]>
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 | |
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> | |
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> |
b3b94faa DT |
20 | |
21 | #include "gfs2.h" | |
5c676f6d SW |
22 | #include "lm_interface.h" |
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 | { |
b3b94faa DT |
68 | int new = create; |
69 | uint64_t dblock; | |
70 | int error; | |
fd88de56 | 71 | int boundary; |
b3b94faa | 72 | |
fd88de56 | 73 | error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary); |
b3b94faa DT |
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); | |
fd88de56 SW |
83 | if (boundary) |
84 | set_buffer_boundary(bh_result); | |
b3b94faa DT |
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 | { | |
b3b94faa DT |
102 | int new = 0; |
103 | uint64_t dblock; | |
104 | int error; | |
fd88de56 | 105 | int boundary; |
b3b94faa | 106 | |
fd88de56 | 107 | error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary); |
b3b94faa DT |
108 | if (error) |
109 | return error; | |
110 | ||
111 | if (dblock) | |
112 | map_bh(bh_result, inode->i_sb, dblock); | |
feaa7bba | 113 | else if (gfs2_assert_withdraw(GFS2_SB(inode), !create)) |
b3b94faa | 114 | error = -EIO; |
fd88de56 SW |
115 | if (boundary) |
116 | set_buffer_boundary(bh_result); | |
b3b94faa DT |
117 | |
118 | return error; | |
119 | } | |
120 | ||
b3b94faa DT |
121 | /** |
122 | * gfs2_writepage - Write complete page | |
123 | * @page: Page to write | |
124 | * | |
125 | * Returns: errno | |
126 | * | |
18ec7d5c SW |
127 | * Some of this is copied from block_write_full_page() although we still |
128 | * call it to do most of the work. | |
b3b94faa DT |
129 | */ |
130 | ||
131 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) | |
132 | { | |
18ec7d5c | 133 | struct inode *inode = page->mapping->host; |
feaa7bba SW |
134 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
135 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
18ec7d5c SW |
136 | loff_t i_size = i_size_read(inode); |
137 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
138 | unsigned offset; | |
b3b94faa | 139 | int error; |
18ec7d5c | 140 | int done_trans = 0; |
b3b94faa | 141 | |
b3b94faa DT |
142 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) { |
143 | unlock_page(page); | |
144 | return -EIO; | |
145 | } | |
5c676f6d | 146 | if (current->journal_info) |
18ec7d5c SW |
147 | goto out_ignore; |
148 | ||
149 | /* Is the page fully outside i_size? (truncate in progress) */ | |
150 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
d2d7b8a2 | 151 | if (page->index > end_index || (page->index == end_index && !offset)) { |
18ec7d5c | 152 | page->mapping->a_ops->invalidatepage(page, 0); |
b3b94faa | 153 | unlock_page(page); |
18ec7d5c | 154 | return 0; /* don't care */ |
b3b94faa DT |
155 | } |
156 | ||
18ec7d5c SW |
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 | } | |
18ec7d5c SW |
164 | error = block_write_full_page(page, get_block_noalloc, wbc); |
165 | if (done_trans) | |
166 | gfs2_trans_end(sdp); | |
b3b94faa | 167 | gfs2_meta_cache_flush(ip); |
b3b94faa | 168 | return error; |
18ec7d5c SW |
169 | |
170 | out_ignore: | |
171 | redirty_page_for_writepage(wbc, page); | |
172 | unlock_page(page); | |
173 | return 0; | |
b3b94faa DT |
174 | } |
175 | ||
fd88de56 SW |
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 | ||
b3b94faa DT |
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 | ||
fd88de56 SW |
203 | /* Only the first page of a stuffed file might contain data */ |
204 | if (unlikely(page->index)) | |
205 | return zero_readpage(page); | |
206 | ||
b3b94faa DT |
207 | error = gfs2_meta_inode_buffer(ip, &dibh); |
208 | if (error) | |
209 | return error; | |
210 | ||
5c4e9e03 | 211 | kaddr = kmap_atomic(page, KM_USER0); |
