xref: /linux/fs/ceph/super.c (revision 894b3c35d1de9cfa4f72b21e280d80d278879c20)
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 	doutc(fsc->client, "begin\n");
50 	ceph_fscrypt_free_dummy_policy(fsc);
51 	ceph_mdsc_close_sessions(fsc->mdsc);
52 	doutc(fsc->client, "done\n");
53 }
54 
ceph_statfs(struct dentry * dentry,struct kstatfs * buf)55 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
56 {
57 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(d_inode(dentry));
58 	struct ceph_mon_client *monc = &fsc->client->monc;
59 	struct ceph_statfs st;
60 	int i, err;
61 	u64 data_pool;
62 
63 	doutc(fsc->client, "begin\n");
64 	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
65 		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
66 	} else {
67 		data_pool = CEPH_NOPOOL;
68 	}
69 
70 	err = ceph_monc_do_statfs(monc, data_pool, &st);
71 	if (err < 0)
72 		return err;
73 
74 	/* fill in kstatfs */
75 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
76 
77 	/*
78 	 * Express utilization in terms of large blocks to avoid
79 	 * overflow on 32-bit machines.
80 	 */
81 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
82 
83 	/*
84 	 * By default use root quota for stats; fallback to overall filesystem
85 	 * usage if using 'noquotadf' mount option or if the root dir doesn't
86 	 * have max_bytes quota set.
87 	 */
88 	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
89 	    !ceph_quota_update_statfs(fsc, buf)) {
90 		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
91 		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
92 		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
93 	}
94 
95 	/*
96 	 * NOTE: for the time being, we make bsize == frsize to humor
97 	 * not-yet-ancient versions of glibc that are broken.
98 	 * Someday, we will probably want to report a real block
99 	 * size...  whatever that may mean for a network file system!
100 	 */
101 	buf->f_bsize = buf->f_frsize;
102 
103 	buf->f_files = le64_to_cpu(st.num_objects);
104 	buf->f_ffree = -1;
105 	buf->f_namelen = NAME_MAX;
106 
107 	/* Must convert the fsid, for consistent values across arches */
108 	buf->f_fsid.val[0] = 0;
109 	mutex_lock(&monc->mutex);
110 	for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
111 		buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
112 	mutex_unlock(&monc->mutex);
113 
114 	/* fold the fs_cluster_id into the upper bits */
115 	buf->f_fsid.val[1] = monc->fs_cluster_id;
116 
117 	doutc(fsc->client, "done\n");
118 	return 0;
119 }
120 
ceph_sync_fs(struct super_block * sb,int wait)121 static int ceph_sync_fs(struct super_block *sb, int wait)
122 {
123 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
124 	struct ceph_client *cl = fsc->client;
125 
126 	if (!wait) {
127 		doutc(cl, "(non-blocking)\n");
128 		ceph_flush_dirty_caps(fsc->mdsc);
129 		ceph_flush_cap_releases(fsc->mdsc);
130 		doutc(cl, "(non-blocking) done\n");
131 		return 0;
132 	}
133 
134 	doutc(cl, "(blocking)\n");
135 	ceph_osdc_sync(&fsc->client->osdc);
136 	ceph_mdsc_sync(fsc->mdsc);
137 	doutc(cl, "(blocking) done\n");
138 	return 0;
139 }
140 
141 /*
142  * mount options
143  */
144 enum {
145 	Opt_wsize,
146 	Opt_rsize,
147 	Opt_rasize,
148 	Opt_caps_wanted_delay_min,
149 	Opt_caps_wanted_delay_max,
150 	Opt_caps_max,
151 	Opt_readdir_max_entries,
152 	Opt_readdir_max_bytes,
153 	Opt_congestion_kb,
154 	/* int args above */
155 	Opt_snapdirname,
156 	Opt_mds_namespace,
157 	Opt_recover_session,
158 	Opt_source,
159 	Opt_mon_addr,
160 	Opt_test_dummy_encryption,
161 	/* string args above */
162 	Opt_dirstat,
163 	Opt_rbytes,
164 	Opt_asyncreaddir,
165 	Opt_dcache,
166 	Opt_ino32,
167 	Opt_fscache,
168 	Opt_poolperm,
169 	Opt_require_active_mds,
170 	Opt_acl,
171 	Opt_quotadf,
172 	Opt_copyfrom,
173 	Opt_wsync,
174 	Opt_pagecache,
175 	Opt_sparseread,
176 };
177 
178 enum ceph_recover_session_mode {
179 	ceph_recover_session_no,
180 	ceph_recover_session_clean
181 };
182 
183 static const struct constant_table ceph_param_recover[] = {
184 	{ "no",		ceph_recover_session_no },
185 	{ "clean",	ceph_recover_session_clean },
186 	{}
187 };
188 
189 static const struct fs_parameter_spec ceph_mount_parameters[] = {
190 	fsparam_flag_no ("acl",				Opt_acl),
191 	fsparam_flag_no ("asyncreaddir",		Opt_asyncreaddir),
192 	fsparam_s32	("caps_max",			Opt_caps_max),
193 	fsparam_u32	("caps_wanted_delay_max",	Opt_caps_wanted_delay_max),
194 	fsparam_u32	("caps_wanted_delay_min",	Opt_caps_wanted_delay_min),
195 	fsparam_u32	("write_congestion_kb",		Opt_congestion_kb),
196 	fsparam_flag_no ("copyfrom",			Opt_copyfrom),
197 	fsparam_flag_no ("dcache",			Opt_dcache),
198 	fsparam_flag_no ("dirstat",			Opt_dirstat),
199 	fsparam_flag_no	("fsc",				Opt_fscache), // fsc|nofsc
200 	fsparam_string	("fsc",				Opt_fscache), // fsc=...
