xref: /linux/fs/ceph/super.c (revision ac2dc6d57425ffa9629941d7c9d7c0e51082cb5a)
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 #include "subvolume_metrics.h"
25 
26 #include <linux/ceph/ceph_features.h>
27 #include <linux/ceph/decode.h>
28 #include <linux/ceph/mon_client.h>
29 #include <linux/ceph/auth.h>
30 #include <linux/ceph/debugfs.h>
31 
32 #include <uapi/linux/magic.h>
33 
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/ceph.h>
36 
37 static DEFINE_SPINLOCK(ceph_fsc_lock);
38 static LIST_HEAD(ceph_fsc_list);
39 
40 /*
41  * Ceph superblock operations
42  *
43  * Handle the basics of mounting, unmounting.
44  */
45 
46 /*
47  * super ops
48  */
ceph_put_super(struct super_block * s)49 static void ceph_put_super(struct super_block *s)
50 {
51 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
52 
53 	doutc(fsc->client, "begin\n");
54 	ceph_fscrypt_free_dummy_policy(fsc);
55 	ceph_mdsc_close_sessions(fsc->mdsc);
56 	doutc(fsc->client, "done\n");
57 }
58 
ceph_statfs(struct dentry * dentry,struct kstatfs * buf)59 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
60 {
61 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(d_inode(dentry));
62 	struct ceph_mon_client *monc = &fsc->client->monc;
63 	struct ceph_statfs st;
64 	int i, err;
65 	u64 data_pool;
66 
67 	doutc(fsc->client, "begin\n");
68 	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
69 		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
70 	} else {
71 		data_pool = CEPH_NOPOOL;
72 	}
73 
74 	err = ceph_monc_do_statfs(monc, data_pool, &st);
75 	if (err < 0)
76 		return err;
77 
78 	/* fill in kstatfs */
79 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
80 
81 	/*
82 	 * Express utilization in terms of large blocks to avoid
83 	 * overflow on 32-bit machines.
84 	 */
85 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
86 
87 	/*
88 	 * By default use root quota for stats; fallback to overall filesystem
89 	 * usage if using 'noquotadf' mount option or if the root dir doesn't
90 	 * have max_bytes quota set.
91 	 */
92 	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
93 	    !ceph_quota_update_statfs(fsc, buf)) {
94 		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
95 		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
96 		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
97 	}
98 
99 	/*
100 	 * NOTE: for the time being, we make bsize == frsize to humor
101 	 * not-yet-ancient versions of glibc that are broken.
102 	 * Someday, we will probably want to report a real block
103 	 * size...  whatever that may mean for a network file system!
104 	 */
105 	buf->f_bsize = buf->f_frsize;
106 
107 	buf->f_files = le64_to_cpu(st.num_objects);
108 	buf->f_ffree = -1;
109 	buf->f_namelen = NAME_MAX;
110 
111 	/* Must convert the fsid, for consistent values across arches */
112 	buf->f_fsid.val[0] = 0;
113 	mutex_lock(&monc->mutex);
114 	for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
115 		buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
116 	mutex_unlock(&monc->mutex);
117 
118 	/* fold the fs_cluster_id into the upper bits */
119 	buf->f_fsid.val[1] = monc->fs_cluster_id;
120 
121 	doutc(fsc->client, "done\n");
122 	return 0;
123 }
124 
ceph_sync_fs(struct super_block * sb,int wait)125 static int ceph_sync_fs(struct super_block *sb, int wait)
126 {
127 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
128 	struct ceph_client *cl = fsc->client;
129 
130 	if (!wait) {
131 		doutc(cl, "(non-blocking)\n");
132 		ceph_flush_dirty_caps(fsc->mdsc);
133 		ceph_flush_cap_releases(fsc->mdsc);
134 		doutc(cl, "(non-blocking) done\n");
135 		return 0;
136 	}
137 
138 	doutc(cl, "(blocking)\n");
139 	ceph_osdc_sync(&fsc->client->osdc);
140 	ceph_mdsc_sync(fsc->mdsc);
141 	doutc(cl, "(blocking) done\n");
142 	return 0;
143 }
144 
145 /*
146  * mount options
147  */
148 enum {
149 	Opt_wsize,
150 	Opt_rsize,
151 	Opt_rasize,
152 	Opt_caps_wanted_delay_min,
153 	Opt_caps_wanted_delay_max,
154 	Opt_caps_max,
155 	Opt_readdir_max_entries,
156 	Opt_readdir_max_bytes,
157 	Opt_congestion_kb,
158 	/* int args above */
159 	Opt_snapdirname,
160 	Opt_mds_namespace,
161 	Opt_recover_session,
162 	Opt_source,
163 	Opt_mon_addr,
164 	Opt_test_dummy_encryption,
165 	/* string args above */
166 	Opt_dirstat,
167 	Opt_rbytes,
168 	Opt_asyncreaddir,
169 	Opt_dcache,
170 	Opt_ino32,
171 	Opt_fscache,
172 	Opt_poolperm,
173 	Opt_require_active_mds,
174 	Opt_acl,
175 	Opt_quotadf,
176 	Opt_copyfrom,
177 	Opt_wsync,
178 	Opt_pagecache,
179 	Opt_sparseread,
180 };
181 
182 enum ceph_recover_session_mode {
183 	ceph_recover_session_no,
184 	ceph_recover_session_clean
185 };
186 
187 static const struct constant_table ceph_param_recover[] = {
188 	{ "no",		ceph_recover_session_no },
189 	{ "clean",	ceph_recover_session_clean },
190 	{}
191 };
192 
193 static const struct fs_parameter_spec ceph_mount_parameters[] = {
194 	fsparam_flag_no ("acl",				Opt_acl),
195 	fsparam_flag_no ("asyncreaddir",		Opt_asyncreaddir),
196 	fsparam_s32	("caps_max",			Opt_caps_max),
197 	fsparam_u32	("caps_wanted_delay_max",	Opt_caps_wanted_delay_max),
198 	fsparam_u32	("caps_wanted_delay_min",	Opt_caps_wanted_delay_min),
199 	fsparam_u32	("write_congestion_kb",		Opt_congestion_kb),
200 	fsparam_flag_no ("copyfrom",			Opt_copyfrom),
201 	fsparam_flag_no ("dcache",			Opt_dcache),
202 	fsparam_flag_no ("dirstat",			Opt_dirstat),
203 	fsparam_flag_no	("fsc",				Opt_fscache), // fsc|nofsc
204 	fsparam_string	("fsc",				Opt_fscache), // fsc=...
