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 | |
7 | * of the GNU General Public License v.2. | |
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> | |
5c676f6d | 20 | #include <linux/gfs2_ondisk.h> |
b3b94faa DT |
21 | |
22 | #include "gfs2.h" | |
5c676f6d SW |
23 | #include "lm_interface.h" |
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 DT |
34 | |
35 | #define buffer_busy(bh) \ | |
36 | ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock) | (1ul << BH_Pinned))) | |
37 | #define buffer_in_io(bh) \ | |
38 | ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock))) | |
39 | ||
40 | static int aspace_get_block(struct inode *inode, sector_t lblock, | |
41 | struct buffer_head *bh_result, int create) | |
42 | { | |
5c676f6d | 43 | gfs2_assert_warn(inode->i_sb->s_fs_info, 0); |
b3b94faa DT |
44 | return -EOPNOTSUPP; |
45 | } | |
46 | ||
47 | static int gfs2_aspace_writepage(struct page *page, | |
48 | struct writeback_control *wbc) | |
49 | { | |
50 | return block_write_full_page(page, aspace_get_block, wbc); | |
51 | } | |
52 | ||
53 | /** | |
54 | * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out. | |
55 | * @bh: the buffer we're stuck on | |
56 | * | |
57 | */ | |
58 | ||
59 | static void stuck_releasepage(struct buffer_head *bh) | |
60 | { | |
b09e593d SW |
61 | struct inode *inode = bh->b_page->mapping->host; |
62 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
5c676f6d | 63 | struct gfs2_bufdata *bd = bh->b_private; |
b3b94faa DT |
64 | struct gfs2_glock *gl; |
65 | ||
b09e593d | 66 | fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode); |
b3b94faa | 67 | fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n", |
382066da | 68 | (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count)); |
b3b94faa | 69 | fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh)); |
5c676f6d | 70 | fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL"); |
b3b94faa DT |
71 | |
72 | if (!bd) | |
73 | return; | |
74 | ||
75 | gl = bd->bd_gl; | |
76 | ||
77 | fs_warn(sdp, "gl = (%u, %llu)\n", | |
382066da | 78 | gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number); |
b3b94faa DT |
79 | |
80 | fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n", | |
81 | (list_empty(&bd->bd_list_tr)) ? "no" : "yes", | |
82 | (list_empty(&bd->bd_le.le_list)) ? "no" : "yes"); | |
83 | ||
84 | if (gl->gl_ops == &gfs2_inode_glops) { | |
5c676f6d | 85 | struct gfs2_inode *ip = gl->gl_object; |
b3b94faa DT |
86 | unsigned int x; |
87 | ||
88 | if (!ip) | |
89 | return; | |
90 | ||
91 | fs_warn(sdp, "ip = %llu %llu\n", | |
382066da SW |
92 | (unsigned long long)ip->i_num.no_formal_ino, |
93 | (unsigned long long)ip->i_num.no_addr); | |
b3b94faa DT |
94 | fs_warn(sdp, "ip->i_count = %d, ip->i_vnode = %s\n", |
95 | atomic_read(&ip->i_count), | |
96 | (ip->i_vnode) ? "!NULL" : "NULL"); | |
97 | ||
98 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) | |
99 | fs_warn(sdp, "ip->i_cache[%u] = %s\n", | |
100 | x, (ip->i_cache[x]) ? "!NULL" : "NULL"); | |
101 | } | |
102 | } | |
103 | ||
104 | /** | |
105 | * gfs2_aspace_releasepage - free the metadata associated with a page | |
106 | * @page: the page that's being released | |
107 | * @gfp_mask: passed from Linux VFS, ignored by us | |
108 | * | |
109 | * Call try_to_free_buffers() if the buffers in this page can be | |
110 | * released. | |
111 | * | |
