xref: /linux/fs/ceph/super.c (revision ab52c59103002b49f2455371e4b9c56ba3ef1781)
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, CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT);
965 	if (!ceph_wb_pagevec_pool)
966 		goto bad_pagevec_pool;
967 
968 	return 0;
969 
970 bad_pagevec_pool:
971 	kmem_cache_destroy(ceph_mds_request_cachep);
972 bad_mds_req:
973 	kmem_cache_destroy(ceph_dir_file_cachep);
974 bad_dir_file:
975 	kmem_cache_destroy(ceph_file_cachep);
976 bad_file:
977 	kmem_cache_destroy(ceph_dentry_cachep);
978 bad_dentry:
979 	kmem_cache_destroy(ceph_cap_flush_cachep);
980 bad_cap_flush:
981 	kmem_cache_destroy(ceph_cap_snap_cachep);
982 bad_cap_snap:
983 	kmem_cache_destroy(ceph_cap_cachep);
984 bad_cap:
985 	kmem_cache_destroy(ceph_inode_cachep);
986 	return error;
987 }
988 
989 static void destroy_caches(void)
990 {
991 	/*
992 	 * Make sure all delayed rcu free inodes are flushed before we
993 	 * destroy cache.
994 	 */
995 	rcu_barrier();
996 
997 	kmem_cache_destroy(ceph_inode_cachep);
998 	kmem_cache_destroy(ceph_cap_cachep);
999 	kmem_cache_destroy(ceph_cap_snap_cachep);
1000 	kmem_cache_destroy(ceph_cap_flush_cachep);
1001 	kmem_cache_destroy(ceph_dentry_cachep);
1002 	kmem_cache_destroy(ceph_file_cachep);
1003 	kmem_cache_destroy(ceph_dir_file_cachep);
1004 	kmem_cache_destroy(ceph_mds_request_cachep);
1005 	mempool_destroy(ceph_wb_pagevec_pool);
1006 }
1007 
1008 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
1009 {
1010 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
1011 	ceph_mdsc_force_umount(fsc->mdsc);
1012 	fsc->filp_gen++; // invalidate open files
1013 }
1014 
1015 /*
1016  * ceph_umount_begin - initiate forced umount.  Tear down the
1017  * mount, skipping steps that may hang while waiting for server(s).
1018  */
1019 void ceph_umount_begin(struct super_block *sb)
1020 {
1021 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1022 
1023 	doutc(fsc->client, "starting forced umount\n");
1024 	if (!fsc)
1025 		return;
1026 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
1027 	__ceph_umount_begin(fsc);
1028 }
1029 
1030 static const struct super_operations ceph_super_ops = {
1031 	.alloc_inode	= ceph_alloc_inode,
1032 	.free_inode	= ceph_free_inode,
1033 	.write_inode    = ceph_write_inode,
1034 	.drop_inode	= generic_delete_inode,
1035 	.evict_inode	= ceph_evict_inode,
1036 	.sync_fs        = ceph_sync_fs,
1037 	.put_super	= ceph_put_super,
1038 	.show_options   = ceph_show_options,
1039 	.statfs		= ceph_statfs,
1040 	.umount_begin   = ceph_umount_begin,
1041 };
1042 
1043 /*
1044  * Bootstrap mount by opening the root directory.  Note the mount
1045  * @started time from caller, and time out if this takes too long.
