xref: /openbmc/linux/fs/ceph/super.c (revision bc5aa3a0)
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 
53 	dout("statfs\n");
54 	err = ceph_monc_do_statfs(&fsc->client->monc, &st);
55 	if (err < 0)
56 		return err;
57 
58 	/* fill in kstatfs */
59 	buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
60 
61 	/*
62 	 * express utilization in terms of large blocks to avoid
63 	 * overflow on 32-bit machines.
64 	 *
65 	 * NOTE: for the time being, we make bsize == frsize to humor
66 	 * not-yet-ancient versions of glibc that are broken.
67 	 * Someday, we will probably want to report a real block
68 	 * size...  whatever that may mean for a network file system!
69 	 */
70 	buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
71 	buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
72 	buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
73 	buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
74 	buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
75 
76 	buf->f_files = le64_to_cpu(st.num_objects);
77 	buf->f_ffree = -1;
78 	buf->f_namelen = NAME_MAX;
79 
80 	/* leave fsid little-endian, regardless of host endianness */
81 	fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
82 	buf->f_fsid.val[0] = fsid & 0xffffffff;
83 	buf->f_fsid.val[1] = fsid >> 32;
84 
85 	return 0;
86 }
87 
88 
89 static int ceph_sync_fs(struct super_block *sb, int wait)
90 {
91 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
92 
93 	if (!wait) {
94 		dout("sync_fs (non-blocking)\n");
95 		ceph_flush_dirty_caps(fsc->mdsc);
96 		dout("sync_fs (non-blocking) done\n");
97 		return 0;
98 	}
99 
100 	dout("sync_fs (blocking)\n");
101 	ceph_osdc_sync(&fsc->client->osdc);
102 	ceph_mdsc_sync(fsc->mdsc);
103 	dout("sync_fs (blocking) done\n");
104 	return 0;
105 }
106 
107 /*
108  * mount options
109  */
110 enum {
111 	Opt_wsize,
112 	Opt_rsize,
113 	Opt_rasize,
114 	Opt_caps_wanted_delay_min,
115 	Opt_caps_wanted_delay_max,
116 	Opt_cap_release_safety,
117 	Opt_readdir_max_entries,
118 	Opt_readdir_max_bytes,
119 	Opt_congestion_kb,
120 	Opt_last_int,
121 	/* int args above */
122 	Opt_snapdirname,
123 	Opt_mds_namespace,
124 	Opt_last_string,
125 	/* string args above */
126 	Opt_dirstat,
127 	Opt_nodirstat,
128 	Opt_rbytes,
129 	Opt_norbytes,
130 	Opt_asyncreaddir,
131 	Opt_noasyncreaddir,
132 	Opt_dcache,
133 	Opt_nodcache,
134 	Opt_ino32,
135 	Opt_noino32,
136 	Opt_fscache,
137 	Opt_nofscache,
138 	Opt_poolperm,
139 	Opt_nopoolperm,
140 #ifdef CONFIG_CEPH_FS_POSIX_ACL
141 	Opt_acl,
142 #endif
143 	Opt_noacl,
144 };
145 
146 static match_table_t fsopt_tokens = {
147 	{Opt_wsize, "wsize=%d"},
148 	{Opt_rsize, "rsize=%d"},
149 	{Opt_rasize, "rasize=%d"},
150 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
151 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
152 	{Opt_cap_release_safety, "cap_release_safety=%d"},
153 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
154 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
155 	{Opt_congestion_kb, "write_congestion_kb=%d"},
156 	/* int args above */
157 	{Opt_snapdirname, "snapdirname=%s"},
158 	{Opt_mds_namespace, "mds_namespace=%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 	{Opt_fscache, "fsc"},
171 	{Opt_nofscache, "nofsc"},
172 	{Opt_poolperm, "poolperm"},
173 	{Opt_nopoolperm, "nopoolperm"},
174 #ifdef CONFIG_CEPH_FS_POSIX_ACL
175 	{Opt_acl, "acl"},
176 #endif
177 	{Opt_noacl, "noacl"},
178 	{-1, NULL}
179 };
180 
181 static int parse_fsopt_token(char *c, void *private)
182 {
183 	struct ceph_mount_options *fsopt = private;
184 	substring_t argstr[MAX_OPT_ARGS];
185 	int token, intval, ret;
186 
187 	token = match_token((char *)c, fsopt_tokens, argstr);
188 	if (token < 0)
189 		return -EINVAL;
190 
191 	if (token < Opt_last_int) {
192 		ret = match_int(&argstr[0], &intval);
193 		if (ret < 0) {
