?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'arm64-stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmarinas...
[linux-2.6.git] / fs / ceph / super.c
blob6627b26a800ca0e74649ecf439076bb9c6a4b095
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/ceph_features.h>
22 #include <linux/ceph/decode.h>
23 #include <linux/ceph/mon_client.h>
24 #include <linux/ceph/auth.h>
25 #include <linux/ceph/debugfs.h>
26
27 /*
28 * Ceph superblock operations
29 *
30 * Handle the basics of mounting, unmounting.
31 */
32
33 /*
34 * super ops
35 */
36 static void ceph_put_super(struct super_block *s)
37 {
38 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
39
40 dout("put_super\n");
41 ceph_mdsc_close_sessions(fsc->mdsc);
42
43 /*
44 * ensure we release the bdi before put_anon_super releases
45 * the device name.
46 */
47 if (s->s_bdi == &fsc->backing_dev_info) {
48 bdi_unregister(&fsc->backing_dev_info);
49 s->s_bdi = NULL;
50 }
51
52 return;
53 }
54
55 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
56 {
57 struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
58 struct ceph_monmap *monmap = fsc->client->monc.monmap;
59 struct ceph_statfs st;
60 u64 fsid;
61 int err;
62
63 dout("statfs\n");
64 err = ceph_monc_do_statfs(&fsc->client->monc, &st);
65 if (err < 0)
66 return err;
67
68 /* fill in kstatfs */
69 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
70
71 /*
72 * express utilization in terms of large blocks to avoid
73 * overflow on 32-bit machines.
74 *
75 * NOTE: for the time being, we make bsize == frsize to humor
76 * not-yet-ancient versions of glibc that are broken.
77 * Someday, we will probably want to report a real block
78 * size... whatever that may mean for a network file system!
79 */
80 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
81 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
82 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
83 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
84 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
85
86 buf->f_files = le64_to_cpu(st.num_objects);
87 buf->f_ffree = -1;
88 buf->f_namelen = NAME_MAX;
89
90 /* leave fsid little-endian, regardless of host endianness */
91 fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
92 buf->f_fsid.val[0] = fsid & 0xffffffff;
93 buf->f_fsid.val[1] = fsid >> 32;
94
95 return 0;
96 }
97
98
99 static int ceph_sync_fs(struct super_block *sb, int wait)
100 {
101 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
102
103 if (!wait) {
104 dout("sync_fs (non-blocking)\n");
105 ceph_flush_dirty_caps(fsc->mdsc);
106 dout("sync_fs (non-blocking) done\n");
107 return 0;
108 }
109
110 dout("sync_fs (blocking)\n");
111 ceph_osdc_sync(&fsc->client->osdc);
112 ceph_mdsc_sync(fsc->mdsc);
113 dout("sync_fs (blocking) done\n");
114 return 0;
115 }
116
117 /*
118 * mount options
119 */
120 enum {
121 Opt_wsize,
122 Opt_rsize,
123 Opt_rasize,
124 Opt_caps_wanted_delay_min,
125 Opt_caps_wanted_delay_max,
126 Opt_cap_release_safety,
127 Opt_readdir_max_entries,
128 Opt_readdir_max_bytes,
129 Opt_congestion_kb,
130 Opt_last_int,
131 /* int args above */
132 Opt_snapdirname,
133 Opt_last_string,
134 /* string args above */
135 Opt_dirstat,
136 Opt_nodirstat,
137 Opt_rbytes,
138 Opt_norbytes,
139 Opt_asyncreaddir,
140 Opt_noasyncreaddir,
141 Opt_dcache,
142 Opt_nodcache,
143 Opt_ino32,
144 Opt_noino32,
145 };
146
147 static match_table_t fsopt_tokens = {
148 {Opt_wsize, "wsize=%d"},
149 {Opt_rsize, "rsize=%d"},
150 {Opt_rasize, "rasize=%d"},
151 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
152 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
153 {Opt_cap_release_safety, "cap_release_safety=%d"},
154 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
155 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
156 {Opt_congestion_kb, "write_congestion_kb=%d"},
157 /* int args above */
158 {Opt_snapdirname, "snapdirname=%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 {-1, NULL}
171 };
172
173 static int parse_fsopt_token(char *c, void *private)
174 {
175 struct ceph_mount_options *fsopt = private;
176 substring_t argstr[MAX_OPT_ARGS];
177 int token, intval, ret;
178
179 token = match_token((char *)c, fsopt_tokens, argstr);
180 if (token < 0)
181 return -EINVAL;
182
183 if (token < Opt_last_int) {
