xref: /openbmc/linux/fs/ceph/super.c (revision 0bea2a65)
1 
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/backing-dev.h>
5 #include <linux/ctype.h>
6 #include <linux/fs.h>
7 #include <linux/inet.h>
8 #include <linux/in6.h>
9 #include <linux/module.h>
10 #include <linux/mount.h>
11 #include <linux/parser.h>
12 #include <linux/sched.h>
13 #include <linux/seq_file.h>
14 #include <linux/slab.h>
15 #include <linux/statfs.h>
16 #include <linux/string.h>
17 
18 #include "super.h"
19 #include "mds_client.h"
20 #include "cache.h"
21 
22 #include <linux/ceph/ceph_features.h>
23 #include <linux/ceph/decode.h>
24 #include <linux/ceph/mon_client.h>
25 #include <linux/ceph/auth.h>
26 #include <linux/ceph/debugfs.h>
27 
28 /*
29  * Ceph superblock operations
30  *
31  * Handle the basics of mounting, unmounting.
32  */
33 
34 /*
35  * super ops
36  */
37 static void ceph_put_super(struct super_block *s)
38 {
39 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
40 
41 	dout("put_super\n");
42 	ceph_mdsc_close_sessions(fsc->mdsc);
43 }
44 
45 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
46 {
47 	struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
48 	struct ceph_monmap *monmap = fsc->client->monc.monmap;
49 	struct ceph_statfs st;
50 	u64 fsid;
51 	int err;
52 	u64 data_pool;
53 
54 	if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
55 		data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
56 	} else {
57 		data_pool = CEPH_NOPOOL;
58 	}
59 
60 	dout("statfs\n");
61 	err = ceph_monc_do_statfs(&fsc->client->monc, data_pool, &st);
62 	if (err < 0)
63 		return err;
64 
65 	/* fill in kstatfs */
66 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
67 
68 	/*
69 	 * express utilization in terms of large blocks to avoid
70 	 * overflow on 32-bit machines.
71 	 *
72 	 * NOTE: for the time being, we make bsize == frsize to humor
73 	 * not-yet-ancient versions of glibc that are broken.
74 	 * Someday, we will probably want to report a real block
75 	 * size...  whatever that may mean for a network file system!
76 	 */
77 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
78 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
79 	buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
80 	buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
81 	buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
82 
83 	buf->f_files = le64_to_cpu(st.num_objects);
84 	buf->f_ffree = -1;
85 	buf->f_namelen = NAME_MAX;
86 
87 	/* Must convert the fsid, for consistent values across arches */
88 	fsid = le64_to_cpu(*(__le64 *)(&monmap->fsid)) ^
89 	       le64_to_cpu(*((__le64 *)&monmap->fsid + 1));
90 	buf->f_fsid.val[0] = fsid & 0xffffffff;
91 	buf->f_fsid.val[1] = fsid >> 32;
92 
93 	return 0;
94 }
95 
96 
97 static int ceph_sync_fs(struct super_block *sb, int wait)
98 {
99 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
100 
101 	if (!wait) {
102 		dout("sync_fs (non-blocking)\n");
103 		ceph_flush_dirty_caps(fsc->mdsc);
104 		dout("sync_fs (non-blocking) done\n");
105 		return 0;
106 	}
107 
108 	dout("sync_fs (blocking)\n");
109 	ceph_osdc_sync(&fsc->client->osdc);
110 	ceph_mdsc_sync(fsc->mdsc);
111 	dout("sync_fs (blocking) done\n");
112 	return 0;
113 }
114 
115 /*
116  * mount options
117  */
118 enum {
119 	Opt_wsize,
120 	Opt_rsize,
121 	Opt_rasize,
122 	Opt_caps_wanted_delay_min,
123 	Opt_caps_wanted_delay_max,
124 	Opt_readdir_max_entries,
125 	Opt_readdir_max_bytes,
126 	Opt_congestion_kb,
127 	Opt_last_int,
128 	/* int args above */
129 	Opt_snapdirname,
130 	Opt_mds_namespace,
131 	Opt_fscache_uniq,
132 	Opt_last_string,
133 	/* string args above */
134 	Opt_dirstat,
135 	Opt_nodirstat,
136 	Opt_rbytes,
137 	Opt_norbytes,
138 	Opt_asyncreaddir,
139 	Opt_noasyncreaddir,
140 	Opt_dcache,
141 	Opt_nodcache,
142 	Opt_ino32,
143 	Opt_noino32,
144 	Opt_fscache,
145 	Opt_nofscache,
146 	Opt_poolperm,
147 	Opt_nopoolperm,
148 	Opt_require_active_mds,
149 	Opt_norequire_active_mds,
150 #ifdef CONFIG_CEPH_FS_POSIX_ACL
151 	Opt_acl,
152 #endif
153 	Opt_noacl,
154 };
155 
156 static match_table_t fsopt_tokens = {
157 	{Opt_wsize, "wsize=%d"},
158 	{Opt_rsize, "rsize=%d"},
159 	{Opt_rasize, "rasize=%d"},
