]>
Commit | Line | Data |
---|---|---|
ccd979bd MF |
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> | |
9517bac6 | 27 | #include <linux/swap.h> |
6af67d82 | 28 | #include <linux/pipe_fs_i.h> |
628a24f5 | 29 | #include <linux/mpage.h> |
a90714c1 | 30 | #include <linux/quotaops.h> |
ccd979bd | 31 | |
ccd979bd MF |
32 | #include <cluster/masklog.h> |
33 | ||
34 | #include "ocfs2.h" | |
35 | ||
36 | #include "alloc.h" | |
37 | #include "aops.h" | |
38 | #include "dlmglue.h" | |
39 | #include "extent_map.h" | |
40 | #include "file.h" | |
41 | #include "inode.h" | |
42 | #include "journal.h" | |
9517bac6 | 43 | #include "suballoc.h" |
ccd979bd MF |
44 | #include "super.h" |
45 | #include "symlink.h" | |
293b2f70 | 46 | #include "refcounttree.h" |
9558156b | 47 | #include "ocfs2_trace.h" |
ccd979bd MF |
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 | ||
9558156b TM |
62 | trace_ocfs2_symlink_get_block( |
63 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
64 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
65 | |
66 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); | |
67 | ||
68 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { | |
69 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", | |
70 | (unsigned long long)iblock); | |
71 | goto bail; | |
72 | } | |
73 | ||
b657c95c | 74 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
75 | if (status < 0) { |
76 | mlog_errno(status); | |
77 | goto bail; | |
78 | } | |
79 | fe = (struct ocfs2_dinode *) bh->b_data; | |
80 | ||
ccd979bd MF |
81 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, |
82 | le32_to_cpu(fe->i_clusters))) { | |
83 | mlog(ML_ERROR, "block offset is outside the allocated size: " | |
84 | "%llu\n", (unsigned long long)iblock); | |
85 | goto bail; | |
86 | } | |
87 | ||
88 | /* We don't use the page cache to create symlink data, so if | |
89 | * need be, copy it over from the buffer cache. */ | |
90 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { | |
91 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + | |
92 | iblock; | |
93 | buffer_cache_bh = sb_getblk(osb->sb, blkno); | |
94 | if (!buffer_cache_bh) { | |
95 | mlog(ML_ERROR, "couldn't getblock for symlink!\n"); | |
96 | goto bail; | |
97 | } | |
98 | ||
99 | /* we haven't locked out transactions, so a commit | |
100 | * could've happened. Since we've got a reference on | |
101 | * the bh, even if it commits while we're doing the | |
102 | * copy, the data is still good. */ | |
103 | if (buffer_jbd(buffer_cache_bh) | |
104 | && ocfs2_inode_is_new(inode)) { | |
c4bc8dcb | 105 | kaddr = kmap_atomic(bh_result->b_page); |
ccd979bd MF |
106 | if (!kaddr) { |
107 | mlog(ML_ERROR, "couldn't kmap!\n"); | |
108 | goto bail; | |
109 | } | |
110 | memcpy(kaddr + (bh_result->b_size * iblock), | |
111 | buffer_cache_bh->b_data, | |
112 | bh_result->b_size); | |
c4bc8dcb | 113 | kunmap_atomic(kaddr); |
ccd979bd MF |
114 | set_buffer_uptodate(bh_result); |
115 | } | |
116 | brelse(buffer_cache_bh); | |
117 | } | |
118 | ||
119 | map_bh(bh_result, inode->i_sb, | |
120 | le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); | |
121 | ||
122 | err = 0; | |
123 | ||
124 | bail: | |
a81cb88b | 125 | brelse(bh); |
ccd979bd | 126 | |
ccd979bd MF |
127 | return err; |
128 | } | |
129 | ||
6f70fa51 TM |
130 | int ocfs2_get_block(struct inode *inode, sector_t iblock, |
131 | struct buffer_head *bh_result, int create) | |
ccd979bd MF |
132 | { |
133 | int err = 0; | |
49cb8d2d | 134 | unsigned int ext_flags; |
628a24f5 MF |
135 | u64 max_blocks = bh_result->b_size >> inode->i_blkbits; |
136 | u64 p_blkno, count, past_eof; | |
25baf2da | 137 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd | 138 | |
9558156b TM |
139 | trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno, |
140 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
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 | ||
628a24f5 | 152 | err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count, |
49cb8d2d | 153 | &ext_flags); |
ccd979bd MF |
154 | if (err) { |
155 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " | |
b0697053 MF |
156 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, |
157 | (unsigned long long)p_blkno); | |
ccd979bd MF |
158 | goto bail; |
159 | } | |
160 | ||
628a24f5 MF |
161 | if (max_blocks < count) |
162 | count = max_blocks; | |
163 | ||
25baf2da MF |
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 | |
ebdec241 | 168 | * allows __block_write_begin() to zero. |
c0420ad2 CL |
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. | |
25baf2da | 174 | */ |
c0420ad2 CL |
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 | } | |
25baf2da | 180 | |
49cb8d2d MF |
181 | /* Treat the unwritten extent as a hole for zeroing purposes. */ |
182 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
25baf2da MF |
183 | map_bh(bh_result, inode->i_sb, p_blkno); |
184 | ||
628a24f5 MF |
185 | bh_result->b_size = count << inode->i_blkbits; |
186 | ||
25baf2da MF |
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(); | |
1f4cea37 | 197 | goto bail; |
25baf2da | 198 | } |
25baf2da | 199 | } |
ccd979bd | 200 | |
5693486b | 201 | past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
9558156b TM |
202 | |
203 | trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno, | |
204 | (unsigned long long)past_eof); | |
5693486b JB |
205 | if (create && (iblock >= past_eof)) |
206 | set_buffer_new(bh_result); | |
207 | ||
ccd979bd MF |
208 | bail: |
209 | if (err < 0) | |
210 | err = -EIO; | |
211 | ||
ccd979bd MF |
212 | return err; |
213 | } | |
214 | ||
1afc32b9 MF |
215 | int ocfs2_read_inline_data(struct inode *inode, struct page *page, |
216 | struct buffer_head *di_bh) | |
6798d35a MF |
217 | { |
218 | void *kaddr; | |
d2849fb2 | 219 | loff_t size; |
6798d35a MF |
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 || | |
d9ae49d6 | 231 | size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) { |
6798d35a | 232 | ocfs2_error(inode->i_sb, |
d2849fb2 JK |
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); | |
6798d35a MF |
236 | return -EROFS; |
237 | } | |
238 | ||
c4bc8dcb | 239 | kaddr = kmap_atomic(page); |
6798d35a MF |
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); | |
c4bc8dcb | 245 | kunmap_atomic(kaddr); |
6798d35a MF |
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; | |
6798d35a MF |
256 | |
257 | BUG_ON(!PageLocked(page)); | |
86c838b0 | 258 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)); |
6798d35a | 259 | |
b657c95c | 260 | ret = ocfs2_read_inode_block(inode, &di_bh); |
6798d35a MF |
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 | ||
ccd979bd MF |
274 | static int ocfs2_readpage(struct file *file, struct page *page) |
275 | { | |
276 | struct inode *inode = page->mapping->host; | |
6798d35a | 277 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd MF |
278 | loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT; |
279 | int ret, unlock = 1; | |
280 | ||
9558156b TM |
281 | trace_ocfs2_readpage((unsigned long long)oi->ip_blkno, |
282 | (page ? page->index : 0)); | |
ccd979bd | 283 | |
e63aecb6 | 284 | ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page); |
ccd979bd MF |
285 | if (ret != 0) { |
286 | if (ret == AOP_TRUNCATED_PAGE) | |
287 | unlock = 0; | |
288 | mlog_errno(ret); | |
289 | goto out; | |
290 | } | |
291 | ||
6798d35a | 292 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { |
c7e25e6e JK |
293 | /* |
294 | * Unlock the page and cycle ip_alloc_sem so that we don't | |
295 | * busyloop waiting for ip_alloc_sem to unlock | |
296 | */ | |
e9dfc0b2 | 297 | ret = AOP_TRUNCATED_PAGE; |
c7e25e6e JK |
298 | unlock_page(page); |
299 | unlock = 0; | |
300 | down_read(&oi->ip_alloc_sem); | |
301 | up_read(&oi->ip_alloc_sem); | |
e63aecb6 | 302 | goto out_inode_unlock; |
e9dfc0b2 | 303 | } |
ccd979bd MF |
304 | |
305 | /* | |
306 | * i_size might have just been updated as we grabed the meta lock. We | |
307 | * might now be discovering a truncate that hit on another node. | |
308 | * block_read_full_page->get_block freaks out if it is asked to read | |
309 | * beyond the end of a file, so we check here. Callers | |
54cb8821 | 310 | * (generic_file_read, vm_ops->fault) are clever enough to check i_size |
ccd979bd MF |
311 | * and notice that the page they just read isn't needed. |
312 | * | |
313 | * XXX sys_readahead() seems to get that wrong? | |
314 | */ | |
315 | if (start >= i_size_read(inode)) { | |
eebd2aa3 | 316 | zero_user(page, 0, PAGE_SIZE); |
ccd979bd MF |
317 | SetPageUptodate(page); |
318 | ret = 0; | |
319 | goto out_alloc; | |
320 | } | |
321 | ||
6798d35a MF |
322 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
323 | ret = ocfs2_readpage_inline(inode, page); | |
324 | else | |
325 | ret = block_read_full_page(page, ocfs2_get_block); | |
ccd979bd MF |
326 | unlock = 0; |
327 | ||
ccd979bd MF |
328 | out_alloc: |
329 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 MF |
330 | out_inode_unlock: |
331 | ocfs2_inode_unlock(inode, 0); | |
ccd979bd MF |
332 | out: |
333 | if (unlock) | |
334 | unlock_page(page); | |
ccd979bd MF |
335 | return ret; |
336 | } | |
337 | ||
628a24f5 MF |
338 | /* |
