Linux Audio

Check our new training course

Loading...
v3.5.6
  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 */
 35static 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
 54static 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_avail) >> (CEPH_BLOCK_SHIFT-10);
 77	buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
 78
 79	buf->f_files = le64_to_cpu(st.num_objects);
 80	buf->f_ffree = -1;
 81	buf->f_namelen = NAME_MAX;
 82	buf->f_frsize = PAGE_CACHE_SIZE;
 83
 84	/* leave fsid little-endian, regardless of host endianness */
 85	fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
 86	buf->f_fsid.val[0] = fsid & 0xffffffff;
 87	buf->f_fsid.val[1] = fsid >> 32;
 88
 89	return 0;
 90}
 91
 92
 93static int ceph_sync_fs(struct super_block *sb, int wait)
 94{
 95	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 96
 97	if (!wait) {
 98		dout("sync_fs (non-blocking)\n");
 99		ceph_flush_dirty_caps(fsc->mdsc);
100		dout("sync_fs (non-blocking) done\n");
101		return 0;
102	}
103
104	dout("sync_fs (blocking)\n");
105	ceph_osdc_sync(&fsc->client->osdc);
106	ceph_mdsc_sync(fsc->mdsc);
107	dout("sync_fs (blocking) done\n");
108	return 0;
109}
110
111/*
112 * mount options
113 */
114enum {
115	Opt_wsize,
116	Opt_rsize,
117	Opt_rasize,
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_asyncreaddir,
134	Opt_noasyncreaddir,
135	Opt_dcache,
136	Opt_nodcache,
137	Opt_ino32,
138	Opt_noino32,
 
 
 
 
 
 
 
 
139};
140
141static match_table_t fsopt_tokens = {
142	{Opt_wsize, "wsize=%d"},
143	{Opt_rsize, "rsize=%d"},
144	{Opt_rasize, "rasize=%d"},
145	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
146	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
147	{Opt_cap_release_safety, "cap_release_safety=%d"},
148	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
149	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
150	{Opt_congestion_kb, "write_congestion_kb=%d"},
151	/* int args above */
152	{Opt_snapdirname, "snapdirname=%s"},
153	/* string args above */
154	{Opt_dirstat, "dirstat"},
155	{Opt_nodirstat, "nodirstat"},
156	{Opt_rbytes, "rbytes"},
157	{Opt_norbytes, "norbytes"},
158	{Opt_asyncreaddir, "asyncreaddir"},
159	{Opt_noasyncreaddir, "noasyncreaddir"},
160	{Opt_dcache, "dcache"},
161	{Opt_nodcache, "nodcache"},
162	{Opt_ino32, "ino32"},
163	{Opt_noino32, "noino32"},
 
 
 
 
 
 
 
 
164	{-1, NULL}
165};
166
167static int parse_fsopt_token(char *c, void *private)
168{
169	struct ceph_mount_options *fsopt = private;
170	substring_t argstr[MAX_OPT_ARGS];
171	int token, intval, ret;
172
173	token = match_token((char *)c, fsopt_tokens, argstr);
174	if (token < 0)
175		return -EINVAL;
176
177	if (token < Opt_last_int) {
178		ret = match_int(&argstr[0], &intval);
179		if (ret < 0) {
180			pr_err("bad mount option arg (not int) "
181			       "at '%s'\n", c);
182			return ret;
183		}
184		dout("got int token %d val %d\n", token, intval);
185	} else if (token > Opt_last_int && token < Opt_last_string) {
186		dout("got string token %d val %s\n", token,
187		     argstr[0].from);
188	} else {
189		dout("got token %d\n", token);
190	}
191
192	switch (token) {
193	case Opt_snapdirname:
194		kfree(fsopt->snapdir_name);
195		fsopt->snapdir_name = kstrndup(argstr[0].from,
196					       argstr[0].to-argstr[0].from,
197					       GFP_KERNEL);
198		if (!fsopt->snapdir_name)
199			return -ENOMEM;
200		break;
201
202		/* misc */
203	case Opt_wsize:
204		fsopt->wsize = intval;
205		break;
206	case Opt_rsize:
207		fsopt->rsize = intval;
208		break;
209	case Opt_rasize:
210		fsopt->rasize = intval;
211		break;
212	case Opt_caps_wanted_delay_min:
213		fsopt->caps_wanted_delay_min = intval;
214		break;
215	case Opt_caps_wanted_delay_max:
216		fsopt->caps_wanted_delay_max = intval;
217		break;
218	case Opt_readdir_max_entries:
219		fsopt->max_readdir = intval;
220		break;
221	case Opt_readdir_max_bytes:
222		fsopt->max_readdir_bytes = intval;
223		break;
224	case Opt_congestion_kb:
225		fsopt->congestion_kb = intval;
226		break;
227	case Opt_dirstat:
228		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
229		break;
230	case Opt_nodirstat:
231		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
232		break;
233	case Opt_rbytes:
234		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
235		break;
236	case Opt_norbytes:
237		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
238		break;
239	case Opt_asyncreaddir:
240		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
241		break;
242	case Opt_noasyncreaddir:
243		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
244		break;
245	case Opt_dcache:
246		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
247		break;
248	case Opt_nodcache:
249		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
250		break;
251	case Opt_ino32:
252		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
253		break;
254	case Opt_noino32:
255		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
256		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
257	default:
258		BUG_ON(token);
259	}
260	return 0;
261}
262
263static void destroy_mount_options(struct ceph_mount_options *args)
264{
265	dout("destroy_mount_options %p\n", args);
266	kfree(args->snapdir_name);
267	kfree(args);
268}
269
270static int strcmp_null(const char *s1, const char *s2)
271{
272	if (!s1 && !s2)
273		return 0;
274	if (s1 && !s2)
275		return -1;
276	if (!s1 && s2)
277		return 1;
278	return strcmp(s1, s2);
279}
280
281static int compare_mount_options(struct ceph_mount_options *new_fsopt,
282				 struct ceph_options *new_opt,
283				 struct ceph_fs_client *fsc)
284{
285	struct ceph_mount_options *fsopt1 = new_fsopt;
286	struct ceph_mount_options *fsopt2 = fsc->mount_options;
287	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
288	int ret;
289
290	ret = memcmp(fsopt1, fsopt2, ofs);
291	if (ret)
292		return ret;
293
294	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
295	if (ret)
296		return ret;
297
298	return ceph_compare_options(new_opt, fsc->client);
299}
300
301static int parse_mount_options(struct ceph_mount_options **pfsopt,
302			       struct ceph_options **popt,
303			       int flags, char *options,
304			       const char *dev_name,
305			       const char **path)
306{
307	struct ceph_mount_options *fsopt;
308	const char *dev_name_end;
309	int err = -ENOMEM;
 
