]>
Commit | Line | Data |
---|---|---|
1 | /* -*- mode: c; c-basic-offset: 8; -*- | |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public | |
8 | * License as published by the Free Software Foundation; either | |
9 | * version 2 of the License, or (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | * General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public | |
17 | * License along with this program; if not, write to the | |
18 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
19 | * Boston, MA 021110-1307, USA. | |
20 | */ | |
21 | ||
22 | #include <linux/fs.h> | |
23 | #include <linux/slab.h> | |
24 | #include <linux/highmem.h> | |
25 | #include <linux/pagemap.h> | |
26 | #include <asm/byteorder.h> | |
27 | #include <linux/swap.h> | |
28 | #include <linux/pipe_fs_i.h> | |
29 | #include <linux/mpage.h> | |
30 | #include <linux/quotaops.h> | |
31 | ||
32 | #define MLOG_MASK_PREFIX ML_FILE_IO | |
33 | #include <cluster/masklog.h> | |
34 | ||
35 | #include "ocfs2.h" | |
36 | ||
37 | #include "alloc.h" | |
38 | #include "aops.h" | |
39 | #include "dlmglue.h" | |
40 | #include "extent_map.h" | |
41 | #include "file.h" | |
42 | #include "inode.h" | |
43 | #include "journal.h" | |
44 | #include "suballoc.h" | |
45 | #include "super.h" | |
46 | #include "symlink.h" | |
47 | #include "refcounttree.h" | |
48 | ||
49 | #include "buffer_head_io.h" | |
50 | ||
51 | static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock, | |
52 | struct buffer_head *bh_result, int create) | |
53 | { | |
54 | int err = -EIO; | |
55 | int status; | |
56 | struct ocfs2_dinode *fe = NULL; | |
57 | struct buffer_head *bh = NULL; | |
58 | struct buffer_head *buffer_cache_bh = NULL; | |
59 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
60 | void *kaddr; | |
61 | ||
62 | mlog(0, "(0x%p, %llu, 0x%p, %d)\n", inode, | |
63 | (unsigned long long)iblock, bh_result, create); | |
64 | ||
65 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); | |
66 | ||
67 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { | |
68 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", | |
69 | (unsigned long long)iblock); | |
70 | goto bail; | |
71 | } | |
72 | ||
73 | status = ocfs2_read_inode_block(inode, &bh); | |
74 | if (status < 0) { | |
75 | mlog_errno(status); | |
76 | goto bail; | |
77 | } | |
78 | fe = (struct ocfs2_dinode *) bh->b_data; | |
79 | ||
80 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, | |
81 | le32_to_cpu(fe->i_clusters))) { | |
82 | mlog(ML_ERROR, "block offset is outside the allocated size: " | |
83 | "%llu\n", (unsigned long long)iblock); | |
84 | goto bail; | |
85 | } | |
86 | ||
87 | /* We don't use the page cache to create symlink data, so if | |
88 | * need be, copy it over from the buffer cache. */ | |
89 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { | |
90 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + | |
91 | iblock; | |
92 | buffer_cache_bh = sb_getblk(osb->sb, blkno); | |
93 | if (!buffer_cache_bh) { | |
94 | mlog(ML_ERROR, "couldn't getblock for symlink!\n"); | |
95 | goto bail; | |
96 | } | |
97 | ||
98 | /* we haven't locked out transactions, so a commit | |
99 | * could've happened. Since we've got a reference on | |
100 | * the bh, even if it commits while we're doing the | |
101 | * copy, the data is still good. */ | |
102 | if (buffer_jbd(buffer_cache_bh) | |
103 | && ocfs2_inode_is_new(inode)) { | |
104 | kaddr = kmap_atomic(bh_result->b_page, KM_USER0); | |
105 | if (!kaddr) { | |
106 | mlog(ML_ERROR, "couldn't kmap!\n"); | |
107 | goto bail; | |
108 | } | |
109 | memcpy(kaddr + (bh_result->b_size * iblock), | |
110 | buffer_cache_bh->b_data, | |
111 | bh_result->b_size); | |
112 | kunmap_atomic(kaddr, KM_USER0); | |
113 | set_buffer_uptodate(bh_result); | |
114 | } | |
115 | brelse(buffer_cache_bh); | |
116 | } | |
117 | ||
118 | map_bh(bh_result, inode->i_sb, | |
119 | le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); | |
120 | ||
121 | err = 0; | |
122 | ||
123 | bail: | |
124 | brelse(bh); | |
125 | ||
126 | mlog_exit(err); | |
127 | return err; | |
128 | } | |
129 | ||
130 | int ocfs2_get_block(struct inode *inode, sector_t iblock, | |
131 | struct buffer_head *bh_result, int create) | |
132 | { | |
133 | int err = 0; | |
134 | unsigned int ext_flags; | |
135 | u64 max_blocks = bh_result->b_size >> inode->i_blkbits; | |
136 | u64 p_blkno, count, past_eof; | |
137 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
138 | ||
139 | mlog(0, "(0x%p, %llu, 0x%p, %d)\n", inode, | |
140 | (unsigned long long)iblock, bh_result, create); | |
141 | ||
142 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) | |
143 | mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", | |
144 | inode, inode->i_ino); | |
145 | ||
146 | if (S_ISLNK(inode->i_mode)) { | |
147 | /* this always does I/O for some reason. */ | |
148 | err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); | |
149 | goto bail; | |
150 | } | |
151 | ||
152 | err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count, | |
153 | &ext_flags); | |
154 | if (err) { | |
155 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " | |
156 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, | |
157 | (unsigned long long)p_blkno); | |
158 | goto bail; | |
159 | } | |
160 | ||
161 | if (max_blocks < count) | |
162 | count = max_blocks; | |
163 | ||
164 | /* | |
165 | * ocfs2 never allocates in this function - the only time we | |
166 | * need to use BH_New is when we're extending i_size on a file | |
167 | * system which doesn't support holes, in which case BH_New | |
168 | * allows __block_write_begin() to zero. | |
169 | * | |
170 | * If we see this on a sparse file system, then a truncate has | |
171 | * raced us and removed the cluster. In this case, we clear | |
172 | * the buffers dirty and uptodate bits and let the buffer code | |
173 | * ignore it as a hole. | |
174 | */ | |
175 | if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) { | |
176 | clear_buffer_dirty(bh_result); | |
177 | clear_buffer_uptodate(bh_result); | |
178 | goto bail; | |
179 | } | |
180 | ||
181 | /* Treat the unwritten extent as a hole for zeroing purposes. */ | |
182 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
183 | map_bh(bh_result, inode->i_sb, p_blkno); | |
184 | ||
185 | bh_result->b_size = count << inode->i_blkbits; | |
186 | ||
187 | if (!ocfs2_sparse_alloc(osb)) { | |
188 | if (p_blkno == 0) { | |
189 | err = -EIO; | |
190 | mlog(ML_ERROR, | |
191 | "iblock = %llu p_blkno = %llu blkno=(%llu)\n", | |
192 | (unsigned long long)iblock, | |
193 | (unsigned long long)p_blkno, | |
194 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
195 | mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters); | |
196 | dump_stack(); | |
197 | goto bail; | |
198 | } | |
199 | } | |
200 | ||
201 | past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); | |
202 | mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino, | |
203 | (unsigned long long)past_eof); | |
204 | if (create && (iblock >= past_eof)) | |
205 | set_buffer_new(bh_result); | |
206 | ||
207 | bail: | |
208 | if (err < 0) | |
209 | err = -EIO; | |
210 | ||
211 | mlog_exit(err); | |
212 | return err; | |
213 | } | |
214 | ||
215 | int ocfs2_read_inline_data(struct inode *inode, struct page *page, | |
216 | struct buffer_head *di_bh) | |
217 | { | |
218 | void *kaddr; | |
219 | loff_t size; | |
220 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
221 | ||
222 | if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) { | |
223 | ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag", | |
224 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
225 | return -EROFS; | |
226 | } | |
227 | ||
228 | size = i_size_read(inode); | |
229 | ||
230 | if (size > PAGE_CACHE_SIZE || | |
231 | size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) { | |
232 | ocfs2_error(inode->i_sb, | |
233 | "Inode %llu has with inline data has bad size: %Lu", | |
234 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
235 | (unsigned long long)size); | |
236 | return -EROFS; | |
237 | } | |
238 | ||
239 | kaddr = kmap_atomic(page, KM_USER0); | |
240 | if (size) | |
241 | memcpy(kaddr, di->id2.i_data.id_data, size); | |
242 | /* Clear the remaining part of the page */ | |
243 | memset(kaddr + size, 0, PAGE_CACHE_SIZE - size); | |
244 | flush_dcache_page(page); | |
245 | kunmap_atomic(kaddr, KM_USER0); | |
246 | ||
247 | SetPageUptodate(page); | |
248 | ||
249 | return 0; | |
250 | } | |
251 | ||
252 | static int ocfs2_readpage_inline(struct inode *inode, struct page *page) | |
