xref: /openbmc/linux/fs/ceph/super.c (revision 55e43d6abd078ed6d219902ce8cb4d68e3c993ba)
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  */
ceph_put_super(struct super_block * s)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 	dout("put_super\n");
50 	ceph_fscrypt_free_dummy_policy(fsc);
51 	ceph_mdsc_close_sessions(fsc->mdsc);
52 }
53 
ceph_statfs(struct dentry * dentry,struct kstatfs * buf)54 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
55 {
56 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(d_inode(dentry));
57 	struct ceph_mon_client *monc = &fsc->client->monc;
58 	struct ceph_statfs st;
59 	int i, err;
60 	u64 data_pool;
61 
62 	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
63 		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
64 	} else {
65 		data_pool = CEPH_NOPOOL;
66 	}
67 
68 	dout("statfs\n");
69 	err = ceph_monc_do_statfs(monc, data_pool, &st);
70 	if (err < 0)
71 		return err;
72 
73 	/* fill in kstatfs */
74 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
75 
76 	/*
77 	 * Express utilization in terms of large blocks to avoid
78 	 * overflow on 32-bit machines.
79 	 */
80 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
81 
82 	/*
83 	 * By default use root quota for stats; fallback to overall filesystem
84 	 * usage if using 'noquotadf' mount option or if the root dir doesn't
85 	 * have max_bytes quota set.
86 	 */
87 	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
88 	    !ceph_quota_update_statfs(fsc, buf)) {
89 		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
90 		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
91 		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
92 	}
93 
94 	/*
95 	 * NOTE: for the time being, we make bsize == frsize to humor
96 	 * not-yet-ancient versions of glibc that are broken.
97 	 * Someday, we will probably want to report a real block
98 	 * size...  whatever that may mean for a network file system!
99 	 */
100 	buf->f_bsize = buf->f_frsize;
101 
102 	buf->f_files = le64_to_cpu(st.num_objects);
103 	buf->f_ffree = -1;
104 	buf->f_namelen = NAME_MAX;
105 
106 	/* Must convert the fsid, for consistent values across arches */
107 	buf->f_fsid.val[0] = 0;
108 	mutex_lock(&monc->mutex);
109 	for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
110 		buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
111 	mutex_unlock(&monc->mutex);
112 
113 	/* fold the fs_cluster_id into the upper bits */
114 	buf->f_fsid.val[1] = monc->fs_cluster_id;
115 
116 	return 0;
117 }
118 
ceph_sync_fs(struct super_block * sb,int wait)119 static int ceph_sync_fs(struct super_block *sb, int wait)
120 {
121 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
122 
123 	if (!wait) {
124 		dout("sync_fs (non-blocking)\n");
125 		ceph_flush_dirty_caps(fsc->mdsc);
126 		dout("sync_fs (non-blocking) done\n");
127 		return 0;
128 	}
129 
130 	dout("sync_fs (blocking)\n");
131 	ceph_osdc_sync(&fsc->client->osdc);
132 	ceph_mdsc_sync(fsc->mdsc);
133 	dout("sync_fs (blocking) done\n");
134 	return 0;
135 }
136 
137 /*
138  * mount options
139  */
140 enum {
141 	Opt_wsize,
142 	Opt_rsize,
143 	Opt_rasize,
144 	Opt_caps_wanted_delay_min,
145 	Opt_caps_wanted_delay_max,
146 	Opt_caps_max,
147 	Opt_readdir_max_entries,
148 	Opt_readdir_max_bytes,
149 	Opt_congestion_kb,
150 	/* int args above */
151 	Opt_snapdirname,
152 	Opt_mds_namespace,
153 	Opt_recover_session,
154 	Opt_source,
155 	Opt_mon_addr,
156 	Opt_test_dummy_encryption,
157 	/* string args above */
158 	Opt_dirstat,
159 	Opt_rbytes,
160 	Opt_asyncreaddir,
161 	Opt_dcache,
162 	Opt_ino32,
163 	Opt_fscache,
164 	Opt_poolperm,
165 	Opt_require_active_mds,
166 	Opt_acl,
167 	Opt_quotadf,
168 	Opt_copyfrom,
169 	Opt_wsync,
170 	Opt_pagecache,
171 	Opt_sparseread,
172 };
173 
174 enum ceph_recover_session_mode {
175 	ceph_recover_session_no,
176 	ceph_recover_session_clean
177 };
178 
179 static const struct constant_table ceph_param_recover[] = {
180 	{ "no",		ceph_recover_session_no },
181 	{ "clean",	ceph_recover_session_clean },
182 	{}
183 };
184 
185 static const struct fs_parameter_spec ceph_mount_parameters[] = {
186 	fsparam_flag_no ("acl",				Opt_acl),
187 	fsparam_flag_no ("asyncreaddir",		Opt_asyncreaddir),
188 	fsparam_s32	("caps_max",			Opt_caps_max),
189 	fsparam_u32	("caps_wanted_delay_max",	Opt_caps_wanted_delay_max),
190 	fsparam_u32	("caps_wanted_delay_min",	Opt_caps_wanted_delay_min),
191 	fsparam_u32	("write_congestion_kb",		Opt_congestion_kb),
192 	fsparam_flag_no ("copyfrom",			Opt_copyfrom),
193 	fsparam_flag_no ("dcache",			Opt_dcache),
194 	fsparam_flag_no ("dirstat",			Opt_dirstat),
195 	fsparam_flag_no	("fsc",				Opt_fscache), // fsc|nofsc
196 	fsparam_string	("fsc",				Opt_fscache), // fsc=...
197 	fsparam_flag_no ("ino32",			Opt_ino32),
198 	fsparam_string	("mds_namespace",		Opt_mds_namespace),
199 	fsparam_string	("mon_addr",			Opt_mon_addr),
200 	fsparam_flag_no ("poolperm",			Opt_poolperm),
201 	fsparam_flag_no ("quotadf",			Opt_quotadf),
202 	fsparam_u32	("rasize",			Opt_rasize),
203 	fsparam_flag_no ("rbytes",			Opt_rbytes),
204 	fsparam_u32	("readdir_max_bytes",		Opt_readdir_max_bytes),
205 	fsparam_u32	("readdir_max_entries",		Opt_readdir_max_entries),
206 	fsparam_enum	("recover_session",		Opt_recover_session, ceph_param_recover),
207 	fsparam_flag_no ("require_active_mds",		Opt_require_active_mds),
208 	fsparam_u32	("rsize",			Opt_rsize),
209 	fsparam_string	("snapdirname",			Opt_snapdirname),
210 	fsparam_string	("source",			Opt_source),
211 	fsparam_flag	("test_dummy_encryption",	Opt_test_dummy_encryption),
212 	fsparam_string	("test_dummy_encryption",	Opt_test_dummy_encryption),
213 	fsparam_u32	("wsize",			Opt_wsize),
214 	fsparam_flag_no	("wsync",			Opt_wsync),
215 	fsparam_flag_no	("pagecache",			Opt_pagecache),
216 	fsparam_flag_no	("sparseread",			Opt_sparseread),
217 	{}
218 };
219 
220 struct ceph_parse_opts_ctx {
221 	struct ceph_options		*copts;
222 	struct ceph_mount_options	*opts;
223 };
224 
225 /*
226  * Remove adjacent slashes and then the trailing slash, unless it is
227  * the only remaining character.
