]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
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 | |
16 | * (sct@redhat.com), 1993, 1998 | |
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 | |
20 | * (jj@sunsite.ms.mff.cuni.cz) | |
21 | * | |
617ba13b | 22 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
23 | */ |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/time.h> | |
dab291af MC |
28 | #include <linux/ext4_jbd2.h> |
29 | #include <linux/jbd2.h> | |
ac27a0ec DK |
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> | |
38 | #include <linux/bio.h> | |
39 | #include "xattr.h" | |
40 | #include "acl.h" | |
41 | ||
ac27a0ec DK |
42 | /* |
43 | * Test whether an inode is a fast symlink. | |
44 | */ | |
617ba13b | 45 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 46 | { |
617ba13b | 47 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
48 | (inode->i_sb->s_blocksize >> 9) : 0; |
49 | ||
50 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
51 | } | |
52 | ||
53 | /* | |
617ba13b | 54 | * The ext4 forget function must perform a revoke if we are freeing data |
ac27a0ec DK |
55 | * which has been journaled. Metadata (eg. indirect blocks) must be |
56 | * revoked in all cases. | |
57 | * | |
58 | * "bh" may be NULL: a metadata block may have been freed from memory | |
59 | * but there may still be a record of it in the journal, and that record | |
60 | * still needs to be revoked. | |
61 | */ | |
617ba13b MC |
62 | int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode, |
63 | struct buffer_head *bh, ext4_fsblk_t blocknr) | |
ac27a0ec DK |
64 | { |
65 | int err; | |
66 | ||
67 | might_sleep(); | |
68 | ||
69 | BUFFER_TRACE(bh, "enter"); | |
70 | ||
71 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " | |
72 | "data mode %lx\n", | |
73 | bh, is_metadata, inode->i_mode, | |
74 | test_opt(inode->i_sb, DATA_FLAGS)); | |
75 | ||
76 | /* Never use the revoke function if we are doing full data | |
77 | * journaling: there is no need to, and a V1 superblock won't | |
78 | * support it. Otherwise, only skip the revoke on un-journaled | |
79 | * data blocks. */ | |
80 | ||
617ba13b MC |
81 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA || |
82 | (!is_metadata && !ext4_should_journal_data(inode))) { | |
ac27a0ec | 83 | if (bh) { |
dab291af | 84 | BUFFER_TRACE(bh, "call jbd2_journal_forget"); |
617ba13b | 85 | return ext4_journal_forget(handle, bh); |
ac27a0ec DK |
86 | } |
87 | return 0; | |
88 | } | |
89 | ||
90 | /* | |
91 | * data!=journal && (is_metadata || should_journal_data(inode)) | |
92 | */ | |
617ba13b MC |
93 | BUFFER_TRACE(bh, "call ext4_journal_revoke"); |
94 | err = ext4_journal_revoke(handle, blocknr, bh); | |
ac27a0ec | 95 | if (err) |
617ba13b | 96 | ext4_abort(inode->i_sb, __FUNCTION__, |
ac27a0ec DK |
97 | "error %d when attempting revoke", err); |
98 | BUFFER_TRACE(bh, "exit"); | |
99 | return err; | |
100 | } | |
101 | ||
102 | /* | |
103 | * Work out how many blocks we need to proceed with the next chunk of a | |
104 | * truncate transaction. | |
105 | */ | |
106 | static unsigned long blocks_for_truncate(struct inode *inode) | |
107 | { | |
725d26d3 | 108 | ext4_lblk_t needed; |
ac27a0ec DK |
109 | |
110 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
111 | ||
112 | /* Give ourselves just enough room to cope with inodes in which | |
113 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
114 | * which resulted in random data in an inode which looked enough | |
617ba13b | 115 | * like a regular file for ext4 to try to delete it. Things |
ac27a0ec DK |
116 | * will go a bit crazy if that happens, but at least we should |
117 | * try not to panic the whole kernel. */ | |
118 | if (needed < 2) | |
119 | needed = 2; | |
120 | ||
121 | /* But we need to bound the transaction so we don't overflow the | |
122 | * journal. */ | |
617ba13b MC |
123 | if (needed > EXT4_MAX_TRANS_DATA) |
124 | needed = EXT4_MAX_TRANS_DATA; | |
ac27a0ec | 125 | |
617ba13b | 126 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
ac27a0ec DK |
127 | } |
128 | ||
129 | /* | |
130 | * Truncate transactions can be complex and absolutely huge. So we need to | |
131 | * be able to restart the transaction at a conventient checkpoint to make | |
132 | * sure we don't overflow the journal. | |
133 | * | |
134 | * start_transaction gets us a new handle for a truncate transaction, | |
135 | * and extend_transaction tries to extend the existing one a bit. If | |
136 | * extend fails, we need to propagate the failure up and restart the | |
137 | * transaction in the top-level truncate loop. --sct | |
138 | */ | |
139 | static handle_t *start_transaction(struct inode *inode) | |
140 | { | |
141 | handle_t *result; | |
142 | ||
617ba13b | 143 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
ac27a0ec DK |
144 | if (!IS_ERR(result)) |
145 | return result; | |
146 | ||
617ba13b | 147 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
ac27a0ec DK |
148 | return result; |
149 | } | |
150 | ||
151 | /* | |
152 | * Try to extend this transaction for the purposes of truncation. | |
153 | * | |
154 | * Returns 0 if we managed to create more room. If we can't create more | |
155 | * room, and the transaction must be restarted we return 1. | |
156 | */ | |
157 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
158 | { | |
617ba13b | 159 | if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS) |
ac27a0ec | 160 | return 0; |
617ba13b | 161 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
ac27a0ec DK |
162 | return 0; |
163 | return 1; | |
164 | } | |
165 | ||
166 | /* | |
167 | * Restart the transaction associated with *handle. This does a commit, | |
168 | * so before we call here everything must be consistently dirtied against | |
169 | * this transaction. | |
170 | */ | |
617ba13b | 171 | static int ext4_journal_test_restart(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
172 | { |
173 | jbd_debug(2, "restarting handle %p\n", handle); | |
617ba13b | 174 | return ext4_journal_restart(handle, blocks_for_truncate(inode)); |
ac27a0ec DK |
175 | } |
176 | ||
177 | /* | |
178 | * Called at the last iput() if i_nlink is zero. | |
179 | */ | |
617ba13b | 180 | void ext4_delete_inode (struct inode * inode) |
ac27a0ec DK |
181 | { |
182 | handle_t *handle; | |
183 | ||
184 | truncate_inode_pages(&inode->i_data, 0); | |
185 | ||
186 | if (is_bad_inode(inode)) | |
187 | goto no_delete; | |
188 | ||
189 | handle = start_transaction(inode); | |
190 | if (IS_ERR(handle)) { | |
191 | /* | |
192 | * If we're going to skip the normal cleanup, we still need to | |
193 | * make sure that the in-core orphan linked list is properly | |
194 | * cleaned up. | |
195 | */ | |
617ba13b | 196 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
197 | goto no_delete; |
198 | } | |
199 | ||
200 | if (IS_SYNC(inode)) | |
201 | handle->h_sync = 1; | |
202 | inode->i_size = 0; | |
203 | if (inode->i_blocks) | |
617ba13b | 204 | ext4_truncate(inode); |
ac27a0ec | 205 | /* |
617ba13b | 206 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 207 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 208 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 209 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 210 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
211 | * (Well, we could do this if we need to, but heck - it works) |
212 | */ | |
617ba13b MC |
213 | ext4_orphan_del(handle, inode); |
214 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
215 | |
216 | /* | |
217 | * One subtle ordering requirement: if anything has gone wrong | |
218 | * (transaction abort, IO errors, whatever), then we can still | |
219 | * do these next steps (the fs will already have been marked as | |
220 | * having errors), but we can't free the inode if the mark_dirty | |
221 | * fails. | |
222 | */ | |
617ba13b | 223 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec DK |
224 | /* If that failed, just do the required in-core inode clear. */ |
225 | clear_inode(inode); | |
226 | else | |
617ba13b MC |
227 | ext4_free_inode(handle, inode); |
228 | ext4_journal_stop(handle); | |
ac27a0ec DK |
229 | return; |
230 | no_delete: | |
231 | clear_inode(inode); /* We must guarantee clearing of inode... */ | |
232 | } | |
233 | ||
234 | typedef struct { | |
235 | __le32 *p; | |
236 | __le32 key; | |
237 | struct buffer_head *bh; | |
238 | } Indirect; | |
239 | ||
240 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
241 | { | |
242 | p->key = *(p->p = v); | |
243 | p->bh = bh; | |
244 | } | |
245 | ||
246 | static int verify_chain(Indirect *from, Indirect *to) | |
247 | { | |
248 | while (from <= to && from->key == *from->p) | |
249 | from++; | |
250 | return (from > to); | |
251 | } | |
252 | ||
253 | /** | |
617ba13b | 254 | * ext4_block_to_path - parse the block number into array of offsets |
ac27a0ec DK |
255 | * @inode: inode in question (we are only interested in its superblock) |
256 | * @i_block: block number to be parsed | |
257 | * @offsets: array to store the offsets in | |
8c55e204 DK |
258 | * @boundary: set this non-zero if the referred-to block is likely to be |
259 | * followed (on disk) by an indirect block. | |
ac27a0ec | 260 | * |
617ba13b | 261 | * To store the locations of file's data ext4 uses a data structure common |
ac27a0ec DK |
262 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
263 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
264 | * This function translates the block number into path in that tree - | |
265 | * return value is the path length and @offsets[n] is the offset of | |
266 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
267 | * (negative or too large) warning is printed and zero returned. | |
268 | * | |
269 | * Note: function doesn't find node addresses, so no IO is needed. All | |
270 | * we need to know is the capacity of indirect blocks (taken from the | |
271 | * inode->i_sb). | |
272 | */ | |
273 | ||
274 | /* | |
275 | * Portability note: the last comparison (check that we fit into triple | |
276 | * indirect block) is spelled differently, because otherwise on an | |
277 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
278 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
279 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
280 | * i_block would have to be negative in the very beginning, so we would not | |
281 | * get there at all. | |
282 | */ | |
283 | ||
617ba13b | 284 | static int ext4_block_to_path(struct inode *inode, |
725d26d3 AK |
285 | ext4_lblk_t i_block, |
286 | ext4_lblk_t offsets[4], int *boundary) | |
ac27a0ec | 287 | { |
617ba13b MC |
288 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
289 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
290 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
ac27a0ec DK |
291 | indirect_blocks = ptrs, |
292 | double_blocks = (1 << (ptrs_bits * 2)); | |
293 | int n = 0; | |
294 | int final = 0; | |
295 | ||
296 | if (i_block < 0) { | |
617ba13b | 297 | ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0"); |
ac27a0ec DK |
298 | } else if (i_block < direct_blocks) { |
299 | offsets[n++] = i_block; | |
300 | final = direct_blocks; | |
301 | } else if ( (i_block -= direct_blocks) < indirect_blocks) { | |
617ba13b | 302 | offsets[n++] = EXT4_IND_BLOCK; |
ac27a0ec DK |
303 | offsets[n++] = i_block; |
304 | final = ptrs; | |
305 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
617ba13b | 306 | offsets[n++] = EXT4_DIND_BLOCK; |
ac27a0ec DK |
307 | offsets[n++] = i_block >> ptrs_bits; |
308 | offsets[n++] = i_block & (ptrs - 1); | |
309 | final = ptrs; | |
310 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
617ba13b | 311 | offsets[n++] = EXT4_TIND_BLOCK; |
ac27a0ec DK |
312 | offsets[n++] = i_block >> (ptrs_bits * 2); |
313 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
314 | offsets[n++] = i_block & (ptrs - 1); | |
315 | final = ptrs; | |
316 | } else { | |
617ba13b | 317 | ext4_warning(inode->i_sb, "ext4_block_to_path", "block > big"); |
ac27a0ec DK |
318 | } |
319 | if (boundary) | |
320 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
321 | return n; | |
322 | } | |
323 | ||
324 | /** | |
617ba13b | 325 | * ext4_get_branch - read the chain of indirect blocks leading to data |
ac27a0ec DK |
326 | * @inode: inode in question |
327 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
328 | * @offsets: offsets of pointers in inode/indirect blocks | |
329 | * @chain: place to store the result | |
330 | * @err: here we store the error value | |
331 | * | |
332 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
333 | * if everything went OK or the pointer to the last filled triple | |
334 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
335 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
336 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
337 | * number (it points into struct inode for i==0 and into the bh->b_data | |
338 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
339 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
340 | * numbers of the chain, addresses they were taken from (and where we can | |
341 | * verify that chain did not change) and buffer_heads hosting these | |
342 | * numbers. | |
343 | * | |
344 | * Function stops when it stumbles upon zero pointer (absent block) | |
345 | * (pointer to last triple returned, *@err == 0) | |
346 | * or when it gets an IO error reading an indirect block | |
347 | * (ditto, *@err == -EIO) | |
348 | * or when it notices that chain had been changed while it was reading | |
349 | * (ditto, *@err == -EAGAIN) | |
350 | * or when it reads all @depth-1 indirect blocks successfully and finds | |
351 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
352 | */ | |
725d26d3 AK |
353 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
354 | ext4_lblk_t *offsets, | |
ac27a0ec DK |
355 | Indirect chain[4], int *err) |
356 | { | |
357 | struct super_block *sb = inode->i_sb; | |
358 | Indirect *p = chain; | |
359 | struct buffer_head *bh; | |
360 | ||
361 | *err = 0; | |
362 | /* i_data is not going away, no lock needed */ | |
617ba13b | 363 | add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets); |
ac27a0ec DK |
364 | if (!p->key) |
365 | goto no_block; | |
366 | while (--depth) { | |
367 | bh = sb_bread(sb, le32_to_cpu(p->key)); | |
368 | if (!bh) | |
369 | goto failure; | |
370 | /* Reader: pointers */ | |
371 | if (!verify_chain(chain, p)) | |
372 | goto changed; | |
373 | add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); | |
374 | /* Reader: end */ | |
375 | if (!p->key) | |
376 | goto no_block; | |
377 | } | |
378 | return NULL; | |
379 | ||
380 | changed: | |
381 | brelse(bh); | |
382 | *err = -EAGAIN; | |
383 | goto no_block; | |
384 | failure: | |
385 | *err = -EIO; | |
386 | no_block: | |
387 | return p; | |
388 | } | |
389 | ||
390 | /** | |
617ba13b | 391 | * ext4_find_near - find a place for allocation with sufficient locality |
ac27a0ec DK |
392 | * @inode: owner |
393 | * @ind: descriptor of indirect block. | |
394 | * | |
395 | * This function returns the prefered place for block allocation. | |
396 | * It is used when heuristic for sequential allocation fails. | |
397 | * Rules are: | |
398 | * + if there is a block to the left of our position - allocate near it. | |
399 | * + if pointer will live in indirect block - allocate near that block. | |
400 | * + if pointer will live in inode - allocate in the same | |
401 | * cylinder group. | |
402 | * | |
403 | * In the latter case we colour the starting block by the callers PID to | |
404 | * prevent it from clashing with concurrent allocations for a different inode | |
405 | * in the same block group. The PID is used here so that functionally related | |
406 | * files will be close-by on-disk. | |
407 | * | |
408 | * Caller must make sure that @ind is valid and will stay that way. | |
409 | */ | |
617ba13b | 410 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
ac27a0ec | 411 | { |
617ba13b | 412 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec DK |
413 | __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data; |
414 | __le32 *p; | |
617ba13b MC |
415 | ext4_fsblk_t bg_start; |
416 | ext4_grpblk_t colour; | |
ac27a0ec DK |
417 | |
418 | /* Try to find previous block */ | |
419 | for (p = ind->p - 1; p >= start; p--) { | |
420 | if (*p) | |
421 | return le32_to_cpu(*p); | |
422 | } | |
423 | ||
424 | /* No such thing, so let's try location of indirect block */ | |
425 | if (ind->bh) | |
426 | return ind->bh->b_blocknr; | |
427 | ||
428 | /* | |
429 | * It is going to be referred to from the inode itself? OK, just put it | |
430 | * into the same cylinder group then. | |
431 | */ | |
617ba13b | 432 | bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group); |
ac27a0ec | 433 | colour = (current->pid % 16) * |
617ba13b | 434 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
