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) { | |
7e9b1955 | 709 | ret = ocfs2_inode_lock(inode, NULL, 1); |
24c40b32 JQ |
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); | |
24c40b32 JQ |
723 | goto clean_orphan; |
724 | } | |
725 | ||
726 | is_overwrite = ocfs2_is_overwrite(osb, inode, offset); | |
727 | if (is_overwrite < 0) { | |
728 | mlog_errno(is_overwrite); | |
729 | ocfs2_inode_unlock(inode, 1); | |
24c40b32 JQ |
730 | goto clean_orphan; |
731 | } | |
732 | ||
733 | ocfs2_inode_unlock(inode, 1); | |
24c40b32 JQ |
734 | } |
735 | ||
736 | written = __blockdev_direct_IO(WRITE, iocb, inode, inode->i_sb->s_bdev, | |
737 | iter, offset, | |
738 | ocfs2_direct_IO_get_blocks, | |
739 | ocfs2_dio_end_io, NULL, 0); | |
740 | if (unlikely(written < 0)) { | |
741 | loff_t i_size = i_size_read(inode); | |
742 | ||
743 | if (offset + count > i_size) { | |
744 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
745 | if (ret < 0) { | |
746 | mlog_errno(ret); | |
747 | goto clean_orphan; | |
748 | } | |
749 | ||
750 | if (i_size == i_size_read(inode)) { | |
751 | ret = ocfs2_truncate_file(inode, di_bh, | |
752 | i_size); | |
753 | if (ret < 0) { | |
754 | if (ret != -ENOSPC) | |
755 | mlog_errno(ret); | |
756 | ||
757 | ocfs2_inode_unlock(inode, 1); | |
758 | brelse(di_bh); | |
759 | goto clean_orphan; | |
760 | } | |
761 | } | |
762 | ||
763 | ocfs2_inode_unlock(inode, 1); | |
764 | brelse(di_bh); | |
765 | ||
766 | ret = jbd2_journal_force_commit(journal); | |
767 | if (ret < 0) | |
768 | mlog_errno(ret); | |
769 | } | |
bdd86215 | 770 | } else if (written > 0 && append_write && !is_overwrite && |
24c40b32 JQ |
771 | !cluster_align) { |
772 | u32 p_cpos = 0; | |
773 | u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset); | |
774 | ||
37a8d89a JQ |
775 | ret = ocfs2_inode_lock(inode, NULL, 0); |
776 | if (ret < 0) { | |
777 | mlog_errno(ret); | |
778 | goto clean_orphan; | |
779 | } | |
780 | ||
24c40b32 JQ |
781 | ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, |
782 | &num_clusters, &ext_flags); | |
783 | if (ret < 0) { | |
784 | mlog_errno(ret); | |
37a8d89a | 785 | ocfs2_inode_unlock(inode, 0); |
24c40b32 JQ |
786 | goto clean_orphan; |
787 | } | |
788 | ||
789 | BUG_ON(!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN)); | |
790 | ||
791 | ret = blkdev_issue_zeroout(osb->sb->s_bdev, | |
792 | p_cpos << (osb->s_clustersize_bits - 9), | |
37a8d89a | 793 | zero_len >> 9, GFP_NOFS, false); |
24c40b32 JQ |
794 | if (ret < 0) |
795 | mlog_errno(ret); | |
37a8d89a JQ |
796 | |
797 | ocfs2_inode_unlock(inode, 0); | |
24c40b32 JQ |
798 | } |
799 | ||
800 | clean_orphan: | |
801 | if (orphaned) { | |
802 | int tmp_ret; | |
803 | int update_isize = written > 0 ? 1 : 0; | |
804 | loff_t end = update_isize ? offset + written : 0; | |
805 | ||
806 | tmp_ret = ocfs2_del_inode_from_orphan(osb, inode, | |
807 | update_isize, end); | |
808 | if (tmp_ret < 0) { | |
809 | ret = tmp_ret; | |
810 | goto out; | |
811 | } | |
812 | ||
813 | tmp_ret = jbd2_journal_force_commit(journal); | |
814 | if (tmp_ret < 0) { | |
815 | ret = tmp_ret; | |
816 | mlog_errno(tmp_ret); | |
817 | } | |
818 | } | |
819 | ||
820 | out: | |
821 | if (ret >= 0) | |
822 | ret = written; | |
823 | return ret; | |
824 | } | |
825 | ||
ccd979bd MF |
826 | static ssize_t ocfs2_direct_IO(int rw, |
827 | struct kiocb *iocb, | |
d8d3d94b AV |
828 | struct iov_iter *iter, |
829 | loff_t offset) | |
ccd979bd MF |
830 | { |
831 | struct file *file = iocb->ki_filp; | |
496ad9aa | 832 | struct inode *inode = file_inode(file)->i_mapping->host; |
24c40b32 JQ |
833 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
834 | int full_coherency = !(osb->s_mount_opt & | |
835 | OCFS2_MOUNT_COHERENCY_BUFFERED); | |
53013cba | 836 | |
6798d35a MF |
837 | /* |
838 | * Fallback to buffered I/O if we see an inode without | |
839 | * extents. | |
840 | */ | |
841 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
842 | return 0; | |
843 | ||
24c40b32 JQ |
844 | /* Fallback to buffered I/O if we are appending and |
845 | * concurrent O_DIRECT writes are allowed. | |
846 | */ | |
847 | if (i_size_read(inode) <= offset && !full_coherency) | |
b80474b4 TM |
848 | return 0; |
849 | ||
24c40b32 JQ |
850 | if (rw == READ) |
851 | return __blockdev_direct_IO(rw, iocb, inode, | |
852 | inode->i_sb->s_bdev, | |
31b14039 | 853 | iter, offset, |
c1e8d35e TM |
854 | ocfs2_direct_IO_get_blocks, |
855 | ocfs2_dio_end_io, NULL, 0); | |
24c40b32 JQ |
856 | else |
857 | return ocfs2_direct_IO_write(iocb, iter, offset); | |
ccd979bd MF |
858 | } |
859 | ||
9517bac6 MF |
860 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, |
861 | u32 cpos, | |
862 | unsigned int *start, | |
863 | unsigned int *end) | |
864 | { | |
865 | unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE; | |
866 | ||
867 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) { | |
868 | unsigned int cpp; | |
869 | ||
870 | cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits); | |
871 | ||
872 | cluster_start = cpos % cpp; | |
873 | cluster_start = cluster_start << osb->s_clustersize_bits; | |
874 | ||
875 | cluster_end = cluster_start + osb->s_clustersize; | |
876 | } | |
877 | ||
878 | BUG_ON(cluster_start > PAGE_SIZE); | |
879 | BUG_ON(cluster_end > PAGE_SIZE); | |
880 | ||
881 | if (start) | |
882 | *start = cluster_start; | |
883 | if (end) | |
884 | *end = cluster_end; | |
885 | } | |
886 | ||
887 | /* | |
888 | * 'from' and 'to' are the region in the page to avoid zeroing. | |
889 | * | |
890 | * If pagesize > clustersize, this function will avoid zeroing outside | |
891 | * of the cluster boundary. | |
892 | * | |
893 | * from == to == 0 is code for "zero the entire cluster region" | |
894 | */ | |
895 | static void ocfs2_clear_page_regions(struct page *page, | |
896 | struct ocfs2_super *osb, u32 cpos, | |
897 | unsigned from, unsigned to) | |
898 | { | |
899 | void *kaddr; | |
900 | unsigned int cluster_start, cluster_end; | |
901 | ||
902 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); | |
903 | ||
c4bc8dcb | 904 | kaddr = kmap_atomic(page); |
9517bac6 MF |
905 | |
906 | if (from || to) { | |
907 | if (from > cluster_start) | |
908 | memset(kaddr + cluster_start, 0, from - cluster_start); | |
909 | if (to < cluster_end) | |
910 | memset(kaddr + to, 0, cluster_end - to); | |
911 | } else { | |
912 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); | |
913 | } | |
914 | ||
c4bc8dcb | 915 | kunmap_atomic(kaddr); |
9517bac6 MF |
916 | } |
917 | ||
4e9563fd MF |
918 | /* |
919 | * Nonsparse file systems fully allocate before we get to the write | |
920 | * code. This prevents ocfs2_write() from tagging the write as an | |
921 | * allocating one, which means ocfs2_map_page_blocks() might try to | |
922 | * read-in the blocks at the tail of our file. Avoid reading them by | |
923 | * testing i_size against each block offset. | |
924 | */ | |
925 | static int ocfs2_should_read_blk(struct inode *inode, struct page *page, | |
926 | unsigned int block_start) | |
927 | { | |
928 | u64 offset = page_offset(page) + block_start; | |
929 | ||
930 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
931 | return 1; | |
932 | ||
933 | if (i_size_read(inode) > offset) | |