201 	fsparam_flag_no ("ino32",			Opt_ino32),
202 	fsparam_string	("mds_namespace",		Opt_mds_namespace),
203 	fsparam_string	("mon_addr",			Opt_mon_addr),
204 	fsparam_flag_no ("poolperm",			Opt_poolperm),
205 	fsparam_flag_no ("quotadf",			Opt_quotadf),
206 	fsparam_u32	("rasize",			Opt_rasize),
207 	fsparam_flag_no ("rbytes",			Opt_rbytes),
208 	fsparam_u32	("readdir_max_bytes",		Opt_readdir_max_bytes),
209 	fsparam_u32	("readdir_max_entries",		Opt_readdir_max_entries),
210 	fsparam_enum	("recover_session",		Opt_recover_session, ceph_param_recover),
211 	fsparam_flag_no ("require_active_mds",		Opt_require_active_mds),
212 	fsparam_u32	("rsize",			Opt_rsize),
213 	fsparam_string	("snapdirname",			Opt_snapdirname),
214 	fsparam_string	("source",			Opt_source),
215 	fsparam_flag	("test_dummy_encryption",	Opt_test_dummy_encryption),
216 	fsparam_string	("test_dummy_encryption",	Opt_test_dummy_encryption),
217 	fsparam_u32	("wsize",			Opt_wsize),
218 	fsparam_flag_no	("wsync",			Opt_wsync),
219 	fsparam_flag_no	("pagecache",			Opt_pagecache),
220 	fsparam_flag_no	("sparseread",			Opt_sparseread),
221 	{}
222 };
223 
224 struct ceph_parse_opts_ctx {
225 	struct ceph_options		*copts;
226 	struct ceph_mount_options	*opts;
227 };
228 
229 /*
230  * Remove adjacent slashes and then the trailing slash, unless it is
231  * the only remaining character.
232  *
233  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
234  */
canonicalize_path(char * path)235 static void canonicalize_path(char *path)
236 {
237 	int i, j = 0;
238 
239 	for (i = 0; path[i] != '\0'; i++) {
240 		if (path[i] != '/' || j < 1 || path[j - 1] != '/')
241 			path[j++] = path[i];
242 	}
243 
244 	if (j > 1 && path[j - 1] == '/')
245 		j--;
246 	path[j] = '\0';
247 }
248 
249 /*
250  * Check if the mds namespace in ceph_mount_options matches
251  * the passed in namespace string. First time match (when
252  * ->mds_namespace is NULL) is treated specially, since
253  * ->mds_namespace needs to be initialized by the caller.
