]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
15 | #include <linux/pagemap.h> | |
fd88de56 | 16 | #include <linux/pagevec.h> |
9b124fbb | 17 | #include <linux/mpage.h> |
d1665e41 | 18 | #include <linux/fs.h> |
5c676f6d | 19 | #include <linux/gfs2_ondisk.h> |
7d308590 | 20 | #include <linux/lm_interface.h> |
b3b94faa DT |
21 | |
22 | #include "gfs2.h" | |
5c676f6d | 23 | #include "incore.h" |
b3b94faa DT |
24 | #include "bmap.h" |
25 | #include "glock.h" | |
26 | #include "inode.h" | |
b3b94faa DT |
27 | #include "log.h" |
28 | #include "meta_io.h" | |
29 | #include "ops_address.h" | |
b3b94faa DT |
30 | #include "quota.h" |
31 | #include "trans.h" | |
18ec7d5c | 32 | #include "rgrp.h" |
61a30dcb | 33 | #include "ops_file.h" |
5c676f6d | 34 | #include "util.h" |
4340fe62 | 35 | #include "glops.h" |
b3b94faa | 36 | |
ba7f7290 SW |
37 | |
38 | static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page, | |
39 | unsigned int from, unsigned int to) | |
40 | { | |
41 | struct buffer_head *head = page_buffers(page); | |
42 | unsigned int bsize = head->b_size; | |
43 | struct buffer_head *bh; | |
44 | unsigned int start, end; | |
45 | ||
46 | for (bh = head, start = 0; bh != head || !start; | |
47 | bh = bh->b_this_page, start = end) { | |
48 | end = start + bsize; | |
49 | if (end <= from || start >= to) | |
50 | continue; | |
51 | gfs2_trans_add_bh(ip->i_gl, bh, 0); | |
52 | } | |
53 | } | |
54 | ||
b3b94faa | 55 | /** |
4ff14670 | 56 | * gfs2_get_block - Fills in a buffer head with details about a block |
b3b94faa DT |
57 | * @inode: The inode |
58 | * @lblock: The block number to look up | |
59 | * @bh_result: The buffer head to return the result in | |
60 | * @create: Non-zero if we may add block to the file | |
61 | * | |
62 | * Returns: errno | |
63 | */ | |
64 | ||
4ff14670 SW |
65 | int gfs2_get_block(struct inode *inode, sector_t lblock, |
66 | struct buffer_head *bh_result, int create) | |
b3b94faa | 67 | { |
23591256 | 68 | return gfs2_block_map(inode, lblock, create, bh_result); |
b3b94faa DT |
69 | } |
70 | ||
71 | /** | |
7a6bbacb | 72 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block |
b3b94faa DT |
73 | * @inode: The inode |
74 | * @lblock: The block number to look up | |
75 | * @bh_result: The buffer head to return the result in | |
76 | * @create: Non-zero if we may add block to the file | |
77 | * | |
78 | * Returns: errno | |
79 | */ | |
80 | ||
7a6bbacb SW |
81 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, |
82 | struct buffer_head *bh_result, int create) | |
b3b94faa | 83 | { |
b3b94faa DT |
84 | int error; |
85 | ||
23591256 | 86 | error = gfs2_block_map(inode, lblock, 0, bh_result); |
b3b94faa DT |
87 | if (error) |
88 | return error; | |
7a6bbacb SW |
89 | if (bh_result->b_blocknr == 0) |
90 | return -EIO; | |
91 | return 0; | |
b3b94faa DT |
92 | } |
93 | ||
7a6bbacb SW |
94 | static int gfs2_get_block_direct(struct inode *inode, sector_t lblock, |
95 | struct buffer_head *bh_result, int create) | |
623d9355 | 96 | { |
23591256 | 97 | return gfs2_block_map(inode, lblock, 0, bh_result); |
623d9355 | 98 | } |
7a6bbacb | 99 | |
b3b94faa DT |
100 | /** |
101 | * gfs2_writepage - Write complete page | |
102 | * @page: Page to write | |
103 | * | |
104 | * Returns: errno | |
105 | * | |
18ec7d5c SW |
106 | * Some of this is copied from block_write_full_page() although we still |
107 | * call it to do most of the work. | |