 
 
310
311	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
312	if (!fsopt)
313		return -ENOMEM;
314
315	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
316
317	fsopt->sb_flags = flags;
318	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
319
320	fsopt->rsize = CEPH_RSIZE_DEFAULT;
321	fsopt->rasize = CEPH_RASIZE_DEFAULT;
322	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
 
 
 
 
 
323	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
324	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
325	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
326	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
327	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
328	fsopt->congestion_kb = default_congestion_kb();
329
330	/* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
331	err = -EINVAL;
332	if (!dev_name)
333		goto out;
334	*path = strstr(dev_name, ":/");
335	if (*path == NULL) {
336		pr_err("device name is missing path (no :/ in %s)\n",
337				dev_name);
338		goto out;
339	}
340	dev_name_end = *path;
341	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
342
343	/* path on server */
344	*path += 2;
345	dout("server path '%s'\n", *path);
346
347	*popt = ceph_parse_options(options, dev_name, dev_name_end,
348				 parse_fsopt_token, (void *)fsopt);
349	if (IS_ERR(*popt)) {
350		err = PTR_ERR(*popt);
351		goto out;
352	}
353
354	/* success */
355	*pfsopt = fsopt;
356	return 0;
357
358out:
359	destroy_mount_options(fsopt);
360	return err;
361}
362
363/**
364 * ceph_show_options - Show mount options in /proc/mounts
365 * @m: seq_file to write to
366 * @root: root of that (sub)tree
367 */
368static int ceph_show_options(struct seq_file *m, struct dentry *root)
369{
370	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
371	struct ceph_mount_options *fsopt = fsc->mount_options;
372	struct ceph_options *opt = fsc->client->options;
 
373
374	if (opt->flags & CEPH_OPT_FSID)
375		seq_printf(m, ",fsid=%pU", &opt->fsid);
376	if (opt->flags & CEPH_OPT_NOSHARE)
377		seq_puts(m, ",noshare");
378	if (opt->flags & CEPH_OPT_NOCRC)
379		seq_puts(m, ",nocrc");
380
381	if (opt->name)
382		seq_printf(m, ",name=%s", opt->name);
383	if (opt->key)
384		seq_puts(m, ",secret=<hidden>");
385
386	if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
387		seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
388	if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
389		seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
390	if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
391		seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
392	if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
393		seq_printf(m, ",osdkeepalivetimeout=%d",
394			   opt->osd_keepalive_timeout);
395
396	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
397		seq_puts(m, ",dirstat");
398	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
399		seq_puts(m, ",norbytes");
400	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
401		seq_puts(m, ",noasyncreaddir");
402	if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
403		seq_puts(m, ",dcache");
404	else
405		seq_puts(m, ",nodcache");
 
 
 
 
 
 
 
 
 
 
 
406
407	if (fsopt->wsize)
408		seq_printf(m, ",wsize=%d", fsopt->wsize);
409	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
410		seq_printf(m, ",rsize=%d", fsopt->rsize);
411	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
412		seq_printf(m, ",rasize=%d", fsopt->rasize);
413	if (fsopt->congestion_kb != default_congestion_kb())
414		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
415	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
416		seq_printf(m, ",caps_wanted_delay_min=%d",
417			 fsopt->caps_wanted_delay_min);
418	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
419		seq_printf(m, ",caps_wanted_delay_max=%d",
420			   fsopt->caps_wanted_delay_max);
421	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
422		seq_printf(m, ",cap_release_safety=%d",
423			   fsopt->cap_release_safety);
424	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
425		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
426	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
427		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
428	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
429		seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
 
430	return 0;
431}
432
433/*
434 * handle any mon messages the standard library doesn't understand.
435 * return error if we don't either.
436 */
437static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
438{
439	struct ceph_fs_client *fsc = client->private;
440	int type = le16_to_cpu(msg->hdr.type);
441
442	switch (type) {
443	case CEPH_MSG_MDS_MAP:
444		ceph_mdsc_handle_map(fsc->mdsc, msg);
445		return 0;
446
447	default:
448		return -1;
449	}
450}
451
452/*
453 * create a new fs client
454 */
455static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
456					struct ceph_options *opt)
457{
458	struct ceph_fs_client *fsc;
459	const unsigned supported_features =
460		CEPH_FEATURE_FLOCK |
461		CEPH_FEATURE_DIRLAYOUTHASH;
462	const unsigned required_features = 0;
 
 
 
463	int err = -ENOMEM;
464
465	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
466	if (!fsc)
467		return ERR_PTR(-ENOMEM);
468
469	fsc->client = ceph_create_client(opt, fsc, supported_features,
470					 required_features);
471	if (IS_ERR(fsc->client)) {
472		err = PTR_ERR(fsc->client);
473		goto fail;
474	}
475	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
476	fsc->client->monc.want_mdsmap = 1;
477
478	fsc->mount_options = fsopt;
479
480	fsc->sb = NULL;
481	fsc->mount_state = CEPH_MOUNT_MOUNTING;
482
483	atomic_long_set(&fsc->writeback_count, 0);
484
485	err = bdi_init(&fsc->backing_dev_info);
486	if (err < 0)
487		goto fail_client;
488
489	err = -ENOMEM;
490	/*
491	 * The number of concurrent works can be high but they don't need
492	 * to be processed in parallel, limit concurrency.
493	 */
494	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
495	if (fsc->wb_wq == NULL)
496		goto fail_bdi;
497	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
498	if (fsc->pg_inv_wq == NULL)
499		goto fail_wb_wq;
500	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
501	if (fsc->trunc_wq == NULL)
502		goto fail_pg_inv_wq;
503
504	/* set up mempools */
505	err = -ENOMEM;
506	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
507			      fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
 
508	if (!fsc->wb_pagevec_pool)
509		goto fail_trunc_wq;
510
 
 
 
 
 
511	/* caps */
512	fsc->min_caps = fsopt->max_readdir;
513
514	return fsc;
515
 
 
516fail_trunc_wq:
517	destroy_workqueue(fsc->trunc_wq);
518fail_pg_inv_wq:
519	destroy_workqueue(fsc->pg_inv_wq);
520fail_wb_wq:
521	destroy_workqueue(fsc->wb_wq);
522fail_bdi:
523	bdi_destroy(&fsc->backing_dev_info);
524fail_client:
525	ceph_destroy_client(fsc->client);
526fail:
527	kfree(fsc);
528	return ERR_PTR(err);
529}
530
531static void destroy_fs_client(struct ceph_fs_client *fsc)
532{
533	dout("destroy_fs_client %p\n", fsc);
534
 
