xref: /linux/fs/gfs2/super.c (revision 17cfcb68af3bc7d5e8ae08779b1853310a2949f3)
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 /**
48  * gfs2_jindex_free - Clear all the journal index information
49  * @sdp: The GFS2 superblock
50  *
51  */
52 
53 void gfs2_jindex_free(struct gfs2_sbd *sdp)
54 {
55 	struct list_head list;
56 	struct gfs2_jdesc *jd;
57 
58 	spin_lock(&sdp->sd_jindex_spin);
59 	list_add(&list, &sdp->sd_jindex_list);
60 	list_del_init(&sdp->sd_jindex_list);
61 	sdp->sd_journals = 0;
62 	spin_unlock(&sdp->sd_jindex_spin);
63 
64 	while (!list_empty(&list)) {
65 		jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
66 		gfs2_free_journal_extents(jd);
67 		list_del(&jd->jd_list);
68 		iput(jd->jd_inode);
69 		kfree(jd);
70 	}
71 }
72 
73 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
74 {
75 	struct gfs2_jdesc *jd;
76 	int found = 0;
77 
78 	list_for_each_entry(jd, head, jd_list) {
79 		if (jd->jd_jid == jid) {
80 			found = 1;
81 			break;
82 		}
83 	}
84 
85 	if (!found)
86 		jd = NULL;
87 
88 	return jd;
89 }
90 
91 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
92 {
93 	struct gfs2_jdesc *jd;
94 
95 	spin_lock(&sdp->sd_jindex_spin);
96 	jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
97 	spin_unlock(&sdp->sd_jindex_spin);
98 
99 	return jd;
100 }
101 
102 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
103 {
104 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
105 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
106 	u64 size = i_size_read(jd->jd_inode);
107 
108 	if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
109 		return -EIO;
110 
111 	jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
112 
113 	if (gfs2_write_alloc_required(ip, 0, size)) {
114 		gfs2_consist_inode(ip);
115 		return -EIO;
116 	}
117 
118 	return 0;
119 }
120 
121 static int init_threads(struct gfs2_sbd *sdp)
122 {
123 	struct task_struct *p;
124 	int error = 0;
125 
126 	p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
127 	if (IS_ERR(p)) {
128 		error = PTR_ERR(p);
129 		fs_err(sdp, "can't start logd thread: %d\n", error);
130 		return error;
131 	}
132 	sdp->sd_logd_process = p;
133 
134 	p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
135 	if (IS_ERR(p)) {
136 		error = PTR_ERR(p);
137 		fs_err(sdp, "can't start quotad thread: %d\n", error);
138 		goto fail;
139 	}
140 	sdp->sd_quotad_process = p;
141 	return 0;
142 
143 fail:
144 	kthread_stop(sdp->sd_logd_process);
145 	sdp->sd_logd_process = NULL;
146 	return error;
147 }
148 
149 /**
150  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
151  * @sdp: the filesystem
152  *
153  * Returns: errno
154  */
155 
156 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
157 {
158 	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
159 	struct gfs2_glock *j_gl = ip->i_gl;
160 	struct gfs2_holder freeze_gh;
161 	struct gfs2_log_header_host head;
162 	int error;
163 
164 	error = init_threads(sdp);
165 	if (error)
166 		return error;
167 
168 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
169 				   &freeze_gh);
170 	if (error)
171 		goto fail_threads;
172 
173 	j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
174 
175 	error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
176 	if (error)
177 		goto fail;
178 
179 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
180 		gfs2_consist(sdp);
181 		error = -EIO;
182 		goto fail;
183 	}
184 
185 	/*  Initialize some head of the log stuff  */
186 	sdp->sd_log_sequence = head.lh_sequence + 1;
187 	gfs2_log_pointers_init(sdp, head.lh_blkno);
188 
189 	error = gfs2_quota_init(sdp);
190 	if (error)
191 		goto fail;
192 
193 	set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
194 
195 	gfs2_glock_dq_uninit(&freeze_gh);
196 
197 	return 0;
198 
199 fail:
200 	freeze_gh.gh_flags |= GL_NOCACHE;
201 	gfs2_glock_dq_uninit(&freeze_gh);
202 fail_threads:
203 	if (sdp->sd_quotad_process)
204 		kthread_stop(sdp->sd_quotad_process);
205 	sdp->sd_quotad_process = NULL;
206 	if (sdp->sd_logd_process)
207 		kthread_stop(sdp->sd_logd_process);
208 	sdp->sd_logd_process = NULL;
209 	return error;
210 }
211 
212 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
213 {
214 	const struct gfs2_statfs_change *str = buf;
215 
216 	sc->sc_total = be64_to_cpu(str->sc_total);
217 	sc->sc_free = be64_to_cpu(str->sc_free);
218 	sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
219 }
220 
221 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
222 {
223 	struct gfs2_statfs_change *str = buf;
224 
225 	str->sc_total = cpu_to_be64(sc->sc_total);
226 	str->sc_free = cpu_to_be64(sc->sc_free);
227 	str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
228 }
229 
230 int gfs2_statfs_init(struct gfs2_sbd *sdp)
231 {
232 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
