 | int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb); | |
351 | int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
352 | const long direct_blocks = EXT3_NDIR_BLOCKS, | |
353 | indirect_blocks = ptrs, | |
354 | double_blocks = (1 << (ptrs_bits * 2)); | |
355 | int n = 0; | |
356 | int final = 0; | |
357 | ||
358 | if (i_block < 0) { | |
359 | ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0"); | |
360 | } else if (i_block < direct_blocks) { | |
361 | offsets[n++] = i_block; | |
362 | final = direct_blocks; | |
363 | } else if ( (i_block -= direct_blocks) < indirect_blocks) { | |
364 | offsets[n++] = EXT3_IND_BLOCK; | |
365 | offsets[n++] = i_block; | |
366 | final = ptrs; | |
367 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
368 | offsets[n++] = EXT3_DIND_BLOCK; | |
369 | offsets[n++] = i_block >> ptrs_bits; | |
370 | offsets[n++] = i_block & (ptrs - 1); | |
371 | final = ptrs; | |
372 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
373 | offsets[n++] = EXT3_TIND_BLOCK; | |
374 | offsets[n++] = i_block >> (ptrs_bits * 2); | |
375 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
376 | offsets[n++] = i_block & (ptrs - 1); | |
377 | final = ptrs; | |
378 | } else { | |
d6859bfc | 379 | ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big"); |
1da177e4 LT |
380 | } |
381 | if (boundary) | |
89747d36 | 382 | *boundary = final - 1 - (i_block & (ptrs - 1)); |
1da177e4 LT |
383 | return n; |
384 | } | |
385 | ||
386 | /** | |
387 | * ext3_get_branch - read the chain of indirect blocks leading to data | |
388 | * @inode: inode in question | |
389 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
390 | * @offsets: offsets of pointers in inode/indirect blocks | |
391 | * @chain: place to store the result | |
392 | * @err: here we store the error value | |
393 | * | |
394 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
395 | * if everything went OK or the pointer to the last filled triple | |
396 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
397 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
398 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
399 | * number (it points into struct inode for i==0 and into the bh->b_data | |
400 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
401 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
402 | * numbers of the chain, addresses they were taken from (and where we can | |
403 | * verify that chain did not change) and buffer_heads hosting these | |
404 | * numbers. | |
405 | * | |
406 | * Function stops when it stumbles upon zero pointer (absent block) | |
407 | * (pointer to last triple returned, *@err == 0) | |
408 | * or when it gets an IO error reading an indirect block | |
409 | * (ditto, *@err == -EIO) | |
410 | * or when it notices that chain had been changed while it was reading | |
411 | * (ditto, *@err == -EAGAIN) | |
412 | * or when it reads all @depth-1 indirect blocks successfully and finds | |
413 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
414 | */ | |
415 | static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets, | |
416 | Indirect chain[4], int *err) | |
417 | { | |
418 | struct super_block *sb = inode->i_sb; | |
419 | Indirect *p = chain; | |
420 | struct buffer_head *bh; | |
421 | ||
422 | *err = 0; | |
423 | /* i_data is not going away, no lock needed */ | |
424 | add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets); | |
425 | if (!p->key) | |
426 | goto no_block; | |
427 | while (--depth) { | |
428 | bh = sb_bread(sb, le32_to_cpu(p->key)); | |
429 | if (!bh) | |
430 | goto failure; | |
431 | /* Reader: pointers */ | |
432 | if (!verify_chain(chain, p)) | |
433 | goto changed; | |
434 | add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); | |
435 | /* Reader: end */ | |
436 | if (!p->key) | |
437 | goto no_block; | |
438 | } | |
439 | return NULL; | |
440 | ||
441 | changed: | |
442 | brelse(bh); | |
443 | *err = -EAGAIN; | |
444 | goto no_block; | |
445 | failure: | |
446 | *err = -EIO; | |
447 | no_block: | |
448 | return p; | |
449 | } | |
450 | ||
451 | /** | |
452 | * ext3_find_near - find a place for allocation with sufficient locality | |
453 | * @inode: owner | |
454 | * @ind: descriptor of indirect block. | |
455 | * | |
1cc8dcf5 | 456 | * This function returns the preferred place for block allocation. |
1da177e4 LT |
457 | * It is used when heuristic for sequential allocation fails. |
458 | * Rules are: | |
459 | * + if there is a block to the left of our position - allocate near it. | |
460 | * + if pointer will live in indirect block - allocate near that block. | |
461 | * + if pointer will live in inode - allocate in the same | |
ae6ddcc5 | 462 | * cylinder group. |
1da177e4 LT |
463 | * |
464 | * In the latter case we colour the starting block by the callers PID to | |
465 | * prevent it from clashing with concurrent allocations for a different inode | |
466 | * in the same block group. The PID is used here so that functionally related | |
467 | * files will be close-by on-disk. | |
468 | * | |
469 | * Caller must make sure that @ind is valid and will stay that way. | |
470 | */ | |
43d23f90 | 471 | static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind) |
1da177e4 LT |
472 | { |
473 | struct ext3_inode_info *ei = EXT3_I(inode); | |
474 | __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data; | |
475 | __le32 *p; | |
43d23f90 MC |
476 | ext3_fsblk_t bg_start; |
477 | ext3_grpblk_t colour; | |
1da177e4 LT |
478 | |
479 | /* Try to find previous block */ | |
d6859bfc | 480 | for (p = ind->p - 1; p >= start; p--) { |
1da177e4 LT |
481 | if (*p) |
482 | return le32_to_cpu(*p); | |
d6859bfc | 483 | } |
1da177e4 LT |
484 | |
485 | /* No such thing, so let's try location of indirect block */ | |
486 | if (ind->bh) | |
487 | return ind->bh->b_blocknr; | |
488 | ||
489 | /* | |
d6859bfc AM |
490 | * It is going to be referred to from the inode itself? OK, just put it |
491 | * into the same cylinder group then. | |
1da177e4 | 492 | */ |
43d23f90 | 493 | bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group); |
1da177e4 LT |
494 | colour = (current->pid % 16) * |
495 | (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16); | |
496 | return bg_start + colour; | |
497 | } | |
498 | ||
499 | /** | |
1cc8dcf5 | 500 | * ext3_find_goal - find a preferred place for allocation. |
1da177e4 LT |
501 | * @inode: owner |
502 | * @block: block we want | |
1da177e4 | 503 | * @partial: pointer to the last triple within a chain |
1da177e4 | 504 | * |
1cc8dcf5 | 505 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 506 | * returns it. |
1da177e4 LT |
507 | */ |
508 | ||
43d23f90 | 509 | static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block, |
fb01bfda | 510 | Indirect *partial) |
1da177e4 | 511 | { |
d6859bfc AM |
512 | struct ext3_block_alloc_info *block_i; |
513 | ||
514 | block_i = EXT3_I(inode)->i_block_alloc_info; | |
1da177e4 LT |
515 | |
516 | /* | |
517 | * try the heuristic for sequential allocation, | |
518 | * failing that at least try to get decent locality. | |
519 | */ | |
520 | if (block_i && (block == block_i->last_alloc_logical_block + 1) | |
521 | && (block_i->last_alloc_physical_block != 0)) { | |
fe55c452 | 522 | return block_i->last_alloc_physical_block + 1; |
1da177e4 LT |
523 | } |
524 | ||
fe55c452 | 525 | return ext3_find_near(inode, partial); |
1da177e4 | 526 | } |
d6859bfc | 527 | |
b47b2478 | 528 | /** |
a4c18ad2 | 529 | * ext3_blks_to_allocate - Look up the block map and count the number |
b47b2478 MC |
530 | * of direct blocks need to be allocated for the given branch. |
531 | * | |
e9ad5620 | 532 | * @branch: chain of indirect blocks |
b47b2478 MC |
533 | * @k: number of blocks need for indirect blocks |
534 | * @blks: number of data blocks to be mapped. | |
535 | * @blocks_to_boundary: the offset in the indirect block | |
536 | * | |
537 | * return the total number of blocks to be allocate, including the | |
538 | * direct and indirect blocks. | |
539 | */ | |
d6859bfc | 540 | static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks, |
b47b2478 MC |
541 | int blocks_to_boundary) |
542 | { | |
543 | unsigned long count = 0; | |
544 | ||
545 | /* | |
546 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
547 | * then it's clear blocks on that path have not allocated | |
548 | */ | |
549 | if (k > 0) { | |
d6859bfc | 550 | /* right now we don't handle cross boundary allocation */ |
b47b2478 MC |
551 | if (blks < blocks_to_boundary + 1) |
552 | count += blks; | |
553 | else | |
554 | count += blocks_to_boundary + 1; | |
555 | return count; | |
556 | } | |
557 | ||
558 | count++; | |
559 | while (count < blks && count <= blocks_to_boundary && | |
560 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
561 | count++; | |
562 | } | |
563 | return count; | |
564 | } | |
565 | ||
566 | /** | |
a4c18ad2 NK |
567 | * ext3_alloc_blocks - multiple allocate blocks needed for a branch |
568 | * @handle: handle for this transaction | |
569 | * @inode: owner | |
570 | * @goal: preferred place for allocation | |
b47b2478 MC |
571 | * @indirect_blks: the number of blocks need to allocate for indirect |
572 | * blocks | |
a4c18ad2 | 573 | * @blks: number of blocks need to allocated for direct blocks |
b47b2478 MC |
574 | * @new_blocks: on return it will store the new block numbers for |
575 | * the indirect blocks(if needed) and the first direct block, | |
a4c18ad2 NK |
576 | * @err: here we store the error value |
577 | * | |
578 | * return the number of direct blocks allocated | |
b47b2478 MC |
579 | */ |
580 | static int ext3_alloc_blocks(handle_t *handle, struct inode *inode, | |
43d23f90 MC |
581 | ext3_fsblk_t goal, int indirect_blks, int blks, |
582 | ext3_fsblk_t new_blocks[4], int *err) | |
b47b2478 MC |
583 | { |
584 | int target, i; | |
585 | unsigned long count = 0; | |
586 | int index = 0; | |
43d23f90 | 587 | ext3_fsblk_t current_block = 0; |
b47b2478 MC |
588 | int ret = 0; |
589 | ||
590 | /* | |
591 | * Here we try to allocate the requested multiple blocks at once, | |
592 | * on a best-effort basis. | |
593 | * To build a branch, we should allocate blocks for | |
594 | * the indirect blocks(if not allocated yet), and at least | |
595 | * the first direct block of this branch. That's the | |
596 | * minimum number of blocks need to allocate(required) | |
597 | */ | |
598 | target = blks + indirect_blks; | |
599 | ||
600 | while (1) { | |
601 | count = target; | |
602 | /* allocating blocks for indirect blocks and direct blocks */ | |
d6859bfc | 603 | current_block = ext3_new_blocks(handle,inode,goal,&count,err); |
b47b2478 MC |
604 | if (*err) |
605 | goto failed_out; | |
606 | ||
607 | target -= count; | |
608 | /* allocate blocks for indirect blocks */ | |
609 | while (index < indirect_blks && count) { | |
610 | new_blocks[index++] = current_block++; | |
611 | count--; | |
612 | } | |
613 | ||
614 | if (count > 0) | |
615 | break; | |
616 | } | |
617 | ||
618 | /* save the new block number for the first direct block */ | |
619 | new_blocks[index] = current_block; | |
620 | ||
621 | /* total number of blocks allocated for direct blocks */ | |
622 | ret = count; | |
623 | *err = 0; | |
624 | return ret; | |
625 | failed_out: | |
626 | for (i = 0; i <index; i++) | |
627 | ext3_free_blocks(handle, inode, new_blocks[i], 1); | |
628 | return ret; | |
629 | } | |
1da177e4 LT |
630 | |
631 | /** | |
632 | * ext3_alloc_branch - allocate and set up a chain of blocks. | |
a4c18ad2 | 633 | * @handle: handle for this transaction |
1da177e4 | 634 | * @inode: owner |
b47b2478 MC |
635 | * @indirect_blks: number of allocated indirect blocks |
636 | * @blks: number of allocated direct blocks | |
a4c18ad2 | 637 | * @goal: preferred place for allocation |
1da177e4 LT |
638 | * @offsets: offsets (in the blocks) to store the pointers to next. |
639 | * @branch: place to store the chain in. | |
640 | * | |
b47b2478 | 641 | * This function allocates blocks, zeroes out all but the last one, |
1da177e4 LT |
642 | * links them into chain and (if we are synchronous) writes them to disk. |
643 | * In other words, it prepares a branch that can be spliced onto the | |
644 | * inode. It stores the information about that chain in the branch[], in | |
645 | * the same format as ext3_get_branch() would do. We are calling it after | |
646 | * we had read the existing part of chain and partial points to the last | |
647 | * triple of that (one with zero ->key). Upon the exit we have the same | |
5b116879 | 648 | * picture as after the successful ext3_get_block(), except that in one |
1da177e4 LT |
649 | * place chain is disconnected - *branch->p is still zero (we did not |
650 | * set the last link), but branch->key contains the number that should | |
651 | * be placed into *branch->p to fill that gap. | |
652 | * | |
653 | * If allocation fails we free all blocks we've allocated (and forget | |
654 | * their buffer_heads) and return the error value the from failed | |
655 | * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain | |
656 | * as described above and return 0. | |
657 | */ | |
1da177e4 | 658 | static int ext3_alloc_branch(handle_t *handle, struct inode *inode, |
43d23f90 | 659 | int indirect_blks, int *blks, ext3_fsblk_t goal, |
b47b2478 | 660 | int *offsets, Indirect *branch) |
1da177e4 LT |
661 | { |
662 | int blocksize = inode->i_sb->s_blocksize; | |
b47b2478 | 663 | int i, n = 0; |
1da177e4 | 664 | int err = 0; |
b47b2478 MC |
665 | struct buffer_head *bh; |
666 | int num; | |
43d23f90 MC |
667 | ext3_fsblk_t new_blocks[4]; |
668 | ext3_fsblk_t current_block; | |
1da177e4 | 669 | |
b47b2478 MC |
670 | num = ext3_alloc_blocks(handle, inode, goal, indirect_blks, |
671 | *blks, new_blocks, &err); | |
672 | if (err) | |
673 | return err; | |
1da177e4 | 674 | |
b47b2478 MC |
675 | branch[0].key = cpu_to_le32(new_blocks[0]); |
676 | /* | |
677 | * metadata blocks and data blocks are allocated. | |
678 | */ | |
679 | for (n = 1; n <= indirect_blks; n++) { | |
680 | /* | |
681 | * Get buffer_head for parent block, zero it out | |
682 | * and set the pointer to new one, then send | |
683 | * parent to disk. | |
684 | */ | |
685 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
686 | branch[n].bh = bh; | |
687 | lock_buffer(bh); | |
688 | BUFFER_TRACE(bh, "call get_create_access"); | |
689 | err = ext3_journal_get_create_access(handle, bh); | |
690 | if (err) { | |
1da177e4 | 691 | unlock_buffer(bh); |
b47b2478 MC |
692 | brelse(bh); |
693 | goto failed; | |
694 | } | |
1da177e4 | 695 | |
b47b2478 MC |
696 | memset(bh->b_data, 0, blocksize); |
697 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
698 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
699 | *branch[n].p = branch[n].key; | |
700 | if ( n == indirect_blks) { | |
701 | current_block = new_blocks[n]; | |
702 | /* | |
703 | * End of chain, update the last new metablock of | |
704 | * the chain to point to the new allocated | |
705 | * data blocks numbers | |
706 | */ | |
707 | for (i=1; i < num; i++) | |
708 | *(branch[n].p + i) = cpu_to_le32(++current_block); | |
1da177e4 | 709 | } |
b47b2478 MC |
710 | BUFFER_TRACE(bh, "marking uptodate"); |
711 | set_buffer_uptodate(bh); | |
712 | unlock_buffer(bh); | |
1da177e4 | 713 | |
b47b2478 MC |
714 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
715 | err = ext3_journal_dirty_metadata(handle, bh); | |
716 | if (err) | |
717 | goto failed; | |
718 | } | |
719 | *blks = num; | |
720 | return err; | |
721 | failed: | |
1da177e4 | 722 | /* Allocation failed, free what we already allocated */ |
b47b2478 | 723 | for (i = 1; i <= n ; i++) { |
1da177e4 LT |
724 | BUFFER_TRACE(branch[i].bh, "call journal_forget"); |
725 | ext3_journal_forget(handle, branch[i].bh); | |
726 | } | |
b47b2478 MC |
727 | for (i = 0; i <indirect_blks; i++) |
728 | ext3_free_blocks(handle, inode, new_blocks[i], 1); | |
729 | ||
730 | ext3_free_blocks(handle, inode, new_blocks[i], num); | |
731 | ||
1da177e4 LT |
732 | return err; |
733 | } | |
734 | ||
735 | /** | |
d6859bfc | 736 | * ext3_splice_branch - splice the allocated branch onto inode. |
a4c18ad2 | 737 | * @handle: handle for this transaction |
d6859bfc AM |
738 | * @inode: owner |
739 | * @block: (logical) number of block we are adding | |
d6859bfc AM |
740 | * @where: location of missing link |
741 | * @num: number of indirect blocks we are adding | |
742 | * @blks: number of direct blocks we are adding | |
743 | * | |
