xref: /openbmc/linux/fs/ceph/super.c (revision 84764a41)
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 
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/decode.h>
23 #include <linux/ceph/mon_client.h>
24 #include <linux/ceph/auth.h>
25 #include <linux/ceph/debugfs.h>
26 
27 /*
28  * Ceph superblock operations
29  *
30  * Handle the basics of mounting, unmounting.
31  */
32 
33 /*
34  * super ops
35  */
36 static void ceph_put_super(struct super_block *s)
37 {
38 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
39 
40 	dout("put_super\n");
41 	ceph_mdsc_close_sessions(fsc->mdsc);
42 
43 	/*
44 	 * ensure we release the bdi before put_anon_super releases
45 	 * the device name.
46 	 */
47 	if (s->s_bdi == &fsc->backing_dev_info) {
48 		bdi_unregister(&fsc->backing_dev_info);
49 		s->s_bdi = NULL;
50 	}
51 
52 	return;
53 }
54 
55 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
56 {
57 	struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
58 	struct ceph_monmap *monmap = fsc->client->monc.monmap;
59 	struct ceph_statfs st;
60 	u64 fsid;
61 	int err;
62 
63 	dout("statfs\n");
64 	err = ceph_monc_do_statfs(&fsc->client->monc, &st);
65 	if (err < 0)
66 		return err;
67 
68 	/* fill in kstatfs */
69 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
70 
71 	/*
72 	 * express utilization in terms of large blocks to avoid
73 	 * overflow on 32-bit machines.
74 	 */
75 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
76 	buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
77 	buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
78 	buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
79 
80 	buf->f_files = le64_to_cpu(st.num_objects);
81 	buf->f_ffree = -1;
82 	buf->f_namelen = NAME_MAX;
83 	buf->f_frsize = PAGE_CACHE_SIZE;
84 
85 	/* leave fsid little-endian, regardless of host endianness */
86 	fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
87 	buf->f_fsid.val[0] = fsid & 0xffffffff;
88 	buf->f_fsid.val[1] = fsid >> 32;
89 
90 	return 0;
91 }
92 
93 
94 static int ceph_sync_fs(struct super_block *sb, int wait)
95 {
96 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
97 
98 	if (!wait) {
99 		dout("sync_fs (non-blocking)\n");
100 		ceph_flush_dirty_caps(fsc->mdsc);
101 		dout("sync_fs (non-blocking) done\n");
102 		return 0;
103 	}
104 
105 	dout("sync_fs (blocking)\n");
106 	ceph_osdc_sync(&fsc->client->osdc);
107 	ceph_mdsc_sync(fsc->mdsc);
108 	dout("sync_fs (blocking) done\n");
109 	return 0;
110 }
111 
112 /*
113  * mount options
114  */
115 enum {
116 	Opt_wsize,
117 	Opt_rsize,
118 	Opt_rasize,
119 	Opt_caps_wanted_delay_min,
120 	Opt_caps_wanted_delay_max,
121 	Opt_cap_release_safety,
122 	Opt_readdir_max_entries,
123 	Opt_readdir_max_bytes,
124 	Opt_congestion_kb,
125 	Opt_last_int,
126 	/* int args above */
127 	Opt_snapdirname,
128 	Opt_last_string,
129 	/* string args above */
130 	Opt_dirstat,
131 	Opt_nodirstat,
132 	Opt_rbytes,
133 	Opt_norbytes,
134 	Opt_asyncreaddir,
135 	Opt_noasyncreaddir,
136 	Opt_dcache,
137 	Opt_nodcache,
138 	Opt_ino32,
139 	Opt_noino32,
140 };
141 
142 static match_table_t fsopt_tokens = {
143 	{Opt_wsize, "wsize=%d"},
144 	{Opt_rsize, "rsize=%d"},
145 	{Opt_rasize, "rasize=%d"},
146 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
147 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
148 	{Opt_cap_release_safety, "cap_release_safety=%d"},
149 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
150 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
151 	{Opt_congestion_kb, "write_congestion_kb=%d"},
152 	/* int args above */
153 	{Opt_snapdirname, "snapdirname=%s"},
154 	/* string args above */
155 	{Opt_dirstat, "dirstat"},
156 	{Opt_nodirstat, "nodirstat"},
157 	{Opt_rbytes, "rbytes"},
158 	{Opt_norbytes, "norbytes"},
159 	{Opt_asyncreaddir, "asyncreaddir"},
160 	{Opt_noasyncreaddir, "noasyncreaddir"},
161 	{Opt_dcache, "dcache"},
162 	{Opt_nodcache, "nodcache"},
163 	{Opt_ino32, "ino32"},
164 	{Opt_noino32, "noino32"},
165 	{-1, NULL}
166 };
167 
168 static int parse_fsopt_token(char *c, void *private)
169 {
170 	struct ceph_mount_options *fsopt = private;
171 	substring_t argstr[MAX_OPT_ARGS];
172 	int token, intval, ret;
173 
174 	token = match_token((char *)c, fsopt_tokens, argstr);
175 	if (token < 0)
176 		return -EINVAL;
177 
178 	if (token < Opt_last_int) {
179 		ret = match_int(&argstr[0], &intval);
180 		if (ret < 0) {
181 			pr_err("bad mount option arg (not int) "
