]>
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 | 28 | #include <linux/jbd2.h> |
ac27a0ec DK |
29 | #include <linux/highuid.h> |
30 | #include <linux/pagemap.h> | |
31 | #include <linux/quotaops.h> | |
32 | #include <linux/string.h> | |
33 | #include <linux/buffer_head.h> | |
34 | #include <linux/writeback.h> | |
64769240 | 35 | #include <linux/pagevec.h> |
ac27a0ec | 36 | #include <linux/mpage.h> |
e83c1397 | 37 | #include <linux/namei.h> |
ac27a0ec DK |
38 | #include <linux/uio.h> |
39 | #include <linux/bio.h> | |
4c0425ff | 40 | #include <linux/workqueue.h> |
744692dc | 41 | #include <linux/kernel.h> |
5a0e3ad6 | 42 | #include <linux/slab.h> |
9bffad1e | 43 | |
3dcf5451 | 44 | #include "ext4_jbd2.h" |
ac27a0ec DK |
45 | #include "xattr.h" |
46 | #include "acl.h" | |
d2a17637 | 47 | #include "ext4_extents.h" |
ac27a0ec | 48 | |
9bffad1e TT |
49 | #include <trace/events/ext4.h> |
50 | ||
a1d6cc56 AK |
51 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
52 | ||
678aaf48 JK |
53 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
54 | loff_t new_size) | |
55 | { | |
7f5aa215 JK |
56 | return jbd2_journal_begin_ordered_truncate( |
57 | EXT4_SB(inode->i_sb)->s_journal, | |
58 | &EXT4_I(inode)->jinode, | |
59 | new_size); | |
678aaf48 JK |
60 | } |
61 | ||
64769240 AT |
62 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
63 | ||
ac27a0ec DK |
64 | /* |
65 | * Test whether an inode is a fast symlink. | |
66 | */ | |
617ba13b | 67 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 68 | { |
617ba13b | 69 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
70 | (inode->i_sb->s_blocksize >> 9) : 0; |
71 | ||
72 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
73 | } | |
74 | ||
ac27a0ec DK |
75 | /* |
76 | * Work out how many blocks we need to proceed with the next chunk of a | |
77 | * truncate transaction. | |
78 | */ | |
79 | static unsigned long blocks_for_truncate(struct inode *inode) | |
80 | { | |
725d26d3 | 81 | ext4_lblk_t needed; |
ac27a0ec DK |
82 | |
83 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
84 | ||
85 | /* Give ourselves just enough room to cope with inodes in which | |
86 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
87 | * which resulted in random data in an inode which looked enough | |
617ba13b | 88 | * like a regular file for ext4 to try to delete it. Things |
ac27a0ec DK |
89 | * will go a bit crazy if that happens, but at least we should |
90 | * try not to panic the whole kernel. */ | |
91 | if (needed < 2) | |
92 | needed = 2; | |
93 | ||
94 | /* But we need to bound the transaction so we don't overflow the | |
95 | * journal. */ | |
617ba13b MC |
96 | if (needed > EXT4_MAX_TRANS_DATA) |
97 | needed = EXT4_MAX_TRANS_DATA; | |
ac27a0ec | 98 | |
617ba13b | 99 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
ac27a0ec DK |
100 | } |
101 | ||
102 | /* | |
103 | * Truncate transactions can be complex and absolutely huge. So we need to | |
104 | * be able to restart the transaction at a conventient checkpoint to make | |
105 | * sure we don't overflow the journal. | |
106 | * | |
107 | * start_transaction gets us a new handle for a truncate transaction, | |
108 | * and extend_transaction tries to extend the existing one a bit. If | |
109 | * extend fails, we need to propagate the failure up and restart the | |
110 | * transaction in the top-level truncate loop. --sct | |
111 | */ | |
112 | static handle_t *start_transaction(struct inode *inode) | |
113 | { | |
114 | handle_t *result; | |
115 | ||
617ba13b | 116 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
ac27a0ec DK |
117 | if (!IS_ERR(result)) |
118 | return result; | |
119 | ||
617ba13b | 120 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
ac27a0ec DK |
121 | return result; |
122 | } | |
123 | ||
124 | /* | |
125 | * Try to extend this transaction for the purposes of truncation. | |
126 | * | |
127 | * Returns 0 if we managed to create more room. If we can't create more | |
128 | * room, and the transaction must be restarted we return 1. | |
129 | */ | |
130 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
131 | { | |
0390131b FM |
132 | if (!ext4_handle_valid(handle)) |
133 | return 0; | |
134 | if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) | |
ac27a0ec | 135 | return 0; |
617ba13b | 136 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
ac27a0ec DK |
137 | return 0; |
138 | return 1; | |
139 | } | |
140 | ||
141 | /* | |
142 | * Restart the transaction associated with *handle. This does a commit, | |
143 | * so before we call here everything must be consistently dirtied against | |
144 | * this transaction. | |
145 | */ | |
fa5d1113 | 146 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 147 | int nblocks) |
ac27a0ec | 148 | { |
487caeef JK |
149 | int ret; |
150 | ||
151 | /* | |
e35fd660 | 152 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
153 | * moment, get_block can be called only for blocks inside i_size since |
154 | * page cache has been already dropped and writes are blocked by | |
155 | * i_mutex. So we can safely drop the i_data_sem here. | |
156 | */ | |
0390131b | 157 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 158 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef JK |
159 | up_write(&EXT4_I(inode)->i_data_sem); |
160 | ret = ext4_journal_restart(handle, blocks_for_truncate(inode)); | |
161 | down_write(&EXT4_I(inode)->i_data_sem); | |
fa5d1113 | 162 | ext4_discard_preallocations(inode); |
487caeef JK |
163 | |
164 | return ret; | |
ac27a0ec DK |
165 | } |
166 | ||
167 | /* | |
168 | * Called at the last iput() if i_nlink is zero. | |
169 | */ | |
af5bc92d | 170 | void ext4_delete_inode(struct inode *inode) |
ac27a0ec DK |
171 | { |
172 | handle_t *handle; | |
bc965ab3 | 173 | int err; |
ac27a0ec | 174 | |
907f4554 | 175 | if (!is_bad_inode(inode)) |
871a2931 | 176 | dquot_initialize(inode); |
907f4554 | 177 | |
678aaf48 JK |
178 | if (ext4_should_order_data(inode)) |
179 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
180 | truncate_inode_pages(&inode->i_data, 0); |
181 | ||
182 | if (is_bad_inode(inode)) | |
183 | goto no_delete; | |
184 | ||
bc965ab3 | 185 | handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3); |
ac27a0ec | 186 | if (IS_ERR(handle)) { |
bc965ab3 | 187 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
188 | /* |
189 | * If we're going to skip the normal cleanup, we still need to | |
190 | * make sure that the in-core orphan linked list is properly | |
191 | * cleaned up. | |
192 | */ | |
617ba13b | 193 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
194 | goto no_delete; |
195 | } | |
196 | ||
197 | if (IS_SYNC(inode)) | |
0390131b | 198 | ext4_handle_sync(handle); |
ac27a0ec | 199 | inode->i_size = 0; |
bc965ab3 TT |
200 | err = ext4_mark_inode_dirty(handle, inode); |
201 | if (err) { | |
12062ddd | 202 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
203 | "couldn't mark inode dirty (err %d)", err); |
204 | goto stop_handle; | |
205 | } | |
ac27a0ec | 206 | if (inode->i_blocks) |
617ba13b | 207 | ext4_truncate(inode); |
bc965ab3 TT |
208 | |
209 | /* | |
210 | * ext4_ext_truncate() doesn't reserve any slop when it | |
211 | * restarts journal transactions; therefore there may not be | |
212 | * enough credits left in the handle to remove the inode from | |
213 | * the orphan list and set the dtime field. | |
214 | */ | |
0390131b | 215 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
216 | err = ext4_journal_extend(handle, 3); |
217 | if (err > 0) | |
218 | err = ext4_journal_restart(handle, 3); | |
219 | if (err != 0) { | |
12062ddd | 220 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
221 | "couldn't extend journal (err %d)", err); |
222 | stop_handle: | |
223 | ext4_journal_stop(handle); | |
224 | goto no_delete; | |
225 | } | |
226 | } | |
227 | ||
ac27a0ec | 228 | /* |
617ba13b | 229 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 230 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 231 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 232 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 233 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
234 | * (Well, we could do this if we need to, but heck - it works) |
235 | */ | |
617ba13b MC |
236 | ext4_orphan_del(handle, inode); |
237 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
238 | |
239 | /* | |
240 | * One subtle ordering requirement: if anything has gone wrong | |
241 | * (transaction abort, IO errors, whatever), then we can still | |
242 | * do these next steps (the fs will already have been marked as | |
243 | * having errors), but we can't free the inode if the mark_dirty | |
244 | * fails. | |
245 | */ | |
617ba13b | 246 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec DK |
247 | /* If that failed, just do the required in-core inode clear. */ |
248 | clear_inode(inode); | |
249 | else | |
617ba13b MC |
250 | ext4_free_inode(handle, inode); |
251 | ext4_journal_stop(handle); | |
ac27a0ec DK |
252 | return; |
253 | no_delete: | |
254 | clear_inode(inode); /* We must guarantee clearing of inode... */ | |
255 | } | |
256 | ||
257 | typedef struct { | |
258 | __le32 *p; | |
259 | __le32 key; | |
260 | struct buffer_head *bh; | |
261 | } Indirect; | |
262 | ||
263 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
264 | { | |
265 | p->key = *(p->p = v); | |
266 | p->bh = bh; | |
267 | } | |
268 | ||
ac27a0ec | 269 | /** |
617ba13b | 270 | * ext4_block_to_path - parse the block number into array of offsets |
ac27a0ec DK |
271 | * @inode: inode in question (we are only interested in its superblock) |
272 | * @i_block: block number to be parsed | |
273 | * @offsets: array to store the offsets in | |
8c55e204 DK |
274 | * @boundary: set this non-zero if the referred-to block is likely to be |
275 | * followed (on disk) by an indirect block. | |
ac27a0ec | 276 | * |
617ba13b | 277 | * To store the locations of file's data ext4 uses a data structure common |
ac27a0ec DK |
278 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
279 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
280 | * This function translates the block number into path in that tree - | |
281 | * return value is the path length and @offsets[n] is the offset of | |
282 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
283 | * (negative or too large) warning is printed and zero returned. | |
284 | * | |
285 | * Note: function doesn't find node addresses, so no IO is needed. All | |
286 | * we need to know is the capacity of indirect blocks (taken from the | |
287 | * inode->i_sb). | |
288 | */ | |
289 | ||
290 | /* | |
291 | * Portability note: the last comparison (check that we fit into triple | |
292 | * indirect block) is spelled differently, because otherwise on an | |
293 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
294 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
295 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
296 | * i_block would have to be negative in the very beginning, so we would not | |
297 | * get there at all. | |
298 | */ | |
299 | ||
617ba13b | 300 | static int ext4_block_to_path(struct inode *inode, |
de9a55b8 TT |
301 | ext4_lblk_t i_block, |
302 | ext4_lblk_t offsets[4], int *boundary) | |
ac27a0ec | 303 | { |
617ba13b MC |
304 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
305 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
306 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
ac27a0ec DK |
307 | indirect_blocks = ptrs, |
308 | double_blocks = (1 << (ptrs_bits * 2)); | |
309 | int n = 0; | |
310 | int final = 0; | |
311 | ||
c333e073 | 312 | if (i_block < direct_blocks) { |
ac27a0ec DK |
313 | offsets[n++] = i_block; |
314 | final = direct_blocks; | |
af5bc92d | 315 | } else if ((i_block -= direct_blocks) < indirect_blocks) { |
617ba13b | 316 | offsets[n++] = EXT4_IND_BLOCK; |
ac27a0ec DK |
317 | offsets[n++] = i_block; |
318 | final = ptrs; | |
319 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
617ba13b | 320 | offsets[n++] = EXT4_DIND_BLOCK; |
ac27a0ec DK |
321 | offsets[n++] = i_block >> ptrs_bits; |
322 | offsets[n++] = i_block & (ptrs - 1); | |
323 | final = ptrs; | |
324 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
617ba13b | 325 | offsets[n++] = EXT4_TIND_BLOCK; |
ac27a0ec DK |
326 | offsets[n++] = i_block >> (ptrs_bits * 2); |
327 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
328 | offsets[n++] = i_block & (ptrs - 1); | |
329 | final = ptrs; | |
330 | } else { | |
12062ddd | 331 | ext4_warning(inode->i_sb, "block %lu > max in inode %lu", |
de9a55b8 TT |
332 | i_block + direct_blocks + |
333 | indirect_blocks + double_blocks, inode->i_ino); | |
ac27a0ec DK |
334 | } |
335 | if (boundary) | |
336 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
337 | return n; | |
338 | } | |
339 | ||
c398eda0 TT |
340 | static int __ext4_check_blockref(const char *function, unsigned int line, |
341 | struct inode *inode, | |
6fd058f7 TT |
342 | __le32 *p, unsigned int max) |
343 | { | |
1c13d5c0 | 344 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; |
f73953c0 | 345 | __le32 *bref = p; |
6fd058f7 TT |
346 | unsigned int blk; |
347 | ||
fe2c8191 | 348 | while (bref < p+max) { |
6fd058f7 | 349 | blk = le32_to_cpu(*bref++); |
de9a55b8 TT |
350 | if (blk && |
351 | unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb), | |
6fd058f7 | 352 | blk, 1))) { |
1c13d5c0 | 353 | es->s_last_error_block = cpu_to_le64(blk); |
c398eda0 TT |
354 | ext4_error_inode(inode, function, line, blk, |
355 | "invalid block"); | |
de9a55b8 TT |
356 | return -EIO; |
357 | } | |
358 | } | |
359 | return 0; | |
fe2c8191 TN |
360 | } |
361 | ||
362 | ||
363 | #define ext4_check_indirect_blockref(inode, bh) \ | |
c398eda0 TT |
364 | __ext4_check_blockref(__func__, __LINE__, inode, \ |
365 | (__le32 *)(bh)->b_data, \ | |
fe2c8191 TN |
366 | EXT4_ADDR_PER_BLOCK((inode)->i_sb)) |
367 | ||
368 | #define ext4_check_inode_blockref(inode) \ | |
c398eda0 TT |
369 | __ext4_check_blockref(__func__, __LINE__, inode, \ |
370 | EXT4_I(inode)->i_data, \ | |
fe2c8191 TN |
371 | EXT4_NDIR_BLOCKS) |
372 | ||
ac27a0ec | 373 | /** |
617ba13b | 374 | * ext4_get_branch - read the chain of indirect blocks leading to data |
ac27a0ec DK |
375 | * @inode: inode in question |
376 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
377 | * @offsets: offsets of pointers in inode/indirect blocks | |
378 | * @chain: place to store the result | |
379 | * @err: here we store the error value | |
380 | * | |
381 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
382 | * if everything went OK or the pointer to the last filled triple | |
383 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
384 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
385 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
386 | * number (it points into struct inode for i==0 and into the bh->b_data | |
387 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
388 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
389 | * numbers of the chain, addresses they were taken from (and where we can | |
390 | * verify that chain did not change) and buffer_heads hosting these | |
391 | * numbers. | |
392 | * | |
393 | * Function stops when it stumbles upon zero pointer (absent block) | |
394 | * (pointer to last triple returned, *@err == 0) | |
395 | * or when it gets an IO error reading an indirect block | |
396 | * (ditto, *@err == -EIO) | |
ac27a0ec DK |
397 | * or when it reads all @depth-1 indirect blocks successfully and finds |
398 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
c278bfec AK |
399 | * |
400 | * Need to be called with | |
0e855ac8 | 401 | * down_read(&EXT4_I(inode)->i_data_sem) |
ac27a0ec | 402 | */ |
725d26d3 AK |
403 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
404 | ext4_lblk_t *offsets, | |
ac27a0ec DK |
405 | Indirect chain[4], int *err) |
406 | { | |
407 | struct super_block *sb = inode->i_sb; | |
408 | Indirect *p = chain; | |
409 | struct buffer_head *bh; | |
410 | ||
411 | *err = 0; | |
412 | /* i_data is not going away, no lock needed */ | |
af5bc92d | 413 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); |
ac27a0ec DK |
414 | if (!p->key) |
415 | goto no_block; | |
416 | while (--depth) { | |
fe2c8191 TN |
417 | bh = sb_getblk(sb, le32_to_cpu(p->key)); |
418 | if (unlikely(!bh)) | |
ac27a0ec | 419 | goto failure; |
de9a55b8 | 420 | |
fe2c8191 TN |
421 | if (!bh_uptodate_or_lock(bh)) { |
422 | if (bh_submit_read(bh) < 0) { | |
423 | put_bh(bh); | |
424 | goto failure; | |
425 | } | |
426 | /* validate block references */ | |
427 | if (ext4_check_indirect_blockref(inode, bh)) { | |
428 | put_bh(bh); | |
429 | goto failure; | |
430 | } | |
431 | } | |
de9a55b8 | 432 | |
af5bc92d | 433 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); |
ac27a0ec DK |
434 | /* Reader: end */ |
435 | if (!p->key) | |
436 | goto no_block; | |
437 | } | |
438 | return NULL; | |
439 | ||
ac27a0ec DK |
440 | failure: |
441 | *err = -EIO; | |
442 | no_block: | |
443 | return p; | |
444 | } | |
445 | ||
446 | /** | |
617ba13b | 447 | * ext4_find_near - find a place for allocation with sufficient locality |
ac27a0ec DK |
448 | * @inode: owner |
449 | * @ind: descriptor of indirect block. | |
450 | * | |
1cc8dcf5 | 451 | * This function returns the preferred place for block allocation. |
ac27a0ec DK |
452 | * It is used when heuristic for sequential allocation fails. |
453 | * Rules are: | |
454 | * + if there is a block to the left of our position - allocate near it. | |
455 | * + if pointer will live in indirect block - allocate near that block. | |
456 | * + if pointer will live in inode - allocate in the same | |
457 | * cylinder group. | |
458 | * | |
459 | * In the latter case we colour the starting block by the callers PID to | |
460 | * prevent it from clashing with concurrent allocations for a different inode | |
461 | * in the same block group. The PID is used here so that functionally related | |
462 | * files will be close-by on-disk. | |
463 | * | |
464 | * Caller must make sure that @ind is valid and will stay that way. | |
465 | */ | |
617ba13b | 466 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
ac27a0ec | 467 | { |
617ba13b | 468 | struct ext4_inode_info *ei = EXT4_I(inode); |
af5bc92d | 469 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; |
ac27a0ec | 470 | __le32 *p; |
617ba13b | 471 | ext4_fsblk_t bg_start; |
74d3487f | 472 | ext4_fsblk_t last_block; |
617ba13b | 473 | ext4_grpblk_t colour; |
a4912123 TT |
474 | ext4_group_t block_group; |
475 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); | |
ac27a0ec DK |
476 | |
477 | /* Try to find previous block */ | |
478 | for (p = ind->p - 1; p >= start; p--) { | |
479 | if (*p) | |
480 | return le32_to_cpu(*p); | |
481 | } | |
482 | ||
483 | /* No such thing, so let's try location of indirect block */ | |
484 | if (ind->bh) | |
485 | return ind->bh->b_blocknr; | |
486 | ||
487 | /* | |
488 | * It is going to be referred to from the inode itself? OK, just put it | |
489 | * into the same cylinder group then. | |
490 | */ | |
a4912123 TT |
491 | block_group = ei->i_block_group; |
492 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { | |
493 | block_group &= ~(flex_size-1); | |
494 | if (S_ISREG(inode->i_mode)) | |
495 | block_group++; | |
496 | } | |
497 | bg_start = ext4_group_first_block_no(inode->i_sb, block_group); | |
74d3487f VC |
498 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
499 | ||
a4912123 TT |
500 | /* |
501 | * If we are doing delayed allocation, we don't need take | |
502 | * colour into account. | |
503 | */ | |
504 | if (test_opt(inode->i_sb, DELALLOC)) | |
505 | return bg_start; | |
506 | ||
74d3487f VC |
507 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
508 | colour = (current->pid % 16) * | |
617ba13b | 509 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
510 | else |
511 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
ac27a0ec DK |
512 | return bg_start + colour; |
513 | } | |
514 | ||
515 | /** | |
1cc8dcf5 | 516 | * ext4_find_goal - find a preferred place for allocation. |
ac27a0ec DK |
517 | * @inode: owner |
518 | * @block: block we want | |
ac27a0ec | 519 | * @partial: pointer to the last triple within a chain |
ac27a0ec | 520 | * |
1cc8dcf5 | 521 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 522 | * returns it. |
fb0a387d ES |
523 | * Because this is only used for non-extent files, we limit the block nr |
524 | * to 32 bits. | |
ac27a0ec | 525 | */ |
725d26d3 | 526 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
de9a55b8 | 527 | Indirect *partial) |
ac27a0ec | 528 | { |
fb0a387d ES |
529 | ext4_fsblk_t goal; |
530 | ||
ac27a0ec | 531 | /* |
c2ea3fde | 532 | * XXX need to get goal block from mballoc's data structures |
ac27a0ec | 533 | */ |
ac27a0ec | 534 | |
fb0a387d ES |
535 | goal = ext4_find_near(inode, partial); |
536 | goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; | |
537 | return goal; | |
ac27a0ec DK |
538 | } |
539 | ||
540 | /** | |
617ba13b | 541 | * ext4_blks_to_allocate: Look up the block map and count the number |
ac27a0ec DK |
542 | * of direct blocks need to be allocated for the given branch. |
543 | * | |
544 | * @branch: chain of indirect blocks | |
545 | * @k: number of blocks need for indirect blocks | |
546 | * @blks: number of data blocks to be mapped. | |
547 | * @blocks_to_boundary: the offset in the indirect block | |
548 | * | |
549 | * return the total number of blocks to be allocate, including the | |
550 | * direct and indirect blocks. | |
551 | */ | |
498e5f24 | 552 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, |
de9a55b8 | 553 | int blocks_to_boundary) |
ac27a0ec | 554 | { |
498e5f24 | 555 | unsigned int count = 0; |
ac27a0ec DK |
556 | |
557 | /* | |
558 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
559 | * then it's clear blocks on that path have not allocated | |
560 | */ | |
561 | if (k > 0) { | |
562 | /* right now we don't handle cross boundary allocation */ | |
563 | if (blks < blocks_to_boundary + 1) | |
564 | count += blks; | |
565 | else | |
566 | count += blocks_to_boundary + 1; | |
567 | return count; | |
568 | } | |
569 | ||
570 | count++; | |
571 | while (count < blks && count <= blocks_to_boundary && | |
572 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
573 | count++; | |
574 | } | |
575 | return count; | |
576 | } | |
577 | ||
578 | /** | |
617ba13b | 579 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
ac27a0ec DK |
580 | * @indirect_blks: the number of blocks need to allocate for indirect |
581 | * blocks | |
582 | * | |
583 | * @new_blocks: on return it will store the new block numbers for | |
584 | * the indirect blocks(if needed) and the first direct block, | |
585 | * @blks: on return it will store the total number of allocated | |
586 | * direct blocks | |
587 | */ | |
617ba13b | 588 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
589 | ext4_lblk_t iblock, ext4_fsblk_t goal, |
590 | int indirect_blks, int blks, | |
591 | ext4_fsblk_t new_blocks[4], int *err) | |
ac27a0ec | 592 | { |
815a1130 | 593 | struct ext4_allocation_request ar; |
ac27a0ec | 594 | int target, i; |
7061eba7 | 595 | unsigned long count = 0, blk_allocated = 0; |
ac27a0ec | 596 | int index = 0; |
617ba13b | 597 | ext4_fsblk_t current_block = 0; |
ac27a0ec DK |
598 | int ret = 0; |
599 | ||
600 | /* | |
601 | * Here we try to allocate the requested multiple blocks at once, | |
602 | * on a best-effort basis. | |
603 | * To build a branch, we should allocate blocks for | |
604 | * the indirect blocks(if not allocated yet), and at least | |
605 | * the first direct block of this branch. That's the | |
606 | * minimum number of blocks need to allocate(required) | |
607 | */ | |
7061eba7 AK |
608 | /* first we try to allocate the indirect blocks */ |
609 | target = indirect_blks; | |
610 | while (target > 0) { | |
ac27a0ec DK |
611 | count = target; |
612 | /* allocating blocks for indirect blocks and direct blocks */ | |
7061eba7 AK |
613 | current_block = ext4_new_meta_blocks(handle, inode, |
614 | goal, &count, err); | |
ac27a0ec DK |
615 | if (*err) |
616 | goto failed_out; | |
617 | ||
273df556 FM |
618 | if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) { |
619 | EXT4_ERROR_INODE(inode, | |
620 | "current_block %llu + count %lu > %d!", | |
621 | current_block, count, | |
622 | EXT4_MAX_BLOCK_FILE_PHYS); | |
623 | *err = -EIO; | |
624 | goto failed_out; | |
625 | } | |
fb0a387d | 626 | |
ac27a0ec DK |
627 | target -= count; |
628 | /* allocate blocks for indirect blocks */ | |
629 | while (index < indirect_blks && count) { | |
630 | new_blocks[index++] = current_block++; | |
631 | count--; | |
632 | } | |
7061eba7 AK |
633 | if (count > 0) { |
634 | /* | |
635 | * save the new block number | |
636 | * for the first direct block | |
637 | */ | |
638 | new_blocks[index] = current_block; | |
639 | printk(KERN_INFO "%s returned more blocks than " | |
640 | "requested\n", __func__); | |
641 | WARN_ON(1); | |
ac27a0ec | 642 | break; |
7061eba7 | 643 | } |
ac27a0ec DK |
644 | } |
645 | ||
7061eba7 AK |
646 | target = blks - count ; |
647 | blk_allocated = count; | |
648 | if (!target) | |
649 | goto allocated; | |
650 | /* Now allocate data blocks */ | |
815a1130 TT |
651 | memset(&ar, 0, sizeof(ar)); |
652 | ar.inode = inode; | |
653 | ar.goal = goal; | |
654 | ar.len = target; | |
655 | ar.logical = iblock; | |
656 | if (S_ISREG(inode->i_mode)) | |
657 | /* enable in-core preallocation only for regular files */ | |
658 | ar.flags = EXT4_MB_HINT_DATA; | |
659 | ||
660 | current_block = ext4_mb_new_blocks(handle, &ar, err); | |
273df556 FM |
661 | if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) { |
662 | EXT4_ERROR_INODE(inode, | |
663 | "current_block %llu + ar.len %d > %d!", | |
664 | current_block, ar.len, | |
665 | EXT4_MAX_BLOCK_FILE_PHYS); | |
666 | *err = -EIO; | |
667 | goto failed_out; | |
668 | } | |
815a1130 | 669 | |
7061eba7 AK |
670 | if (*err && (target == blks)) { |
671 | /* | |
672 | * if the allocation failed and we didn't allocate | |
673 | * any blocks before | |
674 | */ | |
675 | goto failed_out; | |
676 | } | |
677 | if (!*err) { | |
678 | if (target == blks) { | |
de9a55b8 TT |
679 | /* |
680 | * save the new block number | |
681 | * for the first direct block | |
682 | */ | |
7061eba7 AK |
683 | new_blocks[index] = current_block; |
684 | } | |
815a1130 | 685 | blk_allocated += ar.len; |
7061eba7 AK |
686 | } |
687 | allocated: | |
ac27a0ec | 688 | /* total number of blocks allocated for direct blocks */ |
7061eba7 | 689 | ret = blk_allocated; |
ac27a0ec DK |
690 | *err = 0; |
691 | return ret; | |
692 | failed_out: | |
af5bc92d | 693 | for (i = 0; i < index; i++) |
e6362609 | 694 | ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0); |
ac27a0ec DK |
695 | return ret; |
696 | } | |
697 | ||
698 | /** | |
617ba13b | 699 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
ac27a0ec DK |
700 | * @inode: owner |
701 | * @indirect_blks: number of allocated indirect blocks | |
702 | * @blks: number of allocated direct blocks | |
703 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
704 | * @branch: place to store the chain in. | |
705 | * | |
706 | * This function allocates blocks, zeroes out all but the last one, | |
707 | * links them into chain and (if we are synchronous) writes them to disk. | |
708 | * In other words, it prepares a branch that can be spliced onto the | |
709 | * inode. It stores the information about that chain in the branch[], in | |
617ba13b | 710 | * the same format as ext4_get_branch() would do. We are calling it after |
ac27a0ec DK |
711 | * we had read the existing part of chain and partial points to the last |
712 | * triple of that (one with zero ->key). Upon the exit we have the same | |
617ba13b | 713 | * picture as after the successful ext4_get_block(), except that in one |
ac27a0ec DK |
714 | * place chain is disconnected - *branch->p is still zero (we did not |
715 | * set the last link), but branch->key contains the number that should | |
716 | * be placed into *branch->p to fill that gap. | |
717 | * | |
718 | * If allocation fails we free all blocks we've allocated (and forget | |
719 | * their buffer_heads) and return the error value the from failed | |
617ba13b | 720 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
ac27a0ec DK |
721 | * as described above and return 0. |
722 | */ | |
617ba13b | 723 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
724 | ext4_lblk_t iblock, int indirect_blks, |
725 | int *blks, ext4_fsblk_t goal, | |
726 | ext4_lblk_t *offsets, Indirect *branch) | |
ac27a0ec DK |
727 | { |
728 | int blocksize = inode->i_sb->s_blocksize; | |
729 | int i, n = 0; | |
730 | int err = 0; | |
731 | struct buffer_head *bh; | |
732 | int num; | |
617ba13b MC |
733 | ext4_fsblk_t new_blocks[4]; |
734 | ext4_fsblk_t current_block; | |
ac27a0ec | 735 | |
7061eba7 | 736 | num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks, |
ac27a0ec DK |
737 | *blks, new_blocks, &err); |
738 | if (err) | |
739 | return err; | |
740 | ||
741 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
742 | /* | |
743 | * metadata blocks and data blocks are allocated. | |
744 | */ | |
745 | for (n = 1; n <= indirect_blks; n++) { | |
746 | /* | |
747 | * Get buffer_head for parent block, zero it out | |
748 | * and set the pointer to new one, then send | |
749 | * parent to disk. | |
750 | */ | |
751 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
752 | branch[n].bh = bh; | |
753 | lock_buffer(bh); | |
754 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 755 | err = ext4_journal_get_create_access(handle, bh); |
ac27a0ec | 756 | if (err) { |
6487a9d3 CW |
757 | /* Don't brelse(bh) here; it's done in |
758 | * ext4_journal_forget() below */ | |
ac27a0ec | 759 | unlock_buffer(bh); |
ac27a0ec DK |
760 | goto failed; |
761 | } | |
762 | ||
763 | memset(bh->b_data, 0, blocksize); | |
764 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
765 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
766 | *branch[n].p = branch[n].key; | |
af5bc92d | 767 | if (n == indirect_blks) { |
ac27a0ec DK |
768 | current_block = new_blocks[n]; |
769 | /* | |
