xref: /openbmc/linux/fs/ceph/super.c (revision 227a0655)
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 
806 	err = -ENOMEM;
807 	/*
808 	 * The number of concurrent works can be high but they don't need
809 	 * to be processed in parallel, limit concurrency.
810 	 */
811 	fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
812 	if (!fsc->inode_wq)
813 		goto fail_client;
814 	fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
815 	if (!fsc->cap_wq)
816 		goto fail_inode_wq;
817 
818 	spin_lock(&ceph_fsc_lock);
819 	list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
820 	spin_unlock(&ceph_fsc_lock);
821 
822 	return fsc;
823 
824 fail_inode_wq:
825 	destroy_workqueue(fsc->inode_wq);
826 fail_client:
827 	ceph_destroy_client(fsc->client);
828 fail:
829 	kfree(fsc);
830 	if (opt)
831 		ceph_destroy_options(opt);
832 	destroy_mount_options(fsopt);
833 	return ERR_PTR(err);
834 }
835 
836 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
837 {
838 	flush_workqueue(fsc->inode_wq);
839 	flush_workqueue(fsc->cap_wq);
840 }
841 
842 static void destroy_fs_client(struct ceph_fs_client *fsc)
843 {
844 	dout("destroy_fs_client %p\n", fsc);
845 
846 	spin_lock(&ceph_fsc_lock);
847 	list_del(&fsc->metric_wakeup);
848 	spin_unlock(&ceph_fsc_lock);
849 
850 	ceph_mdsc_destroy(fsc);
851 	destroy_workqueue(fsc->inode_wq);
852 	destroy_workqueue(fsc->cap_wq);
853 
854 	destroy_mount_options(fsc->mount_options);
855 
856 	ceph_destroy_client(fsc->client);
857 
858 	kfree(fsc);
859 	dout("destroy_fs_client %p done\n", fsc);
860 }
861 
862 /*
863  * caches
864  */
865 struct kmem_cache *ceph_inode_cachep;
866 struct kmem_cache *ceph_cap_cachep;
867 struct kmem_cache *ceph_cap_flush_cachep;
868 struct kmem_cache *ceph_dentry_cachep;
869 struct kmem_cache *ceph_file_cachep;
870 struct kmem_cache *ceph_dir_file_cachep;
871 struct kmem_cache *ceph_mds_request_cachep;
872 mempool_t *ceph_wb_pagevec_pool;
873 
874 static void ceph_inode_init_once(void *foo)
875 {
876 	struct ceph_inode_info *ci = foo;
877 	inode_init_once(&ci->vfs_inode);
878 }
879 
880 static int __init init_caches(void)
881 {
882 	int error = -ENOMEM;
883 
884 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
885 				      sizeof(struct ceph_inode_info),
886 				      __alignof__(struct ceph_inode_info),
887 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
888 				      SLAB_ACCOUNT, ceph_inode_init_once);
889 	if (!ceph_inode_cachep)
890 		return -ENOMEM;
891 
892 	ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
893 	if (!ceph_cap_cachep)
894 		goto bad_cap;
895 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
896 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
897 	if (!ceph_cap_flush_cachep)
898 		goto bad_cap_flush;
899 
900 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
901 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
902 	if (!ceph_dentry_cachep)
903 		goto bad_dentry;
904 
905 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
906 	if (!ceph_file_cachep)
907 		goto bad_file;
908 
909 	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
910 	if (!ceph_dir_file_cachep)
911 		goto bad_dir_file;
912 
913 	ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, SLAB_MEM_SPREAD);
914 	if (!ceph_mds_request_cachep)
915 		goto bad_mds_req;
916 
917 	ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10, CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT);
918 	if (!ceph_wb_pagevec_pool)
919 		goto bad_pagevec_pool;
920 
921 	return 0;
922 
923 bad_pagevec_pool:
924 	kmem_cache_destroy(ceph_mds_request_cachep);
925 bad_mds_req:
926 	kmem_cache_destroy(ceph_dir_file_cachep);
927 bad_dir_file:
928 	kmem_cache_destroy(ceph_file_cachep);
929 bad_file:
930 	kmem_cache_destroy(ceph_dentry_cachep);
931 bad_dentry:
932 	kmem_cache_destroy(ceph_cap_flush_cachep);
933 bad_cap_flush:
934 	kmem_cache_destroy(ceph_cap_cachep);
935 bad_cap:
936 	kmem_cache_destroy(ceph_inode_cachep);
937 	return error;
938 }
939 
940 static void destroy_caches(void)
941 {
942 	/*
943 	 * Make sure all delayed rcu free inodes are flushed before we
944 	 * destroy cache.
