xref: /linux/fs/ceph/super.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
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(dentry->d_inode);
48 	struct ceph_monmap *monmap = fsc->client->monc.monmap;
49 	struct ceph_statfs st;
50 	u64 fsid;
51 	int err;
52 
53 	dout("statfs\n");
54 	err = ceph_monc_do_statfs(&fsc->client->monc, &st);
55 	if (err < 0)
56 		return err;
57 
58 	/* fill in kstatfs */
59 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
60 
61 	/*
62 	 * express utilization in terms of large blocks to avoid
63 	 * overflow on 32-bit machines.
64 	 *
65 	 * NOTE: for the time being, we make bsize == frsize to humor
66 	 * not-yet-ancient versions of glibc that are broken.
67 	 * Someday, we will probably want to report a real block
68 	 * size...  whatever that may mean for a network file system!
69 	 */
70 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
71 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
72 	buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
73 	buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
74 	buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
75 
76 	buf->f_files = le64_to_cpu(st.num_objects);
77 	buf->f_ffree = -1;
78 	buf->f_namelen = NAME_MAX;
79 
80 	/* leave fsid little-endian, regardless of host endianness */
81 	fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
82 	buf->f_fsid.val[0] = fsid & 0xffffffff;
83 	buf->f_fsid.val[1] = fsid >> 32;
84 
85 	return 0;
86 }
87 
88 
89 static int ceph_sync_fs(struct super_block *sb, int wait)
90 {
91 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
92 
93 	if (!wait) {
94 		dout("sync_fs (non-blocking)\n");
95 		ceph_flush_dirty_caps(fsc->mdsc);
96 		dout("sync_fs (non-blocking) done\n");
97 		return 0;
98 	}
99 
100 	dout("sync_fs (blocking)\n");
101 	ceph_osdc_sync(&fsc->client->osdc);
102 	ceph_mdsc_sync(fsc->mdsc);
103 	dout("sync_fs (blocking) done\n");
104 	return 0;
105 }
106 
107 /*
108  * mount options
109  */
110 enum {
111 	Opt_wsize,
112 	Opt_rsize,
113 	Opt_rasize,
114 	Opt_caps_wanted_delay_min,
115 	Opt_caps_wanted_delay_max,
116 	Opt_cap_release_safety,
117 	Opt_readdir_max_entries,
118 	Opt_readdir_max_bytes,
119 	Opt_congestion_kb,
120 	Opt_last_int,
121 	/* int args above */
122 	Opt_snapdirname,
123 	Opt_last_string,
124 	/* string args above */
125 	Opt_dirstat,
126 	Opt_nodirstat,
127 	Opt_rbytes,
128 	Opt_norbytes,
129 	Opt_asyncreaddir,
130 	Opt_noasyncreaddir,
131 	Opt_dcache,
132 	Opt_nodcache,
133 	Opt_ino32,
134 	Opt_noino32,
135 	Opt_fscache,
136 	Opt_nofscache,
137 #ifdef CONFIG_CEPH_FS_POSIX_ACL
138 	Opt_acl,
139 #endif
140 	Opt_noacl
141 };
142 
143 static match_table_t fsopt_tokens = {
144 	{Opt_wsize, "wsize=%d"},
145 	{Opt_rsize, "rsize=%d"},
146 	{Opt_rasize, "rasize=%d"},
147 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
148 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
149 	{Opt_cap_release_safety, "cap_release_safety=%d"},
150 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
151 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
152 	{Opt_congestion_kb, "write_congestion_kb=%d"},
153 	/* int args above */
154 	{Opt_snapdirname, "snapdirname=%s"},
155 	/* string args above */
156 	{Opt_dirstat, "dirstat"},
157 	{Opt_nodirstat, "nodirstat"},
158 	{Opt_rbytes, "rbytes"},
159 	{Opt_norbytes, "norbytes"},
160 	{Opt_asyncreaddir, "asyncreaddir"},
161 	{Opt_noasyncreaddir, "noasyncreaddir"},
162 	{Opt_dcache, "dcache"},
163 	{Opt_nodcache, "nodcache"},
164 	{Opt_ino32, "ino32"},
165 	{Opt_noino32, "noino32"},
166 	{Opt_fscache, "fsc"},
167 	{Opt_nofscache, "nofsc"},
168 #ifdef CONFIG_CEPH_FS_POSIX_ACL
169 	{Opt_acl, "acl"},
170 #endif
171 	{Opt_noacl, "noacl"},
172 	{-1, NULL}
173 };
174 
175 static int parse_fsopt_token(char *c, void *private)
