]>
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 DT |
105 | /** |
106 | * gfs2_writepage - Write complete page | |
107 | * @page: Page to write | |
108 | * | |
109 | * Returns: errno | |
110 | * | |
18ec7d5c SW |
111 | * Some of this is copied from block_write_full_page() although we still |
112 | * call it to do most of the work. | |
b3b94faa DT |
113 | */ |
114 | ||
115 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) | |
116 | { | |
18ec7d5c | 117 | struct inode *inode = page->mapping->host; |
f4387149 SW |
118 | struct gfs2_inode *ip = GFS2_I(inode); |
119 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
120 | loff_t i_size = i_size_read(inode); |
121 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
122 | unsigned offset; | |
b3b94faa | 123 | int error; |
18ec7d5c | 124 | int done_trans = 0; |
b3b94faa | 125 | |
b3b94faa DT |
126 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) { |
127 | unlock_page(page); | |
128 | return -EIO; | |
129 | } | |
5c676f6d | 130 | if (current->journal_info) |
18ec7d5c SW |
131 | goto out_ignore; |
132 | ||
133 | /* Is the page fully outside i_size? (truncate in progress) */ | |
134 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
d2d7b8a2 | 135 | if (page->index > end_index || (page->index == end_index && !offset)) { |
18ec7d5c | 136 | page->mapping->a_ops->invalidatepage(page, 0); |
b3b94faa | 137 | unlock_page(page); |
18ec7d5c | 138 | return 0; /* don't care */ |
b3b94faa DT |
139 | } |
140 | ||
bf36a713 | 141 | if (PageChecked(page)) { |
18ec7d5c SW |
142 | error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0); |
143 | if (error) | |
144 | goto out_ignore; | |
bf36a713 | 145 | ClearPageChecked(page); |
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 | return error; |
18ec7d5c SW |
157 | |
158 | out_ignore: | |
159 | redirty_page_for_writepage(wbc, page); | |
160 | unlock_page(page); | |
161 | return 0; | |
b3b94faa DT |
162 | } |
163 | ||
a8d638e3 SW |
164 | /** |
165 | * gfs2_writepages - Write a bunch of dirty pages back to disk | |
166 | * @mapping: The mapping to write | |
167 | * @wbc: Write-back control | |
168 | * | |
169 | * For journaled files and/or ordered writes this just falls back to the | |
170 | * kernel's default writepages path for now. We will probably want to change | |
171 | * that eventually (i.e. when we look at allocate on flush). | |
172 | * | |
173 | * For the data=writeback case though we can already ignore buffer heads | |
174 | * and write whole extents at once. This is a big reduction in the | |
175 | * number of I/O requests we send and the bmap calls we make in this case. | |
176 | */ | |
a2cf8222 AB |
177 | static int gfs2_writepages(struct address_space *mapping, |
178 | struct writeback_control *wbc) | |
a8d638e3 SW |
179 | { |
180 | struct inode *inode = mapping->host; | |
181 | struct gfs2_inode *ip = GFS2_I(inode); | |
a8d638e3 | 182 | |
bf36a713 | 183 | if (gfs2_is_writeback(ip)) |
a8d638e3 SW |
184 | return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc); |
185 | ||
186 | return generic_writepages(mapping, wbc); | |
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 | ||
bf126aee SW |
203 | /* |
204 | * Due to the order of unstuffing files and ->nopage(), we can be | |
205 | * asked for a zero page in the case of a stuffed file being extended, | |
206 | * so we need to supply one here. It doesn't happen often. | |
207 | */ | |
208 | if (unlikely(page->index)) { | |
