xref: /openbmc/linux/fs/ceph/super.c (revision 457c8996)
1 
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/backing-dev.h>
5 #include <linux/ctype.h>
6 #include <linux/fs.h>
7 #include <linux/inet.h>
8 #include <linux/in6.h>
9 #include <linux/module.h>
10 #include <linux/mount.h>
11 #include <linux/parser.h>
12 #include <linux/sched.h>
13 #include <linux/seq_file.h>
14 #include <linux/slab.h>
15 #include <linux/statfs.h>
16 #include <linux/string.h>
17 
18 #include "super.h"
19 #include "mds_client.h"
20 #include "cache.h"
21 
22 #include <linux/ceph/ceph_features.h>
23 #include <linux/ceph/decode.h>
24 #include <linux/ceph/mon_client.h>
25 #include <linux/ceph/auth.h>
26 #include <linux/ceph/debugfs.h>
27 
28 /*
29  * Ceph superblock operations
30  *
31  * Handle the basics of mounting, unmounting.
32  */
33 
34 /*
35  * super ops
36  */
37 static void ceph_put_super(struct super_block *s)
38 {
39 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
40 
41 	dout("put_super\n");
42 	ceph_mdsc_close_sessions(fsc->mdsc);
43 }
44 
45 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
46 {
47 	struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
48 	struct ceph_mon_client *monc = &fsc->client->monc;
49 	struct ceph_statfs st;
50 	u64 fsid;
51 	int err;
52 	u64 data_pool;
53 
54 	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
55 		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
56 	} else {
57 		data_pool = CEPH_NOPOOL;
58 	}
59 
60 	dout("statfs\n");
61 	err = ceph_monc_do_statfs(monc, data_pool, &st);
62 	if (err < 0)
63 		return err;
64 
65 	/* fill in kstatfs */
66 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
67 
68 	/*
69 	 * express utilization in terms of large blocks to avoid
70 	 * overflow on 32-bit machines.
71 	 *
72 	 * NOTE: for the time being, we make bsize == frsize to humor
73 	 * not-yet-ancient versions of glibc that are broken.
74 	 * Someday, we will probably want to report a real block
75 	 * size...  whatever that may mean for a network file system!
76 	 */
77 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
78 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
79 
80 	/*
81 	 * By default use root quota for stats; fallback to overall filesystem
82 	 * usage if using 'noquotadf' mount option or if the root dir doesn't
83 	 * have max_bytes quota set.
84 	 */
85 	if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
86 	    !ceph_quota_update_statfs(fsc, buf)) {
87 		buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
88 		buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
89 		buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
90 	}
91 
92 	buf->f_files = le64_to_cpu(st.num_objects);
93 	buf->f_ffree = -1;
94 	buf->f_namelen = NAME_MAX;
95 
96 	/* Must convert the fsid, for consistent values across arches */
97 	mutex_lock(&monc->mutex);
98 	fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
99 	       le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
100 	mutex_unlock(&monc->mutex);
101 
102 	buf->f_fsid.val[0] = fsid & 0xffffffff;
103 	buf->f_fsid.val[1] = fsid >> 32;
104 
105 	return 0;
106 }
107 
108 
109 static int ceph_sync_fs(struct super_block *sb, int wait)
110 {
111 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
112 
113 	if (!wait) {
114 		dout("sync_fs (non-blocking)\n");
115 		ceph_flush_dirty_caps(fsc->mdsc);
116 		dout("sync_fs (non-blocking) done\n");
117 		return 0;
118 	}
119 
120 	dout("sync_fs (blocking)\n");
121 	ceph_osdc_sync(&fsc->client->osdc);
122 	ceph_mdsc_sync(fsc->mdsc);
123 	dout("sync_fs (blocking) done\n");
124 	return 0;
125 }
126 
127 /*
128  * mount options
129  */
130 enum {
131 	Opt_wsize,
132 	Opt_rsize,
133 	Opt_rasize,
134 	Opt_caps_wanted_delay_min,
135 	Opt_caps_wanted_delay_max,
136 	Opt_caps_max,
137 	Opt_readdir_max_entries,
138 	Opt_readdir_max_bytes,
139 	Opt_congestion_kb,
140 	Opt_last_int,
141 	/* int args above */
142 	Opt_snapdirname,
143 	Opt_mds_namespace,
144 	Opt_fscache_uniq,
145 	Opt_last_string,
146 	/* string args above */
147 	Opt_dirstat,
148 	Opt_nodirstat,
149 	Opt_rbytes,
150 	Opt_norbytes,
151 	Opt_asyncreaddir,
152 	Opt_noasyncreaddir,
153 	Opt_dcache,
154 	Opt_nodcache,
155 	Opt_ino32,
156 	Opt_noino32,
157 	Opt_fscache,
158 	Opt_nofscache,
159 	Opt_poolperm,
160 	Opt_nopoolperm,
161 	Opt_require_active_mds,
162 	Opt_norequire_active_mds,
163 #ifdef CONFIG_CEPH_FS_POSIX_ACL
164 	Opt_acl,
165 #endif
166 	Opt_noacl,
167 	Opt_quotadf,
168 	Opt_noquotadf,
169 	Opt_copyfrom,
170 	Opt_nocopyfrom,
171 };
172 
173 static match_table_t fsopt_tokens = {
174 	{Opt_wsize, "wsize=%d"},
