libceph: remove 'osdtimeout' option
[linux-3.10.git] / fs / ceph / super.c
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         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
76         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
77         buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
78         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
79
80         buf->f_files = le64_to_cpu(st.num_objects);
81         buf->f_ffree = -1;
82         buf->f_namelen = NAME_MAX;
83         buf->f_frsize = PAGE_CACHE_SIZE;
84
85         /* leave fsid little-endian, regardless of host endianness */
86         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
87         buf->f_fsid.val[0] = fsid & 0xffffffff;
88         buf->f_fsid.val[1] = fsid >> 32;
89
90         return 0;
91 }
92
93
94 static int ceph_sync_fs(struct super_block *sb, int wait)
95 {
96         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
97
98         if (!wait) {
99                 dout("sync_fs (non-blocking)\n");
100                 ceph_flush_dirty_caps(fsc->mdsc);
101                 dout("sync_fs (non-blocking) done\n");
102                 return 0;
103         }
104
105         dout("sync_fs (blocking)\n");
106         ceph_osdc_sync(&fsc->client->osdc);
107         ceph_mdsc_sync(fsc->mdsc);
108         dout("sync_fs (blocking) done\n");
109         return 0;
110 }
111
112 /*
113  * mount options
114  */
115 enum {
116         Opt_wsize,
117         Opt_rsize,
118         Opt_rasize,
119         Opt_caps_wanted_delay_min,
120         Opt_caps_wanted_delay_max,
121         Opt_cap_release_safety,
122         Opt_readdir_max_entries,
123         Opt_readdir_max_bytes,
124         Opt_congestion_kb,
125         Opt_last_int,
126         /* int args above */
127         Opt_snapdirname,
128         Opt_last_string,
129         /* string args above */
130         Opt_dirstat,
131         Opt_nodirstat,
132         Opt_rbytes,
133         Opt_norbytes,
134         Opt_asyncreaddir,
135         Opt_noasyncreaddir,
136         Opt_dcache,
137         Opt_nodcache,
138         Opt_ino32,
139         Opt_noino32,
140 };
141
142 static match_table_t fsopt_tokens = {
143         {Opt_wsize, "wsize=%d"},
144         {Opt_rsize, "rsize=%d"},
145         {Opt_rasize, "rasize=%d"},
146         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
147         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
148         {Opt_cap_release_safety, "cap_release_safety=%d"},
149         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
150         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
151         {Opt_congestion_kb, "write_congestion_kb=%d"},
152         /* int args above */
153         {Opt_snapdirname, "snapdirname=%s"},
154         /* string args above */
155         {Opt_dirstat, "dirstat"},
156         {Opt_nodirstat, "nodirstat"},
157         {Opt_rbytes, "rbytes"},
158         {Opt_norbytes, "norbytes"},
159         {Opt_asyncreaddir, "asyncreaddir"},
160         {Opt_noasyncreaddir, "noasyncreaddir"},
161         {Opt_dcache, "dcache"},
162         {Opt_nodcache, "nodcache"},
163         {Opt_ino32, "ino32"},
164         {Opt_noino32, "noino32"},
165         {-1, NULL}
166 };
167
168 static int parse_fsopt_token(char *c, void *private)
169 {
170         struct ceph_mount_options *fsopt = private;
171         substring_t argstr[MAX_OPT_ARGS];
172         int token, intval, ret;
173
174         token = match_token((char *)c, fsopt_tokens, argstr);
175         if (token < 0)
176                 return -EINVAL;
177
178         if (token < Opt_last_int) {
179                 ret = match_int(&argstr[0], &intval);
180                 if (ret < 0) {
181                         pr_err("bad mount option arg (not int) "
182                                "at '%s'\n", c);
183                         return ret;
184                 }
185                 dout("got int token %d val %d\n", token, intval);
186         } else if (token > Opt_last_int && token < Opt_last_string) {
187                 dout("got string token %d val %s\n", token,
188                      argstr[0].from);
189         } else {
190                 dout("got token %d\n", token);
191         }
192
193         switch (token) {
194         case Opt_snapdirname:
195                 kfree(fsopt->snapdir_name);
196                 fsopt->snapdir_name = kstrndup(argstr[0].from,
197                                                argstr[0].to-argstr[0].from,
198                                                GFP_KERNEL);
199                 if (!fsopt->snapdir_name)
200                         return -ENOMEM;
201                 break;
202
203                 /* misc */
204         case Opt_wsize:
205                 fsopt->wsize = intval;
206                 break;
207         case Opt_rsize:
208                 fsopt->rsize = intval;