ac27a0ec DK |
435 | return bg_start + colour; |
436 | } | |
437 | ||
438 | /** | |
617ba13b | 439 | * ext4_find_goal - find a prefered place for allocation. |
ac27a0ec DK |
440 | * @inode: owner |
441 | * @block: block we want | |
442 | * @chain: chain of indirect blocks | |
443 | * @partial: pointer to the last triple within a chain | |
444 | * @goal: place to store the result. | |
445 | * | |
446 | * Normally this function find the prefered place for block allocation, | |
447 | * stores it in *@goal and returns zero. | |
448 | */ | |
449 | ||
725d26d3 | 450 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
ac27a0ec DK |
451 | Indirect chain[4], Indirect *partial) |
452 | { | |
617ba13b | 453 | struct ext4_block_alloc_info *block_i; |
ac27a0ec | 454 | |
617ba13b | 455 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
456 | |
457 | /* | |
458 | * try the heuristic for sequential allocation, | |
459 | * failing that at least try to get decent locality. | |
460 | */ | |
461 | if (block_i && (block == block_i->last_alloc_logical_block + 1) | |
462 | && (block_i->last_alloc_physical_block != 0)) { | |
463 | return block_i->last_alloc_physical_block + 1; | |
464 | } | |
465 | ||
617ba13b | 466 | return ext4_find_near(inode, partial); |
ac27a0ec DK |
467 | } |
468 | ||
469 | /** | |
617ba13b | 470 | * ext4_blks_to_allocate: Look up the block map and count the number |
ac27a0ec DK |
471 | * of direct blocks need to be allocated for the given branch. |
472 | * | |
473 | * @branch: chain of indirect blocks | |
474 | * @k: number of blocks need for indirect blocks | |
475 | * @blks: number of data blocks to be mapped. | |
476 | * @blocks_to_boundary: the offset in the indirect block | |
477 | * | |
478 | * return the total number of blocks to be allocate, including the | |
479 | * direct and indirect blocks. | |
480 | */ | |
617ba13b | 481 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks, |
ac27a0ec DK |
482 | int blocks_to_boundary) |
483 | { | |
484 | unsigned long count = 0; | |
485 | ||
486 | /* | |
487 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
488 | * then it's clear blocks on that path have not allocated | |
489 | */ | |
490 | if (k > 0) { | |
491 | /* right now we don't handle cross boundary allocation */ | |
492 | if (blks < blocks_to_boundary + 1) | |
493 | count += blks; | |
494 | else | |
495 | count += blocks_to_boundary + 1; | |
496 | return count; | |
497 | } | |
498 | ||
499 | count++; | |
500 | while (count < blks && count <= blocks_to_boundary && | |
501 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
502 | count++; | |
503 | } | |
504 | return count; | |
505 | } | |
506 | ||
507 | /** | |
617ba13b | 508 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
ac27a0ec DK |
509 | * @indirect_blks: the number of blocks need to allocate for indirect |
510 | * blocks | |
511 | * | |
512 | * @new_blocks: on return it will store the new block numbers for | |
513 | * the indirect blocks(if needed) and the first direct block, | |
514 | * @blks: on return it will store the total number of allocated | |
515 | * direct blocks | |
516 | */ | |
617ba13b MC |
517 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
518 | ext4_fsblk_t goal, int indirect_blks, int blks, | |
519 | ext4_fsblk_t new_blocks[4], int *err) | |
ac27a0ec DK |
520 | { |
521 | int target, i; | |
522 | unsigned long count = 0; | |
523 | int index = 0; | |
617ba13b | 524 | ext4_fsblk_t current_block = 0; |
ac27a0ec DK |
525 | int ret = 0; |
526 | ||
527 | /* | |
528 | * Here we try to allocate the requested multiple blocks at once, | |
529 | * on a best-effort basis. | |
530 | * To build a branch, we should allocate blocks for | |
531 | * the indirect blocks(if not allocated yet), and at least | |
532 | * the first direct block of this branch. That's the | |
533 | * minimum number of blocks need to allocate(required) | |
534 | */ | |
535 | target = blks + indirect_blks; | |
536 | ||
537 | while (1) { | |
538 | count = target; | |
539 | /* allocating blocks for indirect blocks and direct blocks */ | |
617ba13b | 540 | current_block = ext4_new_blocks(handle,inode,goal,&count,err); |
ac27a0ec DK |
541 | if (*err) |
542 | goto failed_out; | |
543 | ||
544 | target -= count; | |
545 | /* allocate blocks for indirect blocks */ | |
546 | while (index < indirect_blks && count) { | |
547 | new_blocks[index++] = current_block++; | |
548 | count--; | |
549 | } | |
550 | ||
551 | if (count > 0) | |
552 | break; | |
553 | } | |
554 | ||
555 | /* save the new block number for the first direct block */ | |
556 | new_blocks[index] = current_block; | |
557 | ||
558 | /* total number of blocks allocated for direct blocks */ | |
559 | ret = count; | |
560 | *err = 0; | |
561 | return ret; | |
562 | failed_out: | |
563 | for (i = 0; i <index; i++) | |
617ba13b | 564 | ext4_free_blocks(handle, inode, new_blocks[i], 1); |
ac27a0ec DK |
565 | return ret; |
566 | } | |
567 | ||
568 | /** | |
617ba13b | 569 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
ac27a0ec DK |
570 | * @inode: owner |
571 | * @indirect_blks: number of allocated indirect blocks | |
572 | * @blks: number of allocated direct blocks | |
573 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
574 | * @branch: place to store the chain in. | |
575 | * | |
576 | * This function allocates blocks, zeroes out all but the last one, | |
577 | * links them into chain and (if we are synchronous) writes them to disk. | |
578 | * In other words, it prepares a branch that can be spliced onto the | |
579 | * inode. It stores the information about that chain in the branch[], in | |
617ba13b | 580 | * the same format as ext4_get_branch() would do. We are calling it after |
ac27a0ec DK |
581 | * we had read the existing part of chain and partial points to the last |
582 | * triple of that (one with zero ->key). Upon the exit we have the same | |
617ba13b | 583 | * picture as after the successful ext4_get_block(), except that in one |
ac27a0ec DK |
584 | * place chain is disconnected - *branch->p is still zero (we did not |
585 | * set the last link), but branch->key contains the number that should | |
586 | * be placed into *branch->p to fill that gap. | |
587 | * | |
588 | * If allocation fails we free all blocks we've allocated (and forget | |
589 | * their buffer_heads) and return the error value the from failed | |
617ba13b | 590 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
ac27a0ec DK |
591 | * as described above and return 0. |
592 | */ | |
617ba13b MC |
593 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
594 | int indirect_blks, int *blks, ext4_fsblk_t goal, | |
725d26d3 | 595 | ext4_lblk_t *offsets, Indirect *branch) |
ac27a0ec DK |
596 | { |
597 | int blocksize = inode->i_sb->s_blocksize; | |
598 | int i, n = 0; | |
599 | int err = 0; | |
600 | struct buffer_head *bh; | |
601 | int num; | |
617ba13b MC |
602 | ext4_fsblk_t new_blocks[4]; |
603 | ext4_fsblk_t current_block; | |
ac27a0ec | 604 | |
617ba13b | 605 | num = ext4_alloc_blocks(handle, inode, goal, indirect_blks, |
ac27a0ec DK |
606 | *blks, new_blocks, &err); |
607 | if (err) | |
608 | return err; | |
609 | ||
610 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
611 | /* | |
612 | * metadata blocks and data blocks are allocated. | |
613 | */ | |
614 | for (n = 1; n <= indirect_blks; n++) { | |
615 | /* | |
616 | * Get buffer_head for parent block, zero it out | |
617 | * and set the pointer to new one, then send | |
618 | * parent to disk. | |
619 | */ | |
620 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
621 | branch[n].bh = bh; | |
622 | lock_buffer(bh); | |
623 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 624 | err = ext4_journal_get_create_access(handle, bh); |
ac27a0ec DK |
625 | if (err) { |
626 | unlock_buffer(bh); | |
627 | brelse(bh); | |
628 | goto failed; | |
629 | } | |
630 | ||
631 | memset(bh->b_data, 0, blocksize); | |
632 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
633 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
634 | *branch[n].p = branch[n].key; | |
635 | if ( n == indirect_blks) { | |
636 | current_block = new_blocks[n]; | |
637 | /* | |
638 | * End of chain, update the last new metablock of | |
639 | * the chain to point to the new allocated | |
640 | * data blocks numbers | |
641 | */ | |
642 | for (i=1; i < num; i++) | |
643 | *(branch[n].p + i) = cpu_to_le32(++current_block); | |
644 | } | |
645 | BUFFER_TRACE(bh, "marking uptodate"); | |
646 | set_buffer_uptodate(bh); | |
647 | unlock_buffer(bh); | |
648 | ||
617ba13b MC |
649 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
650 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
651 | if (err) |
652 | goto failed; | |
653 | } | |
654 | *blks = num; | |
655 | return err; | |
656 | failed: | |
657 | /* Allocation failed, free what we already allocated */ | |
658 | for (i = 1; i <= n ; i++) { | |
dab291af | 659 | BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget"); |
617ba13b | 660 | ext4_journal_forget(handle, branch[i].bh); |
ac27a0ec DK |
661 | } |
662 | for (i = 0; i <indirect_blks; i++) | |
617ba13b | 663 | ext4_free_blocks(handle, inode, new_blocks[i], 1); |
ac27a0ec | 664 | |
617ba13b | 665 | ext4_free_blocks(handle, inode, new_blocks[i], num); |
ac27a0ec DK |
666 | |
667 | return err; | |
668 | } | |
669 | ||
670 | /** | |
617ba13b | 671 | * ext4_splice_branch - splice the allocated branch onto inode. |
ac27a0ec DK |
672 | * @inode: owner |
673 | * @block: (logical) number of block we are adding | |
674 | * @chain: chain of indirect blocks (with a missing link - see | |
617ba13b | 675 | * ext4_alloc_branch) |
ac27a0ec DK |
676 | * @where: location of missing link |
677 | * @num: number of indirect blocks we are adding | |
678 | * @blks: number of direct blocks we are adding | |
679 | * | |
680 | * This function fills the missing link and does all housekeeping needed in | |
681 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
682 | * chain to new block and return 0. | |
683 | */ | |
617ba13b | 684 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
725d26d3 | 685 | ext4_lblk_t block, Indirect *where, int num, int blks) |
ac27a0ec DK |
686 | { |
687 | int i; | |
688 | int err = 0; | |
617ba13b MC |
689 | struct ext4_block_alloc_info *block_i; |
690 | ext4_fsblk_t current_block; | |
ac27a0ec | 691 | |
617ba13b | 692 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
693 | /* |
694 | * If we're splicing into a [td]indirect block (as opposed to the | |
695 | * inode) then we need to get write access to the [td]indirect block | |
696 | * before the splice. | |
697 | */ | |
698 | if (where->bh) { | |
699 | BUFFER_TRACE(where->bh, "get_write_access"); | |
617ba13b | 700 | err = ext4_journal_get_write_access(handle, where->bh); |
ac27a0ec DK |
701 | if (err) |
702 | goto err_out; | |
703 | } | |
704 | /* That's it */ | |
705 | ||
706 | *where->p = where->key; | |
707 | ||
708 | /* | |
709 | * Update the host buffer_head or inode to point to more just allocated | |
710 | * direct blocks blocks | |
711 | */ | |
712 | if (num == 0 && blks > 1) { | |
713 | current_block = le32_to_cpu(where->key) + 1; | |
714 | for (i = 1; i < blks; i++) | |
715 | *(where->p + i ) = cpu_to_le32(current_block++); | |
716 | } | |
717 | ||
718 | /* | |
719 | * update the most recently allocated logical & physical block | |
720 | * in i_block_alloc_info, to assist find the proper goal block for next | |
721 | * allocation | |
722 | */ | |
723 | if (block_i) { | |
724 | block_i->last_alloc_logical_block = block + blks - 1; | |
725 | block_i->last_alloc_physical_block = | |
726 | le32_to_cpu(where[num].key) + blks - 1; | |
727 | } | |
728 | ||
729 | /* We are done with atomic stuff, now do the rest of housekeeping */ | |
730 | ||
ef7f3835 | 731 | inode->i_ctime = ext4_current_time(inode); |
617ba13b | 732 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
733 | |
734 | /* had we spliced it onto indirect block? */ | |
735 | if (where->bh) { | |
736 | /* | |
737 | * If we spliced it onto an indirect block, we haven't | |
738 | * altered the inode. Note however that if it is being spliced | |
739 | * onto an indirect block at the very end of the file (the | |
740 | * file is growing) then we *will* alter the inode to reflect | |
741 | * the new i_size. But that is not done here - it is done in | |
617ba13b | 742 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
ac27a0ec DK |
743 | */ |
744 | jbd_debug(5, "splicing indirect only\n"); | |
617ba13b MC |
745 | BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata"); |
746 | err = ext4_journal_dirty_metadata(handle, where->bh); | |
ac27a0ec DK |
747 | if (err) |
748 | goto err_out; | |
749 | } else { | |
750 | /* | |
751 | * OK, we spliced it into the inode itself on a direct block. | |
752 | * Inode was dirtied above. | |
753 | */ | |
754 | jbd_debug(5, "splicing direct\n"); | |
755 | } | |
756 | return err; | |
757 | ||
758 | err_out: | |
759 | for (i = 1; i <= num; i++) { | |
dab291af | 760 | BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget"); |
617ba13b MC |
761 | ext4_journal_forget(handle, where[i].bh); |
762 | ext4_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1); | |
ac27a0ec | 763 | } |
617ba13b | 764 | ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks); |
ac27a0ec DK |
765 | |
766 | return err; | |
767 | } | |
768 | ||
769 | /* | |
770 | * Allocation strategy is simple: if we have to allocate something, we will | |
771 | * have to go the whole way to leaf. So let's do it before attaching anything | |
772 | * to tree, set linkage between the newborn blocks, write them if sync is | |
773 | * required, recheck the path, free and repeat if check fails, otherwise | |
774 | * set the last missing link (that will protect us from any truncate-generated | |
775 | * removals - all blocks on the path are immune now) and possibly force the | |
776 | * write on the parent block. | |
777 | * That has a nice additional property: no special recovery from the failed | |
778 | * allocations is needed - we simply release blocks and do not touch anything | |
779 | * reachable from inode. | |
780 | * | |
781 | * `handle' can be NULL if create == 0. | |
782 | * | |
783 | * The BKL may not be held on entry here. Be sure to take it early. | |
784 | * return > 0, # of blocks mapped or allocated. | |
785 | * return = 0, if plain lookup failed. | |
786 | * return < 0, error case. | |
787 | */ | |
617ba13b | 788 | int ext4_get_blocks_handle(handle_t *handle, struct inode *inode, |
725d26d3 | 789 | ext4_lblk_t iblock, unsigned long maxblocks, |
ac27a0ec DK |
790 | struct buffer_head *bh_result, |
791 | int create, int extend_disksize) | |
792 | { | |
793 | int err = -EIO; | |
725d26d3 | 794 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
795 | Indirect chain[4]; |
796 | Indirect *partial; | |
617ba13b | 797 | ext4_fsblk_t goal; |
ac27a0ec DK |
798 | int indirect_blks; |
799 | int blocks_to_boundary = 0; | |
800 | int depth; | |
617ba13b | 801 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 802 | int count = 0; |
617ba13b | 803 | ext4_fsblk_t first_block = 0; |
ac27a0ec DK |
804 | |
805 | ||
a86c6181 | 806 | J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)); |
ac27a0ec | 807 | J_ASSERT(handle != NULL || create == 0); |
725d26d3 AK |
808 | depth = ext4_block_to_path(inode, iblock, offsets, |
809 | &blocks_to_boundary); | |
ac27a0ec DK |
810 | |
811 | if (depth == 0) | |
812 | goto out; | |
813 | ||
617ba13b | 814 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
815 | |
816 | /* Simplest case - block found, no allocation needed */ | |
817 | if (!partial) { | |
818 | first_block = le32_to_cpu(chain[depth - 1].key); | |
819 | clear_buffer_new(bh_result); | |
820 | count++; | |
821 | /*map more blocks*/ | |
822 | while (count < maxblocks && count <= blocks_to_boundary) { | |
617ba13b | 823 | ext4_fsblk_t blk; |
ac27a0ec DK |
824 | |
825 | if (!verify_chain(chain, partial)) { | |
826 | /* | |
827 | * Indirect block might be removed by | |
828 | * truncate while we were reading it. | |
829 | * Handling of that case: forget what we've | |
830 | * got now. Flag the err as EAGAIN, so it | |
831 | * will reread. | |
832 | */ | |
833 | err = -EAGAIN; | |
834 | count = 0; | |
835 | break; | |
836 | } | |
837 | blk = le32_to_cpu(*(chain[depth-1].p + count)); | |
838 | ||
839 | if (blk == first_block + count) | |
840 | count++; | |
841 | else | |
842 | break; | |
843 | } | |
844 | if (err != -EAGAIN) | |
845 | goto got_it; | |
846 | } | |
847 | ||
848 | /* Next simple case - plain lookup or failed read of indirect block */ | |
849 | if (!create || err == -EIO) | |
850 | goto cleanup; | |
851 | ||
852 | mutex_lock(&ei->truncate_mutex); | |
853 | ||
854 | /* | |
855 | * If the indirect block is missing while we are reading | |
617ba13b | 856 | * the chain(ext4_get_branch() returns -EAGAIN err), or |
ac27a0ec DK |