228  *
229  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
230  */
canonicalize_path(char * path)231 static void canonicalize_path(char *path)
232 {
233 	int i, j = 0;
234 
235 	for (i = 0; path[i] != '\0'; i++) {
236 		if (path[i] != '/' || j < 1 || path[j - 1] != '/')
237 			path[j++] = path[i];
238 	}
239 
240 	if (j > 1 && path[j - 1] == '/')
241 		j--;
242 	path[j] = '\0';
243 }
244 
245 /*
246  * Check if the mds namespace in ceph_mount_options matches
247  * the passed in namespace string. First time match (when
248  * ->mds_namespace is NULL) is treated specially, since
249  * ->mds_namespace needs to be initialized by the caller.
250  */
namespace_equals(struct ceph_mount_options * fsopt,const char * namespace,size_t len)251 static int namespace_equals(struct ceph_mount_options *fsopt,
252 			    const char *namespace, size_t len)
253 {
254 	return !(fsopt->mds_namespace &&
255 		 (strlen(fsopt->mds_namespace) != len ||
256 		  strncmp(fsopt->mds_namespace, namespace, len)));
257 }
258 
ceph_parse_old_source(const char * dev_name,const char * dev_name_end,struct fs_context * fc)259 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
260 				 struct fs_context *fc)
261 {
262 	int r;
263 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
264 	struct ceph_mount_options *fsopt = pctx->opts;
265 
266 	if (*dev_name_end != ':')
267 		return invalfc(fc, "separator ':' missing in source");
268 
269 	r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
270 			       pctx->copts, fc->log.log, ',');
271 	if (r)
272 		return r;
273 
274 	fsopt->new_dev_syntax = false;
275 	return 0;
276 }
277 
ceph_parse_new_source(const char * dev_name,const char * dev_name_end,struct fs_context * fc)278 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
279 				 struct fs_context *fc)
280 {
281 	size_t len;
282 	struct ceph_fsid fsid;
283 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
284 	struct ceph_options *opts = pctx->copts;
285 	struct ceph_mount_options *fsopt = pctx->opts;
286 	const char *name_start = dev_name;
287 	char *fsid_start, *fs_name_start;
288 
289 	if (*dev_name_end != '=') {
290 		dout("separator '=' missing in source");
291 		return -EINVAL;
292 	}
293 
294 	fsid_start = strchr(dev_name, '@');
295 	if (!fsid_start)
296 		return invalfc(fc, "missing cluster fsid");
297 	len = fsid_start - name_start;
298 	kfree(opts->name);
299 	opts->name = kstrndup(name_start, len, GFP_KERNEL);
300 	if (!opts->name)
301 		return -ENOMEM;
302 	dout("using %s entity name", opts->name);
303 
304 	++fsid_start; /* start of cluster fsid */
305 	fs_name_start = strchr(fsid_start, '.');
306 	if (!fs_name_start)
307 		return invalfc(fc, "missing file system name");
308 
309 	if (ceph_parse_fsid(fsid_start, &fsid))
310 		return invalfc(fc, "Invalid FSID");
311 
312 	++fs_name_start; /* start of file system name */
313 	len = dev_name_end - fs_name_start;
314 
315 	if (!namespace_equals(fsopt, fs_name_start, len))
316 		return invalfc(fc, "Mismatching mds_namespace");
317 	kfree(fsopt->mds_namespace);
318 	fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
319 	if (!fsopt->mds_namespace)
320 		return -ENOMEM;
321 	dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
322 
323 	fsopt->new_dev_syntax = true;
324 	return 0;
325 }
326 
327 /*
328  * Parse the source parameter for new device format. Distinguish the device
329  * spec from the path. Try parsing new device format and fallback to old
330  * format if needed.
331  *
332  * New device syntax will looks like:
333  *     <device_spec>=/<path>
334  * where
335  *     <device_spec> is name@fsid.fsname
336  *     <path> is optional, but if present must begin with '/'
337  * (monitor addresses are passed via mount option)
338  *
339  * Old device syntax is:
340  *     <server_spec>[,<server_spec>...]:[<path>]
341  * where
342  *     <server_spec> is <ip>[:<port>]
343  *     <path> is optional, but if present must begin with '/'
344  */
ceph_parse_source(struct fs_parameter * param,struct fs_context * fc)345 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
346 {
347 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
348 	struct ceph_mount_options *fsopt = pctx->opts;
349 	char *dev_name = param->string, *dev_name_end;
350 	int ret;
351 
352 	dout("%s '%s'\n", __func__, dev_name);
353 	if (!dev_name || !*dev_name)
354 		return invalfc(fc, "Empty source");
355 
356 	dev_name_end = strchr(dev_name, '/');
357 	if (dev_name_end) {
358 		/*
359 		 * The server_path will include the whole chars from userland
360 		 * including the leading '/'.