112 | * Returns: 0 | |
113 | */ | |
114 | ||
115 | static int gfs2_aspace_releasepage(struct page *page, gfp_t gfp_mask) | |
116 | { | |
117 | struct inode *aspace = page->mapping->host; | |
5c676f6d | 118 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; |
b3b94faa DT |
119 | struct buffer_head *bh, *head; |
120 | struct gfs2_bufdata *bd; | |
121 | unsigned long t; | |
122 | ||
123 | if (!page_has_buffers(page)) | |
124 | goto out; | |
125 | ||
126 | head = bh = page_buffers(page); | |
127 | do { | |
128 | t = jiffies; | |
129 | ||
130 | while (atomic_read(&bh->b_count)) { | |
131 | if (atomic_read(&aspace->i_writecount)) { | |
132 | if (time_after_eq(jiffies, t + | |
133 | gfs2_tune_get(sdp, gt_stall_secs) * HZ)) { | |
134 | stuck_releasepage(bh); | |
135 | t = jiffies; | |
136 | } | |
137 | ||
138 | yield(); | |
139 | continue; | |
140 | } | |
141 | ||
142 | return 0; | |
143 | } | |
144 | ||
145 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); | |
146 | ||
5c676f6d | 147 | bd = bh->b_private; |
b3b94faa DT |
148 | if (bd) { |
149 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
150 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
151 | gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list)); | |
152 | gfs2_assert_warn(sdp, !bd->bd_ail); | |
153 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
5c676f6d | 154 | bh->b_private = NULL; |
b3b94faa DT |
155 | } |
156 | ||
157 | bh = bh->b_this_page; | |
158 | } | |
159 | while (bh != head); | |
160 | ||
161 | out: | |
162 | return try_to_free_buffers(page); | |
163 | } | |
164 | ||
165 | static struct address_space_operations aspace_aops = { | |
166 | .writepage = gfs2_aspace_writepage, | |
167 | .releasepage = gfs2_aspace_releasepage, | |
168 | }; | |
169 | ||
170 | /** | |
171 | * gfs2_aspace_get - Create and initialize a struct inode structure | |
172 | * @sdp: the filesystem the aspace is in | |
173 | * | |
174 | * Right now a struct inode is just a struct inode. Maybe Linux | |
175 | * will supply a more lightweight address space construct (that works) | |
176 | * in the future. | |
177 | * | |
178 | * Make sure pages/buffers in this aspace aren't in high memory. | |
179 | * | |
180 | * Returns: the aspace | |
181 | */ | |
182 | ||
183 | struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp) | |
184 | { | |
185 | struct inode *aspace; | |
186 | ||
187 | aspace = new_inode(sdp->sd_vfs); | |
188 | if (aspace) { | |
189 | mapping_set_gfp_mask(aspace->i_mapping, GFP_KERNEL); | |
190 | aspace->i_mapping->a_ops = &aspace_aops; | |
191 | aspace->i_size = ~0ULL; | |
5c676f6d | 192 | aspace->u.generic_ip = NULL; |
b3b94faa DT |
193 | insert_inode_hash(aspace); |
194 | } | |
b3b94faa DT |
195 | return aspace; |
196 | } | |
197 | ||
198 | void gfs2_aspace_put(struct inode *aspace) | |
199 | { | |
200 | remove_inode_hash(aspace); | |
201 | iput(aspace); | |
202 | } | |
203 | ||
204 | /** | |
205 | * gfs2_ail1_start_one - Start I/O on a part of the AIL | |
206 | * @sdp: the filesystem | |
207 | * @tr: the part of the AIL | |
208 | * | |
209 | */ | |
210 | ||
211 | void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
212 | { | |
213 | struct gfs2_bufdata *bd, *s; | |
214 | struct buffer_head *bh; | |
215 | int retry; | |
216 | ||
190562bd SW |
217 | BUG_ON(!spin_is_locked(&sdp->sd_log_lock)); |
218 | ||
b3b94faa DT |
219 | do { |
220 | retry = 0; | |
221 | ||
