xref: /openbmc/linux/fs/ceph/super.c (revision aac5987a)
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 	Opt_require_active_mds,
141 	Opt_norequire_active_mds,
142 #ifdef CONFIG_CEPH_FS_POSIX_ACL
143 	Opt_acl,
144 #endif
145 	Opt_noacl,
146 };
147 
148 static match_table_t fsopt_tokens = {
149 	{Opt_wsize, "wsize=%d"},
150 	{Opt_rsize, "rsize=%d"},
151 	{Opt_rasize, "rasize=%d"},
152 	{Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
153 	{Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
154 	{Opt_cap_release_safety, "cap_release_safety=%d"},
155 	{Opt_readdir_max_entries, "readdir_max_entries=%d"},
156 	{Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
157 	{Opt_congestion_kb, "write_congestion_kb=%d"},
158 	/* int args above */
159 	{Opt_snapdirname, "snapdirname=%s"},
160 	{Opt_mds_namespace, "mds_namespace=%s"},
161 	/* string args above */
162 	{Opt_dirstat, "dirstat"},
163 	{Opt_nodirstat, "nodirstat"},
164 	{Opt_rbytes, "rbytes"},
165 	{Opt_norbytes, "norbytes"},
166 	{Opt_asyncreaddir, "asyncreaddir"},
167 	{Opt_noasyncreaddir, "noasyncreaddir"},
168 	{Opt_dcache, "dcache"},
169 	{Opt_nodcache, "nodcache"},
170 	{Opt_ino32, "ino32"},
171 	{Opt_noino32, "noino32"},
172 	{Opt_fscache, "fsc"},
173 	{Opt_nofscache, "nofsc"},
174 	{Opt_poolperm, "poolperm"},
175 	{Opt_nopoolperm, "nopoolperm"},
176 	{Opt_require_active_mds, "require_active_mds"},
177 	{Opt_norequire_active_mds, "norequire_active_mds"},
178 #ifdef CONFIG_CEPH_FS_POSIX_ACL
179 	{Opt_acl, "acl"},
180 #endif
181 	{Opt_noacl, "noacl"},
182 	{-1, NULL}
183 };
184 
185 static int parse_fsopt_token(char *c, void *private)
186 {
187 	struct ceph_mount_options *fsopt = private;
188 	substring_t argstr[MAX_OPT_ARGS];
189 	int token, intval, ret;
190 
191 	token = match_token((char *)c, fsopt_tokens, argstr);
192 	if (token < 0)
193 		return -EINVAL;
194 
195 	if (token < Opt_last_int) {
196 		ret = match_int(&argstr[0], &intval);
197 		if (ret < 0) {
198 			pr_err("bad mount option arg (not int) "
199 			       "at '%s'\n", c);
200 			return ret;
201 		}
202 		dout("got int token %d val %d\n", token, intval);
203 	} else if (token > Opt_last_int && token < Opt_last_string) {
204 		dout("got string token %d val %s\n", token,
205 		     argstr[0].from);
206 	} else {
207 		dout("got token %d\n", token);
208 	}
209 
210 	switch (token) {
211 	case Opt_snapdirname:
212 		kfree(fsopt->snapdir_name);
213 		fsopt->snapdir_name = kstrndup(argstr[0].from,
214 					       argstr[0].to-argstr[0].from,
215 					       GFP_KERNEL);
216 		if (!fsopt->snapdir_name)
217 			return -ENOMEM;
218 		break;
219 	case Opt_mds_namespace:
220 		fsopt->mds_namespace = kstrndup(argstr[0].from,
221 						argstr[0].to-argstr[0].from,
222 						GFP_KERNEL);
223 		if (!fsopt->mds_namespace)
224 			return -ENOMEM;
225 		break;
226 		/* misc */
227 	case Opt_wsize:
228 		fsopt->wsize = intval;
229 		break;
230 	case Opt_rsize:
231 		fsopt->rsize = intval;
232 		break;
233 	case Opt_rasize:
234 		fsopt->rasize = intval;
235 		break;
236 	case Opt_caps_wanted_delay_min:
237 		fsopt->caps_wanted_delay_min = intval;
238 		break;
239 	case Opt_caps_wanted_delay_max:
240 		fsopt->caps_wanted_delay_max = intval;
241 		break;
242 	case Opt_readdir_max_entries:
243 		fsopt->max_readdir = intval;
244 		break;
245 	case Opt_readdir_max_bytes:
246 		fsopt->max_readdir_bytes = intval;
247 		break;
248 	case Opt_congestion_kb:
249 		fsopt->congestion_kb = intval;
250 		break;
251 	case Opt_dirstat:
252 		fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
253 		break;
254 	case Opt_nodirstat:
255 		fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
256 		break;
257 	case Opt_rbytes:
258 		fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
259 		break;
260 	case Opt_norbytes:
261 		fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
262 		break;
263 	case Opt_asyncreaddir:
264 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
265 		break;
266 	case Opt_noasyncreaddir:
267 		fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
268 		break;
269 	case Opt_dcache:
270 		fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
271 		break;
272 	case Opt_nodcache:
273 		fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
274 		break;
275 	case Opt_ino32:
276 		fsopt->flags |= CEPH_MOUNT_OPT_INO32;
277 		break;
278 	case Opt_noino32:
279 		fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
280 		break;
281 	case Opt_fscache:
282 		fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
283 		break;
284 	case Opt_nofscache:
285 		fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
286 		break;
287 	case Opt_poolperm:
288 		fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
289 		printk ("pool perm");
290 		break;
291 	case Opt_nopoolperm:
292 		fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
293 		break;
294 	case Opt_require_active_mds:
295 		fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
296 		break;
297 	case Opt_norequire_active_mds:
298 		fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
299 		break;
300 #ifdef CONFIG_CEPH_FS_POSIX_ACL
301 	case Opt_acl:
302 		fsopt->sb_flags |= MS_POSIXACL;
303 		break;
304 #endif
305 	case Opt_noacl:
306 		fsopt->sb_flags &= ~MS_POSIXACL;
307 		break;
308 	default:
309 		BUG_ON(token);
310 	}
311 	return 0;
312 }
313 
314 static void destroy_mount_options(struct ceph_mount_options *args)
315 {
316 	dout("destroy_mount_options %p\n", args);
317 	kfree(args->snapdir_name);
318 	kfree(args->mds_namespace);
319 	kfree(args->server_path);
320 	kfree(args);
321 }
322 
323 static int strcmp_null(const char *s1, const char *s2)
324 {
325 	if (!s1 && !s2)
326 		return 0;
327 	if (s1 && !s2)
328 		return -1;
329 	if (!s1 && s2)
330 		return 1;
331 	return strcmp(s1, s2);
332 }
333 
334 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
335 				 struct ceph_options *new_opt,
336 				 struct ceph_fs_client *fsc)
337 {
338 	struct ceph_mount_options *fsopt1 = new_fsopt;
339 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
340 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
341 	int ret;
342 
343 	ret = memcmp(fsopt1, fsopt2, ofs);
344 	if (ret)
345 		return ret;
346 
347 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
348 	if (ret)
349 		return ret;
350 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
351 	if (ret)
352 		return ret;
353 
354 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
355 	if (ret)
356 		return ret;
357 
358 	return ceph_compare_options(new_opt, fsc->client);
359 }
360 
361 static int parse_mount_options(struct ceph_mount_options **pfsopt,
362 			       struct ceph_options **popt,
363 			       int flags, char *options,
364 			       const char *dev_name)
365 {
366 	struct ceph_mount_options *fsopt;
367 	const char *dev_name_end;
368 	int err;
369 
370 	if (!dev_name || !*dev_name)
371 		return -EINVAL;
372 
373 	fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
374 	if (!fsopt)
375 		return -ENOMEM;
376 
377 	dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
378 
379 	fsopt->sb_flags = flags;
380 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
381 
382 	fsopt->rsize = CEPH_RSIZE_DEFAULT;
383 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
384 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
385 	if (!fsopt->snapdir_name) {
386 		err = -ENOMEM;
387 		goto out;
388 	}
389 
390 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
391 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
392 	fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
393 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
394 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
395 	fsopt->congestion_kb = default_congestion_kb();
396 
397 	/*
398 	 * Distinguish the server list from the path in "dev_name".