254  */
namespace_equals(struct ceph_mount_options * fsopt,const char * namespace,size_t len)255 static int namespace_equals(struct ceph_mount_options *fsopt,
256 			    const char *namespace, size_t len)
257 {
258 	return !(fsopt->mds_namespace &&
259 		 (strlen(fsopt->mds_namespace) != len ||
260 		  strncmp(fsopt->mds_namespace, namespace, len)));
261 }
262 
ceph_parse_old_source(const char * dev_name,const char * dev_name_end,struct fs_context * fc)263 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
264 				 struct fs_context *fc)
265 {
266 	int r;
267 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
268 	struct ceph_mount_options *fsopt = pctx->opts;
269 
270 	if (*dev_name_end != ':')
271 		return invalfc(fc, "separator ':' missing in source");
272 
273 	r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
274 			       pctx->copts, fc->log.log, ',');
275 	if (r)
276 		return r;
277 
278 	fsopt->new_dev_syntax = false;
279 	return 0;
280 }
281 
ceph_parse_new_source(const char * dev_name,const char * dev_name_end,struct fs_context * fc)282 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
283 				 struct fs_context *fc)
284 {
285 	size_t len;
286 	struct ceph_fsid fsid;
287 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
288 	struct ceph_mount_options *fsopt = pctx->opts;
289 	char *fsid_start, *fs_name_start;
290 
291 	if (*dev_name_end != '=') {
292 		dout("separator '=' missing in source");
293 		return -EINVAL;
294 	}
295 
296 	fsid_start = strchr(dev_name, '@');
297 	if (!fsid_start)
298 		return invalfc(fc, "missing cluster fsid");
299 	++fsid_start; /* start of cluster fsid */
300 
301 	fs_name_start = strchr(fsid_start, '.');
302 	if (!fs_name_start)
303 		return invalfc(fc, "missing file system name");
304 
305 	if (ceph_parse_fsid(fsid_start, &fsid))
306 		return invalfc(fc, "Invalid FSID");
307 
308 	++fs_name_start; /* start of file system name */
309 	len = dev_name_end - fs_name_start;
310 
311 	if (!namespace_equals(fsopt, fs_name_start, len))
312 		return invalfc(fc, "Mismatching mds_namespace");
313 	kfree(fsopt->mds_namespace);
314 	fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
315 	if (!fsopt->mds_namespace)
316 		return -ENOMEM;
317 	dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
318 
319 	fsopt->new_dev_syntax = true;
320 	return 0;
321 }
322 
323 /*
324  * Parse the source parameter for new device format. Distinguish the device
325  * spec from the path. Try parsing new device format and fallback to old
326  * format if needed.
327  *
328  * New device syntax will looks like:
329  *     <device_spec>=/<path>
330  * where
331  *     <device_spec> is name@fsid.fsname
332  *     <path> is optional, but if present must begin with '/'
333  * (monitor addresses are passed via mount option)
334  *
335  * Old device syntax is:
336  *     <server_spec>[,<server_spec>...]:[<path>]
337  * where
338  *     <server_spec> is <ip>[:<port>]
339  *     <path> is optional, but if present must begin with '/'
340  */
ceph_parse_source(struct fs_parameter * param,struct fs_context * fc)341 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
342 {
343 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
344 	struct ceph_mount_options *fsopt = pctx->opts;
345 	char *dev_name = param->string, *dev_name_end;
346 	int ret;
347 
348 	dout("'%s'\n", dev_name);
349 	if (!dev_name || !*dev_name)
350 		return invalfc(fc, "Empty source");
351 
352 	dev_name_end = strchr(dev_name, '/');
353 	if (dev_name_end) {
354 		/*
355 		 * The server_path will include the whole chars from userland
356 		 * including the leading '/'.
357 		 */
358 		kfree(fsopt->server_path);
359 		fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
360 		if (!fsopt->server_path)
361 			return -ENOMEM;
362 
363 		canonicalize_path(fsopt->server_path);
364 	} else {
365 		dev_name_end = dev_name + strlen(dev_name);
366 	}
367 
368 	dev_name_end--;		/* back up to separator */
369 	if (dev_name_end < dev_name)
370 		return invalfc(fc, "Path missing in source");
371 
372 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
373 	if (fsopt->server_path)
374 		dout("server path '%s'\n", fsopt->server_path);
375 
376 	dout("trying new device syntax");
377 	ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