2840501a | 209 | zero_user_page(page, 0, PAGE_CACHE_SIZE, KM_USER0); |
bf126aee SW |
210 | return 0; |
211 | } | |
fd88de56 | 212 | |
b3b94faa DT |
213 | error = gfs2_meta_inode_buffer(ip, &dibh); |
214 | if (error) | |
215 | return error; | |
216 | ||
5c4e9e03 | 217 | kaddr = kmap_atomic(page, KM_USER0); |
fd88de56 | 218 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), |
b3b94faa | 219 | ip->i_di.di_size); |
fd88de56 | 220 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); |
c312c4fd | 221 | kunmap_atomic(kaddr, KM_USER0); |
bf126aee | 222 | flush_dcache_page(page); |
b3b94faa | 223 | brelse(dibh); |
b3b94faa DT |
224 | SetPageUptodate(page); |
225 | ||
226 | return 0; | |
227 | } | |
228 | ||
b3b94faa | 229 | |
b3b94faa | 230 | /** |
51ff87bd SW |
231 | * __gfs2_readpage - readpage |
232 | * @file: The file to read a page for | |
b3b94faa DT |
233 | * @page: The page to read |
234 | * | |
51ff87bd SW |
235 | * This is the core of gfs2's readpage. Its used by the internal file |
236 | * reading code as in that case we already hold the glock. Also its | |
237 | * called by gfs2_readpage() once the required lock has been granted. | |
238 | * | |
b3b94faa DT |
239 | */ |
240 | ||
51ff87bd | 241 | static int __gfs2_readpage(void *file, struct page *page) |
b3b94faa | 242 | { |
feaa7bba SW |
243 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
244 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
b3b94faa DT |
245 | int error; |
246 | ||
18ec7d5c | 247 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
248 | error = stuffed_readpage(ip, page); |
249 | unlock_page(page); | |
51ff87bd | 250 | } else { |
18ec7d5c | 251 | error = mpage_readpage(page, gfs2_get_block); |
51ff87bd | 252 | } |
b3b94faa DT |
253 | |
254 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
51ff87bd | 255 | return -EIO; |
b3b94faa | 256 | |
51ff87bd SW |
257 | return error; |
258 | } | |
259 | ||
260 | /** | |
261 | * gfs2_readpage - read a page of a file | |
262 | * @file: The file to read | |
263 | * @page: The page of the file | |
264 | * | |
3cc3f710 | 265 | * This deals with the locking required. We use a trylock in order to |
51ff87bd SW |
266 | * avoid the page lock / glock ordering problems returning AOP_TRUNCATED_PAGE |
267 | * in the event that we are unable to get the lock. | |
268 | */ | |
269 | ||
270 | static int gfs2_readpage(struct file *file, struct page *page) | |
271 | { | |
272 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | |
273 | struct gfs2_holder gh; | |
274 | int error; | |
275 | ||
51ff87bd SW |
276 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh); |
277 | error = gfs2_glock_nq_atime(&gh); | |
278 | if (unlikely(error)) { | |
279 | unlock_page(page); | |
280 | goto out; | |
61a30dcb | 281 | } |
51ff87bd SW |
282 | error = __gfs2_readpage(file, page); |
283 | gfs2_glock_dq(&gh); | |
18ec7d5c | 284 | out: |
51ff87bd | 285 | gfs2_holder_uninit(&gh); |
a13cbe37 | 286 | if (error == GLR_TRYFAILED) { |
a13cbe37 | 287 | yield(); |
51ff87bd | 288 | return AOP_TRUNCATED_PAGE; |
a13cbe37 | 289 | } |
51ff87bd SW |
290 | return error; |
291 | } | |
292 | ||
293 | /** | |
294 | * gfs2_internal_read - read an internal file | |
295 | * @ip: The gfs2 inode | |
296 | * @ra_state: The readahead state (or NULL for no readahead) | |
297 | * @buf: The buffer to fill | |
298 | * @pos: The file position | |
299 | * @size: The amount to read | |
300 | * | |
301 | */ | |
302 | ||
303 | int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state, | |
