1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4	* Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
5	*/
6
7	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9	#include <linux/bio.h>
10	#include <linux/sched/signal.h>
11	#include <linux/slab.h>
12	#include <linux/spinlock.h>
13	#include <linux/completion.h>
14	#include <linux/buffer_head.h>
15	#include <linux/statfs.h>
16	#include <linux/seq_file.h>
17	#include <linux/mount.h>
18	#include <linux/kthread.h>
19	#include <linux/delay.h>
20	#include <linux/gfs2_ondisk.h>
21	#include <linux/crc32.h>
22	#include <linux/time.h>
23	#include <linux/wait.h>
24	#include <linux/writeback.h>
25	#include <linux/backing-dev.h>
26	#include <linux/kernel.h>
27
28	#include "gfs2.h"
29	#include "incore.h"
30	#include "bmap.h"
31	#include "dir.h"
32	#include "glock.h"
33	#include "glops.h"
34	#include "inode.h"
35	#include "log.h"
36	#include "meta_io.h"
37	#include "quota.h"
38	#include "recovery.h"
39	#include "rgrp.h"
40	#include "super.h"
41	#include "trans.h"
42	#include "util.h"
43	#include "sys.h"
44	#include "xattr.h"
45	#include "lops.h"
46
47	enum dinode_demise {
48	SHOULD_DELETE_DINODE,
49	SHOULD_NOT_DELETE_DINODE,
50	SHOULD_DEFER_EVICTION,
51	};
52
53	/**
54	* gfs2_jindex_free - Clear all the journal index information
55	* @sdp: The GFS2 superblock
56	*
57	*/
58
59	void gfs2_jindex_free(struct gfs2_sbd *sdp)
60	{
61	struct list_head list;
62	struct gfs2_jdesc *jd;
63
64	spin_lock(lock: &sdp->sd_jindex_spin);
65	list_add(new: &list, head: &sdp->sd_jindex_list);
66	list_del_init(entry: &sdp->sd_jindex_list);
67	sdp->sd_journals = 0;
68	spin_unlock(lock: &sdp->sd_jindex_spin);
69
70	sdp->sd_jdesc = NULL;
71	while (!list_empty(head: &list)) {
72	jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
73	gfs2_free_journal_extents(jd);
74	list_del(entry: &jd->jd_list);
75	iput(jd->jd_inode);
76	jd->jd_inode = NULL;
77	kfree(objp: jd);
78	}
79	}
80
81	static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
82	{
83	struct gfs2_jdesc *jd;
84
85	list_for_each_entry(jd, head, jd_list) {
86	if (jd->jd_jid == jid)
87	return jd;
88	}
89	return NULL;
90	}
91
92	struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
93	{
94	struct gfs2_jdesc *jd;
95
96	spin_lock(lock: &sdp->sd_jindex_spin);
97	jd = jdesc_find_i(head: &sdp->sd_jindex_list, jid);
98	spin_unlock(lock: &sdp->sd_jindex_spin);
99
100	return jd;
101	}
102
103	int gfs2_jdesc_check(struct gfs2_jdesc *jd)
104	{
105	struct gfs2_inode *ip = GFS2_I(inode: jd->jd_inode);
106	struct gfs2_sbd *sdp = GFS2_SB(inode: jd->jd_inode);
107	u64 size = i_size_read(inode: jd->jd_inode);
108
109	if (gfs2_check_internal_file_size(inode: jd->jd_inode, minsize: 8 << 20, BIT(30)))
110	return -EIO;
111
112	jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
113
114	if (gfs2_write_alloc_required(ip, offset: 0, len: size)) {
115	gfs2_consist_inode(ip);
116	return -EIO;
117	}
118
119	return 0;
120	}
121
122	/**
123	* gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
124	* @sdp: the filesystem
125	*
126	* Returns: errno
127	*/
128
129	int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
130	{
131	struct gfs2_inode *ip = GFS2_I(inode: sdp->sd_jdesc->jd_inode);
132	struct gfs2_glock *j_gl = ip->i_gl;
133	struct gfs2_log_header_host head;
134	int error;
135
136	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
137	if (gfs2_withdrawing_or_withdrawn(sdp))
138	return -EIO;
139
140	error = gfs2_find_jhead(jd: sdp->sd_jdesc, head: &head, keep_cache: false);
141	if (error) {
142	gfs2_consist(sdp);
143	return error;
144	}
145
146	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
147	gfs2_consist(sdp);
148	return -EIO;
149	}
150
151	/* Initialize some head of the log stuff */
152	sdp->sd_log_sequence = head.lh_sequence + 1;
153	gfs2_log_pointers_init(sdp, value: head.lh_blkno);
154
155	error = gfs2_quota_init(sdp);
156	if (!error && gfs2_withdrawing_or_withdrawn(sdp))
157	error = -EIO;
158	if (!error)
159	set_bit(nr: SDF_JOURNAL_LIVE, addr: &sdp->sd_flags);
160	return error;
161	}
162
163	void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
164	{
165	const struct gfs2_statfs_change *str = buf;
166
167	sc->sc_total = be64_to_cpu(str->sc_total);
168	sc->sc_free = be64_to_cpu(str->sc_free);
169	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
170	}
171
172	void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
173	{
174	struct gfs2_statfs_change *str = buf;
175
176	str->sc_total = cpu_to_be64(sc->sc_total);
177	str->sc_free = cpu_to_be64(sc->sc_free);
178	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
179	}
180
181	int gfs2_statfs_init(struct gfs2_sbd *sdp)
182	{
183	struct gfs2_inode *m_ip = GFS2_I(inode: sdp->sd_statfs_inode);
184	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
185	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
186	struct buffer_head *m_bh;
187	struct gfs2_holder gh;
188	int error;
189
190	error = gfs2_glock_nq_init(gl: m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
191	gh: &gh);
192	if (error)
193	return error;
194
195	error = gfs2_meta_inode_buffer(ip: m_ip, bhp: &m_bh);
196	if (error)
197	goto out;
198
199	if (sdp->sd_args.ar_spectator) {
200	spin_lock(lock: &sdp->sd_statfs_spin);
201	gfs2_statfs_change_in(sc: m_sc, buf: m_bh->b_data +
202	sizeof(struct gfs2_dinode));
203	spin_unlock(lock: &sdp->sd_statfs_spin);
204	} else {
205	spin_lock(lock: &sdp->sd_statfs_spin);
206	gfs2_statfs_change_in(sc: m_sc, buf: m_bh->b_data +
207	sizeof(struct gfs2_dinode));
208	gfs2_statfs_change_in(sc: l_sc, buf: sdp->sd_sc_bh->b_data +
209	sizeof(struct gfs2_dinode));
210	spin_unlock(lock: &sdp->sd_statfs_spin);
211
212	}
213
214	brelse(bh: m_bh);
215	out:
216	gfs2_glock_dq_uninit(gh: &gh);
217	return 0;
218	}
219
220	void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
221	s64 dinodes)
222	{
223	struct gfs2_inode *l_ip = GFS2_I(inode: sdp->sd_sc_inode);
224	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
225	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
226	s64 x, y;
227	int need_sync = 0;