175 	{Opt_rsize, "rsize=%d"},
176 	{Opt_rasize, "rasize=%d"},
177 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
178 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
179 	{Opt_caps_max, "caps_max=%d"},
180 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
181 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
182 	{Opt_congestion_kb, "write_congestion_kb=%d"},
183 	/* int args above */
184 	{Opt_snapdirname, "snapdirname=%s"},
185 	{Opt_mds_namespace, "mds_namespace=%s"},
186 	{Opt_fscache_uniq, "fsc=%s"},
187 	/* string args above */
188 	{Opt_dirstat, "dirstat"},
189 	{Opt_nodirstat, "nodirstat"},
190 	{Opt_rbytes, "rbytes"},
191 	{Opt_norbytes, "norbytes"},
192 	{Opt_asyncreaddir, "asyncreaddir"},
193 	{Opt_noasyncreaddir, "noasyncreaddir"},
194 	{Opt_dcache, "dcache"},
195 	{Opt_nodcache, "nodcache"},
196 	{Opt_ino32, "ino32"},
197 	{Opt_noino32, "noino32"},
198 	{Opt_fscache, "fsc"},
199 	{Opt_nofscache, "nofsc"},
200 	{Opt_poolperm, "poolperm"},
201 	{Opt_nopoolperm, "nopoolperm"},
202 	{Opt_require_active_mds, "require_active_mds"},
203 	{Opt_norequire_active_mds, "norequire_active_mds"},
204 #ifdef CONFIG_CEPH_FS_POSIX_ACL
205 	{Opt_acl, "acl"},
206 #endif
207 	{Opt_noacl, "noacl"},
208 	{Opt_quotadf, "quotadf"},
209 	{Opt_noquotadf, "noquotadf"},
210 	{Opt_copyfrom, "copyfrom"},
211 	{Opt_nocopyfrom, "nocopyfrom"},
212 	{-1, NULL}
213 };
214 
215 static int parse_fsopt_token(char *c, void *private)
216 {
217 	struct ceph_mount_options *fsopt = private;
218 	substring_t argstr[MAX_OPT_ARGS];
219 	int token, intval, ret;
220 
221 	token = match_token((char *)c, fsopt_tokens, argstr);
222 	if (token < 0)
223 		return -EINVAL;
224 
225 	if (token < Opt_last_int) {
226 		ret = match_int(&argstr[0], &intval);
227 		if (ret < 0) {
228 			pr_err("bad option arg (not int) at '%s'\n", c);
229 			return ret;
230 		}
231 		dout("got int token %d val %d\n", token, intval);
232 	} else if (token > Opt_last_int && token < Opt_last_string) {
233 		dout("got string token %d val %s\n", token,
234 		     argstr[0].from);
235 	} else {
236 		dout("got token %d\n", token);
237 	}
238 
239 	switch (token) {
240 	case Opt_snapdirname:
241 		kfree(fsopt->snapdir_name);
242 		fsopt->snapdir_name = kstrndup(argstr[0].from,
243 					       argstr[0].to-argstr[0].from,
244 					       GFP_KERNEL);
245 		if (!fsopt->snapdir_name)
246 			return -ENOMEM;
247 		break;
248 	case Opt_mds_namespace:
249 		kfree(fsopt->mds_namespace);
250 		fsopt->mds_namespace = kstrndup(argstr[0].from,
251 						argstr[0].to-argstr[0].from,
252 						GFP_KERNEL);
253 		if (!fsopt->mds_namespace)
254 			return -ENOMEM;
255 		break;
256 	case Opt_fscache_uniq:
257 		kfree(fsopt->fscache_uniq);
258 		fsopt->fscache_uniq = kstrndup(argstr[0].from,
259 					       argstr[0].to-argstr[0].from,
260 					       GFP_KERNEL);
261 		if (!fsopt->fscache_uniq)
262 			return -ENOMEM;
263 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
264 		break;
265 		/* misc */
266 	case Opt_wsize:
267 		if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
268 			return -EINVAL;
269 		fsopt->wsize = ALIGN(intval, PAGE_SIZE);
270 		break;
271 	case Opt_rsize:
272 		if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
273 			return -EINVAL;
274 		fsopt->rsize = ALIGN(intval, PAGE_SIZE);
275 		break;
276 	case Opt_rasize:
277 		if (intval < 0)
278 			return -EINVAL;
279 		fsopt->rasize = ALIGN(intval, PAGE_SIZE);
280 		break;
281 	case Opt_caps_wanted_delay_min:
282 		if (intval < 1)
283 			return -EINVAL;
284 		fsopt->caps_wanted_delay_min = intval;
285 		break;
286 	case Opt_caps_wanted_delay_max:
287 		if (intval < 1)
288 			return -EINVAL;
289 		fsopt->caps_wanted_delay_max = intval;
290 		break;
291 	case Opt_caps_max:
292 		if (intval < 0)
293 			return -EINVAL;
294 		fsopt->caps_max = intval;
295 		break;
296 	case Opt_readdir_max_entries:
297 		if (intval < 1)
298 			return -EINVAL;
299 		fsopt->max_readdir = intval;
300 		break;
301 	case Opt_readdir_max_bytes:
302 		if (intval < (int)PAGE_SIZE && intval != 0)
303 			return -EINVAL;
304 		fsopt->max_readdir_bytes = intval;
305 		break;
306 	case Opt_congestion_kb:
307 		if (intval < 1024) /* at least 1M */
308 			return -EINVAL;
309 		fsopt->congestion_kb = intval;
310 		break;
311 	case Opt_dirstat:
312 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
313 		break;
314 	case Opt_nodirstat:
315 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
316 		break;
317 	case Opt_rbytes:
318 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
319 		break;
320 	case Opt_norbytes:
321 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
322 		break;
323 	case Opt_asyncreaddir:
324 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
325 		break;
326 	case Opt_noasyncreaddir:
327 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
328 		break;
329 	case Opt_dcache:
330 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
331 		break;
332 	case Opt_nodcache:
333 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
334 		break;
335 	case Opt_ino32:
336 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
337 		break;
338 	case Opt_noino32:
339 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
340 		break;
341 	case Opt_fscache:
342 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
343 		kfree(fsopt->fscache_uniq);
344 		fsopt->fscache_uniq = NULL;
345 		break;
346 	case Opt_nofscache:
347 		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
348 		kfree(fsopt->fscache_uniq);
349 		fsopt->fscache_uniq = NULL;
350 		break;
351 	case Opt_poolperm:
352 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
353 		break;
354 	case Opt_nopoolperm:
355 		fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
356 		break;
357 	case Opt_require_active_mds:
358 		fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
359 		break;
360 	case Opt_norequire_active_mds:
361 		fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
362 		break;
363 	case Opt_quotadf:
364 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
365 		break;
366 	case Opt_noquotadf:
367 		fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
368 		break;
369 	case Opt_copyfrom:
370 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
371 		break;
372 	case Opt_nocopyfrom:
373 		fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
374 		break;
375 #ifdef CONFIG_CEPH_FS_POSIX_ACL
376 	case Opt_acl:
377 		fsopt->sb_flags |= SB_POSIXACL;
378 		break;
379 #endif
380 	case Opt_noacl:
381 		fsopt->sb_flags &= ~SB_POSIXACL;
382 		break;
383 	default:
384 		BUG_ON(token);
385 	}
386 	return 0;
387 }
388 
389 static void destroy_mount_options(struct ceph_mount_options *args)
390 {
391 	dout("destroy_mount_options %p\n", args);
392 	kfree(args->snapdir_name);
393 	kfree(args->mds_namespace);
394 	kfree(args->server_path);
395 	kfree(args->fscache_uniq);
396 	kfree(args);
397 }
398 
399 static int strcmp_null(const char *s1, const char *s2)
400 {
401 	if (!s1 && !s2)
402 		return 0;
403 	if (s1 && !s2)
404 		return -1;
405 	if (!s1 && s2)
406 		return 1;
407 	return strcmp(s1, s2);
408 }
409 
410 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
411 				 struct ceph_options *new_opt,
412 				 struct ceph_fs_client *fsc)
413 {
414 	struct ceph_mount_options *fsopt1 = new_fsopt;
415 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
416 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
417 	int ret;
418 
419 	ret = memcmp(fsopt1, fsopt2, ofs);
420 	if (ret)
421 		return ret;
422 
423 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
424 	if (ret)
425 		return ret;
426 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
427 	if (ret)
428 		return ret;
429 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
430 	if (ret)
431 		return ret;
432 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
433 	if (ret)
434 		return ret;
435 
436 	return ceph_compare_options(new_opt, fsc->client);
437 }
438 
439 static int parse_mount_options(struct ceph_mount_options **pfsopt,
440 			       struct ceph_options **popt,
441 			       int flags, char *options,
442 			       const char *dev_name)
443 {
444 	struct ceph_mount_options *fsopt;
445 	const char *dev_name_end;
446 	int err;
447 
448 	if (!dev_name || !*dev_name)
449 		return -EINVAL;
450 
451 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
452 	if (!fsopt)
453 		return -ENOMEM;
454 
455 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
456 
457 	fsopt->sb_flags = flags;
458 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
459 
460 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
461 	fsopt->rsize = CEPH_MAX_READ_SIZE;
462 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
463 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
464 	if (!fsopt->snapdir_name) {
465 		err = -ENOMEM;
466 		goto out;
467 	}
468 
469 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
470 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
471 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
472 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
473 	fsopt->congestion_kb = default_congestion_kb();
474 
475 	/*
476 	 * Distinguish the server list from the path in "dev_name".
