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> |
5c676f6d | 20 | #include <linux/gfs2_ondisk.h> |
7d308590 | 21 | #include <linux/lm_interface.h> |
b3b94faa DT |
22 | |
23 | #include "gfs2.h" | |
5c676f6d | 24 | #include "incore.h" |
b3b94faa DT |
25 | #include "bmap.h" |
26 | #include "glock.h" | |
27 | #include "inode.h" | |
b3b94faa DT |
28 | #include "log.h" |
29 | #include "meta_io.h" | |
30 | #include "ops_address.h" | |
b3b94faa DT |
31 | #include "quota.h" |
32 | #include "trans.h" | |
18ec7d5c | 33 | #include "rgrp.h" |
61a30dcb | 34 | #include "ops_file.h" |
cd81a4ba | 35 | #include "super.h" |
5c676f6d | 36 | #include "util.h" |
4340fe62 | 37 | #include "glops.h" |
b3b94faa | 38 | |
ba7f7290 SW |
39 | |
40 | static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page, | |
41 | unsigned int from, unsigned int to) | |
42 | { | |
43 | struct buffer_head *head = page_buffers(page); | |
44 | unsigned int bsize = head->b_size; | |
45 | struct buffer_head *bh; | |
46 | unsigned int start, end; | |
47 | ||
48 | for (bh = head, start = 0; bh != head || !start; | |
49 | bh = bh->b_this_page, start = end) { | |
50 | end = start + bsize; | |
51 | if (end <= from || start >= to) | |
52 | continue; | |
ddf4b426 BM |
53 | if (gfs2_is_jdata(ip)) |
54 | set_buffer_uptodate(bh); | |
ba7f7290 SW |
55 | gfs2_trans_add_bh(ip->i_gl, bh, 0); |
56 | } | |
57 | } | |
58 | ||
b3b94faa | 59 | /** |
4ff14670 | 60 | * gfs2_get_block - Fills in a buffer head with details about a block |
b3b94faa DT |
61 | * @inode: The inode |
62 | * @lblock: The block number to look up | |
63 | * @bh_result: The buffer head to return the result in | |
64 | * @create: Non-zero if we may add block to the file | |
65 | * | |
66 | * Returns: errno | |
67 | */ | |
68 | ||
4ff14670 SW |
69 | int gfs2_get_block(struct inode *inode, sector_t lblock, |
70 | struct buffer_head *bh_result, int create) | |
b3b94faa | 71 | { |
23591256 | 72 | return gfs2_block_map(inode, lblock, create, bh_result); |
b3b94faa DT |
73 | } |
74 | ||
75 | /** | |
7a6bbacb | 76 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block |
b3b94faa DT |
77 | * @inode: The inode |
78 | * @lblock: The block number to look up | |
79 | * @bh_result: The buffer head to return the result in | |
80 | * @create: Non-zero if we may add block to the file | |
81 | * | |
82 | * Returns: errno | |
83 | */ | |
84 | ||
7a6bbacb SW |
85 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, |
86 | struct buffer_head *bh_result, int create) | |
b3b94faa | 87 | { |
b3b94faa DT |
88 | int error; |
89 | ||
23591256 | 90 | error = gfs2_block_map(inode, lblock, 0, bh_result); |
b3b94faa DT |
91 | if (error) |
92 | return error; | |
7a6bbacb SW |
93 | if (bh_result->b_blocknr == 0) |
94 | return -EIO; | |
95 | return 0; | |
b3b94faa DT |
96 | } |
97 | ||
7a6bbacb SW |
98 | static int gfs2_get_block_direct(struct inode *inode, sector_t lblock, |
99 | struct buffer_head *bh_result, int create) | |
623d9355 | 100 | { |
23591256 | 101 | return gfs2_block_map(inode, lblock, 0, bh_result); |
623d9355 | 102 | } |
7a6bbacb | 103 | |
b3b94faa DT |
104 | /** |
105 | * gfs2_writepage - Write complete page | |
106 | * @page: Page to write | |
107 | * | |
108 | * Returns: errno | |
109 | * | |
18ec7d5c SW |
110 | * Some of this is copied from block_write_full_page() although we still |
111 | * call it to do most of the work. | |
b3b94faa DT |
112 | */ |
113 | ||
114 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) | |
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; | |
b3b94faa | 122 | int error; |
18ec7d5c | 123 | int done_trans = 0; |
b3b94faa | 124 | |
b3b94faa DT |
125 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) { |
126 | unlock_page(page); | |
127 | return -EIO; | |
128 | } | |
5c676f6d | 129 | if (current->journal_info) |
18ec7d5c SW |
130 | goto out_ignore; |
131 | ||