770 | * End of chain, update the last new metablock of | |
771 | * the chain to point to the new allocated | |
772 | * data blocks numbers | |
773 | */ | |
de9a55b8 | 774 | for (i = 1; i < num; i++) |
ac27a0ec DK |
775 | *(branch[n].p + i) = cpu_to_le32(++current_block); |
776 | } | |
777 | BUFFER_TRACE(bh, "marking uptodate"); | |
778 | set_buffer_uptodate(bh); | |
779 | unlock_buffer(bh); | |
780 | ||
0390131b FM |
781 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
782 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
783 | if (err) |
784 | goto failed; | |
785 | } | |
786 | *blks = num; | |
787 | return err; | |
788 | failed: | |
789 | /* Allocation failed, free what we already allocated */ | |
e6362609 | 790 | ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0); |
ac27a0ec | 791 | for (i = 1; i <= n ; i++) { |
60e6679e | 792 | /* |
e6362609 TT |
793 | * branch[i].bh is newly allocated, so there is no |
794 | * need to revoke the block, which is why we don't | |
795 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
b7e57e7c | 796 | */ |
e6362609 TT |
797 | ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, |
798 | EXT4_FREE_BLOCKS_FORGET); | |
ac27a0ec | 799 | } |
e6362609 TT |
800 | for (i = n+1; i < indirect_blks; i++) |
801 | ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0); | |
ac27a0ec | 802 | |
e6362609 | 803 | ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0); |
ac27a0ec DK |
804 | |
805 | return err; | |
806 | } | |
807 | ||
808 | /** | |
617ba13b | 809 | * ext4_splice_branch - splice the allocated branch onto inode. |
ac27a0ec DK |
810 | * @inode: owner |
811 | * @block: (logical) number of block we are adding | |
812 | * @chain: chain of indirect blocks (with a missing link - see | |
617ba13b | 813 | * ext4_alloc_branch) |
ac27a0ec DK |
814 | * @where: location of missing link |
815 | * @num: number of indirect blocks we are adding | |
816 | * @blks: number of direct blocks we are adding | |
817 | * | |
818 | * This function fills the missing link and does all housekeeping needed in | |
819 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
820 | * chain to new block and return 0. | |
821 | */ | |
617ba13b | 822 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
823 | ext4_lblk_t block, Indirect *where, int num, |
824 | int blks) | |
ac27a0ec DK |
825 | { |
826 | int i; | |
827 | int err = 0; | |
617ba13b | 828 | ext4_fsblk_t current_block; |
ac27a0ec | 829 | |
ac27a0ec DK |
830 | /* |
831 | * If we're splicing into a [td]indirect block (as opposed to the | |
832 | * inode) then we need to get write access to the [td]indirect block | |
833 | * before the splice. | |
834 | */ | |
835 | if (where->bh) { | |
836 | BUFFER_TRACE(where->bh, "get_write_access"); | |
617ba13b | 837 | err = ext4_journal_get_write_access(handle, where->bh); |
ac27a0ec DK |
838 | if (err) |
839 | goto err_out; | |
840 | } | |
841 | /* That's it */ | |
842 | ||
843 | *where->p = where->key; | |
844 | ||
845 | /* | |
846 | * Update the host buffer_head or inode to point to more just allocated | |
847 | * direct blocks blocks | |
848 | */ | |
849 | if (num == 0 && blks > 1) { | |
850 | current_block = le32_to_cpu(where->key) + 1; | |
851 | for (i = 1; i < blks; i++) | |
af5bc92d | 852 | *(where->p + i) = cpu_to_le32(current_block++); |
ac27a0ec DK |
853 | } |
854 | ||
ac27a0ec | 855 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
ac27a0ec DK |
856 | /* had we spliced it onto indirect block? */ |
857 | if (where->bh) { | |
858 | /* | |
859 | * If we spliced it onto an indirect block, we haven't | |
860 | * altered the inode. Note however that if it is being spliced | |
861 | * onto an indirect block at the very end of the file (the | |
862 | * file is growing) then we *will* alter the inode to reflect | |
863 | * the new i_size. But that is not done here - it is done in | |
617ba13b | 864 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
ac27a0ec DK |
865 | */ |
866 | jbd_debug(5, "splicing indirect only\n"); | |
0390131b FM |
867 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); |
868 | err = ext4_handle_dirty_metadata(handle, inode, where->bh); | |
ac27a0ec DK |
869 | if (err) |
870 | goto err_out; | |
871 | } else { | |
872 | /* | |
873 | * OK, we spliced it into the inode itself on a direct block. | |
ac27a0ec | 874 | */ |
41591750 | 875 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
876 | jbd_debug(5, "splicing direct\n"); |
877 | } | |
878 | return err; | |
879 | ||
880 | err_out: | |
881 | for (i = 1; i <= num; i++) { | |
60e6679e | 882 | /* |
e6362609 TT |
883 | * branch[i].bh is newly allocated, so there is no |
884 | * need to revoke the block, which is why we don't | |
885 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
b7e57e7c | 886 | */ |
e6362609 TT |
887 | ext4_free_blocks(handle, inode, where[i].bh, 0, 1, |
888 | EXT4_FREE_BLOCKS_FORGET); | |
ac27a0ec | 889 | } |
e6362609 TT |
890 | ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key), |
891 | blks, 0); | |
ac27a0ec DK |
892 | |
893 | return err; | |
894 | } | |
895 | ||
896 | /* | |
e35fd660 | 897 | * The ext4_ind_map_blocks() function handles non-extents inodes |
b920c755 | 898 | * (i.e., using the traditional indirect/double-indirect i_blocks |
e35fd660 | 899 | * scheme) for ext4_map_blocks(). |
b920c755 | 900 | * |
ac27a0ec DK |
901 | * Allocation strategy is simple: if we have to allocate something, we will |
902 | * have to go the whole way to leaf. So let's do it before attaching anything | |
903 | * to tree, set linkage between the newborn blocks, write them if sync is | |
904 | * required, recheck the path, free and repeat if check fails, otherwise | |
905 | * set the last missing link (that will protect us from any truncate-generated | |
906 | * removals - all blocks on the path are immune now) and possibly force the | |
907 | * write on the parent block. | |
908 | * That has a nice additional property: no special recovery from the failed | |
909 | * allocations is needed - we simply release blocks and do not touch anything | |
910 | * reachable from inode. | |
911 | * | |
912 | * `handle' can be NULL if create == 0. | |
913 | * | |
ac27a0ec DK |
914 | * return > 0, # of blocks mapped or allocated. |
915 | * return = 0, if plain lookup failed. | |
916 | * return < 0, error case. | |
c278bfec | 917 | * |
b920c755 TT |
918 | * The ext4_ind_get_blocks() function should be called with |
919 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem | |
920 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or | |
921 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system | |
922 | * blocks. | |
ac27a0ec | 923 | */ |
e35fd660 TT |
924 | static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode, |
925 | struct ext4_map_blocks *map, | |
de9a55b8 | 926 | int flags) |
ac27a0ec DK |
927 | { |
928 | int err = -EIO; | |
725d26d3 | 929 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
930 | Indirect chain[4]; |
931 | Indirect *partial; | |
617ba13b | 932 | ext4_fsblk_t goal; |
ac27a0ec DK |
933 | int indirect_blks; |
934 | int blocks_to_boundary = 0; | |
935 | int depth; | |
ac27a0ec | 936 | int count = 0; |
617ba13b | 937 | ext4_fsblk_t first_block = 0; |
ac27a0ec | 938 | |
12e9b892 | 939 | J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); |
c2177057 | 940 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); |
e35fd660 | 941 | depth = ext4_block_to_path(inode, map->m_lblk, offsets, |
de9a55b8 | 942 | &blocks_to_boundary); |
ac27a0ec DK |
943 | |
944 | if (depth == 0) | |
945 | goto out; | |
946 | ||
617ba13b | 947 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
948 | |
949 | /* Simplest case - block found, no allocation needed */ | |
950 | if (!partial) { | |
951 | first_block = le32_to_cpu(chain[depth - 1].key); | |
ac27a0ec DK |
952 | count++; |
953 | /*map more blocks*/ | |
e35fd660 | 954 | while (count < map->m_len && count <= blocks_to_boundary) { |
617ba13b | 955 | ext4_fsblk_t blk; |
ac27a0ec | 956 | |
ac27a0ec DK |
957 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
958 | ||
959 | if (blk == first_block + count) | |
960 | count++; | |
961 | else | |
962 | break; | |
963 | } | |
c278bfec | 964 | goto got_it; |
ac27a0ec DK |
965 | } |
966 | ||
967 | /* Next simple case - plain lookup or failed read of indirect block */ | |
c2177057 | 968 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO) |
ac27a0ec DK |
969 | goto cleanup; |
970 | ||
ac27a0ec | 971 | /* |
c2ea3fde | 972 | * Okay, we need to do block allocation. |
ac27a0ec | 973 | */ |
e35fd660 | 974 | goal = ext4_find_goal(inode, map->m_lblk, partial); |
ac27a0ec DK |
975 | |
976 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
977 | indirect_blks = (chain + depth) - partial - 1; | |
978 | ||
979 | /* | |
980 | * Next look up the indirect map to count the totoal number of | |
981 | * direct blocks to allocate for this branch. | |
982 | */ | |
617ba13b | 983 | count = ext4_blks_to_allocate(partial, indirect_blks, |
e35fd660 | 984 | map->m_len, blocks_to_boundary); |
ac27a0ec | 985 | /* |
617ba13b | 986 | * Block out ext4_truncate while we alter the tree |
ac27a0ec | 987 | */ |
e35fd660 | 988 | err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks, |
de9a55b8 TT |
989 | &count, goal, |
990 | offsets + (partial - chain), partial); | |
ac27a0ec DK |
991 | |
992 | /* | |
617ba13b | 993 | * The ext4_splice_branch call will free and forget any buffers |
ac27a0ec DK |
994 | * on the new chain if there is a failure, but that risks using |
995 | * up transaction credits, especially for bitmaps where the | |
996 | * credits cannot be returned. Can we handle this somehow? We | |
997 | * may need to return -EAGAIN upwards in the worst case. --sct | |
998 | */ | |
999 | if (!err) | |
e35fd660 | 1000 | err = ext4_splice_branch(handle, inode, map->m_lblk, |
de9a55b8 | 1001 | partial, indirect_blks, count); |
2bba702d | 1002 | if (err) |
ac27a0ec DK |
1003 | goto cleanup; |
1004 | ||
e35fd660 | 1005 | map->m_flags |= EXT4_MAP_NEW; |
b436b9be JK |
1006 | |
1007 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
ac27a0ec | 1008 | got_it: |
e35fd660 TT |
1009 | map->m_flags |= EXT4_MAP_MAPPED; |
1010 | map->m_pblk = le32_to_cpu(chain[depth-1].key); | |
1011 | map->m_len = count; | |
ac27a0ec | 1012 | if (count > blocks_to_boundary) |
e35fd660 | 1013 | map->m_flags |= EXT4_MAP_BOUNDARY; |
ac27a0ec DK |
1014 | err = count; |
1015 | /* Clean up and exit */ | |
1016 | partial = chain + depth - 1; /* the whole chain */ | |
1017 | cleanup: | |
1018 | while (partial > chain) { | |
1019 | BUFFER_TRACE(partial->bh, "call brelse"); | |
1020 | brelse(partial->bh); | |
1021 | partial--; | |
1022 | } | |
ac27a0ec DK |
1023 | out: |
1024 | return err; | |
1025 | } | |
1026 | ||
a9e7f447 DM |
1027 | #ifdef CONFIG_QUOTA |
1028 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 1029 | { |
a9e7f447 | 1030 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 1031 | } |
a9e7f447 | 1032 | #endif |
9d0be502 | 1033 | |
12219aea AK |
1034 | /* |
1035 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 1036 | * to allocate a new block at @lblocks for non extent file based file |
12219aea | 1037 | */ |
9d0be502 TT |
1038 | static int ext4_indirect_calc_metadata_amount(struct inode *inode, |
1039 | sector_t lblock) | |
12219aea | 1040 | { |
9d0be502 | 1041 | struct ext4_inode_info *ei = EXT4_I(inode); |
d330a5be | 1042 | sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1); |
9d0be502 | 1043 | int blk_bits; |
12219aea | 1044 | |
9d0be502 TT |
1045 | if (lblock < EXT4_NDIR_BLOCKS) |
1046 | return 0; | |
12219aea | 1047 | |
9d0be502 | 1048 | lblock -= EXT4_NDIR_BLOCKS; |
12219aea | 1049 | |
9d0be502 TT |
1050 | if (ei->i_da_metadata_calc_len && |
1051 | (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) { | |
1052 | ei->i_da_metadata_calc_len++; | |
1053 | return 0; | |
1054 | } | |
1055 | ei->i_da_metadata_calc_last_lblock = lblock & dind_mask; | |
1056 | ei->i_da_metadata_calc_len = 1; | |
d330a5be | 1057 | blk_bits = order_base_2(lblock); |
9d0be502 | 1058 | return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1; |
12219aea AK |
1059 | } |
1060 | ||
1061 | /* | |
1062 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 1063 | * to allocate a block located at @lblock |
12219aea | 1064 | */ |
9d0be502 | 1065 | static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock) |
12219aea | 1066 | { |
12e9b892 | 1067 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 1068 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 1069 | |
9d0be502 | 1070 | return ext4_indirect_calc_metadata_amount(inode, lblock); |
12219aea AK |
1071 | } |
1072 | ||
0637c6f4 TT |
1073 | /* |
1074 | * Called with i_data_sem down, which is important since we can call | |
1075 | * ext4_discard_preallocations() from here. | |
1076 | */ | |
5f634d06 AK |
1077 | void ext4_da_update_reserve_space(struct inode *inode, |
1078 | int used, int quota_claim) | |
12219aea AK |
1079 | { |
1080 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1081 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
1082 | |
1083 | spin_lock(&ei->i_block_reservation_lock); | |
f8ec9d68 | 1084 | trace_ext4_da_update_reserve_space(inode, used); |
0637c6f4 TT |
1085 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
1086 | ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d " | |
1087 | "with only %d reserved data blocks\n", | |
1088 | __func__, inode->i_ino, used, | |
1089 | ei->i_reserved_data_blocks); | |
1090 | WARN_ON(1); | |
1091 | used = ei->i_reserved_data_blocks; | |
1092 | } | |
12219aea | 1093 | |
0637c6f4 TT |
1094 | /* Update per-inode reservations */ |
1095 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 1096 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
72b8ab9d ES |
1097 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, |
1098 | used + ei->i_allocated_meta_blocks); | |
0637c6f4 | 1099 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 1100 | |
0637c6f4 TT |
1101 | if (ei->i_reserved_data_blocks == 0) { |
1102 | /* | |
1103 | * We can release all of the reserved metadata blocks | |
1104 | * only when we have written all of the delayed | |
1105 | * allocation blocks. | |
1106 | */ | |
72b8ab9d ES |
1107 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, |
1108 | ei->i_reserved_meta_blocks); | |
ee5f4d9c | 1109 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1110 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 1111 | } |
12219aea | 1112 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1113 | |
72b8ab9d ES |
1114 | /* Update quota subsystem for data blocks */ |
1115 | if (quota_claim) | |
5dd4056d | 1116 | dquot_claim_block(inode, used); |
72b8ab9d | 1117 | else { |
5f634d06 AK |
1118 | /* |
1119 | * We did fallocate with an offset that is already delayed | |
1120 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 1121 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 1122 | */ |
72b8ab9d | 1123 | dquot_release_reservation_block(inode, used); |
5f634d06 | 1124 | } |
d6014301 AK |
1125 | |
1126 | /* | |
1127 | * If we have done all the pending block allocations and if | |
1128 | * there aren't any writers on the inode, we can discard the | |
1129 | * inode's preallocations. | |
1130 | */ | |
0637c6f4 TT |
1131 | if ((ei->i_reserved_data_blocks == 0) && |
1132 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 1133 | ext4_discard_preallocations(inode); |
12219aea AK |
1134 | } |
1135 | ||
e29136f8 | 1136 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
1137 | unsigned int line, |
1138 | struct ext4_map_blocks *map) | |
6fd058f7 | 1139 | { |
24676da4 TT |
1140 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
1141 | map->m_len)) { | |
c398eda0 TT |
1142 | ext4_error_inode(inode, func, line, map->m_pblk, |
1143 | "lblock %lu mapped to illegal pblock " | |
1144 | "(length %d)", (unsigned long) map->m_lblk, | |
1145 | map->m_len); | |
6fd058f7 TT |
1146 | return -EIO; |
1147 | } | |
1148 | return 0; | |
1149 | } | |
1150 | ||
e29136f8 | 1151 | #define check_block_validity(inode, map) \ |
c398eda0 | 1152 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 1153 | |
55138e0b | 1154 | /* |
1f94533d TT |
1155 | * Return the number of contiguous dirty pages in a given inode |
1156 | * starting at page frame idx. | |
55138e0b TT |
1157 | */ |
1158 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
1159 | unsigned int max_pages) | |
1160 | { | |
1161 | struct address_space *mapping = inode->i_mapping; | |
1162 | pgoff_t index; | |
1163 | struct pagevec pvec; | |
1164 | pgoff_t num = 0; | |
1165 | int i, nr_pages, done = 0; | |
1166 | ||
1167 | if (max_pages == 0) | |
1168 | return 0; | |
1169 | pagevec_init(&pvec, 0); | |
1170 | while (!done) { | |
1171 | index = idx; | |
1172 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
1173 | PAGECACHE_TAG_DIRTY, | |
1174 | (pgoff_t)PAGEVEC_SIZE); | |
1175 | if (nr_pages == 0) | |
1176 | break; | |
1177 | for (i = 0; i < nr_pages; i++) { | |
1178 | struct page *page = pvec.pages[i]; | |
1179 | struct buffer_head *bh, *head; | |
1180 | ||
1181 | lock_page(page); | |
1182 | if (unlikely(page->mapping != mapping) || | |
1183 | !PageDirty(page) || | |
1184 | PageWriteback(page) || | |
1185 | page->index != idx) { | |
1186 | done = 1; | |
1187 | unlock_page(page); | |
1188 | break; | |
1189 | } | |
1f94533d TT |
1190 | if (page_has_buffers(page)) { |
1191 | bh = head = page_buffers(page); | |
1192 | do { | |
1193 | if (!buffer_delay(bh) && | |
1194 | !buffer_unwritten(bh)) | |
1195 | done = 1; | |
1196 | bh = bh->b_this_page; | |
1197 | } while (!done && (bh != head)); | |
1198 | } | |
55138e0b TT |
1199 | unlock_page(page); |
1200 | if (done) | |
1201 | break; | |
1202 | idx++; | |
1203 | num++; | |
1204 | if (num >= max_pages) | |
1205 | break; | |
1206 | } | |
1207 | pagevec_release(&pvec); | |
1208 | } | |
1209 | return num; | |
1210 | } | |
1211 | ||
f5ab0d1f | 1212 | /* |
e35fd660 | 1213 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 1214 | * and returns if the blocks are already mapped. |
f5ab0d1f | 1215 | * |
f5ab0d1f MC |
1216 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
1217 | * and store the allocated blocks in the result buffer head and mark it | |
1218 | * mapped. | |
1219 | * | |
e35fd660 TT |
1220 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
1221 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
1222 | * based files |
1223 | * | |
1224 | * On success, it returns the number of blocks being mapped or allocate. | |
1225 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
1226 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
1227 | * the buffer head is mapped. | |
1228 | * | |
1229 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
1230 | * that casem, buffer head is unmapped | |
1231 | * | |
1232 | * It returns the error in case of allocation failure. | |
1233 | */ | |
e35fd660 TT |
1234 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
1235 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 AK |
1236 | { |
1237 | int retval; | |
f5ab0d1f | 1238 | |
e35fd660 TT |
1239 | map->m_flags = 0; |
1240 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
1241 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
1242 | (unsigned long) map->m_lblk); | |
4df3d265 | 1243 | /* |
b920c755 TT |
1244 | * Try to see if we can get the block without requesting a new |
1245 | * file system block. | |
4df3d265 AK |
1246 | */ |
1247 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 1248 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 1249 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
0e855ac8 | 1250 | } else { |
e35fd660 | 1251 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
0e855ac8 | 1252 | } |
4df3d265 | 1253 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 1254 | |
e35fd660 | 1255 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 1256 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
1257 | if (ret != 0) |
1258 | return ret; | |
1259 | } | |
1260 | ||
f5ab0d1f | 1261 | /* If it is only a block(s) look up */ |
c2177057 | 1262 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
1263 | return retval; |
1264 | ||
1265 | /* | |
1266 | * Returns if the blocks have already allocated | |
1267 | * | |
1268 | * Note that if blocks have been preallocated | |
1269 | * ext4_ext_get_block() returns th create = 0 | |
1270 | * with buffer head unmapped. | |
1271 | */ | |
e35fd660 | 1272 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
1273 | return retval; |
1274 | ||
2a8964d6 AK |
1275 | /* |
1276 | * When we call get_blocks without the create flag, the | |
1277 | * BH_Unwritten flag could have gotten set if the blocks | |
1278 | * requested were part of a uninitialized extent. We need to | |
1279 | * clear this flag now that we are committed to convert all or | |
1280 | * part of the uninitialized extent to be an initialized | |
1281 | * extent. This is because we need to avoid the combination | |
1282 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
1283 | * set on the buffer_head. | |
1284 | */ | |
e35fd660 | 1285 | map->m_flags &= ~EXT4_MAP_UNWRITTEN; |
2a8964d6 | 1286 | |
4df3d265 | 1287 | /* |
f5ab0d1f MC |
1288 | * New blocks allocate and/or writing to uninitialized extent |
1289 | * will possibly result in updating i_data, so we take | |
1290 | * the write lock of i_data_sem, and call get_blocks() | |
1291 | * with create == 1 flag. | |
4df3d265 AK |
1292 | */ |
1293 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
1294 | |
1295 | /* | |
1296 | * if the caller is from delayed allocation writeout path | |
1297 | * we have already reserved fs blocks for allocation | |
1298 | * let the underlying get_block() function know to | |
1299 | * avoid double accounting | |
1300 | */ | |
c2177057 | 1301 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
d2a17637 | 1302 | EXT4_I(inode)->i_delalloc_reserved_flag = 1; |
4df3d265 AK |
1303 | /* |
1304 | * We need to check for EXT4 here because migrate | |
1305 | * could have changed the inode type in between | |
1306 | */ | |
12e9b892 | 1307 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 1308 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 1309 | } else { |
e35fd660 | 1310 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 1311 | |
e35fd660 | 1312 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
1313 | /* |
1314 | * We allocated new blocks which will result in | |
1315 | * i_data's format changing. Force the migrate | |
1316 | * to fail by clearing migrate flags | |
1317 | */ | |
19f5fb7a | 1318 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 1319 | } |
d2a17637 | 1320 | |
5f634d06 AK |
1321 | /* |
1322 | * Update reserved blocks/metadata blocks after successful | |
1323 | * block allocation which had been deferred till now. We don't | |
1324 | * support fallocate for non extent files. So we can update | |
1325 | * reserve space here. | |
1326 | */ | |
1327 | if ((retval > 0) && | |
1296cc85 | 1328 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
1329 | ext4_da_update_reserve_space(inode, retval, 1); |
1330 | } | |
2ac3b6e0 | 1331 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
d2a17637 | 1332 | EXT4_I(inode)->i_delalloc_reserved_flag = 0; |
2ac3b6e0 | 1333 | |
4df3d265 | 1334 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 1335 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 1336 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
1337 | if (ret != 0) |
1338 | return ret; | |
1339 | } | |
0e855ac8 AK |
1340 | return retval; |
1341 | } | |
1342 | ||
f3bd1f3f MC |
1343 | /* Maximum number of blocks we map for direct IO at once. */ |
1344 | #define DIO_MAX_BLOCKS 4096 | |
1345 | ||
2ed88685 TT |
1346 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
1347 | struct buffer_head *bh, int flags) | |
ac27a0ec | 1348 | { |
3e4fdaf8 | 1349 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 1350 | struct ext4_map_blocks map; |
7fb5409d | 1351 | int ret = 0, started = 0; |
f3bd1f3f | 1352 | int dio_credits; |
ac27a0ec | 1353 | |
2ed88685 TT |
1354 | map.m_lblk = iblock; |
1355 | map.m_len = bh->b_size >> inode->i_blkbits; | |
1356 | ||
1357 | if (flags && !handle) { | |
7fb5409d | 1358 | /* Direct IO write... */ |
2ed88685 TT |
1359 | if (map.m_len > DIO_MAX_BLOCKS) |
1360 | map.m_len = DIO_MAX_BLOCKS; | |
1361 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
f3bd1f3f | 1362 | handle = ext4_journal_start(inode, dio_credits); |
7fb5409d | 1363 | if (IS_ERR(handle)) { |
ac27a0ec | 1364 | ret = PTR_ERR(handle); |
2ed88685 | 1365 | return ret; |
ac27a0ec | 1366 | } |
7fb5409d | 1367 | started = 1; |
ac27a0ec DK |
1368 | } |
1369 | ||
2ed88685 | 1370 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 1371 | if (ret > 0) { |
2ed88685 TT |
1372 | map_bh(bh, inode->i_sb, map.m_pblk); |
1373 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1374 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 1375 | ret = 0; |
ac27a0ec | 1376 | } |
7fb5409d JK |
1377 | if (started) |
1378 | ext4_journal_stop(handle); | |
ac27a0ec DK |
1379 | return ret; |
1380 | } | |
1381 | ||
2ed88685 TT |
1382 | int ext4_get_block(struct inode *inode, sector_t iblock, |
1383 | struct buffer_head *bh, int create) | |
1384 | { | |
1385 | return _ext4_get_block(inode, iblock, bh, | |
1386 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
1387 | } | |
1388 | ||
ac27a0ec DK |
1389 | /* |
1390 | * `handle' can be NULL if create is zero | |
1391 | */ | |
617ba13b | 1392 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 1393 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 1394 | { |
2ed88685 TT |
1395 | struct ext4_map_blocks map; |
1396 | struct buffer_head *bh; | |
ac27a0ec DK |
1397 | int fatal = 0, err; |
1398 | ||
1399 | J_ASSERT(handle != NULL || create == 0); | |
1400 | ||
2ed88685 TT |
1401 | map.m_lblk = block; |
1402 | map.m_len = 1; | |
1403 | err = ext4_map_blocks(handle, inode, &map, | |
1404 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 1405 | |
2ed88685 TT |
1406 | if (err < 0) |
1407 | *errp = err; | |
1408 | if (err <= 0) | |
1409 | return NULL; | |
1410 | *errp = 0; | |
1411 | ||
1412 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
1413 | if (!bh) { | |
1414 | *errp = -EIO; | |
1415 | return NULL; | |
ac27a0ec | 1416 | } |
2ed88685 TT |
1417 | if (map.m_flags & EXT4_MAP_NEW) { |
1418 | J_ASSERT(create != 0); | |
1419 | J_ASSERT(handle != NULL); | |
ac27a0ec | 1420 | |
2ed88685 TT |
1421 | /* |
1422 | * Now that we do not always journal data, we should | |
1423 | * keep in mind whether this should always journal the | |
1424 | * new buffer as metadata. For now, regular file | |
1425 | * writes use ext4_get_block instead, so it's not a | |
1426 | * problem. | |
1427 | */ | |
1428 | lock_buffer(bh); | |
1429 | BUFFER_TRACE(bh, "call get_create_access"); | |
1430 | fatal = ext4_journal_get_create_access(handle, bh); | |
1431 | if (!fatal && !buffer_uptodate(bh)) { | |
1432 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
1433 | set_buffer_uptodate(bh); | |
ac27a0ec | 1434 | } |
2ed88685 TT |
1435 | unlock_buffer(bh); |
1436 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
1437 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
1438 | if (!fatal) | |
1439 | fatal = err; | |
1440 | } else { | |
1441 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 1442 | } |
2ed88685 TT |
1443 | if (fatal) { |
1444 | *errp = fatal; | |
1445 | brelse(bh); | |
1446 | bh = NULL; | |
1447 | } | |
1448 | return bh; | |
ac27a0ec DK |
1449 | } |
1450 | ||
617ba13b | 1451 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 1452 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 1453 | { |
af5bc92d | 1454 | struct buffer_head *bh; |
ac27a0ec | 1455 | |
617ba13b | 1456 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
1457 | if (!bh) |
1458 | return bh; | |
1459 | if (buffer_uptodate(bh)) | |
1460 | return bh; | |
1461 | ll_rw_block(READ_META, 1, &bh); | |
1462 | wait_on_buffer(bh); | |
1463 | if (buffer_uptodate(bh)) | |
1464 | return bh; | |
1465 | put_bh(bh); | |
1466 | *err = -EIO; | |
1467 | return NULL; | |
1468 | } | |
1469 | ||
af5bc92d TT |
1470 | static int walk_page_buffers(handle_t *handle, |
1471 | struct buffer_head *head, | |
1472 | unsigned from, | |
1473 | unsigned to, | |
1474 | int *partial, | |
1475 | int (*fn)(handle_t *handle, | |
1476 | struct buffer_head *bh)) | |
ac27a0ec DK |
1477 | { |
1478 | struct buffer_head *bh; | |
1479 | unsigned block_start, block_end; | |
1480 | unsigned blocksize = head->b_size; | |
1481 | int err, ret = 0; | |
1482 | struct buffer_head *next; | |
1483 | ||
af5bc92d TT |
1484 | for (bh = head, block_start = 0; |
1485 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1486 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1487 | next = bh->b_this_page; |
1488 | block_end = block_start + blocksize; | |
1489 | if (block_end <= from || block_start >= to) { | |
1490 | if (partial && !buffer_uptodate(bh)) | |
1491 | *partial = 1; | |
1492 | continue; | |
1493 | } | |
1494 | err = (*fn)(handle, bh); | |
1495 | if (!ret) | |
1496 | ret = err; | |
1497 | } | |
1498 | return ret; | |
1499 | } | |
1500 | ||
1501 | /* | |
1502 | * To preserve ordering, it is essential that the hole instantiation and | |
1503 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1504 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1505 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1506 | * prepare_write() is the right place. |
1507 | * | |
617ba13b MC |
1508 | * Also, this function can nest inside ext4_writepage() -> |
1509 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
1510 | * has generated enough buffer credits to do the whole page. So we won't |
1511 | * block on the journal in that case, which is good, because the caller may | |
1512 | * be PF_MEMALLOC. | |
1513 | * | |
617ba13b | 1514 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1515 | * quota file writes. If we were to commit the transaction while thus |
1516 | * reentered, there can be a deadlock - we would be holding a quota | |
1517 | * lock, and the commit would never complete if another thread had a | |
1518 | * transaction open and was blocking on the quota lock - a ranking | |
