xref: /linux/fs/ceph/super.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
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_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 	if (!fsopt->snapdir_name) {
349 		err = -ENOMEM;
350 		goto out;
351 	}
352 
353 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
354 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
355 	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
356 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
357 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
358 	fsopt->congestion_kb = default_congestion_kb();
359 
360 	/*
361 	 * Distinguish the server list from the path in "dev_name".
362 	 * Internally we do not include the leading '/' in the path.
363 	 *
364 	 * "dev_name" will look like:
365 	 *     <server_spec>[,<server_spec>...]:[<path>]
366 	 * where
367 	 *     <server_spec> is <ip>[:<port>]
368 	 *     <path> is optional, but if present must begin with '/'
369 	 */
370 	dev_name_end = strchr(dev_name, '/');
371 	if (dev_name_end) {
372 		/* skip over leading '/' for path */
373 		*path = dev_name_end + 1;
374 	} else {
375 		/* path is empty */
376 		dev_name_end = dev_name + strlen(dev_name);
377 		*path = dev_name_end;
378 	}
379 	err = -EINVAL;
380 	dev_name_end--;		/* back up to ':' separator */
381 	if (dev_name_end < dev_name || *dev_name_end != ':') {
382 		pr_err("device name is missing path (no : separator in %s)\n",
383 				dev_name);
384 		goto out;
385 	}
386 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
387 	dout("server path '%s'\n", *path);
388 
389 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
390 				 parse_fsopt_token, (void *)fsopt);
391 	if (IS_ERR(*popt)) {
392 		err = PTR_ERR(*popt);
393 		goto out;
394 	}
395 
396 	/* success */
397 	*pfsopt = fsopt;
398 	return 0;
399 
400 out:
401 	destroy_mount_options(fsopt);
402 	return err;
403 }
404 
405 /**
406  * ceph_show_options - Show mount options in /proc/mounts
407  * @m: seq_file to write to
408  * @root: root of that (sub)tree
409  */
410 static int ceph_show_options(struct seq_file *m, struct dentry *root)
411 {
412 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
413 	struct ceph_mount_options *fsopt = fsc->mount_options;
414 	size_t pos;
415 	int ret;
416 
417 	/* a comma between MNT/MS and client options */
418 	seq_putc(m, ',');
419 	pos = m->count;
420 
421 	ret = ceph_print_client_options(m, fsc->client);
422 	if (ret)
423 		return ret;
424 
425 	/* retract our comma if no client options */
426 	if (m->count == pos)
427 		m->count--;
428 
429 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
430 		seq_puts(m, ",dirstat");
431 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
432 		seq_puts(m, ",norbytes");
433 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
434 		seq_puts(m, ",noasyncreaddir");
435 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
436 		seq_puts(m, ",nodcache");
437 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
438 		seq_puts(m, ",fsc");
439 
440 #ifdef CONFIG_CEPH_FS_POSIX_ACL
441 	if (fsopt->sb_flags & MS_POSIXACL)
442 		seq_puts(m, ",acl");
443 	else
444 		seq_puts(m, ",noacl");
445 #endif
446 
447 	if (fsopt->wsize)
448 		seq_printf(m, ",wsize=%d", fsopt->wsize);
449 	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
450 		seq_printf(m, ",rsize=%d", fsopt->rsize);
451 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
452 		seq_printf(m, ",rasize=%d", fsopt->rasize);
453 	if (fsopt->congestion_kb != default_congestion_kb())
454 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
455 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
456 		seq_printf(m, ",caps_wanted_delay_min=%d",
457 			 fsopt->caps_wanted_delay_min);
458 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
459 		seq_printf(m, ",caps_wanted_delay_max=%d",
460 			   fsopt->caps_wanted_delay_max);
461 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
462 		seq_printf(m, ",cap_release_safety=%d",
463 			   fsopt->cap_release_safety);
464 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
465 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
466 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
467 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
468 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
469 		seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
470 
471 	return 0;
472 }
473 
474 /*
475  * handle any mon messages the standard library doesn't understand.
476  * return error if we don't either.