 
535	destroy_workqueue(fsc->wb_wq);
536	destroy_workqueue(fsc->pg_inv_wq);
537	destroy_workqueue(fsc->trunc_wq);
538
539	bdi_destroy(&fsc->backing_dev_info);
540
541	mempool_destroy(fsc->wb_pagevec_pool);
542
543	destroy_mount_options(fsc->mount_options);
544
545	ceph_fs_debugfs_cleanup(fsc);
546
547	ceph_destroy_client(fsc->client);
548
549	kfree(fsc);
550	dout("destroy_fs_client %p done\n", fsc);
551}
552
553/*
554 * caches
555 */
556struct kmem_cache *ceph_inode_cachep;
557struct kmem_cache *ceph_cap_cachep;
 
558struct kmem_cache *ceph_dentry_cachep;
559struct kmem_cache *ceph_file_cachep;
560
561static void ceph_inode_init_once(void *foo)
562{
563	struct ceph_inode_info *ci = foo;
564	inode_init_once(&ci->vfs_inode);
565}
566
567static int __init init_caches(void)
568{
 
 
569	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
570				      sizeof(struct ceph_inode_info),
571				      __alignof__(struct ceph_inode_info),
572				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
573				      ceph_inode_init_once);
574	if (ceph_inode_cachep == NULL)
575		return -ENOMEM;
576
577	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
578				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
579	if (ceph_cap_cachep == NULL)
580		goto bad_cap;
 
 
 
 
581
582	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
583					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
584	if (ceph_dentry_cachep == NULL)
585		goto bad_dentry;
586
587	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
588				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
589	if (ceph_file_cachep == NULL)
590		goto bad_file;
591
592	return 0;
 
593
 
594bad_file:
595	kmem_cache_destroy(ceph_dentry_cachep);
596bad_dentry:
 
 
597	kmem_cache_destroy(ceph_cap_cachep);
598bad_cap:
599	kmem_cache_destroy(ceph_inode_cachep);
600	return -ENOMEM;
601}
602
603static void destroy_caches(void)
604{
 
 
 
 
 
 
605	kmem_cache_destroy(ceph_inode_cachep);
606	kmem_cache_destroy(ceph_cap_cachep);
 
607	kmem_cache_destroy(ceph_dentry_cachep);
608	kmem_cache_destroy(ceph_file_cachep);
 
 
609}
610
611
612/*
613 * ceph_umount_begin - initiate forced umount.  Tear down down the
614 * mount, skipping steps that may hang while waiting for server(s).
615 */
616static void ceph_umount_begin(struct super_block *sb)
617{
618	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
619
620	dout("ceph_umount_begin - starting forced umount\n");
621	if (!fsc)
622		return;
623	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 
624	return;
625}
626
627static const struct super_operations ceph_super_ops = {
628	.alloc_inode	= ceph_alloc_inode,
629	.destroy_inode	= ceph_destroy_inode,
630	.write_inode    = ceph_write_inode,
 
631	.sync_fs        = ceph_sync_fs,
632	.put_super	= ceph_put_super,
633	.show_options   = ceph_show_options,
634	.statfs		= ceph_statfs,
635	.umount_begin   = ceph_umount_begin,
636};
637
638/*
639 * Bootstrap mount by opening the root directory.  Note the mount
640 * @started time from caller, and time out if this takes too long.
641 */
642static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
643				       const char *path,
644				       unsigned long started)
645{
646	struct ceph_mds_client *mdsc = fsc->mdsc;
647	struct ceph_mds_request *req = NULL;
648	int err;
649	struct dentry *root;
650
651	/* open dir */
652	dout("open_root_inode opening '%s'\n", path);
653	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
654	if (IS_ERR(req))
655		return ERR_CAST(req);
656	req->r_path1 = kstrdup(path, GFP_NOFS);
 
 
 
 
 
657	req->r_ino1.ino = CEPH_INO_ROOT;
658	req->r_ino1.snap = CEPH_NOSNAP;
659	req->r_started = started;
660	req->r_timeout = fsc->client->options->mount_timeout * HZ;
661	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
662	req->r_num_caps = 2;
663	err = ceph_mdsc_do_request(mdsc, NULL, req);
664	if (err == 0) {
665		struct inode *inode = req->r_target_inode;
666		req->r_target_inode = NULL;
667		dout("open_root_inode success\n");
668		if (ceph_ino(inode) == CEPH_INO_ROOT &&
669		    fsc->sb->s_root == NULL) {
670			root = d_make_root(inode);
671			if (!root) {
672				root = ERR_PTR(-ENOMEM);
673				goto out;
674			}
675		} else {
676			root = d_obtain_alias(inode);
677		}
678		ceph_init_dentry(root);
679		dout("open_root_inode success, root dentry is %p\n", root);
680	} else {
681		root = ERR_PTR(err);
682	}
683out:
684	ceph_mdsc_put_request(req);
685	return root;
686}
687
688
689
690
691/*
692 * mount: join the ceph cluster, and open root directory.
693 */
694static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
695		      const char *path)
696{
697	int err;
698	unsigned long started = jiffies;  /* note the start time */
699	struct dentry *root;
700	int first = 0;   /* first vfsmount for this super_block */
701
702	dout("mount start\n");
703	mutex_lock(&fsc->client->mount_mutex);
704
705	err = __ceph_open_session(fsc->client, started);
706	if (err < 0)
707		goto out;
 
708
709	dout("mount opening root\n");
710	root = open_root_dentry(fsc, "", started);
711	if (IS_ERR(root)) {
712		err = PTR_ERR(root);
713		goto out;
714	}
715	if (fsc->sb->s_root) {
716		dput(root);
717	} else {
718		fsc->sb->s_root = root;
719		first = 1;
720
721		err = ceph_fs_debugfs_init(fsc);
722		if (err < 0)
723			goto fail;
724	}
725
726	if (path[0] == 0) {
 
727		dget(root);
728	} else {
729		dout("mount opening base mountpoint\n");
730		root = open_root_dentry(fsc, path, started);
731		if (IS_ERR(root)) {
732			err = PTR_ERR(root);
733			goto fail;
734		}
735	}
736
737	fsc->mount_state = CEPH_MOUNT_MOUNTED;
738	dout("mount success\n");
739	mutex_unlock(&fsc->client->mount_mutex);
740	return root;
741
742out:
743	mutex_unlock(&fsc->client->mount_mutex);
744	return ERR_PTR(err);
745
746fail:
747	if (first) {
748		dput(fsc->sb->s_root);
749		fsc->sb->s_root = NULL;
750	}
751	goto out;
 
 
752}
753
754static int ceph_set_super(struct super_block *s, void *data)
755{
756	struct ceph_fs_client *fsc = data;
757	int ret;
758
759	dout("set_super %p data %p\n", s, data);
760
761	s->s_flags = fsc->mount_options->sb_flags;
762	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
763
 
