• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
gfs2_tune_init(struct gfs2_tune * gt)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 
free_sbd(struct gfs2_sbd * sdp)65 void free_sbd(struct gfs2_sbd *sdp)
66 {
67 	struct super_block *sb = sdp->sd_vfs;
68 
69 	if (sdp->sd_lkstats)
70 		free_percpu(sdp->sd_lkstats);
71 	sb->s_fs_info = NULL;
72 	kfree(sdp);
73 }
74 
init_sbd(struct super_block * sb)75 static struct gfs2_sbd *init_sbd(struct super_block *sb)
76 {
77 	struct gfs2_sbd *sdp;
78 
79 	sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
80 	if (!sdp)
81 		return NULL;
82 
83 	sdp->sd_vfs = sb;
84 	sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
85 	if (!sdp->sd_lkstats)
86 		goto fail;
87 	sb->s_fs_info = sdp;
88 
89 	set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
90 	gfs2_tune_init(&sdp->sd_tune);
91 
92 	init_waitqueue_head(&sdp->sd_kill_wait);
93 	init_waitqueue_head(&sdp->sd_async_glock_wait);
94 	atomic_set(&sdp->sd_glock_disposal, 0);
95 	init_completion(&sdp->sd_locking_init);
96 	init_completion(&sdp->sd_wdack);
97 	spin_lock_init(&sdp->sd_statfs_spin);
98 
99 	spin_lock_init(&sdp->sd_rindex_spin);
100 	sdp->sd_rindex_tree.rb_node = NULL;
101 
102 	INIT_LIST_HEAD(&sdp->sd_jindex_list);
103 	spin_lock_init(&sdp->sd_jindex_spin);
104 	mutex_init(&sdp->sd_jindex_mutex);
105 	init_completion(&sdp->sd_journal_ready);
106 
107 	INIT_LIST_HEAD(&sdp->sd_quota_list);
108 	mutex_init(&sdp->sd_quota_sync_mutex);
109 	init_waitqueue_head(&sdp->sd_quota_wait);
110 	spin_lock_init(&sdp->sd_bitmap_lock);
111 
112 	INIT_LIST_HEAD(&sdp->sd_sc_inodes_list);
113 
114 	spin_lock_init(&sdp->sd_log_lock);
115 	atomic_set(&sdp->sd_log_pinned, 0);
116 	INIT_LIST_HEAD(&sdp->sd_log_revokes);
117 	INIT_LIST_HEAD(&sdp->sd_log_ordered);
118 	spin_lock_init(&sdp->sd_ordered_lock);
119 
120 	init_waitqueue_head(&sdp->sd_log_waitq);
121 	init_waitqueue_head(&sdp->sd_logd_waitq);
122 	spin_lock_init(&sdp->sd_ail_lock);
123 	INIT_LIST_HEAD(&sdp->sd_ail1_list);
124 	INIT_LIST_HEAD(&sdp->sd_ail2_list);
125 
126 	init_rwsem(&sdp->sd_log_flush_lock);
127 	atomic_set(&sdp->sd_log_in_flight, 0);
128 	init_waitqueue_head(&sdp->sd_log_flush_wait);
129 	mutex_init(&sdp->sd_freeze_mutex);
130 	INIT_LIST_HEAD(&sdp->sd_dead_glocks);
131 
132 	return sdp;
133 
134 fail:
135 	free_sbd(sdp);
136 	return NULL;
137 }
138 
139 /**
140  * gfs2_check_sb - Check superblock
141  * @sdp: the filesystem
142  * @silent: Don't print a message if the check fails
143  *
144  * Checks the version code of the FS is one that we understand how to
145  * read and that the sizes of the various on-disk structures have not
146  * changed.
147  */
148 
gfs2_check_sb(struct gfs2_sbd * sdp,int silent)149 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
150 {
151 	struct gfs2_sb_host *sb = &sdp->sd_sb;
152 
153 	if (sb->sb_magic != GFS2_MAGIC ||
154 	    sb->sb_type != GFS2_METATYPE_SB) {
155 		if (!silent)
156 			pr_warn("not a GFS2 filesystem\n");
157 		return -EINVAL;
158 	}
159 
160 	if (sb->sb_fs_format < GFS2_FS_FORMAT_MIN ||
161 	    sb->sb_fs_format > GFS2_FS_FORMAT_MAX ||
162 	    sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
163 		fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
164 		return -EINVAL;
165 	}
166 
167 	if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
168 	    (sb->sb_bsize & (sb->sb_bsize - 1))) {
169 		pr_warn("Invalid block size\n");
170 		return -EINVAL;
171 	}
172 	if (sb->sb_bsize_shift != ffs(sb->sb_bsize) - 1) {
173 		pr_warn("Invalid block size shift\n");
174 		return -EINVAL;
175 	}
176 	return 0;
177 }
178 
gfs2_sb_in(struct gfs2_sbd * sdp,const struct gfs2_sb * str)179 static void gfs2_sb_in(struct gfs2_sbd *sdp, const struct gfs2_sb *str)
180 {
181 	struct gfs2_sb_host *sb = &sdp->sd_sb;
182 	struct super_block *s = sdp->sd_vfs;
183 
184 	sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
185 	sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
186 	sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
187 	sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
188 	sb->sb_bsize = be32_to_cpu(str->sb_bsize);
189 	sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
190 	sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
191 	sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
192 	sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
193 	sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
194 
195 	memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
196 	memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
197 	super_set_uuid(s, str->sb_uuid, 16);
198 }
199 
200 /**
201  * gfs2_read_super - Read the gfs2 super block from disk
202  * @sdp: The GFS2 super block
203  * @sector: The location of the super block
204  * @silent: Don't print a message if the check fails
205  *
206  * This uses the bio functions to read the super block from disk
207  * because we want to be 100% sure that we never read cached data.
208  * A super block is read twice only during each GFS2 mount and is
209  * never written to by the filesystem. The first time its read no
210  * locks are held, and the only details which are looked at are those
211  * relating to the locking protocol. Once locking is up and working,
212  * the sb is read again under the lock to establish the location of
213  * the master directory (contains pointers to journals etc) and the
214  * root directory.
215  *
216  * Returns: 0 on success or error
217  */
218 
gfs2_read_super(struct gfs2_sbd * sdp,sector_t sector,int silent)219 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
220 {
221 	struct super_block *sb = sdp->sd_vfs;
222 	struct page *page;
223 	struct bio_vec bvec;
224 	struct bio bio;
225 	int err;
226 
227 	page = alloc_page(GFP_KERNEL);
228 	if (unlikely(!page))
229 		return -ENOMEM;
230 
231 	bio_init(&bio, sb->s_bdev, &bvec, 1, REQ_OP_READ | REQ_META);
232 	bio.bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
233 	__bio_add_page(&bio, page, PAGE_SIZE, 0);
234 
235 	err = submit_bio_wait(&bio);
236 	if (err) {
237 		pr_warn("error %d reading superblock\n", err);
238 		__free_page(page);
239 		return err;
240 	}
241 	gfs2_sb_in(sdp, page_address(page));
242 	__free_page(page);
243 	return gfs2_check_sb(sdp, silent);
244 }
245 
246 /**
247  * gfs2_read_sb - Read super block
248  * @sdp: The GFS2 superblock
249  * @silent: Don't print message if mount fails
250  *
251  */
252 
gfs2_read_sb(struct gfs2_sbd * sdp,int silent)253 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
254 {
255 	u32 hash_blocks, ind_blocks, leaf_blocks;
256 	u32 tmp_blocks;
257 	unsigned int x;
258 	int error;
259 
260 	error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
261 	if (error) {
262 		if (!silent)
263 			fs_err(sdp, "can't read superblock\n");
264 		return error;
265 	}
266 
267 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 9;
268 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
269 	sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
270 			  sizeof(struct gfs2_dinode)) / sizeof(u64);
271 	sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
272 			  sizeof(struct gfs2_meta_header)) / sizeof(u64);
273 	sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
274 			  sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
275 	sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
276 	sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
277 	sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
278 	sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
279 	sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
280 				sizeof(struct gfs2_meta_header)) /
281 			        sizeof(struct gfs2_quota_change);
282 	sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
283 				     sizeof(struct gfs2_meta_header))
284 		* GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
285 
286 	/*
287 	 * We always keep at least one block reserved for revokes in
288 	 * transactions.  This greatly simplifies allocating additional
289 	 * revoke blocks.