160 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
161 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
162 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
163 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
164 	{Opt_congestion_kb, "write_congestion_kb=%d"},
165 	/* int args above */
166 	{Opt_snapdirname, "snapdirname=%s"},
167 	{Opt_mds_namespace, "mds_namespace=%s"},
168 	{Opt_fscache_uniq, "fsc=%s"},
169 	/* string args above */
170 	{Opt_dirstat, "dirstat"},
171 	{Opt_nodirstat, "nodirstat"},
172 	{Opt_rbytes, "rbytes"},
173 	{Opt_norbytes, "norbytes"},
174 	{Opt_asyncreaddir, "asyncreaddir"},
175 	{Opt_noasyncreaddir, "noasyncreaddir"},
176 	{Opt_dcache, "dcache"},
177 	{Opt_nodcache, "nodcache"},
178 	{Opt_ino32, "ino32"},
179 	{Opt_noino32, "noino32"},
180 	{Opt_fscache, "fsc"},
181 	{Opt_nofscache, "nofsc"},
182 	{Opt_poolperm, "poolperm"},
183 	{Opt_nopoolperm, "nopoolperm"},
184 	{Opt_require_active_mds, "require_active_mds"},
185 	{Opt_norequire_active_mds, "norequire_active_mds"},
186 #ifdef CONFIG_CEPH_FS_POSIX_ACL
187 	{Opt_acl, "acl"},
188 #endif
189 	{Opt_noacl, "noacl"},
190 	{-1, NULL}
191 };
192 
193 static int parse_fsopt_token(char *c, void *private)
194 {
195 	struct ceph_mount_options *fsopt = private;
196 	substring_t argstr[MAX_OPT_ARGS];
197 	int token, intval, ret;
198 
199 	token = match_token((char *)c, fsopt_tokens, argstr);
200 	if (token < 0)
201 		return -EINVAL;
202 
203 	if (token < Opt_last_int) {
204 		ret = match_int(&argstr[0], &intval);
205 		if (ret < 0) {
206 			pr_err("bad mount option arg (not int) "
207 			       "at '%s'\n", c);
208 			return ret;
209 		}
210 		dout("got int token %d val %d\n", token, intval);
211 	} else if (token > Opt_last_int && token < Opt_last_string) {
212 		dout("got string token %d val %s\n", token,
213 		     argstr[0].from);
214 	} else {
215 		dout("got token %d\n", token);
216 	}
217 
218 	switch (token) {
219 	case Opt_snapdirname:
220 		kfree(fsopt->snapdir_name);
221 		fsopt->snapdir_name = kstrndup(argstr[0].from,
222 					       argstr[0].to-argstr[0].from,
223 					       GFP_KERNEL);
224 		if (!fsopt->snapdir_name)
225 			return -ENOMEM;
226 		break;
227 	case Opt_mds_namespace:
228 		fsopt->mds_namespace = kstrndup(argstr[0].from,
229 						argstr[0].to-argstr[0].from,
230 						GFP_KERNEL);
231 		if (!fsopt->mds_namespace)
232 			return -ENOMEM;
233 		break;
234 	case Opt_fscache_uniq:
235 		fsopt->fscache_uniq = kstrndup(argstr[0].from,
236 					       argstr[0].to-argstr[0].from,
237 					       GFP_KERNEL);
238 		if (!fsopt->fscache_uniq)
239 			return -ENOMEM;
240 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
241 		break;
242 		/* misc */
243 	case Opt_wsize:
244 		if (intval < PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
245 			return -EINVAL;
246 		fsopt->wsize = ALIGN(intval, PAGE_SIZE);
247 		break;
248 	case Opt_rsize:
249 		if (intval < PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
250 			return -EINVAL;
251 		fsopt->rsize = ALIGN(intval, PAGE_SIZE);
252 		break;
253 	case Opt_rasize:
254 		if (intval < 0)
255 			return -EINVAL;
256 		fsopt->rasize = ALIGN(intval + PAGE_SIZE - 1, PAGE_SIZE);
257 		break;
258 	case Opt_caps_wanted_delay_min:
259 		if (intval < 1)
260 			return -EINVAL;
261 		fsopt->caps_wanted_delay_min = intval;
262 		break;
263 	case Opt_caps_wanted_delay_max:
264 		if (intval < 1)
265 			return -EINVAL;
266 		fsopt->caps_wanted_delay_max = intval;
267 		break;
268 	case Opt_readdir_max_entries:
269 		if (intval < 1)
270 			return -EINVAL;
271 		fsopt->max_readdir = intval;
272 		break;
273 	case Opt_readdir_max_bytes:
274 		if (intval < PAGE_SIZE && intval != 0)
275 			return -EINVAL;
276 		fsopt->max_readdir_bytes = intval;
277 		break;
278 	case Opt_congestion_kb:
279 		if (intval < 1024) /* at least 1M */
280 			return -EINVAL;
281 		fsopt->congestion_kb = intval;
282 		break;
283 	case Opt_dirstat:
284 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
285 		break;
286 	case Opt_nodirstat:
287 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
288 		break;
289 	case Opt_rbytes:
290 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
