Loading...
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 */
37static 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 * ensure we release the bdi before put_anon_super releases
46 * the device name.
47 */
48 if (s->s_bdi == &fsc->backing_dev_info) {
49 bdi_unregister(&fsc->backing_dev_info);
50 s->s_bdi = NULL;
51 }
52
53 return;
54}
55
56static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
57{
58 struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
59 struct ceph_monmap *monmap = fsc->client->monc.monmap;
60 struct ceph_statfs st;
61 u64 fsid;
62 int err;
63
64 dout("statfs\n");
65 err = ceph_monc_do_statfs(&fsc->client->monc, &st);
66 if (err < 0)
67 return err;
68
69 /* fill in kstatfs */
70 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
71
72 /*
73 * express utilization in terms of large blocks to avoid
74 * overflow on 32-bit machines.
75 *
76 * NOTE: for the time being, we make bsize == frsize to humor
77 * not-yet-ancient versions of glibc that are broken.
78 * Someday, we will probably want to report a real block
79 * size... whatever that may mean for a network file system!
80 */
81 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
82 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
83 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
84 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
85 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
86
87 buf->f_files = le64_to_cpu(st.num_objects);
88 buf->f_ffree = -1;
89 buf->f_namelen = NAME_MAX;
90
91 /* leave fsid little-endian, regardless of host endianness */
92 fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
93 buf->f_fsid.val[0] = fsid & 0xffffffff;
94 buf->f_fsid.val[1] = fsid >> 32;
95
96 return 0;
97}
98
99
100static int ceph_sync_fs(struct super_block *sb, int wait)
101{
102 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
103
104 if (!wait) {
105 dout("sync_fs (non-blocking)\n");
106 ceph_flush_dirty_caps(fsc->mdsc);
107 dout("sync_fs (non-blocking) done\n");
108 return 0;
109 }
110
111 dout("sync_fs (blocking)\n");
112 ceph_osdc_sync(&fsc->client->osdc);
113 ceph_mdsc_sync(fsc->mdsc);
114 dout("sync_fs (blocking) done\n");
115 return 0;
116}
117
118/*
119 * mount options
120 */
121enum {
122 Opt_wsize,
123 Opt_rsize,
124 Opt_rasize,
125 Opt_caps_wanted_delay_min,
126 Opt_caps_wanted_delay_max,
127 Opt_cap_release_safety,
128 Opt_readdir_max_entries,
129 Opt_readdir_max_bytes,
130 Opt_congestion_kb,
131 Opt_last_int,
132 /* int args above */
133 Opt_snapdirname,
134 Opt_last_string,
135 /* string args above */
136 Opt_dirstat,
137 Opt_nodirstat,
138 Opt_rbytes,
139 Opt_norbytes,
140 Opt_asyncreaddir,
141 Opt_noasyncreaddir,
142 Opt_dcache,
143 Opt_nodcache,
144 Opt_ino32,
145 Opt_noino32,
146 Opt_fscache,
147 Opt_nofscache,
148#ifdef CONFIG_CEPH_FS_POSIX_ACL
149 Opt_acl,
150#endif
151 Opt_noacl
152};
153
154static match_table_t fsopt_tokens = {
155 {Opt_wsize, "wsize=%d"},
156 {Opt_rsize, "rsize=%d"},
157 {Opt_rasize, "rasize=%d"},
158 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
159 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
160 {Opt_cap_release_safety, "cap_release_safety=%d"},
161 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
162 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
163 {Opt_congestion_kb, "write_congestion_kb=%d"},
164 /* int args above */
165 {Opt_snapdirname, "snapdirname=%s"},
166 /* string args above */
167 {Opt_dirstat, "dirstat"},
168 {Opt_nodirstat, "nodirstat"},
169 {Opt_rbytes, "rbytes"},
170 {Opt_norbytes, "norbytes"},
171 {Opt_asyncreaddir, "asyncreaddir"},
172 {Opt_noasyncreaddir, "noasyncreaddir"},
173 {Opt_dcache, "dcache"},
174 {Opt_nodcache, "nodcache"},
175 {Opt_ino32, "ino32"},
176 {Opt_noino32, "noino32"},
177 {Opt_fscache, "fsc"},
178 {Opt_nofscache, "nofsc"},
179#ifdef CONFIG_CEPH_FS_POSIX_ACL
180 {Opt_acl, "acl"},
181#endif
182 {Opt_noacl, "noacl"},
183 {-1, NULL}
184};
185
186static int parse_fsopt_token(char *c, void *private)
187{
188 struct ceph_mount_options *fsopt = private;
189 substring_t argstr[MAX_OPT_ARGS];
190 int token, intval, ret;
191
192 token = match_token((char *)c, fsopt_tokens, argstr);
193 if (token < 0)
194 return -EINVAL;
195
196 if (token < Opt_last_int) {
197 ret = match_int(&argstr[0], &intval);
198 if (ret < 0) {
199 pr_err("bad mount option arg (not int) "
200 "at '%s'\n", c);
201 return ret;
202 }
203 dout("got int token %d val %d\n", token, intval);
204 } else if (token > Opt_last_int && token < Opt_last_string) {
205 dout("got string token %d val %s\n", token,
206 argstr[0].from);
207 } else {
208 dout("got token %d\n", token);
209 }
210
211 switch (token) {
212 case Opt_snapdirname:
213 kfree(fsopt->snapdir_name);
214 fsopt->snapdir_name = kstrndup(argstr[0].from,
215 argstr[0].to-argstr[0].from,
216 GFP_KERNEL);
217 if (!fsopt->snapdir_name)
218 return -ENOMEM;
219 break;
220
221 /* misc */
222 case Opt_wsize:
223 fsopt->wsize = intval;
224 break;
225 case Opt_rsize:
226 fsopt->rsize = intval;
227 break;
228 case Opt_rasize:
229 fsopt->rasize = intval;
230 break;
231 case Opt_caps_wanted_delay_min:
232 fsopt->caps_wanted_delay_min = intval;
233 break;
234 case Opt_caps_wanted_delay_max:
235 fsopt->caps_wanted_delay_max = intval;
236 break;
237 case Opt_readdir_max_entries:
238 fsopt->max_readdir = intval;
239 break;
240 case Opt_readdir_max_bytes:
241 fsopt->max_readdir_bytes = intval;
242 break;
243 case Opt_congestion_kb:
244 fsopt->congestion_kb = intval;
245 break;
246 case Opt_dirstat:
247 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
248 break;
249 case Opt_nodirstat:
250 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
251 break;
252 case Opt_rbytes:
253 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
254 break;
255 case Opt_norbytes:
256 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
257 break;
258 case Opt_asyncreaddir:
259 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
260 break;
261 case Opt_noasyncreaddir:
262 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
263 break;
264 case Opt_dcache:
265 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
266 break;
267 case Opt_nodcache:
268 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
269 break;
270 case Opt_ino32:
271 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
272 break;
273 case Opt_noino32:
274 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
275 break;
276 case Opt_fscache:
277 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
278 break;
279 case Opt_nofscache:
280 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
281 break;
282#ifdef CONFIG_CEPH_FS_POSIX_ACL
283 case Opt_acl:
284 fsopt->sb_flags |= MS_POSIXACL;
285 break;
286#endif
287 case Opt_noacl:
288 fsopt->sb_flags &= ~MS_POSIXACL;
289 break;
290 default:
291 BUG_ON(token);
292 }
293 return 0;
294}
295
296static void destroy_mount_options(struct ceph_mount_options *args)
297{
298 dout("destroy_mount_options %p\n", args);
299 kfree(args->snapdir_name);
300 kfree(args);
301}
302
303static int strcmp_null(const char *s1, const char *s2)
304{
305 if (!s1 && !s2)
306 return 0;
307 if (s1 && !s2)
308 return -1;
309 if (!s1 && s2)
310 return 1;
311 return strcmp(s1, s2);
312}
313
314static int compare_mount_options(struct ceph_mount_options *new_fsopt,
315 struct ceph_options *new_opt,
316 struct ceph_fs_client *fsc)
317{
318 struct ceph_mount_options *fsopt1 = new_fsopt;
319 struct ceph_mount_options *fsopt2 = fsc->mount_options;
320 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
321 int ret;
322
323 ret = memcmp(fsopt1, fsopt2, ofs);
324 if (ret)
325 return ret;
326
327 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
328 if (ret)
329 return ret;
330
331 return ceph_compare_options(new_opt, fsc->client);
332}
333
334static int parse_mount_options(struct ceph_mount_options **pfsopt,
335 struct ceph_options **popt,
336 int flags, char *options,
337 const char *dev_name,
338 const char **path)
339{
340 struct ceph_mount_options *fsopt;
341 const char *dev_name_end;
342 int err;
343
344 if (!dev_name || !*dev_name)
345 return -EINVAL;
346
347 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
348 if (!fsopt)
349 return -ENOMEM;
350
351 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
352
353 fsopt->sb_flags = flags;
354 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
355
356 fsopt->rsize = CEPH_RSIZE_DEFAULT;
357 fsopt->rasize = CEPH_RASIZE_DEFAULT;
358 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
359 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
360 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
361 fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
362 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
363 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
364 fsopt->congestion_kb = default_congestion_kb();
365
366 /*
367 * Distinguish the server list from the path in "dev_name".