744 | * This function fills the missing link and does all housekeeping needed in | |
745 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
746 | * chain to new block and return 0. | |
1da177e4 | 747 | */ |
d6859bfc AM |
748 | static int ext3_splice_branch(handle_t *handle, struct inode *inode, |
749 | long block, Indirect *where, int num, int blks) | |
1da177e4 LT |
750 | { |
751 | int i; | |
752 | int err = 0; | |
d6859bfc | 753 | struct ext3_block_alloc_info *block_i; |
43d23f90 | 754 | ext3_fsblk_t current_block; |
fe8bc91c | 755 | struct ext3_inode_info *ei = EXT3_I(inode); |
d6859bfc | 756 | |
fe8bc91c | 757 | block_i = ei->i_block_alloc_info; |
1da177e4 LT |
758 | /* |
759 | * If we're splicing into a [td]indirect block (as opposed to the | |
760 | * inode) then we need to get write access to the [td]indirect block | |
761 | * before the splice. | |
762 | */ | |
763 | if (where->bh) { | |
764 | BUFFER_TRACE(where->bh, "get_write_access"); | |
765 | err = ext3_journal_get_write_access(handle, where->bh); | |
766 | if (err) | |
767 | goto err_out; | |
768 | } | |
1da177e4 LT |
769 | /* That's it */ |
770 | ||
771 | *where->p = where->key; | |
d6859bfc AM |
772 | |
773 | /* | |
774 | * Update the host buffer_head or inode to point to more just allocated | |
775 | * direct blocks blocks | |
776 | */ | |
b47b2478 | 777 | if (num == 0 && blks > 1) { |
5dea5176 | 778 | current_block = le32_to_cpu(where->key) + 1; |
b47b2478 MC |
779 | for (i = 1; i < blks; i++) |
780 | *(where->p + i ) = cpu_to_le32(current_block++); | |
781 | } | |
1da177e4 LT |
782 | |
783 | /* | |
784 | * update the most recently allocated logical & physical block | |
785 | * in i_block_alloc_info, to assist find the proper goal block for next | |
786 | * allocation | |
787 | */ | |
788 | if (block_i) { | |
b47b2478 | 789 | block_i->last_alloc_logical_block = block + blks - 1; |
d6859bfc | 790 | block_i->last_alloc_physical_block = |
5dea5176 | 791 | le32_to_cpu(where[num].key) + blks - 1; |
1da177e4 LT |
792 | } |
793 | ||
794 | /* We are done with atomic stuff, now do the rest of housekeeping */ | |
795 | ||
796 | inode->i_ctime = CURRENT_TIME_SEC; | |
797 | ext3_mark_inode_dirty(handle, inode); | |
fe8bc91c JK |
798 | /* ext3_mark_inode_dirty already updated i_sync_tid */ |
799 | atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid); | |
1da177e4 LT |
800 | |
801 | /* had we spliced it onto indirect block? */ | |
802 | if (where->bh) { | |
803 | /* | |
d6859bfc | 804 | * If we spliced it onto an indirect block, we haven't |
1da177e4 LT |
805 | * altered the inode. Note however that if it is being spliced |
806 | * onto an indirect block at the very end of the file (the | |
807 | * file is growing) then we *will* alter the inode to reflect | |
808 | * the new i_size. But that is not done here - it is done in | |
809 | * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode. | |
810 | */ | |
811 | jbd_debug(5, "splicing indirect only\n"); | |
812 | BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata"); | |
813 | err = ext3_journal_dirty_metadata(handle, where->bh); | |
ae6ddcc5 | 814 | if (err) |
1da177e4 LT |
815 | goto err_out; |
816 | } else { | |
817 | /* | |
818 | * OK, we spliced it into the inode itself on a direct block. | |
819 | * Inode was dirtied above. | |
820 | */ | |
821 | jbd_debug(5, "splicing direct\n"); | |
822 | } | |
823 | return err; | |
824 | ||
1da177e4 | 825 | err_out: |
b47b2478 | 826 | for (i = 1; i <= num; i++) { |
1da177e4 LT |
827 | BUFFER_TRACE(where[i].bh, "call journal_forget"); |
828 | ext3_journal_forget(handle, where[i].bh); | |
d6859bfc | 829 | ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1); |
1da177e4 | 830 | } |
b47b2478 MC |
831 | ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks); |
832 | ||
1da177e4 LT |
833 | return err; |
834 | } | |
835 | ||
836 | /* | |
837 | * Allocation strategy is simple: if we have to allocate something, we will | |
838 | * have to go the whole way to leaf. So let's do it before attaching anything | |
839 | * to tree, set linkage between the newborn blocks, write them if sync is | |
840 | * required, recheck the path, free and repeat if check fails, otherwise | |
841 | * set the last missing link (that will protect us from any truncate-generated | |
842 | * removals - all blocks on the path are immune now) and possibly force the | |
843 | * write on the parent block. | |
844 | * That has a nice additional property: no special recovery from the failed | |
845 | * allocations is needed - we simply release blocks and do not touch anything | |
846 | * reachable from inode. | |
847 | * | |
d6859bfc | 848 | * `handle' can be NULL if create == 0. |
1da177e4 LT |
849 | * |
850 | * The BKL may not be held on entry here. Be sure to take it early. | |
89747d36 MC |
851 | * return > 0, # of blocks mapped or allocated. |
852 | * return = 0, if plain lookup failed. | |
853 | * return < 0, error case. | |
1da177e4 | 854 | */ |
d6859bfc AM |
855 | int ext3_get_blocks_handle(handle_t *handle, struct inode *inode, |
856 | sector_t iblock, unsigned long maxblocks, | |
857 | struct buffer_head *bh_result, | |
43237b54 | 858 | int create) |
1da177e4 LT |
859 | { |
860 | int err = -EIO; | |
861 | int offsets[4]; | |
862 | Indirect chain[4]; | |
863 | Indirect *partial; | |
43d23f90 | 864 | ext3_fsblk_t goal; |
b47b2478 | 865 | int indirect_blks; |
89747d36 MC |
866 | int blocks_to_boundary = 0; |
867 | int depth; | |
1da177e4 | 868 | struct ext3_inode_info *ei = EXT3_I(inode); |
89747d36 | 869 | int count = 0; |
43d23f90 | 870 | ext3_fsblk_t first_block = 0; |
89747d36 | 871 | |
1da177e4 | 872 | |
785c4bcc | 873 | trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create); |
1da177e4 | 874 | J_ASSERT(handle != NULL || create == 0); |
d6859bfc | 875 | depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary); |
1da177e4 LT |
876 | |
877 | if (depth == 0) | |
878 | goto out; | |
879 | ||
1da177e4 LT |
880 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
881 | ||
882 | /* Simplest case - block found, no allocation needed */ | |
883 | if (!partial) { | |
5dea5176 | 884 | first_block = le32_to_cpu(chain[depth - 1].key); |
1da177e4 | 885 | clear_buffer_new(bh_result); |
89747d36 MC |
886 | count++; |
887 | /*map more blocks*/ | |
888 | while (count < maxblocks && count <= blocks_to_boundary) { | |
43d23f90 | 889 | ext3_fsblk_t blk; |
5dea5176 | 890 | |
e8ef7aae | 891 | if (!verify_chain(chain, chain + depth - 1)) { |
89747d36 MC |
892 | /* |
893 | * Indirect block might be removed by | |
894 | * truncate while we were reading it. | |
895 | * Handling of that case: forget what we've | |
896 | * got now. Flag the err as EAGAIN, so it | |
897 | * will reread. | |
898 | */ | |
899 | err = -EAGAIN; | |
900 | count = 0; | |
901 | break; | |
902 | } | |
5dea5176 MC |
903 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
904 | ||
905 | if (blk == first_block + count) | |
89747d36 MC |
906 | count++; |
907 | else | |
908 | break; | |
909 | } | |
910 | if (err != -EAGAIN) | |
911 | goto got_it; | |
1da177e4 LT |
912 | } |
913 | ||
914 | /* Next simple case - plain lookup or failed read of indirect block */ | |
fe55c452 MC |
915 | if (!create || err == -EIO) |
916 | goto cleanup; | |
917 | ||
40680f2f JK |
918 | /* |
919 | * Block out ext3_truncate while we alter the tree | |
920 | */ | |
97461518 | 921 | mutex_lock(&ei->truncate_mutex); |
fe55c452 MC |
922 | |
923 | /* | |
924 | * If the indirect block is missing while we are reading | |
925 | * the chain(ext3_get_branch() returns -EAGAIN err), or | |
926 | * if the chain has been changed after we grab the semaphore, | |
927 | * (either because another process truncated this branch, or | |
928 | * another get_block allocated this branch) re-grab the chain to see if | |
929 | * the request block has been allocated or not. | |
930 | * | |
931 | * Since we already block the truncate/other get_block | |
932 | * at this point, we will have the current copy of the chain when we | |
933 | * splice the branch into the tree. | |
934 | */ | |
935 | if (err == -EAGAIN || !verify_chain(chain, partial)) { | |
1da177e4 | 936 | while (partial > chain) { |
1da177e4 LT |
937 | brelse(partial->bh); |
938 | partial--; | |
939 | } | |
fe55c452 MC |
940 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
941 | if (!partial) { | |
89747d36 | 942 | count++; |
97461518 | 943 | mutex_unlock(&ei->truncate_mutex); |
fe55c452 MC |
944 | if (err) |
945 | goto cleanup; | |
946 | clear_buffer_new(bh_result); | |
947 | goto got_it; | |
948 | } | |
1da177e4 LT |
949 | } |
950 | ||
951 | /* | |
fe55c452 MC |
952 | * Okay, we need to do block allocation. Lazily initialize the block |
953 | * allocation info here if necessary | |
954 | */ | |
955 | if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) | |
1da177e4 | 956 | ext3_init_block_alloc_info(inode); |
1da177e4 | 957 | |
fb01bfda | 958 | goal = ext3_find_goal(inode, iblock, partial); |
1da177e4 | 959 | |
b47b2478 MC |
960 | /* the number of blocks need to allocate for [d,t]indirect blocks */ |
961 | indirect_blks = (chain + depth) - partial - 1; | |
1da177e4 | 962 | |
b47b2478 MC |
963 | /* |
964 | * Next look up the indirect map to count the totoal number of | |
965 | * direct blocks to allocate for this branch. | |
966 | */ | |
967 | count = ext3_blks_to_allocate(partial, indirect_blks, | |
968 | maxblocks, blocks_to_boundary); | |
b47b2478 | 969 | err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal, |
fe55c452 | 970 | offsets + (partial - chain), partial); |
1da177e4 | 971 | |
fe55c452 MC |
972 | /* |
973 | * The ext3_splice_branch call will free and forget any buffers | |
1da177e4 LT |
974 | * on the new chain if there is a failure, but that risks using |
975 | * up transaction credits, especially for bitmaps where the | |
976 | * credits cannot be returned. Can we handle this somehow? We | |
fe55c452 MC |
977 | * may need to return -EAGAIN upwards in the worst case. --sct |
978 | */ | |
1da177e4 | 979 | if (!err) |
b47b2478 MC |
980 | err = ext3_splice_branch(handle, inode, iblock, |
981 | partial, indirect_blks, count); | |
97461518 | 982 | mutex_unlock(&ei->truncate_mutex); |
1da177e4 LT |
983 | if (err) |
984 | goto cleanup; | |
985 | ||
986 | set_buffer_new(bh_result); | |
fe55c452 MC |
987 | got_it: |
988 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); | |
20acaa18 | 989 | if (count > blocks_to_boundary) |
fe55c452 | 990 | set_buffer_boundary(bh_result); |
89747d36 | 991 | err = count; |
fe55c452 MC |
992 | /* Clean up and exit */ |
993 | partial = chain + depth - 1; /* the whole chain */ | |
994 | cleanup: | |
1da177e4 | 995 | while (partial > chain) { |
fe55c452 | 996 | BUFFER_TRACE(partial->bh, "call brelse"); |
1da177e4 LT |
997 | brelse(partial->bh); |
998 | partial--; | |
999 | } | |
fe55c452 MC |
1000 | BUFFER_TRACE(bh_result, "returned"); |
1001 | out: | |
785c4bcc LC |
1002 | trace_ext3_get_blocks_exit(inode, iblock, |
1003 | depth ? le32_to_cpu(chain[depth-1].key) : 0, | |
1004 | count, err); | |
fe55c452 | 1005 | return err; |
1da177e4 LT |
1006 | } |
1007 | ||
bd1939de JK |
1008 | /* Maximum number of blocks we map for direct IO at once. */ |
1009 | #define DIO_MAX_BLOCKS 4096 | |
1010 | /* | |
1011 | * Number of credits we need for writing DIO_MAX_BLOCKS: | |
1012 | * We need sb + group descriptor + bitmap + inode -> 4 | |
1013 | * For B blocks with A block pointers per block we need: | |
1014 | * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect). | |
1015 | * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25. | |
1016 | */ | |
1017 | #define DIO_CREDITS 25 | |
1da177e4 | 1018 | |
f91a2ad2 BP |
1019 | static int ext3_get_block(struct inode *inode, sector_t iblock, |
1020 | struct buffer_head *bh_result, int create) | |
1da177e4 | 1021 | { |
3e4fdaf8 | 1022 | handle_t *handle = ext3_journal_current_handle(); |
bd1939de | 1023 | int ret = 0, started = 0; |
1d8fa7a2 | 1024 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
1da177e4 | 1025 | |
bd1939de JK |
1026 | if (create && !handle) { /* Direct IO write... */ |
1027 | if (max_blocks > DIO_MAX_BLOCKS) | |
1028 | max_blocks = DIO_MAX_BLOCKS; | |
1029 | handle = ext3_journal_start(inode, DIO_CREDITS + | |
c459001f | 1030 | EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb)); |
bd1939de | 1031 | if (IS_ERR(handle)) { |
1da177e4 | 1032 | ret = PTR_ERR(handle); |
bd1939de | 1033 | goto out; |
1da177e4 | 1034 | } |
bd1939de | 1035 | started = 1; |
1da177e4 LT |
1036 | } |
1037 | ||
bd1939de | 1038 | ret = ext3_get_blocks_handle(handle, inode, iblock, |
43237b54 | 1039 | max_blocks, bh_result, create); |
bd1939de JK |
1040 | if (ret > 0) { |
1041 | bh_result->b_size = (ret << inode->i_blkbits); | |
1042 | ret = 0; | |
89747d36 | 1043 | } |
bd1939de JK |
1044 | if (started) |
1045 | ext3_journal_stop(handle); | |
1046 | out: | |
1da177e4 LT |
1047 | return ret; |
1048 | } | |
1049 | ||
68c9d702 JB |
1050 | int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
1051 | u64 start, u64 len) | |
1052 | { | |
1053 | return generic_block_fiemap(inode, fieinfo, start, len, | |
1054 | ext3_get_block); | |
1055 | } | |
1056 | ||
1da177e4 LT |
1057 | /* |
1058 | * `handle' can be NULL if create is zero | |
1059 | */ | |
d6859bfc AM |
1060 | struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode, |
1061 | long block, int create, int *errp) | |
1da177e4 LT |
1062 | { |
1063 | struct buffer_head dummy; | |
1064 | int fatal = 0, err; | |
1065 | ||
1066 | J_ASSERT(handle != NULL || create == 0); | |
1067 | ||
1068 | dummy.b_state = 0; | |
1069 | dummy.b_blocknr = -1000; | |
1070 | buffer_trace_init(&dummy.b_history); | |
89747d36 | 1071 | err = ext3_get_blocks_handle(handle, inode, block, 1, |
43237b54 | 1072 | &dummy, create); |
3665d0e5 BP |
1073 | /* |
1074 | * ext3_get_blocks_handle() returns number of blocks | |
1075 | * mapped. 0 in case of a HOLE. | |
1076 | */ | |
1077 | if (err > 0) { | |
1078 | if (err > 1) | |
1079 | WARN_ON(1); | |
89747d36 | 1080 | err = 0; |
89747d36 MC |
1081 | } |
1082 | *errp = err; | |
1083 | if (!err && buffer_mapped(&dummy)) { | |
1da177e4 LT |
1084 | struct buffer_head *bh; |
1085 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); | |
2973dfdb GOC |
1086 | if (!bh) { |
1087 | *errp = -EIO; | |
1088 | goto err; | |
1089 | } | |
1da177e4 LT |
1090 | if (buffer_new(&dummy)) { |
1091 | J_ASSERT(create != 0); | |
c80544dc | 1092 | J_ASSERT(handle != NULL); |
1da177e4 | 1093 | |
d6859bfc AM |
1094 | /* |
1095 | * Now that we do not always journal data, we should | |
1096 | * keep in mind whether this should always journal the | |
1097 | * new buffer as metadata. For now, regular file | |
1098 | * writes use ext3_get_block instead, so it's not a | |
1099 | * problem. | |
1100 | */ | |
1da177e4 LT |
1101 | lock_buffer(bh); |
1102 | BUFFER_TRACE(bh, "call get_create_access"); | |
1103 | fatal = ext3_journal_get_create_access(handle, bh); | |
1104 | if (!fatal && !buffer_uptodate(bh)) { | |
d6859bfc | 1105 | memset(bh->b_data,0,inode->i_sb->s_blocksize); |
1da177e4 LT |
1106 | set_buffer_uptodate(bh); |
1107 | } | |
1108 | unlock_buffer(bh); | |
1109 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
1110 | err = ext3_journal_dirty_metadata(handle, bh); | |
1111 | if (!fatal) | |
1112 | fatal = err; | |
1113 | } else { | |
1114 | BUFFER_TRACE(bh, "not a new buffer"); | |
1115 | } | |
1116 | if (fatal) { | |
1117 | *errp = fatal; | |
1118 | brelse(bh); | |
1119 | bh = NULL; | |
1120 | } | |
1121 | return bh; | |
1122 | } | |
2973dfdb | 1123 | err: |
1da177e4 LT |
1124 | return NULL; |
1125 | } | |
1126 | ||
d6859bfc | 1127 | struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode, |
1da177e4 LT |
1128 | int block, int create, int *err) |
1129 | { | |
1130 | struct buffer_head * bh; | |
1131 | ||
1132 | bh = ext3_getblk(handle, inode, block, create, err); | |
1133 | if (!bh) | |
1134 | return bh; | |
1135 | if (buffer_uptodate(bh)) | |
1136 | return bh; | |
65299a3b | 1137 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
1da177e4 LT |
1138 | wait_on_buffer(bh); |
1139 | if (buffer_uptodate(bh)) | |
1140 | return bh; | |
1141 | put_bh(bh); | |
1142 | *err = -EIO; | |
1143 | return NULL; | |
1144 | } | |
1145 | ||
1146 | static int walk_page_buffers( handle_t *handle, | |
1147 | struct buffer_head *head, | |
1148 | unsigned from, | |
1149 | unsigned to, | |
1150 | int *partial, | |
1151 | int (*fn)( handle_t *handle, | |
1152 | struct buffer_head *bh)) | |
