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_page->mapping, bh->b_page->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;
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 
gfs2_ail_flush(struct gfs2_glock * gl,bool fsync)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 
gfs2_rgrp_metasync(struct gfs2_glock * gl)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 
rgrp_go_sync(struct gfs2_glock * gl)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 (!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 
rgrp_go_inval(struct gfs2_glock * gl,int flags)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 = (rgd->rd_addr * bsize) & PAGE_MASK;
226 	loff_t end = PAGE_ALIGN((rgd->rd_addr + rgd->rd_length) * bsize) - 1;
227 
228 	gfs2_rgrp_brelse(rgd);
229 	WARN_ON_ONCE(!(flags & DIO_METADATA));
230 	truncate_inode_pages_range(mapping, start, end);
231 }
232 
gfs2_rgrp_go_dump(struct seq_file * seq,struct gfs2_glock * gl,const char * fs_id_buf)233 static void gfs2_rgrp_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
234 			      const char *fs_id_buf)
235 {
236 	struct gfs2_rgrpd *rgd = gl->gl_object;
237 
238 	if (rgd)
239 		gfs2_rgrp_dump(seq, rgd, fs_id_buf);
240 }
241 
gfs2_glock2inode(struct gfs2_glock * gl)242 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
243 {
244 	struct gfs2_inode *ip;
245 
246 	spin_lock(&gl->gl_lockref.lock);
247 	ip = gl->gl_object;
248 	if (ip)
249 		set_bit(GIF_GLOP_PENDING, &ip->i_flags);
250 	spin_unlock(&gl->gl_lockref.lock);
251 	return ip;
252 }
253 
gfs2_glock2rgrp(struct gfs2_glock * gl)254 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
255 {
256 	struct gfs2_rgrpd *rgd;
257 
258 	spin_lock(&gl->gl_lockref.lock);
259 	rgd = gl->gl_object;
260 	spin_unlock(&gl->gl_lockref.lock);
261 
262 	return rgd;
263 }
264 
gfs2_clear_glop_pending(struct gfs2_inode * ip)265 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
266 {
267 	if (!ip)
268 		return;
269 
270 	clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
271 	wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
272 }
273 
274 /**
275  * gfs2_inode_metasync - sync out the metadata of an inode
276  * @gl: the glock protecting the inode
277  *
278  */
gfs2_inode_metasync(struct gfs2_glock * gl)279 int gfs2_inode_metasync(struct gfs2_glock *gl)
280 {
281 	struct address_space *metamapping = gfs2_glock2aspace(gl);
282 	int error;
283 
284 	filemap_fdatawrite(metamapping);
285 	error = filemap_fdatawait(metamapping);
286 	if (error)
287 		gfs2_io_error(gl->gl_name.ln_sbd);
288 	return error;
289 }
290 
291 /**
292  * inode_go_sync - Sync the dirty metadata of an inode
293  * @gl: the glock protecting the inode
294  *
295  */
296 
inode_go_sync(struct gfs2_glock * gl)297 static int inode_go_sync(struct gfs2_glock *gl)
298 {
299 	struct gfs2_inode *ip = gfs2_glock2inode(gl);
300 	int isreg = ip && S_ISREG(ip->i_inode.i_mode);
301 	struct address_space *metamapping = gfs2_glock2aspace(gl);
302 	int error = 0, ret;
303 
304 	if (isreg) {
305 		if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
306 			unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
307 		inode_dio_wait(&ip->i_inode);
308 	}
309 	if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
310 		goto out;
311 
312 	GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
313 
314 	gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
315 		       GFS2_LFC_INODE_GO_SYNC);
316 	filemap_fdatawrite(metamapping);
317 	if (isreg) {
318 		struct address_space *mapping = ip->i_inode.i_mapping;
319 		filemap_fdatawrite(mapping);
320 		error = filemap_fdatawait(mapping);
321 		mapping_set_error(mapping, error);
322 	}
323 	ret = gfs2_inode_metasync(gl);
324 	if (!error)
325 		error = ret;
326 	gfs2_ail_empty_gl(gl);
327 	/*
328 	 * Writeback of the data mapping may cause the dirty flag to be set
329 	 * so we have to clear it again here.