368 * Internally we do not include the leading '/' in the path.
369 *
370 * "dev_name" will look like:
371 * <server_spec>[,<server_spec>...]:[<path>]
372 * where
373 * <server_spec> is <ip>[:<port>]
374 * <path> is optional, but if present must begin with '/'
375 */
376 dev_name_end = strchr(dev_name, '/');
377 if (dev_name_end) {
378 /* skip over leading '/' for path */
379 *path = dev_name_end + 1;
380 } else {
381 /* path is empty */
382 dev_name_end = dev_name + strlen(dev_name);
383 *path = dev_name_end;
384 }
385 err = -EINVAL;
386 dev_name_end--; /* back up to ':' separator */
387 if (dev_name_end < dev_name || *dev_name_end != ':') {
388 pr_err("device name is missing path (no : separator in %s)\n",
389 dev_name);
390 goto out;
391 }
392 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
393 dout("server path '%s'\n", *path);
394
395 *popt = ceph_parse_options(options, dev_name, dev_name_end,
396 parse_fsopt_token, (void *)fsopt);
397 if (IS_ERR(*popt)) {
398 err = PTR_ERR(*popt);
399 goto out;
400 }
401
402 /* success */
403 *pfsopt = fsopt;
404 return 0;
405
406out:
407 destroy_mount_options(fsopt);
408 return err;
409}
410
411/**
412 * ceph_show_options - Show mount options in /proc/mounts
413 * @m: seq_file to write to
414 * @root: root of that (sub)tree
415 */
416static int ceph_show_options(struct seq_file *m, struct dentry *root)
417{
418 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
419 struct ceph_mount_options *fsopt = fsc->mount_options;
420 struct ceph_options *opt = fsc->client->options;
421
422 if (opt->flags & CEPH_OPT_FSID)
423 seq_printf(m, ",fsid=%pU", &opt->fsid);
424 if (opt->flags & CEPH_OPT_NOSHARE)
425 seq_puts(m, ",noshare");
426 if (opt->flags & CEPH_OPT_NOCRC)
427 seq_puts(m, ",nocrc");
428
429 if (opt->name)
430 seq_printf(m, ",name=%s", opt->name);
431 if (opt->key)
432 seq_puts(m, ",secret=<hidden>");
433
434 if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
435 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
436 if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
437 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
438 if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
439 seq_printf(m, ",osdkeepalivetimeout=%d",
440 opt->osd_keepalive_timeout);
441
442 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
443 seq_puts(m, ",dirstat");
444 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
445 seq_puts(m, ",norbytes");
446 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
447 seq_puts(m, ",noasyncreaddir");
448 if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
449 seq_puts(m, ",dcache");
450 else
451 seq_puts(m, ",nodcache");
452 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
453 seq_puts(m, ",fsc");
454 else
455 seq_puts(m, ",nofsc");
456
457#ifdef CONFIG_CEPH_FS_POSIX_ACL
458 if (fsopt->sb_flags & MS_POSIXACL)
459 seq_puts(m, ",acl");
460 else
461 seq_puts(m, ",noacl");
462#endif
463
464 if (fsopt->wsize)
465 seq_printf(m, ",wsize=%d", fsopt->wsize);
466 if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
467 seq_printf(m, ",rsize=%d", fsopt->rsize);
468 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
469 seq_printf(m, ",rasize=%d", fsopt->rasize);
470 if (fsopt->congestion_kb != default_congestion_kb())
471 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
472 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
473 seq_printf(m, ",caps_wanted_delay_min=%d",
474 fsopt->caps_wanted_delay_min);
475 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
476 seq_printf(m, ",caps_wanted_delay_max=%d",
477 fsopt->caps_wanted_delay_max);
478 if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
479 seq_printf(m, ",cap_release_safety=%d",
480 fsopt->cap_release_safety);
481 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
482 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
483 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
484 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
485 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
486 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
487 return 0;
488}
489
490/*
491 * handle any mon messages the standard library doesn't understand.
492 * return error if we don't either.
493 */
494static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
495{
496 struct ceph_fs_client *fsc = client->private;
497 int type = le16_to_cpu(msg->hdr.type);
498
499 switch (type) {
500 case CEPH_MSG_MDS_MAP:
501 ceph_mdsc_handle_map(fsc->mdsc, msg);
502 return 0;
503
504 default:
505 return -1;
506 }
507}
508
509/*
510 * create a new fs client
511 */
512static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
513 struct ceph_options *opt)
514{
515 struct ceph_fs_client *fsc;
516 const u64 supported_features =
517 CEPH_FEATURE_FLOCK |
518 CEPH_FEATURE_DIRLAYOUTHASH;
519 const u64 required_features = 0;
520 int page_count;
521 size_t size;
522 int err = -ENOMEM;
523
524 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
525 if (!fsc)
526 return ERR_PTR(-ENOMEM);
527
528 fsc->client = ceph_create_client(opt, fsc, supported_features,
529 required_features);
530 if (IS_ERR(fsc->client)) {
531 err = PTR_ERR(fsc->client);
532 goto fail;
533 }
534 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
535 fsc->client->monc.want_mdsmap = 1;
536
537 fsc->mount_options = fsopt;
538
539 fsc->sb = NULL;
540 fsc->mount_state = CEPH_MOUNT_MOUNTING;
541
542 atomic_long_set(&fsc->writeback_count, 0);
543
544 err = bdi_init(&fsc->backing_dev_info);
545 if (err < 0)
546 goto fail_client;
547
548 err = -ENOMEM;
549 /*
550 * The number of concurrent works can be high but they don't need
551 * to be processed in parallel, limit concurrency.