205 	fsparam_flag_no ("ino32",			Opt_ino32),
206 	fsparam_string	("mds_namespace",		Opt_mds_namespace),
207 	fsparam_string	("mon_addr",			Opt_mon_addr),
208 	fsparam_flag_no ("poolperm",			Opt_poolperm),
209 	fsparam_flag_no ("quotadf",			Opt_quotadf),
210 	fsparam_u32	("rasize",			Opt_rasize),
211 	fsparam_flag_no ("rbytes",			Opt_rbytes),
212 	fsparam_u32	("readdir_max_bytes",		Opt_readdir_max_bytes),
213 	fsparam_u32	("readdir_max_entries",		Opt_readdir_max_entries),
214 	fsparam_enum	("recover_session",		Opt_recover_session, ceph_param_recover),
215 	fsparam_flag_no ("require_active_mds",		Opt_require_active_mds),
216 	fsparam_u32	("rsize",			Opt_rsize),
217 	fsparam_string	("snapdirname",			Opt_snapdirname),
218 	fsparam_string	("source",			Opt_source),
219 	fsparam_flag	("test_dummy_encryption",	Opt_test_dummy_encryption),
220 	fsparam_string	("test_dummy_encryption",	Opt_test_dummy_encryption),
221 	fsparam_u32	("wsize",			Opt_wsize),
222 	fsparam_flag_no	("wsync",			Opt_wsync),
223 	fsparam_flag_no	("pagecache",			Opt_pagecache),
224 	fsparam_flag_no	("sparseread",			Opt_sparseread),
225 	{}
226 };
227 
228 struct ceph_parse_opts_ctx {
229 	struct ceph_options		*copts;
230 	struct ceph_mount_options	*opts;
231 };
232 
233 /*
234  * Remove adjacent slashes and then the trailing slash, unless it is
235  * the only remaining character.
236  *
237  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
238  */
canonicalize_path(char * path)239 static void canonicalize_path(char *path)
240 {
241 	int i, j = 0;
242 
243 	for (i = 0; path[i] != '\0'; i++) {
244 		if (path[i] != '/' || j < 1 || path[j - 1] != '/')
245 			path[j++] = path[i];
246 	}
247 
248 	if (j > 1 && path[j - 1] == '/')
249 		j--;
250 	path[j] = '\0';
251 }
252 
ceph_parse_old_source(const char * dev_name,const char * dev_name_end,struct fs_context * fc)253 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
254 				 struct fs_context *fc)
255 {
256 	int r;
257 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
258 	struct ceph_mount_options *fsopt = pctx->opts;
259 
260 	if (*dev_name_end != ':')
261 		return invalfc(fc, "separator ':' missing in source");
262 
263 	r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
264 			       pctx->copts, fc->log.log, ',');
265 	if (r)
266 		return r;
267 
268 	fsopt->new_dev_syntax = false;
269 	return 0;
270 }
271 
ceph_parse_new_source(const char * dev_name,const char * dev_name_end,struct fs_context * fc)272 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
273 				 struct fs_context *fc)
274 {
275 	size_t len;
276 	struct ceph_fsid fsid;
277 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
278 	struct ceph_options *opts = pctx->copts;
279 	struct ceph_mount_options *fsopt = pctx->opts;
280 	const char *name_start = dev_name;
281 	const char *fsid_start, *fs_name_start;
282 
283 	if (*dev_name_end != '=') {
284 		dout("separator '=' missing in source");
285 		return -EINVAL;
286 	}
287 
288 	fsid_start = strchr(dev_name, '@');
289 	if (!fsid_start)
290 		return invalfc(fc, "missing cluster fsid");
291 	len = fsid_start - name_start;
292 	kfree(opts->name);
293 	opts->name = kstrndup(name_start, len, GFP_KERNEL);
294 	if (!opts->name)
295 		return -ENOMEM;
296 	dout("using %s entity name", opts->name);
297 
298 	++fsid_start; /* start of cluster fsid */
299 	fs_name_start = strchr(fsid_start, '.');
300 	if (!fs_name_start)
301 		return invalfc(fc, "missing file system name");
302 
303 	if (ceph_parse_fsid(fsid_start, &fsid))
304 		return invalfc(fc, "Invalid FSID");
305 
306 	++fs_name_start; /* start of file system name */
307 	len = dev_name_end - fs_name_start;
308 
309 	if (!namespace_equals(fsopt, fs_name_start, len))
310 		return invalfc(fc, "Mismatching mds_namespace");
311 	kfree(fsopt->mds_namespace);
312 	fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
313 	if (!fsopt->mds_namespace)
314 		return -ENOMEM;
315 	dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
316 
317 	fsopt->new_dev_syntax = true;
318 	return 0;
319 }
320 
321 /*
322  * Parse the source parameter for new device format. Distinguish the device
323  * spec from the path. Try parsing new device format and fallback to old
324  * format if needed.
325  *
326  * New device syntax will looks like:
327  *     <device_spec>=/<path>
328  * where
329  *     <device_spec> is name@fsid.fsname
330  *     <path> is optional, but if present must begin with '/'
331  * (monitor addresses are passed via mount option)
332  *
333  * Old device syntax is:
334  *     <server_spec>[,<server_spec>...]:[<path>]
335  * where
336  *     <server_spec> is <ip>[:<port>]
337  *     <path> is optional, but if present must begin with '/'
338  */
ceph_parse_source(struct fs_parameter * param,struct fs_context * fc)339 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
340 {
341 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
342 	struct ceph_mount_options *fsopt = pctx->opts;
343 	char *dev_name = param->string, *dev_name_end;
344 	int ret;
345 
346 	dout("'%s'\n", dev_name);
347 	if (!dev_name || !*dev_name)
348 		return invalfc(fc, "Empty source");
349 
350 	dev_name_end = strchr(dev_name, '/');
351 	if (dev_name_end) {
352 		/*
353 		 * The server_path will include the whole chars from userland
354 		 * including the leading '/'.
