xref: /openbmc/linux/fs/gfs2/ops_fstype.c (revision 66a28915)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
5  */
6 
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/blkdev.h>
15 #include <linux/kthread.h>
16 #include <linux/export.h>
17 #include <linux/namei.h>
18 #include <linux/mount.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/quotaops.h>
21 #include <linux/lockdep.h>
22 #include <linux/module.h>
23 #include <linux/backing-dev.h>
24 #include <linux/fs_parser.h>
25 
26 #include "gfs2.h"
27 #include "incore.h"
28 #include "bmap.h"
29 #include "glock.h"
30 #include "glops.h"
31 #include "inode.h"
32 #include "recovery.h"
33 #include "rgrp.h"
34 #include "super.h"
35 #include "sys.h"
36 #include "util.h"
37 #include "log.h"
38 #include "quota.h"
39 #include "dir.h"
40 #include "meta_io.h"
41 #include "trace_gfs2.h"
42 #include "lops.h"
43 
44 #define DO 0
45 #define UNDO 1
46 
47 /**
48  * gfs2_tune_init - Fill a gfs2_tune structure with default values
49  * @gt: tune
50  *
51  */
52 
53 static void gfs2_tune_init(struct gfs2_tune *gt)
54 {
55 	spin_lock_init(&gt->gt_spin);
56 
57 	gt->gt_quota_warn_period = 10;
58 	gt->gt_quota_scale_num = 1;
59 	gt->gt_quota_scale_den = 1;
60 	gt->gt_new_files_jdata = 0;
61 	gt->gt_max_readahead = BIT(18);
62 	gt->gt_complain_secs = 10;
63 }
64 
65 void free_sbd(struct gfs2_sbd *sdp)
66 {
67 	if (sdp->sd_lkstats)
68 		free_percpu(sdp->sd_lkstats);
69 	kfree(sdp);
70 }
71 
72 static struct gfs2_sbd *init_sbd(struct super_block *sb)
73 {
74 	struct gfs2_sbd *sdp;
75 	struct address_space *mapping;
76 
77 	sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
78 	if (!sdp)
79 		return NULL;
80 
81 	sdp->sd_vfs = sb;
82 	sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
83 	if (!sdp->sd_lkstats)
84 		goto fail;
85 	sb->s_fs_info = sdp;
86 
87 	set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
88 	gfs2_tune_init(&sdp->sd_tune);
89 
90 	init_waitqueue_head(&sdp->sd_glock_wait);
91 	init_waitqueue_head(&sdp->sd_async_glock_wait);
92 	atomic_set(&sdp->sd_glock_disposal, 0);
93 	init_completion(&sdp->sd_locking_init);
94 	init_completion(&sdp->sd_wdack);
95 	spin_lock_init(&sdp->sd_statfs_spin);
96 
97 	spin_lock_init(&sdp->sd_rindex_spin);
98 	sdp->sd_rindex_tree.rb_node = NULL;
99 
100 	INIT_LIST_HEAD(&sdp->sd_jindex_list);
101 	spin_lock_init(&sdp->sd_jindex_spin);
102 	mutex_init(&sdp->sd_jindex_mutex);
103 	init_completion(&sdp->sd_journal_ready);
104 
105 	INIT_LIST_HEAD(&sdp->sd_quota_list);
106 	mutex_init(&sdp->sd_quota_mutex);
107 	mutex_init(&sdp->sd_quota_sync_mutex);
108 	init_waitqueue_head(&sdp->sd_quota_wait);
109 	INIT_LIST_HEAD(&sdp->sd_trunc_list);
110 	spin_lock_init(&sdp->sd_trunc_lock);
111 	spin_lock_init(&sdp->sd_bitmap_lock);
112 
113 	INIT_LIST_HEAD(&sdp->sd_sc_inodes_list);
114 
115 	mapping = &sdp->sd_aspace;
116 
117 	address_space_init_once(mapping);
118 	mapping->a_ops = &gfs2_rgrp_aops;
119 	mapping->host = sb->s_bdev->bd_inode;
120 	mapping->flags = 0;
121 	mapping_set_gfp_mask(mapping, GFP_NOFS);
122 	mapping->private_data = NULL;
123 	mapping->writeback_index = 0;
124 
125 	spin_lock_init(&sdp->sd_log_lock);
126 	atomic_set(&sdp->sd_log_pinned, 0);
127 	INIT_LIST_HEAD(&sdp->sd_log_revokes);
128 	INIT_LIST_HEAD(&sdp->sd_log_ordered);
129 	spin_lock_init(&sdp->sd_ordered_lock);
130 
131 	init_waitqueue_head(&sdp->sd_log_waitq);
132 	init_waitqueue_head(&sdp->sd_logd_waitq);
133 	spin_lock_init(&sdp->sd_ail_lock);
134 	INIT_LIST_HEAD(&sdp->sd_ail1_list);
135 	INIT_LIST_HEAD(&sdp->sd_ail2_list);
136 
137 	init_rwsem(&sdp->sd_log_flush_lock);
138 	atomic_set(&sdp->sd_log_in_flight, 0);
139 	atomic_set(&sdp->sd_reserving_log, 0);
140 	init_waitqueue_head(&sdp->sd_reserving_log_wait);
141 	init_waitqueue_head(&sdp->sd_log_flush_wait);
142 	atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
143 	mutex_init(&sdp->sd_freeze_mutex);
144 
145 	return sdp;
146 
147 fail:
148 	free_sbd(sdp);
149 	return NULL;
150 }
151 
152 /**
153  * gfs2_check_sb - Check superblock
154  * @sdp: the filesystem
155  * @sb: The superblock
156  * @silent: Don't print a message if the check fails
157  *
158  * Checks the version code of the FS is one that we understand how to
159  * read and that the sizes of the various on-disk structures have not
160  * changed.
161  */
162 
163 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
164 {
165 	struct gfs2_sb_host *sb = &sdp->sd_sb;
166 
167 	if (sb->sb_magic != GFS2_MAGIC ||
168 	    sb->sb_type != GFS2_METATYPE_SB) {
169 		if (!silent)
170 			pr_warn("not a GFS2 filesystem\n");
171 		return -EINVAL;
172 	}
173 
174 	if (sb->sb_fs_format != GFS2_FORMAT_FS ||
175 	    sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
176 		fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
177 		return -EINVAL;
178 	}
179 
180 	if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
181 	    (sb->sb_bsize & (sb->sb_bsize - 1))) {
182 		pr_warn("Invalid superblock size\n");
183 		return -EINVAL;
184 	}
185 
186 	return 0;
187 }
188 
189 static void end_bio_io_page(struct bio *bio)
190 {
191 	struct page *page = bio->bi_private;
192 
193 	if (!bio->bi_status)
194 		SetPageUptodate(page);
195 	else
196 		pr_warn("error %d reading superblock\n", bio->bi_status);
197 	unlock_page(page);
198 }
199 
200 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
201 {
202 	struct gfs2_sb_host *sb = &sdp->sd_sb;
203 	struct super_block *s = sdp->sd_vfs;
204 	const struct gfs2_sb *str = buf;
205 
206 	sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
207 	sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
208 	sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
209 	sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
210 	sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
211 	sb->sb_bsize = be32_to_cpu(str->sb_bsize);
212 	sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
213 	sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
214 	sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
215 	sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
216 	sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
217 
218 	memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
219 	memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
220 	memcpy(&s->s_uuid, str->sb_uuid, 16);
221 }
222 
223 /**
224  * gfs2_read_super - Read the gfs2 super block from disk
225  * @sdp: The GFS2 super block
226  * @sector: The location of the super block
227  * @error: The error code to return
228  *
229  * This uses the bio functions to read the super block from disk
230  * because we want to be 100% sure that we never read cached data.
