xref: /linux/fs/ceph/super.c (revision a36e9f5cfe9eb3a1dce8769c7058251c42705357)
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 #include "crypto.h"
24 
25 #include <linux/ceph/ceph_features.h>
26 #include <linux/ceph/decode.h>
27 #include <linux/ceph/mon_client.h>
28 #include <linux/ceph/auth.h>
29 #include <linux/ceph/debugfs.h>
30 
31 #include <uapi/linux/magic.h>
32 
33 static DEFINE_SPINLOCK(ceph_fsc_lock);
34 static LIST_HEAD(ceph_fsc_list);
35 
36 /*
37  * Ceph superblock operations
38  *
39  * Handle the basics of mounting, unmounting.
40  */
41 
42 /*
43  * super ops
44  */
45 static void ceph_put_super(struct super_block *s)
46 {
47 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
48 
49 	doutc(fsc->client, "begin\n");
50 	ceph_fscrypt_free_dummy_policy(fsc);
51 	ceph_mdsc_close_sessions(fsc->mdsc);
52 	doutc(fsc->client, "done\n");
53 }
54 
55 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
56 {
57 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(d_inode(dentry));
58 	struct ceph_mon_client *monc = &fsc->client->monc;
59 	struct ceph_statfs st;
60 	int i, err;
61 	u64 data_pool;
62 
63 	doutc(fsc->client, "begin\n");
64 	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
65 		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
66 	} else {
67 		data_pool = CEPH_NOPOOL;
68 	}
69 
70 	err = ceph_monc_do_statfs(monc, data_pool, &st);
71 	if (err < 0)
72 		return err;
73 
74 	/* fill in kstatfs */
75 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
76 
77 	/*
78 	 * Express utilization in terms of large blocks to avoid
79 	 * overflow on 32-bit machines.
80 	 */
81 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
82 
83 	/*
84 	 * By default use root quota for stats; fallback to overall filesystem
85 	 * usage if using 'noquotadf' mount option or if the root dir doesn't
86 	 * have max_bytes quota set.
87 	 */
88 	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
89 	    !ceph_quota_update_statfs(fsc, buf)) {
90 		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
91 		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
92 		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
93 	}
94 
95 	/*
96 	 * NOTE: for the time being, we make bsize == frsize to humor
97 	 * not-yet-ancient versions of glibc that are broken.
98 	 * Someday, we will probably want to report a real block
99 	 * size...  whatever that may mean for a network file system!
100 	 */
101 	buf->f_bsize = buf->f_frsize;
102 
103 	buf->f_files = le64_to_cpu(st.num_objects);
104 	buf->f_ffree = -1;
105 	buf->f_namelen = NAME_MAX;
106 
107 	/* Must convert the fsid, for consistent values across arches */
108 	buf->f_fsid.val[0] = 0;
109 	mutex_lock(&monc->mutex);
110 	for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
111 		buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
112 	mutex_unlock(&monc->mutex);
113 
114 	/* fold the fs_cluster_id into the upper bits */
115 	buf->f_fsid.val[1] = monc->fs_cluster_id;
116 
117 	doutc(fsc->client, "done\n");
118 	return 0;
119 }
120 
121 static int ceph_sync_fs(struct super_block *sb, int wait)
122 {
123 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
124 	struct ceph_client *cl = fsc->client;
125 
126 	if (!wait) {
127 		doutc(cl, "(non-blocking)\n");
128 		ceph_flush_dirty_caps(fsc->mdsc);
129 		doutc(cl, "(non-blocking) done\n");
130 		return 0;
131 	}
132 
133 	doutc(cl, "(blocking)\n");
134 	ceph_osdc_sync(&fsc->client->osdc);
135 	ceph_mdsc_sync(fsc->mdsc);
136 	doutc(cl, "(blocking) done\n");
137 	return 0;
138 }
139 
140 /*
141  * mount options
142  */
143 enum {
144 	Opt_wsize,
145 	Opt_rsize,
146 	Opt_rasize,
147 	Opt_caps_wanted_delay_min,
148 	Opt_caps_wanted_delay_max,
149 	Opt_caps_max,
150 	Opt_readdir_max_entries,
151 	Opt_readdir_max_bytes,
152 	Opt_congestion_kb,
153 	/* int args above */
154 	Opt_snapdirname,
155 	Opt_mds_namespace,
156 	Opt_recover_session,
157 	Opt_source,
158 	Opt_mon_addr,
159 	Opt_test_dummy_encryption,
160 	/* string args above */
161 	Opt_dirstat,
162 	Opt_rbytes,
163 	Opt_asyncreaddir,
164 	Opt_dcache,
165 	Opt_ino32,
166 	Opt_fscache,
167 	Opt_poolperm,
168 	Opt_require_active_mds,
169 	Opt_acl,
170 	Opt_quotadf,
171 	Opt_copyfrom,
172 	Opt_wsync,
173 	Opt_pagecache,
174 	Opt_sparseread,
175 };
176 
177 enum ceph_recover_session_mode {
178 	ceph_recover_session_no,
179 	ceph_recover_session_clean
180 };
181 
182 static const struct constant_table ceph_param_recover[] = {
183 	{ "no",		ceph_recover_session_no },
184 	{ "clean",	ceph_recover_session_clean },
185 	{}
186 };
187 
188 static const struct fs_parameter_spec ceph_mount_parameters[] = {
189 	fsparam_flag_no ("acl",				Opt_acl),
190 	fsparam_flag_no ("asyncreaddir",		Opt_asyncreaddir),
191 	fsparam_s32	("caps_max",			Opt_caps_max),
192 	fsparam_u32	("caps_wanted_delay_max",	Opt_caps_wanted_delay_max),
193 	fsparam_u32	("caps_wanted_delay_min",	Opt_caps_wanted_delay_min),
194 	fsparam_u32	("write_congestion_kb",		Opt_congestion_kb),
195 	fsparam_flag_no ("copyfrom",			Opt_copyfrom),
196 	fsparam_flag_no ("dcache",			Opt_dcache),
197 	fsparam_flag_no ("dirstat",			Opt_dirstat),
198 	fsparam_flag_no	("fsc",				Opt_fscache), // fsc|nofsc
199 	fsparam_string	("fsc",				Opt_fscache), // fsc=...