1046  */
1047 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1048 				       const char *path,
1049 				       unsigned long started)
1050 {
1051 	struct ceph_client *cl = fsc->client;
1052 	struct ceph_mds_client *mdsc = fsc->mdsc;
1053 	struct ceph_mds_request *req = NULL;
1054 	int err;
1055 	struct dentry *root;
1056 
1057 	/* open dir */
1058 	doutc(cl, "opening '%s'\n", path);
1059 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1060 	if (IS_ERR(req))
1061 		return ERR_CAST(req);
1062 	req->r_path1 = kstrdup(path, GFP_NOFS);
1063 	if (!req->r_path1) {
1064 		root = ERR_PTR(-ENOMEM);
1065 		goto out;
1066 	}
1067 
1068 	req->r_ino1.ino = CEPH_INO_ROOT;
1069 	req->r_ino1.snap = CEPH_NOSNAP;
1070 	req->r_started = started;
1071 	req->r_timeout = fsc->client->options->mount_timeout;
1072 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1073 	req->r_num_caps = 2;
1074 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1075 	if (err == 0) {
1076 		struct inode *inode = req->r_target_inode;
1077 		req->r_target_inode = NULL;
1078 		doutc(cl, "success\n");
1079 		root = d_make_root(inode);
1080 		if (!root) {
1081 			root = ERR_PTR(-ENOMEM);
1082 			goto out;
1083 		}
1084 		doutc(cl, "success, root dentry is %p\n", root);
1085 	} else {
1086 		root = ERR_PTR(err);
1087 	}
1088 out:
1089 	ceph_mdsc_put_request(req);
1090 	return root;
1091 }
1092 
1093 #ifdef CONFIG_FS_ENCRYPTION
1094 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1095 					    struct fs_context *fc,
1096 					    struct ceph_mount_options *fsopt)
1097 {
1098 	struct ceph_fs_client *fsc = sb->s_fs_info;
1099 
1100 	if (!fscrypt_is_dummy_policy_set(&fsopt->dummy_enc_policy))
1101 		return 0;
1102 
1103 	/* No changing encryption context on remount. */
1104 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE &&
1105 	    !fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1106 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1107 						 &fsc->fsc_dummy_enc_policy))
1108 			return 0;
1109 		errorfc(fc, "Can't set test_dummy_encryption on remount");
1110 		return -EINVAL;
1111 	}
1112 
1113 	/* Also make sure fsopt doesn't contain a conflicting value. */
1114 	if (fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1115 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1116 						 &fsc->fsc_dummy_enc_policy))
1117 			return 0;
1118 		errorfc(fc, "Conflicting test_dummy_encryption options");
1119 		return -EINVAL;
1120 	}
1121 
1122 	fsc->fsc_dummy_enc_policy = fsopt->dummy_enc_policy;
1123 	memset(&fsopt->dummy_enc_policy, 0, sizeof(fsopt->dummy_enc_policy));
1124 
1125 	warnfc(fc, "test_dummy_encryption mode enabled");
1126 	return 0;
1127 }
1128 #else
1129 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1130 					    struct fs_context *fc,
1131 					    struct ceph_mount_options *fsopt)
1132 {
1133 	return 0;
1134 }
1135 #endif
1136 
1137 /*
1138  * mount: join the ceph cluster, and open root directory.
1139  */
1140 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1141 				      struct fs_context *fc)
1142 {
1143 	struct ceph_client *cl = fsc->client;
1144 	int err;
1145 	unsigned long started = jiffies;  /* note the start time */
1146 	struct dentry *root;
1147 
1148 	doutc(cl, "mount start %p\n", fsc);
1149 	mutex_lock(&fsc->client->mount_mutex);
1150 
1151 	if (!fsc->sb->s_root) {
1152 		const char *path = fsc->mount_options->server_path ?