355 		 */
356 		kfree(fsopt->server_path);
357 		fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
358 		if (!fsopt->server_path)
359 			return -ENOMEM;
360 
361 		canonicalize_path(fsopt->server_path);
362 	} else {
363 		dev_name_end = dev_name + strlen(dev_name);
364 	}
365 
366 	dev_name_end--;		/* back up to separator */
367 	if (dev_name_end < dev_name)
368 		return invalfc(fc, "Path missing in source");
369 
370 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
371 	if (fsopt->server_path)
372 		dout("server path '%s'\n", fsopt->server_path);
373 
374 	dout("trying new device syntax");
375 	ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
376 	if (ret) {
377 		if (ret != -EINVAL)
378 			return ret;
379 		dout("trying old device syntax");
380 		ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
381 		if (ret)
382 			return ret;
383 	}
384 
385 	fc->source = param->string;
386 	param->string = NULL;
387 	return 0;
388 }
389 
ceph_parse_mon_addr(struct fs_parameter * param,struct fs_context * fc)390 static int ceph_parse_mon_addr(struct fs_parameter *param,
391 			       struct fs_context *fc)
392 {
393 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
394 	struct ceph_mount_options *fsopt = pctx->opts;
395 
396 	kfree(fsopt->mon_addr);
397 	fsopt->mon_addr = param->string;
398 	param->string = NULL;
399 
400 	return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
401 				  pctx->copts, fc->log.log, '/');
402 }
403 
ceph_parse_mount_param(struct fs_context * fc,struct fs_parameter * param)404 static int ceph_parse_mount_param(struct fs_context *fc,
405 				  struct fs_parameter *param)
406 {
407 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
408 	struct ceph_mount_options *fsopt = pctx->opts;
409 	struct fs_parse_result result;
410 	unsigned int mode;
411 	int token, ret;
412 
413 	ret = ceph_parse_param(param, pctx->copts, fc->log.log);
414 	if (ret != -ENOPARAM)
415 		return ret;
416 
417 	token = fs_parse(fc, ceph_mount_parameters, param, &result);
418 	dout("%s: fs_parse '%s' token %d\n",__func__, param->key, token);
419 	if (token < 0)
420 		return token;
421 
422 	switch (token) {
423 	case Opt_snapdirname:
424 		if (strlen(param->string) > NAME_MAX)
425 			return invalfc(fc, "snapdirname too long");
426 		kfree(fsopt->snapdir_name);
427 		fsopt->snapdir_name = param->string;
428 		param->string = NULL;
429 		break;
430 	case Opt_mds_namespace:
431 		if (!namespace_equals(fsopt, param->string, strlen(param->string)))
432 			return invalfc(fc, "Mismatching mds_namespace");
433 		kfree(fsopt->mds_namespace);
434 		fsopt->mds_namespace = param->string;
435 		param->string = NULL;
436 		break;
437 	case Opt_recover_session:
438 		mode = result.uint_32;
439 		if (mode == ceph_recover_session_no)
440 			fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
441 		else if (mode == ceph_recover_session_clean)
442 			fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
443 		else
444 			BUG();
445 		break;
446 	case Opt_source:
447 		if (fc->source)
448 			return invalfc(fc, "Multiple sources specified");
449 		return ceph_parse_source(param, fc);
450 	case Opt_mon_addr:
451 		return ceph_parse_mon_addr(param, fc);
452 	case Opt_wsize:
453 		if (result.uint_32 < PAGE_SIZE ||
454 		    result.uint_32 > CEPH_MAX_WRITE_SIZE)
455 			goto out_of_range;
456 		fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
457 		break;
458 	case Opt_rsize:
459 		if (result.uint_32 < PAGE_SIZE ||
460 		    result.uint_32 > CEPH_MAX_READ_SIZE)
461 			goto out_of_range;
462 		fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
463 		break;
464 	case Opt_rasize:
465 		fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
466 		break;
467 	case Opt_caps_wanted_delay_min:
468 		if (result.uint_32 < 1)
469 			goto out_of_range;
470 		fsopt->caps_wanted_delay_min = result.uint_32;
471 		break;
472 	case Opt_caps_wanted_delay_max:
473 		if (result.uint_32 < 1)
474 			goto out_of_range;
475 		fsopt->caps_wanted_delay_max = result.uint_32;
476 		break;
477 	case Opt_caps_max:
478 		if (result.int_32 < 0)
479 			goto out_of_range;
480 		fsopt->caps_max = result.int_32;
481 		break;
482 	case Opt_readdir_max_entries:
483 		if (result.uint_32 < 1)
484 			goto out_of_range;
485 		fsopt->max_readdir = result.uint_32;
486 		break;
487 	case Opt_readdir_max_bytes:
488 		if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
489 			goto out_of_range;
490 		fsopt->max_readdir_bytes = result.uint_32;
491 		break;
492 	case Opt_congestion_kb:
493 		if (result.uint_32 < 1024) /* at least 1M */
494 			goto out_of_range;
495 		fsopt->congestion_kb = result.uint_32;
496 		break;
497 	case Opt_dirstat:
498 		if (!result.negated)
499 			fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
500 		else
501 			fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
502 		break;
503 	case Opt_rbytes:
504 		if (!result.negated)
505 			fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
506 		else
507 			fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
508 		break;
509 	case Opt_asyncreaddir:
510 		if (!result.negated)
511 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
512 		else
513 			fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
514 		break;
515 	case Opt_dcache:
516 		if (!result.negated)
517 			fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
518 		else
519 			fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
520 		break;
521 	case Opt_ino32:
522 		if (!result.negated)
523 			fsopt->flags |= CEPH_MOUNT_OPT_INO32;
524 		else
525 			fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
526 		break;
527 
528 	case Opt_fscache:
529 #ifdef CONFIG_CEPH_FSCACHE
530 		kfree(fsopt->fscache_uniq);
531 		fsopt->fscache_uniq = NULL;
532 		if (result.negated) {
533 			fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
534 		} else {
535 			fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
536 			fsopt->fscache_uniq = param->string;
537 			param->string = NULL;
538 		}
539 		break;
540 #else
541 		return invalfc(fc, "fscache support is disabled");
542 #endif
543 	case Opt_poolperm:
544 		if (!result.negated)
545 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
546 		else
547 			fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
548 		break;
549 	case Opt_require_active_mds:
550 		if (!result.negated)
551 			fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
552 		else
553 			fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
554 		break;
555 	case Opt_quotadf:
556 		if (!result.negated)
557 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
558 		else
559 			fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
560 		break;
561 	case Opt_copyfrom:
562 		if (!result.negated)
563 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
564 		else
565 			fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
566 		break;
567 	case Opt_acl:
568 		if (!result.negated) {
569 #ifdef CONFIG_CEPH_FS_POSIX_ACL
570 			fc->sb_flags |= SB_POSIXACL;
571 #else
572 			return invalfc(fc, "POSIX ACL support is disabled");
573 #endif
574 		} else {
575 			fc->sb_flags &= ~SB_POSIXACL;
576 		}
577 		break;
578 	case Opt_wsync:
579 		if (!result.negated)
580 			fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
581 		else
582 			fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
583 		break;
584 	case Opt_pagecache:
585 		if (result.negated)
586 			fsopt->flags |= CEPH_MOUNT_OPT_NOPAGECACHE;
587 		else
588 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
589 		break;
590 	case Opt_sparseread:
591 		if (result.negated)
592 			fsopt->flags &= ~CEPH_MOUNT_OPT_SPARSEREAD;
593 		else
594 			fsopt->flags |= CEPH_MOUNT_OPT_SPARSEREAD;
595 		break;
596 	case Opt_test_dummy_encryption:
597 #ifdef CONFIG_FS_ENCRYPTION
598 		fscrypt_free_dummy_policy(&fsopt->dummy_enc_policy);
599 		ret = fscrypt_parse_test_dummy_encryption(param,
600 						&fsopt->dummy_enc_policy);
601 		if (ret == -EINVAL) {