231  * A super block is read twice only during each GFS2 mount and is
232  * never written to by the filesystem. The first time its read no
233  * locks are held, and the only details which are looked at are those
234  * relating to the locking protocol. Once locking is up and working,
235  * the sb is read again under the lock to establish the location of
236  * the master directory (contains pointers to journals etc) and the
237  * root directory.
238  *
239  * Returns: 0 on success or error
240  */
241 
242 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
243 {
244 	struct super_block *sb = sdp->sd_vfs;
245 	struct gfs2_sb *p;
246 	struct page *page;
247 	struct bio *bio;
248 
249 	page = alloc_page(GFP_NOFS);
250 	if (unlikely(!page))
251 		return -ENOMEM;
252 
253 	ClearPageUptodate(page);
254 	ClearPageDirty(page);
255 	lock_page(page);
256 
257 	bio = bio_alloc(GFP_NOFS, 1);
258 	bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
259 	bio_set_dev(bio, sb->s_bdev);
260 	bio_add_page(bio, page, PAGE_SIZE, 0);
261 
262 	bio->bi_end_io = end_bio_io_page;
263 	bio->bi_private = page;
264 	bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
265 	submit_bio(bio);
266 	wait_on_page_locked(page);
267 	bio_put(bio);
268 	if (!PageUptodate(page)) {
269 		__free_page(page);
270 		return -EIO;
271 	}
272 	p = kmap(page);
273 	gfs2_sb_in(sdp, p);
274 	kunmap(page);
275 	__free_page(page);
276 	return gfs2_check_sb(sdp, silent);
277 }
278 
279 /**
280  * gfs2_read_sb - Read super block
281  * @sdp: The GFS2 superblock
282  * @silent: Don't print message if mount fails
283  *
284  */
285 
286 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
287 {
288 	u32 hash_blocks, ind_blocks, leaf_blocks;
289 	u32 tmp_blocks;
290 	unsigned int x;
291 	int error;
292 
293 	error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
294 	if (error) {
295 		if (!silent)
296 			fs_err(sdp, "can't read superblock\n");
297 		return error;
298 	}
299 
300 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
301 			       GFS2_BASIC_BLOCK_SHIFT;
302 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
303 	sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
304 			  sizeof(struct gfs2_dinode)) / sizeof(u64);
305 	sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
306 			  sizeof(struct gfs2_meta_header)) / sizeof(u64);
307 	sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
308 			  sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
309 	sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
310 	sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
311 	sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
312 	sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
313 	sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
314 				sizeof(struct gfs2_meta_header)) /
315 			        sizeof(struct gfs2_quota_change);
316 	sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
317 				     sizeof(struct gfs2_meta_header))
318 		* GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
319 
320 	/* Compute maximum reservation required to add a entry to a directory */
321 
322 	hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
323 			     sdp->sd_jbsize);
324 
325 	ind_blocks = 0;
326 	for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
327 		tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
328 		ind_blocks += tmp_blocks;
329 	}
330 
331 	leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
332 
333 	sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
334 
335 	sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
336 				sizeof(struct gfs2_dinode);
337 	sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
338 	for (x = 2;; x++) {
339 		u64 space, d;
340 		u32 m;
341 
342 		space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
343 		d = space;
344 		m = do_div(d, sdp->sd_inptrs);
345 
346 		if (d != sdp->sd_heightsize[x - 1] || m)
347 			break;
348 		sdp->sd_heightsize[x] = space;
349 	}
350 	sdp->sd_max_height = x;
351 	sdp->sd_heightsize[x] = ~0;
352 	gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
353 
354 	sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
355 				      sizeof(struct gfs2_leaf)) /
356 				     GFS2_MIN_DIRENT_SIZE;
357 	return 0;
358 }
359 
360 static int init_names(struct gfs2_sbd *sdp, int silent)
361 {
362 	char *proto, *table;
363 	int error = 0;
364 
365 	proto = sdp->sd_args.ar_lockproto;
366 	table = sdp->sd_args.ar_locktable;
367 
368 	/*  Try to autodetect  */
369 
370 	if (!proto[0] || !table[0]) {
371 		error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
372 		if (error)
373 			return error;
374 
375 		if (!proto[0])
376 			proto = sdp->sd_sb.sb_lockproto;
377 		if (!table[0])
378 			table = sdp->sd_sb.sb_locktable;
379 	}
380 
381 	if (!table[0])
382 		table = sdp->sd_vfs->s_id;
383 
384 	strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
385 	strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
386 
387 	table = sdp->sd_table_name;
388 	while ((table = strchr(table, '/')))
389 		*table = '_';
390 
391 	return error;
392 }
393 
394 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
395 			int undo)
396 {
397 	int error = 0;
398 
399 	if (undo)
400 		goto fail_trans;
401 
402 	error = gfs2_glock_nq_num(sdp,
403 				  GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
404 				  LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
405 				  mount_gh);
406 	if (error) {
407 		fs_err(sdp, "can't acquire mount glock: %d\n", error);
408 		goto fail;
409 	}
410 
411 	error = gfs2_glock_nq_num(sdp,
412 				  GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
413 				  LM_ST_SHARED,
414 				  LM_FLAG_NOEXP | GL_EXACT,
