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 | 62 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
cb20d518 TT |
63 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, |
64 | struct buffer_head *bh_result, int create); | |
65 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); | |
66 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
67 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); | |
68 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
64769240 | 69 | |
ac27a0ec DK |
70 | /* |
71 | * Test whether an inode is a fast symlink. | |
72 | */ | |
617ba13b | 73 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 74 | { |
617ba13b | 75 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
76 | (inode->i_sb->s_blocksize >> 9) : 0; |
77 | ||
78 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
79 | } | |
80 | ||
ac27a0ec DK |
81 | /* |
82 | * Work out how many blocks we need to proceed with the next chunk of a | |
83 | * truncate transaction. | |
84 | */ | |
85 | static unsigned long blocks_for_truncate(struct inode *inode) | |
86 | { | |
725d26d3 | 87 | ext4_lblk_t needed; |
ac27a0ec DK |
88 | |
89 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
90 | ||
91 | /* Give ourselves just enough room to cope with inodes in which | |
92 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
93 | * which resulted in random data in an inode which looked enough | |
617ba13b | 94 | * like a regular file for ext4 to try to delete it. Things |
ac27a0ec DK |
95 | * will go a bit crazy if that happens, but at least we should |
96 | * try not to panic the whole kernel. */ | |
97 | if (needed < 2) | |
98 | needed = 2; | |
99 | ||
100 | /* But we need to bound the transaction so we don't overflow the | |
101 | * journal. */ | |
617ba13b MC |
102 | if (needed > EXT4_MAX_TRANS_DATA) |
103 | needed = EXT4_MAX_TRANS_DATA; | |
ac27a0ec | 104 | |
617ba13b | 105 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
ac27a0ec DK |
106 | } |
107 | ||
108 | /* | |
109 | * Truncate transactions can be complex and absolutely huge. So we need to | |
110 | * be able to restart the transaction at a conventient checkpoint to make | |
111 | * sure we don't overflow the journal. | |
112 | * | |
113 | * start_transaction gets us a new handle for a truncate transaction, | |
114 | * and extend_transaction tries to extend the existing one a bit. If | |
115 | * extend fails, we need to propagate the failure up and restart the | |
116 | * transaction in the top-level truncate loop. --sct | |
117 | */ | |
118 | static handle_t *start_transaction(struct inode *inode) | |
119 | { | |
120 | handle_t *result; | |
121 | ||
617ba13b | 122 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
ac27a0ec DK |
123 | if (!IS_ERR(result)) |
124 | return result; | |
125 | ||
617ba13b | 126 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
ac27a0ec DK |
127 | return result; |
128 | } | |
129 | ||
130 | /* | |
131 | * Try to extend this transaction for the purposes of truncation. | |
132 | * | |
133 | * Returns 0 if we managed to create more room. If we can't create more | |
134 | * room, and the transaction must be restarted we return 1. | |
135 | */ | |
136 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
137 | { | |
0390131b FM |
138 | if (!ext4_handle_valid(handle)) |
139 | return 0; | |
140 | if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) | |
ac27a0ec | 141 | return 0; |
617ba13b | 142 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
ac27a0ec DK |
143 | return 0; |
144 | return 1; | |
145 | } | |
146 | ||
147 | /* | |
148 | * Restart the transaction associated with *handle. This does a commit, | |
149 | * so before we call here everything must be consistently dirtied against | |
150 | * this transaction. | |
151 | */ | |
fa5d1113 | 152 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 153 | int nblocks) |
ac27a0ec | 154 | { |
487caeef JK |
155 | int ret; |
156 | ||
157 | /* | |
e35fd660 | 158 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
159 | * moment, get_block can be called only for blocks inside i_size since |
160 | * page cache has been already dropped and writes are blocked by | |
161 | * i_mutex. So we can safely drop the i_data_sem here. | |
162 | */ | |
0390131b | 163 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 164 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef JK |
165 | up_write(&EXT4_I(inode)->i_data_sem); |
166 | ret = ext4_journal_restart(handle, blocks_for_truncate(inode)); | |
167 | down_write(&EXT4_I(inode)->i_data_sem); | |
fa5d1113 | 168 | ext4_discard_preallocations(inode); |
487caeef JK |
169 | |
170 | return ret; | |
ac27a0ec DK |
171 | } |
172 | ||
173 | /* | |
174 | * Called at the last iput() if i_nlink is zero. | |
175 | */ | |
0930fcc1 | 176 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
177 | { |
178 | handle_t *handle; | |
bc965ab3 | 179 | int err; |
ac27a0ec | 180 | |
0930fcc1 AV |
181 | if (inode->i_nlink) { |
182 | truncate_inode_pages(&inode->i_data, 0); | |
183 | goto no_delete; | |
184 | } | |
185 | ||
907f4554 | 186 | if (!is_bad_inode(inode)) |
871a2931 | 187 | dquot_initialize(inode); |
907f4554 | 188 | |
678aaf48 JK |
189 | if (ext4_should_order_data(inode)) |
190 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
191 | truncate_inode_pages(&inode->i_data, 0); |
192 | ||
193 | if (is_bad_inode(inode)) | |
194 | goto no_delete; | |
195 | ||
bc965ab3 | 196 | handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3); |
ac27a0ec | 197 | if (IS_ERR(handle)) { |
bc965ab3 | 198 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
199 | /* |
200 | * If we're going to skip the normal cleanup, we still need to | |
201 | * make sure that the in-core orphan linked list is properly | |
202 | * cleaned up. | |
203 | */ | |
617ba13b | 204 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
205 | goto no_delete; |
206 | } | |
207 | ||
208 | if (IS_SYNC(inode)) | |
0390131b | 209 | ext4_handle_sync(handle); |
ac27a0ec | 210 | inode->i_size = 0; |
bc965ab3 TT |
211 | err = ext4_mark_inode_dirty(handle, inode); |
212 | if (err) { | |
12062ddd | 213 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
214 | "couldn't mark inode dirty (err %d)", err); |
215 | goto stop_handle; | |
216 | } | |
ac27a0ec | 217 | if (inode->i_blocks) |
617ba13b | 218 | ext4_truncate(inode); |
bc965ab3 TT |
219 | |
220 | /* | |
221 | * ext4_ext_truncate() doesn't reserve any slop when it | |
222 | * restarts journal transactions; therefore there may not be | |
223 | * enough credits left in the handle to remove the inode from | |
224 | * the orphan list and set the dtime field. | |
225 | */ | |
0390131b | 226 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
227 | err = ext4_journal_extend(handle, 3); |
228 | if (err > 0) | |
229 | err = ext4_journal_restart(handle, 3); | |
230 | if (err != 0) { | |
12062ddd | 231 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
232 | "couldn't extend journal (err %d)", err); |
233 | stop_handle: | |
234 | ext4_journal_stop(handle); | |
45388219 | 235 | ext4_orphan_del(NULL, inode); |
bc965ab3 TT |
236 | goto no_delete; |
237 | } | |
238 | } | |
239 | ||
ac27a0ec | 240 | /* |
617ba13b | 241 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 242 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 243 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 244 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 245 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
246 | * (Well, we could do this if we need to, but heck - it works) |
247 | */ | |
617ba13b MC |
248 | ext4_orphan_del(handle, inode); |
249 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
250 | |
251 | /* | |
252 | * One subtle ordering requirement: if anything has gone wrong | |
253 | * (transaction abort, IO errors, whatever), then we can still | |
254 | * do these next steps (the fs will already have been marked as | |
255 | * having errors), but we can't free the inode if the mark_dirty | |
256 | * fails. | |
257 | */ | |
617ba13b | 258 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 259 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 260 | ext4_clear_inode(inode); |
ac27a0ec | 261 | else |
617ba13b MC |
262 | ext4_free_inode(handle, inode); |
263 | ext4_journal_stop(handle); | |
ac27a0ec DK |
264 | return; |
265 | no_delete: | |
0930fcc1 | 266 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
267 | } |
268 | ||
269 | typedef struct { | |
270 | __le32 *p; | |
271 | __le32 key; | |
272 | struct buffer_head *bh; | |
273 | } Indirect; | |
274 | ||
275 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
276 | { | |
277 | p->key = *(p->p = v); | |
278 | p->bh = bh; | |
279 | } | |
280 | ||
ac27a0ec | 281 | /** |
617ba13b | 282 | * ext4_block_to_path - parse the block number into array of offsets |
ac27a0ec DK |
283 | * @inode: inode in question (we are only interested in its superblock) |
284 | * @i_block: block number to be parsed | |
285 | * @offsets: array to store the offsets in | |
8c55e204 DK |
286 | * @boundary: set this non-zero if the referred-to block is likely to be |
287 | * followed (on disk) by an indirect block. | |
ac27a0ec | 288 | * |
617ba13b | 289 | * To store the locations of file's data ext4 uses a data structure common |
ac27a0ec DK |
290 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
291 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
292 | * This function translates the block number into path in that tree - | |
293 | * return value is the path length and @offsets[n] is the offset of | |
294 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
295 | * (negative or too large) warning is printed and zero returned. | |
296 | * | |
297 | * Note: function doesn't find node addresses, so no IO is needed. All | |
298 | * we need to know is the capacity of indirect blocks (taken from the | |
299 | * inode->i_sb). | |
300 | */ | |
301 | ||
302 | /* | |
303 | * Portability note: the last comparison (check that we fit into triple | |
304 | * indirect block) is spelled differently, because otherwise on an | |
305 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
306 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
307 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
308 | * i_block would have to be negative in the very beginning, so we would not | |
309 | * get there at all. | |
310 | */ | |
311 | ||
617ba13b | 312 | static int ext4_block_to_path(struct inode *inode, |
de9a55b8 TT |
313 | ext4_lblk_t i_block, |
314 | ext4_lblk_t offsets[4], int *boundary) | |
ac27a0ec | 315 | { |
617ba13b MC |
316 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
317 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
318 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
ac27a0ec DK |
319 | indirect_blocks = ptrs, |
320 | double_blocks = (1 << (ptrs_bits * 2)); | |
321 | int n = 0; | |
322 | int final = 0; | |
323 | ||
c333e073 | 324 | if (i_block < direct_blocks) { |
ac27a0ec DK |
325 | offsets[n++] = i_block; |
326 | final = direct_blocks; | |
af5bc92d | 327 | } else if ((i_block -= direct_blocks) < indirect_blocks) { |
617ba13b | 328 | offsets[n++] = EXT4_IND_BLOCK; |
ac27a0ec DK |
329 | offsets[n++] = i_block; |
330 | final = ptrs; | |
331 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
617ba13b | 332 | offsets[n++] = EXT4_DIND_BLOCK; |
ac27a0ec DK |
333 | offsets[n++] = i_block >> ptrs_bits; |
334 | offsets[n++] = i_block & (ptrs - 1); | |
335 | final = ptrs; | |
336 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
617ba13b | 337 | offsets[n++] = EXT4_TIND_BLOCK; |
ac27a0ec DK |
338 | offsets[n++] = i_block >> (ptrs_bits * 2); |
339 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
340 | offsets[n++] = i_block & (ptrs - 1); | |
341 | final = ptrs; | |
342 | } else { | |
12062ddd | 343 | ext4_warning(inode->i_sb, "block %lu > max in inode %lu", |
de9a55b8 TT |
344 | i_block + direct_blocks + |
345 | indirect_blocks + double_blocks, inode->i_ino); | |
ac27a0ec DK |
346 | } |
347 | if (boundary) | |
348 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
349 | return n; | |
350 | } | |
351 | ||
c398eda0 TT |
352 | static int __ext4_check_blockref(const char *function, unsigned int line, |
353 | struct inode *inode, | |
6fd058f7 TT |
354 | __le32 *p, unsigned int max) |
355 | { | |
1c13d5c0 | 356 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; |
f73953c0 | 357 | __le32 *bref = p; |
6fd058f7 TT |
358 | unsigned int blk; |
359 | ||
fe2c8191 | 360 | while (bref < p+max) { |
6fd058f7 | 361 | blk = le32_to_cpu(*bref++); |
de9a55b8 TT |
362 | if (blk && |
363 | unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb), | |
6fd058f7 | 364 | blk, 1))) { |
1c13d5c0 | 365 | es->s_last_error_block = cpu_to_le64(blk); |
c398eda0 TT |
366 | ext4_error_inode(inode, function, line, blk, |
367 | "invalid block"); | |
de9a55b8 TT |
368 | return -EIO; |
369 | } | |
370 | } | |
371 | return 0; | |
fe2c8191 TN |
372 | } |
373 | ||
374 | ||
375 | #define ext4_check_indirect_blockref(inode, bh) \ | |
c398eda0 TT |
376 | __ext4_check_blockref(__func__, __LINE__, inode, \ |
377 | (__le32 *)(bh)->b_data, \ | |
fe2c8191 TN |
378 | EXT4_ADDR_PER_BLOCK((inode)->i_sb)) |
379 | ||
380 | #define ext4_check_inode_blockref(inode) \ | |
c398eda0 TT |
381 | __ext4_check_blockref(__func__, __LINE__, inode, \ |
382 | EXT4_I(inode)->i_data, \ | |
fe2c8191 TN |
383 | EXT4_NDIR_BLOCKS) |
384 | ||
ac27a0ec | 385 | /** |
617ba13b | 386 | * ext4_get_branch - read the chain of indirect blocks leading to data |
ac27a0ec DK |
387 | * @inode: inode in question |
388 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
389 | * @offsets: offsets of pointers in inode/indirect blocks | |
390 | * @chain: place to store the result | |
391 | * @err: here we store the error value | |
392 | * | |
393 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
394 | * if everything went OK or the pointer to the last filled triple | |
395 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
396 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
397 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
398 | * number (it points into struct inode for i==0 and into the bh->b_data | |
399 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
400 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
401 | * numbers of the chain, addresses they were taken from (and where we can | |
402 | * verify that chain did not change) and buffer_heads hosting these | |
403 | * numbers. | |
404 | * | |
405 | * Function stops when it stumbles upon zero pointer (absent block) | |
406 | * (pointer to last triple returned, *@err == 0) | |
407 | * or when it gets an IO error reading an indirect block | |
408 | * (ditto, *@err == -EIO) | |
ac27a0ec DK |
409 | * or when it reads all @depth-1 indirect blocks successfully and finds |
410 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
c278bfec AK |
411 | * |
412 | * Need to be called with | |
0e855ac8 | 413 | * down_read(&EXT4_I(inode)->i_data_sem) |
ac27a0ec | 414 | */ |
725d26d3 AK |
415 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
416 | ext4_lblk_t *offsets, | |
ac27a0ec DK |
417 | Indirect chain[4], int *err) |
418 | { | |
419 | struct super_block *sb = inode->i_sb; | |
420 | Indirect *p = chain; | |
421 | struct buffer_head *bh; | |
422 | ||
423 | *err = 0; | |
424 | /* i_data is not going away, no lock needed */ | |
af5bc92d | 425 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); |
ac27a0ec DK |
426 | if (!p->key) |
427 | goto no_block; | |
428 | while (--depth) { | |
fe2c8191 TN |
429 | bh = sb_getblk(sb, le32_to_cpu(p->key)); |
430 | if (unlikely(!bh)) | |
ac27a0ec | 431 | goto failure; |
de9a55b8 | 432 | |
fe2c8191 TN |
433 | if (!bh_uptodate_or_lock(bh)) { |
434 | if (bh_submit_read(bh) < 0) { | |
435 | put_bh(bh); | |
436 | goto failure; | |
437 | } | |
438 | /* validate block references */ | |
439 | if (ext4_check_indirect_blockref(inode, bh)) { | |
440 | put_bh(bh); | |
441 | goto failure; | |
442 | } | |
443 | } | |
de9a55b8 | 444 | |
af5bc92d | 445 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); |
ac27a0ec DK |
446 | /* Reader: end */ |
447 | if (!p->key) | |
448 | goto no_block; | |
449 | } | |
450 | return NULL; | |
451 | ||
ac27a0ec DK |
452 | failure: |
453 | *err = -EIO; | |
454 | no_block: | |
455 | return p; | |
456 | } | |
457 | ||
458 | /** | |
617ba13b | 459 | * ext4_find_near - find a place for allocation with sufficient locality |
ac27a0ec DK |
460 | * @inode: owner |
461 | * @ind: descriptor of indirect block. | |
462 | * | |
1cc8dcf5 | 463 | * This function returns the preferred place for block allocation. |
ac27a0ec DK |
464 | * It is used when heuristic for sequential allocation fails. |
465 | * Rules are: | |
466 | * + if there is a block to the left of our position - allocate near it. | |
467 | * + if pointer will live in indirect block - allocate near that block. | |
468 | * + if pointer will live in inode - allocate in the same | |
469 | * cylinder group. | |
470 | * | |
471 | * In the latter case we colour the starting block by the callers PID to | |
472 | * prevent it from clashing with concurrent allocations for a different inode | |
473 | * in the same block group. The PID is used here so that functionally related | |
474 | * files will be close-by on-disk. | |
475 | * | |
476 | * Caller must make sure that @ind is valid and will stay that way. | |
477 | */ | |
617ba13b | 478 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
ac27a0ec | 479 | { |
617ba13b | 480 | struct ext4_inode_info *ei = EXT4_I(inode); |
af5bc92d | 481 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; |
ac27a0ec | 482 | __le32 *p; |
617ba13b | 483 | ext4_fsblk_t bg_start; |
74d3487f | 484 | ext4_fsblk_t last_block; |
617ba13b | 485 | ext4_grpblk_t colour; |
a4912123 TT |
486 | ext4_group_t block_group; |
487 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); | |
ac27a0ec DK |
488 | |
489 | /* Try to find previous block */ | |
490 | for (p = ind->p - 1; p >= start; p--) { | |
491 | if (*p) | |
492 | return le32_to_cpu(*p); | |
493 | } | |
494 | ||
495 | /* No such thing, so let's try location of indirect block */ | |
496 | if (ind->bh) | |
497 | return ind->bh->b_blocknr; | |
498 | ||
499 | /* | |
500 | * It is going to be referred to from the inode itself? OK, just put it | |
501 | * into the same cylinder group then. | |
502 | */ | |
a4912123 TT |
503 | block_group = ei->i_block_group; |
504 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { | |
505 | block_group &= ~(flex_size-1); | |
506 | if (S_ISREG(inode->i_mode)) | |
507 | block_group++; | |
508 | } | |
509 | bg_start = ext4_group_first_block_no(inode->i_sb, block_group); | |
74d3487f VC |
510 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
511 | ||
a4912123 TT |
512 | /* |
513 | * If we are doing delayed allocation, we don't need take | |
514 | * colour into account. | |
515 | */ | |
516 | if (test_opt(inode->i_sb, DELALLOC)) | |
517 | return bg_start; | |
518 | ||
74d3487f VC |
519 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
520 | colour = (current->pid % 16) * | |
617ba13b | 521 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
522 | else |
523 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
ac27a0ec DK |
524 | return bg_start + colour; |
525 | } | |
526 | ||
527 | /** | |
1cc8dcf5 | 528 | * ext4_find_goal - find a preferred place for allocation. |
ac27a0ec DK |
529 | * @inode: owner |
530 | * @block: block we want | |
ac27a0ec | 531 | * @partial: pointer to the last triple within a chain |
ac27a0ec | 532 | * |
1cc8dcf5 | 533 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 534 | * returns it. |
fb0a387d ES |
535 | * Because this is only used for non-extent files, we limit the block nr |
536 | * to 32 bits. | |
ac27a0ec | 537 | */ |
725d26d3 | 538 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
de9a55b8 | 539 | Indirect *partial) |
ac27a0ec | 540 | { |
fb0a387d ES |
541 | ext4_fsblk_t goal; |
542 | ||
ac27a0ec | 543 | /* |
c2ea3fde | 544 | * XXX need to get goal block from mballoc's data structures |
ac27a0ec | 545 | */ |
ac27a0ec | 546 | |
fb0a387d ES |
547 | goal = ext4_find_near(inode, partial); |
548 | goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; | |
549 | return goal; | |
ac27a0ec DK |
550 | } |
551 | ||
552 | /** | |
617ba13b | 553 | * ext4_blks_to_allocate: Look up the block map and count the number |
ac27a0ec DK |
554 | * of direct blocks need to be allocated for the given branch. |
555 | * | |
556 | * @branch: chain of indirect blocks | |
557 | * @k: number of blocks need for indirect blocks | |
558 | * @blks: number of data blocks to be mapped. | |
559 | * @blocks_to_boundary: the offset in the indirect block | |
560 | * | |
561 | * return the total number of blocks to be allocate, including the | |
562 | * direct and indirect blocks. | |
563 | */ | |
498e5f24 | 564 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, |
de9a55b8 | 565 | int blocks_to_boundary) |
ac27a0ec | 566 | { |
498e5f24 | 567 | unsigned int count = 0; |
ac27a0ec DK |
568 | |
569 | /* | |
570 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
571 | * then it's clear blocks on that path have not allocated | |
572 | */ | |
573 | if (k > 0) { | |
574 | /* right now we don't handle cross boundary allocation */ | |
575 | if (blks < blocks_to_boundary + 1) | |
576 | count += blks; | |
577 | else | |
578 | count += blocks_to_boundary + 1; | |
579 | return count; | |
580 | } | |
581 | ||
582 | count++; | |
583 | while (count < blks && count <= blocks_to_boundary && | |
584 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
585 | count++; | |
586 | } | |
587 | return count; | |
588 | } | |
589 | ||
590 | /** | |
617ba13b | 591 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
ac27a0ec DK |
592 | * @indirect_blks: the number of blocks need to allocate for indirect |
593 | * blocks | |
594 | * | |
595 | * @new_blocks: on return it will store the new block numbers for | |
596 | * the indirect blocks(if needed) and the first direct block, | |
597 | * @blks: on return it will store the total number of allocated | |
598 | * direct blocks | |
599 | */ | |
617ba13b | 600 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
601 | ext4_lblk_t iblock, ext4_fsblk_t goal, |
602 | int indirect_blks, int blks, | |
603 | ext4_fsblk_t new_blocks[4], int *err) | |
ac27a0ec | 604 | { |
815a1130 | 605 | struct ext4_allocation_request ar; |
ac27a0ec | 606 | int target, i; |
7061eba7 | 607 | unsigned long count = 0, blk_allocated = 0; |
ac27a0ec | 608 | int index = 0; |
617ba13b | 609 | ext4_fsblk_t current_block = 0; |
ac27a0ec DK |
610 | int ret = 0; |
611 | ||
612 | /* | |
613 | * Here we try to allocate the requested multiple blocks at once, | |
614 | * on a best-effort basis. | |
615 | * To build a branch, we should allocate blocks for | |
616 | * the indirect blocks(if not allocated yet), and at least | |
617 | * the first direct block of this branch. That's the | |
618 | * minimum number of blocks need to allocate(required) | |
619 | */ | |
7061eba7 AK |
620 | /* first we try to allocate the indirect blocks */ |
621 | target = indirect_blks; | |
622 | while (target > 0) { | |
ac27a0ec DK |
623 | count = target; |
624 | /* allocating blocks for indirect blocks and direct blocks */ | |
7061eba7 AK |
625 | current_block = ext4_new_meta_blocks(handle, inode, |
626 | goal, &count, err); | |
ac27a0ec DK |
627 | if (*err) |
628 | goto failed_out; | |
629 | ||
273df556 FM |
630 | if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) { |
631 | EXT4_ERROR_INODE(inode, | |
632 | "current_block %llu + count %lu > %d!", | |
633 | current_block, count, | |
634 | EXT4_MAX_BLOCK_FILE_PHYS); | |
635 | *err = -EIO; | |
636 | goto failed_out; | |
637 | } | |
fb0a387d | 638 | |
ac27a0ec DK |
639 | target -= count; |
640 | /* allocate blocks for indirect blocks */ | |
641 | while (index < indirect_blks && count) { | |
642 | new_blocks[index++] = current_block++; | |
643 | count--; | |
644 | } | |
7061eba7 AK |
645 | if (count > 0) { |
646 | /* | |
647 | * save the new block number | |
648 | * for the first direct block | |
649 | */ | |
650 | new_blocks[index] = current_block; | |
651 | printk(KERN_INFO "%s returned more blocks than " | |
652 | "requested\n", __func__); | |
653 | WARN_ON(1); | |
ac27a0ec | 654 | break; |
7061eba7 | 655 | } |
ac27a0ec DK |
656 | } |
657 | ||
7061eba7 AK |
658 | target = blks - count ; |
659 | blk_allocated = count; | |
660 | if (!target) | |
661 | goto allocated; | |
662 | /* Now allocate data blocks */ | |
815a1130 TT |
663 | memset(&ar, 0, sizeof(ar)); |
664 | ar.inode = inode; | |
665 | ar.goal = goal; | |
666 | ar.len = target; | |
667 | ar.logical = iblock; | |
668 | if (S_ISREG(inode->i_mode)) | |
669 | /* enable in-core preallocation only for regular files */ | |
670 | ar.flags = EXT4_MB_HINT_DATA; | |
671 | ||
672 | current_block = ext4_mb_new_blocks(handle, &ar, err); | |
273df556 FM |
673 | if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) { |
674 | EXT4_ERROR_INODE(inode, | |
675 | "current_block %llu + ar.len %d > %d!", | |
676 | current_block, ar.len, | |
677 | EXT4_MAX_BLOCK_FILE_PHYS); | |
678 | *err = -EIO; | |
679 | goto failed_out; | |
680 | } | |
815a1130 | 681 | |
7061eba7 AK |
682 | if (*err && (target == blks)) { |
683 | /* | |
684 | * if the allocation failed and we didn't allocate | |
685 | * any blocks before | |
686 | */ | |
687 | goto failed_out; | |
688 | } | |
689 | if (!*err) { | |
690 | if (target == blks) { | |
de9a55b8 TT |
691 | /* |
692 | * save the new block number | |
693 | * for the first direct block | |
694 | */ | |
7061eba7 AK |
695 | new_blocks[index] = current_block; |
696 | } | |
815a1130 | 697 | blk_allocated += ar.len; |
7061eba7 AK |
698 | } |
699 | allocated: | |
ac27a0ec | 700 | /* total number of blocks allocated for direct blocks */ |
7061eba7 | 701 | ret = blk_allocated; |
ac27a0ec DK |
702 | *err = 0; |
703 | return ret; | |
704 | failed_out: | |
af5bc92d | 705 | for (i = 0; i < index; i++) |
e6362609 | 706 | ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0); |
ac27a0ec DK |
707 | return ret; |
708 | } | |
709 | ||
710 | /** | |
617ba13b | 711 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
ac27a0ec DK |
712 | * @inode: owner |
713 | * @indirect_blks: number of allocated indirect blocks | |
714 | * @blks: number of allocated direct blocks | |
715 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
716 | * @branch: place to store the chain in. | |
717 | * | |
718 | * This function allocates blocks, zeroes out all but the last one, | |
719 | * links them into chain and (if we are synchronous) writes them to disk. | |
720 | * In other words, it prepares a branch that can be spliced onto the | |
721 | * inode. It stores the information about that chain in the branch[], in | |
617ba13b | 722 | * the same format as ext4_get_branch() would do. We are calling it after |
ac27a0ec DK |
723 | * we had read the existing part of chain and partial points to the last |
724 | * triple of that (one with zero ->key). Upon the exit we have the same | |
617ba13b | 725 | * picture as after the successful ext4_get_block(), except that in one |
ac27a0ec DK |
726 | * place chain is disconnected - *branch->p is still zero (we did not |
727 | * set the last link), but branch->key contains the number that should | |
728 | * be placed into *branch->p to fill that gap. | |
729 | * | |
730 | * If allocation fails we free all blocks we've allocated (and forget | |
731 | * their buffer_heads) and return the error value the from failed | |
617ba13b | 732 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
ac27a0ec DK |
733 | * as described above and return 0. |
734 | */ | |
617ba13b | 735 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
736 | ext4_lblk_t iblock, int indirect_blks, |
737 | int *blks, ext4_fsblk_t goal, | |
738 | ext4_lblk_t *offsets, Indirect *branch) | |
ac27a0ec DK |
739 | { |
740 | int blocksize = inode->i_sb->s_blocksize; | |
741 | int i, n = 0; | |
742 | int err = 0; | |
743 | struct buffer_head *bh; | |
744 | int num; | |
617ba13b MC |
745 | ext4_fsblk_t new_blocks[4]; |
746 | ext4_fsblk_t current_block; | |
ac27a0ec | 747 | |
7061eba7 | 748 | num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks, |
ac27a0ec DK |
749 | *blks, new_blocks, &err); |
750 | if (err) | |
751 | return err; | |
752 | ||
753 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
754 | /* | |
755 | * metadata blocks and data blocks are allocated. | |
756 | */ | |
757 | for (n = 1; n <= indirect_blks; n++) { | |
758 | /* | |
759 | * Get buffer_head for parent block, zero it out | |
760 | * and set the pointer to new one, then send | |
761 | * parent to disk. | |
762 | */ | |
763 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
764 | branch[n].bh = bh; | |
765 | lock_buffer(bh); | |
766 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 767 | err = ext4_journal_get_create_access(handle, bh); |
ac27a0ec | 768 | if (err) { |
6487a9d3 CW |
769 | /* Don't brelse(bh) here; it's done in |
770 | * ext4_journal_forget() below */ | |
ac27a0ec | 771 | unlock_buffer(bh); |
ac27a0ec DK |
772 | goto failed; |
773 | } | |
774 | ||
775 | memset(bh->b_data, 0, blocksize); | |
776 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
777 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
778 | *branch[n].p = branch[n].key; | |
af5bc92d | 779 | if (n == indirect_blks) { |
ac27a0ec DK |
780 | current_block = new_blocks[n]; |
781 | /* | |
782 | * End of chain, update the last new metablock of | |
783 | * the chain to point to the new allocated | |
784 | * data blocks numbers | |
785 | */ | |
de9a55b8 | 786 | for (i = 1; i < num; i++) |