176 {
177 	struct ceph_mount_options *fsopt = private;
178 	substring_t argstr[MAX_OPT_ARGS];
179 	int token, intval, ret;
180 
181 	token = match_token((char *)c, fsopt_tokens, argstr);
182 	if (token < 0)
183 		return -EINVAL;
184 
185 	if (token < Opt_last_int) {
186 		ret = match_int(&argstr[0], &intval);
187 		if (ret < 0) {
188 			pr_err("bad mount option arg (not int) "
189 			       "at '%s'\n", c);
190 			return ret;
191 		}
192 		dout("got int token %d val %d\n", token, intval);
193 	} else if (token > Opt_last_int && token < Opt_last_string) {
194 		dout("got string token %d val %s\n", token,
195 		     argstr[0].from);
196 	} else {
197 		dout("got token %d\n", token);
198 	}
199 
200 	switch (token) {
201 	case Opt_snapdirname:
202 		kfree(fsopt->snapdir_name);
203 		fsopt->snapdir_name = kstrndup(argstr[0].from,
204 					       argstr[0].to-argstr[0].from,
205 					       GFP_KERNEL);
206 		if (!fsopt->snapdir_name)
207 			return -ENOMEM;
208 		break;
209 
210 		/* misc */
211 	case Opt_wsize:
212 		fsopt->wsize = intval;
213 		break;
214 	case Opt_rsize:
215 		fsopt->rsize = intval;
216 		break;
217 	case Opt_rasize:
218 		fsopt->rasize = intval;
219 		break;
220 	case Opt_caps_wanted_delay_min:
221 		fsopt->caps_wanted_delay_min = intval;
222 		break;
223 	case Opt_caps_wanted_delay_max:
224 		fsopt->caps_wanted_delay_max = intval;
225 		break;
226 	case Opt_readdir_max_entries:
227 		fsopt->max_readdir = intval;
228 		break;
229 	case Opt_readdir_max_bytes:
230 		fsopt->max_readdir_bytes = intval;
231 		break;
232 	case Opt_congestion_kb:
233 		fsopt->congestion_kb = intval;
234 		break;
235 	case Opt_dirstat:
236 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
237 		break;
238 	case Opt_nodirstat:
239 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
240 		break;
241 	case Opt_rbytes:
242 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
243 		break;
244 	case Opt_norbytes:
245 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
246 		break;
247 	case Opt_asyncreaddir:
248 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
249 		break;
250 	case Opt_noasyncreaddir:
251 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
252 		break;
253 	case Opt_dcache:
254 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
255 		break;
256 	case Opt_nodcache:
257 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
258 		break;
259 	case Opt_ino32:
260 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
261 		break;
262 	case Opt_noino32:
263 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
264 		break;
265 	case Opt_fscache:
266 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
267 		break;
268 	case Opt_nofscache:
269 		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
270 		break;
271 #ifdef CONFIG_CEPH_FS_POSIX_ACL
272 	case Opt_acl:
273 		fsopt->sb_flags |= MS_POSIXACL;
274 		break;
275 #endif
276 	case Opt_noacl:
277 		fsopt->sb_flags &= ~MS_POSIXACL;
278 		break;
279 	default:
280 		BUG_ON(token);
281 	}
282 	return 0;
283 }
284 
285 static void destroy_mount_options(struct ceph_mount_options *args)
286 {
287 	dout("destroy_mount_options %p\n", args);
288 	kfree(args->snapdir_name);
289 	kfree(args);
290 }
291 
292 static int strcmp_null(const char *s1, const char *s2)
293 {
294 	if (!s1 && !s2)
295 		return 0;
296 	if (s1 && !s2)
297 		return -1;
298 	if (!s1 && s2)
299 		return 1;
300 	return strcmp(s1, s2);
301 }
302 
303 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
304 				 struct ceph_options *new_opt,
305 				 struct ceph_fs_client *fsc)
306 {
307 	struct ceph_mount_options *fsopt1 = new_fsopt;
308 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
309 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
310 	int ret;
311 
312 	ret = memcmp(fsopt1, fsopt2, ofs);
