xref: /openbmc/linux/fs/ceph/super.c (revision 6dfcd296)
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 		if (strlen(dev_name_end) > 1) {
400 			fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
401 			if (!fsopt->server_path) {
402 				err = -ENOMEM;
403 				goto out;
404 			}
405 		}
406 	} else {
407 		dev_name_end = dev_name + strlen(dev_name);
408 	}
409 	err = -EINVAL;
410 	dev_name_end--;		/* back up to ':' separator */
411 	if (dev_name_end < dev_name || *dev_name_end != ':') {
412 		pr_err("device name is missing path (no : separator in %s)\n",
413 				dev_name);
414 		goto out;
415 	}
416 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
417 	if (fsopt->server_path)
418 		dout("server path '%s'\n", fsopt->server_path);
419 
420 	*popt = ceph_parse_options(options, dev_name, dev_name_end,
421 				 parse_fsopt_token, (void *)fsopt);
422 	if (IS_ERR(*popt)) {
423 		err = PTR_ERR(*popt);
424 		goto out;
425 	}
426 
427 	/* success */
428 	*pfsopt = fsopt;
429 	return 0;
430 
431 out:
432 	destroy_mount_options(fsopt);
433 	return err;
434 }
435 
436 /**
437  * ceph_show_options - Show mount options in /proc/mounts
438  * @m: seq_file to write to
439  * @root: root of that (sub)tree
440  */
441 static int ceph_show_options(struct seq_file *m, struct dentry *root)
442 {
443 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
444 	struct ceph_mount_options *fsopt = fsc->mount_options;
445 	size_t pos;
446 	int ret;
447 
448 	/* a comma between MNT/MS and client options */
449 	seq_putc(m, ',');
450 	pos = m->count;
451 
452 	ret = ceph_print_client_options(m, fsc->client);
453 	if (ret)
454 		return ret;
455 
456 	/* retract our comma if no client options */
457 	if (m->count == pos)
458 		m->count--;
459 
460 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
461 		seq_puts(m, ",dirstat");
462 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
463 		seq_puts(m, ",rbytes");
464 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
465 		seq_puts(m, ",noasyncreaddir");
466 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
467 		seq_puts(m, ",nodcache");
468 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
469 		seq_puts(m, ",fsc");
470 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
471 		seq_puts(m, ",nopoolperm");
472 
473 #ifdef CONFIG_CEPH_FS_POSIX_ACL
474 	if (fsopt->sb_flags & MS_POSIXACL)
475 		seq_puts(m, ",acl");
476 	else
477 		seq_puts(m, ",noacl");
478 #endif
479 
480 	if (fsopt->mds_namespace)
481 		seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
482 	if (fsopt->wsize)
483 		seq_printf(m, ",wsize=%d", fsopt->wsize);
484 	if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
485 		seq_printf(m, ",rsize=%d", fsopt->rsize);
486 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
487 		seq_printf(m, ",rasize=%d", fsopt->rasize);
488 	if (fsopt->congestion_kb != default_congestion_kb())
489 		seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
490 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
491 		seq_printf(m, ",caps_wanted_delay_min=%d",
492 			 fsopt->caps_wanted_delay_min);
493 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
494 		seq_printf(m, ",caps_wanted_delay_max=%d",
495 			   fsopt->caps_wanted_delay_max);
496 	if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
497 		seq_printf(m, ",cap_release_safety=%d",
498 			   fsopt->cap_release_safety);
499 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
500 		seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
501 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
502 		seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
503 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
504 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
505 
506 	return 0;
507 }
508 
509 /*
510  * handle any mon messages the standard library doesn't understand.
511  * return error if we don't either.