228
229	gfs2_trans_add_meta(gl: l_ip->i_gl, bh: sdp->sd_sc_bh);
230
231	spin_lock(lock: &sdp->sd_statfs_spin);
232	l_sc->sc_total += total;
233	l_sc->sc_free += free;
234	l_sc->sc_dinodes += dinodes;
235	gfs2_statfs_change_out(sc: l_sc, buf: sdp->sd_sc_bh->b_data +
236	sizeof(struct gfs2_dinode));
237	if (sdp->sd_args.ar_statfs_percent) {
238	x = 100 * l_sc->sc_free;
239	y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
240	if (x >= y || x <= -y)
241	need_sync = 1;
242	}
243	spin_unlock(lock: &sdp->sd_statfs_spin);
244
245	if (need_sync)
246	gfs2_wake_up_statfs(sdp);
247	}
248
249	void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
250	{
251	struct gfs2_inode *m_ip = GFS2_I(inode: sdp->sd_statfs_inode);
252	struct gfs2_inode *l_ip = GFS2_I(inode: sdp->sd_sc_inode);
253	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
254	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
255
256	gfs2_trans_add_meta(gl: l_ip->i_gl, bh: sdp->sd_sc_bh);
257	gfs2_trans_add_meta(gl: m_ip->i_gl, bh: m_bh);
258
259	spin_lock(lock: &sdp->sd_statfs_spin);
260	m_sc->sc_total += l_sc->sc_total;
261	m_sc->sc_free += l_sc->sc_free;
262	m_sc->sc_dinodes += l_sc->sc_dinodes;
263	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
264	memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
265	0, sizeof(struct gfs2_statfs_change));
266	gfs2_statfs_change_out(sc: m_sc, buf: m_bh->b_data + sizeof(struct gfs2_dinode));
267	spin_unlock(lock: &sdp->sd_statfs_spin);
268	}
269
270	int gfs2_statfs_sync(struct super_block *sb, int type)
271	{
272	struct gfs2_sbd *sdp = sb->s_fs_info;
273	struct gfs2_inode *m_ip = GFS2_I(inode: sdp->sd_statfs_inode);
274	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
275	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
276	struct gfs2_holder gh;
277	struct buffer_head *m_bh;
278	int error;
279
280	error = gfs2_glock_nq_init(gl: m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
281	gh: &gh);
282	if (error)
283	goto out;
284
285	error = gfs2_meta_inode_buffer(ip: m_ip, bhp: &m_bh);
286	if (error)
287	goto out_unlock;
288
289	spin_lock(lock: &sdp->sd_statfs_spin);
290	gfs2_statfs_change_in(sc: m_sc, buf: m_bh->b_data +
291	sizeof(struct gfs2_dinode));
292	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
293	spin_unlock(lock: &sdp->sd_statfs_spin);
294	goto out_bh;
295	}
296	spin_unlock(lock: &sdp->sd_statfs_spin);
297
298	error = gfs2_trans_begin(sdp, blocks: 2 * RES_DINODE, revokes: 0);
299	if (error)
300	goto out_bh;
301
302	update_statfs(sdp, m_bh);
303	sdp->sd_statfs_force_sync = 0;
304
305	gfs2_trans_end(sdp);
306
307	out_bh:
308	brelse(bh: m_bh);
309	out_unlock:
310	gfs2_glock_dq_uninit(gh: &gh);
311	out:
312	return error;
313	}
314
315	struct lfcc {
316	struct list_head list;
317	struct gfs2_holder gh;
318	};
319
320	/**
321	* gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
322	* journals are clean
323	* @sdp: the file system
324	*
325	* Returns: errno
326	*/
327
328	static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
329	{
330	struct gfs2_inode *ip;
331	struct gfs2_jdesc *jd;
332	struct lfcc *lfcc;
333	LIST_HEAD(list);
334	struct gfs2_log_header_host lh;
335	int error, error2;
336
337	/*
338	* Grab all the journal glocks in SH mode. We are *probably* doing
339	* that to prevent recovery.
340	*/
341
342	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
343	lfcc = kmalloc(size: sizeof(struct lfcc), GFP_KERNEL);
344	if (!lfcc) {
345	error = -ENOMEM;
346	goto out;
347	}
348	ip = GFS2_I(inode: jd->jd_inode);
349	error = gfs2_glock_nq_init(gl: ip->i_gl, LM_ST_SHARED, flags: 0, gh: &lfcc->gh);
350	if (error) {
351	kfree(objp: lfcc);
352	goto out;
353	}
354	list_add(new: &lfcc->list, head: &list);
355	}
356
357	gfs2_freeze_unlock(freeze_gh: &sdp->sd_freeze_gh);
358
359	error = gfs2_glock_nq_init(gl: sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
360	LM_FLAG_NOEXP | GL_NOPID,
361	gh: &sdp->sd_freeze_gh);
362	if (error)
363	goto relock_shared;
364
365	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
366	error = gfs2_jdesc_check(jd);
367	if (error)
368	break;
369	error = gfs2_find_jhead(jd, head: &lh, keep_cache: false);
370	if (error)
371	break;
372	if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
373	error = -EBUSY;
374	break;
375	}
376	}
377
378	if (!error)
379	goto out; /* success */
380
381	gfs2_freeze_unlock(freeze_gh: &sdp->sd_freeze_gh);
382
383	relock_shared:
384	error2 = gfs2_freeze_lock_shared(sdp);
385	gfs2_assert_withdraw(sdp, !error2);
386
387	out:
388	while (!list_empty(head: &list)) {
389	lfcc = list_first_entry(&list, struct lfcc, list);
390	list_del(entry: &lfcc->list);
391	gfs2_glock_dq_uninit(gh: &lfcc->gh);
392	kfree(objp: lfcc);
393	}
394	return error;
395	}
396
397	void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
398	{
399	const struct inode *inode = &ip->i_inode;
400	struct gfs2_dinode *str = buf;
401
402	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
403	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
404	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
405	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
406	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
407	str->di_mode = cpu_to_be32(inode->i_mode);
408	str->di_uid = cpu_to_be32(i_uid_read(inode));
409	str->di_gid = cpu_to_be32(i_gid_read(inode));
410	str->di_nlink = cpu_to_be32(inode->i_nlink);
411	str->di_size = cpu_to_be64(i_size_read(inode));
412	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
413	str->di_atime = cpu_to_be64(inode_get_atime_sec(inode));
414	str->di_mtime = cpu_to_be64(inode_get_mtime_sec(inode));
415	str->di_ctime = cpu_to_be64(inode_get_ctime_sec(inode));
416
417	str->di_goal_meta = cpu_to_be64(ip->i_goal);
418	str->di_goal_data = cpu_to_be64(ip->i_goal);
419	str->di_generation = cpu_to_be64(ip->i_generation);
420
421	str->di_flags = cpu_to_be32(ip->i_diskflags);
422	str->di_height = cpu_to_be16(ip->i_height);
423	str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
424	!(ip->i_diskflags & GFS2_DIF_EXHASH) ?
