xref: /openbmc/linux/fs/ceph/super.c (revision 8684014d)
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 	 * ensure we release the bdi before put_anon_super releases
46 	 * the device name.
47 	 */
48 	if (s->s_bdi == &fsc->backing_dev_info) {
49 		bdi_unregister(&fsc->backing_dev_info);
50 		s->s_bdi = NULL;
51 	}
52 
53 	return;
54 }
55 
56 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
57 {
58 	struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
59 	struct ceph_monmap *monmap = fsc->client->monc.monmap;
60 	struct ceph_statfs st;
61 	u64 fsid;
62 	int err;
63 
64 	dout("statfs\n");
65 	err = ceph_monc_do_statfs(&fsc->client->monc, &st);
66 	if (err < 0)
67 		return err;
68 
69 	/* fill in kstatfs */
70 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
71 
72 	/*
73 	 * express utilization in terms of large blocks to avoid
74 	 * overflow on 32-bit machines.
75 	 *
76 	 * NOTE: for the time being, we make bsize == frsize to humor
77 	 * not-yet-ancient versions of glibc that are broken.
78 	 * Someday, we will probably want to report a real block
79 	 * size...  whatever that may mean for a network file system!
80 	 */
81 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
82 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
83 	buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
84 	buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
85 	buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
86 
87 	buf->f_files = le64_to_cpu(st.num_objects);
88 	buf->f_ffree = -1;
89 	buf->f_namelen = NAME_MAX;
90 
91 	/* leave fsid little-endian, regardless of host endianness */
92 	fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
93 	buf->f_fsid.val[0] = fsid & 0xffffffff;
94 	buf->f_fsid.val[1] = fsid >> 32;
95 
96 	return 0;
97 }
98 
99 
100 static int ceph_sync_fs(struct super_block *sb, int wait)
101 {
102 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
103 
104 	if (!wait) {
105 		dout("sync_fs (non-blocking)\n");
106 		ceph_flush_dirty_caps(fsc->mdsc);
107 		dout("sync_fs (non-blocking) done\n");
108 		return 0;
109 	}
110 
111 	dout("sync_fs (blocking)\n");
112 	ceph_osdc_sync(&fsc->client->osdc);
113 	ceph_mdsc_sync(fsc->mdsc);
114 	dout("sync_fs (blocking) done\n");
115 	return 0;
116 }
117 
118 /*
119  * mount options
120  */
121 enum {
122 	Opt_wsize,
123 	Opt_rsize,
124 	Opt_rasize,
125 	Opt_caps_wanted_delay_min,
126 	Opt_caps_wanted_delay_max,
127 	Opt_cap_release_safety,
128 	Opt_readdir_max_entries,
129 	Opt_readdir_max_bytes,
130 	Opt_congestion_kb,
131 	Opt_last_int,
132 	/* int args above */
133 	Opt_snapdirname,
134 	Opt_last_string,
135 	/* string args above */
136 	Opt_dirstat,
137 	Opt_nodirstat,
138 	Opt_rbytes,
139 	Opt_norbytes,
140 	Opt_asyncreaddir,
141 	Opt_noasyncreaddir,
142 	Opt_dcache,
143 	Opt_nodcache,
144 	Opt_ino32,
145 	Opt_noino32,
146 	Opt_fscache,
147 	Opt_nofscache,
148 #ifdef CONFIG_CEPH_FS_POSIX_ACL
149 	Opt_acl,
150 #endif
151 	Opt_noacl
152 };
153 
154 static match_table_t fsopt_tokens = {
155 	{Opt_wsize, "wsize=%d"},
156 	{Opt_rsize, "rsize=%d"},
157 	{Opt_rasize, "rasize=%d"},
158 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
159 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
160 	{Opt_cap_release_safety, "cap_release_safety=%d"},
161 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
162 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
163 	{Opt_congestion_kb, "write_congestion_kb=%d"},
164 	/* int args above */
165 	{Opt_snapdirname, "snapdirname=%s"},
166 	/* string args above */
167 	{Opt_dirstat, "dirstat"},
168 	{Opt_nodirstat, "nodirstat"},
169 	{Opt_rbytes, "rbytes"},
170 	{Opt_norbytes, "norbytes"},
171 	{Opt_asyncreaddir, "asyncreaddir"},
172 	{Opt_noasyncreaddir, "noasyncreaddir"},
173 	{Opt_dcache, "dcache"},
174 	{Opt_nodcache, "nodcache"},
