xref: /linux/fs/ceph/super.c (revision c145211d1f9e2ef19e7b4c2b943f68366daa97af)
1 
2 #include "ceph_debug.h"
3 
4 #include <linux/backing-dev.h>
5 #include <linux/fs.h>
6 #include <linux/inet.h>
7 #include <linux/in6.h>
8 #include <linux/module.h>
9 #include <linux/mount.h>
10 #include <linux/parser.h>
11 #include <linux/rwsem.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 #include <linux/version.h>
18 #include <linux/vmalloc.h>
19 
20 #include "decode.h"
21 #include "super.h"
22 #include "mon_client.h"
23 #include "auth.h"
24 
25 /*
26  * Ceph superblock operations
27  *
28  * Handle the basics of mounting, unmounting.
29  */
30 
31 
32 /*
33  * find filename portion of a path (/foo/bar/baz -> baz)
34  */
35 const char *ceph_file_part(const char *s, int len)
36 {
37 	const char *e = s + len;
38 
39 	while (e != s && *(e-1) != '/')
40 		e--;
41 	return e;
42 }
43 
44 
45 /*
46  * super ops
47  */
48 static void ceph_put_super(struct super_block *s)
49 {
50 	struct ceph_client *client = ceph_sb_to_client(s);
51 
52 	dout("put_super\n");
53 	ceph_mdsc_close_sessions(&client->mdsc);
54 
55 	/*
56 	 * ensure we release the bdi before put_anon_super releases
57 	 * the device name.
58 	 */
59 	if (s->s_bdi == &client->backing_dev_info) {
60 		bdi_unregister(&client->backing_dev_info);
61 		s->s_bdi = NULL;
62 	}
63 
64 	return;
65 }
66 
67 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
68 {
69 	struct ceph_client *client = ceph_inode_to_client(dentry->d_inode);
70 	struct ceph_monmap *monmap = client->monc.monmap;
71 	struct ceph_statfs st;
72 	u64 fsid;
73 	int err;
74 
75 	dout("statfs\n");
76 	err = ceph_monc_do_statfs(&client->monc, &st);
77 	if (err < 0)
78 		return err;
79 
80 	/* fill in kstatfs */
81 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
82 
83 	/*
84 	 * express utilization in terms of large blocks to avoid
85 	 * overflow on 32-bit machines.
86 	 */
87 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
88 	buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
89 	buf->f_bfree = (le64_to_cpu(st.kb) - le64_to_cpu(st.kb_used)) >>
90 		(CEPH_BLOCK_SHIFT-10);
91 	buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
92 
93 	buf->f_files = le64_to_cpu(st.num_objects);
94 	buf->f_ffree = -1;
95 	buf->f_namelen = PATH_MAX;
96 	buf->f_frsize = PAGE_CACHE_SIZE;
97 
98 	/* leave fsid little-endian, regardless of host endianness */
99 	fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
100 	buf->f_fsid.val[0] = fsid & 0xffffffff;
101 	buf->f_fsid.val[1] = fsid >> 32;
102 
103 	return 0;
104 }
105 
106 
107 static int ceph_syncfs(struct super_block *sb, int wait)
108 {
109 	dout("sync_fs %d\n", wait);
110 	ceph_osdc_sync(&ceph_client(sb)->osdc);
111 	ceph_mdsc_sync(&ceph_client(sb)->mdsc);
112 	dout("sync_fs %d done\n", wait);
113 	return 0;
114 }
115 
116 
117 /**
118  * ceph_show_options - Show mount options in /proc/mounts
119  * @m: seq_file to write to
120  * @mnt: mount descriptor
121  */
122 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
123 {
124 	struct ceph_client *client = ceph_sb_to_client(mnt->mnt_sb);
125 	struct ceph_mount_args *args = client->mount_args;
126 
127 	if (args->flags & CEPH_OPT_FSID)
128 		seq_printf(m, ",fsidmajor=%llu,fsidminor%llu",
129 			   le64_to_cpu(*(__le64 *)&args->fsid.fsid[0]),
130 			   le64_to_cpu(*(__le64 *)&args->fsid.fsid[8]));
131 	if (args->flags & CEPH_OPT_NOSHARE)
132 		seq_puts(m, ",noshare");
133 	if (args->flags & CEPH_OPT_DIRSTAT)
134 		seq_puts(m, ",dirstat");
135 	if ((args->flags & CEPH_OPT_RBYTES) == 0)
136 		seq_puts(m, ",norbytes");
137 	if (args->flags & CEPH_OPT_NOCRC)
138 		seq_puts(m, ",nocrc");
139 	if (args->flags & CEPH_OPT_NOASYNCREADDIR)
140 		seq_puts(m, ",noasyncreaddir");
141 	if (strcmp(args->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
142 		seq_printf(m, ",snapdirname=%s", args->snapdir_name);
143 	if (args->name)
144 		seq_printf(m, ",name=%s", args->name);
145 	if (args->secret)
146 		seq_puts(m, ",secret=<hidden>");
147 	return 0;
148 }
149 
150 /*
151  * caches
152  */
153 struct kmem_cache *ceph_inode_cachep;
154 struct kmem_cache *ceph_cap_cachep;
155 struct kmem_cache *ceph_dentry_cachep;
156 struct kmem_cache *ceph_file_cachep;
157 
158 static void ceph_inode_init_once(void *foo)
159 {
160 	struct ceph_inode_info *ci = foo;
161 	inode_init_once(&ci->vfs_inode);
162 }
163 
164 static int default_congestion_kb(void)
165 {
166 	int congestion_kb;
167 
168 	/*
169 	 * Copied from NFS
170 	 *
171 	 * congestion size, scale with available memory.
