xref: /linux/fs/gfs2/super.c (revision a1c3be890440a1769ed6f822376a3e3ab0d42994)
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(&sdp->sd_jindex_spin);
65 	list_add(&list, &sdp->sd_jindex_list);
66 	list_del_init(&sdp->sd_jindex_list);
67 	sdp->sd_journals = 0;
68 	spin_unlock(&sdp->sd_jindex_spin);
69 
70 	sdp->sd_jdesc = NULL;
71 	while (!list_empty(&list)) {
72 		jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
73 		gfs2_free_journal_extents(jd);
74 		list_del(&jd->jd_list);
75 		iput(jd->jd_inode);
76 		jd->jd_inode = NULL;
77 		kfree(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(&sdp->sd_jindex_spin);
97 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
98 	spin_unlock(&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(jd->jd_inode);
106 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
107 	u64 size = i_size_read(jd->jd_inode);
108 
109 	if (gfs2_check_internal_file_size(jd->jd_inode, 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, 0, size)) {
115 		gfs2_consist_inode(ip);
116 		return -EIO;
117 	}
118 
119 	return 0;
120 }
121 
122 static int init_threads(struct gfs2_sbd *sdp)
123 {
124 	struct task_struct *p;
125 	int error = 0;
126 
127 	p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
128 	if (IS_ERR(p)) {
129 		error = PTR_ERR(p);
130 		fs_err(sdp, "can't start logd thread: %d\n", error);
131 		return error;
132 	}
133 	sdp->sd_logd_process = p;
134 
135 	p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
136 	if (IS_ERR(p)) {
137 		error = PTR_ERR(p);
138 		fs_err(sdp, "can't start quotad thread: %d\n", error);
139 		goto fail;
140 	}
141 	sdp->sd_quotad_process = p;
142 	return 0;
143 
144 fail:
145 	kthread_stop(sdp->sd_logd_process);
146 	sdp->sd_logd_process = NULL;
147 	return error;
148 }
149 
150 /**
151  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
152  * @sdp: the filesystem
153  *
154  * Returns: errno
155  */
156 
157 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
158 {
159 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
160 	struct gfs2_glock *j_gl = ip->i_gl;
161 	struct gfs2_log_header_host head;
162 	int error;
163 
164 	error = init_threads(sdp);
165 	if (error)
166 		return error;
167 
168 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
169 	if (gfs2_withdrawn(sdp)) {
170 		error = -EIO;
171 		goto fail;
172 	}
173 
174 	error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
175 	if (error || gfs2_withdrawn(sdp))
176 		goto fail;
177 
178 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
179 		gfs2_consist(sdp);
180 		error = -EIO;
181 		goto fail;
182 	}
183 
184 	/*  Initialize some head of the log stuff  */
185 	sdp->sd_log_sequence = head.lh_sequence + 1;
186 	gfs2_log_pointers_init(sdp, head.lh_blkno);
187 
188 	error = gfs2_quota_init(sdp);
189 	if (error || gfs2_withdrawn(sdp))
190 		goto fail;
191 
192 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
193 
194 	return 0;
195 
196 fail:
197 	if (sdp->sd_quotad_process)
198 		kthread_stop(sdp->sd_quotad_process);
199 	sdp->sd_quotad_process = NULL;
200 	if (sdp->sd_logd_process)
201 		kthread_stop(sdp->sd_logd_process);
202 	sdp->sd_logd_process = NULL;
203 	return error;
204 }
205 
206 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
207 {
208 	const struct gfs2_statfs_change *str = buf;
209 
210 	sc->sc_total = be64_to_cpu(str->sc_total);
211 	sc->sc_free = be64_to_cpu(str->sc_free);
212 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
213 }
214 
215 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
216 {
217 	struct gfs2_statfs_change *str = buf;
218 
219 	str->sc_total = cpu_to_be64(sc->sc_total);
220 	str->sc_free = cpu_to_be64(sc->sc_free);
221 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
222 }
223 
224 int gfs2_statfs_init(struct gfs2_sbd *sdp)
225 {
226 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
227 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
228 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
229 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
230 	struct buffer_head *m_bh, *l_bh;
231 	struct gfs2_holder gh;
232 	int error;
233 
234 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
235 				   &gh);
236 	if (error)
237 		return error;
238 
239 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
240 	if (error)
241 		goto out;
242 
243 	if (sdp->sd_args.ar_spectator) {
244 		spin_lock(&sdp->sd_statfs_spin);
245 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
246 				      sizeof(struct gfs2_dinode));
247 		spin_unlock(&sdp->sd_statfs_spin);
248 	} else {
249 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
250 		if (error)
251 			goto out_m_bh;
252 
253 		spin_lock(&sdp->sd_statfs_spin);
254 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
255 				      sizeof(struct gfs2_dinode));
256 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
257 				      sizeof(struct gfs2_dinode));
258 		spin_unlock(&sdp->sd_statfs_spin);
259 
260 		brelse(l_bh);
261 	}
262 
263 out_m_bh:
264 	brelse(m_bh);
265 out:
266 	gfs2_glock_dq_uninit(&gh);
267 	return 0;
268 }
269 
270 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
271 			s64 dinodes)
272 {
273 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
274 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
275 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
276 	struct buffer_head *l_bh;
277 	s64 x, y;
278 	int need_sync = 0;
279 	int error;
280 
281 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
282 	if (error)
283 		return;
284 
285 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
286 
287 	spin_lock(&sdp->sd_statfs_spin);
288 	l_sc->sc_total += total;
289 	l_sc->sc_free += free;
290 	l_sc->sc_dinodes += dinodes;
291 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
292 	if (sdp->sd_args.ar_statfs_percent) {
293 		x = 100 * l_sc->sc_free;
294 		y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
295 		if (x >= y || x <= -y)
296 			need_sync = 1;
297 	}
298 	spin_unlock(&sdp->sd_statfs_spin);
299 
300 	brelse(l_bh);
301 	if (need_sync)
302 		gfs2_wake_up_statfs(sdp);
303 }
304 
305 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
306 		   struct buffer_head *l_bh)
307 {
308 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
309 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
310 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
311 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
312 
313 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
314 	gfs2_trans_add_meta(m_ip->i_gl, m_bh);
315 
316 	spin_lock(&sdp->sd_statfs_spin);
317 	m_sc->sc_total += l_sc->sc_total;