182 			       "at '%s'\n", c);
183 			return ret;
184 		}
185 		dout("got int token %d val %d\n", token, intval);
186 	} else if (token > Opt_last_int && token < Opt_last_string) {
187 		dout("got string token %d val %s\n", token,
188 		     argstr[0].from);
189 	} else {
190 		dout("got token %d\n", token);
191 	}
192 
193 	switch (token) {
194 	case Opt_snapdirname:
195 		kfree(fsopt->snapdir_name);
196 		fsopt->snapdir_name = kstrndup(argstr[0].from,
197 					       argstr[0].to-argstr[0].from,
198 					       GFP_KERNEL);
199 		if (!fsopt->snapdir_name)
200 			return -ENOMEM;
201 		break;
202 
203 		/* misc */
204 	case Opt_wsize:
205 		fsopt->wsize = intval;
206 		break;
207 	case Opt_rsize:
208 		fsopt->rsize = intval;
209 		break;
210 	case Opt_rasize:
211 		fsopt->rasize = intval;
212 		break;
213 	case Opt_caps_wanted_delay_min:
214 		fsopt->caps_wanted_delay_min = intval;
215 		break;
216 	case Opt_caps_wanted_delay_max:
217 		fsopt->caps_wanted_delay_max = intval;
218 		break;
219 	case Opt_readdir_max_entries:
220 		fsopt->max_readdir = intval;
221 		break;
222 	case Opt_readdir_max_bytes:
223 		fsopt->max_readdir_bytes = intval;
224 		break;
225 	case Opt_congestion_kb:
226 		fsopt->congestion_kb = intval;
227 		break;
228 	case Opt_dirstat:
229 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
230 		break;
231 	case Opt_nodirstat:
232 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
233 		break;
234 	case Opt_rbytes:
235 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
236 		break;
237 	case Opt_norbytes:
238 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
239 		break;
240 	case Opt_asyncreaddir:
241 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
242 		break;
243 	case Opt_noasyncreaddir:
244 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
245 		break;
246 	case Opt_dcache:
247 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
248 		break;
249 	case Opt_nodcache:
250 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
251 		break;
252 	case Opt_ino32:
253 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
254 		break;
255 	case Opt_noino32:
256 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
257 		break;
258 	default:
259 		BUG_ON(token);
260 	}
261 	return 0;
262 }
263 
264 static void destroy_mount_options(struct ceph_mount_options *args)
265 {
266 	dout("destroy_mount_options %p\n", args);
267 	kfree(args->snapdir_name);
268 	kfree(args);
269 }
270 
271 static int strcmp_null(const char *s1, const char *s2)
272 {
273 	if (!s1 && !s2)
274 		return 0;
275 	if (s1 && !s2)
276 		return -1;
277 	if (!s1 && s2)
278 		return 1;
279 	return strcmp(s1, s2);
280 }
281 
282 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
283 				 struct ceph_options *new_opt,
284 				 struct ceph_fs_client *fsc)
285 {
286 	struct ceph_mount_options *fsopt1 = new_fsopt;
287 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
288 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
289 	int ret;
290 
291 	ret = memcmp(fsopt1, fsopt2, ofs);
292 	if (ret)
293 		return ret;
294 
295 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
296 	if (ret)
297 		return ret;
298 
299 	return ceph_compare_options(new_opt, fsc->client);
300 }
301 
302 static int parse_mount_options(struct ceph_mount_options **pfsopt,
303 			       struct ceph_options **popt,
304 			       int flags, char *options,
305 			       const char *dev_name,
306 			       const char **path)
307 {
308 	struct ceph_mount_options *fsopt;
309 	const char *dev_name_end;
310 	int err;
311 
312 	if (!dev_name || !*dev_name)
313 		return -EINVAL;
314 
315 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
316 	if (!fsopt)
317 		return -ENOMEM;
318 
319 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
320 
321 	fsopt->sb_flags = flags;
322 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
323 
324 	fsopt->rsize = CEPH_RSIZE_DEFAULT;
325 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
326 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
327 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
328 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
329 	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
330 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
331 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
332 	fsopt->congestion_kb = default_congestion_kb();
333 
334 	/*
335 	 * Distinguish the server list from the path in "dev_name".
