1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
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
30 struct workqueue_struct *gfs2_freeze_wq;
31
32 extern struct workqueue_struct *gfs2_control_wq;
33
34 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
35 {
36 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
37
38 fs_err(sdp,
39 "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page "
40 "state 0x%lx\n",
41 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
42 bh->b_folio->mapping, bh->b_folio->flags);
43 fs_err(sdp, "AIL glock %u:%llu mapping %p\n",
44 gl->gl_name.ln_type, gl->gl_name.ln_number,
45 gfs2_glock2aspace(gl));
46 gfs2_lm(sdp, "AIL error\n");
47 gfs2_withdraw_delayed(sdp);
48 }
49
50 /**
51 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
52 * @gl: the glock
53 * @fsync: set when called from fsync (not all buffers will be clean)
54 * @nr_revokes: Number of buffers to revoke
55 *
56 * None of the buffers should be dirty, locked, or pinned.
57 */
58
59 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
60 unsigned int nr_revokes)
61 {
62 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
63 struct list_head *head = &gl->gl_ail_list;
64 struct gfs2_bufdata *bd, *tmp;
65 struct buffer_head *bh;
66 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
67
68 gfs2_log_lock(sdp);
69 spin_lock(&sdp->sd_ail_lock);
70 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
71 if (nr_revokes == 0)
72 break;
73 bh = bd->bd_bh;
74 if (bh->b_state & b_state) {
75 if (fsync)
76 continue;
77 gfs2_ail_error(gl, bh);
78 }
79 gfs2_trans_add_revoke(sdp, bd);
80 nr_revokes--;
81 }
82 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
83 spin_unlock(&sdp->sd_ail_lock);
84 gfs2_log_unlock(sdp);
85
86 if (gfs2_withdrawing(sdp))
87 gfs2_withdraw(sdp);
88 }
89
90
91 static int gfs2_ail_empty_gl(struct gfs2_glock *gl)
92 {
93 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
94 struct gfs2_trans tr;
95 unsigned int revokes;
96 int ret = 0;
97
98 revokes = atomic_read(&gl->gl_ail_count);
99
100 if (!revokes) {
101 bool have_revokes;
102 bool log_in_flight;
103
104 /*
105 * We have nothing on the ail, but there could be revokes on
106 * the sdp revoke queue, in which case, we still want to flush
107 * the log and wait for it to finish.
108 *
109 * If the sdp revoke list is empty too, we might still have an
110 * io outstanding for writing revokes, so we should wait for
111 * it before returning.
112 *
113 * If none of these conditions are true, our revokes are all
114 * flushed and we can return.
115 */
116 gfs2_log_lock(sdp);
117 have_revokes = !list_empty(&sdp->sd_log_revokes);
118 log_in_flight = atomic_read(&sdp->sd_log_in_flight);
119 gfs2_log_unlock(sdp);
120 if (have_revokes)
121 goto flush;
122 if (log_in_flight)
123 log_flush_wait(sdp);
124 return 0;
125 }
126
127 memset(&tr, 0, sizeof(tr));
128 set_bit(TR_ONSTACK, &tr.tr_flags);
129 ret = __gfs2_trans_begin(&tr, sdp, 0, revokes, _RET_IP_);
130 if (ret) {
131 fs_err(sdp, "Transaction error %d: Unable to write revokes.", ret);
132 goto flush;
133 }
134 __gfs2_ail_flush(gl, 0, revokes);
135 gfs2_trans_end(sdp);
136
137 flush:
138 if (!ret)
139 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
140 GFS2_LFC_AIL_EMPTY_GL);
141 return ret;
142 }
143
144 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
145 {
146 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
147 unsigned int revokes = atomic_read(&gl->gl_ail_count);
148 int ret;
149
150 if (!revokes)
151 return;
152
153 ret = gfs2_trans_begin(sdp, 0, revokes);
154 if (ret)
155 return;
156 __gfs2_ail_flush(gl, fsync, revokes);
157 gfs2_trans_end(sdp);
158 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
159 GFS2_LFC_AIL_FLUSH);
160 }
161
162 /**
163 * gfs2_rgrp_metasync - sync out the metadata of a resource group
164 * @gl: the glock protecting the resource group
165 *
166 */
167
168 static int gfs2_rgrp_metasync(struct gfs2_glock *gl)
169 {
170 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
171 struct address_space *metamapping = &sdp->sd_aspace;
172 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
173 const unsigned bsize = sdp->sd_sb.sb_bsize;
174 loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
175 loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
176 int error;
177
178 filemap_fdatawrite_range(metamapping, start, end);
179 error = filemap_fdatawait_range(metamapping, start, end);
180 WARN_ON_ONCE(error && !gfs2_withdrawing_or_withdrawn(sdp));
181 mapping_set_error(metamapping, error);
182 if (error)
183 gfs2_io_error(sdp);
184 return error;
185 }
186
187 /**
188 * rgrp_go_sync - sync out the metadata for this glock
189 * @gl: the glock
190 *
191 * Called when demoting or unlocking an EX glock. We must flush
192 * to disk all dirty buffers/pages relating to this glock, and must not
193 * return to caller to demote/unlock the glock until I/O is complete.
