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 | { |
7a6bbacb | 68 | return gfs2_block_map(inode, lblock, create, bh_result, 32); |
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 | ||
7a6bbacb | 86 | error = gfs2_block_map(inode, lblock, 0, bh_result, 1); |
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 | { |
7a6bbacb | 97 | return gfs2_block_map(inode, lblock, 0, bh_result, 512); |
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 | ||
fd88de56 SW |
159 | static int zero_readpage(struct page *page) |
160 | { | |
161 | void *kaddr; | |
162 | ||
163 | kaddr = kmap_atomic(page, KM_USER0); | |
164 | memset(kaddr, 0, PAGE_CACHE_SIZE); | |
165 | kunmap_atomic(page, KM_USER0); | |
166 | ||
167 | SetPageUptodate(page); | |
168 | ||
169 | return 0; | |
170 | } | |
171 | ||
b3b94faa DT |
172 | /** |
173 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
174 | * @ip: the inode | |
175 | * @page: the page | |
176 | * | |
177 | * Returns: errno | |
178 | */ | |
179 | ||
180 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) | |
181 | { | |
182 | struct buffer_head *dibh; | |
183 | void *kaddr; | |
184 | int error; | |
185 | ||
fd88de56 SW |
186 | /* Only the first page of a stuffed file might contain data */ |
187 | if (unlikely(page->index)) | |
188 | return zero_readpage(page); | |
189 | ||
b3b94faa DT |
190 | error = gfs2_meta_inode_buffer(ip, &dibh); |
191 | if (error) | |
192 | return error; | |
193 | ||
5c4e9e03 | 194 | kaddr = kmap_atomic(page, KM_USER0); |
fd88de56 | 195 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), |
b3b94faa | 196 | ip->i_di.di_size); |
fd88de56 | 197 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); |
5c4e9e03 | 198 | kunmap_atomic(page, KM_USER0); |
b3b94faa DT |
199 | |
200 | brelse(dibh); | |
201 | ||
202 | SetPageUptodate(page); | |
203 | ||
204 | return 0; | |
205 | } | |
206 | ||
b3b94faa | 207 | |
b3b94faa DT |
208 | /** |
209 | * gfs2_readpage - readpage with locking | |
18ec7d5c SW |
210 | * @file: The file to read a page for. N.B. This may be NULL if we are |
211 | * reading an internal file. | |
b3b94faa DT |
212 | * @page: The page to read |
213 | * | |
214 | * Returns: errno | |
215 | */ | |
216 | ||
217 | static int gfs2_readpage(struct file *file, struct page *page) | |
218 | { | |
feaa7bba SW |
219 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
220 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
dc41aeed | 221 | struct gfs2_file *gf = NULL; |
18ec7d5c | 222 | struct gfs2_holder gh; |
b3b94faa | 223 | int error; |
59a1cc6b | 224 | int do_unlock = 0; |
b3b94faa | 225 | |
dd538c83 | 226 | if (likely(file != &gfs2_internal_file_sentinel)) { |
59a1cc6b | 227 | if (file) { |
dc41aeed | 228 | gf = file->private_data; |
59a1cc6b | 229 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) |
dc41aeed | 230 | /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */ |
59a1cc6b SW |
231 | goto skip_lock; |
232 | } | |
fe1bdedc | 233 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh); |
59a1cc6b | 234 | do_unlock = 1; |
61a30dcb | 235 | error = gfs2_glock_nq_m_atime(1, &gh); |
fd88de56 | 236 | if (unlikely(error)) |
61a30dcb SW |
237 | goto out_unlock; |
238 | } | |
b3b94faa | 239 | |
59a1cc6b | 240 | skip_lock: |
18ec7d5c | 241 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
242 | error = stuffed_readpage(ip, page); |
243 | unlock_page(page); | |
b3b94faa | 244 | } else |
18ec7d5c | 245 | error = mpage_readpage(page, gfs2_get_block); |
b3b94faa DT |
246 | |
247 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
248 | error = -EIO; | |
249 | ||
07903c02 | 250 | if (do_unlock) { |
61a30dcb SW |
251 | gfs2_glock_dq_m(1, &gh); |
252 | gfs2_holder_uninit(&gh); | |
253 | } | |
18ec7d5c | 254 | out: |