318 	m_sc->sc_free += l_sc->sc_free;
319 	m_sc->sc_dinodes += l_sc->sc_dinodes;
320 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
321 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
322 	       0, sizeof(struct gfs2_statfs_change));
323 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
324 	spin_unlock(&sdp->sd_statfs_spin);
325 }
326 
327 int gfs2_statfs_sync(struct super_block *sb, int type)
328 {
329 	struct gfs2_sbd *sdp = sb->s_fs_info;
330 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
331 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
332 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
333 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
334 	struct gfs2_holder gh;
335 	struct buffer_head *m_bh, *l_bh;
336 	int error;
337 
338 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
339 				   &gh);
340 	if (error)
341 		goto out;
342 
343 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
344 	if (error)
345 		goto out_unlock;
346 
347 	spin_lock(&sdp->sd_statfs_spin);
348 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
349 			      sizeof(struct gfs2_dinode));
350 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
351 		spin_unlock(&sdp->sd_statfs_spin);
352 		goto out_bh;
353 	}
354 	spin_unlock(&sdp->sd_statfs_spin);
355 
356 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
357 	if (error)
358 		goto out_bh;
359 
360 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
361 	if (error)
362 		goto out_bh2;
363 
364 	update_statfs(sdp, m_bh, l_bh);
365 	sdp->sd_statfs_force_sync = 0;
366 
367 	gfs2_trans_end(sdp);
368 
369 out_bh2:
370 	brelse(l_bh);
371 out_bh:
372 	brelse(m_bh);
373 out_unlock:
374 	gfs2_glock_dq_uninit(&gh);
375 out:
376 	return error;
377 }
378 
379 struct lfcc {
380 	struct list_head list;
381 	struct gfs2_holder gh;
382 };
383 
384 /**
385  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
386  *                            journals are clean
387  * @sdp: the file system
388  * @state: the state to put the transaction lock into
389  * @t_gh: the hold on the transaction lock
390  *
391  * Returns: errno
392  */
393 
394 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
395 {
396 	struct gfs2_inode *ip;
397 	struct gfs2_jdesc *jd;
398 	struct lfcc *lfcc;
399 	LIST_HEAD(list);
400 	struct gfs2_log_header_host lh;
401 	int error;
402 
403 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
404 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
405 		if (!lfcc) {
406 			error = -ENOMEM;
407 			goto out;
408 		}
409 		ip = GFS2_I(jd->jd_inode);
410 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
411 		if (error) {
412 			kfree(lfcc);
413 			goto out;
414 		}
415 		list_add(&lfcc->list, &list);
416 	}
417 
418 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
419 				   LM_FLAG_NOEXP, &sdp->sd_freeze_gh);
420 	if (error)
421 		goto out;
422 
423 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
424 		error = gfs2_jdesc_check(jd);
425 		if (error)
426 			break;
427 		error = gfs2_find_jhead(jd, &lh, false);
428 		if (error)
429 			break;
430 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
431 			error = -EBUSY;
432 			break;
433 		}
434 	}
435 
436 	if (error)
437 		gfs2_freeze_unlock(&sdp->sd_freeze_gh);
438 
439 out:
440 	while (!list_empty(&list)) {
441 		lfcc = list_first_entry(&list, struct lfcc, list);
442 		list_del(&lfcc->list);
443 		gfs2_glock_dq_uninit(&lfcc->gh);
444 		kfree(lfcc);
445 	}
446 	return error;
447 }
448 
449 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
450 {
451 	struct gfs2_dinode *str = buf;
452 
453 	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
454 	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
455 	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
456 	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
457 	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
458 	str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
459 	str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
460 	str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
461 	str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
462 	str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
463 	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
464 	str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
465 	str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
466 	str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
467 
468 	str->di_goal_meta = cpu_to_be64(ip->i_goal);
469 	str->di_goal_data = cpu_to_be64(ip->i_goal);
470 	str->di_generation = cpu_to_be64(ip->i_generation);
471 
472 	str->di_flags = cpu_to_be32(ip->i_diskflags);
473 	str->di_height = cpu_to_be16(ip->i_height);
474 	str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
475 					     !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
476 					     GFS2_FORMAT_DE : 0);
477 	str->di_depth = cpu_to_be16(ip->i_depth);
478 	str->di_entries = cpu_to_be32(ip->i_entries);
479 
480 	str->di_eattr = cpu_to_be64(ip->i_eattr);
481 	str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
482 	str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
483 	str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
484 }
485 
486 /**
487  * gfs2_write_inode - Make sure the inode is stable on the disk
488  * @inode: The inode
489  * @wbc: The writeback control structure
490  *
491  * Returns: errno
492  */
493 
494 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
495 {
496 	struct gfs2_inode *ip = GFS2_I(inode);
497 	struct gfs2_sbd *sdp = GFS2_SB(inode);
498 	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
499 	struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
500 	int ret = 0;
501 	bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
502 
503 	if (flush_all)
504 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
505 			       GFS2_LOG_HEAD_FLUSH_NORMAL |
506 			       GFS2_LFC_WRITE_INODE);
507 	if (bdi->wb.dirty_exceeded)
508 		gfs2_ail1_flush(sdp, wbc);
509 	else
510 		filemap_fdatawrite(metamapping);
511 	if (flush_all)
512 		ret = filemap_fdatawait(metamapping);
513 	if (ret)
514 		mark_inode_dirty_sync(inode);
515 	else {
516 		spin_lock(&inode->i_lock);
517 		if (!(inode->i_flags & I_DIRTY))
518 			gfs2_ordered_del_inode(ip);
519 		spin_unlock(&inode->i_lock);
520 	}
521 	return ret;
522 }
523 
524 /**
525  * gfs2_dirty_inode - check for atime updates
526  * @inode: The inode in question
527  * @flags: The type of dirty
528  *
529  * Unfortunately it can be called under any combination of inode
530  * glock and transaction lock, so we have to check carefully.
