xref: /linux/fs/ceph/super.c (revision 2dbc0838bcf24ca59cabc3130cf3b1d6809cdcd4)
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/parser.h>
13 #include <linux/sched.h>
14 #include <linux/seq_file.h>
15 #include <linux/slab.h>
16 #include <linux/statfs.h>
17 #include <linux/string.h>
18 
19 #include "super.h"
20 #include "mds_client.h"
21 #include "cache.h"
22 
23 #include <linux/ceph/ceph_features.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/mon_client.h>
26 #include <linux/ceph/auth.h>
27 #include <linux/ceph/debugfs.h>
28 
29 /*
30  * Ceph superblock operations
31  *
32  * Handle the basics of mounting, unmounting.
33  */
34 
35 /*
36  * super ops
37  */
38 static void ceph_put_super(struct super_block *s)
39 {
40 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
41 
42 	dout("put_super\n");
43 	ceph_mdsc_close_sessions(fsc->mdsc);
44 }
45 
46 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
47 {
48 	struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
49 	struct ceph_mon_client *monc = &fsc->client->monc;
50 	struct ceph_statfs st;
51 	u64 fsid;
52 	int err;
53 	u64 data_pool;
54 
55 	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
56 		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
57 	} else {
58 		data_pool = CEPH_NOPOOL;
59 	}
60 
61 	dout("statfs\n");
62 	err = ceph_monc_do_statfs(monc, data_pool, &st);
63 	if (err < 0)
64 		return err;
65 
66 	/* fill in kstatfs */
67 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
68 
69 	/*
70 	 * express utilization in terms of large blocks to avoid
71 	 * overflow on 32-bit machines.
72 	 *
73 	 * NOTE: for the time being, we make bsize == frsize to humor
74 	 * not-yet-ancient versions of glibc that are broken.
75 	 * Someday, we will probably want to report a real block
76 	 * size...  whatever that may mean for a network file system!
77 	 */
78 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
79 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
80 
81 	/*
82 	 * By default use root quota for stats; fallback to overall filesystem
83 	 * usage if using 'noquotadf' mount option or if the root dir doesn't
84 	 * have max_bytes quota set.
85 	 */
86 	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
87 	    !ceph_quota_update_statfs(fsc, buf)) {
88 		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
89 		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
90 		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
91 	}
92 
93 	buf->f_files = le64_to_cpu(st.num_objects);
94 	buf->f_ffree = -1;
95 	buf->f_namelen = NAME_MAX;
96 
97 	/* Must convert the fsid, for consistent values across arches */
98 	mutex_lock(&monc->mutex);
99 	fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
100 	       le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
101 	mutex_unlock(&monc->mutex);
102 
103 	buf->f_fsid.val[0] = fsid & 0xffffffff;
104 	buf->f_fsid.val[1] = fsid >> 32;
105 
106 	return 0;
107 }
108 
109 
110 static int ceph_sync_fs(struct super_block *sb, int wait)
111 {
112 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
113 
114 	if (!wait) {
115 		dout("sync_fs (non-blocking)\n");
116 		ceph_flush_dirty_caps(fsc->mdsc);
117 		dout("sync_fs (non-blocking) done\n");
118 		return 0;
119 	}
120 
121 	dout("sync_fs (blocking)\n");
122 	ceph_osdc_sync(&fsc->client->osdc);
123 	ceph_mdsc_sync(fsc->mdsc);
124 	dout("sync_fs (blocking) done\n");
125 	return 0;
126 }
127 
128 /*
129  * mount options
130  */
131 enum {
132 	Opt_wsize,
133 	Opt_rsize,
134 	Opt_rasize,
135 	Opt_caps_wanted_delay_min,
136 	Opt_caps_wanted_delay_max,
137 	Opt_caps_max,
138 	Opt_readdir_max_entries,
139 	Opt_readdir_max_bytes,
140 	Opt_congestion_kb,
141 	Opt_last_int,
142 	/* int args above */
143 	Opt_snapdirname,
144 	Opt_mds_namespace,
145 	Opt_fscache_uniq,
146 	Opt_last_string,
147 	/* string args above */
148 	Opt_dirstat,
149 	Opt_nodirstat,
150 	Opt_rbytes,
151 	Opt_norbytes,
152 	Opt_asyncreaddir,
153 	Opt_noasyncreaddir,
154 	Opt_dcache,
155 	Opt_nodcache,
156 	Opt_ino32,
157 	Opt_noino32,
158 	Opt_fscache,
159 	Opt_nofscache,
160 	Opt_poolperm,
161 	Opt_nopoolperm,
162 	Opt_require_active_mds,
163 	Opt_norequire_active_mds,
164 #ifdef CONFIG_CEPH_FS_POSIX_ACL
165 	Opt_acl,
166 #endif
167 	Opt_noacl,
168 	Opt_quotadf,
169 	Opt_noquotadf,
170 	Opt_copyfrom,
171 	Opt_nocopyfrom,
172 };
173 
174 static match_table_t fsopt_tokens = {
175 	{Opt_wsize, "wsize=%d"},
176 	{Opt_rsize, "rsize=%d"},
177 	{Opt_rasize, "rasize=%d"},
178 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