304 | char *buf, loff_t *pos, unsigned size) | |
305 | { | |
306 | struct address_space *mapping = ip->i_inode.i_mapping; | |
307 | unsigned long index = *pos / PAGE_CACHE_SIZE; | |
308 | unsigned offset = *pos & (PAGE_CACHE_SIZE - 1); | |
309 | unsigned copied = 0; | |
310 | unsigned amt; | |
311 | struct page *page; | |
312 | void *p; | |
313 | ||
314 | do { | |
315 | amt = size - copied; | |
316 | if (offset + size > PAGE_CACHE_SIZE) | |
317 | amt = PAGE_CACHE_SIZE - offset; | |
318 | page = read_cache_page(mapping, index, __gfs2_readpage, NULL); | |
319 | if (IS_ERR(page)) | |
320 | return PTR_ERR(page); | |
321 | p = kmap_atomic(page, KM_USER0); | |
322 | memcpy(buf + copied, p + offset, amt); | |
323 | kunmap_atomic(p, KM_USER0); | |
324 | mark_page_accessed(page); | |
325 | page_cache_release(page); | |
326 | copied += amt; | |
327 | index++; | |
328 | offset = 0; | |
329 | } while(copied < size); | |
330 | (*pos) += size; | |
331 | return size; | |
fd88de56 SW |
332 | } |
333 | ||
fd88de56 SW |
334 | /** |
335 | * gfs2_readpages - Read a bunch of pages at once | |
336 | * | |
337 | * Some notes: | |
338 | * 1. This is only for readahead, so we can simply ignore any things | |
339 | * which are slightly inconvenient (such as locking conflicts between | |
340 | * the page lock and the glock) and return having done no I/O. Its | |
341 | * obviously not something we'd want to do on too regular a basis. | |
342 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 343 | * 2. We don't handle stuffed files here we let readpage do the honours. |
fd88de56 SW |
344 | * 3. mpage_readpages() does most of the heavy lifting in the common case. |
345 | * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places. | |
fd88de56 | 346 | */ |
3cc3f710 | 347 | |
fd88de56 SW |
348 | static int gfs2_readpages(struct file *file, struct address_space *mapping, |
349 | struct list_head *pages, unsigned nr_pages) | |
350 | { | |
351 | struct inode *inode = mapping->host; | |
feaa7bba SW |
352 | struct gfs2_inode *ip = GFS2_I(inode); |
353 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 | 354 | struct gfs2_holder gh; |
3cc3f710 | 355 | int ret; |
fd88de56 | 356 | |
3cc3f710 | 357 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh); |
51ff87bd | 358 | ret = gfs2_glock_nq_atime(&gh); |
51ff87bd | 359 | if (unlikely(ret)) |
3cc3f710 | 360 | goto out_uninit; |
e1d5b18a | 361 | if (!gfs2_is_stuffed(ip)) |
fd88de56 | 362 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block); |
3cc3f710 SW |
363 | gfs2_glock_dq(&gh); |
364 | out_uninit: | |
365 | gfs2_holder_uninit(&gh); | |
fd88de56 SW |
366 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
367 | ret = -EIO; | |
368 | return ret; | |
b3b94faa DT |
369 | } |
370 | ||
371 | /** | |
7765ec26 | 372 | * gfs2_write_begin - Begin to write to a file |
b3b94faa | 373 | * @file: The file to write to |
7765ec26 SW |
374 | * @mapping: The mapping in which to write |
375 | * @pos: The file offset at which to start writing | |
376 | * @len: Length of the write | |
377 | * @flags: Various flags | |
378 | * @pagep: Pointer to return the page | |
379 | * @fsdata: Pointer to return fs data (unused by GFS2) | |
b3b94faa DT |
380 | * |
381 | * Returns: errno | |
382 | */ | |
383 | ||
7765ec26 SW |
384 | static int gfs2_write_begin(struct file *file, struct address_space *mapping, |
385 | loff_t pos, unsigned len, unsigned flags, | |
386 | struct page **pagep, void **fsdata) | |
b3b94faa | 387 | { |
