xref: /openbmc/linux/fs/ceph/super.c (revision e3b9f1e8)
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 		kfree(fsopt->mds_namespace);
229 		fsopt->mds_namespace = kstrndup(argstr[0].from,
230 						argstr[0].to-argstr[0].from,
231 						GFP_KERNEL);
232 		if (!fsopt->mds_namespace)
233 			return -ENOMEM;
234 		break;
235 	case Opt_fscache_uniq:
236 		kfree(fsopt->fscache_uniq);
237 		fsopt->fscache_uniq = kstrndup(argstr[0].from,
238 					       argstr[0].to-argstr[0].from,
239 					       GFP_KERNEL);
240 		if (!fsopt->fscache_uniq)
241 			return -ENOMEM;
242 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
243 		break;
244 		/* misc */
245 	case Opt_wsize:
246 		if (intval < PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
247 			return -EINVAL;
248 		fsopt->wsize = ALIGN(intval, PAGE_SIZE);
249 		break;
250 	case Opt_rsize:
251 		if (intval < PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
252 			return -EINVAL;
253 		fsopt->rsize = ALIGN(intval, PAGE_SIZE);
254 		break;
255 	case Opt_rasize:
256 		if (intval < 0)
257 			return -EINVAL;
258 		fsopt->rasize = ALIGN(intval + PAGE_SIZE - 1, PAGE_SIZE);
259 		break;
260 	case Opt_caps_wanted_delay_min:
261 		if (intval < 1)
262 			return -EINVAL;
263 		fsopt->caps_wanted_delay_min = intval;
264 		break;
265 	case Opt_caps_wanted_delay_max:
266 		if (intval < 1)
267 			return -EINVAL;
268 		fsopt->caps_wanted_delay_max = intval;
269 		break;
270 	case Opt_readdir_max_entries:
271 		if (intval < 1)
272 			return -EINVAL;
273 		fsopt->max_readdir = intval;
274 		break;
275 	case Opt_readdir_max_bytes:
276 		if (intval < PAGE_SIZE && intval != 0)
277 			return -EINVAL;
278 		fsopt->max_readdir_bytes = intval;
279 		break;
280 	case Opt_congestion_kb:
281 		if (intval < 1024) /* at least 1M */
282 			return -EINVAL;
283 		fsopt->congestion_kb = intval;
284 		break;
285 	case Opt_dirstat:
286 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
287 		break;
288 	case Opt_nodirstat:
289 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
290 		break;
291 	case Opt_rbytes:
292 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
293 		break;
294 	case Opt_norbytes:
295 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
296 		break;
297 	case Opt_asyncreaddir:
298 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
299 		break;
300 	case Opt_noasyncreaddir:
301 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
302 		break;
303 	case Opt_dcache:
304 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
305 		break;
306 	case Opt_nodcache:
307 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
308 		break;
309 	case Opt_ino32:
310 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
311 		break;
312 	case Opt_noino32:
313 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
314 		break;
315 	case Opt_fscache:
316 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
317 		break;
318 	case Opt_nofscache:
319 		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
320 		break;
321 	case Opt_poolperm:
322 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
323 		printk ("pool perm");
324 		break;
325 	case Opt_nopoolperm:
326 		fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
327 		break;
328 	case Opt_require_active_mds:
329 		fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
330 		break;
331 	case Opt_norequire_active_mds:
332 		fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
333 		break;
334 #ifdef CONFIG_CEPH_FS_POSIX_ACL
335 	case Opt_acl:
336 		fsopt->sb_flags |= SB_POSIXACL;
337 		break;
338 #endif
339 	case Opt_noacl:
340 		fsopt->sb_flags &= ~SB_POSIXACL;
341 		break;
342 	default:
343 		BUG_ON(token);
344 	}
345 	return 0;
346 }
347 
348 static void destroy_mount_options(struct ceph_mount_options *args)
349 {
350 	dout("destroy_mount_options %p\n", args);
351 	kfree(args->snapdir_name);
352 	kfree(args->mds_namespace);
353 	kfree(args->server_path);
354 	kfree(args->fscache_uniq);
355 	kfree(args);
356 }
