diff options
Diffstat (limited to 'fs/gfs2/ops_address.c')
-rw-r--r-- | fs/gfs2/ops_address.c | 793 |
1 files changed, 793 insertions, 0 deletions
diff --git a/fs/gfs2/ops_address.c b/fs/gfs2/ops_address.c new file mode 100644 index 000000000000..bdd4d6b48721 --- /dev/null +++ b/fs/gfs2/ops_address.c | |||
@@ -0,0 +1,793 @@ | |||
1 | /* | ||
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | ||
3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. | ||
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 | ||
7 | * of the GNU General Public License v.2. | ||
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> | ||
16 | #include <linux/pagevec.h> | ||
17 | #include <linux/mpage.h> | ||
18 | #include <linux/fs.h> | ||
19 | #include <linux/gfs2_ondisk.h> | ||
20 | |||
21 | #include "gfs2.h" | ||
22 | #include "lm_interface.h" | ||
23 | #include "incore.h" | ||
24 | #include "bmap.h" | ||
25 | #include "glock.h" | ||
26 | #include "inode.h" | ||
27 | #include "log.h" | ||
28 | #include "meta_io.h" | ||
29 | #include "ops_address.h" | ||
30 | #include "quota.h" | ||
31 | #include "trans.h" | ||
32 | #include "rgrp.h" | ||
33 | #include "ops_file.h" | ||
34 | #include "util.h" | ||
35 | #include "glops.h" | ||
36 | |||
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 | |||
55 | /** | ||
56 | * gfs2_get_block - Fills in a buffer head with details about a block | ||
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 | |||
65 | int gfs2_get_block(struct inode *inode, sector_t lblock, | ||
66 | struct buffer_head *bh_result, int create) | ||
67 | { | ||
68 | int new = create; | ||
69 | uint64_t dblock; | ||
70 | int error; | ||
71 | int boundary; | ||
72 | |||
73 | error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary); | ||
74 | if (error) | ||
75 | return error; | ||
76 | |||
77 | if (!dblock) | ||
78 | return 0; | ||
79 | |||
80 | map_bh(bh_result, inode->i_sb, dblock); | ||
81 | if (new) | ||
82 | set_buffer_new(bh_result); | ||
83 | if (boundary) | ||
84 | set_buffer_boundary(bh_result); | ||
85 | |||
86 | return 0; | ||
87 | } | ||
88 | |||
89 | /** | ||
90 | * get_block_noalloc - Fills in a buffer head with details about a block | ||
91 | * @inode: The inode | ||
92 | * @lblock: The block number to look up | ||
93 | * @bh_result: The buffer head to return the result in | ||
94 | * @create: Non-zero if we may add block to the file | ||
95 | * | ||
96 | * Returns: errno | ||
97 | */ | ||
98 | |||
99 | static int get_block_noalloc(struct inode *inode, sector_t lblock, | ||
100 | struct buffer_head *bh_result, int create) | ||
101 | { | ||
102 | int new = 0; | ||
103 | uint64_t dblock; | ||
104 | int error; | ||
105 | int boundary; | ||
106 | |||
107 | error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary); | ||
108 | if (error) | ||
109 | return error; | ||
110 | |||
111 | if (dblock) | ||
112 | map_bh(bh_result, inode->i_sb, dblock); | ||
113 | else if (gfs2_assert_withdraw(GFS2_SB(inode), !create)) | ||
114 | error = -EIO; | ||
115 | if (boundary) | ||
116 | set_buffer_boundary(bh_result); | ||
117 | |||
118 | return error; | ||
119 | } | ||
120 | |||
121 | /** | ||
122 | * gfs2_writepage - Write complete page | ||
123 | * @page: Page to write | ||
124 | * | ||
125 | * Returns: errno | ||
126 | * | ||
127 | * Some of this is copied from block_write_full_page() although we still | ||
128 | * call it to do most of the work. | ||
129 | */ | ||
130 | |||
131 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) | ||
132 | { | ||
133 | struct inode *inode = page->mapping->host; | ||
134 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | ||
135 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | ||