175 	{Opt_ino32, "ino32"},
176 	{Opt_noino32, "noino32"},
177 	{Opt_fscache, "fsc"},
178 	{Opt_nofscache, "nofsc"},
179 #ifdef CONFIG_CEPH_FS_POSIX_ACL
180 	{Opt_acl, "acl"},
181 #endif
182 	{Opt_noacl, "noacl"},
183 	{-1, NULL}
184 };
185 
186 static int parse_fsopt_token(char *c, void *private)
187 {
188 	struct ceph_mount_options *fsopt = private;
189 	substring_t argstr[MAX_OPT_ARGS];
190 	int token, intval, ret;
191 
192 	token = match_token((char *)c, fsopt_tokens, argstr);
193 	if (token < 0)
194 		return -EINVAL;
195 
196 	if (token < Opt_last_int) {
197 		ret = match_int(&argstr[0], &intval);
198 		if (ret < 0) {
199 			pr_err("bad mount option arg (not int) "
200 			       "at '%s'\n", c);
201 			return ret;
202 		}
203 		dout("got int token %d val %d\n", token, intval);
204 	} else if (token > Opt_last_int && token < Opt_last_string) {
205 		dout("got string token %d val %s\n", token,
206 		     argstr[0].from);
207 	} else {
208 		dout("got token %d\n", token);
209 	}
210 
211 	switch (token) {
212 	case Opt_snapdirname:
213 		kfree(fsopt->snapdir_name);
214 		fsopt->snapdir_name = kstrndup(argstr[0].from,
215 					       argstr[0].to-argstr[0].from,
216 					       GFP_KERNEL);
217 		if (!fsopt->snapdir_name)
218 			return -ENOMEM;
219 		break;
220 
221 		/* misc */
222 	case Opt_wsize:
223 		fsopt->wsize = intval;
224 		break;
225 	case Opt_rsize:
226 		fsopt->rsize = intval;
227 		break;
228 	case Opt_rasize:
229 		fsopt->rasize = intval;
230 		break;
231 	case Opt_caps_wanted_delay_min:
232 		fsopt->caps_wanted_delay_min = intval;
233 		break;
234 	case Opt_caps_wanted_delay_max:
235 		fsopt->caps_wanted_delay_max = intval;
236 		break;
237 	case Opt_readdir_max_entries:
238 		fsopt->max_readdir = intval;
239 		break;
240 	case Opt_readdir_max_bytes:
241 		fsopt->max_readdir_bytes = intval;
242 		break;
243 	case Opt_congestion_kb:
244 		fsopt->congestion_kb = intval;
245 		break;
246 	case Opt_dirstat:
247 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
248 		break;
249 	case Opt_nodirstat:
250 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
251 		break;
252 	case Opt_rbytes:
253 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
254 		break;
255 	case Opt_norbytes:
256 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
257 		break;
258 	case Opt_asyncreaddir:
259 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
260 		break;
261 	case Opt_noasyncreaddir:
262 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
263 		break;
264 	case Opt_dcache:
265 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
266 		break;
267 	case Opt_nodcache:
268 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
269 		break;
270 	case Opt_ino32:
271 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
272 		break;
273 	case Opt_noino32:
274 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
275 		break;
276 	case Opt_fscache:
277 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
278 		break;
279 	case Opt_nofscache:
280 		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
281 		break;
282 #ifdef CONFIG_CEPH_FS_POSIX_ACL
283 	case Opt_acl:
284 		fsopt->sb_flags |= MS_POSIXACL;
285 		break;
286 #endif
287 	case Opt_noacl:
288 		fsopt->sb_flags &= ~MS_POSIXACL;
289 		break;
290 	default:
291 		BUG_ON(token);
292 	}
293 	return 0;
294 }
295 
296 static void destroy_mount_options(struct ceph_mount_options *args)
297 {
298 	dout("destroy_mount_options %p\n", args);
299 	kfree(args->snapdir_name);
300 	kfree(args);
301 }
302 
303 static int strcmp_null(const char *s1, const char *s2)
304 {
305 	if (!s1 && !s2)
306 		return 0;
307 	if (s1 && !s2)
308 		return -1;
309 	if (!s1 && s2)
310 		return 1;
311 	return strcmp(s1, s2);
312 }
313 
314 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
315 				 struct ceph_options *new_opt,