425	GFS2_FORMAT_DE : 0);
426	str->di_depth = cpu_to_be16(ip->i_depth);
427	str->di_entries = cpu_to_be32(ip->i_entries);
428
429	str->di_eattr = cpu_to_be64(ip->i_eattr);
430	str->di_atime_nsec = cpu_to_be32(inode_get_atime_nsec(inode));
431	str->di_mtime_nsec = cpu_to_be32(inode_get_mtime_nsec(inode));
432	str->di_ctime_nsec = cpu_to_be32(inode_get_ctime_nsec(inode));
433	}
434
435	/**
436	* gfs2_write_inode - Make sure the inode is stable on the disk
437	* @inode: The inode
438	* @wbc: The writeback control structure
439	*
440	* Returns: errno
441	*/
442
443	static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
444	{
445	struct gfs2_inode *ip = GFS2_I(inode);
446	struct gfs2_sbd *sdp = GFS2_SB(inode);
447	struct address_space *metamapping = gfs2_glock2aspace(gl: ip->i_gl);
448	struct backing_dev_info *bdi = inode_to_bdi(inode: metamapping->host);
449	int ret = 0;
450	bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
451
452	if (flush_all)
453	gfs2_log_flush(sdp: GFS2_SB(inode), gl: ip->i_gl,
454	GFS2_LOG_HEAD_FLUSH_NORMAL |
455	GFS2_LFC_WRITE_INODE);
456	if (bdi->wb.dirty_exceeded)
457	gfs2_ail1_flush(sdp, wbc);
458	else
459	filemap_fdatawrite(metamapping);
460	if (flush_all)
461	ret = filemap_fdatawait(mapping: metamapping);
462	if (ret)
463	mark_inode_dirty_sync(inode);
464	else {
465	spin_lock(lock: &inode->i_lock);
466	if (!(inode->i_flags & I_DIRTY))
467	gfs2_ordered_del_inode(ip);
468	spin_unlock(lock: &inode->i_lock);
469	}
470	return ret;
471	}
472
473	/**
474	* gfs2_dirty_inode - check for atime updates
475	* @inode: The inode in question
476	* @flags: The type of dirty
477	*
478	* Unfortunately it can be called under any combination of inode
479	* glock and freeze glock, so we have to check carefully.
480	*
481	* At the moment this deals only with atime - it should be possible
482	* to expand that role in future, once a review of the locking has
483	* been carried out.
484	*/
485
486	static void gfs2_dirty_inode(struct inode *inode, int flags)
487	{
488	struct gfs2_inode *ip = GFS2_I(inode);
489	struct gfs2_sbd *sdp = GFS2_SB(inode);
490	struct buffer_head *bh;
491	struct gfs2_holder gh;
492	int need_unlock = 0;
493	int need_endtrans = 0;
494	int ret;
495
496	if (unlikely(!ip->i_gl)) {
497	/* This can only happen during incomplete inode creation. */
498	BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
499	return;
500	}
501
502	if (gfs2_withdrawing_or_withdrawn(sdp))
503	return;
504	if (!gfs2_glock_is_locked_by_me(gl: ip->i_gl)) {
505	ret = gfs2_glock_nq_init(gl: ip->i_gl, LM_ST_EXCLUSIVE, flags: 0, gh: &gh);
506	if (ret) {
507	fs_err(sdp, "dirty_inode: glock %d\n", ret);
508	gfs2_dump_glock(NULL, gl: ip->i_gl, fsid: true);
509	return;
510	}
511	need_unlock = 1;
512	} else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
513	return;
514
515	if (current->journal_info == NULL) {
516	ret = gfs2_trans_begin(sdp, RES_DINODE, revokes: 0);
517	if (ret) {
518	fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
519	goto out;
520	}
521	need_endtrans = 1;
522	}
523
524	ret = gfs2_meta_inode_buffer(ip, bhp: &bh);
525	if (ret == 0) {
526	gfs2_trans_add_meta(gl: ip->i_gl, bh);
527	gfs2_dinode_out(ip, buf: bh->b_data);
528	brelse(bh);
529	}
530
531	if (need_endtrans)
532	gfs2_trans_end(sdp);
533	out:
534	if (need_unlock)
535	gfs2_glock_dq_uninit(gh: &gh);
536	}
537
538	/**
539	* gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
540	* @sdp: the filesystem
541	*
542	* Returns: errno
543	*/
544
545	void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
546	{
547	int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
548
549	if (!test_bit(SDF_KILL, &sdp->sd_flags))
550	gfs2_flush_delete_work(sdp);
551
552	gfs2_destroy_threads(sdp);
553
554	if (log_write_allowed) {
555	gfs2_quota_sync(sb: sdp->sd_vfs, type: 0);
556	gfs2_statfs_sync(sb: sdp->sd_vfs, type: 0);
557
558	/* We do two log flushes here. The first one commits dirty inodes
559	* and rgrps to the journal, but queues up revokes to the ail list.
560	* The second flush writes out and removes the revokes.