222 | list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, | |
223 | bd_ail_st_list) { | |
224 | bh = bd->bd_bh; | |
225 | ||
226 | gfs2_assert(sdp, bd->bd_ail == ai); | |
227 | ||
228 | if (!buffer_busy(bh)) { | |
1b50259b SW |
229 | if (!buffer_uptodate(bh)) { |
230 | gfs2_log_unlock(sdp); | |
b3b94faa | 231 | gfs2_io_error_bh(sdp, bh); |
1b50259b SW |
232 | gfs2_log_lock(sdp); |
233 | } | |
b3b94faa DT |
234 | list_move(&bd->bd_ail_st_list, |
235 | &ai->ai_ail2_list); | |
236 | continue; | |
237 | } | |
238 | ||
239 | if (!buffer_dirty(bh)) | |
240 | continue; | |
241 | ||
242 | list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list); | |
243 | ||
244 | gfs2_log_unlock(sdp); | |
245 | wait_on_buffer(bh); | |
246 | ll_rw_block(WRITE, 1, &bh); | |
247 | gfs2_log_lock(sdp); | |
248 | ||
249 | retry = 1; | |
250 | break; | |
251 | } | |
252 | } while (retry); | |
253 | } | |
254 | ||
255 | /** | |
256 | * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced | |
257 | * @sdp: the filesystem | |
258 | * @ai: the AIL entry | |
259 | * | |
260 | */ | |
261 | ||
262 | int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags) | |
263 | { | |
264 | struct gfs2_bufdata *bd, *s; | |
265 | struct buffer_head *bh; | |
266 | ||
267 | list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, | |
268 | bd_ail_st_list) { | |
269 | bh = bd->bd_bh; | |
270 | ||
271 | gfs2_assert(sdp, bd->bd_ail == ai); | |
272 | ||
273 | if (buffer_busy(bh)) { | |
274 | if (flags & DIO_ALL) | |
275 | continue; | |
276 | else | |
277 | break; | |
278 | } | |
279 | ||
280 | if (!buffer_uptodate(bh)) | |
281 | gfs2_io_error_bh(sdp, bh); | |
282 | ||
283 | list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); | |
284 | } | |
285 | ||
286 | return list_empty(&ai->ai_ail1_list); | |
287 | } | |
288 | ||
289 | /** | |
290 | * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced | |
291 | * @sdp: the filesystem | |
292 | * @ai: the AIL entry | |
293 | * | |
294 | */ | |
295 | ||
296 | void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
297 | { | |
298 | struct list_head *head = &ai->ai_ail2_list; | |
299 | struct gfs2_bufdata *bd; | |
300 | ||
301 | while (!list_empty(head)) { | |
302 | bd = list_entry(head->prev, struct gfs2_bufdata, | |
303 | bd_ail_st_list); | |
304 | gfs2_assert(sdp, bd->bd_ail == ai); | |
305 | bd->bd_ail = NULL; | |
306 | list_del(&bd->bd_ail_st_list); | |
307 | list_del(&bd->bd_ail_gl_list); | |
308 | atomic_dec(&bd->bd_gl->gl_ail_count); | |
309 | brelse(bd->bd_bh); | |
310 | } | |
311 | } | |
312 | ||
313 | /** | |
314 | * ail_empty_gl - remove all buffers for a given lock from the AIL | |
315 | * @gl: the glock | |
316 | * | |
317 | * None of the buffers should be dirty, locked, or pinned. | |
318 | */ | |
319 | ||
320 | void gfs2_ail_empty_gl(struct gfs2_glock *gl) | |
321 | { | |
322 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
323 | unsigned int blocks; | |
324 | struct list_head *head = &gl->gl_ail_list; | |
325 | struct gfs2_bufdata *bd; | |
326 | struct buffer_head *bh; | |
327 | uint64_t blkno; | |
328 | int error; | |
329 | ||
330 | blocks = atomic_read(&gl->gl_ail_count); | |
331 | if (!blocks) | |
332 | return; | |
333 | ||
334 | error = gfs2_trans_begin(sdp, 0, blocks); | |
335 | if (gfs2_assert_withdraw(sdp, !error)) | |
336 | return; | |
337 | ||
338 | gfs2_log_lock(sdp); | |
339 | while (!list_empty(head)) { | |