512  */
513 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
514 {
515 	struct ceph_fs_client *fsc = client->private;
516 	int type = le16_to_cpu(msg->hdr.type);
517 
518 	switch (type) {
519 	case CEPH_MSG_MDS_MAP:
520 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
521 		return 0;
522 	case CEPH_MSG_FS_MAP_USER:
523 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
524 		return 0;
525 	default:
526 		return -1;
527 	}
528 }
529 
530 /*
531  * create a new fs client
532  */
533 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
534 					struct ceph_options *opt)
535 {
536 	struct ceph_fs_client *fsc;
537 	const u64 supported_features =
538 		CEPH_FEATURE_FLOCK | CEPH_FEATURE_DIRLAYOUTHASH |
539 		CEPH_FEATURE_MDSENC | CEPH_FEATURE_MDS_INLINE_DATA;
540 	const u64 required_features = 0;
541 	int page_count;
542 	size_t size;
543 	int err = -ENOMEM;
544 
545 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
546 	if (!fsc)
547 		return ERR_PTR(-ENOMEM);
548 
549 	fsc->client = ceph_create_client(opt, fsc, supported_features,
550 					 required_features);
551 	if (IS_ERR(fsc->client)) {
552 		err = PTR_ERR(fsc->client);
553 		goto fail;
554 	}
555 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
556 
557 	if (fsopt->mds_namespace == NULL) {
558 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
559 				   0, true);
560 	} else {
561 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
562 				   0, false);
563 	}
564 
565 	fsc->mount_options = fsopt;
566 
567 	fsc->sb = NULL;
568 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
569 
570 	atomic_long_set(&fsc->writeback_count, 0);
571 
572 	err = bdi_init(&fsc->backing_dev_info);
573 	if (err < 0)
574 		goto fail_client;
575 
576 	err = -ENOMEM;
577 	/*
578 	 * The number of concurrent works can be high but they don't need
579 	 * to be processed in parallel, limit concurrency.
580 	 */
581 	fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
582 	if (fsc->wb_wq == NULL)
583 		goto fail_bdi;
584 	fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
585 	if (fsc->pg_inv_wq == NULL)
586 		goto fail_wb_wq;
587 	fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
588 	if (fsc->trunc_wq == NULL)
589 		goto fail_pg_inv_wq;
590 
591 	/* set up mempools */
592 	err = -ENOMEM;
593 	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
594 	size = sizeof (struct page *) * (page_count ? page_count : 1);
595 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
596 	if (!fsc->wb_pagevec_pool)
597 		goto fail_trunc_wq;
598 
599 	/* setup fscache */
600 	if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
601 	    (ceph_fscache_register_fs(fsc) != 0))
602 		goto fail_fscache;
603 
604 	/* caps */
605 	fsc->min_caps = fsopt->max_readdir;
606 
607 	return fsc;
608 
609 fail_fscache:
610 	ceph_fscache_unregister_fs(fsc);
611 fail_trunc_wq:
612 	destroy_workqueue(fsc->trunc_wq);
613 fail_pg_inv_wq:
614 	destroy_workqueue(fsc->pg_inv_wq);
615 fail_wb_wq:
616 	destroy_workqueue(fsc->wb_wq);
617 fail_bdi:
618 	bdi_destroy(&fsc->backing_dev_info);
619 fail_client:
620 	ceph_destroy_client(fsc->client);
621 fail:
622 	kfree(fsc);
623 	return ERR_PTR(err);
624 }
625 
626 static void destroy_fs_client(struct ceph_fs_client *fsc)
627 {
628 	dout("destroy_fs_client %p\n", fsc);
629 
630 	ceph_fscache_unregister_fs(fsc);
631 
632 	destroy_workqueue(fsc->wb_wq);
633 	destroy_workqueue(fsc->pg_inv_wq);
634 	destroy_workqueue(fsc->trunc_wq);
635 
636 	bdi_destroy(&fsc->backing_dev_info);
637 
638 	mempool_destroy(fsc->wb_pagevec_pool);
639 
640 	destroy_mount_options(fsc->mount_options);
641 
642 	ceph_fs_debugfs_cleanup(fsc);
643 
644 	ceph_destroy_client(fsc->client);
645 
646 	kfree(fsc);
647 	dout("destroy_fs_client %p done\n", fsc);
648 }
649 
650 /*
651  * caches
652  */
653 struct kmem_cache *ceph_inode_cachep;
654 struct kmem_cache *ceph_cap_cachep;
655 struct kmem_cache *ceph_cap_flush_cachep;
656 struct kmem_cache *ceph_dentry_cachep;
657 struct kmem_cache *ceph_file_cachep;
658 
659 static void ceph_inode_init_once(void *foo)
660 {
661 	struct ceph_inode_info *ci = foo;
662 	inode_init_once(&ci->vfs_inode);
663 }
664 
665 static int __init init_caches(void)
666 {
667 	int error = -ENOMEM;
668 
669 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
670 				      sizeof(struct ceph_inode_info),
671 				      __alignof__(struct ceph_inode_info),
672 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
673 				      SLAB_ACCOUNT, ceph_inode_init_once);
674 	if (ceph_inode_cachep == NULL)
675 		return -ENOMEM;
676 
677 	ceph_cap_cachep = KMEM_CACHE(ceph_cap,
678 				     SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
679 	if (ceph_cap_cachep == NULL)
680 		goto bad_cap;
681 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
682 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
683 	if (ceph_cap_flush_cachep == NULL)
684 		goto bad_cap_flush;
685 
686 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
687 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
688 	if (ceph_dentry_cachep == NULL)
689 		goto bad_dentry;
690 
691 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
692 
693 	if (ceph_file_cachep == NULL)
694 		goto bad_file;
695 
696 	if ((error = ceph_fscache_register()))
697 		goto bad_file;
698 
699 	return 0;
700 bad_file:
701 	kmem_cache_destroy(ceph_dentry_cachep);
702 bad_dentry:
703 	kmem_cache_destroy(ceph_cap_flush_cachep);
704 bad_cap_flush:
705 	kmem_cache_destroy(ceph_cap_cachep);
706 bad_cap:
707 	kmem_cache_destroy(ceph_inode_cachep);
708 	return error;
709 }
710 
711 static void destroy_caches(void)
712 {
713 	/*
714 	 * Make sure all delayed rcu free inodes are flushed before we
715 	 * destroy cache.
