]>
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> |
a8d638e3 | 19 | #include <linux/writeback.h> |
5c676f6d | 20 | #include <linux/gfs2_ondisk.h> |
7d308590 | 21 | #include <linux/lm_interface.h> |
b3b94faa DT |
22 | |
23 | #include "gfs2.h" | |
5c676f6d | 24 | #include "incore.h" |
b3b94faa DT |
25 | #include "bmap.h" |
26 | #include "glock.h" | |
27 | #include "inode.h" | |
b3b94faa DT |
28 | #include "log.h" |
29 | #include "meta_io.h" | |
30 | #include "ops_address.h" | |
b3b94faa DT |
31 | #include "quota.h" |
32 | #include "trans.h" | |
18ec7d5c | 33 | #include "rgrp.h" |
61a30dcb | 34 | #include "ops_file.h" |
5c676f6d | 35 | #include "util.h" |
4340fe62 | 36 | #include "glops.h" |
b3b94faa | 37 | |
ba7f7290 SW |
38 | |
39 | static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page, | |
40 | unsigned int from, unsigned int to) | |
41 | { | |
42 | struct buffer_head *head = page_buffers(page); | |
43 | unsigned int bsize = head->b_size; | |
44 | struct buffer_head *bh; | |
45 | unsigned int start, end; | |
46 | ||
47 | for (bh = head, start = 0; bh != head || !start; | |
48 | bh = bh->b_this_page, start = end) { | |
49 | end = start + bsize; | |
50 | if (end <= from || start >= to) | |
51 | continue; | |
52 | gfs2_trans_add_bh(ip->i_gl, bh, 0); | |
53 | } | |
54 | } | |
55 | ||
b3b94faa | 56 | /** |
4ff14670 | 57 | * gfs2_get_block - Fills in a buffer head with details about a block |
b3b94faa DT |
58 | * @inode: The inode |
59 | * @lblock: The block number to look up | |
60 | * @bh_result: The buffer head to return the result in | |
61 | * @create: Non-zero if we may add block to the file | |
62 | * | |
63 | * Returns: errno | |
64 | */ | |
65 | ||
4ff14670 SW |
66 | int gfs2_get_block(struct inode *inode, sector_t lblock, |
67 | struct buffer_head *bh_result, int create) | |
b3b94faa | 68 | { |
23591256 | 69 | return gfs2_block_map(inode, lblock, create, bh_result); |
b3b94faa DT |
70 | } |
71 | ||
72 | /** | |
7a6bbacb | 73 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block |
b3b94faa DT |
74 | * @inode: The inode |
75 | * @lblock: The block number to look up | |
76 | * @bh_result: The buffer head to return the result in | |
77 | * @create: Non-zero if we may add block to the file | |
78 | * | |
79 | * Returns: errno | |
80 | */ | |
81 | ||
7a6bbacb SW |
82 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, |
83 | struct buffer_head *bh_result, int create) | |
b3b94faa | 84 | { |
b3b94faa DT |
85 | int error; |
86 | ||
23591256 | 87 | error = gfs2_block_map(inode, lblock, 0, bh_result); |
b3b94faa DT |
88 | if (error) |
89 | return error; | |
7a6bbacb SW |
90 | if (bh_result->b_blocknr == 0) |
91 | return -EIO; | |
92 | return 0; | |
b3b94faa DT |
93 | } |
94 | ||
7a6bbacb SW |
95 | static int gfs2_get_block_direct(struct inode *inode, sector_t lblock, |
96 | struct buffer_head *bh_result, int create) | |
623d9355 | 97 | { |
23591256 | 98 | return gfs2_block_map(inode, lblock, 0, bh_result); |
623d9355 | 99 | } |
7a6bbacb | 100 | |
b3b94faa DT |
101 | /** |
102 | * gfs2_writepage - Write complete page | |
103 | * @page: Page to write | |
104 | * | |
105 | * Returns: errno | |
106 | * | |
18ec7d5c SW |
107 | * Some of this is copied from block_write_full_page() although we still |
108 | * call it to do most of the work. | |
b3b94faa DT |
109 | */ |
110 | ||
111 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) | |
112 | { | |
18ec7d5c | 113 | struct inode *inode = page->mapping->host; |
f4387149 SW |
114 | struct gfs2_inode *ip = GFS2_I(inode); |
115 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