857 | * if the chain has been changed after we grab the semaphore, |
858 | * (either because another process truncated this branch, or | |
859 | * another get_block allocated this branch) re-grab the chain to see if | |
860 | * the request block has been allocated or not. | |
861 | * | |
862 | * Since we already block the truncate/other get_block | |
863 | * at this point, we will have the current copy of the chain when we | |
864 | * splice the branch into the tree. | |
865 | */ | |
866 | if (err == -EAGAIN || !verify_chain(chain, partial)) { | |
867 | while (partial > chain) { | |
868 | brelse(partial->bh); | |
869 | partial--; | |
870 | } | |
617ba13b | 871 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
872 | if (!partial) { |
873 | count++; | |
874 | mutex_unlock(&ei->truncate_mutex); | |
875 | if (err) | |
876 | goto cleanup; | |
877 | clear_buffer_new(bh_result); | |
878 | goto got_it; | |
879 | } | |
880 | } | |
881 | ||
882 | /* | |
883 | * Okay, we need to do block allocation. Lazily initialize the block | |
884 | * allocation info here if necessary | |
885 | */ | |
886 | if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) | |
617ba13b | 887 | ext4_init_block_alloc_info(inode); |
ac27a0ec | 888 | |
617ba13b | 889 | goal = ext4_find_goal(inode, iblock, chain, partial); |
ac27a0ec DK |
890 | |
891 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
892 | indirect_blks = (chain + depth) - partial - 1; | |
893 | ||
894 | /* | |
895 | * Next look up the indirect map to count the totoal number of | |
896 | * direct blocks to allocate for this branch. | |
897 | */ | |
617ba13b | 898 | count = ext4_blks_to_allocate(partial, indirect_blks, |
ac27a0ec DK |
899 | maxblocks, blocks_to_boundary); |
900 | /* | |
617ba13b | 901 | * Block out ext4_truncate while we alter the tree |
ac27a0ec | 902 | */ |
617ba13b | 903 | err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal, |
ac27a0ec DK |
904 | offsets + (partial - chain), partial); |
905 | ||
906 | /* | |
617ba13b | 907 | * The ext4_splice_branch call will free and forget any buffers |
ac27a0ec DK |
908 | * on the new chain if there is a failure, but that risks using |
909 | * up transaction credits, especially for bitmaps where the | |
910 | * credits cannot be returned. Can we handle this somehow? We | |
911 | * may need to return -EAGAIN upwards in the worst case. --sct | |
912 | */ | |
913 | if (!err) | |
617ba13b | 914 | err = ext4_splice_branch(handle, inode, iblock, |
ac27a0ec DK |
915 | partial, indirect_blks, count); |
916 | /* | |
917 | * i_disksize growing is protected by truncate_mutex. Don't forget to | |
918 | * protect it if you're about to implement concurrent | |
617ba13b | 919 | * ext4_get_block() -bzzz |
ac27a0ec DK |
920 | */ |
921 | if (!err && extend_disksize && inode->i_size > ei->i_disksize) | |
922 | ei->i_disksize = inode->i_size; | |
923 | mutex_unlock(&ei->truncate_mutex); | |
924 | if (err) | |
925 | goto cleanup; | |
926 | ||
927 | set_buffer_new(bh_result); | |
928 | got_it: | |
929 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); | |
930 | if (count > blocks_to_boundary) | |
931 | set_buffer_boundary(bh_result); | |
932 | err = count; | |
933 | /* Clean up and exit */ | |
934 | partial = chain + depth - 1; /* the whole chain */ | |
935 | cleanup: | |
936 | while (partial > chain) { | |
937 | BUFFER_TRACE(partial->bh, "call brelse"); | |
938 | brelse(partial->bh); | |
939 | partial--; | |
940 | } | |
941 | BUFFER_TRACE(bh_result, "returned"); | |
942 | out: | |
943 | return err; | |
944 | } | |
945 | ||
617ba13b | 946 | #define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32) |
ac27a0ec | 947 | |
617ba13b | 948 | static int ext4_get_block(struct inode *inode, sector_t iblock, |
ac27a0ec DK |
949 | struct buffer_head *bh_result, int create) |
950 | { | |
3e4fdaf8 | 951 | handle_t *handle = ext4_journal_current_handle(); |
ac27a0ec DK |
952 | int ret = 0; |
953 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; | |
954 | ||
955 | if (!create) | |
956 | goto get_block; /* A read */ | |
957 | ||
958 | if (max_blocks == 1) | |
959 | goto get_block; /* A single block get */ | |
960 | ||
961 | if (handle->h_transaction->t_state == T_LOCKED) { | |
962 | /* | |
963 | * Huge direct-io writes can hold off commits for long | |
964 | * periods of time. Let this commit run. | |
965 | */ | |
617ba13b MC |
966 | ext4_journal_stop(handle); |
967 | handle = ext4_journal_start(inode, DIO_CREDITS); | |
ac27a0ec DK |
968 | if (IS_ERR(handle)) |
969 | ret = PTR_ERR(handle); | |
970 | goto get_block; | |
971 | } | |
972 | ||
617ba13b | 973 | if (handle->h_buffer_credits <= EXT4_RESERVE_TRANS_BLOCKS) { |
ac27a0ec DK |
974 | /* |
975 | * Getting low on buffer credits... | |
976 | */ | |
617ba13b | 977 | ret = ext4_journal_extend(handle, DIO_CREDITS); |
ac27a0ec DK |
978 | if (ret > 0) { |
979 | /* | |
980 | * Couldn't extend the transaction. Start a new one. | |
981 | */ | |
617ba13b | 982 | ret = ext4_journal_restart(handle, DIO_CREDITS); |
ac27a0ec DK |
983 | } |
984 | } | |
985 | ||
986 | get_block: | |
987 | if (ret == 0) { | |
a86c6181 | 988 | ret = ext4_get_blocks_wrap(handle, inode, iblock, |
ac27a0ec DK |
989 | max_blocks, bh_result, create, 0); |
990 | if (ret > 0) { | |
991 | bh_result->b_size = (ret << inode->i_blkbits); | |
992 | ret = 0; | |
993 | } | |
994 | } | |
995 | return ret; | |
996 | } | |
997 | ||
998 | /* | |
999 | * `handle' can be NULL if create is zero | |
1000 | */ | |
617ba13b | 1001 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 1002 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec DK |
1003 | { |
1004 | struct buffer_head dummy; | |
1005 | int fatal = 0, err; | |
1006 | ||
1007 | J_ASSERT(handle != NULL || create == 0); | |
1008 | ||
1009 | dummy.b_state = 0; | |
1010 | dummy.b_blocknr = -1000; | |
1011 | buffer_trace_init(&dummy.b_history); | |
a86c6181 | 1012 | err = ext4_get_blocks_wrap(handle, inode, block, 1, |
ac27a0ec DK |
1013 | &dummy, create, 1); |
1014 | /* | |
617ba13b | 1015 | * ext4_get_blocks_handle() returns number of blocks |
ac27a0ec DK |
1016 | * mapped. 0 in case of a HOLE. |
1017 | */ | |
1018 | if (err > 0) { | |
1019 | if (err > 1) | |
1020 | WARN_ON(1); | |
1021 | err = 0; | |
1022 | } | |
1023 | *errp = err; | |
1024 | if (!err && buffer_mapped(&dummy)) { | |
1025 | struct buffer_head *bh; | |
1026 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); | |
1027 | if (!bh) { | |
1028 | *errp = -EIO; | |
1029 | goto err; | |
1030 | } | |
1031 | if (buffer_new(&dummy)) { | |
1032 | J_ASSERT(create != 0); | |
ac39849d | 1033 | J_ASSERT(handle != NULL); |
ac27a0ec DK |
1034 | |
1035 | /* | |
1036 | * Now that we do not always journal data, we should | |
1037 | * keep in mind whether this should always journal the | |
1038 | * new buffer as metadata. For now, regular file | |
617ba13b | 1039 | * writes use ext4_get_block instead, so it's not a |
ac27a0ec DK |
1040 | * problem. |
1041 | */ | |
1042 | lock_buffer(bh); | |
1043 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 1044 | fatal = ext4_journal_get_create_access(handle, bh); |
ac27a0ec DK |
1045 | if (!fatal && !buffer_uptodate(bh)) { |
1046 | memset(bh->b_data,0,inode->i_sb->s_blocksize); | |
1047 | set_buffer_uptodate(bh); | |
1048 | } | |
1049 | unlock_buffer(bh); | |
617ba13b MC |
1050 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
1051 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
1052 | if (!fatal) |
1053 | fatal = err; | |
1054 | } else { | |
1055 | BUFFER_TRACE(bh, "not a new buffer"); | |
1056 | } | |
1057 | if (fatal) { | |
1058 | *errp = fatal; | |
1059 | brelse(bh); | |
1060 | bh = NULL; | |
1061 | } | |
1062 | return bh; | |
1063 | } | |
1064 | err: | |
1065 | return NULL; | |
1066 | } | |
1067 | ||
617ba13b | 1068 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 1069 | ext4_lblk_t block, int create, int *err) |
ac27a0ec DK |
1070 | { |
1071 | struct buffer_head * bh; | |
1072 | ||
617ba13b | 1073 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
1074 | if (!bh) |
1075 | return bh; | |
1076 | if (buffer_uptodate(bh)) | |
1077 | return bh; | |
1078 | ll_rw_block(READ_META, 1, &bh); | |
1079 | wait_on_buffer(bh); | |
1080 | if (buffer_uptodate(bh)) | |
1081 | return bh; | |
1082 | put_bh(bh); | |
1083 | *err = -EIO; | |
1084 | return NULL; | |
1085 | } | |
1086 | ||
1087 | static int walk_page_buffers( handle_t *handle, | |
1088 | struct buffer_head *head, | |
1089 | unsigned from, | |
1090 | unsigned to, | |
1091 | int *partial, | |
1092 | int (*fn)( handle_t *handle, | |
1093 | struct buffer_head *bh)) | |
1094 | { | |
1095 | struct buffer_head *bh; | |
1096 | unsigned block_start, block_end; | |
1097 | unsigned blocksize = head->b_size; | |
1098 | int err, ret = 0; | |
1099 | struct buffer_head *next; | |
1100 | ||
1101 | for ( bh = head, block_start = 0; | |
1102 | ret == 0 && (bh != head || !block_start); | |
1103 | block_start = block_end, bh = next) | |
1104 | { | |
1105 | next = bh->b_this_page; | |
1106 | block_end = block_start + blocksize; | |
1107 | if (block_end <= from || block_start >= to) { | |
1108 | if (partial && !buffer_uptodate(bh)) | |
1109 | *partial = 1; | |
1110 | continue; | |
1111 | } | |
1112 | err = (*fn)(handle, bh); | |
1113 | if (!ret) | |
1114 | ret = err; | |
1115 | } | |
1116 | return ret; | |
1117 | } | |
1118 | ||
1119 | /* | |
1120 | * To preserve ordering, it is essential that the hole instantiation and | |
1121 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1122 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1123 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1124 | * prepare_write() is the right place. |
1125 | * | |
617ba13b MC |
1126 | * Also, this function can nest inside ext4_writepage() -> |
1127 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
1128 | * has generated enough buffer credits to do the whole page. So we won't |
1129 | * block on the journal in that case, which is good, because the caller may | |
1130 | * be PF_MEMALLOC. | |
1131 | * | |
617ba13b | 1132 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1133 | * quota file writes. If we were to commit the transaction while thus |
1134 | * reentered, there can be a deadlock - we would be holding a quota | |
1135 | * lock, and the commit would never complete if another thread had a | |
1136 | * transaction open and was blocking on the quota lock - a ranking | |
1137 | * violation. | |
1138 | * | |
dab291af | 1139 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1140 | * will _not_ run commit under these circumstances because handle->h_ref |
1141 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1142 | * write. | |
1143 | */ | |
1144 | static int do_journal_get_write_access(handle_t *handle, | |
1145 | struct buffer_head *bh) | |
1146 | { | |
1147 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1148 | return 0; | |
617ba13b | 1149 | return ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1150 | } |
1151 | ||
bfc1af65 NP |
1152 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
1153 | loff_t pos, unsigned len, unsigned flags, | |
1154 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1155 | { |
bfc1af65 | 1156 | struct inode *inode = mapping->host; |
7479d2b9 | 1157 | int ret, needed_blocks = ext4_writepage_trans_blocks(inode); |
ac27a0ec DK |
1158 | handle_t *handle; |
1159 | int retries = 0; | |
bfc1af65 NP |
1160 | struct page *page; |
1161 | pgoff_t index; | |
1162 | unsigned from, to; | |
1163 | ||
1164 | index = pos >> PAGE_CACHE_SHIFT; | |
1165 | from = pos & (PAGE_CACHE_SIZE - 1); | |
1166 | to = from + len; | |
ac27a0ec DK |
1167 | |
1168 | retry: | |
bfc1af65 NP |
1169 | page = __grab_cache_page(mapping, index); |
1170 | if (!page) | |
1171 | return -ENOMEM; | |
1172 | *pagep = page; | |
1173 | ||
1174 | handle = ext4_journal_start(inode, needed_blocks); | |
1175 | if (IS_ERR(handle)) { | |
1176 | unlock_page(page); | |
1177 | page_cache_release(page); | |
1178 | ret = PTR_ERR(handle); | |
1179 | goto out; | |
7479d2b9 | 1180 | } |
ac27a0ec | 1181 | |
bfc1af65 NP |
1182 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
1183 | ext4_get_block); | |
1184 | ||
1185 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
1186 | ret = walk_page_buffers(handle, page_buffers(page), |
1187 | from, to, NULL, do_journal_get_write_access); | |
1188 | } | |
bfc1af65 NP |
1189 | |
1190 | if (ret) { | |
7479d2b9 | 1191 | ext4_journal_stop(handle); |
bfc1af65 NP |
1192 | unlock_page(page); |
1193 | page_cache_release(page); | |
1194 | } | |
1195 | ||
617ba13b | 1196 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 1197 | goto retry; |
7479d2b9 | 1198 | out: |
ac27a0ec DK |
1199 | return ret; |
1200 | } | |
1201 | ||
617ba13b | 1202 | int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh) |
ac27a0ec | 1203 | { |
dab291af | 1204 | int err = jbd2_journal_dirty_data(handle, bh); |
ac27a0ec | 1205 | if (err) |
617ba13b | 1206 | ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__, |
bfc1af65 | 1207 | bh, handle, err); |
ac27a0ec DK |
1208 | return err; |
1209 | } | |
1210 | ||
bfc1af65 NP |
1211 | /* For write_end() in data=journal mode */ |
1212 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
1213 | { |
1214 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1215 | return 0; | |
1216 | set_buffer_uptodate(bh); | |
617ba13b | 1217 | return ext4_journal_dirty_metadata(handle, bh); |
ac27a0ec DK |
1218 | } |
1219 | ||
bfc1af65 NP |
1220 | /* |
1221 | * Generic write_end handler for ordered and writeback ext4 journal modes. | |
1222 | * We can't use generic_write_end, because that unlocks the page and we need to | |
1223 | * unlock the page after ext4_journal_stop, but ext4_journal_stop must run | |
1224 | * after block_write_end. | |
1225 | */ | |
1226 | static int ext4_generic_write_end(struct file *file, | |
1227 | struct address_space *mapping, | |
1228 | loff_t pos, unsigned len, unsigned copied, | |
1229 | struct page *page, void *fsdata) | |
1230 | { | |
1231 | struct inode *inode = file->f_mapping->host; | |
1232 | ||
1233 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
1234 | ||
1235 | if (pos+copied > inode->i_size) { | |
1236 | i_size_write(inode, pos+copied); | |
1237 | mark_inode_dirty(inode); | |
1238 | } | |
1239 | ||
1240 | return copied; | |
1241 | } | |
1242 | ||
ac27a0ec DK |
1243 | /* |
1244 | * We need to pick up the new inode size which generic_commit_write gave us | |
1245 | * `file' can be NULL - eg, when called from page_symlink(). | |
1246 | * | |
617ba13b | 1247 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1248 | * buffers are managed internally. |
1249 | */ | |
bfc1af65 NP |
1250 | static int ext4_ordered_write_end(struct file *file, |
1251 | struct address_space *mapping, | |
1252 | loff_t pos, unsigned len, unsigned copied, | |
1253 | struct page *page, void *fsdata) | |
ac27a0ec | 1254 | { |
617ba13b | 1255 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 NP |
1256 | struct inode *inode = file->f_mapping->host; |
1257 | unsigned from, to; | |
ac27a0ec DK |
1258 | int ret = 0, ret2; |
1259 | ||
bfc1af65 NP |
1260 | from = pos & (PAGE_CACHE_SIZE - 1); |
1261 | to = from + len; | |
1262 | ||
ac27a0ec | 1263 | ret = walk_page_buffers(handle, page_buffers(page), |
617ba13b | 1264 | from, to, NULL, ext4_journal_dirty_data); |
ac27a0ec DK |
1265 | |
1266 | if (ret == 0) { | |
1267 | /* | |
bfc1af65 | 1268 | * generic_write_end() will run mark_inode_dirty() if i_size |
ac27a0ec DK |
1269 | * changes. So let's piggyback the i_disksize mark_inode_dirty |
1270 | * into that. | |
1271 | */ | |
1272 | loff_t new_i_size; | |
1273 | ||
bfc1af65 | 1274 | new_i_size = pos + copied; |
617ba13b MC |
1275 | if (new_i_size > EXT4_I(inode)->i_disksize) |
1276 | EXT4_I(inode)->i_disksize = new_i_size; | |
bfc1af65 NP |
1277 | copied = ext4_generic_write_end(file, mapping, pos, len, copied, |
1278 | page, fsdata); | |
1279 | if (copied < 0) | |
1280 | ret = copied; | |
ac27a0ec | 1281 | } |
617ba13b | 1282 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1283 | if (!ret) |
1284 | ret = ret2; | |
bfc1af65 NP |
1285 | unlock_page(page); |
1286 | page_cache_release(page); | |
1287 | ||
1288 | return ret ? ret : copied; | |
ac27a0ec DK |
1289 | } |
1290 | ||
bfc1af65 NP |
1291 | static int ext4_writeback_write_end(struct file *file, |
1292 | struct address_space *mapping, | |
1293 | loff_t pos, unsigned len, unsigned copied, | |
1294 | struct page *page, void *fsdata) | |
ac27a0ec | 1295 | { |
617ba13b | 1296 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1297 | struct inode *inode = file->f_mapping->host; |
ac27a0ec DK |
1298 | int ret = 0, ret2; |
1299 | loff_t new_i_size; | |
1300 | ||
bfc1af65 | 1301 | new_i_size = pos + copied; |
617ba13b MC |
1302 | if (new_i_size > EXT4_I(inode)->i_disksize) |
1303 | EXT4_I(inode)->i_disksize = new_i_size; | |
ac27a0ec | 1304 | |
bfc1af65 NP |