945 	 */
946 	rcu_barrier();
947 
948 	kmem_cache_destroy(ceph_inode_cachep);
949 	kmem_cache_destroy(ceph_cap_cachep);
950 	kmem_cache_destroy(ceph_cap_flush_cachep);
951 	kmem_cache_destroy(ceph_dentry_cachep);
952 	kmem_cache_destroy(ceph_file_cachep);
953 	kmem_cache_destroy(ceph_dir_file_cachep);
954 	kmem_cache_destroy(ceph_mds_request_cachep);
955 	mempool_destroy(ceph_wb_pagevec_pool);
956 }
957 
958 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
959 {
960 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
961 	ceph_mdsc_force_umount(fsc->mdsc);
962 	fsc->filp_gen++; // invalidate open files
963 }
964 
965 /*
966  * ceph_umount_begin - initiate forced umount.  Tear down the
967  * mount, skipping steps that may hang while waiting for server(s).
968  */
969 void ceph_umount_begin(struct super_block *sb)
970 {
971 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
972 
973 	dout("ceph_umount_begin - starting forced umount\n");
974 	if (!fsc)
975 		return;
976 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
977 	__ceph_umount_begin(fsc);
978 }
979 
980 static const struct super_operations ceph_super_ops = {
981 	.alloc_inode	= ceph_alloc_inode,
982 	.free_inode	= ceph_free_inode,
983 	.write_inode    = ceph_write_inode,
984 	.drop_inode	= generic_delete_inode,
985 	.evict_inode	= ceph_evict_inode,
986 	.sync_fs        = ceph_sync_fs,
987 	.put_super	= ceph_put_super,
988 	.show_options   = ceph_show_options,
989 	.statfs		= ceph_statfs,
990 	.umount_begin   = ceph_umount_begin,
991 };
992 
993 /*
994  * Bootstrap mount by opening the root directory.  Note the mount
995  * @started time from caller, and time out if this takes too long.
996  */
997 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
998 				       const char *path,
999 				       unsigned long started)
1000 {
1001 	struct ceph_mds_client *mdsc = fsc->mdsc;
1002 	struct ceph_mds_request *req = NULL;
1003 	int err;
1004 	struct dentry *root;
1005 
1006 	/* open dir */
1007 	dout("open_root_inode opening '%s'\n", path);
1008 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1009 	if (IS_ERR(req))
1010 		return ERR_CAST(req);
1011 	req->r_path1 = kstrdup(path, GFP_NOFS);
1012 	if (!req->r_path1) {
1013 		root = ERR_PTR(-ENOMEM);
1014 		goto out;
1015 	}
1016 
1017 	req->r_ino1.ino = CEPH_INO_ROOT;
1018 	req->r_ino1.snap = CEPH_NOSNAP;
1019 	req->r_started = started;
1020 	req->r_timeout = fsc->client->options->mount_timeout;
1021 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1022 	req->r_num_caps = 2;
1023 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1024 	if (err == 0) {
1025 		struct inode *inode = req->r_target_inode;
1026 		req->r_target_inode = NULL;
1027 		dout("open_root_inode success\n");
1028 		root = d_make_root(inode);
1029 		if (!root) {
1030 			root = ERR_PTR(-ENOMEM);
1031 			goto out;
1032 		}
1033 		dout("open_root_inode success, root dentry is %p\n", root);
1034 	} else {
1035 		root = ERR_PTR(err);
1036 	}
1037 out:
1038 	ceph_mdsc_put_request(req);
1039 	return root;
1040 }
1041 
1042 /*
1043  * mount: join the ceph cluster, and open root directory.
1044  */
1045 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1046 				      struct fs_context *fc)
1047 {
1048 	int err;
1049 	unsigned long started = jiffies;  /* note the start time */
1050 	struct dentry *root;
1051 
1052 	dout("mount start %p\n", fsc);
1053 	mutex_lock(&fsc->client->mount_mutex);
1054 
1055 	if (!fsc->sb->s_root) {
1056 		const char *path = fsc->mount_options->server_path ?