313 	if (ret)
314 		return ret;
315 
316 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
317 	if (ret)
318 		return ret;
319 
320 	return ceph_compare_options(new_opt, fsc->client);
321 }
322 
323 static int parse_mount_options(struct ceph_mount_options **pfsopt,
324 			       struct ceph_options **popt,
325 			       int flags, char *options,
326 			       const char *dev_name,
327 			       const char **path)
328 {
329 	struct ceph_mount_options *fsopt;
330 	const char *dev_name_end;
331 	int err;
332 
333 	if (!dev_name || !*dev_name)
334 		return -EINVAL;
335 
336 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
337 	if (!fsopt)
338 		return -ENOMEM;
339 
340 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
341 
342 	fsopt->sb_flags = flags;
343 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
344 
345 	fsopt->rsize = CEPH_RSIZE_DEFAULT;
346 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
347 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
348 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
349 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
350 	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
351 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
352 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
353 	fsopt->congestion_kb = default_congestion_kb();
354 
355 	/*
356 	 * Distinguish the server list from the path in "dev_name".
357 	 * Internally we do not include the leading '/' in the path.
358 	 *
359 	 * "dev_name" will look like:
360 	 *     <server_spec>[,<server_spec>...]:[<path>]
361 	 * where
362 	 *     <server_spec> is <ip>[:<port>]
363 	 *     <path> is optional, but if present must begin with '/'
364 	 */
365 	dev_name_end = strchr(dev_name, '/');
366 	if (dev_name_end) {
367 		/* skip over leading '/' for path */
368 		*path = dev_name_end + 1;
369 	} else {
370 		/* path is empty */
371 		dev_name_end = dev_name + strlen(dev_name);
372 		*path = dev_name_end;
373 	}
374 	err = -EINVAL;
375 	dev_name_end--;		/* back up to ':' separator */
376 	if (dev_name_end < dev_name || *dev_name_end != ':') {
377 		pr_err("device name is missing path (no : separator in %s)\n",
378 				dev_name);
379 		goto out;
380 	}
381 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
382 	dout("server path '%s'\n", *path);
383 
384 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
385 				 parse_fsopt_token, (void *)fsopt);
386 	if (IS_ERR(*popt)) {
387 		err = PTR_ERR(*popt);
388 		goto out;
389 	}
390 
391 	/* success */
392 	*pfsopt = fsopt;
393 	return 0;
394 
395 out:
396 	destroy_mount_options(fsopt);
397 	return err;
398 }
399 
400 /**
401  * ceph_show_options - Show mount options in /proc/mounts
402  * @m: seq_file to write to
403  * @root: root of that (sub)tree
404  */
405 static int ceph_show_options(struct seq_file *m, struct dentry *root)
406 {
407 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
408 	struct ceph_mount_options *fsopt = fsc->mount_options;
409 	struct ceph_options *opt = fsc->client->options;
410 
411 	if (opt->flags & CEPH_OPT_FSID)
412 		seq_printf(m, ",fsid=%pU", &opt->fsid);
413 	if (opt->flags & CEPH_OPT_NOSHARE)
414 		seq_puts(m, ",noshare");
415 	if (opt->flags & CEPH_OPT_NOCRC)
416 		seq_puts(m, ",nocrc");
417 	if (opt->flags & CEPH_OPT_NOMSGAUTH)
418 		seq_puts(m, ",nocephx_require_signatures");
419 	if ((opt->flags & CEPH_OPT_TCP_NODELAY) == 0)
420 		seq_puts(m, ",notcp_nodelay");
421 
422 	if (opt->name)
423 		seq_printf(m, ",name=%s", opt->name);
424 	if (opt->key)
425 		seq_puts(m, ",secret=<hidden>");
426 
427 	if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
428 		seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
429 	if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
