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> |
ccd979bd MF |
30 | |
31 | #define MLOG_MASK_PREFIX ML_FILE_IO | |
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" | |
46 | ||
47 | #include "buffer_head_io.h" | |
48 | ||
49 | static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock, | |
50 | struct buffer_head *bh_result, int create) | |
51 | { | |
52 | int err = -EIO; | |
53 | int status; | |
54 | struct ocfs2_dinode *fe = NULL; | |
55 | struct buffer_head *bh = NULL; | |
56 | struct buffer_head *buffer_cache_bh = NULL; | |
57 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
58 | void *kaddr; | |
59 | ||
60 | mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode, | |
61 | (unsigned long long)iblock, bh_result, create); | |
62 | ||
63 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); | |
64 | ||
65 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { | |
66 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", | |
67 | (unsigned long long)iblock); | |
68 | goto bail; | |
69 | } | |
70 | ||
71 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
72 | OCFS2_I(inode)->ip_blkno, | |
73 | &bh, OCFS2_BH_CACHED, inode); | |
74 | if (status < 0) { | |
75 | mlog_errno(status); | |
76 | goto bail; | |
77 | } | |
78 | fe = (struct ocfs2_dinode *) bh->b_data; | |
79 | ||
80 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
b0697053 | 81 | mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n", |
1ca1a111 MF |
82 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, |
83 | fe->i_signature); | |
ccd979bd MF |
84 | goto bail; |
85 | } | |
86 | ||
87 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, | |
88 | le32_to_cpu(fe->i_clusters))) { | |
89 | mlog(ML_ERROR, "block offset is outside the allocated size: " | |
90 | "%llu\n", (unsigned long long)iblock); | |
91 | goto bail; | |
92 | } | |
93 | ||
94 | /* We don't use the page cache to create symlink data, so if | |
95 | * need be, copy it over from the buffer cache. */ | |
96 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { | |
97 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + | |
98 | iblock; | |
99 | buffer_cache_bh = sb_getblk(osb->sb, blkno); | |
100 | if (!buffer_cache_bh) { | |
101 | mlog(ML_ERROR, "couldn't getblock for symlink!\n"); | |
102 | goto bail; | |
103 | } | |
104 | ||
105 | /* we haven't locked out transactions, so a commit | |
106 | * could've happened. Since we've got a reference on | |
107 | * the bh, even if it commits while we're doing the | |
108 | * copy, the data is still good. */ | |
109 | if (buffer_jbd(buffer_cache_bh) | |
110 | && ocfs2_inode_is_new(inode)) { | |
111 | kaddr = kmap_atomic(bh_result->b_page, KM_USER0); | |
112 | if (!kaddr) { | |
113 | mlog(ML_ERROR, "couldn't kmap!\n"); | |
114 | goto bail; | |
115 | } | |
116 | memcpy(kaddr + (bh_result->b_size * iblock), | |
117 | buffer_cache_bh->b_data, | |
118 | bh_result->b_size); | |
119 | kunmap_atomic(kaddr, KM_USER0); | |
120 | set_buffer_uptodate(bh_result); | |
121 | } | |
122 | brelse(buffer_cache_bh); | |
123 | } | |
124 | ||
125 | map_bh(bh_result, inode->i_sb, | |
126 | le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); | |
127 | ||
128 | err = 0; | |
129 | ||
130 | bail: | |
131 | if (bh) | |
132 | brelse(bh); | |
133 | ||
134 | mlog_exit(err); | |
135 | return err; | |
136 | } | |
137 | ||
138 | static int ocfs2_get_block(struct inode *inode, sector_t iblock, | |
139 | struct buffer_head *bh_result, int create) | |
140 | { | |
141 | int err = 0; | |
49cb8d2d | 142 | unsigned int ext_flags; |
628a24f5 MF |
143 | u64 max_blocks = bh_result->b_size >> inode->i_blkbits; |
144 | u64 p_blkno, count, past_eof; | |
25baf2da | 145 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd MF |
146 | |
147 | mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode, | |
148 | (unsigned long long)iblock, bh_result, create); | |
149 | ||
150 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) | |
151 | mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", | |
152 | inode, inode->i_ino); | |
153 | ||
154 | if (S_ISLNK(inode->i_mode)) { | |
155 | /* this always does I/O for some reason. */ | |
156 | err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); | |
157 | goto bail; | |
158 | } | |
159 | ||
628a24f5 | 160 | err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count, |
49cb8d2d | 161 | &ext_flags); |
ccd979bd MF |
162 | if (err) { |
163 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " | |
b0697053 MF |
164 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, |
165 | (unsigned long long)p_blkno); | |
ccd979bd MF |
166 | goto bail; |
167 | } | |
168 | ||
628a24f5 MF |
169 | if (max_blocks < count) |
170 | count = max_blocks; | |
171 | ||
25baf2da MF |
172 | /* |
173 | * ocfs2 never allocates in this function - the only time we | |
174 | * need to use BH_New is when we're extending i_size on a file | |
175 | * system which doesn't support holes, in which case BH_New | |
176 | * allows block_prepare_write() to zero. | |
177 | */ | |
178 | mlog_bug_on_msg(create && p_blkno == 0 && ocfs2_sparse_alloc(osb), | |
179 | "ino %lu, iblock %llu\n", inode->i_ino, | |
180 | (unsigned long long)iblock); | |
181 | ||
49cb8d2d MF |
182 | /* Treat the unwritten extent as a hole for zeroing purposes. */ |
183 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
25baf2da MF |
184 | map_bh(bh_result, inode->i_sb, p_blkno); |
185 | ||
628a24f5 MF |
186 | bh_result->b_size = count << inode->i_blkbits; |
187 | ||
25baf2da MF |
188 | if (!ocfs2_sparse_alloc(osb)) { |
189 | if (p_blkno == 0) { | |
190 | err = -EIO; | |
191 | mlog(ML_ERROR, | |
192 | "iblock = %llu p_blkno = %llu blkno=(%llu)\n", | |
193 | (unsigned long long)iblock, | |
194 | (unsigned long long)p_blkno, | |
195 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
196 | mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters); | |
197 | dump_stack(); | |
198 | } | |
ccd979bd | 199 | |
25baf2da MF |
200 | past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
201 | mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino, | |
202 | (unsigned long long)past_eof); | |
ccd979bd | 203 | |
25baf2da MF |
204 | if (create && (iblock >= past_eof)) |
205 | set_buffer_new(bh_result); | |
206 | } | |
ccd979bd MF |
207 | |
208 | bail: | |
209 | if (err < 0) | |
210 | err = -EIO; | |
211 | ||
212 | mlog_exit(err); | |
213 | return err; | |
214 | } | |
215 | ||
1afc32b9 MF |
216 | int ocfs2_read_inline_data(struct inode *inode, struct page *page, |
217 | struct buffer_head *di_bh) | |
6798d35a MF |
218 | { |
219 | void *kaddr; | |
220 | unsigned int size; | |
221 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
222 | ||
223 | if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) { | |
224 | ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag", | |
225 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
226 | return -EROFS; | |
227 | } | |
228 | ||
229 | size = i_size_read(inode); | |
230 | ||
231 | if (size > PAGE_CACHE_SIZE || | |
232 | size > ocfs2_max_inline_data(inode->i_sb)) { | |
233 | ocfs2_error(inode->i_sb, | |
234 | "Inode %llu has with inline data has bad size: %u", | |
235 | (unsigned long long)OCFS2_I(inode)->ip_blkno, 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 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
257 | ||
258 | BUG_ON(!PageLocked(page)); | |
259 | BUG_ON(!OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL); | |
260 | ||
261 | ret = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &di_bh, | |
262 | OCFS2_BH_CACHED, inode); | |
263 | if (ret) { | |
264 | mlog_errno(ret); | |
265 | goto out; | |
266 | } | |
267 | ||
268 | ret = ocfs2_read_inline_data(inode, page, di_bh); | |
269 | out: | |
270 | unlock_page(page); | |
271 | ||
272 | brelse(di_bh); | |
273 | return ret; | |
274 | } | |
275 | ||
ccd979bd MF |
276 | static int ocfs2_readpage(struct file *file, struct page *page) |
277 | { | |
278 | struct inode *inode = page->mapping->host; | |
6798d35a | 279 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd MF |
280 | loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT; |
281 | int ret, unlock = 1; | |
282 | ||
283 | mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0)); | |
284 | ||
e63aecb6 | 285 | ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page); |
ccd979bd MF |
286 | if (ret != 0) { |
287 | if (ret == AOP_TRUNCATED_PAGE) | |
288 | unlock = 0; | |