415 				  &sdp->sd_live_gh);
416 	if (error) {
417 		fs_err(sdp, "can't acquire live glock: %d\n", error);
418 		goto fail_mount;
419 	}
420 
421 	error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
422 			       CREATE, &sdp->sd_rename_gl);
423 	if (error) {
424 		fs_err(sdp, "can't create rename glock: %d\n", error);
425 		goto fail_live;
426 	}
427 
428 	error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
429 			       CREATE, &sdp->sd_freeze_gl);
430 	if (error) {
431 		fs_err(sdp, "can't create transaction glock: %d\n", error);
432 		goto fail_rename;
433 	}
434 
435 	return 0;
436 
437 fail_trans:
438 	gfs2_glock_put(sdp->sd_freeze_gl);
439 fail_rename:
440 	gfs2_glock_put(sdp->sd_rename_gl);
441 fail_live:
442 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
443 fail_mount:
444 	gfs2_glock_dq_uninit(mount_gh);
445 fail:
446 	return error;
447 }
448 
449 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
450 			    u64 no_addr, const char *name)
451 {
452 	struct gfs2_sbd *sdp = sb->s_fs_info;
453 	struct dentry *dentry;
454 	struct inode *inode;
455 
456 	inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
457 				  GFS2_BLKST_FREE /* ignore */);
458 	if (IS_ERR(inode)) {
459 		fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
460 		return PTR_ERR(inode);
461 	}
462 	dentry = d_make_root(inode);
463 	if (!dentry) {
464 		fs_err(sdp, "can't alloc %s dentry\n", name);
465 		return -ENOMEM;
466 	}
467 	*dptr = dentry;
468 	return 0;
469 }
470 
471 static int init_sb(struct gfs2_sbd *sdp, int silent)
472 {
473 	struct super_block *sb = sdp->sd_vfs;
474 	struct gfs2_holder sb_gh;
475 	u64 no_addr;
476 	int ret;
477 
478 	ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
479 				LM_ST_SHARED, 0, &sb_gh);
480 	if (ret) {
481 		fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
482 		return ret;
483 	}
484 
485 	ret = gfs2_read_sb(sdp, silent);
486 	if (ret) {
487 		fs_err(sdp, "can't read superblock: %d\n", ret);
488 		goto out;
489 	}
490 
491 	/* Set up the buffer cache and SB for real */
492 	if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
493 		ret = -EINVAL;
494 		fs_err(sdp, "FS block size (%u) is too small for device "
495 		       "block size (%u)\n",
496 		       sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
497 		goto out;
498 	}
499 	if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
500 		ret = -EINVAL;
501 		fs_err(sdp, "FS block size (%u) is too big for machine "
502 		       "page size (%u)\n",
503 		       sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
504 		goto out;
505 	}
506 	sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
507 
508 	/* Get the root inode */
509 	no_addr = sdp->sd_sb.sb_root_dir.no_addr;
510 	ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
511 	if (ret)
512 		goto out;
513 
514 	/* Get the master inode */
515 	no_addr = sdp->sd_sb.sb_master_dir.no_addr;
516 	ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
517 	if (ret) {
518 		dput(sdp->sd_root_dir);
519 		goto out;
520 	}
521 	sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
522 out:
523 	gfs2_glock_dq_uninit(&sb_gh);
524 	return ret;
525 }
526 
527 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
528 {
529 	char *message = "FIRSTMOUNT=Done";
530 	char *envp[] = { message, NULL };
531 
532 	fs_info(sdp, "first mount done, others may mount\n");
533 
534 	if (sdp->sd_lockstruct.ls_ops->lm_first_done)
535 		sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
536 
537 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
538 }
539 
540 /**
541  * gfs2_jindex_hold - Grab a lock on the jindex
542  * @sdp: The GFS2 superblock
543  * @ji_gh: the holder for the jindex glock
544  *
545  * Returns: errno
546  */
547 
548 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
549 {
550 	struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
551 	struct qstr name;
552 	char buf[20];
553 	struct gfs2_jdesc *jd;
554 	int error;
555 
556 	name.name = buf;
557 
558 	mutex_lock(&sdp->sd_jindex_mutex);
559 
560 	for (;;) {
561 		struct gfs2_inode *jip;
562 
563 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
564 		if (error)
565 			break;
566 
567 		name.len = sprintf(buf, "journal%u", sdp->sd_journals);
568 		name.hash = gfs2_disk_hash(name.name, name.len);
569 
570 		error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
571 		if (error == -ENOENT) {
572 			error = 0;
573 			break;
574 		}
575 
576 		gfs2_glock_dq_uninit(ji_gh);
577 
578 		if (error)
579 			break;
580 
581 		error = -ENOMEM;
582 		jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
583 		if (!jd)
584 			break;
585 
586 		INIT_LIST_HEAD(&jd->extent_list);
587 		INIT_LIST_HEAD(&jd->jd_revoke_list);
588 
589 		INIT_WORK(&jd->jd_work, gfs2_recover_func);
590 		jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
591 		if (IS_ERR_OR_NULL(jd->jd_inode)) {
592 			if (!jd->jd_inode)
593 				error = -ENOENT;
594 			else
595 				error = PTR_ERR(jd->jd_inode);
596 			kfree(jd);
597 			break;
598 		}
599 
600 		spin_lock(&sdp->sd_jindex_spin);
601 		jd->jd_jid = sdp->sd_journals++;
602 		jip = GFS2_I(jd->jd_inode);
603 		jd->jd_no_addr = jip->i_no_addr;
604 		list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
605 		spin_unlock(&sdp->sd_jindex_spin);
606 	}
607 
608 	mutex_unlock(&sdp->sd_jindex_mutex);
609 
610 	return error;
611 }
612 
613 /**
614  * init_statfs - look up and initialize master and local (per node) statfs inodes
615  * @sdp: The GFS2 superblock
616  *
617  * This should be called after the jindex is initialized in init_journal() and
618  * before gfs2_journal_recovery() is called because we need to be able to write
619  * to these inodes during recovery.
