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 | struct buffer_head *bh, *head; |
38 | int nr_underway = 0; | |
7b6d91da CH |
39 | int write_op = REQ_META | |
40 | (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC_PLUG : WRITE); | |
b3b94faa | 41 | |
4a0f9a32 SW |
42 | BUG_ON(!PageLocked(page)); |
43 | BUG_ON(!page_has_buffers(page)); | |
44 | ||
45 | head = page_buffers(page); | |
46 | bh = head; | |
47 | ||
48 | do { | |
49 | if (!buffer_mapped(bh)) | |
50 | continue; | |
51 | /* | |
52 | * If it's a fully non-blocking write attempt and we cannot | |
53 | * lock the buffer then redirty the page. Note that this can | |
54 | * potentially cause a busy-wait loop from pdflush and kswapd | |
55 | * activity, but those code paths have their own higher-level | |
56 | * throttling. | |
57 | */ | |
1b430bee | 58 | if (wbc->sync_mode != WB_SYNC_NONE) { |
4a0f9a32 SW |
59 | lock_buffer(bh); |
60 | } else if (!trylock_buffer(bh)) { | |
61 | redirty_page_for_writepage(wbc, page); | |
62 | continue; | |
63 | } | |
64 | if (test_clear_buffer_dirty(bh)) { | |
65 | mark_buffer_async_write(bh); | |
66 | } else { | |
67 | unlock_buffer(bh); | |
68 | } | |
69 | } while ((bh = bh->b_this_page) != head); | |
70 | ||
71 | /* | |
72 | * The page and its buffers are protected by PageWriteback(), so we can | |
73 | * drop the bh refcounts early. | |
74 | */ | |
75 | BUG_ON(PageWriteback(page)); | |
76 | set_page_writeback(page); | |
77 | ||
78 | do { | |
79 | struct buffer_head *next = bh->b_this_page; | |
80 | if (buffer_async_write(bh)) { | |
81 | submit_bh(write_op, bh); | |
82 | nr_underway++; | |
83 | } | |
84 | bh = next; | |
85 | } while (bh != head); | |
86 | unlock_page(page); | |
87 | ||
4a0f9a32 SW |
88 | if (nr_underway == 0) |
89 | end_page_writeback(page); | |
90 | ||
eaefbf96 | 91 | return 0; |
b3b94faa DT |
92 | } |
93 | ||
009d8518 | 94 | const struct address_space_operations gfs2_meta_aops = { |
b3b94faa | 95 | .writepage = gfs2_aspace_writepage, |
4340fe62 | 96 | .releasepage = gfs2_releasepage, |
b3b94faa DT |
97 | }; |
98 | ||
b3b94faa DT |
99 | /** |
100 | * gfs2_meta_sync - Sync all buffers associated with a glock | |
101 | * @gl: The glock | |
b3b94faa DT |
102 | * |
103 | */ | |
104 | ||
7276b3b0 | 105 | void gfs2_meta_sync(struct gfs2_glock *gl) |
b3b94faa | 106 | { |
009d8518 | 107 | struct address_space *mapping = gfs2_glock2aspace(gl); |
7276b3b0 | 108 | int error; |
b3b94faa | 109 | |
7276b3b0 SW |
110 | filemap_fdatawrite(mapping); |
111 | error = filemap_fdatawait(mapping); | |
b3b94faa DT |
112 | |
113 | if (error) | |
114 | gfs2_io_error(gl->gl_sbd); | |
115 | } | |
116 | ||
117 | /** | |
6802e340 | 118 | * gfs2_getbuf - Get a buffer with a given address space |
cb4c0313 | 119 | * @gl: the glock |
b3b94faa DT |
120 | * @blkno: the block number (filesystem scope) |
121 | * @create: 1 if the buffer should be created | |
122 | * | |
123 | * Returns: the buffer | |
124 | */ | |
125 | ||
6802e340 | 126 | struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create) |
b3b94faa | 127 | { |
009d8518 | 128 | struct address_space *mapping = gfs2_glock2aspace(gl); |
cb4c0313 | 129 | struct gfs2_sbd *sdp = gl->gl_sbd; |
b3b94faa DT |
130 | struct page *page; |
131 | struct buffer_head *bh; | |
132 | unsigned int shift; | |
133 | unsigned long index; | |
134 | unsigned int bufnum; | |
135 | ||
136 | shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift; | |
137 | index = blkno >> shift; /* convert block to page */ | |
138 | bufnum = blkno - (index << shift); /* block buf index within page */ | |
