1
2
3
4
5
6
7#include <linux/spinlock.h>
8#include <linux/completion.h>
9#include <linux/buffer_head.h>
10#include <linux/gfs2_ondisk.h>
11#include <linux/bio.h>
12#include <linux/posix_acl.h>
13#include <linux/security.h>
14
15#include "gfs2.h"
16#include "incore.h"
17#include "bmap.h"
18#include "glock.h"
19#include "glops.h"
20#include "inode.h"
21#include "log.h"
22#include "meta_io.h"
23#include "recovery.h"
24#include "rgrp.h"
25#include "util.h"
26#include "trans.h"
27#include "dir.h"
28#include "lops.h"
29
30struct workqueue_struct *gfs2_freeze_wq;
31
32static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
33{
34 fs_err(gl->gl_name.ln_sbd,
35 "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page "
36 "state 0x%lx\n",
37 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
38 bh->b_page->mapping, bh->b_page->flags);
39 fs_err(gl->gl_name.ln_sbd, "AIL glock %u:%llu mapping %p\n",
40 gl->gl_name.ln_type, gl->gl_name.ln_number,
41 gfs2_glock2aspace(gl));
42 gfs2_lm_withdraw(gl->gl_name.ln_sbd, "AIL error\n");
43}
44
45
46
47
48
49
50
51
52
53static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
54 unsigned int nr_revokes)
55{
56 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
57 struct list_head *head = &gl->gl_ail_list;
58 struct gfs2_bufdata *bd, *tmp;
59 struct buffer_head *bh;
60 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
61
62 gfs2_log_lock(sdp);
63 spin_lock(&sdp->sd_ail_lock);
64 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
65 if (nr_revokes == 0)
66 break;
67 bh = bd->bd_bh;
68 if (bh->b_state & b_state) {
69 if (fsync)
70 continue;
71 gfs2_ail_error(gl, bh);
72 }
73 gfs2_trans_add_revoke(sdp, bd);
74 nr_revokes--;
75 }
76 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
77 spin_unlock(&sdp->sd_ail_lock);
78 gfs2_log_unlock(sdp);
79}
80
81
82static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
83{
84 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
85 struct gfs2_trans tr;
86
87 memset(&tr, 0, sizeof(tr));
88 INIT_LIST_HEAD(&tr.tr_buf);
89 INIT_LIST_HEAD(&tr.tr_databuf);
90 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
91
92 if (!tr.tr_revokes)
93 return;
94
95
96
97
98 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
99 tr.tr_ip = _RET_IP_;
100 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
101 return;
102 WARN_ON_ONCE(current->journal_info);
103 current->journal_info = &tr;
104
105 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
106
107 gfs2_trans_end(sdp);
108 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
109 GFS2_LFC_AIL_EMPTY_GL);
110}
111
112void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
113{
114 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
115 unsigned int revokes = atomic_read(&gl->gl_ail_count);
116 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
117 int ret;
118
119 if (!revokes)
120 return;
121
122 while (revokes > max_revokes)
123 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
124
125 ret = gfs2_trans_begin(sdp, 0, max_revokes);
126 if (ret)
127 return;
128 __gfs2_ail_flush(gl, fsync, max_revokes);
129 gfs2_trans_end(sdp);
130 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
131 GFS2_LFC_AIL_FLUSH);
132}
133
134
135
136
137
138
139
140
141
142
143static void rgrp_go_sync(struct gfs2_glock *gl)
144{
145 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
146 struct address_space *mapping = &sdp->sd_aspace;
147 struct gfs2_rgrpd *rgd;
148 int error;
149
150 spin_lock(&gl->gl_lockref.lock);
151 rgd = gl->gl_object;
152 if (rgd)
153 gfs2_rgrp_brelse(rgd);
154 spin_unlock(&gl->gl_lockref.lock);
155
156 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
157 return;
158 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
159
160 gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
161 GFS2_LFC_RGRP_GO_SYNC);
162 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
163 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
164 mapping_set_error(mapping, error);