ac27a0ec DK |
787 | *(branch[n].p + i) = cpu_to_le32(++current_block); |
788 | } | |
789 | BUFFER_TRACE(bh, "marking uptodate"); | |
790 | set_buffer_uptodate(bh); | |
791 | unlock_buffer(bh); | |
792 | ||
0390131b FM |
793 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
794 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
795 | if (err) |
796 | goto failed; | |
797 | } | |
798 | *blks = num; | |
799 | return err; | |
800 | failed: | |
801 | /* Allocation failed, free what we already allocated */ | |
e6362609 | 802 | ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0); |
ac27a0ec | 803 | for (i = 1; i <= n ; i++) { |
60e6679e | 804 | /* |
e6362609 TT |
805 | * branch[i].bh is newly allocated, so there is no |
806 | * need to revoke the block, which is why we don't | |
807 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
b7e57e7c | 808 | */ |
e6362609 TT |
809 | ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, |
810 | EXT4_FREE_BLOCKS_FORGET); | |
ac27a0ec | 811 | } |
e6362609 TT |
812 | for (i = n+1; i < indirect_blks; i++) |
813 | ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0); | |
ac27a0ec | 814 | |
e6362609 | 815 | ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0); |
ac27a0ec DK |
816 | |
817 | return err; | |
818 | } | |
819 | ||
820 | /** | |
617ba13b | 821 | * ext4_splice_branch - splice the allocated branch onto inode. |
ac27a0ec DK |
822 | * @inode: owner |
823 | * @block: (logical) number of block we are adding | |
824 | * @chain: chain of indirect blocks (with a missing link - see | |
617ba13b | 825 | * ext4_alloc_branch) |
ac27a0ec DK |
826 | * @where: location of missing link |
827 | * @num: number of indirect blocks we are adding | |
828 | * @blks: number of direct blocks we are adding | |
829 | * | |
830 | * This function fills the missing link and does all housekeeping needed in | |
831 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
832 | * chain to new block and return 0. | |
833 | */ | |
617ba13b | 834 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
835 | ext4_lblk_t block, Indirect *where, int num, |
836 | int blks) | |
ac27a0ec DK |
837 | { |
838 | int i; | |
839 | int err = 0; | |
617ba13b | 840 | ext4_fsblk_t current_block; |
ac27a0ec | 841 | |
ac27a0ec DK |
842 | /* |
843 | * If we're splicing into a [td]indirect block (as opposed to the | |
844 | * inode) then we need to get write access to the [td]indirect block | |
845 | * before the splice. | |
846 | */ | |
847 | if (where->bh) { | |
848 | BUFFER_TRACE(where->bh, "get_write_access"); | |
617ba13b | 849 | err = ext4_journal_get_write_access(handle, where->bh); |
ac27a0ec DK |
850 | if (err) |
851 | goto err_out; | |
852 | } | |
853 | /* That's it */ | |
854 | ||
855 | *where->p = where->key; | |
856 | ||
857 | /* | |
858 | * Update the host buffer_head or inode to point to more just allocated | |
859 | * direct blocks blocks | |
860 | */ | |
861 | if (num == 0 && blks > 1) { | |
862 | current_block = le32_to_cpu(where->key) + 1; | |
863 | for (i = 1; i < blks; i++) | |
af5bc92d | 864 | *(where->p + i) = cpu_to_le32(current_block++); |
ac27a0ec DK |
865 | } |
866 | ||
ac27a0ec | 867 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
ac27a0ec DK |
868 | /* had we spliced it onto indirect block? */ |
869 | if (where->bh) { | |
870 | /* | |
871 | * If we spliced it onto an indirect block, we haven't | |
872 | * altered the inode. Note however that if it is being spliced | |
873 | * onto an indirect block at the very end of the file (the | |
874 | * file is growing) then we *will* alter the inode to reflect | |
875 | * the new i_size. But that is not done here - it is done in | |
617ba13b | 876 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
ac27a0ec DK |
877 | */ |
878 | jbd_debug(5, "splicing indirect only\n"); | |
0390131b FM |
879 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); |
880 | err = ext4_handle_dirty_metadata(handle, inode, where->bh); | |
ac27a0ec DK |
881 | if (err) |
882 | goto err_out; | |
883 | } else { | |
884 | /* | |
885 | * OK, we spliced it into the inode itself on a direct block. | |
ac27a0ec | 886 | */ |
41591750 | 887 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
888 | jbd_debug(5, "splicing direct\n"); |
889 | } | |
890 | return err; | |
891 | ||
892 | err_out: | |
893 | for (i = 1; i <= num; i++) { | |
60e6679e | 894 | /* |
e6362609 TT |
895 | * branch[i].bh is newly allocated, so there is no |
896 | * need to revoke the block, which is why we don't | |
897 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
b7e57e7c | 898 | */ |
e6362609 TT |
899 | ext4_free_blocks(handle, inode, where[i].bh, 0, 1, |
900 | EXT4_FREE_BLOCKS_FORGET); | |
ac27a0ec | 901 | } |
e6362609 TT |
902 | ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key), |
903 | blks, 0); | |
ac27a0ec DK |
904 | |
905 | return err; | |
906 | } | |
907 | ||
908 | /* | |
e35fd660 | 909 | * The ext4_ind_map_blocks() function handles non-extents inodes |
b920c755 | 910 | * (i.e., using the traditional indirect/double-indirect i_blocks |
e35fd660 | 911 | * scheme) for ext4_map_blocks(). |
b920c755 | 912 | * |
ac27a0ec DK |
913 | * Allocation strategy is simple: if we have to allocate something, we will |
914 | * have to go the whole way to leaf. So let's do it before attaching anything | |
915 | * to tree, set linkage between the newborn blocks, write them if sync is | |
916 | * required, recheck the path, free and repeat if check fails, otherwise | |
917 | * set the last missing link (that will protect us from any truncate-generated | |
918 | * removals - all blocks on the path are immune now) and possibly force the | |
919 | * write on the parent block. | |
920 | * That has a nice additional property: no special recovery from the failed | |
921 | * allocations is needed - we simply release blocks and do not touch anything | |
922 | * reachable from inode. | |
923 | * | |
924 | * `handle' can be NULL if create == 0. | |
925 | * | |
ac27a0ec DK |
926 | * return > 0, # of blocks mapped or allocated. |
927 | * return = 0, if plain lookup failed. | |
928 | * return < 0, error case. | |
c278bfec | 929 | * |
b920c755 TT |
930 | * The ext4_ind_get_blocks() function should be called with |
931 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem | |
932 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or | |
933 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system | |
934 | * blocks. | |
ac27a0ec | 935 | */ |
e35fd660 TT |
936 | static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode, |
937 | struct ext4_map_blocks *map, | |
de9a55b8 | 938 | int flags) |
ac27a0ec DK |
939 | { |
940 | int err = -EIO; | |
725d26d3 | 941 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
942 | Indirect chain[4]; |
943 | Indirect *partial; | |
617ba13b | 944 | ext4_fsblk_t goal; |
ac27a0ec DK |
945 | int indirect_blks; |
946 | int blocks_to_boundary = 0; | |
947 | int depth; | |
ac27a0ec | 948 | int count = 0; |
617ba13b | 949 | ext4_fsblk_t first_block = 0; |
ac27a0ec | 950 | |
12e9b892 | 951 | J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); |
c2177057 | 952 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); |
e35fd660 | 953 | depth = ext4_block_to_path(inode, map->m_lblk, offsets, |
de9a55b8 | 954 | &blocks_to_boundary); |
ac27a0ec DK |
955 | |
956 | if (depth == 0) | |
957 | goto out; | |
958 | ||
617ba13b | 959 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
960 | |
961 | /* Simplest case - block found, no allocation needed */ | |
962 | if (!partial) { | |
963 | first_block = le32_to_cpu(chain[depth - 1].key); | |
ac27a0ec DK |
964 | count++; |
965 | /*map more blocks*/ | |
e35fd660 | 966 | while (count < map->m_len && count <= blocks_to_boundary) { |
617ba13b | 967 | ext4_fsblk_t blk; |
ac27a0ec | 968 | |
ac27a0ec DK |
969 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
970 | ||
971 | if (blk == first_block + count) | |
972 | count++; | |
973 | else | |
974 | break; | |
975 | } | |
c278bfec | 976 | goto got_it; |
ac27a0ec DK |
977 | } |
978 | ||
979 | /* Next simple case - plain lookup or failed read of indirect block */ | |
c2177057 | 980 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO) |
ac27a0ec DK |
981 | goto cleanup; |
982 | ||
ac27a0ec | 983 | /* |
c2ea3fde | 984 | * Okay, we need to do block allocation. |
ac27a0ec | 985 | */ |
e35fd660 | 986 | goal = ext4_find_goal(inode, map->m_lblk, partial); |
ac27a0ec DK |
987 | |
988 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
989 | indirect_blks = (chain + depth) - partial - 1; | |
990 | ||
991 | /* | |
992 | * Next look up the indirect map to count the totoal number of | |
993 | * direct blocks to allocate for this branch. | |
994 | */ | |
617ba13b | 995 | count = ext4_blks_to_allocate(partial, indirect_blks, |
e35fd660 | 996 | map->m_len, blocks_to_boundary); |
ac27a0ec | 997 | /* |
617ba13b | 998 | * Block out ext4_truncate while we alter the tree |
ac27a0ec | 999 | */ |
e35fd660 | 1000 | err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks, |
de9a55b8 TT |
1001 | &count, goal, |
1002 | offsets + (partial - chain), partial); | |
ac27a0ec DK |
1003 | |
1004 | /* | |
617ba13b | 1005 | * The ext4_splice_branch call will free and forget any buffers |
ac27a0ec DK |
1006 | * on the new chain if there is a failure, but that risks using |
1007 | * up transaction credits, especially for bitmaps where the | |
1008 | * credits cannot be returned. Can we handle this somehow? We | |
1009 | * may need to return -EAGAIN upwards in the worst case. --sct | |
1010 | */ | |
1011 | if (!err) | |
e35fd660 | 1012 | err = ext4_splice_branch(handle, inode, map->m_lblk, |
de9a55b8 | 1013 | partial, indirect_blks, count); |
2bba702d | 1014 | if (err) |
ac27a0ec DK |
1015 | goto cleanup; |
1016 | ||
e35fd660 | 1017 | map->m_flags |= EXT4_MAP_NEW; |
b436b9be JK |
1018 | |
1019 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
ac27a0ec | 1020 | got_it: |
e35fd660 TT |
1021 | map->m_flags |= EXT4_MAP_MAPPED; |
1022 | map->m_pblk = le32_to_cpu(chain[depth-1].key); | |
1023 | map->m_len = count; | |
ac27a0ec | 1024 | if (count > blocks_to_boundary) |
e35fd660 | 1025 | map->m_flags |= EXT4_MAP_BOUNDARY; |
ac27a0ec DK |
1026 | err = count; |
1027 | /* Clean up and exit */ | |
1028 | partial = chain + depth - 1; /* the whole chain */ | |
1029 | cleanup: | |
1030 | while (partial > chain) { | |
1031 | BUFFER_TRACE(partial->bh, "call brelse"); | |
1032 | brelse(partial->bh); | |
1033 | partial--; | |
1034 | } | |
ac27a0ec DK |
1035 | out: |
1036 | return err; | |
1037 | } | |
1038 | ||
a9e7f447 DM |
1039 | #ifdef CONFIG_QUOTA |
1040 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 1041 | { |
a9e7f447 | 1042 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 1043 | } |
a9e7f447 | 1044 | #endif |
9d0be502 | 1045 | |
12219aea AK |
1046 | /* |
1047 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 1048 | * to allocate a new block at @lblocks for non extent file based file |
12219aea | 1049 | */ |
9d0be502 TT |
1050 | static int ext4_indirect_calc_metadata_amount(struct inode *inode, |
1051 | sector_t lblock) | |
12219aea | 1052 | { |
9d0be502 | 1053 | struct ext4_inode_info *ei = EXT4_I(inode); |
d330a5be | 1054 | sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1); |
9d0be502 | 1055 | int blk_bits; |
12219aea | 1056 | |
9d0be502 TT |
1057 | if (lblock < EXT4_NDIR_BLOCKS) |
1058 | return 0; | |
12219aea | 1059 | |
9d0be502 | 1060 | lblock -= EXT4_NDIR_BLOCKS; |
12219aea | 1061 | |
9d0be502 TT |
1062 | if (ei->i_da_metadata_calc_len && |
1063 | (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) { | |
1064 | ei->i_da_metadata_calc_len++; | |
1065 | return 0; | |
1066 | } | |
1067 | ei->i_da_metadata_calc_last_lblock = lblock & dind_mask; | |
1068 | ei->i_da_metadata_calc_len = 1; | |
d330a5be | 1069 | blk_bits = order_base_2(lblock); |
9d0be502 | 1070 | return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1; |
12219aea AK |
1071 | } |
1072 | ||
1073 | /* | |
1074 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 1075 | * to allocate a block located at @lblock |
12219aea | 1076 | */ |
9d0be502 | 1077 | static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock) |
12219aea | 1078 | { |
12e9b892 | 1079 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 1080 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 1081 | |
9d0be502 | 1082 | return ext4_indirect_calc_metadata_amount(inode, lblock); |
12219aea AK |
1083 | } |
1084 | ||
0637c6f4 TT |
1085 | /* |
1086 | * Called with i_data_sem down, which is important since we can call | |
1087 | * ext4_discard_preallocations() from here. | |
1088 | */ | |
5f634d06 AK |
1089 | void ext4_da_update_reserve_space(struct inode *inode, |
1090 | int used, int quota_claim) | |
12219aea AK |
1091 | { |
1092 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1093 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
1094 | |
1095 | spin_lock(&ei->i_block_reservation_lock); | |
f8ec9d68 | 1096 | trace_ext4_da_update_reserve_space(inode, used); |
0637c6f4 TT |
1097 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
1098 | ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d " | |
1099 | "with only %d reserved data blocks\n", | |
1100 | __func__, inode->i_ino, used, | |
1101 | ei->i_reserved_data_blocks); | |
1102 | WARN_ON(1); | |
1103 | used = ei->i_reserved_data_blocks; | |
1104 | } | |
12219aea | 1105 | |
0637c6f4 TT |
1106 | /* Update per-inode reservations */ |
1107 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 1108 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
72b8ab9d ES |
1109 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, |
1110 | used + ei->i_allocated_meta_blocks); | |
0637c6f4 | 1111 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 1112 | |
0637c6f4 TT |
1113 | if (ei->i_reserved_data_blocks == 0) { |
1114 | /* | |
1115 | * We can release all of the reserved metadata blocks | |
1116 | * only when we have written all of the delayed | |
1117 | * allocation blocks. | |
1118 | */ | |
72b8ab9d ES |
1119 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, |
1120 | ei->i_reserved_meta_blocks); | |
ee5f4d9c | 1121 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1122 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 1123 | } |
12219aea | 1124 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1125 | |
72b8ab9d ES |
1126 | /* Update quota subsystem for data blocks */ |
1127 | if (quota_claim) | |
5dd4056d | 1128 | dquot_claim_block(inode, used); |
72b8ab9d | 1129 | else { |
5f634d06 AK |
1130 | /* |
1131 | * We did fallocate with an offset that is already delayed | |
1132 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 1133 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 1134 | */ |
72b8ab9d | 1135 | dquot_release_reservation_block(inode, used); |
5f634d06 | 1136 | } |
d6014301 AK |
1137 | |
1138 | /* | |
1139 | * If we have done all the pending block allocations and if | |
1140 | * there aren't any writers on the inode, we can discard the | |
1141 | * inode's preallocations. | |
1142 | */ | |
0637c6f4 TT |
1143 | if ((ei->i_reserved_data_blocks == 0) && |
1144 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 1145 | ext4_discard_preallocations(inode); |
12219aea AK |
1146 | } |
1147 | ||
e29136f8 | 1148 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
1149 | unsigned int line, |
1150 | struct ext4_map_blocks *map) | |
6fd058f7 | 1151 | { |
24676da4 TT |
1152 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
1153 | map->m_len)) { | |
c398eda0 TT |
1154 | ext4_error_inode(inode, func, line, map->m_pblk, |
1155 | "lblock %lu mapped to illegal pblock " | |
1156 | "(length %d)", (unsigned long) map->m_lblk, | |
1157 | map->m_len); | |
6fd058f7 TT |
1158 | return -EIO; |
1159 | } | |
1160 | return 0; | |
1161 | } | |
1162 | ||
e29136f8 | 1163 | #define check_block_validity(inode, map) \ |
c398eda0 | 1164 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 1165 | |
55138e0b | 1166 | /* |
1f94533d TT |
1167 | * Return the number of contiguous dirty pages in a given inode |
1168 | * starting at page frame idx. | |
55138e0b TT |
1169 | */ |
1170 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
1171 | unsigned int max_pages) | |
1172 | { | |
1173 | struct address_space *mapping = inode->i_mapping; | |
1174 | pgoff_t index; | |
1175 | struct pagevec pvec; | |
1176 | pgoff_t num = 0; | |
1177 | int i, nr_pages, done = 0; | |
1178 | ||
1179 | if (max_pages == 0) | |
1180 | return 0; | |
1181 | pagevec_init(&pvec, 0); | |
1182 | while (!done) { | |
1183 | index = idx; | |
1184 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
1185 | PAGECACHE_TAG_DIRTY, | |
1186 | (pgoff_t)PAGEVEC_SIZE); | |
1187 | if (nr_pages == 0) | |
1188 | break; | |
1189 | for (i = 0; i < nr_pages; i++) { | |
1190 | struct page *page = pvec.pages[i]; | |
1191 | struct buffer_head *bh, *head; | |
1192 | ||
1193 | lock_page(page); | |
1194 | if (unlikely(page->mapping != mapping) || | |
1195 | !PageDirty(page) || | |
1196 | PageWriteback(page) || | |
1197 | page->index != idx) { | |
1198 | done = 1; | |
1199 | unlock_page(page); | |
1200 | break; | |
1201 | } | |
1f94533d TT |
1202 | if (page_has_buffers(page)) { |
1203 | bh = head = page_buffers(page); | |
1204 | do { | |
1205 | if (!buffer_delay(bh) && | |
1206 | !buffer_unwritten(bh)) | |
1207 | done = 1; | |
1208 | bh = bh->b_this_page; | |
1209 | } while (!done && (bh != head)); | |
1210 | } | |
55138e0b TT |
1211 | unlock_page(page); |
1212 | if (done) | |
1213 | break; | |
1214 | idx++; | |
1215 | num++; | |
659c6009 ES |
1216 | if (num >= max_pages) { |
1217 | done = 1; | |
55138e0b | 1218 | break; |
659c6009 | 1219 | } |
55138e0b TT |
1220 | } |
1221 | pagevec_release(&pvec); | |
1222 | } | |
1223 | return num; | |
1224 | } | |
1225 | ||
f5ab0d1f | 1226 | /* |
e35fd660 | 1227 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 1228 | * and returns if the blocks are already mapped. |
f5ab0d1f | 1229 | * |
f5ab0d1f MC |
1230 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
1231 | * and store the allocated blocks in the result buffer head and mark it | |
1232 | * mapped. | |
1233 | * | |
e35fd660 TT |
1234 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
1235 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
1236 | * based files |
1237 | * | |
1238 | * On success, it returns the number of blocks being mapped or allocate. | |
1239 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
1240 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
1241 | * the buffer head is mapped. | |
1242 | * | |
1243 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
1244 | * that casem, buffer head is unmapped | |
1245 | * | |
1246 | * It returns the error in case of allocation failure. | |
1247 | */ | |
e35fd660 TT |
1248 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
1249 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 AK |
1250 | { |
1251 | int retval; | |
f5ab0d1f | 1252 | |
e35fd660 TT |
1253 | map->m_flags = 0; |
1254 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
1255 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
1256 | (unsigned long) map->m_lblk); | |
4df3d265 | 1257 | /* |
b920c755 TT |
1258 | * Try to see if we can get the block without requesting a new |
1259 | * file system block. | |
4df3d265 AK |
1260 | */ |
1261 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 1262 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 1263 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
0e855ac8 | 1264 | } else { |
e35fd660 | 1265 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
0e855ac8 | 1266 | } |
4df3d265 | 1267 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 1268 | |
e35fd660 | 1269 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 1270 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
1271 | if (ret != 0) |
1272 | return ret; | |
1273 | } | |
1274 | ||
f5ab0d1f | 1275 | /* If it is only a block(s) look up */ |
c2177057 | 1276 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
1277 | return retval; |
1278 | ||
1279 | /* | |
1280 | * Returns if the blocks have already allocated | |
1281 | * | |
1282 | * Note that if blocks have been preallocated | |
1283 | * ext4_ext_get_block() returns th create = 0 | |
1284 | * with buffer head unmapped. | |
1285 | */ | |
e35fd660 | 1286 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
1287 | return retval; |
1288 | ||
2a8964d6 AK |
1289 | /* |
1290 | * When we call get_blocks without the create flag, the | |
1291 | * BH_Unwritten flag could have gotten set if the blocks | |
1292 | * requested were part of a uninitialized extent. We need to | |
1293 | * clear this flag now that we are committed to convert all or | |
1294 | * part of the uninitialized extent to be an initialized | |
1295 | * extent. This is because we need to avoid the combination | |
1296 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
1297 | * set on the buffer_head. | |
1298 | */ | |
e35fd660 | 1299 | map->m_flags &= ~EXT4_MAP_UNWRITTEN; |
2a8964d6 | 1300 | |
4df3d265 | 1301 | /* |
f5ab0d1f MC |
1302 | * New blocks allocate and/or writing to uninitialized extent |
1303 | * will possibly result in updating i_data, so we take | |
1304 | * the write lock of i_data_sem, and call get_blocks() | |
1305 | * with create == 1 flag. | |
4df3d265 AK |
1306 | */ |
1307 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
1308 | |
1309 | /* | |
1310 | * if the caller is from delayed allocation writeout path | |
1311 | * we have already reserved fs blocks for allocation | |
1312 | * let the underlying get_block() function know to | |
1313 | * avoid double accounting | |
1314 | */ | |
c2177057 | 1315 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
d2a17637 | 1316 | EXT4_I(inode)->i_delalloc_reserved_flag = 1; |
4df3d265 AK |
1317 | /* |
1318 | * We need to check for EXT4 here because migrate | |
1319 | * could have changed the inode type in between | |
1320 | */ | |
12e9b892 | 1321 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 1322 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 1323 | } else { |
e35fd660 | 1324 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 1325 | |
e35fd660 | 1326 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
1327 | /* |
1328 | * We allocated new blocks which will result in | |
1329 | * i_data's format changing. Force the migrate | |
1330 | * to fail by clearing migrate flags | |
1331 | */ | |
19f5fb7a | 1332 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 1333 | } |
d2a17637 | 1334 | |
5f634d06 AK |
1335 | /* |
1336 | * Update reserved blocks/metadata blocks after successful | |
1337 | * block allocation which had been deferred till now. We don't | |
1338 | * support fallocate for non extent files. So we can update | |
1339 | * reserve space here. | |
1340 | */ | |
1341 | if ((retval > 0) && | |
1296cc85 | 1342 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
1343 | ext4_da_update_reserve_space(inode, retval, 1); |
1344 | } | |
2ac3b6e0 | 1345 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
d2a17637 | 1346 | EXT4_I(inode)->i_delalloc_reserved_flag = 0; |
2ac3b6e0 | 1347 | |
4df3d265 | 1348 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 1349 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 1350 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
1351 | if (ret != 0) |
1352 | return ret; | |
1353 | } | |
0e855ac8 AK |
1354 | return retval; |
1355 | } | |
1356 | ||
f3bd1f3f MC |
1357 | /* Maximum number of blocks we map for direct IO at once. */ |
1358 | #define DIO_MAX_BLOCKS 4096 | |
1359 | ||
2ed88685 TT |
1360 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
1361 | struct buffer_head *bh, int flags) | |
ac27a0ec | 1362 | { |
3e4fdaf8 | 1363 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 1364 | struct ext4_map_blocks map; |
7fb5409d | 1365 | int ret = 0, started = 0; |
f3bd1f3f | 1366 | int dio_credits; |
ac27a0ec | 1367 | |
2ed88685 TT |
1368 | map.m_lblk = iblock; |
1369 | map.m_len = bh->b_size >> inode->i_blkbits; | |
1370 | ||
1371 | if (flags && !handle) { | |
7fb5409d | 1372 | /* Direct IO write... */ |
2ed88685 TT |
1373 | if (map.m_len > DIO_MAX_BLOCKS) |
1374 | map.m_len = DIO_MAX_BLOCKS; | |
1375 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
f3bd1f3f | 1376 | handle = ext4_journal_start(inode, dio_credits); |
7fb5409d | 1377 | if (IS_ERR(handle)) { |
ac27a0ec | 1378 | ret = PTR_ERR(handle); |
2ed88685 | 1379 | return ret; |
ac27a0ec | 1380 | } |
7fb5409d | 1381 | started = 1; |
ac27a0ec DK |
1382 | } |
1383 | ||
2ed88685 | 1384 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 1385 | if (ret > 0) { |
2ed88685 TT |
1386 | map_bh(bh, inode->i_sb, map.m_pblk); |
1387 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1388 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 1389 | ret = 0; |
ac27a0ec | 1390 | } |
7fb5409d JK |
1391 | if (started) |
1392 | ext4_journal_stop(handle); | |
ac27a0ec DK |
1393 | return ret; |
1394 | } | |
1395 | ||
2ed88685 TT |
1396 | int ext4_get_block(struct inode *inode, sector_t iblock, |
1397 | struct buffer_head *bh, int create) | |
1398 | { | |
1399 | return _ext4_get_block(inode, iblock, bh, | |
1400 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
1401 | } | |
1402 | ||
ac27a0ec DK |
1403 | /* |
1404 | * `handle' can be NULL if create is zero | |
1405 | */ | |
617ba13b | 1406 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 1407 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 1408 | { |
2ed88685 TT |
1409 | struct ext4_map_blocks map; |
1410 | struct buffer_head *bh; | |
ac27a0ec DK |
1411 | int fatal = 0, err; |
1412 | ||
1413 | J_ASSERT(handle != NULL || create == 0); | |
1414 | ||
2ed88685 TT |
1415 | map.m_lblk = block; |
1416 | map.m_len = 1; | |
1417 | err = ext4_map_blocks(handle, inode, &map, | |
1418 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 1419 | |
2ed88685 TT |
1420 | if (err < 0) |
1421 | *errp = err; | |
1422 | if (err <= 0) | |
1423 | return NULL; | |
1424 | *errp = 0; | |
1425 | ||
1426 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
1427 | if (!bh) { | |
1428 | *errp = -EIO; | |
1429 | return NULL; | |
ac27a0ec | 1430 | } |
2ed88685 TT |
1431 | if (map.m_flags & EXT4_MAP_NEW) { |
1432 | J_ASSERT(create != 0); | |
1433 | J_ASSERT(handle != NULL); | |
ac27a0ec | 1434 | |
2ed88685 TT |
1435 | /* |
1436 | * Now that we do not always journal data, we should | |
1437 | * keep in mind whether this should always journal the | |
1438 | * new buffer as metadata. For now, regular file | |
1439 | * writes use ext4_get_block instead, so it's not a | |
1440 | * problem. | |
1441 | */ | |
1442 | lock_buffer(bh); | |
1443 | BUFFER_TRACE(bh, "call get_create_access"); | |
1444 | fatal = ext4_journal_get_create_access(handle, bh); | |
1445 | if (!fatal && !buffer_uptodate(bh)) { | |
1446 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
1447 | set_buffer_uptodate(bh); | |
ac27a0ec | 1448 | } |
2ed88685 TT |
1449 | unlock_buffer(bh); |
1450 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
1451 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
1452 | if (!fatal) | |
1453 | fatal = err; | |
1454 | } else { | |
1455 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 1456 | } |
2ed88685 TT |
1457 | if (fatal) { |
1458 | *errp = fatal; | |
1459 | brelse(bh); | |
1460 | bh = NULL; | |
1461 | } | |
1462 | return bh; | |
ac27a0ec DK |
1463 | } |
1464 | ||
617ba13b | 1465 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 1466 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 1467 | { |
af5bc92d | 1468 | struct buffer_head *bh; |
ac27a0ec | 1469 | |
617ba13b | 1470 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
1471 | if (!bh) |
1472 | return bh; | |
1473 | if (buffer_uptodate(bh)) | |
1474 | return bh; | |
1475 | ll_rw_block(READ_META, 1, &bh); | |
1476 | wait_on_buffer(bh); | |
1477 | if (buffer_uptodate(bh)) | |
1478 | return bh; | |
1479 | put_bh(bh); | |
1480 | *err = -EIO; | |
1481 | return NULL; | |
1482 | } | |
1483 | ||
af5bc92d TT |
1484 | static int walk_page_buffers(handle_t *handle, |
1485 | struct buffer_head *head, | |
1486 | unsigned from, | |
1487 | unsigned to, | |
1488 | int *partial, | |
1489 | int (*fn)(handle_t *handle, | |
1490 | struct buffer_head *bh)) | |
ac27a0ec DK |
1491 | { |
1492 | struct buffer_head *bh; | |
1493 | unsigned block_start, block_end; | |
1494 | unsigned blocksize = head->b_size; | |
1495 | int err, ret = 0; | |
1496 | struct buffer_head *next; | |
1497 | ||
af5bc92d TT |
1498 | for (bh = head, block_start = 0; |
1499 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1500 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1501 | next = bh->b_this_page; |
1502 | block_end = block_start + blocksize; | |
1503 | if (block_end <= from || block_start >= to) { | |
1504 | if (partial && !buffer_uptodate(bh)) | |
1505 | *partial = 1; | |
1506 | continue; | |
1507 | } | |
1508 | err = (*fn)(handle, bh); | |
1509 | if (!ret) | |
1510 | ret = err; | |
1511 | } | |
1512 | return ret; | |
1513 | } | |
1514 | ||
1515 | /* | |
1516 | * To preserve ordering, it is essential that the hole instantiation and | |
1517 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1518 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1519 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1520 | * prepare_write() is the right place. |
1521 | * | |
617ba13b MC |
1522 | * Also, this function can nest inside ext4_writepage() -> |
1523 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
1524 | * has generated enough buffer credits to do the whole page. So we won't |
1525 | * block on the journal in that case, which is good, because the caller may | |
1526 | * be PF_MEMALLOC. | |
