xref: /openbmc/linux/fs/ceph/super.c (revision 2bc7d3e0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 
3 #include <linux/ceph/ceph_debug.h>
4 
5 #include <linux/backing-dev.h>
6 #include <linux/ctype.h>
7 #include <linux/fs.h>
8 #include <linux/inet.h>
9 #include <linux/in6.h>
10 #include <linux/module.h>
11 #include <linux/mount.h>
12 #include <linux/fs_context.h>
13 #include <linux/fs_parser.h>
14 #include <linux/sched.h>
15 #include <linux/seq_file.h>
16 #include <linux/slab.h>
17 #include <linux/statfs.h>
18 #include <linux/string.h>
19 
20 #include "super.h"
21 #include "mds_client.h"
22 #include "cache.h"
23 
24 #include <linux/ceph/ceph_features.h>
25 #include <linux/ceph/decode.h>
26 #include <linux/ceph/mon_client.h>
27 #include <linux/ceph/auth.h>
28 #include <linux/ceph/debugfs.h>
29 
30 #include <uapi/linux/magic.h>
31 
32 static DEFINE_SPINLOCK(ceph_fsc_lock);
33 static LIST_HEAD(ceph_fsc_list);
34 
35 /*
36  * Ceph superblock operations
37  *
38  * Handle the basics of mounting, unmounting.
39  */
40 
41 /*
42  * super ops
43  */
44 static void ceph_put_super(struct super_block *s)
45 {
46 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
47 
48 	dout("put_super\n");
49 	ceph_mdsc_close_sessions(fsc->mdsc);
50 }
51 
52 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
53 {
54 	struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
55 	struct ceph_mon_client *monc = &fsc->client->monc;
56 	struct ceph_statfs st;
57 	int i, err;
58 	u64 data_pool;
59 
60 	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
61 		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
62 	} else {
63 		data_pool = CEPH_NOPOOL;
64 	}
65 
66 	dout("statfs\n");
67 	err = ceph_monc_do_statfs(monc, data_pool, &st);
68 	if (err < 0)
69 		return err;
70 
71 	/* fill in kstatfs */
72 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
73 
74 	/*
75 	 * express utilization in terms of large blocks to avoid
76 	 * overflow on 32-bit machines.
77 	 *
78 	 * NOTE: for the time being, we make bsize == frsize to humor
79 	 * not-yet-ancient versions of glibc that are broken.
80 	 * Someday, we will probably want to report a real block
81 	 * size...  whatever that may mean for a network file system!
82 	 */
83 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
84 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
85 
86 	/*
87 	 * By default use root quota for stats; fallback to overall filesystem
88 	 * usage if using 'noquotadf' mount option or if the root dir doesn't
89 	 * have max_bytes quota set.
90 	 */
91 	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
92 	    !ceph_quota_update_statfs(fsc, buf)) {
93 		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
94 		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
95 		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
96 	}
97 
98 	buf->f_files = le64_to_cpu(st.num_objects);
99 	buf->f_ffree = -1;
100 	buf->f_namelen = NAME_MAX;
101 
102 	/* Must convert the fsid, for consistent values across arches */
103 	buf->f_fsid.val[0] = 0;
104 	mutex_lock(&monc->mutex);
105 	for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
106 		buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
107 	mutex_unlock(&monc->mutex);
108 
109 	/* fold the fs_cluster_id into the upper bits */
110 	buf->f_fsid.val[1] = monc->fs_cluster_id;
111 
112 	return 0;
113 }
114 
115 static int ceph_sync_fs(struct super_block *sb, int wait)
116 {
117 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
118 
119 	if (!wait) {
120 		dout("sync_fs (non-blocking)\n");
121 		ceph_flush_dirty_caps(fsc->mdsc);
122 		dout("sync_fs (non-blocking) done\n");
123 		return 0;
124 	}
125 
126 	dout("sync_fs (blocking)\n");
127 	ceph_osdc_sync(&fsc->client->osdc);
128 	ceph_mdsc_sync(fsc->mdsc);
129 	dout("sync_fs (blocking) done\n");
130 	return 0;
131 }
132 
133 /*
134  * mount options
135  */
136 enum {
137 	Opt_wsize,
138 	Opt_rsize,
139 	Opt_rasize,
140 	Opt_caps_wanted_delay_min,
141 	Opt_caps_wanted_delay_max,
142 	Opt_caps_max,
143 	Opt_readdir_max_entries,
144 	Opt_readdir_max_bytes,
145 	Opt_congestion_kb,
146 	/* int args above */
147 	Opt_snapdirname,
148 	Opt_mds_namespace,
149 	Opt_recover_session,
150 	Opt_source,
151 	Opt_mon_addr,
152 	/* string args above */
153 	Opt_dirstat,
154 	Opt_rbytes,
155 	Opt_asyncreaddir,
156 	Opt_dcache,
157 	Opt_ino32,
158 	Opt_fscache,
159 	Opt_poolperm,
160 	Opt_require_active_mds,
161 	Opt_acl,
162 	Opt_quotadf,
163 	Opt_copyfrom,
164 	Opt_wsync,
165 	Opt_pagecache,
166 };
167 
168 enum ceph_recover_session_mode {
169 	ceph_recover_session_no,
170 	ceph_recover_session_clean
171 };
172 
173 static const struct constant_table ceph_param_recover[] = {
174 	{ "no",		ceph_recover_session_no },
175 	{ "clean",	ceph_recover_session_clean },
176 	{}
177 };
178 
179 static const struct fs_parameter_spec ceph_mount_parameters[] = {
180 	fsparam_flag_no ("acl",				Opt_acl),
181 	fsparam_flag_no ("asyncreaddir",		Opt_asyncreaddir),
182 	fsparam_s32	("caps_max",			Opt_caps_max),
183 	fsparam_u32	("caps_wanted_delay_max",	Opt_caps_wanted_delay_max),
184 	fsparam_u32	("caps_wanted_delay_min",	Opt_caps_wanted_delay_min),
185 	fsparam_u32	("write_congestion_kb",		Opt_congestion_kb),
186 	fsparam_flag_no ("copyfrom",			Opt_copyfrom),
187 	fsparam_flag_no ("dcache",			Opt_dcache),
188 	fsparam_flag_no ("dirstat",			Opt_dirstat),
189 	fsparam_flag_no	("fsc",				Opt_fscache), // fsc|nofsc
190 	fsparam_string	("fsc",				Opt_fscache), // fsc=...
