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> |
e2e40f2c | 32 | #include <linux/uio.h> |
ccd979bd | 33 | |
ccd979bd MF |
34 | #include <cluster/masklog.h> |
35 | ||
36 | #include "ocfs2.h" | |
37 | ||
38 | #include "alloc.h" | |
39 | #include "aops.h" | |
40 | #include "dlmglue.h" | |
41 | #include "extent_map.h" | |
42 | #include "file.h" | |
43 | #include "inode.h" | |
44 | #include "journal.h" | |
9517bac6 | 45 | #include "suballoc.h" |
ccd979bd MF |
46 | #include "super.h" |
47 | #include "symlink.h" | |
293b2f70 | 48 | #include "refcounttree.h" |
9558156b | 49 | #include "ocfs2_trace.h" |
ccd979bd MF |
50 | |
51 | #include "buffer_head_io.h" | |
24c40b32 JQ |
52 | #include "dir.h" |
53 | #include "namei.h" | |
54 | #include "sysfile.h" | |
ccd979bd MF |
55 | |
56 | static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock, | |
57 | struct buffer_head *bh_result, int create) | |
58 | { | |
59 | int err = -EIO; | |
60 | int status; | |
61 | struct ocfs2_dinode *fe = NULL; | |
62 | struct buffer_head *bh = NULL; | |
63 | struct buffer_head *buffer_cache_bh = NULL; | |
64 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
65 | void *kaddr; | |
66 | ||
9558156b TM |
67 | trace_ocfs2_symlink_get_block( |
68 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
69 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
70 | |
71 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); | |
72 | ||
73 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { | |
74 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", | |
75 | (unsigned long long)iblock); | |
76 | goto bail; | |
77 | } | |
78 | ||
b657c95c | 79 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
80 | if (status < 0) { |
81 | mlog_errno(status); | |
82 | goto bail; | |
83 | } | |
84 | fe = (struct ocfs2_dinode *) bh->b_data; | |
85 | ||
ccd979bd MF |
86 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, |
87 | le32_to_cpu(fe->i_clusters))) { | |
7391a294 | 88 | err = -ENOMEM; |
ccd979bd MF |
89 | mlog(ML_ERROR, "block offset is outside the allocated size: " |
90 | "%llu\n", (unsigned long long)iblock); | |
91 | goto bail; | |
92 | } | |
93 | ||
94 | /* We don't use the page cache to create symlink data, so if | |
95 | * need be, copy it over from the buffer cache. */ | |
96 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { | |
97 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + | |
98 | iblock; | |
99 | buffer_cache_bh = sb_getblk(osb->sb, blkno); | |
100 | if (!buffer_cache_bh) { | |
7391a294 | 101 | err = -ENOMEM; |
ccd979bd MF |
102 | mlog(ML_ERROR, "couldn't getblock for symlink!\n"); |
103 | goto bail; | |
104 | } | |
105 | ||
106 | /* we haven't locked out transactions, so a commit | |
107 | * could've happened. Since we've got a reference on | |
108 | * the bh, even if it commits while we're doing the | |
109 | * copy, the data is still good. */ | |
110 | if (buffer_jbd(buffer_cache_bh) | |
111 | && ocfs2_inode_is_new(inode)) { | |
c4bc8dcb | 112 | kaddr = kmap_atomic(bh_result->b_page); |
ccd979bd MF |
113 | if (!kaddr) { |
114 | mlog(ML_ERROR, "couldn't kmap!\n"); | |
115 | goto bail; | |
116 | } | |
117 | memcpy(kaddr + (bh_result->b_size * iblock), | |
118 | buffer_cache_bh->b_data, | |
119 | bh_result->b_size); | |
c4bc8dcb | 120 | kunmap_atomic(kaddr); |
ccd979bd MF |
121 | set_buffer_uptodate(bh_result); |
122 | } | |
123 | brelse(buffer_cache_bh); | |
124 | } | |
125 | ||
126 | map_bh(bh_result, inode->i_sb, | |
127 | le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); | |
128 | ||
129 | err = 0; | |
130 | ||
131 | bail: | |
a81cb88b | 132 | brelse(bh); |
ccd979bd | 133 | |
ccd979bd MF |
134 | return err; |
135 | } | |
136 | ||
6f70fa51 TM |
137 | int ocfs2_get_block(struct inode *inode, sector_t iblock, |
138 | struct buffer_head *bh_result, int create) | |
ccd979bd MF |
139 | { |
140 | int err = 0; | |
49cb8d2d | 141 | unsigned int ext_flags; |
628a24f5 MF |
142 | u64 max_blocks = bh_result->b_size >> inode->i_blkbits; |
143 | u64 p_blkno, count, past_eof; | |
25baf2da | 144 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd | 145 | |
9558156b TM |
146 | trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno, |
147 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
148 | |
149 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) | |
150 | mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", | |
151 | inode, inode->i_ino); | |
152 | ||
153 | if (S_ISLNK(inode->i_mode)) { | |
154 | /* this always does I/O for some reason. */ | |
155 | err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); | |
156 | goto bail; | |
157 | } | |
158 | ||
628a24f5 | 159 | err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count, |
49cb8d2d | 160 | &ext_flags); |
ccd979bd MF |
161 | if (err) { |
162 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " | |
b0697053 MF |
163 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, |
164 | (unsigned long long)p_blkno); | |
ccd979bd MF |
165 | goto bail; |
166 | } | |
167 | ||
628a24f5 MF |
168 | if (max_blocks < count) |
169 | count = max_blocks; | |
170 | ||
25baf2da MF |
171 | /* |
172 | * ocfs2 never allocates in this function - the only time we | |
173 | * need to use BH_New is when we're extending i_size on a file | |
174 | * system which doesn't support holes, in which case BH_New | |
ebdec241 | 175 | * allows __block_write_begin() to zero. |
c0420ad2 CL |
176 | * |
177 | * If we see this on a sparse file system, then a truncate has | |
178 | * raced us and removed the cluster. In this case, we clear | |
179 | * the buffers dirty and uptodate bits and let the buffer code | |
180 | * ignore it as a hole. | |
25baf2da | 181 | */ |
c0420ad2 CL |
182 | if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) { |
183 | clear_buffer_dirty(bh_result); | |
184 | clear_buffer_uptodate(bh_result); | |
185 | goto bail; | |
186 | } | |
25baf2da | 187 | |
49cb8d2d MF |
188 | /* Treat the unwritten extent as a hole for zeroing purposes. */ |
189 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
25baf2da MF |
190 | map_bh(bh_result, inode->i_sb, p_blkno); |
191 | ||
628a24f5 MF |
192 | bh_result->b_size = count << inode->i_blkbits; |
193 | ||
25baf2da MF |
194 | if (!ocfs2_sparse_alloc(osb)) { |
195 | if (p_blkno == 0) { | |
196 | err = -EIO; | |
197 | mlog(ML_ERROR, | |
198 | "iblock = %llu p_blkno = %llu blkno=(%llu)\n", | |
199 | (unsigned long long)iblock, | |
200 | (unsigned long long)p_blkno, | |
201 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
202 | mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters); | |
203 | dump_stack(); | |
1f4cea37 | 204 | goto bail; |
25baf2da | 205 | } |
25baf2da | 206 | } |
ccd979bd | 207 | |
5693486b | 208 | past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
9558156b TM |
209 | |
210 | trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno, | |
211 | (unsigned long long)past_eof); | |
5693486b JB |
212 | if (create && (iblock >= past_eof)) |
213 | set_buffer_new(bh_result); | |
214 | ||
ccd979bd MF |
215 | bail: |
216 | if (err < 0) | |
217 | err = -EIO; | |
218 | ||
ccd979bd MF |
219 | return err; |
220 | } | |
221 | ||
1afc32b9 MF |
222 | int ocfs2_read_inline_data(struct inode *inode, struct page *page, |
223 | struct buffer_head *di_bh) | |
6798d35a MF |
224 | { |
225 | void *kaddr; | |
d2849fb2 | 226 | loff_t size; |
6798d35a MF |
227 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
228 | ||
229 | if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) { | |
230 | ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag", | |
231 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
232 | return -EROFS; | |
233 | } | |
234 | ||
235 | size = i_size_read(inode); | |
236 | ||
237 | if (size > PAGE_CACHE_SIZE || | |
d9ae49d6 | 238 | size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) { |
6798d35a | 239 | ocfs2_error(inode->i_sb, |
d2849fb2 JK |
240 | "Inode %llu has with inline data has bad size: %Lu", |
241 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
242 | (unsigned long long)size); | |
6798d35a MF |
243 | return -EROFS; |
244 | } | |
245 | ||
c4bc8dcb | 246 | kaddr = kmap_atomic(page); |
6798d35a MF |
247 | if (size) |
248 | memcpy(kaddr, di->id2.i_data.id_data, size); | |
249 | /* Clear the remaining part of the page */ | |
250 | memset(kaddr + size, 0, PAGE_CACHE_SIZE - size); | |
251 | flush_dcache_page(page); | |
c4bc8dcb | 252 | kunmap_atomic(kaddr); |
6798d35a MF |
253 | |
254 | SetPageUptodate(page); | |
255 | ||
256 | return 0; | |
257 | } | |
258 | ||
259 | static int ocfs2_readpage_inline(struct inode *inode, struct page *page) | |
260 | { | |
261 | int ret; | |
262 | struct buffer_head *di_bh = NULL; | |
6798d35a MF |
263 | |
264 | BUG_ON(!PageLocked(page)); | |
86c838b0 | 265 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)); |
6798d35a | 266 | |
b657c95c | 267 | ret = ocfs2_read_inode_block(inode, &di_bh); |
6798d35a MF |
268 | if (ret) { |
269 | mlog_errno(ret); | |
270 | goto out; | |
271 | } | |
272 | ||
273 | ret = ocfs2_read_inline_data(inode, page, di_bh); | |
274 | out: | |
275 | unlock_page(page); | |
276 | ||
277 | brelse(di_bh); | |
278 | return ret; | |
279 | } | |
280 | ||
ccd979bd MF |
281 | static int ocfs2_readpage(struct file *file, struct page *page) |
282 | { | |
283 | struct inode *inode = page->mapping->host; | |
6798d35a | 284 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd MF |
285 | loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT; |
286 | int ret, unlock = 1; | |
287 | ||
9558156b TM |
288 | trace_ocfs2_readpage((unsigned long long)oi->ip_blkno, |
289 | (page ? page->index : 0)); | |
ccd979bd | 290 | |
e63aecb6 | 291 | ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page); |
ccd979bd MF |
292 | if (ret != 0) { |
293 | if (ret == AOP_TRUNCATED_PAGE) | |
294 | unlock = 0; | |
295 | mlog_errno(ret); | |
296 | goto out; | |
297 | } | |
298 | ||
6798d35a | 299 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { |
c7e25e6e JK |
300 | /* |
301 | * Unlock the page and cycle ip_alloc_sem so that we don't | |
302 | * busyloop waiting for ip_alloc_sem to unlock | |
303 | */ | |
e9dfc0b2 | 304 | ret = AOP_TRUNCATED_PAGE; |
c7e25e6e JK |
305 | unlock_page(page); |
306 | unlock = 0; | |
307 | down_read(&oi->ip_alloc_sem); | |
308 | up_read(&oi->ip_alloc_sem); | |
e63aecb6 | 309 | goto out_inode_unlock; |
e9dfc0b2 | 310 | } |
ccd979bd MF |
311 | |
312 | /* | |
313 | * i_size might have just been updated as we grabed the meta lock. We | |
314 | * might now be discovering a truncate that hit on another node. | |
315 | * block_read_full_page->get_block freaks out if it is asked to read | |
316 | * beyond the end of a file, so we check here. Callers | |
54cb8821 | 317 | * (generic_file_read, vm_ops->fault) are clever enough to check i_size |
ccd979bd MF |
318 | * and notice that the page they just read isn't needed. |
319 | * | |
320 | * XXX sys_readahead() seems to get that wrong? | |
321 | */ | |
322 | if (start >= i_size_read(inode)) { | |