116 | loff_t i_size = i_size_read(inode); |
117 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
118 | unsigned offset; | |
b3b94faa | 119 | int error; |
18ec7d5c | 120 | int done_trans = 0; |
b3b94faa | 121 | |
b3b94faa DT |
122 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) { |
123 | unlock_page(page); | |
124 | return -EIO; | |
125 | } | |
5c676f6d | 126 | if (current->journal_info) |
18ec7d5c SW |
127 | goto out_ignore; |
128 | ||
129 | /* Is the page fully outside i_size? (truncate in progress) */ | |
130 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
d2d7b8a2 | 131 | if (page->index > end_index || (page->index == end_index && !offset)) { |
18ec7d5c | 132 | page->mapping->a_ops->invalidatepage(page, 0); |
b3b94faa | 133 | unlock_page(page); |
18ec7d5c | 134 | return 0; /* don't care */ |
b3b94faa DT |
135 | } |
136 | ||
18ec7d5c SW |
137 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) { |
138 | error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0); | |
139 | if (error) | |
140 | goto out_ignore; | |
f4387149 SW |
141 | if (!page_has_buffers(page)) { |
142 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
143 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
144 | } | |
18ec7d5c SW |
145 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1); |
146 | done_trans = 1; | |
147 | } | |
7a6bbacb | 148 | error = block_write_full_page(page, gfs2_get_block_noalloc, wbc); |
18ec7d5c SW |
149 | if (done_trans) |
150 | gfs2_trans_end(sdp); | |
b3b94faa | 151 | gfs2_meta_cache_flush(ip); |
b3b94faa | 152 | return error; |
18ec7d5c SW |
153 | |
154 | out_ignore: | |
155 | redirty_page_for_writepage(wbc, page); | |
156 | unlock_page(page); | |
157 | return 0; | |
b3b94faa DT |
158 | } |
159 | ||
a8d638e3 SW |
160 | /** |
161 | * gfs2_writepages - Write a bunch of dirty pages back to disk | |
162 | * @mapping: The mapping to write | |
163 | * @wbc: Write-back control | |
164 | * | |
165 | * For journaled files and/or ordered writes this just falls back to the | |
166 | * kernel's default writepages path for now. We will probably want to change | |
167 | * that eventually (i.e. when we look at allocate on flush). | |
168 | * | |
169 | * For the data=writeback case though we can already ignore buffer heads | |
170 | * and write whole extents at once. This is a big reduction in the | |
171 | * number of I/O requests we send and the bmap calls we make in this case. | |
172 | */ | |
a2cf8222 AB |
173 | static int gfs2_writepages(struct address_space *mapping, |
174 | struct writeback_control *wbc) | |
a8d638e3 SW |
175 | { |
176 | struct inode *inode = mapping->host; | |
177 | struct gfs2_inode *ip = GFS2_I(inode); | |
178 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
179 | ||
180 | if (sdp->sd_args.ar_data == GFS2_DATA_WRITEBACK && !gfs2_is_jdata(ip)) | |
181 | return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc); | |
182 | ||
183 | return generic_writepages(mapping, wbc); | |
184 | } | |
185 | ||
b3b94faa DT |
186 | /** |
187 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
188 | * @ip: the inode | |
189 | * @page: the page | |
190 | * | |
191 | * Returns: errno | |
192 | */ | |
193 | ||
194 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) | |
195 | { | |
196 | struct buffer_head *dibh; | |
197 | void *kaddr; | |
198 | int error; | |
199 | ||
61057c6b | 200 | BUG_ON(page->index); |
fd88de56 | 201 | |
b3b94faa DT |
202 | error = gfs2_meta_inode_buffer(ip, &dibh); |
203 | if (error) | |
204 | return error; | |
205 | ||