764	s->s_fs_info = fsc;
765	fsc->sb = s;
766
767	s->s_op = &ceph_super_ops;
768	s->s_export_op = &ceph_export_ops;
769
770	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
771
772	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
773	if (ret != 0)
774		goto fail;
775
776	return ret;
777
778fail:
779	s->s_fs_info = NULL;
780	fsc->sb = NULL;
781	return ret;
782}
783
784/*
785 * share superblock if same fs AND options
786 */
787static int ceph_compare_super(struct super_block *sb, void *data)
788{
789	struct ceph_fs_client *new = data;
790	struct ceph_mount_options *fsopt = new->mount_options;
791	struct ceph_options *opt = new->client->options;
792	struct ceph_fs_client *other = ceph_sb_to_client(sb);
793
794	dout("ceph_compare_super %p\n", sb);
795
796	if (compare_mount_options(fsopt, opt, other)) {
797		dout("monitor(s)/mount options don't match\n");
798		return 0;
799	}
800	if ((opt->flags & CEPH_OPT_FSID) &&
801	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
802		dout("fsid doesn't match\n");
803		return 0;
804	}
805	if (fsopt->sb_flags != other->mount_options->sb_flags) {
806		dout("flags differ\n");
807		return 0;
808	}
809	return 1;
810}
811
812/*
813 * construct our own bdi so we can control readahead, etc.
814 */
815static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
816
817static int ceph_register_bdi(struct super_block *sb,
818			     struct ceph_fs_client *fsc)
819{
820	int err;
821
822	/* set ra_pages based on rasize mount option? */
823	if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
824		fsc->backing_dev_info.ra_pages =
825			(fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
826			>> PAGE_SHIFT;
827	else
828		fsc->backing_dev_info.ra_pages =
829			default_backing_dev_info.ra_pages;
830
831	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
832			   atomic_long_inc_return(&bdi_seq));
833	if (!err)
834		sb->s_bdi = &fsc->backing_dev_info;
835	return err;
836}
837
838static struct dentry *ceph_mount(struct file_system_type *fs_type,
839		       int flags, const char *dev_name, void *data)
840{
841	struct super_block *sb;
842	struct ceph_fs_client *fsc;
843	struct dentry *res;
844	int err;
845	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
846	const char *path = NULL;
847	struct ceph_mount_options *fsopt = NULL;
848	struct ceph_options *opt = NULL;
849
850	dout("ceph_mount\n");
 
 
 
 
851	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
852	if (err < 0) {
853		res = ERR_PTR(err);
854		goto out_final;
855	}
856
857	/* create client (which we may/may not use) */
858	fsc = create_fs_client(fsopt, opt);
859	if (IS_ERR(fsc)) {
860		res = ERR_CAST(fsc);
861		destroy_mount_options(fsopt);
862		ceph_destroy_options(opt);
863		goto out_final;
864	}
865
866	err = ceph_mdsc_init(fsc);
867	if (err < 0) {
868		res = ERR_PTR(err);
869		goto out;
870	}
871
872	if (ceph_test_opt(fsc->client, NOSHARE))
873		compare_super = NULL;
874	sb = sget(fs_type, compare_super, ceph_set_super, fsc);
875	if (IS_ERR(sb)) {
876		res = ERR_CAST(sb);
877		goto out;
878	}
879
880	if (ceph_sb_to_client(sb) != fsc) {
881		ceph_mdsc_destroy(fsc);
882		destroy_fs_client(fsc);
883		fsc = ceph_sb_to_client(sb);
884		dout("get_sb got existing client %p\n", fsc);
885	} else {
886		dout("get_sb using new client %p\n", fsc);
887		err = ceph_register_bdi(sb, fsc);
888		if (err < 0) {
889			res = ERR_PTR(err);
890			goto out_splat;
891		}
892	}
893
894	res = ceph_real_mount(fsc, path);
895	if (IS_ERR(res))
896		goto out_splat;
897	dout("root %p inode %p ino %llx.%llx\n", res,
898	     res->d_inode, ceph_vinop(res->d_inode));
899	return res;
900
901out_splat:
902	ceph_mdsc_close_sessions(fsc->mdsc);
903	deactivate_locked_super(sb);
904	goto out_final;
905
906out:
907	ceph_mdsc_destroy(fsc);
908	destroy_fs_client(fsc);
909out_final:
910	dout("ceph_mount fail %ld\n", PTR_ERR(res));
911	return res;
912}
913
914static void ceph_kill_sb(struct super_block *s)
915{
916	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
 
 
917	dout("kill_sb %p\n", s);
 
918	ceph_mdsc_pre_umount(fsc->mdsc);
919	kill_anon_super(s);    /* will call put_super after sb is r/o */
920	ceph_mdsc_destroy(fsc);
 
921	destroy_fs_client(fsc);
 
922}
923
924static struct file_system_type ceph_fs_type = {
925	.owner		= THIS_MODULE,
926	.name		= "ceph",
927	.mount		= ceph_mount,
928	.kill_sb	= ceph_kill_sb,
929	.fs_flags	= FS_RENAME_DOES_D_MOVE,
930};
931
932#define _STRINGIFY(x) #x
933#define STRINGIFY(x) _STRINGIFY(x)
934
935static int __init init_ceph(void)
936{
937	int ret = init_caches();
938	if (ret)
939		goto out;
940
 