194 			pr_err("bad mount option arg (not int) "
195 			       "at '%s'\n", c);
196 			return ret;
197 		}
198 		dout("got int token %d val %d\n", token, intval);
199 	} else if (token > Opt_last_int && token < Opt_last_string) {
200 		dout("got string token %d val %s\n", token,
201 		     argstr[0].from);
202 	} else {
203 		dout("got token %d\n", token);
204 	}
205 
206 	switch (token) {
207 	case Opt_snapdirname:
208 		kfree(fsopt->snapdir_name);
209 		fsopt->snapdir_name = kstrndup(argstr[0].from,
210 					       argstr[0].to-argstr[0].from,
211 					       GFP_KERNEL);
212 		if (!fsopt->snapdir_name)
213 			return -ENOMEM;
214 		break;
215 	case Opt_mds_namespace:
216 		fsopt->mds_namespace = kstrndup(argstr[0].from,
217 						argstr[0].to-argstr[0].from,
218 						GFP_KERNEL);
219 		if (!fsopt->mds_namespace)
220 			return -ENOMEM;
221 		break;
222 		/* misc */
223 	case Opt_wsize:
224 		fsopt->wsize = intval;
225 		break;
226 	case Opt_rsize:
227 		fsopt->rsize = intval;
228 		break;
229 	case Opt_rasize:
230 		fsopt->rasize = intval;
231 		break;
232 	case Opt_caps_wanted_delay_min:
233 		fsopt->caps_wanted_delay_min = intval;
234 		break;
235 	case Opt_caps_wanted_delay_max:
236 		fsopt->caps_wanted_delay_max = intval;
237 		break;
238 	case Opt_readdir_max_entries:
239 		fsopt->max_readdir = intval;
240 		break;
241 	case Opt_readdir_max_bytes:
242 		fsopt->max_readdir_bytes = intval;
243 		break;
244 	case Opt_congestion_kb:
245 		fsopt->congestion_kb = intval;
246 		break;
247 	case Opt_dirstat:
248 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
249 		break;
250 	case Opt_nodirstat:
251 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
252 		break;
253 	case Opt_rbytes:
254 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
255 		break;
256 	case Opt_norbytes:
257 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
258 		break;
259 	case Opt_asyncreaddir:
260 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
261 		break;
262 	case Opt_noasyncreaddir:
263 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
264 		break;
265 	case Opt_dcache:
266 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
267 		break;
268 	case Opt_nodcache:
269 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
270 		break;
271 	case Opt_ino32:
272 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
273 		break;
274 	case Opt_noino32:
275 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
276 		break;
277 	case Opt_fscache:
278 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
279 		break;
280 	case Opt_nofscache:
281 		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
282 		break;
283 	case Opt_poolperm:
284 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
285 		printk ("pool perm");
286 		break;
287 	case Opt_nopoolperm:
288 		fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
289 		break;
290 #ifdef CONFIG_CEPH_FS_POSIX_ACL
291 	case Opt_acl:
292 		fsopt->sb_flags |= MS_POSIXACL;
293 		break;
294 #endif
295 	case Opt_noacl:
296 		fsopt->sb_flags &= ~MS_POSIXACL;
297 		break;
298 	default:
299 		BUG_ON(token);
300 	}
301 	return 0;
302 }
303 
304 static void destroy_mount_options(struct ceph_mount_options *args)
305 {
306 	dout("destroy_mount_options %p\n", args);
307 	kfree(args->snapdir_name);
308 	kfree(args->mds_namespace);
309 	kfree(args->server_path);
310 	kfree(args);
311 }
312 
313 static int strcmp_null(const char *s1, const char *s2)
314 {
315 	if (!s1 && !s2)
316 		return 0;
317 	if (s1 && !s2)
318 		return -1;
319 	if (!s1 && s2)
320 		return 1;
321 	return strcmp(s1, s2);
322 }
323 
324 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
325 				 struct ceph_options *new_opt,
326 				 struct ceph_fs_client *fsc)
327 {
328 	struct ceph_mount_options *fsopt1 = new_fsopt;
329 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