1519 | * violation. | |
1520 | * | |
dab291af | 1521 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1522 | * will _not_ run commit under these circumstances because handle->h_ref |
1523 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1524 | * write. | |
1525 | */ | |
1526 | static int do_journal_get_write_access(handle_t *handle, | |
de9a55b8 | 1527 | struct buffer_head *bh) |
ac27a0ec DK |
1528 | { |
1529 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1530 | return 0; | |
617ba13b | 1531 | return ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1532 | } |
1533 | ||
b9a4207d JK |
1534 | /* |
1535 | * Truncate blocks that were not used by write. We have to truncate the | |
1536 | * pagecache as well so that corresponding buffers get properly unmapped. | |
1537 | */ | |
1538 | static void ext4_truncate_failed_write(struct inode *inode) | |
1539 | { | |
1540 | truncate_inode_pages(inode->i_mapping, inode->i_size); | |
1541 | ext4_truncate(inode); | |
1542 | } | |
1543 | ||
744692dc JZ |
1544 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
1545 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 1546 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1547 | loff_t pos, unsigned len, unsigned flags, |
1548 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1549 | { |
af5bc92d | 1550 | struct inode *inode = mapping->host; |
1938a150 | 1551 | int ret, needed_blocks; |
ac27a0ec DK |
1552 | handle_t *handle; |
1553 | int retries = 0; | |
af5bc92d | 1554 | struct page *page; |
de9a55b8 | 1555 | pgoff_t index; |
af5bc92d | 1556 | unsigned from, to; |
bfc1af65 | 1557 | |
9bffad1e | 1558 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1559 | /* |
1560 | * Reserve one block more for addition to orphan list in case | |
1561 | * we allocate blocks but write fails for some reason | |
1562 | */ | |
1563 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 1564 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
1565 | from = pos & (PAGE_CACHE_SIZE - 1); |
1566 | to = from + len; | |
ac27a0ec DK |
1567 | |
1568 | retry: | |
af5bc92d TT |
1569 | handle = ext4_journal_start(inode, needed_blocks); |
1570 | if (IS_ERR(handle)) { | |
1571 | ret = PTR_ERR(handle); | |
1572 | goto out; | |
7479d2b9 | 1573 | } |
ac27a0ec | 1574 | |
ebd3610b JK |
1575 | /* We cannot recurse into the filesystem as the transaction is already |
1576 | * started */ | |
1577 | flags |= AOP_FLAG_NOFS; | |
1578 | ||
54566b2c | 1579 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
1580 | if (!page) { |
1581 | ext4_journal_stop(handle); | |
1582 | ret = -ENOMEM; | |
1583 | goto out; | |
1584 | } | |
1585 | *pagep = page; | |
1586 | ||
744692dc JZ |
1587 | if (ext4_should_dioread_nolock(inode)) |
1588 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, | |
1589 | fsdata, ext4_get_block_write); | |
1590 | else | |
1591 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, | |
1592 | fsdata, ext4_get_block); | |
bfc1af65 NP |
1593 | |
1594 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
1595 | ret = walk_page_buffers(handle, page_buffers(page), |
1596 | from, to, NULL, do_journal_get_write_access); | |
1597 | } | |
bfc1af65 NP |
1598 | |
1599 | if (ret) { | |
af5bc92d | 1600 | unlock_page(page); |
af5bc92d | 1601 | page_cache_release(page); |
ae4d5372 AK |
1602 | /* |
1603 | * block_write_begin may have instantiated a few blocks | |
1604 | * outside i_size. Trim these off again. Don't need | |
1605 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1606 | * |
1607 | * Add inode to orphan list in case we crash before | |
1608 | * truncate finishes | |
ae4d5372 | 1609 | */ |
ffacfa7a | 1610 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1611 | ext4_orphan_add(handle, inode); |
1612 | ||
1613 | ext4_journal_stop(handle); | |
1614 | if (pos + len > inode->i_size) { | |
b9a4207d | 1615 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1616 | /* |
ffacfa7a | 1617 | * If truncate failed early the inode might |
1938a150 AK |
1618 | * still be on the orphan list; we need to |
1619 | * make sure the inode is removed from the | |
1620 | * orphan list in that case. | |
1621 | */ | |
1622 | if (inode->i_nlink) | |
1623 | ext4_orphan_del(NULL, inode); | |
1624 | } | |
bfc1af65 NP |
1625 | } |
1626 | ||
617ba13b | 1627 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 1628 | goto retry; |
7479d2b9 | 1629 | out: |
ac27a0ec DK |
1630 | return ret; |
1631 | } | |
1632 | ||
bfc1af65 NP |
1633 | /* For write_end() in data=journal mode */ |
1634 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
1635 | { |
1636 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1637 | return 0; | |
1638 | set_buffer_uptodate(bh); | |
0390131b | 1639 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
1640 | } |
1641 | ||
f8514083 | 1642 | static int ext4_generic_write_end(struct file *file, |
de9a55b8 TT |
1643 | struct address_space *mapping, |
1644 | loff_t pos, unsigned len, unsigned copied, | |
1645 | struct page *page, void *fsdata) | |
f8514083 AK |
1646 | { |
1647 | int i_size_changed = 0; | |
1648 | struct inode *inode = mapping->host; | |
1649 | handle_t *handle = ext4_journal_current_handle(); | |
1650 | ||
1651 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
1652 | ||
1653 | /* | |
1654 | * No need to use i_size_read() here, the i_size | |
1655 | * cannot change under us because we hold i_mutex. | |
1656 | * | |
1657 | * But it's important to update i_size while still holding page lock: | |
1658 | * page writeout could otherwise come in and zero beyond i_size. | |
1659 | */ | |
1660 | if (pos + copied > inode->i_size) { | |
1661 | i_size_write(inode, pos + copied); | |
1662 | i_size_changed = 1; | |
1663 | } | |
1664 | ||
1665 | if (pos + copied > EXT4_I(inode)->i_disksize) { | |
1666 | /* We need to mark inode dirty even if | |
1667 | * new_i_size is less that inode->i_size | |
1668 | * bu greater than i_disksize.(hint delalloc) | |
1669 | */ | |
1670 | ext4_update_i_disksize(inode, (pos + copied)); | |
1671 | i_size_changed = 1; | |
1672 | } | |
1673 | unlock_page(page); | |
1674 | page_cache_release(page); | |
1675 | ||
1676 | /* | |
1677 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1678 | * makes the holding time of page lock longer. Second, it forces lock | |
1679 | * ordering of page lock and transaction start for journaling | |
1680 | * filesystems. | |
1681 | */ | |
1682 | if (i_size_changed) | |
1683 | ext4_mark_inode_dirty(handle, inode); | |
1684 | ||
1685 | return copied; | |
1686 | } | |
1687 | ||
ac27a0ec DK |
1688 | /* |
1689 | * We need to pick up the new inode size which generic_commit_write gave us | |
1690 | * `file' can be NULL - eg, when called from page_symlink(). | |
1691 | * | |
617ba13b | 1692 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1693 | * buffers are managed internally. |
1694 | */ | |
bfc1af65 | 1695 | static int ext4_ordered_write_end(struct file *file, |
de9a55b8 TT |
1696 | struct address_space *mapping, |
1697 | loff_t pos, unsigned len, unsigned copied, | |
1698 | struct page *page, void *fsdata) | |
ac27a0ec | 1699 | { |
617ba13b | 1700 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1701 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1702 | int ret = 0, ret2; |
1703 | ||
9bffad1e | 1704 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
678aaf48 | 1705 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
1706 | |
1707 | if (ret == 0) { | |
f8514083 | 1708 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1709 | page, fsdata); |
f8a87d89 | 1710 | copied = ret2; |
ffacfa7a | 1711 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1712 | /* if we have allocated more blocks and copied |
1713 | * less. We will have blocks allocated outside | |
1714 | * inode->i_size. So truncate them | |
1715 | */ | |
1716 | ext4_orphan_add(handle, inode); | |
f8a87d89 RK |
1717 | if (ret2 < 0) |
1718 | ret = ret2; | |
ac27a0ec | 1719 | } |
617ba13b | 1720 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1721 | if (!ret) |
1722 | ret = ret2; | |
bfc1af65 | 1723 | |
f8514083 | 1724 | if (pos + len > inode->i_size) { |
b9a4207d | 1725 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1726 | /* |
ffacfa7a | 1727 | * If truncate failed early the inode might still be |
f8514083 AK |
1728 | * on the orphan list; we need to make sure the inode |
1729 | * is removed from the orphan list in that case. | |
1730 | */ | |
1731 | if (inode->i_nlink) | |
1732 | ext4_orphan_del(NULL, inode); | |
1733 | } | |
1734 | ||
1735 | ||
bfc1af65 | 1736 | return ret ? ret : copied; |
ac27a0ec DK |
1737 | } |
1738 | ||
bfc1af65 | 1739 | static int ext4_writeback_write_end(struct file *file, |
de9a55b8 TT |
1740 | struct address_space *mapping, |
1741 | loff_t pos, unsigned len, unsigned copied, | |
1742 | struct page *page, void *fsdata) | |
ac27a0ec | 1743 | { |
617ba13b | 1744 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1745 | struct inode *inode = mapping->host; |
ac27a0ec | 1746 | int ret = 0, ret2; |
ac27a0ec | 1747 | |
9bffad1e | 1748 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
f8514083 | 1749 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1750 | page, fsdata); |
f8a87d89 | 1751 | copied = ret2; |
ffacfa7a | 1752 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1753 | /* if we have allocated more blocks and copied |
1754 | * less. We will have blocks allocated outside | |
1755 | * inode->i_size. So truncate them | |
1756 | */ | |
1757 | ext4_orphan_add(handle, inode); | |
1758 | ||
f8a87d89 RK |
1759 | if (ret2 < 0) |
1760 | ret = ret2; | |
ac27a0ec | 1761 | |
617ba13b | 1762 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1763 | if (!ret) |
1764 | ret = ret2; | |
bfc1af65 | 1765 | |
f8514083 | 1766 | if (pos + len > inode->i_size) { |
b9a4207d | 1767 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1768 | /* |
ffacfa7a | 1769 | * If truncate failed early the inode might still be |
f8514083 AK |
1770 | * on the orphan list; we need to make sure the inode |
1771 | * is removed from the orphan list in that case. | |
1772 | */ | |
1773 | if (inode->i_nlink) | |
1774 | ext4_orphan_del(NULL, inode); | |
1775 | } | |
1776 | ||
bfc1af65 | 1777 | return ret ? ret : copied; |
ac27a0ec DK |
1778 | } |
1779 | ||
bfc1af65 | 1780 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1781 | struct address_space *mapping, |
1782 | loff_t pos, unsigned len, unsigned copied, | |
1783 | struct page *page, void *fsdata) | |
ac27a0ec | 1784 | { |
617ba13b | 1785 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1786 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1787 | int ret = 0, ret2; |
1788 | int partial = 0; | |
bfc1af65 | 1789 | unsigned from, to; |
cf17fea6 | 1790 | loff_t new_i_size; |
ac27a0ec | 1791 | |
9bffad1e | 1792 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1793 | from = pos & (PAGE_CACHE_SIZE - 1); |
1794 | to = from + len; | |
1795 | ||
1796 | if (copied < len) { | |
1797 | if (!PageUptodate(page)) | |
1798 | copied = 0; | |
1799 | page_zero_new_buffers(page, from+copied, to); | |
1800 | } | |
ac27a0ec DK |
1801 | |
1802 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1803 | to, &partial, write_end_fn); |
ac27a0ec DK |
1804 | if (!partial) |
1805 | SetPageUptodate(page); | |
cf17fea6 AK |
1806 | new_i_size = pos + copied; |
1807 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1808 | i_size_write(inode, pos+copied); |
19f5fb7a | 1809 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
cf17fea6 AK |
1810 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1811 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1812 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1813 | if (!ret) |
1814 | ret = ret2; | |
1815 | } | |
bfc1af65 | 1816 | |
cf108bca | 1817 | unlock_page(page); |
f8514083 | 1818 | page_cache_release(page); |
ffacfa7a | 1819 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1820 | /* if we have allocated more blocks and copied |
1821 | * less. We will have blocks allocated outside | |
1822 | * inode->i_size. So truncate them | |
1823 | */ | |
1824 | ext4_orphan_add(handle, inode); | |
1825 | ||
617ba13b | 1826 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1827 | if (!ret) |
1828 | ret = ret2; | |
f8514083 | 1829 | if (pos + len > inode->i_size) { |
b9a4207d | 1830 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1831 | /* |
ffacfa7a | 1832 | * If truncate failed early the inode might still be |
f8514083 AK |
1833 | * on the orphan list; we need to make sure the inode |
1834 | * is removed from the orphan list in that case. | |
1835 | */ | |
1836 | if (inode->i_nlink) | |
1837 | ext4_orphan_del(NULL, inode); | |
1838 | } | |
bfc1af65 NP |
1839 | |
1840 | return ret ? ret : copied; | |
ac27a0ec | 1841 | } |
d2a17637 | 1842 | |
9d0be502 TT |
1843 | /* |
1844 | * Reserve a single block located at lblock | |
1845 | */ | |
1846 | static int ext4_da_reserve_space(struct inode *inode, sector_t lblock) | |
d2a17637 | 1847 | { |
030ba6bc | 1848 | int retries = 0; |
60e58e0f | 1849 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1850 | struct ext4_inode_info *ei = EXT4_I(inode); |
72b8ab9d | 1851 | unsigned long md_needed; |
5dd4056d | 1852 | int ret; |
d2a17637 MC |
1853 | |
1854 | /* | |
1855 | * recalculate the amount of metadata blocks to reserve | |
1856 | * in order to allocate nrblocks | |
1857 | * worse case is one extent per block | |
1858 | */ | |
030ba6bc | 1859 | repeat: |
0637c6f4 | 1860 | spin_lock(&ei->i_block_reservation_lock); |
9d0be502 | 1861 | md_needed = ext4_calc_metadata_amount(inode, lblock); |
f8ec9d68 | 1862 | trace_ext4_da_reserve_space(inode, md_needed); |
0637c6f4 | 1863 | spin_unlock(&ei->i_block_reservation_lock); |
d2a17637 | 1864 | |
60e58e0f | 1865 | /* |
72b8ab9d ES |
1866 | * We will charge metadata quota at writeout time; this saves |
1867 | * us from metadata over-estimation, though we may go over by | |
1868 | * a small amount in the end. Here we just reserve for data. | |
60e58e0f | 1869 | */ |
72b8ab9d | 1870 | ret = dquot_reserve_block(inode, 1); |
5dd4056d CH |
1871 | if (ret) |
1872 | return ret; | |
72b8ab9d ES |
1873 | /* |
1874 | * We do still charge estimated metadata to the sb though; | |
1875 | * we cannot afford to run out of free blocks. | |
1876 | */ | |
9d0be502 | 1877 | if (ext4_claim_free_blocks(sbi, md_needed + 1)) { |
72b8ab9d | 1878 | dquot_release_reservation_block(inode, 1); |
030ba6bc AK |
1879 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1880 | yield(); | |
1881 | goto repeat; | |
1882 | } | |
d2a17637 MC |
1883 | return -ENOSPC; |
1884 | } | |
0637c6f4 | 1885 | spin_lock(&ei->i_block_reservation_lock); |
9d0be502 | 1886 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1887 | ei->i_reserved_meta_blocks += md_needed; |
1888 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1889 | |
d2a17637 MC |
1890 | return 0; /* success */ |
1891 | } | |
1892 | ||
12219aea | 1893 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1894 | { |
1895 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1896 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1897 | |
cd213226 MC |
1898 | if (!to_free) |
1899 | return; /* Nothing to release, exit */ | |
1900 | ||
d2a17637 | 1901 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1902 | |
5a58ec87 | 1903 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1904 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1905 | /* |
0637c6f4 TT |
1906 | * if there aren't enough reserved blocks, then the |
1907 | * counter is messed up somewhere. Since this | |
1908 | * function is called from invalidate page, it's | |
1909 | * harmless to return without any action. | |
cd213226 | 1910 | */ |
0637c6f4 TT |
1911 | ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: " |
1912 | "ino %lu, to_free %d with only %d reserved " | |
1913 | "data blocks\n", inode->i_ino, to_free, | |
1914 | ei->i_reserved_data_blocks); | |
1915 | WARN_ON(1); | |
1916 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1917 | } |
0637c6f4 | 1918 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1919 | |
0637c6f4 TT |
1920 | if (ei->i_reserved_data_blocks == 0) { |
1921 | /* | |
1922 | * We can release all of the reserved metadata blocks | |
1923 | * only when we have written all of the delayed | |
1924 | * allocation blocks. | |
1925 | */ | |
72b8ab9d ES |
1926 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, |
1927 | ei->i_reserved_meta_blocks); | |
ee5f4d9c | 1928 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1929 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1930 | } |
d2a17637 | 1931 | |
72b8ab9d | 1932 | /* update fs dirty data blocks counter */ |
0637c6f4 | 1933 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free); |
d2a17637 | 1934 | |
d2a17637 | 1935 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1936 | |
5dd4056d | 1937 | dquot_release_reservation_block(inode, to_free); |
d2a17637 MC |
1938 | } |
1939 | ||
1940 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1941 | unsigned long offset) |
d2a17637 MC |
1942 | { |
1943 | int to_release = 0; | |
1944 | struct buffer_head *head, *bh; | |
1945 | unsigned int curr_off = 0; | |
1946 | ||
1947 | head = page_buffers(page); | |
1948 | bh = head; | |
1949 | do { | |
1950 | unsigned int next_off = curr_off + bh->b_size; | |
1951 | ||
1952 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1953 | to_release++; | |
1954 | clear_buffer_delay(bh); | |
1955 | } | |
1956 | curr_off = next_off; | |
1957 | } while ((bh = bh->b_this_page) != head); | |
12219aea | 1958 | ext4_da_release_space(page->mapping->host, to_release); |
d2a17637 | 1959 | } |
ac27a0ec | 1960 | |
64769240 AT |
1961 | /* |
1962 | * Delayed allocation stuff | |
1963 | */ | |
1964 | ||
64769240 AT |
1965 | /* |
1966 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1967 | * them with writepage() call back |
64769240 AT |
1968 | * |
1969 | * @mpd->inode: inode | |
1970 | * @mpd->first_page: first page of the extent | |
1971 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1972 | * |
1973 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1974 | * to be allocated. this may be wrong if allocation failed. | |
1975 | * | |
1976 | * As pages are already locked by write_cache_pages(), we can't use it | |
1977 | */ | |
1978 | static int mpage_da_submit_io(struct mpage_da_data *mpd) | |
1979 | { | |
22208ded | 1980 | long pages_skipped; |
791b7f08 AK |
1981 | struct pagevec pvec; |
1982 | unsigned long index, end; | |
1983 | int ret = 0, err, nr_pages, i; | |
1984 | struct inode *inode = mpd->inode; | |
1985 | struct address_space *mapping = inode->i_mapping; | |
64769240 AT |
1986 | |
1987 | BUG_ON(mpd->next_page <= mpd->first_page); | |
791b7f08 AK |
1988 | /* |
1989 | * We need to start from the first_page to the next_page - 1 | |
1990 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1991 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1992 | * at the currently mapped buffer_heads. |
1993 | */ | |
64769240 AT |
1994 | index = mpd->first_page; |
1995 | end = mpd->next_page - 1; | |
1996 | ||
791b7f08 | 1997 | pagevec_init(&pvec, 0); |
64769240 | 1998 | while (index <= end) { |
791b7f08 | 1999 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
2000 | if (nr_pages == 0) |
2001 | break; | |
2002 | for (i = 0; i < nr_pages; i++) { | |
2003 | struct page *page = pvec.pages[i]; | |
2004 | ||
791b7f08 AK |
2005 | index = page->index; |
2006 | if (index > end) | |
2007 | break; | |
2008 | index++; | |
2009 | ||
2010 | BUG_ON(!PageLocked(page)); | |
2011 | BUG_ON(PageWriteback(page)); | |
2012 | ||
22208ded | 2013 | pages_skipped = mpd->wbc->pages_skipped; |
a1d6cc56 | 2014 | err = mapping->a_ops->writepage(page, mpd->wbc); |
22208ded AK |
2015 | if (!err && (pages_skipped == mpd->wbc->pages_skipped)) |
2016 | /* | |
2017 | * have successfully written the page | |
2018 | * without skipping the same | |
2019 | */ | |
a1d6cc56 | 2020 | mpd->pages_written++; |
64769240 AT |
2021 | /* |
2022 | * In error case, we have to continue because | |
2023 | * remaining pages are still locked | |
2024 | * XXX: unlock and re-dirty them? | |
2025 | */ | |
2026 | if (ret == 0) | |
2027 | ret = err; | |
2028 | } | |
2029 | pagevec_release(&pvec); | |
2030 | } | |
64769240 AT |
2031 | return ret; |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers | |
2036 | * | |
64769240 | 2037 | * the function goes through all passed space and put actual disk |
29fa89d0 | 2038 | * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten |
64769240 | 2039 | */ |
2ed88685 TT |
2040 | static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, |
2041 | struct ext4_map_blocks *map) | |
64769240 AT |
2042 | { |
2043 | struct inode *inode = mpd->inode; | |
2044 | struct address_space *mapping = inode->i_mapping; | |
2ed88685 TT |
2045 | int blocks = map->m_len; |
2046 | sector_t pblock = map->m_pblk, cur_logical; | |
64769240 | 2047 | struct buffer_head *head, *bh; |
a1d6cc56 | 2048 | pgoff_t index, end; |
64769240 AT |
2049 | struct pagevec pvec; |
2050 | int nr_pages, i; | |
2051 | ||
2ed88685 TT |
2052 | index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits); |
2053 | end = (map->m_lblk + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
64769240 AT |
2054 | cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
2055 | ||
2056 | pagevec_init(&pvec, 0); | |
2057 | ||
2058 | while (index <= end) { | |
2059 | /* XXX: optimize tail */ | |
2060 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
2061 | if (nr_pages == 0) | |
2062 | break; | |
2063 | for (i = 0; i < nr_pages; i++) { | |
2064 | struct page *page = pvec.pages[i]; | |
2065 | ||
2066 | index = page->index; | |
2067 | if (index > end) | |
2068 | break; | |
2069 | index++; | |
2070 | ||
2071 | BUG_ON(!PageLocked(page)); | |
2072 | BUG_ON(PageWriteback(page)); | |
2073 | BUG_ON(!page_has_buffers(page)); | |
2074 | ||
2075 | bh = page_buffers(page); | |
2076 | head = bh; | |
2077 | ||
2078 | /* skip blocks out of the range */ | |
2079 | do { | |
2ed88685 | 2080 | if (cur_logical >= map->m_lblk) |
64769240 AT |
2081 | break; |
2082 | cur_logical++; | |
2083 | } while ((bh = bh->b_this_page) != head); | |
2084 | ||
2085 | do { | |
2ed88685 | 2086 | if (cur_logical >= map->m_lblk + blocks) |
64769240 | 2087 | break; |
29fa89d0 | 2088 | |
2ed88685 | 2089 | if (buffer_delay(bh) || buffer_unwritten(bh)) { |
29fa89d0 AK |
2090 | |
2091 | BUG_ON(bh->b_bdev != inode->i_sb->s_bdev); | |
2092 | ||
2093 | if (buffer_delay(bh)) { | |
2094 | clear_buffer_delay(bh); | |
2095 | bh->b_blocknr = pblock; | |
2096 | } else { | |
2097 | /* | |
2098 | * unwritten already should have | |
2099 | * blocknr assigned. Verify that | |
2100 | */ | |
2101 | clear_buffer_unwritten(bh); | |
2102 | BUG_ON(bh->b_blocknr != pblock); | |
2103 | } | |
2104 | ||
61628a3f | 2105 | } else if (buffer_mapped(bh)) |
64769240 | 2106 | BUG_ON(bh->b_blocknr != pblock); |
64769240 | 2107 | |
2ed88685 | 2108 | if (map->m_flags & EXT4_MAP_UNINIT) |
744692dc | 2109 | set_buffer_uninit(bh); |
64769240 AT |
2110 | cur_logical++; |
2111 | pblock++; | |
2112 | } while ((bh = bh->b_this_page) != head); | |
2113 | } | |
2114 | pagevec_release(&pvec); | |
2115 | } | |
2116 | } | |
2117 | ||
2118 | ||
c4a0c46e AK |
2119 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, |
2120 | sector_t logical, long blk_cnt) | |
2121 | { | |
2122 | int nr_pages, i; | |
2123 | pgoff_t index, end; | |
2124 | struct pagevec pvec; | |
2125 | struct inode *inode = mpd->inode; | |
2126 | struct address_space *mapping = inode->i_mapping; | |
2127 | ||
2128 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2129 | end = (logical + blk_cnt - 1) >> | |
2130 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2131 | while (index <= end) { | |
2132 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
2133 | if (nr_pages == 0) | |
2134 | break; | |
2135 | for (i = 0; i < nr_pages; i++) { | |
2136 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 2137 | if (page->index > end) |
c4a0c46e | 2138 | break; |
c4a0c46e AK |
2139 | BUG_ON(!PageLocked(page)); |
2140 | BUG_ON(PageWriteback(page)); | |
2141 | block_invalidatepage(page, 0); | |
2142 | ClearPageUptodate(page); | |
2143 | unlock_page(page); | |
2144 | } | |
9b1d0998 JK |
2145 | index = pvec.pages[nr_pages - 1]->index + 1; |
2146 | pagevec_release(&pvec); | |
c4a0c46e AK |
2147 | } |
2148 | return; | |
2149 | } | |
2150 | ||
df22291f AK |
2151 | static void ext4_print_free_blocks(struct inode *inode) |
2152 | { | |
2153 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1693918e TT |
2154 | printk(KERN_CRIT "Total free blocks count %lld\n", |
2155 | ext4_count_free_blocks(inode->i_sb)); | |
2156 | printk(KERN_CRIT "Free/Dirty block details\n"); | |
2157 | printk(KERN_CRIT "free_blocks=%lld\n", | |
2158 | (long long) percpu_counter_sum(&sbi->s_freeblocks_counter)); | |
2159 | printk(KERN_CRIT "dirty_blocks=%lld\n", | |
2160 | (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter)); | |
2161 | printk(KERN_CRIT "Block reservation details\n"); | |
2162 | printk(KERN_CRIT "i_reserved_data_blocks=%u\n", | |
2163 | EXT4_I(inode)->i_reserved_data_blocks); | |
2164 | printk(KERN_CRIT "i_reserved_meta_blocks=%u\n", | |
2165 | EXT4_I(inode)->i_reserved_meta_blocks); | |
df22291f AK |
2166 | return; |
2167 | } | |
2168 | ||
64769240 AT |
2169 | /* |
2170 | * mpage_da_map_blocks - go through given space | |
2171 | * | |
8dc207c0 | 2172 | * @mpd - bh describing space |
64769240 AT |
2173 | * |
2174 | * The function skips space we know is already mapped to disk blocks. | |
2175 | * | |
64769240 | 2176 | */ |
ed5bde0b | 2177 | static int mpage_da_map_blocks(struct mpage_da_data *mpd) |
64769240 | 2178 | { |
2ac3b6e0 | 2179 | int err, blks, get_blocks_flags; |
2ed88685 | 2180 | struct ext4_map_blocks map; |
2fa3cdfb TT |
2181 | sector_t next = mpd->b_blocknr; |
2182 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
2183 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
2184 | handle_t *handle = NULL; | |
64769240 AT |
2185 | |
2186 | /* | |
2187 | * We consider only non-mapped and non-allocated blocks | |
2188 | */ | |
8dc207c0 | 2189 | if ((mpd->b_state & (1 << BH_Mapped)) && |
29fa89d0 AK |
2190 | !(mpd->b_state & (1 << BH_Delay)) && |
2191 | !(mpd->b_state & (1 << BH_Unwritten))) | |
c4a0c46e | 2192 | return 0; |
2fa3cdfb TT |
2193 | |
2194 | /* | |
2195 | * If we didn't accumulate anything to write simply return | |
2196 | */ | |
2197 | if (!mpd->b_size) | |
2198 | return 0; | |
2199 | ||
2200 | handle = ext4_journal_current_handle(); | |
2201 | BUG_ON(!handle); | |
2202 | ||
79ffab34 | 2203 | /* |
79e83036 | 2204 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
2205 | * blocks, or to convert an uninitialized extent to be |
2206 | * initialized (in the case where we have written into | |
2207 | * one or more preallocated blocks). | |
2208 | * | |
2209 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
2210 | * indicate that we are on the delayed allocation path. This | |
2211 | * affects functions in many different parts of the allocation | |
2212 | * call path. This flag exists primarily because we don't | |
79e83036 | 2213 | * want to change *many* call functions, so ext4_map_blocks() |
2ac3b6e0 TT |
2214 | * will set the magic i_delalloc_reserved_flag once the |
2215 | * inode's allocation semaphore is taken. | |
2216 | * | |
2217 | * If the blocks in questions were delalloc blocks, set | |
2218 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
2219 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 2220 | */ |
2ed88685 TT |
2221 | map.m_lblk = next; |
2222 | map.m_len = max_blocks; | |
1296cc85 | 2223 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE; |
744692dc JZ |
2224 | if (ext4_should_dioread_nolock(mpd->inode)) |
2225 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 2226 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
2227 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
2228 | ||
2ed88685 | 2229 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 2230 | if (blks < 0) { |
e3570639 ES |
2231 | struct super_block *sb = mpd->inode->i_sb; |
2232 | ||
2fa3cdfb | 2233 | err = blks; |
ed5bde0b TT |
2234 | /* |
2235 | * If get block returns with error we simply | |
2236 | * return. Later writepage will redirty the page and | |
2237 | * writepages will find the dirty page again | |
c4a0c46e AK |
2238 | */ |
2239 | if (err == -EAGAIN) | |
2240 | return 0; | |
df22291f AK |
2241 | |
2242 | if (err == -ENOSPC && | |
e3570639 | 2243 | ext4_count_free_blocks(sb)) { |
df22291f AK |
2244 | mpd->retval = err; |
2245 | return 0; | |
2246 | } | |
2247 | ||
c4a0c46e | 2248 | /* |
ed5bde0b TT |
2249 | * get block failure will cause us to loop in |
2250 | * writepages, because a_ops->writepage won't be able | |
2251 | * to make progress. The page will be redirtied by | |
2252 | * writepage and writepages will again try to write | |
2253 | * the same. | |
c4a0c46e | 2254 | */ |
e3570639 ES |
2255 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2256 | ext4_msg(sb, KERN_CRIT, | |
2257 | "delayed block allocation failed for inode %lu " | |
2258 | "at logical offset %llu with max blocks %zd " | |
2259 | "with error %d", mpd->inode->i_ino, | |
2260 | (unsigned long long) next, | |
2261 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
2262 | ext4_msg(sb, KERN_CRIT, | |
2263 | "This should not happen!! Data will be lost\n"); | |
2264 | if (err == -ENOSPC) | |
2265 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 2266 | } |
2fa3cdfb | 2267 | /* invalidate all the pages */ |
c4a0c46e | 2268 | ext4_da_block_invalidatepages(mpd, next, |
8dc207c0 | 2269 | mpd->b_size >> mpd->inode->i_blkbits); |
c4a0c46e AK |
2270 | return err; |
2271 | } | |
2fa3cdfb TT |
2272 | BUG_ON(blks == 0); |
2273 | ||
2ed88685 TT |
2274 | if (map.m_flags & EXT4_MAP_NEW) { |
2275 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
2276 | int i; | |
64769240 | 2277 | |
2ed88685 TT |
2278 | for (i = 0; i < map.m_len; i++) |
2279 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
2280 | } | |
64769240 | 2281 | |
a1d6cc56 AK |
2282 | /* |
2283 | * If blocks are delayed marked, we need to | |
