xref: /openbmc/linux/fs/ceph/super.c (revision ee57741c5209154b8ef124bcaa2496da1b69a988)
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_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 
93 static 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  */
114 enum {
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_noasyncreaddir,
134 	Opt_dcache,
135 	Opt_nodcache,
136 	Opt_ino32,
137 };
138 
139 static match_table_t fsopt_tokens = {
140 	{Opt_wsize, "wsize=%d"},
141 	{Opt_rsize, "rsize=%d"},
142 	{Opt_rasize, "rasize=%d"},
143 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
144 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
145 	{Opt_cap_release_safety, "cap_release_safety=%d"},
146 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
147 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
148 	{Opt_congestion_kb, "write_congestion_kb=%d"},
149 	/* int args above */
150 	{Opt_snapdirname, "snapdirname=%s"},
151 	/* string args above */
152 	{Opt_dirstat, "dirstat"},
153 	{Opt_nodirstat, "nodirstat"},
154 	{Opt_rbytes, "rbytes"},
155 	{Opt_norbytes, "norbytes"},
156 	{Opt_noasyncreaddir, "noasyncreaddir"},
157 	{Opt_dcache, "dcache"},
158 	{Opt_nodcache, "nodcache"},
159 	{Opt_ino32, "ino32"},
160 	{-1, NULL}
161 };
162 
163 static int parse_fsopt_token(char *c, void *private)
164 {
165 	struct ceph_mount_options *fsopt = private;
166 	substring_t argstr[MAX_OPT_ARGS];
167 	int token, intval, ret;
168 
169 	token = match_token((char *)c, fsopt_tokens, argstr);
170 	if (token < 0)
171 		return -EINVAL;
172 
173 	if (token < Opt_last_int) {
174 		ret = match_int(&argstr[0], &intval);
175 		if (ret < 0) {
176 			pr_err("bad mount option arg (not int) "
177 			       "at '%s'\n", c);
178 			return ret;
179 		}
180 		dout("got int token %d val %d\n", token, intval);
181 	} else if (token > Opt_last_int && token < Opt_last_string) {
182 		dout("got string token %d val %s\n", token,
183 		     argstr[0].from);
184 	} else {
185 		dout("got token %d\n", token);
186 	}
187 
188 	switch (token) {
189 	case Opt_snapdirname:
190 		kfree(fsopt->snapdir_name);
191 		fsopt->snapdir_name = kstrndup(argstr[0].from,
192 					       argstr[0].to-argstr[0].from,
193 					       GFP_KERNEL);
194 		if (!fsopt->snapdir_name)
195 			return -ENOMEM;
196 		break;
197 
198 		/* misc */
199 	case Opt_wsize:
200 		fsopt->wsize = intval;
201 		break;
202 	case Opt_rsize:
203 		fsopt->rsize = intval;
204 		break;
205 	case Opt_rasize:
206 		fsopt->rasize = intval;
207 		break;
208 	case Opt_caps_wanted_delay_min:
209 		fsopt->caps_wanted_delay_min = intval;
210 		break;
211 	case Opt_caps_wanted_delay_max:
212 		fsopt->caps_wanted_delay_max = intval;
213 		break;
214 	case Opt_readdir_max_entries:
215 		fsopt->max_readdir = intval;
216 		break;
217 	case Opt_readdir_max_bytes:
218 		fsopt->max_readdir_bytes = intval;
219 		break;
220 	case Opt_congestion_kb:
221 		fsopt->congestion_kb = intval;
222 		break;
223 	case Opt_dirstat:
224 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
225 		break;
226 	case Opt_nodirstat:
227 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
228 		break;
229 	case Opt_rbytes:
230 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
231 		break;
232 	case Opt_norbytes:
233 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
234 		break;
235 	case Opt_noasyncreaddir:
236 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
237 		break;
238 	case Opt_dcache:
239 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
240 		break;
241 	case Opt_nodcache:
242 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
243 		break;
244 	case Opt_ino32:
245 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
246 		break;
247 	default:
248 		BUG_ON(token);
249 	}
250 	return 0;
251 }
252 
253 static void destroy_mount_options(struct ceph_mount_options *args)
254 {
255 	dout("destroy_mount_options %p\n", args);
256 	kfree(args->snapdir_name);
257 	kfree(args);
258 }
259 
260 static int strcmp_null(const char *s1, const char *s2)
261 {
262 	if (!s1 && !s2)
263 		return 0;
264 	if (s1 && !s2)
265 		return -1;
266 	if (!s1 && s2)
267 		return 1;
268 	return strcmp(s1, s2);
269 }
270 
271 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
272 				 struct ceph_options *new_opt,
273 				 struct ceph_fs_client *fsc)
274 {
275 	struct ceph_mount_options *fsopt1 = new_fsopt;
276 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
277 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
278 	int ret;
279 
280 	ret = memcmp(fsopt1, fsopt2, ofs);
281 	if (ret)
282 		return ret;
283 
284 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
285 	if (ret)
286 		return ret;
287 
288 	return ceph_compare_options(new_opt, fsc->client);
289 }
290 
291 static int parse_mount_options(struct ceph_mount_options **pfsopt,