330 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
331 	int ret;
332 
333 	ret = memcmp(fsopt1, fsopt2, ofs);
334 	if (ret)
335 		return ret;
336 
337 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
338 	if (ret)
339 		return ret;
340 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
341 	if (ret)
342 		return ret;
343 
344 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
345 	if (ret)
346 		return ret;
347 
348 	return ceph_compare_options(new_opt, fsc->client);
349 }
350 
351 static int parse_mount_options(struct ceph_mount_options **pfsopt,
352 			       struct ceph_options **popt,
353 			       int flags, char *options,
354 			       const char *dev_name)
355 {
356 	struct ceph_mount_options *fsopt;
357 	const char *dev_name_end;
358 	int err;
359 
360 	if (!dev_name || !*dev_name)
361 		return -EINVAL;
362 
363 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
364 	if (!fsopt)
365 		return -ENOMEM;
366 
367 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
368 
369 	fsopt->sb_flags = flags;
370 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
371 
372 	fsopt->rsize = CEPH_RSIZE_DEFAULT;
373 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
374 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
375 	if (!fsopt->snapdir_name) {
376 		err = -ENOMEM;
377 		goto out;
378 	}
379 
380 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
381 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
382 	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
383 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
384 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
385 	fsopt->congestion_kb = default_congestion_kb();
386 
387 	/*
388 	 * Distinguish the server list from the path in "dev_name".
389 	 * Internally we do not include the leading '/' in the path.
390 	 *
391 	 * "dev_name" will look like:
392 	 *     <server_spec>[,<server_spec>...]:[<path>]
393 	 * where
394 	 *     <server_spec> is <ip>[:<port>]
395 	 *     <path> is optional, but if present must begin with '/'
396 	 */
397 	dev_name_end = strchr(dev_name, '/');
398 	if (dev_name_end) {
399 		fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
400 		if (!fsopt->server_path) {
401 			err = -ENOMEM;
402 			goto out;
403 		}
404 	} else {
405 		dev_name_end = dev_name + strlen(dev_name);
406 	}
407 	err = -EINVAL;
408 	dev_name_end--;		/* back up to ':' separator */
409 	if (dev_name_end < dev_name || *dev_name_end != ':') {
410 		pr_err("device name is missing path (no : separator in %s)\n",
411 				dev_name);
412 		goto out;
413 	}
414 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
415 	if (fsopt->server_path)
416 		dout("server path '%s'\n", fsopt->server_path);
417 
418 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
419 				 parse_fsopt_token, (void *)fsopt);
420 	if (IS_ERR(*popt)) {
421 		err = PTR_ERR(*popt);
422 		goto out;
423 	}
424 
425 	/* success */
426 	*pfsopt = fsopt;
427 	return 0;
428 
429 out:
430 	destroy_mount_options(fsopt);
431 	return err;
432 }
433 
434 /**
435  * ceph_show_options - Show mount options in /proc/mounts
436  * @m: seq_file to write to
437  * @root: root of that (sub)tree
438  */
439 static int ceph_show_options(struct seq_file *m, struct dentry *root)
440 {
441 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
442 	struct ceph_mount_options *fsopt = fsc->mount_options;
443 	size_t pos;
444 	int ret;
445 
446 	/* a comma between MNT/MS and client options */
447 	seq_putc(m, ',');
448 	pos = m->count;
449 
450 	ret = ceph_print_client_options(m, fsc->client);
451 	if (ret)
452 		return ret;
453 
454 	/* retract our comma if no client options */
455 	if (m->count == pos)
456 		m->count--;
457 
458 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
459 		seq_puts(m, ",dirstat");
460 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
461 		seq_puts(m, ",rbytes");
