xref: /openbmc/linux/fs/ceph/super.c (revision 97da55fc)
1 
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/backing-dev.h>
5 #include <linux/ctype.h>
6 #include <linux/fs.h>
7 #include <linux/inet.h>
8 #include <linux/in6.h>
9 #include <linux/module.h>
10 #include <linux/mount.h>
11 #include <linux/parser.h>
12 #include <linux/sched.h>
13 #include <linux/seq_file.h>
14 #include <linux/slab.h>
15 #include <linux/statfs.h>
16 #include <linux/string.h>
17 
18 #include "super.h"
19 #include "mds_client.h"
20 
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/decode.h>
23 #include <linux/ceph/mon_client.h>
24 #include <linux/ceph/auth.h>
25 #include <linux/ceph/debugfs.h>
26 
27 /*
28  * Ceph superblock operations
29  *
30  * Handle the basics of mounting, unmounting.
31  */
32 
33 /*
34  * super ops
35  */
36 static void ceph_put_super(struct super_block *s)
37 {
38 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
39 
40 	dout("put_super\n");
41 	ceph_mdsc_close_sessions(fsc->mdsc);
42 
43 	/*
44 	 * ensure we release the bdi before put_anon_super releases
45 	 * the device name.
46 	 */
47 	if (s->s_bdi == &fsc->backing_dev_info) {
48 		bdi_unregister(&fsc->backing_dev_info);
49 		s->s_bdi = NULL;
50 	}
51 
52 	return;
53 }
54 
55 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
56 {
57 	struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
58 	struct ceph_monmap *monmap = fsc->client->monc.monmap;
59 	struct ceph_statfs st;
60 	u64 fsid;
61 	int err;
62 
63 	dout("statfs\n");
64 	err = ceph_monc_do_statfs(&fsc->client->monc, &st);
65 	if (err < 0)
66 		return err;
67 
68 	/* fill in kstatfs */
69 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
70 
71 	/*
72 	 * express utilization in terms of large blocks to avoid
73 	 * overflow on 32-bit machines.
74 	 *
75 	 * NOTE: for the time being, we make bsize == frsize to humor
76 	 * not-yet-ancient versions of glibc that are broken.
77 	 * Someday, we will probably want to report a real block
78 	 * size...  whatever that may mean for a network file system!
79 	 */
80 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
81 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
82 	buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
83 	buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
84 	buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
85 
86 	buf->f_files = le64_to_cpu(st.num_objects);
87 	buf->f_ffree = -1;
88 	buf->f_namelen = NAME_MAX;
89 
90 	/* leave fsid little-endian, regardless of host endianness */
91 	fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
92 	buf->f_fsid.val[0] = fsid & 0xffffffff;
93 	buf->f_fsid.val[1] = fsid >> 32;
94 
95 	return 0;
96 }
97 
98 
99 static int ceph_sync_fs(struct super_block *sb, int wait)
100 {
101 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
102 
103 	if (!wait) {
104 		dout("sync_fs (non-blocking)\n");
105 		ceph_flush_dirty_caps(fsc->mdsc);
106 		dout("sync_fs (non-blocking) done\n");
107 		return 0;
108 	}
109 
110 	dout("sync_fs (blocking)\n");
111 	ceph_osdc_sync(&fsc->client->osdc);
112 	ceph_mdsc_sync(fsc->mdsc);
113 	dout("sync_fs (blocking) done\n");
114 	return 0;
115 }
116 
117 /*
118  * mount options
119  */
120 enum {
121 	Opt_wsize,
122 	Opt_rsize,
123 	Opt_rasize,
124 	Opt_caps_wanted_delay_min,
125 	Opt_caps_wanted_delay_max,
126 	Opt_cap_release_safety,
127 	Opt_readdir_max_entries,
128 	Opt_readdir_max_bytes,
129 	Opt_congestion_kb,
130 	Opt_last_int,
131 	/* int args above */
132 	Opt_snapdirname,
133 	Opt_last_string,
134 	/* string args above */
135 	Opt_dirstat,
136 	Opt_nodirstat,
137 	Opt_rbytes,
138 	Opt_norbytes,
139 	Opt_asyncreaddir,
140 	Opt_noasyncreaddir,
141 	Opt_dcache,
142 	Opt_nodcache,
143 	Opt_ino32,
144 	Opt_noino32,
145 };
146 
147 static match_table_t fsopt_tokens = {
148 	{Opt_wsize, "wsize=%d"},
149 	{Opt_rsize, "rsize=%d"},
150 	{Opt_rasize, "rasize=%d"},
151 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
152 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
153 	{Opt_cap_release_safety, "cap_release_safety=%d"},
154 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
155 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
156 	{Opt_congestion_kb, "write_congestion_kb=%d"},
157 	/* int args above */
158 	{Opt_snapdirname, "snapdirname=%s"},
159 	/* string args above */
160 	{Opt_dirstat, "dirstat"},
161 	{Opt_nodirstat, "nodirstat"},
162 	{Opt_rbytes, "rbytes"},
163 	{Opt_norbytes, "norbytes"},
164 	{Opt_asyncreaddir, "asyncreaddir"},
165 	{Opt_noasyncreaddir, "noasyncreaddir"},