b3b94faa | 255 | return error; |
18ec7d5c SW |
256 | out_unlock: |
257 | unlock_page(page); | |
59a1cc6b | 258 | if (do_unlock) |
fd88de56 SW |
259 | gfs2_holder_uninit(&gh); |
260 | goto out; | |
261 | } | |
262 | ||
fd88de56 SW |
263 | /** |
264 | * gfs2_readpages - Read a bunch of pages at once | |
265 | * | |
266 | * Some notes: | |
267 | * 1. This is only for readahead, so we can simply ignore any things | |
268 | * which are slightly inconvenient (such as locking conflicts between | |
269 | * the page lock and the glock) and return having done no I/O. Its | |
270 | * obviously not something we'd want to do on too regular a basis. | |
271 | * Any I/O we ignore at this time will be done via readpage later. | |
272 | * 2. We have to handle stuffed files here too. | |
273 | * 3. mpage_readpages() does most of the heavy lifting in the common case. | |
274 | * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places. | |
275 | * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as | |
276 | * well as read-ahead. | |
277 | */ | |
278 | static int gfs2_readpages(struct file *file, struct address_space *mapping, | |
279 | struct list_head *pages, unsigned nr_pages) | |
280 | { | |
281 | struct inode *inode = mapping->host; | |
feaa7bba SW |
282 | struct gfs2_inode *ip = GFS2_I(inode); |
283 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 SW |
284 | struct gfs2_holder gh; |
285 | unsigned page_idx; | |
286 | int ret; | |
59a1cc6b | 287 | int do_unlock = 0; |
fd88de56 | 288 | |
dd538c83 | 289 | if (likely(file != &gfs2_internal_file_sentinel)) { |
59a1cc6b SW |
290 | if (file) { |
291 | struct gfs2_file *gf = file->private_data; | |
292 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) | |
293 | goto skip_lock; | |
294 | } | |
fd88de56 SW |
295 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, |
296 | LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh); | |
59a1cc6b | 297 | do_unlock = 1; |
fd88de56 SW |
298 | ret = gfs2_glock_nq_m_atime(1, &gh); |
299 | if (ret == GLR_TRYFAILED) | |
300 | goto out_noerror; | |
301 | if (unlikely(ret)) | |
302 | goto out_unlock; | |
303 | } | |
59a1cc6b | 304 | skip_lock: |
fd88de56 SW |
305 | if (gfs2_is_stuffed(ip)) { |
306 | struct pagevec lru_pvec; | |
307 | pagevec_init(&lru_pvec, 0); | |
308 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
ffeb874b SW |
309 | struct page *page = list_entry(pages->prev, struct page, lru); |
310 | prefetchw(&page->flags); | |
fd88de56 SW |
311 | list_del(&page->lru); |
312 | if (!add_to_page_cache(page, mapping, | |
313 | page->index, GFP_KERNEL)) { | |
314 | ret = stuffed_readpage(ip, page); | |
315 | unlock_page(page); | |
316 | if (!pagevec_add(&lru_pvec, page)) | |
317 | __pagevec_lru_add(&lru_pvec); | |
ffeb874b SW |
318 | } else { |
319 | page_cache_release(page); | |
fd88de56 | 320 | } |
fd88de56 SW |
321 | } |
322 | pagevec_lru_add(&lru_pvec); | |
323 | ret = 0; | |
324 | } else { | |
325 | /* What we really want to do .... */ | |
326 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block); | |
327 | } | |
328 | ||
59a1cc6b | 329 | if (do_unlock) { |
fd88de56 SW |
330 | gfs2_glock_dq_m(1, &gh); |
331 | gfs2_holder_uninit(&gh); | |
332 | } | |
333 | out: | |
334 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
335 | ret = -EIO; | |
336 | return ret; | |
337 | out_noerror: | |
338 | ret = 0; | |
339 | out_unlock: | |
340 | /* unlock all pages, we can't do any I/O right now */ | |
341 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
ffeb874b | 342 | struct page *page = list_entry(pages->prev, struct page, lru); |
fd88de56 SW |
343 | list_del(&page->lru); |
344 | unlock_page(page); | |
345 | page_cache_release(page); | |
346 | } | |
59a1cc6b | 347 | if (do_unlock) |
