Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
7ae8fa84 | 3 | * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
15 | #include <linux/pagemap.h> | |
fd88de56 | 16 | #include <linux/pagevec.h> |
9b124fbb | 17 | #include <linux/mpage.h> |
d1665e41 | 18 | #include <linux/fs.h> |
a8d638e3 | 19 | #include <linux/writeback.h> |
7765ec26 | 20 | #include <linux/swap.h> |
5c676f6d | 21 | #include <linux/gfs2_ondisk.h> |
7d308590 | 22 | #include <linux/lm_interface.h> |
51ff87bd | 23 | #include <linux/swap.h> |
b3b94faa DT |
24 | |
25 | #include "gfs2.h" | |
5c676f6d | 26 | #include "incore.h" |
b3b94faa DT |
27 | #include "bmap.h" |
28 | #include "glock.h" | |
29 | #include "inode.h" | |
b3b94faa DT |
30 | #include "log.h" |
31 | #include "meta_io.h" | |
32 | #include "ops_address.h" | |
b3b94faa DT |
33 | #include "quota.h" |
34 | #include "trans.h" | |
18ec7d5c | 35 | #include "rgrp.h" |
cd81a4ba | 36 | #include "super.h" |
5c676f6d | 37 | #include "util.h" |
4340fe62 | 38 | #include "glops.h" |
b3b94faa | 39 | |
ba7f7290 SW |
40 | |
41 | static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page, | |
42 | unsigned int from, unsigned int to) | |
43 | { | |
44 | struct buffer_head *head = page_buffers(page); | |
45 | unsigned int bsize = head->b_size; | |
46 | struct buffer_head *bh; | |
47 | unsigned int start, end; | |
48 | ||
49 | for (bh = head, start = 0; bh != head || !start; | |
50 | bh = bh->b_this_page, start = end) { | |
51 | end = start + bsize; | |
52 | if (end <= from || start >= to) | |
53 | continue; | |
ddf4b426 BM |
54 | if (gfs2_is_jdata(ip)) |
55 | set_buffer_uptodate(bh); | |
ba7f7290 SW |
56 | gfs2_trans_add_bh(ip->i_gl, bh, 0); |
57 | } | |
58 | } | |
59 | ||
b3b94faa | 60 | /** |
4ff14670 | 61 | * gfs2_get_block - Fills in a buffer head with details about a block |
b3b94faa DT |
62 | * @inode: The inode |
63 | * @lblock: The block number to look up | |
64 | * @bh_result: The buffer head to return the result in | |
65 | * @create: Non-zero if we may add block to the file | |
66 | * | |
67 | * Returns: errno | |
68 | */ | |
69 | ||
4ff14670 SW |
70 | int gfs2_get_block(struct inode *inode, sector_t lblock, |
71 | struct buffer_head *bh_result, int create) | |
b3b94faa | 72 | { |
23591256 | 73 | return gfs2_block_map(inode, lblock, create, bh_result); |
b3b94faa DT |
74 | } |
75 | ||
76 | /** | |
7a6bbacb | 77 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block |
b3b94faa DT |
78 | * @inode: The inode |
79 | * @lblock: The block number to look up | |
80 | * @bh_result: The buffer head to return the result in | |
81 | * @create: Non-zero if we may add block to the file | |
82 | * | |
83 | * Returns: errno | |
84 | */ | |
85 | ||
7a6bbacb SW |
86 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, |
87 | struct buffer_head *bh_result, int create) | |
b3b94faa | 88 | { |
b3b94faa DT |
89 | int error; |
90 | ||
23591256 | 91 | error = gfs2_block_map(inode, lblock, 0, bh_result); |
b3b94faa DT |
92 | if (error) |
93 | return error; | |
de986e85 | 94 | if (!buffer_mapped(bh_result)) |
7a6bbacb SW |
95 | return -EIO; |
96 | return 0; | |
b3b94faa DT |
97 | } |
98 | ||
7a6bbacb SW |
99 | static int gfs2_get_block_direct(struct inode *inode, sector_t lblock, |
100 | struct buffer_head *bh_result, int create) | |
623d9355 | 101 | { |
23591256 | 102 | return gfs2_block_map(inode, lblock, 0, bh_result); |
623d9355 | 103 | } |
7a6bbacb | 104 | |
b3b94faa | 105 | /** |
9ff8ec32 SW |
106 | * gfs2_writepage_common - Common bits of writepage |
107 | * @page: The page to be written | |
108 | * @wbc: The writeback control | |
b3b94faa | 109 | * |
9ff8ec32 | 110 | * Returns: 1 if writepage is ok, otherwise an error code or zero if no error. |
b3b94faa DT |
111 | */ |
112 | ||
9ff8ec32 SW |
113 | static int gfs2_writepage_common(struct page *page, |
114 | struct writeback_control *wbc) | |
b3b94faa | 115 | { |
18ec7d5c | 116 | struct inode *inode = page->mapping->host; |
f4387149 SW |
117 | struct gfs2_inode *ip = GFS2_I(inode); |
118 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
119 | loff_t i_size = i_size_read(inode); |
120 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
121 | unsigned offset; | |
9ff8ec32 | 122 | int ret = -EIO; |
b3b94faa | 123 | |
9ff8ec32 SW |
124 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) |
125 | goto out; | |
126 | ret = 0; | |
5c676f6d | 127 | if (current->journal_info) |
