Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 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> |
7d308590 | 22 | #include <linux/lm_interface.h> |
b3b94faa DT |
23 | |
24 | #include "gfs2.h" | |
5c676f6d | 25 | #include "incore.h" |
b3b94faa DT |
26 | #include "glock.h" |
27 | #include "glops.h" | |
28 | #include "inode.h" | |
29 | #include "log.h" | |
30 | #include "lops.h" | |
31 | #include "meta_io.h" | |
32 | #include "rgrp.h" | |
33 | #include "trans.h" | |
5c676f6d | 34 | #include "util.h" |
4340fe62 | 35 | #include "ops_address.h" |
b3b94faa | 36 | |
b3b94faa DT |
37 | static int aspace_get_block(struct inode *inode, sector_t lblock, |
38 | struct buffer_head *bh_result, int create) | |
39 | { | |
5c676f6d | 40 | gfs2_assert_warn(inode->i_sb->s_fs_info, 0); |
b3b94faa DT |
41 | return -EOPNOTSUPP; |
42 | } | |
43 | ||
44 | static int gfs2_aspace_writepage(struct page *page, | |
45 | struct writeback_control *wbc) | |
46 | { | |
47 | return block_write_full_page(page, aspace_get_block, wbc); | |
48 | } | |
49 | ||
66de045d | 50 | static const struct address_space_operations aspace_aops = { |
b3b94faa | 51 | .writepage = gfs2_aspace_writepage, |
4340fe62 | 52 | .releasepage = gfs2_releasepage, |
b3b94faa DT |
53 | }; |
54 | ||
55 | /** | |
56 | * gfs2_aspace_get - Create and initialize a struct inode structure | |
57 | * @sdp: the filesystem the aspace is in | |
58 | * | |
59 | * Right now a struct inode is just a struct inode. Maybe Linux | |
60 | * will supply a more lightweight address space construct (that works) | |
61 | * in the future. | |
62 | * | |
63 | * Make sure pages/buffers in this aspace aren't in high memory. | |
64 | * | |
65 | * Returns: the aspace | |
66 | */ | |
67 | ||
68 | struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp) | |
69 | { | |
70 | struct inode *aspace; | |
71 | ||
72 | aspace = new_inode(sdp->sd_vfs); | |
73 | if (aspace) { | |
f3bba03f | 74 | mapping_set_gfp_mask(aspace->i_mapping, GFP_NOFS); |
b3b94faa DT |
75 | aspace->i_mapping->a_ops = &aspace_aops; |
76 | aspace->i_size = ~0ULL; | |
bba9dfd8 | 77 | aspace->i_private = NULL; |
b3b94faa DT |
78 | insert_inode_hash(aspace); |
79 | } | |
b3b94faa DT |
80 | return aspace; |
81 | } | |
82 | ||
83 | void gfs2_aspace_put(struct inode *aspace) | |
84 | { | |
85 | remove_inode_hash(aspace); | |
86 | iput(aspace); | |
87 | } | |
88 | ||
b3b94faa DT |
89 | /** |
90 | * gfs2_meta_inval - Invalidate all buffers associated with a glock | |
91 | * @gl: the glock | |
92 | * | |
93 | */ | |
94 | ||
95 | void gfs2_meta_inval(struct gfs2_glock *gl) | |
96 | { | |
97 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
98 | struct inode *aspace = gl->gl_aspace; | |
99 | struct address_space *mapping = gl->gl_aspace->i_mapping; | |
100 | ||
101 | gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count)); | |
102 | ||
103 | atomic_inc(&aspace->i_writecount); | |
104 | truncate_inode_pages(mapping, 0); | |
105 | atomic_dec(&aspace->i_writecount); | |
106 | ||
107 | gfs2_assert_withdraw(sdp, !mapping->nrpages); | |
108 | } | |
109 | ||
110 | /** | |
111 | * gfs2_meta_sync - Sync all buffers associated with a glock | |
112 | * @gl: The glock | |
b3b94faa DT |
113 | * |
114 | */ | |
115 | ||
