Commit | Line | Data |
---|---|---|
328970de | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
fa60ce2c | 2 | /* |
ccd979bd | 3 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. |
ccd979bd MF |
4 | */ |
5 | ||
6 | #include <linux/fs.h> | |
7 | #include <linux/slab.h> | |
8 | #include <linux/highmem.h> | |
9 | #include <linux/pagemap.h> | |
10 | #include <asm/byteorder.h> | |
9517bac6 | 11 | #include <linux/swap.h> |
628a24f5 | 12 | #include <linux/mpage.h> |
a90714c1 | 13 | #include <linux/quotaops.h> |
24c40b32 | 14 | #include <linux/blkdev.h> |
e2e40f2c | 15 | #include <linux/uio.h> |
f86196ea | 16 | #include <linux/mm.h> |
ccd979bd | 17 | |
ccd979bd MF |
18 | #include <cluster/masklog.h> |
19 | ||
20 | #include "ocfs2.h" | |
21 | ||
22 | #include "alloc.h" | |
23 | #include "aops.h" | |
24 | #include "dlmglue.h" | |
25 | #include "extent_map.h" | |
26 | #include "file.h" | |
27 | #include "inode.h" | |
28 | #include "journal.h" | |
9517bac6 | 29 | #include "suballoc.h" |
ccd979bd MF |
30 | #include "super.h" |
31 | #include "symlink.h" | |
293b2f70 | 32 | #include "refcounttree.h" |
9558156b | 33 | #include "ocfs2_trace.h" |
ccd979bd MF |
34 | |
35 | #include "buffer_head_io.h" | |
24c40b32 JQ |
36 | #include "dir.h" |
37 | #include "namei.h" | |
38 | #include "sysfile.h" | |
ccd979bd MF |
39 | |
40 | static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock, | |
41 | struct buffer_head *bh_result, int create) | |
42 | { | |
43 | int err = -EIO; | |
44 | int status; | |
45 | struct ocfs2_dinode *fe = NULL; | |
46 | struct buffer_head *bh = NULL; | |
47 | struct buffer_head *buffer_cache_bh = NULL; | |
48 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
49 | void *kaddr; | |
50 | ||
9558156b TM |
51 | trace_ocfs2_symlink_get_block( |
52 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
53 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
54 | |
55 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); | |
56 | ||
57 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { | |
58 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", | |
59 | (unsigned long long)iblock); | |
60 | goto bail; | |
61 | } | |
62 | ||
b657c95c | 63 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
64 | if (status < 0) { |
65 | mlog_errno(status); | |
66 | goto bail; | |
67 | } | |
68 | fe = (struct ocfs2_dinode *) bh->b_data; | |
69 | ||
ccd979bd MF |
70 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, |
71 | le32_to_cpu(fe->i_clusters))) { | |
7391a294 | 72 | err = -ENOMEM; |
ccd979bd MF |
73 | mlog(ML_ERROR, "block offset is outside the allocated size: " |
74 | "%llu\n", (unsigned long long)iblock); | |
75 | goto bail; | |
76 | } | |
77 | ||
78 | /* We don't use the page cache to create symlink data, so if | |
79 | * need be, copy it over from the buffer cache. */ | |
80 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { | |
81 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + | |
82 | iblock; | |
83 | buffer_cache_bh = sb_getblk(osb->sb, blkno); | |
84 | if (!buffer_cache_bh) { | |
7391a294 | 85 | err = -ENOMEM; |
ccd979bd MF |
86 | mlog(ML_ERROR, "couldn't getblock for symlink!\n"); |
87 | goto bail; | |
88 | } | |
89 | ||
90 | /* we haven't locked out transactions, so a commit | |
91 | * could've happened. Since we've got a reference on | |
92 | * the bh, even if it commits while we're doing the | |
93 | * copy, the data is still good. */ | |
94 | if (buffer_jbd(buffer_cache_bh) | |
95 | && ocfs2_inode_is_new(inode)) { | |
c4bc8dcb | 96 | kaddr = kmap_atomic(bh_result->b_page); |
ccd979bd MF |
97 | if (!kaddr) { |
98 | mlog(ML_ERROR, "couldn't kmap!\n"); | |
99 | goto bail; | |
100 | } | |
101 | memcpy(kaddr + (bh_result->b_size * iblock), | |
102 | buffer_cache_bh->b_data, | |
103 | bh_result->b_size); | |
c4bc8dcb | 104 | kunmap_atomic(kaddr); |
ccd979bd MF |
105 | set_buffer_uptodate(bh_result); |
106 | } | |
107 | brelse(buffer_cache_bh); | |
108 | } | |
109 | ||
110 | map_bh(bh_result, inode->i_sb, | |
111 | le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); | |
112 | ||
113 | err = 0; | |
114 | ||
115 | bail: | |
a81cb88b | 116 | brelse(bh); |
ccd979bd | 117 | |
ccd979bd MF |
118 | return err; |
119 | } | |
120 | ||
3e4c56d4 | 121 | static int ocfs2_lock_get_block(struct inode *inode, sector_t iblock, |
122 | struct buffer_head *bh_result, int create) | |
123 | { | |
124 | int ret = 0; | |
125 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
126 | ||
127 | down_read(&oi->ip_alloc_sem); | |
128 | ret = ocfs2_get_block(inode, iblock, bh_result, create); | |
129 | up_read(&oi->ip_alloc_sem); | |
130 | ||
131 | return ret; | |
132 | } | |
133 | ||
6f70fa51 TM |
134 | int ocfs2_get_block(struct inode *inode, sector_t iblock, |
135 | struct buffer_head *bh_result, int create) | |
ccd979bd MF |
136 | { |
137 | int err = 0; | |
49cb8d2d | 138 | unsigned int ext_flags; |
628a24f5 MF |
139 | u64 max_blocks = bh_result->b_size >> inode->i_blkbits; |
140 | u64 p_blkno, count, past_eof; | |
25baf2da | 141 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd | 142 | |
9558156b TM |
143 | trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno, |
144 | (unsigned long long)iblock, bh_result, create); | |
ccd979bd MF |
145 | |
146 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) | |
147 | mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", | |
148 | inode, inode->i_ino); | |
149 | ||
150 | if (S_ISLNK(inode->i_mode)) { | |
151 | /* this always does I/O for some reason. */ | |
152 | err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); | |
153 | goto bail; | |
154 | } | |
155 | ||
628a24f5 | 156 | err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count, |
49cb8d2d | 157 | &ext_flags); |
ccd979bd MF |
158 | if (err) { |
159 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " | |
b0697053 MF |
160 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, |
161 | (unsigned long long)p_blkno); | |
ccd979bd MF |
162 | goto bail; |
163 | } | |
164 | ||
628a24f5 MF |
165 | if (max_blocks < count) |
166 | count = max_blocks; | |
167 | ||
25baf2da MF |
168 | /* |
169 | * ocfs2 never allocates in this function - the only time we | |
170 | * need to use BH_New is when we're extending i_size on a file | |
171 | * system which doesn't support holes, in which case BH_New | |
ebdec241 | 172 | * allows __block_write_begin() to zero. |
c0420ad2 CL |
173 | * |
174 | * If we see this on a sparse file system, then a truncate has | |
175 | * raced us and removed the cluster. In this case, we clear | |
176 | * the buffers dirty and uptodate bits and let the buffer code | |
177 | * ignore it as a hole. | |
25baf2da | 178 | */ |
c0420ad2 CL |
179 | if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) { |
180 | clear_buffer_dirty(bh_result); | |
181 | clear_buffer_uptodate(bh_result); | |
182 | goto bail; | |
183 | } | |
25baf2da | 184 | |
49cb8d2d MF |
185 | /* Treat the unwritten extent as a hole for zeroing purposes. */ |
186 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | |
25baf2da MF |
187 | map_bh(bh_result, inode->i_sb, p_blkno); |
188 | ||
628a24f5 MF |
189 | bh_result->b_size = count << inode->i_blkbits; |
190 | ||
25baf2da MF |
191 | if (!ocfs2_sparse_alloc(osb)) { |
192 | if (p_blkno == 0) { | |
193 | err = -EIO; | |
194 | mlog(ML_ERROR, | |
195 | "iblock = %llu p_blkno = %llu blkno=(%llu)\n", | |
196 | (unsigned long long)iblock, | |
197 | (unsigned long long)p_blkno, | |
198 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
199 | mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters); | |
200 | dump_stack(); | |
1f4cea37 | 201 | goto bail; |
25baf2da | 202 | } |
25baf2da | 203 | } |
ccd979bd | 204 | |
5693486b | 205 | past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
9558156b TM |
206 | |
207 | trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno, | |
208 | (unsigned long long)past_eof); | |
5693486b JB |
209 | if (create && (iblock >= past_eof)) |
210 | set_buffer_new(bh_result); | |
211 | ||
ccd979bd MF |
212 | bail: |
213 | if (err < 0) | |
214 | err = -EIO; | |
215 | ||
ccd979bd MF |
216 | return err; |
217 | } | |
218 | ||
1afc32b9 MF |
219 | int ocfs2_read_inline_data(struct inode *inode, struct page *page, |
220 | struct buffer_head *di_bh) | |
6798d35a MF |
221 | { |
222 | void *kaddr; | |
d2849fb2 | 223 | loff_t size; |
6798d35a MF |
224 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
225 | ||
226 | if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) { | |
7ecef14a | 227 | ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n", |
6798d35a MF |
228 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
229 | return -EROFS; | |
230 | } | |
231 | ||
232 | size = i_size_read(inode); | |
233 | ||
09cbfeaf | 234 | if (size > PAGE_SIZE || |
d9ae49d6 | 235 | size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) { |
6798d35a | 236 | ocfs2_error(inode->i_sb, |
7ecef14a | 237 | "Inode %llu has with inline data has bad size: %Lu\n", |
d2849fb2 JK |
238 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
239 | (unsigned long long)size); | |
6798d35a MF |
240 | return -EROFS; |
241 | } | |
242 | ||
c4bc8dcb | 243 | kaddr = kmap_atomic(page); |
6798d35a MF |
244 | if (size) |
245 | memcpy(kaddr, di->id2.i_data.id_data, size); | |
246 | /* Clear the remaining part of the page */ | |
09cbfeaf | 247 | memset(kaddr + size, 0, PAGE_SIZE - size); |
6798d35a | 248 | flush_dcache_page(page); |
c4bc8dcb | 249 | kunmap_atomic(kaddr); |
6798d35a MF |
250 | |
251 | SetPageUptodate(page); | |
252 | ||
253 | return 0; | |
254 | } | |
255 | ||
256 | static int ocfs2_readpage_inline(struct inode *inode, struct page *page) | |
257 | { | |
258 | int ret; | |
259 | struct buffer_head *di_bh = NULL; | |
6798d35a MF |
260 | |
261 | BUG_ON(!PageLocked(page)); | |
86c838b0 | 262 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)); |
6798d35a | 263 | |
b657c95c | 264 | ret = ocfs2_read_inode_block(inode, &di_bh); |
6798d35a MF |
265 | if (ret) { |
266 | mlog_errno(ret); | |
267 | goto out; | |
268 | } | |
269 | ||
270 | ret = ocfs2_read_inline_data(inode, page, di_bh); | |
271 | out: | |
272 | unlock_page(page); | |
273 | ||
274 | brelse(di_bh); | |
275 | return ret; | |
276 | } | |
277 | ||
ccd979bd MF |
278 | static int ocfs2_readpage(struct file *file, struct page *page) |
279 | { | |
280 | struct inode *inode = page->mapping->host; | |
6798d35a | 281 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
09cbfeaf | 282 | loff_t start = (loff_t)page->index << PAGE_SHIFT; |
ccd979bd MF |
283 | int ret, unlock = 1; |
284 | ||
9558156b TM |
285 | trace_ocfs2_readpage((unsigned long long)oi->ip_blkno, |
286 | (page ? page->index : 0)); | |
ccd979bd | 287 | |
e63aecb6 | 288 | ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page); |
ccd979bd MF |
289 | if (ret != 0) { |
290 | if (ret == AOP_TRUNCATED_PAGE) | |
291 | unlock = 0; | |
292 | mlog_errno(ret); | |
293 | goto out; | |
294 | } | |
295 | ||
6798d35a | 296 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) { |
c7e25e6e JK |
297 | /* |
298 | * Unlock the page and cycle ip_alloc_sem so that we don't | |
299 | * busyloop waiting for ip_alloc_sem to unlock | |
300 | */ | |
e9dfc0b2 | 301 | ret = AOP_TRUNCATED_PAGE; |
c7e25e6e JK |
302 | unlock_page(page); |
303 | unlock = 0; | |
304 | down_read(&oi->ip_alloc_sem); | |
305 | up_read(&oi->ip_alloc_sem); | |
e63aecb6 | 306 | goto out_inode_unlock; |