139 | ||
140 | if (create) { | |
141 | for (;;) { | |
cb4c0313 | 142 | page = grab_cache_page(mapping, index); |
b3b94faa DT |
143 | if (page) |
144 | break; | |
145 | yield(); | |
146 | } | |
147 | } else { | |
cb4c0313 | 148 | page = find_lock_page(mapping, index); |
b3b94faa DT |
149 | if (!page) |
150 | return NULL; | |
151 | } | |
152 | ||
153 | if (!page_has_buffers(page)) | |
154 | create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); | |
155 | ||
156 | /* Locate header for our buffer within our page */ | |
157 | for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) | |
158 | /* Do nothing */; | |
159 | get_bh(bh); | |
160 | ||
161 | if (!buffer_mapped(bh)) | |
162 | map_bh(bh, sdp->sd_vfs, blkno); | |
163 | ||
164 | unlock_page(page); | |
165 | mark_page_accessed(page); | |
166 | page_cache_release(page); | |
167 | ||
168 | return bh; | |
169 | } | |
170 | ||
171 | static void meta_prep_new(struct buffer_head *bh) | |
172 | { | |
173 | struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; | |
174 | ||
175 | lock_buffer(bh); | |
176 | clear_buffer_dirty(bh); | |
177 | set_buffer_uptodate(bh); | |
178 | unlock_buffer(bh); | |
179 | ||
180 | mh->mh_magic = cpu_to_be32(GFS2_MAGIC); | |
181 | } | |
182 | ||
183 | /** | |
184 | * gfs2_meta_new - Get a block | |
185 | * @gl: The glock associated with this block | |
186 | * @blkno: The block number | |
187 | * | |
188 | * Returns: The buffer | |
189 | */ | |
190 | ||
cd915493 | 191 | struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno) |
b3b94faa DT |
192 | { |
193 | struct buffer_head *bh; | |
6802e340 | 194 | bh = gfs2_getbuf(gl, blkno, CREATE); |
b3b94faa DT |
195 | meta_prep_new(bh); |
196 | return bh; | |
197 | } | |
198 | ||
199 | /** | |
200 | * gfs2_meta_read - Read a block from disk | |
201 | * @gl: The glock covering the block | |
202 | * @blkno: The block number | |
7276b3b0 | 203 | * @flags: flags |
b3b94faa DT |
204 | * @bhp: the place where the buffer is returned (NULL on failure) |
205 | * | |
206 | * Returns: errno | |
207 | */ | |
208 | ||
cd915493 | 209 | int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, |
b3b94faa DT |
210 | struct buffer_head **bhp) |
211 | { | |
c969f58c SW |
212 | struct gfs2_sbd *sdp = gl->gl_sbd; |
213 | struct buffer_head *bh; | |
214 | ||
215 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
216 | return -EIO; | |
217 | ||
218 | *bhp = bh = gfs2_getbuf(gl, blkno, CREATE); | |
219 | ||
220 | lock_buffer(bh); | |
221 | if (buffer_uptodate(bh)) { | |
222 | unlock_buffer(bh); | |
223 | return 0; | |
224 | } | |
225 | bh->b_end_io = end_buffer_read_sync; | |
226 | get_bh(bh); | |
7b6d91da | 227 | submit_bh(READ_SYNC | REQ_META, bh); |
c969f58c SW |
228 | if (!(flags & DIO_WAIT)) |
229 | return 0; | |
230 | ||
231 | wait_on_buffer(bh); | |
232 | if (unlikely(!buffer_uptodate(bh))) { | |
233 | struct gfs2_trans *tr = current->journal_info; | |
234 | if (tr && tr->tr_touched) | |
235 | gfs2_io_error_bh(sdp, bh); | |
236 | brelse(bh); | |
237 | return -EIO; | |
7276b3b0 | 238 | } |
b3b94faa | 239 | |
7276b3b0 | 240 | return 0; |
b3b94faa DT |
241 | } |
242 | ||
243 | /** | |
7276b3b0 | 244 | * gfs2_meta_wait - Reread a block from disk |
b3b94faa | 245 | * @sdp: the filesystem |
7276b3b0 | 246 | * @bh: The block to wait for |
b3b94faa DT |
247 | * |
248 | * Returns: errno | |
249 | */ | |
250 | ||
7276b3b0 | 251 | int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) |
b3b94faa DT |
252 | { |
253 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
254 | return -EIO; | |
