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
gfs2_ail_error(struct gfs2_glock * gl,const struct buffer_head * bh)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
__gfs2_ail_flush(struct gfs2_glock * gl,bool fsync,unsigned int nr_revokes)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
87
gfs2_ail_empty_gl(struct gfs2_glock * gl)88 static int gfs2_ail_empty_gl(struct gfs2_glock *gl)
89 {
90 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
91 struct gfs2_trans tr;
92 unsigned int revokes;
93 int ret = 0;
94
95 revokes = atomic_read(&gl->gl_ail_count);
96
97 if (!revokes) {
98 bool have_revokes;
99 bool log_in_flight;
100
101 /*
102 * We have nothing on the ail, but there could be revokes on
103 * the sdp revoke queue, in which case, we still want to flush
104 * the log and wait for it to finish.
105 *
106 * If the sdp revoke list is empty too, we might still have an
107 * io outstanding for writing revokes, so we should wait for
108 * it before returning.
109 *
110 * If none of these conditions are true, our revokes are all
111 * flushed and we can return.
112 */
113 gfs2_log_lock(sdp);
114 have_revokes = !list_empty(&sdp->sd_log_revokes);
115 log_in_flight = atomic_read(&sdp->sd_log_in_flight);
116 gfs2_log_unlock(sdp);
117 if (have_revokes)
118 goto flush;
119 if (log_in_flight)
120 log_flush_wait(sdp);
121 return 0;
122 }
123
124 memset(&tr, 0, sizeof(tr));
125 set_bit(TR_ONSTACK, &tr.tr_flags);
126 ret = __gfs2_trans_begin(&tr, sdp, 0, revokes, _RET_IP_);
127 if (ret) {
128 fs_err(sdp, "Transaction error %d: Unable to write revokes.", ret);
129 goto flush;
130 }
131 __gfs2_ail_flush(gl, 0, revokes);
132 gfs2_trans_end(sdp);
133
134 flush:
135 if (!ret)
136 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
137 GFS2_LFC_AIL_EMPTY_GL);
138 return ret;
139 }
140
gfs2_ail_flush(struct gfs2_glock * gl,bool fsync)141 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
142 {
143 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
144 unsigned int revokes = atomic_read(&gl->gl_ail_count);
145 int ret;
146
147 if (!revokes)
148 return;
149
150 ret = gfs2_trans_begin(sdp, 0, revokes);
151 if (ret)
152 return;
153 __gfs2_ail_flush(gl, fsync, revokes);
154 gfs2_trans_end(sdp);
155 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
156 GFS2_LFC_AIL_FLUSH);
157 }
158
159 /**
160 * gfs2_rgrp_metasync - sync out the metadata of a resource group
161 * @gl: the glock protecting the resource group
162 *
163 */
164
gfs2_rgrp_metasync(struct gfs2_glock * gl)165 static int gfs2_rgrp_metasync(struct gfs2_glock *gl)
166 {
167 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
168 struct address_space *metamapping = &sdp->sd_aspace;
169 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
170 const unsigned bsize = sdp->sd_sb.sb_bsize;
171 loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
172 loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
173 int error;
174
175 filemap_fdatawrite_range(metamapping, start, end);
176 error = filemap_fdatawait_range(metamapping, start, end);
177 WARN_ON_ONCE(error && !gfs2_withdrawn(sdp));
178 mapping_set_error(metamapping, error);
179 if (error)
180 gfs2_io_error(sdp);
181 return error;
182 }
183
184 /**
185 * rgrp_go_sync - sync out the metadata for this glock
186 * @gl: the glock
187 *
188 * Called when demoting or unlocking an EX glock. We must flush
189 * to disk all dirty buffers/pages relating to this glock, and must not
190 * return to caller to demote/unlock the glock until I/O is complete.