477  */
478 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
479 {
480 	struct ceph_fs_client *fsc = client->private;
481 	int type = le16_to_cpu(msg->hdr.type);
482 
483 	switch (type) {
484 	case CEPH_MSG_MDS_MAP:
485 		ceph_mdsc_handle_map(fsc->mdsc, msg);
486 		return 0;
487 
488 	default:
489 		return -1;
490 	}
491 }
492 
493 /*
494  * create a new fs client
495  */
496 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
497 					struct ceph_options *opt)
498 {
499 	struct ceph_fs_client *fsc;
500 	const u64 supported_features =
501 		CEPH_FEATURE_FLOCK |
502 		CEPH_FEATURE_DIRLAYOUTHASH |
503 		CEPH_FEATURE_MDS_INLINE_DATA;
504 	const u64 required_features = 0;
505 	int page_count;
506 	size_t size;
507 	int err = -ENOMEM;
508 
509 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
510 	if (!fsc)
511 		return ERR_PTR(-ENOMEM);
512 
513 	fsc->client = ceph_create_client(opt, fsc, supported_features,
514 					 required_features);
515 	if (IS_ERR(fsc->client)) {
516 		err = PTR_ERR(fsc->client);
517 		goto fail;
518 	}
519 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
520 	fsc->client->monc.want_mdsmap = 1;
521 
522 	fsc->mount_options = fsopt;
523 
524 	fsc->sb = NULL;
525 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
526 
527 	atomic_long_set(&fsc->writeback_count, 0);
528 
529 	err = bdi_init(&fsc->backing_dev_info);
530 	if (err < 0)
531 		goto fail_client;
532 
533 	err = -ENOMEM;
534 	/*
535 	 * The number of concurrent works can be high but they don't need
536 	 * to be processed in parallel, limit concurrency.
537 	 */
538 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
539 	if (fsc->wb_wq == NULL)
540 		goto fail_bdi;
541 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
542 	if (fsc->pg_inv_wq == NULL)
543 		goto fail_wb_wq;
544 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
545 	if (fsc->trunc_wq == NULL)
546 		goto fail_pg_inv_wq;
547 
548 	/* set up mempools */
549 	err = -ENOMEM;
550 	page_count = fsc->mount_options->wsize >> PAGE_CACHE_SHIFT;
551 	size = sizeof (struct page *) * (page_count ? page_count : 1);
552 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
553 	if (!fsc->wb_pagevec_pool)
554 		goto fail_trunc_wq;
555 
556 	/* setup fscache */
557 	if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
558 	    (ceph_fscache_register_fs(fsc) != 0))
559 		goto fail_fscache;
560 
561 	/* caps */
562 	fsc->min_caps = fsopt->max_readdir;
563 
564 	return fsc;
565 
566 fail_fscache:
567 	ceph_fscache_unregister_fs(fsc);
568 fail_trunc_wq:
569 	destroy_workqueue(fsc->trunc_wq);
570 fail_pg_inv_wq:
571 	destroy_workqueue(fsc->pg_inv_wq);
572 fail_wb_wq:
573 	destroy_workqueue(fsc->wb_wq);
574 fail_bdi:
575 	bdi_destroy(&fsc->backing_dev_info);
576 fail_client:
577 	ceph_destroy_client(fsc->client);
578 fail:
579 	kfree(fsc);
580 	return ERR_PTR(err);
581 }
582 
583 static void destroy_fs_client(struct ceph_fs_client *fsc)
584 {
585 	dout("destroy_fs_client %p\n", fsc);
586 
587 	ceph_fscache_unregister_fs(fsc);
588 
589 	destroy_workqueue(fsc->wb_wq);
590 	destroy_workqueue(fsc->pg_inv_wq);
591 	destroy_workqueue(fsc->trunc_wq);
592 
593 	bdi_destroy(&fsc->backing_dev_info);
594 
595 	mempool_destroy(fsc->wb_pagevec_pool);
596 
597 	destroy_mount_options(fsc->mount_options);
598 
599 	ceph_fs_debugfs_cleanup(fsc);
600 
601 	ceph_destroy_client(fsc->client);
602 
603 	kfree(fsc);
604 	dout("destroy_fs_client %p done\n", fsc);
605 }
606 
607 /*
608  * caches
609  */
610 struct kmem_cache *ceph_inode_cachep;
611 struct kmem_cache *ceph_cap_cachep;
612 struct kmem_cache *ceph_dentry_cachep;
613 struct kmem_cache *ceph_file_cachep;
614 
615 static void ceph_inode_init_once(void *foo)
616 {
617 	struct ceph_inode_info *ci = foo;
618 	inode_init_once(&ci->vfs_inode);
619 }
620 
621 static int __init init_caches(void)
622 {
623 	int error = -ENOMEM;
624 
625 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
626 				      sizeof(struct ceph_inode_info),
627 				      __alignof__(struct ceph_inode_info),
628 				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
629 				      ceph_inode_init_once);
630 	if (ceph_inode_cachep == NULL)
631 		return -ENOMEM;
632 
633 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
634 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
635 	if (ceph_cap_cachep == NULL)
636 		goto bad_cap;
637 
638 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
639 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
640 	if (ceph_dentry_cachep == NULL)
641 		goto bad_dentry;
642 
643 	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
644 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
645 	if (ceph_file_cachep == NULL)
646 		goto bad_file;
647 
648 	if ((error = ceph_fscache_register()))
649 		goto bad_file;
650 
651 	return 0;
652 bad_file:
653 	kmem_cache_destroy(ceph_dentry_cachep);
654 bad_dentry:
655 	kmem_cache_destroy(ceph_cap_cachep);
656 bad_cap:
657 	kmem_cache_destroy(ceph_inode_cachep);
658 	return error;
659 }
660 
661 static void destroy_caches(void)
662 {
663 	/*
664 	 * Make sure all delayed rcu free inodes are flushed before we
665 	 * destroy cache.