361 		 */
362 		kfree(fsopt->server_path);
363 		fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
364 		if (!fsopt->server_path)
365 			return -ENOMEM;
366 
367 		canonicalize_path(fsopt->server_path);
368 	} else {
369 		dev_name_end = dev_name + strlen(dev_name);
370 	}
371 
372 	dev_name_end--;		/* back up to separator */
373 	if (dev_name_end < dev_name)
374 		return invalfc(fc, "Path missing in source");
375 
376 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
377 	if (fsopt->server_path)
378 		dout("server path '%s'\n", fsopt->server_path);
379 
380 	dout("trying new device syntax");
381 	ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
382 	if (ret) {
383 		if (ret != -EINVAL)
384 			return ret;
385 		dout("trying old device syntax");
386 		ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
387 		if (ret)
388 			return ret;
389 	}
390 
391 	fc->source = param->string;
392 	param->string = NULL;
393 	return 0;
394 }
395 
ceph_parse_mon_addr(struct fs_parameter * param,struct fs_context * fc)396 static int ceph_parse_mon_addr(struct fs_parameter *param,
397 			       struct fs_context *fc)
398 {
399 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
400 	struct ceph_mount_options *fsopt = pctx->opts;
401 
402 	kfree(fsopt->mon_addr);
403 	fsopt->mon_addr = param->string;
404 	param->string = NULL;
405 
406 	return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
407 				  pctx->copts, fc->log.log, '/');
408 }
409 
ceph_parse_mount_param(struct fs_context * fc,struct fs_parameter * param)410 static int ceph_parse_mount_param(struct fs_context *fc,
411 				  struct fs_parameter *param)
412 {
413 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
414 	struct ceph_mount_options *fsopt = pctx->opts;
415 	struct fs_parse_result result;
416 	unsigned int mode;
417 	int token, ret;
418 
419 	ret = ceph_parse_param(param, pctx->copts, fc->log.log);
420 	if (ret != -ENOPARAM)
421 		return ret;
422 
423 	token = fs_parse(fc, ceph_mount_parameters, param, &result);
424 	dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
425 	if (token < 0)
426 		return token;
427 
428 	switch (token) {
429 	case Opt_snapdirname:
430 		if (strlen(param->string) > NAME_MAX)
431 			return invalfc(fc, "snapdirname too long");
432 		kfree(fsopt->snapdir_name);
433 		fsopt->snapdir_name = param->string;
434 		param->string = NULL;
435 		break;
436 	case Opt_mds_namespace:
437 		if (!namespace_equals(fsopt, param->string, strlen(param->string)))
438 			return invalfc(fc, "Mismatching mds_namespace");
439 		kfree(fsopt->mds_namespace);
440 		fsopt->mds_namespace = param->string;
441 		param->string = NULL;
442 		break;
443 	case Opt_recover_session:
444 		mode = result.uint_32;
445 		if (mode == ceph_recover_session_no)
446 			fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
447 		else if (mode == ceph_recover_session_clean)
448 			fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
449 		else
450 			BUG();
451 		break;
452 	case Opt_source:
453 		if (fc->source)
454 			return invalfc(fc, "Multiple sources specified");
455 		return ceph_parse_source(param, fc);
456 	case Opt_mon_addr:
457 		return ceph_parse_mon_addr(param, fc);
458 	case Opt_wsize:
459 		if (result.uint_32 < PAGE_SIZE ||
460 		    result.uint_32 > CEPH_MAX_WRITE_SIZE)
461 			goto out_of_range;
462 		fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
463 		break;
464 	case Opt_rsize:
465 		if (result.uint_32 < PAGE_SIZE ||
466 		    result.uint_32 > CEPH_MAX_READ_SIZE)
467 			goto out_of_range;
468 		fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
469 		break;
470 	case Opt_rasize:
471 		fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
472 		break;
473 	case Opt_caps_wanted_delay_min:
474 		if (result.uint_32 < 1)
475 			goto out_of_range;
476 		fsopt->caps_wanted_delay_min = result.uint_32;
477 		break;
478 	case Opt_caps_wanted_delay_max:
479 		if (result.uint_32 < 1)
480 			goto out_of_range;
481 		fsopt->caps_wanted_delay_max = result.uint_32;
482 		break;
483 	case Opt_caps_max:
484 		if (result.int_32 < 0)
485 			goto out_of_range;
486 		fsopt->caps_max = result.int_32;
487 		break;
488 	case Opt_readdir_max_entries:
489 		if (result.uint_32 < 1)
490 			goto out_of_range;
491 		fsopt->max_readdir = result.uint_32;
492 		break;
493 	case Opt_readdir_max_bytes:
494 		if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
495 			goto out_of_range;
496 		fsopt->max_readdir_bytes = result.uint_32;
497 		break;
498 	case Opt_congestion_kb:
499 		if (result.uint_32 < 1024) /* at least 1M */
500 			goto out_of_range;
501 		fsopt->congestion_kb = result.uint_32;
502 		break;
503 	case Opt_dirstat:
504 		if (!result.negated)
505 			fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
506 		else
507 			fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
508 		break;
509 	case Opt_rbytes:
510 		if (!result.negated)
511 			fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
512 		else
513 			fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
514 		break;
515 	case Opt_asyncreaddir:
516 		if (!result.negated)
517 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
518 		else
519 			fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
520 		break;
521 	case Opt_dcache:
522 		if (!result.negated)
523 			fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
524 		else
525 			fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
526 		break;
527 	case Opt_ino32:
528 		if (!result.negated)
529 			fsopt->flags |= CEPH_MOUNT_OPT_INO32;
530 		else
531 			fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
532 		break;
533 
534 	case Opt_fscache:
535 #ifdef CONFIG_CEPH_FSCACHE
536 		kfree(fsopt->fscache_uniq);
537 		fsopt->fscache_uniq = NULL;
538 		if (result.negated) {
539 			fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
540 		} else {
541 			fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
542 			fsopt->fscache_uniq = param->string;
543 			param->string = NULL;
544 		}
545 		break;
546 #else
547 		return invalfc(fc, "fscache support is disabled");
548 #endif
549 	case Opt_poolperm:
550 		if (!result.negated)
551 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
552 		else
553 			fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
554 		break;
555 	case Opt_require_active_mds:
556 		if (!result.negated)
557 			fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
558 		else
559 			fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
560 		break;
561 	case Opt_quotadf:
562 		if (!result.negated)
563 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
564 		else
565 			fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
566 		break;
567 	case Opt_copyfrom:
568 		if (!result.negated)
569 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
570 		else
571 			fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
572 		break;
573 	case Opt_acl:
574 		if (!result.negated) {
575 #ifdef CONFIG_CEPH_FS_POSIX_ACL
576 			fc->sb_flags |= SB_POSIXACL;
577 #else
578 			return invalfc(fc, "POSIX ACL support is disabled");
579 #endif
580 		} else {
581 			fc->sb_flags &= ~SB_POSIXACL;
582 		}
583 		break;
584 	case Opt_wsync:
585 		if (!result.negated)
586 			fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
587 		else
588 			fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
589 		break;