531  *
532  * At the moment this deals only with atime - it should be possible
533  * to expand that role in future, once a review of the locking has
534  * been carried out.
535  */
536 
537 static void gfs2_dirty_inode(struct inode *inode, int flags)
538 {
539 	struct gfs2_inode *ip = GFS2_I(inode);
540 	struct gfs2_sbd *sdp = GFS2_SB(inode);
541 	struct buffer_head *bh;
542 	struct gfs2_holder gh;
543 	int need_unlock = 0;
544 	int need_endtrans = 0;
545 	int ret;
546 
547 	if (unlikely(gfs2_withdrawn(sdp)))
548 		return;
549 	if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
550 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
551 		if (ret) {
552 			fs_err(sdp, "dirty_inode: glock %d\n", ret);
553 			gfs2_dump_glock(NULL, ip->i_gl, true);
554 			return;
555 		}
556 		need_unlock = 1;
557 	} else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
558 		return;
559 
560 	if (current->journal_info == NULL) {
561 		ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
562 		if (ret) {
563 			fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
564 			goto out;
565 		}
566 		need_endtrans = 1;
567 	}
568 
569 	ret = gfs2_meta_inode_buffer(ip, &bh);
570 	if (ret == 0) {
571 		gfs2_trans_add_meta(ip->i_gl, bh);
572 		gfs2_dinode_out(ip, bh->b_data);
573 		brelse(bh);
574 	}
575 
576 	if (need_endtrans)
577 		gfs2_trans_end(sdp);
578 out:
579 	if (need_unlock)
580 		gfs2_glock_dq_uninit(&gh);
581 }
582 
583 /**
584  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
585  * @sdp: the filesystem
586  *
587  * Returns: errno
588  */
589 
590 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
591 {
592 	int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
593 
594 	gfs2_flush_delete_work(sdp);
595 	if (!log_write_allowed && current == sdp->sd_quotad_process)
596 		fs_warn(sdp, "The quotad daemon is withdrawing.\n");
597 	else if (sdp->sd_quotad_process)
598 		kthread_stop(sdp->sd_quotad_process);
599 	sdp->sd_quotad_process = NULL;
600 
601 	if (!log_write_allowed && current == sdp->sd_logd_process)
602 		fs_warn(sdp, "The logd daemon is withdrawing.\n");
603 	else if (sdp->sd_logd_process)
604 		kthread_stop(sdp->sd_logd_process);
605 	sdp->sd_logd_process = NULL;
606 
607 	if (log_write_allowed) {
608 		gfs2_quota_sync(sdp->sd_vfs, 0);
609 		gfs2_statfs_sync(sdp->sd_vfs, 0);
610 
611 		gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
612 			       GFS2_LFC_MAKE_FS_RO);
613 		wait_event_timeout(sdp->sd_log_waitq,
614 				   gfs2_log_is_empty(sdp),
615 				   HZ * 5);
616 		gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
617 	} else {
618 		wait_event_timeout(sdp->sd_log_waitq,
619 				   gfs2_log_is_empty(sdp),
620 				   HZ * 5);
621 	}
622 	gfs2_quota_cleanup(sdp);
623 
624 	if (!log_write_allowed)
625 		sdp->sd_vfs->s_flags |= SB_RDONLY;
626 }
627 
628 /**
629  * gfs2_put_super - Unmount the filesystem
630  * @sb: The VFS superblock
631  *
632  */
633 
634 static void gfs2_put_super(struct super_block *sb)
635 {
636 	struct gfs2_sbd *sdp = sb->s_fs_info;
637 	struct gfs2_jdesc *jd;
638 
639 	/* No more recovery requests */
640 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
641 	smp_mb();
642 
643 	/* Wait on outstanding recovery */
644 restart:
645 	spin_lock(&sdp->sd_jindex_spin);
646 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
647 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
648 			continue;
649 		spin_unlock(&sdp->sd_jindex_spin);
650 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
651 			    TASK_UNINTERRUPTIBLE);
652 		goto restart;
653 	}
654 	spin_unlock(&sdp->sd_jindex_spin);
655 
656 	if (!sb_rdonly(sb)) {
657 		gfs2_make_fs_ro(sdp);
658 	}
659 	WARN_ON(gfs2_withdrawing(sdp));
660 
661 	/*  At this point, we're through modifying the disk  */
662 
663 	/*  Release stuff  */
664 
665 	iput(sdp->sd_jindex);
666 	iput(sdp->sd_statfs_inode);
667 	iput(sdp->sd_rindex);
668 	iput(sdp->sd_quota_inode);
669 
670 	gfs2_glock_put(sdp->sd_rename_gl);
671 	gfs2_glock_put(sdp->sd_freeze_gl);
672 
673 	if (!sdp->sd_args.ar_spectator) {
674 		if (gfs2_holder_initialized(&sdp->sd_journal_gh))
675 			gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
676 		if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
677 			gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
678 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
679 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
680 		free_local_statfs_inodes(sdp);
681 		iput(sdp->sd_qc_inode);
682 	}
683 
684 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
685 	gfs2_clear_rgrpd(sdp);
686 	gfs2_jindex_free(sdp);
687 	/*  Take apart glock structures and buffer lists  */
688 	gfs2_gl_hash_clear(sdp);
689 	truncate_inode_pages_final(&sdp->sd_aspace);
690 	gfs2_delete_debugfs_file(sdp);
691 	/*  Unmount the locking protocol  */
692 	gfs2_lm_unmount(sdp);
693 
694 	/*  At this point, we're through participating in the lockspace  */
695 	gfs2_sys_fs_del(sdp);
696 	free_sbd(sdp);
697 }
698 
699 /**
700  * gfs2_sync_fs - sync the filesystem
701  * @sb: the superblock
702  *
703  * Flushes the log to disk.