209                 break;
210         case Opt_rasize:
211                 fsopt->rasize = intval;
212                 break;
213         case Opt_caps_wanted_delay_min:
214                 fsopt->caps_wanted_delay_min = intval;
215                 break;
216         case Opt_caps_wanted_delay_max:
217                 fsopt->caps_wanted_delay_max = intval;
218                 break;
219         case Opt_readdir_max_entries:
220                 fsopt->max_readdir = intval;
221                 break;
222         case Opt_readdir_max_bytes:
223                 fsopt->max_readdir_bytes = intval;
224                 break;
225         case Opt_congestion_kb:
226                 fsopt->congestion_kb = intval;
227                 break;
228         case Opt_dirstat:
229                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
230                 break;
231         case Opt_nodirstat:
232                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
233                 break;
234         case Opt_rbytes:
235                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
236                 break;
237         case Opt_norbytes:
238                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
239                 break;
240         case Opt_asyncreaddir:
241                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
242                 break;
243         case Opt_noasyncreaddir:
244                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
245                 break;
246         case Opt_dcache:
247                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
248                 break;
249         case Opt_nodcache:
250                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
251                 break;
252         case Opt_ino32:
253                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
254                 break;
255         case Opt_noino32:
256                 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
257                 break;
258         default:
259                 BUG_ON(token);
260         }
261         return 0;
262 }
263
264 static void destroy_mount_options(struct ceph_mount_options *args)
265 {
266         dout("destroy_mount_options %p\n", args);
267         kfree(args->snapdir_name);
268         kfree(args);
269 }
270
271 static int strcmp_null(const char *s1, const char *s2)
272 {
273         if (!s1 && !s2)
274                 return 0;
275         if (s1 && !s2)
276                 return -1;
277         if (!s1 && s2)
278                 return 1;
279         return strcmp(s1, s2);
280 }
281
282 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
283                                  struct ceph_options *new_opt,
284                                  struct ceph_fs_client *fsc)
285 {
286         struct ceph_mount_options *fsopt1 = new_fsopt;
287         struct ceph_mount_options *fsopt2 = fsc->mount_options;
288         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
289         int ret;
290
291         ret = memcmp(fsopt1, fsopt2, ofs);
292         if (ret)
293                 return ret;
294
295         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
296         if (ret)
297                 return ret;
298
299         return ceph_compare_options(new_opt, fsc->client);
300 }
301
302 static int parse_mount_options(struct ceph_mount_options **pfsopt,
303                                struct ceph_options **popt,
304                                int flags, char *options,
305                                const char *dev_name,
306                                const char **path)
307 {
308         struct ceph_mount_options *fsopt;
309         const char *dev_name_end;
310         int err;
311
312         if (!dev_name || !*dev_name)
313                 return -EINVAL;
314
315         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
316         if (!fsopt)
317                 return -ENOMEM;
318
319         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
320
321         fsopt->sb_flags = flags;
322         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
323
324         fsopt->rsize = CEPH_RSIZE_DEFAULT;
325         fsopt->rasize = CEPH_RASIZE_DEFAULT;
326         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
327         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
328         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
329         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
330         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
331         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
332         fsopt->congestion_kb = default_congestion_kb();
333
334         /*
335          * Distinguish the server list from the path in "dev_name".
336          * Internally we do not include the leading '/' in the path.
