gfs2: Don't demote a glock until its revokes are written
[linux-2.6-block.git] / fs / gfs2 / glops.c
CommitLineData
7336d0e6 1// SPDX-License-Identifier: GPL-2.0-only
b3b94faa
DT
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
cf45b752 4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
b3b94faa
DT
5 */
6
b3b94faa
DT
7#include <linux/spinlock.h>
8#include <linux/completion.h>
9#include <linux/buffer_head.h>
5c676f6d 10#include <linux/gfs2_ondisk.h>
6802e340 11#include <linux/bio.h>
c65f7fb5 12#include <linux/posix_acl.h>
f39814f6 13#include <linux/security.h>
b3b94faa
DT
14
15#include "gfs2.h"
5c676f6d 16#include "incore.h"
b3b94faa
DT
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"
b3b94faa
DT
23#include "recovery.h"
24#include "rgrp.h"
5c676f6d 25#include "util.h"
ddacfaf7 26#include "trans.h"
17d539f0 27#include "dir.h"
f4686c26 28#include "lops.h"
b3b94faa 29
2e60d768
BM
30struct workqueue_struct *gfs2_freeze_wq;
31
601ef0d5
BP
32extern struct workqueue_struct *gfs2_control_wq;
33
75549186
SW
34static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
35{
15562c43
BP
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",
75549186
SW
39 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
40 bh->b_page->mapping, bh->b_page->flags);
15562c43 41 fs_err(gl->gl_name.ln_sbd, "AIL glock %u:%llu mapping %p\n",
75549186
SW
42 gl->gl_name.ln_type, gl->gl_name.ln_number,
43 gfs2_glock2aspace(gl));
badb55ec
AG
44 gfs2_lm(gl->gl_name.ln_sbd, "AIL error\n");
45 gfs2_withdraw(gl->gl_name.ln_sbd);
75549186
SW
46}
47
ddacfaf7 48/**
dba898b0 49 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
ddacfaf7 50 * @gl: the glock
b5b24d7a 51 * @fsync: set when called from fsync (not all buffers will be clean)
ddacfaf7
SW
52 *
53 * None of the buffers should be dirty, locked, or pinned.
54 */
55
1bc333f4
BM
56static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
57 unsigned int nr_revokes)
ddacfaf7 58{
15562c43 59 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
ddacfaf7 60 struct list_head *head = &gl->gl_ail_list;
b5b24d7a 61 struct gfs2_bufdata *bd, *tmp;
ddacfaf7 62 struct buffer_head *bh;
b5b24d7a 63 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
d8348de0 64
b5b24d7a 65 gfs2_log_lock(sdp);
d6a079e8 66 spin_lock(&sdp->sd_ail_lock);
1bc333f4
BM
67 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
68 if (nr_revokes == 0)
69 break;
ddacfaf7 70 bh = bd->bd_bh;
b5b24d7a
SW
71 if (bh->b_state & b_state) {
72 if (fsync)
73 continue;
75549186 74 gfs2_ail_error(gl, bh);
b5b24d7a 75 }
1ad38c43 76 gfs2_trans_add_revoke(sdp, bd);
1bc333f4 77 nr_revokes--;
ddacfaf7 78 }
8eae1ca0 79 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
d6a079e8 80 spin_unlock(&sdp->sd_ail_lock);
b5b24d7a 81 gfs2_log_unlock(sdp);
dba898b0
SW
82}
83
84
85static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
86{
15562c43 87 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
dba898b0
SW
88 struct gfs2_trans tr;
89
90 memset(&tr, 0, sizeof(tr));
d69a3c65
SW
91 INIT_LIST_HEAD(&tr.tr_buf);
92 INIT_LIST_HEAD(&tr.tr_databuf);
dba898b0
SW
93 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
94
9ff78289
BP