399 	 * Internally we do not include the leading '/' in the path.
400 	 *
401 	 * "dev_name" will look like:
402 	 *     <server_spec>[,<server_spec>...]:[<path>]
403 	 * where
404 	 *     <server_spec> is <ip>[:<port>]
405 	 *     <path> is optional, but if present must begin with '/'
406 	 */
407 	dev_name_end = strchr(dev_name, '/');
408 	if (dev_name_end) {
409 		if (strlen(dev_name_end) > 1) {
410 			fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
411 			if (!fsopt->server_path) {
412 				err = -ENOMEM;
413 				goto out;
414 			}
415 		}
416 	} else {
417 		dev_name_end = dev_name + strlen(dev_name);
418 	}
419 	err = -EINVAL;
420 	dev_name_end--;		/* back up to ':' separator */
421 	if (dev_name_end < dev_name || *dev_name_end != ':') {
422 		pr_err("device name is missing path (no : separator in %s)\n",
423 				dev_name);
424 		goto out;
425 	}
426 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
427 	if (fsopt->server_path)
428 		dout("server path '%s'\n", fsopt->server_path);
429 
430 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
431 				 parse_fsopt_token, (void *)fsopt);
432 	if (IS_ERR(*popt)) {
433 		err = PTR_ERR(*popt);
434 		goto out;
435 	}
436 
437 	/* success */
438 	*pfsopt = fsopt;
439 	return 0;
440 
441 out:
442 	destroy_mount_options(fsopt);
443 	return err;
444 }
445 
446 /**
447  * ceph_show_options - Show mount options in /proc/mounts
448  * @m: seq_file to write to
449  * @root: root of that (sub)tree
450  */
451 static int ceph_show_options(struct seq_file *m, struct dentry *root)
452 {
453 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
454 	struct ceph_mount_options *fsopt = fsc->mount_options;
455 	size_t pos;
456 	int ret;
457 
458 	/* a comma between MNT/MS and client options */
459 	seq_putc(m, ',');
460 	pos = m->count;
461 
462 	ret = ceph_print_client_options(m, fsc->client);
463 	if (ret)
464 		return ret;
465 
466 	/* retract our comma if no client options */
467 	if (m->count == pos)
468 		m->count--;
469 
470 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
471 		seq_puts(m, ",dirstat");
472 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
473 		seq_puts(m, ",rbytes");
474 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
475 		seq_puts(m, ",noasyncreaddir");
476 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
477 		seq_puts(m, ",nodcache");
478 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
479 		seq_puts(m, ",fsc");
480 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
481 		seq_puts(m, ",nopoolperm");
482 
483 #ifdef CONFIG_CEPH_FS_POSIX_ACL
484 	if (fsopt->sb_flags & MS_POSIXACL)
485 		seq_puts(m, ",acl");
486 	else
487 		seq_puts(m, ",noacl");
488 #endif
489 
490 	if (fsopt->mds_namespace)
491 		seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
492 	if (fsopt->wsize)
493 		seq_printf(m, ",wsize=%d", fsopt->wsize);
494 	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
495 		seq_printf(m, ",rsize=%d", fsopt->rsize);
496 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
497 		seq_printf(m, ",rasize=%d", fsopt->rasize);
498 	if (fsopt->congestion_kb != default_congestion_kb())
499 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
500 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
501 		seq_printf(m, ",caps_wanted_delay_min=%d",
502 			 fsopt->caps_wanted_delay_min);
503 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
504 		seq_printf(m, ",caps_wanted_delay_max=%d",
505 			   fsopt->caps_wanted_delay_max);
506 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
507 		seq_printf(m, ",cap_release_safety=%d",
508 			   fsopt->cap_release_safety);
509 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
510 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
511 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
512 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
513 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
514 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
515 
516 	return 0;
517 }
518 
519 /*
520  * handle any mon messages the standard library doesn't understand.