e9dfc0b2 | 307 | } |
ccd979bd MF |
308 | |
309 | /* | |
310 | * i_size might have just been updated as we grabed the meta lock. We | |
311 | * might now be discovering a truncate that hit on another node. | |
312 | * block_read_full_page->get_block freaks out if it is asked to read | |
313 | * beyond the end of a file, so we check here. Callers | |
54cb8821 | 314 | * (generic_file_read, vm_ops->fault) are clever enough to check i_size |
ccd979bd MF |
315 | * and notice that the page they just read isn't needed. |
316 | * | |
317 | * XXX sys_readahead() seems to get that wrong? | |
318 | */ | |
319 | if (start >= i_size_read(inode)) { | |
eebd2aa3 | 320 | zero_user(page, 0, PAGE_SIZE); |
ccd979bd MF |
321 | SetPageUptodate(page); |
322 | ret = 0; | |
323 | goto out_alloc; | |
324 | } | |
325 | ||
6798d35a MF |
326 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
327 | ret = ocfs2_readpage_inline(inode, page); | |
328 | else | |
329 | ret = block_read_full_page(page, ocfs2_get_block); | |
ccd979bd MF |
330 | unlock = 0; |
331 | ||
ccd979bd | 332 | out_alloc: |
d324cd4c | 333 | up_read(&oi->ip_alloc_sem); |
e63aecb6 MF |
334 | out_inode_unlock: |
335 | ocfs2_inode_unlock(inode, 0); | |
ccd979bd MF |
336 | out: |
337 | if (unlock) | |
338 | unlock_page(page); | |
ccd979bd MF |
339 | return ret; |
340 | } | |
341 | ||
628a24f5 MF |
342 | /* |
343 | * This is used only for read-ahead. Failures or difficult to handle | |
344 | * situations are safe to ignore. | |
345 | * | |
346 | * Right now, we don't bother with BH_Boundary - in-inode extent lists | |
347 | * are quite large (243 extents on 4k blocks), so most inodes don't | |
348 | * grow out to a tree. If need be, detecting boundary extents could | |
349 | * trivially be added in a future version of ocfs2_get_block(). | |
350 | */ | |
d4388340 | 351 | static void ocfs2_readahead(struct readahead_control *rac) |
628a24f5 | 352 | { |
d4388340 MWO |
353 | int ret; |
354 | struct inode *inode = rac->mapping->host; | |
628a24f5 | 355 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
628a24f5 MF |
356 | |
357 | /* | |
358 | * Use the nonblocking flag for the dlm code to avoid page | |
359 | * lock inversion, but don't bother with retrying. | |
360 | */ | |
361 | ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK); | |
362 | if (ret) | |
d4388340 | 363 | return; |
628a24f5 | 364 | |
d4388340 MWO |
365 | if (down_read_trylock(&oi->ip_alloc_sem) == 0) |
366 | goto out_unlock; | |
628a24f5 MF |
367 | |
368 | /* | |
369 | * Don't bother with inline-data. There isn't anything | |
370 | * to read-ahead in that case anyway... | |
371 | */ | |
372 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
d4388340 | 373 | goto out_up; |
628a24f5 MF |
374 | |
375 | /* | |
376 | * Check whether a remote node truncated this file - we just | |
377 | * drop out in that case as it's not worth handling here. | |
378 | */ | |
d4388340 MWO |
379 | if (readahead_pos(rac) >= i_size_read(inode)) |
380 | goto out_up; | |
628a24f5 | 381 | |
d4388340 | 382 | mpage_readahead(rac, ocfs2_get_block); |
628a24f5 | 383 | |
d4388340 | 384 | out_up: |
628a24f5 | 385 | up_read(&oi->ip_alloc_sem); |
d4388340 | 386 | out_unlock: |
628a24f5 | 387 | ocfs2_inode_unlock(inode, 0); |
628a24f5 MF |
388 | } |
389 | ||
ccd979bd MF |
390 | /* Note: Because we don't support holes, our allocation has |
391 | * already happened (allocation writes zeros to the file data) | |
392 | * so we don't have to worry about ordered writes in | |
393 | * ocfs2_writepage. | |
394 | * | |
395 | * ->writepage is called during the process of invalidating the page cache | |
396 | * during blocked lock processing. It can't block on any cluster locks | |
397 | * to during block mapping. It's relying on the fact that the block | |
398 | * mapping can't have disappeared under the dirty pages that it is | |
399 | * being asked to write back. | |
400 | */ | |
401 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) | |
402 | { | |
9558156b TM |
403 | trace_ocfs2_writepage( |
404 | (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno, | |
405 | page->index); | |
ccd979bd | 406 | |
9558156b | 407 | return block_write_full_page(page, ocfs2_get_block, wbc); |
ccd979bd MF |
408 | } |
409 | ||
ccd979bd MF |
410 | /* Taken from ext3. We don't necessarily need the full blown |
411 | * functionality yet, but IMHO it's better to cut and paste the whole | |
412 | * thing so we can avoid introducing our own bugs (and easily pick up | |
413 | * their fixes when they happen) --Mark */ | |
60b11392 MF |
414 | int walk_page_buffers( handle_t *handle, |
415 | struct buffer_head *head, | |
416 | unsigned from, | |
417 | unsigned to, | |
418 | int *partial, | |
419 | int (*fn)( handle_t *handle, | |
420 | struct buffer_head *bh)) | |
ccd979bd MF |
421 | { |
422 | struct buffer_head *bh; | |
423 | unsigned block_start, block_end; | |
424 | unsigned blocksize = head->b_size; | |
425 | int err, ret = 0; | |
426 | struct buffer_head *next; | |
427 | ||
428 | for ( bh = head, block_start = 0; | |
429 | ret == 0 && (bh != head || !block_start); | |
430 | block_start = block_end, bh = next) | |
431 | { | |
432 | next = bh->b_this_page; | |
433 | block_end = block_start + blocksize; | |
434 | if (block_end <= from || block_start >= to) { | |
435 | if (partial && !buffer_uptodate(bh)) | |
436 | *partial = 1; | |
437 | continue; | |
438 | } | |
439 | err = (*fn)(handle, bh); | |
440 | if (!ret) | |
441 | ret = err; | |
442 | } | |
443 | return ret; | |
444 | } | |
445 | ||
ccd979bd MF |
446 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) |
447 | { | |
448 | sector_t status; | |
449 | u64 p_blkno = 0; | |
450 | int err = 0; | |
451 | struct inode *inode = mapping->host; | |
452 | ||
9558156b TM |
453 | trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno, |
454 | (unsigned long long)block); | |
ccd979bd | 455 | |
06a70305 DW |
456 | /* |
457 | * The swap code (ab-)uses ->bmap to get a block mapping and then | |
458 | * bypasseѕ the file system for actual I/O. We really can't allow | |
459 | * that on refcounted inodes, so we have to skip out here. And yes, | |
460 | * 0 is the magic code for a bmap error.. | |
461 | */ | |
462 | if (ocfs2_is_refcount_inode(inode)) | |
463 | return 0; | |
464 | ||
ccd979bd MF |
465 | /* We don't need to lock journal system files, since they aren't |
466 | * accessed concurrently from multiple nodes. | |
467 | */ | |
468 | if (!INODE_JOURNAL(inode)) { | |
e63aecb6 | 469 | err = ocfs2_inode_lock(inode, NULL, 0); |
ccd979bd MF |
470 | if (err) { |
471 | if (err != -ENOENT) | |
472 | mlog_errno(err); | |
473 | goto bail; | |
474 | } | |
475 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | |
476 | } | |
477 | ||
6798d35a MF |
478 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) |
479 | err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, | |
480 | NULL); | |
ccd979bd MF |
481 | |
482 | if (!INODE_JOURNAL(inode)) { | |
483 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
e63aecb6 | 484 | ocfs2_inode_unlock(inode, 0); |
ccd979bd MF |
485 | } |
486 | ||
487 | if (err) { | |
488 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", | |
489 | (unsigned long long)block); | |
490 | mlog_errno(err); | |
491 | goto bail; | |
492 | } | |
493 | ||
ccd979bd MF |
494 | bail: |
495 | status = err ? 0 : p_blkno; | |
496 | ||
ccd979bd MF |
497 | return status; |
498 | } | |
499 | ||
03f981cf JB |
500 | static int ocfs2_releasepage(struct page *page, gfp_t wait) |
501 | { | |
03f981cf JB |
502 | if (!page_has_buffers(page)) |
503 | return 0; | |
41ecc345 | 504 | return try_to_free_buffers(page); |
03f981cf JB |
505 | } |
506 | ||
9517bac6 MF |
507 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, |
508 | u32 cpos, | |
509 | unsigned int *start, | |
510 | unsigned int *end) | |
511 | { | |
09cbfeaf | 512 | unsigned int cluster_start = 0, cluster_end = PAGE_SIZE; |
9517bac6 | 513 | |
09cbfeaf | 514 | if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) { |
9517bac6 MF |
515 | unsigned int cpp; |
516 | ||
09cbfeaf | 517 | cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits); |
9517bac6 MF |
518 | |
519 | cluster_start = cpos % cpp; | |
520 | cluster_start = cluster_start << osb->s_clustersize_bits; | |
521 | ||
522 | cluster_end = cluster_start + osb->s_clustersize; | |
523 | } | |
524 | ||
525 | BUG_ON(cluster_start > PAGE_SIZE); | |
526 | BUG_ON(cluster_end > PAGE_SIZE); | |
527 | ||
528 | if (start) | |
529 | *start = cluster_start; | |
530 | if (end) | |
531 | *end = cluster_end; | |
532 | } | |
533 | ||
534 | /* | |
535 | * 'from' and 'to' are the region in the page to avoid zeroing. | |
536 | * | |
537 | * If pagesize > clustersize, this function will avoid zeroing outside | |
538 | * of the cluster boundary. | |
539 | * | |
540 | * from == to == 0 is code for "zero the entire cluster region" | |
541 | */ | |
542 | static void ocfs2_clear_page_regions(struct page *page, | |
543 | struct ocfs2_super *osb, u32 cpos, | |
544 | unsigned from, unsigned to) | |
545 | { | |
546 | void *kaddr; | |
547 | unsigned int cluster_start, cluster_end; | |
548 | ||
549 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); | |
550 | ||
c4bc8dcb | 551 | kaddr = kmap_atomic(page); |
9517bac6 MF |
552 | |
553 | if (from || to) { | |
554 | if (from > cluster_start) | |
555 | memset(kaddr + cluster_start, 0, from - cluster_start); | |
556 | if (to < cluster_end) | |
557 | memset(kaddr + to, 0, cluster_end - to); | |
558 | } else { | |
559 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); | |
560 | } | |
561 | ||
c4bc8dcb | 562 | kunmap_atomic(kaddr); |
9517bac6 MF |
563 | } |
564 | ||
4e9563fd MF |
565 | /* |
566 | * Nonsparse file systems fully allocate before we get to the write | |
567 | * code. This prevents ocfs2_write() from tagging the write as an | |
568 | * allocating one, which means ocfs2_map_page_blocks() might try to | |
569 | * read-in the blocks at the tail of our file. Avoid reading them by | |
570 | * testing i_size against each block offset. | |
571 | */ | |
572 | static int ocfs2_should_read_blk(struct inode *inode, struct page *page, | |
573 | unsigned int block_start) | |
574 | { | |
575 | u64 offset = page_offset(page) + block_start; | |
576 | ||
577 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
578 | return 1; | |
579 | ||
580 | if (i_size_read(inode) > offset) | |
581 | return 1; | |
582 | ||
583 | return 0; | |
584 | } | |
585 | ||
9517bac6 | 586 | /* |
ebdec241 | 587 | * Some of this taken from __block_write_begin(). We already have our |
9517bac6 MF |
588 | * mapping by now though, and the entire write will be allocating or |
589 | * it won't, so not much need to use BH_New. | |
590 | * | |
591 | * This will also skip zeroing, which is handled externally. | |
592 | */ | |
60b11392 MF |
593 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, |
594 | struct inode *inode, unsigned int from, | |
595 | unsigned int to, int new) | |
9517bac6 MF |
596 | { |
597 | int ret = 0; | |
598 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; | |
599 | unsigned int block_end, block_start; | |
93407472 | 600 | unsigned int bsize = i_blocksize(inode); |
9517bac6 MF |
601 | |
602 | if (!page_has_buffers(page)) | |
603 | create_empty_buffers(page, bsize, 0); | |
604 | ||
605 | head = page_buffers(page); | |
606 | for (bh = head, block_start = 0; bh != head || !block_start; | |