336 	 * Internally we do not include the leading '/' in the path.
337 	 *
338 	 * "dev_name" will look like:
339 	 *     <server_spec>[,<server_spec>...]:[<path>]
340 	 * where
341 	 *     <server_spec> is <ip>[:<port>]
342 	 *     <path> is optional, but if present must begin with '/'
343 	 */
344 	dev_name_end = strchr(dev_name, '/');
345 	if (dev_name_end) {
346 		/* skip over leading '/' for path */
347 		*path = dev_name_end + 1;
348 	} else {
349 		/* path is empty */
350 		dev_name_end = dev_name + strlen(dev_name);
351 		*path = dev_name_end;
352 	}
353 	err = -EINVAL;
354 	dev_name_end--;		/* back up to ':' separator */
355 	if (*dev_name_end != ':') {
356 		pr_err("device name is missing path (no : separator in %s)\n",
357 				dev_name);
358 		goto out;
359 	}
360 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
361 	dout("server path '%s'\n", *path);
362 
363 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
364 				 parse_fsopt_token, (void *)fsopt);
365 	if (IS_ERR(*popt)) {
366 		err = PTR_ERR(*popt);
367 		goto out;
368 	}
369 
370 	/* success */
371 	*pfsopt = fsopt;
372 	return 0;
373 
374 out:
375 	destroy_mount_options(fsopt);
376 	return err;
377 }
378 
379 /**
380  * ceph_show_options - Show mount options in /proc/mounts
381  * @m: seq_file to write to
382  * @root: root of that (sub)tree
383  */
384 static int ceph_show_options(struct seq_file *m, struct dentry *root)
385 {
386 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
387 	struct ceph_mount_options *fsopt = fsc->mount_options;
388 	struct ceph_options *opt = fsc->client->options;
389 
390 	if (opt->flags & CEPH_OPT_FSID)
391 		seq_printf(m, ",fsid=%pU", &opt->fsid);
392 	if (opt->flags & CEPH_OPT_NOSHARE)
393 		seq_puts(m, ",noshare");
394 	if (opt->flags & CEPH_OPT_NOCRC)
395 		seq_puts(m, ",nocrc");
396 
397 	if (opt->name)
398 		seq_printf(m, ",name=%s", opt->name);
399 	if (opt->key)
400 		seq_puts(m, ",secret=<hidden>");
401 
402 	if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
403 		seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
404 	if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
405 		seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
406 	if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
407 		seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
408 	if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
409 		seq_printf(m, ",osdkeepalivetimeout=%d",
410 			   opt->osd_keepalive_timeout);
411 
412 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
413 		seq_puts(m, ",dirstat");
414 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
415 		seq_puts(m, ",norbytes");
416 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
417 		seq_puts(m, ",noasyncreaddir");
418 	if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
419 		seq_puts(m, ",dcache");
420 	else
421 		seq_puts(m, ",nodcache");
422 
423 	if (fsopt->wsize)
424 		seq_printf(m, ",wsize=%d", fsopt->wsize);
425 	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
426 		seq_printf(m, ",rsize=%d", fsopt->rsize);
427 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
428 		seq_printf(m, ",rasize=%d", fsopt->rasize);
429 	if (fsopt->congestion_kb != default_congestion_kb())
430 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
431 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
432 		seq_printf(m, ",caps_wanted_delay_min=%d",
433 			 fsopt->caps_wanted_delay_min);
434 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
435 		seq_printf(m, ",caps_wanted_delay_max=%d",
436 			   fsopt->caps_wanted_delay_max);
437 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
438 		seq_printf(m, ",cap_release_safety=%d",
439 			   fsopt->cap_release_safety);
440 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
441 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
442 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
443 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
444 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
445 		seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
446 	return 0;
447 }
448 
449 /*
450  * handle any mon messages the standard library doesn't understand.