b3b94faa DT |
108 | */ |
109 | ||
110 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) | |
111 | { | |
18ec7d5c | 112 | struct inode *inode = page->mapping->host; |
f4387149 SW |
113 | struct gfs2_inode *ip = GFS2_I(inode); |
114 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
115 | loff_t i_size = i_size_read(inode); |
116 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
117 | unsigned offset; | |
b3b94faa | 118 | int error; |
18ec7d5c | 119 | int done_trans = 0; |
b3b94faa | 120 | |
b3b94faa DT |
121 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) { |
122 | unlock_page(page); | |
123 | return -EIO; | |
124 | } | |
5c676f6d | 125 | if (current->journal_info) |
18ec7d5c SW |
126 | goto out_ignore; |
127 | ||
128 | /* Is the page fully outside i_size? (truncate in progress) */ | |
129 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
d2d7b8a2 | 130 | if (page->index > end_index || (page->index == end_index && !offset)) { |
18ec7d5c | 131 | page->mapping->a_ops->invalidatepage(page, 0); |
b3b94faa | 132 | unlock_page(page); |
18ec7d5c | 133 | return 0; /* don't care */ |
b3b94faa DT |
134 | } |
135 | ||
18ec7d5c SW |
136 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) { |
137 | error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0); | |
138 | if (error) | |
139 | goto out_ignore; | |
f4387149 SW |
140 | if (!page_has_buffers(page)) { |
141 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
142 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
143 | } | |
18ec7d5c SW |
144 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1); |
145 | done_trans = 1; | |
146 | } | |
7a6bbacb | 147 | error = block_write_full_page(page, gfs2_get_block_noalloc, wbc); |
18ec7d5c SW |
148 | if (done_trans) |
149 | gfs2_trans_end(sdp); | |
b3b94faa | 150 | gfs2_meta_cache_flush(ip); |
b3b94faa | 151 | return error; |
18ec7d5c SW |
152 | |
153 | out_ignore: | |
154 | redirty_page_for_writepage(wbc, page); | |
155 | unlock_page(page); | |
156 | return 0; | |
b3b94faa DT |
157 | } |
158 | ||
159 | /** | |
160 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
161 | * @ip: the inode | |
162 | * @page: the page | |
163 | * | |
164 | * Returns: errno | |
165 | */ | |
166 | ||
167 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) | |
168 | { | |
169 | struct buffer_head *dibh; | |
170 | void *kaddr; | |
171 | int error; | |
172 | ||
61057c6b | 173 | BUG_ON(page->index); |
fd88de56 | 174 | |
b3b94faa DT |
175 | error = gfs2_meta_inode_buffer(ip, &dibh); |
176 | if (error) | |
177 | return error; | |
178 | ||
5c4e9e03 | 179 | kaddr = kmap_atomic(page, KM_USER0); |
fd88de56 | 180 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), |
b3b94faa | 181 | ip->i_di.di_size); |
fd88de56 | 182 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); |
c312c4fd | 183 | kunmap_atomic(kaddr, KM_USER0); |
b3b94faa DT |
184 | |
185 | brelse(dibh); | |
186 | ||
187 | SetPageUptodate(page); | |
188 | ||
189 | return 0; | |
190 | } | |
191 | ||
b3b94faa | 192 | |
b3b94faa DT |
193 | /** |
194 | * gfs2_readpage - readpage with locking | |
18ec7d5c SW |
195 | * @file: The file to read a page for. N.B. This may be NULL if we are |
196 | * reading an internal file. | |
b3b94faa DT |
197 | * @page: The page to read |
198 | * | |
199 | * Returns: errno | |
200 | */ | |
201 | ||
202 | static int gfs2_readpage(struct file *file, struct page *page) | |