552 */
553 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
554 if (fsc->wb_wq == NULL)
555 goto fail_bdi;
556 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
557 if (fsc->pg_inv_wq == NULL)
558 goto fail_wb_wq;
559 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
560 if (fsc->trunc_wq == NULL)
561 goto fail_pg_inv_wq;
562
563 /* set up mempools */
564 err = -ENOMEM;
565 page_count = fsc->mount_options->wsize >> PAGE_CACHE_SHIFT;
566 size = sizeof (struct page *) * (page_count ? page_count : 1);
567 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
568 if (!fsc->wb_pagevec_pool)
569 goto fail_trunc_wq;
570
571 /* setup fscache */
572 if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
573 (ceph_fscache_register_fs(fsc) != 0))
574 goto fail_fscache;
575
576 /* caps */
577 fsc->min_caps = fsopt->max_readdir;
578
579 return fsc;
580
581fail_fscache:
582 ceph_fscache_unregister_fs(fsc);
583fail_trunc_wq:
584 destroy_workqueue(fsc->trunc_wq);
585fail_pg_inv_wq:
586 destroy_workqueue(fsc->pg_inv_wq);
587fail_wb_wq:
588 destroy_workqueue(fsc->wb_wq);
589fail_bdi:
590 bdi_destroy(&fsc->backing_dev_info);
591fail_client:
592 ceph_destroy_client(fsc->client);
593fail:
594 kfree(fsc);
595 return ERR_PTR(err);
596}
597
598static void destroy_fs_client(struct ceph_fs_client *fsc)
599{
600 dout("destroy_fs_client %p\n", fsc);
601
602 ceph_fscache_unregister_fs(fsc);
603
604 destroy_workqueue(fsc->wb_wq);
605 destroy_workqueue(fsc->pg_inv_wq);
606 destroy_workqueue(fsc->trunc_wq);
607
608 bdi_destroy(&fsc->backing_dev_info);
609
610 mempool_destroy(fsc->wb_pagevec_pool);
611
612 destroy_mount_options(fsc->mount_options);
613
614 ceph_fs_debugfs_cleanup(fsc);
615
616 ceph_destroy_client(fsc->client);
617
618 kfree(fsc);
619 dout("destroy_fs_client %p done\n", fsc);
620}
621
622/*
623 * caches
624 */
625struct kmem_cache *ceph_inode_cachep;
626struct kmem_cache *ceph_cap_cachep;
627struct kmem_cache *ceph_dentry_cachep;
628struct kmem_cache *ceph_file_cachep;
629
630static void ceph_inode_init_once(void *foo)
631{
632 struct ceph_inode_info *ci = foo;
633 inode_init_once(&ci->vfs_inode);
634}
635
636static int __init init_caches(void)
637{
638 int error = -ENOMEM;
639
640 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
641 sizeof(struct ceph_inode_info),
642 __alignof__(struct ceph_inode_info),
643 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
644 ceph_inode_init_once);
645 if (ceph_inode_cachep == NULL)
646 return -ENOMEM;
647
648 ceph_cap_cachep = KMEM_CACHE(ceph_cap,
649 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
650 if (ceph_cap_cachep == NULL)
651 goto bad_cap;
652
653 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
654 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
655 if (ceph_dentry_cachep == NULL)
656 goto bad_dentry;
657
658 ceph_file_cachep = KMEM_CACHE(ceph_file_info,
659 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
660 if (ceph_file_cachep == NULL)
661 goto bad_file;
662
663 if ((error = ceph_fscache_register()))
664 goto bad_file;
665
666 return 0;
667bad_file:
668 kmem_cache_destroy(ceph_dentry_cachep);
669bad_dentry:
670 kmem_cache_destroy(ceph_cap_cachep);
671bad_cap:
672 kmem_cache_destroy(ceph_inode_cachep);
673 return error;
674}
675
676static void destroy_caches(void)
677{
678 /*
679 * Make sure all delayed rcu free inodes are flushed before we
680 * destroy cache.
681 */
682 rcu_barrier();
683
684 kmem_cache_destroy(ceph_inode_cachep);
685 kmem_cache_destroy(ceph_cap_cachep);
686 kmem_cache_destroy(ceph_dentry_cachep);
687 kmem_cache_destroy(ceph_file_cachep);
688
689 ceph_fscache_unregister();
690}
691
692
693/*
694 * ceph_umount_begin - initiate forced umount. Tear down down the
695 * mount, skipping steps that may hang while waiting for server(s).
696 */
697static void ceph_umount_begin(struct super_block *sb)
698{
699 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
700
701 dout("ceph_umount_begin - starting forced umount\n");
702 if (!fsc)
703 return;
704 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
705 return;
706}
707
708static const struct super_operations ceph_super_ops = {
709 .alloc_inode = ceph_alloc_inode,
710 .destroy_inode = ceph_destroy_inode,
711 .write_inode = ceph_write_inode,
712 .drop_inode = ceph_drop_inode,
713 .sync_fs = ceph_sync_fs,
714 .put_super = ceph_put_super,
715 .show_options = ceph_show_options,
716 .statfs = ceph_statfs,
717 .umount_begin = ceph_umount_begin,
718};
719
720/*
721 * Bootstrap mount by opening the root directory. Note the mount
722 * @started time from caller, and time out if this takes too long.
723 */
724static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
725 const char *path,
726 unsigned long started)
727{
728 struct ceph_mds_client *mdsc = fsc->mdsc;
729 struct ceph_mds_request *req = NULL;
730 int err;
731 struct dentry *root;
732
733 /* open dir */
734 dout("open_root_inode opening '%s'\n", path);
735 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
736 if (IS_ERR(req))
737 return ERR_CAST(req);
738 req->r_path1 = kstrdup(path, GFP_NOFS);
739 req->r_ino1.ino = CEPH_INO_ROOT;
740 req->r_ino1.snap = CEPH_NOSNAP;
741 req->r_started = started;
742 req->r_timeout = fsc->client->options->mount_timeout * HZ;
743 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
744 req->r_num_caps = 2;
745 err = ceph_mdsc_do_request(mdsc, NULL, req);
746 if (err == 0) {
747 struct inode *inode = req->r_target_inode;
748 req->r_target_inode = NULL;
749 dout("open_root_inode success\n");
750 if (ceph_ino(inode) == CEPH_INO_ROOT &&
751 fsc->sb->s_root == NULL) {
752 root = d_make_root(inode);
753 if (!root) {
754 root = ERR_PTR(-ENOMEM);
755 goto out;
756 }
757 } else {
758 root = d_obtain_alias(inode);
759 }
760 ceph_init_dentry(root);
761 dout("open_root_inode success, root dentry is %p\n", root);
762 } else {
763 root = ERR_PTR(err);
764 }
765out:
766 ceph_mdsc_put_request(req);
767 return root;
768}
769
770
771
772
773/*
774 * mount: join the ceph cluster, and open root directory.