7765ec26 SW |
388 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
389 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
18ec7d5c SW |
390 | unsigned int data_blocks, ind_blocks, rblocks; |
391 | int alloc_required; | |
b3b94faa | 392 | int error = 0; |
18ec7d5c | 393 | struct gfs2_alloc *al; |
7765ec26 SW |
394 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
395 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
396 | unsigned to = from + len; | |
397 | struct page *page; | |
52ae7b79 | 398 | |
7765ec26 | 399 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh); |
dcd24799 | 400 | error = gfs2_glock_nq_atime(&ip->i_gh); |
7765ec26 | 401 | if (unlikely(error)) |
18ec7d5c | 402 | goto out_uninit; |
b3b94faa | 403 | |
7765ec26 SW |
404 | error = -ENOMEM; |
405 | page = __grab_cache_page(mapping, index); | |
406 | *pagep = page; | |
407 | if (!page) | |
408 | goto out_unlock; | |
409 | ||
410 | gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks); | |
18ec7d5c | 411 | |
7765ec26 | 412 | error = gfs2_write_alloc_required(ip, pos, len, &alloc_required); |
18ec7d5c | 413 | if (error) |
7765ec26 | 414 | goto out_putpage; |
b3b94faa | 415 | |
18ec7d5c | 416 | |
f5c54804 | 417 | ip->i_alloc.al_requested = 0; |
18ec7d5c SW |
418 | if (alloc_required) { |
419 | al = gfs2_alloc_get(ip); | |
420 | ||
421 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
422 | if (error) | |
423 | goto out_alloc_put; | |
424 | ||
2933f925 | 425 | error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid); |
18ec7d5c SW |
426 | if (error) |
427 | goto out_qunlock; | |
428 | ||
429 | al->al_requested = data_blocks + ind_blocks; | |
430 | error = gfs2_inplace_reserve(ip); | |
431 | if (error) | |
432 | goto out_qunlock; | |
433 | } | |
434 | ||
435 | rblocks = RES_DINODE + ind_blocks; | |
436 | if (gfs2_is_jdata(ip)) | |
437 | rblocks += data_blocks ? data_blocks : 1; | |
438 | if (ind_blocks || data_blocks) | |
439 | rblocks += RES_STATFS + RES_QUOTA; | |
440 | ||
16615be1 SW |
441 | error = gfs2_trans_begin(sdp, rblocks, |
442 | PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); | |
18ec7d5c | 443 | if (error) |
a867bb28 | 444 | goto out_trans_fail; |
18ec7d5c SW |
445 | |
446 | if (gfs2_is_stuffed(ip)) { | |
7765ec26 | 447 | if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { |
f25ef0c1 | 448 | error = gfs2_unstuff_dinode(ip, page); |
5c4e9e03 SW |
449 | if (error == 0) |
450 | goto prepare_write; | |
451 | } else if (!PageUptodate(page)) | |
b3b94faa | 452 | error = stuffed_readpage(ip, page); |
5c4e9e03 | 453 | goto out; |
18ec7d5c SW |
454 | } |
455 | ||
5c4e9e03 | 456 | prepare_write: |
18ec7d5c SW |
457 | error = block_prepare_write(page, from, to, gfs2_get_block); |
458 | ||
459 | out: | |
460 | if (error) { | |
461 | gfs2_trans_end(sdp); | |
a867bb28 | 462 | out_trans_fail: |
18ec7d5c SW |
463 | if (alloc_required) { |
464 | gfs2_inplace_release(ip); | |
465 | out_qunlock: | |
466 | gfs2_quota_unlock(ip); | |
467 | out_alloc_put: | |
468 | gfs2_alloc_put(ip); | |
469 | } | |
7765ec26 SW |
470 | out_putpage: |
471 | page_cache_release(page); | |
472 | if (pos + len > ip->i_inode.i_size) | |
473 | vmtruncate(&ip->i_inode, ip->i_inode.i_size); | |
18ec7d5c SW |
474 | out_unlock: |
475 | gfs2_glock_dq_m(1, &ip->i_gh); | |
476 | out_uninit: | |
477 | gfs2_holder_uninit(&ip->i_gh); | |
478 | } | |
b3b94faa DT |
479 | |
480 | return error; | |
481 | } | |
482 | ||
7ae8fa84 RP |
483 | /** |