357 
358 static int strcmp_null(const char *s1, const char *s2)
359 {
360 	if (!s1 && !s2)
361 		return 0;
362 	if (s1 && !s2)
363 		return -1;
364 	if (!s1 && s2)
365 		return 1;
366 	return strcmp(s1, s2);
367 }
368 
369 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
370 				 struct ceph_options *new_opt,
371 				 struct ceph_fs_client *fsc)
372 {
373 	struct ceph_mount_options *fsopt1 = new_fsopt;
374 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
375 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
376 	int ret;
377 
378 	ret = memcmp(fsopt1, fsopt2, ofs);
379 	if (ret)
380 		return ret;
381 
382 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
383 	if (ret)
384 		return ret;
385 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
386 	if (ret)
387 		return ret;
388 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
389 	if (ret)
390 		return ret;
391 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
392 	if (ret)
393 		return ret;
394 
395 	return ceph_compare_options(new_opt, fsc->client);
396 }
397 
398 static int parse_mount_options(struct ceph_mount_options **pfsopt,
399 			       struct ceph_options **popt,
400 			       int flags, char *options,
401 			       const char *dev_name)
402 {
403 	struct ceph_mount_options *fsopt;
404 	const char *dev_name_end;
405 	int err;
406 
407 	if (!dev_name || !*dev_name)
408 		return -EINVAL;
409 
410 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
411 	if (!fsopt)
412 		return -ENOMEM;
413 
414 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
415 
416 	fsopt->sb_flags = flags;
417 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
418 
419 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
420 	fsopt->rsize = CEPH_MAX_READ_SIZE;
421 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
422 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
423 	if (!fsopt->snapdir_name) {
424 		err = -ENOMEM;
425 		goto out;
426 	}
427 
428 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
429 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
430 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
431 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
432 	fsopt->congestion_kb = default_congestion_kb();
433 
434 	/*
435 	 * Distinguish the server list from the path in "dev_name".
436 	 * Internally we do not include the leading '/' in the path.
437 	 *
438 	 * "dev_name" will look like:
439 	 *     <server_spec>[,<server_spec>...]:[<path>]
440 	 * where
441 	 *     <server_spec> is <ip>[:<port>]
442 	 *     <path> is optional, but if present must begin with '/'
443 	 */
444 	dev_name_end = strchr(dev_name, '/');
445 	if (dev_name_end) {
446 		if (strlen(dev_name_end) > 1) {
447 			fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
448 			if (!fsopt->server_path) {
449 				err = -ENOMEM;
450 				goto out;
451 			}
452 		}
453 	} else {
454 		dev_name_end = dev_name + strlen(dev_name);
455 	}
456 	err = -EINVAL;
457 	dev_name_end--;		/* back up to ':' separator */
458 	if (dev_name_end < dev_name || *dev_name_end != ':') {
459 		pr_err("device name is missing path (no : separator in %s)\n",
460 				dev_name);
461 		goto out;
462 	}
463 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
464 	if (fsopt->server_path)
465 		dout("server path '%s'\n", fsopt->server_path);
466 
467 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
468 				 parse_fsopt_token, (void *)fsopt);
469 	if (IS_ERR(*popt)) {
470 		err = PTR_ERR(*popt);
471 		goto out;
472 	}
473 
474 	/* success */
475 	*pfsopt = fsopt;
476 	return 0;
477 
478 out:
479 	destroy_mount_options(fsopt);
480 	return err;
481 }
482 
483 /**
484  * ceph_show_options - Show mount options in /proc/mounts
485  * @m: seq_file to write to
486  * @root: root of that (sub)tree
487  */
488 static int ceph_show_options(struct seq_file *m, struct dentry *root)
489 {
490 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
491 	struct ceph_mount_options *fsopt = fsc->mount_options;