430 		seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
431 	if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
432 		seq_printf(m, ",osdkeepalivetimeout=%d",
433 			   opt->osd_keepalive_timeout);
434 
435 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
436 		seq_puts(m, ",dirstat");
437 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
438 		seq_puts(m, ",norbytes");
439 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
440 		seq_puts(m, ",noasyncreaddir");
441 	if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
442 		seq_puts(m, ",dcache");
443 	else
444 		seq_puts(m, ",nodcache");
445 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
446 		seq_puts(m, ",fsc");
447 	else
448 		seq_puts(m, ",nofsc");
449 
450 #ifdef CONFIG_CEPH_FS_POSIX_ACL
451 	if (fsopt->sb_flags & MS_POSIXACL)
452 		seq_puts(m, ",acl");
453 	else
454 		seq_puts(m, ",noacl");
455 #endif
456 
457 	if (fsopt->wsize)
458 		seq_printf(m, ",wsize=%d", fsopt->wsize);
459 	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
460 		seq_printf(m, ",rsize=%d", fsopt->rsize);
461 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
462 		seq_printf(m, ",rasize=%d", fsopt->rasize);
463 	if (fsopt->congestion_kb != default_congestion_kb())
464 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
465 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
466 		seq_printf(m, ",caps_wanted_delay_min=%d",
467 			 fsopt->caps_wanted_delay_min);
468 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
469 		seq_printf(m, ",caps_wanted_delay_max=%d",
470 			   fsopt->caps_wanted_delay_max);
471 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
472 		seq_printf(m, ",cap_release_safety=%d",
473 			   fsopt->cap_release_safety);
474 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
475 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
476 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
477 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
478 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
479 		seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
480 	return 0;
481 }
482 
483 /*
484  * handle any mon messages the standard library doesn't understand.
485  * return error if we don't either.
486  */
487 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
488 {
489 	struct ceph_fs_client *fsc = client->private;
490 	int type = le16_to_cpu(msg->hdr.type);
491 
492 	switch (type) {
493 	case CEPH_MSG_MDS_MAP:
494 		ceph_mdsc_handle_map(fsc->mdsc, msg);
495 		return 0;
496 
497 	default:
498 		return -1;
499 	}
500 }
501 
502 /*
503  * create a new fs client
504  */
505 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
506 					struct ceph_options *opt)
507 {
508 	struct ceph_fs_client *fsc;
509 	const u64 supported_features =
510 		CEPH_FEATURE_FLOCK |
511 		CEPH_FEATURE_DIRLAYOUTHASH |
512 		CEPH_FEATURE_MDS_INLINE_DATA;
513 	const u64 required_features = 0;
514 	int page_count;
515 	size_t size;
516 	int err = -ENOMEM;
517 
518 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
519 	if (!fsc)
520 		return ERR_PTR(-ENOMEM);
521 
522 	fsc->client = ceph_create_client(opt, fsc, supported_features,
523 					 required_features);
524 	if (IS_ERR(fsc->client)) {
525 		err = PTR_ERR(fsc->client);
526 		goto fail;
527 	}
528 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
529 	fsc->client->monc.want_mdsmap = 1;
530 
531 	fsc->mount_options = fsopt;
532 
533 	fsc->sb = NULL;
534 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
535 
536 	atomic_long_set(&fsc->writeback_count, 0);
537 
538 	err = bdi_init(&fsc->backing_dev_info);
539 	if (err < 0)
540 		goto fail_client;
541 
542 	err = -ENOMEM;
543 	/*
544 	 * The number of concurrent works can be high but they don't need
545 	 * to be processed in parallel, limit concurrency.
