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> |
24c40b32 | 31 | #include <linux/blkdev.h> |
ccd979bd | 32 | |
ccd979bd MF |
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" |
9558156b | 48 | #include "ocfs2_trace.h" |
ccd979bd MF |
49 | |
50 | #include "buffer_head_io.h" | |
24c40b32 JQ |
51 | #include "dir.h" |
52 | #include "namei.h" | |
53 | #include "sysfile.h" | |
ccd979bd MF |
54 | |
55 | static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock, | |
56 | struct buffer_head *bh_result, int create) | |
57 | { | |
58 | int err = -EIO; | |
59 | int status; | |
60 | struct ocfs2_dinode *fe = NULL; | |
61 | struct buffer_head *bh = NULL; | |
62 | struct buffer_head *buffer_cache_bh = NULL; | |
63 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
64 | void *kaddr; | |
65 | ||
9558156b TM |
66 | trace_ocfs2_symlink_get_block( |
67 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
68 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
69 | |
70 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); | |
71 | ||
72 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { | |
73 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", | |
74 | (unsigned long long)iblock); | |
75 | goto bail; | |
76 | } | |
77 | ||
b657c95c | 78 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
79 | if (status < 0) { |
80 | mlog_errno(status); | |
81 | goto bail; | |
82 | } | |
83 | fe = (struct ocfs2_dinode *) bh->b_data; | |
84 | ||
ccd979bd MF |
85 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, |
86 | le32_to_cpu(fe->i_clusters))) { | |
7391a294 | 87 | err = -ENOMEM; |
ccd979bd MF |
88 | mlog(ML_ERROR, "block offset is outside the allocated size: " |
89 | "%llu\n", (unsigned long long)iblock); | |
90 | goto bail; | |
91 | } | |
92 | ||
93 | /* We don't use the page cache to create symlink data, so if | |
94 | * need be, copy it over from the buffer cache. */ | |
95 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { | |
96 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + | |
97 | iblock; | |
98 | buffer_cache_bh = sb_getblk(osb->sb, blkno); | |
99 | if (!buffer_cache_bh) { | |
7391a294 | 100 | err = -ENOMEM; |
ccd979bd MF |
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)) { | |
c4bc8dcb | 111 | kaddr = kmap_atomic(bh_result->b_page); |
ccd979bd MF |
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); | |
c4bc8dcb | 119 | kunmap_atomic(kaddr); |
ccd979bd MF |
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: | |
a81cb88b | 131 | brelse(bh); |
ccd979bd | 132 | |
ccd979bd MF |
133 | return err; |
134 | } | |
135 | ||
6f70fa51 TM |
136 | int ocfs2_get_block(struct inode *inode, sector_t iblock, |
137 | struct buffer_head *bh_result, int create) | |
ccd979bd MF |
138 | { |
139 | int err = 0; | |
49cb8d2d | 140 | unsigned int ext_flags; |
628a24f5 MF |
141 | u64 max_blocks = bh_result->b_size >> inode->i_blkbits; |
142 | u64 p_blkno, count, past_eof; | |
25baf2da | 143 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd | 144 | |
9558156b TM |
145 | trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno, |
146 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
147 | |
148 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) | |
149 | mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", | |
150 | inode, inode->i_ino); | |
151 | ||
152 | if (S_ISLNK(inode->i_mode)) { | |
153 | /* this always does I/O for some reason. */ | |
154 | err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); | |
155 | goto bail; | |
156 | } | |
157 | ||
628a24f5 | 158 | err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count, |
49cb8d2d | 159 | &ext_flags); |
ccd979bd MF |
160 | if (err) { |
161 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " | |
b0697053 MF |
162 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, |
163 | (unsigned long long)p_blkno); | |
ccd979bd MF |
164 | goto bail; |
165 | } | |
166 | ||
628a24f5 MF |
167 | if (max_blocks < count) |
168 | count = max_blocks; | |
169 | ||
25baf2da MF |
170 | /* |
171 | * ocfs2 never allocates in this function - the only time we | |
172 | * need to use BH_New is when we're extending i_size on a file | |
173 | * system which doesn't support holes, in which case BH_New | |
ebdec241 | 174 | * allows __block_write_begin() to zero. |
c0420ad2 CL |
175 | * |
176 | * If we see this on a sparse file system, then a truncate has | |
177 | * raced us and removed the cluster. In this case, we clear | |
178 | * the buffers dirty and uptodate bits and let the buffer code | |
179 | * ignore it as a hole. | |
25baf2da | 180 | */ |
c0420ad2 CL |
181 | if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) { |
182 | clear_buffer_dirty(bh_result); | |
183 | clear_buffer_uptodate(bh_result); | |
184 | goto bail; | |
185 | } | |
25baf2da | 186 | |
49cb8d2d MF |
187 | /* Treat the unwritten extent as a hole for zeroing purposes. */ |
188 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
25baf2da MF |
189 | map_bh(bh_result, inode->i_sb, p_blkno); |
190 | ||
628a24f5 MF |
191 | bh_result->b_size = count << inode->i_blkbits; |
192 | ||
25baf2da MF |
193 | if (!ocfs2_sparse_alloc(osb)) { |
194 | if (p_blkno == 0) { | |
195 | err = -EIO; | |
196 | mlog(ML_ERROR, | |
197 | "iblock = %llu p_blkno = %llu blkno=(%llu)\n", | |
198 | (unsigned long long)iblock, | |
199 | (unsigned long long)p_blkno, | |
200 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
201 | mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters); | |
202 | dump_stack(); | |
1f4cea37 | 203 | goto bail; |
25baf2da | 204 | } |
25baf2da | 205 | } |
ccd979bd | 206 | |
5693486b | 207 | past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
9558156b TM |
208 | |
209 | trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno, | |
210 | (unsigned long long)past_eof); | |
5693486b JB |
211 | if (create && (iblock >= past_eof)) |
212 | set_buffer_new(bh_result); | |
213 | ||
ccd979bd MF |
214 | bail: |
215 | if (err < 0) | |
216 | err = -EIO; | |
217 | ||
ccd979bd MF |
218 | return err; |
219 | } | |
220 | ||
1afc32b9 MF |
221 | int ocfs2_read_inline_data(struct inode *inode, struct page *page, |
222 | struct buffer_head *di_bh) | |
6798d35a MF |
223 | { |
224 | void *kaddr; | |
d2849fb2 | 225 | loff_t size; |
6798d35a MF |
226 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
227 | ||
228 | if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) { | |
229 | ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag", | |
230 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
231 | return -EROFS; | |
232 | } | |
233 | ||
234 | size = i_size_read(inode); | |
235 | ||
236 | if (size > PAGE_CACHE_SIZE || | |
d9ae49d6 | 237 | size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) { |
6798d35a | 238 | ocfs2_error(inode->i_sb, |
d2849fb2 JK |
239 | "Inode %llu has with inline data has bad size: %Lu", |
240 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
241 | (unsigned long long)size); | |
6798d35a MF |
242 | return -EROFS; |
243 | } | |
244 | ||
c4bc8dcb | 245 | kaddr = kmap_atomic(page); |
6798d35a MF |
246 | if (size) |
247 | memcpy(kaddr, di->id2.i_data.id_data, size); | |
248 | /* Clear the remaining part of the page */ | |
249 | memset(kaddr + size, 0, PAGE_CACHE_SIZE - size); | |
250 | flush_dcache_page(page); | |
c4bc8dcb | 251 | kunmap_atomic(kaddr); |
6798d35a MF |
252 | |
253 | SetPageUptodate(page); | |
254 | ||
255 | return 0; | |
256 | } | |
257 | ||
258 | static int ocfs2_readpage_inline(struct inode *inode, struct page *page) | |
259 | { | |
260 | int ret; | |
261 | struct buffer_head *di_bh = NULL; | |
6798d35a MF |
262 | |
263 | BUG_ON(!PageLocked(page)); | |
86c838b0 | 264 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)); |
6798d35a | 265 | |
b657c95c | 266 | ret = ocfs2_read_inode_block(inode, &di_bh); |
6798d35a MF |
267 | if (ret) { |
268 | mlog_errno(ret); | |
269 | goto out; | |
270 | } | |
271 | ||
272 | ret = ocfs2_read_inline_data(inode, page, di_bh); | |
273 | out: | |
274 | unlock_page(page); | |
275 | ||
276 | brelse(di_bh); | |
277 | return ret; | |
278 | } | |
279 | ||
ccd979bd MF |
280 | static int ocfs2_readpage(struct file *file, struct page *page) |
281 | { | |
282 | struct inode *inode = page->mapping->host; | |
6798d35a | 283 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd MF |
284 | loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT; |
285 | int ret, unlock = 1; | |
286 | ||
9558156b TM |
287 | trace_ocfs2_readpage((unsigned long long)oi->ip_blkno, |
288 | (page ? page->index : 0)); | |
ccd979bd | 289 | |
e63aecb6 | 290 | ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page); |
ccd979bd MF |
291 | if (ret != 0) { |
292 | if (ret == AOP_TRUNCATED_PAGE) | |
293 | unlock = 0; | |
294 | mlog_errno(ret); | |
295 | goto out; | |
296 | } | |
297 | ||
6798d35a | 298 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { |
c7e25e6e JK |
299 | /* |
300 | * Unlock the page and cycle ip_alloc_sem so that we don't | |
301 | * busyloop waiting for ip_alloc_sem to unlock | |
302 | */ | |
e9dfc0b2 | 303 | ret = AOP_TRUNCATED_PAGE; |
c7e25e6e JK |
304 | unlock_page(page); |
305 | unlock = 0; | |
306 | down_read(&oi->ip_alloc_sem); | |
307 | up_read(&oi->ip_alloc_sem); | |
e63aecb6 | 308 | goto out_inode_unlock; |
e9dfc0b2 | 309 | } |
ccd979bd MF |
310 | |
311 | /* | |
312 | * i_size might have just been updated as we grabed the meta lock. We | |
313 | * might now be discovering a truncate that hit on another node. | |
314 | * block_read_full_page->get_block freaks out if it is asked to read | |
315 | * beyond the end of a file, so we check here. Callers | |
54cb8821 | 316 | * (generic_file_read, vm_ops->fault) are clever enough to check i_size |
ccd979bd MF |
317 | * and notice that the page they just read isn't needed. |
318 | * | |
319 | * XXX sys_readahead() seems to get that wrong? | |
320 | */ | |
321 | if (start >= i_size_read(inode)) { | |
eebd2aa3 | 322 | zero_user(page, 0, PAGE_SIZE); |
ccd979bd MF |
323 | SetPageUptodate(page); |
324 | ret = 0; | |
325 | goto out_alloc; | |
326 | } | |
327 | ||
6798d35a MF |
328 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
329 | ret = ocfs2_readpage_inline(inode, page); | |
330 | else | |
331 | ret = block_read_full_page(page, ocfs2_get_block); | |
