]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/ext3/inode.c | |
3 | * | |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
15 | * Goal-directed block allocation by Stephen Tweedie | |
e9ad5620 | 16 | * (sct@redhat.com), 1993, 1998 |
1da177e4 LT |
17 | * Big-endian to little-endian byte-swapping/bitmaps by |
18 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
19 | * 64-bit file support on 64-bit platforms by Jakub Jelinek | |
e9ad5620 | 20 | * (jj@sunsite.ms.mff.cuni.cz) |
1da177e4 LT |
21 | * |
22 | * Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000 | |
23 | */ | |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/time.h> | |
28 | #include <linux/ext3_jbd.h> | |
29 | #include <linux/jbd.h> | |
1da177e4 LT |
30 | #include <linux/highuid.h> |
31 | #include <linux/pagemap.h> | |
32 | #include <linux/quotaops.h> | |
33 | #include <linux/string.h> | |
34 | #include <linux/buffer_head.h> | |
35 | #include <linux/writeback.h> | |
36 | #include <linux/mpage.h> | |
37 | #include <linux/uio.h> | |
caa38fb0 | 38 | #include <linux/bio.h> |
68c9d702 | 39 | #include <linux/fiemap.h> |
b5ed3112 | 40 | #include <linux/namei.h> |
1da177e4 LT |
41 | #include "xattr.h" |
42 | #include "acl.h" | |
43 | ||
44 | static int ext3_writepage_trans_blocks(struct inode *inode); | |
45 | ||
46 | /* | |
47 | * Test whether an inode is a fast symlink. | |
48 | */ | |
d6859bfc | 49 | static int ext3_inode_is_fast_symlink(struct inode *inode) |
1da177e4 LT |
50 | { |
51 | int ea_blocks = EXT3_I(inode)->i_file_acl ? | |
52 | (inode->i_sb->s_blocksize >> 9) : 0; | |
53 | ||
d6859bfc | 54 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
1da177e4 LT |
55 | } |
56 | ||
d6859bfc AM |
57 | /* |
58 | * The ext3 forget function must perform a revoke if we are freeing data | |
1da177e4 | 59 | * which has been journaled. Metadata (eg. indirect blocks) must be |
ae6ddcc5 | 60 | * revoked in all cases. |
1da177e4 LT |
61 | * |
62 | * "bh" may be NULL: a metadata block may have been freed from memory | |
63 | * but there may still be a record of it in the journal, and that record | |
64 | * still needs to be revoked. | |
65 | */ | |
d6859bfc | 66 | int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode, |
1c2bf374 | 67 | struct buffer_head *bh, ext3_fsblk_t blocknr) |
1da177e4 LT |
68 | { |
69 | int err; | |
70 | ||
71 | might_sleep(); | |
72 | ||
73 | BUFFER_TRACE(bh, "enter"); | |
74 | ||
75 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " | |
76 | "data mode %lx\n", | |
77 | bh, is_metadata, inode->i_mode, | |
78 | test_opt(inode->i_sb, DATA_FLAGS)); | |
79 | ||
80 | /* Never use the revoke function if we are doing full data | |
81 | * journaling: there is no need to, and a V1 superblock won't | |
82 | * support it. Otherwise, only skip the revoke on un-journaled | |
83 | * data blocks. */ | |
84 | ||
85 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA || | |
86 | (!is_metadata && !ext3_should_journal_data(inode))) { | |
87 | if (bh) { | |
88 | BUFFER_TRACE(bh, "call journal_forget"); | |
89 | return ext3_journal_forget(handle, bh); | |
90 | } | |
91 | return 0; | |
92 | } | |
93 | ||
94 | /* | |
95 | * data!=journal && (is_metadata || should_journal_data(inode)) | |
96 | */ | |
97 | BUFFER_TRACE(bh, "call ext3_journal_revoke"); | |
98 | err = ext3_journal_revoke(handle, blocknr, bh); | |
99 | if (err) | |
e05b6b52 | 100 | ext3_abort(inode->i_sb, __func__, |
1da177e4 LT |
101 | "error %d when attempting revoke", err); |
102 | BUFFER_TRACE(bh, "exit"); | |
103 | return err; | |
104 | } | |
105 | ||
106 | /* | |
d6859bfc | 107 | * Work out how many blocks we need to proceed with the next chunk of a |
1da177e4 LT |
108 | * truncate transaction. |
109 | */ | |
ae6ddcc5 | 110 | static unsigned long blocks_for_truncate(struct inode *inode) |
1da177e4 LT |
111 | { |
112 | unsigned long needed; | |
113 | ||
114 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
115 | ||
116 | /* Give ourselves just enough room to cope with inodes in which | |
117 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
118 | * which resulted in random data in an inode which looked enough | |
119 | * like a regular file for ext3 to try to delete it. Things | |
120 | * will go a bit crazy if that happens, but at least we should | |
121 | * try not to panic the whole kernel. */ | |
122 | if (needed < 2) | |
123 | needed = 2; | |
124 | ||
125 | /* But we need to bound the transaction so we don't overflow the | |
126 | * journal. */ | |
ae6ddcc5 | 127 | if (needed > EXT3_MAX_TRANS_DATA) |
1da177e4 LT |
128 | needed = EXT3_MAX_TRANS_DATA; |
129 | ||
1f54587b | 130 | return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
1da177e4 LT |
131 | } |
132 | ||
ae6ddcc5 | 133 | /* |
1da177e4 LT |
134 | * Truncate transactions can be complex and absolutely huge. So we need to |
135 | * be able to restart the transaction at a conventient checkpoint to make | |
136 | * sure we don't overflow the journal. | |
137 | * | |
138 | * start_transaction gets us a new handle for a truncate transaction, | |
139 | * and extend_transaction tries to extend the existing one a bit. If | |
140 | * extend fails, we need to propagate the failure up and restart the | |
ae6ddcc5 | 141 | * transaction in the top-level truncate loop. --sct |
1da177e4 | 142 | */ |
ae6ddcc5 | 143 | static handle_t *start_transaction(struct inode *inode) |
1da177e4 LT |
144 | { |
145 | handle_t *result; | |
146 | ||
147 | result = ext3_journal_start(inode, blocks_for_truncate(inode)); | |
148 | if (!IS_ERR(result)) | |
149 | return result; | |
150 | ||
151 | ext3_std_error(inode->i_sb, PTR_ERR(result)); | |
152 | return result; | |
153 | } | |
154 | ||
155 | /* | |
156 | * Try to extend this transaction for the purposes of truncation. | |
157 | * | |
158 | * Returns 0 if we managed to create more room. If we can't create more | |
159 | * room, and the transaction must be restarted we return 1. | |
160 | */ | |
161 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
162 | { | |
163 | if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS) | |
164 | return 0; | |
165 | if (!ext3_journal_extend(handle, blocks_for_truncate(inode))) | |
166 | return 0; | |
167 | return 1; | |
168 | } | |
169 | ||
170 | /* | |
171 | * Restart the transaction associated with *handle. This does a commit, | |
172 | * so before we call here everything must be consistently dirtied against | |
173 | * this transaction. | |
174 | */ | |
00171d3c | 175 | static int truncate_restart_transaction(handle_t *handle, struct inode *inode) |
1da177e4 | 176 | { |
00171d3c JK |
177 | int ret; |
178 | ||
1da177e4 | 179 | jbd_debug(2, "restarting handle %p\n", handle); |
00171d3c JK |
180 | /* |
181 | * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle | |
182 | * At this moment, get_block can be called only for blocks inside | |
183 | * i_size since page cache has been already dropped and writes are | |
184 | * blocked by i_mutex. So we can safely drop the truncate_mutex. | |
185 | */ | |
186 | mutex_unlock(&EXT3_I(inode)->truncate_mutex); | |
187 | ret = ext3_journal_restart(handle, blocks_for_truncate(inode)); | |
188 | mutex_lock(&EXT3_I(inode)->truncate_mutex); | |
189 | return ret; | |
1da177e4 LT |
190 | } |
191 | ||
192 | /* | |
193 | * Called at the last iput() if i_nlink is zero. | |
194 | */ | |
195 | void ext3_delete_inode (struct inode * inode) | |
196 | { | |
197 | handle_t *handle; | |
198 | ||
907f4554 | 199 | if (!is_bad_inode(inode)) |
871a2931 | 200 | dquot_initialize(inode); |
907f4554 | 201 | |
fef26658 MF |
202 | truncate_inode_pages(&inode->i_data, 0); |
203 | ||
1da177e4 LT |
204 | if (is_bad_inode(inode)) |
205 | goto no_delete; | |
206 | ||
207 | handle = start_transaction(inode); | |
208 | if (IS_ERR(handle)) { | |
d6859bfc AM |
209 | /* |
210 | * If we're going to skip the normal cleanup, we still need to | |
211 | * make sure that the in-core orphan linked list is properly | |
212 | * cleaned up. | |
213 | */ | |
1da177e4 LT |
214 | ext3_orphan_del(NULL, inode); |
215 | goto no_delete; | |
216 | } | |
217 | ||
218 | if (IS_SYNC(inode)) | |
219 | handle->h_sync = 1; | |
220 | inode->i_size = 0; | |
221 | if (inode->i_blocks) | |
222 | ext3_truncate(inode); | |
223 | /* | |
224 | * Kill off the orphan record which ext3_truncate created. | |
225 | * AKPM: I think this can be inside the above `if'. | |
226 | * Note that ext3_orphan_del() has to be able to cope with the | |
227 | * deletion of a non-existent orphan - this is because we don't | |
228 | * know if ext3_truncate() actually created an orphan record. | |
229 | * (Well, we could do this if we need to, but heck - it works) | |
230 | */ | |
231 | ext3_orphan_del(handle, inode); | |
232 | EXT3_I(inode)->i_dtime = get_seconds(); | |
233 | ||
ae6ddcc5 | 234 | /* |
1da177e4 LT |
235 | * One subtle ordering requirement: if anything has gone wrong |
236 | * (transaction abort, IO errors, whatever), then we can still | |
237 | * do these next steps (the fs will already have been marked as | |
238 | * having errors), but we can't free the inode if the mark_dirty | |
ae6ddcc5 | 239 | * fails. |
1da177e4 LT |
240 | */ |
241 | if (ext3_mark_inode_dirty(handle, inode)) | |
242 | /* If that failed, just do the required in-core inode clear. */ | |
243 | clear_inode(inode); | |
244 | else | |
245 | ext3_free_inode(handle, inode); | |
246 | ext3_journal_stop(handle); | |
247 | return; | |
248 | no_delete: | |
249 | clear_inode(inode); /* We must guarantee clearing of inode... */ | |
250 | } | |
251 | ||
1da177e4 LT |
252 | typedef struct { |
253 | __le32 *p; | |
254 | __le32 key; | |
255 | struct buffer_head *bh; | |
256 | } Indirect; | |
257 | ||
258 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
259 | { | |
260 | p->key = *(p->p = v); | |
261 | p->bh = bh; | |
262 | } | |
263 | ||
d6859bfc | 264 | static int verify_chain(Indirect *from, Indirect *to) |
1da177e4 LT |
265 | { |
266 | while (from <= to && from->key == *from->p) | |
267 | from++; | |
268 | return (from > to); | |
269 | } | |
270 | ||
271 | /** | |
272 | * ext3_block_to_path - parse the block number into array of offsets | |
273 | * @inode: inode in question (we are only interested in its superblock) | |
274 | * @i_block: block number to be parsed | |
275 | * @offsets: array to store the offsets in | |
276 | * @boundary: set this non-zero if the referred-to block is likely to be | |
277 | * followed (on disk) by an indirect block. | |
278 | * | |
279 | * To store the locations of file's data ext3 uses a data structure common | |
280 | * for UNIX filesystems - tree of pointers anchored in the inode, with | |
281 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
282 | * This function translates the block number into path in that tree - | |
283 | * return value is the path length and @offsets[n] is the offset of | |
284 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
285 | * (negative or too large) warning is printed and zero returned. | |
286 | * | |
287 | * Note: function doesn't find node addresses, so no IO is needed. All | |
288 | * we need to know is the capacity of indirect blocks (taken from the | |
289 | * inode->i_sb). | |
290 | */ | |
291 | ||
292 | /* | |
293 | * Portability note: the last comparison (check that we fit into triple | |
294 | * indirect block) is spelled differently, because otherwise on an | |
295 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
296 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
297 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
298 | * i_block would have to be negative in the very beginning, so we would not | |
299 | * get there at all. | |
300 | */ | |
301 | ||
302 | static int ext3_block_to_path(struct inode *inode, | |
303 | long i_block, int offsets[4], int *boundary) | |
304 | { | |
305 | int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb); | |
306 | int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
307 | const long direct_blocks = EXT3_NDIR_BLOCKS, | |
308 | indirect_blocks = ptrs, | |
309 | double_blocks = (1 << (ptrs_bits * 2)); | |
310 | int n = 0; | |
311 | int final = 0; | |
312 | ||
313 | if (i_block < 0) { | |
314 | ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0"); | |
315 | } else if (i_block < direct_blocks) { | |
316 | offsets[n++] = i_block; | |
317 | final = direct_blocks; | |
318 | } else if ( (i_block -= direct_blocks) < indirect_blocks) { | |
319 | offsets[n++] = EXT3_IND_BLOCK; | |
320 | offsets[n++] = i_block; | |
321 | final = ptrs; | |
322 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
323 | offsets[n++] = EXT3_DIND_BLOCK; | |
324 | offsets[n++] = i_block >> ptrs_bits; | |
325 | offsets[n++] = i_block & (ptrs - 1); | |
326 | final = ptrs; | |
327 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
328 | offsets[n++] = EXT3_TIND_BLOCK; | |
329 | offsets[n++] = i_block >> (ptrs_bits * 2); | |
330 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
331 | offsets[n++] = i_block & (ptrs - 1); | |
332 | final = ptrs; | |
333 | } else { | |
d6859bfc | 334 | ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big"); |
1da177e4 LT |
335 | } |
336 | if (boundary) | |
89747d36 | 337 | *boundary = final - 1 - (i_block & (ptrs - 1)); |
1da177e4 LT |
338 | return n; |
339 | } | |
340 | ||
341 | /** | |
342 | * ext3_get_branch - read the chain of indirect blocks leading to data | |
343 | * @inode: inode in question | |
344 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
345 | * @offsets: offsets of pointers in inode/indirect blocks | |
346 | * @chain: place to store the result | |
347 | * @err: here we store the error value | |
348 | * | |
349 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
350 | * if everything went OK or the pointer to the last filled triple | |
351 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
352 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
353 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
354 | * number (it points into struct inode for i==0 and into the bh->b_data | |
355 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
356 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
357 | * numbers of the chain, addresses they were taken from (and where we can | |
358 | * verify that chain did not change) and buffer_heads hosting these | |
359 | * numbers. | |
360 | * | |
361 | * Function stops when it stumbles upon zero pointer (absent block) | |
362 | * (pointer to last triple returned, *@err == 0) | |
363 | * or when it gets an IO error reading an indirect block | |
364 | * (ditto, *@err == -EIO) | |
365 | * or when it notices that chain had been changed while it was reading | |
366 | * (ditto, *@err == -EAGAIN) | |
367 | * or when it reads all @depth-1 indirect blocks successfully and finds | |
368 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
369 | */ | |
370 | static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets, | |
371 | Indirect chain[4], int *err) | |
372 | { | |
373 | struct super_block *sb = inode->i_sb; | |
374 | Indirect *p = chain; | |
375 | struct buffer_head *bh; | |
376 | ||
377 | *err = 0; | |
378 | /* i_data is not going away, no lock needed */ | |
379 | add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets); | |
380 | if (!p->key) | |
381 | goto no_block; | |
382 | while (--depth) { | |
383 | bh = sb_bread(sb, le32_to_cpu(p->key)); | |
384 | if (!bh) | |
385 | goto failure; | |
386 | /* Reader: pointers */ | |
387 | if (!verify_chain(chain, p)) | |
388 | goto changed; | |
389 | add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); | |
390 | /* Reader: end */ | |
391 | if (!p->key) | |
392 | goto no_block; | |
393 | } | |
394 | return NULL; | |
395 | ||
396 | changed: | |
397 | brelse(bh); | |
398 | *err = -EAGAIN; | |
399 | goto no_block; | |
400 | failure: | |
401 | *err = -EIO; | |
402 | no_block: | |
403 | return p; | |
404 | } | |
405 | ||
406 | /** | |
407 | * ext3_find_near - find a place for allocation with sufficient locality | |
408 | * @inode: owner | |
409 | * @ind: descriptor of indirect block. | |
410 | * | |
1cc8dcf5 | 411 | * This function returns the preferred place for block allocation. |
1da177e4 LT |
412 | * It is used when heuristic for sequential allocation fails. |
413 | * Rules are: | |
414 | * + if there is a block to the left of our position - allocate near it. | |
415 | * + if pointer will live in indirect block - allocate near that block. | |
416 | * + if pointer will live in inode - allocate in the same | |
ae6ddcc5 | 417 | * cylinder group. |
1da177e4 LT |
418 | * |
419 | * In the latter case we colour the starting block by the callers PID to | |
420 | * prevent it from clashing with concurrent allocations for a different inode | |
421 | * in the same block group. The PID is used here so that functionally related | |
422 | * files will be close-by on-disk. | |
423 | * | |
424 | * Caller must make sure that @ind is valid and will stay that way. | |
425 | */ | |
43d23f90 | 426 | static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind) |
1da177e4 LT |
427 | { |
428 | struct ext3_inode_info *ei = EXT3_I(inode); | |
429 | __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data; | |
430 | __le32 *p; | |
43d23f90 MC |
431 | ext3_fsblk_t bg_start; |
432 | ext3_grpblk_t colour; | |
1da177e4 LT |
433 | |
434 | /* Try to find previous block */ | |
d6859bfc | 435 | for (p = ind->p - 1; p >= start; p--) { |