775 */
776static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
777 const char *path)
778{
779 int err;
780 unsigned long started = jiffies; /* note the start time */
781 struct dentry *root;
782 int first = 0; /* first vfsmount for this super_block */
783
784 dout("mount start\n");
785 mutex_lock(&fsc->client->mount_mutex);
786
787 err = __ceph_open_session(fsc->client, started);
788 if (err < 0)
789 goto out;
790
791 dout("mount opening root\n");
792 root = open_root_dentry(fsc, "", started);
793 if (IS_ERR(root)) {
794 err = PTR_ERR(root);
795 goto out;
796 }
797 if (fsc->sb->s_root) {
798 dput(root);
799 } else {
800 fsc->sb->s_root = root;
801 first = 1;
802
803 err = ceph_fs_debugfs_init(fsc);
804 if (err < 0)
805 goto fail;
806 }
807
808 if (path[0] == 0) {
809 dget(root);
810 } else {
811 dout("mount opening base mountpoint\n");
812 root = open_root_dentry(fsc, path, started);
813 if (IS_ERR(root)) {
814 err = PTR_ERR(root);
815 goto fail;
816 }
817 }
818
819 fsc->mount_state = CEPH_MOUNT_MOUNTED;
820 dout("mount success\n");
821 mutex_unlock(&fsc->client->mount_mutex);
822 return root;
823
824out:
825 mutex_unlock(&fsc->client->mount_mutex);
826 return ERR_PTR(err);
827
828fail:
829 if (first) {
830 dput(fsc->sb->s_root);
831 fsc->sb->s_root = NULL;
832 }
833 goto out;
834}
835
836static int ceph_set_super(struct super_block *s, void *data)
837{
838 struct ceph_fs_client *fsc = data;
839 int ret;
840
841 dout("set_super %p data %p\n", s, data);
842
843 s->s_flags = fsc->mount_options->sb_flags;
844 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
845
846 s->s_xattr = ceph_xattr_handlers;
847 s->s_fs_info = fsc;
848 fsc->sb = s;
849
850 s->s_op = &ceph_super_ops;
851 s->s_export_op = &ceph_export_ops;
852
853 s->s_time_gran = 1000; /* 1000 ns == 1 us */
854
855 ret = set_anon_super(s, NULL); /* what is that second arg for? */
856 if (ret != 0)
857 goto fail;
858
859 return ret;
860
861fail:
862 s->s_fs_info = NULL;
863 fsc->sb = NULL;
864 return ret;
865}
866
867/*
868 * share superblock if same fs AND options
869 */
870static int ceph_compare_super(struct super_block *sb, void *data)
871{
872 struct ceph_fs_client *new = data;
873 struct ceph_mount_options *fsopt = new->mount_options;
874 struct ceph_options *opt = new->client->options;
875 struct ceph_fs_client *other = ceph_sb_to_client(sb);
876
877 dout("ceph_compare_super %p\n", sb);
878
879 if (compare_mount_options(fsopt, opt, other)) {
880 dout("monitor(s)/mount options don't match\n");
881 return 0;
882 }
883 if ((opt->flags & CEPH_OPT_FSID) &&
884 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
885 dout("fsid doesn't match\n");
886 return 0;
887 }
888 if (fsopt->sb_flags != other->mount_options->sb_flags) {
889 dout("flags differ\n");
890 return 0;
891 }
892 return 1;
893}
894
895/*
896 * construct our own bdi so we can control readahead, etc.
897 */
898static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
899
900static int ceph_register_bdi(struct super_block *sb,
901 struct ceph_fs_client *fsc)
902{
903 int err;
904
905 /* set ra_pages based on rasize mount option? */
906 if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
907 fsc->backing_dev_info.ra_pages =
908 (fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
909 >> PAGE_SHIFT;
910 else
911 fsc->backing_dev_info.ra_pages =
912 default_backing_dev_info.ra_pages;
913
914 err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
915 atomic_long_inc_return(&bdi_seq));
916 if (!err)
917 sb->s_bdi = &fsc->backing_dev_info;
918 return err;
919}
920
921static struct dentry *ceph_mount(struct file_system_type *fs_type,
922 int flags, const char *dev_name, void *data)
923{
924 struct super_block *sb;
925 struct ceph_fs_client *fsc;
926 struct dentry *res;
927 int err;
928 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
929 const char *path = NULL;
930 struct ceph_mount_options *fsopt = NULL;
931 struct ceph_options *opt = NULL;
932
933 dout("ceph_mount\n");
934
935#ifdef CONFIG_CEPH_FS_POSIX_ACL
936 flags |= MS_POSIXACL;
937#endif
938 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
939 if (err < 0) {
940 res = ERR_PTR(err);
941 goto out_final;
942 }
943
944 /* create client (which we may/may not use) */
945 fsc = create_fs_client(fsopt, opt);
946 if (IS_ERR(fsc)) {
947 res = ERR_CAST(fsc);
948 destroy_mount_options(fsopt);
949 ceph_destroy_options(opt);
950 goto out_final;
951 }
952
953 err = ceph_mdsc_init(fsc);
954 if (err < 0) {
955 res = ERR_PTR(err);
956 goto out;
957 }
958
959 if (ceph_test_opt(fsc->client, NOSHARE))
960 compare_super = NULL;
961 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
962 if (IS_ERR(sb)) {
963 res = ERR_CAST(sb);
964 goto out;
965 }
966
967 if (ceph_sb_to_client(sb) != fsc) {
968 ceph_mdsc_destroy(fsc);
969 destroy_fs_client(fsc);
970 fsc = ceph_sb_to_client(sb);
971 dout("get_sb got existing client %p\n", fsc);
972 } else {
973 dout("get_sb using new client %p\n", fsc);
974 err = ceph_register_bdi(sb, fsc);
975 if (err < 0) {
976 res = ERR_PTR(err);
977 goto out_splat;
978 }
979 }
980
981 res = ceph_real_mount(fsc, path);
982 if (IS_ERR(res))
983 goto out_splat;
984 dout("root %p inode %p ino %llx.%llx\n", res,
985 res->d_inode, ceph_vinop(res->d_inode));
986 return res;
987
988out_splat:
989 ceph_mdsc_close_sessions(fsc->mdsc);
990 deactivate_locked_super(sb);
991 goto out_final;
992
993out:
994 ceph_mdsc_destroy(fsc);
995 destroy_fs_client(fsc);
996out_final:
997 dout("ceph_mount fail %ld\n", PTR_ERR(res));
998 return res;
999}
1000
1001static void ceph_kill_sb(struct super_block *s)
1002{
1003 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1004 dout("kill_sb %p\n", s);
1005 ceph_mdsc_pre_umount(fsc->mdsc);
1006 kill_anon_super(s); /* will call put_super after sb is r/o */
1007 ceph_mdsc_destroy(fsc);
1008 destroy_fs_client(fsc);
1009}
1010
1011static struct file_system_type ceph_fs_type = {
1012 .owner = THIS_MODULE,
1013 .name = "ceph",
1014 .mount = ceph_mount,
1015 .kill_sb = ceph_kill_sb,
1016 .fs_flags = FS_RENAME_DOES_D_MOVE,
1017};
1018MODULE_ALIAS_FS("ceph");
1019
1020#define _STRINGIFY(x) #x
1021#define STRINGIFY(x) _STRINGIFY(x)
1022
1023static int __init init_ceph(void)
1024{
1025 int ret = init_caches();
1026 if (ret)
1027 goto out;
1028
1029 ceph_flock_init();
1030 ceph_xattr_init();
1031 ret = register_filesystem(&ceph_fs_type);
1032 if (ret)
1033 goto out_icache;
1034
1035 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1036
1037 return 0;
1038
1039out_icache:
1040 ceph_xattr_exit();
1041 destroy_caches();
1042out:
1043 return ret;
1044}
1045
1046static void __exit exit_ceph(void)
1047{
1048 dout("exit_ceph\n");
1049 unregister_filesystem(&ceph_fs_type);
1050 ceph_xattr_exit();
1051 destroy_caches();
1052}
1053
1054module_init(init_ceph);
1055module_exit(exit_ceph);
1056
1057MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1058MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1059MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1060MODULE_DESCRIPTION("Ceph filesystem for Linux");
1061MODULE_LICENSE("GPL");
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
21#include <linux/ceph/decode.h>
22#include <linux/ceph/mon_client.h>
23#include <linux/ceph/auth.h>
24#include <linux/ceph/debugfs.h>
25
26/*
27 * Ceph superblock operations
28 *
29 * Handle the basics of mounting, unmounting.