607 | bh = bh->b_this_page, block_start += bsize) { | |
608 | block_end = block_start + bsize; | |
609 | ||
3a307ffc MF |
610 | clear_buffer_new(bh); |
611 | ||
9517bac6 MF |
612 | /* |
613 | * Ignore blocks outside of our i/o range - | |
614 | * they may belong to unallocated clusters. | |
615 | */ | |
60b11392 | 616 | if (block_start >= to || block_end <= from) { |
9517bac6 MF |
617 | if (PageUptodate(page)) |
618 | set_buffer_uptodate(bh); | |
619 | continue; | |
620 | } | |
621 | ||
622 | /* | |
623 | * For an allocating write with cluster size >= page | |
624 | * size, we always write the entire page. | |
625 | */ | |
3a307ffc MF |
626 | if (new) |
627 | set_buffer_new(bh); | |
9517bac6 MF |
628 | |
629 | if (!buffer_mapped(bh)) { | |
630 | map_bh(bh, inode->i_sb, *p_blkno); | |
e64855c6 | 631 | clean_bdev_bh_alias(bh); |
9517bac6 MF |
632 | } |
633 | ||
634 | if (PageUptodate(page)) { | |
635 | if (!buffer_uptodate(bh)) | |
636 | set_buffer_uptodate(bh); | |
637 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
bce99768 | 638 | !buffer_new(bh) && |
4e9563fd | 639 | ocfs2_should_read_blk(inode, page, block_start) && |
bce99768 | 640 | (block_start < from || block_end > to)) { |
dfec8a14 | 641 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
9517bac6 MF |
642 | *wait_bh++=bh; |
643 | } | |
644 | ||
645 | *p_blkno = *p_blkno + 1; | |
646 | } | |
647 | ||
648 | /* | |
649 | * If we issued read requests - let them complete. | |
650 | */ | |
651 | while(wait_bh > wait) { | |
652 | wait_on_buffer(*--wait_bh); | |
653 | if (!buffer_uptodate(*wait_bh)) | |
654 | ret = -EIO; | |
655 | } | |
656 | ||
657 | if (ret == 0 || !new) | |
658 | return ret; | |
659 | ||
660 | /* | |
661 | * If we get -EIO above, zero out any newly allocated blocks | |
662 | * to avoid exposing stale data. | |
663 | */ | |
664 | bh = head; | |
665 | block_start = 0; | |
666 | do { | |
9517bac6 MF |
667 | block_end = block_start + bsize; |
668 | if (block_end <= from) | |
669 | goto next_bh; | |
670 | if (block_start >= to) | |
671 | break; | |
672 | ||
eebd2aa3 | 673 | zero_user(page, block_start, bh->b_size); |
9517bac6 MF |
674 | set_buffer_uptodate(bh); |
675 | mark_buffer_dirty(bh); | |
676 | ||
677 | next_bh: | |
678 | block_start = block_end; | |
679 | bh = bh->b_this_page; | |
680 | } while (bh != head); | |
681 | ||
682 | return ret; | |
683 | } | |
684 | ||
ea1754a0 | 685 | #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) |
3a307ffc MF |
686 | #define OCFS2_MAX_CTXT_PAGES 1 |
687 | #else | |
09cbfeaf | 688 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE) |
3a307ffc MF |
689 | #endif |
690 | ||
09cbfeaf | 691 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE) |
3a307ffc | 692 | |
4506cfb6 RD |
693 | struct ocfs2_unwritten_extent { |
694 | struct list_head ue_node; | |
695 | struct list_head ue_ip_node; | |
696 | u32 ue_cpos; | |
697 | u32 ue_phys; | |
698 | }; | |
699 | ||
6af67d82 | 700 | /* |
3a307ffc | 701 | * Describe the state of a single cluster to be written to. |
6af67d82 | 702 | */ |
3a307ffc MF |
703 | struct ocfs2_write_cluster_desc { |
704 | u32 c_cpos; | |
705 | u32 c_phys; | |
706 | /* | |
707 | * Give this a unique field because c_phys eventually gets | |
708 | * filled. | |
709 | */ | |
710 | unsigned c_new; | |
b46637d5 | 711 | unsigned c_clear_unwritten; |
e7432675 | 712 | unsigned c_needs_zero; |
3a307ffc | 713 | }; |
6af67d82 | 714 | |
3a307ffc MF |
715 | struct ocfs2_write_ctxt { |
716 | /* Logical cluster position / len of write */ | |
717 | u32 w_cpos; | |
718 | u32 w_clen; | |
6af67d82 | 719 | |
e7432675 SM |
720 | /* First cluster allocated in a nonsparse extend */ |
721 | u32 w_first_new_cpos; | |
722 | ||
c1ad1e3c RD |
723 | /* Type of caller. Must be one of buffer, mmap, direct. */ |
724 | ocfs2_write_type_t w_type; | |
725 | ||
3a307ffc | 726 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; |
6af67d82 | 727 | |
3a307ffc MF |
728 | /* |
729 | * This is true if page_size > cluster_size. | |
730 | * | |
731 | * It triggers a set of special cases during write which might | |
732 | * have to deal with allocating writes to partial pages. | |
733 | */ | |
734 | unsigned int w_large_pages; | |
6af67d82 | 735 | |
3a307ffc MF |
736 | /* |
737 | * Pages involved in this write. | |
738 | * | |
739 | * w_target_page is the page being written to by the user. | |
740 | * | |
741 | * w_pages is an array of pages which always contains | |
742 | * w_target_page, and in the case of an allocating write with | |
743 | * page_size < cluster size, it will contain zero'd and mapped | |
744 | * pages adjacent to w_target_page which need to be written | |
745 | * out in so that future reads from that region will get | |
746 | * zero's. | |
747 | */ | |
3a307ffc | 748 | unsigned int w_num_pages; |
83fd9c7f | 749 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; |
3a307ffc | 750 | struct page *w_target_page; |
eeb47d12 | 751 | |
5cffff9e WW |
752 | /* |
753 | * w_target_locked is used for page_mkwrite path indicating no unlocking | |
754 | * against w_target_page in ocfs2_write_end_nolock. | |
755 | */ | |
756 | unsigned int w_target_locked:1; | |
757 | ||
3a307ffc MF |
758 | /* |
759 | * ocfs2_write_end() uses this to know what the real range to | |
760 | * write in the target should be. | |
761 | */ | |
762 | unsigned int w_target_from; | |
763 | unsigned int w_target_to; | |
764 | ||
765 | /* | |
766 | * We could use journal_current_handle() but this is cleaner, | |
767 | * IMHO -Mark | |
768 | */ | |
769 | handle_t *w_handle; | |
770 | ||
771 | struct buffer_head *w_di_bh; | |
b27b7cbc MF |
772 | |
773 | struct ocfs2_cached_dealloc_ctxt w_dealloc; | |
4506cfb6 RD |
774 | |
775 | struct list_head w_unwritten_list; | |
63de8bd9 | 776 | unsigned int w_unwritten_count; |
3a307ffc MF |
777 | }; |
778 | ||
1d410a6e | 779 | void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages) |
3a307ffc MF |
780 | { |
781 | int i; | |
782 | ||
1d410a6e MF |
783 | for(i = 0; i < num_pages; i++) { |
784 | if (pages[i]) { | |
785 | unlock_page(pages[i]); | |
786 | mark_page_accessed(pages[i]); | |
09cbfeaf | 787 | put_page(pages[i]); |
1d410a6e | 788 | } |
6af67d82 | 789 | } |
1d410a6e MF |
790 | } |
791 | ||
136f49b9 | 792 | static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc) |
1d410a6e | 793 | { |
5cffff9e WW |
794 | int i; |
795 | ||
796 | /* | |
797 | * w_target_locked is only set to true in the page_mkwrite() case. | |
798 | * The intent is to allow us to lock the target page from write_begin() | |
799 | * to write_end(). The caller must hold a ref on w_target_page. | |
800 | */ | |
801 | if (wc->w_target_locked) { | |
802 | BUG_ON(!wc->w_target_page); | |
803 | for (i = 0; i < wc->w_num_pages; i++) { | |
804 | if (wc->w_target_page == wc->w_pages[i]) { | |
805 | wc->w_pages[i] = NULL; | |
806 | break; | |
807 | } | |
808 | } | |
809 | mark_page_accessed(wc->w_target_page); | |
09cbfeaf | 810 | put_page(wc->w_target_page); |
5cffff9e | 811 | } |
1d410a6e | 812 | ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages); |
136f49b9 | 813 | } |
6af67d82 | 814 | |
4506cfb6 RD |
815 | static void ocfs2_free_unwritten_list(struct inode *inode, |
816 | struct list_head *head) | |
817 | { | |
818 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
c15471f7 | 819 | struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL; |
4506cfb6 | 820 | |
c15471f7 RD |
821 | list_for_each_entry_safe(ue, tmp, head, ue_node) { |
822 | list_del(&ue->ue_node); | |
4506cfb6 | 823 | spin_lock(&oi->ip_lock); |
c15471f7 | 824 | list_del(&ue->ue_ip_node); |
4506cfb6 | 825 | spin_unlock(&oi->ip_lock); |
c15471f7 | 826 | kfree(ue); |
4506cfb6 RD |
827 | } |
828 | } | |
829 | ||
830 | static void ocfs2_free_write_ctxt(struct inode *inode, | |
831 | struct ocfs2_write_ctxt *wc) | |
136f49b9 | 832 | { |
4506cfb6 | 833 | ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list); |
136f49b9 | 834 | ocfs2_unlock_pages(wc); |
3a307ffc MF |
835 | brelse(wc->w_di_bh); |
836 | kfree(wc); | |
837 | } | |
838 | ||
839 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, | |
840 | struct ocfs2_super *osb, loff_t pos, | |
c1ad1e3c RD |
841 | unsigned len, ocfs2_write_type_t type, |
842 | struct buffer_head *di_bh) | |
3a307ffc | 843 | { |
30b8548f | 844 | u32 cend; |
3a307ffc MF |
845 | struct ocfs2_write_ctxt *wc; |
846 | ||
847 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); | |
848 | if (!wc) | |
849 | return -ENOMEM; | |
6af67d82 | 850 | |
3a307ffc | 851 | wc->w_cpos = pos >> osb->s_clustersize_bits; |
e7432675 | 852 | wc->w_first_new_cpos = UINT_MAX; |
30b8548f | 853 | cend = (pos + len - 1) >> osb->s_clustersize_bits; |
854 | wc->w_clen = cend - wc->w_cpos + 1; | |
607d44aa MF |
855 | get_bh(di_bh); |
856 | wc->w_di_bh = di_bh; | |
c1ad1e3c | 857 | wc->w_type = type; |
6af67d82 | 858 | |
09cbfeaf | 859 | if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) |
3a307ffc MF |
860 | wc->w_large_pages = 1; |
861 | else | |
862 | wc->w_large_pages = 0; | |
863 | ||
b27b7cbc | 864 | ocfs2_init_dealloc_ctxt(&wc->w_dealloc); |
4506cfb6 | 865 | INIT_LIST_HEAD(&wc->w_unwritten_list); |
b27b7cbc | 866 | |
3a307ffc | 867 | *wcp = wc; |
6af67d82 | 868 | |
3a307ffc | 869 | return 0; |
6af67d82 MF |
870 | } |
871 | ||
9517bac6 | 872 | /* |
3a307ffc MF |
873 | * If a page has any new buffers, zero them out here, and mark them uptodate |
874 | * and dirty so they'll be written out (in order to prevent uninitialised | |
875 | * block data from leaking). And clear the new bit. | |
9517bac6 | 876 | */ |
3a307ffc | 877 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) |
9517bac6 | 878 | { |
3a307ffc MF |
879 | unsigned int block_start, block_end; |
880 | struct buffer_head *head, *bh; | |
9517bac6 | 881 | |
3a307ffc MF |
882 | BUG_ON(!PageLocked(page)); |
883 | if (!page_has_buffers(page)) | |
884 | return; | |
9517bac6 | 885 | |
3a307ffc MF |
886 | bh = head = page_buffers(page); |
887 | block_start = 0; | |
888 | do { | |
889 | block_end = block_start + bh->b_size; | |
890 | ||
891 | if (buffer_new(bh)) { | |
892 | if (block_end > from && block_start < to) { | |
893 | if (!PageUptodate(page)) { | |
894 | unsigned start, end; | |
3a307ffc MF |
895 | |
896 | start = max(from, block_start); | |
897 | end = min(to, block_end); | |
898 | ||
eebd2aa3 | 899 | zero_user_segment(page, start, end); |
3a307ffc MF |
900 | set_buffer_uptodate(bh); |
901 | } | |
902 | ||
903 | clear_buffer_new(bh); | |
904 | mark_buffer_dirty(bh); | |
905 | } | |
906 | } | |
9517bac6 | 907 | |
3a307ffc MF |
908 | block_start = block_end; |
909 | bh = bh->b_this_page; | |
910 | } while (bh != head); | |
911 | } | |
912 | ||
913 | /* | |
914 | * Only called when we have a failure during allocating write to write | |
915 | * zero's to the newly allocated region. | |
916 | */ | |
917 | static void ocfs2_write_failure(struct inode *inode, | |
918 | struct ocfs2_write_ctxt *wc, | |
919 | loff_t user_pos, unsigned user_len) | |
920 | { | |
921 | int i; | |
09cbfeaf | 922 | unsigned from = user_pos & (PAGE_SIZE - 1), |
5c26a7b7 | 923 | to = user_pos + user_len; |
3a307ffc MF |
924 | struct page *tmppage; |
925 | ||
65c4db8c RD |
926 | if (wc->w_target_page) |
927 | ocfs2_zero_new_buffers(wc->w_target_page, from, to); | |
9517bac6 | 928 | |
3a307ffc MF |
929 | for(i = 0; i < wc->w_num_pages; i++) { |
930 | tmppage = wc->w_pages[i]; | |
9517bac6 | 931 | |
65c4db8c | 932 | if (tmppage && page_has_buffers(tmppage)) { |