191 	fsparam_flag_no ("ino32",			Opt_ino32),
192 	fsparam_string	("mds_namespace",		Opt_mds_namespace),
193 	fsparam_flag_no ("poolperm",			Opt_poolperm),
194 	fsparam_flag_no ("quotadf",			Opt_quotadf),
195 	fsparam_u32	("rasize",			Opt_rasize),
196 	fsparam_flag_no ("rbytes",			Opt_rbytes),
197 	fsparam_u32	("readdir_max_bytes",		Opt_readdir_max_bytes),
198 	fsparam_u32	("readdir_max_entries",		Opt_readdir_max_entries),
199 	fsparam_enum	("recover_session",		Opt_recover_session, ceph_param_recover),
200 	fsparam_flag_no ("require_active_mds",		Opt_require_active_mds),
201 	fsparam_u32	("rsize",			Opt_rsize),
202 	fsparam_string	("snapdirname",			Opt_snapdirname),
203 	fsparam_string	("source",			Opt_source),
204 	fsparam_string	("mon_addr",			Opt_mon_addr),
205 	fsparam_u32	("wsize",			Opt_wsize),
206 	fsparam_flag_no	("wsync",			Opt_wsync),
207 	fsparam_flag_no	("pagecache",			Opt_pagecache),
208 	{}
209 };
210 
211 struct ceph_parse_opts_ctx {
212 	struct ceph_options		*copts;
213 	struct ceph_mount_options	*opts;
214 };
215 
216 /*
217  * Remove adjacent slashes and then the trailing slash, unless it is
218  * the only remaining character.
219  *
220  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
221  */
222 static void canonicalize_path(char *path)
223 {
224 	int i, j = 0;
225 
226 	for (i = 0; path[i] != '\0'; i++) {
227 		if (path[i] != '/' || j < 1 || path[j - 1] != '/')
228 			path[j++] = path[i];
229 	}
230 
231 	if (j > 1 && path[j - 1] == '/')
232 		j--;
233 	path[j] = '\0';
234 }
235 
236 /*
237  * Check if the mds namespace in ceph_mount_options matches
238  * the passed in namespace string. First time match (when
239  * ->mds_namespace is NULL) is treated specially, since
240  * ->mds_namespace needs to be initialized by the caller.
241  */
242 static int namespace_equals(struct ceph_mount_options *fsopt,
243 			    const char *namespace, size_t len)
244 {
245 	return !(fsopt->mds_namespace &&
246 		 (strlen(fsopt->mds_namespace) != len ||
247 		  strncmp(fsopt->mds_namespace, namespace, len)));
248 }
249 
250 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
251 				 struct fs_context *fc)
252 {
253 	int r;
254 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
255 	struct ceph_mount_options *fsopt = pctx->opts;
256 
257 	if (*dev_name_end != ':')
258 		return invalfc(fc, "separator ':' missing in source");
259 
260 	r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
261 			       pctx->copts, fc->log.log, ',');
262 	if (r)
263 		return r;
264 
265 	fsopt->new_dev_syntax = false;
266 	return 0;
267 }
268 
269 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
270 				 struct fs_context *fc)
271 {
272 	size_t len;
273 	struct ceph_fsid fsid;
274 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
275 	struct ceph_mount_options *fsopt = pctx->opts;
276 	char *fsid_start, *fs_name_start;
277 
278 	if (*dev_name_end != '=') {
279 		dout("separator '=' missing in source");
280 		return -EINVAL;
281 	}
282 
283 	fsid_start = strchr(dev_name, '@');
284 	if (!fsid_start)
285 		return invalfc(fc, "missing cluster fsid");
286 	++fsid_start; /* start of cluster fsid */
287 
288 	fs_name_start = strchr(fsid_start, '.');
289 	if (!fs_name_start)
290 		return invalfc(fc, "missing file system name");
291 
292 	if (ceph_parse_fsid(fsid_start, &fsid))
293 		return invalfc(fc, "Invalid FSID");
294 
295 	++fs_name_start; /* start of file system name */
296 	len = dev_name_end - fs_name_start;
297 
298 	if (!namespace_equals(fsopt, fs_name_start, len))
299 		return invalfc(fc, "Mismatching mds_namespace");
300 	kfree(fsopt->mds_namespace);
301 	fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
302 	if (!fsopt->mds_namespace)
303 		return -ENOMEM;
304 	dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
305 
306 	fsopt->new_dev_syntax = true;
307 	return 0;
308 }
309 
310 /*
311  * Parse the source parameter for new device format. Distinguish the device
312  * spec from the path. Try parsing new device format and fallback to old
313  * format if needed.
314  *
315  * New device syntax will looks like:
316  *     <device_spec>=/<path>
317  * where
318  *     <device_spec> is name@fsid.fsname
319  *     <path> is optional, but if present must begin with '/'
320  * (monitor addresses are passed via mount option)
321  *
322  * Old device syntax is:
323  *     <server_spec>[,<server_spec>...]:[<path>]
324  * where
325  *     <server_spec> is <ip>[:<port>]
326  *     <path> is optional, but if present must begin with '/'
327  */
328 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
329 {
330 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
331 	struct ceph_mount_options *fsopt = pctx->opts;
332 	char *dev_name = param->string, *dev_name_end;
333 	int ret;
334 
335 	dout("%s '%s'\n", __func__, dev_name);
336 	if (!dev_name || !*dev_name)
337 		return invalfc(fc, "Empty source");
338 
339 	dev_name_end = strchr(dev_name, '/');
340 	if (dev_name_end) {
341 		/*
342 		 * The server_path will include the whole chars from userland
343 		 * including the leading '/'.
