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 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(gl->gl_name.ln_sbd, "AIL error\n");
45 gfs2_withdraw(gl->gl_name.ln_sbd);
46 }
47
48 /**
49 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
50 * @gl: the glock
51 * @fsync: set when called from fsync (not all buffers will be clean)
52 * @nr_revokes: Number of buffers to revoke
53 *
54 * None of the buffers should be dirty, locked, or pinned.
55 */
56
__gfs2_ail_flush(struct gfs2_glock * gl,bool fsync,unsigned int nr_revokes)57 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
58 unsigned int nr_revokes)
59 {
60 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
61 struct list_head *head = &gl->gl_ail_list;
62 struct gfs2_bufdata *bd, *tmp;
63 struct buffer_head *bh;
64 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
65
66 gfs2_log_lock(sdp);
67 spin_lock(&sdp->sd_ail_lock);
68 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
69 if (nr_revokes == 0)
70 break;
71 bh = bd->bd_bh;
72 if (bh->b_state & b_state) {
73 if (fsync)
74 continue;
75 gfs2_ail_error(gl, bh);
76 }
77 gfs2_trans_add_revoke(sdp, bd);
78 nr_revokes--;
79 }
80 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
81 spin_unlock(&sdp->sd_ail_lock);
82 gfs2_log_unlock(sdp);
83 }
84
85
gfs2_ail_empty_gl(struct gfs2_glock * gl)86 static int gfs2_ail_empty_gl(struct gfs2_glock *gl)
87 {
88 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
89 struct gfs2_trans tr;
90 unsigned int revokes;
91 int ret;
92
93 revokes = atomic_read(&gl->gl_ail_count);
94
95 if (!revokes) {
96 bool have_revokes;
97 bool log_in_flight;
98
99 /*
100 * We have nothing on the ail, but there could be revokes on
101 * the sdp revoke queue, in which case, we still want to flush
102 * the log and wait for it to finish.
103 *
104 * If the sdp revoke list is empty too, we might still have an
105 * io outstanding for writing revokes, so we should wait for
106 * it before returning.
107 *
108 * If none of these conditions are true, our revokes are all
109 * flushed and we can return.
110 */
111 gfs2_log_lock(sdp);
112 have_revokes = !list_empty(&sdp->sd_log_revokes);
113 log_in_flight = atomic_read(&sdp->sd_log_in_flight);
114 gfs2_log_unlock(sdp);
115 if (have_revokes)
116 goto flush;
117 if (log_in_flight)
118 log_flush_wait(sdp);
119 return 0;
120 }
121
122 memset(&tr, 0, sizeof(tr));
123 set_bit(TR_ONSTACK, &tr.tr_flags);
124 ret = __gfs2_trans_begin(&tr, sdp, 0, revokes, _RET_IP_);
125 if (ret)
126 goto flush;
127 __gfs2_ail_flush(gl, 0, revokes);
128 gfs2_trans_end(sdp);
129
130 flush:
131 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
132 GFS2_LFC_AIL_EMPTY_GL);
133 return 0;
134 }
135
gfs2_ail_flush(struct gfs2_glock * gl,bool fsync)136 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
137 {
138 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
139 unsigned int revokes = atomic_read(&gl->gl_ail_count);
140 int ret;
141
142 if (!revokes)
143 return;
144
145 ret = gfs2_trans_begin(sdp, 0, revokes);
146 if (ret)
147 return;
148 __gfs2_ail_flush(gl, fsync, revokes);
149 gfs2_trans_end(sdp);
150 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
151 GFS2_LFC_AIL_FLUSH);
152 }
153
154 /**
155 * gfs2_rgrp_metasync - sync out the metadata of a resource group
156 * @gl: the glock protecting the resource group
157 *
158 */
159
gfs2_rgrp_metasync(struct gfs2_glock * gl)160 static int gfs2_rgrp_metasync(struct gfs2_glock *gl)
161 {
162 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
163 struct address_space *metamapping = &sdp->sd_aspace;
164 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
165 const unsigned bsize = sdp->sd_sb.sb_bsize;
166 loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
167 loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
168 int error;
169
170 filemap_fdatawrite_range(metamapping, start, end);
171 error = filemap_fdatawait_range(metamapping, start, end);
172 WARN_ON_ONCE(error && !gfs2_withdrawn(sdp));
173 mapping_set_error(metamapping, error);
174 if (error)
175 gfs2_io_error(sdp);
176 return error;
177 }
178
179 /**
180 * rgrp_go_sync - sync out the metadata for this glock
181 * @gl: the glock
182 *
183 * Called when demoting or unlocking an EX glock. We must flush
184 * to disk all dirty buffers/pages relating to this glock, and must not
185 * return to caller to demote/unlock the glock until I/O is complete.