484 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow | |
485 | * @inode: the rindex inode | |
486 | */ | |
487 | static void adjust_fs_space(struct inode *inode) | |
488 | { | |
489 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
490 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; | |
491 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
492 | u64 fs_total, new_free; | |
493 | ||
494 | /* Total up the file system space, according to the latest rindex. */ | |
495 | fs_total = gfs2_ri_total(sdp); | |
496 | ||
497 | spin_lock(&sdp->sd_statfs_spin); | |
498 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) | |
499 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); | |
500 | else | |
501 | new_free = 0; | |
502 | spin_unlock(&sdp->sd_statfs_spin); | |
6c53267f RP |
503 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
504 | (unsigned long long)new_free); | |
7ae8fa84 RP |
505 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
506 | } | |
507 | ||
b3b94faa | 508 | /** |
7765ec26 SW |
509 | * gfs2_stuffed_write_end - Write end for stuffed files |
510 | * @inode: The inode | |
511 | * @dibh: The buffer_head containing the on-disk inode | |
512 | * @pos: The file position | |
513 | * @len: The length of the write | |
514 | * @copied: How much was actually copied by the VFS | |
515 | * @page: The page | |
516 | * | |
517 | * This copies the data from the page into the inode block after | |
518 | * the inode data structure itself. | |
519 | * | |
520 | * Returns: errno | |
521 | */ | |
522 | static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh, | |
523 | loff_t pos, unsigned len, unsigned copied, | |
524 | struct page *page) | |
525 | { | |
526 | struct gfs2_inode *ip = GFS2_I(inode); | |
527 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
528 | u64 to = pos + copied; | |
529 | void *kaddr; | |
530 | unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode); | |
531 | struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data; | |
532 | ||
533 | BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode))); | |
534 | kaddr = kmap_atomic(page, KM_USER0); | |
535 | memcpy(buf + pos, kaddr + pos, copied); | |
536 | memset(kaddr + pos + copied, 0, len - copied); | |
537 | flush_dcache_page(page); | |
538 | kunmap_atomic(kaddr, KM_USER0); | |
539 | ||
540 | if (!PageUptodate(page)) | |
541 | SetPageUptodate(page); | |
542 | unlock_page(page); | |
543 | page_cache_release(page); | |
544 | ||
545 | if (inode->i_size < to) { | |
546 | i_size_write(inode, to); | |
547 | ip->i_di.di_size = inode->i_size; | |
548 | di->di_size = cpu_to_be64(inode->i_size); | |
549 | mark_inode_dirty(inode); | |
550 | } | |
551 | ||
552 | if (inode == sdp->sd_rindex) | |
553 | adjust_fs_space(inode); | |
554 | ||
555 | brelse(dibh); | |
556 | gfs2_trans_end(sdp); | |
557 | gfs2_glock_dq(&ip->i_gh); | |
558 | gfs2_holder_uninit(&ip->i_gh); | |
559 | return copied; | |
560 | } | |
561 | ||
562 | /** | |
563 | * gfs2_write_end | |
b3b94faa | 564 | * @file: The file to write to |
7765ec26 SW |
565 | * @mapping: The address space to write to |
566 | * @pos: The file position | |
567 | * @len: The length of the data | |
568 | * @copied: | |
569 | * @page: The page that has been written | |
570 | * @fsdata: The fsdata (unused in GFS2) | |
571 | * | |
572 | * The main write_end function for GFS2. We have a separate one for | |
573 | * stuffed files as they are slightly different, otherwise we just | |
574 | * put our locking around the VFS provided functions. | |