590 	case Opt_pagecache:
591 		if (result.negated)
592 			fsopt->flags |= CEPH_MOUNT_OPT_NOPAGECACHE;
593 		else
594 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
595 		break;
596 	case Opt_sparseread:
597 		if (result.negated)
598 			fsopt->flags &= ~CEPH_MOUNT_OPT_SPARSEREAD;
599 		else
600 			fsopt->flags |= CEPH_MOUNT_OPT_SPARSEREAD;
601 		break;
602 	case Opt_test_dummy_encryption:
603 #ifdef CONFIG_FS_ENCRYPTION
604 		fscrypt_free_dummy_policy(&fsopt->dummy_enc_policy);
605 		ret = fscrypt_parse_test_dummy_encryption(param,
606 						&fsopt->dummy_enc_policy);
607 		if (ret == -EINVAL) {
608 			warnfc(fc, "Value of option \"%s\" is unrecognized",
609 			       param->key);
610 		} else if (ret == -EEXIST) {
611 			warnfc(fc, "Conflicting test_dummy_encryption options");
612 			ret = -EINVAL;
613 		}
614 #else
615 		warnfc(fc,
616 		       "FS encryption not supported: test_dummy_encryption mount option ignored");
617 #endif
618 		break;
619 	default:
620 		BUG();
621 	}
622 	return 0;
623 
624 out_of_range:
625 	return invalfc(fc, "%s out of range", param->key);
626 }
627 
destroy_mount_options(struct ceph_mount_options * args)628 static void destroy_mount_options(struct ceph_mount_options *args)
629 {
630 	dout("destroy_mount_options %p\n", args);
631 	if (!args)
632 		return;
633 
634 	kfree(args->snapdir_name);
635 	kfree(args->mds_namespace);
636 	kfree(args->server_path);
637 	kfree(args->fscache_uniq);
638 	kfree(args->mon_addr);
639 	fscrypt_free_dummy_policy(&args->dummy_enc_policy);
640 	kfree(args);
641 }
642 
strcmp_null(const char * s1,const char * s2)643 static int strcmp_null(const char *s1, const char *s2)
644 {
645 	if (!s1 && !s2)
646 		return 0;
647 	if (s1 && !s2)
648 		return -1;
649 	if (!s1 && s2)
650 		return 1;
651 	return strcmp(s1, s2);
652 }
653 
compare_mount_options(struct ceph_mount_options * new_fsopt,struct ceph_options * new_opt,struct ceph_fs_client * fsc)654 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
655 				 struct ceph_options *new_opt,
656 				 struct ceph_fs_client *fsc)
657 {
658 	struct ceph_mount_options *fsopt1 = new_fsopt;
659 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
660 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
661 	int ret;
662 
663 	ret = memcmp(fsopt1, fsopt2, ofs);
664 	if (ret)
665 		return ret;
666 
667 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
668 	if (ret)
669 		return ret;
670 
671 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
672 	if (ret)
673 		return ret;
674 
675 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
676 	if (ret)
677 		return ret;
678 
679 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
680 	if (ret)
681 		return ret;
682 
683 	ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
684 	if (ret)
685 		return ret;
686 
687 	return ceph_compare_options(new_opt, fsc->client);
688 }
689 
690 /**
691  * ceph_show_options - Show mount options in /proc/mounts
692  * @m: seq_file to write to
693  * @root: root of that (sub)tree
694  */
ceph_show_options(struct seq_file * m,struct dentry * root)695 static int ceph_show_options(struct seq_file *m, struct dentry *root)
696 {
697 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(root->d_sb);
698 	struct ceph_mount_options *fsopt = fsc->mount_options;
699 	size_t pos;
700 	int ret;
701 
702 	/* a comma between MNT/MS and client options */
703 	seq_putc(m, ',');
704 	pos = m->count;
705 
706 	ret = ceph_print_client_options(m, fsc->client, false);
707 	if (ret)
708 		return ret;
709 
710 	/* retract our comma if no client options */
711 	if (m->count == pos)
712 		m->count--;
713 
714 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
715 		seq_puts(m, ",dirstat");
716 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
717 		seq_puts(m, ",rbytes");
718 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
719 		seq_puts(m, ",noasyncreaddir");
720 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
721 		seq_puts(m, ",nodcache");
722 	if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
723 		seq_puts(m, ",ino32");
724 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
725 		seq_show_option(m, "fsc", fsopt->fscache_uniq);
726 	}
727 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
728 		seq_puts(m, ",nopoolperm");
729 	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
730 		seq_puts(m, ",noquotadf");
731 
732 #ifdef CONFIG_CEPH_FS_POSIX_ACL
733 	if (root->d_sb->s_flags & SB_POSIXACL)
734 		seq_puts(m, ",acl");
735 	else
736 		seq_puts(m, ",noacl");
737 #endif
738 
739 	if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
740 		seq_puts(m, ",copyfrom");
741 
742 	/* dump mds_namespace when old device syntax is in use */
743 	if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
744 		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
745 
746 	if (fsopt->mon_addr)
747 		seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
748 
749 	if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
750 		seq_show_option(m, "recover_session", "clean");
751 
752 	if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
753 		seq_puts(m, ",wsync");
754 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
755 		seq_puts(m, ",nopagecache");
756 	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
757 		seq_puts(m, ",sparseread");
758 
759 	fscrypt_show_test_dummy_encryption(m, ',', root->d_sb);
760 
761 	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
762 		seq_printf(m, ",wsize=%u", fsopt->wsize);
763 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
764 		seq_printf(m, ",rsize=%u", fsopt->rsize);
765 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
766 		seq_printf(m, ",rasize=%u", fsopt->rasize);
767 	if (fsopt->congestion_kb != default_congestion_kb())
768 		seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
769 	if (fsopt->caps_max)
770 		seq_printf(m, ",caps_max=%d", fsopt->caps_max);
771 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
772 		seq_printf(m, ",caps_wanted_delay_min=%u",
773 			 fsopt->caps_wanted_delay_min);
774 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
775 		seq_printf(m, ",caps_wanted_delay_max=%u",
776 			   fsopt->caps_wanted_delay_max);
777 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
778 		seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
779 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
780 		seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
781 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
782 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
783 
784 	return 0;
785 }
786 
787 /*
788  * handle any mon messages the standard library doesn't understand.
789  * return error if we don't either.
790  */
extra_mon_dispatch(struct ceph_client * client,struct ceph_msg * msg)791 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
792 {
793 	struct ceph_fs_client *fsc = client->private;
794 	int type = le16_to_cpu(msg->hdr.type);
795 
796 	switch (type) {
797 	case CEPH_MSG_MDS_MAP:
798 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
799 		return 0;
800 	case CEPH_MSG_FS_MAP_USER:
801 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
802 		return 0;
803 	default:
804 		return -1;
805 	}
806 }
807 
808 /*
809  * create a new fs client
810  *
811  * Success or not, this function consumes @fsopt and @opt.
