Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
091806ed | 3 | * Copyright (C) 2004-2008 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/mm.h> | |
16 | #include <linux/pagemap.h> | |
17 | #include <linux/writeback.h> | |
18 | #include <linux/swap.h> | |
19 | #include <linux/delay.h> | |
2e565bb6 | 20 | #include <linux/bio.h> |
5c676f6d | 21 | #include <linux/gfs2_ondisk.h> |
b3b94faa DT |
22 | |
23 | #include "gfs2.h" | |
5c676f6d | 24 | #include "incore.h" |
b3b94faa DT |
25 | #include "glock.h" |
26 | #include "glops.h" | |
27 | #include "inode.h" | |
28 | #include "log.h" | |
29 | #include "lops.h" | |
30 | #include "meta_io.h" | |
31 | #include "rgrp.h" | |
32 | #include "trans.h" | |
5c676f6d | 33 | #include "util.h" |
b3b94faa | 34 | |
4a0f9a32 | 35 | static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc) |
b3b94faa | 36 | { |
4a0f9a32 SW |
37 | int err; |
38 | struct buffer_head *bh, *head; | |
39 | int nr_underway = 0; | |
40 | int write_op = (1 << BIO_RW_META) | ((wbc->sync_mode == WB_SYNC_ALL ? | |
41 | WRITE_SYNC_PLUG : WRITE)); | |
b3b94faa | 42 | |
4a0f9a32 SW |
43 | BUG_ON(!PageLocked(page)); |
44 | BUG_ON(!page_has_buffers(page)); | |
45 | ||
46 | head = page_buffers(page); | |
47 | bh = head; | |
48 | ||
49 | do { | |
50 | if (!buffer_mapped(bh)) | |
51 | continue; | |
52 | /* | |
53 | * If it's a fully non-blocking write attempt and we cannot | |
54 | * lock the buffer then redirty the page. Note that this can | |
55 | * potentially cause a busy-wait loop from pdflush and kswapd | |
56 | * activity, but those code paths have their own higher-level | |
57 | * throttling. | |
58 | */ | |
59 | if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) { | |
60 | lock_buffer(bh); | |
61 | } else if (!trylock_buffer(bh)) { | |
62 | redirty_page_for_writepage(wbc, page); | |
63 | continue; | |
64 | } | |
65 | if (test_clear_buffer_dirty(bh)) { | |
66 | mark_buffer_async_write(bh); | |
67 | } else { | |
68 | unlock_buffer(bh); | |
69 | } | |
70 | } while ((bh = bh->b_this_page) != head); | |
71 | ||
72 | /* | |
73 | * The page and its buffers are protected by PageWriteback(), so we can | |
74 | * drop the bh refcounts early. | |
75 | */ | |
76 | BUG_ON(PageWriteback(page)); | |
77 | set_page_writeback(page); | |
78 | ||
79 | do { | |
80 | struct buffer_head *next = bh->b_this_page; | |
81 | if (buffer_async_write(bh)) { | |
82 | submit_bh(write_op, bh); | |
83 | nr_underway++; | |
84 | } | |
85 | bh = next; | |
86 | } while (bh != head); | |
87 | unlock_page(page); | |
88 | ||
89 | err = 0; | |
90 | if (nr_underway == 0) | |
91 | end_page_writeback(page); | |
92 | ||
93 | return err; | |
b3b94faa DT |
94 | } |
95 | ||
66de045d | 96 | static const struct address_space_operations aspace_aops = { |
b3b94faa | 97 | .writepage = gfs2_aspace_writepage, |
4340fe62 | 98 | .releasepage = gfs2_releasepage, |
52d4c74b | 99 | .sync_page = block_sync_page, |
b3b94faa DT |
100 | }; |
101 | ||
102 | /** | |
103 | * gfs2_aspace_get - Create and initialize a struct inode structure | |
104 | * @sdp: the filesystem the aspace is in | |
105 | * | |
106 | * Right now a struct inode is just a struct inode. Maybe Linux | |
107 | * will supply a more lightweight address space construct (that works) | |