339 | * This is used only for read-ahead. Failures or difficult to handle | |
340 | * situations are safe to ignore. | |
341 | * | |
342 | * Right now, we don't bother with BH_Boundary - in-inode extent lists | |
343 | * are quite large (243 extents on 4k blocks), so most inodes don't | |
344 | * grow out to a tree. If need be, detecting boundary extents could | |
345 | * trivially be added in a future version of ocfs2_get_block(). | |
346 | */ | |
347 | static int ocfs2_readpages(struct file *filp, struct address_space *mapping, | |
348 | struct list_head *pages, unsigned nr_pages) | |
349 | { | |
350 | int ret, err = -EIO; | |
351 | struct inode *inode = mapping->host; | |
352 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
353 | loff_t start; | |
354 | struct page *last; | |
355 | ||
356 | /* | |
357 | * Use the nonblocking flag for the dlm code to avoid page | |
358 | * lock inversion, but don't bother with retrying. | |
359 | */ | |
360 | ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK); | |
361 | if (ret) | |
362 | return err; | |
363 | ||
364 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { | |
365 | ocfs2_inode_unlock(inode, 0); | |
366 | return err; | |
367 | } | |
368 | ||
369 | /* | |
370 | * Don't bother with inline-data. There isn't anything | |
371 | * to read-ahead in that case anyway... | |
372 | */ | |
373 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
374 | goto out_unlock; | |
375 | ||
376 | /* | |
377 | * Check whether a remote node truncated this file - we just | |
378 | * drop out in that case as it's not worth handling here. | |
379 | */ | |
380 | last = list_entry(pages->prev, struct page, lru); | |
381 | start = (loff_t)last->index << PAGE_CACHE_SHIFT; | |
382 | if (start >= i_size_read(inode)) | |
383 | goto out_unlock; | |
384 | ||
385 | err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block); | |
386 | ||
387 | out_unlock: | |
388 | up_read(&oi->ip_alloc_sem); | |
389 | ocfs2_inode_unlock(inode, 0); | |
390 | ||
391 | return err; | |
392 | } | |
393 | ||
ccd979bd MF |
394 | /* Note: Because we don't support holes, our allocation has |
395 | * already happened (allocation writes zeros to the file data) | |
396 | * so we don't have to worry about ordered writes in | |
397 | * ocfs2_writepage. | |
398 | * | |
399 | * ->writepage is called during the process of invalidating the page cache | |
400 | * during blocked lock processing. It can't block on any cluster locks | |
401 | * to during block mapping. It's relying on the fact that the block | |
402 | * mapping can't have disappeared under the dirty pages that it is | |
403 | * being asked to write back. | |
404 | */ | |
405 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) | |
406 | { | |
9558156b TM |
407 | trace_ocfs2_writepage( |
408 | (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno, | |
409 | page->index); | |
ccd979bd | 410 | |
9558156b | 411 | return block_write_full_page(page, ocfs2_get_block, wbc); |
ccd979bd MF |
412 | } |
413 | ||
ccd979bd MF |
414 | /* Taken from ext3. We don't necessarily need the full blown |
415 | * functionality yet, but IMHO it's better to cut and paste the whole | |
416 | * thing so we can avoid introducing our own bugs (and easily pick up | |
417 | * their fixes when they happen) --Mark */ | |
60b11392 MF |
418 | int walk_page_buffers( handle_t *handle, |
419 | struct buffer_head *head, | |
420 | unsigned from, | |
421 | unsigned to, | |
422 | int *partial, | |
423 | int (*fn)( handle_t *handle, | |
424 | struct buffer_head *bh)) | |
ccd979bd MF |
425 | { |
426 | struct buffer_head *bh; | |
427 | unsigned block_start, block_end; | |
428 | unsigned blocksize = head->b_size; | |
429 | int err, ret = 0; | |
430 | struct buffer_head *next; | |
431 | ||
432 | for ( bh = head, block_start = 0; | |
433 | ret == 0 && (bh != head || !block_start); | |
434 | block_start = block_end, bh = next) | |
435 | { | |
436 | next = bh->b_this_page; | |
437 | block_end = block_start + blocksize; | |
438 | if (block_end <= from || block_start >= to) { | |
439 | if (partial && !buffer_uptodate(bh)) | |
440 | *partial = 1; | |
441 | continue; | |
442 | } | |
443 | err = (*fn)(handle, bh); | |
444 | if (!ret) | |
445 | ret = err; | |
446 | } | |
447 | return ret; | |
448 | } | |
449 | ||
ccd979bd MF |
450 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) |
451 | { | |
452 | sector_t status; | |
453 | u64 p_blkno = 0; | |
454 | int err = 0; | |
455 | struct inode *inode = mapping->host; | |
456 | ||
9558156b TM |
457 | trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno, |
458 | (unsigned long long)block); | |
ccd979bd MF |
459 | |
460 | /* We don't need to lock journal system files, since they aren't | |
461 | * accessed concurrently from multiple nodes. | |
462 | */ | |
463 | if (!INODE_JOURNAL(inode)) { | |
e63aecb6 | 464 | err = ocfs2_inode_lock(inode, NULL, 0); |
ccd979bd MF |
465 | if (err) { |
466 | if (err != -ENOENT) | |
467 | mlog_errno(err); | |
468 | goto bail; | |
469 | } | |
470 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | |
471 | } | |
472 | ||
6798d35a MF |
473 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) |
474 | err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, | |
475 | NULL); | |
ccd979bd MF |
476 | |
477 | if (!INODE_JOURNAL(inode)) { | |
478 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 | 479 | ocfs2_inode_unlock(inode, 0); |
ccd979bd MF |
480 | } |
481 | ||
482 | if (err) { | |
483 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", | |
484 | (unsigned long long)block); | |
485 | mlog_errno(err); | |
486 | goto bail; | |
487 | } | |
488 | ||
ccd979bd MF |
489 | bail: |
490 | status = err ? 0 : p_blkno; | |
491 | ||
ccd979bd MF |
492 | return status; |
493 | } | |
494 | ||
495 | /* | |
496 | * TODO: Make this into a generic get_blocks function. | |
497 | * | |
498 | * From do_direct_io in direct-io.c: | |
499 | * "So what we do is to permit the ->get_blocks function to populate | |
500 | * bh.b_size with the size of IO which is permitted at this offset and | |
501 | * this i_blkbits." | |
502 | * | |
503 | * This function is called directly from get_more_blocks in direct-io.c. | |
504 | * | |
505 | * called like this: dio->get_blocks(dio->inode, fs_startblk, | |
506 | * fs_count, map_bh, dio->rw == WRITE); | |
5fe878ae CH |
507 | * |
508 | * Note that we never bother to allocate blocks here, and thus ignore the | |
509 | * create argument. | |
ccd979bd MF |
510 | */ |
511 | static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock, | |
ccd979bd MF |
512 | struct buffer_head *bh_result, int create) |
513 | { | |
514 | int ret; | |
4f902c37 | 515 | u64 p_blkno, inode_blocks, contig_blocks; |
49cb8d2d | 516 | unsigned int ext_flags; |
184d7d20 | 517 | unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits; |
1d8fa7a2 | 518 | unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits; |
ccd979bd | 519 | |
ccd979bd MF |
520 | /* This function won't even be called if the request isn't all |
521 | * nicely aligned and of the right size, so there's no need | |
522 | * for us to check any of that. */ | |
523 | ||
25baf2da | 524 | inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
564f8a32 | 525 | |
ccd979bd MF |
526 | /* This figures out the size of the next contiguous block, and |
527 | * our logical offset */ | |
363041a5 | 528 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, |
49cb8d2d | 529 | &contig_blocks, &ext_flags); |
ccd979bd MF |
530 | if (ret) { |
531 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", | |
532 | (unsigned long long)iblock); | |
533 | ret = -EIO; | |
534 | goto bail; | |
535 | } | |
536 | ||
cbaee472 TM |
537 | /* We should already CoW the refcounted extent in case of create. */ |
538 | BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED)); | |
539 | ||
25baf2da MF |
540 | /* |
541 | * get_more_blocks() expects us to describe a hole by clearing | |
542 | * the mapped bit on bh_result(). | |
49cb8d2d MF |
543 | * |
544 | * Consider an unwritten extent as a hole. | |
25baf2da | 545 | */ |
49cb8d2d | 546 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) |
25baf2da | 547 | map_bh(bh_result, inode->i_sb, p_blkno); |
5fe878ae | 548 | else |
25baf2da | 549 | clear_buffer_mapped(bh_result); |
ccd979bd MF |
550 | |
551 | /* make sure we don't map more than max_blocks blocks here as | |
552 | that's all the kernel will handle at this point. */ | |
553 | if (max_blocks < contig_blocks) | |
554 | contig_blocks = max_blocks; | |
555 | bh_result->b_size = contig_blocks << blocksize_bits; | |
556 | bail: | |
557 | return ret; | |
558 | } | |
559 | ||
2bd63216 | 560 | /* |
ccd979bd | 561 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're |
bd5fe6c5 CH |
562 | * particularly interested in the aio/dio case. We use the rw_lock DLM lock |
563 | * to protect io on one node from truncation on another. | |
ccd979bd MF |
564 | */ |
565 | static void ocfs2_dio_end_io(struct kiocb *iocb, | |
566 | loff_t offset, | |
567 | ssize_t bytes, | |
40e2e973 CH |
568 | void *private, |
569 | int ret, | |
570 | bool is_async) | |
ccd979bd | 571 | { |
496ad9aa | 572 | struct inode *inode = file_inode(iocb->ki_filp); |