255 | ||
7276b3b0 | 256 | wait_on_buffer(bh); |
b3b94faa | 257 | |
7276b3b0 SW |
258 | if (!buffer_uptodate(bh)) { |
259 | struct gfs2_trans *tr = current->journal_info; | |
260 | if (tr && tr->tr_touched) | |
261 | gfs2_io_error_bh(sdp, bh); | |
262 | return -EIO; | |
b3b94faa | 263 | } |
7276b3b0 SW |
264 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
265 | return -EIO; | |
b3b94faa DT |
266 | |
267 | return 0; | |
268 | } | |
269 | ||
270 | /** | |
586dfdaa | 271 | * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer |
b3b94faa DT |
272 | * @gl: the glock the buffer belongs to |
273 | * @bh: The buffer to be attached to | |
586dfdaa | 274 | * @meta: Flag to indicate whether its metadata or not |
b3b94faa DT |
275 | */ |
276 | ||
568f4c96 SW |
277 | void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh, |
278 | int meta) | |
b3b94faa DT |
279 | { |
280 | struct gfs2_bufdata *bd; | |
281 | ||
18ec7d5c SW |
282 | if (meta) |
283 | lock_page(bh->b_page); | |
b3b94faa | 284 | |
5c676f6d | 285 | if (bh->b_private) { |
18ec7d5c SW |
286 | if (meta) |
287 | unlock_page(bh->b_page); | |
b3b94faa DT |
288 | return; |
289 | } | |
290 | ||
3e5cd087 | 291 | bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL); |
b3b94faa DT |
292 | bd->bd_bh = bh; |
293 | bd->bd_gl = gl; | |
294 | ||
295 | INIT_LIST_HEAD(&bd->bd_list_tr); | |
82ffa516 | 296 | if (meta) |
586dfdaa | 297 | lops_init_le(&bd->bd_le, &gfs2_buf_lops); |
82ffa516 | 298 | else |
586dfdaa | 299 | lops_init_le(&bd->bd_le, &gfs2_databuf_lops); |
5c676f6d | 300 | bh->b_private = bd; |
b3b94faa | 301 | |
18ec7d5c SW |
302 | if (meta) |
303 | unlock_page(bh->b_page); | |
b3b94faa DT |
304 | } |
305 | ||
16615be1 SW |
306 | void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta) |
307 | { | |
009d8518 SW |
308 | struct address_space *mapping = bh->b_page->mapping; |
309 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); | |
16615be1 | 310 | struct gfs2_bufdata *bd = bh->b_private; |
009d8518 | 311 | |
16615be1 | 312 | if (test_clear_buffer_pinned(bh)) { |
5e687eac | 313 | atomic_dec(&sdp->sd_log_pinned); |
16615be1 SW |
314 | list_del_init(&bd->bd_le.le_list); |
315 | if (meta) { | |
316 | gfs2_assert_warn(sdp, sdp->sd_log_num_buf); | |
317 | sdp->sd_log_num_buf--; | |
318 | tr->tr_num_buf_rm++; | |
319 | } else { | |
320 | gfs2_assert_warn(sdp, sdp->sd_log_num_databuf); | |
321 | sdp->sd_log_num_databuf--; | |
322 | tr->tr_num_databuf_rm++; | |
323 | } | |
324 | tr->tr_touched = 1; | |
325 | brelse(bh); | |
326 | } | |
327 | if (bd) { | |
328 | if (bd->bd_ail) { | |
f91a0d3e | 329 | gfs2_remove_from_ail(bd); |
16615be1 SW |
330 | bh->b_private = NULL; |
331 | bd->bd_bh = NULL; | |
332 | bd->bd_blkno = bh->b_blocknr; | |
333 | gfs2_trans_add_revoke(sdp, bd); | |
334 | } | |
335 | } | |
336 | clear_buffer_dirty(bh); | |
337 | clear_buffer_uptodate(bh); | |
338 | } | |
339 | ||
b3b94faa DT |
340 | /** |
341 | * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore | |
342 | * @ip: the inode who owns the buffers | |
343 | * @bstart: the first buffer in the run | |
344 | * @blen: the number of buffers in the run | |
345 | * | |
346 | */ | |
347 | ||
cd915493 | 348 | void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen) |
b3b94faa | 349 | { |
feaa7bba | 350 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
351 | struct buffer_head *bh; |
352 | ||
353 | while (blen) { | |
6802e340 | 354 | bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE); |
b3b94faa | 355 | if (bh) { |
1ad38c43 SW |
356 | lock_buffer(bh); |
357 | gfs2_log_lock(sdp); | |