477 	 * Internally we do not include the leading '/' in the path.
478 	 *
479 	 * "dev_name" will look like:
480 	 *     <server_spec>[,<server_spec>...]:[<path>]
481 	 * where
482 	 *     <server_spec> is <ip>[:<port>]
483 	 *     <path> is optional, but if present must begin with '/'
484 	 */
485 	dev_name_end = strchr(dev_name, '/');
486 	if (dev_name_end) {
487 		if (strlen(dev_name_end) > 1) {
488 			fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
489 			if (!fsopt->server_path) {
490 				err = -ENOMEM;
491 				goto out;
492 			}
493 		}
494 	} else {
495 		dev_name_end = dev_name + strlen(dev_name);
496 	}
497 	err = -EINVAL;
498 	dev_name_end--;		/* back up to ':' separator */
499 	if (dev_name_end < dev_name || *dev_name_end != ':') {
500 		pr_err("device name is missing path (no : separator in %s)\n",
501 				dev_name);
502 		goto out;
503 	}
504 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
505 	if (fsopt->server_path)
506 		dout("server path '%s'\n", fsopt->server_path);
507 
508 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
509 				 parse_fsopt_token, (void *)fsopt);
510 	if (IS_ERR(*popt)) {
511 		err = PTR_ERR(*popt);
512 		goto out;
513 	}
514 
515 	/* success */
516 	*pfsopt = fsopt;
517 	return 0;
518 
519 out:
520 	destroy_mount_options(fsopt);
521 	return err;
522 }
523 
524 /**
525  * ceph_show_options - Show mount options in /proc/mounts
526  * @m: seq_file to write to
527  * @root: root of that (sub)tree
528  */
529 static int ceph_show_options(struct seq_file *m, struct dentry *root)
530 {
531 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
532 	struct ceph_mount_options *fsopt = fsc->mount_options;
533 	size_t pos;
534 	int ret;
535 
536 	/* a comma between MNT/MS and client options */
537 	seq_putc(m, ',');
538 	pos = m->count;
539 
540 	ret = ceph_print_client_options(m, fsc->client, false);
541 	if (ret)
542 		return ret;
543 
544 	/* retract our comma if no client options */
545 	if (m->count == pos)
546 		m->count--;
547 
548 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
549 		seq_puts(m, ",dirstat");
550 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
551 		seq_puts(m, ",rbytes");
552 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
553 		seq_puts(m, ",noasyncreaddir");
554 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
555 		seq_puts(m, ",nodcache");
556 	if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
557 		seq_puts(m, ",ino32");
558 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
559 		seq_show_option(m, "fsc", fsopt->fscache_uniq);
560 	}
561 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
562 		seq_puts(m, ",nopoolperm");
563 	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
564 		seq_puts(m, ",noquotadf");
565 
566 #ifdef CONFIG_CEPH_FS_POSIX_ACL
567 	if (fsopt->sb_flags & SB_POSIXACL)
568 		seq_puts(m, ",acl");
569 	else
570 		seq_puts(m, ",noacl");
571 #endif
572 
573 	if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
574 		seq_puts(m, ",copyfrom");
575 
576 	if (fsopt->mds_namespace)
577 		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
578 	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
579 		seq_printf(m, ",wsize=%d", fsopt->wsize);
580 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
581 		seq_printf(m, ",rsize=%d", fsopt->rsize);
582 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
583 		seq_printf(m, ",rasize=%d", fsopt->rasize);
584 	if (fsopt->congestion_kb != default_congestion_kb())
585 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
586 	if (fsopt->caps_max)
587 		seq_printf(m, ",caps_max=%d", fsopt->caps_max);
588 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
589 		seq_printf(m, ",caps_wanted_delay_min=%d",
590 			 fsopt->caps_wanted_delay_min);
591 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
592 		seq_printf(m, ",caps_wanted_delay_max=%d",
593 			   fsopt->caps_wanted_delay_max);
594 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
595 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
596 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
597 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
598 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
599 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
600 
601 	return 0;
602 }
603 
604 /*
605  * handle any mon messages the standard library doesn't understand.