95 if (!tr.tr_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);
dba898b0 119 return;
9ff78289 120 }
dba898b0 121
24972557
BM
122 /* A shortened, inline version of gfs2_trans_begin()
123 * tr->alloced is not set since the transaction structure is
124 * on the stack */
2e9eeaa1 125 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes);
d29c0afe 126 tr.tr_ip = _RET_IP_;
2e60d768 127 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
24972557 128 return;
8eae1ca0 129 WARN_ON_ONCE(current->journal_info);
dba898b0
SW
130 current->journal_info = &tr;
131
1bc333f4 132 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
dba898b0
SW
133
134 gfs2_trans_end(sdp);
9ff78289 135flush:
805c0907
BP
136 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
137 GFS2_LFC_AIL_EMPTY_GL);
dba898b0 138}
ddacfaf7 139
b5b24d7a 140void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
dba898b0 141{
15562c43 142 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
dba898b0 143 unsigned int revokes = atomic_read(&gl->gl_ail_count);
1bc333f4 144 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
dba898b0
SW
145 int ret;
146
147 if (!revokes)
148 return;
149
1bc333f4
BM
150 while (revokes > max_revokes)
151 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
152
153 ret = gfs2_trans_begin(sdp, 0, max_revokes);
dba898b0
SW
154 if (ret)
155 return;
1bc333f4 156 __gfs2_ail_flush(gl, fsync, max_revokes);
ddacfaf7 157 gfs2_trans_end(sdp);
805c0907
BP
158 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
159 GFS2_LFC_AIL_FLUSH);
ddacfaf7 160}
ba7f7290
SW
161
162/**
6bac243f 163 * rgrp_go_sync - sync out the metadata for this glock
b3b94faa 164 * @gl: the glock
b3b94faa
DT
165 *
166 * Called when demoting or unlocking an EX glock. We must flush
167 * to disk all dirty buffers/pages relating to this glock, and must not
6f6597ba 168 * return to caller to demote/unlock the glock until I/O is complete.
b3b94faa
DT
169 */
170
6bac243f 171static void rgrp_go_sync(struct gfs2_glock *gl)
b3b94faa 172{
15562c43 173 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
70d4ee94 174 struct address_space *mapping = &sdp->sd_aspace;
b3422cac 175 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
6bac243f
SW
176 int error;
177
178 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
b5d32bea 179 return;
8eae1ca0 180 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
b5d32bea 181
805c0907
BP
182 gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
183 GFS2_LFC_RGRP_GO_SYNC);
70d4ee94
SW
184 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
185 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
186 mapping_set_error(mapping, error);
6bac243f 187 gfs2_ail_empty_gl(gl);
7c9ca621 188
f3dd1649 189 spin_lock(&gl->gl_lockref.lock);
8339ee54
SW
190 rgd = gl->gl_object;
191 if (rgd)
192 gfs2_free_clones(rgd);
f3dd1649 193 spin_unlock(&gl->gl_lockref.lock);
b3b94faa
DT
194}
195
196/**
6bac243f 197 * rgrp_go_inval - invalidate the metadata for this glock
b3b94faa
DT
198 * @gl: the glock
199 * @flags:
200 *
6bac243f
SW
201 * We never used LM_ST_DEFERRED with resource groups, so that we
202 * should always see the metadata flag set here.