462 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
463 		seq_puts(m, ",noasyncreaddir");
464 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
465 		seq_puts(m, ",nodcache");
466 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
467 		seq_puts(m, ",fsc");
468 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
469 		seq_puts(m, ",nopoolperm");
470 
471 #ifdef CONFIG_CEPH_FS_POSIX_ACL
472 	if (fsopt->sb_flags & MS_POSIXACL)
473 		seq_puts(m, ",acl");
474 	else
475 		seq_puts(m, ",noacl");
476 #endif
477 
478 	if (fsopt->mds_namespace)
479 		seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
480 	if (fsopt->wsize)
481 		seq_printf(m, ",wsize=%d", fsopt->wsize);
482 	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
483 		seq_printf(m, ",rsize=%d", fsopt->rsize);
484 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
485 		seq_printf(m, ",rasize=%d", fsopt->rasize);
486 	if (fsopt->congestion_kb != default_congestion_kb())
487 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
488 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
489 		seq_printf(m, ",caps_wanted_delay_min=%d",
490 			 fsopt->caps_wanted_delay_min);
491 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
492 		seq_printf(m, ",caps_wanted_delay_max=%d",
493 			   fsopt->caps_wanted_delay_max);
494 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
495 		seq_printf(m, ",cap_release_safety=%d",
496 			   fsopt->cap_release_safety);
497 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
498 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
499 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
500 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
501 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
502 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
503 
504 	return 0;
505 }
506 
507 /*
508  * handle any mon messages the standard library doesn't understand.
509  * return error if we don't either.
510  */
511 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
512 {
513 	struct ceph_fs_client *fsc = client->private;
514 	int type = le16_to_cpu(msg->hdr.type);
515 
516 	switch (type) {
517 	case CEPH_MSG_MDS_MAP:
518 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
519 		return 0;
520 	case CEPH_MSG_FS_MAP_USER:
521 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
522 		return 0;
523 	default:
524 		return -1;
525 	}
526 }
527 
528 /*
529  * create a new fs client
530  */
531 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
532 					struct ceph_options *opt)
533 {
534 	struct ceph_fs_client *fsc;
535 	const u64 supported_features =
536 		CEPH_FEATURE_FLOCK | CEPH_FEATURE_DIRLAYOUTHASH |
537 		CEPH_FEATURE_MDSENC | CEPH_FEATURE_MDS_INLINE_DATA;
538 	const u64 required_features = 0;
539 	int page_count;
540 	size_t size;
541 	int err = -ENOMEM;
542 
543 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
544 	if (!fsc)
545 		return ERR_PTR(-ENOMEM);
546 
547 	fsc->client = ceph_create_client(opt, fsc, supported_features,
548 					 required_features);
549 	if (IS_ERR(fsc->client)) {
550 		err = PTR_ERR(fsc->client);
551 		goto fail;
552 	}
553 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
554 
555 	if (fsopt->mds_namespace == NULL) {
556 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
557 				   0, true);
558 	} else {
559 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
560 				   0, false);
561 	}
562 
563 	fsc->mount_options = fsopt;
564 
565 	fsc->sb = NULL;
566 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
567 
568 	atomic_long_set(&fsc->writeback_count, 0);
569 
570 	err = bdi_init(&fsc->backing_dev_info);
571 	if (err < 0)
572 		goto fail_client;
573 
574 	err = -ENOMEM;
575 	/*
576 	 * The number of concurrent works can be high but they don't need
577 	 * to be processed in parallel, limit concurrency.