292 			       struct ceph_options **popt,
293 			       int flags, char *options,
294 			       const char *dev_name,
295 			       const char **path)
296 {
297 	struct ceph_mount_options *fsopt;
298 	const char *dev_name_end;
299 	int err = -ENOMEM;
300 
301 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
302 	if (!fsopt)
303 		return -ENOMEM;
304 
305 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
306 
307 	fsopt->sb_flags = flags;
308 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
309 
310 	fsopt->rsize = CEPH_RSIZE_DEFAULT;
311 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
312 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
313 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
314 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
315 	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
316 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
317 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
318 	fsopt->congestion_kb = default_congestion_kb();
319 
320 	/* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
321 	err = -EINVAL;
322 	if (!dev_name)
323 		goto out;
324 	*path = strstr(dev_name, ":/");
325 	if (*path == NULL) {
326 		pr_err("device name is missing path (no :/ in %s)\n",
327 				dev_name);
328 		goto out;
329 	}
330 	dev_name_end = *path;
331 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
332 
333 	/* path on server */
334 	*path += 2;
335 	dout("server path '%s'\n", *path);
336 
337 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
338 				 parse_fsopt_token, (void *)fsopt);
339 	if (IS_ERR(*popt)) {
340 		err = PTR_ERR(*popt);
341 		goto out;
342 	}
343 
344 	/* success */
345 	*pfsopt = fsopt;
346 	return 0;
347 
348 out:
349 	destroy_mount_options(fsopt);
350 	return err;
351 }
352 
353 /**
354  * ceph_show_options - Show mount options in /proc/mounts
355  * @m: seq_file to write to
356  * @root: root of that (sub)tree
357  */
358 static int ceph_show_options(struct seq_file *m, struct dentry *root)
359 {
360 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
361 	struct ceph_mount_options *fsopt = fsc->mount_options;
362 	struct ceph_options *opt = fsc->client->options;
363 
364 	if (opt->flags & CEPH_OPT_FSID)
365 		seq_printf(m, ",fsid=%pU", &opt->fsid);
366 	if (opt->flags & CEPH_OPT_NOSHARE)
367 		seq_puts(m, ",noshare");
368 	if (opt->flags & CEPH_OPT_NOCRC)
369 		seq_puts(m, ",nocrc");
370 
371 	if (opt->name)
372 		seq_printf(m, ",name=%s", opt->name);
373 	if (opt->key)
374 		seq_puts(m, ",secret=<hidden>");
375 
376 	if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
377 		seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
378 	if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
379 		seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
380 	if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
381 		seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
382 	if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
383 		seq_printf(m, ",osdkeepalivetimeout=%d",
384 			   opt->osd_keepalive_timeout);
385 
386 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
387 		seq_puts(m, ",dirstat");
388 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
389 		seq_puts(m, ",norbytes");
390 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
391 		seq_puts(m, ",noasyncreaddir");
392 	if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
393 		seq_puts(m, ",dcache");
394 	else
395 		seq_puts(m, ",nodcache");
396 
397 	if (fsopt->wsize)
398 		seq_printf(m, ",wsize=%d", fsopt->wsize);
399 	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
400 		seq_printf(m, ",rsize=%d", fsopt->rsize);
401 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
402 		seq_printf(m, ",rasize=%d", fsopt->rasize);
403 	if (fsopt->congestion_kb != default_congestion_kb())
404 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
405 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
406 		seq_printf(m, ",caps_wanted_delay_min=%d",
407 			 fsopt->caps_wanted_delay_min);
408 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
409 		seq_printf(m, ",caps_wanted_delay_max=%d",
410 			   fsopt->caps_wanted_delay_max);
411 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
412 		seq_printf(m, ",cap_release_safety=%d",
413 			   fsopt->cap_release_safety);
414 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
415 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
416 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
417 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
418 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
419 		seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
420 	return 0;
421 }
422 
423 /*
424  * handle any mon messages the standard library doesn't understand.