620  *
621  * Returns: errno
622  */
623 static int init_statfs(struct gfs2_sbd *sdp)
624 {
625 	int error = 0;
626 	struct inode *master = d_inode(sdp->sd_master_dir);
627 	struct inode *pn = NULL;
628 	char buf[30];
629 	struct gfs2_jdesc *jd;
630 	struct gfs2_inode *ip;
631 
632 	sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
633 	if (IS_ERR(sdp->sd_statfs_inode)) {
634 		error = PTR_ERR(sdp->sd_statfs_inode);
635 		fs_err(sdp, "can't read in statfs inode: %d\n", error);
636 		goto out;
637 	}
638 	if (sdp->sd_args.ar_spectator)
639 		goto out;
640 
641 	pn = gfs2_lookup_simple(master, "per_node");
642 	if (IS_ERR(pn)) {
643 		error = PTR_ERR(pn);
644 		fs_err(sdp, "can't find per_node directory: %d\n", error);
645 		goto put_statfs;
646 	}
647 
648 	/* For each jid, lookup the corresponding local statfs inode in the
649 	 * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
650 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
651 		struct local_statfs_inode *lsi =
652 			kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
653 		if (!lsi) {
654 			error = -ENOMEM;
655 			goto free_local;
656 		}
657 		sprintf(buf, "statfs_change%u", jd->jd_jid);
658 		lsi->si_sc_inode = gfs2_lookup_simple(pn, buf);
659 		if (IS_ERR(lsi->si_sc_inode)) {
660 			error = PTR_ERR(lsi->si_sc_inode);
661 			fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
662 			       jd->jd_jid, error);
663 			goto free_local;
664 		}
665 		lsi->si_jid = jd->jd_jid;
666 		if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
667 			sdp->sd_sc_inode = lsi->si_sc_inode;
668 
669 		list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
670 	}
671 
672 	iput(pn);
673 	ip = GFS2_I(sdp->sd_sc_inode);
674 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
675 				   &sdp->sd_sc_gh);
676 	if (error) {
677 		fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
678 		goto free_local;
679 	}
680 	return 0;
681 
682 free_local:
683 	free_local_statfs_inodes(sdp);
684 	iput(pn);
685 put_statfs:
686 	iput(sdp->sd_statfs_inode);
687 out:
688 	return error;
689 }
690 
691 /* Uninitialize and free up memory used by the list of statfs inodes */
692 static void uninit_statfs(struct gfs2_sbd *sdp)
693 {
694 	if (!sdp->sd_args.ar_spectator) {
695 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
696 		free_local_statfs_inodes(sdp);
697 	}
698 	iput(sdp->sd_statfs_inode);
699 }
700 
701 static int init_journal(struct gfs2_sbd *sdp, int undo)
702 {
703 	struct inode *master = d_inode(sdp->sd_master_dir);
704 	struct gfs2_holder ji_gh;
705 	struct gfs2_inode *ip;
706 	int jindex = 1;
707 	int error = 0;
708 
709 	if (undo) {
710 		jindex = 0;
711 		goto fail_statfs;
712 	}
713 
714 	sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
715 	if (IS_ERR(sdp->sd_jindex)) {
716 		fs_err(sdp, "can't lookup journal index: %d\n", error);
717 		return PTR_ERR(sdp->sd_jindex);
718 	}
719 
720 	/* Load in the journal index special file */
721 
722 	error = gfs2_jindex_hold(sdp, &ji_gh);
723 	if (error) {
724 		fs_err(sdp, "can't read journal index: %d\n", error);
725 		goto fail;
726 	}
727 
728 	error = -EUSERS;
729 	if (!gfs2_jindex_size(sdp)) {
730 		fs_err(sdp, "no journals!\n");
731 		goto fail_jindex;
732 	}
733 
734 	atomic_set(&sdp->sd_log_blks_needed, 0);
735 	if (sdp->sd_args.ar_spectator) {
736 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
737 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
738 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
739 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
740 	} else {
741 		if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
742 			fs_err(sdp, "can't mount journal #%u\n",
743 			       sdp->sd_lockstruct.ls_jid);
744 			fs_err(sdp, "there are only %u journals (0 - %u)\n",
745 			       gfs2_jindex_size(sdp),
746 			       gfs2_jindex_size(sdp) - 1);
747 			goto fail_jindex;
748 		}
749 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
750 
751 		error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
752 					  &gfs2_journal_glops,
753 					  LM_ST_EXCLUSIVE,
754 					  LM_FLAG_NOEXP | GL_NOCACHE,
755 					  &sdp->sd_journal_gh);
756 		if (error) {
757 			fs_err(sdp, "can't acquire journal glock: %d\n", error);
758 			goto fail_jindex;
759 		}
760 
761 		ip = GFS2_I(sdp->sd_jdesc->jd_inode);
762 		sdp->sd_jinode_gl = ip->i_gl;
763 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
764 					   LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
765 					   &sdp->sd_jinode_gh);
766 		if (error) {
767 			fs_err(sdp, "can't acquire journal inode glock: %d\n",
768 			       error);
769 			goto fail_journal_gh;
770 		}
771 
772 		error = gfs2_jdesc_check(sdp->sd_jdesc);
773 		if (error) {
774 			fs_err(sdp, "my journal (%u) is bad: %d\n",
775 			       sdp->sd_jdesc->jd_jid, error);
776 			goto fail_jinode_gh;
777 		}
778 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
779 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
780 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
781 
782 		/* Map the extents for this journal's blocks */
783 		gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
784 	}
785 	trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
786 
787 	/* Lookup statfs inodes here so journal recovery can use them. */
788 	error = init_statfs(sdp);
789 	if (error)
790 		goto fail_jinode_gh;
791 
792 	if (sdp->sd_lockstruct.ls_first) {
793 		unsigned int x;
794 		for (x = 0; x < sdp->sd_journals; x++) {
795 			struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
796 
797 			if (sdp->sd_args.ar_spectator) {
798 				error = check_journal_clean(sdp, jd, true);
799 				if (error)
800 					goto fail_statfs;
801 				continue;
802 			}
803 			error = gfs2_recover_journal(jd, true);
804 			if (error) {
805 				fs_err(sdp, "error recovering journal %u: %d\n",
806 				       x, error);
807 				goto fail_statfs;
808 			}
809 		}
810 
811 		gfs2_others_may_mount(sdp);
812 	} else if (!sdp->sd_args.ar_spectator) {
813 		error = gfs2_recover_journal(sdp->sd_jdesc, true);
814 		if (error) {
815 			fs_err(sdp, "error recovering my journal: %d\n", error);
816 			goto fail_statfs;
817 		}
818 	}
819 
820 	sdp->sd_log_idle = 1;
821 	set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
822 	gfs2_glock_dq_uninit(&ji_gh);
823 	jindex = 0;
824 	INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
825 	return 0;
826 
827 fail_statfs:
828 	uninit_statfs(sdp);
829 fail_jinode_gh:
830 	/* A withdraw may have done dq/uninit so now we need to check it */
831 	if (!sdp->sd_args.ar_spectator &&
832 	    gfs2_holder_initialized(&sdp->sd_jinode_gh))
833 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
834 fail_journal_gh:
835 	if (!sdp->sd_args.ar_spectator &&
836 	    gfs2_holder_initialized(&sdp->sd_journal_gh))
837 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
838 fail_jindex:
839 	gfs2_jindex_free(sdp);
840 	if (jindex)
841 		gfs2_glock_dq_uninit(&ji_gh);
842 fail:
843 	iput(sdp->sd_jindex);
844 	return error;
845 }
846 
847 static struct lock_class_key gfs2_quota_imutex_key;
848 
849 static int init_inodes(struct gfs2_sbd *sdp, int undo)
850 {
851 	int error = 0;
852 	struct inode *master = d_inode(sdp->sd_master_dir);
853 
854 	if (undo)
855 		goto fail_qinode;
856 
857 	error = init_journal(sdp, undo);
858 	complete_all(&sdp->sd_journal_ready);
859 	if (error)
860 		goto fail;
861 
862 	/* Read in the resource index inode */
863 	sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
864 	if (IS_ERR(sdp->sd_rindex)) {
865 		error = PTR_ERR(sdp->sd_rindex);
866 		fs_err(sdp, "can't get resource index inode: %d\n", error);
867 		goto fail_journal;
868 	}
869 	sdp->sd_rindex_uptodate = 0;
870 
871 	/* Read in the quota inode */
872 	sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
873 	if (IS_ERR(sdp->sd_quota_inode)) {
874 		error = PTR_ERR(sdp->sd_quota_inode);
875 		fs_err(sdp, "can't get quota file inode: %d\n", error);
876 		goto fail_rindex;
877 	}
878 	/*
879 	 * i_rwsem on quota files is special. Since this inode is hidden system
880 	 * file, we are safe to define locking ourselves.