53ef99ca | 933 | if (ocfs2_should_order_data(inode)) |
bbd0f327 JQ |
934 | ocfs2_jbd2_inode_add_write(wc->w_handle, inode, |
935 | user_pos, user_len); | |
961cecbe SM |
936 | |
937 | block_commit_write(tmppage, from, to); | |
938 | } | |
9517bac6 | 939 | } |
9517bac6 MF |
940 | } |
941 | ||
3a307ffc MF |
942 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, |
943 | struct ocfs2_write_ctxt *wc, | |
944 | struct page *page, u32 cpos, | |
945 | loff_t user_pos, unsigned user_len, | |
946 | int new) | |
9517bac6 | 947 | { |
3a307ffc MF |
948 | int ret; |
949 | unsigned int map_from = 0, map_to = 0; | |
9517bac6 | 950 | unsigned int cluster_start, cluster_end; |
3a307ffc | 951 | unsigned int user_data_from = 0, user_data_to = 0; |
9517bac6 | 952 | |
3a307ffc | 953 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, |
9517bac6 MF |
954 | &cluster_start, &cluster_end); |
955 | ||
272b62c1 GR |
956 | /* treat the write as new if the a hole/lseek spanned across |
957 | * the page boundary. | |
958 | */ | |
959 | new = new | ((i_size_read(inode) <= page_offset(page)) && | |
960 | (page_offset(page) <= user_pos)); | |
961 | ||
3a307ffc | 962 | if (page == wc->w_target_page) { |
09cbfeaf | 963 | map_from = user_pos & (PAGE_SIZE - 1); |
3a307ffc MF |
964 | map_to = map_from + user_len; |
965 | ||
966 | if (new) | |
967 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
968 | cluster_start, cluster_end, | |
969 | new); | |
970 | else | |
971 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
972 | map_from, map_to, new); | |
973 | if (ret) { | |
9517bac6 MF |
974 | mlog_errno(ret); |
975 | goto out; | |
976 | } | |
977 | ||
3a307ffc MF |
978 | user_data_from = map_from; |
979 | user_data_to = map_to; | |
9517bac6 | 980 | if (new) { |
3a307ffc MF |
981 | map_from = cluster_start; |
982 | map_to = cluster_end; | |
9517bac6 MF |
983 | } |
984 | } else { | |
985 | /* | |
986 | * If we haven't allocated the new page yet, we | |
987 | * shouldn't be writing it out without copying user | |
988 | * data. This is likely a math error from the caller. | |
989 | */ | |
990 | BUG_ON(!new); | |
991 | ||
3a307ffc MF |
992 | map_from = cluster_start; |
993 | map_to = cluster_end; | |
9517bac6 MF |
994 | |
995 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | |
3a307ffc | 996 | cluster_start, cluster_end, new); |
9517bac6 MF |
997 | if (ret) { |
998 | mlog_errno(ret); | |
999 | goto out; | |
1000 | } | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Parts of newly allocated pages need to be zero'd. | |
1005 | * | |
1006 | * Above, we have also rewritten 'to' and 'from' - as far as | |
1007 | * the rest of the function is concerned, the entire cluster | |
1008 | * range inside of a page needs to be written. | |
1009 | * | |
1010 | * We can skip this if the page is up to date - it's already | |
1011 | * been zero'd from being read in as a hole. | |
1012 | */ | |
1013 | if (new && !PageUptodate(page)) | |
1014 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), | |
3a307ffc | 1015 | cpos, user_data_from, user_data_to); |
9517bac6 MF |
1016 | |
1017 | flush_dcache_page(page); | |
1018 | ||
9517bac6 | 1019 | out: |
3a307ffc | 1020 | return ret; |
9517bac6 MF |
1021 | } |
1022 | ||
1023 | /* | |
3a307ffc | 1024 | * This function will only grab one clusters worth of pages. |
9517bac6 | 1025 | */ |
3a307ffc MF |
1026 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, |
1027 | struct ocfs2_write_ctxt *wc, | |
693c241a JB |
1028 | u32 cpos, loff_t user_pos, |
1029 | unsigned user_len, int new, | |
7307de80 | 1030 | struct page *mmap_page) |
9517bac6 | 1031 | { |
3a307ffc | 1032 | int ret = 0, i; |
693c241a | 1033 | unsigned long start, target_index, end_index, index; |
9517bac6 | 1034 | struct inode *inode = mapping->host; |
693c241a | 1035 | loff_t last_byte; |
9517bac6 | 1036 | |
09cbfeaf | 1037 | target_index = user_pos >> PAGE_SHIFT; |
9517bac6 MF |
1038 | |
1039 | /* | |
1040 | * Figure out how many pages we'll be manipulating here. For | |
60b11392 | 1041 | * non allocating write, we just change the one |
693c241a JB |
1042 | * page. Otherwise, we'll need a whole clusters worth. If we're |
1043 | * writing past i_size, we only need enough pages to cover the | |
1044 | * last page of the write. | |
9517bac6 | 1045 | */ |
9517bac6 | 1046 | if (new) { |
3a307ffc MF |
1047 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); |
1048 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); | |
693c241a JB |
1049 | /* |
1050 | * We need the index *past* the last page we could possibly | |
1051 | * touch. This is the page past the end of the write or | |
1052 | * i_size, whichever is greater. | |
1053 | */ | |
1054 | last_byte = max(user_pos + user_len, i_size_read(inode)); | |
1055 | BUG_ON(last_byte < 1); | |
09cbfeaf | 1056 | end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1; |
693c241a JB |
1057 | if ((start + wc->w_num_pages) > end_index) |
1058 | wc->w_num_pages = end_index - start; | |
9517bac6 | 1059 | } else { |
3a307ffc MF |
1060 | wc->w_num_pages = 1; |
1061 | start = target_index; | |
9517bac6 | 1062 | } |
09cbfeaf | 1063 | end_index = (user_pos + user_len - 1) >> PAGE_SHIFT; |
9517bac6 | 1064 | |
3a307ffc | 1065 | for(i = 0; i < wc->w_num_pages; i++) { |
9517bac6 MF |
1066 | index = start + i; |
1067 | ||
65c4db8c RD |
1068 | if (index >= target_index && index <= end_index && |
1069 | wc->w_type == OCFS2_WRITE_MMAP) { | |
7307de80 MF |
1070 | /* |
1071 | * ocfs2_pagemkwrite() is a little different | |
1072 | * and wants us to directly use the page | |
1073 | * passed in. | |
1074 | */ | |
1075 | lock_page(mmap_page); | |
1076 | ||
5cffff9e | 1077 | /* Exit and let the caller retry */ |
7307de80 | 1078 | if (mmap_page->mapping != mapping) { |
5cffff9e | 1079 | WARN_ON(mmap_page->mapping); |
7307de80 | 1080 | unlock_page(mmap_page); |
5cffff9e | 1081 | ret = -EAGAIN; |
7307de80 MF |
1082 | goto out; |
1083 | } | |
1084 | ||
09cbfeaf | 1085 | get_page(mmap_page); |
7307de80 | 1086 | wc->w_pages[i] = mmap_page; |
5cffff9e | 1087 | wc->w_target_locked = true; |
65c4db8c RD |
1088 | } else if (index >= target_index && index <= end_index && |
1089 | wc->w_type == OCFS2_WRITE_DIRECT) { | |
1090 | /* Direct write has no mapping page. */ | |
1091 | wc->w_pages[i] = NULL; | |
1092 | continue; | |
7307de80 MF |
1093 | } else { |
1094 | wc->w_pages[i] = find_or_create_page(mapping, index, | |
1095 | GFP_NOFS); | |
1096 | if (!wc->w_pages[i]) { | |
1097 | ret = -ENOMEM; | |
1098 | mlog_errno(ret); | |
1099 | goto out; | |
1100 | } | |
9517bac6 | 1101 | } |
1269529b | 1102 | wait_for_stable_page(wc->w_pages[i]); |
3a307ffc MF |
1103 | |
1104 | if (index == target_index) | |
1105 | wc->w_target_page = wc->w_pages[i]; | |
9517bac6 | 1106 | } |
3a307ffc | 1107 | out: |
5cffff9e WW |
1108 | if (ret) |
1109 | wc->w_target_locked = false; | |
3a307ffc MF |
1110 | return ret; |
1111 | } | |
1112 | ||
1113 | /* | |
1114 | * Prepare a single cluster for write one cluster into the file. | |
1115 | */ | |
1116 | static int ocfs2_write_cluster(struct address_space *mapping, | |
2de6a3c7 | 1117 | u32 *phys, unsigned int new, |
b46637d5 | 1118 | unsigned int clear_unwritten, |
e7432675 | 1119 | unsigned int should_zero, |
b27b7cbc | 1120 | struct ocfs2_alloc_context *data_ac, |
3a307ffc MF |
1121 | struct ocfs2_alloc_context *meta_ac, |
1122 | struct ocfs2_write_ctxt *wc, u32 cpos, | |
1123 | loff_t user_pos, unsigned user_len) | |
1124 | { | |
b46637d5 | 1125 | int ret, i; |
2de6a3c7 | 1126 | u64 p_blkno; |
3a307ffc | 1127 | struct inode *inode = mapping->host; |
f99b9b7c | 1128 | struct ocfs2_extent_tree et; |
2de6a3c7 | 1129 | int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); |
3a307ffc | 1130 | |
9517bac6 | 1131 | if (new) { |
3a307ffc MF |
1132 | u32 tmp_pos; |
1133 | ||
9517bac6 MF |
1134 | /* |
1135 | * This is safe to call with the page locks - it won't take | |
1136 | * any additional semaphores or cluster locks. | |
1137 | */ | |
3a307ffc | 1138 | tmp_pos = cpos; |
0eb8d47e | 1139 | ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode, |
b46637d5 RD |
1140 | &tmp_pos, 1, !clear_unwritten, |
1141 | wc->w_di_bh, wc->w_handle, | |
1142 | data_ac, meta_ac, NULL); | |
9517bac6 MF |
1143 | /* |
1144 | * This shouldn't happen because we must have already | |
1145 | * calculated the correct meta data allocation required. The | |
1146 | * internal tree allocation code should know how to increase | |
1147 | * transaction credits itself. | |
1148 | * | |
1149 | * If need be, we could handle -EAGAIN for a | |
1150 | * RESTART_TRANS here. | |
1151 | */ | |
1152 | mlog_bug_on_msg(ret == -EAGAIN, | |
1153 | "Inode %llu: EAGAIN return during allocation.\n", | |
1154 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1155 | if (ret < 0) { | |
1156 | mlog_errno(ret); | |
1157 | goto out; | |
1158 | } | |
b46637d5 | 1159 | } else if (clear_unwritten) { |
5e404e9e JB |
1160 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
1161 | wc->w_di_bh); | |
f99b9b7c | 1162 | ret = ocfs2_mark_extent_written(inode, &et, |
2de6a3c7 | 1163 | wc->w_handle, cpos, 1, *phys, |
f99b9b7c | 1164 | meta_ac, &wc->w_dealloc); |
b27b7cbc MF |
1165 | if (ret < 0) { |
1166 | mlog_errno(ret); | |
1167 | goto out; | |
1168 | } | |
1169 | } | |
3a307ffc | 1170 | |
3a307ffc MF |
1171 | /* |
1172 | * The only reason this should fail is due to an inability to | |
1173 | * find the extent added. | |
1174 | */ | |
2de6a3c7 | 1175 | ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL); |
9517bac6 | 1176 | if (ret < 0) { |
61fb9ea4 | 1177 | mlog(ML_ERROR, "Get physical blkno failed for inode %llu, " |
2de6a3c7 RD |
1178 | "at logical cluster %u", |
1179 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
9517bac6 MF |
1180 | goto out; |
1181 | } | |
1182 | ||
2de6a3c7 RD |
1183 | BUG_ON(*phys == 0); |
1184 | ||
1185 | p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys); | |
1186 | if (!should_zero) | |
1187 | p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1); | |
9517bac6 | 1188 | |
3a307ffc MF |
1189 | for(i = 0; i < wc->w_num_pages; i++) { |
1190 | int tmpret; | |
9517bac6 | 1191 | |
65c4db8c RD |
1192 | /* This is the direct io target page. */ |
1193 | if (wc->w_pages[i] == NULL) { | |
1194 | p_blkno++; | |
1195 | continue; | |
1196 | } | |
1197 | ||
3a307ffc MF |
1198 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, |
1199 | wc->w_pages[i], cpos, | |
b27b7cbc MF |
1200 | user_pos, user_len, |
1201 | should_zero); | |
3a307ffc MF |
1202 | if (tmpret) { |
1203 | mlog_errno(tmpret); | |
1204 | if (ret == 0) | |
cbfa9639 | 1205 | ret = tmpret; |
3a307ffc | 1206 | } |
9517bac6 MF |
1207 | } |
1208 | ||
3a307ffc MF |
1209 | /* |
1210 | * We only have cleanup to do in case of allocating write. | |
1211 | */ | |
1212 | if (ret && new) | |
1213 | ocfs2_write_failure(inode, wc, user_pos, user_len); | |
1214 | ||
9517bac6 | 1215 | out: |
9517bac6 | 1216 | |
3a307ffc | 1217 | return ret; |
9517bac6 MF |
1218 | } |
1219 | ||
0d172baa MF |
1220 | static int ocfs2_write_cluster_by_desc(struct address_space *mapping, |
1221 | struct ocfs2_alloc_context *data_ac, | |
1222 | struct ocfs2_alloc_context *meta_ac, | |
1223 | struct ocfs2_write_ctxt *wc, | |
1224 | loff_t pos, unsigned len) | |
1225 | { | |
1226 | int ret, i; | |
db56246c MF |
1227 | loff_t cluster_off; |
1228 | unsigned int local_len = len; | |