30 */
31
32/*
33 * super ops
34 */
35static void ceph_put_super(struct super_block *s)
36{
37 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
38
39 dout("put_super\n");
40 ceph_mdsc_close_sessions(fsc->mdsc);
41
42 /*
43 * ensure we release the bdi before put_anon_super releases
44 * the device name.
45 */
46 if (s->s_bdi == &fsc->backing_dev_info) {
47 bdi_unregister(&fsc->backing_dev_info);
48 s->s_bdi = NULL;
49 }
50
51 return;
52}
53
54static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
55{
56 struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
57 struct ceph_monmap *monmap = fsc->client->monc.monmap;
58 struct ceph_statfs st;
59 u64 fsid;
60 int err;
61
62 dout("statfs\n");
63 err = ceph_monc_do_statfs(&fsc->client->monc, &st);
64 if (err < 0)
65 return err;
66
67 /* fill in kstatfs */
68 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
69
70 /*
71 * express utilization in terms of large blocks to avoid
72 * overflow on 32-bit machines.
73 */
74 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
75 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
76 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
77 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
78
79 buf->f_files = le64_to_cpu(st.num_objects);
80 buf->f_ffree = -1;
81 buf->f_namelen = NAME_MAX;
82 buf->f_frsize = PAGE_CACHE_SIZE;
83
84 /* leave fsid little-endian, regardless of host endianness */
85 fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
86 buf->f_fsid.val[0] = fsid & 0xffffffff;
87 buf->f_fsid.val[1] = fsid >> 32;
88
89 return 0;
90}
91
92
93static int ceph_sync_fs(struct super_block *sb, int wait)
94{
95 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
96
97 if (!wait) {
98 dout("sync_fs (non-blocking)\n");
99 ceph_flush_dirty_caps(fsc->mdsc);
100 dout("sync_fs (non-blocking) done\n");
101 return 0;
102 }
103
104 dout("sync_fs (blocking)\n");
105 ceph_osdc_sync(&fsc->client->osdc);
106 ceph_mdsc_sync(fsc->mdsc);
107 dout("sync_fs (blocking) done\n");
108 return 0;
109}
110
111/*
112 * mount options
113 */
114enum {
115 Opt_wsize,
116 Opt_rsize,
117 Opt_rasize,
118 Opt_caps_wanted_delay_min,
119 Opt_caps_wanted_delay_max,
120 Opt_cap_release_safety,
121 Opt_readdir_max_entries,
122 Opt_readdir_max_bytes,
123 Opt_congestion_kb,
124 Opt_last_int,
125 /* int args above */
126 Opt_snapdirname,
127 Opt_last_string,
128 /* string args above */
129 Opt_dirstat,
130 Opt_nodirstat,
131 Opt_rbytes,
132 Opt_norbytes,
133 Opt_asyncreaddir,
134 Opt_noasyncreaddir,
135 Opt_dcache,
136 Opt_nodcache,
137 Opt_ino32,
138 Opt_noino32,
139};
140
141static match_table_t fsopt_tokens = {
142 {Opt_wsize, "wsize=%d"},
143 {Opt_rsize, "rsize=%d"},
144 {Opt_rasize, "rasize=%d"},
145 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
146 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
147 {Opt_cap_release_safety, "cap_release_safety=%d"},
148 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
149 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
150 {Opt_congestion_kb, "write_congestion_kb=%d"},
151 /* int args above */
152 {Opt_snapdirname, "snapdirname=%s"},
153 /* string args above */
154 {Opt_dirstat, "dirstat"},
155 {Opt_nodirstat, "nodirstat"},
156 {Opt_rbytes, "rbytes"},
157 {Opt_norbytes, "norbytes"},
158 {Opt_asyncreaddir, "asyncreaddir"},
159 {Opt_noasyncreaddir, "noasyncreaddir"},
160 {Opt_dcache, "dcache"},
161 {Opt_nodcache, "nodcache"},
162 {Opt_ino32, "ino32"},
163 {Opt_noino32, "noino32"},
164 {-1, NULL}
165};
166
167static int parse_fsopt_token(char *c, void *private)
168{
169 struct ceph_mount_options *fsopt = private;
170 substring_t argstr[MAX_OPT_ARGS];
171 int token, intval, ret;
172
173 token = match_token((char *)c, fsopt_tokens, argstr);
174 if (token < 0)
175 return -EINVAL;
176
177 if (token < Opt_last_int) {
178 ret = match_int(&argstr[0], &intval);
179 if (ret < 0) {
180 pr_err("bad mount option arg (not int) "
181 "at '%s'\n", c);
182 return ret;
183 }
184 dout("got int token %d val %d\n", token, intval);
185 } else if (token > Opt_last_int && token < Opt_last_string) {
186 dout("got string token %d val %s\n", token,
187 argstr[0].from);
188 } else {
189 dout("got token %d\n", token);
190 }
191
192 switch (token) {
193 case Opt_snapdirname:
194 kfree(fsopt->snapdir_name);
195 fsopt->snapdir_name = kstrndup(argstr[0].from,
196 argstr[0].to-argstr[0].from,
197 GFP_KERNEL);
198 if (!fsopt->snapdir_name)
199 return -ENOMEM;
200 break;
201
202 /* misc */
203 case Opt_wsize:
204 fsopt->wsize = intval;
205 break;
206 case Opt_rsize:
207 fsopt->rsize = intval;
208 break;
209 case Opt_rasize:
210 fsopt->rasize = intval;
211 break;
212 case Opt_caps_wanted_delay_min:
213 fsopt->caps_wanted_delay_min = intval;
214 break;
215 case Opt_caps_wanted_delay_max:
216 fsopt->caps_wanted_delay_max = intval;
217 break;
218 case Opt_readdir_max_entries:
219 fsopt->max_readdir = intval;
220 break;
221 case Opt_readdir_max_bytes:
222 fsopt->max_readdir_bytes = intval;
223 break;
224 case Opt_congestion_kb:
225 fsopt->congestion_kb = intval;
226 break;
227 case Opt_dirstat:
228 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
229 break;
230 case Opt_nodirstat:
231 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
232 break;
233 case Opt_rbytes:
234 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
235 break;
236 case Opt_norbytes:
237 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
238 break;
239 case Opt_asyncreaddir:
240 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
241 break;
242 case Opt_noasyncreaddir:
243 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
244 break;
245 case Opt_dcache:
246 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
247 break;
248 case Opt_nodcache:
249 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
250 break;
251 case Opt_ino32:
252 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
253 break;
254 case Opt_noino32:
255 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
256 break;
257 default:
258 BUG_ON(token);
259 }
260 return 0;
261}
262
263static void destroy_mount_options(struct ceph_mount_options *args)
264{
265 dout("destroy_mount_options %p\n", args);
266 kfree(args->snapdir_name);
267 kfree(args);
268}
269
270static int strcmp_null(const char *s1, const char *s2)
271{
272 if (!s1 && !s2)
273 return 0;
274 if (s1 && !s2)
275 return -1;
276 if (!s1 && s2)
277 return 1;
278 return strcmp(s1, s2);
279}
280
281static int compare_mount_options(struct ceph_mount_options *new_fsopt,
282 struct ceph_options *new_opt,
