]>
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 MF |
31 | |
32 | #define MLOG_MASK_PREFIX ML_FILE_IO | |
33 | #include <cluster/masklog.h> | |
34 | ||
35 | #include "ocfs2.h" | |
36 | ||
37 | #include "alloc.h" | |
38 | #include "aops.h" | |
39 | #include "dlmglue.h" | |
40 | #include "extent_map.h" | |
41 | #include "file.h" | |
42 | #include "inode.h" | |
43 | #include "journal.h" | |
9517bac6 | 44 | #include "suballoc.h" |
ccd979bd MF |
45 | #include "super.h" |
46 | #include "symlink.h" | |
293b2f70 | 47 | #include "refcounttree.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 | ||
ef6b689b TM |
62 | mlog(0, "(0x%p, %llu, 0x%p, %d)\n", inode, |
63 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
64 | |
65 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); | |
66 | ||
67 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { | |
68 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", | |
69 | (unsigned long long)iblock); | |
70 | goto bail; | |
71 | } | |
72 | ||
b657c95c | 73 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
74 | if (status < 0) { |
75 | mlog_errno(status); | |
76 | goto bail; | |
77 | } | |
78 | fe = (struct ocfs2_dinode *) bh->b_data; | |
79 | ||
ccd979bd MF |
80 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, |
81 | le32_to_cpu(fe->i_clusters))) { | |
82 | mlog(ML_ERROR, "block offset is outside the allocated size: " | |
83 | "%llu\n", (unsigned long long)iblock); | |
84 | goto bail; | |
85 | } | |
86 | ||
87 | /* We don't use the page cache to create symlink data, so if | |
88 | * need be, copy it over from the buffer cache. */ | |
89 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { | |
90 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + | |
91 | iblock; | |
92 | buffer_cache_bh = sb_getblk(osb->sb, blkno); | |
93 | if (!buffer_cache_bh) { | |
94 | mlog(ML_ERROR, "couldn't getblock for symlink!\n"); | |
95 | goto bail; | |
96 | } | |
97 | ||
98 | /* we haven't locked out transactions, so a commit | |
99 | * could've happened. Since we've got a reference on | |
100 | * the bh, even if it commits while we're doing the | |
101 | * copy, the data is still good. */ | |
102 | if (buffer_jbd(buffer_cache_bh) | |
103 | && ocfs2_inode_is_new(inode)) { | |
104 | kaddr = kmap_atomic(bh_result->b_page, KM_USER0); | |
105 | if (!kaddr) { | |
106 | mlog(ML_ERROR, "couldn't kmap!\n"); | |
107 | goto bail; | |
108 | } | |
109 | memcpy(kaddr + (bh_result->b_size * iblock), | |
110 | buffer_cache_bh->b_data, | |
111 | bh_result->b_size); | |
112 | kunmap_atomic(kaddr, KM_USER0); | |
113 | set_buffer_uptodate(bh_result); | |
114 | } | |
115 | brelse(buffer_cache_bh); | |
116 | } | |
117 | ||
118 | map_bh(bh_result, inode->i_sb, | |
119 | le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); | |
120 | ||
121 | err = 0; | |
122 | ||
123 | bail: | |
a81cb88b | 124 | brelse(bh); |
ccd979bd | 125 | |
ccd979bd MF |
126 | return err; |
127 | } | |
128 | ||
6f70fa51 TM |
129 | int ocfs2_get_block(struct inode *inode, sector_t iblock, |
130 | struct buffer_head *bh_result, int create) | |
ccd979bd MF |
131 | { |
132 | int err = 0; | |
49cb8d2d | 133 | unsigned int ext_flags; |
628a24f5 MF |
134 | u64 max_blocks = bh_result->b_size >> inode->i_blkbits; |
135 | u64 p_blkno, count, past_eof; | |
25baf2da | 136 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd | 137 | |
ef6b689b TM |
138 | mlog(0, "(0x%p, %llu, 0x%p, %d)\n", inode, |
139 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
140 | |
141 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) | |
142 | mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", | |
143 | inode, inode->i_ino); | |
144 | ||
145 | if (S_ISLNK(inode->i_mode)) { | |
146 | /* this always does I/O for some reason. */ | |
147 | err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); | |
148 | goto bail; | |
149 | } | |
150 | ||
628a24f5 | 151 | err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count, |
49cb8d2d | 152 | &ext_flags); |
ccd979bd MF |
153 | if (err) { |
154 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " | |
b0697053 MF |
155 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, |
156 | (unsigned long long)p_blkno); | |
ccd979bd MF |
157 | goto bail; |
158 | } | |
159 | ||
628a24f5 MF |
160 | if (max_blocks < count) |
161 | count = max_blocks; | |
162 | ||
25baf2da MF |
163 | /* |
164 | * ocfs2 never allocates in this function - the only time we | |
165 | * need to use BH_New is when we're extending i_size on a file | |
166 | * system which doesn't support holes, in which case BH_New | |
ebdec241 | 167 | * allows __block_write_begin() to zero. |
c0420ad2 CL |
168 | * |
169 | * If we see this on a sparse file system, then a truncate has | |
170 | * raced us and removed the cluster. In this case, we clear | |
171 | * the buffers dirty and uptodate bits and let the buffer code | |
172 | * ignore it as a hole. | |
25baf2da | 173 | */ |
c0420ad2 CL |
174 | if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) { |
175 | clear_buffer_dirty(bh_result); | |
176 | clear_buffer_uptodate(bh_result); | |
177 | goto bail; | |
178 | } | |
25baf2da | 179 | |
49cb8d2d MF |
180 | /* Treat the unwritten extent as a hole for zeroing purposes. */ |
181 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
25baf2da MF |
182 | map_bh(bh_result, inode->i_sb, p_blkno); |
183 | ||
628a24f5 MF |
184 | bh_result->b_size = count << inode->i_blkbits; |
185 | ||
25baf2da MF |
186 | if (!ocfs2_sparse_alloc(osb)) { |
187 | if (p_blkno == 0) { | |
188 | err = -EIO; | |
189 | mlog(ML_ERROR, | |
190 | "iblock = %llu p_blkno = %llu blkno=(%llu)\n", | |
191 | (unsigned long long)iblock, | |
192 | (unsigned long long)p_blkno, | |
193 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
194 | mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters); | |
195 | dump_stack(); | |
1f4cea37 | 196 | goto bail; |
25baf2da | 197 | } |
25baf2da | 198 | } |
ccd979bd | 199 | |
5693486b JB |
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); | |
203 | if (create && (iblock >= past_eof)) | |
204 | set_buffer_new(bh_result); | |
205 | ||
ccd979bd MF |
206 | bail: |
207 | if (err < 0) | |
208 | err = -EIO; | |
209 | ||
ccd979bd MF |
210 | return err; |
211 | } | |
212 | ||
1afc32b9 MF |
213 | int ocfs2_read_inline_data(struct inode *inode, struct page *page, |
214 | struct buffer_head *di_bh) | |
6798d35a MF |
215 | { |
216 | void *kaddr; | |
d2849fb2 | 217 | loff_t size; |
6798d35a MF |
218 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
219 | ||
220 | if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) { | |
221 | ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag", | |
222 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
223 | return -EROFS; | |
224 | } | |
225 | ||
226 | size = i_size_read(inode); | |
227 | ||
228 | if (size > PAGE_CACHE_SIZE || | |
d9ae49d6 | 229 | size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) { |
6798d35a | 230 | ocfs2_error(inode->i_sb, |
d2849fb2 JK |
231 | "Inode %llu has with inline data has bad size: %Lu", |
232 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
233 | (unsigned long long)size); | |
6798d35a MF |
234 | return -EROFS; |
235 | } | |
236 | ||
237 | kaddr = kmap_atomic(page, KM_USER0); | |
238 | if (size) | |
239 | memcpy(kaddr, di->id2.i_data.id_data, size); | |
240 | /* Clear the remaining part of the page */ | |
241 | memset(kaddr + size, 0, PAGE_CACHE_SIZE - size); | |
242 | flush_dcache_page(page); | |
243 | kunmap_atomic(kaddr, KM_USER0); | |
244 | ||
245 | SetPageUptodate(page); | |
246 | ||
247 | return 0; | |
248 | } | |
249 | ||
250 | static int ocfs2_readpage_inline(struct inode *inode, struct page *page) | |
251 | { | |
252 | int ret; | |
253 | struct buffer_head *di_bh = NULL; | |
6798d35a MF |
254 | |
255 | BUG_ON(!PageLocked(page)); | |
86c838b0 | 256 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)); |
6798d35a | 257 | |
b657c95c | 258 | ret = ocfs2_read_inode_block(inode, &di_bh); |
6798d35a MF |
259 | if (ret) { |
260 | mlog_errno(ret); | |
261 | goto out; | |
262 | } | |
263 | ||
264 | ret = ocfs2_read_inline_data(inode, page, di_bh); | |
265 | out: | |
266 | unlock_page(page); | |
267 | ||
268 | brelse(di_bh); | |
269 | return ret; | |
270 | } | |
271 | ||
ccd979bd MF |
272 | static int ocfs2_readpage(struct file *file, struct page *page) |
273 | { | |
274 | struct inode *inode = page->mapping->host; | |
6798d35a | 275 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd MF |
276 | loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT; |
277 | int ret, unlock = 1; | |
278 | ||
ef6b689b | 279 | mlog(0, "(0x%p, %lu)\n", file, (page ? page->index : 0)); |
ccd979bd | 280 | |
e63aecb6 | 281 | ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page); |
ccd979bd MF |
282 | if (ret != 0) { |
283 | if (ret == AOP_TRUNCATED_PAGE) | |
284 | unlock = 0; | |
285 | mlog_errno(ret); | |
286 | goto out; | |
287 | } | |
288 | ||
6798d35a | 289 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { |
e9dfc0b2 | 290 | ret = AOP_TRUNCATED_PAGE; |