5c4e9e03 | 206 | kaddr = kmap_atomic(page, KM_USER0); |
fd88de56 | 207 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), |
b3b94faa | 208 | ip->i_di.di_size); |
fd88de56 | 209 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); |
c312c4fd | 210 | kunmap_atomic(kaddr, KM_USER0); |
b3b94faa DT |
211 | |
212 | brelse(dibh); | |
213 | ||
214 | SetPageUptodate(page); | |
215 | ||
216 | return 0; | |
217 | } | |
218 | ||
b3b94faa | 219 | |
b3b94faa DT |
220 | /** |
221 | * gfs2_readpage - readpage with locking | |
18ec7d5c SW |
222 | * @file: The file to read a page for. N.B. This may be NULL if we are |
223 | * reading an internal file. | |
b3b94faa DT |
224 | * @page: The page to read |
225 | * | |
226 | * Returns: errno | |
227 | */ | |
228 | ||
229 | static int gfs2_readpage(struct file *file, struct page *page) | |
230 | { | |
feaa7bba SW |
231 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
232 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
dc41aeed | 233 | struct gfs2_file *gf = NULL; |
18ec7d5c | 234 | struct gfs2_holder gh; |
b3b94faa | 235 | int error; |
59a1cc6b | 236 | int do_unlock = 0; |
b3b94faa | 237 | |
dd538c83 | 238 | if (likely(file != &gfs2_internal_file_sentinel)) { |
59a1cc6b | 239 | if (file) { |
dc41aeed | 240 | gf = file->private_data; |
59a1cc6b | 241 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) |
dc41aeed | 242 | /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */ |
59a1cc6b SW |
243 | goto skip_lock; |
244 | } | |
2ca99501 | 245 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh); |
59a1cc6b | 246 | do_unlock = 1; |
dcd24799 | 247 | error = gfs2_glock_nq_atime(&gh); |
fd88de56 | 248 | if (unlikely(error)) |
61a30dcb SW |
249 | goto out_unlock; |
250 | } | |
b3b94faa | 251 | |
59a1cc6b | 252 | skip_lock: |
18ec7d5c | 253 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
254 | error = stuffed_readpage(ip, page); |
255 | unlock_page(page); | |
b3b94faa | 256 | } else |
18ec7d5c | 257 | error = mpage_readpage(page, gfs2_get_block); |
b3b94faa DT |
258 | |
259 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
260 | error = -EIO; | |
261 | ||
07903c02 | 262 | if (do_unlock) { |
61a30dcb SW |
263 | gfs2_glock_dq_m(1, &gh); |
264 | gfs2_holder_uninit(&gh); | |
265 | } | |
18ec7d5c | 266 | out: |
b3b94faa | 267 | return error; |
18ec7d5c | 268 | out_unlock: |
2ca99501 SW |
269 | if (error == GLR_TRYFAILED) |
270 | error = AOP_TRUNCATED_PAGE; | |
18ec7d5c | 271 | unlock_page(page); |
59a1cc6b | 272 | if (do_unlock) |
fd88de56 SW |
273 | gfs2_holder_uninit(&gh); |
274 | goto out; | |
275 | } | |
276 | ||
fd88de56 SW |
277 | /** |
278 | * gfs2_readpages - Read a bunch of pages at once | |
279 | * | |
280 | * Some notes: | |
281 | * 1. This is only for readahead, so we can simply ignore any things | |
282 | * which are slightly inconvenient (such as locking conflicts between | |
283 | * the page lock and the glock) and return having done no I/O. Its | |
284 | * obviously not something we'd want to do on too regular a basis. | |
285 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 286 | * 2. We don't handle stuffed files here we let readpage do the honours. |
fd88de56 SW |
287 | * 3. mpage_readpages() does most of the heavy lifting in the common case. |
288 | * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places. | |
289 | * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as | |
290 | * well as read-ahead. | |
291 | */ | |
292 | static int gfs2_readpages(struct file *file, struct address_space *mapping, | |
293 | struct list_head *pages, unsigned nr_pages) | |