2284 | * put actual blocknr and drop delayed bit | |
2285 | */ | |
8dc207c0 TT |
2286 | if ((mpd->b_state & (1 << BH_Delay)) || |
2287 | (mpd->b_state & (1 << BH_Unwritten))) | |
2ed88685 | 2288 | mpage_put_bnr_to_bhs(mpd, &map); |
64769240 | 2289 | |
2fa3cdfb TT |
2290 | if (ext4_should_order_data(mpd->inode)) { |
2291 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
2292 | if (err) | |
2293 | return err; | |
2294 | } | |
2295 | ||
2296 | /* | |
03f5d8bc | 2297 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
2298 | */ |
2299 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
2300 | if (disksize > i_size_read(mpd->inode)) | |
2301 | disksize = i_size_read(mpd->inode); | |
2302 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
2303 | ext4_update_i_disksize(mpd->inode, disksize); | |
2304 | return ext4_mark_inode_dirty(handle, mpd->inode); | |
2305 | } | |
2306 | ||
c4a0c46e | 2307 | return 0; |
64769240 AT |
2308 | } |
2309 | ||
bf068ee2 AK |
2310 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
2311 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
2312 | |
2313 | /* | |
2314 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
2315 | * | |
2316 | * @mpd->lbh - extent of blocks | |
2317 | * @logical - logical number of the block in the file | |
2318 | * @bh - bh of the block (used to access block's state) | |
2319 | * | |
2320 | * the function is used to collect contig. blocks in same state | |
2321 | */ | |
2322 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
2323 | sector_t logical, size_t b_size, |
2324 | unsigned long b_state) | |
64769240 | 2325 | { |
64769240 | 2326 | sector_t next; |
8dc207c0 | 2327 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 2328 | |
c445e3e0 ES |
2329 | /* |
2330 | * XXX Don't go larger than mballoc is willing to allocate | |
2331 | * This is a stopgap solution. We eventually need to fold | |
2332 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 2333 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 ES |
2334 | */ |
2335 | if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize) | |
2336 | goto flush_it; | |
2337 | ||
525f4ed8 | 2338 | /* check if thereserved journal credits might overflow */ |
12e9b892 | 2339 | if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) { |
525f4ed8 MC |
2340 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
2341 | /* | |
2342 | * With non-extent format we are limited by the journal | |
2343 | * credit available. Total credit needed to insert | |
2344 | * nrblocks contiguous blocks is dependent on the | |
2345 | * nrblocks. So limit nrblocks. | |
2346 | */ | |
2347 | goto flush_it; | |
2348 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
2349 | EXT4_MAX_TRANS_DATA) { | |
2350 | /* | |
2351 | * Adding the new buffer_head would make it cross the | |
2352 | * allowed limit for which we have journal credit | |
2353 | * reserved. So limit the new bh->b_size | |
2354 | */ | |
2355 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
2356 | mpd->inode->i_blkbits; | |
2357 | /* we will do mpage_da_submit_io in the next loop */ | |
2358 | } | |
2359 | } | |
64769240 AT |
2360 | /* |
2361 | * First block in the extent | |
2362 | */ | |
8dc207c0 TT |
2363 | if (mpd->b_size == 0) { |
2364 | mpd->b_blocknr = logical; | |
2365 | mpd->b_size = b_size; | |
2366 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
2367 | return; |
2368 | } | |
2369 | ||
8dc207c0 | 2370 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
2371 | /* |
2372 | * Can we merge the block to our big extent? | |
2373 | */ | |
8dc207c0 TT |
2374 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
2375 | mpd->b_size += b_size; | |
64769240 AT |
2376 | return; |
2377 | } | |
2378 | ||
525f4ed8 | 2379 | flush_it: |
64769240 AT |
2380 | /* |
2381 | * We couldn't merge the block to our extent, so we | |
2382 | * need to flush current extent and start new one | |
2383 | */ | |
c4a0c46e AK |
2384 | if (mpage_da_map_blocks(mpd) == 0) |
2385 | mpage_da_submit_io(mpd); | |
a1d6cc56 AK |
2386 | mpd->io_done = 1; |
2387 | return; | |
64769240 AT |
2388 | } |
2389 | ||
c364b22c | 2390 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 2391 | { |
c364b22c | 2392 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
2393 | } |
2394 | ||
64769240 AT |
2395 | /* |
2396 | * __mpage_da_writepage - finds extent of pages and blocks | |
2397 | * | |
2398 | * @page: page to consider | |
2399 | * @wbc: not used, we just follow rules | |
2400 | * @data: context | |
2401 | * | |
2402 | * The function finds extents of pages and scan them for all blocks. | |
2403 | */ | |
2404 | static int __mpage_da_writepage(struct page *page, | |
2405 | struct writeback_control *wbc, void *data) | |
2406 | { | |
2407 | struct mpage_da_data *mpd = data; | |
2408 | struct inode *inode = mpd->inode; | |
8dc207c0 | 2409 | struct buffer_head *bh, *head; |
64769240 AT |
2410 | sector_t logical; |
2411 | ||
2412 | /* | |
2413 | * Can we merge this page to current extent? | |
2414 | */ | |
2415 | if (mpd->next_page != page->index) { | |
2416 | /* | |
2417 | * Nope, we can't. So, we map non-allocated blocks | |
a1d6cc56 | 2418 | * and start IO on them using writepage() |
64769240 AT |
2419 | */ |
2420 | if (mpd->next_page != mpd->first_page) { | |
c4a0c46e AK |
2421 | if (mpage_da_map_blocks(mpd) == 0) |
2422 | mpage_da_submit_io(mpd); | |
a1d6cc56 AK |
2423 | /* |
2424 | * skip rest of the page in the page_vec | |
2425 | */ | |
2426 | mpd->io_done = 1; | |
2427 | redirty_page_for_writepage(wbc, page); | |
2428 | unlock_page(page); | |
2429 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2430 | } |
2431 | ||
2432 | /* | |
2433 | * Start next extent of pages ... | |
2434 | */ | |
2435 | mpd->first_page = page->index; | |
2436 | ||
2437 | /* | |
2438 | * ... and blocks | |
2439 | */ | |
8dc207c0 TT |
2440 | mpd->b_size = 0; |
2441 | mpd->b_state = 0; | |
2442 | mpd->b_blocknr = 0; | |
64769240 AT |
2443 | } |
2444 | ||
2445 | mpd->next_page = page->index + 1; | |
2446 | logical = (sector_t) page->index << | |
2447 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2448 | ||
2449 | if (!page_has_buffers(page)) { | |
8dc207c0 TT |
2450 | mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE, |
2451 | (1 << BH_Dirty) | (1 << BH_Uptodate)); | |
a1d6cc56 AK |
2452 | if (mpd->io_done) |
2453 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2454 | } else { |
2455 | /* | |
2456 | * Page with regular buffer heads, just add all dirty ones | |
2457 | */ | |
2458 | head = page_buffers(page); | |
2459 | bh = head; | |
2460 | do { | |
2461 | BUG_ON(buffer_locked(bh)); | |
791b7f08 AK |
2462 | /* |
2463 | * We need to try to allocate | |
2464 | * unmapped blocks in the same page. | |
2465 | * Otherwise we won't make progress | |
43ce1d23 | 2466 | * with the page in ext4_writepage |
791b7f08 | 2467 | */ |
c364b22c | 2468 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { |
8dc207c0 TT |
2469 | mpage_add_bh_to_extent(mpd, logical, |
2470 | bh->b_size, | |
2471 | bh->b_state); | |
a1d6cc56 AK |
2472 | if (mpd->io_done) |
2473 | return MPAGE_DA_EXTENT_TAIL; | |
791b7f08 AK |
2474 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2475 | /* | |
2476 | * mapped dirty buffer. We need to update | |
2477 | * the b_state because we look at | |
2478 | * b_state in mpage_da_map_blocks. We don't | |
2479 | * update b_size because if we find an | |
2480 | * unmapped buffer_head later we need to | |
2481 | * use the b_state flag of that buffer_head. | |
2482 | */ | |
8dc207c0 TT |
2483 | if (mpd->b_size == 0) |
2484 | mpd->b_state = bh->b_state & BH_FLAGS; | |
a1d6cc56 | 2485 | } |
64769240 AT |
2486 | logical++; |
2487 | } while ((bh = bh->b_this_page) != head); | |
2488 | } | |
2489 | ||
2490 | return 0; | |
2491 | } | |
2492 | ||
64769240 | 2493 | /* |
b920c755 TT |
2494 | * This is a special get_blocks_t callback which is used by |
2495 | * ext4_da_write_begin(). It will either return mapped block or | |
2496 | * reserve space for a single block. | |
29fa89d0 AK |
2497 | * |
2498 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2499 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2500 | * | |
2501 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2502 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2503 | * initialized properly. | |
64769240 AT |
2504 | */ |
2505 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2ed88685 | 2506 | struct buffer_head *bh, int create) |
64769240 | 2507 | { |
2ed88685 | 2508 | struct ext4_map_blocks map; |
64769240 | 2509 | int ret = 0; |
33b9817e AK |
2510 | sector_t invalid_block = ~((sector_t) 0xffff); |
2511 | ||
2512 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
2513 | invalid_block = ~0; | |
64769240 AT |
2514 | |
2515 | BUG_ON(create == 0); | |
2ed88685 TT |
2516 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
2517 | ||
2518 | map.m_lblk = iblock; | |
2519 | map.m_len = 1; | |
64769240 AT |
2520 | |
2521 | /* | |
2522 | * first, we need to know whether the block is allocated already | |
2523 | * preallocated blocks are unmapped but should treated | |
2524 | * the same as allocated blocks. | |
2525 | */ | |
2ed88685 TT |
2526 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
2527 | if (ret < 0) | |
2528 | return ret; | |
2529 | if (ret == 0) { | |
2530 | if (buffer_delay(bh)) | |
2531 | return 0; /* Not sure this could or should happen */ | |
64769240 AT |
2532 | /* |
2533 | * XXX: __block_prepare_write() unmaps passed block, | |
2534 | * is it OK? | |
2535 | */ | |
9d0be502 | 2536 | ret = ext4_da_reserve_space(inode, iblock); |
d2a17637 MC |
2537 | if (ret) |
2538 | /* not enough space to reserve */ | |
2539 | return ret; | |
2540 | ||
2ed88685 TT |
2541 | map_bh(bh, inode->i_sb, invalid_block); |
2542 | set_buffer_new(bh); | |
2543 | set_buffer_delay(bh); | |
2544 | return 0; | |
64769240 AT |
2545 | } |
2546 | ||
2ed88685 TT |
2547 | map_bh(bh, inode->i_sb, map.m_pblk); |
2548 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
2549 | ||
2550 | if (buffer_unwritten(bh)) { | |
2551 | /* A delayed write to unwritten bh should be marked | |
2552 | * new and mapped. Mapped ensures that we don't do | |
2553 | * get_block multiple times when we write to the same | |
2554 | * offset and new ensures that we do proper zero out | |
2555 | * for partial write. | |
2556 | */ | |
2557 | set_buffer_new(bh); | |
2558 | set_buffer_mapped(bh); | |
2559 | } | |
2560 | return 0; | |
64769240 | 2561 | } |
61628a3f | 2562 | |
b920c755 TT |
2563 | /* |
2564 | * This function is used as a standard get_block_t calback function | |
2565 | * when there is no desire to allocate any blocks. It is used as a | |
206f7ab4 CH |
2566 | * callback function for block_prepare_write() and block_write_full_page(). |
2567 | * These functions should only try to map a single block at a time. | |
b920c755 TT |
2568 | * |
2569 | * Since this function doesn't do block allocations even if the caller | |
2570 | * requests it by passing in create=1, it is critically important that | |
2571 | * any caller checks to make sure that any buffer heads are returned | |
2572 | * by this function are either all already mapped or marked for | |
206f7ab4 CH |
2573 | * delayed allocation before calling block_write_full_page(). Otherwise, |
2574 | * b_blocknr could be left unitialized, and the page write functions will | |
2575 | * be taken by surprise. | |
b920c755 TT |
2576 | */ |
2577 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
2578 | struct buffer_head *bh_result, int create) |
2579 | { | |
a2dc52b5 | 2580 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2ed88685 | 2581 | return _ext4_get_block(inode, iblock, bh_result, 0); |
61628a3f MC |
2582 | } |
2583 | ||
62e086be AK |
2584 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
2585 | { | |
2586 | get_bh(bh); | |
2587 | return 0; | |
2588 | } | |
2589 | ||
2590 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2591 | { | |
2592 | put_bh(bh); | |
2593 | return 0; | |
2594 | } | |
2595 | ||
2596 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
2597 | unsigned int len) |
2598 | { | |
2599 | struct address_space *mapping = page->mapping; | |
2600 | struct inode *inode = mapping->host; | |
2601 | struct buffer_head *page_bufs; | |
2602 | handle_t *handle = NULL; | |
2603 | int ret = 0; | |
2604 | int err; | |
2605 | ||
2606 | page_bufs = page_buffers(page); | |
2607 | BUG_ON(!page_bufs); | |
2608 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one); | |
2609 | /* As soon as we unlock the page, it can go away, but we have | |
2610 | * references to buffers so we are safe */ | |
2611 | unlock_page(page); | |
2612 | ||
2613 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
2614 | if (IS_ERR(handle)) { | |
2615 | ret = PTR_ERR(handle); | |
2616 | goto out; | |
2617 | } | |
2618 | ||
2619 | ret = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2620 | do_journal_get_write_access); | |
2621 | ||
2622 | err = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2623 | write_end_fn); | |
2624 | if (ret == 0) | |
2625 | ret = err; | |
2626 | err = ext4_journal_stop(handle); | |
2627 | if (!ret) | |
2628 | ret = err; | |
2629 | ||
2630 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one); | |
19f5fb7a | 2631 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be AK |
2632 | out: |
2633 | return ret; | |
2634 | } | |
2635 | ||
744692dc JZ |
2636 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); |
2637 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
2638 | ||
61628a3f | 2639 | /* |
43ce1d23 AK |
2640 | * Note that we don't need to start a transaction unless we're journaling data |
2641 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2642 | * need to file the inode to the transaction's list in ordered mode because if | |
2643 | * we are writing back data added by write(), the inode is already there and if | |
2644 | * we are writing back data modified via mmap(), noone guarantees in which | |
2645 | * transaction the data will hit the disk. In case we are journaling data, we | |
2646 | * cannot start transaction directly because transaction start ranks above page | |
2647 | * lock so we have to do some magic. | |
2648 | * | |
b920c755 TT |
2649 | * This function can get called via... |
2650 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2651 | * - journal_submit_inode_data_buffers (no journal handle) | |
2652 | * - shrink_page_list via pdflush (no journal handle) | |
2653 | * - grab_page_cache when doing write_begin (have journal handle) | |
43ce1d23 AK |
2654 | * |
2655 | * We don't do any block allocation in this function. If we have page with | |
2656 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2657 | * is important for mmaped based write. So if we do with blocksize 1K | |
2658 | * truncate(f, 1024); | |
2659 | * a = mmap(f, 0, 4096); | |
2660 | * a[0] = 'a'; | |
2661 | * truncate(f, 4096); | |
2662 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
2663 | * but other bufer_heads would be unmapped but dirty(dirty done via the | |
2664 | * do_wp_page). So writepage should write the first block. If we modify | |
2665 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2666 | * page_mkwrite callback will do the block allocation and mark the | |
2667 | * buffer_heads mapped. | |
2668 | * | |
2669 | * We redirty the page if we have any buffer_heads that is either delay or | |
2670 | * unwritten in the page. | |
2671 | * | |
2672 | * We can get recursively called as show below. | |
2673 | * | |
2674 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2675 | * ext4_writepage() | |
2676 | * | |
2677 | * But since we don't do any block allocation we should not deadlock. | |
2678 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2679 | */ |
43ce1d23 | 2680 | static int ext4_writepage(struct page *page, |
62e086be | 2681 | struct writeback_control *wbc) |
64769240 | 2682 | { |
64769240 | 2683 | int ret = 0; |
61628a3f | 2684 | loff_t size; |
498e5f24 | 2685 | unsigned int len; |
744692dc | 2686 | struct buffer_head *page_bufs = NULL; |
61628a3f MC |
2687 | struct inode *inode = page->mapping->host; |
2688 | ||
43ce1d23 | 2689 | trace_ext4_writepage(inode, page); |
f0e6c985 AK |
2690 | size = i_size_read(inode); |
2691 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2692 | len = size & ~PAGE_CACHE_MASK; | |
2693 | else | |
2694 | len = PAGE_CACHE_SIZE; | |
64769240 | 2695 | |
f0e6c985 | 2696 | if (page_has_buffers(page)) { |
61628a3f | 2697 | page_bufs = page_buffers(page); |
f0e6c985 | 2698 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
c364b22c | 2699 | ext4_bh_delay_or_unwritten)) { |
61628a3f | 2700 | /* |
f0e6c985 AK |
2701 | * We don't want to do block allocation |
2702 | * So redirty the page and return | |
cd1aac32 AK |
2703 | * We may reach here when we do a journal commit |
2704 | * via journal_submit_inode_data_buffers. | |
2705 | * If we don't have mapping block we just ignore | |
f0e6c985 AK |
2706 | * them. We can also reach here via shrink_page_list |
2707 | */ | |
2708 | redirty_page_for_writepage(wbc, page); | |
2709 | unlock_page(page); | |
2710 | return 0; | |
2711 | } | |
2712 | } else { | |
2713 | /* | |
2714 | * The test for page_has_buffers() is subtle: | |
2715 | * We know the page is dirty but it lost buffers. That means | |
2716 | * that at some moment in time after write_begin()/write_end() | |
2717 | * has been called all buffers have been clean and thus they | |
2718 | * must have been written at least once. So they are all | |
2719 | * mapped and we can happily proceed with mapping them | |
2720 | * and writing the page. | |
2721 | * | |
2722 | * Try to initialize the buffer_heads and check whether | |
2723 | * all are mapped and non delay. We don't want to | |
2724 | * do block allocation here. | |
2725 | */ | |
b767e78a | 2726 | ret = block_prepare_write(page, 0, len, |
b920c755 | 2727 | noalloc_get_block_write); |
f0e6c985 AK |
2728 | if (!ret) { |
2729 | page_bufs = page_buffers(page); | |
2730 | /* check whether all are mapped and non delay */ | |
2731 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
c364b22c | 2732 | ext4_bh_delay_or_unwritten)) { |
f0e6c985 AK |
2733 | redirty_page_for_writepage(wbc, page); |
2734 | unlock_page(page); | |
2735 | return 0; | |
2736 | } | |
2737 | } else { | |
2738 | /* | |
2739 | * We can't do block allocation here | |
2740 | * so just redity the page and unlock | |
2741 | * and return | |
61628a3f | 2742 | */ |
61628a3f MC |
2743 | redirty_page_for_writepage(wbc, page); |
2744 | unlock_page(page); | |
2745 | return 0; | |
2746 | } | |
ed9b3e33 | 2747 | /* now mark the buffer_heads as dirty and uptodate */ |
b767e78a | 2748 | block_commit_write(page, 0, len); |
64769240 AT |
2749 | } |
2750 | ||
43ce1d23 AK |
2751 | if (PageChecked(page) && ext4_should_journal_data(inode)) { |
2752 | /* | |
2753 | * It's mmapped pagecache. Add buffers and journal it. There | |
2754 | * doesn't seem much point in redirtying the page here. | |
2755 | */ | |
2756 | ClearPageChecked(page); | |
3f0ca309 | 2757 | return __ext4_journalled_writepage(page, len); |
43ce1d23 AK |
2758 | } |
2759 | ||
206f7ab4 | 2760 | if (page_bufs && buffer_uninit(page_bufs)) { |
744692dc JZ |
2761 | ext4_set_bh_endio(page_bufs, inode); |
2762 | ret = block_write_full_page_endio(page, noalloc_get_block_write, | |
2763 | wbc, ext4_end_io_buffer_write); | |
2764 | } else | |
b920c755 TT |
2765 | ret = block_write_full_page(page, noalloc_get_block_write, |
2766 | wbc); | |
64769240 | 2767 | |
64769240 AT |
2768 | return ret; |
2769 | } | |
2770 | ||
61628a3f | 2771 | /* |
525f4ed8 MC |
2772 | * This is called via ext4_da_writepages() to |
2773 | * calulate the total number of credits to reserve to fit | |
2774 | * a single extent allocation into a single transaction, | |
2775 | * ext4_da_writpeages() will loop calling this before | |
2776 | * the block allocation. | |
61628a3f | 2777 | */ |
525f4ed8 MC |
2778 | |
2779 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2780 | { | |
2781 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2782 | ||
2783 | /* | |
2784 | * With non-extent format the journal credit needed to | |
2785 | * insert nrblocks contiguous block is dependent on | |
2786 | * number of contiguous block. So we will limit | |
2787 | * number of contiguous block to a sane value | |
2788 | */ | |
12e9b892 | 2789 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
2790 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2791 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2792 | ||
2793 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2794 | } | |
61628a3f | 2795 | |
8e48dcfb TT |
2796 | /* |
2797 | * write_cache_pages_da - walk the list of dirty pages of the given | |
2798 | * address space and call the callback function (which usually writes | |
2799 | * the pages). | |
2800 | * | |
2801 | * This is a forked version of write_cache_pages(). Differences: | |
2802 | * Range cyclic is ignored. | |
2803 | * no_nrwrite_index_update is always presumed true | |
2804 | */ | |
2805 | static int write_cache_pages_da(struct address_space *mapping, | |
2806 | struct writeback_control *wbc, | |
2807 | struct mpage_da_data *mpd) | |
2808 | { | |
2809 | int ret = 0; | |
2810 | int done = 0; | |
2811 | struct pagevec pvec; | |
2812 | int nr_pages; | |
2813 | pgoff_t index; | |
2814 | pgoff_t end; /* Inclusive */ | |
2815 | long nr_to_write = wbc->nr_to_write; | |
2816 | ||
2817 | pagevec_init(&pvec, 0); | |
2818 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2819 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2820 | ||
2821 | while (!done && (index <= end)) { | |
2822 | int i; | |
2823 | ||
2824 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
2825 | PAGECACHE_TAG_DIRTY, | |
2826 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | |
2827 | if (nr_pages == 0) | |
2828 | break; | |
2829 | ||
2830 | for (i = 0; i < nr_pages; i++) { | |
2831 | struct page *page = pvec.pages[i]; | |
2832 | ||
2833 | /* | |
2834 | * At this point, the page may be truncated or | |
2835 | * invalidated (changing page->mapping to NULL), or | |
2836 | * even swizzled back from swapper_space to tmpfs file | |
2837 | * mapping. However, page->index will not change | |
2838 | * because we have a reference on the page. | |
2839 | */ | |
2840 | if (page->index > end) { | |
2841 | done = 1; | |
2842 | break; | |
2843 | } | |
2844 | ||
2845 | lock_page(page); | |
2846 | ||
2847 | /* | |
2848 | * Page truncated or invalidated. We can freely skip it | |
2849 | * then, even for data integrity operations: the page | |
2850 | * has disappeared concurrently, so there could be no | |
2851 | * real expectation of this data interity operation | |
2852 | * even if there is now a new, dirty page at the same | |
2853 | * pagecache address. | |
2854 | */ | |
2855 | if (unlikely(page->mapping != mapping)) { | |
2856 | continue_unlock: | |
2857 | unlock_page(page); | |
2858 | continue; | |
2859 | } | |
2860 | ||
2861 | if (!PageDirty(page)) { | |
2862 | /* someone wrote it for us */ | |
2863 | goto continue_unlock; | |
2864 | } | |
2865 | ||
2866 | if (PageWriteback(page)) { | |
2867 | if (wbc->sync_mode != WB_SYNC_NONE) | |
2868 | wait_on_page_writeback(page); | |
2869 | else | |
2870 | goto continue_unlock; | |
2871 | } | |
2872 | ||
2873 | BUG_ON(PageWriteback(page)); | |
2874 | if (!clear_page_dirty_for_io(page)) | |
2875 | goto continue_unlock; | |
2876 | ||
2877 | ret = __mpage_da_writepage(page, wbc, mpd); | |
2878 | if (unlikely(ret)) { | |
2879 | if (ret == AOP_WRITEPAGE_ACTIVATE) { | |
2880 | unlock_page(page); | |
2881 | ret = 0; | |
2882 | } else { | |
2883 | done = 1; | |
2884 | break; | |
2885 | } | |
2886 | } | |
2887 | ||
2888 | if (nr_to_write > 0) { | |
2889 | nr_to_write--; | |
2890 | if (nr_to_write == 0 && | |
2891 | wbc->sync_mode == WB_SYNC_NONE) { | |
2892 | /* | |
2893 | * We stop writing back only if we are | |
2894 | * not doing integrity sync. In case of | |
2895 | * integrity sync we have to keep going | |
2896 | * because someone may be concurrently | |
2897 | * dirtying pages, and we might have | |
2898 | * synced a lot of newly appeared dirty | |
2899 | * pages, but have not synced all of the | |
2900 | * old dirty pages. | |
2901 | */ | |
2902 | done = 1; | |
2903 | break; | |
2904 | } | |
2905 | } | |
2906 | } | |
2907 | pagevec_release(&pvec); | |
2908 | cond_resched(); | |
2909 | } | |
2910 | return ret; | |
2911 | } | |
2912 | ||
2913 | ||
64769240 | 2914 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2915 | struct writeback_control *wbc) |
64769240 | 2916 | { |
22208ded AK |
2917 | pgoff_t index; |
2918 | int range_whole = 0; | |
61628a3f | 2919 | handle_t *handle = NULL; |
df22291f | 2920 | struct mpage_da_data mpd; |
5e745b04 | 2921 | struct inode *inode = mapping->host; |
498e5f24 TT |
2922 | int pages_written = 0; |
2923 | long pages_skipped; | |
55138e0b | 2924 | unsigned int max_pages; |
2acf2c26 | 2925 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2926 | int needed_blocks, ret = 0; |
2927 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2928 | loff_t range_start = wbc->range_start; |
5e745b04 | 2929 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
61628a3f | 2930 | |
9bffad1e | 2931 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2932 | |
61628a3f MC |
2933 | /* |
2934 | * No pages to write? This is mainly a kludge to avoid starting | |
2935 | * a transaction for special inodes like journal inode on last iput() | |
2936 | * because that could violate lock ordering on umount | |
2937 | */ | |
a1d6cc56 | 2938 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2939 | return 0; |
2a21e37e TT |
2940 | |
2941 | /* | |
2942 | * If the filesystem has aborted, it is read-only, so return | |
2943 | * right away instead of dumping stack traces later on that | |
2944 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2945 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2946 | * the latter could be true if the filesystem is mounted |
2947 | * read-only, and in that case, ext4_da_writepages should | |
2948 | * *never* be called, so if that ever happens, we would want | |
2949 | * the stack trace. | |
2950 | */ | |
4ab2f15b | 2951 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2952 | return -EROFS; |
2953 | ||
22208ded AK |
2954 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2955 | range_whole = 1; | |
61628a3f | 2956 | |
2acf2c26 AK |
2957 | range_cyclic = wbc->range_cyclic; |
2958 | if (wbc->range_cyclic) { | |
22208ded | 2959 | index = mapping->writeback_index; |
2acf2c26 AK |
2960 | if (index) |
2961 | cycled = 0; | |
2962 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2963 | wbc->range_end = LLONG_MAX; | |
2964 | wbc->range_cyclic = 0; | |
2965 | } else | |
22208ded | 2966 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
a1d6cc56 | 2967 | |
55138e0b TT |
2968 | /* |
2969 | * This works around two forms of stupidity. The first is in | |
2970 | * the writeback code, which caps the maximum number of pages | |
2971 | * written to be 1024 pages. This is wrong on multiple | |
2972 | * levels; different architectues have a different page size, | |
2973 | * which changes the maximum amount of data which gets | |
2974 | * written. Secondly, 4 megabytes is way too small. XFS | |
2975 | * forces this value to be 16 megabytes by multiplying | |
2976 | * nr_to_write parameter by four, and then relies on its | |
2977 | * allocator to allocate larger extents to make them | |
2978 | * contiguous. Unfortunately this brings us to the second | |
2979 | * stupidity, which is that ext4's mballoc code only allocates | |
2980 | * at most 2048 blocks. So we force contiguous writes up to | |
2981 | * the number of dirty blocks in the inode, or | |
2982 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2983 | */ | |
2984 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
2985 | if (!range_cyclic && range_whole) | |
2986 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2987 | else | |
2988 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, | |
2989 | max_pages); | |
2990 | if (desired_nr_to_write > max_pages) | |
2991 | desired_nr_to_write = max_pages; | |
2992 | ||
2993 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2994 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2995 | wbc->nr_to_write = desired_nr_to_write; | |
2996 | } | |
2997 | ||
df22291f AK |
2998 | mpd.wbc = wbc; |
2999 | mpd.inode = mapping->host; | |
3000 | ||
22208ded AK |
3001 | pages_skipped = wbc->pages_skipped; |
3002 | ||
2acf2c26 | 3003 | retry: |
22208ded | 3004 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
3005 | |
3006 | /* | |
3007 | * we insert one extent at a time. So we need | |
3008 | * credit needed for single extent allocation. | |
3009 | * journalled mode is currently not supported | |
3010 | * by delalloc | |
3011 | */ | |
3012 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 3013 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 3014 | |
61628a3f MC |
3015 | /* start a new transaction*/ |
3016 | handle = ext4_journal_start(inode, needed_blocks); | |
3017 | if (IS_ERR(handle)) { | |
3018 | ret = PTR_ERR(handle); | |
1693918e | 3019 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 3020 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 3021 | wbc->nr_to_write, inode->i_ino, ret); |
61628a3f MC |
3022 | goto out_writepages; |
3023 | } | |
f63e6005 TT |
3024 | |
3025 | /* | |
3026 | * Now call __mpage_da_writepage to find the next | |
3027 | * contiguous region of logical blocks that need | |
3028 | * blocks to be allocated by ext4. We don't actually | |
3029 | * submit the blocks for I/O here, even though | |
3030 | * write_cache_pages thinks it will, and will set the | |
3031 | * pages as clean for write before calling | |
3032 | * __mpage_da_writepage(). | |
3033 | */ | |
3034 | mpd.b_size = 0; | |
3035 | mpd.b_state = 0; | |
3036 | mpd.b_blocknr = 0; | |
3037 | mpd.first_page = 0; | |
3038 | mpd.next_page = 0; | |
3039 | mpd.io_done = 0; | |
3040 | mpd.pages_written = 0; | |
3041 | mpd.retval = 0; | |
8e48dcfb | 3042 | ret = write_cache_pages_da(mapping, wbc, &mpd); |
f63e6005 | 3043 | /* |
af901ca1 | 3044 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