ccd979bd MF |
332 | unlock = 0; |
333 | ||
ccd979bd MF |
334 | out_alloc: |
335 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 MF |
336 | out_inode_unlock: |
337 | ocfs2_inode_unlock(inode, 0); | |
ccd979bd MF |
338 | out: |
339 | if (unlock) | |
340 | unlock_page(page); | |
ccd979bd MF |
341 | return ret; |
342 | } | |
343 | ||
628a24f5 MF |
344 | /* |
345 | * This is used only for read-ahead. Failures or difficult to handle | |
346 | * situations are safe to ignore. | |
347 | * | |
348 | * Right now, we don't bother with BH_Boundary - in-inode extent lists | |
349 | * are quite large (243 extents on 4k blocks), so most inodes don't | |
350 | * grow out to a tree. If need be, detecting boundary extents could | |
351 | * trivially be added in a future version of ocfs2_get_block(). | |
352 | */ | |
353 | static int ocfs2_readpages(struct file *filp, struct address_space *mapping, | |
354 | struct list_head *pages, unsigned nr_pages) | |
355 | { | |
356 | int ret, err = -EIO; | |
357 | struct inode *inode = mapping->host; | |
358 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
359 | loff_t start; | |
360 | struct page *last; | |
361 | ||
362 | /* | |
363 | * Use the nonblocking flag for the dlm code to avoid page | |
364 | * lock inversion, but don't bother with retrying. | |
365 | */ | |
366 | ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK); | |
367 | if (ret) | |
368 | return err; | |
369 | ||
370 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { | |
371 | ocfs2_inode_unlock(inode, 0); | |
372 | return err; | |
373 | } | |
374 | ||
375 | /* | |
376 | * Don't bother with inline-data. There isn't anything | |
377 | * to read-ahead in that case anyway... | |
378 | */ | |
379 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
380 | goto out_unlock; | |
381 | ||
382 | /* | |
383 | * Check whether a remote node truncated this file - we just | |
384 | * drop out in that case as it's not worth handling here. | |
385 | */ | |
386 | last = list_entry(pages->prev, struct page, lru); | |
387 | start = (loff_t)last->index << PAGE_CACHE_SHIFT; | |
388 | if (start >= i_size_read(inode)) | |
389 | goto out_unlock; | |
390 | ||
391 | err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block); | |
392 | ||
393 | out_unlock: | |
394 | up_read(&oi->ip_alloc_sem); | |
395 | ocfs2_inode_unlock(inode, 0); | |
396 | ||
397 | return err; | |
398 | } | |
399 | ||
ccd979bd MF |
400 | /* Note: Because we don't support holes, our allocation has |
401 | * already happened (allocation writes zeros to the file data) | |
402 | * so we don't have to worry about ordered writes in | |
403 | * ocfs2_writepage. | |
404 | * | |
405 | * ->writepage is called during the process of invalidating the page cache | |
406 | * during blocked lock processing. It can't block on any cluster locks | |
407 | * to during block mapping. It's relying on the fact that the block | |
408 | * mapping can't have disappeared under the dirty pages that it is | |
409 | * being asked to write back. | |
410 | */ | |
411 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) | |
412 | { | |
9558156b TM |
413 | trace_ocfs2_writepage( |
414 | (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno, | |
415 | page->index); | |
ccd979bd | 416 | |
9558156b | 417 | return block_write_full_page(page, ocfs2_get_block, wbc); |
ccd979bd MF |
418 | } |
419 | ||
ccd979bd MF |
420 | /* Taken from ext3. We don't necessarily need the full blown |
421 | * functionality yet, but IMHO it's better to cut and paste the whole | |
422 | * thing so we can avoid introducing our own bugs (and easily pick up | |
423 | * their fixes when they happen) --Mark */ | |
60b11392 MF |
424 | int walk_page_buffers( handle_t *handle, |
425 | struct buffer_head *head, | |
426 | unsigned from, | |
427 | unsigned to, | |
428 | int *partial, | |
429 | int (*fn)( handle_t *handle, | |
430 | struct buffer_head *bh)) | |
ccd979bd MF |
431 | { |
432 | struct buffer_head *bh; | |
433 | unsigned block_start, block_end; | |
434 | unsigned blocksize = head->b_size; | |
435 | int err, ret = 0; | |
436 | struct buffer_head *next; | |
437 | ||
438 | for ( bh = head, block_start = 0; | |
439 | ret == 0 && (bh != head || !block_start); | |
440 | block_start = block_end, bh = next) | |
441 | { | |
442 | next = bh->b_this_page; | |
443 | block_end = block_start + blocksize; | |
444 | if (block_end <= from || block_start >= to) { | |
445 | if (partial && !buffer_uptodate(bh)) | |
446 | *partial = 1; | |
447 | continue; | |
448 | } | |
449 | err = (*fn)(handle, bh); | |
450 | if (!ret) | |
451 | ret = err; | |
452 | } | |
453 | return ret; | |
454 | } | |
455 | ||
ccd979bd MF |
456 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) |
457 | { | |
458 | sector_t status; | |
459 | u64 p_blkno = 0; | |
460 | int err = 0; | |
461 | struct inode *inode = mapping->host; | |
462 | ||
9558156b TM |
463 | trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno, |
464 | (unsigned long long)block); | |
ccd979bd MF |
465 | |
466 | /* We don't need to lock journal system files, since they aren't | |
467 | * accessed concurrently from multiple nodes. | |
468 | */ | |
469 | if (!INODE_JOURNAL(inode)) { | |
e63aecb6 | 470 | err = ocfs2_inode_lock(inode, NULL, 0); |
ccd979bd MF |
471 | if (err) { |
472 | if (err != -ENOENT) | |
473 | mlog_errno(err); | |
474 | goto bail; | |
475 | } | |
476 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | |
477 | } | |
478 | ||
6798d35a MF |
479 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) |
480 | err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, | |
481 | NULL); | |
ccd979bd MF |
482 | |
483 | if (!INODE_JOURNAL(inode)) { | |
484 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 | 485 | ocfs2_inode_unlock(inode, 0); |
ccd979bd MF |
486 | } |
487 | ||
488 | if (err) { | |
489 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", | |
490 | (unsigned long long)block); | |
491 | mlog_errno(err); | |
492 | goto bail; | |
493 | } | |
494 | ||
ccd979bd MF |
495 | bail: |
496 | status = err ? 0 : p_blkno; | |
497 | ||
ccd979bd MF |
498 | return status; |
499 | } | |
500 | ||
501 | /* | |
502 | * TODO: Make this into a generic get_blocks function. | |
503 | * | |
504 | * From do_direct_io in direct-io.c: | |
505 | * "So what we do is to permit the ->get_blocks function to populate | |
506 | * bh.b_size with the size of IO which is permitted at this offset and | |
507 | * this i_blkbits." | |
508 | * | |
509 | * This function is called directly from get_more_blocks in direct-io.c. | |
510 | * | |
511 | * called like this: dio->get_blocks(dio->inode, fs_startblk, | |
512 | * fs_count, map_bh, dio->rw == WRITE); | |
513 | */ | |
514 | static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock, | |
ccd979bd MF |
515 | struct buffer_head *bh_result, int create) |
516 | { | |
517 | int ret; | |
49255dce JQ |
518 | u32 cpos = 0; |
519 | int alloc_locked = 0; | |
4f902c37 | 520 | u64 p_blkno, inode_blocks, contig_blocks; |
49cb8d2d | 521 | unsigned int ext_flags; |
184d7d20 | 522 | unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits; |
1d8fa7a2 | 523 | unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits; |
49255dce JQ |
524 | unsigned long len = bh_result->b_size; |
525 | unsigned int clusters_to_alloc = 0; | |
526 | ||
527 | cpos = ocfs2_blocks_to_clusters(inode->i_sb, iblock); | |
ccd979bd | 528 | |
ccd979bd MF |
529 | /* This function won't even be called if the request isn't all |
530 | * nicely aligned and of the right size, so there's no need | |
531 | * for us to check any of that. */ | |
532 | ||
25baf2da | 533 | inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
564f8a32 | 534 | |
ccd979bd MF |
535 | /* This figures out the size of the next contiguous block, and |
536 | * our logical offset */ | |
363041a5 | 537 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, |
49cb8d2d | 538 | &contig_blocks, &ext_flags); |
ccd979bd MF |
539 | if (ret) { |
540 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", | |
541 | (unsigned long long)iblock); | |
542 | ret = -EIO; | |
543 | goto bail; | |
544 | } | |
545 | ||
cbaee472 TM |
546 | /* We should already CoW the refcounted extent in case of create. */ |
547 | BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED)); | |
548 | ||
49255dce JQ |
549 | /* allocate blocks if no p_blkno is found, and create == 1 */ |
550 | if (!p_blkno && create) { | |
551 | ret = ocfs2_inode_lock(inode, NULL, 1); | |
552 | if (ret < 0) { | |
553 | mlog_errno(ret); | |
554 | goto bail; | |
555 | } | |
556 | ||
557 | alloc_locked = 1; | |
558 | ||
559 | /* fill hole, allocate blocks can't be larger than the size | |
560 | * of the hole */ | |
561 | clusters_to_alloc = ocfs2_clusters_for_bytes(inode->i_sb, len); | |
562 | if (clusters_to_alloc > contig_blocks) | |
563 | clusters_to_alloc = contig_blocks; | |
564 | ||
565 | /* allocate extent and insert them into the extent tree */ | |
566 | ret = ocfs2_extend_allocation(inode, cpos, | |
567 | clusters_to_alloc, 0); | |
568 | if (ret < 0) { | |
569 | mlog_errno(ret); | |
570 | goto bail; | |
571 | } | |
572 | ||
573 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, | |
574 | &contig_blocks, &ext_flags); | |
575 | if (ret < 0) { | |
576 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", | |
577 | (unsigned long long)iblock); | |
578 | ret = -EIO; | |
579 | goto bail; | |
580 | } | |
581 | } | |
582 | ||
25baf2da MF |
583 | /* |
584 | * get_more_blocks() expects us to describe a hole by clearing | |
585 | * the mapped bit on bh_result(). | |
49cb8d2d MF |
586 | * |
587 | * Consider an unwritten extent as a hole. | |
25baf2da | 588 | */ |
49cb8d2d | 589 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) |
25baf2da | 590 | map_bh(bh_result, inode->i_sb, p_blkno); |
5fe878ae | 591 | else |
25baf2da | 592 | clear_buffer_mapped(bh_result); |
ccd979bd MF |
593 | |
594 | /* make sure we don't map more than max_blocks blocks here as | |
595 | that's all the kernel will handle at this point. */ | |
596 | if (max_blocks < contig_blocks) | |
597 | contig_blocks = max_blocks; | |
598 | bh_result->b_size = contig_blocks << blocksize_bits; | |
599 | bail: | |
49255dce JQ |
600 | if (alloc_locked) |
601 | ocfs2_inode_unlock(inode, 1); | |
ccd979bd MF |