fd88de56 | 348 | gfs2_holder_uninit(&gh); |
18ec7d5c | 349 | goto out; |
b3b94faa DT |
350 | } |
351 | ||
352 | /** | |
353 | * gfs2_prepare_write - Prepare to write a page to a file | |
354 | * @file: The file to write to | |
355 | * @page: The page which is to be prepared for writing | |
356 | * @from: From (byte range within page) | |
357 | * @to: To (byte range within page) | |
358 | * | |
359 | * Returns: errno | |
360 | */ | |
361 | ||
362 | static int gfs2_prepare_write(struct file *file, struct page *page, | |
363 | unsigned from, unsigned to) | |
364 | { | |
feaa7bba SW |
365 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
366 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
18ec7d5c SW |
367 | unsigned int data_blocks, ind_blocks, rblocks; |
368 | int alloc_required; | |
b3b94faa | 369 | int error = 0; |
18ec7d5c SW |
370 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from; |
371 | loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
372 | struct gfs2_alloc *al; | |
b3b94faa | 373 | |
fe1bdedc | 374 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh); |
18ec7d5c SW |
375 | error = gfs2_glock_nq_m_atime(1, &ip->i_gh); |
376 | if (error) | |
377 | goto out_uninit; | |
b3b94faa | 378 | |
18ec7d5c SW |
379 | gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks); |
380 | ||
381 | error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required); | |
382 | if (error) | |
383 | goto out_unlock; | |
b3b94faa | 384 | |
18ec7d5c SW |
385 | |
386 | if (alloc_required) { | |
387 | al = gfs2_alloc_get(ip); | |
388 | ||
389 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
390 | if (error) | |
391 | goto out_alloc_put; | |
392 | ||
393 | error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid); | |
394 | if (error) | |
395 | goto out_qunlock; | |
396 | ||
397 | al->al_requested = data_blocks + ind_blocks; | |
398 | error = gfs2_inplace_reserve(ip); | |
399 | if (error) | |
400 | goto out_qunlock; | |
401 | } | |
402 | ||
403 | rblocks = RES_DINODE + ind_blocks; | |
404 | if (gfs2_is_jdata(ip)) | |
405 | rblocks += data_blocks ? data_blocks : 1; | |
406 | if (ind_blocks || data_blocks) | |
407 | rblocks += RES_STATFS + RES_QUOTA; | |
408 | ||
409 | error = gfs2_trans_begin(sdp, rblocks, 0); | |
410 | if (error) | |
411 | goto out; | |
412 | ||
413 | if (gfs2_is_stuffed(ip)) { | |
414 | if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { | |
f25ef0c1 | 415 | error = gfs2_unstuff_dinode(ip, page); |
5c4e9e03 SW |
416 | if (error == 0) |
417 | goto prepare_write; | |
418 | } else if (!PageUptodate(page)) | |
b3b94faa | 419 | error = stuffed_readpage(ip, page); |
5c4e9e03 | 420 | goto out; |
18ec7d5c SW |
421 | } |
422 | ||
5c4e9e03 | 423 | prepare_write: |
18ec7d5c SW |
424 | error = block_prepare_write(page, from, to, gfs2_get_block); |
425 | ||
426 | out: | |
427 | if (error) { | |
428 | gfs2_trans_end(sdp); | |
429 | if (alloc_required) { | |
430 | gfs2_inplace_release(ip); | |
431 | out_qunlock: | |
432 | gfs2_quota_unlock(ip); | |
433 | out_alloc_put: | |
434 | gfs2_alloc_put(ip); | |
435 | } | |
436 | out_unlock: | |
437 | gfs2_glock_dq_m(1, &ip->i_gh); | |
438 | out_uninit: | |
439 | gfs2_holder_uninit(&ip->i_gh); | |
440 | } | |
b3b94faa DT |
441 | |
442 | return error; | |
443 | } | |
444 | ||
445 | /** | |
446 | * gfs2_commit_write - Commit write to a file | |
447 | * @file: The file to write to | |
448 | * @page: The page containing the data | |
449 | * @from: From (byte range within page) | |
450 | * @to: To (byte range within page) | |
451 | * | |
452 | * Returns: errno | |
453 | */ | |
454 | ||
455 | static int gfs2_commit_write(struct file *file, struct page *page, | |
456 | unsigned from, unsigned to) | |
457 | { | |