3045 | * haven't done the I/O yet, map the blocks and submit |
3046 | * them for I/O. | |
3047 | */ | |
3048 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
3049 | if (mpage_da_map_blocks(&mpd) == 0) | |
3050 | mpage_da_submit_io(&mpd); | |
3051 | mpd.io_done = 1; | |
3052 | ret = MPAGE_DA_EXTENT_TAIL; | |
3053 | } | |
b3a3ca8c | 3054 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 3055 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 3056 | |
61628a3f | 3057 | ext4_journal_stop(handle); |
df22291f | 3058 | |
8f64b32e | 3059 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
3060 | /* commit the transaction which would |
3061 | * free blocks released in the transaction | |
3062 | * and try again | |
3063 | */ | |
df22291f | 3064 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
3065 | wbc->pages_skipped = pages_skipped; |
3066 | ret = 0; | |
3067 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 AK |
3068 | /* |
3069 | * got one extent now try with | |
3070 | * rest of the pages | |
3071 | */ | |
22208ded AK |
3072 | pages_written += mpd.pages_written; |
3073 | wbc->pages_skipped = pages_skipped; | |
a1d6cc56 | 3074 | ret = 0; |
2acf2c26 | 3075 | io_done = 1; |
22208ded | 3076 | } else if (wbc->nr_to_write) |
61628a3f MC |
3077 | /* |
3078 | * There is no more writeout needed | |
3079 | * or we requested for a noblocking writeout | |
3080 | * and we found the device congested | |
3081 | */ | |
61628a3f | 3082 | break; |
a1d6cc56 | 3083 | } |
2acf2c26 AK |
3084 | if (!io_done && !cycled) { |
3085 | cycled = 1; | |
3086 | index = 0; | |
3087 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
3088 | wbc->range_end = mapping->writeback_index - 1; | |
3089 | goto retry; | |
3090 | } | |
22208ded | 3091 | if (pages_skipped != wbc->pages_skipped) |
1693918e TT |
3092 | ext4_msg(inode->i_sb, KERN_CRIT, |
3093 | "This should not happen leaving %s " | |
fbe845dd | 3094 | "with nr_to_write = %ld ret = %d", |
1693918e | 3095 | __func__, wbc->nr_to_write, ret); |
22208ded AK |
3096 | |
3097 | /* Update index */ | |
3098 | index += pages_written; | |
2acf2c26 | 3099 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
3100 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
3101 | /* | |
3102 | * set the writeback_index so that range_cyclic | |
3103 | * mode will write it back later | |
3104 | */ | |
3105 | mapping->writeback_index = index; | |
a1d6cc56 | 3106 | |
61628a3f | 3107 | out_writepages: |
2faf2e19 | 3108 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 3109 | wbc->range_start = range_start; |
9bffad1e | 3110 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 3111 | return ret; |
64769240 AT |
3112 | } |
3113 | ||
79f0be8d AK |
3114 | #define FALL_BACK_TO_NONDELALLOC 1 |
3115 | static int ext4_nonda_switch(struct super_block *sb) | |
3116 | { | |
3117 | s64 free_blocks, dirty_blocks; | |
3118 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
3119 | ||
3120 | /* | |
3121 | * switch to non delalloc mode if we are running low | |
3122 | * on free block. The free block accounting via percpu | |
179f7ebf | 3123 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
3124 | * accumulated on each CPU without updating global counters |
3125 | * Delalloc need an accurate free block accounting. So switch | |
3126 | * to non delalloc when we are near to error range. | |
3127 | */ | |
3128 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3129 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter); | |
3130 | if (2 * free_blocks < 3 * dirty_blocks || | |
3131 | free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) { | |
3132 | /* | |
c8afb446 ES |
3133 | * free block count is less than 150% of dirty blocks |
3134 | * or free blocks is less than watermark | |
79f0be8d AK |
3135 | */ |
3136 | return 1; | |
3137 | } | |
c8afb446 ES |
3138 | /* |
3139 | * Even if we don't switch but are nearing capacity, | |
3140 | * start pushing delalloc when 1/2 of free blocks are dirty. | |
3141 | */ | |
3142 | if (free_blocks < 2 * dirty_blocks) | |
3143 | writeback_inodes_sb_if_idle(sb); | |
3144 | ||
79f0be8d AK |
3145 | return 0; |
3146 | } | |
3147 | ||
64769240 | 3148 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
3149 | loff_t pos, unsigned len, unsigned flags, |
3150 | struct page **pagep, void **fsdata) | |
64769240 | 3151 | { |
72b8ab9d | 3152 | int ret, retries = 0; |
64769240 AT |
3153 | struct page *page; |
3154 | pgoff_t index; | |
64769240 AT |
3155 | struct inode *inode = mapping->host; |
3156 | handle_t *handle; | |
3157 | ||
3158 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
3159 | |
3160 | if (ext4_nonda_switch(inode->i_sb)) { | |
3161 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
3162 | return ext4_write_begin(file, mapping, pos, | |
3163 | len, flags, pagep, fsdata); | |
3164 | } | |
3165 | *fsdata = (void *)0; | |
9bffad1e | 3166 | trace_ext4_da_write_begin(inode, pos, len, flags); |
d2a17637 | 3167 | retry: |
64769240 AT |
3168 | /* |
3169 | * With delayed allocation, we don't log the i_disksize update | |
3170 | * if there is delayed block allocation. But we still need | |
3171 | * to journalling the i_disksize update if writes to the end | |
3172 | * of file which has an already mapped buffer. | |
3173 | */ | |
3174 | handle = ext4_journal_start(inode, 1); | |
3175 | if (IS_ERR(handle)) { | |
3176 | ret = PTR_ERR(handle); | |
3177 | goto out; | |
3178 | } | |
ebd3610b JK |
3179 | /* We cannot recurse into the filesystem as the transaction is already |
3180 | * started */ | |
3181 | flags |= AOP_FLAG_NOFS; | |
64769240 | 3182 | |
54566b2c | 3183 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
3184 | if (!page) { |
3185 | ext4_journal_stop(handle); | |
3186 | ret = -ENOMEM; | |
3187 | goto out; | |
3188 | } | |
64769240 AT |
3189 | *pagep = page; |
3190 | ||
3191 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
b920c755 | 3192 | ext4_da_get_block_prep); |
64769240 AT |
3193 | if (ret < 0) { |
3194 | unlock_page(page); | |
3195 | ext4_journal_stop(handle); | |
3196 | page_cache_release(page); | |
ae4d5372 AK |
3197 | /* |
3198 | * block_write_begin may have instantiated a few blocks | |
3199 | * outside i_size. Trim these off again. Don't need | |
3200 | * i_size_read because we hold i_mutex. | |
3201 | */ | |
3202 | if (pos + len > inode->i_size) | |
b9a4207d | 3203 | ext4_truncate_failed_write(inode); |
64769240 AT |
3204 | } |
3205 | ||
d2a17637 MC |
3206 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
3207 | goto retry; | |
64769240 AT |
3208 | out: |
3209 | return ret; | |
3210 | } | |
3211 | ||
632eaeab MC |
3212 | /* |
3213 | * Check if we should update i_disksize | |
3214 | * when write to the end of file but not require block allocation | |
3215 | */ | |
3216 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3217 | unsigned long offset) |
632eaeab MC |
3218 | { |
3219 | struct buffer_head *bh; | |
3220 | struct inode *inode = page->mapping->host; | |
3221 | unsigned int idx; | |
3222 | int i; | |
3223 | ||
3224 | bh = page_buffers(page); | |
3225 | idx = offset >> inode->i_blkbits; | |
3226 | ||
af5bc92d | 3227 | for (i = 0; i < idx; i++) |
632eaeab MC |
3228 | bh = bh->b_this_page; |
3229 | ||
29fa89d0 | 3230 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3231 | return 0; |
3232 | return 1; | |
3233 | } | |
3234 | ||
64769240 | 3235 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3236 | struct address_space *mapping, |
3237 | loff_t pos, unsigned len, unsigned copied, | |
3238 | struct page *page, void *fsdata) | |
64769240 AT |
3239 | { |
3240 | struct inode *inode = mapping->host; | |
3241 | int ret = 0, ret2; | |
3242 | handle_t *handle = ext4_journal_current_handle(); | |
3243 | loff_t new_i_size; | |
632eaeab | 3244 | unsigned long start, end; |
79f0be8d AK |
3245 | int write_mode = (int)(unsigned long)fsdata; |
3246 | ||
3247 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
3248 | if (ext4_should_order_data(inode)) { | |
3249 | return ext4_ordered_write_end(file, mapping, pos, | |
3250 | len, copied, page, fsdata); | |
3251 | } else if (ext4_should_writeback_data(inode)) { | |
3252 | return ext4_writeback_write_end(file, mapping, pos, | |
3253 | len, copied, page, fsdata); | |
3254 | } else { | |
3255 | BUG(); | |
3256 | } | |
3257 | } | |
632eaeab | 3258 | |
9bffad1e | 3259 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 3260 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 3261 | end = start + copied - 1; |
64769240 AT |
3262 | |
3263 | /* | |
3264 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3265 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3266 | * into that. | |
3267 | */ | |
3268 | ||
3269 | new_i_size = pos + copied; | |
632eaeab MC |
3270 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
3271 | if (ext4_da_should_update_i_disksize(page, end)) { | |
3272 | down_write(&EXT4_I(inode)->i_data_sem); | |
3273 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
3274 | /* | |
3275 | * Updating i_disksize when extending file | |
3276 | * without needing block allocation | |
3277 | */ | |
3278 | if (ext4_should_order_data(inode)) | |
3279 | ret = ext4_jbd2_file_inode(handle, | |
3280 | inode); | |
64769240 | 3281 | |
632eaeab MC |
3282 | EXT4_I(inode)->i_disksize = new_i_size; |
3283 | } | |
3284 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
3285 | /* We need to mark inode dirty even if |
3286 | * new_i_size is less that inode->i_size | |
3287 | * bu greater than i_disksize.(hint delalloc) | |
3288 | */ | |
3289 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3290 | } |
632eaeab | 3291 | } |
64769240 AT |
3292 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
3293 | page, fsdata); | |
3294 | copied = ret2; | |
3295 | if (ret2 < 0) | |
3296 | ret = ret2; | |
3297 | ret2 = ext4_journal_stop(handle); | |
3298 | if (!ret) | |
3299 | ret = ret2; | |
3300 | ||
3301 | return ret ? ret : copied; | |
3302 | } | |
3303 | ||
3304 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
3305 | { | |
64769240 AT |
3306 | /* |
3307 | * Drop reserved blocks | |
3308 | */ | |
3309 | BUG_ON(!PageLocked(page)); | |
3310 | if (!page_has_buffers(page)) | |
3311 | goto out; | |
3312 | ||
d2a17637 | 3313 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
3314 | |
3315 | out: | |
3316 | ext4_invalidatepage(page, offset); | |
3317 | ||
3318 | return; | |
3319 | } | |
3320 | ||
ccd2506b TT |
3321 | /* |
3322 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3323 | */ | |
3324 | int ext4_alloc_da_blocks(struct inode *inode) | |
3325 | { | |
fb40ba0d TT |
3326 | trace_ext4_alloc_da_blocks(inode); |
3327 | ||
ccd2506b TT |
3328 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
3329 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
3330 | return 0; | |
3331 | ||
3332 | /* | |
3333 | * We do something simple for now. The filemap_flush() will | |
3334 | * also start triggering a write of the data blocks, which is | |
3335 | * not strictly speaking necessary (and for users of | |
3336 | * laptop_mode, not even desirable). However, to do otherwise | |
3337 | * would require replicating code paths in: | |
de9a55b8 | 3338 | * |
ccd2506b TT |
3339 | * ext4_da_writepages() -> |
3340 | * write_cache_pages() ---> (via passed in callback function) | |
3341 | * __mpage_da_writepage() --> | |
3342 | * mpage_add_bh_to_extent() | |
3343 | * mpage_da_map_blocks() | |
3344 | * | |
3345 | * The problem is that write_cache_pages(), located in | |
3346 | * mm/page-writeback.c, marks pages clean in preparation for | |
3347 | * doing I/O, which is not desirable if we're not planning on | |
3348 | * doing I/O at all. | |
3349 | * | |
3350 | * We could call write_cache_pages(), and then redirty all of | |
3351 | * the pages by calling redirty_page_for_writeback() but that | |
3352 | * would be ugly in the extreme. So instead we would need to | |
3353 | * replicate parts of the code in the above functions, | |
3354 | * simplifying them becuase we wouldn't actually intend to | |
3355 | * write out the pages, but rather only collect contiguous | |
3356 | * logical block extents, call the multi-block allocator, and | |
3357 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3358 | * |
ccd2506b TT |
3359 | * For now, though, we'll cheat by calling filemap_flush(), |
3360 | * which will map the blocks, and start the I/O, but not | |
3361 | * actually wait for the I/O to complete. | |
3362 | */ | |
3363 | return filemap_flush(inode->i_mapping); | |
3364 | } | |
64769240 | 3365 | |
ac27a0ec DK |
3366 | /* |
3367 | * bmap() is special. It gets used by applications such as lilo and by | |
3368 | * the swapper to find the on-disk block of a specific piece of data. | |
3369 | * | |
3370 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3371 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3372 | * filesystem and enables swap, then they may get a nasty shock when the |
3373 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3374 | * the original zero's written out previously to the journal and | |
3375 | * awaiting writeback in the kernel's buffer cache. | |
3376 | * | |
3377 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3378 | * take extra steps to flush any blocks which might be in the cache. | |
3379 | */ | |
617ba13b | 3380 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3381 | { |
3382 | struct inode *inode = mapping->host; | |
3383 | journal_t *journal; | |
3384 | int err; | |
3385 | ||
64769240 AT |
3386 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3387 | test_opt(inode->i_sb, DELALLOC)) { | |
3388 | /* | |
3389 | * With delalloc we want to sync the file | |
3390 | * so that we can make sure we allocate | |
3391 | * blocks for file | |
3392 | */ | |
3393 | filemap_write_and_wait(mapping); | |
3394 | } | |
3395 | ||
19f5fb7a TT |
3396 | if (EXT4_JOURNAL(inode) && |
3397 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3398 | /* |
3399 | * This is a REALLY heavyweight approach, but the use of | |
3400 | * bmap on dirty files is expected to be extremely rare: | |
3401 | * only if we run lilo or swapon on a freshly made file | |
3402 | * do we expect this to happen. | |
3403 | * | |
3404 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3405 | * represent an unprivileged user DOS attack --- we'd be | |
3406 | * in trouble if mortal users could trigger this path at | |
3407 | * will.) | |
3408 | * | |
617ba13b | 3409 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3410 | * regular files. If somebody wants to bmap a directory |
3411 | * or symlink and gets confused because the buffer | |
3412 | * hasn't yet been flushed to disk, they deserve | |
3413 | * everything they get. | |
3414 | */ | |
3415 | ||
19f5fb7a | 3416 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3417 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3418 | jbd2_journal_lock_updates(journal); |
3419 | err = jbd2_journal_flush(journal); | |
3420 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3421 | |
3422 | if (err) | |
3423 | return 0; | |
3424 | } | |
3425 | ||
af5bc92d | 3426 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3427 | } |
3428 | ||
617ba13b | 3429 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3430 | { |
617ba13b | 3431 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
3432 | } |
3433 | ||
3434 | static int | |
617ba13b | 3435 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3436 | struct list_head *pages, unsigned nr_pages) |
3437 | { | |
617ba13b | 3438 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
3439 | } |
3440 | ||
744692dc JZ |
3441 | static void ext4_free_io_end(ext4_io_end_t *io) |
3442 | { | |
3443 | BUG_ON(!io); | |
3444 | if (io->page) | |
3445 | put_page(io->page); | |
3446 | iput(io->inode); | |
3447 | kfree(io); | |
3448 | } | |
3449 | ||
3450 | static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset) | |
3451 | { | |
3452 | struct buffer_head *head, *bh; | |
3453 | unsigned int curr_off = 0; | |
3454 | ||
3455 | if (!page_has_buffers(page)) | |
3456 | return; | |
3457 | head = bh = page_buffers(page); | |
3458 | do { | |
3459 | if (offset <= curr_off && test_clear_buffer_uninit(bh) | |
3460 | && bh->b_private) { | |
3461 | ext4_free_io_end(bh->b_private); | |
3462 | bh->b_private = NULL; | |
3463 | bh->b_end_io = NULL; | |
3464 | } | |
3465 | curr_off = curr_off + bh->b_size; | |
3466 | bh = bh->b_this_page; | |
3467 | } while (bh != head); | |
3468 | } | |
3469 | ||
617ba13b | 3470 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 3471 | { |
617ba13b | 3472 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3473 | |
744692dc JZ |
3474 | /* |
3475 | * free any io_end structure allocated for buffers to be discarded | |
3476 | */ | |
3477 | if (ext4_should_dioread_nolock(page->mapping->host)) | |
3478 | ext4_invalidatepage_free_endio(page, offset); | |
ac27a0ec DK |
3479 | /* |
3480 | * If it's a full truncate we just forget about the pending dirtying | |
3481 | */ | |
3482 | if (offset == 0) | |
3483 | ClearPageChecked(page); | |
3484 | ||
0390131b FM |
3485 | if (journal) |
3486 | jbd2_journal_invalidatepage(journal, page, offset); | |
3487 | else | |
3488 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
3489 | } |
3490 | ||
617ba13b | 3491 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3492 | { |
617ba13b | 3493 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
3494 | |
3495 | WARN_ON(PageChecked(page)); | |
3496 | if (!page_has_buffers(page)) | |
3497 | return 0; | |
0390131b FM |
3498 | if (journal) |
3499 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3500 | else | |
3501 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3502 | } |
3503 | ||
3504 | /* | |
4c0425ff MC |
3505 | * O_DIRECT for ext3 (or indirect map) based files |
3506 | * | |
ac27a0ec DK |
3507 | * If the O_DIRECT write will extend the file then add this inode to the |
3508 | * orphan list. So recovery will truncate it back to the original size | |
3509 | * if the machine crashes during the write. | |
3510 | * | |
3511 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
7fb5409d JK |
3512 | * crashes then stale disk data _may_ be exposed inside the file. But current |
3513 | * VFS code falls back into buffered path in that case so we are safe. | |
ac27a0ec | 3514 | */ |
4c0425ff | 3515 | static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb, |
de9a55b8 TT |
3516 | const struct iovec *iov, loff_t offset, |
3517 | unsigned long nr_segs) | |
ac27a0ec DK |
3518 | { |
3519 | struct file *file = iocb->ki_filp; | |
3520 | struct inode *inode = file->f_mapping->host; | |
617ba13b | 3521 | struct ext4_inode_info *ei = EXT4_I(inode); |
7fb5409d | 3522 | handle_t *handle; |
ac27a0ec DK |
3523 | ssize_t ret; |
3524 | int orphan = 0; | |
3525 | size_t count = iov_length(iov, nr_segs); | |
fbbf6945 | 3526 | int retries = 0; |
ac27a0ec DK |
3527 | |
3528 | if (rw == WRITE) { | |
3529 | loff_t final_size = offset + count; | |
3530 | ||
ac27a0ec | 3531 | if (final_size > inode->i_size) { |
7fb5409d JK |
3532 | /* Credits for sb + inode write */ |
3533 | handle = ext4_journal_start(inode, 2); | |
3534 | if (IS_ERR(handle)) { | |
3535 | ret = PTR_ERR(handle); | |
3536 | goto out; | |
3537 | } | |
617ba13b | 3538 | ret = ext4_orphan_add(handle, inode); |
7fb5409d JK |
3539 | if (ret) { |
3540 | ext4_journal_stop(handle); | |
3541 | goto out; | |
3542 | } | |
ac27a0ec DK |
3543 | orphan = 1; |
3544 | ei->i_disksize = inode->i_size; | |
7fb5409d | 3545 | ext4_journal_stop(handle); |
ac27a0ec DK |
3546 | } |
3547 | } | |
3548 | ||
fbbf6945 | 3549 | retry: |
b7adc1f3 JZ |
3550 | if (rw == READ && ext4_should_dioread_nolock(inode)) |
3551 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, | |
3552 | inode->i_sb->s_bdev, iov, | |
3553 | offset, nr_segs, | |
3554 | ext4_get_block, NULL); | |
3555 | else | |
3556 | ret = blockdev_direct_IO(rw, iocb, inode, | |
3557 | inode->i_sb->s_bdev, iov, | |
ac27a0ec | 3558 | offset, nr_segs, |
617ba13b | 3559 | ext4_get_block, NULL); |
fbbf6945 ES |
3560 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
3561 | goto retry; | |
ac27a0ec | 3562 | |
7fb5409d | 3563 | if (orphan) { |
ac27a0ec DK |
3564 | int err; |
3565 | ||
7fb5409d JK |
3566 | /* Credits for sb + inode write */ |
3567 | handle = ext4_journal_start(inode, 2); | |
3568 | if (IS_ERR(handle)) { | |
3569 | /* This is really bad luck. We've written the data | |
3570 | * but cannot extend i_size. Bail out and pretend | |
3571 | * the write failed... */ | |
3572 | ret = PTR_ERR(handle); | |
da1dafca DM |
3573 | if (inode->i_nlink) |
3574 | ext4_orphan_del(NULL, inode); | |
3575 | ||
7fb5409d JK |
3576 | goto out; |
3577 | } | |
3578 | if (inode->i_nlink) | |
617ba13b | 3579 | ext4_orphan_del(handle, inode); |
7fb5409d | 3580 | if (ret > 0) { |
ac27a0ec DK |
3581 | loff_t end = offset + ret; |
3582 | if (end > inode->i_size) { | |
3583 | ei->i_disksize = end; | |
3584 | i_size_write(inode, end); | |
3585 | /* | |
3586 | * We're going to return a positive `ret' | |
3587 | * here due to non-zero-length I/O, so there's | |
3588 | * no way of reporting error returns from | |
617ba13b | 3589 | * ext4_mark_inode_dirty() to userspace. So |
ac27a0ec DK |
3590 | * ignore it. |
3591 | */ | |
617ba13b | 3592 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
3593 | } |
3594 | } | |
617ba13b | 3595 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
3596 | if (ret == 0) |
3597 | ret = err; | |
3598 | } | |
3599 | out: | |
3600 | return ret; | |
3601 | } | |
3602 | ||
2ed88685 TT |
3603 | /* |
3604 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3605 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3606 | * The extent will be converted to initialized after the IO is complete. | |
3607 | */ | |
c7064ef1 | 3608 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3609 | struct buffer_head *bh_result, int create) |
3610 | { | |
c7064ef1 | 3611 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3612 | inode->i_ino, create); |
2ed88685 TT |
3613 | return _ext4_get_block(inode, iblock, bh_result, |
3614 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3615 | } |
3616 | ||
c7064ef1 | 3617 | static void dump_completed_IO(struct inode * inode) |
8d5d02e6 MC |
3618 | { |
3619 | #ifdef EXT4_DEBUG | |
3620 | struct list_head *cur, *before, *after; | |
3621 | ext4_io_end_t *io, *io0, *io1; | |
744692dc | 3622 | unsigned long flags; |
8d5d02e6 | 3623 | |
c7064ef1 JZ |
3624 | if (list_empty(&EXT4_I(inode)->i_completed_io_list)){ |
3625 | ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino); | |
8d5d02e6 MC |
3626 | return; |
3627 | } | |
3628 | ||
c7064ef1 | 3629 | ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino); |
744692dc | 3630 | spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); |
c7064ef1 | 3631 | list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){ |
8d5d02e6 MC |
3632 | cur = &io->list; |
3633 | before = cur->prev; | |
3634 | io0 = container_of(before, ext4_io_end_t, list); | |
3635 | after = cur->next; | |
3636 | io1 = container_of(after, ext4_io_end_t, list); | |
3637 | ||
3638 | ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", | |
3639 | io, inode->i_ino, io0, io1); | |
3640 | } | |
744692dc | 3641 | spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); |
8d5d02e6 MC |
3642 | #endif |
3643 | } | |
4c0425ff MC |
3644 | |
3645 | /* | |
4c0425ff MC |
3646 | * check a range of space and convert unwritten extents to written. |
3647 | */ | |
c7064ef1 | 3648 | static int ext4_end_io_nolock(ext4_io_end_t *io) |
4c0425ff | 3649 | { |
4c0425ff MC |
3650 | struct inode *inode = io->inode; |
3651 | loff_t offset = io->offset; | |
a1de02dc | 3652 | ssize_t size = io->size; |
4c0425ff | 3653 | int ret = 0; |
4c0425ff | 3654 | |
c7064ef1 | 3655 | ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p," |
8d5d02e6 MC |
3656 | "list->prev 0x%p\n", |
3657 | io, inode->i_ino, io->list.next, io->list.prev); | |
3658 | ||
3659 | if (list_empty(&io->list)) | |
3660 | return ret; | |
3661 | ||
c7064ef1 | 3662 | if (io->flag != EXT4_IO_UNWRITTEN) |
8d5d02e6 MC |
3663 | return ret; |
3664 | ||
744692dc | 3665 | ret = ext4_convert_unwritten_extents(inode, offset, size); |
8d5d02e6 | 3666 | if (ret < 0) { |
4c0425ff | 3667 | printk(KERN_EMERG "%s: failed to convert unwritten" |
8d5d02e6 MC |
3668 | "extents to written extents, error is %d" |
3669 | " io is still on inode %lu aio dio list\n", | |
3670 | __func__, ret, inode->i_ino); | |
3671 | return ret; | |
3672 | } | |
4c0425ff | 3673 | |
5b3ff237 JZ |
3674 | if (io->iocb) |
3675 | aio_complete(io->iocb, io->result, 0); | |
8d5d02e6 MC |
3676 | /* clear the DIO AIO unwritten flag */ |
3677 | io->flag = 0; | |
3678 | return ret; | |
4c0425ff | 3679 | } |
c7064ef1 | 3680 | |
8d5d02e6 MC |
3681 | /* |
3682 | * work on completed aio dio IO, to convert unwritten extents to extents | |
3683 | */ | |
c7064ef1 | 3684 | static void ext4_end_io_work(struct work_struct *work) |
8d5d02e6 | 3685 | { |
744692dc JZ |
3686 | ext4_io_end_t *io = container_of(work, ext4_io_end_t, work); |
3687 | struct inode *inode = io->inode; | |
3688 | struct ext4_inode_info *ei = EXT4_I(inode); | |
3689 | unsigned long flags; | |
3690 | int ret; | |
4c0425ff | 3691 | |
8d5d02e6 | 3692 | mutex_lock(&inode->i_mutex); |
c7064ef1 | 3693 | ret = ext4_end_io_nolock(io); |
744692dc JZ |
3694 | if (ret < 0) { |
3695 | mutex_unlock(&inode->i_mutex); | |
3696 | return; | |
8d5d02e6 | 3697 | } |
744692dc JZ |
3698 | |
3699 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); | |
3700 | if (!list_empty(&io->list)) | |
3701 | list_del_init(&io->list); | |
3702 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
8d5d02e6 | 3703 | mutex_unlock(&inode->i_mutex); |
744692dc | 3704 | ext4_free_io_end(io); |
8d5d02e6 | 3705 | } |
c7064ef1 | 3706 | |
8d5d02e6 MC |
3707 | /* |
3708 | * This function is called from ext4_sync_file(). | |
3709 | * | |
c7064ef1 JZ |
3710 | * When IO is completed, the work to convert unwritten extents to |
3711 | * written is queued on workqueue but may not get immediately | |
8d5d02e6 MC |
3712 | * scheduled. When fsync is called, we need to ensure the |
3713 | * conversion is complete before fsync returns. | |
c7064ef1 JZ |
3714 | * The inode keeps track of a list of pending/completed IO that |
3715 | * might needs to do the conversion. This function walks through | |
3716 | * the list and convert the related unwritten extents for completed IO | |
3717 | * to written. | |
3718 | * The function return the number of pending IOs on success. | |
8d5d02e6 | 3719 | */ |
c7064ef1 | 3720 | int flush_completed_IO(struct inode *inode) |
8d5d02e6 MC |
3721 | { |
3722 | ext4_io_end_t *io; | |
744692dc JZ |
3723 | struct ext4_inode_info *ei = EXT4_I(inode); |
3724 | unsigned long flags; | |
8d5d02e6 MC |
3725 | int ret = 0; |
3726 | int ret2 = 0; | |
3727 | ||
744692dc | 3728 | if (list_empty(&ei->i_completed_io_list)) |
8d5d02e6 MC |
3729 | return ret; |
3730 | ||
c7064ef1 | 3731 | dump_completed_IO(inode); |
744692dc JZ |
3732 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
3733 | while (!list_empty(&ei->i_completed_io_list)){ | |
3734 | io = list_entry(ei->i_completed_io_list.next, | |
8d5d02e6 MC |
3735 | ext4_io_end_t, list); |
3736 | /* | |
c7064ef1 | 3737 | * Calling ext4_end_io_nolock() to convert completed |
8d5d02e6 MC |
3738 | * IO to written. |
3739 | * | |
3740 | * When ext4_sync_file() is called, run_queue() may already | |
3741 | * about to flush the work corresponding to this io structure. | |
3742 | * It will be upset if it founds the io structure related | |
3743 | * to the work-to-be schedule is freed. | |
3744 | * | |
3745 | * Thus we need to keep the io structure still valid here after | |
3746 | * convertion finished. The io structure has a flag to | |
3747 | * avoid double converting from both fsync and background work | |
3748 | * queue work. | |
3749 | */ | |
744692dc | 3750 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
c7064ef1 | 3751 | ret = ext4_end_io_nolock(io); |
744692dc | 3752 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
8d5d02e6 MC |
3753 | if (ret < 0) |
3754 | ret2 = ret; | |
3755 | else | |
3756 | list_del_init(&io->list); | |
3757 | } | |
744692dc | 3758 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
8d5d02e6 MC |
3759 | return (ret2 < 0) ? ret2 : 0; |
3760 | } | |
3761 | ||
744692dc | 3762 | static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags) |
4c0425ff MC |
3763 | { |
3764 | ext4_io_end_t *io = NULL; | |
3765 | ||
744692dc | 3766 | io = kmalloc(sizeof(*io), flags); |
4c0425ff MC |
3767 | |
3768 | if (io) { | |
8d5d02e6 | 3769 | igrab(inode); |
4c0425ff | 3770 | io->inode = inode; |
8d5d02e6 | 3771 | io->flag = 0; |
4c0425ff MC |
3772 | io->offset = 0; |
3773 | io->size = 0; | |
744692dc | 3774 | io->page = NULL; |
5b3ff237 JZ |
3775 | io->iocb = NULL; |
3776 | io->result = 0; | |
c7064ef1 | 3777 | INIT_WORK(&io->work, ext4_end_io_work); |
8d5d02e6 | 3778 | INIT_LIST_HEAD(&io->list); |
4c0425ff MC |
3779 | } |
3780 | ||
3781 | return io; | |
3782 | } | |
3783 | ||
3784 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, | |
552ef802 CH |
3785 | ssize_t size, void *private, int ret, |
3786 | bool is_async) | |
4c0425ff MC |
3787 | { |
3788 | ext4_io_end_t *io_end = iocb->private; | |
3789 | struct workqueue_struct *wq; | |
744692dc JZ |
3790 | unsigned long flags; |
3791 | struct ext4_inode_info *ei; | |
4c0425ff | 3792 | |
4b70df18 M |
3793 | /* if not async direct IO or dio with 0 bytes write, just return */ |
3794 | if (!io_end || !size) | |
552ef802 | 3795 | goto out; |
4b70df18 | 3796 | |
8d5d02e6 MC |
3797 | ext_debug("ext4_end_io_dio(): io_end 0x%p" |
3798 | "for inode %lu, iocb 0x%p, offset %llu, size %llu\n", | |
3799 | iocb->private, io_end->inode->i_ino, iocb, offset, | |
3800 | size); | |
8d5d02e6 MC |
3801 | |
3802 | /* if not aio dio with unwritten extents, just free io and return */ | |
c7064ef1 | 3803 | if (io_end->flag != EXT4_IO_UNWRITTEN){ |
8d5d02e6 MC |
3804 | ext4_free_io_end(io_end); |
3805 | iocb->private = NULL; | |
5b3ff237 JZ |
3806 | out: |
3807 | if (is_async) | |
3808 | aio_complete(iocb, ret, 0); | |
3809 | return; | |
8d5d02e6 MC |
3810 | } |
3811 | ||
4c0425ff MC |
