xref: /linux/fs/gfs2/super.c (revision 9a6b55ac)
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 {
404 	struct gfs2_inode *ip;
405 	struct gfs2_jdesc *jd;
406 	struct lfcc *lfcc;
407 	LIST_HEAD(list);
408 	struct gfs2_log_header_host lh;
409 	int error;
410 
411 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
412 		lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
413 		if (!lfcc) {
414 			error = -ENOMEM;
415 			goto out;
416 		}
417 		ip = GFS2_I(jd->jd_inode);
418 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
419 		if (error) {
420 			kfree(lfcc);
421 			goto out;
422 		}
423 		list_add(&lfcc->list, &list);
424 	}
425 
426 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
427 				   GL_NOCACHE, &sdp->sd_freeze_gh);
428 	if (error)
429 		goto out;
430 
431 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
432 		error = gfs2_jdesc_check(jd);
433 		if (error)
434 			break;
435 		error = gfs2_find_jhead(jd, &lh, false);
436 		if (error)
437 			break;
438 		if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
439 			error = -EBUSY;
440 			break;
441 		}
442 	}
443 
444 	if (error)
445 		gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
446 
447 out:
448 	while (!list_empty(&list)) {
449 		lfcc = list_entry(list.next, struct lfcc, list);
450 		list_del(&lfcc->list);
451 		gfs2_glock_dq_uninit(&lfcc->gh);
452 		kfree(lfcc);
453 	}
454 	return error;
455 }
456 
457 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
458 {
459 	struct gfs2_dinode *str = buf;
460 
461 	str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
462 	str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
463 	str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
464 	str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
465 	str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
466 	str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
467 	str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
468 	str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
469 	str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
470 	str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
471 	str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
472 	str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
473 	str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
474 	str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
475 
476 	str->di_goal_meta = cpu_to_be64(ip->i_goal);
477 	str->di_goal_data = cpu_to_be64(ip->i_goal);
478 	str->di_generation = cpu_to_be64(ip->i_generation);
479 
480 	str->di_flags = cpu_to_be32(ip->i_diskflags);
481 	str->di_height = cpu_to_be16(ip->i_height);
482 	str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
483 					     !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
484 					     GFS2_FORMAT_DE : 0);
485 	str->di_depth = cpu_to_be16(ip->i_depth);
486 	str->di_entries = cpu_to_be32(ip->i_entries);
487 
488 	str->di_eattr = cpu_to_be64(ip->i_eattr);
489 	str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
490 	str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
491 	str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
492 }
493 
494 /**
495  * gfs2_write_inode - Make sure the inode is stable on the disk
496  * @inode: The inode
497  * @wbc: The writeback control structure
498  *
499  * Returns: errno
500  */
501 
502 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
503 {
504 	struct gfs2_inode *ip = GFS2_I(inode);
505 	struct gfs2_sbd *sdp = GFS2_SB(inode);
506 	struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
507 	struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
508 	int ret = 0;
509 	bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
510 
511 	if (flush_all)
512 		gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
513 			       GFS2_LOG_HEAD_FLUSH_NORMAL |
514 			       GFS2_LFC_WRITE_INODE);
515 	if (bdi->wb.dirty_exceeded)
516 		gfs2_ail1_flush(sdp, wbc);
517 	else
518 		filemap_fdatawrite(metamapping);
519 	if (flush_all)
520 		ret = filemap_fdatawait(metamapping);
521 	if (ret)
522 		mark_inode_dirty_sync(inode);
523 	else {
524 		spin_lock(&inode->i_lock);
525 		if (!(inode->i_flags & I_DIRTY))
526 			gfs2_ordered_del_inode(ip);
527 		spin_unlock(&inode->i_lock);
528 	}
529 	return ret;
530 }
531 
532 /**
533  * gfs2_dirty_inode - check for atime updates
534  * @inode: The inode in question
535  * @flags: The type of dirty
536  *
537  * Unfortunately it can be called under any combination of inode
538  * glock and transaction lock, so we have to check carefully.
539  *
540  * At the moment this deals only with atime - it should be possible
541  * to expand that role in future, once a review of the locking has
542  * been carried out.