166 	{Opt_dcache, "dcache"},
167 	{Opt_nodcache, "nodcache"},
168 	{Opt_ino32, "ino32"},
169 	{Opt_noino32, "noino32"},
170 	{-1, NULL}
171 };
172 
173 static int parse_fsopt_token(char *c, void *private)
174 {
175 	struct ceph_mount_options *fsopt = private;
176 	substring_t argstr[MAX_OPT_ARGS];
177 	int token, intval, ret;
178 
179 	token = match_token((char *)c, fsopt_tokens, argstr);
180 	if (token < 0)
181 		return -EINVAL;
182 
183 	if (token < Opt_last_int) {
184 		ret = match_int(&argstr[0], &intval);
185 		if (ret < 0) {
186 			pr_err("bad mount option arg (not int) "
187 			       "at '%s'\n", c);
188 			return ret;
189 		}
190 		dout("got int token %d val %d\n", token, intval);
191 	} else if (token > Opt_last_int && token < Opt_last_string) {
192 		dout("got string token %d val %s\n", token,
193 		     argstr[0].from);
194 	} else {
195 		dout("got token %d\n", token);
196 	}
197 
198 	switch (token) {
199 	case Opt_snapdirname:
200 		kfree(fsopt->snapdir_name);
201 		fsopt->snapdir_name = kstrndup(argstr[0].from,
202 					       argstr[0].to-argstr[0].from,
203 					       GFP_KERNEL);
204 		if (!fsopt->snapdir_name)
205 			return -ENOMEM;
206 		break;
207 
208 		/* misc */
209 	case Opt_wsize:
210 		fsopt->wsize = intval;
211 		break;
212 	case Opt_rsize:
213 		fsopt->rsize = intval;
214 		break;
215 	case Opt_rasize:
216 		fsopt->rasize = intval;
217 		break;
218 	case Opt_caps_wanted_delay_min:
219 		fsopt->caps_wanted_delay_min = intval;
220 		break;
221 	case Opt_caps_wanted_delay_max:
222 		fsopt->caps_wanted_delay_max = intval;
223 		break;
224 	case Opt_readdir_max_entries:
225 		fsopt->max_readdir = intval;
226 		break;
227 	case Opt_readdir_max_bytes:
228 		fsopt->max_readdir_bytes = intval;
229 		break;
230 	case Opt_congestion_kb:
231 		fsopt->congestion_kb = intval;
232 		break;
233 	case Opt_dirstat:
234 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
235 		break;
236 	case Opt_nodirstat:
237 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
238 		break;
239 	case Opt_rbytes:
240 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
241 		break;
242 	case Opt_norbytes:
243 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
244 		break;
245 	case Opt_asyncreaddir:
246 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
247 		break;
248 	case Opt_noasyncreaddir:
249 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
250 		break;
251 	case Opt_dcache:
252 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
253 		break;
254 	case Opt_nodcache:
255 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
256 		break;
257 	case Opt_ino32:
258 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
259 		break;
260 	case Opt_noino32:
261 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
262 		break;
263 	default:
264 		BUG_ON(token);
265 	}
266 	return 0;
267 }
268 
269 static void destroy_mount_options(struct ceph_mount_options *args)
270 {
271 	dout("destroy_mount_options %p\n", args);
272 	kfree(args->snapdir_name);
273 	kfree(args);
274 }
275 
276 static int strcmp_null(const char *s1, const char *s2)
277 {
278 	if (!s1 && !s2)
279 		return 0;
280 	if (s1 && !s2)
281 		return -1;
282 	if (!s1 && s2)
283 		return 1;
284 	return strcmp(s1, s2);
285 }
286 
287 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
288 				 struct ceph_options *new_opt,
289 				 struct ceph_fs_client *fsc)
290 {
291 	struct ceph_mount_options *fsopt1 = new_fsopt;
292 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
293 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
294 	int ret;
295 
296 	ret = memcmp(fsopt1, fsopt2, ofs);
297 	if (ret)
298 		return ret;
299 
300 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
301 	if (ret)
302 		return ret;
303 
304 	return ceph_compare_options(new_opt, fsc->client);
305 }
306 
307 static int parse_mount_options(struct ceph_mount_options **pfsopt,
308 			       struct ceph_options **popt,
309 			       int flags, char *options,
310 			       const char *dev_name,
311 			       const char **path)
312 {
313 	struct ceph_mount_options *fsopt;
314 	const char *dev_name_end;
315 	int err;
316 
317 	if (!dev_name || !*dev_name)
318 		return -EINVAL;
319 
320 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
321 	if (!fsopt)
322 		return -ENOMEM;
323 
324 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
325 
326 	fsopt->sb_flags = flags;
327 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
328 
329 	fsopt->rsize = CEPH_RSIZE_DEFAULT;
330 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
