GFS2: Use rbtree for resource groups and clean up bitmap buffer ref count scheme
[deliverable/linux.git] / fs / gfs2 / glops.c
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
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 version 2.
8 */
9
10 #include <linux/spinlock.h>
11 #include <linux/completion.h>
12 #include <linux/buffer_head.h>
13 #include <linux/gfs2_ondisk.h>
14 #include <linux/bio.h>
15 #include <linux/posix_acl.h>
16
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "bmap.h"
20 #include "glock.h"
21 #include "glops.h"
22 #include "inode.h"
23 #include "log.h"
24 #include "meta_io.h"
25 #include "recovery.h"
26 #include "rgrp.h"
27 #include "util.h"
28 #include "trans.h"
29 #include "dir.h"
30
31 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
32 {
33 fs_err(gl->gl_sbd, "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page state 0x%lx\n",
34 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
35 bh->b_page->mapping, bh->b_page->flags);
36 fs_err(gl->gl_sbd, "AIL glock %u:%llu mapping %p\n",
37 gl->gl_name.ln_type, gl->gl_name.ln_number,
38 gfs2_glock2aspace(gl));
39 gfs2_lm_withdraw(gl->gl_sbd, "AIL error\n");
40 }
41
42 /**
43 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
44 * @gl: the glock
45 *
46 * None of the buffers should be dirty, locked, or pinned.
47 */
48
49 static void __gfs2_ail_flush(struct gfs2_glock *gl, unsigned long b_state)
50 {
51 struct gfs2_sbd *sdp = gl->gl_sbd;
52 struct list_head *head = &gl->gl_ail_list;
53 struct gfs2_bufdata *bd;
54 struct buffer_head *bh;
55 sector_t blocknr;
56
57 spin_lock(&sdp->sd_ail_lock);
58 while (!list_empty(head)) {
59 bd = list_entry(head->next, struct gfs2_bufdata,
60 bd_ail_gl_list);
61 bh = bd->bd_bh;
62 blocknr = bh->b_blocknr;
63 if (bh->b_state & b_state)
64 gfs2_ail_error(gl, bh);
65 bh->b_private = NULL;
66 gfs2_remove_from_ail(bd); /* drops ref on bh */
67 spin_unlock(&sdp->sd_ail_lock);
68
69 bd->bd_bh = NULL;
70 bd->bd_blkno = blocknr;
71
72 gfs2_log_lock(sdp);
73 gfs2_trans_add_revoke(sdp, bd);
74 gfs2_log_unlock(sdp);
75
76 spin_lock(&sdp->sd_ail_lock);
77 }
78 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
79 spin_unlock(&sdp->sd_ail_lock);
80 }
81
82
83 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
84 {
85 struct gfs2_sbd *sdp = gl->gl_sbd;
86 struct gfs2_trans tr;
87
88 memset(&tr, 0, sizeof(tr));
89 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
90
91 if (!tr.tr_revokes)
92 return;
93
94 /* A shortened, inline version of gfs2_trans_begin() */
95 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
96 tr.tr_ip = (unsigned long)__builtin_return_address(0);
97 INIT_LIST_HEAD(&tr.tr_list_buf);
98 gfs2_log_reserve(sdp, tr.tr_reserved);
99 BUG_ON(current->journal_info);
100 current->journal_info = &tr;
101
102 __gfs2_ail_flush(gl, (1ul << BH_Dirty)|(1ul << BH_Pinned)|(1ul << BH_Lock));
103
104 gfs2_trans_end(sdp);
105 gfs2_log_flush(sdp, NULL);
106 }
107
108 void gfs2_ail_flush(struct gfs2_glock *gl)
109 {
110 struct gfs2_sbd *sdp = gl->gl_sbd;
111 unsigned int revokes = atomic_read(&gl->gl_ail_count);
112 int ret;
113
114 if (!revokes)
115 return;
116
117 ret = gfs2_trans_begin(sdp, 0, revokes);
118 if (ret)
119 return;
120 __gfs2_ail_flush(gl, (1ul << BH_Dirty)|(1ul << BH_Pinned));
121 gfs2_trans_end(sdp);
122 gfs2_log_flush(sdp, NULL);
123 }
124
125 /**
126 * rgrp_go_sync - sync out the metadata for this glock
127 * @gl: the glock
128 *
129 * Called when demoting or unlocking an EX glock. We must flush
130 * to disk all dirty buffers/pages relating to this glock, and must not
131 * not return to caller to demote/unlock the glock until I/O is complete.