881 	 */
882 	lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
883 			  &gfs2_quota_imutex_key);
884 
885 	error = gfs2_rindex_update(sdp);
886 	if (error)
887 		goto fail_qinode;
888 
889 	return 0;
890 
891 fail_qinode:
892 	iput(sdp->sd_quota_inode);
893 fail_rindex:
894 	gfs2_clear_rgrpd(sdp);
895 	iput(sdp->sd_rindex);
896 fail_journal:
897 	init_journal(sdp, UNDO);
898 fail:
899 	return error;
900 }
901 
902 static int init_per_node(struct gfs2_sbd *sdp, int undo)
903 {
904 	struct inode *pn = NULL;
905 	char buf[30];
906 	int error = 0;
907 	struct gfs2_inode *ip;
908 	struct inode *master = d_inode(sdp->sd_master_dir);
909 
910 	if (sdp->sd_args.ar_spectator)
911 		return 0;
912 
913 	if (undo)
914 		goto fail_qc_gh;
915 
916 	pn = gfs2_lookup_simple(master, "per_node");
917 	if (IS_ERR(pn)) {
918 		error = PTR_ERR(pn);
919 		fs_err(sdp, "can't find per_node directory: %d\n", error);
920 		return error;
921 	}
922 
923 	sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
924 	sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
925 	if (IS_ERR(sdp->sd_qc_inode)) {
926 		error = PTR_ERR(sdp->sd_qc_inode);
927 		fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
928 		goto fail_ut_i;
929 	}
930 
931 	iput(pn);
932 	pn = NULL;
933 
934 	ip = GFS2_I(sdp->sd_qc_inode);
935 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
936 				   &sdp->sd_qc_gh);
937 	if (error) {
938 		fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
939 		goto fail_qc_i;
940 	}
941 
942 	return 0;
943 
944 fail_qc_gh:
945 	gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
946 fail_qc_i:
947 	iput(sdp->sd_qc_inode);
948 fail_ut_i:
949 	iput(pn);
950 	return error;
951 }
952 
953 static const match_table_t nolock_tokens = {
954 	{ Opt_jid, "jid=%d", },
955 	{ Opt_err, NULL },
956 };
957 
958 static const struct lm_lockops nolock_ops = {
959 	.lm_proto_name = "lock_nolock",
960 	.lm_put_lock = gfs2_glock_free,
961 	.lm_tokens = &nolock_tokens,
962 };
963 
964 /**
965  * gfs2_lm_mount - mount a locking protocol
966  * @sdp: the filesystem
967  * @args: mount arguments
968  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
969  *
970  * Returns: errno
971  */
972 
973 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
974 {
975 	const struct lm_lockops *lm;
976 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
977 	struct gfs2_args *args = &sdp->sd_args;
978 	const char *proto = sdp->sd_proto_name;
979 	const char *table = sdp->sd_table_name;
980 	char *o, *options;
981 	int ret;
982 
983 	if (!strcmp("lock_nolock", proto)) {
984 		lm = &nolock_ops;
985 		sdp->sd_args.ar_localflocks = 1;
986 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
987 	} else if (!strcmp("lock_dlm", proto)) {
988 		lm = &gfs2_dlm_ops;
989 #endif
990 	} else {
991 		pr_info("can't find protocol %s\n", proto);
992 		return -ENOENT;
993 	}
994 
995 	fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
996 
997 	ls->ls_ops = lm;
998 	ls->ls_first = 1;
999 
1000 	for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1001 		substring_t tmp[MAX_OPT_ARGS];
1002 		int token, option;
1003 
1004 		if (!o || !*o)
1005 			continue;
1006 
1007 		token = match_token(o, *lm->lm_tokens, tmp);
1008 		switch (token) {
1009 		case Opt_jid:
1010 			ret = match_int(&tmp[0], &option);
1011 			if (ret || option < 0)
1012 				goto hostdata_error;
1013 			if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1014 				ls->ls_jid = option;
1015 			break;
1016 		case Opt_id:
1017 		case Opt_nodir:
1018 			/* Obsolete, but left for backward compat purposes */
1019 			break;
1020 		case Opt_first:
1021 			ret = match_int(&tmp[0], &option);
1022 			if (ret || (option != 0 && option != 1))
1023 				goto hostdata_error;
1024 			ls->ls_first = option;
1025 			break;
1026 		case Opt_err:
1027 		default:
1028 hostdata_error:
1029 			fs_info(sdp, "unknown hostdata (%s)\n", o);
1030 			return -EINVAL;
1031 		}
1032 	}
1033 
1034 	if (lm->lm_mount == NULL) {
1035 		fs_info(sdp, "Now mounting FS...\n");
1036 		complete_all(&sdp->sd_locking_init);
1037 		return 0;
1038 	}
1039 	ret = lm->lm_mount(sdp, table);
1040 	if (ret == 0)
1041 		fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1042 	complete_all(&sdp->sd_locking_init);
1043 	return ret;
1044 }
1045 
1046 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1047 {
1048 	const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1049 	if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1050 		lm->lm_unmount(sdp);
1051 }
1052 
1053 static int wait_on_journal(struct gfs2_sbd *sdp)
1054 {
1055 	if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1056 		return 0;
1057 
1058 	return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1059 		? -EINTR : 0;
1060 }
1061 
1062 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1063 {
1064 	struct super_block *sb = sdp->sd_vfs;
1065 	char ro[20];
1066 	char spectator[20];
1067 	char *envp[] = { ro, spectator, NULL };
1068 	sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1069 	sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1070 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1071 }
1072 
1073 /**
1074  * gfs2_fill_super - Read in superblock
1075  * @sb: The VFS superblock
1076  * @args: Mount options
1077  * @silent: Don't complain if it's not a GFS2 filesystem
1078  *
1079  * Returns: -errno
1080  */
1081 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1082 {
1083 	struct gfs2_args *args = fc->fs_private;
1084 	int silent = fc->sb_flags & SB_SILENT;
1085 	struct gfs2_sbd *sdp;
1086 	struct gfs2_holder mount_gh;
1087 	int error;
1088 
1089 	sdp = init_sbd(sb);
1090 	if (!sdp) {
1091 		pr_warn("can't alloc struct gfs2_sbd\n");
1092 		return -ENOMEM;
1093 	}
1094 	sdp->sd_args = *args;
1095 
1096 	if (sdp->sd_args.ar_spectator) {
1097                 sb->s_flags |= SB_RDONLY;
1098 		set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1099 	}