331 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
332 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
333 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
334 	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
335 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
336 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
337 	fsopt->congestion_kb = default_congestion_kb();
338 
339 	/*
340 	 * Distinguish the server list from the path in "dev_name".
341 	 * Internally we do not include the leading '/' in the path.
342 	 *
343 	 * "dev_name" will look like:
344 	 *     <server_spec>[,<server_spec>...]:[<path>]
345 	 * where
346 	 *     <server_spec> is <ip>[:<port>]
347 	 *     <path> is optional, but if present must begin with '/'
348 	 */
349 	dev_name_end = strchr(dev_name, '/');
350 	if (dev_name_end) {
351 		/* skip over leading '/' for path */
352 		*path = dev_name_end + 1;
353 	} else {
354 		/* path is empty */
355 		dev_name_end = dev_name + strlen(dev_name);
356 		*path = dev_name_end;
357 	}
358 	err = -EINVAL;
359 	dev_name_end--;		/* back up to ':' separator */
360 	if (*dev_name_end != ':') {
361 		pr_err("device name is missing path (no : separator in %s)\n",
362 				dev_name);
363 		goto out;
364 	}
365 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
366 	dout("server path '%s'\n", *path);
367 
368 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
369 				 parse_fsopt_token, (void *)fsopt);
370 	if (IS_ERR(*popt)) {
371 		err = PTR_ERR(*popt);
372 		goto out;
373 	}
374 
375 	/* success */
376 	*pfsopt = fsopt;
377 	return 0;
378 
379 out:
380 	destroy_mount_options(fsopt);
381 	return err;
382 }
383 
384 /**
385  * ceph_show_options - Show mount options in /proc/mounts
386  * @m: seq_file to write to
387  * @root: root of that (sub)tree
388  */
389 static int ceph_show_options(struct seq_file *m, struct dentry *root)
390 {
391 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
392 	struct ceph_mount_options *fsopt = fsc->mount_options;
393 	struct ceph_options *opt = fsc->client->options;
394 
395 	if (opt->flags & CEPH_OPT_FSID)
396 		seq_printf(m, ",fsid=%pU", &opt->fsid);
397 	if (opt->flags & CEPH_OPT_NOSHARE)
398 		seq_puts(m, ",noshare");
399 	if (opt->flags & CEPH_OPT_NOCRC)
400 		seq_puts(m, ",nocrc");
401 
402 	if (opt->name)
403 		seq_printf(m, ",name=%s", opt->name);
404 	if (opt->key)
405 		seq_puts(m, ",secret=<hidden>");
406 
407 	if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
408 		seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
409 	if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
410 		seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
411 	if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
412 		seq_printf(m, ",osdkeepalivetimeout=%d",
413 			   opt->osd_keepalive_timeout);
414 
415 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
416 		seq_puts(m, ",dirstat");
417 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
418 		seq_puts(m, ",norbytes");
419 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
420 		seq_puts(m, ",noasyncreaddir");
421 	if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
422 		seq_puts(m, ",dcache");
423 	else
424 		seq_puts(m, ",nodcache");
425 
426 	if (fsopt->wsize)
427 		seq_printf(m, ",wsize=%d", fsopt->wsize);
428 	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
429 		seq_printf(m, ",rsize=%d", fsopt->rsize);
430 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
431 		seq_printf(m, ",rasize=%d", fsopt->rasize);
432 	if (fsopt->congestion_kb != default_congestion_kb())
433 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
434 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
435 		seq_printf(m, ",caps_wanted_delay_min=%d",
436 			 fsopt->caps_wanted_delay_min);
437 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
438 		seq_printf(m, ",caps_wanted_delay_max=%d",
439 			   fsopt->caps_wanted_delay_max);
440 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
441 		seq_printf(m, ",cap_release_safety=%d",
442 			   fsopt->cap_release_safety);
443 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
444 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
445 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
446 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
447 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
448 		seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
449 	return 0;
450 }
451 
452 /*
453  * handle any mon messages the standard library doesn't understand.