290 	 */
291 	atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
292 
293 	/* Compute maximum reservation required to add a entry to a directory */
294 
295 	hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
296 			     sdp->sd_jbsize);
297 
298 	ind_blocks = 0;
299 	for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
300 		tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
301 		ind_blocks += tmp_blocks;
302 	}
303 
304 	leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
305 
306 	sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
307 
308 	sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
309 				sizeof(struct gfs2_dinode);
310 	sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
311 	for (x = 2;; x++) {
312 		u64 space, d;
313 		u32 m;
314 
315 		space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
316 		d = space;
317 		m = do_div(d, sdp->sd_inptrs);
318 
319 		if (d != sdp->sd_heightsize[x - 1] || m)
320 			break;
321 		sdp->sd_heightsize[x] = space;
322 	}
323 	sdp->sd_max_height = x;
324 	sdp->sd_heightsize[x] = ~0;
325 	gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
326 
327 	sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
328 				      sizeof(struct gfs2_leaf)) /
329 				     GFS2_MIN_DIRENT_SIZE;
330 	return 0;
331 }
332 
init_names(struct gfs2_sbd * sdp,int silent)333 static int init_names(struct gfs2_sbd *sdp, int silent)
334 {
335 	char *proto, *table;
336 	int error = 0;
337 
338 	proto = sdp->sd_args.ar_lockproto;
339 	table = sdp->sd_args.ar_locktable;
340 
341 	/*  Try to autodetect  */
342 
343 	if (!proto[0] || !table[0]) {
344 		error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
345 		if (error)
346 			return error;
347 
348 		if (!proto[0])
349 			proto = sdp->sd_sb.sb_lockproto;
350 		if (!table[0])
351 			table = sdp->sd_sb.sb_locktable;
352 	}
353 
354 	if (!table[0])
355 		table = sdp->sd_vfs->s_id;
356 
357 	BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN);
358 
359 	strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN);
360 	strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN);
361 
362 	table = sdp->sd_table_name;
363 	while ((table = strchr(table, '/')))
364 		*table = '_';
365 
366 	return error;
367 }
368 
init_locking(struct gfs2_sbd * sdp,struct gfs2_holder * mount_gh,int undo)369 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
370 			int undo)
371 {
372 	int error = 0;
373 
374 	if (undo)
375 		goto fail_trans;
376 
377 	error = gfs2_glock_nq_num(sdp,
378 				  GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
379 				  LM_ST_EXCLUSIVE,
380 				  LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
381 				  mount_gh);
382 	if (error) {
383 		fs_err(sdp, "can't acquire mount glock: %d\n", error);
384 		goto fail;
385 	}
386 
387 	error = gfs2_glock_nq_num(sdp,
388 				  GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
389 				  LM_ST_SHARED,
390 				  LM_FLAG_NOEXP | GL_EXACT | GL_NOPID,
391 				  &sdp->sd_live_gh);
392 	if (error) {
393 		fs_err(sdp, "can't acquire live glock: %d\n", error);
394 		goto fail_mount;
395 	}
396 
397 	error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
398 			       CREATE, &sdp->sd_rename_gl);
399 	if (error) {
400 		fs_err(sdp, "can't create rename glock: %d\n", error);
401 		goto fail_live;
402 	}
403 
404 	error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
405 			       CREATE, &sdp->sd_freeze_gl);
406 	if (error) {
407 		fs_err(sdp, "can't create freeze glock: %d\n", error);
408 		goto fail_rename;
409 	}
410 
411 	return 0;
412 
413 fail_trans:
414 	gfs2_glock_put(sdp->sd_freeze_gl);
415 fail_rename:
416 	gfs2_glock_put(sdp->sd_rename_gl);
417 fail_live:
418 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
419 fail_mount:
420 	gfs2_glock_dq_uninit(mount_gh);
421 fail:
422 	return error;
423 }
424 
gfs2_lookup_root(struct super_block * sb,struct dentry ** dptr,u64 no_addr,const char * name)425 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
426 			    u64 no_addr, const char *name)
427 {
428 	struct gfs2_sbd *sdp = sb->s_fs_info;
429 	struct dentry *dentry;
430 	struct inode *inode;
431 
432 	inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
433 				  GFS2_BLKST_FREE /* ignore */);
434 	if (IS_ERR(inode)) {
435 		fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
436 		return PTR_ERR(inode);
437 	}
438 	dentry = d_make_root(inode);
439 	if (!dentry) {
440 		fs_err(sdp, "can't alloc %s dentry\n", name);
441 		return -ENOMEM;
442 	}
443 	*dptr = dentry;
444 	return 0;
445 }
446 
init_sb(struct gfs2_sbd * sdp,int silent)447 static int init_sb(struct gfs2_sbd *sdp, int silent)
448 {
449 	struct super_block *sb = sdp->sd_vfs;
450 	struct gfs2_holder sb_gh;
451 	u64 no_addr;
452 	int ret;
453 
454 	ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
455 				LM_ST_SHARED, 0, &sb_gh);
456 	if (ret) {
457 		fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
458 		return ret;
459 	}
460 
461 	ret = gfs2_read_sb(sdp, silent);
462 	if (ret) {
463 		fs_err(sdp, "can't read superblock: %d\n", ret);
464 		goto out;
465 	}
466 
467 	switch(sdp->sd_sb.sb_fs_format) {
468 	case GFS2_FS_FORMAT_MAX:
469 		sb->s_xattr = gfs2_xattr_handlers_max;
470 		break;
471 
472 	case GFS2_FS_FORMAT_MIN:
473 		sb->s_xattr = gfs2_xattr_handlers_min;
474 		break;
475 
476 	default:
477 		BUG();
478 	}
479 
480 	/* Set up the buffer cache and SB for real */
481 	if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
482 		ret = -EINVAL;
483 		fs_err(sdp, "FS block size (%u) is too small for device "
484 		       "block size (%u)\n",
485 		       sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
486 		goto out;
487 	}
488 	if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
489 		ret = -EINVAL;
490 		fs_err(sdp, "FS block size (%u) is too big for machine "
491 		       "page size (%u)\n",
492 		       sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
493 		goto out;
494 	}
495 	sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
496 
497 	/* Get the root inode */
498 	no_addr = sdp->sd_sb.sb_root_dir.no_addr;
499 	ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
500 	if (ret)
501 		goto out;
502 
503 	/* Get the master inode */
504 	no_addr = sdp->sd_sb.sb_master_dir.no_addr;
505 	ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
506 	if (ret) {
507 		dput(sdp->sd_root_dir);
508 		goto out;
509 	}
510 	sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