203 | { | |
feaa7bba SW |
204 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
205 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
dc41aeed | 206 | struct gfs2_file *gf = NULL; |
18ec7d5c | 207 | struct gfs2_holder gh; |
b3b94faa | 208 | int error; |
59a1cc6b | 209 | int do_unlock = 0; |
b3b94faa | 210 | |
dd538c83 | 211 | if (likely(file != &gfs2_internal_file_sentinel)) { |
59a1cc6b | 212 | if (file) { |
dc41aeed | 213 | gf = file->private_data; |
59a1cc6b | 214 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) |
dc41aeed | 215 | /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */ |
59a1cc6b SW |
216 | goto skip_lock; |
217 | } | |
2ca99501 | 218 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh); |
59a1cc6b | 219 | do_unlock = 1; |
dcd24799 | 220 | error = gfs2_glock_nq_atime(&gh); |
fd88de56 | 221 | if (unlikely(error)) |
61a30dcb SW |
222 | goto out_unlock; |
223 | } | |
b3b94faa | 224 | |
59a1cc6b | 225 | skip_lock: |
18ec7d5c | 226 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
227 | error = stuffed_readpage(ip, page); |
228 | unlock_page(page); | |
b3b94faa | 229 | } else |
18ec7d5c | 230 | error = mpage_readpage(page, gfs2_get_block); |
b3b94faa DT |
231 | |
232 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
233 | error = -EIO; | |
234 | ||
07903c02 | 235 | if (do_unlock) { |
61a30dcb SW |
236 | gfs2_glock_dq_m(1, &gh); |
237 | gfs2_holder_uninit(&gh); | |
238 | } | |
18ec7d5c | 239 | out: |
b3b94faa | 240 | return error; |
18ec7d5c | 241 | out_unlock: |
2ca99501 SW |
242 | if (error == GLR_TRYFAILED) |
243 | error = AOP_TRUNCATED_PAGE; | |
18ec7d5c | 244 | unlock_page(page); |
59a1cc6b | 245 | if (do_unlock) |
fd88de56 SW |
246 | gfs2_holder_uninit(&gh); |
247 | goto out; | |
248 | } | |
249 | ||
fd88de56 SW |
250 | /** |
251 | * gfs2_readpages - Read a bunch of pages at once | |
252 | * | |
253 | * Some notes: | |
254 | * 1. This is only for readahead, so we can simply ignore any things | |
255 | * which are slightly inconvenient (such as locking conflicts between | |
256 | * the page lock and the glock) and return having done no I/O. Its | |
257 | * obviously not something we'd want to do on too regular a basis. | |
258 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 259 | * 2. We don't handle stuffed files here we let readpage do the honours. |
fd88de56 SW |
260 | * 3. mpage_readpages() does most of the heavy lifting in the common case. |
261 | * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places. | |
262 | * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as | |
263 | * well as read-ahead. | |
264 | */ | |
265 | static int gfs2_readpages(struct file *file, struct address_space *mapping, | |
266 | struct list_head *pages, unsigned nr_pages) | |
267 | { | |
268 | struct inode *inode = mapping->host; | |
feaa7bba SW |
269 | struct gfs2_inode *ip = GFS2_I(inode); |
270 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 | 271 | struct gfs2_holder gh; |
e1d5b18a | 272 | int ret = 0; |
59a1cc6b | 273 | int do_unlock = 0; |
fd88de56 | 274 | |
dd538c83 | 275 | if (likely(file != &gfs2_internal_file_sentinel)) { |
59a1cc6b SW |
276 | if (file) { |
277 | struct gfs2_file *gf = file->private_data; | |
278 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) | |
279 | goto skip_lock; | |
280 | } | |
fd88de56 | 281 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, |
2ca99501 | 282 | LM_FLAG_TRY_1CB|GL_ATIME, &gh); |