7cdfc3a1 | 573 | int level; |
a11f7e63 | 574 | wait_queue_head_t *wq = ocfs2_ioend_wq(inode); |
ccd979bd MF |
575 | |
576 | /* this io's submitter should not have unlocked this before we could */ | |
577 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); | |
7cdfc3a1 | 578 | |
df2d6f26 | 579 | if (ocfs2_iocb_is_sem_locked(iocb)) |
39c99f12 | 580 | ocfs2_iocb_clear_sem_locked(iocb); |
39c99f12 | 581 | |
a11f7e63 MF |
582 | if (ocfs2_iocb_is_unaligned_aio(iocb)) { |
583 | ocfs2_iocb_clear_unaligned_aio(iocb); | |
584 | ||
585 | if (atomic_dec_and_test(&OCFS2_I(inode)->ip_unaligned_aio) && | |
586 | waitqueue_active(wq)) { | |
587 | wake_up_all(wq); | |
588 | } | |
589 | } | |
590 | ||
ccd979bd | 591 | ocfs2_iocb_clear_rw_locked(iocb); |
7cdfc3a1 MF |
592 | |
593 | level = ocfs2_iocb_rw_locked_level(iocb); | |
7cdfc3a1 | 594 | ocfs2_rw_unlock(inode, level); |
40e2e973 | 595 | |
9b171e0c | 596 | inode_dio_done(inode); |
40e2e973 CH |
597 | if (is_async) |
598 | aio_complete(iocb, ret, 0); | |
ccd979bd MF |
599 | } |
600 | ||
03f981cf JB |
601 | /* |
602 | * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen | |
603 | * from ext3. PageChecked() bits have been removed as OCFS2 does not | |
604 | * do journalled data. | |
605 | */ | |
d47992f8 LC |
606 | static void ocfs2_invalidatepage(struct page *page, unsigned int offset, |
607 | unsigned int length) | |
03f981cf JB |
608 | { |
609 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | |
610 | ||
259709b0 LC |
611 | jbd2_journal_invalidatepage(journal, page, offset, |
612 | PAGE_CACHE_SIZE - offset); | |
03f981cf JB |
613 | } |
614 | ||
615 | static int ocfs2_releasepage(struct page *page, gfp_t wait) | |
616 | { | |
617 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | |
618 | ||
619 | if (!page_has_buffers(page)) | |
620 | return 0; | |
2b4e30fb | 621 | return jbd2_journal_try_to_free_buffers(journal, page, wait); |
03f981cf JB |
622 | } |
623 | ||
ccd979bd MF |
624 | static ssize_t ocfs2_direct_IO(int rw, |
625 | struct kiocb *iocb, | |
626 | const struct iovec *iov, | |
627 | loff_t offset, | |
628 | unsigned long nr_segs) | |
629 | { | |
630 | struct file *file = iocb->ki_filp; | |
496ad9aa | 631 | struct inode *inode = file_inode(file)->i_mapping->host; |
53013cba | 632 | |
6798d35a MF |
633 | /* |
634 | * Fallback to buffered I/O if we see an inode without | |
635 | * extents. | |
636 | */ | |
637 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
638 | return 0; | |
639 | ||
b80474b4 TM |
640 | /* Fallback to buffered I/O if we are appending. */ |
641 | if (i_size_read(inode) <= offset) | |
642 | return 0; | |
643 | ||
c1e8d35e TM |
644 | return __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, |
645 | iov, offset, nr_segs, | |
646 | ocfs2_direct_IO_get_blocks, | |
647 | ocfs2_dio_end_io, NULL, 0); | |
ccd979bd MF |
648 | } |
649 | ||
9517bac6 MF |
650 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, |
651 | u32 cpos, | |
652 | unsigned int *start, | |
653 | unsigned int *end) | |
654 | { | |
655 | unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE; | |
656 | ||
657 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) { | |
658 | unsigned int cpp; | |
659 | ||
660 | cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits); | |
661 | ||
662 | cluster_start = cpos % cpp; | |
663 | cluster_start = cluster_start << osb->s_clustersize_bits; | |
664 | ||
665 | cluster_end = cluster_start + osb->s_clustersize; | |
666 | } | |
667 | ||
668 | BUG_ON(cluster_start > PAGE_SIZE); | |
669 | BUG_ON(cluster_end > PAGE_SIZE); | |
670 | ||
671 | if (start) | |
672 | *start = cluster_start; | |
673 | if (end) | |
674 | *end = cluster_end; | |
675 | } | |
676 | ||
677 | /* | |
678 | * 'from' and 'to' are the region in the page to avoid zeroing. | |
679 | * | |
680 | * If pagesize > clustersize, this function will avoid zeroing outside | |
681 | * of the cluster boundary. | |
682 | * | |
683 | * from == to == 0 is code for "zero the entire cluster region" | |
684 | */ | |
685 | static void ocfs2_clear_page_regions(struct page *page, | |
686 | struct ocfs2_super *osb, u32 cpos, | |
687 | unsigned from, unsigned to) | |
688 | { | |
689 | void *kaddr; | |
690 | unsigned int cluster_start, cluster_end; | |
691 | ||
692 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); | |
693 | ||
c4bc8dcb | 694 | kaddr = kmap_atomic(page); |
9517bac6 MF |
695 | |
696 | if (from || to) { | |
697 | if (from > cluster_start) | |
698 | memset(kaddr + cluster_start, 0, from - cluster_start); | |
699 | if (to < cluster_end) | |
700 | memset(kaddr + to, 0, cluster_end - to); | |
701 | } else { | |
702 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); | |
703 | } | |
704 | ||
c4bc8dcb | 705 | kunmap_atomic(kaddr); |
9517bac6 MF |
706 | } |
707 | ||
4e9563fd MF |
708 | /* |
709 | * Nonsparse file systems fully allocate before we get to the write | |
710 | * code. This prevents ocfs2_write() from tagging the write as an | |
711 | * allocating one, which means ocfs2_map_page_blocks() might try to | |
712 | * read-in the blocks at the tail of our file. Avoid reading them by | |
713 | * testing i_size against each block offset. | |
714 | */ | |
715 | static int ocfs2_should_read_blk(struct inode *inode, struct page *page, | |
716 | unsigned int block_start) | |
717 | { | |
718 | u64 offset = page_offset(page) + block_start; | |
719 | ||
720 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
721 | return 1; | |
722 | ||
723 | if (i_size_read(inode) > offset) | |
724 | return 1; | |
725 | ||
726 | return 0; | |
727 | } | |
728 | ||
9517bac6 | 729 | /* |
ebdec241 | 730 | * Some of this taken from __block_write_begin(). We already have our |
9517bac6 MF |
731 | * mapping by now though, and the entire write will be allocating or |
732 | * it won't, so not much need to use BH_New. | |
733 | * | |
734 | * This will also skip zeroing, which is handled externally. | |
735 | */ | |
60b11392 MF |
736 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, |
737 | struct inode *inode, unsigned int from, | |
738 | unsigned int to, int new) | |
9517bac6 MF |
739 | { |
740 | int ret = 0; | |
741 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; | |
742 | unsigned int block_end, block_start; | |
743 | unsigned int bsize = 1 << inode->i_blkbits; | |
744 | ||
745 | if (!page_has_buffers(page)) | |
746 | create_empty_buffers(page, bsize, 0); | |
747 | ||
748 | head = page_buffers(page); | |
749 | for (bh = head, block_start = 0; bh != head || !block_start; | |
750 | bh = bh->b_this_page, block_start += bsize) { | |
751 | block_end = block_start + bsize; | |
752 | ||
3a307ffc MF |
753 | clear_buffer_new(bh); |
754 | ||
9517bac6 MF |
755 | /* |
756 | * Ignore blocks outside of our i/o range - | |
757 | * they may belong to unallocated clusters. | |
758 | */ | |
60b11392 | 759 | if (block_start >= to || block_end <= from) { |
9517bac6 MF |
760 | if (PageUptodate(page)) |
761 | set_buffer_uptodate(bh); | |
762 | continue; | |
763 | } | |
764 | ||
765 | /* | |
766 | * For an allocating write with cluster size >= page | |
767 | * size, we always write the entire page. | |
768 | */ | |
3a307ffc MF |
769 | if (new) |
770 | set_buffer_new(bh); | |
9517bac6 MF |
771 | |
772 | if (!buffer_mapped(bh)) { | |
773 | map_bh(bh, inode->i_sb, *p_blkno); | |
774 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
775 | } | |
776 | ||
777 | if (PageUptodate(page)) { | |
778 | if (!buffer_uptodate(bh)) | |
779 | set_buffer_uptodate(bh); | |
780 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
bce99768 | 781 | !buffer_new(bh) && |
4e9563fd | 782 | ocfs2_should_read_blk(inode, page, block_start) && |
bce99768 | 783 | (block_start < from || block_end > to)) { |
9517bac6 MF |
784 | ll_rw_block(READ, 1, &bh); |
785 | *wait_bh++=bh; | |
786 | } | |
787 | ||
788 | *p_blkno = *p_blkno + 1; | |
789 | } | |
790 | ||
791 | /* | |
792 | * If we issued read requests - let them complete. | |
793 | */ | |
794 | while(wait_bh > wait) { | |
795 | wait_on_buffer(*--wait_bh); | |
796 | if (!buffer_uptodate(*wait_bh)) | |
797 | ret = -EIO; | |
798 | } | |
799 | ||
800 | if (ret == 0 || !new) | |
801 | return ret; | |
802 | ||
803 | /* | |
804 | * If we get -EIO above, zero out any newly allocated blocks | |
805 | * to avoid exposing stale data. | |
806 | */ | |
807 | bh = head; | |
808 | block_start = 0; | |
809 | do { | |
9517bac6 MF |
810 | block_end = block_start + bsize; |
811 | if (block_end <= from) | |
812 | goto next_bh; | |
813 | if (block_start >= to) | |
814 | break; | |
815 | ||
eebd2aa3 | 816 | zero_user(page, block_start, bh->b_size); |
9517bac6 MF |
817 | set_buffer_uptodate(bh); |
818 | mark_buffer_dirty(bh); | |
819 | ||
820 | next_bh: | |
821 | block_start = block_end; | |
822 | bh = bh->b_this_page; | |
823 | } while (bh != head); | |
824 | ||
825 | return ret; | |
826 | } | |
827 | ||
3a307ffc MF |
828 | #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE) |
829 | #define OCFS2_MAX_CTXT_PAGES 1 | |
830 | #else | |
831 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE) | |
832 | #endif | |
833 | ||
834 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE) | |
835 | ||
6af67d82 | 836 | /* |