561	*
562	* The first must be done before the FLUSH_SHUTDOWN code
563	* clears the LIVE flag, otherwise it will not be able to start
564	* a transaction to write its revokes, and the error will cause
565	* a withdraw of the file system. */
566	gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO);
567	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
568	GFS2_LFC_MAKE_FS_RO);
569	wait_event_timeout(sdp->sd_log_waitq,
570	gfs2_log_is_empty(sdp),
571	HZ * 5);
572	gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
573	}
574	gfs2_quota_cleanup(sdp);
575	}
576
577	/**
578	* gfs2_put_super - Unmount the filesystem
579	* @sb: The VFS superblock
580	*
581	*/
582
583	static void gfs2_put_super(struct super_block *sb)
584	{
585	struct gfs2_sbd *sdp = sb->s_fs_info;
586	struct gfs2_jdesc *jd;
587
588	/* No more recovery requests */
589	set_bit(nr: SDF_NORECOVERY, addr: &sdp->sd_flags);
590	smp_mb();
591
592	/* Wait on outstanding recovery */
593	restart:
594	spin_lock(lock: &sdp->sd_jindex_spin);
595	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
596	if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
597	continue;
598	spin_unlock(lock: &sdp->sd_jindex_spin);
599	wait_on_bit(word: &jd->jd_flags, JDF_RECOVERY,
600	TASK_UNINTERRUPTIBLE);
601	goto restart;
602	}
603	spin_unlock(lock: &sdp->sd_jindex_spin);
604
605	if (!sb_rdonly(sb))
606	gfs2_make_fs_ro(sdp);
607	else {
608	if (gfs2_withdrawing_or_withdrawn(sdp))
609	gfs2_destroy_threads(sdp);
610
611	gfs2_quota_cleanup(sdp);
612	}
613
614	WARN_ON(gfs2_withdrawing(sdp));
615
616	/* At this point, we're through modifying the disk */
617
618	/* Release stuff */
619
620	gfs2_freeze_unlock(freeze_gh: &sdp->sd_freeze_gh);
621
622	iput(sdp->sd_jindex);
623	iput(sdp->sd_statfs_inode);
624	iput(sdp->sd_rindex);
625	iput(sdp->sd_quota_inode);
626
627	gfs2_glock_put(gl: sdp->sd_rename_gl);
628	gfs2_glock_put(gl: sdp->sd_freeze_gl);
629
630	if (!sdp->sd_args.ar_spectator) {
631	if (gfs2_holder_initialized(gh: &sdp->sd_journal_gh))
632	gfs2_glock_dq_uninit(gh: &sdp->sd_journal_gh);
633	if (gfs2_holder_initialized(gh: &sdp->sd_jinode_gh))
634	gfs2_glock_dq_uninit(gh: &sdp->sd_jinode_gh);
635	brelse(bh: sdp->sd_sc_bh);
636	gfs2_glock_dq_uninit(gh: &sdp->sd_sc_gh);
637	gfs2_glock_dq_uninit(gh: &sdp->sd_qc_gh);
638	free_local_statfs_inodes(sdp);
639	iput(sdp->sd_qc_inode);
640	}
641
642	gfs2_glock_dq_uninit(gh: &sdp->sd_live_gh);
643	gfs2_clear_rgrpd(sdp);
644	gfs2_jindex_free(sdp);
645	/* Take apart glock structures and buffer lists */
646	gfs2_gl_hash_clear(sdp);
647	truncate_inode_pages_final(&sdp->sd_aspace);
648	gfs2_delete_debugfs_file(sdp);
649	/* Unmount the locking protocol */
650	gfs2_lm_unmount(sdp);
651
652	/* At this point, we're through participating in the lockspace */
653	gfs2_sys_fs_del(sdp);
654	free_sbd(sdp);
655	}
656
657	/**
658	* gfs2_sync_fs - sync the filesystem
659	* @sb: the superblock
660	* @wait: true to wait for completion
661	*
662	* Flushes the log to disk.
663	*/
664
665	static int gfs2_sync_fs(struct super_block *sb, int wait)
666	{
667	struct gfs2_sbd *sdp = sb->s_fs_info;
668
669	gfs2_quota_sync(sb, type: -1);
670	if (wait)
671	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
672	GFS2_LFC_SYNC_FS);
673	return sdp->sd_log_error;
674	}
675
676	static int gfs2_do_thaw(struct gfs2_sbd *sdp)
677	{
678	struct super_block *sb = sdp->sd_vfs;
679	int error;
680
681	error = gfs2_freeze_lock_shared(sdp);
682	if (error)
683	goto fail;
684	error = thaw_super(super: sb, who: FREEZE_HOLDER_USERSPACE);
685	if (!error)
686	return 0;
687
688	fail:
689	fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error);
690	gfs2_assert_withdraw(sdp, 0);
691	return error;
692	}
693
694	void gfs2_freeze_func(struct work_struct *work)
695	{
696	struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
697	struct super_block *sb = sdp->sd_vfs;
698	int error;
699
700	mutex_lock(&sdp->sd_freeze_mutex);
701	error = -EBUSY;
702	if (test_bit(SDF_FROZEN, &sdp->sd_flags))
703	goto freeze_failed;
704
705	error = freeze_super(super: sb, who: FREEZE_HOLDER_USERSPACE);
706	if (error)
707	goto freeze_failed;
708
709	gfs2_freeze_unlock(freeze_gh: &sdp->sd_freeze_gh);
710	set_bit(nr: SDF_FROZEN, addr: &sdp->sd_flags);
711
712	error = gfs2_do_thaw(sdp);
713	if (error)
714	goto out;
715
716	clear_bit(nr: SDF_FROZEN, addr: &sdp->sd_flags);
717	goto out;
718
719	freeze_failed:
720	fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error);
721
722	out:
723	mutex_unlock(lock: &sdp->sd_freeze_mutex);
724	deactivate_super(sb);
725	}
726
727	/**
728	* gfs2_freeze_super - prevent further writes to the filesystem
729	* @sb: the VFS structure for the filesystem
730	*
731	*/
732
733	static int gfs2_freeze_super(struct super_block *sb, enum freeze_holder who)
734	{
735	struct gfs2_sbd *sdp = sb->s_fs_info;
736	int error;
737
738	if (!mutex_trylock(lock: &sdp->sd_freeze_mutex))
739	return -EBUSY;
740	if (test_bit(SDF_FROZEN, &sdp->sd_flags)) {
741	mutex_unlock(lock: &sdp->sd_freeze_mutex);
742	return -EBUSY;
743	}
744
745	for (;;) {
746	error = freeze_super(super: sb, who: FREEZE_HOLDER_USERSPACE);
747	if (error) {
748	fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
749	error);
750	goto out;
751	}
752
753	error = gfs2_lock_fs_check_clean(sdp);
754	if (!error) {
755	set_bit(nr: SDF_FREEZE_INITIATOR, addr: &sdp->sd_flags);
756	set_bit(nr: SDF_FROZEN, addr: &sdp->sd_flags);
757	break;
758	}
759
760	error = gfs2_do_thaw(sdp);
761	if (error)
762	goto out;
763
764	if (error == -EBUSY)
765	fs_err(sdp, "waiting for recovery before freeze\n");
766	else if (error == -EIO) {
767	fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
768	"to recovery error.\n");
769	goto out;
770	} else {
771	fs_err(sdp, "error freezing FS: %d\n", error);
772	}
773	fs_err(sdp, "retrying...\n");
774	msleep(msecs: 1000);
775	}
776
777	out:
778	mutex_unlock(lock: &sdp->sd_freeze_mutex);
779	return error;
780	}
781
782	static int gfs2_freeze_fs(struct super_block *sb)
783	{
784	struct gfs2_sbd *sdp = sb->s_fs_info;
785
786	if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
787	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
788	GFS2_LFC_FREEZE_GO_SYNC);
789	if (gfs2_withdrawing_or_withdrawn(sdp))
790	return -EIO;
791	}
792	return 0;
793	}
794
795	/**
796	* gfs2_thaw_super - reallow writes to the filesystem
797	* @sb: the VFS structure for the filesystem
798	*
799	*/
800
801	static int gfs2_thaw_super(struct super_block *sb, enum freeze_holder who)
802	{
803	struct gfs2_sbd *sdp = sb->s_fs_info;
804	int error;
805
806	if (!mutex_trylock(lock: &sdp->sd_freeze_mutex))
807	return -EBUSY;
808	if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) {
809	mutex_unlock(lock: &sdp->sd_freeze_mutex);
810	return -EINVAL;
811	}
812
813	atomic_inc(v: &sb->s_active);
814	gfs2_freeze_unlock(freeze_gh: &sdp->sd_freeze_gh);
815
816	error = gfs2_do_thaw(sdp);
817
818	if (!error) {
819	clear_bit(nr: SDF_FREEZE_INITIATOR, addr: &sdp->sd_flags);
820	clear_bit(nr: SDF_FROZEN, addr: &sdp->sd_flags);
821	}
822	mutex_unlock(lock: &sdp->sd_freeze_mutex);
823	deactivate_super(sb);
824	return error;
825	}
826
827	void gfs2_thaw_freeze_initiator(struct super_block *sb)
828	{
829	struct gfs2_sbd *sdp = sb->s_fs_info;
830
831	mutex_lock(&sdp->sd_freeze_mutex);
832	if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
833	goto out;
834
835	gfs2_freeze_unlock(freeze_gh: &sdp->sd_freeze_gh);
836
837	out:
838	mutex_unlock(lock: &sdp->sd_freeze_mutex);
839	}
840
841	/**
842	* statfs_slow_fill - fill in the sg for a given RG
843	* @rgd: the RG
844	* @sc: the sc structure
845	*
846	* Returns: 0 on success, -ESTALE if the LVB is invalid
847	*/
848
849	static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
850	struct gfs2_statfs_change_host *sc)
851	{
852	gfs2_rgrp_verify(rgd);
853	sc->sc_total += rgd->rd_data;
854	sc->sc_free += rgd->rd_free;
855	sc->sc_dinodes += rgd->rd_dinodes;
856	return 0;
857	}
858
859	/**
860	* gfs2_statfs_slow - Stat a filesystem using asynchronous locking
861	* @sdp: the filesystem
862	* @sc: the sc info that will be returned
863	*
864	* Any error (other than a signal) will cause this routine to fall back
865	* to the synchronous version.