184 ret = match_int(&argstr[0], &intval);
185 if (ret < 0) {
186 pr_err("bad mount option arg (not int) "
187 "at '%s'\n", c);
188 return ret;
189 }
190 dout("got int token %d val %d\n", token, intval);
191 } else if (token > Opt_last_int && token < Opt_last_string) {
192 dout("got string token %d val %s\n", token,
193 argstr[0].from);
194 } else {
195 dout("got token %d\n", token);
196 }
197
198 switch (token) {
199 case Opt_snapdirname:
200 kfree(fsopt->snapdir_name);
201 fsopt->snapdir_name = kstrndup(argstr[0].from,
202 argstr[0].to-argstr[0].from,
203 GFP_KERNEL);
204 if (!fsopt->snapdir_name)
205 return -ENOMEM;
206 break;
207
208 /* misc */
209 case Opt_wsize:
210 fsopt->wsize = intval;
211 break;
212 case Opt_rsize:
213 fsopt->rsize = intval;
214 break;
215 case Opt_rasize:
216 fsopt->rasize = intval;
217 break;
218 case Opt_caps_wanted_delay_min:
219 fsopt->caps_wanted_delay_min = intval;
220 break;
221 case Opt_caps_wanted_delay_max:
222 fsopt->caps_wanted_delay_max = intval;
223 break;
224 case Opt_readdir_max_entries:
225 fsopt->max_readdir = intval;
226 break;
227 case Opt_readdir_max_bytes:
228 fsopt->max_readdir_bytes = intval;
229 break;
230 case Opt_congestion_kb:
231 fsopt->congestion_kb = intval;
232 break;
233 case Opt_dirstat:
234 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
235 break;
236 case Opt_nodirstat:
237 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
238 break;
239 case Opt_rbytes:
240 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
241 break;
242 case Opt_norbytes:
243 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
244 break;
245 case Opt_asyncreaddir:
246 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
247 break;
248 case Opt_noasyncreaddir:
249 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
250 break;
251 case Opt_dcache:
252 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
253 break;
254 case Opt_nodcache:
255 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
256 break;
257 case Opt_ino32:
258 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
259 break;
260 case Opt_noino32:
261 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
262 break;
263 default:
264 BUG_ON(token);
265 }
266 return 0;
267 }
268
269 static void destroy_mount_options(struct ceph_mount_options *args)
270 {
271 dout("destroy_mount_options %p\n", args);
272 kfree(args->snapdir_name);
273 kfree(args);
274 }
275
276 static int strcmp_null(const char *s1, const char *s2)
277 {
278 if (!s1 && !s2)
279 return 0;
280 if (s1 && !s2)
281 return -1;
282 if (!s1 && s2)
283 return 1;
284 return strcmp(s1, s2);
285 }
286
287 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
288 struct ceph_options *new_opt,
289 struct ceph_fs_client *fsc)
290 {
291 struct ceph_mount_options *fsopt1 = new_fsopt;
292 struct ceph_mount_options *fsopt2 = fsc->mount_options;
293 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
294 int ret;
295
296 ret = memcmp(fsopt1, fsopt2, ofs);
297 if (ret)
298 return ret;
299
300 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
301 if (ret)
302 return ret;
303
304 return ceph_compare_options(new_opt, fsc->client);
305 }
306
307 static int parse_mount_options(struct ceph_mount_options **pfsopt,
308 struct ceph_options **popt,
309 int flags, char *options,
310 const char *dev_name,
311 const char **path)
312 {
313 struct ceph_mount_options *fsopt;
314 const char *dev_name_end;
315 int err;
316
317 if (!dev_name || !*dev_name)
318 return -EINVAL;
319
320 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
321 if (!fsopt)
322 return -ENOMEM;
323
324 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
325
326 fsopt->sb_flags = flags;
327 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
328
329 fsopt->rsize = CEPH_RSIZE_DEFAULT;
330 fsopt->rasize = CEPH_RASIZE_DEFAULT;
331 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
332 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
333 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
334 fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
335 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
336 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
337 fsopt->congestion_kb = default_congestion_kb();
338
339 /*
340 * Distinguish the server list from the path in "dev_name".