289 | mlog_errno(ret); | |
290 | goto out; | |
291 | } | |
292 | ||
6798d35a | 293 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { |
e9dfc0b2 | 294 | ret = AOP_TRUNCATED_PAGE; |
e63aecb6 | 295 | goto out_inode_unlock; |
e9dfc0b2 | 296 | } |
ccd979bd MF |
297 | |
298 | /* | |
299 | * i_size might have just been updated as we grabed the meta lock. We | |
300 | * might now be discovering a truncate that hit on another node. | |
301 | * block_read_full_page->get_block freaks out if it is asked to read | |
302 | * beyond the end of a file, so we check here. Callers | |
54cb8821 | 303 | * (generic_file_read, vm_ops->fault) are clever enough to check i_size |
ccd979bd MF |
304 | * and notice that the page they just read isn't needed. |
305 | * | |
306 | * XXX sys_readahead() seems to get that wrong? | |
307 | */ | |
308 | if (start >= i_size_read(inode)) { | |
5c3c6bb7 | 309 | zero_user_page(page, 0, PAGE_SIZE, KM_USER0); |
ccd979bd MF |
310 | SetPageUptodate(page); |
311 | ret = 0; | |
312 | goto out_alloc; | |
313 | } | |
314 | ||
6798d35a MF |
315 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
316 | ret = ocfs2_readpage_inline(inode, page); | |
317 | else | |
318 | ret = block_read_full_page(page, ocfs2_get_block); | |
ccd979bd MF |
319 | unlock = 0; |
320 | ||
ccd979bd MF |
321 | out_alloc: |
322 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 MF |
323 | out_inode_unlock: |
324 | ocfs2_inode_unlock(inode, 0); | |
ccd979bd MF |
325 | out: |
326 | if (unlock) | |
327 | unlock_page(page); | |
328 | mlog_exit(ret); | |
329 | return ret; | |
330 | } | |
331 | ||
628a24f5 MF |
332 | /* |
333 | * This is used only for read-ahead. Failures or difficult to handle | |
334 | * situations are safe to ignore. | |
335 | * | |
336 | * Right now, we don't bother with BH_Boundary - in-inode extent lists | |
337 | * are quite large (243 extents on 4k blocks), so most inodes don't | |
338 | * grow out to a tree. If need be, detecting boundary extents could | |
339 | * trivially be added in a future version of ocfs2_get_block(). | |
340 | */ | |
341 | static int ocfs2_readpages(struct file *filp, struct address_space *mapping, | |
342 | struct list_head *pages, unsigned nr_pages) | |
343 | { | |
344 | int ret, err = -EIO; | |
345 | struct inode *inode = mapping->host; | |
346 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
347 | loff_t start; | |
348 | struct page *last; | |
349 | ||
350 | /* | |
351 | * Use the nonblocking flag for the dlm code to avoid page | |
352 | * lock inversion, but don't bother with retrying. | |
353 | */ | |
354 | ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK); | |
355 | if (ret) | |
356 | return err; | |
357 | ||
358 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { | |
359 | ocfs2_inode_unlock(inode, 0); | |
360 | return err; | |
361 | } | |
362 | ||
363 | /* | |
364 | * Don't bother with inline-data. There isn't anything | |
365 | * to read-ahead in that case anyway... | |
366 | */ | |
367 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
368 | goto out_unlock; | |
369 | ||
370 | /* | |
371 | * Check whether a remote node truncated this file - we just | |
372 | * drop out in that case as it's not worth handling here. | |
373 | */ | |
374 | last = list_entry(pages->prev, struct page, lru); | |
375 | start = (loff_t)last->index << PAGE_CACHE_SHIFT; | |
376 | if (start >= i_size_read(inode)) | |
377 | goto out_unlock; | |
378 | ||
379 | err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block); | |
380 | ||
381 | out_unlock: | |
382 | up_read(&oi->ip_alloc_sem); | |
383 | ocfs2_inode_unlock(inode, 0); | |
384 | ||
385 | return err; | |
386 | } | |
387 | ||
ccd979bd MF |
388 | /* Note: Because we don't support holes, our allocation has |
389 | * already happened (allocation writes zeros to the file data) | |
390 | * so we don't have to worry about ordered writes in | |
391 | * ocfs2_writepage. | |
392 | * | |
393 | * ->writepage is called during the process of invalidating the page cache | |
394 | * during blocked lock processing. It can't block on any cluster locks | |
395 | * to during block mapping. It's relying on the fact that the block | |
396 | * mapping can't have disappeared under the dirty pages that it is | |
397 | * being asked to write back. | |
398 | */ | |
399 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) | |
400 | { | |
401 | int ret; | |
402 | ||
403 | mlog_entry("(0x%p)\n", page); | |
404 | ||
405 | ret = block_write_full_page(page, ocfs2_get_block, wbc); | |
406 | ||
407 | mlog_exit(ret); | |
408 | ||
409 | return ret; | |
410 | } | |
411 | ||
5069120b MF |
412 | /* |
413 | * This is called from ocfs2_write_zero_page() which has handled it's | |
414 | * own cluster locking and has ensured allocation exists for those | |
415 | * blocks to be written. | |
416 | */ | |
53013cba MF |
417 | int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page, |
418 | unsigned from, unsigned to) | |
419 | { | |
420 | int ret; | |
421 | ||
53013cba MF |
422 | ret = block_prepare_write(page, from, to, ocfs2_get_block); |
423 | ||
53013cba MF |
424 | return ret; |
425 | } | |
426 | ||
ccd979bd MF |
427 | /* Taken from ext3. We don't necessarily need the full blown |
428 | * functionality yet, but IMHO it's better to cut and paste the whole | |
429 | * thing so we can avoid introducing our own bugs (and easily pick up | |
430 | * their fixes when they happen) --Mark */ | |
60b11392 MF |
431 | int walk_page_buffers( handle_t *handle, |
432 | struct buffer_head *head, | |
433 | unsigned from, | |
434 | unsigned to, | |
435 | int *partial, | |
436 | int (*fn)( handle_t *handle, | |
437 | struct buffer_head *bh)) | |
ccd979bd MF |
438 | { |
439 | struct buffer_head *bh; | |
440 | unsigned block_start, block_end; | |
441 | unsigned blocksize = head->b_size; | |
442 | int err, ret = 0; | |
443 | struct buffer_head *next; | |
444 | ||
445 | for ( bh = head, block_start = 0; | |
446 | ret == 0 && (bh != head || !block_start); | |
447 | block_start = block_end, bh = next) | |
448 | { | |
449 | next = bh->b_this_page; | |
450 | block_end = block_start + blocksize; | |
451 | if (block_end <= from || block_start >= to) { | |
452 | if (partial && !buffer_uptodate(bh)) | |
453 | *partial = 1; | |
454 | continue; | |
455 | } | |
456 | err = (*fn)(handle, bh); | |
457 | if (!ret) | |
458 | ret = err; | |
459 | } | |
460 | return ret; | |
461 | } | |
462 | ||
1fabe148 | 463 | handle_t *ocfs2_start_walk_page_trans(struct inode *inode, |
ccd979bd MF |
464 | struct page *page, |
465 | unsigned from, | |
466 | unsigned to) | |
467 | { | |
468 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1fabe148 | 469 | handle_t *handle = NULL; |
ccd979bd MF |
470 | int ret = 0; |
471 | ||
65eff9cc | 472 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
473 | if (!handle) { |
474 | ret = -ENOMEM; | |
475 | mlog_errno(ret); | |
476 | goto out; | |
477 | } | |
478 | ||
479 | if (ocfs2_should_order_data(inode)) { | |
1fabe148 | 480 | ret = walk_page_buffers(handle, |
ccd979bd MF |
481 | page_buffers(page), |
482 | from, to, NULL, | |
483 | ocfs2_journal_dirty_data); | |
484 | if (ret < 0) | |
485 | mlog_errno(ret); | |
486 | } | |
487 | out: | |
488 | if (ret) { | |
489 | if (handle) | |
02dc1af4 | 490 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
491 | handle = ERR_PTR(ret); |
492 | } | |
493 | return handle; | |
494 | } | |
495 | ||
ccd979bd MF |
496 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) |
497 | { | |
498 | sector_t status; | |
499 | u64 p_blkno = 0; | |
500 | int err = 0; | |
501 | struct inode *inode = mapping->host; | |
502 | ||
503 | mlog_entry("(block = %llu)\n", (unsigned long long)block); | |
504 | ||
505 | /* We don't need to lock journal system files, since they aren't | |
506 | * accessed concurrently from multiple nodes. | |
507 | */ | |
508 | if (!INODE_JOURNAL(inode)) { | |
e63aecb6 | 509 | err = ocfs2_inode_lock(inode, NULL, 0); |
ccd979bd MF |
510 | if (err) { |
511 | if (err != -ENOENT) | |
512 | mlog_errno(err); | |
513 | goto bail; | |
514 | } | |
515 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | |
516 | } | |
517 | ||
6798d35a MF |
518 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) |
519 | err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, | |
520 | NULL); | |
ccd979bd MF |
521 | |
522 | if (!INODE_JOURNAL(inode)) { | |