136 | loff_t i_size = i_size_read(inode); | ||
137 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | ||
138 | unsigned offset; | ||
139 | int error; | ||
140 | int done_trans = 0; | ||
141 | |||
142 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) { | ||
143 | unlock_page(page); | ||
144 | return -EIO; | ||
145 | } | ||
146 | if (current->journal_info) | ||
147 | goto out_ignore; | ||
148 | |||
149 | /* Is the page fully outside i_size? (truncate in progress) */ | ||
150 | offset = i_size & (PAGE_CACHE_SIZE-1); | ||
151 | if (page->index > end_index || (page->index == end_index && !offset)) { | ||
152 | page->mapping->a_ops->invalidatepage(page, 0); | ||
153 | unlock_page(page); | ||
154 | return 0; /* don't care */ | ||
155 | } | ||
156 | |||
157 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) { | ||
158 | error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0); | ||
159 | if (error) | ||
160 | goto out_ignore; | ||
161 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1); | ||
162 | done_trans = 1; | ||
163 | } | ||
164 | error = block_write_full_page(page, get_block_noalloc, wbc); | ||
165 | if (done_trans) | ||
166 | gfs2_trans_end(sdp); | ||
167 | gfs2_meta_cache_flush(ip); | ||
168 | return error; | ||
169 | |||
170 | out_ignore: | ||
171 | redirty_page_for_writepage(wbc, page); | ||
172 | unlock_page(page); | ||
173 | return 0; | ||
174 | } | ||
175 | |||
176 | static int zero_readpage(struct page *page) | ||
177 | { | ||
178 | void *kaddr; | ||
179 | |||
180 | kaddr = kmap_atomic(page, KM_USER0); | ||
181 | memset(kaddr, 0, PAGE_CACHE_SIZE); | ||
182 | kunmap_atomic(page, KM_USER0); | ||
183 | |||
184 | SetPageUptodate(page); | ||
185 | |||
186 | return 0; | ||
187 | } | ||
188 | |||
189 | /** | ||
190 | * stuffed_readpage - Fill in a Linux page with stuffed file data | ||
191 | * @ip: the inode | ||
192 | * @page: the page | ||
193 | * | ||
194 | * Returns: errno | ||
195 | */ | ||
196 | |||
197 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) | ||
198 | { | ||
199 | struct buffer_head *dibh; | ||
200 | void *kaddr; | ||
201 | int error; | ||
202 | |||
203 | /* Only the first page of a stuffed file might contain data */ | ||
204 | if (unlikely(page->index)) | ||
205 | return zero_readpage(page); | ||
206 | |||
207 | error = gfs2_meta_inode_buffer(ip, &dibh); | ||
208 | if (error) | ||
209 | return error; | ||
210 | |||
211 | kaddr = kmap_atomic(page, KM_USER0); | ||
212 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), | ||
213 | ip->i_di.di_size); | ||
214 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); | ||
215 | kunmap_atomic(page, KM_USER0); | ||
216 | |||
217 | brelse(dibh); | ||
218 | |||
219 | SetPageUptodate(page); | ||
220 | |||
221 | return 0; | ||
222 | } | ||
223 | |||
224 | |||
225 | /** | ||
226 | * gfs2_readpage - readpage with locking | ||
227 | * @file: The file to read a page for. N.B. This may be NULL if we are | ||
228 | * reading an internal file. | ||
229 | * @page: The page to read | ||
230 | * | ||
231 | * Returns: errno | ||
232 | */ | ||
233 | |||
234 | static int gfs2_readpage(struct file *file, struct page *page) | ||
235 | { | ||
236 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | ||
237 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | ||
238 | struct gfs2_holder gh; | ||
239 | int error; | ||
240 | int do_unlock = 0; | ||
241 | |||
242 | if (likely(file != &gfs2_internal_file_sentinal)) { | ||
243 | if (file) { | ||
244 | struct gfs2_file *gf = file->private_data; | ||
245 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) | ||
246 | goto skip_lock; | ||
247 | } | ||