9ff8ec32 | 128 | goto redirty; |
18ec7d5c | 129 | /* Is the page fully outside i_size? (truncate in progress) */ |
9ff8ec32 | 130 | offset = i_size & (PAGE_CACHE_SIZE-1); |
d2d7b8a2 | 131 | if (page->index > end_index || (page->index == end_index && !offset)) { |
18ec7d5c | 132 | page->mapping->a_ops->invalidatepage(page, 0); |
9ff8ec32 | 133 | goto out; |
b3b94faa | 134 | } |
9ff8ec32 SW |
135 | return 1; |
136 | redirty: | |
137 | redirty_page_for_writepage(wbc, page); | |
138 | out: | |
139 | unlock_page(page); | |
140 | return 0; | |
141 | } | |
142 | ||
143 | /** | |
144 | * gfs2_writeback_writepage - Write page for writeback mappings | |
145 | * @page: The page | |
146 | * @wbc: The writeback control | |
147 | * | |
148 | */ | |
149 | ||
150 | static int gfs2_writeback_writepage(struct page *page, | |
151 | struct writeback_control *wbc) | |
152 | { | |
153 | int ret; | |
154 | ||
155 | ret = gfs2_writepage_common(page, wbc); | |
156 | if (ret <= 0) | |
157 | return ret; | |
158 | ||
159 | ret = mpage_writepage(page, gfs2_get_block_noalloc, wbc); | |
160 | if (ret == -EAGAIN) | |
161 | ret = block_write_full_page(page, gfs2_get_block_noalloc, wbc); | |
162 | return ret; | |
163 | } | |
164 | ||
165 | /** | |
166 | * gfs2_ordered_writepage - Write page for ordered data files | |
167 | * @page: The page to write | |
168 | * @wbc: The writeback control | |
169 | * | |
170 | */ | |
171 | ||
172 | static int gfs2_ordered_writepage(struct page *page, | |
173 | struct writeback_control *wbc) | |
174 | { | |
175 | struct inode *inode = page->mapping->host; | |
176 | struct gfs2_inode *ip = GFS2_I(inode); | |
177 | int ret; | |
178 | ||
179 | ret = gfs2_writepage_common(page, wbc); | |
180 | if (ret <= 0) | |
181 | return ret; | |
182 | ||
183 | if (!page_has_buffers(page)) { | |
184 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
185 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
186 | } | |
187 | gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1); | |
188 | return block_write_full_page(page, gfs2_get_block_noalloc, wbc); | |
189 | } | |
190 | ||
191 | /** | |
192 | * gfs2_jdata_writepage - Write complete page | |
193 | * @page: Page to write | |
194 | * | |
195 | * Returns: errno | |
196 | * | |
197 | */ | |
198 | ||
199 | static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
200 | { | |
201 | struct inode *inode = page->mapping->host; | |
202 | struct gfs2_inode *ip = GFS2_I(inode); | |
203 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
204 | int error; | |
205 | int done_trans = 0; | |
206 | ||
207 | error = gfs2_writepage_common(page, wbc); | |
208 | if (error <= 0) | |
209 | return error; | |
b3b94faa | 210 | |
bf36a713 | 211 | if (PageChecked(page)) { |
18ec7d5c SW |
212 | error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0); |
213 | if (error) | |
214 | goto out_ignore; | |
bf36a713 | 215 | ClearPageChecked(page); |
f4387149 SW |
216 | if (!page_has_buffers(page)) { |
217 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
218 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
219 | } | |
18ec7d5c SW |
220 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1); |
221 | done_trans = 1; | |
222 | } | |
7a6bbacb | 223 | error = block_write_full_page(page, gfs2_get_block_noalloc, wbc); |
18ec7d5c SW |
224 | if (done_trans) |
225 | gfs2_trans_end(sdp); | |
b3b94faa | 226 | return error; |
18ec7d5c SW |
227 | |
228 | out_ignore: | |
229 | redirty_page_for_writepage(wbc, page); | |
230 | unlock_page(page); | |
231 | return 0; | |
b3b94faa DT |
232 | } |
233 | ||
a8d638e3 | 234 | /** |
5561093e | 235 | * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk |
a8d638e3 SW |
236 | * @mapping: The mapping to write |
237 | * @wbc: Write-back control | |
238 | * | |
5561093e | 239 | * For the data=writeback case we can already ignore buffer heads |
a8d638e3 SW |
240 | * and write whole extents at once. This is a big reduction in the |
241 | * number of I/O requests we send and the bmap calls we make in this case. | |
242 | */ | |
5561093e SW |
243 | static int gfs2_writeback_writepages(struct address_space *mapping, |
244 | struct writeback_control *wbc) | |
a8d638e3 | 245 | { |
5561093e | 246 | return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc); |
a8d638e3 SW |
247 | } |
248 | ||
b3b94faa DT |
249 | /** |
250 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
251 | * @ip: the inode | |