1153 				     fsc->mount_options->server_path + 1 : "";
1154 
1155 		err = __ceph_open_session(fsc->client, started);
1156 		if (err < 0)
1157 			goto out;
1158 
1159 		/* setup fscache */
1160 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1161 			err = ceph_fscache_register_fs(fsc, fc);
1162 			if (err < 0)
1163 				goto out;
1164 		}
1165 
1166 		err = ceph_apply_test_dummy_encryption(fsc->sb, fc,
1167 						       fsc->mount_options);
1168 		if (err)
1169 			goto out;
1170 
1171 		doutc(cl, "mount opening path '%s'\n", path);
1172 
1173 		ceph_fs_debugfs_init(fsc);
1174 
1175 		root = open_root_dentry(fsc, path, started);
1176 		if (IS_ERR(root)) {
1177 			err = PTR_ERR(root);
1178 			goto out;
1179 		}
1180 		fsc->sb->s_root = dget(root);
1181 	} else {
1182 		root = dget(fsc->sb->s_root);
1183 	}
1184 
1185 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1186 	doutc(cl, "mount success\n");
1187 	mutex_unlock(&fsc->client->mount_mutex);
1188 	return root;
1189 
1190 out:
1191 	mutex_unlock(&fsc->client->mount_mutex);
1192 	ceph_fscrypt_free_dummy_policy(fsc);
1193 	return ERR_PTR(err);
1194 }
1195 
1196 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1197 {
1198 	struct ceph_fs_client *fsc = s->s_fs_info;
1199 	struct ceph_client *cl = fsc->client;
1200 	int ret;
1201 
1202 	doutc(cl, "%p\n", s);
1203 
1204 	s->s_maxbytes = MAX_LFS_FILESIZE;
1205 
1206 	s->s_xattr = ceph_xattr_handlers;
1207 	fsc->sb = s;
1208 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1209 
1210 	s->s_op = &ceph_super_ops;
1211 	s->s_d_op = &ceph_dentry_ops;
1212 	s->s_export_op = &ceph_export_ops;
1213 
1214 	s->s_time_gran = 1;
1215 	s->s_time_min = 0;
1216 	s->s_time_max = U32_MAX;
1217 	s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1218 
1219 	ceph_fscrypt_set_ops(s);
1220 
1221 	ret = set_anon_super_fc(s, fc);
1222 	if (ret != 0)
1223 		fsc->sb = NULL;
1224 	return ret;
1225 }
1226 
1227 /*
1228  * share superblock if same fs AND options
1229  */
1230 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1231 {
1232 	struct ceph_fs_client *new = fc->s_fs_info;
1233 	struct ceph_mount_options *fsopt = new->mount_options;
1234 	struct ceph_options *opt = new->client->options;
1235 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1236 	struct ceph_client *cl = fsc->client;
1237 
1238 	doutc(cl, "%p\n", sb);
1239 
1240 	if (compare_mount_options(fsopt, opt, fsc)) {
1241 		doutc(cl, "monitor(s)/mount options don't match\n");
1242 		return 0;
1243 	}
1244 	if ((opt->flags & CEPH_OPT_FSID) &&
1245 	    ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1246 		doutc(cl, "fsid doesn't match\n");
1247 		return 0;
1248 	}
1249 	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1250 		doutc(cl, "flags differ\n");
1251 		return 0;
1252 	}
1253 
1254 	if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1255 		doutc(cl, "client is blocklisted (and CLEANRECOVER is not set)\n");
1256 		return 0;
1257 	}
1258 
1259 	if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1260 		doutc(cl, "client has been forcibly unmounted\n");
1261 		return 0;
1262 	}
1263 
1264 	return 1;
1265 }
1266 
1267 /*
1268  * construct our own bdi so we can control readahead, etc.
1269  */
1270 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1271 
1272 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1273 {
1274 	int err;
1275 
1276 	err = super_setup_bdi_name(sb, "ceph-%ld",
1277 				   atomic_long_inc_return(&bdi_seq));
1278 	if (err)
1279 		return err;
1280 
1281 	/* set ra_pages based on rasize mount option? */
1282 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1283 
1284 	/* set io_pages based on max osd read size */
1285 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1286 
1287 	return 0;
1288 }
1289 
1290 static int ceph_get_tree(struct fs_context *fc)
1291 {