1527 | * | |
617ba13b | 1528 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1529 | * quota file writes. If we were to commit the transaction while thus |
1530 | * reentered, there can be a deadlock - we would be holding a quota | |
1531 | * lock, and the commit would never complete if another thread had a | |
1532 | * transaction open and was blocking on the quota lock - a ranking | |
1533 | * violation. | |
1534 | * | |
dab291af | 1535 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1536 | * will _not_ run commit under these circumstances because handle->h_ref |
1537 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1538 | * write. | |
1539 | */ | |
1540 | static int do_journal_get_write_access(handle_t *handle, | |
de9a55b8 | 1541 | struct buffer_head *bh) |
ac27a0ec | 1542 | { |
56d35a4c JK |
1543 | int dirty = buffer_dirty(bh); |
1544 | int ret; | |
1545 | ||
ac27a0ec DK |
1546 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1547 | return 0; | |
56d35a4c JK |
1548 | /* |
1549 | * __block_prepare_write() could have dirtied some buffers. Clean | |
1550 | * the dirty bit as jbd2_journal_get_write_access() could complain | |
1551 | * otherwise about fs integrity issues. Setting of the dirty bit | |
1552 | * by __block_prepare_write() isn't a real problem here as we clear | |
1553 | * the bit before releasing a page lock and thus writeback cannot | |
1554 | * ever write the buffer. | |
1555 | */ | |
1556 | if (dirty) | |
1557 | clear_buffer_dirty(bh); | |
1558 | ret = ext4_journal_get_write_access(handle, bh); | |
1559 | if (!ret && dirty) | |
1560 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1561 | return ret; | |
ac27a0ec DK |
1562 | } |
1563 | ||
b9a4207d JK |
1564 | /* |
1565 | * Truncate blocks that were not used by write. We have to truncate the | |
1566 | * pagecache as well so that corresponding buffers get properly unmapped. | |
1567 | */ | |
1568 | static void ext4_truncate_failed_write(struct inode *inode) | |
1569 | { | |
1570 | truncate_inode_pages(inode->i_mapping, inode->i_size); | |
1571 | ext4_truncate(inode); | |
1572 | } | |
1573 | ||
744692dc JZ |
1574 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
1575 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 1576 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1577 | loff_t pos, unsigned len, unsigned flags, |
1578 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1579 | { |
af5bc92d | 1580 | struct inode *inode = mapping->host; |
1938a150 | 1581 | int ret, needed_blocks; |
ac27a0ec DK |
1582 | handle_t *handle; |
1583 | int retries = 0; | |
af5bc92d | 1584 | struct page *page; |
de9a55b8 | 1585 | pgoff_t index; |
af5bc92d | 1586 | unsigned from, to; |
bfc1af65 | 1587 | |
9bffad1e | 1588 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1589 | /* |
1590 | * Reserve one block more for addition to orphan list in case | |
1591 | * we allocate blocks but write fails for some reason | |
1592 | */ | |
1593 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 1594 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
1595 | from = pos & (PAGE_CACHE_SIZE - 1); |
1596 | to = from + len; | |
ac27a0ec DK |
1597 | |
1598 | retry: | |
af5bc92d TT |
1599 | handle = ext4_journal_start(inode, needed_blocks); |
1600 | if (IS_ERR(handle)) { | |
1601 | ret = PTR_ERR(handle); | |
1602 | goto out; | |
7479d2b9 | 1603 | } |
ac27a0ec | 1604 | |
ebd3610b JK |
1605 | /* We cannot recurse into the filesystem as the transaction is already |
1606 | * started */ | |
1607 | flags |= AOP_FLAG_NOFS; | |
1608 | ||
54566b2c | 1609 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
1610 | if (!page) { |
1611 | ext4_journal_stop(handle); | |
1612 | ret = -ENOMEM; | |
1613 | goto out; | |
1614 | } | |
1615 | *pagep = page; | |
1616 | ||
744692dc | 1617 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1618 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1619 | else |
6e1db88d | 1620 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
1621 | |
1622 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
1623 | ret = walk_page_buffers(handle, page_buffers(page), |
1624 | from, to, NULL, do_journal_get_write_access); | |
1625 | } | |
bfc1af65 NP |
1626 | |
1627 | if (ret) { | |
af5bc92d | 1628 | unlock_page(page); |
af5bc92d | 1629 | page_cache_release(page); |
ae4d5372 | 1630 | /* |
6e1db88d | 1631 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1632 | * outside i_size. Trim these off again. Don't need |
1633 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1634 | * |
1635 | * Add inode to orphan list in case we crash before | |
1636 | * truncate finishes | |
ae4d5372 | 1637 | */ |
ffacfa7a | 1638 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1639 | ext4_orphan_add(handle, inode); |
1640 | ||
1641 | ext4_journal_stop(handle); | |
1642 | if (pos + len > inode->i_size) { | |
b9a4207d | 1643 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1644 | /* |
ffacfa7a | 1645 | * If truncate failed early the inode might |
1938a150 AK |
1646 | * still be on the orphan list; we need to |
1647 | * make sure the inode is removed from the | |
1648 | * orphan list in that case. | |
1649 | */ | |
1650 | if (inode->i_nlink) | |
1651 | ext4_orphan_del(NULL, inode); | |
1652 | } | |
bfc1af65 NP |
1653 | } |
1654 | ||
617ba13b | 1655 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 1656 | goto retry; |
7479d2b9 | 1657 | out: |
ac27a0ec DK |
1658 | return ret; |
1659 | } | |
1660 | ||
bfc1af65 NP |
1661 | /* For write_end() in data=journal mode */ |
1662 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
1663 | { |
1664 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1665 | return 0; | |
1666 | set_buffer_uptodate(bh); | |
0390131b | 1667 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
1668 | } |
1669 | ||
f8514083 | 1670 | static int ext4_generic_write_end(struct file *file, |
de9a55b8 TT |
1671 | struct address_space *mapping, |
1672 | loff_t pos, unsigned len, unsigned copied, | |
1673 | struct page *page, void *fsdata) | |
f8514083 AK |
1674 | { |
1675 | int i_size_changed = 0; | |
1676 | struct inode *inode = mapping->host; | |
1677 | handle_t *handle = ext4_journal_current_handle(); | |
1678 | ||
1679 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
1680 | ||
1681 | /* | |
1682 | * No need to use i_size_read() here, the i_size | |
1683 | * cannot change under us because we hold i_mutex. | |
1684 | * | |
1685 | * But it's important to update i_size while still holding page lock: | |
1686 | * page writeout could otherwise come in and zero beyond i_size. | |
1687 | */ | |
1688 | if (pos + copied > inode->i_size) { | |
1689 | i_size_write(inode, pos + copied); | |
1690 | i_size_changed = 1; | |
1691 | } | |
1692 | ||
1693 | if (pos + copied > EXT4_I(inode)->i_disksize) { | |
1694 | /* We need to mark inode dirty even if | |
1695 | * new_i_size is less that inode->i_size | |
1696 | * bu greater than i_disksize.(hint delalloc) | |
1697 | */ | |
1698 | ext4_update_i_disksize(inode, (pos + copied)); | |
1699 | i_size_changed = 1; | |
1700 | } | |
1701 | unlock_page(page); | |
1702 | page_cache_release(page); | |
1703 | ||
1704 | /* | |
1705 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1706 | * makes the holding time of page lock longer. Second, it forces lock | |
1707 | * ordering of page lock and transaction start for journaling | |
1708 | * filesystems. | |
1709 | */ | |
1710 | if (i_size_changed) | |
1711 | ext4_mark_inode_dirty(handle, inode); | |
1712 | ||
1713 | return copied; | |
1714 | } | |
1715 | ||
ac27a0ec DK |
1716 | /* |
1717 | * We need to pick up the new inode size which generic_commit_write gave us | |
1718 | * `file' can be NULL - eg, when called from page_symlink(). | |
1719 | * | |
617ba13b | 1720 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1721 | * buffers are managed internally. |
1722 | */ | |
bfc1af65 | 1723 | static int ext4_ordered_write_end(struct file *file, |
de9a55b8 TT |
1724 | struct address_space *mapping, |
1725 | loff_t pos, unsigned len, unsigned copied, | |
1726 | struct page *page, void *fsdata) | |
ac27a0ec | 1727 | { |
617ba13b | 1728 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1729 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1730 | int ret = 0, ret2; |
1731 | ||
9bffad1e | 1732 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
678aaf48 | 1733 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
1734 | |
1735 | if (ret == 0) { | |
f8514083 | 1736 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1737 | page, fsdata); |
f8a87d89 | 1738 | copied = ret2; |
ffacfa7a | 1739 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1740 | /* if we have allocated more blocks and copied |
1741 | * less. We will have blocks allocated outside | |
1742 | * inode->i_size. So truncate them | |
1743 | */ | |
1744 | ext4_orphan_add(handle, inode); | |
f8a87d89 RK |
1745 | if (ret2 < 0) |
1746 | ret = ret2; | |
ac27a0ec | 1747 | } |
617ba13b | 1748 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1749 | if (!ret) |
1750 | ret = ret2; | |
bfc1af65 | 1751 | |
f8514083 | 1752 | if (pos + len > inode->i_size) { |
b9a4207d | 1753 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1754 | /* |
ffacfa7a | 1755 | * If truncate failed early the inode might still be |
f8514083 AK |
1756 | * on the orphan list; we need to make sure the inode |
1757 | * is removed from the orphan list in that case. | |
1758 | */ | |
1759 | if (inode->i_nlink) | |
1760 | ext4_orphan_del(NULL, inode); | |
1761 | } | |
1762 | ||
1763 | ||
bfc1af65 | 1764 | return ret ? ret : copied; |
ac27a0ec DK |
1765 | } |
1766 | ||
bfc1af65 | 1767 | static int ext4_writeback_write_end(struct file *file, |
de9a55b8 TT |
1768 | struct address_space *mapping, |
1769 | loff_t pos, unsigned len, unsigned copied, | |
1770 | struct page *page, void *fsdata) | |
ac27a0ec | 1771 | { |
617ba13b | 1772 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1773 | struct inode *inode = mapping->host; |
ac27a0ec | 1774 | int ret = 0, ret2; |
ac27a0ec | 1775 | |
9bffad1e | 1776 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
f8514083 | 1777 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1778 | page, fsdata); |
f8a87d89 | 1779 | copied = ret2; |
ffacfa7a | 1780 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1781 | /* if we have allocated more blocks and copied |
1782 | * less. We will have blocks allocated outside | |
1783 | * inode->i_size. So truncate them | |
1784 | */ | |
1785 | ext4_orphan_add(handle, inode); | |
1786 | ||
f8a87d89 RK |
1787 | if (ret2 < 0) |
1788 | ret = ret2; | |
ac27a0ec | 1789 | |
617ba13b | 1790 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1791 | if (!ret) |
1792 | ret = ret2; | |
bfc1af65 | 1793 | |
f8514083 | 1794 | if (pos + len > inode->i_size) { |
b9a4207d | 1795 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1796 | /* |
ffacfa7a | 1797 | * If truncate failed early the inode might still be |
f8514083 AK |
1798 | * on the orphan list; we need to make sure the inode |
1799 | * is removed from the orphan list in that case. | |
1800 | */ | |
1801 | if (inode->i_nlink) | |
1802 | ext4_orphan_del(NULL, inode); | |
1803 | } | |
1804 | ||
bfc1af65 | 1805 | return ret ? ret : copied; |
ac27a0ec DK |
1806 | } |
1807 | ||
bfc1af65 | 1808 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1809 | struct address_space *mapping, |
1810 | loff_t pos, unsigned len, unsigned copied, | |
1811 | struct page *page, void *fsdata) | |
ac27a0ec | 1812 | { |
617ba13b | 1813 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1814 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1815 | int ret = 0, ret2; |
1816 | int partial = 0; | |
bfc1af65 | 1817 | unsigned from, to; |
cf17fea6 | 1818 | loff_t new_i_size; |
ac27a0ec | 1819 | |
9bffad1e | 1820 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1821 | from = pos & (PAGE_CACHE_SIZE - 1); |
1822 | to = from + len; | |
1823 | ||
1824 | if (copied < len) { | |
1825 | if (!PageUptodate(page)) | |
1826 | copied = 0; | |
1827 | page_zero_new_buffers(page, from+copied, to); | |
1828 | } | |
ac27a0ec DK |
1829 | |
1830 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1831 | to, &partial, write_end_fn); |
ac27a0ec DK |
1832 | if (!partial) |
1833 | SetPageUptodate(page); | |
cf17fea6 AK |
1834 | new_i_size = pos + copied; |
1835 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1836 | i_size_write(inode, pos+copied); |
19f5fb7a | 1837 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
cf17fea6 AK |
1838 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1839 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1840 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1841 | if (!ret) |
1842 | ret = ret2; | |
1843 | } | |
bfc1af65 | 1844 | |
cf108bca | 1845 | unlock_page(page); |
f8514083 | 1846 | page_cache_release(page); |
ffacfa7a | 1847 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1848 | /* if we have allocated more blocks and copied |
1849 | * less. We will have blocks allocated outside | |
1850 | * inode->i_size. So truncate them | |
1851 | */ | |
1852 | ext4_orphan_add(handle, inode); | |
1853 | ||
617ba13b | 1854 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1855 | if (!ret) |
1856 | ret = ret2; | |
f8514083 | 1857 | if (pos + len > inode->i_size) { |
b9a4207d | 1858 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1859 | /* |
ffacfa7a | 1860 | * If truncate failed early the inode might still be |
f8514083 AK |
1861 | * on the orphan list; we need to make sure the inode |
1862 | * is removed from the orphan list in that case. | |
1863 | */ | |
1864 | if (inode->i_nlink) | |
1865 | ext4_orphan_del(NULL, inode); | |
1866 | } | |
bfc1af65 NP |
1867 | |
1868 | return ret ? ret : copied; | |
ac27a0ec | 1869 | } |
d2a17637 | 1870 | |
9d0be502 TT |
1871 | /* |
1872 | * Reserve a single block located at lblock | |
1873 | */ | |
1874 | static int ext4_da_reserve_space(struct inode *inode, sector_t lblock) | |
d2a17637 | 1875 | { |
030ba6bc | 1876 | int retries = 0; |
60e58e0f | 1877 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1878 | struct ext4_inode_info *ei = EXT4_I(inode); |
72b8ab9d | 1879 | unsigned long md_needed; |
5dd4056d | 1880 | int ret; |
d2a17637 MC |
1881 | |
1882 | /* | |
1883 | * recalculate the amount of metadata blocks to reserve | |
1884 | * in order to allocate nrblocks | |
1885 | * worse case is one extent per block | |
1886 | */ | |
030ba6bc | 1887 | repeat: |
0637c6f4 | 1888 | spin_lock(&ei->i_block_reservation_lock); |
9d0be502 | 1889 | md_needed = ext4_calc_metadata_amount(inode, lblock); |
f8ec9d68 | 1890 | trace_ext4_da_reserve_space(inode, md_needed); |
0637c6f4 | 1891 | spin_unlock(&ei->i_block_reservation_lock); |
d2a17637 | 1892 | |
60e58e0f | 1893 | /* |
72b8ab9d ES |
1894 | * We will charge metadata quota at writeout time; this saves |
1895 | * us from metadata over-estimation, though we may go over by | |
1896 | * a small amount in the end. Here we just reserve for data. | |
60e58e0f | 1897 | */ |
72b8ab9d | 1898 | ret = dquot_reserve_block(inode, 1); |
5dd4056d CH |
1899 | if (ret) |
1900 | return ret; | |
72b8ab9d ES |
1901 | /* |
1902 | * We do still charge estimated metadata to the sb though; | |
1903 | * we cannot afford to run out of free blocks. | |
1904 | */ | |
9d0be502 | 1905 | if (ext4_claim_free_blocks(sbi, md_needed + 1)) { |
72b8ab9d | 1906 | dquot_release_reservation_block(inode, 1); |
030ba6bc AK |
1907 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1908 | yield(); | |
1909 | goto repeat; | |
1910 | } | |
d2a17637 MC |
1911 | return -ENOSPC; |
1912 | } | |
0637c6f4 | 1913 | spin_lock(&ei->i_block_reservation_lock); |
9d0be502 | 1914 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1915 | ei->i_reserved_meta_blocks += md_needed; |
1916 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1917 | |
d2a17637 MC |
1918 | return 0; /* success */ |
1919 | } | |
1920 | ||
12219aea | 1921 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1922 | { |
1923 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1924 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1925 | |
cd213226 MC |
1926 | if (!to_free) |
1927 | return; /* Nothing to release, exit */ | |
1928 | ||
d2a17637 | 1929 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1930 | |
5a58ec87 | 1931 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1932 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1933 | /* |
0637c6f4 TT |
1934 | * if there aren't enough reserved blocks, then the |
1935 | * counter is messed up somewhere. Since this | |
1936 | * function is called from invalidate page, it's | |
1937 | * harmless to return without any action. | |
cd213226 | 1938 | */ |
0637c6f4 TT |
1939 | ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: " |
1940 | "ino %lu, to_free %d with only %d reserved " | |
1941 | "data blocks\n", inode->i_ino, to_free, | |
1942 | ei->i_reserved_data_blocks); | |
1943 | WARN_ON(1); | |
1944 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1945 | } |
0637c6f4 | 1946 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1947 | |
0637c6f4 TT |
1948 | if (ei->i_reserved_data_blocks == 0) { |
1949 | /* | |
1950 | * We can release all of the reserved metadata blocks | |
1951 | * only when we have written all of the delayed | |
1952 | * allocation blocks. | |
1953 | */ | |
72b8ab9d ES |
1954 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, |
1955 | ei->i_reserved_meta_blocks); | |
ee5f4d9c | 1956 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1957 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1958 | } |
d2a17637 | 1959 | |
72b8ab9d | 1960 | /* update fs dirty data blocks counter */ |
0637c6f4 | 1961 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free); |
d2a17637 | 1962 | |
d2a17637 | 1963 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1964 | |
5dd4056d | 1965 | dquot_release_reservation_block(inode, to_free); |
d2a17637 MC |
1966 | } |
1967 | ||
1968 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1969 | unsigned long offset) |
d2a17637 MC |
1970 | { |
1971 | int to_release = 0; | |
1972 | struct buffer_head *head, *bh; | |
1973 | unsigned int curr_off = 0; | |
1974 | ||
1975 | head = page_buffers(page); | |
1976 | bh = head; | |
1977 | do { | |
1978 | unsigned int next_off = curr_off + bh->b_size; | |
1979 | ||
1980 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1981 | to_release++; | |
1982 | clear_buffer_delay(bh); | |
1983 | } | |
1984 | curr_off = next_off; | |
1985 | } while ((bh = bh->b_this_page) != head); | |
12219aea | 1986 | ext4_da_release_space(page->mapping->host, to_release); |
d2a17637 | 1987 | } |
ac27a0ec | 1988 | |
64769240 AT |
1989 | /* |
1990 | * Delayed allocation stuff | |
1991 | */ | |
1992 | ||
64769240 AT |
1993 | /* |
1994 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1995 | * them with writepage() call back |
64769240 AT |
1996 | * |
1997 | * @mpd->inode: inode | |
1998 | * @mpd->first_page: first page of the extent | |
1999 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
2000 | * |
2001 | * By the time mpage_da_submit_io() is called we expect all blocks | |
2002 | * to be allocated. this may be wrong if allocation failed. | |
2003 | * | |
2004 | * As pages are already locked by write_cache_pages(), we can't use it | |
2005 | */ | |
1de3e3df TT |
2006 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
2007 | struct ext4_map_blocks *map) | |
64769240 | 2008 | { |
791b7f08 AK |
2009 | struct pagevec pvec; |
2010 | unsigned long index, end; | |
2011 | int ret = 0, err, nr_pages, i; | |
2012 | struct inode *inode = mpd->inode; | |
2013 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 2014 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
2015 | unsigned int len, block_start; |
2016 | struct buffer_head *bh, *page_bufs = NULL; | |
cb20d518 | 2017 | int journal_data = ext4_should_journal_data(inode); |
1de3e3df | 2018 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 2019 | struct ext4_io_submit io_submit; |
64769240 AT |
2020 | |
2021 | BUG_ON(mpd->next_page <= mpd->first_page); | |
bd2d0210 | 2022 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
2023 | /* |
2024 | * We need to start from the first_page to the next_page - 1 | |
2025 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 2026 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
2027 | * at the currently mapped buffer_heads. |
2028 | */ | |
64769240 AT |
2029 | index = mpd->first_page; |
2030 | end = mpd->next_page - 1; | |
2031 | ||
791b7f08 | 2032 | pagevec_init(&pvec, 0); |
64769240 | 2033 | while (index <= end) { |
791b7f08 | 2034 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
2035 | if (nr_pages == 0) |
2036 | break; | |
2037 | for (i = 0; i < nr_pages; i++) { | |
1de3e3df | 2038 | int commit_write = 0, redirty_page = 0; |
64769240 AT |
2039 | struct page *page = pvec.pages[i]; |
2040 | ||
791b7f08 AK |
2041 | index = page->index; |
2042 | if (index > end) | |
2043 | break; | |
cb20d518 TT |
2044 | |
2045 | if (index == size >> PAGE_CACHE_SHIFT) | |
2046 | len = size & ~PAGE_CACHE_MASK; | |
2047 | else | |
2048 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
2049 | if (map) { |
2050 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
2051 | inode->i_blkbits); | |
2052 | pblock = map->m_pblk + (cur_logical - | |
2053 | map->m_lblk); | |
2054 | } | |
791b7f08 AK |
2055 | index++; |
2056 | ||
2057 | BUG_ON(!PageLocked(page)); | |
2058 | BUG_ON(PageWriteback(page)); | |
2059 | ||
cb20d518 TT |
2060 | /* |
2061 | * If the page does not have buffers (for | |
2062 | * whatever reason), try to create them using | |
2063 | * block_prepare_write. If this fails, | |
2064 | * redirty the page and move on. | |
2065 | */ | |
2066 | if (!page_has_buffers(page)) { | |
2067 | if (block_prepare_write(page, 0, len, | |
2068 | noalloc_get_block_write)) { | |
2069 | redirty_page: | |
2070 | redirty_page_for_writepage(mpd->wbc, | |
2071 | page); | |
2072 | unlock_page(page); | |
2073 | continue; | |
2074 | } | |
2075 | commit_write = 1; | |
2076 | } | |
3ecdb3a1 TT |
2077 | |
2078 | bh = page_bufs = page_buffers(page); | |
2079 | block_start = 0; | |
2080 | do { | |
1de3e3df | 2081 | if (!bh) |
3ecdb3a1 | 2082 | goto redirty_page; |
1de3e3df TT |
2083 | if (map && (cur_logical >= map->m_lblk) && |
2084 | (cur_logical <= (map->m_lblk + | |
2085 | (map->m_len - 1)))) { | |
2086 | if (buffer_delay(bh)) { | |
2087 | clear_buffer_delay(bh); | |
2088 | bh->b_blocknr = pblock; | |
2089 | } | |
2090 | if (buffer_unwritten(bh) || | |
2091 | buffer_mapped(bh)) | |
2092 | BUG_ON(bh->b_blocknr != pblock); | |
2093 | if (map->m_flags & EXT4_MAP_UNINIT) | |
2094 | set_buffer_uninit(bh); | |
2095 | clear_buffer_unwritten(bh); | |
2096 | } | |
2097 | ||
2098 | /* redirty page if block allocation undone */ | |
2099 | if (buffer_delay(bh) || buffer_unwritten(bh)) | |
2100 | redirty_page = 1; | |
3ecdb3a1 TT |
2101 | bh = bh->b_this_page; |
2102 | block_start += bh->b_size; | |
1de3e3df TT |
2103 | cur_logical++; |
2104 | pblock++; | |
2105 | } while (bh != page_bufs); | |
2106 | ||
2107 | if (redirty_page) | |
2108 | goto redirty_page; | |
cb20d518 TT |
2109 | |
2110 | if (commit_write) | |
2111 | /* mark the buffer_heads as dirty & uptodate */ | |
2112 | block_commit_write(page, 0, len); | |
2113 | ||
bd2d0210 TT |
2114 | /* |
2115 | * Delalloc doesn't support data journalling, | |
2116 | * but eventually maybe we'll lift this | |
2117 | * restriction. | |
2118 | */ | |
2119 | if (unlikely(journal_data && PageChecked(page))) | |
cb20d518 | 2120 | err = __ext4_journalled_writepage(page, len); |
bd2d0210 TT |
2121 | else |
2122 | err = ext4_bio_write_page(&io_submit, page, | |
2123 | len, mpd->wbc); | |
cb20d518 TT |
2124 | |
2125 | if (!err) | |
a1d6cc56 | 2126 | mpd->pages_written++; |
64769240 AT |
2127 | /* |
2128 | * In error case, we have to continue because | |
2129 | * remaining pages are still locked | |
64769240 AT |
2130 | */ |
2131 | if (ret == 0) | |
2132 | ret = err; | |
2133 | } | |
2134 | pagevec_release(&pvec); | |
2135 | } | |
bd2d0210 | 2136 | ext4_io_submit(&io_submit); |
64769240 AT |
2137 | return ret; |
2138 | } | |
2139 | ||
c4a0c46e AK |
2140 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, |
2141 | sector_t logical, long blk_cnt) | |
2142 | { | |
2143 | int nr_pages, i; | |
2144 | pgoff_t index, end; | |
2145 | struct pagevec pvec; | |
2146 | struct inode *inode = mpd->inode; | |
2147 | struct address_space *mapping = inode->i_mapping; | |
2148 | ||
2149 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2150 | end = (logical + blk_cnt - 1) >> | |
2151 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2152 | while (index <= end) { | |
2153 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
2154 | if (nr_pages == 0) | |
2155 | break; | |
2156 | for (i = 0; i < nr_pages; i++) { | |
2157 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 2158 | if (page->index > end) |
c4a0c46e | 2159 | break; |
c4a0c46e AK |
2160 | BUG_ON(!PageLocked(page)); |
2161 | BUG_ON(PageWriteback(page)); | |
2162 | block_invalidatepage(page, 0); | |
2163 | ClearPageUptodate(page); | |
2164 | unlock_page(page); | |
2165 | } | |
9b1d0998 JK |
2166 | index = pvec.pages[nr_pages - 1]->index + 1; |
2167 | pagevec_release(&pvec); | |
c4a0c46e AK |
2168 | } |
2169 | return; | |
2170 | } | |
2171 | ||
df22291f AK |
2172 | static void ext4_print_free_blocks(struct inode *inode) |
2173 | { | |
2174 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1693918e TT |
2175 | printk(KERN_CRIT "Total free blocks count %lld\n", |
2176 | ext4_count_free_blocks(inode->i_sb)); | |
2177 | printk(KERN_CRIT "Free/Dirty block details\n"); | |
2178 | printk(KERN_CRIT "free_blocks=%lld\n", | |
2179 | (long long) percpu_counter_sum(&sbi->s_freeblocks_counter)); | |
2180 | printk(KERN_CRIT "dirty_blocks=%lld\n", | |
2181 | (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter)); | |
2182 | printk(KERN_CRIT "Block reservation details\n"); | |
2183 | printk(KERN_CRIT "i_reserved_data_blocks=%u\n", | |
2184 | EXT4_I(inode)->i_reserved_data_blocks); | |
2185 | printk(KERN_CRIT "i_reserved_meta_blocks=%u\n", | |
2186 | EXT4_I(inode)->i_reserved_meta_blocks); | |
df22291f AK |
2187 | return; |
2188 | } | |
2189 | ||
64769240 | 2190 | /* |
5a87b7a5 TT |
2191 | * mpage_da_map_and_submit - go through given space, map them |
2192 | * if necessary, and then submit them for I/O | |
64769240 | 2193 | * |
8dc207c0 | 2194 | * @mpd - bh describing space |
64769240 AT |
2195 | * |
2196 | * The function skips space we know is already mapped to disk blocks. | |
2197 | * | |
64769240 | 2198 | */ |
5a87b7a5 | 2199 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 2200 | { |
2ac3b6e0 | 2201 | int err, blks, get_blocks_flags; |
1de3e3df | 2202 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
2203 | sector_t next = mpd->b_blocknr; |
2204 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
2205 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
2206 | handle_t *handle = NULL; | |
64769240 AT |
2207 | |
2208 | /* | |
5a87b7a5 TT |
2209 | * If the blocks are mapped already, or we couldn't accumulate |
2210 | * any blocks, then proceed immediately to the submission stage. | |
64769240 | 2211 | */ |
5a87b7a5 TT |
2212 | if ((mpd->b_size == 0) || |
2213 | ((mpd->b_state & (1 << BH_Mapped)) && | |
2214 | !(mpd->b_state & (1 << BH_Delay)) && | |
2215 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
2216 | goto submit_io; | |
2fa3cdfb TT |
2217 | |
2218 | handle = ext4_journal_current_handle(); | |
2219 | BUG_ON(!handle); | |
2220 | ||
79ffab34 | 2221 | /* |
79e83036 | 2222 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
2223 | * blocks, or to convert an uninitialized extent to be |
2224 | * initialized (in the case where we have written into | |
2225 | * one or more preallocated blocks). | |
2226 | * | |
2227 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
2228 | * indicate that we are on the delayed allocation path. This | |
2229 | * affects functions in many different parts of the allocation | |
2230 | * call path. This flag exists primarily because we don't | |
79e83036 | 2231 | * want to change *many* call functions, so ext4_map_blocks() |
2ac3b6e0 TT |
2232 | * will set the magic i_delalloc_reserved_flag once the |
2233 | * inode's allocation semaphore is taken. | |
2234 | * | |
2235 | * If the blocks in questions were delalloc blocks, set | |
2236 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
2237 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 2238 | */ |
2ed88685 TT |
2239 | map.m_lblk = next; |
2240 | map.m_len = max_blocks; | |
1296cc85 | 2241 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE; |
744692dc JZ |
2242 | if (ext4_should_dioread_nolock(mpd->inode)) |
2243 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 2244 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
2245 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
2246 | ||
2ed88685 | 2247 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 2248 | if (blks < 0) { |
e3570639 ES |
2249 | struct super_block *sb = mpd->inode->i_sb; |
2250 | ||
2fa3cdfb | 2251 | err = blks; |
ed5bde0b | 2252 | /* |
5a87b7a5 TT |
2253 | * If get block returns EAGAIN or ENOSPC and there |
2254 | * appears to be free blocks we will call | |
2255 | * ext4_writepage() for all of the pages which will | |
2256 | * just redirty the pages. | |
c4a0c46e AK |
2257 | */ |
2258 | if (err == -EAGAIN) | |
5a87b7a5 | 2259 | goto submit_io; |
df22291f AK |
2260 | |
2261 | if (err == -ENOSPC && | |
e3570639 | 2262 | ext4_count_free_blocks(sb)) { |
df22291f | 2263 | mpd->retval = err; |
5a87b7a5 | 2264 | goto submit_io; |
df22291f AK |
2265 | } |