330 	 */
331 	smp_mb__before_atomic();
332 	clear_bit(GLF_DIRTY, &gl->gl_flags);
333 
334 out:
335 	gfs2_clear_glop_pending(ip);
336 	return error;
337 }
338 
339 /**
340  * inode_go_inval - prepare a inode glock to be released
341  * @gl: the glock
342  * @flags:
343  *
344  * Normally we invalidate everything, but if we are moving into
345  * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
346  * can keep hold of the metadata, since it won't have changed.
347  *
348  */
349 
inode_go_inval(struct gfs2_glock * gl,int flags)350 static void inode_go_inval(struct gfs2_glock *gl, int flags)
351 {
352 	struct gfs2_inode *ip = gfs2_glock2inode(gl);
353 
354 	if (flags & DIO_METADATA) {
355 		struct address_space *mapping = gfs2_glock2aspace(gl);
356 		truncate_inode_pages(mapping, 0);
357 		if (ip) {
358 			set_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
359 			forget_all_cached_acls(&ip->i_inode);
360 			security_inode_invalidate_secctx(&ip->i_inode);
361 			gfs2_dir_hash_inval(ip);
362 		}
363 	}
364 
365 	if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
366 		gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
367 			       GFS2_LOG_HEAD_FLUSH_NORMAL |
368 			       GFS2_LFC_INODE_GO_INVAL);
369 		gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
370 	}
371 	if (ip && S_ISREG(ip->i_inode.i_mode))
372 		truncate_inode_pages(ip->i_inode.i_mapping, 0);
373 
374 	gfs2_clear_glop_pending(ip);
375 }
376 
377 /**
378  * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
379  * @gl: the glock
380  *
381  * Returns: 1 if it's ok
382  */
383 
inode_go_demote_ok(const struct gfs2_glock * gl)384 static int inode_go_demote_ok(const struct gfs2_glock *gl)
385 {
386 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
387 
388 	if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
389 		return 0;
390 
391 	return 1;
392 }
393 
gfs2_dinode_in(struct gfs2_inode * ip,const void * buf)394 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
395 {
396 	const struct gfs2_dinode *str = buf;
397 	struct timespec64 atime;
398 	u16 height, depth;
399 	umode_t mode = be32_to_cpu(str->di_mode);
400 	bool is_new = ip->i_inode.i_state & I_NEW;
401 
402 	if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
403 		goto corrupt;
404 	if (unlikely(!is_new && inode_wrong_type(&ip->i_inode, mode)))
405 		goto corrupt;
406 	ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
407 	ip->i_inode.i_mode = mode;
408 	if (is_new) {
409 		ip->i_inode.i_rdev = 0;
410 		switch (mode & S_IFMT) {
411 		case S_IFBLK:
412 		case S_IFCHR:
413 			ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
414 						   be32_to_cpu(str->di_minor));
415 			break;
416 		}
417 	}
418 
419 	i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
420 	i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
421 	set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
422 	i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
423 	gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
424 	atime.tv_sec = be64_to_cpu(str->di_atime);
425 	atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
426 	if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
427 		ip->i_inode.i_atime = atime;
428 	ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
429 	ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
430 	ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
431 	ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
432 
433 	ip->i_goal = be64_to_cpu(str->di_goal_meta);
434 	ip->i_generation = be64_to_cpu(str->di_generation);
435 
436 	ip->i_diskflags = be32_to_cpu(str->di_flags);
437 	ip->i_eattr = be64_to_cpu(str->di_eattr);
438 	/* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
439 	gfs2_set_inode_flags(&ip->i_inode);
440 	height = be16_to_cpu(str->di_height);
441 	if (unlikely(height > GFS2_MAX_META_HEIGHT))