191 */
192
rgrp_go_sync(struct gfs2_glock * gl)193 static int rgrp_go_sync(struct gfs2_glock *gl)
194 {
195 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
196 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
197 int error;
198
199 if (!rgd || !test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
200 return 0;
201 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
202
203 gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
204 GFS2_LFC_RGRP_GO_SYNC);
205 error = gfs2_rgrp_metasync(gl);
206 if (!error)
207 error = gfs2_ail_empty_gl(gl);
208 gfs2_free_clones(rgd);
209 return error;
210 }
211
212 /**
213 * rgrp_go_inval - invalidate the metadata for this glock
214 * @gl: the glock
215 * @flags:
216 *
217 * We never used LM_ST_DEFERRED with resource groups, so that we
218 * should always see the metadata flag set here.
219 *
220 */
221
rgrp_go_inval(struct gfs2_glock * gl,int flags)222 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
223 {
224 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
225 struct address_space *mapping = &sdp->sd_aspace;
226 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
227 const unsigned bsize = sdp->sd_sb.sb_bsize;
228 loff_t start, end;
229
230 if (!rgd)
231 return;
232 start = (rgd->rd_addr * bsize) & PAGE_MASK;
233 end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
234 gfs2_rgrp_brelse(rgd);
235 WARN_ON_ONCE(!(flags & DIO_METADATA));
236 truncate_inode_pages_range(mapping, start, end);
237 }
238
gfs2_rgrp_go_dump(struct seq_file * seq,const struct gfs2_glock * gl,const char * fs_id_buf)239 static void gfs2_rgrp_go_dump(struct seq_file *seq, const struct gfs2_glock *gl,
240 const char *fs_id_buf)
241 {
242 struct gfs2_rgrpd *rgd = gl->gl_object;
243
244 if (rgd)
245 gfs2_rgrp_dump(seq, rgd, fs_id_buf);
246 }
247
gfs2_glock2inode(struct gfs2_glock * gl)248 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
249 {
250 struct gfs2_inode *ip;
251
252 spin_lock(&gl->gl_lockref.lock);
253 ip = gl->gl_object;
254 if (ip)
255 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
256 spin_unlock(&gl->gl_lockref.lock);
257 return ip;
258 }
259
gfs2_glock2rgrp(struct gfs2_glock * gl)260 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
261 {
262 struct gfs2_rgrpd *rgd;
263
264 spin_lock(&gl->gl_lockref.lock);
265 rgd = gl->gl_object;
266 spin_unlock(&gl->gl_lockref.lock);
267
268 return rgd;
269 }
270
gfs2_clear_glop_pending(struct gfs2_inode * ip)271 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
272 {
273 if (!ip)
274 return;
275
276 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
277 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
278 }
279
280 /**
281 * gfs2_inode_metasync - sync out the metadata of an inode
282 * @gl: the glock protecting the inode
283 *
284 */
gfs2_inode_metasync(struct gfs2_glock * gl)285 int gfs2_inode_metasync(struct gfs2_glock *gl)
286 {
287 struct address_space *metamapping = gfs2_glock2aspace(gl);
288 int error;
289
290 filemap_fdatawrite(metamapping);
291 error = filemap_fdatawait(metamapping);
292 if (error)
293 gfs2_io_error(gl->gl_name.ln_sbd);
294 return error;
295 }
296
297 /**
298 * inode_go_sync - Sync the dirty metadata of an inode
299 * @gl: the glock protecting the inode
300 *
301 */
302
inode_go_sync(struct gfs2_glock * gl)303 static int inode_go_sync(struct gfs2_glock *gl)
304 {
305 struct gfs2_inode *ip = gfs2_glock2inode(gl);
306 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
307 struct address_space *metamapping = gfs2_glock2aspace(gl);
308 int error = 0, ret;
309
310 if (isreg) {
311 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
312 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
313 inode_dio_wait(&ip->i_inode);
314 }
315 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
316 goto out;
317
318 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
319
320 gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
321 GFS2_LFC_INODE_GO_SYNC);
322 filemap_fdatawrite(metamapping);
323 if (isreg) {
324 struct address_space *mapping = ip->i_inode.i_mapping;
325 filemap_fdatawrite(mapping);
326 error = filemap_fdatawait(mapping);
327 mapping_set_error(mapping, error);
328 }
329 ret = gfs2_inode_metasync(gl);
330 if (!error)
331 error = ret;
332 ret = gfs2_ail_empty_gl(gl);
333 if (!error)
334 error = ret;
335 /*
336 * Writeback of the data mapping may cause the dirty flag to be set
337 * so we have to clear it again here.