2266 | ||
c4a0c46e | 2267 | /* |
ed5bde0b TT |
2268 | * get block failure will cause us to loop in |
2269 | * writepages, because a_ops->writepage won't be able | |
2270 | * to make progress. The page will be redirtied by | |
2271 | * writepage and writepages will again try to write | |
2272 | * the same. | |
c4a0c46e | 2273 | */ |
e3570639 ES |
2274 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2275 | ext4_msg(sb, KERN_CRIT, | |
2276 | "delayed block allocation failed for inode %lu " | |
2277 | "at logical offset %llu with max blocks %zd " | |
2278 | "with error %d", mpd->inode->i_ino, | |
2279 | (unsigned long long) next, | |
2280 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
2281 | ext4_msg(sb, KERN_CRIT, | |
2282 | "This should not happen!! Data will be lost\n"); | |
2283 | if (err == -ENOSPC) | |
2284 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 2285 | } |
2fa3cdfb | 2286 | /* invalidate all the pages */ |
c4a0c46e | 2287 | ext4_da_block_invalidatepages(mpd, next, |
8dc207c0 | 2288 | mpd->b_size >> mpd->inode->i_blkbits); |
5a87b7a5 | 2289 | return; |
c4a0c46e | 2290 | } |
2fa3cdfb TT |
2291 | BUG_ON(blks == 0); |
2292 | ||
1de3e3df | 2293 | mapp = ↦ |
2ed88685 TT |
2294 | if (map.m_flags & EXT4_MAP_NEW) { |
2295 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
2296 | int i; | |
64769240 | 2297 | |
2ed88685 TT |
2298 | for (i = 0; i < map.m_len; i++) |
2299 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
2300 | } | |
64769240 | 2301 | |
2fa3cdfb TT |
2302 | if (ext4_should_order_data(mpd->inode)) { |
2303 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
2304 | if (err) | |
5a87b7a5 TT |
2305 | /* This only happens if the journal is aborted */ |
2306 | return; | |
2fa3cdfb TT |
2307 | } |
2308 | ||
2309 | /* | |
03f5d8bc | 2310 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
2311 | */ |
2312 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
2313 | if (disksize > i_size_read(mpd->inode)) | |
2314 | disksize = i_size_read(mpd->inode); | |
2315 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
2316 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
2317 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
2318 | if (err) | |
2319 | ext4_error(mpd->inode->i_sb, | |
2320 | "Failed to mark inode %lu dirty", | |
2321 | mpd->inode->i_ino); | |
2fa3cdfb TT |
2322 | } |
2323 | ||
5a87b7a5 | 2324 | submit_io: |
1de3e3df | 2325 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 2326 | mpd->io_done = 1; |
64769240 AT |
2327 | } |
2328 | ||
bf068ee2 AK |
2329 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
2330 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
2331 | |
2332 | /* | |
2333 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
2334 | * | |
2335 | * @mpd->lbh - extent of blocks | |
2336 | * @logical - logical number of the block in the file | |
2337 | * @bh - bh of the block (used to access block's state) | |
2338 | * | |
2339 | * the function is used to collect contig. blocks in same state | |
2340 | */ | |
2341 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
2342 | sector_t logical, size_t b_size, |
2343 | unsigned long b_state) | |
64769240 | 2344 | { |
64769240 | 2345 | sector_t next; |
8dc207c0 | 2346 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 2347 | |
c445e3e0 ES |
2348 | /* |
2349 | * XXX Don't go larger than mballoc is willing to allocate | |
2350 | * This is a stopgap solution. We eventually need to fold | |
2351 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 2352 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 ES |
2353 | */ |
2354 | if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize) | |
2355 | goto flush_it; | |
2356 | ||
525f4ed8 | 2357 | /* check if thereserved journal credits might overflow */ |
12e9b892 | 2358 | if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) { |
525f4ed8 MC |
2359 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
2360 | /* | |
2361 | * With non-extent format we are limited by the journal | |
2362 | * credit available. Total credit needed to insert | |
2363 | * nrblocks contiguous blocks is dependent on the | |
2364 | * nrblocks. So limit nrblocks. | |
2365 | */ | |
2366 | goto flush_it; | |
2367 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
2368 | EXT4_MAX_TRANS_DATA) { | |
2369 | /* | |
2370 | * Adding the new buffer_head would make it cross the | |
2371 | * allowed limit for which we have journal credit | |
2372 | * reserved. So limit the new bh->b_size | |
2373 | */ | |
2374 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
2375 | mpd->inode->i_blkbits; | |
2376 | /* we will do mpage_da_submit_io in the next loop */ | |
2377 | } | |
2378 | } | |
64769240 AT |
2379 | /* |
2380 | * First block in the extent | |
2381 | */ | |
8dc207c0 TT |
2382 | if (mpd->b_size == 0) { |
2383 | mpd->b_blocknr = logical; | |
2384 | mpd->b_size = b_size; | |
2385 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
2386 | return; |
2387 | } | |
2388 | ||
8dc207c0 | 2389 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
2390 | /* |
2391 | * Can we merge the block to our big extent? | |
2392 | */ | |
8dc207c0 TT |
2393 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
2394 | mpd->b_size += b_size; | |
64769240 AT |
2395 | return; |
2396 | } | |
2397 | ||
525f4ed8 | 2398 | flush_it: |
64769240 AT |
2399 | /* |
2400 | * We couldn't merge the block to our extent, so we | |
2401 | * need to flush current extent and start new one | |
2402 | */ | |
5a87b7a5 | 2403 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 2404 | return; |
64769240 AT |
2405 | } |
2406 | ||
c364b22c | 2407 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 2408 | { |
c364b22c | 2409 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
2410 | } |
2411 | ||
64769240 AT |
2412 | /* |
2413 | * __mpage_da_writepage - finds extent of pages and blocks | |
2414 | * | |
2415 | * @page: page to consider | |
2416 | * @wbc: not used, we just follow rules | |
2417 | * @data: context | |
2418 | * | |
2419 | * The function finds extents of pages and scan them for all blocks. | |
2420 | */ | |
2421 | static int __mpage_da_writepage(struct page *page, | |
2422 | struct writeback_control *wbc, void *data) | |
2423 | { | |
2424 | struct mpage_da_data *mpd = data; | |
2425 | struct inode *inode = mpd->inode; | |
8dc207c0 | 2426 | struct buffer_head *bh, *head; |
64769240 AT |
2427 | sector_t logical; |
2428 | ||
2429 | /* | |
2430 | * Can we merge this page to current extent? | |
2431 | */ | |
2432 | if (mpd->next_page != page->index) { | |
2433 | /* | |
2434 | * Nope, we can't. So, we map non-allocated blocks | |
5a87b7a5 | 2435 | * and start IO on them |
64769240 AT |
2436 | */ |
2437 | if (mpd->next_page != mpd->first_page) { | |
5a87b7a5 | 2438 | mpage_da_map_and_submit(mpd); |
a1d6cc56 AK |
2439 | /* |
2440 | * skip rest of the page in the page_vec | |
2441 | */ | |
a1d6cc56 AK |
2442 | redirty_page_for_writepage(wbc, page); |
2443 | unlock_page(page); | |
2444 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2445 | } |
2446 | ||
2447 | /* | |
2448 | * Start next extent of pages ... | |
2449 | */ | |
2450 | mpd->first_page = page->index; | |
2451 | ||
2452 | /* | |
2453 | * ... and blocks | |
2454 | */ | |
8dc207c0 TT |
2455 | mpd->b_size = 0; |
2456 | mpd->b_state = 0; | |
2457 | mpd->b_blocknr = 0; | |
64769240 AT |
2458 | } |
2459 | ||
2460 | mpd->next_page = page->index + 1; | |
2461 | logical = (sector_t) page->index << | |
2462 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2463 | ||
2464 | if (!page_has_buffers(page)) { | |
8dc207c0 TT |
2465 | mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE, |
2466 | (1 << BH_Dirty) | (1 << BH_Uptodate)); | |
a1d6cc56 AK |
2467 | if (mpd->io_done) |
2468 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2469 | } else { |
2470 | /* | |
2471 | * Page with regular buffer heads, just add all dirty ones | |
2472 | */ | |
2473 | head = page_buffers(page); | |
2474 | bh = head; | |
2475 | do { | |
2476 | BUG_ON(buffer_locked(bh)); | |
791b7f08 AK |
2477 | /* |
2478 | * We need to try to allocate | |
2479 | * unmapped blocks in the same page. | |
2480 | * Otherwise we won't make progress | |
43ce1d23 | 2481 | * with the page in ext4_writepage |
791b7f08 | 2482 | */ |
c364b22c | 2483 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { |
8dc207c0 TT |
2484 | mpage_add_bh_to_extent(mpd, logical, |
2485 | bh->b_size, | |
2486 | bh->b_state); | |
a1d6cc56 AK |
2487 | if (mpd->io_done) |
2488 | return MPAGE_DA_EXTENT_TAIL; | |
791b7f08 AK |
2489 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2490 | /* | |
2491 | * mapped dirty buffer. We need to update | |
2492 | * the b_state because we look at | |
2493 | * b_state in mpage_da_map_blocks. We don't | |
2494 | * update b_size because if we find an | |
2495 | * unmapped buffer_head later we need to | |
2496 | * use the b_state flag of that buffer_head. | |
2497 | */ | |
8dc207c0 TT |
2498 | if (mpd->b_size == 0) |
2499 | mpd->b_state = bh->b_state & BH_FLAGS; | |
a1d6cc56 | 2500 | } |
64769240 AT |
2501 | logical++; |
2502 | } while ((bh = bh->b_this_page) != head); | |
2503 | } | |
2504 | ||
2505 | return 0; | |
2506 | } | |
2507 | ||
64769240 | 2508 | /* |
b920c755 TT |
2509 | * This is a special get_blocks_t callback which is used by |
2510 | * ext4_da_write_begin(). It will either return mapped block or | |
2511 | * reserve space for a single block. | |
29fa89d0 AK |
2512 | * |
2513 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2514 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2515 | * | |
2516 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2517 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2518 | * initialized properly. | |
64769240 AT |
2519 | */ |
2520 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2ed88685 | 2521 | struct buffer_head *bh, int create) |
64769240 | 2522 | { |
2ed88685 | 2523 | struct ext4_map_blocks map; |
64769240 | 2524 | int ret = 0; |
33b9817e AK |
2525 | sector_t invalid_block = ~((sector_t) 0xffff); |
2526 | ||
2527 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
2528 | invalid_block = ~0; | |
64769240 AT |
2529 | |
2530 | BUG_ON(create == 0); | |
2ed88685 TT |
2531 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
2532 | ||
2533 | map.m_lblk = iblock; | |
2534 | map.m_len = 1; | |
64769240 AT |
2535 | |
2536 | /* | |
2537 | * first, we need to know whether the block is allocated already | |
2538 | * preallocated blocks are unmapped but should treated | |
2539 | * the same as allocated blocks. | |
2540 | */ | |
2ed88685 TT |
2541 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
2542 | if (ret < 0) | |
2543 | return ret; | |
2544 | if (ret == 0) { | |
2545 | if (buffer_delay(bh)) | |
2546 | return 0; /* Not sure this could or should happen */ | |
64769240 AT |
2547 | /* |
2548 | * XXX: __block_prepare_write() unmaps passed block, | |
2549 | * is it OK? | |
2550 | */ | |
9d0be502 | 2551 | ret = ext4_da_reserve_space(inode, iblock); |
d2a17637 MC |
2552 | if (ret) |
2553 | /* not enough space to reserve */ | |
2554 | return ret; | |
2555 | ||
2ed88685 TT |
2556 | map_bh(bh, inode->i_sb, invalid_block); |
2557 | set_buffer_new(bh); | |
2558 | set_buffer_delay(bh); | |
2559 | return 0; | |
64769240 AT |
2560 | } |
2561 | ||
2ed88685 TT |
2562 | map_bh(bh, inode->i_sb, map.m_pblk); |
2563 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
2564 | ||
2565 | if (buffer_unwritten(bh)) { | |
2566 | /* A delayed write to unwritten bh should be marked | |
2567 | * new and mapped. Mapped ensures that we don't do | |
2568 | * get_block multiple times when we write to the same | |
2569 | * offset and new ensures that we do proper zero out | |
2570 | * for partial write. | |
2571 | */ | |
2572 | set_buffer_new(bh); | |
2573 | set_buffer_mapped(bh); | |
2574 | } | |
2575 | return 0; | |
64769240 | 2576 | } |
61628a3f | 2577 | |
b920c755 TT |
2578 | /* |
2579 | * This function is used as a standard get_block_t calback function | |
2580 | * when there is no desire to allocate any blocks. It is used as a | |
206f7ab4 CH |
2581 | * callback function for block_prepare_write() and block_write_full_page(). |
2582 | * These functions should only try to map a single block at a time. | |
b920c755 TT |
2583 | * |
2584 | * Since this function doesn't do block allocations even if the caller | |
2585 | * requests it by passing in create=1, it is critically important that | |
2586 | * any caller checks to make sure that any buffer heads are returned | |
2587 | * by this function are either all already mapped or marked for | |
206f7ab4 CH |
2588 | * delayed allocation before calling block_write_full_page(). Otherwise, |
2589 | * b_blocknr could be left unitialized, and the page write functions will | |
2590 | * be taken by surprise. | |
b920c755 TT |
2591 | */ |
2592 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
2593 | struct buffer_head *bh_result, int create) |
2594 | { | |
a2dc52b5 | 2595 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2ed88685 | 2596 | return _ext4_get_block(inode, iblock, bh_result, 0); |
61628a3f MC |
2597 | } |
2598 | ||
62e086be AK |
2599 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
2600 | { | |
2601 | get_bh(bh); | |
2602 | return 0; | |
2603 | } | |
2604 | ||
2605 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2606 | { | |
2607 | put_bh(bh); | |
2608 | return 0; | |
2609 | } | |
2610 | ||
2611 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
2612 | unsigned int len) |
2613 | { | |
2614 | struct address_space *mapping = page->mapping; | |
2615 | struct inode *inode = mapping->host; | |
2616 | struct buffer_head *page_bufs; | |
2617 | handle_t *handle = NULL; | |
2618 | int ret = 0; | |
2619 | int err; | |
2620 | ||
cb20d518 | 2621 | ClearPageChecked(page); |
62e086be AK |
2622 | page_bufs = page_buffers(page); |
2623 | BUG_ON(!page_bufs); | |
2624 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one); | |
2625 | /* As soon as we unlock the page, it can go away, but we have | |
2626 | * references to buffers so we are safe */ | |
2627 | unlock_page(page); | |
2628 | ||
2629 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
2630 | if (IS_ERR(handle)) { | |
2631 | ret = PTR_ERR(handle); | |
2632 | goto out; | |
2633 | } | |
2634 | ||
2635 | ret = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2636 | do_journal_get_write_access); | |
2637 | ||
2638 | err = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2639 | write_end_fn); | |
2640 | if (ret == 0) | |
2641 | ret = err; | |
2642 | err = ext4_journal_stop(handle); | |
2643 | if (!ret) | |
2644 | ret = err; | |
2645 | ||
2646 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one); | |
19f5fb7a | 2647 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be AK |
2648 | out: |
2649 | return ret; | |
2650 | } | |
2651 | ||
744692dc JZ |
2652 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); |
2653 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
2654 | ||
61628a3f | 2655 | /* |
43ce1d23 AK |
2656 | * Note that we don't need to start a transaction unless we're journaling data |
2657 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2658 | * need to file the inode to the transaction's list in ordered mode because if | |
2659 | * we are writing back data added by write(), the inode is already there and if | |
2660 | * we are writing back data modified via mmap(), noone guarantees in which | |
2661 | * transaction the data will hit the disk. In case we are journaling data, we | |
2662 | * cannot start transaction directly because transaction start ranks above page | |
2663 | * lock so we have to do some magic. | |
2664 | * | |
b920c755 TT |
2665 | * This function can get called via... |
2666 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2667 | * - journal_submit_inode_data_buffers (no journal handle) | |
2668 | * - shrink_page_list via pdflush (no journal handle) | |
2669 | * - grab_page_cache when doing write_begin (have journal handle) | |
43ce1d23 AK |
2670 | * |
2671 | * We don't do any block allocation in this function. If we have page with | |
2672 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2673 | * is important for mmaped based write. So if we do with blocksize 1K | |
2674 | * truncate(f, 1024); | |
2675 | * a = mmap(f, 0, 4096); | |
2676 | * a[0] = 'a'; | |
2677 | * truncate(f, 4096); | |
2678 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
2679 | * but other bufer_heads would be unmapped but dirty(dirty done via the | |
2680 | * do_wp_page). So writepage should write the first block. If we modify | |
2681 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2682 | * page_mkwrite callback will do the block allocation and mark the | |
2683 | * buffer_heads mapped. | |
2684 | * | |
2685 | * We redirty the page if we have any buffer_heads that is either delay or | |
2686 | * unwritten in the page. | |
2687 | * | |
2688 | * We can get recursively called as show below. | |
2689 | * | |
2690 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2691 | * ext4_writepage() | |
2692 | * | |
2693 | * But since we don't do any block allocation we should not deadlock. | |
2694 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2695 | */ |
43ce1d23 | 2696 | static int ext4_writepage(struct page *page, |
62e086be | 2697 | struct writeback_control *wbc) |
64769240 | 2698 | { |
a42afc5f | 2699 | int ret = 0, commit_write = 0; |
61628a3f | 2700 | loff_t size; |
498e5f24 | 2701 | unsigned int len; |
744692dc | 2702 | struct buffer_head *page_bufs = NULL; |
61628a3f MC |
2703 | struct inode *inode = page->mapping->host; |
2704 | ||
43ce1d23 | 2705 | trace_ext4_writepage(inode, page); |
f0e6c985 AK |
2706 | size = i_size_read(inode); |
2707 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2708 | len = size & ~PAGE_CACHE_MASK; | |
2709 | else | |
2710 | len = PAGE_CACHE_SIZE; | |
64769240 | 2711 | |
a42afc5f TT |
2712 | /* |
2713 | * If the page does not have buffers (for whatever reason), | |
2714 | * try to create them using block_prepare_write. If this | |
2715 | * fails, redirty the page and move on. | |
2716 | */ | |
2717 | if (!page_buffers(page)) { | |
2718 | if (block_prepare_write(page, 0, len, | |
2719 | noalloc_get_block_write)) { | |
2720 | redirty_page: | |
f0e6c985 AK |
2721 | redirty_page_for_writepage(wbc, page); |
2722 | unlock_page(page); | |
2723 | return 0; | |
2724 | } | |
a42afc5f TT |
2725 | commit_write = 1; |
2726 | } | |
2727 | page_bufs = page_buffers(page); | |
2728 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
2729 | ext4_bh_delay_or_unwritten)) { | |
f0e6c985 | 2730 | /* |
a42afc5f TT |
2731 | * We don't want to do block allocation So redirty the |
2732 | * page and return We may reach here when we do a | |
2733 | * journal commit via | |
2734 | * journal_submit_inode_data_buffers. If we don't | |
2735 | * have mapping block we just ignore them. We can also | |
2736 | * reach here via shrink_page_list | |
f0e6c985 | 2737 | */ |
a42afc5f TT |
2738 | goto redirty_page; |
2739 | } | |
2740 | if (commit_write) | |
ed9b3e33 | 2741 | /* now mark the buffer_heads as dirty and uptodate */ |
b767e78a | 2742 | block_commit_write(page, 0, len); |
64769240 | 2743 | |
cb20d518 | 2744 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2745 | /* |
2746 | * It's mmapped pagecache. Add buffers and journal it. There | |
2747 | * doesn't seem much point in redirtying the page here. | |
2748 | */ | |
3f0ca309 | 2749 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2750 | |
a42afc5f | 2751 | if (buffer_uninit(page_bufs)) { |
744692dc JZ |
2752 | ext4_set_bh_endio(page_bufs, inode); |
2753 | ret = block_write_full_page_endio(page, noalloc_get_block_write, | |
2754 | wbc, ext4_end_io_buffer_write); | |
2755 | } else | |
b920c755 TT |
2756 | ret = block_write_full_page(page, noalloc_get_block_write, |
2757 | wbc); | |
64769240 | 2758 | |
64769240 AT |
2759 | return ret; |
2760 | } | |
2761 | ||
61628a3f | 2762 | /* |
525f4ed8 MC |
2763 | * This is called via ext4_da_writepages() to |
2764 | * calulate the total number of credits to reserve to fit | |
2765 | * a single extent allocation into a single transaction, | |
2766 | * ext4_da_writpeages() will loop calling this before | |
2767 | * the block allocation. | |
61628a3f | 2768 | */ |
525f4ed8 MC |
2769 | |
2770 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2771 | { | |
2772 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2773 | ||
2774 | /* | |
2775 | * With non-extent format the journal credit needed to | |
2776 | * insert nrblocks contiguous block is dependent on | |
2777 | * number of contiguous block. So we will limit | |
2778 | * number of contiguous block to a sane value | |
2779 | */ | |
12e9b892 | 2780 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
2781 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2782 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2783 | ||
2784 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2785 | } | |
61628a3f | 2786 | |
8e48dcfb TT |
2787 | /* |
2788 | * write_cache_pages_da - walk the list of dirty pages of the given | |
2789 | * address space and call the callback function (which usually writes | |
2790 | * the pages). | |
2791 | * | |
2792 | * This is a forked version of write_cache_pages(). Differences: | |
2793 | * Range cyclic is ignored. | |
2794 | * no_nrwrite_index_update is always presumed true | |
2795 | */ | |
2796 | static int write_cache_pages_da(struct address_space *mapping, | |
2797 | struct writeback_control *wbc, | |
2798 | struct mpage_da_data *mpd) | |
2799 | { | |
2800 | int ret = 0; | |
2801 | int done = 0; | |
2802 | struct pagevec pvec; | |
2803 | int nr_pages; | |
2804 | pgoff_t index; | |
2805 | pgoff_t end; /* Inclusive */ | |
2806 | long nr_to_write = wbc->nr_to_write; | |
2807 | ||
2808 | pagevec_init(&pvec, 0); | |
2809 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2810 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2811 | ||
2812 | while (!done && (index <= end)) { | |
2813 | int i; | |
2814 | ||
2815 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
2816 | PAGECACHE_TAG_DIRTY, | |
2817 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | |
2818 | if (nr_pages == 0) | |
2819 | break; | |
2820 | ||
2821 | for (i = 0; i < nr_pages; i++) { | |
2822 | struct page *page = pvec.pages[i]; | |
2823 | ||
2824 | /* | |
2825 | * At this point, the page may be truncated or | |
2826 | * invalidated (changing page->mapping to NULL), or | |
2827 | * even swizzled back from swapper_space to tmpfs file | |
2828 | * mapping. However, page->index will not change | |
2829 | * because we have a reference on the page. | |
2830 | */ | |
2831 | if (page->index > end) { | |
2832 | done = 1; | |
2833 | break; | |
2834 | } | |
2835 | ||
2836 | lock_page(page); | |
2837 | ||
2838 | /* | |
2839 | * Page truncated or invalidated. We can freely skip it | |
2840 | * then, even for data integrity operations: the page | |
2841 | * has disappeared concurrently, so there could be no | |
2842 | * real expectation of this data interity operation | |
2843 | * even if there is now a new, dirty page at the same | |
2844 | * pagecache address. | |
2845 | */ | |
2846 | if (unlikely(page->mapping != mapping)) { | |
2847 | continue_unlock: | |
2848 | unlock_page(page); | |
2849 | continue; | |
2850 | } | |
2851 | ||
2852 | if (!PageDirty(page)) { | |
2853 | /* someone wrote it for us */ | |
2854 | goto continue_unlock; | |
2855 | } | |
2856 | ||
2857 | if (PageWriteback(page)) { | |
2858 | if (wbc->sync_mode != WB_SYNC_NONE) | |
2859 | wait_on_page_writeback(page); | |
2860 | else | |
2861 | goto continue_unlock; | |
2862 | } | |
2863 | ||
2864 | BUG_ON(PageWriteback(page)); | |
2865 | if (!clear_page_dirty_for_io(page)) | |
2866 | goto continue_unlock; | |
2867 | ||
2868 | ret = __mpage_da_writepage(page, wbc, mpd); | |
2869 | if (unlikely(ret)) { | |
2870 | if (ret == AOP_WRITEPAGE_ACTIVATE) { | |
2871 | unlock_page(page); | |
2872 | ret = 0; | |
2873 | } else { | |
2874 | done = 1; | |
2875 | break; | |
2876 | } | |
2877 | } | |
2878 | ||
2879 | if (nr_to_write > 0) { | |
2880 | nr_to_write--; | |
2881 | if (nr_to_write == 0 && | |
2882 | wbc->sync_mode == WB_SYNC_NONE) { | |
2883 | /* | |
2884 | * We stop writing back only if we are | |
2885 | * not doing integrity sync. In case of | |
2886 | * integrity sync we have to keep going | |
2887 | * because someone may be concurrently | |
2888 | * dirtying pages, and we might have | |
2889 | * synced a lot of newly appeared dirty | |
2890 | * pages, but have not synced all of the | |
2891 | * old dirty pages. | |
2892 | */ | |
2893 | done = 1; | |
2894 | break; | |
2895 | } | |
2896 | } | |
2897 | } | |
2898 | pagevec_release(&pvec); | |
2899 | cond_resched(); | |
2900 | } | |
2901 | return ret; | |
2902 | } | |
2903 | ||
2904 | ||
64769240 | 2905 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2906 | struct writeback_control *wbc) |
64769240 | 2907 | { |
22208ded AK |
2908 | pgoff_t index; |
2909 | int range_whole = 0; | |
61628a3f | 2910 | handle_t *handle = NULL; |
df22291f | 2911 | struct mpage_da_data mpd; |
5e745b04 | 2912 | struct inode *inode = mapping->host; |
498e5f24 TT |
2913 | int pages_written = 0; |
2914 | long pages_skipped; | |
55138e0b | 2915 | unsigned int max_pages; |
2acf2c26 | 2916 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2917 | int needed_blocks, ret = 0; |
2918 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2919 | loff_t range_start = wbc->range_start; |
5e745b04 | 2920 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
61628a3f | 2921 | |
9bffad1e | 2922 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2923 | |
61628a3f MC |
2924 | /* |
2925 | * No pages to write? This is mainly a kludge to avoid starting | |
2926 | * a transaction for special inodes like journal inode on last iput() | |
2927 | * because that could violate lock ordering on umount | |
2928 | */ | |
a1d6cc56 | 2929 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2930 | return 0; |
2a21e37e TT |
2931 | |
2932 | /* | |
2933 | * If the filesystem has aborted, it is read-only, so return | |
2934 | * right away instead of dumping stack traces later on that | |
2935 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2936 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2937 | * the latter could be true if the filesystem is mounted |
2938 | * read-only, and in that case, ext4_da_writepages should | |
2939 | * *never* be called, so if that ever happens, we would want | |
2940 | * the stack trace. | |
2941 | */ | |
4ab2f15b | 2942 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2943 | return -EROFS; |
2944 | ||
22208ded AK |
2945 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2946 | range_whole = 1; | |
61628a3f | 2947 | |
2acf2c26 AK |
2948 | range_cyclic = wbc->range_cyclic; |
2949 | if (wbc->range_cyclic) { | |
22208ded | 2950 | index = mapping->writeback_index; |
2acf2c26 AK |
2951 | if (index) |
2952 | cycled = 0; | |
2953 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2954 | wbc->range_end = LLONG_MAX; | |
2955 | wbc->range_cyclic = 0; | |
2956 | } else | |
22208ded | 2957 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
a1d6cc56 | 2958 | |
55138e0b TT |
2959 | /* |
2960 | * This works around two forms of stupidity. The first is in | |
2961 | * the writeback code, which caps the maximum number of pages | |
2962 | * written to be 1024 pages. This is wrong on multiple | |
2963 | * levels; different architectues have a different page size, | |
2964 | * which changes the maximum amount of data which gets | |
2965 | * written. Secondly, 4 megabytes is way too small. XFS | |
2966 | * forces this value to be 16 megabytes by multiplying | |
2967 | * nr_to_write parameter by four, and then relies on its | |
2968 | * allocator to allocate larger extents to make them | |
2969 | * contiguous. Unfortunately this brings us to the second | |
2970 | * stupidity, which is that ext4's mballoc code only allocates | |
2971 | * at most 2048 blocks. So we force contiguous writes up to | |
2972 | * the number of dirty blocks in the inode, or | |
2973 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2974 | */ | |
2975 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2976 | if (!range_cyclic && range_whole) { |
2977 | if (wbc->nr_to_write == LONG_MAX) | |
2978 | desired_nr_to_write = wbc->nr_to_write; | |
2979 | else | |
2980 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2981 | } else | |
55138e0b TT |
2982 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2983 | max_pages); | |
2984 | if (desired_nr_to_write > max_pages) | |
2985 | desired_nr_to_write = max_pages; | |
2986 | ||
2987 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2988 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2989 | wbc->nr_to_write = desired_nr_to_write; | |
2990 | } | |
2991 | ||
df22291f AK |
2992 | mpd.wbc = wbc; |
2993 | mpd.inode = mapping->host; | |
2994 | ||
22208ded AK |
2995 | pages_skipped = wbc->pages_skipped; |
2996 | ||
2acf2c26 | 2997 | retry: |
22208ded | 2998 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2999 | |
3000 | /* | |
3001 | * we insert one extent at a time. So we need | |
3002 | * credit needed for single extent allocation. | |
3003 | * journalled mode is currently not supported | |
3004 | * by delalloc | |
3005 | */ | |
3006 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 3007 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 3008 | |
61628a3f MC |
3009 | /* start a new transaction*/ |
3010 | handle = ext4_journal_start(inode, needed_blocks); | |
3011 | if (IS_ERR(handle)) { | |
3012 | ret = PTR_ERR(handle); | |