179 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
180 	{Opt_caps_max, "caps_max=%d"},
181 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
182 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
183 	{Opt_congestion_kb, "write_congestion_kb=%d"},
184 	/* int args above */
185 	{Opt_snapdirname, "snapdirname=%s"},
186 	{Opt_mds_namespace, "mds_namespace=%s"},
187 	{Opt_fscache_uniq, "fsc=%s"},
188 	/* string args above */
189 	{Opt_dirstat, "dirstat"},
190 	{Opt_nodirstat, "nodirstat"},
191 	{Opt_rbytes, "rbytes"},
192 	{Opt_norbytes, "norbytes"},
193 	{Opt_asyncreaddir, "asyncreaddir"},
194 	{Opt_noasyncreaddir, "noasyncreaddir"},
195 	{Opt_dcache, "dcache"},
196 	{Opt_nodcache, "nodcache"},
197 	{Opt_ino32, "ino32"},
198 	{Opt_noino32, "noino32"},
199 	{Opt_fscache, "fsc"},
200 	{Opt_nofscache, "nofsc"},
201 	{Opt_poolperm, "poolperm"},
202 	{Opt_nopoolperm, "nopoolperm"},
203 	{Opt_require_active_mds, "require_active_mds"},
204 	{Opt_norequire_active_mds, "norequire_active_mds"},
205 #ifdef CONFIG_CEPH_FS_POSIX_ACL
206 	{Opt_acl, "acl"},
207 #endif
208 	{Opt_noacl, "noacl"},
209 	{Opt_quotadf, "quotadf"},
210 	{Opt_noquotadf, "noquotadf"},
211 	{Opt_copyfrom, "copyfrom"},
212 	{Opt_nocopyfrom, "nocopyfrom"},
213 	{-1, NULL}
214 };
215 
216 static int parse_fsopt_token(char *c, void *private)
217 {
218 	struct ceph_mount_options *fsopt = private;
219 	substring_t argstr[MAX_OPT_ARGS];
220 	int token, intval, ret;
221 
222 	token = match_token((char *)c, fsopt_tokens, argstr);
223 	if (token < 0)
224 		return -EINVAL;
225 
226 	if (token < Opt_last_int) {
227 		ret = match_int(&argstr[0], &intval);
228 		if (ret < 0) {
229 			pr_err("bad option arg (not int) at '%s'\n", c);
230 			return ret;
231 		}
232 		dout("got int token %d val %d\n", token, intval);
233 	} else if (token > Opt_last_int && token < Opt_last_string) {
234 		dout("got string token %d val %s\n", token,
235 		     argstr[0].from);
236 	} else {
237 		dout("got token %d\n", token);
238 	}
239 
240 	switch (token) {
241 	case Opt_snapdirname:
242 		kfree(fsopt->snapdir_name);
243 		fsopt->snapdir_name = kstrndup(argstr[0].from,
244 					       argstr[0].to-argstr[0].from,
245 					       GFP_KERNEL);
246 		if (!fsopt->snapdir_name)
247 			return -ENOMEM;
248 		break;
249 	case Opt_mds_namespace:
250 		kfree(fsopt->mds_namespace);
251 		fsopt->mds_namespace = kstrndup(argstr[0].from,
252 						argstr[0].to-argstr[0].from,
253 						GFP_KERNEL);
254 		if (!fsopt->mds_namespace)
255 			return -ENOMEM;
256 		break;
257 	case Opt_fscache_uniq:
258 		kfree(fsopt->fscache_uniq);
259 		fsopt->fscache_uniq = kstrndup(argstr[0].from,
260 					       argstr[0].to-argstr[0].from,
261 					       GFP_KERNEL);
262 		if (!fsopt->fscache_uniq)
263 			return -ENOMEM;
264 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
265 		break;
266 		/* misc */
267 	case Opt_wsize:
268 		if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
269 			return -EINVAL;
270 		fsopt->wsize = ALIGN(intval, PAGE_SIZE);
271 		break;
272 	case Opt_rsize:
273 		if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
274 			return -EINVAL;
275 		fsopt->rsize = ALIGN(intval, PAGE_SIZE);
276 		break;
277 	case Opt_rasize:
278 		if (intval < 0)
279 			return -EINVAL;
280 		fsopt->rasize = ALIGN(intval, PAGE_SIZE);
281 		break;
282 	case Opt_caps_wanted_delay_min:
283 		if (intval < 1)
284 			return -EINVAL;
285 		fsopt->caps_wanted_delay_min = intval;
286 		break;
287 	case Opt_caps_wanted_delay_max:
288 		if (intval < 1)
289 			return -EINVAL;
290 		fsopt->caps_wanted_delay_max = intval;
291 		break;
292 	case Opt_caps_max:
293 		if (intval < 0)
294 			return -EINVAL;
295 		fsopt->caps_max = intval;
296 		break;
297 	case Opt_readdir_max_entries:
298 		if (intval < 1)
299 			return -EINVAL;
300 		fsopt->max_readdir = intval;
301 		break;
302 	case Opt_readdir_max_bytes:
303 		if (intval < (int)PAGE_SIZE && intval != 0)
304 			return -EINVAL;
305 		fsopt->max_readdir_bytes = intval;
306 		break;
307 	case Opt_congestion_kb:
308 		if (intval < 1024) /* at least 1M */
309 			return -EINVAL;
310 		fsopt->congestion_kb = intval;
311 		break;
312 	case Opt_dirstat:
313 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
314 		break;
315 	case Opt_nodirstat:
316 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
317 		break;
318 	case Opt_rbytes:
319 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
320 		break;
321 	case Opt_norbytes:
322 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
323 		break;
