xref: /openbmc/linux/fs/ceph/super.c (revision 63dc02bd)
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_asyncreaddir,
134 	Opt_noasyncreaddir,
135 	Opt_dcache,
136 	Opt_nodcache,
137 	Opt_ino32,
138 	Opt_noino32,
139 };
140 
141 static 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 
167 static 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 
263 static 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 
270 static 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 
281 static 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 
301 static 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 
358 out:
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  */
368 static 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  */
437 static 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  */
455 static 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 
516 fail_trunc_wq:
517 	destroy_workqueue(fsc->trunc_wq);
518 fail_pg_inv_wq:
519 	destroy_workqueue(fsc->pg_inv_wq);
520 fail_wb_wq:
521 	destroy_workqueue(fsc->wb_wq);
522 fail_bdi:
523 	bdi_destroy(&fsc->backing_dev_info);
524 fail_client:
525 	ceph_destroy_client(fsc->client);
526 fail:
527 	kfree(fsc);
528 	return ERR_PTR(err);
529 }
530 
531 static 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  */
556 struct kmem_cache *ceph_inode_cachep;
557 struct kmem_cache *ceph_cap_cachep;
558 struct kmem_cache *ceph_dentry_cachep;
559 struct kmem_cache *ceph_file_cachep;
560 
561 static void ceph_inode_init_once(void *foo)
562 {
563 	struct ceph_inode_info *ci = foo;
564 	inode_init_once(&ci->vfs_inode);
565 }
566 
567 static 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 
594 bad_file:
595 	kmem_cache_destroy(ceph_dentry_cachep);
596 bad_dentry:
597 	kmem_cache_destroy(ceph_cap_cachep);
598 bad_cap:
599 	kmem_cache_destroy(ceph_inode_cachep);
600 	return -ENOMEM;
601 }
602 
603 static 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  */
616 static 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 
627 static 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  */
642 static 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 	}
683 out:
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  */
694 static 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 
742 out:
743 	mutex_unlock(&fsc->client->mount_mutex);
744 	return ERR_PTR(err);
745 
746 fail:
747 	if (first) {
748 		dput(fsc->sb->s_root);
749 		fsc->sb->s_root = NULL;
750 	}
751 	goto out;
752 }
753 
754 static 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 
778 fail:
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  */
787 static 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  */
815 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
816 
817 static 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 
838 static 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 
901 out_splat:
902 	ceph_mdsc_close_sessions(fsc->mdsc);
903 	deactivate_locked_super(sb);
904 	goto out_final;
905 
906 out:
907 	ceph_mdsc_destroy(fsc);
908 	destroy_fs_client(fsc);
909 out_final:
910 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
911 	return res;
912 }
913 
914 static 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 
924 static 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 
935 static 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 
950 out_icache:
951 	ceph_xattr_exit();
952 	destroy_caches();
953 out:
954 	return ret;
955 }
956 
957 static 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 
965 module_init(init_ceph);
966 module_exit(exit_ceph);
967 
968 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
969 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
970 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
971 MODULE_DESCRIPTION("Ceph filesystem for Linux");
972 MODULE_LICENSE("GPL");
973