602 			warnfc(fc, "Value of option \"%s\" is unrecognized",
603 			       param->key);
604 		} else if (ret == -EEXIST) {
605 			warnfc(fc, "Conflicting test_dummy_encryption options");
606 			ret = -EINVAL;
607 		}
608 #else
609 		warnfc(fc,
610 		       "FS encryption not supported: test_dummy_encryption mount option ignored");
611 #endif
612 		break;
613 	default:
614 		BUG();
615 	}
616 	return 0;
617 
618 out_of_range:
619 	return invalfc(fc, "%s out of range", param->key);
620 }
621 
destroy_mount_options(struct ceph_mount_options * args)622 static void destroy_mount_options(struct ceph_mount_options *args)
623 {
624 	dout("destroy_mount_options %p\n", args);
625 	if (!args)
626 		return;
627 
628 	kfree(args->snapdir_name);
629 	kfree(args->mds_namespace);
630 	kfree(args->server_path);
631 	kfree(args->fscache_uniq);
632 	kfree(args->mon_addr);
633 	fscrypt_free_dummy_policy(&args->dummy_enc_policy);
634 	kfree(args);
635 }
636 
strcmp_null(const char * s1,const char * s2)637 static int strcmp_null(const char *s1, const char *s2)
638 {
639 	if (!s1 && !s2)
640 		return 0;
641 	if (s1 && !s2)
642 		return -1;
643 	if (!s1 && s2)
644 		return 1;
645 	return strcmp(s1, s2);
646 }
647 
compare_mount_options(struct ceph_mount_options * new_fsopt,struct ceph_options * new_opt,struct ceph_fs_client * fsc)648 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
649 				 struct ceph_options *new_opt,
650 				 struct ceph_fs_client *fsc)
651 {
652 	struct ceph_mount_options *fsopt1 = new_fsopt;
653 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
654 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
655 	int ret;
656 
657 	ret = memcmp(fsopt1, fsopt2, ofs);
658 	if (ret)
659 		return ret;
660 
661 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
662 	if (ret)
663 		return ret;
664 
665 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
666 	if (ret)
667 		return ret;
668 
669 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
670 	if (ret)
671 		return ret;
672 
673 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
674 	if (ret)
675 		return ret;
676 
677 	ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
678 	if (ret)
679 		return ret;
680 
681 	return ceph_compare_options(new_opt, fsc->client);
682 }
683 
684 /**
685  * ceph_show_options - Show mount options in /proc/mounts
686  * @m: seq_file to write to
687  * @root: root of that (sub)tree
688  */
ceph_show_options(struct seq_file * m,struct dentry * root)689 static int ceph_show_options(struct seq_file *m, struct dentry *root)
690 {
691 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(root->d_sb);
692 	struct ceph_mount_options *fsopt = fsc->mount_options;
693 	size_t pos;
694 	int ret;
695 
696 	/* a comma between MNT/MS and client options */
697 	seq_putc(m, ',');
698 	pos = m->count;
699 
700 	ret = ceph_print_client_options(m, fsc->client, false);
701 	if (ret)
702 		return ret;
703 
704 	/* retract our comma if no client options */
705 	if (m->count == pos)
706 		m->count--;
707 
708 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
709 		seq_puts(m, ",dirstat");
710 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
711 		seq_puts(m, ",rbytes");
712 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
713 		seq_puts(m, ",noasyncreaddir");
714 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
715 		seq_puts(m, ",nodcache");
716 	if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
717 		seq_puts(m, ",ino32");
718 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
719 		seq_show_option(m, "fsc", fsopt->fscache_uniq);
720 	}
721 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
722 		seq_puts(m, ",nopoolperm");
723 	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
724 		seq_puts(m, ",noquotadf");
725 
726 #ifdef CONFIG_CEPH_FS_POSIX_ACL
727 	if (root->d_sb->s_flags & SB_POSIXACL)
728 		seq_puts(m, ",acl");
729 	else
730 		seq_puts(m, ",noacl");
731 #endif
732 
733 	if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
734 		seq_puts(m, ",copyfrom");
735 
736 	/* dump mds_namespace when old device syntax is in use */
737 	if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
738 		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
739 
740 	if (fsopt->mon_addr)
741 		seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
742 
743 	if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
744 		seq_show_option(m, "recover_session", "clean");
745 
746 	if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
747 		seq_puts(m, ",wsync");
748 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
749 		seq_puts(m, ",nopagecache");
750 	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
751 		seq_puts(m, ",sparseread");
752 
753 	fscrypt_show_test_dummy_encryption(m, ',', root->d_sb);
754 
755 	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
756 		seq_printf(m, ",wsize=%u", fsopt->wsize);
757 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
758 		seq_printf(m, ",rsize=%u", fsopt->rsize);
759 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
760 		seq_printf(m, ",rasize=%u", fsopt->rasize);
761 	if (fsopt->congestion_kb != default_congestion_kb())
762 		seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
763 	if (fsopt->caps_max)
764 		seq_printf(m, ",caps_max=%d", fsopt->caps_max);
765 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
766 		seq_printf(m, ",caps_wanted_delay_min=%u",
767 			 fsopt->caps_wanted_delay_min);
768 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
769 		seq_printf(m, ",caps_wanted_delay_max=%u",
770 			   fsopt->caps_wanted_delay_max);
771 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
772 		seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
773 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
774 		seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
775 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
776 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
777 
778 	return 0;
779 }
780 
781 /*
782  * handle any mon messages the standard library doesn't understand.
783  * return error if we don't either.
784  */
extra_mon_dispatch(struct ceph_client * client,struct ceph_msg * msg)785 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
786 {
787 	struct ceph_fs_client *fsc = client->private;
788 	int type = le16_to_cpu(msg->hdr.type);
789 
790 	switch (type) {
791 	case CEPH_MSG_MDS_MAP:
792 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
793 		return 0;
794 	case CEPH_MSG_FS_MAP_USER:
795 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
796 		return 0;
797 	default:
798 		return -1;
799 	}
800 }
801 
802 /*
803  * create a new fs client
804  *
805  * Success or not, this function consumes @fsopt and @opt.
806  */
create_fs_client(struct ceph_mount_options * fsopt,struct ceph_options * opt)807 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
808 					struct ceph_options *opt)
809 {
810 	struct ceph_fs_client *fsc;
811 	int err;
812 
813 	fsc = kzalloc_obj(*fsc);
814 	if (!fsc) {
815 		err = -ENOMEM;
816 		goto fail;
817 	}
818 
819 	fsc->client = ceph_create_client(opt, fsc);
820 	if (IS_ERR(fsc->client)) {
821 		err = PTR_ERR(fsc->client);
822 		goto fail;
823 	}
824 	opt = NULL; /* fsc->client now owns this */
825 
826 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
827 	ceph_set_opt(fsc->client, ABORT_ON_FULL);
828 
829 	if (!fsopt->mds_namespace) {
830 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
831 				   0, true);
832 	} else {
833 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
834 				   0, false);
835 	}
836 
837 	fsc->mount_options = fsopt;
838 
839 	fsc->sb = NULL;
840 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
841 	fsc->filp_gen = 1;
842 	fsc->have_copy_from2 = true;
843 
844 	atomic_long_set(&fsc->writeback_count, 0);
845 	fsc->write_congested = false;
846 
847 	err = -ENOMEM;
848 	/*
849 	 * The number of concurrent works can be high but they don't need
850 	 * to be processed in parallel, limit concurrency.