132 */
133
134 static void rgrp_go_sync(struct gfs2_glock *gl)
135 {
136 struct address_space *metamapping = gfs2_glock2aspace(gl);
137 struct gfs2_rgrpd *rgd = gl->gl_object;
138 unsigned int x;
139 int error;
140
141 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
142 return;
143 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
144
145 gfs2_log_flush(gl->gl_sbd, gl);
146 filemap_fdatawrite(metamapping);
147 error = filemap_fdatawait(metamapping);
148 mapping_set_error(metamapping, error);
149 gfs2_ail_empty_gl(gl);
150
151 if (!rgd)
152 return;
153
154 for (x = 0; x < rgd->rd_length; x++) {
155 struct gfs2_bitmap *bi = rgd->rd_bits + x;
156 kfree(bi->bi_clone);
157 bi->bi_clone = NULL;
158 }
159 }
160
161 /**
162 * rgrp_go_inval - invalidate the metadata for this glock
163 * @gl: the glock
164 * @flags:
165 *
166 * We never used LM_ST_DEFERRED with resource groups, so that we
167 * should always see the metadata flag set here.
168 *
169 */
170
171 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
172 {
173 struct address_space *mapping = gfs2_glock2aspace(gl);
174
175 BUG_ON(!(flags & DIO_METADATA));
176 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
177 truncate_inode_pages(mapping, 0);
178
179 if (gl->gl_object) {
180 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
181 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
182 }
183 }
184
185 /**
186 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
187 * @gl: the glock protecting the inode
188 *
189 */
190
191 static void inode_go_sync(struct gfs2_glock *gl)
192 {
193 struct gfs2_inode *ip = gl->gl_object;
194 struct address_space *metamapping = gfs2_glock2aspace(gl);
195 int error;
196
197 if (ip && !S_ISREG(ip->i_inode.i_mode))
198 ip = NULL;
199 if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
200 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
201 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
202 return;
203
204 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
205
206 gfs2_log_flush(gl->gl_sbd, gl);
207 filemap_fdatawrite(metamapping);
208 if (ip) {
209 struct address_space *mapping = ip->i_inode.i_mapping;
210 filemap_fdatawrite(mapping);
211 error = filemap_fdatawait(mapping);
212 mapping_set_error(mapping, error);
213 }
214 error = filemap_fdatawait(metamapping);
215 mapping_set_error(metamapping, error);
216 gfs2_ail_empty_gl(gl);
217 /*
218 * Writeback of the data mapping may cause the dirty flag to be set
219 * so we have to clear it again here.
220 */
221 smp_mb__before_clear_bit();
222 clear_bit(GLF_DIRTY, &gl->gl_flags);
223 }
224
225 /**
226 * inode_go_inval - prepare a inode glock to be released
227 * @gl: the glock
228 * @flags:
229 *
230 * Normally we invlidate everything, but if we are moving into
231 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
232 * can keep hold of the metadata, since it won't have changed.
233 *
234 */
235
236 static void inode_go_inval(struct gfs2_glock *gl, int flags)
237 {
238 struct gfs2_inode *ip = gl->gl_object;
239
240 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
241
242 if (flags & DIO_METADATA) {
243 struct address_space *mapping = gfs2_glock2aspace(gl);
244 truncate_inode_pages(mapping, 0);
245 if (ip) {
246 set_bit(GIF_INVALID, &ip->i_flags);
247 forget_all_cached_acls(&ip->i_inode);
248 gfs2_dir_hash_inval(ip);
249 }
250 }
251
252 if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) {
253 gfs2_log_flush(gl->gl_sbd, NULL);
254 gl->gl_sbd->sd_rindex_uptodate = 0;
255 }
256 if (ip && S_ISREG(ip->i_inode.i_mode))
257 truncate_inode_pages(ip->i_inode.i_mapping, 0);
258 }
259
260 /**
261 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
262 * @gl: the glock
263 *
264 * Returns: 1 if it's ok
265 */
266
267 static int inode_go_demote_ok(const struct gfs2_glock *gl)
268 {
269 struct gfs2_sbd *sdp = gl->gl_sbd;
270 struct gfs2_holder *gh;
271
272 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
273 return 0;
274
275 if (!list_empty(&gl->gl_holders)) {
276 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
277 if (gh->gh_list.next != &gl->gl_holders)
278 return 0;
279 }
280
281 return 1;
282 }
283
284 /**
285 * gfs2_set_nlink - Set the inode's link count based on on-disk info
286 * @inode: The inode in question
287 * @nlink: The link count
288 *
289 * If the link count has hit zero, it must never be raised, whatever the
290 * on-disk inode might say. When new struct inodes are created the link
291 * count is set to 1, so that we can safely use this test even when reading
292 * in on disk information for the first time.