606  * return error if we don't either.
607  */
608 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
609 {
610 	struct ceph_fs_client *fsc = client->private;
611 	int type = le16_to_cpu(msg->hdr.type);
612 
613 	switch (type) {
614 	case CEPH_MSG_MDS_MAP:
615 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
616 		return 0;
617 	case CEPH_MSG_FS_MAP_USER:
618 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
619 		return 0;
620 	default:
621 		return -1;
622 	}
623 }
624 
625 /*
626  * create a new fs client
627  *
628  * Success or not, this function consumes @fsopt and @opt.
629  */
630 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
631 					struct ceph_options *opt)
632 {
633 	struct ceph_fs_client *fsc;
634 	int page_count;
635 	size_t size;
636 	int err;
637 
638 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
639 	if (!fsc) {
640 		err = -ENOMEM;
641 		goto fail;
642 	}
643 
644 	fsc->client = ceph_create_client(opt, fsc);
645 	if (IS_ERR(fsc->client)) {
646 		err = PTR_ERR(fsc->client);
647 		goto fail;
648 	}
649 	opt = NULL; /* fsc->client now owns this */
650 
651 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
652 	ceph_set_opt(fsc->client, ABORT_ON_FULL);
653 
654 	if (!fsopt->mds_namespace) {
655 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
656 				   0, true);
657 	} else {
658 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
659 				   0, false);
660 	}
661 
662 	fsc->mount_options = fsopt;
663 
664 	fsc->sb = NULL;
665 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
666 
667 	atomic_long_set(&fsc->writeback_count, 0);
668 
669 	err = -ENOMEM;
670 	/*
671 	 * The number of concurrent works can be high but they don't need
672 	 * to be processed in parallel, limit concurrency.
673 	 */
674 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
675 	if (!fsc->wb_wq)
676 		goto fail_client;
677 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
678 	if (!fsc->pg_inv_wq)
679 		goto fail_wb_wq;
680 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
681 	if (!fsc->trunc_wq)
682 		goto fail_pg_inv_wq;
683 	fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
684 	if (!fsc->cap_wq)
685 		goto fail_trunc_wq;
686 
687 	/* set up mempools */
688 	err = -ENOMEM;
689 	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
690 	size = sizeof (struct page *) * (page_count ? page_count : 1);
691 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
692 	if (!fsc->wb_pagevec_pool)
693 		goto fail_cap_wq;
694 
695 	return fsc;
696 
697 fail_cap_wq:
698 	destroy_workqueue(fsc->cap_wq);
699 fail_trunc_wq:
700 	destroy_workqueue(fsc->trunc_wq);
701 fail_pg_inv_wq:
702 	destroy_workqueue(fsc->pg_inv_wq);
703 fail_wb_wq:
704 	destroy_workqueue(fsc->wb_wq);
705 fail_client:
706 	ceph_destroy_client(fsc->client);
707 fail:
708 	kfree(fsc);
709 	if (opt)
710 		ceph_destroy_options(opt);
711 	destroy_mount_options(fsopt);
712 	return ERR_PTR(err);
713 }
714 
715 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
716 {
717 	flush_workqueue(fsc->wb_wq);
718 	flush_workqueue(fsc->pg_inv_wq);
719 	flush_workqueue(fsc->trunc_wq);
720 	flush_workqueue(fsc->cap_wq);
721 }
722 
723 static void destroy_fs_client(struct ceph_fs_client *fsc)
724 {
725 	dout("destroy_fs_client %p\n", fsc);
726 
727 	destroy_workqueue(fsc->wb_wq);
728 	destroy_workqueue(fsc->pg_inv_wq);
729 	destroy_workqueue(fsc->trunc_wq);
730 	destroy_workqueue(fsc->cap_wq);
731 
732 	mempool_destroy(fsc->wb_pagevec_pool);
733 
734 	destroy_mount_options(fsc->mount_options);
735 
736 	ceph_destroy_client(fsc->client);
737 
738 	kfree(fsc);
739 	dout("destroy_fs_client %p done\n", fsc);
740 }
741 
742 /*
743  * caches
744  */
745 struct kmem_cache *ceph_inode_cachep;