fd88de56 | 212 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), |
b3b94faa | 213 | ip->i_di.di_size); |
fd88de56 | 214 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); |
5c4e9e03 | 215 | kunmap_atomic(page, KM_USER0); |
b3b94faa DT |
216 | |
217 | brelse(dibh); | |
218 | ||
219 | SetPageUptodate(page); | |
220 | ||
221 | return 0; | |
222 | } | |
223 | ||
b3b94faa | 224 | |
b3b94faa DT |
225 | /** |
226 | * gfs2_readpage - readpage with locking | |
18ec7d5c SW |
227 | * @file: The file to read a page for. N.B. This may be NULL if we are |
228 | * reading an internal file. | |
b3b94faa DT |
229 | * @page: The page to read |
230 | * | |
231 | * Returns: errno | |
232 | */ | |
233 | ||
234 | static int gfs2_readpage(struct file *file, struct page *page) | |
235 | { | |
feaa7bba SW |
236 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
237 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
18ec7d5c | 238 | struct gfs2_holder gh; |
b3b94faa DT |
239 | int error; |
240 | ||
fd88de56 | 241 | if (likely(file != &gfs2_internal_file_sentinal)) { |
fe1bdedc | 242 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh); |
61a30dcb | 243 | error = gfs2_glock_nq_m_atime(1, &gh); |
fd88de56 | 244 | if (unlikely(error)) |
61a30dcb SW |
245 | goto out_unlock; |
246 | } | |
b3b94faa | 247 | |
18ec7d5c | 248 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
249 | error = stuffed_readpage(ip, page); |
250 | unlock_page(page); | |
b3b94faa | 251 | } else |
18ec7d5c | 252 | error = mpage_readpage(page, gfs2_get_block); |
b3b94faa DT |
253 | |
254 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
255 | error = -EIO; | |
256 | ||
61a30dcb SW |
257 | if (file != &gfs2_internal_file_sentinal) { |
258 | gfs2_glock_dq_m(1, &gh); | |
259 | gfs2_holder_uninit(&gh); | |
260 | } | |
18ec7d5c | 261 | out: |
b3b94faa | 262 | return error; |
18ec7d5c SW |
263 | out_unlock: |
264 | unlock_page(page); | |
fd88de56 SW |
265 | if (file != &gfs2_internal_file_sentinal) |
266 | gfs2_holder_uninit(&gh); | |
267 | goto out; | |
268 | } | |
269 | ||
fd88de56 SW |
270 | /** |
271 | * gfs2_readpages - Read a bunch of pages at once | |
272 | * | |
273 | * Some notes: | |
274 | * 1. This is only for readahead, so we can simply ignore any things | |
275 | * which are slightly inconvenient (such as locking conflicts between | |
276 | * the page lock and the glock) and return having done no I/O. Its | |
277 | * obviously not something we'd want to do on too regular a basis. | |
278 | * Any I/O we ignore at this time will be done via readpage later. | |
279 | * 2. We have to handle stuffed files here too. | |
280 | * 3. mpage_readpages() does most of the heavy lifting in the common case. | |
281 | * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places. | |
282 | * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as | |
283 | * well as read-ahead. | |
284 | */ | |
285 | static int gfs2_readpages(struct file *file, struct address_space *mapping, | |
286 | struct list_head *pages, unsigned nr_pages) | |
287 | { | |
288 | struct inode *inode = mapping->host; | |
feaa7bba SW |
289 | struct gfs2_inode *ip = GFS2_I(inode); |
290 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 SW |
291 | struct gfs2_holder gh; |
292 | unsigned page_idx; | |
293 | int ret; | |
294 | ||
295 | if (likely(file != &gfs2_internal_file_sentinal)) { | |
296 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, | |
297 | LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh); | |
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 | ||
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 | ||
329 | if (likely(file != &gfs2_internal_file_sentinal)) { | |
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 | } | |
347 | if (likely(file != &gfs2_internal_file_sentinal)) | |