1da177e4 LT |
436 | if (*p) |
437 | return le32_to_cpu(*p); | |
d6859bfc | 438 | } |
1da177e4 LT |
439 | |
440 | /* No such thing, so let's try location of indirect block */ | |
441 | if (ind->bh) | |
442 | return ind->bh->b_blocknr; | |
443 | ||
444 | /* | |
d6859bfc AM |
445 | * It is going to be referred to from the inode itself? OK, just put it |
446 | * into the same cylinder group then. | |
1da177e4 | 447 | */ |
43d23f90 | 448 | bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group); |
1da177e4 LT |
449 | colour = (current->pid % 16) * |
450 | (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16); | |
451 | return bg_start + colour; | |
452 | } | |
453 | ||
454 | /** | |
1cc8dcf5 | 455 | * ext3_find_goal - find a preferred place for allocation. |
1da177e4 LT |
456 | * @inode: owner |
457 | * @block: block we want | |
1da177e4 | 458 | * @partial: pointer to the last triple within a chain |
1da177e4 | 459 | * |
1cc8dcf5 | 460 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 461 | * returns it. |
1da177e4 LT |
462 | */ |
463 | ||
43d23f90 | 464 | static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block, |
fb01bfda | 465 | Indirect *partial) |
1da177e4 | 466 | { |
d6859bfc AM |
467 | struct ext3_block_alloc_info *block_i; |
468 | ||
469 | block_i = EXT3_I(inode)->i_block_alloc_info; | |
1da177e4 LT |
470 | |
471 | /* | |
472 | * try the heuristic for sequential allocation, | |
473 | * failing that at least try to get decent locality. | |
474 | */ | |
475 | if (block_i && (block == block_i->last_alloc_logical_block + 1) | |
476 | && (block_i->last_alloc_physical_block != 0)) { | |
fe55c452 | 477 | return block_i->last_alloc_physical_block + 1; |
1da177e4 LT |
478 | } |
479 | ||
fe55c452 | 480 | return ext3_find_near(inode, partial); |
1da177e4 | 481 | } |
d6859bfc | 482 | |
b47b2478 MC |
483 | /** |
484 | * ext3_blks_to_allocate: Look up the block map and count the number | |
485 | * of direct blocks need to be allocated for the given branch. | |
486 | * | |
e9ad5620 | 487 | * @branch: chain of indirect blocks |
b47b2478 MC |
488 | * @k: number of blocks need for indirect blocks |
489 | * @blks: number of data blocks to be mapped. | |
490 | * @blocks_to_boundary: the offset in the indirect block | |
491 | * | |
492 | * return the total number of blocks to be allocate, including the | |
493 | * direct and indirect blocks. | |
494 | */ | |
d6859bfc | 495 | static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks, |
b47b2478 MC |
496 | int blocks_to_boundary) |
497 | { | |
498 | unsigned long count = 0; | |
499 | ||
500 | /* | |
501 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
502 | * then it's clear blocks on that path have not allocated | |
503 | */ | |
504 | if (k > 0) { | |
d6859bfc | 505 | /* right now we don't handle cross boundary allocation */ |
b47b2478 MC |
506 | if (blks < blocks_to_boundary + 1) |
507 | count += blks; | |
508 | else | |
509 | count += blocks_to_boundary + 1; | |
510 | return count; | |
511 | } | |
512 | ||
513 | count++; | |
514 | while (count < blks && count <= blocks_to_boundary && | |
515 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
516 | count++; | |
517 | } | |
518 | return count; | |
519 | } | |
520 | ||
521 | /** | |
522 | * ext3_alloc_blocks: multiple allocate blocks needed for a branch | |
523 | * @indirect_blks: the number of blocks need to allocate for indirect | |
524 | * blocks | |
525 | * | |
526 | * @new_blocks: on return it will store the new block numbers for | |
527 | * the indirect blocks(if needed) and the first direct block, | |
528 | * @blks: on return it will store the total number of allocated | |
529 | * direct blocks | |
530 | */ | |
531 | static int ext3_alloc_blocks(handle_t *handle, struct inode *inode, | |
43d23f90 MC |
532 | ext3_fsblk_t goal, int indirect_blks, int blks, |
533 | ext3_fsblk_t new_blocks[4], int *err) | |
b47b2478 MC |
534 | { |
535 | int target, i; | |
536 | unsigned long count = 0; | |
537 | int index = 0; | |
43d23f90 | 538 | ext3_fsblk_t current_block = 0; |
b47b2478 MC |
539 | int ret = 0; |
540 | ||
541 | /* | |
542 | * Here we try to allocate the requested multiple blocks at once, | |
543 | * on a best-effort basis. | |
544 | * To build a branch, we should allocate blocks for | |
545 | * the indirect blocks(if not allocated yet), and at least | |
546 | * the first direct block of this branch. That's the | |
547 | * minimum number of blocks need to allocate(required) | |
548 | */ | |
549 | target = blks + indirect_blks; | |
550 | ||
551 | while (1) { | |
552 | count = target; | |
553 | /* allocating blocks for indirect blocks and direct blocks */ | |
d6859bfc | 554 | current_block = ext3_new_blocks(handle,inode,goal,&count,err); |
b47b2478 MC |
555 | if (*err) |
556 | goto failed_out; | |
557 | ||
558 | target -= count; | |
559 | /* allocate blocks for indirect blocks */ | |
560 | while (index < indirect_blks && count) { | |
561 | new_blocks[index++] = current_block++; | |
562 | count--; | |
563 | } | |
564 | ||
565 | if (count > 0) | |
566 | break; | |
567 | } | |
568 | ||
569 | /* save the new block number for the first direct block */ | |
570 | new_blocks[index] = current_block; | |
571 | ||
572 | /* total number of blocks allocated for direct blocks */ | |
573 | ret = count; | |
574 | *err = 0; | |
575 | return ret; | |
576 | failed_out: | |
577 | for (i = 0; i <index; i++) | |
578 | ext3_free_blocks(handle, inode, new_blocks[i], 1); | |
579 | return ret; | |
580 | } | |
1da177e4 LT |
581 | |
582 | /** | |
583 | * ext3_alloc_branch - allocate and set up a chain of blocks. | |
584 | * @inode: owner | |
b47b2478 MC |
585 | * @indirect_blks: number of allocated indirect blocks |
586 | * @blks: number of allocated direct blocks | |
1da177e4 LT |
587 | * @offsets: offsets (in the blocks) to store the pointers to next. |
588 | * @branch: place to store the chain in. | |
589 | * | |
b47b2478 | 590 | * This function allocates blocks, zeroes out all but the last one, |
1da177e4 LT |
591 | * links them into chain and (if we are synchronous) writes them to disk. |
592 | * In other words, it prepares a branch that can be spliced onto the | |
593 | * inode. It stores the information about that chain in the branch[], in | |
594 | * the same format as ext3_get_branch() would do. We are calling it after | |
595 | * we had read the existing part of chain and partial points to the last | |
596 | * triple of that (one with zero ->key). Upon the exit we have the same | |
5b116879 | 597 | * picture as after the successful ext3_get_block(), except that in one |
1da177e4 LT |
598 | * place chain is disconnected - *branch->p is still zero (we did not |
599 | * set the last link), but branch->key contains the number that should | |
600 | * be placed into *branch->p to fill that gap. | |
601 | * | |
602 | * If allocation fails we free all blocks we've allocated (and forget | |
603 | * their buffer_heads) and return the error value the from failed | |
604 | * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain | |
605 | * as described above and return 0. | |
606 | */ | |
1da177e4 | 607 | static int ext3_alloc_branch(handle_t *handle, struct inode *inode, |
43d23f90 | 608 | int indirect_blks, int *blks, ext3_fsblk_t goal, |
b47b2478 | 609 | int *offsets, Indirect *branch) |
1da177e4 LT |
610 | { |
611 | int blocksize = inode->i_sb->s_blocksize; | |
b47b2478 | 612 | int i, n = 0; |
1da177e4 | 613 | int err = 0; |
b47b2478 MC |
614 | struct buffer_head *bh; |
615 | int num; | |
43d23f90 MC |
616 | ext3_fsblk_t new_blocks[4]; |
617 | ext3_fsblk_t current_block; | |
1da177e4 | 618 | |
b47b2478 MC |
619 | num = ext3_alloc_blocks(handle, inode, goal, indirect_blks, |
620 | *blks, new_blocks, &err); | |
621 | if (err) | |
622 | return err; | |
1da177e4 | 623 | |
b47b2478 MC |
624 | branch[0].key = cpu_to_le32(new_blocks[0]); |
625 | /* | |
626 | * metadata blocks and data blocks are allocated. | |
627 | */ | |
628 | for (n = 1; n <= indirect_blks; n++) { | |
629 | /* | |
630 | * Get buffer_head for parent block, zero it out | |
631 | * and set the pointer to new one, then send | |
632 | * parent to disk. | |
633 | */ | |
634 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
635 | branch[n].bh = bh; | |
636 | lock_buffer(bh); | |
637 | BUFFER_TRACE(bh, "call get_create_access"); | |
638 | err = ext3_journal_get_create_access(handle, bh); | |
639 | if (err) { | |
1da177e4 | 640 | unlock_buffer(bh); |
b47b2478 MC |
641 | brelse(bh); |
642 | goto failed; | |
643 | } | |
1da177e4 | 644 | |
b47b2478 MC |
645 | memset(bh->b_data, 0, blocksize); |
646 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
647 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
648 | *branch[n].p = branch[n].key; | |
649 | if ( n == indirect_blks) { | |
650 | current_block = new_blocks[n]; | |
651 | /* | |
652 | * End of chain, update the last new metablock of | |
653 | * the chain to point to the new allocated | |
654 | * data blocks numbers | |
655 | */ | |
656 | for (i=1; i < num; i++) | |
657 | *(branch[n].p + i) = cpu_to_le32(++current_block); | |
1da177e4 | 658 | } |
b47b2478 MC |
659 | BUFFER_TRACE(bh, "marking uptodate"); |
660 | set_buffer_uptodate(bh); | |
661 | unlock_buffer(bh); | |
1da177e4 | 662 | |
b47b2478 MC |
663 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
664 | err = ext3_journal_dirty_metadata(handle, bh); | |
665 | if (err) | |
666 | goto failed; | |
667 | } | |
668 | *blks = num; | |
669 | return err; | |
670 | failed: | |
1da177e4 | 671 | /* Allocation failed, free what we already allocated */ |
b47b2478 | 672 | for (i = 1; i <= n ; i++) { |
1da177e4 LT |
673 | BUFFER_TRACE(branch[i].bh, "call journal_forget"); |
674 | ext3_journal_forget(handle, branch[i].bh); | |
675 | } | |
b47b2478 MC |
676 | for (i = 0; i <indirect_blks; i++) |
677 | ext3_free_blocks(handle, inode, new_blocks[i], 1); | |
678 | ||
679 | ext3_free_blocks(handle, inode, new_blocks[i], num); | |
680 | ||
1da177e4 LT |
681 | return err; |
682 | } | |
683 | ||
684 | /** | |
d6859bfc AM |
685 | * ext3_splice_branch - splice the allocated branch onto inode. |
686 | * @inode: owner | |
687 | * @block: (logical) number of block we are adding | |
688 | * @chain: chain of indirect blocks (with a missing link - see | |
689 | * ext3_alloc_branch) | |
690 | * @where: location of missing link | |
691 | * @num: number of indirect blocks we are adding | |
692 | * @blks: number of direct blocks we are adding | |
693 | * | |
694 | * This function fills the missing link and does all housekeeping needed in | |
695 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
696 | * chain to new block and return 0. | |
1da177e4 | 697 | */ |
d6859bfc AM |
698 | static int ext3_splice_branch(handle_t *handle, struct inode *inode, |
699 | long block, Indirect *where, int num, int blks) | |
1da177e4 LT |
700 | { |
701 | int i; | |
702 | int err = 0; | |
d6859bfc | 703 | struct ext3_block_alloc_info *block_i; |
43d23f90 | 704 | ext3_fsblk_t current_block; |
fe8bc91c | 705 | struct ext3_inode_info *ei = EXT3_I(inode); |
d6859bfc | 706 | |
fe8bc91c | 707 | block_i = ei->i_block_alloc_info; |
1da177e4 LT |
708 | /* |
709 | * If we're splicing into a [td]indirect block (as opposed to the | |
710 | * inode) then we need to get write access to the [td]indirect block | |
711 | * before the splice. | |
712 | */ | |
713 | if (where->bh) { | |
714 | BUFFER_TRACE(where->bh, "get_write_access"); | |
715 | err = ext3_journal_get_write_access(handle, where->bh); | |
716 | if (err) | |
717 | goto err_out; | |
718 | } | |
1da177e4 LT |
719 | /* That's it */ |
720 | ||
721 | *where->p = where->key; | |
d6859bfc AM |
722 | |
723 | /* | |
724 | * Update the host buffer_head or inode to point to more just allocated | |
725 | * direct blocks blocks | |
726 | */ | |
b47b2478 | 727 | if (num == 0 && blks > 1) { |
5dea5176 | 728 | current_block = le32_to_cpu(where->key) + 1; |
b47b2478 MC |
729 | for (i = 1; i < blks; i++) |
730 | *(where->p + i ) = cpu_to_le32(current_block++); | |
731 | } | |
1da177e4 LT |
732 | |
733 | /* | |
734 | * update the most recently allocated logical & physical block | |
735 | * in i_block_alloc_info, to assist find the proper goal block for next | |
736 | * allocation | |
737 | */ | |
738 | if (block_i) { | |
b47b2478 | 739 | block_i->last_alloc_logical_block = block + blks - 1; |
d6859bfc | 740 | block_i->last_alloc_physical_block = |
5dea5176 | 741 | le32_to_cpu(where[num].key) + blks - 1; |
1da177e4 LT |
742 | } |
743 | ||
744 | /* We are done with atomic stuff, now do the rest of housekeeping */ | |
745 | ||
746 | inode->i_ctime = CURRENT_TIME_SEC; | |
747 | ext3_mark_inode_dirty(handle, inode); | |
fe8bc91c JK |
748 | /* ext3_mark_inode_dirty already updated i_sync_tid */ |
749 | atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid); | |
1da177e4 LT |
750 | |
751 | /* had we spliced it onto indirect block? */ | |
752 | if (where->bh) { | |
753 | /* | |
d6859bfc | 754 | * If we spliced it onto an indirect block, we haven't |
1da177e4 LT |
755 | * altered the inode. Note however that if it is being spliced |
756 | * onto an indirect block at the very end of the file (the | |
757 | * file is growing) then we *will* alter the inode to reflect | |
758 | * the new i_size. But that is not done here - it is done in | |
759 | * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode. | |
760 | */ | |
761 | jbd_debug(5, "splicing indirect only\n"); | |
762 | BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata"); | |
763 | err = ext3_journal_dirty_metadata(handle, where->bh); | |
ae6ddcc5 | 764 | if (err) |
1da177e4 LT |
765 | goto err_out; |
766 | } else { | |
767 | /* | |
768 | * OK, we spliced it into the inode itself on a direct block. | |
769 | * Inode was dirtied above. | |
770 | */ | |
771 | jbd_debug(5, "splicing direct\n"); | |
772 | } | |
773 | return err; | |
774 | ||
1da177e4 | 775 | err_out: |
b47b2478 | 776 | for (i = 1; i <= num; i++) { |
1da177e4 LT |
777 | BUFFER_TRACE(where[i].bh, "call journal_forget"); |
778 | ext3_journal_forget(handle, where[i].bh); | |
d6859bfc | 779 | ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1); |
1da177e4 | 780 | } |
b47b2478 MC |
781 | ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks); |
782 | ||
1da177e4 LT |
783 | return err; |
784 | } | |
785 | ||
786 | /* | |
787 | * Allocation strategy is simple: if we have to allocate something, we will | |
788 | * have to go the whole way to leaf. So let's do it before attaching anything | |
789 | * to tree, set linkage between the newborn blocks, write them if sync is | |
790 | * required, recheck the path, free and repeat if check fails, otherwise | |
791 | * set the last missing link (that will protect us from any truncate-generated | |
792 | * removals - all blocks on the path are immune now) and possibly force the | |
793 | * write on the parent block. | |
794 | * That has a nice additional property: no special recovery from the failed | |
795 | * allocations is needed - we simply release blocks and do not touch anything | |
796 | * reachable from inode. | |
797 | * | |
d6859bfc | 798 | * `handle' can be NULL if create == 0. |
1da177e4 LT |
799 | * |
800 | * The BKL may not be held on entry here. Be sure to take it early. | |
89747d36 MC |
801 | * return > 0, # of blocks mapped or allocated. |
802 | * return = 0, if plain lookup failed. | |
803 | * return < 0, error case. | |
1da177e4 | 804 | */ |
d6859bfc AM |
805 | int ext3_get_blocks_handle(handle_t *handle, struct inode *inode, |
806 | sector_t iblock, unsigned long maxblocks, | |
807 | struct buffer_head *bh_result, | |
43237b54 | 808 | int create) |
1da177e4 LT |
809 | { |
810 | int err = -EIO; | |
811 | int offsets[4]; | |
812 | Indirect chain[4]; | |
813 | Indirect *partial; | |
43d23f90 | 814 | ext3_fsblk_t goal; |
b47b2478 | 815 | int indirect_blks; |
89747d36 MC |
816 | int blocks_to_boundary = 0; |
817 | int depth; | |
1da177e4 | 818 | struct ext3_inode_info *ei = EXT3_I(inode); |
89747d36 | 819 | int count = 0; |
43d23f90 | 820 | ext3_fsblk_t first_block = 0; |
89747d36 | 821 | |
1da177e4 LT |
822 | |
823 | J_ASSERT(handle != NULL || create == 0); | |
d6859bfc | 824 | depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary); |
1da177e4 LT |
825 | |
826 | if (depth == 0) | |
827 | goto out; | |
828 | ||
1da177e4 LT |
829 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
830 | ||
831 | /* Simplest case - block found, no allocation needed */ | |
832 | if (!partial) { | |
5dea5176 | 833 | first_block = le32_to_cpu(chain[depth - 1].key); |
1da177e4 | 834 | clear_buffer_new(bh_result); |
89747d36 MC |
835 | count++; |
836 | /*map more blocks*/ | |
837 | while (count < maxblocks && count <= blocks_to_boundary) { | |
43d23f90 | 838 | ext3_fsblk_t blk; |
5dea5176 | 839 | |
e8ef7aae | 840 | if (!verify_chain(chain, chain + depth - 1)) { |
89747d36 MC |
841 | /* |
842 | * Indirect block might be removed by | |
843 | * truncate while we were reading it. | |
844 | * Handling of that case: forget what we've | |
845 | * got now. Flag the err as EAGAIN, so it | |
846 | * will reread. | |
847 | */ | |
848 | err = -EAGAIN; | |
849 | count = 0; | |
850 | break; | |
851 | } | |
5dea5176 MC |
852 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
853 | ||
854 | if (blk == first_block + count) | |
89747d36 MC |
855 | count++; |
856 | else | |