492 	size_t pos;
493 	int ret;
494 
495 	/* a comma between MNT/MS and client options */
496 	seq_putc(m, ',');
497 	pos = m->count;
498 
499 	ret = ceph_print_client_options(m, fsc->client);
500 	if (ret)
501 		return ret;
502 
503 	/* retract our comma if no client options */
504 	if (m->count == pos)
505 		m->count--;
506 
507 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
508 		seq_puts(m, ",dirstat");
509 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
510 		seq_puts(m, ",rbytes");
511 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
512 		seq_puts(m, ",noasyncreaddir");
513 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
514 		seq_puts(m, ",nodcache");
515 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
516 		if (fsopt->fscache_uniq)
517 			seq_printf(m, ",fsc=%s", fsopt->fscache_uniq);
518 		else
519 			seq_puts(m, ",fsc");
520 	}
521 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
522 		seq_puts(m, ",nopoolperm");
523 
524 #ifdef CONFIG_CEPH_FS_POSIX_ACL
525 	if (fsopt->sb_flags & SB_POSIXACL)
526 		seq_puts(m, ",acl");
527 	else
528 		seq_puts(m, ",noacl");
529 #endif
530 
531 	if (fsopt->mds_namespace)
532 		seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
533 	if (fsopt->wsize)
534 		seq_printf(m, ",wsize=%d", fsopt->wsize);
535 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
536 		seq_printf(m, ",rsize=%d", fsopt->rsize);
537 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
538 		seq_printf(m, ",rasize=%d", fsopt->rasize);
539 	if (fsopt->congestion_kb != default_congestion_kb())
540 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
541 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
542 		seq_printf(m, ",caps_wanted_delay_min=%d",
543 			 fsopt->caps_wanted_delay_min);
544 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
545 		seq_printf(m, ",caps_wanted_delay_max=%d",
546 			   fsopt->caps_wanted_delay_max);
547 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
548 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
549 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
550 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
551 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
552 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
553 
554 	return 0;
555 }
556 
557 /*
558  * handle any mon messages the standard library doesn't understand.
559  * return error if we don't either.
560  */
561 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
562 {
563 	struct ceph_fs_client *fsc = client->private;
564 	int type = le16_to_cpu(msg->hdr.type);
565 
566 	switch (type) {
567 	case CEPH_MSG_MDS_MAP:
568 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
569 		return 0;
570 	case CEPH_MSG_FS_MAP_USER:
571 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
572 		return 0;
573 	default:
574 		return -1;
575 	}
576 }
577 
578 /*
579  * create a new fs client
580  */
581 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
582 					struct ceph_options *opt)
583 {
584 	struct ceph_fs_client *fsc;
585 	int page_count;
586 	size_t size;
587 	int err = -ENOMEM;
588 
589 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
590 	if (!fsc)
591 		return ERR_PTR(-ENOMEM);
592 
593 	fsc->client = ceph_create_client(opt, fsc);
594 	if (IS_ERR(fsc->client)) {
595 		err = PTR_ERR(fsc->client);
596 		goto fail;
597 	}
598 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
599 
600 	if (!fsopt->mds_namespace) {
601 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
602 				   0, true);
603 	} else {
604 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
605 				   0, false);
606 	}
607 
608 	fsc->mount_options = fsopt;
609 
610 	fsc->sb = NULL;
611 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
612 
613 	atomic_long_set(&fsc->writeback_count, 0);
614 
615 	err = -ENOMEM;
616 	/*
617 	 * The number of concurrent works can be high but they don't need
618 	 * to be processed in parallel, limit concurrency.