934 | return 1; | |
935 | ||
936 | return 0; | |
937 | } | |
938 | ||
9517bac6 | 939 | /* |
ebdec241 | 940 | * Some of this taken from __block_write_begin(). We already have our |
9517bac6 MF |
941 | * mapping by now though, and the entire write will be allocating or |
942 | * it won't, so not much need to use BH_New. | |
943 | * | |
944 | * This will also skip zeroing, which is handled externally. | |
945 | */ | |
60b11392 MF |
946 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, |
947 | struct inode *inode, unsigned int from, | |
948 | unsigned int to, int new) | |
9517bac6 MF |
949 | { |
950 | int ret = 0; | |
951 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; | |
952 | unsigned int block_end, block_start; | |
953 | unsigned int bsize = 1 << inode->i_blkbits; | |
954 | ||
955 | if (!page_has_buffers(page)) | |
956 | create_empty_buffers(page, bsize, 0); | |
957 | ||
958 | head = page_buffers(page); | |
959 | for (bh = head, block_start = 0; bh != head || !block_start; | |
960 | bh = bh->b_this_page, block_start += bsize) { | |
961 | block_end = block_start + bsize; | |
962 | ||
3a307ffc MF |
963 | clear_buffer_new(bh); |
964 | ||
9517bac6 MF |
965 | /* |
966 | * Ignore blocks outside of our i/o range - | |
967 | * they may belong to unallocated clusters. | |
968 | */ | |
60b11392 | 969 | if (block_start >= to || block_end <= from) { |
9517bac6 MF |
970 | if (PageUptodate(page)) |
971 | set_buffer_uptodate(bh); | |
972 | continue; | |
973 | } | |
974 | ||
975 | /* | |
976 | * For an allocating write with cluster size >= page | |
977 | * size, we always write the entire page. | |
978 | */ | |
3a307ffc MF |
979 | if (new) |
980 | set_buffer_new(bh); | |
9517bac6 MF |
981 | |
982 | if (!buffer_mapped(bh)) { | |
983 | map_bh(bh, inode->i_sb, *p_blkno); | |
984 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
985 | } | |
986 | ||
987 | if (PageUptodate(page)) { | |
988 | if (!buffer_uptodate(bh)) | |
989 | set_buffer_uptodate(bh); | |
990 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
bce99768 | 991 | !buffer_new(bh) && |
4e9563fd | 992 | ocfs2_should_read_blk(inode, page, block_start) && |
bce99768 | 993 | (block_start < from || block_end > to)) { |
9517bac6 MF |
994 | ll_rw_block(READ, 1, &bh); |
995 | *wait_bh++=bh; | |
996 | } | |
997 | ||
998 | *p_blkno = *p_blkno + 1; | |
999 | } | |
1000 | ||
1001 | /* | |
1002 | * If we issued read requests - let them complete. | |
1003 | */ | |
1004 | while(wait_bh > wait) { | |
1005 | wait_on_buffer(*--wait_bh); | |
1006 | if (!buffer_uptodate(*wait_bh)) | |
1007 | ret = -EIO; | |
1008 | } | |
1009 | ||
1010 | if (ret == 0 || !new) | |
1011 | return ret; | |
1012 | ||
1013 | /* | |
1014 | * If we get -EIO above, zero out any newly allocated blocks | |
1015 | * to avoid exposing stale data. | |
1016 | */ | |
1017 | bh = head; | |
1018 | block_start = 0; | |
1019 | do { | |
9517bac6 MF |
1020 | block_end = block_start + bsize; |
1021 | if (block_end <= from) | |
1022 | goto next_bh; | |
1023 | if (block_start >= to) | |
1024 | break; | |
1025 | ||
eebd2aa3 | 1026 | zero_user(page, block_start, bh->b_size); |
9517bac6 MF |
1027 | set_buffer_uptodate(bh); |
1028 | mark_buffer_dirty(bh); | |
1029 | ||
1030 | next_bh: | |
1031 | block_start = block_end; | |
1032 | bh = bh->b_this_page; | |
1033 | } while (bh != head); | |
1034 | ||
1035 | return ret; | |
1036 | } | |
1037 | ||
3a307ffc MF |
1038 | #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE) |
1039 | #define OCFS2_MAX_CTXT_PAGES 1 | |
1040 | #else | |
1041 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE) | |
1042 | #endif | |
1043 | ||
1044 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE) | |
1045 | ||
6af67d82 | 1046 | /* |
3a307ffc | 1047 | * Describe the state of a single cluster to be written to. |
6af67d82 | 1048 | */ |
3a307ffc MF |
1049 | struct ocfs2_write_cluster_desc { |
1050 | u32 c_cpos; | |
1051 | u32 c_phys; | |
1052 | /* | |
1053 | * Give this a unique field because c_phys eventually gets | |
1054 | * filled. | |
1055 | */ | |
1056 | unsigned c_new; | |
b27b7cbc | 1057 | unsigned c_unwritten; |
e7432675 | 1058 | unsigned c_needs_zero; |
3a307ffc | 1059 | }; |
6af67d82 | 1060 | |
3a307ffc MF |
1061 | struct ocfs2_write_ctxt { |
1062 | /* Logical cluster position / len of write */ | |
1063 | u32 w_cpos; | |
1064 | u32 w_clen; | |
6af67d82 | 1065 | |
e7432675 SM |
1066 | /* First cluster allocated in a nonsparse extend */ |
1067 | u32 w_first_new_cpos; | |
1068 | ||
3a307ffc | 1069 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; |
6af67d82 | 1070 | |
3a307ffc MF |
1071 | /* |
1072 | * This is true if page_size > cluster_size. | |
1073 | * | |
1074 | * It triggers a set of special cases during write which might | |
1075 | * have to deal with allocating writes to partial pages. | |
1076 | */ | |
1077 | unsigned int w_large_pages; | |
6af67d82 | 1078 | |
3a307ffc MF |
1079 | /* |
1080 | * Pages involved in this write. | |
1081 | * | |
1082 | * w_target_page is the page being written to by the user. | |
1083 | * | |
1084 | * w_pages is an array of pages which always contains | |
1085 | * w_target_page, and in the case of an allocating write with | |
1086 | * page_size < cluster size, it will contain zero'd and mapped | |
1087 | * pages adjacent to w_target_page which need to be written | |
1088 | * out in so that future reads from that region will get | |
1089 | * zero's. | |
1090 | */ | |
3a307ffc | 1091 | unsigned int w_num_pages; |
83fd9c7f | 1092 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; |
3a307ffc | 1093 | struct page *w_target_page; |
eeb47d12 | 1094 | |
5cffff9e WW |
1095 | /* |
1096 | * w_target_locked is used for page_mkwrite path indicating no unlocking | |
1097 | * against w_target_page in ocfs2_write_end_nolock. | |
1098 | */ | |
1099 | unsigned int w_target_locked:1; | |
1100 | ||
3a307ffc MF |
1101 | /* |
1102 | * ocfs2_write_end() uses this to know what the real range to | |
1103 | * write in the target should be. | |
1104 | */ | |
1105 | unsigned int w_target_from; | |
1106 | unsigned int w_target_to; | |
1107 | ||
1108 | /* | |
1109 | * We could use journal_current_handle() but this is cleaner, | |
1110 | * IMHO -Mark | |
1111 | */ | |
1112 | handle_t *w_handle; | |
1113 | ||
1114 | struct buffer_head *w_di_bh; | |
b27b7cbc MF |
1115 | |
1116 | struct ocfs2_cached_dealloc_ctxt w_dealloc; | |
3a307ffc MF |
1117 | }; |
1118 | ||
1d410a6e | 1119 | void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages) |
3a307ffc MF |
1120 | { |
1121 | int i; | |
1122 | ||
1d410a6e MF |
1123 | for(i = 0; i < num_pages; i++) { |
1124 | if (pages[i]) { | |
1125 | unlock_page(pages[i]); | |
1126 | mark_page_accessed(pages[i]); | |
1127 | page_cache_release(pages[i]); | |
1128 | } | |
6af67d82 | 1129 | } |
1d410a6e MF |
1130 | } |
1131 | ||
136f49b9 | 1132 | static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc) |
1d410a6e | 1133 | { |
5cffff9e WW |
1134 | int i; |
1135 | ||
1136 | /* | |
1137 | * w_target_locked is only set to true in the page_mkwrite() case. | |
1138 | * The intent is to allow us to lock the target page from write_begin() | |
1139 | * to write_end(). The caller must hold a ref on w_target_page. | |
1140 | */ | |
1141 | if (wc->w_target_locked) { | |