442 		goto corrupt;
443 	ip->i_height = (u8)height;
444 
445 	depth = be16_to_cpu(str->di_depth);
446 	if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
447 		goto corrupt;
448 	ip->i_depth = (u8)depth;
449 	ip->i_entries = be32_to_cpu(str->di_entries);
450 
451 	if (S_ISREG(ip->i_inode.i_mode))
452 		gfs2_set_aops(&ip->i_inode);
453 
454 	return 0;
455 corrupt:
456 	gfs2_consist_inode(ip);
457 	return -EIO;
458 }
459 
460 /**
461  * gfs2_inode_refresh - Refresh the incore copy of the dinode
462  * @ip: The GFS2 inode
463  *
464  * Returns: errno
465  */
466 
gfs2_inode_refresh(struct gfs2_inode * ip)467 int gfs2_inode_refresh(struct gfs2_inode *ip)
468 {
469 	struct buffer_head *dibh;
470 	int error;
471 
472 	error = gfs2_meta_inode_buffer(ip, &dibh);
473 	if (error)
474 		return error;
475 
476 	error = gfs2_dinode_in(ip, dibh->b_data);
477 	brelse(dibh);
478 	return error;
479 }
480 
481 /**
482  * inode_go_instantiate - read in an inode if necessary
483  * @gh: The glock holder
484  *
485  * Returns: errno
486  */
487 
inode_go_instantiate(struct gfs2_glock * gl)488 static int inode_go_instantiate(struct gfs2_glock *gl)
489 {
490 	struct gfs2_inode *ip = gl->gl_object;
491 
492 	if (!ip) /* no inode to populate - read it in later */
493 		return 0;
494 
495 	return gfs2_inode_refresh(ip);
496 }
497 
inode_go_held(struct gfs2_holder * gh)498 static int inode_go_held(struct gfs2_holder *gh)
499 {
500 	struct gfs2_glock *gl = gh->gh_gl;
501 	struct gfs2_inode *ip = gl->gl_object;
502 	int error = 0;
503 
504 	if (!ip) /* no inode to populate - read it in later */
505 		return 0;
506 
507 	if (gh->gh_state != LM_ST_DEFERRED)
508 		inode_dio_wait(&ip->i_inode);
509 
510 	if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
511 	    (gl->gl_state == LM_ST_EXCLUSIVE) &&
512 	    (gh->gh_state == LM_ST_EXCLUSIVE))
513 		error = gfs2_truncatei_resume(ip);
514 
515 	return error;
516 }
517 
518 /**
519  * inode_go_dump - print information about an inode
520  * @seq: The iterator
521  * @gl: The glock
522  * @fs_id_buf: file system id (may be empty)
523  *
524  */
525 
inode_go_dump(struct seq_file * seq,struct gfs2_glock * gl,const char * fs_id_buf)526 static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
527 			  const char *fs_id_buf)
528 {
529 	struct gfs2_inode *ip = gl->gl_object;
530 	struct inode *inode = &ip->i_inode;
531 	unsigned long nrpages;
532 
533 	if (ip == NULL)
534 		return;
535 
536 	xa_lock_irq(&inode->i_data.i_pages);
537 	nrpages = inode->i_data.nrpages;
538 	xa_unlock_irq(&inode->i_data.i_pages);
539 
540 	gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
541 		       "p:%lu\n", fs_id_buf,
542 		  (unsigned long long)ip->i_no_formal_ino,
543 		  (unsigned long long)ip->i_no_addr,
544 		  IF2DT(ip->i_inode.i_mode), ip->i_flags,
545 		  (unsigned int)ip->i_diskflags,
546 		  (unsigned long long)i_size_read(inode), nrpages);
547 }
548 
549 /**
550  * freeze_go_sync - promote/demote the freeze glock
551  * @gl: the glock
552  */
553 
freeze_go_sync(struct gfs2_glock * gl)554 static int freeze_go_sync(struct gfs2_glock *gl)
555 {
556 	int error = 0;
557 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
558 
559 	/*
560 	 * We need to check gl_state == LM_ST_SHARED here and not gl_req ==
561 	 * LM_ST_EXCLUSIVE. That's because when any node does a freeze,
562 	 * all the nodes should have the freeze glock in SH mode and they all
563 	 * call do_xmote: One for EX and the others for UN. They ALL must
564 	 * freeze locally, and they ALL must queue freeze work. The freeze_work
565 	 * calls freeze_func, which tries to reacquire the freeze glock in SH,
566 	 * effectively waiting for the thaw on the node who holds it in EX.
567 	 * Once thawed, the work func acquires the freeze glock in
568 	 * SH and everybody goes back to thawed.