59a1cc6b | 283 | do_unlock = 1; |
dcd24799 | 284 | ret = gfs2_glock_nq_atime(&gh); |
907b9bce | 285 | if (ret == GLR_TRYFAILED) |
fd88de56 SW |
286 | goto out_noerror; |
287 | if (unlikely(ret)) | |
288 | goto out_unlock; | |
289 | } | |
59a1cc6b | 290 | skip_lock: |
e1d5b18a | 291 | if (!gfs2_is_stuffed(ip)) |
fd88de56 | 292 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block); |
fd88de56 | 293 | |
59a1cc6b | 294 | if (do_unlock) { |
fd88de56 SW |
295 | gfs2_glock_dq_m(1, &gh); |
296 | gfs2_holder_uninit(&gh); | |
297 | } | |
298 | out: | |
299 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
300 | ret = -EIO; | |
301 | return ret; | |
302 | out_noerror: | |
303 | ret = 0; | |
304 | out_unlock: | |
59a1cc6b | 305 | if (do_unlock) |
fd88de56 | 306 | gfs2_holder_uninit(&gh); |
18ec7d5c | 307 | goto out; |
b3b94faa DT |
308 | } |
309 | ||
310 | /** | |
311 | * gfs2_prepare_write - Prepare to write a page to a file | |
312 | * @file: The file to write to | |
313 | * @page: The page which is to be prepared for writing | |
314 | * @from: From (byte range within page) | |
315 | * @to: To (byte range within page) | |
316 | * | |
317 | * Returns: errno | |
318 | */ | |
319 | ||
320 | static int gfs2_prepare_write(struct file *file, struct page *page, | |
321 | unsigned from, unsigned to) | |
322 | { | |
feaa7bba SW |
323 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
324 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
18ec7d5c SW |
325 | unsigned int data_blocks, ind_blocks, rblocks; |
326 | int alloc_required; | |
b3b94faa | 327 | int error = 0; |
18ec7d5c SW |
328 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from; |
329 | loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
330 | struct gfs2_alloc *al; | |
52ae7b79 RC |
331 | unsigned int write_len = to - from; |
332 | ||
b3b94faa | 333 | |
2ca99501 | 334 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh); |
dcd24799 | 335 | error = gfs2_glock_nq_atime(&ip->i_gh); |
2ca99501 SW |
336 | if (unlikely(error)) { |
337 | if (error == GLR_TRYFAILED) | |
338 | error = AOP_TRUNCATED_PAGE; | |
18ec7d5c | 339 | goto out_uninit; |
2ca99501 | 340 | } |
b3b94faa | 341 | |
52ae7b79 | 342 | gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks); |
18ec7d5c | 343 | |
52ae7b79 | 344 | error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required); |
18ec7d5c SW |
345 | if (error) |
346 | goto out_unlock; | |
b3b94faa | 347 | |
18ec7d5c | 348 | |
f5c54804 | 349 | ip->i_alloc.al_requested = 0; |
18ec7d5c SW |
350 | if (alloc_required) { |
351 | al = gfs2_alloc_get(ip); | |
352 | ||
353 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
354 | if (error) | |
355 | goto out_alloc_put; | |
356 | ||
2933f925 | 357 | error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid); |
18ec7d5c SW |
358 | if (error) |
359 | goto out_qunlock; | |
360 | ||
361 | al->al_requested = data_blocks + ind_blocks; | |
362 | error = gfs2_inplace_reserve(ip); | |
363 | if (error) | |
364 | goto out_qunlock; | |
365 | } | |
366 | ||
367 | rblocks = RES_DINODE + ind_blocks; | |
368 | if (gfs2_is_jdata(ip)) | |
369 | rblocks += data_blocks ? data_blocks : 1; | |
370 | if (ind_blocks || data_blocks) | |
371 | rblocks += RES_STATFS + RES_QUOTA; | |
372 | ||
373 | error = gfs2_trans_begin(sdp, rblocks, 0); | |
374 | if (error) | |
375 | goto out; | |
376 | ||
377 | if (gfs2_is_stuffed(ip)) { | |