1292 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1293 	struct ceph_mount_options *fsopt = pctx->opts;
1294 	struct super_block *sb;
1295 	struct ceph_fs_client *fsc;
1296 	struct dentry *res;
1297 	int (*compare_super)(struct super_block *, struct fs_context *) =
1298 		ceph_compare_super;
1299 	int err;
1300 
1301 	dout("ceph_get_tree\n");
1302 
1303 	if (!fc->source)
1304 		return invalfc(fc, "No source");
1305 	if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1306 		return invalfc(fc, "No monitor address");
1307 
1308 	/* create client (which we may/may not use) */
1309 	fsc = create_fs_client(pctx->opts, pctx->copts);
1310 	pctx->opts = NULL;
1311 	pctx->copts = NULL;
1312 	if (IS_ERR(fsc)) {
1313 		err = PTR_ERR(fsc);
1314 		goto out_final;
1315 	}
1316 
1317 	err = ceph_mdsc_init(fsc);
1318 	if (err < 0)
1319 		goto out;
1320 
1321 	if (ceph_test_opt(fsc->client, NOSHARE))
1322 		compare_super = NULL;
1323 
1324 	fc->s_fs_info = fsc;
1325 	sb = sget_fc(fc, compare_super, ceph_set_super);
1326 	fc->s_fs_info = NULL;
1327 	if (IS_ERR(sb)) {
1328 		err = PTR_ERR(sb);
1329 		goto out;
1330 	}
1331 
1332 	if (ceph_sb_to_fs_client(sb) != fsc) {
1333 		destroy_fs_client(fsc);
1334 		fsc = ceph_sb_to_fs_client(sb);
1335 		dout("get_sb got existing client %p\n", fsc);
1336 	} else {
1337 		dout("get_sb using new client %p\n", fsc);
1338 		err = ceph_setup_bdi(sb, fsc);
1339 		if (err < 0)
1340 			goto out_splat;
1341 	}
1342 
1343 	res = ceph_real_mount(fsc, fc);
1344 	if (IS_ERR(res)) {
1345 		err = PTR_ERR(res);
1346 		goto out_splat;
1347 	}
1348 
1349 	doutc(fsc->client, "root %p inode %p ino %llx.%llx\n", res,
1350 		    d_inode(res), ceph_vinop(d_inode(res)));
1351 	fc->root = fsc->sb->s_root;
1352 	return 0;
1353 
1354 out_splat:
1355 	if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1356 		pr_info("No mds server is up or the cluster is laggy\n");
1357 		err = -EHOSTUNREACH;
1358 	}
1359 
1360 	ceph_mdsc_close_sessions(fsc->mdsc);
1361 	deactivate_locked_super(sb);
1362 	goto out_final;
1363 
1364 out:
1365 	destroy_fs_client(fsc);
1366 out_final:
1367 	dout("ceph_get_tree fail %d\n", err);
1368 	return err;
1369 }
1370 
1371 static void ceph_free_fc(struct fs_context *fc)
1372 {
1373 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1374 
1375 	if (pctx) {
1376 		destroy_mount_options(pctx->opts);
1377 		ceph_destroy_options(pctx->copts);
1378 		kfree(pctx);
1379 	}
1380 }
1381 
1382 static int ceph_reconfigure_fc(struct fs_context *fc)
1383 {
1384 	int err;
1385 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1386 	struct ceph_mount_options *fsopt = pctx->opts;
1387 	struct super_block *sb = fc->root->d_sb;
1388 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1389 
1390 	err = ceph_apply_test_dummy_encryption(sb, fc, fsopt);
1391 	if (err)
1392 		return err;
1393 
1394 	if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1395 		ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1396 	else
1397 		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1398 
1399 	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
1400 		ceph_set_mount_opt(fsc, SPARSEREAD);
1401 	else
1402 		ceph_clear_mount_opt(fsc, SPARSEREAD);
1403 
1404 	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1405 		kfree(fsc->mount_options->mon_addr);
1406 		fsc->mount_options->mon_addr = fsopt->mon_addr;
1407 		fsopt->mon_addr = NULL;
1408 		pr_notice_client(fsc->client,
1409 			"monitor addresses recorded, but not used for reconnection");
1410 	}
1411 
1412 	sync_filesystem(sb);
1413 	return 0;
1414 }
1415 
1416 static const struct fs_context_operations ceph_context_ops = {
1417 	.free		= ceph_free_fc,
1418 	.parse_param	= ceph_parse_mount_param,
1419 	.get_tree	= ceph_get_tree,
1420 	.reconfigure	= ceph_reconfigure_fc,
1421 };
1422 
1423 /*
1424  * Set up the filesystem mount context.