233 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
234 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
235 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
236 	struct buffer_head *m_bh, *l_bh;
237 	struct gfs2_holder gh;
238 	int error;
239 
240 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
241 				   &gh);
242 	if (error)
243 		return error;
244 
245 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
246 	if (error)
247 		goto out;
248 
249 	if (sdp->sd_args.ar_spectator) {
250 		spin_lock(&sdp->sd_statfs_spin);
251 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
252 				      sizeof(struct gfs2_dinode));
253 		spin_unlock(&sdp->sd_statfs_spin);
254 	} else {
255 		error = gfs2_meta_inode_buffer(l_ip, &l_bh);
256 		if (error)
257 			goto out_m_bh;
258 
259 		spin_lock(&sdp->sd_statfs_spin);
260 		gfs2_statfs_change_in(m_sc, m_bh->b_data +
261 				      sizeof(struct gfs2_dinode));
262 		gfs2_statfs_change_in(l_sc, l_bh->b_data +
263 				      sizeof(struct gfs2_dinode));
264 		spin_unlock(&sdp->sd_statfs_spin);
265 
266 		brelse(l_bh);
267 	}
268 
269 out_m_bh:
270 	brelse(m_bh);
271 out:
272 	gfs2_glock_dq_uninit(&gh);
273 	return 0;
274 }
275 
276 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
277 			s64 dinodes)
278 {
279 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
280 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
281 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
282 	struct buffer_head *l_bh;
283 	s64 x, y;
284 	int need_sync = 0;
285 	int error;
286 
287 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
288 	if (error)
289 		return;
290 
291 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
292 
293 	spin_lock(&sdp->sd_statfs_spin);
294 	l_sc->sc_total += total;
295 	l_sc->sc_free += free;
296 	l_sc->sc_dinodes += dinodes;
297 	gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
298 	if (sdp->sd_args.ar_statfs_percent) {
299 		x = 100 * l_sc->sc_free;
300 		y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
301 		if (x >= y || x <= -y)
302 			need_sync = 1;
303 	}
304 	spin_unlock(&sdp->sd_statfs_spin);
305 
306 	brelse(l_bh);
307 	if (need_sync)
308 		gfs2_wake_up_statfs(sdp);
309 }
310 
311 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
312 		   struct buffer_head *l_bh)
313 {
314 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
315 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
316 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
317 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
318 
319 	gfs2_trans_add_meta(l_ip->i_gl, l_bh);
320 	gfs2_trans_add_meta(m_ip->i_gl, m_bh);
321 
322 	spin_lock(&sdp->sd_statfs_spin);
323 	m_sc->sc_total += l_sc->sc_total;
324 	m_sc->sc_free += l_sc->sc_free;
325 	m_sc->sc_dinodes += l_sc->sc_dinodes;
326 	memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
327 	memset(l_bh->b_data + sizeof(struct gfs2_dinode),
328 	       0, sizeof(struct gfs2_statfs_change));
329 	gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
330 	spin_unlock(&sdp->sd_statfs_spin);
331 }
332 
333 int gfs2_statfs_sync(struct super_block *sb, int type)
334 {
335 	struct gfs2_sbd *sdp = sb->s_fs_info;
336 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
337 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
338 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
339 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
340 	struct gfs2_holder gh;
341 	struct buffer_head *m_bh, *l_bh;
342 	int error;
343 
344 	sb_start_write(sb);
345 	error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
346 				   &gh);
347 	if (error)
348 		goto out;
349 
350 	error = gfs2_meta_inode_buffer(m_ip, &m_bh);
351 	if (error)
352 		goto out_unlock;
353 
354 	spin_lock(&sdp->sd_statfs_spin);
355 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
356 			      sizeof(struct gfs2_dinode));
357 	if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
358 		spin_unlock(&sdp->sd_statfs_spin);
359 		goto out_bh;
360 	}
361 	spin_unlock(&sdp->sd_statfs_spin);
362 
363 	error = gfs2_meta_inode_buffer(l_ip, &l_bh);
364 	if (error)
365 		goto out_bh;
366 
367 	error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
368 	if (error)
369 		goto out_bh2;
370 
371 	update_statfs(sdp, m_bh, l_bh);
372 	sdp->sd_statfs_force_sync = 0;
373 
374 	gfs2_trans_end(sdp);
375 
376 out_bh2:
377 	brelse(l_bh);
378 out_bh:
379 	brelse(m_bh);
380 out_unlock:
381 	gfs2_glock_dq_uninit(&gh);
382 out:
383 	sb_end_write(sb);
384 	return error;
385 }
386 
387 struct lfcc {
388 	struct list_head list;
389 	struct gfs2_holder gh;
390 };
391 
392 /**
393  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
394  *                            journals are clean
395  * @sdp: the file system