0d172baa | 1229 | struct ocfs2_write_cluster_desc *desc; |
db56246c | 1230 | struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb); |
0d172baa MF |
1231 | |
1232 | for (i = 0; i < wc->w_clen; i++) { | |
1233 | desc = &wc->w_desc[i]; | |
1234 | ||
db56246c MF |
1235 | /* |
1236 | * We have to make sure that the total write passed in | |
1237 | * doesn't extend past a single cluster. | |
1238 | */ | |
1239 | local_len = len; | |
1240 | cluster_off = pos & (osb->s_clustersize - 1); | |
1241 | if ((cluster_off + local_len) > osb->s_clustersize) | |
1242 | local_len = osb->s_clustersize - cluster_off; | |
1243 | ||
2de6a3c7 | 1244 | ret = ocfs2_write_cluster(mapping, &desc->c_phys, |
b46637d5 RD |
1245 | desc->c_new, |
1246 | desc->c_clear_unwritten, | |
e7432675 SM |
1247 | desc->c_needs_zero, |
1248 | data_ac, meta_ac, | |
db56246c | 1249 | wc, desc->c_cpos, pos, local_len); |
0d172baa MF |
1250 | if (ret) { |
1251 | mlog_errno(ret); | |
1252 | goto out; | |
1253 | } | |
db56246c MF |
1254 | |
1255 | len -= local_len; | |
1256 | pos += local_len; | |
0d172baa MF |
1257 | } |
1258 | ||
1259 | ret = 0; | |
1260 | out: | |
1261 | return ret; | |
1262 | } | |
1263 | ||
3a307ffc MF |
1264 | /* |
1265 | * ocfs2_write_end() wants to know which parts of the target page it | |
1266 | * should complete the write on. It's easiest to compute them ahead of | |
1267 | * time when a more complete view of the write is available. | |
1268 | */ | |
1269 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, | |
1270 | struct ocfs2_write_ctxt *wc, | |
1271 | loff_t pos, unsigned len, int alloc) | |
9517bac6 | 1272 | { |
3a307ffc | 1273 | struct ocfs2_write_cluster_desc *desc; |
9517bac6 | 1274 | |
09cbfeaf | 1275 | wc->w_target_from = pos & (PAGE_SIZE - 1); |
3a307ffc MF |
1276 | wc->w_target_to = wc->w_target_from + len; |
1277 | ||
1278 | if (alloc == 0) | |
1279 | return; | |
1280 | ||
1281 | /* | |
1282 | * Allocating write - we may have different boundaries based | |
1283 | * on page size and cluster size. | |
1284 | * | |
1285 | * NOTE: We can no longer compute one value from the other as | |
1286 | * the actual write length and user provided length may be | |
1287 | * different. | |
1288 | */ | |
9517bac6 | 1289 | |
3a307ffc MF |
1290 | if (wc->w_large_pages) { |
1291 | /* | |
1292 | * We only care about the 1st and last cluster within | |
b27b7cbc | 1293 | * our range and whether they should be zero'd or not. Either |
3a307ffc MF |
1294 | * value may be extended out to the start/end of a |
1295 | * newly allocated cluster. | |
1296 | */ | |
1297 | desc = &wc->w_desc[0]; | |
e7432675 | 1298 | if (desc->c_needs_zero) |
3a307ffc MF |
1299 | ocfs2_figure_cluster_boundaries(osb, |
1300 | desc->c_cpos, | |
1301 | &wc->w_target_from, | |
1302 | NULL); | |
1303 | ||
1304 | desc = &wc->w_desc[wc->w_clen - 1]; | |
e7432675 | 1305 | if (desc->c_needs_zero) |
3a307ffc MF |
1306 | ocfs2_figure_cluster_boundaries(osb, |
1307 | desc->c_cpos, | |
1308 | NULL, | |
1309 | &wc->w_target_to); | |
1310 | } else { | |
1311 | wc->w_target_from = 0; | |
09cbfeaf | 1312 | wc->w_target_to = PAGE_SIZE; |
3a307ffc | 1313 | } |
9517bac6 MF |
1314 | } |
1315 | ||
4506cfb6 RD |
1316 | /* |
1317 | * Check if this extent is marked UNWRITTEN by direct io. If so, we need not to | |
1318 | * do the zero work. And should not to clear UNWRITTEN since it will be cleared | |
1319 | * by the direct io procedure. | |
1320 | * If this is a new extent that allocated by direct io, we should mark it in | |
1321 | * the ip_unwritten_list. | |
1322 | */ | |
1323 | static int ocfs2_unwritten_check(struct inode *inode, | |
1324 | struct ocfs2_write_ctxt *wc, | |
1325 | struct ocfs2_write_cluster_desc *desc) | |
1326 | { | |
1327 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
c15471f7 | 1328 | struct ocfs2_unwritten_extent *ue = NULL, *new = NULL; |
4506cfb6 RD |
1329 | int ret = 0; |
1330 | ||
1331 | if (!desc->c_needs_zero) | |
1332 | return 0; | |
1333 | ||
1334 | retry: | |
1335 | spin_lock(&oi->ip_lock); | |
1336 | /* Needs not to zero no metter buffer or direct. The one who is zero | |
1337 | * the cluster is doing zero. And he will clear unwritten after all | |
1338 | * cluster io finished. */ | |
c15471f7 RD |
1339 | list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) { |
1340 | if (desc->c_cpos == ue->ue_cpos) { | |
4506cfb6 RD |
1341 | BUG_ON(desc->c_new); |
1342 | desc->c_needs_zero = 0; | |
1343 | desc->c_clear_unwritten = 0; | |
1344 | goto unlock; | |
1345 | } | |
1346 | } | |
1347 | ||
1348 | if (wc->w_type != OCFS2_WRITE_DIRECT) | |
1349 | goto unlock; | |
1350 | ||
1351 | if (new == NULL) { | |
1352 | spin_unlock(&oi->ip_lock); | |
1353 | new = kmalloc(sizeof(struct ocfs2_unwritten_extent), | |
1354 | GFP_NOFS); | |
1355 | if (new == NULL) { | |
1356 | ret = -ENOMEM; | |
1357 | goto out; | |
1358 | } | |
1359 | goto retry; | |
1360 | } | |
1361 | /* This direct write will doing zero. */ | |
1362 | new->ue_cpos = desc->c_cpos; | |
1363 | new->ue_phys = desc->c_phys; | |
1364 | desc->c_clear_unwritten = 0; | |
1365 | list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list); | |
1366 | list_add_tail(&new->ue_node, &wc->w_unwritten_list); | |
63de8bd9 | 1367 | wc->w_unwritten_count++; |
4506cfb6 RD |
1368 | new = NULL; |
1369 | unlock: | |
1370 | spin_unlock(&oi->ip_lock); | |
1371 | out: | |
0ae1c2db | 1372 | kfree(new); |
4506cfb6 RD |
1373 | return ret; |
1374 | } | |
1375 | ||
0d172baa MF |
1376 | /* |
1377 | * Populate each single-cluster write descriptor in the write context | |
1378 | * with information about the i/o to be done. | |
b27b7cbc MF |
1379 | * |
1380 | * Returns the number of clusters that will have to be allocated, as | |
1381 | * well as a worst case estimate of the number of extent records that | |
1382 | * would have to be created during a write to an unwritten region. | |
0d172baa MF |
1383 | */ |
1384 | static int ocfs2_populate_write_desc(struct inode *inode, | |
1385 | struct ocfs2_write_ctxt *wc, | |
b27b7cbc MF |
1386 | unsigned int *clusters_to_alloc, |
1387 | unsigned int *extents_to_split) | |
9517bac6 | 1388 | { |
0d172baa | 1389 | int ret; |
3a307ffc | 1390 | struct ocfs2_write_cluster_desc *desc; |
0d172baa | 1391 | unsigned int num_clusters = 0; |
b27b7cbc | 1392 | unsigned int ext_flags = 0; |
0d172baa MF |
1393 | u32 phys = 0; |
1394 | int i; | |
9517bac6 | 1395 | |
b27b7cbc MF |
1396 | *clusters_to_alloc = 0; |
1397 | *extents_to_split = 0; | |
1398 | ||
3a307ffc MF |
1399 | for (i = 0; i < wc->w_clen; i++) { |
1400 | desc = &wc->w_desc[i]; | |
1401 | desc->c_cpos = wc->w_cpos + i; | |
1402 | ||
1403 | if (num_clusters == 0) { | |
b27b7cbc MF |
1404 | /* |
1405 | * Need to look up the next extent record. | |
1406 | */ | |
3a307ffc | 1407 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, |
b27b7cbc | 1408 | &num_clusters, &ext_flags); |
3a307ffc MF |
1409 | if (ret) { |
1410 | mlog_errno(ret); | |
607d44aa | 1411 | goto out; |
3a307ffc | 1412 | } |
b27b7cbc | 1413 | |
293b2f70 TM |
1414 | /* We should already CoW the refcountd extent. */ |
1415 | BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED); | |
1416 | ||
b27b7cbc MF |
1417 | /* |
1418 | * Assume worst case - that we're writing in | |
1419 | * the middle of the extent. | |
1420 | * | |
1421 | * We can assume that the write proceeds from | |
1422 | * left to right, in which case the extent | |
1423 | * insert code is smart enough to coalesce the | |
1424 | * next splits into the previous records created. | |
1425 | */ | |
1426 | if (ext_flags & OCFS2_EXT_UNWRITTEN) | |
1427 | *extents_to_split = *extents_to_split + 2; | |
3a307ffc MF |
1428 | } else if (phys) { |
1429 | /* | |
1430 | * Only increment phys if it doesn't describe | |
1431 | * a hole. | |
1432 | */ | |
1433 | phys++; | |
1434 | } | |
1435 | ||
e7432675 SM |
1436 | /* |
1437 | * If w_first_new_cpos is < UINT_MAX, we have a non-sparse | |
1438 | * file that got extended. w_first_new_cpos tells us | |
1439 | * where the newly allocated clusters are so we can | |
1440 | * zero them. | |
1441 | */ | |
1442 | if (desc->c_cpos >= wc->w_first_new_cpos) { | |
1443 | BUG_ON(phys == 0); | |
1444 | desc->c_needs_zero = 1; | |
1445 | } | |
1446 | ||
3a307ffc MF |
1447 | desc->c_phys = phys; |
1448 | if (phys == 0) { | |
1449 | desc->c_new = 1; | |
e7432675 | 1450 | desc->c_needs_zero = 1; |
b46637d5 | 1451 | desc->c_clear_unwritten = 1; |
0d172baa | 1452 | *clusters_to_alloc = *clusters_to_alloc + 1; |
3a307ffc | 1453 | } |
e7432675 SM |
1454 | |
1455 | if (ext_flags & OCFS2_EXT_UNWRITTEN) { | |
b46637d5 | 1456 | desc->c_clear_unwritten = 1; |
e7432675 SM |
1457 | desc->c_needs_zero = 1; |
1458 | } | |
3a307ffc | 1459 | |
4506cfb6 RD |
1460 | ret = ocfs2_unwritten_check(inode, wc, desc); |
1461 | if (ret) { | |
1462 | mlog_errno(ret); | |
1463 | goto out; | |
1464 | } | |
1465 | ||
3a307ffc | 1466 | num_clusters--; |
9517bac6 MF |
1467 | } |
1468 | ||
0d172baa MF |
1469 | ret = 0; |
1470 | out: | |
1471 | return ret; | |
1472 | } | |
1473 | ||
1afc32b9 MF |
1474 | static int ocfs2_write_begin_inline(struct address_space *mapping, |
1475 | struct inode *inode, | |
1476 | struct ocfs2_write_ctxt *wc) | |
1477 | { | |
1478 | int ret; | |
1479 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1480 | struct page *page; | |
1481 | handle_t *handle; | |
1482 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1483 | ||
f775da2f JB |
1484 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
1485 | if (IS_ERR(handle)) { | |
1486 | ret = PTR_ERR(handle); | |
1487 | mlog_errno(ret); | |
1488 | goto out; | |
1489 | } | |
1490 | ||
1afc32b9 MF |
1491 | page = find_or_create_page(mapping, 0, GFP_NOFS); |
1492 | if (!page) { | |
f775da2f | 1493 | ocfs2_commit_trans(osb, handle); |
1afc32b9 MF |
1494 | ret = -ENOMEM; |
1495 | mlog_errno(ret); | |
1496 | goto out; | |
1497 | } | |
1498 | /* | |
1499 | * If we don't set w_num_pages then this page won't get unlocked | |
1500 | * and freed on cleanup of the write context. | |
1501 | */ | |
1502 | wc->w_pages[0] = wc->w_target_page = page; | |
1503 | wc->w_num_pages = 1; | |
1504 | ||
0cf2f763 | 1505 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 1506 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
1507 | if (ret) { |
1508 | ocfs2_commit_trans(osb, handle); | |
1509 | ||
1510 | mlog_errno(ret); | |
1511 | goto out; | |
1512 | } | |
1513 | ||
1514 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
1515 | ocfs2_set_inode_data_inline(inode, di); | |
1516 | ||
1517 | if (!PageUptodate(page)) { | |
1518 | ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh); | |
1519 | if (ret) { | |
1520 | ocfs2_commit_trans(osb, handle); | |
1521 | ||
1522 | goto out; | |
1523 | } | |
1524 | } | |
1525 | ||
1526 | wc->w_handle = handle; | |
1527 | out: | |
1528 | return ret; | |
1529 | } | |
1530 | ||
1531 | int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size) | |
1532 | { | |
1533 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1534 | ||
0d8a4e0c | 1535 | if (new_size <= le16_to_cpu(di->id2.i_data.id_count)) |
1afc32b9 MF |
1536 | return 1; |
1537 | return 0; | |
1538 | } | |
1539 | ||
1540 | static int ocfs2_try_to_write_inline_data(struct address_space *mapping, | |
1541 | struct inode *inode, loff_t pos, | |
1542 | unsigned len, struct page *mmap_page, | |
1543 | struct ocfs2_write_ctxt *wc) | |
1544 | { | |
1545 | int ret, written = 0; | |
