Merge tag 'fpga-for-6.3-final' of git://git.kernel.org/pub/scm/linux/kernel/git/fpga...
[linux-2.6-microblaze.git] / fs / gfs2 / glops.c
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
87
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;
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                 goto flush;
129         __gfs2_ail_flush(gl, 0, revokes);
130         gfs2_trans_end(sdp);
131
132 flush:
133         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
134                        GFS2_LFC_AIL_EMPTY_GL);
135         return 0;
136 }
137
138 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
139 {
140         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
141         unsigned int revokes = atomic_read(&gl->gl_ail_count);
142         int ret;
143
144         if (!revokes)
145                 return;
146
147         ret = gfs2_trans_begin(sdp, 0, revokes);
148         if (ret)
149                 return;
150         __gfs2_ail_flush(gl, fsync, revokes);
151         gfs2_trans_end(sdp);
152         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
153                        GFS2_LFC_AIL_FLUSH);
154 }
155
156 /**
157  * gfs2_rgrp_metasync - sync out the metadata of a resource group
158  * @gl: the glock protecting the resource group
159  *
160  */
161
162 static int gfs2_rgrp_metasync(struct gfs2_glock *gl)
163 {
164         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
165         struct address_space *metamapping = &sdp->sd_aspace;
166         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
167         const unsigned bsize = sdp->sd_sb.sb_bsize;
168         loff_t start = (rgd->rd_addr * bsize) & PAGE_MASK;
169         loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
170         int error;
171
172         filemap_fdatawrite_range(metamapping, start, end);
173         error = filemap_fdatawait_range(metamapping, start, end);
174         WARN_ON_ONCE(error && !gfs2_withdrawn(sdp));
175         mapping_set_error(metamapping, error);
176         if (error)
177                 gfs2_io_error(sdp);
178         return error;
179 }
180
181 /**
182  * rgrp_go_sync - sync out the metadata for this glock
183  * @gl: the glock
184  *
185  * Called when demoting or unlocking an EX glock.  We must flush
186  * to disk all dirty buffers/pages relating to this glock, and must not
187  * return to caller to demote/unlock the glock until I/O is complete.
188  */
189
190 static int rgrp_go_sync(struct gfs2_glock *gl)
191 {
192         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
193         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
194         int error;
195
196         if (!rgd || !test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
197                 return 0;
198         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
199
200         gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
201                        GFS2_LFC_RGRP_GO_SYNC);
202         error = gfs2_rgrp_metasync(gl);
203         if (!error)
204                 error = gfs2_ail_empty_gl(gl);
205         gfs2_free_clones(rgd);
206         return error;
207 }
208
209 /**
210  * rgrp_go_inval - invalidate the metadata for this glock
211  * @gl: the glock
212  * @flags:
213  *
214  * We never used LM_ST_DEFERRED with resource groups, so that we
215  * should always see the metadata flag set here.
216  *
217  */
218
219 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
220 {
221         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
222         struct address_space *mapping = &sdp->sd_aspace;
223         struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
224         const unsigned bsize = sdp->sd_sb.sb_bsize;
225         loff_t start, end;
226
227         if (!rgd)
228                 return;
229         start = (rgd->rd_addr * bsize) & PAGE_MASK;
230         end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
231         gfs2_rgrp_brelse(rgd);
232         WARN_ON_ONCE(!(flags & DIO_METADATA));
233         truncate_inode_pages_range(mapping, start, end);
234 }
235
236 static void gfs2_rgrp_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
237                               const char *fs_id_buf)
238 {
239         struct gfs2_rgrpd *rgd = gl->gl_object;
240
241         if (rgd)
242                 gfs2_rgrp_dump(seq, rgd, fs_id_buf);
243 }
244
245 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
246 {
247         struct gfs2_inode *ip;
248
249         spin_lock(&gl->gl_lockref.lock);
250         ip = gl->gl_object;
251         if (ip)
252                 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
253         spin_unlock(&gl->gl_lockref.lock);
254         return ip;
255 }
256
257 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
258 {
259         struct gfs2_rgrpd *rgd;
260
261         spin_lock(&gl->gl_lockref.lock);
262         rgd = gl->gl_object;
263         spin_unlock(&gl->gl_lockref.lock);
264
265         return rgd;
266 }
267
268 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
269 {
270         if (!ip)
271                 return;
272
273         clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
274         wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
275 }
276
277 /**