252 | * @page: the page | |
253 | * | |
254 | * Returns: errno | |
255 | */ | |
256 | ||
257 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) | |
258 | { | |
259 | struct buffer_head *dibh; | |
260 | void *kaddr; | |
261 | int error; | |
262 | ||
bf126aee SW |
263 | /* |
264 | * Due to the order of unstuffing files and ->nopage(), we can be | |
265 | * asked for a zero page in the case of a stuffed file being extended, | |
266 | * so we need to supply one here. It doesn't happen often. | |
267 | */ | |
268 | if (unlikely(page->index)) { | |
2840501a | 269 | zero_user_page(page, 0, PAGE_CACHE_SIZE, KM_USER0); |
bf126aee SW |
270 | return 0; |
271 | } | |
fd88de56 | 272 | |
b3b94faa DT |
273 | error = gfs2_meta_inode_buffer(ip, &dibh); |
274 | if (error) | |
275 | return error; | |
276 | ||
5c4e9e03 | 277 | kaddr = kmap_atomic(page, KM_USER0); |
fd88de56 | 278 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), |
b3b94faa | 279 | ip->i_di.di_size); |
fd88de56 | 280 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); |
c312c4fd | 281 | kunmap_atomic(kaddr, KM_USER0); |
bf126aee | 282 | flush_dcache_page(page); |
b3b94faa | 283 | brelse(dibh); |
b3b94faa DT |
284 | SetPageUptodate(page); |
285 | ||
286 | return 0; | |
287 | } | |
288 | ||
b3b94faa | 289 | |
b3b94faa | 290 | /** |
51ff87bd SW |
291 | * __gfs2_readpage - readpage |
292 | * @file: The file to read a page for | |
b3b94faa DT |
293 | * @page: The page to read |
294 | * | |
51ff87bd SW |
295 | * This is the core of gfs2's readpage. Its used by the internal file |
296 | * reading code as in that case we already hold the glock. Also its | |
297 | * called by gfs2_readpage() once the required lock has been granted. | |
298 | * | |
b3b94faa DT |
299 | */ |
300 | ||
51ff87bd | 301 | static int __gfs2_readpage(void *file, struct page *page) |
b3b94faa | 302 | { |
feaa7bba SW |
303 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
304 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
b3b94faa DT |
305 | int error; |
306 | ||
18ec7d5c | 307 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
308 | error = stuffed_readpage(ip, page); |
309 | unlock_page(page); | |
51ff87bd | 310 | } else { |
18ec7d5c | 311 | error = mpage_readpage(page, gfs2_get_block); |
51ff87bd | 312 | } |
b3b94faa DT |
313 | |
314 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
51ff87bd | 315 | return -EIO; |
b3b94faa | 316 | |
51ff87bd SW |
317 | return error; |
318 | } | |
319 | ||
320 | /** | |
321 | * gfs2_readpage - read a page of a file | |
322 | * @file: The file to read | |
323 | * @page: The page of the file | |
324 | * | |
3cc3f710 | 325 | * This deals with the locking required. We use a trylock in order to |
51ff87bd SW |
326 | * avoid the page lock / glock ordering problems returning AOP_TRUNCATED_PAGE |
327 | * in the event that we are unable to get the lock. | |
328 | */ | |
329 | ||
330 | static int gfs2_readpage(struct file *file, struct page *page) | |
331 | { | |
332 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | |
333 | struct gfs2_holder gh; | |
334 | int error; | |
335 | ||
51ff87bd SW |
336 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh); |
337 | error = gfs2_glock_nq_atime(&gh); | |
338 | if (unlikely(error)) { | |
339 | unlock_page(page); | |
340 | goto out; | |
61a30dcb | 341 | } |
51ff87bd SW |
342 | error = __gfs2_readpage(file, page); |
343 | gfs2_glock_dq(&gh); | |
18ec7d5c | 344 | out: |
51ff87bd | 345 | gfs2_holder_uninit(&gh); |
a13cbe37 | 346 | if (error == GLR_TRYFAILED) { |
a13cbe37 | 347 | yield(); |
51ff87bd | 348 | return AOP_TRUNCATED_PAGE; |
a13cbe37 | 349 | } |
51ff87bd SW |
350 | return error; |
351 | } | |
352 | ||
353 | /** | |
354 | * gfs2_internal_read - read an internal file | |
355 | * @ip: The gfs2 inode | |
356 | * @ra_state: The readahead state (or NULL for no readahead) | |
357 | * @buf: The buffer to fill | |
358 | * @pos: The file position | |
359 | * @size: The amount to read | |
360 | * | |
361 | */ | |
362 | ||
363 | int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state, | |
364 | char *buf, loff_t *pos, unsigned size) | |
365 | { | |
366 | struct address_space *mapping = ip->i_inode.i_mapping; | |
367 | unsigned long index = *pos / PAGE_CACHE_SIZE; | |
368 | unsigned offset = *pos & (PAGE_CACHE_SIZE - 1); | |
369 | unsigned copied = 0; | |
370 | unsigned amt; | |
371 | struct page *page; | |
372 | void *p; | |