338 */
339 smp_mb__before_atomic();
340 clear_bit(GLF_DIRTY, &gl->gl_flags);
341
342 out:
343 gfs2_clear_glop_pending(ip);
344 return error;
345 }
346
347 /**
348 * inode_go_inval - prepare a inode glock to be released
349 * @gl: the glock
350 * @flags:
351 *
352 * Normally we invalidate everything, but if we are moving into
353 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
354 * can keep hold of the metadata, since it won't have changed.
355 *
356 */
357
inode_go_inval(struct gfs2_glock * gl,int flags)358 static void inode_go_inval(struct gfs2_glock *gl, int flags)
359 {
360 struct gfs2_inode *ip = gfs2_glock2inode(gl);
361
362 if (flags & DIO_METADATA) {
363 struct address_space *mapping = gfs2_glock2aspace(gl);
364 truncate_inode_pages(mapping, 0);
365 if (ip) {
366 set_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
367 forget_all_cached_acls(&ip->i_inode);
368 security_inode_invalidate_secctx(&ip->i_inode);
369 gfs2_dir_hash_inval(ip);
370 }
371 }
372
373 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
374 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
375 GFS2_LOG_HEAD_FLUSH_NORMAL |
376 GFS2_LFC_INODE_GO_INVAL);
377 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
378 }
379 if (ip && S_ISREG(ip->i_inode.i_mode))
380 truncate_inode_pages(ip->i_inode.i_mapping, 0);
381
382 gfs2_clear_glop_pending(ip);
383 }
384
385 /**
386 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
387 * @gl: the glock
388 *
389 * Returns: 1 if it's ok
390 */
391
inode_go_demote_ok(const struct gfs2_glock * gl)392 static int inode_go_demote_ok(const struct gfs2_glock *gl)
393 {
394 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
395
396 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
397 return 0;
398
399 return 1;
400 }
401
gfs2_dinode_in(struct gfs2_inode * ip,const void * buf)402 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
403 {
404 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
405 const struct gfs2_dinode *str = buf;
406 struct timespec64 atime;
407 u16 height, depth;
408 umode_t mode = be32_to_cpu(str->di_mode);
409 struct inode *inode = &ip->i_inode;
410 bool is_new = inode->i_state & I_NEW;
411
412 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
413 goto corrupt;
414 if (unlikely(!is_new && inode_wrong_type(inode, mode)))
415 goto corrupt;
416 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
417 inode->i_mode = mode;
418 if (is_new) {
419 inode->i_rdev = 0;
420 switch (mode & S_IFMT) {
421 case S_IFBLK:
422 case S_IFCHR:
423 inode->i_rdev = MKDEV(be32_to_cpu(str->di_major),
424 be32_to_cpu(str->di_minor));
425 break;
426 }
427 }
428
429 i_uid_write(inode, be32_to_cpu(str->di_uid));
430 i_gid_write(inode, be32_to_cpu(str->di_gid));
431 set_nlink(inode, be32_to_cpu(str->di_nlink));
432 i_size_write(inode, be64_to_cpu(str->di_size));
433 gfs2_set_inode_blocks(inode, be64_to_cpu(str->di_blocks));
434 atime.tv_sec = be64_to_cpu(str->di_atime);
435 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
436 if (timespec64_compare(&inode->i_atime, &atime) < 0)
437 inode->i_atime = atime;
438 inode->i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
439 inode->i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
440 inode_set_ctime(inode, be64_to_cpu(str->di_ctime),
441 be32_to_cpu(str->di_ctime_nsec));
442
443 ip->i_goal = be64_to_cpu(str->di_goal_meta);
444 ip->i_generation = be64_to_cpu(str->di_generation);
445
446 ip->i_diskflags = be32_to_cpu(str->di_flags);
447 ip->i_eattr = be64_to_cpu(str->di_eattr);
448 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
449 gfs2_set_inode_flags(inode);
450 height = be16_to_cpu(str->di_height);