941	ceph_xattr_init();
942	ret = register_filesystem(&ceph_fs_type);
943	if (ret)
944		goto out_icache;
945
946	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
947
948	return 0;
949
950out_icache:
951	ceph_xattr_exit();
952	destroy_caches();
953out:
954	return ret;
955}
956
957static void __exit exit_ceph(void)
958{
959	dout("exit_ceph\n");
960	unregister_filesystem(&ceph_fs_type);
961	ceph_xattr_exit();
962	destroy_caches();
963}
964
965module_init(init_ceph);
966module_exit(exit_ceph);
967
968MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
969MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
970MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
971MODULE_DESCRIPTION("Ceph filesystem for Linux");
972MODULE_LICENSE("GPL");
v4.6
   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 */
  37static 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
  45static 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
  89static 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 */
 110enum {
 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	Opt_poolperm,
 138	Opt_nopoolperm,
 139#ifdef CONFIG_CEPH_FS_POSIX_ACL
 140	Opt_acl,
 141#endif
 142	Opt_noacl,
 143};
 144
 145static match_table_t fsopt_tokens = {
 146	{Opt_wsize, "wsize=%d"},
 147	{Opt_rsize, "rsize=%d"},
 148	{Opt_rasize, "rasize=%d"},
 149	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
 150	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
 151	{Opt_cap_release_safety, "cap_release_safety=%d"},
 152	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
 153	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
 154	{Opt_congestion_kb, "write_congestion_kb=%d"},
 155	/* int args above */
 156	{Opt_snapdirname, "snapdirname=%s"},
 157	/* string args above */
 158	{Opt_dirstat, "dirstat"},
 159	{Opt_nodirstat, "nodirstat"},
 160	{Opt_rbytes, "rbytes"},
 161	{Opt_norbytes, "norbytes"},
 162	{Opt_asyncreaddir, "asyncreaddir"},
 163	{Opt_noasyncreaddir, "noasyncreaddir"},
 164	{Opt_dcache, "dcache"},
 165	{Opt_nodcache, "nodcache"},
 166	{Opt_ino32, "ino32"},
 167	{Opt_noino32, "noino32"},
 168	{Opt_fscache, "fsc"},
 169	{Opt_nofscache, "nofsc"},
 170	{Opt_poolperm, "poolperm"},
 171	{Opt_nopoolperm, "nopoolperm"},
 172#ifdef CONFIG_CEPH_FS_POSIX_ACL
 173	{Opt_acl, "acl"},
 174#endif
 175	{Opt_noacl, "noacl"},
 176	{-1, NULL}
 177};
 178
 179static int parse_fsopt_token(char *c, void *private)
 180{
 181	struct ceph_mount_options *fsopt = private;
 182	substring_t argstr[MAX_OPT_ARGS];
 183	int token, intval, ret;
 184
 185	token = match_token((char *)c, fsopt_tokens, argstr);
 186	if (token < 0)
 187		return -EINVAL;
 188
 189	if (token < Opt_last_int) {
 190		ret = match_int(&argstr[0], &intval);
 191		if (ret < 0) {
 192			pr_err("bad mount option arg (not int) "
 193			       "at '%s'\n", c);
 194			return ret;
 195		}
 196		dout("got int token %d val %d\n", token, intval);
 197	} else if (token > Opt_last_int && token < Opt_last_string) {
 198		dout("got string token %d val %s\n", token,
 199		     argstr[0].from);
 200	} else {
 201		dout("got token %d\n", token);
 202	}
 203
 204	switch (token) {
 205	case Opt_snapdirname:
 206		kfree(fsopt->snapdir_name);
 207		fsopt->snapdir_name = kstrndup(argstr[0].from,
 208					       argstr[0].to-argstr[0].from,
 209					       GFP_KERNEL);
 210		if (!fsopt->snapdir_name)
 211			return -ENOMEM;
 212		break;
 213
 214		/* misc */
 215	case Opt_wsize:
 216		fsopt->wsize = intval;
 217		break;
 218	case Opt_rsize:
 219		fsopt->rsize = intval;
 220		break;
 221	case Opt_rasize:
 222		fsopt->rasize = intval;
 223		break;
 224	case Opt_caps_wanted_delay_min:
 225		fsopt->caps_wanted_delay_min = intval;
 226		break;
 227	case Opt_caps_wanted_delay_max:
 228		fsopt->caps_wanted_delay_max = intval;
 229		break;
 230	case Opt_readdir_max_entries:
 231		fsopt->max_readdir = intval;
 232		break;
 233	case Opt_readdir_max_bytes:
 234		fsopt->max_readdir_bytes = intval;
 235		break;
 236	case Opt_congestion_kb:
 237		fsopt->congestion_kb = intval;
 238		break;
 239	case Opt_dirstat:
 240		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 241		break;
 242	case Opt_nodirstat:
 243		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 244		break;
 245	case Opt_rbytes:
 246		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 247		break;
 248	case Opt_norbytes:
 249		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 250		break;
 251	case Opt_asyncreaddir:
 252		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 253		break;
 254	case Opt_noasyncreaddir:
 255		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 256		break;
 257	case Opt_dcache:
 258		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 259		break;
 260	case Opt_nodcache:
 261		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 262		break;
 263	case Opt_ino32:
 264		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 265		break;
 266	case Opt_noino32:
 267		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 268		break;
 269	case Opt_fscache:
 270		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 271		break;
 272	case Opt_nofscache:
 273		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 274		break;
 275	case Opt_poolperm:
 276		fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 277		printk ("pool perm");
 278		break;
 279	case Opt_nopoolperm:
 280		fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 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
 296static 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
 303static 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
 314static 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
 334static 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	if (!fsopt->snapdir_name) {
 360		err = -ENOMEM;
 361		goto out;
 362	}
 363
 364	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
 365	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
 366	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
 367	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
 368	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
 369	fsopt->congestion_kb = default_congestion_kb();
 370
 371	/*
 372	 * Distinguish the server list from the path in "dev_name".
 373	 * Internally we do not include the leading '/' in the path.
 374	 *
 375	 * "dev_name" will look like:
 376	 *     <server_spec>[,<server_spec>...]:[<path>]
 377	 * where
 378	 *     <server_spec> is <ip>[:<port>]
 379	 *     <path> is optional, but if present must begin with '/'
 380	 */
 381	dev_name_end = strchr(dev_name, '/');
 382	if (dev_name_end) {
 383		/* skip over leading '/' for path */
 384		*path = dev_name_end + 1;
 385	} else {
 386		/* path is empty */
 387		dev_name_end = dev_name + strlen(dev_name);
 388		*path = dev_name_end;
 389	}
 390	err = -EINVAL;
 391	dev_name_end--;		/* back up to ':' separator */
 392	if (dev_name_end < dev_name || *dev_name_end != ':') {
 393		pr_err("device name is missing path (no : separator in %s)\n",
 
 
 394				dev_name);
 395		goto out;
 396	}
 
 397	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 
 
 
 398	dout("server path '%s'\n", *path);
 399
 400	*popt = ceph_parse_options(options, dev_name, dev_name_end,
 401				 parse_fsopt_token, (void *)fsopt);
 402	if (IS_ERR(*popt)) {
 403		err = PTR_ERR(*popt);
 404		goto out;
 405	}
 406
 407	/* success */
 408	*pfsopt = fsopt;
 409	return 0;
 410
 411out:
 412	destroy_mount_options(fsopt);
 413	return err;
 414}
 415
 416/**
 417 * ceph_show_options - Show mount options in /proc/mounts
 418 * @m: seq_file to write to
 419 * @root: root of that (sub)tree
 420 */
 421static int ceph_show_options(struct seq_file *m, struct dentry *root)
 422{
 423	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 424	struct ceph_mount_options *fsopt = fsc->mount_options;
 425	size_t pos;
 426	int ret;
 427
 428	/* a comma between MNT/MS and client options */
 429	seq_putc(m, ',');
 430	pos = m->count;
 431
 432	ret = ceph_print_client_options(m, fsc->client);
 433	if (ret)
 434		return ret;
 435
 436	/* retract our comma if no client options */
 437	if (m->count == pos)
 438		m->count--;
 