200 	fsparam_flag_no ("ino32",			Opt_ino32),
201 	fsparam_string	("mds_namespace",		Opt_mds_namespace),
202 	fsparam_string	("mon_addr",			Opt_mon_addr),
203 	fsparam_flag_no ("poolperm",			Opt_poolperm),
204 	fsparam_flag_no ("quotadf",			Opt_quotadf),
205 	fsparam_u32	("rasize",			Opt_rasize),
206 	fsparam_flag_no ("rbytes",			Opt_rbytes),
207 	fsparam_u32	("readdir_max_bytes",		Opt_readdir_max_bytes),
208 	fsparam_u32	("readdir_max_entries",		Opt_readdir_max_entries),
209 	fsparam_enum	("recover_session",		Opt_recover_session, ceph_param_recover),
210 	fsparam_flag_no ("require_active_mds",		Opt_require_active_mds),
211 	fsparam_u32	("rsize",			Opt_rsize),
212 	fsparam_string	("snapdirname",			Opt_snapdirname),
213 	fsparam_string	("source",			Opt_source),
214 	fsparam_flag	("test_dummy_encryption",	Opt_test_dummy_encryption),
215 	fsparam_string	("test_dummy_encryption",	Opt_test_dummy_encryption),
216 	fsparam_u32	("wsize",			Opt_wsize),
217 	fsparam_flag_no	("wsync",			Opt_wsync),
218 	fsparam_flag_no	("pagecache",			Opt_pagecache),
219 	fsparam_flag_no	("sparseread",			Opt_sparseread),
220 	{}
221 };
222 
223 struct ceph_parse_opts_ctx {
224 	struct ceph_options		*copts;
225 	struct ceph_mount_options	*opts;
226 };
227 
228 /*
229  * Remove adjacent slashes and then the trailing slash, unless it is
230  * the only remaining character.
231  *
232  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
233  */
234 static void canonicalize_path(char *path)
235 {
236 	int i, j = 0;
237 
238 	for (i = 0; path[i] != '\0'; i++) {
239 		if (path[i] != '/' || j < 1 || path[j - 1] != '/')
240 			path[j++] = path[i];
241 	}
242 
243 	if (j > 1 && path[j - 1] == '/')
244 		j--;
245 	path[j] = '\0';
246 }
247 
248 /*
249  * Check if the mds namespace in ceph_mount_options matches
250  * the passed in namespace string. First time match (when
251  * ->mds_namespace is NULL) is treated specially, since
252  * ->mds_namespace needs to be initialized by the caller.
253  */
254 static int namespace_equals(struct ceph_mount_options *fsopt,
255 			    const char *namespace, size_t len)
256 {
257 	return !(fsopt->mds_namespace &&
258 		 (strlen(fsopt->mds_namespace) != len ||
259 		  strncmp(fsopt->mds_namespace, namespace, len)));
260 }
261 
262 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
263 				 struct fs_context *fc)
264 {
265 	int r;
266 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
267 	struct ceph_mount_options *fsopt = pctx->opts;
268 
269 	if (*dev_name_end != ':')
270 		return invalfc(fc, "separator ':' missing in source");
271 
272 	r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
273 			       pctx->copts, fc->log.log, ',');
274 	if (r)
275 		return r;
276 
277 	fsopt->new_dev_syntax = false;
278 	return 0;
279 }
280 
281 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
282 				 struct fs_context *fc)
283 {
284 	size_t len;
285 	struct ceph_fsid fsid;
286 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
287 	struct ceph_mount_options *fsopt = pctx->opts;
288 	char *fsid_start, *fs_name_start;
289 
290 	if (*dev_name_end != '=') {
291 		dout("separator '=' missing in source");
292 		return -EINVAL;
293 	}
294 
295 	fsid_start = strchr(dev_name, '@');
296 	if (!fsid_start)
297 		return invalfc(fc, "missing cluster fsid");
298 	++fsid_start; /* start of cluster fsid */
299 
300 	fs_name_start = strchr(fsid_start, '.');
301 	if (!fs_name_start)
302 		return invalfc(fc, "missing file system name");
303 
304 	if (ceph_parse_fsid(fsid_start, &fsid))
305 		return invalfc(fc, "Invalid FSID");
306 
307 	++fs_name_start; /* start of file system name */
308 	len = dev_name_end - fs_name_start;
309 
310 	if (!namespace_equals(fsopt, fs_name_start, len))
311 		return invalfc(fc, "Mismatching mds_namespace");
312 	kfree(fsopt->mds_namespace);
313 	fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
314 	if (!fsopt->mds_namespace)
315 		return -ENOMEM;
316 	dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
317 
318 	fsopt->new_dev_syntax = true;
319 	return 0;
320 }
321 
322 /*
323  * Parse the source parameter for new device format. Distinguish the device
324  * spec from the path. Try parsing new device format and fallback to old
325  * format if needed.
326  *
327  * New device syntax will looks like:
328  *     <device_spec>=/<path>
329  * where
330  *     <device_spec> is name@fsid.fsname
331  *     <path> is optional, but if present must begin with '/'
332  * (monitor addresses are passed via mount option)
333  *
334  * Old device syntax is:
335  *     <server_spec>[,<server_spec>...]:[<path>]
336  * where
337  *     <server_spec> is <ip>[:<port>]
338  *     <path> is optional, but if present must begin with '/'
339  */
340 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
341 {
342 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
343 	struct ceph_mount_options *fsopt = pctx->opts;
344 	char *dev_name = param->string, *dev_name_end;
345 	int ret;
346 
347 	dout("'%s'\n", dev_name);
348 	if (!dev_name || !*dev_name)
349 		return invalfc(fc, "Empty source");
350 
351 	dev_name_end = strchr(dev_name, '/');
352 	if (dev_name_end) {
353 		/*
354 		 * The server_path will include the whole chars from userland
355 		 * including the leading '/'.