132 | /* Is the page fully outside i_size? (truncate in progress) */ | |
133 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
d2d7b8a2 | 134 | if (page->index > end_index || (page->index == end_index && !offset)) { |
18ec7d5c | 135 | page->mapping->a_ops->invalidatepage(page, 0); |
b3b94faa | 136 | unlock_page(page); |
18ec7d5c | 137 | return 0; /* don't care */ |
b3b94faa DT |
138 | } |
139 | ||
8fb68595 RP |
140 | if ((sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) && |
141 | PageChecked(page)) { | |
142 | ClearPageChecked(page); | |
18ec7d5c SW |
143 | error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0); |
144 | if (error) | |
145 | goto out_ignore; | |
f4387149 SW |
146 | if (!page_has_buffers(page)) { |
147 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
148 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
149 | } | |
18ec7d5c SW |
150 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1); |
151 | done_trans = 1; | |
152 | } | |
7a6bbacb | 153 | error = block_write_full_page(page, gfs2_get_block_noalloc, wbc); |
18ec7d5c SW |
154 | if (done_trans) |
155 | gfs2_trans_end(sdp); | |
b3b94faa | 156 | gfs2_meta_cache_flush(ip); |
b3b94faa | 157 | return error; |
18ec7d5c SW |
158 | |
159 | out_ignore: | |
160 | redirty_page_for_writepage(wbc, page); | |
161 | unlock_page(page); | |
162 | return 0; | |
b3b94faa DT |
163 | } |
164 | ||
a8d638e3 SW |
165 | /** |
166 | * gfs2_writepages - Write a bunch of dirty pages back to disk | |
167 | * @mapping: The mapping to write | |
168 | * @wbc: Write-back control | |
169 | * | |
170 | * For journaled files and/or ordered writes this just falls back to the | |
171 | * kernel's default writepages path for now. We will probably want to change | |
172 | * that eventually (i.e. when we look at allocate on flush). | |
173 | * | |
174 | * For the data=writeback case though we can already ignore buffer heads | |
175 | * and write whole extents at once. This is a big reduction in the | |
176 | * number of I/O requests we send and the bmap calls we make in this case. | |
177 | */ | |
a2cf8222 AB |
178 | static int gfs2_writepages(struct address_space *mapping, |
179 | struct writeback_control *wbc) | |
a8d638e3 SW |
180 | { |
181 | struct inode *inode = mapping->host; | |
182 | struct gfs2_inode *ip = GFS2_I(inode); | |
183 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
184 | ||
185 | if (sdp->sd_args.ar_data == GFS2_DATA_WRITEBACK && !gfs2_is_jdata(ip)) | |
186 | return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc); | |
187 | ||
188 | return generic_writepages(mapping, wbc); | |
189 | } | |
190 | ||
b3b94faa DT |
191 | /** |
192 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
193 | * @ip: the inode | |
194 | * @page: the page | |
195 | * | |
196 | * Returns: errno | |
197 | */ | |
198 | ||
199 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) | |
200 | { | |
201 | struct buffer_head *dibh; | |
202 | void *kaddr; | |
203 | int error; | |
204 | ||
bf126aee SW |
205 | /* |
206 | * Due to the order of unstuffing files and ->nopage(), we can be | |
207 | * asked for a zero page in the case of a stuffed file being extended, | |
208 | * so we need to supply one here. It doesn't happen often. | |
209 | */ | |
210 | if (unlikely(page->index)) { | |
2840501a | 211 | zero_user_page(page, 0, PAGE_CACHE_SIZE, KM_USER0); |
bf126aee SW |
212 | return 0; |
213 | } | |
fd88de56 | 214 | |
b3b94faa DT |
215 | error = gfs2_meta_inode_buffer(ip, &dibh); |
216 | if (error) | |
217 | return error; | |
218 | ||
5c4e9e03 | 219 | kaddr = kmap_atomic(page, KM_USER0); |
fd88de56 | 220 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), |
b3b94faa | 221 | ip->i_di.di_size); |
fd88de56 | 222 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); |
c312c4fd | 223 | kunmap_atomic(kaddr, KM_USER0); |
bf126aee | 224 | flush_dcache_page(page); |
b3b94faa | 225 | brelse(dibh); |
b3b94faa DT |
226 | SetPageUptodate(page); |