521  * return error if we don't either.
522  */
523 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
524 {
525 	struct ceph_fs_client *fsc = client->private;
526 	int type = le16_to_cpu(msg->hdr.type);
527 
528 	switch (type) {
529 	case CEPH_MSG_MDS_MAP:
530 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
531 		return 0;
532 	case CEPH_MSG_FS_MAP_USER:
533 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
534 		return 0;
535 	default:
536 		return -1;
537 	}
538 }
539 
540 /*
541  * create a new fs client
542  */
543 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
544 					struct ceph_options *opt)
545 {
546 	struct ceph_fs_client *fsc;
547 	const u64 supported_features =
548 		CEPH_FEATURE_FLOCK | CEPH_FEATURE_DIRLAYOUTHASH |
549 		CEPH_FEATURE_MDSENC | CEPH_FEATURE_MDS_INLINE_DATA;
550 	const u64 required_features = 0;
551 	int page_count;
552 	size_t size;
553 	int err = -ENOMEM;
554 
555 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
556 	if (!fsc)
557 		return ERR_PTR(-ENOMEM);
558 
559 	fsc->client = ceph_create_client(opt, fsc, supported_features,
560 					 required_features);
561 	if (IS_ERR(fsc->client)) {
562 		err = PTR_ERR(fsc->client);
563 		goto fail;
564 	}
565 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
566 
567 	if (fsopt->mds_namespace == NULL) {
568 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
569 				   0, true);
570 	} else {
571 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
572 				   0, false);
573 	}
574 
575 	fsc->mount_options = fsopt;
576 
577 	fsc->sb = NULL;
578 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
579 
580 	atomic_long_set(&fsc->writeback_count, 0);
581 
582 	err = bdi_init(&fsc->backing_dev_info);
583 	if (err < 0)
584 		goto fail_client;
585 
586 	err = -ENOMEM;
587 	/*
588 	 * The number of concurrent works can be high but they don't need
589 	 * to be processed in parallel, limit concurrency.
590 	 */
591 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
592 	if (fsc->wb_wq == NULL)
593 		goto fail_bdi;
594 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
595 	if (fsc->pg_inv_wq == NULL)
596 		goto fail_wb_wq;
597 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
598 	if (fsc->trunc_wq == NULL)
599 		goto fail_pg_inv_wq;
600 
601 	/* set up mempools */
602 	err = -ENOMEM;
603 	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
604 	size = sizeof (struct page *) * (page_count ? page_count : 1);
605 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
606 	if (!fsc->wb_pagevec_pool)
607 		goto fail_trunc_wq;
608 
609 	/* setup fscache */
610 	if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
611 	    (ceph_fscache_register_fs(fsc) != 0))
612 		goto fail_fscache;
613 
614 	/* caps */
615 	fsc->min_caps = fsopt->max_readdir;
616 
617 	return fsc;
618 
619 fail_fscache:
620 	ceph_fscache_unregister_fs(fsc);
621 fail_trunc_wq:
622 	destroy_workqueue(fsc->trunc_wq);
623 fail_pg_inv_wq:
624 	destroy_workqueue(fsc->pg_inv_wq);
625 fail_wb_wq:
626 	destroy_workqueue(fsc->wb_wq);
627 fail_bdi:
628 	bdi_destroy(&fsc->backing_dev_info);