523 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 | 524 | ocfs2_inode_unlock(inode, 0); |
ccd979bd MF |
525 | } |
526 | ||
527 | if (err) { | |
528 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", | |
529 | (unsigned long long)block); | |
530 | mlog_errno(err); | |
531 | goto bail; | |
532 | } | |
533 | ||
ccd979bd MF |
534 | bail: |
535 | status = err ? 0 : p_blkno; | |
536 | ||
537 | mlog_exit((int)status); | |
538 | ||
539 | return status; | |
540 | } | |
541 | ||
542 | /* | |
543 | * TODO: Make this into a generic get_blocks function. | |
544 | * | |
545 | * From do_direct_io in direct-io.c: | |
546 | * "So what we do is to permit the ->get_blocks function to populate | |
547 | * bh.b_size with the size of IO which is permitted at this offset and | |
548 | * this i_blkbits." | |
549 | * | |
550 | * This function is called directly from get_more_blocks in direct-io.c. | |
551 | * | |
552 | * called like this: dio->get_blocks(dio->inode, fs_startblk, | |
553 | * fs_count, map_bh, dio->rw == WRITE); | |
554 | */ | |
555 | static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock, | |
ccd979bd MF |
556 | struct buffer_head *bh_result, int create) |
557 | { | |
558 | int ret; | |
4f902c37 | 559 | u64 p_blkno, inode_blocks, contig_blocks; |
49cb8d2d | 560 | unsigned int ext_flags; |
184d7d20 | 561 | unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits; |
1d8fa7a2 | 562 | unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits; |
ccd979bd | 563 | |
ccd979bd MF |
564 | /* This function won't even be called if the request isn't all |
565 | * nicely aligned and of the right size, so there's no need | |
566 | * for us to check any of that. */ | |
567 | ||
25baf2da | 568 | inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
564f8a32 MF |
569 | |
570 | /* | |
571 | * Any write past EOF is not allowed because we'd be extending. | |
572 | */ | |
573 | if (create && (iblock + max_blocks) > inode_blocks) { | |
ccd979bd MF |
574 | ret = -EIO; |
575 | goto bail; | |
576 | } | |
ccd979bd MF |
577 | |
578 | /* This figures out the size of the next contiguous block, and | |
579 | * our logical offset */ | |
363041a5 | 580 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, |
49cb8d2d | 581 | &contig_blocks, &ext_flags); |
ccd979bd MF |
582 | if (ret) { |
583 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", | |
584 | (unsigned long long)iblock); | |
585 | ret = -EIO; | |
586 | goto bail; | |
587 | } | |
588 | ||
25baf2da MF |
589 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno) { |
590 | ocfs2_error(inode->i_sb, | |
591 | "Inode %llu has a hole at block %llu\n", | |
592 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
593 | (unsigned long long)iblock); | |
594 | ret = -EROFS; | |
595 | goto bail; | |
596 | } | |
597 | ||
598 | /* | |
599 | * get_more_blocks() expects us to describe a hole by clearing | |
600 | * the mapped bit on bh_result(). | |
49cb8d2d MF |
601 | * |
602 | * Consider an unwritten extent as a hole. | |
25baf2da | 603 | */ |
49cb8d2d | 604 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) |
25baf2da MF |
605 | map_bh(bh_result, inode->i_sb, p_blkno); |
606 | else { | |
607 | /* | |
608 | * ocfs2_prepare_inode_for_write() should have caught | |
609 | * the case where we'd be filling a hole and triggered | |
610 | * a buffered write instead. | |
611 | */ | |
612 | if (create) { | |
613 | ret = -EIO; | |
614 | mlog_errno(ret); | |
615 | goto bail; | |
616 | } | |
617 | ||
618 | clear_buffer_mapped(bh_result); | |
619 | } | |
ccd979bd MF |
620 | |
621 | /* make sure we don't map more than max_blocks blocks here as | |
622 | that's all the kernel will handle at this point. */ | |
623 | if (max_blocks < contig_blocks) | |
624 | contig_blocks = max_blocks; | |
625 | bh_result->b_size = contig_blocks << blocksize_bits; | |
626 | bail: | |
627 | return ret; | |
628 | } | |
629 | ||
630 | /* | |
631 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're | |
632 | * particularly interested in the aio/dio case. Like the core uses | |
633 | * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from | |
634 | * truncation on another. | |
635 | */ | |
636 | static void ocfs2_dio_end_io(struct kiocb *iocb, | |
637 | loff_t offset, | |
638 | ssize_t bytes, | |
639 | void *private) | |
640 | { | |
d28c9174 | 641 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; |
7cdfc3a1 | 642 | int level; |
ccd979bd MF |
643 | |
644 | /* this io's submitter should not have unlocked this before we could */ | |
645 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); | |
7cdfc3a1 | 646 | |
ccd979bd | 647 | ocfs2_iocb_clear_rw_locked(iocb); |
7cdfc3a1 MF |
648 | |
649 | level = ocfs2_iocb_rw_locked_level(iocb); | |
650 | if (!level) | |
651 | up_read(&inode->i_alloc_sem); | |
652 | ocfs2_rw_unlock(inode, level); | |
ccd979bd MF |
653 | } |
654 | ||
03f981cf JB |
655 | /* |
656 | * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen | |
657 | * from ext3. PageChecked() bits have been removed as OCFS2 does not | |
658 | * do journalled data. | |
659 | */ | |
660 | static void ocfs2_invalidatepage(struct page *page, unsigned long offset) | |
661 | { | |
662 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | |
663 | ||
664 | journal_invalidatepage(journal, page, offset); | |
665 | } | |
666 | ||
667 | static int ocfs2_releasepage(struct page *page, gfp_t wait) | |
668 | { | |
669 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | |
670 | ||
671 | if (!page_has_buffers(page)) | |
672 | return 0; | |
673 | return journal_try_to_free_buffers(journal, page, wait); | |
674 | } | |
675 | ||
ccd979bd MF |
676 | static ssize_t ocfs2_direct_IO(int rw, |
677 | struct kiocb *iocb, | |
678 | const struct iovec *iov, | |
679 | loff_t offset, | |
680 | unsigned long nr_segs) | |
681 | { | |
682 | struct file *file = iocb->ki_filp; | |
d28c9174 | 683 | struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host; |
ccd979bd MF |
684 | int ret; |
685 | ||
686 | mlog_entry_void(); | |
53013cba | 687 | |
6798d35a MF |
688 | /* |
689 | * Fallback to buffered I/O if we see an inode without | |
690 | * extents. | |
691 | */ | |
692 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
693 | return 0; | |
694 | ||
ccd979bd MF |
695 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, |
696 | inode->i_sb->s_bdev, iov, offset, | |
697 | nr_segs, | |
698 | ocfs2_direct_IO_get_blocks, | |
699 | ocfs2_dio_end_io); | |
c934a92d | 700 | |
ccd979bd MF |
701 | mlog_exit(ret); |
702 | return ret; | |
703 | } | |
704 | ||
9517bac6 MF |
705 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, |
706 | u32 cpos, | |
707 | unsigned int *start, | |
708 | unsigned int *end) | |
709 | { | |
710 | unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE; | |
711 | ||
712 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) { | |
713 | unsigned int cpp; | |
714 | ||
715 | cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits); | |
716 | ||
717 | cluster_start = cpos % cpp; | |
718 | cluster_start = cluster_start << osb->s_clustersize_bits; | |
719 | ||
720 | cluster_end = cluster_start + osb->s_clustersize; | |
721 | } | |
722 | ||
723 | BUG_ON(cluster_start > PAGE_SIZE); | |
724 | BUG_ON(cluster_end > PAGE_SIZE); | |
725 | ||
726 | if (start) | |
727 | *start = cluster_start; | |
728 | if (end) | |
729 | *end = cluster_end; | |
730 | } | |
731 | ||
732 | /* | |
733 | * 'from' and 'to' are the region in the page to avoid zeroing. | |
734 | * | |
735 | * If pagesize > clustersize, this function will avoid zeroing outside | |
736 | * of the cluster boundary. | |
737 | * | |
738 | * from == to == 0 is code for "zero the entire cluster region" | |
739 | */ | |
740 | static void ocfs2_clear_page_regions(struct page *page, | |
741 | struct ocfs2_super *osb, u32 cpos, | |
742 | unsigned from, unsigned to) | |
743 | { | |
744 | void *kaddr; | |
745 | unsigned int cluster_start, cluster_end; | |
746 | ||
747 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); | |
748 | ||
749 | kaddr = kmap_atomic(page, KM_USER0); | |
750 | ||
751 | if (from || to) { | |
752 | if (from > cluster_start) | |
753 | memset(kaddr + cluster_start, 0, from - cluster_start); | |
754 | if (to < cluster_end) | |
755 | memset(kaddr + to, 0, cluster_end - to); | |
756 | } else { | |
757 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); | |
758 | } | |
759 | ||
760 | kunmap_atomic(kaddr, KM_USER0); | |
761 | } | |
762 | ||
4e9563fd MF |