294 | { | |
295 | struct inode *inode = mapping->host; | |
feaa7bba SW |
296 | struct gfs2_inode *ip = GFS2_I(inode); |
297 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 | 298 | struct gfs2_holder gh; |
e1d5b18a | 299 | int ret = 0; |
59a1cc6b | 300 | int do_unlock = 0; |
fd88de56 | 301 | |
dd538c83 | 302 | if (likely(file != &gfs2_internal_file_sentinel)) { |
59a1cc6b SW |
303 | if (file) { |
304 | struct gfs2_file *gf = file->private_data; | |
305 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) | |
306 | goto skip_lock; | |
307 | } | |
fd88de56 | 308 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, |
2ca99501 | 309 | LM_FLAG_TRY_1CB|GL_ATIME, &gh); |
59a1cc6b | 310 | do_unlock = 1; |
dcd24799 | 311 | ret = gfs2_glock_nq_atime(&gh); |
907b9bce | 312 | if (ret == GLR_TRYFAILED) |
fd88de56 SW |
313 | goto out_noerror; |
314 | if (unlikely(ret)) | |
315 | goto out_unlock; | |
316 | } | |
59a1cc6b | 317 | skip_lock: |
e1d5b18a | 318 | if (!gfs2_is_stuffed(ip)) |
fd88de56 | 319 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block); |
fd88de56 | 320 | |
59a1cc6b | 321 | if (do_unlock) { |
fd88de56 SW |
322 | gfs2_glock_dq_m(1, &gh); |
323 | gfs2_holder_uninit(&gh); | |
324 | } | |
325 | out: | |
326 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
327 | ret = -EIO; | |
328 | return ret; | |
329 | out_noerror: | |
330 | ret = 0; | |
331 | out_unlock: | |
59a1cc6b | 332 | if (do_unlock) |
fd88de56 | 333 | gfs2_holder_uninit(&gh); |
18ec7d5c | 334 | goto out; |
b3b94faa DT |
335 | } |
336 | ||
337 | /** | |
338 | * gfs2_prepare_write - Prepare to write a page to a file | |
339 | * @file: The file to write to | |
340 | * @page: The page which is to be prepared for writing | |
341 | * @from: From (byte range within page) | |
342 | * @to: To (byte range within page) | |
343 | * | |
344 | * Returns: errno | |
345 | */ | |
346 | ||
347 | static int gfs2_prepare_write(struct file *file, struct page *page, | |
348 | unsigned from, unsigned to) | |
349 | { | |
feaa7bba SW |
350 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
351 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
18ec7d5c SW |
352 | unsigned int data_blocks, ind_blocks, rblocks; |
353 | int alloc_required; | |
b3b94faa | 354 | int error = 0; |
18ec7d5c SW |
355 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from; |
356 | loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
357 | struct gfs2_alloc *al; | |
52ae7b79 RC |
358 | unsigned int write_len = to - from; |
359 | ||
b3b94faa | 360 | |
2ca99501 | 361 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh); |
dcd24799 | 362 | error = gfs2_glock_nq_atime(&ip->i_gh); |
2ca99501 | 363 | if (unlikely(error)) { |
2d72e710 SW |
364 | if (error == GLR_TRYFAILED) { |
365 | unlock_page(page); | |
2ca99501 | 366 | error = AOP_TRUNCATED_PAGE; |
2d72e710 | 367 | } |
18ec7d5c | 368 | goto out_uninit; |
2ca99501 | 369 | } |
b3b94faa | 370 | |
52ae7b79 | 371 | gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks); |
18ec7d5c | 372 | |
52ae7b79 | 373 | error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required); |
18ec7d5c SW |
374 | if (error) |
375 | goto out_unlock; | |
b3b94faa | 376 | |
18ec7d5c | 377 | |
f5c54804 | 378 | ip->i_alloc.al_requested = 0; |
18ec7d5c SW |
379 | if (alloc_required) { |
380 | al = gfs2_alloc_get(ip); | |
381 | ||
382 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
383 | if (error) | |
384 | goto out_alloc_put; | |
385 | ||
2933f925 | 386 | error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid); |
18ec7d5c SW |