546 	 */
547 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
548 	if (fsc->wb_wq == NULL)
549 		goto fail_bdi;
550 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
551 	if (fsc->pg_inv_wq == NULL)
552 		goto fail_wb_wq;
553 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
554 	if (fsc->trunc_wq == NULL)
555 		goto fail_pg_inv_wq;
556 
557 	/* set up mempools */
558 	err = -ENOMEM;
559 	page_count = fsc->mount_options->wsize >> PAGE_CACHE_SHIFT;
560 	size = sizeof (struct page *) * (page_count ? page_count : 1);
561 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
562 	if (!fsc->wb_pagevec_pool)
563 		goto fail_trunc_wq;
564 
565 	/* setup fscache */
566 	if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
567 	    (ceph_fscache_register_fs(fsc) != 0))
568 		goto fail_fscache;
569 
570 	/* caps */
571 	fsc->min_caps = fsopt->max_readdir;
572 
573 	return fsc;
574 
575 fail_fscache:
576 	ceph_fscache_unregister_fs(fsc);
577 fail_trunc_wq:
578 	destroy_workqueue(fsc->trunc_wq);
579 fail_pg_inv_wq:
580 	destroy_workqueue(fsc->pg_inv_wq);
581 fail_wb_wq:
582 	destroy_workqueue(fsc->wb_wq);
583 fail_bdi:
584 	bdi_destroy(&fsc->backing_dev_info);
585 fail_client:
586 	ceph_destroy_client(fsc->client);
587 fail:
588 	kfree(fsc);
589 	return ERR_PTR(err);
590 }
591 
592 static void destroy_fs_client(struct ceph_fs_client *fsc)
593 {
594 	dout("destroy_fs_client %p\n", fsc);
595 
596 	ceph_fscache_unregister_fs(fsc);
597 
598 	destroy_workqueue(fsc->wb_wq);
599 	destroy_workqueue(fsc->pg_inv_wq);
600 	destroy_workqueue(fsc->trunc_wq);
601 
602 	bdi_destroy(&fsc->backing_dev_info);
603 
604 	mempool_destroy(fsc->wb_pagevec_pool);
605 
606 	destroy_mount_options(fsc->mount_options);
607 
608 	ceph_fs_debugfs_cleanup(fsc);
609 
610 	ceph_destroy_client(fsc->client);
611 
612 	kfree(fsc);
613 	dout("destroy_fs_client %p done\n", fsc);
614 }
615 
616 /*
617  * caches
618  */
619 struct kmem_cache *ceph_inode_cachep;
620 struct kmem_cache *ceph_cap_cachep;
621 struct kmem_cache *ceph_dentry_cachep;
622 struct kmem_cache *ceph_file_cachep;
623 
624 static void ceph_inode_init_once(void *foo)
625 {
626 	struct ceph_inode_info *ci = foo;
627 	inode_init_once(&ci->vfs_inode);
628 }
629 
630 static int __init init_caches(void)
631 {
632 	int error = -ENOMEM;
633 
634 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
635 				      sizeof(struct ceph_inode_info),
636 				      __alignof__(struct ceph_inode_info),
637 				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
638 				      ceph_inode_init_once);
639 	if (ceph_inode_cachep == NULL)
640 		return -ENOMEM;
641 
642 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
643 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
644 	if (ceph_cap_cachep == NULL)
645 		goto bad_cap;
646 
647 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
648 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
649 	if (ceph_dentry_cachep == NULL)
650 		goto bad_dentry;
651 
652 	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
653 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
654 	if (ceph_file_cachep == NULL)
655 		goto bad_file;
656 
657 	if ((error = ceph_fscache_register()))
658 		goto bad_file;
659 
660 	return 0;
661 bad_file:
662 	kmem_cache_destroy(ceph_dentry_cachep);
663 bad_dentry:
664 	kmem_cache_destroy(ceph_cap_cachep);
665 bad_cap:
666 	kmem_cache_destroy(ceph_inode_cachep);
667 	return error;
668 }
669 
670 static void destroy_caches(void)
671 {
672 	/*
673 	 * Make sure all delayed rcu free inodes are flushed before we
674 	 * destroy cache.