186 */
187
rgrp_go_sync(struct gfs2_glock * gl)188 static int rgrp_go_sync(struct gfs2_glock *gl)
189 {
190 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
191 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
192 int error;
193
194 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
195 return 0;
196 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
197
198 gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
199 GFS2_LFC_RGRP_GO_SYNC);
200 error = gfs2_rgrp_metasync(gl);
201 if (!error)
202 error = gfs2_ail_empty_gl(gl);
203 gfs2_free_clones(rgd);
204 return error;
205 }
206
207 /**
208 * rgrp_go_inval - invalidate the metadata for this glock
209 * @gl: the glock
210 * @flags:
211 *
212 * We never used LM_ST_DEFERRED with resource groups, so that we
213 * should always see the metadata flag set here.
214 *
215 */
216
rgrp_go_inval(struct gfs2_glock * gl,int flags)217 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
218 {
219 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
220 struct address_space *mapping = &sdp->sd_aspace;
221 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
222 const unsigned bsize = sdp->sd_sb.sb_bsize;
223 loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
224 loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
225
226 gfs2_rgrp_brelse(rgd);
227 WARN_ON_ONCE(!(flags & DIO_METADATA));
228 truncate_inode_pages_range(mapping, start, end);
229 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
230 }
231
gfs2_rgrp_go_dump(struct seq_file * seq,struct gfs2_glock * gl,const char * fs_id_buf)232 static void gfs2_rgrp_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
233 const char *fs_id_buf)
234 {
235 struct gfs2_rgrpd *rgd = gl->gl_object;
236
237 if (rgd)
238 gfs2_rgrp_dump(seq, rgd, fs_id_buf);
239 }
240
gfs2_glock2inode(struct gfs2_glock * gl)241 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
242 {
243 struct gfs2_inode *ip;
244
245 spin_lock(&gl->gl_lockref.lock);
246 ip = gl->gl_object;
247 if (ip)
248 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
249 spin_unlock(&gl->gl_lockref.lock);
250 return ip;
251 }
252
gfs2_glock2rgrp(struct gfs2_glock * gl)253 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
254 {
255 struct gfs2_rgrpd *rgd;
256
257 spin_lock(&gl->gl_lockref.lock);
258 rgd = gl->gl_object;
259 spin_unlock(&gl->gl_lockref.lock);
260
261 return rgd;
262 }
263
gfs2_clear_glop_pending(struct gfs2_inode * ip)264 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
265 {
266 if (!ip)
267 return;
268
269 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
270 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
271 }
272
273 /**
274 * gfs2_inode_metasync - sync out the metadata of an inode
275 * @gl: the glock protecting the inode
276 *
277 */
gfs2_inode_metasync(struct gfs2_glock * gl)278 int gfs2_inode_metasync(struct gfs2_glock *gl)
279 {
280 struct address_space *metamapping = gfs2_glock2aspace(gl);
281 int error;
282
283 filemap_fdatawrite(metamapping);
284 error = filemap_fdatawait(metamapping);
285 if (error)
286 gfs2_io_error(gl->gl_name.ln_sbd);
287 return error;
288 }
289
290 /**
291 * inode_go_sync - Sync the dirty metadata of an inode
292 * @gl: the glock protecting the inode
293 *
294 */
295
inode_go_sync(struct gfs2_glock * gl)296 static int inode_go_sync(struct gfs2_glock *gl)
297 {
298 struct gfs2_inode *ip = gfs2_glock2inode(gl);
299 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
300 struct address_space *metamapping = gfs2_glock2aspace(gl);
301 int error = 0, ret;
302
303 if (isreg) {
304 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
305 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
306 inode_dio_wait(&ip->i_inode);
307 }
308 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
309 goto out;
310
311 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
312
313 gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
314 GFS2_LFC_INODE_GO_SYNC);
315 filemap_fdatawrite(metamapping);
316 if (isreg) {
317 struct address_space *mapping = ip->i_inode.i_mapping;
318 filemap_fdatawrite(mapping);
319 error = filemap_fdatawait(mapping);
320 mapping_set_error(mapping, error);
321 }
322 ret = gfs2_inode_metasync(gl);
323 if (!error)
324 error = ret;
325 gfs2_ail_empty_gl(gl);
326 /*
327 * Writeback of the data mapping may cause the dirty flag to be set
328 * so we have to clear it again here.