425  * return error if we don't either.
426  */
427 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
428 {
429 	struct ceph_fs_client *fsc = client->private;
430 	int type = le16_to_cpu(msg->hdr.type);
431 
432 	switch (type) {
433 	case CEPH_MSG_MDS_MAP:
434 		ceph_mdsc_handle_map(fsc->mdsc, msg);
435 		return 0;
436 
437 	default:
438 		return -1;
439 	}
440 }
441 
442 /*
443  * create a new fs client
444  */
445 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
446 					struct ceph_options *opt)
447 {
448 	struct ceph_fs_client *fsc;
449 	const unsigned supported_features =
450 		CEPH_FEATURE_FLOCK |
451 		CEPH_FEATURE_DIRLAYOUTHASH;
452 	const unsigned required_features = 0;
453 	int err = -ENOMEM;
454 
455 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
456 	if (!fsc)
457 		return ERR_PTR(-ENOMEM);
458 
459 	fsc->client = ceph_create_client(opt, fsc, supported_features,
460 					 required_features);
461 	if (IS_ERR(fsc->client)) {
462 		err = PTR_ERR(fsc->client);
463 		goto fail;
464 	}
465 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
466 	fsc->client->monc.want_mdsmap = 1;
467 
468 	fsc->mount_options = fsopt;
469 
470 	fsc->sb = NULL;
471 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
472 
473 	atomic_long_set(&fsc->writeback_count, 0);
474 
475 	err = bdi_init(&fsc->backing_dev_info);
476 	if (err < 0)
477 		goto fail_client;
478 
479 	err = -ENOMEM;
480 	/*
481 	 * The number of concurrent works can be high but they don't need
482 	 * to be processed in parallel, limit concurrency.
483 	 */
484 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
485 	if (fsc->wb_wq == NULL)
486 		goto fail_bdi;
487 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
488 	if (fsc->pg_inv_wq == NULL)
489 		goto fail_wb_wq;
490 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
491 	if (fsc->trunc_wq == NULL)
492 		goto fail_pg_inv_wq;
493 
494 	/* set up mempools */
495 	err = -ENOMEM;
496 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
497 			      fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
498 	if (!fsc->wb_pagevec_pool)
499 		goto fail_trunc_wq;
500 
501 	/* caps */
502 	fsc->min_caps = fsopt->max_readdir;
503 
504 	return fsc;
505 
506 fail_trunc_wq:
507 	destroy_workqueue(fsc->trunc_wq);
508 fail_pg_inv_wq:
509 	destroy_workqueue(fsc->pg_inv_wq);
510 fail_wb_wq:
511 	destroy_workqueue(fsc->wb_wq);
512 fail_bdi:
513 	bdi_destroy(&fsc->backing_dev_info);
514 fail_client:
515 	ceph_destroy_client(fsc->client);
516 fail:
517 	kfree(fsc);
518 	return ERR_PTR(err);
519 }
520 
521 static void destroy_fs_client(struct ceph_fs_client *fsc)
522 {
523 	dout("destroy_fs_client %p\n", fsc);
524 
525 	destroy_workqueue(fsc->wb_wq);
526 	destroy_workqueue(fsc->pg_inv_wq);
527 	destroy_workqueue(fsc->trunc_wq);
528 
529 	bdi_destroy(&fsc->backing_dev_info);
530 
531 	mempool_destroy(fsc->wb_pagevec_pool);
532 
533 	destroy_mount_options(fsc->mount_options);
534 
535 	ceph_fs_debugfs_cleanup(fsc);
536 
537 	ceph_destroy_client(fsc->client);
538 
539 	kfree(fsc);
540 	dout("destroy_fs_client %p done\n", fsc);
541 }
542 
543 /*
544  * caches
545  */
546 struct kmem_cache *ceph_inode_cachep;
547 struct kmem_cache *ceph_cap_cachep;
548 struct kmem_cache *ceph_dentry_cachep;
549 struct kmem_cache *ceph_file_cachep;
550 
551 static void ceph_inode_init_once(void *foo)
552 {
553 	struct ceph_inode_info *ci = foo;
554 	inode_init_once(&ci->vfs_inode);
555 }
556 
557 static int __init init_caches(void)
558 {
559 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
560 				      sizeof(struct ceph_inode_info),
561 				      __alignof__(struct ceph_inode_info),