16615be1 | 358 | gfs2_remove_from_journal(bh, current->journal_info, 1); |
1ad38c43 | 359 | gfs2_log_unlock(sdp); |
b3b94faa | 360 | unlock_buffer(bh); |
b3b94faa DT |
361 | brelse(bh); |
362 | } | |
363 | ||
364 | bstart++; | |
365 | blen--; | |
366 | } | |
367 | } | |
368 | ||
b3b94faa DT |
369 | /** |
370 | * gfs2_meta_indirect_buffer - Get a metadata buffer | |
371 | * @ip: The GFS2 inode | |
372 | * @height: The level of this buf in the metadata (indir addr) tree (if any) | |
373 | * @num: The block number (device relative) of the buffer | |
374 | * @new: Non-zero if we may create a new buffer | |
375 | * @bhp: the buffer is returned here | |
376 | * | |
b3b94faa DT |
377 | * Returns: errno |
378 | */ | |
379 | ||
cd915493 | 380 | int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num, |
b3b94faa DT |
381 | int new, struct buffer_head **bhp) |
382 | { | |
7276b3b0 SW |
383 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
384 | struct gfs2_glock *gl = ip->i_gl; | |
f91a0d3e SW |
385 | struct buffer_head *bh; |
386 | int ret = 0; | |
b3b94faa DT |
387 | |
388 | if (new) { | |
f91a0d3e SW |
389 | BUG_ON(height == 0); |
390 | bh = gfs2_meta_new(gl, num); | |
d4e9c4c3 | 391 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
392 | gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
393 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
7276b3b0 SW |
394 | } else { |
395 | u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI; | |
f91a0d3e SW |
396 | ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh); |
397 | if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) { | |
398 | brelse(bh); | |
399 | ret = -EIO; | |
7276b3b0 | 400 | } |
b3b94faa | 401 | } |
b3b94faa | 402 | *bhp = bh; |
f91a0d3e | 403 | return ret; |
b3b94faa DT |
404 | } |
405 | ||
406 | /** | |
407 | * gfs2_meta_ra - start readahead on an extent of a file | |
408 | * @gl: the glock the blocks belong to | |
409 | * @dblock: the starting disk block | |
410 | * @extlen: the number of blocks in the extent | |
411 | * | |
7276b3b0 | 412 | * returns: the first buffer in the extent |
b3b94faa DT |
413 | */ |
414 | ||
7276b3b0 | 415 | struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen) |
b3b94faa DT |
416 | { |
417 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
b3b94faa | 418 | struct buffer_head *first_bh, *bh; |
cd915493 | 419 | u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> |
568f4c96 | 420 | sdp->sd_sb.sb_bsize_shift; |
b3b94faa | 421 | |
7276b3b0 SW |
422 | BUG_ON(!extlen); |
423 | ||
424 | if (max_ra < 1) | |
425 | max_ra = 1; | |
b3b94faa DT |
426 | if (extlen > max_ra) |
427 | extlen = max_ra; | |
428 | ||
6802e340 | 429 | first_bh = gfs2_getbuf(gl, dblock, CREATE); |
b3b94faa DT |
430 | |
431 | if (buffer_uptodate(first_bh)) | |
432 | goto out; | |
7276b3b0 | 433 | if (!buffer_locked(first_bh)) |
7b6d91da | 434 | ll_rw_block(READ_SYNC | REQ_META, 1, &first_bh); |
b3b94faa DT |
435 | |
436 | dblock++; | |
437 | extlen--; | |
438 | ||
439 | while (extlen) { | |
6802e340 | 440 | bh = gfs2_getbuf(gl, dblock, CREATE); |
b3b94faa | 441 | |
7276b3b0 SW |
442 | if (!buffer_uptodate(bh) && !buffer_locked(bh)) |
443 | ll_rw_block(READA, 1, &bh); | |
444 | brelse(bh); | |
b3b94faa DT |
445 | dblock++; |
446 | extlen--; | |
7276b3b0 SW |
447 | if (!buffer_locked(first_bh) && buffer_uptodate(first_bh)) |
448 | goto out; | |
b3b94faa DT |
449 | } |
450 | ||
7276b3b0 | 451 | wait_on_buffer(first_bh); |
a91ea69f | 452 | out: |
7276b3b0 | 453 | return first_bh; |
b3b94faa DT |
454 | } |
455 |