578 	 */
579 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
580 	if (fsc->wb_wq == NULL)
581 		goto fail_bdi;
582 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
583 	if (fsc->pg_inv_wq == NULL)
584 		goto fail_wb_wq;
585 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
586 	if (fsc->trunc_wq == NULL)
587 		goto fail_pg_inv_wq;
588 
589 	/* set up mempools */
590 	err = -ENOMEM;
591 	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
592 	size = sizeof (struct page *) * (page_count ? page_count : 1);
593 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
594 	if (!fsc->wb_pagevec_pool)
595 		goto fail_trunc_wq;
596 
597 	/* setup fscache */
598 	if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
599 	    (ceph_fscache_register_fs(fsc) != 0))
600 		goto fail_fscache;
601 
602 	/* caps */
603 	fsc->min_caps = fsopt->max_readdir;
604 
605 	return fsc;
606 
607 fail_fscache:
608 	ceph_fscache_unregister_fs(fsc);
609 fail_trunc_wq:
610 	destroy_workqueue(fsc->trunc_wq);
611 fail_pg_inv_wq:
612 	destroy_workqueue(fsc->pg_inv_wq);
613 fail_wb_wq:
614 	destroy_workqueue(fsc->wb_wq);
615 fail_bdi:
616 	bdi_destroy(&fsc->backing_dev_info);
617 fail_client:
618 	ceph_destroy_client(fsc->client);
619 fail:
620 	kfree(fsc);
621 	return ERR_PTR(err);
622 }
623 
624 static void destroy_fs_client(struct ceph_fs_client *fsc)
625 {
626 	dout("destroy_fs_client %p\n", fsc);
627 
628 	ceph_fscache_unregister_fs(fsc);
629 
630 	destroy_workqueue(fsc->wb_wq);
631 	destroy_workqueue(fsc->pg_inv_wq);
632 	destroy_workqueue(fsc->trunc_wq);
633 
634 	bdi_destroy(&fsc->backing_dev_info);
635 
636 	mempool_destroy(fsc->wb_pagevec_pool);
637 
638 	destroy_mount_options(fsc->mount_options);
639 
640 	ceph_fs_debugfs_cleanup(fsc);
641 
642 	ceph_destroy_client(fsc->client);
643 
644 	kfree(fsc);
645 	dout("destroy_fs_client %p done\n", fsc);
646 }
647 
648 /*
649  * caches
650  */
651 struct kmem_cache *ceph_inode_cachep;
652 struct kmem_cache *ceph_cap_cachep;
653 struct kmem_cache *ceph_cap_flush_cachep;
654 struct kmem_cache *ceph_dentry_cachep;
655 struct kmem_cache *ceph_file_cachep;
656 
657 static void ceph_inode_init_once(void *foo)
658 {
659 	struct ceph_inode_info *ci = foo;
660 	inode_init_once(&ci->vfs_inode);
661 }
662 
663 static int __init init_caches(void)
664 {
665 	int error = -ENOMEM;
666 
667 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
668 				      sizeof(struct ceph_inode_info),
669 				      __alignof__(struct ceph_inode_info),
670 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
671 				      SLAB_ACCOUNT, ceph_inode_init_once);
672 	if (ceph_inode_cachep == NULL)
673 		return -ENOMEM;
674 
675 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
676 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
677 	if (ceph_cap_cachep == NULL)
678 		goto bad_cap;
679 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
680 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
681 	if (ceph_cap_flush_cachep == NULL)
682 		goto bad_cap_flush;
683 
684 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
685 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
686 	if (ceph_dentry_cachep == NULL)
687 		goto bad_dentry;
688 
689 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
690 
691 	if (ceph_file_cachep == NULL)
692 		goto bad_file;
693 
694 	if ((error = ceph_fscache_register()))
695 		goto bad_file;
696 
697 	return 0;
698 bad_file:
699 	kmem_cache_destroy(ceph_dentry_cachep);
700 bad_dentry:
701 	kmem_cache_destroy(ceph_cap_flush_cachep);
702 bad_cap_flush:
703 	kmem_cache_destroy(ceph_cap_cachep);
704 bad_cap:
705 	kmem_cache_destroy(ceph_inode_cachep);
706 	return error;
707 }
708 
709 static void destroy_caches(void)
710 {
711 	/*
712 	 * Make sure all delayed rcu free inodes are flushed before we
713 	 * destroy cache.