e63aecb6 | 291 | goto out_inode_unlock; |
e9dfc0b2 | 292 | } |
ccd979bd MF |
293 | |
294 | /* | |
295 | * i_size might have just been updated as we grabed the meta lock. We | |
296 | * might now be discovering a truncate that hit on another node. | |
297 | * block_read_full_page->get_block freaks out if it is asked to read | |
298 | * beyond the end of a file, so we check here. Callers | |
54cb8821 | 299 | * (generic_file_read, vm_ops->fault) are clever enough to check i_size |
ccd979bd MF |
300 | * and notice that the page they just read isn't needed. |
301 | * | |
302 | * XXX sys_readahead() seems to get that wrong? | |
303 | */ | |
304 | if (start >= i_size_read(inode)) { | |
eebd2aa3 | 305 | zero_user(page, 0, PAGE_SIZE); |
ccd979bd MF |
306 | SetPageUptodate(page); |
307 | ret = 0; | |
308 | goto out_alloc; | |
309 | } | |
310 | ||
6798d35a MF |
311 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
312 | ret = ocfs2_readpage_inline(inode, page); | |
313 | else | |
314 | ret = block_read_full_page(page, ocfs2_get_block); | |
ccd979bd MF |
315 | unlock = 0; |
316 | ||
ccd979bd MF |
317 | out_alloc: |
318 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 MF |
319 | out_inode_unlock: |
320 | ocfs2_inode_unlock(inode, 0); | |
ccd979bd MF |
321 | out: |
322 | if (unlock) | |
323 | unlock_page(page); | |
ccd979bd MF |
324 | return ret; |
325 | } | |
326 | ||
628a24f5 MF |
327 | /* |
328 | * This is used only for read-ahead. Failures or difficult to handle | |
329 | * situations are safe to ignore. | |
330 | * | |
331 | * Right now, we don't bother with BH_Boundary - in-inode extent lists | |
332 | * are quite large (243 extents on 4k blocks), so most inodes don't | |
333 | * grow out to a tree. If need be, detecting boundary extents could | |
334 | * trivially be added in a future version of ocfs2_get_block(). | |
335 | */ | |
336 | static int ocfs2_readpages(struct file *filp, struct address_space *mapping, | |
337 | struct list_head *pages, unsigned nr_pages) | |
338 | { | |
339 | int ret, err = -EIO; | |
340 | struct inode *inode = mapping->host; | |
341 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
342 | loff_t start; | |
343 | struct page *last; | |
344 | ||
345 | /* | |
346 | * Use the nonblocking flag for the dlm code to avoid page | |
347 | * lock inversion, but don't bother with retrying. | |
348 | */ | |
349 | ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK); | |
350 | if (ret) | |
351 | return err; | |
352 | ||
353 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { | |
354 | ocfs2_inode_unlock(inode, 0); | |
355 | return err; | |
356 | } | |
357 | ||
358 | /* | |
359 | * Don't bother with inline-data. There isn't anything | |
360 | * to read-ahead in that case anyway... | |
361 | */ | |
362 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
363 | goto out_unlock; | |
364 | ||
365 | /* | |
366 | * Check whether a remote node truncated this file - we just | |
367 | * drop out in that case as it's not worth handling here. | |
368 | */ | |
369 | last = list_entry(pages->prev, struct page, lru); | |
370 | start = (loff_t)last->index << PAGE_CACHE_SHIFT; | |
371 | if (start >= i_size_read(inode)) | |
372 | goto out_unlock; | |
373 | ||
374 | err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block); | |
375 | ||
376 | out_unlock: | |
377 | up_read(&oi->ip_alloc_sem); | |
378 | ocfs2_inode_unlock(inode, 0); | |
379 | ||
380 | return err; | |
381 | } | |
382 | ||
ccd979bd MF |
383 | /* Note: Because we don't support holes, our allocation has |
384 | * already happened (allocation writes zeros to the file data) | |
385 | * so we don't have to worry about ordered writes in | |
386 | * ocfs2_writepage. | |
387 | * | |
388 | * ->writepage is called during the process of invalidating the page cache | |
389 | * during blocked lock processing. It can't block on any cluster locks | |
390 | * to during block mapping. It's relying on the fact that the block | |
391 | * mapping can't have disappeared under the dirty pages that it is | |
392 | * being asked to write back. | |
393 | */ | |
394 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) | |
395 | { | |
396 | int ret; | |
397 | ||
ef6b689b | 398 | mlog(0, "(0x%p)\n", page); |
ccd979bd MF |
399 | |
400 | ret = block_write_full_page(page, ocfs2_get_block, wbc); | |
401 | ||
ccd979bd MF |
402 | return ret; |
403 | } | |
404 | ||
ccd979bd MF |
405 | /* Taken from ext3. We don't necessarily need the full blown |
406 | * functionality yet, but IMHO it's better to cut and paste the whole | |
407 | * thing so we can avoid introducing our own bugs (and easily pick up | |
408 | * their fixes when they happen) --Mark */ | |
60b11392 MF |
409 | int walk_page_buffers( handle_t *handle, |
410 | struct buffer_head *head, | |
411 | unsigned from, | |
412 | unsigned to, | |
413 | int *partial, | |
414 | int (*fn)( handle_t *handle, | |
415 | struct buffer_head *bh)) | |
ccd979bd MF |
416 | { |
417 | struct buffer_head *bh; | |
418 | unsigned block_start, block_end; | |
419 | unsigned blocksize = head->b_size; | |
420 | int err, ret = 0; | |
421 | struct buffer_head *next; | |
422 | ||
423 | for ( bh = head, block_start = 0; | |
424 | ret == 0 && (bh != head || !block_start); | |
425 | block_start = block_end, bh = next) | |
426 | { | |
427 | next = bh->b_this_page; | |
428 | block_end = block_start + blocksize; | |
429 | if (block_end <= from || block_start >= to) { | |
430 | if (partial && !buffer_uptodate(bh)) | |
431 | *partial = 1; | |
432 | continue; | |
433 | } | |
434 | err = (*fn)(handle, bh); | |
435 | if (!ret) | |
436 | ret = err; | |
437 | } | |
438 | return ret; | |
439 | } | |
440 | ||
ccd979bd MF |
441 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) |
442 | { | |
443 | sector_t status; | |
444 | u64 p_blkno = 0; | |
445 | int err = 0; | |
446 | struct inode *inode = mapping->host; | |
447 | ||
ef6b689b | 448 | mlog(0, "(block = %llu)\n", (unsigned long long)block); |
ccd979bd MF |
449 | |
450 | /* We don't need to lock journal system files, since they aren't | |
451 | * accessed concurrently from multiple nodes. | |
452 | */ | |
453 | if (!INODE_JOURNAL(inode)) { | |
e63aecb6 | 454 | err = ocfs2_inode_lock(inode, NULL, 0); |
ccd979bd MF |
455 | if (err) { |
456 | if (err != -ENOENT) | |
457 | mlog_errno(err); | |
458 | goto bail; | |
459 | } | |
460 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | |
461 | } | |
462 | ||
6798d35a MF |
463 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) |
464 | err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, | |
465 | NULL); | |
ccd979bd MF |
466 | |
467 | if (!INODE_JOURNAL(inode)) { | |
468 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 | 469 | ocfs2_inode_unlock(inode, 0); |
ccd979bd MF |
470 | } |
471 | ||
472 | if (err) { | |
473 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", | |
474 | (unsigned long long)block); | |
475 | mlog_errno(err); | |
476 | goto bail; | |
477 | } | |
478 | ||
ccd979bd MF |
479 | bail: |
480 | status = err ? 0 : p_blkno; | |
481 | ||
ccd979bd MF |
482 | return status; |
483 | } | |
484 | ||
485 | /* | |
486 | * TODO: Make this into a generic get_blocks function. | |
487 | * | |
488 | * From do_direct_io in direct-io.c: | |
489 | * "So what we do is to permit the ->get_blocks function to populate | |
490 | * bh.b_size with the size of IO which is permitted at this offset and | |
491 | * this i_blkbits." | |
492 | * | |
493 | * This function is called directly from get_more_blocks in direct-io.c. | |
494 | * | |
495 | * called like this: dio->get_blocks(dio->inode, fs_startblk, | |
496 | * fs_count, map_bh, dio->rw == WRITE); | |
5fe878ae CH |
497 | * |
498 | * Note that we never bother to allocate blocks here, and thus ignore the | |
499 | * create argument. | |
ccd979bd MF |
500 | */ |
501 | static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock, | |
ccd979bd MF |
502 | struct buffer_head *bh_result, int create) |
503 | { | |
504 | int ret; | |
4f902c37 | 505 | u64 p_blkno, inode_blocks, contig_blocks; |
49cb8d2d | 506 | unsigned int ext_flags; |
184d7d20 | 507 | unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits; |
1d8fa7a2 | 508 | unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits; |
ccd979bd | 509 | |
ccd979bd MF |
510 | /* This function won't even be called if the request isn't all |
511 | * nicely aligned and of the right size, so there's no need | |
512 | * for us to check any of that. */ | |
513 | ||
25baf2da | 514 | inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
564f8a32 | 515 | |
ccd979bd MF |
516 | /* This figures out the size of the next contiguous block, and |
517 | * our logical offset */ | |
363041a5 | 518 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, |
49cb8d2d | 519 | &contig_blocks, &ext_flags); |
ccd979bd MF |
520 | if (ret) { |
521 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", | |
522 | (unsigned long long)iblock); | |
523 | ret = -EIO; | |
524 | goto bail; | |
525 | } | |
526 | ||
cbaee472 TM |
527 | /* We should already CoW the refcounted extent in case of create. */ |
