xref: /openbmc/linux/fs/ceph/super.c (revision 4800cd83)
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->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
294 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
295         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
296         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
297         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
298         fsopt->congestion_kb = default_congestion_kb();
299 
300         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
301         err = -EINVAL;
302         if (!dev_name)
303                 goto out;
304         *path = strstr(dev_name, ":/");
305         if (*path == NULL) {
306                 pr_err("device name is missing path (no :/ in %s)\n",
307                        dev_name);
308                 goto out;
309         }
310 	dev_name_end = *path;
311 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
312 
313 	/* path on server */
314 	*path += 2;
315 	dout("server path '%s'\n", *path);
316 
317 	err = ceph_parse_options(popt, options, dev_name, dev_name_end,
318 				 parse_fsopt_token, (void *)fsopt);
319 	if (err)
320 		goto out;
321 
322 	/* success */
323 	*pfsopt = fsopt;
324 	return 0;
325 
326 out:
327 	destroy_mount_options(fsopt);
328 	return err;
329 }
330 
331 /**
332  * ceph_show_options - Show mount options in /proc/mounts
333  * @m: seq_file to write to
334  * @mnt: mount descriptor
335  */
336 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
337 {
338 	struct ceph_fs_client *fsc = ceph_sb_to_client(mnt->mnt_sb);
339 	struct ceph_mount_options *fsopt = fsc->mount_options;
340 	struct ceph_options *opt = fsc->client->options;
341 
342 	if (opt->flags & CEPH_OPT_FSID)
343 		seq_printf(m, ",fsid=%pU", &opt->fsid);
344 	if (opt->flags & CEPH_OPT_NOSHARE)
345 		seq_puts(m, ",noshare");
346 	if (opt->flags & CEPH_OPT_NOCRC)
347 		seq_puts(m, ",nocrc");
348 
349 	if (opt->name)
350 		seq_printf(m, ",name=%s", opt->name);
351 	if (opt->secret)
352 		seq_puts(m, ",secret=<hidden>");
353 
354 	if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
355 		seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
356 	if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
357 		seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
358 	if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
359 		seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
360 	if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
361 		seq_printf(m, ",osdkeepalivetimeout=%d",
362 			   opt->osd_keepalive_timeout);
363 
364 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
365 		seq_puts(m, ",dirstat");
366 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
367 		seq_puts(m, ",norbytes");
368 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
369 		seq_puts(m, ",noasyncreaddir");
370 
371 	if (fsopt->wsize)
372 		seq_printf(m, ",wsize=%d", fsopt->wsize);
373 	if (fsopt->rsize != CEPH_MOUNT_RSIZE_DEFAULT)
374 		seq_printf(m, ",rsize=%d", fsopt->rsize);
375 	if (fsopt->congestion_kb != default_congestion_kb())
376 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
377 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
378 		seq_printf(m, ",caps_wanted_delay_min=%d",
379 			 fsopt->caps_wanted_delay_min);
380 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
381 		seq_printf(m, ",caps_wanted_delay_max=%d",
382 			   fsopt->caps_wanted_delay_max);
383 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
384 		seq_printf(m, ",cap_release_safety=%d",
385 			   fsopt->cap_release_safety);
386 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
387 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
388 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
389 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
390 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
391 		seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
392 	return 0;
393 }
394 
395 /*
396  * handle any mon messages the standard library doesn't understand.
397  * return error if we don't either.
398  */
399 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
400 {
401 	struct ceph_fs_client *fsc = client->private;
402 	int type = le16_to_cpu(msg->hdr.type);
403 
404 	switch (type) {
405 	case CEPH_MSG_MDS_MAP:
406 		ceph_mdsc_handle_map(fsc->mdsc, msg);
407 		return 0;
408 
409 	default:
410 		return -1;
411 	}
412 }
413 
414 /*
415  * create a new fs client
416  */
417 struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
418 					struct ceph_options *opt)
419 {
420 	struct ceph_fs_client *fsc;
421 	int err = -ENOMEM;
422 
423 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
424 	if (!fsc)
425 		return ERR_PTR(-ENOMEM);
426 
427 	fsc->client = ceph_create_client(opt, fsc);
428 	if (IS_ERR(fsc->client)) {
429 		err = PTR_ERR(fsc->client);
430 		goto fail;
431 	}
432 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
433 	fsc->client->supported_features |= CEPH_FEATURE_FLOCK |
434 		CEPH_FEATURE_DIRLAYOUTHASH;
435 	fsc->client->monc.want_mdsmap = 1;
436 
437 	fsc->mount_options = fsopt;
438 
439 	fsc->sb = NULL;
440 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
441 
442 	atomic_long_set(&fsc->writeback_count, 0);
443 
444 	err = bdi_init(&fsc->backing_dev_info);
445 	if (err < 0)
446 		goto fail_client;
447 
448 	err = -ENOMEM;
449 	/*
450 	 * The number of concurrent works can be high but they don't need
451 	 * to be processed in parallel, limit concurrency.