1693918e | 3013 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 3014 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 3015 | wbc->nr_to_write, inode->i_ino, ret); |
61628a3f MC |
3016 | goto out_writepages; |
3017 | } | |
f63e6005 TT |
3018 | |
3019 | /* | |
3020 | * Now call __mpage_da_writepage to find the next | |
3021 | * contiguous region of logical blocks that need | |
3022 | * blocks to be allocated by ext4. We don't actually | |
3023 | * submit the blocks for I/O here, even though | |
3024 | * write_cache_pages thinks it will, and will set the | |
3025 | * pages as clean for write before calling | |
3026 | * __mpage_da_writepage(). | |
3027 | */ | |
3028 | mpd.b_size = 0; | |
3029 | mpd.b_state = 0; | |
3030 | mpd.b_blocknr = 0; | |
3031 | mpd.first_page = 0; | |
3032 | mpd.next_page = 0; | |
3033 | mpd.io_done = 0; | |
3034 | mpd.pages_written = 0; | |
3035 | mpd.retval = 0; | |
8e48dcfb | 3036 | ret = write_cache_pages_da(mapping, wbc, &mpd); |
f63e6005 | 3037 | /* |
af901ca1 | 3038 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
3039 | * haven't done the I/O yet, map the blocks and submit |
3040 | * them for I/O. | |
3041 | */ | |
3042 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 3043 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
3044 | ret = MPAGE_DA_EXTENT_TAIL; |
3045 | } | |
b3a3ca8c | 3046 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 3047 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 3048 | |
61628a3f | 3049 | ext4_journal_stop(handle); |
df22291f | 3050 | |
8f64b32e | 3051 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
3052 | /* commit the transaction which would |
3053 | * free blocks released in the transaction | |
3054 | * and try again | |
3055 | */ | |
df22291f | 3056 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
3057 | wbc->pages_skipped = pages_skipped; |
3058 | ret = 0; | |
3059 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 AK |
3060 | /* |
3061 | * got one extent now try with | |
3062 | * rest of the pages | |
3063 | */ | |
22208ded AK |
3064 | pages_written += mpd.pages_written; |
3065 | wbc->pages_skipped = pages_skipped; | |
a1d6cc56 | 3066 | ret = 0; |
2acf2c26 | 3067 | io_done = 1; |
22208ded | 3068 | } else if (wbc->nr_to_write) |
61628a3f MC |
3069 | /* |
3070 | * There is no more writeout needed | |
3071 | * or we requested for a noblocking writeout | |
3072 | * and we found the device congested | |
3073 | */ | |
61628a3f | 3074 | break; |
a1d6cc56 | 3075 | } |
2acf2c26 AK |
3076 | if (!io_done && !cycled) { |
3077 | cycled = 1; | |
3078 | index = 0; | |
3079 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
3080 | wbc->range_end = mapping->writeback_index - 1; | |
3081 | goto retry; | |
3082 | } | |
22208ded | 3083 | if (pages_skipped != wbc->pages_skipped) |
1693918e TT |
3084 | ext4_msg(inode->i_sb, KERN_CRIT, |
3085 | "This should not happen leaving %s " | |
fbe845dd | 3086 | "with nr_to_write = %ld ret = %d", |
1693918e | 3087 | __func__, wbc->nr_to_write, ret); |
22208ded AK |
3088 | |
3089 | /* Update index */ | |
3090 | index += pages_written; | |
2acf2c26 | 3091 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
3092 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
3093 | /* | |
3094 | * set the writeback_index so that range_cyclic | |
3095 | * mode will write it back later | |
3096 | */ | |
3097 | mapping->writeback_index = index; | |
a1d6cc56 | 3098 | |
61628a3f | 3099 | out_writepages: |
2faf2e19 | 3100 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 3101 | wbc->range_start = range_start; |
9bffad1e | 3102 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 3103 | return ret; |
64769240 AT |
3104 | } |
3105 | ||
79f0be8d AK |
3106 | #define FALL_BACK_TO_NONDELALLOC 1 |
3107 | static int ext4_nonda_switch(struct super_block *sb) | |
3108 | { | |
3109 | s64 free_blocks, dirty_blocks; | |
3110 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
3111 | ||
3112 | /* | |
3113 | * switch to non delalloc mode if we are running low | |
3114 | * on free block. The free block accounting via percpu | |
179f7ebf | 3115 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
3116 | * accumulated on each CPU without updating global counters |
3117 | * Delalloc need an accurate free block accounting. So switch | |
3118 | * to non delalloc when we are near to error range. | |
3119 | */ | |
3120 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3121 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter); | |
3122 | if (2 * free_blocks < 3 * dirty_blocks || | |
3123 | free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) { | |
3124 | /* | |
c8afb446 ES |
3125 | * free block count is less than 150% of dirty blocks |
3126 | * or free blocks is less than watermark | |
79f0be8d AK |
3127 | */ |
3128 | return 1; | |
3129 | } | |
c8afb446 ES |
3130 | /* |
3131 | * Even if we don't switch but are nearing capacity, | |
3132 | * start pushing delalloc when 1/2 of free blocks are dirty. | |
3133 | */ | |
3134 | if (free_blocks < 2 * dirty_blocks) | |
3135 | writeback_inodes_sb_if_idle(sb); | |
3136 | ||
79f0be8d AK |
3137 | return 0; |
3138 | } | |
3139 | ||
64769240 | 3140 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
3141 | loff_t pos, unsigned len, unsigned flags, |
3142 | struct page **pagep, void **fsdata) | |
64769240 | 3143 | { |
72b8ab9d | 3144 | int ret, retries = 0; |
64769240 AT |
3145 | struct page *page; |
3146 | pgoff_t index; | |
64769240 AT |
3147 | struct inode *inode = mapping->host; |
3148 | handle_t *handle; | |
3149 | ||
3150 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
3151 | |
3152 | if (ext4_nonda_switch(inode->i_sb)) { | |
3153 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
3154 | return ext4_write_begin(file, mapping, pos, | |
3155 | len, flags, pagep, fsdata); | |
3156 | } | |
3157 | *fsdata = (void *)0; | |
9bffad1e | 3158 | trace_ext4_da_write_begin(inode, pos, len, flags); |
d2a17637 | 3159 | retry: |
64769240 AT |
3160 | /* |
3161 | * With delayed allocation, we don't log the i_disksize update | |
3162 | * if there is delayed block allocation. But we still need | |
3163 | * to journalling the i_disksize update if writes to the end | |
3164 | * of file which has an already mapped buffer. | |
3165 | */ | |
3166 | handle = ext4_journal_start(inode, 1); | |
3167 | if (IS_ERR(handle)) { | |
3168 | ret = PTR_ERR(handle); | |
3169 | goto out; | |
3170 | } | |
ebd3610b JK |
3171 | /* We cannot recurse into the filesystem as the transaction is already |
3172 | * started */ | |
3173 | flags |= AOP_FLAG_NOFS; | |
64769240 | 3174 | |
54566b2c | 3175 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
3176 | if (!page) { |
3177 | ext4_journal_stop(handle); | |
3178 | ret = -ENOMEM; | |
3179 | goto out; | |
3180 | } | |
64769240 AT |
3181 | *pagep = page; |
3182 | ||
6e1db88d | 3183 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
3184 | if (ret < 0) { |
3185 | unlock_page(page); | |
3186 | ext4_journal_stop(handle); | |
3187 | page_cache_release(page); | |
ae4d5372 AK |
3188 | /* |
3189 | * block_write_begin may have instantiated a few blocks | |
3190 | * outside i_size. Trim these off again. Don't need | |
3191 | * i_size_read because we hold i_mutex. | |
3192 | */ | |
3193 | if (pos + len > inode->i_size) | |
b9a4207d | 3194 | ext4_truncate_failed_write(inode); |
64769240 AT |
3195 | } |
3196 | ||
d2a17637 MC |
3197 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
3198 | goto retry; | |
64769240 AT |
3199 | out: |
3200 | return ret; | |
3201 | } | |
3202 | ||
632eaeab MC |
3203 | /* |
3204 | * Check if we should update i_disksize | |
3205 | * when write to the end of file but not require block allocation | |
3206 | */ | |
3207 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3208 | unsigned long offset) |
632eaeab MC |
3209 | { |
3210 | struct buffer_head *bh; | |
3211 | struct inode *inode = page->mapping->host; | |
3212 | unsigned int idx; | |
3213 | int i; | |
3214 | ||
3215 | bh = page_buffers(page); | |
3216 | idx = offset >> inode->i_blkbits; | |
3217 | ||
af5bc92d | 3218 | for (i = 0; i < idx; i++) |
632eaeab MC |
3219 | bh = bh->b_this_page; |
3220 | ||
29fa89d0 | 3221 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3222 | return 0; |
3223 | return 1; | |
3224 | } | |
3225 | ||
64769240 | 3226 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3227 | struct address_space *mapping, |
3228 | loff_t pos, unsigned len, unsigned copied, | |
3229 | struct page *page, void *fsdata) | |
64769240 AT |
3230 | { |
3231 | struct inode *inode = mapping->host; | |
3232 | int ret = 0, ret2; | |
3233 | handle_t *handle = ext4_journal_current_handle(); | |
3234 | loff_t new_i_size; | |
632eaeab | 3235 | unsigned long start, end; |
79f0be8d AK |
3236 | int write_mode = (int)(unsigned long)fsdata; |
3237 | ||
3238 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
3239 | if (ext4_should_order_data(inode)) { | |
3240 | return ext4_ordered_write_end(file, mapping, pos, | |
3241 | len, copied, page, fsdata); | |
3242 | } else if (ext4_should_writeback_data(inode)) { | |
3243 | return ext4_writeback_write_end(file, mapping, pos, | |
3244 | len, copied, page, fsdata); | |
3245 | } else { | |
3246 | BUG(); | |
3247 | } | |
3248 | } | |
632eaeab | 3249 | |
9bffad1e | 3250 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 3251 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 3252 | end = start + copied - 1; |
64769240 AT |
3253 | |
3254 | /* | |
3255 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3256 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3257 | * into that. | |
3258 | */ | |
3259 | ||
3260 | new_i_size = pos + copied; | |
632eaeab MC |
3261 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
3262 | if (ext4_da_should_update_i_disksize(page, end)) { | |
3263 | down_write(&EXT4_I(inode)->i_data_sem); | |
3264 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
3265 | /* | |
3266 | * Updating i_disksize when extending file | |
3267 | * without needing block allocation | |
3268 | */ | |
3269 | if (ext4_should_order_data(inode)) | |
3270 | ret = ext4_jbd2_file_inode(handle, | |
3271 | inode); | |
64769240 | 3272 | |
632eaeab MC |
3273 | EXT4_I(inode)->i_disksize = new_i_size; |
3274 | } | |
3275 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
3276 | /* We need to mark inode dirty even if |
3277 | * new_i_size is less that inode->i_size | |
3278 | * bu greater than i_disksize.(hint delalloc) | |
3279 | */ | |
3280 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3281 | } |
632eaeab | 3282 | } |
64769240 AT |
3283 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
3284 | page, fsdata); | |
3285 | copied = ret2; | |
3286 | if (ret2 < 0) | |
3287 | ret = ret2; | |
3288 | ret2 = ext4_journal_stop(handle); | |
3289 | if (!ret) | |
3290 | ret = ret2; | |
3291 | ||
3292 | return ret ? ret : copied; | |
3293 | } | |
3294 | ||
3295 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
3296 | { | |
64769240 AT |
3297 | /* |
3298 | * Drop reserved blocks | |
3299 | */ | |
3300 | BUG_ON(!PageLocked(page)); | |
3301 | if (!page_has_buffers(page)) | |
3302 | goto out; | |
3303 | ||
d2a17637 | 3304 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
3305 | |
3306 | out: | |
3307 | ext4_invalidatepage(page, offset); | |
3308 | ||
3309 | return; | |
3310 | } | |
3311 | ||
ccd2506b TT |
3312 | /* |
3313 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3314 | */ | |
3315 | int ext4_alloc_da_blocks(struct inode *inode) | |
3316 | { | |
fb40ba0d TT |
3317 | trace_ext4_alloc_da_blocks(inode); |
3318 | ||
ccd2506b TT |
3319 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
3320 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
3321 | return 0; | |
3322 | ||
3323 | /* | |
3324 | * We do something simple for now. The filemap_flush() will | |
3325 | * also start triggering a write of the data blocks, which is | |
3326 | * not strictly speaking necessary (and for users of | |
3327 | * laptop_mode, not even desirable). However, to do otherwise | |
3328 | * would require replicating code paths in: | |
de9a55b8 | 3329 | * |
ccd2506b TT |
3330 | * ext4_da_writepages() -> |
3331 | * write_cache_pages() ---> (via passed in callback function) | |
3332 | * __mpage_da_writepage() --> | |
3333 | * mpage_add_bh_to_extent() | |
3334 | * mpage_da_map_blocks() | |
3335 | * | |
3336 | * The problem is that write_cache_pages(), located in | |
3337 | * mm/page-writeback.c, marks pages clean in preparation for | |
3338 | * doing I/O, which is not desirable if we're not planning on | |
3339 | * doing I/O at all. | |
3340 | * | |
3341 | * We could call write_cache_pages(), and then redirty all of | |
3342 | * the pages by calling redirty_page_for_writeback() but that | |
3343 | * would be ugly in the extreme. So instead we would need to | |
3344 | * replicate parts of the code in the above functions, | |
3345 | * simplifying them becuase we wouldn't actually intend to | |
3346 | * write out the pages, but rather only collect contiguous | |
3347 | * logical block extents, call the multi-block allocator, and | |
3348 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3349 | * |
ccd2506b TT |
3350 | * For now, though, we'll cheat by calling filemap_flush(), |
3351 | * which will map the blocks, and start the I/O, but not | |
3352 | * actually wait for the I/O to complete. | |
3353 | */ | |
3354 | return filemap_flush(inode->i_mapping); | |
3355 | } | |
64769240 | 3356 | |
ac27a0ec DK |
3357 | /* |
3358 | * bmap() is special. It gets used by applications such as lilo and by | |
3359 | * the swapper to find the on-disk block of a specific piece of data. | |
3360 | * | |
3361 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3362 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3363 | * filesystem and enables swap, then they may get a nasty shock when the |
3364 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3365 | * the original zero's written out previously to the journal and | |
3366 | * awaiting writeback in the kernel's buffer cache. | |
3367 | * | |
3368 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3369 | * take extra steps to flush any blocks which might be in the cache. | |
3370 | */ | |
617ba13b | 3371 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3372 | { |
3373 | struct inode *inode = mapping->host; | |
3374 | journal_t *journal; | |
3375 | int err; | |
3376 | ||
64769240 AT |
3377 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3378 | test_opt(inode->i_sb, DELALLOC)) { | |
3379 | /* | |
3380 | * With delalloc we want to sync the file | |
3381 | * so that we can make sure we allocate | |
3382 | * blocks for file | |
3383 | */ | |
3384 | filemap_write_and_wait(mapping); | |
3385 | } | |
3386 | ||
19f5fb7a TT |
3387 | if (EXT4_JOURNAL(inode) && |
3388 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3389 | /* |
3390 | * This is a REALLY heavyweight approach, but the use of | |
3391 | * bmap on dirty files is expected to be extremely rare: | |
3392 | * only if we run lilo or swapon on a freshly made file | |
3393 | * do we expect this to happen. | |
3394 | * | |
3395 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3396 | * represent an unprivileged user DOS attack --- we'd be | |
3397 | * in trouble if mortal users could trigger this path at | |
3398 | * will.) | |
3399 | * | |
617ba13b | 3400 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3401 | * regular files. If somebody wants to bmap a directory |
3402 | * or symlink and gets confused because the buffer | |
3403 | * hasn't yet been flushed to disk, they deserve | |
3404 | * everything they get. | |
3405 | */ | |
3406 | ||
19f5fb7a | 3407 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3408 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3409 | jbd2_journal_lock_updates(journal); |
3410 | err = jbd2_journal_flush(journal); | |
3411 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3412 | |
3413 | if (err) | |
3414 | return 0; | |
3415 | } | |
3416 | ||
af5bc92d | 3417 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3418 | } |
3419 | ||
617ba13b | 3420 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3421 | { |
617ba13b | 3422 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
3423 | } |
3424 | ||
3425 | static int | |
617ba13b | 3426 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3427 | struct list_head *pages, unsigned nr_pages) |
3428 | { | |
617ba13b | 3429 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
3430 | } |
3431 | ||
744692dc JZ |
3432 | static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset) |
3433 | { | |
3434 | struct buffer_head *head, *bh; | |
3435 | unsigned int curr_off = 0; | |
3436 | ||
3437 | if (!page_has_buffers(page)) | |
3438 | return; | |
3439 | head = bh = page_buffers(page); | |
3440 | do { | |
3441 | if (offset <= curr_off && test_clear_buffer_uninit(bh) | |
3442 | && bh->b_private) { | |
3443 | ext4_free_io_end(bh->b_private); | |
3444 | bh->b_private = NULL; | |
3445 | bh->b_end_io = NULL; | |
3446 | } | |
3447 | curr_off = curr_off + bh->b_size; | |
3448 | bh = bh->b_this_page; | |
3449 | } while (bh != head); | |
3450 | } | |
3451 | ||
617ba13b | 3452 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 3453 | { |
617ba13b | 3454 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3455 | |
744692dc JZ |
3456 | /* |
3457 | * free any io_end structure allocated for buffers to be discarded | |
3458 | */ | |
3459 | if (ext4_should_dioread_nolock(page->mapping->host)) | |
3460 | ext4_invalidatepage_free_endio(page, offset); | |
ac27a0ec DK |
3461 | /* |
3462 | * If it's a full truncate we just forget about the pending dirtying | |
3463 | */ | |
3464 | if (offset == 0) | |
3465 | ClearPageChecked(page); | |
3466 | ||
0390131b FM |
3467 | if (journal) |
3468 | jbd2_journal_invalidatepage(journal, page, offset); | |
3469 | else | |
3470 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
3471 | } |
3472 | ||
617ba13b | 3473 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3474 | { |
617ba13b | 3475 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
3476 | |
3477 | WARN_ON(PageChecked(page)); | |
3478 | if (!page_has_buffers(page)) | |
3479 | return 0; | |
0390131b FM |
3480 | if (journal) |
3481 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3482 | else | |
3483 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3484 | } |
3485 | ||
3486 | /* | |
4c0425ff MC |
3487 | * O_DIRECT for ext3 (or indirect map) based files |
3488 | * | |
ac27a0ec DK |
3489 | * If the O_DIRECT write will extend the file then add this inode to the |
3490 | * orphan list. So recovery will truncate it back to the original size | |
3491 | * if the machine crashes during the write. | |
3492 | * | |
3493 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
7fb5409d JK |
3494 | * crashes then stale disk data _may_ be exposed inside the file. But current |
3495 | * VFS code falls back into buffered path in that case so we are safe. | |
ac27a0ec | 3496 | */ |
4c0425ff | 3497 | static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb, |
de9a55b8 TT |
3498 | const struct iovec *iov, loff_t offset, |
3499 | unsigned long nr_segs) | |
ac27a0ec DK |
3500 | { |
3501 | struct file *file = iocb->ki_filp; | |
3502 | struct inode *inode = file->f_mapping->host; | |
617ba13b | 3503 | struct ext4_inode_info *ei = EXT4_I(inode); |
7fb5409d | 3504 | handle_t *handle; |
ac27a0ec DK |
3505 | ssize_t ret; |
3506 | int orphan = 0; | |
3507 | size_t count = iov_length(iov, nr_segs); | |
fbbf6945 | 3508 | int retries = 0; |
ac27a0ec DK |
3509 | |
3510 | if (rw == WRITE) { | |
3511 | loff_t final_size = offset + count; | |
3512 | ||
ac27a0ec | 3513 | if (final_size > inode->i_size) { |
7fb5409d JK |
3514 | /* Credits for sb + inode write */ |
3515 | handle = ext4_journal_start(inode, 2); | |
3516 | if (IS_ERR(handle)) { | |
3517 | ret = PTR_ERR(handle); | |
3518 | goto out; | |
3519 | } | |
617ba13b | 3520 | ret = ext4_orphan_add(handle, inode); |
7fb5409d JK |
3521 | if (ret) { |
3522 | ext4_journal_stop(handle); | |
3523 | goto out; | |
3524 | } | |
ac27a0ec DK |
3525 | orphan = 1; |
3526 | ei->i_disksize = inode->i_size; | |
7fb5409d | 3527 | ext4_journal_stop(handle); |
ac27a0ec DK |
3528 | } |
3529 | } | |
3530 | ||
fbbf6945 | 3531 | retry: |
b7adc1f3 | 3532 | if (rw == READ && ext4_should_dioread_nolock(inode)) |
eafdc7d1 | 3533 | ret = __blockdev_direct_IO(rw, iocb, inode, |
b7adc1f3 JZ |
3534 | inode->i_sb->s_bdev, iov, |
3535 | offset, nr_segs, | |
eafdc7d1 CH |
3536 | ext4_get_block, NULL, NULL, 0); |
3537 | else { | |
b7adc1f3 JZ |
3538 | ret = blockdev_direct_IO(rw, iocb, inode, |
3539 | inode->i_sb->s_bdev, iov, | |
ac27a0ec | 3540 | offset, nr_segs, |
617ba13b | 3541 | ext4_get_block, NULL); |
eafdc7d1 CH |
3542 | |
3543 | if (unlikely((rw & WRITE) && ret < 0)) { | |
3544 | loff_t isize = i_size_read(inode); | |
3545 | loff_t end = offset + iov_length(iov, nr_segs); | |
3546 | ||
3547 | if (end > isize) | |
3548 | vmtruncate(inode, isize); | |
3549 | } | |
3550 | } | |
fbbf6945 ES |
3551 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
3552 | goto retry; | |
ac27a0ec | 3553 | |
7fb5409d | 3554 | if (orphan) { |
ac27a0ec DK |
3555 | int err; |
3556 | ||
7fb5409d JK |
3557 | /* Credits for sb + inode write */ |
3558 | handle = ext4_journal_start(inode, 2); | |
3559 | if (IS_ERR(handle)) { | |
3560 | /* This is really bad luck. We've written the data | |
3561 | * but cannot extend i_size. Bail out and pretend | |
3562 | * the write failed... */ | |
3563 | ret = PTR_ERR(handle); | |
da1dafca DM |
3564 | if (inode->i_nlink) |
3565 | ext4_orphan_del(NULL, inode); | |
3566 | ||
7fb5409d JK |
3567 | goto out; |
3568 | } | |
3569 | if (inode->i_nlink) | |
617ba13b | 3570 | ext4_orphan_del(handle, inode); |
7fb5409d | 3571 | if (ret > 0) { |
ac27a0ec DK |
3572 | loff_t end = offset + ret; |
3573 | if (end > inode->i_size) { | |
3574 | ei->i_disksize = end; | |
3575 | i_size_write(inode, end); | |
3576 | /* | |
3577 | * We're going to return a positive `ret' | |
3578 | * here due to non-zero-length I/O, so there's | |
3579 | * no way of reporting error returns from | |
617ba13b | 3580 | * ext4_mark_inode_dirty() to userspace. So |
ac27a0ec DK |
3581 | * ignore it. |
3582 | */ | |
617ba13b | 3583 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
3584 | } |
3585 | } | |
617ba13b | 3586 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
3587 | if (ret == 0) |
3588 | ret = err; | |
3589 | } | |
3590 | out: | |
3591 | return ret; | |
3592 | } | |
3593 | ||
2ed88685 TT |
3594 | /* |
3595 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3596 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3597 | * The extent will be converted to initialized after the IO is complete. | |
3598 | */ | |
c7064ef1 | 3599 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3600 | struct buffer_head *bh_result, int create) |
3601 | { | |
c7064ef1 | 3602 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3603 | inode->i_ino, create); |
2ed88685 TT |
3604 | return _ext4_get_block(inode, iblock, bh_result, |
3605 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3606 | } |
3607 | ||
c7064ef1 | 3608 | static void dump_completed_IO(struct inode * inode) |
8d5d02e6 MC |
3609 | { |
3610 | #ifdef EXT4_DEBUG | |
3611 | struct list_head *cur, *before, *after; | |
3612 | ext4_io_end_t *io, *io0, *io1; | |
744692dc | 3613 | unsigned long flags; |
8d5d02e6 | 3614 | |
c7064ef1 JZ |
3615 | if (list_empty(&EXT4_I(inode)->i_completed_io_list)){ |
3616 | ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino); | |
8d5d02e6 MC |
3617 | return; |
3618 | } | |
3619 | ||
c7064ef1 | 3620 | ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino); |
744692dc | 3621 | spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); |
c7064ef1 | 3622 | list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){ |
8d5d02e6 MC |
3623 | cur = &io->list; |
3624 | before = cur->prev; | |
3625 | io0 = container_of(before, ext4_io_end_t, list); | |
3626 | after = cur->next; | |
3627 | io1 = container_of(after, ext4_io_end_t, list); | |
3628 | ||
3629 | ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", | |
3630 | io, inode->i_ino, io0, io1); | |
3631 | } | |
744692dc | 3632 | spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); |
8d5d02e6 MC |
3633 | #endif |
3634 | } | |
4c0425ff | 3635 | |
8d5d02e6 MC |
3636 | /* |
3637 | * This function is called from ext4_sync_file(). | |
3638 | * | |
c7064ef1 JZ |
3639 | * When IO is completed, the work to convert unwritten extents to |
3640 | * written is queued on workqueue but may not get immediately | |
8d5d02e6 MC |
3641 | * scheduled. When fsync is called, we need to ensure the |
3642 | * conversion is complete before fsync returns. | |
c7064ef1 JZ |
3643 | * The inode keeps track of a list of pending/completed IO that |
3644 | * might needs to do the conversion. This function walks through | |
3645 | * the list and convert the related unwritten extents for completed IO | |
3646 | * to written. | |
3647 | * The function return the number of pending IOs on success. | |
8d5d02e6 | 3648 | */ |
c7064ef1 | 3649 | int flush_completed_IO(struct inode *inode) |
8d5d02e6 MC |
3650 | { |
3651 | ext4_io_end_t *io; | |
744692dc JZ |
3652 | struct ext4_inode_info *ei = EXT4_I(inode); |
3653 | unsigned long flags; | |
8d5d02e6 MC |
3654 | int ret = 0; |
3655 | int ret2 = 0; | |
3656 | ||
744692dc | 3657 | if (list_empty(&ei->i_completed_io_list)) |
8d5d02e6 MC |
3658 | return ret; |
3659 | ||
c7064ef1 | 3660 | dump_completed_IO(inode); |
744692dc JZ |
3661 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
3662 | while (!list_empty(&ei->i_completed_io_list)){ | |
3663 | io = list_entry(ei->i_completed_io_list.next, | |
8d5d02e6 MC |
3664 | ext4_io_end_t, list); |
3665 | /* | |
c7064ef1 | 3666 | * Calling ext4_end_io_nolock() to convert completed |
8d5d02e6 MC |
3667 | * IO to written. |
3668 | * | |
3669 | * When ext4_sync_file() is called, run_queue() may already | |
3670 | * about to flush the work corresponding to this io structure. | |
3671 | * It will be upset if it founds the io structure related | |
3672 | * to the work-to-be schedule is freed. | |
3673 | * | |
3674 | * Thus we need to keep the io structure still valid here after | |
3675 | * convertion finished. The io structure has a flag to | |
3676 | * avoid double converting from both fsync and background work | |
3677 | * queue work. | |
3678 | */ | |
744692dc | 3679 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
c7064ef1 | 3680 | ret = ext4_end_io_nolock(io); |
744692dc | 3681 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
8d5d02e6 MC |
3682 | if (ret < 0) |
3683 | ret2 = ret; | |
3684 | else | |
3685 | list_del_init(&io->list); | |
3686 | } | |
744692dc | 3687 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
8d5d02e6 MC |
3688 | return (ret2 < 0) ? ret2 : 0; |
3689 | } | |
3690 | ||
4c0425ff | 3691 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
3692 | ssize_t size, void *private, int ret, |
3693 | bool is_async) | |
4c0425ff MC |
3694 | { |
3695 | ext4_io_end_t *io_end = iocb->private; | |
3696 | struct workqueue_struct *wq; | |
744692dc JZ |
3697 | unsigned long flags; |
3698 | struct ext4_inode_info *ei; | |
4c0425ff | 3699 | |
4b70df18 M |
3700 | /* if not async direct IO or dio with 0 bytes write, just return */ |
3701 | if (!io_end || !size) | |
552ef802 | 3702 | goto out; |
4b70df18 | 3703 | |
8d5d02e6 MC |
3704 | ext_debug("ext4_end_io_dio(): io_end 0x%p" |
3705 | "for inode %lu, iocb 0x%p, offset %llu, size %llu\n", | |
3706 | iocb->private, io_end->inode->i_ino, iocb, offset, | |
3707 | size); | |
8d5d02e6 MC |
3708 | |
3709 | /* if not aio dio with unwritten extents, just free io and return */ | |
bd2d0210 | 3710 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
8d5d02e6 MC |
3711 | ext4_free_io_end(io_end); |
3712 | iocb->private = NULL; | |
5b3ff237 JZ |
3713 | out: |
3714 | if (is_async) | |
3715 | aio_complete(iocb, ret, 0); | |
3716 | return; | |
8d5d02e6 MC |
3717 | } |
3718 | ||
4c0425ff MC |
3719 | io_end->offset = offset; |
3720 | io_end->size = size; | |
5b3ff237 JZ |
3721 | if (is_async) { |
3722 | io_end->iocb = iocb; | |
3723 | io_end->result = ret; | |
3724 | } | |
4c0425ff MC |
3725 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; |
3726 | ||
8d5d02e6 | 3727 | /* Add the io_end to per-inode completed aio dio list*/ |
744692dc JZ |
3728 | ei = EXT4_I(io_end->inode); |
3729 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); | |
3730 | list_add_tail(&io_end->list, &ei->i_completed_io_list); | |
3731 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
c999af2b ES |
3732 | |
3733 | /* queue the work to convert unwritten extents to written */ | |
3734 | queue_work(wq, &io_end->work); | |
4c0425ff MC |
3735 | iocb->private = NULL; |
3736 | } | |
c7064ef1 | 3737 | |
744692dc JZ |
3738 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate) |
3739 | { | |
3740 | ext4_io_end_t *io_end = bh->b_private; | |
3741 | struct workqueue_struct *wq; | |
3742 | struct inode *inode; | |
3743 | unsigned long flags; | |
3744 | ||
3745 | if (!test_clear_buffer_uninit(bh) || !io_end) | |
3746 | goto out; | |
3747 | ||
3748 | if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) { | |
3749 | printk("sb umounted, discard end_io request for inode %lu\n", | |
3750 | io_end->inode->i_ino); | |
3751 | ext4_free_io_end(io_end); | |
3752 | goto out; | |
3753 | } | |
3754 | ||
bd2d0210 | 3755 | io_end->flag = EXT4_IO_END_UNWRITTEN; |
744692dc JZ |
3756 | inode = io_end->inode; |
3757 | ||
3758 | /* Add the io_end to per-inode completed io list*/ | |
3759 | spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); | |