291 		break;
292 	case Opt_norbytes:
293 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
294 		break;
295 	case Opt_asyncreaddir:
296 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
297 		break;
298 	case Opt_noasyncreaddir:
299 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
300 		break;
301 	case Opt_dcache:
302 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
303 		break;
304 	case Opt_nodcache:
305 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
306 		break;
307 	case Opt_ino32:
308 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
309 		break;
310 	case Opt_noino32:
311 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
312 		break;
313 	case Opt_fscache:
314 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
315 		break;
316 	case Opt_nofscache:
317 		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
318 		break;
319 	case Opt_poolperm:
320 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
321 		printk ("pool perm");
322 		break;
323 	case Opt_nopoolperm:
324 		fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
325 		break;
326 	case Opt_require_active_mds:
327 		fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
328 		break;
329 	case Opt_norequire_active_mds:
330 		fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
331 		break;
332 #ifdef CONFIG_CEPH_FS_POSIX_ACL
333 	case Opt_acl:
334 		fsopt->sb_flags |= MS_POSIXACL;
335 		break;
336 #endif
337 	case Opt_noacl:
338 		fsopt->sb_flags &= ~MS_POSIXACL;
339 		break;
340 	default:
341 		BUG_ON(token);
342 	}
343 	return 0;
344 }
345 
346 static void destroy_mount_options(struct ceph_mount_options *args)
347 {
348 	dout("destroy_mount_options %p\n", args);
349 	kfree(args->snapdir_name);
350 	kfree(args->mds_namespace);
351 	kfree(args->server_path);
352 	kfree(args->fscache_uniq);
353 	kfree(args);
354 }
355 
356 static int strcmp_null(const char *s1, const char *s2)
357 {
358 	if (!s1 && !s2)
359 		return 0;
360 	if (s1 && !s2)
361 		return -1;
362 	if (!s1 && s2)
363 		return 1;
364 	return strcmp(s1, s2);
365 }
366 
367 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
368 				 struct ceph_options *new_opt,
369 				 struct ceph_fs_client *fsc)
370 {
371 	struct ceph_mount_options *fsopt1 = new_fsopt;
372 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
373 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
374 	int ret;
375 
376 	ret = memcmp(fsopt1, fsopt2, ofs);
377 	if (ret)
378 		return ret;
379 
380 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
381 	if (ret)
382 		return ret;
383 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
384 	if (ret)
385 		return ret;
386 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
387 	if (ret)
388 		return ret;
389 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
390 	if (ret)
391 		return ret;
392 
393 	return ceph_compare_options(new_opt, fsc->client);
394 }
395 
396 static int parse_mount_options(struct ceph_mount_options **pfsopt,
397 			       struct ceph_options **popt,
398 			       int flags, char *options,
399 			       const char *dev_name)
400 {
401 	struct ceph_mount_options *fsopt;
402 	const char *dev_name_end;
403 	int err;
404 
405 	if (!dev_name || !*dev_name)
406 		return -EINVAL;
407 
408 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
409 	if (!fsopt)
410 		return -ENOMEM;
411 
412 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
413 
414 	fsopt->sb_flags = flags;
415 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
416 
417 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
418 	fsopt->rsize = CEPH_MAX_READ_SIZE;
419 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
420 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
421 	if (!fsopt->snapdir_name) {
422 		err = -ENOMEM;
423 		goto out;
424 	}
425 
426 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
427 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
428 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
429 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
430 	fsopt->congestion_kb = default_congestion_kb();
431 
432 	/*
433 	 * Distinguish the server list from the path in "dev_name".