619 	 */
620 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
621 	if (!fsc->wb_wq)
622 		goto fail_client;
623 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
624 	if (!fsc->pg_inv_wq)
625 		goto fail_wb_wq;
626 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
627 	if (!fsc->trunc_wq)
628 		goto fail_pg_inv_wq;
629 
630 	/* set up mempools */
631 	err = -ENOMEM;
632 	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
633 	size = sizeof (struct page *) * (page_count ? page_count : 1);
634 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
635 	if (!fsc->wb_pagevec_pool)
636 		goto fail_trunc_wq;
637 
638 	/* caps */
639 	fsc->min_caps = fsopt->max_readdir;
640 
641 	return fsc;
642 
643 fail_trunc_wq:
644 	destroy_workqueue(fsc->trunc_wq);
645 fail_pg_inv_wq:
646 	destroy_workqueue(fsc->pg_inv_wq);
647 fail_wb_wq:
648 	destroy_workqueue(fsc->wb_wq);
649 fail_client:
650 	ceph_destroy_client(fsc->client);
651 fail:
652 	kfree(fsc);
653 	return ERR_PTR(err);
654 }
655 
656 static void destroy_fs_client(struct ceph_fs_client *fsc)
657 {
658 	dout("destroy_fs_client %p\n", fsc);
659 
660 	destroy_workqueue(fsc->wb_wq);
661 	destroy_workqueue(fsc->pg_inv_wq);
662 	destroy_workqueue(fsc->trunc_wq);
663 
664 	mempool_destroy(fsc->wb_pagevec_pool);
665 
666 	destroy_mount_options(fsc->mount_options);
667 
668 	ceph_destroy_client(fsc->client);
669 
670 	kfree(fsc);
671 	dout("destroy_fs_client %p done\n", fsc);
672 }
673 
674 /*
675  * caches
676  */
677 struct kmem_cache *ceph_inode_cachep;
678 struct kmem_cache *ceph_cap_cachep;
679 struct kmem_cache *ceph_cap_flush_cachep;
680 struct kmem_cache *ceph_dentry_cachep;
681 struct kmem_cache *ceph_file_cachep;
682 
683 static void ceph_inode_init_once(void *foo)
684 {
685 	struct ceph_inode_info *ci = foo;
686 	inode_init_once(&ci->vfs_inode);
687 }
688 
689 static int __init init_caches(void)
690 {
691 	int error = -ENOMEM;
692 
693 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
694 				      sizeof(struct ceph_inode_info),
695 				      __alignof__(struct ceph_inode_info),
696 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
697 				      SLAB_ACCOUNT, ceph_inode_init_once);
698 	if (!ceph_inode_cachep)
699 		return -ENOMEM;
700 
701 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
702 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
703 	if (!ceph_cap_cachep)
704 		goto bad_cap;
705 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
706 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
707 	if (!ceph_cap_flush_cachep)
708 		goto bad_cap_flush;
709 
710 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
711 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
712 	if (!ceph_dentry_cachep)
713 		goto bad_dentry;
714 
715 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
716 	if (!ceph_file_cachep)
717 		goto bad_file;
718 
719 	error = ceph_fscache_register();
720 	if (error)
721 		goto bad_fscache;
722 
723 	return 0;
724 
725 bad_fscache:
726 	kmem_cache_destroy(ceph_file_cachep);
727 bad_file:
728 	kmem_cache_destroy(ceph_dentry_cachep);
729 bad_dentry:
730 	kmem_cache_destroy(ceph_cap_flush_cachep);
731 bad_cap_flush:
732 	kmem_cache_destroy(ceph_cap_cachep);
733 bad_cap:
734 	kmem_cache_destroy(ceph_inode_cachep);
735 	return error;
736 }
737 
738 static void destroy_caches(void)
739 {
740 	/*
741 	 * Make sure all delayed rcu free inodes are flushed before we
742 	 * destroy cache.