396  * @state: the state to put the transaction lock into
397  * @t_gh: the hold on the transaction lock
398  *
399  * Returns: errno
400  */
401 
402 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
403 				    struct gfs2_holder *freeze_gh)
404 {
405 	struct gfs2_inode *ip;
406 	struct gfs2_jdesc *jd;
407 	struct lfcc *lfcc;
408 	LIST_HEAD(list);
409 	struct gfs2_log_header_host lh;
410 	int error;
411 
412 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
413 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
414 		if (!lfcc) {
415 			error = -ENOMEM;
416 			goto out;
417 		}
418 		ip = GFS2_I(jd->jd_inode);
419 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
420 		if (error) {
421 			kfree(lfcc);
422 			goto out;
423 		}
424 		list_add(&lfcc->list, &list);
425 	}
426 
427 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
428 				   GL_NOCACHE, freeze_gh);
429 
430 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
431 		error = gfs2_jdesc_check(jd);
432 		if (error)
433 			break;
434 		error = gfs2_find_jhead(jd, &lh, false);
435 		if (error)
436 			break;
437 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
438 			error = -EBUSY;
439 			break;
440 		}
441 	}
442 
443 	if (error)
444 		gfs2_glock_dq_uninit(freeze_gh);
445 
446 out:
447 	while (!list_empty(&list)) {
448 		lfcc = list_entry(list.next, struct lfcc, list);
449 		list_del(&lfcc->list);
450 		gfs2_glock_dq_uninit(&lfcc->gh);
451 		kfree(lfcc);
452 	}
453 	return error;
454 }
455 
456 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
457 {
458 	struct gfs2_dinode *str = buf;
459 
460 	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
461 	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
462 	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
463 	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
464 	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
465 	str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
466 	str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
467 	str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
468 	str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
469 	str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
470 	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
471 	str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
472 	str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
473 	str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
474 
475 	str->di_goal_meta = cpu_to_be64(ip->i_goal);
476 	str->di_goal_data = cpu_to_be64(ip->i_goal);
477 	str->di_generation = cpu_to_be64(ip->i_generation);
478 
479 	str->di_flags = cpu_to_be32(ip->i_diskflags);
480 	str->di_height = cpu_to_be16(ip->i_height);
481 	str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
482 					     !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
483 					     GFS2_FORMAT_DE : 0);
484 	str->di_depth = cpu_to_be16(ip->i_depth);
485 	str->di_entries = cpu_to_be32(ip->i_entries);
486 
487 	str->di_eattr = cpu_to_be64(ip->i_eattr);
488 	str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
489 	str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
490 	str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
491 }
492 
493 /**
494  * gfs2_write_inode - Make sure the inode is stable on the disk
495  * @inode: The inode
496  * @wbc: The writeback control structure
497  *
498  * Returns: errno
499  */
500 
501 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
502 {
503 	struct gfs2_inode *ip = GFS2_I(inode);
504 	struct gfs2_sbd *sdp = GFS2_SB(inode);
505 	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
506 	struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
507 	int ret = 0;
508 	bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
509 
510 	if (flush_all)
511 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
512 			       GFS2_LOG_HEAD_FLUSH_NORMAL |
513 			       GFS2_LFC_WRITE_INODE);
514 	if (bdi->wb.dirty_exceeded)
515 		gfs2_ail1_flush(sdp, wbc);
516 	else
517 		filemap_fdatawrite(metamapping);
518 	if (flush_all)
519 		ret = filemap_fdatawait(metamapping);
520 	if (ret)
521 		mark_inode_dirty_sync(inode);
522 	else {
523 		spin_lock(&inode->i_lock);
524 		if (!(inode->i_flags & I_DIRTY))
525 			gfs2_ordered_del_inode(ip);
526 		spin_unlock(&inode->i_lock);
527 	}
528 	return ret;
529 }
530 
531 /**
532  * gfs2_dirty_inode - check for atime updates
533  * @inode: The inode in question
534  * @flags: The type of dirty
535  *
536  * Unfortunately it can be called under any combination of inode
537  * glock and transaction lock, so we have to check carefully.
538  *
539  * At the moment this deals only with atime - it should be possible
540  * to expand that role in future, once a review of the locking has
541  * been carried out.