3a307ffc | 837 | * Describe the state of a single cluster to be written to. |
6af67d82 | 838 | */ |
3a307ffc MF |
839 | struct ocfs2_write_cluster_desc { |
840 | u32 c_cpos; | |
841 | u32 c_phys; | |
842 | /* | |
843 | * Give this a unique field because c_phys eventually gets | |
844 | * filled. | |
845 | */ | |
846 | unsigned c_new; | |
b27b7cbc | 847 | unsigned c_unwritten; |
e7432675 | 848 | unsigned c_needs_zero; |
3a307ffc | 849 | }; |
6af67d82 | 850 | |
3a307ffc MF |
851 | struct ocfs2_write_ctxt { |
852 | /* Logical cluster position / len of write */ | |
853 | u32 w_cpos; | |
854 | u32 w_clen; | |
6af67d82 | 855 | |
e7432675 SM |
856 | /* First cluster allocated in a nonsparse extend */ |
857 | u32 w_first_new_cpos; | |
858 | ||
3a307ffc | 859 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; |
6af67d82 | 860 | |
3a307ffc MF |
861 | /* |
862 | * This is true if page_size > cluster_size. | |
863 | * | |
864 | * It triggers a set of special cases during write which might | |
865 | * have to deal with allocating writes to partial pages. | |
866 | */ | |
867 | unsigned int w_large_pages; | |
6af67d82 | 868 | |
3a307ffc MF |
869 | /* |
870 | * Pages involved in this write. | |
871 | * | |
872 | * w_target_page is the page being written to by the user. | |
873 | * | |
874 | * w_pages is an array of pages which always contains | |
875 | * w_target_page, and in the case of an allocating write with | |
876 | * page_size < cluster size, it will contain zero'd and mapped | |
877 | * pages adjacent to w_target_page which need to be written | |
878 | * out in so that future reads from that region will get | |
879 | * zero's. | |
880 | */ | |
3a307ffc | 881 | unsigned int w_num_pages; |
83fd9c7f | 882 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; |
3a307ffc | 883 | struct page *w_target_page; |
eeb47d12 | 884 | |
5cffff9e WW |
885 | /* |
886 | * w_target_locked is used for page_mkwrite path indicating no unlocking | |
887 | * against w_target_page in ocfs2_write_end_nolock. | |
888 | */ | |
889 | unsigned int w_target_locked:1; | |
890 | ||
3a307ffc MF |
891 | /* |
892 | * ocfs2_write_end() uses this to know what the real range to | |
893 | * write in the target should be. | |
894 | */ | |
895 | unsigned int w_target_from; | |
896 | unsigned int w_target_to; | |
897 | ||
898 | /* | |
899 | * We could use journal_current_handle() but this is cleaner, | |
900 | * IMHO -Mark | |
901 | */ | |
902 | handle_t *w_handle; | |
903 | ||
904 | struct buffer_head *w_di_bh; | |
b27b7cbc MF |
905 | |
906 | struct ocfs2_cached_dealloc_ctxt w_dealloc; | |
3a307ffc MF |
907 | }; |
908 | ||
1d410a6e | 909 | void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages) |
3a307ffc MF |
910 | { |
911 | int i; | |
912 | ||
1d410a6e MF |
913 | for(i = 0; i < num_pages; i++) { |
914 | if (pages[i]) { | |
915 | unlock_page(pages[i]); | |
916 | mark_page_accessed(pages[i]); | |
917 | page_cache_release(pages[i]); | |
918 | } | |
6af67d82 | 919 | } |
1d410a6e MF |
920 | } |
921 | ||
922 | static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc) | |
923 | { | |
5cffff9e WW |
924 | int i; |
925 | ||
926 | /* | |
927 | * w_target_locked is only set to true in the page_mkwrite() case. | |
928 | * The intent is to allow us to lock the target page from write_begin() | |
929 | * to write_end(). The caller must hold a ref on w_target_page. | |
930 | */ | |
931 | if (wc->w_target_locked) { | |
932 | BUG_ON(!wc->w_target_page); | |
933 | for (i = 0; i < wc->w_num_pages; i++) { | |
934 | if (wc->w_target_page == wc->w_pages[i]) { | |
935 | wc->w_pages[i] = NULL; | |
936 | break; | |
937 | } | |
938 | } | |
939 | mark_page_accessed(wc->w_target_page); | |
940 | page_cache_release(wc->w_target_page); | |
941 | } | |
1d410a6e | 942 | ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages); |
6af67d82 | 943 | |
3a307ffc MF |
944 | brelse(wc->w_di_bh); |
945 | kfree(wc); | |
946 | } | |
947 | ||
948 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, | |
949 | struct ocfs2_super *osb, loff_t pos, | |
607d44aa | 950 | unsigned len, struct buffer_head *di_bh) |
3a307ffc | 951 | { |
30b8548f | 952 | u32 cend; |
3a307ffc MF |
953 | struct ocfs2_write_ctxt *wc; |
954 | ||
955 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); | |
956 | if (!wc) | |
957 | return -ENOMEM; | |
6af67d82 | 958 | |
3a307ffc | 959 | wc->w_cpos = pos >> osb->s_clustersize_bits; |
e7432675 | 960 | wc->w_first_new_cpos = UINT_MAX; |
30b8548f | 961 | cend = (pos + len - 1) >> osb->s_clustersize_bits; |
962 | wc->w_clen = cend - wc->w_cpos + 1; | |
607d44aa MF |
963 | get_bh(di_bh); |
964 | wc->w_di_bh = di_bh; | |
6af67d82 | 965 | |
3a307ffc MF |
966 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) |
967 | wc->w_large_pages = 1; | |
968 | else | |
969 | wc->w_large_pages = 0; | |
970 | ||
b27b7cbc MF |
971 | ocfs2_init_dealloc_ctxt(&wc->w_dealloc); |
972 | ||
3a307ffc | 973 | *wcp = wc; |
6af67d82 | 974 | |
3a307ffc | 975 | return 0; |
6af67d82 MF |
976 | } |
977 | ||
9517bac6 | 978 | /* |
3a307ffc MF |
979 | * If a page has any new buffers, zero them out here, and mark them uptodate |
980 | * and dirty so they'll be written out (in order to prevent uninitialised | |
981 | * block data from leaking). And clear the new bit. | |
9517bac6 | 982 | */ |
3a307ffc | 983 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) |
9517bac6 | 984 | { |
3a307ffc MF |
985 | unsigned int block_start, block_end; |
986 | struct buffer_head *head, *bh; | |
9517bac6 | 987 | |
3a307ffc MF |
988 | BUG_ON(!PageLocked(page)); |
989 | if (!page_has_buffers(page)) | |
990 | return; | |
9517bac6 | 991 | |
3a307ffc MF |
992 | bh = head = page_buffers(page); |
993 | block_start = 0; | |
994 | do { | |
995 | block_end = block_start + bh->b_size; | |
996 | ||
997 | if (buffer_new(bh)) { | |
998 | if (block_end > from && block_start < to) { | |
999 | if (!PageUptodate(page)) { | |
1000 | unsigned start, end; | |
3a307ffc MF |
1001 | |
1002 | start = max(from, block_start); | |
1003 | end = min(to, block_end); | |
1004 | ||
eebd2aa3 | 1005 | zero_user_segment(page, start, end); |
3a307ffc MF |
1006 | set_buffer_uptodate(bh); |
1007 | } | |
1008 | ||
1009 | clear_buffer_new(bh); | |
1010 | mark_buffer_dirty(bh); | |
1011 | } | |
1012 | } | |
9517bac6 | 1013 | |
3a307ffc MF |
1014 | block_start = block_end; |
1015 | bh = bh->b_this_page; | |
1016 | } while (bh != head); | |
1017 | } | |
1018 | ||
1019 | /* | |
1020 | * Only called when we have a failure during allocating write to write | |
1021 | * zero's to the newly allocated region. | |
1022 | */ | |
1023 | static void ocfs2_write_failure(struct inode *inode, | |
1024 | struct ocfs2_write_ctxt *wc, | |
1025 | loff_t user_pos, unsigned user_len) | |
1026 | { | |
1027 | int i; | |
5c26a7b7 MF |
1028 | unsigned from = user_pos & (PAGE_CACHE_SIZE - 1), |
1029 | to = user_pos + user_len; | |
3a307ffc MF |
1030 | struct page *tmppage; |
1031 | ||
5c26a7b7 | 1032 | ocfs2_zero_new_buffers(wc->w_target_page, from, to); |
9517bac6 | 1033 | |
3a307ffc MF |
1034 | for(i = 0; i < wc->w_num_pages; i++) { |
1035 | tmppage = wc->w_pages[i]; | |
9517bac6 | 1036 | |
961cecbe | 1037 | if (page_has_buffers(tmppage)) { |
53ef99ca | 1038 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 1039 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
1040 | |
1041 | block_commit_write(tmppage, from, to); | |
1042 | } | |
9517bac6 | 1043 | } |
9517bac6 MF |
1044 | } |
1045 | ||
3a307ffc MF |
1046 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, |
1047 | struct ocfs2_write_ctxt *wc, | |
1048 | struct page *page, u32 cpos, | |
1049 | loff_t user_pos, unsigned user_len, | |
1050 | int new) | |
9517bac6 | 1051 | { |
3a307ffc MF |
1052 | int ret; |
1053 | unsigned int map_from = 0, map_to = 0; | |
9517bac6 | 1054 | unsigned int cluster_start, cluster_end; |
3a307ffc | 1055 | unsigned int user_data_from = 0, user_data_to = 0; |
9517bac6 | 1056 | |
3a307ffc | 1057 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, |
9517bac6 MF |
1058 | &cluster_start, &cluster_end); |
1059 | ||
272b62c1 GR |
1060 | /* treat the write as new if the a hole/lseek spanned across |
1061 | * the page boundary. | |
1062 | */ | |
1063 | new = new | ((i_size_read(inode) <= page_offset(page)) && | |
1064 | (page_offset(page) <= user_pos)); | |
1065 | ||
3a307ffc MF |
1066 | if (page == wc->w_target_page) { |
1067 | map_from = user_pos & (PAGE_CACHE_SIZE - 1); | |
1068 | map_to = map_from + user_len; | |
1069 | ||
1070 | if (new) | |
1071 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1072 | cluster_start, cluster_end, | |
1073 | new); | |
1074 | else | |
1075 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1076 | map_from, map_to, new); | |
1077 | if (ret) { | |
9517bac6 MF |
1078 | mlog_errno(ret); |
1079 | goto out; | |
1080 | } | |
1081 | ||
3a307ffc MF |
1082 | user_data_from = map_from; |
1083 | user_data_to = map_to; | |
9517bac6 | 1084 | if (new) { |
3a307ffc MF |
1085 | map_from = cluster_start; |
1086 | map_to = cluster_end; | |
9517bac6 MF |
1087 | } |
1088 | } else { | |
1089 | /* | |
1090 | * If we haven't allocated the new page yet, we | |
1091 | * shouldn't be writing it out without copying user | |