344 		 */
345 		kfree(fsopt->server_path);
346 		fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
347 		if (!fsopt->server_path)
348 			return -ENOMEM;
349 
350 		canonicalize_path(fsopt->server_path);
351 	} else {
352 		dev_name_end = dev_name + strlen(dev_name);
353 	}
354 
355 	dev_name_end--;		/* back up to separator */
356 	if (dev_name_end < dev_name)
357 		return invalfc(fc, "Path missing in source");
358 
359 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
360 	if (fsopt->server_path)
361 		dout("server path '%s'\n", fsopt->server_path);
362 
363 	dout("trying new device syntax");
364 	ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
365 	if (ret) {
366 		if (ret != -EINVAL)
367 			return ret;
368 		dout("trying old device syntax");
369 		ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
370 		if (ret)
371 			return ret;
372 	}
373 
374 	fc->source = param->string;
375 	param->string = NULL;
376 	return 0;
377 }
378 
379 static int ceph_parse_mon_addr(struct fs_parameter *param,
380 			       struct fs_context *fc)
381 {
382 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
383 	struct ceph_mount_options *fsopt = pctx->opts;
384 
385 	kfree(fsopt->mon_addr);
386 	fsopt->mon_addr = param->string;
387 	param->string = NULL;
388 
389 	return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
390 				  pctx->copts, fc->log.log, '/');
391 }
392 
393 static int ceph_parse_mount_param(struct fs_context *fc,
394 				  struct fs_parameter *param)
395 {
396 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
397 	struct ceph_mount_options *fsopt = pctx->opts;
398 	struct fs_parse_result result;
399 	unsigned int mode;
400 	int token, ret;
401 
402 	ret = ceph_parse_param(param, pctx->copts, fc->log.log);
403 	if (ret != -ENOPARAM)
404 		return ret;
405 
406 	token = fs_parse(fc, ceph_mount_parameters, param, &result);
407 	dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
408 	if (token < 0)
409 		return token;
410 
411 	switch (token) {
412 	case Opt_snapdirname:
413 		kfree(fsopt->snapdir_name);
414 		fsopt->snapdir_name = param->string;
415 		param->string = NULL;
416 		break;
417 	case Opt_mds_namespace:
418 		if (!namespace_equals(fsopt, param->string, strlen(param->string)))
419 			return invalfc(fc, "Mismatching mds_namespace");
420 		kfree(fsopt->mds_namespace);
421 		fsopt->mds_namespace = param->string;
422 		param->string = NULL;
423 		break;
424 	case Opt_recover_session:
425 		mode = result.uint_32;
426 		if (mode == ceph_recover_session_no)
427 			fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
428 		else if (mode == ceph_recover_session_clean)
429 			fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
430 		else
431 			BUG();
432 		break;
433 	case Opt_source:
434 		if (fc->source)
435 			return invalfc(fc, "Multiple sources specified");
436 		return ceph_parse_source(param, fc);
437 	case Opt_mon_addr:
438 		return ceph_parse_mon_addr(param, fc);
439 	case Opt_wsize:
440 		if (result.uint_32 < PAGE_SIZE ||
441 		    result.uint_32 > CEPH_MAX_WRITE_SIZE)
442 			goto out_of_range;
443 		fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
444 		break;
445 	case Opt_rsize:
446 		if (result.uint_32 < PAGE_SIZE ||
447 		    result.uint_32 > CEPH_MAX_READ_SIZE)
448 			goto out_of_range;
449 		fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
450 		break;
451 	case Opt_rasize:
452 		fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
453 		break;
454 	case Opt_caps_wanted_delay_min:
455 		if (result.uint_32 < 1)
456 			goto out_of_range;
457 		fsopt->caps_wanted_delay_min = result.uint_32;
458 		break;
459 	case Opt_caps_wanted_delay_max:
460 		if (result.uint_32 < 1)
461 			goto out_of_range;
462 		fsopt->caps_wanted_delay_max = result.uint_32;
463 		break;
464 	case Opt_caps_max:
465 		if (result.int_32 < 0)
466 			goto out_of_range;
467 		fsopt->caps_max = result.int_32;
468 		break;
469 	case Opt_readdir_max_entries:
470 		if (result.uint_32 < 1)
471 			goto out_of_range;
472 		fsopt->max_readdir = result.uint_32;
473 		break;
474 	case Opt_readdir_max_bytes:
475 		if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
476 			goto out_of_range;
477 		fsopt->max_readdir_bytes = result.uint_32;
478 		break;
479 	case Opt_congestion_kb:
480 		if (result.uint_32 < 1024) /* at least 1M */
481 			goto out_of_range;
482 		fsopt->congestion_kb = result.uint_32;
483 		break;
484 	case Opt_dirstat:
485 		if (!result.negated)
486 			fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
487 		else
488 			fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
489 		break;
490 	case Opt_rbytes:
491 		if (!result.negated)
492 			fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
493 		else
494 			fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
495 		break;
496 	case Opt_asyncreaddir:
497 		if (!result.negated)
498 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
499 		else
500 			fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
501 		break;
502 	case Opt_dcache:
503 		if (!result.negated)
504 			fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
505 		else
506 			fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
507 		break;
508 	case Opt_ino32:
509 		if (!result.negated)
510 			fsopt->flags |= CEPH_MOUNT_OPT_INO32;
511 		else
512 			fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
513 		break;
514 
515 	case Opt_fscache:
516 #ifdef CONFIG_CEPH_FSCACHE
517 		kfree(fsopt->fscache_uniq);
518 		fsopt->fscache_uniq = NULL;
519 		if (result.negated) {
520 			fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
521 		} else {
522 			fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
523 			fsopt->fscache_uniq = param->string;
524 			param->string = NULL;
525 		}
526 		break;
527 #else
528 		return invalfc(fc, "fscache support is disabled");
529 #endif
530 	case Opt_poolperm:
531 		if (!result.negated)
532 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
533 		else
534 			fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
535 		break;
536 	case Opt_require_active_mds:
537 		if (!result.negated)
538 			fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
539 		else
540 			fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
541 		break;
542 	case Opt_quotadf:
543 		if (!result.negated)
544 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
545 		else
546 			fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
547 		break;
548 	case Opt_copyfrom:
549 		if (!result.negated)
550 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
551 		else
552 			fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
553 		break;
554 	case Opt_acl:
555 		if (!result.negated) {
556 #ifdef CONFIG_CEPH_FS_POSIX_ACL
557 			fc->sb_flags |= SB_POSIXACL;
558 #else
559 			return invalfc(fc, "POSIX ACL support is disabled");
560 #endif
561 		} else {
562 			fc->sb_flags &= ~SB_POSIXACL;
563 		}
564 		break;
565 	case Opt_wsync:
566 		if (!result.negated)
567 			fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
568 		else
569 			fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
570 		break;
571 	case Opt_pagecache:
572 		if (result.negated)
573 			fsopt->flags |= CEPH_MOUNT_OPT_NOPAGECACHE;
574 		else
575 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
576 		break;
577 	default:
578 		BUG();
579 	}
580 	return 0;
581 
582 out_of_range:
583 	return invalfc(fc, "%s out of range", param->key);
584 }
585 
586 static void destroy_mount_options(struct ceph_mount_options *args)
587 {
588 	dout("destroy_mount_options %p\n", args);
589 	if (!args)
590 		return;
591 
592 	kfree(args->snapdir_name);
593 	kfree(args->mds_namespace);
594 	kfree(args->server_path);
595 	kfree(args->fscache_uniq);
596 	kfree(args->mon_addr);
597 	kfree(args);
598 }
599 
600 static int strcmp_null(const char *s1, const char *s2)
601 {
602 	if (!s1 && !s2)
603 		return 0;
604 	if (s1 && !s2)
605 		return -1;
606 	if (!s1 && s2)
607 		return 1;
608 	return strcmp(s1, s2);
609 }
610 
611 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
612 				 struct ceph_options *new_opt,
613 				 struct ceph_fs_client *fsc)
614 {
615 	struct ceph_mount_options *fsopt1 = new_fsopt;
616 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
617 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
618 	int ret;
619 
620 	ret = memcmp(fsopt1, fsopt2, ofs);
621 	if (ret)
622 		return ret;
623 
624 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
625 	if (ret)
626 		return ret;
627 
628 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
629 	if (ret)
630 		return ret;
631 
632 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
633 	if (ret)
634 		return ret;
635 
636 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
637 	if (ret)
638 		return ret;
639 
640 	ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
641 	if (ret)
642 		return ret;
643 
644 	return ceph_compare_options(new_opt, fsc->client);
645 }
646 
647 /**
648  * ceph_show_options - Show mount options in /proc/mounts
649  * @m: seq_file to write to
650  * @root: root of that (sub)tree
651  */
652 static int ceph_show_options(struct seq_file *m, struct dentry *root)
653 {
654 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
655 	struct ceph_mount_options *fsopt = fsc->mount_options;
656 	size_t pos;
657 	int ret;
658 
659 	/* a comma between MNT/MS and client options */
660 	seq_putc(m, ',');
661 	pos = m->count;
662 
663 	ret = ceph_print_client_options(m, fsc->client, false);
664 	if (ret)
665 		return ret;
666 
667 	/* retract our comma if no client options */
668 	if (m->count == pos)
669 		m->count--;
670 
671 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
672 		seq_puts(m, ",dirstat");
673 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
674 		seq_puts(m, ",rbytes");
675 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
676 		seq_puts(m, ",noasyncreaddir");
677 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
678 		seq_puts(m, ",nodcache");
679 	if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
680 		seq_puts(m, ",ino32");
681 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
682 		seq_show_option(m, "fsc", fsopt->fscache_uniq);
683 	}
684 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
685 		seq_puts(m, ",nopoolperm");
686 	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
687 		seq_puts(m, ",noquotadf");
688 
689 #ifdef CONFIG_CEPH_FS_POSIX_ACL
690 	if (root->d_sb->s_flags & SB_POSIXACL)
691 		seq_puts(m, ",acl");
692 	else
693 		seq_puts(m, ",noacl");
694 #endif
695 
696 	if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
697 		seq_puts(m, ",copyfrom");
698 
699 	/* dump mds_namespace when old device syntax is in use */
700 	if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
701 		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
702 
703 	if (fsopt->mon_addr)
704 		seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
705 
706 	if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
707 		seq_show_option(m, "recover_session", "clean");
708 
709 	if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
710 		seq_puts(m, ",wsync");
711 
712 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
713 		seq_puts(m, ",nopagecache");
714 
715 	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
716 		seq_printf(m, ",wsize=%u", fsopt->wsize);
717 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
718 		seq_printf(m, ",rsize=%u", fsopt->rsize);
719 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
720 		seq_printf(m, ",rasize=%u", fsopt->rasize);
721 	if (fsopt->congestion_kb != default_congestion_kb())
722 		seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
723 	if (fsopt->caps_max)
724 		seq_printf(m, ",caps_max=%d", fsopt->caps_max);
725 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
726 		seq_printf(m, ",caps_wanted_delay_min=%u",
727 			 fsopt->caps_wanted_delay_min);
728 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
729 		seq_printf(m, ",caps_wanted_delay_max=%u",
730 			   fsopt->caps_wanted_delay_max);
731 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
732 		seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
733 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
734 		seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
735 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
736 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
737 
738 	return 0;
739 }
740 
741 /*
742  * handle any mon messages the standard library doesn't understand.
743  * return error if we don't either.