851 	 */
852 	fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
853 	if (!fsc->inode_wq)
854 		goto fail_client;
855 	fsc->cap_wq = alloc_workqueue("ceph-cap", WQ_PERCPU, 1);
856 	if (!fsc->cap_wq)
857 		goto fail_inode_wq;
858 
859 	hash_init(fsc->async_unlink_conflict);
860 	spin_lock_init(&fsc->async_unlink_conflict_lock);
861 
862 	spin_lock(&ceph_fsc_lock);
863 	list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
864 	spin_unlock(&ceph_fsc_lock);
865 
866 	return fsc;
867 
868 fail_inode_wq:
869 	destroy_workqueue(fsc->inode_wq);
870 fail_client:
871 	ceph_destroy_client(fsc->client);
872 fail:
873 	kfree(fsc);
874 	if (opt)
875 		ceph_destroy_options(opt);
876 	destroy_mount_options(fsopt);
877 	return ERR_PTR(err);
878 }
879 
flush_fs_workqueues(struct ceph_fs_client * fsc)880 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
881 {
882 	flush_workqueue(fsc->inode_wq);
883 	flush_workqueue(fsc->cap_wq);
884 }
885 
destroy_fs_client(struct ceph_fs_client * fsc)886 static void destroy_fs_client(struct ceph_fs_client *fsc)
887 {
888 	doutc(fsc->client, "%p\n", fsc);
889 
890 	spin_lock(&ceph_fsc_lock);
891 	list_del(&fsc->metric_wakeup);
892 	spin_unlock(&ceph_fsc_lock);
893 
894 	ceph_mdsc_destroy(fsc);
895 	destroy_workqueue(fsc->inode_wq);
896 	destroy_workqueue(fsc->cap_wq);
897 
898 	destroy_mount_options(fsc->mount_options);
899 
900 	ceph_destroy_client(fsc->client);
901 
902 	kfree(fsc);
903 	dout("%s: %p done\n", __func__, fsc);
904 }
905 
906 /*
907  * caches
908  */
909 struct kmem_cache *ceph_inode_cachep;
910 struct kmem_cache *ceph_cap_cachep;
911 struct kmem_cache *ceph_cap_snap_cachep;
912 struct kmem_cache *ceph_cap_flush_cachep;
913 struct kmem_cache *ceph_dentry_cachep;
914 struct kmem_cache *ceph_file_cachep;
915 struct kmem_cache *ceph_dir_file_cachep;
916 struct kmem_cache *ceph_mds_request_cachep;
917 mempool_t *ceph_wb_pagevec_pool;
918 
ceph_inode_init_once(void * foo)919 static void ceph_inode_init_once(void *foo)
920 {
921 	struct ceph_inode_info *ci = foo;
922 	inode_init_once(&ci->netfs.inode);
923 }
924 
init_caches(void)925 static int __init init_caches(void)
926 {
927 	int error = -ENOMEM;
928 
929 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
930 				      sizeof(struct ceph_inode_info),
931 				      __alignof__(struct ceph_inode_info),
932 				      SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
933 				      ceph_inode_init_once);
934 	if (!ceph_inode_cachep)
935 		return -ENOMEM;
936 
937 	ceph_cap_cachep = KMEM_CACHE(ceph_cap, 0);
938 	if (!ceph_cap_cachep)
939 		goto bad_cap;
940 	ceph_cap_snap_cachep = KMEM_CACHE(ceph_cap_snap, 0);
941 	if (!ceph_cap_snap_cachep)
942 		goto bad_cap_snap;
943 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
944 					   SLAB_RECLAIM_ACCOUNT);
945 	if (!ceph_cap_flush_cachep)
946 		goto bad_cap_flush;
947 
948 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
949 					SLAB_RECLAIM_ACCOUNT);
950 	if (!ceph_dentry_cachep)
951 		goto bad_dentry;
952 
953 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, 0);
954 	if (!ceph_file_cachep)
955 		goto bad_file;
956 
957 	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, 0);
958 	if (!ceph_dir_file_cachep)
959 		goto bad_dir_file;
960 
961 	ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, 0);
962 	if (!ceph_mds_request_cachep)
963 		goto bad_mds_req;
964 
965 	ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10,
966 	    (CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT) * sizeof(struct page *));
967 	if (!ceph_wb_pagevec_pool)
968 		goto bad_pagevec_pool;
969 
970 	error = ceph_subvolume_metrics_cache_init();
971 	if (error)
972 		goto bad_subvol_metrics;
973 
974 	return 0;
975 
976 bad_subvol_metrics:
977 	mempool_destroy(ceph_wb_pagevec_pool);
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 	ceph_subvolume_metrics_cache_destroy();
1015 }
1016 
__ceph_umount_begin(struct ceph_fs_client * fsc)1017 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
1018 {
1019 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
1020 	ceph_mdsc_force_umount(fsc->mdsc);
1021 	fsc->filp_gen++; // invalidate open files
1022 }
1023 
1024 /*
1025  * ceph_umount_begin - initiate forced umount.  Tear down the
1026  * mount, skipping steps that may hang while waiting for server(s).
1027  */
ceph_umount_begin(struct super_block * sb)1028 void ceph_umount_begin(struct super_block *sb)
1029 {
1030 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1031 
1032 	doutc(fsc->client, "starting forced umount\n");
1033 
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	= inode_just_drop,
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_client *cl = fsc->client;
1060 	struct ceph_mds_client *mdsc = fsc->mdsc;
1061 	struct ceph_mds_request *req = NULL;
1062 	int err;
1063 	struct dentry *root;
1064 
1065 	/* open dir */
1066 	doutc(cl, "opening '%s'\n", path);
1067 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1068 	if (IS_ERR(req))
1069 		return ERR_CAST(req);
1070 	req->r_path1 = kstrdup(path, GFP_NOFS);
1071 	if (!req->r_path1) {
1072 		root = ERR_PTR(-ENOMEM);
1073 		goto out;
1074 	}
1075 
1076 	req->r_ino1.ino = CEPH_INO_ROOT;
1077 	req->r_ino1.snap = CEPH_NOSNAP;
1078 	req->r_started = started;
1079 	req->r_timeout = fsc->client->options->mount_timeout;
1080 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1081 	req->r_num_caps = 2;
1082 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1083 	if (err == 0) {
1084 		struct inode *inode = req->r_target_inode;
1085 		req->r_target_inode = NULL;
1086 		doutc(cl, "success\n");
1087 		root = d_make_root(inode);
1088 		if (!root) {
1089 			root = ERR_PTR(-ENOMEM);
1090 			goto out;
1091 		}
1092 		doutc(cl, "success, root dentry is %p\n", root);
1093 	} else {
1094 		root = ERR_PTR(err);
1095 	}
1096 out:
1097 	ceph_mdsc_put_request(req);
1098 	return root;
1099 }
1100 
1101 #ifdef CONFIG_FS_ENCRYPTION
ceph_apply_test_dummy_encryption(struct super_block * sb,struct fs_context * fc,struct ceph_mount_options * fsopt)1102 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1103 					    struct fs_context *fc,
1104 					    struct ceph_mount_options *fsopt)
1105 {
1106 	struct ceph_fs_client *fsc = sb->s_fs_info;
1107 
1108 	if (!fscrypt_is_dummy_policy_set(&fsopt->dummy_enc_policy))
1109 		return 0;
1110 
1111 	/* No changing encryption context on remount. */
1112 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE &&
1113 	    !fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1114 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1115 						 &fsc->fsc_dummy_enc_policy))
1116 			return 0;
1117 		errorfc(fc, "Can't set test_dummy_encryption on remount");
1118 		return -EINVAL;
1119 	}
1120 
1121 	/* Also make sure fsopt doesn't contain a conflicting value. */
1122 	if (fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1123 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1124 						 &fsc->fsc_dummy_enc_policy))
1125 			return 0;
1126 		errorfc(fc, "Conflicting test_dummy_encryption options");
1127 		return -EINVAL;
1128 	}
1129 
1130 	fsc->fsc_dummy_enc_policy = fsopt->dummy_enc_policy;
1131 	memset(&fsopt->dummy_enc_policy, 0, sizeof(fsopt->dummy_enc_policy));
1132 
1133 	warnfc(fc, "test_dummy_encryption mode enabled");
1134 	return 0;
1135 }
1136 #else
ceph_apply_test_dummy_encryption(struct super_block * sb,struct fs_context * fc,struct ceph_mount_options * fsopt)1137 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1138 					    struct fs_context *fc,
1139 					    struct ceph_mount_options *fsopt)
1140 {
1141 	return 0;
1142 }
1143 #endif
1144 
1145 /*
1146  * mount: join the ceph cluster, and open root directory.