763 | /* |
764 | * Nonsparse file systems fully allocate before we get to the write | |
765 | * code. This prevents ocfs2_write() from tagging the write as an | |
766 | * allocating one, which means ocfs2_map_page_blocks() might try to | |
767 | * read-in the blocks at the tail of our file. Avoid reading them by | |
768 | * testing i_size against each block offset. | |
769 | */ | |
770 | static int ocfs2_should_read_blk(struct inode *inode, struct page *page, | |
771 | unsigned int block_start) | |
772 | { | |
773 | u64 offset = page_offset(page) + block_start; | |
774 | ||
775 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
776 | return 1; | |
777 | ||
778 | if (i_size_read(inode) > offset) | |
779 | return 1; | |
780 | ||
781 | return 0; | |
782 | } | |
783 | ||
9517bac6 MF |
784 | /* |
785 | * Some of this taken from block_prepare_write(). We already have our | |
786 | * mapping by now though, and the entire write will be allocating or | |
787 | * it won't, so not much need to use BH_New. | |
788 | * | |
789 | * This will also skip zeroing, which is handled externally. | |
790 | */ | |
60b11392 MF |
791 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, |
792 | struct inode *inode, unsigned int from, | |
793 | unsigned int to, int new) | |
9517bac6 MF |
794 | { |
795 | int ret = 0; | |
796 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; | |
797 | unsigned int block_end, block_start; | |
798 | unsigned int bsize = 1 << inode->i_blkbits; | |
799 | ||
800 | if (!page_has_buffers(page)) | |
801 | create_empty_buffers(page, bsize, 0); | |
802 | ||
803 | head = page_buffers(page); | |
804 | for (bh = head, block_start = 0; bh != head || !block_start; | |
805 | bh = bh->b_this_page, block_start += bsize) { | |
806 | block_end = block_start + bsize; | |
807 | ||
3a307ffc MF |
808 | clear_buffer_new(bh); |
809 | ||
9517bac6 MF |
810 | /* |
811 | * Ignore blocks outside of our i/o range - | |
812 | * they may belong to unallocated clusters. | |
813 | */ | |
60b11392 | 814 | if (block_start >= to || block_end <= from) { |
9517bac6 MF |
815 | if (PageUptodate(page)) |
816 | set_buffer_uptodate(bh); | |
817 | continue; | |
818 | } | |
819 | ||
820 | /* | |
821 | * For an allocating write with cluster size >= page | |
822 | * size, we always write the entire page. | |
823 | */ | |
3a307ffc MF |
824 | if (new) |
825 | set_buffer_new(bh); | |
9517bac6 MF |
826 | |
827 | if (!buffer_mapped(bh)) { | |
828 | map_bh(bh, inode->i_sb, *p_blkno); | |
829 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
830 | } | |
831 | ||
832 | if (PageUptodate(page)) { | |
833 | if (!buffer_uptodate(bh)) | |
834 | set_buffer_uptodate(bh); | |
835 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
bce99768 | 836 | !buffer_new(bh) && |
4e9563fd | 837 | ocfs2_should_read_blk(inode, page, block_start) && |
bce99768 | 838 | (block_start < from || block_end > to)) { |
9517bac6 MF |
839 | ll_rw_block(READ, 1, &bh); |
840 | *wait_bh++=bh; | |
841 | } | |
842 | ||
843 | *p_blkno = *p_blkno + 1; | |
844 | } | |
845 | ||
846 | /* | |
847 | * If we issued read requests - let them complete. | |
848 | */ | |
849 | while(wait_bh > wait) { | |
850 | wait_on_buffer(*--wait_bh); | |
851 | if (!buffer_uptodate(*wait_bh)) | |
852 | ret = -EIO; | |
853 | } | |
854 | ||
855 | if (ret == 0 || !new) | |
856 | return ret; | |
857 | ||
858 | /* | |
859 | * If we get -EIO above, zero out any newly allocated blocks | |
860 | * to avoid exposing stale data. | |
861 | */ | |
862 | bh = head; | |
863 | block_start = 0; | |
864 | do { | |
9517bac6 MF |
865 | block_end = block_start + bsize; |
866 | if (block_end <= from) | |
867 | goto next_bh; | |
868 | if (block_start >= to) | |
869 | break; | |
870 | ||
54c57dc3 | 871 | zero_user_page(page, block_start, bh->b_size, KM_USER0); |
9517bac6 MF |
872 | set_buffer_uptodate(bh); |
873 | mark_buffer_dirty(bh); | |
874 | ||
875 | next_bh: | |
876 | block_start = block_end; | |
877 | bh = bh->b_this_page; | |
878 | } while (bh != head); | |
879 | ||
880 | return ret; | |
881 | } | |
882 | ||
3a307ffc MF |
883 | #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE) |
884 | #define OCFS2_MAX_CTXT_PAGES 1 | |
885 | #else | |
886 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE) | |
887 | #endif | |
888 | ||
889 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE) | |
890 | ||
6af67d82 | 891 | /* |
3a307ffc | 892 | * Describe the state of a single cluster to be written to. |
6af67d82 | 893 | */ |
3a307ffc MF |
894 | struct ocfs2_write_cluster_desc { |
895 | u32 c_cpos; | |
896 | u32 c_phys; | |
897 | /* | |
898 | * Give this a unique field because c_phys eventually gets | |
899 | * filled. | |
900 | */ | |
901 | unsigned c_new; | |
b27b7cbc | 902 | unsigned c_unwritten; |
3a307ffc | 903 | }; |
6af67d82 | 904 | |
b27b7cbc MF |
905 | static inline int ocfs2_should_zero_cluster(struct ocfs2_write_cluster_desc *d) |
906 | { | |
907 | return d->c_new || d->c_unwritten; | |
908 | } | |
909 | ||
3a307ffc MF |
910 | struct ocfs2_write_ctxt { |
911 | /* Logical cluster position / len of write */ | |
912 | u32 w_cpos; | |
913 | u32 w_clen; | |
6af67d82 | 914 | |
3a307ffc | 915 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; |
6af67d82 | 916 | |
3a307ffc MF |
917 | /* |
918 | * This is true if page_size > cluster_size. | |
919 | * | |
920 | * It triggers a set of special cases during write which might | |
921 | * have to deal with allocating writes to partial pages. | |
922 | */ | |
923 | unsigned int w_large_pages; | |
6af67d82 | 924 | |
3a307ffc MF |
925 | /* |
926 | * Pages involved in this write. | |
927 | * | |
928 | * w_target_page is the page being written to by the user. | |
929 | * | |
930 | * w_pages is an array of pages which always contains | |
931 | * w_target_page, and in the case of an allocating write with | |
932 | * page_size < cluster size, it will contain zero'd and mapped | |
933 | * pages adjacent to w_target_page which need to be written | |
934 | * out in so that future reads from that region will get | |
935 | * zero's. | |
936 | */ | |
937 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; | |
938 | unsigned int w_num_pages; | |
939 | struct page *w_target_page; | |
eeb47d12 | 940 | |
3a307ffc MF |
941 | /* |
942 | * ocfs2_write_end() uses this to know what the real range to | |
943 | * write in the target should be. | |
944 | */ | |
945 | unsigned int w_target_from; | |
946 | unsigned int w_target_to; | |
947 | ||
948 | /* | |
949 | * We could use journal_current_handle() but this is cleaner, | |
950 | * IMHO -Mark | |
951 | */ | |
952 | handle_t *w_handle; | |
953 | ||
954 | struct buffer_head *w_di_bh; | |
b27b7cbc MF |
955 | |
956 | struct ocfs2_cached_dealloc_ctxt w_dealloc; | |
3a307ffc MF |
957 | }; |
958 | ||
1d410a6e | 959 | void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages) |
3a307ffc MF |
960 | { |
961 | int i; | |
962 | ||
1d410a6e MF |
963 | for(i = 0; i < num_pages; i++) { |
964 | if (pages[i]) { | |
965 | unlock_page(pages[i]); | |
966 | mark_page_accessed(pages[i]); | |
967 | page_cache_release(pages[i]); | |
968 | } | |
6af67d82 | 969 | } |
1d410a6e MF |
970 | } |
971 | ||
972 | static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc) | |
973 | { | |
974 | ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages); | |
6af67d82 | 975 | |
3a307ffc MF |
976 | brelse(wc->w_di_bh); |
977 | kfree(wc); | |
978 | } | |
979 | ||
980 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, | |
981 | struct ocfs2_super *osb, loff_t pos, | |
607d44aa | 982 | unsigned len, struct buffer_head *di_bh) |
3a307ffc | 983 | { |
30b8548f | 984 | u32 cend; |
3a307ffc MF |
985 | struct ocfs2_write_ctxt *wc; |
986 | ||
987 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); | |
988 | if (!wc) | |
989 | return -ENOMEM; | |
6af67d82 | 990 | |
3a307ffc | 991 | wc->w_cpos = pos >> osb->s_clustersize_bits; |
30b8548f | 992 | cend = (pos + len - 1) >> osb->s_clustersize_bits; |
993 | wc->w_clen = cend - wc->w_cpos + 1; | |
607d44aa MF |
994 | get_bh(di_bh); |
995 | wc->w_di_bh = di_bh; | |
6af67d82 | 996 | |
3a307ffc MF |
997 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) |
998 | wc->w_large_pages = 1; | |
999 | else | |
1000 | wc->w_large_pages = 0; | |
1001 | ||
b27b7cbc MF |
1002 | ocfs2_init_dealloc_ctxt(&wc->w_dealloc); |
1003 | ||
3a307ffc | 1004 | *wcp = wc; |
6af67d82 | 1005 | |
3a307ffc | 1006 | return 0; |
6af67d82 MF |