203 *
b3b94faa
DT
204 */
205
6bac243f 206static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
b3b94faa 207{
15562c43 208 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
70d4ee94 209 struct address_space *mapping = &sdp->sd_aspace;
6f6597ba 210 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
39b0f1e9
BP
211
212 if (rgd)
213 gfs2_rgrp_brelse(rgd);
b3b94faa 214
8eae1ca0 215 WARN_ON_ONCE(!(flags & DIO_METADATA));
7005c3e4 216 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
cf45b752 217
39b0f1e9 218 if (rgd)
cf45b752 219 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
b3b94faa
DT
220}
221
4fd1a579
AG
222static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
223{
224 struct gfs2_inode *ip;
225
226 spin_lock(&gl->gl_lockref.lock);
227 ip = gl->gl_object;
228 if (ip)
229 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
230 spin_unlock(&gl->gl_lockref.lock);
231 return ip;
232}
233
6f6597ba
AG
234struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
235{
236 struct gfs2_rgrpd *rgd;
237
238 spin_lock(&gl->gl_lockref.lock);
239 rgd = gl->gl_object;
240 spin_unlock(&gl->gl_lockref.lock);
241
242 return rgd;
243}
244
4fd1a579
AG
245static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
246{
247 if (!ip)
248 return;
249
250 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
251 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
252}
253
b5d32bea
SW
254/**
255 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
256 * @gl: the glock protecting the inode
257 *
258 */
259
260static void inode_go_sync(struct gfs2_glock *gl)
261{
4fd1a579
AG
262 struct gfs2_inode *ip = gfs2_glock2inode(gl);
263 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
009d8518 264 struct address_space *metamapping = gfs2_glock2aspace(gl);
3042a2cc
SW
265 int error;
266
4fd1a579 267 if (isreg) {
582d2f7a
SW
268 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
269 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
270 inode_dio_wait(&ip->i_inode);
271 }
6bac243f 272 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
4fd1a579 273 goto out;
b5d32bea 274
8eae1ca0 275 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
6bac243f 276
805c0907
BP
277 gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
278 GFS2_LFC_INODE_GO_SYNC);
6bac243f 279 filemap_fdatawrite(metamapping);
4fd1a579 280 if (isreg) {
6bac243f
SW
281 struct address_space *mapping = ip->i_inode.i_mapping;
282 filemap_fdatawrite(mapping);
283 error = filemap_fdatawait(mapping);
284 mapping_set_error(mapping, error);
b5d32bea 285 }
6bac243f
SW
286 error = filemap_fdatawait(metamapping);
287 mapping_set_error(metamapping, error);
288 gfs2_ail_empty_gl(gl);
52fcd11c
SW
289 /*
290 * Writeback of the data mapping may cause the dirty flag to be set
291 * so we have to clear it again here.
292 */
4e857c58 293 smp_mb__before_atomic();
52fcd11c 294 clear_bit(GLF_DIRTY, &gl->gl_flags);
4fd1a579
AG
295
296out:
297 gfs2_clear_glop_pending(ip);
b5d32bea
SW
298}
299
b3b94faa
DT
300/**
301 * inode_go_inval - prepare a inode glock to be released
302 * @gl: the glock
303 * @flags:
6b49d1d9
GU
304 *
305 * Normally we invalidate everything, but if we are moving into
6bac243f
SW
306 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
307 * can keep hold of the metadata, since it won't have changed.
b3b94faa
DT
308 *
309 */
310
311static void inode_go_inval(struct gfs2_glock *gl, int flags)
312{
4fd1a579 313 struct gfs2_inode *ip = gfs2_glock2inode(gl);
b3b94faa 314
6bac243f 315 if (flags & DIO_METADATA) {
009d8518 316 struct address_space *mapping = gfs2_glock2aspace(gl);
6bac243f 317 truncate_inode_pages(mapping, 0);
c65f7fb5 318 if (ip) {
b004157a 319 set_bit(GIF_INVALID, &ip->i_flags);
c65f7fb5 320 forget_all_cached_acls(&ip->i_inode);
f39814f6 321 security_inode_invalidate_secctx(&ip->i_inode);