528 | BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED)); | |
529 | ||
25baf2da MF |
530 | /* |
531 | * get_more_blocks() expects us to describe a hole by clearing | |
532 | * the mapped bit on bh_result(). | |
49cb8d2d MF |
533 | * |
534 | * Consider an unwritten extent as a hole. | |
25baf2da | 535 | */ |
49cb8d2d | 536 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) |
25baf2da | 537 | map_bh(bh_result, inode->i_sb, p_blkno); |
5fe878ae | 538 | else |
25baf2da | 539 | clear_buffer_mapped(bh_result); |
ccd979bd MF |
540 | |
541 | /* make sure we don't map more than max_blocks blocks here as | |
542 | that's all the kernel will handle at this point. */ | |
543 | if (max_blocks < contig_blocks) | |
544 | contig_blocks = max_blocks; | |
545 | bh_result->b_size = contig_blocks << blocksize_bits; | |
546 | bail: | |
547 | return ret; | |
548 | } | |
549 | ||
2bd63216 | 550 | /* |
ccd979bd MF |
551 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're |
552 | * particularly interested in the aio/dio case. Like the core uses | |
553 | * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from | |
554 | * truncation on another. | |
555 | */ | |
556 | static void ocfs2_dio_end_io(struct kiocb *iocb, | |
557 | loff_t offset, | |
558 | ssize_t bytes, | |
40e2e973 CH |
559 | void *private, |
560 | int ret, | |
561 | bool is_async) | |
ccd979bd | 562 | { |
d28c9174 | 563 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; |
7cdfc3a1 | 564 | int level; |
ccd979bd MF |
565 | |
566 | /* this io's submitter should not have unlocked this before we could */ | |
567 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); | |
7cdfc3a1 | 568 | |
39c99f12 TY |
569 | if (ocfs2_iocb_is_sem_locked(iocb)) { |
570 | up_read(&inode->i_alloc_sem); | |
571 | ocfs2_iocb_clear_sem_locked(iocb); | |
572 | } | |
573 | ||
ccd979bd | 574 | ocfs2_iocb_clear_rw_locked(iocb); |
7cdfc3a1 MF |
575 | |
576 | level = ocfs2_iocb_rw_locked_level(iocb); | |
7cdfc3a1 | 577 | ocfs2_rw_unlock(inode, level); |
40e2e973 CH |
578 | |
579 | if (is_async) | |
580 | aio_complete(iocb, ret, 0); | |
ccd979bd MF |
581 | } |
582 | ||
03f981cf JB |
583 | /* |
584 | * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen | |
585 | * from ext3. PageChecked() bits have been removed as OCFS2 does not | |
586 | * do journalled data. | |
587 | */ | |
588 | static void ocfs2_invalidatepage(struct page *page, unsigned long offset) | |
589 | { | |
590 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | |
591 | ||
2b4e30fb | 592 | jbd2_journal_invalidatepage(journal, page, offset); |
03f981cf JB |
593 | } |
594 | ||
595 | static int ocfs2_releasepage(struct page *page, gfp_t wait) | |
596 | { | |
597 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | |
598 | ||
599 | if (!page_has_buffers(page)) | |
600 | return 0; | |
2b4e30fb | 601 | return jbd2_journal_try_to_free_buffers(journal, page, wait); |
03f981cf JB |
602 | } |
603 | ||
ccd979bd MF |
604 | static ssize_t ocfs2_direct_IO(int rw, |
605 | struct kiocb *iocb, | |
606 | const struct iovec *iov, | |
607 | loff_t offset, | |
608 | unsigned long nr_segs) | |
609 | { | |
610 | struct file *file = iocb->ki_filp; | |
d28c9174 | 611 | struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host; |
ccd979bd | 612 | |
6798d35a MF |
613 | /* |
614 | * Fallback to buffered I/O if we see an inode without | |
615 | * extents. | |
616 | */ | |
617 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
618 | return 0; | |
619 | ||
b80474b4 TM |
620 | /* Fallback to buffered I/O if we are appending. */ |
621 | if (i_size_read(inode) <= offset) | |
622 | return 0; | |
623 | ||
c1e8d35e TM |
624 | return __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, |
625 | iov, offset, nr_segs, | |
626 | ocfs2_direct_IO_get_blocks, | |
627 | ocfs2_dio_end_io, NULL, 0); | |
ccd979bd MF |
628 | } |
629 | ||
9517bac6 MF |
630 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, |
631 | u32 cpos, | |
632 | unsigned int *start, | |
633 | unsigned int *end) | |
634 | { | |
635 | unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE; | |
636 | ||
637 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) { | |
638 | unsigned int cpp; | |
639 | ||
640 | cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits); | |
641 | ||
642 | cluster_start = cpos % cpp; | |
643 | cluster_start = cluster_start << osb->s_clustersize_bits; | |
644 | ||
645 | cluster_end = cluster_start + osb->s_clustersize; | |
646 | } | |
647 | ||
648 | BUG_ON(cluster_start > PAGE_SIZE); | |
649 | BUG_ON(cluster_end > PAGE_SIZE); | |
650 | ||
651 | if (start) | |
652 | *start = cluster_start; | |
653 | if (end) | |
654 | *end = cluster_end; | |
655 | } | |
656 | ||
657 | /* | |
658 | * 'from' and 'to' are the region in the page to avoid zeroing. | |
659 | * | |
660 | * If pagesize > clustersize, this function will avoid zeroing outside | |
661 | * of the cluster boundary. | |
662 | * | |
663 | * from == to == 0 is code for "zero the entire cluster region" | |
664 | */ | |
665 | static void ocfs2_clear_page_regions(struct page *page, | |
666 | struct ocfs2_super *osb, u32 cpos, | |
667 | unsigned from, unsigned to) | |
668 | { | |
669 | void *kaddr; | |
670 | unsigned int cluster_start, cluster_end; | |
671 | ||
672 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); | |
673 | ||
674 | kaddr = kmap_atomic(page, KM_USER0); | |
675 | ||
676 | if (from || to) { | |
677 | if (from > cluster_start) | |
678 | memset(kaddr + cluster_start, 0, from - cluster_start); | |
679 | if (to < cluster_end) | |
680 | memset(kaddr + to, 0, cluster_end - to); | |
681 | } else { | |
682 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); | |
683 | } | |
684 | ||
685 | kunmap_atomic(kaddr, KM_USER0); | |
686 | } | |
687 | ||
4e9563fd MF |
688 | /* |
689 | * Nonsparse file systems fully allocate before we get to the write | |
690 | * code. This prevents ocfs2_write() from tagging the write as an | |
691 | * allocating one, which means ocfs2_map_page_blocks() might try to | |
692 | * read-in the blocks at the tail of our file. Avoid reading them by | |
693 | * testing i_size against each block offset. | |
694 | */ | |
695 | static int ocfs2_should_read_blk(struct inode *inode, struct page *page, | |
696 | unsigned int block_start) | |
697 | { | |
698 | u64 offset = page_offset(page) + block_start; | |
699 | ||
700 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
701 | return 1; | |
702 | ||
703 | if (i_size_read(inode) > offset) | |
704 | return 1; | |
705 | ||
706 | return 0; | |
707 | } | |
708 | ||
9517bac6 | 709 | /* |
ebdec241 | 710 | * Some of this taken from __block_write_begin(). We already have our |
9517bac6 MF |
711 | * mapping by now though, and the entire write will be allocating or |
712 | * it won't, so not much need to use BH_New. | |
713 | * | |
714 | * This will also skip zeroing, which is handled externally. | |
715 | */ | |
60b11392 MF |
716 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, |
717 | struct inode *inode, unsigned int from, | |
718 | unsigned int to, int new) | |
9517bac6 MF |
719 | { |
720 | int ret = 0; | |
721 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; | |
722 | unsigned int block_end, block_start; | |
723 | unsigned int bsize = 1 << inode->i_blkbits; | |
724 | ||
725 | if (!page_has_buffers(page)) | |
726 | create_empty_buffers(page, bsize, 0); | |
727 | ||
728 | head = page_buffers(page); | |
729 | for (bh = head, block_start = 0; bh != head || !block_start; | |
730 | bh = bh->b_this_page, block_start += bsize) { | |
731 | block_end = block_start + bsize; | |
732 | ||
3a307ffc MF |
733 | clear_buffer_new(bh); |
734 | ||
9517bac6 MF |
735 | /* |
736 | * Ignore blocks outside of our i/o range - | |
737 | * they may belong to unallocated clusters. | |
738 | */ | |
60b11392 | 739 | if (block_start >= to || block_end <= from) { |
9517bac6 MF |
740 | if (PageUptodate(page)) |
741 | set_buffer_uptodate(bh); | |
742 | continue; | |
743 | } | |
744 | ||
745 | /* | |
746 | * For an allocating write with cluster size >= page | |
747 | * size, we always write the entire page. | |
748 | */ | |
3a307ffc MF |
749 | if (new) |
750 | set_buffer_new(bh); | |
9517bac6 MF |
751 | |
752 | if (!buffer_mapped(bh)) { | |
753 | map_bh(bh, inode->i_sb, *p_blkno); | |
754 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
755 | } | |
756 | ||
757 | if (PageUptodate(page)) { | |
758 | if (!buffer_uptodate(bh)) | |
759 | set_buffer_uptodate(bh); | |
760 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
bce99768 | 761 | !buffer_new(bh) && |
4e9563fd | 762 | ocfs2_should_read_blk(inode, page, block_start) && |
bce99768 | 763 | (block_start < from || block_end > to)) { |
9517bac6 MF |
764 | ll_rw_block(READ, 1, &bh); |
765 | *wait_bh++=bh; | |
766 | } | |
767 | ||
768 | *p_blkno = *p_blkno + 1; | |
769 | } | |
770 | ||
771 | /* | |
772 | * If we issued read requests - let them complete. | |
773 | */ | |
774 | while(wait_bh > wait) { | |
775 | wait_on_buffer(*--wait_bh); | |
776 | if (!buffer_uptodate(*wait_bh)) | |
777 | ret = -EIO; | |