387 | if (error) |
388 | goto out_qunlock; | |
389 | ||
390 | al->al_requested = data_blocks + ind_blocks; | |
391 | error = gfs2_inplace_reserve(ip); | |
392 | if (error) | |
393 | goto out_qunlock; | |
394 | } | |
395 | ||
396 | rblocks = RES_DINODE + ind_blocks; | |
397 | if (gfs2_is_jdata(ip)) | |
398 | rblocks += data_blocks ? data_blocks : 1; | |
399 | if (ind_blocks || data_blocks) | |
400 | rblocks += RES_STATFS + RES_QUOTA; | |
401 | ||
402 | error = gfs2_trans_begin(sdp, rblocks, 0); | |
403 | if (error) | |
404 | goto out; | |
405 | ||
406 | if (gfs2_is_stuffed(ip)) { | |
407 | if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { | |
f25ef0c1 | 408 | error = gfs2_unstuff_dinode(ip, page); |
5c4e9e03 SW |
409 | if (error == 0) |
410 | goto prepare_write; | |
411 | } else if (!PageUptodate(page)) | |
b3b94faa | 412 | error = stuffed_readpage(ip, page); |
5c4e9e03 | 413 | goto out; |
18ec7d5c SW |
414 | } |
415 | ||
5c4e9e03 | 416 | prepare_write: |
18ec7d5c SW |
417 | error = block_prepare_write(page, from, to, gfs2_get_block); |
418 | ||
419 | out: | |
420 | if (error) { | |
421 | gfs2_trans_end(sdp); | |
422 | if (alloc_required) { | |
423 | gfs2_inplace_release(ip); | |
424 | out_qunlock: | |
425 | gfs2_quota_unlock(ip); | |
426 | out_alloc_put: | |
427 | gfs2_alloc_put(ip); | |
428 | } | |
429 | out_unlock: | |
430 | gfs2_glock_dq_m(1, &ip->i_gh); | |
431 | out_uninit: | |
432 | gfs2_holder_uninit(&ip->i_gh); | |
433 | } | |
b3b94faa DT |
434 | |
435 | return error; | |
436 | } | |
437 | ||
438 | /** | |
439 | * gfs2_commit_write - Commit write to a file | |
440 | * @file: The file to write to | |
441 | * @page: The page containing the data | |
442 | * @from: From (byte range within page) | |
443 | * @to: To (byte range within page) | |
444 | * | |
445 | * Returns: errno | |
446 | */ | |
447 | ||
448 | static int gfs2_commit_write(struct file *file, struct page *page, | |
449 | unsigned from, unsigned to) | |
450 | { | |
451 | struct inode *inode = page->mapping->host; | |
feaa7bba SW |
452 | struct gfs2_inode *ip = GFS2_I(inode); |
453 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
454 | int error = -EOPNOTSUPP; |
455 | struct buffer_head *dibh; | |
48516ced SW |
456 | struct gfs2_alloc *al = &ip->i_alloc; |
457 | struct gfs2_dinode *di; | |
b3b94faa | 458 | |
18ec7d5c SW |
459 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl))) |
460 | goto fail_nounlock; | |
461 | ||
462 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
463 | if (error) | |
464 | goto fail_endtrans; | |
465 | ||
466 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
48516ced | 467 | di = (struct gfs2_dinode *)dibh->b_data; |
18ec7d5c | 468 | |
b3b94faa | 469 | if (gfs2_is_stuffed(ip)) { |
cd915493 | 470 | u64 file_size; |
b3b94faa DT |
471 | void *kaddr; |
472 | ||
cd915493 | 473 | file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to; |
b3b94faa | 474 | |
18ec7d5c | 475 | kaddr = kmap_atomic(page, KM_USER0); |
b3b94faa | 476 | memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from, |
0bd5996a | 477 | kaddr + from, to - from); |
c312c4fd | 478 | kunmap_atomic(kaddr, KM_USER0); |
b3b94faa DT |
479 | |
480 | SetPageUptodate(page); | |
481 | ||
ae619320 | 482 | if (inode->i_size < file_size) { |
b3b94faa | 483 | i_size_write(inode, file_size); |
ae619320 SW |
484 | mark_inode_dirty(inode); |
485 | } | |
b3b94faa | 486 | } else { |
568f4c96 SW |
487 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || |
488 | gfs2_is_jdata(ip)) | |
257f9b4e | 489 | gfs2_page_add_databufs(ip, page, from, to); |
b3b94faa DT |