 
 
 
 
 
 
 
 
 
 439
 440	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 441		seq_puts(m, ",dirstat");
 442	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 443		seq_puts(m, ",rbytes");
 444	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 445		seq_puts(m, ",noasyncreaddir");
 446	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 
 
 447		seq_puts(m, ",nodcache");
 448	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
 449		seq_puts(m, ",fsc");
 450	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 451		seq_puts(m, ",nopoolperm");
 452
 453#ifdef CONFIG_CEPH_FS_POSIX_ACL
 454	if (fsopt->sb_flags & MS_POSIXACL)
 455		seq_puts(m, ",acl");
 456	else
 457		seq_puts(m, ",noacl");
 458#endif
 459
 460	if (fsopt->wsize)
 461		seq_printf(m, ",wsize=%d", fsopt->wsize);
 462	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
 463		seq_printf(m, ",rsize=%d", fsopt->rsize);
 464	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 465		seq_printf(m, ",rasize=%d", fsopt->rasize);
 466	if (fsopt->congestion_kb != default_congestion_kb())
 467		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
 468	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 469		seq_printf(m, ",caps_wanted_delay_min=%d",
 470			 fsopt->caps_wanted_delay_min);
 471	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 472		seq_printf(m, ",caps_wanted_delay_max=%d",
 473			   fsopt->caps_wanted_delay_max);
 474	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
 475		seq_printf(m, ",cap_release_safety=%d",
 476			   fsopt->cap_release_safety);
 477	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 478		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
 479	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 480		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
 481	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 482		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 483
 484	return 0;
 485}
 486
 487/*
 488 * handle any mon messages the standard library doesn't understand.
 489 * return error if we don't either.
 490 */
 491static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 492{
 493	struct ceph_fs_client *fsc = client->private;
 494	int type = le16_to_cpu(msg->hdr.type);
 495
 496	switch (type) {
 497	case CEPH_MSG_MDS_MAP:
 498		ceph_mdsc_handle_map(fsc->mdsc, msg);
 499		return 0;
 500
 501	default:
 502		return -1;
 503	}
 504}
 505
 506/*
 507 * create a new fs client
 508 */
 509static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 510					struct ceph_options *opt)
 511{
 512	struct ceph_fs_client *fsc;
 513	const u64 supported_features =
 514		CEPH_FEATURE_FLOCK |
 515		CEPH_FEATURE_DIRLAYOUTHASH |
 516		CEPH_FEATURE_MDS_INLINE_DATA;
 517	const u64 required_features = 0;
 518	int page_count;
 519	size_t size;
 520	int err = -ENOMEM;
 521
 522	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 523	if (!fsc)
 524		return ERR_PTR(-ENOMEM);
 525
 526	fsc->client = ceph_create_client(opt, fsc, supported_features,
 527					 required_features);
 528	if (IS_ERR(fsc->client)) {
 529		err = PTR_ERR(fsc->client);
 530		goto fail;
 531	}
 532	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 533	ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP, 0, true);
 534
 535	fsc->mount_options = fsopt;
 536
 537	fsc->sb = NULL;
 538	fsc->mount_state = CEPH_MOUNT_MOUNTING;
 539
 540	atomic_long_set(&fsc->writeback_count, 0);
 541
 542	err = bdi_init(&fsc->backing_dev_info);
 543	if (err < 0)
 544		goto fail_client;
 545
 546	err = -ENOMEM;
 547	/*
 548	 * The number of concurrent works can be high but they don't need
 549	 * to be processed in parallel, limit concurrency.
 550	 */
 551	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
 552	if (fsc->wb_wq == NULL)
 553		goto fail_bdi;
 554	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
 555	if (fsc->pg_inv_wq == NULL)
 556		goto fail_wb_wq;
 557	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
 558	if (fsc->trunc_wq == NULL)
 559		goto fail_pg_inv_wq;
 560
 561	/* set up mempools */
 562	err = -ENOMEM;
 563	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
 564	size = sizeof (struct page *) * (page_count ? page_count : 1);
 565	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
 566	if (!fsc->wb_pagevec_pool)
 567		goto fail_trunc_wq;
 568
 569	/* setup fscache */
 570	if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
 571	    (ceph_fscache_register_fs(fsc) != 0))
 572		goto fail_fscache;
 573
 574	/* caps */
 575	fsc->min_caps = fsopt->max_readdir;
 576
 577	return fsc;
 578
 579fail_fscache:
 580	ceph_fscache_unregister_fs(fsc);
 581fail_trunc_wq:
 582	destroy_workqueue(fsc->trunc_wq);
 583fail_pg_inv_wq:
 584	destroy_workqueue(fsc->pg_inv_wq);
 585fail_wb_wq:
 586	destroy_workqueue(fsc->wb_wq);
 587fail_bdi:
 588	bdi_destroy(&fsc->backing_dev_info);
 589fail_client:
 590	ceph_destroy_client(fsc->client);
 591fail:
 592	kfree(fsc);
 593	return ERR_PTR(err);
 594}
 595
 596static void destroy_fs_client(struct ceph_fs_client *fsc)
 597{
 598	dout("destroy_fs_client %p\n", fsc);
 599
 600	ceph_fscache_unregister_fs(fsc);
 601
 602	destroy_workqueue(fsc->wb_wq);
 603	destroy_workqueue(fsc->pg_inv_wq);
 604	destroy_workqueue(fsc->trunc_wq);
 605
 606	bdi_destroy(&fsc->backing_dev_info);
 607
 608	mempool_destroy(fsc->wb_pagevec_pool);
 609
 610	destroy_mount_options(fsc->mount_options);
 611
 612	ceph_fs_debugfs_cleanup(fsc);
 613
 614	ceph_destroy_client(fsc->client);
 615
 616	kfree(fsc);
 617	dout("destroy_fs_client %p done\n", fsc);
 618}
 619
 620/*
 621 * caches
 622 */