165 gfs2_ail_empty_gl(gl);
166
167 spin_lock(&gl->gl_lockref.lock);
168 rgd = gl->gl_object;
169 if (rgd)
170 gfs2_free_clones(rgd);
171 spin_unlock(&gl->gl_lockref.lock);
172}
173
174
175
176
177
178
179
180
181
182
183
184static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
185{
186 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
187 struct address_space *mapping = &sdp->sd_aspace;
188 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
189
190 if (rgd)
191 gfs2_rgrp_brelse(rgd);
192
193 WARN_ON_ONCE(!(flags & DIO_METADATA));
194 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
195 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
196
197 if (rgd)
198 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
199}
200
201static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
202{
203 struct gfs2_inode *ip;
204
205 spin_lock(&gl->gl_lockref.lock);
206 ip = gl->gl_object;
207 if (ip)
208 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
209 spin_unlock(&gl->gl_lockref.lock);
210 return ip;
211}
212
213struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
214{
215 struct gfs2_rgrpd *rgd;
216
217 spin_lock(&gl->gl_lockref.lock);
218 rgd = gl->gl_object;
219 spin_unlock(&gl->gl_lockref.lock);
220
221 return rgd;
222}
223
224static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
225{
226 if (!ip)
227 return;
228
229 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
230 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
231}
232
233
234
235
236
237
238
239static void inode_go_sync(struct gfs2_glock *gl)
240{
241 struct gfs2_inode *ip = gfs2_glock2inode(gl);
242 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
243 struct address_space *metamapping = gfs2_glock2aspace(gl);
244 int error;
245
246 if (isreg) {
247 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
248 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
249 inode_dio_wait(&ip->i_inode);
250 }
251 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
252 goto out;
253
254 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
255
256 gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
257 GFS2_LFC_INODE_GO_SYNC);
258 filemap_fdatawrite(metamapping);
259 if (isreg) {
260 struct address_space *mapping = ip->i_inode.i_mapping;
261 filemap_fdatawrite(mapping);
262 error = filemap_fdatawait(mapping);
263 mapping_set_error(mapping, error);
264 }
265 error = filemap_fdatawait(metamapping);
266 mapping_set_error(metamapping, error);
267 gfs2_ail_empty_gl(gl);
268
269
270
271
272 smp_mb__before_atomic();
273 clear_bit(GLF_DIRTY, &gl->gl_flags);
274
275out:
276 gfs2_clear_glop_pending(ip);
277}
278
279
280
281
282
283
284
285
286
287
288
289
290static void inode_go_inval(struct gfs2_glock *gl, int flags)
291{
292 struct gfs2_inode *ip = gfs2_glock2inode(gl);
293
294 gfs2_assert_withdraw(gl->gl_name.ln_sbd, !atomic_read(&gl->gl_ail_count));
295
296 if (flags & DIO_METADATA) {
297 struct address_space *mapping = gfs2_glock2aspace(gl);
298 truncate_inode_pages(mapping, 0);
299 if (ip) {
300 set_bit(GIF_INVALID, &ip->i_flags);
301 forget_all_cached_acls(&ip->i_inode);
302 security_inode_invalidate_secctx(&ip->i_inode);
303 gfs2_dir_hash_inval(ip);
304 }
305 }
306
307 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
308 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
309 GFS2_LOG_HEAD_FLUSH_NORMAL |
310 GFS2_LFC_INODE_GO_INVAL);
311 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
312 }
313 if (ip && S_ISREG(ip->i_inode.i_mode))
314 truncate_inode_pages(ip->i_inode.i_mapping, 0);
315
316 gfs2_clear_glop_pending(ip);
317}
318
319
320
321
322
323
324
325
326static int inode_go_demote_ok(const struct gfs2_glock *gl)
327{
328 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
329
330 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
331 return 0;
332
333 return 1;
334}
335
336static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
337{
338 const struct gfs2_dinode *str = buf;
339 struct timespec64 atime;