629 fail_client:
630 	ceph_destroy_client(fsc->client);
631 fail:
632 	kfree(fsc);
633 	return ERR_PTR(err);
634 }
635 
636 static void destroy_fs_client(struct ceph_fs_client *fsc)
637 {
638 	dout("destroy_fs_client %p\n", fsc);
639 
640 	ceph_fscache_unregister_fs(fsc);
641 
642 	destroy_workqueue(fsc->wb_wq);
643 	destroy_workqueue(fsc->pg_inv_wq);
644 	destroy_workqueue(fsc->trunc_wq);
645 
646 	bdi_destroy(&fsc->backing_dev_info);
647 
648 	mempool_destroy(fsc->wb_pagevec_pool);
649 
650 	destroy_mount_options(fsc->mount_options);
651 
652 	ceph_fs_debugfs_cleanup(fsc);
653 
654 	ceph_destroy_client(fsc->client);
655 
656 	kfree(fsc);
657 	dout("destroy_fs_client %p done\n", fsc);
658 }
659 
660 /*
661  * caches
662  */
663 struct kmem_cache *ceph_inode_cachep;
664 struct kmem_cache *ceph_cap_cachep;
665 struct kmem_cache *ceph_cap_flush_cachep;
666 struct kmem_cache *ceph_dentry_cachep;
667 struct kmem_cache *ceph_file_cachep;
668 
669 static void ceph_inode_init_once(void *foo)
670 {
671 	struct ceph_inode_info *ci = foo;
672 	inode_init_once(&ci->vfs_inode);
673 }
674 
675 static int __init init_caches(void)
676 {
677 	int error = -ENOMEM;
678 
679 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
680 				      sizeof(struct ceph_inode_info),
681 				      __alignof__(struct ceph_inode_info),
682 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
683 				      SLAB_ACCOUNT, ceph_inode_init_once);
684 	if (ceph_inode_cachep == NULL)
685 		return -ENOMEM;
686 
687 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
688 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
689 	if (ceph_cap_cachep == NULL)
690 		goto bad_cap;
691 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
692 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
693 	if (ceph_cap_flush_cachep == NULL)
694 		goto bad_cap_flush;
695 
696 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
697 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
698 	if (ceph_dentry_cachep == NULL)
699 		goto bad_dentry;
700 
701 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
702 
703 	if (ceph_file_cachep == NULL)
704 		goto bad_file;
705 
706 	if ((error = ceph_fscache_register()))
707 		goto bad_file;
708 
709 	return 0;
710 bad_file:
711 	kmem_cache_destroy(ceph_dentry_cachep);
712 bad_dentry:
713 	kmem_cache_destroy(ceph_cap_flush_cachep);
714 bad_cap_flush:
715 	kmem_cache_destroy(ceph_cap_cachep);
716 bad_cap:
717 	kmem_cache_destroy(ceph_inode_cachep);
718 	return error;
719 }
720 
721 static void destroy_caches(void)
722 {
723 	/*
724 	 * Make sure all delayed rcu free inodes are flushed before we
725 	 * destroy cache.
726 	 */
727 	rcu_barrier();
728 
729 	kmem_cache_destroy(ceph_inode_cachep);
730 	kmem_cache_destroy(ceph_cap_cachep);
731 	kmem_cache_destroy(ceph_cap_flush_cachep);
732 	kmem_cache_destroy(ceph_dentry_cachep);
733 	kmem_cache_destroy(ceph_file_cachep);
734 
735 	ceph_fscache_unregister();
736 }
737 
738 
739 /*
740  * ceph_umount_begin - initiate forced umount.  Tear down down the
741  * mount, skipping steps that may hang while waiting for server(s).