452 	 */
453 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
454 	if (fsc->wb_wq == NULL)
455 		goto fail_bdi;
456 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
457 	if (fsc->pg_inv_wq == NULL)
458 		goto fail_wb_wq;
459 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
460 	if (fsc->trunc_wq == NULL)
461 		goto fail_pg_inv_wq;
462 
463 	/* set up mempools */
464 	err = -ENOMEM;
465 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
466 			      fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
467 	if (!fsc->wb_pagevec_pool)
468 		goto fail_trunc_wq;
469 
470 	/* caps */
471 	fsc->min_caps = fsopt->max_readdir;
472 
473 	return fsc;
474 
475 fail_trunc_wq:
476 	destroy_workqueue(fsc->trunc_wq);
477 fail_pg_inv_wq:
478 	destroy_workqueue(fsc->pg_inv_wq);
479 fail_wb_wq:
480 	destroy_workqueue(fsc->wb_wq);
481 fail_bdi:
482 	bdi_destroy(&fsc->backing_dev_info);
483 fail_client:
484 	ceph_destroy_client(fsc->client);
485 fail:
486 	kfree(fsc);
487 	return ERR_PTR(err);
488 }
489 
490 void destroy_fs_client(struct ceph_fs_client *fsc)
491 {
492 	dout("destroy_fs_client %p\n", fsc);
493 
494 	destroy_workqueue(fsc->wb_wq);
495 	destroy_workqueue(fsc->pg_inv_wq);
496 	destroy_workqueue(fsc->trunc_wq);
497 
498 	bdi_destroy(&fsc->backing_dev_info);
499 
500 	mempool_destroy(fsc->wb_pagevec_pool);
501 
502 	destroy_mount_options(fsc->mount_options);
503 
504 	ceph_fs_debugfs_cleanup(fsc);
505 
506 	ceph_destroy_client(fsc->client);
507 
508 	kfree(fsc);
509 	dout("destroy_fs_client %p done\n", fsc);
510 }
511 
512 /*
513  * caches
514  */
515 struct kmem_cache *ceph_inode_cachep;
516 struct kmem_cache *ceph_cap_cachep;
517 struct kmem_cache *ceph_dentry_cachep;
518 struct kmem_cache *ceph_file_cachep;
519 
520 static void ceph_inode_init_once(void *foo)
521 {
522 	struct ceph_inode_info *ci = foo;
523 	inode_init_once(&ci->vfs_inode);
524 }
525 
526 static int __init init_caches(void)
527 {
528 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
529 				      sizeof(struct ceph_inode_info),
530 				      __alignof__(struct ceph_inode_info),
531 				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
532 				      ceph_inode_init_once);
533 	if (ceph_inode_cachep == NULL)
534 		return -ENOMEM;
535 
536 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
537 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
538 	if (ceph_cap_cachep == NULL)
539 		goto bad_cap;
540 
541 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
542 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
543 	if (ceph_dentry_cachep == NULL)
544 		goto bad_dentry;
545 
546 	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
547 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
548 	if (ceph_file_cachep == NULL)
549 		goto bad_file;
550 
551 	return 0;
552 
553 bad_file:
554 	kmem_cache_destroy(ceph_dentry_cachep);
555 bad_dentry:
556 	kmem_cache_destroy(ceph_cap_cachep);
557 bad_cap:
558 	kmem_cache_destroy(ceph_inode_cachep);
559 	return -ENOMEM;
560 }
561 
562 static void destroy_caches(void)
563 {
564 	kmem_cache_destroy(ceph_inode_cachep);
565 	kmem_cache_destroy(ceph_cap_cachep);
566 	kmem_cache_destroy(ceph_dentry_cachep);
567 	kmem_cache_destroy(ceph_file_cachep);
568 }
569 
570 
571 /*
572  * ceph_umount_begin - initiate forced umount.  Tear down down the
573  * mount, skipping steps that may hang while waiting for server(s).
574  */
575 static void ceph_umount_begin(struct super_block *sb)
576 {
577 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
578 
579 	dout("ceph_umount_begin - starting forced umount\n");
580 	if (!fsc)
581 		return;
582 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
583 	return;
584 }
585 
586 static const struct super_operations ceph_super_ops = {
587 	.alloc_inode	= ceph_alloc_inode,
588 	.destroy_inode	= ceph_destroy_inode,
589 	.write_inode    = ceph_write_inode,
590 	.sync_fs        = ceph_sync_fs,
591 	.put_super	= ceph_put_super,
592 	.show_options   = ceph_show_options,
593 	.statfs		= ceph_statfs,
594 	.umount_begin   = ceph_umount_begin,
595 };
596 
597 /*
598  * Bootstrap mount by opening the root directory.  Note the mount
599  * @started time from caller, and time out if this takes too long.