316 				 struct ceph_fs_client *fsc)
317 {
318 	struct ceph_mount_options *fsopt1 = new_fsopt;
319 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
320 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
321 	int ret;
322 
323 	ret = memcmp(fsopt1, fsopt2, ofs);
324 	if (ret)
325 		return ret;
326 
327 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
328 	if (ret)
329 		return ret;
330 
331 	return ceph_compare_options(new_opt, fsc->client);
332 }
333 
334 static int parse_mount_options(struct ceph_mount_options **pfsopt,
335 			       struct ceph_options **popt,
336 			       int flags, char *options,
337 			       const char *dev_name,
338 			       const char **path)
339 {
340 	struct ceph_mount_options *fsopt;
341 	const char *dev_name_end;
342 	int err;
343 
344 	if (!dev_name || !*dev_name)
345 		return -EINVAL;
346 
347 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
348 	if (!fsopt)
349 		return -ENOMEM;
350 
351 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
352 
353 	fsopt->sb_flags = flags;
354 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
355 
356 	fsopt->rsize = CEPH_RSIZE_DEFAULT;
357 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
358 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
359 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
360 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
361 	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
362 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
363 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
364 	fsopt->congestion_kb = default_congestion_kb();
365 
366 	/*
367 	 * Distinguish the server list from the path in "dev_name".
368 	 * Internally we do not include the leading '/' in the path.
369 	 *
370 	 * "dev_name" will look like:
371 	 *     <server_spec>[,<server_spec>...]:[<path>]
372 	 * where
373 	 *     <server_spec> is <ip>[:<port>]
374 	 *     <path> is optional, but if present must begin with '/'
375 	 */
376 	dev_name_end = strchr(dev_name, '/');
377 	if (dev_name_end) {
378 		/* skip over leading '/' for path */
379 		*path = dev_name_end + 1;
380 	} else {
381 		/* path is empty */
382 		dev_name_end = dev_name + strlen(dev_name);
383 		*path = dev_name_end;
384 	}
385 	err = -EINVAL;
386 	dev_name_end--;		/* back up to ':' separator */
387 	if (dev_name_end < dev_name || *dev_name_end != ':') {
388 		pr_err("device name is missing path (no : separator in %s)\n",
389 				dev_name);
390 		goto out;
391 	}
392 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
393 	dout("server path '%s'\n", *path);
394 
395 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
396 				 parse_fsopt_token, (void *)fsopt);
397 	if (IS_ERR(*popt)) {
398 		err = PTR_ERR(*popt);
399 		goto out;
400 	}
401 
402 	/* success */
403 	*pfsopt = fsopt;
404 	return 0;
405 
406 out:
407 	destroy_mount_options(fsopt);
408 	return err;
409 }
410 
411 /**
412  * ceph_show_options - Show mount options in /proc/mounts
413  * @m: seq_file to write to
414  * @root: root of that (sub)tree
415  */
416 static int ceph_show_options(struct seq_file *m, struct dentry *root)
417 {
418 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
419 	struct ceph_mount_options *fsopt = fsc->mount_options;
420 	struct ceph_options *opt = fsc->client->options;
421 
422 	if (opt->flags & CEPH_OPT_FSID)
423 		seq_printf(m, ",fsid=%pU", &opt->fsid);
424 	if (opt->flags & CEPH_OPT_NOSHARE)
425 		seq_puts(m, ",noshare");
426 	if (opt->flags & CEPH_OPT_NOCRC)
427 		seq_puts(m, ",nocrc");
428 
429 	if (opt->name)
430 		seq_printf(m, ",name=%s", opt->name);
431 	if (opt->key)
432 		seq_puts(m, ",secret=<hidden>");
433 
434 	if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
435 		seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
436 	if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