562 				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
563 				      ceph_inode_init_once);
564 	if (ceph_inode_cachep == NULL)
565 		return -ENOMEM;
566 
567 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
568 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
569 	if (ceph_cap_cachep == NULL)
570 		goto bad_cap;
571 
572 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
573 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
574 	if (ceph_dentry_cachep == NULL)
575 		goto bad_dentry;
576 
577 	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
578 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
579 	if (ceph_file_cachep == NULL)
580 		goto bad_file;
581 
582 	return 0;
583 
584 bad_file:
585 	kmem_cache_destroy(ceph_dentry_cachep);
586 bad_dentry:
587 	kmem_cache_destroy(ceph_cap_cachep);
588 bad_cap:
589 	kmem_cache_destroy(ceph_inode_cachep);
590 	return -ENOMEM;
591 }
592 
593 static void destroy_caches(void)
594 {
595 	kmem_cache_destroy(ceph_inode_cachep);
596 	kmem_cache_destroy(ceph_cap_cachep);
597 	kmem_cache_destroy(ceph_dentry_cachep);
598 	kmem_cache_destroy(ceph_file_cachep);
599 }
600 
601 
602 /*
603  * ceph_umount_begin - initiate forced umount.  Tear down down the
604  * mount, skipping steps that may hang while waiting for server(s).
605  */
606 static void ceph_umount_begin(struct super_block *sb)
607 {
608 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
609 
610 	dout("ceph_umount_begin - starting forced umount\n");
611 	if (!fsc)
612 		return;
613 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
614 	return;
615 }
616 
617 static const struct super_operations ceph_super_ops = {
618 	.alloc_inode	= ceph_alloc_inode,
619 	.destroy_inode	= ceph_destroy_inode,
620 	.write_inode    = ceph_write_inode,
621 	.sync_fs        = ceph_sync_fs,
622 	.put_super	= ceph_put_super,
623 	.show_options   = ceph_show_options,
624 	.statfs		= ceph_statfs,
625 	.umount_begin   = ceph_umount_begin,
626 };
627 
628 /*
629  * Bootstrap mount by opening the root directory.  Note the mount
630  * @started time from caller, and time out if this takes too long.
631  */
632 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
633 				       const char *path,
634 				       unsigned long started)
635 {
636 	struct ceph_mds_client *mdsc = fsc->mdsc;
637 	struct ceph_mds_request *req = NULL;
638 	int err;
639 	struct dentry *root;
640 
641 	/* open dir */
642 	dout("open_root_inode opening '%s'\n", path);
643 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
644 	if (IS_ERR(req))
645 		return ERR_CAST(req);
646 	req->r_path1 = kstrdup(path, GFP_NOFS);
647 	req->r_ino1.ino = CEPH_INO_ROOT;
648 	req->r_ino1.snap = CEPH_NOSNAP;
649 	req->r_started = started;
650 	req->r_timeout = fsc->client->options->mount_timeout * HZ;
651 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
652 	req->r_num_caps = 2;
653 	err = ceph_mdsc_do_request(mdsc, NULL, req);
654 	if (err == 0) {
655 		struct inode *inode = req->r_target_inode;
656 		req->r_target_inode = NULL;
657 		dout("open_root_inode success\n");
658 		if (ceph_ino(inode) == CEPH_INO_ROOT &&
659 		    fsc->sb->s_root == NULL) {
660 			root = d_alloc_root(inode);
661 			if (!root) {
662 				iput(inode);
663 				root = ERR_PTR(-ENOMEM);
664 				goto out;
665 			}
666 		} else {
667 			root = d_obtain_alias(inode);
668 		}
669 		ceph_init_dentry(root);
670 		dout("open_root_inode success, root dentry is %p\n", root);
671 	} else {
672 		root = ERR_PTR(err);
673 	}
674 out:
675 	ceph_mdsc_put_request(req);
676 	return root;
677 }
678 
679 
680 
681 
682 /*
683  * mount: join the ceph cluster, and open root directory.