340 | bd = list_entry(head->next, struct gfs2_bufdata, | |
341 | bd_ail_gl_list); | |
342 | bh = bd->bd_bh; | |
343 | blkno = bh->b_blocknr; | |
344 | gfs2_assert_withdraw(sdp, !buffer_busy(bh)); | |
345 | ||
346 | bd->bd_ail = NULL; | |
347 | list_del(&bd->bd_ail_st_list); | |
348 | list_del(&bd->bd_ail_gl_list); | |
349 | atomic_dec(&gl->gl_ail_count); | |
350 | brelse(bh); | |
351 | gfs2_log_unlock(sdp); | |
352 | ||
353 | gfs2_trans_add_revoke(sdp, blkno); | |
354 | ||
355 | gfs2_log_lock(sdp); | |
356 | } | |
357 | gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count)); | |
358 | gfs2_log_unlock(sdp); | |
359 | ||
360 | gfs2_trans_end(sdp); | |
b09e593d | 361 | gfs2_log_flush(sdp, NULL); |
b3b94faa DT |
362 | } |
363 | ||
364 | /** | |
365 | * gfs2_meta_inval - Invalidate all buffers associated with a glock | |
366 | * @gl: the glock | |
367 | * | |
368 | */ | |
369 | ||
370 | void gfs2_meta_inval(struct gfs2_glock *gl) | |
371 | { | |
372 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
373 | struct inode *aspace = gl->gl_aspace; | |
374 | struct address_space *mapping = gl->gl_aspace->i_mapping; | |
375 | ||
376 | gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count)); | |
377 | ||
378 | atomic_inc(&aspace->i_writecount); | |
379 | truncate_inode_pages(mapping, 0); | |
380 | atomic_dec(&aspace->i_writecount); | |
381 | ||
382 | gfs2_assert_withdraw(sdp, !mapping->nrpages); | |
383 | } | |
384 | ||
385 | /** | |
386 | * gfs2_meta_sync - Sync all buffers associated with a glock | |
387 | * @gl: The glock | |
388 | * @flags: DIO_START | DIO_WAIT | |
389 | * | |
390 | */ | |
391 | ||
392 | void gfs2_meta_sync(struct gfs2_glock *gl, int flags) | |
393 | { | |
394 | struct address_space *mapping = gl->gl_aspace->i_mapping; | |
395 | int error = 0; | |
396 | ||
397 | if (flags & DIO_START) | |
398 | filemap_fdatawrite(mapping); | |
399 | if (!error && (flags & DIO_WAIT)) | |
400 | error = filemap_fdatawait(mapping); | |
401 | ||
402 | if (error) | |
403 | gfs2_io_error(gl->gl_sbd); | |
404 | } | |
405 | ||
406 | /** | |
407 | * getbuf - Get a buffer with a given address space | |
408 | * @sdp: the filesystem | |
409 | * @aspace: the address space | |
410 | * @blkno: the block number (filesystem scope) | |
411 | * @create: 1 if the buffer should be created | |
412 | * | |
413 | * Returns: the buffer | |
414 | */ | |
415 | ||
416 | static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace, | |
417 | uint64_t blkno, int create) | |
418 | { | |
419 | struct page *page; | |
420 | struct buffer_head *bh; | |
421 | unsigned int shift; | |
422 | unsigned long index; | |
423 | unsigned int bufnum; | |
424 | ||
425 | shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift; | |
426 | index = blkno >> shift; /* convert block to page */ | |
427 | bufnum = blkno - (index << shift); /* block buf index within page */ | |
428 | ||
429 | if (create) { | |
430 | for (;;) { | |
431 | page = grab_cache_page(aspace->i_mapping, index); | |
432 | if (page) | |
433 | break; | |
434 | yield(); | |
435 | } | |
436 | } else { | |
437 | page = find_lock_page(aspace->i_mapping, index); | |
438 | if (!page) | |
439 | return NULL; | |
440 | } | |
441 | ||
442 | if (!page_has_buffers(page)) | |
443 | create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); | |
444 | ||
445 | /* Locate header for our buffer within our page */ | |