451  * return error if we don't either.
452  */
453 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
454 {
455 	struct ceph_fs_client *fsc = client->private;
456 	int type = le16_to_cpu(msg->hdr.type);
457 
458 	switch (type) {
459 	case CEPH_MSG_MDS_MAP:
460 		ceph_mdsc_handle_map(fsc->mdsc, msg);
461 		return 0;
462 
463 	default:
464 		return -1;
465 	}
466 }
467 
468 /*
469  * create a new fs client
470  */
471 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
472 					struct ceph_options *opt)
473 {
474 	struct ceph_fs_client *fsc;
475 	const unsigned supported_features =
476 		CEPH_FEATURE_FLOCK |
477 		CEPH_FEATURE_DIRLAYOUTHASH;
478 	const unsigned required_features = 0;
479 	int err = -ENOMEM;
480 
481 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
482 	if (!fsc)
483 		return ERR_PTR(-ENOMEM);
484 
485 	fsc->client = ceph_create_client(opt, fsc, supported_features,
486 					 required_features);
487 	if (IS_ERR(fsc->client)) {
488 		err = PTR_ERR(fsc->client);
489 		goto fail;
490 	}
491 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
492 	fsc->client->monc.want_mdsmap = 1;
493 
494 	fsc->mount_options = fsopt;
495 
496 	fsc->sb = NULL;
497 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
498 
499 	atomic_long_set(&fsc->writeback_count, 0);
500 
501 	err = bdi_init(&fsc->backing_dev_info);
502 	if (err < 0)
503 		goto fail_client;
504 
505 	err = -ENOMEM;
506 	/*
507 	 * The number of concurrent works can be high but they don't need
508 	 * to be processed in parallel, limit concurrency.
509 	 */
510 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
511 	if (fsc->wb_wq == NULL)
512 		goto fail_bdi;
513 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
514 	if (fsc->pg_inv_wq == NULL)
515 		goto fail_wb_wq;
516 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
517 	if (fsc->trunc_wq == NULL)
518 		goto fail_pg_inv_wq;
519 
520 	/* set up mempools */
521 	err = -ENOMEM;
522 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
523 			      fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
524 	if (!fsc->wb_pagevec_pool)
525 		goto fail_trunc_wq;
526 
527 	/* caps */
528 	fsc->min_caps = fsopt->max_readdir;
529 
530 	return fsc;
531 
532 fail_trunc_wq:
533 	destroy_workqueue(fsc->trunc_wq);
534 fail_pg_inv_wq:
535 	destroy_workqueue(fsc->pg_inv_wq);
536 fail_wb_wq:
537 	destroy_workqueue(fsc->wb_wq);
538 fail_bdi:
539 	bdi_destroy(&fsc->backing_dev_info);
540 fail_client:
541 	ceph_destroy_client(fsc->client);
542 fail:
543 	kfree(fsc);
544 	return ERR_PTR(err);
545 }
546 
547 static void destroy_fs_client(struct ceph_fs_client *fsc)
548 {
549 	dout("destroy_fs_client %p\n", fsc);
550 
551 	destroy_workqueue(fsc->wb_wq);
552 	destroy_workqueue(fsc->pg_inv_wq);
553 	destroy_workqueue(fsc->trunc_wq);
554 
555 	bdi_destroy(&fsc->backing_dev_info);
556 
557 	mempool_destroy(fsc->wb_pagevec_pool);
558 
559 	destroy_mount_options(fsc->mount_options);
560 
561 	ceph_fs_debugfs_cleanup(fsc);
562 
563 	ceph_destroy_client(fsc->client);
564 
565 	kfree(fsc);
566 	dout("destroy_fs_client %p done\n", fsc);
567 }
568 
569 /*
570  * caches
571  */
572 struct kmem_cache *ceph_inode_cachep;
573 struct kmem_cache *ceph_cap_cachep;
574 struct kmem_cache *ceph_dentry_cachep;
575 struct kmem_cache *ceph_file_cachep;
576 
577 static void ceph_inode_init_once(void *foo)
578 {
579 	struct ceph_inode_info *ci = foo;
580 	inode_init_once(&ci->vfs_inode);
581 }
582 
583 static int __init init_caches(void)
584 {
585 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
586 				      sizeof(struct ceph_inode_info),
587 				      __alignof__(struct ceph_inode_info),
588 				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