542  */
543 
544 static void gfs2_dirty_inode(struct inode *inode, int flags)
545 {
546 	struct gfs2_inode *ip = GFS2_I(inode);
547 	struct gfs2_sbd *sdp = GFS2_SB(inode);
548 	struct buffer_head *bh;
549 	struct gfs2_holder gh;
550 	int need_unlock = 0;
551 	int need_endtrans = 0;
552 	int ret;
553 
554 	if (!(flags & I_DIRTY_INODE))
555 		return;
556 	if (unlikely(test_bit(SDF_WITHDRAWN, &sdp->sd_flags)))
557 		return;
558 	if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
559 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
560 		if (ret) {
561 			fs_err(sdp, "dirty_inode: glock %d\n", ret);
562 			return;
563 		}
564 		need_unlock = 1;
565 	} else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
566 		return;
567 
568 	if (current->journal_info == NULL) {
569 		ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
570 		if (ret) {
571 			fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
572 			goto out;
573 		}
574 		need_endtrans = 1;
575 	}
576 
577 	ret = gfs2_meta_inode_buffer(ip, &bh);
578 	if (ret == 0) {
579 		gfs2_trans_add_meta(ip->i_gl, bh);
580 		gfs2_dinode_out(ip, bh->b_data);
581 		brelse(bh);
582 	}
583 
584 	if (need_endtrans)
585 		gfs2_trans_end(sdp);
586 out:
587 	if (need_unlock)
588 		gfs2_glock_dq_uninit(&gh);
589 }
590 
591 /**
592  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
593  * @sdp: the filesystem
594  *
595  * Returns: errno
596  */
597 
598 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
599 {
600 	struct gfs2_holder freeze_gh;
601 	int error;
602 
603 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
604 				   &freeze_gh);
605 	if (error && !test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
606 		return error;
607 
608 	flush_workqueue(gfs2_delete_workqueue);
609 	if (sdp->sd_quotad_process)
610 		kthread_stop(sdp->sd_quotad_process);
611 	sdp->sd_quotad_process = NULL;
612 	if (sdp->sd_logd_process)
613 		kthread_stop(sdp->sd_logd_process);
614 	sdp->sd_logd_process = NULL;
615 
616 	gfs2_quota_sync(sdp->sd_vfs, 0);
617 	gfs2_statfs_sync(sdp->sd_vfs, 0);
618 
619 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
620 		       GFS2_LFC_MAKE_FS_RO);
621 	wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
622 	gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
623 
624 	if (gfs2_holder_initialized(&freeze_gh))
625 		gfs2_glock_dq_uninit(&freeze_gh);
626 
627 	gfs2_quota_cleanup(sdp);
628 
629 	return error;
630 }
631 
632 /**
633  * gfs2_put_super - Unmount the filesystem
634  * @sb: The VFS superblock
635  *
636  */
637 
638 static void gfs2_put_super(struct super_block *sb)
639 {
640 	struct gfs2_sbd *sdp = sb->s_fs_info;
641 	int error;
642 	struct gfs2_jdesc *jd;
643 
644 	/* No more recovery requests */
645 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
646 	smp_mb();
647 
648 	/* Wait on outstanding recovery */
649 restart:
650 	spin_lock(&sdp->sd_jindex_spin);
651 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
652 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
653 			continue;
654 		spin_unlock(&sdp->sd_jindex_spin);
655 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
656 			    TASK_UNINTERRUPTIBLE);
657 		goto restart;
658 	}
659 	spin_unlock(&sdp->sd_jindex_spin);
660 
661 	if (!sb_rdonly(sb)) {
662 		error = gfs2_make_fs_ro(sdp);
663 		if (error)
664 			gfs2_io_error(sdp);
665 	}
666 	/*  At this point, we're through modifying the disk  */
667 
668 	/*  Release stuff  */
669 
670 	iput(sdp->sd_jindex);
671 	iput(sdp->sd_statfs_inode);
672 	iput(sdp->sd_rindex);
673 	iput(sdp->sd_quota_inode);
674 
675 	gfs2_glock_put(sdp->sd_rename_gl);
676 	gfs2_glock_put(sdp->sd_freeze_gl);
677 
678 	if (!sdp->sd_args.ar_spectator) {
679 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
680 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
681 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
682 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
683 		iput(sdp->sd_sc_inode);
684 		iput(sdp->sd_qc_inode);
685 	}
686 
687 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
688 	gfs2_clear_rgrpd(sdp);
689 	gfs2_jindex_free(sdp);
690 	/*  Take apart glock structures and buffer lists  */
691 	gfs2_gl_hash_clear(sdp);
692 	gfs2_delete_debugfs_file(sdp);
693 	/*  Unmount the locking protocol  */
694 	gfs2_lm_unmount(sdp);
695 
696 	/*  At this point, we're through participating in the lockspace  */
697 	gfs2_sys_fs_del(sdp);
698 }
699 
700 /**
701  * gfs2_sync_fs - sync the filesystem
702  * @sb: the superblock
703  *
704  * Flushes the log to disk.