337          *
338          * "dev_name" will look like:
339          *     <server_spec>[,<server_spec>...]:[<path>]
340          * where
341          *     <server_spec> is <ip>[:<port>]
342          *     <path> is optional, but if present must begin with '/'
343          */
344         dev_name_end = strchr(dev_name, '/');
345         if (dev_name_end) {
346                 /* skip over leading '/' for path */
347                 *path = dev_name_end + 1;
348         } else {
349                 /* path is empty */
350                 dev_name_end = dev_name + strlen(dev_name);
351                 *path = dev_name_end;
352         }
353         err = -EINVAL;
354         dev_name_end--;         /* back up to ':' separator */
355         if (*dev_name_end != ':') {
356                 pr_err("device name is missing path (no : separator in %s)\n",
357                                 dev_name);
358                 goto out;
359         }
360         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
361         dout("server path '%s'\n", *path);
362
363         *popt = ceph_parse_options(options, dev_name, dev_name_end,
364                                  parse_fsopt_token, (void *)fsopt);
365         if (IS_ERR(*popt)) {
366                 err = PTR_ERR(*popt);
367                 goto out;
368         }
369
370         /* success */
371         *pfsopt = fsopt;
372         return 0;
373
374 out:
375         destroy_mount_options(fsopt);
376         return err;
377 }
378
379 /**
380  * ceph_show_options - Show mount options in /proc/mounts
381  * @m: seq_file to write to
382  * @root: root of that (sub)tree
383  */
384 static int ceph_show_options(struct seq_file *m, struct dentry *root)
385 {
386         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
387         struct ceph_mount_options *fsopt = fsc->mount_options;
388         struct ceph_options *opt = fsc->client->options;
389
390         if (opt->flags & CEPH_OPT_FSID)
391                 seq_printf(m, ",fsid=%pU", &opt->fsid);
392         if (opt->flags & CEPH_OPT_NOSHARE)
393                 seq_puts(m, ",noshare");
394         if (opt->flags & CEPH_OPT_NOCRC)
395                 seq_puts(m, ",nocrc");
396
397         if (opt->name)
398                 seq_printf(m, ",name=%s", opt->name);
399         if (opt->key)
400                 seq_puts(m, ",secret=<hidden>");
401
402         if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
403                 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
404         if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
405                 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
406         if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
407                 seq_printf(m, ",osdkeepalivetimeout=%d",
408                            opt->osd_keepalive_timeout);
409
410         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
411                 seq_puts(m, ",dirstat");
412         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
413                 seq_puts(m, ",norbytes");
414         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
415                 seq_puts(m, ",noasyncreaddir");
416         if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
417                 seq_puts(m, ",dcache");
418         else
419                 seq_puts(m, ",nodcache");
420
421         if (fsopt->wsize)
422                 seq_printf(m, ",wsize=%d", fsopt->wsize);
423         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
424                 seq_printf(m, ",rsize=%d", fsopt->rsize);
425         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
426                 seq_printf(m, ",rasize=%d", fsopt->rasize);
427         if (fsopt->congestion_kb != default_congestion_kb())
428                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
429         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
430                 seq_printf(m, ",caps_wanted_delay_min=%d",
431                          fsopt->caps_wanted_delay_min);
432         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
433                 seq_printf(m, ",caps_wanted_delay_max=%d",
434                            fsopt->caps_wanted_delay_max);
435         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
436                 seq_printf(m, ",cap_release_safety=%d",
437                            fsopt->cap_release_safety);
438         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
439                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
440         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
441                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
442         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
443                 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
444         return 0;
445 }
446
447 /*
448  * handle any mon messages the standard library doesn't understand.
449  * return error if we don't either.