704  */
705 
706 static int gfs2_sync_fs(struct super_block *sb, int wait)
707 {
708 	struct gfs2_sbd *sdp = sb->s_fs_info;
709 
710 	gfs2_quota_sync(sb, -1);
711 	if (wait)
712 		gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
713 			       GFS2_LFC_SYNC_FS);
714 	return sdp->sd_log_error;
715 }
716 
717 void gfs2_freeze_func(struct work_struct *work)
718 {
719 	int error;
720 	struct gfs2_holder freeze_gh;
721 	struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
722 	struct super_block *sb = sdp->sd_vfs;
723 
724 	atomic_inc(&sb->s_active);
725 	error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
726 	if (error) {
727 		gfs2_assert_withdraw(sdp, 0);
728 	} else {
729 		atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
730 		error = thaw_super(sb);
731 		if (error) {
732 			fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
733 				error);
734 			gfs2_assert_withdraw(sdp, 0);
735 		}
736 		gfs2_freeze_unlock(&freeze_gh);
737 	}
738 	deactivate_super(sb);
739 	clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
740 	wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
741 	return;
742 }
743 
744 /**
745  * gfs2_freeze - prevent further writes to the filesystem
746  * @sb: the VFS structure for the filesystem
747  *
748  */
749 
750 static int gfs2_freeze(struct super_block *sb)
751 {
752 	struct gfs2_sbd *sdp = sb->s_fs_info;
753 	int error = 0;
754 
755 	mutex_lock(&sdp->sd_freeze_mutex);
756 	if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
757 		goto out;
758 
759 	for (;;) {
760 		if (gfs2_withdrawn(sdp)) {
761 			error = -EINVAL;
762 			goto out;
763 		}
764 
765 		error = gfs2_lock_fs_check_clean(sdp);
766 		if (!error)
767 			break;
768 
769 		if (error == -EBUSY)
770 			fs_err(sdp, "waiting for recovery before freeze\n");
771 		else if (error == -EIO) {
772 			fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
773 			       "to recovery error.\n");
774 			goto out;
775 		} else {
776 			fs_err(sdp, "error freezing FS: %d\n", error);
777 		}
778 		fs_err(sdp, "retrying...\n");
779 		msleep(1000);
780 	}
781 	set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
782 out:
783 	mutex_unlock(&sdp->sd_freeze_mutex);
784 	return error;
785 }
786 
787 /**
788  * gfs2_unfreeze - reallow writes to the filesystem
789  * @sb: the VFS structure for the filesystem
790  *
791  */
792 
793 static int gfs2_unfreeze(struct super_block *sb)
794 {
795 	struct gfs2_sbd *sdp = sb->s_fs_info;
796 
797 	mutex_lock(&sdp->sd_freeze_mutex);
798         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
799 	    !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
800 		mutex_unlock(&sdp->sd_freeze_mutex);
801                 return 0;
802 	}
803 
804 	gfs2_freeze_unlock(&sdp->sd_freeze_gh);
805 	mutex_unlock(&sdp->sd_freeze_mutex);
806 	return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
807 }
808 
809 /**
810  * statfs_fill - fill in the sg for a given RG
811  * @rgd: the RG
812  * @sc: the sc structure
813  *
814  * Returns: 0 on success, -ESTALE if the LVB is invalid
815  */
816 
817 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
818 			    struct gfs2_statfs_change_host *sc)
819 {
820 	gfs2_rgrp_verify(rgd);
821 	sc->sc_total += rgd->rd_data;
822 	sc->sc_free += rgd->rd_free;
823 	sc->sc_dinodes += rgd->rd_dinodes;
824 	return 0;
825 }
826 
827 /**
828  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
829  * @sdp: the filesystem
830  * @sc: the sc info that will be returned
831  *
832  * Any error (other than a signal) will cause this routine to fall back
833  * to the synchronous version.
834  *
835  * FIXME: This really shouldn't busy wait like this.