451 if (unlikely(height > sdp->sd_max_height))
452 goto corrupt;
453 ip->i_height = (u8)height;
454
455 depth = be16_to_cpu(str->di_depth);
456 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
457 goto corrupt;
458 ip->i_depth = (u8)depth;
459 ip->i_entries = be32_to_cpu(str->di_entries);
460
461 if (gfs2_is_stuffed(ip) && inode->i_size > gfs2_max_stuffed_size(ip))
462 goto corrupt;
463
464 if (S_ISREG(inode->i_mode))
465 gfs2_set_aops(inode);
466
467 return 0;
468 corrupt:
469 gfs2_consist_inode(ip);
470 return -EIO;
471 }
472
473 /**
474 * gfs2_inode_refresh - Refresh the incore copy of the dinode
475 * @ip: The GFS2 inode
476 *
477 * Returns: errno
478 */
479
gfs2_inode_refresh(struct gfs2_inode * ip)480 int gfs2_inode_refresh(struct gfs2_inode *ip)
481 {
482 struct buffer_head *dibh;
483 int error;
484
485 error = gfs2_meta_inode_buffer(ip, &dibh);
486 if (error)
487 return error;
488
489 error = gfs2_dinode_in(ip, dibh->b_data);
490 brelse(dibh);
491 return error;
492 }
493
494 /**
495 * inode_go_instantiate - read in an inode if necessary
496 * @gh: The glock holder
497 *
498 * Returns: errno
499 */
500
inode_go_instantiate(struct gfs2_glock * gl)501 static int inode_go_instantiate(struct gfs2_glock *gl)
502 {
503 struct gfs2_inode *ip = gl->gl_object;
504
505 if (!ip) /* no inode to populate - read it in later */
506 return 0;
507
508 return gfs2_inode_refresh(ip);
509 }
510
inode_go_held(struct gfs2_holder * gh)511 static int inode_go_held(struct gfs2_holder *gh)
512 {
513 struct gfs2_glock *gl = gh->gh_gl;
514 struct gfs2_inode *ip = gl->gl_object;
515 int error = 0;
516
517 if (!ip) /* no inode to populate - read it in later */
518 return 0;
519
520 if (gh->gh_state != LM_ST_DEFERRED)
521 inode_dio_wait(&ip->i_inode);
522
523 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
524 (gl->gl_state == LM_ST_EXCLUSIVE) &&
525 (gh->gh_state == LM_ST_EXCLUSIVE))
526 error = gfs2_truncatei_resume(ip);
527
528 return error;
529 }
530
531 /**
532 * inode_go_dump - print information about an inode
533 * @seq: The iterator
534 * @gl: The glock
535 * @fs_id_buf: file system id (may be empty)
536 *
537 */
538
inode_go_dump(struct seq_file * seq,const struct gfs2_glock * gl,const char * fs_id_buf)539 static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl,
540 const char *fs_id_buf)
541 {
542 struct gfs2_inode *ip = gl->gl_object;
543 const struct inode *inode = &ip->i_inode;
544
545 if (ip == NULL)
546 return;
547
548 gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
549 "p:%lu\n", fs_id_buf,
550 (unsigned long long)ip->i_no_formal_ino,
551 (unsigned long long)ip->i_no_addr,
552 IF2DT(inode->i_mode), ip->i_flags,
553 (unsigned int)ip->i_diskflags,
554 (unsigned long long)i_size_read(inode),
555 inode->i_data.nrpages);
556 }
557
558 /**
559 * freeze_go_callback - A cluster node is requesting a freeze
560 * @gl: the glock
561 * @remote: true if this came from a different cluster node
562 */
563
freeze_go_callback(struct gfs2_glock * gl,bool remote)564 static void freeze_go_callback(struct gfs2_glock *gl, bool remote)
565 {
566 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
567 struct super_block *sb = sdp->sd_vfs;
568
569 if (!remote ||
570 (gl->gl_state != LM_ST_SHARED &&
571 gl->gl_state != LM_ST_UNLOCKED) ||
572 gl->gl_demote_state != LM_ST_UNLOCKED)
573 return;
574
575 /*
576 * Try to get an active super block reference to prevent racing with
577 * unmount (see super_trylock_shared()). But note that unmount isn't
578 * the only place where a write lock on s_umount is taken, and we can
579 * fail here because of things like remount as well.