454  * return error if we don't either.
455  */
456 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
457 {
458 	struct ceph_fs_client *fsc = client->private;
459 	int type = le16_to_cpu(msg->hdr.type);
460 
461 	switch (type) {
462 	case CEPH_MSG_MDS_MAP:
463 		ceph_mdsc_handle_map(fsc->mdsc, msg);
464 		return 0;
465 
466 	default:
467 		return -1;
468 	}
469 }
470 
471 /*
472  * create a new fs client
473  */
474 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
475 					struct ceph_options *opt)
476 {
477 	struct ceph_fs_client *fsc;
478 	const unsigned supported_features =
479 		CEPH_FEATURE_FLOCK |
480 		CEPH_FEATURE_DIRLAYOUTHASH;
481 	const unsigned required_features = 0;
482 	int err = -ENOMEM;
483 
484 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
485 	if (!fsc)
486 		return ERR_PTR(-ENOMEM);
487 
488 	fsc->client = ceph_create_client(opt, fsc, supported_features,
489 					 required_features);
490 	if (IS_ERR(fsc->client)) {
491 		err = PTR_ERR(fsc->client);
492 		goto fail;
493 	}
494 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
495 	fsc->client->monc.want_mdsmap = 1;
496 
497 	fsc->mount_options = fsopt;
498 
499 	fsc->sb = NULL;
500 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
501 
502 	atomic_long_set(&fsc->writeback_count, 0);
503 
504 	err = bdi_init(&fsc->backing_dev_info);
505 	if (err < 0)
506 		goto fail_client;
507 
508 	err = -ENOMEM;
509 	/*
510 	 * The number of concurrent works can be high but they don't need
511 	 * to be processed in parallel, limit concurrency.
512 	 */
513 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
514 	if (fsc->wb_wq == NULL)
515 		goto fail_bdi;
516 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
517 	if (fsc->pg_inv_wq == NULL)
518 		goto fail_wb_wq;
519 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
520 	if (fsc->trunc_wq == NULL)
521 		goto fail_pg_inv_wq;
522 
523 	/* set up mempools */
524 	err = -ENOMEM;
525 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
526 			      fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
527 	if (!fsc->wb_pagevec_pool)
528 		goto fail_trunc_wq;
529 
530 	/* caps */
531 	fsc->min_caps = fsopt->max_readdir;
532 
533 	return fsc;
534 
535 fail_trunc_wq:
536 	destroy_workqueue(fsc->trunc_wq);
537 fail_pg_inv_wq:
538 	destroy_workqueue(fsc->pg_inv_wq);
539 fail_wb_wq:
540 	destroy_workqueue(fsc->wb_wq);
541 fail_bdi:
542 	bdi_destroy(&fsc->backing_dev_info);
543 fail_client:
544 	ceph_destroy_client(fsc->client);
545 fail:
546 	kfree(fsc);
547 	return ERR_PTR(err);
548 }
549 
550 static void destroy_fs_client(struct ceph_fs_client *fsc)
551 {
552 	dout("destroy_fs_client %p\n", fsc);
553 
554 	destroy_workqueue(fsc->wb_wq);
555 	destroy_workqueue(fsc->pg_inv_wq);
556 	destroy_workqueue(fsc->trunc_wq);
557 
558 	bdi_destroy(&fsc->backing_dev_info);
559 
560 	mempool_destroy(fsc->wb_pagevec_pool);
561 
562 	destroy_mount_options(fsc->mount_options);
563 
564 	ceph_fs_debugfs_cleanup(fsc);
565 
566 	ceph_destroy_client(fsc->client);
567 
568 	kfree(fsc);
569 	dout("destroy_fs_client %p done\n", fsc);
570 }
571 
572 /*
573  * caches
574  */
575 struct kmem_cache *ceph_inode_cachep;
576 struct kmem_cache *ceph_cap_cachep;
577 struct kmem_cache *ceph_dentry_cachep;
578 struct kmem_cache *ceph_file_cachep;
579 
580 static void ceph_inode_init_once(void *foo)
581 {
582 	struct ceph_inode_info *ci = foo;
583 	inode_init_once(&ci->vfs_inode);
584 }
585 
586 static int __init init_caches(void)
587 {
588 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
589 				      sizeof(struct ceph_inode_info),
590 				      __alignof__(struct ceph_inode_info),
591 				      (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