eebd2aa3 | 323 | zero_user(page, 0, PAGE_SIZE); |
ccd979bd MF |
324 | SetPageUptodate(page); |
325 | ret = 0; | |
326 | goto out_alloc; | |
327 | } | |
328 | ||
6798d35a MF |
329 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
330 | ret = ocfs2_readpage_inline(inode, page); | |
331 | else | |
332 | ret = block_read_full_page(page, ocfs2_get_block); | |
ccd979bd MF |
333 | unlock = 0; |
334 | ||
ccd979bd MF |
335 | out_alloc: |
336 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 MF |
337 | out_inode_unlock: |
338 | ocfs2_inode_unlock(inode, 0); | |
ccd979bd MF |
339 | out: |
340 | if (unlock) | |
341 | unlock_page(page); | |
ccd979bd MF |
342 | return ret; |
343 | } | |
344 | ||
628a24f5 MF |
345 | /* |
346 | * This is used only for read-ahead. Failures or difficult to handle | |
347 | * situations are safe to ignore. | |
348 | * | |
349 | * Right now, we don't bother with BH_Boundary - in-inode extent lists | |
350 | * are quite large (243 extents on 4k blocks), so most inodes don't | |
351 | * grow out to a tree. If need be, detecting boundary extents could | |
352 | * trivially be added in a future version of ocfs2_get_block(). | |
353 | */ | |
354 | static int ocfs2_readpages(struct file *filp, struct address_space *mapping, | |
355 | struct list_head *pages, unsigned nr_pages) | |
356 | { | |
357 | int ret, err = -EIO; | |
358 | struct inode *inode = mapping->host; | |
359 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
360 | loff_t start; | |
361 | struct page *last; | |
362 | ||
363 | /* | |
364 | * Use the nonblocking flag for the dlm code to avoid page | |
365 | * lock inversion, but don't bother with retrying. | |
366 | */ | |
367 | ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK); | |
368 | if (ret) | |
369 | return err; | |
370 | ||
371 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { | |
372 | ocfs2_inode_unlock(inode, 0); | |
373 | return err; | |
374 | } | |
375 | ||
376 | /* | |
377 | * Don't bother with inline-data. There isn't anything | |
378 | * to read-ahead in that case anyway... | |
379 | */ | |
380 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
381 | goto out_unlock; | |
382 | ||
383 | /* | |
384 | * Check whether a remote node truncated this file - we just | |
385 | * drop out in that case as it's not worth handling here. | |
386 | */ | |
387 | last = list_entry(pages->prev, struct page, lru); | |
388 | start = (loff_t)last->index << PAGE_CACHE_SHIFT; | |
389 | if (start >= i_size_read(inode)) | |
390 | goto out_unlock; | |
391 | ||
392 | err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block); | |
393 | ||
394 | out_unlock: | |
395 | up_read(&oi->ip_alloc_sem); | |
396 | ocfs2_inode_unlock(inode, 0); | |
397 | ||
398 | return err; | |
399 | } | |
400 | ||
ccd979bd MF |
401 | /* Note: Because we don't support holes, our allocation has |
402 | * already happened (allocation writes zeros to the file data) | |
403 | * so we don't have to worry about ordered writes in | |
404 | * ocfs2_writepage. | |
405 | * | |
406 | * ->writepage is called during the process of invalidating the page cache | |
407 | * during blocked lock processing. It can't block on any cluster locks | |
408 | * to during block mapping. It's relying on the fact that the block | |
409 | * mapping can't have disappeared under the dirty pages that it is | |
410 | * being asked to write back. | |
411 | */ | |
412 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) | |
413 | { | |
9558156b TM |
414 | trace_ocfs2_writepage( |
415 | (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno, | |
416 | page->index); | |
ccd979bd | 417 | |
9558156b | 418 | return block_write_full_page(page, ocfs2_get_block, wbc); |
ccd979bd MF |
419 | } |
420 | ||
ccd979bd MF |
421 | /* Taken from ext3. We don't necessarily need the full blown |
422 | * functionality yet, but IMHO it's better to cut and paste the whole | |
423 | * thing so we can avoid introducing our own bugs (and easily pick up | |
424 | * their fixes when they happen) --Mark */ | |
60b11392 MF |
425 | int walk_page_buffers( handle_t *handle, |
426 | struct buffer_head *head, | |
427 | unsigned from, | |
428 | unsigned to, | |
429 | int *partial, | |
430 | int (*fn)( handle_t *handle, | |
431 | struct buffer_head *bh)) | |
ccd979bd MF |
432 | { |
433 | struct buffer_head *bh; | |
434 | unsigned block_start, block_end; | |
435 | unsigned blocksize = head->b_size; | |
436 | int err, ret = 0; | |
437 | struct buffer_head *next; | |
438 | ||
439 | for ( bh = head, block_start = 0; | |
440 | ret == 0 && (bh != head || !block_start); | |
441 | block_start = block_end, bh = next) | |
442 | { | |
443 | next = bh->b_this_page; | |
444 | block_end = block_start + blocksize; | |
445 | if (block_end <= from || block_start >= to) { | |
446 | if (partial && !buffer_uptodate(bh)) | |
447 | *partial = 1; | |
448 | continue; | |
449 | } | |
450 | err = (*fn)(handle, bh); | |
451 | if (!ret) | |
452 | ret = err; | |
453 | } | |
454 | return ret; | |
455 | } | |
456 | ||
ccd979bd MF |
457 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) |
458 | { | |
459 | sector_t status; | |
460 | u64 p_blkno = 0; | |
461 | int err = 0; | |
462 | struct inode *inode = mapping->host; | |
463 | ||
9558156b TM |
464 | trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno, |
465 | (unsigned long long)block); | |
ccd979bd MF |
466 | |
467 | /* We don't need to lock journal system files, since they aren't | |
468 | * accessed concurrently from multiple nodes. | |
469 | */ | |
470 | if (!INODE_JOURNAL(inode)) { | |
e63aecb6 | 471 | err = ocfs2_inode_lock(inode, NULL, 0); |
ccd979bd MF |
472 | if (err) { |
473 | if (err != -ENOENT) | |
474 | mlog_errno(err); | |
475 | goto bail; | |
476 | } | |
477 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | |
478 | } | |
479 | ||
6798d35a MF |
480 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) |
481 | err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, | |
482 | NULL); | |
ccd979bd MF |
483 | |
484 | if (!INODE_JOURNAL(inode)) { | |
485 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 | 486 | ocfs2_inode_unlock(inode, 0); |
ccd979bd MF |
487 | } |
488 | ||
489 | if (err) { | |
490 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", | |
491 | (unsigned long long)block); | |
492 | mlog_errno(err); | |
493 | goto bail; | |
494 | } | |
495 | ||
ccd979bd MF |
496 | bail: |
497 | status = err ? 0 : p_blkno; | |
498 | ||
ccd979bd MF |
499 | return status; |
500 | } | |
501 | ||
502 | /* | |
503 | * TODO: Make this into a generic get_blocks function. | |
504 | * | |
505 | * From do_direct_io in direct-io.c: | |
506 | * "So what we do is to permit the ->get_blocks function to populate | |
507 | * bh.b_size with the size of IO which is permitted at this offset and | |
508 | * this i_blkbits." | |
509 | * | |
510 | * This function is called directly from get_more_blocks in direct-io.c. | |
511 | * | |
512 | * called like this: dio->get_blocks(dio->inode, fs_startblk, | |
513 | * fs_count, map_bh, dio->rw == WRITE); | |
514 | */ | |
515 | static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock, | |
ccd979bd MF |
516 | struct buffer_head *bh_result, int create) |
517 | { | |
518 | int ret; | |
49255dce JQ |
519 | u32 cpos = 0; |
520 | int alloc_locked = 0; | |
4f902c37 | 521 | u64 p_blkno, inode_blocks, contig_blocks; |
49cb8d2d | 522 | unsigned int ext_flags; |
184d7d20 | 523 | unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits; |
1d8fa7a2 | 524 | unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits; |
49255dce JQ |
525 | unsigned long len = bh_result->b_size; |
526 | unsigned int clusters_to_alloc = 0; | |
527 | ||
528 | cpos = ocfs2_blocks_to_clusters(inode->i_sb, iblock); | |
ccd979bd | 529 | |
ccd979bd MF |
530 | /* This function won't even be called if the request isn't all |
531 | * nicely aligned and of the right size, so there's no need | |
532 | * for us to check any of that. */ | |
533 | ||
25baf2da | 534 | inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
564f8a32 | 535 | |
ccd979bd MF |
536 | /* This figures out the size of the next contiguous block, and |
537 | * our logical offset */ | |
363041a5 | 538 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, |
49cb8d2d | 539 | &contig_blocks, &ext_flags); |
ccd979bd MF |
540 | if (ret) { |
541 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", | |
542 | (unsigned long long)iblock); | |
543 | ret = -EIO; | |
544 | goto bail; | |
545 | } | |
546 | ||
cbaee472 TM |
547 | /* We should already CoW the refcounted extent in case of create. */ |
548 | BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED)); | |
549 | ||
49255dce JQ |
550 | /* allocate blocks if no p_blkno is found, and create == 1 */ |
551 | if (!p_blkno && create) { | |
552 | ret = ocfs2_inode_lock(inode, NULL, 1); | |
553 | if (ret < 0) { | |
554 | mlog_errno(ret); | |
555 | goto bail; | |
556 | } | |
557 | ||
558 | alloc_locked = 1; | |
559 | ||
560 | /* fill hole, allocate blocks can't be larger than the size | |
561 | * of the hole */ | |
562 | clusters_to_alloc = ocfs2_clusters_for_bytes(inode->i_sb, len); | |
563 | if (clusters_to_alloc > contig_blocks) | |
564 | clusters_to_alloc = contig_blocks; | |
565 | ||
566 | /* allocate extent and insert them into the extent tree */ | |
567 | ret = ocfs2_extend_allocation(inode, cpos, | |
568 | clusters_to_alloc, 0); | |
569 | if (ret < 0) { | |
570 | mlog_errno(ret); | |
571 | goto bail; | |
572 | } | |
573 | ||
574 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, | |
575 | &contig_blocks, &ext_flags); | |
576 | if (ret < 0) { | |
577 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", | |
578 | (unsigned long long)iblock); | |
579 | ret = -EIO; | |
580 | goto bail; | |
581 | } | |
582 | } | |
583 | ||
25baf2da MF |
584 | /* |
585 | * get_more_blocks() expects us to describe a hole by clearing | |
586 | * the mapped bit on bh_result(). | |
49cb8d2d MF |
587 | * |
588 | * Consider an unwritten extent as a hole. | |
25baf2da | 589 | */ |
49cb8d2d | 590 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) |
25baf2da | 591 | map_bh(bh_result, inode->i_sb, p_blkno); |
5fe878ae | 592 | else |
25baf2da | 593 | clear_buffer_mapped(bh_result); |
ccd979bd MF |
594 | |
595 | /* make sure we don't map more than max_blocks blocks here as | |
596 | that's all the kernel will handle at this point. */ | |
597 | if (max_blocks < contig_blocks) | |
598 | contig_blocks = max_blocks; | |
599 | bh_result->b_size = contig_blocks << blocksize_bits; | |
600 | bail: | |
49255dce JQ |
601 | if (alloc_locked) |
602 | ocfs2_inode_unlock(inode, 1); | |
ccd979bd MF |
603 | return ret; |
604 | } | |
605 | ||
2bd63216 | 606 | /* |
ccd979bd | 607 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're |
bd5fe6c5 CH |
608 | * particularly interested in the aio/dio case. We use the rw_lock DLM lock |
609 | * to protect io on one node from truncation on another. | |
ccd979bd MF |
610 | */ |
611 | static void ocfs2_dio_end_io(struct kiocb *iocb, | |