1546 | loff_t end = pos + len; | |
1547 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
d9ae49d6 | 1548 | struct ocfs2_dinode *di = NULL; |
1afc32b9 | 1549 | |
9558156b TM |
1550 | trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno, |
1551 | len, (unsigned long long)pos, | |
1552 | oi->ip_dyn_features); | |
1afc32b9 MF |
1553 | |
1554 | /* | |
1555 | * Handle inodes which already have inline data 1st. | |
1556 | */ | |
1557 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1558 | if (mmap_page == NULL && | |
1559 | ocfs2_size_fits_inline_data(wc->w_di_bh, end)) | |
1560 | goto do_inline_write; | |
1561 | ||
1562 | /* | |
1563 | * The write won't fit - we have to give this inode an | |
1564 | * inline extent list now. | |
1565 | */ | |
1566 | ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh); | |
1567 | if (ret) | |
1568 | mlog_errno(ret); | |
1569 | goto out; | |
1570 | } | |
1571 | ||
1572 | /* | |
1573 | * Check whether the inode can accept inline data. | |
1574 | */ | |
1575 | if (oi->ip_clusters != 0 || i_size_read(inode) != 0) | |
1576 | return 0; | |
1577 | ||
1578 | /* | |
1579 | * Check whether the write can fit. | |
1580 | */ | |
d9ae49d6 TY |
1581 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; |
1582 | if (mmap_page || | |
1583 | end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) | |
1afc32b9 MF |
1584 | return 0; |
1585 | ||
1586 | do_inline_write: | |
1587 | ret = ocfs2_write_begin_inline(mapping, inode, wc); | |
1588 | if (ret) { | |
1589 | mlog_errno(ret); | |
1590 | goto out; | |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * This signals to the caller that the data can be written | |
1595 | * inline. | |
1596 | */ | |
1597 | written = 1; | |
1598 | out: | |
1599 | return written ? written : ret; | |
1600 | } | |
1601 | ||
65ed39d6 MF |
1602 | /* |
1603 | * This function only does anything for file systems which can't | |
1604 | * handle sparse files. | |
1605 | * | |
1606 | * What we want to do here is fill in any hole between the current end | |
1607 | * of allocation and the end of our write. That way the rest of the | |
1608 | * write path can treat it as an non-allocating write, which has no | |
1609 | * special case code for sparse/nonsparse files. | |
1610 | */ | |
5693486b JB |
1611 | static int ocfs2_expand_nonsparse_inode(struct inode *inode, |
1612 | struct buffer_head *di_bh, | |
1613 | loff_t pos, unsigned len, | |
65ed39d6 MF |
1614 | struct ocfs2_write_ctxt *wc) |
1615 | { | |
1616 | int ret; | |
65ed39d6 MF |
1617 | loff_t newsize = pos + len; |
1618 | ||
5693486b | 1619 | BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); |
65ed39d6 MF |
1620 | |
1621 | if (newsize <= i_size_read(inode)) | |
1622 | return 0; | |
1623 | ||
5693486b | 1624 | ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos); |
65ed39d6 MF |
1625 | if (ret) |
1626 | mlog_errno(ret); | |
1627 | ||
46e62556 RD |
1628 | /* There is no wc if this is call from direct. */ |
1629 | if (wc) | |
1630 | wc->w_first_new_cpos = | |
1631 | ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)); | |
e7432675 | 1632 | |
65ed39d6 MF |
1633 | return ret; |
1634 | } | |
1635 | ||
5693486b JB |
1636 | static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh, |
1637 | loff_t pos) | |
1638 | { | |
1639 | int ret = 0; | |
1640 | ||
1641 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); | |
1642 | if (pos > i_size_read(inode)) | |
1643 | ret = ocfs2_zero_extend(inode, di_bh, pos); | |
1644 | ||
1645 | return ret; | |
1646 | } | |
1647 | ||
c1ad1e3c RD |
1648 | int ocfs2_write_begin_nolock(struct address_space *mapping, |
1649 | loff_t pos, unsigned len, ocfs2_write_type_t type, | |
0d172baa MF |
1650 | struct page **pagep, void **fsdata, |
1651 | struct buffer_head *di_bh, struct page *mmap_page) | |
1652 | { | |
e7432675 | 1653 | int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS; |
50308d81 | 1654 | unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0; |
0d172baa MF |
1655 | struct ocfs2_write_ctxt *wc; |
1656 | struct inode *inode = mapping->host; | |
1657 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1658 | struct ocfs2_dinode *di; | |
1659 | struct ocfs2_alloc_context *data_ac = NULL; | |
1660 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1661 | handle_t *handle; | |
f99b9b7c | 1662 | struct ocfs2_extent_tree et; |
50308d81 | 1663 | int try_free = 1, ret1; |
0d172baa | 1664 | |
50308d81 | 1665 | try_again: |
c1ad1e3c | 1666 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh); |
0d172baa MF |
1667 | if (ret) { |
1668 | mlog_errno(ret); | |
1669 | return ret; | |
1670 | } | |
1671 | ||
1afc32b9 MF |
1672 | if (ocfs2_supports_inline_data(osb)) { |
1673 | ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len, | |
1674 | mmap_page, wc); | |
1675 | if (ret == 1) { | |
1676 | ret = 0; | |
1677 | goto success; | |
1678 | } | |
1679 | if (ret < 0) { | |
1680 | mlog_errno(ret); | |
1681 | goto out; | |
1682 | } | |
1683 | } | |
1684 | ||
46e62556 RD |
1685 | /* Direct io change i_size late, should not zero tail here. */ |
1686 | if (type != OCFS2_WRITE_DIRECT) { | |
1687 | if (ocfs2_sparse_alloc(osb)) | |
1688 | ret = ocfs2_zero_tail(inode, di_bh, pos); | |
1689 | else | |
1690 | ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, | |
1691 | len, wc); | |
1692 | if (ret) { | |
1693 | mlog_errno(ret); | |
1694 | goto out; | |
1695 | } | |
65ed39d6 MF |
1696 | } |
1697 | ||
293b2f70 TM |
1698 | ret = ocfs2_check_range_for_refcount(inode, pos, len); |
1699 | if (ret < 0) { | |
1700 | mlog_errno(ret); | |
1701 | goto out; | |
1702 | } else if (ret == 1) { | |
50308d81 | 1703 | clusters_need = wc->w_clen; |
c7dd3392 | 1704 | ret = ocfs2_refcount_cow(inode, di_bh, |
37f8a2bf | 1705 | wc->w_cpos, wc->w_clen, UINT_MAX); |
293b2f70 TM |
1706 | if (ret) { |
1707 | mlog_errno(ret); | |
1708 | goto out; | |
1709 | } | |
1710 | } | |
1711 | ||
b27b7cbc MF |
1712 | ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc, |
1713 | &extents_to_split); | |
0d172baa MF |
1714 | if (ret) { |
1715 | mlog_errno(ret); | |
1716 | goto out; | |
1717 | } | |
50308d81 | 1718 | clusters_need += clusters_to_alloc; |
0d172baa MF |
1719 | |
1720 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1721 | ||
9558156b TM |
1722 | trace_ocfs2_write_begin_nolock( |
1723 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1724 | (long long)i_size_read(inode), | |
1725 | le32_to_cpu(di->i_clusters), | |
c1ad1e3c | 1726 | pos, len, type, mmap_page, |
9558156b TM |
1727 | clusters_to_alloc, extents_to_split); |
1728 | ||
3a307ffc MF |
1729 | /* |
1730 | * We set w_target_from, w_target_to here so that | |
1731 | * ocfs2_write_end() knows which range in the target page to | |
1732 | * write out. An allocation requires that we write the entire | |
1733 | * cluster range. | |
1734 | */ | |
b27b7cbc | 1735 | if (clusters_to_alloc || extents_to_split) { |
3a307ffc MF |
1736 | /* |
1737 | * XXX: We are stretching the limits of | |
b27b7cbc | 1738 | * ocfs2_lock_allocators(). It greatly over-estimates |
3a307ffc MF |
1739 | * the work to be done. |
1740 | */ | |
5e404e9e JB |
1741 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), |
1742 | wc->w_di_bh); | |
f99b9b7c | 1743 | ret = ocfs2_lock_allocators(inode, &et, |
231b87d1 | 1744 | clusters_to_alloc, extents_to_split, |
f99b9b7c | 1745 | &data_ac, &meta_ac); |
9517bac6 MF |
1746 | if (ret) { |
1747 | mlog_errno(ret); | |
607d44aa | 1748 | goto out; |
9517bac6 MF |
1749 | } |
1750 | ||
4fe370af MF |
1751 | if (data_ac) |
1752 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; | |
1753 | ||
811f933d | 1754 | credits = ocfs2_calc_extend_credits(inode->i_sb, |
06f9da6e | 1755 | &di->id2.i_list); |
46e62556 RD |
1756 | } else if (type == OCFS2_WRITE_DIRECT) |
1757 | /* direct write needs not to start trans if no extents alloc. */ | |
1758 | goto success; | |
9517bac6 | 1759 | |
e7432675 SM |
1760 | /* |
1761 | * We have to zero sparse allocated clusters, unwritten extent clusters, | |
1762 | * and non-sparse clusters we just extended. For non-sparse writes, | |
1763 | * we know zeros will only be needed in the first and/or last cluster. | |
1764 | */ | |
4506cfb6 RD |
1765 | if (wc->w_clen && (wc->w_desc[0].c_needs_zero || |
1766 | wc->w_desc[wc->w_clen - 1].c_needs_zero)) | |
e7432675 SM |
1767 | cluster_of_pages = 1; |
1768 | else | |
1769 | cluster_of_pages = 0; | |
1770 | ||
1771 | ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages); | |
3a307ffc | 1772 | |
9517bac6 MF |
1773 | handle = ocfs2_start_trans(osb, credits); |
1774 | if (IS_ERR(handle)) { | |
1775 | ret = PTR_ERR(handle); | |
1776 | mlog_errno(ret); | |
607d44aa | 1777 | goto out; |
9517bac6 MF |
1778 | } |
1779 | ||
3a307ffc MF |
1780 | wc->w_handle = handle; |
1781 | ||
5dd4056d CH |
1782 | if (clusters_to_alloc) { |
1783 | ret = dquot_alloc_space_nodirty(inode, | |
1784 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); | |
1785 | if (ret) | |
1786 | goto out_commit; | |
a90714c1 | 1787 | } |
7f27ec97 | 1788 | |
0cf2f763 | 1789 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, |
13723d00 | 1790 | OCFS2_JOURNAL_ACCESS_WRITE); |
3a307ffc | 1791 | if (ret) { |
9517bac6 | 1792 | mlog_errno(ret); |
a90714c1 | 1793 | goto out_quota; |
9517bac6 MF |
1794 | } |
1795 | ||
3a307ffc MF |
1796 | /* |
1797 | * Fill our page array first. That way we've grabbed enough so | |
1798 | * that we can zero and flush if we error after adding the | |
1799 | * extent. | |
1800 | */ | |
693c241a | 1801 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len, |
e7432675 | 1802 | cluster_of_pages, mmap_page); |
5cffff9e | 1803 | if (ret && ret != -EAGAIN) { |
9517bac6 | 1804 | mlog_errno(ret); |
a90714c1 | 1805 | goto out_quota; |
9517bac6 MF |
1806 | } |
1807 | ||
5cffff9e WW |
1808 | /* |
1809 | * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock | |
1810 | * the target page. In this case, we exit with no error and no target | |
1811 | * page. This will trigger the caller, page_mkwrite(), to re-try | |
1812 | * the operation. | |
1813 | */ | |
1814 | if (ret == -EAGAIN) { | |
1815 | BUG_ON(wc->w_target_page); | |
1816 | ret = 0; | |
1817 | goto out_quota; | |
1818 | } | |
1819 | ||
0d172baa MF |
1820 | ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos, |
1821 | len); | |
1822 | if (ret) { | |
1823 | mlog_errno(ret); | |
a90714c1 | 1824 | goto out_quota; |
9517bac6 | 1825 | } |
9517bac6 | 1826 | |
3a307ffc MF |
1827 | if (data_ac) |
1828 | ocfs2_free_alloc_context(data_ac); | |
1829 | if (meta_ac) | |
1830 | ocfs2_free_alloc_context(meta_ac); | |
9517bac6 | 1831 | |
1afc32b9 | 1832 | success: |
65c4db8c RD |
1833 | if (pagep) |
1834 | *pagep = wc->w_target_page; | |
3a307ffc MF |
1835 | *fsdata = wc; |
1836 | return 0; | |
a90714c1 JK |
1837 | out_quota: |
1838 | if (clusters_to_alloc) | |
5dd4056d | 1839 | dquot_free_space(inode, |
a90714c1 | 1840 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); |
9517bac6 MF |
1841 | out_commit: |
1842 | ocfs2_commit_trans(osb, handle); | |
1843 | ||
9517bac6 | 1844 | out: |
c33f0785 ER |
1845 | /* |
1846 | * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(), | |
1847 | * even in case of error here like ENOSPC and ENOMEM. So, we need | |
1848 | * to unlock the target page manually to prevent deadlocks when | |
1849 | * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED | |
1850 | * to VM code. | |
1851 | */ | |
1852 | if (wc->w_target_locked) | |