1153 | { | |
1154 | struct buffer_head *bh; | |
1155 | unsigned block_start, block_end; | |
1156 | unsigned blocksize = head->b_size; | |
1157 | int err, ret = 0; | |
1158 | struct buffer_head *next; | |
1159 | ||
1160 | for ( bh = head, block_start = 0; | |
1161 | ret == 0 && (bh != head || !block_start); | |
e9ad5620 | 1162 | block_start = block_end, bh = next) |
1da177e4 LT |
1163 | { |
1164 | next = bh->b_this_page; | |
1165 | block_end = block_start + blocksize; | |
1166 | if (block_end <= from || block_start >= to) { | |
1167 | if (partial && !buffer_uptodate(bh)) | |
1168 | *partial = 1; | |
1169 | continue; | |
1170 | } | |
1171 | err = (*fn)(handle, bh); | |
1172 | if (!ret) | |
1173 | ret = err; | |
1174 | } | |
1175 | return ret; | |
1176 | } | |
1177 | ||
1178 | /* | |
1179 | * To preserve ordering, it is essential that the hole instantiation and | |
1180 | * the data write be encapsulated in a single transaction. We cannot | |
1181 | * close off a transaction and start a new one between the ext3_get_block() | |
1182 | * and the commit_write(). So doing the journal_start at the start of | |
1183 | * prepare_write() is the right place. | |
1184 | * | |
1185 | * Also, this function can nest inside ext3_writepage() -> | |
1186 | * block_write_full_page(). In that case, we *know* that ext3_writepage() | |
1187 | * has generated enough buffer credits to do the whole page. So we won't | |
1188 | * block on the journal in that case, which is good, because the caller may | |
1189 | * be PF_MEMALLOC. | |
1190 | * | |
1191 | * By accident, ext3 can be reentered when a transaction is open via | |
1192 | * quota file writes. If we were to commit the transaction while thus | |
1193 | * reentered, there can be a deadlock - we would be holding a quota | |
1194 | * lock, and the commit would never complete if another thread had a | |
1195 | * transaction open and was blocking on the quota lock - a ranking | |
1196 | * violation. | |
1197 | * | |
1198 | * So what we do is to rely on the fact that journal_stop/journal_start | |
1199 | * will _not_ run commit under these circumstances because handle->h_ref | |
1200 | * is elevated. We'll still have enough credits for the tiny quotafile | |
ae6ddcc5 | 1201 | * write. |
1da177e4 | 1202 | */ |
d6859bfc AM |
1203 | static int do_journal_get_write_access(handle_t *handle, |
1204 | struct buffer_head *bh) | |
1da177e4 | 1205 | { |
5f11e6a4 JK |
1206 | int dirty = buffer_dirty(bh); |
1207 | int ret; | |
1208 | ||
1da177e4 LT |
1209 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1210 | return 0; | |
5f11e6a4 JK |
1211 | /* |
1212 | * __block_prepare_write() could have dirtied some buffers. Clean | |
1213 | * the dirty bit as jbd2_journal_get_write_access() could complain | |
1214 | * otherwise about fs integrity issues. Setting of the dirty bit | |
1215 | * by __block_prepare_write() isn't a real problem here as we clear | |
1216 | * the bit before releasing a page lock and thus writeback cannot | |
1217 | * ever write the buffer. | |
1218 | */ | |
1219 | if (dirty) | |
1220 | clear_buffer_dirty(bh); | |
1221 | ret = ext3_journal_get_write_access(handle, bh); | |
1222 | if (!ret && dirty) | |
1223 | ret = ext3_journal_dirty_metadata(handle, bh); | |
1224 | return ret; | |
1da177e4 LT |
1225 | } |
1226 | ||
68eb3db0 JK |
1227 | /* |
1228 | * Truncate blocks that were not used by write. We have to truncate the | |
1229 | * pagecache as well so that corresponding buffers get properly unmapped. | |
1230 | */ | |
1231 | static void ext3_truncate_failed_write(struct inode *inode) | |
1232 | { | |
1233 | truncate_inode_pages(inode->i_mapping, inode->i_size); | |
1234 | ext3_truncate(inode); | |
1235 | } | |
1236 | ||
ee3e77f1 JK |
1237 | /* |
1238 | * Truncate blocks that were not used by direct IO write. We have to zero out | |
1239 | * the last file block as well because direct IO might have written to it. | |
1240 | */ | |
1241 | static void ext3_truncate_failed_direct_write(struct inode *inode) | |
1242 | { | |
1243 | ext3_block_truncate_page(inode, inode->i_size); | |
1244 | ext3_truncate(inode); | |
1245 | } | |
1246 | ||
f4fc66a8 NP |
1247 | static int ext3_write_begin(struct file *file, struct address_space *mapping, |
1248 | loff_t pos, unsigned len, unsigned flags, | |
1249 | struct page **pagep, void **fsdata) | |
1da177e4 | 1250 | { |
f4fc66a8 | 1251 | struct inode *inode = mapping->host; |
695f6ae0 | 1252 | int ret; |
1da177e4 LT |
1253 | handle_t *handle; |
1254 | int retries = 0; | |
f4fc66a8 NP |
1255 | struct page *page; |
1256 | pgoff_t index; | |
1257 | unsigned from, to; | |
695f6ae0 JK |
1258 | /* Reserve one block more for addition to orphan list in case |
1259 | * we allocate blocks but write fails for some reason */ | |
1260 | int needed_blocks = ext3_writepage_trans_blocks(inode) + 1; | |
f4fc66a8 | 1261 | |
785c4bcc LC |
1262 | trace_ext3_write_begin(inode, pos, len, flags); |
1263 | ||
f4fc66a8 NP |
1264 | index = pos >> PAGE_CACHE_SHIFT; |
1265 | from = pos & (PAGE_CACHE_SIZE - 1); | |
1266 | to = from + len; | |
1da177e4 LT |
1267 | |
1268 | retry: | |
54566b2c | 1269 | page = grab_cache_page_write_begin(mapping, index, flags); |
f4fc66a8 NP |
1270 | if (!page) |
1271 | return -ENOMEM; | |
1272 | *pagep = page; | |
1273 | ||
1da177e4 | 1274 | handle = ext3_journal_start(inode, needed_blocks); |
1aa9b4b9 | 1275 | if (IS_ERR(handle)) { |
f4fc66a8 NP |
1276 | unlock_page(page); |
1277 | page_cache_release(page); | |
1aa9b4b9 AM |
1278 | ret = PTR_ERR(handle); |
1279 | goto out; | |
1280 | } | |
6e1db88d | 1281 | ret = __block_write_begin(page, pos, len, ext3_get_block); |
1da177e4 | 1282 | if (ret) |
f4fc66a8 | 1283 | goto write_begin_failed; |
1da177e4 LT |
1284 | |
1285 | if (ext3_should_journal_data(inode)) { | |
1286 | ret = walk_page_buffers(handle, page_buffers(page), | |
1287 | from, to, NULL, do_journal_get_write_access); | |
1288 | } | |
f4fc66a8 NP |
1289 | write_begin_failed: |
1290 | if (ret) { | |
5ec8b75e AK |
1291 | /* |
1292 | * block_write_begin may have instantiated a few blocks | |
1293 | * outside i_size. Trim these off again. Don't need | |
1294 | * i_size_read because we hold i_mutex. | |
695f6ae0 JK |
1295 | * |
1296 | * Add inode to orphan list in case we crash before truncate | |
9eaaa2d5 JK |
1297 | * finishes. Do this only if ext3_can_truncate() agrees so |
1298 | * that orphan processing code is happy. | |
5ec8b75e | 1299 | */ |
9eaaa2d5 | 1300 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 JK |
1301 | ext3_orphan_add(handle, inode); |
1302 | ext3_journal_stop(handle); | |
1303 | unlock_page(page); | |
1304 | page_cache_release(page); | |
5ec8b75e | 1305 | if (pos + len > inode->i_size) |
68eb3db0 | 1306 | ext3_truncate_failed_write(inode); |
f4fc66a8 | 1307 | } |
1da177e4 LT |
1308 | if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries)) |
1309 | goto retry; | |
1aa9b4b9 | 1310 | out: |
1da177e4 LT |
1311 | return ret; |
1312 | } | |
1313 | ||
f4fc66a8 | 1314 | |
d6859bfc | 1315 | int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh) |
1da177e4 LT |
1316 | { |
1317 | int err = journal_dirty_data(handle, bh); | |
1318 | if (err) | |
e05b6b52 | 1319 | ext3_journal_abort_handle(__func__, __func__, |
f4fc66a8 | 1320 | bh, handle, err); |
1da177e4 LT |
1321 | return err; |
1322 | } | |
1323 | ||
695f6ae0 JK |
1324 | /* For ordered writepage and write_end functions */ |
1325 | static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh) | |
1326 | { | |
1327 | /* | |
1328 | * Write could have mapped the buffer but it didn't copy the data in | |
1329 | * yet. So avoid filing such buffer into a transaction. | |
1330 | */ | |
1331 | if (buffer_mapped(bh) && buffer_uptodate(bh)) | |
1332 | return ext3_journal_dirty_data(handle, bh); | |
1333 | return 0; | |
1334 | } | |
1335 | ||
f4fc66a8 NP |
1336 | /* For write_end() in data=journal mode */ |
1337 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
1da177e4 LT |
1338 | { |
1339 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1340 | return 0; | |
1341 | set_buffer_uptodate(bh); | |
1342 | return ext3_journal_dirty_metadata(handle, bh); | |
1343 | } | |
1344 | ||
f4fc66a8 | 1345 | /* |
695f6ae0 JK |
1346 | * This is nasty and subtle: ext3_write_begin() could have allocated blocks |
1347 | * for the whole page but later we failed to copy the data in. Update inode | |
1348 | * size according to what we managed to copy. The rest is going to be | |
1349 | * truncated in write_end function. | |
f4fc66a8 | 1350 | */ |
695f6ae0 | 1351 | static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied) |
f4fc66a8 | 1352 | { |
695f6ae0 JK |
1353 | /* What matters to us is i_disksize. We don't write i_size anywhere */ |
1354 | if (pos + copied > inode->i_size) | |
1355 | i_size_write(inode, pos + copied); | |
1356 | if (pos + copied > EXT3_I(inode)->i_disksize) { | |
1357 | EXT3_I(inode)->i_disksize = pos + copied; | |
f4fc66a8 NP |
1358 | mark_inode_dirty(inode); |
1359 | } | |
f4fc66a8 NP |
1360 | } |
1361 | ||
1da177e4 LT |
1362 | /* |
1363 | * We need to pick up the new inode size which generic_commit_write gave us | |
1364 | * `file' can be NULL - eg, when called from page_symlink(). | |
1365 | * | |
1366 | * ext3 never places buffers on inode->i_mapping->private_list. metadata | |
1367 | * buffers are managed internally. | |
1368 | */ | |
f4fc66a8 NP |
1369 | static int ext3_ordered_write_end(struct file *file, |
1370 | struct address_space *mapping, | |
1371 | loff_t pos, unsigned len, unsigned copied, | |
1372 | struct page *page, void *fsdata) | |
1da177e4 LT |
1373 | { |
1374 | handle_t *handle = ext3_journal_current_handle(); | |
f4fc66a8 NP |
1375 | struct inode *inode = file->f_mapping->host; |
1376 | unsigned from, to; | |
1da177e4 LT |
1377 | int ret = 0, ret2; |
1378 | ||
785c4bcc | 1379 | trace_ext3_ordered_write_end(inode, pos, len, copied); |
695f6ae0 | 1380 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
f4fc66a8 | 1381 | |
695f6ae0 JK |
1382 | from = pos & (PAGE_CACHE_SIZE - 1); |
1383 | to = from + copied; | |
1da177e4 | 1384 | ret = walk_page_buffers(handle, page_buffers(page), |
695f6ae0 | 1385 | from, to, NULL, journal_dirty_data_fn); |
1da177e4 | 1386 | |
695f6ae0 JK |
1387 | if (ret == 0) |
1388 | update_file_sizes(inode, pos, copied); | |
1389 | /* | |
1390 | * There may be allocated blocks outside of i_size because | |
1391 | * we failed to copy some data. Prepare for truncate. | |
1392 | */ | |
9eaaa2d5 | 1393 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 | 1394 | ext3_orphan_add(handle, inode); |
1da177e4 LT |
1395 | ret2 = ext3_journal_stop(handle); |
1396 | if (!ret) | |
1397 | ret = ret2; | |
f4fc66a8 NP |
1398 | unlock_page(page); |
1399 | page_cache_release(page); | |
1400 | ||
695f6ae0 | 1401 | if (pos + len > inode->i_size) |
68eb3db0 | 1402 | ext3_truncate_failed_write(inode); |
f4fc66a8 | 1403 | return ret ? ret : copied; |
1da177e4 LT |
1404 | } |
1405 | ||
f4fc66a8 NP |
1406 | static int ext3_writeback_write_end(struct file *file, |
1407 | struct address_space *mapping, | |
1408 | loff_t pos, unsigned len, unsigned copied, | |
1409 | struct page *page, void *fsdata) | |
1da177e4 LT |
1410 | { |
1411 | handle_t *handle = ext3_journal_current_handle(); | |
f4fc66a8 | 1412 | struct inode *inode = file->f_mapping->host; |
695f6ae0 | 1413 | int ret; |
1da177e4 | 1414 | |
785c4bcc | 1415 | trace_ext3_writeback_write_end(inode, pos, len, copied); |
695f6ae0 JK |
1416 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
1417 | update_file_sizes(inode, pos, copied); | |
1418 | /* | |
1419 | * There may be allocated blocks outside of i_size because | |
1420 | * we failed to copy some data. Prepare for truncate. | |
1421 | */ | |
9eaaa2d5 | 1422 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 JK |
1423 | ext3_orphan_add(handle, inode); |
1424 | ret = ext3_journal_stop(handle); | |
f4fc66a8 NP |
1425 | unlock_page(page); |
1426 | page_cache_release(page); | |
1427 | ||
695f6ae0 | 1428 | if (pos + len > inode->i_size) |
68eb3db0 | 1429 | ext3_truncate_failed_write(inode); |
f4fc66a8 | 1430 | return ret ? ret : copied; |
1da177e4 LT |
1431 | } |
1432 | ||
f4fc66a8 NP |
1433 | static int ext3_journalled_write_end(struct file *file, |
1434 | struct address_space *mapping, | |
1435 | loff_t pos, unsigned len, unsigned copied, | |
1436 | struct page *page, void *fsdata) | |
1da177e4 LT |
1437 | { |
1438 | handle_t *handle = ext3_journal_current_handle(); | |
f4fc66a8 | 1439 | struct inode *inode = mapping->host; |
b22570d9 | 1440 | struct ext3_inode_info *ei = EXT3_I(inode); |
1da177e4 LT |
1441 | int ret = 0, ret2; |
1442 | int partial = 0; | |
f4fc66a8 | 1443 | unsigned from, to; |
1da177e4 | 1444 | |
785c4bcc | 1445 | trace_ext3_journalled_write_end(inode, pos, len, copied); |
f4fc66a8 NP |
1446 | from = pos & (PAGE_CACHE_SIZE - 1); |
1447 | to = from + len; | |
1448 | ||
1449 | if (copied < len) { | |
1450 | if (!PageUptodate(page)) | |
1451 | copied = 0; | |
695f6ae0 JK |
1452 | page_zero_new_buffers(page, from + copied, to); |
1453 | to = from + copied; | |
f4fc66a8 | 1454 | } |
1da177e4 LT |
1455 | |
1456 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
f4fc66a8 | 1457 | to, &partial, write_end_fn); |
1da177e4 LT |
1458 | if (!partial) |
1459 | SetPageUptodate(page); | |
695f6ae0 JK |
1460 | |
1461 | if (pos + copied > inode->i_size) | |
1462 | i_size_write(inode, pos + copied); | |
1463 | /* | |
1464 | * There may be allocated blocks outside of i_size because | |
1465 | * we failed to copy some data. Prepare for truncate. | |
1466 | */ | |
9eaaa2d5 | 1467 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 | 1468 | ext3_orphan_add(handle, inode); |
9df93939 | 1469 | ext3_set_inode_state(inode, EXT3_STATE_JDATA); |
b22570d9 JK |
1470 | atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid); |
1471 | if (inode->i_size > ei->i_disksize) { | |
1472 | ei->i_disksize = inode->i_size; | |
1da177e4 | 1473 | ret2 = ext3_mark_inode_dirty(handle, inode); |
ae6ddcc5 | 1474 | if (!ret) |
1da177e4 LT |
1475 | ret = ret2; |
1476 | } | |
f4fc66a8 | 1477 | |
1da177e4 LT |
1478 | ret2 = ext3_journal_stop(handle); |
1479 | if (!ret) | |
1480 | ret = ret2; | |
f4fc66a8 NP |
1481 | unlock_page(page); |
1482 | page_cache_release(page); | |
1483 | ||
695f6ae0 | 1484 | if (pos + len > inode->i_size) |
68eb3db0 | 1485 | ext3_truncate_failed_write(inode); |
f4fc66a8 | 1486 | return ret ? ret : copied; |
1da177e4 LT |
1487 | } |
1488 | ||
ae6ddcc5 | 1489 | /* |
1da177e4 LT |
1490 | * bmap() is special. It gets used by applications such as lilo and by |
1491 | * the swapper to find the on-disk block of a specific piece of data. | |
1492 | * | |
1493 | * Naturally, this is dangerous if the block concerned is still in the | |
1494 | * journal. If somebody makes a swapfile on an ext3 data-journaling | |
1495 | * filesystem and enables swap, then they may get a nasty shock when the | |
1496 | * data getting swapped to that swapfile suddenly gets overwritten by | |
1497 | * the original zero's written out previously to the journal and | |
ae6ddcc5 | 1498 | * awaiting writeback in the kernel's buffer cache. |
1da177e4 LT |
1499 | * |
1500 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
ae6ddcc5 | 1501 | * take extra steps to flush any blocks which might be in the cache. |
1da177e4 LT |
1502 | */ |
1503 | static sector_t ext3_bmap(struct address_space *mapping, sector_t block) | |
1504 | { | |
1505 | struct inode *inode = mapping->host; | |
1506 | journal_t *journal; | |
1507 | int err; | |
1508 | ||
9df93939 | 1509 | if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) { |
ae6ddcc5 | 1510 | /* |
1da177e4 LT |
1511 | * This is a REALLY heavyweight approach, but the use of |
1512 | * bmap on dirty files is expected to be extremely rare: | |
1513 | * only if we run lilo or swapon on a freshly made file | |
ae6ddcc5 | 1514 | * do we expect this to happen. |
1da177e4 LT |
1515 | * |
1516 | * (bmap requires CAP_SYS_RAWIO so this does not | |
1517 | * represent an unprivileged user DOS attack --- we'd be | |
1518 | * in trouble if mortal users could trigger this path at | |
ae6ddcc5 | 1519 | * will.) |
1da177e4 LT |
1520 | * |
1521 | * NB. EXT3_STATE_JDATA is not set on files other than | |
1522 | * regular files. If somebody wants to bmap a directory | |
1523 | * or symlink and gets confused because the buffer | |
1524 | * hasn't yet been flushed to disk, they deserve | |
1525 | * everything they get. | |
1526 | */ | |