248 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh); | ||
249 | do_unlock = 1; | ||
250 | error = gfs2_glock_nq_m_atime(1, &gh); | ||
251 | if (unlikely(error)) | ||
252 | goto out_unlock; | ||
253 | } | ||
254 | |||
255 | skip_lock: | ||
256 | if (gfs2_is_stuffed(ip)) { | ||
257 | error = stuffed_readpage(ip, page); | ||
258 | unlock_page(page); | ||
259 | } else | ||
260 | error = mpage_readpage(page, gfs2_get_block); | ||
261 | |||
262 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | ||
263 | error = -EIO; | ||
264 | |||
265 | if (file != &gfs2_internal_file_sentinal) { | ||
266 | gfs2_glock_dq_m(1, &gh); | ||
267 | gfs2_holder_uninit(&gh); | ||
268 | } | ||
269 | out: | ||
270 | return error; | ||
271 | out_unlock: | ||
272 | unlock_page(page); | ||
273 | if (do_unlock) | ||
274 | gfs2_holder_uninit(&gh); | ||
275 | goto out; | ||
276 | } | ||
277 | |||
278 | /** | ||
279 | * gfs2_readpages - Read a bunch of pages at once | ||
280 | * | ||
281 | * Some notes: | ||
282 | * 1. This is only for readahead, so we can simply ignore any things | ||
283 | * which are slightly inconvenient (such as locking conflicts between | ||
284 | * the page lock and the glock) and return having done no I/O. Its | ||
285 | * obviously not something we'd want to do on too regular a basis. | ||
286 | * Any I/O we ignore at this time will be done via readpage later. | ||
287 | * 2. We have to handle stuffed files here too. | ||
288 | * 3. mpage_readpages() does most of the heavy lifting in the common case. | ||
289 | * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places. | ||
290 | * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as | ||
291 | * well as read-ahead. | ||
292 | */ | ||
293 | static int gfs2_readpages(struct file *file, struct address_space *mapping, | ||
294 | struct list_head *pages, unsigned nr_pages) | ||
295 | { | ||
296 | struct inode *inode = mapping->host; | ||
297 | struct gfs2_inode *ip = GFS2_I(inode); | ||
298 | struct gfs2_sbd *sdp = GFS2_SB(inode); | ||
299 | struct gfs2_holder gh; | ||
300 | unsigned page_idx; | ||
301 | int ret; | ||
302 | int do_unlock = 0; | ||
303 | |||
304 | if (likely(file != &gfs2_internal_file_sentinal)) { | ||
305 | if (file) { | ||
306 | struct gfs2_file *gf = file->private_data; | ||
307 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) | ||
308 | goto skip_lock; | ||
309 | } | ||
310 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, | ||
311 | LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh); | ||
312 | do_unlock = 1; | ||
313 | ret = gfs2_glock_nq_m_atime(1, &gh); | ||
314 | if (ret == GLR_TRYFAILED) | ||
315 | goto out_noerror; | ||
316 | if (unlikely(ret)) | ||
317 | goto out_unlock; | ||
318 | } | ||
319 | skip_lock: | ||
320 | if (gfs2_is_stuffed(ip)) { | ||
321 | struct pagevec lru_pvec; | ||
322 | pagevec_init(&lru_pvec, 0); | ||
323 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | ||
324 | struct page *page = list_entry(pages->prev, struct page, lru); | ||
325 | prefetchw(&page->flags); | ||
326 | list_del(&page->lru); | ||
327 | if (!add_to_page_cache(page, mapping, | ||
328 | page->index, GFP_KERNEL)) { | ||
329 | ret = stuffed_readpage(ip, page); | ||
330 | unlock_page(page); | ||
331 | if (!pagevec_add(&lru_pvec, page)) | ||
332 | __pagevec_lru_add(&lru_pvec); | ||
333 | } else { | ||
334 | page_cache_release(page); | ||
335 | } | ||
336 | } | ||
337 | pagevec_lru_add(&lru_pvec); | ||
338 | ret = 0; | ||
339 | } else { | ||
340 | /* What we really want to do .... */ | ||
341 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block); | ||
342 | } | ||
343 | |||
344 | if (do_unlock) { | ||
345 | gfs2_glock_dq_m(1, &gh); | ||