1092 | * data. This is likely a math error from the caller. | |
1093 | */ | |
1094 | BUG_ON(!new); | |
1095 | ||
3a307ffc MF |
1096 | map_from = cluster_start; |
1097 | map_to = cluster_end; | |
9517bac6 MF |
1098 | |
1099 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
3a307ffc | 1100 | cluster_start, cluster_end, new); |
9517bac6 MF |
1101 | if (ret) { |
1102 | mlog_errno(ret); | |
1103 | goto out; | |
1104 | } | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * Parts of newly allocated pages need to be zero'd. | |
1109 | * | |
1110 | * Above, we have also rewritten 'to' and 'from' - as far as | |
1111 | * the rest of the function is concerned, the entire cluster | |
1112 | * range inside of a page needs to be written. | |
1113 | * | |
1114 | * We can skip this if the page is up to date - it's already | |
1115 | * been zero'd from being read in as a hole. | |
1116 | */ | |
1117 | if (new && !PageUptodate(page)) | |
1118 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), | |
3a307ffc | 1119 | cpos, user_data_from, user_data_to); |
9517bac6 MF |
1120 | |
1121 | flush_dcache_page(page); | |
1122 | ||
9517bac6 | 1123 | out: |
3a307ffc | 1124 | return ret; |
9517bac6 MF |
1125 | } |
1126 | ||
1127 | /* | |
3a307ffc | 1128 | * This function will only grab one clusters worth of pages. |
9517bac6 | 1129 | */ |
3a307ffc MF |
1130 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, |
1131 | struct ocfs2_write_ctxt *wc, | |
693c241a JB |
1132 | u32 cpos, loff_t user_pos, |
1133 | unsigned user_len, int new, | |
7307de80 | 1134 | struct page *mmap_page) |
9517bac6 | 1135 | { |
3a307ffc | 1136 | int ret = 0, i; |
693c241a | 1137 | unsigned long start, target_index, end_index, index; |
9517bac6 | 1138 | struct inode *inode = mapping->host; |
693c241a | 1139 | loff_t last_byte; |
9517bac6 | 1140 | |
3a307ffc | 1141 | target_index = user_pos >> PAGE_CACHE_SHIFT; |
9517bac6 MF |
1142 | |
1143 | /* | |
1144 | * Figure out how many pages we'll be manipulating here. For | |
60b11392 | 1145 | * non allocating write, we just change the one |
693c241a JB |
1146 | * page. Otherwise, we'll need a whole clusters worth. If we're |
1147 | * writing past i_size, we only need enough pages to cover the | |
1148 | * last page of the write. | |
9517bac6 | 1149 | */ |
9517bac6 | 1150 | if (new) { |
3a307ffc MF |
1151 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); |
1152 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); | |
693c241a JB |
1153 | /* |
1154 | * We need the index *past* the last page we could possibly | |
1155 | * touch. This is the page past the end of the write or | |
1156 | * i_size, whichever is greater. | |
1157 | */ | |
1158 | last_byte = max(user_pos + user_len, i_size_read(inode)); | |
1159 | BUG_ON(last_byte < 1); | |
1160 | end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1; | |
1161 | if ((start + wc->w_num_pages) > end_index) | |
1162 | wc->w_num_pages = end_index - start; | |
9517bac6 | 1163 | } else { |
3a307ffc MF |
1164 | wc->w_num_pages = 1; |
1165 | start = target_index; | |
9517bac6 MF |
1166 | } |
1167 | ||
3a307ffc | 1168 | for(i = 0; i < wc->w_num_pages; i++) { |
9517bac6 MF |
1169 | index = start + i; |
1170 | ||
7307de80 MF |
1171 | if (index == target_index && mmap_page) { |
1172 | /* | |
1173 | * ocfs2_pagemkwrite() is a little different | |
1174 | * and wants us to directly use the page | |
1175 | * passed in. | |
1176 | */ | |
1177 | lock_page(mmap_page); | |
1178 | ||
5cffff9e | 1179 | /* Exit and let the caller retry */ |
7307de80 | 1180 | if (mmap_page->mapping != mapping) { |
5cffff9e | 1181 | WARN_ON(mmap_page->mapping); |
7307de80 | 1182 | unlock_page(mmap_page); |
5cffff9e | 1183 | ret = -EAGAIN; |
7307de80 MF |
1184 | goto out; |
1185 | } | |
1186 | ||
1187 | page_cache_get(mmap_page); | |
1188 | wc->w_pages[i] = mmap_page; | |
5cffff9e | 1189 | wc->w_target_locked = true; |
7307de80 MF |
1190 | } else { |
1191 | wc->w_pages[i] = find_or_create_page(mapping, index, | |
1192 | GFP_NOFS); | |
1193 | if (!wc->w_pages[i]) { | |
1194 | ret = -ENOMEM; | |
1195 | mlog_errno(ret); | |
1196 | goto out; | |
1197 | } | |
9517bac6 | 1198 | } |
1269529b | 1199 | wait_for_stable_page(wc->w_pages[i]); |
3a307ffc MF |
1200 | |
1201 | if (index == target_index) | |
1202 | wc->w_target_page = wc->w_pages[i]; | |
9517bac6 | 1203 | } |
3a307ffc | 1204 | out: |
5cffff9e WW |
1205 | if (ret) |
1206 | wc->w_target_locked = false; | |
3a307ffc MF |
1207 | return ret; |
1208 | } | |
1209 | ||
1210 | /* | |
1211 | * Prepare a single cluster for write one cluster into the file. | |
1212 | */ | |
1213 | static int ocfs2_write_cluster(struct address_space *mapping, | |
b27b7cbc | 1214 | u32 phys, unsigned int unwritten, |
e7432675 | 1215 | unsigned int should_zero, |
b27b7cbc | 1216 | struct ocfs2_alloc_context *data_ac, |
3a307ffc MF |
1217 | struct ocfs2_alloc_context *meta_ac, |
1218 | struct ocfs2_write_ctxt *wc, u32 cpos, | |
1219 | loff_t user_pos, unsigned user_len) | |
1220 | { | |
e7432675 | 1221 | int ret, i, new; |
3a307ffc MF |
1222 | u64 v_blkno, p_blkno; |
1223 | struct inode *inode = mapping->host; | |
f99b9b7c | 1224 | struct ocfs2_extent_tree et; |
3a307ffc MF |
1225 | |
1226 | new = phys == 0 ? 1 : 0; | |
9517bac6 | 1227 | if (new) { |
3a307ffc MF |
1228 | u32 tmp_pos; |
1229 | ||
9517bac6 MF |
1230 | /* |
1231 | * This is safe to call with the page locks - it won't take | |
1232 | * any additional semaphores or cluster locks. | |
1233 | */ | |
3a307ffc | 1234 | tmp_pos = cpos; |
0eb8d47e TM |
1235 | ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode, |
1236 | &tmp_pos, 1, 0, wc->w_di_bh, | |
1237 | wc->w_handle, data_ac, | |
1238 | meta_ac, NULL); | |
9517bac6 MF |
1239 | /* |
1240 | * This shouldn't happen because we must have already | |
1241 | * calculated the correct meta data allocation required. The | |
1242 | * internal tree allocation code should know how to increase | |
1243 | * transaction credits itself. | |
1244 | * | |
1245 | * If need be, we could handle -EAGAIN for a | |
1246 | * RESTART_TRANS here. | |
1247 | */ | |
1248 | mlog_bug_on_msg(ret == -EAGAIN, | |
1249 | "Inode %llu: EAGAIN return during allocation.\n", | |
1250 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1251 | if (ret < 0) { | |
1252 | mlog_errno(ret); | |
1253 | goto out; | |
1254 | } | |
b27b7cbc | 1255 | } else if (unwritten) { |
5e404e9e JB |
1256 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
1257 | wc->w_di_bh); | |
f99b9b7c | 1258 | ret = ocfs2_mark_extent_written(inode, &et, |
b27b7cbc | 1259 | wc->w_handle, cpos, 1, phys, |
f99b9b7c | 1260 | meta_ac, &wc->w_dealloc); |
b27b7cbc MF |
1261 | if (ret < 0) { |
1262 | mlog_errno(ret); | |
1263 | goto out; | |
1264 | } | |
1265 | } | |
3a307ffc | 1266 | |
b27b7cbc | 1267 | if (should_zero) |
3a307ffc | 1268 | v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos); |
b27b7cbc | 1269 | else |
3a307ffc | 1270 | v_blkno = user_pos >> inode->i_sb->s_blocksize_bits; |
9517bac6 | 1271 | |
3a307ffc MF |
1272 | /* |
1273 | * The only reason this should fail is due to an inability to | |
1274 | * find the extent added. | |
1275 | */ | |
49cb8d2d MF |
1276 | ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL, |
1277 | NULL); | |
9517bac6 | 1278 | if (ret < 0) { |
3a307ffc MF |
1279 | ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, " |
1280 | "at logical block %llu", | |
1281 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1282 | (unsigned long long)v_blkno); | |
9517bac6 MF |
1283 | goto out; |
1284 | } | |
1285 | ||
1286 | BUG_ON(p_blkno == 0); | |
1287 | ||
3a307ffc MF |
1288 | for(i = 0; i < wc->w_num_pages; i++) { |
1289 | int tmpret; | |
9517bac6 | 1290 | |
3a307ffc MF |
1291 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, |
1292 | wc->w_pages[i], cpos, | |
b27b7cbc MF |
1293 | user_pos, user_len, |
1294 | should_zero); | |
3a307ffc MF |
1295 | if (tmpret) { |
1296 | mlog_errno(tmpret); | |
1297 | if (ret == 0) | |
cbfa9639 | 1298 | ret = tmpret; |
3a307ffc | 1299 | } |
9517bac6 MF |
1300 | } |
1301 | ||
3a307ffc MF |
1302 | /* |
1303 | * We only have cleanup to do in case of allocating write. | |
1304 | */ | |
1305 | if (ret && new) | |
1306 | ocfs2_write_failure(inode, wc, user_pos, user_len); | |
1307 | ||
9517bac6 | 1308 | out: |
9517bac6 | 1309 | |
3a307ffc | 1310 | return ret; |
9517bac6 MF |
1311 | } |
1312 | ||
0d172baa MF |
1313 | static int ocfs2_write_cluster_by_desc(struct address_space *mapping, |
1314 | struct ocfs2_alloc_context *data_ac, | |
1315 | struct ocfs2_alloc_context *meta_ac, | |
1316 | struct ocfs2_write_ctxt *wc, | |
1317 | loff_t pos, unsigned len) | |
1318 | { | |
1319 | int ret, i; | |
db56246c MF |
1320 | loff_t cluster_off; |
1321 | unsigned int local_len = len; | |
0d172baa | 1322 | struct ocfs2_write_cluster_desc *desc; |
db56246c | 1323 | struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb); |
0d172baa MF |
1324 | |
1325 | for (i = 0; i < wc->w_clen; i++) { | |
1326 | desc = &wc->w_desc[i]; | |
1327 | ||
db56246c MF |
1328 | /* |
1329 | * We have to make sure that the total write passed in | |
1330 | * doesn't extend past a single cluster. | |
1331 | */ | |
