xref: /linux/fs/ceph/super.c (revision 092e0e7e520a1fca03e13c9f2d157432a8657ff2)
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/decode.h>
22 #include <linux/ceph/mon_client.h>
23 #include <linux/ceph/auth.h>
24 #include <linux/ceph/debugfs.h>
25 
26 /*
27  * Ceph superblock operations
28  *
29  * Handle the basics of mounting, unmounting.
30  */
31 
32 /*
33  * super ops
34  */
35 static void ceph_put_super(struct super_block *s)
36 {
37 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
38 
39 	dout("put_super\n");
40 	ceph_mdsc_close_sessions(fsc->mdsc);
41 
42 	/*
43 	 * ensure we release the bdi before put_anon_super releases
44 	 * the device name.
45 	 */
46 	if (s->s_bdi == &fsc->backing_dev_info) {
47 		bdi_unregister(&fsc->backing_dev_info);
48 		s->s_bdi = NULL;
49 	}
50 
51 	return;
52 }
53 
54 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
55 {
56 	struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
57 	struct ceph_monmap *monmap = fsc->client->monc.monmap;
58 	struct ceph_statfs st;
59 	u64 fsid;
60 	int err;
61 
62 	dout("statfs\n");
63 	err = ceph_monc_do_statfs(&fsc->client->monc, &st);
64 	if (err < 0)
65 		return err;
66 
67 	/* fill in kstatfs */
68 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
69 
70 	/*
71 	 * express utilization in terms of large blocks to avoid
72 	 * overflow on 32-bit machines.
73 	 */
74 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
75 	buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
76 	buf->f_bfree = (le64_to_cpu(st.kb) - le64_to_cpu(st.kb_used)) >>
77 		(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_caps_wanted_delay_min,
119 	Opt_caps_wanted_delay_max,
120 	Opt_cap_release_safety,
121 	Opt_readdir_max_entries,
122 	Opt_readdir_max_bytes,
123 	Opt_congestion_kb,
124 	Opt_last_int,
125 	/* int args above */
126 	Opt_snapdirname,
127 	Opt_last_string,
128 	/* string args above */
129 	Opt_dirstat,
130 	Opt_nodirstat,
131 	Opt_rbytes,
132 	Opt_norbytes,
133 	Opt_noasyncreaddir,
134 };
135 
136 static match_table_t fsopt_tokens = {
137 	{Opt_wsize, "wsize=%d"},
138 	{Opt_rsize, "rsize=%d"},
139 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
140 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
141 	{Opt_cap_release_safety, "cap_release_safety=%d"},
142 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
143 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
144 	{Opt_congestion_kb, "write_congestion_kb=%d"},
145 	/* int args above */
146 	{Opt_snapdirname, "snapdirname=%s"},
147 	/* string args above */
148 	{Opt_dirstat, "dirstat"},
149 	{Opt_nodirstat, "nodirstat"},
150 	{Opt_rbytes, "rbytes"},
151 	{Opt_norbytes, "norbytes"},
152 	{Opt_noasyncreaddir, "noasyncreaddir"},
153 	{-1, NULL}
154 };
155 
156 static int parse_fsopt_token(char *c, void *private)
157 {
158 	struct ceph_mount_options *fsopt = private;
159 	substring_t argstr[MAX_OPT_ARGS];
160 	int token, intval, ret;
161 
162 	token = match_token((char *)c, fsopt_tokens, argstr);
163 	if (token < 0)
164 		return -EINVAL;
165 
166 	if (token < Opt_last_int) {
167 		ret = match_int(&argstr[0], &intval);
168 		if (ret < 0) {
169 			pr_err("bad mount option arg (not int) "
170 			       "at '%s'\n", c);
171 			return ret;
172 		}
173 		dout("got int token %d val %d\n", token, intval);
174 	} else if (token > Opt_last_int && token < Opt_last_string) {
175 		dout("got string token %d val %s\n", token,
176 		     argstr[0].from);
177 	} else {
178 		dout("got token %d\n", token);
179 	}
180 
181 	switch (token) {
182 	case Opt_snapdirname:
183 		kfree(fsopt->snapdir_name);
184 		fsopt->snapdir_name = kstrndup(argstr[0].from,
185 					       argstr[0].to-argstr[0].from,
186 					       GFP_KERNEL);
187 		if (!fsopt->snapdir_name)
188 			return -ENOMEM;
189 		break;
190 
191 		/* misc */
192 	case Opt_wsize:
193 		fsopt->wsize = intval;
194 		break;
195 	case Opt_rsize:
196 		fsopt->rsize = intval;
197 		break;
198 	case Opt_caps_wanted_delay_min:
199 		fsopt->caps_wanted_delay_min = intval;
200 		break;
201 	case Opt_caps_wanted_delay_max:
202 		fsopt->caps_wanted_delay_max = intval;
203 		break;
204 	case Opt_readdir_max_entries:
205 		fsopt->max_readdir = intval;
206 		break;
207 	case Opt_readdir_max_bytes:
208 		fsopt->max_readdir_bytes = intval;
209 		break;
210 	case Opt_congestion_kb:
211 		fsopt->congestion_kb = intval;
212 		break;
213 	case Opt_dirstat:
214 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
215 		break;
216 	case Opt_nodirstat:
217 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
218 		break;
219 	case Opt_rbytes:
220 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
221 		break;
222 	case Opt_norbytes:
223 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
224 		break;
225 	case Opt_noasyncreaddir:
226 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
227 		break;
228 	default:
229 		BUG_ON(token);
230 	}
231 	return 0;
232 }
233 
234 static void destroy_mount_options(struct ceph_mount_options *args)
235 {
236 	dout("destroy_mount_options %p\n", args);
237 	kfree(args->snapdir_name);
238 	kfree(args);
239 }
240 
241 static int strcmp_null(const char *s1, const char *s2)
242 {
243 	if (!s1 && !s2)
244 		return 0;
245 	if (s1 && !s2)
246 		return -1;
247 	if (!s1 && s2)
248 		return 1;
249 	return strcmp(s1, s2);
250 }
251 
252 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
253 				 struct ceph_options *new_opt,
254 				 struct ceph_fs_client *fsc)
255 {
256 	struct ceph_mount_options *fsopt1 = new_fsopt;
257 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
258 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
259 	int ret;
260 
261 	ret = memcmp(fsopt1, fsopt2, ofs);
262 	if (ret)
263 		return ret;
264 
265 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
266 	if (ret)
267 		return ret;
268 
269 	return ceph_compare_options(new_opt, fsc->client);
270 }
271 
272 static int parse_mount_options(struct ceph_mount_options **pfsopt,
273 			       struct ceph_options **popt,
274 			       int flags, char *options,
275 			       const char *dev_name,
276 			       const char **path)
277 {
278 	struct ceph_mount_options *fsopt;
279 	const char *dev_name_end;
280 	int err = -ENOMEM;
281 
282 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
283 	if (!fsopt)
284 		return -ENOMEM;
285 
286 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
287 
288         fsopt->sb_flags = flags;
289         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
290 
291         fsopt->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
292         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
293         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
294         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
295         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
296         fsopt->congestion_kb = default_congestion_kb();
297 
298         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
299         err = -EINVAL;
300         if (!dev_name)
301                 goto out;
302         *path = strstr(dev_name, ":/");
303         if (*path == NULL) {
304                 pr_err("device name is missing path (no :/ in %s)\n",
305                        dev_name);
306                 goto out;
307         }
308 	dev_name_end = *path;
309 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
310 
311 	/* path on server */
312 	*path += 2;
313 	dout("server path '%s'\n", *path);
314 
315 	err = ceph_parse_options(popt, options, dev_name, dev_name_end,
316 				 parse_fsopt_token, (void *)fsopt);
317 	if (err)
318 		goto out;
319 
320 	/* success */
321 	*pfsopt = fsopt;
322 	return 0;
323 
324 out:
325 	destroy_mount_options(fsopt);
326 	return err;
327 }
328 
329 /**
330  * ceph_show_options - Show mount options in /proc/mounts
331  * @m: seq_file to write to
332  * @mnt: mount descriptor
333  */
334 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
335 {
336 	struct ceph_fs_client *fsc = ceph_sb_to_client(mnt->mnt_sb);