172 	 *
173 	 *  64MB:    8192k
174 	 * 128MB:   11585k
175 	 * 256MB:   16384k
176 	 * 512MB:   23170k
177 	 *   1GB:   32768k
178 	 *   2GB:   46340k
179 	 *   4GB:   65536k
180 	 *   8GB:   92681k
181 	 *  16GB:  131072k
182 	 *
183 	 * This allows larger machines to have larger/more transfers.
184 	 * Limit the default to 256M
185 	 */
186 	congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
187 	if (congestion_kb > 256*1024)
188 		congestion_kb = 256*1024;
189 
190 	return congestion_kb;
191 }
192 
193 static int __init init_caches(void)
194 {
195 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
196 				      sizeof(struct ceph_inode_info),
197 				      __alignof__(struct ceph_inode_info),
198 				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
199 				      ceph_inode_init_once);
200 	if (ceph_inode_cachep == NULL)
201 		return -ENOMEM;
202 
203 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
204 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
205 	if (ceph_cap_cachep == NULL)
206 		goto bad_cap;
207 
208 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
209 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
210 	if (ceph_dentry_cachep == NULL)
211 		goto bad_dentry;
212 
213 	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
214 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
215 	if (ceph_file_cachep == NULL)
216 		goto bad_file;
217 
218 	return 0;
219 
220 bad_file:
221 	kmem_cache_destroy(ceph_dentry_cachep);
222 bad_dentry:
223 	kmem_cache_destroy(ceph_cap_cachep);
224 bad_cap:
225 	kmem_cache_destroy(ceph_inode_cachep);
226 	return -ENOMEM;
227 }
228 
229 static void destroy_caches(void)
230 {
231 	kmem_cache_destroy(ceph_inode_cachep);
232 	kmem_cache_destroy(ceph_cap_cachep);
233 	kmem_cache_destroy(ceph_dentry_cachep);
234 	kmem_cache_destroy(ceph_file_cachep);
235 }
236 
237 
238 /*
239  * ceph_umount_begin - initiate forced umount.  Tear down down the
240  * mount, skipping steps that may hang while waiting for server(s).
241  */
242 static void ceph_umount_begin(struct super_block *sb)
243 {
244 	struct ceph_client *client = ceph_sb_to_client(sb);
245 
246 	dout("ceph_umount_begin - starting forced umount\n");
247 	if (!client)
248 		return;
249 	client->mount_state = CEPH_MOUNT_SHUTDOWN;
250 	return;
251 }
252 
253 static const struct super_operations ceph_super_ops = {
254 	.alloc_inode	= ceph_alloc_inode,
255 	.destroy_inode	= ceph_destroy_inode,
256 	.write_inode    = ceph_write_inode,
257 	.sync_fs        = ceph_syncfs,
258 	.put_super	= ceph_put_super,
259 	.show_options   = ceph_show_options,
260 	.statfs		= ceph_statfs,
261 	.umount_begin   = ceph_umount_begin,
262 };
263 
264 
265 const char *ceph_msg_type_name(int type)
266 {
267 	switch (type) {
268 	case CEPH_MSG_SHUTDOWN: return "shutdown";
269 	case CEPH_MSG_PING: return "ping";
270 	case CEPH_MSG_AUTH: return "auth";
271 	case CEPH_MSG_AUTH_REPLY: return "auth_reply";
272 	case CEPH_MSG_MON_MAP: return "mon_map";
273 	case CEPH_MSG_MON_GET_MAP: return "mon_get_map";
274 	case CEPH_MSG_MON_SUBSCRIBE: return "mon_subscribe";
275 	case CEPH_MSG_MON_SUBSCRIBE_ACK: return "mon_subscribe_ack";
276 	case CEPH_MSG_STATFS: return "statfs";
277 	case CEPH_MSG_STATFS_REPLY: return "statfs_reply";
278 	case CEPH_MSG_MDS_MAP: return "mds_map";
279 	case CEPH_MSG_CLIENT_SESSION: return "client_session";
280 	case CEPH_MSG_CLIENT_RECONNECT: return "client_reconnect";
281 	case CEPH_MSG_CLIENT_REQUEST: return "client_request";
282 	case CEPH_MSG_CLIENT_REQUEST_FORWARD: return "client_request_forward";