324 	case Opt_asyncreaddir:
325 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
326 		break;
327 	case Opt_noasyncreaddir:
328 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
329 		break;
330 	case Opt_dcache:
331 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
332 		break;
333 	case Opt_nodcache:
334 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
335 		break;
336 	case Opt_ino32:
337 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
338 		break;
339 	case Opt_noino32:
340 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
341 		break;
342 	case Opt_fscache:
343 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
344 		kfree(fsopt->fscache_uniq);
345 		fsopt->fscache_uniq = NULL;
346 		break;
347 	case Opt_nofscache:
348 		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
349 		kfree(fsopt->fscache_uniq);
350 		fsopt->fscache_uniq = NULL;
351 		break;
352 	case Opt_poolperm:
353 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
354 		break;
355 	case Opt_nopoolperm:
356 		fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
357 		break;
358 	case Opt_require_active_mds:
359 		fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
360 		break;
361 	case Opt_norequire_active_mds:
362 		fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
363 		break;
364 	case Opt_quotadf:
365 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
366 		break;
367 	case Opt_noquotadf:
368 		fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
369 		break;
370 	case Opt_copyfrom:
371 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
372 		break;
373 	case Opt_nocopyfrom:
374 		fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
375 		break;
376 #ifdef CONFIG_CEPH_FS_POSIX_ACL
377 	case Opt_acl:
378 		fsopt->sb_flags |= SB_POSIXACL;
379 		break;
380 #endif
381 	case Opt_noacl:
382 		fsopt->sb_flags &= ~SB_POSIXACL;
383 		break;
384 	default:
385 		BUG_ON(token);
386 	}
387 	return 0;
388 }
389 
390 static void destroy_mount_options(struct ceph_mount_options *args)
391 {
392 	dout("destroy_mount_options %p\n", args);
393 	kfree(args->snapdir_name);
394 	kfree(args->mds_namespace);
395 	kfree(args->server_path);
396 	kfree(args->fscache_uniq);
397 	kfree(args);
398 }
399 
400 static int strcmp_null(const char *s1, const char *s2)
401 {
402 	if (!s1 && !s2)
403 		return 0;
404 	if (s1 && !s2)
405 		return -1;
406 	if (!s1 && s2)
407 		return 1;
408 	return strcmp(s1, s2);
409 }
410 
411 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
412 				 struct ceph_options *new_opt,
413 				 struct ceph_fs_client *fsc)
414 {
415 	struct ceph_mount_options *fsopt1 = new_fsopt;
416 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
417 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
418 	int ret;
419 
420 	ret = memcmp(fsopt1, fsopt2, ofs);
421 	if (ret)
422 		return ret;
423 
424 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
425 	if (ret)
426 		return ret;
427 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
428 	if (ret)
429 		return ret;
430 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
431 	if (ret)
432 		return ret;
433 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
434 	if (ret)
435 		return ret;
436 
437 	return ceph_compare_options(new_opt, fsc->client);
438 }
439 
440 static int parse_mount_options(struct ceph_mount_options **pfsopt,
441 			       struct ceph_options **popt,
442 			       int flags, char *options,
443 			       const char *dev_name)
444 {
445 	struct ceph_mount_options *fsopt;
446 	const char *dev_name_end;
447 	int err;
448 
449 	if (!dev_name || !*dev_name)
450 		return -EINVAL;
451 
452 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
453 	if (!fsopt)
454 		return -ENOMEM;
455 
456 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
457 
458 	fsopt->sb_flags = flags;
459 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
460 
461 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
462 	fsopt->rsize = CEPH_MAX_READ_SIZE;
463 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
464 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
465 	if (!fsopt->snapdir_name) {
466 		err = -ENOMEM;
467 		goto out;
468 	}
469 
470 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
471 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
472 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
473 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
474 	fsopt->congestion_kb = default_congestion_kb();
475 
476 	/*
477 	 * Distinguish the server list from the path in "dev_name".