1142 | BUG_ON(!wc->w_target_page); | |
1143 | for (i = 0; i < wc->w_num_pages; i++) { | |
1144 | if (wc->w_target_page == wc->w_pages[i]) { | |
1145 | wc->w_pages[i] = NULL; | |
1146 | break; | |
1147 | } | |
1148 | } | |
1149 | mark_page_accessed(wc->w_target_page); | |
1150 | page_cache_release(wc->w_target_page); | |
1151 | } | |
1d410a6e | 1152 | ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages); |
136f49b9 | 1153 | } |
6af67d82 | 1154 | |
136f49b9 JB |
1155 | static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc) |
1156 | { | |
1157 | ocfs2_unlock_pages(wc); | |
3a307ffc MF |
1158 | brelse(wc->w_di_bh); |
1159 | kfree(wc); | |
1160 | } | |
1161 | ||
1162 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, | |
1163 | struct ocfs2_super *osb, loff_t pos, | |
607d44aa | 1164 | unsigned len, struct buffer_head *di_bh) |
3a307ffc | 1165 | { |
30b8548f | 1166 | u32 cend; |
3a307ffc MF |
1167 | struct ocfs2_write_ctxt *wc; |
1168 | ||
1169 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); | |
1170 | if (!wc) | |
1171 | return -ENOMEM; | |
6af67d82 | 1172 | |
3a307ffc | 1173 | wc->w_cpos = pos >> osb->s_clustersize_bits; |
e7432675 | 1174 | wc->w_first_new_cpos = UINT_MAX; |
30b8548f | 1175 | cend = (pos + len - 1) >> osb->s_clustersize_bits; |
1176 | wc->w_clen = cend - wc->w_cpos + 1; | |
607d44aa MF |
1177 | get_bh(di_bh); |
1178 | wc->w_di_bh = di_bh; | |
6af67d82 | 1179 | |
3a307ffc MF |
1180 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) |
1181 | wc->w_large_pages = 1; | |
1182 | else | |
1183 | wc->w_large_pages = 0; | |
1184 | ||
b27b7cbc MF |
1185 | ocfs2_init_dealloc_ctxt(&wc->w_dealloc); |
1186 | ||
3a307ffc | 1187 | *wcp = wc; |
6af67d82 | 1188 | |
3a307ffc | 1189 | return 0; |
6af67d82 MF |
1190 | } |
1191 | ||
9517bac6 | 1192 | /* |
3a307ffc MF |
1193 | * If a page has any new buffers, zero them out here, and mark them uptodate |
1194 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1195 | * block data from leaking). And clear the new bit. | |
9517bac6 | 1196 | */ |
3a307ffc | 1197 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) |
9517bac6 | 1198 | { |
3a307ffc MF |
1199 | unsigned int block_start, block_end; |
1200 | struct buffer_head *head, *bh; | |
9517bac6 | 1201 | |
3a307ffc MF |
1202 | BUG_ON(!PageLocked(page)); |
1203 | if (!page_has_buffers(page)) | |
1204 | return; | |
9517bac6 | 1205 | |
3a307ffc MF |
1206 | bh = head = page_buffers(page); |
1207 | block_start = 0; | |
1208 | do { | |
1209 | block_end = block_start + bh->b_size; | |
1210 | ||
1211 | if (buffer_new(bh)) { | |
1212 | if (block_end > from && block_start < to) { | |
1213 | if (!PageUptodate(page)) { | |
1214 | unsigned start, end; | |
3a307ffc MF |
1215 | |
1216 | start = max(from, block_start); | |
1217 | end = min(to, block_end); | |
1218 | ||
eebd2aa3 | 1219 | zero_user_segment(page, start, end); |
3a307ffc MF |
1220 | set_buffer_uptodate(bh); |
1221 | } | |
1222 | ||
1223 | clear_buffer_new(bh); | |
1224 | mark_buffer_dirty(bh); | |
1225 | } | |
1226 | } | |
9517bac6 | 1227 | |
3a307ffc MF |
1228 | block_start = block_end; |
1229 | bh = bh->b_this_page; | |
1230 | } while (bh != head); | |
1231 | } | |
1232 | ||
1233 | /* | |
1234 | * Only called when we have a failure during allocating write to write | |
1235 | * zero's to the newly allocated region. | |
1236 | */ | |
1237 | static void ocfs2_write_failure(struct inode *inode, | |
1238 | struct ocfs2_write_ctxt *wc, | |
1239 | loff_t user_pos, unsigned user_len) | |
1240 | { | |
1241 | int i; | |
5c26a7b7 MF |
1242 | unsigned from = user_pos & (PAGE_CACHE_SIZE - 1), |
1243 | to = user_pos + user_len; | |
3a307ffc MF |
1244 | struct page *tmppage; |
1245 | ||
5c26a7b7 | 1246 | ocfs2_zero_new_buffers(wc->w_target_page, from, to); |
9517bac6 | 1247 | |
3a307ffc MF |
1248 | for(i = 0; i < wc->w_num_pages; i++) { |
1249 | tmppage = wc->w_pages[i]; | |
9517bac6 | 1250 | |
961cecbe | 1251 | if (page_has_buffers(tmppage)) { |
53ef99ca | 1252 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 1253 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
1254 | |
1255 | block_commit_write(tmppage, from, to); | |
1256 | } | |
9517bac6 | 1257 | } |
9517bac6 MF |
1258 | } |
1259 | ||
3a307ffc MF |
1260 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, |
1261 | struct ocfs2_write_ctxt *wc, | |
1262 | struct page *page, u32 cpos, | |
1263 | loff_t user_pos, unsigned user_len, | |
1264 | int new) | |
9517bac6 | 1265 | { |
3a307ffc MF |
1266 | int ret; |
1267 | unsigned int map_from = 0, map_to = 0; | |
9517bac6 | 1268 | unsigned int cluster_start, cluster_end; |
3a307ffc | 1269 | unsigned int user_data_from = 0, user_data_to = 0; |
9517bac6 | 1270 | |
3a307ffc | 1271 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, |
9517bac6 MF |
1272 | &cluster_start, &cluster_end); |
1273 | ||
272b62c1 GR |
1274 | /* treat the write as new if the a hole/lseek spanned across |
1275 | * the page boundary. | |
1276 | */ | |
1277 | new = new | ((i_size_read(inode) <= page_offset(page)) && | |
1278 | (page_offset(page) <= user_pos)); | |
1279 | ||
3a307ffc MF |
1280 | if (page == wc->w_target_page) { |
1281 | map_from = user_pos & (PAGE_CACHE_SIZE - 1); | |
1282 | map_to = map_from + user_len; | |
1283 | ||
1284 | if (new) | |
1285 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1286 | cluster_start, cluster_end, | |
1287 | new); | |
1288 | else | |
1289 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1290 | map_from, map_to, new); | |
1291 | if (ret) { | |
9517bac6 MF |
1292 | mlog_errno(ret); |
1293 | goto out; | |
1294 | } | |
1295 | ||
3a307ffc MF |
1296 | user_data_from = map_from; |
1297 | user_data_to = map_to; | |
9517bac6 | 1298 | if (new) { |
3a307ffc MF |
1299 | map_from = cluster_start; |
1300 | map_to = cluster_end; | |
9517bac6 MF |
1301 | } |
1302 | } else { | |
1303 | /* | |
1304 | * If we haven't allocated the new page yet, we | |
1305 | * shouldn't be writing it out without copying user | |
1306 | * data. This is likely a math error from the caller. | |
1307 | */ | |
1308 | BUG_ON(!new); | |
1309 | ||
3a307ffc MF |
1310 | map_from = cluster_start; |
1311 | map_to = cluster_end; | |
9517bac6 MF |
1312 | |
1313 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
3a307ffc | 1314 | cluster_start, cluster_end, new); |
9517bac6 MF |
1315 | if (ret) { |
1316 | mlog_errno(ret); | |
1317 | goto out; | |
1318 | } | |
1319 | } | |
1320 | ||
1321 | /* | |
1322 | * Parts of newly allocated pages need to be zero'd. | |
1323 | * | |
1324 | * Above, we have also rewritten 'to' and 'from' - as far as | |
1325 | * the rest of the function is concerned, the entire cluster | |
1326 | * range inside of a page needs to be written. | |
1327 | * | |
1328 | * We can skip this if the page is up to date - it's already | |
1329 | * been zero'd from being read in as a hole. | |
1330 | */ | |
1331 | if (new && !PageUptodate(page)) | |
1332 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), | |
3a307ffc | 1333 | cpos, user_data_from, user_data_to); |
9517bac6 MF |
1334 | |
1335 | flush_dcache_page(page); | |
1336 | ||
9517bac6 | 1337 | out: |
3a307ffc | 1338 | return ret; |
9517bac6 MF |
1339 | } |
1340 | ||
1341 | /* | |