341 * Internally we do not include the leading '/' in the path.
342 *
343 * "dev_name" will look like:
344 * <server_spec>[,<server_spec>...]:[<path>]
345 * where
346 * <server_spec> is <ip>[:<port>]
347 * <path> is optional, but if present must begin with '/'
348 */
349 dev_name_end = strchr(dev_name, '/');
350 if (dev_name_end) {
351 /* skip over leading '/' for path */
352 *path = dev_name_end + 1;
353 } else {
354 /* path is empty */
355 dev_name_end = dev_name + strlen(dev_name);
356 *path = dev_name_end;
357 }
358 err = -EINVAL;
359 dev_name_end--; /* back up to ':' separator */
360 if (dev_name_end < dev_name || *dev_name_end != ':') {
361 pr_err("device name is missing path (no : separator in %s)\n",
362 dev_name);
363 goto out;
364 }
365 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
366 dout("server path '%s'\n", *path);
367
368 *popt = ceph_parse_options(options, dev_name, dev_name_end,
369 parse_fsopt_token, (void *)fsopt);
370 if (IS_ERR(*popt)) {
371 err = PTR_ERR(*popt);
372 goto out;
373 }
374
375 /* success */
376 *pfsopt = fsopt;
377 return 0;
378
379 out:
380 destroy_mount_options(fsopt);
381 return err;
382 }
383
384 /**
385 * ceph_show_options - Show mount options in /proc/mounts
386 * @m: seq_file to write to
387 * @root: root of that (sub)tree
388 */
389 static int ceph_show_options(struct seq_file *m, struct dentry *root)
390 {
391 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
392 struct ceph_mount_options *fsopt = fsc->mount_options;
393 struct ceph_options *opt = fsc->client->options;
394
395 if (opt->flags & CEPH_OPT_FSID)
396 seq_printf(m, ",fsid=%pU", &opt->fsid);
397 if (opt->flags & CEPH_OPT_NOSHARE)
398 seq_puts(m, ",noshare");
399 if (opt->flags & CEPH_OPT_NOCRC)
400 seq_puts(m, ",nocrc");
401
402 if (opt->name)
403 seq_printf(m, ",name=%s", opt->name);
404 if (opt->key)
405 seq_puts(m, ",secret=<hidden>");
406
407 if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
408 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
409 if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
410 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
411 if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
412 seq_printf(m, ",osdkeepalivetimeout=%d",
413 opt->osd_keepalive_timeout);
414
415 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
416 seq_puts(m, ",dirstat");
417 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
418 seq_puts(m, ",norbytes");
419 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
420 seq_puts(m, ",noasyncreaddir");
421 if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
422 seq_puts(m, ",dcache");
423 else
424 seq_puts(m, ",nodcache");
425
426 if (fsopt->wsize)
427 seq_printf(m, ",wsize=%d", fsopt->wsize);
428 if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
429 seq_printf(m, ",rsize=%d", fsopt->rsize);
430 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
431 seq_printf(m, ",rasize=%d", fsopt->rasize);
432 if (fsopt->congestion_kb != default_congestion_kb())
433 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
434 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
435 seq_printf(m, ",caps_wanted_delay_min=%d",
436 fsopt->caps_wanted_delay_min);
437 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
438 seq_printf(m, ",caps_wanted_delay_max=%d",
439 fsopt->caps_wanted_delay_max);
440 if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
441 seq_printf(m, ",cap_release_safety=%d",
442 fsopt->cap_release_safety);
443 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
444 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
445 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
446 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
447 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
448 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
449 return 0;
450 }
451
452 /*
453 * handle any mon messages the standard library doesn't understand.