684  */
685 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
686 		      const char *path)
687 {
688 	int err;
689 	unsigned long started = jiffies;  /* note the start time */
690 	struct dentry *root;
691 	int first = 0;   /* first vfsmount for this super_block */
692 
693 	dout("mount start\n");
694 	mutex_lock(&fsc->client->mount_mutex);
695 
696 	err = __ceph_open_session(fsc->client, started);
697 	if (err < 0)
698 		goto out;
699 
700 	dout("mount opening root\n");
701 	root = open_root_dentry(fsc, "", started);
702 	if (IS_ERR(root)) {
703 		err = PTR_ERR(root);
704 		goto out;
705 	}
706 	if (fsc->sb->s_root) {
707 		dput(root);
708 	} else {
709 		fsc->sb->s_root = root;
710 		first = 1;
711 
712 		err = ceph_fs_debugfs_init(fsc);
713 		if (err < 0)
714 			goto fail;
715 	}
716 
717 	if (path[0] == 0) {
718 		dget(root);
719 	} else {
720 		dout("mount opening base mountpoint\n");
721 		root = open_root_dentry(fsc, path, started);
722 		if (IS_ERR(root)) {
723 			err = PTR_ERR(root);
724 			goto fail;
725 		}
726 	}
727 
728 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
729 	dout("mount success\n");
730 	mutex_unlock(&fsc->client->mount_mutex);
731 	return root;
732 
733 out:
734 	mutex_unlock(&fsc->client->mount_mutex);
735 	return ERR_PTR(err);
736 
737 fail:
738 	if (first) {
739 		dput(fsc->sb->s_root);
740 		fsc->sb->s_root = NULL;
741 	}
742 	goto out;
743 }
744 
745 static int ceph_set_super(struct super_block *s, void *data)
746 {
747 	struct ceph_fs_client *fsc = data;
748 	int ret;
749 
750 	dout("set_super %p data %p\n", s, data);
751 
752 	s->s_flags = fsc->mount_options->sb_flags;
753 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
754 
755 	s->s_fs_info = fsc;
756 	fsc->sb = s;
757 
758 	s->s_op = &ceph_super_ops;
759 	s->s_export_op = &ceph_export_ops;
760 
761 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
762 
763 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
764 	if (ret != 0)
765 		goto fail;
766 
767 	return ret;
768 
769 fail:
770 	s->s_fs_info = NULL;
771 	fsc->sb = NULL;
772 	return ret;
773 }
774 
775 /*
776  * share superblock if same fs AND options
777  */
778 static int ceph_compare_super(struct super_block *sb, void *data)
779 {
780 	struct ceph_fs_client *new = data;
781 	struct ceph_mount_options *fsopt = new->mount_options;
782 	struct ceph_options *opt = new->client->options;
783 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
784 
785 	dout("ceph_compare_super %p\n", sb);
786 
787 	if (compare_mount_options(fsopt, opt, other)) {
788 		dout("monitor(s)/mount options don't match\n");
789 		return 0;
790 	}
791 	if ((opt->flags & CEPH_OPT_FSID) &&
792 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
793 		dout("fsid doesn't match\n");
794 		return 0;
795 	}
796 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
797 		dout("flags differ\n");
798 		return 0;
799 	}
800 	return 1;
801 }
802 
803 /*
804  * construct our own bdi so we can control readahead, etc.