283 	case CEPH_MSG_CLIENT_REPLY: return "client_reply";
284 	case CEPH_MSG_CLIENT_CAPS: return "client_caps";
285 	case CEPH_MSG_CLIENT_CAPRELEASE: return "client_cap_release";
286 	case CEPH_MSG_CLIENT_SNAP: return "client_snap";
287 	case CEPH_MSG_CLIENT_LEASE: return "client_lease";
288 	case CEPH_MSG_OSD_MAP: return "osd_map";
289 	case CEPH_MSG_OSD_OP: return "osd_op";
290 	case CEPH_MSG_OSD_OPREPLY: return "osd_opreply";
291 	default: return "unknown";
292 	}
293 }
294 
295 
296 /*
297  * mount options
298  */
299 enum {
300 	Opt_fsidmajor,
301 	Opt_fsidminor,
302 	Opt_monport,
303 	Opt_wsize,
304 	Opt_rsize,
305 	Opt_osdtimeout,
306 	Opt_osdkeepalivetimeout,
307 	Opt_mount_timeout,
308 	Opt_osd_idle_ttl,
309 	Opt_caps_wanted_delay_min,
310 	Opt_caps_wanted_delay_max,
311 	Opt_readdir_max_entries,
312 	Opt_congestion_kb,
313 	Opt_last_int,
314 	/* int args above */
315 	Opt_snapdirname,
316 	Opt_name,
317 	Opt_secret,
318 	Opt_last_string,
319 	/* string args above */
320 	Opt_ip,
321 	Opt_noshare,
322 	Opt_dirstat,
323 	Opt_nodirstat,
324 	Opt_rbytes,
325 	Opt_norbytes,
326 	Opt_nocrc,
327 	Opt_noasyncreaddir,
328 };
329 
330 static match_table_t arg_tokens = {
331 	{Opt_fsidmajor, "fsidmajor=%ld"},
332 	{Opt_fsidminor, "fsidminor=%ld"},
333 	{Opt_monport, "monport=%d"},
334 	{Opt_wsize, "wsize=%d"},
335 	{Opt_rsize, "rsize=%d"},
336 	{Opt_osdtimeout, "osdtimeout=%d"},
337 	{Opt_osdkeepalivetimeout, "osdkeepalive=%d"},
338 	{Opt_mount_timeout, "mount_timeout=%d"},
339 	{Opt_osd_idle_ttl, "osd_idle_ttl=%d"},
340 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
341 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
342 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
343 	{Opt_congestion_kb, "write_congestion_kb=%d"},
344 	/* int args above */
345 	{Opt_snapdirname, "snapdirname=%s"},
346 	{Opt_name, "name=%s"},
347 	{Opt_secret, "secret=%s"},
348 	/* string args above */
349 	{Opt_ip, "ip=%s"},
350 	{Opt_noshare, "noshare"},
351 	{Opt_dirstat, "dirstat"},
352 	{Opt_nodirstat, "nodirstat"},
353 	{Opt_rbytes, "rbytes"},
354 	{Opt_norbytes, "norbytes"},
355 	{Opt_nocrc, "nocrc"},
356 	{Opt_noasyncreaddir, "noasyncreaddir"},
357 	{-1, NULL}
358 };
359 
360 
361 static struct ceph_mount_args *parse_mount_args(int flags, char *options,
362 						const char *dev_name,
363 						const char **path)
364 {
365 	struct ceph_mount_args *args;
366 	const char *c;
367 	int err = -ENOMEM;
368 	substring_t argstr[MAX_OPT_ARGS];
369 
370 	args = kzalloc(sizeof(*args), GFP_KERNEL);
371 	if (!args)
372 		return ERR_PTR(-ENOMEM);
373 	args->mon_addr = kcalloc(CEPH_MAX_MON, sizeof(*args->mon_addr),
374 				 GFP_KERNEL);
375 	if (!args->mon_addr)
376 		goto out;
377 
378 	dout("parse_mount_args %p, dev_name '%s'\n", args, dev_name);
379 
380 	/* start with defaults */
381 	args->sb_flags = flags;
382 	args->flags = CEPH_OPT_DEFAULT;
383 	args->osd_timeout = CEPH_OSD_TIMEOUT_DEFAULT;
384 	args->osd_keepalive_timeout = CEPH_OSD_KEEPALIVE_DEFAULT;
385 	args->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
386 	args->osd_idle_ttl = CEPH_OSD_IDLE_TTL_DEFAULT;   /* seconds */
387 	args->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
388 	args->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
389 	args->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
390 	args->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
391 	args->cap_release_safety = CEPH_CAPS_PER_RELEASE * 4;
392 	args->max_readdir = 1024;
393 	args->congestion_kb = default_congestion_kb();
394 
395 	/* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