543  */
544 
545 static void gfs2_dirty_inode(struct inode *inode, int flags)
546 {
547 	struct gfs2_inode *ip = GFS2_I(inode);
548 	struct gfs2_sbd *sdp = GFS2_SB(inode);
549 	struct buffer_head *bh;
550 	struct gfs2_holder gh;
551 	int need_unlock = 0;
552 	int need_endtrans = 0;
553 	int ret;
554 
555 	if (!(flags & I_DIRTY_INODE))
556 		return;
557 	if (unlikely(gfs2_withdrawn(sdp)))
558 		return;
559 	if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
560 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
561 		if (ret) {
562 			fs_err(sdp, "dirty_inode: glock %d\n", ret);
563 			return;
564 		}
565 		need_unlock = 1;
566 	} else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
567 		return;
568 
569 	if (current->journal_info == NULL) {
570 		ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
571 		if (ret) {
572 			fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
573 			goto out;
574 		}
575 		need_endtrans = 1;
576 	}
577 
578 	ret = gfs2_meta_inode_buffer(ip, &bh);
579 	if (ret == 0) {
580 		gfs2_trans_add_meta(ip->i_gl, bh);
581 		gfs2_dinode_out(ip, bh->b_data);
582 		brelse(bh);
583 	}
584 
585 	if (need_endtrans)
586 		gfs2_trans_end(sdp);
587 out:
588 	if (need_unlock)
589 		gfs2_glock_dq_uninit(&gh);
590 }
591 
592 /**
593  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
594  * @sdp: the filesystem
595  *
596  * Returns: errno
597  */
598 
599 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
600 {
601 	struct gfs2_holder freeze_gh;
602 	int error;
603 
604 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
605 				   &freeze_gh);
606 	if (error && !gfs2_withdrawn(sdp))
607 		return error;
608 
609 	flush_workqueue(gfs2_delete_workqueue);
610 	if (sdp->sd_quotad_process)
611 		kthread_stop(sdp->sd_quotad_process);
612 	sdp->sd_quotad_process = NULL;
613 	if (sdp->sd_logd_process)
614 		kthread_stop(sdp->sd_logd_process);
615 	sdp->sd_logd_process = NULL;
616 
617 	gfs2_quota_sync(sdp->sd_vfs, 0);
618 	gfs2_statfs_sync(sdp->sd_vfs, 0);
619 
620 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
621 		       GFS2_LFC_MAKE_FS_RO);
622 	wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
623 	gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
624 
625 	if (gfs2_holder_initialized(&freeze_gh))
626 		gfs2_glock_dq_uninit(&freeze_gh);
627 
628 	gfs2_quota_cleanup(sdp);
629 
630 	return error;
631 }
632 
633 /**
634  * gfs2_put_super - Unmount the filesystem
635  * @sb: The VFS superblock
636  *
637  */
638 
639 static void gfs2_put_super(struct super_block *sb)
640 {
641 	struct gfs2_sbd *sdp = sb->s_fs_info;
642 	int error;
643 	struct gfs2_jdesc *jd;
644 
645 	/* No more recovery requests */
646 	set_bit(SDF_NORECOVERY, &sdp->sd_flags);
647 	smp_mb();
648 
649 	/* Wait on outstanding recovery */
650 restart:
651 	spin_lock(&sdp->sd_jindex_spin);
652 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
653 		if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
654 			continue;
655 		spin_unlock(&sdp->sd_jindex_spin);
656 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
657 			    TASK_UNINTERRUPTIBLE);
658 		goto restart;
659 	}
660 	spin_unlock(&sdp->sd_jindex_spin);
661 
662 	if (!sb_rdonly(sb)) {
663 		error = gfs2_make_fs_ro(sdp);
664 		if (error)
665 			gfs2_io_error(sdp);
666 	}
667 	/*  At this point, we're through modifying the disk  */
668 
669 	/*  Release stuff  */
670 
671 	iput(sdp->sd_jindex);
672 	iput(sdp->sd_statfs_inode);
673 	iput(sdp->sd_rindex);
674 	iput(sdp->sd_quota_inode);
675 
676 	gfs2_glock_put(sdp->sd_rename_gl);
677 	gfs2_glock_put(sdp->sd_freeze_gl);
678 
679 	if (!sdp->sd_args.ar_spectator) {
680 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
681 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
682 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
683 		gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
684 		iput(sdp->sd_sc_inode);
685 		iput(sdp->sd_qc_inode);
686 	}
687 
688 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
689 	gfs2_clear_rgrpd(sdp);
690 	gfs2_jindex_free(sdp);
691 	/*  Take apart glock structures and buffer lists  */
692 	gfs2_gl_hash_clear(sdp);
693 	gfs2_delete_debugfs_file(sdp);
694 	/*  Unmount the locking protocol  */
695 	gfs2_lm_unmount(sdp);
696 
697 	/*  At this point, we're through participating in the lockspace  */
698 	gfs2_sys_fs_del(sdp);
699 }
700 
701 /**
702  * gfs2_sync_fs - sync the filesystem
703  * @sb: the superblock
704  *
705  * Flushes the log to disk.