253 | { | |
254 | int ret; | |
255 | struct buffer_head *di_bh = NULL; | |
256 | ||
257 | BUG_ON(!PageLocked(page)); | |
258 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)); | |
259 | ||
260 | ret = ocfs2_read_inode_block(inode, &di_bh); | |
261 | if (ret) { | |
262 | mlog_errno(ret); | |
263 | goto out; | |
264 | } | |
265 | ||
266 | ret = ocfs2_read_inline_data(inode, page, di_bh); | |
267 | out: | |
268 | unlock_page(page); | |
269 | ||
270 | brelse(di_bh); | |
271 | return ret; | |
272 | } | |
273 | ||
274 | static int ocfs2_readpage(struct file *file, struct page *page) | |
275 | { | |
276 | struct inode *inode = page->mapping->host; | |
277 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
278 | loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT; | |
279 | int ret, unlock = 1; | |
280 | ||
281 | mlog(0, "(0x%p, %lu)\n", file, (page ? page->index : 0)); | |
282 | ||
283 | ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page); | |
284 | if (ret != 0) { | |
285 | if (ret == AOP_TRUNCATED_PAGE) | |
286 | unlock = 0; | |
287 | mlog_errno(ret); | |
288 | goto out; | |
289 | } | |
290 | ||
291 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { | |
292 | ret = AOP_TRUNCATED_PAGE; | |
293 | goto out_inode_unlock; | |
294 | } | |
295 | ||
296 | /* | |
297 | * i_size might have just been updated as we grabed the meta lock. We | |
298 | * might now be discovering a truncate that hit on another node. | |
299 | * block_read_full_page->get_block freaks out if it is asked to read | |
300 | * beyond the end of a file, so we check here. Callers | |
301 | * (generic_file_read, vm_ops->fault) are clever enough to check i_size | |
302 | * and notice that the page they just read isn't needed. | |
303 | * | |
304 | * XXX sys_readahead() seems to get that wrong? | |
305 | */ | |
306 | if (start >= i_size_read(inode)) { | |
307 | zero_user(page, 0, PAGE_SIZE); | |
308 | SetPageUptodate(page); | |
309 | ret = 0; | |
310 | goto out_alloc; | |
311 | } | |
312 | ||
313 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
314 | ret = ocfs2_readpage_inline(inode, page); | |
315 | else | |
316 | ret = block_read_full_page(page, ocfs2_get_block); | |
317 | unlock = 0; | |
318 | ||
319 | out_alloc: | |
320 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
321 | out_inode_unlock: | |
322 | ocfs2_inode_unlock(inode, 0); | |
323 | out: | |
324 | if (unlock) | |
325 | unlock_page(page); | |
326 | mlog_exit(ret); | |
327 | return ret; | |
328 | } | |
329 | ||
330 | /* | |
331 | * This is used only for read-ahead. Failures or difficult to handle | |
332 | * situations are safe to ignore. | |
333 | * | |
334 | * Right now, we don't bother with BH_Boundary - in-inode extent lists | |
335 | * are quite large (243 extents on 4k blocks), so most inodes don't | |
336 | * grow out to a tree. If need be, detecting boundary extents could | |
337 | * trivially be added in a future version of ocfs2_get_block(). | |
338 | */ | |
339 | static int ocfs2_readpages(struct file *filp, struct address_space *mapping, | |
340 | struct list_head *pages, unsigned nr_pages) | |
341 | { | |
342 | int ret, err = -EIO; | |
343 | struct inode *inode = mapping->host; | |
344 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
345 | loff_t start; | |
346 | struct page *last; | |
347 | ||
348 | /* | |
349 | * Use the nonblocking flag for the dlm code to avoid page | |
350 | * lock inversion, but don't bother with retrying. | |
351 | */ | |
352 | ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK); | |
353 | if (ret) | |
354 | return err; | |
355 | ||
356 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { | |
357 | ocfs2_inode_unlock(inode, 0); | |
358 | return err; | |
359 | } | |
360 | ||
361 | /* | |
362 | * Don't bother with inline-data. There isn't anything | |
363 | * to read-ahead in that case anyway... | |
364 | */ | |
365 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
366 | goto out_unlock; | |
367 | ||
368 | /* | |
369 | * Check whether a remote node truncated this file - we just | |
370 | * drop out in that case as it's not worth handling here. | |
371 | */ | |
372 | last = list_entry(pages->prev, struct page, lru); | |
373 | start = (loff_t)last->index << PAGE_CACHE_SHIFT; | |
374 | if (start >= i_size_read(inode)) | |
375 | goto out_unlock; | |
376 | ||
377 | err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block); | |
378 | ||
379 | out_unlock: | |
380 | up_read(&oi->ip_alloc_sem); | |
381 | ocfs2_inode_unlock(inode, 0); | |
382 | ||
383 | return err; | |
384 | } | |
385 | ||
386 | /* Note: Because we don't support holes, our allocation has | |
387 | * already happened (allocation writes zeros to the file data) | |
388 | * so we don't have to worry about ordered writes in | |
389 | * ocfs2_writepage. | |
390 | * | |
391 | * ->writepage is called during the process of invalidating the page cache | |
392 | * during blocked lock processing. It can't block on any cluster locks | |
393 | * to during block mapping. It's relying on the fact that the block | |
394 | * mapping can't have disappeared under the dirty pages that it is | |
395 | * being asked to write back. | |
396 | */ | |
397 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) | |
398 | { | |
399 | int ret; | |
400 | ||
401 | mlog(0, "(0x%p)\n", page); | |
402 | ||
403 | ret = block_write_full_page(page, ocfs2_get_block, wbc); | |
404 | ||
405 | mlog_exit(ret); | |
406 | ||
407 | return ret; | |
408 | } | |
409 | ||
410 | /* Taken from ext3. We don't necessarily need the full blown | |
411 | * functionality yet, but IMHO it's better to cut and paste the whole | |
412 | * thing so we can avoid introducing our own bugs (and easily pick up | |
413 | * their fixes when they happen) --Mark */ | |
414 | int walk_page_buffers( handle_t *handle, | |
415 | struct buffer_head *head, | |
416 | unsigned from, | |
417 | unsigned to, | |
418 | int *partial, | |
419 | int (*fn)( handle_t *handle, | |
420 | struct buffer_head *bh)) | |
421 | { | |
422 | struct buffer_head *bh; | |
423 | unsigned block_start, block_end; | |
424 | unsigned blocksize = head->b_size; | |
425 | int err, ret = 0; | |
426 | struct buffer_head *next; | |
427 | ||
428 | for ( bh = head, block_start = 0; | |
429 | ret == 0 && (bh != head || !block_start); | |
430 | block_start = block_end, bh = next) | |
431 | { | |
432 | next = bh->b_this_page; | |
433 | block_end = block_start + blocksize; | |
434 | if (block_end <= from || block_start >= to) { | |
435 | if (partial && !buffer_uptodate(bh)) | |
436 | *partial = 1; | |
437 | continue; | |
438 | } | |
439 | err = (*fn)(handle, bh); | |
440 | if (!ret) | |
441 | ret = err; | |
442 | } | |
443 | return ret; | |
444 | } | |
445 | ||
446 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) | |
447 | { | |
448 | sector_t status; | |
449 | u64 p_blkno = 0; | |
450 | int err = 0; | |
451 | struct inode *inode = mapping->host; | |
452 | ||
453 | mlog(0, "(block = %llu)\n", (unsigned long long)block); | |
454 | ||
455 | /* We don't need to lock journal system files, since they aren't | |
456 | * accessed concurrently from multiple nodes. | |
457 | */ | |
458 | if (!INODE_JOURNAL(inode)) { | |
459 | err = ocfs2_inode_lock(inode, NULL, 0); | |
460 | if (err) { | |
461 | if (err != -ENOENT) | |
462 | mlog_errno(err); | |
463 | goto bail; | |
464 | } | |
465 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | |
466 | } | |
467 | ||
468 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
469 | err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, | |
470 | NULL); | |
471 | ||
472 | if (!INODE_JOURNAL(inode)) { | |
473 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
474 | ocfs2_inode_unlock(inode, 0); | |
475 | } | |
476 | ||
477 | if (err) { | |
478 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", | |
479 | (unsigned long long)block); | |
480 | mlog_errno(err); | |
481 | goto bail; | |
482 | } | |
483 | ||
484 | bail: | |
485 | status = err ? 0 : p_blkno; | |
486 | ||
487 | mlog_exit((int)status); | |
488 | ||
489 | return status; | |
490 | } | |
491 | ||
492 | /* | |
493 | * TODO: Make this into a generic get_blocks function. | |
494 | * | |
495 | * From do_direct_io in direct-io.c: | |
496 | * "So what we do is to permit the ->get_blocks function to populate | |
497 | * bh.b_size with the size of IO which is permitted at this offset and | |
498 | * this i_blkbits." | |
499 | * | |
500 | * This function is called directly from get_more_blocks in direct-io.c. | |
501 | * | |
502 | * called like this: dio->get_blocks(dio->inode, fs_startblk, | |