602 | return ret; |
603 | } | |
604 | ||
2bd63216 | 605 | /* |
ccd979bd | 606 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're |
bd5fe6c5 CH |
607 | * particularly interested in the aio/dio case. We use the rw_lock DLM lock |
608 | * to protect io on one node from truncation on another. | |
ccd979bd MF |
609 | */ |
610 | static void ocfs2_dio_end_io(struct kiocb *iocb, | |
611 | loff_t offset, | |
612 | ssize_t bytes, | |
7b7a8665 | 613 | void *private) |
ccd979bd | 614 | { |
496ad9aa | 615 | struct inode *inode = file_inode(iocb->ki_filp); |
7cdfc3a1 | 616 | int level; |
ccd979bd MF |
617 | |
618 | /* this io's submitter should not have unlocked this before we could */ | |
619 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); | |
7cdfc3a1 | 620 | |
df2d6f26 | 621 | if (ocfs2_iocb_is_sem_locked(iocb)) |
39c99f12 | 622 | ocfs2_iocb_clear_sem_locked(iocb); |
39c99f12 | 623 | |
a11f7e63 MF |
624 | if (ocfs2_iocb_is_unaligned_aio(iocb)) { |
625 | ocfs2_iocb_clear_unaligned_aio(iocb); | |
626 | ||
c18ceab0 | 627 | mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio); |
a11f7e63 MF |
628 | } |
629 | ||
ccd979bd | 630 | ocfs2_iocb_clear_rw_locked(iocb); |
7cdfc3a1 MF |
631 | |
632 | level = ocfs2_iocb_rw_locked_level(iocb); | |
7cdfc3a1 | 633 | ocfs2_rw_unlock(inode, level); |
ccd979bd MF |
634 | } |
635 | ||
03f981cf JB |
636 | static int ocfs2_releasepage(struct page *page, gfp_t wait) |
637 | { | |
03f981cf JB |
638 | if (!page_has_buffers(page)) |
639 | return 0; | |
41ecc345 | 640 | return try_to_free_buffers(page); |
03f981cf JB |
641 | } |
642 | ||
24c40b32 JQ |
643 | static int ocfs2_is_overwrite(struct ocfs2_super *osb, |
644 | struct inode *inode, loff_t offset) | |
645 | { | |
646 | int ret = 0; | |
647 | u32 v_cpos = 0; | |
648 | u32 p_cpos = 0; | |
649 | unsigned int num_clusters = 0; | |
650 | unsigned int ext_flags = 0; | |
651 | ||
652 | v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset); | |
653 | ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, | |
654 | &num_clusters, &ext_flags); | |
655 | if (ret < 0) { | |
656 | mlog_errno(ret); | |
657 | return ret; | |
658 | } | |
659 | ||
660 | if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
661 | return 1; | |
662 | ||
663 | return 0; | |
664 | } | |
665 | ||
666 | static ssize_t ocfs2_direct_IO_write(struct kiocb *iocb, | |
667 | struct iov_iter *iter, | |
668 | loff_t offset) | |
669 | { | |
670 | ssize_t ret = 0; | |
671 | ssize_t written = 0; | |
672 | bool orphaned = false; | |
673 | int is_overwrite = 0; | |
674 | struct file *file = iocb->ki_filp; | |
675 | struct inode *inode = file_inode(file)->i_mapping->host; | |
676 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
677 | struct buffer_head *di_bh = NULL; | |
678 | size_t count = iter->count; | |
679 | journal_t *journal = osb->journal->j_journal; | |
680 | u32 zero_len; | |
681 | int cluster_align; | |
682 | loff_t final_size = offset + count; | |
683 | int append_write = offset >= i_size_read(inode) ? 1 : 0; | |
684 | unsigned int num_clusters = 0; | |
685 | unsigned int ext_flags = 0; | |
686 | ||
687 | { | |
688 | u64 o = offset; | |
689 | ||
690 | zero_len = do_div(o, 1 << osb->s_clustersize_bits); | |
691 | cluster_align = !zero_len; | |
692 | } | |
693 | ||
694 | /* | |
695 | * when final_size > inode->i_size, inode->i_size will be | |
696 | * updated after direct write, so add the inode to orphan | |
697 | * dir first. | |
698 | */ | |
699 | if (final_size > i_size_read(inode)) { | |
700 | ret = ocfs2_add_inode_to_orphan(osb, inode); | |
701 | if (ret < 0) { | |
702 | mlog_errno(ret); | |
703 | goto out; | |
704 | } | |
705 | orphaned = true; | |
706 | } | |
707 | ||
708 | if (append_write) { | |
709 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
710 | if (ret < 0) { | |
711 | mlog_errno(ret); | |
712 | goto clean_orphan; | |
713 | } | |
714 | ||
715 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
716 | ret = ocfs2_zero_extend(inode, di_bh, offset); | |
717 | else | |
718 | ret = ocfs2_extend_no_holes(inode, di_bh, offset, | |
719 | offset); | |
720 | if (ret < 0) { | |
721 | mlog_errno(ret); | |
722 | ocfs2_inode_unlock(inode, 1); | |
723 | brelse(di_bh); | |
724 | goto clean_orphan; | |
725 | } | |
726 | ||
727 | is_overwrite = ocfs2_is_overwrite(osb, inode, offset); | |
728 | if (is_overwrite < 0) { | |
729 | mlog_errno(is_overwrite); | |
730 | ocfs2_inode_unlock(inode, 1); | |
731 | brelse(di_bh); | |
732 | goto clean_orphan; | |
733 | } | |
734 | ||
735 | ocfs2_inode_unlock(inode, 1); | |
736 | brelse(di_bh); | |
737 | di_bh = NULL; | |
738 | } | |
739 | ||
740 | written = __blockdev_direct_IO(WRITE, iocb, inode, inode->i_sb->s_bdev, | |
741 | iter, offset, | |
742 | ocfs2_direct_IO_get_blocks, | |
743 | ocfs2_dio_end_io, NULL, 0); | |
744 | if (unlikely(written < 0)) { | |
745 | loff_t i_size = i_size_read(inode); | |
746 | ||
747 | if (offset + count > i_size) { | |
748 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
749 | if (ret < 0) { | |
750 | mlog_errno(ret); | |
751 | goto clean_orphan; | |
752 | } | |
753 | ||
754 | if (i_size == i_size_read(inode)) { | |
755 | ret = ocfs2_truncate_file(inode, di_bh, | |
756 | i_size); | |
757 | if (ret < 0) { | |
758 | if (ret != -ENOSPC) | |
759 | mlog_errno(ret); | |
760 | ||
761 | ocfs2_inode_unlock(inode, 1); | |
762 | brelse(di_bh); | |
763 | goto clean_orphan; | |
764 | } | |
765 | } | |
766 | ||
767 | ocfs2_inode_unlock(inode, 1); | |
768 | brelse(di_bh); | |
769 | ||
770 | ret = jbd2_journal_force_commit(journal); | |
771 | if (ret < 0) | |
772 | mlog_errno(ret); | |
773 | } | |
774 | } else if (written < 0 && append_write && !is_overwrite && | |
775 | !cluster_align) { | |
776 | u32 p_cpos = 0; | |
777 | u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset); | |
778 | ||
779 | ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, | |
780 | &num_clusters, &ext_flags); | |
781 | if (ret < 0) { | |
782 | mlog_errno(ret); | |
783 | goto clean_orphan; | |
784 | } | |
785 | ||
786 | BUG_ON(!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN)); | |
787 | ||
788 | ret = blkdev_issue_zeroout(osb->sb->s_bdev, | |
789 | p_cpos << (osb->s_clustersize_bits - 9), | |
790 | zero_len >> 9, GFP_KERNEL, false); | |
791 | if (ret < 0) | |
792 | mlog_errno(ret); | |
793 | } | |
794 | ||
795 | clean_orphan: | |
796 | if (orphaned) { | |
797 | int tmp_ret; | |
798 | int update_isize = written > 0 ? 1 : 0; | |
799 | loff_t end = update_isize ? offset + written : 0; | |
800 | ||
801 | tmp_ret = ocfs2_del_inode_from_orphan(osb, inode, | |
802 | update_isize, end); | |
803 | if (tmp_ret < 0) { | |
804 | ret = tmp_ret; | |
805 | goto out; | |
806 | } | |
807 | ||
808 | tmp_ret = jbd2_journal_force_commit(journal); | |
809 | if (tmp_ret < 0) { | |
810 | ret = tmp_ret; | |
811 | mlog_errno(tmp_ret); | |
812 | } | |
813 | } | |
814 | ||
815 | out: | |
816 | if (ret >= 0) | |
817 | ret = written; | |
818 | return ret; | |
819 | } | |
820 | ||
ccd979bd MF |
821 | static ssize_t ocfs2_direct_IO(int rw, |
822 | struct kiocb *iocb, | |
d8d3d94b AV |
823 | struct iov_iter *iter, |
824 | loff_t offset) | |
ccd979bd MF |
825 | { |
826 | struct file *file = iocb->ki_filp; | |
496ad9aa | 827 | struct inode *inode = file_inode(file)->i_mapping->host; |
24c40b32 JQ |
828 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
829 | int full_coherency = !(osb->s_mount_opt & | |
830 | OCFS2_MOUNT_COHERENCY_BUFFERED); | |
53013cba | 831 | |
6798d35a MF |
832 | /* |
833 | * Fallback to buffered I/O if we see an inode without | |
834 | * extents. | |
835 | */ | |
836 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
837 | return 0; | |
838 | ||
24c40b32 JQ |
839 | /* Fallback to buffered I/O if we are appending and |
840 | * concurrent O_DIRECT writes are allowed. | |
841 | */ | |
842 | if (i_size_read(inode) <= offset && !full_coherency) | |
b80474b4 TM |
843 | return 0; |
844 | ||
24c40b32 JQ |
845 | if (rw == READ) |
846 | return __blockdev_direct_IO(rw, iocb, inode, | |
847 | inode->i_sb->s_bdev, | |
31b14039 | 848 | iter, offset, |
c1e8d35e TM |
849 | ocfs2_direct_IO_get_blocks, |
850 | ocfs2_dio_end_io, NULL, 0); | |
24c40b32 JQ |
851 | else |
852 | return ocfs2_direct_IO_write(iocb, iter, offset); | |
ccd979bd MF |
853 | } |
854 | ||
9517bac6 MF |
855 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, |
856 | u32 cpos, | |
857 | unsigned int *start, | |
858 | unsigned int *end) | |
859 | { | |
860 | unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE; | |
861 | ||
862 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) { | |
863 | unsigned int cpp; | |
864 | ||
865 | cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits); | |
866 | ||
867 | cluster_start = cpos % cpp; | |
868 | cluster_start = cluster_start << osb->s_clustersize_bits; | |
869 | ||
870 | cluster_end = cluster_start + osb->s_clustersize; | |
871 | } | |
872 | ||
873 | BUG_ON(cluster_start > PAGE_SIZE); | |
874 | BUG_ON(cluster_end > PAGE_SIZE); | |
875 | ||
876 | if (start) | |
877 | *start = cluster_start; | |
878 | if (end) | |
879 | *end = cluster_end; | |
880 | } | |
881 | ||
882 | /* | |
883 | * 'from' and 'to' are the region in the page to avoid zeroing. | |
884 | * | |
885 | * If pagesize > clustersize, this function will avoid zeroing outside | |
886 | * of the cluster boundary. | |
887 | * | |
888 | * from == to == 0 is code for "zero the entire cluster region" | |
889 | */ | |
890 | static void ocfs2_clear_page_regions(struct page *page, | |
891 | struct ocfs2_super *osb, u32 cpos, | |
892 | unsigned from, unsigned to) | |
893 | { | |
894 | void *kaddr; | |
895 | unsigned int cluster_start, cluster_end; | |
896 | ||
897 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); | |
898 | ||
c4bc8dcb | 899 | kaddr = kmap_atomic(page); |
9517bac6 MF |
900 | |
901 | if (from || to) { | |
902 | if (from > cluster_start) | |
903 | memset(kaddr + cluster_start, 0, from - cluster_start); | |
904 | if (to < cluster_end) | |
905 | memset(kaddr + to, 0, cluster_end - to); | |
906 | } else { | |
907 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); | |
908 | } | |
909 | ||
c4bc8dcb | 910 | kunmap_atomic(kaddr); |
9517bac6 MF |
911 | } |
912 | ||
4e9563fd MF |
913 | /* |