7276b3b0 | 116 | void gfs2_meta_sync(struct gfs2_glock *gl) |
b3b94faa DT |
117 | { |
118 | struct address_space *mapping = gl->gl_aspace->i_mapping; | |
7276b3b0 | 119 | int error; |
b3b94faa | 120 | |
7276b3b0 SW |
121 | filemap_fdatawrite(mapping); |
122 | error = filemap_fdatawait(mapping); | |
b3b94faa DT |
123 | |
124 | if (error) | |
125 | gfs2_io_error(gl->gl_sbd); | |
126 | } | |
127 | ||
128 | /** | |
129 | * getbuf - Get a buffer with a given address space | |
cb4c0313 | 130 | * @gl: the glock |
b3b94faa DT |
131 | * @blkno: the block number (filesystem scope) |
132 | * @create: 1 if the buffer should be created | |
133 | * | |
134 | * Returns: the buffer | |
135 | */ | |
136 | ||
cb4c0313 | 137 | static struct buffer_head *getbuf(struct gfs2_glock *gl, u64 blkno, int create) |
b3b94faa | 138 | { |
cb4c0313 SW |
139 | struct address_space *mapping = gl->gl_aspace->i_mapping; |
140 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
b3b94faa DT |
141 | struct page *page; |
142 | struct buffer_head *bh; | |
143 | unsigned int shift; | |
144 | unsigned long index; | |
145 | unsigned int bufnum; | |
146 | ||
147 | shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift; | |
148 | index = blkno >> shift; /* convert block to page */ | |
149 | bufnum = blkno - (index << shift); /* block buf index within page */ | |
150 | ||
151 | if (create) { | |
152 | for (;;) { | |
cb4c0313 | 153 | page = grab_cache_page(mapping, index); |
b3b94faa DT |
154 | if (page) |
155 | break; | |
156 | yield(); | |
157 | } | |
158 | } else { | |
cb4c0313 | 159 | page = find_lock_page(mapping, index); |
b3b94faa DT |
160 | if (!page) |
161 | return NULL; | |
162 | } | |
163 | ||
164 | if (!page_has_buffers(page)) | |
165 | create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); | |
166 | ||
167 | /* Locate header for our buffer within our page */ | |
168 | for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) | |
169 | /* Do nothing */; | |
170 | get_bh(bh); | |
171 | ||
172 | if (!buffer_mapped(bh)) | |
173 | map_bh(bh, sdp->sd_vfs, blkno); | |
174 | ||
175 | unlock_page(page); | |
176 | mark_page_accessed(page); | |
177 | page_cache_release(page); | |
178 | ||
179 | return bh; | |
180 | } | |
181 | ||
182 | static void meta_prep_new(struct buffer_head *bh) | |
183 | { | |
184 | struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; | |
185 | ||
186 | lock_buffer(bh); | |
187 | clear_buffer_dirty(bh); | |
188 | set_buffer_uptodate(bh); | |
189 | unlock_buffer(bh); | |
190 | ||
191 | mh->mh_magic = cpu_to_be32(GFS2_MAGIC); | |
192 | } | |
193 | ||
194 | /** | |
195 | * gfs2_meta_new - Get a block | |
196 | * @gl: The glock associated with this block | |
197 | * @blkno: The block number | |
198 | * | |
199 | * Returns: The buffer | |
200 | */ | |
201 | ||
cd915493 | 202 | struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno) |
b3b94faa DT |
203 | { |
204 | struct buffer_head *bh; | |
cb4c0313 | 205 | bh = getbuf(gl, blkno, CREATE); |
b3b94faa DT |
206 | meta_prep_new(bh); |
207 | return bh; | |
208 | } | |
209 | ||
210 | /** | |
211 | * gfs2_meta_read - Read a block from disk | |
212 | * @gl: The glock covering the block | |
213 | * @blkno: The block number | |
7276b3b0 | 214 | * @flags: flags |
b3b94faa DT |