511 out:
512 	gfs2_glock_dq_uninit(&sb_gh);
513 	return ret;
514 }
515 
gfs2_others_may_mount(struct gfs2_sbd * sdp)516 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
517 {
518 	char *message = "FIRSTMOUNT=Done";
519 	char *envp[] = { message, NULL };
520 
521 	fs_info(sdp, "first mount done, others may mount\n");
522 
523 	if (sdp->sd_lockstruct.ls_ops->lm_first_done)
524 		sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
525 
526 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
527 }
528 
529 /**
530  * gfs2_jindex_hold - Grab a lock on the jindex
531  * @sdp: The GFS2 superblock
532  * @ji_gh: the holder for the jindex glock
533  *
534  * Returns: errno
535  */
536 
gfs2_jindex_hold(struct gfs2_sbd * sdp,struct gfs2_holder * ji_gh)537 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
538 {
539 	struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
540 	struct qstr name;
541 	char buf[20];
542 	struct gfs2_jdesc *jd;
543 	int error;
544 
545 	name.name = buf;
546 
547 	mutex_lock(&sdp->sd_jindex_mutex);
548 
549 	for (;;) {
550 		struct gfs2_inode *jip;
551 
552 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
553 		if (error)
554 			break;
555 
556 		name.len = sprintf(buf, "journal%u", sdp->sd_journals);
557 		name.hash = gfs2_disk_hash(name.name, name.len);
558 
559 		error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
560 		if (error == -ENOENT) {
561 			error = 0;
562 			break;
563 		}
564 
565 		gfs2_glock_dq_uninit(ji_gh);
566 
567 		if (error)
568 			break;
569 
570 		error = -ENOMEM;
571 		jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
572 		if (!jd)
573 			break;
574 
575 		INIT_LIST_HEAD(&jd->extent_list);
576 		INIT_LIST_HEAD(&jd->jd_revoke_list);
577 
578 		INIT_WORK(&jd->jd_work, gfs2_recover_func);
579 		jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
580 		if (IS_ERR_OR_NULL(jd->jd_inode)) {
581 			if (!jd->jd_inode)
582 				error = -ENOENT;
583 			else
584 				error = PTR_ERR(jd->jd_inode);
585 			kfree(jd);
586 			break;
587 		}
588 
589 		d_mark_dontcache(jd->jd_inode);
590 		spin_lock(&sdp->sd_jindex_spin);
591 		jd->jd_jid = sdp->sd_journals++;
592 		jip = GFS2_I(jd->jd_inode);
593 		jd->jd_no_addr = jip->i_no_addr;
594 		list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
595 		spin_unlock(&sdp->sd_jindex_spin);
596 	}
597 
598 	mutex_unlock(&sdp->sd_jindex_mutex);
599 
600 	return error;
601 }
602 
603 /**
604  * init_statfs - look up and initialize master and local (per node) statfs inodes
605  * @sdp: The GFS2 superblock
606  *
607  * This should be called after the jindex is initialized in init_journal() and
608  * before gfs2_journal_recovery() is called because we need to be able to write
609  * to these inodes during recovery.
610  *
611  * Returns: errno
612  */
init_statfs(struct gfs2_sbd * sdp)613 static int init_statfs(struct gfs2_sbd *sdp)
614 {
615 	int error = 0;
616 	struct inode *master = d_inode(sdp->sd_master_dir);
617 	struct inode *pn = NULL;
618 	char buf[30];
619 	struct gfs2_jdesc *jd;
620 	struct gfs2_inode *ip;
621 
622 	sdp->sd_statfs_inode = gfs2_lookup_meta(master, "statfs");
623 	if (IS_ERR(sdp->sd_statfs_inode)) {
624 		error = PTR_ERR(sdp->sd_statfs_inode);
625 		fs_err(sdp, "can't read in statfs inode: %d\n", error);
626 		goto out;
627 	}
628 	if (sdp->sd_args.ar_spectator)
629 		goto out;
630 
631 	pn = gfs2_lookup_meta(master, "per_node");
632 	if (IS_ERR(pn)) {
633 		error = PTR_ERR(pn);
634 		fs_err(sdp, "can't find per_node directory: %d\n", error);
635 		goto put_statfs;
636 	}
637 
638 	/* For each jid, lookup the corresponding local statfs inode in the
639 	 * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
640 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
641 		struct local_statfs_inode *lsi =
642 			kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
643 		if (!lsi) {
644 			error = -ENOMEM;
645 			goto free_local;
646 		}
647 		sprintf(buf, "statfs_change%u", jd->jd_jid);
648 		lsi->si_sc_inode = gfs2_lookup_meta(pn, buf);
649 		if (IS_ERR(lsi->si_sc_inode)) {
650 			error = PTR_ERR(lsi->si_sc_inode);
651 			fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
652 			       jd->jd_jid, error);
653 			kfree(lsi);
654 			goto free_local;
655 		}
656 		lsi->si_jid = jd->jd_jid;
657 		if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
658 			sdp->sd_sc_inode = lsi->si_sc_inode;
659 
660 		list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
661 	}
662 
663 	iput(pn);
664 	pn = NULL;
665 	ip = GFS2_I(sdp->sd_sc_inode);
666 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
667 				   &sdp->sd_sc_gh);
668 	if (error) {
669 		fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
670 		goto free_local;
671 	}
672 	/* read in the local statfs buffer - other nodes don't change it. */
673 	error = gfs2_meta_inode_buffer(ip, &sdp->sd_sc_bh);
674 	if (error) {
675 		fs_err(sdp, "Cannot read in local statfs: %d\n", error);
676 		goto unlock_sd_gh;
677 	}
678 	return 0;
679 
680 unlock_sd_gh:
681 	gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
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 */
uninit_statfs(struct gfs2_sbd * sdp)692 static void uninit_statfs(struct gfs2_sbd *sdp)
693 {
694 	if (!sdp->sd_args.ar_spectator) {
695 		brelse(sdp->sd_sc_bh);
696 		gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
697 		free_local_statfs_inodes(sdp);
698 	}
699 	iput(sdp->sd_statfs_inode);
700 }
701 
init_journal(struct gfs2_sbd * sdp,int undo)702 static int init_journal(struct gfs2_sbd *sdp, int undo)
703 {
704 	struct inode *master = d_inode(sdp->sd_master_dir);
705 	struct gfs2_holder ji_gh;
706 	struct gfs2_inode *ip;
707 	int error = 0;
708 
709 	gfs2_holder_mark_uninitialized(&ji_gh);
710 	if (undo)
711 		goto fail_statfs;
712 
713 	sdp->sd_jindex = gfs2_lookup_meta(master, "jindex");
714 	if (IS_ERR(sdp->sd_jindex)) {
715 		fs_err(sdp, "can't lookup journal index: %d\n", error);
716 		return PTR_ERR(sdp->sd_jindex);
717 	}
718 
719 	/* Load in the journal index special file */
720 
721 	error = gfs2_jindex_hold(sdp, &ji_gh);
722 	if (error) {
723 		fs_err(sdp, "can't read journal index: %d\n", error);
724 		goto fail;
725 	}
726 
727 	error = -EUSERS;
728 	if (!gfs2_jindex_size(sdp)) {
729 		fs_err(sdp, "no journals!\n");
730 		goto fail_jindex;
731 	}
732 
733 	atomic_set(&sdp->sd_log_blks_needed, 0);
734 	if (sdp->sd_args.ar_spectator) {