914 | * Nonsparse file systems fully allocate before we get to the write | |
915 | * code. This prevents ocfs2_write() from tagging the write as an | |
916 | * allocating one, which means ocfs2_map_page_blocks() might try to | |
917 | * read-in the blocks at the tail of our file. Avoid reading them by | |
918 | * testing i_size against each block offset. | |
919 | */ | |
920 | static int ocfs2_should_read_blk(struct inode *inode, struct page *page, | |
921 | unsigned int block_start) | |
922 | { | |
923 | u64 offset = page_offset(page) + block_start; | |
924 | ||
925 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
926 | return 1; | |
927 | ||
928 | if (i_size_read(inode) > offset) | |
929 | return 1; | |
930 | ||
931 | return 0; | |
932 | } | |
933 | ||
9517bac6 | 934 | /* |
ebdec241 | 935 | * Some of this taken from __block_write_begin(). We already have our |
9517bac6 MF |
936 | * mapping by now though, and the entire write will be allocating or |
937 | * it won't, so not much need to use BH_New. | |
938 | * | |
939 | * This will also skip zeroing, which is handled externally. | |
940 | */ | |
60b11392 MF |
941 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, |
942 | struct inode *inode, unsigned int from, | |
943 | unsigned int to, int new) | |
9517bac6 MF |
944 | { |
945 | int ret = 0; | |
946 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; | |
947 | unsigned int block_end, block_start; | |
948 | unsigned int bsize = 1 << inode->i_blkbits; | |
949 | ||
950 | if (!page_has_buffers(page)) | |
951 | create_empty_buffers(page, bsize, 0); | |
952 | ||
953 | head = page_buffers(page); | |
954 | for (bh = head, block_start = 0; bh != head || !block_start; | |
955 | bh = bh->b_this_page, block_start += bsize) { | |
956 | block_end = block_start + bsize; | |
957 | ||
3a307ffc MF |
958 | clear_buffer_new(bh); |
959 | ||
9517bac6 MF |
960 | /* |
961 | * Ignore blocks outside of our i/o range - | |
962 | * they may belong to unallocated clusters. | |
963 | */ | |
60b11392 | 964 | if (block_start >= to || block_end <= from) { |
9517bac6 MF |
965 | if (PageUptodate(page)) |
966 | set_buffer_uptodate(bh); | |
967 | continue; | |
968 | } | |
969 | ||
970 | /* | |
971 | * For an allocating write with cluster size >= page | |
972 | * size, we always write the entire page. | |
973 | */ | |
3a307ffc MF |
974 | if (new) |
975 | set_buffer_new(bh); | |
9517bac6 MF |
976 | |
977 | if (!buffer_mapped(bh)) { | |
978 | map_bh(bh, inode->i_sb, *p_blkno); | |
979 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
980 | } | |
981 | ||
982 | if (PageUptodate(page)) { | |
983 | if (!buffer_uptodate(bh)) | |
984 | set_buffer_uptodate(bh); | |
985 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
bce99768 | 986 | !buffer_new(bh) && |
4e9563fd | 987 | ocfs2_should_read_blk(inode, page, block_start) && |
bce99768 | 988 | (block_start < from || block_end > to)) { |
9517bac6 MF |
989 | ll_rw_block(READ, 1, &bh); |
990 | *wait_bh++=bh; | |
991 | } | |
992 | ||
993 | *p_blkno = *p_blkno + 1; | |
994 | } | |
995 | ||
996 | /* | |
997 | * If we issued read requests - let them complete. | |
998 | */ | |
999 | while(wait_bh > wait) { | |
1000 | wait_on_buffer(*--wait_bh); | |
1001 | if (!buffer_uptodate(*wait_bh)) | |
1002 | ret = -EIO; | |
1003 | } | |
1004 | ||
1005 | if (ret == 0 || !new) | |
1006 | return ret; | |
1007 | ||
1008 | /* | |
1009 | * If we get -EIO above, zero out any newly allocated blocks | |
1010 | * to avoid exposing stale data. | |
1011 | */ | |
1012 | bh = head; | |
1013 | block_start = 0; | |
1014 | do { | |
9517bac6 MF |
1015 | block_end = block_start + bsize; |
1016 | if (block_end <= from) | |
1017 | goto next_bh; | |
1018 | if (block_start >= to) | |
1019 | break; | |
1020 | ||
eebd2aa3 | 1021 | zero_user(page, block_start, bh->b_size); |
9517bac6 MF |
1022 | set_buffer_uptodate(bh); |
1023 | mark_buffer_dirty(bh); | |
1024 | ||
1025 | next_bh: | |
1026 | block_start = block_end; | |
1027 | bh = bh->b_this_page; | |
1028 | } while (bh != head); | |
1029 | ||
1030 | return ret; | |
1031 | } | |
1032 | ||
3a307ffc MF |
1033 | #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE) |
1034 | #define OCFS2_MAX_CTXT_PAGES 1 | |
1035 | #else | |
1036 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE) | |
1037 | #endif | |
1038 | ||
1039 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE) | |
1040 | ||
6af67d82 | 1041 | /* |
3a307ffc | 1042 | * Describe the state of a single cluster to be written to. |
6af67d82 | 1043 | */ |
3a307ffc MF |
1044 | struct ocfs2_write_cluster_desc { |
1045 | u32 c_cpos; | |
1046 | u32 c_phys; | |
1047 | /* | |
1048 | * Give this a unique field because c_phys eventually gets | |
1049 | * filled. | |
1050 | */ | |
1051 | unsigned c_new; | |
b27b7cbc | 1052 | unsigned c_unwritten; |
e7432675 | 1053 | unsigned c_needs_zero; |
3a307ffc | 1054 | }; |
6af67d82 | 1055 | |
3a307ffc MF |
1056 | struct ocfs2_write_ctxt { |
1057 | /* Logical cluster position / len of write */ | |
1058 | u32 w_cpos; | |
1059 | u32 w_clen; | |
6af67d82 | 1060 | |
e7432675 SM |
1061 | /* First cluster allocated in a nonsparse extend */ |
1062 | u32 w_first_new_cpos; | |
1063 | ||
3a307ffc | 1064 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; |
6af67d82 | 1065 | |
3a307ffc MF |
1066 | /* |
1067 | * This is true if page_size > cluster_size. | |
1068 | * | |
1069 | * It triggers a set of special cases during write which might | |
1070 | * have to deal with allocating writes to partial pages. | |
1071 | */ | |
1072 | unsigned int w_large_pages; | |
6af67d82 | 1073 | |
3a307ffc MF |
1074 | /* |
1075 | * Pages involved in this write. | |
1076 | * | |
1077 | * w_target_page is the page being written to by the user. | |
1078 | * | |
1079 | * w_pages is an array of pages which always contains | |
1080 | * w_target_page, and in the case of an allocating write with | |
1081 | * page_size < cluster size, it will contain zero'd and mapped | |
1082 | * pages adjacent to w_target_page which need to be written | |
1083 | * out in so that future reads from that region will get | |
1084 | * zero's. | |
1085 | */ | |
3a307ffc | 1086 | unsigned int w_num_pages; |
83fd9c7f | 1087 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; |
3a307ffc | 1088 | struct page *w_target_page; |
eeb47d12 | 1089 | |
5cffff9e WW |
1090 | /* |
1091 | * w_target_locked is used for page_mkwrite path indicating no unlocking | |
1092 | * against w_target_page in ocfs2_write_end_nolock. | |
1093 | */ | |
1094 | unsigned int w_target_locked:1; | |
1095 | ||
3a307ffc MF |
1096 | /* |
1097 | * ocfs2_write_end() uses this to know what the real range to | |
1098 | * write in the target should be. | |
1099 | */ | |
1100 | unsigned int w_target_from; | |
1101 | unsigned int w_target_to; | |
1102 | ||
1103 | /* | |
1104 | * We could use journal_current_handle() but this is cleaner, | |
1105 | * IMHO -Mark | |
1106 | */ | |
1107 | handle_t *w_handle; | |
1108 | ||
1109 | struct buffer_head *w_di_bh; | |
b27b7cbc MF |
1110 | |
1111 | struct ocfs2_cached_dealloc_ctxt w_dealloc; | |
3a307ffc MF |
1112 | }; |
1113 | ||
1d410a6e | 1114 | void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages) |
3a307ffc MF |
1115 | { |
1116 | int i; | |
1117 | ||
1d410a6e MF |
1118 | for(i = 0; i < num_pages; i++) { |
1119 | if (pages[i]) { | |
1120 | unlock_page(pages[i]); | |
1121 | mark_page_accessed(pages[i]); | |
1122 | page_cache_release(pages[i]); | |
1123 | } | |
6af67d82 | 1124 | } |
1d410a6e MF |
1125 | } |
1126 | ||
136f49b9 | 1127 | static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc) |
1d410a6e | 1128 | { |
5cffff9e WW |
1129 | int i; |
1130 | ||
1131 | /* | |
1132 | * w_target_locked is only set to true in the page_mkwrite() case. | |
1133 | * The intent is to allow us to lock the target page from write_begin() | |
1134 | * to write_end(). The caller must hold a ref on w_target_page. | |
1135 | */ | |
1136 | if (wc->w_target_locked) { | |
1137 | BUG_ON(!wc->w_target_page); | |
1138 | for (i = 0; i < wc->w_num_pages; i++) { | |
1139 | if (wc->w_target_page == wc->w_pages[i]) { | |
1140 | wc->w_pages[i] = NULL; | |
1141 | break; | |
1142 | } | |
1143 | } | |
1144 | mark_page_accessed(wc->w_target_page); | |
1145 | page_cache_release(wc->w_target_page); | |
1146 | } | |
1d410a6e | 1147 | ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages); |
136f49b9 | 1148 | } |
6af67d82 | 1149 | |
136f49b9 JB |
1150 | static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc) |
1151 | { | |
1152 | ocfs2_unlock_pages(wc); | |
3a307ffc MF |
1153 | brelse(wc->w_di_bh); |
1154 | kfree(wc); | |
1155 | } | |
1156 | ||
1157 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, | |
1158 | struct ocfs2_super *osb, loff_t pos, | |
607d44aa | 1159 | unsigned len, struct buffer_head *di_bh) |
3a307ffc | 1160 | { |
30b8548f | 1161 | u32 cend; |
3a307ffc MF |
1162 | struct ocfs2_write_ctxt *wc; |
1163 | ||
1164 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); | |
1165 | if (!wc) | |
1166 | return -ENOMEM; | |
6af67d82 | 1167 | |
3a307ffc | 1168 | wc->w_cpos = pos >> osb->s_clustersize_bits; |
e7432675 | 1169 | wc->w_first_new_cpos = UINT_MAX; |
30b8548f | 1170 | cend = (pos + len - 1) >> osb->s_clustersize_bits; |
1171 | wc->w_clen = cend - wc->w_cpos + 1; | |
607d44aa MF |
1172 | get_bh(di_bh); |
1173 | wc->w_di_bh = di_bh; | |
6af67d82 | 1174 | |
3a307ffc MF |
1175 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) |
1176 | wc->w_large_pages = 1; | |
1177 | else | |
1178 | wc->w_large_pages = 0; | |
1179 | ||
b27b7cbc MF |
1180 | ocfs2_init_dealloc_ctxt(&wc->w_dealloc); |
1181 | ||
3a307ffc | 1182 | *wcp = wc; |
6af67d82 | 1183 | |
3a307ffc | 1184 | return 0; |
6af67d82 MF |
1185 | } |
1186 | ||
9517bac6 | 1187 | /* |
3a307ffc MF |
1188 | * If a page has any new buffers, zero them out here, and mark them uptodate |
1189 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1190 | * block data from leaking). And clear the new bit. | |
9517bac6 | 1191 | */ |
3a307ffc | 1192 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) |
9517bac6 | 1193 | { |
3a307ffc MF |
1194 | unsigned int block_start, block_end; |
1195 | struct buffer_head *head, *bh; | |
9517bac6 | 1196 | |
3a307ffc MF |
1197 | BUG_ON(!PageLocked(page)); |
1198 | if (!page_has_buffers(page)) | |