589 				      ceph_inode_init_once);
590 	if (ceph_inode_cachep == NULL)
591 		return -ENOMEM;
592 
593 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
594 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
595 	if (ceph_cap_cachep == NULL)
596 		goto bad_cap;
597 
598 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
599 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
600 	if (ceph_dentry_cachep == NULL)
601 		goto bad_dentry;
602 
603 	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
604 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
605 	if (ceph_file_cachep == NULL)
606 		goto bad_file;
607 
608 	return 0;
609 
610 bad_file:
611 	kmem_cache_destroy(ceph_dentry_cachep);
612 bad_dentry:
613 	kmem_cache_destroy(ceph_cap_cachep);
614 bad_cap:
615 	kmem_cache_destroy(ceph_inode_cachep);
616 	return -ENOMEM;
617 }
618 
619 static void destroy_caches(void)
620 {
621 	/*
622 	 * Make sure all delayed rcu free inodes are flushed before we
623 	 * destroy cache.
624 	 */
625 	rcu_barrier();
626 	kmem_cache_destroy(ceph_inode_cachep);
627 	kmem_cache_destroy(ceph_cap_cachep);
628 	kmem_cache_destroy(ceph_dentry_cachep);
629 	kmem_cache_destroy(ceph_file_cachep);
630 }
631 
632 
633 /*
634  * ceph_umount_begin - initiate forced umount.  Tear down down the
635  * mount, skipping steps that may hang while waiting for server(s).
636  */
637 static void ceph_umount_begin(struct super_block *sb)
638 {
639 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
640 
641 	dout("ceph_umount_begin - starting forced umount\n");
642 	if (!fsc)
643 		return;
644 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
645 	return;
646 }
647 
648 static const struct super_operations ceph_super_ops = {
649 	.alloc_inode	= ceph_alloc_inode,
650 	.destroy_inode	= ceph_destroy_inode,
651 	.write_inode    = ceph_write_inode,
652 	.sync_fs        = ceph_sync_fs,
653 	.put_super	= ceph_put_super,
654 	.show_options   = ceph_show_options,
655 	.statfs		= ceph_statfs,
656 	.umount_begin   = ceph_umount_begin,
657 };
658 
659 /*
660  * Bootstrap mount by opening the root directory.  Note the mount
661  * @started time from caller, and time out if this takes too long.
662  */
663 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
664 				       const char *path,
665 				       unsigned long started)
666 {
667 	struct ceph_mds_client *mdsc = fsc->mdsc;
668 	struct ceph_mds_request *req = NULL;
669 	int err;
670 	struct dentry *root;
671 
672 	/* open dir */
673 	dout("open_root_inode opening '%s'\n", path);
674 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
675 	if (IS_ERR(req))
676 		return ERR_CAST(req);
677 	req->r_path1 = kstrdup(path, GFP_NOFS);
678 	req->r_ino1.ino = CEPH_INO_ROOT;
679 	req->r_ino1.snap = CEPH_NOSNAP;
680 	req->r_started = started;
681 	req->r_timeout = fsc->client->options->mount_timeout * HZ;
682 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
683 	req->r_num_caps = 2;
684 	err = ceph_mdsc_do_request(mdsc, NULL, req);
685 	if (err == 0) {
686 		struct inode *inode = req->r_target_inode;
687 		req->r_target_inode = NULL;
688 		dout("open_root_inode success\n");
689 		if (ceph_ino(inode) == CEPH_INO_ROOT &&
690 		    fsc->sb->s_root == NULL) {
691 			root = d_make_root(inode);
692 			if (!root) {
693 				root = ERR_PTR(-ENOMEM);
694 				goto out;
695 			}
696 		} else {
697 			root = d_obtain_alias(inode);
698 		}
699 		ceph_init_dentry(root);
700 		dout("open_root_inode success, root dentry is %p\n", root);
701 	} else {
702 		root = ERR_PTR(err);
703 	}
704 out:
705 	ceph_mdsc_put_request(req);
706 	return root;
707 }
708 
709 
710 
711 
712 /*
713  * mount: join the ceph cluster, and open root directory.