3a307ffc | 1342 | * This function will only grab one clusters worth of pages. |
9517bac6 | 1343 | */ |
3a307ffc MF |
1344 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, |
1345 | struct ocfs2_write_ctxt *wc, | |
693c241a JB |
1346 | u32 cpos, loff_t user_pos, |
1347 | unsigned user_len, int new, | |
7307de80 | 1348 | struct page *mmap_page) |
9517bac6 | 1349 | { |
3a307ffc | 1350 | int ret = 0, i; |
693c241a | 1351 | unsigned long start, target_index, end_index, index; |
9517bac6 | 1352 | struct inode *inode = mapping->host; |
693c241a | 1353 | loff_t last_byte; |
9517bac6 | 1354 | |
3a307ffc | 1355 | target_index = user_pos >> PAGE_CACHE_SHIFT; |
9517bac6 MF |
1356 | |
1357 | /* | |
1358 | * Figure out how many pages we'll be manipulating here. For | |
60b11392 | 1359 | * non allocating write, we just change the one |
693c241a JB |
1360 | * page. Otherwise, we'll need a whole clusters worth. If we're |
1361 | * writing past i_size, we only need enough pages to cover the | |
1362 | * last page of the write. | |
9517bac6 | 1363 | */ |
9517bac6 | 1364 | if (new) { |
3a307ffc MF |
1365 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); |
1366 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); | |
693c241a JB |
1367 | /* |
1368 | * We need the index *past* the last page we could possibly | |
1369 | * touch. This is the page past the end of the write or | |
1370 | * i_size, whichever is greater. | |
1371 | */ | |
1372 | last_byte = max(user_pos + user_len, i_size_read(inode)); | |
1373 | BUG_ON(last_byte < 1); | |
1374 | end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1; | |
1375 | if ((start + wc->w_num_pages) > end_index) | |
1376 | wc->w_num_pages = end_index - start; | |
9517bac6 | 1377 | } else { |
3a307ffc MF |
1378 | wc->w_num_pages = 1; |
1379 | start = target_index; | |
9517bac6 MF |
1380 | } |
1381 | ||
3a307ffc | 1382 | for(i = 0; i < wc->w_num_pages; i++) { |
9517bac6 MF |
1383 | index = start + i; |
1384 | ||
7307de80 MF |
1385 | if (index == target_index && mmap_page) { |
1386 | /* | |
1387 | * ocfs2_pagemkwrite() is a little different | |
1388 | * and wants us to directly use the page | |
1389 | * passed in. | |
1390 | */ | |
1391 | lock_page(mmap_page); | |
1392 | ||
5cffff9e | 1393 | /* Exit and let the caller retry */ |
7307de80 | 1394 | if (mmap_page->mapping != mapping) { |
5cffff9e | 1395 | WARN_ON(mmap_page->mapping); |
7307de80 | 1396 | unlock_page(mmap_page); |
5cffff9e | 1397 | ret = -EAGAIN; |
7307de80 MF |
1398 | goto out; |
1399 | } | |
1400 | ||
1401 | page_cache_get(mmap_page); | |
1402 | wc->w_pages[i] = mmap_page; | |
5cffff9e | 1403 | wc->w_target_locked = true; |
7307de80 MF |
1404 | } else { |
1405 | wc->w_pages[i] = find_or_create_page(mapping, index, | |
1406 | GFP_NOFS); | |
1407 | if (!wc->w_pages[i]) { | |
1408 | ret = -ENOMEM; | |
1409 | mlog_errno(ret); | |
1410 | goto out; | |
1411 | } | |
9517bac6 | 1412 | } |
1269529b | 1413 | wait_for_stable_page(wc->w_pages[i]); |
3a307ffc MF |
1414 | |
1415 | if (index == target_index) | |
1416 | wc->w_target_page = wc->w_pages[i]; | |
9517bac6 | 1417 | } |
3a307ffc | 1418 | out: |
5cffff9e WW |
1419 | if (ret) |
1420 | wc->w_target_locked = false; | |
3a307ffc MF |
1421 | return ret; |
1422 | } | |
1423 | ||
1424 | /* | |
1425 | * Prepare a single cluster for write one cluster into the file. | |
1426 | */ | |
1427 | static int ocfs2_write_cluster(struct address_space *mapping, | |
b27b7cbc | 1428 | u32 phys, unsigned int unwritten, |
e7432675 | 1429 | unsigned int should_zero, |
b27b7cbc | 1430 | struct ocfs2_alloc_context *data_ac, |
3a307ffc MF |
1431 | struct ocfs2_alloc_context *meta_ac, |
1432 | struct ocfs2_write_ctxt *wc, u32 cpos, | |
1433 | loff_t user_pos, unsigned user_len) | |
1434 | { | |
e7432675 | 1435 | int ret, i, new; |
3a307ffc MF |
1436 | u64 v_blkno, p_blkno; |
1437 | struct inode *inode = mapping->host; | |
f99b9b7c | 1438 | struct ocfs2_extent_tree et; |
3a307ffc MF |
1439 | |
1440 | new = phys == 0 ? 1 : 0; | |
9517bac6 | 1441 | if (new) { |
3a307ffc MF |
1442 | u32 tmp_pos; |
1443 | ||
9517bac6 MF |
1444 | /* |
1445 | * This is safe to call with the page locks - it won't take | |
1446 | * any additional semaphores or cluster locks. | |
1447 | */ | |
3a307ffc | 1448 | tmp_pos = cpos; |
0eb8d47e TM |
1449 | ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode, |
1450 | &tmp_pos, 1, 0, wc->w_di_bh, | |
1451 | wc->w_handle, data_ac, | |
1452 | meta_ac, NULL); | |
9517bac6 MF |
1453 | /* |
1454 | * This shouldn't happen because we must have already | |
1455 | * calculated the correct meta data allocation required. The | |
1456 | * internal tree allocation code should know how to increase | |
1457 | * transaction credits itself. | |
1458 | * | |
1459 | * If need be, we could handle -EAGAIN for a | |
1460 | * RESTART_TRANS here. | |
1461 | */ | |
1462 | mlog_bug_on_msg(ret == -EAGAIN, | |
1463 | "Inode %llu: EAGAIN return during allocation.\n", | |
1464 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1465 | if (ret < 0) { | |
1466 | mlog_errno(ret); | |
1467 | goto out; | |
1468 | } | |
b27b7cbc | 1469 | } else if (unwritten) { |
5e404e9e JB |
1470 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
1471 | wc->w_di_bh); | |
f99b9b7c | 1472 | ret = ocfs2_mark_extent_written(inode, &et, |
b27b7cbc | 1473 | wc->w_handle, cpos, 1, phys, |
f99b9b7c | 1474 | meta_ac, &wc->w_dealloc); |
b27b7cbc MF |
1475 | if (ret < 0) { |
1476 | mlog_errno(ret); | |
1477 | goto out; | |
1478 | } | |
1479 | } | |
3a307ffc | 1480 | |
b27b7cbc | 1481 | if (should_zero) |
3a307ffc | 1482 | v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos); |
b27b7cbc | 1483 | else |
3a307ffc | 1484 | v_blkno = user_pos >> inode->i_sb->s_blocksize_bits; |
9517bac6 | 1485 | |
3a307ffc MF |
1486 | /* |
1487 | * The only reason this should fail is due to an inability to | |
1488 | * find the extent added. | |
1489 | */ | |
49cb8d2d MF |
1490 | ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL, |
1491 | NULL); | |
9517bac6 | 1492 | if (ret < 0) { |
61fb9ea4 | 1493 | mlog(ML_ERROR, "Get physical blkno failed for inode %llu, " |
3a307ffc MF |
1494 | "at logical block %llu", |
1495 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1496 | (unsigned long long)v_blkno); | |
9517bac6 MF |
1497 | goto out; |
1498 | } | |
1499 | ||
1500 | BUG_ON(p_blkno == 0); | |
1501 | ||
3a307ffc MF |
1502 | for(i = 0; i < wc->w_num_pages; i++) { |
1503 | int tmpret; | |
9517bac6 | 1504 | |
3a307ffc MF |
1505 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, |
1506 | wc->w_pages[i], cpos, | |
b27b7cbc MF |
1507 | user_pos, user_len, |
1508 | should_zero); | |
3a307ffc MF |
1509 | if (tmpret) { |
1510 | mlog_errno(tmpret); | |
1511 | if (ret == 0) | |
cbfa9639 | 1512 | ret = tmpret; |
3a307ffc | 1513 | } |
9517bac6 MF |
1514 | } |
1515 | ||
3a307ffc MF |
1516 | /* |
1517 | * We only have cleanup to do in case of allocating write. | |
1518 | */ | |
1519 | if (ret && new) | |
1520 | ocfs2_write_failure(inode, wc, user_pos, user_len); | |
1521 | ||
9517bac6 | 1522 | out: |
9517bac6 | 1523 | |
3a307ffc | 1524 | return ret; |
9517bac6 MF |
1525 | } |
1526 | ||
0d172baa MF |
1527 | static int ocfs2_write_cluster_by_desc(struct address_space *mapping, |