666 	 */
667 	rcu_barrier();
668 
669 	kmem_cache_destroy(ceph_inode_cachep);
670 	kmem_cache_destroy(ceph_cap_cachep);
671 	kmem_cache_destroy(ceph_dentry_cachep);
672 	kmem_cache_destroy(ceph_file_cachep);
673 
674 	ceph_fscache_unregister();
675 }
676 
677 
678 /*
679  * ceph_umount_begin - initiate forced umount.  Tear down down the
680  * mount, skipping steps that may hang while waiting for server(s).
681  */
682 static void ceph_umount_begin(struct super_block *sb)
683 {
684 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
685 
686 	dout("ceph_umount_begin - starting forced umount\n");
687 	if (!fsc)
688 		return;
689 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
690 	return;
691 }
692 
693 static const struct super_operations ceph_super_ops = {
694 	.alloc_inode	= ceph_alloc_inode,
695 	.destroy_inode	= ceph_destroy_inode,
696 	.write_inode    = ceph_write_inode,
697 	.drop_inode	= ceph_drop_inode,
698 	.sync_fs        = ceph_sync_fs,
699 	.put_super	= ceph_put_super,
700 	.show_options   = ceph_show_options,
701 	.statfs		= ceph_statfs,
702 	.umount_begin   = ceph_umount_begin,
703 };
704 
705 /*
706  * Bootstrap mount by opening the root directory.  Note the mount
707  * @started time from caller, and time out if this takes too long.
708  */
709 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
710 				       const char *path,
711 				       unsigned long started)
712 {
713 	struct ceph_mds_client *mdsc = fsc->mdsc;
714 	struct ceph_mds_request *req = NULL;
715 	int err;
716 	struct dentry *root;
717 
718 	/* open dir */
719 	dout("open_root_inode opening '%s'\n", path);
720 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
721 	if (IS_ERR(req))
722 		return ERR_CAST(req);
723 	req->r_path1 = kstrdup(path, GFP_NOFS);
724 	if (!req->r_path1) {
725 		root = ERR_PTR(-ENOMEM);
726 		goto out;
727 	}
728 
729 	req->r_ino1.ino = CEPH_INO_ROOT;
730 	req->r_ino1.snap = CEPH_NOSNAP;
731 	req->r_started = started;
732 	req->r_timeout = fsc->client->options->mount_timeout * HZ;
733 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
734 	req->r_num_caps = 2;
735 	err = ceph_mdsc_do_request(mdsc, NULL, req);
736 	if (err == 0) {
737 		struct inode *inode = req->r_target_inode;
738 		req->r_target_inode = NULL;
739 		dout("open_root_inode success\n");
740 		if (ceph_ino(inode) == CEPH_INO_ROOT &&
741 		    fsc->sb->s_root == NULL) {
742 			root = d_make_root(inode);
743 			if (!root) {
744 				root = ERR_PTR(-ENOMEM);
745 				goto out;
746 			}
747 		} else {
748 			root = d_obtain_root(inode);
749 		}
750 		ceph_init_dentry(root);
751 		dout("open_root_inode success, root dentry is %p\n", root);
752 	} else {
753 		root = ERR_PTR(err);
754 	}
755 out:
756 	ceph_mdsc_put_request(req);
757 	return root;
758 }
759 
760 
761 
762 
763 /*
764  * mount: join the ceph cluster, and open root directory.
