diff options
Diffstat (limited to 'fs/gfs2/ops_address.c')
-rw-r--r-- | fs/gfs2/ops_address.c | 790 |
1 files changed, 790 insertions, 0 deletions
diff --git a/fs/gfs2/ops_address.c b/fs/gfs2/ops_address.c new file mode 100644 index 000000000000..4fb743f4e4a4 --- /dev/null +++ b/fs/gfs2/ops_address.c | |||
@@ -0,0 +1,790 @@ | |||
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 version 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 | #include <linux/lm_interface.h> | ||
21 | |||
22 | #include "gfs2.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 | return gfs2_block_map(inode, lblock, create, bh_result, 32); | ||
69 | } | ||
70 | |||
71 | /** | ||
72 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block | ||
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 | |||
81 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, | ||
82 | struct buffer_head *bh_result, int create) | ||
83 | { | ||
84 | int error; | ||
85 | |||
86 | error = gfs2_block_map(inode, lblock, 0, bh_result, 1); | ||
87 | if (error) | ||
88 | return error; | ||
89 | if (bh_result->b_blocknr == 0) | ||
90 | return -EIO; | ||
91 | return 0; | ||
92 | } | ||
93 | |||
94 | static int gfs2_get_block_direct(struct inode *inode, sector_t lblock, | ||
95 | struct buffer_head *bh_result, int create) | ||
96 | { | ||
97 | return gfs2_block_map(inode, lblock, 0, bh_result, 32); | ||
98 | } | ||
99 | |||
100 | /** | ||
101 | * gfs2_writepage - Write complete page | ||
102 | * @page: Page to write | ||
103 | * | ||
104 | * Returns: errno | ||
105 | * | ||
106 | * Some of this is copied from block_write_full_page() although we still | ||
107 | * call it to do most of the work. | ||
108 | */ | ||
109 | |||
110 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) | ||
111 | { | ||
112 | struct inode *inode = page->mapping->host; | ||
113 | struct gfs2_inode *ip = GFS2_I(inode); | ||
114 | struct gfs2_sbd *sdp = GFS2_SB(inode); | ||
115 | loff_t i_size = i_size_read(inode); | ||
116 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | ||
117 | unsigned offset; | ||
118 | int error; | ||
119 | int done_trans = 0; | ||
120 | |||
121 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) { | ||
122 | unlock_page(page); | ||
123 | return -EIO; | ||
124 | } | ||
125 | if (current->journal_info) | ||
126 | goto out_ignore; | ||
127 | |||
128 | /* Is the page fully outside i_size? (truncate in progress) */ | ||
129 | offset = i_size & (PAGE_CACHE_SIZE-1); | ||
130 | if (page->index > end_index || (page->index == end_index && !offset)) { | ||
131 | page->mapping->a_ops->invalidatepage(page, 0); | ||
132 | unlock_page(page); | ||
133 | return 0; /* don't care */ | ||
134 | } | ||
135 | |||
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; | ||
140 | if (!page_has_buffers(page)) { | ||
141 | create_empty_buffers(page, inode->i_sb->s_blocksize, | ||
142 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | ||
143 | } | ||
144 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1); | ||
145 | done_trans = 1; | ||
146 | } | ||
147 | error = block_write_full_page(page, gfs2_get_block_noalloc, wbc); | ||
148 | if (done_trans) | ||
149 | gfs2_trans_end(sdp); | ||
150 | gfs2_meta_cache_flush(ip); | ||
151 | return error; | ||
152 | |||
153 | out_ignore: | ||
154 | redirty_page_for_writepage(wbc, page); | ||
155 | unlock_page(page); | ||
156 | return 0; | ||
157 | } | ||
158 | |||
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 | |||
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 | |||
186 | /* Only the first page of a stuffed file might contain data */ | ||
187 | if (unlikely(page->index)) | ||
188 | return zero_readpage(page); | ||
189 | |||
190 | error = gfs2_meta_inode_buffer(ip, &dibh); | ||
191 | if (error) | ||
192 | return error; | ||
193 | |||
194 | kaddr = kmap_atomic(page, KM_USER0); | ||