340 u16 height, depth;
341
342 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
343 goto corrupt;
344 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
345 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
346 ip->i_inode.i_rdev = 0;
347 switch (ip->i_inode.i_mode & S_IFMT) {
348 case S_IFBLK:
349 case S_IFCHR:
350 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
351 be32_to_cpu(str->di_minor));
352 break;
353 };
354
355 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
356 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
357 set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
358 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
359 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
360 atime.tv_sec = be64_to_cpu(str->di_atime);
361 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
362 if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
363 ip->i_inode.i_atime = atime;
364 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
365 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
366 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
367 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
368
369 ip->i_goal = be64_to_cpu(str->di_goal_meta);
370 ip->i_generation = be64_to_cpu(str->di_generation);
371
372 ip->i_diskflags = be32_to_cpu(str->di_flags);
373 ip->i_eattr = be64_to_cpu(str->di_eattr);
374
375 gfs2_set_inode_flags(&ip->i_inode);
376 height = be16_to_cpu(str->di_height);
377 if (unlikely(height > GFS2_MAX_META_HEIGHT))
378 goto corrupt;
379 ip->i_height = (u8)height;
380
381 depth = be16_to_cpu(str->di_depth);
382 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
383 goto corrupt;
384 ip->i_depth = (u8)depth;
385 ip->i_entries = be32_to_cpu(str->di_entries);
386
387 if (S_ISREG(ip->i_inode.i_mode))
388 gfs2_set_aops(&ip->i_inode);
389
390 return 0;
391corrupt:
392 gfs2_consist_inode(ip);
393 return -EIO;
394}
395
396
397
398
399
400
401
402
403int gfs2_inode_refresh(struct gfs2_inode *ip)
404{
405 struct buffer_head *dibh;
406 int error;
407
408 error = gfs2_meta_inode_buffer(ip, &dibh);
409 if (error)
410 return error;
411
412 error = gfs2_dinode_in(ip, dibh->b_data);
413 brelse(dibh);
414 clear_bit(GIF_INVALID, &ip->i_flags);
415
416 return error;
417}
418
419
420
421
422
423
424
425
426
427static int inode_go_lock(struct gfs2_holder *gh)
428{
429 struct gfs2_glock *gl = gh->gh_gl;
430 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
431 struct gfs2_inode *ip = gl->gl_object;
432 int error = 0;
433
434 if (!ip || (gh->gh_flags & GL_SKIP))
435 return 0;
436
437 if (test_bit(GIF_INVALID, &ip->i_flags)) {
438 error = gfs2_inode_refresh(ip);
439 if (error)
440 return error;
441 }
442
443 if (gh->gh_state != LM_ST_DEFERRED)
444 inode_dio_wait(&ip->i_inode);
445
446 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
447 (gl->gl_state == LM_ST_EXCLUSIVE) &&
448 (gh->gh_state == LM_ST_EXCLUSIVE)) {
449 spin_lock(&sdp->sd_trunc_lock);
450 if (list_empty(&ip->i_trunc_list))
451 list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
452 spin_unlock(&sdp->sd_trunc_lock);
453 wake_up(&sdp->sd_quota_wait);
454 return 1;
455 }
456
457 return error;
458}
459
460
461
462
463
464
465
466
467
468static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
469 const char *fs_id_buf)
470{
471 struct gfs2_inode *ip = gl->gl_object;
472 struct inode *inode = &ip->i_inode;
473 unsigned long nrpages;
474
475 if (ip == NULL)
476 return;
477
478 xa_lock_irq(&inode->i_data.i_pages);
479 nrpages = inode->i_data.nrpages;
480 xa_unlock_irq(&inode->i_data.i_pages);
481
482 gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
483 "p:%lu\n", fs_id_buf,
484 (unsigned long long)ip->i_no_formal_ino,
485 (unsigned long long)ip->i_no_addr,
486 IF2DT(ip->i_inode.i_mode), ip->i_flags,
487 (unsigned int)ip->i_diskflags,
488 (unsigned long long)i_size_read(inode), nrpages);
489}
490
491
492
493
494
495
496
497
498
499static void freeze_go_sync(struct gfs2_glock *gl)
500{
501 int error = 0;