735 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
736 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
737 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
738 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
739 	} else {
740 		if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
741 			fs_err(sdp, "can't mount journal #%u\n",
742 			       sdp->sd_lockstruct.ls_jid);
743 			fs_err(sdp, "there are only %u journals (0 - %u)\n",
744 			       gfs2_jindex_size(sdp),
745 			       gfs2_jindex_size(sdp) - 1);
746 			goto fail_jindex;
747 		}
748 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
749 
750 		error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
751 					  &gfs2_journal_glops,
752 					  LM_ST_EXCLUSIVE,
753 					  LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
754 					  &sdp->sd_journal_gh);
755 		if (error) {
756 			fs_err(sdp, "can't acquire journal glock: %d\n", error);
757 			goto fail_jindex;
758 		}
759 
760 		ip = GFS2_I(sdp->sd_jdesc->jd_inode);
761 		sdp->sd_jinode_gl = ip->i_gl;
762 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
763 					   LM_FLAG_NOEXP | GL_EXACT |
764 					   GL_NOCACHE | GL_NOPID,
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 	INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
824 	return 0;
825 
826 fail_statfs:
827 	uninit_statfs(sdp);
828 fail_jinode_gh:
829 	/* A withdraw may have done dq/uninit so now we need to check it */
830 	if (!sdp->sd_args.ar_spectator &&
831 	    gfs2_holder_initialized(&sdp->sd_jinode_gh))
832 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
833 fail_journal_gh:
834 	if (!sdp->sd_args.ar_spectator &&
835 	    gfs2_holder_initialized(&sdp->sd_journal_gh))
836 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
837 fail_jindex:
838 	gfs2_jindex_free(sdp);
839 	if (gfs2_holder_initialized(&ji_gh))
840 		gfs2_glock_dq_uninit(&ji_gh);
841 fail:
842 	iput(sdp->sd_jindex);
843 	return error;
844 }
845 
846 static struct lock_class_key gfs2_quota_imutex_key;
847 
init_inodes(struct gfs2_sbd * sdp,int undo)848 static int init_inodes(struct gfs2_sbd *sdp, int undo)
849 {
850 	int error = 0;
851 	struct inode *master = d_inode(sdp->sd_master_dir);
852 
853 	if (undo)
854 		goto fail_qinode;
855 
856 	error = init_journal(sdp, undo);
857 	complete_all(&sdp->sd_journal_ready);
858 	if (error)
859 		goto fail;
860 
861 	/* Read in the resource index inode */
862 	sdp->sd_rindex = gfs2_lookup_meta(master, "rindex");
863 	if (IS_ERR(sdp->sd_rindex)) {
864 		error = PTR_ERR(sdp->sd_rindex);
865 		fs_err(sdp, "can't get resource index inode: %d\n", error);
866 		goto fail_journal;
867 	}
868 	sdp->sd_rindex_uptodate = 0;
869 
870 	/* Read in the quota inode */
871 	sdp->sd_quota_inode = gfs2_lookup_meta(master, "quota");
872 	if (IS_ERR(sdp->sd_quota_inode)) {
873 		error = PTR_ERR(sdp->sd_quota_inode);
874 		fs_err(sdp, "can't get quota file inode: %d\n", error);
875 		goto fail_rindex;
876 	}
877 	/*
878 	 * i_rwsem on quota files is special. Since this inode is hidden system
879 	 * file, we are safe to define locking ourselves.
880 	 */
881 	lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
882 			  &gfs2_quota_imutex_key);
883 
884 	error = gfs2_rindex_update(sdp);
885 	if (error)
886 		goto fail_qinode;
887 
888 	return 0;
889 
890 fail_qinode:
891 	iput(sdp->sd_quota_inode);
892 fail_rindex:
893 	gfs2_clear_rgrpd(sdp);
894 	iput(sdp->sd_rindex);
895 fail_journal:
896 	init_journal(sdp, UNDO);
897 fail:
898 	return error;
899 }
900 
init_per_node(struct gfs2_sbd * sdp,int undo)901 static int init_per_node(struct gfs2_sbd *sdp, int undo)
902 {
903 	struct inode *pn = NULL;
904 	char buf[30];
905 	int error = 0;
906 	struct gfs2_inode *ip;
907 	struct inode *master = d_inode(sdp->sd_master_dir);
908 
909 	if (sdp->sd_args.ar_spectator)
910 		return 0;
911 
912 	if (undo)
913 		goto fail_qc_gh;
914 
915 	pn = gfs2_lookup_meta(master, "per_node");
916 	if (IS_ERR(pn)) {
917 		error = PTR_ERR(pn);
918 		fs_err(sdp, "can't find per_node directory: %d\n", error);
919 		return error;
920 	}
921 
922 	sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
923 	sdp->sd_qc_inode = gfs2_lookup_meta(pn, buf);
924 	if (IS_ERR(sdp->sd_qc_inode)) {
925 		error = PTR_ERR(sdp->sd_qc_inode);
926 		fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
927 		goto fail_ut_i;
928 	}
929 
930 	iput(pn);
931 	pn = NULL;
932 
933 	ip = GFS2_I(sdp->sd_qc_inode);
934 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
935 				   &sdp->sd_qc_gh);
936 	if (error) {
937 		fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
938 		goto fail_qc_i;
939 	}
940 
941 	return 0;
942 
943 fail_qc_gh:
944 	gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
945 fail_qc_i:
946 	iput(sdp->sd_qc_inode);
947 fail_ut_i:
948 	iput(pn);
949 	return error;
950 }
951 
952 static const match_table_t nolock_tokens = {
953 	{ Opt_jid, "jid=%d", },
954 	{ Opt_err, NULL },
955 };
956 
957 static const struct lm_lockops nolock_ops = {
958 	.lm_proto_name = "lock_nolock",
959 	.lm_put_lock = gfs2_glock_free,
960 	.lm_tokens = &nolock_tokens,
961 };
962 
963 /**
964  * gfs2_lm_mount - mount a locking protocol
965  * @sdp: the filesystem
966  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
967  *
968  * Returns: errno
969  */
970 
gfs2_lm_mount(struct gfs2_sbd * sdp,int silent)971 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
972 {
973 	const struct lm_lockops *lm;
974 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
975 	struct gfs2_args *args = &sdp->sd_args;
976 	const char *proto = sdp->sd_proto_name;
977 	const char *table = sdp->sd_table_name;
978 	char *o, *options;
979 	int ret;
980 
981 	if (!strcmp("lock_nolock", proto)) {
982 		lm = &nolock_ops;
983 		sdp->sd_args.ar_localflocks = 1;
984 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
985 	} else if (!strcmp("lock_dlm", proto)) {
986 		lm = &gfs2_dlm_ops;
987 #endif
988 	} else {
989 		pr_info("can't find protocol %s\n", proto);
990 		return -ENOENT;
991 	}
992 
993 	fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
994 
995 	ls->ls_ops = lm;
996 	ls->ls_first = 1;
997 
998 	for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
999 		substring_t tmp[MAX_OPT_ARGS];
1000 		int token, option;
1001 
1002 		if (!o || !*o)
1003 			continue;
1004 
1005 		token = match_token(o, *lm->lm_tokens, tmp);
1006 		switch (token) {
1007 		case Opt_jid:
1008 			ret = match_int(&tmp[0], &option);