450  */
451 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
452 {
453         struct ceph_fs_client *fsc = client->private;
454         int type = le16_to_cpu(msg->hdr.type);
455
456         switch (type) {
457         case CEPH_MSG_MDS_MAP:
458                 ceph_mdsc_handle_map(fsc->mdsc, msg);
459                 return 0;
460
461         default:
462                 return -1;
463         }
464 }
465
466 /*
467  * create a new fs client
468  */
469 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
470                                         struct ceph_options *opt)
471 {
472         struct ceph_fs_client *fsc;
473         const unsigned supported_features =
474                 CEPH_FEATURE_FLOCK |
475                 CEPH_FEATURE_DIRLAYOUTHASH;
476         const unsigned required_features = 0;
477         int err = -ENOMEM;
478
479         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
480         if (!fsc)
481                 return ERR_PTR(-ENOMEM);
482
483         fsc->client = ceph_create_client(opt, fsc, supported_features,
484                                          required_features);
485         if (IS_ERR(fsc->client)) {
486                 err = PTR_ERR(fsc->client);
487                 goto fail;
488         }
489         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
490         fsc->client->monc.want_mdsmap = 1;
491
492         fsc->mount_options = fsopt;
493
494         fsc->sb = NULL;
495         fsc->mount_state = CEPH_MOUNT_MOUNTING;
496
497         atomic_long_set(&fsc->writeback_count, 0);
498
499         err = bdi_init(&fsc->backing_dev_info);
500         if (err < 0)
501                 goto fail_client;
502
503         err = -ENOMEM;
504         /*
505          * The number of concurrent works can be high but they don't need
506          * to be processed in parallel, limit concurrency.
507          */
508         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
509         if (fsc->wb_wq == NULL)
510                 goto fail_bdi;
511         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
512         if (fsc->pg_inv_wq == NULL)
513                 goto fail_wb_wq;
514         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
515         if (fsc->trunc_wq == NULL)
516                 goto fail_pg_inv_wq;
517
518         /* set up mempools */
519         err = -ENOMEM;
520         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
521                               fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
522         if (!fsc->wb_pagevec_pool)
523                 goto fail_trunc_wq;
524
525         /* caps */
526         fsc->min_caps = fsopt->max_readdir;
527
528         return fsc;
529
530 fail_trunc_wq:
531         destroy_workqueue(fsc->trunc_wq);
532 fail_pg_inv_wq:
533         destroy_workqueue(fsc->pg_inv_wq);
534 fail_wb_wq:
535         destroy_workqueue(fsc->wb_wq);
536 fail_bdi:
537         bdi_destroy(&fsc->backing_dev_info);
538 fail_client:
539         ceph_destroy_client(fsc->client);
540 fail:
541         kfree(fsc);
542         return ERR_PTR(err);
543 }
544
545 static void destroy_fs_client(struct ceph_fs_client *fsc)
546 {
547         dout("destroy_fs_client %p\n", fsc);
548
549         destroy_workqueue(fsc->wb_wq);
550         destroy_workqueue(fsc->pg_inv_wq);
551         destroy_workqueue(fsc->trunc_wq);
552
553         bdi_destroy(&fsc->backing_dev_info);
554
555         mempool_destroy(fsc->wb_pagevec_pool);
556
557         destroy_mount_options(fsc->mount_options);
558
559         ceph_fs_debugfs_cleanup(fsc);
560
561         ceph_destroy_client(fsc->client);
562
563         kfree(fsc);
564         dout("destroy_fs_client %p done\n", fsc);
565 }
566
567 /*
568  * caches
569  */
570 struct kmem_cache *ceph_inode_cachep;
571 struct kmem_cache *ceph_cap_cachep;
572 struct kmem_cache *ceph_dentry_cachep;
573 struct kmem_cache *ceph_file_cachep;
574
575 static void ceph_inode_init_once(void *foo)
576 {
577         struct ceph_inode_info *ci = foo;
578         inode_init_once(&ci->vfs_inode);
579 }
580
581 static int __init init_caches(void)
582 {
583         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
584                                       sizeof(struct ceph_inode_info),
585                                       __alignof__(struct ceph_inode_info),
586                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
587                                       ceph_inode_init_once);
588         if (ceph_inode_cachep == NULL)
589                 return -ENOMEM;
590
591         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
592                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
593         if (ceph_cap_cachep == NULL)
594                 goto bad_cap;
595
596         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
597                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
598         if (ceph_dentry_cachep == NULL)
599                 goto bad_dentry;
600
601         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
602                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
603         if (ceph_file_cachep == NULL)
604                 goto bad_file;
605
606         return 0;
607
608 bad_file:
609         kmem_cache_destroy(ceph_dentry_cachep);
610 bad_dentry:
611         kmem_cache_destroy(ceph_cap_cachep);
612 bad_cap:
613         kmem_cache_destroy(ceph_inode_cachep);
614         return -ENOMEM;
615 }
616
617 static void destroy_caches(void)
618 {
619         kmem_cache_destroy(ceph_inode_cachep);
620         kmem_cache_destroy(ceph_cap_cachep);
621         kmem_cache_destroy(ceph_dentry_cachep);
622         kmem_cache_destroy(ceph_file_cachep);
623 }
624
625
626 /*
627  * ceph_umount_begin - initiate forced umount.  Tear down down the
628  * mount, skipping steps that may hang while waiting for server(s).