 623struct kmem_cache *ceph_inode_cachep;
 624struct kmem_cache *ceph_cap_cachep;
 625struct kmem_cache *ceph_cap_flush_cachep;
 626struct kmem_cache *ceph_dentry_cachep;
 627struct kmem_cache *ceph_file_cachep;
 628
 629static void ceph_inode_init_once(void *foo)
 630{
 631	struct ceph_inode_info *ci = foo;
 632	inode_init_once(&ci->vfs_inode);
 633}
 634
 635static int __init init_caches(void)
 636{
 637	int error = -ENOMEM;
 638
 639	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 640				      sizeof(struct ceph_inode_info),
 641				      __alignof__(struct ceph_inode_info),
 642				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 643				      SLAB_ACCOUNT, ceph_inode_init_once);
 644	if (ceph_inode_cachep == NULL)
 645		return -ENOMEM;
 646
 647	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
 648				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 649	if (ceph_cap_cachep == NULL)
 650		goto bad_cap;
 651	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 652					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 653	if (ceph_cap_flush_cachep == NULL)
 654		goto bad_cap_flush;
 655
 656	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 657					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 658	if (ceph_dentry_cachep == NULL)
 659		goto bad_dentry;
 660
 661	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
 662				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 663	if (ceph_file_cachep == NULL)
 664		goto bad_file;
 665
 666	if ((error = ceph_fscache_register()))
 667		goto bad_file;
 668
 669	return 0;
 670bad_file:
 671	kmem_cache_destroy(ceph_dentry_cachep);
 672bad_dentry:
 673	kmem_cache_destroy(ceph_cap_flush_cachep);
 674bad_cap_flush:
 675	kmem_cache_destroy(ceph_cap_cachep);
 676bad_cap:
 677	kmem_cache_destroy(ceph_inode_cachep);
 678	return error;
 679}
 680
 681static void destroy_caches(void)
 682{
 683	/*
 684	 * Make sure all delayed rcu free inodes are flushed before we
 685	 * destroy cache.
 686	 */
 687	rcu_barrier();
 688
 689	kmem_cache_destroy(ceph_inode_cachep);
 690	kmem_cache_destroy(ceph_cap_cachep);
 691	kmem_cache_destroy(ceph_cap_flush_cachep);
 692	kmem_cache_destroy(ceph_dentry_cachep);
 693	kmem_cache_destroy(ceph_file_cachep);
 694
 695	ceph_fscache_unregister();
 696}
 697
 698
 699/*
 700 * ceph_umount_begin - initiate forced umount.  Tear down down the
 701 * mount, skipping steps that may hang while waiting for server(s).
 702 */
 703static void ceph_umount_begin(struct super_block *sb)
 704{
 705	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 706
 707	dout("ceph_umount_begin - starting forced umount\n");
 708	if (!fsc)
 709		return;
 710	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 711	ceph_mdsc_force_umount(fsc->mdsc);
 712	return;
 713}
 714
 715static const struct super_operations ceph_super_ops = {
 716	.alloc_inode	= ceph_alloc_inode,
 717	.destroy_inode	= ceph_destroy_inode,
 718	.write_inode    = ceph_write_inode,
 719	.drop_inode	= ceph_drop_inode,
 720	.sync_fs        = ceph_sync_fs,
 721	.put_super	= ceph_put_super,
 722	.show_options   = ceph_show_options,
 723	.statfs		= ceph_statfs,
 724	.umount_begin   = ceph_umount_begin,
 725};
 726
 727/*
 728 * Bootstrap mount by opening the root directory.  Note the mount
 729 * @started time from caller, and time out if this takes too long.
 730 */
 731static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
 732				       const char *path,
 733				       unsigned long started)
 734{
 735	struct ceph_mds_client *mdsc = fsc->mdsc;
 736	struct ceph_mds_request *req = NULL;
 737	int err;
 738	struct dentry *root;
 739
 740	/* open dir */
 741	dout("open_root_inode opening '%s'\n", path);
 742	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
 743	if (IS_ERR(req))
 744		return ERR_CAST(req);
 745	req->r_path1 = kstrdup(path, GFP_NOFS);
 746	if (!req->r_path1) {
 747		root = ERR_PTR(-ENOMEM);
 748		goto out;
 749	}
 750
 751	req->r_ino1.ino = CEPH_INO_ROOT;
 752	req->r_ino1.snap = CEPH_NOSNAP;
 753	req->r_started = started;
 754	req->r_timeout = fsc->client->options->mount_timeout;
 755	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 756	req->r_num_caps = 2;
 757	err = ceph_mdsc_do_request(mdsc, NULL, req);
 758	if (err == 0) {
 759		struct inode *inode = req->r_target_inode;
 760		req->r_target_inode = NULL;
 761		dout("open_root_inode success\n");
 762		if (ceph_ino(inode) == CEPH_INO_ROOT &&
 763		    fsc->sb->s_root == NULL) {
 764			root = d_make_root(inode);
 765			if (!root) {
 766				root = ERR_PTR(-ENOMEM);
 767				goto out;
 768			}
 769		} else {
 770			root = d_obtain_root(inode);
 771		}
 772		ceph_init_dentry(root);
 773		dout("open_root_inode success, root dentry is %p\n", root);
 774	} else {
 775		root = ERR_PTR(err);
 776	}
 777out:
 778	ceph_mdsc_put_request(req);
 779	return root;
 780}
 781
 782
 783
 784
 785/*
 786 * mount: join the ceph cluster, and open root directory.
 787 */
 788static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
 789		      const char *path)
 790{
 791	int err;
 792	unsigned long started = jiffies;  /* note the start time */
 793	struct dentry *root;
 794	int first = 0;   /* first vfsmount for this super_block */
 795
 796	dout("mount start %p\n", fsc);
 797	mutex_lock(&fsc->client->mount_mutex);
 798
 799	if (!fsc->sb->s_root) {
 800		err = __ceph_open_session(fsc->client, started);
 801		if (err < 0)
 802			goto out;
 803
 804		dout("mount opening root\n");
 805		root = open_root_dentry(fsc, "", started);
 806		if (IS_ERR(root)) {
 807			err = PTR_ERR(root);
 808			goto out;
 809		}
 