1527 | ||
9df93939 | 1528 | ext3_clear_inode_state(inode, EXT3_STATE_JDATA); |
1da177e4 LT |
1529 | journal = EXT3_JOURNAL(inode); |
1530 | journal_lock_updates(journal); | |
1531 | err = journal_flush(journal); | |
1532 | journal_unlock_updates(journal); | |
1533 | ||
1534 | if (err) | |
1535 | return 0; | |
1536 | } | |
1537 | ||
1538 | return generic_block_bmap(mapping,block,ext3_get_block); | |
1539 | } | |
1540 | ||
1541 | static int bget_one(handle_t *handle, struct buffer_head *bh) | |
1542 | { | |
1543 | get_bh(bh); | |
1544 | return 0; | |
1545 | } | |
1546 | ||
1547 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1548 | { | |
1549 | put_bh(bh); | |
1550 | return 0; | |
1551 | } | |
1552 | ||
9e80d407 JK |
1553 | static int buffer_unmapped(handle_t *handle, struct buffer_head *bh) |
1554 | { | |
1555 | return !buffer_mapped(bh); | |
1556 | } | |
695f6ae0 | 1557 | |
1da177e4 LT |
1558 | /* |
1559 | * Note that we always start a transaction even if we're not journalling | |
1560 | * data. This is to preserve ordering: any hole instantiation within | |
1561 | * __block_write_full_page -> ext3_get_block() should be journalled | |
1562 | * along with the data so we don't crash and then get metadata which | |
1563 | * refers to old data. | |
1564 | * | |
1565 | * In all journalling modes block_write_full_page() will start the I/O. | |
1566 | * | |
1567 | * Problem: | |
1568 | * | |
1569 | * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1570 | * ext3_writepage() | |
1571 | * | |
1572 | * Similar for: | |
1573 | * | |
1574 | * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ... | |
1575 | * | |
1576 | * Same applies to ext3_get_block(). We will deadlock on various things like | |
97461518 | 1577 | * lock_journal and i_truncate_mutex. |
1da177e4 LT |
1578 | * |
1579 | * Setting PF_MEMALLOC here doesn't work - too many internal memory | |
1580 | * allocations fail. | |
1581 | * | |
1582 | * 16May01: If we're reentered then journal_current_handle() will be | |
1583 | * non-zero. We simply *return*. | |
1584 | * | |
1585 | * 1 July 2001: @@@ FIXME: | |
1586 | * In journalled data mode, a data buffer may be metadata against the | |
1587 | * current transaction. But the same file is part of a shared mapping | |
1588 | * and someone does a writepage() on it. | |
1589 | * | |
1590 | * We will move the buffer onto the async_data list, but *after* it has | |
1591 | * been dirtied. So there's a small window where we have dirty data on | |
1592 | * BJ_Metadata. | |
1593 | * | |
1594 | * Note that this only applies to the last partial page in the file. The | |
1595 | * bit which block_write_full_page() uses prepare/commit for. (That's | |
1596 | * broken code anyway: it's wrong for msync()). | |
1597 | * | |
1598 | * It's a rare case: affects the final partial page, for journalled data | |
1599 | * where the file is subject to bith write() and writepage() in the same | |
1600 | * transction. To fix it we'll need a custom block_write_full_page(). | |
1601 | * We'll probably need that anyway for journalling writepage() output. | |
1602 | * | |
1603 | * We don't honour synchronous mounts for writepage(). That would be | |
1604 | * disastrous. Any write() or metadata operation will sync the fs for | |
1605 | * us. | |
1606 | * | |
1607 | * AKPM2: if all the page's buffers are mapped to disk and !data=journal, | |
1608 | * we don't need to open a transaction here. | |
1609 | */ | |
1610 | static int ext3_ordered_writepage(struct page *page, | |
d6859bfc | 1611 | struct writeback_control *wbc) |
1da177e4 LT |
1612 | { |
1613 | struct inode *inode = page->mapping->host; | |
1614 | struct buffer_head *page_bufs; | |
1615 | handle_t *handle = NULL; | |
1616 | int ret = 0; | |
1617 | int err; | |
1618 | ||
1619 | J_ASSERT(PageLocked(page)); | |
49792c80 | 1620 | WARN_ON_ONCE(IS_RDONLY(inode)); |
1da177e4 LT |
1621 | |
1622 | /* | |
1623 | * We give up here if we're reentered, because it might be for a | |
1624 | * different filesystem. | |
1625 | */ | |
1626 | if (ext3_journal_current_handle()) | |
1627 | goto out_fail; | |
1628 | ||
785c4bcc | 1629 | trace_ext3_ordered_writepage(page); |
9e80d407 JK |
1630 | if (!page_has_buffers(page)) { |
1631 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
1632 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
430db323 JK |
1633 | page_bufs = page_buffers(page); |
1634 | } else { | |
1635 | page_bufs = page_buffers(page); | |
1636 | if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE, | |
1637 | NULL, buffer_unmapped)) { | |
1638 | /* Provide NULL get_block() to catch bugs if buffers | |
1639 | * weren't really mapped */ | |
1640 | return block_write_full_page(page, NULL, wbc); | |
1641 | } | |
9e80d407 | 1642 | } |
1da177e4 LT |
1643 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
1644 | ||
1645 | if (IS_ERR(handle)) { | |
1646 | ret = PTR_ERR(handle); | |
1647 | goto out_fail; | |
1648 | } | |
1649 | ||
1da177e4 LT |
1650 | walk_page_buffers(handle, page_bufs, 0, |
1651 | PAGE_CACHE_SIZE, NULL, bget_one); | |
1652 | ||
1653 | ret = block_write_full_page(page, ext3_get_block, wbc); | |
1654 | ||
1655 | /* | |
1656 | * The page can become unlocked at any point now, and | |
1657 | * truncate can then come in and change things. So we | |
1658 | * can't touch *page from now on. But *page_bufs is | |
1659 | * safe due to elevated refcount. | |
1660 | */ | |
1661 | ||
1662 | /* | |
ae6ddcc5 | 1663 | * And attach them to the current transaction. But only if |
1da177e4 LT |
1664 | * block_write_full_page() succeeded. Otherwise they are unmapped, |
1665 | * and generally junk. | |
1666 | */ | |
1667 | if (ret == 0) { | |
1668 | err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, | |
1669 | NULL, journal_dirty_data_fn); | |
1670 | if (!ret) | |
1671 | ret = err; | |
1672 | } | |
1673 | walk_page_buffers(handle, page_bufs, 0, | |
1674 | PAGE_CACHE_SIZE, NULL, bput_one); | |
1675 | err = ext3_journal_stop(handle); | |
1676 | if (!ret) | |
1677 | ret = err; | |
1678 | return ret; | |
1679 | ||
1680 | out_fail: | |
1681 | redirty_page_for_writepage(wbc, page); | |
1682 | unlock_page(page); | |
1683 | return ret; | |
1684 | } | |
1685 | ||
1da177e4 LT |
1686 | static int ext3_writeback_writepage(struct page *page, |
1687 | struct writeback_control *wbc) | |
1688 | { | |
1689 | struct inode *inode = page->mapping->host; | |
1690 | handle_t *handle = NULL; | |
1691 | int ret = 0; | |
1692 | int err; | |
1693 | ||
49792c80 DM |
1694 | J_ASSERT(PageLocked(page)); |
1695 | WARN_ON_ONCE(IS_RDONLY(inode)); | |
1696 | ||
1da177e4 LT |
1697 | if (ext3_journal_current_handle()) |
1698 | goto out_fail; | |
1699 | ||
785c4bcc | 1700 | trace_ext3_writeback_writepage(page); |
430db323 JK |
1701 | if (page_has_buffers(page)) { |
1702 | if (!walk_page_buffers(NULL, page_buffers(page), 0, | |
1703 | PAGE_CACHE_SIZE, NULL, buffer_unmapped)) { | |
1704 | /* Provide NULL get_block() to catch bugs if buffers | |
1705 | * weren't really mapped */ | |
1706 | return block_write_full_page(page, NULL, wbc); | |
1707 | } | |
1708 | } | |
1709 | ||
1da177e4 LT |
1710 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
1711 | if (IS_ERR(handle)) { | |
1712 | ret = PTR_ERR(handle); | |
1713 | goto out_fail; | |
1714 | } | |
1715 | ||
4c4d3901 | 1716 | ret = block_write_full_page(page, ext3_get_block, wbc); |
1da177e4 LT |
1717 | |
1718 | err = ext3_journal_stop(handle); | |
1719 | if (!ret) | |
1720 | ret = err; | |
1721 | return ret; | |
1722 | ||
1723 | out_fail: | |
1724 | redirty_page_for_writepage(wbc, page); | |
1725 | unlock_page(page); | |
1726 | return ret; | |
1727 | } | |
1728 | ||
1729 | static int ext3_journalled_writepage(struct page *page, | |
1730 | struct writeback_control *wbc) | |
1731 | { | |
1732 | struct inode *inode = page->mapping->host; | |
1733 | handle_t *handle = NULL; | |
1734 | int ret = 0; | |
1735 | int err; | |
1736 | ||
49792c80 DM |
1737 | J_ASSERT(PageLocked(page)); |
1738 | WARN_ON_ONCE(IS_RDONLY(inode)); | |
1739 | ||
1da177e4 LT |
1740 | if (ext3_journal_current_handle()) |
1741 | goto no_write; | |
1742 | ||
785c4bcc | 1743 | trace_ext3_journalled_writepage(page); |
1da177e4 LT |
1744 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
1745 | if (IS_ERR(handle)) { | |
1746 | ret = PTR_ERR(handle); | |
1747 | goto no_write; | |
1748 | } | |
1749 | ||
1750 | if (!page_has_buffers(page) || PageChecked(page)) { | |
1751 | /* | |
1752 | * It's mmapped pagecache. Add buffers and journal it. There | |
1753 | * doesn't seem much point in redirtying the page here. | |
1754 | */ | |
1755 | ClearPageChecked(page); | |
ebdec241 CH |
1756 | ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE, |
1757 | ext3_get_block); | |
ab4eb43c DL |
1758 | if (ret != 0) { |
1759 | ext3_journal_stop(handle); | |
1da177e4 | 1760 | goto out_unlock; |
ab4eb43c | 1761 | } |
1da177e4 LT |
1762 | ret = walk_page_buffers(handle, page_buffers(page), 0, |
1763 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); | |
1764 | ||
1765 | err = walk_page_buffers(handle, page_buffers(page), 0, | |
f4fc66a8 | 1766 | PAGE_CACHE_SIZE, NULL, write_end_fn); |
1da177e4 LT |
1767 | if (ret == 0) |
1768 | ret = err; | |
9df93939 | 1769 | ext3_set_inode_state(inode, EXT3_STATE_JDATA); |
b22570d9 JK |
1770 | atomic_set(&EXT3_I(inode)->i_datasync_tid, |
1771 | handle->h_transaction->t_tid); | |
1da177e4 LT |
1772 | unlock_page(page); |
1773 | } else { | |
1774 | /* | |
1775 | * It may be a page full of checkpoint-mode buffers. We don't | |
1776 | * really know unless we go poke around in the buffer_heads. | |
1777 | * But block_write_full_page will do the right thing. | |
1778 | */ | |
1779 | ret = block_write_full_page(page, ext3_get_block, wbc); | |
1780 | } | |
1781 | err = ext3_journal_stop(handle); | |
1782 | if (!ret) | |
1783 | ret = err; | |
1784 | out: | |
1785 | return ret; | |
1786 | ||
1787 | no_write: | |
1788 | redirty_page_for_writepage(wbc, page); | |
1789 | out_unlock: | |
1790 | unlock_page(page); | |
1791 | goto out; | |
1792 | } | |
1793 | ||
1794 | static int ext3_readpage(struct file *file, struct page *page) | |
1795 | { | |
785c4bcc | 1796 | trace_ext3_readpage(page); |
1da177e4 LT |
1797 | return mpage_readpage(page, ext3_get_block); |
1798 | } | |
1799 | ||
1800 | static int | |
1801 | ext3_readpages(struct file *file, struct address_space *mapping, | |
1802 | struct list_head *pages, unsigned nr_pages) | |
1803 | { | |
1804 | return mpage_readpages(mapping, pages, nr_pages, ext3_get_block); | |
1805 | } | |
1806 | ||
2ff28e22 | 1807 | static void ext3_invalidatepage(struct page *page, unsigned long offset) |
1da177e4 LT |
1808 | { |
1809 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); | |
1810 | ||
785c4bcc LC |
1811 | trace_ext3_invalidatepage(page, offset); |
1812 | ||
1da177e4 LT |
1813 | /* |
1814 | * If it's a full truncate we just forget about the pending dirtying | |
1815 | */ | |
1816 | if (offset == 0) | |
1817 | ClearPageChecked(page); | |
1818 | ||
2ff28e22 | 1819 | journal_invalidatepage(journal, page, offset); |
1da177e4 LT |
1820 | } |
1821 | ||
27496a8c | 1822 | static int ext3_releasepage(struct page *page, gfp_t wait) |
1da177e4 LT |
1823 | { |
1824 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); | |
1825 | ||
785c4bcc | 1826 | trace_ext3_releasepage(page); |
1da177e4 LT |
1827 | WARN_ON(PageChecked(page)); |
1828 | if (!page_has_buffers(page)) | |
1829 | return 0; | |
1830 | return journal_try_to_free_buffers(journal, page, wait); | |
1831 | } | |
1832 | ||
1833 | /* | |
1834 | * If the O_DIRECT write will extend the file then add this inode to the | |
1835 | * orphan list. So recovery will truncate it back to the original size | |
1836 | * if the machine crashes during the write. | |
1837 | * | |
1838 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
bd1939de JK |
1839 | * crashes then stale disk data _may_ be exposed inside the file. But current |
1840 | * VFS code falls back into buffered path in that case so we are safe. | |
1da177e4 LT |
1841 | */ |
1842 | static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb, | |
1843 | const struct iovec *iov, loff_t offset, | |
1844 | unsigned long nr_segs) | |
1845 | { | |
1846 | struct file *file = iocb->ki_filp; | |
1847 | struct inode *inode = file->f_mapping->host; | |
1848 | struct ext3_inode_info *ei = EXT3_I(inode); | |
bd1939de | 1849 | handle_t *handle; |
1da177e4 LT |
1850 | ssize_t ret; |
1851 | int orphan = 0; | |
1852 | size_t count = iov_length(iov, nr_segs); | |
ea0174a7 | 1853 | int retries = 0; |
1da177e4 | 1854 | |
785c4bcc LC |
1855 | trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
1856 | ||
1da177e4 LT |
1857 | if (rw == WRITE) { |
1858 | loff_t final_size = offset + count; | |
1859 | ||
1da177e4 | 1860 | if (final_size > inode->i_size) { |
bd1939de JK |
1861 | /* Credits for sb + inode write */ |
1862 | handle = ext3_journal_start(inode, 2); | |
1863 | if (IS_ERR(handle)) { | |
1864 | ret = PTR_ERR(handle); | |
1865 | goto out; | |
1866 | } | |
1da177e4 | 1867 | ret = ext3_orphan_add(handle, inode); |
bd1939de JK |
1868 | if (ret) { |
1869 | ext3_journal_stop(handle); | |
1870 | goto out; | |
1871 | } | |
1da177e4 LT |
1872 | orphan = 1; |
1873 | ei->i_disksize = inode->i_size; | |
bd1939de | 1874 | ext3_journal_stop(handle); |
1da177e4 LT |
1875 | } |
1876 | } | |
1877 | ||
ea0174a7 | 1878 | retry: |
aacfc19c CH |
1879 | ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs, |
1880 | ext3_get_block); | |
eafdc7d1 CH |
1881 | /* |
1882 | * In case of error extending write may have instantiated a few | |
1883 | * blocks outside i_size. Trim these off again. | |
1884 | */ | |
1885 | if (unlikely((rw & WRITE) && ret < 0)) { | |
1886 | loff_t isize = i_size_read(inode); | |
1887 | loff_t end = offset + iov_length(iov, nr_segs); | |
1888 | ||
1889 | if (end > isize) | |
ee3e77f1 | 1890 | ext3_truncate_failed_direct_write(inode); |
eafdc7d1 | 1891 | } |
ea0174a7 ES |
1892 | if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries)) |
1893 | goto retry; | |
1da177e4 | 1894 | |
bd1939de | 1895 | if (orphan) { |
1da177e4 LT |
1896 | int err; |
1897 | ||
bd1939de JK |
1898 | /* Credits for sb + inode write */ |
1899 | handle = ext3_journal_start(inode, 2); | |
1900 | if (IS_ERR(handle)) { | |
1901 | /* This is really bad luck. We've written the data | |
7eb4969e JK |
1902 | * but cannot extend i_size. Truncate allocated blocks |
1903 | * and pretend the write failed... */ | |
ee3e77f1 | 1904 | ext3_truncate_failed_direct_write(inode); |
bd1939de JK |
1905 | ret = PTR_ERR(handle); |
1906 | goto out; | |
1907 | } | |
1908 | if (inode->i_nlink) | |
1da177e4 | 1909 | ext3_orphan_del(handle, inode); |
bd1939de | 1910 | if (ret > 0) { |
1da177e4 LT |
1911 | loff_t end = offset + ret; |
1912 | if (end > inode->i_size) { | |
1913 | ei->i_disksize = end; | |
1914 | i_size_write(inode, end); | |
1915 | /* | |
1916 | * We're going to return a positive `ret' | |
1917 | * here due to non-zero-length I/O, so there's | |
1918 | * no way of reporting error returns from | |
1919 | * ext3_mark_inode_dirty() to userspace. So | |
1920 | * ignore it. | |
1921 | */ | |
1922 | ext3_mark_inode_dirty(handle, inode); | |
1923 | } | |
1924 | } | |
1925 | err = ext3_journal_stop(handle); | |
1926 | if (ret == 0) | |
1927 | ret = err; | |
1928 | } | |
1929 | out: | |
785c4bcc LC |
1930 | trace_ext3_direct_IO_exit(inode, offset, |
1931 | iov_length(iov, nr_segs), rw, ret); | |
1da177e4 LT |
1932 | return ret; |
1933 | } | |
1934 | ||
1935 | /* | |
1936 | * Pages can be marked dirty completely asynchronously from ext3's journalling | |
1937 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do | |
1938 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
1939 | * not necessarily locked. | |
1940 | * | |
1941 | * We cannot just dirty the page and leave attached buffers clean, because the | |
1942 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
1943 | * or jbddirty because all the journalling code will explode. | |
1944 | * | |
1945 | * So what we do is to mark the page "pending dirty" and next time writepage | |
1946 | * is called, propagate that into the buffers appropriately. | |
1947 | */ | |
1948 | static int ext3_journalled_set_page_dirty(struct page *page) | |
1949 | { | |
1950 | SetPageChecked(page); | |
1951 | return __set_page_dirty_nobuffers(page); | |
1952 | } | |
1953 | ||
f5e54d6e | 1954 | static const struct address_space_operations ext3_ordered_aops = { |