612 | loff_t offset, | |
613 | ssize_t bytes, | |
7b7a8665 | 614 | void *private) |
ccd979bd | 615 | { |
496ad9aa | 616 | struct inode *inode = file_inode(iocb->ki_filp); |
7cdfc3a1 | 617 | int level; |
ccd979bd MF |
618 | |
619 | /* this io's submitter should not have unlocked this before we could */ | |
620 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); | |
7cdfc3a1 | 621 | |
df2d6f26 | 622 | if (ocfs2_iocb_is_sem_locked(iocb)) |
39c99f12 | 623 | ocfs2_iocb_clear_sem_locked(iocb); |
39c99f12 | 624 | |
a11f7e63 MF |
625 | if (ocfs2_iocb_is_unaligned_aio(iocb)) { |
626 | ocfs2_iocb_clear_unaligned_aio(iocb); | |
627 | ||
c18ceab0 | 628 | mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio); |
a11f7e63 MF |
629 | } |
630 | ||
ccd979bd | 631 | ocfs2_iocb_clear_rw_locked(iocb); |
7cdfc3a1 MF |
632 | |
633 | level = ocfs2_iocb_rw_locked_level(iocb); | |
7cdfc3a1 | 634 | ocfs2_rw_unlock(inode, level); |
ccd979bd MF |
635 | } |
636 | ||
03f981cf JB |
637 | static int ocfs2_releasepage(struct page *page, gfp_t wait) |
638 | { | |
03f981cf JB |
639 | if (!page_has_buffers(page)) |
640 | return 0; | |
41ecc345 | 641 | return try_to_free_buffers(page); |
03f981cf JB |
642 | } |
643 | ||
24c40b32 JQ |
644 | static int ocfs2_is_overwrite(struct ocfs2_super *osb, |
645 | struct inode *inode, loff_t offset) | |
646 | { | |
647 | int ret = 0; | |
648 | u32 v_cpos = 0; | |
649 | u32 p_cpos = 0; | |
650 | unsigned int num_clusters = 0; | |
651 | unsigned int ext_flags = 0; | |
652 | ||
653 | v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset); | |
654 | ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, | |
655 | &num_clusters, &ext_flags); | |
656 | if (ret < 0) { | |
657 | mlog_errno(ret); | |
658 | return ret; | |
659 | } | |
660 | ||
661 | if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
662 | return 1; | |
663 | ||
664 | return 0; | |
665 | } | |
666 | ||
14a5275d JQ |
667 | static int ocfs2_direct_IO_zero_extend(struct ocfs2_super *osb, |
668 | struct inode *inode, loff_t offset, | |
669 | u64 zero_len, int cluster_align) | |
670 | { | |
671 | u32 p_cpos = 0; | |
672 | u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, i_size_read(inode)); | |
673 | unsigned int num_clusters = 0; | |
674 | unsigned int ext_flags = 0; | |
675 | int ret = 0; | |
676 | ||
677 | if (offset <= i_size_read(inode) || cluster_align) | |
678 | return 0; | |
679 | ||
680 | ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, &num_clusters, | |
681 | &ext_flags); | |
682 | if (ret < 0) { | |
683 | mlog_errno(ret); | |
684 | return ret; | |
685 | } | |
686 | ||
687 | if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) { | |
688 | u64 s = i_size_read(inode); | |
689 | sector_t sector = (p_cpos << (osb->s_clustersize_bits - 9)) + | |
690 | (do_div(s, osb->s_clustersize) >> 9); | |
691 | ||
692 | ret = blkdev_issue_zeroout(osb->sb->s_bdev, sector, | |
693 | zero_len >> 9, GFP_NOFS, false); | |
694 | if (ret < 0) | |
695 | mlog_errno(ret); | |
696 | } | |
697 | ||
698 | return ret; | |
699 | } | |
700 | ||
701 | static int ocfs2_direct_IO_extend_no_holes(struct ocfs2_super *osb, | |
702 | struct inode *inode, loff_t offset) | |
703 | { | |
704 | u64 zero_start, zero_len, total_zero_len; | |
705 | u32 p_cpos = 0, clusters_to_add; | |
706 | u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, i_size_read(inode)); | |
707 | unsigned int num_clusters = 0; | |
708 | unsigned int ext_flags = 0; | |
709 | u32 size_div, offset_div; | |
710 | int ret = 0; | |
711 | ||
712 | { | |
713 | u64 o = offset; | |
714 | u64 s = i_size_read(inode); | |
715 | ||
716 | offset_div = do_div(o, osb->s_clustersize); | |
717 | size_div = do_div(s, osb->s_clustersize); | |
718 | } | |
719 | ||
720 | if (offset <= i_size_read(inode)) | |
721 | return 0; | |
722 | ||
723 | clusters_to_add = ocfs2_bytes_to_clusters(inode->i_sb, offset) - | |
724 | ocfs2_bytes_to_clusters(inode->i_sb, i_size_read(inode)); | |
725 | total_zero_len = offset - i_size_read(inode); | |
726 | if (clusters_to_add) | |
727 | total_zero_len -= offset_div; | |
728 | ||
729 | /* Allocate clusters to fill out holes, and this is only needed | |
730 | * when we add more than one clusters. Otherwise the cluster will | |
731 | * be allocated during direct IO */ | |
732 | if (clusters_to_add > 1) { | |
733 | ret = ocfs2_extend_allocation(inode, | |
734 | OCFS2_I(inode)->ip_clusters, | |
735 | clusters_to_add - 1, 0); | |
736 | if (ret) { | |
737 | mlog_errno(ret); | |
738 | goto out; | |
739 | } | |
740 | } | |
741 | ||
742 | while (total_zero_len) { | |
743 | ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, &num_clusters, | |
744 | &ext_flags); | |
745 | if (ret < 0) { | |
746 | mlog_errno(ret); | |
747 | goto out; | |
748 | } | |
749 | ||
750 | zero_start = ocfs2_clusters_to_bytes(osb->sb, p_cpos) + | |
751 | size_div; | |
752 | zero_len = ocfs2_clusters_to_bytes(osb->sb, num_clusters) - | |
753 | size_div; | |
754 | zero_len = min(total_zero_len, zero_len); | |
755 | ||
756 | if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) { | |
757 | ret = blkdev_issue_zeroout(osb->sb->s_bdev, | |
758 | zero_start >> 9, zero_len >> 9, | |
759 | GFP_NOFS, false); | |
760 | if (ret < 0) { | |
761 | mlog_errno(ret); | |
762 | goto out; | |
763 | } | |
764 | } | |
765 | ||
766 | total_zero_len -= zero_len; | |
767 | v_cpos += ocfs2_bytes_to_clusters(osb->sb, zero_len + size_div); | |
768 | ||
769 | /* Only at first iteration can be cluster not aligned. | |
770 | * So set size_div to 0 for the rest */ | |
771 | size_div = 0; | |
772 | } | |
773 | ||
774 | out: | |
775 | return ret; | |
776 | } | |
777 | ||
24c40b32 JQ |
778 | static ssize_t ocfs2_direct_IO_write(struct kiocb *iocb, |
779 | struct iov_iter *iter, | |
780 | loff_t offset) | |
781 | { | |
782 | ssize_t ret = 0; | |
783 | ssize_t written = 0; | |
784 | bool orphaned = false; | |
785 | int is_overwrite = 0; | |
786 | struct file *file = iocb->ki_filp; | |
787 | struct inode *inode = file_inode(file)->i_mapping->host; | |
788 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
789 | struct buffer_head *di_bh = NULL; | |
790 | size_t count = iter->count; | |
791 | journal_t *journal = osb->journal->j_journal; | |
14a5275d JQ |
792 | u64 zero_len_head, zero_len_tail; |
793 | int cluster_align_head, cluster_align_tail; | |
24c40b32 JQ |
794 | loff_t final_size = offset + count; |
795 | int append_write = offset >= i_size_read(inode) ? 1 : 0; | |
796 | unsigned int num_clusters = 0; | |
797 | unsigned int ext_flags = 0; | |
798 | ||
799 | { | |
800 | u64 o = offset; | |
14a5275d JQ |
801 | u64 s = i_size_read(inode); |
802 | ||
803 | zero_len_head = do_div(o, 1 << osb->s_clustersize_bits); | |
804 | cluster_align_head = !zero_len_head; | |
24c40b32 | 805 | |
14a5275d JQ |
806 | zero_len_tail = osb->s_clustersize - |
807 | do_div(s, osb->s_clustersize); | |
808 | if ((offset - i_size_read(inode)) < zero_len_tail) | |
809 | zero_len_tail = offset - i_size_read(inode); | |
810 | cluster_align_tail = !zero_len_tail; | |
24c40b32 JQ |
811 | } |
812 | ||
813 | /* | |
814 | * when final_size > inode->i_size, inode->i_size will be | |
815 | * updated after direct write, so add the inode to orphan | |
816 | * dir first. | |
817 | */ | |
818 | if (final_size > i_size_read(inode)) { | |
819 | ret = ocfs2_add_inode_to_orphan(osb, inode); | |
820 | if (ret < 0) { | |
821 | mlog_errno(ret); | |
822 | goto out; | |
823 | } | |
824 | orphaned = true; | |
825 | } | |
826 | ||
827 | if (append_write) { | |
7e9b1955 | 828 | ret = ocfs2_inode_lock(inode, NULL, 1); |
24c40b32 JQ |
829 | if (ret < 0) { |
830 | mlog_errno(ret); | |
831 | goto clean_orphan; | |
832 | } | |
833 | ||
14a5275d JQ |
834 | /* zeroing out the previously allocated cluster tail |
835 | * that but not zeroed */ | |
24c40b32 | 836 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
14a5275d JQ |
837 | ret = ocfs2_direct_IO_zero_extend(osb, inode, offset, |
838 | zero_len_tail, cluster_align_tail); | |
24c40b32 | 839 | else |
14a5275d | 840 | ret = ocfs2_direct_IO_extend_no_holes(osb, inode, |
24c40b32 JQ |
841 | offset); |
842 | if (ret < 0) { | |
843 | mlog_errno(ret); | |
844 | ocfs2_inode_unlock(inode, 1); | |
24c40b32 JQ |
845 | goto clean_orphan; |
846 | } | |
847 | ||
848 | is_overwrite = ocfs2_is_overwrite(osb, inode, offset); | |
849 | if (is_overwrite < 0) { | |
850 | mlog_errno(is_overwrite); | |
851 | ocfs2_inode_unlock(inode, 1); | |
24c40b32 JQ |
852 | goto clean_orphan; |
853 | } | |
854 | ||
855 | ocfs2_inode_unlock(inode, 1); | |
24c40b32 JQ |
856 | } |
857 | ||
17f8c842 OS |
858 | written = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter, |
859 | offset, ocfs2_direct_IO_get_blocks, | |
860 | ocfs2_dio_end_io, NULL, 0); | |
24c40b32 JQ |
861 | if (unlikely(written < 0)) { |
862 | loff_t i_size = i_size_read(inode); | |
863 | ||
864 | if (offset + count > i_size) { | |
865 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
866 | if (ret < 0) { | |
867 | mlog_errno(ret); | |
868 | goto clean_orphan; | |
869 | } | |
870 | ||
871 | if (i_size == i_size_read(inode)) { | |
872 | ret = ocfs2_truncate_file(inode, di_bh, | |
873 | i_size); | |
874 | if (ret < 0) { | |
875 | if (ret != -ENOSPC) | |
876 | mlog_errno(ret); | |
877 | ||
878 | ocfs2_inode_unlock(inode, 1); | |
879 | brelse(di_bh); | |
880 | goto clean_orphan; | |
881 | } | |
882 | } | |
883 | ||
884 | ocfs2_inode_unlock(inode, 1); | |
885 | brelse(di_bh); | |
886 | ||
887 | ret = jbd2_journal_force_commit(journal); | |
888 | if (ret < 0) | |
889 | mlog_errno(ret); | |
890 | } | |
bdd86215 | 891 | } else if (written > 0 && append_write && !is_overwrite && |
14a5275d JQ |
892 | !cluster_align_head) { |
893 | /* zeroing out the allocated cluster head */ | |
24c40b32 JQ |
894 | u32 p_cpos = 0; |
895 | u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset); | |
896 | ||
37a8d89a JQ |
897 | ret = ocfs2_inode_lock(inode, NULL, 0); |
898 | if (ret < 0) { | |
899 | mlog_errno(ret); | |
900 | goto clean_orphan; | |
901 | } | |
902 | ||
24c40b32 JQ |
903 | ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, |
904 | &num_clusters, &ext_flags); | |
905 | if (ret < 0) { | |
906 | mlog_errno(ret); | |
37a8d89a | 907 | ocfs2_inode_unlock(inode, 0); |
24c40b32 JQ |
908 | goto clean_orphan; |
909 | } | |
910 | ||
911 | BUG_ON(!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN)); | |
912 | ||
913 | ret = blkdev_issue_zeroout(osb->sb->s_bdev, | |
914 | p_cpos << (osb->s_clustersize_bits - 9), | |
14a5275d | 915 | zero_len_head >> 9, GFP_NOFS, false); |
24c40b32 JQ |
916 | if (ret < 0) |
917 | mlog_errno(ret); | |
37a8d89a JQ |
918 | |
919 | ocfs2_inode_unlock(inode, 0); | |
24c40b32 JQ |
920 | } |
921 | ||
922 | clean_orphan: | |
923 | if (orphaned) { | |
924 | int tmp_ret; | |
925 | int update_isize = written > 0 ? 1 : 0; | |
926 | loff_t end = update_isize ? offset + written : 0; | |
927 | ||
928 | tmp_ret = ocfs2_del_inode_from_orphan(osb, inode, | |
929 | update_isize, end); | |
930 | if (tmp_ret < 0) { | |