857 | break; | |
858 | } | |
859 | if (err != -EAGAIN) | |
860 | goto got_it; | |
1da177e4 LT |
861 | } |
862 | ||
863 | /* Next simple case - plain lookup or failed read of indirect block */ | |
fe55c452 MC |
864 | if (!create || err == -EIO) |
865 | goto cleanup; | |
866 | ||
97461518 | 867 | mutex_lock(&ei->truncate_mutex); |
fe55c452 MC |
868 | |
869 | /* | |
870 | * If the indirect block is missing while we are reading | |
871 | * the chain(ext3_get_branch() returns -EAGAIN err), or | |
872 | * if the chain has been changed after we grab the semaphore, | |
873 | * (either because another process truncated this branch, or | |
874 | * another get_block allocated this branch) re-grab the chain to see if | |
875 | * the request block has been allocated or not. | |
876 | * | |
877 | * Since we already block the truncate/other get_block | |
878 | * at this point, we will have the current copy of the chain when we | |
879 | * splice the branch into the tree. | |
880 | */ | |
881 | if (err == -EAGAIN || !verify_chain(chain, partial)) { | |
1da177e4 | 882 | while (partial > chain) { |
1da177e4 LT |
883 | brelse(partial->bh); |
884 | partial--; | |
885 | } | |
fe55c452 MC |
886 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
887 | if (!partial) { | |
89747d36 | 888 | count++; |
97461518 | 889 | mutex_unlock(&ei->truncate_mutex); |
fe55c452 MC |
890 | if (err) |
891 | goto cleanup; | |
892 | clear_buffer_new(bh_result); | |
893 | goto got_it; | |
894 | } | |
1da177e4 LT |
895 | } |
896 | ||
897 | /* | |
fe55c452 MC |
898 | * Okay, we need to do block allocation. Lazily initialize the block |
899 | * allocation info here if necessary | |
900 | */ | |
901 | if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) | |
1da177e4 | 902 | ext3_init_block_alloc_info(inode); |
1da177e4 | 903 | |
fb01bfda | 904 | goal = ext3_find_goal(inode, iblock, partial); |
1da177e4 | 905 | |
b47b2478 MC |
906 | /* the number of blocks need to allocate for [d,t]indirect blocks */ |
907 | indirect_blks = (chain + depth) - partial - 1; | |
1da177e4 | 908 | |
b47b2478 MC |
909 | /* |
910 | * Next look up the indirect map to count the totoal number of | |
911 | * direct blocks to allocate for this branch. | |
912 | */ | |
913 | count = ext3_blks_to_allocate(partial, indirect_blks, | |
914 | maxblocks, blocks_to_boundary); | |
1da177e4 LT |
915 | /* |
916 | * Block out ext3_truncate while we alter the tree | |
917 | */ | |
b47b2478 | 918 | err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal, |
fe55c452 | 919 | offsets + (partial - chain), partial); |
1da177e4 | 920 | |
fe55c452 MC |
921 | /* |
922 | * The ext3_splice_branch call will free and forget any buffers | |
1da177e4 LT |
923 | * on the new chain if there is a failure, but that risks using |
924 | * up transaction credits, especially for bitmaps where the | |
925 | * credits cannot be returned. Can we handle this somehow? We | |
fe55c452 MC |
926 | * may need to return -EAGAIN upwards in the worst case. --sct |
927 | */ | |
1da177e4 | 928 | if (!err) |
b47b2478 MC |
929 | err = ext3_splice_branch(handle, inode, iblock, |
930 | partial, indirect_blks, count); | |
97461518 | 931 | mutex_unlock(&ei->truncate_mutex); |
1da177e4 LT |
932 | if (err) |
933 | goto cleanup; | |
934 | ||
935 | set_buffer_new(bh_result); | |
fe55c452 MC |
936 | got_it: |
937 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); | |
20acaa18 | 938 | if (count > blocks_to_boundary) |
fe55c452 | 939 | set_buffer_boundary(bh_result); |
89747d36 | 940 | err = count; |
fe55c452 MC |
941 | /* Clean up and exit */ |
942 | partial = chain + depth - 1; /* the whole chain */ | |
943 | cleanup: | |
1da177e4 | 944 | while (partial > chain) { |
fe55c452 | 945 | BUFFER_TRACE(partial->bh, "call brelse"); |
1da177e4 LT |
946 | brelse(partial->bh); |
947 | partial--; | |
948 | } | |
fe55c452 MC |
949 | BUFFER_TRACE(bh_result, "returned"); |
950 | out: | |
951 | return err; | |
1da177e4 LT |
952 | } |
953 | ||
bd1939de JK |
954 | /* Maximum number of blocks we map for direct IO at once. */ |
955 | #define DIO_MAX_BLOCKS 4096 | |
956 | /* | |
957 | * Number of credits we need for writing DIO_MAX_BLOCKS: | |
958 | * We need sb + group descriptor + bitmap + inode -> 4 | |
959 | * For B blocks with A block pointers per block we need: | |
960 | * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect). | |
961 | * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25. | |
962 | */ | |
963 | #define DIO_CREDITS 25 | |
1da177e4 | 964 | |
f91a2ad2 BP |
965 | static int ext3_get_block(struct inode *inode, sector_t iblock, |
966 | struct buffer_head *bh_result, int create) | |
1da177e4 | 967 | { |
3e4fdaf8 | 968 | handle_t *handle = ext3_journal_current_handle(); |
bd1939de | 969 | int ret = 0, started = 0; |
1d8fa7a2 | 970 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
1da177e4 | 971 | |
bd1939de JK |
972 | if (create && !handle) { /* Direct IO write... */ |
973 | if (max_blocks > DIO_MAX_BLOCKS) | |
974 | max_blocks = DIO_MAX_BLOCKS; | |
975 | handle = ext3_journal_start(inode, DIO_CREDITS + | |
c459001f | 976 | EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb)); |
bd1939de | 977 | if (IS_ERR(handle)) { |
1da177e4 | 978 | ret = PTR_ERR(handle); |
bd1939de | 979 | goto out; |
1da177e4 | 980 | } |
bd1939de | 981 | started = 1; |
1da177e4 LT |
982 | } |
983 | ||
bd1939de | 984 | ret = ext3_get_blocks_handle(handle, inode, iblock, |
43237b54 | 985 | max_blocks, bh_result, create); |
bd1939de JK |
986 | if (ret > 0) { |
987 | bh_result->b_size = (ret << inode->i_blkbits); | |
988 | ret = 0; | |
89747d36 | 989 | } |
bd1939de JK |
990 | if (started) |
991 | ext3_journal_stop(handle); | |
992 | out: | |
1da177e4 LT |
993 | return ret; |
994 | } | |
995 | ||
68c9d702 JB |
996 | int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
997 | u64 start, u64 len) | |
998 | { | |
999 | return generic_block_fiemap(inode, fieinfo, start, len, | |
1000 | ext3_get_block); | |
1001 | } | |
1002 | ||
1da177e4 LT |
1003 | /* |
1004 | * `handle' can be NULL if create is zero | |
1005 | */ | |
d6859bfc AM |
1006 | struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode, |
1007 | long block, int create, int *errp) | |
1da177e4 LT |
1008 | { |
1009 | struct buffer_head dummy; | |
1010 | int fatal = 0, err; | |
1011 | ||
1012 | J_ASSERT(handle != NULL || create == 0); | |
1013 | ||
1014 | dummy.b_state = 0; | |
1015 | dummy.b_blocknr = -1000; | |
1016 | buffer_trace_init(&dummy.b_history); | |
89747d36 | 1017 | err = ext3_get_blocks_handle(handle, inode, block, 1, |
43237b54 | 1018 | &dummy, create); |
3665d0e5 BP |
1019 | /* |
1020 | * ext3_get_blocks_handle() returns number of blocks | |
1021 | * mapped. 0 in case of a HOLE. | |
1022 | */ | |
1023 | if (err > 0) { | |
1024 | if (err > 1) | |
1025 | WARN_ON(1); | |
89747d36 | 1026 | err = 0; |
89747d36 MC |
1027 | } |
1028 | *errp = err; | |
1029 | if (!err && buffer_mapped(&dummy)) { | |
1da177e4 LT |
1030 | struct buffer_head *bh; |
1031 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); | |
2973dfdb GOC |
1032 | if (!bh) { |
1033 | *errp = -EIO; | |
1034 | goto err; | |
1035 | } | |
1da177e4 LT |
1036 | if (buffer_new(&dummy)) { |
1037 | J_ASSERT(create != 0); | |
c80544dc | 1038 | J_ASSERT(handle != NULL); |
1da177e4 | 1039 | |
d6859bfc AM |
1040 | /* |
1041 | * Now that we do not always journal data, we should | |
1042 | * keep in mind whether this should always journal the | |
1043 | * new buffer as metadata. For now, regular file | |
1044 | * writes use ext3_get_block instead, so it's not a | |
1045 | * problem. | |
1046 | */ | |
1da177e4 LT |
1047 | lock_buffer(bh); |
1048 | BUFFER_TRACE(bh, "call get_create_access"); | |
1049 | fatal = ext3_journal_get_create_access(handle, bh); | |
1050 | if (!fatal && !buffer_uptodate(bh)) { | |
d6859bfc | 1051 | memset(bh->b_data,0,inode->i_sb->s_blocksize); |
1da177e4 LT |
1052 | set_buffer_uptodate(bh); |
1053 | } | |
1054 | unlock_buffer(bh); | |
1055 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
1056 | err = ext3_journal_dirty_metadata(handle, bh); | |
1057 | if (!fatal) | |
1058 | fatal = err; | |
1059 | } else { | |
1060 | BUFFER_TRACE(bh, "not a new buffer"); | |
1061 | } | |
1062 | if (fatal) { | |
1063 | *errp = fatal; | |
1064 | brelse(bh); | |
1065 | bh = NULL; | |
1066 | } | |
1067 | return bh; | |
1068 | } | |
2973dfdb | 1069 | err: |
1da177e4 LT |
1070 | return NULL; |
1071 | } | |
1072 | ||
d6859bfc | 1073 | struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode, |
1da177e4 LT |
1074 | int block, int create, int *err) |
1075 | { | |
1076 | struct buffer_head * bh; | |
1077 | ||
1078 | bh = ext3_getblk(handle, inode, block, create, err); | |
1079 | if (!bh) | |
1080 | return bh; | |
1081 | if (buffer_uptodate(bh)) | |
1082 | return bh; | |
caa38fb0 | 1083 | ll_rw_block(READ_META, 1, &bh); |
1da177e4 LT |
1084 | wait_on_buffer(bh); |
1085 | if (buffer_uptodate(bh)) | |
1086 | return bh; | |
1087 | put_bh(bh); | |
1088 | *err = -EIO; | |
1089 | return NULL; | |
1090 | } | |
1091 | ||
1092 | static int walk_page_buffers( handle_t *handle, | |
1093 | struct buffer_head *head, | |
1094 | unsigned from, | |
1095 | unsigned to, | |
1096 | int *partial, | |
1097 | int (*fn)( handle_t *handle, | |
1098 | struct buffer_head *bh)) | |
1099 | { | |
1100 | struct buffer_head *bh; | |
1101 | unsigned block_start, block_end; | |
1102 | unsigned blocksize = head->b_size; | |
1103 | int err, ret = 0; | |
1104 | struct buffer_head *next; | |
1105 | ||
1106 | for ( bh = head, block_start = 0; | |
1107 | ret == 0 && (bh != head || !block_start); | |
e9ad5620 | 1108 | block_start = block_end, bh = next) |
1da177e4 LT |
1109 | { |
1110 | next = bh->b_this_page; | |
1111 | block_end = block_start + blocksize; | |
1112 | if (block_end <= from || block_start >= to) { | |
1113 | if (partial && !buffer_uptodate(bh)) | |
1114 | *partial = 1; | |
1115 | continue; | |
1116 | } | |
1117 | err = (*fn)(handle, bh); | |
1118 | if (!ret) | |
1119 | ret = err; | |
1120 | } | |
1121 | return ret; | |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * To preserve ordering, it is essential that the hole instantiation and | |
1126 | * the data write be encapsulated in a single transaction. We cannot | |
1127 | * close off a transaction and start a new one between the ext3_get_block() | |
1128 | * and the commit_write(). So doing the journal_start at the start of | |
1129 | * prepare_write() is the right place. | |
1130 | * | |
1131 | * Also, this function can nest inside ext3_writepage() -> | |
1132 | * block_write_full_page(). In that case, we *know* that ext3_writepage() | |
1133 | * has generated enough buffer credits to do the whole page. So we won't | |
1134 | * block on the journal in that case, which is good, because the caller may | |
1135 | * be PF_MEMALLOC. | |
1136 | * | |
1137 | * By accident, ext3 can be reentered when a transaction is open via | |
1138 | * quota file writes. If we were to commit the transaction while thus | |
1139 | * reentered, there can be a deadlock - we would be holding a quota | |
1140 | * lock, and the commit would never complete if another thread had a | |
1141 | * transaction open and was blocking on the quota lock - a ranking | |
1142 | * violation. | |
1143 | * | |
1144 | * So what we do is to rely on the fact that journal_stop/journal_start | |
1145 | * will _not_ run commit under these circumstances because handle->h_ref | |
1146 | * is elevated. We'll still have enough credits for the tiny quotafile | |
ae6ddcc5 | 1147 | * write. |
1da177e4 | 1148 | */ |
d6859bfc AM |
1149 | static int do_journal_get_write_access(handle_t *handle, |
1150 | struct buffer_head *bh) | |
1da177e4 LT |
1151 | { |
1152 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1153 | return 0; | |
1154 | return ext3_journal_get_write_access(handle, bh); | |
1155 | } | |
1156 | ||
68eb3db0 JK |
1157 | /* |
1158 | * Truncate blocks that were not used by write. We have to truncate the | |
1159 | * pagecache as well so that corresponding buffers get properly unmapped. | |
1160 | */ | |
1161 | static void ext3_truncate_failed_write(struct inode *inode) | |
1162 | { | |
1163 | truncate_inode_pages(inode->i_mapping, inode->i_size); | |
1164 | ext3_truncate(inode); | |
1165 | } | |
1166 | ||
f4fc66a8 NP |
1167 | static int ext3_write_begin(struct file *file, struct address_space *mapping, |
1168 | loff_t pos, unsigned len, unsigned flags, | |
1169 | struct page **pagep, void **fsdata) | |
1da177e4 | 1170 | { |
f4fc66a8 | 1171 | struct inode *inode = mapping->host; |
695f6ae0 | 1172 | int ret; |
1da177e4 LT |
1173 | handle_t *handle; |
1174 | int retries = 0; | |
f4fc66a8 NP |
1175 | struct page *page; |
1176 | pgoff_t index; | |
1177 | unsigned from, to; | |
695f6ae0 JK |
1178 | /* Reserve one block more for addition to orphan list in case |
1179 | * we allocate blocks but write fails for some reason */ | |
1180 | int needed_blocks = ext3_writepage_trans_blocks(inode) + 1; | |
f4fc66a8 NP |
1181 | |
1182 | index = pos >> PAGE_CACHE_SHIFT; | |
1183 | from = pos & (PAGE_CACHE_SIZE - 1); | |
1184 | to = from + len; | |
1da177e4 LT |
1185 | |
1186 | retry: | |
54566b2c | 1187 | page = grab_cache_page_write_begin(mapping, index, flags); |
f4fc66a8 NP |
1188 | if (!page) |
1189 | return -ENOMEM; | |
1190 | *pagep = page; | |
1191 | ||
1da177e4 | 1192 | handle = ext3_journal_start(inode, needed_blocks); |
1aa9b4b9 | 1193 | if (IS_ERR(handle)) { |
f4fc66a8 NP |
1194 | unlock_page(page); |
1195 | page_cache_release(page); | |
1aa9b4b9 AM |
1196 | ret = PTR_ERR(handle); |
1197 | goto out; | |
1198 | } | |
f4fc66a8 NP |
1199 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
1200 | ext3_get_block); | |
1da177e4 | 1201 | if (ret) |
f4fc66a8 | 1202 | goto write_begin_failed; |
1da177e4 LT |
1203 | |
1204 | if (ext3_should_journal_data(inode)) { | |
1205 | ret = walk_page_buffers(handle, page_buffers(page), | |
1206 | from, to, NULL, do_journal_get_write_access); | |
1207 | } | |
f4fc66a8 NP |
1208 | write_begin_failed: |
1209 | if (ret) { | |
5ec8b75e AK |
1210 | /* |
1211 | * block_write_begin may have instantiated a few blocks | |
1212 | * outside i_size. Trim these off again. Don't need | |
1213 | * i_size_read because we hold i_mutex. | |
695f6ae0 JK |
1214 | * |
1215 | * Add inode to orphan list in case we crash before truncate | |
9eaaa2d5 JK |
1216 | * finishes. Do this only if ext3_can_truncate() agrees so |
1217 | * that orphan processing code is happy. | |
5ec8b75e | 1218 | */ |
9eaaa2d5 | 1219 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 JK |
1220 | ext3_orphan_add(handle, inode); |
1221 | ext3_journal_stop(handle); | |
1222 | unlock_page(page); | |
1223 | page_cache_release(page); | |
5ec8b75e | 1224 | if (pos + len > inode->i_size) |
68eb3db0 | 1225 | ext3_truncate_failed_write(inode); |
f4fc66a8 | 1226 | } |
1da177e4 LT |
1227 | if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries)) |
1228 | goto retry; | |
1aa9b4b9 | 1229 | out: |
1da177e4 LT |
1230 | return ret; |
1231 | } | |
1232 | ||
f4fc66a8 | 1233 | |
d6859bfc | 1234 | int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh) |
1da177e4 LT |
1235 | { |
1236 | int err = journal_dirty_data(handle, bh); | |
1237 | if (err) | |
e05b6b52 | 1238 | ext3_journal_abort_handle(__func__, __func__, |
f4fc66a8 | 1239 | bh, handle, err); |
1da177e4 LT |
1240 | return err; |
1241 | } | |
1242 | ||
695f6ae0 JK |
1243 | /* For ordered writepage and write_end functions */ |
1244 | static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh) | |
1245 | { | |
1246 | /* | |
1247 | * Write could have mapped the buffer but it didn't copy the data in | |
1248 | * yet. So avoid filing such buffer into a transaction. | |
1249 | */ | |
1250 | if (buffer_mapped(bh) && buffer_uptodate(bh)) | |
1251 | return ext3_journal_dirty_data(handle, bh); | |
1252 | return 0; | |
1253 | } | |
1254 | ||
f4fc66a8 NP |
1255 | /* For write_end() in data=journal mode */ |
1256 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
1da177e4 LT |
1257 | { |
1258 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1259 | return 0; | |
1260 | set_buffer_uptodate(bh); | |
1261 | return ext3_journal_dirty_metadata(handle, bh); | |
1262 | } | |
1263 | ||
f4fc66a8 | 1264 | /* |
695f6ae0 JK |
1265 | * This is nasty and subtle: ext3_write_begin() could have allocated blocks |
1266 | * for the whole page but later we failed to copy the data in. Update inode | |
1267 | * size according to what we managed to copy. The rest is going to be | |
1268 | * truncated in write_end function. | |
f4fc66a8 | 1269 | */ |
695f6ae0 | 1270 | static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied) |
f4fc66a8 | 1271 | { |
695f6ae0 JK |
1272 | /* What matters to us is i_disksize. We don't write i_size anywhere */ |
1273 | if (pos + copied > inode->i_size) | |
1274 | i_size_write(inode, pos + copied); | |
1275 | if (pos + copied > EXT3_I(inode)->i_disksize) { | |
1276 | EXT3_I(inode)->i_disksize = pos + copied; | |
f4fc66a8 NP |
1277 | mark_inode_dirty(inode); |
1278 | } | |
f4fc66a8 NP |
1279 | } |
1280 | ||
1da177e4 LT |
1281 | /* |
1282 | * We need to pick up the new inode size which generic_commit_write gave us | |
1283 | * `file' can be NULL - eg, when called from page_symlink(). | |
1284 | * | |
1285 | * ext3 never places buffers on inode->i_mapping->private_list. metadata | |