396 	err = -EINVAL;
397 	if (!dev_name)
398 		goto out;
399 	*path = strstr(dev_name, ":/");
400 	if (*path == NULL) {
401 		pr_err("device name is missing path (no :/ in %s)\n",
402 		       dev_name);
403 		goto out;
404 	}
405 
406 	/* get mon ip(s) */
407 	err = ceph_parse_ips(dev_name, *path, args->mon_addr,
408 			     CEPH_MAX_MON, &args->num_mon);
409 	if (err < 0)
410 		goto out;
411 
412 	/* path on server */
413 	*path += 2;
414 	dout("server path '%s'\n", *path);
415 
416 	/* parse mount options */
417 	while ((c = strsep(&options, ",")) != NULL) {
418 		int token, intval, ret;
419 		if (!*c)
420 			continue;
421 		err = -EINVAL;
422 		token = match_token((char *)c, arg_tokens, argstr);
423 		if (token < 0) {
424 			pr_err("bad mount option at '%s'\n", c);
425 			goto out;
426 		}
427 		if (token < Opt_last_int) {
428 			ret = match_int(&argstr[0], &intval);
429 			if (ret < 0) {
430 				pr_err("bad mount option arg (not int) "
431 				       "at '%s'\n", c);
432 				continue;
433 			}
434 			dout("got int token %d val %d\n", token, intval);
435 		} else if (token > Opt_last_int && token < Opt_last_string) {
436 			dout("got string token %d val %s\n", token,
437 			     argstr[0].from);
438 		} else {
439 			dout("got token %d\n", token);
440 		}
441 		switch (token) {
442 		case Opt_fsidmajor:
443 			*(__le64 *)&args->fsid.fsid[0] = cpu_to_le64(intval);
444 			break;
445 		case Opt_fsidminor:
446 			*(__le64 *)&args->fsid.fsid[8] = cpu_to_le64(intval);
447 			break;
448 		case Opt_ip:
449 			err = ceph_parse_ips(argstr[0].from,
450 					     argstr[0].to,
451 					     &args->my_addr,
452 					     1, NULL);
453 			if (err < 0)
454 				goto out;
455 			args->flags |= CEPH_OPT_MYIP;
456 			break;
457 
458 		case Opt_snapdirname:
459 			kfree(args->snapdir_name);
460 			args->snapdir_name = kstrndup(argstr[0].from,
461 					      argstr[0].to-argstr[0].from,
462 					      GFP_KERNEL);
463 			break;
464 		case Opt_name:
465 			args->name = kstrndup(argstr[0].from,
466 					      argstr[0].to-argstr[0].from,
467 					      GFP_KERNEL);
468 			break;
469 		case Opt_secret:
470 			args->secret = kstrndup(argstr[0].from,
471 						argstr[0].to-argstr[0].from,
472 						GFP_KERNEL);
473 			break;
474 
475 			/* misc */
476 		case Opt_wsize:
477 			args->wsize = intval;
478 			break;
479 		case Opt_rsize:
480 			args->rsize = intval;
481 			break;
482 		case Opt_osdtimeout:
483 			args->osd_timeout = intval;
484 			break;
485 		case Opt_osdkeepalivetimeout:
486 			args->osd_keepalive_timeout = intval;
487 			break;
488 		case Opt_mount_timeout:
489 			args->mount_timeout = intval;
490 			break;
491 		case Opt_caps_wanted_delay_min:
492 			args->caps_wanted_delay_min = intval;
493 			break;
494 		case Opt_caps_wanted_delay_max:
495 			args->caps_wanted_delay_max = intval;
496 			break;
497 		case Opt_readdir_max_entries:
498 			args->max_readdir = intval;
499 			break;
500 		case Opt_congestion_kb:
501 			args->congestion_kb = intval;
502 			break;
503 
504 		case Opt_noshare:
505 			args->flags |= CEPH_OPT_NOSHARE;
506 			break;
507 
508 		case Opt_dirstat:
509 			args->flags |= CEPH_OPT_DIRSTAT;
510 			break;
511 		case Opt_nodirstat:
512 			args->flags &= ~CEPH_OPT_DIRSTAT;
513 			break;
514 		case Opt_rbytes:
515 			args->flags |= CEPH_OPT_RBYTES;
516 			break;
517 		case Opt_norbytes:
518 			args->flags &= ~CEPH_OPT_RBYTES;
519 			break;
520 		case Opt_nocrc:
521 			args->flags |= CEPH_OPT_NOCRC;
522 			break;
523 		case Opt_noasyncreaddir:
524 			args->flags |= CEPH_OPT_NOASYNCREADDIR;