1147  */
ceph_real_mount(struct ceph_fs_client * fsc,struct fs_context * fc)1148 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1149 				      struct fs_context *fc)
1150 {
1151 	struct ceph_client *cl = fsc->client;
1152 	int err;
1153 	unsigned long started = jiffies;  /* note the start time */
1154 	struct dentry *root;
1155 
1156 	doutc(cl, "mount start %p\n", fsc);
1157 	mutex_lock(&fsc->client->mount_mutex);
1158 
1159 	if (!fsc->sb->s_root) {
1160 		const char *path = fsc->mount_options->server_path ?
1161 				     fsc->mount_options->server_path + 1 : "";
1162 
1163 		err = __ceph_open_session(fsc->client);
1164 		if (err < 0)
1165 			goto out;
1166 
1167 		/* setup fscache */
1168 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1169 			err = ceph_fscache_register_fs(fsc, fc);
1170 			if (err < 0)
1171 				goto out;
1172 		}
1173 
1174 		err = ceph_apply_test_dummy_encryption(fsc->sb, fc,
1175 						       fsc->mount_options);
1176 		if (err)
1177 			goto out;
1178 
1179 		doutc(cl, "mount opening path '%s'\n", path);
1180 
1181 		ceph_fs_debugfs_init(fsc);
1182 
1183 		root = open_root_dentry(fsc, path, started);
1184 		if (IS_ERR(root)) {
1185 			err = PTR_ERR(root);
1186 			goto out;
1187 		}
1188 		fsc->sb->s_root = dget(root);
1189 	} else {
1190 		root = dget(fsc->sb->s_root);
1191 	}
1192 
1193 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1194 	doutc(cl, "mount success\n");
1195 	mutex_unlock(&fsc->client->mount_mutex);
1196 	return root;
1197 
1198 out:
1199 	mutex_unlock(&fsc->client->mount_mutex);
1200 	ceph_fscrypt_free_dummy_policy(fsc);
1201 	return ERR_PTR(err);
1202 }
1203 
ceph_set_super(struct super_block * s,struct fs_context * fc)1204 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1205 {
1206 	struct ceph_fs_client *fsc = s->s_fs_info;
1207 	struct ceph_client *cl = fsc->client;
1208 	int ret;
1209 
1210 	doutc(cl, "%p\n", s);
1211 
1212 	s->s_maxbytes = MAX_LFS_FILESIZE;
1213 
1214 	s->s_xattr = ceph_xattr_handlers;
1215 	fsc->sb = s;
1216 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1217 
1218 	s->s_op = &ceph_super_ops;
1219 	set_default_d_op(s, &ceph_dentry_ops);
1220 	s->s_export_op = &ceph_export_ops;
1221 
1222 	s->s_time_gran = 1;
1223 	s->s_time_min = 0;
1224 	s->s_time_max = U32_MAX;
1225 	s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1226 	s->s_magic = CEPH_SUPER_MAGIC;
1227 
1228 	ceph_fscrypt_set_ops(s);
1229 
1230 	ret = set_anon_super_fc(s, fc);
1231 	if (ret != 0)
1232 		fsc->sb = NULL;
1233 	return ret;
1234 }
1235 
1236 /*
1237  * share superblock if same fs AND options
1238  */
ceph_compare_super(struct super_block * sb,struct fs_context * fc)1239 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1240 {
1241 	struct ceph_fs_client *new = fc->s_fs_info;
1242 	struct ceph_mount_options *fsopt = new->mount_options;
1243 	struct ceph_options *opt = new->client->options;
1244 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1245 	struct ceph_client *cl = fsc->client;
1246 
1247 	doutc(cl, "%p\n", sb);
1248 
1249 	if (compare_mount_options(fsopt, opt, fsc)) {
1250 		doutc(cl, "monitor(s)/mount options don't match\n");
1251 		return 0;
1252 	}
1253 	if ((opt->flags & CEPH_OPT_FSID) &&
1254 	    ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1255 		doutc(cl, "fsid doesn't match\n");
1256 		return 0;
1257 	}
1258 	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1259 		doutc(cl, "flags differ\n");
1260 		return 0;
1261 	}
1262 
1263 	if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1264 		doutc(cl, "client is blocklisted (and CLEANRECOVER is not set)\n");
1265 		return 0;
1266 	}
1267 
1268 	if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1269 		doutc(cl, "client has been forcibly unmounted\n");
1270 		return 0;
1271 	}
1272 
1273 	return 1;
1274 }
1275 
1276 /*
1277  * construct our own bdi so we can control readahead, etc.