1057 				     fsc->mount_options->server_path + 1 : "";
1058 
1059 		err = __ceph_open_session(fsc->client, started);
1060 		if (err < 0)
1061 			goto out;
1062 
1063 		/* setup fscache */
1064 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1065 			err = ceph_fscache_register_fs(fsc, fc);
1066 			if (err < 0)
1067 				goto out;
1068 		}
1069 
1070 		dout("mount opening path '%s'\n", path);
1071 
1072 		ceph_fs_debugfs_init(fsc);
1073 
1074 		root = open_root_dentry(fsc, path, started);
1075 		if (IS_ERR(root)) {
1076 			err = PTR_ERR(root);
1077 			goto out;
1078 		}
1079 		fsc->sb->s_root = dget(root);
1080 	} else {
1081 		root = dget(fsc->sb->s_root);
1082 	}
1083 
1084 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1085 	dout("mount success\n");
1086 	mutex_unlock(&fsc->client->mount_mutex);
1087 	return root;
1088 
1089 out:
1090 	mutex_unlock(&fsc->client->mount_mutex);
1091 	return ERR_PTR(err);
1092 }
1093 
1094 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1095 {
1096 	struct ceph_fs_client *fsc = s->s_fs_info;
1097 	int ret;
1098 
1099 	dout("set_super %p\n", s);
1100 
1101 	s->s_maxbytes = MAX_LFS_FILESIZE;
1102 
1103 	s->s_xattr = ceph_xattr_handlers;
1104 	fsc->sb = s;
1105 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1106 
1107 	s->s_op = &ceph_super_ops;
1108 	s->s_d_op = &ceph_dentry_ops;
1109 	s->s_export_op = &ceph_export_ops;
1110 
1111 	s->s_time_gran = 1;
1112 	s->s_time_min = 0;
1113 	s->s_time_max = U32_MAX;
1114 
1115 	ret = set_anon_super_fc(s, fc);
1116 	if (ret != 0)
1117 		fsc->sb = NULL;
1118 	return ret;
1119 }
1120 
1121 /*
1122  * share superblock if same fs AND options
1123  */
1124 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1125 {
1126 	struct ceph_fs_client *new = fc->s_fs_info;
1127 	struct ceph_mount_options *fsopt = new->mount_options;
1128 	struct ceph_options *opt = new->client->options;
1129 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1130 
1131 	dout("ceph_compare_super %p\n", sb);
1132 
1133 	if (compare_mount_options(fsopt, opt, fsc)) {
1134 		dout("monitor(s)/mount options don't match\n");
1135 		return 0;
1136 	}
1137 	if ((opt->flags & CEPH_OPT_FSID) &&
1138 	    ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1139 		dout("fsid doesn't match\n");
1140 		return 0;
1141 	}
1142 	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1143 		dout("flags differ\n");
1144 		return 0;
1145 	}
1146 
1147 	if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1148 		dout("client is blocklisted (and CLEANRECOVER is not set)\n");
1149 		return 0;
1150 	}
1151 
1152 	if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1153 		dout("client has been forcibly unmounted\n");
1154 		return 0;
1155 	}
1156 
1157 	return 1;
1158 }
1159 
1160 /*
1161  * construct our own bdi so we can control readahead, etc.