1007 | } |
1008 | ||
9517bac6 | 1009 | /* |
3a307ffc MF |
1010 | * If a page has any new buffers, zero them out here, and mark them uptodate |
1011 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1012 | * block data from leaking). And clear the new bit. | |
9517bac6 | 1013 | */ |
3a307ffc | 1014 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) |
9517bac6 | 1015 | { |
3a307ffc MF |
1016 | unsigned int block_start, block_end; |
1017 | struct buffer_head *head, *bh; | |
9517bac6 | 1018 | |
3a307ffc MF |
1019 | BUG_ON(!PageLocked(page)); |
1020 | if (!page_has_buffers(page)) | |
1021 | return; | |
9517bac6 | 1022 | |
3a307ffc MF |
1023 | bh = head = page_buffers(page); |
1024 | block_start = 0; | |
1025 | do { | |
1026 | block_end = block_start + bh->b_size; | |
1027 | ||
1028 | if (buffer_new(bh)) { | |
1029 | if (block_end > from && block_start < to) { | |
1030 | if (!PageUptodate(page)) { | |
1031 | unsigned start, end; | |
3a307ffc MF |
1032 | |
1033 | start = max(from, block_start); | |
1034 | end = min(to, block_end); | |
1035 | ||
54c57dc3 | 1036 | zero_user_page(page, start, end - start, KM_USER0); |
3a307ffc MF |
1037 | set_buffer_uptodate(bh); |
1038 | } | |
1039 | ||
1040 | clear_buffer_new(bh); | |
1041 | mark_buffer_dirty(bh); | |
1042 | } | |
1043 | } | |
9517bac6 | 1044 | |
3a307ffc MF |
1045 | block_start = block_end; |
1046 | bh = bh->b_this_page; | |
1047 | } while (bh != head); | |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | * Only called when we have a failure during allocating write to write | |
1052 | * zero's to the newly allocated region. | |
1053 | */ | |
1054 | static void ocfs2_write_failure(struct inode *inode, | |
1055 | struct ocfs2_write_ctxt *wc, | |
1056 | loff_t user_pos, unsigned user_len) | |
1057 | { | |
1058 | int i; | |
5c26a7b7 MF |
1059 | unsigned from = user_pos & (PAGE_CACHE_SIZE - 1), |
1060 | to = user_pos + user_len; | |
3a307ffc MF |
1061 | struct page *tmppage; |
1062 | ||
5c26a7b7 | 1063 | ocfs2_zero_new_buffers(wc->w_target_page, from, to); |
9517bac6 | 1064 | |
3a307ffc MF |
1065 | for(i = 0; i < wc->w_num_pages; i++) { |
1066 | tmppage = wc->w_pages[i]; | |
9517bac6 | 1067 | |
3a307ffc MF |
1068 | if (ocfs2_should_order_data(inode)) |
1069 | walk_page_buffers(wc->w_handle, page_buffers(tmppage), | |
1070 | from, to, NULL, | |
1071 | ocfs2_journal_dirty_data); | |
eeb47d12 | 1072 | |
3a307ffc | 1073 | block_commit_write(tmppage, from, to); |
9517bac6 | 1074 | } |
9517bac6 MF |
1075 | } |
1076 | ||
3a307ffc MF |
1077 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, |
1078 | struct ocfs2_write_ctxt *wc, | |
1079 | struct page *page, u32 cpos, | |
1080 | loff_t user_pos, unsigned user_len, | |
1081 | int new) | |
9517bac6 | 1082 | { |
3a307ffc MF |
1083 | int ret; |
1084 | unsigned int map_from = 0, map_to = 0; | |
9517bac6 | 1085 | unsigned int cluster_start, cluster_end; |
3a307ffc | 1086 | unsigned int user_data_from = 0, user_data_to = 0; |
9517bac6 | 1087 | |
3a307ffc | 1088 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, |
9517bac6 MF |
1089 | &cluster_start, &cluster_end); |
1090 | ||
3a307ffc MF |
1091 | if (page == wc->w_target_page) { |
1092 | map_from = user_pos & (PAGE_CACHE_SIZE - 1); | |
1093 | map_to = map_from + user_len; | |
1094 | ||
1095 | if (new) | |
1096 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1097 | cluster_start, cluster_end, | |
1098 | new); | |
1099 | else | |
1100 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1101 | map_from, map_to, new); | |
1102 | if (ret) { | |
9517bac6 MF |
1103 | mlog_errno(ret); |
1104 | goto out; | |
1105 | } | |
1106 | ||
3a307ffc MF |
1107 | user_data_from = map_from; |
1108 | user_data_to = map_to; | |
9517bac6 | 1109 | if (new) { |
3a307ffc MF |
1110 | map_from = cluster_start; |
1111 | map_to = cluster_end; | |
9517bac6 MF |
1112 | } |
1113 | } else { | |
1114 | /* | |
1115 | * If we haven't allocated the new page yet, we | |
1116 | * shouldn't be writing it out without copying user | |
1117 | * data. This is likely a math error from the caller. | |
1118 | */ | |
1119 | BUG_ON(!new); | |
1120 | ||
3a307ffc MF |
1121 | map_from = cluster_start; |
1122 | map_to = cluster_end; | |
9517bac6 MF |
1123 | |
1124 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
3a307ffc | 1125 | cluster_start, cluster_end, new); |
9517bac6 MF |
1126 | if (ret) { |
1127 | mlog_errno(ret); | |
1128 | goto out; | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * Parts of newly allocated pages need to be zero'd. | |
1134 | * | |
1135 | * Above, we have also rewritten 'to' and 'from' - as far as | |
1136 | * the rest of the function is concerned, the entire cluster | |
1137 | * range inside of a page needs to be written. | |
1138 | * | |
1139 | * We can skip this if the page is up to date - it's already | |
1140 | * been zero'd from being read in as a hole. | |
1141 | */ | |
1142 | if (new && !PageUptodate(page)) | |
1143 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), | |
3a307ffc | 1144 | cpos, user_data_from, user_data_to); |
9517bac6 MF |
1145 | |
1146 | flush_dcache_page(page); | |
1147 | ||
9517bac6 | 1148 | out: |
3a307ffc | 1149 | return ret; |
9517bac6 MF |
1150 | } |
1151 | ||
1152 | /* | |
3a307ffc | 1153 | * This function will only grab one clusters worth of pages. |
9517bac6 | 1154 | */ |
3a307ffc MF |
1155 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, |
1156 | struct ocfs2_write_ctxt *wc, | |
7307de80 MF |
1157 | u32 cpos, loff_t user_pos, int new, |
1158 | struct page *mmap_page) | |
9517bac6 | 1159 | { |
3a307ffc MF |
1160 | int ret = 0, i; |
1161 | unsigned long start, target_index, index; | |
9517bac6 | 1162 | struct inode *inode = mapping->host; |
9517bac6 | 1163 | |
3a307ffc | 1164 | target_index = user_pos >> PAGE_CACHE_SHIFT; |
9517bac6 MF |
1165 | |
1166 | /* | |
1167 | * Figure out how many pages we'll be manipulating here. For | |
60b11392 MF |
1168 | * non allocating write, we just change the one |
1169 | * page. Otherwise, we'll need a whole clusters worth. | |
9517bac6 | 1170 | */ |
9517bac6 | 1171 | if (new) { |
3a307ffc MF |
1172 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); |
1173 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); | |
9517bac6 | 1174 | } else { |
3a307ffc MF |
1175 | wc->w_num_pages = 1; |
1176 | start = target_index; | |
9517bac6 MF |
1177 | } |
1178 | ||
3a307ffc | 1179 | for(i = 0; i < wc->w_num_pages; i++) { |
9517bac6 MF |
1180 | index = start + i; |
1181 | ||
7307de80 MF |
1182 | if (index == target_index && mmap_page) { |
1183 | /* | |
1184 | * ocfs2_pagemkwrite() is a little different | |
1185 | * and wants us to directly use the page | |
1186 | * passed in. | |
1187 | */ | |
1188 | lock_page(mmap_page); | |
1189 | ||
1190 | if (mmap_page->mapping != mapping) { | |
1191 | unlock_page(mmap_page); | |
1192 | /* | |
1193 | * Sanity check - the locking in | |
1194 | * ocfs2_pagemkwrite() should ensure | |
1195 | * that this code doesn't trigger. | |
1196 | */ | |
1197 | ret = -EINVAL; | |
1198 | mlog_errno(ret); | |
1199 | goto out; | |
1200 | } | |
1201 | ||
1202 | page_cache_get(mmap_page); | |
1203 | wc->w_pages[i] = mmap_page; | |
1204 | } else { | |
1205 | wc->w_pages[i] = find_or_create_page(mapping, index, | |
1206 | GFP_NOFS); | |
1207 | if (!wc->w_pages[i]) { | |
1208 | ret = -ENOMEM; | |
1209 | mlog_errno(ret); | |
1210 | goto out; | |
1211 | } | |
9517bac6 | 1212 | } |
3a307ffc MF |
1213 | |
1214 | if (index == target_index) | |
1215 | wc->w_target_page = wc->w_pages[i]; | |
9517bac6 | 1216 | } |
3a307ffc MF |
1217 | out: |
1218 | return ret; | |
1219 | } | |
1220 | ||
1221 | /* | |
1222 | * Prepare a single cluster for write one cluster into the file. | |
1223 | */ | |
1224 | static int ocfs2_write_cluster(struct address_space *mapping, | |
b27b7cbc MF |
1225 | u32 phys, unsigned int unwritten, |
1226 | struct ocfs2_alloc_context *data_ac, | |
3a307ffc MF |
1227 | struct ocfs2_alloc_context *meta_ac, |
1228 | struct ocfs2_write_ctxt *wc, u32 cpos, | |
1229 | loff_t user_pos, unsigned user_len) | |
1230 | { | |
b27b7cbc | 1231 | int ret, i, new, should_zero = 0; |
3a307ffc MF |
1232 | u64 v_blkno, p_blkno; |
1233 | struct inode *inode = mapping->host; | |
1234 | ||
1235 | new = phys == 0 ? 1 : 0; | |
b27b7cbc MF |
1236 | if (new || unwritten) |
1237 | should_zero = 1; | |
9517bac6 MF |
1238 | |
1239 | if (new) { | |
3a307ffc MF |
1240 | u32 tmp_pos; |