716 	 */
717 	rcu_barrier();
718 
719 	kmem_cache_destroy(ceph_inode_cachep);
720 	kmem_cache_destroy(ceph_cap_cachep);
721 	kmem_cache_destroy(ceph_cap_flush_cachep);
722 	kmem_cache_destroy(ceph_dentry_cachep);
723 	kmem_cache_destroy(ceph_file_cachep);
724 
725 	ceph_fscache_unregister();
726 }
727 
728 
729 /*
730  * ceph_umount_begin - initiate forced umount.  Tear down down the
731  * mount, skipping steps that may hang while waiting for server(s).
732  */
733 static void ceph_umount_begin(struct super_block *sb)
734 {
735 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
736 
737 	dout("ceph_umount_begin - starting forced umount\n");
738 	if (!fsc)
739 		return;
740 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
741 	ceph_mdsc_force_umount(fsc->mdsc);
742 	return;
743 }
744 
745 static const struct super_operations ceph_super_ops = {
746 	.alloc_inode	= ceph_alloc_inode,
747 	.destroy_inode	= ceph_destroy_inode,
748 	.write_inode    = ceph_write_inode,
749 	.drop_inode	= ceph_drop_inode,
750 	.evict_inode	= ceph_evict_inode,
751 	.sync_fs        = ceph_sync_fs,
752 	.put_super	= ceph_put_super,
753 	.show_options   = ceph_show_options,
754 	.statfs		= ceph_statfs,
755 	.umount_begin   = ceph_umount_begin,
756 };
757 
758 /*
759  * Bootstrap mount by opening the root directory.  Note the mount
760  * @started time from caller, and time out if this takes too long.
761  */
762 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
763 				       const char *path,
764 				       unsigned long started)
765 {
766 	struct ceph_mds_client *mdsc = fsc->mdsc;
767 	struct ceph_mds_request *req = NULL;
768 	int err;
769 	struct dentry *root;
770 
771 	/* open dir */
772 	dout("open_root_inode opening '%s'\n", path);
773 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
774 	if (IS_ERR(req))
775 		return ERR_CAST(req);
776 	req->r_path1 = kstrdup(path, GFP_NOFS);
777 	if (!req->r_path1) {
778 		root = ERR_PTR(-ENOMEM);
779 		goto out;
780 	}
781 
782 	req->r_ino1.ino = CEPH_INO_ROOT;
783 	req->r_ino1.snap = CEPH_NOSNAP;
784 	req->r_started = started;
785 	req->r_timeout = fsc->client->options->mount_timeout;
786 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
787 	req->r_num_caps = 2;
788 	err = ceph_mdsc_do_request(mdsc, NULL, req);
789 	if (err == 0) {
790 		struct inode *inode = req->r_target_inode;
791 		req->r_target_inode = NULL;
792 		dout("open_root_inode success\n");
793 		root = d_make_root(inode);
794 		if (!root) {
795 			root = ERR_PTR(-ENOMEM);
796 			goto out;
797 		}
798 		ceph_init_dentry(root);
799 		dout("open_root_inode success, root dentry is %p\n", root);
800 	} else {
801 		root = ERR_PTR(err);
802 	}
803 out:
804 	ceph_mdsc_put_request(req);
805 	return root;
806 }
807 
808 
809 
810 
811 /*
812  * mount: join the ceph cluster, and open root directory.