490 | error = generic_commit_write(file, page, from, to); |
491 | if (error) | |
492 | goto fail; | |
493 | } | |
494 | ||
48516ced | 495 | if (ip->i_di.di_size < inode->i_size) { |
18ec7d5c | 496 | ip->i_di.di_size = inode->i_size; |
48516ced SW |
497 | di->di_size = cpu_to_be64(inode->i_size); |
498 | } | |
499 | ||
18ec7d5c SW |
500 | brelse(dibh); |
501 | gfs2_trans_end(sdp); | |
502 | if (al->al_requested) { | |
503 | gfs2_inplace_release(ip); | |
504 | gfs2_quota_unlock(ip); | |
505 | gfs2_alloc_put(ip); | |
506 | } | |
507 | gfs2_glock_dq_m(1, &ip->i_gh); | |
508 | gfs2_holder_uninit(&ip->i_gh); | |
b3b94faa DT |
509 | return 0; |
510 | ||
18ec7d5c SW |
511 | fail: |
512 | brelse(dibh); | |
513 | fail_endtrans: | |
514 | gfs2_trans_end(sdp); | |
515 | if (al->al_requested) { | |
516 | gfs2_inplace_release(ip); | |
517 | gfs2_quota_unlock(ip); | |
518 | gfs2_alloc_put(ip); | |
519 | } | |
520 | gfs2_glock_dq_m(1, &ip->i_gh); | |
521 | gfs2_holder_uninit(&ip->i_gh); | |
522 | fail_nounlock: | |
b3b94faa | 523 | ClearPageUptodate(page); |
b3b94faa DT |
524 | return error; |
525 | } | |
526 | ||
527 | /** | |
528 | * gfs2_bmap - Block map function | |
529 | * @mapping: Address space info | |
530 | * @lblock: The block to map | |
531 | * | |
532 | * Returns: The disk address for the block or 0 on hole or error | |
533 | */ | |
534 | ||
535 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
536 | { | |
feaa7bba | 537 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
538 | struct gfs2_holder i_gh; |
539 | sector_t dblock = 0; | |
540 | int error; | |
541 | ||
b3b94faa DT |
542 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
543 | if (error) | |
544 | return 0; | |
545 | ||
546 | if (!gfs2_is_stuffed(ip)) | |
4ff14670 | 547 | dblock = generic_block_bmap(mapping, lblock, gfs2_get_block); |
b3b94faa DT |
548 | |
549 | gfs2_glock_dq_uninit(&i_gh); | |
550 | ||
551 | return dblock; | |
552 | } | |
553 | ||
554 | static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh) | |
555 | { | |
64fb4eb7 | 556 | struct gfs2_bufdata *bd; |
b3b94faa DT |
557 | |
558 | gfs2_log_lock(sdp); | |
5c676f6d | 559 | bd = bh->b_private; |
64fb4eb7 SW |
560 | if (bd) { |
561 | bd->bd_bh = NULL; | |
5c676f6d | 562 | bh->b_private = NULL; |
15d00c0b SW |
563 | } |
564 | gfs2_log_unlock(sdp); | |
b3b94faa DT |
565 | |
566 | lock_buffer(bh); | |
567 | clear_buffer_dirty(bh); | |
568 | bh->b_bdev = NULL; | |
569 | clear_buffer_mapped(bh); | |
570 | clear_buffer_req(bh); | |
571 | clear_buffer_new(bh); | |
572 | clear_buffer_delay(bh); | |
573 | unlock_buffer(bh); | |
574 | } | |
575 | ||
8628de05 | 576 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) |
b3b94faa | 577 | { |
15d00c0b | 578 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
b3b94faa DT |
579 | struct buffer_head *head, *bh, *next; |
580 | unsigned int curr_off = 0; | |
b3b94faa DT |
581 | |
582 | BUG_ON(!PageLocked(page)); | |
583 | if (!page_has_buffers(page)) | |
8628de05 | 584 | return; |
b3b94faa DT |
585 | |
586 | bh = head = page_buffers(page); | |
587 | do { | |
588 | unsigned int next_off = curr_off + bh->b_size; | |
589 | next = bh->b_this_page; | |
590 | ||
591 | if (offset <= curr_off) | |
592 | discard_buffer(sdp, bh); | |
593 | ||
594 | curr_off = next_off; | |
595 | bh = next; | |
596 | } while (bh != head); | |
597 | ||
598 | if (!offset) | |
8628de05 | 599 | try_to_release_page(page, 0); |
b3b94faa | 600 | |
8628de05 | 601 | return; |
b3b94faa DT |
602 | } |
603 | ||
c7b33834 SW |
604 | /** |
605 | * gfs2_ok_for_dio - check that dio is valid on this file | |