778 | } | |
779 | ||
780 | if (ret == 0 || !new) | |
781 | return ret; | |
782 | ||
783 | /* | |
784 | * If we get -EIO above, zero out any newly allocated blocks | |
785 | * to avoid exposing stale data. | |
786 | */ | |
787 | bh = head; | |
788 | block_start = 0; | |
789 | do { | |
9517bac6 MF |
790 | block_end = block_start + bsize; |
791 | if (block_end <= from) | |
792 | goto next_bh; | |
793 | if (block_start >= to) | |
794 | break; | |
795 | ||
eebd2aa3 | 796 | zero_user(page, block_start, bh->b_size); |
9517bac6 MF |
797 | set_buffer_uptodate(bh); |
798 | mark_buffer_dirty(bh); | |
799 | ||
800 | next_bh: | |
801 | block_start = block_end; | |
802 | bh = bh->b_this_page; | |
803 | } while (bh != head); | |
804 | ||
805 | return ret; | |
806 | } | |
807 | ||
3a307ffc MF |
808 | #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE) |
809 | #define OCFS2_MAX_CTXT_PAGES 1 | |
810 | #else | |
811 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE) | |
812 | #endif | |
813 | ||
814 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE) | |
815 | ||
6af67d82 | 816 | /* |
3a307ffc | 817 | * Describe the state of a single cluster to be written to. |
6af67d82 | 818 | */ |
3a307ffc MF |
819 | struct ocfs2_write_cluster_desc { |
820 | u32 c_cpos; | |
821 | u32 c_phys; | |
822 | /* | |
823 | * Give this a unique field because c_phys eventually gets | |
824 | * filled. | |
825 | */ | |
826 | unsigned c_new; | |
b27b7cbc | 827 | unsigned c_unwritten; |
e7432675 | 828 | unsigned c_needs_zero; |
3a307ffc | 829 | }; |
6af67d82 | 830 | |
3a307ffc MF |
831 | struct ocfs2_write_ctxt { |
832 | /* Logical cluster position / len of write */ | |
833 | u32 w_cpos; | |
834 | u32 w_clen; | |
6af67d82 | 835 | |
e7432675 SM |
836 | /* First cluster allocated in a nonsparse extend */ |
837 | u32 w_first_new_cpos; | |
838 | ||
3a307ffc | 839 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; |
6af67d82 | 840 | |
3a307ffc MF |
841 | /* |
842 | * This is true if page_size > cluster_size. | |
843 | * | |
844 | * It triggers a set of special cases during write which might | |
845 | * have to deal with allocating writes to partial pages. | |
846 | */ | |
847 | unsigned int w_large_pages; | |
6af67d82 | 848 | |
3a307ffc MF |
849 | /* |
850 | * Pages involved in this write. | |
851 | * | |
852 | * w_target_page is the page being written to by the user. | |
853 | * | |
854 | * w_pages is an array of pages which always contains | |
855 | * w_target_page, and in the case of an allocating write with | |
856 | * page_size < cluster size, it will contain zero'd and mapped | |
857 | * pages adjacent to w_target_page which need to be written | |
858 | * out in so that future reads from that region will get | |
859 | * zero's. | |
860 | */ | |
3a307ffc | 861 | unsigned int w_num_pages; |
83fd9c7f | 862 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; |
3a307ffc | 863 | struct page *w_target_page; |
eeb47d12 | 864 | |
3a307ffc MF |
865 | /* |
866 | * ocfs2_write_end() uses this to know what the real range to | |
867 | * write in the target should be. | |
868 | */ | |
869 | unsigned int w_target_from; | |
870 | unsigned int w_target_to; | |
871 | ||
872 | /* | |
873 | * We could use journal_current_handle() but this is cleaner, | |
874 | * IMHO -Mark | |
875 | */ | |
876 | handle_t *w_handle; | |
877 | ||
878 | struct buffer_head *w_di_bh; | |
b27b7cbc MF |
879 | |
880 | struct ocfs2_cached_dealloc_ctxt w_dealloc; | |
3a307ffc MF |
881 | }; |
882 | ||
1d410a6e | 883 | void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages) |
3a307ffc MF |
884 | { |
885 | int i; | |
886 | ||
1d410a6e MF |
887 | for(i = 0; i < num_pages; i++) { |
888 | if (pages[i]) { | |
889 | unlock_page(pages[i]); | |
890 | mark_page_accessed(pages[i]); | |
891 | page_cache_release(pages[i]); | |
892 | } | |
6af67d82 | 893 | } |
1d410a6e MF |
894 | } |
895 | ||
896 | static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc) | |
897 | { | |
898 | ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages); | |
6af67d82 | 899 | |
3a307ffc MF |
900 | brelse(wc->w_di_bh); |
901 | kfree(wc); | |
902 | } | |
903 | ||
904 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, | |
905 | struct ocfs2_super *osb, loff_t pos, | |
607d44aa | 906 | unsigned len, struct buffer_head *di_bh) |
3a307ffc | 907 | { |
30b8548f | 908 | u32 cend; |
3a307ffc MF |
909 | struct ocfs2_write_ctxt *wc; |
910 | ||
911 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); | |
912 | if (!wc) | |
913 | return -ENOMEM; | |
6af67d82 | 914 | |
3a307ffc | 915 | wc->w_cpos = pos >> osb->s_clustersize_bits; |
e7432675 | 916 | wc->w_first_new_cpos = UINT_MAX; |
30b8548f | 917 | cend = (pos + len - 1) >> osb->s_clustersize_bits; |
918 | wc->w_clen = cend - wc->w_cpos + 1; | |
607d44aa MF |
919 | get_bh(di_bh); |
920 | wc->w_di_bh = di_bh; | |
6af67d82 | 921 | |
3a307ffc MF |
922 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) |
923 | wc->w_large_pages = 1; | |
924 | else | |
925 | wc->w_large_pages = 0; | |
926 | ||
b27b7cbc MF |
927 | ocfs2_init_dealloc_ctxt(&wc->w_dealloc); |
928 | ||
3a307ffc | 929 | *wcp = wc; |
6af67d82 | 930 | |
3a307ffc | 931 | return 0; |
6af67d82 MF |
932 | } |
933 | ||
9517bac6 | 934 | /* |
3a307ffc MF |
935 | * If a page has any new buffers, zero them out here, and mark them uptodate |
936 | * and dirty so they'll be written out (in order to prevent uninitialised | |
937 | * block data from leaking). And clear the new bit. | |
9517bac6 | 938 | */ |
3a307ffc | 939 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) |
9517bac6 | 940 | { |
3a307ffc MF |
941 | unsigned int block_start, block_end; |
942 | struct buffer_head *head, *bh; | |
9517bac6 | 943 | |
3a307ffc MF |
944 | BUG_ON(!PageLocked(page)); |
945 | if (!page_has_buffers(page)) | |
946 | return; | |
9517bac6 | 947 | |
3a307ffc MF |
948 | bh = head = page_buffers(page); |
949 | block_start = 0; | |
950 | do { | |
951 | block_end = block_start + bh->b_size; | |
952 | ||
953 | if (buffer_new(bh)) { | |
954 | if (block_end > from && block_start < to) { | |
955 | if (!PageUptodate(page)) { | |
956 | unsigned start, end; | |
3a307ffc MF |
957 | |
958 | start = max(from, block_start); | |
959 | end = min(to, block_end); | |
960 | ||
eebd2aa3 | 961 | zero_user_segment(page, start, end); |
3a307ffc MF |
962 | set_buffer_uptodate(bh); |
963 | } | |
964 | ||
965 | clear_buffer_new(bh); | |
966 | mark_buffer_dirty(bh); | |
967 | } | |
968 | } | |
9517bac6 | 969 | |
3a307ffc MF |
970 | block_start = block_end; |
971 | bh = bh->b_this_page; | |
972 | } while (bh != head); | |
973 | } | |
974 | ||
975 | /* | |
976 | * Only called when we have a failure during allocating write to write | |
977 | * zero's to the newly allocated region. | |
978 | */ | |
979 | static void ocfs2_write_failure(struct inode *inode, | |
980 | struct ocfs2_write_ctxt *wc, | |
981 | loff_t user_pos, unsigned user_len) | |
982 | { | |
983 | int i; | |
5c26a7b7 MF |
984 | unsigned from = user_pos & (PAGE_CACHE_SIZE - 1), |
985 | to = user_pos + user_len; | |
3a307ffc MF |
986 | struct page *tmppage; |
987 | ||
5c26a7b7 | 988 | ocfs2_zero_new_buffers(wc->w_target_page, from, to); |
9517bac6 | 989 | |
3a307ffc MF |
990 | for(i = 0; i < wc->w_num_pages; i++) { |
991 | tmppage = wc->w_pages[i]; | |
9517bac6 | 992 | |
961cecbe | 993 | if (page_has_buffers(tmppage)) { |
53ef99ca | 994 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 995 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
996 | |
997 | block_commit_write(tmppage, from, to); | |
998 | } | |
9517bac6 | 999 | } |
9517bac6 MF |
1000 | } |
1001 | ||
3a307ffc MF |
1002 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, |
1003 | struct ocfs2_write_ctxt *wc, | |
1004 | struct page *page, u32 cpos, | |
1005 | loff_t user_pos, unsigned user_len, | |
1006 | int new) | |
9517bac6 | 1007 | { |
3a307ffc MF |
1008 | int ret; |
1009 | unsigned int map_from = 0, map_to = 0; | |
9517bac6 | 1010 | unsigned int cluster_start, cluster_end; |
3a307ffc | 1011 | unsigned int user_data_from = 0, user_data_to = 0; |
9517bac6 | 1012 | |
3a307ffc | 1013 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, |
9517bac6 MF |
1014 | &cluster_start, &cluster_end); |
1015 | ||
3a307ffc MF |
1016 | if (page == wc->w_target_page) { |
1017 | map_from = user_pos & (PAGE_CACHE_SIZE - 1); | |
1018 | map_to = map_from + user_len; | |
1019 | ||
1020 | if (new) | |
1021 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1022 | cluster_start, cluster_end, | |
1023 | new); | |
1024 | else | |
1025 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1026 | map_from, map_to, new); | |
1027 | if (ret) { | |
9517bac6 MF |
1028 | mlog_errno(ret); |
1029 | goto out; | |
1030 | } | |
1031 | ||
3a307ffc MF |
1032 | user_data_from = map_from; |
1033 | user_data_to = map_to; | |
9517bac6 | 1034 | if (new) { |
3a307ffc MF |
1035 | map_from = cluster_start; |
1036 | map_to = cluster_end; | |
9517bac6 MF |
1037 | } |
1038 | } else { | |
1039 | /* | |
1040 | * If we haven't allocated the new page yet, we | |