17d539f0 322 gfs2_dir_hash_inval(ip);
c65f7fb5 323 }
b004157a
SW
324 }
325
15562c43 326 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
c1696fb8 327 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
805c0907
BP
328 GFS2_LOG_HEAD_FLUSH_NORMAL |
329 GFS2_LFC_INODE_GO_INVAL);
15562c43 330 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
1ce53368 331 }
3cc3f710 332 if (ip && S_ISREG(ip->i_inode.i_mode))
b004157a 333 truncate_inode_pages(ip->i_inode.i_mapping, 0);
4fd1a579
AG
334
335 gfs2_clear_glop_pending(ip);
b3b94faa
DT
336}
337
338/**
339 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
340 * @gl: the glock
341 *
342 * Returns: 1 if it's ok
343 */
344
97cc1025 345static int inode_go_demote_ok(const struct gfs2_glock *gl)
b3b94faa 346{
15562c43 347 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
bc015cb8 348
97cc1025
SW
349 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
350 return 0;
bc015cb8 351
97cc1025 352 return 1;
b3b94faa
DT
353}
354
d4b2cf1b
SW
355static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
356{
357 const struct gfs2_dinode *str = buf;
95582b00 358 struct timespec64 atime;
d4b2cf1b
SW
359 u16 height, depth;
360
361 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
362 goto corrupt;
363 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
364 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
365 ip->i_inode.i_rdev = 0;
366 switch (ip->i_inode.i_mode & S_IFMT) {
367 case S_IFBLK:
368 case S_IFCHR:
369 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
370 be32_to_cpu(str->di_minor));
371 break;
098b9c14 372 }
d4b2cf1b 373
d0546426
EB
374 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
375 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
eebd2e81 376 set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
d4b2cf1b
SW
377 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
378 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
379 atime.tv_sec = be64_to_cpu(str->di_atime);
380 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
95582b00 381 if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
d4b2cf1b
SW
382 ip->i_inode.i_atime = atime;
383 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
384 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
385 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
386 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
387
388 ip->i_goal = be64_to_cpu(str->di_goal_meta);
389 ip->i_generation = be64_to_cpu(str->di_generation);
390
391 ip->i_diskflags = be32_to_cpu(str->di_flags);
9964afbb
SW
392 ip->i_eattr = be64_to_cpu(str->di_eattr);
393 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
d4b2cf1b
SW
394 gfs2_set_inode_flags(&ip->i_inode);
395 height = be16_to_cpu(str->di_height);
396 if (unlikely(height > GFS2_MAX_META_HEIGHT))
397 goto corrupt;
398 ip->i_height = (u8)height;
399
400 depth = be16_to_cpu(str->di_depth);
401 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
402 goto corrupt;
403 ip->i_depth = (u8)depth;
404 ip->i_entries = be32_to_cpu(str->di_entries);
405
d4b2cf1b
SW
406 if (S_ISREG(ip->i_inode.i_mode))
407 gfs2_set_aops(&ip->i_inode);
408
409 return 0;
410corrupt:
411 gfs2_consist_inode(ip);
412 return -EIO;
413}
414
415/**
416 * gfs2_inode_refresh - Refresh the incore copy of the dinode
417 * @ip: The GFS2 inode
418 *
419 * Returns: errno
420 */
421
422int gfs2_inode_refresh(struct gfs2_inode *ip)
423{
424 struct buffer_head *dibh;
425 int error;
426
427 error = gfs2_meta_inode_buffer(ip, &dibh);
428 if (error)
429 return error;
430
d4b2cf1b
SW
431 error = gfs2_dinode_in(ip, dibh->b_data);
432 brelse(dibh);
433 clear_bit(GIF_INVALID, &ip->i_flags);
434
435 return error;
436}
437
b3b94faa
DT
438/**
439 * inode_go_lock - operation done after an inode lock is locked by a process
440 * @gl: the glock
441 * @flags:
442 *
443 * Returns: errno