525 			break;
526 
527 		default:
528 			BUG_ON(token);
529 		}
530 	}
531 	return args;
532 
533 out:
534 	kfree(args->mon_addr);
535 	kfree(args);
536 	return ERR_PTR(err);
537 }
538 
539 static void destroy_mount_args(struct ceph_mount_args *args)
540 {
541 	dout("destroy_mount_args %p\n", args);
542 	kfree(args->snapdir_name);
543 	args->snapdir_name = NULL;
544 	kfree(args->name);
545 	args->name = NULL;
546 	kfree(args->secret);
547 	args->secret = NULL;
548 	kfree(args);
549 }
550 
551 /*
552  * create a fresh client instance
553  */
554 static struct ceph_client *ceph_create_client(struct ceph_mount_args *args)
555 {
556 	struct ceph_client *client;
557 	int err = -ENOMEM;
558 
559 	client = kzalloc(sizeof(*client), GFP_KERNEL);
560 	if (client == NULL)
561 		return ERR_PTR(-ENOMEM);
562 
563 	mutex_init(&client->mount_mutex);
564 
565 	init_waitqueue_head(&client->auth_wq);
566 
567 	client->sb = NULL;
568 	client->mount_state = CEPH_MOUNT_MOUNTING;
569 	client->mount_args = args;
570 
571 	client->msgr = NULL;
572 
573 	client->auth_err = 0;
574 	atomic_long_set(&client->writeback_count, 0);
575 
576 	err = bdi_init(&client->backing_dev_info);
577 	if (err < 0)
578 		goto fail;
579 
580 	err = -ENOMEM;
581 	client->wb_wq = create_workqueue("ceph-writeback");
582 	if (client->wb_wq == NULL)
583 		goto fail_bdi;
584 	client->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
585 	if (client->pg_inv_wq == NULL)
586 		goto fail_wb_wq;
587 	client->trunc_wq = create_singlethread_workqueue("ceph-trunc");
588 	if (client->trunc_wq == NULL)
589 		goto fail_pg_inv_wq;
590 
591 	/* set up mempools */
592 	err = -ENOMEM;
593 	client->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
594 			      client->mount_args->wsize >> PAGE_CACHE_SHIFT);
595 	if (!client->wb_pagevec_pool)
596 		goto fail_trunc_wq;
597 
598 	/* caps */
599 	client->min_caps = args->max_readdir;
600 	ceph_adjust_min_caps(client->min_caps);
601 
602 	/* subsystems */
603 	err = ceph_monc_init(&client->monc, client);
604 	if (err < 0)
605 		goto fail_mempool;
606 	err = ceph_osdc_init(&client->osdc, client);
607 	if (err < 0)
608 		goto fail_monc;
609 	err = ceph_mdsc_init(&client->mdsc, client);
610 	if (err < 0)
611 		goto fail_osdc;
612 	return client;
613 
614 fail_osdc:
615 	ceph_osdc_stop(&client->osdc);
616 fail_monc:
617 	ceph_monc_stop(&client->monc);
618 fail_mempool:
619 	mempool_destroy(client->wb_pagevec_pool);
620 fail_trunc_wq:
621 	destroy_workqueue(client->trunc_wq);
622 fail_pg_inv_wq:
623 	destroy_workqueue(client->pg_inv_wq);
624 fail_wb_wq:
625 	destroy_workqueue(client->wb_wq);
626 fail_bdi:
627 	bdi_destroy(&client->backing_dev_info);
628 fail:
629 	kfree(client);
630 	return ERR_PTR(err);
631 }
632 
633 static void ceph_destroy_client(struct ceph_client *client)
634 {
635 	dout("destroy_client %p\n", client);
636 
637 	/* unmount */
638 	ceph_mdsc_stop(&client->mdsc);
639 	ceph_monc_stop(&client->monc);
640 	ceph_osdc_stop(&client->osdc);
641 
642 	ceph_adjust_min_caps(-client->min_caps);
643 
644 	ceph_debugfs_client_cleanup(client);
645 	destroy_workqueue(client->wb_wq);
646 	destroy_workqueue(client->pg_inv_wq);
647 	destroy_workqueue(client->trunc_wq);
648 
649 	bdi_destroy(&client->backing_dev_info);
650 
651 	if (client->msgr)
652 		ceph_messenger_destroy(client->msgr);
653 	mempool_destroy(client->wb_pagevec_pool);
654 
655 	destroy_mount_args(client->mount_args);
656 
657 	kfree(client);
658 	dout("destroy_client %p done\n", client);
659 }
660 
661 /*
662  * Initially learn our fsid, or verify an fsid matches.