437 		seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
438 	if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
439 		seq_printf(m, ",osdkeepalivetimeout=%d",
440 			   opt->osd_keepalive_timeout);
441 
442 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
443 		seq_puts(m, ",dirstat");
444 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
445 		seq_puts(m, ",norbytes");
446 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
447 		seq_puts(m, ",noasyncreaddir");
448 	if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
449 		seq_puts(m, ",dcache");
450 	else
451 		seq_puts(m, ",nodcache");
452 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
453 		seq_puts(m, ",fsc");
454 	else
455 		seq_puts(m, ",nofsc");
456 
457 #ifdef CONFIG_CEPH_FS_POSIX_ACL
458 	if (fsopt->sb_flags & MS_POSIXACL)
459 		seq_puts(m, ",acl");
460 	else
461 		seq_puts(m, ",noacl");
462 #endif
463 
464 	if (fsopt->wsize)
465 		seq_printf(m, ",wsize=%d", fsopt->wsize);
466 	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
467 		seq_printf(m, ",rsize=%d", fsopt->rsize);
468 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
469 		seq_printf(m, ",rasize=%d", fsopt->rasize);
470 	if (fsopt->congestion_kb != default_congestion_kb())
471 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
472 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
473 		seq_printf(m, ",caps_wanted_delay_min=%d",
474 			 fsopt->caps_wanted_delay_min);
475 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
476 		seq_printf(m, ",caps_wanted_delay_max=%d",
477 			   fsopt->caps_wanted_delay_max);
478 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
479 		seq_printf(m, ",cap_release_safety=%d",
480 			   fsopt->cap_release_safety);
481 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
482 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
483 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
484 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
485 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
486 		seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
487 	return 0;
488 }
489 
490 /*
491  * handle any mon messages the standard library doesn't understand.
492  * return error if we don't either.
493  */
494 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
495 {
496 	struct ceph_fs_client *fsc = client->private;
497 	int type = le16_to_cpu(msg->hdr.type);
498 
499 	switch (type) {
500 	case CEPH_MSG_MDS_MAP:
501 		ceph_mdsc_handle_map(fsc->mdsc, msg);
502 		return 0;
503 
504 	default:
505 		return -1;
506 	}
507 }
508 
509 /*
510  * create a new fs client
511  */
512 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
513 					struct ceph_options *opt)
514 {
515 	struct ceph_fs_client *fsc;
516 	const u64 supported_features =
517 		CEPH_FEATURE_FLOCK |
518 		CEPH_FEATURE_DIRLAYOUTHASH |
519 		CEPH_FEATURE_MDS_INLINE_DATA;
520 	const u64 required_features = 0;
521 	int page_count;
522 	size_t size;
523 	int err = -ENOMEM;
524 
525 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
526 	if (!fsc)
527 		return ERR_PTR(-ENOMEM);
528 
529 	fsc->client = ceph_create_client(opt, fsc, supported_features,
530 					 required_features);
531 	if (IS_ERR(fsc->client)) {
532 		err = PTR_ERR(fsc->client);
533 		goto fail;
534 	}
535 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
536 	fsc->client->monc.want_mdsmap = 1;
537 
538 	fsc->mount_options = fsopt;
539 
540 	fsc->sb = NULL;
541 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
542 
543 	atomic_long_set(&fsc->writeback_count, 0);
544 
545 	err = bdi_init(&fsc->backing_dev_info);
546 	if (err < 0)
547 		goto fail_client;
548 
549 	err = -ENOMEM;
550 	/*
551 	 * The number of concurrent works can be high but they don't need
552 	 * to be processed in parallel, limit concurrency.