805  */
806 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
807 
808 static int ceph_register_bdi(struct super_block *sb,
809 			     struct ceph_fs_client *fsc)
810 {
811 	int err;
812 
813 	/* set ra_pages based on rasize mount option? */
814 	if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
815 		fsc->backing_dev_info.ra_pages =
816 			(fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
817 			>> PAGE_SHIFT;
818 	else
819 		fsc->backing_dev_info.ra_pages =
820 			default_backing_dev_info.ra_pages;
821 
822 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
823 			   atomic_long_inc_return(&bdi_seq));
824 	if (!err)
825 		sb->s_bdi = &fsc->backing_dev_info;
826 	return err;
827 }
828 
829 static struct dentry *ceph_mount(struct file_system_type *fs_type,
830 		       int flags, const char *dev_name, void *data)
831 {
832 	struct super_block *sb;
833 	struct ceph_fs_client *fsc;
834 	struct dentry *res;
835 	int err;
836 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
837 	const char *path = NULL;
838 	struct ceph_mount_options *fsopt = NULL;
839 	struct ceph_options *opt = NULL;
840 
841 	dout("ceph_mount\n");
842 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
843 	if (err < 0) {
844 		res = ERR_PTR(err);
845 		goto out_final;
846 	}
847 
848 	/* create client (which we may/may not use) */
849 	fsc = create_fs_client(fsopt, opt);
850 	if (IS_ERR(fsc)) {
851 		res = ERR_CAST(fsc);
852 		destroy_mount_options(fsopt);
853 		ceph_destroy_options(opt);
854 		goto out_final;
855 	}
856 
857 	err = ceph_mdsc_init(fsc);
858 	if (err < 0) {
859 		res = ERR_PTR(err);
860 		goto out;
861 	}
862 
863 	if (ceph_test_opt(fsc->client, NOSHARE))
864 		compare_super = NULL;
865 	sb = sget(fs_type, compare_super, ceph_set_super, fsc);
866 	if (IS_ERR(sb)) {
867 		res = ERR_CAST(sb);
868 		goto out;
869 	}
870 
871 	if (ceph_sb_to_client(sb) != fsc) {
872 		ceph_mdsc_destroy(fsc);
873 		destroy_fs_client(fsc);
874 		fsc = ceph_sb_to_client(sb);
875 		dout("get_sb got existing client %p\n", fsc);
876 	} else {
877 		dout("get_sb using new client %p\n", fsc);
878 		err = ceph_register_bdi(sb, fsc);
879 		if (err < 0) {
880 			res = ERR_PTR(err);
881 			goto out_splat;
882 		}
883 	}
884 
885 	res = ceph_real_mount(fsc, path);
886 	if (IS_ERR(res))
887 		goto out_splat;
888 	dout("root %p inode %p ino %llx.%llx\n", res,
889 	     res->d_inode, ceph_vinop(res->d_inode));
890 	return res;
891 
892 out_splat:
893 	ceph_mdsc_close_sessions(fsc->mdsc);
894 	deactivate_locked_super(sb);
895 	goto out_final;
896 
897 out:
898 	ceph_mdsc_destroy(fsc);
899 	destroy_fs_client(fsc);
900 out_final:
901 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
902 	return res;
903 }
904 
905 static void ceph_kill_sb(struct super_block *s)
906 {
907 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
908 	dout("kill_sb %p\n", s);
909 	ceph_mdsc_pre_umount(fsc->mdsc);
910 	kill_anon_super(s);    /* will call put_super after sb is r/o */
911 	ceph_mdsc_destroy(fsc);
912 	destroy_fs_client(fsc);
913 }
914 
915 static struct file_system_type ceph_fs_type = {
916 	.owner		= THIS_MODULE,
917 	.name		= "ceph",
918 	.mount		= ceph_mount,
919 	.kill_sb	= ceph_kill_sb,
920 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
921 };
922 
923 #define _STRINGIFY(x) #x
924 #define STRINGIFY(x) _STRINGIFY(x)
925 
926 static int __init init_ceph(void)
927 {
928 	int ret = init_caches();
929 	if (ret)
930 		goto out;
931 
932 	ceph_xattr_init();
933 	ret = register_filesystem(&ceph_fs_type);
934 	if (ret)
935 		goto out_icache;
936 
937 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
938 
939 	return 0;
940 
941 out_icache:
942 	ceph_xattr_exit();
943 	destroy_caches();
944 out:
945 	return ret;
946 }
947 
948 static void __exit exit_ceph(void)
949 {
950 	dout("exit_ceph\n");
951 	unregister_filesystem(&ceph_fs_type);
952 	ceph_xattr_exit();
953 	destroy_caches();
954 }
955 
956 module_init(init_ceph);
957 module_exit(exit_ceph);
958 
959 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
960 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
961 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
962 MODULE_DESCRIPTION("Ceph filesystem for Linux");
963 MODULE_LICENSE("GPL");
964