714 	 */
715 	rcu_barrier();
716 
717 	kmem_cache_destroy(ceph_inode_cachep);
718 	kmem_cache_destroy(ceph_cap_cachep);
719 	kmem_cache_destroy(ceph_cap_flush_cachep);
720 	kmem_cache_destroy(ceph_dentry_cachep);
721 	kmem_cache_destroy(ceph_file_cachep);
722 
723 	ceph_fscache_unregister();
724 }
725 
726 
727 /*
728  * ceph_umount_begin - initiate forced umount.  Tear down down the
729  * mount, skipping steps that may hang while waiting for server(s).
730  */
731 static void ceph_umount_begin(struct super_block *sb)
732 {
733 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
734 
735 	dout("ceph_umount_begin - starting forced umount\n");
736 	if (!fsc)
737 		return;
738 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
739 	ceph_mdsc_force_umount(fsc->mdsc);
740 	return;
741 }
742 
743 static const struct super_operations ceph_super_ops = {
744 	.alloc_inode	= ceph_alloc_inode,
745 	.destroy_inode	= ceph_destroy_inode,
746 	.write_inode    = ceph_write_inode,
747 	.drop_inode	= ceph_drop_inode,
748 	.evict_inode	= ceph_evict_inode,
749 	.sync_fs        = ceph_sync_fs,
750 	.put_super	= ceph_put_super,
751 	.show_options   = ceph_show_options,
752 	.statfs		= ceph_statfs,
753 	.umount_begin   = ceph_umount_begin,
754 };
755 
756 /*
757  * Bootstrap mount by opening the root directory.  Note the mount
758  * @started time from caller, and time out if this takes too long.
759  */
760 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
761 				       const char *path,
762 				       unsigned long started)
763 {
764 	struct ceph_mds_client *mdsc = fsc->mdsc;
765 	struct ceph_mds_request *req = NULL;
766 	int err;
767 	struct dentry *root;
768 
769 	/* open dir */
770 	dout("open_root_inode opening '%s'\n", path);
771 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
772 	if (IS_ERR(req))
773 		return ERR_CAST(req);
774 	req->r_path1 = kstrdup(path, GFP_NOFS);
775 	if (!req->r_path1) {
776 		root = ERR_PTR(-ENOMEM);
777 		goto out;
778 	}
779 
780 	req->r_ino1.ino = CEPH_INO_ROOT;
781 	req->r_ino1.snap = CEPH_NOSNAP;
782 	req->r_started = started;
783 	req->r_timeout = fsc->client->options->mount_timeout;
784 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
785 	req->r_num_caps = 2;
786 	err = ceph_mdsc_do_request(mdsc, NULL, req);
787 	if (err == 0) {
788 		struct inode *inode = req->r_target_inode;
789 		req->r_target_inode = NULL;
790 		dout("open_root_inode success\n");
791 		if (ceph_ino(inode) == CEPH_INO_ROOT &&
792 		    fsc->sb->s_root == NULL) {
793 			root = d_make_root(inode);
794 			if (!root) {
795 				root = ERR_PTR(-ENOMEM);
796 				goto out;
797 			}
798 		} else {
799 			root = d_obtain_root(inode);
800 		}
801 		ceph_init_dentry(root);
802 		dout("open_root_inode success, root dentry is %p\n", root);
803 	} else {
804 		root = ERR_PTR(err);
805 	}
806 out:
807 	ceph_mdsc_put_request(req);
808 	return root;
809 }
810 
811 
812 
813 
814 /*
815  * mount: join the ceph cluster, and open root directory.