706  */
707 
708 static int gfs2_sync_fs(struct super_block *sb, int wait)
709 {
710 	struct gfs2_sbd *sdp = sb->s_fs_info;
711 
712 	gfs2_quota_sync(sb, -1);
713 	if (wait)
714 		gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
715 			       GFS2_LFC_SYNC_FS);
716 	return sdp->sd_log_error;
717 }
718 
719 void gfs2_freeze_func(struct work_struct *work)
720 {
721 	int error;
722 	struct gfs2_holder freeze_gh;
723 	struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
724 	struct super_block *sb = sdp->sd_vfs;
725 
726 	atomic_inc(&sb->s_active);
727 	error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
728 				   &freeze_gh);
729 	if (error) {
730 		fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error);
731 		gfs2_assert_withdraw(sdp, 0);
732 	} else {
733 		atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
734 		error = thaw_super(sb);
735 		if (error) {
736 			fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
737 				error);
738 			gfs2_assert_withdraw(sdp, 0);
739 		}
740 		if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
741 			freeze_gh.gh_flags |= GL_NOCACHE;
742 		gfs2_glock_dq_uninit(&freeze_gh);
743 	}
744 	deactivate_super(sb);
745 	clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
746 	wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
747 	return;
748 }
749 
750 /**
751  * gfs2_freeze - prevent further writes to the filesystem
752  * @sb: the VFS structure for the filesystem
753  *
754  */
755 
756 static int gfs2_freeze(struct super_block *sb)
757 {
758 	struct gfs2_sbd *sdp = sb->s_fs_info;
759 	int error = 0;
760 
761 	mutex_lock(&sdp->sd_freeze_mutex);
762 	if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
763 		goto out;
764 
765 	for (;;) {
766 		if (gfs2_withdrawn(sdp)) {
767 			error = -EINVAL;
768 			goto out;
769 		}
770 
771 		error = gfs2_lock_fs_check_clean(sdp);
772 		if (!error)
773 			break;
774 
775 		if (error == -EBUSY)
776 			fs_err(sdp, "waiting for recovery before freeze\n");
777 		else if (error == -EIO) {
778 			fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
779 			       "to recovery error.\n");
780 			goto out;
781 		} else {
782 			fs_err(sdp, "error freezing FS: %d\n", error);
783 		}
784 		fs_err(sdp, "retrying...\n");
785 		msleep(1000);
786 	}
787 	set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
788 out:
789 	mutex_unlock(&sdp->sd_freeze_mutex);
790 	return error;
791 }
792 
793 /**
794  * gfs2_unfreeze - reallow writes to the filesystem
795  * @sb: the VFS structure for the filesystem
796  *
797  */
798 
799 static int gfs2_unfreeze(struct super_block *sb)
800 {
801 	struct gfs2_sbd *sdp = sb->s_fs_info;
802 
803 	mutex_lock(&sdp->sd_freeze_mutex);
804         if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
805 	    !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
806 		mutex_unlock(&sdp->sd_freeze_mutex);
807                 return 0;
808 	}
809 
810 	gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
811 	mutex_unlock(&sdp->sd_freeze_mutex);
812 	return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
813 }
814 
815 /**
816  * statfs_fill - fill in the sg for a given RG
817  * @rgd: the RG
818  * @sc: the sc structure
819  *
820  * Returns: 0 on success, -ESTALE if the LVB is invalid
821  */
822 
823 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
824 			    struct gfs2_statfs_change_host *sc)
825 {
826 	gfs2_rgrp_verify(rgd);
827 	sc->sc_total += rgd->rd_data;
828 	sc->sc_free += rgd->rd_free;
829 	sc->sc_dinodes += rgd->rd_dinodes;
830 	return 0;
831 }
832 
833 /**
834  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
835  * @sdp: the filesystem
836  * @sc: the sc info that will be returned
837  *
838  * Any error (other than a signal) will cause this routine to fall back
839  * to the synchronous version.