1100 	if (sdp->sd_args.ar_posix_acl)
1101 		sb->s_flags |= SB_POSIXACL;
1102 	if (sdp->sd_args.ar_nobarrier)
1103 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1104 
1105 	sb->s_flags |= SB_NOSEC;
1106 	sb->s_magic = GFS2_MAGIC;
1107 	sb->s_op = &gfs2_super_ops;
1108 	sb->s_d_op = &gfs2_dops;
1109 	sb->s_export_op = &gfs2_export_ops;
1110 	sb->s_xattr = gfs2_xattr_handlers;
1111 	sb->s_qcop = &gfs2_quotactl_ops;
1112 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1113 	sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1114 	sb->s_time_gran = 1;
1115 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1116 
1117 	/* Set up the buffer cache and fill in some fake block size values
1118 	   to allow us to read-in the on-disk superblock. */
1119 	sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1120 	sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1121 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1122                                GFS2_BASIC_BLOCK_SHIFT;
1123 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1124 
1125 	sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1126 	sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1127 	if (sdp->sd_args.ar_statfs_quantum) {
1128 		sdp->sd_tune.gt_statfs_slow = 0;
1129 		sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1130 	} else {
1131 		sdp->sd_tune.gt_statfs_slow = 1;
1132 		sdp->sd_tune.gt_statfs_quantum = 30;
1133 	}
1134 
1135 	error = init_names(sdp, silent);
1136 	if (error)
1137 		goto fail_free;
1138 
1139 	snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1140 
1141 	error = gfs2_sys_fs_add(sdp);
1142 	if (error)
1143 		goto fail_free;
1144 
1145 	gfs2_create_debugfs_file(sdp);
1146 
1147 	error = gfs2_lm_mount(sdp, silent);
1148 	if (error)
1149 		goto fail_debug;
1150 
1151 	error = init_locking(sdp, &mount_gh, DO);
1152 	if (error)
1153 		goto fail_lm;
1154 
1155 	error = init_sb(sdp, silent);
1156 	if (error)
1157 		goto fail_locking;
1158 
1159 	error = wait_on_journal(sdp);
1160 	if (error)
1161 		goto fail_sb;
1162 
1163 	/*
1164 	 * If user space has failed to join the cluster or some similar
1165 	 * failure has occurred, then the journal id will contain a
1166 	 * negative (error) number. This will then be returned to the
1167 	 * caller (of the mount syscall). We do this even for spectator
1168 	 * mounts (which just write a jid of 0 to indicate "ok" even though
1169 	 * the jid is unused in the spectator case)
1170 	 */
1171 	if (sdp->sd_lockstruct.ls_jid < 0) {
1172 		error = sdp->sd_lockstruct.ls_jid;
1173 		sdp->sd_lockstruct.ls_jid = 0;
1174 		goto fail_sb;
1175 	}
1176 
1177 	if (sdp->sd_args.ar_spectator)
1178 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1179 			 sdp->sd_table_name);
1180 	else
1181 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1182 			 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1183 
1184 	error = init_inodes(sdp, DO);
1185 	if (error)
1186 		goto fail_sb;
1187 
1188 	error = init_per_node(sdp, DO);
1189 	if (error)
1190 		goto fail_inodes;
1191 
1192 	error = gfs2_statfs_init(sdp);
1193 	if (error) {
1194 		fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1195 		goto fail_per_node;
1196 	}
1197 
1198 	if (sb_rdonly(sb)) {
1199 		struct gfs2_holder freeze_gh;
1200 
1201 		error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
1202 					   LM_FLAG_NOEXP | GL_EXACT,
1203 					   &freeze_gh);
1204 		if (error) {
1205 			fs_err(sdp, "can't make FS RO: %d\n", error);
1206 			goto fail_per_node;
1207 		}
1208 		gfs2_glock_dq_uninit(&freeze_gh);
1209 	} else {
1210 		error = gfs2_make_fs_rw(sdp);
1211 		if (error) {
1212 			fs_err(sdp, "can't make FS RW: %d\n", error);
1213 			goto fail_per_node;
1214 		}
1215 	}
1216 
1217 	gfs2_glock_dq_uninit(&mount_gh);
1218 	gfs2_online_uevent(sdp);
1219 	return 0;
1220 
1221 fail_per_node:
1222 	init_per_node(sdp, UNDO);
1223 fail_inodes:
1224 	init_inodes(sdp, UNDO);
1225 fail_sb:
1226 	if (sdp->sd_root_dir)
1227 		dput(sdp->sd_root_dir);
1228 	if (sdp->sd_master_dir)
1229 		dput(sdp->sd_master_dir);
1230 	if (sb->s_root)
1231 		dput(sb->s_root);
1232 	sb->s_root = NULL;
1233 fail_locking:
1234 	init_locking(sdp, &mount_gh, UNDO);
1235 fail_lm:
1236 	complete_all(&sdp->sd_journal_ready);
1237 	gfs2_gl_hash_clear(sdp);
1238 	gfs2_lm_unmount(sdp);
1239 fail_debug:
1240 	gfs2_delete_debugfs_file(sdp);
1241 	gfs2_sys_fs_del(sdp);
1242 fail_free:
1243 	free_sbd(sdp);
1244 	sb->s_fs_info = NULL;
1245 	return error;
1246 }
1247 
1248 /**
1249  * gfs2_get_tree - Get the GFS2 superblock and root directory
1250  * @fc: The filesystem context
1251  *
1252  * Returns: 0 or -errno on error
1253  */
1254 static int gfs2_get_tree(struct fs_context *fc)
1255 {
1256 	struct gfs2_args *args = fc->fs_private;
1257 	struct gfs2_sbd *sdp;
1258 	int error;
1259 
1260 	error = get_tree_bdev(fc, gfs2_fill_super);
1261 	if (error)
1262 		return error;
1263 
1264 	sdp = fc->root->d_sb->s_fs_info;
1265 	dput(fc->root);
1266 	if (args->ar_meta)
1267 		fc->root = dget(sdp->sd_master_dir);
1268 	else
1269 		fc->root = dget(sdp->sd_root_dir);
1270 	return 0;
1271 }
1272 
1273 static void gfs2_fc_free(struct fs_context *fc)
1274 {
1275 	struct gfs2_args *args = fc->fs_private;
1276 
1277 	kfree(args);
1278 }
1279 
1280 enum gfs2_param {
1281 	Opt_lockproto,
1282 	Opt_locktable,
1283 	Opt_hostdata,
1284 	Opt_spectator,
1285 	Opt_ignore_local_fs,
1286 	Opt_localflocks,
1287 	Opt_localcaching,
1288 	Opt_debug,
1289 	Opt_upgrade,
1290 	Opt_acl,
1291 	Opt_quota,
1292 	Opt_quota_flag,
1293 	Opt_suiddir,
1294 	Opt_data,
1295 	Opt_meta,
1296 	Opt_discard,
1297 	Opt_commit,
1298 	Opt_errors,
1299 	Opt_statfs_quantum,
1300 	Opt_statfs_percent,
1301 	Opt_quota_quantum,
1302 	Opt_barrier,
1303 	Opt_rgrplvb,
1304 	Opt_loccookie,
1305 };
1306 
1307 static const struct constant_table gfs2_param_quota[] = {
1308 	{"off",        GFS2_QUOTA_OFF},
1309 	{"account",    GFS2_QUOTA_ACCOUNT},