705  */
706 
707 static int gfs2_sync_fs(struct super_block *sb, int wait)
708 {
709 	struct gfs2_sbd *sdp = sb->s_fs_info;
710 
711 	gfs2_quota_sync(sb, -1);
712 	if (wait)
713 		gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
714 			       GFS2_LFC_SYNC_FS);
715 	return sdp->sd_log_error;
716 }
717 
718 void gfs2_freeze_func(struct work_struct *work)
719 {
720 	int error;
721 	struct gfs2_holder freeze_gh;
722 	struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
723 	struct super_block *sb = sdp->sd_vfs;
724 
725 	atomic_inc(&sb->s_active);
726 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
727 				   &freeze_gh);
728 	if (error) {
729 		fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error);
730 		gfs2_assert_withdraw(sdp, 0);
731 	} else {
732 		atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
733 		error = thaw_super(sb);
734 		if (error) {
735 			fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
736 				error);
737 			gfs2_assert_withdraw(sdp, 0);
738 		}
739 		if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
740 			freeze_gh.gh_flags |= GL_NOCACHE;
741 		gfs2_glock_dq_uninit(&freeze_gh);
742 	}
743 	deactivate_super(sb);
744 	clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
745 	wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
746 	return;
747 }
748 
749 /**
750  * gfs2_freeze - prevent further writes to the filesystem
751  * @sb: the VFS structure for the filesystem
752  *
753  */
754 
755 static int gfs2_freeze(struct super_block *sb)
756 {
757 	struct gfs2_sbd *sdp = sb->s_fs_info;
758 	int error = 0;
759 
760 	mutex_lock(&sdp->sd_freeze_mutex);
761 	if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
762 		goto out;
763 
764 	if (test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
765 		error = -EINVAL;
766 		goto out;
767 	}
768 
769 	for (;;) {
770 		error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
771 		if (!error)
772 			break;
773 
774 		if (error == -EBUSY)
775 			fs_err(sdp, "waiting for recovery before freeze\n");
776 		else
777 			fs_err(sdp, "error freezing FS: %d\n", error);
778 
779 		fs_err(sdp, "retrying...\n");
780 		msleep(1000);
781 	}
782 	set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
783 out:
784 	mutex_unlock(&sdp->sd_freeze_mutex);
785 	return error;
786 }
787 
788 /**
789  * gfs2_unfreeze - reallow writes to the filesystem
790  * @sb: the VFS structure for the filesystem
791  *
792  */
793 
794 static int gfs2_unfreeze(struct super_block *sb)
795 {
796 	struct gfs2_sbd *sdp = sb->s_fs_info;
797 
798 	mutex_lock(&sdp->sd_freeze_mutex);
799         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
800 	    !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
801 		mutex_unlock(&sdp->sd_freeze_mutex);
802                 return 0;
803 	}
804 
805 	gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
806 	mutex_unlock(&sdp->sd_freeze_mutex);
807 	return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
808 }
809 
810 /**
811  * statfs_fill - fill in the sg for a given RG
812  * @rgd: the RG
813  * @sc: the sc structure
814  *
815  * Returns: 0 on success, -ESTALE if the LVB is invalid
816  */
817 
818 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
819 			    struct gfs2_statfs_change_host *sc)
820 {
821 	gfs2_rgrp_verify(rgd);
822 	sc->sc_total += rgd->rd_data;
823 	sc->sc_free += rgd->rd_free;
824 	sc->sc_dinodes += rgd->rd_dinodes;
825 	return 0;
826 }
827 
828 /**
829  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
830  * @sdp: the filesystem
831  * @sc: the sc info that will be returned
832  *
833  * Any error (other than a signal) will cause this routine to fall back
834  * to the synchronous version.
835  *
836  * FIXME: This really shouldn't busy wait like this.
837  *
838  * Returns: errno
839  */
840 
841 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
842 {
843 	struct gfs2_rgrpd *rgd_next;
844 	struct gfs2_holder *gha, *gh;
845 	unsigned int slots = 64;
846 	unsigned int x;
847 	int done;
848 	int error = 0, err;
849 
850 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
851 	gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
852 	if (!gha)
853 		return -ENOMEM;
854 	for (x = 0; x < slots; x++)
855 		gfs2_holder_mark_uninitialized(gha + x);
856 
857 	rgd_next = gfs2_rgrpd_get_first(sdp);
858 
859 	for (;;) {
860 		done = 1;
861 
862 		for (x = 0; x < slots; x++) {
863 			gh = gha + x;
864 
865 			if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
866 				err = gfs2_glock_wait(gh);
867 				if (err) {
868 					gfs2_holder_uninit(gh);
869 					error = err;
870 				} else {
871 					if (!error) {
872 						struct gfs2_rgrpd *rgd =
873 							gfs2_glock2rgrp(gh->gh_gl);
874 
875 						error = statfs_slow_fill(rgd, sc);
876 					}
877 					gfs2_glock_dq_uninit(gh);
878 				}
879 			}
880 
881 			if (gfs2_holder_initialized(gh))
882 				done = 0;
883 			else if (rgd_next && !error) {
884 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
885 							   LM_ST_SHARED,
886 							   GL_ASYNC,
887 							   gh);
888 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
889 				done = 0;
890 			}
891 
892 			if (signal_pending(current))
893 				error = -ERESTARTSYS;
894 		}
895 
896 		if (done)
897 			break;
898 
899 		yield();
900 	}
901 
902 	kfree(gha);