580 */
581 if (down_read_trylock(&sb->s_umount)) {
582 atomic_inc(&sb->s_active);
583 up_read(&sb->s_umount);
584 if (!queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work))
585 deactivate_super(sb);
586 }
587 }
588
589 /**
590 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
591 * @gl: the glock
592 */
freeze_go_xmote_bh(struct gfs2_glock * gl)593 static int freeze_go_xmote_bh(struct gfs2_glock *gl)
594 {
595 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
596 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
597 struct gfs2_glock *j_gl = ip->i_gl;
598 struct gfs2_log_header_host head;
599 int error;
600
601 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
602 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
603
604 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
605 if (gfs2_assert_withdraw_delayed(sdp, !error))
606 return error;
607 if (gfs2_assert_withdraw_delayed(sdp, head.lh_flags &
608 GFS2_LOG_HEAD_UNMOUNT))
609 return -EIO;
610 sdp->sd_log_sequence = head.lh_sequence + 1;
611 gfs2_log_pointers_init(sdp, head.lh_blkno);
612 }
613 return 0;
614 }
615
616 /**
617 * freeze_go_demote_ok
618 * @gl: the glock
619 *
620 * Always returns 0
621 */
622
freeze_go_demote_ok(const struct gfs2_glock * gl)623 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
624 {
625 return 0;
626 }
627
628 /**
629 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
630 * @gl: the glock
631 * @remote: true if this came from a different cluster node
632 *
633 * gl_lockref.lock lock is held while calling this
634 */
iopen_go_callback(struct gfs2_glock * gl,bool remote)635 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
636 {
637 struct gfs2_inode *ip = gl->gl_object;
638 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
639
640 if (!remote || sb_rdonly(sdp->sd_vfs) ||
641 test_bit(SDF_KILL, &sdp->sd_flags))
642 return;
643
644 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
645 gl->gl_state == LM_ST_SHARED && ip) {
646 gl->gl_lockref.count++;
647 if (!gfs2_queue_try_to_evict(gl))
648 gl->gl_lockref.count--;
649 }
650 }
651
652 /**
653 * inode_go_free - wake up anyone waiting for dlm's unlock ast to free it
654 * @gl: glock being freed
655 *
656 * For now, this is only used for the journal inode glock. In withdraw
657 * situations, we need to wait for the glock to be freed so that we know
658 * other nodes may proceed with recovery / journal replay.
659 */
inode_go_free(struct gfs2_glock * gl)660 static void inode_go_free(struct gfs2_glock *gl)
661 {
662 /* Note that we cannot reference gl_object because it's already set
663 * to NULL by this point in its lifecycle. */
664 if (!test_bit(GLF_FREEING, &gl->gl_flags))
665 return;
666 clear_bit_unlock(GLF_FREEING, &gl->gl_flags);
667 wake_up_bit(&gl->gl_flags, GLF_FREEING);
668 }
669
670 /**
671 * nondisk_go_callback - used to signal when a node did a withdraw
672 * @gl: the nondisk glock
673 * @remote: true if this came from a different cluster node
674 *
675 */
nondisk_go_callback(struct gfs2_glock * gl,bool remote)676 static void nondisk_go_callback(struct gfs2_glock *gl, bool remote)
677 {
678 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
679
680 /* Ignore the callback unless it's from another node, and it's the
681 live lock. */
682 if (!remote || gl->gl_name.ln_number != GFS2_LIVE_LOCK)
683 return;
684
685 /* First order of business is to cancel the demote request. We don't
686 * really want to demote a nondisk glock. At best it's just to inform
687 * us of another node's withdraw. We'll keep it in SH mode. */
688 clear_bit(GLF_DEMOTE, &gl->gl_flags);
689 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
690
691 /* Ignore the unlock if we're withdrawn, unmounting, or in recovery. */
692 if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) ||
693 test_bit(SDF_WITHDRAWN, &sdp->sd_flags) ||
694 test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags))
695 return;
696
697 /* We only care when a node wants us to unlock, because that means
698 * they want a journal recovered. */
699 if (gl->gl_demote_state != LM_ST_UNLOCKED)
700 return;
701
702 if (sdp->sd_args.ar_spectator) {
703 fs_warn(sdp, "Spectator node cannot recover journals.\n");
704 return;
705 }
706
707 fs_warn(sdp, "Some node has withdrawn; checking for recovery.\n");
708 set_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
709 /*
710 * We can't call remote_withdraw directly here or gfs2_recover_journal
711 * because this is called from the glock unlock function and the
712 * remote_withdraw needs to enqueue and dequeue the same "live" glock
713 * we were called from. So we queue it to the control work queue in
714 * lock_dlm.