215 | * @bhp: the place where the buffer is returned (NULL on failure) |
216 | * | |
217 | * Returns: errno | |
218 | */ | |
219 | ||
cd915493 | 220 | int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, |
b3b94faa DT |
221 | struct buffer_head **bhp) |
222 | { | |
cb4c0313 | 223 | *bhp = getbuf(gl, blkno, CREATE); |
7276b3b0 | 224 | if (!buffer_uptodate(*bhp)) |
2e565bb6 | 225 | ll_rw_block(READ_META, 1, bhp); |
7276b3b0 SW |
226 | if (flags & DIO_WAIT) { |
227 | int error = gfs2_meta_wait(gl->gl_sbd, *bhp); | |
228 | if (error) { | |
229 | brelse(*bhp); | |
230 | return error; | |
231 | } | |
232 | } | |
b3b94faa | 233 | |
7276b3b0 | 234 | return 0; |
b3b94faa DT |
235 | } |
236 | ||
237 | /** | |
7276b3b0 | 238 | * gfs2_meta_wait - Reread a block from disk |
b3b94faa | 239 | * @sdp: the filesystem |
7276b3b0 | 240 | * @bh: The block to wait for |
b3b94faa DT |
241 | * |
242 | * Returns: errno | |
243 | */ | |
244 | ||
7276b3b0 | 245 | int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) |
b3b94faa DT |
246 | { |
247 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
248 | return -EIO; | |
249 | ||
7276b3b0 | 250 | wait_on_buffer(bh); |
b3b94faa | 251 | |
7276b3b0 SW |
252 | if (!buffer_uptodate(bh)) { |
253 | struct gfs2_trans *tr = current->journal_info; | |
254 | if (tr && tr->tr_touched) | |
255 | gfs2_io_error_bh(sdp, bh); | |
256 | return -EIO; | |
b3b94faa | 257 | } |
7276b3b0 SW |
258 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
259 | return -EIO; | |
b3b94faa DT |
260 | |
261 | return 0; | |
262 | } | |
263 | ||
264 | /** | |
586dfdaa | 265 | * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer |
b3b94faa DT |
266 | * @gl: the glock the buffer belongs to |
267 | * @bh: The buffer to be attached to | |
586dfdaa | 268 | * @meta: Flag to indicate whether its metadata or not |
b3b94faa DT |
269 | */ |
270 | ||
568f4c96 SW |
271 | void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh, |
272 | int meta) | |
b3b94faa DT |
273 | { |
274 | struct gfs2_bufdata *bd; | |
275 | ||
18ec7d5c SW |
276 | if (meta) |
277 | lock_page(bh->b_page); | |
b3b94faa | 278 | |
5c676f6d | 279 | if (bh->b_private) { |
18ec7d5c SW |
280 | if (meta) |
281 | unlock_page(bh->b_page); | |
b3b94faa DT |
282 | return; |
283 | } | |
284 | ||
c3762229 | 285 | bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL), |
b3b94faa DT |
286 | bd->bd_bh = bh; |
287 | bd->bd_gl = gl; | |
288 | ||
289 | INIT_LIST_HEAD(&bd->bd_list_tr); | |
82ffa516 | 290 | if (meta) |
586dfdaa | 291 | lops_init_le(&bd->bd_le, &gfs2_buf_lops); |
82ffa516 | 292 | else |
586dfdaa | 293 | lops_init_le(&bd->bd_le, &gfs2_databuf_lops); |
5c676f6d | 294 | bh->b_private = bd; |
b3b94faa | 295 | |
18ec7d5c SW |
296 | if (meta) |
297 | unlock_page(bh->b_page); | |
b3b94faa DT |
298 | } |
299 | ||
16615be1 SW |
300 | void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta) |
301 | { | |
302 | struct gfs2_sbd *sdp = GFS2_SB(bh->b_page->mapping->host); | |
303 | struct gfs2_bufdata *bd = bh->b_private; | |
304 | if (test_clear_buffer_pinned(bh)) { | |
305 | list_del_init(&bd->bd_le.le_list); | |
306 | if (meta) { | |
307 | gfs2_assert_warn(sdp, sdp->sd_log_num_buf); | |
308 | sdp->sd_log_num_buf--; | |
309 | tr->tr_num_buf_rm++; | |
310 | } else { | |