1425  */
1426 static int ceph_init_fs_context(struct fs_context *fc)
1427 {
1428 	struct ceph_parse_opts_ctx *pctx;
1429 	struct ceph_mount_options *fsopt;
1430 
1431 	pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1432 	if (!pctx)
1433 		return -ENOMEM;
1434 
1435 	pctx->copts = ceph_alloc_options();
1436 	if (!pctx->copts)
1437 		goto nomem;
1438 
1439 	pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1440 	if (!pctx->opts)
1441 		goto nomem;
1442 
1443 	fsopt = pctx->opts;
1444 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1445 
1446 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1447 	fsopt->rsize = CEPH_MAX_READ_SIZE;
1448 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
1449 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1450 	if (!fsopt->snapdir_name)
1451 		goto nomem;
1452 
1453 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1454 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1455 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1456 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1457 	fsopt->congestion_kb = default_congestion_kb();
1458 
1459 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1460 	fc->sb_flags |= SB_POSIXACL;
1461 #endif
1462 
1463 	fc->fs_private = pctx;
1464 	fc->ops = &ceph_context_ops;
1465 	return 0;
1466 
1467 nomem:
1468 	destroy_mount_options(pctx->opts);
1469 	ceph_destroy_options(pctx->copts);
1470 	kfree(pctx);
1471 	return -ENOMEM;
1472 }
1473 
1474 /*
1475  * Return true if it successfully increases the blocker counter,
1476  * or false if the mdsc is in stopping and flushed state.
1477  */
1478 static bool __inc_stopping_blocker(struct ceph_mds_client *mdsc)
1479 {
1480 	spin_lock(&mdsc->stopping_lock);
1481 	if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING) {
1482 		spin_unlock(&mdsc->stopping_lock);
1483 		return false;
1484 	}
1485 	atomic_inc(&mdsc->stopping_blockers);
1486 	spin_unlock(&mdsc->stopping_lock);
1487 	return true;
1488 }
1489 
1490 static void __dec_stopping_blocker(struct ceph_mds_client *mdsc)
1491 {
1492 	spin_lock(&mdsc->stopping_lock);
1493 	if (!atomic_dec_return(&mdsc->stopping_blockers) &&
1494 	    mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING)
1495 		complete_all(&mdsc->stopping_waiter);
1496 	spin_unlock(&mdsc->stopping_lock);
1497 }
1498 
1499 /* For metadata IO requests */
1500 bool ceph_inc_mds_stopping_blocker(struct ceph_mds_client *mdsc,
1501 				   struct ceph_mds_session *session)
1502 {
1503 	mutex_lock(&session->s_mutex);
1504 	inc_session_sequence(session);
1505 	mutex_unlock(&session->s_mutex);
1506 
1507 	return __inc_stopping_blocker(mdsc);
1508 }
1509 
1510 void ceph_dec_mds_stopping_blocker(struct ceph_mds_client *mdsc)
1511 {
1512 	__dec_stopping_blocker(mdsc);
1513 }
1514 
1515 /* For data IO requests */
1516 bool ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1517 {
1518 	return __inc_stopping_blocker(mdsc);
1519 }
1520 
1521 void ceph_dec_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1522 {
1523 	__dec_stopping_blocker(mdsc);
1524 }
1525 
1526 static void ceph_kill_sb(struct super_block *s)
1527 {
1528 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
1529 	struct ceph_client *cl = fsc->client;
1530 	struct ceph_mds_client *mdsc = fsc->mdsc;
1531 	bool wait;
1532 
1533 	doutc(cl, "%p\n", s);
1534 
1535 	ceph_mdsc_pre_umount(mdsc);
1536 	flush_fs_workqueues(fsc);
1537 
1538 	/*
1539 	 * Though the kill_anon_super() will finally trigger the
1540 	 * sync_filesystem() anyway, we still need to do it here and
1541 	 * then bump the stage of shutdown. This will allow us to
1542 	 * drop any further message, which will increase the inodes'
1543 	 * i_count reference counters but makes no sense any more,
1544 	 * from MDSs.
1545 	 *
1546 	 * Without this when evicting the inodes it may fail in the
1547 	 * kill_anon_super(), which will trigger a warning when
1548 	 * destroying the fscrypt keyring and then possibly trigger
1549 	 * a further crash in ceph module when the iput() tries to
1550 	 * evict the inodes later.