293 */
294
295 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
296 {
297 /*
298 * We will need to review setting the nlink count here in the
299 * light of the forthcoming ro bind mount work. This is a reminder
300 * to do that.
301 */
302 if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
303 if (nlink == 0)
304 clear_nlink(inode);
305 else
306 inode->i_nlink = nlink;
307 }
308 }
309
310 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
311 {
312 const struct gfs2_dinode *str = buf;
313 struct timespec atime;
314 u16 height, depth;
315
316 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
317 goto corrupt;
318 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
319 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
320 ip->i_inode.i_rdev = 0;
321 switch (ip->i_inode.i_mode & S_IFMT) {
322 case S_IFBLK:
323 case S_IFCHR:
324 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
325 be32_to_cpu(str->di_minor));
326 break;
327 };
328
329 ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
330 ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
331 gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
332 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
333 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
334 atime.tv_sec = be64_to_cpu(str->di_atime);
335 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
336 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
337 ip->i_inode.i_atime = atime;
338 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
339 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
340 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
341 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
342
343 ip->i_goal = be64_to_cpu(str->di_goal_meta);
344 ip->i_generation = be64_to_cpu(str->di_generation);
345
346 ip->i_diskflags = be32_to_cpu(str->di_flags);
347 ip->i_eattr = be64_to_cpu(str->di_eattr);
348 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
349 gfs2_set_inode_flags(&ip->i_inode);
350 height = be16_to_cpu(str->di_height);
351 if (unlikely(height > GFS2_MAX_META_HEIGHT))
352 goto corrupt;
353 ip->i_height = (u8)height;
354
355 depth = be16_to_cpu(str->di_depth);
356 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
357 goto corrupt;
358 ip->i_depth = (u8)depth;
359 ip->i_entries = be32_to_cpu(str->di_entries);
360
361 if (S_ISREG(ip->i_inode.i_mode))
362 gfs2_set_aops(&ip->i_inode);
363
364 return 0;
365 corrupt:
366 gfs2_consist_inode(ip);
367 return -EIO;
368 }
369
370 /**
371 * gfs2_inode_refresh - Refresh the incore copy of the dinode
372 * @ip: The GFS2 inode
373 *
374 * Returns: errno
375 */
376
377 int gfs2_inode_refresh(struct gfs2_inode *ip)
378 {
379 struct buffer_head *dibh;
380 int error;
381
382 error = gfs2_meta_inode_buffer(ip, &dibh);
383 if (error)
384 return error;
385
386 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) {
387 brelse(dibh);
388 return -EIO;
389 }
390
391 error = gfs2_dinode_in(ip, dibh->b_data);
392 brelse(dibh);
393 clear_bit(GIF_INVALID, &ip->i_flags);
394
395 return error;
396 }
397
398 /**
399 * inode_go_lock - operation done after an inode lock is locked by a process
400 * @gl: the glock
401 * @flags:
402 *
403 * Returns: errno
404 */
405
406 static int inode_go_lock(struct gfs2_holder *gh)
407 {
408 struct gfs2_glock *gl = gh->gh_gl;
409 struct gfs2_sbd *sdp = gl->gl_sbd;
410 struct gfs2_inode *ip = gl->gl_object;
411 int error = 0;
412
413 if (!ip || (gh->gh_flags & GL_SKIP))
414 return 0;
415
416 if (test_bit(GIF_INVALID, &ip->i_flags)) {
417 error = gfs2_inode_refresh(ip);
418 if (error)
419 return error;
420 }
421
422 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
423 (gl->gl_state == LM_ST_EXCLUSIVE) &&
424 (gh->gh_state == LM_ST_EXCLUSIVE)) {
425 spin_lock(&sdp->sd_trunc_lock);
426 if (list_empty(&ip->i_trunc_list))
427 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
428 spin_unlock(&sdp->sd_trunc_lock);
429 wake_up(&sdp->sd_quota_wait);
430 return 1;
431 }
432
433 return error;
434 }
435
436 /**
437 * inode_go_dump - print information about an inode
438 * @seq: The iterator
439 * @ip: the inode
440 *
441 * Returns: 0 on success, -ENOBUFS when we run out of space
442 */
443
444 static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
445 {
446 const struct gfs2_inode *ip = gl->gl_object;
447 if (ip == NULL)
448 return 0;
449 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