227 | ||
228 | return 0; | |
229 | } | |
230 | ||
b3b94faa | 231 | |
b3b94faa DT |
232 | /** |
233 | * gfs2_readpage - readpage with locking | |
18ec7d5c SW |
234 | * @file: The file to read a page for. N.B. This may be NULL if we are |
235 | * reading an internal file. | |
b3b94faa DT |
236 | * @page: The page to read |
237 | * | |
238 | * Returns: errno | |
239 | */ | |
240 | ||
241 | static int gfs2_readpage(struct file *file, struct page *page) | |
242 | { | |
feaa7bba SW |
243 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
244 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
dc41aeed | 245 | struct gfs2_file *gf = NULL; |
18ec7d5c | 246 | struct gfs2_holder gh; |
b3b94faa | 247 | int error; |
59a1cc6b | 248 | int do_unlock = 0; |
b3b94faa | 249 | |
dd538c83 | 250 | if (likely(file != &gfs2_internal_file_sentinel)) { |
59a1cc6b | 251 | if (file) { |
dc41aeed | 252 | gf = file->private_data; |
59a1cc6b | 253 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) |
54cb8821 | 254 | /* gfs2_sharewrite_fault has grabbed the ip->i_gl already */ |
59a1cc6b SW |
255 | goto skip_lock; |
256 | } | |
2ca99501 | 257 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh); |
59a1cc6b | 258 | do_unlock = 1; |
dcd24799 | 259 | error = gfs2_glock_nq_atime(&gh); |
fd88de56 | 260 | if (unlikely(error)) |
61a30dcb SW |
261 | goto out_unlock; |
262 | } | |
b3b94faa | 263 | |
59a1cc6b | 264 | skip_lock: |
18ec7d5c | 265 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
266 | error = stuffed_readpage(ip, page); |
267 | unlock_page(page); | |
b3b94faa | 268 | } else |
18ec7d5c | 269 | error = mpage_readpage(page, gfs2_get_block); |
b3b94faa DT |
270 | |
271 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
272 | error = -EIO; | |
273 | ||
07903c02 | 274 | if (do_unlock) { |
61a30dcb SW |
275 | gfs2_glock_dq_m(1, &gh); |
276 | gfs2_holder_uninit(&gh); | |
277 | } | |
18ec7d5c | 278 | out: |
b3b94faa | 279 | return error; |
18ec7d5c SW |
280 | out_unlock: |
281 | unlock_page(page); | |
a13cbe37 JW |
282 | if (error == GLR_TRYFAILED) { |
283 | error = AOP_TRUNCATED_PAGE; | |
284 | yield(); | |
285 | } | |
59a1cc6b | 286 | if (do_unlock) |
fd88de56 SW |
287 | gfs2_holder_uninit(&gh); |
288 | goto out; | |
289 | } | |
290 | ||
fd88de56 SW |
291 | /** |
292 | * gfs2_readpages - Read a bunch of pages at once | |
293 | * | |
294 | * Some notes: | |
295 | * 1. This is only for readahead, so we can simply ignore any things | |
296 | * which are slightly inconvenient (such as locking conflicts between | |
297 | * the page lock and the glock) and return having done no I/O. Its | |
298 | * obviously not something we'd want to do on too regular a basis. | |
299 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 300 | * 2. We don't handle stuffed files here we let readpage do the honours. |
fd88de56 SW |
301 | * 3. mpage_readpages() does most of the heavy lifting in the common case. |
302 | * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places. | |
303 | * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as | |
304 | * well as read-ahead. | |
305 | */ | |
306 | static int gfs2_readpages(struct file *file, struct address_space *mapping, | |
307 | struct list_head *pages, unsigned nr_pages) | |
308 | { | |
309 | struct inode *inode = mapping->host; | |
feaa7bba SW |
310 | struct gfs2_inode *ip = GFS2_I(inode); |
311 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 | 312 | struct gfs2_holder gh; |
e1d5b18a | 313 | int ret = 0; |
59a1cc6b | 314 | int do_unlock = 0; |
fd88de56 | 315 | |
dd538c83 | 316 | if (likely(file != &gfs2_internal_file_sentinel)) { |
59a1cc6b SW |
317 | if (file) { |
318 | struct gfs2_file *gf = file->private_data; | |
319 | if (test_bit(GFF_EXLOCK, &gf->f_flags)) | |
320 | goto skip_lock; | |
321 | } | |
fd88de56 | 322 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, |
2ca99501 | 323 | LM_FLAG_TRY_1CB|GL_ATIME, &gh); |