3812 | io_end->offset = offset; |
3813 | io_end->size = size; | |
5b3ff237 JZ |
3814 | if (is_async) { |
3815 | io_end->iocb = iocb; | |
3816 | io_end->result = ret; | |
3817 | } | |
4c0425ff MC |
3818 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; |
3819 | ||
8d5d02e6 | 3820 | /* queue the work to convert unwritten extents to written */ |
4c0425ff MC |
3821 | queue_work(wq, &io_end->work); |
3822 | ||
8d5d02e6 | 3823 | /* Add the io_end to per-inode completed aio dio list*/ |
744692dc JZ |
3824 | ei = EXT4_I(io_end->inode); |
3825 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); | |
3826 | list_add_tail(&io_end->list, &ei->i_completed_io_list); | |
3827 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
4c0425ff MC |
3828 | iocb->private = NULL; |
3829 | } | |
c7064ef1 | 3830 | |
744692dc JZ |
3831 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate) |
3832 | { | |
3833 | ext4_io_end_t *io_end = bh->b_private; | |
3834 | struct workqueue_struct *wq; | |
3835 | struct inode *inode; | |
3836 | unsigned long flags; | |
3837 | ||
3838 | if (!test_clear_buffer_uninit(bh) || !io_end) | |
3839 | goto out; | |
3840 | ||
3841 | if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) { | |
3842 | printk("sb umounted, discard end_io request for inode %lu\n", | |
3843 | io_end->inode->i_ino); | |
3844 | ext4_free_io_end(io_end); | |
3845 | goto out; | |
3846 | } | |
3847 | ||
3848 | io_end->flag = EXT4_IO_UNWRITTEN; | |
3849 | inode = io_end->inode; | |
3850 | ||
3851 | /* Add the io_end to per-inode completed io list*/ | |
3852 | spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); | |
3853 | list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list); | |
3854 | spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); | |
3855 | ||
3856 | wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq; | |
3857 | /* queue the work to convert unwritten extents to written */ | |
3858 | queue_work(wq, &io_end->work); | |
3859 | out: | |
3860 | bh->b_private = NULL; | |
3861 | bh->b_end_io = NULL; | |
3862 | clear_buffer_uninit(bh); | |
3863 | end_buffer_async_write(bh, uptodate); | |
3864 | } | |
3865 | ||
3866 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode) | |
3867 | { | |
3868 | ext4_io_end_t *io_end; | |
3869 | struct page *page = bh->b_page; | |
3870 | loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT; | |
3871 | size_t size = bh->b_size; | |
3872 | ||
3873 | retry: | |
3874 | io_end = ext4_init_io_end(inode, GFP_ATOMIC); | |
3875 | if (!io_end) { | |
3876 | if (printk_ratelimit()) | |
3877 | printk(KERN_WARNING "%s: allocation fail\n", __func__); | |
3878 | schedule(); | |
3879 | goto retry; | |
3880 | } | |
3881 | io_end->offset = offset; | |
3882 | io_end->size = size; | |
3883 | /* | |
3884 | * We need to hold a reference to the page to make sure it | |
3885 | * doesn't get evicted before ext4_end_io_work() has a chance | |
3886 | * to convert the extent from written to unwritten. | |
3887 | */ | |
3888 | io_end->page = page; | |
3889 | get_page(io_end->page); | |
3890 | ||
3891 | bh->b_private = io_end; | |
3892 | bh->b_end_io = ext4_end_io_buffer_write; | |
3893 | return 0; | |
3894 | } | |
3895 | ||
4c0425ff MC |
3896 | /* |
3897 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3898 | * preallocated extents, and those write extend the file, no need to | |
3899 | * fall back to buffered IO. | |
3900 | * | |
3901 | * For holes, we fallocate those blocks, mark them as unintialized | |
3902 | * If those blocks were preallocated, we mark sure they are splited, but | |
3903 | * still keep the range to write as unintialized. | |
3904 | * | |
8d5d02e6 MC |
3905 | * The unwrritten extents will be converted to written when DIO is completed. |
3906 | * For async direct IO, since the IO may still pending when return, we | |
3907 | * set up an end_io call back function, which will do the convertion | |
3908 | * when async direct IO completed. | |
4c0425ff MC |
3909 | * |
3910 | * If the O_DIRECT write will extend the file then add this inode to the | |
3911 | * orphan list. So recovery will truncate it back to the original size | |
3912 | * if the machine crashes during the write. | |
3913 | * | |
3914 | */ | |
3915 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
3916 | const struct iovec *iov, loff_t offset, | |
3917 | unsigned long nr_segs) | |
3918 | { | |
3919 | struct file *file = iocb->ki_filp; | |
3920 | struct inode *inode = file->f_mapping->host; | |
3921 | ssize_t ret; | |
3922 | size_t count = iov_length(iov, nr_segs); | |
3923 | ||
3924 | loff_t final_size = offset + count; | |
3925 | if (rw == WRITE && final_size <= inode->i_size) { | |
3926 | /* | |
8d5d02e6 MC |
3927 | * We could direct write to holes and fallocate. |
3928 | * | |
3929 | * Allocated blocks to fill the hole are marked as uninitialized | |
4c0425ff MC |
3930 | * to prevent paralel buffered read to expose the stale data |
3931 | * before DIO complete the data IO. | |
8d5d02e6 MC |
3932 | * |
3933 | * As to previously fallocated extents, ext4 get_block | |
4c0425ff MC |
3934 | * will just simply mark the buffer mapped but still |
3935 | * keep the extents uninitialized. | |
3936 | * | |
8d5d02e6 MC |
3937 | * for non AIO case, we will convert those unwritten extents |
3938 | * to written after return back from blockdev_direct_IO. | |
3939 | * | |
3940 | * for async DIO, the conversion needs to be defered when | |
3941 | * the IO is completed. The ext4 end_io callback function | |
3942 | * will be called to take care of the conversion work. | |
3943 | * Here for async case, we allocate an io_end structure to | |
3944 | * hook to the iocb. | |
4c0425ff | 3945 | */ |
8d5d02e6 MC |
3946 | iocb->private = NULL; |
3947 | EXT4_I(inode)->cur_aio_dio = NULL; | |
3948 | if (!is_sync_kiocb(iocb)) { | |
744692dc | 3949 | iocb->private = ext4_init_io_end(inode, GFP_NOFS); |
8d5d02e6 MC |
3950 | if (!iocb->private) |
3951 | return -ENOMEM; | |
3952 | /* | |
3953 | * we save the io structure for current async | |
79e83036 | 3954 | * direct IO, so that later ext4_map_blocks() |
8d5d02e6 MC |
3955 | * could flag the io structure whether there |
3956 | * is a unwritten extents needs to be converted | |
3957 | * when IO is completed. | |
3958 | */ | |
3959 | EXT4_I(inode)->cur_aio_dio = iocb->private; | |
3960 | } | |
3961 | ||
4c0425ff MC |
3962 | ret = blockdev_direct_IO(rw, iocb, inode, |
3963 | inode->i_sb->s_bdev, iov, | |
3964 | offset, nr_segs, | |
c7064ef1 | 3965 | ext4_get_block_write, |
4c0425ff | 3966 | ext4_end_io_dio); |
8d5d02e6 MC |
3967 | if (iocb->private) |
3968 | EXT4_I(inode)->cur_aio_dio = NULL; | |
3969 | /* | |
3970 | * The io_end structure takes a reference to the inode, | |
3971 | * that structure needs to be destroyed and the | |
3972 | * reference to the inode need to be dropped, when IO is | |
3973 | * complete, even with 0 byte write, or failed. | |
3974 | * | |
3975 | * In the successful AIO DIO case, the io_end structure will be | |
3976 | * desctroyed and the reference to the inode will be dropped | |
3977 | * after the end_io call back function is called. | |
3978 | * | |
3979 | * In the case there is 0 byte write, or error case, since | |
3980 | * VFS direct IO won't invoke the end_io call back function, | |
3981 | * we need to free the end_io structure here. | |
3982 | */ | |
3983 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { | |
3984 | ext4_free_io_end(iocb->private); | |
3985 | iocb->private = NULL; | |
19f5fb7a TT |
3986 | } else if (ret > 0 && ext4_test_inode_state(inode, |
3987 | EXT4_STATE_DIO_UNWRITTEN)) { | |
109f5565 | 3988 | int err; |
8d5d02e6 MC |
3989 | /* |
3990 | * for non AIO case, since the IO is already | |
3991 | * completed, we could do the convertion right here | |
3992 | */ | |
109f5565 M |
3993 | err = ext4_convert_unwritten_extents(inode, |
3994 | offset, ret); | |
3995 | if (err < 0) | |
3996 | ret = err; | |
19f5fb7a | 3997 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
109f5565 | 3998 | } |
4c0425ff MC |
3999 | return ret; |
4000 | } | |
8d5d02e6 MC |
4001 | |
4002 | /* for write the the end of file case, we fall back to old way */ | |
4c0425ff MC |
4003 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
4004 | } | |
4005 | ||
4006 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
4007 | const struct iovec *iov, loff_t offset, | |
4008 | unsigned long nr_segs) | |
4009 | { | |
4010 | struct file *file = iocb->ki_filp; | |
4011 | struct inode *inode = file->f_mapping->host; | |
4012 | ||
12e9b892 | 4013 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4c0425ff MC |
4014 | return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
4015 | ||
4016 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
4017 | } | |
4018 | ||
ac27a0ec | 4019 | /* |
617ba13b | 4020 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
4021 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
4022 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
4023 | * not necessarily locked. | |
4024 | * | |
4025 | * We cannot just dirty the page and leave attached buffers clean, because the | |
4026 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
4027 | * or jbddirty because all the journalling code will explode. | |
4028 | * | |
4029 | * So what we do is to mark the page "pending dirty" and next time writepage | |
4030 | * is called, propagate that into the buffers appropriately. | |
4031 | */ | |
617ba13b | 4032 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
4033 | { |
4034 | SetPageChecked(page); | |
4035 | return __set_page_dirty_nobuffers(page); | |
4036 | } | |
4037 | ||
617ba13b | 4038 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
4039 | .readpage = ext4_readpage, |
4040 | .readpages = ext4_readpages, | |
43ce1d23 | 4041 | .writepage = ext4_writepage, |
8ab22b9a HH |
4042 | .sync_page = block_sync_page, |
4043 | .write_begin = ext4_write_begin, | |
4044 | .write_end = ext4_ordered_write_end, | |
4045 | .bmap = ext4_bmap, | |
4046 | .invalidatepage = ext4_invalidatepage, | |
4047 | .releasepage = ext4_releasepage, | |
4048 | .direct_IO = ext4_direct_IO, | |
4049 | .migratepage = buffer_migrate_page, | |
4050 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 4051 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
4052 | }; |
4053 | ||
617ba13b | 4054 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
4055 | .readpage = ext4_readpage, |
4056 | .readpages = ext4_readpages, | |
43ce1d23 | 4057 | .writepage = ext4_writepage, |
8ab22b9a HH |
4058 | .sync_page = block_sync_page, |
4059 | .write_begin = ext4_write_begin, | |
4060 | .write_end = ext4_writeback_write_end, | |
4061 | .bmap = ext4_bmap, | |
4062 | .invalidatepage = ext4_invalidatepage, | |
4063 | .releasepage = ext4_releasepage, | |
4064 | .direct_IO = ext4_direct_IO, | |
4065 | .migratepage = buffer_migrate_page, | |
4066 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 4067 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
4068 | }; |
4069 | ||
617ba13b | 4070 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
4071 | .readpage = ext4_readpage, |
4072 | .readpages = ext4_readpages, | |
43ce1d23 | 4073 | .writepage = ext4_writepage, |
8ab22b9a HH |
4074 | .sync_page = block_sync_page, |
4075 | .write_begin = ext4_write_begin, | |
4076 | .write_end = ext4_journalled_write_end, | |
4077 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
4078 | .bmap = ext4_bmap, | |
4079 | .invalidatepage = ext4_invalidatepage, | |
4080 | .releasepage = ext4_releasepage, | |
4081 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 4082 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
4083 | }; |
4084 | ||
64769240 | 4085 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
4086 | .readpage = ext4_readpage, |
4087 | .readpages = ext4_readpages, | |
43ce1d23 | 4088 | .writepage = ext4_writepage, |
8ab22b9a HH |
4089 | .writepages = ext4_da_writepages, |
4090 | .sync_page = block_sync_page, | |
4091 | .write_begin = ext4_da_write_begin, | |
4092 | .write_end = ext4_da_write_end, | |
4093 | .bmap = ext4_bmap, | |
4094 | .invalidatepage = ext4_da_invalidatepage, | |
4095 | .releasepage = ext4_releasepage, | |
4096 | .direct_IO = ext4_direct_IO, | |
4097 | .migratepage = buffer_migrate_page, | |
4098 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 4099 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
4100 | }; |
4101 | ||
617ba13b | 4102 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 4103 | { |
cd1aac32 AK |
4104 | if (ext4_should_order_data(inode) && |
4105 | test_opt(inode->i_sb, DELALLOC)) | |
4106 | inode->i_mapping->a_ops = &ext4_da_aops; | |
4107 | else if (ext4_should_order_data(inode)) | |
617ba13b | 4108 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
64769240 AT |
4109 | else if (ext4_should_writeback_data(inode) && |
4110 | test_opt(inode->i_sb, DELALLOC)) | |
4111 | inode->i_mapping->a_ops = &ext4_da_aops; | |
617ba13b MC |
4112 | else if (ext4_should_writeback_data(inode)) |
4113 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 4114 | else |
617ba13b | 4115 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
4116 | } |
4117 | ||
4118 | /* | |
617ba13b | 4119 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
4120 | * up to the end of the block which corresponds to `from'. |
4121 | * This required during truncate. We need to physically zero the tail end | |
4122 | * of that block so it doesn't yield old data if the file is later grown. | |
4123 | */ | |
cf108bca | 4124 | int ext4_block_truncate_page(handle_t *handle, |
ac27a0ec DK |
4125 | struct address_space *mapping, loff_t from) |
4126 | { | |
617ba13b | 4127 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 4128 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
725d26d3 AK |
4129 | unsigned blocksize, length, pos; |
4130 | ext4_lblk_t iblock; | |
ac27a0ec DK |
4131 | struct inode *inode = mapping->host; |
4132 | struct buffer_head *bh; | |
cf108bca | 4133 | struct page *page; |
ac27a0ec | 4134 | int err = 0; |
ac27a0ec | 4135 | |
f4a01017 TT |
4136 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, |
4137 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
cf108bca JK |
4138 | if (!page) |
4139 | return -EINVAL; | |
4140 | ||
ac27a0ec DK |
4141 | blocksize = inode->i_sb->s_blocksize; |
4142 | length = blocksize - (offset & (blocksize - 1)); | |
4143 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
4144 | ||
ac27a0ec DK |
4145 | if (!page_has_buffers(page)) |
4146 | create_empty_buffers(page, blocksize, 0); | |
4147 | ||
4148 | /* Find the buffer that contains "offset" */ | |
4149 | bh = page_buffers(page); | |
4150 | pos = blocksize; | |
4151 | while (offset >= pos) { | |
4152 | bh = bh->b_this_page; | |
4153 | iblock++; | |
4154 | pos += blocksize; | |
4155 | } | |
4156 | ||
4157 | err = 0; | |
4158 | if (buffer_freed(bh)) { | |
4159 | BUFFER_TRACE(bh, "freed: skip"); | |
4160 | goto unlock; | |
4161 | } | |
4162 | ||
4163 | if (!buffer_mapped(bh)) { | |
4164 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 4165 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
4166 | /* unmapped? It's a hole - nothing to do */ |
4167 | if (!buffer_mapped(bh)) { | |
4168 | BUFFER_TRACE(bh, "still unmapped"); | |
4169 | goto unlock; | |
4170 | } | |
4171 | } | |
4172 | ||
4173 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
4174 | if (PageUptodate(page)) | |
4175 | set_buffer_uptodate(bh); | |
4176 | ||
4177 | if (!buffer_uptodate(bh)) { | |
4178 | err = -EIO; | |
4179 | ll_rw_block(READ, 1, &bh); | |
4180 | wait_on_buffer(bh); | |
4181 | /* Uhhuh. Read error. Complain and punt. */ | |
4182 | if (!buffer_uptodate(bh)) | |
4183 | goto unlock; | |
4184 | } | |
4185 | ||
617ba13b | 4186 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 4187 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 4188 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
4189 | if (err) |
4190 | goto unlock; | |
4191 | } | |
4192 | ||
eebd2aa3 | 4193 | zero_user(page, offset, length); |
ac27a0ec DK |
4194 | |
4195 | BUFFER_TRACE(bh, "zeroed end of block"); | |
4196 | ||
4197 | err = 0; | |
617ba13b | 4198 | if (ext4_should_journal_data(inode)) { |
0390131b | 4199 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
ac27a0ec | 4200 | } else { |
617ba13b | 4201 | if (ext4_should_order_data(inode)) |
678aaf48 | 4202 | err = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
4203 | mark_buffer_dirty(bh); |
4204 | } | |
4205 | ||
4206 | unlock: | |
4207 | unlock_page(page); | |
4208 | page_cache_release(page); | |
4209 | return err; | |
4210 | } | |
4211 | ||
4212 | /* | |
4213 | * Probably it should be a library function... search for first non-zero word | |
4214 | * or memcmp with zero_page, whatever is better for particular architecture. | |
4215 | * Linus? | |
4216 | */ | |
4217 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
4218 | { | |
4219 | while (p < q) | |
4220 | if (*p++) | |
4221 | return 0; | |
4222 | return 1; | |
4223 | } | |
4224 | ||
4225 | /** | |
617ba13b | 4226 | * ext4_find_shared - find the indirect blocks for partial truncation. |
ac27a0ec DK |
4227 | * @inode: inode in question |
4228 | * @depth: depth of the affected branch | |
617ba13b | 4229 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
ac27a0ec DK |
4230 | * @chain: place to store the pointers to partial indirect blocks |
4231 | * @top: place to the (detached) top of branch | |
4232 | * | |
617ba13b | 4233 | * This is a helper function used by ext4_truncate(). |
ac27a0ec DK |
4234 | * |
4235 | * When we do truncate() we may have to clean the ends of several | |
4236 | * indirect blocks but leave the blocks themselves alive. Block is | |
4237 | * partially truncated if some data below the new i_size is refered | |
4238 | * from it (and it is on the path to the first completely truncated | |
4239 | * data block, indeed). We have to free the top of that path along | |
4240 | * with everything to the right of the path. Since no allocation | |
617ba13b | 4241 | * past the truncation point is possible until ext4_truncate() |
ac27a0ec DK |
4242 | * finishes, we may safely do the latter, but top of branch may |
4243 | * require special attention - pageout below the truncation point | |
4244 | * might try to populate it. | |
4245 | * | |
4246 | * We atomically detach the top of branch from the tree, store the | |
4247 | * block number of its root in *@top, pointers to buffer_heads of | |
4248 | * partially truncated blocks - in @chain[].bh and pointers to | |
4249 | * their last elements that should not be removed - in | |
4250 | * @chain[].p. Return value is the pointer to last filled element | |
4251 | * of @chain. | |
4252 | * | |
4253 | * The work left to caller to do the actual freeing of subtrees: | |
4254 | * a) free the subtree starting from *@top | |
4255 | * b) free the subtrees whose roots are stored in | |
4256 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
4257 | * c) free the subtrees growing from the inode past the @chain[0]. | |
4258 | * (no partially truncated stuff there). */ | |
4259 | ||
617ba13b | 4260 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
de9a55b8 TT |
4261 | ext4_lblk_t offsets[4], Indirect chain[4], |
4262 | __le32 *top) | |
ac27a0ec DK |
4263 | { |
4264 | Indirect *partial, *p; | |
4265 | int k, err; | |
4266 | ||
4267 | *top = 0; | |
bf48aabb | 4268 | /* Make k index the deepest non-null offset + 1 */ |
ac27a0ec DK |
4269 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
4270 | ; | |
617ba13b | 4271 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
ac27a0ec DK |
4272 | /* Writer: pointers */ |
4273 | if (!partial) | |
4274 | partial = chain + k-1; | |
4275 | /* | |
4276 | * If the branch acquired continuation since we've looked at it - | |
4277 | * fine, it should all survive and (new) top doesn't belong to us. | |
4278 | */ | |
4279 | if (!partial->key && *partial->p) | |
4280 | /* Writer: end */ | |
4281 | goto no_top; | |
af5bc92d | 4282 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) |
ac27a0ec DK |
4283 | ; |
4284 | /* | |
4285 | * OK, we've found the last block that must survive. The rest of our | |
4286 | * branch should be detached before unlocking. However, if that rest | |
4287 | * of branch is all ours and does not grow immediately from the inode | |
4288 | * it's easier to cheat and just decrement partial->p. | |
4289 | */ | |
4290 | if (p == chain + k - 1 && p > chain) { | |
4291 | p->p--; | |
4292 | } else { | |
4293 | *top = *p->p; | |
617ba13b | 4294 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
ac27a0ec DK |
4295 | #if 0 |
4296 | *p->p = 0; | |
4297 | #endif | |
4298 | } | |
4299 | /* Writer: end */ | |
4300 | ||
af5bc92d | 4301 | while (partial > p) { |
ac27a0ec DK |
4302 | brelse(partial->bh); |
4303 | partial--; | |
4304 | } | |
4305 | no_top: | |
4306 | return partial; | |
4307 | } | |
4308 | ||
4309 | /* | |
4310 | * Zero a number of block pointers in either an inode or an indirect block. | |
4311 | * If we restart the transaction we must again get write access to the | |
4312 | * indirect block for further modification. | |
4313 | * | |
4314 | * We release `count' blocks on disk, but (last - first) may be greater | |
4315 | * than `count' because there can be holes in there. | |
4316 | */ | |
1f2acb60 TT |
4317 | static int ext4_clear_blocks(handle_t *handle, struct inode *inode, |
4318 | struct buffer_head *bh, | |
4319 | ext4_fsblk_t block_to_free, | |
4320 | unsigned long count, __le32 *first, | |
4321 | __le32 *last) | |
ac27a0ec DK |
4322 | { |
4323 | __le32 *p; | |
1f2acb60 | 4324 | int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED; |
e6362609 TT |
4325 | |
4326 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
4327 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
50689696 | 4328 | |
1f2acb60 TT |
4329 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free, |
4330 | count)) { | |
24676da4 TT |
4331 | EXT4_ERROR_INODE(inode, "attempt to clear invalid " |
4332 | "blocks %llu len %lu", | |
4333 | (unsigned long long) block_to_free, count); | |
1f2acb60 TT |
4334 | return 1; |
4335 | } | |
4336 | ||
ac27a0ec DK |
4337 | if (try_to_extend_transaction(handle, inode)) { |
4338 | if (bh) { | |
0390131b FM |
4339 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
4340 | ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec | 4341 | } |
617ba13b | 4342 | ext4_mark_inode_dirty(handle, inode); |
487caeef JK |
4343 | ext4_truncate_restart_trans(handle, inode, |
4344 | blocks_for_truncate(inode)); | |
ac27a0ec DK |
4345 | if (bh) { |
4346 | BUFFER_TRACE(bh, "retaking write access"); | |
617ba13b | 4347 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
4348 | } |
4349 | } | |
4350 | ||
e6362609 TT |
4351 | for (p = first; p < last; p++) |
4352 | *p = 0; | |
ac27a0ec | 4353 | |
e6362609 | 4354 | ext4_free_blocks(handle, inode, 0, block_to_free, count, flags); |
1f2acb60 | 4355 | return 0; |
ac27a0ec DK |
4356 | } |
4357 | ||
4358 | /** | |
617ba13b | 4359 | * ext4_free_data - free a list of data blocks |
ac27a0ec DK |
4360 | * @handle: handle for this transaction |
4361 | * @inode: inode we are dealing with | |
4362 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
4363 | * @first: array of block numbers | |
4364 | * @last: points immediately past the end of array | |
4365 | * | |
4366 | * We are freeing all blocks refered from that array (numbers are stored as | |
4367 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
4368 | * | |
4369 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
4370 | * blocks are contiguous then releasing them at one time will only affect one | |
4371 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
4372 | * actually use a lot of journal space. | |
4373 | * | |
4374 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
4375 | * block pointers. | |
4376 | */ | |
617ba13b | 4377 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
4378 | struct buffer_head *this_bh, |
4379 | __le32 *first, __le32 *last) | |
4380 | { | |
617ba13b | 4381 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ac27a0ec DK |
4382 | unsigned long count = 0; /* Number of blocks in the run */ |
4383 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
4384 | corresponding to | |
4385 | block_to_free */ | |
617ba13b | 4386 | ext4_fsblk_t nr; /* Current block # */ |
ac27a0ec DK |
4387 | __le32 *p; /* Pointer into inode/ind |
4388 | for current block */ | |
4389 | int err; | |
4390 | ||
4391 | if (this_bh) { /* For indirect block */ | |
4392 | BUFFER_TRACE(this_bh, "get_write_access"); | |
617ba13b | 4393 | err = ext4_journal_get_write_access(handle, this_bh); |
ac27a0ec DK |
4394 | /* Important: if we can't update the indirect pointers |
4395 | * to the blocks, we can't free them. */ | |
4396 | if (err) | |
4397 | return; | |
4398 | } | |
4399 | ||
4400 | for (p = first; p < last; p++) { | |
4401 | nr = le32_to_cpu(*p); | |
4402 | if (nr) { | |
4403 | /* accumulate blocks to free if they're contiguous */ | |
4404 | if (count == 0) { | |
4405 | block_to_free = nr; | |
4406 | block_to_free_p = p; | |
4407 | count = 1; | |
4408 | } else if (nr == block_to_free + count) { | |
4409 | count++; | |
4410 | } else { | |
1f2acb60 TT |
4411 | if (ext4_clear_blocks(handle, inode, this_bh, |
4412 | block_to_free, count, | |
4413 | block_to_free_p, p)) | |
4414 | break; | |
ac27a0ec DK |
4415 | block_to_free = nr; |
4416 | block_to_free_p = p; | |
4417 | count = 1; | |
4418 | } | |
4419 | } | |
4420 | } | |
4421 | ||
4422 | if (count > 0) | |
617ba13b | 4423 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
ac27a0ec DK |
4424 | count, block_to_free_p, p); |
4425 | ||
4426 | if (this_bh) { | |
0390131b | 4427 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); |
71dc8fbc DG |
4428 | |
4429 | /* | |
4430 | * The buffer head should have an attached journal head at this | |
4431 | * point. However, if the data is corrupted and an indirect | |
4432 | * block pointed to itself, it would have been detached when | |
4433 | * the block was cleared. Check for this instead of OOPSing. | |
4434 | */ | |
e7f07968 | 4435 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) |
0390131b | 4436 | ext4_handle_dirty_metadata(handle, inode, this_bh); |
71dc8fbc | 4437 | else |
24676da4 TT |
4438 | EXT4_ERROR_INODE(inode, |
4439 | "circular indirect block detected at " | |
4440 | "block %llu", | |
4441 | (unsigned long long) this_bh->b_blocknr); | |
ac27a0ec DK |
4442 | } |
4443 | } | |
4444 | ||
4445 | /** | |
617ba13b | 4446 | * ext4_free_branches - free an array of branches |
ac27a0ec DK |
4447 | * @handle: JBD handle for this transaction |
4448 | * @inode: inode we are dealing with | |
4449 | * @parent_bh: the buffer_head which contains *@first and *@last | |
4450 | * @first: array of block numbers | |
4451 | * @last: pointer immediately past the end of array | |
4452 | * @depth: depth of the branches to free | |
4453 | * | |
4454 | * We are freeing all blocks refered from these branches (numbers are | |
4455 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
4456 | * appropriately. | |
4457 | */ | |
617ba13b | 4458 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
4459 | struct buffer_head *parent_bh, |
4460 | __le32 *first, __le32 *last, int depth) | |
4461 | { | |
617ba13b | 4462 | ext4_fsblk_t nr; |
ac27a0ec DK |
4463 | __le32 *p; |
4464 | ||
0390131b | 4465 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
4466 | return; |
4467 | ||
4468 | if (depth--) { | |
4469 | struct buffer_head *bh; | |
617ba13b | 4470 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
4471 | p = last; |
4472 | while (--p >= first) { | |
4473 | nr = le32_to_cpu(*p); | |
4474 | if (!nr) | |
4475 | continue; /* A hole */ | |
4476 | ||
1f2acb60 TT |
4477 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), |
4478 | nr, 1)) { | |
24676da4 TT |
4479 | EXT4_ERROR_INODE(inode, |
4480 | "invalid indirect mapped " | |
4481 | "block %lu (level %d)", | |
4482 | (unsigned long) nr, depth); | |
1f2acb60 TT |
4483 | break; |
4484 | } | |
4485 | ||
ac27a0ec DK |
4486 | /* Go read the buffer for the next level down */ |
4487 | bh = sb_bread(inode->i_sb, nr); | |
4488 | ||
4489 | /* | |
4490 | * A read failure? Report error and clear slot | |
4491 | * (should be rare). | |
4492 | */ | |
4493 | if (!bh) { | |
c398eda0 TT |
4494 | EXT4_ERROR_INODE_BLOCK(inode, nr, |
4495 | "Read failure"); | |
ac27a0ec DK |
4496 | continue; |
4497 | } | |
4498 | ||
4499 | /* This zaps the entire block. Bottom up. */ | |
4500 | BUFFER_TRACE(bh, "free child branches"); | |
617ba13b | 4501 | ext4_free_branches(handle, inode, bh, |
af5bc92d TT |
4502 | (__le32 *) bh->b_data, |
4503 | (__le32 *) bh->b_data + addr_per_block, | |
4504 | depth); | |
ac27a0ec | 4505 | |
ac27a0ec DK |
4506 | /* |
4507 | * Everything below this this pointer has been | |
4508 | * released. Now let this top-of-subtree go. | |
4509 | * | |
4510 | * We want the freeing of this indirect block to be | |
4511 | * atomic in the journal with the updating of the | |
4512 | * bitmap block which owns it. So make some room in | |
4513 | * the journal. | |
4514 | * | |
4515 | * We zero the parent pointer *after* freeing its | |
4516 | * pointee in the bitmaps, so if extend_transaction() | |
4517 | * for some reason fails to put the bitmap changes and | |
4518 | * the release into the same transaction, recovery | |
4519 | * will merely complain about releasing a free block, | |
4520 | * rather than leaking blocks. | |
4521 | */ | |
0390131b | 4522 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
4523 | return; |
4524 | if (try_to_extend_transaction(handle, inode)) { | |
617ba13b | 4525 | ext4_mark_inode_dirty(handle, inode); |
487caeef JK |
4526 | ext4_truncate_restart_trans(handle, inode, |
4527 | blocks_for_truncate(inode)); | |
ac27a0ec DK |
4528 | } |
4529 | ||
40389687 A |
4530 | /* |
4531 | * The forget flag here is critical because if | |
4532 | * we are journaling (and not doing data | |
4533 | * journaling), we have to make sure a revoke | |
4534 | * record is written to prevent the journal | |
4535 | * replay from overwriting the (former) | |
4536 | * indirect block if it gets reallocated as a | |
4537 | * data block. This must happen in the same | |