714  */
715 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
716 		      const char *path)
717 {
718 	int err;
719 	unsigned long started = jiffies;  /* note the start time */
720 	struct dentry *root;
721 	int first = 0;   /* first vfsmount for this super_block */
722 
723 	dout("mount start\n");
724 	mutex_lock(&fsc->client->mount_mutex);
725 
726 	err = __ceph_open_session(fsc->client, started);
727 	if (err < 0)
728 		goto out;
729 
730 	dout("mount opening root\n");
731 	root = open_root_dentry(fsc, "", started);
732 	if (IS_ERR(root)) {
733 		err = PTR_ERR(root);
734 		goto out;
735 	}
736 	if (fsc->sb->s_root) {
737 		dput(root);
738 	} else {
739 		fsc->sb->s_root = root;
740 		first = 1;
741 
742 		err = ceph_fs_debugfs_init(fsc);
743 		if (err < 0)
744 			goto fail;
745 	}
746 
747 	if (path[0] == 0) {
748 		dget(root);
749 	} else {
750 		dout("mount opening base mountpoint\n");
751 		root = open_root_dentry(fsc, path, started);
752 		if (IS_ERR(root)) {
753 			err = PTR_ERR(root);
754 			goto fail;
755 		}
756 	}
757 
758 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
759 	dout("mount success\n");
760 	mutex_unlock(&fsc->client->mount_mutex);
761 	return root;
762 
763 out:
764 	mutex_unlock(&fsc->client->mount_mutex);
765 	return ERR_PTR(err);
766 
767 fail:
768 	if (first) {
769 		dput(fsc->sb->s_root);
770 		fsc->sb->s_root = NULL;
771 	}
772 	goto out;
773 }
774 
775 static int ceph_set_super(struct super_block *s, void *data)
776 {
777 	struct ceph_fs_client *fsc = data;
778 	int ret;
779 
780 	dout("set_super %p data %p\n", s, data);
781 
782 	s->s_flags = fsc->mount_options->sb_flags;
783 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
784 
785 	s->s_fs_info = fsc;
786 	fsc->sb = s;
787 
788 	s->s_op = &ceph_super_ops;
789 	s->s_export_op = &ceph_export_ops;
790 
791 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
792 
793 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
794 	if (ret != 0)
795 		goto fail;
796 
797 	return ret;
798 
799 fail:
800 	s->s_fs_info = NULL;
801 	fsc->sb = NULL;
802 	return ret;
803 }
804 
805 /*
806  * share superblock if same fs AND options
807  */
808 static int ceph_compare_super(struct super_block *sb, void *data)
809 {
810 	struct ceph_fs_client *new = data;
811 	struct ceph_mount_options *fsopt = new->mount_options;
812 	struct ceph_options *opt = new->client->options;
813 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
814 
815 	dout("ceph_compare_super %p\n", sb);
816 
817 	if (compare_mount_options(fsopt, opt, other)) {
818 		dout("monitor(s)/mount options don't match\n");
819 		return 0;
820 	}
821 	if ((opt->flags & CEPH_OPT_FSID) &&
822 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
823 		dout("fsid doesn't match\n");
824 		return 0;
825 	}
826 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
827 		dout("flags differ\n");
828 		return 0;
829 	}
830 	return 1;
831 }
832 
833 /*
834  * construct our own bdi so we can control readahead, etc.