903 	return error;
904 }
905 
906 /**
907  * gfs2_statfs_i - Do a statfs
908  * @sdp: the filesystem
909  * @sg: the sg structure
910  *
911  * Returns: errno
912  */
913 
914 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
915 {
916 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
917 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
918 
919 	spin_lock(&sdp->sd_statfs_spin);
920 
921 	*sc = *m_sc;
922 	sc->sc_total += l_sc->sc_total;
923 	sc->sc_free += l_sc->sc_free;
924 	sc->sc_dinodes += l_sc->sc_dinodes;
925 
926 	spin_unlock(&sdp->sd_statfs_spin);
927 
928 	if (sc->sc_free < 0)
929 		sc->sc_free = 0;
930 	if (sc->sc_free > sc->sc_total)
931 		sc->sc_free = sc->sc_total;
932 	if (sc->sc_dinodes < 0)
933 		sc->sc_dinodes = 0;
934 
935 	return 0;
936 }
937 
938 /**
939  * gfs2_statfs - Gather and return stats about the filesystem
940  * @sb: The superblock
941  * @statfsbuf: The buffer
942  *
943  * Returns: 0 on success or error code
944  */
945 
946 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
947 {
948 	struct super_block *sb = dentry->d_sb;
949 	struct gfs2_sbd *sdp = sb->s_fs_info;
950 	struct gfs2_statfs_change_host sc;
951 	int error;
952 
953 	error = gfs2_rindex_update(sdp);
954 	if (error)
955 		return error;
956 
957 	if (gfs2_tune_get(sdp, gt_statfs_slow))
958 		error = gfs2_statfs_slow(sdp, &sc);
959 	else
960 		error = gfs2_statfs_i(sdp, &sc);
961 
962 	if (error)
963 		return error;
964 
965 	buf->f_type = GFS2_MAGIC;
966 	buf->f_bsize = sdp->sd_sb.sb_bsize;
967 	buf->f_blocks = sc.sc_total;
968 	buf->f_bfree = sc.sc_free;
969 	buf->f_bavail = sc.sc_free;
970 	buf->f_files = sc.sc_dinodes + sc.sc_free;
971 	buf->f_ffree = sc.sc_free;
972 	buf->f_namelen = GFS2_FNAMESIZE;
973 
974 	return 0;
975 }
976 
977 /**
978  * gfs2_drop_inode - Drop an inode (test for remote unlink)
979  * @inode: The inode to drop
980  *
981  * If we've received a callback on an iopen lock then it's because a
982  * remote node tried to deallocate the inode but failed due to this node
983  * still having the inode open. Here we mark the link count zero
984  * since we know that it must have reached zero if the GLF_DEMOTE flag
985  * is set on the iopen glock. If we didn't do a disk read since the
986  * remote node removed the final link then we might otherwise miss
987  * this event. This check ensures that this node will deallocate the
988  * inode's blocks, or alternatively pass the baton on to another
989  * node for later deallocation.
990  */
991 
992 static int gfs2_drop_inode(struct inode *inode)
993 {
994 	struct gfs2_inode *ip = GFS2_I(inode);
995 
996 	if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
997 	    inode->i_nlink &&
998 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
999 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1000 		if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1001 			clear_nlink(inode);
1002 	}
1003 
1004 	/*
1005 	 * When under memory pressure when an inode's link count has dropped to
1006 	 * zero, defer deleting the inode to the delete workqueue.  This avoids
1007 	 * calling into DLM under memory pressure, which can deadlock.
1008 	 */
1009 	if (!inode->i_nlink &&
1010 	    unlikely(current->flags & PF_MEMALLOC) &&
1011 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
1012 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1013 
1014 		gfs2_glock_hold(gl);
1015 		if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1016 			gfs2_glock_queue_put(gl);
1017 		return false;
1018 	}
1019 
1020 	return generic_drop_inode(inode);
1021 }
1022 
1023 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1024 {
1025 	do {
1026 		if (d1 == d2)
1027 			return 1;
1028 		d1 = d1->d_parent;
1029 	} while (!IS_ROOT(d1));
1030 	return 0;
1031 }
1032 
1033 /**
1034  * gfs2_show_options - Show mount options for /proc/mounts
1035  * @s: seq_file structure
1036  * @root: root of this (sub)tree
1037  *
1038  * Returns: 0 on success or error code
1039  */
1040 
1041 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1042 {
1043 	struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1044 	struct gfs2_args *args = &sdp->sd_args;
1045 	int val;
1046 
1047 	if (is_ancestor(root, sdp->sd_master_dir))
1048 		seq_puts(s, ",meta");
1049 	if (args->ar_lockproto[0])
1050 		seq_show_option(s, "lockproto", args->ar_lockproto);
1051 	if (args->ar_locktable[0])
1052 		seq_show_option(s, "locktable", args->ar_locktable);
1053 	if (args->ar_hostdata[0])
1054 		seq_show_option(s, "hostdata", args->ar_hostdata);
1055 	if (args->ar_spectator)
1056 		seq_puts(s, ",spectator");
1057 	if (args->ar_localflocks)
1058 		seq_puts(s, ",localflocks");
1059 	if (args->ar_debug)
1060 		seq_puts(s, ",debug");
1061 	if (args->ar_posix_acl)
1062 		seq_puts(s, ",acl");
1063 	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1064 		char *state;
1065 		switch (args->ar_quota) {
1066 		case GFS2_QUOTA_OFF:
1067 			state = "off";
1068 			break;
1069 		case GFS2_QUOTA_ACCOUNT:
1070 			state = "account";
1071 			break;
1072 		case GFS2_QUOTA_ON:
1073 			state = "on";
1074 			break;
1075 		default:
1076 			state = "unknown";
1077 			break;
1078 		}
1079 		seq_printf(s, ",quota=%s", state);
1080 	}
1081 	if (args->ar_suiddir)
1082 		seq_puts(s, ",suiddir");
1083 	if (args->ar_data != GFS2_DATA_DEFAULT) {