931 | ret = tmp_ret; | |
932 | goto out; | |
933 | } | |
934 | ||
935 | tmp_ret = jbd2_journal_force_commit(journal); | |
936 | if (tmp_ret < 0) { | |
937 | ret = tmp_ret; | |
938 | mlog_errno(tmp_ret); | |
939 | } | |
940 | } | |
941 | ||
942 | out: | |
943 | if (ret >= 0) | |
944 | ret = written; | |
945 | return ret; | |
946 | } | |
947 | ||
22c6186e | 948 | static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter, |
d8d3d94b | 949 | loff_t offset) |
ccd979bd MF |
950 | { |
951 | struct file *file = iocb->ki_filp; | |
496ad9aa | 952 | struct inode *inode = file_inode(file)->i_mapping->host; |
24c40b32 JQ |
953 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
954 | int full_coherency = !(osb->s_mount_opt & | |
955 | OCFS2_MOUNT_COHERENCY_BUFFERED); | |
53013cba | 956 | |
6798d35a MF |
957 | /* |
958 | * Fallback to buffered I/O if we see an inode without | |
959 | * extents. | |
960 | */ | |
961 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
962 | return 0; | |
963 | ||
24c40b32 JQ |
964 | /* Fallback to buffered I/O if we are appending and |
965 | * concurrent O_DIRECT writes are allowed. | |
966 | */ | |
967 | if (i_size_read(inode) <= offset && !full_coherency) | |
b80474b4 TM |
968 | return 0; |
969 | ||
6f673763 | 970 | if (iov_iter_rw(iter) == READ) |
17f8c842 OS |
971 | return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, |
972 | iter, offset, | |
973 | ocfs2_direct_IO_get_blocks, | |
974 | ocfs2_dio_end_io, NULL, 0); | |
24c40b32 JQ |
975 | else |
976 | return ocfs2_direct_IO_write(iocb, iter, offset); | |
ccd979bd MF |
977 | } |
978 | ||
9517bac6 MF |
979 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, |
980 | u32 cpos, | |
981 | unsigned int *start, | |
982 | unsigned int *end) | |
983 | { | |
984 | unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE; | |
985 | ||
986 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) { | |
987 | unsigned int cpp; | |
988 | ||
989 | cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits); | |
990 | ||
991 | cluster_start = cpos % cpp; | |
992 | cluster_start = cluster_start << osb->s_clustersize_bits; | |
993 | ||
994 | cluster_end = cluster_start + osb->s_clustersize; | |
995 | } | |
996 | ||
997 | BUG_ON(cluster_start > PAGE_SIZE); | |
998 | BUG_ON(cluster_end > PAGE_SIZE); | |
999 | ||
1000 | if (start) | |
1001 | *start = cluster_start; | |
1002 | if (end) | |
1003 | *end = cluster_end; | |
1004 | } | |
1005 | ||
1006 | /* | |
1007 | * 'from' and 'to' are the region in the page to avoid zeroing. | |
1008 | * | |
1009 | * If pagesize > clustersize, this function will avoid zeroing outside | |
1010 | * of the cluster boundary. | |
1011 | * | |
1012 | * from == to == 0 is code for "zero the entire cluster region" | |
1013 | */ | |
1014 | static void ocfs2_clear_page_regions(struct page *page, | |
1015 | struct ocfs2_super *osb, u32 cpos, | |
1016 | unsigned from, unsigned to) | |
1017 | { | |
1018 | void *kaddr; | |
1019 | unsigned int cluster_start, cluster_end; | |
1020 | ||
1021 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); | |
1022 | ||
c4bc8dcb | 1023 | kaddr = kmap_atomic(page); |
9517bac6 MF |
1024 | |
1025 | if (from || to) { | |
1026 | if (from > cluster_start) | |
1027 | memset(kaddr + cluster_start, 0, from - cluster_start); | |
1028 | if (to < cluster_end) | |
1029 | memset(kaddr + to, 0, cluster_end - to); | |
1030 | } else { | |
1031 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); | |
1032 | } | |
1033 | ||
c4bc8dcb | 1034 | kunmap_atomic(kaddr); |
9517bac6 MF |
1035 | } |
1036 | ||
4e9563fd MF |
1037 | /* |
1038 | * Nonsparse file systems fully allocate before we get to the write | |
1039 | * code. This prevents ocfs2_write() from tagging the write as an | |
1040 | * allocating one, which means ocfs2_map_page_blocks() might try to | |
1041 | * read-in the blocks at the tail of our file. Avoid reading them by | |
1042 | * testing i_size against each block offset. | |
1043 | */ | |
1044 | static int ocfs2_should_read_blk(struct inode *inode, struct page *page, | |
1045 | unsigned int block_start) | |
1046 | { | |
1047 | u64 offset = page_offset(page) + block_start; | |
1048 | ||
1049 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
1050 | return 1; | |
1051 | ||
1052 | if (i_size_read(inode) > offset) | |
1053 | return 1; | |
1054 | ||
1055 | return 0; | |
1056 | } | |
1057 | ||
9517bac6 | 1058 | /* |
ebdec241 | 1059 | * Some of this taken from __block_write_begin(). We already have our |
9517bac6 MF |
1060 | * mapping by now though, and the entire write will be allocating or |
1061 | * it won't, so not much need to use BH_New. | |
1062 | * | |
1063 | * This will also skip zeroing, which is handled externally. | |
1064 | */ | |
60b11392 MF |
1065 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, |
1066 | struct inode *inode, unsigned int from, | |
1067 | unsigned int to, int new) | |
9517bac6 MF |
1068 | { |
1069 | int ret = 0; | |
1070 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; | |
1071 | unsigned int block_end, block_start; | |
1072 | unsigned int bsize = 1 << inode->i_blkbits; | |
1073 | ||
1074 | if (!page_has_buffers(page)) | |
1075 | create_empty_buffers(page, bsize, 0); | |
1076 | ||
1077 | head = page_buffers(page); | |
1078 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1079 | bh = bh->b_this_page, block_start += bsize) { | |
1080 | block_end = block_start + bsize; | |
1081 | ||
3a307ffc MF |
1082 | clear_buffer_new(bh); |
1083 | ||
9517bac6 MF |
1084 | /* |
1085 | * Ignore blocks outside of our i/o range - | |
1086 | * they may belong to unallocated clusters. | |
1087 | */ | |
60b11392 | 1088 | if (block_start >= to || block_end <= from) { |
9517bac6 MF |
1089 | if (PageUptodate(page)) |
1090 | set_buffer_uptodate(bh); | |
1091 | continue; | |
1092 | } | |
1093 | ||
1094 | /* | |
1095 | * For an allocating write with cluster size >= page | |
1096 | * size, we always write the entire page. | |
1097 | */ | |
3a307ffc MF |
1098 | if (new) |
1099 | set_buffer_new(bh); | |
9517bac6 MF |
1100 | |
1101 | if (!buffer_mapped(bh)) { | |
1102 | map_bh(bh, inode->i_sb, *p_blkno); | |
1103 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
1104 | } | |
1105 | ||
1106 | if (PageUptodate(page)) { | |
1107 | if (!buffer_uptodate(bh)) | |
1108 | set_buffer_uptodate(bh); | |
1109 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
bce99768 | 1110 | !buffer_new(bh) && |
4e9563fd | 1111 | ocfs2_should_read_blk(inode, page, block_start) && |
bce99768 | 1112 | (block_start < from || block_end > to)) { |
9517bac6 MF |
1113 | ll_rw_block(READ, 1, &bh); |
1114 | *wait_bh++=bh; | |
1115 | } | |
1116 | ||
1117 | *p_blkno = *p_blkno + 1; | |
1118 | } | |
1119 | ||
1120 | /* | |
1121 | * If we issued read requests - let them complete. | |
1122 | */ | |
1123 | while(wait_bh > wait) { | |
1124 | wait_on_buffer(*--wait_bh); | |
1125 | if (!buffer_uptodate(*wait_bh)) | |
1126 | ret = -EIO; | |
1127 | } | |
1128 | ||
1129 | if (ret == 0 || !new) | |
1130 | return ret; | |
1131 | ||
1132 | /* | |
1133 | * If we get -EIO above, zero out any newly allocated blocks | |
1134 | * to avoid exposing stale data. | |
1135 | */ | |
1136 | bh = head; | |
1137 | block_start = 0; | |
1138 | do { | |
9517bac6 MF |
1139 | block_end = block_start + bsize; |
1140 | if (block_end <= from) | |
1141 | goto next_bh; | |
1142 | if (block_start >= to) | |
1143 | break; | |
1144 | ||
eebd2aa3 | 1145 | zero_user(page, block_start, bh->b_size); |
9517bac6 MF |
1146 | set_buffer_uptodate(bh); |
1147 | mark_buffer_dirty(bh); | |
1148 | ||
1149 | next_bh: | |
1150 | block_start = block_end; | |
1151 | bh = bh->b_this_page; | |
1152 | } while (bh != head); | |
1153 | ||
1154 | return ret; | |
1155 | } | |
1156 | ||
3a307ffc MF |
1157 | #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE) |
1158 | #define OCFS2_MAX_CTXT_PAGES 1 | |
1159 | #else | |
1160 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE) | |
1161 | #endif | |
1162 | ||
1163 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE) | |
1164 | ||
6af67d82 | 1165 | /* |
3a307ffc | 1166 | * Describe the state of a single cluster to be written to. |
6af67d82 | 1167 | */ |
3a307ffc MF |
1168 | struct ocfs2_write_cluster_desc { |
1169 | u32 c_cpos; | |
1170 | u32 c_phys; | |
1171 | /* | |
1172 | * Give this a unique field because c_phys eventually gets | |
1173 | * filled. | |
1174 | */ | |
1175 | unsigned c_new; | |
b27b7cbc | 1176 | unsigned c_unwritten; |
e7432675 | 1177 | unsigned c_needs_zero; |
3a307ffc | 1178 | }; |
6af67d82 | 1179 | |
3a307ffc MF |
1180 | struct ocfs2_write_ctxt { |
1181 | /* Logical cluster position / len of write */ | |
1182 | u32 w_cpos; | |
1183 | u32 w_clen; | |
6af67d82 | 1184 | |
e7432675 SM |
1185 | /* First cluster allocated in a nonsparse extend */ |
1186 | u32 w_first_new_cpos; | |
1187 | ||
3a307ffc | 1188 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; |
6af67d82 | 1189 | |
3a307ffc MF |
1190 | /* |
1191 | * This is true if page_size > cluster_size. | |
1192 | * | |
1193 | * It triggers a set of special cases during write which might | |
1194 | * have to deal with allocating writes to partial pages. | |
1195 | */ | |
1196 | unsigned int w_large_pages; | |
6af67d82 | 1197 | |
3a307ffc MF |
1198 | /* |
1199 | * Pages involved in this write. | |
1200 | * | |
1201 | * w_target_page is the page being written to by the user. | |
1202 | * | |
1203 | * w_pages is an array of pages which always contains | |
1204 | * w_target_page, and in the case of an allocating write with | |
1205 | * page_size < cluster size, it will contain zero'd and mapped | |
1206 | * pages adjacent to w_target_page which need to be written | |
1207 | * out in so that future reads from that region will get | |
1208 | * zero's. | |
1209 | */ | |
3a307ffc | 1210 | unsigned int w_num_pages; |
83fd9c7f | 1211 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; |
3a307ffc | 1212 | struct page *w_target_page; |
eeb47d12 | 1213 | |
5cffff9e WW |
1214 | /* |
1215 | * w_target_locked is used for page_mkwrite path indicating no unlocking | |
1216 | * against w_target_page in ocfs2_write_end_nolock. | |
1217 | */ | |
1218 | unsigned int w_target_locked:1; | |
1219 | ||
3a307ffc MF |
1220 | /* |
1221 | * ocfs2_write_end() uses this to know what the real range to | |
1222 | * write in the target should be. | |
1223 | */ | |
1224 | unsigned int w_target_from; | |
1225 | unsigned int w_target_to; | |
1226 | ||
1227 | /* | |
1228 | * We could use journal_current_handle() but this is cleaner, | |
1229 | * IMHO -Mark | |
1230 | */ | |
1231 | handle_t *w_handle; | |
1232 | ||
1233 | struct buffer_head *w_di_bh; | |
b27b7cbc MF |
1234 | |
1235 | struct ocfs2_cached_dealloc_ctxt w_dealloc; | |
3a307ffc MF |
1236 | }; |
1237 | ||
1d410a6e | 1238 | void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages) |
3a307ffc MF |
1239 | { |
1240 | int i; | |
1241 | ||
1d410a6e MF |
1242 | for(i = 0; i < num_pages; i++) { |
1243 | if (pages[i]) { | |
1244 | unlock_page(pages[i]); | |