840  *
841  * FIXME: This really shouldn't busy wait like this.
842  *
843  * Returns: errno
844  */
845 
846 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
847 {
848 	struct gfs2_rgrpd *rgd_next;
849 	struct gfs2_holder *gha, *gh;
850 	unsigned int slots = 64;
851 	unsigned int x;
852 	int done;
853 	int error = 0, err;
854 
855 	memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
856 	gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
857 	if (!gha)
858 		return -ENOMEM;
859 	for (x = 0; x < slots; x++)
860 		gfs2_holder_mark_uninitialized(gha + x);
861 
862 	rgd_next = gfs2_rgrpd_get_first(sdp);
863 
864 	for (;;) {
865 		done = 1;
866 
867 		for (x = 0; x < slots; x++) {
868 			gh = gha + x;
869 
870 			if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
871 				err = gfs2_glock_wait(gh);
872 				if (err) {
873 					gfs2_holder_uninit(gh);
874 					error = err;
875 				} else {
876 					if (!error) {
877 						struct gfs2_rgrpd *rgd =
878 							gfs2_glock2rgrp(gh->gh_gl);
879 
880 						error = statfs_slow_fill(rgd, sc);
881 					}
882 					gfs2_glock_dq_uninit(gh);
883 				}
884 			}
885 
886 			if (gfs2_holder_initialized(gh))
887 				done = 0;
888 			else if (rgd_next && !error) {
889 				error = gfs2_glock_nq_init(rgd_next->rd_gl,
890 							   LM_ST_SHARED,
891 							   GL_ASYNC,
892 							   gh);
893 				rgd_next = gfs2_rgrpd_get_next(rgd_next);
894 				done = 0;
895 			}
896 
897 			if (signal_pending(current))
898 				error = -ERESTARTSYS;
899 		}
900 
901 		if (done)
902 			break;
903 
904 		yield();
905 	}
906 
907 	kfree(gha);
908 	return error;
909 }
910 
911 /**
912  * gfs2_statfs_i - Do a statfs
913  * @sdp: the filesystem
914  * @sg: the sg structure
915  *
916  * Returns: errno
917  */
918 
919 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
920 {
921 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
922 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
923 
924 	spin_lock(&sdp->sd_statfs_spin);
925 
926 	*sc = *m_sc;
927 	sc->sc_total += l_sc->sc_total;
928 	sc->sc_free += l_sc->sc_free;
929 	sc->sc_dinodes += l_sc->sc_dinodes;
930 
931 	spin_unlock(&sdp->sd_statfs_spin);
932 
933 	if (sc->sc_free < 0)
934 		sc->sc_free = 0;
935 	if (sc->sc_free > sc->sc_total)
936 		sc->sc_free = sc->sc_total;
937 	if (sc->sc_dinodes < 0)
938 		sc->sc_dinodes = 0;
939 
940 	return 0;
941 }
942 
943 /**
944  * gfs2_statfs - Gather and return stats about the filesystem
945  * @sb: The superblock
946  * @statfsbuf: The buffer
947  *
948  * Returns: 0 on success or error code
949  */
950 
951 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
952 {
953 	struct super_block *sb = dentry->d_sb;
954 	struct gfs2_sbd *sdp = sb->s_fs_info;
955 	struct gfs2_statfs_change_host sc;
956 	int error;
957 
958 	error = gfs2_rindex_update(sdp);
959 	if (error)
960 		return error;
961 
962 	if (gfs2_tune_get(sdp, gt_statfs_slow))
963 		error = gfs2_statfs_slow(sdp, &sc);
964 	else
965 		error = gfs2_statfs_i(sdp, &sc);
966 
967 	if (error)
968 		return error;
969 
970 	buf->f_type = GFS2_MAGIC;
971 	buf->f_bsize = sdp->sd_sb.sb_bsize;
972 	buf->f_blocks = sc.sc_total;
973 	buf->f_bfree = sc.sc_free;
974 	buf->f_bavail = sc.sc_free;
975 	buf->f_files = sc.sc_dinodes + sc.sc_free;
976 	buf->f_ffree = sc.sc_free;
977 	buf->f_namelen = GFS2_FNAMESIZE;
978 
979 	return 0;
980 }
981 
982 /**
983  * gfs2_drop_inode - Drop an inode (test for remote unlink)
984  * @inode: The inode to drop