1551 	 */
1552 	sync_filesystem(s);
1553 
1554 	spin_lock(&mdsc->stopping_lock);
1555 	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHING;
1556 	wait = !!atomic_read(&mdsc->stopping_blockers);
1557 	spin_unlock(&mdsc->stopping_lock);
1558 
1559 	if (wait && atomic_read(&mdsc->stopping_blockers)) {
1560 		long timeleft = wait_for_completion_killable_timeout(
1561 					&mdsc->stopping_waiter,
1562 					fsc->client->options->mount_timeout);
1563 		if (!timeleft) /* timed out */
1564 			pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1565 		else if (timeleft < 0) /* killed */
1566 			pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1567 	}
1568 
1569 	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHED;
1570 	kill_anon_super(s);
1571 
1572 	fsc->client->extra_mon_dispatch = NULL;
1573 	ceph_fs_debugfs_cleanup(fsc);
1574 
1575 	ceph_fscache_unregister_fs(fsc);
1576 
1577 	destroy_fs_client(fsc);
1578 }
1579 
1580 static struct file_system_type ceph_fs_type = {
1581 	.owner		= THIS_MODULE,
1582 	.name		= "ceph",
1583 	.init_fs_context = ceph_init_fs_context,
1584 	.kill_sb	= ceph_kill_sb,
1585 	.fs_flags	= FS_RENAME_DOES_D_MOVE | FS_ALLOW_IDMAP,
1586 };
1587 MODULE_ALIAS_FS("ceph");
1588 
1589 int ceph_force_reconnect(struct super_block *sb)
1590 {
1591 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1592 	int err = 0;
1593 
1594 	fsc->mount_state = CEPH_MOUNT_RECOVER;
1595 	__ceph_umount_begin(fsc);
1596 
1597 	/* Make sure all page caches get invalidated.
1598 	 * see remove_session_caps_cb() */
1599 	flush_workqueue(fsc->inode_wq);
1600 
1601 	/* In case that we were blocklisted. This also reset
1602 	 * all mon/osd connections */
1603 	ceph_reset_client_addr(fsc->client);
1604 
1605 	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1606 
1607 	fsc->blocklisted = false;
1608 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1609 
1610 	if (sb->s_root) {
1611 		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1612 					CEPH_STAT_CAP_INODE, true);
1613 	}
1614 	return err;
1615 }
1616 
1617 static int __init init_ceph(void)
1618 {
1619 	int ret = init_caches();
1620 	if (ret)
1621 		goto out;
1622 
1623 	ceph_flock_init();
1624 	ret = register_filesystem(&ceph_fs_type);
1625 	if (ret)
1626 		goto out_caches;
1627 
1628 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1629 
1630 	return 0;
1631 
1632 out_caches:
1633 	destroy_caches();
1634 out:
1635 	return ret;
1636 }
1637 
1638 static void __exit exit_ceph(void)
1639 {
1640 	dout("exit_ceph\n");
1641 	unregister_filesystem(&ceph_fs_type);
1642 	destroy_caches();
1643 }
1644 
1645 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1646 {
1647 	struct ceph_fs_client *fsc;
1648 	int ret;
1649 
1650 	ret = param_set_bool(val, kp);
1651 	if (ret) {
1652 		pr_err("Failed to parse sending metrics switch value '%s'\n",
1653 		       val);
1654 		return ret;
1655 	} else if (!disable_send_metrics) {
1656 		// wake up all the mds clients
1657 		spin_lock(&ceph_fsc_lock);
1658 		list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1659 			metric_schedule_delayed(&fsc->mdsc->metric);
1660 		}
1661 		spin_unlock(&ceph_fsc_lock);
1662 	}
1663 
1664 	return 0;
1665 }
1666 
1667 static const struct kernel_param_ops param_ops_metrics = {
1668 	.set = param_set_metrics,
1669 	.get = param_get_bool,
1670 };
1671 
1672 bool disable_send_metrics = false;
1673 module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1674 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1675 
1676 /* for both v1 and v2 syntax */
1677 static bool mount_support = true;
1678 static const struct kernel_param_ops param_ops_mount_syntax = {
1679 	.get = param_get_bool,
1680 };
1681 module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1682 module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1683 
1684 bool enable_unsafe_idmap = false;
1685 module_param(enable_unsafe_idmap, bool, 0644);
1686 MODULE_PARM_DESC(enable_unsafe_idmap,
1687 		 "Allow to use idmapped mounts with MDS without CEPHFS_FEATURE_HAS_OWNER_UIDGID");
1688 
1689 module_init(init_ceph);
1690 module_exit(exit_ceph);
1691 
1692 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1693 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1694 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1695 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1696 MODULE_LICENSE("GPL");
1697