283 struct ceph_fs_client *fsc)
284{
285 struct ceph_mount_options *fsopt1 = new_fsopt;
286 struct ceph_mount_options *fsopt2 = fsc->mount_options;
287 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
288 int ret;
289
290 ret = memcmp(fsopt1, fsopt2, ofs);
291 if (ret)
292 return ret;
293
294 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
295 if (ret)
296 return ret;
297
298 return ceph_compare_options(new_opt, fsc->client);
299}
300
301static int parse_mount_options(struct ceph_mount_options **pfsopt,
302 struct ceph_options **popt,
303 int flags, char *options,
304 const char *dev_name,
305 const char **path)
306{
307 struct ceph_mount_options *fsopt;
308 const char *dev_name_end;
309 int err = -ENOMEM;
310
311 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
312 if (!fsopt)
313 return -ENOMEM;
314
315 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
316
317 fsopt->sb_flags = flags;
318 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
319
320 fsopt->rsize = CEPH_RSIZE_DEFAULT;
321 fsopt->rasize = CEPH_RASIZE_DEFAULT;
322 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
323 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
324 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
325 fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
326 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
327 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
328 fsopt->congestion_kb = default_congestion_kb();
329
330 /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
331 err = -EINVAL;
332 if (!dev_name)
333 goto out;
334 *path = strstr(dev_name, ":/");
335 if (*path == NULL) {
336 pr_err("device name is missing path (no :/ in %s)\n",
337 dev_name);
338 goto out;
339 }
340 dev_name_end = *path;
341 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
342
343 /* path on server */
344 *path += 2;
345 dout("server path '%s'\n", *path);
346
347 *popt = ceph_parse_options(options, dev_name, dev_name_end,
348 parse_fsopt_token, (void *)fsopt);
349 if (IS_ERR(*popt)) {
350 err = PTR_ERR(*popt);
351 goto out;
352 }
353
354 /* success */
355 *pfsopt = fsopt;
356 return 0;
357
358out:
359 destroy_mount_options(fsopt);
360 return err;
361}
362
363/**
364 * ceph_show_options - Show mount options in /proc/mounts
365 * @m: seq_file to write to
366 * @root: root of that (sub)tree
367 */
368static int ceph_show_options(struct seq_file *m, struct dentry *root)
369{
370 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
371 struct ceph_mount_options *fsopt = fsc->mount_options;
372 struct ceph_options *opt = fsc->client->options;
373
374 if (opt->flags & CEPH_OPT_FSID)
375 seq_printf(m, ",fsid=%pU", &opt->fsid);
376 if (opt->flags & CEPH_OPT_NOSHARE)
377 seq_puts(m, ",noshare");
378 if (opt->flags & CEPH_OPT_NOCRC)
379 seq_puts(m, ",nocrc");
380
381 if (opt->name)
382 seq_printf(m, ",name=%s", opt->name);
383 if (opt->key)
384 seq_puts(m, ",secret=<hidden>");
385
386 if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
387 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
388 if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
389 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
390 if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
391 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
392 if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
393 seq_printf(m, ",osdkeepalivetimeout=%d",
394 opt->osd_keepalive_timeout);
395
396 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
397 seq_puts(m, ",dirstat");
398 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
399 seq_puts(m, ",norbytes");
400 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
401 seq_puts(m, ",noasyncreaddir");
402 if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
403 seq_puts(m, ",dcache");
404 else
405 seq_puts(m, ",nodcache");
406
407 if (fsopt->wsize)
408 seq_printf(m, ",wsize=%d", fsopt->wsize);
409 if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
410 seq_printf(m, ",rsize=%d", fsopt->rsize);
411 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
412 seq_printf(m, ",rasize=%d", fsopt->rasize);
413 if (fsopt->congestion_kb != default_congestion_kb())
414 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
415 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
416 seq_printf(m, ",caps_wanted_delay_min=%d",
417 fsopt->caps_wanted_delay_min);
418 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
419 seq_printf(m, ",caps_wanted_delay_max=%d",
420 fsopt->caps_wanted_delay_max);
421 if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
422 seq_printf(m, ",cap_release_safety=%d",
423 fsopt->cap_release_safety);
424 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
425 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
426 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
427 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
428 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
429 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
430 return 0;
431}
432
433/*
434 * handle any mon messages the standard library doesn't understand.
435 * return error if we don't either.
436 */
437static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
438{
439 struct ceph_fs_client *fsc = client->private;
440 int type = le16_to_cpu(msg->hdr.type);
441
442 switch (type) {
443 case CEPH_MSG_MDS_MAP:
444 ceph_mdsc_handle_map(fsc->mdsc, msg);
445 return 0;
446
447 default:
448 return -1;
449 }
450}
451
452/*
453 * create a new fs client
454 */
455static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
456 struct ceph_options *opt)
457{
458 struct ceph_fs_client *fsc;
459 const unsigned supported_features =
460 CEPH_FEATURE_FLOCK |
461 CEPH_FEATURE_DIRLAYOUTHASH;
462 const unsigned required_features = 0;
463 int err = -ENOMEM;
464
465 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
466 if (!fsc)
467 return ERR_PTR(-ENOMEM);
468
469 fsc->client = ceph_create_client(opt, fsc, supported_features,
470 required_features);
471 if (IS_ERR(fsc->client)) {
472 err = PTR_ERR(fsc->client);
473 goto fail;
474 }
475 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
476 fsc->client->monc.want_mdsmap = 1;
477
478 fsc->mount_options = fsopt;
479
480 fsc->sb = NULL;
481 fsc->mount_state = CEPH_MOUNT_MOUNTING;
482
483 atomic_long_set(&fsc->writeback_count, 0);
484
485 err = bdi_init(&fsc->backing_dev_info);
486 if (err < 0)
487 goto fail_client;
488
489 err = -ENOMEM;
490 /*
491 * The number of concurrent works can be high but they don't need
492 * to be processed in parallel, limit concurrency.