663  */
664 int ceph_check_fsid(struct ceph_client *client, struct ceph_fsid *fsid)
665 {
666 	if (client->have_fsid) {
667 		if (ceph_fsid_compare(&client->fsid, fsid)) {
668 			pr_err("bad fsid, had " FSID_FORMAT " got " FSID_FORMAT,
669 			       PR_FSID(&client->fsid), PR_FSID(fsid));
670 			return -1;
671 		}
672 	} else {
673 		pr_info("client%lld fsid " FSID_FORMAT "\n",
674 			client->monc.auth->global_id, PR_FSID(fsid));
675 		memcpy(&client->fsid, fsid, sizeof(*fsid));
676 		ceph_debugfs_client_init(client);
677 		client->have_fsid = true;
678 	}
679 	return 0;
680 }
681 
682 /*
683  * true if we have the mon map (and have thus joined the cluster)
684  */
685 static int have_mon_map(struct ceph_client *client)
686 {
687 	return client->monc.monmap && client->monc.monmap->epoch;
688 }
689 
690 /*
691  * Bootstrap mount by opening the root directory.  Note the mount
692  * @started time from caller, and time out if this takes too long.
693  */
694 static struct dentry *open_root_dentry(struct ceph_client *client,
695 				       const char *path,
696 				       unsigned long started)
697 {
698 	struct ceph_mds_client *mdsc = &client->mdsc;
699 	struct ceph_mds_request *req = NULL;
700 	int err;
701 	struct dentry *root;
702 
703 	/* open dir */
704 	dout("open_root_inode opening '%s'\n", path);
705 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
706 	if (IS_ERR(req))
707 		return ERR_PTR(PTR_ERR(req));
708 	req->r_path1 = kstrdup(path, GFP_NOFS);
709 	req->r_ino1.ino = CEPH_INO_ROOT;
710 	req->r_ino1.snap = CEPH_NOSNAP;
711 	req->r_started = started;
712 	req->r_timeout = client->mount_args->mount_timeout * HZ;
713 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
714 	req->r_num_caps = 2;
715 	err = ceph_mdsc_do_request(mdsc, NULL, req);
716 	if (err == 0) {
717 		dout("open_root_inode success\n");
718 		if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
719 		    client->sb->s_root == NULL)
720 			root = d_alloc_root(req->r_target_inode);
721 		else
722 			root = d_obtain_alias(req->r_target_inode);
723 		req->r_target_inode = NULL;
724 		dout("open_root_inode success, root dentry is %p\n", root);
725 	} else {
726 		root = ERR_PTR(err);
727 	}
728 	ceph_mdsc_put_request(req);
729 	return root;
730 }
731 
732 /*
733  * mount: join the ceph cluster, and open root directory.