985  *
986  * If we've received a callback on an iopen lock then it's because a
987  * remote node tried to deallocate the inode but failed due to this node
988  * still having the inode open. Here we mark the link count zero
989  * since we know that it must have reached zero if the GLF_DEMOTE flag
990  * is set on the iopen glock. If we didn't do a disk read since the
991  * remote node removed the final link then we might otherwise miss
992  * this event. This check ensures that this node will deallocate the
993  * inode's blocks, or alternatively pass the baton on to another
994  * node for later deallocation.
995  */
996 
997 static int gfs2_drop_inode(struct inode *inode)
998 {
999 	struct gfs2_inode *ip = GFS2_I(inode);
1000 
1001 	if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1002 	    inode->i_nlink &&
1003 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
1004 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1005 		if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1006 			clear_nlink(inode);
1007 	}
1008 
1009 	/*
1010 	 * When under memory pressure when an inode's link count has dropped to
1011 	 * zero, defer deleting the inode to the delete workqueue.  This avoids
1012 	 * calling into DLM under memory pressure, which can deadlock.
1013 	 */
1014 	if (!inode->i_nlink &&
1015 	    unlikely(current->flags & PF_MEMALLOC) &&
1016 	    gfs2_holder_initialized(&ip->i_iopen_gh)) {
1017 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1018 
1019 		gfs2_glock_hold(gl);
1020 		if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1021 			gfs2_glock_queue_put(gl);
1022 		return false;
1023 	}
1024 
1025 	return generic_drop_inode(inode);
1026 }
1027 
1028 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1029 {
1030 	do {
1031 		if (d1 == d2)
1032 			return 1;
1033 		d1 = d1->d_parent;
1034 	} while (!IS_ROOT(d1));
1035 	return 0;
1036 }
1037 
1038 /**
1039  * gfs2_show_options - Show mount options for /proc/mounts
1040  * @s: seq_file structure
1041  * @root: root of this (sub)tree
1042  *
1043  * Returns: 0 on success or error code
1044  */
1045 
1046 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1047 {
1048 	struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1049 	struct gfs2_args *args = &sdp->sd_args;
1050 	int val;
1051 
1052 	if (is_ancestor(root, sdp->sd_master_dir))
1053 		seq_puts(s, ",meta");
1054 	if (args->ar_lockproto[0])
1055 		seq_show_option(s, "lockproto", args->ar_lockproto);
1056 	if (args->ar_locktable[0])
1057 		seq_show_option(s, "locktable", args->ar_locktable);
1058 	if (args->ar_hostdata[0])
1059 		seq_show_option(s, "hostdata", args->ar_hostdata);
1060 	if (args->ar_spectator)
1061 		seq_puts(s, ",spectator");
1062 	if (args->ar_localflocks)
1063 		seq_puts(s, ",localflocks");
1064 	if (args->ar_debug)
1065 		seq_puts(s, ",debug");
1066 	if (args->ar_posix_acl)
1067 		seq_puts(s, ",acl");
1068 	if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1069 		char *state;
1070 		switch (args->ar_quota) {
1071 		case GFS2_QUOTA_OFF:
1072 			state = "off";
1073 			break;
1074 		case GFS2_QUOTA_ACCOUNT:
1075 			state = "account";
1076 			break;
1077 		case GFS2_QUOTA_ON:
1078 			state = "on";
1079 			break;
1080 		default:
1081 			state = "unknown";
1082 			break;
1083 		}
1084 		seq_printf(s, ",quota=%s", state);
1085 	}
1086 	if (args->ar_suiddir)
1087 		seq_puts(s, ",suiddir");
1088 	if (args->ar_data != GFS2_DATA_DEFAULT) {
1089 		char *state;