378 	if (ret) {
379 		if (ret != -EINVAL)
380 			return ret;
381 		dout("trying old device syntax");
382 		ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
383 		if (ret)
384 			return ret;
385 	}
386 
387 	fc->source = param->string;
388 	param->string = NULL;
389 	return 0;
390 }
391 
ceph_parse_mon_addr(struct fs_parameter * param,struct fs_context * fc)392 static int ceph_parse_mon_addr(struct fs_parameter *param,
393 			       struct fs_context *fc)
394 {
395 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
396 	struct ceph_mount_options *fsopt = pctx->opts;
397 
398 	kfree(fsopt->mon_addr);
399 	fsopt->mon_addr = param->string;
400 	param->string = NULL;
401 
402 	return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
403 				  pctx->copts, fc->log.log, '/');
404 }
405 
ceph_parse_mount_param(struct fs_context * fc,struct fs_parameter * param)406 static int ceph_parse_mount_param(struct fs_context *fc,
407 				  struct fs_parameter *param)
408 {
409 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
410 	struct ceph_mount_options *fsopt = pctx->opts;
411 	struct fs_parse_result result;
412 	unsigned int mode;
413 	int token, ret;
414 
415 	ret = ceph_parse_param(param, pctx->copts, fc->log.log);
416 	if (ret != -ENOPARAM)
417 		return ret;
418 
419 	token = fs_parse(fc, ceph_mount_parameters, param, &result);
420 	dout("%s: fs_parse '%s' token %d\n",__func__, param->key, token);
421 	if (token < 0)
422 		return token;
423 
424 	switch (token) {
425 	case Opt_snapdirname:
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(sizeof(*fsc), GFP_KERNEL);
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", 0, 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 	return 0;
971 
972 bad_pagevec_pool:
973 	kmem_cache_destroy(ceph_mds_request_cachep);
974 bad_mds_req:
975 	kmem_cache_destroy(ceph_dir_file_cachep);
976 bad_dir_file:
977 	kmem_cache_destroy(ceph_file_cachep);
978 bad_file:
979 	kmem_cache_destroy(ceph_dentry_cachep);
980 bad_dentry:
981 	kmem_cache_destroy(ceph_cap_flush_cachep);
982 bad_cap_flush:
983 	kmem_cache_destroy(ceph_cap_snap_cachep);
984 bad_cap_snap:
985 	kmem_cache_destroy(ceph_cap_cachep);
986 bad_cap:
987 	kmem_cache_destroy(ceph_inode_cachep);
988 	return error;
989 }
990 
destroy_caches(void)991 static void destroy_caches(void)
992 {
993 	/*
994 	 * Make sure all delayed rcu free inodes are flushed before we
995 	 * destroy cache.
996 	 */
997 	rcu_barrier();
998 
999 	kmem_cache_destroy(ceph_inode_cachep);
1000 	kmem_cache_destroy(ceph_cap_cachep);
1001 	kmem_cache_destroy(ceph_cap_snap_cachep);
1002 	kmem_cache_destroy(ceph_cap_flush_cachep);
1003 	kmem_cache_destroy(ceph_dentry_cachep);
1004 	kmem_cache_destroy(ceph_file_cachep);
1005 	kmem_cache_destroy(ceph_dir_file_cachep);
1006 	kmem_cache_destroy(ceph_mds_request_cachep);
1007 	mempool_destroy(ceph_wb_pagevec_pool);
1008 }
1009 
__ceph_umount_begin(struct ceph_fs_client * fsc)1010 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
1011 {
1012 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
1013 	ceph_mdsc_force_umount(fsc->mdsc);
1014 	fsc->filp_gen++; // invalidate open files
1015 }
1016 
1017 /*
1018  * ceph_umount_begin - initiate forced umount.  Tear down the
1019  * mount, skipping steps that may hang while waiting for server(s).
1020  */
ceph_umount_begin(struct super_block * sb)1021 void ceph_umount_begin(struct super_block *sb)
1022 {
1023 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1024 
1025 	doutc(fsc->client, "starting forced umount\n");
1026 	if (!fsc)
1027 		return;
1028 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
1029 	__ceph_umount_begin(fsc);
1030 }
1031 
1032 static const struct super_operations ceph_super_ops = {
1033 	.alloc_inode	= ceph_alloc_inode,
1034 	.free_inode	= ceph_free_inode,
1035 	.write_inode    = ceph_write_inode,
1036 	.drop_inode	= generic_delete_inode,
1037 	.evict_inode	= ceph_evict_inode,
1038 	.sync_fs        = ceph_sync_fs,
1039 	.put_super	= ceph_put_super,
1040 	.show_options   = ceph_show_options,
1041 	.statfs		= ceph_statfs,
1042 	.umount_begin   = ceph_umount_begin,
1043 };
1044 
1045 /*
1046  * Bootstrap mount by opening the root directory.  Note the mount
1047  * @started time from caller, and time out if this takes too long.