337 	struct ceph_mount_options *fsopt = fsc->mount_options;
338 	struct ceph_options *opt = fsc->client->options;
339 
340 	if (opt->flags & CEPH_OPT_FSID)
341 		seq_printf(m, ",fsid=%pU", &opt->fsid);
342 	if (opt->flags & CEPH_OPT_NOSHARE)
343 		seq_puts(m, ",noshare");
344 	if (opt->flags & CEPH_OPT_NOCRC)
345 		seq_puts(m, ",nocrc");
346 
347 	if (opt->name)
348 		seq_printf(m, ",name=%s", opt->name);
349 	if (opt->secret)
350 		seq_puts(m, ",secret=<hidden>");
351 
352 	if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
353 		seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
354 	if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
355 		seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
356 	if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
357 		seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
358 	if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
359 		seq_printf(m, ",osdkeepalivetimeout=%d",
360 			   opt->osd_keepalive_timeout);
361 
362 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
363 		seq_puts(m, ",dirstat");
364 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
365 		seq_puts(m, ",norbytes");
366 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
367 		seq_puts(m, ",noasyncreaddir");
368 
369 	if (fsopt->wsize)
370 		seq_printf(m, ",wsize=%d", fsopt->wsize);
371 	if (fsopt->rsize != CEPH_MOUNT_RSIZE_DEFAULT)
372 		seq_printf(m, ",rsize=%d", fsopt->rsize);
373 	if (fsopt->congestion_kb != default_congestion_kb())
374 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
375 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
376 		seq_printf(m, ",caps_wanted_delay_min=%d",
377 			 fsopt->caps_wanted_delay_min);
378 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
379 		seq_printf(m, ",caps_wanted_delay_max=%d",
380 			   fsopt->caps_wanted_delay_max);
381 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
382 		seq_printf(m, ",cap_release_safety=%d",
383 			   fsopt->cap_release_safety);
384 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
385 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
386 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
387 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
388 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
389 		seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
390 	return 0;
391 }
392 
393 /*
394  * handle any mon messages the standard library doesn't understand.
395  * return error if we don't either.
396  */
397 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
398 {
399 	struct ceph_fs_client *fsc = client->private;
400 	int type = le16_to_cpu(msg->hdr.type);
401 
402 	switch (type) {
403 	case CEPH_MSG_MDS_MAP:
404 		ceph_mdsc_handle_map(fsc->mdsc, msg);
405 		return 0;
406 
407 	default:
408 		return -1;
409 	}
410 }
411 
412 /*
413  * create a new fs client
414  */
415 struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
416 					struct ceph_options *opt)
417 {
418 	struct ceph_fs_client *fsc;
419 	int err = -ENOMEM;
420 
421 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
422 	if (!fsc)
423 		return ERR_PTR(-ENOMEM);
424 
425 	fsc->client = ceph_create_client(opt, fsc);
426 	if (IS_ERR(fsc->client)) {
427 		err = PTR_ERR(fsc->client);
428 		goto fail;
429 	}
430 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
431 	fsc->client->supported_features |= CEPH_FEATURE_FLOCK;
432 	fsc->client->monc.want_mdsmap = 1;
433 
434 	fsc->mount_options = fsopt;
435 
436 	fsc->sb = NULL;
437 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
438 
439 	atomic_long_set(&fsc->writeback_count, 0);
440 
441 	err = bdi_init(&fsc->backing_dev_info);
442 	if (err < 0)
443 		goto fail_client;
444 
445 	err = -ENOMEM;
446 	fsc->wb_wq = create_workqueue("ceph-writeback");
447 	if (fsc->wb_wq == NULL)
448 		goto fail_bdi;
449 	fsc->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
450 	if (fsc->pg_inv_wq == NULL)
451 		goto fail_wb_wq;