3760 | list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list); | |
3761 | spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); | |
3762 | ||
3763 | wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq; | |
3764 | /* queue the work to convert unwritten extents to written */ | |
3765 | queue_work(wq, &io_end->work); | |
3766 | out: | |
3767 | bh->b_private = NULL; | |
3768 | bh->b_end_io = NULL; | |
3769 | clear_buffer_uninit(bh); | |
3770 | end_buffer_async_write(bh, uptodate); | |
3771 | } | |
3772 | ||
3773 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode) | |
3774 | { | |
3775 | ext4_io_end_t *io_end; | |
3776 | struct page *page = bh->b_page; | |
3777 | loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT; | |
3778 | size_t size = bh->b_size; | |
3779 | ||
3780 | retry: | |
3781 | io_end = ext4_init_io_end(inode, GFP_ATOMIC); | |
3782 | if (!io_end) { | |
3783 | if (printk_ratelimit()) | |
3784 | printk(KERN_WARNING "%s: allocation fail\n", __func__); | |
3785 | schedule(); | |
3786 | goto retry; | |
3787 | } | |
3788 | io_end->offset = offset; | |
3789 | io_end->size = size; | |
3790 | /* | |
3791 | * We need to hold a reference to the page to make sure it | |
3792 | * doesn't get evicted before ext4_end_io_work() has a chance | |
3793 | * to convert the extent from written to unwritten. | |
3794 | */ | |
3795 | io_end->page = page; | |
3796 | get_page(io_end->page); | |
3797 | ||
3798 | bh->b_private = io_end; | |
3799 | bh->b_end_io = ext4_end_io_buffer_write; | |
3800 | return 0; | |
3801 | } | |
3802 | ||
4c0425ff MC |
3803 | /* |
3804 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3805 | * preallocated extents, and those write extend the file, no need to | |
3806 | * fall back to buffered IO. | |
3807 | * | |
3808 | * For holes, we fallocate those blocks, mark them as unintialized | |
3809 | * If those blocks were preallocated, we mark sure they are splited, but | |
3810 | * still keep the range to write as unintialized. | |
3811 | * | |
8d5d02e6 MC |
3812 | * The unwrritten extents will be converted to written when DIO is completed. |
3813 | * For async direct IO, since the IO may still pending when return, we | |
3814 | * set up an end_io call back function, which will do the convertion | |
3815 | * when async direct IO completed. | |
4c0425ff MC |
3816 | * |
3817 | * If the O_DIRECT write will extend the file then add this inode to the | |
3818 | * orphan list. So recovery will truncate it back to the original size | |
3819 | * if the machine crashes during the write. | |
3820 | * | |
3821 | */ | |
3822 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
3823 | const struct iovec *iov, loff_t offset, | |
3824 | unsigned long nr_segs) | |
3825 | { | |
3826 | struct file *file = iocb->ki_filp; | |
3827 | struct inode *inode = file->f_mapping->host; | |
3828 | ssize_t ret; | |
3829 | size_t count = iov_length(iov, nr_segs); | |
3830 | ||
3831 | loff_t final_size = offset + count; | |
3832 | if (rw == WRITE && final_size <= inode->i_size) { | |
3833 | /* | |
8d5d02e6 MC |
3834 | * We could direct write to holes and fallocate. |
3835 | * | |
3836 | * Allocated blocks to fill the hole are marked as uninitialized | |
4c0425ff MC |
3837 | * to prevent paralel buffered read to expose the stale data |
3838 | * before DIO complete the data IO. | |
8d5d02e6 MC |
3839 | * |
3840 | * As to previously fallocated extents, ext4 get_block | |
4c0425ff MC |
3841 | * will just simply mark the buffer mapped but still |
3842 | * keep the extents uninitialized. | |
3843 | * | |
8d5d02e6 MC |
3844 | * for non AIO case, we will convert those unwritten extents |
3845 | * to written after return back from blockdev_direct_IO. | |
3846 | * | |
3847 | * for async DIO, the conversion needs to be defered when | |
3848 | * the IO is completed. The ext4 end_io callback function | |
3849 | * will be called to take care of the conversion work. | |
3850 | * Here for async case, we allocate an io_end structure to | |
3851 | * hook to the iocb. | |
4c0425ff | 3852 | */ |
8d5d02e6 MC |
3853 | iocb->private = NULL; |
3854 | EXT4_I(inode)->cur_aio_dio = NULL; | |
3855 | if (!is_sync_kiocb(iocb)) { | |
744692dc | 3856 | iocb->private = ext4_init_io_end(inode, GFP_NOFS); |
8d5d02e6 MC |
3857 | if (!iocb->private) |
3858 | return -ENOMEM; | |
3859 | /* | |
3860 | * we save the io structure for current async | |
79e83036 | 3861 | * direct IO, so that later ext4_map_blocks() |
8d5d02e6 MC |
3862 | * could flag the io structure whether there |
3863 | * is a unwritten extents needs to be converted | |
3864 | * when IO is completed. | |
3865 | */ | |
3866 | EXT4_I(inode)->cur_aio_dio = iocb->private; | |
3867 | } | |
3868 | ||
4c0425ff MC |
3869 | ret = blockdev_direct_IO(rw, iocb, inode, |
3870 | inode->i_sb->s_bdev, iov, | |
3871 | offset, nr_segs, | |
c7064ef1 | 3872 | ext4_get_block_write, |
4c0425ff | 3873 | ext4_end_io_dio); |
8d5d02e6 MC |
3874 | if (iocb->private) |
3875 | EXT4_I(inode)->cur_aio_dio = NULL; | |
3876 | /* | |
3877 | * The io_end structure takes a reference to the inode, | |
3878 | * that structure needs to be destroyed and the | |
3879 | * reference to the inode need to be dropped, when IO is | |
3880 | * complete, even with 0 byte write, or failed. | |
3881 | * | |
3882 | * In the successful AIO DIO case, the io_end structure will be | |
3883 | * desctroyed and the reference to the inode will be dropped | |
3884 | * after the end_io call back function is called. | |
3885 | * | |
3886 | * In the case there is 0 byte write, or error case, since | |
3887 | * VFS direct IO won't invoke the end_io call back function, | |
3888 | * we need to free the end_io structure here. | |
3889 | */ | |
3890 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { | |
3891 | ext4_free_io_end(iocb->private); | |
3892 | iocb->private = NULL; | |
19f5fb7a TT |
3893 | } else if (ret > 0 && ext4_test_inode_state(inode, |
3894 | EXT4_STATE_DIO_UNWRITTEN)) { | |
109f5565 | 3895 | int err; |
8d5d02e6 MC |
3896 | /* |
3897 | * for non AIO case, since the IO is already | |
3898 | * completed, we could do the convertion right here | |
3899 | */ | |
109f5565 M |
3900 | err = ext4_convert_unwritten_extents(inode, |
3901 | offset, ret); | |
3902 | if (err < 0) | |
3903 | ret = err; | |
19f5fb7a | 3904 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
109f5565 | 3905 | } |
4c0425ff MC |
3906 | return ret; |
3907 | } | |
8d5d02e6 MC |
3908 | |
3909 | /* for write the the end of file case, we fall back to old way */ | |
4c0425ff MC |
3910 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
3911 | } | |
3912 | ||
3913 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
3914 | const struct iovec *iov, loff_t offset, | |
3915 | unsigned long nr_segs) | |
3916 | { | |
3917 | struct file *file = iocb->ki_filp; | |
3918 | struct inode *inode = file->f_mapping->host; | |
3919 | ||
12e9b892 | 3920 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4c0425ff MC |
3921 | return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3922 | ||
3923 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3924 | } | |
3925 | ||
ac27a0ec | 3926 | /* |
617ba13b | 3927 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3928 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3929 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3930 | * not necessarily locked. | |
3931 | * | |
3932 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3933 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3934 | * or jbddirty because all the journalling code will explode. | |
3935 | * | |
3936 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3937 | * is called, propagate that into the buffers appropriately. | |
3938 | */ | |
617ba13b | 3939 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3940 | { |
3941 | SetPageChecked(page); | |
3942 | return __set_page_dirty_nobuffers(page); | |
3943 | } | |
3944 | ||
617ba13b | 3945 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
3946 | .readpage = ext4_readpage, |
3947 | .readpages = ext4_readpages, | |
43ce1d23 | 3948 | .writepage = ext4_writepage, |
8ab22b9a HH |
3949 | .sync_page = block_sync_page, |
3950 | .write_begin = ext4_write_begin, | |
3951 | .write_end = ext4_ordered_write_end, | |
3952 | .bmap = ext4_bmap, | |
3953 | .invalidatepage = ext4_invalidatepage, | |
3954 | .releasepage = ext4_releasepage, | |
3955 | .direct_IO = ext4_direct_IO, | |
3956 | .migratepage = buffer_migrate_page, | |
3957 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3958 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3959 | }; |
3960 | ||
617ba13b | 3961 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
3962 | .readpage = ext4_readpage, |
3963 | .readpages = ext4_readpages, | |
43ce1d23 | 3964 | .writepage = ext4_writepage, |
8ab22b9a HH |
3965 | .sync_page = block_sync_page, |
3966 | .write_begin = ext4_write_begin, | |
3967 | .write_end = ext4_writeback_write_end, | |
3968 | .bmap = ext4_bmap, | |
3969 | .invalidatepage = ext4_invalidatepage, | |
3970 | .releasepage = ext4_releasepage, | |
3971 | .direct_IO = ext4_direct_IO, | |
3972 | .migratepage = buffer_migrate_page, | |
3973 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3974 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3975 | }; |
3976 | ||
617ba13b | 3977 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3978 | .readpage = ext4_readpage, |
3979 | .readpages = ext4_readpages, | |
43ce1d23 | 3980 | .writepage = ext4_writepage, |
8ab22b9a HH |
3981 | .sync_page = block_sync_page, |
3982 | .write_begin = ext4_write_begin, | |
3983 | .write_end = ext4_journalled_write_end, | |
3984 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3985 | .bmap = ext4_bmap, | |
3986 | .invalidatepage = ext4_invalidatepage, | |
3987 | .releasepage = ext4_releasepage, | |
3988 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3989 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3990 | }; |
3991 | ||
64769240 | 3992 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3993 | .readpage = ext4_readpage, |
3994 | .readpages = ext4_readpages, | |
43ce1d23 | 3995 | .writepage = ext4_writepage, |
8ab22b9a HH |
3996 | .writepages = ext4_da_writepages, |
3997 | .sync_page = block_sync_page, | |
3998 | .write_begin = ext4_da_write_begin, | |
3999 | .write_end = ext4_da_write_end, | |
4000 | .bmap = ext4_bmap, | |
4001 | .invalidatepage = ext4_da_invalidatepage, | |
4002 | .releasepage = ext4_releasepage, | |
4003 | .direct_IO = ext4_direct_IO, | |
4004 | .migratepage = buffer_migrate_page, | |
4005 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 4006 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
4007 | }; |
4008 | ||
617ba13b | 4009 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 4010 | { |
cd1aac32 AK |
4011 | if (ext4_should_order_data(inode) && |
4012 | test_opt(inode->i_sb, DELALLOC)) | |
4013 | inode->i_mapping->a_ops = &ext4_da_aops; | |
4014 | else if (ext4_should_order_data(inode)) | |
617ba13b | 4015 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
64769240 AT |
4016 | else if (ext4_should_writeback_data(inode) && |
4017 | test_opt(inode->i_sb, DELALLOC)) | |
4018 | inode->i_mapping->a_ops = &ext4_da_aops; | |
617ba13b MC |
4019 | else if (ext4_should_writeback_data(inode)) |
4020 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 4021 | else |
617ba13b | 4022 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
4023 | } |
4024 | ||
4025 | /* | |
617ba13b | 4026 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
4027 | * up to the end of the block which corresponds to `from'. |
4028 | * This required during truncate. We need to physically zero the tail end | |
4029 | * of that block so it doesn't yield old data if the file is later grown. | |
4030 | */ | |
cf108bca | 4031 | int ext4_block_truncate_page(handle_t *handle, |
ac27a0ec DK |
4032 | struct address_space *mapping, loff_t from) |
4033 | { | |
617ba13b | 4034 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 4035 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
725d26d3 AK |
4036 | unsigned blocksize, length, pos; |
4037 | ext4_lblk_t iblock; | |
ac27a0ec DK |
4038 | struct inode *inode = mapping->host; |
4039 | struct buffer_head *bh; | |
cf108bca | 4040 | struct page *page; |
ac27a0ec | 4041 | int err = 0; |
ac27a0ec | 4042 | |
f4a01017 TT |
4043 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, |
4044 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
cf108bca JK |
4045 | if (!page) |
4046 | return -EINVAL; | |
4047 | ||
ac27a0ec DK |
4048 | blocksize = inode->i_sb->s_blocksize; |
4049 | length = blocksize - (offset & (blocksize - 1)); | |
4050 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
4051 | ||
ac27a0ec DK |
4052 | if (!page_has_buffers(page)) |
4053 | create_empty_buffers(page, blocksize, 0); | |
4054 | ||
4055 | /* Find the buffer that contains "offset" */ | |
4056 | bh = page_buffers(page); | |
4057 | pos = blocksize; | |
4058 | while (offset >= pos) { | |
4059 | bh = bh->b_this_page; | |
4060 | iblock++; | |
4061 | pos += blocksize; | |
4062 | } | |
4063 | ||
4064 | err = 0; | |
4065 | if (buffer_freed(bh)) { | |
4066 | BUFFER_TRACE(bh, "freed: skip"); | |
4067 | goto unlock; | |
4068 | } | |
4069 | ||
4070 | if (!buffer_mapped(bh)) { | |
4071 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 4072 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
4073 | /* unmapped? It's a hole - nothing to do */ |
4074 | if (!buffer_mapped(bh)) { | |
4075 | BUFFER_TRACE(bh, "still unmapped"); | |
4076 | goto unlock; | |
4077 | } | |
4078 | } | |
4079 | ||
4080 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
4081 | if (PageUptodate(page)) | |
4082 | set_buffer_uptodate(bh); | |
4083 | ||
4084 | if (!buffer_uptodate(bh)) { | |
4085 | err = -EIO; | |
4086 | ll_rw_block(READ, 1, &bh); | |
4087 | wait_on_buffer(bh); | |
4088 | /* Uhhuh. Read error. Complain and punt. */ | |
4089 | if (!buffer_uptodate(bh)) | |
4090 | goto unlock; | |
4091 | } | |
4092 | ||
617ba13b | 4093 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 4094 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 4095 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
4096 | if (err) |
4097 | goto unlock; | |
4098 | } | |
4099 | ||
eebd2aa3 | 4100 | zero_user(page, offset, length); |
ac27a0ec DK |
4101 | |
4102 | BUFFER_TRACE(bh, "zeroed end of block"); | |
4103 | ||
4104 | err = 0; | |
617ba13b | 4105 | if (ext4_should_journal_data(inode)) { |
0390131b | 4106 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
ac27a0ec | 4107 | } else { |
617ba13b | 4108 | if (ext4_should_order_data(inode)) |
678aaf48 | 4109 | err = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
4110 | mark_buffer_dirty(bh); |
4111 | } | |
4112 | ||
4113 | unlock: | |
4114 | unlock_page(page); | |
4115 | page_cache_release(page); | |
4116 | return err; | |
4117 | } | |
4118 | ||
4119 | /* | |
4120 | * Probably it should be a library function... search for first non-zero word | |
4121 | * or memcmp with zero_page, whatever is better for particular architecture. | |
4122 | * Linus? | |
4123 | */ | |
4124 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
4125 | { | |
4126 | while (p < q) | |
4127 | if (*p++) | |
4128 | return 0; | |
4129 | return 1; | |
4130 | } | |
4131 | ||
4132 | /** | |
617ba13b | 4133 | * ext4_find_shared - find the indirect blocks for partial truncation. |
ac27a0ec DK |
4134 | * @inode: inode in question |
4135 | * @depth: depth of the affected branch | |
617ba13b | 4136 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
ac27a0ec DK |
4137 | * @chain: place to store the pointers to partial indirect blocks |
4138 | * @top: place to the (detached) top of branch | |
4139 | * | |
617ba13b | 4140 | * This is a helper function used by ext4_truncate(). |
ac27a0ec DK |
4141 | * |
4142 | * When we do truncate() we may have to clean the ends of several | |
4143 | * indirect blocks but leave the blocks themselves alive. Block is | |
4144 | * partially truncated if some data below the new i_size is refered | |
4145 | * from it (and it is on the path to the first completely truncated | |
4146 | * data block, indeed). We have to free the top of that path along | |
4147 | * with everything to the right of the path. Since no allocation | |
617ba13b | 4148 | * past the truncation point is possible until ext4_truncate() |
ac27a0ec DK |
4149 | * finishes, we may safely do the latter, but top of branch may |
4150 | * require special attention - pageout below the truncation point | |
4151 | * might try to populate it. | |
4152 | * | |
4153 | * We atomically detach the top of branch from the tree, store the | |
4154 | * block number of its root in *@top, pointers to buffer_heads of | |
4155 | * partially truncated blocks - in @chain[].bh and pointers to | |
4156 | * their last elements that should not be removed - in | |
4157 | * @chain[].p. Return value is the pointer to last filled element | |
4158 | * of @chain. | |
4159 | * | |
4160 | * The work left to caller to do the actual freeing of subtrees: | |
4161 | * a) free the subtree starting from *@top | |
4162 | * b) free the subtrees whose roots are stored in | |
4163 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
4164 | * c) free the subtrees growing from the inode past the @chain[0]. | |
4165 | * (no partially truncated stuff there). */ | |
4166 | ||
617ba13b | 4167 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
de9a55b8 TT |
4168 | ext4_lblk_t offsets[4], Indirect chain[4], |
4169 | __le32 *top) | |
ac27a0ec DK |
4170 | { |
4171 | Indirect *partial, *p; | |
4172 | int k, err; | |
4173 | ||
4174 | *top = 0; | |
bf48aabb | 4175 | /* Make k index the deepest non-null offset + 1 */ |
ac27a0ec DK |
4176 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
4177 | ; | |
617ba13b | 4178 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
ac27a0ec DK |
4179 | /* Writer: pointers */ |
4180 | if (!partial) | |
4181 | partial = chain + k-1; | |
4182 | /* | |
4183 | * If the branch acquired continuation since we've looked at it - | |
4184 | * fine, it should all survive and (new) top doesn't belong to us. | |
4185 | */ | |
4186 | if (!partial->key && *partial->p) | |
4187 | /* Writer: end */ | |
4188 | goto no_top; | |
af5bc92d | 4189 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) |
ac27a0ec DK |
4190 | ; |
4191 | /* | |
4192 | * OK, we've found the last block that must survive. The rest of our | |
4193 | * branch should be detached before unlocking. However, if that rest | |
4194 | * of branch is all ours and does not grow immediately from the inode | |
4195 | * it's easier to cheat and just decrement partial->p. | |
4196 | */ | |
4197 | if (p == chain + k - 1 && p > chain) { | |
4198 | p->p--; | |
4199 | } else { | |
4200 | *top = *p->p; | |
617ba13b | 4201 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
ac27a0ec DK |
4202 | #if 0 |
4203 | *p->p = 0; | |
4204 | #endif | |
4205 | } | |
4206 | /* Writer: end */ | |
4207 | ||
af5bc92d | 4208 | while (partial > p) { |
ac27a0ec DK |
4209 | brelse(partial->bh); |
4210 | partial--; | |
4211 | } | |
4212 | no_top: | |
4213 | return partial; | |
4214 | } | |
4215 | ||
4216 | /* | |
4217 | * Zero a number of block pointers in either an inode or an indirect block. | |
4218 | * If we restart the transaction we must again get write access to the | |
4219 | * indirect block for further modification. | |
4220 | * | |
4221 | * We release `count' blocks on disk, but (last - first) may be greater | |
4222 | * than `count' because there can be holes in there. | |
4223 | */ | |
1f2acb60 TT |
4224 | static int ext4_clear_blocks(handle_t *handle, struct inode *inode, |
4225 | struct buffer_head *bh, | |
4226 | ext4_fsblk_t block_to_free, | |
4227 | unsigned long count, __le32 *first, | |
4228 | __le32 *last) | |
ac27a0ec DK |
4229 | { |
4230 | __le32 *p; | |
1f2acb60 | 4231 | int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED; |
e6362609 TT |
4232 | |
4233 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
4234 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
50689696 | 4235 | |
1f2acb60 TT |
4236 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free, |
4237 | count)) { | |
24676da4 TT |
4238 | EXT4_ERROR_INODE(inode, "attempt to clear invalid " |
4239 | "blocks %llu len %lu", | |
4240 | (unsigned long long) block_to_free, count); | |
1f2acb60 TT |
4241 | return 1; |
4242 | } | |
4243 | ||
ac27a0ec DK |
4244 | if (try_to_extend_transaction(handle, inode)) { |
4245 | if (bh) { | |
0390131b FM |
4246 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
4247 | ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec | 4248 | } |
617ba13b | 4249 | ext4_mark_inode_dirty(handle, inode); |
487caeef JK |
4250 | ext4_truncate_restart_trans(handle, inode, |
4251 | blocks_for_truncate(inode)); | |
ac27a0ec DK |
4252 | if (bh) { |
4253 | BUFFER_TRACE(bh, "retaking write access"); | |
617ba13b | 4254 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
4255 | } |
4256 | } | |
4257 | ||
e6362609 TT |
4258 | for (p = first; p < last; p++) |
4259 | *p = 0; | |
ac27a0ec | 4260 | |
e6362609 | 4261 | ext4_free_blocks(handle, inode, 0, block_to_free, count, flags); |
1f2acb60 | 4262 | return 0; |
ac27a0ec DK |
4263 | } |
4264 | ||
4265 | /** | |
617ba13b | 4266 | * ext4_free_data - free a list of data blocks |
ac27a0ec DK |
4267 | * @handle: handle for this transaction |
4268 | * @inode: inode we are dealing with | |
4269 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
4270 | * @first: array of block numbers | |
4271 | * @last: points immediately past the end of array | |
4272 | * | |
4273 | * We are freeing all blocks refered from that array (numbers are stored as | |
4274 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
4275 | * | |
4276 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
4277 | * blocks are contiguous then releasing them at one time will only affect one | |
4278 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
4279 | * actually use a lot of journal space. | |
4280 | * | |
4281 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
4282 | * block pointers. | |
4283 | */ | |
617ba13b | 4284 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
4285 | struct buffer_head *this_bh, |
4286 | __le32 *first, __le32 *last) | |
4287 | { | |
617ba13b | 4288 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ac27a0ec DK |
4289 | unsigned long count = 0; /* Number of blocks in the run */ |
4290 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
4291 | corresponding to | |
4292 | block_to_free */ | |
617ba13b | 4293 | ext4_fsblk_t nr; /* Current block # */ |
ac27a0ec DK |
4294 | __le32 *p; /* Pointer into inode/ind |
4295 | for current block */ | |
4296 | int err; | |
4297 | ||
4298 | if (this_bh) { /* For indirect block */ | |
4299 | BUFFER_TRACE(this_bh, "get_write_access"); | |
617ba13b | 4300 | err = ext4_journal_get_write_access(handle, this_bh); |
ac27a0ec DK |
4301 | /* Important: if we can't update the indirect pointers |
4302 | * to the blocks, we can't free them. */ | |
4303 | if (err) | |
4304 | return; | |
4305 | } | |
4306 | ||
4307 | for (p = first; p < last; p++) { | |
4308 | nr = le32_to_cpu(*p); | |
4309 | if (nr) { | |
4310 | /* accumulate blocks to free if they're contiguous */ | |
4311 | if (count == 0) { | |
4312 | block_to_free = nr; | |
4313 | block_to_free_p = p; | |
4314 | count = 1; | |
4315 | } else if (nr == block_to_free + count) { | |
4316 | count++; | |
4317 | } else { | |
1f2acb60 TT |
4318 | if (ext4_clear_blocks(handle, inode, this_bh, |
4319 | block_to_free, count, | |
4320 | block_to_free_p, p)) | |
4321 | break; | |
ac27a0ec DK |
4322 | block_to_free = nr; |
4323 | block_to_free_p = p; | |
4324 | count = 1; | |
4325 | } | |
4326 | } | |
4327 | } | |
4328 | ||
4329 | if (count > 0) | |
617ba13b | 4330 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
ac27a0ec DK |
4331 | count, block_to_free_p, p); |
4332 | ||
4333 | if (this_bh) { | |
0390131b | 4334 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); |
71dc8fbc DG |
4335 | |
4336 | /* | |
4337 | * The buffer head should have an attached journal head at this | |
4338 | * point. However, if the data is corrupted and an indirect | |
4339 | * block pointed to itself, it would have been detached when | |
4340 | * the block was cleared. Check for this instead of OOPSing. | |
4341 | */ | |
e7f07968 | 4342 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) |
0390131b | 4343 | ext4_handle_dirty_metadata(handle, inode, this_bh); |
71dc8fbc | 4344 | else |
24676da4 TT |
4345 | EXT4_ERROR_INODE(inode, |
4346 | "circular indirect block detected at " | |
4347 | "block %llu", | |
4348 | (unsigned long long) this_bh->b_blocknr); | |
ac27a0ec DK |
4349 | } |
4350 | } | |
4351 | ||
4352 | /** | |
617ba13b | 4353 | * ext4_free_branches - free an array of branches |
ac27a0ec DK |
4354 | * @handle: JBD handle for this transaction |
4355 | * @inode: inode we are dealing with | |
4356 | * @parent_bh: the buffer_head which contains *@first and *@last | |
4357 | * @first: array of block numbers | |
4358 | * @last: pointer immediately past the end of array | |
4359 | * @depth: depth of the branches to free | |
4360 | * | |
4361 | * We are freeing all blocks refered from these branches (numbers are | |
4362 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
4363 | * appropriately. | |
4364 | */ | |
617ba13b | 4365 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
4366 | struct buffer_head *parent_bh, |
4367 | __le32 *first, __le32 *last, int depth) | |
4368 | { | |
617ba13b | 4369 | ext4_fsblk_t nr; |
ac27a0ec DK |
4370 | __le32 *p; |
4371 | ||
0390131b | 4372 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
4373 | return; |
4374 | ||
4375 | if (depth--) { | |
4376 | struct buffer_head *bh; | |
617ba13b | 4377 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
4378 | p = last; |
4379 | while (--p >= first) { | |
4380 | nr = le32_to_cpu(*p); | |
4381 | if (!nr) | |
4382 | continue; /* A hole */ | |
4383 | ||
1f2acb60 TT |
4384 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), |
4385 | nr, 1)) { | |
24676da4 TT |
4386 | EXT4_ERROR_INODE(inode, |
4387 | "invalid indirect mapped " | |
4388 | "block %lu (level %d)", | |
4389 | (unsigned long) nr, depth); | |
1f2acb60 TT |
4390 | break; |
4391 | } | |
4392 | ||
ac27a0ec DK |
4393 | /* Go read the buffer for the next level down */ |
4394 | bh = sb_bread(inode->i_sb, nr); | |
4395 | ||
4396 | /* | |
4397 | * A read failure? Report error and clear slot | |
4398 | * (should be rare). | |
4399 | */ | |
4400 | if (!bh) { | |
c398eda0 TT |
4401 | EXT4_ERROR_INODE_BLOCK(inode, nr, |
4402 | "Read failure"); | |
ac27a0ec DK |
4403 | continue; |
4404 | } | |
4405 | ||
4406 | /* This zaps the entire block. Bottom up. */ | |
4407 | BUFFER_TRACE(bh, "free child branches"); | |
617ba13b | 4408 | ext4_free_branches(handle, inode, bh, |
af5bc92d TT |
4409 | (__le32 *) bh->b_data, |
4410 | (__le32 *) bh->b_data + addr_per_block, | |
4411 | depth); | |
ac27a0ec | 4412 | |
ac27a0ec DK |
4413 | /* |
4414 | * Everything below this this pointer has been | |
4415 | * released. Now let this top-of-subtree go. | |
4416 | * | |
4417 | * We want the freeing of this indirect block to be | |
4418 | * atomic in the journal with the updating of the | |
4419 | * bitmap block which owns it. So make some room in | |
4420 | * the journal. | |
4421 | * | |
4422 | * We zero the parent pointer *after* freeing its | |
4423 | * pointee in the bitmaps, so if extend_transaction() | |
4424 | * for some reason fails to put the bitmap changes and | |
4425 | * the release into the same transaction, recovery | |
4426 | * will merely complain about releasing a free block, | |
4427 | * rather than leaking blocks. | |
4428 | */ | |
0390131b | 4429 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
4430 | return; |
4431 | if (try_to_extend_transaction(handle, inode)) { | |
617ba13b | 4432 | ext4_mark_inode_dirty(handle, inode); |
487caeef JK |
4433 | ext4_truncate_restart_trans(handle, inode, |
4434 | blocks_for_truncate(inode)); | |
ac27a0ec DK |
4435 | } |
4436 | ||
40389687 A |
4437 | /* |
4438 | * The forget flag here is critical because if | |
4439 | * we are journaling (and not doing data | |
4440 | * journaling), we have to make sure a revoke | |
4441 | * record is written to prevent the journal | |
4442 | * replay from overwriting the (former) | |
4443 | * indirect block if it gets reallocated as a | |
4444 | * data block. This must happen in the same | |
4445 | * transaction where the data blocks are | |
4446 | * actually freed. | |
4447 | */ | |
e6362609 | 4448 | ext4_free_blocks(handle, inode, 0, nr, 1, |
40389687 A |
4449 | EXT4_FREE_BLOCKS_METADATA| |
4450 | EXT4_FREE_BLOCKS_FORGET); | |
ac27a0ec DK |
4451 | |
4452 | if (parent_bh) { | |
4453 | /* | |
4454 | * The block which we have just freed is | |
4455 | * pointed to by an indirect block: journal it | |
4456 | */ | |
4457 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
617ba13b | 4458 | if (!ext4_journal_get_write_access(handle, |
ac27a0ec DK |
4459 | parent_bh)){ |
4460 | *p = 0; | |
4461 | BUFFER_TRACE(parent_bh, | |
0390131b FM |
4462 | "call ext4_handle_dirty_metadata"); |
4463 | ext4_handle_dirty_metadata(handle, | |
4464 | inode, | |
4465 | parent_bh); | |
ac27a0ec DK |
4466 | } |
4467 | } | |
4468 | } | |
4469 | } else { | |
4470 | /* We have reached the bottom of the tree. */ | |
4471 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
617ba13b | 4472 | ext4_free_data(handle, inode, parent_bh, first, last); |
ac27a0ec DK |
4473 | } |
4474 | } | |
4475 | ||
91ef4caf DG |
4476 | int ext4_can_truncate(struct inode *inode) |
4477 | { | |
4478 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4479 | return 0; | |