1332 | local_len = len; | |
1333 | cluster_off = pos & (osb->s_clustersize - 1); | |
1334 | if ((cluster_off + local_len) > osb->s_clustersize) | |
1335 | local_len = osb->s_clustersize - cluster_off; | |
1336 | ||
b27b7cbc | 1337 | ret = ocfs2_write_cluster(mapping, desc->c_phys, |
e7432675 SM |
1338 | desc->c_unwritten, |
1339 | desc->c_needs_zero, | |
1340 | data_ac, meta_ac, | |
db56246c | 1341 | wc, desc->c_cpos, pos, local_len); |
0d172baa MF |
1342 | if (ret) { |
1343 | mlog_errno(ret); | |
1344 | goto out; | |
1345 | } | |
db56246c MF |
1346 | |
1347 | len -= local_len; | |
1348 | pos += local_len; | |
0d172baa MF |
1349 | } |
1350 | ||
1351 | ret = 0; | |
1352 | out: | |
1353 | return ret; | |
1354 | } | |
1355 | ||
3a307ffc MF |
1356 | /* |
1357 | * ocfs2_write_end() wants to know which parts of the target page it | |
1358 | * should complete the write on. It's easiest to compute them ahead of | |
1359 | * time when a more complete view of the write is available. | |
1360 | */ | |
1361 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, | |
1362 | struct ocfs2_write_ctxt *wc, | |
1363 | loff_t pos, unsigned len, int alloc) | |
9517bac6 | 1364 | { |
3a307ffc | 1365 | struct ocfs2_write_cluster_desc *desc; |
9517bac6 | 1366 | |
3a307ffc MF |
1367 | wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1); |
1368 | wc->w_target_to = wc->w_target_from + len; | |
1369 | ||
1370 | if (alloc == 0) | |
1371 | return; | |
1372 | ||
1373 | /* | |
1374 | * Allocating write - we may have different boundaries based | |
1375 | * on page size and cluster size. | |
1376 | * | |
1377 | * NOTE: We can no longer compute one value from the other as | |
1378 | * the actual write length and user provided length may be | |
1379 | * different. | |
1380 | */ | |
9517bac6 | 1381 | |
3a307ffc MF |
1382 | if (wc->w_large_pages) { |
1383 | /* | |
1384 | * We only care about the 1st and last cluster within | |
b27b7cbc | 1385 | * our range and whether they should be zero'd or not. Either |
3a307ffc MF |
1386 | * value may be extended out to the start/end of a |
1387 | * newly allocated cluster. | |
1388 | */ | |
1389 | desc = &wc->w_desc[0]; | |
e7432675 | 1390 | if (desc->c_needs_zero) |
3a307ffc MF |
1391 | ocfs2_figure_cluster_boundaries(osb, |
1392 | desc->c_cpos, | |
1393 | &wc->w_target_from, | |
1394 | NULL); | |
1395 | ||
1396 | desc = &wc->w_desc[wc->w_clen - 1]; | |
e7432675 | 1397 | if (desc->c_needs_zero) |
3a307ffc MF |
1398 | ocfs2_figure_cluster_boundaries(osb, |
1399 | desc->c_cpos, | |
1400 | NULL, | |
1401 | &wc->w_target_to); | |
1402 | } else { | |
1403 | wc->w_target_from = 0; | |
1404 | wc->w_target_to = PAGE_CACHE_SIZE; | |
1405 | } | |
9517bac6 MF |
1406 | } |
1407 | ||
0d172baa MF |
1408 | /* |
1409 | * Populate each single-cluster write descriptor in the write context | |
1410 | * with information about the i/o to be done. | |
b27b7cbc MF |
1411 | * |
1412 | * Returns the number of clusters that will have to be allocated, as | |
1413 | * well as a worst case estimate of the number of extent records that | |
1414 | * would have to be created during a write to an unwritten region. | |
0d172baa MF |
1415 | */ |
1416 | static int ocfs2_populate_write_desc(struct inode *inode, | |
1417 | struct ocfs2_write_ctxt *wc, | |
b27b7cbc MF |
1418 | unsigned int *clusters_to_alloc, |
1419 | unsigned int *extents_to_split) | |
9517bac6 | 1420 | { |
0d172baa | 1421 | int ret; |
3a307ffc | 1422 | struct ocfs2_write_cluster_desc *desc; |
0d172baa | 1423 | unsigned int num_clusters = 0; |
b27b7cbc | 1424 | unsigned int ext_flags = 0; |
0d172baa MF |
1425 | u32 phys = 0; |
1426 | int i; | |
9517bac6 | 1427 | |
b27b7cbc MF |
1428 | *clusters_to_alloc = 0; |
1429 | *extents_to_split = 0; | |
1430 | ||
3a307ffc MF |
1431 | for (i = 0; i < wc->w_clen; i++) { |
1432 | desc = &wc->w_desc[i]; | |
1433 | desc->c_cpos = wc->w_cpos + i; | |
1434 | ||
1435 | if (num_clusters == 0) { | |
b27b7cbc MF |
1436 | /* |
1437 | * Need to look up the next extent record. | |
1438 | */ | |
3a307ffc | 1439 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, |
b27b7cbc | 1440 | &num_clusters, &ext_flags); |
3a307ffc MF |
1441 | if (ret) { |
1442 | mlog_errno(ret); | |
607d44aa | 1443 | goto out; |
3a307ffc | 1444 | } |
b27b7cbc | 1445 | |
293b2f70 TM |
1446 | /* We should already CoW the refcountd extent. */ |
1447 | BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED); | |
1448 | ||
b27b7cbc MF |
1449 | /* |
1450 | * Assume worst case - that we're writing in | |
1451 | * the middle of the extent. | |
1452 | * | |
1453 | * We can assume that the write proceeds from | |
1454 | * left to right, in which case the extent | |
1455 | * insert code is smart enough to coalesce the | |
1456 | * next splits into the previous records created. | |
1457 | */ | |
1458 | if (ext_flags & OCFS2_EXT_UNWRITTEN) | |
1459 | *extents_to_split = *extents_to_split + 2; | |
3a307ffc MF |
1460 | } else if (phys) { |
1461 | /* | |
1462 | * Only increment phys if it doesn't describe | |
1463 | * a hole. | |
1464 | */ | |
1465 | phys++; | |
1466 | } | |
1467 | ||
e7432675 SM |
1468 | /* |
1469 | * If w_first_new_cpos is < UINT_MAX, we have a non-sparse | |
1470 | * file that got extended. w_first_new_cpos tells us | |
1471 | * where the newly allocated clusters are so we can | |
1472 | * zero them. | |
1473 | */ | |
1474 | if (desc->c_cpos >= wc->w_first_new_cpos) { | |
1475 | BUG_ON(phys == 0); | |
1476 | desc->c_needs_zero = 1; | |
1477 | } | |
1478 | ||
3a307ffc MF |
1479 | desc->c_phys = phys; |
1480 | if (phys == 0) { | |
1481 | desc->c_new = 1; | |
e7432675 | 1482 | desc->c_needs_zero = 1; |
0d172baa | 1483 | *clusters_to_alloc = *clusters_to_alloc + 1; |
3a307ffc | 1484 | } |
e7432675 SM |
1485 | |
1486 | if (ext_flags & OCFS2_EXT_UNWRITTEN) { | |
b27b7cbc | 1487 | desc->c_unwritten = 1; |
e7432675 SM |
1488 | desc->c_needs_zero = 1; |
1489 | } | |
3a307ffc MF |
1490 | |
1491 | num_clusters--; | |
9517bac6 MF |
1492 | } |
1493 | ||
0d172baa MF |
1494 | ret = 0; |
1495 | out: | |
1496 | return ret; | |
1497 | } | |
1498 | ||
1afc32b9 MF |
1499 | static int ocfs2_write_begin_inline(struct address_space *mapping, |
1500 | struct inode *inode, | |
1501 | struct ocfs2_write_ctxt *wc) | |
1502 | { | |
1503 | int ret; | |
1504 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1505 | struct page *page; | |
1506 | handle_t *handle; | |
1507 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1508 | ||
1509 | page = find_or_create_page(mapping, 0, GFP_NOFS); | |
1510 | if (!page) { | |
1511 | ret = -ENOMEM; | |
1512 | mlog_errno(ret); | |
1513 | goto out; | |
1514 | } | |
1515 | /* | |
1516 | * If we don't set w_num_pages then this page won't get unlocked | |
1517 | * and freed on cleanup of the write context. | |
1518 | */ | |
1519 | wc->w_pages[0] = wc->w_target_page = page; | |
1520 | wc->w_num_pages = 1; | |
1521 | ||
1522 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1523 | if (IS_ERR(handle)) { | |
1524 | ret = PTR_ERR(handle); | |
1525 | mlog_errno(ret); | |
1526 | goto out; | |
1527 | } | |
1528 | ||
0cf2f763 | 1529 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 1530 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
1531 | if (ret) { |
1532 | ocfs2_commit_trans(osb, handle); | |
1533 | ||
1534 | mlog_errno(ret); | |
1535 | goto out; | |
1536 | } | |
1537 | ||
1538 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
1539 | ocfs2_set_inode_data_inline(inode, di); | |
1540 | ||
1541 | if (!PageUptodate(page)) { | |
1542 | ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh); | |
1543 | if (ret) { | |
1544 | ocfs2_commit_trans(osb, handle); | |
1545 | ||
1546 | goto out; | |
1547 | } | |
1548 | } | |
1549 | ||
1550 | wc->w_handle = handle; | |
1551 | out: | |
1552 | return ret; | |
1553 | } | |
1554 | ||
1555 | int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size) | |
1556 | { | |
1557 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1558 | ||
0d8a4e0c | 1559 | if (new_size <= le16_to_cpu(di->id2.i_data.id_count)) |
1afc32b9 MF |
1560 | return 1; |
1561 | return 0; | |
1562 | } | |
1563 | ||
1564 | static int ocfs2_try_to_write_inline_data(struct address_space *mapping, | |
1565 | struct inode *inode, loff_t pos, | |
1566 | unsigned len, struct page *mmap_page, | |
1567 | struct ocfs2_write_ctxt *wc) | |
1568 | { | |
1569 | int ret, written = 0; | |
1570 | loff_t end = pos + len; | |
1571 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
d9ae49d6 | 1572 | struct ocfs2_dinode *di = NULL; |
1afc32b9 | 1573 | |
9558156b TM |
1574 | trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno, |
1575 | len, (unsigned long long)pos, | |
1576 | oi->ip_dyn_features); | |
1afc32b9 MF |
1577 | |
1578 | /* | |
1579 | * Handle inodes which already have inline data 1st. | |
1580 | */ | |
1581 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1582 | if (mmap_page == NULL && | |
1583 | ocfs2_size_fits_inline_data(wc->w_di_bh, end)) | |
1584 | goto do_inline_write; | |
1585 | ||
1586 | /* | |
1587 | * The write won't fit - we have to give this inode an | |
1588 | * inline extent list now. | |
1589 | */ | |
1590 | ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh); | |