108 | * in the future. | |
109 | * | |
110 | * Make sure pages/buffers in this aspace aren't in high memory. | |
111 | * | |
112 | * Returns: the aspace | |
113 | */ | |
114 | ||
115 | struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp) | |
116 | { | |
117 | struct inode *aspace; | |
091806ed | 118 | struct gfs2_inode *ip; |
b3b94faa DT |
119 | |
120 | aspace = new_inode(sdp->sd_vfs); | |
121 | if (aspace) { | |
f3bba03f | 122 | mapping_set_gfp_mask(aspace->i_mapping, GFP_NOFS); |
b3b94faa | 123 | aspace->i_mapping->a_ops = &aspace_aops; |
ba198098 | 124 | aspace->i_size = MAX_LFS_FILESIZE; |
091806ed BP |
125 | ip = GFS2_I(aspace); |
126 | clear_bit(GIF_USER, &ip->i_flags); | |
b3b94faa DT |
127 | insert_inode_hash(aspace); |
128 | } | |
b3b94faa DT |
129 | return aspace; |
130 | } | |
131 | ||
132 | void gfs2_aspace_put(struct inode *aspace) | |
133 | { | |
134 | remove_inode_hash(aspace); | |
135 | iput(aspace); | |
136 | } | |
137 | ||
b3b94faa DT |
138 | /** |
139 | * gfs2_meta_sync - Sync all buffers associated with a glock | |
140 | * @gl: The glock | |
b3b94faa DT |
141 | * |
142 | */ | |
143 | ||
7276b3b0 | 144 | void gfs2_meta_sync(struct gfs2_glock *gl) |
b3b94faa DT |
145 | { |
146 | struct address_space *mapping = gl->gl_aspace->i_mapping; | |
7276b3b0 | 147 | int error; |
b3b94faa | 148 | |
7276b3b0 SW |
149 | filemap_fdatawrite(mapping); |
150 | error = filemap_fdatawait(mapping); | |
b3b94faa DT |
151 | |
152 | if (error) | |
153 | gfs2_io_error(gl->gl_sbd); | |
154 | } | |
155 | ||
156 | /** | |
6802e340 | 157 | * gfs2_getbuf - Get a buffer with a given address space |
cb4c0313 | 158 | * @gl: the glock |
b3b94faa DT |
159 | * @blkno: the block number (filesystem scope) |
160 | * @create: 1 if the buffer should be created | |
161 | * | |
162 | * Returns: the buffer | |
163 | */ | |
164 | ||
6802e340 | 165 | struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create) |
b3b94faa | 166 | { |
cb4c0313 SW |
167 | struct address_space *mapping = gl->gl_aspace->i_mapping; |
168 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
b3b94faa DT |
169 | struct page *page; |
170 | struct buffer_head *bh; | |
171 | unsigned int shift; | |
172 | unsigned long index; | |
173 | unsigned int bufnum; | |
174 | ||
175 | shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift; | |
176 | index = blkno >> shift; /* convert block to page */ | |
177 | bufnum = blkno - (index << shift); /* block buf index within page */ | |
178 | ||
179 | if (create) { | |
180 | for (;;) { | |
cb4c0313 | 181 | page = grab_cache_page(mapping, index); |
b3b94faa DT |
182 | if (page) |
183 | break; | |
184 | yield(); | |
185 | } | |
186 | } else { | |
cb4c0313 | 187 | page = find_lock_page(mapping, index); |
b3b94faa DT |
188 | if (!page) |
189 | return NULL; | |
190 | } | |
191 | ||
192 | if (!page_has_buffers(page)) | |
193 | create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); | |
194 | ||
195 | /* Locate header for our buffer within our page */ | |
196 | for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) | |
197 | /* Do nothing */; | |
198 | get_bh(bh); | |
199 | ||
200 | if (!buffer_mapped(bh)) | |
201 | map_bh(bh, sdp->sd_vfs, blkno); | |
202 | ||
203 | unlock_page(page); | |