742  */
743 static void ceph_umount_begin(struct super_block *sb)
744 {
745 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
746 
747 	dout("ceph_umount_begin - starting forced umount\n");
748 	if (!fsc)
749 		return;
750 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
751 	ceph_mdsc_force_umount(fsc->mdsc);
752 	return;
753 }
754 
755 static const struct super_operations ceph_super_ops = {
756 	.alloc_inode	= ceph_alloc_inode,
757 	.destroy_inode	= ceph_destroy_inode,
758 	.write_inode    = ceph_write_inode,
759 	.drop_inode	= ceph_drop_inode,
760 	.sync_fs        = ceph_sync_fs,
761 	.put_super	= ceph_put_super,
762 	.show_options   = ceph_show_options,
763 	.statfs		= ceph_statfs,
764 	.umount_begin   = ceph_umount_begin,
765 };
766 
767 /*
768  * Bootstrap mount by opening the root directory.  Note the mount
769  * @started time from caller, and time out if this takes too long.
770  */
771 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
772 				       const char *path,
773 				       unsigned long started)
774 {
775 	struct ceph_mds_client *mdsc = fsc->mdsc;
776 	struct ceph_mds_request *req = NULL;
777 	int err;
778 	struct dentry *root;
779 
780 	/* open dir */
781 	dout("open_root_inode opening '%s'\n", path);
782 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
783 	if (IS_ERR(req))
784 		return ERR_CAST(req);
785 	req->r_path1 = kstrdup(path, GFP_NOFS);
786 	if (!req->r_path1) {
787 		root = ERR_PTR(-ENOMEM);
788 		goto out;
789 	}
790 
791 	req->r_ino1.ino = CEPH_INO_ROOT;
792 	req->r_ino1.snap = CEPH_NOSNAP;
793 	req->r_started = started;
794 	req->r_timeout = fsc->client->options->mount_timeout;
795 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
796 	req->r_num_caps = 2;
797 	err = ceph_mdsc_do_request(mdsc, NULL, req);
798 	if (err == 0) {
799 		struct inode *inode = req->r_target_inode;
800 		req->r_target_inode = NULL;
801 		dout("open_root_inode success\n");
802 		root = d_make_root(inode);
803 		if (!root) {
804 			root = ERR_PTR(-ENOMEM);
805 			goto out;
806 		}
807 		dout("open_root_inode success, root dentry is %p\n", root);
808 	} else {
809 		root = ERR_PTR(err);
810 	}
811 out:
812 	ceph_mdsc_put_request(req);
813 	return root;
814 }
815 
816 
817 
818 
819 /*
820  * mount: join the ceph cluster, and open root directory.
821  */
822 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
823 {
824 	int err;
825 	unsigned long started = jiffies;  /* note the start time */
826 	struct dentry *root;
827 	int first = 0;   /* first vfsmount for this super_block */
828 
829 	dout("mount start %p\n", fsc);
830 	mutex_lock(&fsc->client->mount_mutex);
831 
832 	if (!fsc->sb->s_root) {
833 		const char *path;
834 		err = __ceph_open_session(fsc->client, started);
835 		if (err < 0)
836 			goto out;
837 
838 		if (!fsc->mount_options->server_path) {
839 			path = "";
840 			dout("mount opening path \\t\n");
841 		} else {
842 			path = fsc->mount_options->server_path + 1;
843 			dout("mount opening path %s\n", path);
844 		}
845 		root = open_root_dentry(fsc, path, started);
846 		if (IS_ERR(root)) {
847 			err = PTR_ERR(root);
848 			goto out;
849 		}
850 		fsc->sb->s_root = dget(root);
851 		first = 1;
852 
853 		err = ceph_fs_debugfs_init(fsc);
854 		if (err < 0)
855 			goto fail;
856 	} else {
857 		root = dget(fsc->sb->s_root);
858 	}
859 