1278  */
1279 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1280 
ceph_setup_bdi(struct super_block * sb,struct ceph_fs_client * fsc)1281 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1282 {
1283 	int err;
1284 
1285 	err = super_setup_bdi_name(sb, "ceph-%ld",
1286 				   atomic_long_inc_return(&bdi_seq));
1287 	if (err)
1288 		return err;
1289 
1290 	/* set ra_pages based on rasize mount option? */
1291 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1292 
1293 	/* set io_pages based on max osd read size */
1294 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1295 
1296 	return 0;
1297 }
1298 
ceph_get_tree(struct fs_context * fc)1299 static int ceph_get_tree(struct fs_context *fc)
1300 {
1301 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1302 	struct ceph_mount_options *fsopt = pctx->opts;
1303 	struct super_block *sb;
1304 	struct ceph_fs_client *fsc;
1305 	struct dentry *res;
1306 	int (*compare_super)(struct super_block *, struct fs_context *) =
1307 		ceph_compare_super;
1308 	int err;
1309 
1310 	dout("ceph_get_tree\n");
1311 
1312 	if (!fc->source)
1313 		return invalfc(fc, "No source");
1314 	if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1315 		return invalfc(fc, "No monitor address");
1316 
1317 	/* create client (which we may/may not use) */
1318 	fsc = create_fs_client(pctx->opts, pctx->copts);
1319 	pctx->opts = NULL;
1320 	pctx->copts = NULL;
1321 	if (IS_ERR(fsc)) {
1322 		err = PTR_ERR(fsc);
1323 		goto out_final;
1324 	}
1325 
1326 	err = ceph_mdsc_init(fsc);
1327 	if (err < 0)
1328 		goto out;
1329 
1330 	if (ceph_test_opt(fsc->client, NOSHARE))
1331 		compare_super = NULL;
1332 
1333 	fc->s_fs_info = fsc;
1334 	sb = sget_fc(fc, compare_super, ceph_set_super);
1335 	fc->s_fs_info = NULL;
1336 	if (IS_ERR(sb)) {
1337 		err = PTR_ERR(sb);
1338 		goto out;
1339 	}
1340 
1341 	if (ceph_sb_to_fs_client(sb) != fsc) {
1342 		destroy_fs_client(fsc);
1343 		fsc = ceph_sb_to_fs_client(sb);
1344 		dout("get_sb got existing client %p\n", fsc);
1345 	} else {
1346 		dout("get_sb using new client %p\n", fsc);
1347 		err = ceph_setup_bdi(sb, fsc);
1348 		if (err < 0)
1349 			goto out_splat;
1350 	}
1351 
1352 	res = ceph_real_mount(fsc, fc);
1353 	if (IS_ERR(res)) {
1354 		err = PTR_ERR(res);
1355 		goto out_splat;
1356 	}
1357 
1358 	doutc(fsc->client, "root %p inode %p ino %llx.%llx\n", res,
1359 		    d_inode(res), ceph_vinop(d_inode(res)));
1360 	fc->root = fsc->sb->s_root;
1361 	return 0;
1362 
1363 out_splat:
1364 	if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1365 		pr_info("No mds server is up or the cluster is laggy\n");
1366 		err = -EHOSTUNREACH;
1367 	}
1368 
1369 	ceph_mdsc_close_sessions(fsc->mdsc);
1370 	deactivate_locked_super(sb);
1371 	goto out_final;
1372 
1373 out:
1374 	destroy_fs_client(fsc);
1375 out_final:
1376 	dout("ceph_get_tree fail %d\n", err);
1377 	return err;
1378 }
1379 
ceph_free_fc(struct fs_context * fc)1380 static void ceph_free_fc(struct fs_context *fc)
1381 {
1382 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1383 
1384 	if (pctx) {
1385 		destroy_mount_options(pctx->opts);
1386 		ceph_destroy_options(pctx->copts);
1387 		kfree(pctx);
1388 	}
1389 }
1390 
ceph_reconfigure_fc(struct fs_context * fc)1391 static int ceph_reconfigure_fc(struct fs_context *fc)
1392 {
1393 	int err;
1394 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1395 	struct ceph_mount_options *fsopt = pctx->opts;
1396 	struct super_block *sb = fc->root->d_sb;
1397 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1398 
1399 	err = ceph_apply_test_dummy_encryption(sb, fc, fsopt);
1400 	if (err)
1401 		return err;
1402 
1403 	if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1404 		ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1405 	else
1406 		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1407 
1408 	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
1409 		ceph_set_mount_opt(fsc, SPARSEREAD);
1410 	else
1411 		ceph_clear_mount_opt(fsc, SPARSEREAD);
1412 
1413 	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1414 		kfree(fsc->mount_options->mon_addr);
1415 		fsc->mount_options->mon_addr = fsopt->mon_addr;
1416 		fsopt->mon_addr = NULL;
1417 		pr_notice_client(fsc->client,
1418 			"monitor addresses recorded, but not used for reconnection");
1419 	}
1420 
1421 	sync_filesystem(sb);
1422 	return 0;
1423 }
1424 
1425 static const struct fs_context_operations ceph_context_ops = {
1426 	.free		= ceph_free_fc,
1427 	.parse_param	= ceph_parse_mount_param,
1428 	.get_tree	= ceph_get_tree,
1429 	.reconfigure	= ceph_reconfigure_fc,
1430 };
1431 
1432 /*
1433  * Set up the filesystem mount context.
1434  */
ceph_init_fs_context(struct fs_context * fc)1435 static int ceph_init_fs_context(struct fs_context *fc)
1436 {
1437 	struct ceph_parse_opts_ctx *pctx;
1438 	struct ceph_mount_options *fsopt;
1439 
1440 	pctx = kzalloc_obj(*pctx);
1441 	if (!pctx)
1442 		return -ENOMEM;
1443 
1444 	pctx->copts = ceph_alloc_options();
1445 	if (!pctx->copts)
1446 		goto nomem;
1447 
1448 	pctx->opts = kzalloc_obj(*pctx->opts);
1449 	if (!pctx->opts)
1450 		goto nomem;
1451 
1452 	fsopt = pctx->opts;
1453 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1454 
1455 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1456 	fsopt->rsize = CEPH_MAX_READ_SIZE;
1457 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
1458 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1459 	if (!fsopt->snapdir_name)
1460 		goto nomem;
1461 
1462 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1463 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1464 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1465 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1466 	fsopt->congestion_kb = default_congestion_kb();
1467 
1468 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1469 	fc->sb_flags |= SB_POSIXACL;
1470 #endif
1471 
1472 	fc->fs_private = pctx;
1473 	fc->ops = &ceph_context_ops;
1474 	return 0;
1475 
1476 nomem:
1477 	destroy_mount_options(pctx->opts);
1478 	ceph_destroy_options(pctx->copts);
1479 	kfree(pctx);
1480 	return -ENOMEM;
1481 }
1482 
1483 /*
1484  * Return true if it successfully increases the blocker counter,
1485  * or false if the mdsc is in stopping and flushed state.
1486  */
__inc_stopping_blocker(struct ceph_mds_client * mdsc)1487 static bool __inc_stopping_blocker(struct ceph_mds_client *mdsc)
1488 {
1489 	spin_lock(&mdsc->stopping_lock);
1490 	if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING) {
1491 		spin_unlock(&mdsc->stopping_lock);
1492 		return false;
1493 	}
1494 	atomic_inc(&mdsc->stopping_blockers);
1495 	spin_unlock(&mdsc->stopping_lock);
1496 	return true;
1497 }
1498 
__dec_stopping_blocker(struct ceph_mds_client * mdsc)1499 static void __dec_stopping_blocker(struct ceph_mds_client *mdsc)
1500 {
1501 	spin_lock(&mdsc->stopping_lock);
1502 	if (!atomic_dec_return(&mdsc->stopping_blockers) &&
1503 	    mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING)
1504 		complete_all(&mdsc->stopping_waiter);
1505 	spin_unlock(&mdsc->stopping_lock);
1506 }
1507 
1508 /* For metadata IO requests */
ceph_inc_mds_stopping_blocker(struct ceph_mds_client * mdsc,struct ceph_mds_session * session)1509 bool ceph_inc_mds_stopping_blocker(struct ceph_mds_client *mdsc,
1510 				   struct ceph_mds_session *session)
1511 {
1512 	mutex_lock(&session->s_mutex);
1513 	inc_session_sequence(session);
1514 	mutex_unlock(&session->s_mutex);
1515 
1516 	return __inc_stopping_blocker(mdsc);
1517 }
1518 
ceph_dec_mds_stopping_blocker(struct ceph_mds_client * mdsc)1519 void ceph_dec_mds_stopping_blocker(struct ceph_mds_client *mdsc)
1520 {
1521 	__dec_stopping_blocker(mdsc);
1522 }
1523 
1524 /* For data IO requests */
ceph_inc_osd_stopping_blocker(struct ceph_mds_client * mdsc)1525 bool ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1526 {
1527 	return __inc_stopping_blocker(mdsc);
1528 }
1529 
ceph_dec_osd_stopping_blocker(struct ceph_mds_client * mdsc)1530 void ceph_dec_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1531 {
1532 	__dec_stopping_blocker(mdsc);
1533 }
1534 
ceph_kill_sb(struct super_block * s)1535 static void ceph_kill_sb(struct super_block *s)
1536 {
1537 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
1538 	struct ceph_client *cl = fsc->client;
1539 	struct ceph_mds_client *mdsc = fsc->mdsc;
1540 	bool wait;
1541 
1542 	doutc(cl, "%p\n", s);
1543 
1544 	ceph_mdsc_pre_umount(mdsc);
1545 	flush_fs_workqueues(fsc);
1546 
1547 	/*
1548 	 * Though the kill_anon_super() will finally trigger the
1549 	 * sync_filesystem() anyway, we still need to do it here and
1550 	 * then bump the stage of shutdown. This will allow us to
1551 	 * drop any further message, which will increase the inodes'
1552 	 * i_count reference counters but makes no sense any more,
1553 	 * from MDSs.