194 */
195
196 static int rgrp_go_sync(struct gfs2_glock *gl)
197 {
198 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
199 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
200 int error;
201
202 if (!rgd || !test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
203 return 0;
204 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
205
206 gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
207 GFS2_LFC_RGRP_GO_SYNC);
208 error = gfs2_rgrp_metasync(gl);
209 if (!error)
210 error = gfs2_ail_empty_gl(gl);
211 gfs2_free_clones(rgd);
212 return error;
213 }
214
215 /**
216 * rgrp_go_inval - invalidate the metadata for this glock
217 * @gl: the glock
218 * @flags:
219 *
220 * We never used LM_ST_DEFERRED with resource groups, so that we
221 * should always see the metadata flag set here.
222 *
223 */
224
225 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
226 {
227 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
228 struct address_space *mapping = &sdp->sd_aspace;
229 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
230 const unsigned bsize = sdp->sd_sb.sb_bsize;
231 loff_t start, end;
232
233 if (!rgd)
234 return;
235 start = (rgd->rd_addr * bsize) & PAGE_MASK;
236 end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
237 gfs2_rgrp_brelse(rgd);
238 WARN_ON_ONCE(!(flags & DIO_METADATA));
239 truncate_inode_pages_range(mapping, start, end);
240 }
241
242 static void gfs2_rgrp_go_dump(struct seq_file *seq, const struct gfs2_glock *gl,
243 const char *fs_id_buf)
244 {
245 struct gfs2_rgrpd *rgd = gl->gl_object;
246
247 if (rgd)
248 gfs2_rgrp_dump(seq, rgd, fs_id_buf);
249 }
250
251 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
252 {
253 struct gfs2_inode *ip;
254
255 spin_lock(&gl->gl_lockref.lock);
256 ip = gl->gl_object;
257 if (ip)
258 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
259 spin_unlock(&gl->gl_lockref.lock);
260 return ip;
261 }
262
263 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
264 {
265 struct gfs2_rgrpd *rgd;
266
267 spin_lock(&gl->gl_lockref.lock);
268 rgd = gl->gl_object;
269 spin_unlock(&gl->gl_lockref.lock);
270
271 return rgd;
272 }
273
274 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
275 {
276 if (!ip)
277 return;
278
279 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
280 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
281 }
282
283 /**
284 * gfs2_inode_metasync - sync out the metadata of an inode
285 * @gl: the glock protecting the inode
286 *
287 */
288 int gfs2_inode_metasync(struct gfs2_glock *gl)
289 {
290 struct address_space *metamapping = gfs2_glock2aspace(gl);
291 int error;
292
293 filemap_fdatawrite(metamapping);
294 error = filemap_fdatawait(metamapping);
295 if (error)
296 gfs2_io_error(gl->gl_name.ln_sbd);
297 return error;
298 }
299
300 /**
301 * inode_go_sync - Sync the dirty metadata of an inode
302 * @gl: the glock protecting the inode
303 *
304 */
305
306 static int inode_go_sync(struct gfs2_glock *gl)
307 {
308 struct gfs2_inode *ip = gfs2_glock2inode(gl);
309 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
310 struct address_space *metamapping = gfs2_glock2aspace(gl);
311 int error = 0, ret;
312
313 if (isreg) {
314 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
315 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
316 inode_dio_wait(&ip->i_inode);
317 }
318 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
319 goto out;
320
321 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
322
323 gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
324 GFS2_LFC_INODE_GO_SYNC);
325 filemap_fdatawrite(metamapping);
326 if (isreg) {
327 struct address_space *mapping = ip->i_inode.i_mapping;
328 filemap_fdatawrite(mapping);
329 error = filemap_fdatawait(mapping);
330 mapping_set_error(mapping, error);
331 }
332 ret = gfs2_inode_metasync(gl);
333 if (!error)
334 error = ret;
335 ret = gfs2_ail_empty_gl(gl);
336 if (!error)
337 error = ret;
338 /*
339 * Writeback of the data mapping may cause the dirty flag to be set
340 * so we have to clear it again here.