346 | gfs2_holder_uninit(&gh); | ||
347 | } | ||
348 | out: | ||
349 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | ||
350 | ret = -EIO; | ||
351 | return ret; | ||
352 | out_noerror: | ||
353 | ret = 0; | ||
354 | out_unlock: | ||
355 | /* unlock all pages, we can't do any I/O right now */ | ||
356 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | ||
357 | struct page *page = list_entry(pages->prev, struct page, lru); | ||
358 | list_del(&page->lru); | ||
359 | unlock_page(page); | ||
360 | page_cache_release(page); | ||
361 | } | ||
362 | if (do_unlock) | ||
363 | gfs2_holder_uninit(&gh); | ||
364 | goto out; | ||
365 | } | ||
366 | |||
367 | /** | ||
368 | * gfs2_prepare_write - Prepare to write a page to a file | ||
369 | * @file: The file to write to | ||
370 | * @page: The page which is to be prepared for writing | ||
371 | * @from: From (byte range within page) | ||
372 | * @to: To (byte range within page) | ||
373 | * | ||
374 | * Returns: errno | ||
375 | */ | ||
376 | |||
377 | static int gfs2_prepare_write(struct file *file, struct page *page, | ||
378 | unsigned from, unsigned to) | ||
379 | { | ||
380 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | ||
381 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | ||
382 | unsigned int data_blocks, ind_blocks, rblocks; | ||
383 | int alloc_required; | ||
384 | int error = 0; | ||
385 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from; | ||
386 | loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | ||
387 | struct gfs2_alloc *al; | ||
388 | |||
389 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh); | ||
390 | error = gfs2_glock_nq_m_atime(1, &ip->i_gh); | ||
391 | if (error) | ||
392 | goto out_uninit; | ||
393 | |||
394 | gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks); | ||
395 | |||
396 | error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required); | ||
397 | if (error) | ||
398 | goto out_unlock; | ||
399 | |||
400 | |||
401 | if (alloc_required) { | ||
402 | al = gfs2_alloc_get(ip); | ||
403 | |||
404 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | ||
405 | if (error) | ||
406 | goto out_alloc_put; | ||
407 | |||
408 | error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid); | ||
409 | if (error) | ||
410 | goto out_qunlock; | ||
411 | |||
412 | al->al_requested = data_blocks + ind_blocks; | ||
413 | error = gfs2_inplace_reserve(ip); | ||
414 | if (error) | ||
415 | goto out_qunlock; | ||
416 | } | ||
417 | |||
418 | rblocks = RES_DINODE + ind_blocks; | ||
419 | if (gfs2_is_jdata(ip)) | ||
420 | rblocks += data_blocks ? data_blocks : 1; | ||
421 | if (ind_blocks || data_blocks) | ||
422 | rblocks += RES_STATFS + RES_QUOTA; | ||
423 | |||
424 | error = gfs2_trans_begin(sdp, rblocks, 0); | ||
425 | if (error) | ||
426 | goto out; | ||
427 | |||
428 | if (gfs2_is_stuffed(ip)) { | ||
429 | if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { | ||
430 | error = gfs2_unstuff_dinode(ip, page); | ||
431 | if (error == 0) | ||
432 | goto prepare_write; | ||
433 | } else if (!PageUptodate(page)) | ||
434 | error = stuffed_readpage(ip, page); | ||
435 | goto out; | ||
436 | } | ||
437 | |||
438 | prepare_write: | ||
439 | error = block_prepare_write(page, from, to, gfs2_get_block); | ||
440 | |||
441 | out: | ||
442 | if (error) { | ||
443 | gfs2_trans_end(sdp); | ||
444 | if (alloc_required) { | ||
445 | gfs2_inplace_release(ip); | ||
446 | out_qunlock: | ||
447 | gfs2_quota_unlock(ip); | ||
448 | out_alloc_put: | ||
449 | gfs2_alloc_put(ip); | ||
450 | } | ||
451 | out_unlock: | ||
452 | gfs2_glock_dq_m(1, &ip->i_gh); | ||
453 | out_uninit: | ||
454 | gfs2_holder_uninit(&ip->i_gh); | ||