1041 | * shouldn't be writing it out without copying user | |
1042 | * data. This is likely a math error from the caller. | |
1043 | */ | |
1044 | BUG_ON(!new); | |
1045 | ||
3a307ffc MF |
1046 | map_from = cluster_start; |
1047 | map_to = cluster_end; | |
9517bac6 MF |
1048 | |
1049 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
3a307ffc | 1050 | cluster_start, cluster_end, new); |
9517bac6 MF |
1051 | if (ret) { |
1052 | mlog_errno(ret); | |
1053 | goto out; | |
1054 | } | |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * Parts of newly allocated pages need to be zero'd. | |
1059 | * | |
1060 | * Above, we have also rewritten 'to' and 'from' - as far as | |
1061 | * the rest of the function is concerned, the entire cluster | |
1062 | * range inside of a page needs to be written. | |
1063 | * | |
1064 | * We can skip this if the page is up to date - it's already | |
1065 | * been zero'd from being read in as a hole. | |
1066 | */ | |
1067 | if (new && !PageUptodate(page)) | |
1068 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), | |
3a307ffc | 1069 | cpos, user_data_from, user_data_to); |
9517bac6 MF |
1070 | |
1071 | flush_dcache_page(page); | |
1072 | ||
9517bac6 | 1073 | out: |
3a307ffc | 1074 | return ret; |
9517bac6 MF |
1075 | } |
1076 | ||
1077 | /* | |
3a307ffc | 1078 | * This function will only grab one clusters worth of pages. |
9517bac6 | 1079 | */ |
3a307ffc MF |
1080 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, |
1081 | struct ocfs2_write_ctxt *wc, | |
693c241a JB |
1082 | u32 cpos, loff_t user_pos, |
1083 | unsigned user_len, int new, | |
7307de80 | 1084 | struct page *mmap_page) |
9517bac6 | 1085 | { |
3a307ffc | 1086 | int ret = 0, i; |
693c241a | 1087 | unsigned long start, target_index, end_index, index; |
9517bac6 | 1088 | struct inode *inode = mapping->host; |
693c241a | 1089 | loff_t last_byte; |
9517bac6 | 1090 | |
3a307ffc | 1091 | target_index = user_pos >> PAGE_CACHE_SHIFT; |
9517bac6 MF |
1092 | |
1093 | /* | |
1094 | * Figure out how many pages we'll be manipulating here. For | |
60b11392 | 1095 | * non allocating write, we just change the one |
693c241a JB |
1096 | * page. Otherwise, we'll need a whole clusters worth. If we're |
1097 | * writing past i_size, we only need enough pages to cover the | |
1098 | * last page of the write. | |
9517bac6 | 1099 | */ |
9517bac6 | 1100 | if (new) { |
3a307ffc MF |
1101 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); |
1102 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); | |
693c241a JB |
1103 | /* |
1104 | * We need the index *past* the last page we could possibly | |
1105 | * touch. This is the page past the end of the write or | |
1106 | * i_size, whichever is greater. | |
1107 | */ | |
1108 | last_byte = max(user_pos + user_len, i_size_read(inode)); | |
1109 | BUG_ON(last_byte < 1); | |
1110 | end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1; | |
1111 | if ((start + wc->w_num_pages) > end_index) | |
1112 | wc->w_num_pages = end_index - start; | |
9517bac6 | 1113 | } else { |
3a307ffc MF |
1114 | wc->w_num_pages = 1; |
1115 | start = target_index; | |
9517bac6 MF |
1116 | } |
1117 | ||
3a307ffc | 1118 | for(i = 0; i < wc->w_num_pages; i++) { |
9517bac6 MF |
1119 | index = start + i; |
1120 | ||
7307de80 MF |
1121 | if (index == target_index && mmap_page) { |
1122 | /* | |
1123 | * ocfs2_pagemkwrite() is a little different | |
1124 | * and wants us to directly use the page | |
1125 | * passed in. | |
1126 | */ | |
1127 | lock_page(mmap_page); | |
1128 | ||
1129 | if (mmap_page->mapping != mapping) { | |
1130 | unlock_page(mmap_page); | |
1131 | /* | |
1132 | * Sanity check - the locking in | |
1133 | * ocfs2_pagemkwrite() should ensure | |
1134 | * that this code doesn't trigger. | |
1135 | */ | |
1136 | ret = -EINVAL; | |
1137 | mlog_errno(ret); | |
1138 | goto out; | |
1139 | } | |
1140 | ||
1141 | page_cache_get(mmap_page); | |
1142 | wc->w_pages[i] = mmap_page; | |
1143 | } else { | |
1144 | wc->w_pages[i] = find_or_create_page(mapping, index, | |
1145 | GFP_NOFS); | |
1146 | if (!wc->w_pages[i]) { | |
1147 | ret = -ENOMEM; | |
1148 | mlog_errno(ret); | |
1149 | goto out; | |
1150 | } | |
9517bac6 | 1151 | } |
3a307ffc MF |
1152 | |
1153 | if (index == target_index) | |
1154 | wc->w_target_page = wc->w_pages[i]; | |
9517bac6 | 1155 | } |
3a307ffc MF |
1156 | out: |
1157 | return ret; | |
1158 | } | |
1159 | ||
1160 | /* | |
1161 | * Prepare a single cluster for write one cluster into the file. | |
1162 | */ | |
1163 | static int ocfs2_write_cluster(struct address_space *mapping, | |
b27b7cbc | 1164 | u32 phys, unsigned int unwritten, |
e7432675 | 1165 | unsigned int should_zero, |
b27b7cbc | 1166 | struct ocfs2_alloc_context *data_ac, |
3a307ffc MF |
1167 | struct ocfs2_alloc_context *meta_ac, |
1168 | struct ocfs2_write_ctxt *wc, u32 cpos, | |
1169 | loff_t user_pos, unsigned user_len) | |
1170 | { | |
e7432675 | 1171 | int ret, i, new; |
3a307ffc MF |
1172 | u64 v_blkno, p_blkno; |
1173 | struct inode *inode = mapping->host; | |
f99b9b7c | 1174 | struct ocfs2_extent_tree et; |
3a307ffc MF |
1175 | |
1176 | new = phys == 0 ? 1 : 0; | |
9517bac6 | 1177 | if (new) { |
3a307ffc MF |
1178 | u32 tmp_pos; |
1179 | ||
9517bac6 MF |
1180 | /* |
1181 | * This is safe to call with the page locks - it won't take | |
1182 | * any additional semaphores or cluster locks. | |
1183 | */ | |
3a307ffc | 1184 | tmp_pos = cpos; |
0eb8d47e TM |
1185 | ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode, |
1186 | &tmp_pos, 1, 0, wc->w_di_bh, | |
1187 | wc->w_handle, data_ac, | |
1188 | meta_ac, NULL); | |
9517bac6 MF |
1189 | /* |
1190 | * This shouldn't happen because we must have already | |
1191 | * calculated the correct meta data allocation required. The | |
1192 | * internal tree allocation code should know how to increase | |
1193 | * transaction credits itself. | |
1194 | * | |
1195 | * If need be, we could handle -EAGAIN for a | |
1196 | * RESTART_TRANS here. | |
1197 | */ | |
1198 | mlog_bug_on_msg(ret == -EAGAIN, | |
1199 | "Inode %llu: EAGAIN return during allocation.\n", | |
1200 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1201 | if (ret < 0) { | |
1202 | mlog_errno(ret); | |
1203 | goto out; | |
1204 | } | |
b27b7cbc | 1205 | } else if (unwritten) { |
5e404e9e JB |
1206 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
1207 | wc->w_di_bh); | |
f99b9b7c | 1208 | ret = ocfs2_mark_extent_written(inode, &et, |
b27b7cbc | 1209 | wc->w_handle, cpos, 1, phys, |
f99b9b7c | 1210 | meta_ac, &wc->w_dealloc); |
b27b7cbc MF |
1211 | if (ret < 0) { |
1212 | mlog_errno(ret); | |
1213 | goto out; | |
1214 | } | |
1215 | } | |
3a307ffc | 1216 | |
b27b7cbc | 1217 | if (should_zero) |
3a307ffc | 1218 | v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos); |
b27b7cbc | 1219 | else |
3a307ffc | 1220 | v_blkno = user_pos >> inode->i_sb->s_blocksize_bits; |
9517bac6 | 1221 | |
3a307ffc MF |
1222 | /* |
1223 | * The only reason this should fail is due to an inability to | |
1224 | * find the extent added. | |
1225 | */ | |
49cb8d2d MF |
1226 | ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL, |
1227 | NULL); | |
9517bac6 | 1228 | if (ret < 0) { |
3a307ffc MF |
1229 | ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, " |
1230 | "at logical block %llu", | |
1231 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1232 | (unsigned long long)v_blkno); | |
9517bac6 MF |
1233 | goto out; |
1234 | } | |
1235 | ||
1236 | BUG_ON(p_blkno == 0); | |
1237 | ||
3a307ffc MF |
1238 | for(i = 0; i < wc->w_num_pages; i++) { |
1239 | int tmpret; | |
9517bac6 | 1240 | |
3a307ffc MF |
1241 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, |
1242 | wc->w_pages[i], cpos, | |
b27b7cbc MF |
1243 | user_pos, user_len, |
1244 | should_zero); | |
3a307ffc MF |
1245 | if (tmpret) { |
1246 | mlog_errno(tmpret); | |
1247 | if (ret == 0) | |
cbfa9639 | 1248 | ret = tmpret; |
3a307ffc | 1249 | } |
9517bac6 MF |
1250 | } |
1251 | ||
3a307ffc MF |
1252 | /* |
1253 | * We only have cleanup to do in case of allocating write. | |
1254 | */ | |
1255 | if (ret && new) | |
1256 | ocfs2_write_failure(inode, wc, user_pos, user_len); | |
1257 | ||
9517bac6 | 1258 | out: |
9517bac6 | 1259 | |
3a307ffc | 1260 | return ret; |
9517bac6 MF |
1261 | } |
1262 | ||
0d172baa MF |
1263 | static int ocfs2_write_cluster_by_desc(struct address_space *mapping, |
1264 | struct ocfs2_alloc_context *data_ac, | |
1265 | struct ocfs2_alloc_context *meta_ac, | |
1266 | struct ocfs2_write_ctxt *wc, | |
1267 | loff_t pos, unsigned len) | |
1268 | { | |
1269 | int ret, i; | |
db56246c MF |
1270 | loff_t cluster_off; |
1271 | unsigned int local_len = len; | |
0d172baa | 1272 | struct ocfs2_write_cluster_desc *desc; |
db56246c | 1273 | struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb); |
0d172baa MF |
1274 | |
1275 | for (i = 0; i < wc->w_clen; i++) { | |
1276 | desc = &wc->w_desc[i]; | |
1277 | ||
db56246c MF |
1278 | /* |
1279 | * We have to make sure that the total write passed in | |
1280 | * doesn't extend past a single cluster. | |
1281 | */ | |