1305 | copied = ext4_generic_write_end(file, mapping, pos, len, copied, |
1306 | page, fsdata); | |
1307 | if (copied < 0) | |
1308 | ret = copied; | |
ac27a0ec | 1309 | |
617ba13b | 1310 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1311 | if (!ret) |
1312 | ret = ret2; | |
bfc1af65 NP |
1313 | unlock_page(page); |
1314 | page_cache_release(page); | |
1315 | ||
1316 | return ret ? ret : copied; | |
ac27a0ec DK |
1317 | } |
1318 | ||
bfc1af65 NP |
1319 | static int ext4_journalled_write_end(struct file *file, |
1320 | struct address_space *mapping, | |
1321 | loff_t pos, unsigned len, unsigned copied, | |
1322 | struct page *page, void *fsdata) | |
ac27a0ec | 1323 | { |
617ba13b | 1324 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1325 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1326 | int ret = 0, ret2; |
1327 | int partial = 0; | |
bfc1af65 | 1328 | unsigned from, to; |
ac27a0ec | 1329 | |
bfc1af65 NP |
1330 | from = pos & (PAGE_CACHE_SIZE - 1); |
1331 | to = from + len; | |
1332 | ||
1333 | if (copied < len) { | |
1334 | if (!PageUptodate(page)) | |
1335 | copied = 0; | |
1336 | page_zero_new_buffers(page, from+copied, to); | |
1337 | } | |
ac27a0ec DK |
1338 | |
1339 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1340 | to, &partial, write_end_fn); |
ac27a0ec DK |
1341 | if (!partial) |
1342 | SetPageUptodate(page); | |
bfc1af65 NP |
1343 | if (pos+copied > inode->i_size) |
1344 | i_size_write(inode, pos+copied); | |
617ba13b MC |
1345 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
1346 | if (inode->i_size > EXT4_I(inode)->i_disksize) { | |
1347 | EXT4_I(inode)->i_disksize = inode->i_size; | |
1348 | ret2 = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
1349 | if (!ret) |
1350 | ret = ret2; | |
1351 | } | |
bfc1af65 | 1352 | |
617ba13b | 1353 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1354 | if (!ret) |
1355 | ret = ret2; | |
bfc1af65 NP |
1356 | unlock_page(page); |
1357 | page_cache_release(page); | |
1358 | ||
1359 | return ret ? ret : copied; | |
ac27a0ec DK |
1360 | } |
1361 | ||
1362 | /* | |
1363 | * bmap() is special. It gets used by applications such as lilo and by | |
1364 | * the swapper to find the on-disk block of a specific piece of data. | |
1365 | * | |
1366 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 1367 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
1368 | * filesystem and enables swap, then they may get a nasty shock when the |
1369 | * data getting swapped to that swapfile suddenly gets overwritten by | |
1370 | * the original zero's written out previously to the journal and | |
1371 | * awaiting writeback in the kernel's buffer cache. | |
1372 | * | |
1373 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
1374 | * take extra steps to flush any blocks which might be in the cache. | |
1375 | */ | |
617ba13b | 1376 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
1377 | { |
1378 | struct inode *inode = mapping->host; | |
1379 | journal_t *journal; | |
1380 | int err; | |
1381 | ||
617ba13b | 1382 | if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) { |
ac27a0ec DK |
1383 | /* |
1384 | * This is a REALLY heavyweight approach, but the use of | |
1385 | * bmap on dirty files is expected to be extremely rare: | |
1386 | * only if we run lilo or swapon on a freshly made file | |
1387 | * do we expect this to happen. | |
1388 | * | |
1389 | * (bmap requires CAP_SYS_RAWIO so this does not | |
1390 | * represent an unprivileged user DOS attack --- we'd be | |
1391 | * in trouble if mortal users could trigger this path at | |
1392 | * will.) | |
1393 | * | |
617ba13b | 1394 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
1395 | * regular files. If somebody wants to bmap a directory |
1396 | * or symlink and gets confused because the buffer | |
1397 | * hasn't yet been flushed to disk, they deserve | |
1398 | * everything they get. | |
1399 | */ | |
1400 | ||
617ba13b MC |
1401 | EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA; |
1402 | journal = EXT4_JOURNAL(inode); | |
dab291af MC |
1403 | jbd2_journal_lock_updates(journal); |
1404 | err = jbd2_journal_flush(journal); | |
1405 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
1406 | |
1407 | if (err) | |
1408 | return 0; | |
1409 | } | |
1410 | ||
617ba13b | 1411 | return generic_block_bmap(mapping,block,ext4_get_block); |
ac27a0ec DK |
1412 | } |
1413 | ||
1414 | static int bget_one(handle_t *handle, struct buffer_head *bh) | |
1415 | { | |
1416 | get_bh(bh); | |
1417 | return 0; | |
1418 | } | |
1419 | ||
1420 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1421 | { | |
1422 | put_bh(bh); | |
1423 | return 0; | |
1424 | } | |
1425 | ||
dab291af | 1426 | static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh) |
ac27a0ec DK |
1427 | { |
1428 | if (buffer_mapped(bh)) | |
617ba13b | 1429 | return ext4_journal_dirty_data(handle, bh); |
ac27a0ec DK |
1430 | return 0; |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * Note that we always start a transaction even if we're not journalling | |
1435 | * data. This is to preserve ordering: any hole instantiation within | |
617ba13b | 1436 | * __block_write_full_page -> ext4_get_block() should be journalled |
ac27a0ec DK |
1437 | * along with the data so we don't crash and then get metadata which |
1438 | * refers to old data. | |
1439 | * | |
1440 | * In all journalling modes block_write_full_page() will start the I/O. | |
1441 | * | |
1442 | * Problem: | |
1443 | * | |
617ba13b MC |
1444 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
1445 | * ext4_writepage() | |
ac27a0ec DK |
1446 | * |
1447 | * Similar for: | |
1448 | * | |
617ba13b | 1449 | * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ... |
ac27a0ec | 1450 | * |
617ba13b | 1451 | * Same applies to ext4_get_block(). We will deadlock on various things like |
ac27a0ec DK |
1452 | * lock_journal and i_truncate_mutex. |
1453 | * | |
1454 | * Setting PF_MEMALLOC here doesn't work - too many internal memory | |
1455 | * allocations fail. | |
1456 | * | |
1457 | * 16May01: If we're reentered then journal_current_handle() will be | |
1458 | * non-zero. We simply *return*. | |
1459 | * | |
1460 | * 1 July 2001: @@@ FIXME: | |
1461 | * In journalled data mode, a data buffer may be metadata against the | |
1462 | * current transaction. But the same file is part of a shared mapping | |
1463 | * and someone does a writepage() on it. | |
1464 | * | |
1465 | * We will move the buffer onto the async_data list, but *after* it has | |
1466 | * been dirtied. So there's a small window where we have dirty data on | |
1467 | * BJ_Metadata. | |
1468 | * | |
1469 | * Note that this only applies to the last partial page in the file. The | |
1470 | * bit which block_write_full_page() uses prepare/commit for. (That's | |
1471 | * broken code anyway: it's wrong for msync()). | |
1472 | * | |
1473 | * It's a rare case: affects the final partial page, for journalled data | |
1474 | * where the file is subject to bith write() and writepage() in the same | |
1475 | * transction. To fix it we'll need a custom block_write_full_page(). | |
1476 | * We'll probably need that anyway for journalling writepage() output. | |
1477 | * | |
1478 | * We don't honour synchronous mounts for writepage(). That would be | |
1479 | * disastrous. Any write() or metadata operation will sync the fs for | |
1480 | * us. | |
1481 | * | |
1482 | * AKPM2: if all the page's buffers are mapped to disk and !data=journal, | |
1483 | * we don't need to open a transaction here. | |
1484 | */ | |
617ba13b | 1485 | static int ext4_ordered_writepage(struct page *page, |
ac27a0ec DK |
1486 | struct writeback_control *wbc) |
1487 | { | |
1488 | struct inode *inode = page->mapping->host; | |
1489 | struct buffer_head *page_bufs; | |
1490 | handle_t *handle = NULL; | |
1491 | int ret = 0; | |
1492 | int err; | |
1493 | ||
1494 | J_ASSERT(PageLocked(page)); | |
1495 | ||
1496 | /* | |
1497 | * We give up here if we're reentered, because it might be for a | |
1498 | * different filesystem. | |
1499 | */ | |
617ba13b | 1500 | if (ext4_journal_current_handle()) |
ac27a0ec DK |
1501 | goto out_fail; |
1502 | ||
617ba13b | 1503 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
ac27a0ec DK |
1504 | |
1505 | if (IS_ERR(handle)) { | |
1506 | ret = PTR_ERR(handle); | |
1507 | goto out_fail; | |
1508 | } | |
1509 | ||
1510 | if (!page_has_buffers(page)) { | |
1511 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
1512 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
1513 | } | |
1514 | page_bufs = page_buffers(page); | |
1515 | walk_page_buffers(handle, page_bufs, 0, | |
1516 | PAGE_CACHE_SIZE, NULL, bget_one); | |
1517 | ||
617ba13b | 1518 | ret = block_write_full_page(page, ext4_get_block, wbc); |
ac27a0ec DK |
1519 | |
1520 | /* | |
1521 | * The page can become unlocked at any point now, and | |
1522 | * truncate can then come in and change things. So we | |
1523 | * can't touch *page from now on. But *page_bufs is | |
1524 | * safe due to elevated refcount. | |
1525 | */ | |
1526 | ||
1527 | /* | |
1528 | * And attach them to the current transaction. But only if | |
1529 | * block_write_full_page() succeeded. Otherwise they are unmapped, | |
1530 | * and generally junk. | |
1531 | */ | |
1532 | if (ret == 0) { | |
1533 | err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, | |
dab291af | 1534 | NULL, jbd2_journal_dirty_data_fn); |
ac27a0ec DK |
1535 | if (!ret) |
1536 | ret = err; | |
1537 | } | |
1538 | walk_page_buffers(handle, page_bufs, 0, | |
1539 | PAGE_CACHE_SIZE, NULL, bput_one); | |
617ba13b | 1540 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
1541 | if (!ret) |
1542 | ret = err; | |
1543 | return ret; | |
1544 | ||
1545 | out_fail: | |
1546 | redirty_page_for_writepage(wbc, page); | |
1547 | unlock_page(page); | |
1548 | return ret; | |
1549 | } | |
1550 | ||
617ba13b | 1551 | static int ext4_writeback_writepage(struct page *page, |
ac27a0ec DK |
1552 | struct writeback_control *wbc) |
1553 | { | |
1554 | struct inode *inode = page->mapping->host; | |
1555 | handle_t *handle = NULL; | |
1556 | int ret = 0; | |
1557 | int err; | |
1558 | ||
617ba13b | 1559 | if (ext4_journal_current_handle()) |
ac27a0ec DK |
1560 | goto out_fail; |
1561 | ||
617ba13b | 1562 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
ac27a0ec DK |
1563 | if (IS_ERR(handle)) { |
1564 | ret = PTR_ERR(handle); | |
1565 | goto out_fail; | |
1566 | } | |
1567 | ||
617ba13b MC |
1568 | if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) |
1569 | ret = nobh_writepage(page, ext4_get_block, wbc); | |
ac27a0ec | 1570 | else |
617ba13b | 1571 | ret = block_write_full_page(page, ext4_get_block, wbc); |
ac27a0ec | 1572 | |
617ba13b | 1573 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
1574 | if (!ret) |
1575 | ret = err; | |
1576 | return ret; | |
1577 | ||
1578 | out_fail: | |
1579 | redirty_page_for_writepage(wbc, page); | |
1580 | unlock_page(page); | |
1581 | return ret; | |
1582 | } | |
1583 | ||
617ba13b | 1584 | static int ext4_journalled_writepage(struct page *page, |
ac27a0ec DK |
1585 | struct writeback_control *wbc) |
1586 | { | |
1587 | struct inode *inode = page->mapping->host; | |
1588 | handle_t *handle = NULL; | |
1589 | int ret = 0; | |
1590 | int err; | |
1591 | ||
617ba13b | 1592 | if (ext4_journal_current_handle()) |
ac27a0ec DK |
1593 | goto no_write; |
1594 | ||
617ba13b | 1595 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
ac27a0ec DK |
1596 | if (IS_ERR(handle)) { |
1597 | ret = PTR_ERR(handle); | |
1598 | goto no_write; | |
1599 | } | |
1600 | ||
1601 | if (!page_has_buffers(page) || PageChecked(page)) { | |
1602 | /* | |
1603 | * It's mmapped pagecache. Add buffers and journal it. There | |
1604 | * doesn't seem much point in redirtying the page here. | |
1605 | */ | |
1606 | ClearPageChecked(page); | |
1607 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, | |
617ba13b | 1608 | ext4_get_block); |
ac27a0ec | 1609 | if (ret != 0) { |
617ba13b | 1610 | ext4_journal_stop(handle); |
ac27a0ec DK |
1611 | goto out_unlock; |
1612 | } | |
1613 | ret = walk_page_buffers(handle, page_buffers(page), 0, | |
1614 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); | |
1615 | ||
1616 | err = walk_page_buffers(handle, page_buffers(page), 0, | |
bfc1af65 | 1617 | PAGE_CACHE_SIZE, NULL, write_end_fn); |
ac27a0ec DK |
1618 | if (ret == 0) |
1619 | ret = err; | |
617ba13b | 1620 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
ac27a0ec DK |
1621 | unlock_page(page); |
1622 | } else { | |
1623 | /* | |
1624 | * It may be a page full of checkpoint-mode buffers. We don't | |
1625 | * really know unless we go poke around in the buffer_heads. | |
1626 | * But block_write_full_page will do the right thing. | |
1627 | */ | |
617ba13b | 1628 | ret = block_write_full_page(page, ext4_get_block, wbc); |
ac27a0ec | 1629 | } |
617ba13b | 1630 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
1631 | if (!ret) |
1632 | ret = err; | |
1633 | out: | |
1634 | return ret; | |
1635 | ||
1636 | no_write: | |
1637 | redirty_page_for_writepage(wbc, page); | |
1638 | out_unlock: | |
1639 | unlock_page(page); | |
1640 | goto out; | |
1641 | } | |
1642 | ||
617ba13b | 1643 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 1644 | { |
617ba13b | 1645 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
1646 | } |
1647 | ||
1648 | static int | |
617ba13b | 1649 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
1650 | struct list_head *pages, unsigned nr_pages) |
1651 | { | |
617ba13b | 1652 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
1653 | } |
1654 | ||
617ba13b | 1655 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 1656 | { |
617ba13b | 1657 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
1658 | |
1659 | /* | |
1660 | * If it's a full truncate we just forget about the pending dirtying | |
1661 | */ | |
1662 | if (offset == 0) | |
1663 | ClearPageChecked(page); | |
1664 | ||
dab291af | 1665 | jbd2_journal_invalidatepage(journal, page, offset); |
ac27a0ec DK |
1666 | } |
1667 | ||
617ba13b | 1668 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 1669 | { |
617ba13b | 1670 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
1671 | |
1672 | WARN_ON(PageChecked(page)); | |
1673 | if (!page_has_buffers(page)) | |
1674 | return 0; | |
dab291af | 1675 | return jbd2_journal_try_to_free_buffers(journal, page, wait); |
ac27a0ec DK |
1676 | } |
1677 | ||
1678 | /* | |
1679 | * If the O_DIRECT write will extend the file then add this inode to the | |
1680 | * orphan list. So recovery will truncate it back to the original size | |
1681 | * if the machine crashes during the write. | |
1682 | * | |
1683 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
1684 | * crashes then stale disk data _may_ be exposed inside the file. | |
1685 | */ | |
617ba13b | 1686 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, |
ac27a0ec DK |
1687 | const struct iovec *iov, loff_t offset, |
1688 | unsigned long nr_segs) | |
1689 | { | |
1690 | struct file *file = iocb->ki_filp; | |
1691 | struct inode *inode = file->f_mapping->host; | |
617ba13b | 1692 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec DK |
1693 | handle_t *handle = NULL; |
1694 | ssize_t ret; | |
1695 | int orphan = 0; | |
1696 | size_t count = iov_length(iov, nr_segs); | |
1697 | ||
1698 | if (rw == WRITE) { | |
1699 | loff_t final_size = offset + count; | |
1700 | ||
617ba13b | 1701 | handle = ext4_journal_start(inode, DIO_CREDITS); |
ac27a0ec DK |
1702 | if (IS_ERR(handle)) { |
1703 | ret = PTR_ERR(handle); | |
1704 | goto out; | |
1705 | } | |
1706 | if (final_size > inode->i_size) { | |
617ba13b | 1707 | ret = ext4_orphan_add(handle, inode); |
ac27a0ec DK |
1708 | if (ret) |
1709 | goto out_stop; | |
1710 | orphan = 1; | |
1711 | ei->i_disksize = inode->i_size; | |
1712 | } | |
1713 | } | |
1714 | ||
1715 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, | |
1716 | offset, nr_segs, | |
617ba13b | 1717 | ext4_get_block, NULL); |
ac27a0ec DK |
1718 | |
1719 | /* | |
617ba13b | 1720 | * Reacquire the handle: ext4_get_block() can restart the transaction |
ac27a0ec | 1721 | */ |
3e4fdaf8 | 1722 | handle = ext4_journal_current_handle(); |
ac27a0ec DK |
1723 | |
1724 | out_stop: | |
1725 | if (handle) { | |
1726 | int err; | |
1727 | ||
1728 | if (orphan && inode->i_nlink) | |
617ba13b | 1729 | ext4_orphan_del(handle, inode); |
ac27a0ec DK |
1730 | if (orphan && ret > 0) { |
1731 | loff_t end = offset + ret; | |
1732 | if (end > inode->i_size) { | |
1733 | ei->i_disksize = end; | |
1734 | i_size_write(inode, end); | |
1735 | /* | |
1736 | * We're going to return a positive `ret' | |
1737 | * here due to non-zero-length I/O, so there's | |
1738 | * no way of reporting error returns from | |
617ba13b | 1739 | * ext4_mark_inode_dirty() to userspace. So |
ac27a0ec DK |
1740 | * ignore it. |
1741 | */ | |