503 | * fs_count, map_bh, dio->rw == WRITE); | |
504 | * | |
505 | * Note that we never bother to allocate blocks here, and thus ignore the | |
506 | * create argument. | |
507 | */ | |
508 | static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock, | |
509 | struct buffer_head *bh_result, int create) | |
510 | { | |
511 | int ret; | |
512 | u64 p_blkno, inode_blocks, contig_blocks; | |
513 | unsigned int ext_flags; | |
514 | unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits; | |
515 | unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits; | |
516 | ||
517 | /* This function won't even be called if the request isn't all | |
518 | * nicely aligned and of the right size, so there's no need | |
519 | * for us to check any of that. */ | |
520 | ||
521 | inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); | |
522 | ||
523 | /* This figures out the size of the next contiguous block, and | |
524 | * our logical offset */ | |
525 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, | |
526 | &contig_blocks, &ext_flags); | |
527 | if (ret) { | |
528 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", | |
529 | (unsigned long long)iblock); | |
530 | ret = -EIO; | |
531 | goto bail; | |
532 | } | |
533 | ||
534 | /* We should already CoW the refcounted extent in case of create. */ | |
535 | BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED)); | |
536 | ||
537 | /* | |
538 | * get_more_blocks() expects us to describe a hole by clearing | |
539 | * the mapped bit on bh_result(). | |
540 | * | |
541 | * Consider an unwritten extent as a hole. | |
542 | */ | |
543 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
544 | map_bh(bh_result, inode->i_sb, p_blkno); | |
545 | else | |
546 | clear_buffer_mapped(bh_result); | |
547 | ||
548 | /* make sure we don't map more than max_blocks blocks here as | |
549 | that's all the kernel will handle at this point. */ | |
550 | if (max_blocks < contig_blocks) | |
551 | contig_blocks = max_blocks; | |
552 | bh_result->b_size = contig_blocks << blocksize_bits; | |
553 | bail: | |
554 | return ret; | |
555 | } | |
556 | ||
557 | /* | |
558 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're | |
559 | * particularly interested in the aio/dio case. Like the core uses | |
560 | * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from | |
561 | * truncation on another. | |
562 | */ | |
563 | static void ocfs2_dio_end_io(struct kiocb *iocb, | |
564 | loff_t offset, | |
565 | ssize_t bytes, | |
566 | void *private, | |
567 | int ret, | |
568 | bool is_async) | |
569 | { | |
570 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; | |
571 | int level; | |
572 | ||
573 | /* this io's submitter should not have unlocked this before we could */ | |
574 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); | |
575 | ||
576 | if (ocfs2_iocb_is_sem_locked(iocb)) { | |
577 | up_read(&inode->i_alloc_sem); | |
578 | ocfs2_iocb_clear_sem_locked(iocb); | |
579 | } | |
580 | ||
581 | ocfs2_iocb_clear_rw_locked(iocb); | |
582 | ||
583 | level = ocfs2_iocb_rw_locked_level(iocb); | |
584 | ocfs2_rw_unlock(inode, level); | |
585 | ||
586 | if (is_async) | |
587 | aio_complete(iocb, ret, 0); | |
588 | } | |
589 | ||
590 | /* | |
591 | * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen | |
592 | * from ext3. PageChecked() bits have been removed as OCFS2 does not | |
593 | * do journalled data. | |
594 | */ | |
595 | static void ocfs2_invalidatepage(struct page *page, unsigned long offset) | |
596 | { | |
597 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | |
598 | ||
599 | jbd2_journal_invalidatepage(journal, page, offset); | |
600 | } | |
601 | ||
602 | static int ocfs2_releasepage(struct page *page, gfp_t wait) | |
603 | { | |
604 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | |
605 | ||
606 | if (!page_has_buffers(page)) | |
607 | return 0; | |
608 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
609 | } | |
610 | ||
611 | static ssize_t ocfs2_direct_IO(int rw, | |
612 | struct kiocb *iocb, | |
613 | const struct iovec *iov, | |
614 | loff_t offset, | |
615 | unsigned long nr_segs) | |
616 | { | |
617 | struct file *file = iocb->ki_filp; | |
618 | struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host; | |
619 | int ret; | |
620 | ||
621 | /* | |
622 | * Fallback to buffered I/O if we see an inode without | |
623 | * extents. | |
624 | */ | |
625 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
626 | return 0; | |
627 | ||
628 | /* Fallback to buffered I/O if we are appending. */ | |
629 | if (i_size_read(inode) <= offset) | |
630 | return 0; | |
631 | ||
632 | ret = __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, | |
633 | iov, offset, nr_segs, | |
634 | ocfs2_direct_IO_get_blocks, | |
635 | ocfs2_dio_end_io, NULL, 0); | |
636 | ||
637 | mlog_exit(ret); | |
638 | return ret; | |
639 | } | |
640 | ||
641 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, | |
642 | u32 cpos, | |
643 | unsigned int *start, | |
644 | unsigned int *end) | |
645 | { | |
646 | unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE; | |
647 | ||
648 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) { | |
649 | unsigned int cpp; | |
650 | ||
651 | cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits); | |
652 | ||
653 | cluster_start = cpos % cpp; | |
654 | cluster_start = cluster_start << osb->s_clustersize_bits; | |
655 | ||
656 | cluster_end = cluster_start + osb->s_clustersize; | |
657 | } | |
658 | ||
659 | BUG_ON(cluster_start > PAGE_SIZE); | |
660 | BUG_ON(cluster_end > PAGE_SIZE); | |
661 | ||
662 | if (start) | |
663 | *start = cluster_start; | |
664 | if (end) | |
665 | *end = cluster_end; | |
666 | } | |
667 | ||
668 | /* | |
669 | * 'from' and 'to' are the region in the page to avoid zeroing. | |
670 | * | |
671 | * If pagesize > clustersize, this function will avoid zeroing outside | |
672 | * of the cluster boundary. | |
673 | * | |
674 | * from == to == 0 is code for "zero the entire cluster region" | |
675 | */ | |
676 | static void ocfs2_clear_page_regions(struct page *page, | |
677 | struct ocfs2_super *osb, u32 cpos, | |
678 | unsigned from, unsigned to) | |
679 | { | |
680 | void *kaddr; | |
681 | unsigned int cluster_start, cluster_end; | |
682 | ||
683 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); | |
684 | ||
685 | kaddr = kmap_atomic(page, KM_USER0); | |
686 | ||
687 | if (from || to) { | |
688 | if (from > cluster_start) | |
689 | memset(kaddr + cluster_start, 0, from - cluster_start); | |
690 | if (to < cluster_end) | |
691 | memset(kaddr + to, 0, cluster_end - to); | |
692 | } else { | |
693 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); | |
694 | } | |
695 | ||
696 | kunmap_atomic(kaddr, KM_USER0); | |
697 | } | |
698 | ||
699 | /* | |
700 | * Nonsparse file systems fully allocate before we get to the write | |
701 | * code. This prevents ocfs2_write() from tagging the write as an | |
702 | * allocating one, which means ocfs2_map_page_blocks() might try to | |
703 | * read-in the blocks at the tail of our file. Avoid reading them by | |
704 | * testing i_size against each block offset. | |
705 | */ | |
706 | static int ocfs2_should_read_blk(struct inode *inode, struct page *page, | |
707 | unsigned int block_start) | |
708 | { | |
709 | u64 offset = page_offset(page) + block_start; | |
710 | ||
711 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
712 | return 1; | |
713 | ||
714 | if (i_size_read(inode) > offset) | |
715 | return 1; | |
716 | ||
717 | return 0; | |
718 | } | |
719 | ||
720 | /* | |
721 | * Some of this taken from __block_write_begin(). We already have our | |
722 | * mapping by now though, and the entire write will be allocating or | |
723 | * it won't, so not much need to use BH_New. | |
724 | * | |
725 | * This will also skip zeroing, which is handled externally. | |
726 | */ | |
727 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, | |
728 | struct inode *inode, unsigned int from, | |
729 | unsigned int to, int new) | |
730 | { | |
731 | int ret = 0; | |
732 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; | |
733 | unsigned int block_end, block_start; | |
734 | unsigned int bsize = 1 << inode->i_blkbits; | |
735 | ||
736 | if (!page_has_buffers(page)) | |
737 | create_empty_buffers(page, bsize, 0); | |
738 | ||
739 | head = page_buffers(page); | |
740 | for (bh = head, block_start = 0; bh != head || !block_start; | |
741 | bh = bh->b_this_page, block_start += bsize) { | |
742 | block_end = block_start + bsize; | |
743 | ||