455 | } | ||
456 | |||
457 | return error; | ||
458 | } | ||
459 | |||
460 | /** | ||
461 | * gfs2_commit_write - Commit write to a file | ||
462 | * @file: The file to write to | ||
463 | * @page: The page containing the data | ||
464 | * @from: From (byte range within page) | ||
465 | * @to: To (byte range within page) | ||
466 | * | ||
467 | * Returns: errno | ||
468 | */ | ||
469 | |||
470 | static int gfs2_commit_write(struct file *file, struct page *page, | ||
471 | unsigned from, unsigned to) | ||
472 | { | ||
473 | struct inode *inode = page->mapping->host; | ||
474 | struct gfs2_inode *ip = GFS2_I(inode); | ||
475 | struct gfs2_sbd *sdp = GFS2_SB(inode); | ||
476 | int error = -EOPNOTSUPP; | ||
477 | struct buffer_head *dibh; | ||
478 | struct gfs2_alloc *al = &ip->i_alloc;; | ||
479 | |||
480 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl))) | ||
481 | goto fail_nounlock; | ||
482 | |||
483 | error = gfs2_meta_inode_buffer(ip, &dibh); | ||
484 | if (error) | ||
485 | goto fail_endtrans; | ||
486 | |||
487 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | ||
488 | |||
489 | if (gfs2_is_stuffed(ip)) { | ||
490 | uint64_t file_size; | ||
491 | void *kaddr; | ||
492 | |||
493 | file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to; | ||
494 | |||
495 | kaddr = kmap_atomic(page, KM_USER0); | ||
496 | memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from, | ||
497 | (char *)kaddr + from, to - from); | ||
498 | kunmap_atomic(page, KM_USER0); | ||
499 | |||
500 | SetPageUptodate(page); | ||
501 | |||
502 | if (inode->i_size < file_size) | ||
503 | i_size_write(inode, file_size); | ||
504 | } else { | ||
505 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || | ||
506 | gfs2_is_jdata(ip)) | ||
507 | gfs2_page_add_databufs(ip, page, from, to); | ||
508 | error = generic_commit_write(file, page, from, to); | ||
509 | if (error) | ||
510 | goto fail; | ||
511 | } | ||
512 | |||
513 | if (ip->i_di.di_size < inode->i_size) | ||
514 | ip->i_di.di_size = inode->i_size; | ||
515 | |||
516 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | ||
517 | brelse(dibh); | ||
518 | gfs2_trans_end(sdp); | ||
519 | if (al->al_requested) { | ||
520 | gfs2_inplace_release(ip); | ||
521 | gfs2_quota_unlock(ip); | ||
522 | gfs2_alloc_put(ip); | ||
523 | } | ||
524 | gfs2_glock_dq_m(1, &ip->i_gh); | ||
525 | gfs2_holder_uninit(&ip->i_gh); | ||
526 | return 0; | ||
527 | |||
528 | fail: | ||
529 | brelse(dibh); | ||
530 | fail_endtrans: | ||
531 | gfs2_trans_end(sdp); | ||
532 | if (al->al_requested) { | ||
533 | gfs2_inplace_release(ip); | ||
534 | gfs2_quota_unlock(ip); | ||
535 | gfs2_alloc_put(ip); | ||
536 | } | ||
537 | gfs2_glock_dq_m(1, &ip->i_gh); | ||
538 | gfs2_holder_uninit(&ip->i_gh); | ||
539 | fail_nounlock: | ||
540 | ClearPageUptodate(page); | ||
541 | return error; | ||
542 | } | ||
543 | |||
544 | /** | ||
545 | * gfs2_bmap - Block map function | ||
546 | * @mapping: Address space info | ||
547 | * @lblock: The block to map | ||
548 | * | ||
549 | * Returns: The disk address for the block or 0 on hole or error | ||
550 | */ | ||
551 | |||
552 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | ||
553 | { | ||
554 | struct gfs2_inode *ip = GFS2_I(mapping->host); | ||
555 | struct gfs2_holder i_gh; | ||
556 | sector_t dblock = 0; | ||
557 | int error; | ||
558 | |||
559 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); | ||
560 | if (error) | ||
561 | return 0; | ||
562 | |||
563 | if (!gfs2_is_stuffed(ip)) | ||
564 | dblock = generic_block_bmap(mapping, lblock, gfs2_get_block); | ||
565 | |||
566 | gfs2_glock_dq_uninit(&i_gh); | ||
567 | |||
568 | return dblock; | ||