8ab22b9a HH |
1955 | .readpage = ext3_readpage, |
1956 | .readpages = ext3_readpages, | |
1957 | .writepage = ext3_ordered_writepage, | |
8ab22b9a HH |
1958 | .write_begin = ext3_write_begin, |
1959 | .write_end = ext3_ordered_write_end, | |
1960 | .bmap = ext3_bmap, | |
1961 | .invalidatepage = ext3_invalidatepage, | |
1962 | .releasepage = ext3_releasepage, | |
1963 | .direct_IO = ext3_direct_IO, | |
1964 | .migratepage = buffer_migrate_page, | |
1965 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 1966 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
1967 | }; |
1968 | ||
f5e54d6e | 1969 | static const struct address_space_operations ext3_writeback_aops = { |
8ab22b9a HH |
1970 | .readpage = ext3_readpage, |
1971 | .readpages = ext3_readpages, | |
1972 | .writepage = ext3_writeback_writepage, | |
8ab22b9a HH |
1973 | .write_begin = ext3_write_begin, |
1974 | .write_end = ext3_writeback_write_end, | |
1975 | .bmap = ext3_bmap, | |
1976 | .invalidatepage = ext3_invalidatepage, | |
1977 | .releasepage = ext3_releasepage, | |
1978 | .direct_IO = ext3_direct_IO, | |
1979 | .migratepage = buffer_migrate_page, | |
1980 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 1981 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
1982 | }; |
1983 | ||
f5e54d6e | 1984 | static const struct address_space_operations ext3_journalled_aops = { |
8ab22b9a HH |
1985 | .readpage = ext3_readpage, |
1986 | .readpages = ext3_readpages, | |
1987 | .writepage = ext3_journalled_writepage, | |
8ab22b9a HH |
1988 | .write_begin = ext3_write_begin, |
1989 | .write_end = ext3_journalled_write_end, | |
1990 | .set_page_dirty = ext3_journalled_set_page_dirty, | |
1991 | .bmap = ext3_bmap, | |
1992 | .invalidatepage = ext3_invalidatepage, | |
1993 | .releasepage = ext3_releasepage, | |
1994 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 1995 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
1996 | }; |
1997 | ||
1998 | void ext3_set_aops(struct inode *inode) | |
1999 | { | |
2000 | if (ext3_should_order_data(inode)) | |
2001 | inode->i_mapping->a_ops = &ext3_ordered_aops; | |
2002 | else if (ext3_should_writeback_data(inode)) | |
2003 | inode->i_mapping->a_ops = &ext3_writeback_aops; | |
2004 | else | |
2005 | inode->i_mapping->a_ops = &ext3_journalled_aops; | |
2006 | } | |
2007 | ||
2008 | /* | |
2009 | * ext3_block_truncate_page() zeroes out a mapping from file offset `from' | |
2010 | * up to the end of the block which corresponds to `from'. | |
2011 | * This required during truncate. We need to physically zero the tail end | |
2012 | * of that block so it doesn't yield old data if the file is later grown. | |
2013 | */ | |
ee3e77f1 | 2014 | static int ext3_block_truncate_page(struct inode *inode, loff_t from) |
1da177e4 | 2015 | { |
43d23f90 | 2016 | ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ee3e77f1 | 2017 | unsigned offset = from & (PAGE_CACHE_SIZE - 1); |
1da177e4 | 2018 | unsigned blocksize, iblock, length, pos; |
ee3e77f1 JK |
2019 | struct page *page; |
2020 | handle_t *handle = NULL; | |
1da177e4 LT |
2021 | struct buffer_head *bh; |
2022 | int err = 0; | |
1da177e4 | 2023 | |
ee3e77f1 | 2024 | /* Truncated on block boundary - nothing to do */ |
1da177e4 | 2025 | blocksize = inode->i_sb->s_blocksize; |
ee3e77f1 JK |
2026 | if ((from & (blocksize - 1)) == 0) |
2027 | return 0; | |
2028 | ||
2029 | page = grab_cache_page(inode->i_mapping, index); | |
2030 | if (!page) | |
2031 | return -ENOMEM; | |
1da177e4 LT |
2032 | length = blocksize - (offset & (blocksize - 1)); |
2033 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
2034 | ||
1da177e4 LT |
2035 | if (!page_has_buffers(page)) |
2036 | create_empty_buffers(page, blocksize, 0); | |
2037 | ||
2038 | /* Find the buffer that contains "offset" */ | |
2039 | bh = page_buffers(page); | |
2040 | pos = blocksize; | |
2041 | while (offset >= pos) { | |
2042 | bh = bh->b_this_page; | |
2043 | iblock++; | |
2044 | pos += blocksize; | |
2045 | } | |
2046 | ||
2047 | err = 0; | |
2048 | if (buffer_freed(bh)) { | |
2049 | BUFFER_TRACE(bh, "freed: skip"); | |
2050 | goto unlock; | |
2051 | } | |
2052 | ||
2053 | if (!buffer_mapped(bh)) { | |
2054 | BUFFER_TRACE(bh, "unmapped"); | |
2055 | ext3_get_block(inode, iblock, bh, 0); | |
2056 | /* unmapped? It's a hole - nothing to do */ | |
2057 | if (!buffer_mapped(bh)) { | |
2058 | BUFFER_TRACE(bh, "still unmapped"); | |
2059 | goto unlock; | |
2060 | } | |
2061 | } | |
2062 | ||
2063 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2064 | if (PageUptodate(page)) | |
2065 | set_buffer_uptodate(bh); | |
2066 | ||
2067 | if (!buffer_uptodate(bh)) { | |
2068 | err = -EIO; | |
2069 | ll_rw_block(READ, 1, &bh); | |
2070 | wait_on_buffer(bh); | |
2071 | /* Uhhuh. Read error. Complain and punt. */ | |
2072 | if (!buffer_uptodate(bh)) | |
2073 | goto unlock; | |
2074 | } | |
2075 | ||
ee3e77f1 JK |
2076 | /* data=writeback mode doesn't need transaction to zero-out data */ |
2077 | if (!ext3_should_writeback_data(inode)) { | |
2078 | /* We journal at most one block */ | |
2079 | handle = ext3_journal_start(inode, 1); | |
2080 | if (IS_ERR(handle)) { | |
2081 | clear_highpage(page); | |
2082 | flush_dcache_page(page); | |
2083 | err = PTR_ERR(handle); | |
2084 | goto unlock; | |
2085 | } | |
2086 | } | |
2087 | ||
1da177e4 LT |
2088 | if (ext3_should_journal_data(inode)) { |
2089 | BUFFER_TRACE(bh, "get write access"); | |
2090 | err = ext3_journal_get_write_access(handle, bh); | |
2091 | if (err) | |
ee3e77f1 | 2092 | goto stop; |
1da177e4 LT |
2093 | } |
2094 | ||
eebd2aa3 | 2095 | zero_user(page, offset, length); |
1da177e4 LT |
2096 | BUFFER_TRACE(bh, "zeroed end of block"); |
2097 | ||
2098 | err = 0; | |
2099 | if (ext3_should_journal_data(inode)) { | |
2100 | err = ext3_journal_dirty_metadata(handle, bh); | |
2101 | } else { | |
2102 | if (ext3_should_order_data(inode)) | |
2103 | err = ext3_journal_dirty_data(handle, bh); | |
2104 | mark_buffer_dirty(bh); | |
2105 | } | |
ee3e77f1 JK |
2106 | stop: |
2107 | if (handle) | |
2108 | ext3_journal_stop(handle); | |
1da177e4 LT |
2109 | |
2110 | unlock: | |
2111 | unlock_page(page); | |
2112 | page_cache_release(page); | |
2113 | return err; | |
2114 | } | |
2115 | ||
2116 | /* | |
2117 | * Probably it should be a library function... search for first non-zero word | |
2118 | * or memcmp with zero_page, whatever is better for particular architecture. | |
2119 | * Linus? | |
2120 | */ | |
2121 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
2122 | { | |
2123 | while (p < q) | |
2124 | if (*p++) | |
2125 | return 0; | |
2126 | return 1; | |
2127 | } | |
2128 | ||
2129 | /** | |
2130 | * ext3_find_shared - find the indirect blocks for partial truncation. | |
2131 | * @inode: inode in question | |
2132 | * @depth: depth of the affected branch | |
2133 | * @offsets: offsets of pointers in that branch (see ext3_block_to_path) | |
2134 | * @chain: place to store the pointers to partial indirect blocks | |
2135 | * @top: place to the (detached) top of branch | |
2136 | * | |
2137 | * This is a helper function used by ext3_truncate(). | |
2138 | * | |
2139 | * When we do truncate() we may have to clean the ends of several | |
2140 | * indirect blocks but leave the blocks themselves alive. Block is | |
25985edc | 2141 | * partially truncated if some data below the new i_size is referred |
1da177e4 LT |
2142 | * from it (and it is on the path to the first completely truncated |
2143 | * data block, indeed). We have to free the top of that path along | |
2144 | * with everything to the right of the path. Since no allocation | |
2145 | * past the truncation point is possible until ext3_truncate() | |
2146 | * finishes, we may safely do the latter, but top of branch may | |
2147 | * require special attention - pageout below the truncation point | |
2148 | * might try to populate it. | |
2149 | * | |
2150 | * We atomically detach the top of branch from the tree, store the | |
2151 | * block number of its root in *@top, pointers to buffer_heads of | |
2152 | * partially truncated blocks - in @chain[].bh and pointers to | |
2153 | * their last elements that should not be removed - in | |
2154 | * @chain[].p. Return value is the pointer to last filled element | |
2155 | * of @chain. | |
2156 | * | |
2157 | * The work left to caller to do the actual freeing of subtrees: | |
2158 | * a) free the subtree starting from *@top | |
2159 | * b) free the subtrees whose roots are stored in | |
2160 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
2161 | * c) free the subtrees growing from the inode past the @chain[0]. | |
2162 | * (no partially truncated stuff there). */ | |
2163 | ||
d6859bfc AM |
2164 | static Indirect *ext3_find_shared(struct inode *inode, int depth, |
2165 | int offsets[4], Indirect chain[4], __le32 *top) | |
1da177e4 LT |
2166 | { |
2167 | Indirect *partial, *p; | |
2168 | int k, err; | |
2169 | ||
2170 | *top = 0; | |
bf48aabb | 2171 | /* Make k index the deepest non-null offset + 1 */ |
1da177e4 LT |
2172 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
2173 | ; | |
2174 | partial = ext3_get_branch(inode, k, offsets, chain, &err); | |
2175 | /* Writer: pointers */ | |
2176 | if (!partial) | |
2177 | partial = chain + k-1; | |
2178 | /* | |
2179 | * If the branch acquired continuation since we've looked at it - | |
2180 | * fine, it should all survive and (new) top doesn't belong to us. | |
2181 | */ | |
2182 | if (!partial->key && *partial->p) | |
2183 | /* Writer: end */ | |
2184 | goto no_top; | |
2185 | for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) | |
2186 | ; | |
2187 | /* | |
2188 | * OK, we've found the last block that must survive. The rest of our | |
2189 | * branch should be detached before unlocking. However, if that rest | |
2190 | * of branch is all ours and does not grow immediately from the inode | |
2191 | * it's easier to cheat and just decrement partial->p. | |
2192 | */ | |
2193 | if (p == chain + k - 1 && p > chain) { | |
2194 | p->p--; | |
2195 | } else { | |
2196 | *top = *p->p; | |
2197 | /* Nope, don't do this in ext3. Must leave the tree intact */ | |
2198 | #if 0 | |
2199 | *p->p = 0; | |
2200 | #endif | |
2201 | } | |
2202 | /* Writer: end */ | |
2203 | ||
d6859bfc | 2204 | while(partial > p) { |
1da177e4 LT |
2205 | brelse(partial->bh); |
2206 | partial--; | |
2207 | } | |
2208 | no_top: | |
2209 | return partial; | |
2210 | } | |
2211 | ||
2212 | /* | |
2213 | * Zero a number of block pointers in either an inode or an indirect block. | |
2214 | * If we restart the transaction we must again get write access to the | |
2215 | * indirect block for further modification. | |
2216 | * | |
2217 | * We release `count' blocks on disk, but (last - first) may be greater | |
2218 | * than `count' because there can be holes in there. | |
2219 | */ | |
d6859bfc | 2220 | static void ext3_clear_blocks(handle_t *handle, struct inode *inode, |
43d23f90 | 2221 | struct buffer_head *bh, ext3_fsblk_t block_to_free, |
d6859bfc | 2222 | unsigned long count, __le32 *first, __le32 *last) |
1da177e4 LT |
2223 | { |
2224 | __le32 *p; | |
2225 | if (try_to_extend_transaction(handle, inode)) { | |
2226 | if (bh) { | |
2227 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
156e7431 NK |
2228 | if (ext3_journal_dirty_metadata(handle, bh)) |
2229 | return; | |
1da177e4 LT |
2230 | } |
2231 | ext3_mark_inode_dirty(handle, inode); | |
00171d3c | 2232 | truncate_restart_transaction(handle, inode); |
1da177e4 LT |
2233 | if (bh) { |
2234 | BUFFER_TRACE(bh, "retaking write access"); | |
156e7431 NK |
2235 | if (ext3_journal_get_write_access(handle, bh)) |
2236 | return; | |
1da177e4 LT |
2237 | } |
2238 | } | |
2239 | ||
2240 | /* | |
2241 | * Any buffers which are on the journal will be in memory. We find | |
2242 | * them on the hash table so journal_revoke() will run journal_forget() | |
2243 | * on them. We've already detached each block from the file, so | |
2244 | * bforget() in journal_forget() should be safe. | |
2245 | * | |
2246 | * AKPM: turn on bforget in journal_forget()!!! | |
2247 | */ | |
2248 | for (p = first; p < last; p++) { | |
2249 | u32 nr = le32_to_cpu(*p); | |
2250 | if (nr) { | |
2251 | struct buffer_head *bh; | |
2252 | ||
2253 | *p = 0; | |
2254 | bh = sb_find_get_block(inode->i_sb, nr); | |
2255 | ext3_forget(handle, 0, inode, bh, nr); | |
2256 | } | |
2257 | } | |
2258 | ||
2259 | ext3_free_blocks(handle, inode, block_to_free, count); | |
2260 | } | |
2261 | ||
2262 | /** | |
2263 | * ext3_free_data - free a list of data blocks | |
2264 | * @handle: handle for this transaction | |
2265 | * @inode: inode we are dealing with | |
2266 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
2267 | * @first: array of block numbers | |
2268 | * @last: points immediately past the end of array | |
2269 | * | |
25985edc | 2270 | * We are freeing all blocks referred from that array (numbers are stored as |
1da177e4 LT |
2271 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. |
2272 | * | |
2273 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
2274 | * blocks are contiguous then releasing them at one time will only affect one | |
2275 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
2276 | * actually use a lot of journal space. | |
2277 | * | |
2278 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
2279 | * block pointers. | |
2280 | */ | |
2281 | static void ext3_free_data(handle_t *handle, struct inode *inode, | |
2282 | struct buffer_head *this_bh, | |
2283 | __le32 *first, __le32 *last) | |
2284 | { | |
43d23f90 | 2285 | ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ae6ddcc5 | 2286 | unsigned long count = 0; /* Number of blocks in the run */ |
1da177e4 LT |
2287 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind |
2288 | corresponding to | |
2289 | block_to_free */ | |
43d23f90 | 2290 | ext3_fsblk_t nr; /* Current block # */ |
1da177e4 LT |
2291 | __le32 *p; /* Pointer into inode/ind |
2292 | for current block */ | |
2293 | int err; | |
2294 | ||
2295 | if (this_bh) { /* For indirect block */ | |
2296 | BUFFER_TRACE(this_bh, "get_write_access"); | |
2297 | err = ext3_journal_get_write_access(handle, this_bh); | |
2298 | /* Important: if we can't update the indirect pointers | |
2299 | * to the blocks, we can't free them. */ | |
2300 | if (err) | |
2301 | return; | |
2302 | } | |
2303 | ||
2304 | for (p = first; p < last; p++) { | |
2305 | nr = le32_to_cpu(*p); | |
2306 | if (nr) { | |
2307 | /* accumulate blocks to free if they're contiguous */ | |
2308 | if (count == 0) { | |
2309 | block_to_free = nr; | |
2310 | block_to_free_p = p; | |
2311 | count = 1; | |
2312 | } else if (nr == block_to_free + count) { | |
2313 | count++; | |
2314 | } else { | |
ae6ddcc5 | 2315 | ext3_clear_blocks(handle, inode, this_bh, |
1da177e4 LT |
2316 | block_to_free, |
2317 | count, block_to_free_p, p); | |
2318 | block_to_free = nr; | |
2319 | block_to_free_p = p; | |
2320 | count = 1; | |
2321 | } | |
2322 | } | |
2323 | } | |
2324 | ||
2325 | if (count > 0) | |
2326 | ext3_clear_blocks(handle, inode, this_bh, block_to_free, | |
2327 | count, block_to_free_p, p); | |
2328 | ||
2329 | if (this_bh) { | |
2330 | BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata"); | |
3ccc3167 DG |
2331 | |
2332 | /* | |
2333 | * The buffer head should have an attached journal head at this | |
2334 | * point. However, if the data is corrupted and an indirect | |
2335 | * block pointed to itself, it would have been detached when | |
2336 | * the block was cleared. Check for this instead of OOPSing. | |
2337 | */ | |
2338 | if (bh2jh(this_bh)) | |
2339 | ext3_journal_dirty_metadata(handle, this_bh); | |
2340 | else | |
2341 | ext3_error(inode->i_sb, "ext3_free_data", | |
2342 | "circular indirect block detected, " | |
2343 | "inode=%lu, block=%llu", | |
2344 | inode->i_ino, | |
2345 | (unsigned long long)this_bh->b_blocknr); | |
1da177e4 LT |
2346 | } |
2347 | } | |
2348 | ||
2349 | /** | |
2350 | * ext3_free_branches - free an array of branches | |
2351 | * @handle: JBD handle for this transaction | |
2352 | * @inode: inode we are dealing with | |
2353 | * @parent_bh: the buffer_head which contains *@first and *@last | |
2354 | * @first: array of block numbers | |
2355 | * @last: pointer immediately past the end of array | |
2356 | * @depth: depth of the branches to free | |
2357 | * | |
25985edc | 2358 | * We are freeing all blocks referred from these branches (numbers are |
1da177e4 LT |