1241 | ||
9517bac6 MF |
1242 | /* |
1243 | * This is safe to call with the page locks - it won't take | |
1244 | * any additional semaphores or cluster locks. | |
1245 | */ | |
3a307ffc | 1246 | tmp_pos = cpos; |
9517bac6 | 1247 | ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode, |
2ae99a60 | 1248 | &tmp_pos, 1, 0, wc->w_di_bh, |
3a307ffc MF |
1249 | wc->w_handle, data_ac, |
1250 | meta_ac, NULL); | |
9517bac6 MF |
1251 | /* |
1252 | * This shouldn't happen because we must have already | |
1253 | * calculated the correct meta data allocation required. The | |
1254 | * internal tree allocation code should know how to increase | |
1255 | * transaction credits itself. | |
1256 | * | |
1257 | * If need be, we could handle -EAGAIN for a | |
1258 | * RESTART_TRANS here. | |
1259 | */ | |
1260 | mlog_bug_on_msg(ret == -EAGAIN, | |
1261 | "Inode %llu: EAGAIN return during allocation.\n", | |
1262 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1263 | if (ret < 0) { | |
1264 | mlog_errno(ret); | |
1265 | goto out; | |
1266 | } | |
b27b7cbc MF |
1267 | } else if (unwritten) { |
1268 | ret = ocfs2_mark_extent_written(inode, wc->w_di_bh, | |
1269 | wc->w_handle, cpos, 1, phys, | |
1270 | meta_ac, &wc->w_dealloc); | |
1271 | if (ret < 0) { | |
1272 | mlog_errno(ret); | |
1273 | goto out; | |
1274 | } | |
1275 | } | |
3a307ffc | 1276 | |
b27b7cbc | 1277 | if (should_zero) |
3a307ffc | 1278 | v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos); |
b27b7cbc | 1279 | else |
3a307ffc | 1280 | v_blkno = user_pos >> inode->i_sb->s_blocksize_bits; |
9517bac6 | 1281 | |
3a307ffc MF |
1282 | /* |
1283 | * The only reason this should fail is due to an inability to | |
1284 | * find the extent added. | |
1285 | */ | |
49cb8d2d MF |
1286 | ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL, |
1287 | NULL); | |
9517bac6 | 1288 | if (ret < 0) { |
3a307ffc MF |
1289 | ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, " |
1290 | "at logical block %llu", | |
1291 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1292 | (unsigned long long)v_blkno); | |
9517bac6 MF |
1293 | goto out; |
1294 | } | |
1295 | ||
1296 | BUG_ON(p_blkno == 0); | |
1297 | ||
3a307ffc MF |
1298 | for(i = 0; i < wc->w_num_pages; i++) { |
1299 | int tmpret; | |
9517bac6 | 1300 | |
3a307ffc MF |
1301 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, |
1302 | wc->w_pages[i], cpos, | |
b27b7cbc MF |
1303 | user_pos, user_len, |
1304 | should_zero); | |
3a307ffc MF |
1305 | if (tmpret) { |
1306 | mlog_errno(tmpret); | |
1307 | if (ret == 0) | |
1308 | tmpret = ret; | |
1309 | } | |
9517bac6 MF |
1310 | } |
1311 | ||
3a307ffc MF |
1312 | /* |
1313 | * We only have cleanup to do in case of allocating write. | |
1314 | */ | |
1315 | if (ret && new) | |
1316 | ocfs2_write_failure(inode, wc, user_pos, user_len); | |
1317 | ||
9517bac6 | 1318 | out: |
9517bac6 | 1319 | |
3a307ffc | 1320 | return ret; |
9517bac6 MF |
1321 | } |
1322 | ||
0d172baa MF |
1323 | static int ocfs2_write_cluster_by_desc(struct address_space *mapping, |
1324 | struct ocfs2_alloc_context *data_ac, | |
1325 | struct ocfs2_alloc_context *meta_ac, | |
1326 | struct ocfs2_write_ctxt *wc, | |
1327 | loff_t pos, unsigned len) | |
1328 | { | |
1329 | int ret, i; | |
db56246c MF |
1330 | loff_t cluster_off; |
1331 | unsigned int local_len = len; | |
0d172baa | 1332 | struct ocfs2_write_cluster_desc *desc; |
db56246c | 1333 | struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb); |
0d172baa MF |
1334 | |
1335 | for (i = 0; i < wc->w_clen; i++) { | |
1336 | desc = &wc->w_desc[i]; | |
1337 | ||
db56246c MF |
1338 | /* |
1339 | * We have to make sure that the total write passed in | |
1340 | * doesn't extend past a single cluster. | |
1341 | */ | |
1342 | local_len = len; | |
1343 | cluster_off = pos & (osb->s_clustersize - 1); | |
1344 | if ((cluster_off + local_len) > osb->s_clustersize) | |
1345 | local_len = osb->s_clustersize - cluster_off; | |
1346 | ||
b27b7cbc MF |
1347 | ret = ocfs2_write_cluster(mapping, desc->c_phys, |
1348 | desc->c_unwritten, data_ac, meta_ac, | |
db56246c | 1349 | wc, desc->c_cpos, pos, local_len); |
0d172baa MF |
1350 | if (ret) { |
1351 | mlog_errno(ret); | |
1352 | goto out; | |
1353 | } | |
db56246c MF |
1354 | |
1355 | len -= local_len; | |
1356 | pos += local_len; | |
0d172baa MF |
1357 | } |
1358 | ||
1359 | ret = 0; | |
1360 | out: | |
1361 | return ret; | |
1362 | } | |
1363 | ||
3a307ffc MF |
1364 | /* |
1365 | * ocfs2_write_end() wants to know which parts of the target page it | |
1366 | * should complete the write on. It's easiest to compute them ahead of | |
1367 | * time when a more complete view of the write is available. | |
1368 | */ | |
1369 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, | |
1370 | struct ocfs2_write_ctxt *wc, | |
1371 | loff_t pos, unsigned len, int alloc) | |
9517bac6 | 1372 | { |
3a307ffc | 1373 | struct ocfs2_write_cluster_desc *desc; |
9517bac6 | 1374 | |
3a307ffc MF |
1375 | wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1); |
1376 | wc->w_target_to = wc->w_target_from + len; | |
1377 | ||
1378 | if (alloc == 0) | |
1379 | return; | |
1380 | ||
1381 | /* | |
1382 | * Allocating write - we may have different boundaries based | |
1383 | * on page size and cluster size. | |
1384 | * | |
1385 | * NOTE: We can no longer compute one value from the other as | |
1386 | * the actual write length and user provided length may be | |
1387 | * different. | |
1388 | */ | |
9517bac6 | 1389 | |
3a307ffc MF |
1390 | if (wc->w_large_pages) { |
1391 | /* | |
1392 | * We only care about the 1st and last cluster within | |
b27b7cbc | 1393 | * our range and whether they should be zero'd or not. Either |
3a307ffc MF |
1394 | * value may be extended out to the start/end of a |
1395 | * newly allocated cluster. | |
1396 | */ | |
1397 | desc = &wc->w_desc[0]; | |
b27b7cbc | 1398 | if (ocfs2_should_zero_cluster(desc)) |
3a307ffc MF |
1399 | ocfs2_figure_cluster_boundaries(osb, |
1400 | desc->c_cpos, | |
1401 | &wc->w_target_from, | |
1402 | NULL); | |
1403 | ||
1404 | desc = &wc->w_desc[wc->w_clen - 1]; | |
b27b7cbc | 1405 | if (ocfs2_should_zero_cluster(desc)) |
3a307ffc MF |
1406 | ocfs2_figure_cluster_boundaries(osb, |
1407 | desc->c_cpos, | |
1408 | NULL, | |
1409 | &wc->w_target_to); | |
1410 | } else { | |
1411 | wc->w_target_from = 0; | |
1412 | wc->w_target_to = PAGE_CACHE_SIZE; | |
1413 | } | |
9517bac6 MF |
1414 | } |
1415 | ||
0d172baa MF |
1416 | /* |
1417 | * Populate each single-cluster write descriptor in the write context | |
1418 | * with information about the i/o to be done. | |
b27b7cbc MF |
1419 | * |
1420 | * Returns the number of clusters that will have to be allocated, as | |
1421 | * well as a worst case estimate of the number of extent records that | |
1422 | * would have to be created during a write to an unwritten region. | |
0d172baa MF |
1423 | */ |
1424 | static int ocfs2_populate_write_desc(struct inode *inode, | |
1425 | struct ocfs2_write_ctxt *wc, | |
b27b7cbc MF |
1426 | unsigned int *clusters_to_alloc, |
1427 | unsigned int *extents_to_split) | |
9517bac6 | 1428 | { |
0d172baa | 1429 | int ret; |
3a307ffc | 1430 | struct ocfs2_write_cluster_desc *desc; |
0d172baa | 1431 | unsigned int num_clusters = 0; |
b27b7cbc | 1432 | unsigned int ext_flags = 0; |
0d172baa MF |
1433 | u32 phys = 0; |
1434 | int i; | |
9517bac6 | 1435 | |
b27b7cbc MF |
1436 | *clusters_to_alloc = 0; |
1437 | *extents_to_split = 0; | |
1438 | ||
3a307ffc MF |
1439 | for (i = 0; i < wc->w_clen; i++) { |
1440 | desc = &wc->w_desc[i]; | |
1441 | desc->c_cpos = wc->w_cpos + i; | |
1442 | ||
1443 | if (num_clusters == 0) { | |
b27b7cbc MF |
1444 | /* |
1445 | * Need to look up the next extent record. | |
1446 | */ | |
3a307ffc | 1447 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, |
b27b7cbc | 1448 | &num_clusters, &ext_flags); |
3a307ffc MF |
1449 | if (ret) { |
1450 | mlog_errno(ret); | |
607d44aa | 1451 | goto out; |
3a307ffc | 1452 | } |
b27b7cbc MF |
1453 | |
1454 | /* | |
1455 | * Assume worst case - that we're writing in | |
1456 | * the middle of the extent. | |
1457 | * | |
1458 | * We can assume that the write proceeds from | |
1459 | * left to right, in which case the extent | |
1460 | * insert code is smart enough to coalesce the | |
1461 | * next splits into the previous records created. | |
1462 | */ | |
1463 | if (ext_flags & OCFS2_EXT_UNWRITTEN) | |