1528 | struct ocfs2_alloc_context *data_ac, | |
1529 | struct ocfs2_alloc_context *meta_ac, | |
1530 | struct ocfs2_write_ctxt *wc, | |
1531 | loff_t pos, unsigned len) | |
1532 | { | |
1533 | int ret, i; | |
db56246c MF |
1534 | loff_t cluster_off; |
1535 | unsigned int local_len = len; | |
0d172baa | 1536 | struct ocfs2_write_cluster_desc *desc; |
db56246c | 1537 | struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb); |
0d172baa MF |
1538 | |
1539 | for (i = 0; i < wc->w_clen; i++) { | |
1540 | desc = &wc->w_desc[i]; | |
1541 | ||
db56246c MF |
1542 | /* |
1543 | * We have to make sure that the total write passed in | |
1544 | * doesn't extend past a single cluster. | |
1545 | */ | |
1546 | local_len = len; | |
1547 | cluster_off = pos & (osb->s_clustersize - 1); | |
1548 | if ((cluster_off + local_len) > osb->s_clustersize) | |
1549 | local_len = osb->s_clustersize - cluster_off; | |
1550 | ||
b27b7cbc | 1551 | ret = ocfs2_write_cluster(mapping, desc->c_phys, |
e7432675 SM |
1552 | desc->c_unwritten, |
1553 | desc->c_needs_zero, | |
1554 | data_ac, meta_ac, | |
db56246c | 1555 | wc, desc->c_cpos, pos, local_len); |
0d172baa MF |
1556 | if (ret) { |
1557 | mlog_errno(ret); | |
1558 | goto out; | |
1559 | } | |
db56246c MF |
1560 | |
1561 | len -= local_len; | |
1562 | pos += local_len; | |
0d172baa MF |
1563 | } |
1564 | ||
1565 | ret = 0; | |
1566 | out: | |
1567 | return ret; | |
1568 | } | |
1569 | ||
3a307ffc MF |
1570 | /* |
1571 | * ocfs2_write_end() wants to know which parts of the target page it | |
1572 | * should complete the write on. It's easiest to compute them ahead of | |
1573 | * time when a more complete view of the write is available. | |
1574 | */ | |
1575 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, | |
1576 | struct ocfs2_write_ctxt *wc, | |
1577 | loff_t pos, unsigned len, int alloc) | |
9517bac6 | 1578 | { |
3a307ffc | 1579 | struct ocfs2_write_cluster_desc *desc; |
9517bac6 | 1580 | |
3a307ffc MF |
1581 | wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1); |
1582 | wc->w_target_to = wc->w_target_from + len; | |
1583 | ||
1584 | if (alloc == 0) | |
1585 | return; | |
1586 | ||
1587 | /* | |
1588 | * Allocating write - we may have different boundaries based | |
1589 | * on page size and cluster size. | |
1590 | * | |
1591 | * NOTE: We can no longer compute one value from the other as | |
1592 | * the actual write length and user provided length may be | |
1593 | * different. | |
1594 | */ | |
9517bac6 | 1595 | |
3a307ffc MF |
1596 | if (wc->w_large_pages) { |
1597 | /* | |
1598 | * We only care about the 1st and last cluster within | |
b27b7cbc | 1599 | * our range and whether they should be zero'd or not. Either |
3a307ffc MF |
1600 | * value may be extended out to the start/end of a |
1601 | * newly allocated cluster. | |
1602 | */ | |
1603 | desc = &wc->w_desc[0]; | |
e7432675 | 1604 | if (desc->c_needs_zero) |
3a307ffc MF |
1605 | ocfs2_figure_cluster_boundaries(osb, |
1606 | desc->c_cpos, | |
1607 | &wc->w_target_from, | |
1608 | NULL); | |
1609 | ||
1610 | desc = &wc->w_desc[wc->w_clen - 1]; | |
e7432675 | 1611 | if (desc->c_needs_zero) |
3a307ffc MF |
1612 | ocfs2_figure_cluster_boundaries(osb, |
1613 | desc->c_cpos, | |
1614 | NULL, | |
1615 | &wc->w_target_to); | |
1616 | } else { | |
1617 | wc->w_target_from = 0; | |
1618 | wc->w_target_to = PAGE_CACHE_SIZE; | |
1619 | } | |
9517bac6 MF |
1620 | } |
1621 | ||
0d172baa MF |
1622 | /* |
1623 | * Populate each single-cluster write descriptor in the write context | |
1624 | * with information about the i/o to be done. | |
b27b7cbc MF |
1625 | * |
1626 | * Returns the number of clusters that will have to be allocated, as | |
1627 | * well as a worst case estimate of the number of extent records that | |
1628 | * would have to be created during a write to an unwritten region. | |
0d172baa MF |
1629 | */ |
1630 | static int ocfs2_populate_write_desc(struct inode *inode, | |
1631 | struct ocfs2_write_ctxt *wc, | |
b27b7cbc MF |
1632 | unsigned int *clusters_to_alloc, |
1633 | unsigned int *extents_to_split) | |
9517bac6 | 1634 | { |
0d172baa | 1635 | int ret; |
3a307ffc | 1636 | struct ocfs2_write_cluster_desc *desc; |
0d172baa | 1637 | unsigned int num_clusters = 0; |
b27b7cbc | 1638 | unsigned int ext_flags = 0; |
0d172baa MF |
1639 | u32 phys = 0; |
1640 | int i; | |
9517bac6 | 1641 | |
b27b7cbc MF |
1642 | *clusters_to_alloc = 0; |
1643 | *extents_to_split = 0; | |
1644 | ||
3a307ffc MF |
1645 | for (i = 0; i < wc->w_clen; i++) { |
1646 | desc = &wc->w_desc[i]; | |
1647 | desc->c_cpos = wc->w_cpos + i; | |
1648 | ||
1649 | if (num_clusters == 0) { | |
b27b7cbc MF |
1650 | /* |
1651 | * Need to look up the next extent record. | |
1652 | */ | |
3a307ffc | 1653 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, |
b27b7cbc | 1654 | &num_clusters, &ext_flags); |
3a307ffc MF |
1655 | if (ret) { |
1656 | mlog_errno(ret); | |
607d44aa | 1657 | goto out; |
3a307ffc | 1658 | } |
b27b7cbc | 1659 | |
293b2f70 TM |
1660 | /* We should already CoW the refcountd extent. */ |
1661 | BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED); | |
1662 | ||
b27b7cbc MF |
1663 | /* |
1664 | * Assume worst case - that we're writing in | |
1665 | * the middle of the extent. | |
1666 | * | |
1667 | * We can assume that the write proceeds from | |
1668 | * left to right, in which case the extent | |
1669 | * insert code is smart enough to coalesce the | |
1670 | * next splits into the previous records created. | |
1671 | */ | |
1672 | if (ext_flags & OCFS2_EXT_UNWRITTEN) | |
1673 | *extents_to_split = *extents_to_split + 2; | |
3a307ffc MF |
1674 | } else if (phys) { |
1675 | /* | |
1676 | * Only increment phys if it doesn't describe | |
1677 | * a hole. | |
1678 | */ | |
1679 | phys++; | |
1680 | } | |
1681 | ||
e7432675 SM |
1682 | /* |
1683 | * If w_first_new_cpos is < UINT_MAX, we have a non-sparse | |
1684 | * file that got extended. w_first_new_cpos tells us | |
1685 | * where the newly allocated clusters are so we can | |
1686 | * zero them. | |
1687 | */ | |
1688 | if (desc->c_cpos >= wc->w_first_new_cpos) { | |
1689 | BUG_ON(phys == 0); | |
1690 | desc->c_needs_zero = 1; | |
1691 | } | |
1692 | ||
3a307ffc MF |
1693 | desc->c_phys = phys; |
1694 | if (phys == 0) { | |
1695 | desc->c_new = 1; | |
e7432675 | 1696 | desc->c_needs_zero = 1; |
0d172baa | 1697 | *clusters_to_alloc = *clusters_to_alloc + 1; |
3a307ffc | 1698 | } |
e7432675 SM |
1699 | |
1700 | if (ext_flags & OCFS2_EXT_UNWRITTEN) { | |
b27b7cbc | 1701 | desc->c_unwritten = 1; |
e7432675 SM |
1702 | desc->c_needs_zero = 1; |
1703 | } | |
3a307ffc MF |
1704 | |
1705 | num_clusters--; | |
9517bac6 MF |
1706 | } |
1707 | ||
0d172baa MF |
1708 | ret = 0; |
1709 | out: | |
1710 | return ret; | |
1711 | } | |
1712 | ||
1afc32b9 MF |
1713 | static int ocfs2_write_begin_inline(struct address_space *mapping, |
1714 | struct inode *inode, | |
1715 | struct ocfs2_write_ctxt *wc) | |
1716 | { | |
1717 | int ret; | |
1718 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1719 | struct page *page; | |
1720 | handle_t *handle; | |
1721 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1722 | ||
f775da2f JB |
1723 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
1724 | if (IS_ERR(handle)) { | |