629  */
630 static void ceph_umount_begin(struct super_block *sb)
631 {
632         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
633
634         dout("ceph_umount_begin - starting forced umount\n");
635         if (!fsc)
636                 return;
637         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
638         return;
639 }
640
641 static const struct super_operations ceph_super_ops = {
642         .alloc_inode    = ceph_alloc_inode,
643         .destroy_inode  = ceph_destroy_inode,
644         .write_inode    = ceph_write_inode,
645         .sync_fs        = ceph_sync_fs,
646         .put_super      = ceph_put_super,
647         .show_options   = ceph_show_options,
648         .statfs         = ceph_statfs,
649         .umount_begin   = ceph_umount_begin,
650 };
651
652 /*
653  * Bootstrap mount by opening the root directory.  Note the mount
654  * @started time from caller, and time out if this takes too long.
655  */
656 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
657                                        const char *path,
658                                        unsigned long started)
659 {
660         struct ceph_mds_client *mdsc = fsc->mdsc;
661         struct ceph_mds_request *req = NULL;
662         int err;
663         struct dentry *root;
664
665         /* open dir */
666         dout("open_root_inode opening '%s'\n", path);
667         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
668         if (IS_ERR(req))
669                 return ERR_CAST(req);
670         req->r_path1 = kstrdup(path, GFP_NOFS);
671         req->r_ino1.ino = CEPH_INO_ROOT;
672         req->r_ino1.snap = CEPH_NOSNAP;
673         req->r_started = started;
674         req->r_timeout = fsc->client->options->mount_timeout * HZ;
675         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
676         req->r_num_caps = 2;
677         err = ceph_mdsc_do_request(mdsc, NULL, req);
678         if (err == 0) {
679                 struct inode *inode = req->r_target_inode;
680                 req->r_target_inode = NULL;
681                 dout("open_root_inode success\n");
682                 if (ceph_ino(inode) == CEPH_INO_ROOT &&
683                     fsc->sb->s_root == NULL) {
684                         root = d_make_root(inode);
685                         if (!root) {
686                                 root = ERR_PTR(-ENOMEM);
687                                 goto out;
688                         }
689                 } else {
690                         root = d_obtain_alias(inode);
691                 }
692                 ceph_init_dentry(root);
693                 dout("open_root_inode success, root dentry is %p\n", root);
694         } else {
695                 root = ERR_PTR(err);
696         }
697 out:
698         ceph_mdsc_put_request(req);
699         return root;
700 }
701
702
703
704
705 /*
706  * mount: join the ceph cluster, and open root directory.
