1
2
3
4
5
6
7
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#include <linux/security.h>
17
18#include "gfs2.h"
19#include "incore.h"
20#include "bmap.h"
21#include "glock.h"
22#include "glops.h"
23#include "inode.h"
24#include "log.h"
25#include "meta_io.h"
26#include "recovery.h"
27#include "rgrp.h"
28#include "util.h"
29#include "trans.h"
30#include "dir.h"
31
32struct workqueue_struct *gfs2_freeze_wq;
33
34static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
35{
36 fs_err(gl->gl_name.ln_sbd,
37 "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page "
38 "state 0x%lx\n",
39 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
40 bh->b_page->mapping, bh->b_page->flags);
41 fs_err(gl->gl_name.ln_sbd, "AIL glock %u:%llu mapping %p\n",
42 gl->gl_name.ln_type, gl->gl_name.ln_number,
43 gfs2_glock2aspace(gl));
44 gfs2_lm_withdraw(gl->gl_name.ln_sbd, "AIL error\n");
45}
46
47
48
49
50
51
52
53
54
55static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
56 unsigned int nr_revokes)
57{
58 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
59 struct list_head *head = &gl->gl_ail_list;
60 struct gfs2_bufdata *bd, *tmp;
61 struct buffer_head *bh;
62 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
63
64 gfs2_log_lock(sdp);
65 spin_lock(&sdp->sd_ail_lock);
66 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
67 if (nr_revokes == 0)
68 break;
69 bh = bd->bd_bh;
70 if (bh->b_state & b_state) {
71 if (fsync)
72 continue;
73 gfs2_ail_error(gl, bh);
74 }
75 gfs2_trans_add_revoke(sdp, bd);
76 nr_revokes--;
77 }
78 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
79 spin_unlock(&sdp->sd_ail_lock);
80 gfs2_log_unlock(sdp);
81}
82
83
84static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
85{
86 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
87 struct gfs2_trans tr;
88
89 memset(&tr, 0, sizeof(tr));
90 INIT_LIST_HEAD(&tr.tr_buf);
91 INIT_LIST_HEAD(&tr.tr_databuf);
92 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
93
94 if (!tr.tr_revokes)
95 return;
96
97
98
99
100 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
101 tr.tr_ip = _RET_IP_;
102 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
103 return;
104 WARN_ON_ONCE(current->journal_info);
105 current->journal_info = &tr;
106
107 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
108
109 gfs2_trans_end(sdp);
110 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
111 GFS2_LFC_AIL_EMPTY_GL);
112}
113
114void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
115{
116 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
117 unsigned int revokes = atomic_read(&gl->gl_ail_count);
118 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
119 int ret;
120
121 if (!revokes)
122 return;
123
124 while (revokes > max_revokes)
125 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
126
127 ret = gfs2_trans_begin(sdp, 0, max_revokes);
128 if (ret)
129 return;
130 __gfs2_ail_flush(gl, fsync, max_revokes);
131 gfs2_trans_end(sdp);
132 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
133 GFS2_LFC_AIL_FLUSH);
134}
135
136
137
138
139
140
141
142
143
144
145static void rgrp_go_sync(struct gfs2_glock *gl)
146{
147 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
148 struct address_space *mapping = &sdp->sd_aspace;
149 struct gfs2_rgrpd *rgd;
150 int error;
151
152 spin_lock(&gl->gl_lockref.lock);
153 rgd = gl->gl_object;
154 if (rgd)
155 gfs2_rgrp_brelse(rgd);
156 spin_unlock(&gl->gl_lockref.lock);
157
158 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
159 return;
160 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
161
162 gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
163 GFS2_LFC_RGRP_GO_SYNC);
164 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
165 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
166 mapping_set_error(mapping, error);
167 gfs2_ail_empty_gl(gl);
168