617ba13b | 1742 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1743 | } |
1744 | } | |
617ba13b | 1745 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
1746 | if (ret == 0) |
1747 | ret = err; | |
1748 | } | |
1749 | out: | |
1750 | return ret; | |
1751 | } | |
1752 | ||
1753 | /* | |
617ba13b | 1754 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
1755 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
1756 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
1757 | * not necessarily locked. | |
1758 | * | |
1759 | * We cannot just dirty the page and leave attached buffers clean, because the | |
1760 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
1761 | * or jbddirty because all the journalling code will explode. | |
1762 | * | |
1763 | * So what we do is to mark the page "pending dirty" and next time writepage | |
1764 | * is called, propagate that into the buffers appropriately. | |
1765 | */ | |
617ba13b | 1766 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
1767 | { |
1768 | SetPageChecked(page); | |
1769 | return __set_page_dirty_nobuffers(page); | |
1770 | } | |
1771 | ||
617ba13b MC |
1772 | static const struct address_space_operations ext4_ordered_aops = { |
1773 | .readpage = ext4_readpage, | |
1774 | .readpages = ext4_readpages, | |
1775 | .writepage = ext4_ordered_writepage, | |
ac27a0ec | 1776 | .sync_page = block_sync_page, |
bfc1af65 NP |
1777 | .write_begin = ext4_write_begin, |
1778 | .write_end = ext4_ordered_write_end, | |
617ba13b MC |
1779 | .bmap = ext4_bmap, |
1780 | .invalidatepage = ext4_invalidatepage, | |
1781 | .releasepage = ext4_releasepage, | |
1782 | .direct_IO = ext4_direct_IO, | |
ac27a0ec DK |
1783 | .migratepage = buffer_migrate_page, |
1784 | }; | |
1785 | ||
617ba13b MC |
1786 | static const struct address_space_operations ext4_writeback_aops = { |
1787 | .readpage = ext4_readpage, | |
1788 | .readpages = ext4_readpages, | |
1789 | .writepage = ext4_writeback_writepage, | |
ac27a0ec | 1790 | .sync_page = block_sync_page, |
bfc1af65 NP |
1791 | .write_begin = ext4_write_begin, |
1792 | .write_end = ext4_writeback_write_end, | |
617ba13b MC |
1793 | .bmap = ext4_bmap, |
1794 | .invalidatepage = ext4_invalidatepage, | |
1795 | .releasepage = ext4_releasepage, | |
1796 | .direct_IO = ext4_direct_IO, | |
ac27a0ec DK |
1797 | .migratepage = buffer_migrate_page, |
1798 | }; | |
1799 | ||
617ba13b MC |
1800 | static const struct address_space_operations ext4_journalled_aops = { |
1801 | .readpage = ext4_readpage, | |
1802 | .readpages = ext4_readpages, | |
1803 | .writepage = ext4_journalled_writepage, | |
ac27a0ec | 1804 | .sync_page = block_sync_page, |
bfc1af65 NP |
1805 | .write_begin = ext4_write_begin, |
1806 | .write_end = ext4_journalled_write_end, | |
617ba13b MC |
1807 | .set_page_dirty = ext4_journalled_set_page_dirty, |
1808 | .bmap = ext4_bmap, | |
1809 | .invalidatepage = ext4_invalidatepage, | |
1810 | .releasepage = ext4_releasepage, | |
ac27a0ec DK |
1811 | }; |
1812 | ||
617ba13b | 1813 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 1814 | { |
617ba13b MC |
1815 | if (ext4_should_order_data(inode)) |
1816 | inode->i_mapping->a_ops = &ext4_ordered_aops; | |
1817 | else if (ext4_should_writeback_data(inode)) | |
1818 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 1819 | else |
617ba13b | 1820 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
1821 | } |
1822 | ||
1823 | /* | |
617ba13b | 1824 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
1825 | * up to the end of the block which corresponds to `from'. |
1826 | * This required during truncate. We need to physically zero the tail end | |
1827 | * of that block so it doesn't yield old data if the file is later grown. | |
1828 | */ | |
a86c6181 | 1829 | int ext4_block_truncate_page(handle_t *handle, struct page *page, |
ac27a0ec DK |
1830 | struct address_space *mapping, loff_t from) |
1831 | { | |
617ba13b | 1832 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 1833 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
725d26d3 AK |
1834 | unsigned blocksize, length, pos; |
1835 | ext4_lblk_t iblock; | |
ac27a0ec DK |
1836 | struct inode *inode = mapping->host; |
1837 | struct buffer_head *bh; | |
1838 | int err = 0; | |
ac27a0ec DK |
1839 | |
1840 | blocksize = inode->i_sb->s_blocksize; | |
1841 | length = blocksize - (offset & (blocksize - 1)); | |
1842 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
1843 | ||
1844 | /* | |
1845 | * For "nobh" option, we can only work if we don't need to | |
1846 | * read-in the page - otherwise we create buffers to do the IO. | |
1847 | */ | |
1848 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && | |
617ba13b | 1849 | ext4_should_writeback_data(inode) && PageUptodate(page)) { |
fc0e15a6 | 1850 | zero_user_page(page, offset, length, KM_USER0); |
ac27a0ec DK |
1851 | set_page_dirty(page); |
1852 | goto unlock; | |
1853 | } | |
1854 | ||
1855 | if (!page_has_buffers(page)) | |
1856 | create_empty_buffers(page, blocksize, 0); | |
1857 | ||
1858 | /* Find the buffer that contains "offset" */ | |
1859 | bh = page_buffers(page); | |
1860 | pos = blocksize; | |
1861 | while (offset >= pos) { | |
1862 | bh = bh->b_this_page; | |
1863 | iblock++; | |
1864 | pos += blocksize; | |
1865 | } | |
1866 | ||
1867 | err = 0; | |
1868 | if (buffer_freed(bh)) { | |
1869 | BUFFER_TRACE(bh, "freed: skip"); | |
1870 | goto unlock; | |
1871 | } | |
1872 | ||
1873 | if (!buffer_mapped(bh)) { | |
1874 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 1875 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
1876 | /* unmapped? It's a hole - nothing to do */ |
1877 | if (!buffer_mapped(bh)) { | |
1878 | BUFFER_TRACE(bh, "still unmapped"); | |
1879 | goto unlock; | |
1880 | } | |
1881 | } | |
1882 | ||
1883 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
1884 | if (PageUptodate(page)) | |
1885 | set_buffer_uptodate(bh); | |
1886 | ||
1887 | if (!buffer_uptodate(bh)) { | |
1888 | err = -EIO; | |
1889 | ll_rw_block(READ, 1, &bh); | |
1890 | wait_on_buffer(bh); | |
1891 | /* Uhhuh. Read error. Complain and punt. */ | |
1892 | if (!buffer_uptodate(bh)) | |
1893 | goto unlock; | |
1894 | } | |
1895 | ||
617ba13b | 1896 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 1897 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 1898 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1899 | if (err) |
1900 | goto unlock; | |
1901 | } | |
1902 | ||
fc0e15a6 | 1903 | zero_user_page(page, offset, length, KM_USER0); |
ac27a0ec DK |
1904 | |
1905 | BUFFER_TRACE(bh, "zeroed end of block"); | |
1906 | ||
1907 | err = 0; | |
617ba13b MC |
1908 | if (ext4_should_journal_data(inode)) { |
1909 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec | 1910 | } else { |
617ba13b MC |
1911 | if (ext4_should_order_data(inode)) |
1912 | err = ext4_journal_dirty_data(handle, bh); | |
ac27a0ec DK |
1913 | mark_buffer_dirty(bh); |
1914 | } | |
1915 | ||
1916 | unlock: | |
1917 | unlock_page(page); | |
1918 | page_cache_release(page); | |
1919 | return err; | |
1920 | } | |
1921 | ||
1922 | /* | |
1923 | * Probably it should be a library function... search for first non-zero word | |
1924 | * or memcmp with zero_page, whatever is better for particular architecture. | |
1925 | * Linus? | |
1926 | */ | |
1927 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
1928 | { | |
1929 | while (p < q) | |
1930 | if (*p++) | |
1931 | return 0; | |
1932 | return 1; | |
1933 | } | |
1934 | ||
1935 | /** | |
617ba13b | 1936 | * ext4_find_shared - find the indirect blocks for partial truncation. |
ac27a0ec DK |
1937 | * @inode: inode in question |
1938 | * @depth: depth of the affected branch | |
617ba13b | 1939 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
ac27a0ec DK |
1940 | * @chain: place to store the pointers to partial indirect blocks |
1941 | * @top: place to the (detached) top of branch | |
1942 | * | |
617ba13b | 1943 | * This is a helper function used by ext4_truncate(). |
ac27a0ec DK |
1944 | * |
1945 | * When we do truncate() we may have to clean the ends of several | |
1946 | * indirect blocks but leave the blocks themselves alive. Block is | |
1947 | * partially truncated if some data below the new i_size is refered | |
1948 | * from it (and it is on the path to the first completely truncated | |
1949 | * data block, indeed). We have to free the top of that path along | |
1950 | * with everything to the right of the path. Since no allocation | |
617ba13b | 1951 | * past the truncation point is possible until ext4_truncate() |
ac27a0ec DK |
1952 | * finishes, we may safely do the latter, but top of branch may |
1953 | * require special attention - pageout below the truncation point | |
1954 | * might try to populate it. | |
1955 | * | |
1956 | * We atomically detach the top of branch from the tree, store the | |
1957 | * block number of its root in *@top, pointers to buffer_heads of | |
1958 | * partially truncated blocks - in @chain[].bh and pointers to | |
1959 | * their last elements that should not be removed - in | |
1960 | * @chain[].p. Return value is the pointer to last filled element | |
1961 | * of @chain. | |
1962 | * | |
1963 | * The work left to caller to do the actual freeing of subtrees: | |
1964 | * a) free the subtree starting from *@top | |
1965 | * b) free the subtrees whose roots are stored in | |
1966 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
1967 | * c) free the subtrees growing from the inode past the @chain[0]. | |
1968 | * (no partially truncated stuff there). */ | |
1969 | ||
617ba13b | 1970 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
725d26d3 | 1971 | ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top) |
ac27a0ec DK |
1972 | { |
1973 | Indirect *partial, *p; | |
1974 | int k, err; | |
1975 | ||
1976 | *top = 0; | |
1977 | /* Make k index the deepest non-null offest + 1 */ | |
1978 | for (k = depth; k > 1 && !offsets[k-1]; k--) | |
1979 | ; | |
617ba13b | 1980 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
ac27a0ec DK |
1981 | /* Writer: pointers */ |
1982 | if (!partial) | |
1983 | partial = chain + k-1; | |
1984 | /* | |
1985 | * If the branch acquired continuation since we've looked at it - | |
1986 | * fine, it should all survive and (new) top doesn't belong to us. | |
1987 | */ | |
1988 | if (!partial->key && *partial->p) | |
1989 | /* Writer: end */ | |
1990 | goto no_top; | |
1991 | for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) | |
1992 | ; | |
1993 | /* | |
1994 | * OK, we've found the last block that must survive. The rest of our | |
1995 | * branch should be detached before unlocking. However, if that rest | |
1996 | * of branch is all ours and does not grow immediately from the inode | |
1997 | * it's easier to cheat and just decrement partial->p. | |
1998 | */ | |
1999 | if (p == chain + k - 1 && p > chain) { | |
2000 | p->p--; | |
2001 | } else { | |
2002 | *top = *p->p; | |
617ba13b | 2003 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
ac27a0ec DK |
2004 | #if 0 |
2005 | *p->p = 0; | |
2006 | #endif | |
2007 | } | |
2008 | /* Writer: end */ | |
2009 | ||
2010 | while(partial > p) { | |
2011 | brelse(partial->bh); | |
2012 | partial--; | |
2013 | } | |
2014 | no_top: | |
2015 | return partial; | |
2016 | } | |
2017 | ||
2018 | /* | |
2019 | * Zero a number of block pointers in either an inode or an indirect block. | |
2020 | * If we restart the transaction we must again get write access to the | |
2021 | * indirect block for further modification. | |
2022 | * | |
2023 | * We release `count' blocks on disk, but (last - first) may be greater | |
2024 | * than `count' because there can be holes in there. | |
2025 | */ | |
617ba13b MC |
2026 | static void ext4_clear_blocks(handle_t *handle, struct inode *inode, |
2027 | struct buffer_head *bh, ext4_fsblk_t block_to_free, | |
ac27a0ec DK |
2028 | unsigned long count, __le32 *first, __le32 *last) |
2029 | { | |
2030 | __le32 *p; | |
2031 | if (try_to_extend_transaction(handle, inode)) { | |
2032 | if (bh) { | |
617ba13b MC |
2033 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
2034 | ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec | 2035 | } |
617ba13b MC |
2036 | ext4_mark_inode_dirty(handle, inode); |
2037 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
2038 | if (bh) { |
2039 | BUFFER_TRACE(bh, "retaking write access"); | |
617ba13b | 2040 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
2041 | } |
2042 | } | |
2043 | ||
2044 | /* | |
2045 | * Any buffers which are on the journal will be in memory. We find | |
dab291af | 2046 | * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget() |
ac27a0ec | 2047 | * on them. We've already detached each block from the file, so |
dab291af | 2048 | * bforget() in jbd2_journal_forget() should be safe. |
ac27a0ec | 2049 | * |
dab291af | 2050 | * AKPM: turn on bforget in jbd2_journal_forget()!!! |
ac27a0ec DK |
2051 | */ |
2052 | for (p = first; p < last; p++) { | |
2053 | u32 nr = le32_to_cpu(*p); | |
2054 | if (nr) { | |
1d03ec98 | 2055 | struct buffer_head *tbh; |
ac27a0ec DK |
2056 | |
2057 | *p = 0; | |
1d03ec98 AK |
2058 | tbh = sb_find_get_block(inode->i_sb, nr); |
2059 | ext4_forget(handle, 0, inode, tbh, nr); | |
ac27a0ec DK |
2060 | } |
2061 | } | |
2062 | ||
617ba13b | 2063 | ext4_free_blocks(handle, inode, block_to_free, count); |
ac27a0ec DK |
2064 | } |
2065 | ||
2066 | /** | |
617ba13b | 2067 | * ext4_free_data - free a list of data blocks |
ac27a0ec DK |
2068 | * @handle: handle for this transaction |
2069 | * @inode: inode we are dealing with | |
2070 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
2071 | * @first: array of block numbers | |
2072 | * @last: points immediately past the end of array | |
2073 | * | |
2074 | * We are freeing all blocks refered from that array (numbers are stored as | |
2075 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
2076 | * | |
2077 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
2078 | * blocks are contiguous then releasing them at one time will only affect one | |
2079 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
2080 | * actually use a lot of journal space. | |
2081 | * | |
2082 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
2083 | * block pointers. | |
2084 | */ | |
617ba13b | 2085 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
2086 | struct buffer_head *this_bh, |
2087 | __le32 *first, __le32 *last) | |
2088 | { | |
617ba13b | 2089 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ac27a0ec DK |
2090 | unsigned long count = 0; /* Number of blocks in the run */ |
2091 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
2092 | corresponding to | |
2093 | block_to_free */ | |
617ba13b | 2094 | ext4_fsblk_t nr; /* Current block # */ |
ac27a0ec DK |
2095 | __le32 *p; /* Pointer into inode/ind |
2096 | for current block */ | |
2097 | int err; | |
2098 | ||
2099 | if (this_bh) { /* For indirect block */ | |
2100 | BUFFER_TRACE(this_bh, "get_write_access"); | |
617ba13b | 2101 | err = ext4_journal_get_write_access(handle, this_bh); |
ac27a0ec DK |
2102 | /* Important: if we can't update the indirect pointers |
2103 | * to the blocks, we can't free them. */ | |
2104 | if (err) | |
2105 | return; | |
2106 | } | |
2107 | ||
2108 | for (p = first; p < last; p++) { | |
2109 | nr = le32_to_cpu(*p); | |
2110 | if (nr) { | |
2111 | /* accumulate blocks to free if they're contiguous */ | |
2112 | if (count == 0) { | |
2113 | block_to_free = nr; | |
2114 | block_to_free_p = p; | |
2115 | count = 1; | |
2116 | } else if (nr == block_to_free + count) { | |
2117 | count++; | |
2118 | } else { | |
617ba13b | 2119 | ext4_clear_blocks(handle, inode, this_bh, |
ac27a0ec DK |
2120 | block_to_free, |
2121 | count, block_to_free_p, p); | |
2122 | block_to_free = nr; | |
2123 | block_to_free_p = p; | |
2124 | count = 1; | |
2125 | } | |
2126 | } | |
2127 | } | |
2128 | ||
2129 | if (count > 0) | |
617ba13b | 2130 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
ac27a0ec DK |
2131 | count, block_to_free_p, p); |
2132 | ||
2133 | if (this_bh) { | |
617ba13b MC |
2134 | BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata"); |
2135 | ext4_journal_dirty_metadata(handle, this_bh); | |
ac27a0ec DK |
2136 | } |
2137 | } | |
2138 | ||
2139 | /** | |
617ba13b | 2140 | * ext4_free_branches - free an array of branches |
ac27a0ec DK |
2141 | * @handle: JBD handle for this transaction |
2142 | * @inode: inode we are dealing with | |
2143 | * @parent_bh: the buffer_head which contains *@first and *@last | |
2144 | * @first: array of block numbers | |
2145 | * @last: pointer immediately past the end of array | |
2146 | * @depth: depth of the branches to free | |
2147 | * | |
2148 | * We are freeing all blocks refered from these branches (numbers are | |
2149 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
2150 | * appropriately. | |
2151 | */ | |
617ba13b | 2152 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