1286 | * buffers are managed internally. | |
1287 | */ | |
f4fc66a8 NP |
1288 | static int ext3_ordered_write_end(struct file *file, |
1289 | struct address_space *mapping, | |
1290 | loff_t pos, unsigned len, unsigned copied, | |
1291 | struct page *page, void *fsdata) | |
1da177e4 LT |
1292 | { |
1293 | handle_t *handle = ext3_journal_current_handle(); | |
f4fc66a8 NP |
1294 | struct inode *inode = file->f_mapping->host; |
1295 | unsigned from, to; | |
1da177e4 LT |
1296 | int ret = 0, ret2; |
1297 | ||
695f6ae0 | 1298 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
f4fc66a8 | 1299 | |
695f6ae0 JK |
1300 | from = pos & (PAGE_CACHE_SIZE - 1); |
1301 | to = from + copied; | |
1da177e4 | 1302 | ret = walk_page_buffers(handle, page_buffers(page), |
695f6ae0 | 1303 | from, to, NULL, journal_dirty_data_fn); |
1da177e4 | 1304 | |
695f6ae0 JK |
1305 | if (ret == 0) |
1306 | update_file_sizes(inode, pos, copied); | |
1307 | /* | |
1308 | * There may be allocated blocks outside of i_size because | |
1309 | * we failed to copy some data. Prepare for truncate. | |
1310 | */ | |
9eaaa2d5 | 1311 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 | 1312 | ext3_orphan_add(handle, inode); |
1da177e4 LT |
1313 | ret2 = ext3_journal_stop(handle); |
1314 | if (!ret) | |
1315 | ret = ret2; | |
f4fc66a8 NP |
1316 | unlock_page(page); |
1317 | page_cache_release(page); | |
1318 | ||
695f6ae0 | 1319 | if (pos + len > inode->i_size) |
68eb3db0 | 1320 | ext3_truncate_failed_write(inode); |
f4fc66a8 | 1321 | return ret ? ret : copied; |
1da177e4 LT |
1322 | } |
1323 | ||
f4fc66a8 NP |
1324 | static int ext3_writeback_write_end(struct file *file, |
1325 | struct address_space *mapping, | |
1326 | loff_t pos, unsigned len, unsigned copied, | |
1327 | struct page *page, void *fsdata) | |
1da177e4 LT |
1328 | { |
1329 | handle_t *handle = ext3_journal_current_handle(); | |
f4fc66a8 | 1330 | struct inode *inode = file->f_mapping->host; |
695f6ae0 | 1331 | int ret; |
1da177e4 | 1332 | |
695f6ae0 JK |
1333 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
1334 | update_file_sizes(inode, pos, copied); | |
1335 | /* | |
1336 | * There may be allocated blocks outside of i_size because | |
1337 | * we failed to copy some data. Prepare for truncate. | |
1338 | */ | |
9eaaa2d5 | 1339 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 JK |
1340 | ext3_orphan_add(handle, inode); |
1341 | ret = ext3_journal_stop(handle); | |
f4fc66a8 NP |
1342 | unlock_page(page); |
1343 | page_cache_release(page); | |
1344 | ||
695f6ae0 | 1345 | if (pos + len > inode->i_size) |
68eb3db0 | 1346 | ext3_truncate_failed_write(inode); |
f4fc66a8 | 1347 | return ret ? ret : copied; |
1da177e4 LT |
1348 | } |
1349 | ||
f4fc66a8 NP |
1350 | static int ext3_journalled_write_end(struct file *file, |
1351 | struct address_space *mapping, | |
1352 | loff_t pos, unsigned len, unsigned copied, | |
1353 | struct page *page, void *fsdata) | |
1da177e4 LT |
1354 | { |
1355 | handle_t *handle = ext3_journal_current_handle(); | |
f4fc66a8 | 1356 | struct inode *inode = mapping->host; |
1da177e4 LT |
1357 | int ret = 0, ret2; |
1358 | int partial = 0; | |
f4fc66a8 | 1359 | unsigned from, to; |
1da177e4 | 1360 | |
f4fc66a8 NP |
1361 | from = pos & (PAGE_CACHE_SIZE - 1); |
1362 | to = from + len; | |
1363 | ||
1364 | if (copied < len) { | |
1365 | if (!PageUptodate(page)) | |
1366 | copied = 0; | |
695f6ae0 JK |
1367 | page_zero_new_buffers(page, from + copied, to); |
1368 | to = from + copied; | |
f4fc66a8 | 1369 | } |
1da177e4 LT |
1370 | |
1371 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
f4fc66a8 | 1372 | to, &partial, write_end_fn); |
1da177e4 LT |
1373 | if (!partial) |
1374 | SetPageUptodate(page); | |
695f6ae0 JK |
1375 | |
1376 | if (pos + copied > inode->i_size) | |
1377 | i_size_write(inode, pos + copied); | |
1378 | /* | |
1379 | * There may be allocated blocks outside of i_size because | |
1380 | * we failed to copy some data. Prepare for truncate. | |
1381 | */ | |
9eaaa2d5 | 1382 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
695f6ae0 | 1383 | ext3_orphan_add(handle, inode); |
9df93939 | 1384 | ext3_set_inode_state(inode, EXT3_STATE_JDATA); |
1da177e4 LT |
1385 | if (inode->i_size > EXT3_I(inode)->i_disksize) { |
1386 | EXT3_I(inode)->i_disksize = inode->i_size; | |
1387 | ret2 = ext3_mark_inode_dirty(handle, inode); | |
ae6ddcc5 | 1388 | if (!ret) |
1da177e4 LT |
1389 | ret = ret2; |
1390 | } | |
f4fc66a8 | 1391 | |
1da177e4 LT |
1392 | ret2 = ext3_journal_stop(handle); |
1393 | if (!ret) | |
1394 | ret = ret2; | |
f4fc66a8 NP |
1395 | unlock_page(page); |
1396 | page_cache_release(page); | |
1397 | ||
695f6ae0 | 1398 | if (pos + len > inode->i_size) |
68eb3db0 | 1399 | ext3_truncate_failed_write(inode); |
f4fc66a8 | 1400 | return ret ? ret : copied; |
1da177e4 LT |
1401 | } |
1402 | ||
ae6ddcc5 | 1403 | /* |
1da177e4 LT |
1404 | * bmap() is special. It gets used by applications such as lilo and by |
1405 | * the swapper to find the on-disk block of a specific piece of data. | |
1406 | * | |
1407 | * Naturally, this is dangerous if the block concerned is still in the | |
1408 | * journal. If somebody makes a swapfile on an ext3 data-journaling | |
1409 | * filesystem and enables swap, then they may get a nasty shock when the | |
1410 | * data getting swapped to that swapfile suddenly gets overwritten by | |
1411 | * the original zero's written out previously to the journal and | |
ae6ddcc5 | 1412 | * awaiting writeback in the kernel's buffer cache. |
1da177e4 LT |
1413 | * |
1414 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
ae6ddcc5 | 1415 | * take extra steps to flush any blocks which might be in the cache. |
1da177e4 LT |
1416 | */ |
1417 | static sector_t ext3_bmap(struct address_space *mapping, sector_t block) | |
1418 | { | |
1419 | struct inode *inode = mapping->host; | |
1420 | journal_t *journal; | |
1421 | int err; | |
1422 | ||
9df93939 | 1423 | if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) { |
ae6ddcc5 | 1424 | /* |
1da177e4 LT |
1425 | * This is a REALLY heavyweight approach, but the use of |
1426 | * bmap on dirty files is expected to be extremely rare: | |
1427 | * only if we run lilo or swapon on a freshly made file | |
ae6ddcc5 | 1428 | * do we expect this to happen. |
1da177e4 LT |
1429 | * |
1430 | * (bmap requires CAP_SYS_RAWIO so this does not | |
1431 | * represent an unprivileged user DOS attack --- we'd be | |
1432 | * in trouble if mortal users could trigger this path at | |
ae6ddcc5 | 1433 | * will.) |
1da177e4 LT |
1434 | * |
1435 | * NB. EXT3_STATE_JDATA is not set on files other than | |
1436 | * regular files. If somebody wants to bmap a directory | |
1437 | * or symlink and gets confused because the buffer | |
1438 | * hasn't yet been flushed to disk, they deserve | |
1439 | * everything they get. | |
1440 | */ | |
1441 | ||
9df93939 | 1442 | ext3_clear_inode_state(inode, EXT3_STATE_JDATA); |
1da177e4 LT |
1443 | journal = EXT3_JOURNAL(inode); |
1444 | journal_lock_updates(journal); | |
1445 | err = journal_flush(journal); | |
1446 | journal_unlock_updates(journal); | |
1447 | ||
1448 | if (err) | |
1449 | return 0; | |
1450 | } | |
1451 | ||
1452 | return generic_block_bmap(mapping,block,ext3_get_block); | |
1453 | } | |
1454 | ||
1455 | static int bget_one(handle_t *handle, struct buffer_head *bh) | |
1456 | { | |
1457 | get_bh(bh); | |
1458 | return 0; | |
1459 | } | |
1460 | ||
1461 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1462 | { | |
1463 | put_bh(bh); | |
1464 | return 0; | |
1465 | } | |
1466 | ||
9e80d407 JK |
1467 | static int buffer_unmapped(handle_t *handle, struct buffer_head *bh) |
1468 | { | |
1469 | return !buffer_mapped(bh); | |
1470 | } | |
695f6ae0 | 1471 | |
1da177e4 LT |
1472 | /* |
1473 | * Note that we always start a transaction even if we're not journalling | |
1474 | * data. This is to preserve ordering: any hole instantiation within | |
1475 | * __block_write_full_page -> ext3_get_block() should be journalled | |
1476 | * along with the data so we don't crash and then get metadata which | |
1477 | * refers to old data. | |
1478 | * | |
1479 | * In all journalling modes block_write_full_page() will start the I/O. | |
1480 | * | |
1481 | * Problem: | |
1482 | * | |
1483 | * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1484 | * ext3_writepage() | |
1485 | * | |
1486 | * Similar for: | |
1487 | * | |
1488 | * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ... | |
1489 | * | |
1490 | * Same applies to ext3_get_block(). We will deadlock on various things like | |
97461518 | 1491 | * lock_journal and i_truncate_mutex. |
1da177e4 LT |
1492 | * |
1493 | * Setting PF_MEMALLOC here doesn't work - too many internal memory | |
1494 | * allocations fail. | |
1495 | * | |
1496 | * 16May01: If we're reentered then journal_current_handle() will be | |
1497 | * non-zero. We simply *return*. | |
1498 | * | |
1499 | * 1 July 2001: @@@ FIXME: | |
1500 | * In journalled data mode, a data buffer may be metadata against the | |
1501 | * current transaction. But the same file is part of a shared mapping | |
1502 | * and someone does a writepage() on it. | |
1503 | * | |
1504 | * We will move the buffer onto the async_data list, but *after* it has | |
1505 | * been dirtied. So there's a small window where we have dirty data on | |
1506 | * BJ_Metadata. | |
1507 | * | |
1508 | * Note that this only applies to the last partial page in the file. The | |
1509 | * bit which block_write_full_page() uses prepare/commit for. (That's | |
1510 | * broken code anyway: it's wrong for msync()). | |
1511 | * | |
1512 | * It's a rare case: affects the final partial page, for journalled data | |
1513 | * where the file is subject to bith write() and writepage() in the same | |
1514 | * transction. To fix it we'll need a custom block_write_full_page(). | |
1515 | * We'll probably need that anyway for journalling writepage() output. | |
1516 | * | |
1517 | * We don't honour synchronous mounts for writepage(). That would be | |
1518 | * disastrous. Any write() or metadata operation will sync the fs for | |
1519 | * us. | |
1520 | * | |
1521 | * AKPM2: if all the page's buffers are mapped to disk and !data=journal, | |
1522 | * we don't need to open a transaction here. | |
1523 | */ | |
1524 | static int ext3_ordered_writepage(struct page *page, | |
d6859bfc | 1525 | struct writeback_control *wbc) |
1da177e4 LT |
1526 | { |
1527 | struct inode *inode = page->mapping->host; | |
1528 | struct buffer_head *page_bufs; | |
1529 | handle_t *handle = NULL; | |
1530 | int ret = 0; | |
1531 | int err; | |
1532 | ||
1533 | J_ASSERT(PageLocked(page)); | |
49792c80 | 1534 | WARN_ON_ONCE(IS_RDONLY(inode)); |
1da177e4 LT |
1535 | |
1536 | /* | |
1537 | * We give up here if we're reentered, because it might be for a | |
1538 | * different filesystem. | |
1539 | */ | |
1540 | if (ext3_journal_current_handle()) | |
1541 | goto out_fail; | |
1542 | ||
9e80d407 JK |
1543 | if (!page_has_buffers(page)) { |
1544 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
1545 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
430db323 JK |
1546 | page_bufs = page_buffers(page); |
1547 | } else { | |
1548 | page_bufs = page_buffers(page); | |
1549 | if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE, | |
1550 | NULL, buffer_unmapped)) { | |
1551 | /* Provide NULL get_block() to catch bugs if buffers | |
1552 | * weren't really mapped */ | |
1553 | return block_write_full_page(page, NULL, wbc); | |
1554 | } | |
9e80d407 | 1555 | } |
1da177e4 LT |
1556 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
1557 | ||
1558 | if (IS_ERR(handle)) { | |
1559 | ret = PTR_ERR(handle); | |
1560 | goto out_fail; | |
1561 | } | |
1562 | ||
1da177e4 LT |
1563 | walk_page_buffers(handle, page_bufs, 0, |
1564 | PAGE_CACHE_SIZE, NULL, bget_one); | |
1565 | ||
1566 | ret = block_write_full_page(page, ext3_get_block, wbc); | |
1567 | ||
1568 | /* | |
1569 | * The page can become unlocked at any point now, and | |
1570 | * truncate can then come in and change things. So we | |
1571 | * can't touch *page from now on. But *page_bufs is | |
1572 | * safe due to elevated refcount. | |
1573 | */ | |
1574 | ||
1575 | /* | |
ae6ddcc5 | 1576 | * And attach them to the current transaction. But only if |
1da177e4 LT |
1577 | * block_write_full_page() succeeded. Otherwise they are unmapped, |
1578 | * and generally junk. | |
1579 | */ | |
1580 | if (ret == 0) { | |
1581 | err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, | |
1582 | NULL, journal_dirty_data_fn); | |
1583 | if (!ret) | |
1584 | ret = err; | |
1585 | } | |
1586 | walk_page_buffers(handle, page_bufs, 0, | |
1587 | PAGE_CACHE_SIZE, NULL, bput_one); | |
1588 | err = ext3_journal_stop(handle); | |
1589 | if (!ret) | |
1590 | ret = err; | |
1591 | return ret; | |
1592 | ||
1593 | out_fail: | |
1594 | redirty_page_for_writepage(wbc, page); | |
1595 | unlock_page(page); | |
1596 | return ret; | |
1597 | } | |
1598 | ||
1da177e4 LT |
1599 | static int ext3_writeback_writepage(struct page *page, |
1600 | struct writeback_control *wbc) | |
1601 | { | |
1602 | struct inode *inode = page->mapping->host; | |
1603 | handle_t *handle = NULL; | |
1604 | int ret = 0; | |
1605 | int err; | |
1606 | ||
49792c80 DM |
1607 | J_ASSERT(PageLocked(page)); |
1608 | WARN_ON_ONCE(IS_RDONLY(inode)); | |
1609 | ||
1da177e4 LT |
1610 | if (ext3_journal_current_handle()) |
1611 | goto out_fail; | |
1612 | ||
430db323 JK |
1613 | if (page_has_buffers(page)) { |
1614 | if (!walk_page_buffers(NULL, page_buffers(page), 0, | |
1615 | PAGE_CACHE_SIZE, NULL, buffer_unmapped)) { | |
1616 | /* Provide NULL get_block() to catch bugs if buffers | |
1617 | * weren't really mapped */ | |
1618 | return block_write_full_page(page, NULL, wbc); | |
1619 | } | |
1620 | } | |
1621 | ||
1da177e4 LT |
1622 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
1623 | if (IS_ERR(handle)) { | |
1624 | ret = PTR_ERR(handle); | |
1625 | goto out_fail; | |
1626 | } | |
1627 | ||
0e31f51d | 1628 | if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode)) |
1da177e4 LT |
1629 | ret = nobh_writepage(page, ext3_get_block, wbc); |
1630 | else | |
1631 | ret = block_write_full_page(page, ext3_get_block, wbc); | |
1632 | ||
1633 | err = ext3_journal_stop(handle); | |
1634 | if (!ret) | |
1635 | ret = err; | |
1636 | return ret; | |
1637 | ||
1638 | out_fail: | |
1639 | redirty_page_for_writepage(wbc, page); | |
1640 | unlock_page(page); | |
1641 | return ret; | |
1642 | } | |
1643 | ||
1644 | static int ext3_journalled_writepage(struct page *page, | |
1645 | struct writeback_control *wbc) | |
1646 | { | |
1647 | struct inode *inode = page->mapping->host; | |
1648 | handle_t *handle = NULL; | |
1649 | int ret = 0; | |
1650 | int err; | |
1651 | ||
49792c80 DM |
1652 | J_ASSERT(PageLocked(page)); |
1653 | WARN_ON_ONCE(IS_RDONLY(inode)); | |
1654 | ||
1da177e4 LT |
1655 | if (ext3_journal_current_handle()) |
1656 | goto no_write; | |
1657 | ||
1658 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); | |
1659 | if (IS_ERR(handle)) { | |
1660 | ret = PTR_ERR(handle); | |
1661 | goto no_write; | |
1662 | } | |
1663 | ||
1664 | if (!page_has_buffers(page) || PageChecked(page)) { | |
1665 | /* | |
1666 | * It's mmapped pagecache. Add buffers and journal it. There | |
1667 | * doesn't seem much point in redirtying the page here. | |
1668 | */ | |
1669 | ClearPageChecked(page); | |
1670 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, | |
1671 | ext3_get_block); | |
ab4eb43c DL |
1672 | if (ret != 0) { |
1673 | ext3_journal_stop(handle); | |
1da177e4 | 1674 | goto out_unlock; |
ab4eb43c | 1675 | } |
1da177e4 LT |
1676 | ret = walk_page_buffers(handle, page_buffers(page), 0, |
1677 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); | |
1678 | ||
1679 | err = walk_page_buffers(handle, page_buffers(page), 0, | |
f4fc66a8 | 1680 | PAGE_CACHE_SIZE, NULL, write_end_fn); |
1da177e4 LT |
1681 | if (ret == 0) |
1682 | ret = err; | |
9df93939 | 1683 | ext3_set_inode_state(inode, EXT3_STATE_JDATA); |
1da177e4 LT |
1684 | unlock_page(page); |
1685 | } else { | |
1686 | /* | |
1687 | * It may be a page full of checkpoint-mode buffers. We don't | |
1688 | * really know unless we go poke around in the buffer_heads. | |
1689 | * But block_write_full_page will do the right thing. | |
1690 | */ | |
1691 | ret = block_write_full_page(page, ext3_get_block, wbc); | |
1692 | } | |
1693 | err = ext3_journal_stop(handle); | |
1694 | if (!ret) | |
1695 | ret = err; | |
1696 | out: | |
1697 | return ret; | |
1698 | ||
1699 | no_write: | |
1700 | redirty_page_for_writepage(wbc, page); | |
1701 | out_unlock: | |
1702 | unlock_page(page); | |
1703 | goto out; | |
1704 | } | |
1705 | ||
1706 | static int ext3_readpage(struct file *file, struct page *page) | |
1707 | { | |
1708 | return mpage_readpage(page, ext3_get_block); | |
1709 | } | |
1710 | ||
1711 | static int | |
1712 | ext3_readpages(struct file *file, struct address_space *mapping, | |
1713 | struct list_head *pages, unsigned nr_pages) | |
1714 | { | |
1715 | return mpage_readpages(mapping, pages, nr_pages, ext3_get_block); | |
1716 | } | |
1717 | ||
2ff28e22 | 1718 | static void ext3_invalidatepage(struct page *page, unsigned long offset) |
1da177e4 LT |
1719 | { |
1720 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); | |
1721 | ||
1722 | /* | |