606 | * @ip: The inode | |
607 | * @rw: READ or WRITE | |
608 | * @offset: The offset at which we are reading or writing | |
609 | * | |
610 | * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o) | |
611 | * 1 (to accept the i/o request) | |
612 | */ | |
613 | static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset) | |
614 | { | |
615 | /* | |
616 | * Should we return an error here? I can't see that O_DIRECT for | |
617 | * a journaled file makes any sense. For now we'll silently fall | |
618 | * back to buffered I/O, likewise we do the same for stuffed | |
619 | * files since they are (a) small and (b) unaligned. | |
620 | */ | |
621 | if (gfs2_is_jdata(ip)) | |
622 | return 0; | |
623 | ||
624 | if (gfs2_is_stuffed(ip)) | |
625 | return 0; | |
626 | ||
627 | if (offset > i_size_read(&ip->i_inode)) | |
628 | return 0; | |
629 | return 1; | |
630 | } | |
631 | ||
632 | ||
633 | ||
a9e5f4d0 SW |
634 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, |
635 | const struct iovec *iov, loff_t offset, | |
636 | unsigned long nr_segs) | |
d1665e41 SW |
637 | { |
638 | struct file *file = iocb->ki_filp; | |
639 | struct inode *inode = file->f_mapping->host; | |
feaa7bba | 640 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
641 | struct gfs2_holder gh; |
642 | int rv; | |
643 | ||
644 | /* | |
c7b33834 SW |
645 | * Deferred lock, even if its a write, since we do no allocation |
646 | * on this path. All we need change is atime, and this lock mode | |
647 | * ensures that other nodes have flushed their buffered read caches | |
648 | * (i.e. their page cache entries for this inode). We do not, | |
649 | * unfortunately have the option of only flushing a range like | |
650 | * the VFS does. | |
d1665e41 | 651 | */ |
c7b33834 | 652 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh); |
dcd24799 | 653 | rv = gfs2_glock_nq_atime(&gh); |
d1665e41 | 654 | if (rv) |
c7b33834 SW |
655 | return rv; |
656 | rv = gfs2_ok_for_dio(ip, rw, offset); | |
657 | if (rv != 1) | |
658 | goto out; /* dio not valid, fall back to buffered i/o */ | |
659 | ||
660 | rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev, | |
661 | iov, offset, nr_segs, | |
662 | gfs2_get_block_direct, NULL); | |
d1665e41 SW |
663 | out: |
664 | gfs2_glock_dq_m(1, &gh); | |
665 | gfs2_holder_uninit(&gh); | |
d1665e41 SW |
666 | return rv; |
667 | } | |
668 | ||
4340fe62 SW |
669 | /** |
670 | * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out. | |
671 | * @bh: the buffer we're stuck on | |
672 | * | |
673 | */ | |
674 | ||
675 | static void stuck_releasepage(struct buffer_head *bh) | |
676 | { | |
677 | struct inode *inode = bh->b_page->mapping->host; | |
678 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
679 | struct gfs2_bufdata *bd = bh->b_private; | |
680 | struct gfs2_glock *gl; | |
166afccd SW |
681 | static unsigned limit = 0; |
682 | ||
0bd5996a | 683 | if (limit > 3) |
166afccd | 684 | return; |
0bd5996a | 685 | limit++; |
4340fe62 SW |
686 | |
687 | fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode); | |
688 | fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n", | |
689 | (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count)); | |
690 | fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh)); | |
691 | fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL"); | |
692 | ||
693 | if (!bd) | |
694 | return; | |
695 | ||
696 | gl = bd->bd_gl; | |
697 | ||
907b9bce | 698 | fs_warn(sdp, "gl = (%u, %llu)\n", |
4340fe62 SW |
699 | gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number); |