734  */
735 static int ceph_mount(struct ceph_client *client, struct vfsmount *mnt,
736 		      const char *path)
737 {
738 	struct ceph_entity_addr *myaddr = NULL;
739 	int err;
740 	unsigned long timeout = client->mount_args->mount_timeout * HZ;
741 	unsigned long started = jiffies;  /* note the start time */
742 	struct dentry *root;
743 
744 	dout("mount start\n");
745 	mutex_lock(&client->mount_mutex);
746 
747 	/* initialize the messenger */
748 	if (client->msgr == NULL) {
749 		if (ceph_test_opt(client, MYIP))
750 			myaddr = &client->mount_args->my_addr;
751 		client->msgr = ceph_messenger_create(myaddr);
752 		if (IS_ERR(client->msgr)) {
753 			err = PTR_ERR(client->msgr);
754 			client->msgr = NULL;
755 			goto out;
756 		}
757 		client->msgr->nocrc = ceph_test_opt(client, NOCRC);
758 	}
759 
760 	/* open session, and wait for mon, mds, and osd maps */
761 	err = ceph_monc_open_session(&client->monc);
762 	if (err < 0)
763 		goto out;
764 
765 	while (!have_mon_map(client)) {
766 		err = -EIO;
767 		if (timeout && time_after_eq(jiffies, started + timeout))
768 			goto out;
769 
770 		/* wait */
771 		dout("mount waiting for mon_map\n");
772 		err = wait_event_interruptible_timeout(client->auth_wq,
773 			       have_mon_map(client) || (client->auth_err < 0),
774 			       timeout);
775 		if (err == -EINTR || err == -ERESTARTSYS)
776 			goto out;
777 		if (client->auth_err < 0) {
778 			err = client->auth_err;
779 			goto out;
780 		}
781 	}
782 
783 	dout("mount opening root\n");
784 	root = open_root_dentry(client, "", started);
785 	if (IS_ERR(root)) {
786 		err = PTR_ERR(root);
787 		goto out;
788 	}
789 	if (client->sb->s_root)
790 		dput(root);
791 	else
792 		client->sb->s_root = root;
793 
794 	if (path[0] == 0) {
795 		dget(root);
796 	} else {
797 		dout("mount opening base mountpoint\n");
798 		root = open_root_dentry(client, path, started);
799 		if (IS_ERR(root)) {
800 			err = PTR_ERR(root);
801 			dput(client->sb->s_root);
802 			client->sb->s_root = NULL;
803 			goto out;
804 		}
805 	}
806 
807 	mnt->mnt_root = root;
808 	mnt->mnt_sb = client->sb;
809 
810 	client->mount_state = CEPH_MOUNT_MOUNTED;
811 	dout("mount success\n");
812 	err = 0;
813 
814 out:
815 	mutex_unlock(&client->mount_mutex);
816 	return err;
817 }
818 
819 static int ceph_set_super(struct super_block *s, void *data)
820 {
821 	struct ceph_client *client = data;
822 	int ret;
823 
824 	dout("set_super %p data %p\n", s, data);
825 
826 	s->s_flags = client->mount_args->sb_flags;
827 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
828 
829 	s->s_fs_info = client;
830 	client->sb = s;
831 
832 	s->s_op = &ceph_super_ops;
833 	s->s_export_op = &ceph_export_ops;
834 
835 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
836 
837 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
838 	if (ret != 0)
839 		goto fail;
840 
841 	return ret;
842 
843 fail:
844 	s->s_fs_info = NULL;
845 	client->sb = NULL;
846 	return ret;
847 }
848 
849 /*
850  * share superblock if same fs AND options
851  */
852 static int ceph_compare_super(struct super_block *sb, void *data)
853 {
854 	struct ceph_client *new = data;
855 	struct ceph_mount_args *args = new->mount_args;
856 	struct ceph_client *other = ceph_sb_to_client(sb);
857 	int i;
858 
859 	dout("ceph_compare_super %p\n", sb);
860 	if (args->flags & CEPH_OPT_FSID) {
861 		if (ceph_fsid_compare(&args->fsid, &other->fsid)) {
862 			dout("fsid doesn't match\n");
863 			return 0;
864 		}
865 	} else {
866 		/* do we share (a) monitor? */
867 		for (i = 0; i < new->monc.monmap->num_mon; i++)
868 			if (ceph_monmap_contains(other->monc.monmap,
869 					 &new->monc.monmap->mon_inst[i].addr))
870 				break;
871 		if (i == new->monc.monmap->num_mon) {
872 			dout("mon ip not part of monmap\n");
873 			return 0;
874 		}
875 		dout("mon ip matches existing sb %p\n", sb);
876 	}
877 	if (args->sb_flags != other->mount_args->sb_flags) {
878 		dout("flags differ\n");
879 		return 0;
880 	}
881 	return 1;
882 }
883 
884 /*
885  * construct our own bdi so we can control readahead, etc.