1591 | if (ret) | |
1592 | mlog_errno(ret); | |
1593 | goto out; | |
1594 | } | |
1595 | ||
1596 | /* | |
1597 | * Check whether the inode can accept inline data. | |
1598 | */ | |
1599 | if (oi->ip_clusters != 0 || i_size_read(inode) != 0) | |
1600 | return 0; | |
1601 | ||
1602 | /* | |
1603 | * Check whether the write can fit. | |
1604 | */ | |
d9ae49d6 TY |
1605 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; |
1606 | if (mmap_page || | |
1607 | end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) | |
1afc32b9 MF |
1608 | return 0; |
1609 | ||
1610 | do_inline_write: | |
1611 | ret = ocfs2_write_begin_inline(mapping, inode, wc); | |
1612 | if (ret) { | |
1613 | mlog_errno(ret); | |
1614 | goto out; | |
1615 | } | |
1616 | ||
1617 | /* | |
1618 | * This signals to the caller that the data can be written | |
1619 | * inline. | |
1620 | */ | |
1621 | written = 1; | |
1622 | out: | |
1623 | return written ? written : ret; | |
1624 | } | |
1625 | ||
65ed39d6 MF |
1626 | /* |
1627 | * This function only does anything for file systems which can't | |
1628 | * handle sparse files. | |
1629 | * | |
1630 | * What we want to do here is fill in any hole between the current end | |
1631 | * of allocation and the end of our write. That way the rest of the | |
1632 | * write path can treat it as an non-allocating write, which has no | |
1633 | * special case code for sparse/nonsparse files. | |
1634 | */ | |
5693486b JB |
1635 | static int ocfs2_expand_nonsparse_inode(struct inode *inode, |
1636 | struct buffer_head *di_bh, | |
1637 | loff_t pos, unsigned len, | |
65ed39d6 MF |
1638 | struct ocfs2_write_ctxt *wc) |
1639 | { | |
1640 | int ret; | |
65ed39d6 MF |
1641 | loff_t newsize = pos + len; |
1642 | ||
5693486b | 1643 | BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); |
65ed39d6 MF |
1644 | |
1645 | if (newsize <= i_size_read(inode)) | |
1646 | return 0; | |
1647 | ||
5693486b | 1648 | ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos); |
65ed39d6 MF |
1649 | if (ret) |
1650 | mlog_errno(ret); | |
1651 | ||
e7432675 SM |
1652 | wc->w_first_new_cpos = |
1653 | ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)); | |
1654 | ||
65ed39d6 MF |
1655 | return ret; |
1656 | } | |
1657 | ||
5693486b JB |
1658 | static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh, |
1659 | loff_t pos) | |
1660 | { | |
1661 | int ret = 0; | |
1662 | ||
1663 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); | |
1664 | if (pos > i_size_read(inode)) | |
1665 | ret = ocfs2_zero_extend(inode, di_bh, pos); | |
1666 | ||
1667 | return ret; | |
1668 | } | |
1669 | ||
50308d81 TM |
1670 | /* |
1671 | * Try to flush truncate logs if we can free enough clusters from it. | |
1672 | * As for return value, "< 0" means error, "0" no space and "1" means | |
1673 | * we have freed enough spaces and let the caller try to allocate again. | |
1674 | */ | |
1675 | static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb, | |
1676 | unsigned int needed) | |
1677 | { | |
1678 | tid_t target; | |
1679 | int ret = 0; | |
1680 | unsigned int truncated_clusters; | |
1681 | ||
1682 | mutex_lock(&osb->osb_tl_inode->i_mutex); | |
1683 | truncated_clusters = osb->truncated_clusters; | |
1684 | mutex_unlock(&osb->osb_tl_inode->i_mutex); | |
1685 | ||
1686 | /* | |
1687 | * Check whether we can succeed in allocating if we free | |
1688 | * the truncate log. | |
1689 | */ | |
1690 | if (truncated_clusters < needed) | |
1691 | goto out; | |
1692 | ||
1693 | ret = ocfs2_flush_truncate_log(osb); | |
1694 | if (ret) { | |
1695 | mlog_errno(ret); | |
1696 | goto out; | |
1697 | } | |
1698 | ||
1699 | if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) { | |
1700 | jbd2_log_wait_commit(osb->journal->j_journal, target); | |
1701 | ret = 1; | |
1702 | } | |
1703 | out: | |
1704 | return ret; | |
1705 | } | |
1706 | ||
0378da0f TM |
1707 | int ocfs2_write_begin_nolock(struct file *filp, |
1708 | struct address_space *mapping, | |
0d172baa MF |
1709 | loff_t pos, unsigned len, unsigned flags, |
1710 | struct page **pagep, void **fsdata, | |
1711 | struct buffer_head *di_bh, struct page *mmap_page) | |
1712 | { | |
e7432675 | 1713 | int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS; |
50308d81 | 1714 | unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0; |
0d172baa MF |
1715 | struct ocfs2_write_ctxt *wc; |
1716 | struct inode *inode = mapping->host; | |
1717 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1718 | struct ocfs2_dinode *di; | |
1719 | struct ocfs2_alloc_context *data_ac = NULL; | |
1720 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1721 | handle_t *handle; | |
f99b9b7c | 1722 | struct ocfs2_extent_tree et; |
50308d81 | 1723 | int try_free = 1, ret1; |
0d172baa | 1724 | |
50308d81 | 1725 | try_again: |
0d172baa MF |
1726 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh); |
1727 | if (ret) { | |
1728 | mlog_errno(ret); | |
1729 | return ret; | |
1730 | } | |
1731 | ||
1afc32b9 MF |
1732 | if (ocfs2_supports_inline_data(osb)) { |
1733 | ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len, | |
1734 | mmap_page, wc); | |
1735 | if (ret == 1) { | |
1736 | ret = 0; | |
1737 | goto success; | |
1738 | } | |
1739 | if (ret < 0) { | |
1740 | mlog_errno(ret); | |
1741 | goto out; | |
1742 | } | |
1743 | } | |
1744 | ||
5693486b JB |
1745 | if (ocfs2_sparse_alloc(osb)) |
1746 | ret = ocfs2_zero_tail(inode, di_bh, pos); | |
1747 | else | |
1748 | ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len, | |
1749 | wc); | |
65ed39d6 MF |
1750 | if (ret) { |
1751 | mlog_errno(ret); | |
1752 | goto out; | |
1753 | } | |
1754 | ||
293b2f70 TM |
1755 | ret = ocfs2_check_range_for_refcount(inode, pos, len); |
1756 | if (ret < 0) { | |
1757 | mlog_errno(ret); | |
1758 | goto out; | |
1759 | } else if (ret == 1) { | |
50308d81 | 1760 | clusters_need = wc->w_clen; |
15502712 | 1761 | ret = ocfs2_refcount_cow(inode, filp, di_bh, |
37f8a2bf | 1762 | wc->w_cpos, wc->w_clen, UINT_MAX); |
293b2f70 TM |
1763 | if (ret) { |
1764 | mlog_errno(ret); | |
1765 | goto out; | |
1766 | } | |
1767 | } | |
1768 | ||
b27b7cbc MF |
1769 | ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc, |
1770 | &extents_to_split); | |
0d172baa MF |
1771 | if (ret) { |
1772 | mlog_errno(ret); | |
1773 | goto out; | |
1774 | } | |
50308d81 | 1775 | clusters_need += clusters_to_alloc; |
0d172baa MF |
1776 | |
1777 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1778 | ||
9558156b TM |
1779 | trace_ocfs2_write_begin_nolock( |
1780 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1781 | (long long)i_size_read(inode), | |
1782 | le32_to_cpu(di->i_clusters), | |
1783 | pos, len, flags, mmap_page, | |
1784 | clusters_to_alloc, extents_to_split); | |
1785 | ||
3a307ffc MF |
1786 | /* |
1787 | * We set w_target_from, w_target_to here so that | |
1788 | * ocfs2_write_end() knows which range in the target page to | |
1789 | * write out. An allocation requires that we write the entire | |
1790 | * cluster range. | |
1791 | */ | |
b27b7cbc | 1792 | if (clusters_to_alloc || extents_to_split) { |
3a307ffc MF |
1793 | /* |
1794 | * XXX: We are stretching the limits of | |
b27b7cbc | 1795 | * ocfs2_lock_allocators(). It greatly over-estimates |
3a307ffc MF |
1796 | * the work to be done. |
1797 | */ | |
5e404e9e JB |
1798 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
1799 | wc->w_di_bh); | |
f99b9b7c | 1800 | ret = ocfs2_lock_allocators(inode, &et, |
231b87d1 | 1801 | clusters_to_alloc, extents_to_split, |
f99b9b7c | 1802 | &data_ac, &meta_ac); |
9517bac6 MF |
1803 | if (ret) { |
1804 | mlog_errno(ret); | |
607d44aa | 1805 | goto out; |
9517bac6 MF |
1806 | } |
1807 | ||
4fe370af MF |
1808 | if (data_ac) |
1809 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; | |
1810 | ||
811f933d TM |
1811 | credits = ocfs2_calc_extend_credits(inode->i_sb, |
1812 | &di->id2.i_list, | |
3a307ffc MF |
1813 | clusters_to_alloc); |
1814 | ||
9517bac6 MF |
1815 | } |
1816 | ||
e7432675 SM |
1817 | /* |
1818 | * We have to zero sparse allocated clusters, unwritten extent clusters, | |
1819 | * and non-sparse clusters we just extended. For non-sparse writes, | |
1820 | * we know zeros will only be needed in the first and/or last cluster. | |
1821 | */ | |
1822 | if (clusters_to_alloc || extents_to_split || | |
8379e7c4 SM |
1823 | (wc->w_clen && (wc->w_desc[0].c_needs_zero || |
1824 | wc->w_desc[wc->w_clen - 1].c_needs_zero))) | |
e7432675 SM |
1825 | cluster_of_pages = 1; |
1826 | else | |
1827 | cluster_of_pages = 0; | |
1828 | ||
1829 | ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages); | |
3a307ffc | 1830 | |
9517bac6 MF |
1831 | handle = ocfs2_start_trans(osb, credits); |
1832 | if (IS_ERR(handle)) { | |
1833 | ret = PTR_ERR(handle); | |
1834 | mlog_errno(ret); | |
607d44aa | 1835 | goto out; |
9517bac6 MF |
1836 | } |
1837 | ||
3a307ffc MF |
1838 | wc->w_handle = handle; |
1839 | ||
5dd4056d CH |
1840 | if (clusters_to_alloc) { |
1841 | ret = dquot_alloc_space_nodirty(inode, | |
1842 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); | |
1843 | if (ret) | |
1844 | goto out_commit; | |
a90714c1 | 1845 | } |
3a307ffc MF |
1846 | /* |