329 */
330 smp_mb__before_atomic();
331 clear_bit(GLF_DIRTY, &gl->gl_flags);
332
333 out:
334 gfs2_clear_glop_pending(ip);
335 return error;
336 }
337
338 /**
339 * inode_go_inval - prepare a inode glock to be released
340 * @gl: the glock
341 * @flags:
342 *
343 * Normally we invalidate everything, but if we are moving into
344 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
345 * can keep hold of the metadata, since it won't have changed.
346 *
347 */
348
inode_go_inval(struct gfs2_glock * gl,int flags)349 static void inode_go_inval(struct gfs2_glock *gl, int flags)
350 {
351 struct gfs2_inode *ip = gfs2_glock2inode(gl);
352
353 if (flags & DIO_METADATA) {
354 struct address_space *mapping = gfs2_glock2aspace(gl);
355 truncate_inode_pages(mapping, 0);
356 if (ip) {
357 set_bit(GIF_INVALID, &ip->i_flags);
358 forget_all_cached_acls(&ip->i_inode);
359 security_inode_invalidate_secctx(&ip->i_inode);
360 gfs2_dir_hash_inval(ip);
361 }
362 }
363
364 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
365 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
366 GFS2_LOG_HEAD_FLUSH_NORMAL |
367 GFS2_LFC_INODE_GO_INVAL);
368 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
369 }
370 if (ip && S_ISREG(ip->i_inode.i_mode))
371 truncate_inode_pages(ip->i_inode.i_mapping, 0);
372
373 gfs2_clear_glop_pending(ip);
374 }
375
376 /**
377 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
378 * @gl: the glock
379 *
380 * Returns: 1 if it's ok
381 */
382
inode_go_demote_ok(const struct gfs2_glock * gl)383 static int inode_go_demote_ok(const struct gfs2_glock *gl)
384 {
385 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
386
387 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
388 return 0;
389
390 return 1;
391 }
392
gfs2_dinode_in(struct gfs2_inode * ip,const void * buf)393 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
394 {
395 const struct gfs2_dinode *str = buf;
396 struct timespec64 atime;
397 u16 height, depth;
398 umode_t mode = be32_to_cpu(str->di_mode);
399 bool is_new = ip->i_inode.i_flags & I_NEW;
400
401 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
402 goto corrupt;
403 if (unlikely(!is_new && inode_wrong_type(&ip->i_inode, mode)))
404 goto corrupt;
405 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
406 ip->i_inode.i_mode = mode;
407 if (is_new) {
408 ip->i_inode.i_rdev = 0;
409 switch (mode & S_IFMT) {
410 case S_IFBLK:
411 case S_IFCHR:
412 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
413 be32_to_cpu(str->di_minor));
414 break;
415 }
416 }
417
418 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
419 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
420 set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
421 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
422 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
423 atime.tv_sec = be64_to_cpu(str->di_atime);
424 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
425 if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
426 ip->i_inode.i_atime = atime;
427 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
428 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
429 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
430 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
431
432 ip->i_goal = be64_to_cpu(str->di_goal_meta);
433 ip->i_generation = be64_to_cpu(str->di_generation);
434
435 ip->i_diskflags = be32_to_cpu(str->di_flags);
436 ip->i_eattr = be64_to_cpu(str->di_eattr);
437 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
438 gfs2_set_inode_flags(&ip->i_inode);
439 height = be16_to_cpu(str->di_height);
440 if (unlikely(height > GFS2_MAX_META_HEIGHT))
441 goto corrupt;