452 	fsc->trunc_wq = create_singlethread_workqueue("ceph-trunc");
453 	if (fsc->trunc_wq == NULL)
454 		goto fail_pg_inv_wq;
455 
456 	/* set up mempools */
457 	err = -ENOMEM;
458 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
459 			      fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
460 	if (!fsc->wb_pagevec_pool)
461 		goto fail_trunc_wq;
462 
463 	/* caps */
464 	fsc->min_caps = fsopt->max_readdir;
465 
466 	return fsc;
467 
468 fail_trunc_wq:
469 	destroy_workqueue(fsc->trunc_wq);
470 fail_pg_inv_wq:
471 	destroy_workqueue(fsc->pg_inv_wq);
472 fail_wb_wq:
473 	destroy_workqueue(fsc->wb_wq);
474 fail_bdi:
475 	bdi_destroy(&fsc->backing_dev_info);
476 fail_client:
477 	ceph_destroy_client(fsc->client);
478 fail:
479 	kfree(fsc);
480 	return ERR_PTR(err);
481 }
482 
483 void destroy_fs_client(struct ceph_fs_client *fsc)
484 {
485 	dout("destroy_fs_client %p\n", fsc);
486 
487 	destroy_workqueue(fsc->wb_wq);
488 	destroy_workqueue(fsc->pg_inv_wq);
489 	destroy_workqueue(fsc->trunc_wq);
490 
491 	bdi_destroy(&fsc->backing_dev_info);
492 
493 	mempool_destroy(fsc->wb_pagevec_pool);
494 
495 	destroy_mount_options(fsc->mount_options);
496 
497 	ceph_fs_debugfs_cleanup(fsc);
498 
499 	ceph_destroy_client(fsc->client);
500 
501 	kfree(fsc);
502 	dout("destroy_fs_client %p done\n", fsc);
503 }
504 
505 /*
506  * caches
507  */
508 struct kmem_cache *ceph_inode_cachep;
509 struct kmem_cache *ceph_cap_cachep;
510 struct kmem_cache *ceph_dentry_cachep;
511 struct kmem_cache *ceph_file_cachep;
512 
513 static void ceph_inode_init_once(void *foo)
514 {
515 	struct ceph_inode_info *ci = foo;
516 	inode_init_once(&ci->vfs_inode);
517 }
518 
519 static int __init init_caches(void)
520 {
521 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
522 				      sizeof(struct ceph_inode_info),
523 				      __alignof__(struct ceph_inode_info),
524 				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
525 				      ceph_inode_init_once);
526 	if (ceph_inode_cachep == NULL)
527 		return -ENOMEM;
528 
529 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
530 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
531 	if (ceph_cap_cachep == NULL)
532 		goto bad_cap;
533 
534 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
535 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
536 	if (ceph_dentry_cachep == NULL)
537 		goto bad_dentry;
538 
539 	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
540 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
541 	if (ceph_file_cachep == NULL)
542 		goto bad_file;
543 
544 	return 0;
545 
546 bad_file:
547 	kmem_cache_destroy(ceph_dentry_cachep);
548 bad_dentry:
549 	kmem_cache_destroy(ceph_cap_cachep);
550 bad_cap:
551 	kmem_cache_destroy(ceph_inode_cachep);
552 	return -ENOMEM;
553 }
554 
555 static void destroy_caches(void)
556 {
557 	kmem_cache_destroy(ceph_inode_cachep);
558 	kmem_cache_destroy(ceph_cap_cachep);
559 	kmem_cache_destroy(ceph_dentry_cachep);
560 	kmem_cache_destroy(ceph_file_cachep);
561 }
562 
563 
564 /*
565  * ceph_umount_begin - initiate forced umount.  Tear down down the
566  * mount, skipping steps that may hang while waiting for server(s).
567  */
568 static void ceph_umount_begin(struct super_block *sb)
569 {
570 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
571 
572 	dout("ceph_umount_begin - starting forced umount\n");
573 	if (!fsc)
574 		return;
575 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
576 	return;
577 }
578 
579 static const struct super_operations ceph_super_ops = {
580 	.alloc_inode	= ceph_alloc_inode,
581 	.destroy_inode	= ceph_destroy_inode,
582 	.write_inode    = ceph_write_inode,
583 	.sync_fs        = ceph_sync_fs,
584 	.put_super	= ceph_put_super,
585 	.show_options   = ceph_show_options,
586 	.statfs		= ceph_statfs,
587 	.umount_begin   = ceph_umount_begin,
588 };
589 
590 /*
591  * Bootstrap mount by opening the root directory.  Note the mount
592  * @started time from caller, and time out if this takes too long.