278  * gfs2_inode_metasync - sync out the metadata of an inode
279  * @gl: the glock protecting the inode
280  *
281  */
282 int gfs2_inode_metasync(struct gfs2_glock *gl)
283 {
284         struct address_space *metamapping = gfs2_glock2aspace(gl);
285         int error;
286
287         filemap_fdatawrite(metamapping);
288         error = filemap_fdatawait(metamapping);
289         if (error)
290                 gfs2_io_error(gl->gl_name.ln_sbd);
291         return error;
292 }
293
294 /**
295  * inode_go_sync - Sync the dirty metadata of an inode
296  * @gl: the glock protecting the inode
297  *
298  */
299
300 static int inode_go_sync(struct gfs2_glock *gl)
301 {
302         struct gfs2_inode *ip = gfs2_glock2inode(gl);
303         int isreg = ip && S_ISREG(ip->i_inode.i_mode);
304         struct address_space *metamapping = gfs2_glock2aspace(gl);
305         int error = 0, ret;
306
307         if (isreg) {
308                 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
309                         unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
310                 inode_dio_wait(&ip->i_inode);
311         }
312         if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
313                 goto out;
314
315         GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
316
317         gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
318                        GFS2_LFC_INODE_GO_SYNC);
319         filemap_fdatawrite(metamapping);
320         if (isreg) {
321                 struct address_space *mapping = ip->i_inode.i_mapping;
322                 filemap_fdatawrite(mapping);
323                 error = filemap_fdatawait(mapping);
324                 mapping_set_error(mapping, error);
325         }
326         ret = gfs2_inode_metasync(gl);
327         if (!error)
328                 error = ret;
329         gfs2_ail_empty_gl(gl);
330         /*
331          * Writeback of the data mapping may cause the dirty flag to be set
332          * so we have to clear it again here.
333          */
334         smp_mb__before_atomic();
335         clear_bit(GLF_DIRTY, &gl->gl_flags);
336
337 out:
338         gfs2_clear_glop_pending(ip);
339         return error;
340 }
341
342 /**
343  * inode_go_inval - prepare a inode glock to be released
344  * @gl: the glock
345  * @flags:
346  *
347  * Normally we invalidate everything, but if we are moving into
348  * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
349  * can keep hold of the metadata, since it won't have changed.
350  *
351  */
352
353 static void inode_go_inval(struct gfs2_glock *gl, int flags)
354 {
355         struct gfs2_inode *ip = gfs2_glock2inode(gl);
356
357         if (flags & DIO_METADATA) {
358                 struct address_space *mapping = gfs2_glock2aspace(gl);
359                 truncate_inode_pages(mapping, 0);
360                 if (ip) {
361                         set_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
362                         forget_all_cached_acls(&ip->i_inode);
363                         security_inode_invalidate_secctx(&ip->i_inode);
364                         gfs2_dir_hash_inval(ip);
365                 }
366         }
367
368         if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
369                 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
370                                GFS2_LOG_HEAD_FLUSH_NORMAL |
371                                GFS2_LFC_INODE_GO_INVAL);
372                 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
373         }
374         if (ip && S_ISREG(ip->i_inode.i_mode))
375                 truncate_inode_pages(ip->i_inode.i_mapping, 0);
376
377         gfs2_clear_glop_pending(ip);
378 }
379
380 /**
381  * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
382  * @gl: the glock
383  *
384  * Returns: 1 if it's ok
385  */
386
387 static int inode_go_demote_ok(const struct gfs2_glock *gl)
388 {
389         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
390
391         if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
392                 return 0;
393
394         return 1;
395 }
396
397 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
398 {
399         const struct gfs2_dinode *str = buf;
400         struct timespec64 atime;
401         u16 height, depth;
402         umode_t mode = be32_to_cpu(str->di_mode);
403         struct inode *inode = &ip->i_inode;
404         bool is_new = inode->i_state & I_NEW;
405
406         if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
407                 goto corrupt;
408         if (unlikely(!is_new && inode_wrong_type(inode, mode)))
409                 goto corrupt;
410         ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
411         inode->i_mode = mode;
412         if (is_new) {
413                 inode->i_rdev = 0;
414                 switch (mode & S_IFMT) {
415                 case S_IFBLK:
416                 case S_IFCHR:
417                         inode->i_rdev = MKDEV(be32_to_cpu(str->di_major),
418                                               be32_to_cpu(str->di_minor));
419                         break;
420                 }
421         }
422
423         i_uid_write(inode, be32_to_cpu(str->di_uid));
424         i_gid_write(inode, be32_to_cpu(str->di_gid));