378 | if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { | |
f25ef0c1 | 379 | error = gfs2_unstuff_dinode(ip, page); |
5c4e9e03 SW |
380 | if (error == 0) |
381 | goto prepare_write; | |
382 | } else if (!PageUptodate(page)) | |
b3b94faa | 383 | error = stuffed_readpage(ip, page); |
5c4e9e03 | 384 | goto out; |
18ec7d5c SW |
385 | } |
386 | ||
5c4e9e03 | 387 | prepare_write: |
18ec7d5c SW |
388 | error = block_prepare_write(page, from, to, gfs2_get_block); |
389 | ||
390 | out: | |
391 | if (error) { | |
392 | gfs2_trans_end(sdp); | |
393 | if (alloc_required) { | |
394 | gfs2_inplace_release(ip); | |
395 | out_qunlock: | |
396 | gfs2_quota_unlock(ip); | |
397 | out_alloc_put: | |
398 | gfs2_alloc_put(ip); | |
399 | } | |
400 | out_unlock: | |
401 | gfs2_glock_dq_m(1, &ip->i_gh); | |
402 | out_uninit: | |
403 | gfs2_holder_uninit(&ip->i_gh); | |
404 | } | |
b3b94faa DT |
405 | |
406 | return error; | |
407 | } | |
408 | ||
409 | /** | |
410 | * gfs2_commit_write - Commit write to a file | |
411 | * @file: The file to write to | |
412 | * @page: The page containing the data | |
413 | * @from: From (byte range within page) | |
414 | * @to: To (byte range within page) | |
415 | * | |
416 | * Returns: errno | |
417 | */ | |
418 | ||
419 | static int gfs2_commit_write(struct file *file, struct page *page, | |
420 | unsigned from, unsigned to) | |
421 | { | |
422 | struct inode *inode = page->mapping->host; | |
feaa7bba SW |
423 | struct gfs2_inode *ip = GFS2_I(inode); |
424 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
425 | int error = -EOPNOTSUPP; |
426 | struct buffer_head *dibh; | |
48516ced SW |
427 | struct gfs2_alloc *al = &ip->i_alloc; |
428 | struct gfs2_dinode *di; | |
b3b94faa | 429 | |
18ec7d5c SW |
430 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl))) |
431 | goto fail_nounlock; | |
432 | ||
433 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
434 | if (error) | |
435 | goto fail_endtrans; | |
436 | ||
437 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
48516ced | 438 | di = (struct gfs2_dinode *)dibh->b_data; |
18ec7d5c | 439 | |
b3b94faa | 440 | if (gfs2_is_stuffed(ip)) { |
cd915493 | 441 | u64 file_size; |
b3b94faa DT |
442 | void *kaddr; |
443 | ||
cd915493 | 444 | file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to; |
b3b94faa | 445 | |
18ec7d5c | 446 | kaddr = kmap_atomic(page, KM_USER0); |
b3b94faa | 447 | memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from, |
0bd5996a | 448 | kaddr + from, to - from); |
c312c4fd | 449 | kunmap_atomic(kaddr, KM_USER0); |
b3b94faa DT |
450 | |
451 | SetPageUptodate(page); | |
452 | ||
ae619320 | 453 | if (inode->i_size < file_size) { |
b3b94faa | 454 | i_size_write(inode, file_size); |
ae619320 SW |
455 | mark_inode_dirty(inode); |
456 | } | |
b3b94faa | 457 | } else { |
568f4c96 SW |
458 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || |
459 | gfs2_is_jdata(ip)) | |
257f9b4e | 460 | gfs2_page_add_databufs(ip, page, from, to); |
b3b94faa DT |
461 | error = generic_commit_write(file, page, from, to); |
462 | if (error) | |
463 | goto fail; | |
464 | } | |
465 | ||
48516ced | 466 | if (ip->i_di.di_size < inode->i_size) { |
18ec7d5c | 467 | ip->i_di.di_size = inode->i_size; |
48516ced SW |
468 | di->di_size = cpu_to_be64(inode->i_size); |
469 | } | |
470 | ||
18ec7d5c SW |
471 | brelse(dibh); |
472 | gfs2_trans_end(sdp); | |
473 | if (al->al_requested) { | |