1090 		switch (args->ar_data) {
1091 		case GFS2_DATA_WRITEBACK:
1092 			state = "writeback";
1093 			break;
1094 		case GFS2_DATA_ORDERED:
1095 			state = "ordered";
1096 			break;
1097 		default:
1098 			state = "unknown";
1099 			break;
1100 		}
1101 		seq_printf(s, ",data=%s", state);
1102 	}
1103 	if (args->ar_discard)
1104 		seq_puts(s, ",discard");
1105 	val = sdp->sd_tune.gt_logd_secs;
1106 	if (val != 30)
1107 		seq_printf(s, ",commit=%d", val);
1108 	val = sdp->sd_tune.gt_statfs_quantum;
1109 	if (val != 30)
1110 		seq_printf(s, ",statfs_quantum=%d", val);
1111 	else if (sdp->sd_tune.gt_statfs_slow)
1112 		seq_puts(s, ",statfs_quantum=0");
1113 	val = sdp->sd_tune.gt_quota_quantum;
1114 	if (val != 60)
1115 		seq_printf(s, ",quota_quantum=%d", val);
1116 	if (args->ar_statfs_percent)
1117 		seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1118 	if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1119 		const char *state;
1120 
1121 		switch (args->ar_errors) {
1122 		case GFS2_ERRORS_WITHDRAW:
1123 			state = "withdraw";
1124 			break;
1125 		case GFS2_ERRORS_PANIC:
1126 			state = "panic";
1127 			break;
1128 		default:
1129 			state = "unknown";
1130 			break;
1131 		}
1132 		seq_printf(s, ",errors=%s", state);
1133 	}
1134 	if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1135 		seq_puts(s, ",nobarrier");
1136 	if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1137 		seq_puts(s, ",demote_interface_used");
1138 	if (args->ar_rgrplvb)
1139 		seq_puts(s, ",rgrplvb");
1140 	if (args->ar_loccookie)
1141 		seq_puts(s, ",loccookie");
1142 	return 0;
1143 }
1144 
1145 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1146 {
1147 	struct inode *inode = &ip->i_inode;
1148 	struct gfs2_glock *gl = ip->i_gl;
1149 
1150 	truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1151 	truncate_inode_pages(&inode->i_data, 0);
1152 
1153 	if (atomic_read(&gl->gl_revokes) == 0) {
1154 		clear_bit(GLF_LFLUSH, &gl->gl_flags);
1155 		clear_bit(GLF_DIRTY, &gl->gl_flags);
1156 	}
1157 }
1158 
1159 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1160 {
1161 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1162 	struct gfs2_rgrpd *rgd;
1163 	struct gfs2_holder gh;
1164 	int error;
1165 
1166 	if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1167 		gfs2_consist_inode(ip);
1168 		return -EIO;
1169 	}
1170 
1171 	error = gfs2_rindex_update(sdp);
1172 	if (error)
1173 		return error;
1174 
1175 	error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1176 	if (error)
1177 		return error;
1178 
1179 	rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1180 	if (!rgd) {
1181 		gfs2_consist_inode(ip);
1182 		error = -EIO;
1183 		goto out_qs;
1184 	}
1185 
1186 	error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1187 	if (error)
1188 		goto out_qs;
1189 
1190 	error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1191 				 sdp->sd_jdesc->jd_blocks);
1192 	if (error)
1193 		goto out_rg_gunlock;
1194 
1195 	gfs2_free_di(rgd, ip);
1196 
1197 	gfs2_final_release_pages(ip);
1198 
1199 	gfs2_trans_end(sdp);
1200 
1201 out_rg_gunlock:
1202 	gfs2_glock_dq_uninit(&gh);
1203 out_qs:
1204 	gfs2_quota_unhold(ip);
1205 	return error;
1206 }
1207 
1208 /**
1209  * gfs2_glock_put_eventually
1210  * @gl:	The glock to put
1211  *
1212  * When under memory pressure, trigger a deferred glock put to make sure we
1213  * won't call into DLM and deadlock.  Otherwise, put the glock directly.