1725 | ret = PTR_ERR(handle); | |
1726 | mlog_errno(ret); | |
1727 | goto out; | |
1728 | } | |
1729 | ||
1afc32b9 MF |
1730 | page = find_or_create_page(mapping, 0, GFP_NOFS); |
1731 | if (!page) { | |
f775da2f | 1732 | ocfs2_commit_trans(osb, handle); |
1afc32b9 MF |
1733 | ret = -ENOMEM; |
1734 | mlog_errno(ret); | |
1735 | goto out; | |
1736 | } | |
1737 | /* | |
1738 | * If we don't set w_num_pages then this page won't get unlocked | |
1739 | * and freed on cleanup of the write context. | |
1740 | */ | |
1741 | wc->w_pages[0] = wc->w_target_page = page; | |
1742 | wc->w_num_pages = 1; | |
1743 | ||
0cf2f763 | 1744 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 1745 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
1746 | if (ret) { |
1747 | ocfs2_commit_trans(osb, handle); | |
1748 | ||
1749 | mlog_errno(ret); | |
1750 | goto out; | |
1751 | } | |
1752 | ||
1753 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
1754 | ocfs2_set_inode_data_inline(inode, di); | |
1755 | ||
1756 | if (!PageUptodate(page)) { | |
1757 | ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh); | |
1758 | if (ret) { | |
1759 | ocfs2_commit_trans(osb, handle); | |
1760 | ||
1761 | goto out; | |
1762 | } | |
1763 | } | |
1764 | ||
1765 | wc->w_handle = handle; | |
1766 | out: | |
1767 | return ret; | |
1768 | } | |
1769 | ||
1770 | int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size) | |
1771 | { | |
1772 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1773 | ||
0d8a4e0c | 1774 | if (new_size <= le16_to_cpu(di->id2.i_data.id_count)) |
1afc32b9 MF |
1775 | return 1; |
1776 | return 0; | |
1777 | } | |
1778 | ||
1779 | static int ocfs2_try_to_write_inline_data(struct address_space *mapping, | |
1780 | struct inode *inode, loff_t pos, | |
1781 | unsigned len, struct page *mmap_page, | |
1782 | struct ocfs2_write_ctxt *wc) | |
1783 | { | |
1784 | int ret, written = 0; | |
1785 | loff_t end = pos + len; | |
1786 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
d9ae49d6 | 1787 | struct ocfs2_dinode *di = NULL; |
1afc32b9 | 1788 | |
9558156b TM |
1789 | trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno, |
1790 | len, (unsigned long long)pos, | |
1791 | oi->ip_dyn_features); | |
1afc32b9 MF |
1792 | |
1793 | /* | |
1794 | * Handle inodes which already have inline data 1st. | |
1795 | */ | |
1796 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1797 | if (mmap_page == NULL && | |
1798 | ocfs2_size_fits_inline_data(wc->w_di_bh, end)) | |
1799 | goto do_inline_write; | |
1800 | ||
1801 | /* | |
1802 | * The write won't fit - we have to give this inode an | |
1803 | * inline extent list now. | |
1804 | */ | |
1805 | ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh); | |
1806 | if (ret) | |
1807 | mlog_errno(ret); | |
1808 | goto out; | |
1809 | } | |
1810 | ||
1811 | /* | |
1812 | * Check whether the inode can accept inline data. | |
1813 | */ | |
1814 | if (oi->ip_clusters != 0 || i_size_read(inode) != 0) | |
1815 | return 0; | |
1816 | ||
1817 | /* | |
1818 | * Check whether the write can fit. | |
1819 | */ | |
d9ae49d6 TY |
1820 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; |
1821 | if (mmap_page || | |
1822 | end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) | |
1afc32b9 MF |
1823 | return 0; |
1824 | ||
1825 | do_inline_write: | |
1826 | ret = ocfs2_write_begin_inline(mapping, inode, wc); | |
1827 | if (ret) { | |
1828 | mlog_errno(ret); | |
1829 | goto out; | |
1830 | } | |
1831 | ||
1832 | /* | |
1833 | * This signals to the caller that the data can be written | |
1834 | * inline. | |
1835 | */ | |
1836 | written = 1; | |
1837 | out: | |
1838 | return written ? written : ret; | |
1839 | } | |
1840 | ||
65ed39d6 MF |
1841 | /* |
1842 | * This function only does anything for file systems which can't | |
1843 | * handle sparse files. | |
1844 | * | |
1845 | * What we want to do here is fill in any hole between the current end | |
1846 | * of allocation and the end of our write. That way the rest of the | |
1847 | * write path can treat it as an non-allocating write, which has no | |
1848 | * special case code for sparse/nonsparse files. | |
1849 | */ | |
5693486b JB |
1850 | static int ocfs2_expand_nonsparse_inode(struct inode *inode, |
1851 | struct buffer_head *di_bh, | |
1852 | loff_t pos, unsigned len, | |
65ed39d6 MF |
1853 | struct ocfs2_write_ctxt *wc) |
1854 | { | |
1855 | int ret; | |
65ed39d6 MF |
1856 | loff_t newsize = pos + len; |
1857 | ||
5693486b | 1858 | BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); |
65ed39d6 MF |
1859 | |
1860 | if (newsize <= i_size_read(inode)) | |
1861 | return 0; | |
1862 | ||
5693486b | 1863 | ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos); |
65ed39d6 MF |
1864 | if (ret) |
1865 | mlog_errno(ret); | |
1866 | ||
e7432675 SM |
1867 | wc->w_first_new_cpos = |
1868 | ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)); | |
1869 | ||
65ed39d6 MF |
1870 | return ret; |
1871 | } | |
1872 | ||
5693486b JB |
1873 | static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh, |
1874 | loff_t pos) | |
1875 | { | |
1876 | int ret = 0; | |
1877 | ||
1878 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); | |
1879 | if (pos > i_size_read(inode)) | |
1880 | ret = ocfs2_zero_extend(inode, di_bh, pos); | |
1881 | ||
1882 | return ret; | |
1883 | } | |
1884 | ||
50308d81 TM |
1885 | /* |
1886 | * Try to flush truncate logs if we can free enough clusters from it. | |
1887 | * As for return value, "< 0" means error, "0" no space and "1" means | |
1888 | * we have freed enough spaces and let the caller try to allocate again. | |
1889 | */ | |
1890 | static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb, | |
1891 | unsigned int needed) | |
1892 | { | |
1893 | tid_t target; | |
1894 | int ret = 0; | |
1895 | unsigned int truncated_clusters; | |
1896 | ||
1897 | mutex_lock(&osb->osb_tl_inode->i_mutex); | |
1898 | truncated_clusters = osb->truncated_clusters; | |
1899 | mutex_unlock(&osb->osb_tl_inode->i_mutex); | |
1900 | ||
1901 | /* | |
1902 | * Check whether we can succeed in allocating if we free | |
1903 | * the truncate log. | |
1904 | */ | |
1905 | if (truncated_clusters < needed) | |
1906 | goto out; | |
1907 | ||
1908 | ret = ocfs2_flush_truncate_log(osb); | |
1909 | if (ret) { | |
1910 | mlog_errno(ret); | |
1911 | goto out; | |
1912 | } | |
1913 | ||
1914 | if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) { | |
1915 | jbd2_log_wait_commit(osb->journal->j_journal, target); | |
1916 | ret = 1; | |
1917 | } | |
1918 | out: | |
1919 | return ret; | |
1920 | } | |
1921 | ||
0378da0f TM |
1922 | int ocfs2_write_begin_nolock(struct file *filp, |
1923 | struct address_space *mapping, | |
0d172baa MF |
1924 | loff_t pos, unsigned len, unsigned flags, |
1925 | struct page **pagep, void **fsdata, | |
1926 | struct buffer_head *di_bh, struct page *mmap_page) | |
1927 | { | |
e7432675 | 1928 | int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS; |
50308d81 | 1929 | unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0; |
0d172baa MF |
1930 | struct ocfs2_write_ctxt *wc; |
1931 | struct inode *inode = mapping->host; | |
1932 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1933 | struct ocfs2_dinode *di; | |