866	*
867	* FIXME: This really shouldn't busy wait like this.
868	*
869	* Returns: errno
870	*/
871
872	static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
873	{
874	struct gfs2_rgrpd *rgd_next;
875	struct gfs2_holder *gha, *gh;
876	unsigned int slots = 64;
877	unsigned int x;
878	int done;
879	int error = 0, err;
880
881	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
882	gha = kmalloc_array(n: slots, size: sizeof(struct gfs2_holder), GFP_KERNEL);
883	if (!gha)
884	return -ENOMEM;
885	for (x = 0; x < slots; x++)
886	gfs2_holder_mark_uninitialized(gh: gha + x);
887
888	rgd_next = gfs2_rgrpd_get_first(sdp);
889
890	for (;;) {
891	done = 1;
892
893	for (x = 0; x < slots; x++) {
894	gh = gha + x;
895
896	if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
897	err = gfs2_glock_wait(gh);
898	if (err) {
899	gfs2_holder_uninit(gh);
900	error = err;
901	} else {
902	if (!error) {
903	struct gfs2_rgrpd *rgd =
904	gfs2_glock2rgrp(gl: gh->gh_gl);
905
906	error = statfs_slow_fill(rgd, sc);
907	}
908	gfs2_glock_dq_uninit(gh);
909	}
910	}
911
912	if (gfs2_holder_initialized(gh))
913	done = 0;
914	else if (rgd_next && !error) {
915	error = gfs2_glock_nq_init(gl: rgd_next->rd_gl,
916	LM_ST_SHARED,
917	GL_ASYNC,
918	gh);
919	rgd_next = gfs2_rgrpd_get_next(rgd: rgd_next);
920	done = 0;
921	}
922
923	if (signal_pending(current))
924	error = -ERESTARTSYS;
925	}
926
927	if (done)
928	break;
929
930	yield();
931	}
932
933	kfree(objp: gha);
934	return error;
935	}
936
937	/**
938	* gfs2_statfs_i - Do a statfs
939	* @sdp: the filesystem
940	* @sc: the sc structure
941	*
942	* Returns: errno
943	*/
944
945	static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
946	{
947	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
948	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
949
950	spin_lock(lock: &sdp->sd_statfs_spin);
951
952	*sc = *m_sc;
953	sc->sc_total += l_sc->sc_total;
954	sc->sc_free += l_sc->sc_free;
955	sc->sc_dinodes += l_sc->sc_dinodes;
956
957	spin_unlock(lock: &sdp->sd_statfs_spin);
958
959	if (sc->sc_free < 0)
960	sc->sc_free = 0;
961	if (sc->sc_free > sc->sc_total)
962	sc->sc_free = sc->sc_total;
963	if (sc->sc_dinodes < 0)
964	sc->sc_dinodes = 0;
965
966	return 0;
967	}
968
969	/**
970	* gfs2_statfs - Gather and return stats about the filesystem
971	* @dentry: The name of the link
972	* @buf: The buffer
973	*
974	* Returns: 0 on success or error code
975	*/
976
977	static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
978	{
979	struct super_block *sb = dentry->d_sb;
980	struct gfs2_sbd *sdp = sb->s_fs_info;
981	struct gfs2_statfs_change_host sc;
982	int error;
983
984	error = gfs2_rindex_update(sdp);
985	if (error)
986	return error;
987
988	if (gfs2_tune_get(sdp, gt_statfs_slow))
989	error = gfs2_statfs_slow(sdp, sc: &sc);
990	else
991	error = gfs2_statfs_i(sdp, sc: &sc);
992
993	if (error)
994	return error;
995
996	buf->f_type = GFS2_MAGIC;
997	buf->f_bsize = sdp->sd_sb.sb_bsize;
998	buf->f_blocks = sc.sc_total;
999	buf->f_bfree = sc.sc_free;
1000	buf->f_bavail = sc.sc_free;
1001	buf->f_files = sc.sc_dinodes + sc.sc_free;
1002	buf->f_ffree = sc.sc_free;
1003	buf->f_namelen = GFS2_FNAMESIZE;
1004	buf->f_fsid = uuid_to_fsid(uuid: sb->s_uuid.b);
1005
1006	return 0;
1007	}
1008
1009	/**
1010	* gfs2_drop_inode - Drop an inode (test for remote unlink)
1011	* @inode: The inode to drop
1012	*
1013	* If we've received a callback on an iopen lock then it's because a
1014	* remote node tried to deallocate the inode but failed due to this node
1015	* still having the inode open. Here we mark the link count zero
1016	* since we know that it must have reached zero if the GLF_DEMOTE flag
1017	* is set on the iopen glock. If we didn't do a disk read since the
1018	* remote node removed the final link then we might otherwise miss
1019	* this event. This check ensures that this node will deallocate the
1020	* inode's blocks, or alternatively pass the baton on to another
1021	* node for later deallocation.
1022	*/
1023
1024	static int gfs2_drop_inode(struct inode *inode)
1025	{
1026	struct gfs2_inode *ip = GFS2_I(inode);
1027	struct gfs2_sbd *sdp = GFS2_SB(inode);
1028
1029	if (inode->i_nlink &&
1030	gfs2_holder_initialized(gh: &ip->i_iopen_gh)) {
1031	struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1032	if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1033	clear_nlink(inode);
1034	}
1035
1036	/*
1037	* When under memory pressure when an inode's link count has dropped to
1038	* zero, defer deleting the inode to the delete workqueue. This avoids
1039	* calling into DLM under memory pressure, which can deadlock.