356 		 */
357 		kfree(fsopt->server_path);
358 		fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
359 		if (!fsopt->server_path)
360 			return -ENOMEM;
361 
362 		canonicalize_path(fsopt->server_path);
363 	} else {
364 		dev_name_end = dev_name + strlen(dev_name);
365 	}
366 
367 	dev_name_end--;		/* back up to separator */
368 	if (dev_name_end < dev_name)
369 		return invalfc(fc, "Path missing in source");
370 
371 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
372 	if (fsopt->server_path)
373 		dout("server path '%s'\n", fsopt->server_path);
374 
375 	dout("trying new device syntax");
376 	ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
377 	if (ret) {
378 		if (ret != -EINVAL)
379 			return ret;
380 		dout("trying old device syntax");
381 		ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
382 		if (ret)
383 			return ret;
384 	}
385 
386 	fc->source = param->string;
387 	param->string = NULL;
388 	return 0;
389 }
390 
391 static int ceph_parse_mon_addr(struct fs_parameter *param,
392 			       struct fs_context *fc)
393 {
394 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
395 	struct ceph_mount_options *fsopt = pctx->opts;
396 
397 	kfree(fsopt->mon_addr);
398 	fsopt->mon_addr = param->string;
399 	param->string = NULL;
400 
401 	return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
402 				  pctx->copts, fc->log.log, '/');
403 }
404 
405 static int ceph_parse_mount_param(struct fs_context *fc,
406 				  struct fs_parameter *param)
407 {
408 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
409 	struct ceph_mount_options *fsopt = pctx->opts;
410 	struct fs_parse_result result;
411 	unsigned int mode;
412 	int token, ret;
413 
414 	ret = ceph_parse_param(param, pctx->copts, fc->log.log);
415 	if (ret != -ENOPARAM)
416 		return ret;
417 
418 	token = fs_parse(fc, ceph_mount_parameters, param, &result);
419 	dout("%s: fs_parse '%s' token %d\n",__func__, param->key, token);
420 	if (token < 0)
421 		return token;
422 
423 	switch (token) {
424 	case Opt_snapdirname:
425 		kfree(fsopt->snapdir_name);
426 		fsopt->snapdir_name = param->string;
427 		param->string = NULL;
428 		break;
429 	case Opt_mds_namespace:
430 		if (!namespace_equals(fsopt, param->string, strlen(param->string)))
431 			return invalfc(fc, "Mismatching mds_namespace");
432 		kfree(fsopt->mds_namespace);
433 		fsopt->mds_namespace = param->string;
434 		param->string = NULL;
435 		break;
436 	case Opt_recover_session:
437 		mode = result.uint_32;
438 		if (mode == ceph_recover_session_no)
439 			fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
440 		else if (mode == ceph_recover_session_clean)
441 			fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
442 		else
443 			BUG();
444 		break;
445 	case Opt_source:
446 		if (fc->source)
447 			return invalfc(fc, "Multiple sources specified");
448 		return ceph_parse_source(param, fc);
449 	case Opt_mon_addr:
450 		return ceph_parse_mon_addr(param, fc);
451 	case Opt_wsize:
452 		if (result.uint_32 < PAGE_SIZE ||
453 		    result.uint_32 > CEPH_MAX_WRITE_SIZE)
454 			goto out_of_range;
455 		fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
456 		break;
457 	case Opt_rsize:
458 		if (result.uint_32 < PAGE_SIZE ||
459 		    result.uint_32 > CEPH_MAX_READ_SIZE)
460 			goto out_of_range;
461 		fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
462 		break;
463 	case Opt_rasize:
464 		fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
465 		break;
466 	case Opt_caps_wanted_delay_min:
467 		if (result.uint_32 < 1)
468 			goto out_of_range;
469 		fsopt->caps_wanted_delay_min = result.uint_32;
470 		break;
471 	case Opt_caps_wanted_delay_max:
472 		if (result.uint_32 < 1)
473 			goto out_of_range;
474 		fsopt->caps_wanted_delay_max = result.uint_32;
475 		break;
476 	case Opt_caps_max:
477 		if (result.int_32 < 0)
478 			goto out_of_range;
479 		fsopt->caps_max = result.int_32;
480 		break;
481 	case Opt_readdir_max_entries:
482 		if (result.uint_32 < 1)
483 			goto out_of_range;
484 		fsopt->max_readdir = result.uint_32;
485 		break;
486 	case Opt_readdir_max_bytes:
487 		if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
488 			goto out_of_range;
489 		fsopt->max_readdir_bytes = result.uint_32;
490 		break;
491 	case Opt_congestion_kb:
492 		if (result.uint_32 < 1024) /* at least 1M */
493 			goto out_of_range;
494 		fsopt->congestion_kb = result.uint_32;
495 		break;
496 	case Opt_dirstat:
497 		if (!result.negated)
498 			fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
499 		else
500 			fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
501 		break;
502 	case Opt_rbytes:
503 		if (!result.negated)
504 			fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
505 		else
506 			fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
507 		break;
508 	case Opt_asyncreaddir:
509 		if (!result.negated)
510 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
511 		else
512 			fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
513 		break;
514 	case Opt_dcache:
515 		if (!result.negated)
516 			fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
517 		else
518 			fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
519 		break;
520 	case Opt_ino32:
521 		if (!result.negated)
522 			fsopt->flags |= CEPH_MOUNT_OPT_INO32;
523 		else
524 			fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
525 		break;
526 
527 	case Opt_fscache:
528 #ifdef CONFIG_CEPH_FSCACHE
529 		kfree(fsopt->fscache_uniq);
530 		fsopt->fscache_uniq = NULL;
531 		if (result.negated) {
532 			fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
533 		} else {
534 			fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
535 			fsopt->fscache_uniq = param->string;
536 			param->string = NULL;
537 		}
538 		break;
539 #else
540 		return invalfc(fc, "fscache support is disabled");
541 #endif
542 	case Opt_poolperm:
543 		if (!result.negated)
544 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
545 		else
546 			fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
547 		break;
548 	case Opt_require_active_mds:
549 		if (!result.negated)
550 			fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
551 		else
552 			fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
553 		break;
554 	case Opt_quotadf:
555 		if (!result.negated)
556 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
557 		else
558 			fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
559 		break;
560 	case Opt_copyfrom:
561 		if (!result.negated)
562 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
563 		else
564 			fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
565 		break;
566 	case Opt_acl:
567 		if (!result.negated) {
568 #ifdef CONFIG_CEPH_FS_POSIX_ACL
569 			fc->sb_flags |= SB_POSIXACL;
570 #else
571 			return invalfc(fc, "POSIX ACL support is disabled");
572 #endif
573 		} else {
574 			fc->sb_flags &= ~SB_POSIXACL;
575 		}
576 		break;
577 	case Opt_wsync:
578 		if (!result.negated)
579 			fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
580 		else
581 			fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
582 		break;
583 	case Opt_pagecache:
584 		if (result.negated)
585 			fsopt->flags |= CEPH_MOUNT_OPT_NOPAGECACHE;
586 		else
587 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
588 		break;
589 	case Opt_sparseread:
590 		if (result.negated)
591 			fsopt->flags &= ~CEPH_MOUNT_OPT_SPARSEREAD;
592 		else
593 			fsopt->flags |= CEPH_MOUNT_OPT_SPARSEREAD;
594 		break;
595 	case Opt_test_dummy_encryption:
596 #ifdef CONFIG_FS_ENCRYPTION
597 		fscrypt_free_dummy_policy(&fsopt->dummy_enc_policy);
598 		ret = fscrypt_parse_test_dummy_encryption(param,