1847 | * We don't want this to fail in ocfs2_write_end(), so do it | |
1848 | * here. | |
1849 | */ | |
0cf2f763 | 1850 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 1851 | OCFS2_JOURNAL_ACCESS_WRITE); |
3a307ffc | 1852 | if (ret) { |
9517bac6 | 1853 | mlog_errno(ret); |
a90714c1 | 1854 | goto out_quota; |
9517bac6 MF |
1855 | } |
1856 | ||
3a307ffc MF |
1857 | /* |
1858 | * Fill our page array first. That way we've grabbed enough so | |
1859 | * that we can zero and flush if we error after adding the | |
1860 | * extent. | |
1861 | */ | |
693c241a | 1862 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len, |
e7432675 | 1863 | cluster_of_pages, mmap_page); |
5cffff9e | 1864 | if (ret && ret != -EAGAIN) { |
9517bac6 | 1865 | mlog_errno(ret); |
a90714c1 | 1866 | goto out_quota; |
9517bac6 MF |
1867 | } |
1868 | ||
5cffff9e WW |
1869 | /* |
1870 | * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock | |
1871 | * the target page. In this case, we exit with no error and no target | |
1872 | * page. This will trigger the caller, page_mkwrite(), to re-try | |
1873 | * the operation. | |
1874 | */ | |
1875 | if (ret == -EAGAIN) { | |
1876 | BUG_ON(wc->w_target_page); | |
1877 | ret = 0; | |
1878 | goto out_quota; | |
1879 | } | |
1880 | ||
0d172baa MF |
1881 | ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos, |
1882 | len); | |
1883 | if (ret) { | |
1884 | mlog_errno(ret); | |
a90714c1 | 1885 | goto out_quota; |
9517bac6 | 1886 | } |
9517bac6 | 1887 | |
3a307ffc MF |
1888 | if (data_ac) |
1889 | ocfs2_free_alloc_context(data_ac); | |
1890 | if (meta_ac) | |
1891 | ocfs2_free_alloc_context(meta_ac); | |
9517bac6 | 1892 | |
1afc32b9 | 1893 | success: |
3a307ffc MF |
1894 | *pagep = wc->w_target_page; |
1895 | *fsdata = wc; | |
1896 | return 0; | |
a90714c1 JK |
1897 | out_quota: |
1898 | if (clusters_to_alloc) | |
5dd4056d | 1899 | dquot_free_space(inode, |
a90714c1 | 1900 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); |
9517bac6 MF |
1901 | out_commit: |
1902 | ocfs2_commit_trans(osb, handle); | |
1903 | ||
9517bac6 | 1904 | out: |
3a307ffc MF |
1905 | ocfs2_free_write_ctxt(wc); |
1906 | ||
9517bac6 MF |
1907 | if (data_ac) |
1908 | ocfs2_free_alloc_context(data_ac); | |
1909 | if (meta_ac) | |
1910 | ocfs2_free_alloc_context(meta_ac); | |
50308d81 TM |
1911 | |
1912 | if (ret == -ENOSPC && try_free) { | |
1913 | /* | |
1914 | * Try to free some truncate log so that we can have enough | |
1915 | * clusters to allocate. | |
1916 | */ | |
1917 | try_free = 0; | |
1918 | ||
1919 | ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); | |
1920 | if (ret1 == 1) | |
1921 | goto try_again; | |
1922 | ||
1923 | if (ret1 < 0) | |
1924 | mlog_errno(ret1); | |
1925 | } | |
1926 | ||
3a307ffc MF |
1927 | return ret; |
1928 | } | |
1929 | ||
b6af1bcd NP |
1930 | static int ocfs2_write_begin(struct file *file, struct address_space *mapping, |
1931 | loff_t pos, unsigned len, unsigned flags, | |
1932 | struct page **pagep, void **fsdata) | |
607d44aa MF |
1933 | { |
1934 | int ret; | |
1935 | struct buffer_head *di_bh = NULL; | |
1936 | struct inode *inode = mapping->host; | |
1937 | ||
e63aecb6 | 1938 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
607d44aa MF |
1939 | if (ret) { |
1940 | mlog_errno(ret); | |
1941 | return ret; | |
1942 | } | |
1943 | ||
1944 | /* | |
1945 | * Take alloc sem here to prevent concurrent lookups. That way | |
1946 | * the mapping, zeroing and tree manipulation within | |
1947 | * ocfs2_write() will be safe against ->readpage(). This | |
1948 | * should also serve to lock out allocation from a shared | |
1949 | * writeable region. | |
1950 | */ | |
1951 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1952 | ||
0378da0f | 1953 | ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep, |
7307de80 | 1954 | fsdata, di_bh, NULL); |
607d44aa MF |
1955 | if (ret) { |
1956 | mlog_errno(ret); | |
c934a92d | 1957 | goto out_fail; |
607d44aa MF |
1958 | } |
1959 | ||
1960 | brelse(di_bh); | |
1961 | ||
1962 | return 0; | |
1963 | ||
607d44aa MF |
1964 | out_fail: |
1965 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1966 | ||
1967 | brelse(di_bh); | |
e63aecb6 | 1968 | ocfs2_inode_unlock(inode, 1); |
607d44aa MF |
1969 | |
1970 | return ret; | |
1971 | } | |
1972 | ||
1afc32b9 MF |
1973 | static void ocfs2_write_end_inline(struct inode *inode, loff_t pos, |
1974 | unsigned len, unsigned *copied, | |
1975 | struct ocfs2_dinode *di, | |
1976 | struct ocfs2_write_ctxt *wc) | |
1977 | { | |
1978 | void *kaddr; | |
1979 | ||
1980 | if (unlikely(*copied < len)) { | |
1981 | if (!PageUptodate(wc->w_target_page)) { | |
1982 | *copied = 0; | |
1983 | return; | |
1984 | } | |
1985 | } | |
1986 | ||
c4bc8dcb | 1987 | kaddr = kmap_atomic(wc->w_target_page); |
1afc32b9 | 1988 | memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied); |
c4bc8dcb | 1989 | kunmap_atomic(kaddr); |
1afc32b9 | 1990 | |
9558156b TM |
1991 | trace_ocfs2_write_end_inline( |
1992 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1afc32b9 MF |
1993 | (unsigned long long)pos, *copied, |
1994 | le16_to_cpu(di->id2.i_data.id_count), | |
1995 | le16_to_cpu(di->i_dyn_features)); | |
1996 | } | |
1997 | ||
7307de80 MF |
1998 | int ocfs2_write_end_nolock(struct address_space *mapping, |
1999 | loff_t pos, unsigned len, unsigned copied, | |
2000 | struct page *page, void *fsdata) | |
3a307ffc MF |
2001 | { |
2002 | int i; | |
2003 | unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1); | |
2004 | struct inode *inode = mapping->host; | |
2005 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2006 | struct ocfs2_write_ctxt *wc = fsdata; | |
2007 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
2008 | handle_t *handle = wc->w_handle; | |
2009 | struct page *tmppage; | |
2010 | ||
1afc32b9 MF |
2011 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
2012 | ocfs2_write_end_inline(inode, pos, len, &copied, di, wc); | |
2013 | goto out_write_size; | |
2014 | } | |
2015 | ||
3a307ffc MF |
2016 | if (unlikely(copied < len)) { |
2017 | if (!PageUptodate(wc->w_target_page)) | |
2018 | copied = 0; | |
2019 | ||
2020 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, | |
2021 | start+len); | |
2022 | } | |
2023 | flush_dcache_page(wc->w_target_page); | |
2024 | ||
2025 | for(i = 0; i < wc->w_num_pages; i++) { | |
2026 | tmppage = wc->w_pages[i]; | |
2027 | ||
2028 | if (tmppage == wc->w_target_page) { | |
2029 | from = wc->w_target_from; | |
2030 | to = wc->w_target_to; | |
2031 | ||
2032 | BUG_ON(from > PAGE_CACHE_SIZE || | |
2033 | to > PAGE_CACHE_SIZE || | |
2034 | to < from); | |
2035 | } else { | |
2036 | /* | |
2037 | * Pages adjacent to the target (if any) imply | |
2038 | * a hole-filling write in which case we want | |
2039 | * to flush their entire range. | |
2040 | */ | |
2041 | from = 0; | |
2042 | to = PAGE_CACHE_SIZE; | |
2043 | } | |
2044 | ||
961cecbe | 2045 | if (page_has_buffers(tmppage)) { |
53ef99ca | 2046 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 2047 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
2048 | block_commit_write(tmppage, from, to); |
2049 | } | |
3a307ffc MF |
2050 | } |
2051 | ||
1afc32b9 | 2052 | out_write_size: |
3a307ffc MF |
2053 | pos += copied; |
2054 | if (pos > inode->i_size) { | |
2055 | i_size_write(inode, pos); | |
2056 | mark_inode_dirty(inode); | |
2057 | } | |
2058 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
2059 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
2060 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
2061 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); | |
2062 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
3a307ffc MF |
2063 | ocfs2_journal_dirty(handle, wc->w_di_bh); |
2064 | ||
2065 | ocfs2_commit_trans(osb, handle); | |
59a5e416 | 2066 | |
b27b7cbc MF |
2067 | ocfs2_run_deallocs(osb, &wc->w_dealloc); |
2068 | ||
607d44aa MF |
2069 | ocfs2_free_write_ctxt(wc); |
2070 | ||
2071 | return copied; | |
2072 | } | |
2073 | ||
b6af1bcd NP |
2074 | static int ocfs2_write_end(struct file *file, struct address_space *mapping, |
2075 | loff_t pos, unsigned len, unsigned copied, | |
2076 | struct page *page, void *fsdata) | |
607d44aa MF |
2077 | { |
2078 | int ret; | |
2079 | struct inode *inode = mapping->host; | |
2080 | ||
2081 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata); | |
2082 | ||
3a307ffc | 2083 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
e63aecb6 | 2084 | ocfs2_inode_unlock(inode, 1); |
9517bac6 | 2085 | |
607d44aa | 2086 | return ret; |
9517bac6 MF |
2087 | } |
2088 | ||
f5e54d6e | 2089 | const struct address_space_operations ocfs2_aops = { |
1fca3a05 HH |
2090 | .readpage = ocfs2_readpage, |
2091 | .readpages = ocfs2_readpages, | |
2092 | .writepage = ocfs2_writepage, | |
2093 | .write_begin = ocfs2_write_begin, | |
2094 | .write_end = ocfs2_write_end, | |
2095 | .bmap = ocfs2_bmap, | |
1fca3a05 HH |
2096 | .direct_IO = ocfs2_direct_IO, |
2097 | .invalidatepage = ocfs2_invalidatepage, | |
2098 | .releasepage = ocfs2_releasepage, | |
2099 | .migratepage = buffer_migrate_page, | |
2100 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 2101 | .error_remove_page = generic_error_remove_page, |
ccd979bd | 2102 | }; |