1048  */
open_root_dentry(struct ceph_fs_client * fsc,const char * path,unsigned long started)1049 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1050 				       const char *path,
1051 				       unsigned long started)
1052 {
1053 	struct ceph_client *cl = fsc->client;
1054 	struct ceph_mds_client *mdsc = fsc->mdsc;
1055 	struct ceph_mds_request *req = NULL;
1056 	int err;
1057 	struct dentry *root;
1058 
1059 	/* open dir */
1060 	doutc(cl, "opening '%s'\n", path);
1061 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1062 	if (IS_ERR(req))
1063 		return ERR_CAST(req);
1064 	req->r_path1 = kstrdup(path, GFP_NOFS);
1065 	if (!req->r_path1) {
1066 		root = ERR_PTR(-ENOMEM);
1067 		goto out;
1068 	}
1069 
1070 	req->r_ino1.ino = CEPH_INO_ROOT;
1071 	req->r_ino1.snap = CEPH_NOSNAP;
1072 	req->r_started = started;
1073 	req->r_timeout = fsc->client->options->mount_timeout;
1074 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1075 	req->r_num_caps = 2;
1076 	err = ceph_mdsc_do_request(mdsc, NULL, req);
1077 	if (err == 0) {
1078 		struct inode *inode = req->r_target_inode;
1079 		req->r_target_inode = NULL;
1080 		doutc(cl, "success\n");
1081 		root = d_make_root(inode);
1082 		if (!root) {
1083 			root = ERR_PTR(-ENOMEM);
1084 			goto out;
1085 		}
1086 		doutc(cl, "success, root dentry is %p\n", root);
1087 	} else {
1088 		root = ERR_PTR(err);
1089 	}
1090 out:
1091 	ceph_mdsc_put_request(req);
1092 	return root;
1093 }
1094 
1095 #ifdef CONFIG_FS_ENCRYPTION
ceph_apply_test_dummy_encryption(struct super_block * sb,struct fs_context * fc,struct ceph_mount_options * fsopt)1096 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1097 					    struct fs_context *fc,
1098 					    struct ceph_mount_options *fsopt)
1099 {
1100 	struct ceph_fs_client *fsc = sb->s_fs_info;
1101 
1102 	if (!fscrypt_is_dummy_policy_set(&fsopt->dummy_enc_policy))
1103 		return 0;
1104 
1105 	/* No changing encryption context on remount. */
1106 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE &&
1107 	    !fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1108 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1109 						 &fsc->fsc_dummy_enc_policy))
1110 			return 0;
1111 		errorfc(fc, "Can't set test_dummy_encryption on remount");
1112 		return -EINVAL;
1113 	}
1114 
1115 	/* Also make sure fsopt doesn't contain a conflicting value. */
1116 	if (fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1117 		if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1118 						 &fsc->fsc_dummy_enc_policy))
1119 			return 0;
1120 		errorfc(fc, "Conflicting test_dummy_encryption options");
1121 		return -EINVAL;
1122 	}
1123 
1124 	fsc->fsc_dummy_enc_policy = fsopt->dummy_enc_policy;
1125 	memset(&fsopt->dummy_enc_policy, 0, sizeof(fsopt->dummy_enc_policy));
1126 
1127 	warnfc(fc, "test_dummy_encryption mode enabled");
1128 	return 0;
1129 }
1130 #else
ceph_apply_test_dummy_encryption(struct super_block * sb,struct fs_context * fc,struct ceph_mount_options * fsopt)1131 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1132 					    struct fs_context *fc,
1133 					    struct ceph_mount_options *fsopt)
1134 {
1135 	return 0;
1136 }
1137 #endif
1138 
1139 /*
1140  * mount: join the ceph cluster, and open root directory.
1141  */
ceph_real_mount(struct ceph_fs_client * fsc,struct fs_context * fc)1142 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1143 				      struct fs_context *fc)
1144 {
1145 	struct ceph_client *cl = fsc->client;
1146 	int err;
1147 	unsigned long started = jiffies;  /* note the start time */
1148 	struct dentry *root;
1149 
1150 	doutc(cl, "mount start %p\n", fsc);
1151 	mutex_lock(&fsc->client->mount_mutex);
1152 
1153 	if (!fsc->sb->s_root) {
1154 		const char *path = fsc->mount_options->server_path ?