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)) { |
b3b94faa DT |
475 | uint64_t file_size; |
476 | void *kaddr; | |
477 | ||
478 | file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to; | |
479 | ||
18ec7d5c | 480 | kaddr = kmap_atomic(page, KM_USER0); |
b3b94faa | 481 | memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from, |
18ec7d5c SW |
482 | (char *)kaddr + from, to - from); |
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; |
b3b94faa DT |
565 | gfs2_log_unlock(sdp); |
566 | brelse(bh); | |
567 | } else | |
568 | gfs2_log_unlock(sdp); | |
569 | ||
570 | lock_buffer(bh); | |
571 | clear_buffer_dirty(bh); | |
572 | bh->b_bdev = NULL; | |
573 | clear_buffer_mapped(bh); | |
574 | clear_buffer_req(bh); | |
575 | clear_buffer_new(bh); | |
576 | clear_buffer_delay(bh); | |
577 | unlock_buffer(bh); | |
578 | } | |
579 | ||
8628de05 | 580 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) |
b3b94faa | 581 | { |
5c676f6d | 582 | struct gfs2_sbd *sdp = page->mapping->host->i_sb->s_fs_info; |
b3b94faa DT |
583 | struct buffer_head *head, *bh, *next; |
584 | unsigned int curr_off = 0; | |
b3b94faa DT |
585 | |
586 | BUG_ON(!PageLocked(page)); | |
587 | if (!page_has_buffers(page)) | |
8628de05 | 588 | return; |
b3b94faa DT |
589 | |
590 | bh = head = page_buffers(page); | |
591 | do { | |
592 | unsigned int next_off = curr_off + bh->b_size; | |
593 | next = bh->b_this_page; | |
594 | ||
595 | if (offset <= curr_off) | |
596 | discard_buffer(sdp, bh); | |
597 | ||
598 | curr_off = next_off; | |
599 | bh = next; | |
600 | } while (bh != head); | |
601 | ||
602 | if (!offset) | |
8628de05 | 603 | try_to_release_page(page, 0); |
b3b94faa | 604 | |
8628de05 | 605 | return; |
b3b94faa DT |
606 | } |
607 | ||
a9e5f4d0 SW |
608 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, |
609 | const struct iovec *iov, loff_t offset, | |
610 | unsigned long nr_segs) | |
d1665e41 SW |
611 | { |
612 | struct file *file = iocb->ki_filp; | |
613 | struct inode *inode = file->f_mapping->host; | |
feaa7bba | 614 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
615 | struct gfs2_holder gh; |
616 | int rv; | |
617 | ||
a9e5f4d0 SW |
618 | if (rw == READ) |
619 | mutex_lock(&inode->i_mutex); | |
d1665e41 | 620 | /* |
a9e5f4d0 | 621 | * Shared lock, even if its a write, since we do no allocation |
d1665e41 SW |
622 | * on this path. All we need change is atime. |
623 | */ | |
624 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh); | |
625 | rv = gfs2_glock_nq_m_atime(1, &gh); | |
626 | if (rv) | |
627 | goto out; | |
628 | ||
a9e5f4d0 SW |
629 | if (offset > i_size_read(inode)) |
630 | goto out; | |
631 | ||
d1665e41 SW |
632 | /* |
633 | * Should we return an error here? I can't see that O_DIRECT for | |
634 | * a journaled file makes any sense. For now we'll silently fall | |
635 | * back to buffered I/O, likewise we do the same for stuffed | |
636 | * files since they are (a) small and (b) unaligned. | |
637 | */ | |
638 | if (gfs2_is_jdata(ip)) | |
639 | goto out; | |
640 | ||
641 | if (gfs2_is_stuffed(ip)) | |
642 | goto out; | |
643 | ||
a9e5f4d0 SW |
644 | rv = blockdev_direct_IO_own_locking(rw, iocb, inode, |
645 | inode->i_sb->s_bdev, | |
646 | iov, offset, nr_segs, | |
647 | gfs2_get_block, NULL); | |
d1665e41 SW |
648 | out: |
649 | gfs2_glock_dq_m(1, &gh); | |
650 | gfs2_holder_uninit(&gh); | |
a9e5f4d0 SW |
651 | if (rw == READ) |
652 | mutex_unlock(&inode->i_mutex); | |
d1665e41 SW |
653 | |
654 | return rv; | |
655 | } | |
656 | ||
4340fe62 SW |
657 | /** |
658 | * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out. | |
659 | * @bh: the buffer we're stuck on | |
660 | * | |
661 | */ | |
662 | ||
663 | static void stuck_releasepage(struct buffer_head *bh) | |
664 | { | |
665 | struct inode *inode = bh->b_page->mapping->host; | |
