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" |
627c10b7 | 34 | #include "trace_gfs2.h" |
b3b94faa | 35 | |
4a0f9a32 | 36 | static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc) |
b3b94faa | 37 | { |
4a0f9a32 SW |
38 | struct buffer_head *bh, *head; |
39 | int nr_underway = 0; | |
7637241e | 40 | int write_flags = REQ_META | REQ_PRIO | wbc_to_write_flags(wbc); |
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 | |
e76e0ec9 | 54 | * potentially cause a busy-wait loop from flusher thread and kswapd |
4a0f9a32 SW |
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)) { | |
2a222ca9 | 81 | submit_bh(REQ_OP_WRITE, write_flags, bh); |
4a0f9a32 SW |
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 | ||
1b2ad412 SW |
99 | const struct address_space_operations gfs2_rgrp_aops = { |
100 | .writepage = gfs2_aspace_writepage, | |
101 | .releasepage = gfs2_releasepage, | |
102 | }; | |
103 | ||
b3b94faa | 104 | /** |
6802e340 | 105 | * gfs2_getbuf - Get a buffer with a given address space |
cb4c0313 | 106 | * @gl: the glock |
b3b94faa DT |
107 | * @blkno: the block number (filesystem scope) |
108 | * @create: 1 if the buffer should be created | |
109 | * | |
110 | * Returns: the buffer | |
111 | */ | |
112 | ||
6802e340 | 113 | struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create) |
b3b94faa | 114 | { |
009d8518 | 115 | struct address_space *mapping = gfs2_glock2aspace(gl); |
15562c43 | 116 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
b3b94faa DT |
117 | struct page *page; |
118 | struct buffer_head *bh; | |
119 | unsigned int shift; | |
120 | unsigned long index; | |
121 | unsigned int bufnum; | |
122 | ||
70d4ee94 SW |
123 | if (mapping == NULL) |
124 | mapping = &sdp->sd_aspace; | |
125 | ||
09cbfeaf | 126 | shift = PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift; |
b3b94faa DT |
127 | index = blkno >> shift; /* convert block to page */ |
128 | bufnum = blkno - (index << shift); /* block buf index within page */ | |
129 | ||
130 | if (create) { | |
131 | for (;;) { | |
cb4c0313 | 132 | page = grab_cache_page(mapping, index); |
b3b94faa DT |
133 | if (page) |
134 | break; | |
135 | yield(); | |
136 | } | |
137 | } else { | |
2457aec6 MG |
138 | page = find_get_page_flags(mapping, index, |
139 | FGP_LOCK|FGP_ACCESSED); | |
b3b94faa DT |
140 | if (!page) |
141 | return NULL; | |
142 | } | |
143 | ||
144 | if (!page_has_buffers(page)) | |
145 | create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); | |
146 | ||
147 | /* Locate header for our buffer within our page */ | |
148 | for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) | |
149 | /* Do nothing */; | |
150 | get_bh(bh); | |
151 | ||
152 | if (!buffer_mapped(bh)) | |
153 | map_bh(bh, sdp->sd_vfs, blkno); | |
154 | ||
155 | unlock_page(page); | |
09cbfeaf | 156 | put_page(page); |
b3b94faa DT |
157 | |
158 | return bh; | |
159 | } | |
160 | ||
161 | static void meta_prep_new(struct buffer_head *bh) | |
162 | { | |
163 | struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; | |
164 | ||
165 | lock_buffer(bh); | |
166 | clear_buffer_dirty(bh); | |
167 | set_buffer_uptodate(bh); | |
168 | unlock_buffer(bh); | |
169 | ||
170 | mh->mh_magic = cpu_to_be32(GFS2_MAGIC); | |
171 | } | |
172 | ||
173 | /** | |
174 | * gfs2_meta_new - Get a block | |
175 | * @gl: The glock associated with this block | |
176 | * @blkno: The block number | |
177 | * | |
178 | * Returns: The buffer | |
179 | */ | |
180 | ||