450 (unsigned long long)ip->i_no_formal_ino,
451 (unsigned long long)ip->i_no_addr,
452 IF2DT(ip->i_inode.i_mode), ip->i_flags,
453 (unsigned int)ip->i_diskflags,
454 (unsigned long long)i_size_read(&ip->i_inode));
455 return 0;
456 }
457
458 /**
459 * trans_go_sync - promote/demote the transaction glock
460 * @gl: the glock
461 * @state: the requested state
462 * @flags:
463 *
464 */
465
466 static void trans_go_sync(struct gfs2_glock *gl)
467 {
468 struct gfs2_sbd *sdp = gl->gl_sbd;
469
470 if (gl->gl_state != LM_ST_UNLOCKED &&
471 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
472 gfs2_meta_syncfs(sdp);
473 gfs2_log_shutdown(sdp);
474 }
475 }
476
477 /**
478 * trans_go_xmote_bh - After promoting/demoting the transaction glock
479 * @gl: the glock
480 *
481 */
482
483 static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
484 {
485 struct gfs2_sbd *sdp = gl->gl_sbd;
486 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
487 struct gfs2_glock *j_gl = ip->i_gl;
488 struct gfs2_log_header_host head;
489 int error;
490
491 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
492 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
493
494 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
495 if (error)
496 gfs2_consist(sdp);
497 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
498 gfs2_consist(sdp);
499
500 /* Initialize some head of the log stuff */
501 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
502 sdp->sd_log_sequence = head.lh_sequence + 1;
503 gfs2_log_pointers_init(sdp, head.lh_blkno);
504 }
505 }
506 return 0;
507 }
508
509 /**
510 * trans_go_demote_ok
511 * @gl: the glock
512 *
513 * Always returns 0
514 */
515
516 static int trans_go_demote_ok(const struct gfs2_glock *gl)
517 {
518 return 0;
519 }
520
521 /**
522 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
523 * @gl: the glock
524 *
525 * gl_spin lock is held while calling this
526 */
527 static void iopen_go_callback(struct gfs2_glock *gl)
528 {
529 struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
530 struct gfs2_sbd *sdp = gl->gl_sbd;
531
532 if (sdp->sd_vfs->s_flags & MS_RDONLY)
533 return;
534
535 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
536 gl->gl_state == LM_ST_SHARED && ip) {
537 gfs2_glock_hold(gl);
538 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
539 gfs2_glock_put_nolock(gl);
540 }
541 }
542
543 const struct gfs2_glock_operations gfs2_meta_glops = {
544 .go_type = LM_TYPE_META,
545 };
546
547 const struct gfs2_glock_operations gfs2_inode_glops = {
548 .go_xmote_th = inode_go_sync,
549 .go_inval = inode_go_inval,
550 .go_demote_ok = inode_go_demote_ok,
551 .go_lock = inode_go_lock,
552 .go_dump = inode_go_dump,
553 .go_type = LM_TYPE_INODE,
554 .go_flags = GLOF_ASPACE,
555 };
556
557 const struct gfs2_glock_operations gfs2_rgrp_glops = {
558 .go_xmote_th = rgrp_go_sync,
559 .go_inval = rgrp_go_inval,
560 .go_lock = gfs2_rgrp_go_lock,
561 .go_unlock = gfs2_rgrp_go_unlock,
562 .go_dump = gfs2_rgrp_dump,
563 .go_type = LM_TYPE_RGRP,
564 .go_flags = GLOF_ASPACE,
565 };
566
567 const struct gfs2_glock_operations gfs2_trans_glops = {
568 .go_xmote_th = trans_go_sync,
569 .go_xmote_bh = trans_go_xmote_bh,
570 .go_demote_ok = trans_go_demote_ok,
571 .go_type = LM_TYPE_NONDISK,
572 };
573
574 const struct gfs2_glock_operations gfs2_iopen_glops = {
575 .go_type = LM_TYPE_IOPEN,
576 .go_callback = iopen_go_callback,
577 };
578
579 const struct gfs2_glock_operations gfs2_flock_glops = {
580 .go_type = LM_TYPE_FLOCK,
581 };
582
583 const struct gfs2_glock_operations gfs2_nondisk_glops = {
584 .go_type = LM_TYPE_NONDISK,
585 };
586
587 const struct gfs2_glock_operations gfs2_quota_glops = {
588 .go_type = LM_TYPE_QUOTA,
589 };
590
591 const struct gfs2_glock_operations gfs2_journal_glops = {
592 .go_type = LM_TYPE_JOURNAL,
593 };
594
595 const struct gfs2_glock_operations *gfs2_glops_list[] = {
596 [LM_TYPE_META] = &gfs2_meta_glops,
597 [LM_TYPE_INODE] = &gfs2_inode_glops,
598 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
599 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
600 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
601 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
602 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
603 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
604 };
605
This page took 0.046857 seconds and 6 git commands to generate.