4538 | * transaction where the data blocks are | |
4539 | * actually freed. | |
4540 | */ | |
e6362609 | 4541 | ext4_free_blocks(handle, inode, 0, nr, 1, |
40389687 A |
4542 | EXT4_FREE_BLOCKS_METADATA| |
4543 | EXT4_FREE_BLOCKS_FORGET); | |
ac27a0ec DK |
4544 | |
4545 | if (parent_bh) { | |
4546 | /* | |
4547 | * The block which we have just freed is | |
4548 | * pointed to by an indirect block: journal it | |
4549 | */ | |
4550 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
617ba13b | 4551 | if (!ext4_journal_get_write_access(handle, |
ac27a0ec DK |
4552 | parent_bh)){ |
4553 | *p = 0; | |
4554 | BUFFER_TRACE(parent_bh, | |
0390131b FM |
4555 | "call ext4_handle_dirty_metadata"); |
4556 | ext4_handle_dirty_metadata(handle, | |
4557 | inode, | |
4558 | parent_bh); | |
ac27a0ec DK |
4559 | } |
4560 | } | |
4561 | } | |
4562 | } else { | |
4563 | /* We have reached the bottom of the tree. */ | |
4564 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
617ba13b | 4565 | ext4_free_data(handle, inode, parent_bh, first, last); |
ac27a0ec DK |
4566 | } |
4567 | } | |
4568 | ||
91ef4caf DG |
4569 | int ext4_can_truncate(struct inode *inode) |
4570 | { | |
4571 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4572 | return 0; | |
4573 | if (S_ISREG(inode->i_mode)) | |
4574 | return 1; | |
4575 | if (S_ISDIR(inode->i_mode)) | |
4576 | return 1; | |
4577 | if (S_ISLNK(inode->i_mode)) | |
4578 | return !ext4_inode_is_fast_symlink(inode); | |
4579 | return 0; | |
4580 | } | |
4581 | ||
ac27a0ec | 4582 | /* |
617ba13b | 4583 | * ext4_truncate() |
ac27a0ec | 4584 | * |
617ba13b MC |
4585 | * We block out ext4_get_block() block instantiations across the entire |
4586 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4587 | * simultaneously on behalf of the same inode. |
4588 | * | |
4589 | * As we work through the truncate and commmit bits of it to the journal there | |
4590 | * is one core, guiding principle: the file's tree must always be consistent on | |
4591 | * disk. We must be able to restart the truncate after a crash. | |
4592 | * | |
4593 | * The file's tree may be transiently inconsistent in memory (although it | |
4594 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4595 | * the contents of (the filesystem + the journal) must be consistent and | |
4596 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4597 | * left-to-right works OK too). | |
4598 | * | |
4599 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4600 | * truncate against the orphan inode list. | |
4601 | * | |
4602 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4603 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4604 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4605 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4606 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4607 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4608 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4609 | */ |
617ba13b | 4610 | void ext4_truncate(struct inode *inode) |
ac27a0ec DK |
4611 | { |
4612 | handle_t *handle; | |
617ba13b | 4613 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 4614 | __le32 *i_data = ei->i_data; |
617ba13b | 4615 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec | 4616 | struct address_space *mapping = inode->i_mapping; |
725d26d3 | 4617 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
4618 | Indirect chain[4]; |
4619 | Indirect *partial; | |
4620 | __le32 nr = 0; | |
4621 | int n; | |
725d26d3 | 4622 | ext4_lblk_t last_block; |
ac27a0ec | 4623 | unsigned blocksize = inode->i_sb->s_blocksize; |
ac27a0ec | 4624 | |
91ef4caf | 4625 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
4626 | return; |
4627 | ||
12e9b892 | 4628 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 4629 | |
5534fb5b | 4630 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4631 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4632 | |
12e9b892 | 4633 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
cf108bca | 4634 | ext4_ext_truncate(inode); |
1d03ec98 AK |
4635 | return; |
4636 | } | |
a86c6181 | 4637 | |
ac27a0ec | 4638 | handle = start_transaction(inode); |
cf108bca | 4639 | if (IS_ERR(handle)) |
ac27a0ec | 4640 | return; /* AKPM: return what? */ |
ac27a0ec DK |
4641 | |
4642 | last_block = (inode->i_size + blocksize-1) | |
617ba13b | 4643 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
ac27a0ec | 4644 | |
cf108bca JK |
4645 | if (inode->i_size & (blocksize - 1)) |
4646 | if (ext4_block_truncate_page(handle, mapping, inode->i_size)) | |
4647 | goto out_stop; | |
ac27a0ec | 4648 | |
617ba13b | 4649 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
ac27a0ec DK |
4650 | if (n == 0) |
4651 | goto out_stop; /* error */ | |
4652 | ||
4653 | /* | |
4654 | * OK. This truncate is going to happen. We add the inode to the | |
4655 | * orphan list, so that if this truncate spans multiple transactions, | |
4656 | * and we crash, we will resume the truncate when the filesystem | |
4657 | * recovers. It also marks the inode dirty, to catch the new size. | |
4658 | * | |
4659 | * Implication: the file must always be in a sane, consistent | |
4660 | * truncatable state while each transaction commits. | |
4661 | */ | |
617ba13b | 4662 | if (ext4_orphan_add(handle, inode)) |
ac27a0ec DK |
4663 | goto out_stop; |
4664 | ||
632eaeab MC |
4665 | /* |
4666 | * From here we block out all ext4_get_block() callers who want to | |
4667 | * modify the block allocation tree. | |
4668 | */ | |
4669 | down_write(&ei->i_data_sem); | |
b4df2030 | 4670 | |
c2ea3fde | 4671 | ext4_discard_preallocations(inode); |
b4df2030 | 4672 | |
ac27a0ec DK |
4673 | /* |
4674 | * The orphan list entry will now protect us from any crash which | |
4675 | * occurs before the truncate completes, so it is now safe to propagate | |
4676 | * the new, shorter inode size (held for now in i_size) into the | |
4677 | * on-disk inode. We do this via i_disksize, which is the value which | |
617ba13b | 4678 | * ext4 *really* writes onto the disk inode. |
ac27a0ec DK |
4679 | */ |
4680 | ei->i_disksize = inode->i_size; | |
4681 | ||
ac27a0ec | 4682 | if (n == 1) { /* direct blocks */ |
617ba13b MC |
4683 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
4684 | i_data + EXT4_NDIR_BLOCKS); | |
ac27a0ec DK |
4685 | goto do_indirects; |
4686 | } | |
4687 | ||
617ba13b | 4688 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
ac27a0ec DK |
4689 | /* Kill the top of shared branch (not detached) */ |
4690 | if (nr) { | |
4691 | if (partial == chain) { | |
4692 | /* Shared branch grows from the inode */ | |
617ba13b | 4693 | ext4_free_branches(handle, inode, NULL, |
ac27a0ec DK |
4694 | &nr, &nr+1, (chain+n-1) - partial); |
4695 | *partial->p = 0; | |
4696 | /* | |
4697 | * We mark the inode dirty prior to restart, | |
4698 | * and prior to stop. No need for it here. | |
4699 | */ | |
4700 | } else { | |
4701 | /* Shared branch grows from an indirect block */ | |
4702 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
617ba13b | 4703 | ext4_free_branches(handle, inode, partial->bh, |
ac27a0ec DK |
4704 | partial->p, |
4705 | partial->p+1, (chain+n-1) - partial); | |
4706 | } | |
4707 | } | |
4708 | /* Clear the ends of indirect blocks on the shared branch */ | |
4709 | while (partial > chain) { | |
617ba13b | 4710 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
ac27a0ec DK |
4711 | (__le32*)partial->bh->b_data+addr_per_block, |
4712 | (chain+n-1) - partial); | |
4713 | BUFFER_TRACE(partial->bh, "call brelse"); | |
de9a55b8 | 4714 | brelse(partial->bh); |
ac27a0ec DK |
4715 | partial--; |
4716 | } | |
4717 | do_indirects: | |
4718 | /* Kill the remaining (whole) subtrees */ | |
4719 | switch (offsets[0]) { | |
4720 | default: | |
617ba13b | 4721 | nr = i_data[EXT4_IND_BLOCK]; |
ac27a0ec | 4722 | if (nr) { |
617ba13b MC |
4723 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
4724 | i_data[EXT4_IND_BLOCK] = 0; | |
ac27a0ec | 4725 | } |
617ba13b MC |
4726 | case EXT4_IND_BLOCK: |
4727 | nr = i_data[EXT4_DIND_BLOCK]; | |
ac27a0ec | 4728 | if (nr) { |
617ba13b MC |
4729 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
4730 | i_data[EXT4_DIND_BLOCK] = 0; | |
ac27a0ec | 4731 | } |
617ba13b MC |
4732 | case EXT4_DIND_BLOCK: |
4733 | nr = i_data[EXT4_TIND_BLOCK]; | |
ac27a0ec | 4734 | if (nr) { |
617ba13b MC |
4735 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
4736 | i_data[EXT4_TIND_BLOCK] = 0; | |
ac27a0ec | 4737 | } |
617ba13b | 4738 | case EXT4_TIND_BLOCK: |
ac27a0ec DK |
4739 | ; |
4740 | } | |
4741 | ||
0e855ac8 | 4742 | up_write(&ei->i_data_sem); |
ef7f3835 | 4743 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
617ba13b | 4744 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
4745 | |
4746 | /* | |
4747 | * In a multi-transaction truncate, we only make the final transaction | |
4748 | * synchronous | |
4749 | */ | |
4750 | if (IS_SYNC(inode)) | |
0390131b | 4751 | ext4_handle_sync(handle); |
ac27a0ec DK |
4752 | out_stop: |
4753 | /* | |
4754 | * If this was a simple ftruncate(), and the file will remain alive | |
4755 | * then we need to clear up the orphan record which we created above. | |
4756 | * However, if this was a real unlink then we were called by | |
617ba13b | 4757 | * ext4_delete_inode(), and we allow that function to clean up the |
ac27a0ec DK |
4758 | * orphan info for us. |
4759 | */ | |
4760 | if (inode->i_nlink) | |
617ba13b | 4761 | ext4_orphan_del(handle, inode); |
ac27a0ec | 4762 | |
617ba13b | 4763 | ext4_journal_stop(handle); |
ac27a0ec DK |
4764 | } |
4765 | ||
ac27a0ec | 4766 | /* |
617ba13b | 4767 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4768 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4769 | * data in memory that is needed to recreate the on-disk version of this | |
4770 | * inode. | |
4771 | */ | |
617ba13b MC |
4772 | static int __ext4_get_inode_loc(struct inode *inode, |
4773 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4774 | { |
240799cd TT |
4775 | struct ext4_group_desc *gdp; |
4776 | struct buffer_head *bh; | |
4777 | struct super_block *sb = inode->i_sb; | |
4778 | ext4_fsblk_t block; | |
4779 | int inodes_per_block, inode_offset; | |
4780 | ||
3a06d778 | 4781 | iloc->bh = NULL; |
240799cd TT |
4782 | if (!ext4_valid_inum(sb, inode->i_ino)) |
4783 | return -EIO; | |
ac27a0ec | 4784 | |
240799cd TT |
4785 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4786 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4787 | if (!gdp) | |
ac27a0ec DK |
4788 | return -EIO; |
4789 | ||
240799cd TT |
4790 | /* |
4791 | * Figure out the offset within the block group inode table | |
4792 | */ | |
4793 | inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb)); | |
4794 | inode_offset = ((inode->i_ino - 1) % | |
4795 | EXT4_INODES_PER_GROUP(sb)); | |
4796 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4797 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4798 | ||
4799 | bh = sb_getblk(sb, block); | |
ac27a0ec | 4800 | if (!bh) { |
c398eda0 TT |
4801 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4802 | "unable to read itable block"); | |
ac27a0ec DK |
4803 | return -EIO; |
4804 | } | |
4805 | if (!buffer_uptodate(bh)) { | |
4806 | lock_buffer(bh); | |
9c83a923 HK |
4807 | |
4808 | /* | |
4809 | * If the buffer has the write error flag, we have failed | |
4810 | * to write out another inode in the same block. In this | |
4811 | * case, we don't have to read the block because we may | |
4812 | * read the old inode data successfully. | |
4813 | */ | |
4814 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4815 | set_buffer_uptodate(bh); | |
4816 | ||
ac27a0ec DK |
4817 | if (buffer_uptodate(bh)) { |
4818 | /* someone brought it uptodate while we waited */ | |
4819 | unlock_buffer(bh); | |
4820 | goto has_buffer; | |
4821 | } | |
4822 | ||
4823 | /* | |
4824 | * If we have all information of the inode in memory and this | |
4825 | * is the only valid inode in the block, we need not read the | |
4826 | * block. | |
4827 | */ | |
4828 | if (in_mem) { | |
4829 | struct buffer_head *bitmap_bh; | |
240799cd | 4830 | int i, start; |
ac27a0ec | 4831 | |
240799cd | 4832 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4833 | |
240799cd TT |
4834 | /* Is the inode bitmap in cache? */ |
4835 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
4836 | if (!bitmap_bh) |
4837 | goto make_io; | |
4838 | ||
4839 | /* | |
4840 | * If the inode bitmap isn't in cache then the | |
4841 | * optimisation may end up performing two reads instead | |
4842 | * of one, so skip it. | |
4843 | */ | |
4844 | if (!buffer_uptodate(bitmap_bh)) { | |
4845 | brelse(bitmap_bh); | |
4846 | goto make_io; | |
4847 | } | |
240799cd | 4848 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4849 | if (i == inode_offset) |
4850 | continue; | |
617ba13b | 4851 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4852 | break; |
4853 | } | |
4854 | brelse(bitmap_bh); | |
240799cd | 4855 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4856 | /* all other inodes are free, so skip I/O */ |
4857 | memset(bh->b_data, 0, bh->b_size); | |
4858 | set_buffer_uptodate(bh); | |
4859 | unlock_buffer(bh); | |
4860 | goto has_buffer; | |
4861 | } | |
4862 | } | |
4863 | ||
4864 | make_io: | |
240799cd TT |
4865 | /* |
4866 | * If we need to do any I/O, try to pre-readahead extra | |
4867 | * blocks from the inode table. | |
4868 | */ | |
4869 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4870 | ext4_fsblk_t b, end, table; | |
4871 | unsigned num; | |
4872 | ||
4873 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4874 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
4875 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
4876 | if (table > b) | |
4877 | b = table; | |
4878 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
4879 | num = EXT4_INODES_PER_GROUP(sb); | |
4880 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4881 | EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) | |
560671a0 | 4882 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4883 | table += num / inodes_per_block; |
4884 | if (end > table) | |
4885 | end = table; | |
4886 | while (b <= end) | |
4887 | sb_breadahead(sb, b++); | |
4888 | } | |
4889 | ||
ac27a0ec DK |
4890 | /* |
4891 | * There are other valid inodes in the buffer, this inode | |
4892 | * has in-inode xattrs, or we don't have this inode in memory. | |
4893 | * Read the block from disk. | |
4894 | */ | |
4895 | get_bh(bh); | |
4896 | bh->b_end_io = end_buffer_read_sync; | |
4897 | submit_bh(READ_META, bh); | |
4898 | wait_on_buffer(bh); | |
4899 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4900 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4901 | "unable to read itable block"); | |
ac27a0ec DK |
4902 | brelse(bh); |
4903 | return -EIO; | |
4904 | } | |
4905 | } | |
4906 | has_buffer: | |
4907 | iloc->bh = bh; | |
4908 | return 0; | |
4909 | } | |
4910 | ||
617ba13b | 4911 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4912 | { |
4913 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4914 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4915 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4916 | } |
4917 | ||
617ba13b | 4918 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4919 | { |
617ba13b | 4920 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
4921 | |
4922 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 4923 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 4924 | inode->i_flags |= S_SYNC; |
617ba13b | 4925 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 4926 | inode->i_flags |= S_APPEND; |
617ba13b | 4927 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 4928 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 4929 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 4930 | inode->i_flags |= S_NOATIME; |
617ba13b | 4931 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
4932 | inode->i_flags |= S_DIRSYNC; |
4933 | } | |
4934 | ||
ff9ddf7e JK |
4935 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4936 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4937 | { | |
84a8dce2 DM |
4938 | unsigned int vfs_fl; |
4939 | unsigned long old_fl, new_fl; | |
4940 | ||
4941 | do { | |
4942 | vfs_fl = ei->vfs_inode.i_flags; | |
4943 | old_fl = ei->i_flags; | |
4944 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4945 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4946 | EXT4_DIRSYNC_FL); | |
4947 | if (vfs_fl & S_SYNC) | |
4948 | new_fl |= EXT4_SYNC_FL; | |
4949 | if (vfs_fl & S_APPEND) | |
4950 | new_fl |= EXT4_APPEND_FL; | |
4951 | if (vfs_fl & S_IMMUTABLE) | |
4952 | new_fl |= EXT4_IMMUTABLE_FL; | |
4953 | if (vfs_fl & S_NOATIME) | |
4954 | new_fl |= EXT4_NOATIME_FL; | |
4955 | if (vfs_fl & S_DIRSYNC) | |
4956 | new_fl |= EXT4_DIRSYNC_FL; | |
4957 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4958 | } |
de9a55b8 | 4959 | |
0fc1b451 | 4960 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4961 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4962 | { |
4963 | blkcnt_t i_blocks ; | |
8180a562 AK |
4964 | struct inode *inode = &(ei->vfs_inode); |
4965 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4966 | |
4967 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4968 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4969 | /* we are using combined 48 bit field */ | |
4970 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4971 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4972 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4973 | /* i_blocks represent file system block size */ |
4974 | return i_blocks << (inode->i_blkbits - 9); | |
4975 | } else { | |
4976 | return i_blocks; | |
4977 | } | |
0fc1b451 AK |
4978 | } else { |
4979 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4980 | } | |
4981 | } | |
ff9ddf7e | 4982 | |
1d1fe1ee | 4983 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4984 | { |
617ba13b MC |
4985 | struct ext4_iloc iloc; |
4986 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4987 | struct ext4_inode_info *ei; |
1d1fe1ee | 4988 | struct inode *inode; |
b436b9be | 4989 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4990 | long ret; |
ac27a0ec DK |
4991 | int block; |
4992 | ||
1d1fe1ee DH |
4993 | inode = iget_locked(sb, ino); |
4994 | if (!inode) | |
4995 | return ERR_PTR(-ENOMEM); | |
4996 | if (!(inode->i_state & I_NEW)) | |
4997 | return inode; | |
4998 | ||
4999 | ei = EXT4_I(inode); | |
567f3e9a | 5000 | iloc.bh = 0; |
ac27a0ec | 5001 | |
1d1fe1ee DH |
5002 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
5003 | if (ret < 0) | |
ac27a0ec | 5004 | goto bad_inode; |
617ba13b | 5005 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
5006 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
5007 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
5008 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 5009 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
5010 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
5011 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
5012 | } | |
5013 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
ac27a0ec | 5014 | |
19f5fb7a | 5015 | ei->i_state_flags = 0; |
ac27a0ec DK |
5016 | ei->i_dir_start_lookup = 0; |
5017 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
5018 | /* We now have enough fields to check if the inode was active or not. | |
5019 | * This is needed because nfsd might try to access dead inodes | |
5020 | * the test is that same one that e2fsck uses | |
5021 | * NeilBrown 1999oct15 | |
5022 | */ | |
5023 | if (inode->i_nlink == 0) { | |
5024 | if (inode->i_mode == 0 || | |
617ba13b | 5025 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 5026 | /* this inode is deleted */ |
1d1fe1ee | 5027 | ret = -ESTALE; |
ac27a0ec DK |
5028 | goto bad_inode; |
5029 | } | |
5030 | /* The only unlinked inodes we let through here have | |
5031 | * valid i_mode and are being read by the orphan | |
5032 | * recovery code: that's fine, we're about to complete | |
5033 | * the process of deleting those. */ | |
5034 | } | |
ac27a0ec | 5035 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 5036 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 5037 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 5038 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
5039 | ei->i_file_acl |= |
5040 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 5041 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 5042 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
5043 | #ifdef CONFIG_QUOTA |
5044 | ei->i_reserved_quota = 0; | |
5045 | #endif | |
ac27a0ec DK |
5046 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
5047 | ei->i_block_group = iloc.block_group; | |
a4912123 | 5048 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
5049 | /* |
5050 | * NOTE! The in-memory inode i_data array is in little-endian order | |
5051 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
5052 | */ | |
617ba13b | 5053 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
5054 | ei->i_data[block] = raw_inode->i_block[block]; |
5055 | INIT_LIST_HEAD(&ei->i_orphan); | |
5056 | ||
b436b9be JK |
5057 | /* |
5058 | * Set transaction id's of transactions that have to be committed | |
5059 | * to finish f[data]sync. We set them to currently running transaction | |
5060 | * as we cannot be sure that the inode or some of its metadata isn't | |
5061 | * part of the transaction - the inode could have been reclaimed and | |
5062 | * now it is reread from disk. | |
5063 | */ | |
5064 | if (journal) { | |
5065 | transaction_t *transaction; | |
5066 | tid_t tid; | |
5067 | ||
5068 | spin_lock(&journal->j_state_lock); | |
5069 | if (journal->j_running_transaction) | |
5070 | transaction = journal->j_running_transaction; | |
5071 | else | |
5072 | transaction = journal->j_committing_transaction; | |
5073 | if (transaction) | |
5074 | tid = transaction->t_tid; | |
5075 | else | |
5076 | tid = journal->j_commit_sequence; | |
5077 | spin_unlock(&journal->j_state_lock); | |
5078 | ei->i_sync_tid = tid; | |
5079 | ei->i_datasync_tid = tid; | |
5080 | } | |
5081 | ||
0040d987 | 5082 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec | 5083 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
617ba13b | 5084 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f | 5085 | EXT4_INODE_SIZE(inode->i_sb)) { |
1d1fe1ee | 5086 | ret = -EIO; |
ac27a0ec | 5087 | goto bad_inode; |
e5d2861f | 5088 | } |
ac27a0ec DK |
5089 | if (ei->i_extra_isize == 0) { |
5090 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
5091 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
5092 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
5093 | } else { |
5094 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 5095 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 5096 | ei->i_extra_isize; |
617ba13b | 5097 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
19f5fb7a | 5098 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
ac27a0ec DK |
5099 | } |
5100 | } else | |
5101 | ei->i_extra_isize = 0; | |
5102 | ||
ef7f3835 KS |
5103 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
5104 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
5105 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
5106 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
5107 | ||
25ec56b5 JNC |
5108 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
5109 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
5110 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5111 | inode->i_version |= | |
5112 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
5113 | } | |
5114 | ||
c4b5a614 | 5115 | ret = 0; |
485c26ec | 5116 | if (ei->i_file_acl && |
1032988c | 5117 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
5118 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
5119 | ei->i_file_acl); | |
485c26ec TT |
5120 | ret = -EIO; |
5121 | goto bad_inode; | |
07a03824 | 5122 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
c4b5a614 TT |
5123 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
5124 | (S_ISLNK(inode->i_mode) && | |
5125 | !ext4_inode_is_fast_symlink(inode))) | |
5126 | /* Validate extent which is part of inode */ | |
5127 | ret = ext4_ext_check_inode(inode); | |
de9a55b8 | 5128 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
fe2c8191 TN |
5129 | (S_ISLNK(inode->i_mode) && |
5130 | !ext4_inode_is_fast_symlink(inode))) { | |
de9a55b8 | 5131 | /* Validate block references which are part of inode */ |
fe2c8191 TN |
5132 | ret = ext4_check_inode_blockref(inode); |
5133 | } | |
567f3e9a | 5134 | if (ret) |
de9a55b8 | 5135 | goto bad_inode; |
7a262f7c | 5136 | |
ac27a0ec | 5137 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
5138 | inode->i_op = &ext4_file_inode_operations; |
5139 | inode->i_fop = &ext4_file_operations; | |
5140 | ext4_set_aops(inode); | |
ac27a0ec | 5141 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
5142 | inode->i_op = &ext4_dir_inode_operations; |
5143 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 5144 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 5145 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 5146 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
5147 | nd_terminate_link(ei->i_data, inode->i_size, |
5148 | sizeof(ei->i_data) - 1); | |
5149 | } else { | |
617ba13b MC |
5150 | inode->i_op = &ext4_symlink_inode_operations; |
5151 | ext4_set_aops(inode); | |
ac27a0ec | 5152 | } |
563bdd61 TT |
5153 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
5154 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 5155 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
5156 | if (raw_inode->i_block[0]) |
5157 | init_special_inode(inode, inode->i_mode, | |
5158 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
5159 | else | |
5160 | init_special_inode(inode, inode->i_mode, | |
5161 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 | 5162 | } else { |
563bdd61 | 5163 | ret = -EIO; |
24676da4 | 5164 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 5165 | goto bad_inode; |
ac27a0ec | 5166 | } |
af5bc92d | 5167 | brelse(iloc.bh); |
617ba13b | 5168 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
5169 | unlock_new_inode(inode); |
5170 | return inode; | |
ac27a0ec DK |
5171 | |
5172 | bad_inode: | |
567f3e9a | 5173 | brelse(iloc.bh); |
1d1fe1ee DH |
5174 | iget_failed(inode); |
5175 | return ERR_PTR(ret); | |
ac27a0ec DK |
5176 | } |
5177 | ||
0fc1b451 AK |
5178 | static int ext4_inode_blocks_set(handle_t *handle, |
5179 | struct ext4_inode *raw_inode, | |
5180 | struct ext4_inode_info *ei) | |
5181 | { | |
5182 | struct inode *inode = &(ei->vfs_inode); | |
5183 | u64 i_blocks = inode->i_blocks; | |
5184 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
5185 | |
5186 | if (i_blocks <= ~0U) { | |
5187 | /* | |
5188 | * i_blocks can be represnted in a 32 bit variable | |
5189 | * as multiple of 512 bytes | |
5190 | */ | |
8180a562 | 5191 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5192 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 5193 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
5194 | return 0; |
5195 | } | |
5196 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
5197 | return -EFBIG; | |
5198 | ||
5199 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
5200 | /* |
5201 | * i_blocks can be represented in a 48 bit variable | |
5202 | * as multiple of 512 bytes | |
5203 | */ | |
8180a562 | 5204 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5205 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 5206 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 5207 | } else { |
84a8dce2 | 5208 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
5209 | /* i_block is stored in file system block size */ |
5210 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
5211 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
5212 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 5213 | } |
f287a1a5 | 5214 | return 0; |
0fc1b451 AK |
5215 | } |
5216 | ||
ac27a0ec DK |
5217 | /* |
5218 | * Post the struct inode info into an on-disk inode location in the | |
5219 | * buffer-cache. This gobbles the caller's reference to the | |
5220 | * buffer_head in the inode location struct. | |
5221 | * | |
5222 | * The caller must have write access to iloc->bh. | |
5223 | */ | |
617ba13b | 5224 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 5225 | struct inode *inode, |
830156c7 | 5226 | struct ext4_iloc *iloc) |
ac27a0ec | 5227 | { |
617ba13b MC |
5228 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
5229 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
5230 | struct buffer_head *bh = iloc->bh; |
5231 | int err = 0, rc, block; | |
5232 | ||
5233 | /* For fields not not tracking in the in-memory inode, | |
5234 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 5235 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 5236 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 5237 | |
ff9ddf7e | 5238 | ext4_get_inode_flags(ei); |
ac27a0ec | 5239 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
af5bc92d | 5240 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
5241 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
5242 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
5243 | /* | |
5244 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
5245 | * re-used with the upper 16 bits of the uid/gid intact | |
5246 | */ | |
af5bc92d | 5247 | if (!ei->i_dtime) { |
ac27a0ec DK |
5248 | raw_inode->i_uid_high = |
5249 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
5250 | raw_inode->i_gid_high = | |
5251 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
5252 | } else { | |
5253 | raw_inode->i_uid_high = 0; | |
5254 | raw_inode->i_gid_high = 0; | |
5255 | } | |
5256 | } else { | |
5257 | raw_inode->i_uid_low = | |
5258 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
5259 | raw_inode->i_gid_low = | |
5260 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
5261 | raw_inode->i_uid_high = 0; | |
5262 | raw_inode->i_gid_high = 0; | |
5263 | } | |
5264 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5265 | |