553 	 */
554 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
555 	if (fsc->wb_wq == NULL)
556 		goto fail_bdi;
557 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
558 	if (fsc->pg_inv_wq == NULL)
559 		goto fail_wb_wq;
560 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
561 	if (fsc->trunc_wq == NULL)
562 		goto fail_pg_inv_wq;
563 
564 	/* set up mempools */
565 	err = -ENOMEM;
566 	page_count = fsc->mount_options->wsize >> PAGE_CACHE_SHIFT;
567 	size = sizeof (struct page *) * (page_count ? page_count : 1);
568 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
569 	if (!fsc->wb_pagevec_pool)
570 		goto fail_trunc_wq;
571 
572 	/* setup fscache */
573 	if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
574 	    (ceph_fscache_register_fs(fsc) != 0))
575 		goto fail_fscache;
576 
577 	/* caps */
578 	fsc->min_caps = fsopt->max_readdir;
579 
580 	return fsc;
581 
582 fail_fscache:
583 	ceph_fscache_unregister_fs(fsc);
584 fail_trunc_wq:
585 	destroy_workqueue(fsc->trunc_wq);
586 fail_pg_inv_wq:
587 	destroy_workqueue(fsc->pg_inv_wq);
588 fail_wb_wq:
589 	destroy_workqueue(fsc->wb_wq);
590 fail_bdi:
591 	bdi_destroy(&fsc->backing_dev_info);
592 fail_client:
593 	ceph_destroy_client(fsc->client);
594 fail:
595 	kfree(fsc);
596 	return ERR_PTR(err);
597 }
598 
599 static void destroy_fs_client(struct ceph_fs_client *fsc)
600 {
601 	dout("destroy_fs_client %p\n", fsc);
602 
603 	ceph_fscache_unregister_fs(fsc);
604 
605 	destroy_workqueue(fsc->wb_wq);
606 	destroy_workqueue(fsc->pg_inv_wq);
607 	destroy_workqueue(fsc->trunc_wq);
608 
609 	bdi_destroy(&fsc->backing_dev_info);
610 
611 	mempool_destroy(fsc->wb_pagevec_pool);
612 
613 	destroy_mount_options(fsc->mount_options);
614 
615 	ceph_fs_debugfs_cleanup(fsc);
616 
617 	ceph_destroy_client(fsc->client);
618 
619 	kfree(fsc);
620 	dout("destroy_fs_client %p done\n", fsc);
621 }
622 
623 /*
624  * caches
625  */
626 struct kmem_cache *ceph_inode_cachep;
627 struct kmem_cache *ceph_cap_cachep;
628 struct kmem_cache *ceph_dentry_cachep;
629 struct kmem_cache *ceph_file_cachep;
630 
631 static void ceph_inode_init_once(void *foo)
632 {
633 	struct ceph_inode_info *ci = foo;
634 	inode_init_once(&ci->vfs_inode);
635 }
636 
637 static int __init init_caches(void)
638 {
639 	int error = -ENOMEM;
640 
641 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
642 				      sizeof(struct ceph_inode_info),
643 				      __alignof__(struct ceph_inode_info),
644 				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
645 				      ceph_inode_init_once);
646 	if (ceph_inode_cachep == NULL)
647 		return -ENOMEM;
648 
649 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
650 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
651 	if (ceph_cap_cachep == NULL)
652 		goto bad_cap;
653 
654 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
655 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
656 	if (ceph_dentry_cachep == NULL)
657 		goto bad_dentry;
658 
659 	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
660 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
661 	if (ceph_file_cachep == NULL)
662 		goto bad_file;
663 
664 	if ((error = ceph_fscache_register()))
665 		goto bad_file;
666 
667 	return 0;
668 bad_file:
669 	kmem_cache_destroy(ceph_dentry_cachep);
670 bad_dentry:
671 	kmem_cache_destroy(ceph_cap_cachep);
672 bad_cap:
673 	kmem_cache_destroy(ceph_inode_cachep);
674 	return error;
675 }
676 
677 static void destroy_caches(void)
678 {
679 	/*
680 	 * Make sure all delayed rcu free inodes are flushed before we
681 	 * destroy cache.