 
 
 810		fsc->sb->s_root = root;
 811		first = 1;
 812
 813		err = ceph_fs_debugfs_init(fsc);
 814		if (err < 0)
 815			goto fail;
 816	}
 817
 818	if (path[0] == 0) {
 819		root = fsc->sb->s_root;
 820		dget(root);
 821	} else {
 822		dout("mount opening base mountpoint\n");
 823		root = open_root_dentry(fsc, path, started);
 824		if (IS_ERR(root)) {
 825			err = PTR_ERR(root);
 826			goto fail;
 827		}
 828	}
 829
 830	fsc->mount_state = CEPH_MOUNT_MOUNTED;
 831	dout("mount success\n");
 832	mutex_unlock(&fsc->client->mount_mutex);
 833	return root;
 834
 
 
 
 
 835fail:
 836	if (first) {
 837		dput(fsc->sb->s_root);
 838		fsc->sb->s_root = NULL;
 839	}
 840out:
 841	mutex_unlock(&fsc->client->mount_mutex);
 842	return ERR_PTR(err);
 843}
 844
 845static int ceph_set_super(struct super_block *s, void *data)
 846{
 847	struct ceph_fs_client *fsc = data;
 848	int ret;
 849
 850	dout("set_super %p data %p\n", s, data);
 851
 852	s->s_flags = fsc->mount_options->sb_flags;
 853	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
 854
 855	s->s_xattr = ceph_xattr_handlers;
 856	s->s_fs_info = fsc;
 857	fsc->sb = s;
 858
 859	s->s_op = &ceph_super_ops;
 860	s->s_export_op = &ceph_export_ops;
 861
 862	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
 863
 864	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
 865	if (ret != 0)
 866		goto fail;
 867
 868	return ret;
 869
 870fail:
 871	s->s_fs_info = NULL;
 872	fsc->sb = NULL;
 873	return ret;
 874}
 875
 876/*
 877 * share superblock if same fs AND options
 878 */
 879static int ceph_compare_super(struct super_block *sb, void *data)
 880{
 881	struct ceph_fs_client *new = data;
 882	struct ceph_mount_options *fsopt = new->mount_options;
 883	struct ceph_options *opt = new->client->options;
 884	struct ceph_fs_client *other = ceph_sb_to_client(sb);
 885
 886	dout("ceph_compare_super %p\n", sb);
 887
 888	if (compare_mount_options(fsopt, opt, other)) {
 889		dout("monitor(s)/mount options don't match\n");
 890		return 0;
 891	}
 892	if ((opt->flags & CEPH_OPT_FSID) &&
 893	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
 894		dout("fsid doesn't match\n");
 895		return 0;
 896	}
 897	if (fsopt->sb_flags != other->mount_options->sb_flags) {
 898		dout("flags differ\n");
 899		return 0;
 900	}
 901	return 1;
 902}
 903
 904/*
 905 * construct our own bdi so we can control readahead, etc.
 906 */
 907static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 908
 909static int ceph_register_bdi(struct super_block *sb,
 910			     struct ceph_fs_client *fsc)
 911{
 912	int err;
 913
 914	/* set ra_pages based on rasize mount option? */
 915	if (fsc->mount_options->rasize >= PAGE_SIZE)
 916		fsc->backing_dev_info.ra_pages =
 917			(fsc->mount_options->rasize + PAGE_SIZE - 1)
 918			>> PAGE_SHIFT;
 919	else
 920		fsc->backing_dev_info.ra_pages =
 921			VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
 922
 923	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
 924			   atomic_long_inc_return(&bdi_seq));
 925	if (!err)
 926		sb->s_bdi = &fsc->backing_dev_info;
 927	return err;
 928}
 929
 930static struct dentry *ceph_mount(struct file_system_type *fs_type,
 931		       int flags, const char *dev_name, void *data)
 932{
 933	struct super_block *sb;
 934	struct ceph_fs_client *fsc;
 935	struct dentry *res;
 936	int err;
 937	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
 938	const char *path = NULL;
 939	struct ceph_mount_options *fsopt = NULL;
 940	struct ceph_options *opt = NULL;
 941
 942	dout("ceph_mount\n");
 943
 944#ifdef CONFIG_CEPH_FS_POSIX_ACL
 945	flags |= MS_POSIXACL;
 946#endif
 947	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
 948	if (err < 0) {
 949		res = ERR_PTR(err);
 950		goto out_final;
 951	}
 952
 953	/* create client (which we may/may not use) */
 954	fsc = create_fs_client(fsopt, opt);
 955	if (IS_ERR(fsc)) {
 956		res = ERR_CAST(fsc);
 957		destroy_mount_options(fsopt);
 958		ceph_destroy_options(opt);
 959		goto out_final;
 960	}
 961
 962	err = ceph_mdsc_init(fsc);
 963	if (err < 0) {
 964		res = ERR_PTR(err);
 965		goto out;
 966	}
 967
 968	if (ceph_test_opt(fsc->client, NOSHARE))
 969		compare_super = NULL;
 970	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
 971	if (IS_ERR(sb)) {
 972		res = ERR_CAST(sb);
 973		goto out;
 974	}
 975
 976	if (ceph_sb_to_client(sb) != fsc) {
 977		ceph_mdsc_destroy(fsc);
 978		destroy_fs_client(fsc);
 979		fsc = ceph_sb_to_client(sb);
 980		dout("get_sb got existing client %p\n", fsc);
 981	} else {
 982		dout("get_sb using new client %p\n", fsc);
 983		err = ceph_register_bdi(sb, fsc);
 984		if (err < 0) {
 985			res = ERR_PTR(err);
 986			goto out_splat;
 987		}
 988	}
 989
 990	res = ceph_real_mount(fsc, path);
 991	if (IS_ERR(res))
 992		goto out_splat;
 993	dout("root %p inode %p ino %llx.%llx\n", res,
 994	     d_inode(res), ceph_vinop(d_inode(res)));
 995	return res;
 996
 997out_splat:
 998	ceph_mdsc_close_sessions(fsc->mdsc);
 999	deactivate_locked_super(sb);
1000	goto out_final;
1001
1002out:
1003	ceph_mdsc_destroy(fsc);
1004	destroy_fs_client(fsc);
1005out_final:
1006	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1007	return res;
1008}
1009
1010static void ceph_kill_sb(struct super_block *s)
1011{
1012	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1013	dev_t dev = s->s_dev;
1014
1015	dout("kill_sb %p\n", s);
1016
1017	ceph_mdsc_pre_umount(fsc->mdsc);
1018	generic_shutdown_super(s);
1019	ceph_mdsc_destroy(fsc);
1020
1021	destroy_fs_client(fsc);
1022	free_anon_bdev(dev);
1023}
1024
1025static struct file_system_type ceph_fs_type = {
1026	.owner		= THIS_MODULE,
1027	.name		= "ceph",
1028	.mount		= ceph_mount,
1029	.kill_sb	= ceph_kill_sb,
1030	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1031};
1032MODULE_ALIAS_FS("ceph");
 
 
1033
1034static int __init init_ceph(void)
1035{
1036	int ret = init_caches();
1037	if (ret)
1038		goto out;
1039
1040	ceph_flock_init();
1041	ceph_xattr_init();
1042	ret = register_filesystem(&ceph_fs_type);
1043	if (ret)
1044		goto out_xattr;
1045
1046	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1047
1048	return 0;
1049
1050out_xattr:
1051	ceph_xattr_exit();
1052	destroy_caches();
1053out:
1054	return ret;
1055}
1056
1057static void __exit exit_ceph(void)
1058{
1059	dout("exit_ceph\n");
1060	unregister_filesystem(&ceph_fs_type);
1061	ceph_xattr_exit();
1062	destroy_caches();
1063}
1064
1065module_init(init_ceph);
1066module_exit(exit_ceph);
1067
1068MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1069MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1070MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1071MODULE_DESCRIPTION("Ceph filesystem for Linux");
1072MODULE_LICENSE("GPL");