442 ip->i_height = (u8)height;
443
444 depth = be16_to_cpu(str->di_depth);
445 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
446 goto corrupt;
447 ip->i_depth = (u8)depth;
448 ip->i_entries = be32_to_cpu(str->di_entries);
449
450 if (S_ISREG(ip->i_inode.i_mode))
451 gfs2_set_aops(&ip->i_inode);
452
453 return 0;
454 corrupt:
455 gfs2_consist_inode(ip);
456 return -EIO;
457 }
458
459 /**
460 * gfs2_inode_refresh - Refresh the incore copy of the dinode
461 * @ip: The GFS2 inode
462 *
463 * Returns: errno
464 */
465
gfs2_inode_refresh(struct gfs2_inode * ip)466 int gfs2_inode_refresh(struct gfs2_inode *ip)
467 {
468 struct buffer_head *dibh;
469 int error;
470
471 error = gfs2_meta_inode_buffer(ip, &dibh);
472 if (error)
473 return error;
474
475 error = gfs2_dinode_in(ip, dibh->b_data);
476 brelse(dibh);
477 clear_bit(GIF_INVALID, &ip->i_flags);
478
479 return error;
480 }
481
482 /**
483 * inode_go_lock - operation done after an inode lock is locked by a process
484 * @gh: The glock holder
485 *
486 * Returns: errno
487 */
488
inode_go_lock(struct gfs2_holder * gh)489 static int inode_go_lock(struct gfs2_holder *gh)
490 {
491 struct gfs2_glock *gl = gh->gh_gl;
492 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
493 struct gfs2_inode *ip = gl->gl_object;
494 int error = 0;
495
496 if (!ip || (gh->gh_flags & GL_SKIP))
497 return 0;
498
499 if (test_bit(GIF_INVALID, &ip->i_flags)) {
500 error = gfs2_inode_refresh(ip);
501 if (error)
502 return error;
503 }
504
505 if (gh->gh_state != LM_ST_DEFERRED)
506 inode_dio_wait(&ip->i_inode);
507
508 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
509 (gl->gl_state == LM_ST_EXCLUSIVE) &&
510 (gh->gh_state == LM_ST_EXCLUSIVE)) {
511 spin_lock(&sdp->sd_trunc_lock);
512 if (list_empty(&ip->i_trunc_list))
513 list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
514 spin_unlock(&sdp->sd_trunc_lock);
515 wake_up(&sdp->sd_quota_wait);
516 return 1;
517 }
518
519 return error;
520 }
521
522 /**
523 * inode_go_dump - print information about an inode
524 * @seq: The iterator
525 * @gl: The glock
526 * @fs_id_buf: file system id (may be empty)
527 *
528 */
529
inode_go_dump(struct seq_file * seq,struct gfs2_glock * gl,const char * fs_id_buf)530 static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
531 const char *fs_id_buf)
532 {
533 struct gfs2_inode *ip = gl->gl_object;
534 struct inode *inode = &ip->i_inode;
535 unsigned long nrpages;
536
537 if (ip == NULL)
538 return;
539
540 xa_lock_irq(&inode->i_data.i_pages);
541 nrpages = inode->i_data.nrpages;
542 xa_unlock_irq(&inode->i_data.i_pages);
543
544 gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
545 "p:%lu\n", fs_id_buf,
546 (unsigned long long)ip->i_no_formal_ino,
547 (unsigned long long)ip->i_no_addr,
548 IF2DT(ip->i_inode.i_mode), ip->i_flags,
549 (unsigned int)ip->i_diskflags,
550 (unsigned long long)i_size_read(inode), nrpages);
551 }
552
553 /**
554 * freeze_go_sync - promote/demote the freeze glock
555 * @gl: the glock
556 */
557
freeze_go_sync(struct gfs2_glock * gl)558 static int freeze_go_sync(struct gfs2_glock *gl)
559 {
560 int error = 0;
561 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
562
563 /*
564 * We need to check gl_state == LM_ST_SHARED here and not gl_req ==
565 * LM_ST_EXCLUSIVE. That's because when any node does a freeze,
566 * all the nodes should have the freeze glock in SH mode and they all
567 * call do_xmote: One for EX and the others for UN. They ALL must
568 * freeze locally, and they ALL must queue freeze work. The freeze_work
569 * calls freeze_func, which tries to reacquire the freeze glock in SH,
570 * effectively waiting for the thaw on the node who holds it in EX.