2153 | struct buffer_head *parent_bh, |
2154 | __le32 *first, __le32 *last, int depth) | |
2155 | { | |
617ba13b | 2156 | ext4_fsblk_t nr; |
ac27a0ec DK |
2157 | __le32 *p; |
2158 | ||
2159 | if (is_handle_aborted(handle)) | |
2160 | return; | |
2161 | ||
2162 | if (depth--) { | |
2163 | struct buffer_head *bh; | |
617ba13b | 2164 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
2165 | p = last; |
2166 | while (--p >= first) { | |
2167 | nr = le32_to_cpu(*p); | |
2168 | if (!nr) | |
2169 | continue; /* A hole */ | |
2170 | ||
2171 | /* Go read the buffer for the next level down */ | |
2172 | bh = sb_bread(inode->i_sb, nr); | |
2173 | ||
2174 | /* | |
2175 | * A read failure? Report error and clear slot | |
2176 | * (should be rare). | |
2177 | */ | |
2178 | if (!bh) { | |
617ba13b | 2179 | ext4_error(inode->i_sb, "ext4_free_branches", |
2ae02107 | 2180 | "Read failure, inode=%lu, block=%llu", |
ac27a0ec DK |
2181 | inode->i_ino, nr); |
2182 | continue; | |
2183 | } | |
2184 | ||
2185 | /* This zaps the entire block. Bottom up. */ | |
2186 | BUFFER_TRACE(bh, "free child branches"); | |
617ba13b | 2187 | ext4_free_branches(handle, inode, bh, |
ac27a0ec DK |
2188 | (__le32*)bh->b_data, |
2189 | (__le32*)bh->b_data + addr_per_block, | |
2190 | depth); | |
2191 | ||
2192 | /* | |
2193 | * We've probably journalled the indirect block several | |
2194 | * times during the truncate. But it's no longer | |
2195 | * needed and we now drop it from the transaction via | |
dab291af | 2196 | * jbd2_journal_revoke(). |
ac27a0ec DK |
2197 | * |
2198 | * That's easy if it's exclusively part of this | |
2199 | * transaction. But if it's part of the committing | |
dab291af | 2200 | * transaction then jbd2_journal_forget() will simply |
ac27a0ec | 2201 | * brelse() it. That means that if the underlying |
617ba13b | 2202 | * block is reallocated in ext4_get_block(), |
ac27a0ec DK |
2203 | * unmap_underlying_metadata() will find this block |
2204 | * and will try to get rid of it. damn, damn. | |
2205 | * | |
2206 | * If this block has already been committed to the | |
2207 | * journal, a revoke record will be written. And | |
2208 | * revoke records must be emitted *before* clearing | |
2209 | * this block's bit in the bitmaps. | |
2210 | */ | |
617ba13b | 2211 | ext4_forget(handle, 1, inode, bh, bh->b_blocknr); |
ac27a0ec DK |
2212 | |
2213 | /* | |
2214 | * Everything below this this pointer has been | |
2215 | * released. Now let this top-of-subtree go. | |
2216 | * | |
2217 | * We want the freeing of this indirect block to be | |
2218 | * atomic in the journal with the updating of the | |
2219 | * bitmap block which owns it. So make some room in | |
2220 | * the journal. | |
2221 | * | |
2222 | * We zero the parent pointer *after* freeing its | |
2223 | * pointee in the bitmaps, so if extend_transaction() | |
2224 | * for some reason fails to put the bitmap changes and | |
2225 | * the release into the same transaction, recovery | |
2226 | * will merely complain about releasing a free block, | |
2227 | * rather than leaking blocks. | |
2228 | */ | |
2229 | if (is_handle_aborted(handle)) | |
2230 | return; | |
2231 | if (try_to_extend_transaction(handle, inode)) { | |
617ba13b MC |
2232 | ext4_mark_inode_dirty(handle, inode); |
2233 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
2234 | } |
2235 | ||
617ba13b | 2236 | ext4_free_blocks(handle, inode, nr, 1); |
ac27a0ec DK |
2237 | |
2238 | if (parent_bh) { | |
2239 | /* | |
2240 | * The block which we have just freed is | |
2241 | * pointed to by an indirect block: journal it | |
2242 | */ | |
2243 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
617ba13b | 2244 | if (!ext4_journal_get_write_access(handle, |
ac27a0ec DK |
2245 | parent_bh)){ |
2246 | *p = 0; | |
2247 | BUFFER_TRACE(parent_bh, | |
617ba13b MC |
2248 | "call ext4_journal_dirty_metadata"); |
2249 | ext4_journal_dirty_metadata(handle, | |
ac27a0ec DK |
2250 | parent_bh); |
2251 | } | |
2252 | } | |
2253 | } | |
2254 | } else { | |
2255 | /* We have reached the bottom of the tree. */ | |
2256 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
617ba13b | 2257 | ext4_free_data(handle, inode, parent_bh, first, last); |
ac27a0ec DK |
2258 | } |
2259 | } | |
2260 | ||
2261 | /* | |
617ba13b | 2262 | * ext4_truncate() |
ac27a0ec | 2263 | * |
617ba13b MC |
2264 | * We block out ext4_get_block() block instantiations across the entire |
2265 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
2266 | * simultaneously on behalf of the same inode. |
2267 | * | |
2268 | * As we work through the truncate and commmit bits of it to the journal there | |
2269 | * is one core, guiding principle: the file's tree must always be consistent on | |
2270 | * disk. We must be able to restart the truncate after a crash. | |
2271 | * | |
2272 | * The file's tree may be transiently inconsistent in memory (although it | |
2273 | * probably isn't), but whenever we close off and commit a journal transaction, | |
2274 | * the contents of (the filesystem + the journal) must be consistent and | |
2275 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
2276 | * left-to-right works OK too). | |
2277 | * | |
2278 | * Note that at recovery time, journal replay occurs *before* the restart of | |
2279 | * truncate against the orphan inode list. | |
2280 | * | |
2281 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 2282 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 2283 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
2284 | * ext4_truncate() to have another go. So there will be instantiated blocks |
2285 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 2286 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 2287 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 2288 | */ |
617ba13b | 2289 | void ext4_truncate(struct inode *inode) |
ac27a0ec DK |
2290 | { |
2291 | handle_t *handle; | |
617ba13b | 2292 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 2293 | __le32 *i_data = ei->i_data; |
617ba13b | 2294 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec | 2295 | struct address_space *mapping = inode->i_mapping; |
725d26d3 | 2296 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
2297 | Indirect chain[4]; |
2298 | Indirect *partial; | |
2299 | __le32 nr = 0; | |
2300 | int n; | |
725d26d3 | 2301 | ext4_lblk_t last_block; |
ac27a0ec DK |
2302 | unsigned blocksize = inode->i_sb->s_blocksize; |
2303 | struct page *page; | |
2304 | ||
2305 | if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
2306 | S_ISLNK(inode->i_mode))) | |
2307 | return; | |
617ba13b | 2308 | if (ext4_inode_is_fast_symlink(inode)) |
ac27a0ec DK |
2309 | return; |
2310 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
2311 | return; | |
2312 | ||
2313 | /* | |
2314 | * We have to lock the EOF page here, because lock_page() nests | |
dab291af | 2315 | * outside jbd2_journal_start(). |
ac27a0ec DK |
2316 | */ |
2317 | if ((inode->i_size & (blocksize - 1)) == 0) { | |
2318 | /* Block boundary? Nothing to do */ | |
2319 | page = NULL; | |
2320 | } else { | |
2321 | page = grab_cache_page(mapping, | |
2322 | inode->i_size >> PAGE_CACHE_SHIFT); | |
2323 | if (!page) | |
2324 | return; | |
2325 | } | |
2326 | ||
1d03ec98 AK |
2327 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
2328 | ext4_ext_truncate(inode, page); | |
2329 | return; | |
2330 | } | |
a86c6181 | 2331 | |
ac27a0ec DK |
2332 | handle = start_transaction(inode); |
2333 | if (IS_ERR(handle)) { | |
2334 | if (page) { | |
2335 | clear_highpage(page); | |
2336 | flush_dcache_page(page); | |
2337 | unlock_page(page); | |
2338 | page_cache_release(page); | |
2339 | } | |
2340 | return; /* AKPM: return what? */ | |
2341 | } | |
2342 | ||
2343 | last_block = (inode->i_size + blocksize-1) | |
617ba13b | 2344 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
ac27a0ec DK |
2345 | |
2346 | if (page) | |
617ba13b | 2347 | ext4_block_truncate_page(handle, page, mapping, inode->i_size); |
ac27a0ec | 2348 | |
617ba13b | 2349 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
ac27a0ec DK |
2350 | if (n == 0) |
2351 | goto out_stop; /* error */ | |
2352 | ||
2353 | /* | |
2354 | * OK. This truncate is going to happen. We add the inode to the | |
2355 | * orphan list, so that if this truncate spans multiple transactions, | |
2356 | * and we crash, we will resume the truncate when the filesystem | |
2357 | * recovers. It also marks the inode dirty, to catch the new size. | |
2358 | * | |
2359 | * Implication: the file must always be in a sane, consistent | |
2360 | * truncatable state while each transaction commits. | |
2361 | */ | |
617ba13b | 2362 | if (ext4_orphan_add(handle, inode)) |
ac27a0ec DK |
2363 | goto out_stop; |
2364 | ||
2365 | /* | |
2366 | * The orphan list entry will now protect us from any crash which | |
2367 | * occurs before the truncate completes, so it is now safe to propagate | |
2368 | * the new, shorter inode size (held for now in i_size) into the | |
2369 | * on-disk inode. We do this via i_disksize, which is the value which | |
617ba13b | 2370 | * ext4 *really* writes onto the disk inode. |
ac27a0ec DK |
2371 | */ |
2372 | ei->i_disksize = inode->i_size; | |
2373 | ||
2374 | /* | |
617ba13b | 2375 | * From here we block out all ext4_get_block() callers who want to |
ac27a0ec DK |
2376 | * modify the block allocation tree. |
2377 | */ | |
2378 | mutex_lock(&ei->truncate_mutex); | |
2379 | ||
2380 | if (n == 1) { /* direct blocks */ | |
617ba13b MC |
2381 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
2382 | i_data + EXT4_NDIR_BLOCKS); | |
ac27a0ec DK |
2383 | goto do_indirects; |
2384 | } | |
2385 | ||
617ba13b | 2386 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
ac27a0ec DK |
2387 | /* Kill the top of shared branch (not detached) */ |
2388 | if (nr) { | |
2389 | if (partial == chain) { | |
2390 | /* Shared branch grows from the inode */ | |
617ba13b | 2391 | ext4_free_branches(handle, inode, NULL, |
ac27a0ec DK |
2392 | &nr, &nr+1, (chain+n-1) - partial); |
2393 | *partial->p = 0; | |
2394 | /* | |
2395 | * We mark the inode dirty prior to restart, | |
2396 | * and prior to stop. No need for it here. | |
2397 | */ | |
2398 | } else { | |
2399 | /* Shared branch grows from an indirect block */ | |
2400 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
617ba13b | 2401 | ext4_free_branches(handle, inode, partial->bh, |
ac27a0ec DK |
2402 | partial->p, |
2403 | partial->p+1, (chain+n-1) - partial); | |
2404 | } | |
2405 | } | |
2406 | /* Clear the ends of indirect blocks on the shared branch */ | |
2407 | while (partial > chain) { | |
617ba13b | 2408 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
ac27a0ec DK |
2409 | (__le32*)partial->bh->b_data+addr_per_block, |
2410 | (chain+n-1) - partial); | |
2411 | BUFFER_TRACE(partial->bh, "call brelse"); | |
2412 | brelse (partial->bh); | |
2413 | partial--; | |
2414 | } | |
2415 | do_indirects: | |
2416 | /* Kill the remaining (whole) subtrees */ | |
2417 | switch (offsets[0]) { | |
2418 | default: | |
617ba13b | 2419 | nr = i_data[EXT4_IND_BLOCK]; |
ac27a0ec | 2420 | if (nr) { |
617ba13b MC |
2421 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
2422 | i_data[EXT4_IND_BLOCK] = 0; | |
ac27a0ec | 2423 | } |
617ba13b MC |
2424 | case EXT4_IND_BLOCK: |
2425 | nr = i_data[EXT4_DIND_BLOCK]; | |
ac27a0ec | 2426 | if (nr) { |
617ba13b MC |
2427 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
2428 | i_data[EXT4_DIND_BLOCK] = 0; | |
ac27a0ec | 2429 | } |
617ba13b MC |
2430 | case EXT4_DIND_BLOCK: |
2431 | nr = i_data[EXT4_TIND_BLOCK]; | |
ac27a0ec | 2432 | if (nr) { |
617ba13b MC |
2433 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
2434 | i_data[EXT4_TIND_BLOCK] = 0; | |
ac27a0ec | 2435 | } |
617ba13b | 2436 | case EXT4_TIND_BLOCK: |
ac27a0ec DK |
2437 | ; |
2438 | } | |
2439 | ||
617ba13b | 2440 | ext4_discard_reservation(inode); |
ac27a0ec DK |
2441 | |
2442 | mutex_unlock(&ei->truncate_mutex); | |
ef7f3835 | 2443 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
617ba13b | 2444 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
2445 | |
2446 | /* | |
2447 | * In a multi-transaction truncate, we only make the final transaction | |
2448 | * synchronous | |
2449 | */ | |
2450 | if (IS_SYNC(inode)) | |
2451 | handle->h_sync = 1; | |
2452 | out_stop: | |
2453 | /* | |
2454 | * If this was a simple ftruncate(), and the file will remain alive | |
2455 | * then we need to clear up the orphan record which we created above. | |
2456 | * However, if this was a real unlink then we were called by | |
617ba13b | 2457 | * ext4_delete_inode(), and we allow that function to clean up the |
ac27a0ec DK |
2458 | * orphan info for us. |
2459 | */ | |
2460 | if (inode->i_nlink) | |
617ba13b | 2461 | ext4_orphan_del(handle, inode); |
ac27a0ec | 2462 | |
617ba13b | 2463 | ext4_journal_stop(handle); |
ac27a0ec DK |
2464 | } |
2465 | ||
617ba13b MC |
2466 | static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb, |
2467 | unsigned long ino, struct ext4_iloc *iloc) | |
ac27a0ec | 2468 | { |
fd2d4291 AM |
2469 | unsigned long desc, group_desc; |
2470 | ext4_group_t block_group; | |
ac27a0ec | 2471 | unsigned long offset; |
617ba13b | 2472 | ext4_fsblk_t block; |
ac27a0ec | 2473 | struct buffer_head *bh; |
617ba13b | 2474 | struct ext4_group_desc * gdp; |
ac27a0ec | 2475 | |
617ba13b | 2476 | if (!ext4_valid_inum(sb, ino)) { |
ac27a0ec DK |
2477 | /* |
2478 | * This error is already checked for in namei.c unless we are | |
2479 | * looking at an NFS filehandle, in which case no error | |
2480 | * report is needed | |
2481 | */ | |
2482 | return 0; | |
2483 | } | |
2484 | ||
617ba13b MC |
2485 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
2486 | if (block_group >= EXT4_SB(sb)->s_groups_count) { | |
2487 | ext4_error(sb,"ext4_get_inode_block","group >= groups count"); | |
ac27a0ec DK |
2488 | return 0; |
2489 | } | |
2490 | smp_rmb(); | |
617ba13b MC |
2491 | group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); |
2492 | desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); | |
2493 | bh = EXT4_SB(sb)->s_group_desc[group_desc]; | |
ac27a0ec | 2494 | if (!bh) { |
617ba13b | 2495 | ext4_error (sb, "ext4_get_inode_block", |
ac27a0ec DK |
2496 | "Descriptor not loaded"); |
2497 | return 0; | |
2498 | } | |
2499 | ||
0d1ee42f AR |
2500 | gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data + |
2501 | desc * EXT4_DESC_SIZE(sb)); | |
ac27a0ec DK |
2502 | /* |
2503 | * Figure out the offset within the block group inode table | |
2504 | */ | |
617ba13b MC |
2505 | offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) * |
2506 | EXT4_INODE_SIZE(sb); | |
8fadc143 AR |
2507 | block = ext4_inode_table(sb, gdp) + |
2508 | (offset >> EXT4_BLOCK_SIZE_BITS(sb)); | |
ac27a0ec DK |
2509 | |
2510 | iloc->block_group = block_group; | |
617ba13b | 2511 | iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1); |
ac27a0ec DK |
2512 | return block; |
2513 | } | |
2514 | ||
2515 | /* | |
617ba13b | 2516 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
2517 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
2518 | * data in memory that is needed to recreate the on-disk version of this | |
2519 | * inode. | |
2520 | */ | |
617ba13b MC |
2521 | static int __ext4_get_inode_loc(struct inode *inode, |
2522 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 2523 | { |
617ba13b | 2524 | ext4_fsblk_t block; |
ac27a0ec DK |
2525 | struct buffer_head *bh; |
2526 | ||
617ba13b | 2527 | block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc); |
ac27a0ec DK |
2528 | if (!block) |
2529 | return -EIO; | |
2530 | ||
2531 | bh = sb_getblk(inode->i_sb, block); | |
2532 | if (!bh) { | |
617ba13b | 2533 | ext4_error (inode->i_sb, "ext4_get_inode_loc", |
ac27a0ec | 2534 | "unable to read inode block - " |
2ae02107 | 2535 | "inode=%lu, block=%llu", |
ac27a0ec DK |
2536 | inode->i_ino, block); |
2537 | return -EIO; | |
2538 | } | |
2539 | if (!buffer_uptodate(bh)) { | |
2540 | lock_buffer(bh); | |
2541 | if (buffer_uptodate(bh)) { | |
2542 | /* someone brought it uptodate while we waited */ | |
2543 | unlock_buffer(bh); | |
2544 | goto has_buffer; | |
2545 | } | |
2546 | ||
2547 | /* | |
2548 | * If we have all information of the inode in memory and this | |
2549 | * is the only valid inode in the block, we need not read the | |
2550 | * block. | |
2551 | */ | |
2552 | if (in_mem) { | |
2553 | struct buffer_head *bitmap_bh; | |
617ba13b | 2554 | struct ext4_group_desc *desc; |
ac27a0ec DK |
2555 | int inodes_per_buffer; |
2556 | int inode_offset, i; | |
fd2d4291 | 2557 | ext4_group_t block_group; |
ac27a0ec DK |
2558 | int start; |
2559 | ||
2560 | block_group = (inode->i_ino - 1) / | |
617ba13b | 2561 | EXT4_INODES_PER_GROUP(inode->i_sb); |
ac27a0ec | 2562 | inodes_per_buffer = bh->b_size / |
617ba13b | 2563 | EXT4_INODE_SIZE(inode->i_sb); |