1282 | local_len = len; | |
1283 | cluster_off = pos & (osb->s_clustersize - 1); | |
1284 | if ((cluster_off + local_len) > osb->s_clustersize) | |
1285 | local_len = osb->s_clustersize - cluster_off; | |
1286 | ||
b27b7cbc | 1287 | ret = ocfs2_write_cluster(mapping, desc->c_phys, |
e7432675 SM |
1288 | desc->c_unwritten, |
1289 | desc->c_needs_zero, | |
1290 | data_ac, meta_ac, | |
db56246c | 1291 | wc, desc->c_cpos, pos, local_len); |
0d172baa MF |
1292 | if (ret) { |
1293 | mlog_errno(ret); | |
1294 | goto out; | |
1295 | } | |
db56246c MF |
1296 | |
1297 | len -= local_len; | |
1298 | pos += local_len; | |
0d172baa MF |
1299 | } |
1300 | ||
1301 | ret = 0; | |
1302 | out: | |
1303 | return ret; | |
1304 | } | |
1305 | ||
3a307ffc MF |
1306 | /* |
1307 | * ocfs2_write_end() wants to know which parts of the target page it | |
1308 | * should complete the write on. It's easiest to compute them ahead of | |
1309 | * time when a more complete view of the write is available. | |
1310 | */ | |
1311 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, | |
1312 | struct ocfs2_write_ctxt *wc, | |
1313 | loff_t pos, unsigned len, int alloc) | |
9517bac6 | 1314 | { |
3a307ffc | 1315 | struct ocfs2_write_cluster_desc *desc; |
9517bac6 | 1316 | |
3a307ffc MF |
1317 | wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1); |
1318 | wc->w_target_to = wc->w_target_from + len; | |
1319 | ||
1320 | if (alloc == 0) | |
1321 | return; | |
1322 | ||
1323 | /* | |
1324 | * Allocating write - we may have different boundaries based | |
1325 | * on page size and cluster size. | |
1326 | * | |
1327 | * NOTE: We can no longer compute one value from the other as | |
1328 | * the actual write length and user provided length may be | |
1329 | * different. | |
1330 | */ | |
9517bac6 | 1331 | |
3a307ffc MF |
1332 | if (wc->w_large_pages) { |
1333 | /* | |
1334 | * We only care about the 1st and last cluster within | |
b27b7cbc | 1335 | * our range and whether they should be zero'd or not. Either |
3a307ffc MF |
1336 | * value may be extended out to the start/end of a |
1337 | * newly allocated cluster. | |
1338 | */ | |
1339 | desc = &wc->w_desc[0]; | |
e7432675 | 1340 | if (desc->c_needs_zero) |
3a307ffc MF |
1341 | ocfs2_figure_cluster_boundaries(osb, |
1342 | desc->c_cpos, | |
1343 | &wc->w_target_from, | |
1344 | NULL); | |
1345 | ||
1346 | desc = &wc->w_desc[wc->w_clen - 1]; | |
e7432675 | 1347 | if (desc->c_needs_zero) |
3a307ffc MF |
1348 | ocfs2_figure_cluster_boundaries(osb, |
1349 | desc->c_cpos, | |
1350 | NULL, | |
1351 | &wc->w_target_to); | |
1352 | } else { | |
1353 | wc->w_target_from = 0; | |
1354 | wc->w_target_to = PAGE_CACHE_SIZE; | |
1355 | } | |
9517bac6 MF |
1356 | } |
1357 | ||
0d172baa MF |
1358 | /* |
1359 | * Populate each single-cluster write descriptor in the write context | |
1360 | * with information about the i/o to be done. | |
b27b7cbc MF |
1361 | * |
1362 | * Returns the number of clusters that will have to be allocated, as | |
1363 | * well as a worst case estimate of the number of extent records that | |
1364 | * would have to be created during a write to an unwritten region. | |
0d172baa MF |
1365 | */ |
1366 | static int ocfs2_populate_write_desc(struct inode *inode, | |
1367 | struct ocfs2_write_ctxt *wc, | |
b27b7cbc MF |
1368 | unsigned int *clusters_to_alloc, |
1369 | unsigned int *extents_to_split) | |
9517bac6 | 1370 | { |
0d172baa | 1371 | int ret; |
3a307ffc | 1372 | struct ocfs2_write_cluster_desc *desc; |
0d172baa | 1373 | unsigned int num_clusters = 0; |
b27b7cbc | 1374 | unsigned int ext_flags = 0; |
0d172baa MF |
1375 | u32 phys = 0; |
1376 | int i; | |
9517bac6 | 1377 | |
b27b7cbc MF |
1378 | *clusters_to_alloc = 0; |
1379 | *extents_to_split = 0; | |
1380 | ||
3a307ffc MF |
1381 | for (i = 0; i < wc->w_clen; i++) { |
1382 | desc = &wc->w_desc[i]; | |
1383 | desc->c_cpos = wc->w_cpos + i; | |
1384 | ||
1385 | if (num_clusters == 0) { | |
b27b7cbc MF |
1386 | /* |
1387 | * Need to look up the next extent record. | |
1388 | */ | |
3a307ffc | 1389 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, |
b27b7cbc | 1390 | &num_clusters, &ext_flags); |
3a307ffc MF |
1391 | if (ret) { |
1392 | mlog_errno(ret); | |
607d44aa | 1393 | goto out; |
3a307ffc | 1394 | } |
b27b7cbc | 1395 | |
293b2f70 TM |
1396 | /* We should already CoW the refcountd extent. */ |
1397 | BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED); | |
1398 | ||
b27b7cbc MF |
1399 | /* |
1400 | * Assume worst case - that we're writing in | |
1401 | * the middle of the extent. | |
1402 | * | |
1403 | * We can assume that the write proceeds from | |
1404 | * left to right, in which case the extent | |
1405 | * insert code is smart enough to coalesce the | |
1406 | * next splits into the previous records created. | |
1407 | */ | |
1408 | if (ext_flags & OCFS2_EXT_UNWRITTEN) | |
1409 | *extents_to_split = *extents_to_split + 2; | |
3a307ffc MF |
1410 | } else if (phys) { |
1411 | /* | |
1412 | * Only increment phys if it doesn't describe | |
1413 | * a hole. | |
1414 | */ | |
1415 | phys++; | |
1416 | } | |
1417 | ||
e7432675 SM |
1418 | /* |
1419 | * If w_first_new_cpos is < UINT_MAX, we have a non-sparse | |
1420 | * file that got extended. w_first_new_cpos tells us | |
1421 | * where the newly allocated clusters are so we can | |
1422 | * zero them. | |
1423 | */ | |
1424 | if (desc->c_cpos >= wc->w_first_new_cpos) { | |
1425 | BUG_ON(phys == 0); | |
1426 | desc->c_needs_zero = 1; | |
1427 | } | |
1428 | ||
3a307ffc MF |
1429 | desc->c_phys = phys; |
1430 | if (phys == 0) { | |
1431 | desc->c_new = 1; | |
e7432675 | 1432 | desc->c_needs_zero = 1; |
0d172baa | 1433 | *clusters_to_alloc = *clusters_to_alloc + 1; |
3a307ffc | 1434 | } |
e7432675 SM |
1435 | |
1436 | if (ext_flags & OCFS2_EXT_UNWRITTEN) { | |
b27b7cbc | 1437 | desc->c_unwritten = 1; |
e7432675 SM |
1438 | desc->c_needs_zero = 1; |
1439 | } | |
3a307ffc MF |
1440 | |
1441 | num_clusters--; | |
9517bac6 MF |
1442 | } |
1443 | ||
0d172baa MF |
1444 | ret = 0; |
1445 | out: | |
1446 | return ret; | |
1447 | } | |
1448 | ||
1afc32b9 MF |
1449 | static int ocfs2_write_begin_inline(struct address_space *mapping, |
1450 | struct inode *inode, | |
1451 | struct ocfs2_write_ctxt *wc) | |
1452 | { | |
1453 | int ret; | |
1454 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1455 | struct page *page; | |
1456 | handle_t *handle; | |
1457 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1458 | ||
1459 | page = find_or_create_page(mapping, 0, GFP_NOFS); | |
1460 | if (!page) { | |
1461 | ret = -ENOMEM; | |
1462 | mlog_errno(ret); | |
1463 | goto out; | |
1464 | } | |
1465 | /* | |
1466 | * If we don't set w_num_pages then this page won't get unlocked | |
1467 | * and freed on cleanup of the write context. | |
1468 | */ | |
1469 | wc->w_pages[0] = wc->w_target_page = page; | |
1470 | wc->w_num_pages = 1; | |
1471 | ||
1472 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1473 | if (IS_ERR(handle)) { | |
1474 | ret = PTR_ERR(handle); | |
1475 | mlog_errno(ret); | |
1476 | goto out; | |
1477 | } | |
1478 | ||
0cf2f763 | 1479 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 1480 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
1481 | if (ret) { |
1482 | ocfs2_commit_trans(osb, handle); | |
1483 | ||
1484 | mlog_errno(ret); | |
1485 | goto out; | |
1486 | } | |
1487 | ||
1488 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
1489 | ocfs2_set_inode_data_inline(inode, di); | |
1490 | ||
1491 | if (!PageUptodate(page)) { | |
1492 | ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh); | |
1493 | if (ret) { | |
1494 | ocfs2_commit_trans(osb, handle); | |
1495 | ||
1496 | goto out; | |
1497 | } | |
1498 | } | |
1499 | ||
1500 | wc->w_handle = handle; | |
1501 | out: | |
1502 | return ret; | |
1503 | } | |
1504 | ||
1505 | int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size) | |
1506 | { | |
1507 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1508 | ||
0d8a4e0c | 1509 | if (new_size <= le16_to_cpu(di->id2.i_data.id_count)) |
1afc32b9 MF |
1510 | return 1; |
1511 | return 0; | |
1512 | } | |
1513 | ||
1514 | static int ocfs2_try_to_write_inline_data(struct address_space *mapping, | |
1515 | struct inode *inode, loff_t pos, | |
1516 | unsigned len, struct page *mmap_page, | |
1517 | struct ocfs2_write_ctxt *wc) | |
1518 | { | |
1519 | int ret, written = 0; | |
1520 | loff_t end = pos + len; | |
1521 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
d9ae49d6 | 1522 | struct ocfs2_dinode *di = NULL; |
1afc32b9 MF |
1523 | |
1524 | mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n", | |
1525 | (unsigned long long)oi->ip_blkno, len, (unsigned long long)pos, | |
1526 | oi->ip_dyn_features); | |
1527 | ||
1528 | /* | |
1529 | * Handle inodes which already have inline data 1st. | |
1530 | */ | |
1531 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1532 | if (mmap_page == NULL && | |
1533 | ocfs2_size_fits_inline_data(wc->w_di_bh, end)) | |
1534 | goto do_inline_write; | |
1535 | ||
1536 | /* | |
1537 | * The write won't fit - we have to give this inode an | |