569 	 */
570 	if (gl->gl_state == LM_ST_SHARED && !gfs2_withdrawn(sdp) &&
571 	    !test_bit(SDF_NORECOVERY, &sdp->sd_flags)) {
572 		atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
573 		error = freeze_super(sdp->sd_vfs);
574 		if (error) {
575 			fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
576 				error);
577 			if (gfs2_withdrawn(sdp)) {
578 				atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
579 				return 0;
580 			}
581 			gfs2_assert_withdraw(sdp, 0);
582 		}
583 		queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
584 		if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
585 			gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
586 				       GFS2_LFC_FREEZE_GO_SYNC);
587 		else /* read-only mounts */
588 			atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
589 	}
590 	return 0;
591 }
592 
593 /**
594  * freeze_go_xmote_bh - After promoting/demoting the freeze glock
595  * @gl: the glock
596  */
freeze_go_xmote_bh(struct gfs2_glock * gl)597 static int freeze_go_xmote_bh(struct gfs2_glock *gl)
598 {
599 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
600 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
601 	struct gfs2_glock *j_gl = ip->i_gl;
602 	struct gfs2_log_header_host head;
603 	int error;
604 
605 	if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
606 		j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
607 
608 		error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
609 		if (gfs2_assert_withdraw_delayed(sdp, !error))
610 			return error;
611 		if (gfs2_assert_withdraw_delayed(sdp, head.lh_flags &
612 						 GFS2_LOG_HEAD_UNMOUNT))
613 			return -EIO;
614 		sdp->sd_log_sequence = head.lh_sequence + 1;
615 		gfs2_log_pointers_init(sdp, head.lh_blkno);
616 	}
617 	return 0;
618 }
619 
620 /**
621  * freeze_go_demote_ok
622  * @gl: the glock
623  *
624  * Always returns 0
625  */
626 
freeze_go_demote_ok(const struct gfs2_glock * gl)627 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
628 {
629 	return 0;
630 }
631 
632 /**
633  * iopen_go_callback - schedule the dcache entry for the inode to be deleted
634  * @gl: the glock
635  * @remote: true if this came from a different cluster node
636  *
637  * gl_lockref.lock lock is held while calling this
638  */
iopen_go_callback(struct gfs2_glock * gl,bool remote)639 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
640 {
641 	struct gfs2_inode *ip = gl->gl_object;
642 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
643 
644 	if (!remote || sb_rdonly(sdp->sd_vfs))
645 		return;
646 
647 	if (gl->gl_demote_state == LM_ST_UNLOCKED &&
648 	    gl->gl_state == LM_ST_SHARED && ip) {
649 		gl->gl_lockref.count++;
650 		if (!queue_delayed_work(gfs2_delete_workqueue,
651 					&gl->gl_delete, 0))
652 			gl->gl_lockref.count--;
653 	}
654 }
655 
iopen_go_demote_ok(const struct gfs2_glock * gl)656 static int iopen_go_demote_ok(const struct gfs2_glock *gl)
657 {
658        return !gfs2_delete_work_queued(gl);
659 }
660 
661 /**
662  * inode_go_free - wake up anyone waiting for dlm's unlock ast to free it
663  * @gl: glock being freed
664  *
665  * For now, this is only used for the journal inode glock. In withdraw
666  * situations, we need to wait for the glock to be freed so that we know
667  * other nodes may proceed with recovery / journal replay.
668  */
inode_go_free(struct gfs2_glock * gl)669 static void inode_go_free(struct gfs2_glock *gl)
670 {
671 	/* Note that we cannot reference gl_object because it's already set
672 	 * to NULL by this point in its lifecycle. */
673 	if (!test_bit(GLF_FREEING, &gl->gl_flags))
674 		return;
675 	clear_bit_unlock(GLF_FREEING, &gl->gl_flags);
676 	wake_up_bit(&gl->gl_flags, GLF_FREEING);
677 }
678 
679 /**
680  * nondisk_go_callback - used to signal when a node did a withdraw
681  * @gl: the nondisk glock
682  * @remote: true if this came from a different cluster node
683  *
684  */
nondisk_go_callback(struct gfs2_glock * gl,bool remote)685 static void nondisk_go_callback(struct gfs2_glock *gl, bool remote)
686 {
687 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
688 
689 	/* Ignore the callback unless it's from another node, and it's the
690 	   live lock. */
691 	if (!remote || gl->gl_name.ln_number != GFS2_LIVE_LOCK)
692 		return;
693 
694 	/* First order of business is to cancel the demote request. We don't
695 	 * really want to demote a nondisk glock. At best it's just to inform
696 	 * us of another node's withdraw. We'll keep it in SH mode. */
697 	clear_bit(GLF_DEMOTE, &gl->gl_flags);
698 	clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
699 
700 	/* Ignore the unlock if we're withdrawn, unmounting, or in recovery. */
701 	if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) ||
702 	    test_bit(SDF_WITHDRAWN, &sdp->sd_flags) ||
703 	    test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags))
704 		return;
705 
706 	/* We only care when a node wants us to unlock, because that means
707 	 * they want a journal recovered. */
708 	if (gl->gl_demote_state != LM_ST_UNLOCKED)
709 		return;
710 
711 	if (sdp->sd_args.ar_spectator) {
712 		fs_warn(sdp, "Spectator node cannot recover journals.\n");
713 		return;
714 	}
715 
716 	fs_warn(sdp, "Some node has withdrawn; checking for recovery.\n");
717 	set_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
718 	/*
719 	 * We can't call remote_withdraw directly here or gfs2_recover_journal
720 	 * because this is called from the glock unlock function and the
721 	 * remote_withdraw needs to enqueue and dequeue the same "live" glock
722 	 * we were called from. So we queue it to the control work queue in
723 	 * lock_dlm.