746 struct kmem_cache *ceph_cap_cachep;
747 struct kmem_cache *ceph_cap_flush_cachep;
748 struct kmem_cache *ceph_dentry_cachep;
749 struct kmem_cache *ceph_file_cachep;
750 struct kmem_cache *ceph_dir_file_cachep;
751 
752 static void ceph_inode_init_once(void *foo)
753 {
754 	struct ceph_inode_info *ci = foo;
755 	inode_init_once(&ci->vfs_inode);
756 }
757 
758 static int __init init_caches(void)
759 {
760 	int error = -ENOMEM;
761 
762 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
763 				      sizeof(struct ceph_inode_info),
764 				      __alignof__(struct ceph_inode_info),
765 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
766 				      SLAB_ACCOUNT, ceph_inode_init_once);
767 	if (!ceph_inode_cachep)
768 		return -ENOMEM;
769 
770 	ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
771 	if (!ceph_cap_cachep)
772 		goto bad_cap;
773 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
774 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
775 	if (!ceph_cap_flush_cachep)
776 		goto bad_cap_flush;
777 
778 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
779 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
780 	if (!ceph_dentry_cachep)
781 		goto bad_dentry;
782 
783 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
784 	if (!ceph_file_cachep)
785 		goto bad_file;
786 
787 	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
788 	if (!ceph_dir_file_cachep)
789 		goto bad_dir_file;
790 
791 	error = ceph_fscache_register();
792 	if (error)
793 		goto bad_fscache;
794 
795 	return 0;
796 
797 bad_fscache:
798 	kmem_cache_destroy(ceph_dir_file_cachep);
799 bad_dir_file:
800 	kmem_cache_destroy(ceph_file_cachep);
801 bad_file:
802 	kmem_cache_destroy(ceph_dentry_cachep);
803 bad_dentry:
804 	kmem_cache_destroy(ceph_cap_flush_cachep);
805 bad_cap_flush:
806 	kmem_cache_destroy(ceph_cap_cachep);
807 bad_cap:
808 	kmem_cache_destroy(ceph_inode_cachep);
809 	return error;
810 }
811 
812 static void destroy_caches(void)
813 {
814 	/*
815 	 * Make sure all delayed rcu free inodes are flushed before we
816 	 * destroy cache.
817 	 */
818 	rcu_barrier();
819 
820 	kmem_cache_destroy(ceph_inode_cachep);
821 	kmem_cache_destroy(ceph_cap_cachep);
822 	kmem_cache_destroy(ceph_cap_flush_cachep);
823 	kmem_cache_destroy(ceph_dentry_cachep);
824 	kmem_cache_destroy(ceph_file_cachep);
825 	kmem_cache_destroy(ceph_dir_file_cachep);
826 
827 	ceph_fscache_unregister();
828 }
829 
830 
831 /*
832  * ceph_umount_begin - initiate forced umount.  Tear down down the
833  * mount, skipping steps that may hang while waiting for server(s).
834  */
835 static void ceph_umount_begin(struct super_block *sb)
836 {
837 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
838 
839 	dout("ceph_umount_begin - starting forced umount\n");
840 	if (!fsc)
841 		return;
842 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
843 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
844 	ceph_mdsc_force_umount(fsc->mdsc);
845 	return;
846 }
847 
848 static int ceph_remount(struct super_block *sb, int *flags, char *data)
849 {
850 	sync_filesystem(sb);
851 	return 0;
852 }
853 
854 static const struct super_operations ceph_super_ops = {
855 	.alloc_inode	= ceph_alloc_inode,
856 	.destroy_inode	= ceph_destroy_inode,
857 	.free_inode	= ceph_free_inode,
858 	.write_inode    = ceph_write_inode,
859 	.drop_inode	= ceph_drop_inode,
860 	.sync_fs        = ceph_sync_fs,
861 	.put_super	= ceph_put_super,
862 	.remount_fs	= ceph_remount,
863 	.show_options   = ceph_show_options,
864 	.statfs		= ceph_statfs,
865 	.umount_begin   = ceph_umount_begin,
866 };
867 
868 /*
869  * Bootstrap mount by opening the root directory.  Note the mount
870  * @started time from caller, and time out if this takes too long.