1464 | *extents_to_split = *extents_to_split + 2; | |
3a307ffc MF |
1465 | } else if (phys) { |
1466 | /* | |
1467 | * Only increment phys if it doesn't describe | |
1468 | * a hole. | |
1469 | */ | |
1470 | phys++; | |
1471 | } | |
1472 | ||
1473 | desc->c_phys = phys; | |
1474 | if (phys == 0) { | |
1475 | desc->c_new = 1; | |
0d172baa | 1476 | *clusters_to_alloc = *clusters_to_alloc + 1; |
3a307ffc | 1477 | } |
b27b7cbc MF |
1478 | if (ext_flags & OCFS2_EXT_UNWRITTEN) |
1479 | desc->c_unwritten = 1; | |
3a307ffc MF |
1480 | |
1481 | num_clusters--; | |
9517bac6 MF |
1482 | } |
1483 | ||
0d172baa MF |
1484 | ret = 0; |
1485 | out: | |
1486 | return ret; | |
1487 | } | |
1488 | ||
1afc32b9 MF |
1489 | static int ocfs2_write_begin_inline(struct address_space *mapping, |
1490 | struct inode *inode, | |
1491 | struct ocfs2_write_ctxt *wc) | |
1492 | { | |
1493 | int ret; | |
1494 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1495 | struct page *page; | |
1496 | handle_t *handle; | |
1497 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1498 | ||
1499 | page = find_or_create_page(mapping, 0, GFP_NOFS); | |
1500 | if (!page) { | |
1501 | ret = -ENOMEM; | |
1502 | mlog_errno(ret); | |
1503 | goto out; | |
1504 | } | |
1505 | /* | |
1506 | * If we don't set w_num_pages then this page won't get unlocked | |
1507 | * and freed on cleanup of the write context. | |
1508 | */ | |
1509 | wc->w_pages[0] = wc->w_target_page = page; | |
1510 | wc->w_num_pages = 1; | |
1511 | ||
1512 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1513 | if (IS_ERR(handle)) { | |
1514 | ret = PTR_ERR(handle); | |
1515 | mlog_errno(ret); | |
1516 | goto out; | |
1517 | } | |
1518 | ||
1519 | ret = ocfs2_journal_access(handle, inode, wc->w_di_bh, | |
1520 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1521 | if (ret) { | |
1522 | ocfs2_commit_trans(osb, handle); | |
1523 | ||
1524 | mlog_errno(ret); | |
1525 | goto out; | |
1526 | } | |
1527 | ||
1528 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
1529 | ocfs2_set_inode_data_inline(inode, di); | |
1530 | ||
1531 | if (!PageUptodate(page)) { | |
1532 | ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh); | |
1533 | if (ret) { | |
1534 | ocfs2_commit_trans(osb, handle); | |
1535 | ||
1536 | goto out; | |
1537 | } | |
1538 | } | |
1539 | ||
1540 | wc->w_handle = handle; | |
1541 | out: | |
1542 | return ret; | |
1543 | } | |
1544 | ||
1545 | int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size) | |
1546 | { | |
1547 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1548 | ||
0d8a4e0c | 1549 | if (new_size <= le16_to_cpu(di->id2.i_data.id_count)) |
1afc32b9 MF |
1550 | return 1; |
1551 | return 0; | |
1552 | } | |
1553 | ||
1554 | static int ocfs2_try_to_write_inline_data(struct address_space *mapping, | |
1555 | struct inode *inode, loff_t pos, | |
1556 | unsigned len, struct page *mmap_page, | |
1557 | struct ocfs2_write_ctxt *wc) | |
1558 | { | |
1559 | int ret, written = 0; | |
1560 | loff_t end = pos + len; | |
1561 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1562 | ||
1563 | mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n", | |
1564 | (unsigned long long)oi->ip_blkno, len, (unsigned long long)pos, | |
1565 | oi->ip_dyn_features); | |
1566 | ||
1567 | /* | |
1568 | * Handle inodes which already have inline data 1st. | |
1569 | */ | |
1570 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1571 | if (mmap_page == NULL && | |
1572 | ocfs2_size_fits_inline_data(wc->w_di_bh, end)) | |
1573 | goto do_inline_write; | |
1574 | ||
1575 | /* | |
1576 | * The write won't fit - we have to give this inode an | |
1577 | * inline extent list now. | |
1578 | */ | |
1579 | ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh); | |
1580 | if (ret) | |
1581 | mlog_errno(ret); | |
1582 | goto out; | |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * Check whether the inode can accept inline data. | |
1587 | */ | |
1588 | if (oi->ip_clusters != 0 || i_size_read(inode) != 0) | |
1589 | return 0; | |
1590 | ||
1591 | /* | |
1592 | * Check whether the write can fit. | |
1593 | */ | |
1594 | if (mmap_page || end > ocfs2_max_inline_data(inode->i_sb)) | |
1595 | return 0; | |
1596 | ||
1597 | do_inline_write: | |
1598 | ret = ocfs2_write_begin_inline(mapping, inode, wc); | |
1599 | if (ret) { | |
1600 | mlog_errno(ret); | |
1601 | goto out; | |
1602 | } | |
1603 | ||
1604 | /* | |
1605 | * This signals to the caller that the data can be written | |
1606 | * inline. | |
1607 | */ | |
1608 | written = 1; | |
1609 | out: | |
1610 | return written ? written : ret; | |
1611 | } | |
1612 | ||
65ed39d6 MF |
1613 | /* |
1614 | * This function only does anything for file systems which can't | |
1615 | * handle sparse files. | |
1616 | * | |
1617 | * What we want to do here is fill in any hole between the current end | |
1618 | * of allocation and the end of our write. That way the rest of the | |
1619 | * write path can treat it as an non-allocating write, which has no | |
1620 | * special case code for sparse/nonsparse files. | |
1621 | */ | |
1622 | static int ocfs2_expand_nonsparse_inode(struct inode *inode, loff_t pos, | |
1623 | unsigned len, | |
1624 | struct ocfs2_write_ctxt *wc) | |
1625 | { | |
1626 | int ret; | |
1627 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1628 | loff_t newsize = pos + len; | |
1629 | ||
1630 | if (ocfs2_sparse_alloc(osb)) | |
1631 | return 0; | |
1632 | ||
1633 | if (newsize <= i_size_read(inode)) | |
1634 | return 0; | |
1635 | ||
1636 | ret = ocfs2_extend_no_holes(inode, newsize, newsize - len); | |
1637 | if (ret) | |
1638 | mlog_errno(ret); | |
1639 | ||
1640 | return ret; | |
1641 | } | |
1642 | ||
0d172baa MF |
1643 | int ocfs2_write_begin_nolock(struct address_space *mapping, |
1644 | loff_t pos, unsigned len, unsigned flags, | |
1645 | struct page **pagep, void **fsdata, | |
1646 | struct buffer_head *di_bh, struct page *mmap_page) | |
1647 | { | |
1648 | int ret, credits = OCFS2_INODE_UPDATE_CREDITS; | |
b27b7cbc | 1649 | unsigned int clusters_to_alloc, extents_to_split; |
0d172baa MF |
1650 | struct ocfs2_write_ctxt *wc; |
1651 | struct inode *inode = mapping->host; | |
1652 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1653 | struct ocfs2_dinode *di; | |
1654 | struct ocfs2_alloc_context *data_ac = NULL; | |
1655 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1656 | handle_t *handle; | |
1657 | ||
1658 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh); | |
1659 | if (ret) { | |
1660 | mlog_errno(ret); | |
1661 | return ret; | |
1662 | } | |
1663 | ||
1afc32b9 MF |
1664 | if (ocfs2_supports_inline_data(osb)) { |
1665 | ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len, | |
1666 | mmap_page, wc); | |
1667 | if (ret == 1) { | |
1668 | ret = 0; | |
1669 | goto success; | |
1670 | } | |
1671 | if (ret < 0) { | |
1672 | mlog_errno(ret); | |
1673 | goto out; | |
1674 | } | |
1675 | } | |
1676 | ||
65ed39d6 MF |
1677 | ret = ocfs2_expand_nonsparse_inode(inode, pos, len, wc); |
1678 | if (ret) { | |
1679 | mlog_errno(ret); | |
1680 | goto out; | |
1681 | } | |
1682 | ||
b27b7cbc MF |
1683 | ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc, |
1684 | &extents_to_split); | |
0d172baa MF |
1685 | if (ret) { |
1686 | mlog_errno(ret); | |
1687 | goto out; | |
1688 | } | |
1689 | ||
1690 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1691 | ||
3a307ffc MF |
1692 | /* |
1693 | * We set w_target_from, w_target_to here so that | |
1694 | * ocfs2_write_end() knows which range in the target page to | |
1695 | * write out. An allocation requires that we write the entire | |
1696 | * cluster range. | |
1697 | */ | |
b27b7cbc | 1698 | if (clusters_to_alloc || extents_to_split) { |
3a307ffc MF |
1699 | /* |
1700 | * XXX: We are stretching the limits of | |
b27b7cbc | 1701 | * ocfs2_lock_allocators(). It greatly over-estimates |
3a307ffc MF |
1702 | * the work to be done. |
1703 | */ | |
1704 | ret = ocfs2_lock_allocators(inode, di, clusters_to_alloc, | |
b27b7cbc | 1705 | extents_to_split, &data_ac, &meta_ac); |
9517bac6 MF |
1706 | if (ret) { |
1707 | mlog_errno(ret); | |
607d44aa | 1708 | goto out; |
9517bac6 MF |
1709 | } |
1710 | ||
3a307ffc MF |
1711 | credits = ocfs2_calc_extend_credits(inode->i_sb, di, |
1712 | clusters_to_alloc); | |
1713 | ||
9517bac6 MF |
1714 | } |
1715 | ||
b27b7cbc MF |
1716 | ocfs2_set_target_boundaries(osb, wc, pos, len, |