592 				      ceph_inode_init_once);
593 	if (ceph_inode_cachep == NULL)
594 		return -ENOMEM;
595 
596 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
597 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
598 	if (ceph_cap_cachep == NULL)
599 		goto bad_cap;
600 
601 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
602 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
603 	if (ceph_dentry_cachep == NULL)
604 		goto bad_dentry;
605 
606 	ceph_file_cachep = KMEM_CACHE(ceph_file_info,
607 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
608 	if (ceph_file_cachep == NULL)
609 		goto bad_file;
610 
611 	return 0;
612 
613 bad_file:
614 	kmem_cache_destroy(ceph_dentry_cachep);
615 bad_dentry:
616 	kmem_cache_destroy(ceph_cap_cachep);
617 bad_cap:
618 	kmem_cache_destroy(ceph_inode_cachep);
619 	return -ENOMEM;
620 }
621 
622 static void destroy_caches(void)
623 {
624 	/*
625 	 * Make sure all delayed rcu free inodes are flushed before we
626 	 * destroy cache.
627 	 */
628 	rcu_barrier();
629 	kmem_cache_destroy(ceph_inode_cachep);
630 	kmem_cache_destroy(ceph_cap_cachep);
631 	kmem_cache_destroy(ceph_dentry_cachep);
632 	kmem_cache_destroy(ceph_file_cachep);
633 }
634 
635 
636 /*
637  * ceph_umount_begin - initiate forced umount.  Tear down down the
638  * mount, skipping steps that may hang while waiting for server(s).
639  */
640 static void ceph_umount_begin(struct super_block *sb)
641 {
642 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
643 
644 	dout("ceph_umount_begin - starting forced umount\n");
645 	if (!fsc)
646 		return;
647 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
648 	return;
649 }
650 
651 static const struct super_operations ceph_super_ops = {
652 	.alloc_inode	= ceph_alloc_inode,
653 	.destroy_inode	= ceph_destroy_inode,
654 	.write_inode    = ceph_write_inode,
655 	.sync_fs        = ceph_sync_fs,
656 	.put_super	= ceph_put_super,
657 	.show_options   = ceph_show_options,
658 	.statfs		= ceph_statfs,
659 	.umount_begin   = ceph_umount_begin,
660 };
661 
662 /*
663  * Bootstrap mount by opening the root directory.  Note the mount
664  * @started time from caller, and time out if this takes too long.
665  */
666 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
667 				       const char *path,
668 				       unsigned long started)
669 {
670 	struct ceph_mds_client *mdsc = fsc->mdsc;
671 	struct ceph_mds_request *req = NULL;
672 	int err;
673 	struct dentry *root;
674 
675 	/* open dir */
676 	dout("open_root_inode opening '%s'\n", path);
677 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
678 	if (IS_ERR(req))
679 		return ERR_CAST(req);
680 	req->r_path1 = kstrdup(path, GFP_NOFS);
681 	req->r_ino1.ino = CEPH_INO_ROOT;
682 	req->r_ino1.snap = CEPH_NOSNAP;
683 	req->r_started = started;
684 	req->r_timeout = fsc->client->options->mount_timeout * HZ;
685 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
686 	req->r_num_caps = 2;
687 	err = ceph_mdsc_do_request(mdsc, NULL, req);
688 	if (err == 0) {
689 		struct inode *inode = req->r_target_inode;
690 		req->r_target_inode = NULL;
691 		dout("open_root_inode success\n");
692 		if (ceph_ino(inode) == CEPH_INO_ROOT &&
693 		    fsc->sb->s_root == NULL) {
694 			root = d_make_root(inode);
695 			if (!root) {
696 				root = ERR_PTR(-ENOMEM);
697 				goto out;
698 			}
699 		} else {
700 			root = d_obtain_alias(inode);
701 		}
702 		ceph_init_dentry(root);
703 		dout("open_root_inode success, root dentry is %p\n", root);
704 	} else {
705 		root = ERR_PTR(err);
706 	}
707 out:
708 	ceph_mdsc_put_request(req);
709 	return root;
710 }
711 
712 
713 
714 
715 /*
716  * mount: join the ceph cluster, and open root directory.