599 						&fsopt->dummy_enc_policy);
600 		if (ret == -EINVAL) {
601 			warnfc(fc, "Value of option \"%s\" is unrecognized",
602 			       param->key);
603 		} else if (ret == -EEXIST) {
604 			warnfc(fc, "Conflicting test_dummy_encryption options");
605 			ret = -EINVAL;
606 		}
607 #else
608 		warnfc(fc,
609 		       "FS encryption not supported: test_dummy_encryption mount option ignored");
610 #endif
611 		break;
612 	default:
613 		BUG();
614 	}
615 	return 0;
616 
617 out_of_range:
618 	return invalfc(fc, "%s out of range", param->key);
619 }
620 
621 static void destroy_mount_options(struct ceph_mount_options *args)
622 {
623 	dout("destroy_mount_options %p\n", args);
624 	if (!args)
625 		return;
626 
627 	kfree(args->snapdir_name);
628 	kfree(args->mds_namespace);
629 	kfree(args->server_path);
630 	kfree(args->fscache_uniq);
631 	kfree(args->mon_addr);
632 	fscrypt_free_dummy_policy(&args->dummy_enc_policy);
633 	kfree(args);
634 }
635 
636 static int strcmp_null(const char *s1, const char *s2)
637 {
638 	if (!s1 && !s2)
639 		return 0;
640 	if (s1 && !s2)
641 		return -1;
642 	if (!s1 && s2)
643 		return 1;
644 	return strcmp(s1, s2);
645 }
646 
647 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
648 				 struct ceph_options *new_opt,
649 				 struct ceph_fs_client *fsc)
650 {
651 	struct ceph_mount_options *fsopt1 = new_fsopt;
652 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
653 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
654 	int ret;
655 
656 	ret = memcmp(fsopt1, fsopt2, ofs);
657 	if (ret)
658 		return ret;
659 
660 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
661 	if (ret)
662 		return ret;
663 
664 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
665 	if (ret)
666 		return ret;
667 
668 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
669 	if (ret)
670 		return ret;
671 
672 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
673 	if (ret)
674 		return ret;
675 
676 	ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
677 	if (ret)
678 		return ret;
679 
680 	return ceph_compare_options(new_opt, fsc->client);
681 }
682 
683 /**
684  * ceph_show_options - Show mount options in /proc/mounts
685  * @m: seq_file to write to
686  * @root: root of that (sub)tree
687  */
688 static int ceph_show_options(struct seq_file *m, struct dentry *root)
689 {
690 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(root->d_sb);
691 	struct ceph_mount_options *fsopt = fsc->mount_options;
692 	size_t pos;
693 	int ret;
694 
695 	/* a comma between MNT/MS and client options */
696 	seq_putc(m, ',');
697 	pos = m->count;
698 
699 	ret = ceph_print_client_options(m, fsc->client, false);
700 	if (ret)
701 		return ret;
702 
703 	/* retract our comma if no client options */
704 	if (m->count == pos)
705 		m->count--;
706 
707 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
708 		seq_puts(m, ",dirstat");
709 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
710 		seq_puts(m, ",rbytes");
711 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
712 		seq_puts(m, ",noasyncreaddir");
713 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
714 		seq_puts(m, ",nodcache");
715 	if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
716 		seq_puts(m, ",ino32");
717 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
718 		seq_show_option(m, "fsc", fsopt->fscache_uniq);
719 	}
720 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
721 		seq_puts(m, ",nopoolperm");
722 	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
723 		seq_puts(m, ",noquotadf");
724 
725 #ifdef CONFIG_CEPH_FS_POSIX_ACL
726 	if (root->d_sb->s_flags & SB_POSIXACL)
727 		seq_puts(m, ",acl");
728 	else
729 		seq_puts(m, ",noacl");
730 #endif
731 
732 	if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
733 		seq_puts(m, ",copyfrom");
734 
735 	/* dump mds_namespace when old device syntax is in use */
736 	if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
737 		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
738 
739 	if (fsopt->mon_addr)
740 		seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
741 
742 	if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
743 		seq_show_option(m, "recover_session", "clean");
744 
745 	if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
746 		seq_puts(m, ",wsync");
747 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
748 		seq_puts(m, ",nopagecache");
749 	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
750 		seq_puts(m, ",sparseread");
751 
752 	fscrypt_show_test_dummy_encryption(m, ',', root->d_sb);
753 
754 	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
755 		seq_printf(m, ",wsize=%u", fsopt->wsize);
756 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
757 		seq_printf(m, ",rsize=%u", fsopt->rsize);
758 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
759 		seq_printf(m, ",rasize=%u", fsopt->rasize);
760 	if (fsopt->congestion_kb != default_congestion_kb())
761 		seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
762 	if (fsopt->caps_max)
763 		seq_printf(m, ",caps_max=%d", fsopt->caps_max);
764 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
765 		seq_printf(m, ",caps_wanted_delay_min=%u",
766 			 fsopt->caps_wanted_delay_min);
767 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
768 		seq_printf(m, ",caps_wanted_delay_max=%u",
769 			   fsopt->caps_wanted_delay_max);
770 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
771 		seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
772 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
773 		seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
774 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
775 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
776 
777 	return 0;
778 }
779 
780 /*
781  * handle any mon messages the standard library doesn't understand.
782  * return error if we don't either.
783  */
784 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
785 {
786 	struct ceph_fs_client *fsc = client->private;
787 	int type = le16_to_cpu(msg->hdr.type);
788 
789 	switch (type) {
790 	case CEPH_MSG_MDS_MAP:
791 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
792 		return 0;
793 	case CEPH_MSG_FS_MAP_USER:
794 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
795 		return 0;
796 	default:
797 		return -1;
798 	}
799 }
800 
801 /*
802  * create a new fs client
803  *
804  * Success or not, this function consumes @fsopt and @opt.
805  */
806 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
807 					struct ceph_options *opt)
808 {
809 	struct ceph_fs_client *fsc;
810 	int err;
811 
812 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
813 	if (!fsc) {
814 		err = -ENOMEM;
815 		goto fail;
816 	}
817 
818 	fsc->client = ceph_create_client(opt, fsc);
819 	if (IS_ERR(fsc->client)) {
820 		err = PTR_ERR(fsc->client);
821 		goto fail;
822 	}
823 	opt = NULL; /* fsc->client now owns this */
824 
825 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
826 	ceph_set_opt(fsc->client, ABORT_ON_FULL);
827 
828 	if (!fsopt->mds_namespace) {
829 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
830 				   0, true);
831 	} else {
832 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
833 				   0, false);
834 	}
835 
836 	fsc->mount_options = fsopt;
837 
838 	fsc->sb = NULL;
839 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
840 	fsc->filp_gen = 1;
841 	fsc->have_copy_from2 = true;
842 
843 	atomic_long_set(&fsc->writeback_count, 0);
844 	fsc->write_congested = false;
845 
846 	err = -ENOMEM;
847 	/*
848 	 * The number of concurrent works can be high but they don't need
849 	 * to be processed in parallel, limit concurrency.