446 | for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) | |
447 | /* Do nothing */; | |
448 | get_bh(bh); | |
449 | ||
450 | if (!buffer_mapped(bh)) | |
451 | map_bh(bh, sdp->sd_vfs, blkno); | |
452 | ||
453 | unlock_page(page); | |
454 | mark_page_accessed(page); | |
455 | page_cache_release(page); | |
456 | ||
457 | return bh; | |
458 | } | |
459 | ||
460 | static void meta_prep_new(struct buffer_head *bh) | |
461 | { | |
462 | struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; | |
463 | ||
464 | lock_buffer(bh); | |
465 | clear_buffer_dirty(bh); | |
466 | set_buffer_uptodate(bh); | |
467 | unlock_buffer(bh); | |
468 | ||
469 | mh->mh_magic = cpu_to_be32(GFS2_MAGIC); | |
470 | } | |
471 | ||
472 | /** | |
473 | * gfs2_meta_new - Get a block | |
474 | * @gl: The glock associated with this block | |
475 | * @blkno: The block number | |
476 | * | |
477 | * Returns: The buffer | |
478 | */ | |
479 | ||
480 | struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, uint64_t blkno) | |
481 | { | |
482 | struct buffer_head *bh; | |
483 | bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE); | |
484 | meta_prep_new(bh); | |
485 | return bh; | |
486 | } | |
487 | ||
488 | /** | |
489 | * gfs2_meta_read - Read a block from disk | |
490 | * @gl: The glock covering the block | |
491 | * @blkno: The block number | |
492 | * @flags: flags to gfs2_dreread() | |
493 | * @bhp: the place where the buffer is returned (NULL on failure) | |
494 | * | |
495 | * Returns: errno | |
496 | */ | |
497 | ||
498 | int gfs2_meta_read(struct gfs2_glock *gl, uint64_t blkno, int flags, | |
499 | struct buffer_head **bhp) | |
500 | { | |
501 | int error; | |
502 | ||
503 | *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE); | |
504 | error = gfs2_meta_reread(gl->gl_sbd, *bhp, flags); | |
505 | if (error) | |
506 | brelse(*bhp); | |
507 | ||
508 | return error; | |
509 | } | |
510 | ||
511 | /** | |
512 | * gfs2_meta_reread - Reread a block from disk | |
513 | * @sdp: the filesystem | |
514 | * @bh: The block to read | |
515 | * @flags: Flags that control the read | |
516 | * | |
517 | * Returns: errno | |
518 | */ | |
519 | ||
520 | int gfs2_meta_reread(struct gfs2_sbd *sdp, struct buffer_head *bh, int flags) | |
521 | { | |
522 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
523 | return -EIO; | |
524 | ||
525 | if (flags & DIO_FORCE) | |
526 | clear_buffer_uptodate(bh); | |
527 | ||
528 | if ((flags & DIO_START) && !buffer_uptodate(bh)) | |
529 | ll_rw_block(READ, 1, &bh); | |
530 | ||
531 | if (flags & DIO_WAIT) { | |
532 | wait_on_buffer(bh); | |
533 | ||
534 | if (!buffer_uptodate(bh)) { | |
5c676f6d | 535 | struct gfs2_trans *tr = current->journal_info; |
b3b94faa DT |
536 | if (tr && tr->tr_touched) |
537 | gfs2_io_error_bh(sdp, bh); | |
538 | return -EIO; | |
539 | } | |
540 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
541 | return -EIO; | |
542 | } | |
543 | ||
544 | return 0; | |
545 | } | |
546 | ||
547 | /** | |
586dfdaa | 548 | * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer |
b3b94faa DT |
549 | * @gl: the glock the buffer belongs to |
550 | * @bh: The buffer to be attached to | |
586dfdaa | 551 | * @meta: Flag to indicate whether its metadata or not |
b3b94faa DT |
552 | */ |
553 | ||
568f4c96 SW |
554 | void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh, |
555 | int meta) | |
b3b94faa DT |
556 | { |