744  */
745 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
746 {
747 	struct ceph_fs_client *fsc = client->private;
748 	int type = le16_to_cpu(msg->hdr.type);
749 
750 	switch (type) {
751 	case CEPH_MSG_MDS_MAP:
752 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
753 		return 0;
754 	case CEPH_MSG_FS_MAP_USER:
755 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
756 		return 0;
757 	default:
758 		return -1;
759 	}
760 }
761 
762 /*
763  * create a new fs client
764  *
765  * Success or not, this function consumes @fsopt and @opt.
766  */
767 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
768 					struct ceph_options *opt)
769 {
770 	struct ceph_fs_client *fsc;
771 	int err;
772 
773 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
774 	if (!fsc) {
775 		err = -ENOMEM;
776 		goto fail;
777 	}
778 
779 	fsc->client = ceph_create_client(opt, fsc);
780 	if (IS_ERR(fsc->client)) {
781 		err = PTR_ERR(fsc->client);
782 		goto fail;
783 	}
784 	opt = NULL; /* fsc->client now owns this */
785 
786 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
787 	ceph_set_opt(fsc->client, ABORT_ON_FULL);
788 
789 	if (!fsopt->mds_namespace) {
790 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
791 				   0, true);
792 	} else {
793 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
794 				   0, false);
795 	}
796 
797 	fsc->mount_options = fsopt;
798 
799 	fsc->sb = NULL;
800 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
801 	fsc->filp_gen = 1;
802 	fsc->have_copy_from2 = true;
803 
804 	atomic_long_set(&fsc->writeback_count, 0);
805 	fsc->write_congested = false;
806 
807 	err = -ENOMEM;
808 	/*
809 	 * The number of concurrent works can be high but they don't need
810 	 * to be processed in parallel, limit concurrency.
811 	 */
812 	fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
813 	if (!fsc->inode_wq)
814 		goto fail_client;
815 	fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
816 	if (!fsc->cap_wq)
817 		goto fail_inode_wq;
818 
819 	spin_lock(&ceph_fsc_lock);
820 	list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
821 	spin_unlock(&ceph_fsc_lock);
822 
823 	return fsc;
824 
825 fail_inode_wq:
826 	destroy_workqueue(fsc->inode_wq);
827 fail_client:
828 	ceph_destroy_client(fsc->client);
829 fail:
830 	kfree(fsc);
831 	if (opt)
832 		ceph_destroy_options(opt);
833 	destroy_mount_options(fsopt);
834 	return ERR_PTR(err);
835 }
836 
837 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
838 {
839 	flush_workqueue(fsc->inode_wq);
840 	flush_workqueue(fsc->cap_wq);
841 }
842 
843 static void destroy_fs_client(struct ceph_fs_client *fsc)
844 {
845 	dout("destroy_fs_client %p\n", fsc);
846 
847 	spin_lock(&ceph_fsc_lock);
848 	list_del(&fsc->metric_wakeup);
849 	spin_unlock(&ceph_fsc_lock);
850 
851 	ceph_mdsc_destroy(fsc);
852 	destroy_workqueue(fsc->inode_wq);
853 	destroy_workqueue(fsc->cap_wq);
854 
855 	destroy_mount_options(fsc->mount_options);
856 
857 	ceph_destroy_client(fsc->client);
858 
859 	kfree(fsc);
860 	dout("destroy_fs_client %p done\n", fsc);
861 }
862 
863 /*
864  * caches
865  */
866 struct kmem_cache *ceph_inode_cachep;
867 struct kmem_cache *ceph_cap_cachep;
868 struct kmem_cache *ceph_cap_snap_cachep;
869 struct kmem_cache *ceph_cap_flush_cachep;
870 struct kmem_cache *ceph_dentry_cachep;
871 struct kmem_cache *ceph_file_cachep;
872 struct kmem_cache *ceph_dir_file_cachep;
873 struct kmem_cache *ceph_mds_request_cachep;
874 mempool_t *ceph_wb_pagevec_pool;
875 
876 static void ceph_inode_init_once(void *foo)
877 {
878 	struct ceph_inode_info *ci = foo;
879 	inode_init_once(&ci->vfs_inode);
880 }
881 
882 static int __init init_caches(void)
883 {
884 	int error = -ENOMEM;
885 
886 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
887 				      sizeof(struct ceph_inode_info),
888 				      __alignof__(struct ceph_inode_info),
889 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
890 				      SLAB_ACCOUNT, ceph_inode_init_once);
891 	if (!ceph_inode_cachep)
892 		return -ENOMEM;
893 
894 	ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
895 	if (!ceph_cap_cachep)
896 		goto bad_cap;
897 	ceph_cap_snap_cachep = KMEM_CACHE(ceph_cap_snap, SLAB_MEM_SPREAD);
898 	if (!ceph_cap_snap_cachep)
899 		goto bad_cap_snap;
900 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
901 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
902 	if (!ceph_cap_flush_cachep)
903 		goto bad_cap_flush;
904 
905 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
906 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
907 	if (!ceph_dentry_cachep)
908 		goto bad_dentry;
909 
910 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
911 	if (!ceph_file_cachep)
912 		goto bad_file;
913 
914 	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
915 	if (!ceph_dir_file_cachep)
916 		goto bad_dir_file;
917 
918 	ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, SLAB_MEM_SPREAD);
919 	if (!ceph_mds_request_cachep)
920 		goto bad_mds_req;
921 
922 	ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10, CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT);
923 	if (!ceph_wb_pagevec_pool)
924 		goto bad_pagevec_pool;
925 
926 	return 0;
927 
928 bad_pagevec_pool:
929 	kmem_cache_destroy(ceph_mds_request_cachep);
930 bad_mds_req:
931 	kmem_cache_destroy(ceph_dir_file_cachep);
932 bad_dir_file:
933 	kmem_cache_destroy(ceph_file_cachep);
934 bad_file:
935 	kmem_cache_destroy(ceph_dentry_cachep);
936 bad_dentry:
937 	kmem_cache_destroy(ceph_cap_flush_cachep);
938 bad_cap_flush:
939 	kmem_cache_destroy(ceph_cap_snap_cachep);
940 bad_cap_snap:
941 	kmem_cache_destroy(ceph_cap_cachep);
942 bad_cap:
943 	kmem_cache_destroy(ceph_inode_cachep);
944 	return error;
945 }
946 
947 static void destroy_caches(void)
948 {
949 	/*
950 	 * Make sure all delayed rcu free inodes are flushed before we
951 	 * destroy cache.