195 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), | ||
196 | ip->i_di.di_size); | ||
197 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); | ||
198 | kunmap_atomic(page, KM_USER0); | ||
199 | |||
200 | brelse(dibh); | ||
201 | |||
202 | SetPageUptodate(page); | ||
203 | |||
204 | return 0; | ||
205 | } | ||
206 | |||
207 | |||
208 | /** | ||
209 | * gfs2_readpage - readpage with locking | ||
210 | * @file: The file to read a page for. N.B. This may be NULL if we are | ||
211 | * reading an internal file. | ||
212 | * @page: The page to read | ||
213 | * | ||
214 | * Returns: errno | ||
215 | */ | ||
216 | |||
217 | static int gfs2_readpage(struct file *file, struct page *page) | ||
218 | { | ||
219 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | ||
220 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | ||
221 | struct gfs2_file *gf = NULL; | ||
222 | struct gfs2_holder gh; | ||
223 | int error; | ||
224 | int do_unlock = 0; | ||
225 | |||
226 | if (likely(file != &gfs2_internal_file_sentinel)) { | ||
227 | if (file) { | ||
228 | gf = file->private_data; | ||
229 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) | ||
230 | /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */ | ||
231 | goto skip_lock; | ||
232 | } | ||
233 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh); | ||
234 | do_unlock = 1; | ||
235 | error = gfs2_glock_nq_m_atime(1, &gh); | ||
236 | if (unlikely(error)) | ||
237 | goto out_unlock; | ||
238 | } | ||
239 | |||
240 | skip_lock: | ||
241 | if (gfs2_is_stuffed(ip)) { | ||
242 | error = stuffed_readpage(ip, page); | ||
243 | unlock_page(page); | ||
244 | } else | ||
245 | error = mpage_readpage(page, gfs2_get_block); | ||
246 | |||
247 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | ||
248 | error = -EIO; | ||
249 | |||
250 | if (do_unlock) { | ||
251 | gfs2_glock_dq_m(1, &gh); | ||
252 | gfs2_holder_uninit(&gh); | ||
253 | } | ||
254 | out: | ||
255 | return error; | ||
256 | out_unlock: | ||
257 | unlock_page(page); | ||
258 | if (do_unlock) | ||
259 | gfs2_holder_uninit(&gh); | ||
260 | goto out; | ||
261 | } | ||
262 | |||
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; | ||
282 | struct gfs2_inode *ip = GFS2_I(inode); | ||
283 | struct gfs2_sbd *sdp = GFS2_SB(inode); | ||
284 | struct gfs2_holder gh; | ||
285 | unsigned page_idx; | ||
286 | int ret; | ||
287 | int do_unlock = 0; | ||
288 | |||
289 | if (likely(file != &gfs2_internal_file_sentinel)) { | ||
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 | } | ||
295 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, | ||
296 | LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh); | ||
297 | do_unlock = 1; | ||
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 | } | ||
304 | skip_lock: | ||
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++) { | ||
309 | struct page *page = list_entry(pages->prev, struct page, lru); | ||
310 | prefetchw(&page->flags); | ||
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); | ||
318 | } else { | ||
319 | page_cache_release(page); | ||
320 | } | ||
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 | |||
329 | if (do_unlock) { | ||
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++) { | ||
342 | struct page *page = list_entry(pages->prev, struct page, lru); | ||
343 | list_del(&page->lru); | ||
344 | unlock_page(page); | ||
345 | page_cache_release(page); | ||
346 | } | ||
347 | if (do_unlock) | ||
348 | gfs2_holder_uninit(&gh); | ||
349 | goto out; | ||
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 | { | ||
365 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | ||
366 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | ||
367 | unsigned int data_blocks, ind_blocks, rblocks; | ||
368 | int alloc_required; | ||
369 | int error = 0; | ||
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; | ||
373 | |||