454 * return error if we don't either.
455 */
456 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
457 {
458 struct ceph_fs_client *fsc = client->private;
459 int type = le16_to_cpu(msg->hdr.type);
460
461 switch (type) {
462 case CEPH_MSG_MDS_MAP:
463 ceph_mdsc_handle_map(fsc->mdsc, msg);
464 return 0;
465
466 default:
467 return -1;
468 }
469 }
470
471 /*
472 * create a new fs client
473 */
474 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
475 struct ceph_options *opt)
476 {
477 struct ceph_fs_client *fsc;
478 const unsigned supported_features =
479 CEPH_FEATURE_FLOCK |
480 CEPH_FEATURE_DIRLAYOUTHASH;
481 const unsigned required_features = 0;
482 int page_count;
483 size_t size;
484 int err = -ENOMEM;
485
486 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
487 if (!fsc)
488 return ERR_PTR(-ENOMEM);
489
490 fsc->client = ceph_create_client(opt, fsc, supported_features,
491 required_features);
492 if (IS_ERR(fsc->client)) {
493 err = PTR_ERR(fsc->client);
494 goto fail;
495 }
496 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
497 fsc->client->monc.want_mdsmap = 1;
498
499 fsc->mount_options = fsopt;
500
501 fsc->sb = NULL;
502 fsc->mount_state = CEPH_MOUNT_MOUNTING;
503
504 atomic_long_set(&fsc->writeback_count, 0);
505
506 err = bdi_init(&fsc->backing_dev_info);
507 if (err < 0)
508 goto fail_client;
509
510 err = -ENOMEM;
511 /*
512 * The number of concurrent works can be high but they don't need
513 * to be processed in parallel, limit concurrency.
514 */
515 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
516 if (fsc->wb_wq == NULL)
517 goto fail_bdi;
518 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
519 if (fsc->pg_inv_wq == NULL)
520 goto fail_wb_wq;
521 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
522 if (fsc->trunc_wq == NULL)
523 goto fail_pg_inv_wq;
524
525 /* set up mempools */
526 err = -ENOMEM;
527 page_count = fsc->mount_options->wsize >> PAGE_CACHE_SHIFT;
528 size = sizeof (struct page *) * (page_count ? page_count : 1);
529 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
530 if (!fsc->wb_pagevec_pool)
531 goto fail_trunc_wq;
532
533 /* caps */
534 fsc->min_caps = fsopt->max_readdir;
535
536 return fsc;
537
538 fail_trunc_wq:
539 destroy_workqueue(fsc->trunc_wq);
540 fail_pg_inv_wq:
541 destroy_workqueue(fsc->pg_inv_wq);
542 fail_wb_wq:
543 destroy_workqueue(fsc->wb_wq);
544 fail_bdi:
545 bdi_destroy(&fsc->backing_dev_info);
546 fail_client:
547 ceph_destroy_client(fsc->client);
548 fail:
549 kfree(fsc);
550 return ERR_PTR(err);
551 }
552
553 static void destroy_fs_client(struct ceph_fs_client *fsc)
554 {
555 dout("destroy_fs_client %p\n", fsc);
556
557 destroy_workqueue(fsc->wb_wq);
558 destroy_workqueue(fsc->pg_inv_wq);
559 destroy_workqueue(fsc->trunc_wq);
560
561 bdi_destroy(&fsc->backing_dev_info);
562
563 mempool_destroy(fsc->wb_pagevec_pool);
564
565 destroy_mount_options(fsc->mount_options);
566
567 ceph_fs_debugfs_cleanup(fsc);
568
569 ceph_destroy_client(fsc->client);
570
571 kfree(fsc);
572 dout("destroy_fs_client %p done\n", fsc);
573 }
574
575 /*
576 * caches
577 */
578 struct kmem_cache *ceph_inode_cachep;
579 struct kmem_cache *ceph_cap_cachep;
580 struct kmem_cache *ceph_dentry_cachep;
581 struct kmem_cache *ceph_file_cachep;
582
583 static void ceph_inode_init_once(void *foo)
584 {
585 struct ceph_inode_info *ci = foo;
586 inode_init_once(&ci->vfs_inode);
587 }
588
589 static int __init init_caches(void)
590 {
591 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
592 sizeof(struct ceph_inode_info),
593 __alignof__(struct ceph_inode_info),