812  */
create_fs_client(struct ceph_mount_options * fsopt,struct ceph_options * opt)813 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
814 					struct ceph_options *opt)
815 {
816 	struct ceph_fs_client *fsc;
817 	int err;
818 
819 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
820 	if (!fsc) {
821 		err = -ENOMEM;
822 		goto fail;
823 	}
824 
825 	fsc->client = ceph_create_client(opt, fsc);
826 	if (IS_ERR(fsc->client)) {
827 		err = PTR_ERR(fsc->client);
828 		goto fail;
829 	}
830 	opt = NULL; /* fsc->client now owns this */
831 
832 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
833 	ceph_set_opt(fsc->client, ABORT_ON_FULL);
834 
835 	if (!fsopt->mds_namespace) {
836 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
837 				   0, true);
838 	} else {
839 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
840 				   0, false);
841 	}
842 
843 	fsc->mount_options = fsopt;
844 
845 	fsc->sb = NULL;
846 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
847 	fsc->filp_gen = 1;
848 	fsc->have_copy_from2 = true;
849 
850 	atomic_long_set(&fsc->writeback_count, 0);
851 	fsc->write_congested = false;
852 
853 	err = -ENOMEM;
854 	/*
855 	 * The number of concurrent works can be high but they don't need
856 	 * to be processed in parallel, limit concurrency.
857 	 */
858 	fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
859 	if (!fsc->inode_wq)
860 		goto fail_client;
861 	fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
862 	if (!fsc->cap_wq)
863 		goto fail_inode_wq;
864 
865 	hash_init(fsc->async_unlink_conflict);
866 	spin_lock_init(&fsc->async_unlink_conflict_lock);
867 
868 	spin_lock(&ceph_fsc_lock);
869 	list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
870 	spin_unlock(&ceph_fsc_lock);
871 
872 	return fsc;
873 
874 fail_inode_wq:
875 	destroy_workqueue(fsc->inode_wq);
876 fail_client:
877 	ceph_destroy_client(fsc->client);
878 fail:
879 	kfree(fsc);
880 	if (opt)
881 		ceph_destroy_options(opt);
882 	destroy_mount_options(fsopt);
883 	return ERR_PTR(err);
884 }
885 
flush_fs_workqueues(struct ceph_fs_client * fsc)886 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
887 {
888 	flush_workqueue(fsc->inode_wq);
889 	flush_workqueue(fsc->cap_wq);
890 }
891 
destroy_fs_client(struct ceph_fs_client * fsc)892 static void destroy_fs_client(struct ceph_fs_client *fsc)
893 {
894 	dout("destroy_fs_client %p\n", fsc);
895 
896 	spin_lock(&ceph_fsc_lock);
897 	list_del(&fsc->metric_wakeup);
898 	spin_unlock(&ceph_fsc_lock);
899 
900 	ceph_mdsc_destroy(fsc);
901 	destroy_workqueue(fsc->inode_wq);
902 	destroy_workqueue(fsc->cap_wq);
903 
904 	destroy_mount_options(fsc->mount_options);
905 
906 	ceph_destroy_client(fsc->client);
907 
908 	kfree(fsc);
909 	dout("destroy_fs_client %p done\n", fsc);
910 }
911 
912 /*
913  * caches
914  */
915 struct kmem_cache *ceph_inode_cachep;
916 struct kmem_cache *ceph_cap_cachep;
917 struct kmem_cache *ceph_cap_snap_cachep;
918 struct kmem_cache *ceph_cap_flush_cachep;
919 struct kmem_cache *ceph_dentry_cachep;
920 struct kmem_cache *ceph_file_cachep;
921 struct kmem_cache *ceph_dir_file_cachep;
922 struct kmem_cache *ceph_mds_request_cachep;
923 mempool_t *ceph_wb_pagevec_pool;
924 
ceph_inode_init_once(void * foo)925 static void ceph_inode_init_once(void *foo)
926 {
927 	struct ceph_inode_info *ci = foo;
928 	inode_init_once(&ci->netfs.inode);
929 }
930 
init_caches(void)931 static int __init init_caches(void)
932 {
933 	int error = -ENOMEM;
934 
935 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
936 				      sizeof(struct ceph_inode_info),
937 				      __alignof__(struct ceph_inode_info),
938 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
939 				      SLAB_ACCOUNT, ceph_inode_init_once);
940 	if (!ceph_inode_cachep)
941 		return -ENOMEM;
942 
943 	ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
944 	if (!ceph_cap_cachep)
945 		goto bad_cap;
946 	ceph_cap_snap_cachep = KMEM_CACHE(ceph_cap_snap, SLAB_MEM_SPREAD);
947 	if (!ceph_cap_snap_cachep)
948 		goto bad_cap_snap;
949 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
950 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
951 	if (!ceph_cap_flush_cachep)
952 		goto bad_cap_flush;
953 
954 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
955 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
956 	if (!ceph_dentry_cachep)
957 		goto bad_dentry;
958 
959 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
960 	if (!ceph_file_cachep)
961 		goto bad_file;
962 
963 	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
964 	if (!ceph_dir_file_cachep)
965 		goto bad_dir_file;
966 
967 	ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, SLAB_MEM_SPREAD);
968 	if (!ceph_mds_request_cachep)
969 		goto bad_mds_req;
970 
971 	ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10,
972 	    (CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT) * sizeof(struct page *));
973 	if (!ceph_wb_pagevec_pool)
974 		goto bad_pagevec_pool;
975 
976 	return 0;
977 
978 bad_pagevec_pool:
979 	kmem_cache_destroy(ceph_mds_request_cachep);
980 bad_mds_req:
981 	kmem_cache_destroy(ceph_dir_file_cachep);
982 bad_dir_file:
983 	kmem_cache_destroy(ceph_file_cachep);
984 bad_file:
985 	kmem_cache_destroy(ceph_dentry_cachep);
986 bad_dentry:
987 	kmem_cache_destroy(ceph_cap_flush_cachep);
988 bad_cap_flush:
989 	kmem_cache_destroy(ceph_cap_snap_cachep);
990 bad_cap_snap:
991 	kmem_cache_destroy(ceph_cap_cachep);
992 bad_cap:
993 	kmem_cache_destroy(ceph_inode_cachep);
994 	return error;
995 }
996 
destroy_caches(void)997 static void destroy_caches(void)
998 {
999 	/*
1000 	 * Make sure all delayed rcu free inodes are flushed before we
1001 	 * destroy cache.