4480 | if (S_ISREG(inode->i_mode)) | |
4481 | return 1; | |
4482 | if (S_ISDIR(inode->i_mode)) | |
4483 | return 1; | |
4484 | if (S_ISLNK(inode->i_mode)) | |
4485 | return !ext4_inode_is_fast_symlink(inode); | |
4486 | return 0; | |
4487 | } | |
4488 | ||
ac27a0ec | 4489 | /* |
617ba13b | 4490 | * ext4_truncate() |
ac27a0ec | 4491 | * |
617ba13b MC |
4492 | * We block out ext4_get_block() block instantiations across the entire |
4493 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4494 | * simultaneously on behalf of the same inode. |
4495 | * | |
4496 | * As we work through the truncate and commmit bits of it to the journal there | |
4497 | * is one core, guiding principle: the file's tree must always be consistent on | |
4498 | * disk. We must be able to restart the truncate after a crash. | |
4499 | * | |
4500 | * The file's tree may be transiently inconsistent in memory (although it | |
4501 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4502 | * the contents of (the filesystem + the journal) must be consistent and | |
4503 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4504 | * left-to-right works OK too). | |
4505 | * | |
4506 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4507 | * truncate against the orphan inode list. | |
4508 | * | |
4509 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4510 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4511 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4512 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4513 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4514 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4515 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4516 | */ |
617ba13b | 4517 | void ext4_truncate(struct inode *inode) |
ac27a0ec DK |
4518 | { |
4519 | handle_t *handle; | |
617ba13b | 4520 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 4521 | __le32 *i_data = ei->i_data; |
617ba13b | 4522 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec | 4523 | struct address_space *mapping = inode->i_mapping; |
725d26d3 | 4524 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
4525 | Indirect chain[4]; |
4526 | Indirect *partial; | |
4527 | __le32 nr = 0; | |
4528 | int n; | |
725d26d3 | 4529 | ext4_lblk_t last_block; |
ac27a0ec | 4530 | unsigned blocksize = inode->i_sb->s_blocksize; |
ac27a0ec | 4531 | |
91ef4caf | 4532 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
4533 | return; |
4534 | ||
12e9b892 | 4535 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 4536 | |
5534fb5b | 4537 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4538 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4539 | |
12e9b892 | 4540 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
cf108bca | 4541 | ext4_ext_truncate(inode); |
1d03ec98 AK |
4542 | return; |
4543 | } | |
a86c6181 | 4544 | |
ac27a0ec | 4545 | handle = start_transaction(inode); |
cf108bca | 4546 | if (IS_ERR(handle)) |
ac27a0ec | 4547 | return; /* AKPM: return what? */ |
ac27a0ec DK |
4548 | |
4549 | last_block = (inode->i_size + blocksize-1) | |
617ba13b | 4550 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
ac27a0ec | 4551 | |
cf108bca JK |
4552 | if (inode->i_size & (blocksize - 1)) |
4553 | if (ext4_block_truncate_page(handle, mapping, inode->i_size)) | |
4554 | goto out_stop; | |
ac27a0ec | 4555 | |
617ba13b | 4556 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
ac27a0ec DK |
4557 | if (n == 0) |
4558 | goto out_stop; /* error */ | |
4559 | ||
4560 | /* | |
4561 | * OK. This truncate is going to happen. We add the inode to the | |
4562 | * orphan list, so that if this truncate spans multiple transactions, | |
4563 | * and we crash, we will resume the truncate when the filesystem | |
4564 | * recovers. It also marks the inode dirty, to catch the new size. | |
4565 | * | |
4566 | * Implication: the file must always be in a sane, consistent | |
4567 | * truncatable state while each transaction commits. | |
4568 | */ | |
617ba13b | 4569 | if (ext4_orphan_add(handle, inode)) |
ac27a0ec DK |
4570 | goto out_stop; |
4571 | ||
632eaeab MC |
4572 | /* |
4573 | * From here we block out all ext4_get_block() callers who want to | |
4574 | * modify the block allocation tree. | |
4575 | */ | |
4576 | down_write(&ei->i_data_sem); | |
b4df2030 | 4577 | |
c2ea3fde | 4578 | ext4_discard_preallocations(inode); |
b4df2030 | 4579 | |
ac27a0ec DK |
4580 | /* |
4581 | * The orphan list entry will now protect us from any crash which | |
4582 | * occurs before the truncate completes, so it is now safe to propagate | |
4583 | * the new, shorter inode size (held for now in i_size) into the | |
4584 | * on-disk inode. We do this via i_disksize, which is the value which | |
617ba13b | 4585 | * ext4 *really* writes onto the disk inode. |
ac27a0ec DK |
4586 | */ |
4587 | ei->i_disksize = inode->i_size; | |
4588 | ||
ac27a0ec | 4589 | if (n == 1) { /* direct blocks */ |
617ba13b MC |
4590 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
4591 | i_data + EXT4_NDIR_BLOCKS); | |
ac27a0ec DK |
4592 | goto do_indirects; |
4593 | } | |
4594 | ||
617ba13b | 4595 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
ac27a0ec DK |
4596 | /* Kill the top of shared branch (not detached) */ |
4597 | if (nr) { | |
4598 | if (partial == chain) { | |
4599 | /* Shared branch grows from the inode */ | |
617ba13b | 4600 | ext4_free_branches(handle, inode, NULL, |
ac27a0ec DK |
4601 | &nr, &nr+1, (chain+n-1) - partial); |
4602 | *partial->p = 0; | |
4603 | /* | |
4604 | * We mark the inode dirty prior to restart, | |
4605 | * and prior to stop. No need for it here. | |
4606 | */ | |
4607 | } else { | |
4608 | /* Shared branch grows from an indirect block */ | |
4609 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
617ba13b | 4610 | ext4_free_branches(handle, inode, partial->bh, |
ac27a0ec DK |
4611 | partial->p, |
4612 | partial->p+1, (chain+n-1) - partial); | |
4613 | } | |
4614 | } | |
4615 | /* Clear the ends of indirect blocks on the shared branch */ | |
4616 | while (partial > chain) { | |
617ba13b | 4617 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
ac27a0ec DK |
4618 | (__le32*)partial->bh->b_data+addr_per_block, |
4619 | (chain+n-1) - partial); | |
4620 | BUFFER_TRACE(partial->bh, "call brelse"); | |
de9a55b8 | 4621 | brelse(partial->bh); |
ac27a0ec DK |
4622 | partial--; |
4623 | } | |
4624 | do_indirects: | |
4625 | /* Kill the remaining (whole) subtrees */ | |
4626 | switch (offsets[0]) { | |
4627 | default: | |
617ba13b | 4628 | nr = i_data[EXT4_IND_BLOCK]; |
ac27a0ec | 4629 | if (nr) { |
617ba13b MC |
4630 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
4631 | i_data[EXT4_IND_BLOCK] = 0; | |
ac27a0ec | 4632 | } |
617ba13b MC |
4633 | case EXT4_IND_BLOCK: |
4634 | nr = i_data[EXT4_DIND_BLOCK]; | |
ac27a0ec | 4635 | if (nr) { |
617ba13b MC |
4636 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
4637 | i_data[EXT4_DIND_BLOCK] = 0; | |
ac27a0ec | 4638 | } |
617ba13b MC |
4639 | case EXT4_DIND_BLOCK: |
4640 | nr = i_data[EXT4_TIND_BLOCK]; | |
ac27a0ec | 4641 | if (nr) { |
617ba13b MC |
4642 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
4643 | i_data[EXT4_TIND_BLOCK] = 0; | |
ac27a0ec | 4644 | } |
617ba13b | 4645 | case EXT4_TIND_BLOCK: |
ac27a0ec DK |
4646 | ; |
4647 | } | |
4648 | ||
0e855ac8 | 4649 | up_write(&ei->i_data_sem); |
ef7f3835 | 4650 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
617ba13b | 4651 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
4652 | |
4653 | /* | |
4654 | * In a multi-transaction truncate, we only make the final transaction | |
4655 | * synchronous | |
4656 | */ | |
4657 | if (IS_SYNC(inode)) | |
0390131b | 4658 | ext4_handle_sync(handle); |
ac27a0ec DK |
4659 | out_stop: |
4660 | /* | |
4661 | * If this was a simple ftruncate(), and the file will remain alive | |
4662 | * then we need to clear up the orphan record which we created above. | |
4663 | * However, if this was a real unlink then we were called by | |
617ba13b | 4664 | * ext4_delete_inode(), and we allow that function to clean up the |
ac27a0ec DK |
4665 | * orphan info for us. |
4666 | */ | |
4667 | if (inode->i_nlink) | |
617ba13b | 4668 | ext4_orphan_del(handle, inode); |
ac27a0ec | 4669 | |
617ba13b | 4670 | ext4_journal_stop(handle); |
ac27a0ec DK |
4671 | } |
4672 | ||
ac27a0ec | 4673 | /* |
617ba13b | 4674 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4675 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4676 | * data in memory that is needed to recreate the on-disk version of this | |
4677 | * inode. | |
4678 | */ | |
617ba13b MC |
4679 | static int __ext4_get_inode_loc(struct inode *inode, |
4680 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4681 | { |
240799cd TT |
4682 | struct ext4_group_desc *gdp; |
4683 | struct buffer_head *bh; | |
4684 | struct super_block *sb = inode->i_sb; | |
4685 | ext4_fsblk_t block; | |
4686 | int inodes_per_block, inode_offset; | |
4687 | ||
3a06d778 | 4688 | iloc->bh = NULL; |
240799cd TT |
4689 | if (!ext4_valid_inum(sb, inode->i_ino)) |
4690 | return -EIO; | |
ac27a0ec | 4691 | |
240799cd TT |
4692 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4693 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4694 | if (!gdp) | |
ac27a0ec DK |
4695 | return -EIO; |
4696 | ||
240799cd TT |
4697 | /* |
4698 | * Figure out the offset within the block group inode table | |
4699 | */ | |
4700 | inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb)); | |
4701 | inode_offset = ((inode->i_ino - 1) % | |
4702 | EXT4_INODES_PER_GROUP(sb)); | |
4703 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4704 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4705 | ||
4706 | bh = sb_getblk(sb, block); | |
ac27a0ec | 4707 | if (!bh) { |
c398eda0 TT |
4708 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4709 | "unable to read itable block"); | |
ac27a0ec DK |
4710 | return -EIO; |
4711 | } | |
4712 | if (!buffer_uptodate(bh)) { | |
4713 | lock_buffer(bh); | |
9c83a923 HK |
4714 | |
4715 | /* | |
4716 | * If the buffer has the write error flag, we have failed | |
4717 | * to write out another inode in the same block. In this | |
4718 | * case, we don't have to read the block because we may | |
4719 | * read the old inode data successfully. | |
4720 | */ | |
4721 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4722 | set_buffer_uptodate(bh); | |
4723 | ||
ac27a0ec DK |
4724 | if (buffer_uptodate(bh)) { |
4725 | /* someone brought it uptodate while we waited */ | |
4726 | unlock_buffer(bh); | |
4727 | goto has_buffer; | |
4728 | } | |
4729 | ||
4730 | /* | |
4731 | * If we have all information of the inode in memory and this | |
4732 | * is the only valid inode in the block, we need not read the | |
4733 | * block. | |
4734 | */ | |
4735 | if (in_mem) { | |
4736 | struct buffer_head *bitmap_bh; | |
240799cd | 4737 | int i, start; |
ac27a0ec | 4738 | |
240799cd | 4739 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4740 | |
240799cd TT |
4741 | /* Is the inode bitmap in cache? */ |
4742 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
4743 | if (!bitmap_bh) |
4744 | goto make_io; | |
4745 | ||
4746 | /* | |
4747 | * If the inode bitmap isn't in cache then the | |
4748 | * optimisation may end up performing two reads instead | |
4749 | * of one, so skip it. | |
4750 | */ | |
4751 | if (!buffer_uptodate(bitmap_bh)) { | |
4752 | brelse(bitmap_bh); | |
4753 | goto make_io; | |
4754 | } | |
240799cd | 4755 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4756 | if (i == inode_offset) |
4757 | continue; | |
617ba13b | 4758 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4759 | break; |
4760 | } | |
4761 | brelse(bitmap_bh); | |
240799cd | 4762 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4763 | /* all other inodes are free, so skip I/O */ |
4764 | memset(bh->b_data, 0, bh->b_size); | |
4765 | set_buffer_uptodate(bh); | |
4766 | unlock_buffer(bh); | |
4767 | goto has_buffer; | |
4768 | } | |
4769 | } | |
4770 | ||
4771 | make_io: | |
240799cd TT |
4772 | /* |
4773 | * If we need to do any I/O, try to pre-readahead extra | |
4774 | * blocks from the inode table. | |
4775 | */ | |
4776 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4777 | ext4_fsblk_t b, end, table; | |
4778 | unsigned num; | |
4779 | ||
4780 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4781 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
4782 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
4783 | if (table > b) | |
4784 | b = table; | |
4785 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
4786 | num = EXT4_INODES_PER_GROUP(sb); | |
4787 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4788 | EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) | |
560671a0 | 4789 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4790 | table += num / inodes_per_block; |
4791 | if (end > table) | |
4792 | end = table; | |
4793 | while (b <= end) | |
4794 | sb_breadahead(sb, b++); | |
4795 | } | |
4796 | ||
ac27a0ec DK |
4797 | /* |
4798 | * There are other valid inodes in the buffer, this inode | |
4799 | * has in-inode xattrs, or we don't have this inode in memory. | |
4800 | * Read the block from disk. | |
4801 | */ | |
4802 | get_bh(bh); | |
4803 | bh->b_end_io = end_buffer_read_sync; | |
4804 | submit_bh(READ_META, bh); | |
4805 | wait_on_buffer(bh); | |
4806 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4807 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4808 | "unable to read itable block"); | |
ac27a0ec DK |
4809 | brelse(bh); |
4810 | return -EIO; | |
4811 | } | |
4812 | } | |
4813 | has_buffer: | |
4814 | iloc->bh = bh; | |
4815 | return 0; | |
4816 | } | |
4817 | ||
617ba13b | 4818 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4819 | { |
4820 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4821 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4822 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4823 | } |
4824 | ||
617ba13b | 4825 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4826 | { |
617ba13b | 4827 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
4828 | |
4829 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 4830 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 4831 | inode->i_flags |= S_SYNC; |
617ba13b | 4832 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 4833 | inode->i_flags |= S_APPEND; |
617ba13b | 4834 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 4835 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 4836 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 4837 | inode->i_flags |= S_NOATIME; |
617ba13b | 4838 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
4839 | inode->i_flags |= S_DIRSYNC; |
4840 | } | |
4841 | ||
ff9ddf7e JK |
4842 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4843 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4844 | { | |
84a8dce2 DM |
4845 | unsigned int vfs_fl; |
4846 | unsigned long old_fl, new_fl; | |
4847 | ||
4848 | do { | |
4849 | vfs_fl = ei->vfs_inode.i_flags; | |
4850 | old_fl = ei->i_flags; | |
4851 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4852 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4853 | EXT4_DIRSYNC_FL); | |
4854 | if (vfs_fl & S_SYNC) | |
4855 | new_fl |= EXT4_SYNC_FL; | |
4856 | if (vfs_fl & S_APPEND) | |
4857 | new_fl |= EXT4_APPEND_FL; | |
4858 | if (vfs_fl & S_IMMUTABLE) | |
4859 | new_fl |= EXT4_IMMUTABLE_FL; | |
4860 | if (vfs_fl & S_NOATIME) | |
4861 | new_fl |= EXT4_NOATIME_FL; | |
4862 | if (vfs_fl & S_DIRSYNC) | |
4863 | new_fl |= EXT4_DIRSYNC_FL; | |
4864 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4865 | } |
de9a55b8 | 4866 | |
0fc1b451 | 4867 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4868 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4869 | { |
4870 | blkcnt_t i_blocks ; | |
8180a562 AK |
4871 | struct inode *inode = &(ei->vfs_inode); |
4872 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4873 | |
4874 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4875 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4876 | /* we are using combined 48 bit field */ | |
4877 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4878 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4879 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4880 | /* i_blocks represent file system block size */ |
4881 | return i_blocks << (inode->i_blkbits - 9); | |
4882 | } else { | |
4883 | return i_blocks; | |
4884 | } | |
0fc1b451 AK |
4885 | } else { |
4886 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4887 | } | |
4888 | } | |
ff9ddf7e | 4889 | |
1d1fe1ee | 4890 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4891 | { |
617ba13b MC |
4892 | struct ext4_iloc iloc; |
4893 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4894 | struct ext4_inode_info *ei; |
1d1fe1ee | 4895 | struct inode *inode; |
b436b9be | 4896 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4897 | long ret; |
ac27a0ec DK |
4898 | int block; |
4899 | ||
1d1fe1ee DH |
4900 | inode = iget_locked(sb, ino); |
4901 | if (!inode) | |
4902 | return ERR_PTR(-ENOMEM); | |
4903 | if (!(inode->i_state & I_NEW)) | |
4904 | return inode; | |
4905 | ||
4906 | ei = EXT4_I(inode); | |
567f3e9a | 4907 | iloc.bh = 0; |
ac27a0ec | 4908 | |
1d1fe1ee DH |
4909 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4910 | if (ret < 0) | |
ac27a0ec | 4911 | goto bad_inode; |
617ba13b | 4912 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
4913 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
4914 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
4915 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4916 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
4917 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4918 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
4919 | } | |
4920 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
ac27a0ec | 4921 | |
19f5fb7a | 4922 | ei->i_state_flags = 0; |
ac27a0ec DK |
4923 | ei->i_dir_start_lookup = 0; |
4924 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4925 | /* We now have enough fields to check if the inode was active or not. | |
4926 | * This is needed because nfsd might try to access dead inodes | |
4927 | * the test is that same one that e2fsck uses | |
4928 | * NeilBrown 1999oct15 | |
4929 | */ | |
4930 | if (inode->i_nlink == 0) { | |
4931 | if (inode->i_mode == 0 || | |
617ba13b | 4932 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 4933 | /* this inode is deleted */ |
1d1fe1ee | 4934 | ret = -ESTALE; |
ac27a0ec DK |
4935 | goto bad_inode; |
4936 | } | |
4937 | /* The only unlinked inodes we let through here have | |
4938 | * valid i_mode and are being read by the orphan | |
4939 | * recovery code: that's fine, we're about to complete | |
4940 | * the process of deleting those. */ | |
4941 | } | |
ac27a0ec | 4942 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4943 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4944 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4945 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4946 | ei->i_file_acl |= |
4947 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4948 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4949 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4950 | #ifdef CONFIG_QUOTA |
4951 | ei->i_reserved_quota = 0; | |
4952 | #endif | |
ac27a0ec DK |
4953 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4954 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4955 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4956 | /* |
4957 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4958 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4959 | */ | |
617ba13b | 4960 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4961 | ei->i_data[block] = raw_inode->i_block[block]; |
4962 | INIT_LIST_HEAD(&ei->i_orphan); | |
4963 | ||
b436b9be JK |
4964 | /* |
4965 | * Set transaction id's of transactions that have to be committed | |
4966 | * to finish f[data]sync. We set them to currently running transaction | |
4967 | * as we cannot be sure that the inode or some of its metadata isn't | |
4968 | * part of the transaction - the inode could have been reclaimed and | |
4969 | * now it is reread from disk. | |
4970 | */ | |
4971 | if (journal) { | |
4972 | transaction_t *transaction; | |
4973 | tid_t tid; | |
4974 | ||
a931da6a | 4975 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4976 | if (journal->j_running_transaction) |
4977 | transaction = journal->j_running_transaction; | |
4978 | else | |
4979 | transaction = journal->j_committing_transaction; | |
4980 | if (transaction) | |
4981 | tid = transaction->t_tid; | |
4982 | else | |
4983 | tid = journal->j_commit_sequence; | |
a931da6a | 4984 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4985 | ei->i_sync_tid = tid; |
4986 | ei->i_datasync_tid = tid; | |
4987 | } | |
4988 | ||
0040d987 | 4989 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec | 4990 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
617ba13b | 4991 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f | 4992 | EXT4_INODE_SIZE(inode->i_sb)) { |
1d1fe1ee | 4993 | ret = -EIO; |
ac27a0ec | 4994 | goto bad_inode; |
e5d2861f | 4995 | } |
ac27a0ec DK |
4996 | if (ei->i_extra_isize == 0) { |
4997 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4998 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4999 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
5000 | } else { |
5001 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 5002 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 5003 | ei->i_extra_isize; |
617ba13b | 5004 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
19f5fb7a | 5005 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
ac27a0ec DK |
5006 | } |
5007 | } else | |
5008 | ei->i_extra_isize = 0; | |
5009 | ||
ef7f3835 KS |
5010 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
5011 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
5012 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
5013 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
5014 | ||
25ec56b5 JNC |
5015 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
5016 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
5017 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5018 | inode->i_version |= | |
5019 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
5020 | } | |
5021 | ||
c4b5a614 | 5022 | ret = 0; |
485c26ec | 5023 | if (ei->i_file_acl && |
1032988c | 5024 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
5025 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
5026 | ei->i_file_acl); | |
485c26ec TT |
5027 | ret = -EIO; |
5028 | goto bad_inode; | |
07a03824 | 5029 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
c4b5a614 TT |
5030 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
5031 | (S_ISLNK(inode->i_mode) && | |
5032 | !ext4_inode_is_fast_symlink(inode))) | |
5033 | /* Validate extent which is part of inode */ | |
5034 | ret = ext4_ext_check_inode(inode); | |
de9a55b8 | 5035 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
fe2c8191 TN |
5036 | (S_ISLNK(inode->i_mode) && |
5037 | !ext4_inode_is_fast_symlink(inode))) { | |
de9a55b8 | 5038 | /* Validate block references which are part of inode */ |
fe2c8191 TN |
5039 | ret = ext4_check_inode_blockref(inode); |
5040 | } | |
567f3e9a | 5041 | if (ret) |
de9a55b8 | 5042 | goto bad_inode; |
7a262f7c | 5043 | |
ac27a0ec | 5044 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
5045 | inode->i_op = &ext4_file_inode_operations; |
5046 | inode->i_fop = &ext4_file_operations; | |
5047 | ext4_set_aops(inode); | |
ac27a0ec | 5048 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
5049 | inode->i_op = &ext4_dir_inode_operations; |
5050 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 5051 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 5052 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 5053 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
5054 | nd_terminate_link(ei->i_data, inode->i_size, |
5055 | sizeof(ei->i_data) - 1); | |
5056 | } else { | |
617ba13b MC |
5057 | inode->i_op = &ext4_symlink_inode_operations; |
5058 | ext4_set_aops(inode); | |
ac27a0ec | 5059 | } |
563bdd61 TT |
5060 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
5061 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 5062 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
5063 | if (raw_inode->i_block[0]) |
5064 | init_special_inode(inode, inode->i_mode, | |
5065 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
5066 | else | |
5067 | init_special_inode(inode, inode->i_mode, | |
5068 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 | 5069 | } else { |
563bdd61 | 5070 | ret = -EIO; |
24676da4 | 5071 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 5072 | goto bad_inode; |
ac27a0ec | 5073 | } |
af5bc92d | 5074 | brelse(iloc.bh); |
617ba13b | 5075 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
5076 | unlock_new_inode(inode); |
5077 | return inode; | |
ac27a0ec DK |
5078 | |
5079 | bad_inode: | |
567f3e9a | 5080 | brelse(iloc.bh); |
1d1fe1ee DH |
5081 | iget_failed(inode); |
5082 | return ERR_PTR(ret); | |
ac27a0ec DK |
5083 | } |
5084 | ||
0fc1b451 AK |
5085 | static int ext4_inode_blocks_set(handle_t *handle, |
5086 | struct ext4_inode *raw_inode, | |
5087 | struct ext4_inode_info *ei) | |
5088 | { | |
5089 | struct inode *inode = &(ei->vfs_inode); | |
5090 | u64 i_blocks = inode->i_blocks; | |
5091 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
5092 | |
5093 | if (i_blocks <= ~0U) { | |
5094 | /* | |
5095 | * i_blocks can be represnted in a 32 bit variable | |
5096 | * as multiple of 512 bytes | |
5097 | */ | |
8180a562 | 5098 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5099 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 5100 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
5101 | return 0; |
5102 | } | |
5103 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
5104 | return -EFBIG; | |
5105 | ||
5106 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
5107 | /* |
5108 | * i_blocks can be represented in a 48 bit variable | |
5109 | * as multiple of 512 bytes | |
5110 | */ | |
8180a562 | 5111 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5112 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 5113 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 5114 | } else { |
84a8dce2 | 5115 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
5116 | /* i_block is stored in file system block size */ |
5117 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
5118 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
5119 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 5120 | } |
f287a1a5 | 5121 | return 0; |
0fc1b451 AK |
5122 | } |
5123 | ||
ac27a0ec DK |
5124 | /* |
5125 | * Post the struct inode info into an on-disk inode location in the | |
5126 | * buffer-cache. This gobbles the caller's reference to the | |
5127 | * buffer_head in the inode location struct. | |
5128 | * | |
5129 | * The caller must have write access to iloc->bh. | |
5130 | */ | |
617ba13b | 5131 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 5132 | struct inode *inode, |
830156c7 | 5133 | struct ext4_iloc *iloc) |
ac27a0ec | 5134 | { |
617ba13b MC |
5135 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
5136 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
5137 | struct buffer_head *bh = iloc->bh; |
5138 | int err = 0, rc, block; | |
5139 | ||
5140 | /* For fields not not tracking in the in-memory inode, | |
5141 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 5142 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 5143 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 5144 | |
ff9ddf7e | 5145 | ext4_get_inode_flags(ei); |
ac27a0ec | 5146 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
af5bc92d | 5147 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
5148 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
5149 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
5150 | /* | |
5151 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
5152 | * re-used with the upper 16 bits of the uid/gid intact | |
5153 | */ | |
af5bc92d | 5154 | if (!ei->i_dtime) { |
ac27a0ec DK |
5155 | raw_inode->i_uid_high = |
5156 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
5157 | raw_inode->i_gid_high = | |
5158 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
5159 | } else { | |
5160 | raw_inode->i_uid_high = 0; | |
5161 | raw_inode->i_gid_high = 0; | |
5162 | } | |
5163 | } else { | |
5164 | raw_inode->i_uid_low = | |
5165 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
5166 | raw_inode->i_gid_low = | |
5167 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
5168 | raw_inode->i_uid_high = 0; | |
5169 | raw_inode->i_gid_high = 0; | |
5170 | } | |
5171 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5172 | |
5173 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5174 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5175 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5176 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5177 | ||
0fc1b451 AK |
5178 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
5179 | goto out_brelse; | |
ac27a0ec | 5180 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
1b9c12f4 | 5181 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); |
9b8f1f01 MC |
5182 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
5183 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
5184 | raw_inode->i_file_acl_high = |
5185 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5186 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
5187 | ext4_isize_set(raw_inode, ei->i_disksize); |