557 | struct gfs2_bufdata *bd; | |
558 | ||
18ec7d5c SW |
559 | if (meta) |
560 | lock_page(bh->b_page); | |
b3b94faa | 561 | |
5c676f6d | 562 | if (bh->b_private) { |
18ec7d5c SW |
563 | if (meta) |
564 | unlock_page(bh->b_page); | |
b3b94faa DT |
565 | return; |
566 | } | |
567 | ||
f55ab26a | 568 | bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL), |
b3b94faa DT |
569 | memset(bd, 0, sizeof(struct gfs2_bufdata)); |
570 | ||
571 | bd->bd_bh = bh; | |
572 | bd->bd_gl = gl; | |
573 | ||
574 | INIT_LIST_HEAD(&bd->bd_list_tr); | |
18ec7d5c | 575 | if (meta) { |
586dfdaa | 576 | lops_init_le(&bd->bd_le, &gfs2_buf_lops); |
18ec7d5c | 577 | } else { |
586dfdaa | 578 | lops_init_le(&bd->bd_le, &gfs2_databuf_lops); |
18ec7d5c SW |
579 | get_bh(bh); |
580 | } | |
5c676f6d | 581 | bh->b_private = bd; |
b3b94faa | 582 | |
18ec7d5c SW |
583 | if (meta) |
584 | unlock_page(bh->b_page); | |
b3b94faa DT |
585 | } |
586 | ||
587 | /** | |
a98ab220 | 588 | * gfs2_pin - Pin a buffer in memory |
b3b94faa DT |
589 | * @sdp: the filesystem the buffer belongs to |
590 | * @bh: The buffer to be pinned | |
591 | * | |
592 | */ | |
593 | ||
a98ab220 | 594 | void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh) |
b3b94faa | 595 | { |
5c676f6d | 596 | struct gfs2_bufdata *bd = bh->b_private; |
b3b94faa DT |
597 | |
598 | gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)); | |
599 | ||
600 | if (test_set_buffer_pinned(bh)) | |
601 | gfs2_assert_withdraw(sdp, 0); | |
602 | ||
603 | wait_on_buffer(bh); | |
604 | ||
605 | /* If this buffer is in the AIL and it has already been written | |
606 | to in-place disk block, remove it from the AIL. */ | |
607 | ||
608 | gfs2_log_lock(sdp); | |
609 | if (bd->bd_ail && !buffer_in_io(bh)) | |
610 | list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list); | |
611 | gfs2_log_unlock(sdp); | |
612 | ||
613 | clear_buffer_dirty(bh); | |
614 | wait_on_buffer(bh); | |
615 | ||
616 | if (!buffer_uptodate(bh)) | |
617 | gfs2_io_error_bh(sdp, bh); | |
618 | ||
619 | get_bh(bh); | |
620 | } | |
621 | ||
622 | /** | |
a98ab220 | 623 | * gfs2_unpin - Unpin a buffer |
b3b94faa DT |
624 | * @sdp: the filesystem the buffer belongs to |
625 | * @bh: The buffer to unpin | |
626 | * @ai: | |
627 | * | |
628 | */ | |
629 | ||
a98ab220 SW |
630 | void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh, |
631 | struct gfs2_ail *ai) | |
b3b94faa | 632 | { |
5c676f6d | 633 | struct gfs2_bufdata *bd = bh->b_private; |
b3b94faa DT |
634 | |
635 | gfs2_assert_withdraw(sdp, buffer_uptodate(bh)); | |
636 | ||
637 | if (!buffer_pinned(bh)) | |
638 | gfs2_assert_withdraw(sdp, 0); | |
639 | ||
640 | mark_buffer_dirty(bh); | |
641 | clear_buffer_pinned(bh); | |
642 | ||
643 | gfs2_log_lock(sdp); | |
644 | if (bd->bd_ail) { | |
645 | list_del(&bd->bd_ail_st_list); | |
646 | brelse(bh); | |
647 | } else { | |
648 | struct gfs2_glock *gl = bd->bd_gl; | |
649 | list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list); | |
650 | atomic_inc(&gl->gl_ail_count); | |
651 | } | |
652 | bd->bd_ail = ai; | |
653 | list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list); | |
654 | gfs2_log_unlock(sdp); | |
655 | } | |
656 | ||
657 | /** | |
658 | * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore | |
659 | * @ip: the inode who owns the buffers | |
660 | * @bstart: the first buffer in the run | |