204 | mark_page_accessed(page); | |
205 | page_cache_release(page); | |
206 | ||
207 | return bh; | |
208 | } | |
209 | ||
210 | static void meta_prep_new(struct buffer_head *bh) | |
211 | { | |
212 | struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; | |
213 | ||
214 | lock_buffer(bh); | |
215 | clear_buffer_dirty(bh); | |
216 | set_buffer_uptodate(bh); | |
217 | unlock_buffer(bh); | |
218 | ||
219 | mh->mh_magic = cpu_to_be32(GFS2_MAGIC); | |
220 | } | |
221 | ||
222 | /** | |
223 | * gfs2_meta_new - Get a block | |
224 | * @gl: The glock associated with this block | |
225 | * @blkno: The block number | |
226 | * | |
227 | * Returns: The buffer | |
228 | */ | |
229 | ||
cd915493 | 230 | struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno) |
b3b94faa DT |
231 | { |
232 | struct buffer_head *bh; | |
6802e340 | 233 | bh = gfs2_getbuf(gl, blkno, CREATE); |
b3b94faa DT |
234 | meta_prep_new(bh); |
235 | return bh; | |
236 | } | |
237 | ||
238 | /** | |
239 | * gfs2_meta_read - Read a block from disk | |
240 | * @gl: The glock covering the block | |
241 | * @blkno: The block number | |
7276b3b0 | 242 | * @flags: flags |
b3b94faa DT |
243 | * @bhp: the place where the buffer is returned (NULL on failure) |
244 | * | |
245 | * Returns: errno | |
246 | */ | |
247 | ||
cd915493 | 248 | int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, |
b3b94faa DT |
249 | struct buffer_head **bhp) |
250 | { | |
c969f58c SW |
251 | struct gfs2_sbd *sdp = gl->gl_sbd; |
252 | struct buffer_head *bh; | |
253 | ||
254 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
255 | return -EIO; | |
256 | ||
257 | *bhp = bh = gfs2_getbuf(gl, blkno, CREATE); | |
258 | ||
259 | lock_buffer(bh); | |
260 | if (buffer_uptodate(bh)) { | |
261 | unlock_buffer(bh); | |
262 | return 0; | |
263 | } | |
264 | bh->b_end_io = end_buffer_read_sync; | |
265 | get_bh(bh); | |
266 | submit_bh(READ_SYNC | (1 << BIO_RW_META), bh); | |
267 | if (!(flags & DIO_WAIT)) | |
268 | return 0; | |
269 | ||
270 | wait_on_buffer(bh); | |
271 | if (unlikely(!buffer_uptodate(bh))) { | |
272 | struct gfs2_trans *tr = current->journal_info; | |
273 | if (tr && tr->tr_touched) | |
274 | gfs2_io_error_bh(sdp, bh); | |
275 | brelse(bh); | |
276 | return -EIO; | |
7276b3b0 | 277 | } |
b3b94faa | 278 | |
7276b3b0 | 279 | return 0; |
b3b94faa DT |
280 | } |
281 | ||
282 | /** | |
7276b3b0 | 283 | * gfs2_meta_wait - Reread a block from disk |
b3b94faa | 284 | * @sdp: the filesystem |
7276b3b0 | 285 | * @bh: The block to wait for |
b3b94faa DT |
286 | * |
287 | * Returns: errno | |
288 | */ | |
289 | ||
7276b3b0 | 290 | int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) |
b3b94faa DT |
291 | { |
292 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
293 | return -EIO; | |
294 | ||
7276b3b0 | 295 | wait_on_buffer(bh); |
b3b94faa | 296 | |
7276b3b0 SW |
297 | if (!buffer_uptodate(bh)) { |
298 | struct gfs2_trans *tr = current->journal_info; | |
299 | if (tr && tr->tr_touched) | |
300 | gfs2_io_error_bh(sdp, bh); | |
301 | return -EIO; | |
b3b94faa | 302 | } |
7276b3b0 SW |
303 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
304 | return -EIO; | |
b3b94faa DT |
305 | |
306 | return 0; | |
307 | } | |
308 | ||
309 | /** | |