813  */
814 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
815 {
816 	int err;
817 	unsigned long started = jiffies;  /* note the start time */
818 	struct dentry *root;
819 	int first = 0;   /* first vfsmount for this super_block */
820 
821 	dout("mount start %p\n", fsc);
822 	mutex_lock(&fsc->client->mount_mutex);
823 
824 	if (!fsc->sb->s_root) {
825 		const char *path;
826 		err = __ceph_open_session(fsc->client, started);
827 		if (err < 0)
828 			goto out;
829 
830 		if (!fsc->mount_options->server_path) {
831 			path = "";
832 			dout("mount opening path \\t\n");
833 		} else {
834 			path = fsc->mount_options->server_path + 1;
835 			dout("mount opening path %s\n", path);
836 		}
837 		root = open_root_dentry(fsc, path, started);
838 		if (IS_ERR(root)) {
839 			err = PTR_ERR(root);
840 			goto out;
841 		}
842 		fsc->sb->s_root = dget(root);
843 		first = 1;
844 
845 		err = ceph_fs_debugfs_init(fsc);
846 		if (err < 0)
847 			goto fail;
848 	}
849 
850 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
851 	dout("mount success\n");
852 	mutex_unlock(&fsc->client->mount_mutex);
853 	return root;
854 
855 fail:
856 	if (first) {
857 		dput(fsc->sb->s_root);
858 		fsc->sb->s_root = NULL;
859 	}
860 out:
861 	mutex_unlock(&fsc->client->mount_mutex);
862 	return ERR_PTR(err);
863 }
864 
865 static int ceph_set_super(struct super_block *s, void *data)
866 {
867 	struct ceph_fs_client *fsc = data;
868 	int ret;
869 
870 	dout("set_super %p data %p\n", s, data);
871 
872 	s->s_flags = fsc->mount_options->sb_flags;
873 	s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
874 
875 	s->s_xattr = ceph_xattr_handlers;
876 	s->s_fs_info = fsc;
877 	fsc->sb = s;
878 
879 	s->s_op = &ceph_super_ops;
880 	s->s_export_op = &ceph_export_ops;
881 
882 	s->s_time_gran = 1000;  /* 1000 ns == 1 us */
883 
884 	ret = set_anon_super(s, NULL);  /* what is that second arg for? */
885 	if (ret != 0)
886 		goto fail;
887 
888 	return ret;
889 
890 fail:
891 	s->s_fs_info = NULL;
892 	fsc->sb = NULL;
893 	return ret;
894 }
895 
896 /*
897  * share superblock if same fs AND options
898  */
899 static int ceph_compare_super(struct super_block *sb, void *data)
900 {
901 	struct ceph_fs_client *new = data;
902 	struct ceph_mount_options *fsopt = new->mount_options;
903 	struct ceph_options *opt = new->client->options;
904 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
905 
906 	dout("ceph_compare_super %p\n", sb);
907 
908 	if (compare_mount_options(fsopt, opt, other)) {
909 		dout("monitor(s)/mount options don't match\n");
910 		return 0;
911 	}
912 	if ((opt->flags & CEPH_OPT_FSID) &&
913 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
914 		dout("fsid doesn't match\n");
915 		return 0;
916 	}
917 	if (fsopt->sb_flags != other->mount_options->sb_flags) {
918 		dout("flags differ\n");
919 		return 0;
920 	}
921 	return 1;
922 }
923 
924 /*
925  * construct our own bdi so we can control readahead, etc.