59a1cc6b | 324 | do_unlock = 1; |
dcd24799 | 325 | ret = gfs2_glock_nq_atime(&gh); |
907b9bce | 326 | if (ret == GLR_TRYFAILED) |
fd88de56 SW |
327 | goto out_noerror; |
328 | if (unlikely(ret)) | |
329 | goto out_unlock; | |
330 | } | |
59a1cc6b | 331 | skip_lock: |
e1d5b18a | 332 | if (!gfs2_is_stuffed(ip)) |
fd88de56 | 333 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block); |
fd88de56 | 334 | |
59a1cc6b | 335 | if (do_unlock) { |
fd88de56 SW |
336 | gfs2_glock_dq_m(1, &gh); |
337 | gfs2_holder_uninit(&gh); | |
338 | } | |
339 | out: | |
340 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
341 | ret = -EIO; | |
342 | return ret; | |
343 | out_noerror: | |
344 | ret = 0; | |
345 | out_unlock: | |
59a1cc6b | 346 | if (do_unlock) |
fd88de56 | 347 | gfs2_holder_uninit(&gh); |
18ec7d5c | 348 | goto out; |
b3b94faa DT |
349 | } |
350 | ||
351 | /** | |
352 | * gfs2_prepare_write - Prepare to write a page to a file | |
353 | * @file: The file to write to | |
354 | * @page: The page which is to be prepared for writing | |
355 | * @from: From (byte range within page) | |
356 | * @to: To (byte range within page) | |
357 | * | |
358 | * Returns: errno | |
359 | */ | |
360 | ||
361 | static int gfs2_prepare_write(struct file *file, struct page *page, | |
362 | unsigned from, unsigned to) | |
363 | { | |
feaa7bba SW |
364 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
365 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
18ec7d5c SW |
366 | unsigned int data_blocks, ind_blocks, rblocks; |
367 | int alloc_required; | |
b3b94faa | 368 | int error = 0; |
18ec7d5c SW |
369 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from; |
370 | loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
371 | struct gfs2_alloc *al; | |
52ae7b79 RC |
372 | unsigned int write_len = to - from; |
373 | ||
b3b94faa | 374 | |
2ca99501 | 375 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh); |
dcd24799 | 376 | error = gfs2_glock_nq_atime(&ip->i_gh); |
2ca99501 | 377 | if (unlikely(error)) { |
2d72e710 SW |
378 | if (error == GLR_TRYFAILED) { |
379 | unlock_page(page); | |
2ca99501 | 380 | error = AOP_TRUNCATED_PAGE; |
a13cbe37 | 381 | yield(); |
2d72e710 | 382 | } |
18ec7d5c | 383 | goto out_uninit; |
2ca99501 | 384 | } |
b3b94faa | 385 | |
52ae7b79 | 386 | gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks); |
18ec7d5c | 387 | |
52ae7b79 | 388 | error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required); |
18ec7d5c SW |
389 | if (error) |
390 | goto out_unlock; | |
b3b94faa | 391 | |
18ec7d5c | 392 | |
f5c54804 | 393 | ip->i_alloc.al_requested = 0; |
18ec7d5c SW |
394 | if (alloc_required) { |
395 | al = gfs2_alloc_get(ip); | |
396 | ||
397 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
398 | if (error) | |
399 | goto out_alloc_put; | |
400 | ||
2933f925 | 401 | error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid); |
18ec7d5c SW |
402 | if (error) |
403 | goto out_qunlock; | |
404 | ||
405 | al->al_requested = data_blocks + ind_blocks; | |
406 | error = gfs2_inplace_reserve(ip); | |
407 | if (error) | |
408 | goto out_qunlock; | |
409 | } | |
410 | ||
411 | rblocks = RES_DINODE + ind_blocks; | |
412 | if (gfs2_is_jdata(ip)) | |
413 | rblocks += data_blocks ? data_blocks : 1; | |
414 | if (ind_blocks || data_blocks) | |
415 | rblocks += RES_STATFS + RES_QUOTA; | |
416 | ||
417 | error = gfs2_trans_begin(sdp, rblocks, 0); | |
418 | if (error) | |
a867bb28 | 419 | goto out_trans_fail; |
18ec7d5c SW |
420 | |
421 | if (gfs2_is_stuffed(ip)) { | |
422 | if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { | |
f25ef0c1 | 423 | error = gfs2_unstuff_dinode(ip, page); |
5c4e9e03 SW |
424 | if (error == 0) |
425 | goto prepare_write; | |
426 | } else if (!PageUptodate(page)) | |
b3b94faa | 427 | error = stuffed_readpage(ip, page); |
5c4e9e03 | 428 | goto out; |
18ec7d5c SW |
429 | } |
430 | ||