600  */
601 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
602 				       const char *path,
603 				       unsigned long started)
604 {
605 	struct ceph_mds_client *mdsc = fsc->mdsc;
606 	struct ceph_mds_request *req = NULL;
607 	int err;
608 	struct dentry *root;
609 
610 	/* open dir */
611 	dout("open_root_inode opening '%s'\n", path);
612 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
613 	if (IS_ERR(req))
614 		return ERR_CAST(req);
615 	req->r_path1 = kstrdup(path, GFP_NOFS);
616 	req->r_ino1.ino = CEPH_INO_ROOT;
617 	req->r_ino1.snap = CEPH_NOSNAP;
618 	req->r_started = started;
619 	req->r_timeout = fsc->client->options->mount_timeout * HZ;
620 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
621 	req->r_num_caps = 2;
622 	err = ceph_mdsc_do_request(mdsc, NULL, req);
623 	if (err == 0) {
624 		dout("open_root_inode success\n");
625 		if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
626 		    fsc->sb->s_root == NULL)
627 			root = d_alloc_root(req->r_target_inode);
628 		else
629 			root = d_obtain_alias(req->r_target_inode);
630 		req->r_target_inode = NULL;
631 		dout("open_root_inode success, root dentry is %p\n", root);
632 	} else {
633 		root = ERR_PTR(err);
634 	}
635 	ceph_mdsc_put_request(req);
636 	return root;
637 }
638 
639 
640 
641 
642 /*
643  * mount: join the ceph cluster, and open root directory.
644  */
645 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
646 		      const char *path)
647 {
648 	int err;
649 	unsigned long started = jiffies;  /* note the start time */
650 	struct dentry *root;
651 	int first = 0;   /* first vfsmount for this super_block */
652 
653 	dout("mount start\n");
654 	mutex_lock(&fsc->client->mount_mutex);
655 
656 	err = __ceph_open_session(fsc->client, started);
657 	if (err < 0)
658 		goto out;
659 
660 	dout("mount opening root\n");
661 	root = open_root_dentry(fsc, "", started);
662 	if (IS_ERR(root)) {
663 		err = PTR_ERR(root);
664 		goto out;
665 	}
666 	if (fsc->sb->s_root) {
667 		dput(root);
668 	} else {
669 		fsc->sb->s_root = root;
670 		first = 1;
671 
672 		err = ceph_fs_debugfs_init(fsc);
673 		if (err < 0)
674 			goto fail;
675 	}
676 
677 	if (path[0] == 0) {
678 		dget(root);
679 	} else {
680 		dout("mount opening base mountpoint\n");
681 		root = open_root_dentry(fsc, path, started);
682 		if (IS_ERR(root)) {
683 			err = PTR_ERR(root);
684 			goto fail;
685 		}
686 	}
687 
688 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
689 	dout("mount success\n");
690 	mutex_unlock(&fsc->client->mount_mutex);
691 	return root;
692 
693 out:
694 	mutex_unlock(&fsc->client->mount_mutex);
695 	return ERR_PTR(err);
696 
697 fail:
698 	if (first) {
699 		dput(fsc->sb->s_root);
700 		fsc->sb->s_root = NULL;
701 	}
702 	goto out;
703 }
704 
705 static int ceph_set_super(struct super_block *s, void *data)
706 {
707 	struct ceph_fs_client *fsc = data;
708 	int ret;
709 
710 	dout("set_super %p data %p\n", s, data);
711 
712 	s->s_flags = fsc->mount_options->sb_flags;
713 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
714 
715 	s->s_fs_info = fsc;
716 	fsc->sb = s;
717 
718 	s->s_op = &ceph_super_ops;
719 	s->s_export_op = &ceph_export_ops;
720 
721 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
722 
723 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
724 	if (ret != 0)
725 		goto fail;
726 
727 	return ret;
728 
729 fail:
730 	s->s_fs_info = NULL;
731 	fsc->sb = NULL;
732 	return ret;
733 }
734 
735 /*
736  * share superblock if same fs AND options
737  */
738 static int ceph_compare_super(struct super_block *sb, void *data)
739 {
740 	struct ceph_fs_client *new = data;
741 	struct ceph_mount_options *fsopt = new->mount_options;
742 	struct ceph_options *opt = new->client->options;
743 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
744 
745 	dout("ceph_compare_super %p\n", sb);
746 
747 	if (compare_mount_options(fsopt, opt, other)) {
748 		dout("monitor(s)/mount options don't match\n");
749 		return 0;
750 	}
751 	if ((opt->flags & CEPH_OPT_FSID) &&
752 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
753 		dout("fsid doesn't match\n");
754 		return 0;
755 	}
756 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
757 		dout("flags differ\n");
758 		return 0;
759 	}
760 	return 1;
761 }
762 
763 /*
764  * construct our own bdi so we can control readahead, etc.