744 | clear_buffer_new(bh); | |
745 | ||
746 | /* | |
747 | * Ignore blocks outside of our i/o range - | |
748 | * they may belong to unallocated clusters. | |
749 | */ | |
750 | if (block_start >= to || block_end <= from) { | |
751 | if (PageUptodate(page)) | |
752 | set_buffer_uptodate(bh); | |
753 | continue; | |
754 | } | |
755 | ||
756 | /* | |
757 | * For an allocating write with cluster size >= page | |
758 | * size, we always write the entire page. | |
759 | */ | |
760 | if (new) | |
761 | set_buffer_new(bh); | |
762 | ||
763 | if (!buffer_mapped(bh)) { | |
764 | map_bh(bh, inode->i_sb, *p_blkno); | |
765 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
766 | } | |
767 | ||
768 | if (PageUptodate(page)) { | |
769 | if (!buffer_uptodate(bh)) | |
770 | set_buffer_uptodate(bh); | |
771 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
772 | !buffer_new(bh) && | |
773 | ocfs2_should_read_blk(inode, page, block_start) && | |
774 | (block_start < from || block_end > to)) { | |
775 | ll_rw_block(READ, 1, &bh); | |
776 | *wait_bh++=bh; | |
777 | } | |
778 | ||
779 | *p_blkno = *p_blkno + 1; | |
780 | } | |
781 | ||
782 | /* | |
783 | * If we issued read requests - let them complete. | |
784 | */ | |
785 | while(wait_bh > wait) { | |
786 | wait_on_buffer(*--wait_bh); | |
787 | if (!buffer_uptodate(*wait_bh)) | |
788 | ret = -EIO; | |
789 | } | |
790 | ||
791 | if (ret == 0 || !new) | |
792 | return ret; | |
793 | ||
794 | /* | |
795 | * If we get -EIO above, zero out any newly allocated blocks | |
796 | * to avoid exposing stale data. | |
797 | */ | |
798 | bh = head; | |
799 | block_start = 0; | |
800 | do { | |
801 | block_end = block_start + bsize; | |
802 | if (block_end <= from) | |
803 | goto next_bh; | |
804 | if (block_start >= to) | |
805 | break; | |
806 | ||
807 | zero_user(page, block_start, bh->b_size); | |
808 | set_buffer_uptodate(bh); | |
809 | mark_buffer_dirty(bh); | |
810 | ||
811 | next_bh: | |
812 | block_start = block_end; | |
813 | bh = bh->b_this_page; | |
814 | } while (bh != head); | |
815 | ||
816 | return ret; | |
817 | } | |
818 | ||
819 | #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE) | |
820 | #define OCFS2_MAX_CTXT_PAGES 1 | |
821 | #else | |
822 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE) | |
823 | #endif | |
824 | ||
825 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE) | |
826 | ||
827 | /* | |
828 | * Describe the state of a single cluster to be written to. | |
829 | */ | |
830 | struct ocfs2_write_cluster_desc { | |
831 | u32 c_cpos; | |
832 | u32 c_phys; | |
833 | /* | |
834 | * Give this a unique field because c_phys eventually gets | |
835 | * filled. | |
836 | */ | |
837 | unsigned c_new; | |
838 | unsigned c_unwritten; | |
839 | unsigned c_needs_zero; | |
840 | }; | |
841 | ||
842 | struct ocfs2_write_ctxt { | |
843 | /* Logical cluster position / len of write */ | |
844 | u32 w_cpos; | |
845 | u32 w_clen; | |
846 | ||
847 | /* First cluster allocated in a nonsparse extend */ | |
848 | u32 w_first_new_cpos; | |
849 | ||
850 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; | |
851 | ||
852 | /* | |
853 | * This is true if page_size > cluster_size. | |
854 | * | |
855 | * It triggers a set of special cases during write which might | |
856 | * have to deal with allocating writes to partial pages. | |
857 | */ | |
858 | unsigned int w_large_pages; | |
859 | ||
860 | /* | |
861 | * Pages involved in this write. | |
862 | * | |
863 | * w_target_page is the page being written to by the user. | |
864 | * | |
865 | * w_pages is an array of pages which always contains | |
866 | * w_target_page, and in the case of an allocating write with | |
867 | * page_size < cluster size, it will contain zero'd and mapped | |
868 | * pages adjacent to w_target_page which need to be written | |
869 | * out in so that future reads from that region will get | |
870 | * zero's. | |
871 | */ | |
872 | unsigned int w_num_pages; | |
873 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; | |
874 | struct page *w_target_page; | |
875 | ||
876 | /* | |
877 | * ocfs2_write_end() uses this to know what the real range to | |
878 | * write in the target should be. | |
879 | */ | |
880 | unsigned int w_target_from; | |
881 | unsigned int w_target_to; | |
882 | ||
883 | /* | |
884 | * We could use journal_current_handle() but this is cleaner, | |
885 | * IMHO -Mark | |
886 | */ | |
887 | handle_t *w_handle; | |
888 | ||
889 | struct buffer_head *w_di_bh; | |
890 | ||
891 | struct ocfs2_cached_dealloc_ctxt w_dealloc; | |
892 | }; | |
893 | ||
894 | void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages) | |
895 | { | |
896 | int i; | |
897 | ||
898 | for(i = 0; i < num_pages; i++) { | |
899 | if (pages[i]) { | |
900 | unlock_page(pages[i]); | |
901 | mark_page_accessed(pages[i]); | |
902 | page_cache_release(pages[i]); | |
903 | } | |
904 | } | |
905 | } | |
906 | ||
907 | static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc) | |
908 | { | |
909 | ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages); | |
910 | ||
911 | brelse(wc->w_di_bh); | |
912 | kfree(wc); | |
913 | } | |
914 | ||
915 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, | |
916 | struct ocfs2_super *osb, loff_t pos, | |
917 | unsigned len, struct buffer_head *di_bh) | |
918 | { | |
919 | u32 cend; | |
920 | struct ocfs2_write_ctxt *wc; | |
921 | ||
922 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); | |
923 | if (!wc) | |
924 | return -ENOMEM; | |
925 | ||
926 | wc->w_cpos = pos >> osb->s_clustersize_bits; | |
927 | wc->w_first_new_cpos = UINT_MAX; | |
928 | cend = (pos + len - 1) >> osb->s_clustersize_bits; | |
929 | wc->w_clen = cend - wc->w_cpos + 1; | |
930 | get_bh(di_bh); | |
931 | wc->w_di_bh = di_bh; | |
932 | ||
933 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) | |
934 | wc->w_large_pages = 1; | |
935 | else | |
936 | wc->w_large_pages = 0; | |
937 | ||
938 | ocfs2_init_dealloc_ctxt(&wc->w_dealloc); | |
939 | ||
940 | *wcp = wc; | |
941 | ||
942 | return 0; | |
943 | } | |
944 | ||
945 | /* | |
946 | * If a page has any new buffers, zero them out here, and mark them uptodate | |
947 | * and dirty so they'll be written out (in order to prevent uninitialised | |
948 | * block data from leaking). And clear the new bit. | |
949 | */ | |
950 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
951 | { | |
952 | unsigned int block_start, block_end; | |
953 | struct buffer_head *head, *bh; | |
954 | ||
955 | BUG_ON(!PageLocked(page)); | |
956 | if (!page_has_buffers(page)) | |
957 | return; | |
958 | ||
959 | bh = head = page_buffers(page); | |
960 | block_start = 0; | |
961 | do { | |
962 | block_end = block_start + bh->b_size; | |
963 | ||
964 | if (buffer_new(bh)) { | |
965 | if (block_end > from && block_start < to) { | |
966 | if (!PageUptodate(page)) { | |
967 | unsigned start, end; | |
968 | ||
969 | start = max(from, block_start); | |
970 | end = min(to, block_end); | |
971 | ||
972 | zero_user_segment(page, start, end); | |
973 | set_buffer_uptodate(bh); | |
974 | } | |
975 | ||
976 | clear_buffer_new(bh); | |
977 | mark_buffer_dirty(bh); | |
978 | } | |
979 | } | |
980 | ||
981 | block_start = block_end; | |
982 | bh = bh->b_this_page; | |
983 | } while (bh != head); | |
984 | } | |
985 | ||
986 | /* | |
987 | * Only called when we have a failure during allocating write to write | |
988 | * zero's to the newly allocated region. | |
989 | */ | |
990 | static void ocfs2_write_failure(struct inode *inode, | |
991 | struct ocfs2_write_ctxt *wc, | |
992 | loff_t user_pos, unsigned user_len) | |
993 | { | |
994 | int i; | |
995 | unsigned from = user_pos & (PAGE_CACHE_SIZE - 1), | |
996 | to = user_pos + user_len; | |
997 | struct page *tmppage; | |
998 | ||
999 | ocfs2_zero_new_buffers(wc->w_target_page, from, to); | |
1000 | ||
1001 | for(i = 0; i < wc->w_num_pages; i++) { | |
1002 | tmppage = wc->w_pages[i]; | |
1003 | ||
1004 | if (page_has_buffers(tmppage)) { | |
1005 | if (ocfs2_should_order_data(inode)) | |
1006 | ocfs2_jbd2_file_inode(wc->w_handle, inode); | |
1007 | ||
1008 | block_commit_write(tmppage, from, to); | |
1009 | } | |
1010 | } | |
1011 | } | |
1012 | ||
1013 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, | |
1014 | struct ocfs2_write_ctxt *wc, | |
1015 | struct page *page, u32 cpos, | |
1016 | loff_t user_pos, unsigned user_len, | |
1017 | int new) | |
1018 | { | |
1019 | int ret; | |
1020 | unsigned int map_from = 0, map_to = 0; | |
1021 | unsigned int cluster_start, cluster_end; | |
1022 | unsigned int user_data_from = 0, user_data_to = 0; | |