1853 | unlock_page(mmap_page); | |
1854 | ||
4506cfb6 | 1855 | ocfs2_free_write_ctxt(inode, wc); |
3a307ffc | 1856 | |
b1214e47 | 1857 | if (data_ac) { |
9517bac6 | 1858 | ocfs2_free_alloc_context(data_ac); |
b1214e47 X |
1859 | data_ac = NULL; |
1860 | } | |
1861 | if (meta_ac) { | |
9517bac6 | 1862 | ocfs2_free_alloc_context(meta_ac); |
b1214e47 X |
1863 | meta_ac = NULL; |
1864 | } | |
50308d81 TM |
1865 | |
1866 | if (ret == -ENOSPC && try_free) { | |
1867 | /* | |
1868 | * Try to free some truncate log so that we can have enough | |
1869 | * clusters to allocate. | |
1870 | */ | |
1871 | try_free = 0; | |
1872 | ||
1873 | ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); | |
1874 | if (ret1 == 1) | |
1875 | goto try_again; | |
1876 | ||
1877 | if (ret1 < 0) | |
1878 | mlog_errno(ret1); | |
1879 | } | |
1880 | ||
3a307ffc MF |
1881 | return ret; |
1882 | } | |
1883 | ||
b6af1bcd NP |
1884 | static int ocfs2_write_begin(struct file *file, struct address_space *mapping, |
1885 | loff_t pos, unsigned len, unsigned flags, | |
1886 | struct page **pagep, void **fsdata) | |
607d44aa MF |
1887 | { |
1888 | int ret; | |
1889 | struct buffer_head *di_bh = NULL; | |
1890 | struct inode *inode = mapping->host; | |
1891 | ||
e63aecb6 | 1892 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
607d44aa MF |
1893 | if (ret) { |
1894 | mlog_errno(ret); | |
1895 | return ret; | |
1896 | } | |
1897 | ||
1898 | /* | |
1899 | * Take alloc sem here to prevent concurrent lookups. That way | |
1900 | * the mapping, zeroing and tree manipulation within | |
1901 | * ocfs2_write() will be safe against ->readpage(). This | |
1902 | * should also serve to lock out allocation from a shared | |
1903 | * writeable region. | |
1904 | */ | |
1905 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1906 | ||
c1ad1e3c RD |
1907 | ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER, |
1908 | pagep, fsdata, di_bh, NULL); | |
607d44aa MF |
1909 | if (ret) { |
1910 | mlog_errno(ret); | |
c934a92d | 1911 | goto out_fail; |
607d44aa MF |
1912 | } |
1913 | ||
1914 | brelse(di_bh); | |
1915 | ||
1916 | return 0; | |
1917 | ||
607d44aa MF |
1918 | out_fail: |
1919 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1920 | ||
1921 | brelse(di_bh); | |
e63aecb6 | 1922 | ocfs2_inode_unlock(inode, 1); |
607d44aa MF |
1923 | |
1924 | return ret; | |
1925 | } | |
1926 | ||
1afc32b9 MF |
1927 | static void ocfs2_write_end_inline(struct inode *inode, loff_t pos, |
1928 | unsigned len, unsigned *copied, | |
1929 | struct ocfs2_dinode *di, | |
1930 | struct ocfs2_write_ctxt *wc) | |
1931 | { | |
1932 | void *kaddr; | |
1933 | ||
1934 | if (unlikely(*copied < len)) { | |
1935 | if (!PageUptodate(wc->w_target_page)) { | |
1936 | *copied = 0; | |
1937 | return; | |
1938 | } | |
1939 | } | |
1940 | ||
c4bc8dcb | 1941 | kaddr = kmap_atomic(wc->w_target_page); |
1afc32b9 | 1942 | memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied); |
c4bc8dcb | 1943 | kunmap_atomic(kaddr); |
1afc32b9 | 1944 | |
9558156b TM |
1945 | trace_ocfs2_write_end_inline( |
1946 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1afc32b9 MF |
1947 | (unsigned long long)pos, *copied, |
1948 | le16_to_cpu(di->id2.i_data.id_count), | |
1949 | le16_to_cpu(di->i_dyn_features)); | |
1950 | } | |
1951 | ||
7307de80 | 1952 | int ocfs2_write_end_nolock(struct address_space *mapping, |
07f38d97 | 1953 | loff_t pos, unsigned len, unsigned copied, void *fsdata) |
3a307ffc | 1954 | { |
7f27ec97 | 1955 | int i, ret; |
09cbfeaf | 1956 | unsigned from, to, start = pos & (PAGE_SIZE - 1); |
3a307ffc MF |
1957 | struct inode *inode = mapping->host; |
1958 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1959 | struct ocfs2_write_ctxt *wc = fsdata; | |
1960 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | |
1961 | handle_t *handle = wc->w_handle; | |
1962 | struct page *tmppage; | |
1963 | ||
4506cfb6 RD |
1964 | BUG_ON(!list_empty(&wc->w_unwritten_list)); |
1965 | ||
46e62556 RD |
1966 | if (handle) { |
1967 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), | |
1968 | wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE); | |
1969 | if (ret) { | |
1970 | copied = ret; | |
1971 | mlog_errno(ret); | |
1972 | goto out; | |
1973 | } | |
7f27ec97 | 1974 | } |
1975 | ||
1afc32b9 MF |
1976 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
1977 | ocfs2_write_end_inline(inode, pos, len, &copied, di, wc); | |
1978 | goto out_write_size; | |
1979 | } | |
1980 | ||
65c4db8c | 1981 | if (unlikely(copied < len) && wc->w_target_page) { |
3a307ffc MF |
1982 | if (!PageUptodate(wc->w_target_page)) |
1983 | copied = 0; | |
1984 | ||
1985 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, | |
1986 | start+len); | |
1987 | } | |
65c4db8c RD |
1988 | if (wc->w_target_page) |
1989 | flush_dcache_page(wc->w_target_page); | |
3a307ffc MF |
1990 | |
1991 | for(i = 0; i < wc->w_num_pages; i++) { | |
1992 | tmppage = wc->w_pages[i]; | |
1993 | ||
65c4db8c RD |
1994 | /* This is the direct io target page. */ |
1995 | if (tmppage == NULL) | |
1996 | continue; | |
1997 | ||
3a307ffc MF |
1998 | if (tmppage == wc->w_target_page) { |
1999 | from = wc->w_target_from; | |
2000 | to = wc->w_target_to; | |
2001 | ||
09cbfeaf KS |
2002 | BUG_ON(from > PAGE_SIZE || |
2003 | to > PAGE_SIZE || | |
3a307ffc MF |
2004 | to < from); |
2005 | } else { | |
2006 | /* | |
2007 | * Pages adjacent to the target (if any) imply | |
2008 | * a hole-filling write in which case we want | |
2009 | * to flush their entire range. | |
2010 | */ | |
2011 | from = 0; | |
09cbfeaf | 2012 | to = PAGE_SIZE; |
3a307ffc MF |
2013 | } |
2014 | ||
961cecbe | 2015 | if (page_has_buffers(tmppage)) { |
bbd0f327 JQ |
2016 | if (handle && ocfs2_should_order_data(inode)) { |
2017 | loff_t start_byte = | |
2018 | ((loff_t)tmppage->index << PAGE_SHIFT) + | |
2019 | from; | |
2020 | loff_t length = to - from; | |
2021 | ocfs2_jbd2_inode_add_write(handle, inode, | |
2022 | start_byte, length); | |
2023 | } | |
961cecbe SM |
2024 | block_commit_write(tmppage, from, to); |
2025 | } | |
3a307ffc MF |
2026 | } |
2027 | ||
1afc32b9 | 2028 | out_write_size: |
46e62556 RD |
2029 | /* Direct io do not update i_size here. */ |
2030 | if (wc->w_type != OCFS2_WRITE_DIRECT) { | |
2031 | pos += copied; | |
2032 | if (pos > i_size_read(inode)) { | |
2033 | i_size_write(inode, pos); | |
2034 | mark_inode_dirty(inode); | |
2035 | } | |
2036 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
2037 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
078cd827 | 2038 | inode->i_mtime = inode->i_ctime = current_time(inode); |
46e62556 RD |
2039 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); |
2040 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
583fee3e JJB |
2041 | if (handle) |
2042 | ocfs2_update_inode_fsync_trans(handle, inode, 1); | |
46e62556 RD |
2043 | } |
2044 | if (handle) | |
2045 | ocfs2_journal_dirty(handle, wc->w_di_bh); | |
3a307ffc | 2046 | |
7f27ec97 | 2047 | out: |
136f49b9 JB |
2048 | /* unlock pages before dealloc since it needs acquiring j_trans_barrier |
2049 | * lock, or it will cause a deadlock since journal commit threads holds | |
2050 | * this lock and will ask for the page lock when flushing the data. | |
2051 | * put it here to preserve the unlock order. | |
2052 | */ | |
2053 | ocfs2_unlock_pages(wc); | |
2054 | ||
46e62556 RD |
2055 | if (handle) |
2056 | ocfs2_commit_trans(osb, handle); | |
59a5e416 | 2057 | |
b27b7cbc MF |
2058 | ocfs2_run_deallocs(osb, &wc->w_dealloc); |
2059 | ||
136f49b9 JB |
2060 | brelse(wc->w_di_bh); |
2061 | kfree(wc); | |
607d44aa MF |
2062 | |
2063 | return copied; | |
2064 | } | |
2065 | ||
b6af1bcd NP |
2066 | static int ocfs2_write_end(struct file *file, struct address_space *mapping, |
2067 | loff_t pos, unsigned len, unsigned copied, | |
2068 | struct page *page, void *fsdata) | |
607d44aa MF |
2069 | { |
2070 | int ret; | |
2071 | struct inode *inode = mapping->host; | |
2072 | ||
07f38d97 | 2073 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, fsdata); |
607d44aa | 2074 | |
3a307ffc | 2075 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
e63aecb6 | 2076 | ocfs2_inode_unlock(inode, 1); |
9517bac6 | 2077 | |
607d44aa | 2078 | return ret; |
9517bac6 MF |
2079 | } |
2080 | ||
c15471f7 RD |
2081 | struct ocfs2_dio_write_ctxt { |
2082 | struct list_head dw_zero_list; | |
2083 | unsigned dw_zero_count; | |
2084 | int dw_orphaned; | |
2085 | pid_t dw_writer_pid; | |
2086 | }; | |
2087 | ||
2088 | static struct ocfs2_dio_write_ctxt * | |
2089 | ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc) | |
2090 | { | |
2091 | struct ocfs2_dio_write_ctxt *dwc = NULL; | |
2092 | ||
2093 | if (bh->b_private) | |
2094 | return bh->b_private; | |
2095 | ||
2096 | dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS); | |
2097 | if (dwc == NULL) | |
2098 | return NULL; | |
2099 | INIT_LIST_HEAD(&dwc->dw_zero_list); | |
2100 | dwc->dw_zero_count = 0; | |
2101 | dwc->dw_orphaned = 0; | |
2102 | dwc->dw_writer_pid = task_pid_nr(current); | |
2103 | bh->b_private = dwc; | |
2104 | *alloc = 1; | |
2105 | ||
2106 | return dwc; | |
2107 | } | |
2108 | ||
2109 | static void ocfs2_dio_free_write_ctx(struct inode *inode, | |
2110 | struct ocfs2_dio_write_ctxt *dwc) | |
2111 | { | |
2112 | ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list); | |
2113 | kfree(dwc); | |
2114 | } | |
2115 | ||
2116 | /* | |
2117 | * TODO: Make this into a generic get_blocks function. | |
2118 | * | |
2119 | * From do_direct_io in direct-io.c: | |
2120 | * "So what we do is to permit the ->get_blocks function to populate | |
2121 | * bh.b_size with the size of IO which is permitted at this offset and | |
2122 | * this i_blkbits." | |
2123 | * | |
2124 | * This function is called directly from get_more_blocks in direct-io.c. | |
2125 | * | |
2126 | * called like this: dio->get_blocks(dio->inode, fs_startblk, | |
2127 | * fs_count, map_bh, dio->rw == WRITE); | |
2128 | */ | |
3e4c56d4 | 2129 | static int ocfs2_dio_wr_get_block(struct inode *inode, sector_t iblock, |
c15471f7 RD |
2130 | struct buffer_head *bh_result, int create) |
2131 | { | |
2132 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
a86a72a4 | 2133 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
c15471f7 RD |
2134 | struct ocfs2_write_ctxt *wc; |
2135 | struct ocfs2_write_cluster_desc *desc = NULL; | |
2136 | struct ocfs2_dio_write_ctxt *dwc = NULL; | |
2137 | struct buffer_head *di_bh = NULL; | |
2138 | u64 p_blkno; | |
7a243c82 JG |
2139 | unsigned int i_blkbits = inode->i_sb->s_blocksize_bits; |
2140 | loff_t pos = iblock << i_blkbits; | |
2141 | sector_t endblk = (i_size_read(inode) - 1) >> i_blkbits; | |
c15471f7 RD |
2142 | unsigned len, total_len = bh_result->b_size; |
2143 | int ret = 0, first_get_block = 0; | |
2144 | ||
2145 | len = osb->s_clustersize - (pos & (osb->s_clustersize - 1)); | |
2146 | len = min(total_len, len); | |
2147 | ||
7a243c82 JG |
2148 | /* |
2149 | * bh_result->b_size is count in get_more_blocks according to write | |
2150 | * "pos" and "end", we need map twice to return different buffer state: | |
2151 | * 1. area in file size, not set NEW; | |
2152 | * 2. area out file size, set NEW. | |
2153 | * | |
2154 | * iblock endblk | |
2155 | * |--------|---------|---------|--------- | |
2156 | * |<-------area in file------->| | |
2157 | */ | |
2158 | ||
2159 | if ((iblock <= endblk) && | |