675 	 */
676 	rcu_barrier();
677 
678 	kmem_cache_destroy(ceph_inode_cachep);
679 	kmem_cache_destroy(ceph_cap_cachep);
680 	kmem_cache_destroy(ceph_dentry_cachep);
681 	kmem_cache_destroy(ceph_file_cachep);
682 
683 	ceph_fscache_unregister();
684 }
685 
686 
687 /*
688  * ceph_umount_begin - initiate forced umount.  Tear down down the
689  * mount, skipping steps that may hang while waiting for server(s).
690  */
691 static void ceph_umount_begin(struct super_block *sb)
692 {
693 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
694 
695 	dout("ceph_umount_begin - starting forced umount\n");
696 	if (!fsc)
697 		return;
698 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
699 	return;
700 }
701 
702 static const struct super_operations ceph_super_ops = {
703 	.alloc_inode	= ceph_alloc_inode,
704 	.destroy_inode	= ceph_destroy_inode,
705 	.write_inode    = ceph_write_inode,
706 	.drop_inode	= ceph_drop_inode,
707 	.sync_fs        = ceph_sync_fs,
708 	.put_super	= ceph_put_super,
709 	.show_options   = ceph_show_options,
710 	.statfs		= ceph_statfs,
711 	.umount_begin   = ceph_umount_begin,
712 };
713 
714 /*
715  * Bootstrap mount by opening the root directory.  Note the mount
716  * @started time from caller, and time out if this takes too long.
717  */
718 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
719 				       const char *path,
720 				       unsigned long started)
721 {
722 	struct ceph_mds_client *mdsc = fsc->mdsc;
723 	struct ceph_mds_request *req = NULL;
724 	int err;
725 	struct dentry *root;
726 
727 	/* open dir */
728 	dout("open_root_inode opening '%s'\n", path);
729 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
730 	if (IS_ERR(req))
731 		return ERR_CAST(req);
732 	req->r_path1 = kstrdup(path, GFP_NOFS);
733 	req->r_ino1.ino = CEPH_INO_ROOT;
734 	req->r_ino1.snap = CEPH_NOSNAP;
735 	req->r_started = started;
736 	req->r_timeout = fsc->client->options->mount_timeout * HZ;
737 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
738 	req->r_num_caps = 2;
739 	err = ceph_mdsc_do_request(mdsc, NULL, req);
740 	if (err == 0) {
741 		struct inode *inode = req->r_target_inode;
742 		req->r_target_inode = NULL;
743 		dout("open_root_inode success\n");
744 		if (ceph_ino(inode) == CEPH_INO_ROOT &&
745 		    fsc->sb->s_root == NULL) {
746 			root = d_make_root(inode);
747 			if (!root) {
748 				root = ERR_PTR(-ENOMEM);
749 				goto out;
750 			}
751 		} else {
752 			root = d_obtain_root(inode);
753 		}
754 		ceph_init_dentry(root);
755 		dout("open_root_inode success, root dentry is %p\n", root);
756 	} else {
757 		root = ERR_PTR(err);
758 	}
759 out:
760 	ceph_mdsc_put_request(req);
761 	return root;
762 }
763 
764 
765 
766 
767 /*
768  * mount: join the ceph cluster, and open root directory.