1538 | * inline extent list now. | |
1539 | */ | |
1540 | ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh); | |
1541 | if (ret) | |
1542 | mlog_errno(ret); | |
1543 | goto out; | |
1544 | } | |
1545 | ||
1546 | /* | |
1547 | * Check whether the inode can accept inline data. | |
1548 | */ | |
1549 | if (oi->ip_clusters != 0 || i_size_read(inode) != 0) | |
1550 | return 0; | |
1551 | ||
1552 | /* | |
1553 | * Check whether the write can fit. | |
1554 | */ | |
d9ae49d6 TY |
1555 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; |
1556 | if (mmap_page || | |
1557 | end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) | |
1afc32b9 MF |
1558 | return 0; |
1559 | ||
1560 | do_inline_write: | |
1561 | ret = ocfs2_write_begin_inline(mapping, inode, wc); | |
1562 | if (ret) { | |
1563 | mlog_errno(ret); | |
1564 | goto out; | |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * This signals to the caller that the data can be written | |
1569 | * inline. | |
1570 | */ | |
1571 | written = 1; | |
1572 | out: | |
1573 | return written ? written : ret; | |
1574 | } | |
1575 | ||
65ed39d6 MF |
1576 | /* |
1577 | * This function only does anything for file systems which can't | |
1578 | * handle sparse files. | |
1579 | * | |
1580 | * What we want to do here is fill in any hole between the current end | |
1581 | * of allocation and the end of our write. That way the rest of the | |
1582 | * write path can treat it as an non-allocating write, which has no | |
1583 | * special case code for sparse/nonsparse files. | |
1584 | */ | |
5693486b JB |
1585 | static int ocfs2_expand_nonsparse_inode(struct inode *inode, |
1586 | struct buffer_head *di_bh, | |
1587 | loff_t pos, unsigned len, | |
65ed39d6 MF |
1588 | struct ocfs2_write_ctxt *wc) |
1589 | { | |
1590 | int ret; | |
65ed39d6 MF |
1591 | loff_t newsize = pos + len; |
1592 | ||
5693486b | 1593 | BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); |
65ed39d6 MF |
1594 | |
1595 | if (newsize <= i_size_read(inode)) | |
1596 | return 0; | |
1597 | ||
5693486b | 1598 | ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos); |
65ed39d6 MF |
1599 | if (ret) |
1600 | mlog_errno(ret); | |
1601 | ||
e7432675 SM |
1602 | wc->w_first_new_cpos = |
1603 | ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)); | |
1604 | ||
65ed39d6 MF |
1605 | return ret; |
1606 | } | |
1607 | ||
5693486b JB |
1608 | static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh, |
1609 | loff_t pos) | |
1610 | { | |
1611 | int ret = 0; | |
1612 | ||
1613 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); | |
1614 | if (pos > i_size_read(inode)) | |
1615 | ret = ocfs2_zero_extend(inode, di_bh, pos); | |
1616 | ||
1617 | return ret; | |
1618 | } | |
1619 | ||
50308d81 TM |
1620 | /* |
1621 | * Try to flush truncate logs if we can free enough clusters from it. | |
1622 | * As for return value, "< 0" means error, "0" no space and "1" means | |
1623 | * we have freed enough spaces and let the caller try to allocate again. | |
1624 | */ | |
1625 | static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb, | |
1626 | unsigned int needed) | |
1627 | { | |
1628 | tid_t target; | |
1629 | int ret = 0; | |
1630 | unsigned int truncated_clusters; | |
1631 | ||
1632 | mutex_lock(&osb->osb_tl_inode->i_mutex); | |
1633 | truncated_clusters = osb->truncated_clusters; | |
1634 | mutex_unlock(&osb->osb_tl_inode->i_mutex); | |
1635 | ||
1636 | /* | |
1637 | * Check whether we can succeed in allocating if we free | |
1638 | * the truncate log. | |
1639 | */ | |
1640 | if (truncated_clusters < needed) | |
1641 | goto out; | |
1642 | ||
1643 | ret = ocfs2_flush_truncate_log(osb); | |
1644 | if (ret) { | |
1645 | mlog_errno(ret); | |
1646 | goto out; | |
1647 | } | |
1648 | ||
1649 | if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) { | |
1650 | jbd2_log_wait_commit(osb->journal->j_journal, target); | |
1651 | ret = 1; | |
1652 | } | |
1653 | out: | |
1654 | return ret; | |
1655 | } | |
1656 | ||
0378da0f TM |
1657 | int ocfs2_write_begin_nolock(struct file *filp, |
1658 | struct address_space *mapping, | |
0d172baa MF |
1659 | loff_t pos, unsigned len, unsigned flags, |
1660 | struct page **pagep, void **fsdata, | |
1661 | struct buffer_head *di_bh, struct page *mmap_page) | |
1662 | { | |
e7432675 | 1663 | int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS; |
50308d81 | 1664 | unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0; |
0d172baa MF |
1665 | struct ocfs2_write_ctxt *wc; |
1666 | struct inode *inode = mapping->host; | |
1667 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1668 | struct ocfs2_dinode *di; | |
1669 | struct ocfs2_alloc_context *data_ac = NULL; | |
1670 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1671 | handle_t *handle; | |
f99b9b7c | 1672 | struct ocfs2_extent_tree et; |
50308d81 | 1673 | int try_free = 1, ret1; |
0d172baa | 1674 | |
50308d81 | 1675 | try_again: |
0d172baa MF |
1676 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh); |
1677 | if (ret) { | |
1678 | mlog_errno(ret); | |
1679 | return ret; | |
1680 | } | |
1681 | ||
1afc32b9 MF |
1682 | if (ocfs2_supports_inline_data(osb)) { |
1683 | ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len, | |
1684 | mmap_page, wc); | |
1685 | if (ret == 1) { | |
1686 | ret = 0; | |
1687 | goto success; | |
1688 | } | |
1689 | if (ret < 0) { | |
1690 | mlog_errno(ret); | |
1691 | goto out; | |
1692 | } | |
1693 | } | |
1694 | ||
5693486b JB |
1695 | if (ocfs2_sparse_alloc(osb)) |
1696 | ret = ocfs2_zero_tail(inode, di_bh, pos); | |
1697 | else | |
1698 | ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len, | |
1699 | wc); | |
65ed39d6 MF |
1700 | if (ret) { |
1701 | mlog_errno(ret); | |
1702 | goto out; | |
1703 | } | |
1704 | ||
293b2f70 TM |
1705 | ret = ocfs2_check_range_for_refcount(inode, pos, len); |
1706 | if (ret < 0) { | |
1707 | mlog_errno(ret); | |
1708 | goto out; | |
1709 | } else if (ret == 1) { | |
50308d81 | 1710 | clusters_need = wc->w_clen; |
15502712 | 1711 | ret = ocfs2_refcount_cow(inode, filp, di_bh, |
37f8a2bf | 1712 | wc->w_cpos, wc->w_clen, UINT_MAX); |
293b2f70 TM |
1713 | if (ret) { |
1714 | mlog_errno(ret); | |
1715 | goto out; | |
1716 | } | |
1717 | } | |
1718 | ||
b27b7cbc MF |
1719 | ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc, |
1720 | &extents_to_split); | |
0d172baa MF |
1721 | if (ret) { |
1722 | mlog_errno(ret); | |
1723 | goto out; | |
1724 | } | |
50308d81 | 1725 | clusters_need += clusters_to_alloc; |
0d172baa MF |
1726 | |
1727 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1728 | ||
3a307ffc MF |
1729 | /* |
1730 | * We set w_target_from, w_target_to here so that | |
1731 | * ocfs2_write_end() knows which range in the target page to | |
1732 | * write out. An allocation requires that we write the entire | |
1733 | * cluster range. | |
1734 | */ | |
b27b7cbc | 1735 | if (clusters_to_alloc || extents_to_split) { |
3a307ffc MF |
1736 | /* |
1737 | * XXX: We are stretching the limits of | |
b27b7cbc | 1738 | * ocfs2_lock_allocators(). It greatly over-estimates |
3a307ffc MF |
1739 | * the work to be done. |
1740 | */ | |
e7d4cb6b TM |
1741 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u," |
1742 | " clusters_to_add = %u, extents_to_split = %u\n", | |
1743 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1744 | (long long)i_size_read(inode), le32_to_cpu(di->i_clusters), | |
1745 | clusters_to_alloc, extents_to_split); | |
1746 | ||
5e404e9e JB |
1747 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
1748 | wc->w_di_bh); | |
f99b9b7c | 1749 | ret = ocfs2_lock_allocators(inode, &et, |
231b87d1 | 1750 | clusters_to_alloc, extents_to_split, |
f99b9b7c | 1751 | &data_ac, &meta_ac); |
9517bac6 MF |
1752 | if (ret) { |
1753 | mlog_errno(ret); | |
607d44aa | 1754 | goto out; |
9517bac6 MF |
1755 | } |
1756 | ||
4fe370af MF |
1757 | if (data_ac) |
1758 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; | |
1759 | ||
811f933d TM |
1760 | credits = ocfs2_calc_extend_credits(inode->i_sb, |
1761 | &di->id2.i_list, | |
3a307ffc MF |
1762 | clusters_to_alloc); |
1763 | ||
9517bac6 MF |
1764 | } |
1765 | ||
e7432675 SM |
1766 | /* |
1767 | * We have to zero sparse allocated clusters, unwritten extent clusters, | |
1768 | * and non-sparse clusters we just extended. For non-sparse writes, | |
1769 | * we know zeros will only be needed in the first and/or last cluster. | |
1770 | */ | |
1771 | if (clusters_to_alloc || extents_to_split || | |
8379e7c4 SM |
1772 | (wc->w_clen && (wc->w_desc[0].c_needs_zero || |
1773 | wc->w_desc[wc->w_clen - 1].c_needs_zero))) | |
e7432675 SM |
1774 | cluster_of_pages = 1; |
1775 | else | |
1776 | cluster_of_pages = 0; | |
1777 | ||
1778 | ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages); | |
3a307ffc | 1779 | |
9517bac6 MF |
1780 | handle = ocfs2_start_trans(osb, credits); |
1781 | if (IS_ERR(handle)) { | |
1782 | ret = PTR_ERR(handle); | |
1783 | mlog_errno(ret); | |
607d44aa | 1784 | goto out; |
9517bac6 MF |
1785 | } |
1786 | ||
3a307ffc MF |
1787 | wc->w_handle = handle; |