860 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
861 	dout("mount success\n");
862 	mutex_unlock(&fsc->client->mount_mutex);
863 	return root;
864 
865 fail:
866 	if (first) {
867 		dput(fsc->sb->s_root);
868 		fsc->sb->s_root = NULL;
869 	}
870 out:
871 	mutex_unlock(&fsc->client->mount_mutex);
872 	return ERR_PTR(err);
873 }
874 
875 static int ceph_set_super(struct super_block *s, void *data)
876 {
877 	struct ceph_fs_client *fsc = data;
878 	int ret;
879 
880 	dout("set_super %p data %p\n", s, data);
881 
882 	s->s_flags = fsc->mount_options->sb_flags;
883 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
884 
885 	s->s_xattr = ceph_xattr_handlers;
886 	s->s_fs_info = fsc;
887 	fsc->sb = s;
888 
889 	s->s_op = &ceph_super_ops;
890 	s->s_d_op = &ceph_dentry_ops;
891 	s->s_export_op = &ceph_export_ops;
892 
893 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
894 
895 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
896 	if (ret != 0)
897 		goto fail;
898 
899 	return ret;
900 
901 fail:
902 	s->s_fs_info = NULL;
903 	fsc->sb = NULL;
904 	return ret;
905 }
906 
907 /*
908  * share superblock if same fs AND options
909  */
910 static int ceph_compare_super(struct super_block *sb, void *data)
911 {
912 	struct ceph_fs_client *new = data;
913 	struct ceph_mount_options *fsopt = new->mount_options;
914 	struct ceph_options *opt = new->client->options;
915 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
916 
917 	dout("ceph_compare_super %p\n", sb);
918 
919 	if (compare_mount_options(fsopt, opt, other)) {
920 		dout("monitor(s)/mount options don't match\n");
921 		return 0;
922 	}
923 	if ((opt->flags & CEPH_OPT_FSID) &&
924 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
925 		dout("fsid doesn't match\n");
926 		return 0;
927 	}
928 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
929 		dout("flags differ\n");
930 		return 0;
931 	}
932 	return 1;
933 }
934 
935 /*
936  * construct our own bdi so we can control readahead, etc.
937  */
938 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
939 
940 static int ceph_register_bdi(struct super_block *sb,
941 			     struct ceph_fs_client *fsc)
942 {
943 	int err;
944 
945 	/* set ra_pages based on rasize mount option? */
946 	if (fsc->mount_options->rasize >= PAGE_SIZE)
947 		fsc->backing_dev_info.ra_pages =
948 			(fsc->mount_options->rasize + PAGE_SIZE - 1)
949 			>> PAGE_SHIFT;
950 	else
951 		fsc->backing_dev_info.ra_pages =
952 			VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
953 
954 	if (fsc->mount_options->rsize > fsc->mount_options->rasize &&
955 	    fsc->mount_options->rsize >= PAGE_SIZE)
956 		fsc->backing_dev_info.io_pages =
957 			(fsc->mount_options->rsize + PAGE_SIZE - 1)
958 			>> PAGE_SHIFT;
959 	else if (fsc->mount_options->rsize == 0)
960 		fsc->backing_dev_info.io_pages = ULONG_MAX;
961 
962 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
963 			   atomic_long_inc_return(&bdi_seq));
964 	if (!err)
965 		sb->s_bdi = &fsc->backing_dev_info;
966 	return err;
967 }
968 
969 static struct dentry *ceph_mount(struct file_system_type *fs_type,
970 		       int flags, const char *dev_name, void *data)
971 {
972 	struct super_block *sb;
973 	struct ceph_fs_client *fsc;
974 	struct dentry *res;
975 	int err;
976 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
977 	struct ceph_mount_options *fsopt = NULL;