474 | gfs2_inplace_release(ip); | |
475 | gfs2_quota_unlock(ip); | |
476 | gfs2_alloc_put(ip); | |
477 | } | |
478 | gfs2_glock_dq_m(1, &ip->i_gh); | |
479 | gfs2_holder_uninit(&ip->i_gh); | |
b3b94faa DT |
480 | return 0; |
481 | ||
18ec7d5c SW |
482 | fail: |
483 | brelse(dibh); | |
484 | fail_endtrans: | |
485 | gfs2_trans_end(sdp); | |
486 | if (al->al_requested) { | |
487 | gfs2_inplace_release(ip); | |
488 | gfs2_quota_unlock(ip); | |
489 | gfs2_alloc_put(ip); | |
490 | } | |
491 | gfs2_glock_dq_m(1, &ip->i_gh); | |
492 | gfs2_holder_uninit(&ip->i_gh); | |
493 | fail_nounlock: | |
b3b94faa | 494 | ClearPageUptodate(page); |
b3b94faa DT |
495 | return error; |
496 | } | |
497 | ||
498 | /** | |
499 | * gfs2_bmap - Block map function | |
500 | * @mapping: Address space info | |
501 | * @lblock: The block to map | |
502 | * | |
503 | * Returns: The disk address for the block or 0 on hole or error | |
504 | */ | |
505 | ||
506 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
507 | { | |
feaa7bba | 508 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
509 | struct gfs2_holder i_gh; |
510 | sector_t dblock = 0; | |
511 | int error; | |
512 | ||
b3b94faa DT |
513 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
514 | if (error) | |
515 | return 0; | |
516 | ||
517 | if (!gfs2_is_stuffed(ip)) | |
4ff14670 | 518 | dblock = generic_block_bmap(mapping, lblock, gfs2_get_block); |
b3b94faa DT |
519 | |
520 | gfs2_glock_dq_uninit(&i_gh); | |
521 | ||
522 | return dblock; | |
523 | } | |
524 | ||
525 | static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh) | |
526 | { | |
64fb4eb7 | 527 | struct gfs2_bufdata *bd; |
b3b94faa DT |
528 | |
529 | gfs2_log_lock(sdp); | |
5c676f6d | 530 | bd = bh->b_private; |
64fb4eb7 SW |
531 | if (bd) { |
532 | bd->bd_bh = NULL; | |
5c676f6d | 533 | bh->b_private = NULL; |
15d00c0b SW |
534 | } |
535 | gfs2_log_unlock(sdp); | |
b3b94faa DT |
536 | |
537 | lock_buffer(bh); | |
538 | clear_buffer_dirty(bh); | |
539 | bh->b_bdev = NULL; | |
540 | clear_buffer_mapped(bh); | |
541 | clear_buffer_req(bh); | |
542 | clear_buffer_new(bh); | |
543 | clear_buffer_delay(bh); | |
544 | unlock_buffer(bh); | |
545 | } | |
546 | ||
8628de05 | 547 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) |
b3b94faa | 548 | { |
15d00c0b | 549 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
b3b94faa DT |
550 | struct buffer_head *head, *bh, *next; |
551 | unsigned int curr_off = 0; | |
b3b94faa DT |
552 | |
553 | BUG_ON(!PageLocked(page)); | |
554 | if (!page_has_buffers(page)) | |
8628de05 | 555 | return; |
b3b94faa DT |
556 | |
557 | bh = head = page_buffers(page); | |
558 | do { | |
559 | unsigned int next_off = curr_off + bh->b_size; | |
560 | next = bh->b_this_page; | |
561 | ||
562 | if (offset <= curr_off) | |
563 | discard_buffer(sdp, bh); | |
564 | ||
565 | curr_off = next_off; | |
566 | bh = next; | |
567 | } while (bh != head); | |
568 | ||
569 | if (!offset) | |
8628de05 | 570 | try_to_release_page(page, 0); |
b3b94faa | 571 | |
8628de05 | 572 | return; |
b3b94faa DT |
573 | } |
574 | ||
c7b33834 SW |
575 | /** |
576 | * gfs2_ok_for_dio - check that dio is valid on this file | |
577 | * @ip: The inode | |
578 | * @rw: READ or WRITE | |
579 | * @offset: The offset at which we are reading or writing | |
580 | * | |
581 | * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o) | |
582 | * 1 (to accept the i/o request) | |