816  */
817 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
818 {
819 	int err;
820 	unsigned long started = jiffies;  /* note the start time */
821 	struct dentry *root;
822 	int first = 0;   /* first vfsmount for this super_block */
823 
824 	dout("mount start %p\n", fsc);
825 	mutex_lock(&fsc->client->mount_mutex);
826 
827 	if (!fsc->sb->s_root) {
828 		err = __ceph_open_session(fsc->client, started);
829 		if (err < 0)
830 			goto out;
831 
832 		dout("mount opening root\n");
833 		root = open_root_dentry(fsc, "", started);
834 		if (IS_ERR(root)) {
835 			err = PTR_ERR(root);
836 			goto out;
837 		}
838 		fsc->sb->s_root = root;
839 		first = 1;
840 
841 		err = ceph_fs_debugfs_init(fsc);
842 		if (err < 0)
843 			goto fail;
844 	}
845 
846 	if (!fsc->mount_options->server_path) {
847 		root = fsc->sb->s_root;
848 		dget(root);
849 	} else {
850 		const char *path = fsc->mount_options->server_path + 1;
851 		dout("mount opening path %s\n", path);
852 		root = open_root_dentry(fsc, path, started);
853 		if (IS_ERR(root)) {
854 			err = PTR_ERR(root);
855 			goto fail;
856 		}
857 	}
858 
859 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
860 	dout("mount success\n");
861 	mutex_unlock(&fsc->client->mount_mutex);
862 	return root;
863 
864 fail:
865 	if (first) {
866 		dput(fsc->sb->s_root);
867 		fsc->sb->s_root = NULL;
868 	}
869 out:
870 	mutex_unlock(&fsc->client->mount_mutex);
871 	return ERR_PTR(err);
872 }
873 
874 static int ceph_set_super(struct super_block *s, void *data)
875 {
876 	struct ceph_fs_client *fsc = data;
877 	int ret;
878 
879 	dout("set_super %p data %p\n", s, data);
880 
881 	s->s_flags = fsc->mount_options->sb_flags;
882 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
883 
884 	s->s_xattr = ceph_xattr_handlers;
885 	s->s_fs_info = fsc;
886 	fsc->sb = s;
887 
888 	s->s_op = &ceph_super_ops;
889 	s->s_export_op = &ceph_export_ops;
890 
891 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
892 
893 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
894 	if (ret != 0)
895 		goto fail;
896 
897 	return ret;
898 
899 fail:
900 	s->s_fs_info = NULL;
901 	fsc->sb = NULL;
902 	return ret;
903 }
904 
905 /*
906  * share superblock if same fs AND options
907  */
908 static int ceph_compare_super(struct super_block *sb, void *data)
909 {
910 	struct ceph_fs_client *new = data;
911 	struct ceph_mount_options *fsopt = new->mount_options;
912 	struct ceph_options *opt = new->client->options;
913 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
914 
915 	dout("ceph_compare_super %p\n", sb);
916 
917 	if (compare_mount_options(fsopt, opt, other)) {
918 		dout("monitor(s)/mount options don't match\n");
919 		return 0;
920 	}
921 	if ((opt->flags & CEPH_OPT_FSID) &&
922 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
923 		dout("fsid doesn't match\n");
924 		return 0;
925 	}
926 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
927 		dout("flags differ\n");
928 		return 0;
929 	}
930 	return 1;
931 }
932 
933 /*
934  * construct our own bdi so we can control readahead, etc.