707  */
708 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
709                       const char *path)
710 {
711         int err;
712         unsigned long started = jiffies;  /* note the start time */
713         struct dentry *root;
714         int first = 0;   /* first vfsmount for this super_block */
715
716         dout("mount start\n");
717         mutex_lock(&fsc->client->mount_mutex);
718
719         err = __ceph_open_session(fsc->client, started);
720         if (err < 0)
721                 goto out;
722
723         dout("mount opening root\n");
724         root = open_root_dentry(fsc, "", started);
725         if (IS_ERR(root)) {
726                 err = PTR_ERR(root);
727                 goto out;
728         }
729         if (fsc->sb->s_root) {
730                 dput(root);
731         } else {
732                 fsc->sb->s_root = root;
733                 first = 1;
734
735                 err = ceph_fs_debugfs_init(fsc);
736                 if (err < 0)
737                         goto fail;
738         }
739
740         if (path[0] == 0) {
741                 dget(root);
742         } else {
743                 dout("mount opening base mountpoint\n");
744                 root = open_root_dentry(fsc, path, started);
745                 if (IS_ERR(root)) {
746                         err = PTR_ERR(root);
747                         goto fail;
748                 }
749         }
750
751         fsc->mount_state = CEPH_MOUNT_MOUNTED;
752         dout("mount success\n");
753         mutex_unlock(&fsc->client->mount_mutex);
754         return root;
755
756 out:
757         mutex_unlock(&fsc->client->mount_mutex);
758         return ERR_PTR(err);
759
760 fail:
761         if (first) {
762                 dput(fsc->sb->s_root);
763                 fsc->sb->s_root = NULL;
764         }
765         goto out;
766 }
767
768 static int ceph_set_super(struct super_block *s, void *data)
769 {
770         struct ceph_fs_client *fsc = data;
771         int ret;
772
773         dout("set_super %p data %p\n", s, data);
774
775         s->s_flags = fsc->mount_options->sb_flags;
776         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
777
778         s->s_fs_info = fsc;
779         fsc->sb = s;
780
781         s->s_op = &ceph_super_ops;
782         s->s_export_op = &ceph_export_ops;
783
784         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
785
786         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
787         if (ret != 0)
788                 goto fail;
789
790         return ret;
791
792 fail:
793         s->s_fs_info = NULL;
794         fsc->sb = NULL;
795         return ret;
796 }
797
798 /*
799  * share superblock if same fs AND options
800  */
801 static int ceph_compare_super(struct super_block *sb, void *data)
802 {
803         struct ceph_fs_client *new = data;
804         struct ceph_mount_options *fsopt = new->mount_options;
805         struct ceph_options *opt = new->client->options;
806         struct ceph_fs_client *other = ceph_sb_to_client(sb);
807
808         dout("ceph_compare_super %p\n", sb);
809
810         if (compare_mount_options(fsopt, opt, other)) {
811                 dout("monitor(s)/mount options don't match\n");
812                 return 0;
813         }
814         if ((opt->flags & CEPH_OPT_FSID) &&
815             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
816                 dout("fsid doesn't match\n");
817                 return 0;
818         }
819         if (fsopt->sb_flags != other->mount_options->sb_flags) {
820                 dout("flags differ\n");
821                 return 0;
822         }
823         return 1;
824 }
825
826 /*
827  * construct our own bdi so we can control readahead, etc.