5c4e9e03 | 431 | prepare_write: |
18ec7d5c SW |
432 | error = block_prepare_write(page, from, to, gfs2_get_block); |
433 | ||
434 | out: | |
435 | if (error) { | |
436 | gfs2_trans_end(sdp); | |
a867bb28 | 437 | out_trans_fail: |
18ec7d5c SW |
438 | if (alloc_required) { |
439 | gfs2_inplace_release(ip); | |
440 | out_qunlock: | |
441 | gfs2_quota_unlock(ip); | |
442 | out_alloc_put: | |
443 | gfs2_alloc_put(ip); | |
444 | } | |
445 | out_unlock: | |
446 | gfs2_glock_dq_m(1, &ip->i_gh); | |
447 | out_uninit: | |
448 | gfs2_holder_uninit(&ip->i_gh); | |
449 | } | |
b3b94faa DT |
450 | |
451 | return error; | |
452 | } | |
453 | ||
7ae8fa84 RP |
454 | /** |
455 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow | |
456 | * @inode: the rindex inode | |
457 | */ | |
458 | static void adjust_fs_space(struct inode *inode) | |
459 | { | |
460 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
461 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; | |
462 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
463 | u64 fs_total, new_free; | |
464 | ||
465 | /* Total up the file system space, according to the latest rindex. */ | |
466 | fs_total = gfs2_ri_total(sdp); | |
467 | ||
468 | spin_lock(&sdp->sd_statfs_spin); | |
469 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) | |
470 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); | |
471 | else | |
472 | new_free = 0; | |
473 | spin_unlock(&sdp->sd_statfs_spin); | |
6c53267f RP |
474 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
475 | (unsigned long long)new_free); | |
7ae8fa84 RP |
476 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
477 | } | |
478 | ||
b3b94faa DT |
479 | /** |
480 | * gfs2_commit_write - Commit write to a file | |
481 | * @file: The file to write to | |
482 | * @page: The page containing the data | |
483 | * @from: From (byte range within page) | |
484 | * @to: To (byte range within page) | |
485 | * | |
486 | * Returns: errno | |
487 | */ | |
488 | ||
489 | static int gfs2_commit_write(struct file *file, struct page *page, | |
490 | unsigned from, unsigned to) | |
491 | { | |
492 | struct inode *inode = page->mapping->host; | |
feaa7bba SW |
493 | struct gfs2_inode *ip = GFS2_I(inode); |
494 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
495 | int error = -EOPNOTSUPP; |
496 | struct buffer_head *dibh; | |
48516ced SW |
497 | struct gfs2_alloc *al = &ip->i_alloc; |
498 | struct gfs2_dinode *di; | |
b3b94faa | 499 | |
18ec7d5c SW |
500 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl))) |
501 | goto fail_nounlock; | |
502 | ||
503 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
504 | if (error) | |
505 | goto fail_endtrans; | |
506 | ||
507 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
48516ced | 508 | di = (struct gfs2_dinode *)dibh->b_data; |
18ec7d5c | 509 | |
b3b94faa | 510 | if (gfs2_is_stuffed(ip)) { |
cd915493 | 511 | u64 file_size; |
b3b94faa DT |
512 | void *kaddr; |
513 | ||
cd915493 | 514 | file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to; |
b3b94faa | 515 | |
18ec7d5c | 516 | kaddr = kmap_atomic(page, KM_USER0); |
b3b94faa | 517 | memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from, |
0bd5996a | 518 | kaddr + from, to - from); |
c312c4fd | 519 | kunmap_atomic(kaddr, KM_USER0); |
b3b94faa DT |
520 | |
521 | SetPageUptodate(page); | |
522 | ||
ae619320 | 523 | if (inode->i_size < file_size) { |
b3b94faa | 524 | i_size_write(inode, file_size); |
ae619320 SW |
525 | mark_inode_dirty(inode); |
526 | } | |
b3b94faa | 527 | } else { |
568f4c96 SW |
528 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || |
529 | gfs2_is_jdata(ip)) | |
257f9b4e | 530 | gfs2_page_add_databufs(ip, page, from, to); |
b3b94faa DT |
531 | error = generic_commit_write(file, page, from, to); |
532 | if (error) | |
533 | goto fail; | |
534 | } | |
535 | ||
48516ced | 536 | if (ip->i_di.di_size < inode->i_size) { |