478 	 * Internally we do not include the leading '/' in the path.
479 	 *
480 	 * "dev_name" will look like:
481 	 *     <server_spec>[,<server_spec>...]:[<path>]
482 	 * where
483 	 *     <server_spec> is <ip>[:<port>]
484 	 *     <path> is optional, but if present must begin with '/'
485 	 */
486 	dev_name_end = strchr(dev_name, '/');
487 	if (dev_name_end) {
488 		if (strlen(dev_name_end) > 1) {
489 			fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
490 			if (!fsopt->server_path) {
491 				err = -ENOMEM;
492 				goto out;
493 			}
494 		}
495 	} else {
496 		dev_name_end = dev_name + strlen(dev_name);
497 	}
498 	err = -EINVAL;
499 	dev_name_end--;		/* back up to ':' separator */
500 	if (dev_name_end < dev_name || *dev_name_end != ':') {
501 		pr_err("device name is missing path (no : separator in %s)\n",
502 				dev_name);
503 		goto out;
504 	}
505 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
506 	if (fsopt->server_path)
507 		dout("server path '%s'\n", fsopt->server_path);
508 
509 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
510 				 parse_fsopt_token, (void *)fsopt);
511 	if (IS_ERR(*popt)) {
512 		err = PTR_ERR(*popt);
513 		goto out;
514 	}
515 
516 	/* success */
517 	*pfsopt = fsopt;
518 	return 0;
519 
520 out:
521 	destroy_mount_options(fsopt);
522 	return err;
523 }
524 
525 /**
526  * ceph_show_options - Show mount options in /proc/mounts
527  * @m: seq_file to write to
528  * @root: root of that (sub)tree
529  */
530 static int ceph_show_options(struct seq_file *m, struct dentry *root)
531 {
532 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
533 	struct ceph_mount_options *fsopt = fsc->mount_options;
534 	size_t pos;
535 	int ret;
536 
537 	/* a comma between MNT/MS and client options */
538 	seq_putc(m, ',');
539 	pos = m->count;
540 
541 	ret = ceph_print_client_options(m, fsc->client, false);
542 	if (ret)
543 		return ret;
544 
545 	/* retract our comma if no client options */
546 	if (m->count == pos)
547 		m->count--;
548 
549 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
550 		seq_puts(m, ",dirstat");
551 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
552 		seq_puts(m, ",rbytes");
553 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
554 		seq_puts(m, ",noasyncreaddir");
555 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
556 		seq_puts(m, ",nodcache");
557 	if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
558 		seq_puts(m, ",ino32");
559 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
560 		seq_show_option(m, "fsc", fsopt->fscache_uniq);
561 	}
562 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
563 		seq_puts(m, ",nopoolperm");
564 	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
565 		seq_puts(m, ",noquotadf");
566 
567 #ifdef CONFIG_CEPH_FS_POSIX_ACL
568 	if (fsopt->sb_flags & SB_POSIXACL)
569 		seq_puts(m, ",acl");
570 	else
571 		seq_puts(m, ",noacl");
572 #endif
573 
574 	if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
575 		seq_puts(m, ",copyfrom");
576 
577 	if (fsopt->mds_namespace)
578 		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
579 	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
580 		seq_printf(m, ",wsize=%d", fsopt->wsize);
581 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
582 		seq_printf(m, ",rsize=%d", fsopt->rsize);
583 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
584 		seq_printf(m, ",rasize=%d", fsopt->rasize);
585 	if (fsopt->congestion_kb != default_congestion_kb())
586 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
587 	if (fsopt->caps_max)
588 		seq_printf(m, ",caps_max=%d", fsopt->caps_max);
589 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
590 		seq_printf(m, ",caps_wanted_delay_min=%d",
591 			 fsopt->caps_wanted_delay_min);
592 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
593 		seq_printf(m, ",caps_wanted_delay_max=%d",
594 			   fsopt->caps_wanted_delay_max);
595 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
596 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
597 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
598 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
599 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
600 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
601 
602 	return 0;
603 }
604 
605 /*
606  * handle any mon messages the standard library doesn't understand.