850 	 */
851 	fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
852 	if (!fsc->inode_wq)
853 		goto fail_client;
854 	fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
855 	if (!fsc->cap_wq)
856 		goto fail_inode_wq;
857 
858 	hash_init(fsc->async_unlink_conflict);
859 	spin_lock_init(&fsc->async_unlink_conflict_lock);
860 
861 	spin_lock(&ceph_fsc_lock);
862 	list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
863 	spin_unlock(&ceph_fsc_lock);
864 
865 	return fsc;
866 
867 fail_inode_wq:
868 	destroy_workqueue(fsc->inode_wq);
869 fail_client:
870 	ceph_destroy_client(fsc->client);
871 fail:
872 	kfree(fsc);
873 	if (opt)
874 		ceph_destroy_options(opt);
875 	destroy_mount_options(fsopt);
876 	return ERR_PTR(err);
877 }
878 
879 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
880 {
881 	flush_workqueue(fsc->inode_wq);
882 	flush_workqueue(fsc->cap_wq);
883 }
884 
885 static void destroy_fs_client(struct ceph_fs_client *fsc)
886 {
887 	doutc(fsc->client, "%p\n", fsc);
888 
889 	spin_lock(&ceph_fsc_lock);
890 	list_del(&fsc->metric_wakeup);
891 	spin_unlock(&ceph_fsc_lock);
892 
893 	ceph_mdsc_destroy(fsc);
894 	destroy_workqueue(fsc->inode_wq);
895 	destroy_workqueue(fsc->cap_wq);
896 
897 	destroy_mount_options(fsc->mount_options);
898 
899 	ceph_destroy_client(fsc->client);
900 
901 	kfree(fsc);
902 	dout("%s: %p done\n", __func__, fsc);
903 }
904 
905 /*
906  * caches
907  */
908 struct kmem_cache *ceph_inode_cachep;
909 struct kmem_cache *ceph_cap_cachep;
910 struct kmem_cache *ceph_cap_snap_cachep;
911 struct kmem_cache *ceph_cap_flush_cachep;
912 struct kmem_cache *ceph_dentry_cachep;
913 struct kmem_cache *ceph_file_cachep;
914 struct kmem_cache *ceph_dir_file_cachep;
915 struct kmem_cache *ceph_mds_request_cachep;
916 mempool_t *ceph_wb_pagevec_pool;
917 
918 static void ceph_inode_init_once(void *foo)
919 {
920 	struct ceph_inode_info *ci = foo;
921 	inode_init_once(&ci->netfs.inode);
922 }
923 
924 static int __init init_caches(void)
925 {
926 	int error = -ENOMEM;
927 
928 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
929 				      sizeof(struct ceph_inode_info),
930 				      __alignof__(struct ceph_inode_info),
931 				      SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
932 				      ceph_inode_init_once);
933 	if (!ceph_inode_cachep)
934 		return -ENOMEM;
935 
936 	ceph_cap_cachep = KMEM_CACHE(ceph_cap, 0);
937 	if (!ceph_cap_cachep)
938 		goto bad_cap;
939 	ceph_cap_snap_cachep = KMEM_CACHE(ceph_cap_snap, 0);
940 	if (!ceph_cap_snap_cachep)
941 		goto bad_cap_snap;
942 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
943 					   SLAB_RECLAIM_ACCOUNT);
944 	if (!ceph_cap_flush_cachep)
945 		goto bad_cap_flush;
946 
947 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
948 					SLAB_RECLAIM_ACCOUNT);
949 	if (!ceph_dentry_cachep)
950 		goto bad_dentry;
951 
952 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, 0);
953 	if (!ceph_file_cachep)
954 		goto bad_file;
955 
956 	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, 0);
957 	if (!ceph_dir_file_cachep)
958 		goto bad_dir_file;
959 
960 	ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, 0);
961 	if (!ceph_mds_request_cachep)
962 		goto bad_mds_req;
963 
964 	ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10,
965 	    (CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT) * sizeof(struct page *));
966 	if (!ceph_wb_pagevec_pool)
967 		goto bad_pagevec_pool;
968 
969 	return 0;
970 
971 bad_pagevec_pool:
972 	kmem_cache_destroy(ceph_mds_request_cachep);
973 bad_mds_req:
974 	kmem_cache_destroy(ceph_dir_file_cachep);
975 bad_dir_file:
976 	kmem_cache_destroy(ceph_file_cachep);
977 bad_file:
978 	kmem_cache_destroy(ceph_dentry_cachep);
979 bad_dentry:
980 	kmem_cache_destroy(ceph_cap_flush_cachep);
981 bad_cap_flush:
982 	kmem_cache_destroy(ceph_cap_snap_cachep);
983 bad_cap_snap:
984 	kmem_cache_destroy(ceph_cap_cachep);
985 bad_cap:
986 	kmem_cache_destroy(ceph_inode_cachep);
987 	return error;
988 }
989 
990 static void destroy_caches(void)
991 {
992 	/*
993 	 * Make sure all delayed rcu free inodes are flushed before we
994 	 * destroy cache.
995 	 */
996 	rcu_barrier();
997 
998 	kmem_cache_destroy(ceph_inode_cachep);
999 	kmem_cache_destroy(ceph_cap_cachep);
1000 	kmem_cache_destroy(ceph_cap_snap_cachep);
1001 	kmem_cache_destroy(ceph_cap_flush_cachep);
1002 	kmem_cache_destroy(ceph_dentry_cachep);
1003 	kmem_cache_destroy(ceph_file_cachep);
1004 	kmem_cache_destroy(ceph_dir_file_cachep);
1005 	kmem_cache_destroy(ceph_mds_request_cachep);
1006 	mempool_destroy(ceph_wb_pagevec_pool);
1007 }
1008 
1009 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
1010 {
1011 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
1012 	ceph_mdsc_force_umount(fsc->mdsc);
1013 	fsc->filp_gen++; // invalidate open files
1014 }
1015 
1016 /*
1017  * ceph_umount_begin - initiate forced umount.  Tear down the
1018  * mount, skipping steps that may hang while waiting for server(s).
1019  */
1020 void ceph_umount_begin(struct super_block *sb)
1021 {
1022 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1023 
1024 	doutc(fsc->client, "starting forced umount\n");
1025 	if (!fsc)
1026 		return;
1027 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
1028 	__ceph_umount_begin(fsc);
1029 }
1030 
1031 static const struct super_operations ceph_super_ops = {
1032 	.alloc_inode	= ceph_alloc_inode,
1033 	.free_inode	= ceph_free_inode,
1034 	.write_inode    = ceph_write_inode,
1035 	.drop_inode	= generic_delete_inode,
1036 	.evict_inode	= ceph_evict_inode,
1037 	.sync_fs        = ceph_sync_fs,
1038 	.put_super	= ceph_put_super,
1039 	.show_options   = ceph_show_options,
1040 	.statfs		= ceph_statfs,
1041 	.umount_begin   = ceph_umount_begin,
1042 };
1043 
1044 /*
1045  * Bootstrap mount by opening the root directory.  Note the mount
1046  * @started time from caller, and time out if this takes too long.