717  */
718 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
719 		      const char *path)
720 {
721 	int err;
722 	unsigned long started = jiffies;  /* note the start time */
723 	struct dentry *root;
724 	int first = 0;   /* first vfsmount for this super_block */
725 
726 	dout("mount start\n");
727 	mutex_lock(&fsc->client->mount_mutex);
728 
729 	err = __ceph_open_session(fsc->client, started);
730 	if (err < 0)
731 		goto out;
732 
733 	dout("mount opening root\n");
734 	root = open_root_dentry(fsc, "", started);
735 	if (IS_ERR(root)) {
736 		err = PTR_ERR(root);
737 		goto out;
738 	}
739 	if (fsc->sb->s_root) {
740 		dput(root);
741 	} else {
742 		fsc->sb->s_root = root;
743 		first = 1;
744 
745 		err = ceph_fs_debugfs_init(fsc);
746 		if (err < 0)
747 			goto fail;
748 	}
749 
750 	if (path[0] == 0) {
751 		dget(root);
752 	} else {
753 		dout("mount opening base mountpoint\n");
754 		root = open_root_dentry(fsc, path, started);
755 		if (IS_ERR(root)) {
756 			err = PTR_ERR(root);
757 			goto fail;
758 		}
759 	}
760 
761 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
762 	dout("mount success\n");
763 	mutex_unlock(&fsc->client->mount_mutex);
764 	return root;
765 
766 out:
767 	mutex_unlock(&fsc->client->mount_mutex);
768 	return ERR_PTR(err);
769 
770 fail:
771 	if (first) {
772 		dput(fsc->sb->s_root);
773 		fsc->sb->s_root = NULL;
774 	}
775 	goto out;
776 }
777 
778 static int ceph_set_super(struct super_block *s, void *data)
779 {
780 	struct ceph_fs_client *fsc = data;
781 	int ret;
782 
783 	dout("set_super %p data %p\n", s, data);
784 
785 	s->s_flags = fsc->mount_options->sb_flags;
786 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
787 
788 	s->s_fs_info = fsc;
789 	fsc->sb = s;
790 
791 	s->s_op = &ceph_super_ops;
792 	s->s_export_op = &ceph_export_ops;
793 
794 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
795 
796 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
797 	if (ret != 0)
798 		goto fail;
799 
800 	return ret;
801 
802 fail:
803 	s->s_fs_info = NULL;
804 	fsc->sb = NULL;
805 	return ret;
806 }
807 
808 /*
809  * share superblock if same fs AND options
810  */
811 static int ceph_compare_super(struct super_block *sb, void *data)
812 {
813 	struct ceph_fs_client *new = data;
814 	struct ceph_mount_options *fsopt = new->mount_options;
815 	struct ceph_options *opt = new->client->options;
816 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
817 
818 	dout("ceph_compare_super %p\n", sb);
819 
820 	if (compare_mount_options(fsopt, opt, other)) {
821 		dout("monitor(s)/mount options don't match\n");
822 		return 0;
823 	}
824 	if ((opt->flags & CEPH_OPT_FSID) &&
825 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
826 		dout("fsid doesn't match\n");
827 		return 0;
828 	}
829 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
830 		dout("flags differ\n");
831 		return 0;
832 	}
833 	return 1;
834 }
835 
836 /*
837  * construct our own bdi so we can control readahead, etc.