1199 | return; | |
9517bac6 | 1200 | |
3a307ffc MF |
1201 | bh = head = page_buffers(page); |
1202 | block_start = 0; | |
1203 | do { | |
1204 | block_end = block_start + bh->b_size; | |
1205 | ||
1206 | if (buffer_new(bh)) { | |
1207 | if (block_end > from && block_start < to) { | |
1208 | if (!PageUptodate(page)) { | |
1209 | unsigned start, end; | |
3a307ffc MF |
1210 | |
1211 | start = max(from, block_start); | |
1212 | end = min(to, block_end); | |
1213 | ||
eebd2aa3 | 1214 | zero_user_segment(page, start, end); |
3a307ffc MF |
1215 | set_buffer_uptodate(bh); |
1216 | } | |
1217 | ||
1218 | clear_buffer_new(bh); | |
1219 | mark_buffer_dirty(bh); | |
1220 | } | |
1221 | } | |
9517bac6 | 1222 | |
3a307ffc MF |
1223 | block_start = block_end; |
1224 | bh = bh->b_this_page; | |
1225 | } while (bh != head); | |
1226 | } | |
1227 | ||
1228 | /* | |
1229 | * Only called when we have a failure during allocating write to write | |
1230 | * zero's to the newly allocated region. | |
1231 | */ | |
1232 | static void ocfs2_write_failure(struct inode *inode, | |
1233 | struct ocfs2_write_ctxt *wc, | |
1234 | loff_t user_pos, unsigned user_len) | |
1235 | { | |
1236 | int i; | |
5c26a7b7 MF |
1237 | unsigned from = user_pos & (PAGE_CACHE_SIZE - 1), |
1238 | to = user_pos + user_len; | |
3a307ffc MF |
1239 | struct page *tmppage; |
1240 | ||
5c26a7b7 | 1241 | ocfs2_zero_new_buffers(wc->w_target_page, from, to); |
9517bac6 | 1242 | |
3a307ffc MF |
1243 | for(i = 0; i < wc->w_num_pages; i++) { |
1244 | tmppage = wc->w_pages[i]; | |
9517bac6 | 1245 | |
961cecbe | 1246 | if (page_has_buffers(tmppage)) { |
53ef99ca | 1247 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 1248 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
1249 | |
1250 | block_commit_write(tmppage, from, to); | |
1251 | } | |
9517bac6 | 1252 | } |
9517bac6 MF |
1253 | } |
1254 | ||
3a307ffc MF |
1255 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, |
1256 | struct ocfs2_write_ctxt *wc, | |
1257 | struct page *page, u32 cpos, | |
1258 | loff_t user_pos, unsigned user_len, | |
1259 | int new) | |
9517bac6 | 1260 | { |
3a307ffc MF |
1261 | int ret; |
1262 | unsigned int map_from = 0, map_to = 0; | |
9517bac6 | 1263 | unsigned int cluster_start, cluster_end; |
3a307ffc | 1264 | unsigned int user_data_from = 0, user_data_to = 0; |
9517bac6 | 1265 | |
3a307ffc | 1266 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, |
9517bac6 MF |
1267 | &cluster_start, &cluster_end); |
1268 | ||
272b62c1 GR |
1269 | /* treat the write as new if the a hole/lseek spanned across |
1270 | * the page boundary. | |
1271 | */ | |
1272 | new = new | ((i_size_read(inode) <= page_offset(page)) && | |
1273 | (page_offset(page) <= user_pos)); | |
1274 | ||
3a307ffc MF |
1275 | if (page == wc->w_target_page) { |
1276 | map_from = user_pos & (PAGE_CACHE_SIZE - 1); | |
1277 | map_to = map_from + user_len; | |
1278 | ||
1279 | if (new) | |
1280 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1281 | cluster_start, cluster_end, | |
1282 | new); | |
1283 | else | |
1284 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1285 | map_from, map_to, new); | |
1286 | if (ret) { | |
9517bac6 MF |
1287 | mlog_errno(ret); |
1288 | goto out; | |
1289 | } | |
1290 | ||
3a307ffc MF |
1291 | user_data_from = map_from; |
1292 | user_data_to = map_to; | |
9517bac6 | 1293 | if (new) { |
3a307ffc MF |
1294 | map_from = cluster_start; |
1295 | map_to = cluster_end; | |
9517bac6 MF |
1296 | } |
1297 | } else { | |
1298 | /* | |
1299 | * If we haven't allocated the new page yet, we | |
1300 | * shouldn't be writing it out without copying user | |
1301 | * data. This is likely a math error from the caller. | |
1302 | */ | |
1303 | BUG_ON(!new); | |
1304 | ||
3a307ffc MF |
1305 | map_from = cluster_start; |
1306 | map_to = cluster_end; | |
9517bac6 MF |
1307 | |
1308 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
3a307ffc | 1309 | cluster_start, cluster_end, new); |
9517bac6 MF |
1310 | if (ret) { |
1311 | mlog_errno(ret); | |
1312 | goto out; | |
1313 | } | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * Parts of newly allocated pages need to be zero'd. | |
1318 | * | |
1319 | * Above, we have also rewritten 'to' and 'from' - as far as | |
1320 | * the rest of the function is concerned, the entire cluster | |
1321 | * range inside of a page needs to be written. | |
1322 | * | |
1323 | * We can skip this if the page is up to date - it's already | |
1324 | * been zero'd from being read in as a hole. | |
1325 | */ | |
1326 | if (new && !PageUptodate(page)) | |
1327 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), | |
3a307ffc | 1328 | cpos, user_data_from, user_data_to); |
9517bac6 MF |
1329 | |
1330 | flush_dcache_page(page); | |
1331 | ||
9517bac6 | 1332 | out: |
3a307ffc | 1333 | return ret; |
9517bac6 MF |
1334 | } |
1335 | ||
1336 | /* | |
3a307ffc | 1337 | * This function will only grab one clusters worth of pages. |
9517bac6 | 1338 | */ |
3a307ffc MF |
1339 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, |
1340 | struct ocfs2_write_ctxt *wc, | |
693c241a JB |
1341 | u32 cpos, loff_t user_pos, |
1342 | unsigned user_len, int new, | |
7307de80 | 1343 | struct page *mmap_page) |
9517bac6 | 1344 | { |
3a307ffc | 1345 | int ret = 0, i; |
693c241a | 1346 | unsigned long start, target_index, end_index, index; |
9517bac6 | 1347 | struct inode *inode = mapping->host; |
693c241a | 1348 | loff_t last_byte; |
9517bac6 | 1349 | |
3a307ffc | 1350 | target_index = user_pos >> PAGE_CACHE_SHIFT; |
9517bac6 MF |
1351 | |
1352 | /* | |
1353 | * Figure out how many pages we'll be manipulating here. For | |
60b11392 | 1354 | * non allocating write, we just change the one |
693c241a JB |
1355 | * page. Otherwise, we'll need a whole clusters worth. If we're |
1356 | * writing past i_size, we only need enough pages to cover the | |
1357 | * last page of the write. | |
9517bac6 | 1358 | */ |
9517bac6 | 1359 | if (new) { |
3a307ffc MF |
1360 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); |
1361 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); | |
693c241a JB |
1362 | /* |
1363 | * We need the index *past* the last page we could possibly | |
1364 | * touch. This is the page past the end of the write or | |
1365 | * i_size, whichever is greater. | |
1366 | */ | |
1367 | last_byte = max(user_pos + user_len, i_size_read(inode)); | |
1368 | BUG_ON(last_byte < 1); | |
1369 | end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1; | |
1370 | if ((start + wc->w_num_pages) > end_index) | |
1371 | wc->w_num_pages = end_index - start; | |
9517bac6 | 1372 | } else { |
3a307ffc MF |
1373 | wc->w_num_pages = 1; |
1374 | start = target_index; | |
9517bac6 MF |
1375 | } |
1376 | ||
3a307ffc | 1377 | for(i = 0; i < wc->w_num_pages; i++) { |
9517bac6 MF |
1378 | index = start + i; |
1379 | ||
7307de80 MF |
1380 | if (index == target_index && mmap_page) { |
1381 | /* | |
1382 | * ocfs2_pagemkwrite() is a little different | |
1383 | * and wants us to directly use the page | |
1384 | * passed in. | |
1385 | */ | |
1386 | lock_page(mmap_page); | |
1387 | ||
5cffff9e | 1388 | /* Exit and let the caller retry */ |
7307de80 | 1389 | if (mmap_page->mapping != mapping) { |
5cffff9e | 1390 | WARN_ON(mmap_page->mapping); |
7307de80 | 1391 | unlock_page(mmap_page); |
5cffff9e | 1392 | ret = -EAGAIN; |
7307de80 MF |
1393 | goto out; |
1394 | } | |
1395 | ||
1396 | page_cache_get(mmap_page); | |
1397 | wc->w_pages[i] = mmap_page; | |
5cffff9e | 1398 | wc->w_target_locked = true; |
7307de80 MF |
1399 | } else { |
1400 | wc->w_pages[i] = find_or_create_page(mapping, index, | |
1401 | GFP_NOFS); | |
1402 | if (!wc->w_pages[i]) { | |
1403 | ret = -ENOMEM; | |
1404 | mlog_errno(ret); | |
1405 | goto out; | |
1406 | } | |
9517bac6 | 1407 | } |
1269529b | 1408 | wait_for_stable_page(wc->w_pages[i]); |
3a307ffc MF |
1409 | |
1410 | if (index == target_index) | |
1411 | wc->w_target_page = wc->w_pages[i]; | |
9517bac6 | 1412 | } |
3a307ffc | 1413 | out: |
5cffff9e WW |
1414 | if (ret) |
1415 | wc->w_target_locked = false; | |
3a307ffc MF |
1416 | return ret; |
1417 | } | |
1418 | ||
1419 | /* | |
1420 | * Prepare a single cluster for write one cluster into the file. | |
1421 | */ | |
1422 | static int ocfs2_write_cluster(struct address_space *mapping, | |
b27b7cbc | 1423 | u32 phys, unsigned int unwritten, |
e7432675 | 1424 | unsigned int should_zero, |
b27b7cbc | 1425 | struct ocfs2_alloc_context *data_ac, |
3a307ffc MF |
1426 | struct ocfs2_alloc_context *meta_ac, |
1427 | struct ocfs2_write_ctxt *wc, u32 cpos, | |
1428 | loff_t user_pos, unsigned user_len) | |
1429 | { | |
e7432675 | 1430 | int ret, i, new; |
3a307ffc MF |
1431 | u64 v_blkno, p_blkno; |
1432 | struct inode *inode = mapping->host; | |
f99b9b7c | 1433 | struct ocfs2_extent_tree et; |
3a307ffc MF |
1434 | |
1435 | new = phys == 0 ? 1 : 0; | |
9517bac6 | 1436 | if (new) { |
3a307ffc MF |
1437 | u32 tmp_pos; |
1438 | ||
9517bac6 MF |
1439 | /* |
1440 | * This is safe to call with the page locks - it won't take | |
1441 | * any additional semaphores or cluster locks. | |
1442 | */ | |
3a307ffc | 1443 | tmp_pos = cpos; |
0eb8d47e TM |
1444 | ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode, |
1445 | &tmp_pos, 1, 0, wc->w_di_bh, | |
1446 | wc->w_handle, data_ac, | |
1447 | meta_ac, NULL); | |
9517bac6 MF |
1448 | /* |
1449 | * This shouldn't happen because we must have already | |
1450 | * calculated the correct meta data allocation required. The | |
1451 | * internal tree allocation code should know how to increase | |
1452 | * transaction credits itself. | |
1453 | * | |
1454 | * If need be, we could handle -EAGAIN for a | |
1455 | * RESTART_TRANS here. | |
1456 | */ | |
1457 | mlog_bug_on_msg(ret == -EAGAIN, | |
1458 | "Inode %llu: EAGAIN return during allocation.\n", | |
1459 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1460 | if (ret < 0) { | |
1461 | mlog_errno(ret); | |
1462 | goto out; | |
1463 | } | |
b27b7cbc | 1464 | } else if (unwritten) { |
5e404e9e JB |
1465 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
1466 | wc->w_di_bh); | |
f99b9b7c | 1467 | ret = ocfs2_mark_extent_written(inode, &et, |
b27b7cbc | 1468 | wc->w_handle, cpos, 1, phys, |
f99b9b7c | 1469 | meta_ac, &wc->w_dealloc); |
b27b7cbc MF |