666 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
667 | struct gfs2_bufdata *bd = bh->b_private; | |
668 | struct gfs2_glock *gl; | |
669 | ||
670 | fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode); | |
671 | fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n", | |
672 | (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count)); | |
673 | fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh)); | |
674 | fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL"); | |
675 | ||
676 | if (!bd) | |
677 | return; | |
678 | ||
679 | gl = bd->bd_gl; | |
680 | ||
681 | fs_warn(sdp, "gl = (%u, %llu)\n", | |
682 | gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number); | |
683 | ||
684 | fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n", | |
685 | (list_empty(&bd->bd_list_tr)) ? "no" : "yes", | |
686 | (list_empty(&bd->bd_le.le_list)) ? "no" : "yes"); | |
687 | ||
688 | if (gl->gl_ops == &gfs2_inode_glops) { | |
689 | struct gfs2_inode *ip = gl->gl_object; | |
690 | unsigned int x; | |
691 | ||
692 | if (!ip) | |
693 | return; | |
694 | ||
695 | fs_warn(sdp, "ip = %llu %llu\n", | |
696 | (unsigned long long)ip->i_num.no_formal_ino, | |
697 | (unsigned long long)ip->i_num.no_addr); | |
698 | ||
699 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) | |
700 | fs_warn(sdp, "ip->i_cache[%u] = %s\n", | |
701 | x, (ip->i_cache[x]) ? "!NULL" : "NULL"); | |
702 | } | |
703 | } | |
704 | ||
705 | /** | |
706 | * gfs2_aspace_releasepage - free the metadata associated with a page | |
707 | * @page: the page that's being released | |
708 | * @gfp_mask: passed from Linux VFS, ignored by us | |
709 | * | |
710 | * Call try_to_free_buffers() if the buffers in this page can be | |
711 | * released. | |
712 | * | |
713 | * Returns: 0 | |
714 | */ | |
715 | ||
716 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
717 | { | |
718 | struct inode *aspace = page->mapping->host; | |
719 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | |
720 | struct buffer_head *bh, *head; | |
721 | struct gfs2_bufdata *bd; | |
722 | unsigned long t; | |
723 | ||
724 | if (!page_has_buffers(page)) | |
725 | goto out; | |
726 | ||
727 | head = bh = page_buffers(page); | |
728 | do { | |
729 | t = jiffies; | |
730 | ||
731 | while (atomic_read(&bh->b_count)) { | |
732 | if (atomic_read(&aspace->i_writecount)) { | |
733 | if (time_after_eq(jiffies, t + | |
734 | gfs2_tune_get(sdp, gt_stall_secs) * HZ)) { | |
735 | stuck_releasepage(bh); | |
736 | t = jiffies; | |
737 | } | |
738 | ||
739 | yield(); | |
740 | continue; | |
741 | } | |
742 | ||
743 | return 0; | |
744 | } | |
745 | ||
746 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); | |
747 | ||
748 | bd = bh->b_private; | |
749 | if (bd) { | |
750 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
751 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
752 | gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list)); | |
753 | gfs2_assert_warn(sdp, !bd->bd_ail); | |
754 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
755 | bh->b_private = NULL; | |
756 | } | |
757 | ||
758 | bh = bh->b_this_page; | |
759 | } | |
760 | while (bh != head); | |
761 | ||
a9e5f4d0 | 762 | out: |
4340fe62 SW |
763 | return try_to_free_buffers(page); |
764 | } | |
765 | ||
66de045d | 766 | const struct address_space_operations gfs2_file_aops = { |
b3b94faa DT |
767 | .writepage = gfs2_writepage, |
768 | .readpage = gfs2_readpage, | |
fd88de56 | 769 | .readpages = gfs2_readpages, |
b3b94faa DT |
770 | .sync_page = block_sync_page, |
771 | .prepare_write = gfs2_prepare_write, | |
772 | .commit_write = gfs2_commit_write, | |
773 | .bmap = gfs2_bmap, | |
774 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 775 | .releasepage = gfs2_releasepage, |
b3b94faa DT |
776 | .direct_IO = gfs2_direct_IO, |
777 | }; | |
778 |