311 | gfs2_assert_warn(sdp, sdp->sd_log_num_databuf); | |
312 | sdp->sd_log_num_databuf--; | |
313 | tr->tr_num_databuf_rm++; | |
314 | } | |
315 | tr->tr_touched = 1; | |
316 | brelse(bh); | |
317 | } | |
318 | if (bd) { | |
319 | if (bd->bd_ail) { | |
320 | gfs2_remove_from_ail(NULL, bd); | |
321 | bh->b_private = NULL; | |
322 | bd->bd_bh = NULL; | |
323 | bd->bd_blkno = bh->b_blocknr; | |
324 | gfs2_trans_add_revoke(sdp, bd); | |
325 | } | |
326 | } | |
327 | clear_buffer_dirty(bh); | |
328 | clear_buffer_uptodate(bh); | |
329 | } | |
330 | ||
b3b94faa DT |
331 | /** |
332 | * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore | |
333 | * @ip: the inode who owns the buffers | |
334 | * @bstart: the first buffer in the run | |
335 | * @blen: the number of buffers in the run | |
336 | * | |
337 | */ | |
338 | ||
cd915493 | 339 | void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen) |
b3b94faa | 340 | { |
feaa7bba | 341 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
b3b94faa DT |
342 | struct buffer_head *bh; |
343 | ||
344 | while (blen) { | |
cb4c0313 | 345 | bh = getbuf(ip->i_gl, bstart, NO_CREATE); |
b3b94faa | 346 | if (bh) { |
1ad38c43 SW |
347 | lock_buffer(bh); |
348 | gfs2_log_lock(sdp); | |
16615be1 | 349 | gfs2_remove_from_journal(bh, current->journal_info, 1); |
1ad38c43 | 350 | gfs2_log_unlock(sdp); |
b3b94faa | 351 | unlock_buffer(bh); |
b3b94faa DT |
352 | brelse(bh); |
353 | } | |
354 | ||
355 | bstart++; | |
356 | blen--; | |
357 | } | |
358 | } | |
359 | ||
360 | /** | |
361 | * gfs2_meta_cache_flush - get rid of any references on buffers for this inode | |
362 | * @ip: The GFS2 inode | |
363 | * | |
364 | * This releases buffers that are in the most-recently-used array of | |
365 | * blocks used for indirect block addressing for this inode. | |
366 | */ | |
367 | ||
368 | void gfs2_meta_cache_flush(struct gfs2_inode *ip) | |
369 | { | |
370 | struct buffer_head **bh_slot; | |
371 | unsigned int x; | |
372 | ||
373 | spin_lock(&ip->i_spin); | |
374 | ||
375 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) { | |
376 | bh_slot = &ip->i_cache[x]; | |
de986e85 WC |
377 | if (*bh_slot) { |
378 | brelse(*bh_slot); | |
379 | *bh_slot = NULL; | |
380 | } | |
b3b94faa DT |
381 | } |
382 | ||
383 | spin_unlock(&ip->i_spin); | |
384 | } | |
385 | ||
386 | /** | |
387 | * gfs2_meta_indirect_buffer - Get a metadata buffer | |
388 | * @ip: The GFS2 inode | |
389 | * @height: The level of this buf in the metadata (indir addr) tree (if any) | |
390 | * @num: The block number (device relative) of the buffer | |
391 | * @new: Non-zero if we may create a new buffer | |
392 | * @bhp: the buffer is returned here | |
393 | * | |
394 | * Try to use the gfs2_inode's MRU metadata tree cache. | |
395 | * | |
396 | * Returns: errno | |
397 | */ | |
398 | ||
cd915493 | 399 | int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num, |
b3b94faa DT |
400 | int new, struct buffer_head **bhp) |
401 | { | |
7276b3b0 SW |
402 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
403 | struct gfs2_glock *gl = ip->i_gl; | |
404 | struct buffer_head *bh = NULL, **bh_slot = ip->i_cache + height; | |
405 | int in_cache = 0; | |
b3b94faa | 406 | |
b004157a SW |
407 | BUG_ON(!gl); |
408 | BUG_ON(!sdp); | |