2359 | * stored as little-endian 32-bit) and updating @inode->i_blocks |
2360 | * appropriately. | |
2361 | */ | |
2362 | static void ext3_free_branches(handle_t *handle, struct inode *inode, | |
2363 | struct buffer_head *parent_bh, | |
2364 | __le32 *first, __le32 *last, int depth) | |
2365 | { | |
43d23f90 | 2366 | ext3_fsblk_t nr; |
1da177e4 LT |
2367 | __le32 *p; |
2368 | ||
2369 | if (is_handle_aborted(handle)) | |
2370 | return; | |
2371 | ||
2372 | if (depth--) { | |
2373 | struct buffer_head *bh; | |
2374 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); | |
2375 | p = last; | |
2376 | while (--p >= first) { | |
2377 | nr = le32_to_cpu(*p); | |
2378 | if (!nr) | |
2379 | continue; /* A hole */ | |
2380 | ||
2381 | /* Go read the buffer for the next level down */ | |
2382 | bh = sb_bread(inode->i_sb, nr); | |
2383 | ||
2384 | /* | |
2385 | * A read failure? Report error and clear slot | |
2386 | * (should be rare). | |
2387 | */ | |
2388 | if (!bh) { | |
2389 | ext3_error(inode->i_sb, "ext3_free_branches", | |
eee194e7 | 2390 | "Read failure, inode=%lu, block="E3FSBLK, |
1da177e4 LT |
2391 | inode->i_ino, nr); |
2392 | continue; | |
2393 | } | |
2394 | ||
2395 | /* This zaps the entire block. Bottom up. */ | |
2396 | BUFFER_TRACE(bh, "free child branches"); | |
2397 | ext3_free_branches(handle, inode, bh, | |
2398 | (__le32*)bh->b_data, | |
2399 | (__le32*)bh->b_data + addr_per_block, | |
2400 | depth); | |
2401 | ||
1da177e4 LT |
2402 | /* |
2403 | * Everything below this this pointer has been | |
2404 | * released. Now let this top-of-subtree go. | |
2405 | * | |
2406 | * We want the freeing of this indirect block to be | |
2407 | * atomic in the journal with the updating of the | |
2408 | * bitmap block which owns it. So make some room in | |
2409 | * the journal. | |
2410 | * | |
2411 | * We zero the parent pointer *after* freeing its | |
2412 | * pointee in the bitmaps, so if extend_transaction() | |
2413 | * for some reason fails to put the bitmap changes and | |
2414 | * the release into the same transaction, recovery | |
2415 | * will merely complain about releasing a free block, | |
2416 | * rather than leaking blocks. | |
2417 | */ | |
2418 | if (is_handle_aborted(handle)) | |
2419 | return; | |
2420 | if (try_to_extend_transaction(handle, inode)) { | |
2421 | ext3_mark_inode_dirty(handle, inode); | |
00171d3c | 2422 | truncate_restart_transaction(handle, inode); |
1da177e4 LT |
2423 | } |
2424 | ||
f25f6242 JK |
2425 | /* |
2426 | * We've probably journalled the indirect block several | |
2427 | * times during the truncate. But it's no longer | |
2428 | * needed and we now drop it from the transaction via | |
2429 | * journal_revoke(). | |
2430 | * | |
2431 | * That's easy if it's exclusively part of this | |
2432 | * transaction. But if it's part of the committing | |
2433 | * transaction then journal_forget() will simply | |
2434 | * brelse() it. That means that if the underlying | |
2435 | * block is reallocated in ext3_get_block(), | |
2436 | * unmap_underlying_metadata() will find this block | |
2437 | * and will try to get rid of it. damn, damn. Thus | |
2438 | * we don't allow a block to be reallocated until | |
2439 | * a transaction freeing it has fully committed. | |
2440 | * | |
2441 | * We also have to make sure journal replay after a | |
2442 | * crash does not overwrite non-journaled data blocks | |
2443 | * with old metadata when the block got reallocated for | |
2444 | * data. Thus we have to store a revoke record for a | |
2445 | * block in the same transaction in which we free the | |
2446 | * block. | |
2447 | */ | |
2448 | ext3_forget(handle, 1, inode, bh, bh->b_blocknr); | |
2449 | ||
1da177e4 LT |
2450 | ext3_free_blocks(handle, inode, nr, 1); |
2451 | ||
2452 | if (parent_bh) { | |
2453 | /* | |
2454 | * The block which we have just freed is | |
2455 | * pointed to by an indirect block: journal it | |
2456 | */ | |
2457 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
2458 | if (!ext3_journal_get_write_access(handle, | |
2459 | parent_bh)){ | |
2460 | *p = 0; | |
2461 | BUFFER_TRACE(parent_bh, | |
2462 | "call ext3_journal_dirty_metadata"); | |
ae6ddcc5 | 2463 | ext3_journal_dirty_metadata(handle, |
1da177e4 LT |
2464 | parent_bh); |
2465 | } | |
2466 | } | |
2467 | } | |
2468 | } else { | |
2469 | /* We have reached the bottom of the tree. */ | |
2470 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
2471 | ext3_free_data(handle, inode, parent_bh, first, last); | |
2472 | } | |
2473 | } | |
2474 | ||
ae76dd9a DG |
2475 | int ext3_can_truncate(struct inode *inode) |
2476 | { | |
ae76dd9a DG |
2477 | if (S_ISREG(inode->i_mode)) |
2478 | return 1; | |
2479 | if (S_ISDIR(inode->i_mode)) | |
2480 | return 1; | |
2481 | if (S_ISLNK(inode->i_mode)) | |
2482 | return !ext3_inode_is_fast_symlink(inode); | |
2483 | return 0; | |
2484 | } | |
2485 | ||
1da177e4 LT |
2486 | /* |
2487 | * ext3_truncate() | |
2488 | * | |
2489 | * We block out ext3_get_block() block instantiations across the entire | |
2490 | * transaction, and VFS/VM ensures that ext3_truncate() cannot run | |
2491 | * simultaneously on behalf of the same inode. | |
2492 | * | |
2493 | * As we work through the truncate and commmit bits of it to the journal there | |
2494 | * is one core, guiding principle: the file's tree must always be consistent on | |
2495 | * disk. We must be able to restart the truncate after a crash. | |
2496 | * | |
2497 | * The file's tree may be transiently inconsistent in memory (although it | |
2498 | * probably isn't), but whenever we close off and commit a journal transaction, | |
2499 | * the contents of (the filesystem + the journal) must be consistent and | |
2500 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
2501 | * left-to-right works OK too). | |
2502 | * | |
2503 | * Note that at recovery time, journal replay occurs *before* the restart of | |
2504 | * truncate against the orphan inode list. | |
2505 | * | |
2506 | * The committed inode has the new, desired i_size (which is the same as | |
2507 | * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see | |
2508 | * that this inode's truncate did not complete and it will again call | |
2509 | * ext3_truncate() to have another go. So there will be instantiated blocks | |
2510 | * to the right of the truncation point in a crashed ext3 filesystem. But | |
2511 | * that's fine - as long as they are linked from the inode, the post-crash | |
2512 | * ext3_truncate() run will find them and release them. | |
2513 | */ | |
d6859bfc | 2514 | void ext3_truncate(struct inode *inode) |
1da177e4 LT |
2515 | { |
2516 | handle_t *handle; | |
2517 | struct ext3_inode_info *ei = EXT3_I(inode); | |
2518 | __le32 *i_data = ei->i_data; | |
2519 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); | |
1da177e4 LT |
2520 | int offsets[4]; |
2521 | Indirect chain[4]; | |
2522 | Indirect *partial; | |
2523 | __le32 nr = 0; | |
2524 | int n; | |
2525 | long last_block; | |
2526 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1da177e4 | 2527 | |
785c4bcc | 2528 | trace_ext3_truncate_enter(inode); |
1da177e4 | 2529 | |
ae76dd9a | 2530 | if (!ext3_can_truncate(inode)) |
ef43618a | 2531 | goto out_notrans; |
1da177e4 | 2532 | |
f7ab34ea | 2533 | if (inode->i_size == 0 && ext3_should_writeback_data(inode)) |
9df93939 | 2534 | ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE); |
f7ab34ea | 2535 | |
1da177e4 | 2536 | handle = start_transaction(inode); |
ee3e77f1 | 2537 | if (IS_ERR(handle)) |
ef43618a | 2538 | goto out_notrans; |
1da177e4 LT |
2539 | |
2540 | last_block = (inode->i_size + blocksize-1) | |
2541 | >> EXT3_BLOCK_SIZE_BITS(inode->i_sb); | |
1da177e4 LT |
2542 | n = ext3_block_to_path(inode, last_block, offsets, NULL); |
2543 | if (n == 0) | |
2544 | goto out_stop; /* error */ | |
2545 | ||
2546 | /* | |
2547 | * OK. This truncate is going to happen. We add the inode to the | |
2548 | * orphan list, so that if this truncate spans multiple transactions, | |
2549 | * and we crash, we will resume the truncate when the filesystem | |
2550 | * recovers. It also marks the inode dirty, to catch the new size. | |
2551 | * | |
2552 | * Implication: the file must always be in a sane, consistent | |
2553 | * truncatable state while each transaction commits. | |
2554 | */ | |
2555 | if (ext3_orphan_add(handle, inode)) | |
2556 | goto out_stop; | |
2557 | ||
2558 | /* | |
2559 | * The orphan list entry will now protect us from any crash which | |
2560 | * occurs before the truncate completes, so it is now safe to propagate | |
2561 | * the new, shorter inode size (held for now in i_size) into the | |
2562 | * on-disk inode. We do this via i_disksize, which is the value which | |
2563 | * ext3 *really* writes onto the disk inode. | |
2564 | */ | |
2565 | ei->i_disksize = inode->i_size; | |
2566 | ||
2567 | /* | |
2568 | * From here we block out all ext3_get_block() callers who want to | |
2569 | * modify the block allocation tree. | |
2570 | */ | |
97461518 | 2571 | mutex_lock(&ei->truncate_mutex); |
1da177e4 LT |
2572 | |
2573 | if (n == 1) { /* direct blocks */ | |
2574 | ext3_free_data(handle, inode, NULL, i_data+offsets[0], | |
2575 | i_data + EXT3_NDIR_BLOCKS); | |
2576 | goto do_indirects; | |
2577 | } | |
2578 | ||
2579 | partial = ext3_find_shared(inode, n, offsets, chain, &nr); | |
2580 | /* Kill the top of shared branch (not detached) */ | |
2581 | if (nr) { | |
2582 | if (partial == chain) { | |
2583 | /* Shared branch grows from the inode */ | |
2584 | ext3_free_branches(handle, inode, NULL, | |
2585 | &nr, &nr+1, (chain+n-1) - partial); | |
2586 | *partial->p = 0; | |
2587 | /* | |
2588 | * We mark the inode dirty prior to restart, | |
2589 | * and prior to stop. No need for it here. | |
2590 | */ | |
2591 | } else { | |
2592 | /* Shared branch grows from an indirect block */ | |
1da177e4 LT |
2593 | ext3_free_branches(handle, inode, partial->bh, |
2594 | partial->p, | |
2595 | partial->p+1, (chain+n-1) - partial); | |
2596 | } | |
2597 | } | |
2598 | /* Clear the ends of indirect blocks on the shared branch */ | |
2599 | while (partial > chain) { | |
2600 | ext3_free_branches(handle, inode, partial->bh, partial->p + 1, | |
2601 | (__le32*)partial->bh->b_data+addr_per_block, | |
2602 | (chain+n-1) - partial); | |
2603 | BUFFER_TRACE(partial->bh, "call brelse"); | |
2604 | brelse (partial->bh); | |
2605 | partial--; | |
2606 | } | |
2607 | do_indirects: | |
2608 | /* Kill the remaining (whole) subtrees */ | |
2609 | switch (offsets[0]) { | |
d6859bfc AM |
2610 | default: |
2611 | nr = i_data[EXT3_IND_BLOCK]; | |
2612 | if (nr) { | |
2613 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1); | |
2614 | i_data[EXT3_IND_BLOCK] = 0; | |
2615 | } | |
2616 | case EXT3_IND_BLOCK: | |
2617 | nr = i_data[EXT3_DIND_BLOCK]; | |
2618 | if (nr) { | |
2619 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2); | |
2620 | i_data[EXT3_DIND_BLOCK] = 0; | |
2621 | } | |
2622 | case EXT3_DIND_BLOCK: | |
2623 | nr = i_data[EXT3_TIND_BLOCK]; | |
2624 | if (nr) { | |
2625 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3); | |
2626 | i_data[EXT3_TIND_BLOCK] = 0; | |
2627 | } | |
2628 | case EXT3_TIND_BLOCK: | |
2629 | ; | |
1da177e4 LT |
2630 | } |
2631 | ||
2632 | ext3_discard_reservation(inode); | |
2633 | ||
97461518 | 2634 | mutex_unlock(&ei->truncate_mutex); |
1da177e4 LT |
2635 | inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC; |
2636 | ext3_mark_inode_dirty(handle, inode); | |
2637 | ||
d6859bfc AM |
2638 | /* |
2639 | * In a multi-transaction truncate, we only make the final transaction | |
2640 | * synchronous | |
2641 | */ | |
1da177e4 LT |
2642 | if (IS_SYNC(inode)) |
2643 | handle->h_sync = 1; | |
2644 | out_stop: | |
2645 | /* | |
2646 | * If this was a simple ftruncate(), and the file will remain alive | |
2647 | * then we need to clear up the orphan record which we created above. | |
2648 | * However, if this was a real unlink then we were called by | |
ac14a95b | 2649 | * ext3_evict_inode(), and we allow that function to clean up the |
1da177e4 LT |
2650 | * orphan info for us. |
2651 | */ | |
2652 | if (inode->i_nlink) | |
2653 | ext3_orphan_del(handle, inode); | |
2654 | ||
2655 | ext3_journal_stop(handle); | |
785c4bcc | 2656 | trace_ext3_truncate_exit(inode); |
ef43618a JK |
2657 | return; |
2658 | out_notrans: | |
2659 | /* | |
2660 | * Delete the inode from orphan list so that it doesn't stay there | |
2661 | * forever and trigger assertion on umount. | |
2662 | */ | |
2663 | if (inode->i_nlink) | |
2664 | ext3_orphan_del(NULL, inode); | |
785c4bcc | 2665 | trace_ext3_truncate_exit(inode); |
1da177e4 LT |
2666 | } |
2667 | ||
43d23f90 | 2668 | static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb, |
1da177e4 LT |
2669 | unsigned long ino, struct ext3_iloc *iloc) |
2670 | { | |
e0e369a7 | 2671 | unsigned long block_group; |
43d23f90 MC |
2672 | unsigned long offset; |
2673 | ext3_fsblk_t block; | |
e0e369a7 | 2674 | struct ext3_group_desc *gdp; |
1da177e4 | 2675 | |
2ccb48eb NB |
2676 | if (!ext3_valid_inum(sb, ino)) { |
2677 | /* | |
2678 | * This error is already checked for in namei.c unless we are | |
2679 | * looking at an NFS filehandle, in which case no error | |
2680 | * report is needed | |
2681 | */ | |
1da177e4 LT |
2682 | return 0; |
2683 | } | |
2ccb48eb | 2684 | |
1da177e4 | 2685 | block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); |
e0e369a7 AM |
2686 | gdp = ext3_get_group_desc(sb, block_group, NULL); |
2687 | if (!gdp) | |
1da177e4 | 2688 | return 0; |
1da177e4 LT |
2689 | /* |
2690 | * Figure out the offset within the block group inode table | |
2691 | */ | |
2692 | offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) * | |
2693 | EXT3_INODE_SIZE(sb); | |
e0e369a7 | 2694 | block = le32_to_cpu(gdp->bg_inode_table) + |
1da177e4 LT |
2695 | (offset >> EXT3_BLOCK_SIZE_BITS(sb)); |
2696 | ||
2697 | iloc->block_group = block_group; | |
2698 | iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1); | |
2699 | return block; | |
2700 | } | |
2701 | ||
2702 | /* | |
2703 | * ext3_get_inode_loc returns with an extra refcount against the inode's | |
2704 | * underlying buffer_head on success. If 'in_mem' is true, we have all | |
2705 | * data in memory that is needed to recreate the on-disk version of this | |
2706 | * inode. | |
2707 | */ | |
2708 | static int __ext3_get_inode_loc(struct inode *inode, | |
2709 | struct ext3_iloc *iloc, int in_mem) | |
2710 | { | |
43d23f90 | 2711 | ext3_fsblk_t block; |
1da177e4 LT |
2712 | struct buffer_head *bh; |
2713 | ||
2714 | block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc); | |
2715 | if (!block) | |
2716 | return -EIO; | |
2717 | ||
2718 | bh = sb_getblk(inode->i_sb, block); | |
2719 | if (!bh) { | |
2720 | ext3_error (inode->i_sb, "ext3_get_inode_loc", | |
2721 | "unable to read inode block - " | |
43d23f90 MC |
2722 | "inode=%lu, block="E3FSBLK, |
2723 | inode->i_ino, block); | |
1da177e4 LT |
2724 | return -EIO; |
2725 | } | |
2726 | if (!buffer_uptodate(bh)) { | |
2727 | lock_buffer(bh); | |
95450f5a HK |
2728 | |
2729 | /* | |
2730 | * If the buffer has the write error flag, we have failed | |
2731 | * to write out another inode in the same block. In this | |
2732 | * case, we don't have to read the block because we may | |
2733 | * read the old inode data successfully. | |
2734 | */ | |
2735 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
2736 | set_buffer_uptodate(bh); | |
2737 | ||
1da177e4 LT |
2738 | if (buffer_uptodate(bh)) { |
2739 | /* someone brought it uptodate while we waited */ | |
2740 | unlock_buffer(bh); | |
2741 | goto has_buffer; | |
2742 | } | |
2743 | ||
2744 | /* | |
2745 | * If we have all information of the inode in memory and this | |
2746 | * is the only valid inode in the block, we need not read the | |
2747 | * block. | |
2748 | */ | |
2749 | if (in_mem) { | |
2750 | struct buffer_head *bitmap_bh; | |
2751 | struct ext3_group_desc *desc; | |
2752 | int inodes_per_buffer; | |
2753 | int inode_offset, i; | |
2754 | int block_group; | |
2755 | int start; | |
2756 | ||
2757 | block_group = (inode->i_ino - 1) / | |
2758 | EXT3_INODES_PER_GROUP(inode->i_sb); | |
2759 | inodes_per_buffer = bh->b_size / | |
2760 | EXT3_INODE_SIZE(inode->i_sb); | |
2761 | inode_offset = ((inode->i_ino - 1) % | |
2762 | EXT3_INODES_PER_GROUP(inode->i_sb)); | |
2763 | start = inode_offset & ~(inodes_per_buffer - 1); | |
2764 | ||
2765 | /* Is the inode bitmap in cache? */ | |