1040	*/
1041	if (!inode->i_nlink &&
1042	unlikely(current->flags & PF_MEMALLOC) &&
1043	gfs2_holder_initialized(gh: &ip->i_iopen_gh)) {
1044	struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1045
1046	gfs2_glock_hold(gl);
1047	if (!gfs2_queue_try_to_evict(gl))
1048	gfs2_glock_queue_put(gl);
1049	return 0;
1050	}
1051
1052	/*
1053	* No longer cache inodes when trying to evict them all.
1054	*/
1055	if (test_bit(SDF_EVICTING, &sdp->sd_flags))
1056	return 1;
1057
1058	return generic_drop_inode(inode);
1059	}
1060
1061	/**
1062	* gfs2_show_options - Show mount options for /proc/mounts
1063	* @s: seq_file structure
1064	* @root: root of this (sub)tree
1065	*
1066	* Returns: 0 on success or error code
1067	*/
1068
1069	static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1070	{
1071	struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1072	struct gfs2_args *args = &sdp->sd_args;
1073	unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1074
1075	spin_lock(lock: &sdp->sd_tune.gt_spin);
1076	logd_secs = sdp->sd_tune.gt_logd_secs;
1077	quota_quantum = sdp->sd_tune.gt_quota_quantum;
1078	statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1079	statfs_slow = sdp->sd_tune.gt_statfs_slow;
1080	spin_unlock(lock: &sdp->sd_tune.gt_spin);
1081
1082	if (is_subdir(root, sdp->sd_master_dir))
1083	seq_puts(m: s, s: ",meta");
1084	if (args->ar_lockproto[0])
1085	seq_show_option(m: s, name: "lockproto", value: args->ar_lockproto);
1086	if (args->ar_locktable[0])
1087	seq_show_option(m: s, name: "locktable", value: args->ar_locktable);
1088	if (args->ar_hostdata[0])
1089	seq_show_option(m: s, name: "hostdata", value: args->ar_hostdata);
1090	if (args->ar_spectator)
1091	seq_puts(m: s, s: ",spectator");
1092	if (args->ar_localflocks)
1093	seq_puts(m: s, s: ",localflocks");
1094	if (args->ar_debug)
1095	seq_puts(m: s, s: ",debug");
1096	if (args->ar_posix_acl)
1097	seq_puts(m: s, s: ",acl");
1098	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1099	char *state;
1100	switch (args->ar_quota) {
1101	case GFS2_QUOTA_OFF:
1102	state = "off";
1103	break;
1104	case GFS2_QUOTA_ACCOUNT:
1105	state = "account";
1106	break;
1107	case GFS2_QUOTA_ON:
1108	state = "on";
1109	break;
1110	case GFS2_QUOTA_QUIET:
1111	state = "quiet";
1112	break;
1113	default:
1114	state = "unknown";
1115	break;
1116	}
1117	seq_printf(m: s, fmt: ",quota=%s", state);
1118	}
1119	if (args->ar_suiddir)
1120	seq_puts(m: s, s: ",suiddir");
1121	if (args->ar_data != GFS2_DATA_DEFAULT) {
1122	char *state;
1123	switch (args->ar_data) {
1124	case GFS2_DATA_WRITEBACK:
1125	state = "writeback";
1126	break;
1127	case GFS2_DATA_ORDERED:
1128	state = "ordered";
1129	break;
1130	default:
1131	state = "unknown";
1132	break;
1133	}
1134	seq_printf(m: s, fmt: ",data=%s", state);
1135	}
1136	if (args->ar_discard)
1137	seq_puts(m: s, s: ",discard");
1138	if (logd_secs != 30)
1139	seq_printf(m: s, fmt: ",commit=%d", logd_secs);
1140	if (statfs_quantum != 30)
1141	seq_printf(m: s, fmt: ",statfs_quantum=%d", statfs_quantum);
1142	else if (statfs_slow)
1143	seq_puts(m: s, s: ",statfs_quantum=0");
1144	if (quota_quantum != 60)
1145	seq_printf(m: s, fmt: ",quota_quantum=%d", quota_quantum);
1146	if (args->ar_statfs_percent)
1147	seq_printf(m: s, fmt: ",statfs_percent=%d", args->ar_statfs_percent);
1148	if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1149	const char *state;
1150
1151	switch (args->ar_errors) {
1152	case GFS2_ERRORS_WITHDRAW:
1153	state = "withdraw";
1154	break;
1155	case GFS2_ERRORS_PANIC:
1156	state = "panic";
1157	break;
1158	default:
1159	state = "unknown";
1160	break;
1161	}
1162	seq_printf(m: s, fmt: ",errors=%s", state);
1163	}
1164	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1165	seq_puts(m: s, s: ",nobarrier");
1166	if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1167	seq_puts(m: s, s: ",demote_interface_used");
1168	if (args->ar_rgrplvb)
1169	seq_puts(m: s, s: ",rgrplvb");
1170	if (args->ar_loccookie)
1171	seq_puts(m: s, s: ",loccookie");
1172	return 0;
1173	}
1174
1175	static void gfs2_final_release_pages(struct gfs2_inode *ip)
1176	{
1177	struct inode *inode = &ip->i_inode;
1178	struct gfs2_glock *gl = ip->i_gl;
1179
1180	if (unlikely(!gl)) {
1181	/* This can only happen during incomplete inode creation. */
1182	BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1183	return;
1184	}
1185
1186	truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1187	truncate_inode_pages(&inode->i_data, 0);
1188
1189	if (atomic_read(v: &gl->gl_revokes) == 0) {
1190	clear_bit(nr: GLF_LFLUSH, addr: &gl->gl_flags);
1191	clear_bit(nr: GLF_DIRTY, addr: &gl->gl_flags);
1192	}
1193	}
1194
1195	static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1196	{
1197	struct gfs2_sbd *sdp = GFS2_SB(inode: &ip->i_inode);
1198	struct gfs2_rgrpd *rgd;
1199	struct gfs2_holder gh;
1200	int error;
1201
1202	if (gfs2_get_inode_blocks(inode: &ip->i_inode) != 1) {
1203	gfs2_consist_inode(ip);
1204	return -EIO;
1205	}
1206
1207	gfs2_rindex_update(sdp);
1208
1209	error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1210	if (error)
1211	return error;
1212
1213	rgd = gfs2_blk2rgrpd(sdp, blk: ip->i_no_addr, exact: 1);
1214	if (!rgd) {
1215	gfs2_consist_inode(ip);
1216	error = -EIO;
1217	goto out_qs;
1218	}
1219
1220	error = gfs2_glock_nq_init(gl: rgd->rd_gl, LM_ST_EXCLUSIVE,
1221	LM_FLAG_NODE_SCOPE, gh: &gh);
1222	if (error)
1223	goto out_qs;
1224
1225	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1226	revokes: sdp->sd_jdesc->jd_blocks);
1227	if (error)
1228	goto out_rg_gunlock;
1229
1230	gfs2_free_di(rgd, ip);
1231
1232	gfs2_final_release_pages(ip);
1233
1234	gfs2_trans_end(sdp);
1235
1236	out_rg_gunlock:
1237	gfs2_glock_dq_uninit(gh: &gh);
1238	out_qs:
1239	gfs2_quota_unhold(ip);
1240	return error;
1241	}
1242
1243	/**
1244	* gfs2_glock_put_eventually
1245	* @gl: The glock to put
1246	*
1247	* When under memory pressure, trigger a deferred glock put to make sure we
1248	* won't call into DLM and deadlock. Otherwise, put the glock directly.