871  */
872 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
873 				       const char *path,
874 				       unsigned long started)
875 {
876 	struct ceph_mds_client *mdsc = fsc->mdsc;
877 	struct ceph_mds_request *req = NULL;
878 	int err;
879 	struct dentry *root;
880 
881 	/* open dir */
882 	dout("open_root_inode opening '%s'\n", path);
883 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
884 	if (IS_ERR(req))
885 		return ERR_CAST(req);
886 	req->r_path1 = kstrdup(path, GFP_NOFS);
887 	if (!req->r_path1) {
888 		root = ERR_PTR(-ENOMEM);
889 		goto out;
890 	}
891 
892 	req->r_ino1.ino = CEPH_INO_ROOT;
893 	req->r_ino1.snap = CEPH_NOSNAP;
894 	req->r_started = started;
895 	req->r_timeout = fsc->client->options->mount_timeout;
896 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
897 	req->r_num_caps = 2;
898 	err = ceph_mdsc_do_request(mdsc, NULL, req);
899 	if (err == 0) {
900 		struct inode *inode = req->r_target_inode;
901 		req->r_target_inode = NULL;
902 		dout("open_root_inode success\n");
903 		root = d_make_root(inode);
904 		if (!root) {
905 			root = ERR_PTR(-ENOMEM);
906 			goto out;
907 		}
908 		dout("open_root_inode success, root dentry is %p\n", root);
909 	} else {
910 		root = ERR_PTR(err);
911 	}
912 out:
913 	ceph_mdsc_put_request(req);
914 	return root;
915 }
916 
917 
918 
919 
920 /*
921  * mount: join the ceph cluster, and open root directory.
922  */
923 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
924 {
925 	int err;
926 	unsigned long started = jiffies;  /* note the start time */
927 	struct dentry *root;
928 
929 	dout("mount start %p\n", fsc);
930 	mutex_lock(&fsc->client->mount_mutex);
931 
932 	if (!fsc->sb->s_root) {
933 		const char *path;
934 		err = __ceph_open_session(fsc->client, started);
935 		if (err < 0)
936 			goto out;
937 
938 		/* setup fscache */
939 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
940 			err = ceph_fscache_register_fs(fsc);
941 			if (err < 0)
942 				goto out;
943 		}
944 
945 		if (!fsc->mount_options->server_path) {
946 			path = "";
947 			dout("mount opening path \\t\n");
948 		} else {
949 			path = fsc->mount_options->server_path + 1;
950 			dout("mount opening path %s\n", path);
951 		}
952 
953 		err = ceph_fs_debugfs_init(fsc);
954 		if (err < 0)
955 			goto out;
956 
957 		root = open_root_dentry(fsc, path, started);
958 		if (IS_ERR(root)) {
959 			err = PTR_ERR(root);
960 			goto out;
961 		}
962 		fsc->sb->s_root = dget(root);
963 	} else {
964 		root = dget(fsc->sb->s_root);
965 	}
966 
967 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
968 	dout("mount success\n");
969 	mutex_unlock(&fsc->client->mount_mutex);
970 	return root;
971 
972 out:
973 	mutex_unlock(&fsc->client->mount_mutex);
974 	return ERR_PTR(err);
975 }
976 
977 static int ceph_set_super(struct super_block *s, void *data)
978 {
979 	struct ceph_fs_client *fsc = data;
980 	int ret;
981 
982 	dout("set_super %p data %p\n", s, data);
983 
984 	s->s_flags = fsc->mount_options->sb_flags;
985 	s->s_maxbytes = MAX_LFS_FILESIZE;
986 
987 	s->s_xattr = ceph_xattr_handlers;
988 	s->s_fs_info = fsc;
989 	fsc->sb = s;
990 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
991 
992 	s->s_op = &ceph_super_ops;
993 	s->s_d_op = &ceph_dentry_ops;
994 	s->s_export_op = &ceph_export_ops;
995 
996 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
997 
998 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
999 	if (ret != 0)
1000 		goto fail;
1001 
1002 	return ret;
1003 
1004 fail:
1005 	s->s_fs_info = NULL;
1006 	fsc->sb = NULL;
1007 	return ret;
1008 }
1009 
1010 /*
1011  * share superblock if same fs AND options
1012  */
1013 static int ceph_compare_super(struct super_block *sb, void *data)
1014 {
1015 	struct ceph_fs_client *new = data;
1016 	struct ceph_mount_options *fsopt = new->mount_options;
1017 	struct ceph_options *opt = new->client->options;
1018 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
1019 
1020 	dout("ceph_compare_super %p\n", sb);
1021 
1022 	if (compare_mount_options(fsopt, opt, other)) {
1023 		dout("monitor(s)/mount options don't match\n");
1024 		return 0;
1025 	}
1026 	if ((opt->flags & CEPH_OPT_FSID) &&
1027 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
1028 		dout("fsid doesn't match\n");
1029 		return 0;
1030 	}
1031 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
1032 		dout("flags differ\n");
1033 		return 0;
1034 	}
1035 	return 1;
1036 }
1037 
1038 /*
1039  * construct our own bdi so we can control readahead, etc.