5266 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5267 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5268 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5269 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5270 | ||
0fc1b451 AK |
5271 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
5272 | goto out_brelse; | |
ac27a0ec | 5273 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
1b9c12f4 | 5274 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); |
9b8f1f01 MC |
5275 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
5276 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
5277 | raw_inode->i_file_acl_high = |
5278 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5279 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
5280 | ext4_isize_set(raw_inode, ei->i_disksize); |
5281 | if (ei->i_disksize > 0x7fffffffULL) { | |
5282 | struct super_block *sb = inode->i_sb; | |
5283 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
5284 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
5285 | EXT4_SB(sb)->s_es->s_rev_level == | |
5286 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
5287 | /* If this is the first large file | |
5288 | * created, add a flag to the superblock. | |
5289 | */ | |
5290 | err = ext4_journal_get_write_access(handle, | |
5291 | EXT4_SB(sb)->s_sbh); | |
5292 | if (err) | |
5293 | goto out_brelse; | |
5294 | ext4_update_dynamic_rev(sb); | |
5295 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 5296 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
a48380f7 | 5297 | sb->s_dirt = 1; |
0390131b | 5298 | ext4_handle_sync(handle); |
73b50c1c | 5299 | err = ext4_handle_dirty_metadata(handle, NULL, |
a48380f7 | 5300 | EXT4_SB(sb)->s_sbh); |
ac27a0ec DK |
5301 | } |
5302 | } | |
5303 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5304 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5305 | if (old_valid_dev(inode->i_rdev)) { | |
5306 | raw_inode->i_block[0] = | |
5307 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5308 | raw_inode->i_block[1] = 0; | |
5309 | } else { | |
5310 | raw_inode->i_block[0] = 0; | |
5311 | raw_inode->i_block[1] = | |
5312 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5313 | raw_inode->i_block[2] = 0; | |
5314 | } | |
de9a55b8 TT |
5315 | } else |
5316 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
5317 | raw_inode->i_block[block] = ei->i_data[block]; | |
ac27a0ec | 5318 | |
25ec56b5 JNC |
5319 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
5320 | if (ei->i_extra_isize) { | |
5321 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5322 | raw_inode->i_version_hi = | |
5323 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 5324 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
5325 | } |
5326 | ||
830156c7 | 5327 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 5328 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
5329 | if (!err) |
5330 | err = rc; | |
19f5fb7a | 5331 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 5332 | |
b436b9be | 5333 | ext4_update_inode_fsync_trans(handle, inode, 0); |
ac27a0ec | 5334 | out_brelse: |
af5bc92d | 5335 | brelse(bh); |
617ba13b | 5336 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5337 | return err; |
5338 | } | |
5339 | ||
5340 | /* | |
617ba13b | 5341 | * ext4_write_inode() |
ac27a0ec DK |
5342 | * |
5343 | * We are called from a few places: | |
5344 | * | |
5345 | * - Within generic_file_write() for O_SYNC files. | |
5346 | * Here, there will be no transaction running. We wait for any running | |
5347 | * trasnaction to commit. | |
5348 | * | |
5349 | * - Within sys_sync(), kupdate and such. | |
5350 | * We wait on commit, if tol to. | |
5351 | * | |
5352 | * - Within prune_icache() (PF_MEMALLOC == true) | |
5353 | * Here we simply return. We can't afford to block kswapd on the | |
5354 | * journal commit. | |
5355 | * | |
5356 | * In all cases it is actually safe for us to return without doing anything, | |
5357 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 5358 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
5359 | * knfsd. |
5360 | * | |
5361 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5362 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5363 | * which we are interested. | |
5364 | * | |
5365 | * It would be a bug for them to not do this. The code: | |
5366 | * | |
5367 | * mark_inode_dirty(inode) | |
5368 | * stuff(); | |
5369 | * inode->i_size = expr; | |
5370 | * | |
5371 | * is in error because a kswapd-driven write_inode() could occur while | |
5372 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
5373 | * will no longer be on the superblock's dirty inode list. | |
5374 | */ | |
a9185b41 | 5375 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5376 | { |
91ac6f43 FM |
5377 | int err; |
5378 | ||
ac27a0ec DK |
5379 | if (current->flags & PF_MEMALLOC) |
5380 | return 0; | |
5381 | ||
91ac6f43 FM |
5382 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5383 | if (ext4_journal_current_handle()) { | |
5384 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5385 | dump_stack(); | |
5386 | return -EIO; | |
5387 | } | |
ac27a0ec | 5388 | |
a9185b41 | 5389 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
5390 | return 0; |
5391 | ||
5392 | err = ext4_force_commit(inode->i_sb); | |
5393 | } else { | |
5394 | struct ext4_iloc iloc; | |
ac27a0ec | 5395 | |
8b472d73 | 5396 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
5397 | if (err) |
5398 | return err; | |
a9185b41 | 5399 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
5400 | sync_dirty_buffer(iloc.bh); |
5401 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
5402 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
5403 | "IO error syncing inode"); | |
830156c7 FM |
5404 | err = -EIO; |
5405 | } | |
fd2dd9fb | 5406 | brelse(iloc.bh); |
91ac6f43 FM |
5407 | } |
5408 | return err; | |
ac27a0ec DK |
5409 | } |
5410 | ||
5411 | /* | |
617ba13b | 5412 | * ext4_setattr() |
ac27a0ec DK |
5413 | * |
5414 | * Called from notify_change. | |
5415 | * | |
5416 | * We want to trap VFS attempts to truncate the file as soon as | |
5417 | * possible. In particular, we want to make sure that when the VFS | |
5418 | * shrinks i_size, we put the inode on the orphan list and modify | |
5419 | * i_disksize immediately, so that during the subsequent flushing of | |
5420 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5421 | * commit will leave the blocks being flushed in an unused state on | |
5422 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5423 | * be freed, so we have a strong guarantee that no future commit will | |
5424 | * leave these blocks visible to the user.) | |
5425 | * | |
678aaf48 JK |
5426 | * Another thing we have to assure is that if we are in ordered mode |
5427 | * and inode is still attached to the committing transaction, we must | |
5428 | * we start writeout of all the dirty pages which are being truncated. | |
5429 | * This way we are sure that all the data written in the previous | |
5430 | * transaction are already on disk (truncate waits for pages under | |
5431 | * writeback). | |
5432 | * | |
5433 | * Called with inode->i_mutex down. | |
ac27a0ec | 5434 | */ |
617ba13b | 5435 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
5436 | { |
5437 | struct inode *inode = dentry->d_inode; | |
5438 | int error, rc = 0; | |
5439 | const unsigned int ia_valid = attr->ia_valid; | |
5440 | ||
5441 | error = inode_change_ok(inode, attr); | |
5442 | if (error) | |
5443 | return error; | |
5444 | ||
12755627 | 5445 | if (is_quota_modification(inode, attr)) |
871a2931 | 5446 | dquot_initialize(inode); |
ac27a0ec DK |
5447 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
5448 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
5449 | handle_t *handle; | |
5450 | ||
5451 | /* (user+group)*(old+new) structure, inode write (sb, | |
5452 | * inode block, ? - but truncate inode update has it) */ | |
5aca07eb | 5453 | handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
194074ac | 5454 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3); |
ac27a0ec DK |
5455 | if (IS_ERR(handle)) { |
5456 | error = PTR_ERR(handle); | |
5457 | goto err_out; | |
5458 | } | |
b43fa828 | 5459 | error = dquot_transfer(inode, attr); |
ac27a0ec | 5460 | if (error) { |
617ba13b | 5461 | ext4_journal_stop(handle); |
ac27a0ec DK |
5462 | return error; |
5463 | } | |
5464 | /* Update corresponding info in inode so that everything is in | |
5465 | * one transaction */ | |
5466 | if (attr->ia_valid & ATTR_UID) | |
5467 | inode->i_uid = attr->ia_uid; | |
5468 | if (attr->ia_valid & ATTR_GID) | |
5469 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5470 | error = ext4_mark_inode_dirty(handle, inode); |
5471 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5472 | } |
5473 | ||
e2b46574 | 5474 | if (attr->ia_valid & ATTR_SIZE) { |
12e9b892 | 5475 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5476 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5477 | ||
0c095c7f TT |
5478 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5479 | return -EFBIG; | |
e2b46574 ES |
5480 | } |
5481 | } | |
5482 | ||
ac27a0ec | 5483 | if (S_ISREG(inode->i_mode) && |
c8d46e41 JZ |
5484 | attr->ia_valid & ATTR_SIZE && |
5485 | (attr->ia_size < inode->i_size || | |
12e9b892 | 5486 | (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) { |
ac27a0ec DK |
5487 | handle_t *handle; |
5488 | ||
617ba13b | 5489 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
5490 | if (IS_ERR(handle)) { |
5491 | error = PTR_ERR(handle); | |
5492 | goto err_out; | |
5493 | } | |
5494 | ||
617ba13b MC |
5495 | error = ext4_orphan_add(handle, inode); |
5496 | EXT4_I(inode)->i_disksize = attr->ia_size; | |
5497 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
5498 | if (!error) |
5499 | error = rc; | |
617ba13b | 5500 | ext4_journal_stop(handle); |
678aaf48 JK |
5501 | |
5502 | if (ext4_should_order_data(inode)) { | |
5503 | error = ext4_begin_ordered_truncate(inode, | |
5504 | attr->ia_size); | |
5505 | if (error) { | |
5506 | /* Do as much error cleanup as possible */ | |
5507 | handle = ext4_journal_start(inode, 3); | |
5508 | if (IS_ERR(handle)) { | |
5509 | ext4_orphan_del(NULL, inode); | |
5510 | goto err_out; | |
5511 | } | |
5512 | ext4_orphan_del(handle, inode); | |
5513 | ext4_journal_stop(handle); | |
5514 | goto err_out; | |
5515 | } | |
5516 | } | |
c8d46e41 | 5517 | /* ext4_truncate will clear the flag */ |
12e9b892 | 5518 | if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))) |
c8d46e41 | 5519 | ext4_truncate(inode); |
ac27a0ec DK |
5520 | } |
5521 | ||
5522 | rc = inode_setattr(inode, attr); | |
5523 | ||
617ba13b | 5524 | /* If inode_setattr's call to ext4_truncate failed to get a |
ac27a0ec DK |
5525 | * transaction handle at all, we need to clean up the in-core |
5526 | * orphan list manually. */ | |
5527 | if (inode->i_nlink) | |
617ba13b | 5528 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
5529 | |
5530 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 5531 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
5532 | |
5533 | err_out: | |
617ba13b | 5534 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5535 | if (!error) |
5536 | error = rc; | |
5537 | return error; | |
5538 | } | |
5539 | ||
3e3398a0 MC |
5540 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
5541 | struct kstat *stat) | |
5542 | { | |
5543 | struct inode *inode; | |
5544 | unsigned long delalloc_blocks; | |
5545 | ||
5546 | inode = dentry->d_inode; | |
5547 | generic_fillattr(inode, stat); | |
5548 | ||
5549 | /* | |
5550 | * We can't update i_blocks if the block allocation is delayed | |
5551 | * otherwise in the case of system crash before the real block | |
5552 | * allocation is done, we will have i_blocks inconsistent with | |
5553 | * on-disk file blocks. | |
5554 | * We always keep i_blocks updated together with real | |
5555 | * allocation. But to not confuse with user, stat | |
5556 | * will return the blocks that include the delayed allocation | |
5557 | * blocks for this file. | |
5558 | */ | |
5559 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
5560 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
5561 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
5562 | ||
5563 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
5564 | return 0; | |
5565 | } | |
ac27a0ec | 5566 | |
a02908f1 MC |
5567 | static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks, |
5568 | int chunk) | |
5569 | { | |
5570 | int indirects; | |
5571 | ||
5572 | /* if nrblocks are contiguous */ | |
5573 | if (chunk) { | |
5574 | /* | |
5575 | * With N contiguous data blocks, it need at most | |
5576 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks | |
5577 | * 2 dindirect blocks | |
5578 | * 1 tindirect block | |
5579 | */ | |
5580 | indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
5581 | return indirects + 3; | |
5582 | } | |
5583 | /* | |
5584 | * if nrblocks are not contiguous, worse case, each block touch | |
5585 | * a indirect block, and each indirect block touch a double indirect | |
5586 | * block, plus a triple indirect block | |
5587 | */ | |
5588 | indirects = nrblocks * 2 + 1; | |
5589 | return indirects; | |
5590 | } | |
5591 | ||
5592 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
5593 | { | |
12e9b892 | 5594 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
ac51d837 TT |
5595 | return ext4_indirect_trans_blocks(inode, nrblocks, chunk); |
5596 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); | |
a02908f1 | 5597 | } |
ac51d837 | 5598 | |
ac27a0ec | 5599 | /* |
a02908f1 MC |
5600 | * Account for index blocks, block groups bitmaps and block group |
5601 | * descriptor blocks if modify datablocks and index blocks | |
5602 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5603 | * |
a02908f1 | 5604 | * If datablocks are discontiguous, they are possible to spread over |
af901ca1 | 5605 | * different block groups too. If they are contiuguous, with flexbg, |
a02908f1 | 5606 | * they could still across block group boundary. |
ac27a0ec | 5607 | * |
a02908f1 MC |
5608 | * Also account for superblock, inode, quota and xattr blocks |
5609 | */ | |
5610 | int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
5611 | { | |
8df9675f TT |
5612 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5613 | int gdpblocks; | |
a02908f1 MC |
5614 | int idxblocks; |
5615 | int ret = 0; | |
5616 | ||
5617 | /* | |
5618 | * How many index blocks need to touch to modify nrblocks? | |
5619 | * The "Chunk" flag indicating whether the nrblocks is | |
5620 | * physically contiguous on disk | |
5621 | * | |
5622 | * For Direct IO and fallocate, they calls get_block to allocate | |
5623 | * one single extent at a time, so they could set the "Chunk" flag | |
5624 | */ | |
5625 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
5626 | ||
5627 | ret = idxblocks; | |
5628 | ||
5629 | /* | |
5630 | * Now let's see how many group bitmaps and group descriptors need | |
5631 | * to account | |
5632 | */ | |
5633 | groups = idxblocks; | |
5634 | if (chunk) | |
5635 | groups += 1; | |
5636 | else | |
5637 | groups += nrblocks; | |
5638 | ||
5639 | gdpblocks = groups; | |
8df9675f TT |
5640 | if (groups > ngroups) |
5641 | groups = ngroups; | |
a02908f1 MC |
5642 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5643 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5644 | ||
5645 | /* bitmaps and block group descriptor blocks */ | |
5646 | ret += groups + gdpblocks; | |
5647 | ||
5648 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5649 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5650 | ||
5651 | return ret; | |
5652 | } | |
5653 | ||
5654 | /* | |
5655 | * Calulate the total number of credits to reserve to fit | |
f3bd1f3f MC |
5656 | * the modification of a single pages into a single transaction, |
5657 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5658 | * |
525f4ed8 | 5659 | * This could be called via ext4_write_begin() |
ac27a0ec | 5660 | * |
525f4ed8 | 5661 | * We need to consider the worse case, when |
a02908f1 | 5662 | * one new block per extent. |
ac27a0ec | 5663 | */ |
a86c6181 | 5664 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5665 | { |
617ba13b | 5666 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5667 | int ret; |
5668 | ||
a02908f1 | 5669 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 5670 | |
a02908f1 | 5671 | /* Account for data blocks for journalled mode */ |
617ba13b | 5672 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5673 | ret += bpp; |
ac27a0ec DK |
5674 | return ret; |
5675 | } | |
f3bd1f3f MC |
5676 | |
5677 | /* | |
5678 | * Calculate the journal credits for a chunk of data modification. | |
5679 | * | |
5680 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5681 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5682 | * |
5683 | * journal buffers for data blocks are not included here, as DIO | |
5684 | * and fallocate do no need to journal data buffers. | |
5685 | */ | |
5686 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5687 | { | |
5688 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5689 | } | |
5690 | ||
ac27a0ec | 5691 | /* |
617ba13b | 5692 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5693 | * Give this, we know that the caller already has write access to iloc->bh. |
5694 | */ | |
617ba13b | 5695 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5696 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5697 | { |
5698 | int err = 0; | |
5699 | ||
25ec56b5 JNC |
5700 | if (test_opt(inode->i_sb, I_VERSION)) |
5701 | inode_inc_iversion(inode); | |
5702 | ||
ac27a0ec DK |
5703 | /* the do_update_inode consumes one bh->b_count */ |
5704 | get_bh(iloc->bh); | |
5705 | ||
dab291af | 5706 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5707 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5708 | put_bh(iloc->bh); |
5709 | return err; | |
5710 | } | |
5711 | ||
5712 | /* | |
5713 | * On success, We end up with an outstanding reference count against | |
5714 | * iloc->bh. This _must_ be cleaned up later. | |
5715 | */ | |
5716 | ||
5717 | int | |
617ba13b MC |
5718 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5719 | struct ext4_iloc *iloc) | |
ac27a0ec | 5720 | { |
0390131b FM |
5721 | int err; |
5722 | ||
5723 | err = ext4_get_inode_loc(inode, iloc); | |
5724 | if (!err) { | |
5725 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5726 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5727 | if (err) { | |
5728 | brelse(iloc->bh); | |
5729 | iloc->bh = NULL; | |
ac27a0ec DK |
5730 | } |
5731 | } | |
617ba13b | 5732 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5733 | return err; |
5734 | } | |
5735 | ||
6dd4ee7c KS |
5736 | /* |
5737 | * Expand an inode by new_extra_isize bytes. | |
5738 | * Returns 0 on success or negative error number on failure. | |
5739 | */ | |
1d03ec98 AK |
5740 | static int ext4_expand_extra_isize(struct inode *inode, |
5741 | unsigned int new_extra_isize, | |
5742 | struct ext4_iloc iloc, | |
5743 | handle_t *handle) | |
6dd4ee7c KS |
5744 | { |
5745 | struct ext4_inode *raw_inode; | |
5746 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
5747 | |
5748 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5749 | return 0; | |
5750 | ||
5751 | raw_inode = ext4_raw_inode(&iloc); | |
5752 | ||
5753 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
5754 | |
5755 | /* No extended attributes present */ | |
19f5fb7a TT |
5756 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
5757 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
5758 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
5759 | new_extra_isize); | |
5760 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5761 | return 0; | |
5762 | } | |
5763 | ||
5764 | /* try to expand with EAs present */ | |
5765 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5766 | raw_inode, handle); | |
5767 | } | |
5768 | ||
ac27a0ec DK |
5769 | /* |
5770 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5771 | * dirtiness (it may still be data-dirty). | |
5772 | * This means that the in-core inode may be reaped by prune_icache | |
5773 | * without having to perform any I/O. This is a very good thing, | |
5774 | * because *any* task may call prune_icache - even ones which | |
5775 | * have a transaction open against a different journal. | |
5776 | * | |
5777 | * Is this cheating? Not really. Sure, we haven't written the | |
5778 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5779 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5780 | * we start and wait on commits. | |
5781 | * | |
5782 | * Is this efficient/effective? Well, we're being nice to the system | |
5783 | * by cleaning up our inodes proactively so they can be reaped | |
5784 | * without I/O. But we are potentially leaving up to five seconds' | |
5785 | * worth of inodes floating about which prune_icache wants us to | |
5786 | * write out. One way to fix that would be to get prune_icache() | |
5787 | * to do a write_super() to free up some memory. It has the desired | |
5788 | * effect. | |
5789 | */ | |
617ba13b | 5790 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5791 | { |
617ba13b | 5792 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5793 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5794 | static unsigned int mnt_count; | |
5795 | int err, ret; | |
ac27a0ec DK |
5796 | |
5797 | might_sleep(); | |
617ba13b | 5798 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5799 | if (ext4_handle_valid(handle) && |
5800 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5801 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5802 | /* |
5803 | * We need extra buffer credits since we may write into EA block | |
5804 | * with this same handle. If journal_extend fails, then it will | |
5805 | * only result in a minor loss of functionality for that inode. | |
5806 | * If this is felt to be critical, then e2fsck should be run to | |
5807 | * force a large enough s_min_extra_isize. | |
5808 | */ | |
5809 | if ((jbd2_journal_extend(handle, | |
5810 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5811 | ret = ext4_expand_extra_isize(inode, | |
5812 | sbi->s_want_extra_isize, | |
5813 | iloc, handle); | |
5814 | if (ret) { | |
19f5fb7a TT |
5815 | ext4_set_inode_state(inode, |
5816 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
5817 | if (mnt_count != |
5818 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5819 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5820 | "Unable to expand inode %lu. Delete" |
5821 | " some EAs or run e2fsck.", | |
5822 | inode->i_ino); | |
c1bddad9 AK |
5823 | mnt_count = |
5824 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5825 | } |
5826 | } | |
5827 | } | |
5828 | } | |
ac27a0ec | 5829 | if (!err) |
617ba13b | 5830 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5831 | return err; |
5832 | } | |
5833 | ||
5834 | /* | |
617ba13b | 5835 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5836 | * |
5837 | * We're really interested in the case where a file is being extended. | |
5838 | * i_size has been changed by generic_commit_write() and we thus need | |
5839 | * to include the updated inode in the current transaction. | |
5840 | * | |
5dd4056d | 5841 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5842 | * are allocated to the file. |
5843 | * | |
5844 | * If the inode is marked synchronous, we don't honour that here - doing | |
5845 | * so would cause a commit on atime updates, which we don't bother doing. | |
5846 | * We handle synchronous inodes at the highest possible level. | |
5847 | */ | |
617ba13b | 5848 | void ext4_dirty_inode(struct inode *inode) |
ac27a0ec | 5849 | { |
ac27a0ec DK |
5850 | handle_t *handle; |
5851 | ||
617ba13b | 5852 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
5853 | if (IS_ERR(handle)) |
5854 | goto out; | |
f3dc272f | 5855 | |
f3dc272f CW |
5856 | ext4_mark_inode_dirty(handle, inode); |
5857 | ||
617ba13b | 5858 | ext4_journal_stop(handle); |
ac27a0ec DK |
5859 | out: |
5860 | return; | |
5861 | } | |
5862 | ||
5863 | #if 0 | |
5864 | /* | |
5865 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5866 | * it from being flushed to disk early. Unlike | |
617ba13b | 5867 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5868 | * returns no iloc structure, so the caller needs to repeat the iloc |
5869 | * lookup to mark the inode dirty later. | |
5870 | */ | |
617ba13b | 5871 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5872 | { |
617ba13b | 5873 | struct ext4_iloc iloc; |
ac27a0ec DK |
5874 | |
5875 | int err = 0; | |
5876 | if (handle) { | |
617ba13b | 5877 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5878 | if (!err) { |
5879 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5880 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5881 | if (!err) |
0390131b | 5882 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5883 | NULL, |
0390131b | 5884 | iloc.bh); |
ac27a0ec DK |
5885 | brelse(iloc.bh); |
5886 | } | |
5887 | } | |
617ba13b | 5888 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5889 | return err; |
5890 | } | |
5891 | #endif | |
5892 | ||
617ba13b | 5893 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5894 | { |
5895 | journal_t *journal; | |
5896 | handle_t *handle; | |
5897 | int err; | |
5898 | ||
5899 | /* | |
5900 | * We have to be very careful here: changing a data block's | |
5901 | * journaling status dynamically is dangerous. If we write a | |
5902 | * data block to the journal, change the status and then delete | |
5903 | * that block, we risk forgetting to revoke the old log record | |
5904 | * from the journal and so a subsequent replay can corrupt data. | |
5905 | * So, first we make sure that the journal is empty and that | |
5906 | * nobody is changing anything. | |
5907 | */ | |
5908 | ||
617ba13b | 5909 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5910 | if (!journal) |
5911 | return 0; | |
d699594d | 5912 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5913 | return -EROFS; |
5914 | ||
dab291af MC |
5915 | jbd2_journal_lock_updates(journal); |
5916 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
5917 | |
5918 | /* | |
5919 | * OK, there are no updates running now, and all cached data is | |
5920 | * synced to disk. We are now in a completely consistent state | |
5921 | * which doesn't have anything in the journal, and we know that | |
5922 | * no filesystem updates are running, so it is safe to modify | |
5923 | * the inode's in-core data-journaling state flag now. | |
5924 | */ | |
5925 | ||
5926 | if (val) | |
12e9b892 | 5927 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
ac27a0ec | 5928 | else |
12e9b892 | 5929 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
617ba13b | 5930 | ext4_set_aops(inode); |
ac27a0ec | 5931 | |
dab291af | 5932 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
5933 | |
5934 | /* Finally we can mark the inode as dirty. */ | |
5935 | ||
617ba13b | 5936 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
5937 | if (IS_ERR(handle)) |
5938 | return PTR_ERR(handle); | |
5939 | ||
617ba13b | 5940 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5941 | ext4_handle_sync(handle); |
617ba13b MC |
5942 | ext4_journal_stop(handle); |
5943 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5944 | |
5945 | return err; | |
5946 | } | |
2e9ee850 AK |
5947 | |
5948 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5949 | { | |
5950 | return !buffer_mapped(bh); | |
5951 | } | |
5952 | ||
c2ec175c | 5953 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5954 | { |
c2ec175c | 5955 | struct page *page = vmf->page; |
2e9ee850 AK |
5956 | loff_t size; |
5957 | unsigned long len; | |
5958 | int ret = -EINVAL; | |
79f0be8d | 5959 | void *fsdata; |
2e9ee850 AK |
5960 | struct file *file = vma->vm_file; |
5961 | struct inode *inode = file->f_path.dentry->d_inode; | |
5962 | struct address_space *mapping = inode->i_mapping; | |
5963 | ||
5964 | /* | |
5965 | * Get i_alloc_sem to stop truncates messing with the inode. We cannot | |
5966 | * get i_mutex because we are already holding mmap_sem. | |
5967 | */ | |
5968 | down_read(&inode->i_alloc_sem); | |
5969 | size = i_size_read(inode); | |
5970 | if (page->mapping != mapping || size <= page_offset(page) | |
5971 | || !PageUptodate(page)) { | |
5972 | /* page got truncated from under us? */ | |
5973 | goto out_unlock; | |
5974 | } | |
5975 | ret = 0; | |
5976 | if (PageMappedToDisk(page)) | |
5977 | goto out_unlock; | |
5978 | ||
5979 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5980 | len = size & ~PAGE_CACHE_MASK; | |
5981 | else | |
5982 | len = PAGE_CACHE_SIZE; | |
5983 | ||
a827eaff AK |
5984 | lock_page(page); |
5985 | /* | |
5986 | * return if we have all the buffers mapped. This avoid | |
5987 | * the need to call write_begin/write_end which does a | |
5988 | * journal_start/journal_stop which can block and take | |
5989 | * long time | |
5990 | */ | |
2e9ee850 | 5991 | if (page_has_buffers(page)) { |
2e9ee850 | 5992 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, |
a827eaff AK |
5993 | ext4_bh_unmapped)) { |
5994 | unlock_page(page); | |
2e9ee850 | 5995 | goto out_unlock; |
a827eaff | 5996 | } |
2e9ee850 | 5997 | } |
a827eaff | 5998 | unlock_page(page); |
2e9ee850 AK |
5999 | /* |
6000 | * OK, we need to fill the hole... Do write_begin write_end | |
6001 | * to do block allocation/reservation.We are not holding | |
6002 | * inode.i__mutex here. That allow * parallel write_begin, | |
6003 | * write_end call. lock_page prevent this from happening | |
6004 | * on the same page though | |
6005 | */ | |
6006 | ret = mapping->a_ops->write_begin(file, mapping, page_offset(page), | |
79f0be8d | 6007 | len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); |
2e9ee850 AK |
6008 | if (ret < 0) |
6009 | goto out_unlock; | |
6010 | ret = mapping->a_ops->write_end(file, mapping, page_offset(page), | |
79f0be8d | 6011 | len, len, page, fsdata); |
2e9ee850 AK |
6012 | if (ret < 0) |
6013 | goto out_unlock; | |
6014 | ret = 0; | |
6015 | out_unlock: | |
c2ec175c NP |
6016 | if (ret) |
6017 | ret = VM_FAULT_SIGBUS; | |
2e9ee850 AK |
6018 | up_read(&inode->i_alloc_sem); |
6019 | return ret; | |
6020 | } |