594 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
595 ceph_inode_init_once);
596 if (ceph_inode_cachep == NULL)
597 return -ENOMEM;
598
599 ceph_cap_cachep = KMEM_CACHE(ceph_cap,
600 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
601 if (ceph_cap_cachep == NULL)
602 goto bad_cap;
603
604 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
605 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
606 if (ceph_dentry_cachep == NULL)
607 goto bad_dentry;
608
609 ceph_file_cachep = KMEM_CACHE(ceph_file_info,
610 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
611 if (ceph_file_cachep == NULL)
612 goto bad_file;
613
614 return 0;
615
616 bad_file:
617 kmem_cache_destroy(ceph_dentry_cachep);
618 bad_dentry:
619 kmem_cache_destroy(ceph_cap_cachep);
620 bad_cap:
621 kmem_cache_destroy(ceph_inode_cachep);
622 return -ENOMEM;
623 }
624
625 static void destroy_caches(void)
626 {
627 /*
628 * Make sure all delayed rcu free inodes are flushed before we
629 * destroy cache.
630 */
631 rcu_barrier();
632 kmem_cache_destroy(ceph_inode_cachep);
633 kmem_cache_destroy(ceph_cap_cachep);
634 kmem_cache_destroy(ceph_dentry_cachep);
635 kmem_cache_destroy(ceph_file_cachep);
636 }
637
638
639 /*
640 * ceph_umount_begin - initiate forced umount. Tear down down the
641 * mount, skipping steps that may hang while waiting for server(s).
642 */
643 static void ceph_umount_begin(struct super_block *sb)
644 {
645 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
646
647 dout("ceph_umount_begin - starting forced umount\n");
648 if (!fsc)
649 return;
650 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
651 return;
652 }
653
654 static const struct super_operations ceph_super_ops = {
655 .alloc_inode = ceph_alloc_inode,
656 .destroy_inode = ceph_destroy_inode,
657 .write_inode = ceph_write_inode,
658 .sync_fs = ceph_sync_fs,
659 .put_super = ceph_put_super,
660 .show_options = ceph_show_options,
661 .statfs = ceph_statfs,
662 .umount_begin = ceph_umount_begin,
663 };
664
665 /*
666 * Bootstrap mount by opening the root directory. Note the mount
667 * @started time from caller, and time out if this takes too long.
668 */
669 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
670 const char *path,
671 unsigned long started)
672 {
673 struct ceph_mds_client *mdsc = fsc->mdsc;
674 struct ceph_mds_request *req = NULL;
675 int err;
676 struct dentry *root;
677
678 /* open dir */
679 dout("open_root_inode opening '%s'\n", path);
680 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
681 if (IS_ERR(req))
682 return ERR_CAST(req);
683 req->r_path1 = kstrdup(path, GFP_NOFS);
684 req->r_ino1.ino = CEPH_INO_ROOT;
685 req->r_ino1.snap = CEPH_NOSNAP;
686 req->r_started = started;
687 req->r_timeout = fsc->client->options->mount_timeout * HZ;
688 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
689 req->r_num_caps = 2;
690 err = ceph_mdsc_do_request(mdsc, NULL, req);
691 if (err == 0) {
692 struct inode *inode = req->r_target_inode;
693 req->r_target_inode = NULL;
694 dout("open_root_inode success\n");
695 if (ceph_ino(inode) == CEPH_INO_ROOT &&
696 fsc->sb->s_root == NULL) {
697 root = d_make_root(inode);
698 if (!root) {
699 root = ERR_PTR(-ENOMEM);
700 goto out;
701 }
702 } else {
703 root = d_obtain_alias(inode);
704 }
705 ceph_init_dentry(root);
706 dout("open_root_inode success, root dentry is %p\n", root);
707 } else {
708 root = ERR_PTR(err);
709 }
710 out:
711 ceph_mdsc_put_request(req);
712 return root;
713 }
714
715
716
717
718 /*
719 * mount: join the ceph cluster, and open root directory.