583 | */ | |
584 | static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset) | |
585 | { | |
586 | /* | |
587 | * Should we return an error here? I can't see that O_DIRECT for | |
588 | * a journaled file makes any sense. For now we'll silently fall | |
589 | * back to buffered I/O, likewise we do the same for stuffed | |
590 | * files since they are (a) small and (b) unaligned. | |
591 | */ | |
592 | if (gfs2_is_jdata(ip)) | |
593 | return 0; | |
594 | ||
595 | if (gfs2_is_stuffed(ip)) | |
596 | return 0; | |
597 | ||
598 | if (offset > i_size_read(&ip->i_inode)) | |
599 | return 0; | |
600 | return 1; | |
601 | } | |
602 | ||
603 | ||
604 | ||
a9e5f4d0 SW |
605 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, |
606 | const struct iovec *iov, loff_t offset, | |
607 | unsigned long nr_segs) | |
d1665e41 SW |
608 | { |
609 | struct file *file = iocb->ki_filp; | |
610 | struct inode *inode = file->f_mapping->host; | |
feaa7bba | 611 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
612 | struct gfs2_holder gh; |
613 | int rv; | |
614 | ||
615 | /* | |
c7b33834 SW |
616 | * Deferred lock, even if its a write, since we do no allocation |
617 | * on this path. All we need change is atime, and this lock mode | |
618 | * ensures that other nodes have flushed their buffered read caches | |
619 | * (i.e. their page cache entries for this inode). We do not, | |
620 | * unfortunately have the option of only flushing a range like | |
621 | * the VFS does. | |
d1665e41 | 622 | */ |
c7b33834 | 623 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh); |
dcd24799 | 624 | rv = gfs2_glock_nq_atime(&gh); |
d1665e41 | 625 | if (rv) |
c7b33834 SW |
626 | return rv; |
627 | rv = gfs2_ok_for_dio(ip, rw, offset); | |
628 | if (rv != 1) | |
629 | goto out; /* dio not valid, fall back to buffered i/o */ | |
630 | ||
631 | rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev, | |
632 | iov, offset, nr_segs, | |
633 | gfs2_get_block_direct, NULL); | |
d1665e41 SW |
634 | out: |
635 | gfs2_glock_dq_m(1, &gh); | |
636 | gfs2_holder_uninit(&gh); | |
d1665e41 SW |
637 | return rv; |
638 | } | |
639 | ||
4340fe62 SW |
640 | /** |
641 | * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out. | |
642 | * @bh: the buffer we're stuck on | |
643 | * | |
644 | */ | |
645 | ||
646 | static void stuck_releasepage(struct buffer_head *bh) | |
647 | { | |
648 | struct inode *inode = bh->b_page->mapping->host; | |
649 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
650 | struct gfs2_bufdata *bd = bh->b_private; | |
651 | struct gfs2_glock *gl; | |
166afccd SW |
652 | static unsigned limit = 0; |
653 | ||
0bd5996a | 654 | if (limit > 3) |
166afccd | 655 | return; |
0bd5996a | 656 | limit++; |
4340fe62 SW |
657 | |
658 | fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode); | |
659 | fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n", | |
660 | (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count)); | |
661 | fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh)); | |
662 | fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL"); | |
663 | ||
664 | if (!bd) | |
665 | return; | |
666 | ||
667 | gl = bd->bd_gl; | |
668 | ||
907b9bce | 669 | fs_warn(sdp, "gl = (%u, %llu)\n", |
4340fe62 SW |
670 | gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number); |
671 | ||
672 | fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n", | |