743 	 */
744 	rcu_barrier();
745 
746 	kmem_cache_destroy(ceph_inode_cachep);
747 	kmem_cache_destroy(ceph_cap_cachep);
748 	kmem_cache_destroy(ceph_cap_flush_cachep);
749 	kmem_cache_destroy(ceph_dentry_cachep);
750 	kmem_cache_destroy(ceph_file_cachep);
751 
752 	ceph_fscache_unregister();
753 }
754 
755 
756 /*
757  * ceph_umount_begin - initiate forced umount.  Tear down down the
758  * mount, skipping steps that may hang while waiting for server(s).
759  */
760 static void ceph_umount_begin(struct super_block *sb)
761 {
762 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
763 
764 	dout("ceph_umount_begin - starting forced umount\n");
765 	if (!fsc)
766 		return;
767 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
768 	ceph_mdsc_force_umount(fsc->mdsc);
769 	return;
770 }
771 
772 static const struct super_operations ceph_super_ops = {
773 	.alloc_inode	= ceph_alloc_inode,
774 	.destroy_inode	= ceph_destroy_inode,
775 	.write_inode    = ceph_write_inode,
776 	.drop_inode	= ceph_drop_inode,
777 	.sync_fs        = ceph_sync_fs,
778 	.put_super	= ceph_put_super,
779 	.show_options   = ceph_show_options,
780 	.statfs		= ceph_statfs,
781 	.umount_begin   = ceph_umount_begin,
782 };
783 
784 /*
785  * Bootstrap mount by opening the root directory.  Note the mount
786  * @started time from caller, and time out if this takes too long.
787  */
788 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
789 				       const char *path,
790 				       unsigned long started)
791 {
792 	struct ceph_mds_client *mdsc = fsc->mdsc;
793 	struct ceph_mds_request *req = NULL;
794 	int err;
795 	struct dentry *root;
796 
797 	/* open dir */
798 	dout("open_root_inode opening '%s'\n", path);
799 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
800 	if (IS_ERR(req))
801 		return ERR_CAST(req);
802 	req->r_path1 = kstrdup(path, GFP_NOFS);
803 	if (!req->r_path1) {
804 		root = ERR_PTR(-ENOMEM);
805 		goto out;
806 	}
807 
808 	req->r_ino1.ino = CEPH_INO_ROOT;
809 	req->r_ino1.snap = CEPH_NOSNAP;
810 	req->r_started = started;
811 	req->r_timeout = fsc->client->options->mount_timeout;
812 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
813 	req->r_num_caps = 2;
814 	err = ceph_mdsc_do_request(mdsc, NULL, req);
815 	if (err == 0) {
816 		struct inode *inode = req->r_target_inode;
817 		req->r_target_inode = NULL;
818 		dout("open_root_inode success\n");
819 		root = d_make_root(inode);
820 		if (!root) {
821 			root = ERR_PTR(-ENOMEM);
822 			goto out;
823 		}
824 		dout("open_root_inode success, root dentry is %p\n", root);
825 	} else {
826 		root = ERR_PTR(err);
827 	}
828 out:
829 	ceph_mdsc_put_request(req);
830 	return root;
831 }
832 
833 
834 
835 
836 /*
837  * mount: join the ceph cluster, and open root directory.