586dfdaa | 310 | * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer |
b3b94faa DT |
311 | * @gl: the glock the buffer belongs to |
312 | * @bh: The buffer to be attached to | |
586dfdaa | 313 | * @meta: Flag to indicate whether its metadata or not |
b3b94faa DT |
314 | */ |
315 | ||
568f4c96 SW |
316 | void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh, |
317 | int meta) | |
b3b94faa DT |
318 | { |
319 | struct gfs2_bufdata *bd; | |
320 | ||
18ec7d5c SW |
321 | if (meta) |
322 | lock_page(bh->b_page); | |
b3b94faa | 323 | |
5c676f6d | 324 | if (bh->b_private) { |
18ec7d5c SW |
325 | if (meta) |
326 | unlock_page(bh->b_page); | |
b3b94faa DT |
327 | return; |
328 | } | |
329 | ||
3e5cd087 | 330 | bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
331 | bd->bd_bh = bh; |
332 | bd->bd_gl = gl; | |
333 | ||
334 | INIT_LIST_HEAD(&bd->bd_list_tr); | |
82ffa516 | 335 | if (meta) |
586dfdaa | 336 | lops_init_le(&bd->bd_le, &gfs2_buf_lops); |
82ffa516 | 337 | else |
586dfdaa | 338 | lops_init_le(&bd->bd_le, &gfs2_databuf_lops); |
5c676f6d | 339 | bh->b_private = bd; |
b3b94faa | 340 | |
18ec7d5c SW |
341 | if (meta) |
342 | unlock_page(bh->b_page); | |
b3b94faa DT |
343 | } |
344 | ||
16615be1 SW |
345 | void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta) |
346 | { | |
347 | struct gfs2_sbd *sdp = GFS2_SB(bh->b_page->mapping->host); | |
348 | struct gfs2_bufdata *bd = bh->b_private; | |
349 | if (test_clear_buffer_pinned(bh)) { | |
350 | list_del_init(&bd->bd_le.le_list); | |
351 | if (meta) { | |
352 | gfs2_assert_warn(sdp, sdp->sd_log_num_buf); | |
353 | sdp->sd_log_num_buf--; | |
354 | tr->tr_num_buf_rm++; | |
355 | } else { | |
356 | gfs2_assert_warn(sdp, sdp->sd_log_num_databuf); | |
357 | sdp->sd_log_num_databuf--; | |
358 | tr->tr_num_databuf_rm++; | |
359 | } | |
360 | tr->tr_touched = 1; | |
361 | brelse(bh); | |
362 | } | |
363 | if (bd) { | |
364 | if (bd->bd_ail) { | |
f91a0d3e | 365 | gfs2_remove_from_ail(bd); |
16615be1 SW |
366 | bh->b_private = NULL; |
367 | bd->bd_bh = NULL; | |
368 | bd->bd_blkno = bh->b_blocknr; | |
369 | gfs2_trans_add_revoke(sdp, bd); | |
370 | } | |
371 | } | |
372 | clear_buffer_dirty(bh); | |
373 | clear_buffer_uptodate(bh); | |
374 | } | |
375 | ||
b3b94faa DT |
376 | /** |
377 | * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore | |
378 | * @ip: the inode who owns the buffers | |
379 | * @bstart: the first buffer in the run | |
380 | * @blen: the number of buffers in the run | |
381 | * | |
382 | */ | |
383 | ||
cd915493 | 384 | void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen) |
b3b94faa | 385 | { |
feaa7bba | 386 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
387 | struct buffer_head *bh; |
388 | ||
389 | while (blen) { | |
6802e340 | 390 | bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE); |
b3b94faa | 391 | if (bh) { |
1ad38c43 SW |
392 | lock_buffer(bh); |
393 | gfs2_log_lock(sdp); | |
16615be1 | 394 | gfs2_remove_from_journal(bh, current->journal_info, 1); |
1ad38c43 | 395 | gfs2_log_unlock(sdp); |
b3b94faa | 396 | unlock_buffer(bh); |
b3b94faa DT |
397 | brelse(bh); |
398 | } | |
399 | ||
400 | bstart++; | |
401 | blen--; | |
402 | } | |
403 | } | |
404 | ||
b3b94faa DT |
405 | /** |