434 	 * Internally we do not include the leading '/' in the path.
435 	 *
436 	 * "dev_name" will look like:
437 	 *     <server_spec>[,<server_spec>...]:[<path>]
438 	 * where
439 	 *     <server_spec> is <ip>[:<port>]
440 	 *     <path> is optional, but if present must begin with '/'
441 	 */
442 	dev_name_end = strchr(dev_name, '/');
443 	if (dev_name_end) {
444 		if (strlen(dev_name_end) > 1) {
445 			fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
446 			if (!fsopt->server_path) {
447 				err = -ENOMEM;
448 				goto out;
449 			}
450 		}
451 	} else {
452 		dev_name_end = dev_name + strlen(dev_name);
453 	}
454 	err = -EINVAL;
455 	dev_name_end--;		/* back up to ':' separator */
456 	if (dev_name_end < dev_name || *dev_name_end != ':') {
457 		pr_err("device name is missing path (no : separator in %s)\n",
458 				dev_name);
459 		goto out;
460 	}
461 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
462 	if (fsopt->server_path)
463 		dout("server path '%s'\n", fsopt->server_path);
464 
465 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
466 				 parse_fsopt_token, (void *)fsopt);
467 	if (IS_ERR(*popt)) {
468 		err = PTR_ERR(*popt);
469 		goto out;
470 	}
471 
472 	/* success */
473 	*pfsopt = fsopt;
474 	return 0;
475 
476 out:
477 	destroy_mount_options(fsopt);
478 	return err;
479 }
480 
481 /**
482  * ceph_show_options - Show mount options in /proc/mounts
483  * @m: seq_file to write to
484  * @root: root of that (sub)tree
485  */
486 static int ceph_show_options(struct seq_file *m, struct dentry *root)
487 {
488 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
489 	struct ceph_mount_options *fsopt = fsc->mount_options;
490 	size_t pos;
491 	int ret;
492 
493 	/* a comma between MNT/MS and client options */
494 	seq_putc(m, ',');
495 	pos = m->count;
496 
497 	ret = ceph_print_client_options(m, fsc->client);
498 	if (ret)
499 		return ret;
500 
501 	/* retract our comma if no client options */
502 	if (m->count == pos)
503 		m->count--;
504 
505 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
506 		seq_puts(m, ",dirstat");
507 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
508 		seq_puts(m, ",rbytes");
509 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
510 		seq_puts(m, ",noasyncreaddir");
511 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
512 		seq_puts(m, ",nodcache");
513 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
514 		if (fsopt->fscache_uniq)
515 			seq_printf(m, ",fsc=%s", fsopt->fscache_uniq);
516 		else
517 			seq_puts(m, ",fsc");
518 	}
519 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
520 		seq_puts(m, ",nopoolperm");
521 
522 #ifdef CONFIG_CEPH_FS_POSIX_ACL
523 	if (fsopt->sb_flags & MS_POSIXACL)
524 		seq_puts(m, ",acl");
525 	else
526 		seq_puts(m, ",noacl");
527 #endif
528 
529 	if (fsopt->mds_namespace)
530 		seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
531 	if (fsopt->wsize)
532 		seq_printf(m, ",wsize=%d", fsopt->wsize);
533 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
534 		seq_printf(m, ",rsize=%d", fsopt->rsize);
535 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
536 		seq_printf(m, ",rasize=%d", fsopt->rasize);
537 	if (fsopt->congestion_kb != default_congestion_kb())
538 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
539 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
540 		seq_printf(m, ",caps_wanted_delay_min=%d",
541 			 fsopt->caps_wanted_delay_min);
542 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
543 		seq_printf(m, ",caps_wanted_delay_max=%d",
544 			   fsopt->caps_wanted_delay_max);
545 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
546 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
547 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
548 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
549 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
550 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
551 
552 	return 0;
553 }
554 
555 /*
556  * handle any mon messages the standard library doesn't understand.