444 */
445
446static int inode_go_lock(struct gfs2_holder *gh)
447{
448 struct gfs2_glock *gl = gh->gh_gl;
15562c43 449 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
5c676f6d 450 struct gfs2_inode *ip = gl->gl_object;
b3b94faa
DT
451 int error = 0;
452
091806ed 453 if (!ip || (gh->gh_flags & GL_SKIP))
b3b94faa
DT
454 return 0;
455
bfded27b 456 if (test_bit(GIF_INVALID, &ip->i_flags)) {
b3b94faa
DT
457 error = gfs2_inode_refresh(ip);
458 if (error)
459 return error;
b3b94faa
DT
460 }
461
582d2f7a
SW
462 if (gh->gh_state != LM_ST_DEFERRED)
463 inode_dio_wait(&ip->i_inode);
464
383f01fb 465 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
b3b94faa 466 (gl->gl_state == LM_ST_EXCLUSIVE) &&
813e0c46
SW
467 (gh->gh_state == LM_ST_EXCLUSIVE)) {
468 spin_lock(&sdp->sd_trunc_lock);
469 if (list_empty(&ip->i_trunc_list))
e7cb550d 470 list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
813e0c46
SW
471 spin_unlock(&sdp->sd_trunc_lock);
472 wake_up(&sdp->sd_quota_wait);
473 return 1;
474 }
b3b94faa
DT
475
476 return error;
477}
478
6802e340
SW
479/**
480 * inode_go_dump - print information about an inode
481 * @seq: The iterator
482 * @ip: the inode
3792ce97 483 * @fs_id_buf: file system id (may be empty)
6802e340 484 *
6802e340
SW
485 */
486
3792ce97
BP
487static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
488 const char *fs_id_buf)
6802e340 489{
27a2660f
BP
490 struct gfs2_inode *ip = gl->gl_object;
491 struct inode *inode = &ip->i_inode;
492 unsigned long nrpages;
493
6802e340 494 if (ip == NULL)
ac3beb6a 495 return;
27a2660f
BP
496
497 xa_lock_irq(&inode->i_data.i_pages);
498 nrpages = inode->i_data.nrpages;
499 xa_unlock_irq(&inode->i_data.i_pages);
500
3792ce97
BP
501 gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
502 "p:%lu\n", fs_id_buf,
6802e340
SW
503 (unsigned long long)ip->i_no_formal_ino,
504 (unsigned long long)ip->i_no_addr,
fa75cedc
SW
505 IF2DT(ip->i_inode.i_mode), ip->i_flags,
506 (unsigned int)ip->i_diskflags,
27a2660f 507 (unsigned long long)i_size_read(inode), nrpages);
6802e340
SW
508}
509
b3b94faa 510/**
24972557 511 * freeze_go_sync - promote/demote the freeze glock
b3b94faa
DT
512 * @gl: the glock
513 * @state: the requested state
514 * @flags:
515 *
516 */
517
24972557 518static void freeze_go_sync(struct gfs2_glock *gl)
b3b94faa 519{
2e60d768 520 int error = 0;
15562c43 521 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
b3b94faa 522
601ef0d5 523 if (gl->gl_state == LM_ST_SHARED && !gfs2_withdrawn(sdp) &&
b3b94faa 524 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
2e60d768
BM
525 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
526 error = freeze_super(sdp->sd_vfs);
527 if (error) {
f29e62ee
BP
528 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
529 error);
601ef0d5
BP
530 if (gfs2_withdrawn(sdp)) {
531 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
532 return;
533 }
2e60d768
BM
534 gfs2_assert_withdraw(sdp, 0);
535 }
536 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
805c0907
BP
537 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
538 GFS2_LFC_FREEZE_GO_SYNC);
b3b94faa 539 }
b3b94faa
DT
540}
541
542/**
24972557 543 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
b3b94faa
DT
544 * @gl: the glock
545 *
546 */
547
24972557 548static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
b3b94faa 549{
15562c43 550 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
feaa7bba 551 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
5c676f6d 552 struct gfs2_glock *j_gl = ip->i_gl;
55167622 553 struct gfs2_log_header_host head;
b3b94faa
DT
554 int error;
555
6802e340 556 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
1a14d3a6 557 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
b3b94faa 558
f4686c26 559 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
b3b94faa
DT
560 if (error)