1245 | mark_page_accessed(pages[i]); | |
1246 | page_cache_release(pages[i]); | |
1247 | } | |
6af67d82 | 1248 | } |
1d410a6e MF |
1249 | } |
1250 | ||
136f49b9 | 1251 | static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc) |
1d410a6e | 1252 | { |
5cffff9e WW |
1253 | int i; |
1254 | ||
1255 | /* | |
1256 | * w_target_locked is only set to true in the page_mkwrite() case. | |
1257 | * The intent is to allow us to lock the target page from write_begin() | |
1258 | * to write_end(). The caller must hold a ref on w_target_page. | |
1259 | */ | |
1260 | if (wc->w_target_locked) { | |
1261 | BUG_ON(!wc->w_target_page); | |
1262 | for (i = 0; i < wc->w_num_pages; i++) { | |
1263 | if (wc->w_target_page == wc->w_pages[i]) { | |
1264 | wc->w_pages[i] = NULL; | |
1265 | break; | |
1266 | } | |
1267 | } | |
1268 | mark_page_accessed(wc->w_target_page); | |
1269 | page_cache_release(wc->w_target_page); | |
1270 | } | |
1d410a6e | 1271 | ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages); |
136f49b9 | 1272 | } |
6af67d82 | 1273 | |
136f49b9 JB |
1274 | static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc) |
1275 | { | |
1276 | ocfs2_unlock_pages(wc); | |
3a307ffc MF |
1277 | brelse(wc->w_di_bh); |
1278 | kfree(wc); | |
1279 | } | |
1280 | ||
1281 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, | |
1282 | struct ocfs2_super *osb, loff_t pos, | |
607d44aa | 1283 | unsigned len, struct buffer_head *di_bh) |
3a307ffc | 1284 | { |
30b8548f | 1285 | u32 cend; |
3a307ffc MF |
1286 | struct ocfs2_write_ctxt *wc; |
1287 | ||
1288 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); | |
1289 | if (!wc) | |
1290 | return -ENOMEM; | |
6af67d82 | 1291 | |
3a307ffc | 1292 | wc->w_cpos = pos >> osb->s_clustersize_bits; |
e7432675 | 1293 | wc->w_first_new_cpos = UINT_MAX; |
30b8548f | 1294 | cend = (pos + len - 1) >> osb->s_clustersize_bits; |
1295 | wc->w_clen = cend - wc->w_cpos + 1; | |
607d44aa MF |
1296 | get_bh(di_bh); |
1297 | wc->w_di_bh = di_bh; | |
6af67d82 | 1298 | |
3a307ffc MF |
1299 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) |
1300 | wc->w_large_pages = 1; | |
1301 | else | |
1302 | wc->w_large_pages = 0; | |
1303 | ||
b27b7cbc MF |
1304 | ocfs2_init_dealloc_ctxt(&wc->w_dealloc); |
1305 | ||
3a307ffc | 1306 | *wcp = wc; |
6af67d82 | 1307 | |
3a307ffc | 1308 | return 0; |
6af67d82 MF |
1309 | } |
1310 | ||
9517bac6 | 1311 | /* |
3a307ffc MF |
1312 | * If a page has any new buffers, zero them out here, and mark them uptodate |
1313 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1314 | * block data from leaking). And clear the new bit. | |
9517bac6 | 1315 | */ |
3a307ffc | 1316 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) |
9517bac6 | 1317 | { |
3a307ffc MF |
1318 | unsigned int block_start, block_end; |
1319 | struct buffer_head *head, *bh; | |
9517bac6 | 1320 | |
3a307ffc MF |
1321 | BUG_ON(!PageLocked(page)); |
1322 | if (!page_has_buffers(page)) | |
1323 | return; | |
9517bac6 | 1324 | |
3a307ffc MF |
1325 | bh = head = page_buffers(page); |
1326 | block_start = 0; | |
1327 | do { | |
1328 | block_end = block_start + bh->b_size; | |
1329 | ||
1330 | if (buffer_new(bh)) { | |
1331 | if (block_end > from && block_start < to) { | |
1332 | if (!PageUptodate(page)) { | |
1333 | unsigned start, end; | |
3a307ffc MF |
1334 | |
1335 | start = max(from, block_start); | |
1336 | end = min(to, block_end); | |
1337 | ||
eebd2aa3 | 1338 | zero_user_segment(page, start, end); |
3a307ffc MF |
1339 | set_buffer_uptodate(bh); |
1340 | } | |
1341 | ||
1342 | clear_buffer_new(bh); | |
1343 | mark_buffer_dirty(bh); | |
1344 | } | |
1345 | } | |
9517bac6 | 1346 | |
3a307ffc MF |
1347 | block_start = block_end; |
1348 | bh = bh->b_this_page; | |
1349 | } while (bh != head); | |
1350 | } | |
1351 | ||
1352 | /* | |
1353 | * Only called when we have a failure during allocating write to write | |
1354 | * zero's to the newly allocated region. | |
1355 | */ | |
1356 | static void ocfs2_write_failure(struct inode *inode, | |
1357 | struct ocfs2_write_ctxt *wc, | |
1358 | loff_t user_pos, unsigned user_len) | |
1359 | { | |
1360 | int i; | |
5c26a7b7 MF |
1361 | unsigned from = user_pos & (PAGE_CACHE_SIZE - 1), |
1362 | to = user_pos + user_len; | |
3a307ffc MF |
1363 | struct page *tmppage; |
1364 | ||
5c26a7b7 | 1365 | ocfs2_zero_new_buffers(wc->w_target_page, from, to); |
9517bac6 | 1366 | |
3a307ffc MF |
1367 | for(i = 0; i < wc->w_num_pages; i++) { |
1368 | tmppage = wc->w_pages[i]; | |
9517bac6 | 1369 | |
961cecbe | 1370 | if (page_has_buffers(tmppage)) { |
53ef99ca | 1371 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 1372 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
1373 | |
1374 | block_commit_write(tmppage, from, to); | |
1375 | } | |
9517bac6 | 1376 | } |
9517bac6 MF |
1377 | } |
1378 | ||
3a307ffc MF |
1379 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, |
1380 | struct ocfs2_write_ctxt *wc, | |
1381 | struct page *page, u32 cpos, | |
1382 | loff_t user_pos, unsigned user_len, | |
1383 | int new) | |
9517bac6 | 1384 | { |
3a307ffc MF |
1385 | int ret; |
1386 | unsigned int map_from = 0, map_to = 0; | |
9517bac6 | 1387 | unsigned int cluster_start, cluster_end; |
3a307ffc | 1388 | unsigned int user_data_from = 0, user_data_to = 0; |
9517bac6 | 1389 | |
3a307ffc | 1390 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, |
9517bac6 MF |
1391 | &cluster_start, &cluster_end); |
1392 | ||
272b62c1 GR |
1393 | /* treat the write as new if the a hole/lseek spanned across |
1394 | * the page boundary. | |
1395 | */ | |
1396 | new = new | ((i_size_read(inode) <= page_offset(page)) && | |
1397 | (page_offset(page) <= user_pos)); | |
1398 | ||
3a307ffc MF |
1399 | if (page == wc->w_target_page) { |
1400 | map_from = user_pos & (PAGE_CACHE_SIZE - 1); | |
1401 | map_to = map_from + user_len; | |
1402 | ||
1403 | if (new) | |
1404 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1405 | cluster_start, cluster_end, | |
1406 | new); | |
1407 | else | |
1408 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
1409 | map_from, map_to, new); | |
1410 | if (ret) { | |
9517bac6 MF |
1411 | mlog_errno(ret); |
1412 | goto out; | |
1413 | } | |
1414 | ||
3a307ffc MF |
1415 | user_data_from = map_from; |
1416 | user_data_to = map_to; | |
9517bac6 | 1417 | if (new) { |
3a307ffc MF |
1418 | map_from = cluster_start; |
1419 | map_to = cluster_end; | |
9517bac6 MF |
1420 | } |
1421 | } else { | |
1422 | /* | |
1423 | * If we haven't allocated the new page yet, we | |
1424 | * shouldn't be writing it out without copying user | |
1425 | * data. This is likely a math error from the caller. | |
1426 | */ | |
1427 | BUG_ON(!new); | |
1428 | ||
3a307ffc MF |
1429 | map_from = cluster_start; |
1430 | map_to = cluster_end; | |
9517bac6 MF |
1431 | |
1432 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
3a307ffc | 1433 | cluster_start, cluster_end, new); |
9517bac6 MF |
1434 | if (ret) { |
1435 | mlog_errno(ret); | |
1436 | goto out; | |
1437 | } | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * Parts of newly allocated pages need to be zero'd. | |
1442 | * | |
1443 | * Above, we have also rewritten 'to' and 'from' - as far as | |
1444 | * the rest of the function is concerned, the entire cluster | |
1445 | * range inside of a page needs to be written. | |
1446 | * | |
1447 | * We can skip this if the page is up to date - it's already | |
1448 | * been zero'd from being read in as a hole. | |
1449 | */ | |
1450 | if (new && !PageUptodate(page)) | |
1451 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), | |
3a307ffc | 1452 | cpos, user_data_from, user_data_to); |
9517bac6 MF |
1453 | |
1454 | flush_dcache_page(page); | |
1455 | ||
9517bac6 | 1456 | out: |
3a307ffc | 1457 | return ret; |
9517bac6 MF |
1458 | } |
1459 | ||
1460 | /* | |
3a307ffc | 1461 | * This function will only grab one clusters worth of pages. |
9517bac6 | 1462 | */ |
3a307ffc MF |
1463 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, |
1464 | struct ocfs2_write_ctxt *wc, | |
693c241a JB |
1465 | u32 cpos, loff_t user_pos, |
1466 | unsigned user_len, int new, | |
7307de80 | 1467 | struct page *mmap_page) |
9517bac6 | 1468 | { |
3a307ffc | 1469 | int ret = 0, i; |
693c241a | 1470 | unsigned long start, target_index, end_index, index; |
9517bac6 | 1471 | struct inode *inode = mapping->host; |
693c241a | 1472 | loff_t last_byte; |
9517bac6 | 1473 | |
3a307ffc | 1474 | target_index = user_pos >> PAGE_CACHE_SHIFT; |
9517bac6 MF |
1475 | |
1476 | /* | |
1477 | * Figure out how many pages we'll be manipulating here. For | |
60b11392 | 1478 | * non allocating write, we just change the one |
693c241a JB |
1479 | * page. Otherwise, we'll need a whole clusters worth. If we're |
1480 | * writing past i_size, we only need enough pages to cover the | |
1481 | * last page of the write. | |
9517bac6 | 1482 | */ |
9517bac6 | 1483 | if (new) { |
3a307ffc MF |
1484 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); |
1485 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); | |
693c241a JB |
1486 | /* |
1487 | * We need the index *past* the last page we could possibly | |
1488 | * touch. This is the page past the end of the write or | |
1489 | * i_size, whichever is greater. | |
1490 | */ | |
1491 | last_byte = max(user_pos + user_len, i_size_read(inode)); | |
1492 | BUG_ON(last_byte < 1); | |
1493 | end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1; | |
1494 | if ((start + wc->w_num_pages) > end_index) | |
1495 | wc->w_num_pages = end_index - start; | |
9517bac6 | 1496 | } else { |
3a307ffc MF |
1497 | wc->w_num_pages = 1; |
1498 | start = target_index; | |
9517bac6 MF |
1499 | } |
1500 | ||
3a307ffc | 1501 | for(i = 0; i < wc->w_num_pages; i++) { |
9517bac6 MF |
1502 | index = start + i; |
1503 | ||
7307de80 MF |
1504 | if (index == target_index && mmap_page) { |
1505 | /* | |
1506 | * ocfs2_pagemkwrite() is a little different | |
1507 | * and wants us to directly use the page | |
1508 | * passed in. | |
1509 | */ | |
1510 | lock_page(mmap_page); | |
1511 | ||
5cffff9e | 1512 | /* Exit and let the caller retry */ |
7307de80 | 1513 | if (mmap_page->mapping != mapping) { |
5cffff9e | 1514 | WARN_ON(mmap_page->mapping); |
7307de80 | 1515 | unlock_page(mmap_page); |
5cffff9e | 1516 | ret = -EAGAIN; |
7307de80 MF |
1517 | goto out; |
1518 | } | |
1519 | ||
1520 | page_cache_get(mmap_page); | |
1521 | wc->w_pages[i] = mmap_page; | |
5cffff9e | 1522 | wc->w_target_locked = true; |
7307de80 MF |
1523 | } else { |
1524 | wc->w_pages[i] = find_or_create_page(mapping, index, | |
1525 | GFP_NOFS); | |
1526 | if (!wc->w_pages[i]) { | |
1527 | ret = -ENOMEM; | |
1528 | mlog_errno(ret); | |
1529 | goto out; | |
1530 | } | |
9517bac6 | 1531 | } |
1269529b | 1532 | wait_for_stable_page(wc->w_pages[i]); |
3a307ffc MF |
1533 | |
1534 | if (index == target_index) | |
1535 | wc->w_target_page = wc->w_pages[i]; | |
9517bac6 | 1536 | } |
3a307ffc | 1537 | out: |
5cffff9e WW |
1538 | if (ret) |
1539 | wc->w_target_locked = false; | |