425         set_nlink(inode, be32_to_cpu(str->di_nlink));
426         i_size_write(inode, be64_to_cpu(str->di_size));
427         gfs2_set_inode_blocks(inode, be64_to_cpu(str->di_blocks));
428         atime.tv_sec = be64_to_cpu(str->di_atime);
429         atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
430         if (timespec64_compare(&inode->i_atime, &atime) < 0)
431                 inode->i_atime = atime;
432         inode->i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
433         inode->i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
434         inode->i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
435         inode->i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
436
437         ip->i_goal = be64_to_cpu(str->di_goal_meta);
438         ip->i_generation = be64_to_cpu(str->di_generation);
439
440         ip->i_diskflags = be32_to_cpu(str->di_flags);
441         ip->i_eattr = be64_to_cpu(str->di_eattr);
442         /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
443         gfs2_set_inode_flags(inode);
444         height = be16_to_cpu(str->di_height);
445         if (unlikely(height > GFS2_MAX_META_HEIGHT))
446                 goto corrupt;
447         ip->i_height = (u8)height;
448
449         depth = be16_to_cpu(str->di_depth);
450         if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
451                 goto corrupt;
452         ip->i_depth = (u8)depth;
453         ip->i_entries = be32_to_cpu(str->di_entries);
454
455         if (gfs2_is_stuffed(ip) && inode->i_size > gfs2_max_stuffed_size(ip))
456                 goto corrupt;
457
458         if (S_ISREG(inode->i_mode))
459                 gfs2_set_aops(inode);
460
461         return 0;
462 corrupt:
463         gfs2_consist_inode(ip);
464         return -EIO;
465 }
466
467 /**
468  * gfs2_inode_refresh - Refresh the incore copy of the dinode
469  * @ip: The GFS2 inode
470  *
471  * Returns: errno
472  */
473
474 int gfs2_inode_refresh(struct gfs2_inode *ip)
475 {
476         struct buffer_head *dibh;
477         int error;
478
479         error = gfs2_meta_inode_buffer(ip, &dibh);
480         if (error)
481                 return error;
482
483         error = gfs2_dinode_in(ip, dibh->b_data);
484         brelse(dibh);
485         return error;
486 }
487
488 /**
489  * inode_go_instantiate - read in an inode if necessary
490  * @gh: The glock holder
491  *
492  * Returns: errno
493  */
494
495 static int inode_go_instantiate(struct gfs2_glock *gl)
496 {
497         struct gfs2_inode *ip = gl->gl_object;
498
499         if (!ip) /* no inode to populate - read it in later */
500                 return 0;
501
502         return gfs2_inode_refresh(ip);
503 }
504
505 static int inode_go_held(struct gfs2_holder *gh)
506 {
507         struct gfs2_glock *gl = gh->gh_gl;
508         struct gfs2_inode *ip = gl->gl_object;
509         int error = 0;
510
511         if (!ip) /* no inode to populate - read it in later */
512                 return 0;
513
514         if (gh->gh_state != LM_ST_DEFERRED)
515                 inode_dio_wait(&ip->i_inode);
516
517         if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
518             (gl->gl_state == LM_ST_EXCLUSIVE) &&
519             (gh->gh_state == LM_ST_EXCLUSIVE))
520                 error = gfs2_truncatei_resume(ip);
521
522         return error;
523 }
524
525 /**
526  * inode_go_dump - print information about an inode
527  * @seq: The iterator
528  * @gl: The glock
529  * @fs_id_buf: file system id (may be empty)
530  *
531  */
532
533 static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
534                           const char *fs_id_buf)
535 {
536         struct gfs2_inode *ip = gl->gl_object;
537         struct inode *inode = &ip->i_inode;
538         unsigned long nrpages;
539
540         if (ip == NULL)
541                 return;
542
543         xa_lock_irq(&inode->i_data.i_pages);
544         nrpages = inode->i_data.nrpages;
545         xa_unlock_irq(&inode->i_data.i_pages);
546
547         gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
548                        "p:%lu\n", fs_id_buf,
549                   (unsigned long long)ip->i_no_formal_ino,
550                   (unsigned long long)ip->i_no_addr,
551                   IF2DT(ip->i_inode.i_mode), ip->i_flags,
552                   (unsigned int)ip->i_diskflags,
553                   (unsigned long long)i_size_read(inode), nrpages);
554 }
555
556 /**
557  * freeze_go_sync - promote/demote the freeze glock
558  * @gl: the glock
559  */
560
561 static int freeze_go_sync(struct gfs2_glock *gl)
562 {
563         int error = 0;
564         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
565
566         /*
567          * We need to check gl_state == LM_ST_SHARED here and not gl_req ==
568          * LM_ST_EXCLUSIVE. That's because when any node does a freeze,
569          * all the nodes should have the freeze glock in SH mode and they all
570          * call do_xmote: One for EX and the others for UN. They ALL must
571          * freeze locally, and they ALL must queue freeze work. The freeze_work
572          * calls freeze_func, which tries to reacquire the freeze glock in SH,
573          * effectively waiting for the thaw on the node who holds it in EX.