1249	*/
1250
1251	static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1252	{
1253	if (current->flags & PF_MEMALLOC)
1254	gfs2_glock_queue_put(gl);
1255	else
1256	gfs2_glock_put(gl);
1257	}
1258
1259	static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1260	{
1261	struct gfs2_inode *ip = GFS2_I(inode);
1262	struct gfs2_sbd *sdp = GFS2_SB(inode);
1263	struct gfs2_holder *gh = &ip->i_iopen_gh;
1264	long timeout = 5 * HZ;
1265	int error;
1266
1267	gh->gh_flags |= GL_NOCACHE;
1268	gfs2_glock_dq_wait(gh);
1269
1270	/*
1271	* If there are no other lock holders, we will immediately get
1272	* exclusive access to the iopen glock here.
1273	*
1274	* Otherwise, the other nodes holding the lock will be notified about
1275	* our locking request. If they do not have the inode open, they are
1276	* expected to evict the cached inode and release the lock, allowing us
1277	* to proceed.
1278	*
1279	* Otherwise, if they cannot evict the inode, they are expected to poke
1280	* the inode glock (note: not the iopen glock). We will notice that
1281	* and stop waiting for the iopen glock immediately. The other node(s)
1282	* are then expected to take care of deleting the inode when they no
1283	* longer use it.
1284	*
1285	* As a last resort, if another node keeps holding the iopen glock
1286	* without showing any activity on the inode glock, we will eventually
1287	* time out and fail the iopen glock upgrade.
1288	*/
1289
1290	gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1291	error = gfs2_glock_nq(gh);
1292	if (error)
1293	return false;
1294
1295	timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1296	!test_bit(HIF_WAIT, &gh->gh_iflags) ||
1297	test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1298	timeout);
1299	if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1300	gfs2_glock_dq(gh);
1301	return false;
1302	}
1303	return gfs2_glock_holder_ready(gh) == 0;
1304	}
1305
1306	/**
1307	* evict_should_delete - determine whether the inode is eligible for deletion
1308	* @inode: The inode to evict
1309	* @gh: The glock holder structure
1310	*
1311	* This function determines whether the evicted inode is eligible to be deleted
1312	* and locks the inode glock.
1313	*
1314	* Returns: the fate of the dinode
1315	*/
1316	static enum dinode_demise evict_should_delete(struct inode *inode,
1317	struct gfs2_holder *gh)
1318	{
1319	struct gfs2_inode *ip = GFS2_I(inode);
1320	struct super_block *sb = inode->i_sb;
1321	struct gfs2_sbd *sdp = sb->s_fs_info;
1322	int ret;
1323
1324	if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1325	goto should_delete;
1326
1327	if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1328	return SHOULD_DEFER_EVICTION;
1329
1330	/* Deletes should never happen under memory pressure anymore. */
1331	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1332	return SHOULD_DEFER_EVICTION;
1333
1334	/* Must not read inode block until block type has been verified */
1335	ret = gfs2_glock_nq_init(gl: ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1336	if (unlikely(ret)) {
1337	glock_clear_object(gl: ip->i_iopen_gh.gh_gl, object: ip);
1338	ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1339	gfs2_glock_dq_uninit(gh: &ip->i_iopen_gh);
1340	return SHOULD_DEFER_EVICTION;
1341	}
1342
1343	if (gfs2_inode_already_deleted(gl: ip->i_gl, generation: ip->i_no_formal_ino))
1344	return SHOULD_NOT_DELETE_DINODE;
1345	ret = gfs2_check_blk_type(sdp, no_addr: ip->i_no_addr, GFS2_BLKST_UNLINKED);
1346	if (ret)
1347	return SHOULD_NOT_DELETE_DINODE;
1348
1349	ret = gfs2_instantiate(gh);
1350	if (ret)
1351	return SHOULD_NOT_DELETE_DINODE;
1352
1353	/*
1354	* The inode may have been recreated in the meantime.
1355	*/
1356	if (inode->i_nlink)
1357	return SHOULD_NOT_DELETE_DINODE;
1358
1359	should_delete:
1360	if (gfs2_holder_initialized(gh: &ip->i_iopen_gh) &&
1361	test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1362	if (!gfs2_upgrade_iopen_glock(inode)) {
1363	gfs2_holder_uninit(gh: &ip->i_iopen_gh);
1364	return SHOULD_NOT_DELETE_DINODE;
1365	}
1366	}
1367	return SHOULD_DELETE_DINODE;
1368	}
1369
1370	/**
1371	* evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1372	* @inode: The inode to evict
1373	*/
1374	static int evict_unlinked_inode(struct inode *inode)
1375	{
1376	struct gfs2_inode *ip = GFS2_I(inode);
1377	int ret;
1378
1379	if (S_ISDIR(inode->i_mode) &&
1380	(ip->i_diskflags & GFS2_DIF_EXHASH)) {
1381	ret = gfs2_dir_exhash_dealloc(dip: ip);
1382	if (ret)
1383	goto out;
1384	}
1385
1386	if (ip->i_eattr) {
1387	ret = gfs2_ea_dealloc(ip);
1388	if (ret)
1389	goto out;
1390	}
1391
1392	if (!gfs2_is_stuffed(ip)) {
1393	ret = gfs2_file_dealloc(ip);
1394	if (ret)
1395	goto out;
1396	}
1397
1398	/*
1399	* As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1400	* can get called to recreate it, or even gfs2_inode_lookup() if the
1401	* inode was recreated on another node in the meantime.
1402	*
1403	* However, inserting the new inode into the inode hash table will not
1404	* succeed until the old inode is removed, and that only happens after
1405	* ->evict_inode() returns. The new inode is attached to its inode and
1406	* iopen glocks after inserting it into the inode hash table, so at
1407	* that point we can be sure that both glocks are unused.