838  */
839 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
840 
841 static int ceph_register_bdi(struct super_block *sb,
842 			     struct ceph_fs_client *fsc)
843 {
844 	int err;
845 
846 	/* set ra_pages based on rasize mount option? */
847 	if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
848 		fsc->backing_dev_info.ra_pages =
849 			(fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
850 			>> PAGE_SHIFT;
851 	else
852 		fsc->backing_dev_info.ra_pages =
853 			default_backing_dev_info.ra_pages;
854 
855 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
856 			   atomic_long_inc_return(&bdi_seq));
857 	if (!err)
858 		sb->s_bdi = &fsc->backing_dev_info;
859 	return err;
860 }
861 
862 static struct dentry *ceph_mount(struct file_system_type *fs_type,
863 		       int flags, const char *dev_name, void *data)
864 {
865 	struct super_block *sb;
866 	struct ceph_fs_client *fsc;
867 	struct dentry *res;
868 	int err;
869 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
870 	const char *path = NULL;
871 	struct ceph_mount_options *fsopt = NULL;
872 	struct ceph_options *opt = NULL;
873 
874 	dout("ceph_mount\n");
875 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
876 	if (err < 0) {
877 		res = ERR_PTR(err);
878 		goto out_final;
879 	}
880 
881 	/* create client (which we may/may not use) */
882 	fsc = create_fs_client(fsopt, opt);
883 	if (IS_ERR(fsc)) {
884 		res = ERR_CAST(fsc);
885 		destroy_mount_options(fsopt);
886 		ceph_destroy_options(opt);
887 		goto out_final;
888 	}
889 
890 	err = ceph_mdsc_init(fsc);
891 	if (err < 0) {
892 		res = ERR_PTR(err);
893 		goto out;
894 	}
895 
896 	if (ceph_test_opt(fsc->client, NOSHARE))
897 		compare_super = NULL;
898 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
899 	if (IS_ERR(sb)) {
900 		res = ERR_CAST(sb);
901 		goto out;
902 	}
903 
904 	if (ceph_sb_to_client(sb) != fsc) {
905 		ceph_mdsc_destroy(fsc);
906 		destroy_fs_client(fsc);
907 		fsc = ceph_sb_to_client(sb);
908 		dout("get_sb got existing client %p\n", fsc);
909 	} else {
910 		dout("get_sb using new client %p\n", fsc);
911 		err = ceph_register_bdi(sb, fsc);
912 		if (err < 0) {
913 			res = ERR_PTR(err);
914 			goto out_splat;
915 		}
916 	}
917 
918 	res = ceph_real_mount(fsc, path);
919 	if (IS_ERR(res))
920 		goto out_splat;
921 	dout("root %p inode %p ino %llx.%llx\n", res,
922 	     res->d_inode, ceph_vinop(res->d_inode));
923 	return res;
924 
925 out_splat:
926 	ceph_mdsc_close_sessions(fsc->mdsc);
927 	deactivate_locked_super(sb);
928 	goto out_final;
929 
930 out:
931 	ceph_mdsc_destroy(fsc);
932 	destroy_fs_client(fsc);
933 out_final:
934 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
935 	return res;
936 }
937 
938 static void ceph_kill_sb(struct super_block *s)
939 {
940 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
941 	dout("kill_sb %p\n", s);
942 	ceph_mdsc_pre_umount(fsc->mdsc);
943 	kill_anon_super(s);    /* will call put_super after sb is r/o */
944 	ceph_mdsc_destroy(fsc);
945 	destroy_fs_client(fsc);
946 }
947 
948 static struct file_system_type ceph_fs_type = {
949 	.owner		= THIS_MODULE,
950 	.name		= "ceph",
951 	.mount		= ceph_mount,
952 	.kill_sb	= ceph_kill_sb,
953 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
954 };
955 MODULE_ALIAS_FS("ceph");
956 
957 #define _STRINGIFY(x) #x
958 #define STRINGIFY(x) _STRINGIFY(x)
959 
960 static int __init init_ceph(void)
961 {
962 	int ret = init_caches();
963 	if (ret)
964 		goto out;
965 
966 	ceph_xattr_init();
967 	ret = register_filesystem(&ceph_fs_type);
968 	if (ret)
969 		goto out_icache;
970 
971 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
972 
973 	return 0;
974 
975 out_icache:
976 	ceph_xattr_exit();
977 	destroy_caches();
978 out:
979 	return ret;
980 }
981 
982 static void __exit exit_ceph(void)
983 {
984 	dout("exit_ceph\n");
985 	unregister_filesystem(&ceph_fs_type);
986 	ceph_xattr_exit();
987 	destroy_caches();
988 }
989 
990 module_init(init_ceph);
991 module_exit(exit_ceph);
992 
993 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
994 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
995 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
996 MODULE_DESCRIPTION("Ceph filesystem for Linux");
997 MODULE_LICENSE("GPL");
998