18ec7d5c | 537 | ip->i_di.di_size = inode->i_size; |
48516ced SW |
538 | di->di_size = cpu_to_be64(inode->i_size); |
539 | } | |
540 | ||
7ae8fa84 RP |
541 | if (inode == sdp->sd_rindex) |
542 | adjust_fs_space(inode); | |
543 | ||
18ec7d5c SW |
544 | brelse(dibh); |
545 | gfs2_trans_end(sdp); | |
546 | if (al->al_requested) { | |
547 | gfs2_inplace_release(ip); | |
548 | gfs2_quota_unlock(ip); | |
549 | gfs2_alloc_put(ip); | |
550 | } | |
1de91390 | 551 | unlock_page(page); |
18ec7d5c | 552 | gfs2_glock_dq_m(1, &ip->i_gh); |
1de91390 | 553 | lock_page(page); |
18ec7d5c | 554 | gfs2_holder_uninit(&ip->i_gh); |
b3b94faa DT |
555 | return 0; |
556 | ||
18ec7d5c SW |
557 | fail: |
558 | brelse(dibh); | |
559 | fail_endtrans: | |
560 | gfs2_trans_end(sdp); | |
561 | if (al->al_requested) { | |
562 | gfs2_inplace_release(ip); | |
563 | gfs2_quota_unlock(ip); | |
564 | gfs2_alloc_put(ip); | |
565 | } | |
1de91390 | 566 | unlock_page(page); |
18ec7d5c | 567 | gfs2_glock_dq_m(1, &ip->i_gh); |
1de91390 | 568 | lock_page(page); |
18ec7d5c SW |
569 | gfs2_holder_uninit(&ip->i_gh); |
570 | fail_nounlock: | |
b3b94faa | 571 | ClearPageUptodate(page); |
b3b94faa DT |
572 | return error; |
573 | } | |
574 | ||
8fb68595 RP |
575 | /** |
576 | * gfs2_set_page_dirty - Page dirtying function | |
577 | * @page: The page to dirty | |
578 | * | |
579 | * Returns: 1 if it dirtyed the page, or 0 otherwise | |
580 | */ | |
581 | ||
582 | static int gfs2_set_page_dirty(struct page *page) | |
583 | { | |
584 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | |
585 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
586 | ||
587 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) | |
588 | SetPageChecked(page); | |
589 | return __set_page_dirty_buffers(page); | |
590 | } | |
591 | ||
b3b94faa DT |
592 | /** |
593 | * gfs2_bmap - Block map function | |
594 | * @mapping: Address space info | |
595 | * @lblock: The block to map | |
596 | * | |
597 | * Returns: The disk address for the block or 0 on hole or error | |
598 | */ | |
599 | ||
600 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
601 | { | |
feaa7bba | 602 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
603 | struct gfs2_holder i_gh; |
604 | sector_t dblock = 0; | |
605 | int error; | |
606 | ||
b3b94faa DT |
607 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
608 | if (error) | |
609 | return 0; | |
610 | ||
611 | if (!gfs2_is_stuffed(ip)) | |
4ff14670 | 612 | dblock = generic_block_bmap(mapping, lblock, gfs2_get_block); |
b3b94faa DT |
613 | |
614 | gfs2_glock_dq_uninit(&i_gh); | |
615 | ||
616 | return dblock; | |
617 | } | |
618 | ||
619 | static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh) | |
620 | { | |
64fb4eb7 | 621 | struct gfs2_bufdata *bd; |
b3b94faa DT |
622 | |
623 | gfs2_log_lock(sdp); | |
5c676f6d | 624 | bd = bh->b_private; |
64fb4eb7 SW |
625 | if (bd) { |
626 | bd->bd_bh = NULL; | |
5c676f6d | 627 | bh->b_private = NULL; |
8fb68595 RP |
628 | if (!bd->bd_ail && list_empty(&bd->bd_le.le_list)) |
629 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
15d00c0b SW |
630 | } |
631 | gfs2_log_unlock(sdp); | |
b3b94faa DT |
632 | |
633 | lock_buffer(bh); | |
634 | clear_buffer_dirty(bh); | |
635 | bh->b_bdev = NULL; | |
636 | clear_buffer_mapped(bh); | |
637 | clear_buffer_req(bh); | |
638 | clear_buffer_new(bh); | |
639 | clear_buffer_delay(bh); | |
640 | unlock_buffer(bh); | |
641 | } | |
642 | ||
8628de05 | 643 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) |
b3b94faa | 644 | { |
15d00c0b | 645 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
b3b94faa DT |
646 | struct buffer_head *head, *bh, *next; |
647 | unsigned int curr_off = 0; | |
b3b94faa DT |
648 | |
649 | BUG_ON(!PageLocked(page)); | |
8fb68595 RP |
650 | if (offset == 0) |
651 | ClearPageChecked(page); | |
b3b94faa | 652 | if (!page_has_buffers(page)) |
8628de05 | 653 | return; |