1009 			if (ret || option < 0)
1010 				goto hostdata_error;
1011 			if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1012 				ls->ls_jid = option;
1013 			break;
1014 		case Opt_id:
1015 		case Opt_nodir:
1016 			/* Obsolete, but left for backward compat purposes */
1017 			break;
1018 		case Opt_first:
1019 			ret = match_int(&tmp[0], &option);
1020 			if (ret || (option != 0 && option != 1))
1021 				goto hostdata_error;
1022 			ls->ls_first = option;
1023 			break;
1024 		case Opt_err:
1025 		default:
1026 hostdata_error:
1027 			fs_info(sdp, "unknown hostdata (%s)\n", o);
1028 			return -EINVAL;
1029 		}
1030 	}
1031 
1032 	if (lm->lm_mount == NULL) {
1033 		fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format);
1034 		complete_all(&sdp->sd_locking_init);
1035 		return 0;
1036 	}
1037 	ret = lm->lm_mount(sdp, table);
1038 	if (ret == 0)
1039 		fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n",
1040 		        sdp->sd_sb.sb_fs_format);
1041 	complete_all(&sdp->sd_locking_init);
1042 	return ret;
1043 }
1044 
gfs2_lm_unmount(struct gfs2_sbd * sdp)1045 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1046 {
1047 	const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1048 	if (!gfs2_withdrawing_or_withdrawn(sdp) && lm->lm_unmount)
1049 		lm->lm_unmount(sdp);
1050 }
1051 
wait_on_journal(struct gfs2_sbd * sdp)1052 static int wait_on_journal(struct gfs2_sbd *sdp)
1053 {
1054 	if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1055 		return 0;
1056 
1057 	return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1058 		? -EINTR : 0;
1059 }
1060 
gfs2_online_uevent(struct gfs2_sbd * sdp)1061 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1062 {
1063 	struct super_block *sb = sdp->sd_vfs;
1064 	char ro[20];
1065 	char spectator[20];
1066 	char *envp[] = { ro, spectator, NULL };
1067 	sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1068 	sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1069 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1070 }
1071 
init_threads(struct gfs2_sbd * sdp)1072 static int init_threads(struct gfs2_sbd *sdp)
1073 {
1074 	struct task_struct *p;
1075 	int error = 0;
1076 
1077 	p = kthread_create(gfs2_logd, sdp, "gfs2_logd/%s", sdp->sd_fsname);
1078 	if (IS_ERR(p)) {
1079 		error = PTR_ERR(p);
1080 		fs_err(sdp, "can't create logd thread: %d\n", error);
1081 		return error;
1082 	}
1083 	get_task_struct(p);
1084 	sdp->sd_logd_process = p;
1085 
1086 	p = kthread_create(gfs2_quotad, sdp, "gfs2_quotad/%s", sdp->sd_fsname);
1087 	if (IS_ERR(p)) {
1088 		error = PTR_ERR(p);
1089 		fs_err(sdp, "can't create quotad thread: %d\n", error);
1090 		goto fail;
1091 	}
1092 	get_task_struct(p);
1093 	sdp->sd_quotad_process = p;
1094 
1095 	wake_up_process(sdp->sd_logd_process);
1096 	wake_up_process(sdp->sd_quotad_process);
1097 	return 0;
1098 
1099 fail:
1100 	kthread_stop_put(sdp->sd_logd_process);
1101 	sdp->sd_logd_process = NULL;
1102 	return error;
1103 }
1104 
gfs2_destroy_threads(struct gfs2_sbd * sdp)1105 void gfs2_destroy_threads(struct gfs2_sbd *sdp)
1106 {
1107 	if (sdp->sd_logd_process) {
1108 		kthread_stop_put(sdp->sd_logd_process);
1109 		sdp->sd_logd_process = NULL;
1110 	}
1111 	if (sdp->sd_quotad_process) {
1112 		kthread_stop_put(sdp->sd_quotad_process);
1113 		sdp->sd_quotad_process = NULL;
1114 	}
1115 }
1116 
1117 /**
1118  * gfs2_fill_super - Read in superblock
1119  * @sb: The VFS superblock
1120  * @fc: Mount options and flags
1121  *
1122  * Returns: -errno
1123  */
gfs2_fill_super(struct super_block * sb,struct fs_context * fc)1124 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1125 {
1126 	struct gfs2_args *args = fc->fs_private;
1127 	int silent = fc->sb_flags & SB_SILENT;
1128 	struct gfs2_sbd *sdp;
1129 	struct gfs2_holder mount_gh;
1130 	struct address_space *mapping;
1131 	int error;
1132 
1133 	sdp = init_sbd(sb);
1134 	if (!sdp) {
1135 		pr_warn("can't alloc struct gfs2_sbd\n");
1136 		return -ENOMEM;
1137 	}
1138 	sdp->sd_args = *args;
1139 
1140 	if (sdp->sd_args.ar_spectator) {
1141                 sb->s_flags |= SB_RDONLY;
1142 		set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1143 	}
1144 	if (sdp->sd_args.ar_posix_acl)
1145 		sb->s_flags |= SB_POSIXACL;
1146 	if (sdp->sd_args.ar_nobarrier)
1147 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1148 
1149 	sb->s_flags |= SB_NOSEC;
1150 	sb->s_magic = GFS2_MAGIC;
1151 	sb->s_op = &gfs2_super_ops;
1152 
1153 	sb->s_d_op = &gfs2_dops;
1154 	sb->s_export_op = &gfs2_export_ops;
1155 	sb->s_qcop = &gfs2_quotactl_ops;
1156 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1157 	sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1158 	sb->s_time_gran = 1;
1159 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1160 
1161 	/* Set up the buffer cache and fill in some fake block size values
1162 	   to allow us to read-in the on-disk superblock. */
1163 	sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, 512);
1164 	sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1165 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 9;
1166 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1167 
1168 	sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1169 	sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1170 	if (sdp->sd_args.ar_statfs_quantum) {
1171 		sdp->sd_tune.gt_statfs_slow = 0;
1172 		sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1173 	} else {
1174 		sdp->sd_tune.gt_statfs_slow = 1;
1175 		sdp->sd_tune.gt_statfs_quantum = 30;
1176 	}
1177 
1178 	/* Set up an address space for metadata writes */
1179 	sdp->sd_inode = new_inode(sb);
1180 	error = -ENOMEM;
1181 	if (!sdp->sd_inode)
1182 		goto fail_free;
1183 	sdp->sd_inode->i_ino = GFS2_BAD_INO;
1184 	sdp->sd_inode->i_size = OFFSET_MAX;
1185 
1186 	mapping = gfs2_aspace(sdp);
1187 	mapping->a_ops = &gfs2_rgrp_aops;
1188 	mapping_set_gfp_mask(mapping, GFP_NOFS);
1189 
1190 	error = init_names(sdp, silent);
1191 	if (error)
1192 		goto fail_iput;
1193 
1194 	snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1195 
1196 	error = -ENOMEM;
1197 	sdp->sd_glock_wq = alloc_workqueue("gfs2-glock/%s",
1198 			WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_FREEZABLE, 0,
1199 			sdp->sd_fsname);
1200 	if (!sdp->sd_glock_wq)
1201 		goto fail_iput;
1202 
1203 	sdp->sd_delete_wq = alloc_workqueue("gfs2-delete/%s",
1204 			WQ_MEM_RECLAIM | WQ_FREEZABLE, 0, sdp->sd_fsname);
1205 	if (!sdp->sd_delete_wq)