574          * Once thawed, the work func acquires the freeze glock in
575          * SH and everybody goes back to thawed.
576          */
577         if (gl->gl_state == LM_ST_SHARED && !gfs2_withdrawn(sdp) &&
578             !test_bit(SDF_NORECOVERY, &sdp->sd_flags)) {
579                 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
580                 error = freeze_super(sdp->sd_vfs);
581                 if (error) {
582                         fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
583                                 error);
584                         if (gfs2_withdrawn(sdp)) {
585                                 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
586                                 return 0;
587                         }
588                         gfs2_assert_withdraw(sdp, 0);
589                 }
590                 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
591                 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
592                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
593                                        GFS2_LFC_FREEZE_GO_SYNC);
594                 else /* read-only mounts */
595                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
596         }
597         return 0;
598 }
599
600 /**
601  * freeze_go_xmote_bh - After promoting/demoting the freeze glock
602  * @gl: the glock
603  */
604 static int freeze_go_xmote_bh(struct gfs2_glock *gl)
605 {
606         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
607         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
608         struct gfs2_glock *j_gl = ip->i_gl;
609         struct gfs2_log_header_host head;
610         int error;
611
612         if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
613                 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
614
615                 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
616                 if (gfs2_assert_withdraw_delayed(sdp, !error))
617                         return error;
618                 if (gfs2_assert_withdraw_delayed(sdp, head.lh_flags &
619                                                  GFS2_LOG_HEAD_UNMOUNT))
620                         return -EIO;
621                 sdp->sd_log_sequence = head.lh_sequence + 1;
622                 gfs2_log_pointers_init(sdp, head.lh_blkno);
623         }
624         return 0;
625 }
626
627 /**
628  * freeze_go_demote_ok
629  * @gl: the glock
630  *
631  * Always returns 0
632  */
633
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  */
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             test_bit(SDF_DEACTIVATING, &sdp->sd_flags))
653                 return;
654
655         if (gl->gl_demote_state == LM_ST_UNLOCKED &&
656             gl->gl_state == LM_ST_SHARED && ip) {
657                 gl->gl_lockref.count++;
658                 if (!gfs2_queue_try_to_evict(gl))
659                         gl->gl_lockref.count--;
660         }
661 }
662
663 /**
664  * inode_go_free - wake up anyone waiting for dlm's unlock ast to free it
665  * @gl: glock being freed
666  *
667  * For now, this is only used for the journal inode glock. In withdraw
668  * situations, we need to wait for the glock to be freed so that we know
669  * other nodes may proceed with recovery / journal replay.