1723 | * If it's a full truncate we just forget about the pending dirtying | |
1724 | */ | |
1725 | if (offset == 0) | |
1726 | ClearPageChecked(page); | |
1727 | ||
2ff28e22 | 1728 | journal_invalidatepage(journal, page, offset); |
1da177e4 LT |
1729 | } |
1730 | ||
27496a8c | 1731 | static int ext3_releasepage(struct page *page, gfp_t wait) |
1da177e4 LT |
1732 | { |
1733 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); | |
1734 | ||
1735 | WARN_ON(PageChecked(page)); | |
1736 | if (!page_has_buffers(page)) | |
1737 | return 0; | |
1738 | return journal_try_to_free_buffers(journal, page, wait); | |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * If the O_DIRECT write will extend the file then add this inode to the | |
1743 | * orphan list. So recovery will truncate it back to the original size | |
1744 | * if the machine crashes during the write. | |
1745 | * | |
1746 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
bd1939de JK |
1747 | * crashes then stale disk data _may_ be exposed inside the file. But current |
1748 | * VFS code falls back into buffered path in that case so we are safe. | |
1da177e4 LT |
1749 | */ |
1750 | static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb, | |
1751 | const struct iovec *iov, loff_t offset, | |
1752 | unsigned long nr_segs) | |
1753 | { | |
1754 | struct file *file = iocb->ki_filp; | |
1755 | struct inode *inode = file->f_mapping->host; | |
1756 | struct ext3_inode_info *ei = EXT3_I(inode); | |
bd1939de | 1757 | handle_t *handle; |
1da177e4 LT |
1758 | ssize_t ret; |
1759 | int orphan = 0; | |
1760 | size_t count = iov_length(iov, nr_segs); | |
ea0174a7 | 1761 | int retries = 0; |
1da177e4 LT |
1762 | |
1763 | if (rw == WRITE) { | |
1764 | loff_t final_size = offset + count; | |
1765 | ||
1da177e4 | 1766 | if (final_size > inode->i_size) { |
bd1939de JK |
1767 | /* Credits for sb + inode write */ |
1768 | handle = ext3_journal_start(inode, 2); | |
1769 | if (IS_ERR(handle)) { | |
1770 | ret = PTR_ERR(handle); | |
1771 | goto out; | |
1772 | } | |
1da177e4 | 1773 | ret = ext3_orphan_add(handle, inode); |
bd1939de JK |
1774 | if (ret) { |
1775 | ext3_journal_stop(handle); | |
1776 | goto out; | |
1777 | } | |
1da177e4 LT |
1778 | orphan = 1; |
1779 | ei->i_disksize = inode->i_size; | |
bd1939de | 1780 | ext3_journal_stop(handle); |
1da177e4 LT |
1781 | } |
1782 | } | |
1783 | ||
ea0174a7 | 1784 | retry: |
ae6ddcc5 | 1785 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, |
1da177e4 | 1786 | offset, nr_segs, |
f91a2ad2 | 1787 | ext3_get_block, NULL); |
ea0174a7 ES |
1788 | if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries)) |
1789 | goto retry; | |
1da177e4 | 1790 | |
bd1939de | 1791 | if (orphan) { |
1da177e4 LT |
1792 | int err; |
1793 | ||
bd1939de JK |
1794 | /* Credits for sb + inode write */ |
1795 | handle = ext3_journal_start(inode, 2); | |
1796 | if (IS_ERR(handle)) { | |
1797 | /* This is really bad luck. We've written the data | |
7eb4969e JK |
1798 | * but cannot extend i_size. Truncate allocated blocks |
1799 | * and pretend the write failed... */ | |
1800 | ext3_truncate(inode); | |
bd1939de JK |
1801 | ret = PTR_ERR(handle); |
1802 | goto out; | |
1803 | } | |
1804 | if (inode->i_nlink) | |
1da177e4 | 1805 | ext3_orphan_del(handle, inode); |
bd1939de | 1806 | if (ret > 0) { |
1da177e4 LT |
1807 | loff_t end = offset + ret; |
1808 | if (end > inode->i_size) { | |
1809 | ei->i_disksize = end; | |
1810 | i_size_write(inode, end); | |
1811 | /* | |
1812 | * We're going to return a positive `ret' | |
1813 | * here due to non-zero-length I/O, so there's | |
1814 | * no way of reporting error returns from | |
1815 | * ext3_mark_inode_dirty() to userspace. So | |
1816 | * ignore it. | |
1817 | */ | |
1818 | ext3_mark_inode_dirty(handle, inode); | |
1819 | } | |
1820 | } | |
1821 | err = ext3_journal_stop(handle); | |
1822 | if (ret == 0) | |
1823 | ret = err; | |
1824 | } | |
1825 | out: | |
1826 | return ret; | |
1827 | } | |
1828 | ||
1829 | /* | |
1830 | * Pages can be marked dirty completely asynchronously from ext3's journalling | |
1831 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do | |
1832 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
1833 | * not necessarily locked. | |
1834 | * | |
1835 | * We cannot just dirty the page and leave attached buffers clean, because the | |
1836 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
1837 | * or jbddirty because all the journalling code will explode. | |
1838 | * | |
1839 | * So what we do is to mark the page "pending dirty" and next time writepage | |
1840 | * is called, propagate that into the buffers appropriately. | |
1841 | */ | |
1842 | static int ext3_journalled_set_page_dirty(struct page *page) | |
1843 | { | |
1844 | SetPageChecked(page); | |
1845 | return __set_page_dirty_nobuffers(page); | |
1846 | } | |
1847 | ||
f5e54d6e | 1848 | static const struct address_space_operations ext3_ordered_aops = { |
8ab22b9a HH |
1849 | .readpage = ext3_readpage, |
1850 | .readpages = ext3_readpages, | |
1851 | .writepage = ext3_ordered_writepage, | |
1852 | .sync_page = block_sync_page, | |
1853 | .write_begin = ext3_write_begin, | |
1854 | .write_end = ext3_ordered_write_end, | |
1855 | .bmap = ext3_bmap, | |
1856 | .invalidatepage = ext3_invalidatepage, | |
1857 | .releasepage = ext3_releasepage, | |
1858 | .direct_IO = ext3_direct_IO, | |
1859 | .migratepage = buffer_migrate_page, | |
1860 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 1861 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
1862 | }; |
1863 | ||
f5e54d6e | 1864 | static const struct address_space_operations ext3_writeback_aops = { |
8ab22b9a HH |
1865 | .readpage = ext3_readpage, |
1866 | .readpages = ext3_readpages, | |
1867 | .writepage = ext3_writeback_writepage, | |
1868 | .sync_page = block_sync_page, | |
1869 | .write_begin = ext3_write_begin, | |
1870 | .write_end = ext3_writeback_write_end, | |
1871 | .bmap = ext3_bmap, | |
1872 | .invalidatepage = ext3_invalidatepage, | |
1873 | .releasepage = ext3_releasepage, | |
1874 | .direct_IO = ext3_direct_IO, | |
1875 | .migratepage = buffer_migrate_page, | |
1876 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 1877 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
1878 | }; |
1879 | ||
f5e54d6e | 1880 | static const struct address_space_operations ext3_journalled_aops = { |
8ab22b9a HH |
1881 | .readpage = ext3_readpage, |
1882 | .readpages = ext3_readpages, | |
1883 | .writepage = ext3_journalled_writepage, | |
1884 | .sync_page = block_sync_page, | |
1885 | .write_begin = ext3_write_begin, | |
1886 | .write_end = ext3_journalled_write_end, | |
1887 | .set_page_dirty = ext3_journalled_set_page_dirty, | |
1888 | .bmap = ext3_bmap, | |
1889 | .invalidatepage = ext3_invalidatepage, | |
1890 | .releasepage = ext3_releasepage, | |
1891 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 1892 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
1893 | }; |
1894 | ||
1895 | void ext3_set_aops(struct inode *inode) | |
1896 | { | |
1897 | if (ext3_should_order_data(inode)) | |
1898 | inode->i_mapping->a_ops = &ext3_ordered_aops; | |
1899 | else if (ext3_should_writeback_data(inode)) | |
1900 | inode->i_mapping->a_ops = &ext3_writeback_aops; | |
1901 | else | |
1902 | inode->i_mapping->a_ops = &ext3_journalled_aops; | |
1903 | } | |
1904 | ||
1905 | /* | |
1906 | * ext3_block_truncate_page() zeroes out a mapping from file offset `from' | |
1907 | * up to the end of the block which corresponds to `from'. | |
1908 | * This required during truncate. We need to physically zero the tail end | |
1909 | * of that block so it doesn't yield old data if the file is later grown. | |
1910 | */ | |
1911 | static int ext3_block_truncate_page(handle_t *handle, struct page *page, | |
1912 | struct address_space *mapping, loff_t from) | |
1913 | { | |
43d23f90 | 1914 | ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
1915 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
1916 | unsigned blocksize, iblock, length, pos; | |
1917 | struct inode *inode = mapping->host; | |
1918 | struct buffer_head *bh; | |
1919 | int err = 0; | |
1da177e4 LT |
1920 | |
1921 | blocksize = inode->i_sb->s_blocksize; | |
1922 | length = blocksize - (offset & (blocksize - 1)); | |
1923 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
1924 | ||
1925 | /* | |
1926 | * For "nobh" option, we can only work if we don't need to | |
1927 | * read-in the page - otherwise we create buffers to do the IO. | |
1928 | */ | |
cd6ef84e BP |
1929 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && |
1930 | ext3_should_writeback_data(inode) && PageUptodate(page)) { | |
eebd2aa3 | 1931 | zero_user(page, offset, length); |
cd6ef84e BP |
1932 | set_page_dirty(page); |
1933 | goto unlock; | |
1da177e4 LT |
1934 | } |
1935 | ||
1936 | if (!page_has_buffers(page)) | |
1937 | create_empty_buffers(page, blocksize, 0); | |
1938 | ||
1939 | /* Find the buffer that contains "offset" */ | |
1940 | bh = page_buffers(page); | |
1941 | pos = blocksize; | |
1942 | while (offset >= pos) { | |
1943 | bh = bh->b_this_page; | |
1944 | iblock++; | |
1945 | pos += blocksize; | |
1946 | } | |
1947 | ||
1948 | err = 0; | |
1949 | if (buffer_freed(bh)) { | |
1950 | BUFFER_TRACE(bh, "freed: skip"); | |
1951 | goto unlock; | |
1952 | } | |
1953 | ||
1954 | if (!buffer_mapped(bh)) { | |
1955 | BUFFER_TRACE(bh, "unmapped"); | |
1956 | ext3_get_block(inode, iblock, bh, 0); | |
1957 | /* unmapped? It's a hole - nothing to do */ | |
1958 | if (!buffer_mapped(bh)) { | |
1959 | BUFFER_TRACE(bh, "still unmapped"); | |
1960 | goto unlock; | |
1961 | } | |
1962 | } | |
1963 | ||
1964 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
1965 | if (PageUptodate(page)) | |
1966 | set_buffer_uptodate(bh); | |
1967 | ||
1968 | if (!buffer_uptodate(bh)) { | |
1969 | err = -EIO; | |
1970 | ll_rw_block(READ, 1, &bh); | |
1971 | wait_on_buffer(bh); | |
1972 | /* Uhhuh. Read error. Complain and punt. */ | |
1973 | if (!buffer_uptodate(bh)) | |
1974 | goto unlock; | |
1975 | } | |
1976 | ||
1977 | if (ext3_should_journal_data(inode)) { | |
1978 | BUFFER_TRACE(bh, "get write access"); | |
1979 | err = ext3_journal_get_write_access(handle, bh); | |
1980 | if (err) | |
1981 | goto unlock; | |
1982 | } | |
1983 | ||
eebd2aa3 | 1984 | zero_user(page, offset, length); |
1da177e4 LT |
1985 | BUFFER_TRACE(bh, "zeroed end of block"); |
1986 | ||
1987 | err = 0; | |
1988 | if (ext3_should_journal_data(inode)) { | |
1989 | err = ext3_journal_dirty_metadata(handle, bh); | |
1990 | } else { | |
1991 | if (ext3_should_order_data(inode)) | |
1992 | err = ext3_journal_dirty_data(handle, bh); | |
1993 | mark_buffer_dirty(bh); | |
1994 | } | |
1995 | ||
1996 | unlock: | |
1997 | unlock_page(page); | |
1998 | page_cache_release(page); | |
1999 | return err; | |
2000 | } | |
2001 | ||
2002 | /* | |
2003 | * Probably it should be a library function... search for first non-zero word | |
2004 | * or memcmp with zero_page, whatever is better for particular architecture. | |
2005 | * Linus? | |
2006 | */ | |
2007 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
2008 | { | |
2009 | while (p < q) | |
2010 | if (*p++) | |
2011 | return 0; | |
2012 | return 1; | |
2013 | } | |
2014 | ||
2015 | /** | |
2016 | * ext3_find_shared - find the indirect blocks for partial truncation. | |
2017 | * @inode: inode in question | |
2018 | * @depth: depth of the affected branch | |
2019 | * @offsets: offsets of pointers in that branch (see ext3_block_to_path) | |
2020 | * @chain: place to store the pointers to partial indirect blocks | |
2021 | * @top: place to the (detached) top of branch | |
2022 | * | |
2023 | * This is a helper function used by ext3_truncate(). | |
2024 | * | |
2025 | * When we do truncate() we may have to clean the ends of several | |
2026 | * indirect blocks but leave the blocks themselves alive. Block is | |
2027 | * partially truncated if some data below the new i_size is refered | |
2028 | * from it (and it is on the path to the first completely truncated | |
2029 | * data block, indeed). We have to free the top of that path along | |
2030 | * with everything to the right of the path. Since no allocation | |
2031 | * past the truncation point is possible until ext3_truncate() | |
2032 | * finishes, we may safely do the latter, but top of branch may | |
2033 | * require special attention - pageout below the truncation point | |
2034 | * might try to populate it. | |
2035 | * | |
2036 | * We atomically detach the top of branch from the tree, store the | |
2037 | * block number of its root in *@top, pointers to buffer_heads of | |
2038 | * partially truncated blocks - in @chain[].bh and pointers to | |
2039 | * their last elements that should not be removed - in | |
2040 | * @chain[].p. Return value is the pointer to last filled element | |
2041 | * of @chain. | |
2042 | * | |
2043 | * The work left to caller to do the actual freeing of subtrees: | |
2044 | * a) free the subtree starting from *@top | |
2045 | * b) free the subtrees whose roots are stored in | |
2046 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
2047 | * c) free the subtrees growing from the inode past the @chain[0]. | |
2048 | * (no partially truncated stuff there). */ | |
2049 | ||
d6859bfc AM |
2050 | static Indirect *ext3_find_shared(struct inode *inode, int depth, |
2051 | int offsets[4], Indirect chain[4], __le32 *top) | |
1da177e4 LT |
2052 | { |
2053 | Indirect *partial, *p; | |
2054 | int k, err; | |
2055 | ||
2056 | *top = 0; | |
bf48aabb | 2057 | /* Make k index the deepest non-null offset + 1 */ |
1da177e4 LT |
2058 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
2059 | ; | |
2060 | partial = ext3_get_branch(inode, k, offsets, chain, &err); | |
2061 | /* Writer: pointers */ | |
2062 | if (!partial) | |
2063 | partial = chain + k-1; | |
2064 | /* | |
2065 | * If the branch acquired continuation since we've looked at it - | |
2066 | * fine, it should all survive and (new) top doesn't belong to us. | |
2067 | */ | |
2068 | if (!partial->key && *partial->p) | |
2069 | /* Writer: end */ | |
2070 | goto no_top; | |
2071 | for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) | |
2072 | ; | |
2073 | /* | |
2074 | * OK, we've found the last block that must survive. The rest of our | |
2075 | * branch should be detached before unlocking. However, if that rest | |
2076 | * of branch is all ours and does not grow immediately from the inode | |
2077 | * it's easier to cheat and just decrement partial->p. | |
2078 | */ | |
2079 | if (p == chain + k - 1 && p > chain) { | |
2080 | p->p--; | |
2081 | } else { | |
2082 | *top = *p->p; | |
2083 | /* Nope, don't do this in ext3. Must leave the tree intact */ | |
2084 | #if 0 | |
2085 | *p->p = 0; | |
2086 | #endif | |
2087 | } | |
2088 | /* Writer: end */ | |
2089 | ||
d6859bfc | 2090 | while(partial > p) { |
1da177e4 LT |
2091 | brelse(partial->bh); |
2092 | partial--; | |
2093 | } | |
2094 | no_top: | |
2095 | return partial; | |
2096 | } | |
2097 | ||
2098 | /* | |
2099 | * Zero a number of block pointers in either an inode or an indirect block. | |
2100 | * If we restart the transaction we must again get write access to the | |
2101 | * indirect block for further modification. | |
2102 | * | |
2103 | * We release `count' blocks on disk, but (last - first) may be greater | |
2104 | * than `count' because there can be holes in there. | |
2105 | */ | |
d6859bfc | 2106 | static void ext3_clear_blocks(handle_t *handle, struct inode *inode, |
43d23f90 | 2107 | struct buffer_head *bh, ext3_fsblk_t block_to_free, |
d6859bfc | 2108 | unsigned long count, __le32 *first, __le32 *last) |
1da177e4 LT |
2109 | { |
2110 | __le32 *p; | |
2111 | if (try_to_extend_transaction(handle, inode)) { | |
2112 | if (bh) { | |
2113 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
2114 | ext3_journal_dirty_metadata(handle, bh); | |
2115 | } | |
2116 | ext3_mark_inode_dirty(handle, inode); | |
00171d3c | 2117 | truncate_restart_transaction(handle, inode); |
1da177e4 LT |
2118 | if (bh) { |
2119 | BUFFER_TRACE(bh, "retaking write access"); | |
2120 | ext3_journal_get_write_access(handle, bh); | |
2121 | } | |
2122 | } | |
2123 | ||
2124 | /* | |
2125 | * Any buffers which are on the journal will be in memory. We find | |
2126 | * them on the hash table so journal_revoke() will run journal_forget() | |
2127 | * on them. We've already detached each block from the file, so | |
2128 | * bforget() in journal_forget() should be safe. | |
2129 | * | |
2130 | * AKPM: turn on bforget in journal_forget()!!! | |
2131 | */ | |
2132 | for (p = first; p < last; p++) { | |
2133 | u32 nr = le32_to_cpu(*p); | |
2134 | if (nr) { | |
2135 | struct buffer_head *bh; | |
2136 | ||
2137 | *p = 0; | |
2138 | bh = sb_find_get_block(inode->i_sb, nr); | |
2139 | ext3_forget(handle, 0, inode, bh, nr); | |
2140 | } | |
2141 | } | |
2142 | ||
2143 | ext3_free_blocks(handle, inode, block_to_free, count); | |
2144 | } | |
2145 | ||
2146 | /** | |
2147 | * ext3_free_data - free a list of data blocks | |
2148 | * @handle: handle for this transaction | |
2149 | * @inode: inode we are dealing with | |
2150 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
2151 | * @first: array of block numbers | |
2152 | * @last: points immediately past the end of array | |
2153 | * | |