1788 | ||
5dd4056d CH |
1789 | if (clusters_to_alloc) { |
1790 | ret = dquot_alloc_space_nodirty(inode, | |
1791 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); | |
1792 | if (ret) | |
1793 | goto out_commit; | |
a90714c1 | 1794 | } |
3a307ffc MF |
1795 | /* |
1796 | * We don't want this to fail in ocfs2_write_end(), so do it | |
1797 | * here. | |
1798 | */ | |
0cf2f763 | 1799 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 1800 | OCFS2_JOURNAL_ACCESS_WRITE); |
3a307ffc | 1801 | if (ret) { |
9517bac6 | 1802 | mlog_errno(ret); |
a90714c1 | 1803 | goto out_quota; |
9517bac6 MF |
1804 | } |
1805 | ||
3a307ffc MF |
1806 | /* |
1807 | * Fill our page array first. That way we've grabbed enough so | |
1808 | * that we can zero and flush if we error after adding the | |
1809 | * extent. | |
1810 | */ | |
693c241a | 1811 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len, |
e7432675 | 1812 | cluster_of_pages, mmap_page); |
9517bac6 MF |
1813 | if (ret) { |
1814 | mlog_errno(ret); | |
a90714c1 | 1815 | goto out_quota; |
9517bac6 MF |
1816 | } |
1817 | ||
0d172baa MF |
1818 | ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos, |
1819 | len); | |
1820 | if (ret) { | |
1821 | mlog_errno(ret); | |
a90714c1 | 1822 | goto out_quota; |
9517bac6 | 1823 | } |
9517bac6 | 1824 | |
3a307ffc MF |
1825 | if (data_ac) |
1826 | ocfs2_free_alloc_context(data_ac); | |
1827 | if (meta_ac) | |
1828 | ocfs2_free_alloc_context(meta_ac); | |
9517bac6 | 1829 | |
1afc32b9 | 1830 | success: |
3a307ffc MF |
1831 | *pagep = wc->w_target_page; |
1832 | *fsdata = wc; | |
1833 | return 0; | |
a90714c1 JK |
1834 | out_quota: |
1835 | if (clusters_to_alloc) | |
5dd4056d | 1836 | dquot_free_space(inode, |
a90714c1 | 1837 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); |
9517bac6 MF |
1838 | out_commit: |
1839 | ocfs2_commit_trans(osb, handle); | |
1840 | ||
9517bac6 | 1841 | out: |
3a307ffc MF |
1842 | ocfs2_free_write_ctxt(wc); |
1843 | ||
9517bac6 MF |
1844 | if (data_ac) |
1845 | ocfs2_free_alloc_context(data_ac); | |
1846 | if (meta_ac) | |
1847 | ocfs2_free_alloc_context(meta_ac); | |
50308d81 TM |
1848 | |
1849 | if (ret == -ENOSPC && try_free) { | |
1850 | /* | |
1851 | * Try to free some truncate log so that we can have enough | |
1852 | * clusters to allocate. | |
1853 | */ | |
1854 | try_free = 0; | |
1855 | ||
1856 | ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); | |
1857 | if (ret1 == 1) | |
1858 | goto try_again; | |
1859 | ||
1860 | if (ret1 < 0) | |
1861 | mlog_errno(ret1); | |
1862 | } | |
1863 | ||
3a307ffc MF |
1864 | return ret; |
1865 | } | |
1866 | ||
b6af1bcd NP |
1867 | static int ocfs2_write_begin(struct file *file, struct address_space *mapping, |
1868 | loff_t pos, unsigned len, unsigned flags, | |
1869 | struct page **pagep, void **fsdata) | |
607d44aa MF |
1870 | { |
1871 | int ret; | |
1872 | struct buffer_head *di_bh = NULL; | |
1873 | struct inode *inode = mapping->host; | |
1874 | ||
e63aecb6 | 1875 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
607d44aa MF |
1876 | if (ret) { |
1877 | mlog_errno(ret); | |
1878 | return ret; | |
1879 | } | |
1880 | ||
1881 | /* | |
1882 | * Take alloc sem here to prevent concurrent lookups. That way | |
1883 | * the mapping, zeroing and tree manipulation within | |
1884 | * ocfs2_write() will be safe against ->readpage(). This | |
1885 | * should also serve to lock out allocation from a shared | |
1886 | * writeable region. | |
1887 | */ | |
1888 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1889 | ||
0378da0f | 1890 | ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep, |
7307de80 | 1891 | fsdata, di_bh, NULL); |
607d44aa MF |
1892 | if (ret) { |
1893 | mlog_errno(ret); | |
c934a92d | 1894 | goto out_fail; |
607d44aa MF |
1895 | } |
1896 | ||
1897 | brelse(di_bh); | |
1898 | ||
1899 | return 0; | |
1900 | ||
607d44aa MF |
1901 | out_fail: |
1902 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1903 | ||
1904 | brelse(di_bh); | |
e63aecb6 | 1905 | ocfs2_inode_unlock(inode, 1); |
607d44aa MF |
1906 | |
1907 | return ret; | |
1908 | } | |
1909 | ||
1afc32b9 MF |
1910 | static void ocfs2_write_end_inline(struct inode *inode, loff_t pos, |
1911 | unsigned len, unsigned *copied, | |
1912 | struct ocfs2_dinode *di, | |
1913 | struct ocfs2_write_ctxt *wc) | |
1914 | { | |
1915 | void *kaddr; | |
1916 | ||
1917 | if (unlikely(*copied < len)) { | |
1918 | if (!PageUptodate(wc->w_target_page)) { | |
1919 | *copied = 0; | |
1920 | return; | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | kaddr = kmap_atomic(wc->w_target_page, KM_USER0); | |
1925 | memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied); | |
1926 | kunmap_atomic(kaddr, KM_USER0); | |
1927 | ||
1928 | mlog(0, "Data written to inode at offset %llu. " | |
1929 | "id_count = %u, copied = %u, i_dyn_features = 0x%x\n", | |
1930 | (unsigned long long)pos, *copied, | |
1931 | le16_to_cpu(di->id2.i_data.id_count), | |
1932 | le16_to_cpu(di->i_dyn_features)); | |
1933 | } | |
1934 | ||
7307de80 MF |
1935 | int ocfs2_write_end_nolock(struct address_space *mapping, |
1936 | loff_t pos, unsigned len, unsigned copied, | |
1937 | struct page *page, void *fsdata) | |
3a307ffc MF |
1938 | { |
1939 | int i; | |
1940 | unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1); | |
1941 | struct inode *inode = mapping->host; | |
1942 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1943 | struct ocfs2_write_ctxt *wc = fsdata; | |
1944 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1945 | handle_t *handle = wc->w_handle; | |
1946 | struct page *tmppage; | |
1947 | ||
1afc32b9 MF |
1948 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
1949 | ocfs2_write_end_inline(inode, pos, len, &copied, di, wc); | |
1950 | goto out_write_size; | |
1951 | } | |
1952 | ||
3a307ffc MF |
1953 | if (unlikely(copied < len)) { |
1954 | if (!PageUptodate(wc->w_target_page)) | |
1955 | copied = 0; | |
1956 | ||
1957 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, | |
1958 | start+len); | |
1959 | } | |
1960 | flush_dcache_page(wc->w_target_page); | |
1961 | ||
1962 | for(i = 0; i < wc->w_num_pages; i++) { | |
1963 | tmppage = wc->w_pages[i]; | |
1964 | ||
1965 | if (tmppage == wc->w_target_page) { | |
1966 | from = wc->w_target_from; | |
1967 | to = wc->w_target_to; | |
1968 | ||
1969 | BUG_ON(from > PAGE_CACHE_SIZE || | |
1970 | to > PAGE_CACHE_SIZE || | |
1971 | to < from); | |
1972 | } else { | |
1973 | /* | |
1974 | * Pages adjacent to the target (if any) imply | |
1975 | * a hole-filling write in which case we want | |
1976 | * to flush their entire range. | |
1977 | */ | |
1978 | from = 0; | |
1979 | to = PAGE_CACHE_SIZE; | |
1980 | } | |
1981 | ||
961cecbe | 1982 | if (page_has_buffers(tmppage)) { |
53ef99ca | 1983 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 1984 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
1985 | block_commit_write(tmppage, from, to); |
1986 | } | |
3a307ffc MF |
1987 | } |
1988 | ||
1afc32b9 | 1989 | out_write_size: |
3a307ffc MF |
1990 | pos += copied; |
1991 | if (pos > inode->i_size) { | |
1992 | i_size_write(inode, pos); | |
1993 | mark_inode_dirty(inode); | |
1994 | } | |
1995 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
1996 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
1997 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
1998 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1999 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
3a307ffc MF |
2000 | ocfs2_journal_dirty(handle, wc->w_di_bh); |
2001 | ||
2002 | ocfs2_commit_trans(osb, handle); | |
59a5e416 | 2003 | |
b27b7cbc MF |
2004 | ocfs2_run_deallocs(osb, &wc->w_dealloc); |
2005 | ||
607d44aa MF |
2006 | ocfs2_free_write_ctxt(wc); |
2007 | ||
2008 | return copied; | |
2009 | } | |
2010 | ||
b6af1bcd NP |
2011 | static int ocfs2_write_end(struct file *file, struct address_space *mapping, |
2012 | loff_t pos, unsigned len, unsigned copied, | |
2013 | struct page *page, void *fsdata) | |
607d44aa MF |
2014 | { |
2015 | int ret; | |
2016 | struct inode *inode = mapping->host; | |
2017 | ||
2018 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata); | |
2019 | ||
3a307ffc | 2020 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
e63aecb6 | 2021 | ocfs2_inode_unlock(inode, 1); |
9517bac6 | 2022 | |
607d44aa | 2023 | return ret; |
9517bac6 MF |
2024 | } |
2025 | ||
f5e54d6e | 2026 | const struct address_space_operations ocfs2_aops = { |
1fca3a05 HH |
2027 | .readpage = ocfs2_readpage, |
2028 | .readpages = ocfs2_readpages, | |
2029 | .writepage = ocfs2_writepage, | |
2030 | .write_begin = ocfs2_write_begin, | |
2031 | .write_end = ocfs2_write_end, | |
2032 | .bmap = ocfs2_bmap, | |
2033 | .sync_page = block_sync_page, | |
2034 | .direct_IO = ocfs2_direct_IO, | |
2035 | .invalidatepage = ocfs2_invalidatepage, | |
2036 | .releasepage = ocfs2_releasepage, | |
2037 | .migratepage = buffer_migrate_page, | |
2038 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 2039 | .error_remove_page = generic_error_remove_page, |
ccd979bd | 2040 | }; |