458 | struct inode *inode = page->mapping->host; | |
feaa7bba SW |
459 | struct gfs2_inode *ip = GFS2_I(inode); |
460 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
461 | int error = -EOPNOTSUPP; |
462 | struct buffer_head *dibh; | |
463 | struct gfs2_alloc *al = &ip->i_alloc;; | |
b3b94faa | 464 | |
18ec7d5c SW |
465 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl))) |
466 | goto fail_nounlock; | |
467 | ||
468 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
469 | if (error) | |
470 | goto fail_endtrans; | |
471 | ||
472 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
473 | ||
b3b94faa | 474 | if (gfs2_is_stuffed(ip)) { |
cd915493 | 475 | u64 file_size; |
b3b94faa DT |
476 | void *kaddr; |
477 | ||
cd915493 | 478 | file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to; |
b3b94faa | 479 | |
18ec7d5c | 480 | kaddr = kmap_atomic(page, KM_USER0); |
b3b94faa | 481 | memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from, |
0bd5996a | 482 | kaddr + from, to - from); |
18ec7d5c | 483 | kunmap_atomic(page, KM_USER0); |
b3b94faa DT |
484 | |
485 | SetPageUptodate(page); | |
486 | ||
487 | if (inode->i_size < file_size) | |
488 | i_size_write(inode, file_size); | |
489 | } else { | |
568f4c96 SW |
490 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || |
491 | gfs2_is_jdata(ip)) | |
257f9b4e | 492 | gfs2_page_add_databufs(ip, page, from, to); |
b3b94faa DT |
493 | error = generic_commit_write(file, page, from, to); |
494 | if (error) | |
495 | goto fail; | |
496 | } | |
497 | ||
18ec7d5c SW |
498 | if (ip->i_di.di_size < inode->i_size) |
499 | ip->i_di.di_size = inode->i_size; | |
500 | ||
501 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
502 | brelse(dibh); | |
503 | gfs2_trans_end(sdp); | |
504 | if (al->al_requested) { | |
505 | gfs2_inplace_release(ip); | |
506 | gfs2_quota_unlock(ip); | |
507 | gfs2_alloc_put(ip); | |
508 | } | |
509 | gfs2_glock_dq_m(1, &ip->i_gh); | |
510 | gfs2_holder_uninit(&ip->i_gh); | |
b3b94faa DT |
511 | return 0; |
512 | ||
18ec7d5c SW |
513 | fail: |
514 | brelse(dibh); | |
515 | fail_endtrans: | |
516 | gfs2_trans_end(sdp); | |
517 | if (al->al_requested) { | |
518 | gfs2_inplace_release(ip); | |
519 | gfs2_quota_unlock(ip); | |
520 | gfs2_alloc_put(ip); | |
521 | } | |
522 | gfs2_glock_dq_m(1, &ip->i_gh); | |
523 | gfs2_holder_uninit(&ip->i_gh); | |
524 | fail_nounlock: | |
b3b94faa | 525 | ClearPageUptodate(page); |
b3b94faa DT |
526 | return error; |
527 | } | |
528 | ||
529 | /** | |
530 | * gfs2_bmap - Block map function | |
531 | * @mapping: Address space info | |
532 | * @lblock: The block to map | |
533 | * | |
534 | * Returns: The disk address for the block or 0 on hole or error | |
535 | */ | |
536 | ||
537 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
538 | { | |
feaa7bba | 539 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
540 | struct gfs2_holder i_gh; |
541 | sector_t dblock = 0; | |
542 | int error; | |
543 | ||
b3b94faa DT |
544 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
545 | if (error) | |
546 | return 0; | |
547 | ||
548 | if (!gfs2_is_stuffed(ip)) | |
4ff14670 | 549 | dblock = generic_block_bmap(mapping, lblock, gfs2_get_block); |
b3b94faa DT |
550 | |
551 | gfs2_glock_dq_uninit(&i_gh); | |
552 | ||
553 | return dblock; | |
554 | } | |
555 | ||
556 | static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh) | |
557 | { | |
64fb4eb7 | 558 | struct gfs2_bufdata *bd; |
b3b94faa DT |
559 | |
560 | gfs2_log_lock(sdp); | |
5c676f6d | 561 | bd = bh->b_private; |
64fb4eb7 SW |
562 | if (bd) { |
563 | bd->bd_bh = NULL; | |
5c676f6d | 564 | bh->b_private = NULL; |
15d00c0b SW |
565 | } |
566 | gfs2_log_unlock(sdp); | |
b3b94faa DT |
567 | |
568 | lock_buffer(bh); | |