1002 	 */
1003 	rcu_barrier();
1004 
1005 	kmem_cache_destroy(ceph_inode_cachep);
1006 	kmem_cache_destroy(ceph_cap_cachep);
1007 	kmem_cache_destroy(ceph_cap_snap_cachep);
1008 	kmem_cache_destroy(ceph_cap_flush_cachep);
1009 	kmem_cache_destroy(ceph_dentry_cachep);
1010 	kmem_cache_destroy(ceph_file_cachep);
1011 	kmem_cache_destroy(ceph_dir_file_cachep);
1012 	kmem_cache_destroy(ceph_mds_request_cachep);
1013 	mempool_destroy(ceph_wb_pagevec_pool);
1014 }
1015 
__ceph_umount_begin(struct ceph_fs_client * fsc)1016 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
1017 {
1018 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
1019 	ceph_mdsc_force_umount(fsc->mdsc);
1020 	fsc->filp_gen++; // invalidate open files
1021 }
1022 
1023 /*
1024  * ceph_umount_begin - initiate forced umount.  Tear down the
1025  * mount, skipping steps that may hang while waiting for server(s).
1026  */
ceph_umount_begin(struct super_block * sb)1027 void ceph_umount_begin(struct super_block *sb)
1028 {
1029 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1030 
1031 	dout("ceph_umount_begin - starting forced umount\n");
1032 	if (!fsc)
1033 		return;
1034 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
1035 	__ceph_umount_begin(fsc);
1036 }
1037 
1038 static const struct super_operations ceph_super_ops = {
1039 	.alloc_inode	= ceph_alloc_inode,
1040 	.free_inode	= ceph_free_inode,
1041 	.write_inode    = ceph_write_inode,
1042 	.drop_inode	= generic_delete_inode,
1043 	.evict_inode	= ceph_evict_inode,
1044 	.sync_fs        = ceph_sync_fs,
1045 	.put_super	= ceph_put_super,
1046 	.show_options   = ceph_show_options,
1047 	.statfs		= ceph_statfs,
1048 	.umount_begin   = ceph_umount_begin,
1049 };
1050 
1051 /*
1052  * Bootstrap mount by opening the root directory.  Note the mount
1053  * @started time from caller, and time out if this takes too long.
1054  */
open_root_dentry(struct ceph_fs_client * fsc,const char * path,unsigned long started)1055 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1056 				       const char *path,
1057 				       unsigned long started)
1058 {
1059 	struct ceph_mds_client *mdsc = fsc->mdsc;
1060 	struct ceph_mds_request *req = NULL;
1061 	int err;
1062 	struct dentry *root;
1063 
1064 	/* open dir */
1065 	dout("open_root_inode opening '%s'\n", path);
1066 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1067 	if (IS_ERR(req))
1068 		return ERR_CAST(req);
1069 	req->r_path1 = kstrdup(path, GFP_NOFS);
1070 	if (!req->r_path1) {
1071 		root = ERR_PTR(-ENOMEM);
1072 		goto out;
1073 	}
1074 
1075 	req->r_ino1.ino = CEPH_INO_ROOT;
1076 	req->r_ino1.snap = CEPH_NOSNAP;
1077 	req->r_started = started;
1078 	req->r_timeout = fsc->client->options->mount_timeout;
1079 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1080 	req->r_num_caps = 2;
1081 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1082 	if (err == 0) {
1083 		struct inode *inode = req->r_target_inode;
1084 		req->r_target_inode = NULL;
1085 		dout("open_root_inode success\n");
1086 		root = d_make_root(inode);
1087 		if (!root) {
1088 			root = ERR_PTR(-ENOMEM);
1089 			goto out;
1090 		}
1091 		dout("open_root_inode success, root dentry is %p\n", root);
1092 	} else {
1093 		root = ERR_PTR(err);
1094 	}
1095 out:
1096 	ceph_mdsc_put_request(req);
1097 	return root;
1098 }
1099 
1100 #ifdef CONFIG_FS_ENCRYPTION
ceph_apply_test_dummy_encryption(struct super_block * sb,struct fs_context * fc,struct ceph_mount_options * fsopt)1101 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1102 					    struct fs_context *fc,
1103 					    struct ceph_mount_options *fsopt)
1104 {
1105 	struct ceph_fs_client *fsc = sb->s_fs_info;
1106 
1107 	if (!fscrypt_is_dummy_policy_set(&fsopt->dummy_enc_policy))
1108 		return 0;
1109 
1110 	/* No changing encryption context on remount. */
1111 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE &&
1112 	    !fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1113 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1114 						 &fsc->fsc_dummy_enc_policy))
1115 			return 0;
1116 		errorfc(fc, "Can't set test_dummy_encryption on remount");
1117 		return -EINVAL;
1118 	}
1119 
1120 	/* Also make sure fsopt doesn't contain a conflicting value. */
1121 	if (fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1122 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1123 						 &fsc->fsc_dummy_enc_policy))
1124 			return 0;
1125 		errorfc(fc, "Conflicting test_dummy_encryption options");
1126 		return -EINVAL;
1127 	}
1128 
1129 	fsc->fsc_dummy_enc_policy = fsopt->dummy_enc_policy;
1130 	memset(&fsopt->dummy_enc_policy, 0, sizeof(fsopt->dummy_enc_policy));
1131 
1132 	warnfc(fc, "test_dummy_encryption mode enabled");
1133 	return 0;
1134 }
1135 #else
ceph_apply_test_dummy_encryption(struct super_block * sb,struct fs_context * fc,struct ceph_mount_options * fsopt)1136 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1137 					    struct fs_context *fc,
1138 					    struct ceph_mount_options *fsopt)
1139 {
1140 	return 0;
1141 }
1142 #endif
1143 
1144 /*
1145  * mount: join the ceph cluster, and open root directory.
1146  */
ceph_real_mount(struct ceph_fs_client * fsc,struct fs_context * fc)1147 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1148 				      struct fs_context *fc)
1149 {
1150 	int err;
1151 	unsigned long started = jiffies;  /* note the start time */
1152 	struct dentry *root;
1153 
1154 	dout("mount start %p\n", fsc);
1155 	mutex_lock(&fsc->client->mount_mutex);
1156 
1157 	if (!fsc->sb->s_root) {
1158 		const char *path = fsc->mount_options->server_path ?