673 | (list_empty(&bd->bd_list_tr)) ? "no" : "yes", | |
674 | (list_empty(&bd->bd_le.le_list)) ? "no" : "yes"); | |
675 | ||
676 | if (gl->gl_ops == &gfs2_inode_glops) { | |
677 | struct gfs2_inode *ip = gl->gl_object; | |
678 | unsigned int x; | |
679 | ||
680 | if (!ip) | |
681 | return; | |
682 | ||
683 | fs_warn(sdp, "ip = %llu %llu\n", | |
684 | (unsigned long long)ip->i_num.no_formal_ino, | |
685 | (unsigned long long)ip->i_num.no_addr); | |
686 | ||
687 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) | |
688 | fs_warn(sdp, "ip->i_cache[%u] = %s\n", | |
689 | x, (ip->i_cache[x]) ? "!NULL" : "NULL"); | |
690 | } | |
691 | } | |
692 | ||
693 | /** | |
623d9355 | 694 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
695 | * @page: the page that's being released |
696 | * @gfp_mask: passed from Linux VFS, ignored by us | |
697 | * | |
698 | * Call try_to_free_buffers() if the buffers in this page can be | |
699 | * released. | |
700 | * | |
701 | * Returns: 0 | |
702 | */ | |
703 | ||
704 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
705 | { | |
706 | struct inode *aspace = page->mapping->host; | |
707 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | |
708 | struct buffer_head *bh, *head; | |
709 | struct gfs2_bufdata *bd; | |
166afccd | 710 | unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ; |
4340fe62 SW |
711 | |
712 | if (!page_has_buffers(page)) | |
713 | goto out; | |
714 | ||
715 | head = bh = page_buffers(page); | |
716 | do { | |
4340fe62 | 717 | while (atomic_read(&bh->b_count)) { |
166afccd SW |
718 | if (!atomic_read(&aspace->i_writecount)) |
719 | return 0; | |
720 | ||
61057c6b RC |
721 | if (!(gfp_mask & __GFP_WAIT)) |
722 | return 0; | |
723 | ||
166afccd SW |
724 | if (time_after_eq(jiffies, t)) { |
725 | stuck_releasepage(bh); | |
726 | /* should we withdraw here? */ | |
727 | return 0; | |
4340fe62 SW |
728 | } |
729 | ||
166afccd | 730 | yield(); |
4340fe62 SW |
731 | } |
732 | ||
733 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); | |
623d9355 | 734 | gfs2_assert_warn(sdp, !buffer_dirty(bh)); |
4340fe62 | 735 | |
623d9355 | 736 | gfs2_log_lock(sdp); |
4340fe62 SW |
737 | bd = bh->b_private; |
738 | if (bd) { | |
739 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
740 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
4340fe62 | 741 | gfs2_assert_warn(sdp, !bd->bd_ail); |
623d9355 SW |
742 | bd->bd_bh = NULL; |
743 | if (!list_empty(&bd->bd_le.le_list)) | |
744 | bd = NULL; | |
4340fe62 SW |
745 | bh->b_private = NULL; |
746 | } | |
623d9355 SW |
747 | gfs2_log_unlock(sdp); |
748 | if (bd) | |
749 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
4340fe62 SW |
750 | |
751 | bh = bh->b_this_page; | |
166afccd | 752 | } while (bh != head); |
4340fe62 | 753 | |
a9e5f4d0 | 754 | out: |
4340fe62 SW |
755 | return try_to_free_buffers(page); |
756 | } | |
757 | ||
66de045d | 758 | const struct address_space_operations gfs2_file_aops = { |
b3b94faa DT |
759 | .writepage = gfs2_writepage, |
760 | .readpage = gfs2_readpage, | |
fd88de56 | 761 | .readpages = gfs2_readpages, |
b3b94faa DT |
762 | .sync_page = block_sync_page, |
763 | .prepare_write = gfs2_prepare_write, | |
764 | .commit_write = gfs2_commit_write, | |
765 | .bmap = gfs2_bmap, | |
766 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 767 | .releasepage = gfs2_releasepage, |
b3b94faa DT |
768 | .direct_IO = gfs2_direct_IO, |
769 | }; | |
770 |