169 spin_lock(&gl->gl_lockref.lock);
170 rgd = gl->gl_object;
171 if (rgd)
172 gfs2_free_clones(rgd);
173 spin_unlock(&gl->gl_lockref.lock);
174}
175
176
177
178
179
180
181
182
183
184
185
186static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
187{
188 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
189 struct address_space *mapping = &sdp->sd_aspace;
190 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
191
192 if (rgd)
193 gfs2_rgrp_brelse(rgd);
194
195 WARN_ON_ONCE(!(flags & DIO_METADATA));
196 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
197 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
198
199 if (rgd)
200 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
201}
202
203static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
204{
205 struct gfs2_inode *ip;
206
207 spin_lock(&gl->gl_lockref.lock);
208 ip = gl->gl_object;
209 if (ip)
210 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
211 spin_unlock(&gl->gl_lockref.lock);
212 return ip;
213}
214
215struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
216{
217 struct gfs2_rgrpd *rgd;
218
219 spin_lock(&gl->gl_lockref.lock);
220 rgd = gl->gl_object;
221 spin_unlock(&gl->gl_lockref.lock);
222
223 return rgd;
224}
225
226static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
227{
228 if (!ip)
229 return;
230
231 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
232 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
233}
234
235
236
237
238
239
240
241static void inode_go_sync(struct gfs2_glock *gl)
242{
243 struct gfs2_inode *ip = gfs2_glock2inode(gl);
244 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
245 struct address_space *metamapping = gfs2_glock2aspace(gl);
246 int error;
247
248 if (isreg) {
249 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
250 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
251 inode_dio_wait(&ip->i_inode);
252 }
253 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
254 goto out;
255
256 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
257
258 gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
259 GFS2_LFC_INODE_GO_SYNC);
260 filemap_fdatawrite(metamapping);
261 if (isreg) {
262 struct address_space *mapping = ip->i_inode.i_mapping;
263 filemap_fdatawrite(mapping);
264 error = filemap_fdatawait(mapping);
265 mapping_set_error(mapping, error);
266 }
267 error = filemap_fdatawait(metamapping);
268 mapping_set_error(metamapping, error);
269 gfs2_ail_empty_gl(gl);
270
271
272
273
274 smp_mb__before_atomic();
275 clear_bit(GLF_DIRTY, &gl->gl_flags);
276
277out:
278 gfs2_clear_glop_pending(ip);
279}
280
281
282
283
284
285
286
287
288
289
290
291
292static void inode_go_inval(struct gfs2_glock *gl, int flags)
293{
294 struct gfs2_inode *ip = gfs2_glock2inode(gl);
295
296 gfs2_assert_withdraw(gl->gl_name.ln_sbd, !atomic_read(&gl->gl_ail_count));
297
298 if (flags & DIO_METADATA) {
299 struct address_space *mapping = gfs2_glock2aspace(gl);
300 truncate_inode_pages(mapping, 0);
301 if (ip) {
302 set_bit(GIF_INVALID, &ip->i_flags);
303 forget_all_cached_acls(&ip->i_inode);
304 security_inode_invalidate_secctx(&ip->i_inode);
305 gfs2_dir_hash_inval(ip);
306 }
307 }
308
309 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
310 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
311 GFS2_LOG_HEAD_FLUSH_NORMAL |
312 GFS2_LFC_INODE_GO_INVAL);
313 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
314 }
315 if (ip && S_ISREG(ip->i_inode.i_mode))
316 truncate_inode_pages(ip->i_inode.i_mapping, 0);
317
318 gfs2_clear_glop_pending(ip);
319}
320
321
322
323
324
325
326
327
328static int inode_go_demote_ok(const struct gfs2_glock *gl)
329{
330 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
331
332 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
333 return 0;
334
335 return 1;
336}
337
338static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
339{