493 */
494 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
495 if (fsc->wb_wq == NULL)
496 goto fail_bdi;
497 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
498 if (fsc->pg_inv_wq == NULL)
499 goto fail_wb_wq;
500 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
501 if (fsc->trunc_wq == NULL)
502 goto fail_pg_inv_wq;
503
504 /* set up mempools */
505 err = -ENOMEM;
506 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
507 fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
508 if (!fsc->wb_pagevec_pool)
509 goto fail_trunc_wq;
510
511 /* caps */
512 fsc->min_caps = fsopt->max_readdir;
513
514 return fsc;
515
516fail_trunc_wq:
517 destroy_workqueue(fsc->trunc_wq);
518fail_pg_inv_wq:
519 destroy_workqueue(fsc->pg_inv_wq);
520fail_wb_wq:
521 destroy_workqueue(fsc->wb_wq);
522fail_bdi:
523 bdi_destroy(&fsc->backing_dev_info);
524fail_client:
525 ceph_destroy_client(fsc->client);
526fail:
527 kfree(fsc);
528 return ERR_PTR(err);
529}
530
531static void destroy_fs_client(struct ceph_fs_client *fsc)
532{
533 dout("destroy_fs_client %p\n", fsc);
534
535 destroy_workqueue(fsc->wb_wq);
536 destroy_workqueue(fsc->pg_inv_wq);
537 destroy_workqueue(fsc->trunc_wq);
538
539 bdi_destroy(&fsc->backing_dev_info);
540
541 mempool_destroy(fsc->wb_pagevec_pool);
542
543 destroy_mount_options(fsc->mount_options);
544
545 ceph_fs_debugfs_cleanup(fsc);
546
547 ceph_destroy_client(fsc->client);
548
549 kfree(fsc);
550 dout("destroy_fs_client %p done\n", fsc);
551}
552
553/*
554 * caches
555 */
556struct kmem_cache *ceph_inode_cachep;
557struct kmem_cache *ceph_cap_cachep;
558struct kmem_cache *ceph_dentry_cachep;
559struct kmem_cache *ceph_file_cachep;
560
561static void ceph_inode_init_once(void *foo)
562{
563 struct ceph_inode_info *ci = foo;
564 inode_init_once(&ci->vfs_inode);
565}
566
567static int __init init_caches(void)
568{
569 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
570 sizeof(struct ceph_inode_info),
571 __alignof__(struct ceph_inode_info),
572 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
573 ceph_inode_init_once);
574 if (ceph_inode_cachep == NULL)
575 return -ENOMEM;
576
577 ceph_cap_cachep = KMEM_CACHE(ceph_cap,
578 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
579 if (ceph_cap_cachep == NULL)
580 goto bad_cap;
581
582 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
583 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
584 if (ceph_dentry_cachep == NULL)
585 goto bad_dentry;
586
587 ceph_file_cachep = KMEM_CACHE(ceph_file_info,
588 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
589 if (ceph_file_cachep == NULL)
590 goto bad_file;
591
592 return 0;
593
594bad_file:
595 kmem_cache_destroy(ceph_dentry_cachep);
596bad_dentry:
597 kmem_cache_destroy(ceph_cap_cachep);
598bad_cap:
599 kmem_cache_destroy(ceph_inode_cachep);
600 return -ENOMEM;
601}
602
603static void destroy_caches(void)
604{
605 kmem_cache_destroy(ceph_inode_cachep);
606 kmem_cache_destroy(ceph_cap_cachep);
607 kmem_cache_destroy(ceph_dentry_cachep);
608 kmem_cache_destroy(ceph_file_cachep);
609}
610
611
612/*
613 * ceph_umount_begin - initiate forced umount. Tear down down the
614 * mount, skipping steps that may hang while waiting for server(s).
615 */
616static void ceph_umount_begin(struct super_block *sb)
617{
618 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
619
620 dout("ceph_umount_begin - starting forced umount\n");
621 if (!fsc)
622 return;
623 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
624 return;
625}
626
627static const struct super_operations ceph_super_ops = {
628 .alloc_inode = ceph_alloc_inode,
629 .destroy_inode = ceph_destroy_inode,
630 .write_inode = ceph_write_inode,
631 .sync_fs = ceph_sync_fs,
632 .put_super = ceph_put_super,
633 .show_options = ceph_show_options,
634 .statfs = ceph_statfs,
635 .umount_begin = ceph_umount_begin,
636};
637
638/*
639 * Bootstrap mount by opening the root directory. Note the mount
640 * @started time from caller, and time out if this takes too long.
641 */
642static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
643 const char *path,
644 unsigned long started)
645{
646 struct ceph_mds_client *mdsc = fsc->mdsc;
647 struct ceph_mds_request *req = NULL;
648 int err;
649 struct dentry *root;
650
651 /* open dir */
652 dout("open_root_inode opening '%s'\n", path);
653 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
654 if (IS_ERR(req))
655 return ERR_CAST(req);
656 req->r_path1 = kstrdup(path, GFP_NOFS);
657 req->r_ino1.ino = CEPH_INO_ROOT;
658 req->r_ino1.snap = CEPH_NOSNAP;
659 req->r_started = started;
660 req->r_timeout = fsc->client->options->mount_timeout * HZ;
661 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
662 req->r_num_caps = 2;
663 err = ceph_mdsc_do_request(mdsc, NULL, req);
664 if (err == 0) {
665 struct inode *inode = req->r_target_inode;
666 req->r_target_inode = NULL;
667 dout("open_root_inode success\n");
668 if (ceph_ino(inode) == CEPH_INO_ROOT &&
669 fsc->sb->s_root == NULL) {
670 root = d_make_root(inode);
671 if (!root) {
672 root = ERR_PTR(-ENOMEM);
673 goto out;
674 }
675 } else {
676 root = d_obtain_alias(inode);
677 }
678 ceph_init_dentry(root);
679 dout("open_root_inode success, root dentry is %p\n", root);
680 } else {
681 root = ERR_PTR(err);
682 }
683out:
684 ceph_mdsc_put_request(req);
685 return root;
686}
687
688
689
690
691/*
692 * mount: join the ceph cluster, and open root directory.