374 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh); | ||
375 | error = gfs2_glock_nq_m_atime(1, &ip->i_gh); | ||
376 | if (error) | ||
377 | goto out_uninit; | ||
378 | |||
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; | ||
384 | |||
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)) { | ||
415 | error = gfs2_unstuff_dinode(ip, page); | ||
416 | if (error == 0) | ||
417 | goto prepare_write; | ||
418 | } else if (!PageUptodate(page)) | ||
419 | error = stuffed_readpage(ip, page); | ||
420 | goto out; | ||
421 | } | ||
422 | |||
423 | prepare_write: | ||
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 | } | ||
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; | ||
459 | struct gfs2_inode *ip = GFS2_I(inode); | ||
460 | struct gfs2_sbd *sdp = GFS2_SB(inode); | ||
461 | int error = -EOPNOTSUPP; | ||
462 | struct buffer_head *dibh; | ||
463 | struct gfs2_alloc *al = &ip->i_alloc; | ||
464 | struct gfs2_dinode *di; | ||
465 | |||
466 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl))) | ||
467 | goto fail_nounlock; | ||
468 | |||
469 | error = gfs2_meta_inode_buffer(ip, &dibh); | ||
470 | if (error) | ||
471 | goto fail_endtrans; | ||
472 | |||
473 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | ||
474 | di = (struct gfs2_dinode *)dibh->b_data; | ||
475 | |||
476 | if (gfs2_is_stuffed(ip)) { | ||
477 | u64 file_size; | ||
478 | void *kaddr; | ||
479 | |||
480 | file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to; | ||
481 | |||
482 | kaddr = kmap_atomic(page, KM_USER0); | ||
483 | memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from, | ||
484 | kaddr + from, to - from); | ||
485 | kunmap_atomic(page, KM_USER0); | ||
486 | |||
487 | SetPageUptodate(page); | ||
488 | |||
489 | if (inode->i_size < file_size) | ||
490 | i_size_write(inode, file_size); | ||
491 | } else { | ||
492 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || | ||
493 | gfs2_is_jdata(ip)) | ||
494 | gfs2_page_add_databufs(ip, page, from, to); | ||
495 | error = generic_commit_write(file, page, from, to); | ||
496 | if (error) | ||
497 | goto fail; | ||
498 | } | ||
499 | |||
500 | if (ip->i_di.di_size < inode->i_size) { | ||
501 | ip->i_di.di_size = inode->i_size; | ||
502 | di->di_size = cpu_to_be64(inode->i_size); | ||
503 | } | ||
504 | |||
505 | di->di_mode = cpu_to_be32(inode->i_mode); | ||
506 | di->di_atime = cpu_to_be64(inode->i_atime.tv_sec); | ||
507 | di->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec); | ||
508 | di->di_ctime = cpu_to_be64(inode->i_ctime.tv_sec); | ||
509 | |||
510 | brelse(dibh); | ||
511 | gfs2_trans_end(sdp); | ||
512 | if (al->al_requested) { | ||
513 | gfs2_inplace_release(ip); | ||
514 | gfs2_quota_unlock(ip); | ||
515 | gfs2_alloc_put(ip); | ||
516 | } | ||
517 | gfs2_glock_dq_m(1, &ip->i_gh); | ||
518 | gfs2_holder_uninit(&ip->i_gh); | ||
519 | return 0; | ||
520 | |||
521 | fail: | ||
522 | brelse(dibh); | ||
523 | fail_endtrans: | ||
524 | gfs2_trans_end(sdp); | ||
525 | if (al->al_requested) { | ||
526 | gfs2_inplace_release(ip); | ||
527 | gfs2_quota_unlock(ip); | ||
528 | gfs2_alloc_put(ip); | ||
529 | } | ||
530 | gfs2_glock_dq_m(1, &ip->i_gh); | ||
531 | gfs2_holder_uninit(&ip->i_gh); | ||
532 | fail_nounlock: | ||
533 | ClearPageUptodate(page); | ||
534 | return error; | ||
535 | } | ||
536 | |||
537 | /** | ||
538 | * gfs2_bmap - Block map function | ||
539 | * @mapping: Address space info | ||
540 | * @lblock: The block to map | ||
541 | * | ||
542 | * Returns: The disk address for the block or 0 on hole or error | ||
543 | */ | ||
544 | |||
545 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | ||
546 | { | ||
547 | struct gfs2_inode *ip = GFS2_I(mapping->host); | ||
548 | struct gfs2_holder i_gh; | ||
549 | sector_t dblock = 0; | ||
550 | int error; | ||
551 | |||
552 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); | ||