ac27a0ec | 2564 | inode_offset = ((inode->i_ino - 1) % |
617ba13b | 2565 | EXT4_INODES_PER_GROUP(inode->i_sb)); |
ac27a0ec DK |
2566 | start = inode_offset & ~(inodes_per_buffer - 1); |
2567 | ||
2568 | /* Is the inode bitmap in cache? */ | |
617ba13b | 2569 | desc = ext4_get_group_desc(inode->i_sb, |
ac27a0ec DK |
2570 | block_group, NULL); |
2571 | if (!desc) | |
2572 | goto make_io; | |
2573 | ||
2574 | bitmap_bh = sb_getblk(inode->i_sb, | |
8fadc143 | 2575 | ext4_inode_bitmap(inode->i_sb, desc)); |
ac27a0ec DK |
2576 | if (!bitmap_bh) |
2577 | goto make_io; | |
2578 | ||
2579 | /* | |
2580 | * If the inode bitmap isn't in cache then the | |
2581 | * optimisation may end up performing two reads instead | |
2582 | * of one, so skip it. | |
2583 | */ | |
2584 | if (!buffer_uptodate(bitmap_bh)) { | |
2585 | brelse(bitmap_bh); | |
2586 | goto make_io; | |
2587 | } | |
2588 | for (i = start; i < start + inodes_per_buffer; i++) { | |
2589 | if (i == inode_offset) | |
2590 | continue; | |
617ba13b | 2591 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
2592 | break; |
2593 | } | |
2594 | brelse(bitmap_bh); | |
2595 | if (i == start + inodes_per_buffer) { | |
2596 | /* all other inodes are free, so skip I/O */ | |
2597 | memset(bh->b_data, 0, bh->b_size); | |
2598 | set_buffer_uptodate(bh); | |
2599 | unlock_buffer(bh); | |
2600 | goto has_buffer; | |
2601 | } | |
2602 | } | |
2603 | ||
2604 | make_io: | |
2605 | /* | |
2606 | * There are other valid inodes in the buffer, this inode | |
2607 | * has in-inode xattrs, or we don't have this inode in memory. | |
2608 | * Read the block from disk. | |
2609 | */ | |
2610 | get_bh(bh); | |
2611 | bh->b_end_io = end_buffer_read_sync; | |
2612 | submit_bh(READ_META, bh); | |
2613 | wait_on_buffer(bh); | |
2614 | if (!buffer_uptodate(bh)) { | |
617ba13b | 2615 | ext4_error(inode->i_sb, "ext4_get_inode_loc", |
ac27a0ec | 2616 | "unable to read inode block - " |
2ae02107 | 2617 | "inode=%lu, block=%llu", |
ac27a0ec DK |
2618 | inode->i_ino, block); |
2619 | brelse(bh); | |
2620 | return -EIO; | |
2621 | } | |
2622 | } | |
2623 | has_buffer: | |
2624 | iloc->bh = bh; | |
2625 | return 0; | |
2626 | } | |
2627 | ||
617ba13b | 2628 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
2629 | { |
2630 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b MC |
2631 | return __ext4_get_inode_loc(inode, iloc, |
2632 | !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR)); | |
ac27a0ec DK |
2633 | } |
2634 | ||
617ba13b | 2635 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 2636 | { |
617ba13b | 2637 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
2638 | |
2639 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 2640 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 2641 | inode->i_flags |= S_SYNC; |
617ba13b | 2642 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 2643 | inode->i_flags |= S_APPEND; |
617ba13b | 2644 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 2645 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 2646 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 2647 | inode->i_flags |= S_NOATIME; |
617ba13b | 2648 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
2649 | inode->i_flags |= S_DIRSYNC; |
2650 | } | |
2651 | ||
ff9ddf7e JK |
2652 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
2653 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
2654 | { | |
2655 | unsigned int flags = ei->vfs_inode.i_flags; | |
2656 | ||
2657 | ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
2658 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL); | |
2659 | if (flags & S_SYNC) | |
2660 | ei->i_flags |= EXT4_SYNC_FL; | |
2661 | if (flags & S_APPEND) | |
2662 | ei->i_flags |= EXT4_APPEND_FL; | |
2663 | if (flags & S_IMMUTABLE) | |
2664 | ei->i_flags |= EXT4_IMMUTABLE_FL; | |
2665 | if (flags & S_NOATIME) | |
2666 | ei->i_flags |= EXT4_NOATIME_FL; | |
2667 | if (flags & S_DIRSYNC) | |
2668 | ei->i_flags |= EXT4_DIRSYNC_FL; | |
2669 | } | |
2670 | ||
617ba13b | 2671 | void ext4_read_inode(struct inode * inode) |
ac27a0ec | 2672 | { |
617ba13b MC |
2673 | struct ext4_iloc iloc; |
2674 | struct ext4_inode *raw_inode; | |
2675 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
2676 | struct buffer_head *bh; |
2677 | int block; | |
2678 | ||
617ba13b MC |
2679 | #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL |
2680 | ei->i_acl = EXT4_ACL_NOT_CACHED; | |
2681 | ei->i_default_acl = EXT4_ACL_NOT_CACHED; | |
ac27a0ec DK |
2682 | #endif |
2683 | ei->i_block_alloc_info = NULL; | |
2684 | ||
617ba13b | 2685 | if (__ext4_get_inode_loc(inode, &iloc, 0)) |
ac27a0ec DK |
2686 | goto bad_inode; |
2687 | bh = iloc.bh; | |
617ba13b | 2688 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
2689 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
2690 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
2691 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
2692 | if(!(test_opt (inode->i_sb, NO_UID32))) { | |
2693 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; | |
2694 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
2695 | } | |
2696 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
2697 | inode->i_size = le32_to_cpu(raw_inode->i_size); | |
ac27a0ec DK |
2698 | |
2699 | ei->i_state = 0; | |
2700 | ei->i_dir_start_lookup = 0; | |
2701 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
2702 | /* We now have enough fields to check if the inode was active or not. | |
2703 | * This is needed because nfsd might try to access dead inodes | |
2704 | * the test is that same one that e2fsck uses | |
2705 | * NeilBrown 1999oct15 | |
2706 | */ | |
2707 | if (inode->i_nlink == 0) { | |
2708 | if (inode->i_mode == 0 || | |
617ba13b | 2709 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec DK |
2710 | /* this inode is deleted */ |
2711 | brelse (bh); | |
2712 | goto bad_inode; | |
2713 | } | |
2714 | /* The only unlinked inodes we let through here have | |
2715 | * valid i_mode and are being read by the orphan | |
2716 | * recovery code: that's fine, we're about to complete | |
2717 | * the process of deleting those. */ | |
2718 | } | |
2719 | inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); | |
2720 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); | |
ac27a0ec | 2721 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); |
9b8f1f01 MC |
2722 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
2723 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
2724 | ei->i_file_acl |= |
2725 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
ac27a0ec DK |
2726 | if (!S_ISREG(inode->i_mode)) { |
2727 | ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); | |
2728 | } else { | |
2729 | inode->i_size |= | |
2730 | ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; | |
2731 | } | |
2732 | ei->i_disksize = inode->i_size; | |
2733 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); | |
2734 | ei->i_block_group = iloc.block_group; | |
2735 | /* | |
2736 | * NOTE! The in-memory inode i_data array is in little-endian order | |
2737 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
2738 | */ | |
617ba13b | 2739 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
2740 | ei->i_data[block] = raw_inode->i_block[block]; |
2741 | INIT_LIST_HEAD(&ei->i_orphan); | |
2742 | ||
617ba13b MC |
2743 | if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 && |
2744 | EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
ac27a0ec DK |
2745 | /* |
2746 | * When mke2fs creates big inodes it does not zero out | |
617ba13b | 2747 | * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE, |
ac27a0ec DK |
2748 | * so ignore those first few inodes. |
2749 | */ | |
2750 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
617ba13b | 2751 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f KK |
2752 | EXT4_INODE_SIZE(inode->i_sb)) { |
2753 | brelse (bh); | |
ac27a0ec | 2754 | goto bad_inode; |
e5d2861f | 2755 | } |
ac27a0ec DK |
2756 | if (ei->i_extra_isize == 0) { |
2757 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
2758 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
2759 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
2760 | } else { |
2761 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 2762 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 2763 | ei->i_extra_isize; |
617ba13b MC |
2764 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
2765 | ei->i_state |= EXT4_STATE_XATTR; | |
ac27a0ec DK |
2766 | } |
2767 | } else | |
2768 | ei->i_extra_isize = 0; | |
2769 | ||
ef7f3835 KS |
2770 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
2771 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
2772 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
2773 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
2774 | ||
ac27a0ec | 2775 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
2776 | inode->i_op = &ext4_file_inode_operations; |
2777 | inode->i_fop = &ext4_file_operations; | |
2778 | ext4_set_aops(inode); | |
ac27a0ec | 2779 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
2780 | inode->i_op = &ext4_dir_inode_operations; |
2781 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 2782 | } else if (S_ISLNK(inode->i_mode)) { |
617ba13b MC |
2783 | if (ext4_inode_is_fast_symlink(inode)) |
2784 | inode->i_op = &ext4_fast_symlink_inode_operations; | |
ac27a0ec | 2785 | else { |
617ba13b MC |
2786 | inode->i_op = &ext4_symlink_inode_operations; |
2787 | ext4_set_aops(inode); | |
ac27a0ec DK |
2788 | } |
2789 | } else { | |
617ba13b | 2790 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
2791 | if (raw_inode->i_block[0]) |
2792 | init_special_inode(inode, inode->i_mode, | |
2793 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
2794 | else | |
2795 | init_special_inode(inode, inode->i_mode, | |
2796 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
2797 | } | |
2798 | brelse (iloc.bh); | |
617ba13b | 2799 | ext4_set_inode_flags(inode); |
ac27a0ec DK |
2800 | return; |
2801 | ||
2802 | bad_inode: | |
2803 | make_bad_inode(inode); | |
2804 | return; | |
2805 | } | |
2806 | ||
2807 | /* | |
2808 | * Post the struct inode info into an on-disk inode location in the | |
2809 | * buffer-cache. This gobbles the caller's reference to the | |
2810 | * buffer_head in the inode location struct. | |
2811 | * | |
2812 | * The caller must have write access to iloc->bh. | |
2813 | */ | |
617ba13b | 2814 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 2815 | struct inode *inode, |
617ba13b | 2816 | struct ext4_iloc *iloc) |
ac27a0ec | 2817 | { |
617ba13b MC |
2818 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
2819 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
2820 | struct buffer_head *bh = iloc->bh; |
2821 | int err = 0, rc, block; | |
2822 | ||
2823 | /* For fields not not tracking in the in-memory inode, | |
2824 | * initialise them to zero for new inodes. */ | |
617ba13b MC |
2825 | if (ei->i_state & EXT4_STATE_NEW) |
2826 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); | |
ac27a0ec | 2827 | |
ff9ddf7e | 2828 | ext4_get_inode_flags(ei); |
ac27a0ec DK |
2829 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
2830 | if(!(test_opt(inode->i_sb, NO_UID32))) { | |
2831 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); | |
2832 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
2833 | /* | |
2834 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
2835 | * re-used with the upper 16 bits of the uid/gid intact | |
2836 | */ | |
2837 | if(!ei->i_dtime) { | |
2838 | raw_inode->i_uid_high = | |
2839 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
2840 | raw_inode->i_gid_high = | |
2841 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
2842 | } else { | |
2843 | raw_inode->i_uid_high = 0; | |
2844 | raw_inode->i_gid_high = 0; | |
2845 | } | |
2846 | } else { | |
2847 | raw_inode->i_uid_low = | |
2848 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
2849 | raw_inode->i_gid_low = | |
2850 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
2851 | raw_inode->i_uid_high = 0; | |
2852 | raw_inode->i_gid_high = 0; | |
2853 | } | |
2854 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
2855 | raw_inode->i_size = cpu_to_le32(ei->i_disksize); | |
ef7f3835 KS |
2856 | |
2857 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
2858 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
2859 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
2860 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
2861 | ||
ac27a0ec DK |
2862 | raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); |
2863 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); | |
2864 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); | |
9b8f1f01 MC |
2865 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
2866 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
2867 | raw_inode->i_file_acl_high = |
2868 | cpu_to_le16(ei->i_file_acl >> 32); | |
ac27a0ec DK |
2869 | raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); |
2870 | if (!S_ISREG(inode->i_mode)) { | |
2871 | raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); | |
2872 | } else { | |
2873 | raw_inode->i_size_high = | |
2874 | cpu_to_le32(ei->i_disksize >> 32); | |
2875 | if (ei->i_disksize > 0x7fffffffULL) { | |
2876 | struct super_block *sb = inode->i_sb; | |
617ba13b MC |
2877 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
2878 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
2879 | EXT4_SB(sb)->s_es->s_rev_level == | |
2880 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
ac27a0ec DK |
2881 | /* If this is the first large file |
2882 | * created, add a flag to the superblock. | |
2883 | */ | |
617ba13b MC |
2884 | err = ext4_journal_get_write_access(handle, |
2885 | EXT4_SB(sb)->s_sbh); | |
ac27a0ec DK |
2886 | if (err) |
2887 | goto out_brelse; | |
617ba13b MC |
2888 | ext4_update_dynamic_rev(sb); |
2889 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
2890 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); | |
ac27a0ec DK |
2891 | sb->s_dirt = 1; |
2892 | handle->h_sync = 1; | |
617ba13b MC |
2893 | err = ext4_journal_dirty_metadata(handle, |
2894 | EXT4_SB(sb)->s_sbh); | |
ac27a0ec DK |
2895 | } |
2896 | } | |
2897 | } | |
2898 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
2899 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
2900 | if (old_valid_dev(inode->i_rdev)) { | |
2901 | raw_inode->i_block[0] = | |
2902 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
2903 | raw_inode->i_block[1] = 0; | |
2904 | } else { | |
2905 | raw_inode->i_block[0] = 0; | |
2906 | raw_inode->i_block[1] = | |
2907 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
2908 | raw_inode->i_block[2] = 0; | |
2909 | } | |
617ba13b | 2910 | } else for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
2911 | raw_inode->i_block[block] = ei->i_data[block]; |
2912 | ||
2913 | if (ei->i_extra_isize) | |
2914 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); | |
2915 | ||
617ba13b MC |
2916 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
2917 | rc = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
2918 | if (!err) |
2919 | err = rc; | |
617ba13b | 2920 | ei->i_state &= ~EXT4_STATE_NEW; |
ac27a0ec DK |
2921 | |
2922 | out_brelse: | |
2923 | brelse (bh); | |
617ba13b | 2924 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
2925 | return err; |
2926 | } | |
2927 | ||
2928 | /* | |
617ba13b | 2929 | * ext4_write_inode() |
ac27a0ec DK |
2930 | * |
2931 | * We are called from a few places: | |
2932 | * | |
2933 | * - Within generic_file_write() for O_SYNC files. | |
2934 | * Here, there will be no transaction running. We wait for any running | |
2935 | * trasnaction to commit. | |
2936 | * | |
2937 | * - Within sys_sync(), kupdate and such. | |
2938 | * We wait on commit, if tol to. | |
2939 | * | |
2940 | * - Within prune_icache() (PF_MEMALLOC == true) | |
2941 | * Here we simply return. We can't afford to block kswapd on the | |
2942 | * journal commit. | |
2943 | * | |
2944 | * In all cases it is actually safe for us to return without doing anything, | |
2945 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 2946 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
2947 | * knfsd. |
2948 | * | |
2949 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
2950 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
2951 | * which we are interested. | |
2952 | * | |
2953 | * It would be a bug for them to not do this. The code: | |
2954 | * | |
2955 | * mark_inode_dirty(inode) | |
2956 | * stuff(); | |
2957 | * inode->i_size = expr; | |
2958 | * | |
2959 | * is in error because a kswapd-driven write_inode() could occur while | |
2960 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
2961 | * will no longer be on the superblock's dirty inode list. | |
2962 | */ | |
617ba13b | 2963 | int ext4_write_inode(struct inode *inode, int wait) |
ac27a0ec DK |