b3b94faa DT |
575 | * |
576 | * Returns: errno | |
577 | */ | |
578 | ||
7765ec26 SW |
579 | static int gfs2_write_end(struct file *file, struct address_space *mapping, |
580 | loff_t pos, unsigned len, unsigned copied, | |
581 | struct page *page, void *fsdata) | |
b3b94faa DT |
582 | { |
583 | struct inode *inode = page->mapping->host; | |
feaa7bba SW |
584 | struct gfs2_inode *ip = GFS2_I(inode); |
585 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c | 586 | struct buffer_head *dibh; |
48516ced SW |
587 | struct gfs2_alloc *al = &ip->i_alloc; |
588 | struct gfs2_dinode *di; | |
7765ec26 SW |
589 | unsigned int from = pos & (PAGE_CACHE_SIZE - 1); |
590 | unsigned int to = from + len; | |
591 | int ret; | |
b3b94faa | 592 | |
7765ec26 | 593 | BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == 0); |
18ec7d5c | 594 | |
7765ec26 SW |
595 | ret = gfs2_meta_inode_buffer(ip, &dibh); |
596 | if (unlikely(ret)) { | |
597 | unlock_page(page); | |
598 | page_cache_release(page); | |
599 | goto failed; | |
600 | } | |
18ec7d5c SW |
601 | |
602 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
b3b94faa | 603 | |
7765ec26 SW |
604 | if (gfs2_is_stuffed(ip)) |
605 | return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page); | |
b3b94faa | 606 | |
bf36a713 | 607 | if (!gfs2_is_writeback(ip)) |
7765ec26 | 608 | gfs2_page_add_databufs(ip, page, from, to); |
b3b94faa | 609 | |
7765ec26 | 610 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
b3b94faa | 611 | |
7765ec26 SW |
612 | if (likely(ret >= 0)) { |
613 | copied = ret; | |
614 | if ((pos + copied) > inode->i_size) { | |
615 | di = (struct gfs2_dinode *)dibh->b_data; | |
616 | ip->i_di.di_size = inode->i_size; | |
617 | di->di_size = cpu_to_be64(inode->i_size); | |
ae619320 SW |
618 | mark_inode_dirty(inode); |
619 | } | |
48516ced SW |
620 | } |
621 | ||
7ae8fa84 RP |
622 | if (inode == sdp->sd_rindex) |
623 | adjust_fs_space(inode); | |
624 | ||
18ec7d5c SW |
625 | brelse(dibh); |
626 | gfs2_trans_end(sdp); | |
7765ec26 | 627 | failed: |
18ec7d5c SW |
628 | if (al->al_requested) { |
629 | gfs2_inplace_release(ip); | |
630 | gfs2_quota_unlock(ip); | |
631 | gfs2_alloc_put(ip); | |
632 | } | |
7765ec26 | 633 | gfs2_glock_dq(&ip->i_gh); |
18ec7d5c | 634 | gfs2_holder_uninit(&ip->i_gh); |
7765ec26 | 635 | return ret; |
b3b94faa DT |
636 | } |
637 | ||
8fb68595 RP |
638 | /** |
639 | * gfs2_set_page_dirty - Page dirtying function | |
640 | * @page: The page to dirty | |
641 | * | |
642 | * Returns: 1 if it dirtyed the page, or 0 otherwise | |
643 | */ | |
644 | ||
645 | static int gfs2_set_page_dirty(struct page *page) | |
646 | { | |
647 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | |
648 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
649 | ||
650 | if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) | |
651 | SetPageChecked(page); | |
652 | return __set_page_dirty_buffers(page); | |
653 | } | |
654 | ||
b3b94faa DT |
655 | /** |
656 | * gfs2_bmap - Block map function | |
657 | * @mapping: Address space info | |
658 | * @lblock: The block to map | |
659 | * | |
660 | * Returns: The disk address for the block or 0 on hole or error | |
661 | */ | |
662 | ||
663 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
664 | { | |
feaa7bba | 665 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
666 | struct gfs2_holder i_gh; |
667 | sector_t dblock = 0; | |
668 | int error; | |
669 | ||
b3b94faa DT |
670 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