409 | ||
b3b94faa | 410 | spin_lock(&ip->i_spin); |
7276b3b0 SW |
411 | if (*bh_slot && (*bh_slot)->b_blocknr == num) { |
412 | bh = *bh_slot; | |
413 | get_bh(bh); | |
414 | in_cache = 1; | |
b3b94faa DT |
415 | } |
416 | spin_unlock(&ip->i_spin); | |
417 | ||
7276b3b0 | 418 | if (!bh) |
cb4c0313 | 419 | bh = getbuf(gl, num, CREATE); |
b3b94faa | 420 | |
7276b3b0 SW |
421 | if (!bh) |
422 | return -ENOBUFS; | |
b3b94faa DT |
423 | |
424 | if (new) { | |
7276b3b0 SW |
425 | if (gfs2_assert_warn(sdp, height)) |
426 | goto err; | |
427 | meta_prep_new(bh); | |
d4e9c4c3 | 428 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
429 | gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
430 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
7276b3b0 SW |
431 | } else { |
432 | u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI; | |
433 | if (!buffer_uptodate(bh)) { | |
2e565bb6 | 434 | ll_rw_block(READ_META, 1, &bh); |
7276b3b0 SW |
435 | if (gfs2_meta_wait(sdp, bh)) |
436 | goto err; | |
437 | } | |
438 | if (gfs2_metatype_check(sdp, bh, mtype)) | |
439 | goto err; | |
440 | } | |
b3b94faa | 441 | |
7276b3b0 SW |
442 | if (!in_cache) { |
443 | spin_lock(&ip->i_spin); | |
444 | if (*bh_slot) | |
445 | brelse(*bh_slot); | |
446 | *bh_slot = bh; | |
447 | get_bh(bh); | |
448 | spin_unlock(&ip->i_spin); | |
b3b94faa DT |
449 | } |
450 | ||
451 | *bhp = bh; | |
b3b94faa | 452 | return 0; |
7276b3b0 SW |
453 | err: |
454 | brelse(bh); | |
455 | return -EIO; | |
b3b94faa DT |
456 | } |
457 | ||
458 | /** | |
459 | * gfs2_meta_ra - start readahead on an extent of a file | |
460 | * @gl: the glock the blocks belong to | |
461 | * @dblock: the starting disk block | |
462 | * @extlen: the number of blocks in the extent | |
463 | * | |
7276b3b0 | 464 | * returns: the first buffer in the extent |
b3b94faa DT |
465 | */ |
466 | ||
7276b3b0 | 467 | struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen) |
b3b94faa DT |
468 | { |
469 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
b3b94faa | 470 | struct buffer_head *first_bh, *bh; |
cd915493 | 471 | u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> |
568f4c96 | 472 | sdp->sd_sb.sb_bsize_shift; |
b3b94faa | 473 | |
7276b3b0 SW |
474 | BUG_ON(!extlen); |
475 | ||
476 | if (max_ra < 1) | |
477 | max_ra = 1; | |
b3b94faa DT |
478 | if (extlen > max_ra) |
479 | extlen = max_ra; | |
480 | ||
cb4c0313 | 481 | first_bh = getbuf(gl, dblock, CREATE); |
b3b94faa DT |
482 | |
483 | if (buffer_uptodate(first_bh)) | |
484 | goto out; | |
7276b3b0 | 485 | if (!buffer_locked(first_bh)) |
2e565bb6 | 486 | ll_rw_block(READ_META, 1, &first_bh); |
b3b94faa DT |
487 | |
488 | dblock++; | |
489 | extlen--; | |
490 | ||
491 | while (extlen) { | |
cb4c0313 | 492 | bh = getbuf(gl, dblock, CREATE); |
b3b94faa | 493 | |
7276b3b0 SW |
494 | if (!buffer_uptodate(bh) && !buffer_locked(bh)) |
495 | ll_rw_block(READA, 1, &bh); | |
496 | brelse(bh); | |
b3b94faa DT |
497 | dblock++; |
498 | extlen--; | |
7276b3b0 SW |
499 | if (!buffer_locked(first_bh) && buffer_uptodate(first_bh)) |
500 | goto out; | |
b3b94faa DT |
501 | } |
502 | ||
7276b3b0 | 503 | wait_on_buffer(first_bh); |
a91ea69f | 504 | out: |
7276b3b0 | 505 | return first_bh; |
b3b94faa DT |
506 | } |
507 |