724 	 */
725 	queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
726 }
727 
728 const struct gfs2_glock_operations gfs2_meta_glops = {
729 	.go_type = LM_TYPE_META,
730 	.go_flags = GLOF_NONDISK,
731 };
732 
733 const struct gfs2_glock_operations gfs2_inode_glops = {
734 	.go_sync = inode_go_sync,
735 	.go_inval = inode_go_inval,
736 	.go_demote_ok = inode_go_demote_ok,
737 	.go_instantiate = inode_go_instantiate,
738 	.go_held = inode_go_held,
739 	.go_dump = inode_go_dump,
740 	.go_type = LM_TYPE_INODE,
741 	.go_flags = GLOF_ASPACE | GLOF_LRU | GLOF_LVB,
742 	.go_free = inode_go_free,
743 };
744 
745 const struct gfs2_glock_operations gfs2_rgrp_glops = {
746 	.go_sync = rgrp_go_sync,
747 	.go_inval = rgrp_go_inval,
748 	.go_instantiate = gfs2_rgrp_go_instantiate,
749 	.go_dump = gfs2_rgrp_go_dump,
750 	.go_type = LM_TYPE_RGRP,
751 	.go_flags = GLOF_LVB,
752 };
753 
754 const struct gfs2_glock_operations gfs2_freeze_glops = {
755 	.go_sync = freeze_go_sync,
756 	.go_xmote_bh = freeze_go_xmote_bh,
757 	.go_demote_ok = freeze_go_demote_ok,
758 	.go_type = LM_TYPE_NONDISK,
759 	.go_flags = GLOF_NONDISK,
760 };
761 
762 const struct gfs2_glock_operations gfs2_iopen_glops = {
763 	.go_type = LM_TYPE_IOPEN,
764 	.go_callback = iopen_go_callback,
765 	.go_dump = inode_go_dump,
766 	.go_demote_ok = iopen_go_demote_ok,
767 	.go_flags = GLOF_LRU | GLOF_NONDISK,
768 	.go_subclass = 1,
769 };
770 
771 const struct gfs2_glock_operations gfs2_flock_glops = {
772 	.go_type = LM_TYPE_FLOCK,
773 	.go_flags = GLOF_LRU | GLOF_NONDISK,
774 };
775 
776 const struct gfs2_glock_operations gfs2_nondisk_glops = {
777 	.go_type = LM_TYPE_NONDISK,
778 	.go_flags = GLOF_NONDISK,
779 	.go_callback = nondisk_go_callback,
780 };
781 
782 const struct gfs2_glock_operations gfs2_quota_glops = {
783 	.go_type = LM_TYPE_QUOTA,
784 	.go_flags = GLOF_LVB | GLOF_LRU | GLOF_NONDISK,
785 };
786 
787 const struct gfs2_glock_operations gfs2_journal_glops = {
788 	.go_type = LM_TYPE_JOURNAL,
789 	.go_flags = GLOF_NONDISK,
790 };
791 
792 const struct gfs2_glock_operations *gfs2_glops_list[] = {
793 	[LM_TYPE_META] = &gfs2_meta_glops,
794 	[LM_TYPE_INODE] = &gfs2_inode_glops,
795 	[LM_TYPE_RGRP] = &gfs2_rgrp_glops,
796 	[LM_TYPE_IOPEN] = &gfs2_iopen_glops,
797 	[LM_TYPE_FLOCK] = &gfs2_flock_glops,
798 	[LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
799 	[LM_TYPE_QUOTA] = &gfs2_quota_glops,
800 	[LM_TYPE_JOURNAL] = &gfs2_journal_glops,
801 };
802 
803