373 | ||
374 | do { | |
375 | amt = size - copied; | |
376 | if (offset + size > PAGE_CACHE_SIZE) | |
377 | amt = PAGE_CACHE_SIZE - offset; | |
378 | page = read_cache_page(mapping, index, __gfs2_readpage, NULL); | |
379 | if (IS_ERR(page)) | |
380 | return PTR_ERR(page); | |
381 | p = kmap_atomic(page, KM_USER0); | |
382 | memcpy(buf + copied, p + offset, amt); | |
383 | kunmap_atomic(p, KM_USER0); | |
384 | mark_page_accessed(page); | |
385 | page_cache_release(page); | |
386 | copied += amt; | |
387 | index++; | |
388 | offset = 0; | |
389 | } while(copied < size); | |
390 | (*pos) += size; | |
391 | return size; | |
fd88de56 SW |
392 | } |
393 | ||
fd88de56 SW |
394 | /** |
395 | * gfs2_readpages - Read a bunch of pages at once | |
396 | * | |
397 | * Some notes: | |
398 | * 1. This is only for readahead, so we can simply ignore any things | |
399 | * which are slightly inconvenient (such as locking conflicts between | |
400 | * the page lock and the glock) and return having done no I/O. Its | |
401 | * obviously not something we'd want to do on too regular a basis. | |
402 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 403 | * 2. We don't handle stuffed files here we let readpage do the honours. |
fd88de56 SW |
404 | * 3. mpage_readpages() does most of the heavy lifting in the common case. |
405 | * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places. | |
fd88de56 | 406 | */ |
3cc3f710 | 407 | |
fd88de56 SW |
408 | static int gfs2_readpages(struct file *file, struct address_space *mapping, |
409 | struct list_head *pages, unsigned nr_pages) | |
410 | { | |
411 | struct inode *inode = mapping->host; | |
feaa7bba SW |
412 | struct gfs2_inode *ip = GFS2_I(inode); |
413 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 | 414 | struct gfs2_holder gh; |
3cc3f710 | 415 | int ret; |
fd88de56 | 416 | |
3cc3f710 | 417 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh); |
51ff87bd | 418 | ret = gfs2_glock_nq_atime(&gh); |
51ff87bd | 419 | if (unlikely(ret)) |
3cc3f710 | 420 | goto out_uninit; |
e1d5b18a | 421 | if (!gfs2_is_stuffed(ip)) |
fd88de56 | 422 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block); |
3cc3f710 SW |
423 | gfs2_glock_dq(&gh); |
424 | out_uninit: | |
425 | gfs2_holder_uninit(&gh); | |
fd88de56 SW |
426 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
427 | ret = -EIO; | |
428 | return ret; | |
b3b94faa DT |
429 | } |
430 | ||
431 | /** | |
7765ec26 | 432 | * gfs2_write_begin - Begin to write to a file |
b3b94faa | 433 | * @file: The file to write to |
7765ec26 SW |
434 | * @mapping: The mapping in which to write |
435 | * @pos: The file offset at which to start writing | |
436 | * @len: Length of the write | |
437 | * @flags: Various flags | |
438 | * @pagep: Pointer to return the page | |
439 | * @fsdata: Pointer to return fs data (unused by GFS2) | |
b3b94faa DT |
440 | * |
441 | * Returns: errno | |
442 | */ | |
443 | ||
7765ec26 SW |
444 | static int gfs2_write_begin(struct file *file, struct address_space *mapping, |
445 | loff_t pos, unsigned len, unsigned flags, | |
446 | struct page **pagep, void **fsdata) | |
b3b94faa | 447 | { |
7765ec26 SW |
448 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
449 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
18ec7d5c SW |
450 | unsigned int data_blocks, ind_blocks, rblocks; |
451 | int alloc_required; | |
b3b94faa | 452 | int error = 0; |
18ec7d5c | 453 | struct gfs2_alloc *al; |
7765ec26 SW |
454 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
455 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
456 | unsigned to = from + len; | |
457 | struct page *page; | |
52ae7b79 | 458 | |
7765ec26 | 459 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh); |
dcd24799 | 460 | error = gfs2_glock_nq_atime(&ip->i_gh); |
7765ec26 | 461 | if (unlikely(error)) |
18ec7d5c | 462 | goto out_uninit; |
b3b94faa | 463 | |
7765ec26 SW |
464 | error = -ENOMEM; |
465 | page = __grab_cache_page(mapping, index); | |
466 | *pagep = page; | |
467 | if (!page) | |
468 | goto out_unlock; | |
469 | ||
470 | gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks); | |
18ec7d5c | 471 | |
7765ec26 | 472 | error = gfs2_write_alloc_required(ip, pos, len, &alloc_required); |
18ec7d5c | 473 | if (error) |
7765ec26 | 474 | goto out_putpage; |
b3b94faa | 475 | |
18ec7d5c | 476 | |
f5c54804 | 477 | ip->i_alloc.al_requested = 0; |
18ec7d5c SW |