836  *
837  * Returns: errno
838  */
839 
840 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
841 {
842 	struct gfs2_rgrpd *rgd_next;
843 	struct gfs2_holder *gha, *gh;
844 	unsigned int slots = 64;
845 	unsigned int x;
846 	int done;
847 	int error = 0, err;
848 
849 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
850 	gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
851 	if (!gha)
852 		return -ENOMEM;
853 	for (x = 0; x < slots; x++)
854 		gfs2_holder_mark_uninitialized(gha + x);
855 
856 	rgd_next = gfs2_rgrpd_get_first(sdp);
857 
858 	for (;;) {
859 		done = 1;
860 
861 		for (x = 0; x < slots; x++) {
862 			gh = gha + x;
863 
864 			if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
865 				err = gfs2_glock_wait(gh);
866 				if (err) {
867 					gfs2_holder_uninit(gh);
868 					error = err;
869 				} else {
870 					if (!error) {
871 						struct gfs2_rgrpd *rgd =
872 							gfs2_glock2rgrp(gh->gh_gl);
873 
874 						error = statfs_slow_fill(rgd, sc);
875 					}
876 					gfs2_glock_dq_uninit(gh);
877 				}
878 			}
879 
880 			if (gfs2_holder_initialized(gh))
881 				done = 0;
882 			else if (rgd_next && !error) {
883 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
884 							   LM_ST_SHARED,
885 							   GL_ASYNC,
886 							   gh);
887 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
888 				done = 0;
889 			}
890 
891 			if (signal_pending(current))
892 				error = -ERESTARTSYS;
893 		}
894 
895 		if (done)
896 			break;
897 
898 		yield();
899 	}
900 
901 	kfree(gha);
902 	return error;
903 }
904 
905 /**
906  * gfs2_statfs_i - Do a statfs
907  * @sdp: the filesystem
908  * @sg: the sg structure
909  *
910  * Returns: errno
911  */
912 
913 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
914 {
915 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
916 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
917 
918 	spin_lock(&sdp->sd_statfs_spin);
919 
920 	*sc = *m_sc;
921 	sc->sc_total += l_sc->sc_total;
922 	sc->sc_free += l_sc->sc_free;
923 	sc->sc_dinodes += l_sc->sc_dinodes;
924 
925 	spin_unlock(&sdp->sd_statfs_spin);
926 
927 	if (sc->sc_free < 0)
928 		sc->sc_free = 0;
929 	if (sc->sc_free > sc->sc_total)
930 		sc->sc_free = sc->sc_total;
931 	if (sc->sc_dinodes < 0)
932 		sc->sc_dinodes = 0;
933 
934 	return 0;
935 }
936 
937 /**
938  * gfs2_statfs - Gather and return stats about the filesystem
939  * @sb: The superblock
940  * @statfsbuf: The buffer
941  *
942  * Returns: 0 on success or error code
943  */
944 
945 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
946 {
947 	struct super_block *sb = dentry->d_sb;
948 	struct gfs2_sbd *sdp = sb->s_fs_info;
949 	struct gfs2_statfs_change_host sc;
950 	int error;
951 
952 	error = gfs2_rindex_update(sdp);
953 	if (error)
954 		return error;
955 
956 	if (gfs2_tune_get(sdp, gt_statfs_slow))
957 		error = gfs2_statfs_slow(sdp, &sc);
958 	else
959 		error = gfs2_statfs_i(sdp, &sc);
960 
961 	if (error)
962 		return error;
963 
964 	buf->f_type = GFS2_MAGIC;
965 	buf->f_bsize = sdp->sd_sb.sb_bsize;
966 	buf->f_blocks = sc.sc_total;
967 	buf->f_bfree = sc.sc_free;
968 	buf->f_bavail = sc.sc_free;
969 	buf->f_files = sc.sc_dinodes + sc.sc_free;
970 	buf->f_ffree = sc.sc_free;
971 	buf->f_namelen = GFS2_FNAMESIZE;
972 
973 	return 0;
974 }
975 
976 /**
977  * gfs2_drop_inode - Drop an inode (test for remote unlink)
978  * @inode: The inode to drop
979  *
980  * If we've received a callback on an iopen lock then it's because a
981  * remote node tried to deallocate the inode but failed due to this node
982  * still having the inode open. Here we mark the link count zero
983  * since we know that it must have reached zero if the GLF_DEMOTE flag
984  * is set on the iopen glock. If we didn't do a disk read since the
985  * remote node removed the final link then we might otherwise miss
986  * this event. This check ensures that this node will deallocate the
987  * inode's blocks, or alternatively pass the baton on to another
988  * node for later deallocation.
989  */
990 
991 static int gfs2_drop_inode(struct inode *inode)
992 {
993 	struct gfs2_inode *ip = GFS2_I(inode);
994 
995 	if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
996 	    inode->i_nlink &&
997 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
998 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
999 		if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1000 			clear_nlink(inode);
1001 	}
1002 
1003 	/*
1004 	 * When under memory pressure when an inode's link count has dropped to
1005 	 * zero, defer deleting the inode to the delete workqueue.  This avoids
1006 	 * calling into DLM under memory pressure, which can deadlock.