670  */
671 static void inode_go_free(struct gfs2_glock *gl)
672 {
673         /* Note that we cannot reference gl_object because it's already set
674          * to NULL by this point in its lifecycle. */
675         if (!test_bit(GLF_FREEING, &gl->gl_flags))
676                 return;
677         clear_bit_unlock(GLF_FREEING, &gl->gl_flags);
678         wake_up_bit(&gl->gl_flags, GLF_FREEING);
679 }
680
681 /**
682  * nondisk_go_callback - used to signal when a node did a withdraw
683  * @gl: the nondisk glock
684  * @remote: true if this came from a different cluster node
685  *
686  */
687 static void nondisk_go_callback(struct gfs2_glock *gl, bool remote)
688 {
689         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
690
691         /* Ignore the callback unless it's from another node, and it's the
692            live lock. */
693         if (!remote || gl->gl_name.ln_number != GFS2_LIVE_LOCK)
694                 return;
695
696         /* First order of business is to cancel the demote request. We don't
697          * really want to demote a nondisk glock. At best it's just to inform
698          * us of another node's withdraw. We'll keep it in SH mode. */
699         clear_bit(GLF_DEMOTE, &gl->gl_flags);
700         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
701
702         /* Ignore the unlock if we're withdrawn, unmounting, or in recovery. */
703         if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) ||
704             test_bit(SDF_WITHDRAWN, &sdp->sd_flags) ||
705             test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags))
706                 return;
707
708         /* We only care when a node wants us to unlock, because that means
709          * they want a journal recovered. */
710         if (gl->gl_demote_state != LM_ST_UNLOCKED)
711                 return;
712
713         if (sdp->sd_args.ar_spectator) {
714                 fs_warn(sdp, "Spectator node cannot recover journals.\n");
715                 return;
716         }
717
718         fs_warn(sdp, "Some node has withdrawn; checking for recovery.\n");
719         set_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
720         /*
721          * We can't call remote_withdraw directly here or gfs2_recover_journal
722          * because this is called from the glock unlock function and the
723          * remote_withdraw needs to enqueue and dequeue the same "live" glock
724          * we were called from. So we queue it to the control work queue in
725          * lock_dlm.
726          */
727         queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
728 }
729
730 const struct gfs2_glock_operations gfs2_meta_glops = {
731         .go_type = LM_TYPE_META,
732         .go_flags = GLOF_NONDISK,
733 };
734
735 const struct gfs2_glock_operations gfs2_inode_glops = {
736         .go_sync = inode_go_sync,
737         .go_inval = inode_go_inval,
738         .go_demote_ok = inode_go_demote_ok,
739         .go_instantiate = inode_go_instantiate,
740         .go_held = inode_go_held,
741         .go_dump = inode_go_dump,
742         .go_type = LM_TYPE_INODE,
743         .go_flags = GLOF_ASPACE | GLOF_LRU | GLOF_LVB,
744         .go_free = inode_go_free,
745 };
746
747 const struct gfs2_glock_operations gfs2_rgrp_glops = {
748         .go_sync = rgrp_go_sync,
749         .go_inval = rgrp_go_inval,
750         .go_instantiate = gfs2_rgrp_go_instantiate,
751         .go_dump = gfs2_rgrp_go_dump,
752         .go_type = LM_TYPE_RGRP,
753         .go_flags = GLOF_LVB,
754 };
755
756 const struct gfs2_glock_operations gfs2_freeze_glops = {
757         .go_sync = freeze_go_sync,
758         .go_xmote_bh = freeze_go_xmote_bh,
759         .go_demote_ok = freeze_go_demote_ok,
760         .go_type = LM_TYPE_NONDISK,
761         .go_flags = GLOF_NONDISK,
762 };
763
764 const struct gfs2_glock_operations gfs2_iopen_glops = {
765         .go_type = LM_TYPE_IOPEN,
766         .go_callback = iopen_go_callback,
767         .go_dump = inode_go_dump,
768         .go_flags = GLOF_LRU | GLOF_NONDISK,
769         .go_subclass = 1,
770 };
771
772 const struct gfs2_glock_operations gfs2_flock_glops = {
773         .go_type = LM_TYPE_FLOCK,
774         .go_flags = GLOF_LRU | GLOF_NONDISK,
775 };
776
777 const struct gfs2_glock_operations gfs2_nondisk_glops = {
778         .go_type = LM_TYPE_NONDISK,
779         .go_flags = GLOF_NONDISK,
780         .go_callback = nondisk_go_callback,
781 };
782
783 const struct gfs2_glock_operations gfs2_quota_glops = {
784         .go_type = LM_TYPE_QUOTA,
785         .go_flags = GLOF_LVB | GLOF_LRU | GLOF_NONDISK,
786 };
787
788 const struct gfs2_glock_operations gfs2_journal_glops = {
789         .go_type = LM_TYPE_JOURNAL,
790         .go_flags = GLOF_NONDISK,
791 };
792
793 const struct gfs2_glock_operations *gfs2_glops_list[] = {
794         [LM_TYPE_META] = &gfs2_meta_glops,
795         [LM_TYPE_INODE] = &gfs2_inode_glops,
796         [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
797         [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
798         [LM_TYPE_FLOCK] = &gfs2_flock_glops,
799         [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
800         [LM_TYPE_QUOTA] = &gfs2_quota_glops,
801         [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
802 };
803