693 */
694static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
695 const char *path)
696{
697 int err;
698 unsigned long started = jiffies; /* note the start time */
699 struct dentry *root;
700 int first = 0; /* first vfsmount for this super_block */
701
702 dout("mount start\n");
703 mutex_lock(&fsc->client->mount_mutex);
704
705 err = __ceph_open_session(fsc->client, started);
706 if (err < 0)
707 goto out;
708
709 dout("mount opening root\n");
710 root = open_root_dentry(fsc, "", started);
711 if (IS_ERR(root)) {
712 err = PTR_ERR(root);
713 goto out;
714 }
715 if (fsc->sb->s_root) {
716 dput(root);
717 } else {
718 fsc->sb->s_root = root;
719 first = 1;
720
721 err = ceph_fs_debugfs_init(fsc);
722 if (err < 0)
723 goto fail;
724 }
725
726 if (path[0] == 0) {
727 dget(root);
728 } else {
729 dout("mount opening base mountpoint\n");
730 root = open_root_dentry(fsc, path, started);
731 if (IS_ERR(root)) {
732 err = PTR_ERR(root);
733 goto fail;
734 }
735 }
736
737 fsc->mount_state = CEPH_MOUNT_MOUNTED;
738 dout("mount success\n");
739 mutex_unlock(&fsc->client->mount_mutex);
740 return root;
741
742out:
743 mutex_unlock(&fsc->client->mount_mutex);
744 return ERR_PTR(err);
745
746fail:
747 if (first) {
748 dput(fsc->sb->s_root);
749 fsc->sb->s_root = NULL;
750 }
751 goto out;
752}
753
754static int ceph_set_super(struct super_block *s, void *data)
755{
756 struct ceph_fs_client *fsc = data;
757 int ret;
758
759 dout("set_super %p data %p\n", s, data);
760
761 s->s_flags = fsc->mount_options->sb_flags;
762 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
763
764 s->s_fs_info = fsc;
765 fsc->sb = s;
766
767 s->s_op = &ceph_super_ops;
768 s->s_export_op = &ceph_export_ops;
769
770 s->s_time_gran = 1000; /* 1000 ns == 1 us */
771
772 ret = set_anon_super(s, NULL); /* what is that second arg for? */
773 if (ret != 0)
774 goto fail;
775
776 return ret;
777
778fail:
779 s->s_fs_info = NULL;
780 fsc->sb = NULL;
781 return ret;
782}
783
784/*
785 * share superblock if same fs AND options
786 */
787static int ceph_compare_super(struct super_block *sb, void *data)
788{
789 struct ceph_fs_client *new = data;
790 struct ceph_mount_options *fsopt = new->mount_options;
791 struct ceph_options *opt = new->client->options;
792 struct ceph_fs_client *other = ceph_sb_to_client(sb);
793
794 dout("ceph_compare_super %p\n", sb);
795
796 if (compare_mount_options(fsopt, opt, other)) {
797 dout("monitor(s)/mount options don't match\n");
798 return 0;
799 }
800 if ((opt->flags & CEPH_OPT_FSID) &&
801 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
802 dout("fsid doesn't match\n");
803 return 0;
804 }
805 if (fsopt->sb_flags != other->mount_options->sb_flags) {
806 dout("flags differ\n");
807 return 0;
808 }
809 return 1;
810}
811
812/*
813 * construct our own bdi so we can control readahead, etc.
814 */
815static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
816
817static int ceph_register_bdi(struct super_block *sb,
818 struct ceph_fs_client *fsc)
819{
820 int err;
821
822 /* set ra_pages based on rasize mount option? */
823 if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
824 fsc->backing_dev_info.ra_pages =
825 (fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
826 >> PAGE_SHIFT;
827 else
828 fsc->backing_dev_info.ra_pages =
829 default_backing_dev_info.ra_pages;
830
831 err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
832 atomic_long_inc_return(&bdi_seq));
833 if (!err)
834 sb->s_bdi = &fsc->backing_dev_info;
835 return err;
836}
837
838static struct dentry *ceph_mount(struct file_system_type *fs_type,
839 int flags, const char *dev_name, void *data)
840{
841 struct super_block *sb;
842 struct ceph_fs_client *fsc;
843 struct dentry *res;
844 int err;
845 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
846 const char *path = NULL;
847 struct ceph_mount_options *fsopt = NULL;
848 struct ceph_options *opt = NULL;
849
850 dout("ceph_mount\n");
851 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
852 if (err < 0) {
853 res = ERR_PTR(err);
854 goto out_final;
855 }
856
857 /* create client (which we may/may not use) */
858 fsc = create_fs_client(fsopt, opt);
859 if (IS_ERR(fsc)) {
860 res = ERR_CAST(fsc);
861 destroy_mount_options(fsopt);
862 ceph_destroy_options(opt);
863 goto out_final;
864 }
865
866 err = ceph_mdsc_init(fsc);
867 if (err < 0) {
868 res = ERR_PTR(err);
869 goto out;
870 }
871
872 if (ceph_test_opt(fsc->client, NOSHARE))
873 compare_super = NULL;
874 sb = sget(fs_type, compare_super, ceph_set_super, fsc);
875 if (IS_ERR(sb)) {
876 res = ERR_CAST(sb);
877 goto out;
878 }
879
880 if (ceph_sb_to_client(sb) != fsc) {
881 ceph_mdsc_destroy(fsc);
882 destroy_fs_client(fsc);
883 fsc = ceph_sb_to_client(sb);
884 dout("get_sb got existing client %p\n", fsc);
885 } else {
886 dout("get_sb using new client %p\n", fsc);
887 err = ceph_register_bdi(sb, fsc);
888 if (err < 0) {
889 res = ERR_PTR(err);
890 goto out_splat;
891 }
892 }
893
894 res = ceph_real_mount(fsc, path);
895 if (IS_ERR(res))
896 goto out_splat;
897 dout("root %p inode %p ino %llx.%llx\n", res,
898 res->d_inode, ceph_vinop(res->d_inode));
899 return res;
900
901out_splat:
902 ceph_mdsc_close_sessions(fsc->mdsc);
903 deactivate_locked_super(sb);
904 goto out_final;
905
906out:
907 ceph_mdsc_destroy(fsc);
908 destroy_fs_client(fsc);
909out_final:
910 dout("ceph_mount fail %ld\n", PTR_ERR(res));
911 return res;
912}
913
914static void ceph_kill_sb(struct super_block *s)
915{
916 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
917 dout("kill_sb %p\n", s);
918 ceph_mdsc_pre_umount(fsc->mdsc);
919 kill_anon_super(s); /* will call put_super after sb is r/o */
920 ceph_mdsc_destroy(fsc);
921 destroy_fs_client(fsc);
922}
923
924static struct file_system_type ceph_fs_type = {
925 .owner = THIS_MODULE,
926 .name = "ceph",
927 .mount = ceph_mount,
928 .kill_sb = ceph_kill_sb,
929 .fs_flags = FS_RENAME_DOES_D_MOVE,
930};
931
932#define _STRINGIFY(x) #x
933#define STRINGIFY(x) _STRINGIFY(x)
934
935static int __init init_ceph(void)
936{
937 int ret = init_caches();
938 if (ret)
939 goto out;
940
941 ceph_xattr_init();
942 ret = register_filesystem(&ceph_fs_type);
943 if (ret)
944 goto out_icache;
945
946 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
947
948 return 0;
949
950out_icache:
951 ceph_xattr_exit();
952 destroy_caches();
953out:
954 return ret;
955}
956
957static void __exit exit_ceph(void)
958{
959 dout("exit_ceph\n");
960 unregister_filesystem(&ceph_fs_type);
961 ceph_xattr_exit();
962 destroy_caches();
963}
964
965module_init(init_ceph);
966module_exit(exit_ceph);
967
968MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
969MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
970MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
971MODULE_DESCRIPTION("Ceph filesystem for Linux");
972MODULE_LICENSE("GPL");