cd915493 | 181 | struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno) |
b3b94faa DT |
182 | { |
183 | struct buffer_head *bh; | |
6802e340 | 184 | bh = gfs2_getbuf(gl, blkno, CREATE); |
b3b94faa DT |
185 | meta_prep_new(bh); |
186 | return bh; | |
187 | } | |
188 | ||
39b0555f | 189 | static void gfs2_meta_read_endio(struct bio *bio) |
c8d57703 | 190 | { |
39b0555f AG |
191 | struct bio_vec *bvec; |
192 | int i; | |
193 | ||
194 | bio_for_each_segment_all(bvec, bio, i) { | |
195 | struct page *page = bvec->bv_page; | |
196 | struct buffer_head *bh = page_buffers(page); | |
197 | unsigned int len = bvec->bv_len; | |
198 | ||
199 | while (bh_offset(bh) < bvec->bv_offset) | |
200 | bh = bh->b_this_page; | |
201 | do { | |
202 | struct buffer_head *next = bh->b_this_page; | |
203 | len -= bh->b_size; | |
4e4cbee9 | 204 | bh->b_end_io(bh, !bio->bi_status); |
39b0555f AG |
205 | bh = next; |
206 | } while (bh && len); | |
207 | } | |
208 | bio_put(bio); | |
209 | } | |
c8d57703 | 210 | |
39b0555f AG |
211 | /* |
212 | * Submit several consecutive buffer head I/O requests as a single bio I/O | |
213 | * request. (See submit_bh_wbc.) | |
214 | */ | |
e1b1afa6 MC |
215 | static void gfs2_submit_bhs(int op, int op_flags, struct buffer_head *bhs[], |
216 | int num) | |
39b0555f | 217 | { |
23e5671a AG |
218 | while (num > 0) { |
219 | struct buffer_head *bh = *bhs; | |
220 | struct bio *bio; | |
221 | ||
222 | bio = bio_alloc(GFP_NOIO, num); | |
223 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
224 | bio->bi_bdev = bh->b_bdev; | |
225 | while (num > 0) { | |
226 | bh = *bhs; | |
227 | if (!bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh))) { | |
228 | BUG_ON(bio->bi_iter.bi_size == 0); | |
229 | break; | |
230 | } | |
231 | bhs++; | |
232 | num--; | |
233 | } | |
234 | bio->bi_end_io = gfs2_meta_read_endio; | |
235 | bio_set_op_attrs(bio, op, op_flags); | |
236 | submit_bio(bio); | |
c8d57703 | 237 | } |
c8d57703 AG |
238 | } |
239 | ||
b3b94faa DT |
240 | /** |
241 | * gfs2_meta_read - Read a block from disk | |
242 | * @gl: The glock covering the block | |
243 | * @blkno: The block number | |
7276b3b0 | 244 | * @flags: flags |
b3b94faa DT |
245 | * @bhp: the place where the buffer is returned (NULL on failure) |
246 | * | |
247 | * Returns: errno | |
248 | */ | |
249 | ||
cd915493 | 250 | int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, |
c8d57703 | 251 | int rahead, struct buffer_head **bhp) |
b3b94faa | 252 | { |
15562c43 | 253 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
39b0555f AG |
254 | struct buffer_head *bh, *bhs[2]; |
255 | int num = 0; | |
c969f58c | 256 | |
44b8db13 MY |
257 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { |
258 | *bhp = NULL; | |
c969f58c | 259 | return -EIO; |
44b8db13 | 260 | } |
c969f58c SW |
261 | |
262 | *bhp = bh = gfs2_getbuf(gl, blkno, CREATE); | |
263 | ||
264 | lock_buffer(bh); | |
265 | if (buffer_uptodate(bh)) { | |
266 | unlock_buffer(bh); | |
39b0555f AG |
267 | flags &= ~DIO_WAIT; |
268 | } else { | |
269 | bh->b_end_io = end_buffer_read_sync; | |
270 | get_bh(bh); | |
271 | bhs[num++] = bh; | |
c969f58c | 272 | } |
39b0555f AG |
273 | |
274 | if (rahead) { | |
275 | bh = gfs2_getbuf(gl, blkno + 1, CREATE); | |
276 | ||
277 | lock_buffer(bh); | |
278 | if (buffer_uptodate(bh)) { | |
279 | unlock_buffer(bh); | |
280 | brelse(bh); | |
281 | } else { | |
282 | bh->b_end_io = end_buffer_read_sync; | |
283 | bhs[num++] = bh; | |
284 | } | |
285 | } | |
286 | ||