1023 | ||
1024 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, | |
1025 | &cluster_start, &cluster_end); | |
1026 | ||
1027 | if (page == wc->w_target_page) { | |
1028 | map_from = user_pos & (PAGE_CACHE_SIZE - 1); | |
1029 | map_to = map_from + user_len; | |
1030 | ||
1031 | if (new) | |
1032 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1033 | cluster_start, cluster_end, | |
1034 | new); | |
1035 | else | |
1036 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1037 | map_from, map_to, new); | |
1038 | if (ret) { | |
1039 | mlog_errno(ret); | |
1040 | goto out; | |
1041 | } | |
1042 | ||
1043 | user_data_from = map_from; | |
1044 | user_data_to = map_to; | |
1045 | if (new) { | |
1046 | map_from = cluster_start; | |
1047 | map_to = cluster_end; | |
1048 | } | |
1049 | } else { | |
1050 | /* | |
1051 | * If we haven't allocated the new page yet, we | |
1052 | * shouldn't be writing it out without copying user | |
1053 | * data. This is likely a math error from the caller. | |
1054 | */ | |
1055 | BUG_ON(!new); | |
1056 | ||
1057 | map_from = cluster_start; | |
1058 | map_to = cluster_end; | |
1059 | ||
1060 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1061 | cluster_start, cluster_end, new); | |
1062 | if (ret) { | |
1063 | mlog_errno(ret); | |
1064 | goto out; | |
1065 | } | |
1066 | } | |
1067 | ||
1068 | /* | |
1069 | * Parts of newly allocated pages need to be zero'd. | |
1070 | * | |
1071 | * Above, we have also rewritten 'to' and 'from' - as far as | |
1072 | * the rest of the function is concerned, the entire cluster | |
1073 | * range inside of a page needs to be written. | |
1074 | * | |
1075 | * We can skip this if the page is up to date - it's already | |
1076 | * been zero'd from being read in as a hole. | |
1077 | */ | |
1078 | if (new && !PageUptodate(page)) | |
1079 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), | |
1080 | cpos, user_data_from, user_data_to); | |
1081 | ||
1082 | flush_dcache_page(page); | |
1083 | ||
1084 | out: | |
1085 | return ret; | |
1086 | } | |
1087 | ||
1088 | /* | |
1089 | * This function will only grab one clusters worth of pages. | |
1090 | */ | |
1091 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, | |
1092 | struct ocfs2_write_ctxt *wc, | |
1093 | u32 cpos, loff_t user_pos, | |
1094 | unsigned user_len, int new, | |
1095 | struct page *mmap_page) | |
1096 | { | |
1097 | int ret = 0, i; | |
1098 | unsigned long start, target_index, end_index, index; | |
1099 | struct inode *inode = mapping->host; | |
1100 | loff_t last_byte; | |
1101 | ||
1102 | target_index = user_pos >> PAGE_CACHE_SHIFT; | |
1103 | ||
1104 | /* | |
1105 | * Figure out how many pages we'll be manipulating here. For | |
1106 | * non allocating write, we just change the one | |
1107 | * page. Otherwise, we'll need a whole clusters worth. If we're | |
1108 | * writing past i_size, we only need enough pages to cover the | |
1109 | * last page of the write. | |
1110 | */ | |
1111 | if (new) { | |
1112 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); | |
1113 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); | |
1114 | /* | |
1115 | * We need the index *past* the last page we could possibly | |
1116 | * touch. This is the page past the end of the write or | |
1117 | * i_size, whichever is greater. | |
1118 | */ | |
1119 | last_byte = max(user_pos + user_len, i_size_read(inode)); | |
1120 | BUG_ON(last_byte < 1); | |
1121 | end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1; | |
1122 | if ((start + wc->w_num_pages) > end_index) | |
1123 | wc->w_num_pages = end_index - start; | |
1124 | } else { | |
1125 | wc->w_num_pages = 1; | |
1126 | start = target_index; | |
1127 | } | |
1128 | ||
1129 | for(i = 0; i < wc->w_num_pages; i++) { | |
1130 | index = start + i; | |
1131 | ||
1132 | if (index == target_index && mmap_page) { | |
1133 | /* | |
1134 | * ocfs2_pagemkwrite() is a little different | |
1135 | * and wants us to directly use the page | |
1136 | * passed in. | |
1137 | */ | |
1138 | lock_page(mmap_page); | |
1139 | ||
1140 | if (mmap_page->mapping != mapping) { | |
1141 | unlock_page(mmap_page); | |
1142 | /* | |
1143 | * Sanity check - the locking in | |
1144 | * ocfs2_pagemkwrite() should ensure | |
1145 | * that this code doesn't trigger. | |
1146 | */ | |
1147 | ret = -EINVAL; | |
1148 | mlog_errno(ret); | |
1149 | goto out; | |
1150 | } | |
1151 | ||
1152 | page_cache_get(mmap_page); | |
1153 | wc->w_pages[i] = mmap_page; | |
1154 | } else { | |
1155 | wc->w_pages[i] = find_or_create_page(mapping, index, | |
1156 | GFP_NOFS); | |
1157 | if (!wc->w_pages[i]) { | |
1158 | ret = -ENOMEM; | |
1159 | mlog_errno(ret); | |
1160 | goto out; | |
1161 | } | |
1162 | } | |
1163 | ||
1164 | if (index == target_index) | |
1165 | wc->w_target_page = wc->w_pages[i]; | |
1166 | } | |
1167 | out: | |
1168 | return ret; | |
1169 | } | |
1170 | ||
1171 | /* | |
1172 | * Prepare a single cluster for write one cluster into the file. | |
1173 | */ | |
1174 | static int ocfs2_write_cluster(struct address_space *mapping, | |
1175 | u32 phys, unsigned int unwritten, | |
1176 | unsigned int should_zero, | |
1177 | struct ocfs2_alloc_context *data_ac, | |
1178 | struct ocfs2_alloc_context *meta_ac, | |
1179 | struct ocfs2_write_ctxt *wc, u32 cpos, | |
1180 | loff_t user_pos, unsigned user_len) | |
1181 | { | |
1182 | int ret, i, new; | |
1183 | u64 v_blkno, p_blkno; | |
1184 | struct inode *inode = mapping->host; | |
1185 | struct ocfs2_extent_tree et; | |
1186 | ||
1187 | new = phys == 0 ? 1 : 0; | |
1188 | if (new) { | |
1189 | u32 tmp_pos; | |
1190 | ||
1191 | /* | |
1192 | * This is safe to call with the page locks - it won't take | |
1193 | * any additional semaphores or cluster locks. | |
1194 | */ | |
1195 | tmp_pos = cpos; | |
1196 | ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode, | |
1197 | &tmp_pos, 1, 0, wc->w_di_bh, | |
1198 | wc->w_handle, data_ac, | |
1199 | meta_ac, NULL); | |
1200 | /* | |
1201 | * This shouldn't happen because we must have already | |
1202 | * calculated the correct meta data allocation required. The | |
1203 | * internal tree allocation code should know how to increase | |
1204 | * transaction credits itself. | |
1205 | * | |
1206 | * If need be, we could handle -EAGAIN for a | |
1207 | * RESTART_TRANS here. | |
1208 | */ | |
1209 | mlog_bug_on_msg(ret == -EAGAIN, | |
1210 | "Inode %llu: EAGAIN return during allocation.\n", | |
1211 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1212 | if (ret < 0) { | |
1213 | mlog_errno(ret); | |
1214 | goto out; | |
1215 | } | |
1216 | } else if (unwritten) { | |
1217 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), | |
1218 | wc->w_di_bh); | |
1219 | ret = ocfs2_mark_extent_written(inode, &et, | |
1220 | wc->w_handle, cpos, 1, phys, | |
1221 | meta_ac, &wc->w_dealloc); | |
1222 | if (ret < 0) { | |
1223 | mlog_errno(ret); | |
1224 | goto out; | |
1225 | } | |
1226 | } | |
1227 | ||
1228 | if (should_zero) | |
1229 | v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos); | |
1230 | else | |
1231 | v_blkno = user_pos >> inode->i_sb->s_blocksize_bits; | |
1232 | ||
1233 | /* | |
1234 | * The only reason this should fail is due to an inability to | |
1235 | * find the extent added. | |
1236 | */ | |
1237 | ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL, | |
1238 | NULL); | |
1239 | if (ret < 0) { | |
1240 | ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, " | |
1241 | "at logical block %llu", | |
1242 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1243 | (unsigned long long)v_blkno); | |
1244 | goto out; | |
1245 | } | |
1246 | ||
1247 | BUG_ON(p_blkno == 0); | |
1248 | ||
1249 | for(i = 0; i < wc->w_num_pages; i++) { | |
1250 | int tmpret; | |
1251 | ||
1252 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, | |
1253 | wc->w_pages[i], cpos, | |
1254 | user_pos, user_len, | |
1255 | should_zero); | |
1256 | if (tmpret) { | |
1257 | mlog_errno(tmpret); | |
1258 | if (ret == 0) | |
1259 | ret = tmpret; | |
1260 | } | |
1261 | } | |
1262 | ||
1263 | /* | |
1264 | * We only have cleanup to do in case of allocating write. | |
1265 | */ | |
1266 | if (ret && new) | |
1267 | ocfs2_write_failure(inode, wc, user_pos, user_len); | |
1268 | ||
1269 | out: | |
1270 | ||
1271 | return ret; | |
1272 | } | |
1273 | ||
1274 | static int ocfs2_write_cluster_by_desc(struct address_space *mapping, | |
1275 | struct ocfs2_alloc_context *data_ac, | |
1276 | struct ocfs2_alloc_context *meta_ac, | |
1277 | struct ocfs2_write_ctxt *wc, | |