978 	struct ceph_options *opt = NULL;
979 
980 	dout("ceph_mount\n");
981 
982 #ifdef CONFIG_CEPH_FS_POSIX_ACL
983 	flags |= MS_POSIXACL;
984 #endif
985 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
986 	if (err < 0) {
987 		res = ERR_PTR(err);
988 		goto out_final;
989 	}
990 
991 	/* create client (which we may/may not use) */
992 	fsc = create_fs_client(fsopt, opt);
993 	if (IS_ERR(fsc)) {
994 		res = ERR_CAST(fsc);
995 		destroy_mount_options(fsopt);
996 		ceph_destroy_options(opt);
997 		goto out_final;
998 	}
999 
1000 	err = ceph_mdsc_init(fsc);
1001 	if (err < 0) {
1002 		res = ERR_PTR(err);
1003 		goto out;
1004 	}
1005 
1006 	if (ceph_test_opt(fsc->client, NOSHARE))
1007 		compare_super = NULL;
1008 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1009 	if (IS_ERR(sb)) {
1010 		res = ERR_CAST(sb);
1011 		goto out;
1012 	}
1013 
1014 	if (ceph_sb_to_client(sb) != fsc) {
1015 		ceph_mdsc_destroy(fsc);
1016 		destroy_fs_client(fsc);
1017 		fsc = ceph_sb_to_client(sb);
1018 		dout("get_sb got existing client %p\n", fsc);
1019 	} else {
1020 		dout("get_sb using new client %p\n", fsc);
1021 		err = ceph_register_bdi(sb, fsc);
1022 		if (err < 0) {
1023 			res = ERR_PTR(err);
1024 			goto out_splat;
1025 		}
1026 	}
1027 
1028 	res = ceph_real_mount(fsc);
1029 	if (IS_ERR(res))
1030 		goto out_splat;
1031 	dout("root %p inode %p ino %llx.%llx\n", res,
1032 	     d_inode(res), ceph_vinop(d_inode(res)));
1033 	return res;
1034 
1035 out_splat:
1036 	ceph_mdsc_close_sessions(fsc->mdsc);
1037 	deactivate_locked_super(sb);
1038 	goto out_final;
1039 
1040 out:
1041 	ceph_mdsc_destroy(fsc);
1042 	destroy_fs_client(fsc);
1043 out_final:
1044 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1045 	return res;
1046 }
1047 
1048 static void ceph_kill_sb(struct super_block *s)
1049 {
1050 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1051 	dev_t dev = s->s_dev;
1052 
1053 	dout("kill_sb %p\n", s);
1054 
1055 	ceph_mdsc_pre_umount(fsc->mdsc);
1056 	generic_shutdown_super(s);
1057 	ceph_mdsc_destroy(fsc);
1058 
1059 	destroy_fs_client(fsc);
1060 	free_anon_bdev(dev);
1061 }
1062 
1063 static struct file_system_type ceph_fs_type = {
1064 	.owner		= THIS_MODULE,
1065 	.name		= "ceph",
1066 	.mount		= ceph_mount,
1067 	.kill_sb	= ceph_kill_sb,
1068 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1069 };
1070 MODULE_ALIAS_FS("ceph");
1071 
1072 static int __init init_ceph(void)
1073 {
1074 	int ret = init_caches();
1075 	if (ret)
1076 		goto out;
1077 
1078 	ceph_flock_init();
1079 	ceph_xattr_init();
1080 	ret = register_filesystem(&ceph_fs_type);
1081 	if (ret)
1082 		goto out_xattr;
1083 
1084 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1085 
1086 	return 0;
1087 
1088 out_xattr:
1089 	ceph_xattr_exit();
1090 	destroy_caches();
1091 out:
1092 	return ret;
1093 }
1094 
1095 static void __exit exit_ceph(void)
1096 {
1097 	dout("exit_ceph\n");
1098 	unregister_filesystem(&ceph_fs_type);
1099 	ceph_xattr_exit();
1100 	destroy_caches();
1101 }
1102 
1103 module_init(init_ceph);
1104 module_exit(exit_ceph);
1105 
1106 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1107 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1108 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1109 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1110 MODULE_LICENSE("GPL");
1111