1047  */
1048 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1049 				       const char *path,
1050 				       unsigned long started)
1051 {
1052 	struct ceph_client *cl = fsc->client;
1053 	struct ceph_mds_client *mdsc = fsc->mdsc;
1054 	struct ceph_mds_request *req = NULL;
1055 	int err;
1056 	struct dentry *root;
1057 
1058 	/* open dir */
1059 	doutc(cl, "opening '%s'\n", path);
1060 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1061 	if (IS_ERR(req))
1062 		return ERR_CAST(req);
1063 	req->r_path1 = kstrdup(path, GFP_NOFS);
1064 	if (!req->r_path1) {
1065 		root = ERR_PTR(-ENOMEM);
1066 		goto out;
1067 	}
1068 
1069 	req->r_ino1.ino = CEPH_INO_ROOT;
1070 	req->r_ino1.snap = CEPH_NOSNAP;
1071 	req->r_started = started;
1072 	req->r_timeout = fsc->client->options->mount_timeout;
1073 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1074 	req->r_num_caps = 2;
1075 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1076 	if (err == 0) {
1077 		struct inode *inode = req->r_target_inode;
1078 		req->r_target_inode = NULL;
1079 		doutc(cl, "success\n");
1080 		root = d_make_root(inode);
1081 		if (!root) {
1082 			root = ERR_PTR(-ENOMEM);
1083 			goto out;
1084 		}
1085 		doutc(cl, "success, root dentry is %p\n", root);
1086 	} else {
1087 		root = ERR_PTR(err);
1088 	}
1089 out:
1090 	ceph_mdsc_put_request(req);
1091 	return root;
1092 }
1093 
1094 #ifdef CONFIG_FS_ENCRYPTION
1095 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1096 					    struct fs_context *fc,
1097 					    struct ceph_mount_options *fsopt)
1098 {
1099 	struct ceph_fs_client *fsc = sb->s_fs_info;
1100 
1101 	if (!fscrypt_is_dummy_policy_set(&fsopt->dummy_enc_policy))
1102 		return 0;
1103 
1104 	/* No changing encryption context on remount. */
1105 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE &&
1106 	    !fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1107 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1108 						 &fsc->fsc_dummy_enc_policy))
1109 			return 0;
1110 		errorfc(fc, "Can't set test_dummy_encryption on remount");
1111 		return -EINVAL;
1112 	}
1113 
1114 	/* Also make sure fsopt doesn't contain a conflicting value. */
1115 	if (fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1116 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1117 						 &fsc->fsc_dummy_enc_policy))
1118 			return 0;
1119 		errorfc(fc, "Conflicting test_dummy_encryption options");
1120 		return -EINVAL;
1121 	}
1122 
1123 	fsc->fsc_dummy_enc_policy = fsopt->dummy_enc_policy;
1124 	memset(&fsopt->dummy_enc_policy, 0, sizeof(fsopt->dummy_enc_policy));
1125 
1126 	warnfc(fc, "test_dummy_encryption mode enabled");
1127 	return 0;
1128 }
1129 #else
1130 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1131 					    struct fs_context *fc,
1132 					    struct ceph_mount_options *fsopt)
1133 {
1134 	return 0;
1135 }
1136 #endif
1137 
1138 /*
1139  * mount: join the ceph cluster, and open root directory.
1140  */
1141 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1142 				      struct fs_context *fc)
1143 {
1144 	struct ceph_client *cl = fsc->client;
1145 	int err;
1146 	unsigned long started = jiffies;  /* note the start time */
1147 	struct dentry *root;
1148 
1149 	doutc(cl, "mount start %p\n", fsc);
1150 	mutex_lock(&fsc->client->mount_mutex);
1151 
1152 	if (!fsc->sb->s_root) {
1153 		const char *path = fsc->mount_options->server_path ?
1154 				     fsc->mount_options->server_path + 1 : "";
1155 
1156 		err = __ceph_open_session(fsc->client, started);
1157 		if (err < 0)
1158 			goto out;
1159 
1160 		/* setup fscache */
1161 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1162 			err = ceph_fscache_register_fs(fsc, fc);
1163 			if (err < 0)
1164 				goto out;
1165 		}
1166 
1167 		err = ceph_apply_test_dummy_encryption(fsc->sb, fc,
1168 						       fsc->mount_options);
1169 		if (err)
1170 			goto out;
1171 
1172 		doutc(cl, "mount opening path '%s'\n", path);
1173 
1174 		ceph_fs_debugfs_init(fsc);
1175 
1176 		root = open_root_dentry(fsc, path, started);
1177 		if (IS_ERR(root)) {
1178 			err = PTR_ERR(root);
1179 			goto out;
1180 		}
1181 		fsc->sb->s_root = dget(root);
1182 	} else {
1183 		root = dget(fsc->sb->s_root);
1184 	}
1185 
1186 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1187 	doutc(cl, "mount success\n");
1188 	mutex_unlock(&fsc->client->mount_mutex);
1189 	return root;
1190 
1191 out:
1192 	mutex_unlock(&fsc->client->mount_mutex);
1193 	ceph_fscrypt_free_dummy_policy(fsc);
1194 	return ERR_PTR(err);
1195 }
1196 
1197 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1198 {
1199 	struct ceph_fs_client *fsc = s->s_fs_info;
1200 	struct ceph_client *cl = fsc->client;
1201 	int ret;
1202 
1203 	doutc(cl, "%p\n", s);
1204 
1205 	s->s_maxbytes = MAX_LFS_FILESIZE;
1206 
1207 	s->s_xattr = ceph_xattr_handlers;
1208 	fsc->sb = s;
1209 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1210 
1211 	s->s_op = &ceph_super_ops;
1212 	s->s_d_op = &ceph_dentry_ops;
1213 	s->s_export_op = &ceph_export_ops;
1214 
1215 	s->s_time_gran = 1;
1216 	s->s_time_min = 0;
1217 	s->s_time_max = U32_MAX;
1218 	s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1219 
1220 	ceph_fscrypt_set_ops(s);
1221 
1222 	ret = set_anon_super_fc(s, fc);
1223 	if (ret != 0)
1224 		fsc->sb = NULL;
1225 	return ret;
1226 }
1227 
1228 /*
1229  * share superblock if same fs AND options
1230  */
1231 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1232 {
1233 	struct ceph_fs_client *new = fc->s_fs_info;
1234 	struct ceph_mount_options *fsopt = new->mount_options;
1235 	struct ceph_options *opt = new->client->options;
1236 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1237 	struct ceph_client *cl = fsc->client;
1238 
1239 	doutc(cl, "%p\n", sb);
1240 
1241 	if (compare_mount_options(fsopt, opt, fsc)) {
1242 		doutc(cl, "monitor(s)/mount options don't match\n");
1243 		return 0;
1244 	}
1245 	if ((opt->flags & CEPH_OPT_FSID) &&
1246 	    ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1247 		doutc(cl, "fsid doesn't match\n");
1248 		return 0;
1249 	}
1250 	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1251 		doutc(cl, "flags differ\n");
1252 		return 0;
1253 	}
1254 
1255 	if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1256 		doutc(cl, "client is blocklisted (and CLEANRECOVER is not set)\n");
1257 		return 0;
1258 	}
1259 
1260 	if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1261 		doutc(cl, "client has been forcibly unmounted\n");
1262 		return 0;
1263 	}
1264 
1265 	return 1;
1266 }
1267 
1268 /*
1269  * construct our own bdi so we can control readahead, etc.