1470 | if (ret < 0) { |
1471 | mlog_errno(ret); | |
1472 | goto out; | |
1473 | } | |
1474 | } | |
3a307ffc | 1475 | |
b27b7cbc | 1476 | if (should_zero) |
3a307ffc | 1477 | v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos); |
b27b7cbc | 1478 | else |
3a307ffc | 1479 | v_blkno = user_pos >> inode->i_sb->s_blocksize_bits; |
9517bac6 | 1480 | |
3a307ffc MF |
1481 | /* |
1482 | * The only reason this should fail is due to an inability to | |
1483 | * find the extent added. | |
1484 | */ | |
49cb8d2d MF |
1485 | ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL, |
1486 | NULL); | |
9517bac6 | 1487 | if (ret < 0) { |
61fb9ea4 | 1488 | mlog(ML_ERROR, "Get physical blkno failed for inode %llu, " |
3a307ffc MF |
1489 | "at logical block %llu", |
1490 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1491 | (unsigned long long)v_blkno); | |
9517bac6 MF |
1492 | goto out; |
1493 | } | |
1494 | ||
1495 | BUG_ON(p_blkno == 0); | |
1496 | ||
3a307ffc MF |
1497 | for(i = 0; i < wc->w_num_pages; i++) { |
1498 | int tmpret; | |
9517bac6 | 1499 | |
3a307ffc MF |
1500 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, |
1501 | wc->w_pages[i], cpos, | |
b27b7cbc MF |
1502 | user_pos, user_len, |
1503 | should_zero); | |
3a307ffc MF |
1504 | if (tmpret) { |
1505 | mlog_errno(tmpret); | |
1506 | if (ret == 0) | |
cbfa9639 | 1507 | ret = tmpret; |
3a307ffc | 1508 | } |
9517bac6 MF |
1509 | } |
1510 | ||
3a307ffc MF |
1511 | /* |
1512 | * We only have cleanup to do in case of allocating write. | |
1513 | */ | |
1514 | if (ret && new) | |
1515 | ocfs2_write_failure(inode, wc, user_pos, user_len); | |
1516 | ||
9517bac6 | 1517 | out: |
9517bac6 | 1518 | |
3a307ffc | 1519 | return ret; |
9517bac6 MF |
1520 | } |
1521 | ||
0d172baa MF |
1522 | static int ocfs2_write_cluster_by_desc(struct address_space *mapping, |
1523 | struct ocfs2_alloc_context *data_ac, | |
1524 | struct ocfs2_alloc_context *meta_ac, | |
1525 | struct ocfs2_write_ctxt *wc, | |
1526 | loff_t pos, unsigned len) | |
1527 | { | |
1528 | int ret, i; | |
db56246c MF |
1529 | loff_t cluster_off; |
1530 | unsigned int local_len = len; | |
0d172baa | 1531 | struct ocfs2_write_cluster_desc *desc; |
db56246c | 1532 | struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb); |
0d172baa MF |
1533 | |
1534 | for (i = 0; i < wc->w_clen; i++) { | |
1535 | desc = &wc->w_desc[i]; | |
1536 | ||
db56246c MF |
1537 | /* |
1538 | * We have to make sure that the total write passed in | |
1539 | * doesn't extend past a single cluster. | |
1540 | */ | |
1541 | local_len = len; | |
1542 | cluster_off = pos & (osb->s_clustersize - 1); | |
1543 | if ((cluster_off + local_len) > osb->s_clustersize) | |
1544 | local_len = osb->s_clustersize - cluster_off; | |
1545 | ||
b27b7cbc | 1546 | ret = ocfs2_write_cluster(mapping, desc->c_phys, |
e7432675 SM |
1547 | desc->c_unwritten, |
1548 | desc->c_needs_zero, | |
1549 | data_ac, meta_ac, | |
db56246c | 1550 | wc, desc->c_cpos, pos, local_len); |
0d172baa MF |
1551 | if (ret) { |
1552 | mlog_errno(ret); | |
1553 | goto out; | |
1554 | } | |
db56246c MF |
1555 | |
1556 | len -= local_len; | |
1557 | pos += local_len; | |
0d172baa MF |
1558 | } |
1559 | ||
1560 | ret = 0; | |
1561 | out: | |
1562 | return ret; | |
1563 | } | |
1564 | ||
3a307ffc MF |
1565 | /* |
1566 | * ocfs2_write_end() wants to know which parts of the target page it | |
1567 | * should complete the write on. It's easiest to compute them ahead of | |
1568 | * time when a more complete view of the write is available. | |
1569 | */ | |
1570 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, | |
1571 | struct ocfs2_write_ctxt *wc, | |
1572 | loff_t pos, unsigned len, int alloc) | |
9517bac6 | 1573 | { |
3a307ffc | 1574 | struct ocfs2_write_cluster_desc *desc; |
9517bac6 | 1575 | |
3a307ffc MF |
1576 | wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1); |
1577 | wc->w_target_to = wc->w_target_from + len; | |
1578 | ||
1579 | if (alloc == 0) | |
1580 | return; | |
1581 | ||
1582 | /* | |
1583 | * Allocating write - we may have different boundaries based | |
1584 | * on page size and cluster size. | |
1585 | * | |
1586 | * NOTE: We can no longer compute one value from the other as | |
1587 | * the actual write length and user provided length may be | |
1588 | * different. | |
1589 | */ | |
9517bac6 | 1590 | |
3a307ffc MF |
1591 | if (wc->w_large_pages) { |
1592 | /* | |
1593 | * We only care about the 1st and last cluster within | |
b27b7cbc | 1594 | * our range and whether they should be zero'd or not. Either |
3a307ffc MF |
1595 | * value may be extended out to the start/end of a |
1596 | * newly allocated cluster. | |
1597 | */ | |
1598 | desc = &wc->w_desc[0]; | |
e7432675 | 1599 | if (desc->c_needs_zero) |
3a307ffc MF |
1600 | ocfs2_figure_cluster_boundaries(osb, |
1601 | desc->c_cpos, | |
1602 | &wc->w_target_from, | |
1603 | NULL); | |
1604 | ||
1605 | desc = &wc->w_desc[wc->w_clen - 1]; | |
e7432675 | 1606 | if (desc->c_needs_zero) |
3a307ffc MF |
1607 | ocfs2_figure_cluster_boundaries(osb, |
1608 | desc->c_cpos, | |
1609 | NULL, | |
1610 | &wc->w_target_to); | |
1611 | } else { | |
1612 | wc->w_target_from = 0; | |
1613 | wc->w_target_to = PAGE_CACHE_SIZE; | |
1614 | } | |
9517bac6 MF |
1615 | } |
1616 | ||
0d172baa MF |
1617 | /* |
1618 | * Populate each single-cluster write descriptor in the write context | |
1619 | * with information about the i/o to be done. | |
b27b7cbc MF |
1620 | * |
1621 | * Returns the number of clusters that will have to be allocated, as | |
1622 | * well as a worst case estimate of the number of extent records that | |
1623 | * would have to be created during a write to an unwritten region. | |
0d172baa MF |
1624 | */ |
1625 | static int ocfs2_populate_write_desc(struct inode *inode, | |
1626 | struct ocfs2_write_ctxt *wc, | |
b27b7cbc MF |
1627 | unsigned int *clusters_to_alloc, |
1628 | unsigned int *extents_to_split) | |
9517bac6 | 1629 | { |
0d172baa | 1630 | int ret; |
3a307ffc | 1631 | struct ocfs2_write_cluster_desc *desc; |
0d172baa | 1632 | unsigned int num_clusters = 0; |
b27b7cbc | 1633 | unsigned int ext_flags = 0; |
0d172baa MF |
1634 | u32 phys = 0; |
1635 | int i; | |
9517bac6 | 1636 | |
b27b7cbc MF |
1637 | *clusters_to_alloc = 0; |
1638 | *extents_to_split = 0; | |
1639 | ||
3a307ffc MF |
1640 | for (i = 0; i < wc->w_clen; i++) { |
1641 | desc = &wc->w_desc[i]; | |
1642 | desc->c_cpos = wc->w_cpos + i; | |
1643 | ||
1644 | if (num_clusters == 0) { | |
b27b7cbc MF |
1645 | /* |
1646 | * Need to look up the next extent record. | |
1647 | */ | |
3a307ffc | 1648 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, |
b27b7cbc | 1649 | &num_clusters, &ext_flags); |
3a307ffc MF |
1650 | if (ret) { |
1651 | mlog_errno(ret); | |
607d44aa | 1652 | goto out; |
3a307ffc | 1653 | } |
b27b7cbc | 1654 | |
293b2f70 TM |
1655 | /* We should already CoW the refcountd extent. */ |
1656 | BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED); | |
1657 | ||
b27b7cbc MF |
1658 | /* |
1659 | * Assume worst case - that we're writing in | |
1660 | * the middle of the extent. | |
1661 | * | |
1662 | * We can assume that the write proceeds from | |
1663 | * left to right, in which case the extent | |
1664 | * insert code is smart enough to coalesce the | |
1665 | * next splits into the previous records created. | |
1666 | */ | |
1667 | if (ext_flags & OCFS2_EXT_UNWRITTEN) | |
1668 | *extents_to_split = *extents_to_split + 2; | |
3a307ffc MF |
1669 | } else if (phys) { |
1670 | /* | |
1671 | * Only increment phys if it doesn't describe | |
1672 | * a hole. | |
1673 | */ | |
1674 | phys++; | |
1675 | } | |
1676 | ||
e7432675 SM |
1677 | /* |
1678 | * If w_first_new_cpos is < UINT_MAX, we have a non-sparse | |
1679 | * file that got extended. w_first_new_cpos tells us | |
1680 | * where the newly allocated clusters are so we can | |
1681 | * zero them. | |
1682 | */ | |
1683 | if (desc->c_cpos >= wc->w_first_new_cpos) { | |
1684 | BUG_ON(phys == 0); | |
1685 | desc->c_needs_zero = 1; | |
1686 | } | |
1687 | ||
3a307ffc MF |
1688 | desc->c_phys = phys; |
1689 | if (phys == 0) { | |
1690 | desc->c_new = 1; | |
e7432675 | 1691 | desc->c_needs_zero = 1; |
0d172baa | 1692 | *clusters_to_alloc = *clusters_to_alloc + 1; |
3a307ffc | 1693 | } |
e7432675 SM |
1694 | |
1695 | if (ext_flags & OCFS2_EXT_UNWRITTEN) { | |
b27b7cbc | 1696 | desc->c_unwritten = 1; |
e7432675 SM |
1697 | desc->c_needs_zero = 1; |
1698 | } | |
3a307ffc MF |
1699 | |
1700 | num_clusters--; | |
9517bac6 MF |
1701 | } |
1702 | ||
0d172baa MF |
1703 | ret = 0; |
1704 | out: | |
1705 | return ret; | |
1706 | } | |
1707 | ||
1afc32b9 MF |
1708 | static int ocfs2_write_begin_inline(struct address_space *mapping, |
1709 | struct inode *inode, | |
1710 | struct ocfs2_write_ctxt *wc) | |
1711 | { | |
1712 | int ret; | |
1713 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1714 | struct page *page; | |
1715 | handle_t *handle; | |
1716 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1717 | ||
f775da2f JB |
1718 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
1719 | if (IS_ERR(handle)) { | |
1720 | ret = PTR_ERR(handle); | |
1721 | mlog_errno(ret); | |
1722 | goto out; | |
1723 | } | |
1724 | ||
1afc32b9 MF |
1725 | page = find_or_create_page(mapping, 0, GFP_NOFS); |
1726 | if (!page) { | |
f775da2f | 1727 | ocfs2_commit_trans(osb, handle); |
1afc32b9 MF |
1728 | ret = -ENOMEM; |
1729 | mlog_errno(ret); | |
1730 | goto out; | |
1731 | } | |
1732 | /* | |
1733 | * If we don't set w_num_pages then this page won't get unlocked | |
1734 | * and freed on cleanup of the write context. | |
1735 | */ | |
1736 | wc->w_pages[0] = wc->w_target_page = page; | |
1737 | wc->w_num_pages = 1; | |
1738 | ||
0cf2f763 | 1739 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 1740 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
1741 | if (ret) { |
1742 | ocfs2_commit_trans(osb, handle); | |
1743 | ||
1744 | mlog_errno(ret); | |
1745 | goto out; | |
1746 | } | |
1747 | ||
1748 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
1749 | ocfs2_set_inode_data_inline(inode, di); | |
1750 | ||
1751 | if (!PageUptodate(page)) { | |
1752 | ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh); | |
1753 | if (ret) { | |
1754 | ocfs2_commit_trans(osb, handle); | |
1755 | ||