569 | clear_buffer_dirty(bh); | |
570 | bh->b_bdev = NULL; | |
571 | clear_buffer_mapped(bh); | |
572 | clear_buffer_req(bh); | |
573 | clear_buffer_new(bh); | |
574 | clear_buffer_delay(bh); | |
575 | unlock_buffer(bh); | |
576 | } | |
577 | ||
8628de05 | 578 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) |
b3b94faa | 579 | { |
15d00c0b | 580 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
b3b94faa DT |
581 | struct buffer_head *head, *bh, *next; |
582 | unsigned int curr_off = 0; | |
b3b94faa DT |
583 | |
584 | BUG_ON(!PageLocked(page)); | |
585 | if (!page_has_buffers(page)) | |
8628de05 | 586 | return; |
b3b94faa DT |
587 | |
588 | bh = head = page_buffers(page); | |
589 | do { | |
590 | unsigned int next_off = curr_off + bh->b_size; | |
591 | next = bh->b_this_page; | |
592 | ||
593 | if (offset <= curr_off) | |
594 | discard_buffer(sdp, bh); | |
595 | ||
596 | curr_off = next_off; | |
597 | bh = next; | |
598 | } while (bh != head); | |
599 | ||
600 | if (!offset) | |
8628de05 | 601 | try_to_release_page(page, 0); |
b3b94faa | 602 | |
8628de05 | 603 | return; |
b3b94faa DT |
604 | } |
605 | ||
a9e5f4d0 SW |
606 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, |
607 | const struct iovec *iov, loff_t offset, | |
608 | unsigned long nr_segs) | |
d1665e41 SW |
609 | { |
610 | struct file *file = iocb->ki_filp; | |
611 | struct inode *inode = file->f_mapping->host; | |
feaa7bba | 612 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
613 | struct gfs2_holder gh; |
614 | int rv; | |
615 | ||
a9e5f4d0 SW |
616 | if (rw == READ) |
617 | mutex_lock(&inode->i_mutex); | |
d1665e41 | 618 | /* |
a9e5f4d0 | 619 | * Shared lock, even if its a write, since we do no allocation |
d1665e41 SW |
620 | * on this path. All we need change is atime. |
621 | */ | |
622 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh); | |
623 | rv = gfs2_glock_nq_m_atime(1, &gh); | |
624 | if (rv) | |
625 | goto out; | |
626 | ||
a9e5f4d0 SW |
627 | if (offset > i_size_read(inode)) |
628 | goto out; | |
629 | ||
d1665e41 SW |
630 | /* |
631 | * Should we return an error here? I can't see that O_DIRECT for | |
632 | * a journaled file makes any sense. For now we'll silently fall | |
633 | * back to buffered I/O, likewise we do the same for stuffed | |
634 | * files since they are (a) small and (b) unaligned. | |
635 | */ | |
636 | if (gfs2_is_jdata(ip)) | |
637 | goto out; | |
638 | ||
639 | if (gfs2_is_stuffed(ip)) | |
640 | goto out; | |
641 | ||
a9e5f4d0 SW |
642 | rv = blockdev_direct_IO_own_locking(rw, iocb, inode, |
643 | inode->i_sb->s_bdev, | |
644 | iov, offset, nr_segs, | |
7a6bbacb | 645 | gfs2_get_block_direct, NULL); |
d1665e41 SW |
646 | out: |
647 | gfs2_glock_dq_m(1, &gh); | |
648 | gfs2_holder_uninit(&gh); | |
a9e5f4d0 SW |
649 | if (rw == READ) |
650 | mutex_unlock(&inode->i_mutex); | |
d1665e41 SW |
651 | |
652 | return rv; | |
653 | } | |
654 | ||
4340fe62 SW |
655 | /** |
656 | * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out. | |
657 | * @bh: the buffer we're stuck on | |
658 | * | |
659 | */ | |
660 | ||
661 | static void stuck_releasepage(struct buffer_head *bh) | |
662 | { | |
663 | struct inode *inode = bh->b_page->mapping->host; | |
664 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
665 | struct gfs2_bufdata *bd = bh->b_private; | |
666 | struct gfs2_glock *gl; | |
166afccd SW |
667 | static unsigned limit = 0; |
668 | ||
0bd5996a | 669 | if (limit > 3) |
166afccd | 670 | return; |
0bd5996a | 671 | limit++; |
4340fe62 SW |
672 | |
673 | fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode); | |
674 | fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n", | |
675 | (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count)); | |