3a307ffc MF |
1540 | return ret; |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * Prepare a single cluster for write one cluster into the file. | |
1545 | */ | |
1546 | static int ocfs2_write_cluster(struct address_space *mapping, | |
b27b7cbc | 1547 | u32 phys, unsigned int unwritten, |
e7432675 | 1548 | unsigned int should_zero, |
b27b7cbc | 1549 | struct ocfs2_alloc_context *data_ac, |
3a307ffc MF |
1550 | struct ocfs2_alloc_context *meta_ac, |
1551 | struct ocfs2_write_ctxt *wc, u32 cpos, | |
1552 | loff_t user_pos, unsigned user_len) | |
1553 | { | |
e7432675 | 1554 | int ret, i, new; |
3a307ffc MF |
1555 | u64 v_blkno, p_blkno; |
1556 | struct inode *inode = mapping->host; | |
f99b9b7c | 1557 | struct ocfs2_extent_tree et; |
3a307ffc MF |
1558 | |
1559 | new = phys == 0 ? 1 : 0; | |
9517bac6 | 1560 | if (new) { |
3a307ffc MF |
1561 | u32 tmp_pos; |
1562 | ||
9517bac6 MF |
1563 | /* |
1564 | * This is safe to call with the page locks - it won't take | |
1565 | * any additional semaphores or cluster locks. | |
1566 | */ | |
3a307ffc | 1567 | tmp_pos = cpos; |
0eb8d47e TM |
1568 | ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode, |
1569 | &tmp_pos, 1, 0, wc->w_di_bh, | |
1570 | wc->w_handle, data_ac, | |
1571 | meta_ac, NULL); | |
9517bac6 MF |
1572 | /* |
1573 | * This shouldn't happen because we must have already | |
1574 | * calculated the correct meta data allocation required. The | |
1575 | * internal tree allocation code should know how to increase | |
1576 | * transaction credits itself. | |
1577 | * | |
1578 | * If need be, we could handle -EAGAIN for a | |
1579 | * RESTART_TRANS here. | |
1580 | */ | |
1581 | mlog_bug_on_msg(ret == -EAGAIN, | |
1582 | "Inode %llu: EAGAIN return during allocation.\n", | |
1583 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1584 | if (ret < 0) { | |
1585 | mlog_errno(ret); | |
1586 | goto out; | |
1587 | } | |
b27b7cbc | 1588 | } else if (unwritten) { |
5e404e9e JB |
1589 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
1590 | wc->w_di_bh); | |
f99b9b7c | 1591 | ret = ocfs2_mark_extent_written(inode, &et, |
b27b7cbc | 1592 | wc->w_handle, cpos, 1, phys, |
f99b9b7c | 1593 | meta_ac, &wc->w_dealloc); |
b27b7cbc MF |
1594 | if (ret < 0) { |
1595 | mlog_errno(ret); | |
1596 | goto out; | |
1597 | } | |
1598 | } | |
3a307ffc | 1599 | |
b27b7cbc | 1600 | if (should_zero) |
3a307ffc | 1601 | v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos); |
b27b7cbc | 1602 | else |
3a307ffc | 1603 | v_blkno = user_pos >> inode->i_sb->s_blocksize_bits; |
9517bac6 | 1604 | |
3a307ffc MF |
1605 | /* |
1606 | * The only reason this should fail is due to an inability to | |
1607 | * find the extent added. | |
1608 | */ | |
49cb8d2d MF |
1609 | ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL, |
1610 | NULL); | |
9517bac6 | 1611 | if (ret < 0) { |
61fb9ea4 | 1612 | mlog(ML_ERROR, "Get physical blkno failed for inode %llu, " |
3a307ffc MF |
1613 | "at logical block %llu", |
1614 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1615 | (unsigned long long)v_blkno); | |
9517bac6 MF |
1616 | goto out; |
1617 | } | |
1618 | ||
1619 | BUG_ON(p_blkno == 0); | |
1620 | ||
3a307ffc MF |
1621 | for(i = 0; i < wc->w_num_pages; i++) { |
1622 | int tmpret; | |
9517bac6 | 1623 | |
3a307ffc MF |
1624 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, |
1625 | wc->w_pages[i], cpos, | |
b27b7cbc MF |
1626 | user_pos, user_len, |
1627 | should_zero); | |
3a307ffc MF |
1628 | if (tmpret) { |
1629 | mlog_errno(tmpret); | |
1630 | if (ret == 0) | |
cbfa9639 | 1631 | ret = tmpret; |
3a307ffc | 1632 | } |
9517bac6 MF |
1633 | } |
1634 | ||
3a307ffc MF |
1635 | /* |
1636 | * We only have cleanup to do in case of allocating write. | |
1637 | */ | |
1638 | if (ret && new) | |
1639 | ocfs2_write_failure(inode, wc, user_pos, user_len); | |
1640 | ||
9517bac6 | 1641 | out: |
9517bac6 | 1642 | |
3a307ffc | 1643 | return ret; |
9517bac6 MF |
1644 | } |
1645 | ||
0d172baa MF |
1646 | static int ocfs2_write_cluster_by_desc(struct address_space *mapping, |
1647 | struct ocfs2_alloc_context *data_ac, | |
1648 | struct ocfs2_alloc_context *meta_ac, | |
1649 | struct ocfs2_write_ctxt *wc, | |
1650 | loff_t pos, unsigned len) | |
1651 | { | |
1652 | int ret, i; | |
db56246c MF |
1653 | loff_t cluster_off; |
1654 | unsigned int local_len = len; | |
0d172baa | 1655 | struct ocfs2_write_cluster_desc *desc; |
db56246c | 1656 | struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb); |
0d172baa MF |
1657 | |
1658 | for (i = 0; i < wc->w_clen; i++) { | |
1659 | desc = &wc->w_desc[i]; | |
1660 | ||
db56246c MF |
1661 | /* |
1662 | * We have to make sure that the total write passed in | |
1663 | * doesn't extend past a single cluster. | |
1664 | */ | |
1665 | local_len = len; | |
1666 | cluster_off = pos & (osb->s_clustersize - 1); | |
1667 | if ((cluster_off + local_len) > osb->s_clustersize) | |
1668 | local_len = osb->s_clustersize - cluster_off; | |
1669 | ||
b27b7cbc | 1670 | ret = ocfs2_write_cluster(mapping, desc->c_phys, |
e7432675 SM |
1671 | desc->c_unwritten, |
1672 | desc->c_needs_zero, | |
1673 | data_ac, meta_ac, | |
db56246c | 1674 | wc, desc->c_cpos, pos, local_len); |
0d172baa MF |
1675 | if (ret) { |
1676 | mlog_errno(ret); | |
1677 | goto out; | |
1678 | } | |
db56246c MF |
1679 | |
1680 | len -= local_len; | |
1681 | pos += local_len; | |
0d172baa MF |
1682 | } |
1683 | ||
1684 | ret = 0; | |
1685 | out: | |
1686 | return ret; | |
1687 | } | |
1688 | ||
3a307ffc MF |
1689 | /* |
1690 | * ocfs2_write_end() wants to know which parts of the target page it | |
1691 | * should complete the write on. It's easiest to compute them ahead of | |
1692 | * time when a more complete view of the write is available. | |
1693 | */ | |
1694 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, | |
1695 | struct ocfs2_write_ctxt *wc, | |
1696 | loff_t pos, unsigned len, int alloc) | |
9517bac6 | 1697 | { |
3a307ffc | 1698 | struct ocfs2_write_cluster_desc *desc; |
9517bac6 | 1699 | |
3a307ffc MF |
1700 | wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1); |
1701 | wc->w_target_to = wc->w_target_from + len; | |
1702 | ||
1703 | if (alloc == 0) | |
1704 | return; | |
1705 | ||
1706 | /* | |
1707 | * Allocating write - we may have different boundaries based | |
1708 | * on page size and cluster size. | |
1709 | * | |
1710 | * NOTE: We can no longer compute one value from the other as | |
1711 | * the actual write length and user provided length may be | |
1712 | * different. | |
1713 | */ | |
9517bac6 | 1714 | |
3a307ffc MF |
1715 | if (wc->w_large_pages) { |
1716 | /* | |
1717 | * We only care about the 1st and last cluster within | |
b27b7cbc | 1718 | * our range and whether they should be zero'd or not. Either |
3a307ffc MF |
1719 | * value may be extended out to the start/end of a |
1720 | * newly allocated cluster. | |
1721 | */ | |
1722 | desc = &wc->w_desc[0]; | |
e7432675 | 1723 | if (desc->c_needs_zero) |
3a307ffc MF |
1724 | ocfs2_figure_cluster_boundaries(osb, |
1725 | desc->c_cpos, | |
1726 | &wc->w_target_from, | |
1727 | NULL); | |
1728 | ||
1729 | desc = &wc->w_desc[wc->w_clen - 1]; | |
e7432675 | 1730 | if (desc->c_needs_zero) |
3a307ffc MF |
1731 | ocfs2_figure_cluster_boundaries(osb, |
1732 | desc->c_cpos, | |
1733 | NULL, | |
1734 | &wc->w_target_to); | |
1735 | } else { | |
1736 | wc->w_target_from = 0; | |
1737 | wc->w_target_to = PAGE_CACHE_SIZE; | |
1738 | } | |
9517bac6 MF |
1739 | } |
1740 | ||
0d172baa MF |
1741 | /* |
1742 | * Populate each single-cluster write descriptor in the write context | |
1743 | * with information about the i/o to be done. | |
b27b7cbc MF |
1744 | * |
1745 | * Returns the number of clusters that will have to be allocated, as | |
1746 | * well as a worst case estimate of the number of extent records that | |
1747 | * would have to be created during a write to an unwritten region. | |
0d172baa MF |
1748 | */ |
1749 | static int ocfs2_populate_write_desc(struct inode *inode, | |
1750 | struct ocfs2_write_ctxt *wc, | |
b27b7cbc MF |
1751 | unsigned int *clusters_to_alloc, |
1752 | unsigned int *extents_to_split) | |
9517bac6 | 1753 | { |
0d172baa | 1754 | int ret; |
3a307ffc | 1755 | struct ocfs2_write_cluster_desc *desc; |
0d172baa | 1756 | unsigned int num_clusters = 0; |
b27b7cbc | 1757 | unsigned int ext_flags = 0; |
0d172baa MF |
1758 | u32 phys = 0; |
1759 | int i; | |
9517bac6 | 1760 | |
b27b7cbc MF |
1761 | *clusters_to_alloc = 0; |
1762 | *extents_to_split = 0; | |
1763 | ||
3a307ffc MF |
1764 | for (i = 0; i < wc->w_clen; i++) { |
1765 | desc = &wc->w_desc[i]; | |
1766 | desc->c_cpos = wc->w_cpos + i; | |
1767 | ||
1768 | if (num_clusters == 0) { | |
b27b7cbc MF |
1769 | /* |
1770 | * Need to look up the next extent record. | |
1771 | */ | |
3a307ffc | 1772 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, |
b27b7cbc | 1773 | &num_clusters, &ext_flags); |
3a307ffc MF |
1774 | if (ret) { |
1775 | mlog_errno(ret); | |
607d44aa | 1776 | goto out; |
3a307ffc | 1777 | } |
b27b7cbc | 1778 | |
293b2f70 TM |
1779 | /* We should already CoW the refcountd extent. */ |
1780 | BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED); | |
1781 | ||
b27b7cbc MF |
1782 | /* |
1783 | * Assume worst case - that we're writing in | |
1784 | * the middle of the extent. | |
1785 | * | |
1786 | * We can assume that the write proceeds from | |
1787 | * left to right, in which case the extent | |
1788 | * insert code is smart enough to coalesce the | |
1789 | * next splits into the previous records created. | |
1790 | */ | |
1791 | if (ext_flags & OCFS2_EXT_UNWRITTEN) | |
1792 | *extents_to_split = *extents_to_split + 2; | |
3a307ffc MF |
1793 | } else if (phys) { |
1794 | /* | |
1795 | * Only increment phys if it doesn't describe | |
1796 | * a hole. | |
1797 | */ | |
1798 | phys++; | |
1799 | } | |
1800 | ||
e7432675 SM |
1801 | /* |
1802 | * If w_first_new_cpos is < UINT_MAX, we have a non-sparse | |
1803 | * file that got extended. w_first_new_cpos tells us | |
1804 | * where the newly allocated clusters are so we can | |
1805 | * zero them. | |
1806 | */ | |
1807 | if (desc->c_cpos >= wc->w_first_new_cpos) { | |
1808 | BUG_ON(phys == 0); | |
1809 | desc->c_needs_zero = 1; | |
1810 | } | |
1811 | ||
3a307ffc MF |
1812 | desc->c_phys = phys; |
1813 | if (phys == 0) { | |
1814 | desc->c_new = 1; | |
e7432675 | 1815 | desc->c_needs_zero = 1; |
0d172baa | 1816 | *clusters_to_alloc = *clusters_to_alloc + 1; |
3a307ffc | 1817 | } |
e7432675 SM |
1818 | |
1819 | if (ext_flags & OCFS2_EXT_UNWRITTEN) { | |
b27b7cbc | 1820 | desc->c_unwritten = 1; |
e7432675 SM |
1821 | desc->c_needs_zero = 1; |
1822 | } | |
3a307ffc MF |
1823 | |
1824 | num_clusters--; | |
9517bac6 MF |
1825 | } |
1826 | ||
0d172baa MF |
1827 | ret = 0; |
1828 | out: | |
1829 | return ret; | |
1830 | } | |
1831 | ||
1afc32b9 MF |
1832 | static int ocfs2_write_begin_inline(struct address_space *mapping, |
1833 | struct inode *inode, | |
1834 | struct ocfs2_write_ctxt *wc) | |
1835 | { | |
1836 | int ret; | |