671 | if (error) | |
672 | return 0; | |
673 | ||
674 | if (!gfs2_is_stuffed(ip)) | |
4ff14670 | 675 | dblock = generic_block_bmap(mapping, lblock, gfs2_get_block); |
b3b94faa DT |
676 | |
677 | gfs2_glock_dq_uninit(&i_gh); | |
678 | ||
679 | return dblock; | |
680 | } | |
681 | ||
d7b616e2 SW |
682 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) |
683 | { | |
684 | struct gfs2_bufdata *bd; | |
685 | ||
686 | lock_buffer(bh); | |
687 | gfs2_log_lock(sdp); | |
688 | clear_buffer_dirty(bh); | |
689 | bd = bh->b_private; | |
690 | if (bd) { | |
16615be1 SW |
691 | if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh)) |
692 | list_del_init(&bd->bd_le.le_list); | |
693 | else | |
694 | gfs2_remove_from_journal(bh, current->journal_info, 0); | |
d7b616e2 SW |
695 | } |
696 | bh->b_bdev = NULL; | |
697 | clear_buffer_mapped(bh); | |
698 | clear_buffer_req(bh); | |
699 | clear_buffer_new(bh); | |
700 | gfs2_log_unlock(sdp); | |
701 | unlock_buffer(bh); | |
702 | } | |
703 | ||
8628de05 | 704 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) |
b3b94faa | 705 | { |
d7b616e2 SW |
706 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
707 | struct buffer_head *bh, *head; | |
708 | unsigned long pos = 0; | |
709 | ||
b3b94faa | 710 | BUG_ON(!PageLocked(page)); |
8fb68595 RP |
711 | if (offset == 0) |
712 | ClearPageChecked(page); | |
d7b616e2 SW |
713 | if (!page_has_buffers(page)) |
714 | goto out; | |
b3b94faa | 715 | |
d7b616e2 SW |
716 | bh = head = page_buffers(page); |
717 | do { | |
718 | if (offset <= pos) | |
719 | gfs2_discard(sdp, bh); | |
720 | pos += bh->b_size; | |
721 | bh = bh->b_this_page; | |
722 | } while (bh != head); | |
723 | out: | |
724 | if (offset == 0) | |
725 | try_to_release_page(page, 0); | |
b3b94faa DT |
726 | } |
727 | ||
c7b33834 SW |
728 | /** |
729 | * gfs2_ok_for_dio - check that dio is valid on this file | |
730 | * @ip: The inode | |
731 | * @rw: READ or WRITE | |
732 | * @offset: The offset at which we are reading or writing | |
733 | * | |
734 | * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o) | |
735 | * 1 (to accept the i/o request) | |
736 | */ | |
737 | static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset) | |
738 | { | |
739 | /* | |
740 | * Should we return an error here? I can't see that O_DIRECT for | |
741 | * a journaled file makes any sense. For now we'll silently fall | |
742 | * back to buffered I/O, likewise we do the same for stuffed | |
743 | * files since they are (a) small and (b) unaligned. | |
744 | */ | |
745 | if (gfs2_is_jdata(ip)) | |
746 | return 0; | |
747 | ||
748 | if (gfs2_is_stuffed(ip)) | |
749 | return 0; | |
750 | ||
751 | if (offset > i_size_read(&ip->i_inode)) | |
752 | return 0; | |
753 | return 1; | |
754 | } | |
755 | ||
756 | ||
757 | ||
a9e5f4d0 SW |
758 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, |
759 | const struct iovec *iov, loff_t offset, | |
760 | unsigned long nr_segs) | |
d1665e41 SW |
761 | { |
762 | struct file *file = iocb->ki_filp; | |
763 | struct inode *inode = file->f_mapping->host; | |
feaa7bba | 764 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
765 | struct gfs2_holder gh; |
766 | int rv; | |
767 | ||
768 | /* | |
c7b33834 SW |
769 | * Deferred lock, even if its a write, since we do no allocation |
770 | * on this path. All we need change is atime, and this lock mode | |
771 | * ensures that other nodes have flushed their buffered read caches | |