1206 		goto fail_glock_wq;
1207 
1208 	error = gfs2_sys_fs_add(sdp);
1209 	if (error)
1210 		goto fail_delete_wq;
1211 
1212 	gfs2_create_debugfs_file(sdp);
1213 
1214 	error = gfs2_lm_mount(sdp, silent);
1215 	if (error)
1216 		goto fail_debug;
1217 
1218 	error = init_locking(sdp, &mount_gh, DO);
1219 	if (error)
1220 		goto fail_lm;
1221 
1222 	error = init_sb(sdp, silent);
1223 	if (error)
1224 		goto fail_locking;
1225 
1226 	/* Turn rgrplvb on by default if fs format is recent enough */
1227 	if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801)
1228 		sdp->sd_args.ar_rgrplvb = 1;
1229 
1230 	error = wait_on_journal(sdp);
1231 	if (error)
1232 		goto fail_sb;
1233 
1234 	/*
1235 	 * If user space has failed to join the cluster or some similar
1236 	 * failure has occurred, then the journal id will contain a
1237 	 * negative (error) number. This will then be returned to the
1238 	 * caller (of the mount syscall). We do this even for spectator
1239 	 * mounts (which just write a jid of 0 to indicate "ok" even though
1240 	 * the jid is unused in the spectator case)
1241 	 */
1242 	if (sdp->sd_lockstruct.ls_jid < 0) {
1243 		error = sdp->sd_lockstruct.ls_jid;
1244 		sdp->sd_lockstruct.ls_jid = 0;
1245 		goto fail_sb;
1246 	}
1247 
1248 	if (sdp->sd_args.ar_spectator)
1249 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1250 			 sdp->sd_table_name);
1251 	else
1252 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1253 			 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1254 
1255 	error = init_inodes(sdp, DO);
1256 	if (error)
1257 		goto fail_sb;
1258 
1259 	error = init_per_node(sdp, DO);
1260 	if (error)
1261 		goto fail_inodes;
1262 
1263 	error = gfs2_statfs_init(sdp);
1264 	if (error) {
1265 		fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1266 		goto fail_per_node;
1267 	}
1268 
1269 	if (!sb_rdonly(sb)) {
1270 		error = init_threads(sdp);
1271 		if (error)
1272 			goto fail_per_node;
1273 	}
1274 
1275 	error = gfs2_freeze_lock_shared(sdp);
1276 	if (error)
1277 		goto fail_per_node;
1278 
1279 	if (!sb_rdonly(sb))
1280 		error = gfs2_make_fs_rw(sdp);
1281 
1282 	if (error) {
1283 		gfs2_freeze_unlock(sdp);
1284 		gfs2_destroy_threads(sdp);
1285 		fs_err(sdp, "can't make FS RW: %d\n", error);
1286 		goto fail_per_node;
1287 	}
1288 	gfs2_glock_dq_uninit(&mount_gh);
1289 	gfs2_online_uevent(sdp);
1290 	return 0;
1291 
1292 fail_per_node:
1293 	init_per_node(sdp, UNDO);
1294 fail_inodes:
1295 	init_inodes(sdp, UNDO);
1296 fail_sb:
1297 	if (sdp->sd_root_dir)
1298 		dput(sdp->sd_root_dir);
1299 	if (sdp->sd_master_dir)
1300 		dput(sdp->sd_master_dir);
1301 	if (sb->s_root)
1302 		dput(sb->s_root);
1303 	sb->s_root = NULL;
1304 fail_locking:
1305 	init_locking(sdp, &mount_gh, UNDO);
1306 fail_lm:
1307 	complete_all(&sdp->sd_journal_ready);
1308 	gfs2_gl_hash_clear(sdp);
1309 	gfs2_lm_unmount(sdp);
1310 fail_debug:
1311 	gfs2_delete_debugfs_file(sdp);
1312 	gfs2_sys_fs_del(sdp);
1313 fail_delete_wq:
1314 	destroy_workqueue(sdp->sd_delete_wq);
1315 fail_glock_wq:
1316 	if (sdp->sd_glock_wq)
1317 		destroy_workqueue(sdp->sd_glock_wq);
1318 fail_iput:
1319 	iput(sdp->sd_inode);
1320 fail_free:
1321 	free_sbd(sdp);
1322 	return error;
1323 }
1324 
1325 /**
1326  * gfs2_get_tree - Get the GFS2 superblock and root directory
1327  * @fc: The filesystem context
1328  *
1329  * Returns: 0 or -errno on error
1330  */
gfs2_get_tree(struct fs_context * fc)1331 static int gfs2_get_tree(struct fs_context *fc)
1332 {
1333 	struct gfs2_args *args = fc->fs_private;
1334 	struct gfs2_sbd *sdp;
1335 	int error;
1336 
1337 	error = get_tree_bdev(fc, gfs2_fill_super);
1338 	if (error)
1339 		return error;
1340 
1341 	sdp = fc->root->d_sb->s_fs_info;
1342 	dput(fc->root);
1343 	if (args->ar_meta)
1344 		fc->root = dget(sdp->sd_master_dir);
1345 	else
1346 		fc->root = dget(sdp->sd_root_dir);
1347 	return 0;
1348 }
1349 
gfs2_fc_free(struct fs_context * fc)1350 static void gfs2_fc_free(struct fs_context *fc)
1351 {
1352 	struct gfs2_args *args = fc->fs_private;
1353 
1354 	kfree(args);
1355 }
1356 
1357 enum gfs2_param {
1358 	Opt_lockproto,
1359 	Opt_locktable,
1360 	Opt_hostdata,
1361 	Opt_spectator,
1362 	Opt_ignore_local_fs,
1363 	Opt_localflocks,
1364 	Opt_localcaching,
1365 	Opt_debug,
1366 	Opt_upgrade,
1367 	Opt_acl,
1368 	Opt_quota,
1369 	Opt_quota_flag,
1370 	Opt_suiddir,
1371 	Opt_data,
1372 	Opt_meta,
1373 	Opt_discard,
1374 	Opt_commit,
1375 	Opt_errors,
1376 	Opt_statfs_quantum,
1377 	Opt_statfs_percent,
1378 	Opt_quota_quantum,
1379 	Opt_barrier,
1380 	Opt_rgrplvb,
1381 	Opt_loccookie,
1382 };
1383 
1384 static const struct constant_table gfs2_param_quota[] = {
1385 	{"off",        GFS2_QUOTA_OFF},
1386 	{"account",    GFS2_QUOTA_ACCOUNT},
1387 	{"on",         GFS2_QUOTA_ON},
1388 	{"quiet",      GFS2_QUOTA_QUIET},
1389 	{}
1390 };
1391 
1392 enum opt_data {
1393 	Opt_data_writeback = GFS2_DATA_WRITEBACK,
1394 	Opt_data_ordered   = GFS2_DATA_ORDERED,
1395 };
1396 
1397 static const struct constant_table gfs2_param_data[] = {
1398 	{"writeback",  Opt_data_writeback },
1399 	{"ordered",    Opt_data_ordered },
1400 	{}
1401 };
1402 
1403 enum opt_errors {
1404 	Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1405 	Opt_errors_panic    = GFS2_ERRORS_PANIC,
1406 };
1407 
1408 static const struct constant_table gfs2_param_errors[] = {
1409 	{"withdraw",   Opt_errors_withdraw },
1410 	{"panic",      Opt_errors_panic },
1411 	{}
1412 };
1413 
1414 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1415 	fsparam_string ("lockproto",          Opt_lockproto),
1416 	fsparam_string ("locktable",          Opt_locktable),
1417 	fsparam_string ("hostdata",           Opt_hostdata),
1418 	fsparam_flag   ("spectator",          Opt_spectator),
1419 	fsparam_flag   ("norecovery",         Opt_spectator),
1420 	fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1421 	fsparam_flag   ("localflocks",        Opt_localflocks),
1422 	fsparam_flag   ("localcaching",       Opt_localcaching),
1423 	fsparam_flag_no("debug",              Opt_debug),
1424 	fsparam_flag   ("upgrade",            Opt_upgrade),
1425 	fsparam_flag_no("acl",                Opt_acl),
1426 	fsparam_flag_no("suiddir",            Opt_suiddir),
1427 	fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1428 	fsparam_flag   ("meta",               Opt_meta),
1429 	fsparam_flag_no("discard",            Opt_discard),
1430 	fsparam_s32    ("commit",             Opt_commit),