1155 				     fsc->mount_options->server_path + 1 : "";
1156 
1157 		err = __ceph_open_session(fsc->client, started);
1158 		if (err < 0)
1159 			goto out;
1160 
1161 		/* setup fscache */
1162 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1163 			err = ceph_fscache_register_fs(fsc, fc);
1164 			if (err < 0)
1165 				goto out;
1166 		}
1167 
1168 		err = ceph_apply_test_dummy_encryption(fsc->sb, fc,
1169 						       fsc->mount_options);
1170 		if (err)
1171 			goto out;
1172 
1173 		doutc(cl, "mount opening path '%s'\n", path);
1174 
1175 		ceph_fs_debugfs_init(fsc);
1176 
1177 		root = open_root_dentry(fsc, path, started);
1178 		if (IS_ERR(root)) {
1179 			err = PTR_ERR(root);
1180 			goto out;
1181 		}
1182 		fsc->sb->s_root = dget(root);
1183 	} else {
1184 		root = dget(fsc->sb->s_root);
1185 	}
1186 
1187 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1188 	doutc(cl, "mount success\n");
1189 	mutex_unlock(&fsc->client->mount_mutex);
1190 	return root;
1191 
1192 out:
1193 	mutex_unlock(&fsc->client->mount_mutex);
1194 	ceph_fscrypt_free_dummy_policy(fsc);
1195 	return ERR_PTR(err);
1196 }
1197 
ceph_set_super(struct super_block * s,struct fs_context * fc)1198 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1199 {
1200 	struct ceph_fs_client *fsc = s->s_fs_info;
1201 	struct ceph_client *cl = fsc->client;
1202 	int ret;
1203 
1204 	doutc(cl, "%p\n", s);
1205 
1206 	s->s_maxbytes = MAX_LFS_FILESIZE;
1207 
1208 	s->s_xattr = ceph_xattr_handlers;
1209 	fsc->sb = s;
1210 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1211 
1212 	s->s_op = &ceph_super_ops;
1213 	s->s_d_op = &ceph_dentry_ops;
1214 	s->s_export_op = &ceph_export_ops;
1215 
1216 	s->s_time_gran = 1;
1217 	s->s_time_min = 0;
1218 	s->s_time_max = U32_MAX;
1219 	s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1220 
1221 	ceph_fscrypt_set_ops(s);
1222 
1223 	ret = set_anon_super_fc(s, fc);
1224 	if (ret != 0)
1225 		fsc->sb = NULL;
1226 	return ret;
1227 }
1228 
1229 /*
1230  * share superblock if same fs AND options
1231  */
ceph_compare_super(struct super_block * sb,struct fs_context * fc)1232 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1233 {
1234 	struct ceph_fs_client *new = fc->s_fs_info;
1235 	struct ceph_mount_options *fsopt = new->mount_options;
1236 	struct ceph_options *opt = new->client->options;
1237 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1238 	struct ceph_client *cl = fsc->client;
1239 
1240 	doutc(cl, "%p\n", sb);
1241 
1242 	if (compare_mount_options(fsopt, opt, fsc)) {
1243 		doutc(cl, "monitor(s)/mount options don't match\n");
1244 		return 0;
1245 	}
1246 	if ((opt->flags & CEPH_OPT_FSID) &&
1247 	    ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1248 		doutc(cl, "fsid doesn't match\n");
1249 		return 0;
1250 	}
1251 	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1252 		doutc(cl, "flags differ\n");
1253 		return 0;
1254 	}
1255 
1256 	if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1257 		doutc(cl, "client is blocklisted (and CLEANRECOVER is not set)\n");
1258 		return 0;
1259 	}
1260 
1261 	if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1262 		doutc(cl, "client has been forcibly unmounted\n");
1263 		return 0;
1264 	}
1265 
1266 	return 1;
1267 }
1268 
1269 /*
1270  * construct our own bdi so we can control readahead, etc.
1271  */
1272 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1273 
ceph_setup_bdi(struct super_block * sb,struct ceph_fs_client * fsc)1274 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1275 {
1276 	int err;
1277 
1278 	err = super_setup_bdi_name(sb, "ceph-%ld",
1279 				   atomic_long_inc_return(&bdi_seq));
1280 	if (err)
1281 		return err;
1282 
1283 	/* set ra_pages based on rasize mount option? */
1284 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1285 
1286 	/* set io_pages based on max osd read size */
1287 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1288 
1289 	return 0;
1290 }
1291 
ceph_get_tree(struct fs_context * fc)1292 static int ceph_get_tree(struct fs_context *fc)
1293 {
1294 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1295 	struct ceph_mount_options *fsopt = pctx->opts;
1296 	struct super_block *sb;
1297 	struct ceph_fs_client *fsc;
1298 	struct dentry *res;