70fd7614 | 287 | gfs2_submit_bhs(REQ_OP_READ, REQ_META | REQ_PRIO, bhs, num); |
c969f58c SW |
288 | if (!(flags & DIO_WAIT)) |
289 | return 0; | |
290 | ||
39b0555f | 291 | bh = *bhp; |
c969f58c SW |
292 | wait_on_buffer(bh); |
293 | if (unlikely(!buffer_uptodate(bh))) { | |
294 | struct gfs2_trans *tr = current->journal_info; | |
9862ca05 | 295 | if (tr && test_bit(TR_TOUCHED, &tr->tr_flags)) |
c969f58c SW |
296 | gfs2_io_error_bh(sdp, bh); |
297 | brelse(bh); | |
44b8db13 | 298 | *bhp = NULL; |
c969f58c | 299 | return -EIO; |
7276b3b0 | 300 | } |
b3b94faa | 301 | |
7276b3b0 | 302 | return 0; |
b3b94faa DT |
303 | } |
304 | ||
305 | /** | |
7276b3b0 | 306 | * gfs2_meta_wait - Reread a block from disk |
b3b94faa | 307 | * @sdp: the filesystem |
7276b3b0 | 308 | * @bh: The block to wait for |
b3b94faa DT |
309 | * |
310 | * Returns: errno | |
311 | */ | |
312 | ||
7276b3b0 | 313 | int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) |
b3b94faa DT |
314 | { |
315 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
316 | return -EIO; | |
317 | ||
7276b3b0 | 318 | wait_on_buffer(bh); |
b3b94faa | 319 | |
7276b3b0 SW |
320 | if (!buffer_uptodate(bh)) { |
321 | struct gfs2_trans *tr = current->journal_info; | |
9862ca05 | 322 | if (tr && test_bit(TR_TOUCHED, &tr->tr_flags)) |
7276b3b0 SW |
323 | gfs2_io_error_bh(sdp, bh); |
324 | return -EIO; | |
b3b94faa | 325 | } |
7276b3b0 SW |
326 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
327 | return -EIO; | |
b3b94faa DT |
328 | |
329 | return 0; | |
330 | } | |
331 | ||
68cd4ce2 | 332 | void gfs2_remove_from_journal(struct buffer_head *bh, int meta) |
16615be1 | 333 | { |
009d8518 SW |
334 | struct address_space *mapping = bh->b_page->mapping; |
335 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); | |
16615be1 | 336 | struct gfs2_bufdata *bd = bh->b_private; |
68cd4ce2 | 337 | struct gfs2_trans *tr = current->journal_info; |
502be2a3 | 338 | int was_pinned = 0; |
009d8518 | 339 | |
16615be1 | 340 | if (test_clear_buffer_pinned(bh)) { |
627c10b7 | 341 | trace_gfs2_pin(bd, 0); |
5e687eac | 342 | atomic_dec(&sdp->sd_log_pinned); |
c0752aa7 | 343 | list_del_init(&bd->bd_list); |
68cd4ce2 | 344 | if (meta == REMOVE_META) |
16615be1 | 345 | tr->tr_num_buf_rm++; |
022ef4fe | 346 | else |
16615be1 | 347 | tr->tr_num_databuf_rm++; |
9862ca05 | 348 | set_bit(TR_TOUCHED, &tr->tr_flags); |
502be2a3 | 349 | was_pinned = 1; |
16615be1 SW |
350 | brelse(bh); |
351 | } | |
352 | if (bd) { | |
c618e87a | 353 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 | 354 | if (bd->bd_tr) { |
16615be1 | 355 | gfs2_trans_add_revoke(sdp, bd); |
502be2a3 BP |
356 | } else if (was_pinned) { |
357 | bh->b_private = NULL; | |
358 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
16615be1 | 359 | } |
c618e87a | 360 | spin_unlock(&sdp->sd_ail_lock); |
16615be1 SW |
361 | } |
362 | clear_buffer_dirty(bh); | |
363 | clear_buffer_uptodate(bh); | |
364 | } | |
365 | ||
b3b94faa DT |
366 | /** |
367 | * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore | |
368 | * @ip: the inode who owns the buffers | |
369 | * @bstart: the first buffer in the run | |
370 | * @blen: the number of buffers in the run | |
371 | * | |
372 | */ | |
373 | ||
cd915493 | 374 | void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen) |
b3b94faa | 375 | { |
feaa7bba | 376 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
377 | struct buffer_head *bh; |
378 | ||
379 | while (blen) { | |