341 */
342 smp_mb__before_atomic();
343 clear_bit(GLF_DIRTY, &gl->gl_flags);
344
345 out:
346 gfs2_clear_glop_pending(ip);
347 return error;
348 }
349
350 /**
351 * inode_go_inval - prepare a inode glock to be released
352 * @gl: the glock
353 * @flags:
354 *
355 * Normally we invalidate everything, but if we are moving into
356 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
357 * can keep hold of the metadata, since it won't have changed.
358 *
359 */
360
361 static void inode_go_inval(struct gfs2_glock *gl, int flags)
362 {
363 struct gfs2_inode *ip = gfs2_glock2inode(gl);
364
365 if (flags & DIO_METADATA) {
366 struct address_space *mapping = gfs2_glock2aspace(gl);
367 truncate_inode_pages(mapping, 0);
368 if (ip) {
369 set_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
370 forget_all_cached_acls(&ip->i_inode);
371 security_inode_invalidate_secctx(&ip->i_inode);
372 gfs2_dir_hash_inval(ip);
373 }
374 }
375
376 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
377 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
378 GFS2_LOG_HEAD_FLUSH_NORMAL |
379 GFS2_LFC_INODE_GO_INVAL);
380 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
381 }
382 if (ip && S_ISREG(ip->i_inode.i_mode))
383 truncate_inode_pages(ip->i_inode.i_mapping, 0);
384
385 gfs2_clear_glop_pending(ip);
386 }
387
388 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
389 {
390 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
391 const struct gfs2_dinode *str = buf;
392 struct timespec64 atime, iatime;
393 u16 height, depth;
394 umode_t mode = be32_to_cpu(str->di_mode);
395 struct inode *inode = &ip->i_inode;
396 bool is_new = inode->i_state & I_NEW;
397
398 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr))) {
399 gfs2_consist_inode(ip);
400 return -EIO;
401 }
402 if (unlikely(!is_new && inode_wrong_type(inode, mode))) {
403 gfs2_consist_inode(ip);
404 return -EIO;
405 }
406 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
407 inode->i_mode = mode;
408 if (is_new) {
409 inode->i_rdev = 0;
410 switch (mode & S_IFMT) {
411 case S_IFBLK:
412 case S_IFCHR:
413 inode->i_rdev = MKDEV(be32_to_cpu(str->di_major),
414 be32_to_cpu(str->di_minor));
415 break;
416 }
417 }
418
419 i_uid_write(inode, be32_to_cpu(str->di_uid));
420 i_gid_write(inode, be32_to_cpu(str->di_gid));
421 set_nlink(inode, be32_to_cpu(str->di_nlink));
422 i_size_write(inode, be64_to_cpu(str->di_size));
423 gfs2_set_inode_blocks(inode, be64_to_cpu(str->di_blocks));
424 atime.tv_sec = be64_to_cpu(str->di_atime);
425 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
426 iatime = inode_get_atime(inode);
427 if (timespec64_compare(&iatime, &atime) < 0)
428 inode_set_atime_to_ts(inode, atime);
429 inode_set_mtime(inode, be64_to_cpu(str->di_mtime),
430 be32_to_cpu(str->di_mtime_nsec));
431 inode_set_ctime(inode, be64_to_cpu(str->di_ctime),
432 be32_to_cpu(str->di_ctime_nsec));
433
434 ip->i_goal = be64_to_cpu(str->di_goal_meta);
435 ip->i_generation = be64_to_cpu(str->di_generation);
436
437 ip->i_diskflags = be32_to_cpu(str->di_flags);
438 ip->i_eattr = be64_to_cpu(str->di_eattr);
439 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
440 gfs2_set_inode_flags(inode);
441 height = be16_to_cpu(str->di_height);
442 if (unlikely(height > sdp->sd_max_height)) {
443 gfs2_consist_inode(ip);