478 | if (alloc_required) { |
479 | al = gfs2_alloc_get(ip); | |
480 | ||
481 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
482 | if (error) | |
483 | goto out_alloc_put; | |
484 | ||
2933f925 | 485 | error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid); |
18ec7d5c SW |
486 | if (error) |
487 | goto out_qunlock; | |
488 | ||
489 | al->al_requested = data_blocks + ind_blocks; | |
490 | error = gfs2_inplace_reserve(ip); | |
491 | if (error) | |
492 | goto out_qunlock; | |
493 | } | |
494 | ||
495 | rblocks = RES_DINODE + ind_blocks; | |
496 | if (gfs2_is_jdata(ip)) | |
497 | rblocks += data_blocks ? data_blocks : 1; | |
498 | if (ind_blocks || data_blocks) | |
499 | rblocks += RES_STATFS + RES_QUOTA; | |
500 | ||
16615be1 SW |
501 | error = gfs2_trans_begin(sdp, rblocks, |
502 | PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); | |
18ec7d5c | 503 | if (error) |
a867bb28 | 504 | goto out_trans_fail; |
18ec7d5c SW |
505 | |
506 | if (gfs2_is_stuffed(ip)) { | |
7765ec26 | 507 | if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { |
f25ef0c1 | 508 | error = gfs2_unstuff_dinode(ip, page); |
5c4e9e03 SW |
509 | if (error == 0) |
510 | goto prepare_write; | |
511 | } else if (!PageUptodate(page)) | |
b3b94faa | 512 | error = stuffed_readpage(ip, page); |
5c4e9e03 | 513 | goto out; |
18ec7d5c SW |
514 | } |
515 | ||
5c4e9e03 | 516 | prepare_write: |
18ec7d5c SW |
517 | error = block_prepare_write(page, from, to, gfs2_get_block); |
518 | ||
519 | out: | |
520 | if (error) { | |
521 | gfs2_trans_end(sdp); | |
a867bb28 | 522 | out_trans_fail: |
18ec7d5c SW |
523 | if (alloc_required) { |
524 | gfs2_inplace_release(ip); | |
525 | out_qunlock: | |
526 | gfs2_quota_unlock(ip); | |
527 | out_alloc_put: | |
528 | gfs2_alloc_put(ip); | |
529 | } | |
7765ec26 SW |
530 | out_putpage: |
531 | page_cache_release(page); | |
532 | if (pos + len > ip->i_inode.i_size) | |
533 | vmtruncate(&ip->i_inode, ip->i_inode.i_size); | |
18ec7d5c SW |
534 | out_unlock: |
535 | gfs2_glock_dq_m(1, &ip->i_gh); | |
536 | out_uninit: | |
537 | gfs2_holder_uninit(&ip->i_gh); | |
538 | } | |
b3b94faa DT |
539 | |
540 | return error; | |
541 | } | |
542 | ||
7ae8fa84 RP |
543 | /** |
544 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow | |
545 | * @inode: the rindex inode | |
546 | */ | |
547 | static void adjust_fs_space(struct inode *inode) | |
548 | { | |
549 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
550 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; | |
551 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
552 | u64 fs_total, new_free; | |
553 | ||
554 | /* Total up the file system space, according to the latest rindex. */ | |
555 | fs_total = gfs2_ri_total(sdp); | |
556 | ||
557 | spin_lock(&sdp->sd_statfs_spin); | |
558 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) | |
559 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); | |
560 | else | |
561 | new_free = 0; | |
562 | spin_unlock(&sdp->sd_statfs_spin); | |
6c53267f RP |
563 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
564 | (unsigned long long)new_free); | |
7ae8fa84 RP |
565 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
566 | } | |
567 | ||
b3b94faa | 568 | /** |
7765ec26 SW |
569 | * gfs2_stuffed_write_end - Write end for stuffed files |
570 | * @inode: The inode | |
571 | * @dibh: The buffer_head containing the on-disk inode | |
572 | * @pos: The file position | |
573 | * @len: The length of the write | |
574 | * @copied: How much was actually copied by the VFS | |
575 | * @page: The page | |
576 | * | |
577 | * This copies the data from the page into the inode block after | |
578 | * the inode data structure itself. | |
579 | * | |
580 | * Returns: errno | |
581 | */ | |
582 | static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh, | |
583 | loff_t pos, unsigned len, unsigned copied, | |
584 | struct page *page) | |
585 | { | |
586 | struct gfs2_inode *ip = GFS2_I(inode); | |
587 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
588 | u64 to = pos + copied; | |
589 | void *kaddr; | |
590 | unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode); | |
591 | struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data; | |
592 | ||
593 | BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode))); | |
594 | kaddr = kmap_atomic(page, KM_USER0); | |
595 | memcpy(buf + pos, kaddr + pos, copied); | |
596 | memset(kaddr + pos + copied, 0, len - copied); | |