1162  */
1163 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1164 
1165 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1166 {
1167 	int err;
1168 
1169 	err = super_setup_bdi_name(sb, "ceph-%ld",
1170 				   atomic_long_inc_return(&bdi_seq));
1171 	if (err)
1172 		return err;
1173 
1174 	/* set ra_pages based on rasize mount option? */
1175 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1176 
1177 	/* set io_pages based on max osd read size */
1178 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1179 
1180 	return 0;
1181 }
1182 
1183 static int ceph_get_tree(struct fs_context *fc)
1184 {
1185 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1186 	struct ceph_mount_options *fsopt = pctx->opts;
1187 	struct super_block *sb;
1188 	struct ceph_fs_client *fsc;
1189 	struct dentry *res;
1190 	int (*compare_super)(struct super_block *, struct fs_context *) =
1191 		ceph_compare_super;
1192 	int err;
1193 
1194 	dout("ceph_get_tree\n");
1195 
1196 	if (!fc->source)
1197 		return invalfc(fc, "No source");
1198 	if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1199 		return invalfc(fc, "No monitor address");
1200 
1201 	/* create client (which we may/may not use) */
1202 	fsc = create_fs_client(pctx->opts, pctx->copts);
1203 	pctx->opts = NULL;
1204 	pctx->copts = NULL;
1205 	if (IS_ERR(fsc)) {
1206 		err = PTR_ERR(fsc);
1207 		goto out_final;
1208 	}
1209 
1210 	err = ceph_mdsc_init(fsc);
1211 	if (err < 0)
1212 		goto out;
1213 
1214 	if (ceph_test_opt(fsc->client, NOSHARE))
1215 		compare_super = NULL;
1216 
1217 	fc->s_fs_info = fsc;
1218 	sb = sget_fc(fc, compare_super, ceph_set_super);
1219 	fc->s_fs_info = NULL;
1220 	if (IS_ERR(sb)) {
1221 		err = PTR_ERR(sb);
1222 		goto out;
1223 	}
1224 
1225 	if (ceph_sb_to_client(sb) != fsc) {
1226 		destroy_fs_client(fsc);
1227 		fsc = ceph_sb_to_client(sb);
1228 		dout("get_sb got existing client %p\n", fsc);
1229 	} else {
1230 		dout("get_sb using new client %p\n", fsc);
1231 		err = ceph_setup_bdi(sb, fsc);
1232 		if (err < 0)
1233 			goto out_splat;
1234 	}
1235 
1236 	res = ceph_real_mount(fsc, fc);
1237 	if (IS_ERR(res)) {
1238 		err = PTR_ERR(res);
1239 		goto out_splat;
1240 	}
1241 	dout("root %p inode %p ino %llx.%llx\n", res,
1242 	     d_inode(res), ceph_vinop(d_inode(res)));
1243 	fc->root = fsc->sb->s_root;
1244 	return 0;
1245 
1246 out_splat:
1247 	if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1248 		pr_info("No mds server is up or the cluster is laggy\n");
1249 		err = -EHOSTUNREACH;
1250 	}
1251 
1252 	ceph_mdsc_close_sessions(fsc->mdsc);
1253 	deactivate_locked_super(sb);
1254 	goto out_final;
1255 
1256 out:
1257 	destroy_fs_client(fsc);
1258 out_final:
1259 	dout("ceph_get_tree fail %d\n", err);
1260 	return err;
1261 }
1262 
1263 static void ceph_free_fc(struct fs_context *fc)
1264 {
1265 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1266 
1267 	if (pctx) {
1268 		destroy_mount_options(pctx->opts);
1269 		ceph_destroy_options(pctx->copts);
1270 		kfree(pctx);
1271 	}
1272 }
1273 
1274 static int ceph_reconfigure_fc(struct fs_context *fc)
1275 {
1276 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1277 	struct ceph_mount_options *fsopt = pctx->opts;
1278 	struct ceph_fs_client *fsc = ceph_sb_to_client(fc->root->d_sb);
1279 
1280 	if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1281 		ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1282 	else
1283 		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1284 
1285 	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1286 		kfree(fsc->mount_options->mon_addr);
1287 		fsc->mount_options->mon_addr = fsopt->mon_addr;
1288 		fsopt->mon_addr = NULL;
1289 		pr_notice("ceph: monitor addresses recorded, but not used for reconnection");
1290 	}
1291 
1292 	sync_filesystem(fc->root->d_sb);
1293 	return 0;
1294 }
1295 
1296 static const struct fs_context_operations ceph_context_ops = {
1297 	.free		= ceph_free_fc,
1298 	.parse_param	= ceph_parse_mount_param,
1299 	.get_tree	= ceph_get_tree,
1300 	.reconfigure	= ceph_reconfigure_fc,
1301 };
1302 
1303 /*
1304  * Set up the filesystem mount context.