720 */
721 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
722 const char *path)
723 {
724 int err;
725 unsigned long started = jiffies; /* note the start time */
726 struct dentry *root;
727 int first = 0; /* first vfsmount for this super_block */
728
729 dout("mount start\n");
730 mutex_lock(&fsc->client->mount_mutex);
731
732 err = __ceph_open_session(fsc->client, started);
733 if (err < 0)
734 goto out;
735
736 dout("mount opening root\n");
737 root = open_root_dentry(fsc, "", started);
738 if (IS_ERR(root)) {
739 err = PTR_ERR(root);
740 goto out;
741 }
742 if (fsc->sb->s_root) {
743 dput(root);
744 } else {
745 fsc->sb->s_root = root;
746 first = 1;
747
748 err = ceph_fs_debugfs_init(fsc);
749 if (err < 0)
750 goto fail;
751 }
752
753 if (path[0] == 0) {
754 dget(root);
755 } else {
756 dout("mount opening base mountpoint\n");
757 root = open_root_dentry(fsc, path, started);
758 if (IS_ERR(root)) {
759 err = PTR_ERR(root);
760 goto fail;
761 }
762 }
763
764 fsc->mount_state = CEPH_MOUNT_MOUNTED;
765 dout("mount success\n");
766 mutex_unlock(&fsc->client->mount_mutex);
767 return root;
768
769 out:
770 mutex_unlock(&fsc->client->mount_mutex);
771 return ERR_PTR(err);
772
773 fail:
774 if (first) {
775 dput(fsc->sb->s_root);
776 fsc->sb->s_root = NULL;
777 }
778 goto out;
779 }
780
781 static int ceph_set_super(struct super_block *s, void *data)
782 {
783 struct ceph_fs_client *fsc = data;
784 int ret;
785
786 dout("set_super %p data %p\n", s, data);
787
788 s->s_flags = fsc->mount_options->sb_flags;
789 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
790
791 s->s_fs_info = fsc;
792 fsc->sb = s;
793
794 s->s_op = &ceph_super_ops;
795 s->s_export_op = &ceph_export_ops;
796
797 s->s_time_gran = 1000; /* 1000 ns == 1 us */
798
799 ret = set_anon_super(s, NULL); /* what is that second arg for? */
800 if (ret != 0)
801 goto fail;
802
803 return ret;
804
805 fail:
806 s->s_fs_info = NULL;
807 fsc->sb = NULL;
808 return ret;
809 }
810
811 /*
812 * share superblock if same fs AND options
813 */
814 static int ceph_compare_super(struct super_block *sb, void *data)
815 {
816 struct ceph_fs_client *new = data;
817 struct ceph_mount_options *fsopt = new->mount_options;
818 struct ceph_options *opt = new->client->options;
819 struct ceph_fs_client *other = ceph_sb_to_client(sb);
820
821 dout("ceph_compare_super %p\n", sb);
822
823 if (compare_mount_options(fsopt, opt, other)) {
824 dout("monitor(s)/mount options don't match\n");
825 return 0;
826 }
827 if ((opt->flags & CEPH_OPT_FSID) &&
828 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
829 dout("fsid doesn't match\n");
830 return 0;
831 }
832 if (fsopt->sb_flags != other->mount_options->sb_flags) {
833 dout("flags differ\n");
834 return 0;
835 }
836 return 1;
837 }
838
839 /*
840 * construct our own bdi so we can control readahead, etc.