1717 | clusters_to_alloc + extents_to_split); | |
3a307ffc | 1718 | |
9517bac6 MF |
1719 | handle = ocfs2_start_trans(osb, credits); |
1720 | if (IS_ERR(handle)) { | |
1721 | ret = PTR_ERR(handle); | |
1722 | mlog_errno(ret); | |
607d44aa | 1723 | goto out; |
9517bac6 MF |
1724 | } |
1725 | ||
3a307ffc MF |
1726 | wc->w_handle = handle; |
1727 | ||
1728 | /* | |
1729 | * We don't want this to fail in ocfs2_write_end(), so do it | |
1730 | * here. | |
1731 | */ | |
1732 | ret = ocfs2_journal_access(handle, inode, wc->w_di_bh, | |
1733 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1734 | if (ret) { | |
9517bac6 MF |
1735 | mlog_errno(ret); |
1736 | goto out_commit; | |
1737 | } | |
1738 | ||
3a307ffc MF |
1739 | /* |
1740 | * Fill our page array first. That way we've grabbed enough so | |
1741 | * that we can zero and flush if we error after adding the | |
1742 | * extent. | |
1743 | */ | |
1744 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, | |
b27b7cbc MF |
1745 | clusters_to_alloc + extents_to_split, |
1746 | mmap_page); | |
9517bac6 MF |
1747 | if (ret) { |
1748 | mlog_errno(ret); | |
1749 | goto out_commit; | |
1750 | } | |
1751 | ||
0d172baa MF |
1752 | ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos, |
1753 | len); | |
1754 | if (ret) { | |
1755 | mlog_errno(ret); | |
1756 | goto out_commit; | |
9517bac6 | 1757 | } |
9517bac6 | 1758 | |
3a307ffc MF |
1759 | if (data_ac) |
1760 | ocfs2_free_alloc_context(data_ac); | |
1761 | if (meta_ac) | |
1762 | ocfs2_free_alloc_context(meta_ac); | |
9517bac6 | 1763 | |
1afc32b9 | 1764 | success: |
3a307ffc MF |
1765 | *pagep = wc->w_target_page; |
1766 | *fsdata = wc; | |
1767 | return 0; | |
9517bac6 MF |
1768 | out_commit: |
1769 | ocfs2_commit_trans(osb, handle); | |
1770 | ||
9517bac6 | 1771 | out: |
3a307ffc MF |
1772 | ocfs2_free_write_ctxt(wc); |
1773 | ||
9517bac6 MF |
1774 | if (data_ac) |
1775 | ocfs2_free_alloc_context(data_ac); | |
1776 | if (meta_ac) | |
1777 | ocfs2_free_alloc_context(meta_ac); | |
3a307ffc MF |
1778 | return ret; |
1779 | } | |
1780 | ||
b6af1bcd NP |
1781 | static int ocfs2_write_begin(struct file *file, struct address_space *mapping, |
1782 | loff_t pos, unsigned len, unsigned flags, | |
1783 | struct page **pagep, void **fsdata) | |
607d44aa MF |
1784 | { |
1785 | int ret; | |
1786 | struct buffer_head *di_bh = NULL; | |
1787 | struct inode *inode = mapping->host; | |
1788 | ||
e63aecb6 | 1789 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
607d44aa MF |
1790 | if (ret) { |
1791 | mlog_errno(ret); | |
1792 | return ret; | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * Take alloc sem here to prevent concurrent lookups. That way | |
1797 | * the mapping, zeroing and tree manipulation within | |
1798 | * ocfs2_write() will be safe against ->readpage(). This | |
1799 | * should also serve to lock out allocation from a shared | |
1800 | * writeable region. | |
1801 | */ | |
1802 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1803 | ||
607d44aa | 1804 | ret = ocfs2_write_begin_nolock(mapping, pos, len, flags, pagep, |
7307de80 | 1805 | fsdata, di_bh, NULL); |
607d44aa MF |
1806 | if (ret) { |
1807 | mlog_errno(ret); | |
c934a92d | 1808 | goto out_fail; |
607d44aa MF |
1809 | } |
1810 | ||
1811 | brelse(di_bh); | |
1812 | ||
1813 | return 0; | |
1814 | ||
607d44aa MF |
1815 | out_fail: |
1816 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1817 | ||
1818 | brelse(di_bh); | |
e63aecb6 | 1819 | ocfs2_inode_unlock(inode, 1); |
607d44aa MF |
1820 | |
1821 | return ret; | |
1822 | } | |
1823 | ||
1afc32b9 MF |
1824 | static void ocfs2_write_end_inline(struct inode *inode, loff_t pos, |
1825 | unsigned len, unsigned *copied, | |
1826 | struct ocfs2_dinode *di, | |
1827 | struct ocfs2_write_ctxt *wc) | |
1828 | { | |
1829 | void *kaddr; | |
1830 | ||
1831 | if (unlikely(*copied < len)) { | |
1832 | if (!PageUptodate(wc->w_target_page)) { | |
1833 | *copied = 0; | |
1834 | return; | |
1835 | } | |
1836 | } | |
1837 | ||
1838 | kaddr = kmap_atomic(wc->w_target_page, KM_USER0); | |
1839 | memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied); | |
1840 | kunmap_atomic(kaddr, KM_USER0); | |
1841 | ||
1842 | mlog(0, "Data written to inode at offset %llu. " | |
1843 | "id_count = %u, copied = %u, i_dyn_features = 0x%x\n", | |
1844 | (unsigned long long)pos, *copied, | |
1845 | le16_to_cpu(di->id2.i_data.id_count), | |
1846 | le16_to_cpu(di->i_dyn_features)); | |
1847 | } | |
1848 | ||
7307de80 MF |
1849 | int ocfs2_write_end_nolock(struct address_space *mapping, |
1850 | loff_t pos, unsigned len, unsigned copied, | |
1851 | struct page *page, void *fsdata) | |
3a307ffc MF |
1852 | { |
1853 | int i; | |
1854 | unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1); | |
1855 | struct inode *inode = mapping->host; | |
1856 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1857 | struct ocfs2_write_ctxt *wc = fsdata; | |
1858 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1859 | handle_t *handle = wc->w_handle; | |
1860 | struct page *tmppage; | |
1861 | ||
1afc32b9 MF |
1862 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
1863 | ocfs2_write_end_inline(inode, pos, len, &copied, di, wc); | |
1864 | goto out_write_size; | |
1865 | } | |
1866 | ||
3a307ffc MF |
1867 | if (unlikely(copied < len)) { |
1868 | if (!PageUptodate(wc->w_target_page)) | |
1869 | copied = 0; | |
1870 | ||
1871 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, | |
1872 | start+len); | |
1873 | } | |
1874 | flush_dcache_page(wc->w_target_page); | |
1875 | ||
1876 | for(i = 0; i < wc->w_num_pages; i++) { | |
1877 | tmppage = wc->w_pages[i]; | |
1878 | ||
1879 | if (tmppage == wc->w_target_page) { | |
1880 | from = wc->w_target_from; | |
1881 | to = wc->w_target_to; | |
1882 | ||
1883 | BUG_ON(from > PAGE_CACHE_SIZE || | |
1884 | to > PAGE_CACHE_SIZE || | |
1885 | to < from); | |
1886 | } else { | |
1887 | /* | |
1888 | * Pages adjacent to the target (if any) imply | |
1889 | * a hole-filling write in which case we want | |
1890 | * to flush their entire range. | |
1891 | */ | |
1892 | from = 0; | |
1893 | to = PAGE_CACHE_SIZE; | |
1894 | } | |
1895 | ||
1896 | if (ocfs2_should_order_data(inode)) | |
1897 | walk_page_buffers(wc->w_handle, page_buffers(tmppage), | |
1898 | from, to, NULL, | |
1899 | ocfs2_journal_dirty_data); | |
1900 | ||
1901 | block_commit_write(tmppage, from, to); | |
1902 | } | |
1903 | ||
1afc32b9 | 1904 | out_write_size: |
3a307ffc MF |
1905 | pos += copied; |
1906 | if (pos > inode->i_size) { | |
1907 | i_size_write(inode, pos); | |
1908 | mark_inode_dirty(inode); | |
1909 | } | |
1910 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
1911 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
1912 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
1913 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1914 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
3a307ffc MF |
1915 | ocfs2_journal_dirty(handle, wc->w_di_bh); |
1916 | ||
1917 | ocfs2_commit_trans(osb, handle); | |
59a5e416 | 1918 | |
b27b7cbc MF |
1919 | ocfs2_run_deallocs(osb, &wc->w_dealloc); |
1920 | ||
607d44aa MF |
1921 | ocfs2_free_write_ctxt(wc); |
1922 | ||
1923 | return copied; | |
1924 | } | |
1925 | ||
b6af1bcd NP |
1926 | static int ocfs2_write_end(struct file *file, struct address_space *mapping, |
1927 | loff_t pos, unsigned len, unsigned copied, | |
1928 | struct page *page, void *fsdata) | |
607d44aa MF |
1929 | { |
1930 | int ret; | |
1931 | struct inode *inode = mapping->host; | |
1932 | ||
1933 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata); | |
1934 | ||
3a307ffc | 1935 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
e63aecb6 | 1936 | ocfs2_inode_unlock(inode, 1); |
9517bac6 | 1937 | |
607d44aa | 1938 | return ret; |
9517bac6 MF |
1939 | } |
1940 | ||
f5e54d6e | 1941 | const struct address_space_operations ocfs2_aops = { |
ccd979bd | 1942 | .readpage = ocfs2_readpage, |
628a24f5 | 1943 | .readpages = ocfs2_readpages, |
ccd979bd | 1944 | .writepage = ocfs2_writepage, |
b6af1bcd NP |
1945 | .write_begin = ocfs2_write_begin, |
1946 | .write_end = ocfs2_write_end, | |
ccd979bd MF |
1947 | .bmap = ocfs2_bmap, |
1948 | .sync_page = block_sync_page, | |
03f981cf JB |
1949 | .direct_IO = ocfs2_direct_IO, |
1950 | .invalidatepage = ocfs2_invalidatepage, | |
1951 | .releasepage = ocfs2_releasepage, | |
1952 | .migratepage = buffer_migrate_page, | |
ccd979bd | 1953 | }; |