557  * return error if we don't either.
558  */
559 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
560 {
561 	struct ceph_fs_client *fsc = client->private;
562 	int type = le16_to_cpu(msg->hdr.type);
563 
564 	switch (type) {
565 	case CEPH_MSG_MDS_MAP:
566 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
567 		return 0;
568 	case CEPH_MSG_FS_MAP_USER:
569 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
570 		return 0;
571 	default:
572 		return -1;
573 	}
574 }
575 
576 /*
577  * create a new fs client
578  */
579 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
580 					struct ceph_options *opt)
581 {
582 	struct ceph_fs_client *fsc;
583 	int page_count;
584 	size_t size;
585 	int err = -ENOMEM;
586 
587 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
588 	if (!fsc)
589 		return ERR_PTR(-ENOMEM);
590 
591 	fsc->client = ceph_create_client(opt, fsc);
592 	if (IS_ERR(fsc->client)) {
593 		err = PTR_ERR(fsc->client);
594 		goto fail;
595 	}
596 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
597 
598 	if (!fsopt->mds_namespace) {
599 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
600 				   0, true);
601 	} else {
602 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
603 				   0, false);
604 	}
605 
606 	fsc->mount_options = fsopt;
607 
608 	fsc->sb = NULL;
609 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
610 
611 	atomic_long_set(&fsc->writeback_count, 0);
612 
613 	err = -ENOMEM;
614 	/*
615 	 * The number of concurrent works can be high but they don't need
616 	 * to be processed in parallel, limit concurrency.
617 	 */
618 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
619 	if (!fsc->wb_wq)
620 		goto fail_client;
621 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
622 	if (!fsc->pg_inv_wq)
623 		goto fail_wb_wq;
624 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
625 	if (!fsc->trunc_wq)
626 		goto fail_pg_inv_wq;
627 
628 	/* set up mempools */
629 	err = -ENOMEM;
630 	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
631 	size = sizeof (struct page *) * (page_count ? page_count : 1);
632 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
633 	if (!fsc->wb_pagevec_pool)
634 		goto fail_trunc_wq;
635 
636 	/* caps */
637 	fsc->min_caps = fsopt->max_readdir;
638 
639 	return fsc;
640 
641 fail_trunc_wq:
642 	destroy_workqueue(fsc->trunc_wq);
643 fail_pg_inv_wq:
644 	destroy_workqueue(fsc->pg_inv_wq);
645 fail_wb_wq:
646 	destroy_workqueue(fsc->wb_wq);
647 fail_client:
648 	ceph_destroy_client(fsc->client);
649 fail:
650 	kfree(fsc);
651 	return ERR_PTR(err);
652 }
653 
654 static void destroy_fs_client(struct ceph_fs_client *fsc)
655 {
656 	dout("destroy_fs_client %p\n", fsc);
657 
658 	destroy_workqueue(fsc->wb_wq);
659 	destroy_workqueue(fsc->pg_inv_wq);
660 	destroy_workqueue(fsc->trunc_wq);
661 
662 	mempool_destroy(fsc->wb_pagevec_pool);
663 
664 	destroy_mount_options(fsc->mount_options);
665 
666 	ceph_destroy_client(fsc->client);
667 
668 	kfree(fsc);
669 	dout("destroy_fs_client %p done\n", fsc);
670 }
671 
672 /*
673  * caches
674  */
675 struct kmem_cache *ceph_inode_cachep;
676 struct kmem_cache *ceph_cap_cachep;
677 struct kmem_cache *ceph_cap_flush_cachep;
678 struct kmem_cache *ceph_dentry_cachep;
679 struct kmem_cache *ceph_file_cachep;
680 
681 static void ceph_inode_init_once(void *foo)
682 {
683 	struct ceph_inode_info *ci = foo;
684 	inode_init_once(&ci->vfs_inode);
685 }
686 
687 static int __init init_caches(void)
688 {
689 	int error = -ENOMEM;
690 
691 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
692 				      sizeof(struct ceph_inode_info),
693 				      __alignof__(struct ceph_inode_info),
694 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
695 				      SLAB_ACCOUNT, ceph_inode_init_once);
696 	if (!ceph_inode_cachep)
697 		return -ENOMEM;
698 
699 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
700 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
701 	if (!ceph_cap_cachep)
702 		goto bad_cap;
703 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
704 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
705 	if (!ceph_cap_flush_cachep)
706 		goto bad_cap_flush;
707 
708 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
709 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
710 	if (!ceph_dentry_cachep)
711 		goto bad_dentry;
712 
713 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
714 
715 	if (!ceph_file_cachep)
716 		goto bad_file;
717 
718 	if ((error = ceph_fscache_register()))
719 		goto bad_file;
720 
721 	return 0;
722 bad_file:
723 	kmem_cache_destroy(ceph_dentry_cachep);
724 bad_dentry:
725 	kmem_cache_destroy(ceph_cap_flush_cachep);
726 bad_cap_flush:
727 	kmem_cache_destroy(ceph_cap_cachep);
728 bad_cap:
729 	kmem_cache_destroy(ceph_inode_cachep);
730 	return error;
731 }
732 
733 static void destroy_caches(void)
734 {
735 	/*
736 	 * Make sure all delayed rcu free inodes are flushed before we
737 	 * destroy cache.