1934 | struct ocfs2_alloc_context *data_ac = NULL; | |
1935 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1936 | handle_t *handle; | |
f99b9b7c | 1937 | struct ocfs2_extent_tree et; |
50308d81 | 1938 | int try_free = 1, ret1; |
0d172baa | 1939 | |
50308d81 | 1940 | try_again: |
0d172baa MF |
1941 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh); |
1942 | if (ret) { | |
1943 | mlog_errno(ret); | |
1944 | return ret; | |
1945 | } | |
1946 | ||
1afc32b9 MF |
1947 | if (ocfs2_supports_inline_data(osb)) { |
1948 | ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len, | |
1949 | mmap_page, wc); | |
1950 | if (ret == 1) { | |
1951 | ret = 0; | |
1952 | goto success; | |
1953 | } | |
1954 | if (ret < 0) { | |
1955 | mlog_errno(ret); | |
1956 | goto out; | |
1957 | } | |
1958 | } | |
1959 | ||
5693486b JB |
1960 | if (ocfs2_sparse_alloc(osb)) |
1961 | ret = ocfs2_zero_tail(inode, di_bh, pos); | |
1962 | else | |
1963 | ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len, | |
1964 | wc); | |
65ed39d6 MF |
1965 | if (ret) { |
1966 | mlog_errno(ret); | |
1967 | goto out; | |
1968 | } | |
1969 | ||
293b2f70 TM |
1970 | ret = ocfs2_check_range_for_refcount(inode, pos, len); |
1971 | if (ret < 0) { | |
1972 | mlog_errno(ret); | |
1973 | goto out; | |
1974 | } else if (ret == 1) { | |
50308d81 | 1975 | clusters_need = wc->w_clen; |
c7dd3392 | 1976 | ret = ocfs2_refcount_cow(inode, di_bh, |
37f8a2bf | 1977 | wc->w_cpos, wc->w_clen, UINT_MAX); |
293b2f70 TM |
1978 | if (ret) { |
1979 | mlog_errno(ret); | |
1980 | goto out; | |
1981 | } | |
1982 | } | |
1983 | ||
b27b7cbc MF |
1984 | ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc, |
1985 | &extents_to_split); | |
0d172baa MF |
1986 | if (ret) { |
1987 | mlog_errno(ret); | |
1988 | goto out; | |
1989 | } | |
50308d81 | 1990 | clusters_need += clusters_to_alloc; |
0d172baa MF |
1991 | |
1992 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1993 | ||
9558156b TM |
1994 | trace_ocfs2_write_begin_nolock( |
1995 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1996 | (long long)i_size_read(inode), | |
1997 | le32_to_cpu(di->i_clusters), | |
1998 | pos, len, flags, mmap_page, | |
1999 | clusters_to_alloc, extents_to_split); | |
2000 | ||
3a307ffc MF |
2001 | /* |
2002 | * We set w_target_from, w_target_to here so that | |
2003 | * ocfs2_write_end() knows which range in the target page to | |
2004 | * write out. An allocation requires that we write the entire | |
2005 | * cluster range. | |
2006 | */ | |
b27b7cbc | 2007 | if (clusters_to_alloc || extents_to_split) { |
3a307ffc MF |
2008 | /* |
2009 | * XXX: We are stretching the limits of | |
b27b7cbc | 2010 | * ocfs2_lock_allocators(). It greatly over-estimates |
3a307ffc MF |
2011 | * the work to be done. |
2012 | */ | |
5e404e9e JB |
2013 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
2014 | wc->w_di_bh); | |
f99b9b7c | 2015 | ret = ocfs2_lock_allocators(inode, &et, |
231b87d1 | 2016 | clusters_to_alloc, extents_to_split, |
f99b9b7c | 2017 | &data_ac, &meta_ac); |
9517bac6 MF |
2018 | if (ret) { |
2019 | mlog_errno(ret); | |
607d44aa | 2020 | goto out; |
9517bac6 MF |
2021 | } |
2022 | ||
4fe370af MF |
2023 | if (data_ac) |
2024 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; | |
2025 | ||
811f933d | 2026 | credits = ocfs2_calc_extend_credits(inode->i_sb, |
06f9da6e | 2027 | &di->id2.i_list); |
3a307ffc | 2028 | |
9517bac6 MF |
2029 | } |
2030 | ||
e7432675 SM |
2031 | /* |
2032 | * We have to zero sparse allocated clusters, unwritten extent clusters, | |
2033 | * and non-sparse clusters we just extended. For non-sparse writes, | |
2034 | * we know zeros will only be needed in the first and/or last cluster. | |
2035 | */ | |
2036 | if (clusters_to_alloc || extents_to_split || | |
8379e7c4 SM |
2037 | (wc->w_clen && (wc->w_desc[0].c_needs_zero || |
2038 | wc->w_desc[wc->w_clen - 1].c_needs_zero))) | |
e7432675 SM |
2039 | cluster_of_pages = 1; |
2040 | else | |
2041 | cluster_of_pages = 0; | |
2042 | ||
2043 | ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages); | |
3a307ffc | 2044 | |
9517bac6 MF |
2045 | handle = ocfs2_start_trans(osb, credits); |
2046 | if (IS_ERR(handle)) { | |
2047 | ret = PTR_ERR(handle); | |
2048 | mlog_errno(ret); | |
607d44aa | 2049 | goto out; |
9517bac6 MF |
2050 | } |
2051 | ||
3a307ffc MF |
2052 | wc->w_handle = handle; |
2053 | ||
5dd4056d CH |
2054 | if (clusters_to_alloc) { |
2055 | ret = dquot_alloc_space_nodirty(inode, | |
2056 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); | |
2057 | if (ret) | |
2058 | goto out_commit; | |
a90714c1 | 2059 | } |
3a307ffc MF |
2060 | /* |
2061 | * We don't want this to fail in ocfs2_write_end(), so do it | |
2062 | * here. | |
2063 | */ | |
0cf2f763 | 2064 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 2065 | OCFS2_JOURNAL_ACCESS_WRITE); |
3a307ffc | 2066 | if (ret) { |
9517bac6 | 2067 | mlog_errno(ret); |
a90714c1 | 2068 | goto out_quota; |
9517bac6 MF |
2069 | } |
2070 | ||
3a307ffc MF |
2071 | /* |
2072 | * Fill our page array first. That way we've grabbed enough so | |
2073 | * that we can zero and flush if we error after adding the | |
2074 | * extent. | |
2075 | */ | |
693c241a | 2076 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len, |
e7432675 | 2077 | cluster_of_pages, mmap_page); |
5cffff9e | 2078 | if (ret && ret != -EAGAIN) { |
9517bac6 | 2079 | mlog_errno(ret); |
a90714c1 | 2080 | goto out_quota; |
9517bac6 MF |
2081 | } |
2082 | ||
5cffff9e WW |
2083 | /* |
2084 | * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock | |
2085 | * the target page. In this case, we exit with no error and no target | |
2086 | * page. This will trigger the caller, page_mkwrite(), to re-try | |
2087 | * the operation. | |
2088 | */ | |
2089 | if (ret == -EAGAIN) { | |
2090 | BUG_ON(wc->w_target_page); | |
2091 | ret = 0; | |
2092 | goto out_quota; | |
2093 | } | |
2094 | ||
0d172baa MF |
2095 | ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos, |
2096 | len); | |
2097 | if (ret) { | |
2098 | mlog_errno(ret); | |
a90714c1 | 2099 | goto out_quota; |
9517bac6 | 2100 | } |
9517bac6 | 2101 | |
3a307ffc MF |
2102 | if (data_ac) |
2103 | ocfs2_free_alloc_context(data_ac); | |
2104 | if (meta_ac) | |
2105 | ocfs2_free_alloc_context(meta_ac); | |
9517bac6 | 2106 | |
1afc32b9 | 2107 | success: |
3a307ffc MF |
2108 | *pagep = wc->w_target_page; |
2109 | *fsdata = wc; | |
2110 | return 0; | |
a90714c1 JK |
2111 | out_quota: |
2112 | if (clusters_to_alloc) | |
5dd4056d | 2113 | dquot_free_space(inode, |
a90714c1 | 2114 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); |
9517bac6 MF |
2115 | out_commit: |
2116 | ocfs2_commit_trans(osb, handle); | |
2117 | ||
9517bac6 | 2118 | out: |
3a307ffc MF |
2119 | ocfs2_free_write_ctxt(wc); |
2120 | ||
b1214e47 | 2121 | if (data_ac) { |
9517bac6 | 2122 | ocfs2_free_alloc_context(data_ac); |
b1214e47 X |
2123 | data_ac = NULL; |
2124 | } | |
2125 | if (meta_ac) { | |
9517bac6 | 2126 | ocfs2_free_alloc_context(meta_ac); |
b1214e47 X |
2127 | meta_ac = NULL; |
2128 | } | |
50308d81 TM |
2129 | |
2130 | if (ret == -ENOSPC && try_free) { | |
2131 | /* | |
2132 | * Try to free some truncate log so that we can have enough | |
2133 | * clusters to allocate. | |
2134 | */ | |