1270  */
1271 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1272 
1273 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1274 {
1275 	int err;
1276 
1277 	err = super_setup_bdi_name(sb, "ceph-%ld",
1278 				   atomic_long_inc_return(&bdi_seq));
1279 	if (err)
1280 		return err;
1281 
1282 	/* set ra_pages based on rasize mount option? */
1283 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1284 
1285 	/* set io_pages based on max osd read size */
1286 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1287 
1288 	return 0;
1289 }
1290 
1291 static int ceph_get_tree(struct fs_context *fc)
1292 {
1293 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1294 	struct ceph_mount_options *fsopt = pctx->opts;
1295 	struct super_block *sb;
1296 	struct ceph_fs_client *fsc;
1297 	struct dentry *res;
1298 	int (*compare_super)(struct super_block *, struct fs_context *) =
1299 		ceph_compare_super;
1300 	int err;
1301 
1302 	dout("ceph_get_tree\n");
1303 
1304 	if (!fc->source)
1305 		return invalfc(fc, "No source");
1306 	if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1307 		return invalfc(fc, "No monitor address");
1308 
1309 	/* create client (which we may/may not use) */
1310 	fsc = create_fs_client(pctx->opts, pctx->copts);
1311 	pctx->opts = NULL;
1312 	pctx->copts = NULL;
1313 	if (IS_ERR(fsc)) {
1314 		err = PTR_ERR(fsc);
1315 		goto out_final;
1316 	}
1317 
1318 	err = ceph_mdsc_init(fsc);
1319 	if (err < 0)
1320 		goto out;
1321 
1322 	if (ceph_test_opt(fsc->client, NOSHARE))
1323 		compare_super = NULL;
1324 
1325 	fc->s_fs_info = fsc;
1326 	sb = sget_fc(fc, compare_super, ceph_set_super);
1327 	fc->s_fs_info = NULL;
1328 	if (IS_ERR(sb)) {
1329 		err = PTR_ERR(sb);
1330 		goto out;
1331 	}
1332 
1333 	if (ceph_sb_to_fs_client(sb) != fsc) {
1334 		destroy_fs_client(fsc);
1335 		fsc = ceph_sb_to_fs_client(sb);
1336 		dout("get_sb got existing client %p\n", fsc);
1337 	} else {
1338 		dout("get_sb using new client %p\n", fsc);
1339 		err = ceph_setup_bdi(sb, fsc);
1340 		if (err < 0)
1341 			goto out_splat;
1342 	}
1343 
1344 	res = ceph_real_mount(fsc, fc);
1345 	if (IS_ERR(res)) {
1346 		err = PTR_ERR(res);
1347 		goto out_splat;
1348 	}
1349 
1350 	doutc(fsc->client, "root %p inode %p ino %llx.%llx\n", res,
1351 		    d_inode(res), ceph_vinop(d_inode(res)));
1352 	fc->root = fsc->sb->s_root;
1353 	return 0;
1354 
1355 out_splat:
1356 	if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1357 		pr_info("No mds server is up or the cluster is laggy\n");
1358 		err = -EHOSTUNREACH;
1359 	}
1360 
1361 	ceph_mdsc_close_sessions(fsc->mdsc);
1362 	deactivate_locked_super(sb);
1363 	goto out_final;
1364 
1365 out:
1366 	destroy_fs_client(fsc);
1367 out_final:
1368 	dout("ceph_get_tree fail %d\n", err);
1369 	return err;
1370 }
1371 
1372 static void ceph_free_fc(struct fs_context *fc)
1373 {
1374 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1375 
1376 	if (pctx) {
1377 		destroy_mount_options(pctx->opts);
1378 		ceph_destroy_options(pctx->copts);
1379 		kfree(pctx);
1380 	}
1381 }
1382 
1383 static int ceph_reconfigure_fc(struct fs_context *fc)
1384 {
1385 	int err;
1386 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1387 	struct ceph_mount_options *fsopt = pctx->opts;
1388 	struct super_block *sb = fc->root->d_sb;
1389 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1390 
1391 	err = ceph_apply_test_dummy_encryption(sb, fc, fsopt);
1392 	if (err)
1393 		return err;
1394 
1395 	if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1396 		ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1397 	else
1398 		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1399 
1400 	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
1401 		ceph_set_mount_opt(fsc, SPARSEREAD);
1402 	else
1403 		ceph_clear_mount_opt(fsc, SPARSEREAD);
1404 
1405 	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1406 		kfree(fsc->mount_options->mon_addr);
1407 		fsc->mount_options->mon_addr = fsopt->mon_addr;
1408 		fsopt->mon_addr = NULL;
1409 		pr_notice_client(fsc->client,
1410 			"monitor addresses recorded, but not used for reconnection");
1411 	}
1412 
1413 	sync_filesystem(sb);
1414 	return 0;
1415 }
1416 
1417 static const struct fs_context_operations ceph_context_ops = {
1418 	.free		= ceph_free_fc,
1419 	.parse_param	= ceph_parse_mount_param,
1420 	.get_tree	= ceph_get_tree,
1421 	.reconfigure	= ceph_reconfigure_fc,
1422 };
1423 
1424 /*
1425  * Set up the filesystem mount context.
1426  */
1427 static int ceph_init_fs_context(struct fs_context *fc)
1428 {
1429 	struct ceph_parse_opts_ctx *pctx;
1430 	struct ceph_mount_options *fsopt;
1431 
1432 	pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1433 	if (!pctx)
1434 		return -ENOMEM;
1435 
1436 	pctx->copts = ceph_alloc_options();
1437 	if (!pctx->copts)
1438 		goto nomem;
1439 
1440 	pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1441 	if (!pctx->opts)
1442 		goto nomem;
1443 
1444 	fsopt = pctx->opts;
1445 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1446 
1447 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1448 	fsopt->rsize = CEPH_MAX_READ_SIZE;
1449 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
1450 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1451 	if (!fsopt->snapdir_name)
1452 		goto nomem;
1453 
1454 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1455 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1456 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1457 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1458 	fsopt->congestion_kb = default_congestion_kb();
1459 
1460 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1461 	fc->sb_flags |= SB_POSIXACL;
1462 #endif
1463 
1464 	fc->fs_private = pctx;
1465 	fc->ops = &ceph_context_ops;
1466 	return 0;
1467 
1468 nomem:
1469 	destroy_mount_options(pctx->opts);
1470 	ceph_destroy_options(pctx->copts);
1471 	kfree(pctx);
1472 	return -ENOMEM;
1473 }
1474 
1475 /*
1476  * Return true if it successfully increases the blocker counter,
1477  * or false if the mdsc is in stopping and flushed state.