1305  */
1306 static int ceph_init_fs_context(struct fs_context *fc)
1307 {
1308 	struct ceph_parse_opts_ctx *pctx;
1309 	struct ceph_mount_options *fsopt;
1310 
1311 	pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1312 	if (!pctx)
1313 		return -ENOMEM;
1314 
1315 	pctx->copts = ceph_alloc_options();
1316 	if (!pctx->copts)
1317 		goto nomem;
1318 
1319 	pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1320 	if (!pctx->opts)
1321 		goto nomem;
1322 
1323 	fsopt = pctx->opts;
1324 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1325 
1326 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1327 	fsopt->rsize = CEPH_MAX_READ_SIZE;
1328 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
1329 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1330 	if (!fsopt->snapdir_name)
1331 		goto nomem;
1332 
1333 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1334 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1335 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1336 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1337 	fsopt->congestion_kb = default_congestion_kb();
1338 
1339 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1340 	fc->sb_flags |= SB_POSIXACL;
1341 #endif
1342 
1343 	fc->fs_private = pctx;
1344 	fc->ops = &ceph_context_ops;
1345 	return 0;
1346 
1347 nomem:
1348 	destroy_mount_options(pctx->opts);
1349 	ceph_destroy_options(pctx->copts);
1350 	kfree(pctx);
1351 	return -ENOMEM;
1352 }
1353 
1354 static void ceph_kill_sb(struct super_block *s)
1355 {
1356 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1357 
1358 	dout("kill_sb %p\n", s);
1359 
1360 	ceph_mdsc_pre_umount(fsc->mdsc);
1361 	flush_fs_workqueues(fsc);
1362 
1363 	kill_anon_super(s);
1364 
1365 	fsc->client->extra_mon_dispatch = NULL;
1366 	ceph_fs_debugfs_cleanup(fsc);
1367 
1368 	ceph_fscache_unregister_fs(fsc);
1369 
1370 	destroy_fs_client(fsc);
1371 }
1372 
1373 static struct file_system_type ceph_fs_type = {
1374 	.owner		= THIS_MODULE,
1375 	.name		= "ceph",
1376 	.init_fs_context = ceph_init_fs_context,
1377 	.kill_sb	= ceph_kill_sb,
1378 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1379 };
1380 MODULE_ALIAS_FS("ceph");
1381 
1382 int ceph_force_reconnect(struct super_block *sb)
1383 {
1384 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1385 	int err = 0;
1386 
1387 	fsc->mount_state = CEPH_MOUNT_RECOVER;
1388 	__ceph_umount_begin(fsc);
1389 
1390 	/* Make sure all page caches get invalidated.
1391 	 * see remove_session_caps_cb() */
1392 	flush_workqueue(fsc->inode_wq);
1393 
1394 	/* In case that we were blocklisted. This also reset
1395 	 * all mon/osd connections */
1396 	ceph_reset_client_addr(fsc->client);
1397 
1398 	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1399 
1400 	fsc->blocklisted = false;
1401 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1402 
1403 	if (sb->s_root) {
1404 		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1405 					CEPH_STAT_CAP_INODE, true);
1406 	}
1407 	return err;
1408 }
1409 
1410 static int __init init_ceph(void)
1411 {
1412 	int ret = init_caches();
1413 	if (ret)
1414 		goto out;
1415 
1416 	ceph_flock_init();
1417 	ret = register_filesystem(&ceph_fs_type);
1418 	if (ret)
1419 		goto out_caches;
1420 
1421 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1422 
1423 	return 0;
1424 
1425 out_caches:
1426 	destroy_caches();
1427 out:
1428 	return ret;
1429 }
1430 
1431 static void __exit exit_ceph(void)
1432 {
1433 	dout("exit_ceph\n");
1434 	unregister_filesystem(&ceph_fs_type);
1435 	destroy_caches();
1436 }
1437 
1438 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1439 {
1440 	struct ceph_fs_client *fsc;
1441 	int ret;
1442 
1443 	ret = param_set_bool(val, kp);
1444 	if (ret) {
1445 		pr_err("Failed to parse sending metrics switch value '%s'\n",
1446 		       val);
1447 		return ret;
1448 	} else if (!disable_send_metrics) {
1449 		// wake up all the mds clients
1450 		spin_lock(&ceph_fsc_lock);
1451 		list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1452 			metric_schedule_delayed(&fsc->mdsc->metric);
1453 		}
1454 		spin_unlock(&ceph_fsc_lock);
1455 	}
1456 
1457 	return 0;
1458 }
1459 
1460 static const struct kernel_param_ops param_ops_metrics = {
1461 	.set = param_set_metrics,
1462 	.get = param_get_bool,
1463 };
1464 
1465 bool disable_send_metrics = false;
1466 module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1467 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1468 
1469 /* for both v1 and v2 syntax */
1470 static bool mount_support = true;
1471 static const struct kernel_param_ops param_ops_mount_syntax = {
1472 	.get = param_get_bool,
1473 };
1474 module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1475 module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1476 
1477 module_init(init_ceph);
1478 module_exit(exit_ceph);
1479 
1480 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1481 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1482 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1483 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1484 MODULE_LICENSE("GPL");
1485