738 	 */
739 	rcu_barrier();
740 
741 	kmem_cache_destroy(ceph_inode_cachep);
742 	kmem_cache_destroy(ceph_cap_cachep);
743 	kmem_cache_destroy(ceph_cap_flush_cachep);
744 	kmem_cache_destroy(ceph_dentry_cachep);
745 	kmem_cache_destroy(ceph_file_cachep);
746 
747 	ceph_fscache_unregister();
748 }
749 
750 
751 /*
752  * ceph_umount_begin - initiate forced umount.  Tear down down the
753  * mount, skipping steps that may hang while waiting for server(s).
754  */
755 static void ceph_umount_begin(struct super_block *sb)
756 {
757 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
758 
759 	dout("ceph_umount_begin - starting forced umount\n");
760 	if (!fsc)
761 		return;
762 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
763 	ceph_mdsc_force_umount(fsc->mdsc);
764 	return;
765 }
766 
767 static const struct super_operations ceph_super_ops = {
768 	.alloc_inode	= ceph_alloc_inode,
769 	.destroy_inode	= ceph_destroy_inode,
770 	.write_inode    = ceph_write_inode,
771 	.drop_inode	= ceph_drop_inode,
772 	.sync_fs        = ceph_sync_fs,
773 	.put_super	= ceph_put_super,
774 	.show_options   = ceph_show_options,
775 	.statfs		= ceph_statfs,
776 	.umount_begin   = ceph_umount_begin,
777 };
778 
779 /*
780  * Bootstrap mount by opening the root directory.  Note the mount
781  * @started time from caller, and time out if this takes too long.
782  */
783 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
784 				       const char *path,
785 				       unsigned long started)
786 {
787 	struct ceph_mds_client *mdsc = fsc->mdsc;
788 	struct ceph_mds_request *req = NULL;
789 	int err;
790 	struct dentry *root;
791 
792 	/* open dir */
793 	dout("open_root_inode opening '%s'\n", path);
794 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
795 	if (IS_ERR(req))
796 		return ERR_CAST(req);
797 	req->r_path1 = kstrdup(path, GFP_NOFS);
798 	if (!req->r_path1) {
799 		root = ERR_PTR(-ENOMEM);
800 		goto out;
801 	}
802 
803 	req->r_ino1.ino = CEPH_INO_ROOT;
804 	req->r_ino1.snap = CEPH_NOSNAP;
805 	req->r_started = started;
806 	req->r_timeout = fsc->client->options->mount_timeout;
807 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
808 	req->r_num_caps = 2;
809 	err = ceph_mdsc_do_request(mdsc, NULL, req);
810 	if (err == 0) {
811 		struct inode *inode = req->r_target_inode;
812 		req->r_target_inode = NULL;
813 		dout("open_root_inode success\n");
814 		root = d_make_root(inode);
815 		if (!root) {
816 			root = ERR_PTR(-ENOMEM);
817 			goto out;
818 		}
819 		dout("open_root_inode success, root dentry is %p\n", root);
820 	} else {
821 		root = ERR_PTR(err);
822 	}
823 out:
824 	ceph_mdsc_put_request(req);
825 	return root;
826 }
827 
828 
829 
830 
831 /*
832  * mount: join the ceph cluster, and open root directory.