682 	 */
683 	rcu_barrier();
684 
685 	kmem_cache_destroy(ceph_inode_cachep);
686 	kmem_cache_destroy(ceph_cap_cachep);
687 	kmem_cache_destroy(ceph_dentry_cachep);
688 	kmem_cache_destroy(ceph_file_cachep);
689 
690 	ceph_fscache_unregister();
691 }
692 
693 
694 /*
695  * ceph_umount_begin - initiate forced umount.  Tear down down the
696  * mount, skipping steps that may hang while waiting for server(s).
697  */
698 static void ceph_umount_begin(struct super_block *sb)
699 {
700 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
701 
702 	dout("ceph_umount_begin - starting forced umount\n");
703 	if (!fsc)
704 		return;
705 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
706 	return;
707 }
708 
709 static const struct super_operations ceph_super_ops = {
710 	.alloc_inode	= ceph_alloc_inode,
711 	.destroy_inode	= ceph_destroy_inode,
712 	.write_inode    = ceph_write_inode,
713 	.drop_inode	= ceph_drop_inode,
714 	.sync_fs        = ceph_sync_fs,
715 	.put_super	= ceph_put_super,
716 	.show_options   = ceph_show_options,
717 	.statfs		= ceph_statfs,
718 	.umount_begin   = ceph_umount_begin,
719 };
720 
721 /*
722  * Bootstrap mount by opening the root directory.  Note the mount
723  * @started time from caller, and time out if this takes too long.
724  */
725 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
726 				       const char *path,
727 				       unsigned long started)
728 {
729 	struct ceph_mds_client *mdsc = fsc->mdsc;
730 	struct ceph_mds_request *req = NULL;
731 	int err;
732 	struct dentry *root;
733 
734 	/* open dir */
735 	dout("open_root_inode opening '%s'\n", path);
736 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
737 	if (IS_ERR(req))
738 		return ERR_CAST(req);
739 	req->r_path1 = kstrdup(path, GFP_NOFS);
740 	req->r_ino1.ino = CEPH_INO_ROOT;
741 	req->r_ino1.snap = CEPH_NOSNAP;
742 	req->r_started = started;
743 	req->r_timeout = fsc->client->options->mount_timeout * HZ;
744 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
745 	req->r_num_caps = 2;
746 	err = ceph_mdsc_do_request(mdsc, NULL, req);
747 	if (err == 0) {
748 		struct inode *inode = req->r_target_inode;
749 		req->r_target_inode = NULL;
750 		dout("open_root_inode success\n");
751 		if (ceph_ino(inode) == CEPH_INO_ROOT &&
752 		    fsc->sb->s_root == NULL) {
753 			root = d_make_root(inode);
754 			if (!root) {
755 				root = ERR_PTR(-ENOMEM);
756 				goto out;
757 			}
758 		} else {
759 			root = d_obtain_root(inode);
760 		}
761 		ceph_init_dentry(root);
762 		dout("open_root_inode success, root dentry is %p\n", root);
763 	} else {
764 		root = ERR_PTR(err);
765 	}
766 out:
767 	ceph_mdsc_put_request(req);
768 	return root;
769 }
770 
771 
772 
773 
774 /*
775  * mount: join the ceph cluster, and open root directory.