6802e340 | 380 | bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE); |
b3b94faa | 381 | if (bh) { |
1ad38c43 SW |
382 | lock_buffer(bh); |
383 | gfs2_log_lock(sdp); | |
68cd4ce2 | 384 | gfs2_remove_from_journal(bh, REMOVE_META); |
1ad38c43 | 385 | gfs2_log_unlock(sdp); |
b3b94faa | 386 | unlock_buffer(bh); |
b3b94faa DT |
387 | brelse(bh); |
388 | } | |
389 | ||
390 | bstart++; | |
391 | blen--; | |
392 | } | |
393 | } | |
394 | ||
b3b94faa DT |
395 | /** |
396 | * gfs2_meta_indirect_buffer - Get a metadata buffer | |
397 | * @ip: The GFS2 inode | |
398 | * @height: The level of this buf in the metadata (indir addr) tree (if any) | |
399 | * @num: The block number (device relative) of the buffer | |
b3b94faa DT |
400 | * @bhp: the buffer is returned here |
401 | * | |
b3b94faa DT |
402 | * Returns: errno |
403 | */ | |
404 | ||
cd915493 | 405 | int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num, |
f2f9c812 | 406 | struct buffer_head **bhp) |
b3b94faa | 407 | { |
7276b3b0 SW |
408 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
409 | struct gfs2_glock *gl = ip->i_gl; | |
f91a0d3e SW |
410 | struct buffer_head *bh; |
411 | int ret = 0; | |
f2f9c812 | 412 | u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI; |
c8d57703 AG |
413 | int rahead = 0; |
414 | ||
415 | if (num == ip->i_no_addr) | |
416 | rahead = ip->i_rahead; | |
b3b94faa | 417 | |
c8d57703 | 418 | ret = gfs2_meta_read(gl, num, DIO_WAIT, rahead, &bh); |
f2f9c812 BP |
419 | if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) { |
420 | brelse(bh); | |
421 | ret = -EIO; | |
b3b94faa | 422 | } |
b3b94faa | 423 | *bhp = bh; |
f91a0d3e | 424 | return ret; |
b3b94faa DT |
425 | } |
426 | ||
427 | /** | |
428 | * gfs2_meta_ra - start readahead on an extent of a file | |
429 | * @gl: the glock the blocks belong to | |
430 | * @dblock: the starting disk block | |
431 | * @extlen: the number of blocks in the extent | |
432 | * | |
7276b3b0 | 433 | * returns: the first buffer in the extent |
b3b94faa DT |
434 | */ |
435 | ||
7276b3b0 | 436 | struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen) |
b3b94faa | 437 | { |
15562c43 | 438 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
b3b94faa | 439 | struct buffer_head *first_bh, *bh; |
cd915493 | 440 | u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> |
568f4c96 | 441 | sdp->sd_sb.sb_bsize_shift; |
b3b94faa | 442 | |
7276b3b0 SW |
443 | BUG_ON(!extlen); |
444 | ||
445 | if (max_ra < 1) | |
446 | max_ra = 1; | |
b3b94faa DT |
447 | if (extlen > max_ra) |
448 | extlen = max_ra; | |
449 | ||
6802e340 | 450 | first_bh = gfs2_getbuf(gl, dblock, CREATE); |
b3b94faa DT |
451 | |
452 | if (buffer_uptodate(first_bh)) | |
453 | goto out; | |
7276b3b0 | 454 | if (!buffer_locked(first_bh)) |
70fd7614 | 455 | ll_rw_block(REQ_OP_READ, REQ_META, 1, &first_bh); |
b3b94faa DT |
456 | |
457 | dblock++; | |
458 | extlen--; | |
459 | ||
460 | while (extlen) { | |
6802e340 | 461 | bh = gfs2_getbuf(gl, dblock, CREATE); |
b3b94faa | 462 | |
7276b3b0 | 463 | if (!buffer_uptodate(bh) && !buffer_locked(bh)) |
70246286 | 464 | ll_rw_block(REQ_OP_READ, REQ_RAHEAD | REQ_META, 1, &bh); |
7276b3b0 | 465 | brelse(bh); |
b3b94faa DT |
466 | dblock++; |
467 | extlen--; | |
7276b3b0 SW |
468 | if (!buffer_locked(first_bh) && buffer_uptodate(first_bh)) |
469 | goto out; | |
b3b94faa DT |
470 | } |
471 | ||
7276b3b0 | 472 | wait_on_buffer(first_bh); |
a91ea69f | 473 | out: |
7276b3b0 | 474 | return first_bh; |
b3b94faa DT |
475 | } |
476 |