444 return -EIO;
445 }
446 ip->i_height = (u8)height;
447
448 depth = be16_to_cpu(str->di_depth);
449 if (unlikely(depth > GFS2_DIR_MAX_DEPTH)) {
450 gfs2_consist_inode(ip);
451 return -EIO;
452 }
453 ip->i_depth = (u8)depth;
454 ip->i_entries = be32_to_cpu(str->di_entries);
455
456 if (gfs2_is_stuffed(ip) && inode->i_size > gfs2_max_stuffed_size(ip)) {
457 gfs2_consist_inode(ip);
458 return -EIO;
459 }
460 if (S_ISREG(inode->i_mode))
461 gfs2_set_aops(inode);
462
463 return 0;
464 }
465
466 /**
467 * gfs2_inode_refresh - Refresh the incore copy of the dinode
468 * @ip: The GFS2 inode
469 *
470 * Returns: errno
471 */
472
473 int gfs2_inode_refresh(struct gfs2_inode *ip)
474 {
475 struct buffer_head *dibh;
476 int error;
477
478 error = gfs2_meta_inode_buffer(ip, &dibh);
479 if (error)
480 return error;
481
482 error = gfs2_dinode_in(ip, dibh->b_data);
483 brelse(dibh);
484 return error;
485 }
486
487 /**
488 * inode_go_instantiate - read in an inode if necessary
489 * @gl: The glock
490 *
491 * Returns: errno
492 */
493
494 static int inode_go_instantiate(struct gfs2_glock *gl)
495 {
496 struct gfs2_inode *ip = gl->gl_object;
497
498 if (!ip) /* no inode to populate - read it in later */
499 return 0;
500
501 return gfs2_inode_refresh(ip);
502 }
503
504 static int inode_go_held(struct gfs2_holder *gh)
505 {
506 struct gfs2_glock *gl = gh->gh_gl;
507 struct gfs2_inode *ip = gl->gl_object;
508 int error = 0;
509
510 if (!ip) /* no inode to populate - read it in later */
511 return 0;
512
513 if (gh->gh_state != LM_ST_DEFERRED)
514 inode_dio_wait(&ip->i_inode);
515
516 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
517 (gl->gl_state == LM_ST_EXCLUSIVE) &&
518 (gh->gh_state == LM_ST_EXCLUSIVE))
519 error = gfs2_truncatei_resume(ip);
520
521 return error;
522 }
523
524 /**
525 * inode_go_dump - print information about an inode
526 * @seq: The iterator
527 * @gl: The glock
528 * @fs_id_buf: file system id (may be empty)
529 *
530 */
531
532 static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl,
533 const char *fs_id_buf)
534 {
535 struct gfs2_inode *ip = gl->gl_object;
536 const struct inode *inode = &ip->i_inode;
537
538 if (ip == NULL)
539 return;
540
541 gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
542 "p:%lu\n", fs_id_buf,
543 (unsigned long long)ip->i_no_formal_ino,
544 (unsigned long long)ip->i_no_addr,
545 IF2DT(inode->i_mode), ip->i_flags,
546 (unsigned int)ip->i_diskflags,
547 (unsigned long long)i_size_read(inode),
548 inode->i_data.nrpages);
549 }
550
551 /**
552 * freeze_go_callback - A cluster node is requesting a freeze
553 * @gl: the glock
554 * @remote: true if this came from a different cluster node
555 */
556
557 static void freeze_go_callback(struct gfs2_glock *gl, bool remote)
558 {
559 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
560 struct super_block *sb = sdp->sd_vfs;
561
562 if (!remote ||
563 (gl->gl_state != LM_ST_SHARED &&
564 gl->gl_state != LM_ST_UNLOCKED) ||
565 gl->gl_demote_state != LM_ST_UNLOCKED)
566 return;
567
568 /*
569 * Try to get an active super block reference to prevent racing with
570 * unmount (see super_trylock_shared()). But note that unmount isn't
571 * the only place where a write lock on s_umount is taken, and we can
572 * fail here because of things like remount as well.