1756 | goto out; | |
1757 | } | |
1758 | } | |
1759 | ||
1760 | wc->w_handle = handle; | |
1761 | out: | |
1762 | return ret; | |
1763 | } | |
1764 | ||
1765 | int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size) | |
1766 | { | |
1767 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1768 | ||
0d8a4e0c | 1769 | if (new_size <= le16_to_cpu(di->id2.i_data.id_count)) |
1afc32b9 MF |
1770 | return 1; |
1771 | return 0; | |
1772 | } | |
1773 | ||
1774 | static int ocfs2_try_to_write_inline_data(struct address_space *mapping, | |
1775 | struct inode *inode, loff_t pos, | |
1776 | unsigned len, struct page *mmap_page, | |
1777 | struct ocfs2_write_ctxt *wc) | |
1778 | { | |
1779 | int ret, written = 0; | |
1780 | loff_t end = pos + len; | |
1781 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
d9ae49d6 | 1782 | struct ocfs2_dinode *di = NULL; |
1afc32b9 | 1783 | |
9558156b TM |
1784 | trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno, |
1785 | len, (unsigned long long)pos, | |
1786 | oi->ip_dyn_features); | |
1afc32b9 MF |
1787 | |
1788 | /* | |
1789 | * Handle inodes which already have inline data 1st. | |
1790 | */ | |
1791 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1792 | if (mmap_page == NULL && | |
1793 | ocfs2_size_fits_inline_data(wc->w_di_bh, end)) | |
1794 | goto do_inline_write; | |
1795 | ||
1796 | /* | |
1797 | * The write won't fit - we have to give this inode an | |
1798 | * inline extent list now. | |
1799 | */ | |
1800 | ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh); | |
1801 | if (ret) | |
1802 | mlog_errno(ret); | |
1803 | goto out; | |
1804 | } | |
1805 | ||
1806 | /* | |
1807 | * Check whether the inode can accept inline data. | |
1808 | */ | |
1809 | if (oi->ip_clusters != 0 || i_size_read(inode) != 0) | |
1810 | return 0; | |
1811 | ||
1812 | /* | |
1813 | * Check whether the write can fit. | |
1814 | */ | |
d9ae49d6 TY |
1815 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; |
1816 | if (mmap_page || | |
1817 | end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) | |
1afc32b9 MF |
1818 | return 0; |
1819 | ||
1820 | do_inline_write: | |
1821 | ret = ocfs2_write_begin_inline(mapping, inode, wc); | |
1822 | if (ret) { | |
1823 | mlog_errno(ret); | |
1824 | goto out; | |
1825 | } | |
1826 | ||
1827 | /* | |
1828 | * This signals to the caller that the data can be written | |
1829 | * inline. | |
1830 | */ | |
1831 | written = 1; | |
1832 | out: | |
1833 | return written ? written : ret; | |
1834 | } | |
1835 | ||
65ed39d6 MF |
1836 | /* |
1837 | * This function only does anything for file systems which can't | |
1838 | * handle sparse files. | |
1839 | * | |
1840 | * What we want to do here is fill in any hole between the current end | |
1841 | * of allocation and the end of our write. That way the rest of the | |
1842 | * write path can treat it as an non-allocating write, which has no | |
1843 | * special case code for sparse/nonsparse files. | |
1844 | */ | |
5693486b JB |
1845 | static int ocfs2_expand_nonsparse_inode(struct inode *inode, |
1846 | struct buffer_head *di_bh, | |
1847 | loff_t pos, unsigned len, | |
65ed39d6 MF |
1848 | struct ocfs2_write_ctxt *wc) |
1849 | { | |
1850 | int ret; | |
65ed39d6 MF |
1851 | loff_t newsize = pos + len; |
1852 | ||
5693486b | 1853 | BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); |
65ed39d6 MF |
1854 | |
1855 | if (newsize <= i_size_read(inode)) | |
1856 | return 0; | |
1857 | ||
5693486b | 1858 | ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos); |
65ed39d6 MF |
1859 | if (ret) |
1860 | mlog_errno(ret); | |
1861 | ||
e7432675 SM |
1862 | wc->w_first_new_cpos = |
1863 | ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)); | |
1864 | ||
65ed39d6 MF |
1865 | return ret; |
1866 | } | |
1867 | ||
5693486b JB |
1868 | static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh, |
1869 | loff_t pos) | |
1870 | { | |
1871 | int ret = 0; | |
1872 | ||
1873 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); | |
1874 | if (pos > i_size_read(inode)) | |
1875 | ret = ocfs2_zero_extend(inode, di_bh, pos); | |
1876 | ||
1877 | return ret; | |
1878 | } | |
1879 | ||
50308d81 TM |
1880 | /* |
1881 | * Try to flush truncate logs if we can free enough clusters from it. | |
1882 | * As for return value, "< 0" means error, "0" no space and "1" means | |
1883 | * we have freed enough spaces and let the caller try to allocate again. | |
1884 | */ | |
1885 | static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb, | |
1886 | unsigned int needed) | |
1887 | { | |
1888 | tid_t target; | |
1889 | int ret = 0; | |
1890 | unsigned int truncated_clusters; | |
1891 | ||
1892 | mutex_lock(&osb->osb_tl_inode->i_mutex); | |
1893 | truncated_clusters = osb->truncated_clusters; | |
1894 | mutex_unlock(&osb->osb_tl_inode->i_mutex); | |
1895 | ||
1896 | /* | |
1897 | * Check whether we can succeed in allocating if we free | |
1898 | * the truncate log. | |
1899 | */ | |
1900 | if (truncated_clusters < needed) | |
1901 | goto out; | |
1902 | ||
1903 | ret = ocfs2_flush_truncate_log(osb); | |
1904 | if (ret) { | |
1905 | mlog_errno(ret); | |
1906 | goto out; | |
1907 | } | |
1908 | ||
1909 | if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) { | |
1910 | jbd2_log_wait_commit(osb->journal->j_journal, target); | |
1911 | ret = 1; | |
1912 | } | |
1913 | out: | |
1914 | return ret; | |
1915 | } | |
1916 | ||
0378da0f TM |
1917 | int ocfs2_write_begin_nolock(struct file *filp, |
1918 | struct address_space *mapping, | |
0d172baa MF |
1919 | loff_t pos, unsigned len, unsigned flags, |
1920 | struct page **pagep, void **fsdata, | |
1921 | struct buffer_head *di_bh, struct page *mmap_page) | |
1922 | { | |
e7432675 | 1923 | int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS; |
50308d81 | 1924 | unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0; |
0d172baa MF |
1925 | struct ocfs2_write_ctxt *wc; |
1926 | struct inode *inode = mapping->host; | |
1927 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1928 | struct ocfs2_dinode *di; | |
1929 | struct ocfs2_alloc_context *data_ac = NULL; | |
1930 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1931 | handle_t *handle; | |
f99b9b7c | 1932 | struct ocfs2_extent_tree et; |
50308d81 | 1933 | int try_free = 1, ret1; |
0d172baa | 1934 | |
50308d81 | 1935 | try_again: |
0d172baa MF |
1936 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh); |
1937 | if (ret) { | |
1938 | mlog_errno(ret); | |
1939 | return ret; | |
1940 | } | |
1941 | ||
1afc32b9 MF |
1942 | if (ocfs2_supports_inline_data(osb)) { |
1943 | ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len, | |
1944 | mmap_page, wc); | |
1945 | if (ret == 1) { | |
1946 | ret = 0; | |
1947 | goto success; | |
1948 | } | |
1949 | if (ret < 0) { | |
1950 | mlog_errno(ret); | |
1951 | goto out; | |
1952 | } | |
1953 | } | |
1954 | ||
5693486b JB |
1955 | if (ocfs2_sparse_alloc(osb)) |
1956 | ret = ocfs2_zero_tail(inode, di_bh, pos); | |
1957 | else | |
1958 | ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len, | |
1959 | wc); | |
65ed39d6 MF |
1960 | if (ret) { |
1961 | mlog_errno(ret); | |
1962 | goto out; | |
1963 | } | |
1964 | ||
293b2f70 TM |
1965 | ret = ocfs2_check_range_for_refcount(inode, pos, len); |
1966 | if (ret < 0) { | |
1967 | mlog_errno(ret); | |
1968 | goto out; | |
1969 | } else if (ret == 1) { | |
50308d81 | 1970 | clusters_need = wc->w_clen; |
c7dd3392 | 1971 | ret = ocfs2_refcount_cow(inode, di_bh, |
37f8a2bf | 1972 | wc->w_cpos, wc->w_clen, UINT_MAX); |
293b2f70 TM |
1973 | if (ret) { |
1974 | mlog_errno(ret); | |
1975 | goto out; | |
1976 | } | |
1977 | } | |
1978 | ||
b27b7cbc MF |
1979 | ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc, |
1980 | &extents_to_split); | |
0d172baa MF |
1981 | if (ret) { |
1982 | mlog_errno(ret); | |
1983 | goto out; | |
1984 | } | |
50308d81 | 1985 | clusters_need += clusters_to_alloc; |
0d172baa MF |
1986 | |
1987 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1988 | ||
9558156b TM |
1989 | trace_ocfs2_write_begin_nolock( |
1990 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1991 | (long long)i_size_read(inode), | |
1992 | le32_to_cpu(di->i_clusters), | |
1993 | pos, len, flags, mmap_page, | |
1994 | clusters_to_alloc, extents_to_split); | |
1995 | ||
3a307ffc MF |
1996 | /* |
1997 | * We set w_target_from, w_target_to here so that | |
1998 | * ocfs2_write_end() knows which range in the target page to | |
1999 | * write out. An allocation requires that we write the entire | |
2000 | * cluster range. | |
2001 | */ | |
b27b7cbc | 2002 | if (clusters_to_alloc || extents_to_split) { |
3a307ffc MF |
2003 | /* |
2004 | * XXX: We are stretching the limits of | |
b27b7cbc | 2005 | * ocfs2_lock_allocators(). It greatly over-estimates |
3a307ffc MF |
2006 | * the work to be done. |
2007 | */ | |
5e404e9e JB |
2008 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
2009 | wc->w_di_bh); | |
f99b9b7c | 2010 | ret = ocfs2_lock_allocators(inode, &et, |
231b87d1 | 2011 | clusters_to_alloc, extents_to_split, |
f99b9b7c | 2012 | &data_ac, &meta_ac); |
9517bac6 MF |
2013 | if (ret) { |
2014 | mlog_errno(ret); | |
607d44aa | 2015 | goto out; |
9517bac6 MF |
2016 | } |
2017 | ||
4fe370af MF |
2018 | if (data_ac) |
2019 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; | |
2020 | ||
811f933d | 2021 | credits = ocfs2_calc_extend_credits(inode->i_sb, |
06f9da6e | 2022 | &di->id2.i_list); |
3a307ffc | 2023 | |
9517bac6 MF |
2024 | } |
2025 | ||
e7432675 SM |
2026 | /* |
2027 | * We have to zero sparse allocated clusters, unwritten extent clusters, | |
2028 | * and non-sparse clusters we just extended. For non-sparse writes, | |
2029 | * we know zeros will only be needed in the first and/or last cluster. | |
2030 | */ | |
2031 | if (clusters_to_alloc || extents_to_split || | |
8379e7c4 SM |
2032 | (wc->w_clen && (wc->w_desc[0].c_needs_zero || |
2033 | wc->w_desc[wc->w_clen - 1].c_needs_zero))) | |
e7432675 SM |
2034 | cluster_of_pages = 1; |
2035 | else | |
2036 | cluster_of_pages = 0; | |
2037 | ||
2038 | ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages); | |
3a307ffc | 2039 | |
9517bac6 MF |
2040 | handle = ocfs2_start_trans(osb, credits); |
2041 | if (IS_ERR(handle)) { | |
2042 | ret = PTR_ERR(handle); | |