926  */
927 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
928 
929 static int ceph_register_bdi(struct super_block *sb,
930 			     struct ceph_fs_client *fsc)
931 {
932 	int err;
933 
934 	/* set ra_pages based on rasize mount option? */
935 	if (fsc->mount_options->rasize >= PAGE_SIZE)
936 		fsc->backing_dev_info.ra_pages =
937 			(fsc->mount_options->rasize + PAGE_SIZE - 1)
938 			>> PAGE_SHIFT;
939 	else
940 		fsc->backing_dev_info.ra_pages =
941 			VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
942 
943 	err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
944 			   atomic_long_inc_return(&bdi_seq));
945 	if (!err)
946 		sb->s_bdi = &fsc->backing_dev_info;
947 	return err;
948 }
949 
950 static struct dentry *ceph_mount(struct file_system_type *fs_type,
951 		       int flags, const char *dev_name, void *data)
952 {
953 	struct super_block *sb;
954 	struct ceph_fs_client *fsc;
955 	struct dentry *res;
956 	int err;
957 	int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
958 	struct ceph_mount_options *fsopt = NULL;
959 	struct ceph_options *opt = NULL;
960 
961 	dout("ceph_mount\n");
962 
963 #ifdef CONFIG_CEPH_FS_POSIX_ACL
964 	flags |= MS_POSIXACL;
965 #endif
966 	err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
967 	if (err < 0) {
968 		res = ERR_PTR(err);
969 		goto out_final;
970 	}
971 
972 	/* create client (which we may/may not use) */
973 	fsc = create_fs_client(fsopt, opt);
974 	if (IS_ERR(fsc)) {
975 		res = ERR_CAST(fsc);
976 		destroy_mount_options(fsopt);
977 		ceph_destroy_options(opt);
978 		goto out_final;
979 	}
980 
981 	err = ceph_mdsc_init(fsc);
982 	if (err < 0) {
983 		res = ERR_PTR(err);
984 		goto out;
985 	}
986 
987 	if (ceph_test_opt(fsc->client, NOSHARE))
988 		compare_super = NULL;
989 	sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
990 	if (IS_ERR(sb)) {
991 		res = ERR_CAST(sb);
992 		goto out;
993 	}
994 
995 	if (ceph_sb_to_client(sb) != fsc) {
996 		ceph_mdsc_destroy(fsc);
997 		destroy_fs_client(fsc);
998 		fsc = ceph_sb_to_client(sb);
999 		dout("get_sb got existing client %p\n", fsc);
1000 	} else {
1001 		dout("get_sb using new client %p\n", fsc);
1002 		err = ceph_register_bdi(sb, fsc);
1003 		if (err < 0) {
1004 			res = ERR_PTR(err);
1005 			goto out_splat;
1006 		}
1007 	}
1008 
1009 	res = ceph_real_mount(fsc);
1010 	if (IS_ERR(res))
1011 		goto out_splat;
1012 	dout("root %p inode %p ino %llx.%llx\n", res,
1013 	     d_inode(res), ceph_vinop(d_inode(res)));
1014 	return res;
1015 
1016 out_splat:
1017 	ceph_mdsc_close_sessions(fsc->mdsc);
1018 	deactivate_locked_super(sb);
1019 	goto out_final;
1020 
1021 out:
1022 	ceph_mdsc_destroy(fsc);
1023 	destroy_fs_client(fsc);
1024 out_final:
1025 	dout("ceph_mount fail %ld\n", PTR_ERR(res));
1026 	return res;
1027 }
1028 
1029 static void ceph_kill_sb(struct super_block *s)
1030 {
1031 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1032 	dev_t dev = s->s_dev;
1033 
1034 	dout("kill_sb %p\n", s);
1035 
1036 	ceph_mdsc_pre_umount(fsc->mdsc);
1037 	generic_shutdown_super(s);
1038 	ceph_mdsc_destroy(fsc);
1039 
1040 	destroy_fs_client(fsc);
1041 	free_anon_bdev(dev);
1042 }
1043 
1044 static struct file_system_type ceph_fs_type = {
1045 	.owner		= THIS_MODULE,
1046 	.name		= "ceph",
1047 	.mount		= ceph_mount,
1048 	.kill_sb	= ceph_kill_sb,
1049 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1050 };
1051 MODULE_ALIAS_FS("ceph");
1052 
1053 static int __init init_ceph(void)
1054 {
1055 	int ret = init_caches();
1056 	if (ret)
1057 		goto out;
1058 
1059 	ceph_flock_init();
1060 	ceph_xattr_init();
1061 	ret = register_filesystem(&ceph_fs_type);
1062 	if (ret)
1063 		goto out_xattr;
1064 
1065 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1066 
1067 	return 0;
1068 
1069 out_xattr:
1070 	ceph_xattr_exit();
1071 	destroy_caches();
1072 out:
1073 	return ret;
1074 }
1075 
1076 static void __exit exit_ceph(void)
1077 {
1078 	dout("exit_ceph\n");
1079 	unregister_filesystem(&ceph_fs_type);
1080 	ceph_xattr_exit();
1081 	destroy_caches();
1082 }
1083 
1084 module_init(init_ceph);
1085 module_exit(exit_ceph);
1086 
1087 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1088 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1089 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1090 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1091 MODULE_LICENSE("GPL");
1092