1278 | loff_t pos, unsigned len) | |
1279 | { | |
1280 | int ret, i; | |
1281 | loff_t cluster_off; | |
1282 | unsigned int local_len = len; | |
1283 | struct ocfs2_write_cluster_desc *desc; | |
1284 | struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb); | |
1285 | ||
1286 | for (i = 0; i < wc->w_clen; i++) { | |
1287 | desc = &wc->w_desc[i]; | |
1288 | ||
1289 | /* | |
1290 | * We have to make sure that the total write passed in | |
1291 | * doesn't extend past a single cluster. | |
1292 | */ | |
1293 | local_len = len; | |
1294 | cluster_off = pos & (osb->s_clustersize - 1); | |
1295 | if ((cluster_off + local_len) > osb->s_clustersize) | |
1296 | local_len = osb->s_clustersize - cluster_off; | |
1297 | ||
1298 | ret = ocfs2_write_cluster(mapping, desc->c_phys, | |
1299 | desc->c_unwritten, | |
1300 | desc->c_needs_zero, | |
1301 | data_ac, meta_ac, | |
1302 | wc, desc->c_cpos, pos, local_len); | |
1303 | if (ret) { | |
1304 | mlog_errno(ret); | |
1305 | goto out; | |
1306 | } | |
1307 | ||
1308 | len -= local_len; | |
1309 | pos += local_len; | |
1310 | } | |
1311 | ||
1312 | ret = 0; | |
1313 | out: | |
1314 | return ret; | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * ocfs2_write_end() wants to know which parts of the target page it | |
1319 | * should complete the write on. It's easiest to compute them ahead of | |
1320 | * time when a more complete view of the write is available. | |
1321 | */ | |
1322 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, | |
1323 | struct ocfs2_write_ctxt *wc, | |
1324 | loff_t pos, unsigned len, int alloc) | |
1325 | { | |
1326 | struct ocfs2_write_cluster_desc *desc; | |
1327 | ||
1328 | wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1); | |
1329 | wc->w_target_to = wc->w_target_from + len; | |
1330 | ||
1331 | if (alloc == 0) | |
1332 | return; | |
1333 | ||
1334 | /* | |
1335 | * Allocating write - we may have different boundaries based | |
1336 | * on page size and cluster size. | |
1337 | * | |
1338 | * NOTE: We can no longer compute one value from the other as | |
1339 | * the actual write length and user provided length may be | |
1340 | * different. | |
1341 | */ | |
1342 | ||
1343 | if (wc->w_large_pages) { | |
1344 | /* | |
1345 | * We only care about the 1st and last cluster within | |
1346 | * our range and whether they should be zero'd or not. Either | |
1347 | * value may be extended out to the start/end of a | |
1348 | * newly allocated cluster. | |
1349 | */ | |
1350 | desc = &wc->w_desc[0]; | |
1351 | if (desc->c_needs_zero) | |
1352 | ocfs2_figure_cluster_boundaries(osb, | |
1353 | desc->c_cpos, | |
1354 | &wc->w_target_from, | |
1355 | NULL); | |
1356 | ||
1357 | desc = &wc->w_desc[wc->w_clen - 1]; | |
1358 | if (desc->c_needs_zero) | |
1359 | ocfs2_figure_cluster_boundaries(osb, | |
1360 | desc->c_cpos, | |
1361 | NULL, | |
1362 | &wc->w_target_to); | |
1363 | } else { | |
1364 | wc->w_target_from = 0; | |
1365 | wc->w_target_to = PAGE_CACHE_SIZE; | |
1366 | } | |
1367 | } | |
1368 | ||
1369 | /* | |
1370 | * Populate each single-cluster write descriptor in the write context | |
1371 | * with information about the i/o to be done. | |
1372 | * | |
1373 | * Returns the number of clusters that will have to be allocated, as | |
1374 | * well as a worst case estimate of the number of extent records that | |
1375 | * would have to be created during a write to an unwritten region. | |
1376 | */ | |
1377 | static int ocfs2_populate_write_desc(struct inode *inode, | |
1378 | struct ocfs2_write_ctxt *wc, | |
1379 | unsigned int *clusters_to_alloc, | |
1380 | unsigned int *extents_to_split) | |
1381 | { | |
1382 | int ret; | |
1383 | struct ocfs2_write_cluster_desc *desc; | |
1384 | unsigned int num_clusters = 0; | |
1385 | unsigned int ext_flags = 0; | |
1386 | u32 phys = 0; | |
1387 | int i; | |
1388 | ||
1389 | *clusters_to_alloc = 0; | |
1390 | *extents_to_split = 0; | |
1391 | ||
1392 | for (i = 0; i < wc->w_clen; i++) { | |
1393 | desc = &wc->w_desc[i]; | |
1394 | desc->c_cpos = wc->w_cpos + i; | |
1395 | ||
1396 | if (num_clusters == 0) { | |
1397 | /* | |
1398 | * Need to look up the next extent record. | |
1399 | */ | |
1400 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, | |
1401 | &num_clusters, &ext_flags); | |
1402 | if (ret) { | |
1403 | mlog_errno(ret); | |
1404 | goto out; | |
1405 | } | |
1406 | ||
1407 | /* We should already CoW the refcountd extent. */ | |
1408 | BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED); | |
1409 | ||
1410 | /* | |
1411 | * Assume worst case - that we're writing in | |
1412 | * the middle of the extent. | |
1413 | * | |
1414 | * We can assume that the write proceeds from | |
1415 | * left to right, in which case the extent | |
1416 | * insert code is smart enough to coalesce the | |
1417 | * next splits into the previous records created. | |
1418 | */ | |
1419 | if (ext_flags & OCFS2_EXT_UNWRITTEN) | |
1420 | *extents_to_split = *extents_to_split + 2; | |
1421 | } else if (phys) { | |
1422 | /* | |
1423 | * Only increment phys if it doesn't describe | |
1424 | * a hole. | |
1425 | */ | |
1426 | phys++; | |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * If w_first_new_cpos is < UINT_MAX, we have a non-sparse | |
1431 | * file that got extended. w_first_new_cpos tells us | |
1432 | * where the newly allocated clusters are so we can | |
1433 | * zero them. | |
1434 | */ | |
1435 | if (desc->c_cpos >= wc->w_first_new_cpos) { | |
1436 | BUG_ON(phys == 0); | |
1437 | desc->c_needs_zero = 1; | |
1438 | } | |
1439 | ||
1440 | desc->c_phys = phys; | |
1441 | if (phys == 0) { | |
1442 | desc->c_new = 1; | |
1443 | desc->c_needs_zero = 1; | |
1444 | *clusters_to_alloc = *clusters_to_alloc + 1; | |
1445 | } | |
1446 | ||
1447 | if (ext_flags & OCFS2_EXT_UNWRITTEN) { | |
1448 | desc->c_unwritten = 1; | |
1449 | desc->c_needs_zero = 1; | |
1450 | } | |
1451 | ||
1452 | num_clusters--; | |
1453 | } | |
1454 | ||
1455 | ret = 0; | |
1456 | out: | |
1457 | return ret; | |
1458 | } | |
1459 | ||
1460 | static int ocfs2_write_begin_inline(struct address_space *mapping, | |
1461 | struct inode *inode, | |
1462 | struct ocfs2_write_ctxt *wc) | |
1463 | { | |
1464 | int ret; | |
1465 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1466 | struct page *page; | |
1467 | handle_t *handle; | |
1468 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1469 | ||
1470 | page = find_or_create_page(mapping, 0, GFP_NOFS); | |
1471 | if (!page) { | |
1472 | ret = -ENOMEM; | |
1473 | mlog_errno(ret); | |
1474 | goto out; | |
1475 | } | |
1476 | /* | |
1477 | * If we don't set w_num_pages then this page won't get unlocked | |
1478 | * and freed on cleanup of the write context. | |
1479 | */ | |
1480 | wc->w_pages[0] = wc->w_target_page = page; | |
1481 | wc->w_num_pages = 1; | |
1482 | ||
1483 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1484 | if (IS_ERR(handle)) { | |
1485 | ret = PTR_ERR(handle); | |
1486 | mlog_errno(ret); | |
1487 | goto out; | |
1488 | } | |
1489 | ||
1490 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, | |
1491 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1492 | if (ret) { | |
1493 | ocfs2_commit_trans(osb, handle); | |
1494 | ||
1495 | mlog_errno(ret); | |
1496 | goto out; | |
1497 | } | |
1498 | ||
1499 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
1500 | ocfs2_set_inode_data_inline(inode, di); | |
1501 | ||
1502 | if (!PageUptodate(page)) { | |
1503 | ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh); | |
1504 | if (ret) { | |
1505 | ocfs2_commit_trans(osb, handle); | |
1506 | ||
1507 | goto out; | |
1508 | } | |
1509 | } | |
1510 | ||
1511 | wc->w_handle = handle; | |
1512 | out: | |
1513 | return ret; | |
1514 | } | |
1515 | ||
1516 | int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size) | |
1517 | { | |
1518 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1519 | ||
1520 | if (new_size <= le16_to_cpu(di->id2.i_data.id_count)) | |
1521 | return 1; | |
1522 | return 0; | |
1523 | } | |
1524 | ||
1525 | static int ocfs2_try_to_write_inline_data(struct address_space *mapping, | |
1526 | struct inode *inode, loff_t pos, | |
1527 | unsigned len, struct page *mmap_page, | |
1528 | struct ocfs2_write_ctxt *wc) | |
1529 | { | |
1530 | int ret, written = 0; | |
1531 | loff_t end = pos + len; | |
1532 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1533 | struct ocfs2_dinode *di = NULL; | |
1534 | ||
1535 | mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n", | |
1536 | (unsigned long long)oi->ip_blkno, len, (unsigned long long)pos, | |