1837 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1838 | struct page *page; | |
1839 | handle_t *handle; | |
1840 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1841 | ||
f775da2f JB |
1842 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
1843 | if (IS_ERR(handle)) { | |
1844 | ret = PTR_ERR(handle); | |
1845 | mlog_errno(ret); | |
1846 | goto out; | |
1847 | } | |
1848 | ||
1afc32b9 MF |
1849 | page = find_or_create_page(mapping, 0, GFP_NOFS); |
1850 | if (!page) { | |
f775da2f | 1851 | ocfs2_commit_trans(osb, handle); |
1afc32b9 MF |
1852 | ret = -ENOMEM; |
1853 | mlog_errno(ret); | |
1854 | goto out; | |
1855 | } | |
1856 | /* | |
1857 | * If we don't set w_num_pages then this page won't get unlocked | |
1858 | * and freed on cleanup of the write context. | |
1859 | */ | |
1860 | wc->w_pages[0] = wc->w_target_page = page; | |
1861 | wc->w_num_pages = 1; | |
1862 | ||
0cf2f763 | 1863 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 1864 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
1865 | if (ret) { |
1866 | ocfs2_commit_trans(osb, handle); | |
1867 | ||
1868 | mlog_errno(ret); | |
1869 | goto out; | |
1870 | } | |
1871 | ||
1872 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
1873 | ocfs2_set_inode_data_inline(inode, di); | |
1874 | ||
1875 | if (!PageUptodate(page)) { | |
1876 | ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh); | |
1877 | if (ret) { | |
1878 | ocfs2_commit_trans(osb, handle); | |
1879 | ||
1880 | goto out; | |
1881 | } | |
1882 | } | |
1883 | ||
1884 | wc->w_handle = handle; | |
1885 | out: | |
1886 | return ret; | |
1887 | } | |
1888 | ||
1889 | int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size) | |
1890 | { | |
1891 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1892 | ||
0d8a4e0c | 1893 | if (new_size <= le16_to_cpu(di->id2.i_data.id_count)) |
1afc32b9 MF |
1894 | return 1; |
1895 | return 0; | |
1896 | } | |
1897 | ||
1898 | static int ocfs2_try_to_write_inline_data(struct address_space *mapping, | |
1899 | struct inode *inode, loff_t pos, | |
1900 | unsigned len, struct page *mmap_page, | |
1901 | struct ocfs2_write_ctxt *wc) | |
1902 | { | |
1903 | int ret, written = 0; | |
1904 | loff_t end = pos + len; | |
1905 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
d9ae49d6 | 1906 | struct ocfs2_dinode *di = NULL; |
1afc32b9 | 1907 | |
9558156b TM |
1908 | trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno, |
1909 | len, (unsigned long long)pos, | |
1910 | oi->ip_dyn_features); | |
1afc32b9 MF |
1911 | |
1912 | /* | |
1913 | * Handle inodes which already have inline data 1st. | |
1914 | */ | |
1915 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1916 | if (mmap_page == NULL && | |
1917 | ocfs2_size_fits_inline_data(wc->w_di_bh, end)) | |
1918 | goto do_inline_write; | |
1919 | ||
1920 | /* | |
1921 | * The write won't fit - we have to give this inode an | |
1922 | * inline extent list now. | |
1923 | */ | |
1924 | ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh); | |
1925 | if (ret) | |
1926 | mlog_errno(ret); | |
1927 | goto out; | |
1928 | } | |
1929 | ||
1930 | /* | |
1931 | * Check whether the inode can accept inline data. | |
1932 | */ | |
1933 | if (oi->ip_clusters != 0 || i_size_read(inode) != 0) | |
1934 | return 0; | |
1935 | ||
1936 | /* | |
1937 | * Check whether the write can fit. | |
1938 | */ | |
d9ae49d6 TY |
1939 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; |
1940 | if (mmap_page || | |
1941 | end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) | |
1afc32b9 MF |
1942 | return 0; |
1943 | ||
1944 | do_inline_write: | |
1945 | ret = ocfs2_write_begin_inline(mapping, inode, wc); | |
1946 | if (ret) { | |
1947 | mlog_errno(ret); | |
1948 | goto out; | |
1949 | } | |
1950 | ||
1951 | /* | |
1952 | * This signals to the caller that the data can be written | |
1953 | * inline. | |
1954 | */ | |
1955 | written = 1; | |
1956 | out: | |
1957 | return written ? written : ret; | |
1958 | } | |
1959 | ||
65ed39d6 MF |
1960 | /* |
1961 | * This function only does anything for file systems which can't | |
1962 | * handle sparse files. | |
1963 | * | |
1964 | * What we want to do here is fill in any hole between the current end | |
1965 | * of allocation and the end of our write. That way the rest of the | |
1966 | * write path can treat it as an non-allocating write, which has no | |
1967 | * special case code for sparse/nonsparse files. | |
1968 | */ | |
5693486b JB |
1969 | static int ocfs2_expand_nonsparse_inode(struct inode *inode, |
1970 | struct buffer_head *di_bh, | |
1971 | loff_t pos, unsigned len, | |
65ed39d6 MF |
1972 | struct ocfs2_write_ctxt *wc) |
1973 | { | |
1974 | int ret; | |
65ed39d6 MF |
1975 | loff_t newsize = pos + len; |
1976 | ||
5693486b | 1977 | BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); |
65ed39d6 MF |
1978 | |
1979 | if (newsize <= i_size_read(inode)) | |
1980 | return 0; | |
1981 | ||
5693486b | 1982 | ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos); |
65ed39d6 MF |
1983 | if (ret) |
1984 | mlog_errno(ret); | |
1985 | ||
e7432675 SM |
1986 | wc->w_first_new_cpos = |
1987 | ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)); | |
1988 | ||
65ed39d6 MF |
1989 | return ret; |
1990 | } | |
1991 | ||
5693486b JB |
1992 | static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh, |
1993 | loff_t pos) | |
1994 | { | |
1995 | int ret = 0; | |
1996 | ||
1997 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); | |
1998 | if (pos > i_size_read(inode)) | |
1999 | ret = ocfs2_zero_extend(inode, di_bh, pos); | |
2000 | ||
2001 | return ret; | |
2002 | } | |
2003 | ||
50308d81 TM |
2004 | /* |
2005 | * Try to flush truncate logs if we can free enough clusters from it. | |
2006 | * As for return value, "< 0" means error, "0" no space and "1" means | |
2007 | * we have freed enough spaces and let the caller try to allocate again. | |
2008 | */ | |
2009 | static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb, | |
2010 | unsigned int needed) | |
2011 | { | |
2012 | tid_t target; | |
2013 | int ret = 0; | |
2014 | unsigned int truncated_clusters; | |
2015 | ||
2016 | mutex_lock(&osb->osb_tl_inode->i_mutex); | |
2017 | truncated_clusters = osb->truncated_clusters; | |
2018 | mutex_unlock(&osb->osb_tl_inode->i_mutex); | |
2019 | ||
2020 | /* | |
2021 | * Check whether we can succeed in allocating if we free | |
2022 | * the truncate log. | |
2023 | */ | |
2024 | if (truncated_clusters < needed) | |
2025 | goto out; | |
2026 | ||
2027 | ret = ocfs2_flush_truncate_log(osb); | |
2028 | if (ret) { | |
2029 | mlog_errno(ret); | |
2030 | goto out; | |
2031 | } | |
2032 | ||
2033 | if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) { | |
2034 | jbd2_log_wait_commit(osb->journal->j_journal, target); | |
2035 | ret = 1; | |
2036 | } | |
2037 | out: | |
2038 | return ret; | |
2039 | } | |
2040 | ||
0378da0f TM |
2041 | int ocfs2_write_begin_nolock(struct file *filp, |
2042 | struct address_space *mapping, | |
0d172baa MF |
2043 | loff_t pos, unsigned len, unsigned flags, |
2044 | struct page **pagep, void **fsdata, | |
2045 | struct buffer_head *di_bh, struct page *mmap_page) | |
2046 | { | |
e7432675 | 2047 | int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS; |
50308d81 | 2048 | unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0; |
0d172baa MF |
2049 | struct ocfs2_write_ctxt *wc; |
2050 | struct inode *inode = mapping->host; | |
2051 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2052 | struct ocfs2_dinode *di; | |
2053 | struct ocfs2_alloc_context *data_ac = NULL; | |
2054 | struct ocfs2_alloc_context *meta_ac = NULL; | |
2055 | handle_t *handle; | |
f99b9b7c | 2056 | struct ocfs2_extent_tree et; |
50308d81 | 2057 | int try_free = 1, ret1; |
0d172baa | 2058 | |
50308d81 | 2059 | try_again: |
0d172baa MF |
2060 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh); |
2061 | if (ret) { | |
2062 | mlog_errno(ret); | |
2063 | return ret; | |
2064 | } | |
2065 | ||
1afc32b9 MF |
2066 | if (ocfs2_supports_inline_data(osb)) { |
2067 | ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len, | |
2068 | mmap_page, wc); | |
2069 | if (ret == 1) { | |
2070 | ret = 0; | |
2071 | goto success; | |
2072 | } | |
2073 | if (ret < 0) { | |
2074 | mlog_errno(ret); | |
2075 | goto out; | |
2076 | } | |
2077 | } | |
2078 | ||
5693486b JB |
2079 | if (ocfs2_sparse_alloc(osb)) |
2080 | ret = ocfs2_zero_tail(inode, di_bh, pos); | |
2081 | else | |
2082 | ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len, | |
2083 | wc); | |
65ed39d6 MF |
2084 | if (ret) { |
2085 | mlog_errno(ret); | |
2086 | goto out; | |
2087 | } | |
2088 | ||
293b2f70 TM |
2089 | ret = ocfs2_check_range_for_refcount(inode, pos, len); |
2090 | if (ret < 0) { | |
2091 | mlog_errno(ret); | |
2092 | goto out; | |
2093 | } else if (ret == 1) { | |
50308d81 | 2094 | clusters_need = wc->w_clen; |
c7dd3392 | 2095 | ret = ocfs2_refcount_cow(inode, di_bh, |
37f8a2bf | 2096 | wc->w_cpos, wc->w_clen, UINT_MAX); |
293b2f70 TM |
2097 | if (ret) { |
2098 | mlog_errno(ret); | |
2099 | goto out; | |
2100 | } | |
2101 | } | |
2102 | ||
b27b7cbc MF |
2103 | ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc, |
2104 | &extents_to_split); | |
0d172baa MF |
2105 | if (ret) { |
2106 | mlog_errno(ret); | |
2107 | goto out; | |
2108 | } | |
50308d81 | 2109 | clusters_need += clusters_to_alloc; |
0d172baa MF |
2110 | |
2111 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
2112 | ||
9558156b TM |
2113 | trace_ocfs2_write_begin_nolock( |
2114 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2115 | (long long)i_size_read(inode), | |
2116 | le32_to_cpu(di->i_clusters), | |
2117 | pos, len, flags, mmap_page, | |
2118 | clusters_to_alloc, extents_to_split); | |
2119 | ||
3a307ffc MF |
2120 | /* |
2121 | * We set w_target_from, w_target_to here so that | |
2122 | * ocfs2_write_end() knows which range in the target page to | |
2123 | * write out. An allocation requires that we write the entire | |
2124 | * cluster range. | |
2125 | */ | |
b27b7cbc | 2126 | if (clusters_to_alloc || extents_to_split) { |
3a307ffc MF |
2127 | /* |
2128 | * XXX: We are stretching the limits of | |
b27b7cbc | 2129 | * ocfs2_lock_allocators(). It greatly over-estimates |
3a307ffc MF |
2130 | * the work to be done. |
2131 | */ | |
5e404e9e JB |
2132 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
2133 | wc->w_di_bh); | |
f99b9b7c | 2134 | ret = ocfs2_lock_allocators(inode, &et, |
231b87d1 | 2135 | clusters_to_alloc, extents_to_split, |
f99b9b7c | 2136 | &data_ac, &meta_ac); |
9517bac6 MF |
2137 | if (ret) { |
2138 | mlog_errno(ret); | |
607d44aa | 2139 | goto out; |
9517bac6 MF |
2140 | } |
2141 | ||
4fe370af MF |
2142 | if (data_ac) |
2143 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; | |
2144 | ||
811f933d | 2145 | credits = ocfs2_calc_extend_credits(inode->i_sb, |
06f9da6e | 2146 | &di->id2.i_list); |
3a307ffc | 2147 | |
9517bac6 MF |
2148 | } |
2149 | ||
e7432675 SM |
2150 | /* |