661 | * @blen: the number of buffers in the run | |
662 | * | |
663 | */ | |
664 | ||
665 | void gfs2_meta_wipe(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen) | |
666 | { | |
667 | struct gfs2_sbd *sdp = ip->i_sbd; | |
668 | struct inode *aspace = ip->i_gl->gl_aspace; | |
669 | struct buffer_head *bh; | |
670 | ||
671 | while (blen) { | |
672 | bh = getbuf(sdp, aspace, bstart, NO_CREATE); | |
673 | if (bh) { | |
5c676f6d | 674 | struct gfs2_bufdata *bd = bh->b_private; |
b3b94faa DT |
675 | |
676 | if (test_clear_buffer_pinned(bh)) { | |
5c676f6d | 677 | struct gfs2_trans *tr = current->journal_info; |
b3b94faa DT |
678 | gfs2_log_lock(sdp); |
679 | list_del_init(&bd->bd_le.le_list); | |
680 | gfs2_assert_warn(sdp, sdp->sd_log_num_buf); | |
681 | sdp->sd_log_num_buf--; | |
682 | gfs2_log_unlock(sdp); | |
5c676f6d | 683 | tr->tr_num_buf_rm++; |
b3b94faa DT |
684 | brelse(bh); |
685 | } | |
686 | if (bd) { | |
687 | gfs2_log_lock(sdp); | |
688 | if (bd->bd_ail) { | |
689 | uint64_t blkno = bh->b_blocknr; | |
690 | bd->bd_ail = NULL; | |
691 | list_del(&bd->bd_ail_st_list); | |
692 | list_del(&bd->bd_ail_gl_list); | |
693 | atomic_dec(&bd->bd_gl->gl_ail_count); | |
694 | brelse(bh); | |
695 | gfs2_log_unlock(sdp); | |
696 | gfs2_trans_add_revoke(sdp, blkno); | |
697 | } else | |
698 | gfs2_log_unlock(sdp); | |
699 | } | |
700 | ||
701 | lock_buffer(bh); | |
702 | clear_buffer_dirty(bh); | |
703 | clear_buffer_uptodate(bh); | |
704 | unlock_buffer(bh); | |
705 | ||
706 | brelse(bh); | |
707 | } | |
708 | ||
709 | bstart++; | |
710 | blen--; | |
711 | } | |
712 | } | |
713 | ||
714 | /** | |
715 | * gfs2_meta_cache_flush - get rid of any references on buffers for this inode | |
716 | * @ip: The GFS2 inode | |
717 | * | |
718 | * This releases buffers that are in the most-recently-used array of | |
719 | * blocks used for indirect block addressing for this inode. | |
720 | */ | |
721 | ||
722 | void gfs2_meta_cache_flush(struct gfs2_inode *ip) | |
723 | { | |
724 | struct buffer_head **bh_slot; | |
725 | unsigned int x; | |
726 | ||
727 | spin_lock(&ip->i_spin); | |
728 | ||
729 | for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) { | |
730 | bh_slot = &ip->i_cache[x]; | |
731 | if (!*bh_slot) | |
732 | break; | |
733 | brelse(*bh_slot); | |
734 | *bh_slot = NULL; | |
735 | } | |
736 | ||
737 | spin_unlock(&ip->i_spin); | |
738 | } | |
739 | ||
740 | /** | |
741 | * gfs2_meta_indirect_buffer - Get a metadata buffer | |
742 | * @ip: The GFS2 inode | |
743 | * @height: The level of this buf in the metadata (indir addr) tree (if any) | |
744 | * @num: The block number (device relative) of the buffer | |
745 | * @new: Non-zero if we may create a new buffer | |
746 | * @bhp: the buffer is returned here | |
747 | * | |
748 | * Try to use the gfs2_inode's MRU metadata tree cache. | |
749 | * | |
750 | * Returns: errno | |
751 | */ | |
752 | ||
753 | int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, uint64_t num, | |
754 | int new, struct buffer_head **bhp) | |
755 | { | |
756 | struct buffer_head *bh, **bh_slot = ip->i_cache + height; | |
757 | int error; | |
758 | ||
759 | spin_lock(&ip->i_spin); | |
760 | bh = *bh_slot; | |
761 | if (bh) { | |
762 | if (bh->b_blocknr == num) | |
763 | get_bh(bh); | |
764 | else | |
765 | bh = NULL; | |
766 | } | |
767 | spin_unlock(&ip->i_spin); | |
768 | ||
769 | if (bh) { | |
770 | if (new) | |
771 | meta_prep_new(bh); | |