952 	 */
953 	rcu_barrier();
954 
955 	kmem_cache_destroy(ceph_inode_cachep);
956 	kmem_cache_destroy(ceph_cap_cachep);
957 	kmem_cache_destroy(ceph_cap_snap_cachep);
958 	kmem_cache_destroy(ceph_cap_flush_cachep);
959 	kmem_cache_destroy(ceph_dentry_cachep);
960 	kmem_cache_destroy(ceph_file_cachep);
961 	kmem_cache_destroy(ceph_dir_file_cachep);
962 	kmem_cache_destroy(ceph_mds_request_cachep);
963 	mempool_destroy(ceph_wb_pagevec_pool);
964 }
965 
966 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
967 {
968 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
969 	ceph_mdsc_force_umount(fsc->mdsc);
970 	fsc->filp_gen++; // invalidate open files
971 }
972 
973 /*
974  * ceph_umount_begin - initiate forced umount.  Tear down the
975  * mount, skipping steps that may hang while waiting for server(s).
976  */
977 void ceph_umount_begin(struct super_block *sb)
978 {
979 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
980 
981 	dout("ceph_umount_begin - starting forced umount\n");
982 	if (!fsc)
983 		return;
984 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
985 	__ceph_umount_begin(fsc);
986 }
987 
988 static const struct super_operations ceph_super_ops = {
989 	.alloc_inode	= ceph_alloc_inode,
990 	.free_inode	= ceph_free_inode,
991 	.write_inode    = ceph_write_inode,
992 	.drop_inode	= generic_delete_inode,
993 	.evict_inode	= ceph_evict_inode,
994 	.sync_fs        = ceph_sync_fs,
995 	.put_super	= ceph_put_super,
996 	.show_options   = ceph_show_options,
997 	.statfs		= ceph_statfs,
998 	.umount_begin   = ceph_umount_begin,
999 };
1000 
1001 /*
1002  * Bootstrap mount by opening the root directory.  Note the mount
1003  * @started time from caller, and time out if this takes too long.
1004  */
1005 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1006 				       const char *path,
1007 				       unsigned long started)
1008 {
1009 	struct ceph_mds_client *mdsc = fsc->mdsc;
1010 	struct ceph_mds_request *req = NULL;
1011 	int err;
1012 	struct dentry *root;
1013 
1014 	/* open dir */
1015 	dout("open_root_inode opening '%s'\n", path);
1016 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1017 	if (IS_ERR(req))
1018 		return ERR_CAST(req);
1019 	req->r_path1 = kstrdup(path, GFP_NOFS);
1020 	if (!req->r_path1) {
1021 		root = ERR_PTR(-ENOMEM);
1022 		goto out;
1023 	}
1024 
1025 	req->r_ino1.ino = CEPH_INO_ROOT;
1026 	req->r_ino1.snap = CEPH_NOSNAP;
1027 	req->r_started = started;
1028 	req->r_timeout = fsc->client->options->mount_timeout;
1029 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1030 	req->r_num_caps = 2;
1031 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1032 	if (err == 0) {
1033 		struct inode *inode = req->r_target_inode;
1034 		req->r_target_inode = NULL;
1035 		dout("open_root_inode success\n");
1036 		root = d_make_root(inode);
1037 		if (!root) {
1038 			root = ERR_PTR(-ENOMEM);
1039 			goto out;
1040 		}
1041 		dout("open_root_inode success, root dentry is %p\n", root);
1042 	} else {
1043 		root = ERR_PTR(err);
1044 	}
1045 out:
1046 	ceph_mdsc_put_request(req);
1047 	return root;
1048 }
1049 
1050 /*
1051  * mount: join the ceph cluster, and open root directory.
1052  */
1053 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1054 				      struct fs_context *fc)
1055 {
1056 	int err;
1057 	unsigned long started = jiffies;  /* note the start time */
1058 	struct dentry *root;
1059 
1060 	dout("mount start %p\n", fsc);
1061 	mutex_lock(&fsc->client->mount_mutex);
1062 
1063 	if (!fsc->sb->s_root) {
1064 		const char *path = fsc->mount_options->server_path ?