2151 | * We have to zero sparse allocated clusters, unwritten extent clusters, | |
2152 | * and non-sparse clusters we just extended. For non-sparse writes, | |
2153 | * we know zeros will only be needed in the first and/or last cluster. | |
2154 | */ | |
2155 | if (clusters_to_alloc || extents_to_split || | |
8379e7c4 SM |
2156 | (wc->w_clen && (wc->w_desc[0].c_needs_zero || |
2157 | wc->w_desc[wc->w_clen - 1].c_needs_zero))) | |
e7432675 SM |
2158 | cluster_of_pages = 1; |
2159 | else | |
2160 | cluster_of_pages = 0; | |
2161 | ||
2162 | ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages); | |
3a307ffc | 2163 | |
9517bac6 MF |
2164 | handle = ocfs2_start_trans(osb, credits); |
2165 | if (IS_ERR(handle)) { | |
2166 | ret = PTR_ERR(handle); | |
2167 | mlog_errno(ret); | |
607d44aa | 2168 | goto out; |
9517bac6 MF |
2169 | } |
2170 | ||
3a307ffc MF |
2171 | wc->w_handle = handle; |
2172 | ||
5dd4056d CH |
2173 | if (clusters_to_alloc) { |
2174 | ret = dquot_alloc_space_nodirty(inode, | |
2175 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); | |
2176 | if (ret) | |
2177 | goto out_commit; | |
a90714c1 | 2178 | } |
3a307ffc MF |
2179 | /* |
2180 | * We don't want this to fail in ocfs2_write_end(), so do it | |
2181 | * here. | |
2182 | */ | |
0cf2f763 | 2183 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 2184 | OCFS2_JOURNAL_ACCESS_WRITE); |
3a307ffc | 2185 | if (ret) { |
9517bac6 | 2186 | mlog_errno(ret); |
a90714c1 | 2187 | goto out_quota; |
9517bac6 MF |
2188 | } |
2189 | ||
3a307ffc MF |
2190 | /* |
2191 | * Fill our page array first. That way we've grabbed enough so | |
2192 | * that we can zero and flush if we error after adding the | |
2193 | * extent. | |
2194 | */ | |
693c241a | 2195 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len, |
e7432675 | 2196 | cluster_of_pages, mmap_page); |
5cffff9e | 2197 | if (ret && ret != -EAGAIN) { |
9517bac6 | 2198 | mlog_errno(ret); |
a90714c1 | 2199 | goto out_quota; |
9517bac6 MF |
2200 | } |
2201 | ||
5cffff9e WW |
2202 | /* |
2203 | * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock | |
2204 | * the target page. In this case, we exit with no error and no target | |
2205 | * page. This will trigger the caller, page_mkwrite(), to re-try | |
2206 | * the operation. | |
2207 | */ | |
2208 | if (ret == -EAGAIN) { | |
2209 | BUG_ON(wc->w_target_page); | |
2210 | ret = 0; | |
2211 | goto out_quota; | |
2212 | } | |
2213 | ||
0d172baa MF |
2214 | ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos, |
2215 | len); | |
2216 | if (ret) { | |
2217 | mlog_errno(ret); | |
a90714c1 | 2218 | goto out_quota; |
9517bac6 | 2219 | } |
9517bac6 | 2220 | |
3a307ffc MF |
2221 | if (data_ac) |
2222 | ocfs2_free_alloc_context(data_ac); | |
2223 | if (meta_ac) | |
2224 | ocfs2_free_alloc_context(meta_ac); | |
9517bac6 | 2225 | |
1afc32b9 | 2226 | success: |
3a307ffc MF |
2227 | *pagep = wc->w_target_page; |
2228 | *fsdata = wc; | |
2229 | return 0; | |
a90714c1 JK |
2230 | out_quota: |
2231 | if (clusters_to_alloc) | |
5dd4056d | 2232 | dquot_free_space(inode, |
a90714c1 | 2233 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); |
9517bac6 MF |
2234 | out_commit: |
2235 | ocfs2_commit_trans(osb, handle); | |
2236 | ||
9517bac6 | 2237 | out: |
3a307ffc MF |
2238 | ocfs2_free_write_ctxt(wc); |
2239 | ||
b1214e47 | 2240 | if (data_ac) { |
9517bac6 | 2241 | ocfs2_free_alloc_context(data_ac); |
b1214e47 X |
2242 | data_ac = NULL; |
2243 | } | |
2244 | if (meta_ac) { | |
9517bac6 | 2245 | ocfs2_free_alloc_context(meta_ac); |
b1214e47 X |
2246 | meta_ac = NULL; |
2247 | } | |
50308d81 TM |
2248 | |
2249 | if (ret == -ENOSPC && try_free) { | |
2250 | /* | |
2251 | * Try to free some truncate log so that we can have enough | |
2252 | * clusters to allocate. | |
2253 | */ | |
2254 | try_free = 0; | |
2255 | ||
2256 | ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); | |
2257 | if (ret1 == 1) | |
2258 | goto try_again; | |
2259 | ||
2260 | if (ret1 < 0) | |
2261 | mlog_errno(ret1); | |
2262 | } | |
2263 | ||
3a307ffc MF |
2264 | return ret; |
2265 | } | |
2266 | ||
b6af1bcd NP |
2267 | static int ocfs2_write_begin(struct file *file, struct address_space *mapping, |
2268 | loff_t pos, unsigned len, unsigned flags, | |
2269 | struct page **pagep, void **fsdata) | |
607d44aa MF |
2270 | { |
2271 | int ret; | |
2272 | struct buffer_head *di_bh = NULL; | |
2273 | struct inode *inode = mapping->host; | |
2274 | ||
e63aecb6 | 2275 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
607d44aa MF |
2276 | if (ret) { |
2277 | mlog_errno(ret); | |
2278 | return ret; | |
2279 | } | |
2280 | ||
2281 | /* | |
2282 | * Take alloc sem here to prevent concurrent lookups. That way | |
2283 | * the mapping, zeroing and tree manipulation within | |
2284 | * ocfs2_write() will be safe against ->readpage(). This | |
2285 | * should also serve to lock out allocation from a shared | |
2286 | * writeable region. | |
2287 | */ | |
2288 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
2289 | ||
0378da0f | 2290 | ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep, |
7307de80 | 2291 | fsdata, di_bh, NULL); |
607d44aa MF |
2292 | if (ret) { |
2293 | mlog_errno(ret); | |
c934a92d | 2294 | goto out_fail; |
607d44aa MF |
2295 | } |
2296 | ||
2297 | brelse(di_bh); | |
2298 | ||
2299 | return 0; | |
2300 | ||
607d44aa MF |
2301 | out_fail: |
2302 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
2303 | ||
2304 | brelse(di_bh); | |
e63aecb6 | 2305 | ocfs2_inode_unlock(inode, 1); |
607d44aa MF |
2306 | |
2307 | return ret; | |
2308 | } | |
2309 | ||
1afc32b9 MF |
2310 | static void ocfs2_write_end_inline(struct inode *inode, loff_t pos, |
2311 | unsigned len, unsigned *copied, | |
2312 | struct ocfs2_dinode *di, | |
2313 | struct ocfs2_write_ctxt *wc) | |
2314 | { | |
2315 | void *kaddr; | |
2316 | ||
2317 | if (unlikely(*copied < len)) { | |
2318 | if (!PageUptodate(wc->w_target_page)) { | |
2319 | *copied = 0; | |
2320 | return; | |
2321 | } | |
2322 | } | |
2323 | ||
c4bc8dcb | 2324 | kaddr = kmap_atomic(wc->w_target_page); |
1afc32b9 | 2325 | memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied); |
c4bc8dcb | 2326 | kunmap_atomic(kaddr); |
1afc32b9 | 2327 | |
9558156b TM |
2328 | trace_ocfs2_write_end_inline( |
2329 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1afc32b9 MF |
2330 | (unsigned long long)pos, *copied, |
2331 | le16_to_cpu(di->id2.i_data.id_count), | |
2332 | le16_to_cpu(di->i_dyn_features)); | |
2333 | } | |
2334 | ||
7307de80 MF |
2335 | int ocfs2_write_end_nolock(struct address_space *mapping, |
2336 | loff_t pos, unsigned len, unsigned copied, | |
2337 | struct page *page, void *fsdata) | |
3a307ffc MF |
2338 | { |
2339 | int i; | |
2340 | unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1); | |
2341 | struct inode *inode = mapping->host; | |
2342 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2343 | struct ocfs2_write_ctxt *wc = fsdata; | |
2344 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
2345 | handle_t *handle = wc->w_handle; | |
2346 | struct page *tmppage; | |
2347 | ||
1afc32b9 MF |
2348 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
2349 | ocfs2_write_end_inline(inode, pos, len, &copied, di, wc); | |
2350 | goto out_write_size; | |
2351 | } | |
2352 | ||
3a307ffc MF |
2353 | if (unlikely(copied < len)) { |
2354 | if (!PageUptodate(wc->w_target_page)) | |
2355 | copied = 0; | |
2356 | ||
2357 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, | |
2358 | start+len); | |
2359 | } | |
2360 | flush_dcache_page(wc->w_target_page); | |
2361 | ||
2362 | for(i = 0; i < wc->w_num_pages; i++) { | |
2363 | tmppage = wc->w_pages[i]; | |
2364 | ||
2365 | if (tmppage == wc->w_target_page) { | |
2366 | from = wc->w_target_from; | |
2367 | to = wc->w_target_to; | |
2368 | ||
2369 | BUG_ON(from > PAGE_CACHE_SIZE || | |
2370 | to > PAGE_CACHE_SIZE || | |
2371 | to < from); | |
2372 | } else { | |
2373 | /* | |
2374 | * Pages adjacent to the target (if any) imply | |
2375 | * a hole-filling write in which case we want | |
2376 | * to flush their entire range. | |
2377 | */ | |
2378 | from = 0; | |
2379 | to = PAGE_CACHE_SIZE; | |
2380 | } | |
2381 | ||
961cecbe | 2382 | if (page_has_buffers(tmppage)) { |
53ef99ca | 2383 | if (ocfs2_should_order_data(inode)) |
2b4e30fb | 2384 | ocfs2_jbd2_file_inode(wc->w_handle, inode); |
961cecbe SM |
2385 | block_commit_write(tmppage, from, to); |
2386 | } | |
3a307ffc MF |
2387 | } |
2388 | ||
1afc32b9 | 2389 | out_write_size: |
3a307ffc | 2390 | pos += copied; |
f17c20dd | 2391 | if (pos > i_size_read(inode)) { |
3a307ffc MF |
2392 | i_size_write(inode, pos); |
2393 | mark_inode_dirty(inode); | |
2394 | } | |
2395 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
2396 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
2397 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
2398 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); | |
2399 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
2931cdcb | 2400 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
3a307ffc MF |
2401 | ocfs2_journal_dirty(handle, wc->w_di_bh); |
2402 | ||
136f49b9 JB |
2403 | /* unlock pages before dealloc since it needs acquiring j_trans_barrier |
2404 | * lock, or it will cause a deadlock since journal commit threads holds | |
2405 | * this lock and will ask for the page lock when flushing the data. | |
2406 | * put it here to preserve the unlock order. | |
2407 | */ | |
2408 | ocfs2_unlock_pages(wc); | |
2409 | ||
3a307ffc | 2410 | ocfs2_commit_trans(osb, handle); |
59a5e416 | 2411 | |
b27b7cbc MF |
2412 | ocfs2_run_deallocs(osb, &wc->w_dealloc); |
2413 | ||
136f49b9 JB |
2414 | brelse(wc->w_di_bh); |
2415 | kfree(wc); | |
607d44aa MF |
2416 | |
2417 | return copied; | |
2418 | } | |
2419 | ||
b6af1bcd NP |
2420 | static int ocfs2_write_end(struct file *file, struct address_space *mapping, |
2421 | loff_t pos, unsigned len, unsigned copied, | |
2422 | struct page *page, void *fsdata) | |
607d44aa MF |
2423 | { |
2424 | int ret; | |
2425 | struct inode *inode = mapping->host; | |
2426 | ||
2427 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata); | |
2428 | ||
3a307ffc | 2429 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
e63aecb6 | 2430 | ocfs2_inode_unlock(inode, 1); |
9517bac6 | 2431 | |
607d44aa | 2432 | return ret; |
9517bac6 MF |
2433 | } |
2434 | ||
f5e54d6e | 2435 | const struct address_space_operations ocfs2_aops = { |
1fca3a05 HH |
2436 | .readpage = ocfs2_readpage, |
2437 | .readpages = ocfs2_readpages, | |
2438 | .writepage = ocfs2_writepage, | |
2439 | .write_begin = ocfs2_write_begin, | |
2440 | .write_end = ocfs2_write_end, | |
2441 | .bmap = ocfs2_bmap, | |
1fca3a05 | 2442 | .direct_IO = ocfs2_direct_IO, |
41ecc345 | 2443 | .invalidatepage = block_invalidatepage, |
1fca3a05 HH |
2444 | .releasepage = ocfs2_releasepage, |
2445 | .migratepage = buffer_migrate_page, | |
2446 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 2447 | .error_remove_page = generic_error_remove_page, |
ccd979bd | 2448 | }; |