b3b94faa DT |
654 | |
655 | bh = head = page_buffers(page); | |
656 | do { | |
657 | unsigned int next_off = curr_off + bh->b_size; | |
658 | next = bh->b_this_page; | |
659 | ||
660 | if (offset <= curr_off) | |
661 | discard_buffer(sdp, bh); | |
662 | ||
663 | curr_off = next_off; | |
664 | bh = next; | |
665 | } while (bh != head); | |
666 | ||
667 | if (!offset) | |
8628de05 | 668 | try_to_release_page(page, 0); |
b3b94faa | 669 | |
8628de05 | 670 | return; |
b3b94faa DT |
671 | } |
672 | ||
c7b33834 SW |
673 | /** |
674 | * gfs2_ok_for_dio - check that dio is valid on this file | |
675 | * @ip: The inode | |
676 | * @rw: READ or WRITE | |
677 | * @offset: The offset at which we are reading or writing | |
678 | * | |
679 | * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o) | |
680 | * 1 (to accept the i/o request) | |
681 | */ | |
682 | static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset) | |
683 | { | |
684 | /* | |
685 | * Should we return an error here? I can't see that O_DIRECT for | |
686 | * a journaled file makes any sense. For now we'll silently fall | |
687 | * back to buffered I/O, likewise we do the same for stuffed | |
688 | * files since they are (a) small and (b) unaligned. | |
689 | */ | |
690 | if (gfs2_is_jdata(ip)) | |
691 | return 0; | |
692 | ||
693 | if (gfs2_is_stuffed(ip)) | |
694 | return 0; | |
695 | ||
696 | if (offset > i_size_read(&ip->i_inode)) | |
697 | return 0; | |
698 | return 1; | |
699 | } | |
700 | ||
701 | ||
702 | ||
a9e5f4d0 SW |
703 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, |
704 | const struct iovec *iov, loff_t offset, | |
705 | unsigned long nr_segs) | |
d1665e41 SW |
706 | { |
707 | struct file *file = iocb->ki_filp; | |
708 | struct inode *inode = file->f_mapping->host; | |
feaa7bba | 709 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
710 | struct gfs2_holder gh; |
711 | int rv; | |
712 | ||
713 | /* | |
c7b33834 SW |
714 | * Deferred lock, even if its a write, since we do no allocation |
715 | * on this path. All we need change is atime, and this lock mode | |
716 | * ensures that other nodes have flushed their buffered read caches | |
717 | * (i.e. their page cache entries for this inode). We do not, | |
718 | * unfortunately have the option of only flushing a range like | |
719 | * the VFS does. | |
d1665e41 | 720 | */ |
c7b33834 | 721 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh); |
dcd24799 | 722 | rv = gfs2_glock_nq_atime(&gh); |
d1665e41 | 723 | if (rv) |
c7b33834 SW |
724 | return rv; |
725 | rv = gfs2_ok_for_dio(ip, rw, offset); | |
726 | if (rv != 1) | |
727 | goto out; /* dio not valid, fall back to buffered i/o */ | |
728 | ||
729 | rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev, | |
730 | iov, offset, nr_segs, | |
731 | gfs2_get_block_direct, NULL); | |
d1665e41 SW |
732 | out: |
733 | gfs2_glock_dq_m(1, &gh); | |
734 | gfs2_holder_uninit(&gh); | |
d1665e41 SW |
735 | return rv; |
736 | } | |
737 | ||
4340fe62 SW |
738 | /** |
739 | * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out. | |
740 | * @bh: the buffer we're stuck on | |
741 | * | |
742 | */ | |
743 | ||
744 | static void stuck_releasepage(struct buffer_head *bh) | |
745 | { | |
746 | struct inode *inode = bh->b_page->mapping->host; | |
747 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
748 | struct gfs2_bufdata *bd = bh->b_private; | |
749 | struct gfs2_glock *gl; | |
166afccd SW |
750 | static unsigned limit = 0; |
751 | ||
0bd5996a | 752 | if (limit > 3) |
166afccd | 753 | return; |
0bd5996a | 754 | limit++; |
4340fe62 SW |
755 | |
756 | fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode); | |
757 | fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n", | |
758 | (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count)); | |
759 | fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh)); | |
760 | fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL"); | |