1159 				     fsc->mount_options->server_path + 1 : "";
1160 
1161 		err = __ceph_open_session(fsc->client, started);
1162 		if (err < 0)
1163 			goto out;
1164 
1165 		/* setup fscache */
1166 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1167 			err = ceph_fscache_register_fs(fsc, fc);
1168 			if (err < 0)
1169 				goto out;
1170 		}
1171 
1172 		err = ceph_apply_test_dummy_encryption(fsc->sb, fc,
1173 						       fsc->mount_options);
1174 		if (err)
1175 			goto out;
1176 
1177 		dout("mount opening path '%s'\n", path);
1178 
1179 		ceph_fs_debugfs_init(fsc);
1180 
1181 		root = open_root_dentry(fsc, path, started);
1182 		if (IS_ERR(root)) {
1183 			err = PTR_ERR(root);
1184 			goto out;
1185 		}
1186 		fsc->sb->s_root = dget(root);
1187 	} else {
1188 		root = dget(fsc->sb->s_root);
1189 	}
1190 
1191 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1192 	dout("mount success\n");
1193 	mutex_unlock(&fsc->client->mount_mutex);
1194 	return root;
1195 
1196 out:
1197 	mutex_unlock(&fsc->client->mount_mutex);
1198 	ceph_fscrypt_free_dummy_policy(fsc);
1199 	return ERR_PTR(err);
1200 }
1201 
ceph_set_super(struct super_block * s,struct fs_context * fc)1202 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1203 {
1204 	struct ceph_fs_client *fsc = s->s_fs_info;
1205 	int ret;
1206 
1207 	dout("set_super %p\n", s);
1208 
1209 	s->s_maxbytes = MAX_LFS_FILESIZE;
1210 
1211 	s->s_xattr = ceph_xattr_handlers;
1212 	fsc->sb = s;
1213 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1214 
1215 	s->s_op = &ceph_super_ops;
1216 	s->s_d_op = &ceph_dentry_ops;
1217 	s->s_export_op = &ceph_export_ops;
1218 
1219 	s->s_time_gran = 1;
1220 	s->s_time_min = 0;
1221 	s->s_time_max = U32_MAX;
1222 	s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1223 
1224 	ceph_fscrypt_set_ops(s);
1225 
1226 	ret = set_anon_super_fc(s, fc);
1227 	if (ret != 0)
1228 		fsc->sb = NULL;
1229 	return ret;
1230 }
1231 
1232 /*
1233  * share superblock if same fs AND options
1234  */
ceph_compare_super(struct super_block * sb,struct fs_context * fc)1235 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1236 {
1237 	struct ceph_fs_client *new = fc->s_fs_info;
1238 	struct ceph_mount_options *fsopt = new->mount_options;
1239 	struct ceph_options *opt = new->client->options;
1240 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1241 
1242 	dout("ceph_compare_super %p\n", sb);
1243 
1244 	if (compare_mount_options(fsopt, opt, fsc)) {
1245 		dout("monitor(s)/mount options don't match\n");
1246 		return 0;
1247 	}
1248 	if ((opt->flags & CEPH_OPT_FSID) &&
1249 	    ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1250 		dout("fsid doesn't match\n");
1251 		return 0;
1252 	}
1253 	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1254 		dout("flags differ\n");
1255 		return 0;
1256 	}
1257 
1258 	if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1259 		dout("client is blocklisted (and CLEANRECOVER is not set)\n");
1260 		return 0;
1261 	}
1262 
1263 	if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1264 		dout("client has been forcibly unmounted\n");
1265 		return 0;
1266 	}
1267 
1268 	return 1;
1269 }
1270 
1271 /*
1272  * construct our own bdi so we can control readahead, etc.
1273  */
1274 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1275 
ceph_setup_bdi(struct super_block * sb,struct ceph_fs_client * fsc)1276 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1277 {
1278 	int err;
1279 
1280 	err = super_setup_bdi_name(sb, "ceph-%ld",
1281 				   atomic_long_inc_return(&bdi_seq));
1282 	if (err)
1283 		return err;
1284 
1285 	/* set ra_pages based on rasize mount option? */
1286 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1287 
1288 	/* set io_pages based on max osd read size */
1289 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1290 
1291 	return 0;
1292 }
1293 
ceph_get_tree(struct fs_context * fc)1294 static int ceph_get_tree(struct fs_context *fc)
1295 {
1296 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1297 	struct ceph_mount_options *fsopt = pctx->opts;
1298 	struct super_block *sb;
1299 	struct ceph_fs_client *fsc;
1300 	struct dentry *res;
1301 	int (*compare_super)(struct super_block *, struct fs_context *) =
1302 		ceph_compare_super;
1303 	int err;
1304 
1305 	dout("ceph_get_tree\n");
1306 
1307 	if (!fc->source)
1308 		return invalfc(fc, "No source");
1309 	if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1310 		return invalfc(fc, "No monitor address");
1311 
1312 	/* create client (which we may/may not use) */
1313 	fsc = create_fs_client(pctx->opts, pctx->copts);
1314 	pctx->opts = NULL;
1315 	pctx->copts = NULL;
1316 	if (IS_ERR(fsc)) {
1317 		err = PTR_ERR(fsc);
1318 		goto out_final;
1319 	}
1320 
1321 	err = ceph_mdsc_init(fsc);
1322 	if (err < 0)
1323 		goto out;
1324 
1325 	if (ceph_test_opt(fsc->client, NOSHARE))
1326 		compare_super = NULL;
1327 
1328 	fc->s_fs_info = fsc;
1329 	sb = sget_fc(fc, compare_super, ceph_set_super);
1330 	fc->s_fs_info = NULL;
1331 	if (IS_ERR(sb)) {
1332 		err = PTR_ERR(sb);
1333 		goto out;
1334 	}
1335 
1336 	if (ceph_sb_to_fs_client(sb) != fsc) {
1337 		destroy_fs_client(fsc);
1338 		fsc = ceph_sb_to_fs_client(sb);
1339 		dout("get_sb got existing client %p\n", fsc);
1340 	} else {
1341 		dout("get_sb using new client %p\n", fsc);
1342 		err = ceph_setup_bdi(sb, fsc);
1343 		if (err < 0)
1344 			goto out_splat;
1345 	}
1346 
1347 	res = ceph_real_mount(fsc, fc);
1348 	if (IS_ERR(res)) {
1349 		err = PTR_ERR(res);
1350 		goto out_splat;
1351 	}
1352 	dout("root %p inode %p ino %llx.%llx\n", res,
1353 	     d_inode(res), ceph_vinop(d_inode(res)));
1354 	fc->root = fsc->sb->s_root;
1355 	return 0;
1356 
1357 out_splat:
1358 	if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1359 		pr_info("No mds server is up or the cluster is laggy\n");
1360 		err = -EHOSTUNREACH;
1361 	}
1362 
1363 	ceph_mdsc_close_sessions(fsc->mdsc);
1364 	deactivate_locked_super(sb);
1365 	goto out_final;
1366 
1367 out:
1368 	destroy_fs_client(fsc);
1369 out_final:
1370 	dout("ceph_get_tree fail %d\n", err);
1371 	return err;
1372 }
1373 
ceph_free_fc(struct fs_context * fc)1374 static void ceph_free_fc(struct fs_context *fc)
1375 {
1376 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1377 
1378 	if (pctx) {
1379 		destroy_mount_options(pctx->opts);
1380 		ceph_destroy_options(pctx->copts);
1381 		kfree(pctx);
1382 	}
1383 }
1384 
ceph_reconfigure_fc(struct fs_context * fc)1385 static int ceph_reconfigure_fc(struct fs_context *fc)
1386 {
1387 	int err;
1388 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1389 	struct ceph_mount_options *fsopt = pctx->opts;
1390 	struct super_block *sb = fc->root->d_sb;
1391 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1392 
1393 	err = ceph_apply_test_dummy_encryption(sb, fc, fsopt);
1394 	if (err)
1395 		return err;
1396 
1397 	if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1398 		ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1399 	else
1400 		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1401 
1402 	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
1403 		ceph_set_mount_opt(fsc, SPARSEREAD);
1404 	else
1405 		ceph_clear_mount_opt(fsc, SPARSEREAD);
1406 
1407 	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1408 		kfree(fsc->mount_options->mon_addr);
1409 		fsc->mount_options->mon_addr = fsopt->mon_addr;
1410 		fsopt->mon_addr = NULL;
1411 		pr_notice("ceph: monitor addresses recorded, but not used for reconnection");
1412 	}
1413 
1414 	sync_filesystem(sb);
1415 	return 0;
1416 }
1417 
1418 static const struct fs_context_operations ceph_context_ops = {
1419 	.free		= ceph_free_fc,
1420 	.parse_param	= ceph_parse_mount_param,
1421 	.get_tree	= ceph_get_tree,
1422 	.reconfigure	= ceph_reconfigure_fc,
1423 };
1424 
1425 /*
1426  * Set up the filesystem mount context.