776  */
777 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
778 		      const char *path)
779 {
780 	int err;
781 	unsigned long started = jiffies;  /* note the start time */
782 	struct dentry *root;
783 	int first = 0;   /* first vfsmount for this super_block */
784 
785 	dout("mount start\n");
786 	mutex_lock(&fsc->client->mount_mutex);
787 
788 	err = __ceph_open_session(fsc->client, started);
789 	if (err < 0)
790 		goto out;
791 
792 	dout("mount opening root\n");
793 	root = open_root_dentry(fsc, "", started);
794 	if (IS_ERR(root)) {
795 		err = PTR_ERR(root);
796 		goto out;
797 	}
798 	if (fsc->sb->s_root) {
799 		dput(root);
800 	} else {
801 		fsc->sb->s_root = root;
802 		first = 1;
803 
804 		err = ceph_fs_debugfs_init(fsc);
805 		if (err < 0)
806 			goto fail;
807 	}
808 
809 	if (path[0] == 0) {
810 		dget(root);
811 	} else {
812 		dout("mount opening base mountpoint\n");
813 		root = open_root_dentry(fsc, path, started);
814 		if (IS_ERR(root)) {
815 			err = PTR_ERR(root);
816 			goto fail;
817 		}
818 	}
819 
820 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
821 	dout("mount success\n");
822 	mutex_unlock(&fsc->client->mount_mutex);
823 	return root;
824 
825 out:
826 	mutex_unlock(&fsc->client->mount_mutex);
827 	return ERR_PTR(err);
828 
829 fail:
830 	if (first) {
831 		dput(fsc->sb->s_root);
832 		fsc->sb->s_root = NULL;
833 	}
834 	goto out;
835 }
836 
837 static int ceph_set_super(struct super_block *s, void *data)
838 {
839 	struct ceph_fs_client *fsc = data;
840 	int ret;
841 
842 	dout("set_super %p data %p\n", s, data);
843 
844 	s->s_flags = fsc->mount_options->sb_flags;
845 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
846 
847 	s->s_xattr = ceph_xattr_handlers;
848 	s->s_fs_info = fsc;
849 	fsc->sb = s;
850 
851 	s->s_op = &ceph_super_ops;
852 	s->s_export_op = &ceph_export_ops;
853 
854 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
855 
856 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
857 	if (ret != 0)
858 		goto fail;
859 
860 	return ret;
861 
862 fail:
863 	s->s_fs_info = NULL;
864 	fsc->sb = NULL;
865 	return ret;
866 }
867 
868 /*
869  * share superblock if same fs AND options
870  */
871 static int ceph_compare_super(struct super_block *sb, void *data)
872 {
873 	struct ceph_fs_client *new = data;
874 	struct ceph_mount_options *fsopt = new->mount_options;
875 	struct ceph_options *opt = new->client->options;
876 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
877 
878 	dout("ceph_compare_super %p\n", sb);
879 
880 	if (compare_mount_options(fsopt, opt, other)) {
881 		dout("monitor(s)/mount options don't match\n");
882 		return 0;
883 	}
884 	if ((opt->flags & CEPH_OPT_FSID) &&
885 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
886 		dout("fsid doesn't match\n");
887 		return 0;
888 	}
889 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
890 		dout("flags differ\n");
891 		return 0;
892 	}
893 	return 1;
894 }
895 
896 /*
897  * construct our own bdi so we can control readahead, etc.
898  */
899 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
900 
901 static int ceph_register_bdi(struct super_block *sb,
902 			     struct ceph_fs_client *fsc)
903 {
904 	int err;
905 
906 	/* set ra_pages based on rasize mount option? */
907 	if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
908 		fsc->backing_dev_info.ra_pages =
909 			(fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
910 			>> PAGE_SHIFT;
911 	else
912 		fsc->backing_dev_info.ra_pages =
913 			default_backing_dev_info.ra_pages;
914 
915 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
916 			   atomic_long_inc_return(&bdi_seq));
917 	if (!err)
918 		sb->s_bdi = &fsc->backing_dev_info;
919 	return err;
920 }
921 