841 */
842 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
843
844 static int ceph_register_bdi(struct super_block *sb,
845 struct ceph_fs_client *fsc)
846 {
847 int err;
848
849 /* set ra_pages based on rasize mount option? */
850 if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
851 fsc->backing_dev_info.ra_pages =
852 (fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
853 >> PAGE_SHIFT;
854 else
855 fsc->backing_dev_info.ra_pages =
856 default_backing_dev_info.ra_pages;
857
858 err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
859 atomic_long_inc_return(&bdi_seq));
860 if (!err)
861 sb->s_bdi = &fsc->backing_dev_info;
862 return err;
863 }
864
865 static struct dentry *ceph_mount(struct file_system_type *fs_type,
866 int flags, const char *dev_name, void *data)
867 {
868 struct super_block *sb;
869 struct ceph_fs_client *fsc;
870 struct dentry *res;
871 int err;
872 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
873 const char *path = NULL;
874 struct ceph_mount_options *fsopt = NULL;
875 struct ceph_options *opt = NULL;
876
877 dout("ceph_mount\n");
878 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
879 if (err < 0) {
880 res = ERR_PTR(err);
881 goto out_final;
882 }
883
884 /* create client (which we may/may not use) */
885 fsc = create_fs_client(fsopt, opt);
886 if (IS_ERR(fsc)) {
887 res = ERR_CAST(fsc);
888 destroy_mount_options(fsopt);
889 ceph_destroy_options(opt);
890 goto out_final;
891 }
892
893 err = ceph_mdsc_init(fsc);
894 if (err < 0) {
895 res = ERR_PTR(err);
896 goto out;
897 }
898
899 if (ceph_test_opt(fsc->client, NOSHARE))
900 compare_super = NULL;
901 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
902 if (IS_ERR(sb)) {
903 res = ERR_CAST(sb);
904 goto out;
905 }
906
907 if (ceph_sb_to_client(sb) != fsc) {
908 ceph_mdsc_destroy(fsc);
909 destroy_fs_client(fsc);
910 fsc = ceph_sb_to_client(sb);
911 dout("get_sb got existing client %p\n", fsc);
912 } else {
913 dout("get_sb using new client %p\n", fsc);
914 err = ceph_register_bdi(sb, fsc);
915 if (err < 0) {
916 res = ERR_PTR(err);
917 goto out_splat;
918 }
919 }
920
921 res = ceph_real_mount(fsc, path);
922 if (IS_ERR(res))
923 goto out_splat;
924 dout("root %p inode %p ino %llx.%llx\n", res,
925 res->d_inode, ceph_vinop(res->d_inode));
926 return res;
927
928 out_splat:
929 ceph_mdsc_close_sessions(fsc->mdsc);
930 deactivate_locked_super(sb);
931 goto out_final;
932
933 out:
934 ceph_mdsc_destroy(fsc);
935 destroy_fs_client(fsc);
936 out_final:
937 dout("ceph_mount fail %ld\n", PTR_ERR(res));
938 return res;
939 }
940
941 static void ceph_kill_sb(struct super_block *s)
942 {
943 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
944 dout("kill_sb %p\n", s);
945 ceph_mdsc_pre_umount(fsc->mdsc);
946 kill_anon_super(s); /* will call put_super after sb is r/o */
947 ceph_mdsc_destroy(fsc);
948 destroy_fs_client(fsc);
949 }
950
951 static struct file_system_type ceph_fs_type = {
952 .owner = THIS_MODULE,
953 .name = "ceph",
954 .mount = ceph_mount,
955 .kill_sb = ceph_kill_sb,
956 .fs_flags = FS_RENAME_DOES_D_MOVE,
957 };
958 MODULE_ALIAS_FS("ceph");
959
960 #define _STRINGIFY(x) #x
961 #define STRINGIFY(x) _STRINGIFY(x)
962
963 static int __init init_ceph(void)
964 {
965 int ret = init_caches();
966 if (ret)
967 goto out;
968
969 ceph_xattr_init();
970 ret = register_filesystem(&ceph_fs_type);
971 if (ret)
972 goto out_icache;
973
974 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
975
976 return 0;
977
978 out_icache:
979 ceph_xattr_exit();
980 destroy_caches();
981 out:
982 return ret;
983 }
984
985 static void __exit exit_ceph(void)
986 {
987 dout("exit_ceph\n");
988 unregister_filesystem(&ceph_fs_type);
989 ceph_xattr_exit();
990 destroy_caches();
991 }
992
993 module_init(init_ceph);
994 module_exit(exit_ceph);
995
996 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
997 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
998 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
999 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1000 MODULE_LICENSE("GPL");
Linus' kernel tree