553 | if (error) | ||
554 | return 0; | ||
555 | |||
556 | if (!gfs2_is_stuffed(ip)) | ||
557 | dblock = generic_block_bmap(mapping, lblock, gfs2_get_block); | ||
558 | |||
559 | gfs2_glock_dq_uninit(&i_gh); | ||
560 | |||
561 | return dblock; | ||
562 | } | ||
563 | |||
564 | static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh) | ||
565 | { | ||
566 | struct gfs2_bufdata *bd; | ||
567 | |||
568 | gfs2_log_lock(sdp); | ||
569 | bd = bh->b_private; | ||
570 | if (bd) { | ||
571 | bd->bd_bh = NULL; | ||
572 | bh->b_private = NULL; | ||
573 | } | ||
574 | gfs2_log_unlock(sdp); | ||
575 | |||
576 | lock_buffer(bh); | ||
577 | clear_buffer_dirty(bh); | ||
578 | bh->b_bdev = NULL; | ||
579 | clear_buffer_mapped(bh); | ||
580 | clear_buffer_req(bh); | ||
581 | clear_buffer_new(bh); | ||
582 | clear_buffer_delay(bh); | ||
583 | unlock_buffer(bh); | ||
584 | } | ||
585 | |||
586 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) | ||
587 | { | ||
588 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | ||
589 | struct buffer_head *head, *bh, *next; | ||
590 | unsigned int curr_off = 0; | ||
591 | |||
592 | BUG_ON(!PageLocked(page)); | ||
593 | if (!page_has_buffers(page)) | ||
594 | return; | ||
595 | |||
596 | bh = head = page_buffers(page); | ||
597 | do { | ||
598 | unsigned int next_off = curr_off + bh->b_size; | ||
599 | next = bh->b_this_page; | ||
600 | |||
601 | if (offset <= curr_off) | ||
602 | discard_buffer(sdp, bh); | ||
603 | |||
604 | curr_off = next_off; | ||
605 | bh = next; | ||
606 | } while (bh != head); | ||
607 | |||
608 | if (!offset) | ||
609 | try_to_release_page(page, 0); | ||
610 | |||
611 | return; | ||
612 | } | ||
613 | |||
614 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, | ||
615 | const struct iovec *iov, loff_t offset, | ||
616 | unsigned long nr_segs) | ||
617 | { | ||
618 | struct file *file = iocb->ki_filp; | ||
619 | struct inode *inode = file->f_mapping->host; | ||
620 | struct gfs2_inode *ip = GFS2_I(inode); | ||
621 | struct gfs2_holder gh; | ||
622 | int rv; | ||
623 | |||
624 | if (rw == READ) | ||
625 | mutex_lock(&inode->i_mutex); | ||
626 | /* | ||
627 | * Shared lock, even if its a write, since we do no allocation | ||
628 | * on this path. All we need change is atime. | ||
629 | */ | ||
630 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh); | ||
631 | rv = gfs2_glock_nq_m_atime(1, &gh); | ||
632 | if (rv) | ||
633 | goto out; | ||
634 | |||
635 | if (offset > i_size_read(inode)) | ||
636 | goto out; | ||
637 | |||
638 | /* | ||
639 | * Should we return an error here? I can't see that O_DIRECT for | ||
640 | * a journaled file makes any sense. For now we'll silently fall | ||
641 | * back to buffered I/O, likewise we do the same for stuffed | ||
642 | * files since they are (a) small and (b) unaligned. | ||
643 | */ | ||
644 | if (gfs2_is_jdata(ip)) | ||
645 | goto out; | ||
646 | |||
647 | if (gfs2_is_stuffed(ip)) | ||
648 | goto out; | ||
649 | |||
650 | rv = blockdev_direct_IO_own_locking(rw, iocb, inode, | ||
651 | inode->i_sb->s_bdev, | ||
652 | iov, offset, nr_segs, | ||
653 | gfs2_get_block_direct, NULL); | ||
654 | out: | ||
655 | gfs2_glock_dq_m(1, &gh); | ||
656 | gfs2_holder_uninit(&gh); | ||
657 | if (rw == READ) | ||
658 | mutex_unlock(&inode->i_mutex); | ||
659 | |||
660 | return rv; | ||
661 | } | ||
662 | |||
663 | /** | ||
664 | * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out. | ||
665 | * @bh: the buffer we're stuck on | ||
666 | * | ||
667 | */ | ||
668 | |||
669 | static void stuck_releasepage(struct buffer_head *bh) | ||
670 | { | ||
671 | struct inode *inode = bh->b_page->mapping->host; | ||
672 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | ||
673 | struct gfs2_bufdata *bd = bh->b_private; | ||
674 | struct gfs2_glock *gl; | ||
675 | static unsigned limit = 0; | ||
676 | |||
677 | if (limit > 3) | ||