340 const struct gfs2_dinode *str = buf;
341 struct timespec atime;
342 u16 height, depth;
343
344 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
345 goto corrupt;
346 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
347 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
348 ip->i_inode.i_rdev = 0;
349 switch (ip->i_inode.i_mode & S_IFMT) {
350 case S_IFBLK:
351 case S_IFCHR:
352 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
353 be32_to_cpu(str->di_minor));
354 break;
355 };
356
357 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
358 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
359 set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
360 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
361 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
362 atime.tv_sec = be64_to_cpu(str->di_atime);
363 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
364 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
365 ip->i_inode.i_atime = atime;
366 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
367 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
368 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
369 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
370
371 ip->i_goal = be64_to_cpu(str->di_goal_meta);
372 ip->i_generation = be64_to_cpu(str->di_generation);
373
374 ip->i_diskflags = be32_to_cpu(str->di_flags);
375 ip->i_eattr = be64_to_cpu(str->di_eattr);
376
377 gfs2_set_inode_flags(&ip->i_inode);
378 height = be16_to_cpu(str->di_height);
379 if (unlikely(height > GFS2_MAX_META_HEIGHT))
380 goto corrupt;
381 ip->i_height = (u8)height;
382
383 depth = be16_to_cpu(str->di_depth);
384 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
385 goto corrupt;
386 ip->i_depth = (u8)depth;
387 ip->i_entries = be32_to_cpu(str->di_entries);
388
389 if (S_ISREG(ip->i_inode.i_mode))
390 gfs2_set_aops(&ip->i_inode);
391
392 return 0;
393corrupt:
394 gfs2_consist_inode(ip);
395 return -EIO;
396}
397
398
399
400
401
402
403
404
405int gfs2_inode_refresh(struct gfs2_inode *ip)
406{
407 struct buffer_head *dibh;
408 int error;
409
410 error = gfs2_meta_inode_buffer(ip, &dibh);
411 if (error)
412 return error;
413
414 error = gfs2_dinode_in(ip, dibh->b_data);
415 brelse(dibh);
416 clear_bit(GIF_INVALID, &ip->i_flags);
417
418 return error;
419}
420
421
422
423
424
425
426
427
428
429static int inode_go_lock(struct gfs2_holder *gh)
430{
431 struct gfs2_glock *gl = gh->gh_gl;
432 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
433 struct gfs2_inode *ip = gl->gl_object;
434 int error = 0;
435
436 if (!ip || (gh->gh_flags & GL_SKIP))
437 return 0;
438
439 if (test_bit(GIF_INVALID, &ip->i_flags)) {
440 error = gfs2_inode_refresh(ip);
441 if (error)
442 return error;
443 }
444
445 if (gh->gh_state != LM_ST_DEFERRED)
446 inode_dio_wait(&ip->i_inode);
447
448 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
449 (gl->gl_state == LM_ST_EXCLUSIVE) &&
450 (gh->gh_state == LM_ST_EXCLUSIVE)) {
451 spin_lock(&sdp->sd_trunc_lock);
452 if (list_empty(&ip->i_trunc_list))
453 list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
454 spin_unlock(&sdp->sd_trunc_lock);
455 wake_up(&sdp->sd_quota_wait);
456 return 1;
457 }
458
459 return error;
460}
461
462
463
464
465
466
467
468
469static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
470{
471 const struct gfs2_inode *ip = gl->gl_object;
472 if (ip == NULL)
473 return;
474 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
475 (unsigned long long)ip->i_no_formal_ino,
476 (unsigned long long)ip->i_no_addr,
477 IF2DT(ip->i_inode.i_mode), ip->i_flags,
478 (unsigned int)ip->i_diskflags,
479 (unsigned long long)i_size_read(&ip->i_inode));
480}
481
482
483
484
485
486
487
488
489
490static void freeze_go_sync(struct gfs2_glock *gl)
491{
492 int error = 0;
493 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
494
495 if (gl->gl_state == LM_ST_SHARED &&