765  */
766 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
767 		      const char *path)
768 {
769 	int err;
770 	unsigned long started = jiffies;  /* note the start time */
771 	struct dentry *root;
772 	int first = 0;   /* first vfsmount for this super_block */
773 
774 	dout("mount start\n");
775 	mutex_lock(&fsc->client->mount_mutex);
776 
777 	err = __ceph_open_session(fsc->client, started);
778 	if (err < 0)
779 		goto out;
780 
781 	dout("mount opening root\n");
782 	root = open_root_dentry(fsc, "", started);
783 	if (IS_ERR(root)) {
784 		err = PTR_ERR(root);
785 		goto out;
786 	}
787 	if (fsc->sb->s_root) {
788 		dput(root);
789 	} else {
790 		fsc->sb->s_root = root;
791 		first = 1;
792 
793 		err = ceph_fs_debugfs_init(fsc);
794 		if (err < 0)
795 			goto fail;
796 	}
797 
798 	if (path[0] == 0) {
799 		dget(root);
800 	} else {
801 		dout("mount opening base mountpoint\n");
802 		root = open_root_dentry(fsc, path, started);
803 		if (IS_ERR(root)) {
804 			err = PTR_ERR(root);
805 			goto fail;
806 		}
807 	}
808 
809 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
810 	dout("mount success\n");
811 	mutex_unlock(&fsc->client->mount_mutex);
812 	return root;
813 
814 out:
815 	mutex_unlock(&fsc->client->mount_mutex);
816 	return ERR_PTR(err);
817 
818 fail:
819 	if (first) {
820 		dput(fsc->sb->s_root);
821 		fsc->sb->s_root = NULL;
822 	}
823 	goto out;
824 }
825 
826 static int ceph_set_super(struct super_block *s, void *data)
827 {
828 	struct ceph_fs_client *fsc = data;
829 	int ret;
830 
831 	dout("set_super %p data %p\n", s, data);
832 
833 	s->s_flags = fsc->mount_options->sb_flags;
834 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
835 
836 	s->s_xattr = ceph_xattr_handlers;
837 	s->s_fs_info = fsc;
838 	fsc->sb = s;
839 
840 	s->s_op = &ceph_super_ops;
841 	s->s_export_op = &ceph_export_ops;
842 
843 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
844 
845 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
846 	if (ret != 0)
847 		goto fail;
848 
849 	return ret;
850 
851 fail:
852 	s->s_fs_info = NULL;
853 	fsc->sb = NULL;
854 	return ret;
855 }
856 
857 /*
858  * share superblock if same fs AND options
859  */
860 static int ceph_compare_super(struct super_block *sb, void *data)
861 {
862 	struct ceph_fs_client *new = data;
863 	struct ceph_mount_options *fsopt = new->mount_options;
864 	struct ceph_options *opt = new->client->options;
865 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
866 
867 	dout("ceph_compare_super %p\n", sb);
868 
869 	if (compare_mount_options(fsopt, opt, other)) {
870 		dout("monitor(s)/mount options don't match\n");
871 		return 0;
872 	}
873 	if ((opt->flags & CEPH_OPT_FSID) &&
874 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
875 		dout("fsid doesn't match\n");
876 		return 0;
877 	}
878 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
879 		dout("flags differ\n");
880 		return 0;
881 	}
882 	return 1;
883 }
884 
885 /*
886  * construct our own bdi so we can control readahead, etc.