833  */
834 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
835 {
836 	int err;
837 	unsigned long started = jiffies;  /* note the start time */
838 	struct dentry *root;
839 	int first = 0;   /* first vfsmount for this super_block */
840 
841 	dout("mount start %p\n", fsc);
842 	mutex_lock(&fsc->client->mount_mutex);
843 
844 	if (!fsc->sb->s_root) {
845 		const char *path;
846 		err = __ceph_open_session(fsc->client, started);
847 		if (err < 0)
848 			goto out;
849 
850 		/* setup fscache */
851 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
852 			err = ceph_fscache_register_fs(fsc);
853 			if (err < 0)
854 				goto out;
855 		}
856 
857 		if (!fsc->mount_options->server_path) {
858 			path = "";
859 			dout("mount opening path \\t\n");
860 		} else {
861 			path = fsc->mount_options->server_path + 1;
862 			dout("mount opening path %s\n", path);
863 		}
864 		root = open_root_dentry(fsc, path, started);
865 		if (IS_ERR(root)) {
866 			err = PTR_ERR(root);
867 			goto out;
868 		}
869 		fsc->sb->s_root = dget(root);
870 		first = 1;
871 
872 		err = ceph_fs_debugfs_init(fsc);
873 		if (err < 0)
874 			goto fail;
875 	} else {
876 		root = dget(fsc->sb->s_root);
877 	}
878 
879 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
880 	dout("mount success\n");
881 	mutex_unlock(&fsc->client->mount_mutex);
882 	return root;
883 
884 fail:
885 	if (first) {
886 		dput(fsc->sb->s_root);
887 		fsc->sb->s_root = NULL;
888 	}
889 out:
890 	mutex_unlock(&fsc->client->mount_mutex);
891 	return ERR_PTR(err);
892 }
893 
894 static int ceph_set_super(struct super_block *s, void *data)
895 {
896 	struct ceph_fs_client *fsc = data;
897 	int ret;
898 
899 	dout("set_super %p data %p\n", s, data);
900 
901 	s->s_flags = fsc->mount_options->sb_flags;
902 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
903 
904 	s->s_xattr = ceph_xattr_handlers;
905 	s->s_fs_info = fsc;
906 	fsc->sb = s;
907 
908 	s->s_op = &ceph_super_ops;
909 	s->s_d_op = &ceph_dentry_ops;
910 	s->s_export_op = &ceph_export_ops;
911 
912 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
913 
914 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
915 	if (ret != 0)
916 		goto fail;
917 
918 	return ret;
919 
920 fail:
921 	s->s_fs_info = NULL;
922 	fsc->sb = NULL;
923 	return ret;
924 }
925 
926 /*
927  * share superblock if same fs AND options
928  */
929 static int ceph_compare_super(struct super_block *sb, void *data)
930 {
931 	struct ceph_fs_client *new = data;
932 	struct ceph_mount_options *fsopt = new->mount_options;
933 	struct ceph_options *opt = new->client->options;
934 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
935 
936 	dout("ceph_compare_super %p\n", sb);
937 
938 	if (compare_mount_options(fsopt, opt, other)) {
939 		dout("monitor(s)/mount options don't match\n");
940 		return 0;
941 	}
942 	if ((opt->flags & CEPH_OPT_FSID) &&
943 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
944 		dout("fsid doesn't match\n");
945 		return 0;
946 	}
947 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
948 		dout("flags differ\n");
949 		return 0;
950 	}
951 	return 1;
952 }
953 
954 /*
955  * construct our own bdi so we can control readahead, etc.