765  */
766 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
767 
768 static int ceph_register_bdi(struct super_block *sb,
769 			     struct ceph_fs_client *fsc)
770 {
771 	int err;
772 
773 	/* set ra_pages based on rsize mount option? */
774 	if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
775 		fsc->backing_dev_info.ra_pages =
776 			(fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
777 			>> PAGE_SHIFT;
778 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
779 			   atomic_long_inc_return(&bdi_seq));
780 	if (!err)
781 		sb->s_bdi = &fsc->backing_dev_info;
782 	return err;
783 }
784 
785 static struct dentry *ceph_mount(struct file_system_type *fs_type,
786 		       int flags, const char *dev_name, void *data)
787 {
788 	struct super_block *sb;
789 	struct ceph_fs_client *fsc;
790 	struct dentry *res;
791 	int err;
792 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
793 	const char *path = NULL;
794 	struct ceph_mount_options *fsopt = NULL;
795 	struct ceph_options *opt = NULL;
796 
797 	dout("ceph_mount\n");
798 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
799 	if (err < 0) {
800 		res = ERR_PTR(err);
801 		goto out_final;
802 	}
803 
804 	/* create client (which we may/may not use) */
805 	fsc = create_fs_client(fsopt, opt);
806 	if (IS_ERR(fsc)) {
807 		res = ERR_CAST(fsc);
808 		kfree(fsopt);
809 		kfree(opt);
810 		goto out_final;
811 	}
812 
813 	err = ceph_mdsc_init(fsc);
814 	if (err < 0) {
815 		res = ERR_PTR(err);
816 		goto out;
817 	}
818 
819 	if (ceph_test_opt(fsc->client, NOSHARE))
820 		compare_super = NULL;
821 	sb = sget(fs_type, compare_super, ceph_set_super, fsc);
822 	if (IS_ERR(sb)) {
823 		res = ERR_CAST(sb);
824 		goto out;
825 	}
826 
827 	if (ceph_sb_to_client(sb) != fsc) {
828 		ceph_mdsc_destroy(fsc);
829 		destroy_fs_client(fsc);
830 		fsc = ceph_sb_to_client(sb);
831 		dout("get_sb got existing client %p\n", fsc);
832 	} else {
833 		dout("get_sb using new client %p\n", fsc);
834 		err = ceph_register_bdi(sb, fsc);
835 		if (err < 0) {
836 			res = ERR_PTR(err);
837 			goto out_splat;
838 		}
839 	}
840 
841 	res = ceph_real_mount(fsc, path);
842 	if (IS_ERR(res))
843 		goto out_splat;
844 	dout("root %p inode %p ino %llx.%llx\n", res,
845 	     res->d_inode, ceph_vinop(res->d_inode));
846 	return res;
847 
848 out_splat:
849 	ceph_mdsc_close_sessions(fsc->mdsc);
850 	deactivate_locked_super(sb);
851 	goto out_final;
852 
853 out:
854 	ceph_mdsc_destroy(fsc);
855 	destroy_fs_client(fsc);
856 out_final:
857 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
858 	return res;
859 }
860 
861 static void ceph_kill_sb(struct super_block *s)
862 {
863 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
864 	dout("kill_sb %p\n", s);
865 	ceph_mdsc_pre_umount(fsc->mdsc);
866 	kill_anon_super(s);    /* will call put_super after sb is r/o */
867 	ceph_mdsc_destroy(fsc);
868 	destroy_fs_client(fsc);
869 }
870 
871 static struct file_system_type ceph_fs_type = {
872 	.owner		= THIS_MODULE,
873 	.name		= "ceph",
874 	.mount		= ceph_mount,
875 	.kill_sb	= ceph_kill_sb,
876 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
877 };
878 
879 #define _STRINGIFY(x) #x
880 #define STRINGIFY(x) _STRINGIFY(x)
881 
882 static int __init init_ceph(void)
883 {
884 	int ret = init_caches();
885 	if (ret)
886 		goto out;
887 
888 	ret = register_filesystem(&ceph_fs_type);
889 	if (ret)
890 		goto out_icache;
891 
892 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
893 
894 	return 0;
895 
896 out_icache:
897 	destroy_caches();
898 out:
899 	return ret;
900 }
901 
902 static void __exit exit_ceph(void)
903 {
904 	dout("exit_ceph\n");
905 	unregister_filesystem(&ceph_fs_type);
906 	destroy_caches();
907 }
908 
909 module_init(init_ceph);
910 module_exit(exit_ceph);
911 
912 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
913 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
914 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
915 MODULE_DESCRIPTION("Ceph filesystem for Linux");
916 MODULE_LICENSE("GPL");
917