1299 	int (*compare_super)(struct super_block *, struct fs_context *) =
1300 		ceph_compare_super;
1301 	int err;
1302 
1303 	dout("ceph_get_tree\n");
1304 
1305 	if (!fc->source)
1306 		return invalfc(fc, "No source");
1307 	if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1308 		return invalfc(fc, "No monitor address");
1309 
1310 	/* create client (which we may/may not use) */
1311 	fsc = create_fs_client(pctx->opts, pctx->copts);
1312 	pctx->opts = NULL;
1313 	pctx->copts = NULL;
1314 	if (IS_ERR(fsc)) {
1315 		err = PTR_ERR(fsc);
1316 		goto out_final;
1317 	}
1318 
1319 	err = ceph_mdsc_init(fsc);
1320 	if (err < 0)
1321 		goto out;
1322 
1323 	if (ceph_test_opt(fsc->client, NOSHARE))
1324 		compare_super = NULL;
1325 
1326 	fc->s_fs_info = fsc;
1327 	sb = sget_fc(fc, compare_super, ceph_set_super);
1328 	fc->s_fs_info = NULL;
1329 	if (IS_ERR(sb)) {
1330 		err = PTR_ERR(sb);
1331 		goto out;
1332 	}
1333 
1334 	if (ceph_sb_to_fs_client(sb) != fsc) {
1335 		destroy_fs_client(fsc);
1336 		fsc = ceph_sb_to_fs_client(sb);
1337 		dout("get_sb got existing client %p\n", fsc);
1338 	} else {
1339 		dout("get_sb using new client %p\n", fsc);
1340 		err = ceph_setup_bdi(sb, fsc);
1341 		if (err < 0)
1342 			goto out_splat;
1343 	}
1344 
1345 	res = ceph_real_mount(fsc, fc);
1346 	if (IS_ERR(res)) {
1347 		err = PTR_ERR(res);
1348 		goto out_splat;
1349 	}
1350 
1351 	doutc(fsc->client, "root %p inode %p ino %llx.%llx\n", res,
1352 		    d_inode(res), ceph_vinop(d_inode(res)));
1353 	fc->root = fsc->sb->s_root;
1354 	return 0;
1355 
1356 out_splat:
1357 	if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1358 		pr_info("No mds server is up or the cluster is laggy\n");
1359 		err = -EHOSTUNREACH;
1360 	}
1361 
1362 	ceph_mdsc_close_sessions(fsc->mdsc);
1363 	deactivate_locked_super(sb);
1364 	goto out_final;
1365 
1366 out:
1367 	destroy_fs_client(fsc);
1368 out_final:
1369 	dout("ceph_get_tree fail %d\n", err);
1370 	return err;
1371 }
1372 
ceph_free_fc(struct fs_context * fc)1373 static void ceph_free_fc(struct fs_context *fc)
1374 {
1375 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1376 
1377 	if (pctx) {
1378 		destroy_mount_options(pctx->opts);
1379 		ceph_destroy_options(pctx->copts);
1380 		kfree(pctx);
1381 	}
1382 }
1383 
ceph_reconfigure_fc(struct fs_context * fc)1384 static int ceph_reconfigure_fc(struct fs_context *fc)
1385 {
1386 	int err;
1387 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1388 	struct ceph_mount_options *fsopt = pctx->opts;
1389 	struct super_block *sb = fc->root->d_sb;
1390 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1391 
1392 	err = ceph_apply_test_dummy_encryption(sb, fc, fsopt);
1393 	if (err)
1394 		return err;
1395 
1396 	if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1397 		ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1398 	else
1399 		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1400 
1401 	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
1402 		ceph_set_mount_opt(fsc, SPARSEREAD);
1403 	else
1404 		ceph_clear_mount_opt(fsc, SPARSEREAD);
1405 
1406 	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1407 		kfree(fsc->mount_options->mon_addr);
1408 		fsc->mount_options->mon_addr = fsopt->mon_addr;
1409 		fsopt->mon_addr = NULL;
1410 		pr_notice_client(fsc->client,
1411 			"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_client *cl = fsc->client;
1532 	struct ceph_mds_client *mdsc = fsc->mdsc;
1533 	bool wait;
1534 
1535 	doutc(cl, "%p\n", s);
1536 
1537 	ceph_mdsc_pre_umount(mdsc);
1538 	flush_fs_workqueues(fsc);
1539 
1540 	/*
1541 	 * Though the kill_anon_super() will finally trigger the
1542 	 * sync_filesystem() anyway, we still need to do it here and
1543 	 * then bump the stage of shutdown. This will allow us to
1544 	 * drop any further message, which will increase the inodes'
1545 	 * i_count reference counters but makes no sense any more,
1546 	 * from MDSs.
1547 	 *
1548 	 * Without this when evicting the inodes it may fail in the
1549 	 * kill_anon_super(), which will trigger a warning when
1550 	 * destroying the fscrypt keyring and then possibly trigger
1551 	 * a further crash in ceph module when the iput() tries to
1552 	 * evict the inodes later.