561 gfs2_consist(sdp);
562 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
563 gfs2_consist(sdp);
564
565 /* Initialize some head of the log stuff */
eb43e660 566 if (!gfs2_withdrawn(sdp)) {
b3b94faa
DT
567 sdp->sd_log_sequence = head.lh_sequence + 1;
568 gfs2_log_pointers_init(sdp, head.lh_blkno);
569 }
570 }
6802e340 571 return 0;
b3b94faa
DT
572}
573
97cc1025
SW
574/**
575 * trans_go_demote_ok
576 * @gl: the glock
577 *
578 * Always returns 0
579 */
580
24972557 581static int freeze_go_demote_ok(const struct gfs2_glock *gl)
97cc1025
SW
582{
583 return 0;
584}
585
b94a170e
BM
586/**
587 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
588 * @gl: the glock
589 *
f3dd1649 590 * gl_lockref.lock lock is held while calling this
b94a170e 591 */
81ffbf65 592static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
b94a170e 593{
6f6597ba 594 struct gfs2_inode *ip = gl->gl_object;
15562c43 595 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
001e8e8d 596
bc98a42c 597 if (!remote || sb_rdonly(sdp->sd_vfs))
001e8e8d 598 return;
b94a170e
BM
599
600 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
009d8518 601 gl->gl_state == LM_ST_SHARED && ip) {
e66cf161 602 gl->gl_lockref.count++;
b94a170e 603 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
e66cf161 604 gl->gl_lockref.count--;
b94a170e
BM
605 }
606}
607
601ef0d5
BP
608/**
609 * inode_go_free - wake up anyone waiting for dlm's unlock ast to free it
610 * @gl: glock being freed
611 *
612 * For now, this is only used for the journal inode glock. In withdraw
613 * situations, we need to wait for the glock to be freed so that we know
614 * other nodes may proceed with recovery / journal replay.
615 */
616static void inode_go_free(struct gfs2_glock *gl)
617{
618 /* Note that we cannot reference gl_object because it's already set
619 * to NULL by this point in its lifecycle. */
620 if (!test_bit(GLF_FREEING, &gl->gl_flags))
621 return;
622 clear_bit_unlock(GLF_FREEING, &gl->gl_flags);
623 wake_up_bit(&gl->gl_flags, GLF_FREEING);
624}
625
626/**
627 * nondisk_go_callback - used to signal when a node did a withdraw
628 * @gl: the nondisk glock
629 * @remote: true if this came from a different cluster node
630 *
631 */
632static void nondisk_go_callback(struct gfs2_glock *gl, bool remote)
633{
634 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
635
636 /* Ignore the callback unless it's from another node, and it's the
637 live lock. */
638 if (!remote || gl->gl_name.ln_number != GFS2_LIVE_LOCK)
639 return;
640
641 /* First order of business is to cancel the demote request. We don't
642 * really want to demote a nondisk glock. At best it's just to inform
643 * us of another node's withdraw. We'll keep it in SH mode. */
644 clear_bit(GLF_DEMOTE, &gl->gl_flags);
645 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
646
647 /* Ignore the unlock if we're withdrawn, unmounting, or in recovery. */
648 if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) ||
649 test_bit(SDF_WITHDRAWN, &sdp->sd_flags) ||
650 test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags))
651 return;
652
653 /* We only care when a node wants us to unlock, because that means
654 * they want a journal recovered. */
655 if (gl->gl_demote_state != LM_ST_UNLOCKED)
656 return;
657
658 if (sdp->sd_args.ar_spectator) {
659 fs_warn(sdp, "Spectator node cannot recover journals.\n");
660 return;
661 }
662
663 fs_warn(sdp, "Some node has withdrawn; checking for recovery.\n");
664 set_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
665 /*
666 * We can't call remote_withdraw directly here or gfs2_recover_journal
667 * because this is called from the glock unlock function and the
668 * remote_withdraw needs to enqueue and dequeue the same "live" glock
669 * we were called from. So we queue it to the control work queue in
670 * lock_dlm.