2154 | * We are freeing all blocks refered from that array (numbers are stored as | |
2155 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
2156 | * | |
2157 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
2158 | * blocks are contiguous then releasing them at one time will only affect one | |
2159 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
2160 | * actually use a lot of journal space. | |
2161 | * | |
2162 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
2163 | * block pointers. | |
2164 | */ | |
2165 | static void ext3_free_data(handle_t *handle, struct inode *inode, | |
2166 | struct buffer_head *this_bh, | |
2167 | __le32 *first, __le32 *last) | |
2168 | { | |
43d23f90 | 2169 | ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ae6ddcc5 | 2170 | unsigned long count = 0; /* Number of blocks in the run */ |
1da177e4 LT |
2171 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind |
2172 | corresponding to | |
2173 | block_to_free */ | |
43d23f90 | 2174 | ext3_fsblk_t nr; /* Current block # */ |
1da177e4 LT |
2175 | __le32 *p; /* Pointer into inode/ind |
2176 | for current block */ | |
2177 | int err; | |
2178 | ||
2179 | if (this_bh) { /* For indirect block */ | |
2180 | BUFFER_TRACE(this_bh, "get_write_access"); | |
2181 | err = ext3_journal_get_write_access(handle, this_bh); | |
2182 | /* Important: if we can't update the indirect pointers | |
2183 | * to the blocks, we can't free them. */ | |
2184 | if (err) | |
2185 | return; | |
2186 | } | |
2187 | ||
2188 | for (p = first; p < last; p++) { | |
2189 | nr = le32_to_cpu(*p); | |
2190 | if (nr) { | |
2191 | /* accumulate blocks to free if they're contiguous */ | |
2192 | if (count == 0) { | |
2193 | block_to_free = nr; | |
2194 | block_to_free_p = p; | |
2195 | count = 1; | |
2196 | } else if (nr == block_to_free + count) { | |
2197 | count++; | |
2198 | } else { | |
ae6ddcc5 | 2199 | ext3_clear_blocks(handle, inode, this_bh, |
1da177e4 LT |
2200 | block_to_free, |
2201 | count, block_to_free_p, p); | |
2202 | block_to_free = nr; | |
2203 | block_to_free_p = p; | |
2204 | count = 1; | |
2205 | } | |
2206 | } | |
2207 | } | |
2208 | ||
2209 | if (count > 0) | |
2210 | ext3_clear_blocks(handle, inode, this_bh, block_to_free, | |
2211 | count, block_to_free_p, p); | |
2212 | ||
2213 | if (this_bh) { | |
2214 | BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata"); | |
3ccc3167 DG |
2215 | |
2216 | /* | |
2217 | * The buffer head should have an attached journal head at this | |
2218 | * point. However, if the data is corrupted and an indirect | |
2219 | * block pointed to itself, it would have been detached when | |
2220 | * the block was cleared. Check for this instead of OOPSing. | |
2221 | */ | |
2222 | if (bh2jh(this_bh)) | |
2223 | ext3_journal_dirty_metadata(handle, this_bh); | |
2224 | else | |
2225 | ext3_error(inode->i_sb, "ext3_free_data", | |
2226 | "circular indirect block detected, " | |
2227 | "inode=%lu, block=%llu", | |
2228 | inode->i_ino, | |
2229 | (unsigned long long)this_bh->b_blocknr); | |
1da177e4 LT |
2230 | } |
2231 | } | |
2232 | ||
2233 | /** | |
2234 | * ext3_free_branches - free an array of branches | |
2235 | * @handle: JBD handle for this transaction | |
2236 | * @inode: inode we are dealing with | |
2237 | * @parent_bh: the buffer_head which contains *@first and *@last | |
2238 | * @first: array of block numbers | |
2239 | * @last: pointer immediately past the end of array | |
2240 | * @depth: depth of the branches to free | |
2241 | * | |
2242 | * We are freeing all blocks refered from these branches (numbers are | |
2243 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
2244 | * appropriately. | |
2245 | */ | |
2246 | static void ext3_free_branches(handle_t *handle, struct inode *inode, | |
2247 | struct buffer_head *parent_bh, | |
2248 | __le32 *first, __le32 *last, int depth) | |
2249 | { | |
43d23f90 | 2250 | ext3_fsblk_t nr; |
1da177e4 LT |
2251 | __le32 *p; |
2252 | ||
2253 | if (is_handle_aborted(handle)) | |
2254 | return; | |
2255 | ||
2256 | if (depth--) { | |
2257 | struct buffer_head *bh; | |
2258 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); | |
2259 | p = last; | |
2260 | while (--p >= first) { | |
2261 | nr = le32_to_cpu(*p); | |
2262 | if (!nr) | |
2263 | continue; /* A hole */ | |
2264 | ||
2265 | /* Go read the buffer for the next level down */ | |
2266 | bh = sb_bread(inode->i_sb, nr); | |
2267 | ||
2268 | /* | |
2269 | * A read failure? Report error and clear slot | |
2270 | * (should be rare). | |
2271 | */ | |
2272 | if (!bh) { | |
2273 | ext3_error(inode->i_sb, "ext3_free_branches", | |
eee194e7 | 2274 | "Read failure, inode=%lu, block="E3FSBLK, |
1da177e4 LT |
2275 | inode->i_ino, nr); |
2276 | continue; | |
2277 | } | |
2278 | ||
2279 | /* This zaps the entire block. Bottom up. */ | |
2280 | BUFFER_TRACE(bh, "free child branches"); | |
2281 | ext3_free_branches(handle, inode, bh, | |
2282 | (__le32*)bh->b_data, | |
2283 | (__le32*)bh->b_data + addr_per_block, | |
2284 | depth); | |
2285 | ||
2286 | /* | |
2287 | * We've probably journalled the indirect block several | |
2288 | * times during the truncate. But it's no longer | |
2289 | * needed and we now drop it from the transaction via | |
2290 | * journal_revoke(). | |
2291 | * | |
2292 | * That's easy if it's exclusively part of this | |
2293 | * transaction. But if it's part of the committing | |
2294 | * transaction then journal_forget() will simply | |
2295 | * brelse() it. That means that if the underlying | |
2296 | * block is reallocated in ext3_get_block(), | |
2297 | * unmap_underlying_metadata() will find this block | |
2298 | * and will try to get rid of it. damn, damn. | |
2299 | * | |
2300 | * If this block has already been committed to the | |
2301 | * journal, a revoke record will be written. And | |
2302 | * revoke records must be emitted *before* clearing | |
2303 | * this block's bit in the bitmaps. | |
2304 | */ | |
2305 | ext3_forget(handle, 1, inode, bh, bh->b_blocknr); | |
2306 | ||
2307 | /* | |
2308 | * Everything below this this pointer has been | |
2309 | * released. Now let this top-of-subtree go. | |
2310 | * | |
2311 | * We want the freeing of this indirect block to be | |
2312 | * atomic in the journal with the updating of the | |
2313 | * bitmap block which owns it. So make some room in | |
2314 | * the journal. | |
2315 | * | |
2316 | * We zero the parent pointer *after* freeing its | |
2317 | * pointee in the bitmaps, so if extend_transaction() | |
2318 | * for some reason fails to put the bitmap changes and | |
2319 | * the release into the same transaction, recovery | |
2320 | * will merely complain about releasing a free block, | |
2321 | * rather than leaking blocks. | |
2322 | */ | |
2323 | if (is_handle_aborted(handle)) | |
2324 | return; | |
2325 | if (try_to_extend_transaction(handle, inode)) { | |
2326 | ext3_mark_inode_dirty(handle, inode); | |
00171d3c | 2327 | truncate_restart_transaction(handle, inode); |
1da177e4 LT |
2328 | } |
2329 | ||
2330 | ext3_free_blocks(handle, inode, nr, 1); | |
2331 | ||
2332 | if (parent_bh) { | |
2333 | /* | |
2334 | * The block which we have just freed is | |
2335 | * pointed to by an indirect block: journal it | |
2336 | */ | |
2337 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
2338 | if (!ext3_journal_get_write_access(handle, | |
2339 | parent_bh)){ | |
2340 | *p = 0; | |
2341 | BUFFER_TRACE(parent_bh, | |
2342 | "call ext3_journal_dirty_metadata"); | |
ae6ddcc5 | 2343 | ext3_journal_dirty_metadata(handle, |
1da177e4 LT |
2344 | parent_bh); |
2345 | } | |
2346 | } | |
2347 | } | |
2348 | } else { | |
2349 | /* We have reached the bottom of the tree. */ | |
2350 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
2351 | ext3_free_data(handle, inode, parent_bh, first, last); | |
2352 | } | |
2353 | } | |
2354 | ||
ae76dd9a DG |
2355 | int ext3_can_truncate(struct inode *inode) |
2356 | { | |
2357 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
2358 | return 0; | |
2359 | if (S_ISREG(inode->i_mode)) | |
2360 | return 1; | |
2361 | if (S_ISDIR(inode->i_mode)) | |
2362 | return 1; | |
2363 | if (S_ISLNK(inode->i_mode)) | |
2364 | return !ext3_inode_is_fast_symlink(inode); | |
2365 | return 0; | |
2366 | } | |
2367 | ||
1da177e4 LT |
2368 | /* |
2369 | * ext3_truncate() | |
2370 | * | |
2371 | * We block out ext3_get_block() block instantiations across the entire | |
2372 | * transaction, and VFS/VM ensures that ext3_truncate() cannot run | |
2373 | * simultaneously on behalf of the same inode. | |
2374 | * | |
2375 | * As we work through the truncate and commmit bits of it to the journal there | |
2376 | * is one core, guiding principle: the file's tree must always be consistent on | |
2377 | * disk. We must be able to restart the truncate after a crash. | |
2378 | * | |
2379 | * The file's tree may be transiently inconsistent in memory (although it | |
2380 | * probably isn't), but whenever we close off and commit a journal transaction, | |
2381 | * the contents of (the filesystem + the journal) must be consistent and | |
2382 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
2383 | * left-to-right works OK too). | |
2384 | * | |
2385 | * Note that at recovery time, journal replay occurs *before* the restart of | |
2386 | * truncate against the orphan inode list. | |
2387 | * | |
2388 | * The committed inode has the new, desired i_size (which is the same as | |
2389 | * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see | |
2390 | * that this inode's truncate did not complete and it will again call | |
2391 | * ext3_truncate() to have another go. So there will be instantiated blocks | |
2392 | * to the right of the truncation point in a crashed ext3 filesystem. But | |
2393 | * that's fine - as long as they are linked from the inode, the post-crash | |
2394 | * ext3_truncate() run will find them and release them. | |
2395 | */ | |
d6859bfc | 2396 | void ext3_truncate(struct inode *inode) |
1da177e4 LT |
2397 | { |
2398 | handle_t *handle; | |
2399 | struct ext3_inode_info *ei = EXT3_I(inode); | |
2400 | __le32 *i_data = ei->i_data; | |
2401 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); | |
2402 | struct address_space *mapping = inode->i_mapping; | |
2403 | int offsets[4]; | |
2404 | Indirect chain[4]; | |
2405 | Indirect *partial; | |
2406 | __le32 nr = 0; | |
2407 | int n; | |
2408 | long last_block; | |
2409 | unsigned blocksize = inode->i_sb->s_blocksize; | |
2410 | struct page *page; | |
2411 | ||
ae76dd9a | 2412 | if (!ext3_can_truncate(inode)) |
ef43618a | 2413 | goto out_notrans; |
1da177e4 | 2414 | |
f7ab34ea | 2415 | if (inode->i_size == 0 && ext3_should_writeback_data(inode)) |
9df93939 | 2416 | ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE); |
f7ab34ea | 2417 | |
1da177e4 LT |
2418 | /* |
2419 | * We have to lock the EOF page here, because lock_page() nests | |
2420 | * outside journal_start(). | |
2421 | */ | |
2422 | if ((inode->i_size & (blocksize - 1)) == 0) { | |
2423 | /* Block boundary? Nothing to do */ | |
2424 | page = NULL; | |
2425 | } else { | |
2426 | page = grab_cache_page(mapping, | |
2427 | inode->i_size >> PAGE_CACHE_SHIFT); | |
2428 | if (!page) | |
ef43618a | 2429 | goto out_notrans; |
1da177e4 LT |
2430 | } |
2431 | ||
2432 | handle = start_transaction(inode); | |
2433 | if (IS_ERR(handle)) { | |
2434 | if (page) { | |
2435 | clear_highpage(page); | |
2436 | flush_dcache_page(page); | |
2437 | unlock_page(page); | |
2438 | page_cache_release(page); | |
2439 | } | |
ef43618a | 2440 | goto out_notrans; |
1da177e4 LT |
2441 | } |
2442 | ||
2443 | last_block = (inode->i_size + blocksize-1) | |
2444 | >> EXT3_BLOCK_SIZE_BITS(inode->i_sb); | |
2445 | ||
2446 | if (page) | |
2447 | ext3_block_truncate_page(handle, page, mapping, inode->i_size); | |
2448 | ||
2449 | n = ext3_block_to_path(inode, last_block, offsets, NULL); | |
2450 | if (n == 0) | |
2451 | goto out_stop; /* error */ | |
2452 | ||
2453 | /* | |
2454 | * OK. This truncate is going to happen. We add the inode to the | |
2455 | * orphan list, so that if this truncate spans multiple transactions, | |
2456 | * and we crash, we will resume the truncate when the filesystem | |
2457 | * recovers. It also marks the inode dirty, to catch the new size. | |
2458 | * | |
2459 | * Implication: the file must always be in a sane, consistent | |
2460 | * truncatable state while each transaction commits. | |
2461 | */ | |
2462 | if (ext3_orphan_add(handle, inode)) | |
2463 | goto out_stop; | |
2464 | ||
2465 | /* | |
2466 | * The orphan list entry will now protect us from any crash which | |
2467 | * occurs before the truncate completes, so it is now safe to propagate | |
2468 | * the new, shorter inode size (held for now in i_size) into the | |
2469 | * on-disk inode. We do this via i_disksize, which is the value which | |
2470 | * ext3 *really* writes onto the disk inode. | |
2471 | */ | |
2472 | ei->i_disksize = inode->i_size; | |
2473 | ||
2474 | /* | |
2475 | * From here we block out all ext3_get_block() callers who want to | |
2476 | * modify the block allocation tree. | |
2477 | */ | |
97461518 | 2478 | mutex_lock(&ei->truncate_mutex); |
1da177e4 LT |
2479 | |
2480 | if (n == 1) { /* direct blocks */ | |
2481 | ext3_free_data(handle, inode, NULL, i_data+offsets[0], | |
2482 | i_data + EXT3_NDIR_BLOCKS); | |
2483 | goto do_indirects; | |
2484 | } | |
2485 | ||
2486 | partial = ext3_find_shared(inode, n, offsets, chain, &nr); | |
2487 | /* Kill the top of shared branch (not detached) */ | |
2488 | if (nr) { | |
2489 | if (partial == chain) { | |
2490 | /* Shared branch grows from the inode */ | |
2491 | ext3_free_branches(handle, inode, NULL, | |
2492 | &nr, &nr+1, (chain+n-1) - partial); | |
2493 | *partial->p = 0; | |
2494 | /* | |
2495 | * We mark the inode dirty prior to restart, | |
2496 | * and prior to stop. No need for it here. | |
2497 | */ | |
2498 | } else { | |
2499 | /* Shared branch grows from an indirect block */ | |
2500 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
2501 | ext3_free_branches(handle, inode, partial->bh, | |
2502 | partial->p, | |
2503 | partial->p+1, (chain+n-1) - partial); | |
2504 | } | |
2505 | } | |
2506 | /* Clear the ends of indirect blocks on the shared branch */ | |
2507 | while (partial > chain) { | |
2508 | ext3_free_branches(handle, inode, partial->bh, partial->p + 1, | |
2509 | (__le32*)partial->bh->b_data+addr_per_block, | |
2510 | (chain+n-1) - partial); | |
2511 | BUFFER_TRACE(partial->bh, "call brelse"); | |
2512 | brelse (partial->bh); | |
2513 | partial--; | |
2514 | } | |
2515 | do_indirects: | |
2516 | /* Kill the remaining (whole) subtrees */ | |
2517 | switch (offsets[0]) { | |
d6859bfc AM |
2518 | default: |
2519 | nr = i_data[EXT3_IND_BLOCK]; | |
2520 | if (nr) { | |
2521 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1); | |
2522 | i_data[EXT3_IND_BLOCK] = 0; | |
2523 | } | |
2524 | case EXT3_IND_BLOCK: | |
2525 | nr = i_data[EXT3_DIND_BLOCK]; | |
2526 | if (nr) { | |
2527 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2); | |
2528 | i_data[EXT3_DIND_BLOCK] = 0; | |
2529 | } | |
2530 | case EXT3_DIND_BLOCK: | |
2531 | nr = i_data[EXT3_TIND_BLOCK]; | |
2532 | if (nr) { | |
2533 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3); | |
2534 | i_data[EXT3_TIND_BLOCK] = 0; | |
2535 | } | |
2536 | case EXT3_TIND_BLOCK: | |
2537 | ; | |
1da177e4 LT |
2538 | } |
2539 | ||
2540 | ext3_discard_reservation(inode); | |
2541 | ||
97461518 | 2542 | mutex_unlock(&ei->truncate_mutex); |
1da177e4 LT |
2543 | inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC; |
2544 | ext3_mark_inode_dirty(handle, inode); | |
2545 | ||
d6859bfc AM |
2546 | /* |
2547 | * In a multi-transaction truncate, we only make the final transaction | |
2548 | * synchronous | |
2549 | */ | |
1da177e4 LT |
2550 | if (IS_SYNC(inode)) |
2551 | handle->h_sync = 1; | |
2552 | out_stop: | |
2553 | /* | |
2554 | * If this was a simple ftruncate(), and the file will remain alive | |
2555 | * then we need to clear up the orphan record which we created above. | |
2556 | * However, if this was a real unlink then we were called by | |
2557 | * ext3_delete_inode(), and we allow that function to clean up the | |
2558 | * orphan info for us. | |
2559 | */ | |
2560 | if (inode->i_nlink) | |
2561 | ext3_orphan_del(handle, inode); | |
2562 | ||
2563 | ext3_journal_stop(handle); | |
ef43618a JK |
2564 | return; |
2565 | out_notrans: | |
2566 | /* | |
2567 | * Delete the inode from orphan list so that it doesn't stay there | |
2568 | * forever and trigger assertion on umount. | |
2569 | */ | |
2570 | if (inode->i_nlink) | |
2571 | ext3_orphan_del(NULL, inode); | |
1da177e4 LT |
2572 | } |
2573 | ||
43d23f90 | 2574 | static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb, |
1da177e4 LT |
2575 | unsigned long ino, struct ext3_iloc *iloc) |
2576 | { | |
e0e369a7 | 2577 | unsigned long block_group; |
43d23f90 MC |
2578 | unsigned long offset; |
2579 | ext3_fsblk_t block; | |
e0e369a7 | 2580 | struct ext3_group_desc *gdp; |
1da177e4 | 2581 | |
2ccb48eb NB |
2582 | if (!ext3_valid_inum(sb, ino)) { |
2583 | /* | |
2584 | * This error is already checked for in namei.c unless we are | |
2585 | * looking at an NFS filehandle, in which case no error | |
2586 | * report is needed | |
2587 | */ | |
1da177e4 LT |
2588 | return 0; |
2589 | } | |
2ccb48eb | 2590 | |