956  */
957 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
958 
959 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
960 {
961 	int err;
962 
963 	err = super_setup_bdi_name(sb, "ceph-%ld",
964 				   atomic_long_inc_return(&bdi_seq));
965 	if (err)
966 		return err;
967 
968 	/* set ra_pages based on rasize mount option? */
969 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
970 
971 	/* set io_pages based on max osd read size */
972 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
973 
974 	return 0;
975 }
976 
977 static struct dentry *ceph_mount(struct file_system_type *fs_type,
978 		       int flags, const char *dev_name, void *data)
979 {
980 	struct super_block *sb;
981 	struct ceph_fs_client *fsc;
982 	struct dentry *res;
983 	int err;
984 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
985 	struct ceph_mount_options *fsopt = NULL;
986 	struct ceph_options *opt = NULL;
987 
988 	dout("ceph_mount\n");
989 
990 #ifdef CONFIG_CEPH_FS_POSIX_ACL
991 	flags |= MS_POSIXACL;
992 #endif
993 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
994 	if (err < 0) {
995 		res = ERR_PTR(err);
996 		goto out_final;
997 	}
998 
999 	/* create client (which we may/may not use) */
1000 	fsc = create_fs_client(fsopt, opt);
1001 	if (IS_ERR(fsc)) {
1002 		res = ERR_CAST(fsc);
1003 		destroy_mount_options(fsopt);
1004 		ceph_destroy_options(opt);
1005 		goto out_final;
1006 	}
1007 
1008 	err = ceph_mdsc_init(fsc);
1009 	if (err < 0) {
1010 		res = ERR_PTR(err);
1011 		goto out;
1012 	}
1013 
1014 	if (ceph_test_opt(fsc->client, NOSHARE))
1015 		compare_super = NULL;
1016 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1017 	if (IS_ERR(sb)) {
1018 		res = ERR_CAST(sb);
1019 		goto out;
1020 	}
1021 
1022 	if (ceph_sb_to_client(sb) != fsc) {
1023 		ceph_mdsc_destroy(fsc);
1024 		destroy_fs_client(fsc);
1025 		fsc = ceph_sb_to_client(sb);
1026 		dout("get_sb got existing client %p\n", fsc);
1027 	} else {
1028 		dout("get_sb using new client %p\n", fsc);
1029 		err = ceph_setup_bdi(sb, fsc);
1030 		if (err < 0) {
1031 			res = ERR_PTR(err);
1032 			goto out_splat;
1033 		}
1034 	}
1035 
1036 	res = ceph_real_mount(fsc);
1037 	if (IS_ERR(res))
1038 		goto out_splat;
1039 	dout("root %p inode %p ino %llx.%llx\n", res,
1040 	     d_inode(res), ceph_vinop(d_inode(res)));
1041 	return res;
1042 
1043 out_splat:
1044 	ceph_mdsc_close_sessions(fsc->mdsc);
1045 	deactivate_locked_super(sb);
1046 	goto out_final;
1047 
1048 out:
1049 	ceph_mdsc_destroy(fsc);
1050 	destroy_fs_client(fsc);
1051 out_final:
1052 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1053 	return res;
1054 }
1055 
1056 static void ceph_kill_sb(struct super_block *s)
1057 {
1058 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1059 	dev_t dev = s->s_dev;
1060 
1061 	dout("kill_sb %p\n", s);
1062 
1063 	ceph_mdsc_pre_umount(fsc->mdsc);
1064 	generic_shutdown_super(s);
1065 
1066 	fsc->client->extra_mon_dispatch = NULL;
1067 	ceph_fs_debugfs_cleanup(fsc);
1068 
1069 	ceph_fscache_unregister_fs(fsc);
1070 
1071 	ceph_mdsc_destroy(fsc);
1072 
1073 	destroy_fs_client(fsc);
1074 	free_anon_bdev(dev);
1075 }
1076 
1077 static struct file_system_type ceph_fs_type = {
1078 	.owner		= THIS_MODULE,
1079 	.name		= "ceph",
1080 	.mount		= ceph_mount,
1081 	.kill_sb	= ceph_kill_sb,
1082 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1083 };
1084 MODULE_ALIAS_FS("ceph");
1085 
1086 static int __init init_ceph(void)
1087 {
1088 	int ret = init_caches();
1089 	if (ret)
1090 		goto out;
1091 
1092 	ceph_flock_init();
1093 	ceph_xattr_init();
1094 	ret = register_filesystem(&ceph_fs_type);
1095 	if (ret)
1096 		goto out_xattr;
1097 
1098 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1099 
1100 	return 0;
1101 
1102 out_xattr:
1103 	ceph_xattr_exit();
1104 	destroy_caches();
1105 out:
1106 	return ret;
1107 }
1108 
1109 static void __exit exit_ceph(void)
1110 {
1111 	dout("exit_ceph\n");
1112 	unregister_filesystem(&ceph_fs_type);
1113 	ceph_xattr_exit();
1114 	destroy_caches();
1115 }
1116 
1117 module_init(init_ceph);
1118 module_exit(exit_ceph);
1119 
1120 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1121 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1122 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1123 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1124 MODULE_LICENSE("GPL");
1125