1040  */
1041 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1042 
1043 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1044 {
1045 	int err;
1046 
1047 	err = super_setup_bdi_name(sb, "ceph-%ld",
1048 				   atomic_long_inc_return(&bdi_seq));
1049 	if (err)
1050 		return err;
1051 
1052 	/* set ra_pages based on rasize mount option? */
1053 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1054 
1055 	/* set io_pages based on max osd read size */
1056 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1057 
1058 	return 0;
1059 }
1060 
1061 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1062 		       int flags, const char *dev_name, void *data)
1063 {
1064 	struct super_block *sb;
1065 	struct ceph_fs_client *fsc;
1066 	struct dentry *res;
1067 	int err;
1068 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1069 	struct ceph_mount_options *fsopt = NULL;
1070 	struct ceph_options *opt = NULL;
1071 
1072 	dout("ceph_mount\n");
1073 
1074 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1075 	flags |= SB_POSIXACL;
1076 #endif
1077 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1078 	if (err < 0) {
1079 		res = ERR_PTR(err);
1080 		goto out_final;
1081 	}
1082 
1083 	/* create client (which we may/may not use) */
1084 	fsc = create_fs_client(fsopt, opt);
1085 	if (IS_ERR(fsc)) {
1086 		res = ERR_CAST(fsc);
1087 		goto out_final;
1088 	}
1089 
1090 	err = ceph_mdsc_init(fsc);
1091 	if (err < 0) {
1092 		res = ERR_PTR(err);
1093 		goto out;
1094 	}
1095 
1096 	if (ceph_test_opt(fsc->client, NOSHARE))
1097 		compare_super = NULL;
1098 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1099 	if (IS_ERR(sb)) {
1100 		res = ERR_CAST(sb);
1101 		goto out;
1102 	}
1103 
1104 	if (ceph_sb_to_client(sb) != fsc) {
1105 		ceph_mdsc_destroy(fsc);
1106 		destroy_fs_client(fsc);
1107 		fsc = ceph_sb_to_client(sb);
1108 		dout("get_sb got existing client %p\n", fsc);
1109 	} else {
1110 		dout("get_sb using new client %p\n", fsc);
1111 		err = ceph_setup_bdi(sb, fsc);
1112 		if (err < 0) {
1113 			res = ERR_PTR(err);
1114 			goto out_splat;
1115 		}
1116 	}
1117 
1118 	res = ceph_real_mount(fsc);
1119 	if (IS_ERR(res))
1120 		goto out_splat;
1121 	dout("root %p inode %p ino %llx.%llx\n", res,
1122 	     d_inode(res), ceph_vinop(d_inode(res)));
1123 	return res;
1124 
1125 out_splat:
1126 	ceph_mdsc_close_sessions(fsc->mdsc);
1127 	deactivate_locked_super(sb);
1128 	goto out_final;
1129 
1130 out:
1131 	ceph_mdsc_destroy(fsc);
1132 	destroy_fs_client(fsc);
1133 out_final:
1134 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1135 	return res;
1136 }
1137 
1138 static void ceph_kill_sb(struct super_block *s)
1139 {
1140 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1141 	dev_t dev = s->s_dev;
1142 
1143 	dout("kill_sb %p\n", s);
1144 
1145 	ceph_mdsc_pre_umount(fsc->mdsc);
1146 	flush_fs_workqueues(fsc);
1147 
1148 	generic_shutdown_super(s);
1149 
1150 	fsc->client->extra_mon_dispatch = NULL;
1151 	ceph_fs_debugfs_cleanup(fsc);
1152 
1153 	ceph_fscache_unregister_fs(fsc);
1154 
1155 	ceph_mdsc_destroy(fsc);
1156 
1157 	destroy_fs_client(fsc);
1158 	free_anon_bdev(dev);
1159 }
1160 
1161 static struct file_system_type ceph_fs_type = {
1162 	.owner		= THIS_MODULE,
1163 	.name		= "ceph",
1164 	.mount		= ceph_mount,
1165 	.kill_sb	= ceph_kill_sb,
1166 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1167 };
1168 MODULE_ALIAS_FS("ceph");
1169 
1170 static int __init init_ceph(void)
1171 {
1172 	int ret = init_caches();
1173 	if (ret)
1174 		goto out;
1175 
1176 	ceph_flock_init();
1177 	ceph_xattr_init();
1178 	ret = register_filesystem(&ceph_fs_type);
1179 	if (ret)
1180 		goto out_xattr;
1181 
1182 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1183 
1184 	return 0;
1185 
1186 out_xattr:
1187 	ceph_xattr_exit();
1188 	destroy_caches();
1189 out:
1190 	return ret;
1191 }
1192 
1193 static void __exit exit_ceph(void)
1194 {
1195 	dout("exit_ceph\n");
1196 	unregister_filesystem(&ceph_fs_type);
1197 	ceph_xattr_exit();
1198 	destroy_caches();
1199 }
1200 
1201 module_init(init_ceph);
1202 module_exit(exit_ceph);
1203 
1204 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1205 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1206 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1207 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1208 MODULE_LICENSE("GPL");
1209