496 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
497 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
498 error = freeze_super(sdp->sd_vfs);
499 if (error) {
500 printk(KERN_INFO "GFS2: couldn't freeze filesystem: %d\n", error);
501 gfs2_assert_withdraw(sdp, 0);
502 }
503 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
504 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
505 GFS2_LFC_FREEZE_GO_SYNC);
506 }
507}
508
509
510
511
512
513
514
515static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
516{
517 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
518 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
519 struct gfs2_glock *j_gl = ip->i_gl;
520 struct gfs2_log_header_host head;
521 int error;
522
523 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
524 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
525
526 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
527 if (error)
528 gfs2_consist(sdp);
529 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
530 gfs2_consist(sdp);
531
532
533 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
534 sdp->sd_log_sequence = head.lh_sequence + 1;
535 gfs2_log_pointers_init(sdp, head.lh_blkno);
536 }
537 }
538 return 0;
539}
540
541
542
543
544
545
546
547
548static int freeze_go_demote_ok(const struct gfs2_glock *gl)
549{
550 return 0;
551}
552
553
554
555
556
557
558
559static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
560{
561 struct gfs2_inode *ip = gl->gl_object;
562 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
563
564 if (!remote || sb_rdonly(sdp->sd_vfs))
565 return;
566
567 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
568 gl->gl_state == LM_ST_SHARED && ip) {
569 gl->gl_lockref.count++;
570 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
571 gl->gl_lockref.count--;
572 }
573}
574
575const struct gfs2_glock_operations gfs2_meta_glops = {
576 .go_type = LM_TYPE_META,
577};
578
579const struct gfs2_glock_operations gfs2_inode_glops = {
580 .go_sync = inode_go_sync,
581 .go_inval = inode_go_inval,
582 .go_demote_ok = inode_go_demote_ok,
583 .go_lock = inode_go_lock,
584 .go_dump = inode_go_dump,
585 .go_type = LM_TYPE_INODE,
586 .go_flags = GLOF_ASPACE | GLOF_LRU,
587};
588
589const struct gfs2_glock_operations gfs2_rgrp_glops = {
590 .go_sync = rgrp_go_sync,
591 .go_inval = rgrp_go_inval,
592 .go_lock = gfs2_rgrp_go_lock,
593 .go_unlock = gfs2_rgrp_go_unlock,
594 .go_dump = gfs2_rgrp_dump,
595 .go_type = LM_TYPE_RGRP,
596 .go_flags = GLOF_LVB,
597};
598
599const struct gfs2_glock_operations gfs2_freeze_glops = {
600 .go_sync = freeze_go_sync,
601 .go_xmote_bh = freeze_go_xmote_bh,
602 .go_demote_ok = freeze_go_demote_ok,
603 .go_type = LM_TYPE_NONDISK,
604};
605
606const struct gfs2_glock_operations gfs2_iopen_glops = {
607 .go_type = LM_TYPE_IOPEN,
608 .go_callback = iopen_go_callback,
609 .go_flags = GLOF_LRU,
610};
611
612const struct gfs2_glock_operations gfs2_flock_glops = {
613 .go_type = LM_TYPE_FLOCK,
614 .go_flags = GLOF_LRU,
615};
616
617const struct gfs2_glock_operations gfs2_nondisk_glops = {
618 .go_type = LM_TYPE_NONDISK,
619};
620
621const struct gfs2_glock_operations gfs2_quota_glops = {
622 .go_type = LM_TYPE_QUOTA,
623 .go_flags = GLOF_LVB | GLOF_LRU,
624};
625
626const struct gfs2_glock_operations gfs2_journal_glops = {
627 .go_type = LM_TYPE_JOURNAL,
628};
629
630const struct gfs2_glock_operations *gfs2_glops_list[] = {
631 [LM_TYPE_META] = &gfs2_meta_glops,
632 [LM_TYPE_INODE] = &gfs2_inode_glops,
633 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
634 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
635 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
636 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
637 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
638 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
639};
640
641