571 * Once thawed, the work func acquires the freeze glock in
572 * SH and everybody goes back to thawed.
573 */
574 if (gl->gl_state == LM_ST_SHARED && !gfs2_withdrawn(sdp) &&
575 !test_bit(SDF_NORECOVERY, &sdp->sd_flags)) {
576 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
577 error = freeze_super(sdp->sd_vfs);
578 if (error) {
579 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
580 error);
581 if (gfs2_withdrawn(sdp)) {
582 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
583 return 0;
584 }
585 gfs2_assert_withdraw(sdp, 0);
586 }
587 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
588 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
589 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
590 GFS2_LFC_FREEZE_GO_SYNC);
591 else /* read-only mounts */
592 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
593 }
594 return 0;
595 }
596
597 /**
598 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
599 * @gl: the glock
600 */
freeze_go_xmote_bh(struct gfs2_glock * gl)601 static int freeze_go_xmote_bh(struct gfs2_glock *gl)
602 {
603 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
604 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
605 struct gfs2_glock *j_gl = ip->i_gl;
606 struct gfs2_log_header_host head;
607 int error;
608
609 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
610 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
611
612 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
613 if (error)
614 gfs2_consist(sdp);
615 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
616 gfs2_consist(sdp);
617
618 /* Initialize some head of the log stuff */
619 if (!gfs2_withdrawn(sdp)) {
620 sdp->sd_log_sequence = head.lh_sequence + 1;
621 gfs2_log_pointers_init(sdp, head.lh_blkno);
622 }
623 }
624 return 0;
625 }
626
627 /**
628 * freeze_go_demote_ok
629 * @gl: the glock
630 *
631 * Always returns 0
632 */
633
freeze_go_demote_ok(const struct gfs2_glock * gl)634 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
635 {
636 return 0;
637 }
638
639 /**
640 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
641 * @gl: the glock
642 * @remote: true if this came from a different cluster node
643 *
644 * gl_lockref.lock lock is held while calling this
645 */
iopen_go_callback(struct gfs2_glock * gl,bool remote)646 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
647 {
648 struct gfs2_inode *ip = gl->gl_object;
649 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
650
651 if (!remote || sb_rdonly(sdp->sd_vfs))
652 return;
653
654 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
655 gl->gl_state == LM_ST_SHARED && ip) {
656 gl->gl_lockref.count++;
657 if (!queue_delayed_work(gfs2_delete_workqueue,
658 &gl->gl_delete, 0))
659 gl->gl_lockref.count--;
660 }
661 }
662
iopen_go_demote_ok(const struct gfs2_glock * gl)663 static int iopen_go_demote_ok(const struct gfs2_glock *gl)
664 {
665 return !gfs2_delete_work_queued(gl);
666 }
667
668 /**
669 * inode_go_free - wake up anyone waiting for dlm's unlock ast to free it
670 * @gl: glock being freed
671 *
672 * For now, this is only used for the journal inode glock. In withdraw
673 * situations, we need to wait for the glock to be freed so that we know
674 * other nodes may proceed with recovery / journal replay.