1431 	fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1432 	fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1433 	fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1434 	fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1435 	fsparam_flag_no("barrier",            Opt_barrier),
1436 	fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1437 	fsparam_flag_no("loccookie",          Opt_loccookie),
1438 	/* quota can be a flag or an enum so it gets special treatment */
1439 	fsparam_flag_no("quota",	      Opt_quota_flag),
1440 	fsparam_enum("quota",		      Opt_quota, gfs2_param_quota),
1441 	{}
1442 };
1443 
1444 /* Parse a single mount parameter */
gfs2_parse_param(struct fs_context * fc,struct fs_parameter * param)1445 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1446 {
1447 	struct gfs2_args *args = fc->fs_private;
1448 	struct fs_parse_result result;
1449 	int o;
1450 
1451 	o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1452 	if (o < 0)
1453 		return o;
1454 
1455 	switch (o) {
1456 	case Opt_lockproto:
1457 		strscpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1458 		break;
1459 	case Opt_locktable:
1460 		strscpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1461 		break;
1462 	case Opt_hostdata:
1463 		strscpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1464 		break;
1465 	case Opt_spectator:
1466 		args->ar_spectator = 1;
1467 		break;
1468 	case Opt_ignore_local_fs:
1469 		/* Retained for backwards compat only */
1470 		break;
1471 	case Opt_localflocks:
1472 		args->ar_localflocks = 1;
1473 		break;
1474 	case Opt_localcaching:
1475 		/* Retained for backwards compat only */
1476 		break;
1477 	case Opt_debug:
1478 		if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1479 			return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1480 		args->ar_debug = result.boolean;
1481 		break;
1482 	case Opt_upgrade:
1483 		/* Retained for backwards compat only */
1484 		break;
1485 	case Opt_acl:
1486 		args->ar_posix_acl = result.boolean;
1487 		break;
1488 	case Opt_quota_flag:
1489 		args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1490 		break;
1491 	case Opt_quota:
1492 		args->ar_quota = result.int_32;
1493 		break;
1494 	case Opt_suiddir:
1495 		args->ar_suiddir = result.boolean;
1496 		break;
1497 	case Opt_data:
1498 		/* The uint_32 result maps directly to GFS2_DATA_* */
1499 		args->ar_data = result.uint_32;
1500 		break;
1501 	case Opt_meta:
1502 		args->ar_meta = 1;
1503 		break;
1504 	case Opt_discard:
1505 		args->ar_discard = result.boolean;
1506 		break;
1507 	case Opt_commit:
1508 		if (result.int_32 <= 0)
1509 			return invalfc(fc, "commit mount option requires a positive numeric argument");
1510 		args->ar_commit = result.int_32;
1511 		break;
1512 	case Opt_statfs_quantum:
1513 		if (result.int_32 < 0)
1514 			return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1515 		args->ar_statfs_quantum = result.int_32;
1516 		break;
1517 	case Opt_quota_quantum:
1518 		if (result.int_32 <= 0)
1519 			return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1520 		args->ar_quota_quantum = result.int_32;
1521 		break;
1522 	case Opt_statfs_percent:
1523 		if (result.int_32 < 0 || result.int_32 > 100)
1524 			return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1525 		args->ar_statfs_percent = result.int_32;
1526 		break;
1527 	case Opt_errors:
1528 		if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1529 			return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1530 		args->ar_errors = result.uint_32;
1531 		break;
1532 	case Opt_barrier:
1533 		args->ar_nobarrier = result.boolean;
1534 		break;
1535 	case Opt_rgrplvb:
1536 		args->ar_rgrplvb = result.boolean;
1537 		args->ar_got_rgrplvb = 1;
1538 		break;
1539 	case Opt_loccookie:
1540 		args->ar_loccookie = result.boolean;
1541 		break;
1542 	default:
1543 		return invalfc(fc, "invalid mount option: %s", param->key);
1544 	}
1545 	return 0;
1546 }
1547 
gfs2_reconfigure(struct fs_context * fc)1548 static int gfs2_reconfigure(struct fs_context *fc)
1549 {
1550 	struct super_block *sb = fc->root->d_sb;
1551 	struct gfs2_sbd *sdp = sb->s_fs_info;
1552 	struct gfs2_args *oldargs = &sdp->sd_args;
1553 	struct gfs2_args *newargs = fc->fs_private;
1554 	struct gfs2_tune *gt = &sdp->sd_tune;
1555 	int error = 0;
1556 
1557 	sync_filesystem(sb);
1558 
1559 	spin_lock(&gt->gt_spin);
1560 	oldargs->ar_commit = gt->gt_logd_secs;
1561 	oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1562 	if (gt->gt_statfs_slow)
1563 		oldargs->ar_statfs_quantum = 0;
1564 	else
1565 		oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1566 	spin_unlock(&gt->gt_spin);
1567 
1568 	if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1569 		errorfc(fc, "reconfiguration of locking protocol not allowed");
1570 		return -EINVAL;
1571 	}
1572 	if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1573 		errorfc(fc, "reconfiguration of lock table not allowed");
1574 		return -EINVAL;
1575 	}
1576 	if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1577 		errorfc(fc, "reconfiguration of host data not allowed");
1578 		return -EINVAL;
1579 	}
1580 	if (newargs->ar_spectator != oldargs->ar_spectator) {
1581 		errorfc(fc, "reconfiguration of spectator mode not allowed");
1582 		return -EINVAL;
1583 	}
1584 	if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1585 		errorfc(fc, "reconfiguration of localflocks not allowed");
1586 		return -EINVAL;
1587 	}
1588 	if (newargs->ar_meta != oldargs->ar_meta) {
1589 		errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1590 		return -EINVAL;
1591 	}
1592 	if (oldargs->ar_spectator)
1593 		fc->sb_flags |= SB_RDONLY;
1594 
1595 	if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1596 		if (fc->sb_flags & SB_RDONLY) {
1597 			gfs2_make_fs_ro(sdp);
1598 		} else {
1599 			error = gfs2_make_fs_rw(sdp);
1600 			if (error)
1601 				errorfc(fc, "unable to remount read-write");
1602 		}
1603 	}
1604 	sdp->sd_args = *newargs;
1605 
1606 	if (sdp->sd_args.ar_posix_acl)
1607 		sb->s_flags |= SB_POSIXACL;
1608 	else
1609 		sb->s_flags &= ~SB_POSIXACL;
1610 	if (sdp->sd_args.ar_nobarrier)
1611 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1612 	else
1613 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1614 	spin_lock(&gt->gt_spin);
1615 	gt->gt_logd_secs = newargs->ar_commit;
1616 	gt->gt_quota_quantum = newargs->ar_quota_quantum;