607  * return error if we don't either.
608  */
609 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
610 {
611 	struct ceph_fs_client *fsc = client->private;
612 	int type = le16_to_cpu(msg->hdr.type);
613 
614 	switch (type) {
615 	case CEPH_MSG_MDS_MAP:
616 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
617 		return 0;
618 	case CEPH_MSG_FS_MAP_USER:
619 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
620 		return 0;
621 	default:
622 		return -1;
623 	}
624 }
625 
626 /*
627  * create a new fs client
628  *
629  * Success or not, this function consumes @fsopt and @opt.
630  */
631 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
632 					struct ceph_options *opt)
633 {
634 	struct ceph_fs_client *fsc;
635 	int page_count;
636 	size_t size;
637 	int err;
638 
639 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
640 	if (!fsc) {
641 		err = -ENOMEM;
642 		goto fail;
643 	}
644 
645 	fsc->client = ceph_create_client(opt, fsc);
646 	if (IS_ERR(fsc->client)) {
647 		err = PTR_ERR(fsc->client);
648 		goto fail;
649 	}
650 	opt = NULL; /* fsc->client now owns this */
651 
652 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
653 	ceph_set_opt(fsc->client, ABORT_ON_FULL);
654 
655 	if (!fsopt->mds_namespace) {
656 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
657 				   0, true);
658 	} else {
659 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
660 				   0, false);
661 	}
662 
663 	fsc->mount_options = fsopt;
664 
665 	fsc->sb = NULL;
666 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
667 
668 	atomic_long_set(&fsc->writeback_count, 0);
669 
670 	err = -ENOMEM;
671 	/*
672 	 * The number of concurrent works can be high but they don't need
673 	 * to be processed in parallel, limit concurrency.
674 	 */
675 	fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
676 	if (!fsc->inode_wq)
677 		goto fail_client;
678 	fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
679 	if (!fsc->cap_wq)
680 		goto fail_inode_wq;
681 
682 	/* set up mempools */
683 	err = -ENOMEM;
684 	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
685 	size = sizeof (struct page *) * (page_count ? page_count : 1);
686 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
687 	if (!fsc->wb_pagevec_pool)
688 		goto fail_cap_wq;
689 
690 	return fsc;
691 
692 fail_cap_wq:
693 	destroy_workqueue(fsc->cap_wq);
694 fail_inode_wq:
695 	destroy_workqueue(fsc->inode_wq);
696 fail_client:
697 	ceph_destroy_client(fsc->client);
698 fail:
699 	kfree(fsc);
700 	if (opt)
701 		ceph_destroy_options(opt);
702 	destroy_mount_options(fsopt);
703 	return ERR_PTR(err);
704 }
705 
706 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
707 {
708 	flush_workqueue(fsc->inode_wq);
709 	flush_workqueue(fsc->cap_wq);
710 }
711 
712 static void destroy_fs_client(struct ceph_fs_client *fsc)
713 {
714 	dout("destroy_fs_client %p\n", fsc);
715 
716 	destroy_workqueue(fsc->inode_wq);
717 	destroy_workqueue(fsc->cap_wq);
718 
719 	mempool_destroy(fsc->wb_pagevec_pool);
720 
721 	destroy_mount_options(fsc->mount_options);
722 
723 	ceph_destroy_client(fsc->client);
724 
725 	kfree(fsc);
726 	dout("destroy_fs_client %p done\n", fsc);
727 }
728 
729 /*
730  * caches
731  */
732 struct kmem_cache *ceph_inode_cachep;
733 struct kmem_cache *ceph_cap_cachep;
734 struct kmem_cache *ceph_cap_flush_cachep;
735 struct kmem_cache *ceph_dentry_cachep;
736 struct kmem_cache *ceph_file_cachep;
737 struct kmem_cache *ceph_dir_file_cachep;
738 
739 static void ceph_inode_init_once(void *foo)
740 {
741 	struct ceph_inode_info *ci = foo;
742 	inode_init_once(&ci->vfs_inode);
743 }
744 
745 static int __init init_caches(void)
746 {
747 	int error = -ENOMEM;
748 
749 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
750 				      sizeof(struct ceph_inode_info),
751 				      __alignof__(struct ceph_inode_info),
752 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
753 				      SLAB_ACCOUNT, ceph_inode_init_once);
754 	if (!ceph_inode_cachep)
755 		return -ENOMEM;
756 
757 	ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
758 	if (!ceph_cap_cachep)
759 		goto bad_cap;
760 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
761 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