1310 	{"on",         GFS2_QUOTA_ON},
1311 	{}
1312 };
1313 
1314 enum opt_data {
1315 	Opt_data_writeback = GFS2_DATA_WRITEBACK,
1316 	Opt_data_ordered   = GFS2_DATA_ORDERED,
1317 };
1318 
1319 static const struct constant_table gfs2_param_data[] = {
1320 	{"writeback",  Opt_data_writeback },
1321 	{"ordered",    Opt_data_ordered },
1322 	{}
1323 };
1324 
1325 enum opt_errors {
1326 	Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1327 	Opt_errors_panic    = GFS2_ERRORS_PANIC,
1328 };
1329 
1330 static const struct constant_table gfs2_param_errors[] = {
1331 	{"withdraw",   Opt_errors_withdraw },
1332 	{"panic",      Opt_errors_panic },
1333 	{}
1334 };
1335 
1336 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1337 	fsparam_string ("lockproto",          Opt_lockproto),
1338 	fsparam_string ("locktable",          Opt_locktable),
1339 	fsparam_string ("hostdata",           Opt_hostdata),
1340 	fsparam_flag   ("spectator",          Opt_spectator),
1341 	fsparam_flag   ("norecovery",         Opt_spectator),
1342 	fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1343 	fsparam_flag   ("localflocks",        Opt_localflocks),
1344 	fsparam_flag   ("localcaching",       Opt_localcaching),
1345 	fsparam_flag_no("debug",              Opt_debug),
1346 	fsparam_flag   ("upgrade",            Opt_upgrade),
1347 	fsparam_flag_no("acl",                Opt_acl),
1348 	fsparam_flag_no("suiddir",            Opt_suiddir),
1349 	fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1350 	fsparam_flag   ("meta",               Opt_meta),
1351 	fsparam_flag_no("discard",            Opt_discard),
1352 	fsparam_s32    ("commit",             Opt_commit),
1353 	fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1354 	fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1355 	fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1356 	fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1357 	fsparam_flag_no("barrier",            Opt_barrier),
1358 	fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1359 	fsparam_flag_no("loccookie",          Opt_loccookie),
1360 	/* quota can be a flag or an enum so it gets special treatment */
1361 	fsparam_flag_no("quota",	      Opt_quota_flag),
1362 	fsparam_enum("quota",		      Opt_quota, gfs2_param_quota),
1363 	{}
1364 };
1365 
1366 /* Parse a single mount parameter */
1367 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1368 {
1369 	struct gfs2_args *args = fc->fs_private;
1370 	struct fs_parse_result result;
1371 	int o;
1372 
1373 	o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1374 	if (o < 0)
1375 		return o;
1376 
1377 	switch (o) {
1378 	case Opt_lockproto:
1379 		strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1380 		break;
1381 	case Opt_locktable:
1382 		strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1383 		break;
1384 	case Opt_hostdata:
1385 		strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1386 		break;
1387 	case Opt_spectator:
1388 		args->ar_spectator = 1;
1389 		break;
1390 	case Opt_ignore_local_fs:
1391 		/* Retained for backwards compat only */
1392 		break;
1393 	case Opt_localflocks:
1394 		args->ar_localflocks = 1;
1395 		break;
1396 	case Opt_localcaching:
1397 		/* Retained for backwards compat only */
1398 		break;
1399 	case Opt_debug:
1400 		if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1401 			return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1402 		args->ar_debug = result.boolean;
1403 		break;
1404 	case Opt_upgrade:
1405 		/* Retained for backwards compat only */
1406 		break;
1407 	case Opt_acl:
1408 		args->ar_posix_acl = result.boolean;
1409 		break;
1410 	case Opt_quota_flag:
1411 		args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1412 		break;
1413 	case Opt_quota:
1414 		args->ar_quota = result.int_32;
1415 		break;
1416 	case Opt_suiddir:
1417 		args->ar_suiddir = result.boolean;
1418 		break;
1419 	case Opt_data:
1420 		/* The uint_32 result maps directly to GFS2_DATA_* */
1421 		args->ar_data = result.uint_32;
1422 		break;
1423 	case Opt_meta:
1424 		args->ar_meta = 1;
1425 		break;
1426 	case Opt_discard:
1427 		args->ar_discard = result.boolean;
1428 		break;
1429 	case Opt_commit:
1430 		if (result.int_32 <= 0)
1431 			return invalfc(fc, "commit mount option requires a positive numeric argument");
1432 		args->ar_commit = result.int_32;
1433 		break;
1434 	case Opt_statfs_quantum:
1435 		if (result.int_32 < 0)
1436 			return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1437 		args->ar_statfs_quantum = result.int_32;
1438 		break;
1439 	case Opt_quota_quantum:
1440 		if (result.int_32 <= 0)
1441 			return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1442 		args->ar_quota_quantum = result.int_32;
1443 		break;
1444 	case Opt_statfs_percent:
1445 		if (result.int_32 < 0 || result.int_32 > 100)
1446 			return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1447 		args->ar_statfs_percent = result.int_32;
1448 		break;
1449 	case Opt_errors:
1450 		if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1451 			return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1452 		args->ar_errors = result.uint_32;
1453 		break;
1454 	case Opt_barrier:
1455 		args->ar_nobarrier = result.boolean;
1456 		break;
1457 	case Opt_rgrplvb:
1458 		args->ar_rgrplvb = result.boolean;
1459 		break;
1460 	case Opt_loccookie:
1461 		args->ar_loccookie = result.boolean;
1462 		break;
1463 	default:
1464 		return invalfc(fc, "invalid mount option: %s", param->key);
1465 	}
1466 	return 0;
1467 }
1468 
1469 static int gfs2_reconfigure(struct fs_context *fc)
1470 {
1471 	struct super_block *sb = fc->root->d_sb;
1472 	struct gfs2_sbd *sdp = sb->s_fs_info;
1473 	struct gfs2_args *oldargs = &sdp->sd_args;
1474 	struct gfs2_args *newargs = fc->fs_private;
1475 	struct gfs2_tune *gt = &sdp->sd_tune;
1476 	int error = 0;
1477 
1478 	sync_filesystem(sb);
1479 
1480 	spin_lock(&gt->gt_spin);
1481 	oldargs->ar_commit = gt->gt_logd_secs;