1553 	 */
1554 	sync_filesystem(s);
1555 
1556 	spin_lock(&mdsc->stopping_lock);
1557 	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHING;
1558 	wait = !!atomic_read(&mdsc->stopping_blockers);
1559 	spin_unlock(&mdsc->stopping_lock);
1560 
1561 	if (wait && atomic_read(&mdsc->stopping_blockers)) {
1562 		long timeleft = wait_for_completion_killable_timeout(
1563 					&mdsc->stopping_waiter,
1564 					fsc->client->options->mount_timeout);
1565 		if (!timeleft) /* timed out */
1566 			pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1567 		else if (timeleft < 0) /* killed */
1568 			pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1569 	}
1570 
1571 	mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHED;
1572 	kill_anon_super(s);
1573 
1574 	fsc->client->extra_mon_dispatch = NULL;
1575 	ceph_fs_debugfs_cleanup(fsc);
1576 
1577 	ceph_fscache_unregister_fs(fsc);
1578 
1579 	destroy_fs_client(fsc);
1580 }
1581 
1582 static struct file_system_type ceph_fs_type = {
1583 	.owner		= THIS_MODULE,
1584 	.name		= "ceph",
1585 	.init_fs_context = ceph_init_fs_context,
1586 	.kill_sb	= ceph_kill_sb,
1587 	.fs_flags	= FS_RENAME_DOES_D_MOVE | FS_ALLOW_IDMAP,
1588 };
1589 MODULE_ALIAS_FS("ceph");
1590 
ceph_force_reconnect(struct super_block * sb)1591 int ceph_force_reconnect(struct super_block *sb)
1592 {
1593 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1594 	int err = 0;
1595 
1596 	fsc->mount_state = CEPH_MOUNT_RECOVER;
1597 	__ceph_umount_begin(fsc);
1598 
1599 	/* Make sure all page caches get invalidated.
1600 	 * see remove_session_caps_cb() */
1601 	flush_workqueue(fsc->inode_wq);
1602 
1603 	/* In case that we were blocklisted. This also reset
1604 	 * all mon/osd connections */
1605 	ceph_reset_client_addr(fsc->client);
1606 
1607 	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1608 
1609 	fsc->blocklisted = false;
1610 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1611 
1612 	if (sb->s_root) {
1613 		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1614 					CEPH_STAT_CAP_INODE, true);
1615 	}
1616 	return err;
1617 }
1618 
init_ceph(void)1619 static int __init init_ceph(void)
1620 {
1621 	int ret = init_caches();
1622 	if (ret)
1623 		goto out;
1624 
1625 	ceph_flock_init();
1626 	ret = register_filesystem(&ceph_fs_type);
1627 	if (ret)
1628 		goto out_caches;
1629 
1630 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1631 
1632 	return 0;
1633 
1634 out_caches:
1635 	destroy_caches();
1636 out:
1637 	return ret;
1638 }
1639 
exit_ceph(void)1640 static void __exit exit_ceph(void)
1641 {
1642 	dout("exit_ceph\n");
1643 	unregister_filesystem(&ceph_fs_type);
1644 	destroy_caches();
1645 }
1646 
param_set_metrics(const char * val,const struct kernel_param * kp)1647 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1648 {
1649 	struct ceph_fs_client *fsc;
1650 	int ret;
1651 
1652 	ret = param_set_bool(val, kp);
1653 	if (ret) {
1654 		pr_err("Failed to parse sending metrics switch value '%s'\n",
1655 		       val);
1656 		return ret;
1657 	} else if (!disable_send_metrics) {
1658 		// wake up all the mds clients
1659 		spin_lock(&ceph_fsc_lock);
1660 		list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1661 			metric_schedule_delayed(&fsc->mdsc->metric);
1662 		}
1663 		spin_unlock(&ceph_fsc_lock);
1664 	}
1665 
1666 	return 0;
1667 }
1668 
1669 static const struct kernel_param_ops param_ops_metrics = {
1670 	.set = param_set_metrics,
1671 	.get = param_get_bool,
1672 };
1673 
1674 bool disable_send_metrics = false;
1675 module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1676 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1677 
1678 /* for both v1 and v2 syntax */
1679 static bool mount_support = true;
1680 static const struct kernel_param_ops param_ops_mount_syntax = {
1681 	.get = param_get_bool,
1682 };
1683 module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1684 module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1685 
1686 bool enable_unsafe_idmap = false;
1687 module_param(enable_unsafe_idmap, bool, 0644);
1688 MODULE_PARM_DESC(enable_unsafe_idmap,
1689 		 "Allow to use idmapped mounts with MDS without CEPHFS_FEATURE_HAS_OWNER_UIDGID");
1690 
1691 module_init(init_ceph);
1692 module_exit(exit_ceph);
1693 
1694 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1695 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1696 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1697 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1698 MODULE_LICENSE("GPL");
1699