2964 | { |
2965 | if (current->flags & PF_MEMALLOC) | |
2966 | return 0; | |
2967 | ||
617ba13b | 2968 | if (ext4_journal_current_handle()) { |
b38bd33a | 2969 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
ac27a0ec DK |
2970 | dump_stack(); |
2971 | return -EIO; | |
2972 | } | |
2973 | ||
2974 | if (!wait) | |
2975 | return 0; | |
2976 | ||
617ba13b | 2977 | return ext4_force_commit(inode->i_sb); |
ac27a0ec DK |
2978 | } |
2979 | ||
2980 | /* | |
617ba13b | 2981 | * ext4_setattr() |
ac27a0ec DK |
2982 | * |
2983 | * Called from notify_change. | |
2984 | * | |
2985 | * We want to trap VFS attempts to truncate the file as soon as | |
2986 | * possible. In particular, we want to make sure that when the VFS | |
2987 | * shrinks i_size, we put the inode on the orphan list and modify | |
2988 | * i_disksize immediately, so that during the subsequent flushing of | |
2989 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
2990 | * commit will leave the blocks being flushed in an unused state on | |
2991 | * disk. (On recovery, the inode will get truncated and the blocks will | |
2992 | * be freed, so we have a strong guarantee that no future commit will | |
2993 | * leave these blocks visible to the user.) | |
2994 | * | |
2995 | * Called with inode->sem down. | |
2996 | */ | |
617ba13b | 2997 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
2998 | { |
2999 | struct inode *inode = dentry->d_inode; | |
3000 | int error, rc = 0; | |
3001 | const unsigned int ia_valid = attr->ia_valid; | |
3002 | ||
3003 | error = inode_change_ok(inode, attr); | |
3004 | if (error) | |
3005 | return error; | |
3006 | ||
3007 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || | |
3008 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
3009 | handle_t *handle; | |
3010 | ||
3011 | /* (user+group)*(old+new) structure, inode write (sb, | |
3012 | * inode block, ? - but truncate inode update has it) */ | |
617ba13b MC |
3013 | handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+ |
3014 | EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3); | |
ac27a0ec DK |
3015 | if (IS_ERR(handle)) { |
3016 | error = PTR_ERR(handle); | |
3017 | goto err_out; | |
3018 | } | |
3019 | error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0; | |
3020 | if (error) { | |
617ba13b | 3021 | ext4_journal_stop(handle); |
ac27a0ec DK |
3022 | return error; |
3023 | } | |
3024 | /* Update corresponding info in inode so that everything is in | |
3025 | * one transaction */ | |
3026 | if (attr->ia_valid & ATTR_UID) | |
3027 | inode->i_uid = attr->ia_uid; | |
3028 | if (attr->ia_valid & ATTR_GID) | |
3029 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
3030 | error = ext4_mark_inode_dirty(handle, inode); |
3031 | ext4_journal_stop(handle); | |
ac27a0ec DK |
3032 | } |
3033 | ||
3034 | if (S_ISREG(inode->i_mode) && | |
3035 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { | |
3036 | handle_t *handle; | |
3037 | ||
617ba13b | 3038 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
3039 | if (IS_ERR(handle)) { |
3040 | error = PTR_ERR(handle); | |
3041 | goto err_out; | |
3042 | } | |
3043 | ||
617ba13b MC |
3044 | error = ext4_orphan_add(handle, inode); |
3045 | EXT4_I(inode)->i_disksize = attr->ia_size; | |
3046 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
3047 | if (!error) |
3048 | error = rc; | |
617ba13b | 3049 | ext4_journal_stop(handle); |
ac27a0ec DK |
3050 | } |
3051 | ||
3052 | rc = inode_setattr(inode, attr); | |
3053 | ||
617ba13b | 3054 | /* If inode_setattr's call to ext4_truncate failed to get a |
ac27a0ec DK |
3055 | * transaction handle at all, we need to clean up the in-core |
3056 | * orphan list manually. */ | |
3057 | if (inode->i_nlink) | |
617ba13b | 3058 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
3059 | |
3060 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 3061 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
3062 | |
3063 | err_out: | |
617ba13b | 3064 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
3065 | if (!error) |
3066 | error = rc; | |
3067 | return error; | |
3068 | } | |
3069 | ||
3070 | ||
3071 | /* | |
3072 | * How many blocks doth make a writepage()? | |
3073 | * | |
3074 | * With N blocks per page, it may be: | |
3075 | * N data blocks | |
3076 | * 2 indirect block | |
3077 | * 2 dindirect | |
3078 | * 1 tindirect | |
3079 | * N+5 bitmap blocks (from the above) | |
3080 | * N+5 group descriptor summary blocks | |
3081 | * 1 inode block | |
3082 | * 1 superblock. | |
617ba13b | 3083 | * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files |
ac27a0ec | 3084 | * |
617ba13b | 3085 | * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS |
ac27a0ec DK |
3086 | * |
3087 | * With ordered or writeback data it's the same, less the N data blocks. | |
3088 | * | |
3089 | * If the inode's direct blocks can hold an integral number of pages then a | |
3090 | * page cannot straddle two indirect blocks, and we can only touch one indirect | |
3091 | * and dindirect block, and the "5" above becomes "3". | |
3092 | * | |
3093 | * This still overestimates under most circumstances. If we were to pass the | |
3094 | * start and end offsets in here as well we could do block_to_path() on each | |
3095 | * block and work out the exact number of indirects which are touched. Pah. | |
3096 | */ | |
3097 | ||
a86c6181 | 3098 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 3099 | { |
617ba13b MC |
3100 | int bpp = ext4_journal_blocks_per_page(inode); |
3101 | int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3; | |
ac27a0ec DK |
3102 | int ret; |
3103 | ||
a86c6181 AT |
3104 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) |
3105 | return ext4_ext_writepage_trans_blocks(inode, bpp); | |
3106 | ||
617ba13b | 3107 | if (ext4_should_journal_data(inode)) |
ac27a0ec DK |
3108 | ret = 3 * (bpp + indirects) + 2; |
3109 | else | |
3110 | ret = 2 * (bpp + indirects) + 2; | |
3111 | ||
3112 | #ifdef CONFIG_QUOTA | |
3113 | /* We know that structure was already allocated during DQUOT_INIT so | |
3114 | * we will be updating only the data blocks + inodes */ | |
617ba13b | 3115 | ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); |
ac27a0ec DK |
3116 | #endif |
3117 | ||
3118 | return ret; | |
3119 | } | |
3120 | ||
3121 | /* | |
617ba13b | 3122 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
3123 | * Give this, we know that the caller already has write access to iloc->bh. |
3124 | */ | |
617ba13b MC |
3125 | int ext4_mark_iloc_dirty(handle_t *handle, |
3126 | struct inode *inode, struct ext4_iloc *iloc) | |
ac27a0ec DK |
3127 | { |
3128 | int err = 0; | |
3129 | ||
3130 | /* the do_update_inode consumes one bh->b_count */ | |
3131 | get_bh(iloc->bh); | |
3132 | ||
dab291af | 3133 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
617ba13b | 3134 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
3135 | put_bh(iloc->bh); |
3136 | return err; | |
3137 | } | |
3138 | ||
3139 | /* | |
3140 | * On success, We end up with an outstanding reference count against | |
3141 | * iloc->bh. This _must_ be cleaned up later. | |
3142 | */ | |
3143 | ||
3144 | int | |
617ba13b MC |
3145 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
3146 | struct ext4_iloc *iloc) | |
ac27a0ec DK |
3147 | { |
3148 | int err = 0; | |
3149 | if (handle) { | |
617ba13b | 3150 | err = ext4_get_inode_loc(inode, iloc); |
ac27a0ec DK |
3151 | if (!err) { |
3152 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
617ba13b | 3153 | err = ext4_journal_get_write_access(handle, iloc->bh); |
ac27a0ec DK |
3154 | if (err) { |
3155 | brelse(iloc->bh); | |
3156 | iloc->bh = NULL; | |
3157 | } | |
3158 | } | |
3159 | } | |
617ba13b | 3160 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
3161 | return err; |
3162 | } | |
3163 | ||
6dd4ee7c KS |
3164 | /* |
3165 | * Expand an inode by new_extra_isize bytes. | |
3166 | * Returns 0 on success or negative error number on failure. | |
3167 | */ | |
1d03ec98 AK |
3168 | static int ext4_expand_extra_isize(struct inode *inode, |
3169 | unsigned int new_extra_isize, | |
3170 | struct ext4_iloc iloc, | |
3171 | handle_t *handle) | |
6dd4ee7c KS |
3172 | { |
3173 | struct ext4_inode *raw_inode; | |
3174 | struct ext4_xattr_ibody_header *header; | |
3175 | struct ext4_xattr_entry *entry; | |
3176 | ||
3177 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
3178 | return 0; | |
3179 | ||
3180 | raw_inode = ext4_raw_inode(&iloc); | |
3181 | ||
3182 | header = IHDR(inode, raw_inode); | |
3183 | entry = IFIRST(header); | |
3184 | ||
3185 | /* No extended attributes present */ | |
3186 | if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) || | |
3187 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
3188 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, | |
3189 | new_extra_isize); | |
3190 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
3191 | return 0; | |
3192 | } | |
3193 | ||
3194 | /* try to expand with EAs present */ | |
3195 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
3196 | raw_inode, handle); | |
3197 | } | |
3198 | ||
ac27a0ec DK |
3199 | /* |
3200 | * What we do here is to mark the in-core inode as clean with respect to inode | |
3201 | * dirtiness (it may still be data-dirty). | |
3202 | * This means that the in-core inode may be reaped by prune_icache | |
3203 | * without having to perform any I/O. This is a very good thing, | |
3204 | * because *any* task may call prune_icache - even ones which | |
3205 | * have a transaction open against a different journal. | |
3206 | * | |
3207 | * Is this cheating? Not really. Sure, we haven't written the | |
3208 | * inode out, but prune_icache isn't a user-visible syncing function. | |
3209 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
3210 | * we start and wait on commits. | |
3211 | * | |
3212 | * Is this efficient/effective? Well, we're being nice to the system | |
3213 | * by cleaning up our inodes proactively so they can be reaped | |
3214 | * without I/O. But we are potentially leaving up to five seconds' | |
3215 | * worth of inodes floating about which prune_icache wants us to | |
3216 | * write out. One way to fix that would be to get prune_icache() | |
3217 | * to do a write_super() to free up some memory. It has the desired | |
3218 | * effect. | |
3219 | */ | |
617ba13b | 3220 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 3221 | { |
617ba13b | 3222 | struct ext4_iloc iloc; |
6dd4ee7c KS |
3223 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
3224 | static unsigned int mnt_count; | |
3225 | int err, ret; | |
ac27a0ec DK |
3226 | |
3227 | might_sleep(); | |
617ba13b | 3228 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
6dd4ee7c KS |
3229 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && |
3230 | !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) { | |
3231 | /* | |
3232 | * We need extra buffer credits since we may write into EA block | |
3233 | * with this same handle. If journal_extend fails, then it will | |
3234 | * only result in a minor loss of functionality for that inode. | |
3235 | * If this is felt to be critical, then e2fsck should be run to | |
3236 | * force a large enough s_min_extra_isize. | |
3237 | */ | |
3238 | if ((jbd2_journal_extend(handle, | |
3239 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
3240 | ret = ext4_expand_extra_isize(inode, | |
3241 | sbi->s_want_extra_isize, | |
3242 | iloc, handle); | |
3243 | if (ret) { | |
3244 | EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND; | |
c1bddad9 AK |
3245 | if (mnt_count != |
3246 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
6dd4ee7c KS |
3247 | ext4_warning(inode->i_sb, __FUNCTION__, |
3248 | "Unable to expand inode %lu. Delete" | |
3249 | " some EAs or run e2fsck.", | |
3250 | inode->i_ino); | |
c1bddad9 AK |
3251 | mnt_count = |
3252 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
3253 | } |
3254 | } | |
3255 | } | |
3256 | } | |
ac27a0ec | 3257 | if (!err) |
617ba13b | 3258 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
3259 | return err; |
3260 | } | |
3261 | ||
3262 | /* | |
617ba13b | 3263 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
3264 | * |
3265 | * We're really interested in the case where a file is being extended. | |
3266 | * i_size has been changed by generic_commit_write() and we thus need | |
3267 | * to include the updated inode in the current transaction. | |
3268 | * | |
3269 | * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks | |
3270 | * are allocated to the file. | |
3271 | * | |
3272 | * If the inode is marked synchronous, we don't honour that here - doing | |
3273 | * so would cause a commit on atime updates, which we don't bother doing. | |
3274 | * We handle synchronous inodes at the highest possible level. | |
3275 | */ | |
617ba13b | 3276 | void ext4_dirty_inode(struct inode *inode) |
ac27a0ec | 3277 | { |
617ba13b | 3278 | handle_t *current_handle = ext4_journal_current_handle(); |
ac27a0ec DK |
3279 | handle_t *handle; |
3280 | ||
617ba13b | 3281 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
3282 | if (IS_ERR(handle)) |
3283 | goto out; | |
3284 | if (current_handle && | |
3285 | current_handle->h_transaction != handle->h_transaction) { | |
3286 | /* This task has a transaction open against a different fs */ | |
3287 | printk(KERN_EMERG "%s: transactions do not match!\n", | |
3288 | __FUNCTION__); | |
3289 | } else { | |
3290 | jbd_debug(5, "marking dirty. outer handle=%p\n", | |
3291 | current_handle); | |
617ba13b | 3292 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 3293 | } |
617ba13b | 3294 | ext4_journal_stop(handle); |
ac27a0ec DK |
3295 | out: |
3296 | return; | |
3297 | } | |
3298 | ||
3299 | #if 0 | |
3300 | /* | |
3301 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
3302 | * it from being flushed to disk early. Unlike | |
617ba13b | 3303 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
3304 | * returns no iloc structure, so the caller needs to repeat the iloc |
3305 | * lookup to mark the inode dirty later. | |
3306 | */ | |
617ba13b | 3307 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 3308 | { |
617ba13b | 3309 | struct ext4_iloc iloc; |
ac27a0ec DK |
3310 | |
3311 | int err = 0; | |
3312 | if (handle) { | |
617ba13b | 3313 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
3314 | if (!err) { |
3315 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 3316 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 3317 | if (!err) |
617ba13b | 3318 | err = ext4_journal_dirty_metadata(handle, |
ac27a0ec DK |
3319 | iloc.bh); |
3320 | brelse(iloc.bh); | |
3321 | } | |
3322 | } | |
617ba13b | 3323 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
3324 | return err; |
3325 | } | |
3326 | #endif | |
3327 | ||
617ba13b | 3328 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
3329 | { |
3330 | journal_t *journal; | |
3331 | handle_t *handle; | |
3332 | int err; | |
3333 | ||
3334 | /* | |
3335 | * We have to be very careful here: changing a data block's | |
3336 | * journaling status dynamically is dangerous. If we write a | |
3337 | * data block to the journal, change the status and then delete | |
3338 | * that block, we risk forgetting to revoke the old log record | |
3339 | * from the journal and so a subsequent replay can corrupt data. | |
3340 | * So, first we make sure that the journal is empty and that | |
3341 | * nobody is changing anything. | |
3342 | */ | |
3343 | ||
617ba13b | 3344 | journal = EXT4_JOURNAL(inode); |
d699594d | 3345 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
3346 | return -EROFS; |
3347 | ||
dab291af MC |
3348 | jbd2_journal_lock_updates(journal); |
3349 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
3350 | |
3351 | /* | |
3352 | * OK, there are no updates running now, and all cached data is | |
3353 | * synced to disk. We are now in a completely consistent state | |
3354 | * which doesn't have anything in the journal, and we know that | |
3355 | * no filesystem updates are running, so it is safe to modify | |
3356 | * the inode's in-core data-journaling state flag now. | |
3357 | */ | |
3358 | ||
3359 | if (val) | |
617ba13b | 3360 | EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL; |
ac27a0ec | 3361 | else |
617ba13b MC |
3362 | EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL; |
3363 | ext4_set_aops(inode); | |
ac27a0ec | 3364 | |
dab291af | 3365 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
3366 | |
3367 | /* Finally we can mark the inode as dirty. */ | |
3368 | ||
617ba13b | 3369 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
3370 | if (IS_ERR(handle)) |
3371 | return PTR_ERR(handle); | |
3372 | ||
617ba13b | 3373 | err = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 3374 | handle->h_sync = 1; |
617ba13b MC |
3375 | ext4_journal_stop(handle); |
3376 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
3377 | |
3378 | return err; | |
3379 | } |