838  */
839 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
840 {
841 	int err;
842 	unsigned long started = jiffies;  /* note the start time */
843 	struct dentry *root;
844 
845 	dout("mount start %p\n", fsc);
846 	mutex_lock(&fsc->client->mount_mutex);
847 
848 	if (!fsc->sb->s_root) {
849 		const char *path;
850 		err = __ceph_open_session(fsc->client, started);
851 		if (err < 0)
852 			goto out;
853 
854 		/* setup fscache */
855 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
856 			err = ceph_fscache_register_fs(fsc);
857 			if (err < 0)
858 				goto out;
859 		}
860 
861 		if (!fsc->mount_options->server_path) {
862 			path = "";
863 			dout("mount opening path \\t\n");
864 		} else {
865 			path = fsc->mount_options->server_path + 1;
866 			dout("mount opening path %s\n", path);
867 		}
868 
869 		err = ceph_fs_debugfs_init(fsc);
870 		if (err < 0)
871 			goto out;
872 
873 		root = open_root_dentry(fsc, path, started);
874 		if (IS_ERR(root)) {
875 			err = PTR_ERR(root);
876 			goto out;
877 		}
878 		fsc->sb->s_root = dget(root);
879 	} else {
880 		root = dget(fsc->sb->s_root);
881 	}
882 
883 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
884 	dout("mount success\n");
885 	mutex_unlock(&fsc->client->mount_mutex);
886 	return root;
887 
888 out:
889 	mutex_unlock(&fsc->client->mount_mutex);
890 	return ERR_PTR(err);
891 }
892 
893 static int ceph_set_super(struct super_block *s, void *data)
894 {
895 	struct ceph_fs_client *fsc = data;
896 	int ret;
897 
898 	dout("set_super %p data %p\n", s, data);
899 
900 	s->s_flags = fsc->mount_options->sb_flags;
901 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
902 
903 	s->s_xattr = ceph_xattr_handlers;
904 	s->s_fs_info = fsc;
905 	fsc->sb = s;
906 
907 	s->s_op = &ceph_super_ops;
908 	s->s_d_op = &ceph_dentry_ops;
909 	s->s_export_op = &ceph_export_ops;
910 
911 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
912 
913 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
914 	if (ret != 0)
915 		goto fail;
916 
917 	return ret;
918 
919 fail:
920 	s->s_fs_info = NULL;
921 	fsc->sb = NULL;
922 	return ret;
923 }
924 
925 /*
926  * share superblock if same fs AND options
927  */
928 static int ceph_compare_super(struct super_block *sb, void *data)
929 {
930 	struct ceph_fs_client *new = data;
931 	struct ceph_mount_options *fsopt = new->mount_options;
932 	struct ceph_options *opt = new->client->options;
933 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
934 
935 	dout("ceph_compare_super %p\n", sb);
936 
937 	if (compare_mount_options(fsopt, opt, other)) {
938 		dout("monitor(s)/mount options don't match\n");
939 		return 0;
940 	}
941 	if ((opt->flags & CEPH_OPT_FSID) &&
942 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
943 		dout("fsid doesn't match\n");
944 		return 0;
945 	}
946 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
947 		dout("flags differ\n");
948 		return 0;
949 	}
950 	return 1;
951 }
952 
953 /*
954  * construct our own bdi so we can control readahead, etc.
955  */
956 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
957 
958 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
959 {
960 	int err;
961 
962 	err = super_setup_bdi_name(sb, "ceph-%ld",
963 				   atomic_long_inc_return(&bdi_seq));
964 	if (err)
965 		return err;
966 
967 	/* set ra_pages based on rasize mount option? */
968 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
969 
970 	/* set io_pages based on max osd read size */
971 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
972 
973 	return 0;
974 }
975 
976 static struct dentry *ceph_mount(struct file_system_type *fs_type,
977 		       int flags, const char *dev_name, void *data)