1427  */
ceph_init_fs_context(struct fs_context * fc)1428 static int ceph_init_fs_context(struct fs_context *fc)
1429 {
1430 	struct ceph_parse_opts_ctx *pctx;
1431 	struct ceph_mount_options *fsopt;
1432 
1433 	pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1434 	if (!pctx)
1435 		return -ENOMEM;
1436 
1437 	pctx->copts = ceph_alloc_options();
1438 	if (!pctx->copts)
1439 		goto nomem;
1440 
1441 	pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1442 	if (!pctx->opts)
1443 		goto nomem;
1444 
1445 	fsopt = pctx->opts;
1446 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1447 
1448 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1449 	fsopt->rsize = CEPH_MAX_READ_SIZE;
1450 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
1451 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1452 	if (!fsopt->snapdir_name)
1453 		goto nomem;
1454 
1455 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1456 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1457 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1458 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1459 	fsopt->congestion_kb = default_congestion_kb();
1460 
1461 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1462 	fc->sb_flags |= SB_POSIXACL;
1463 #endif
1464 
1465 	fc->fs_private = pctx;
1466 	fc->ops = &ceph_context_ops;
1467 	return 0;
1468 
1469 nomem:
1470 	destroy_mount_options(pctx->opts);
1471 	ceph_destroy_options(pctx->copts);
1472 	kfree(pctx);
1473 	return -ENOMEM;
1474 }
1475 
1476 /*
1477  * Return true if it successfully increases the blocker counter,
1478  * or false if the mdsc is in stopping and flushed state.
1479  */
__inc_stopping_blocker(struct ceph_mds_client * mdsc)1480 static bool __inc_stopping_blocker(struct ceph_mds_client *mdsc)
1481 {
1482 	spin_lock(&mdsc->stopping_lock);
1483 	if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING) {
1484 		spin_unlock(&mdsc->stopping_lock);
1485 		return false;
1486 	}
1487 	atomic_inc(&mdsc->stopping_blockers);
1488 	spin_unlock(&mdsc->stopping_lock);
1489 	return true;
1490 }
1491 
__dec_stopping_blocker(struct ceph_mds_client * mdsc)1492 static void __dec_stopping_blocker(struct ceph_mds_client *mdsc)
1493 {
1494 	spin_lock(&mdsc->stopping_lock);
1495 	if (!atomic_dec_return(&mdsc->stopping_blockers) &&
1496 	    mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING)
1497 		complete_all(&mdsc->stopping_waiter);
1498 	spin_unlock(&mdsc->stopping_lock);
1499 }
1500 
1501 /* For metadata IO requests */
ceph_inc_mds_stopping_blocker(struct ceph_mds_client * mdsc,struct ceph_mds_session * session)1502 bool ceph_inc_mds_stopping_blocker(struct ceph_mds_client *mdsc,
1503 				   struct ceph_mds_session *session)
1504 {
1505 	mutex_lock(&session->s_mutex);
1506 	inc_session_sequence(session);
1507 	mutex_unlock(&session->s_mutex);
1508 
1509 	return __inc_stopping_blocker(mdsc);
1510 }
1511 
ceph_dec_mds_stopping_blocker(struct ceph_mds_client * mdsc)1512 void ceph_dec_mds_stopping_blocker(struct ceph_mds_client *mdsc)
1513 {
1514 	__dec_stopping_blocker(mdsc);
1515 }
1516 
1517 /* For data IO requests */
ceph_inc_osd_stopping_blocker(struct ceph_mds_client * mdsc)1518 bool ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1519 {
1520 	return __inc_stopping_blocker(mdsc);
1521 }
1522 
ceph_dec_osd_stopping_blocker(struct ceph_mds_client * mdsc)1523 void ceph_dec_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1524 {
1525 	__dec_stopping_blocker(mdsc);
1526 }
1527 
ceph_kill_sb(struct super_block * s)1528 static void ceph_kill_sb(struct super_block *s)
1529 {
1530 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
1531 	struct ceph_mds_client *mdsc = fsc->mdsc;
1532 	bool wait;
1533 
1534 	dout("kill_sb %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("umount timed out, %ld\n", timeleft);
1566 		else if (timeleft < 0) /* killed */
1567 			pr_warn("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,
1587 };
1588 MODULE_ALIAS_FS("ceph");
1589 
ceph_force_reconnect(struct super_block * sb)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 
init_ceph(void)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 
exit_ceph(void)1639 static void __exit exit_ceph(void)
1640 {
1641 	dout("exit_ceph\n");
1642 	unregister_filesystem(&ceph_fs_type);
1643 	destroy_caches();
1644 }
1645 
param_set_metrics(const char * val,const struct kernel_param * kp)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 module_init(init_ceph);
1686 module_exit(exit_ceph);
1687 
1688 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1689 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1690 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1691 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1692 MODULE_LICENSE("GPL");
1693