922 static struct dentry *ceph_mount(struct file_system_type *fs_type,
923 		       int flags, const char *dev_name, void *data)
924 {
925 	struct super_block *sb;
926 	struct ceph_fs_client *fsc;
927 	struct dentry *res;
928 	int err;
929 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
930 	const char *path = NULL;
931 	struct ceph_mount_options *fsopt = NULL;
932 	struct ceph_options *opt = NULL;
933 
934 	dout("ceph_mount\n");
935 
936 #ifdef CONFIG_CEPH_FS_POSIX_ACL
937 	flags |= MS_POSIXACL;
938 #endif
939 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
940 	if (err < 0) {
941 		res = ERR_PTR(err);
942 		goto out_final;
943 	}
944 
945 	/* create client (which we may/may not use) */
946 	fsc = create_fs_client(fsopt, opt);
947 	if (IS_ERR(fsc)) {
948 		res = ERR_CAST(fsc);
949 		destroy_mount_options(fsopt);
950 		ceph_destroy_options(opt);
951 		goto out_final;
952 	}
953 
954 	err = ceph_mdsc_init(fsc);
955 	if (err < 0) {
956 		res = ERR_PTR(err);
957 		goto out;
958 	}
959 
960 	if (ceph_test_opt(fsc->client, NOSHARE))
961 		compare_super = NULL;
962 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
963 	if (IS_ERR(sb)) {
964 		res = ERR_CAST(sb);
965 		goto out;
966 	}
967 
968 	if (ceph_sb_to_client(sb) != fsc) {
969 		ceph_mdsc_destroy(fsc);
970 		destroy_fs_client(fsc);
971 		fsc = ceph_sb_to_client(sb);
972 		dout("get_sb got existing client %p\n", fsc);
973 	} else {
974 		dout("get_sb using new client %p\n", fsc);
975 		err = ceph_register_bdi(sb, fsc);
976 		if (err < 0) {
977 			res = ERR_PTR(err);
978 			goto out_splat;
979 		}
980 	}
981 
982 	res = ceph_real_mount(fsc, path);
983 	if (IS_ERR(res))
984 		goto out_splat;
985 	dout("root %p inode %p ino %llx.%llx\n", res,
986 	     res->d_inode, ceph_vinop(res->d_inode));
987 	return res;
988 
989 out_splat:
990 	ceph_mdsc_close_sessions(fsc->mdsc);
991 	deactivate_locked_super(sb);
992 	goto out_final;
993 
994 out:
995 	ceph_mdsc_destroy(fsc);
996 	destroy_fs_client(fsc);
997 out_final:
998 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
999 	return res;
1000 }
1001 
1002 static void ceph_kill_sb(struct super_block *s)
1003 {
1004 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1005 	dout("kill_sb %p\n", s);
1006 	ceph_mdsc_pre_umount(fsc->mdsc);
1007 	kill_anon_super(s);    /* will call put_super after sb is r/o */
1008 	ceph_mdsc_destroy(fsc);
1009 	destroy_fs_client(fsc);
1010 }
1011 
1012 static struct file_system_type ceph_fs_type = {
1013 	.owner		= THIS_MODULE,
1014 	.name		= "ceph",
1015 	.mount		= ceph_mount,
1016 	.kill_sb	= ceph_kill_sb,
1017 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1018 };
1019 MODULE_ALIAS_FS("ceph");
1020 
1021 static int __init init_ceph(void)
1022 {
1023 	int ret = init_caches();
1024 	if (ret)
1025 		goto out;
1026 
1027 	ceph_flock_init();
1028 	ceph_xattr_init();
1029 	ret = ceph_snap_init();
1030 	if (ret)
1031 		goto out_xattr;
1032 	ret = register_filesystem(&ceph_fs_type);
1033 	if (ret)
1034 		goto out_snap;
1035 
1036 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1037 
1038 	return 0;
1039 
1040 out_snap:
1041 	ceph_snap_exit();
1042 out_xattr:
1043 	ceph_xattr_exit();
1044 	destroy_caches();
1045 out:
1046 	return ret;
1047 }
1048 
1049 static void __exit exit_ceph(void)
1050 {
1051 	dout("exit_ceph\n");
1052 	unregister_filesystem(&ceph_fs_type);
1053 	ceph_snap_exit();
1054 	ceph_xattr_exit();
1055 	destroy_caches();
1056 }
1057 
1058 module_init(init_ceph);
1059 module_exit(exit_ceph);
1060 
1061 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1062 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1063 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1064 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1065 MODULE_LICENSE("GPL");
1066