1617 	if (newargs->ar_statfs_quantum) {
1618 		gt->gt_statfs_slow = 0;
1619 		gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1620 	}
1621 	else {
1622 		gt->gt_statfs_slow = 1;
1623 		gt->gt_statfs_quantum = 30;
1624 	}
1625 	spin_unlock(&gt->gt_spin);
1626 
1627 	gfs2_online_uevent(sdp);
1628 	return error;
1629 }
1630 
1631 static const struct fs_context_operations gfs2_context_ops = {
1632 	.free        = gfs2_fc_free,
1633 	.parse_param = gfs2_parse_param,
1634 	.get_tree    = gfs2_get_tree,
1635 	.reconfigure = gfs2_reconfigure,
1636 };
1637 
1638 /* Set up the filesystem mount context */
gfs2_init_fs_context(struct fs_context * fc)1639 static int gfs2_init_fs_context(struct fs_context *fc)
1640 {
1641 	struct gfs2_args *args;
1642 
1643 	args = kmalloc(sizeof(*args), GFP_KERNEL);
1644 	if (args == NULL)
1645 		return -ENOMEM;
1646 
1647 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1648 		struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1649 
1650 		*args = sdp->sd_args;
1651 	} else {
1652 		memset(args, 0, sizeof(*args));
1653 		args->ar_quota = GFS2_QUOTA_DEFAULT;
1654 		args->ar_data = GFS2_DATA_DEFAULT;
1655 		args->ar_commit = 30;
1656 		args->ar_statfs_quantum = 30;
1657 		args->ar_quota_quantum = 60;
1658 		args->ar_errors = GFS2_ERRORS_DEFAULT;
1659 	}
1660 	fc->fs_private = args;
1661 	fc->ops = &gfs2_context_ops;
1662 	return 0;
1663 }
1664 
set_meta_super(struct super_block * s,struct fs_context * fc)1665 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1666 {
1667 	return -EINVAL;
1668 }
1669 
test_meta_super(struct super_block * s,struct fs_context * fc)1670 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1671 {
1672 	return (fc->sget_key == s->s_bdev);
1673 }
1674 
gfs2_meta_get_tree(struct fs_context * fc)1675 static int gfs2_meta_get_tree(struct fs_context *fc)
1676 {
1677 	struct super_block *s;
1678 	struct gfs2_sbd *sdp;
1679 	struct path path;
1680 	int error;
1681 
1682 	if (!fc->source || !*fc->source)
1683 		return -EINVAL;
1684 
1685 	error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1686 	if (error) {
1687 		pr_warn("path_lookup on %s returned error %d\n",
1688 		        fc->source, error);
1689 		return error;
1690 	}
1691 	fc->fs_type = &gfs2_fs_type;
1692 	fc->sget_key = path.dentry->d_sb->s_bdev;
1693 	s = sget_fc(fc, test_meta_super, set_meta_super);
1694 	path_put(&path);
1695 	if (IS_ERR(s)) {
1696 		pr_warn("gfs2 mount does not exist\n");
1697 		return PTR_ERR(s);
1698 	}
1699 	if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1700 		deactivate_locked_super(s);
1701 		return -EBUSY;
1702 	}
1703 	sdp = s->s_fs_info;
1704 	fc->root = dget(sdp->sd_master_dir);
1705 	return 0;
1706 }
1707 
1708 static const struct fs_context_operations gfs2_meta_context_ops = {
1709 	.free        = gfs2_fc_free,
1710 	.get_tree    = gfs2_meta_get_tree,
1711 };
1712 
gfs2_meta_init_fs_context(struct fs_context * fc)1713 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1714 {
1715 	int ret = gfs2_init_fs_context(fc);
1716 
1717 	if (ret)
1718 		return ret;
1719 
1720 	fc->ops = &gfs2_meta_context_ops;
1721 	return 0;
1722 }
1723 
1724 /**
1725  * gfs2_evict_inodes - evict inodes cooperatively
1726  * @sb: the superblock
1727  *
1728  * When evicting an inode with a zero link count, we are trying to upgrade the
1729  * inode's iopen glock from SH to EX mode in order to determine if we can
1730  * delete the inode.  The other nodes are supposed to evict the inode from
1731  * their caches if they can, and to poke the inode's inode glock if they cannot
1732  * do so.  Either behavior allows gfs2_upgrade_iopen_glock() to proceed
1733  * quickly, but if the other nodes are not cooperating, the lock upgrading
1734  * attempt will time out.  Since inodes are evicted sequentially, this can add
1735  * up quickly.
1736  *
1737  * Function evict_inodes() tries to keep the s_inode_list_lock list locked over
1738  * a long time, which prevents other inodes from being evicted concurrently.
1739  * This precludes the cooperative behavior we are looking for.  This special
1740  * version of evict_inodes() avoids that.
1741  *
1742  * Modeled after drop_pagecache_sb().
1743  */
gfs2_evict_inodes(struct super_block * sb)1744 static void gfs2_evict_inodes(struct super_block *sb)
1745 {
1746 	struct inode *inode, *toput_inode = NULL;
1747 	struct gfs2_sbd *sdp = sb->s_fs_info;
1748 
1749 	set_bit(SDF_EVICTING, &sdp->sd_flags);
1750 
1751 	spin_lock(&sb->s_inode_list_lock);
1752 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1753 		spin_lock(&inode->i_lock);
1754 		if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) &&
1755 		    !need_resched()) {
1756 			spin_unlock(&inode->i_lock);
1757 			continue;
1758 		}
1759 		atomic_inc(&inode->i_count);
1760 		spin_unlock(&inode->i_lock);
1761 		spin_unlock(&sb->s_inode_list_lock);
1762 
1763 		iput(toput_inode);
1764 		toput_inode = inode;
1765 
1766 		cond_resched();
1767 		spin_lock(&sb->s_inode_list_lock);
1768 	}
1769 	spin_unlock(&sb->s_inode_list_lock);
1770 	iput(toput_inode);
1771 }
1772 
gfs2_kill_sb(struct super_block * sb)1773 static void gfs2_kill_sb(struct super_block *sb)
1774 {
1775 	struct gfs2_sbd *sdp = sb->s_fs_info;
1776 
1777 	if (sdp == NULL) {
1778 		kill_block_super(sb);
1779 		return;
1780 	}
1781 
1782 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1783 	dput(sdp->sd_root_dir);
1784 	dput(sdp->sd_master_dir);
1785 	sdp->sd_root_dir = NULL;
1786 	sdp->sd_master_dir = NULL;
1787 	shrink_dcache_sb(sb);
1788 
1789 	gfs2_evict_inodes(sb);
1790 
1791 	/*
1792 	 * Flush and then drain the delete workqueue here (via
1793 	 * destroy_workqueue()) to ensure that any delete work that
1794 	 * may be running will also see the SDF_KILL flag.
1795 	 */
1796 	set_bit(SDF_KILL, &sdp->sd_flags);
1797 	gfs2_flush_delete_work(sdp);
1798 	destroy_workqueue(sdp->sd_delete_wq);
1799 
1800 	kill_block_super(sb);
1801 }
1802 
1803 struct file_system_type gfs2_fs_type = {
1804 	.name = "gfs2",
1805 	.fs_flags = FS_REQUIRES_DEV,
1806 	.init_fs_context = gfs2_init_fs_context,
1807 	.parameters = gfs2_fs_parameters,
1808 	.kill_sb = gfs2_kill_sb,
1809 	.owner = THIS_MODULE,
1810 };
1811 MODULE_ALIAS_FS("gfs2");
1812 
1813 struct file_system_type gfs2meta_fs_type = {
1814 	.name = "gfs2meta",
1815 	.fs_flags = FS_REQUIRES_DEV,
1816 	.init_fs_context = gfs2_meta_init_fs_context,
1817 	.owner = THIS_MODULE,
1818 };
1819 MODULE_ALIAS_FS("gfs2meta");
1820