1007 	 */
1008 	if (!inode->i_nlink &&
1009 	    unlikely(current->flags & PF_MEMALLOC) &&
1010 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
1011 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1012 
1013 		gfs2_glock_hold(gl);
1014 		if (!gfs2_queue_delete_work(gl, 0))
1015 			gfs2_glock_queue_put(gl);
1016 		return false;
1017 	}
1018 
1019 	return generic_drop_inode(inode);
1020 }
1021 
1022 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1023 {
1024 	do {
1025 		if (d1 == d2)
1026 			return 1;
1027 		d1 = d1->d_parent;
1028 	} while (!IS_ROOT(d1));
1029 	return 0;
1030 }
1031 
1032 /**
1033  * gfs2_show_options - Show mount options for /proc/mounts
1034  * @s: seq_file structure
1035  * @root: root of this (sub)tree
1036  *
1037  * Returns: 0 on success or error code
1038  */
1039 
1040 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1041 {
1042 	struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1043 	struct gfs2_args *args = &sdp->sd_args;
1044 	int val;
1045 
1046 	if (is_ancestor(root, sdp->sd_master_dir))
1047 		seq_puts(s, ",meta");
1048 	if (args->ar_lockproto[0])
1049 		seq_show_option(s, "lockproto", args->ar_lockproto);
1050 	if (args->ar_locktable[0])
1051 		seq_show_option(s, "locktable", args->ar_locktable);
1052 	if (args->ar_hostdata[0])
1053 		seq_show_option(s, "hostdata", args->ar_hostdata);
1054 	if (args->ar_spectator)
1055 		seq_puts(s, ",spectator");
1056 	if (args->ar_localflocks)
1057 		seq_puts(s, ",localflocks");
1058 	if (args->ar_debug)
1059 		seq_puts(s, ",debug");
1060 	if (args->ar_posix_acl)
1061 		seq_puts(s, ",acl");
1062 	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1063 		char *state;
1064 		switch (args->ar_quota) {
1065 		case GFS2_QUOTA_OFF:
1066 			state = "off";
1067 			break;
1068 		case GFS2_QUOTA_ACCOUNT:
1069 			state = "account";
1070 			break;
1071 		case GFS2_QUOTA_ON:
1072 			state = "on";
1073 			break;
1074 		default:
1075 			state = "unknown";
1076 			break;
1077 		}
1078 		seq_printf(s, ",quota=%s", state);
1079 	}
1080 	if (args->ar_suiddir)
1081 		seq_puts(s, ",suiddir");
1082 	if (args->ar_data != GFS2_DATA_DEFAULT) {
1083 		char *state;
1084 		switch (args->ar_data) {
1085 		case GFS2_DATA_WRITEBACK:
1086 			state = "writeback";
1087 			break;
1088 		case GFS2_DATA_ORDERED:
1089 			state = "ordered";
1090 			break;
1091 		default:
1092 			state = "unknown";
1093 			break;
1094 		}
1095 		seq_printf(s, ",data=%s", state);
1096 	}
1097 	if (args->ar_discard)
1098 		seq_puts(s, ",discard");
1099 	val = sdp->sd_tune.gt_logd_secs;
1100 	if (val != 30)
1101 		seq_printf(s, ",commit=%d", val);
1102 	val = sdp->sd_tune.gt_statfs_quantum;
1103 	if (val != 30)
1104 		seq_printf(s, ",statfs_quantum=%d", val);
1105 	else if (sdp->sd_tune.gt_statfs_slow)
1106 		seq_puts(s, ",statfs_quantum=0");
1107 	val = sdp->sd_tune.gt_quota_quantum;
1108 	if (val != 60)
1109 		seq_printf(s, ",quota_quantum=%d", val);
1110 	if (args->ar_statfs_percent)
1111 		seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1112 	if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1113 		const char *state;
1114 
1115 		switch (args->ar_errors) {
1116 		case GFS2_ERRORS_WITHDRAW:
1117 			state = "withdraw";
1118 			break;
1119 		case GFS2_ERRORS_PANIC:
1120 			state = "panic";
1121 			break;
1122 		default:
1123 			state = "unknown";
1124 			break;
1125 		}
1126 		seq_printf(s, ",errors=%s", state);
1127 	}
1128 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1129 		seq_puts(s, ",nobarrier");
1130 	if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1131 		seq_puts(s, ",demote_interface_used");
1132 	if (args->ar_rgrplvb)
1133 		seq_puts(s, ",rgrplvb");
1134 	if (args->ar_loccookie)
1135 		seq_puts(s, ",loccookie");
1136 	return 0;
1137 }
1138 
1139 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1140 {
1141 	struct inode *inode = &ip->i_inode;
1142 	struct gfs2_glock *gl = ip->i_gl;
1143 
1144 	truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1145 	truncate_inode_pages(&inode->i_data, 0);
1146 
1147 	if (atomic_read(&gl->gl_revokes) == 0) {
1148 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
1149 		clear_bit(GLF_DIRTY, &gl->gl_flags);
1150 	}
1151 }
1152 
1153 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1154 {
1155 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1156 	struct gfs2_rgrpd *rgd;
1157 	struct gfs2_holder gh;
1158 	int error;
1159 
1160 	if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1161 		gfs2_consist_inode(ip);
1162 		return -EIO;
1163 	}
1164 
1165 	error = gfs2_rindex_update(sdp);
1166 	if (error)
1167 		return error;
1168 
1169 	error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1170 	if (error)
1171 		return error;
1172 
1173 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1174 	if (!rgd) {
1175 		gfs2_consist_inode(ip);
1176 		error = -EIO;
1177 		goto out_qs;
1178 	}
1179 
1180 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1181 				   LM_FLAG_NODE_SCOPE, &gh);
1182 	if (error)
1183 		goto out_qs;
1184 
1185 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1186 				 sdp->sd_jdesc->jd_blocks);
1187 	if (error)
1188 		goto out_rg_gunlock;
1189 
1190 	gfs2_free_di(rgd, ip);
1191 
1192 	gfs2_final_release_pages(ip);
1193 
1194 	gfs2_trans_end(sdp);
1195 
1196 out_rg_gunlock:
1197 	gfs2_glock_dq_uninit(&gh);
1198 out_qs:
1199 	gfs2_quota_unhold(ip);
1200 	return error;
1201 }
1202 
1203 /**
1204  * gfs2_glock_put_eventually
1205  * @gl:	The glock to put
1206  *
1207  * When under memory pressure, trigger a deferred glock put to make sure we
1208  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1209  */
1210 
1211 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1212 {
1213 	if (current->flags & PF_MEMALLOC)
1214 		gfs2_glock_queue_put(gl);
1215 	else
1216 		gfs2_glock_put(gl);
1217 }
1218 
1219 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1220 {
1221 	struct gfs2_inode *ip = GFS2_I(inode);
1222 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1223 	struct gfs2_holder *gh = &ip->i_iopen_gh;
1224 	long timeout = 5 * HZ;
1225 	int error;
1226 
1227 	gh->gh_flags |= GL_NOCACHE;
1228 	gfs2_glock_dq_wait(gh);
1229 
1230 	/*
1231 	 * If there are no other lock holders, we'll get the lock immediately.