675 */
inode_go_free(struct gfs2_glock * gl)676 static void inode_go_free(struct gfs2_glock *gl)
677 {
678 /* Note that we cannot reference gl_object because it's already set
679 * to NULL by this point in its lifecycle. */
680 if (!test_bit(GLF_FREEING, &gl->gl_flags))
681 return;
682 clear_bit_unlock(GLF_FREEING, &gl->gl_flags);
683 wake_up_bit(&gl->gl_flags, GLF_FREEING);
684 }
685
686 /**
687 * nondisk_go_callback - used to signal when a node did a withdraw
688 * @gl: the nondisk glock
689 * @remote: true if this came from a different cluster node
690 *
691 */
nondisk_go_callback(struct gfs2_glock * gl,bool remote)692 static void nondisk_go_callback(struct gfs2_glock *gl, bool remote)
693 {
694 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
695
696 /* Ignore the callback unless it's from another node, and it's the
697 live lock. */
698 if (!remote || gl->gl_name.ln_number != GFS2_LIVE_LOCK)
699 return;
700
701 /* First order of business is to cancel the demote request. We don't
702 * really want to demote a nondisk glock. At best it's just to inform
703 * us of another node's withdraw. We'll keep it in SH mode. */
704 clear_bit(GLF_DEMOTE, &gl->gl_flags);
705 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
706
707 /* Ignore the unlock if we're withdrawn, unmounting, or in recovery. */
708 if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) ||
709 test_bit(SDF_WITHDRAWN, &sdp->sd_flags) ||
710 test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags))
711 return;
712
713 /* We only care when a node wants us to unlock, because that means
714 * they want a journal recovered. */
715 if (gl->gl_demote_state != LM_ST_UNLOCKED)
716 return;
717
718 if (sdp->sd_args.ar_spectator) {
719 fs_warn(sdp, "Spectator node cannot recover journals.\n");
720 return;
721 }
722
723 fs_warn(sdp, "Some node has withdrawn; checking for recovery.\n");
724 set_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
725 /*
726 * We can't call remote_withdraw directly here or gfs2_recover_journal
727 * because this is called from the glock unlock function and the
728 * remote_withdraw needs to enqueue and dequeue the same "live" glock
729 * we were called from. So we queue it to the control work queue in
730 * lock_dlm.
731 */
732 queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
733 }
734
735 const struct gfs2_glock_operations gfs2_meta_glops = {
736 .go_type = LM_TYPE_META,
737 .go_flags = GLOF_NONDISK,
738 };
739
740 const struct gfs2_glock_operations gfs2_inode_glops = {
741 .go_sync = inode_go_sync,
742 .go_inval = inode_go_inval,
743 .go_demote_ok = inode_go_demote_ok,
744 .go_lock = inode_go_lock,
745 .go_dump = inode_go_dump,
746 .go_type = LM_TYPE_INODE,
747 .go_flags = GLOF_ASPACE | GLOF_LRU | GLOF_LVB,
748 .go_free = inode_go_free,
749 };
750
751 const struct gfs2_glock_operations gfs2_rgrp_glops = {
752 .go_sync = rgrp_go_sync,
753 .go_inval = rgrp_go_inval,
754 .go_lock = gfs2_rgrp_go_lock,
755 .go_dump = gfs2_rgrp_go_dump,
756 .go_type = LM_TYPE_RGRP,
757 .go_flags = GLOF_LVB,
758 };
759
760 const struct gfs2_glock_operations gfs2_freeze_glops = {
761 .go_sync = freeze_go_sync,
762 .go_xmote_bh = freeze_go_xmote_bh,
763 .go_demote_ok = freeze_go_demote_ok,
764 .go_type = LM_TYPE_NONDISK,
765 .go_flags = GLOF_NONDISK,
766 };
767
768 const struct gfs2_glock_operations gfs2_iopen_glops = {
769 .go_type = LM_TYPE_IOPEN,
770 .go_callback = iopen_go_callback,
771 .go_demote_ok = iopen_go_demote_ok,
772 .go_flags = GLOF_LRU | GLOF_NONDISK,
773 .go_subclass = 1,
774 };
775
776 const struct gfs2_glock_operations gfs2_flock_glops = {
777 .go_type = LM_TYPE_FLOCK,
778 .go_flags = GLOF_LRU | GLOF_NONDISK,
779 };
780
781 const struct gfs2_glock_operations gfs2_nondisk_glops = {
782 .go_type = LM_TYPE_NONDISK,
783 .go_flags = GLOF_NONDISK,
784 .go_callback = nondisk_go_callback,
785 };
786
787 const struct gfs2_glock_operations gfs2_quota_glops = {
788 .go_type = LM_TYPE_QUOTA,
789 .go_flags = GLOF_LVB | GLOF_LRU | GLOF_NONDISK,
790 };
791
792 const struct gfs2_glock_operations gfs2_journal_glops = {
793 .go_type = LM_TYPE_JOURNAL,
794 .go_flags = GLOF_NONDISK,
795 };
796
797 const struct gfs2_glock_operations *gfs2_glops_list[] = {
798 [LM_TYPE_META] = &gfs2_meta_glops,
799 [LM_TYPE_INODE] = &gfs2_inode_glops,
800 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
801 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
802 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
803 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
804 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
805 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
806 };
807
808