769  */
770 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
771 		      const char *path)
772 {
773 	int err;
774 	unsigned long started = jiffies;  /* note the start time */
775 	struct dentry *root;
776 	int first = 0;   /* first vfsmount for this super_block */
777 
778 	dout("mount start\n");
779 	mutex_lock(&fsc->client->mount_mutex);
780 
781 	err = __ceph_open_session(fsc->client, started);
782 	if (err < 0)
783 		goto out;
784 
785 	dout("mount opening root\n");
786 	root = open_root_dentry(fsc, "", started);
787 	if (IS_ERR(root)) {
788 		err = PTR_ERR(root);
789 		goto out;
790 	}
791 	if (fsc->sb->s_root) {
792 		dput(root);
793 	} else {
794 		fsc->sb->s_root = root;
795 		first = 1;
796 
797 		err = ceph_fs_debugfs_init(fsc);
798 		if (err < 0)
799 			goto fail;
800 	}
801 
802 	if (path[0] == 0) {
803 		dget(root);
804 	} else {
805 		dout("mount opening base mountpoint\n");
806 		root = open_root_dentry(fsc, path, started);
807 		if (IS_ERR(root)) {
808 			err = PTR_ERR(root);
809 			goto fail;
810 		}
811 	}
812 
813 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
814 	dout("mount success\n");
815 	mutex_unlock(&fsc->client->mount_mutex);
816 	return root;
817 
818 out:
819 	mutex_unlock(&fsc->client->mount_mutex);
820 	return ERR_PTR(err);
821 
822 fail:
823 	if (first) {
824 		dput(fsc->sb->s_root);
825 		fsc->sb->s_root = NULL;
826 	}
827 	goto out;
828 }
829 
830 static int ceph_set_super(struct super_block *s, void *data)
831 {
832 	struct ceph_fs_client *fsc = data;
833 	int ret;
834 
835 	dout("set_super %p data %p\n", s, data);
836 
837 	s->s_flags = fsc->mount_options->sb_flags;
838 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
839 
840 	s->s_xattr = ceph_xattr_handlers;
841 	s->s_fs_info = fsc;
842 	fsc->sb = s;
843 
844 	s->s_op = &ceph_super_ops;
845 	s->s_export_op = &ceph_export_ops;
846 
847 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
848 
849 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
850 	if (ret != 0)
851 		goto fail;
852 
853 	return ret;
854 
855 fail:
856 	s->s_fs_info = NULL;
857 	fsc->sb = NULL;
858 	return ret;
859 }
860 
861 /*
862  * share superblock if same fs AND options
863  */
864 static int ceph_compare_super(struct super_block *sb, void *data)
865 {
866 	struct ceph_fs_client *new = data;
867 	struct ceph_mount_options *fsopt = new->mount_options;
868 	struct ceph_options *opt = new->client->options;
869 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
870 
871 	dout("ceph_compare_super %p\n", sb);
872 
873 	if (compare_mount_options(fsopt, opt, other)) {
874 		dout("monitor(s)/mount options don't match\n");
875 		return 0;
876 	}
877 	if ((opt->flags & CEPH_OPT_FSID) &&
878 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
879 		dout("fsid doesn't match\n");
880 		return 0;
881 	}
882 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
883 		dout("flags differ\n");
884 		return 0;
885 	}
886 	return 1;
887 }
888 
889 /*
890  * construct our own bdi so we can control readahead, etc.
891  */
892 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
893 
894 static int ceph_register_bdi(struct super_block *sb,
895 			     struct ceph_fs_client *fsc)
896 {
897 	int err;
898 
899 	/* set ra_pages based on rasize mount option? */
900 	if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
901 		fsc->backing_dev_info.ra_pages =
902 			(fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
903 			>> PAGE_SHIFT;
904 	else
905 		fsc->backing_dev_info.ra_pages =
906 			VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
907 
908 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
909 			   atomic_long_inc_return(&bdi_seq));
910 	if (!err)
911 		sb->s_bdi = &fsc->backing_dev_info;
912 	return err;
913 }
914 