1232 	 * Otherwise, the other nodes holding the lock will be notified about
1233 	 * our locking request.  If they don't have the inode open, they'll
1234 	 * evict the cached inode and release the lock.  Otherwise, if they
1235 	 * poke the inode glock, we'll take this as an indication that they
1236 	 * still need the iopen glock and that they'll take care of deleting
1237 	 * the inode when they're done.  As a last resort, if another node
1238 	 * keeps holding the iopen glock without showing any activity on the
1239 	 * inode glock, we'll eventually time out.
1240 	 *
1241 	 * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1242 	 * locking request as an optimization to notify lock holders as soon as
1243 	 * possible.  Without that flag, they'd be notified implicitly by the
1244 	 * second locking request.
1245 	 */
1246 
1247 	gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1248 	error = gfs2_glock_nq(gh);
1249 	if (error != GLR_TRYFAILED)
1250 		return !error;
1251 
1252 	gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1253 	error = gfs2_glock_nq(gh);
1254 	if (error)
1255 		return false;
1256 
1257 	timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1258 		!test_bit(HIF_WAIT, &gh->gh_iflags) ||
1259 		test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1260 		timeout);
1261 	if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1262 		gfs2_glock_dq(gh);
1263 		return false;
1264 	}
1265 	return true;
1266 }
1267 
1268 /**
1269  * evict_should_delete - determine whether the inode is eligible for deletion
1270  * @inode: The inode to evict
1271  *
1272  * This function determines whether the evicted inode is eligible to be deleted
1273  * and locks the inode glock.
1274  *
1275  * Returns: the fate of the dinode
1276  */
1277 static enum dinode_demise evict_should_delete(struct inode *inode,
1278 					      struct gfs2_holder *gh)
1279 {
1280 	struct gfs2_inode *ip = GFS2_I(inode);
1281 	struct super_block *sb = inode->i_sb;
1282 	struct gfs2_sbd *sdp = sb->s_fs_info;
1283 	int ret;
1284 
1285 	if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1286 		BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1287 		goto should_delete;
1288 	}
1289 
1290 	if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1291 		return SHOULD_DEFER_EVICTION;
1292 
1293 	/* Deletes should never happen under memory pressure anymore.  */
1294 	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1295 		return SHOULD_DEFER_EVICTION;
1296 
1297 	/* Must not read inode block until block type has been verified */
1298 	ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1299 	if (unlikely(ret)) {
1300 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1301 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1302 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1303 		return SHOULD_DEFER_EVICTION;
1304 	}
1305 
1306 	if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1307 		return SHOULD_NOT_DELETE_DINODE;
1308 	ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1309 	if (ret)
1310 		return SHOULD_NOT_DELETE_DINODE;
1311 
1312 	if (test_bit(GIF_INVALID, &ip->i_flags)) {
1313 		ret = gfs2_inode_refresh(ip);
1314 		if (ret)
1315 			return SHOULD_NOT_DELETE_DINODE;
1316 	}
1317 
1318 	/*
1319 	 * The inode may have been recreated in the meantime.
1320 	 */
1321 	if (inode->i_nlink)
1322 		return SHOULD_NOT_DELETE_DINODE;
1323 
1324 should_delete:
1325 	if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1326 	    test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1327 		if (!gfs2_upgrade_iopen_glock(inode)) {
1328 			gfs2_holder_uninit(&ip->i_iopen_gh);
1329 			return SHOULD_NOT_DELETE_DINODE;
1330 		}
1331 	}
1332 	return SHOULD_DELETE_DINODE;
1333 }
1334 
1335 /**
1336  * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1337  * @inode: The inode to evict
1338  */
1339 static int evict_unlinked_inode(struct inode *inode)
1340 {
1341 	struct gfs2_inode *ip = GFS2_I(inode);
1342 	int ret;
1343 
1344 	if (S_ISDIR(inode->i_mode) &&
1345 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1346 		ret = gfs2_dir_exhash_dealloc(ip);
1347 		if (ret)
1348 			goto out;
1349 	}
1350 
1351 	if (ip->i_eattr) {
1352 		ret = gfs2_ea_dealloc(ip);
1353 		if (ret)
1354 			goto out;
1355 	}
1356 
1357 	if (!gfs2_is_stuffed(ip)) {
1358 		ret = gfs2_file_dealloc(ip);
1359 		if (ret)
1360 			goto out;
1361 	}
1362 
1363 	/* We're about to clear the bitmap for the dinode, but as soon as we
1364 	   do, gfs2_create_inode can create another inode at the same block
1365 	   location and try to set gl_object again. We clear gl_object here so
1366 	   that subsequent inode creates don't see an old gl_object. */
1367 	glock_clear_object(ip->i_gl, ip);
1368 	ret = gfs2_dinode_dealloc(ip);
1369 	gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1370 out:
1371 	return ret;
1372 }
1373 
1374 /*
1375  * evict_linked_inode - evict an inode whose dinode has not been unlinked
1376  * @inode: The inode to evict
1377  */
1378 static int evict_linked_inode(struct inode *inode)
1379 {
1380 	struct super_block *sb = inode->i_sb;
1381 	struct gfs2_sbd *sdp = sb->s_fs_info;
1382 	struct gfs2_inode *ip = GFS2_I(inode);
1383 	struct address_space *metamapping;
1384 	int ret;
1385 
1386 	gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1387 		       GFS2_LFC_EVICT_INODE);
1388 	metamapping = gfs2_glock2aspace(ip->i_gl);
1389 	if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1390 		filemap_fdatawrite(metamapping);
1391 		filemap_fdatawait(metamapping);
1392 	}
1393 	write_inode_now(inode, 1);
1394 	gfs2_ail_flush(ip->i_gl, 0);
1395 
1396 	ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1397 	if (ret)
1398 		return ret;
1399 
1400 	/* Needs to be done before glock release & also in a transaction */
1401 	truncate_inode_pages(&inode->i_data, 0);
1402 	truncate_inode_pages(metamapping, 0);
1403 	gfs2_trans_end(sdp);
1404 	return 0;
1405 }
1406 
1407 /**
1408  * gfs2_evict_inode - Remove an inode from cache
1409  * @inode: The inode to evict
1410  *
1411  * There are three cases to consider:
1412  * 1. i_nlink == 0, we are final opener (and must deallocate)
1413  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1414  * 3. i_nlink > 0
1415  *
1416  * If the fs is read only, then we have to treat all cases as per #3
1417  * since we are unable to do any deallocation. The inode will be
1418  * deallocated by the next read/write node to attempt an allocation
1419  * in the same resource group
1420  *
1421  * We have to (at the moment) hold the inodes main lock to cover
1422  * the gap between unlocking the shared lock on the iopen lock and
1423  * taking the exclusive lock. I'd rather do a shared -> exclusive
1424  * conversion on the iopen lock, but we can change that later. This
1425  * is safe, just less efficient.