772 | else { | |
773 | error = gfs2_meta_reread(ip->i_sbd, bh, | |
774 | DIO_START | DIO_WAIT); | |
775 | if (error) { | |
776 | brelse(bh); | |
777 | return error; | |
778 | } | |
779 | } | |
780 | } else { | |
781 | if (new) | |
782 | bh = gfs2_meta_new(ip->i_gl, num); | |
783 | else { | |
784 | error = gfs2_meta_read(ip->i_gl, num, | |
785 | DIO_START | DIO_WAIT, &bh); | |
786 | if (error) | |
787 | return error; | |
788 | } | |
789 | ||
790 | spin_lock(&ip->i_spin); | |
791 | if (*bh_slot != bh) { | |
792 | brelse(*bh_slot); | |
793 | *bh_slot = bh; | |
794 | get_bh(bh); | |
795 | } | |
796 | spin_unlock(&ip->i_spin); | |
797 | } | |
798 | ||
799 | if (new) { | |
800 | if (gfs2_assert_warn(ip->i_sbd, height)) { | |
801 | brelse(bh); | |
802 | return -EIO; | |
803 | } | |
d4e9c4c3 | 804 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
b3b94faa DT |
805 | gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN); |
806 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
807 | ||
808 | } else if (gfs2_metatype_check(ip->i_sbd, bh, | |
809 | (height) ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)) { | |
810 | brelse(bh); | |
811 | return -EIO; | |
812 | } | |
813 | ||
814 | *bhp = bh; | |
815 | ||
816 | return 0; | |
817 | } | |
818 | ||
819 | /** | |
820 | * gfs2_meta_ra - start readahead on an extent of a file | |
821 | * @gl: the glock the blocks belong to | |
822 | * @dblock: the starting disk block | |
823 | * @extlen: the number of blocks in the extent | |
824 | * | |
825 | */ | |
826 | ||
827 | void gfs2_meta_ra(struct gfs2_glock *gl, uint64_t dblock, uint32_t extlen) | |
828 | { | |
829 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
830 | struct inode *aspace = gl->gl_aspace; | |
831 | struct buffer_head *first_bh, *bh; | |
568f4c96 SW |
832 | uint32_t max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> |
833 | sdp->sd_sb.sb_bsize_shift; | |
b3b94faa DT |
834 | int error; |
835 | ||
836 | if (!extlen || !max_ra) | |
837 | return; | |
838 | if (extlen > max_ra) | |
839 | extlen = max_ra; | |
840 | ||
841 | first_bh = getbuf(sdp, aspace, dblock, CREATE); | |
842 | ||
843 | if (buffer_uptodate(first_bh)) | |
844 | goto out; | |
845 | if (!buffer_locked(first_bh)) { | |
846 | error = gfs2_meta_reread(sdp, first_bh, DIO_START); | |
847 | if (error) | |
848 | goto out; | |
849 | } | |
850 | ||
851 | dblock++; | |
852 | extlen--; | |
853 | ||
854 | while (extlen) { | |
855 | bh = getbuf(sdp, aspace, dblock, CREATE); | |
856 | ||
857 | if (!buffer_uptodate(bh) && !buffer_locked(bh)) { | |
858 | error = gfs2_meta_reread(sdp, bh, DIO_START); | |
859 | brelse(bh); | |
860 | if (error) | |
861 | goto out; | |
862 | } else | |
863 | brelse(bh); | |
864 | ||
865 | dblock++; | |
866 | extlen--; | |
867 | ||
868 | if (buffer_uptodate(first_bh)) | |
869 | break; | |
870 | } | |
871 | ||
872 | out: | |
873 | brelse(first_bh); | |
874 | } | |
875 | ||
876 | /** | |
877 | * gfs2_meta_syncfs - sync all the buffers in a filesystem | |
878 | * @sdp: the filesystem | |
879 | * | |
880 | */ | |
881 | ||
882 | void gfs2_meta_syncfs(struct gfs2_sbd *sdp) | |
883 | { | |
b09e593d | 884 | gfs2_log_flush(sdp, NULL); |
b3b94faa DT |
885 | for (;;) { |
886 | gfs2_ail1_start(sdp, DIO_ALL); | |
887 | if (gfs2_ail1_empty(sdp, DIO_ALL)) | |
888 | break; | |
fe1bdedc | 889 | msleep(10); |
b3b94faa DT |
890 | } |
891 | } | |
892 |