886  */
887 static int ceph_register_bdi(struct super_block *sb, struct ceph_client *client)
888 {
889 	int err;
890 
891 	/* set ra_pages based on rsize mount option? */
892 	if (client->mount_args->rsize >= PAGE_CACHE_SIZE)
893 		client->backing_dev_info.ra_pages =
894 			(client->mount_args->rsize + PAGE_CACHE_SIZE - 1)
895 			>> PAGE_SHIFT;
896 	err = bdi_register_dev(&client->backing_dev_info, sb->s_dev);
897 	if (!err)
898 		sb->s_bdi = &client->backing_dev_info;
899 	return err;
900 }
901 
902 static int ceph_get_sb(struct file_system_type *fs_type,
903 		       int flags, const char *dev_name, void *data,
904 		       struct vfsmount *mnt)
905 {
906 	struct super_block *sb;
907 	struct ceph_client *client;
908 	int err;
909 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
910 	const char *path = NULL;
911 	struct ceph_mount_args *args;
912 
913 	dout("ceph_get_sb\n");
914 	args = parse_mount_args(flags, data, dev_name, &path);
915 	if (IS_ERR(args)) {
916 		err = PTR_ERR(args);
917 		goto out_final;
918 	}
919 
920 	/* create client (which we may/may not use) */
921 	client = ceph_create_client(args);
922 	if (IS_ERR(client)) {
923 		err = PTR_ERR(client);
924 		goto out_final;
925 	}
926 
927 	if (client->mount_args->flags & CEPH_OPT_NOSHARE)
928 		compare_super = NULL;
929 	sb = sget(fs_type, compare_super, ceph_set_super, client);
930 	if (IS_ERR(sb)) {
931 		err = PTR_ERR(sb);
932 		goto out;
933 	}
934 
935 	if (ceph_client(sb) != client) {
936 		ceph_destroy_client(client);
937 		client = ceph_client(sb);
938 		dout("get_sb got existing client %p\n", client);
939 	} else {
940 		dout("get_sb using new client %p\n", client);
941 		err = ceph_register_bdi(sb, client);
942 		if (err < 0)
943 			goto out_splat;
944 	}
945 
946 	err = ceph_mount(client, mnt, path);
947 	if (err < 0)
948 		goto out_splat;
949 	dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
950 	     mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
951 	return 0;
952 
953 out_splat:
954 	ceph_mdsc_close_sessions(&client->mdsc);
955 	up_write(&sb->s_umount);
956 	deactivate_super(sb);
957 	goto out_final;
958 
959 out:
960 	ceph_destroy_client(client);
961 out_final:
962 	dout("ceph_get_sb fail %d\n", err);
963 	return err;
964 }
965 
966 static void ceph_kill_sb(struct super_block *s)
967 {
968 	struct ceph_client *client = ceph_sb_to_client(s);
969 	dout("kill_sb %p\n", s);
970 	ceph_mdsc_pre_umount(&client->mdsc);
971 	kill_anon_super(s);    /* will call put_super after sb is r/o */
972 	ceph_destroy_client(client);
973 }
974 
975 static struct file_system_type ceph_fs_type = {
976 	.owner		= THIS_MODULE,
977 	.name		= "ceph",
978 	.get_sb		= ceph_get_sb,
979 	.kill_sb	= ceph_kill_sb,
980 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
981 };
982 
983 #define _STRINGIFY(x) #x
984 #define STRINGIFY(x) _STRINGIFY(x)
985 
986 static int __init init_ceph(void)
987 {
988 	int ret = 0;
989 
990 	ret = ceph_debugfs_init();
991 	if (ret < 0)
992 		goto out;
993 
994 	ret = ceph_msgr_init();
995 	if (ret < 0)
996 		goto out_debugfs;
997 
998 	ret = init_caches();
999 	if (ret)
1000 		goto out_msgr;
1001 
1002 	ceph_caps_init();
1003 
1004 	ret = register_filesystem(&ceph_fs_type);
1005 	if (ret)
1006 		goto out_icache;
1007 
1008 	pr_info("loaded (mon/mds/osd proto %d/%d/%d, osdmap %d/%d %d/%d)\n",
1009 		CEPH_MONC_PROTOCOL, CEPH_MDSC_PROTOCOL, CEPH_OSDC_PROTOCOL,
1010 		CEPH_OSDMAP_VERSION, CEPH_OSDMAP_VERSION_EXT,
1011 		CEPH_OSDMAP_INC_VERSION, CEPH_OSDMAP_INC_VERSION_EXT);
1012 	return 0;
1013 
1014 out_icache:
1015 	destroy_caches();
1016 out_msgr:
1017 	ceph_msgr_exit();
1018 out_debugfs:
1019 	ceph_debugfs_cleanup();
1020 out:
1021 	return ret;
1022 }
1023 
1024 static void __exit exit_ceph(void)
1025 {
1026 	dout("exit_ceph\n");
1027 	unregister_filesystem(&ceph_fs_type);
1028 	ceph_caps_finalize();
1029 	destroy_caches();
1030 	ceph_msgr_exit();
1031 	ceph_debugfs_cleanup();
1032 }
1033 
1034 module_init(init_ceph);
1035 module_exit(exit_ceph);
1036 
1037 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1038 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1039 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1040 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1041 MODULE_LICENSE("GPL");
1042