569 | } | ||
570 | |||
571 | static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh) | ||
572 | { | ||
573 | struct gfs2_bufdata *bd; | ||
574 | |||
575 | gfs2_log_lock(sdp); | ||
576 | bd = bh->b_private; | ||
577 | if (bd) { | ||
578 | bd->bd_bh = NULL; | ||
579 | bh->b_private = NULL; | ||
580 | gfs2_log_unlock(sdp); | ||
581 | brelse(bh); | ||
582 | } else | ||
583 | gfs2_log_unlock(sdp); | ||
584 | |||
585 | lock_buffer(bh); | ||
586 | clear_buffer_dirty(bh); | ||
587 | bh->b_bdev = NULL; | ||
588 | clear_buffer_mapped(bh); | ||
589 | clear_buffer_req(bh); | ||
590 | clear_buffer_new(bh); | ||
591 | clear_buffer_delay(bh); | ||
592 | unlock_buffer(bh); | ||
593 | } | ||
594 | |||
595 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) | ||
596 | { | ||
597 | struct gfs2_sbd *sdp = page->mapping->host->i_sb->s_fs_info; | ||
598 | struct buffer_head *head, *bh, *next; | ||
599 | unsigned int curr_off = 0; | ||
600 | |||
601 | BUG_ON(!PageLocked(page)); | ||
602 | if (!page_has_buffers(page)) | ||
603 | return; | ||
604 | |||
605 | bh = head = page_buffers(page); | ||
606 | do { | ||
607 | unsigned int next_off = curr_off + bh->b_size; | ||
608 | next = bh->b_this_page; | ||
609 | |||
610 | if (offset <= curr_off) | ||
611 | discard_buffer(sdp, bh); | ||
612 | |||
613 | curr_off = next_off; | ||
614 | bh = next; | ||
615 | } while (bh != head); | ||
616 | |||
617 | if (!offset) | ||
618 | try_to_release_page(page, 0); | ||
619 | |||
620 | return; | ||
621 | } | ||
622 | |||
623 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, | ||
624 | const struct iovec *iov, loff_t offset, | ||
625 | unsigned long nr_segs) | ||
626 | { | ||
627 | struct file *file = iocb->ki_filp; | ||
628 | struct inode *inode = file->f_mapping->host; | ||
629 | struct gfs2_inode *ip = GFS2_I(inode); | ||
630 | struct gfs2_holder gh; | ||
631 | int rv; | ||
632 | |||
633 | if (rw == READ) | ||
634 | mutex_lock(&inode->i_mutex); | ||
635 | /* | ||
636 | * Shared lock, even if its a write, since we do no allocation | ||
637 | * on this path. All we need change is atime. | ||
638 | */ | ||
639 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh); | ||
640 | rv = gfs2_glock_nq_m_atime(1, &gh); | ||
641 | if (rv) | ||
642 | goto out; | ||
643 | |||
644 | if (offset > i_size_read(inode)) | ||
645 | goto out; | ||
646 | |||
647 | /* | ||
648 | * Should we return an error here? I can't see that O_DIRECT for | ||
649 | * a journaled file makes any sense. For now we'll silently fall | ||
650 | * back to buffered I/O, likewise we do the same for stuffed | ||
651 | * files since they are (a) small and (b) unaligned. | ||
652 | */ | ||
653 | if (gfs2_is_jdata(ip)) | ||
654 | goto out; | ||
655 | |||
656 | if (gfs2_is_stuffed(ip)) | ||
657 | goto out; | ||
658 | |||
659 | rv = blockdev_direct_IO_own_locking(rw, iocb, inode, | ||
660 | inode->i_sb->s_bdev, | ||
661 | iov, offset, nr_segs, | ||
662 | gfs2_get_block, NULL); | ||
663 | out: | ||
664 | gfs2_glock_dq_m(1, &gh); | ||
665 | gfs2_holder_uninit(&gh); | ||
666 | if (rw == READ) | ||
667 | mutex_unlock(&inode->i_mutex); | ||
668 | |||
669 | return rv; | ||
670 | } | ||
671 | |||
672 | /** | ||
673 | * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out. | ||
674 | * @bh: the buffer we're stuck on | ||
675 | * | ||
676 | */ | ||
677 | |||
678 | static void stuck_releasepage(struct buffer_head *bh) | ||
679 | { | ||
680 | struct inode *inode = bh->b_page->mapping->host; | ||
681 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | ||
682 | struct gfs2_bufdata *bd = bh->b_private; | ||
683 | struct gfs2_glock *gl; | ||
684 | |||
685 | fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode); | ||