761 | ||
762 | if (!bd) | |
763 | return; | |
764 | ||
765 | gl = bd->bd_gl; | |
766 | ||
907b9bce | 767 | fs_warn(sdp, "gl = (%u, %llu)\n", |
4340fe62 SW |
768 | gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number); |
769 | ||
770 | fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n", | |
771 | (list_empty(&bd->bd_list_tr)) ? "no" : "yes", | |
772 | (list_empty(&bd->bd_le.le_list)) ? "no" : "yes"); | |
773 | ||
774 | if (gl->gl_ops == &gfs2_inode_glops) { | |
775 | struct gfs2_inode *ip = gl->gl_object; | |
776 | unsigned int x; | |
777 | ||
778 | if (!ip) | |
779 | return; | |
780 | ||
781 | fs_warn(sdp, "ip = %llu %llu\n", | |
dbb7cae2 SW |
782 | (unsigned long long)ip->i_no_formal_ino, |
783 | (unsigned long long)ip->i_no_addr); | |
4340fe62 SW |
784 | |
785 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) | |
786 | fs_warn(sdp, "ip->i_cache[%u] = %s\n", | |
787 | x, (ip->i_cache[x]) ? "!NULL" : "NULL"); | |
788 | } | |
789 | } | |
790 | ||
791 | /** | |
623d9355 | 792 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
793 | * @page: the page that's being released |
794 | * @gfp_mask: passed from Linux VFS, ignored by us | |
795 | * | |
796 | * Call try_to_free_buffers() if the buffers in this page can be | |
797 | * released. | |
798 | * | |
799 | * Returns: 0 | |
800 | */ | |
801 | ||
802 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
803 | { | |
804 | struct inode *aspace = page->mapping->host; | |
805 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | |
806 | struct buffer_head *bh, *head; | |
807 | struct gfs2_bufdata *bd; | |
166afccd | 808 | unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ; |
4340fe62 SW |
809 | |
810 | if (!page_has_buffers(page)) | |
811 | goto out; | |
812 | ||
813 | head = bh = page_buffers(page); | |
814 | do { | |
4340fe62 | 815 | while (atomic_read(&bh->b_count)) { |
166afccd SW |
816 | if (!atomic_read(&aspace->i_writecount)) |
817 | return 0; | |
818 | ||
61057c6b RC |
819 | if (!(gfp_mask & __GFP_WAIT)) |
820 | return 0; | |
821 | ||
166afccd SW |
822 | if (time_after_eq(jiffies, t)) { |
823 | stuck_releasepage(bh); | |
824 | /* should we withdraw here? */ | |
825 | return 0; | |
4340fe62 SW |
826 | } |
827 | ||
166afccd | 828 | yield(); |
4340fe62 SW |
829 | } |
830 | ||
831 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); | |
623d9355 | 832 | gfs2_assert_warn(sdp, !buffer_dirty(bh)); |
4340fe62 | 833 | |
623d9355 | 834 | gfs2_log_lock(sdp); |
4340fe62 SW |
835 | bd = bh->b_private; |
836 | if (bd) { | |
837 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
838 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
4340fe62 | 839 | gfs2_assert_warn(sdp, !bd->bd_ail); |
623d9355 SW |
840 | bd->bd_bh = NULL; |
841 | if (!list_empty(&bd->bd_le.le_list)) | |
842 | bd = NULL; | |
4340fe62 SW |
843 | bh->b_private = NULL; |
844 | } | |
623d9355 SW |
845 | gfs2_log_unlock(sdp); |
846 | if (bd) | |
847 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
4340fe62 SW |
848 | |
849 | bh = bh->b_this_page; | |
166afccd | 850 | } while (bh != head); |
4340fe62 | 851 | |
a9e5f4d0 | 852 | out: |
4340fe62 SW |
853 | return try_to_free_buffers(page); |
854 | } | |
855 | ||
66de045d | 856 | const struct address_space_operations gfs2_file_aops = { |
b3b94faa | 857 | .writepage = gfs2_writepage, |
a8d638e3 | 858 | .writepages = gfs2_writepages, |
b3b94faa | 859 | .readpage = gfs2_readpage, |
fd88de56 | 860 | .readpages = gfs2_readpages, |
b3b94faa DT |
861 | .sync_page = block_sync_page, |
862 | .prepare_write = gfs2_prepare_write, | |
863 | .commit_write = gfs2_commit_write, | |
8fb68595 | 864 | .set_page_dirty = gfs2_set_page_dirty, |
b3b94faa DT |
865 | .bmap = gfs2_bmap, |
866 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 867 | .releasepage = gfs2_releasepage, |
b3b94faa DT |
868 | .direct_IO = gfs2_direct_IO, |
869 | }; | |
870 |