1214  */
1215 
1216 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1217 {
1218 	if (current->flags & PF_MEMALLOC)
1219 		gfs2_glock_queue_put(gl);
1220 	else
1221 		gfs2_glock_put(gl);
1222 }
1223 
1224 /**
1225  * gfs2_evict_inode - Remove an inode from cache
1226  * @inode: The inode to evict
1227  *
1228  * There are three cases to consider:
1229  * 1. i_nlink == 0, we are final opener (and must deallocate)
1230  * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1231  * 3. i_nlink > 0
1232  *
1233  * If the fs is read only, then we have to treat all cases as per #3
1234  * since we are unable to do any deallocation. The inode will be
1235  * deallocated by the next read/write node to attempt an allocation
1236  * in the same resource group
1237  *
1238  * We have to (at the moment) hold the inodes main lock to cover
1239  * the gap between unlocking the shared lock on the iopen lock and
1240  * taking the exclusive lock. I'd rather do a shared -> exclusive
1241  * conversion on the iopen lock, but we can change that later. This
1242  * is safe, just less efficient.
1243  */
1244 
1245 static void gfs2_evict_inode(struct inode *inode)
1246 {
1247 	struct super_block *sb = inode->i_sb;
1248 	struct gfs2_sbd *sdp = sb->s_fs_info;
1249 	struct gfs2_inode *ip = GFS2_I(inode);
1250 	struct gfs2_holder gh;
1251 	struct address_space *metamapping;
1252 	int error;
1253 
1254 	if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1255 		clear_inode(inode);
1256 		return;
1257 	}
1258 
1259 	if (inode->i_nlink || sb_rdonly(sb))
1260 		goto out;
1261 
1262 	if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1263 		BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1264 		gfs2_holder_mark_uninitialized(&gh);
1265 		goto alloc_failed;
1266 	}
1267 
1268 	/* Deletes should never happen under memory pressure anymore.  */
1269 	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1270 		goto out;
1271 
1272 	/* Must not read inode block until block type has been verified */
1273 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1274 	if (unlikely(error)) {
1275 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1276 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1277 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1278 		goto out;
1279 	}
1280 
1281 	error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1282 	if (error)
1283 		goto out_truncate;
1284 
1285 	if (test_bit(GIF_INVALID, &ip->i_flags)) {
1286 		error = gfs2_inode_refresh(ip);
1287 		if (error)
1288 			goto out_truncate;
1289 	}
1290 
1291 	/*
1292 	 * The inode may have been recreated in the meantime.