1478  */
1479 static bool __inc_stopping_blocker(struct ceph_mds_client *mdsc)
1480 {
1481 	spin_lock(&mdsc->stopping_lock);
1482 	if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING) {
1483 		spin_unlock(&mdsc->stopping_lock);
1484 		return false;
1485 	}
1486 	atomic_inc(&mdsc->stopping_blockers);
1487 	spin_unlock(&mdsc->stopping_lock);
1488 	return true;
1489 }
1490 
1491 static void __dec_stopping_blocker(struct ceph_mds_client *mdsc)
1492 {
1493 	spin_lock(&mdsc->stopping_lock);
1494 	if (!atomic_dec_return(&mdsc->stopping_blockers) &&
1495 	    mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING)
1496 		complete_all(&mdsc->stopping_waiter);
1497 	spin_unlock(&mdsc->stopping_lock);
1498 }
1499 
1500 /* For metadata IO requests */
1501 bool ceph_inc_mds_stopping_blocker(struct ceph_mds_client *mdsc,
1502 				   struct ceph_mds_session *session)
1503 {
1504 	mutex_lock(&session->s_mutex);
1505 	inc_session_sequence(session);
1506 	mutex_unlock(&session->s_mutex);
1507 
1508 	return __inc_stopping_blocker(mdsc);
1509 }
1510 
1511 void ceph_dec_mds_stopping_blocker(struct ceph_mds_client *mdsc)
1512 {
1513 	__dec_stopping_blocker(mdsc);
1514 }
1515 
1516 /* For data IO requests */
1517 bool ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1518 {
1519 	return __inc_stopping_blocker(mdsc);
1520 }
1521 
1522 void ceph_dec_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1523 {
1524 	__dec_stopping_blocker(mdsc);
1525 }
1526 
1527 static void ceph_kill_sb(struct super_block *s)
1528 {
1529 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
1530 	struct ceph_client *cl = fsc->client;
1531 	struct ceph_mds_client *mdsc = fsc->mdsc;
1532 	bool wait;
1533 
1534 	doutc(cl, "%p\n", s);
1535 
1536 	ceph_mdsc_pre_umount(mdsc);
1537 	flush_fs_workqueues(fsc);
1538 
1539 	/*
1540 	 * Though the kill_anon_super() will finally trigger the
1541 	 * sync_filesystem() anyway, we still need to do it here and
1542 	 * then bump the stage of shutdown. This will allow us to
1543 	 * drop any further message, which will increase the inodes'
1544 	 * i_count reference counters but makes no sense any more,
1545 	 * from MDSs.
1546 	 *
1547 	 * Without this when evicting the inodes it may fail in the
1548 	 * kill_anon_super(), which will trigger a warning when
1549 	 * destroying the fscrypt keyring and then possibly trigger
1550 	 * a further crash in ceph module when the iput() tries to
1551 	 * evict the inodes later.
1552 	 */
1553 	sync_filesystem(s);
1554 
1555 	spin_lock(&mdsc->stopping_lock);
1556 	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHING;
1557 	wait = !!atomic_read(&mdsc->stopping_blockers);
1558 	spin_unlock(&mdsc->stopping_lock);
1559 
1560 	if (wait && atomic_read(&mdsc->stopping_blockers)) {
1561 		long timeleft = wait_for_completion_killable_timeout(
1562 					&mdsc->stopping_waiter,
1563 					fsc->client->options->mount_timeout);
1564 		if (!timeleft) /* timed out */
1565 			pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1566 		else if (timeleft < 0) /* killed */
1567 			pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1568 	}
1569 
1570 	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHED;
1571 	kill_anon_super(s);
1572 
1573 	fsc->client->extra_mon_dispatch = NULL;
1574 	ceph_fs_debugfs_cleanup(fsc);
1575 
1576 	ceph_fscache_unregister_fs(fsc);
1577 
1578 	destroy_fs_client(fsc);
1579 }
1580 
1581 static struct file_system_type ceph_fs_type = {
1582 	.owner		= THIS_MODULE,
1583 	.name		= "ceph",
1584 	.init_fs_context = ceph_init_fs_context,
1585 	.kill_sb	= ceph_kill_sb,
1586 	.fs_flags	= FS_RENAME_DOES_D_MOVE | FS_ALLOW_IDMAP,
1587 };
1588 MODULE_ALIAS_FS("ceph");
1589 
1590 int ceph_force_reconnect(struct super_block *sb)
1591 {
1592 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1593 	int err = 0;
1594 
1595 	fsc->mount_state = CEPH_MOUNT_RECOVER;
1596 	__ceph_umount_begin(fsc);
1597 
1598 	/* Make sure all page caches get invalidated.
1599 	 * see remove_session_caps_cb() */
1600 	flush_workqueue(fsc->inode_wq);
1601 
1602 	/* In case that we were blocklisted. This also reset
1603 	 * all mon/osd connections */
1604 	ceph_reset_client_addr(fsc->client);
1605 
1606 	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1607 
1608 	fsc->blocklisted = false;
1609 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1610 
1611 	if (sb->s_root) {
1612 		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1613 					CEPH_STAT_CAP_INODE, true);
1614 	}
1615 	return err;
1616 }
1617 
1618 static int __init init_ceph(void)
1619 {
1620 	int ret = init_caches();
1621 	if (ret)
1622 		goto out;
1623 
1624 	ceph_flock_init();
1625 	ret = register_filesystem(&ceph_fs_type);
1626 	if (ret)
1627 		goto out_caches;
1628 
1629 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1630 
1631 	return 0;
1632 
1633 out_caches:
1634 	destroy_caches();
1635 out:
1636 	return ret;
1637 }
1638 
1639 static void __exit exit_ceph(void)
1640 {
1641 	dout("exit_ceph\n");
1642 	unregister_filesystem(&ceph_fs_type);
1643 	destroy_caches();
1644 }
1645 
1646 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1647 {
1648 	struct ceph_fs_client *fsc;
1649 	int ret;
1650 
1651 	ret = param_set_bool(val, kp);
1652 	if (ret) {
1653 		pr_err("Failed to parse sending metrics switch value '%s'\n",
1654 		       val);
1655 		return ret;
1656 	} else if (!disable_send_metrics) {
1657 		// wake up all the mds clients
1658 		spin_lock(&ceph_fsc_lock);
1659 		list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1660 			metric_schedule_delayed(&fsc->mdsc->metric);
1661 		}
1662 		spin_unlock(&ceph_fsc_lock);
1663 	}
1664 
1665 	return 0;
1666 }
1667 
1668 static const struct kernel_param_ops param_ops_metrics = {
1669 	.set = param_set_metrics,
1670 	.get = param_get_bool,
1671 };
1672 
1673 bool disable_send_metrics = false;
1674 module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1675 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1676 
1677 /* for both v1 and v2 syntax */
1678 static bool mount_support = true;
1679 static const struct kernel_param_ops param_ops_mount_syntax = {
1680 	.get = param_get_bool,
1681 };
1682 module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1683 module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1684 
1685 bool enable_unsafe_idmap = false;
1686 module_param(enable_unsafe_idmap, bool, 0644);
1687 MODULE_PARM_DESC(enable_unsafe_idmap,
1688 		 "Allow to use idmapped mounts with MDS without CEPHFS_FEATURE_HAS_OWNER_UIDGID");
1689 
1690 module_init(init_ceph);
1691 module_exit(exit_ceph);
1692 
1693 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1694 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1695 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1696 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1697 MODULE_LICENSE("GPL");
1698