2135 | try_free = 0; | |
2136 | ||
2137 | ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); | |
2138 | if (ret1 == 1) | |
2139 | goto try_again; | |
2140 | ||
2141 | if (ret1 < 0) | |
2142 | mlog_errno(ret1); | |
2143 | } | |
2144 | ||
3a307ffc MF |
2145 | return ret; |
2146 | } | |
2147 | ||
b6af1bcd NP |
2148 | static int ocfs2_write_begin(struct file *file, struct address_space *mapping, |
2149 | loff_t pos, unsigned len, unsigned flags, | |
2150 | struct page **pagep, void **fsdata) | |
607d44aa MF |
2151 | { |
2152 | int ret; | |
2153 | struct buffer_head *di_bh = NULL; | |
2154 | struct inode *inode = mapping->host; | |
2155 | ||
e63aecb6 | 2156 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
607d44aa MF |
2157 | if (ret) { |
2158 | mlog_errno(ret); | |
2159 | return ret; | |
2160 | } | |
2161 | ||
2162 | /* | |
2163 | * Take alloc sem here to prevent concurrent lookups. That way | |
2164 | * the mapping, zeroing and tree manipulation within | |
2165 | * ocfs2_write() will be safe against ->readpage(). This | |
2166 | * should also serve to lock out allocation from a shared | |
2167 | * writeable region. | |
2168 | */ | |
2169 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
2170 | ||
0378da0f | 2171 | ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep, |
7307de80 | 2172 | fsdata, di_bh, NULL); |
607d44aa MF |
2173 | if (ret) { |
2174 | mlog_errno(ret); | |
c934a92d | 2175 | goto out_fail; |
607d44aa MF |
2176 | } |
2177 | ||
2178 | brelse(di_bh); | |
2179 | ||
2180 | return 0; | |
2181 | ||
607d44aa MF |
2182 | out_fail: |
2183 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
2184 | ||
2185 | brelse(di_bh); | |
e63aecb6 | 2186 | ocfs2_inode_unlock(inode, 1); |
607d44aa MF |
2187 | |
2188 | return ret; | |
2189 | } | |
2190 | ||
1afc32b9 MF |
2191 | static void ocfs2_write_end_inline(struct inode *inode, loff_t pos, |
2192 | unsigned len, unsigned *copied, | |
2193 | struct ocfs2_dinode *di, | |
2194 | struct ocfs2_write_ctxt *wc) | |
2195 | { | |
2196 | void *kaddr; | |
2197 | ||
2198 | if (unlikely(*copied < len)) { | |
2199 | if (!PageUptodate(wc->w_target_page)) { | |
2200 | *copied = 0; | |
2201 | return; | |
2202 | } | |
2203 | } | |
2204 | ||
c4bc8dcb | 2205 | kaddr = kmap_atomic(wc->w_target_page); |
1afc32b9 | 2206 | memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied); |
c4bc8dcb | 2207 | kunmap_atomic(kaddr); |
1afc32b9 | 2208 | |
9558156b TM |
2209 | trace_ocfs2_write_end_inline( |
2210 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1afc32b9 MF |
2211 | (unsigned long long)pos, *copied, |
2212 | le16_to_cpu(di->id2.i_data.id_count), | |
2213 | le16_to_cpu(di->i_dyn_features)); | |
2214 | } | |
2215 | ||
7307de80 MF |
2216 | int ocfs2_write_end_nolock(struct address_space *mapping, |
2217 | loff_t pos, unsigned len, unsigned copied, | |
2218 | struct page *page, void *fsdata) | |
3a307ffc MF |
2219 | { |
2220 | int i; | |
2221 | unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1); | |
2222 | struct inode *inode = mapping->host; | |
2223 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2224 | struct ocfs2_write_ctxt *wc = fsdata; | |
2225 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
2226 | handle_t *handle = wc->w_handle; | |
2227 | struct page *tmppage; | |
2228 | ||
1afc32b9 MF |
2229 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
2230 | ocfs2_write_end_inline(inode, pos, len, &copied, di, wc); | |
2231 | goto out_write_size; | |
2232 | } | |
2233 | ||
3a307ffc MF |
2234 | if (unlikely(copied < len)) { |
2235 | if (!PageUptodate(wc->w_target_page)) | |
2236 | copied = 0; | |
2237 | ||
2238 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, | |
2239 | start+len); | |
2240 | } | |
2241 | flush_dcache_page(wc->w_target_page); | |
2242 | ||
2243 | for(i = 0; i < wc->w_num_pages; i++) { | |
2244 | tmppage = wc->w_pages[i]; | |
2245 | ||
2246 | if (tmppage == wc->w_target_page) { | |
2247 | from = wc->w_target_from; | |
2248 | to = wc->w_target_to; | |
2249 | ||
2250 | BUG_ON(from > PAGE_CACHE_SIZE || | |
2251 | to > PAGE_CACHE_SIZE || | |
2252 | to < from); | |
2253 | } else { | |
2254 | /* | |
2255 | * Pages adjacent to the target (if any) imply | |
2256 | * a hole-filling write in which case we want | |
2257 | * to flush their entire range. | |
2258 | */ | |
2259 | from = 0; | |
2260 | to = PAGE_CACHE_SIZE; | |
2261 | } | |
2262 | ||
961cecbe | 2263 | if (page_has_buffers(tmppage)) { |
53ef99ca | 2264 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 2265 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
2266 | block_commit_write(tmppage, from, to); |
2267 | } | |
3a307ffc MF |
2268 | } |
2269 | ||
1afc32b9 | 2270 | out_write_size: |
3a307ffc | 2271 | pos += copied; |
f17c20dd | 2272 | if (pos > i_size_read(inode)) { |
3a307ffc MF |
2273 | i_size_write(inode, pos); |
2274 | mark_inode_dirty(inode); | |
2275 | } | |
2276 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
2277 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
2278 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
2279 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); | |
2280 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
2931cdcb | 2281 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
3a307ffc MF |
2282 | ocfs2_journal_dirty(handle, wc->w_di_bh); |
2283 | ||
136f49b9 JB |
2284 | /* unlock pages before dealloc since it needs acquiring j_trans_barrier |
2285 | * lock, or it will cause a deadlock since journal commit threads holds | |
2286 | * this lock and will ask for the page lock when flushing the data. | |
2287 | * put it here to preserve the unlock order. | |
2288 | */ | |
2289 | ocfs2_unlock_pages(wc); | |
2290 | ||
3a307ffc | 2291 | ocfs2_commit_trans(osb, handle); |
59a5e416 | 2292 | |
b27b7cbc MF |
2293 | ocfs2_run_deallocs(osb, &wc->w_dealloc); |
2294 | ||
136f49b9 JB |
2295 | brelse(wc->w_di_bh); |
2296 | kfree(wc); | |
607d44aa MF |
2297 | |
2298 | return copied; | |
2299 | } | |
2300 | ||
b6af1bcd NP |
2301 | static int ocfs2_write_end(struct file *file, struct address_space *mapping, |
2302 | loff_t pos, unsigned len, unsigned copied, | |
2303 | struct page *page, void *fsdata) | |
607d44aa MF |
2304 | { |
2305 | int ret; | |
2306 | struct inode *inode = mapping->host; | |
2307 | ||
2308 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata); | |
2309 | ||
3a307ffc | 2310 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
e63aecb6 | 2311 | ocfs2_inode_unlock(inode, 1); |
9517bac6 | 2312 | |
607d44aa | 2313 | return ret; |
9517bac6 MF |
2314 | } |
2315 | ||
f5e54d6e | 2316 | const struct address_space_operations ocfs2_aops = { |
1fca3a05 HH |
2317 | .readpage = ocfs2_readpage, |
2318 | .readpages = ocfs2_readpages, | |
2319 | .writepage = ocfs2_writepage, | |
2320 | .write_begin = ocfs2_write_begin, | |
2321 | .write_end = ocfs2_write_end, | |
2322 | .bmap = ocfs2_bmap, | |
1fca3a05 | 2323 | .direct_IO = ocfs2_direct_IO, |
41ecc345 | 2324 | .invalidatepage = block_invalidatepage, |
1fca3a05 HH |
2325 | .releasepage = ocfs2_releasepage, |
2326 | .migratepage = buffer_migrate_page, | |
2327 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 2328 | .error_remove_page = generic_error_remove_page, |
ccd979bd | 2329 | }; |