2766 | desc = ext3_get_group_desc(inode->i_sb, | |
2767 | block_group, NULL); | |
2768 | if (!desc) | |
2769 | goto make_io; | |
2770 | ||
2771 | bitmap_bh = sb_getblk(inode->i_sb, | |
2772 | le32_to_cpu(desc->bg_inode_bitmap)); | |
2773 | if (!bitmap_bh) | |
2774 | goto make_io; | |
2775 | ||
2776 | /* | |
2777 | * If the inode bitmap isn't in cache then the | |
2778 | * optimisation may end up performing two reads instead | |
2779 | * of one, so skip it. | |
2780 | */ | |
2781 | if (!buffer_uptodate(bitmap_bh)) { | |
2782 | brelse(bitmap_bh); | |
2783 | goto make_io; | |
2784 | } | |
2785 | for (i = start; i < start + inodes_per_buffer; i++) { | |
2786 | if (i == inode_offset) | |
2787 | continue; | |
2788 | if (ext3_test_bit(i, bitmap_bh->b_data)) | |
2789 | break; | |
2790 | } | |
2791 | brelse(bitmap_bh); | |
2792 | if (i == start + inodes_per_buffer) { | |
2793 | /* all other inodes are free, so skip I/O */ | |
2794 | memset(bh->b_data, 0, bh->b_size); | |
2795 | set_buffer_uptodate(bh); | |
2796 | unlock_buffer(bh); | |
2797 | goto has_buffer; | |
2798 | } | |
2799 | } | |
2800 | ||
2801 | make_io: | |
2802 | /* | |
2803 | * There are other valid inodes in the buffer, this inode | |
2804 | * has in-inode xattrs, or we don't have this inode in memory. | |
2805 | * Read the block from disk. | |
2806 | */ | |
785c4bcc | 2807 | trace_ext3_load_inode(inode); |
1da177e4 LT |
2808 | get_bh(bh); |
2809 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 2810 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
1da177e4 LT |
2811 | wait_on_buffer(bh); |
2812 | if (!buffer_uptodate(bh)) { | |
2813 | ext3_error(inode->i_sb, "ext3_get_inode_loc", | |
2814 | "unable to read inode block - " | |
43d23f90 | 2815 | "inode=%lu, block="E3FSBLK, |
1da177e4 LT |
2816 | inode->i_ino, block); |
2817 | brelse(bh); | |
2818 | return -EIO; | |
2819 | } | |
2820 | } | |
2821 | has_buffer: | |
2822 | iloc->bh = bh; | |
2823 | return 0; | |
2824 | } | |
2825 | ||
2826 | int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc) | |
2827 | { | |
2828 | /* We have all inode data except xattrs in memory here. */ | |
2829 | return __ext3_get_inode_loc(inode, iloc, | |
9df93939 | 2830 | !ext3_test_inode_state(inode, EXT3_STATE_XATTR)); |
1da177e4 LT |
2831 | } |
2832 | ||
2833 | void ext3_set_inode_flags(struct inode *inode) | |
2834 | { | |
2835 | unsigned int flags = EXT3_I(inode)->i_flags; | |
2836 | ||
2837 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
2838 | if (flags & EXT3_SYNC_FL) | |
2839 | inode->i_flags |= S_SYNC; | |
2840 | if (flags & EXT3_APPEND_FL) | |
2841 | inode->i_flags |= S_APPEND; | |
2842 | if (flags & EXT3_IMMUTABLE_FL) | |
2843 | inode->i_flags |= S_IMMUTABLE; | |
2844 | if (flags & EXT3_NOATIME_FL) | |
2845 | inode->i_flags |= S_NOATIME; | |
2846 | if (flags & EXT3_DIRSYNC_FL) | |
2847 | inode->i_flags |= S_DIRSYNC; | |
2848 | } | |
2849 | ||
28be5abb JK |
2850 | /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */ |
2851 | void ext3_get_inode_flags(struct ext3_inode_info *ei) | |
2852 | { | |
2853 | unsigned int flags = ei->vfs_inode.i_flags; | |
2854 | ||
2855 | ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL| | |
2856 | EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL); | |
2857 | if (flags & S_SYNC) | |
2858 | ei->i_flags |= EXT3_SYNC_FL; | |
2859 | if (flags & S_APPEND) | |
2860 | ei->i_flags |= EXT3_APPEND_FL; | |
2861 | if (flags & S_IMMUTABLE) | |
2862 | ei->i_flags |= EXT3_IMMUTABLE_FL; | |
2863 | if (flags & S_NOATIME) | |
2864 | ei->i_flags |= EXT3_NOATIME_FL; | |
2865 | if (flags & S_DIRSYNC) | |
2866 | ei->i_flags |= EXT3_DIRSYNC_FL; | |
2867 | } | |
2868 | ||
473043dc | 2869 | struct inode *ext3_iget(struct super_block *sb, unsigned long ino) |
1da177e4 LT |
2870 | { |
2871 | struct ext3_iloc iloc; | |
2872 | struct ext3_inode *raw_inode; | |
473043dc | 2873 | struct ext3_inode_info *ei; |
1da177e4 | 2874 | struct buffer_head *bh; |
473043dc | 2875 | struct inode *inode; |
fe8bc91c JK |
2876 | journal_t *journal = EXT3_SB(sb)->s_journal; |
2877 | transaction_t *transaction; | |
473043dc | 2878 | long ret; |
1da177e4 LT |
2879 | int block; |
2880 | ||
473043dc DH |
2881 | inode = iget_locked(sb, ino); |
2882 | if (!inode) | |
2883 | return ERR_PTR(-ENOMEM); | |
2884 | if (!(inode->i_state & I_NEW)) | |
2885 | return inode; | |
2886 | ||
2887 | ei = EXT3_I(inode); | |
1da177e4 LT |
2888 | ei->i_block_alloc_info = NULL; |
2889 | ||
473043dc DH |
2890 | ret = __ext3_get_inode_loc(inode, &iloc, 0); |
2891 | if (ret < 0) | |
1da177e4 LT |
2892 | goto bad_inode; |
2893 | bh = iloc.bh; | |
2894 | raw_inode = ext3_raw_inode(&iloc); | |
2895 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); | |
2896 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
2897 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
2898 | if(!(test_opt (inode->i_sb, NO_UID32))) { | |
2899 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; | |
2900 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
2901 | } | |
2902 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
2903 | inode->i_size = le32_to_cpu(raw_inode->i_size); | |
4d7bf11d MR |
2904 | inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime); |
2905 | inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime); | |
2906 | inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime); | |
1da177e4 LT |
2907 | inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0; |
2908 | ||
de329820 | 2909 | ei->i_state_flags = 0; |
1da177e4 LT |
2910 | ei->i_dir_start_lookup = 0; |
2911 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
2912 | /* We now have enough fields to check if the inode was active or not. | |
2913 | * This is needed because nfsd might try to access dead inodes | |
2914 | * the test is that same one that e2fsck uses | |
2915 | * NeilBrown 1999oct15 | |
2916 | */ | |
2917 | if (inode->i_nlink == 0) { | |
2918 | if (inode->i_mode == 0 || | |
2919 | !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) { | |
2920 | /* this inode is deleted */ | |
2921 | brelse (bh); | |
473043dc | 2922 | ret = -ESTALE; |
1da177e4 LT |
2923 | goto bad_inode; |
2924 | } | |
2925 | /* The only unlinked inodes we let through here have | |
2926 | * valid i_mode and are being read by the orphan | |
2927 | * recovery code: that's fine, we're about to complete | |
2928 | * the process of deleting those. */ | |
2929 | } | |
1da177e4 LT |
2930 | inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); |
2931 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); | |
2932 | #ifdef EXT3_FRAGMENTS | |
2933 | ei->i_faddr = le32_to_cpu(raw_inode->i_faddr); | |
2934 | ei->i_frag_no = raw_inode->i_frag; | |
2935 | ei->i_frag_size = raw_inode->i_fsize; | |
2936 | #endif | |
2937 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); | |
2938 | if (!S_ISREG(inode->i_mode)) { | |
2939 | ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); | |
2940 | } else { | |
2941 | inode->i_size |= | |
2942 | ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; | |
2943 | } | |
2944 | ei->i_disksize = inode->i_size; | |
2945 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); | |
2946 | ei->i_block_group = iloc.block_group; | |
2947 | /* | |
2948 | * NOTE! The in-memory inode i_data array is in little-endian order | |
2949 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
2950 | */ | |
2951 | for (block = 0; block < EXT3_N_BLOCKS; block++) | |
2952 | ei->i_data[block] = raw_inode->i_block[block]; | |
2953 | INIT_LIST_HEAD(&ei->i_orphan); | |
2954 | ||
fe8bc91c JK |
2955 | /* |
2956 | * Set transaction id's of transactions that have to be committed | |
2957 | * to finish f[data]sync. We set them to currently running transaction | |
2958 | * as we cannot be sure that the inode or some of its metadata isn't | |
2959 | * part of the transaction - the inode could have been reclaimed and | |
2960 | * now it is reread from disk. | |
2961 | */ | |
2962 | if (journal) { | |
2963 | tid_t tid; | |
2964 | ||
2965 | spin_lock(&journal->j_state_lock); | |
2966 | if (journal->j_running_transaction) | |
2967 | transaction = journal->j_running_transaction; | |
2968 | else | |
2969 | transaction = journal->j_committing_transaction; | |
2970 | if (transaction) | |
2971 | tid = transaction->t_tid; | |
2972 | else | |
2973 | tid = journal->j_commit_sequence; | |
2974 | spin_unlock(&journal->j_state_lock); | |
2975 | atomic_set(&ei->i_sync_tid, tid); | |
2976 | atomic_set(&ei->i_datasync_tid, tid); | |
2977 | } | |
2978 | ||
1da177e4 LT |
2979 | if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 && |
2980 | EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) { | |
2981 | /* | |
2982 | * When mke2fs creates big inodes it does not zero out | |
2983 | * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE, | |
2984 | * so ignore those first few inodes. | |
2985 | */ | |
2986 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
2987 | if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
e4a10a36 KK |
2988 | EXT3_INODE_SIZE(inode->i_sb)) { |
2989 | brelse (bh); | |
473043dc | 2990 | ret = -EIO; |
1da177e4 | 2991 | goto bad_inode; |
e4a10a36 | 2992 | } |
1da177e4 LT |
2993 | if (ei->i_extra_isize == 0) { |
2994 | /* The extra space is currently unused. Use it. */ | |
2995 | ei->i_extra_isize = sizeof(struct ext3_inode) - | |
2996 | EXT3_GOOD_OLD_INODE_SIZE; | |
2997 | } else { | |
2998 | __le32 *magic = (void *)raw_inode + | |
2999 | EXT3_GOOD_OLD_INODE_SIZE + | |
3000 | ei->i_extra_isize; | |
3001 | if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC)) | |
9df93939 | 3002 | ext3_set_inode_state(inode, EXT3_STATE_XATTR); |
1da177e4 LT |
3003 | } |
3004 | } else | |
3005 | ei->i_extra_isize = 0; | |
3006 | ||
3007 | if (S_ISREG(inode->i_mode)) { | |
3008 | inode->i_op = &ext3_file_inode_operations; | |
3009 | inode->i_fop = &ext3_file_operations; | |
3010 | ext3_set_aops(inode); | |
3011 | } else if (S_ISDIR(inode->i_mode)) { | |
3012 | inode->i_op = &ext3_dir_inode_operations; | |
3013 | inode->i_fop = &ext3_dir_operations; | |
3014 | } else if (S_ISLNK(inode->i_mode)) { | |
b5ed3112 | 3015 | if (ext3_inode_is_fast_symlink(inode)) { |
1da177e4 | 3016 | inode->i_op = &ext3_fast_symlink_inode_operations; |
b5ed3112 DG |
3017 | nd_terminate_link(ei->i_data, inode->i_size, |
3018 | sizeof(ei->i_data) - 1); | |
3019 | } else { | |
1da177e4 LT |
3020 | inode->i_op = &ext3_symlink_inode_operations; |
3021 | ext3_set_aops(inode); | |
3022 | } | |
3023 | } else { | |
3024 | inode->i_op = &ext3_special_inode_operations; | |
3025 | if (raw_inode->i_block[0]) | |
3026 | init_special_inode(inode, inode->i_mode, | |
3027 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
ae6ddcc5 | 3028 | else |
1da177e4 LT |
3029 | init_special_inode(inode, inode->i_mode, |
3030 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
3031 | } | |
3032 | brelse (iloc.bh); | |
3033 | ext3_set_inode_flags(inode); | |
473043dc DH |
3034 | unlock_new_inode(inode); |
3035 | return inode; | |
1da177e4 LT |
3036 | |
3037 | bad_inode: | |
473043dc DH |
3038 | iget_failed(inode); |
3039 | return ERR_PTR(ret); | |
1da177e4 LT |
3040 | } |
3041 | ||
3042 | /* | |
3043 | * Post the struct inode info into an on-disk inode location in the | |
3044 | * buffer-cache. This gobbles the caller's reference to the | |
3045 | * buffer_head in the inode location struct. | |
3046 | * | |
3047 | * The caller must have write access to iloc->bh. | |
3048 | */ | |
ae6ddcc5 MC |
3049 | static int ext3_do_update_inode(handle_t *handle, |
3050 | struct inode *inode, | |
1da177e4 LT |
3051 | struct ext3_iloc *iloc) |
3052 | { | |
3053 | struct ext3_inode *raw_inode = ext3_raw_inode(iloc); | |
3054 | struct ext3_inode_info *ei = EXT3_I(inode); | |
3055 | struct buffer_head *bh = iloc->bh; | |
3056 | int err = 0, rc, block; | |
3057 | ||
4f003fd3 CM |
3058 | again: |
3059 | /* we can't allow multiple procs in here at once, its a bit racey */ | |
3060 | lock_buffer(bh); | |
3061 | ||
1da177e4 LT |
3062 | /* For fields not not tracking in the in-memory inode, |
3063 | * initialise them to zero for new inodes. */ | |
9df93939 | 3064 | if (ext3_test_inode_state(inode, EXT3_STATE_NEW)) |
1da177e4 LT |
3065 | memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size); |
3066 | ||
28be5abb | 3067 | ext3_get_inode_flags(ei); |
1da177e4 LT |
3068 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
3069 | if(!(test_opt(inode->i_sb, NO_UID32))) { | |
3070 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); | |
3071 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
3072 | /* | |
3073 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
3074 | * re-used with the upper 16 bits of the uid/gid intact | |
3075 | */ | |
3076 | if(!ei->i_dtime) { | |
3077 | raw_inode->i_uid_high = | |
3078 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
3079 | raw_inode->i_gid_high = | |
3080 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
3081 | } else { | |
3082 | raw_inode->i_uid_high = 0; | |
3083 | raw_inode->i_gid_high = 0; | |
3084 | } | |
3085 | } else { | |
3086 | raw_inode->i_uid_low = | |
3087 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
3088 | raw_inode->i_gid_low = | |
3089 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
3090 | raw_inode->i_uid_high = 0; | |
3091 | raw_inode->i_gid_high = 0; | |
3092 | } | |
3093 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
3094 | raw_inode->i_size = cpu_to_le32(ei->i_disksize); | |
3095 | raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec); | |
3096 | raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec); | |
3097 | raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec); | |
3098 | raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); | |
3099 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); | |
3100 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); | |
3101 | #ifdef EXT3_FRAGMENTS | |
3102 | raw_inode->i_faddr = cpu_to_le32(ei->i_faddr); | |
3103 | raw_inode->i_frag = ei->i_frag_no; | |
3104 | raw_inode->i_fsize = ei->i_frag_size; | |
3105 | #endif | |
3106 | raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); | |
3107 | if (!S_ISREG(inode->i_mode)) { | |
3108 | raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); | |
3109 | } else { | |
3110 | raw_inode->i_size_high = | |
3111 | cpu_to_le32(ei->i_disksize >> 32); | |
3112 | if (ei->i_disksize > 0x7fffffffULL) { | |
3113 | struct super_block *sb = inode->i_sb; | |
3114 | if (!EXT3_HAS_RO_COMPAT_FEATURE(sb, | |
3115 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE) || | |
3116 | EXT3_SB(sb)->s_es->s_rev_level == | |
3117 | cpu_to_le32(EXT3_GOOD_OLD_REV)) { | |
3118 | /* If this is the first large file | |
3119 | * created, add a flag to the superblock. | |
3120 | */ | |
4f003fd3 | 3121 | unlock_buffer(bh); |
1da177e4 LT |
3122 | err = ext3_journal_get_write_access(handle, |
3123 | EXT3_SB(sb)->s_sbh); | |
3124 | if (err) | |
3125 | goto out_brelse; | |
4f003fd3 | 3126 | |
1da177e4 LT |
3127 | ext3_update_dynamic_rev(sb); |
3128 | EXT3_SET_RO_COMPAT_FEATURE(sb, | |
3129 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE); | |
1da177e4 LT |
3130 | handle->h_sync = 1; |
3131 | err = ext3_journal_dirty_metadata(handle, | |
3132 | EXT3_SB(sb)->s_sbh); | |
4f003fd3 CM |
3133 | /* get our lock and start over */ |
3134 | goto again; | |
1da177e4 LT |
3135 | } |
3136 | } | |
3137 | } | |
3138 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
3139 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
3140 | if (old_valid_dev(inode->i_rdev)) { | |
3141 | raw_inode->i_block[0] = | |
3142 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
3143 | raw_inode->i_block[1] = 0; | |
3144 | } else { | |
3145 | raw_inode->i_block[0] = 0; | |
3146 | raw_inode->i_block[1] = | |
3147 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
3148 | raw_inode->i_block[2] = 0; | |
3149 | } | |
3150 | } else for (block = 0; block < EXT3_N_BLOCKS; block++) | |
3151 | raw_inode->i_block[block] = ei->i_data[block]; | |
3152 | ||
ff87b37d | 3153 | if (ei->i_extra_isize) |
1da177e4 LT |
3154 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
3155 | ||
3156 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
4f003fd3 | 3157 | unlock_buffer(bh); |
1da177e4 LT |
3158 | rc = ext3_journal_dirty_metadata(handle, bh); |
3159 | if (!err) | |
3160 | err = rc; | |
9df93939 | 3161 | ext3_clear_inode_state(inode, EXT3_STATE_NEW); |
1da177e4 | 3162 | |