828  */
829 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
830
831 static int ceph_register_bdi(struct super_block *sb,
832                              struct ceph_fs_client *fsc)
833 {
834         int err;
835
836         /* set ra_pages based on rasize mount option? */
837         if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
838                 fsc->backing_dev_info.ra_pages =
839                         (fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
840                         >> PAGE_SHIFT;
841         else
842                 fsc->backing_dev_info.ra_pages =
843                         default_backing_dev_info.ra_pages;
844
845         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
846                            atomic_long_inc_return(&bdi_seq));
847         if (!err)
848                 sb->s_bdi = &fsc->backing_dev_info;
849         return err;
850 }
851
852 static struct dentry *ceph_mount(struct file_system_type *fs_type,
853                        int flags, const char *dev_name, void *data)
854 {
855         struct super_block *sb;
856         struct ceph_fs_client *fsc;
857         struct dentry *res;
858         int err;
859         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
860         const char *path = NULL;
861         struct ceph_mount_options *fsopt = NULL;
862         struct ceph_options *opt = NULL;
863
864         dout("ceph_mount\n");
865         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
866         if (err < 0) {
867                 res = ERR_PTR(err);
868                 goto out_final;
869         }
870
871         /* create client (which we may/may not use) */
872         fsc = create_fs_client(fsopt, opt);
873         if (IS_ERR(fsc)) {
874                 res = ERR_CAST(fsc);
875                 destroy_mount_options(fsopt);
876                 ceph_destroy_options(opt);
877                 goto out_final;
878         }
879
880         err = ceph_mdsc_init(fsc);
881         if (err < 0) {
882                 res = ERR_PTR(err);
883                 goto out;
884         }
885
886         if (ceph_test_opt(fsc->client, NOSHARE))
887                 compare_super = NULL;
888         sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
889         if (IS_ERR(sb)) {
890                 res = ERR_CAST(sb);
891                 goto out;
892         }
893
894         if (ceph_sb_to_client(sb) != fsc) {
895                 ceph_mdsc_destroy(fsc);
896                 destroy_fs_client(fsc);
897                 fsc = ceph_sb_to_client(sb);
898                 dout("get_sb got existing client %p\n", fsc);
899         } else {
900                 dout("get_sb using new client %p\n", fsc);
901                 err = ceph_register_bdi(sb, fsc);
902                 if (err < 0) {
903                         res = ERR_PTR(err);
904                         goto out_splat;
905                 }
906         }
907
908         res = ceph_real_mount(fsc, path);
909         if (IS_ERR(res))
910                 goto out_splat;
911         dout("root %p inode %p ino %llx.%llx\n", res,
912              res->d_inode, ceph_vinop(res->d_inode));
913         return res;
914
915 out_splat:
916         ceph_mdsc_close_sessions(fsc->mdsc);
917         deactivate_locked_super(sb);
918         goto out_final;
919
920 out:
921         ceph_mdsc_destroy(fsc);
922         destroy_fs_client(fsc);
923 out_final:
924         dout("ceph_mount fail %ld\n", PTR_ERR(res));
925         return res;
926 }
927
928 static void ceph_kill_sb(struct super_block *s)
929 {
930         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
931         dout("kill_sb %p\n", s);
932         ceph_mdsc_pre_umount(fsc->mdsc);
933         kill_anon_super(s);    /* will call put_super after sb is r/o */
934         ceph_mdsc_destroy(fsc);
935         destroy_fs_client(fsc);
936 }
937
938 static struct file_system_type ceph_fs_type = {
939         .owner          = THIS_MODULE,
940         .name           = "ceph",
941         .mount          = ceph_mount,
942         .kill_sb        = ceph_kill_sb,
943         .fs_flags       = FS_RENAME_DOES_D_MOVE,
944 };
945
946 #define _STRINGIFY(x) #x
947 #define STRINGIFY(x) _STRINGIFY(x)
948
949 static int __init init_ceph(void)
950 {
951         int ret = init_caches();
952         if (ret)
953                 goto out;
954
955         ceph_xattr_init();
956         ret = register_filesystem(&ceph_fs_type);
957         if (ret)
958                 goto out_icache;
959
960         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
961
962         return 0;
963
964 out_icache:
965         ceph_xattr_exit();
966         destroy_caches();
967 out:
968         return ret;
969 }
970
971 static void __exit exit_ceph(void)
972 {
973         dout("exit_ceph\n");
974         unregister_filesystem(&ceph_fs_type);
975         ceph_xattr_exit();
976         destroy_caches();
977 }
978
979 module_init(init_ceph);
980 module_exit(exit_ceph);
981
982 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
983 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
984 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
985 MODULE_DESCRIPTION("Ceph filesystem for Linux");
986 MODULE_LICENSE("GPL");