597 | flush_dcache_page(page); | |
598 | kunmap_atomic(kaddr, KM_USER0); | |
599 | ||
600 | if (!PageUptodate(page)) | |
601 | SetPageUptodate(page); | |
602 | unlock_page(page); | |
603 | page_cache_release(page); | |
604 | ||
605 | if (inode->i_size < to) { | |
606 | i_size_write(inode, to); | |
607 | ip->i_di.di_size = inode->i_size; | |
608 | di->di_size = cpu_to_be64(inode->i_size); | |
609 | mark_inode_dirty(inode); | |
610 | } | |
611 | ||
612 | if (inode == sdp->sd_rindex) | |
613 | adjust_fs_space(inode); | |
614 | ||
615 | brelse(dibh); | |
616 | gfs2_trans_end(sdp); | |
617 | gfs2_glock_dq(&ip->i_gh); | |
618 | gfs2_holder_uninit(&ip->i_gh); | |
619 | return copied; | |
620 | } | |
621 | ||
622 | /** | |
623 | * gfs2_write_end | |
b3b94faa | 624 | * @file: The file to write to |
7765ec26 SW |
625 | * @mapping: The address space to write to |
626 | * @pos: The file position | |
627 | * @len: The length of the data | |
628 | * @copied: | |
629 | * @page: The page that has been written | |
630 | * @fsdata: The fsdata (unused in GFS2) | |
631 | * | |
632 | * The main write_end function for GFS2. We have a separate one for | |
633 | * stuffed files as they are slightly different, otherwise we just | |
634 | * put our locking around the VFS provided functions. | |
b3b94faa DT |
635 | * |
636 | * Returns: errno | |
637 | */ | |
638 | ||
7765ec26 SW |
639 | static int gfs2_write_end(struct file *file, struct address_space *mapping, |
640 | loff_t pos, unsigned len, unsigned copied, | |
641 | struct page *page, void *fsdata) | |
b3b94faa DT |
642 | { |
643 | struct inode *inode = page->mapping->host; | |
feaa7bba SW |
644 | struct gfs2_inode *ip = GFS2_I(inode); |
645 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c | 646 | struct buffer_head *dibh; |
48516ced SW |
647 | struct gfs2_alloc *al = &ip->i_alloc; |
648 | struct gfs2_dinode *di; | |
7765ec26 SW |
649 | unsigned int from = pos & (PAGE_CACHE_SIZE - 1); |
650 | unsigned int to = from + len; | |
651 | int ret; | |
b3b94faa | 652 | |
7765ec26 | 653 | BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == 0); |
18ec7d5c | 654 | |
7765ec26 SW |
655 | ret = gfs2_meta_inode_buffer(ip, &dibh); |
656 | if (unlikely(ret)) { | |
657 | unlock_page(page); | |
658 | page_cache_release(page); | |
659 | goto failed; | |
660 | } | |
18ec7d5c SW |
661 | |
662 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
b3b94faa | 663 | |
7765ec26 SW |
664 | if (gfs2_is_stuffed(ip)) |
665 | return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page); | |
b3b94faa | 666 | |
bf36a713 | 667 | if (!gfs2_is_writeback(ip)) |
7765ec26 | 668 | gfs2_page_add_databufs(ip, page, from, to); |
b3b94faa | 669 | |
7765ec26 | 670 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
b3b94faa | 671 | |
7765ec26 SW |
672 | if (likely(ret >= 0)) { |
673 | copied = ret; | |
674 | if ((pos + copied) > inode->i_size) { | |
675 | di = (struct gfs2_dinode *)dibh->b_data; | |
676 | ip->i_di.di_size = inode->i_size; | |
677 | di->di_size = cpu_to_be64(inode->i_size); | |
ae619320 SW |
678 | mark_inode_dirty(inode); |
679 | } | |
48516ced SW |
680 | } |
681 | ||
7ae8fa84 RP |
682 | if (inode == sdp->sd_rindex) |
683 | adjust_fs_space(inode); | |
684 | ||
18ec7d5c SW |
685 | brelse(dibh); |
686 | gfs2_trans_end(sdp); | |
7765ec26 | 687 | failed: |
18ec7d5c SW |
688 | if (al->al_requested) { |
689 | gfs2_inplace_release(ip); | |
690 | gfs2_quota_unlock(ip); | |
691 | gfs2_alloc_put(ip); | |
692 | } | |
7765ec26 | 693 | gfs2_glock_dq(&ip->i_gh); |
18ec7d5c | 694 | gfs2_holder_uninit(&ip->i_gh); |
7765ec26 | 695 | return ret; |
b3b94faa DT |
696 | } |
697 | ||
8fb68595 RP |
698 | /** |
699 | * gfs2_set_page_dirty - Page dirtying function | |
700 | * @page: The page to dirty | |
701 | * | |
702 | * Returns: 1 if it dirtyed the page, or 0 otherwise | |
703 | */ | |
704 | ||
705 | static int gfs2_set_page_dirty(struct page *page) | |
706 | { | |
5561093e | 707 | SetPageChecked(page); |
8fb68595 RP |
708 | return __set_page_dirty_buffers(page); |
709 | } | |
710 | ||
b3b94faa DT |
711 | /** |
712 | * gfs2_bmap - Block map function | |
713 | * @mapping: Address space info | |
714 | * @lblock: The block to map | |
715 | * | |
716 | * Returns: The disk address for the block or 0 on hole or error | |
717 | */ | |
718 | ||
719 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
720 | { | |