1554 	 *
1555 	 * Without this when evicting the inodes it may fail in the
1556 	 * kill_anon_super(), which will trigger a warning when
1557 	 * destroying the fscrypt keyring and then possibly trigger
1558 	 * a further crash in ceph module when the iput() tries to
1559 	 * evict the inodes later.
1560 	 */
1561 	sync_filesystem(s);
1562 
1563 	if (atomic64_read(&mdsc->dirty_folios) > 0) {
1564 		wait_queue_head_t *wq = &mdsc->flush_end_wq;
1565 		long timeleft = wait_event_killable_timeout(*wq,
1566 					atomic64_read(&mdsc->dirty_folios) <= 0,
1567 					fsc->client->options->mount_timeout);
1568 		if (!timeleft) /* timed out */
1569 			pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1570 		else if (timeleft < 0) /* killed */
1571 			pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1572 	}
1573 
1574 	spin_lock(&mdsc->stopping_lock);
1575 	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHING;
1576 	wait = !!atomic_read(&mdsc->stopping_blockers);
1577 	spin_unlock(&mdsc->stopping_lock);
1578 
1579 	if (wait && atomic_read(&mdsc->stopping_blockers)) {
1580 		long timeleft = wait_for_completion_killable_timeout(
1581 					&mdsc->stopping_waiter,
1582 					fsc->client->options->mount_timeout);
1583 		if (!timeleft) /* timed out */
1584 			pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1585 		else if (timeleft < 0) /* killed */
1586 			pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1587 	}
1588 
1589 	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHED;
1590 	kill_anon_super(s);
1591 
1592 	fsc->client->extra_mon_dispatch = NULL;
1593 	ceph_fs_debugfs_cleanup(fsc);
1594 
1595 	ceph_fscache_unregister_fs(fsc);
1596 
1597 	destroy_fs_client(fsc);
1598 }
1599 
1600 static struct file_system_type ceph_fs_type = {
1601 	.owner		= THIS_MODULE,
1602 	.name		= "ceph",
1603 	.init_fs_context = ceph_init_fs_context,
1604 	.kill_sb	= ceph_kill_sb,
1605 	.fs_flags	= FS_RENAME_DOES_D_MOVE | FS_ALLOW_IDMAP,
1606 };
1607 MODULE_ALIAS_FS("ceph");
1608 
ceph_force_reconnect(struct super_block * sb)1609 int ceph_force_reconnect(struct super_block *sb)
1610 {
1611 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1612 	int err = 0;
1613 
1614 	fsc->mount_state = CEPH_MOUNT_RECOVER;
1615 	__ceph_umount_begin(fsc);
1616 
1617 	/* Make sure all page caches get invalidated.
1618 	 * see remove_session_caps_cb() */
1619 	flush_workqueue(fsc->inode_wq);
1620 
1621 	/* In case that we were blocklisted. This also reset
1622 	 * all mon/osd connections */
1623 	ceph_reset_client_addr(fsc->client);
1624 
1625 	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1626 
1627 	fsc->blocklisted = false;
1628 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1629 
1630 	if (sb->s_root) {
1631 		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1632 					CEPH_STAT_CAP_INODE, true);
1633 	}
1634 	return err;
1635 }
1636 
init_ceph(void)1637 static int __init init_ceph(void)
1638 {
1639 	int ret = init_caches();
1640 	if (ret)
1641 		goto out;
1642 
1643 	ceph_flock_init();
1644 	ret = register_filesystem(&ceph_fs_type);
1645 	if (ret)
1646 		goto out_caches;
1647 
1648 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1649 
1650 	return 0;
1651 
1652 out_caches:
1653 	destroy_caches();
1654 out:
1655 	return ret;
1656 }
1657 
exit_ceph(void)1658 static void __exit exit_ceph(void)
1659 {
1660 	dout("exit_ceph\n");
1661 	unregister_filesystem(&ceph_fs_type);
1662 	destroy_caches();
1663 }
1664 
param_set_metrics(const char * val,const struct kernel_param * kp)1665 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1666 {
1667 	struct ceph_fs_client *fsc;
1668 	int ret;
1669 
1670 	ret = param_set_bool(val, kp);
1671 	if (ret) {
1672 		pr_err("Failed to parse sending metrics switch value '%s'\n",
1673 		       val);
1674 		return ret;
1675 	} else if (!disable_send_metrics) {
1676 		// wake up all the mds clients
1677 		spin_lock(&ceph_fsc_lock);
1678 		list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1679 			metric_schedule_delayed(&fsc->mdsc->metric);
1680 		}
1681 		spin_unlock(&ceph_fsc_lock);
1682 	}
1683 
1684 	return 0;
1685 }
1686 
1687 static const struct kernel_param_ops param_ops_metrics = {
1688 	.set = param_set_metrics,
1689 	.get = param_get_bool,
1690 };
1691 
1692 bool disable_send_metrics = false;
1693 module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1694 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1695 
1696 /* for both v1 and v2 syntax */
1697 static bool mount_support = true;
1698 static const struct kernel_param_ops param_ops_mount_syntax = {
1699 	.get = param_get_bool,
1700 };
1701 module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1702 module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1703 
1704 bool enable_unsafe_idmap = false;
1705 module_param(enable_unsafe_idmap, bool, 0644);
1706 MODULE_PARM_DESC(enable_unsafe_idmap,
1707 		 "Allow to use idmapped mounts with MDS without CEPHFS_FEATURE_HAS_OWNER_UIDGID");
1708 
1709 module_init(init_ceph);
1710 module_exit(exit_ceph);
1711 
1712 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1713 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1714 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1715 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1716 MODULE_LICENSE("GPL");
1717