678 | return; | ||
679 | limit++; | ||
680 | |||
681 | fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode); | ||
682 | fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n", | ||
683 | (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count)); | ||
684 | fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh)); | ||
685 | fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL"); | ||
686 | |||
687 | if (!bd) | ||
688 | return; | ||
689 | |||
690 | gl = bd->bd_gl; | ||
691 | |||
692 | fs_warn(sdp, "gl = (%u, %llu)\n", | ||
693 | gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number); | ||
694 | |||
695 | fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n", | ||
696 | (list_empty(&bd->bd_list_tr)) ? "no" : "yes", | ||
697 | (list_empty(&bd->bd_le.le_list)) ? "no" : "yes"); | ||
698 | |||
699 | if (gl->gl_ops == &gfs2_inode_glops) { | ||
700 | struct gfs2_inode *ip = gl->gl_object; | ||
701 | unsigned int x; | ||
702 | |||
703 | if (!ip) | ||
704 | return; | ||
705 | |||
706 | fs_warn(sdp, "ip = %llu %llu\n", | ||
707 | (unsigned long long)ip->i_num.no_formal_ino, | ||
708 | (unsigned long long)ip->i_num.no_addr); | ||
709 | |||
710 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) | ||
711 | fs_warn(sdp, "ip->i_cache[%u] = %s\n", | ||
712 | x, (ip->i_cache[x]) ? "!NULL" : "NULL"); | ||
713 | } | ||
714 | } | ||
715 | |||
716 | /** | ||
717 | * gfs2_releasepage - free the metadata associated with a page | ||
718 | * @page: the page that's being released | ||
719 | * @gfp_mask: passed from Linux VFS, ignored by us | ||
720 | * | ||
721 | * Call try_to_free_buffers() if the buffers in this page can be | ||
722 | * released. | ||
723 | * | ||
724 | * Returns: 0 | ||
725 | */ | ||
726 | |||
727 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | ||
728 | { | ||
729 | struct inode *aspace = page->mapping->host; | ||
730 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | ||
731 | struct buffer_head *bh, *head; | ||
732 | struct gfs2_bufdata *bd; | ||
733 | unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ; | ||
734 | |||
735 | if (!page_has_buffers(page)) | ||
736 | goto out; | ||
737 | |||
738 | head = bh = page_buffers(page); | ||
739 | do { | ||
740 | while (atomic_read(&bh->b_count)) { | ||
741 | if (!atomic_read(&aspace->i_writecount)) | ||
742 | return 0; | ||
743 | |||
744 | if (time_after_eq(jiffies, t)) { | ||
745 | stuck_releasepage(bh); | ||
746 | /* should we withdraw here? */ | ||
747 | return 0; | ||
748 | } | ||
749 | |||
750 | yield(); | ||
751 | } | ||
752 | |||
753 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); | ||
754 | gfs2_assert_warn(sdp, !buffer_dirty(bh)); | ||
755 | |||
756 | gfs2_log_lock(sdp); | ||
757 | bd = bh->b_private; | ||
758 | if (bd) { | ||
759 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | ||
760 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | ||
761 | gfs2_assert_warn(sdp, !bd->bd_ail); | ||
762 | bd->bd_bh = NULL; | ||
763 | if (!list_empty(&bd->bd_le.le_list)) | ||
764 | bd = NULL; | ||
765 | bh->b_private = NULL; | ||
766 | } | ||
767 | gfs2_log_unlock(sdp); | ||
768 | if (bd) | ||
769 | kmem_cache_free(gfs2_bufdata_cachep, bd); | ||
770 | |||
771 | bh = bh->b_this_page; | ||
772 | } while (bh != head); | ||
773 | |||
774 | out: | ||
775 | return try_to_free_buffers(page); | ||
776 | } | ||
777 | |||
778 | const struct address_space_operations gfs2_file_aops = { | ||
779 | .writepage = gfs2_writepage, | ||
780 | .readpage = gfs2_readpage, | ||
781 | .readpages = gfs2_readpages, | ||
782 | .sync_page = block_sync_page, | ||
783 | .prepare_write = gfs2_prepare_write, | ||
784 | .commit_write = gfs2_commit_write, | ||
785 | .bmap = gfs2_bmap, | ||
786 | .invalidatepage = gfs2_invalidatepage, | ||
787 | .releasepage = gfs2_releasepage, | ||
788 | .direct_IO = gfs2_direct_IO, | ||
789 | }; | ||
790 | |||