1482 	oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1483 	if (gt->gt_statfs_slow)
1484 		oldargs->ar_statfs_quantum = 0;
1485 	else
1486 		oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1487 	spin_unlock(&gt->gt_spin);
1488 
1489 	if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1490 		errorfc(fc, "reconfiguration of locking protocol not allowed");
1491 		return -EINVAL;
1492 	}
1493 	if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1494 		errorfc(fc, "reconfiguration of lock table not allowed");
1495 		return -EINVAL;
1496 	}
1497 	if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1498 		errorfc(fc, "reconfiguration of host data not allowed");
1499 		return -EINVAL;
1500 	}
1501 	if (newargs->ar_spectator != oldargs->ar_spectator) {
1502 		errorfc(fc, "reconfiguration of spectator mode not allowed");
1503 		return -EINVAL;
1504 	}
1505 	if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1506 		errorfc(fc, "reconfiguration of localflocks not allowed");
1507 		return -EINVAL;
1508 	}
1509 	if (newargs->ar_meta != oldargs->ar_meta) {
1510 		errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1511 		return -EINVAL;
1512 	}
1513 	if (oldargs->ar_spectator)
1514 		fc->sb_flags |= SB_RDONLY;
1515 
1516 	if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1517 		if (fc->sb_flags & SB_RDONLY) {
1518 			error = gfs2_make_fs_ro(sdp);
1519 			if (error)
1520 				errorfc(fc, "unable to remount read-only");
1521 		} else {
1522 			error = gfs2_make_fs_rw(sdp);
1523 			if (error)
1524 				errorfc(fc, "unable to remount read-write");
1525 		}
1526 	}
1527 	sdp->sd_args = *newargs;
1528 
1529 	if (sdp->sd_args.ar_posix_acl)
1530 		sb->s_flags |= SB_POSIXACL;
1531 	else
1532 		sb->s_flags &= ~SB_POSIXACL;
1533 	if (sdp->sd_args.ar_nobarrier)
1534 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1535 	else
1536 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1537 	spin_lock(&gt->gt_spin);
1538 	gt->gt_logd_secs = newargs->ar_commit;
1539 	gt->gt_quota_quantum = newargs->ar_quota_quantum;
1540 	if (newargs->ar_statfs_quantum) {
1541 		gt->gt_statfs_slow = 0;
1542 		gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1543 	}
1544 	else {
1545 		gt->gt_statfs_slow = 1;
1546 		gt->gt_statfs_quantum = 30;
1547 	}
1548 	spin_unlock(&gt->gt_spin);
1549 
1550 	gfs2_online_uevent(sdp);
1551 	return error;
1552 }
1553 
1554 static const struct fs_context_operations gfs2_context_ops = {
1555 	.free        = gfs2_fc_free,
1556 	.parse_param = gfs2_parse_param,
1557 	.get_tree    = gfs2_get_tree,
1558 	.reconfigure = gfs2_reconfigure,
1559 };
1560 
1561 /* Set up the filesystem mount context */
1562 static int gfs2_init_fs_context(struct fs_context *fc)
1563 {
1564 	struct gfs2_args *args;
1565 
1566 	args = kmalloc(sizeof(*args), GFP_KERNEL);
1567 	if (args == NULL)
1568 		return -ENOMEM;
1569 
1570 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1571 		struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1572 
1573 		*args = sdp->sd_args;
1574 	} else {
1575 		memset(args, 0, sizeof(*args));
1576 		args->ar_quota = GFS2_QUOTA_DEFAULT;
1577 		args->ar_data = GFS2_DATA_DEFAULT;
1578 		args->ar_commit = 30;
1579 		args->ar_statfs_quantum = 30;
1580 		args->ar_quota_quantum = 60;
1581 		args->ar_errors = GFS2_ERRORS_DEFAULT;
1582 	}
1583 	fc->fs_private = args;
1584 	fc->ops = &gfs2_context_ops;
1585 	return 0;
1586 }
1587 
1588 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1589 {
1590 	return -EINVAL;
1591 }
1592 
1593 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1594 {
1595 	return (fc->sget_key == s->s_bdev);
1596 }
1597 
1598 static int gfs2_meta_get_tree(struct fs_context *fc)
1599 {
1600 	struct super_block *s;
1601 	struct gfs2_sbd *sdp;
1602 	struct path path;
1603 	int error;
1604 
1605 	if (!fc->source || !*fc->source)
1606 		return -EINVAL;
1607 
1608 	error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1609 	if (error) {
1610 		pr_warn("path_lookup on %s returned error %d\n",
1611 		        fc->source, error);
1612 		return error;
1613 	}
1614 	fc->fs_type = &gfs2_fs_type;
1615 	fc->sget_key = path.dentry->d_sb->s_bdev;
1616 	s = sget_fc(fc, test_meta_super, set_meta_super);
1617 	path_put(&path);
1618 	if (IS_ERR(s)) {
1619 		pr_warn("gfs2 mount does not exist\n");
1620 		return PTR_ERR(s);
1621 	}
1622 	if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1623 		deactivate_locked_super(s);
1624 		return -EBUSY;
1625 	}
1626 	sdp = s->s_fs_info;
1627 	fc->root = dget(sdp->sd_master_dir);
1628 	return 0;
1629 }
1630 
1631 static const struct fs_context_operations gfs2_meta_context_ops = {
1632 	.free        = gfs2_fc_free,
1633 	.get_tree    = gfs2_meta_get_tree,
1634 };
1635 
1636 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1637 {
1638 	int ret = gfs2_init_fs_context(fc);
1639 
1640 	if (ret)
1641 		return ret;
1642 
1643 	fc->ops = &gfs2_meta_context_ops;
1644 	return 0;
1645 }
1646 
1647 static void gfs2_kill_sb(struct super_block *sb)
1648 {
1649 	struct gfs2_sbd *sdp = sb->s_fs_info;
1650 
1651 	if (sdp == NULL) {
1652 		kill_block_super(sb);
1653 		return;
1654 	}
1655 
1656 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1657 	dput(sdp->sd_root_dir);
1658 	dput(sdp->sd_master_dir);
1659 	sdp->sd_root_dir = NULL;
1660 	sdp->sd_master_dir = NULL;
1661 	shrink_dcache_sb(sb);
1662 	kill_block_super(sb);
1663 }
1664 
1665 struct file_system_type gfs2_fs_type = {
1666 	.name = "gfs2",
1667 	.fs_flags = FS_REQUIRES_DEV,
1668 	.init_fs_context = gfs2_init_fs_context,
1669 	.parameters = gfs2_fs_parameters,
1670 	.kill_sb = gfs2_kill_sb,
1671 	.owner = THIS_MODULE,
1672 };
1673 MODULE_ALIAS_FS("gfs2");
1674 
1675 struct file_system_type gfs2meta_fs_type = {
1676 	.name = "gfs2meta",
1677 	.fs_flags = FS_REQUIRES_DEV,
1678 	.init_fs_context = gfs2_meta_init_fs_context,
1679 	.owner = THIS_MODULE,
1680 };
1681 MODULE_ALIAS_FS("gfs2meta");
1682