feaa7bba | 721 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
722 | struct gfs2_holder i_gh; |
723 | sector_t dblock = 0; | |
724 | int error; | |
725 | ||
b3b94faa DT |
726 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
727 | if (error) | |
728 | return 0; | |
729 | ||
730 | if (!gfs2_is_stuffed(ip)) | |
4ff14670 | 731 | dblock = generic_block_bmap(mapping, lblock, gfs2_get_block); |
b3b94faa DT |
732 | |
733 | gfs2_glock_dq_uninit(&i_gh); | |
734 | ||
735 | return dblock; | |
736 | } | |
737 | ||
d7b616e2 SW |
738 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) |
739 | { | |
740 | struct gfs2_bufdata *bd; | |
741 | ||
742 | lock_buffer(bh); | |
743 | gfs2_log_lock(sdp); | |
744 | clear_buffer_dirty(bh); | |
745 | bd = bh->b_private; | |
746 | if (bd) { | |
16615be1 SW |
747 | if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh)) |
748 | list_del_init(&bd->bd_le.le_list); | |
749 | else | |
750 | gfs2_remove_from_journal(bh, current->journal_info, 0); | |
d7b616e2 SW |
751 | } |
752 | bh->b_bdev = NULL; | |
753 | clear_buffer_mapped(bh); | |
754 | clear_buffer_req(bh); | |
755 | clear_buffer_new(bh); | |
756 | gfs2_log_unlock(sdp); | |
757 | unlock_buffer(bh); | |
758 | } | |
759 | ||
8628de05 | 760 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) |
b3b94faa | 761 | { |
d7b616e2 SW |
762 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
763 | struct buffer_head *bh, *head; | |
764 | unsigned long pos = 0; | |
765 | ||
b3b94faa | 766 | BUG_ON(!PageLocked(page)); |
8fb68595 RP |
767 | if (offset == 0) |
768 | ClearPageChecked(page); | |
d7b616e2 SW |
769 | if (!page_has_buffers(page)) |
770 | goto out; | |
b3b94faa | 771 | |
d7b616e2 SW |
772 | bh = head = page_buffers(page); |
773 | do { | |
774 | if (offset <= pos) | |
775 | gfs2_discard(sdp, bh); | |
776 | pos += bh->b_size; | |
777 | bh = bh->b_this_page; | |
778 | } while (bh != head); | |
779 | out: | |
780 | if (offset == 0) | |
781 | try_to_release_page(page, 0); | |
b3b94faa DT |
782 | } |
783 | ||
c7b33834 SW |
784 | /** |
785 | * gfs2_ok_for_dio - check that dio is valid on this file | |
786 | * @ip: The inode | |
787 | * @rw: READ or WRITE | |
788 | * @offset: The offset at which we are reading or writing | |
789 | * | |
790 | * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o) | |
791 | * 1 (to accept the i/o request) | |
792 | */ | |
793 | static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset) | |
794 | { | |
795 | /* | |
796 | * Should we return an error here? I can't see that O_DIRECT for | |
5561093e SW |
797 | * a stuffed file makes any sense. For now we'll silently fall |
798 | * back to buffered I/O | |
c7b33834 | 799 | */ |
c7b33834 SW |
800 | if (gfs2_is_stuffed(ip)) |
801 | return 0; | |
802 | ||
803 | if (offset > i_size_read(&ip->i_inode)) | |
804 | return 0; | |
805 | return 1; | |
806 | } | |
807 | ||
808 | ||
809 | ||
a9e5f4d0 SW |
810 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, |
811 | const struct iovec *iov, loff_t offset, | |
812 | unsigned long nr_segs) | |
d1665e41 SW |
813 | { |
814 | struct file *file = iocb->ki_filp; | |
815 | struct inode *inode = file->f_mapping->host; | |
feaa7bba | 816 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
817 | struct gfs2_holder gh; |
818 | int rv; | |
819 | ||
820 | /* | |
c7b33834 SW |
821 | * Deferred lock, even if its a write, since we do no allocation |
822 | * on this path. All we need change is atime, and this lock mode | |
823 | * ensures that other nodes have flushed their buffered read caches | |
824 | * (i.e. their page cache entries for this inode). We do not, | |
825 | * unfortunately have the option of only flushing a range like | |
826 | * the VFS does. | |
d1665e41 | 827 | */ |
c7b33834 | 828 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh); |
dcd24799 | 829 | rv = gfs2_glock_nq_atime(&gh); |
d1665e41 | 830 | if (rv) |
c7b33834 SW |
831 | return rv; |
832 | rv = gfs2_ok_for_dio(ip, rw, offset); | |
833 | if (rv != 1) | |
834 | goto out; /* dio not valid, fall back to buffered i/o */ | |
835 | ||
836 | rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev, | |
837 | iov, offset, nr_segs, | |
838 | gfs2_get_block_direct, NULL); | |
d1665e41 SW |
839 | out: |
840 | gfs2_glock_dq_m(1, &gh); | |
841 | gfs2_holder_uninit(&gh); | |
d1665e41 SW |
842 | return rv; |
843 | } | |
844 | ||
4340fe62 | 845 | /** |
623d9355 | 846 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