1537 | oi->ip_dyn_features); | |
1538 | ||
1539 | /* | |
1540 | * Handle inodes which already have inline data 1st. | |
1541 | */ | |
1542 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1543 | if (mmap_page == NULL && | |
1544 | ocfs2_size_fits_inline_data(wc->w_di_bh, end)) | |
1545 | goto do_inline_write; | |
1546 | ||
1547 | /* | |
1548 | * The write won't fit - we have to give this inode an | |
1549 | * inline extent list now. | |
1550 | */ | |
1551 | ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh); | |
1552 | if (ret) | |
1553 | mlog_errno(ret); | |
1554 | goto out; | |
1555 | } | |
1556 | ||
1557 | /* | |
1558 | * Check whether the inode can accept inline data. | |
1559 | */ | |
1560 | if (oi->ip_clusters != 0 || i_size_read(inode) != 0) | |
1561 | return 0; | |
1562 | ||
1563 | /* | |
1564 | * Check whether the write can fit. | |
1565 | */ | |
1566 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1567 | if (mmap_page || | |
1568 | end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) | |
1569 | return 0; | |
1570 | ||
1571 | do_inline_write: | |
1572 | ret = ocfs2_write_begin_inline(mapping, inode, wc); | |
1573 | if (ret) { | |
1574 | mlog_errno(ret); | |
1575 | goto out; | |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * This signals to the caller that the data can be written | |
1580 | * inline. | |
1581 | */ | |
1582 | written = 1; | |
1583 | out: | |
1584 | return written ? written : ret; | |
1585 | } | |
1586 | ||
1587 | /* | |
1588 | * This function only does anything for file systems which can't | |
1589 | * handle sparse files. | |
1590 | * | |
1591 | * What we want to do here is fill in any hole between the current end | |
1592 | * of allocation and the end of our write. That way the rest of the | |
1593 | * write path can treat it as an non-allocating write, which has no | |
1594 | * special case code for sparse/nonsparse files. | |
1595 | */ | |
1596 | static int ocfs2_expand_nonsparse_inode(struct inode *inode, | |
1597 | struct buffer_head *di_bh, | |
1598 | loff_t pos, unsigned len, | |
1599 | struct ocfs2_write_ctxt *wc) | |
1600 | { | |
1601 | int ret; | |
1602 | loff_t newsize = pos + len; | |
1603 | ||
1604 | BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); | |
1605 | ||
1606 | if (newsize <= i_size_read(inode)) | |
1607 | return 0; | |
1608 | ||
1609 | ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos); | |
1610 | if (ret) | |
1611 | mlog_errno(ret); | |
1612 | ||
1613 | wc->w_first_new_cpos = | |
1614 | ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)); | |
1615 | ||
1616 | return ret; | |
1617 | } | |
1618 | ||
1619 | static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh, | |
1620 | loff_t pos) | |
1621 | { | |
1622 | int ret = 0; | |
1623 | ||
1624 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); | |
1625 | if (pos > i_size_read(inode)) | |
1626 | ret = ocfs2_zero_extend(inode, di_bh, pos); | |
1627 | ||
1628 | return ret; | |
1629 | } | |
1630 | ||
1631 | /* | |
1632 | * Try to flush truncate logs if we can free enough clusters from it. | |
1633 | * As for return value, "< 0" means error, "0" no space and "1" means | |
1634 | * we have freed enough spaces and let the caller try to allocate again. | |
1635 | */ | |
1636 | static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb, | |
1637 | unsigned int needed) | |
1638 | { | |
1639 | tid_t target; | |
1640 | int ret = 0; | |
1641 | unsigned int truncated_clusters; | |
1642 | ||
1643 | mutex_lock(&osb->osb_tl_inode->i_mutex); | |
1644 | truncated_clusters = osb->truncated_clusters; | |
1645 | mutex_unlock(&osb->osb_tl_inode->i_mutex); | |
1646 | ||
1647 | /* | |
1648 | * Check whether we can succeed in allocating if we free | |
1649 | * the truncate log. | |
1650 | */ | |
1651 | if (truncated_clusters < needed) | |
1652 | goto out; | |
1653 | ||
1654 | ret = ocfs2_flush_truncate_log(osb); | |
1655 | if (ret) { | |
1656 | mlog_errno(ret); | |
1657 | goto out; | |
1658 | } | |
1659 | ||
1660 | if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) { | |
1661 | jbd2_log_wait_commit(osb->journal->j_journal, target); | |
1662 | ret = 1; | |
1663 | } | |
1664 | out: | |
1665 | return ret; | |
1666 | } | |
1667 | ||
1668 | int ocfs2_write_begin_nolock(struct file *filp, | |
1669 | struct address_space *mapping, | |
1670 | loff_t pos, unsigned len, unsigned flags, | |
1671 | struct page **pagep, void **fsdata, | |
1672 | struct buffer_head *di_bh, struct page *mmap_page) | |
1673 | { | |
1674 | int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS; | |
1675 | unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0; | |
1676 | struct ocfs2_write_ctxt *wc; | |
1677 | struct inode *inode = mapping->host; | |
1678 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1679 | struct ocfs2_dinode *di; | |
1680 | struct ocfs2_alloc_context *data_ac = NULL; | |
1681 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1682 | handle_t *handle; | |
1683 | struct ocfs2_extent_tree et; | |
1684 | int try_free = 1, ret1; | |
1685 | ||
1686 | try_again: | |
1687 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh); | |
1688 | if (ret) { | |
1689 | mlog_errno(ret); | |
1690 | return ret; | |
1691 | } | |
1692 | ||
1693 | if (ocfs2_supports_inline_data(osb)) { | |
1694 | ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len, | |
1695 | mmap_page, wc); | |
1696 | if (ret == 1) { | |
1697 | ret = 0; | |
1698 | goto success; | |
1699 | } | |
1700 | if (ret < 0) { | |
1701 | mlog_errno(ret); | |
1702 | goto out; | |
1703 | } | |
1704 | } | |
1705 | ||
1706 | if (ocfs2_sparse_alloc(osb)) | |
1707 | ret = ocfs2_zero_tail(inode, di_bh, pos); | |
1708 | else | |
1709 | ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len, | |
1710 | wc); | |
1711 | if (ret) { | |
1712 | mlog_errno(ret); | |
1713 | goto out; | |
1714 | } | |
1715 | ||
1716 | ret = ocfs2_check_range_for_refcount(inode, pos, len); | |
1717 | if (ret < 0) { | |
1718 | mlog_errno(ret); | |
1719 | goto out; | |
1720 | } else if (ret == 1) { | |
1721 | clusters_need = wc->w_clen; | |
1722 | ret = ocfs2_refcount_cow(inode, filp, di_bh, | |
1723 | wc->w_cpos, wc->w_clen, UINT_MAX); | |
1724 | if (ret) { | |
1725 | mlog_errno(ret); | |
1726 | goto out; | |
1727 | } | |
1728 | } | |
1729 | ||
1730 | ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc, | |
1731 | &extents_to_split); | |
1732 | if (ret) { | |
1733 | mlog_errno(ret); | |
1734 | goto out; | |
1735 | } | |
1736 | clusters_need += clusters_to_alloc; | |
1737 | ||
1738 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1739 | ||
1740 | /* | |
1741 | * We set w_target_from, w_target_to here so that | |
1742 | * ocfs2_write_end() knows which range in the target page to | |
1743 | * write out. An allocation requires that we write the entire | |
1744 | * cluster range. | |
1745 | */ | |
1746 | if (clusters_to_alloc || extents_to_split) { | |
1747 | /* | |
1748 | * XXX: We are stretching the limits of | |
1749 | * ocfs2_lock_allocators(). It greatly over-estimates | |
1750 | * the work to be done. | |
1751 | */ | |
1752 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u," | |
1753 | " clusters_to_add = %u, extents_to_split = %u\n", | |
1754 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1755 | (long long)i_size_read(inode), le32_to_cpu(di->i_clusters), | |
1756 | clusters_to_alloc, extents_to_split); | |
1757 | ||
1758 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), | |
1759 | wc->w_di_bh); | |
1760 | ret = ocfs2_lock_allocators(inode, &et, | |
1761 | clusters_to_alloc, extents_to_split, | |
1762 | &data_ac, &meta_ac); | |
1763 | if (ret) { | |
1764 | mlog_errno(ret); | |
1765 | goto out; | |
1766 | } | |
1767 | ||
1768 | if (data_ac) | |
1769 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; | |
1770 | ||
1771 | credits = ocfs2_calc_extend_credits(inode->i_sb, | |
1772 | &di->id2.i_list, | |
1773 | clusters_to_alloc); | |
1774 | ||
1775 | } | |
1776 | ||
1777 | /* | |
1778 | * We have to zero sparse allocated clusters, unwritten extent clusters, | |
1779 | * and non-sparse clusters we just extended. For non-sparse writes, | |
1780 | * we know zeros will only be needed in the first and/or last cluster. | |
1781 | */ | |
1782 | if (clusters_to_alloc || extents_to_split || | |
1783 | (wc->w_clen && (wc->w_desc[0].c_needs_zero || | |
1784 | wc->w_desc[wc->w_clen - 1].c_needs_zero))) | |
1785 | cluster_of_pages = 1; | |
1786 | else | |
1787 | cluster_of_pages = 0; | |
1788 | ||
1789 | ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages); | |
1790 | ||