5188 | if (ei->i_disksize > 0x7fffffffULL) { | |
5189 | struct super_block *sb = inode->i_sb; | |
5190 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
5191 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
5192 | EXT4_SB(sb)->s_es->s_rev_level == | |
5193 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
5194 | /* If this is the first large file | |
5195 | * created, add a flag to the superblock. | |
5196 | */ | |
5197 | err = ext4_journal_get_write_access(handle, | |
5198 | EXT4_SB(sb)->s_sbh); | |
5199 | if (err) | |
5200 | goto out_brelse; | |
5201 | ext4_update_dynamic_rev(sb); | |
5202 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 5203 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
a48380f7 | 5204 | sb->s_dirt = 1; |
0390131b | 5205 | ext4_handle_sync(handle); |
73b50c1c | 5206 | err = ext4_handle_dirty_metadata(handle, NULL, |
a48380f7 | 5207 | EXT4_SB(sb)->s_sbh); |
ac27a0ec DK |
5208 | } |
5209 | } | |
5210 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5211 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5212 | if (old_valid_dev(inode->i_rdev)) { | |
5213 | raw_inode->i_block[0] = | |
5214 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5215 | raw_inode->i_block[1] = 0; | |
5216 | } else { | |
5217 | raw_inode->i_block[0] = 0; | |
5218 | raw_inode->i_block[1] = | |
5219 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5220 | raw_inode->i_block[2] = 0; | |
5221 | } | |
de9a55b8 TT |
5222 | } else |
5223 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
5224 | raw_inode->i_block[block] = ei->i_data[block]; | |
ac27a0ec | 5225 | |
25ec56b5 JNC |
5226 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
5227 | if (ei->i_extra_isize) { | |
5228 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5229 | raw_inode->i_version_hi = | |
5230 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 5231 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
5232 | } |
5233 | ||
830156c7 | 5234 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 5235 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
5236 | if (!err) |
5237 | err = rc; | |
19f5fb7a | 5238 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 5239 | |
b436b9be | 5240 | ext4_update_inode_fsync_trans(handle, inode, 0); |
ac27a0ec | 5241 | out_brelse: |
af5bc92d | 5242 | brelse(bh); |
617ba13b | 5243 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5244 | return err; |
5245 | } | |
5246 | ||
5247 | /* | |
617ba13b | 5248 | * ext4_write_inode() |
ac27a0ec DK |
5249 | * |
5250 | * We are called from a few places: | |
5251 | * | |
5252 | * - Within generic_file_write() for O_SYNC files. | |
5253 | * Here, there will be no transaction running. We wait for any running | |
5254 | * trasnaction to commit. | |
5255 | * | |
5256 | * - Within sys_sync(), kupdate and such. | |
5257 | * We wait on commit, if tol to. | |
5258 | * | |
5259 | * - Within prune_icache() (PF_MEMALLOC == true) | |
5260 | * Here we simply return. We can't afford to block kswapd on the | |
5261 | * journal commit. | |
5262 | * | |
5263 | * In all cases it is actually safe for us to return without doing anything, | |
5264 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 5265 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
5266 | * knfsd. |
5267 | * | |
5268 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5269 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5270 | * which we are interested. | |
5271 | * | |
5272 | * It would be a bug for them to not do this. The code: | |
5273 | * | |
5274 | * mark_inode_dirty(inode) | |
5275 | * stuff(); | |
5276 | * inode->i_size = expr; | |
5277 | * | |
5278 | * is in error because a kswapd-driven write_inode() could occur while | |
5279 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
5280 | * will no longer be on the superblock's dirty inode list. | |
5281 | */ | |
a9185b41 | 5282 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5283 | { |
91ac6f43 FM |
5284 | int err; |
5285 | ||
ac27a0ec DK |
5286 | if (current->flags & PF_MEMALLOC) |
5287 | return 0; | |
5288 | ||
91ac6f43 FM |
5289 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5290 | if (ext4_journal_current_handle()) { | |
5291 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5292 | dump_stack(); | |
5293 | return -EIO; | |
5294 | } | |
ac27a0ec | 5295 | |
a9185b41 | 5296 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
5297 | return 0; |
5298 | ||
5299 | err = ext4_force_commit(inode->i_sb); | |
5300 | } else { | |
5301 | struct ext4_iloc iloc; | |
ac27a0ec | 5302 | |
8b472d73 | 5303 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
5304 | if (err) |
5305 | return err; | |
a9185b41 | 5306 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
5307 | sync_dirty_buffer(iloc.bh); |
5308 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
5309 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
5310 | "IO error syncing inode"); | |
830156c7 FM |
5311 | err = -EIO; |
5312 | } | |
fd2dd9fb | 5313 | brelse(iloc.bh); |
91ac6f43 FM |
5314 | } |
5315 | return err; | |
ac27a0ec DK |
5316 | } |
5317 | ||
5318 | /* | |
617ba13b | 5319 | * ext4_setattr() |
ac27a0ec DK |
5320 | * |
5321 | * Called from notify_change. | |
5322 | * | |
5323 | * We want to trap VFS attempts to truncate the file as soon as | |
5324 | * possible. In particular, we want to make sure that when the VFS | |
5325 | * shrinks i_size, we put the inode on the orphan list and modify | |
5326 | * i_disksize immediately, so that during the subsequent flushing of | |
5327 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5328 | * commit will leave the blocks being flushed in an unused state on | |
5329 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5330 | * be freed, so we have a strong guarantee that no future commit will | |
5331 | * leave these blocks visible to the user.) | |
5332 | * | |
678aaf48 JK |
5333 | * Another thing we have to assure is that if we are in ordered mode |
5334 | * and inode is still attached to the committing transaction, we must | |
5335 | * we start writeout of all the dirty pages which are being truncated. | |
5336 | * This way we are sure that all the data written in the previous | |
5337 | * transaction are already on disk (truncate waits for pages under | |
5338 | * writeback). | |
5339 | * | |
5340 | * Called with inode->i_mutex down. | |
ac27a0ec | 5341 | */ |
617ba13b | 5342 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
5343 | { |
5344 | struct inode *inode = dentry->d_inode; | |
5345 | int error, rc = 0; | |
5346 | const unsigned int ia_valid = attr->ia_valid; | |
5347 | ||
5348 | error = inode_change_ok(inode, attr); | |
5349 | if (error) | |
5350 | return error; | |
5351 | ||
12755627 | 5352 | if (is_quota_modification(inode, attr)) |
871a2931 | 5353 | dquot_initialize(inode); |
ac27a0ec DK |
5354 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
5355 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
5356 | handle_t *handle; | |
5357 | ||
5358 | /* (user+group)*(old+new) structure, inode write (sb, | |
5359 | * inode block, ? - but truncate inode update has it) */ | |
5aca07eb | 5360 | handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
194074ac | 5361 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3); |
ac27a0ec DK |
5362 | if (IS_ERR(handle)) { |
5363 | error = PTR_ERR(handle); | |
5364 | goto err_out; | |
5365 | } | |
b43fa828 | 5366 | error = dquot_transfer(inode, attr); |
ac27a0ec | 5367 | if (error) { |
617ba13b | 5368 | ext4_journal_stop(handle); |
ac27a0ec DK |
5369 | return error; |
5370 | } | |
5371 | /* Update corresponding info in inode so that everything is in | |
5372 | * one transaction */ | |
5373 | if (attr->ia_valid & ATTR_UID) | |
5374 | inode->i_uid = attr->ia_uid; | |
5375 | if (attr->ia_valid & ATTR_GID) | |
5376 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5377 | error = ext4_mark_inode_dirty(handle, inode); |
5378 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5379 | } |
5380 | ||
e2b46574 | 5381 | if (attr->ia_valid & ATTR_SIZE) { |
12e9b892 | 5382 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5383 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5384 | ||
0c095c7f TT |
5385 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5386 | return -EFBIG; | |
e2b46574 ES |
5387 | } |
5388 | } | |
5389 | ||
ac27a0ec | 5390 | if (S_ISREG(inode->i_mode) && |
c8d46e41 JZ |
5391 | attr->ia_valid & ATTR_SIZE && |
5392 | (attr->ia_size < inode->i_size || | |
12e9b892 | 5393 | (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) { |
ac27a0ec DK |
5394 | handle_t *handle; |
5395 | ||
617ba13b | 5396 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
5397 | if (IS_ERR(handle)) { |
5398 | error = PTR_ERR(handle); | |
5399 | goto err_out; | |
5400 | } | |
5401 | ||
617ba13b MC |
5402 | error = ext4_orphan_add(handle, inode); |
5403 | EXT4_I(inode)->i_disksize = attr->ia_size; | |
5404 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
5405 | if (!error) |
5406 | error = rc; | |
617ba13b | 5407 | ext4_journal_stop(handle); |
678aaf48 JK |
5408 | |
5409 | if (ext4_should_order_data(inode)) { | |
5410 | error = ext4_begin_ordered_truncate(inode, | |
5411 | attr->ia_size); | |
5412 | if (error) { | |
5413 | /* Do as much error cleanup as possible */ | |
5414 | handle = ext4_journal_start(inode, 3); | |
5415 | if (IS_ERR(handle)) { | |
5416 | ext4_orphan_del(NULL, inode); | |
5417 | goto err_out; | |
5418 | } | |
5419 | ext4_orphan_del(handle, inode); | |
5420 | ext4_journal_stop(handle); | |
5421 | goto err_out; | |
5422 | } | |
5423 | } | |
c8d46e41 | 5424 | /* ext4_truncate will clear the flag */ |
12e9b892 | 5425 | if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))) |
c8d46e41 | 5426 | ext4_truncate(inode); |
ac27a0ec DK |
5427 | } |
5428 | ||
1025774c CH |
5429 | if ((attr->ia_valid & ATTR_SIZE) && |
5430 | attr->ia_size != i_size_read(inode)) | |
5431 | rc = vmtruncate(inode, attr->ia_size); | |
ac27a0ec | 5432 | |
1025774c CH |
5433 | if (!rc) { |
5434 | setattr_copy(inode, attr); | |
5435 | mark_inode_dirty(inode); | |
5436 | } | |
5437 | ||
5438 | /* | |
5439 | * If the call to ext4_truncate failed to get a transaction handle at | |
5440 | * all, we need to clean up the in-core orphan list manually. | |
5441 | */ | |
ac27a0ec | 5442 | if (inode->i_nlink) |
617ba13b | 5443 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
5444 | |
5445 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 5446 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
5447 | |
5448 | err_out: | |
617ba13b | 5449 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5450 | if (!error) |
5451 | error = rc; | |
5452 | return error; | |
5453 | } | |
5454 | ||
3e3398a0 MC |
5455 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
5456 | struct kstat *stat) | |
5457 | { | |
5458 | struct inode *inode; | |
5459 | unsigned long delalloc_blocks; | |
5460 | ||
5461 | inode = dentry->d_inode; | |
5462 | generic_fillattr(inode, stat); | |
5463 | ||
5464 | /* | |
5465 | * We can't update i_blocks if the block allocation is delayed | |
5466 | * otherwise in the case of system crash before the real block | |
5467 | * allocation is done, we will have i_blocks inconsistent with | |
5468 | * on-disk file blocks. | |
5469 | * We always keep i_blocks updated together with real | |
5470 | * allocation. But to not confuse with user, stat | |
5471 | * will return the blocks that include the delayed allocation | |
5472 | * blocks for this file. | |
5473 | */ | |
5474 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
5475 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
5476 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
5477 | ||
5478 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
5479 | return 0; | |
5480 | } | |
ac27a0ec | 5481 | |
a02908f1 MC |
5482 | static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks, |
5483 | int chunk) | |
5484 | { | |
5485 | int indirects; | |
5486 | ||
5487 | /* if nrblocks are contiguous */ | |
5488 | if (chunk) { | |
5489 | /* | |
5490 | * With N contiguous data blocks, it need at most | |
5491 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks | |
5492 | * 2 dindirect blocks | |
5493 | * 1 tindirect block | |
5494 | */ | |
5495 | indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
5496 | return indirects + 3; | |
5497 | } | |
5498 | /* | |
5499 | * if nrblocks are not contiguous, worse case, each block touch | |
5500 | * a indirect block, and each indirect block touch a double indirect | |
5501 | * block, plus a triple indirect block | |
5502 | */ | |
5503 | indirects = nrblocks * 2 + 1; | |
5504 | return indirects; | |
5505 | } | |
5506 | ||
5507 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
5508 | { | |
12e9b892 | 5509 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
ac51d837 TT |
5510 | return ext4_indirect_trans_blocks(inode, nrblocks, chunk); |
5511 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); | |
a02908f1 | 5512 | } |
ac51d837 | 5513 | |
ac27a0ec | 5514 | /* |
a02908f1 MC |
5515 | * Account for index blocks, block groups bitmaps and block group |
5516 | * descriptor blocks if modify datablocks and index blocks | |
5517 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5518 | * |
a02908f1 | 5519 | * If datablocks are discontiguous, they are possible to spread over |
af901ca1 | 5520 | * different block groups too. If they are contiuguous, with flexbg, |
a02908f1 | 5521 | * they could still across block group boundary. |
ac27a0ec | 5522 | * |
a02908f1 MC |
5523 | * Also account for superblock, inode, quota and xattr blocks |
5524 | */ | |
5525 | int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
5526 | { | |
8df9675f TT |
5527 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5528 | int gdpblocks; | |
a02908f1 MC |
5529 | int idxblocks; |
5530 | int ret = 0; | |
5531 | ||
5532 | /* | |
5533 | * How many index blocks need to touch to modify nrblocks? | |
5534 | * The "Chunk" flag indicating whether the nrblocks is | |
5535 | * physically contiguous on disk | |
5536 | * | |
5537 | * For Direct IO and fallocate, they calls get_block to allocate | |
5538 | * one single extent at a time, so they could set the "Chunk" flag | |
5539 | */ | |
5540 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
5541 | ||
5542 | ret = idxblocks; | |
5543 | ||
5544 | /* | |
5545 | * Now let's see how many group bitmaps and group descriptors need | |
5546 | * to account | |
5547 | */ | |
5548 | groups = idxblocks; | |
5549 | if (chunk) | |
5550 | groups += 1; | |
5551 | else | |
5552 | groups += nrblocks; | |
5553 | ||
5554 | gdpblocks = groups; | |
8df9675f TT |
5555 | if (groups > ngroups) |
5556 | groups = ngroups; | |
a02908f1 MC |
5557 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5558 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5559 | ||
5560 | /* bitmaps and block group descriptor blocks */ | |
5561 | ret += groups + gdpblocks; | |
5562 | ||
5563 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5564 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5565 | ||
5566 | return ret; | |
5567 | } | |
5568 | ||
5569 | /* | |
5570 | * Calulate the total number of credits to reserve to fit | |
f3bd1f3f MC |
5571 | * the modification of a single pages into a single transaction, |
5572 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5573 | * |
525f4ed8 | 5574 | * This could be called via ext4_write_begin() |
ac27a0ec | 5575 | * |
525f4ed8 | 5576 | * We need to consider the worse case, when |
a02908f1 | 5577 | * one new block per extent. |
ac27a0ec | 5578 | */ |
a86c6181 | 5579 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5580 | { |
617ba13b | 5581 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5582 | int ret; |
5583 | ||
a02908f1 | 5584 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 5585 | |
a02908f1 | 5586 | /* Account for data blocks for journalled mode */ |
617ba13b | 5587 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5588 | ret += bpp; |
ac27a0ec DK |
5589 | return ret; |
5590 | } | |
f3bd1f3f MC |
5591 | |
5592 | /* | |
5593 | * Calculate the journal credits for a chunk of data modification. | |
5594 | * | |
5595 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5596 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5597 | * |
5598 | * journal buffers for data blocks are not included here, as DIO | |
5599 | * and fallocate do no need to journal data buffers. | |
5600 | */ | |
5601 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5602 | { | |
5603 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5604 | } | |
5605 | ||
ac27a0ec | 5606 | /* |
617ba13b | 5607 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5608 | * Give this, we know that the caller already has write access to iloc->bh. |
5609 | */ | |
617ba13b | 5610 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5611 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5612 | { |
5613 | int err = 0; | |
5614 | ||
25ec56b5 JNC |
5615 | if (test_opt(inode->i_sb, I_VERSION)) |
5616 | inode_inc_iversion(inode); | |
5617 | ||
ac27a0ec DK |
5618 | /* the do_update_inode consumes one bh->b_count */ |
5619 | get_bh(iloc->bh); | |
5620 | ||
dab291af | 5621 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5622 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5623 | put_bh(iloc->bh); |
5624 | return err; | |
5625 | } | |
5626 | ||
5627 | /* | |
5628 | * On success, We end up with an outstanding reference count against | |
5629 | * iloc->bh. This _must_ be cleaned up later. | |
5630 | */ | |
5631 | ||
5632 | int | |
617ba13b MC |
5633 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5634 | struct ext4_iloc *iloc) | |
ac27a0ec | 5635 | { |
0390131b FM |
5636 | int err; |
5637 | ||
5638 | err = ext4_get_inode_loc(inode, iloc); | |
5639 | if (!err) { | |
5640 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5641 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5642 | if (err) { | |
5643 | brelse(iloc->bh); | |
5644 | iloc->bh = NULL; | |
ac27a0ec DK |
5645 | } |
5646 | } | |
617ba13b | 5647 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5648 | return err; |
5649 | } | |
5650 | ||
6dd4ee7c KS |
5651 | /* |
5652 | * Expand an inode by new_extra_isize bytes. | |
5653 | * Returns 0 on success or negative error number on failure. | |
5654 | */ | |
1d03ec98 AK |
5655 | static int ext4_expand_extra_isize(struct inode *inode, |
5656 | unsigned int new_extra_isize, | |
5657 | struct ext4_iloc iloc, | |
5658 | handle_t *handle) | |
6dd4ee7c KS |
5659 | { |
5660 | struct ext4_inode *raw_inode; | |
5661 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
5662 | |
5663 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5664 | return 0; | |
5665 | ||
5666 | raw_inode = ext4_raw_inode(&iloc); | |
5667 | ||
5668 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
5669 | |
5670 | /* No extended attributes present */ | |
19f5fb7a TT |
5671 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
5672 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
5673 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
5674 | new_extra_isize); | |
5675 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5676 | return 0; | |
5677 | } | |
5678 | ||
5679 | /* try to expand with EAs present */ | |
5680 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5681 | raw_inode, handle); | |
5682 | } | |
5683 | ||
ac27a0ec DK |
5684 | /* |
5685 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5686 | * dirtiness (it may still be data-dirty). | |
5687 | * This means that the in-core inode may be reaped by prune_icache | |
5688 | * without having to perform any I/O. This is a very good thing, | |
5689 | * because *any* task may call prune_icache - even ones which | |
5690 | * have a transaction open against a different journal. | |
5691 | * | |
5692 | * Is this cheating? Not really. Sure, we haven't written the | |
5693 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5694 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5695 | * we start and wait on commits. | |
5696 | * | |
5697 | * Is this efficient/effective? Well, we're being nice to the system | |
5698 | * by cleaning up our inodes proactively so they can be reaped | |
5699 | * without I/O. But we are potentially leaving up to five seconds' | |
5700 | * worth of inodes floating about which prune_icache wants us to | |
5701 | * write out. One way to fix that would be to get prune_icache() | |
5702 | * to do a write_super() to free up some memory. It has the desired | |
5703 | * effect. | |
5704 | */ | |
617ba13b | 5705 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5706 | { |
617ba13b | 5707 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5708 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5709 | static unsigned int mnt_count; | |
5710 | int err, ret; | |
ac27a0ec DK |
5711 | |
5712 | might_sleep(); | |
617ba13b | 5713 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5714 | if (ext4_handle_valid(handle) && |
5715 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5716 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5717 | /* |
5718 | * We need extra buffer credits since we may write into EA block | |
5719 | * with this same handle. If journal_extend fails, then it will | |
5720 | * only result in a minor loss of functionality for that inode. | |
5721 | * If this is felt to be critical, then e2fsck should be run to | |
5722 | * force a large enough s_min_extra_isize. | |
5723 | */ | |
5724 | if ((jbd2_journal_extend(handle, | |
5725 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5726 | ret = ext4_expand_extra_isize(inode, | |
5727 | sbi->s_want_extra_isize, | |
5728 | iloc, handle); | |
5729 | if (ret) { | |
19f5fb7a TT |
5730 | ext4_set_inode_state(inode, |
5731 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
5732 | if (mnt_count != |
5733 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5734 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5735 | "Unable to expand inode %lu. Delete" |
5736 | " some EAs or run e2fsck.", | |
5737 | inode->i_ino); | |
c1bddad9 AK |
5738 | mnt_count = |
5739 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5740 | } |
5741 | } | |
5742 | } | |
5743 | } | |
ac27a0ec | 5744 | if (!err) |
617ba13b | 5745 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5746 | return err; |
5747 | } | |
5748 | ||
5749 | /* | |
617ba13b | 5750 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5751 | * |
5752 | * We're really interested in the case where a file is being extended. | |
5753 | * i_size has been changed by generic_commit_write() and we thus need | |
5754 | * to include the updated inode in the current transaction. | |
5755 | * | |
5dd4056d | 5756 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5757 | * are allocated to the file. |
5758 | * | |
5759 | * If the inode is marked synchronous, we don't honour that here - doing | |
5760 | * so would cause a commit on atime updates, which we don't bother doing. | |
5761 | * We handle synchronous inodes at the highest possible level. | |
5762 | */ | |
617ba13b | 5763 | void ext4_dirty_inode(struct inode *inode) |
ac27a0ec | 5764 | { |
ac27a0ec DK |
5765 | handle_t *handle; |
5766 | ||
617ba13b | 5767 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
5768 | if (IS_ERR(handle)) |
5769 | goto out; | |
f3dc272f | 5770 | |
f3dc272f CW |
5771 | ext4_mark_inode_dirty(handle, inode); |
5772 | ||
617ba13b | 5773 | ext4_journal_stop(handle); |
ac27a0ec DK |
5774 | out: |
5775 | return; | |
5776 | } | |
5777 | ||
5778 | #if 0 | |
5779 | /* | |
5780 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5781 | * it from being flushed to disk early. Unlike | |
617ba13b | 5782 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5783 | * returns no iloc structure, so the caller needs to repeat the iloc |
5784 | * lookup to mark the inode dirty later. | |
5785 | */ | |
617ba13b | 5786 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5787 | { |
617ba13b | 5788 | struct ext4_iloc iloc; |
ac27a0ec DK |
5789 | |
5790 | int err = 0; | |
5791 | if (handle) { | |
617ba13b | 5792 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5793 | if (!err) { |
5794 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5795 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5796 | if (!err) |
0390131b | 5797 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5798 | NULL, |
0390131b | 5799 | iloc.bh); |
ac27a0ec DK |
5800 | brelse(iloc.bh); |
5801 | } | |
5802 | } | |
617ba13b | 5803 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5804 | return err; |
5805 | } | |
5806 | #endif | |
5807 | ||
617ba13b | 5808 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5809 | { |
5810 | journal_t *journal; | |
5811 | handle_t *handle; | |
5812 | int err; | |
5813 | ||
5814 | /* | |
5815 | * We have to be very careful here: changing a data block's | |
5816 | * journaling status dynamically is dangerous. If we write a | |
5817 | * data block to the journal, change the status and then delete | |
5818 | * that block, we risk forgetting to revoke the old log record | |
5819 | * from the journal and so a subsequent replay can corrupt data. | |
5820 | * So, first we make sure that the journal is empty and that | |
5821 | * nobody is changing anything. | |
5822 | */ | |
5823 | ||
617ba13b | 5824 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5825 | if (!journal) |
5826 | return 0; | |
d699594d | 5827 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5828 | return -EROFS; |
5829 | ||
dab291af MC |
5830 | jbd2_journal_lock_updates(journal); |
5831 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
5832 | |
5833 | /* | |
5834 | * OK, there are no updates running now, and all cached data is | |
5835 | * synced to disk. We are now in a completely consistent state | |
5836 | * which doesn't have anything in the journal, and we know that | |
5837 | * no filesystem updates are running, so it is safe to modify | |
5838 | * the inode's in-core data-journaling state flag now. | |
5839 | */ | |
5840 | ||
5841 | if (val) | |
12e9b892 | 5842 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
ac27a0ec | 5843 | else |
12e9b892 | 5844 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
617ba13b | 5845 | ext4_set_aops(inode); |
ac27a0ec | 5846 | |
dab291af | 5847 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
5848 | |
5849 | /* Finally we can mark the inode as dirty. */ | |
5850 | ||
617ba13b | 5851 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
5852 | if (IS_ERR(handle)) |
5853 | return PTR_ERR(handle); | |
5854 | ||
617ba13b | 5855 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5856 | ext4_handle_sync(handle); |
617ba13b MC |
5857 | ext4_journal_stop(handle); |
5858 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5859 | |
5860 | return err; | |
5861 | } | |
2e9ee850 AK |
5862 | |
5863 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5864 | { | |
5865 | return !buffer_mapped(bh); | |
5866 | } | |
5867 | ||
c2ec175c | 5868 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5869 | { |
c2ec175c | 5870 | struct page *page = vmf->page; |
2e9ee850 AK |
5871 | loff_t size; |
5872 | unsigned long len; | |
5873 | int ret = -EINVAL; | |
79f0be8d | 5874 | void *fsdata; |
2e9ee850 AK |
5875 | struct file *file = vma->vm_file; |
5876 | struct inode *inode = file->f_path.dentry->d_inode; | |
5877 | struct address_space *mapping = inode->i_mapping; | |
5878 | ||
5879 | /* | |
5880 | * Get i_alloc_sem to stop truncates messing with the inode. We cannot | |
5881 | * get i_mutex because we are already holding mmap_sem. | |
5882 | */ | |
5883 | down_read(&inode->i_alloc_sem); | |
5884 | size = i_size_read(inode); | |
5885 | if (page->mapping != mapping || size <= page_offset(page) | |
5886 | || !PageUptodate(page)) { | |
5887 | /* page got truncated from under us? */ | |
5888 | goto out_unlock; | |
5889 | } | |
5890 | ret = 0; | |
5891 | if (PageMappedToDisk(page)) | |
5892 | goto out_unlock; | |
5893 | ||
5894 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5895 | len = size & ~PAGE_CACHE_MASK; | |
5896 | else | |
5897 | len = PAGE_CACHE_SIZE; | |
5898 | ||
a827eaff AK |
5899 | lock_page(page); |
5900 | /* | |
5901 | * return if we have all the buffers mapped. This avoid | |
5902 | * the need to call write_begin/write_end which does a | |
5903 | * journal_start/journal_stop which can block and take | |
5904 | * long time | |
5905 | */ | |
2e9ee850 | 5906 | if (page_has_buffers(page)) { |
2e9ee850 | 5907 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, |
a827eaff AK |
5908 | ext4_bh_unmapped)) { |
5909 | unlock_page(page); | |
2e9ee850 | 5910 | goto out_unlock; |
a827eaff | 5911 | } |
2e9ee850 | 5912 | } |
a827eaff | 5913 | unlock_page(page); |
2e9ee850 AK |
5914 | /* |
5915 | * OK, we need to fill the hole... Do write_begin write_end | |
5916 | * to do block allocation/reservation.We are not holding | |
5917 | * inode.i__mutex here. That allow * parallel write_begin, | |
5918 | * write_end call. lock_page prevent this from happening | |
5919 | * on the same page though | |
5920 | */ | |
5921 | ret = mapping->a_ops->write_begin(file, mapping, page_offset(page), | |
79f0be8d | 5922 | len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); |
2e9ee850 AK |
5923 | if (ret < 0) |
5924 | goto out_unlock; | |
5925 | ret = mapping->a_ops->write_end(file, mapping, page_offset(page), | |
79f0be8d | 5926 | len, len, page, fsdata); |
2e9ee850 AK |
5927 | if (ret < 0) |
5928 | goto out_unlock; | |
5929 | ret = 0; | |
5930 | out_unlock: | |
c2ec175c NP |
5931 | if (ret) |
5932 | ret = VM_FAULT_SIGBUS; | |
2e9ee850 AK |
5933 | up_read(&inode->i_alloc_sem); |
5934 | return ret; | |
5935 | } |