978 {
979 	struct super_block *sb;
980 	struct ceph_fs_client *fsc;
981 	struct dentry *res;
982 	int err;
983 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
984 	struct ceph_mount_options *fsopt = NULL;
985 	struct ceph_options *opt = NULL;
986 
987 	dout("ceph_mount\n");
988 
989 #ifdef CONFIG_CEPH_FS_POSIX_ACL
990 	flags |= SB_POSIXACL;
991 #endif
992 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
993 	if (err < 0) {
994 		res = ERR_PTR(err);
995 		goto out_final;
996 	}
997 
998 	/* create client (which we may/may not use) */
999 	fsc = create_fs_client(fsopt, opt);
1000 	if (IS_ERR(fsc)) {
1001 		res = ERR_CAST(fsc);
1002 		destroy_mount_options(fsopt);
1003 		ceph_destroy_options(opt);
1004 		goto out_final;
1005 	}
1006 
1007 	err = ceph_mdsc_init(fsc);
1008 	if (err < 0) {
1009 		res = ERR_PTR(err);
1010 		goto out;
1011 	}
1012 
1013 	if (ceph_test_opt(fsc->client, NOSHARE))
1014 		compare_super = NULL;
1015 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1016 	if (IS_ERR(sb)) {
1017 		res = ERR_CAST(sb);
1018 		goto out;
1019 	}
1020 
1021 	if (ceph_sb_to_client(sb) != fsc) {
1022 		ceph_mdsc_destroy(fsc);
1023 		destroy_fs_client(fsc);
1024 		fsc = ceph_sb_to_client(sb);
1025 		dout("get_sb got existing client %p\n", fsc);
1026 	} else {
1027 		dout("get_sb using new client %p\n", fsc);
1028 		err = ceph_setup_bdi(sb, fsc);
1029 		if (err < 0) {
1030 			res = ERR_PTR(err);
1031 			goto out_splat;
1032 		}
1033 	}
1034 
1035 	res = ceph_real_mount(fsc);
1036 	if (IS_ERR(res))
1037 		goto out_splat;
1038 	dout("root %p inode %p ino %llx.%llx\n", res,
1039 	     d_inode(res), ceph_vinop(d_inode(res)));
1040 	return res;
1041 
1042 out_splat:
1043 	ceph_mdsc_close_sessions(fsc->mdsc);
1044 	deactivate_locked_super(sb);
1045 	goto out_final;
1046 
1047 out:
1048 	ceph_mdsc_destroy(fsc);
1049 	destroy_fs_client(fsc);
1050 out_final:
1051 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1052 	return res;
1053 }
1054 
1055 static void ceph_kill_sb(struct super_block *s)
1056 {
1057 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1058 	dev_t dev = s->s_dev;
1059 
1060 	dout("kill_sb %p\n", s);
1061 
1062 	ceph_mdsc_pre_umount(fsc->mdsc);
1063 	generic_shutdown_super(s);
1064 
1065 	fsc->client->extra_mon_dispatch = NULL;
1066 	ceph_fs_debugfs_cleanup(fsc);
1067 
1068 	ceph_fscache_unregister_fs(fsc);
1069 
1070 	ceph_mdsc_destroy(fsc);
1071 
1072 	destroy_fs_client(fsc);
1073 	free_anon_bdev(dev);
1074 }
1075 
1076 static struct file_system_type ceph_fs_type = {
1077 	.owner		= THIS_MODULE,
1078 	.name		= "ceph",
1079 	.mount		= ceph_mount,
1080 	.kill_sb	= ceph_kill_sb,
1081 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1082 };
1083 MODULE_ALIAS_FS("ceph");
1084 
1085 static int __init init_ceph(void)
1086 {
1087 	int ret = init_caches();
1088 	if (ret)
1089 		goto out;
1090 
1091 	ceph_flock_init();
1092 	ceph_xattr_init();
1093 	ret = register_filesystem(&ceph_fs_type);
1094 	if (ret)
1095 		goto out_xattr;
1096 
1097 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1098 
1099 	return 0;
1100 
1101 out_xattr:
1102 	ceph_xattr_exit();
1103 	destroy_caches();
1104 out:
1105 	return ret;
1106 }
1107 
1108 static void __exit exit_ceph(void)
1109 {
1110 	dout("exit_ceph\n");
1111 	unregister_filesystem(&ceph_fs_type);
1112 	ceph_xattr_exit();
1113 	destroy_caches();
1114 }
1115 
1116 module_init(init_ceph);
1117 module_exit(exit_ceph);
1118 
1119 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1120 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1121 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1122 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1123 MODULE_LICENSE("GPL");
1124