1791 | handle = ocfs2_start_trans(osb, credits); | |
1792 | if (IS_ERR(handle)) { | |
1793 | ret = PTR_ERR(handle); | |
1794 | mlog_errno(ret); | |
1795 | goto out; | |
1796 | } | |
1797 | ||
1798 | wc->w_handle = handle; | |
1799 | ||
1800 | if (clusters_to_alloc) { | |
1801 | ret = dquot_alloc_space_nodirty(inode, | |
1802 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); | |
1803 | if (ret) | |
1804 | goto out_commit; | |
1805 | } | |
1806 | /* | |
1807 | * We don't want this to fail in ocfs2_write_end(), so do it | |
1808 | * here. | |
1809 | */ | |
1810 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, | |
1811 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1812 | if (ret) { | |
1813 | mlog_errno(ret); | |
1814 | goto out_quota; | |
1815 | } | |
1816 | ||
1817 | /* | |
1818 | * Fill our page array first. That way we've grabbed enough so | |
1819 | * that we can zero and flush if we error after adding the | |
1820 | * extent. | |
1821 | */ | |
1822 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len, | |
1823 | cluster_of_pages, mmap_page); | |
1824 | if (ret) { | |
1825 | mlog_errno(ret); | |
1826 | goto out_quota; | |
1827 | } | |
1828 | ||
1829 | ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos, | |
1830 | len); | |
1831 | if (ret) { | |
1832 | mlog_errno(ret); | |
1833 | goto out_quota; | |
1834 | } | |
1835 | ||
1836 | if (data_ac) | |
1837 | ocfs2_free_alloc_context(data_ac); | |
1838 | if (meta_ac) | |
1839 | ocfs2_free_alloc_context(meta_ac); | |
1840 | ||
1841 | success: | |
1842 | *pagep = wc->w_target_page; | |
1843 | *fsdata = wc; | |
1844 | return 0; | |
1845 | out_quota: | |
1846 | if (clusters_to_alloc) | |
1847 | dquot_free_space(inode, | |
1848 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); | |
1849 | out_commit: | |
1850 | ocfs2_commit_trans(osb, handle); | |
1851 | ||
1852 | out: | |
1853 | ocfs2_free_write_ctxt(wc); | |
1854 | ||
1855 | if (data_ac) | |
1856 | ocfs2_free_alloc_context(data_ac); | |
1857 | if (meta_ac) | |
1858 | ocfs2_free_alloc_context(meta_ac); | |
1859 | ||
1860 | if (ret == -ENOSPC && try_free) { | |
1861 | /* | |
1862 | * Try to free some truncate log so that we can have enough | |
1863 | * clusters to allocate. | |
1864 | */ | |
1865 | try_free = 0; | |
1866 | ||
1867 | ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); | |
1868 | if (ret1 == 1) | |
1869 | goto try_again; | |
1870 | ||
1871 | if (ret1 < 0) | |
1872 | mlog_errno(ret1); | |
1873 | } | |
1874 | ||
1875 | return ret; | |
1876 | } | |
1877 | ||
1878 | static int ocfs2_write_begin(struct file *file, struct address_space *mapping, | |
1879 | loff_t pos, unsigned len, unsigned flags, | |
1880 | struct page **pagep, void **fsdata) | |
1881 | { | |
1882 | int ret; | |
1883 | struct buffer_head *di_bh = NULL; | |
1884 | struct inode *inode = mapping->host; | |
1885 | ||
1886 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
1887 | if (ret) { | |
1888 | mlog_errno(ret); | |
1889 | return ret; | |
1890 | } | |
1891 | ||
1892 | /* | |
1893 | * Take alloc sem here to prevent concurrent lookups. That way | |
1894 | * the mapping, zeroing and tree manipulation within | |
1895 | * ocfs2_write() will be safe against ->readpage(). This | |
1896 | * should also serve to lock out allocation from a shared | |
1897 | * writeable region. | |
1898 | */ | |
1899 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1900 | ||
1901 | ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep, | |
1902 | fsdata, di_bh, NULL); | |
1903 | if (ret) { | |
1904 | mlog_errno(ret); | |
1905 | goto out_fail; | |
1906 | } | |
1907 | ||
1908 | brelse(di_bh); | |
1909 | ||
1910 | return 0; | |
1911 | ||
1912 | out_fail: | |
1913 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1914 | ||
1915 | brelse(di_bh); | |
1916 | ocfs2_inode_unlock(inode, 1); | |
1917 | ||
1918 | return ret; | |
1919 | } | |
1920 | ||
1921 | static void ocfs2_write_end_inline(struct inode *inode, loff_t pos, | |
1922 | unsigned len, unsigned *copied, | |
1923 | struct ocfs2_dinode *di, | |
1924 | struct ocfs2_write_ctxt *wc) | |
1925 | { | |
1926 | void *kaddr; | |
1927 | ||
1928 | if (unlikely(*copied < len)) { | |
1929 | if (!PageUptodate(wc->w_target_page)) { | |
1930 | *copied = 0; | |
1931 | return; | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | kaddr = kmap_atomic(wc->w_target_page, KM_USER0); | |
1936 | memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied); | |
1937 | kunmap_atomic(kaddr, KM_USER0); | |
1938 | ||
1939 | mlog(0, "Data written to inode at offset %llu. " | |
1940 | "id_count = %u, copied = %u, i_dyn_features = 0x%x\n", | |
1941 | (unsigned long long)pos, *copied, | |
1942 | le16_to_cpu(di->id2.i_data.id_count), | |
1943 | le16_to_cpu(di->i_dyn_features)); | |
1944 | } | |
1945 | ||
1946 | int ocfs2_write_end_nolock(struct address_space *mapping, | |
1947 | loff_t pos, unsigned len, unsigned copied, | |
1948 | struct page *page, void *fsdata) | |
1949 | { | |
1950 | int i; | |
1951 | unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1); | |
1952 | struct inode *inode = mapping->host; | |
1953 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1954 | struct ocfs2_write_ctxt *wc = fsdata; | |
1955 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1956 | handle_t *handle = wc->w_handle; | |
1957 | struct page *tmppage; | |
1958 | ||
1959 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1960 | ocfs2_write_end_inline(inode, pos, len, &copied, di, wc); | |
1961 | goto out_write_size; | |
1962 | } | |
1963 | ||
1964 | if (unlikely(copied < len)) { | |
1965 | if (!PageUptodate(wc->w_target_page)) | |
1966 | copied = 0; | |
1967 | ||
1968 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, | |
1969 | start+len); | |
1970 | } | |
1971 | flush_dcache_page(wc->w_target_page); | |
1972 | ||
1973 | for(i = 0; i < wc->w_num_pages; i++) { | |
1974 | tmppage = wc->w_pages[i]; | |
1975 | ||
1976 | if (tmppage == wc->w_target_page) { | |
1977 | from = wc->w_target_from; | |
1978 | to = wc->w_target_to; | |
1979 | ||
1980 | BUG_ON(from > PAGE_CACHE_SIZE || | |
1981 | to > PAGE_CACHE_SIZE || | |
1982 | to < from); | |
1983 | } else { | |
1984 | /* | |
1985 | * Pages adjacent to the target (if any) imply | |
1986 | * a hole-filling write in which case we want | |
1987 | * to flush their entire range. | |
1988 | */ | |
1989 | from = 0; | |
1990 | to = PAGE_CACHE_SIZE; | |
1991 | } | |
1992 | ||
1993 | if (page_has_buffers(tmppage)) { | |
1994 | if (ocfs2_should_order_data(inode)) | |
1995 | ocfs2_jbd2_file_inode(wc->w_handle, inode); | |
1996 | block_commit_write(tmppage, from, to); | |
1997 | } | |
1998 | } | |
1999 | ||
2000 | out_write_size: | |
2001 | pos += copied; | |
2002 | if (pos > inode->i_size) { | |
2003 | i_size_write(inode, pos); | |
2004 | mark_inode_dirty(inode); | |
2005 | } | |
2006 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
2007 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
2008 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
2009 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); | |
2010 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
2011 | ocfs2_journal_dirty(handle, wc->w_di_bh); | |
2012 | ||
2013 | ocfs2_commit_trans(osb, handle); | |
2014 | ||
2015 | ocfs2_run_deallocs(osb, &wc->w_dealloc); | |
2016 | ||
2017 | ocfs2_free_write_ctxt(wc); | |
2018 | ||
2019 | return copied; | |
2020 | } | |
2021 | ||
2022 | static int ocfs2_write_end(struct file *file, struct address_space *mapping, | |
2023 | loff_t pos, unsigned len, unsigned copied, | |
2024 | struct page *page, void *fsdata) | |
2025 | { | |
2026 | int ret; | |
2027 | struct inode *inode = mapping->host; | |
2028 | ||
2029 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata); | |
2030 | ||
2031 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
2032 | ocfs2_inode_unlock(inode, 1); | |
2033 | ||
2034 | return ret; | |
2035 | } | |
2036 | ||
2037 | const struct address_space_operations ocfs2_aops = { | |
2038 | .readpage = ocfs2_readpage, | |
2039 | .readpages = ocfs2_readpages, | |
2040 | .writepage = ocfs2_writepage, | |
2041 | .write_begin = ocfs2_write_begin, | |
2042 | .write_end = ocfs2_write_end, | |
2043 | .bmap = ocfs2_bmap, | |
2044 | .sync_page = block_sync_page, | |
2045 | .direct_IO = ocfs2_direct_IO, | |
2046 | .invalidatepage = ocfs2_invalidatepage, | |
2047 | .releasepage = ocfs2_releasepage, | |
2048 | .migratepage = buffer_migrate_page, | |
2049 | .is_partially_uptodate = block_is_partially_uptodate, | |
2050 | .error_remove_page = generic_error_remove_page, | |
2051 | }; |