772 | * (i.e. their page cache entries for this inode). We do not, | |
773 | * unfortunately have the option of only flushing a range like | |
774 | * the VFS does. | |
d1665e41 | 775 | */ |
c7b33834 | 776 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh); |
dcd24799 | 777 | rv = gfs2_glock_nq_atime(&gh); |
d1665e41 | 778 | if (rv) |
c7b33834 SW |
779 | return rv; |
780 | rv = gfs2_ok_for_dio(ip, rw, offset); | |
781 | if (rv != 1) | |
782 | goto out; /* dio not valid, fall back to buffered i/o */ | |
783 | ||
784 | rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev, | |
785 | iov, offset, nr_segs, | |
786 | gfs2_get_block_direct, NULL); | |
d1665e41 SW |
787 | out: |
788 | gfs2_glock_dq_m(1, &gh); | |
789 | gfs2_holder_uninit(&gh); | |
d1665e41 SW |
790 | return rv; |
791 | } | |
792 | ||
4340fe62 | 793 | /** |
623d9355 | 794 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
795 | * @page: the page that's being released |
796 | * @gfp_mask: passed from Linux VFS, ignored by us | |
797 | * | |
798 | * Call try_to_free_buffers() if the buffers in this page can be | |
799 | * released. | |
800 | * | |
801 | * Returns: 0 | |
802 | */ | |
803 | ||
804 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
805 | { | |
806 | struct inode *aspace = page->mapping->host; | |
807 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | |
808 | struct buffer_head *bh, *head; | |
809 | struct gfs2_bufdata *bd; | |
4340fe62 SW |
810 | |
811 | if (!page_has_buffers(page)) | |
891ba6d4 | 812 | return 0; |
4340fe62 | 813 | |
bb3b0e3d | 814 | gfs2_log_lock(sdp); |
4340fe62 SW |
815 | head = bh = page_buffers(page); |
816 | do { | |
bb3b0e3d SW |
817 | if (atomic_read(&bh->b_count)) |
818 | goto cannot_release; | |
819 | bd = bh->b_private; | |
820 | if (bd && bd->bd_ail) | |
821 | goto cannot_release; | |
4340fe62 | 822 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); |
623d9355 | 823 | gfs2_assert_warn(sdp, !buffer_dirty(bh)); |
bb3b0e3d SW |
824 | bh = bh->b_this_page; |
825 | } while(bh != head); | |
826 | gfs2_log_unlock(sdp); | |
4340fe62 | 827 | |
bb3b0e3d SW |
828 | head = bh = page_buffers(page); |
829 | do { | |
623d9355 | 830 | gfs2_log_lock(sdp); |
4340fe62 SW |
831 | bd = bh->b_private; |
832 | if (bd) { | |
833 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
834 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
d7b616e2 SW |
835 | if (!list_empty(&bd->bd_le.le_list)) { |
836 | if (!buffer_pinned(bh)) | |
837 | list_del_init(&bd->bd_le.le_list); | |
838 | else | |
839 | bd = NULL; | |
840 | } | |
841 | if (bd) | |
842 | bd->bd_bh = 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 | |
4340fe62 | 852 | return try_to_free_buffers(page); |
bb3b0e3d SW |
853 | cannot_release: |
854 | gfs2_log_unlock(sdp); | |
855 | return 0; | |
4340fe62 SW |
856 | } |
857 | ||
66de045d | 858 | const struct address_space_operations gfs2_file_aops = { |
b3b94faa | 859 | .writepage = gfs2_writepage, |
a8d638e3 | 860 | .writepages = gfs2_writepages, |
b3b94faa | 861 | .readpage = gfs2_readpage, |
fd88de56 | 862 | .readpages = gfs2_readpages, |
b3b94faa | 863 | .sync_page = block_sync_page, |
7765ec26 SW |
864 | .write_begin = gfs2_write_begin, |
865 | .write_end = gfs2_write_end, | |
8fb68595 | 866 | .set_page_dirty = gfs2_set_page_dirty, |
b3b94faa DT |
867 | .bmap = gfs2_bmap, |
868 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 869 | .releasepage = gfs2_releasepage, |
b3b94faa DT |
870 | .direct_IO = gfs2_direct_IO, |
871 | }; | |
872 |