835  */
836 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
837 
838 static int ceph_register_bdi(struct super_block *sb,
839 			     struct ceph_fs_client *fsc)
840 {
841 	int err;
842 
843 	/* set ra_pages based on rasize mount option? */
844 	if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
845 		fsc->backing_dev_info.ra_pages =
846 			(fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
847 			>> PAGE_SHIFT;
848 	else
849 		fsc->backing_dev_info.ra_pages =
850 			default_backing_dev_info.ra_pages;
851 
852 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
853 			   atomic_long_inc_return(&bdi_seq));
854 	if (!err)
855 		sb->s_bdi = &fsc->backing_dev_info;
856 	return err;
857 }
858 
859 static struct dentry *ceph_mount(struct file_system_type *fs_type,
860 		       int flags, const char *dev_name, void *data)
861 {
862 	struct super_block *sb;
863 	struct ceph_fs_client *fsc;
864 	struct dentry *res;
865 	int err;
866 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
867 	const char *path = NULL;
868 	struct ceph_mount_options *fsopt = NULL;
869 	struct ceph_options *opt = NULL;
870 
871 	dout("ceph_mount\n");
872 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
873 	if (err < 0) {
874 		res = ERR_PTR(err);
875 		goto out_final;
876 	}
877 
878 	/* create client (which we may/may not use) */
879 	fsc = create_fs_client(fsopt, opt);
880 	if (IS_ERR(fsc)) {
881 		res = ERR_CAST(fsc);
882 		destroy_mount_options(fsopt);
883 		ceph_destroy_options(opt);
884 		goto out_final;
885 	}
886 
887 	err = ceph_mdsc_init(fsc);
888 	if (err < 0) {
889 		res = ERR_PTR(err);
890 		goto out;
891 	}
892 
893 	if (ceph_test_opt(fsc->client, NOSHARE))
894 		compare_super = NULL;
895 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
896 	if (IS_ERR(sb)) {
897 		res = ERR_CAST(sb);
898 		goto out;
899 	}
900 
901 	if (ceph_sb_to_client(sb) != fsc) {
902 		ceph_mdsc_destroy(fsc);
903 		destroy_fs_client(fsc);
904 		fsc = ceph_sb_to_client(sb);
905 		dout("get_sb got existing client %p\n", fsc);
906 	} else {
907 		dout("get_sb using new client %p\n", fsc);
908 		err = ceph_register_bdi(sb, fsc);
909 		if (err < 0) {
910 			res = ERR_PTR(err);
911 			goto out_splat;
912 		}
913 	}
914 
915 	res = ceph_real_mount(fsc, path);
916 	if (IS_ERR(res))
917 		goto out_splat;
918 	dout("root %p inode %p ino %llx.%llx\n", res,
919 	     res->d_inode, ceph_vinop(res->d_inode));
920 	return res;
921 
922 out_splat:
923 	ceph_mdsc_close_sessions(fsc->mdsc);
924 	deactivate_locked_super(sb);
925 	goto out_final;
926 
927 out:
928 	ceph_mdsc_destroy(fsc);
929 	destroy_fs_client(fsc);
930 out_final:
931 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
932 	return res;
933 }
934 
935 static void ceph_kill_sb(struct super_block *s)
936 {
937 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
938 	dout("kill_sb %p\n", s);
939 	ceph_mdsc_pre_umount(fsc->mdsc);
940 	kill_anon_super(s);    /* will call put_super after sb is r/o */
941 	ceph_mdsc_destroy(fsc);
942 	destroy_fs_client(fsc);
943 }
944 
945 static struct file_system_type ceph_fs_type = {
946 	.owner		= THIS_MODULE,
947 	.name		= "ceph",
948 	.mount		= ceph_mount,
949 	.kill_sb	= ceph_kill_sb,
950 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
951 };
952 
953 #define _STRINGIFY(x) #x
954 #define STRINGIFY(x) _STRINGIFY(x)
955 
956 static int __init init_ceph(void)
957 {
958 	int ret = init_caches();
959 	if (ret)
960 		goto out;
961 
962 	ceph_xattr_init();
963 	ret = register_filesystem(&ceph_fs_type);
964 	if (ret)
965 		goto out_icache;
966 
967 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
968 
969 	return 0;
970 
971 out_icache:
972 	ceph_xattr_exit();
973 	destroy_caches();
974 out:
975 	return ret;
976 }
977 
978 static void __exit exit_ceph(void)
979 {
980 	dout("exit_ceph\n");
981 	unregister_filesystem(&ceph_fs_type);
982 	ceph_xattr_exit();
983 	destroy_caches();
984 }
985 
986 module_init(init_ceph);
987 module_exit(exit_ceph);
988 
989 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
990 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
991 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
992 MODULE_DESCRIPTION("Ceph filesystem for Linux");
993 MODULE_LICENSE("GPL");
994