2160 | ((iblock + ((len - 1) >> i_blkbits)) > endblk)) | |
2161 | len = (endblk - iblock + 1) << i_blkbits; | |
2162 | ||
c15471f7 RD |
2163 | mlog(0, "get block of %lu at %llu:%u req %u\n", |
2164 | inode->i_ino, pos, len, total_len); | |
2165 | ||
ce170828 RD |
2166 | /* |
2167 | * Because we need to change file size in ocfs2_dio_end_io_write(), or | |
2168 | * we may need to add it to orphan dir. So can not fall to fast path | |
2169 | * while file size will be changed. | |
2170 | */ | |
2171 | if (pos + total_len <= i_size_read(inode)) { | |
a86a72a4 | 2172 | |
3e4c56d4 | 2173 | /* This is the fast path for re-write. */ |
2174 | ret = ocfs2_lock_get_block(inode, iblock, bh_result, create); | |
ce170828 RD |
2175 | if (buffer_mapped(bh_result) && |
2176 | !buffer_new(bh_result) && | |
2177 | ret == 0) | |
2178 | goto out; | |
c15471f7 | 2179 | |
ce170828 RD |
2180 | /* Clear state set by ocfs2_get_block. */ |
2181 | bh_result->b_state = 0; | |
2182 | } | |
c15471f7 RD |
2183 | |
2184 | dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block); | |
2185 | if (unlikely(dwc == NULL)) { | |
2186 | ret = -ENOMEM; | |
2187 | mlog_errno(ret); | |
2188 | goto out; | |
2189 | } | |
2190 | ||
2191 | if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) > | |
2192 | ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) && | |
2193 | !dwc->dw_orphaned) { | |
2194 | /* | |
2195 | * when we are going to alloc extents beyond file size, add the | |
2196 | * inode to orphan dir, so we can recall those spaces when | |
2197 | * system crashed during write. | |
2198 | */ | |
2199 | ret = ocfs2_add_inode_to_orphan(osb, inode); | |
2200 | if (ret < 0) { | |
2201 | mlog_errno(ret); | |
2202 | goto out; | |
2203 | } | |
2204 | dwc->dw_orphaned = 1; | |
2205 | } | |
2206 | ||
2207 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
2208 | if (ret) { | |
2209 | mlog_errno(ret); | |
2210 | goto out; | |
2211 | } | |
2212 | ||
a86a72a4 RD |
2213 | down_write(&oi->ip_alloc_sem); |
2214 | ||
c15471f7 | 2215 | if (first_get_block) { |
1119d3c0 | 2216 | if (ocfs2_sparse_alloc(osb)) |
c15471f7 RD |
2217 | ret = ocfs2_zero_tail(inode, di_bh, pos); |
2218 | else | |
2219 | ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, | |
2220 | total_len, NULL); | |
2221 | if (ret < 0) { | |
2222 | mlog_errno(ret); | |
2223 | goto unlock; | |
2224 | } | |
2225 | } | |
2226 | ||
2227 | ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len, | |
2228 | OCFS2_WRITE_DIRECT, NULL, | |
2229 | (void **)&wc, di_bh, NULL); | |
2230 | if (ret) { | |
2231 | mlog_errno(ret); | |
2232 | goto unlock; | |
2233 | } | |
2234 | ||
2235 | desc = &wc->w_desc[0]; | |
2236 | ||
2237 | p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys); | |
2238 | BUG_ON(p_blkno == 0); | |
2239 | p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1); | |
2240 | ||
2241 | map_bh(bh_result, inode->i_sb, p_blkno); | |
2242 | bh_result->b_size = len; | |
2243 | if (desc->c_needs_zero) | |
2244 | set_buffer_new(bh_result); | |
2245 | ||
7a243c82 JG |
2246 | if (iblock > endblk) |
2247 | set_buffer_new(bh_result); | |
2248 | ||
c15471f7 RD |
2249 | /* May sleep in end_io. It should not happen in a irq context. So defer |
2250 | * it to dio work queue. */ | |
2251 | set_buffer_defer_completion(bh_result); | |
2252 | ||
2253 | if (!list_empty(&wc->w_unwritten_list)) { | |
2254 | struct ocfs2_unwritten_extent *ue = NULL; | |
2255 | ||
2256 | ue = list_first_entry(&wc->w_unwritten_list, | |
2257 | struct ocfs2_unwritten_extent, | |
2258 | ue_node); | |
2259 | BUG_ON(ue->ue_cpos != desc->c_cpos); | |
2260 | /* The physical address may be 0, fill it. */ | |
2261 | ue->ue_phys = desc->c_phys; | |
2262 | ||
2263 | list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list); | |
63de8bd9 | 2264 | dwc->dw_zero_count += wc->w_unwritten_count; |
c15471f7 RD |
2265 | } |
2266 | ||
07f38d97 | 2267 | ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, wc); |
c15471f7 RD |
2268 | BUG_ON(ret != len); |
2269 | ret = 0; | |
2270 | unlock: | |
a86a72a4 | 2271 | up_write(&oi->ip_alloc_sem); |
c15471f7 RD |
2272 | ocfs2_inode_unlock(inode, 1); |
2273 | brelse(di_bh); | |
2274 | out: | |
2275 | if (ret < 0) | |
2276 | ret = -EIO; | |
2277 | return ret; | |
2278 | } | |
2279 | ||
08554955 DW |
2280 | static int ocfs2_dio_end_io_write(struct inode *inode, |
2281 | struct ocfs2_dio_write_ctxt *dwc, | |
2282 | loff_t offset, | |
2283 | ssize_t bytes) | |
c15471f7 RD |
2284 | { |
2285 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
2286 | struct ocfs2_extent_tree et; | |
2287 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
a86a72a4 | 2288 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
c15471f7 RD |
2289 | struct ocfs2_unwritten_extent *ue = NULL; |
2290 | struct buffer_head *di_bh = NULL; | |
2291 | struct ocfs2_dinode *di; | |
2292 | struct ocfs2_alloc_context *data_ac = NULL; | |
2293 | struct ocfs2_alloc_context *meta_ac = NULL; | |
2294 | handle_t *handle = NULL; | |
2295 | loff_t end = offset + bytes; | |
90bd070a | 2296 | int ret = 0, credits = 0; |
c15471f7 RD |
2297 | |
2298 | ocfs2_init_dealloc_ctxt(&dealloc); | |
2299 | ||
2300 | /* We do clear unwritten, delete orphan, change i_size here. If neither | |
2301 | * of these happen, we can skip all this. */ | |
2302 | if (list_empty(&dwc->dw_zero_list) && | |
2303 | end <= i_size_read(inode) && | |
2304 | !dwc->dw_orphaned) | |
2305 | goto out; | |
2306 | ||
a86a72a4 RD |
2307 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2308 | if (ret < 0) { | |
2309 | mlog_errno(ret); | |
2310 | goto out; | |
2311 | } | |
2312 | ||
2313 | down_write(&oi->ip_alloc_sem); | |
2314 | ||
c15471f7 RD |
2315 | /* Delete orphan before acquire i_mutex. */ |
2316 | if (dwc->dw_orphaned) { | |
2317 | BUG_ON(dwc->dw_writer_pid != task_pid_nr(current)); | |
2318 | ||
2319 | end = end > i_size_read(inode) ? end : 0; | |
2320 | ||
2321 | ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh, | |
2322 | !!end, end); | |
2323 | if (ret < 0) | |
2324 | mlog_errno(ret); | |
2325 | } | |
2326 | ||
aef73a61 | 2327 | di = (struct ocfs2_dinode *)di_bh->b_data; |
c15471f7 RD |
2328 | |
2329 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); | |
2330 | ||
71a36944 CG |
2331 | /* Attach dealloc with extent tree in case that we may reuse extents |
2332 | * which are already unlinked from current extent tree due to extent | |
2333 | * rotation and merging. | |
2334 | */ | |
2335 | et.et_dealloc = &dealloc; | |
2336 | ||
c15471f7 RD |
2337 | ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2, |
2338 | &data_ac, &meta_ac); | |
28888681 RD |
2339 | if (ret) { |
2340 | mlog_errno(ret); | |
2341 | goto unlock; | |
2342 | } | |
c15471f7 RD |
2343 | |
2344 | credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list); | |
2345 | ||
2346 | handle = ocfs2_start_trans(osb, credits); | |
2347 | if (IS_ERR(handle)) { | |
2348 | ret = PTR_ERR(handle); | |
2349 | mlog_errno(ret); | |
2350 | goto unlock; | |
2351 | } | |
2352 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, | |
2353 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2354 | if (ret) { | |
2355 | mlog_errno(ret); | |
2356 | goto commit; | |
2357 | } | |
2358 | ||
2359 | list_for_each_entry(ue, &dwc->dw_zero_list, ue_node) { | |
2360 | ret = ocfs2_mark_extent_written(inode, &et, handle, | |
2361 | ue->ue_cpos, 1, | |
2362 | ue->ue_phys, | |
2363 | meta_ac, &dealloc); | |
2364 | if (ret < 0) { | |
2365 | mlog_errno(ret); | |
2366 | break; | |
2367 | } | |
2368 | } | |
2369 | ||
2370 | if (end > i_size_read(inode)) { | |
2371 | ret = ocfs2_set_inode_size(handle, inode, di_bh, end); | |
2372 | if (ret < 0) | |
2373 | mlog_errno(ret); | |
2374 | } | |
2375 | commit: | |
2376 | ocfs2_commit_trans(osb, handle); | |
2377 | unlock: | |
a86a72a4 | 2378 | up_write(&oi->ip_alloc_sem); |
c15471f7 RD |
2379 | ocfs2_inode_unlock(inode, 1); |
2380 | brelse(di_bh); | |
2381 | out: | |
c15471f7 RD |
2382 | if (data_ac) |
2383 | ocfs2_free_alloc_context(data_ac); | |
2384 | if (meta_ac) | |
2385 | ocfs2_free_alloc_context(meta_ac); | |
28888681 | 2386 | ocfs2_run_deallocs(osb, &dealloc); |
28888681 | 2387 | ocfs2_dio_free_write_ctx(inode, dwc); |
08554955 DW |
2388 | |
2389 | return ret; | |
c15471f7 RD |
2390 | } |
2391 | ||
2392 | /* | |
2393 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're | |
2394 | * particularly interested in the aio/dio case. We use the rw_lock DLM lock | |
2395 | * to protect io on one node from truncation on another. | |
2396 | */ | |
2397 | static int ocfs2_dio_end_io(struct kiocb *iocb, | |
2398 | loff_t offset, | |
2399 | ssize_t bytes, | |
2400 | void *private) | |
2401 | { | |
2402 | struct inode *inode = file_inode(iocb->ki_filp); | |
2403 | int level; | |
08554955 | 2404 | int ret = 0; |
c15471f7 RD |
2405 | |
2406 | /* this io's submitter should not have unlocked this before we could */ | |
2407 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); | |
2408 | ||
5040f8df WW |
2409 | if (bytes <= 0) |
2410 | mlog_ratelimited(ML_ERROR, "Direct IO failed, bytes = %lld", | |
2411 | (long long)bytes); | |
2412 | if (private) { | |
2413 | if (bytes > 0) | |
2414 | ret = ocfs2_dio_end_io_write(inode, private, offset, | |
2415 | bytes); | |
2416 | else | |
2417 | ocfs2_dio_free_write_ctx(inode, private); | |
2418 | } | |
c15471f7 RD |
2419 | |
2420 | ocfs2_iocb_clear_rw_locked(iocb); | |
2421 | ||
2422 | level = ocfs2_iocb_rw_locked_level(iocb); | |
2423 | ocfs2_rw_unlock(inode, level); | |
08554955 | 2424 | return ret; |
c15471f7 RD |
2425 | } |
2426 | ||
c8b8e32d | 2427 | static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
c15471f7 RD |
2428 | { |
2429 | struct file *file = iocb->ki_filp; | |
93c76a3d | 2430 | struct inode *inode = file->f_mapping->host; |
c15471f7 | 2431 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
c15471f7 RD |
2432 | get_block_t *get_block; |
2433 | ||
2434 | /* | |
2435 | * Fallback to buffered I/O if we see an inode without | |
2436 | * extents. | |
2437 | */ | |
2438 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
2439 | return 0; | |
2440 | ||
2441 | /* Fallback to buffered I/O if we do not support append dio. */ | |
c8b8e32d CH |
2442 | if (iocb->ki_pos + iter->count > i_size_read(inode) && |
2443 | !ocfs2_supports_append_dio(osb)) | |
c15471f7 RD |
2444 | return 0; |
2445 | ||
2446 | if (iov_iter_rw(iter) == READ) | |
3e4c56d4 | 2447 | get_block = ocfs2_lock_get_block; |
c15471f7 | 2448 | else |
3e4c56d4 | 2449 | get_block = ocfs2_dio_wr_get_block; |
c15471f7 RD |
2450 | |
2451 | return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, | |
c8b8e32d | 2452 | iter, get_block, |
c15471f7 RD |
2453 | ocfs2_dio_end_io, NULL, 0); |
2454 | } | |
2455 | ||
f5e54d6e | 2456 | const struct address_space_operations ocfs2_aops = { |
1fca3a05 | 2457 | .readpage = ocfs2_readpage, |
d4388340 | 2458 | .readahead = ocfs2_readahead, |
1fca3a05 HH |
2459 | .writepage = ocfs2_writepage, |
2460 | .write_begin = ocfs2_write_begin, | |
2461 | .write_end = ocfs2_write_end, | |
2462 | .bmap = ocfs2_bmap, | |
1fca3a05 | 2463 | .direct_IO = ocfs2_direct_IO, |
41ecc345 | 2464 | .invalidatepage = block_invalidatepage, |
1fca3a05 HH |
2465 | .releasepage = ocfs2_releasepage, |
2466 | .migratepage = buffer_migrate_page, | |
2467 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 2468 | .error_remove_page = generic_error_remove_page, |
ccd979bd | 2469 | }; |