887  */
888 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
889 
890 static int ceph_register_bdi(struct super_block *sb,
891 			     struct ceph_fs_client *fsc)
892 {
893 	int err;
894 
895 	/* set ra_pages based on rasize mount option? */
896 	if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
897 		fsc->backing_dev_info.ra_pages =
898 			(fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
899 			>> PAGE_SHIFT;
900 	else
901 		fsc->backing_dev_info.ra_pages =
902 			VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
903 
904 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
905 			   atomic_long_inc_return(&bdi_seq));
906 	if (!err)
907 		sb->s_bdi = &fsc->backing_dev_info;
908 	return err;
909 }
910 
911 static struct dentry *ceph_mount(struct file_system_type *fs_type,
912 		       int flags, const char *dev_name, void *data)
913 {
914 	struct super_block *sb;
915 	struct ceph_fs_client *fsc;
916 	struct dentry *res;
917 	int err;
918 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
919 	const char *path = NULL;
920 	struct ceph_mount_options *fsopt = NULL;
921 	struct ceph_options *opt = NULL;
922 
923 	dout("ceph_mount\n");
924 
925 #ifdef CONFIG_CEPH_FS_POSIX_ACL
926 	flags |= MS_POSIXACL;
927 #endif
928 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
929 	if (err < 0) {
930 		res = ERR_PTR(err);
931 		goto out_final;
932 	}
933 
934 	/* create client (which we may/may not use) */
935 	fsc = create_fs_client(fsopt, opt);
936 	if (IS_ERR(fsc)) {
937 		res = ERR_CAST(fsc);
938 		destroy_mount_options(fsopt);
939 		ceph_destroy_options(opt);
940 		goto out_final;
941 	}
942 
943 	err = ceph_mdsc_init(fsc);
944 	if (err < 0) {
945 		res = ERR_PTR(err);
946 		goto out;
947 	}
948 
949 	if (ceph_test_opt(fsc->client, NOSHARE))
950 		compare_super = NULL;
951 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
952 	if (IS_ERR(sb)) {
953 		res = ERR_CAST(sb);
954 		goto out;
955 	}
956 
957 	if (ceph_sb_to_client(sb) != fsc) {
958 		ceph_mdsc_destroy(fsc);
959 		destroy_fs_client(fsc);
960 		fsc = ceph_sb_to_client(sb);
961 		dout("get_sb got existing client %p\n", fsc);
962 	} else {
963 		dout("get_sb using new client %p\n", fsc);
964 		err = ceph_register_bdi(sb, fsc);
965 		if (err < 0) {
966 			res = ERR_PTR(err);
967 			goto out_splat;
968 		}
969 	}
970 
971 	res = ceph_real_mount(fsc, path);
972 	if (IS_ERR(res))
973 		goto out_splat;
974 	dout("root %p inode %p ino %llx.%llx\n", res,
975 	     d_inode(res), ceph_vinop(d_inode(res)));
976 	return res;
977 
978 out_splat:
979 	ceph_mdsc_close_sessions(fsc->mdsc);
980 	deactivate_locked_super(sb);
981 	goto out_final;
982 
983 out:
984 	ceph_mdsc_destroy(fsc);
985 	destroy_fs_client(fsc);
986 out_final:
987 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
988 	return res;
989 }
990 
991 static void ceph_kill_sb(struct super_block *s)
992 {
993 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
994 	dev_t dev = s->s_dev;
995 
996 	dout("kill_sb %p\n", s);
997 
998 	ceph_mdsc_pre_umount(fsc->mdsc);
999 	generic_shutdown_super(s);
1000 	ceph_mdsc_destroy(fsc);
1001 
1002 	destroy_fs_client(fsc);
1003 	free_anon_bdev(dev);
1004 }
1005 
1006 static struct file_system_type ceph_fs_type = {
1007 	.owner		= THIS_MODULE,
1008 	.name		= "ceph",
1009 	.mount		= ceph_mount,
1010 	.kill_sb	= ceph_kill_sb,
1011 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1012 };
1013 MODULE_ALIAS_FS("ceph");
1014 
1015 static int __init init_ceph(void)
1016 {
1017 	int ret = init_caches();
1018 	if (ret)
1019 		goto out;
1020 
1021 	ceph_flock_init();
1022 	ceph_xattr_init();
1023 	ret = ceph_snap_init();
1024 	if (ret)
1025 		goto out_xattr;
1026 	ret = register_filesystem(&ceph_fs_type);
1027 	if (ret)
1028 		goto out_snap;
1029 
1030 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1031 
1032 	return 0;
1033 
1034 out_snap:
1035 	ceph_snap_exit();
1036 out_xattr:
1037 	ceph_xattr_exit();
1038 	destroy_caches();
1039 out:
1040 	return ret;
1041 }
1042 
1043 static void __exit exit_ceph(void)
1044 {
1045 	dout("exit_ceph\n");
1046 	unregister_filesystem(&ceph_fs_type);
1047 	ceph_snap_exit();
1048 	ceph_xattr_exit();
1049 	destroy_caches();
1050 }
1051 
1052 module_init(init_ceph);
1053 module_exit(exit_ceph);
1054 
1055 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1056 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1057 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1058 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1059 MODULE_LICENSE("GPL");
1060