915 static struct dentry *ceph_mount(struct file_system_type *fs_type,
916 		       int flags, const char *dev_name, void *data)
917 {
918 	struct super_block *sb;
919 	struct ceph_fs_client *fsc;
920 	struct dentry *res;
921 	int err;
922 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
923 	const char *path = NULL;
924 	struct ceph_mount_options *fsopt = NULL;
925 	struct ceph_options *opt = NULL;
926 
927 	dout("ceph_mount\n");
928 
929 #ifdef CONFIG_CEPH_FS_POSIX_ACL
930 	flags |= MS_POSIXACL;
931 #endif
932 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
933 	if (err < 0) {
934 		res = ERR_PTR(err);
935 		goto out_final;
936 	}
937 
938 	/* create client (which we may/may not use) */
939 	fsc = create_fs_client(fsopt, opt);
940 	if (IS_ERR(fsc)) {
941 		res = ERR_CAST(fsc);
942 		destroy_mount_options(fsopt);
943 		ceph_destroy_options(opt);
944 		goto out_final;
945 	}
946 
947 	err = ceph_mdsc_init(fsc);
948 	if (err < 0) {
949 		res = ERR_PTR(err);
950 		goto out;
951 	}
952 
953 	if (ceph_test_opt(fsc->client, NOSHARE))
954 		compare_super = NULL;
955 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
956 	if (IS_ERR(sb)) {
957 		res = ERR_CAST(sb);
958 		goto out;
959 	}
960 
961 	if (ceph_sb_to_client(sb) != fsc) {
962 		ceph_mdsc_destroy(fsc);
963 		destroy_fs_client(fsc);
964 		fsc = ceph_sb_to_client(sb);
965 		dout("get_sb got existing client %p\n", fsc);
966 	} else {
967 		dout("get_sb using new client %p\n", fsc);
968 		err = ceph_register_bdi(sb, fsc);
969 		if (err < 0) {
970 			res = ERR_PTR(err);
971 			goto out_splat;
972 		}
973 	}
974 
975 	res = ceph_real_mount(fsc, path);
976 	if (IS_ERR(res))
977 		goto out_splat;
978 	dout("root %p inode %p ino %llx.%llx\n", res,
979 	     res->d_inode, ceph_vinop(res->d_inode));
980 	return res;
981 
982 out_splat:
983 	ceph_mdsc_close_sessions(fsc->mdsc);
984 	deactivate_locked_super(sb);
985 	goto out_final;
986 
987 out:
988 	ceph_mdsc_destroy(fsc);
989 	destroy_fs_client(fsc);
990 out_final:
991 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
992 	return res;
993 }
994 
995 static void ceph_kill_sb(struct super_block *s)
996 {
997 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
998 	dev_t dev = s->s_dev;
999 
1000 	dout("kill_sb %p\n", s);
1001 
1002 	ceph_mdsc_pre_umount(fsc->mdsc);
1003 	generic_shutdown_super(s);
1004 	ceph_mdsc_destroy(fsc);
1005 
1006 	destroy_fs_client(fsc);
1007 	free_anon_bdev(dev);
1008 }
1009 
1010 static struct file_system_type ceph_fs_type = {
1011 	.owner		= THIS_MODULE,
1012 	.name		= "ceph",
1013 	.mount		= ceph_mount,
1014 	.kill_sb	= ceph_kill_sb,
1015 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1016 };
1017 MODULE_ALIAS_FS("ceph");
1018 
1019 static int __init init_ceph(void)
1020 {
1021 	int ret = init_caches();
1022 	if (ret)
1023 		goto out;
1024 
1025 	ceph_flock_init();
1026 	ceph_xattr_init();
1027 	ret = ceph_snap_init();
1028 	if (ret)
1029 		goto out_xattr;
1030 	ret = register_filesystem(&ceph_fs_type);
1031 	if (ret)
1032 		goto out_snap;
1033 
1034 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1035 
1036 	return 0;
1037 
1038 out_snap:
1039 	ceph_snap_exit();
1040 out_xattr:
1041 	ceph_xattr_exit();
1042 	destroy_caches();
1043 out:
1044 	return ret;
1045 }
1046 
1047 static void __exit exit_ceph(void)
1048 {
1049 	dout("exit_ceph\n");
1050 	unregister_filesystem(&ceph_fs_type);
1051 	ceph_snap_exit();
1052 	ceph_xattr_exit();
1053 	destroy_caches();
1054 }
1055 
1056 module_init(init_ceph);
1057 module_exit(exit_ceph);
1058 
1059 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1060 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1061 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1062 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1063 MODULE_LICENSE("GPL");
1064