676 | fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh)); | |
677 | fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL"); | |
678 | ||
679 | if (!bd) | |
680 | return; | |
681 | ||
682 | gl = bd->bd_gl; | |
683 | ||
684 | fs_warn(sdp, "gl = (%u, %llu)\n", | |
685 | gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number); | |
686 | ||
687 | fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n", | |
688 | (list_empty(&bd->bd_list_tr)) ? "no" : "yes", | |
689 | (list_empty(&bd->bd_le.le_list)) ? "no" : "yes"); | |
690 | ||
691 | if (gl->gl_ops == &gfs2_inode_glops) { | |
692 | struct gfs2_inode *ip = gl->gl_object; | |
693 | unsigned int x; | |
694 | ||
695 | if (!ip) | |
696 | return; | |
697 | ||
698 | fs_warn(sdp, "ip = %llu %llu\n", | |
699 | (unsigned long long)ip->i_num.no_formal_ino, | |
700 | (unsigned long long)ip->i_num.no_addr); | |
701 | ||
702 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) | |
703 | fs_warn(sdp, "ip->i_cache[%u] = %s\n", | |
704 | x, (ip->i_cache[x]) ? "!NULL" : "NULL"); | |
705 | } | |
706 | } | |
707 | ||
708 | /** | |
623d9355 | 709 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
710 | * @page: the page that's being released |
711 | * @gfp_mask: passed from Linux VFS, ignored by us | |
712 | * | |
713 | * Call try_to_free_buffers() if the buffers in this page can be | |
714 | * released. | |
715 | * | |
716 | * Returns: 0 | |
717 | */ | |
718 | ||
719 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
720 | { | |
721 | struct inode *aspace = page->mapping->host; | |
722 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | |
723 | struct buffer_head *bh, *head; | |
724 | struct gfs2_bufdata *bd; | |
166afccd | 725 | unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ; |
4340fe62 SW |
726 | |
727 | if (!page_has_buffers(page)) | |
728 | goto out; | |
729 | ||
730 | head = bh = page_buffers(page); | |
731 | do { | |
4340fe62 | 732 | while (atomic_read(&bh->b_count)) { |
166afccd SW |
733 | if (!atomic_read(&aspace->i_writecount)) |
734 | return 0; | |
735 | ||
736 | if (time_after_eq(jiffies, t)) { | |
737 | stuck_releasepage(bh); | |
738 | /* should we withdraw here? */ | |
739 | return 0; | |
4340fe62 SW |
740 | } |
741 | ||
166afccd | 742 | yield(); |
4340fe62 SW |
743 | } |
744 | ||
745 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); | |
623d9355 | 746 | gfs2_assert_warn(sdp, !buffer_dirty(bh)); |
4340fe62 | 747 | |
623d9355 | 748 | gfs2_log_lock(sdp); |
4340fe62 SW |
749 | bd = bh->b_private; |
750 | if (bd) { | |
751 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
752 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
4340fe62 | 753 | gfs2_assert_warn(sdp, !bd->bd_ail); |
623d9355 SW |
754 | bd->bd_bh = NULL; |
755 | if (!list_empty(&bd->bd_le.le_list)) | |
756 | bd = NULL; | |
4340fe62 SW |
757 | bh->b_private = NULL; |
758 | } | |
623d9355 SW |
759 | gfs2_log_unlock(sdp); |
760 | if (bd) | |
761 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
4340fe62 SW |
762 | |
763 | bh = bh->b_this_page; | |
166afccd | 764 | } while (bh != head); |
4340fe62 | 765 | |
a9e5f4d0 | 766 | out: |
4340fe62 SW |
767 | return try_to_free_buffers(page); |
768 | } | |
769 | ||
66de045d | 770 | const struct address_space_operations gfs2_file_aops = { |
b3b94faa DT |
771 | .writepage = gfs2_writepage, |
772 | .readpage = gfs2_readpage, | |
fd88de56 | 773 | .readpages = gfs2_readpages, |
b3b94faa DT |
774 | .sync_page = block_sync_page, |
775 | .prepare_write = gfs2_prepare_write, | |
776 | .commit_write = gfs2_commit_write, | |
777 | .bmap = gfs2_bmap, | |
778 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 779 | .releasepage = gfs2_releasepage, |
b3b94faa DT |
780 | .direct_IO = gfs2_direct_IO, |
781 | }; | |
782 |