2043 | mlog_errno(ret); | |
607d44aa | 2044 | goto out; |
9517bac6 MF |
2045 | } |
2046 | ||
3a307ffc MF |
2047 | wc->w_handle = handle; |
2048 | ||
5dd4056d CH |
2049 | if (clusters_to_alloc) { |
2050 | ret = dquot_alloc_space_nodirty(inode, | |
2051 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); | |
2052 | if (ret) | |
2053 | goto out_commit; | |
a90714c1 | 2054 | } |
3a307ffc MF |
2055 | /* |
2056 | * We don't want this to fail in ocfs2_write_end(), so do it | |
2057 | * here. | |
2058 | */ | |
0cf2f763 | 2059 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 2060 | OCFS2_JOURNAL_ACCESS_WRITE); |
3a307ffc | 2061 | if (ret) { |
9517bac6 | 2062 | mlog_errno(ret); |
a90714c1 | 2063 | goto out_quota; |
9517bac6 MF |
2064 | } |
2065 | ||
3a307ffc MF |
2066 | /* |
2067 | * Fill our page array first. That way we've grabbed enough so | |
2068 | * that we can zero and flush if we error after adding the | |
2069 | * extent. | |
2070 | */ | |
693c241a | 2071 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len, |
e7432675 | 2072 | cluster_of_pages, mmap_page); |
5cffff9e | 2073 | if (ret && ret != -EAGAIN) { |
9517bac6 | 2074 | mlog_errno(ret); |
a90714c1 | 2075 | goto out_quota; |
9517bac6 MF |
2076 | } |
2077 | ||
5cffff9e WW |
2078 | /* |
2079 | * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock | |
2080 | * the target page. In this case, we exit with no error and no target | |
2081 | * page. This will trigger the caller, page_mkwrite(), to re-try | |
2082 | * the operation. | |
2083 | */ | |
2084 | if (ret == -EAGAIN) { | |
2085 | BUG_ON(wc->w_target_page); | |
2086 | ret = 0; | |
2087 | goto out_quota; | |
2088 | } | |
2089 | ||
0d172baa MF |
2090 | ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos, |
2091 | len); | |
2092 | if (ret) { | |
2093 | mlog_errno(ret); | |
a90714c1 | 2094 | goto out_quota; |
9517bac6 | 2095 | } |
9517bac6 | 2096 | |
3a307ffc MF |
2097 | if (data_ac) |
2098 | ocfs2_free_alloc_context(data_ac); | |
2099 | if (meta_ac) | |
2100 | ocfs2_free_alloc_context(meta_ac); | |
9517bac6 | 2101 | |
1afc32b9 | 2102 | success: |
3a307ffc MF |
2103 | *pagep = wc->w_target_page; |
2104 | *fsdata = wc; | |
2105 | return 0; | |
a90714c1 JK |
2106 | out_quota: |
2107 | if (clusters_to_alloc) | |
5dd4056d | 2108 | dquot_free_space(inode, |
a90714c1 | 2109 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); |
9517bac6 MF |
2110 | out_commit: |
2111 | ocfs2_commit_trans(osb, handle); | |
2112 | ||
9517bac6 | 2113 | out: |
3a307ffc MF |
2114 | ocfs2_free_write_ctxt(wc); |
2115 | ||
b1214e47 | 2116 | if (data_ac) { |
9517bac6 | 2117 | ocfs2_free_alloc_context(data_ac); |
b1214e47 X |
2118 | data_ac = NULL; |
2119 | } | |
2120 | if (meta_ac) { | |
9517bac6 | 2121 | ocfs2_free_alloc_context(meta_ac); |
b1214e47 X |
2122 | meta_ac = NULL; |
2123 | } | |
50308d81 TM |
2124 | |
2125 | if (ret == -ENOSPC && try_free) { | |
2126 | /* | |
2127 | * Try to free some truncate log so that we can have enough | |
2128 | * clusters to allocate. | |
2129 | */ | |
2130 | try_free = 0; | |
2131 | ||
2132 | ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); | |
2133 | if (ret1 == 1) | |
2134 | goto try_again; | |
2135 | ||
2136 | if (ret1 < 0) | |
2137 | mlog_errno(ret1); | |
2138 | } | |
2139 | ||
3a307ffc MF |
2140 | return ret; |
2141 | } | |
2142 | ||
b6af1bcd NP |
2143 | static int ocfs2_write_begin(struct file *file, struct address_space *mapping, |
2144 | loff_t pos, unsigned len, unsigned flags, | |
2145 | struct page **pagep, void **fsdata) | |
607d44aa MF |
2146 | { |
2147 | int ret; | |
2148 | struct buffer_head *di_bh = NULL; | |
2149 | struct inode *inode = mapping->host; | |
2150 | ||
e63aecb6 | 2151 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
607d44aa MF |
2152 | if (ret) { |
2153 | mlog_errno(ret); | |
2154 | return ret; | |
2155 | } | |
2156 | ||
2157 | /* | |
2158 | * Take alloc sem here to prevent concurrent lookups. That way | |
2159 | * the mapping, zeroing and tree manipulation within | |
2160 | * ocfs2_write() will be safe against ->readpage(). This | |
2161 | * should also serve to lock out allocation from a shared | |
2162 | * writeable region. | |
2163 | */ | |
2164 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
2165 | ||
0378da0f | 2166 | ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep, |
7307de80 | 2167 | fsdata, di_bh, NULL); |
607d44aa MF |
2168 | if (ret) { |
2169 | mlog_errno(ret); | |
c934a92d | 2170 | goto out_fail; |
607d44aa MF |
2171 | } |
2172 | ||
2173 | brelse(di_bh); | |
2174 | ||
2175 | return 0; | |
2176 | ||
607d44aa MF |
2177 | out_fail: |
2178 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
2179 | ||
2180 | brelse(di_bh); | |
e63aecb6 | 2181 | ocfs2_inode_unlock(inode, 1); |
607d44aa MF |
2182 | |
2183 | return ret; | |
2184 | } | |
2185 | ||
1afc32b9 MF |
2186 | static void ocfs2_write_end_inline(struct inode *inode, loff_t pos, |
2187 | unsigned len, unsigned *copied, | |
2188 | struct ocfs2_dinode *di, | |
2189 | struct ocfs2_write_ctxt *wc) | |
2190 | { | |
2191 | void *kaddr; | |
2192 | ||
2193 | if (unlikely(*copied < len)) { | |
2194 | if (!PageUptodate(wc->w_target_page)) { | |
2195 | *copied = 0; | |
2196 | return; | |
2197 | } | |
2198 | } | |
2199 | ||
c4bc8dcb | 2200 | kaddr = kmap_atomic(wc->w_target_page); |
1afc32b9 | 2201 | memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied); |
c4bc8dcb | 2202 | kunmap_atomic(kaddr); |
1afc32b9 | 2203 | |
9558156b TM |
2204 | trace_ocfs2_write_end_inline( |
2205 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1afc32b9 MF |
2206 | (unsigned long long)pos, *copied, |
2207 | le16_to_cpu(di->id2.i_data.id_count), | |
2208 | le16_to_cpu(di->i_dyn_features)); | |
2209 | } | |
2210 | ||
7307de80 MF |
2211 | int ocfs2_write_end_nolock(struct address_space *mapping, |
2212 | loff_t pos, unsigned len, unsigned copied, | |
2213 | struct page *page, void *fsdata) | |
3a307ffc MF |
2214 | { |
2215 | int i; | |
2216 | unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1); | |
2217 | struct inode *inode = mapping->host; | |
2218 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2219 | struct ocfs2_write_ctxt *wc = fsdata; | |
2220 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
2221 | handle_t *handle = wc->w_handle; | |
2222 | struct page *tmppage; | |
2223 | ||
1afc32b9 MF |
2224 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
2225 | ocfs2_write_end_inline(inode, pos, len, &copied, di, wc); | |
2226 | goto out_write_size; | |
2227 | } | |
2228 | ||
3a307ffc MF |
2229 | if (unlikely(copied < len)) { |
2230 | if (!PageUptodate(wc->w_target_page)) | |
2231 | copied = 0; | |
2232 | ||
2233 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, | |
2234 | start+len); | |
2235 | } | |
2236 | flush_dcache_page(wc->w_target_page); | |
2237 | ||
2238 | for(i = 0; i < wc->w_num_pages; i++) { | |
2239 | tmppage = wc->w_pages[i]; | |
2240 | ||
2241 | if (tmppage == wc->w_target_page) { | |
2242 | from = wc->w_target_from; | |
2243 | to = wc->w_target_to; | |
2244 | ||
2245 | BUG_ON(from > PAGE_CACHE_SIZE || | |
2246 | to > PAGE_CACHE_SIZE || | |
2247 | to < from); | |
2248 | } else { | |
2249 | /* | |
2250 | * Pages adjacent to the target (if any) imply | |
2251 | * a hole-filling write in which case we want | |
2252 | * to flush their entire range. | |
2253 | */ | |
2254 | from = 0; | |
2255 | to = PAGE_CACHE_SIZE; | |
2256 | } | |
2257 | ||
961cecbe | 2258 | if (page_has_buffers(tmppage)) { |
53ef99ca | 2259 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 2260 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
2261 | block_commit_write(tmppage, from, to); |
2262 | } | |
3a307ffc MF |
2263 | } |
2264 | ||
1afc32b9 | 2265 | out_write_size: |
3a307ffc | 2266 | pos += copied; |
f17c20dd | 2267 | if (pos > i_size_read(inode)) { |
3a307ffc MF |
2268 | i_size_write(inode, pos); |
2269 | mark_inode_dirty(inode); | |
2270 | } | |
2271 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
2272 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
2273 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
2274 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); | |
2275 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
2931cdcb | 2276 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
3a307ffc MF |
2277 | ocfs2_journal_dirty(handle, wc->w_di_bh); |
2278 | ||
136f49b9 JB |
2279 | /* unlock pages before dealloc since it needs acquiring j_trans_barrier |
2280 | * lock, or it will cause a deadlock since journal commit threads holds | |
2281 | * this lock and will ask for the page lock when flushing the data. | |
2282 | * put it here to preserve the unlock order. | |
2283 | */ | |
2284 | ocfs2_unlock_pages(wc); | |
2285 | ||
3a307ffc | 2286 | ocfs2_commit_trans(osb, handle); |
59a5e416 | 2287 | |
b27b7cbc MF |
2288 | ocfs2_run_deallocs(osb, &wc->w_dealloc); |
2289 | ||
136f49b9 JB |
2290 | brelse(wc->w_di_bh); |
2291 | kfree(wc); | |
607d44aa MF |
2292 | |
2293 | return copied; | |
2294 | } | |
2295 | ||
b6af1bcd NP |
2296 | static int ocfs2_write_end(struct file *file, struct address_space *mapping, |
2297 | loff_t pos, unsigned len, unsigned copied, | |
2298 | struct page *page, void *fsdata) | |
607d44aa MF |
2299 | { |
2300 | int ret; | |
2301 | struct inode *inode = mapping->host; | |
2302 | ||
2303 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata); | |
2304 | ||
3a307ffc | 2305 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
e63aecb6 | 2306 | ocfs2_inode_unlock(inode, 1); |
9517bac6 | 2307 | |
607d44aa | 2308 | return ret; |
9517bac6 MF |
2309 | } |
2310 | ||
f5e54d6e | 2311 | const struct address_space_operations ocfs2_aops = { |
1fca3a05 HH |
2312 | .readpage = ocfs2_readpage, |
2313 | .readpages = ocfs2_readpages, | |
2314 | .writepage = ocfs2_writepage, | |
2315 | .write_begin = ocfs2_write_begin, | |
2316 | .write_end = ocfs2_write_end, | |
2317 | .bmap = ocfs2_bmap, | |
1fca3a05 | 2318 | .direct_IO = ocfs2_direct_IO, |
41ecc345 | 2319 | .invalidatepage = block_invalidatepage, |
1fca3a05 HH |
2320 | .releasepage = ocfs2_releasepage, |
2321 | .migratepage = buffer_migrate_page, | |
2322 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 2323 | .error_remove_page = generic_error_remove_page, |
ccd979bd | 2324 | }; |