406 | * gfs2_meta_indirect_buffer - Get a metadata buffer | |
407 | * @ip: The GFS2 inode | |
408 | * @height: The level of this buf in the metadata (indir addr) tree (if any) | |
409 | * @num: The block number (device relative) of the buffer | |
410 | * @new: Non-zero if we may create a new buffer | |
411 | * @bhp: the buffer is returned here | |
412 | * | |
b3b94faa DT |
413 | * Returns: errno |
414 | */ | |
415 | ||
cd915493 | 416 | int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num, |
b3b94faa DT |
417 | int new, struct buffer_head **bhp) |
418 | { | |
7276b3b0 SW |
419 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
420 | struct gfs2_glock *gl = ip->i_gl; | |
f91a0d3e SW |
421 | struct buffer_head *bh; |
422 | int ret = 0; | |
b3b94faa DT |
423 | |
424 | if (new) { | |
f91a0d3e SW |
425 | BUG_ON(height == 0); |
426 | bh = gfs2_meta_new(gl, num); | |
d4e9c4c3 | 427 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
428 | gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
429 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
7276b3b0 SW |
430 | } else { |
431 | u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI; | |
f91a0d3e SW |
432 | ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh); |
433 | if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) { | |
434 | brelse(bh); | |
435 | ret = -EIO; | |
7276b3b0 | 436 | } |
b3b94faa | 437 | } |
b3b94faa | 438 | *bhp = bh; |
f91a0d3e | 439 | return ret; |
b3b94faa DT |
440 | } |
441 | ||
442 | /** | |
443 | * gfs2_meta_ra - start readahead on an extent of a file | |
444 | * @gl: the glock the blocks belong to | |
445 | * @dblock: the starting disk block | |
446 | * @extlen: the number of blocks in the extent | |
447 | * | |
7276b3b0 | 448 | * returns: the first buffer in the extent |
b3b94faa DT |
449 | */ |
450 | ||
7276b3b0 | 451 | struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen) |
b3b94faa DT |
452 | { |
453 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
b3b94faa | 454 | struct buffer_head *first_bh, *bh; |
cd915493 | 455 | u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> |
568f4c96 | 456 | sdp->sd_sb.sb_bsize_shift; |
b3b94faa | 457 | |
7276b3b0 SW |
458 | BUG_ON(!extlen); |
459 | ||
460 | if (max_ra < 1) | |
461 | max_ra = 1; | |
b3b94faa DT |
462 | if (extlen > max_ra) |
463 | extlen = max_ra; | |
464 | ||
6802e340 | 465 | first_bh = gfs2_getbuf(gl, dblock, CREATE); |
b3b94faa DT |
466 | |
467 | if (buffer_uptodate(first_bh)) | |
468 | goto out; | |
7276b3b0 | 469 | if (!buffer_locked(first_bh)) |
c969f58c | 470 | ll_rw_block(READ_SYNC | (1 << BIO_RW_META), 1, &first_bh); |
b3b94faa DT |
471 | |
472 | dblock++; | |
473 | extlen--; | |
474 | ||
475 | while (extlen) { | |
6802e340 | 476 | bh = gfs2_getbuf(gl, dblock, CREATE); |
b3b94faa | 477 | |
7276b3b0 SW |
478 | if (!buffer_uptodate(bh) && !buffer_locked(bh)) |
479 | ll_rw_block(READA, 1, &bh); | |
480 | brelse(bh); | |
b3b94faa DT |
481 | dblock++; |
482 | extlen--; | |
7276b3b0 SW |
483 | if (!buffer_locked(first_bh) && buffer_uptodate(first_bh)) |
484 | goto out; | |
b3b94faa DT |
485 | } |
486 | ||
7276b3b0 | 487 | wait_on_buffer(first_bh); |
a91ea69f | 488 | out: |
7276b3b0 | 489 | return first_bh; |
b3b94faa DT |
490 | } |
491 |