762 	if (!ceph_cap_flush_cachep)
763 		goto bad_cap_flush;
764 
765 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
766 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
767 	if (!ceph_dentry_cachep)
768 		goto bad_dentry;
769 
770 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
771 	if (!ceph_file_cachep)
772 		goto bad_file;
773 
774 	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
775 	if (!ceph_dir_file_cachep)
776 		goto bad_dir_file;
777 
778 	error = ceph_fscache_register();
779 	if (error)
780 		goto bad_fscache;
781 
782 	return 0;
783 
784 bad_fscache:
785 	kmem_cache_destroy(ceph_dir_file_cachep);
786 bad_dir_file:
787 	kmem_cache_destroy(ceph_file_cachep);
788 bad_file:
789 	kmem_cache_destroy(ceph_dentry_cachep);
790 bad_dentry:
791 	kmem_cache_destroy(ceph_cap_flush_cachep);
792 bad_cap_flush:
793 	kmem_cache_destroy(ceph_cap_cachep);
794 bad_cap:
795 	kmem_cache_destroy(ceph_inode_cachep);
796 	return error;
797 }
798 
799 static void destroy_caches(void)
800 {
801 	/*
802 	 * Make sure all delayed rcu free inodes are flushed before we
803 	 * destroy cache.
804 	 */
805 	rcu_barrier();
806 
807 	kmem_cache_destroy(ceph_inode_cachep);
808 	kmem_cache_destroy(ceph_cap_cachep);
809 	kmem_cache_destroy(ceph_cap_flush_cachep);
810 	kmem_cache_destroy(ceph_dentry_cachep);
811 	kmem_cache_destroy(ceph_file_cachep);
812 	kmem_cache_destroy(ceph_dir_file_cachep);
813 
814 	ceph_fscache_unregister();
815 }
816 
817 
818 /*
819  * ceph_umount_begin - initiate forced umount.  Tear down down the
820  * mount, skipping steps that may hang while waiting for server(s).
821  */
822 static void ceph_umount_begin(struct super_block *sb)
823 {
824 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
825 
826 	dout("ceph_umount_begin - starting forced umount\n");
827 	if (!fsc)
828 		return;
829 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
830 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
831 	ceph_mdsc_force_umount(fsc->mdsc);
832 	return;
833 }
834 
835 static int ceph_remount(struct super_block *sb, int *flags, char *data)
836 {
837 	sync_filesystem(sb);
838 	return 0;
839 }
840 
841 static const struct super_operations ceph_super_ops = {
842 	.alloc_inode	= ceph_alloc_inode,
843 	.destroy_inode	= ceph_destroy_inode,
844 	.free_inode	= ceph_free_inode,
845 	.write_inode    = ceph_write_inode,
846 	.drop_inode	= ceph_drop_inode,
847 	.sync_fs        = ceph_sync_fs,
848 	.put_super	= ceph_put_super,
849 	.remount_fs	= ceph_remount,
850 	.show_options   = ceph_show_options,
851 	.statfs		= ceph_statfs,
852 	.umount_begin   = ceph_umount_begin,
853 };
854 
855 /*
856  * Bootstrap mount by opening the root directory.  Note the mount
857  * @started time from caller, and time out if this takes too long.
858  */
859 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
860 				       const char *path,
861 				       unsigned long started)
862 {
863 	struct ceph_mds_client *mdsc = fsc->mdsc;
864 	struct ceph_mds_request *req = NULL;
865 	int err;
866 	struct dentry *root;
867 
868 	/* open dir */
869 	dout("open_root_inode opening '%s'\n", path);
870 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
871 	if (IS_ERR(req))
872 		return ERR_CAST(req);
873 	req->r_path1 = kstrdup(path, GFP_NOFS);
874 	if (!req->r_path1) {
875 		root = ERR_PTR(-ENOMEM);
876 		goto out;
877 	}
878 
879 	req->r_ino1.ino = CEPH_INO_ROOT;
880 	req->r_ino1.snap = CEPH_NOSNAP;
881 	req->r_started = started;
882 	req->r_timeout = fsc->client->options->mount_timeout;
883 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
884 	req->r_num_caps = 2;
885 	err = ceph_mdsc_do_request(mdsc, NULL, req);
886 	if (err == 0) {
887 		struct inode *inode = req->r_target_inode;
888 		req->r_target_inode = NULL;
889 		dout("open_root_inode success\n");
890 		root = d_make_root(inode);
891 		if (!root) {
892 			root = ERR_PTR(-ENOMEM);
893 			goto out;
894 		}
895 		dout("open_root_inode success, root dentry is %p\n", root);
896 	} else {
897 		root = ERR_PTR(err);
898 	}
899 out:
900 	ceph_mdsc_put_request(req);
901 	return root;
902 }
903 
904 
905 
906 
907 /*
908  * mount: join the ceph cluster, and open root directory.