1065 				     fsc->mount_options->server_path + 1 : "";
1066 
1067 		err = __ceph_open_session(fsc->client, started);
1068 		if (err < 0)
1069 			goto out;
1070 
1071 		/* setup fscache */
1072 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1073 			err = ceph_fscache_register_fs(fsc, fc);
1074 			if (err < 0)
1075 				goto out;
1076 		}
1077 
1078 		dout("mount opening path '%s'\n", path);
1079 
1080 		ceph_fs_debugfs_init(fsc);
1081 
1082 		root = open_root_dentry(fsc, path, started);
1083 		if (IS_ERR(root)) {
1084 			err = PTR_ERR(root);
1085 			goto out;
1086 		}
1087 		fsc->sb->s_root = dget(root);
1088 	} else {
1089 		root = dget(fsc->sb->s_root);
1090 	}
1091 
1092 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1093 	dout("mount success\n");
1094 	mutex_unlock(&fsc->client->mount_mutex);
1095 	return root;
1096 
1097 out:
1098 	mutex_unlock(&fsc->client->mount_mutex);
1099 	return ERR_PTR(err);
1100 }
1101 
1102 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1103 {
1104 	struct ceph_fs_client *fsc = s->s_fs_info;
1105 	int ret;
1106 
1107 	dout("set_super %p\n", s);
1108 
1109 	s->s_maxbytes = MAX_LFS_FILESIZE;
1110 
1111 	s->s_xattr = ceph_xattr_handlers;
1112 	fsc->sb = s;
1113 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1114 
1115 	s->s_op = &ceph_super_ops;
1116 	s->s_d_op = &ceph_dentry_ops;
1117 	s->s_export_op = &ceph_export_ops;
1118 
1119 	s->s_time_gran = 1;
1120 	s->s_time_min = 0;
1121 	s->s_time_max = U32_MAX;
1122 	s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1123 
1124 	ret = set_anon_super_fc(s, fc);
1125 	if (ret != 0)
1126 		fsc->sb = NULL;
1127 	return ret;
1128 }
1129 
1130 /*
1131  * share superblock if same fs AND options
1132  */
1133 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1134 {
1135 	struct ceph_fs_client *new = fc->s_fs_info;
1136 	struct ceph_mount_options *fsopt = new->mount_options;
1137 	struct ceph_options *opt = new->client->options;
1138 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1139 
1140 	dout("ceph_compare_super %p\n", sb);
1141 
1142 	if (compare_mount_options(fsopt, opt, fsc)) {
1143 		dout("monitor(s)/mount options don't match\n");
1144 		return 0;
1145 	}
1146 	if ((opt->flags & CEPH_OPT_FSID) &&
1147 	    ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1148 		dout("fsid doesn't match\n");
1149 		return 0;
1150 	}
1151 	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1152 		dout("flags differ\n");
1153 		return 0;
1154 	}
1155 
1156 	if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1157 		dout("client is blocklisted (and CLEANRECOVER is not set)\n");
1158 		return 0;
1159 	}
1160 
1161 	if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1162 		dout("client has been forcibly unmounted\n");
1163 		return 0;
1164 	}
1165 
1166 	return 1;
1167 }
1168 
1169 /*
1170  * construct our own bdi so we can control readahead, etc.
1171  */
1172 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1173 
1174 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1175 {
1176 	int err;
1177 
1178 	err = super_setup_bdi_name(sb, "ceph-%ld",
1179 				   atomic_long_inc_return(&bdi_seq));
1180 	if (err)
1181 		return err;
1182 
1183 	/* set ra_pages based on rasize mount option? */
1184 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1185 
1186 	/* set io_pages based on max osd read size */
1187 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1188 
1189 	return 0;
1190 }
1191 
1192 static int ceph_get_tree(struct fs_context *fc)
1193 {
1194 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1195 	struct ceph_mount_options *fsopt = pctx->opts;
1196 	struct super_block *sb;
1197 	struct ceph_fs_client *fsc;
1198 	struct dentry *res;
1199 	int (*compare_super)(struct super_block *, struct fs_context *) =
1200 		ceph_compare_super;
1201 	int err;
1202 
1203 	dout("ceph_get_tree\n");
1204 
1205 	if (!fc->source)
1206 		return invalfc(fc, "No source");
1207 	if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1208 		return invalfc(fc, "No monitor address");
1209 
1210 	/* create client (which we may/may not use) */
1211 	fsc = create_fs_client(pctx->opts, pctx->copts);
1212 	pctx->opts = NULL;
1213 	pctx->copts = NULL;
1214 	if (IS_ERR(fsc)) {
1215 		err = PTR_ERR(fsc);
1216 		goto out_final;
1217 	}
1218 
1219 	err = ceph_mdsc_init(fsc);
1220 	if (err < 0)
1221 		goto out;
1222 
1223 	if (ceph_test_opt(fsc->client, NOSHARE))
1224 		compare_super = NULL;
1225 
1226 	fc->s_fs_info = fsc;
1227 	sb = sget_fc(fc, compare_super, ceph_set_super);
1228 	fc->s_fs_info = NULL;
1229 	if (IS_ERR(sb)) {
1230 		err = PTR_ERR(sb);
1231 		goto out;
1232 	}
1233 
1234 	if (ceph_sb_to_client(sb) != fsc) {
1235 		destroy_fs_client(fsc);
1236 		fsc = ceph_sb_to_client(sb);
1237 		dout("get_sb got existing client %p\n", fsc);
1238 	} else {
1239 		dout("get_sb using new client %p\n", fsc);
1240 		err = ceph_setup_bdi(sb, fsc);
1241 		if (err < 0)
1242 			goto out_splat;
1243 	}
1244 
1245 	res = ceph_real_mount(fsc, fc);
1246 	if (IS_ERR(res)) {
1247 		err = PTR_ERR(res);
1248 		goto out_splat;
1249 	}
1250 	dout("root %p inode %p ino %llx.%llx\n", res,
1251 	     d_inode(res), ceph_vinop(d_inode(res)));
1252 	fc->root = fsc->sb->s_root;
1253 	return 0;
1254 
1255 out_splat:
1256 	if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1257 		pr_info("No mds server is up or the cluster is laggy\n");
1258 		err = -EHOSTUNREACH;
1259 	}
1260 
1261 	ceph_mdsc_close_sessions(fsc->mdsc);
1262 	deactivate_locked_super(sb);
1263 	goto out_final;
1264 
1265 out:
1266 	destroy_fs_client(fsc);
1267 out_final:
1268 	dout("ceph_get_tree fail %d\n", err);
1269 	return err;
1270 }
1271 
1272 static void ceph_free_fc(struct fs_context *fc)
1273 {
1274 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1275 
1276 	if (pctx) {
1277 		destroy_mount_options(pctx->opts);
1278 		ceph_destroy_options(pctx->copts);
1279 		kfree(pctx);
1280 	}
1281 }
1282 
1283 static int ceph_reconfigure_fc(struct fs_context *fc)
1284 {
1285 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1286 	struct ceph_mount_options *fsopt = pctx->opts;
1287 	struct ceph_fs_client *fsc = ceph_sb_to_client(fc->root->d_sb);
1288 
1289 	if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1290 		ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1291 	else
1292 		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1293 
1294 	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1295 		kfree(fsc->mount_options->mon_addr);
1296 		fsc->mount_options->mon_addr = fsopt->mon_addr;
1297 		fsopt->mon_addr = NULL;
1298 		pr_notice("ceph: monitor addresses recorded, but not used for reconnection");
1299 	}
1300 
1301 	sync_filesystem(fc->root->d_sb);
1302 	return 0;
1303 }
1304 
1305 static const struct fs_context_operations ceph_context_ops = {
1306 	.free		= ceph_free_fc,
1307 	.parse_param	= ceph_parse_mount_param,
1308 	.get_tree	= ceph_get_tree,
1309 	.reconfigure	= ceph_reconfigure_fc,
1310 };
1311 
1312 /*
1313  * Set up the filesystem mount context.