1084 		char *state;
1085 		switch (args->ar_data) {
1086 		case GFS2_DATA_WRITEBACK:
1087 			state = "writeback";
1088 			break;
1089 		case GFS2_DATA_ORDERED:
1090 			state = "ordered";
1091 			break;
1092 		default:
1093 			state = "unknown";
1094 			break;
1095 		}
1096 		seq_printf(s, ",data=%s", state);
1097 	}
1098 	if (args->ar_discard)
1099 		seq_puts(s, ",discard");
1100 	val = sdp->sd_tune.gt_logd_secs;
1101 	if (val != 30)
1102 		seq_printf(s, ",commit=%d", val);
1103 	val = sdp->sd_tune.gt_statfs_quantum;
1104 	if (val != 30)
1105 		seq_printf(s, ",statfs_quantum=%d", val);
1106 	else if (sdp->sd_tune.gt_statfs_slow)
1107 		seq_puts(s, ",statfs_quantum=0");
1108 	val = sdp->sd_tune.gt_quota_quantum;
1109 	if (val != 60)
1110 		seq_printf(s, ",quota_quantum=%d", val);
1111 	if (args->ar_statfs_percent)
1112 		seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1113 	if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1114 		const char *state;
1115 
1116 		switch (args->ar_errors) {
1117 		case GFS2_ERRORS_WITHDRAW:
1118 			state = "withdraw";
1119 			break;
1120 		case GFS2_ERRORS_PANIC:
1121 			state = "panic";
1122 			break;
1123 		default:
1124 			state = "unknown";
1125 			break;
1126 		}
1127 		seq_printf(s, ",errors=%s", state);
1128 	}
1129 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1130 		seq_puts(s, ",nobarrier");
1131 	if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1132 		seq_puts(s, ",demote_interface_used");
1133 	if (args->ar_rgrplvb)
1134 		seq_puts(s, ",rgrplvb");
1135 	if (args->ar_loccookie)
1136 		seq_puts(s, ",loccookie");
1137 	return 0;
1138 }
1139 
1140 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1141 {
1142 	struct inode *inode = &ip->i_inode;
1143 	struct gfs2_glock *gl = ip->i_gl;
1144 
1145 	truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1146 	truncate_inode_pages(&inode->i_data, 0);
1147 
1148 	if (atomic_read(&gl->gl_revokes) == 0) {
1149 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
1150 		clear_bit(GLF_DIRTY, &gl->gl_flags);
1151 	}
1152 }
1153 
1154 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1155 {
1156 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1157 	struct gfs2_rgrpd *rgd;
1158 	struct gfs2_holder gh;
1159 	int error;
1160 
1161 	if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1162 		gfs2_consist_inode(ip);
1163 		return -EIO;
1164 	}
1165 
1166 	error = gfs2_rindex_update(sdp);
1167 	if (error)
1168 		return error;
1169 
1170 	error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1171 	if (error)
1172 		return error;
1173 
1174 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1175 	if (!rgd) {
1176 		gfs2_consist_inode(ip);
1177 		error = -EIO;
1178 		goto out_qs;
1179 	}
1180 
1181 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &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 /**
1220  * gfs2_evict_inode - Remove an inode from cache
1221  * @inode: The inode to evict
1222  *
1223  * There are three cases to consider:
1224  * 1. i_nlink == 0, we are final opener (and must deallocate)
1225  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1226  * 3. i_nlink > 0
1227  *
1228  * If the fs is read only, then we have to treat all cases as per #3
1229  * since we are unable to do any deallocation. The inode will be
1230  * deallocated by the next read/write node to attempt an allocation
1231  * in the same resource group
1232  *
1233  * We have to (at the moment) hold the inodes main lock to cover
1234  * the gap between unlocking the shared lock on the iopen lock and
1235  * taking the exclusive lock. I'd rather do a shared -> exclusive
1236  * conversion on the iopen lock, but we can change that later. This
1237  * is safe, just less efficient.
1238  */
1239 
1240 static void gfs2_evict_inode(struct inode *inode)
1241 {
1242 	struct super_block *sb = inode->i_sb;
1243 	struct gfs2_sbd *sdp = sb->s_fs_info;
1244 	struct gfs2_inode *ip = GFS2_I(inode);
1245 	struct gfs2_holder gh;
1246 	struct address_space *metamapping;
1247 	int error;
1248 
1249 	if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1250 		clear_inode(inode);
1251 		return;
1252 	}
1253 
1254 	if (inode->i_nlink || sb_rdonly(sb))
1255 		goto out;
1256 
1257 	if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1258 		BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1259 		gfs2_holder_mark_uninitialized(&gh);
1260 		goto alloc_failed;
1261 	}
1262 
1263 	/* Deletes should never happen under memory pressure anymore.  */
1264 	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1265 		goto out;
1266 
1267 	/* Must not read inode block until block type has been verified */
1268 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1269 	if (unlikely(error)) {
1270 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1271 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1272 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1273 		goto out;
1274 	}
1275 
1276 	error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1277 	if (error)
1278 		goto out_truncate;
1279 
1280 	if (test_bit(GIF_INVALID, &ip->i_flags)) {
1281 		error = gfs2_inode_refresh(ip);
1282 		if (error)
1283 			goto out_truncate;
1284 	}
1285 
1286 	/*
1287 	 * The inode may have been recreated in the meantime.