909  */
910 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
911 {
912 	int err;
913 	unsigned long started = jiffies;  /* note the start time */
914 	struct dentry *root;
915 
916 	dout("mount start %p\n", fsc);
917 	mutex_lock(&fsc->client->mount_mutex);
918 
919 	if (!fsc->sb->s_root) {
920 		const char *path;
921 		err = __ceph_open_session(fsc->client, started);
922 		if (err < 0)
923 			goto out;
924 
925 		/* setup fscache */
926 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
927 			err = ceph_fscache_register_fs(fsc);
928 			if (err < 0)
929 				goto out;
930 		}
931 
932 		if (!fsc->mount_options->server_path) {
933 			path = "";
934 			dout("mount opening path \\t\n");
935 		} else {
936 			path = fsc->mount_options->server_path + 1;
937 			dout("mount opening path %s\n", path);
938 		}
939 
940 		ceph_fs_debugfs_init(fsc);
941 
942 		root = open_root_dentry(fsc, path, started);
943 		if (IS_ERR(root)) {
944 			err = PTR_ERR(root);
945 			goto out;
946 		}
947 		fsc->sb->s_root = dget(root);
948 	} else {
949 		root = dget(fsc->sb->s_root);
950 	}
951 
952 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
953 	dout("mount success\n");
954 	mutex_unlock(&fsc->client->mount_mutex);
955 	return root;
956 
957 out:
958 	mutex_unlock(&fsc->client->mount_mutex);
959 	return ERR_PTR(err);
960 }
961 
962 static int ceph_set_super(struct super_block *s, void *data)
963 {
964 	struct ceph_fs_client *fsc = data;
965 	int ret;
966 
967 	dout("set_super %p data %p\n", s, data);
968 
969 	s->s_flags = fsc->mount_options->sb_flags;
970 	s->s_maxbytes = MAX_LFS_FILESIZE;
971 
972 	s->s_xattr = ceph_xattr_handlers;
973 	s->s_fs_info = fsc;
974 	fsc->sb = s;
975 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
976 
977 	s->s_op = &ceph_super_ops;
978 	s->s_d_op = &ceph_dentry_ops;
979 	s->s_export_op = &ceph_export_ops;
980 
981 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
982 
983 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
984 	if (ret != 0)
985 		goto fail;
986 
987 	return ret;
988 
989 fail:
990 	s->s_fs_info = NULL;
991 	fsc->sb = NULL;
992 	return ret;
993 }
994 
995 /*
996  * share superblock if same fs AND options
997  */
998 static int ceph_compare_super(struct super_block *sb, void *data)
999 {
1000 	struct ceph_fs_client *new = data;
1001 	struct ceph_mount_options *fsopt = new->mount_options;
1002 	struct ceph_options *opt = new->client->options;
1003 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
1004 
1005 	dout("ceph_compare_super %p\n", sb);
1006 
1007 	if (compare_mount_options(fsopt, opt, other)) {
1008 		dout("monitor(s)/mount options don't match\n");
1009 		return 0;
1010 	}
1011 	if ((opt->flags & CEPH_OPT_FSID) &&
1012 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
1013 		dout("fsid doesn't match\n");
1014 		return 0;
1015 	}
1016 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
1017 		dout("flags differ\n");
1018 		return 0;
1019 	}
1020 	return 1;
1021 }
1022 
1023 /*
1024  * construct our own bdi so we can control readahead, etc.