502 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
503
504 if (gl->gl_state == LM_ST_SHARED &&
505 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
506 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
507 error = freeze_super(sdp->sd_vfs);
508 if (error) {
509 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
510 error);
511 gfs2_assert_withdraw(sdp, 0);
512 }
513 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
514 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
515 GFS2_LFC_FREEZE_GO_SYNC);
516 }
517}
518
519
520
521
522
523
524
525static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
526{
527 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
528 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
529 struct gfs2_glock *j_gl = ip->i_gl;
530 struct gfs2_log_header_host head;
531 int error;
532
533 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
534 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
535
536 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
537 if (error)
538 gfs2_consist(sdp);
539 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
540 gfs2_consist(sdp);
541
542
543 if (!test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
544 sdp->sd_log_sequence = head.lh_sequence + 1;
545 gfs2_log_pointers_init(sdp, head.lh_blkno);
546 }
547 }
548 return 0;
549}
550
551
552
553
554
555
556
557
558static int freeze_go_demote_ok(const struct gfs2_glock *gl)
559{
560 return 0;
561}
562
563
564
565
566
567
568
569static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
570{
571 struct gfs2_inode *ip = gl->gl_object;
572 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
573
574 if (!remote || sb_rdonly(sdp->sd_vfs))
575 return;
576
577 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
578 gl->gl_state == LM_ST_SHARED && ip) {
579 gl->gl_lockref.count++;
580 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
581 gl->gl_lockref.count--;
582 }
583}
584
585const struct gfs2_glock_operations gfs2_meta_glops = {
586 .go_type = LM_TYPE_META,
587};
588
589const struct gfs2_glock_operations gfs2_inode_glops = {
590 .go_sync = inode_go_sync,
591 .go_inval = inode_go_inval,
592 .go_demote_ok = inode_go_demote_ok,
593 .go_lock = inode_go_lock,
594 .go_dump = inode_go_dump,
595 .go_type = LM_TYPE_INODE,
596 .go_flags = GLOF_ASPACE | GLOF_LRU,
597};
598
599const struct gfs2_glock_operations gfs2_rgrp_glops = {
600 .go_sync = rgrp_go_sync,
601 .go_inval = rgrp_go_inval,
602 .go_lock = gfs2_rgrp_go_lock,
603 .go_unlock = gfs2_rgrp_go_unlock,
604 .go_dump = gfs2_rgrp_dump,
605 .go_type = LM_TYPE_RGRP,
606 .go_flags = GLOF_LVB,
607};
608
609const struct gfs2_glock_operations gfs2_freeze_glops = {
610 .go_sync = freeze_go_sync,
611 .go_xmote_bh = freeze_go_xmote_bh,
612 .go_demote_ok = freeze_go_demote_ok,
613 .go_type = LM_TYPE_NONDISK,
614};
615
616const struct gfs2_glock_operations gfs2_iopen_glops = {
617 .go_type = LM_TYPE_IOPEN,
618 .go_callback = iopen_go_callback,
619 .go_flags = GLOF_LRU,
620};
621
622const struct gfs2_glock_operations gfs2_flock_glops = {
623 .go_type = LM_TYPE_FLOCK,
624 .go_flags = GLOF_LRU,
625};
626
627const struct gfs2_glock_operations gfs2_nondisk_glops = {
628 .go_type = LM_TYPE_NONDISK,
629};
630
631const struct gfs2_glock_operations gfs2_quota_glops = {
632 .go_type = LM_TYPE_QUOTA,
633 .go_flags = GLOF_LVB | GLOF_LRU,
634};
635
636const struct gfs2_glock_operations gfs2_journal_glops = {
637 .go_type = LM_TYPE_JOURNAL,
638};
639
640const struct gfs2_glock_operations *gfs2_glops_list[] = {
641 [LM_TYPE_META] = &gfs2_meta_glops,
642 [LM_TYPE_INODE] = &gfs2_inode_glops,
643 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
644 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
645 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
646 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
647 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
648 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
649};
650
651