593  */
594 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
595 				       const char *path,
596 				       unsigned long started)
597 {
598 	struct ceph_mds_client *mdsc = fsc->mdsc;
599 	struct ceph_mds_request *req = NULL;
600 	int err;
601 	struct dentry *root;
602 
603 	/* open dir */
604 	dout("open_root_inode opening '%s'\n", path);
605 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
606 	if (IS_ERR(req))
607 		return ERR_CAST(req);
608 	req->r_path1 = kstrdup(path, GFP_NOFS);
609 	req->r_ino1.ino = CEPH_INO_ROOT;
610 	req->r_ino1.snap = CEPH_NOSNAP;
611 	req->r_started = started;
612 	req->r_timeout = fsc->client->options->mount_timeout * HZ;
613 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
614 	req->r_num_caps = 2;
615 	err = ceph_mdsc_do_request(mdsc, NULL, req);
616 	if (err == 0) {
617 		dout("open_root_inode success\n");
618 		if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
619 		    fsc->sb->s_root == NULL)
620 			root = d_alloc_root(req->r_target_inode);
621 		else
622 			root = d_obtain_alias(req->r_target_inode);
623 		req->r_target_inode = NULL;
624 		dout("open_root_inode success, root dentry is %p\n", root);
625 	} else {
626 		root = ERR_PTR(err);
627 	}
628 	ceph_mdsc_put_request(req);
629 	return root;
630 }
631 
632 
633 
634 
635 /*
636  * mount: join the ceph cluster, and open root directory.
637  */
638 static int ceph_mount(struct ceph_fs_client *fsc, struct vfsmount *mnt,
639 		      const char *path)
640 {
641 	int err;
642 	unsigned long started = jiffies;  /* note the start time */
643 	struct dentry *root;
644 	int first = 0;   /* first vfsmount for this super_block */
645 
646 	dout("mount start\n");
647 	mutex_lock(&fsc->client->mount_mutex);
648 
649 	err = __ceph_open_session(fsc->client, started);
650 	if (err < 0)
651 		goto out;
652 
653 	dout("mount opening root\n");
654 	root = open_root_dentry(fsc, "", started);
655 	if (IS_ERR(root)) {
656 		err = PTR_ERR(root);
657 		goto out;
658 	}
659 	if (fsc->sb->s_root) {
660 		dput(root);
661 	} else {
662 		fsc->sb->s_root = root;
663 		first = 1;
664 
665 		err = ceph_fs_debugfs_init(fsc);
666 		if (err < 0)
667 			goto fail;
668 	}
669 
670 	if (path[0] == 0) {
671 		dget(root);
672 	} else {
673 		dout("mount opening base mountpoint\n");
674 		root = open_root_dentry(fsc, path, started);
675 		if (IS_ERR(root)) {
676 			err = PTR_ERR(root);
677 			goto fail;
678 		}
679 	}
680 
681 	mnt->mnt_root = root;
682 	mnt->mnt_sb = fsc->sb;
683 
684 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
685 	dout("mount success\n");
686 	err = 0;
687 
688 out:
689 	mutex_unlock(&fsc->client->mount_mutex);
690 	return err;
691 
692 fail:
693 	if (first) {
694 		dput(fsc->sb->s_root);
695 		fsc->sb->s_root = NULL;
696 	}
697 	goto out;
698 }
699 
700 static int ceph_set_super(struct super_block *s, void *data)
701 {
702 	struct ceph_fs_client *fsc = data;
703 	int ret;
704 
705 	dout("set_super %p data %p\n", s, data);
706 
707 	s->s_flags = fsc->mount_options->sb_flags;
708 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
709 
710 	s->s_fs_info = fsc;
711 	fsc->sb = s;
712 
713 	s->s_op = &ceph_super_ops;
714 	s->s_export_op = &ceph_export_ops;
715 
716 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
717 
718 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
719 	if (ret != 0)
720 		goto fail;
721 
722 	return ret;
723 
724 fail:
725 	s->s_fs_info = NULL;
726 	fsc->sb = NULL;
727 	return ret;
728 }
729 
730 /*
731  * share superblock if same fs AND options
732  */
733 static int ceph_compare_super(struct super_block *sb, void *data)
734 {
735 	struct ceph_fs_client *new = data;
736 	struct ceph_mount_options *fsopt = new->mount_options;
737 	struct ceph_options *opt = new->client->options;
738 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
739 
740 	dout("ceph_compare_super %p\n", sb);
741 
742 	if (compare_mount_options(fsopt, opt, other)) {
743 		dout("monitor(s)/mount options don't match\n");
744 		return 0;
745 	}
746 	if ((opt->flags & CEPH_OPT_FSID) &&
747 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
748 		dout("fsid doesn't match\n");
749 		return 0;
750 	}
751 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
752 		dout("flags differ\n");
753 		return 0;
754 	}
755 	return 1;
756 }
757 
758 /*
759  * construct our own bdi so we can control readahead, etc.