fe8bc91c | 3163 | atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid); |
1da177e4 LT |
3164 | out_brelse: |
3165 | brelse (bh); | |
3166 | ext3_std_error(inode->i_sb, err); | |
3167 | return err; | |
3168 | } | |
3169 | ||
3170 | /* | |
3171 | * ext3_write_inode() | |
3172 | * | |
3173 | * We are called from a few places: | |
3174 | * | |
3175 | * - Within generic_file_write() for O_SYNC files. | |
3176 | * Here, there will be no transaction running. We wait for any running | |
3177 | * trasnaction to commit. | |
3178 | * | |
3179 | * - Within sys_sync(), kupdate and such. | |
3180 | * We wait on commit, if tol to. | |
3181 | * | |
3182 | * - Within prune_icache() (PF_MEMALLOC == true) | |
3183 | * Here we simply return. We can't afford to block kswapd on the | |
3184 | * journal commit. | |
3185 | * | |
3186 | * In all cases it is actually safe for us to return without doing anything, | |
3187 | * because the inode has been copied into a raw inode buffer in | |
3188 | * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for | |
3189 | * knfsd. | |
3190 | * | |
3191 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
3192 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
3193 | * which we are interested. | |
3194 | * | |
3195 | * It would be a bug for them to not do this. The code: | |
3196 | * | |
3197 | * mark_inode_dirty(inode) | |
3198 | * stuff(); | |
3199 | * inode->i_size = expr; | |
3200 | * | |
3201 | * is in error because a kswapd-driven write_inode() could occur while | |
3202 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
3203 | * will no longer be on the superblock's dirty inode list. | |
3204 | */ | |
a9185b41 | 3205 | int ext3_write_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 LT |
3206 | { |
3207 | if (current->flags & PF_MEMALLOC) | |
3208 | return 0; | |
3209 | ||
3210 | if (ext3_journal_current_handle()) { | |
9ad163ae | 3211 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
1da177e4 LT |
3212 | dump_stack(); |
3213 | return -EIO; | |
3214 | } | |
3215 | ||
a9185b41 | 3216 | if (wbc->sync_mode != WB_SYNC_ALL) |
1da177e4 LT |
3217 | return 0; |
3218 | ||
3219 | return ext3_force_commit(inode->i_sb); | |
3220 | } | |
3221 | ||
3222 | /* | |
3223 | * ext3_setattr() | |
3224 | * | |
3225 | * Called from notify_change. | |
3226 | * | |
3227 | * We want to trap VFS attempts to truncate the file as soon as | |
3228 | * possible. In particular, we want to make sure that when the VFS | |
3229 | * shrinks i_size, we put the inode on the orphan list and modify | |
3230 | * i_disksize immediately, so that during the subsequent flushing of | |
3231 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
3232 | * commit will leave the blocks being flushed in an unused state on | |
3233 | * disk. (On recovery, the inode will get truncated and the blocks will | |
3234 | * be freed, so we have a strong guarantee that no future commit will | |
ae6ddcc5 | 3235 | * leave these blocks visible to the user.) |
1da177e4 LT |
3236 | * |
3237 | * Called with inode->sem down. | |
3238 | */ | |
3239 | int ext3_setattr(struct dentry *dentry, struct iattr *attr) | |
3240 | { | |
3241 | struct inode *inode = dentry->d_inode; | |
3242 | int error, rc = 0; | |
3243 | const unsigned int ia_valid = attr->ia_valid; | |
3244 | ||
3245 | error = inode_change_ok(inode, attr); | |
3246 | if (error) | |
3247 | return error; | |
3248 | ||
12755627 | 3249 | if (is_quota_modification(inode, attr)) |
871a2931 | 3250 | dquot_initialize(inode); |
1da177e4 LT |
3251 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
3252 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
3253 | handle_t *handle; | |
3254 | ||
3255 | /* (user+group)*(old+new) structure, inode write (sb, | |
3256 | * inode block, ? - but truncate inode update has it) */ | |
c459001f DM |
3257 | handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
3258 | EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3); | |
1da177e4 LT |
3259 | if (IS_ERR(handle)) { |
3260 | error = PTR_ERR(handle); | |
3261 | goto err_out; | |
3262 | } | |
b43fa828 | 3263 | error = dquot_transfer(inode, attr); |
1da177e4 LT |
3264 | if (error) { |
3265 | ext3_journal_stop(handle); | |
3266 | return error; | |
3267 | } | |
3268 | /* Update corresponding info in inode so that everything is in | |
3269 | * one transaction */ | |
3270 | if (attr->ia_valid & ATTR_UID) | |
3271 | inode->i_uid = attr->ia_uid; | |
3272 | if (attr->ia_valid & ATTR_GID) | |
3273 | inode->i_gid = attr->ia_gid; | |
3274 | error = ext3_mark_inode_dirty(handle, inode); | |
3275 | ext3_journal_stop(handle); | |
3276 | } | |
3277 | ||
562c72aa CH |
3278 | if (attr->ia_valid & ATTR_SIZE) |
3279 | inode_dio_wait(inode); | |
3280 | ||
1da177e4 LT |
3281 | if (S_ISREG(inode->i_mode) && |
3282 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { | |
3283 | handle_t *handle; | |
3284 | ||
3285 | handle = ext3_journal_start(inode, 3); | |
3286 | if (IS_ERR(handle)) { | |
3287 | error = PTR_ERR(handle); | |
3288 | goto err_out; | |
3289 | } | |
3290 | ||
3291 | error = ext3_orphan_add(handle, inode); | |
ee3e77f1 JK |
3292 | if (error) { |
3293 | ext3_journal_stop(handle); | |
3294 | goto err_out; | |
3295 | } | |
1da177e4 | 3296 | EXT3_I(inode)->i_disksize = attr->ia_size; |
ee3e77f1 | 3297 | error = ext3_mark_inode_dirty(handle, inode); |
1da177e4 | 3298 | ext3_journal_stop(handle); |
ee3e77f1 JK |
3299 | if (error) { |
3300 | /* Some hard fs error must have happened. Bail out. */ | |
3301 | ext3_orphan_del(NULL, inode); | |
3302 | goto err_out; | |
3303 | } | |
3304 | rc = ext3_block_truncate_page(inode, attr->ia_size); | |
3305 | if (rc) { | |
3306 | /* Cleanup orphan list and exit */ | |
3307 | handle = ext3_journal_start(inode, 3); | |
3308 | if (IS_ERR(handle)) { | |
3309 | ext3_orphan_del(NULL, inode); | |
3310 | goto err_out; | |
3311 | } | |
3312 | ext3_orphan_del(handle, inode); | |
3313 | ext3_journal_stop(handle); | |
3314 | goto err_out; | |
3315 | } | |
1da177e4 LT |
3316 | } |
3317 | ||
1025774c CH |
3318 | if ((attr->ia_valid & ATTR_SIZE) && |
3319 | attr->ia_size != i_size_read(inode)) { | |
40680f2f JK |
3320 | truncate_setsize(inode, attr->ia_size); |
3321 | ext3_truncate(inode); | |
1025774c CH |
3322 | } |
3323 | ||
3324 | setattr_copy(inode, attr); | |
3325 | mark_inode_dirty(inode); | |
1da177e4 | 3326 | |
1025774c | 3327 | if (ia_valid & ATTR_MODE) |
1da177e4 LT |
3328 | rc = ext3_acl_chmod(inode); |
3329 | ||
3330 | err_out: | |
3331 | ext3_std_error(inode->i_sb, error); | |
3332 | if (!error) | |
3333 | error = rc; | |
3334 | return error; | |
3335 | } | |
3336 | ||
3337 | ||
3338 | /* | |
d6859bfc | 3339 | * How many blocks doth make a writepage()? |
1da177e4 LT |
3340 | * |
3341 | * With N blocks per page, it may be: | |
3342 | * N data blocks | |
3343 | * 2 indirect block | |
3344 | * 2 dindirect | |
3345 | * 1 tindirect | |
3346 | * N+5 bitmap blocks (from the above) | |
3347 | * N+5 group descriptor summary blocks | |
3348 | * 1 inode block | |
3349 | * 1 superblock. | |
3350 | * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files | |
3351 | * | |
3352 | * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS | |
3353 | * | |
3354 | * With ordered or writeback data it's the same, less the N data blocks. | |
3355 | * | |
3356 | * If the inode's direct blocks can hold an integral number of pages then a | |
3357 | * page cannot straddle two indirect blocks, and we can only touch one indirect | |
3358 | * and dindirect block, and the "5" above becomes "3". | |
3359 | * | |
3360 | * This still overestimates under most circumstances. If we were to pass the | |
3361 | * start and end offsets in here as well we could do block_to_path() on each | |
3362 | * block and work out the exact number of indirects which are touched. Pah. | |
3363 | */ | |
3364 | ||
3365 | static int ext3_writepage_trans_blocks(struct inode *inode) | |
3366 | { | |
3367 | int bpp = ext3_journal_blocks_per_page(inode); | |
3368 | int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3; | |
3369 | int ret; | |
3370 | ||
3371 | if (ext3_should_journal_data(inode)) | |
3372 | ret = 3 * (bpp + indirects) + 2; | |
3373 | else | |
523334ba | 3374 | ret = 2 * (bpp + indirects) + indirects + 2; |
1da177e4 LT |
3375 | |
3376 | #ifdef CONFIG_QUOTA | |
871a2931 | 3377 | /* We know that structure was already allocated during dquot_initialize so |
1da177e4 | 3378 | * we will be updating only the data blocks + inodes */ |
c459001f | 3379 | ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
1da177e4 LT |
3380 | #endif |
3381 | ||
3382 | return ret; | |
3383 | } | |
3384 | ||
3385 | /* | |
3386 | * The caller must have previously called ext3_reserve_inode_write(). | |
3387 | * Give this, we know that the caller already has write access to iloc->bh. | |
3388 | */ | |
3389 | int ext3_mark_iloc_dirty(handle_t *handle, | |
3390 | struct inode *inode, struct ext3_iloc *iloc) | |
3391 | { | |
3392 | int err = 0; | |
3393 | ||
3394 | /* the do_update_inode consumes one bh->b_count */ | |
3395 | get_bh(iloc->bh); | |
3396 | ||
3397 | /* ext3_do_update_inode() does journal_dirty_metadata */ | |
3398 | err = ext3_do_update_inode(handle, inode, iloc); | |
3399 | put_bh(iloc->bh); | |
3400 | return err; | |
3401 | } | |
3402 | ||
ae6ddcc5 | 3403 | /* |
1da177e4 | 3404 | * On success, We end up with an outstanding reference count against |
ae6ddcc5 | 3405 | * iloc->bh. This _must_ be cleaned up later. |
1da177e4 LT |
3406 | */ |
3407 | ||
3408 | int | |
ae6ddcc5 | 3409 | ext3_reserve_inode_write(handle_t *handle, struct inode *inode, |
1da177e4 LT |
3410 | struct ext3_iloc *iloc) |
3411 | { | |
3412 | int err = 0; | |
3413 | if (handle) { | |
3414 | err = ext3_get_inode_loc(inode, iloc); | |
3415 | if (!err) { | |
3416 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
3417 | err = ext3_journal_get_write_access(handle, iloc->bh); | |
3418 | if (err) { | |
3419 | brelse(iloc->bh); | |
3420 | iloc->bh = NULL; | |
3421 | } | |
3422 | } | |
3423 | } | |
3424 | ext3_std_error(inode->i_sb, err); | |
3425 | return err; | |
3426 | } | |
3427 | ||
3428 | /* | |
d6859bfc AM |
3429 | * What we do here is to mark the in-core inode as clean with respect to inode |
3430 | * dirtiness (it may still be data-dirty). | |
1da177e4 LT |
3431 | * This means that the in-core inode may be reaped by prune_icache |
3432 | * without having to perform any I/O. This is a very good thing, | |
3433 | * because *any* task may call prune_icache - even ones which | |
3434 | * have a transaction open against a different journal. | |
3435 | * | |
3436 | * Is this cheating? Not really. Sure, we haven't written the | |
3437 | * inode out, but prune_icache isn't a user-visible syncing function. | |
3438 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
3439 | * we start and wait on commits. | |
3440 | * | |
3441 | * Is this efficient/effective? Well, we're being nice to the system | |
3442 | * by cleaning up our inodes proactively so they can be reaped | |
3443 | * without I/O. But we are potentially leaving up to five seconds' | |
3444 | * worth of inodes floating about which prune_icache wants us to | |
3445 | * write out. One way to fix that would be to get prune_icache() | |
3446 | * to do a write_super() to free up some memory. It has the desired | |
3447 | * effect. | |
3448 | */ | |
3449 | int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode) | |
3450 | { | |
3451 | struct ext3_iloc iloc; | |
3452 | int err; | |
3453 | ||
3454 | might_sleep(); | |
785c4bcc | 3455 | trace_ext3_mark_inode_dirty(inode, _RET_IP_); |
1da177e4 LT |
3456 | err = ext3_reserve_inode_write(handle, inode, &iloc); |
3457 | if (!err) | |
3458 | err = ext3_mark_iloc_dirty(handle, inode, &iloc); | |
3459 | return err; | |
3460 | } | |
3461 | ||
3462 | /* | |
d6859bfc | 3463 | * ext3_dirty_inode() is called from __mark_inode_dirty() |
1da177e4 LT |
3464 | * |
3465 | * We're really interested in the case where a file is being extended. | |
3466 | * i_size has been changed by generic_commit_write() and we thus need | |
3467 | * to include the updated inode in the current transaction. | |
3468 | * | |
5dd4056d | 3469 | * Also, dquot_alloc_space() will always dirty the inode when blocks |
1da177e4 LT |
3470 | * are allocated to the file. |
3471 | * | |
3472 | * If the inode is marked synchronous, we don't honour that here - doing | |
3473 | * so would cause a commit on atime updates, which we don't bother doing. | |
3474 | * We handle synchronous inodes at the highest possible level. | |
3475 | */ | |
aa385729 | 3476 | void ext3_dirty_inode(struct inode *inode, int flags) |
1da177e4 LT |
3477 | { |
3478 | handle_t *current_handle = ext3_journal_current_handle(); | |
3479 | handle_t *handle; | |
3480 | ||
3481 | handle = ext3_journal_start(inode, 2); | |
3482 | if (IS_ERR(handle)) | |
3483 | goto out; | |
3484 | if (current_handle && | |
3485 | current_handle->h_transaction != handle->h_transaction) { | |
3486 | /* This task has a transaction open against a different fs */ | |
3487 | printk(KERN_EMERG "%s: transactions do not match!\n", | |
e05b6b52 | 3488 | __func__); |
1da177e4 LT |
3489 | } else { |
3490 | jbd_debug(5, "marking dirty. outer handle=%p\n", | |
3491 | current_handle); | |
3492 | ext3_mark_inode_dirty(handle, inode); | |
3493 | } | |
3494 | ext3_journal_stop(handle); | |
3495 | out: | |
3496 | return; | |
3497 | } | |
3498 | ||
d6859bfc | 3499 | #if 0 |
ae6ddcc5 | 3500 | /* |
1da177e4 LT |
3501 | * Bind an inode's backing buffer_head into this transaction, to prevent |
3502 | * it from being flushed to disk early. Unlike | |
3503 | * ext3_reserve_inode_write, this leaves behind no bh reference and | |
3504 | * returns no iloc structure, so the caller needs to repeat the iloc | |
3505 | * lookup to mark the inode dirty later. | |
3506 | */ | |
d6859bfc | 3507 | static int ext3_pin_inode(handle_t *handle, struct inode *inode) |
1da177e4 LT |
3508 | { |
3509 | struct ext3_iloc iloc; | |
3510 | ||
3511 | int err = 0; | |
3512 | if (handle) { | |
3513 | err = ext3_get_inode_loc(inode, &iloc); | |
3514 | if (!err) { | |
3515 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
3516 | err = journal_get_write_access(handle, iloc.bh); | |
3517 | if (!err) | |
ae6ddcc5 | 3518 | err = ext3_journal_dirty_metadata(handle, |
1da177e4 LT |
3519 | iloc.bh); |
3520 | brelse(iloc.bh); | |
3521 | } | |
3522 | } | |
3523 | ext3_std_error(inode->i_sb, err); | |
3524 | return err; | |
3525 | } | |
3526 | #endif | |
3527 | ||
3528 | int ext3_change_inode_journal_flag(struct inode *inode, int val) | |
3529 | { | |
3530 | journal_t *journal; | |
3531 | handle_t *handle; | |
3532 | int err; | |
3533 | ||
3534 | /* | |
3535 | * We have to be very careful here: changing a data block's | |
3536 | * journaling status dynamically is dangerous. If we write a | |
3537 | * data block to the journal, change the status and then delete | |
3538 | * that block, we risk forgetting to revoke the old log record | |
3539 | * from the journal and so a subsequent replay can corrupt data. | |
3540 | * So, first we make sure that the journal is empty and that | |
3541 | * nobody is changing anything. | |
3542 | */ | |
3543 | ||
3544 | journal = EXT3_JOURNAL(inode); | |
e3a68e30 | 3545 | if (is_journal_aborted(journal)) |
1da177e4 LT |
3546 | return -EROFS; |
3547 | ||
3548 | journal_lock_updates(journal); | |
3549 | journal_flush(journal); | |
3550 | ||
3551 | /* | |
3552 | * OK, there are no updates running now, and all cached data is | |
3553 | * synced to disk. We are now in a completely consistent state | |
3554 | * which doesn't have anything in the journal, and we know that | |
3555 | * no filesystem updates are running, so it is safe to modify | |
3556 | * the inode's in-core data-journaling state flag now. | |
3557 | */ | |
3558 | ||
3559 | if (val) | |
3560 | EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL; | |
3561 | else | |
3562 | EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL; | |
3563 | ext3_set_aops(inode); | |
3564 | ||
3565 | journal_unlock_updates(journal); | |
3566 | ||
3567 | /* Finally we can mark the inode as dirty. */ | |
3568 | ||
3569 | handle = ext3_journal_start(inode, 1); | |
3570 | if (IS_ERR(handle)) | |
3571 | return PTR_ERR(handle); | |
3572 | ||
3573 | err = ext3_mark_inode_dirty(handle, inode); | |
3574 | handle->h_sync = 1; | |
3575 | ext3_journal_stop(handle); | |
3576 | ext3_std_error(inode->i_sb, err); | |
3577 | ||
3578 | return err; | |
3579 | } |