1314  */
1315 static int ceph_init_fs_context(struct fs_context *fc)
1316 {
1317 	struct ceph_parse_opts_ctx *pctx;
1318 	struct ceph_mount_options *fsopt;
1319 
1320 	pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1321 	if (!pctx)
1322 		return -ENOMEM;
1323 
1324 	pctx->copts = ceph_alloc_options();
1325 	if (!pctx->copts)
1326 		goto nomem;
1327 
1328 	pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1329 	if (!pctx->opts)
1330 		goto nomem;
1331 
1332 	fsopt = pctx->opts;
1333 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1334 
1335 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1336 	fsopt->rsize = CEPH_MAX_READ_SIZE;
1337 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
1338 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1339 	if (!fsopt->snapdir_name)
1340 		goto nomem;
1341 
1342 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1343 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1344 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1345 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1346 	fsopt->congestion_kb = default_congestion_kb();
1347 
1348 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1349 	fc->sb_flags |= SB_POSIXACL;
1350 #endif
1351 
1352 	fc->fs_private = pctx;
1353 	fc->ops = &ceph_context_ops;
1354 	return 0;
1355 
1356 nomem:
1357 	destroy_mount_options(pctx->opts);
1358 	ceph_destroy_options(pctx->copts);
1359 	kfree(pctx);
1360 	return -ENOMEM;
1361 }
1362 
1363 static void ceph_kill_sb(struct super_block *s)
1364 {
1365 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1366 
1367 	dout("kill_sb %p\n", s);
1368 
1369 	ceph_mdsc_pre_umount(fsc->mdsc);
1370 	flush_fs_workqueues(fsc);
1371 
1372 	kill_anon_super(s);
1373 
1374 	fsc->client->extra_mon_dispatch = NULL;
1375 	ceph_fs_debugfs_cleanup(fsc);
1376 
1377 	ceph_fscache_unregister_fs(fsc);
1378 
1379 	destroy_fs_client(fsc);
1380 }
1381 
1382 static struct file_system_type ceph_fs_type = {
1383 	.owner		= THIS_MODULE,
1384 	.name		= "ceph",
1385 	.init_fs_context = ceph_init_fs_context,
1386 	.kill_sb	= ceph_kill_sb,
1387 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1388 };
1389 MODULE_ALIAS_FS("ceph");
1390 
1391 int ceph_force_reconnect(struct super_block *sb)
1392 {
1393 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1394 	int err = 0;
1395 
1396 	fsc->mount_state = CEPH_MOUNT_RECOVER;
1397 	__ceph_umount_begin(fsc);
1398 
1399 	/* Make sure all page caches get invalidated.
1400 	 * see remove_session_caps_cb() */
1401 	flush_workqueue(fsc->inode_wq);
1402 
1403 	/* In case that we were blocklisted. This also reset
1404 	 * all mon/osd connections */
1405 	ceph_reset_client_addr(fsc->client);
1406 
1407 	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1408 
1409 	fsc->blocklisted = false;
1410 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1411 
1412 	if (sb->s_root) {
1413 		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1414 					CEPH_STAT_CAP_INODE, true);
1415 	}
1416 	return err;
1417 }
1418 
1419 static int __init init_ceph(void)
1420 {
1421 	int ret = init_caches();
1422 	if (ret)
1423 		goto out;
1424 
1425 	ceph_flock_init();
1426 	ret = register_filesystem(&ceph_fs_type);
1427 	if (ret)
1428 		goto out_caches;
1429 
1430 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1431 
1432 	return 0;
1433 
1434 out_caches:
1435 	destroy_caches();
1436 out:
1437 	return ret;
1438 }
1439 
1440 static void __exit exit_ceph(void)
1441 {
1442 	dout("exit_ceph\n");
1443 	unregister_filesystem(&ceph_fs_type);
1444 	destroy_caches();
1445 }
1446 
1447 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1448 {
1449 	struct ceph_fs_client *fsc;
1450 	int ret;
1451 
1452 	ret = param_set_bool(val, kp);
1453 	if (ret) {
1454 		pr_err("Failed to parse sending metrics switch value '%s'\n",
1455 		       val);
1456 		return ret;
1457 	} else if (!disable_send_metrics) {
1458 		// wake up all the mds clients
1459 		spin_lock(&ceph_fsc_lock);
1460 		list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1461 			metric_schedule_delayed(&fsc->mdsc->metric);
1462 		}
1463 		spin_unlock(&ceph_fsc_lock);
1464 	}
1465 
1466 	return 0;
1467 }
1468 
1469 static const struct kernel_param_ops param_ops_metrics = {
1470 	.set = param_set_metrics,
1471 	.get = param_get_bool,
1472 };
1473 
1474 bool disable_send_metrics = false;
1475 module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1476 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1477 
1478 /* for both v1 and v2 syntax */
1479 static bool mount_support = true;
1480 static const struct kernel_param_ops param_ops_mount_syntax = {
1481 	.get = param_get_bool,
1482 };
1483 module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1484 module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1485 
1486 module_init(init_ceph);
1487 module_exit(exit_ceph);
1488 
1489 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1490 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1491 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1492 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1493 MODULE_LICENSE("GPL");
1494