1288 	 */
1289 	if (inode->i_nlink)
1290 		goto out_truncate;
1291 
1292 alloc_failed:
1293 	if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1294 	    test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1295 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1296 		gfs2_glock_dq_wait(&ip->i_iopen_gh);
1297 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1298 				   &ip->i_iopen_gh);
1299 		error = gfs2_glock_nq(&ip->i_iopen_gh);
1300 		if (error)
1301 			goto out_truncate;
1302 	}
1303 
1304 	if (S_ISDIR(inode->i_mode) &&
1305 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1306 		error = gfs2_dir_exhash_dealloc(ip);
1307 		if (error)
1308 			goto out_unlock;
1309 	}
1310 
1311 	if (ip->i_eattr) {
1312 		error = gfs2_ea_dealloc(ip);
1313 		if (error)
1314 			goto out_unlock;
1315 	}
1316 
1317 	if (!gfs2_is_stuffed(ip)) {
1318 		error = gfs2_file_dealloc(ip);
1319 		if (error)
1320 			goto out_unlock;
1321 	}
1322 
1323 	/* We're about to clear the bitmap for the dinode, but as soon as we
1324 	   do, gfs2_create_inode can create another inode at the same block
1325 	   location and try to set gl_object again. We clear gl_object here so
1326 	   that subsequent inode creates don't see an old gl_object. */
1327 	glock_clear_object(ip->i_gl, ip);
1328 	error = gfs2_dinode_dealloc(ip);
1329 	goto out_unlock;
1330 
1331 out_truncate:
1332 	gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1333 		       GFS2_LFC_EVICT_INODE);
1334 	metamapping = gfs2_glock2aspace(ip->i_gl);
1335 	if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1336 		filemap_fdatawrite(metamapping);
1337 		filemap_fdatawait(metamapping);
1338 	}
1339 	write_inode_now(inode, 1);
1340 	gfs2_ail_flush(ip->i_gl, 0);
1341 
1342 	error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1343 	if (error)
1344 		goto out_unlock;
1345 	/* Needs to be done before glock release & also in a transaction */
1346 	truncate_inode_pages(&inode->i_data, 0);
1347 	truncate_inode_pages(metamapping, 0);
1348 	gfs2_trans_end(sdp);
1349 
1350 out_unlock:
1351 	if (gfs2_rs_active(&ip->i_res))
1352 		gfs2_rs_deltree(&ip->i_res);
1353 
1354 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1355 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1356 		if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1357 			ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1358 			gfs2_glock_dq(&ip->i_iopen_gh);
1359 		}
1360 		gfs2_holder_uninit(&ip->i_iopen_gh);
1361 	}
1362 	if (gfs2_holder_initialized(&gh)) {
1363 		glock_clear_object(ip->i_gl, ip);
1364 		gfs2_glock_dq_uninit(&gh);
1365 	}
1366 	if (error && error != GLR_TRYFAILED && error != -EROFS)
1367 		fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1368 out:
1369 	truncate_inode_pages_final(&inode->i_data);
1370 	gfs2_rsqa_delete(ip, NULL);
1371 	gfs2_ordered_del_inode(ip);
1372 	clear_inode(inode);
1373 	gfs2_dir_hash_inval(ip);
1374 	glock_clear_object(ip->i_gl, ip);
1375 	wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1376 	gfs2_glock_add_to_lru(ip->i_gl);
1377 	gfs2_glock_put_eventually(ip->i_gl);
1378 	ip->i_gl = NULL;
1379 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1380 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1381 
1382 		glock_clear_object(gl, ip);
1383 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1384 		gfs2_glock_hold(gl);
1385 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1386 		gfs2_glock_put_eventually(gl);
1387 	}
1388 }
1389 
1390 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1391 {
1392 	struct gfs2_inode *ip;
1393 
1394 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1395 	if (!ip)
1396 		return NULL;
1397 	ip->i_flags = 0;
1398 	ip->i_gl = NULL;
1399 	memset(&ip->i_res, 0, sizeof(ip->i_res));
1400 	RB_CLEAR_NODE(&ip->i_res.rs_node);
1401 	ip->i_rahead = 0;
1402 	return &ip->i_inode;
1403 }
1404 
1405 static void gfs2_free_inode(struct inode *inode)
1406 {
1407 	kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1408 }
1409 
1410 const struct super_operations gfs2_super_ops = {
1411 	.alloc_inode		= gfs2_alloc_inode,
1412 	.free_inode		= gfs2_free_inode,
1413 	.write_inode		= gfs2_write_inode,
1414 	.dirty_inode		= gfs2_dirty_inode,
1415 	.evict_inode		= gfs2_evict_inode,
1416 	.put_super		= gfs2_put_super,
1417 	.sync_fs		= gfs2_sync_fs,
1418 	.freeze_super		= gfs2_freeze,
1419 	.thaw_super		= gfs2_unfreeze,
1420 	.statfs			= gfs2_statfs,
1421 	.drop_inode		= gfs2_drop_inode,
1422 	.show_options		= gfs2_show_options,
1423 };
1424 
1425