573 */
574 if (down_read_trylock(&sb->s_umount)) {
575 atomic_inc(&sb->s_active);
576 up_read(&sb->s_umount);
577 if (!queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work))
578 deactivate_super(sb);
579 }
580 }
581
582 /**
583 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
584 * @gl: the glock
585 */
586 static int freeze_go_xmote_bh(struct gfs2_glock *gl)
587 {
588 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
589 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
590 struct gfs2_glock *j_gl = ip->i_gl;
591 struct gfs2_log_header_host head;
592 int error;
593
594 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
595 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
596
597 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
598 if (gfs2_assert_withdraw_delayed(sdp, !error))
599 return error;
600 if (gfs2_assert_withdraw_delayed(sdp, head.lh_flags &
601 GFS2_LOG_HEAD_UNMOUNT))
602 return -EIO;
603 sdp->sd_log_sequence = head.lh_sequence + 1;
604 gfs2_log_pointers_init(sdp, head.lh_blkno);
605 }
606 return 0;
607 }
608
609 /**
610 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
611 * @gl: the glock
612 * @remote: true if this came from a different cluster node
613 *
614 * gl_lockref.lock lock is held while calling this
615 */
616 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
617 {
618 struct gfs2_inode *ip = gl->gl_object;
619 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
620
621 if (!remote || sb_rdonly(sdp->sd_vfs) ||
622 test_bit(SDF_KILL, &sdp->sd_flags))
623 return;
624
625 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
626 gl->gl_state == LM_ST_SHARED && ip) {
627 gl->gl_lockref.count++;
628 if (!gfs2_queue_try_to_evict(gl))
629 gl->gl_lockref.count--;
630 }
631 }
632
633 /**
634 * inode_go_unlocked - wake up anyone waiting for dlm's unlock ast
635 * @gl: glock being unlocked
636 *
637 * For now, this is only used for the journal inode glock. In withdraw
638 * situations, we need to wait for the glock to be unlocked so that we know
639 * other nodes may proceed with recovery / journal replay.