686 | fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n", | ||
687 | (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count)); | ||
688 | fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh)); | ||
689 | fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL"); | ||
690 | |||
691 | if (!bd) | ||
692 | return; | ||
693 | |||
694 | gl = bd->bd_gl; | ||
695 | |||
696 | fs_warn(sdp, "gl = (%u, %llu)\n", | ||
697 | gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number); | ||
698 | |||
699 | fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n", | ||
700 | (list_empty(&bd->bd_list_tr)) ? "no" : "yes", | ||
701 | (list_empty(&bd->bd_le.le_list)) ? "no" : "yes"); | ||
702 | |||
703 | if (gl->gl_ops == &gfs2_inode_glops) { | ||
704 | struct gfs2_inode *ip = gl->gl_object; | ||
705 | unsigned int x; | ||
706 | |||
707 | if (!ip) | ||
708 | return; | ||
709 | |||
710 | fs_warn(sdp, "ip = %llu %llu\n", | ||
711 | (unsigned long long)ip->i_num.no_formal_ino, | ||
712 | (unsigned long long)ip->i_num.no_addr); | ||
713 | |||
714 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) | ||
715 | fs_warn(sdp, "ip->i_cache[%u] = %s\n", | ||
716 | x, (ip->i_cache[x]) ? "!NULL" : "NULL"); | ||
717 | } | ||
718 | } | ||
719 | |||
720 | /** | ||
721 | * gfs2_aspace_releasepage - free the metadata associated with a page | ||
722 | * @page: the page that's being released | ||
723 | * @gfp_mask: passed from Linux VFS, ignored by us | ||
724 | * | ||
725 | * Call try_to_free_buffers() if the buffers in this page can be | ||
726 | * released. | ||
727 | * | ||
728 | * Returns: 0 | ||
729 | */ | ||
730 | |||
731 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | ||
732 | { | ||
733 | struct inode *aspace = page->mapping->host; | ||
734 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | ||
735 | struct buffer_head *bh, *head; | ||
736 | struct gfs2_bufdata *bd; | ||
737 | unsigned long t; | ||
738 | |||
739 | if (!page_has_buffers(page)) | ||
740 | goto out; | ||
741 | |||
742 | head = bh = page_buffers(page); | ||
743 | do { | ||
744 | t = jiffies; | ||
745 | |||
746 | while (atomic_read(&bh->b_count)) { | ||
747 | if (atomic_read(&aspace->i_writecount)) { | ||
748 | if (time_after_eq(jiffies, t + | ||
749 | gfs2_tune_get(sdp, gt_stall_secs) * HZ)) { | ||
750 | stuck_releasepage(bh); | ||
751 | t = jiffies; | ||
752 | } | ||
753 | |||
754 | yield(); | ||
755 | continue; | ||
756 | } | ||
757 | |||
758 | return 0; | ||
759 | } | ||
760 | |||
761 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); | ||
762 | |||
763 | bd = bh->b_private; | ||
764 | if (bd) { | ||
765 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | ||
766 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | ||
767 | gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list)); | ||
768 | gfs2_assert_warn(sdp, !bd->bd_ail); | ||
769 | kmem_cache_free(gfs2_bufdata_cachep, bd); | ||
770 | bh->b_private = NULL; | ||
771 | } | ||
772 | |||
773 | bh = bh->b_this_page; | ||
774 | } | ||
775 | while (bh != head); | ||
776 | |||
777 | out: | ||
778 | return try_to_free_buffers(page); | ||
779 | } | ||
780 | |||
781 | const struct address_space_operations gfs2_file_aops = { | ||
782 | .writepage = gfs2_writepage, | ||
783 | .readpage = gfs2_readpage, | ||
784 | .readpages = gfs2_readpages, | ||
785 | .sync_page = block_sync_page, | ||
786 | .prepare_write = gfs2_prepare_write, | ||
787 | .commit_write = gfs2_commit_write, | ||
788 | .bmap = gfs2_bmap, | ||
789 | .invalidatepage = gfs2_invalidatepage, | ||
790 | .releasepage = gfs2_releasepage, | ||
791 | .direct_IO = gfs2_direct_IO, | ||
792 | }; | ||
793 | |||