1025  */
1026 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1027 
1028 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1029 {
1030 	int err;
1031 
1032 	err = super_setup_bdi_name(sb, "ceph-%ld",
1033 				   atomic_long_inc_return(&bdi_seq));
1034 	if (err)
1035 		return err;
1036 
1037 	/* set ra_pages based on rasize mount option? */
1038 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1039 
1040 	/* set io_pages based on max osd read size */
1041 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1042 
1043 	return 0;
1044 }
1045 
1046 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1047 		       int flags, const char *dev_name, void *data)
1048 {
1049 	struct super_block *sb;
1050 	struct ceph_fs_client *fsc;
1051 	struct dentry *res;
1052 	int err;
1053 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1054 	struct ceph_mount_options *fsopt = NULL;
1055 	struct ceph_options *opt = NULL;
1056 
1057 	dout("ceph_mount\n");
1058 
1059 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1060 	flags |= SB_POSIXACL;
1061 #endif
1062 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1063 	if (err < 0) {
1064 		res = ERR_PTR(err);
1065 		goto out_final;
1066 	}
1067 
1068 	/* create client (which we may/may not use) */
1069 	fsc = create_fs_client(fsopt, opt);
1070 	if (IS_ERR(fsc)) {
1071 		res = ERR_CAST(fsc);
1072 		goto out_final;
1073 	}
1074 
1075 	err = ceph_mdsc_init(fsc);
1076 	if (err < 0) {
1077 		res = ERR_PTR(err);
1078 		goto out;
1079 	}
1080 
1081 	if (ceph_test_opt(fsc->client, NOSHARE))
1082 		compare_super = NULL;
1083 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1084 	if (IS_ERR(sb)) {
1085 		res = ERR_CAST(sb);
1086 		goto out;
1087 	}
1088 
1089 	if (ceph_sb_to_client(sb) != fsc) {
1090 		ceph_mdsc_destroy(fsc);
1091 		destroy_fs_client(fsc);
1092 		fsc = ceph_sb_to_client(sb);
1093 		dout("get_sb got existing client %p\n", fsc);
1094 	} else {
1095 		dout("get_sb using new client %p\n", fsc);
1096 		err = ceph_setup_bdi(sb, fsc);
1097 		if (err < 0) {
1098 			res = ERR_PTR(err);
1099 			goto out_splat;
1100 		}
1101 	}
1102 
1103 	res = ceph_real_mount(fsc);
1104 	if (IS_ERR(res))
1105 		goto out_splat;
1106 	dout("root %p inode %p ino %llx.%llx\n", res,
1107 	     d_inode(res), ceph_vinop(d_inode(res)));
1108 	return res;
1109 
1110 out_splat:
1111 	ceph_mdsc_close_sessions(fsc->mdsc);
1112 	deactivate_locked_super(sb);
1113 	goto out_final;
1114 
1115 out:
1116 	ceph_mdsc_destroy(fsc);
1117 	destroy_fs_client(fsc);
1118 out_final:
1119 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1120 	return res;
1121 }
1122 
1123 static void ceph_kill_sb(struct super_block *s)
1124 {
1125 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1126 	dev_t dev = s->s_dev;
1127 
1128 	dout("kill_sb %p\n", s);
1129 
1130 	ceph_mdsc_pre_umount(fsc->mdsc);
1131 	flush_fs_workqueues(fsc);
1132 
1133 	generic_shutdown_super(s);
1134 
1135 	fsc->client->extra_mon_dispatch = NULL;
1136 	ceph_fs_debugfs_cleanup(fsc);
1137 
1138 	ceph_fscache_unregister_fs(fsc);
1139 
1140 	ceph_mdsc_destroy(fsc);
1141 
1142 	destroy_fs_client(fsc);
1143 	free_anon_bdev(dev);
1144 }
1145 
1146 static struct file_system_type ceph_fs_type = {
1147 	.owner		= THIS_MODULE,
1148 	.name		= "ceph",
1149 	.mount		= ceph_mount,
1150 	.kill_sb	= ceph_kill_sb,
1151 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1152 };
1153 MODULE_ALIAS_FS("ceph");
1154 
1155 static int __init init_ceph(void)
1156 {
1157 	int ret = init_caches();
1158 	if (ret)
1159 		goto out;
1160 
1161 	ceph_flock_init();
1162 	ceph_xattr_init();
1163 	ret = register_filesystem(&ceph_fs_type);
1164 	if (ret)
1165 		goto out_xattr;
1166 
1167 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1168 
1169 	return 0;
1170 
1171 out_xattr:
1172 	ceph_xattr_exit();
1173 	destroy_caches();
1174 out:
1175 	return ret;
1176 }
1177 
1178 static void __exit exit_ceph(void)
1179 {
1180 	dout("exit_ceph\n");
1181 	unregister_filesystem(&ceph_fs_type);
1182 	ceph_xattr_exit();
1183 	destroy_caches();
1184 }
1185 
1186 module_init(init_ceph);
1187 module_exit(exit_ceph);
1188 
1189 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1190 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1191 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1192 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1193 MODULE_LICENSE("GPL");
1194