1293 	 */
1294 	if (inode->i_nlink)
1295 		goto out_truncate;
1296 
1297 alloc_failed:
1298 	if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1299 	    test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1300 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1301 		gfs2_glock_dq_wait(&ip->i_iopen_gh);
1302 		gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1303 				   &ip->i_iopen_gh);
1304 		error = gfs2_glock_nq(&ip->i_iopen_gh);
1305 		if (error)
1306 			goto out_truncate;
1307 	}
1308 
1309 	if (S_ISDIR(inode->i_mode) &&
1310 	    (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1311 		error = gfs2_dir_exhash_dealloc(ip);
1312 		if (error)
1313 			goto out_unlock;
1314 	}
1315 
1316 	if (ip->i_eattr) {
1317 		error = gfs2_ea_dealloc(ip);
1318 		if (error)
1319 			goto out_unlock;
1320 	}
1321 
1322 	if (!gfs2_is_stuffed(ip)) {
1323 		error = gfs2_file_dealloc(ip);
1324 		if (error)
1325 			goto out_unlock;
1326 	}
1327 
1328 	/* We're about to clear the bitmap for the dinode, but as soon as we
1329 	   do, gfs2_create_inode can create another inode at the same block
1330 	   location and try to set gl_object again. We clear gl_object here so
1331 	   that subsequent inode creates don't see an old gl_object. */
1332 	glock_clear_object(ip->i_gl, ip);
1333 	error = gfs2_dinode_dealloc(ip);
1334 	goto out_unlock;
1335 
1336 out_truncate:
1337 	gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1338 		       GFS2_LFC_EVICT_INODE);
1339 	metamapping = gfs2_glock2aspace(ip->i_gl);
1340 	if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1341 		filemap_fdatawrite(metamapping);
1342 		filemap_fdatawait(metamapping);
1343 	}
1344 	write_inode_now(inode, 1);
1345 	gfs2_ail_flush(ip->i_gl, 0);
1346 
1347 	error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1348 	if (error)
1349 		goto out_unlock;
1350 	/* Needs to be done before glock release & also in a transaction */
1351 	truncate_inode_pages(&inode->i_data, 0);
1352 	truncate_inode_pages(metamapping, 0);
1353 	gfs2_trans_end(sdp);
1354 
1355 out_unlock:
1356 	if (gfs2_rs_active(&ip->i_res))
1357 		gfs2_rs_deltree(&ip->i_res);
1358 
1359 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1360 		glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1361 		if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1362 			ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1363 			gfs2_glock_dq(&ip->i_iopen_gh);
1364 		}
1365 		gfs2_holder_uninit(&ip->i_iopen_gh);
1366 	}
1367 	if (gfs2_holder_initialized(&gh)) {
1368 		glock_clear_object(ip->i_gl, ip);
1369 		gfs2_glock_dq_uninit(&gh);
1370 	}
1371 	if (error && error != GLR_TRYFAILED && error != -EROFS)
1372 		fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1373 out:
1374 	truncate_inode_pages_final(&inode->i_data);
1375 	gfs2_rsqa_delete(ip, NULL);
1376 	gfs2_ordered_del_inode(ip);
1377 	clear_inode(inode);
1378 	gfs2_dir_hash_inval(ip);
1379 	glock_clear_object(ip->i_gl, ip);
1380 	wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1381 	gfs2_glock_add_to_lru(ip->i_gl);
1382 	gfs2_glock_put_eventually(ip->i_gl);
1383 	ip->i_gl = NULL;
1384 	if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1385 		struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1386 
1387 		glock_clear_object(gl, ip);
1388 		ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1389 		gfs2_glock_hold(gl);
1390 		gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1391 		gfs2_glock_put_eventually(gl);
1392 	}
1393 }
1394 
1395 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1396 {
1397 	struct gfs2_inode *ip;
1398 
1399 	ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1400 	if (!ip)
1401 		return NULL;
1402 	ip->i_flags = 0;
1403 	ip->i_gl = NULL;
1404 	memset(&ip->i_res, 0, sizeof(ip->i_res));
1405 	RB_CLEAR_NODE(&ip->i_res.rs_node);
1406 	ip->i_rahead = 0;
1407 	return &ip->i_inode;
1408 }
1409 
1410 static void gfs2_free_inode(struct inode *inode)
1411 {
1412 	kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1413 }
1414 
1415 const struct super_operations gfs2_super_ops = {
1416 	.alloc_inode		= gfs2_alloc_inode,
1417 	.free_inode		= gfs2_free_inode,
1418 	.write_inode		= gfs2_write_inode,
1419 	.dirty_inode		= gfs2_dirty_inode,
1420 	.evict_inode		= gfs2_evict_inode,
1421 	.put_super		= gfs2_put_super,
1422 	.sync_fs		= gfs2_sync_fs,
1423 	.freeze_super		= gfs2_freeze,
1424 	.thaw_super		= gfs2_unfreeze,
1425 	.statfs			= gfs2_statfs,
1426 	.drop_inode		= gfs2_drop_inode,
1427 	.show_options		= gfs2_show_options,
1428 };
1429 
1430