847 | * @page: the page that's being released |
848 | * @gfp_mask: passed from Linux VFS, ignored by us | |
849 | * | |
850 | * Call try_to_free_buffers() if the buffers in this page can be | |
851 | * released. | |
852 | * | |
853 | * Returns: 0 | |
854 | */ | |
855 | ||
856 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
857 | { | |
858 | struct inode *aspace = page->mapping->host; | |
859 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | |
860 | struct buffer_head *bh, *head; | |
861 | struct gfs2_bufdata *bd; | |
4340fe62 SW |
862 | |
863 | if (!page_has_buffers(page)) | |
891ba6d4 | 864 | return 0; |
4340fe62 | 865 | |
bb3b0e3d | 866 | gfs2_log_lock(sdp); |
4340fe62 SW |
867 | head = bh = page_buffers(page); |
868 | do { | |
bb3b0e3d SW |
869 | if (atomic_read(&bh->b_count)) |
870 | goto cannot_release; | |
871 | bd = bh->b_private; | |
872 | if (bd && bd->bd_ail) | |
873 | goto cannot_release; | |
4340fe62 | 874 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); |
623d9355 | 875 | gfs2_assert_warn(sdp, !buffer_dirty(bh)); |
bb3b0e3d SW |
876 | bh = bh->b_this_page; |
877 | } while(bh != head); | |
878 | gfs2_log_unlock(sdp); | |
4340fe62 | 879 | |
bb3b0e3d SW |
880 | head = bh = page_buffers(page); |
881 | do { | |
623d9355 | 882 | gfs2_log_lock(sdp); |
4340fe62 SW |
883 | bd = bh->b_private; |
884 | if (bd) { | |
885 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
886 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
d7b616e2 SW |
887 | if (!list_empty(&bd->bd_le.le_list)) { |
888 | if (!buffer_pinned(bh)) | |
889 | list_del_init(&bd->bd_le.le_list); | |
890 | else | |
891 | bd = NULL; | |
892 | } | |
893 | if (bd) | |
894 | bd->bd_bh = NULL; | |
4340fe62 SW |
895 | bh->b_private = NULL; |
896 | } | |
623d9355 SW |
897 | gfs2_log_unlock(sdp); |
898 | if (bd) | |
899 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
4340fe62 SW |
900 | |
901 | bh = bh->b_this_page; | |
166afccd | 902 | } while (bh != head); |
4340fe62 | 903 | |
4340fe62 | 904 | return try_to_free_buffers(page); |
bb3b0e3d SW |
905 | cannot_release: |
906 | gfs2_log_unlock(sdp); | |
907 | return 0; | |
4340fe62 SW |
908 | } |
909 | ||
5561093e | 910 | static const struct address_space_operations gfs2_writeback_aops = { |
9ff8ec32 | 911 | .writepage = gfs2_writeback_writepage, |
5561093e SW |
912 | .writepages = gfs2_writeback_writepages, |
913 | .readpage = gfs2_readpage, | |
914 | .readpages = gfs2_readpages, | |
915 | .sync_page = block_sync_page, | |
916 | .write_begin = gfs2_write_begin, | |
917 | .write_end = gfs2_write_end, | |
918 | .bmap = gfs2_bmap, | |
919 | .invalidatepage = gfs2_invalidatepage, | |
920 | .releasepage = gfs2_releasepage, | |
921 | .direct_IO = gfs2_direct_IO, | |
922 | }; | |
923 | ||
924 | static const struct address_space_operations gfs2_ordered_aops = { | |
9ff8ec32 | 925 | .writepage = gfs2_ordered_writepage, |
b3b94faa | 926 | .readpage = gfs2_readpage, |
fd88de56 | 927 | .readpages = gfs2_readpages, |
b3b94faa | 928 | .sync_page = block_sync_page, |
7765ec26 SW |
929 | .write_begin = gfs2_write_begin, |
930 | .write_end = gfs2_write_end, | |
8fb68595 | 931 | .set_page_dirty = gfs2_set_page_dirty, |
b3b94faa DT |
932 | .bmap = gfs2_bmap, |
933 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 934 | .releasepage = gfs2_releasepage, |
b3b94faa DT |
935 | .direct_IO = gfs2_direct_IO, |
936 | }; | |
937 | ||
5561093e | 938 | static const struct address_space_operations gfs2_jdata_aops = { |
9ff8ec32 | 939 | .writepage = gfs2_jdata_writepage, |
5561093e SW |
940 | .readpage = gfs2_readpage, |
941 | .readpages = gfs2_readpages, | |
942 | .sync_page = block_sync_page, | |
943 | .write_begin = gfs2_write_begin, | |
944 | .write_end = gfs2_write_end, | |
945 | .set_page_dirty = gfs2_set_page_dirty, | |
946 | .bmap = gfs2_bmap, | |
947 | .invalidatepage = gfs2_invalidatepage, | |
948 | .releasepage = gfs2_releasepage, | |
949 | }; | |
950 | ||
951 | void gfs2_set_aops(struct inode *inode) | |
952 | { | |
953 | struct gfs2_inode *ip = GFS2_I(inode); | |
954 | ||
955 | if (gfs2_is_writeback(ip)) | |
956 | inode->i_mapping->a_ops = &gfs2_writeback_aops; | |
957 | else if (gfs2_is_ordered(ip)) | |
958 | inode->i_mapping->a_ops = &gfs2_ordered_aops; | |
959 | else if (gfs2_is_jdata(ip)) | |
960 | inode->i_mapping->a_ops = &gfs2_jdata_aops; | |
961 | else | |
962 | BUG(); | |
963 | } | |
964 |