671 */
672 queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
673}
674
8fb4b536 675const struct gfs2_glock_operations gfs2_meta_glops = {
ea67eedb 676 .go_type = LM_TYPE_META,
a72d2401 677 .go_flags = GLOF_NONDISK,
b3b94faa
DT
678};
679
8fb4b536 680const struct gfs2_glock_operations gfs2_inode_glops = {
06dfc306 681 .go_sync = inode_go_sync,
b3b94faa
DT
682 .go_inval = inode_go_inval,
683 .go_demote_ok = inode_go_demote_ok,
684 .go_lock = inode_go_lock,
6802e340 685 .go_dump = inode_go_dump,
ea67eedb 686 .go_type = LM_TYPE_INODE,
e7ccaf5f 687 .go_flags = GLOF_ASPACE | GLOF_LRU,
601ef0d5 688 .go_free = inode_go_free,
b3b94faa
DT
689};
690
8fb4b536 691const struct gfs2_glock_operations gfs2_rgrp_glops = {
06dfc306 692 .go_sync = rgrp_go_sync,
6bac243f 693 .go_inval = rgrp_go_inval,
7c9ca621 694 .go_lock = gfs2_rgrp_go_lock,
09010978 695 .go_dump = gfs2_rgrp_dump,
ea67eedb 696 .go_type = LM_TYPE_RGRP,
70d4ee94 697 .go_flags = GLOF_LVB,
b3b94faa
DT
698};
699
24972557
BM
700const struct gfs2_glock_operations gfs2_freeze_glops = {
701 .go_sync = freeze_go_sync,
702 .go_xmote_bh = freeze_go_xmote_bh,
703 .go_demote_ok = freeze_go_demote_ok,
ea67eedb 704 .go_type = LM_TYPE_NONDISK,
a72d2401 705 .go_flags = GLOF_NONDISK,
b3b94faa
DT
706};
707
8fb4b536 708const struct gfs2_glock_operations gfs2_iopen_glops = {
ea67eedb 709 .go_type = LM_TYPE_IOPEN,
b94a170e 710 .go_callback = iopen_go_callback,
a72d2401 711 .go_flags = GLOF_LRU | GLOF_NONDISK,
b3b94faa
DT
712};
713
8fb4b536 714const struct gfs2_glock_operations gfs2_flock_glops = {
ea67eedb 715 .go_type = LM_TYPE_FLOCK,
a72d2401 716 .go_flags = GLOF_LRU | GLOF_NONDISK,
b3b94faa
DT
717};
718
8fb4b536 719const struct gfs2_glock_operations gfs2_nondisk_glops = {
ea67eedb 720 .go_type = LM_TYPE_NONDISK,
a72d2401 721 .go_flags = GLOF_NONDISK,
601ef0d5 722 .go_callback = nondisk_go_callback,
b3b94faa
DT
723};
724
8fb4b536 725const struct gfs2_glock_operations gfs2_quota_glops = {
ea67eedb 726 .go_type = LM_TYPE_QUOTA,
a72d2401 727 .go_flags = GLOF_LVB | GLOF_LRU | GLOF_NONDISK,
b3b94faa
DT
728};
729
8fb4b536 730const struct gfs2_glock_operations gfs2_journal_glops = {
ea67eedb 731 .go_type = LM_TYPE_JOURNAL,
a72d2401 732 .go_flags = GLOF_NONDISK,
b3b94faa
DT
733};
734
64d576ba
SW
735const struct gfs2_glock_operations *gfs2_glops_list[] = {
736 [LM_TYPE_META] = &gfs2_meta_glops,
737 [LM_TYPE_INODE] = &gfs2_inode_glops,
738 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
64d576ba
SW
739 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
740 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
741 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
742 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
743 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
744};
745