760  */
761 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
762 
763 static int ceph_register_bdi(struct super_block *sb,
764 			     struct ceph_fs_client *fsc)
765 {
766 	int err;
767 
768 	/* set ra_pages based on rsize mount option? */
769 	if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
770 		fsc->backing_dev_info.ra_pages =
771 			(fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
772 			>> PAGE_SHIFT;
773 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
774 			   atomic_long_inc_return(&bdi_seq));
775 	if (!err)
776 		sb->s_bdi = &fsc->backing_dev_info;
777 	return err;
778 }
779 
780 static int ceph_get_sb(struct file_system_type *fs_type,
781 		       int flags, const char *dev_name, void *data,
782 		       struct vfsmount *mnt)
783 {
784 	struct super_block *sb;
785 	struct ceph_fs_client *fsc;
786 	int err;
787 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
788 	const char *path = NULL;
789 	struct ceph_mount_options *fsopt = NULL;
790 	struct ceph_options *opt = NULL;
791 
792 	dout("ceph_get_sb\n");
793 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
794 	if (err < 0)
795 		goto out_final;
796 
797 	/* create client (which we may/may not use) */
798 	fsc = create_fs_client(fsopt, opt);
799 	if (IS_ERR(fsc)) {
800 		err = PTR_ERR(fsc);
801 		kfree(fsopt);
802 		kfree(opt);
803 		goto out_final;
804 	}
805 
806 	err = ceph_mdsc_init(fsc);
807 	if (err < 0)
808 		goto out;
809 
810 	if (ceph_test_opt(fsc->client, NOSHARE))
811 		compare_super = NULL;
812 	sb = sget(fs_type, compare_super, ceph_set_super, fsc);
813 	if (IS_ERR(sb)) {
814 		err = PTR_ERR(sb);
815 		goto out;
816 	}
817 
818 	if (ceph_sb_to_client(sb) != fsc) {
819 		ceph_mdsc_destroy(fsc);
820 		destroy_fs_client(fsc);
821 		fsc = ceph_sb_to_client(sb);
822 		dout("get_sb got existing client %p\n", fsc);
823 	} else {
824 		dout("get_sb using new client %p\n", fsc);
825 		err = ceph_register_bdi(sb, fsc);
826 		if (err < 0)
827 			goto out_splat;
828 	}
829 
830 	err = ceph_mount(fsc, mnt, path);
831 	if (err < 0)
832 		goto out_splat;
833 	dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
834 	     mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
835 	return 0;
836 
837 out_splat:
838 	ceph_mdsc_close_sessions(fsc->mdsc);
839 	deactivate_locked_super(sb);
840 	goto out_final;
841 
842 out:
843 	ceph_mdsc_destroy(fsc);
844 	destroy_fs_client(fsc);
845 out_final:
846 	dout("ceph_get_sb fail %d\n", err);
847 	return err;
848 }
849 
850 static void ceph_kill_sb(struct super_block *s)
851 {
852 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
853 	dout("kill_sb %p\n", s);
854 	ceph_mdsc_pre_umount(fsc->mdsc);
855 	kill_anon_super(s);    /* will call put_super after sb is r/o */
856 	ceph_mdsc_destroy(fsc);
857 	destroy_fs_client(fsc);
858 }
859 
860 static struct file_system_type ceph_fs_type = {
861 	.owner		= THIS_MODULE,
862 	.name		= "ceph",
863 	.get_sb		= ceph_get_sb,
864 	.kill_sb	= ceph_kill_sb,
865 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
866 };
867 
868 #define _STRINGIFY(x) #x
869 #define STRINGIFY(x) _STRINGIFY(x)
870 
871 static int __init init_ceph(void)
872 {
873 	int ret = init_caches();
874 	if (ret)
875 		goto out;
876 
877 	ret = register_filesystem(&ceph_fs_type);
878 	if (ret)
879 		goto out_icache;
880 
881 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
882 
883 	return 0;
884 
885 out_icache:
886 	destroy_caches();
887 out:
888 	return ret;
889 }
890 
891 static void __exit exit_ceph(void)
892 {
893 	dout("exit_ceph\n");
894 	unregister_filesystem(&ceph_fs_type);
895 	destroy_caches();
896 }
897 
898 module_init(init_ceph);
899 module_exit(exit_ceph);
900 
901 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
902 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
903 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
904 MODULE_DESCRIPTION("Ceph filesystem for Linux");
905 MODULE_LICENSE("GPL");
906