715 */
716 queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
717 }
718
719 const struct gfs2_glock_operations gfs2_meta_glops = {
720 .go_type = LM_TYPE_META,
721 .go_flags = GLOF_NONDISK,
722 };
723
724 const struct gfs2_glock_operations gfs2_inode_glops = {
725 .go_sync = inode_go_sync,
726 .go_inval = inode_go_inval,
727 .go_demote_ok = inode_go_demote_ok,
728 .go_instantiate = inode_go_instantiate,
729 .go_held = inode_go_held,
730 .go_dump = inode_go_dump,
731 .go_type = LM_TYPE_INODE,
732 .go_flags = GLOF_ASPACE | GLOF_LRU | GLOF_LVB,
733 .go_free = inode_go_free,
734 };
735
736 const struct gfs2_glock_operations gfs2_rgrp_glops = {
737 .go_sync = rgrp_go_sync,
738 .go_inval = rgrp_go_inval,
739 .go_instantiate = gfs2_rgrp_go_instantiate,
740 .go_dump = gfs2_rgrp_go_dump,
741 .go_type = LM_TYPE_RGRP,
742 .go_flags = GLOF_LVB,
743 };
744
745 const struct gfs2_glock_operations gfs2_freeze_glops = {
746 .go_xmote_bh = freeze_go_xmote_bh,
747 .go_demote_ok = freeze_go_demote_ok,
748 .go_callback = freeze_go_callback,
749 .go_type = LM_TYPE_NONDISK,
750 .go_flags = GLOF_NONDISK,
751 };
752
753 const struct gfs2_glock_operations gfs2_iopen_glops = {
754 .go_type = LM_TYPE_IOPEN,
755 .go_callback = iopen_go_callback,
756 .go_dump = inode_go_dump,
757 .go_flags = GLOF_LRU | GLOF_NONDISK,
758 .go_subclass = 1,
759 };
760
761 const struct gfs2_glock_operations gfs2_flock_glops = {
762 .go_type = LM_TYPE_FLOCK,
763 .go_flags = GLOF_LRU | GLOF_NONDISK,
764 };
765
766 const struct gfs2_glock_operations gfs2_nondisk_glops = {
767 .go_type = LM_TYPE_NONDISK,
768 .go_flags = GLOF_NONDISK,
769 .go_callback = nondisk_go_callback,
770 };
771
772 const struct gfs2_glock_operations gfs2_quota_glops = {
773 .go_type = LM_TYPE_QUOTA,
774 .go_flags = GLOF_LVB | GLOF_LRU | GLOF_NONDISK,
775 };
776
777 const struct gfs2_glock_operations gfs2_journal_glops = {
778 .go_type = LM_TYPE_JOURNAL,
779 .go_flags = GLOF_NONDISK,
780 };
781
782 const struct gfs2_glock_operations *gfs2_glops_list[] = {
783 [LM_TYPE_META] = &gfs2_meta_glops,
784 [LM_TYPE_INODE] = &gfs2_inode_glops,
785 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
786 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
787 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
788 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
789 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
790 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
791 };
792
793