700 | ||
701 | fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n", | |
702 | (list_empty(&bd->bd_list_tr)) ? "no" : "yes", | |
703 | (list_empty(&bd->bd_le.le_list)) ? "no" : "yes"); | |
704 | ||
705 | if (gl->gl_ops == &gfs2_inode_glops) { | |
706 | struct gfs2_inode *ip = gl->gl_object; | |
707 | unsigned int x; | |
708 | ||
709 | if (!ip) | |
710 | return; | |
711 | ||
712 | fs_warn(sdp, "ip = %llu %llu\n", | |
713 | (unsigned long long)ip->i_num.no_formal_ino, | |
714 | (unsigned long long)ip->i_num.no_addr); | |
715 | ||
716 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) | |
717 | fs_warn(sdp, "ip->i_cache[%u] = %s\n", | |
718 | x, (ip->i_cache[x]) ? "!NULL" : "NULL"); | |
719 | } | |
720 | } | |
721 | ||
722 | /** | |
623d9355 | 723 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
724 | * @page: the page that's being released |
725 | * @gfp_mask: passed from Linux VFS, ignored by us | |
726 | * | |
727 | * Call try_to_free_buffers() if the buffers in this page can be | |
728 | * released. | |
729 | * | |
730 | * Returns: 0 | |
731 | */ | |
732 | ||
733 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
734 | { | |
735 | struct inode *aspace = page->mapping->host; | |
736 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | |
737 | struct buffer_head *bh, *head; | |
738 | struct gfs2_bufdata *bd; | |
166afccd | 739 | unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ; |
4340fe62 SW |
740 | |
741 | if (!page_has_buffers(page)) | |
742 | goto out; | |
743 | ||
744 | head = bh = page_buffers(page); | |
745 | do { | |
4340fe62 | 746 | while (atomic_read(&bh->b_count)) { |
166afccd SW |
747 | if (!atomic_read(&aspace->i_writecount)) |
748 | return 0; | |
749 | ||
61057c6b RC |
750 | if (!(gfp_mask & __GFP_WAIT)) |
751 | return 0; | |
752 | ||
166afccd SW |
753 | if (time_after_eq(jiffies, t)) { |
754 | stuck_releasepage(bh); | |
755 | /* should we withdraw here? */ | |
756 | return 0; | |
4340fe62 SW |
757 | } |
758 | ||
166afccd | 759 | yield(); |
4340fe62 SW |
760 | } |
761 | ||
762 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); | |
623d9355 | 763 | gfs2_assert_warn(sdp, !buffer_dirty(bh)); |
4340fe62 | 764 | |
623d9355 | 765 | gfs2_log_lock(sdp); |
4340fe62 SW |
766 | bd = bh->b_private; |
767 | if (bd) { | |
768 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
769 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
4340fe62 | 770 | gfs2_assert_warn(sdp, !bd->bd_ail); |
623d9355 SW |
771 | bd->bd_bh = NULL; |
772 | if (!list_empty(&bd->bd_le.le_list)) | |
773 | bd = NULL; | |
4340fe62 SW |
774 | bh->b_private = NULL; |
775 | } | |
623d9355 SW |
776 | gfs2_log_unlock(sdp); |
777 | if (bd) | |
778 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
4340fe62 SW |
779 | |
780 | bh = bh->b_this_page; | |
166afccd | 781 | } while (bh != head); |
4340fe62 | 782 | |
a9e5f4d0 | 783 | out: |
4340fe62 SW |
784 | return try_to_free_buffers(page); |
785 | } | |
786 | ||
66de045d | 787 | const struct address_space_operations gfs2_file_aops = { |
b3b94faa | 788 | .writepage = gfs2_writepage, |
a8d638e3 | 789 | .writepages = gfs2_writepages, |
b3b94faa | 790 | .readpage = gfs2_readpage, |
fd88de56 | 791 | .readpages = gfs2_readpages, |
b3b94faa DT |
792 | .sync_page = block_sync_page, |
793 | .prepare_write = gfs2_prepare_write, | |
794 | .commit_write = gfs2_commit_write, | |
795 | .bmap = gfs2_bmap, | |
796 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 797 | .releasepage = gfs2_releasepage, |
b3b94faa DT |
798 | .direct_IO = gfs2_direct_IO, |
799 | }; | |
800 |