1da177e4 | 2591 | block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); |
e0e369a7 AM |
2592 | gdp = ext3_get_group_desc(sb, block_group, NULL); |
2593 | if (!gdp) | |
1da177e4 | 2594 | return 0; |
1da177e4 LT |
2595 | /* |
2596 | * Figure out the offset within the block group inode table | |
2597 | */ | |
2598 | offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) * | |
2599 | EXT3_INODE_SIZE(sb); | |
e0e369a7 | 2600 | block = le32_to_cpu(gdp->bg_inode_table) + |
1da177e4 LT |
2601 | (offset >> EXT3_BLOCK_SIZE_BITS(sb)); |
2602 | ||
2603 | iloc->block_group = block_group; | |
2604 | iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1); | |
2605 | return block; | |
2606 | } | |
2607 | ||
2608 | /* | |
2609 | * ext3_get_inode_loc returns with an extra refcount against the inode's | |
2610 | * underlying buffer_head on success. If 'in_mem' is true, we have all | |
2611 | * data in memory that is needed to recreate the on-disk version of this | |
2612 | * inode. | |
2613 | */ | |
2614 | static int __ext3_get_inode_loc(struct inode *inode, | |
2615 | struct ext3_iloc *iloc, int in_mem) | |
2616 | { | |
43d23f90 | 2617 | ext3_fsblk_t block; |
1da177e4 LT |
2618 | struct buffer_head *bh; |
2619 | ||
2620 | block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc); | |
2621 | if (!block) | |
2622 | return -EIO; | |
2623 | ||
2624 | bh = sb_getblk(inode->i_sb, block); | |
2625 | if (!bh) { | |
2626 | ext3_error (inode->i_sb, "ext3_get_inode_loc", | |
2627 | "unable to read inode block - " | |
43d23f90 MC |
2628 | "inode=%lu, block="E3FSBLK, |
2629 | inode->i_ino, block); | |
1da177e4 LT |
2630 | return -EIO; |
2631 | } | |
2632 | if (!buffer_uptodate(bh)) { | |
2633 | lock_buffer(bh); | |
95450f5a HK |
2634 | |
2635 | /* | |
2636 | * If the buffer has the write error flag, we have failed | |
2637 | * to write out another inode in the same block. In this | |
2638 | * case, we don't have to read the block because we may | |
2639 | * read the old inode data successfully. | |
2640 | */ | |
2641 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
2642 | set_buffer_uptodate(bh); | |
2643 | ||
1da177e4 LT |
2644 | if (buffer_uptodate(bh)) { |
2645 | /* someone brought it uptodate while we waited */ | |
2646 | unlock_buffer(bh); | |
2647 | goto has_buffer; | |
2648 | } | |
2649 | ||
2650 | /* | |
2651 | * If we have all information of the inode in memory and this | |
2652 | * is the only valid inode in the block, we need not read the | |
2653 | * block. | |
2654 | */ | |
2655 | if (in_mem) { | |
2656 | struct buffer_head *bitmap_bh; | |
2657 | struct ext3_group_desc *desc; | |
2658 | int inodes_per_buffer; | |
2659 | int inode_offset, i; | |
2660 | int block_group; | |
2661 | int start; | |
2662 | ||
2663 | block_group = (inode->i_ino - 1) / | |
2664 | EXT3_INODES_PER_GROUP(inode->i_sb); | |
2665 | inodes_per_buffer = bh->b_size / | |
2666 | EXT3_INODE_SIZE(inode->i_sb); | |
2667 | inode_offset = ((inode->i_ino - 1) % | |
2668 | EXT3_INODES_PER_GROUP(inode->i_sb)); | |
2669 | start = inode_offset & ~(inodes_per_buffer - 1); | |
2670 | ||
2671 | /* Is the inode bitmap in cache? */ | |
2672 | desc = ext3_get_group_desc(inode->i_sb, | |
2673 | block_group, NULL); | |
2674 | if (!desc) | |
2675 | goto make_io; | |
2676 | ||
2677 | bitmap_bh = sb_getblk(inode->i_sb, | |
2678 | le32_to_cpu(desc->bg_inode_bitmap)); | |
2679 | if (!bitmap_bh) | |
2680 | goto make_io; | |
2681 | ||
2682 | /* | |
2683 | * If the inode bitmap isn't in cache then the | |
2684 | * optimisation may end up performing two reads instead | |
2685 | * of one, so skip it. | |
2686 | */ | |
2687 | if (!buffer_uptodate(bitmap_bh)) { | |
2688 | brelse(bitmap_bh); | |
2689 | goto make_io; | |
2690 | } | |
2691 | for (i = start; i < start + inodes_per_buffer; i++) { | |
2692 | if (i == inode_offset) | |
2693 | continue; | |
2694 | if (ext3_test_bit(i, bitmap_bh->b_data)) | |
2695 | break; | |
2696 | } | |
2697 | brelse(bitmap_bh); | |
2698 | if (i == start + inodes_per_buffer) { | |
2699 | /* all other inodes are free, so skip I/O */ | |
2700 | memset(bh->b_data, 0, bh->b_size); | |
2701 | set_buffer_uptodate(bh); | |
2702 | unlock_buffer(bh); | |
2703 | goto has_buffer; | |
2704 | } | |
2705 | } | |
2706 | ||
2707 | make_io: | |
2708 | /* | |
2709 | * There are other valid inodes in the buffer, this inode | |
2710 | * has in-inode xattrs, or we don't have this inode in memory. | |
2711 | * Read the block from disk. | |
2712 | */ | |
2713 | get_bh(bh); | |
2714 | bh->b_end_io = end_buffer_read_sync; | |
caa38fb0 | 2715 | submit_bh(READ_META, bh); |
1da177e4 LT |
2716 | wait_on_buffer(bh); |
2717 | if (!buffer_uptodate(bh)) { | |
2718 | ext3_error(inode->i_sb, "ext3_get_inode_loc", | |
2719 | "unable to read inode block - " | |
43d23f90 | 2720 | "inode=%lu, block="E3FSBLK, |
1da177e4 LT |
2721 | inode->i_ino, block); |
2722 | brelse(bh); | |
2723 | return -EIO; | |
2724 | } | |
2725 | } | |
2726 | has_buffer: | |
2727 | iloc->bh = bh; | |
2728 | return 0; | |
2729 | } | |
2730 | ||
2731 | int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc) | |
2732 | { | |
2733 | /* We have all inode data except xattrs in memory here. */ | |
2734 | return __ext3_get_inode_loc(inode, iloc, | |
9df93939 | 2735 | !ext3_test_inode_state(inode, EXT3_STATE_XATTR)); |
1da177e4 LT |
2736 | } |
2737 | ||
2738 | void ext3_set_inode_flags(struct inode *inode) | |
2739 | { | |
2740 | unsigned int flags = EXT3_I(inode)->i_flags; | |
2741 | ||
2742 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
2743 | if (flags & EXT3_SYNC_FL) | |
2744 | inode->i_flags |= S_SYNC; | |
2745 | if (flags & EXT3_APPEND_FL) | |
2746 | inode->i_flags |= S_APPEND; | |
2747 | if (flags & EXT3_IMMUTABLE_FL) | |
2748 | inode->i_flags |= S_IMMUTABLE; | |
2749 | if (flags & EXT3_NOATIME_FL) | |
2750 | inode->i_flags |= S_NOATIME; | |
2751 | if (flags & EXT3_DIRSYNC_FL) | |
2752 | inode->i_flags |= S_DIRSYNC; | |
2753 | } | |
2754 | ||
28be5abb JK |
2755 | /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */ |
2756 | void ext3_get_inode_flags(struct ext3_inode_info *ei) | |
2757 | { | |
2758 | unsigned int flags = ei->vfs_inode.i_flags; | |
2759 | ||
2760 | ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL| | |
2761 | EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL); | |
2762 | if (flags & S_SYNC) | |
2763 | ei->i_flags |= EXT3_SYNC_FL; | |
2764 | if (flags & S_APPEND) | |
2765 | ei->i_flags |= EXT3_APPEND_FL; | |
2766 | if (flags & S_IMMUTABLE) | |
2767 | ei->i_flags |= EXT3_IMMUTABLE_FL; | |
2768 | if (flags & S_NOATIME) | |
2769 | ei->i_flags |= EXT3_NOATIME_FL; | |
2770 | if (flags & S_DIRSYNC) | |
2771 | ei->i_flags |= EXT3_DIRSYNC_FL; | |
2772 | } | |
2773 | ||
473043dc | 2774 | struct inode *ext3_iget(struct super_block *sb, unsigned long ino) |
1da177e4 LT |
2775 | { |
2776 | struct ext3_iloc iloc; | |
2777 | struct ext3_inode *raw_inode; | |
473043dc | 2778 | struct ext3_inode_info *ei; |
1da177e4 | 2779 | struct buffer_head *bh; |
473043dc | 2780 | struct inode *inode; |
fe8bc91c JK |
2781 | journal_t *journal = EXT3_SB(sb)->s_journal; |
2782 | transaction_t *transaction; | |
473043dc | 2783 | long ret; |
1da177e4 LT |
2784 | int block; |
2785 | ||
473043dc DH |
2786 | inode = iget_locked(sb, ino); |
2787 | if (!inode) | |
2788 | return ERR_PTR(-ENOMEM); | |
2789 | if (!(inode->i_state & I_NEW)) | |
2790 | return inode; | |
2791 | ||
2792 | ei = EXT3_I(inode); | |
1da177e4 LT |
2793 | ei->i_block_alloc_info = NULL; |
2794 | ||
473043dc DH |
2795 | ret = __ext3_get_inode_loc(inode, &iloc, 0); |
2796 | if (ret < 0) | |
1da177e4 LT |
2797 | goto bad_inode; |
2798 | bh = iloc.bh; | |
2799 | raw_inode = ext3_raw_inode(&iloc); | |
2800 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); | |
2801 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
2802 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
2803 | if(!(test_opt (inode->i_sb, NO_UID32))) { | |
2804 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; | |
2805 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
2806 | } | |
2807 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
2808 | inode->i_size = le32_to_cpu(raw_inode->i_size); | |
4d7bf11d MR |
2809 | inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime); |
2810 | inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime); | |
2811 | inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime); | |
1da177e4 LT |
2812 | inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0; |
2813 | ||
de329820 | 2814 | ei->i_state_flags = 0; |
1da177e4 LT |
2815 | ei->i_dir_start_lookup = 0; |
2816 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
2817 | /* We now have enough fields to check if the inode was active or not. | |
2818 | * This is needed because nfsd might try to access dead inodes | |
2819 | * the test is that same one that e2fsck uses | |
2820 | * NeilBrown 1999oct15 | |
2821 | */ | |
2822 | if (inode->i_nlink == 0) { | |
2823 | if (inode->i_mode == 0 || | |
2824 | !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) { | |
2825 | /* this inode is deleted */ | |
2826 | brelse (bh); | |
473043dc | 2827 | ret = -ESTALE; |
1da177e4 LT |
2828 | goto bad_inode; |
2829 | } | |
2830 | /* The only unlinked inodes we let through here have | |
2831 | * valid i_mode and are being read by the orphan | |
2832 | * recovery code: that's fine, we're about to complete | |
2833 | * the process of deleting those. */ | |
2834 | } | |
1da177e4 LT |
2835 | inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); |
2836 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); | |
2837 | #ifdef EXT3_FRAGMENTS | |
2838 | ei->i_faddr = le32_to_cpu(raw_inode->i_faddr); | |
2839 | ei->i_frag_no = raw_inode->i_frag; | |
2840 | ei->i_frag_size = raw_inode->i_fsize; | |
2841 | #endif | |
2842 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); | |
2843 | if (!S_ISREG(inode->i_mode)) { | |
2844 | ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); | |
2845 | } else { | |
2846 | inode->i_size |= | |
2847 | ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; | |
2848 | } | |
2849 | ei->i_disksize = inode->i_size; | |
2850 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); | |
2851 | ei->i_block_group = iloc.block_group; | |
2852 | /* | |
2853 | * NOTE! The in-memory inode i_data array is in little-endian order | |
2854 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
2855 | */ | |
2856 | for (block = 0; block < EXT3_N_BLOCKS; block++) | |
2857 | ei->i_data[block] = raw_inode->i_block[block]; | |
2858 | INIT_LIST_HEAD(&ei->i_orphan); | |
2859 | ||
fe8bc91c JK |
2860 | /* |
2861 | * Set transaction id's of transactions that have to be committed | |
2862 | * to finish f[data]sync. We set them to currently running transaction | |
2863 | * as we cannot be sure that the inode or some of its metadata isn't | |
2864 | * part of the transaction - the inode could have been reclaimed and | |
2865 | * now it is reread from disk. | |
2866 | */ | |
2867 | if (journal) { | |
2868 | tid_t tid; | |
2869 | ||
2870 | spin_lock(&journal->j_state_lock); | |
2871 | if (journal->j_running_transaction) | |
2872 | transaction = journal->j_running_transaction; | |
2873 | else | |
2874 | transaction = journal->j_committing_transaction; | |
2875 | if (transaction) | |
2876 | tid = transaction->t_tid; | |
2877 | else | |
2878 | tid = journal->j_commit_sequence; | |
2879 | spin_unlock(&journal->j_state_lock); | |
2880 | atomic_set(&ei->i_sync_tid, tid); | |
2881 | atomic_set(&ei->i_datasync_tid, tid); | |
2882 | } | |
2883 | ||
1da177e4 LT |
2884 | if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 && |
2885 | EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) { | |
2886 | /* | |
2887 | * When mke2fs creates big inodes it does not zero out | |
2888 | * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE, | |
2889 | * so ignore those first few inodes. | |
2890 | */ | |
2891 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
2892 | if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
e4a10a36 KK |
2893 | EXT3_INODE_SIZE(inode->i_sb)) { |
2894 | brelse (bh); | |
473043dc | 2895 | ret = -EIO; |
1da177e4 | 2896 | goto bad_inode; |
e4a10a36 | 2897 | } |
1da177e4 LT |
2898 | if (ei->i_extra_isize == 0) { |
2899 | /* The extra space is currently unused. Use it. */ | |
2900 | ei->i_extra_isize = sizeof(struct ext3_inode) - | |
2901 | EXT3_GOOD_OLD_INODE_SIZE; | |
2902 | } else { | |
2903 | __le32 *magic = (void *)raw_inode + | |
2904 | EXT3_GOOD_OLD_INODE_SIZE + | |
2905 | ei->i_extra_isize; | |
2906 | if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC)) | |
9df93939 | 2907 | ext3_set_inode_state(inode, EXT3_STATE_XATTR); |
1da177e4 LT |
2908 | } |
2909 | } else | |
2910 | ei->i_extra_isize = 0; | |
2911 | ||
2912 | if (S_ISREG(inode->i_mode)) { | |
2913 | inode->i_op = &ext3_file_inode_operations; | |
2914 | inode->i_fop = &ext3_file_operations; | |
2915 | ext3_set_aops(inode); | |
2916 | } else if (S_ISDIR(inode->i_mode)) { | |
2917 | inode->i_op = &ext3_dir_inode_operations; | |
2918 | inode->i_fop = &ext3_dir_operations; | |
2919 | } else if (S_ISLNK(inode->i_mode)) { | |
b5ed3112 | 2920 | if (ext3_inode_is_fast_symlink(inode)) { |
1da177e4 | 2921 | inode->i_op = &ext3_fast_symlink_inode_operations; |
b5ed3112 DG |
2922 | nd_terminate_link(ei->i_data, inode->i_size, |
2923 | sizeof(ei->i_data) - 1); | |
2924 | } else { | |
1da177e4 LT |
2925 | inode->i_op = &ext3_symlink_inode_operations; |
2926 | ext3_set_aops(inode); | |
2927 | } | |
2928 | } else { | |
2929 | inode->i_op = &ext3_special_inode_operations; | |
2930 | if (raw_inode->i_block[0]) | |
2931 | init_special_inode(inode, inode->i_mode, | |
2932 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
ae6ddcc5 | 2933 | else |
1da177e4 LT |
2934 | init_special_inode(inode, inode->i_mode, |
2935 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
2936 | } | |
2937 | brelse (iloc.bh); | |
2938 | ext3_set_inode_flags(inode); | |
473043dc DH |
2939 | unlock_new_inode(inode); |
2940 | return inode; | |
1da177e4 LT |
2941 | |
2942 | bad_inode: | |
473043dc DH |
2943 | iget_failed(inode); |
2944 | return ERR_PTR(ret); | |
1da177e4 LT |
2945 | } |
2946 | ||
2947 | /* | |
2948 | * Post the struct inode info into an on-disk inode location in the | |
2949 | * buffer-cache. This gobbles the caller's reference to the | |
2950 | * buffer_head in the inode location struct. | |
2951 | * | |
2952 | * The caller must have write access to iloc->bh. | |
2953 | */ | |
ae6ddcc5 MC |
2954 | static int ext3_do_update_inode(handle_t *handle, |
2955 | struct inode *inode, | |
1da177e4 LT |
2956 | struct ext3_iloc *iloc) |
2957 | { | |
2958 | struct ext3_inode *raw_inode = ext3_raw_inode(iloc); | |
2959 | struct ext3_inode_info *ei = EXT3_I(inode); | |
2960 | struct buffer_head *bh = iloc->bh; | |
2961 | int err = 0, rc, block; | |
2962 | ||
4f003fd3 CM |
2963 | again: |
2964 | /* we can't allow multiple procs in here at once, its a bit racey */ | |
2965 | lock_buffer(bh); | |
2966 | ||
1da177e4 LT |
2967 | /* For fields not not tracking in the in-memory inode, |
2968 | * initialise them to zero for new inodes. */ | |
9df93939 | 2969 | if (ext3_test_inode_state(inode, EXT3_STATE_NEW)) |
1da177e4 LT |
2970 | memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size); |
2971 | ||
28be5abb | 2972 | ext3_get_inode_flags(ei); |
1da177e4 LT |
2973 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
2974 | if(!(test_opt(inode->i_sb, NO_UID32))) { | |
2975 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); | |
2976 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
2977 | /* | |
2978 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
2979 | * re-used with the upper 16 bits of the uid/gid intact | |
2980 | */ | |
2981 | if(!ei->i_dtime) { | |
2982 | raw_inode->i_uid_high = | |
2983 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
2984 | raw_inode->i_gid_high = | |
2985 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
2986 | } else { | |
2987 | raw_inode->i_uid_high = 0; | |
2988 | raw_inode->i_gid_high = 0; | |
2989 | } | |
2990 | } else { | |
2991 | raw_inode->i_uid_low = | |
2992 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
2993 | raw_inode->i_gid_low = | |
2994 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
2995 | raw_inode->i_uid_high = 0; | |
2996 | raw_inode->i_gid_high = 0; | |
2997 | } | |
2998 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
2999 | raw_inode->i_size = cpu_to_le32(ei->i_disksize); | |
3000 | raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec); | |
3001 | raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec); | |
3002 | raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec); | |
3003 | raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); | |
3004 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); | |
3005 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); | |
3006 | #ifdef EXT3_FRAGMENTS | |
3007 | raw_inode->i_faddr = cpu_to_le32(ei->i_faddr); | |
3008 | raw_inode->i_frag = ei->i_frag_no; | |
3009 | raw_inode->i_fsize = ei->i_frag_size; | |
3010 | #endif | |
3011 | raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); | |
3012 | if (!S_ISREG(inode->i_mode)) { | |
3013 | raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); | |
3014 | } else { | |
3015 | raw_inode->i_size_high = | |
3016 | cpu_to_le32(ei->i_disksize >> 32); | |