1408	*/
1409
1410	ret = gfs2_dinode_dealloc(ip);
1411	if (!ret && ip->i_gl)
1412	gfs2_inode_remember_delete(gl: ip->i_gl, generation: ip->i_no_formal_ino);
1413
1414	out:
1415	return ret;
1416	}
1417
1418	/*
1419	* evict_linked_inode - evict an inode whose dinode has not been unlinked
1420	* @inode: The inode to evict
1421	*/
1422	static int evict_linked_inode(struct inode *inode)
1423	{
1424	struct super_block *sb = inode->i_sb;
1425	struct gfs2_sbd *sdp = sb->s_fs_info;
1426	struct gfs2_inode *ip = GFS2_I(inode);
1427	struct address_space *metamapping;
1428	int ret;
1429
1430	gfs2_log_flush(sdp, gl: ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1431	GFS2_LFC_EVICT_INODE);
1432	metamapping = gfs2_glock2aspace(gl: ip->i_gl);
1433	if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1434	filemap_fdatawrite(metamapping);
1435	filemap_fdatawait(mapping: metamapping);
1436	}
1437	write_inode_now(inode, sync: 1);
1438	gfs2_ail_flush(gl: ip->i_gl, fsync: 0);
1439
1440	ret = gfs2_trans_begin(sdp, blocks: 0, revokes: sdp->sd_jdesc->jd_blocks);
1441	if (ret)
1442	return ret;
1443
1444	/* Needs to be done before glock release & also in a transaction */
1445	truncate_inode_pages(&inode->i_data, 0);
1446	truncate_inode_pages(metamapping, 0);
1447	gfs2_trans_end(sdp);
1448	return 0;
1449	}
1450
1451	/**
1452	* gfs2_evict_inode - Remove an inode from cache
1453	* @inode: The inode to evict
1454	*
1455	* There are three cases to consider:
1456	* 1. i_nlink == 0, we are final opener (and must deallocate)
1457	* 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1458	* 3. i_nlink > 0
1459	*
1460	* If the fs is read only, then we have to treat all cases as per #3
1461	* since we are unable to do any deallocation. The inode will be
1462	* deallocated by the next read/write node to attempt an allocation
1463	* in the same resource group
1464	*
1465	* We have to (at the moment) hold the inodes main lock to cover
1466	* the gap between unlocking the shared lock on the iopen lock and
1467	* taking the exclusive lock. I'd rather do a shared -> exclusive
1468	* conversion on the iopen lock, but we can change that later. This
1469	* is safe, just less efficient.
1470	*/
1471
1472	static void gfs2_evict_inode(struct inode *inode)
1473	{
1474	struct super_block *sb = inode->i_sb;
1475	struct gfs2_sbd *sdp = sb->s_fs_info;
1476	struct gfs2_inode *ip = GFS2_I(inode);
1477	struct gfs2_holder gh;
1478	int ret;
1479
1480	if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1481	goto out;
1482
1483	/*
1484	* In case of an incomplete mount, gfs2_evict_inode() may be called for
1485	* system files without having an active journal to write to. In that
1486	* case, skip the filesystem evict.
1487	*/
1488	if (!sdp->sd_jdesc)
1489	goto out;
1490
1491	gfs2_holder_mark_uninitialized(gh: &gh);
1492	ret = evict_should_delete(inode, gh: &gh);
1493	if (ret == SHOULD_DEFER_EVICTION)
1494	goto out;
1495	if (ret == SHOULD_DELETE_DINODE)
1496	ret = evict_unlinked_inode(inode);
1497	else
1498	ret = evict_linked_inode(inode);
1499
1500	if (gfs2_rs_active(rs: &ip->i_res))
1501	gfs2_rs_deltree(rs: &ip->i_res);
1502
1503	if (gfs2_holder_initialized(gh: &gh))
1504	gfs2_glock_dq_uninit(gh: &gh);
1505	if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1506	fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1507	out:
1508	truncate_inode_pages_final(&inode->i_data);
1509	if (ip->i_qadata)
1510	gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1511	gfs2_rs_deltree(rs: &ip->i_res);
1512	gfs2_ordered_del_inode(ip);
1513	clear_inode(inode);
1514	gfs2_dir_hash_inval(ip);
1515	if (gfs2_holder_initialized(gh: &ip->i_iopen_gh)) {
1516	struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1517
1518	glock_clear_object(gl, object: ip);
1519	gfs2_glock_hold(gl);
1520	ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1521	gfs2_glock_dq_uninit(gh: &ip->i_iopen_gh);
1522	gfs2_glock_put_eventually(gl);
1523	}
1524	if (ip->i_gl) {
1525	glock_clear_object(gl: ip->i_gl, object: ip);
1526	wait_on_bit_io(word: &ip->i_flags, bit: GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1527	gfs2_glock_add_to_lru(gl: ip->i_gl);
1528	gfs2_glock_put_eventually(gl: ip->i_gl);
1529	rcu_assign_pointer(ip->i_gl, NULL);
1530	}
1531	}
1532
1533	static struct inode *gfs2_alloc_inode(struct super_block *sb)
1534	{
1535	struct gfs2_inode *ip;
1536
1537	ip = alloc_inode_sb(sb, cache: gfs2_inode_cachep, GFP_KERNEL);
1538	if (!ip)
1539	return NULL;
1540	ip->i_no_addr = 0;
1541	ip->i_flags = 0;
1542	ip->i_gl = NULL;
1543	gfs2_holder_mark_uninitialized(gh: &ip->i_iopen_gh);
1544	memset(&ip->i_res, 0, sizeof(ip->i_res));
1545	RB_CLEAR_NODE(&ip->i_res.rs_node);
1546	ip->i_rahead = 0;
1547	return &ip->i_inode;
1548	}
1549
1550	static void gfs2_free_inode(struct inode *inode)
1551	{
1552	kmem_cache_free(s: gfs2_inode_cachep, objp: GFS2_I(inode));
1553	}
1554
1555	void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1556	{
1557	struct local_statfs_inode *lsi, *safe;
1558
1559	/* Run through the statfs inodes list to iput and free memory */
1560	list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1561	if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1562	sdp->sd_sc_inode = NULL; /* belongs to this node */
1563	if (lsi->si_sc_inode)
1564	iput(lsi->si_sc_inode);
1565	list_del(entry: &lsi->si_list);
1566	kfree(objp: lsi);
1567	}
1568	}
1569
1570	struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1571	unsigned int index)
1572	{
1573	struct local_statfs_inode *lsi;
1574
1575	/* Return the local (per node) statfs inode in the
1576	* sdp->sd_sc_inodes_list corresponding to the 'index'. */
1577	list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1578	if (lsi->si_jid == index)
1579	return lsi->si_sc_inode;
1580	}
1581	return NULL;
1582	}
1583
1584	const struct super_operations gfs2_super_ops = {
1585	.alloc_inode = gfs2_alloc_inode,
1586	.free_inode = gfs2_free_inode,
1587	.write_inode = gfs2_write_inode,
1588	.dirty_inode = gfs2_dirty_inode,
1589	.evict_inode = gfs2_evict_inode,
1590	.put_super = gfs2_put_super,
1591	.sync_fs = gfs2_sync_fs,
1592	.freeze_super = gfs2_freeze_super,
1593	.freeze_fs = gfs2_freeze_fs,
1594	.thaw_super = gfs2_thaw_super,
1595	.statfs = gfs2_statfs,
1596	.drop_inode = gfs2_drop_inode,
1597	.show_options = gfs2_show_options,
1598	};
1599
1600