1426  */
1427 
1428 static void gfs2_evict_inode(struct inode *inode)
1429 {
1430 	struct super_block *sb = inode->i_sb;
1431 	struct gfs2_sbd *sdp = sb->s_fs_info;
1432 	struct gfs2_inode *ip = GFS2_I(inode);
1433 	struct gfs2_holder gh;
1434 	int ret;
1435 
1436 	if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1437 		clear_inode(inode);
1438 		return;
1439 	}
1440 
1441 	if (inode->i_nlink || sb_rdonly(sb))
1442 		goto out;
1443 
1444 	gfs2_holder_mark_uninitialized(&gh);
1445 	ret = evict_should_delete(inode, &gh);
1446 	if (ret == SHOULD_DEFER_EVICTION)
1447 		goto out;
1448 	if (ret == SHOULD_DELETE_DINODE)
1449 		ret = evict_unlinked_inode(inode);
1450 	else
1451 		ret = evict_linked_inode(inode);
1452 
1453 	if (gfs2_rs_active(&ip->i_res))
1454 		gfs2_rs_deltree(&ip->i_res);
1455 
1456 	if (gfs2_holder_initialized(&gh)) {
1457 		glock_clear_object(ip->i_gl, ip);
1458 		gfs2_glock_dq_uninit(&gh);
1459 	}
1460 	if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1461 		fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1462 out:
1463 	truncate_inode_pages_final(&inode->i_data);
1464 	if (ip->i_qadata)
1465 		gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1466 	gfs2_rs_delete(ip, NULL);
1467 	gfs2_ordered_del_inode(ip);
1468 	clear_inode(inode);
1469 	gfs2_dir_hash_inval(ip);
1470 	if (ip->i_gl) {
1471 		glock_clear_object(ip->i_gl, ip);
1472 		wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1473 		gfs2_glock_add_to_lru(ip->i_gl);
1474 		gfs2_glock_put_eventually(ip->i_gl);
1475 		ip->i_gl = NULL;
1476 	}
1477 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1478 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1479 
1480 		glock_clear_object(gl, ip);
1481 		if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1482 			ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1483 			gfs2_glock_dq(&ip->i_iopen_gh);
1484 		}
1485 		gfs2_glock_hold(gl);
1486 		gfs2_holder_uninit(&ip->i_iopen_gh);
1487 		gfs2_glock_put_eventually(gl);
1488 	}
1489 }
1490 
1491 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1492 {
1493 	struct gfs2_inode *ip;
1494 
1495 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1496 	if (!ip)
1497 		return NULL;
1498 	ip->i_flags = 0;
1499 	ip->i_gl = NULL;
1500 	gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1501 	memset(&ip->i_res, 0, sizeof(ip->i_res));
1502 	RB_CLEAR_NODE(&ip->i_res.rs_node);
1503 	ip->i_rahead = 0;
1504 	return &ip->i_inode;
1505 }
1506 
1507 static void gfs2_free_inode(struct inode *inode)
1508 {
1509 	kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1510 }
1511 
1512 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1513 {
1514 	struct local_statfs_inode *lsi, *safe;
1515 
1516 	/* Run through the statfs inodes list to iput and free memory */
1517 	list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1518 		if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1519 			sdp->sd_sc_inode = NULL; /* belongs to this node */
1520 		if (lsi->si_sc_inode)
1521 			iput(lsi->si_sc_inode);
1522 		list_del(&lsi->si_list);
1523 		kfree(lsi);
1524 	}
1525 }
1526 
1527 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1528 					     unsigned int index)
1529 {
1530 	struct local_statfs_inode *lsi;
1531 
1532 	/* Return the local (per node) statfs inode in the
1533 	 * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1534 	list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1535 		if (lsi->si_jid == index)
1536 			return lsi->si_sc_inode;
1537 	}
1538 	return NULL;
1539 }
1540 
1541 const struct super_operations gfs2_super_ops = {
1542 	.alloc_inode		= gfs2_alloc_inode,
1543 	.free_inode		= gfs2_free_inode,
1544 	.write_inode		= gfs2_write_inode,
1545 	.dirty_inode		= gfs2_dirty_inode,
1546 	.evict_inode		= gfs2_evict_inode,
1547 	.put_super		= gfs2_put_super,
1548 	.sync_fs		= gfs2_sync_fs,
1549 	.freeze_super		= gfs2_freeze,
1550 	.thaw_super		= gfs2_unfreeze,
1551 	.statfs			= gfs2_statfs,
1552 	.drop_inode		= gfs2_drop_inode,
1553 	.show_options		= gfs2_show_options,
1554 };
1555 
1556