3017 | if (ei->i_disksize > 0x7fffffffULL) { | |
3018 | struct super_block *sb = inode->i_sb; | |
3019 | if (!EXT3_HAS_RO_COMPAT_FEATURE(sb, | |
3020 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE) || | |
3021 | EXT3_SB(sb)->s_es->s_rev_level == | |
3022 | cpu_to_le32(EXT3_GOOD_OLD_REV)) { | |
3023 | /* If this is the first large file | |
3024 | * created, add a flag to the superblock. | |
3025 | */ | |
4f003fd3 | 3026 | unlock_buffer(bh); |
1da177e4 LT |
3027 | err = ext3_journal_get_write_access(handle, |
3028 | EXT3_SB(sb)->s_sbh); | |
3029 | if (err) | |
3030 | goto out_brelse; | |
4f003fd3 | 3031 | |
1da177e4 LT |
3032 | ext3_update_dynamic_rev(sb); |
3033 | EXT3_SET_RO_COMPAT_FEATURE(sb, | |
3034 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE); | |
1da177e4 LT |
3035 | handle->h_sync = 1; |
3036 | err = ext3_journal_dirty_metadata(handle, | |
3037 | EXT3_SB(sb)->s_sbh); | |
4f003fd3 CM |
3038 | /* get our lock and start over */ |
3039 | goto again; | |
1da177e4 LT |
3040 | } |
3041 | } | |
3042 | } | |
3043 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
3044 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
3045 | if (old_valid_dev(inode->i_rdev)) { | |
3046 | raw_inode->i_block[0] = | |
3047 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
3048 | raw_inode->i_block[1] = 0; | |
3049 | } else { | |
3050 | raw_inode->i_block[0] = 0; | |
3051 | raw_inode->i_block[1] = | |
3052 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
3053 | raw_inode->i_block[2] = 0; | |
3054 | } | |
3055 | } else for (block = 0; block < EXT3_N_BLOCKS; block++) | |
3056 | raw_inode->i_block[block] = ei->i_data[block]; | |
3057 | ||
ff87b37d | 3058 | if (ei->i_extra_isize) |
1da177e4 LT |
3059 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
3060 | ||
3061 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
4f003fd3 | 3062 | unlock_buffer(bh); |
1da177e4 LT |
3063 | rc = ext3_journal_dirty_metadata(handle, bh); |
3064 | if (!err) | |
3065 | err = rc; | |
9df93939 | 3066 | ext3_clear_inode_state(inode, EXT3_STATE_NEW); |
1da177e4 | 3067 | |
fe8bc91c | 3068 | atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid); |
1da177e4 LT |
3069 | out_brelse: |
3070 | brelse (bh); | |
3071 | ext3_std_error(inode->i_sb, err); | |
3072 | return err; | |
3073 | } | |
3074 | ||
3075 | /* | |
3076 | * ext3_write_inode() | |
3077 | * | |
3078 | * We are called from a few places: | |
3079 | * | |
3080 | * - Within generic_file_write() for O_SYNC files. | |
3081 | * Here, there will be no transaction running. We wait for any running | |
3082 | * trasnaction to commit. | |
3083 | * | |
3084 | * - Within sys_sync(), kupdate and such. | |
3085 | * We wait on commit, if tol to. | |
3086 | * | |
3087 | * - Within prune_icache() (PF_MEMALLOC == true) | |
3088 | * Here we simply return. We can't afford to block kswapd on the | |
3089 | * journal commit. | |
3090 | * | |
3091 | * In all cases it is actually safe for us to return without doing anything, | |
3092 | * because the inode has been copied into a raw inode buffer in | |
3093 | * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for | |
3094 | * knfsd. | |
3095 | * | |
3096 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
3097 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
3098 | * which we are interested. | |
3099 | * | |
3100 | * It would be a bug for them to not do this. The code: | |
3101 | * | |
3102 | * mark_inode_dirty(inode) | |
3103 | * stuff(); | |
3104 | * inode->i_size = expr; | |
3105 | * | |
3106 | * is in error because a kswapd-driven write_inode() could occur while | |
3107 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
3108 | * will no longer be on the superblock's dirty inode list. | |
3109 | */ | |
a9185b41 | 3110 | int ext3_write_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 LT |
3111 | { |
3112 | if (current->flags & PF_MEMALLOC) | |
3113 | return 0; | |
3114 | ||
3115 | if (ext3_journal_current_handle()) { | |
9ad163ae | 3116 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
1da177e4 LT |
3117 | dump_stack(); |
3118 | return -EIO; | |
3119 | } | |
3120 | ||
a9185b41 | 3121 | if (wbc->sync_mode != WB_SYNC_ALL) |
1da177e4 LT |
3122 | return 0; |
3123 | ||
3124 | return ext3_force_commit(inode->i_sb); | |
3125 | } | |
3126 | ||
3127 | /* | |
3128 | * ext3_setattr() | |
3129 | * | |
3130 | * Called from notify_change. | |
3131 | * | |
3132 | * We want to trap VFS attempts to truncate the file as soon as | |
3133 | * possible. In particular, we want to make sure that when the VFS | |
3134 | * shrinks i_size, we put the inode on the orphan list and modify | |
3135 | * i_disksize immediately, so that during the subsequent flushing of | |
3136 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
3137 | * commit will leave the blocks being flushed in an unused state on | |
3138 | * disk. (On recovery, the inode will get truncated and the blocks will | |
3139 | * be freed, so we have a strong guarantee that no future commit will | |
ae6ddcc5 | 3140 | * leave these blocks visible to the user.) |
1da177e4 LT |
3141 | * |
3142 | * Called with inode->sem down. | |
3143 | */ | |
3144 | int ext3_setattr(struct dentry *dentry, struct iattr *attr) | |
3145 | { | |
3146 | struct inode *inode = dentry->d_inode; | |
3147 | int error, rc = 0; | |
3148 | const unsigned int ia_valid = attr->ia_valid; | |
3149 | ||
3150 | error = inode_change_ok(inode, attr); | |
3151 | if (error) | |
3152 | return error; | |
3153 | ||
12755627 | 3154 | if (is_quota_modification(inode, attr)) |
871a2931 | 3155 | dquot_initialize(inode); |
1da177e4 LT |
3156 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
3157 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
3158 | handle_t *handle; | |
3159 | ||
3160 | /* (user+group)*(old+new) structure, inode write (sb, | |
3161 | * inode block, ? - but truncate inode update has it) */ | |
c459001f DM |
3162 | handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
3163 | EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3); | |
1da177e4 LT |
3164 | if (IS_ERR(handle)) { |
3165 | error = PTR_ERR(handle); | |
3166 | goto err_out; | |
3167 | } | |
b43fa828 | 3168 | error = dquot_transfer(inode, attr); |
1da177e4 LT |
3169 | if (error) { |
3170 | ext3_journal_stop(handle); | |
3171 | return error; | |
3172 | } | |
3173 | /* Update corresponding info in inode so that everything is in | |
3174 | * one transaction */ | |
3175 | if (attr->ia_valid & ATTR_UID) | |
3176 | inode->i_uid = attr->ia_uid; | |
3177 | if (attr->ia_valid & ATTR_GID) | |
3178 | inode->i_gid = attr->ia_gid; | |
3179 | error = ext3_mark_inode_dirty(handle, inode); | |
3180 | ext3_journal_stop(handle); | |
3181 | } | |
3182 | ||
3183 | if (S_ISREG(inode->i_mode) && | |
3184 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { | |
3185 | handle_t *handle; | |
3186 | ||
3187 | handle = ext3_journal_start(inode, 3); | |
3188 | if (IS_ERR(handle)) { | |
3189 | error = PTR_ERR(handle); | |
3190 | goto err_out; | |
3191 | } | |
3192 | ||
3193 | error = ext3_orphan_add(handle, inode); | |
3194 | EXT3_I(inode)->i_disksize = attr->ia_size; | |
3195 | rc = ext3_mark_inode_dirty(handle, inode); | |
3196 | if (!error) | |
3197 | error = rc; | |
3198 | ext3_journal_stop(handle); | |
3199 | } | |
3200 | ||
3201 | rc = inode_setattr(inode, attr); | |
3202 | ||
1da177e4 LT |
3203 | if (!rc && (ia_valid & ATTR_MODE)) |
3204 | rc = ext3_acl_chmod(inode); | |
3205 | ||
3206 | err_out: | |
3207 | ext3_std_error(inode->i_sb, error); | |
3208 | if (!error) | |
3209 | error = rc; | |
3210 | return error; | |
3211 | } | |
3212 | ||
3213 | ||
3214 | /* | |
d6859bfc | 3215 | * How many blocks doth make a writepage()? |
1da177e4 LT |
3216 | * |
3217 | * With N blocks per page, it may be: | |
3218 | * N data blocks | |
3219 | * 2 indirect block | |
3220 | * 2 dindirect | |
3221 | * 1 tindirect | |
3222 | * N+5 bitmap blocks (from the above) | |
3223 | * N+5 group descriptor summary blocks | |
3224 | * 1 inode block | |
3225 | * 1 superblock. | |
3226 | * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files | |
3227 | * | |
3228 | * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS | |
3229 | * | |
3230 | * With ordered or writeback data it's the same, less the N data blocks. | |
3231 | * | |
3232 | * If the inode's direct blocks can hold an integral number of pages then a | |
3233 | * page cannot straddle two indirect blocks, and we can only touch one indirect | |
3234 | * and dindirect block, and the "5" above becomes "3". | |
3235 | * | |
3236 | * This still overestimates under most circumstances. If we were to pass the | |
3237 | * start and end offsets in here as well we could do block_to_path() on each | |
3238 | * block and work out the exact number of indirects which are touched. Pah. | |
3239 | */ | |
3240 | ||
3241 | static int ext3_writepage_trans_blocks(struct inode *inode) | |
3242 | { | |
3243 | int bpp = ext3_journal_blocks_per_page(inode); | |
3244 | int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3; | |
3245 | int ret; | |
3246 | ||
3247 | if (ext3_should_journal_data(inode)) | |
3248 | ret = 3 * (bpp + indirects) + 2; | |
3249 | else | |
3250 | ret = 2 * (bpp + indirects) + 2; | |
3251 | ||
3252 | #ifdef CONFIG_QUOTA | |
871a2931 | 3253 | /* We know that structure was already allocated during dquot_initialize so |
1da177e4 | 3254 | * we will be updating only the data blocks + inodes */ |
c459001f | 3255 | ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
1da177e4 LT |
3256 | #endif |
3257 | ||
3258 | return ret; | |
3259 | } | |
3260 | ||
3261 | /* | |
3262 | * The caller must have previously called ext3_reserve_inode_write(). | |
3263 | * Give this, we know that the caller already has write access to iloc->bh. | |
3264 | */ | |
3265 | int ext3_mark_iloc_dirty(handle_t *handle, | |
3266 | struct inode *inode, struct ext3_iloc *iloc) | |
3267 | { | |
3268 | int err = 0; | |
3269 | ||
3270 | /* the do_update_inode consumes one bh->b_count */ | |
3271 | get_bh(iloc->bh); | |
3272 | ||
3273 | /* ext3_do_update_inode() does journal_dirty_metadata */ | |
3274 | err = ext3_do_update_inode(handle, inode, iloc); | |
3275 | put_bh(iloc->bh); | |
3276 | return err; | |
3277 | } | |
3278 | ||
ae6ddcc5 | 3279 | /* |
1da177e4 | 3280 | * On success, We end up with an outstanding reference count against |
ae6ddcc5 | 3281 | * iloc->bh. This _must_ be cleaned up later. |
1da177e4 LT |
3282 | */ |
3283 | ||
3284 | int | |
ae6ddcc5 | 3285 | ext3_reserve_inode_write(handle_t *handle, struct inode *inode, |
1da177e4 LT |
3286 | struct ext3_iloc *iloc) |
3287 | { | |
3288 | int err = 0; | |
3289 | if (handle) { | |
3290 | err = ext3_get_inode_loc(inode, iloc); | |
3291 | if (!err) { | |
3292 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
3293 | err = ext3_journal_get_write_access(handle, iloc->bh); | |
3294 | if (err) { | |
3295 | brelse(iloc->bh); | |
3296 | iloc->bh = NULL; | |
3297 | } | |
3298 | } | |
3299 | } | |
3300 | ext3_std_error(inode->i_sb, err); | |
3301 | return err; | |
3302 | } | |
3303 | ||
3304 | /* | |
d6859bfc AM |
3305 | * What we do here is to mark the in-core inode as clean with respect to inode |
3306 | * dirtiness (it may still be data-dirty). | |
1da177e4 LT |
3307 | * This means that the in-core inode may be reaped by prune_icache |
3308 | * without having to perform any I/O. This is a very good thing, | |
3309 | * because *any* task may call prune_icache - even ones which | |
3310 | * have a transaction open against a different journal. | |
3311 | * | |
3312 | * Is this cheating? Not really. Sure, we haven't written the | |
3313 | * inode out, but prune_icache isn't a user-visible syncing function. | |
3314 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
3315 | * we start and wait on commits. | |
3316 | * | |
3317 | * Is this efficient/effective? Well, we're being nice to the system | |
3318 | * by cleaning up our inodes proactively so they can be reaped | |
3319 | * without I/O. But we are potentially leaving up to five seconds' | |
3320 | * worth of inodes floating about which prune_icache wants us to | |
3321 | * write out. One way to fix that would be to get prune_icache() | |
3322 | * to do a write_super() to free up some memory. It has the desired | |
3323 | * effect. | |
3324 | */ | |
3325 | int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode) | |
3326 | { | |
3327 | struct ext3_iloc iloc; | |
3328 | int err; | |
3329 | ||
3330 | might_sleep(); | |
3331 | err = ext3_reserve_inode_write(handle, inode, &iloc); | |
3332 | if (!err) | |
3333 | err = ext3_mark_iloc_dirty(handle, inode, &iloc); | |
3334 | return err; | |
3335 | } | |
3336 | ||
3337 | /* | |
d6859bfc | 3338 | * ext3_dirty_inode() is called from __mark_inode_dirty() |
1da177e4 LT |
3339 | * |
3340 | * We're really interested in the case where a file is being extended. | |
3341 | * i_size has been changed by generic_commit_write() and we thus need | |
3342 | * to include the updated inode in the current transaction. | |
3343 | * | |
5dd4056d | 3344 | * Also, dquot_alloc_space() will always dirty the inode when blocks |
1da177e4 LT |
3345 | * are allocated to the file. |
3346 | * | |
3347 | * If the inode is marked synchronous, we don't honour that here - doing | |
3348 | * so would cause a commit on atime updates, which we don't bother doing. | |
3349 | * We handle synchronous inodes at the highest possible level. | |
3350 | */ | |
3351 | void ext3_dirty_inode(struct inode *inode) | |
3352 | { | |
3353 | handle_t *current_handle = ext3_journal_current_handle(); | |
3354 | handle_t *handle; | |
3355 | ||
3356 | handle = ext3_journal_start(inode, 2); | |
3357 | if (IS_ERR(handle)) | |
3358 | goto out; | |
3359 | if (current_handle && | |
3360 | current_handle->h_transaction != handle->h_transaction) { | |
3361 | /* This task has a transaction open against a different fs */ | |
3362 | printk(KERN_EMERG "%s: transactions do not match!\n", | |
e05b6b52 | 3363 | __func__); |
1da177e4 LT |
3364 | } else { |
3365 | jbd_debug(5, "marking dirty. outer handle=%p\n", | |
3366 | current_handle); | |
3367 | ext3_mark_inode_dirty(handle, inode); | |
3368 | } | |
3369 | ext3_journal_stop(handle); | |
3370 | out: | |
3371 | return; | |
3372 | } | |
3373 | ||
d6859bfc | 3374 | #if 0 |
ae6ddcc5 | 3375 | /* |
1da177e4 LT |
3376 | * Bind an inode's backing buffer_head into this transaction, to prevent |
3377 | * it from being flushed to disk early. Unlike | |
3378 | * ext3_reserve_inode_write, this leaves behind no bh reference and | |
3379 | * returns no iloc structure, so the caller needs to repeat the iloc | |
3380 | * lookup to mark the inode dirty later. | |
3381 | */ | |
d6859bfc | 3382 | static int ext3_pin_inode(handle_t *handle, struct inode *inode) |
1da177e4 LT |
3383 | { |
3384 | struct ext3_iloc iloc; | |
3385 | ||
3386 | int err = 0; | |
3387 | if (handle) { | |
3388 | err = ext3_get_inode_loc(inode, &iloc); | |
3389 | if (!err) { | |
3390 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
3391 | err = journal_get_write_access(handle, iloc.bh); | |
3392 | if (!err) | |
ae6ddcc5 | 3393 | err = ext3_journal_dirty_metadata(handle, |
1da177e4 LT |
3394 | iloc.bh); |
3395 | brelse(iloc.bh); | |
3396 | } | |
3397 | } | |
3398 | ext3_std_error(inode->i_sb, err); | |
3399 | return err; | |
3400 | } | |
3401 | #endif | |
3402 | ||
3403 | int ext3_change_inode_journal_flag(struct inode *inode, int val) | |
3404 | { | |
3405 | journal_t *journal; | |
3406 | handle_t *handle; | |
3407 | int err; | |
3408 | ||
3409 | /* | |
3410 | * We have to be very careful here: changing a data block's | |
3411 | * journaling status dynamically is dangerous. If we write a | |
3412 | * data block to the journal, change the status and then delete | |
3413 | * that block, we risk forgetting to revoke the old log record | |
3414 | * from the journal and so a subsequent replay can corrupt data. | |
3415 | * So, first we make sure that the journal is empty and that | |
3416 | * nobody is changing anything. | |
3417 | */ | |
3418 | ||
3419 | journal = EXT3_JOURNAL(inode); | |
e3a68e30 | 3420 | if (is_journal_aborted(journal)) |
1da177e4 LT |
3421 | return -EROFS; |
3422 | ||
3423 | journal_lock_updates(journal); | |
3424 | journal_flush(journal); | |
3425 | ||
3426 | /* | |
3427 | * OK, there are no updates running now, and all cached data is | |
3428 | * synced to disk. We are now in a completely consistent state | |
3429 | * which doesn't have anything in the journal, and we know that | |
3430 | * no filesystem updates are running, so it is safe to modify | |
3431 | * the inode's in-core data-journaling state flag now. | |
3432 | */ | |
3433 | ||
3434 | if (val) | |
3435 | EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL; | |
3436 | else | |
3437 | EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL; | |
3438 | ext3_set_aops(inode); | |
3439 | ||
3440 | journal_unlock_updates(journal); | |
3441 | ||
3442 | /* Finally we can mark the inode as dirty. */ | |
3443 | ||
3444 | handle = ext3_journal_start(inode, 1); | |
3445 | if (IS_ERR(handle)) | |
3446 | return PTR_ERR(handle); | |
3447 | ||
3448 | err = ext3_mark_inode_dirty(handle, inode); | |
3449 | handle->h_sync = 1; | |
3450 | ext3_journal_stop(handle); | |
3451 | ext3_std_error(inode->i_sb, err); | |
3452 | ||
3453 | return err; | |
3454 | } |