640 */
641 static void inode_go_unlocked(struct gfs2_glock *gl)
642 {
643 /* Note that we cannot reference gl_object because it's already set
644 * to NULL by this point in its lifecycle. */
645 if (!test_bit(GLF_UNLOCKED, &gl->gl_flags))
646 return;
647 clear_bit_unlock(GLF_UNLOCKED, &gl->gl_flags);
648 wake_up_bit(&gl->gl_flags, GLF_UNLOCKED);
649 }
650
651 /**
652 * nondisk_go_callback - used to signal when a node did a withdraw
653 * @gl: the nondisk glock
654 * @remote: true if this came from a different cluster node
655 *
656 */
657 static void nondisk_go_callback(struct gfs2_glock *gl, bool remote)
658 {
659 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
660
661 /* Ignore the callback unless it's from another node, and it's the
662 live lock. */
663 if (!remote || gl->gl_name.ln_number != GFS2_LIVE_LOCK)
664 return;
665
666 /* First order of business is to cancel the demote request. We don't
667 * really want to demote a nondisk glock. At best it's just to inform
668 * us of another node's withdraw. We'll keep it in SH mode. */
669 clear_bit(GLF_DEMOTE, &gl->gl_flags);
670 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
671
672 /* Ignore the unlock if we're withdrawn, unmounting, or in recovery. */
673 if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) ||
674 test_bit(SDF_WITHDRAWN, &sdp->sd_flags) ||
675 test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags))
676 return;
677
678 /* We only care when a node wants us to unlock, because that means
679 * they want a journal recovered. */
680 if (gl->gl_demote_state != LM_ST_UNLOCKED)
681 return;
682
683 if (sdp->sd_args.ar_spectator) {
684 fs_warn(sdp, "Spectator node cannot recover journals.\n");
685 return;
686 }
687
688 fs_warn(sdp, "Some node has withdrawn; checking for recovery.\n");
689 set_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
690 /*
691 * We can't call remote_withdraw directly here or gfs2_recover_journal
692 * because this is called from the glock unlock function and the
693 * remote_withdraw needs to enqueue and dequeue the same "live" glock
694 * we were called from. So we queue it to the control work queue in
695 * lock_dlm.
696 */
697 queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
698 }
699
700 const struct gfs2_glock_operations gfs2_meta_glops = {
701 .go_type = LM_TYPE_META,
702 .go_flags = GLOF_NONDISK,
703 };
704
705 const struct gfs2_glock_operations gfs2_inode_glops = {
706 .go_sync = inode_go_sync,
707 .go_inval = inode_go_inval,
708 .go_instantiate = inode_go_instantiate,
709 .go_held = inode_go_held,
710 .go_dump = inode_go_dump,
711 .go_type = LM_TYPE_INODE,
712 .go_flags = GLOF_ASPACE | GLOF_LVB,
713 .go_unlocked = inode_go_unlocked,
714 };
715
716 const struct gfs2_glock_operations gfs2_rgrp_glops = {
717 .go_sync = rgrp_go_sync,
718 .go_inval = rgrp_go_inval,
719 .go_instantiate = gfs2_rgrp_go_instantiate,
720 .go_dump = gfs2_rgrp_go_dump,
721 .go_type = LM_TYPE_RGRP,
722 .go_flags = GLOF_LVB,
723 };
724
725 const struct gfs2_glock_operations gfs2_freeze_glops = {
726 .go_xmote_bh = freeze_go_xmote_bh,
727 .go_callback = freeze_go_callback,
728 .go_type = LM_TYPE_NONDISK,
729 .go_flags = GLOF_NONDISK,
730 };
731
732 const struct gfs2_glock_operations gfs2_iopen_glops = {
733 .go_type = LM_TYPE_IOPEN,
734 .go_callback = iopen_go_callback,
735 .go_dump = inode_go_dump,
736 .go_flags = GLOF_NONDISK,
737 .go_subclass = 1,
738 };
739
740 const struct gfs2_glock_operations gfs2_flock_glops = {
741 .go_type = LM_TYPE_FLOCK,
742 .go_flags = GLOF_NONDISK,
743 };
744
745 const struct gfs2_glock_operations gfs2_nondisk_glops = {
746 .go_type = LM_TYPE_NONDISK,
747 .go_flags = GLOF_NONDISK,
748 .go_callback = nondisk_go_callback,
749 };
750
751 const struct gfs2_glock_operations gfs2_quota_glops = {
752 .go_type = LM_TYPE_QUOTA,
753 .go_flags = GLOF_LVB | GLOF_NONDISK,
754 };
755
756 const struct gfs2_glock_operations gfs2_journal_glops = {
757 .go_type = LM_TYPE_JOURNAL,
758 .go_flags = GLOF_NONDISK,
759 };
760
761 const struct gfs2_glock_operations *gfs2_glops_list[] = {
762 [LM_TYPE_META] = &gfs2_meta_glops,
763 [LM_TYPE_INODE] = &gfs2_inode_glops,
764 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
765 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
766 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
767 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
768 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
769 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
770 };
771
772
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.