935  */
936 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
937 
938 static int ceph_register_bdi(struct super_block *sb,
939 			     struct ceph_fs_client *fsc)
940 {
941 	int err;
942 
943 	/* set ra_pages based on rasize mount option? */
944 	if (fsc->mount_options->rasize >= PAGE_SIZE)
945 		fsc->backing_dev_info.ra_pages =
946 			(fsc->mount_options->rasize + PAGE_SIZE - 1)
947 			>> PAGE_SHIFT;
948 	else
949 		fsc->backing_dev_info.ra_pages =
950 			VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
951 
952 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
953 			   atomic_long_inc_return(&bdi_seq));
954 	if (!err)
955 		sb->s_bdi = &fsc->backing_dev_info;
956 	return err;
957 }
958 
959 static struct dentry *ceph_mount(struct file_system_type *fs_type,
960 		       int flags, const char *dev_name, void *data)
961 {
962 	struct super_block *sb;
963 	struct ceph_fs_client *fsc;
964 	struct dentry *res;
965 	int err;
966 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
967 	struct ceph_mount_options *fsopt = NULL;
968 	struct ceph_options *opt = NULL;
969 
970 	dout("ceph_mount\n");
971 
972 #ifdef CONFIG_CEPH_FS_POSIX_ACL
973 	flags |= MS_POSIXACL;
974 #endif
975 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
976 	if (err < 0) {
977 		res = ERR_PTR(err);
978 		goto out_final;
979 	}
980 
981 	/* create client (which we may/may not use) */
982 	fsc = create_fs_client(fsopt, opt);
983 	if (IS_ERR(fsc)) {
984 		res = ERR_CAST(fsc);
985 		destroy_mount_options(fsopt);
986 		ceph_destroy_options(opt);
987 		goto out_final;
988 	}
989 
990 	err = ceph_mdsc_init(fsc);
991 	if (err < 0) {
992 		res = ERR_PTR(err);
993 		goto out;
994 	}
995 
996 	if (ceph_test_opt(fsc->client, NOSHARE))
997 		compare_super = NULL;
998 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
999 	if (IS_ERR(sb)) {
1000 		res = ERR_CAST(sb);
1001 		goto out;
1002 	}
1003 
1004 	if (ceph_sb_to_client(sb) != fsc) {
1005 		ceph_mdsc_destroy(fsc);
1006 		destroy_fs_client(fsc);
1007 		fsc = ceph_sb_to_client(sb);
1008 		dout("get_sb got existing client %p\n", fsc);
1009 	} else {
1010 		dout("get_sb using new client %p\n", fsc);
1011 		err = ceph_register_bdi(sb, fsc);
1012 		if (err < 0) {
1013 			res = ERR_PTR(err);
1014 			goto out_splat;
1015 		}
1016 	}
1017 
1018 	res = ceph_real_mount(fsc);
1019 	if (IS_ERR(res))
1020 		goto out_splat;
1021 	dout("root %p inode %p ino %llx.%llx\n", res,
1022 	     d_inode(res), ceph_vinop(d_inode(res)));
1023 	return res;
1024 
1025 out_splat:
1026 	ceph_mdsc_close_sessions(fsc->mdsc);
1027 	deactivate_locked_super(sb);
1028 	goto out_final;
1029 
1030 out:
1031 	ceph_mdsc_destroy(fsc);
1032 	destroy_fs_client(fsc);
1033 out_final:
1034 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1035 	return res;
1036 }
1037 
1038 static void ceph_kill_sb(struct super_block *s)
1039 {
1040 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1041 	dev_t dev = s->s_dev;
1042 
1043 	dout("kill_sb %p\n", s);
1044 
1045 	ceph_mdsc_pre_umount(fsc->mdsc);
1046 	generic_shutdown_super(s);
1047 	ceph_mdsc_destroy(fsc);
1048 
1049 	destroy_fs_client(fsc);
1050 	free_anon_bdev(dev);
1051 }
1052 
1053 static struct file_system_type ceph_fs_type = {
1054 	.owner		= THIS_MODULE,
1055 	.name		= "ceph",
1056 	.mount		= ceph_mount,
1057 	.kill_sb	= ceph_kill_sb,
1058 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1059 };
1060 MODULE_ALIAS_FS("ceph");
1061 
1062 static int __init init_ceph(void)
1063 {
1064 	int ret = init_caches();
1065 	if (ret)
1066 		goto out;
1067 
1068 	ceph_flock_init();
1069 	ceph_xattr_init();
1070 	ret = register_filesystem(&ceph_fs_type);
1071 	if (ret)
1072 		goto out_xattr;
1073 
1074 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1075 
1076 	return 0;
1077 
1078 out_xattr:
1079 	ceph_xattr_exit();
1080 	destroy_caches();
1081 out:
1082 	return ret;
1083 }
1084 
1085 static void __exit exit_ceph(void)
1086 {
1087 	dout("exit_ceph\n");
1088 	unregister_filesystem(&ceph_fs_type);
1089 	ceph_xattr_exit();
1090 	destroy_caches();
1091 }
1092 
1093 module_init(init_ceph);
1094 module_exit(exit_ceph);
1095 
1096 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1097 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1098 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1099 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1100 MODULE_LICENSE("GPL");
1101