Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[linux-2.6.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/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  */
35 static 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
54 static 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
93 static 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  */
114 enum {
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_noasyncreaddir,
134         Opt_dcache,
135         Opt_nodcache,
136         Opt_ino32,
137 };
138
139 static match_table_t fsopt_tokens = {
140         {Opt_wsize, "wsize=%d"},
141         {Opt_rsize, "rsize=%d"},
142         {Opt_rasize, "rasize=%d"},
143         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
144         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
145         {Opt_cap_release_safety, "cap_release_safety=%d"},
146         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
147         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
148         {Opt_congestion_kb, "write_congestion_kb=%d"},
149         /* int args above */
150         {Opt_snapdirname, "snapdirname=%s"},
151         /* string args above */
152         {Opt_dirstat, "dirstat"},
153         {Opt_nodirstat, "nodirstat"},
154         {Opt_rbytes, "rbytes"},
155         {Opt_norbytes, "norbytes"},
156         {Opt_noasyncreaddir, "noasyncreaddir"},
157         {Opt_dcache, "dcache"},
158         {Opt_nodcache, "nodcache"},
159         {Opt_ino32, "ino32"},
160         {-1, NULL}
161 };
162
163 static int parse_fsopt_token(char *c, void *private)
164 {
165         struct ceph_mount_options *fsopt = private;
166         substring_t argstr[MAX_OPT_ARGS];
167         int token, intval, ret;
168
169         token = match_token((char *)c, fsopt_tokens, argstr);
170         if (token < 0)
171                 return -EINVAL;
172
173         if (token < Opt_last_int) {
174                 ret = match_int(&argstr[0], &intval);
175                 if (ret < 0) {
176                         pr_err("bad mount option arg (not int) "
177                                "at '%s'\n", c);
178                         return ret;
179                 }
180                 dout("got int token %d val %d\n", token, intval);
181         } else if (token > Opt_last_int && token < Opt_last_string) {
182                 dout("got string token %d val %s\n", token,
183                      argstr[0].from);
184         } else {
185                 dout("got token %d\n", token);
186         }
187
188         switch (token) {
189         case Opt_snapdirname:
190                 kfree(fsopt->snapdir_name);
191                 fsopt->snapdir_name = kstrndup(argstr[0].from,
192                                                argstr[0].to-argstr[0].from,
193                                                GFP_KERNEL);
194                 if (!fsopt->snapdir_name)
195                         return -ENOMEM;
196                 break;
197
198                 /* misc */
199         case Opt_wsize:
200                 fsopt->wsize = intval;
201                 break;
202         case Opt_rsize:
203                 fsopt->rsize = intval;
204                 break;
205         case Opt_rasize:
206                 fsopt->rasize = intval;
207                 break;
208         case Opt_caps_wanted_delay_min:
209                 fsopt->caps_wanted_delay_min = intval;
210                 break;
211         case Opt_caps_wanted_delay_max:
212                 fsopt->caps_wanted_delay_max = intval;
213                 break;
214         case Opt_readdir_max_entries:
215                 fsopt->max_readdir = intval;
216                 break;
217         case Opt_readdir_max_bytes:
218                 fsopt->max_readdir_bytes = intval;
219                 break;
220         case Opt_congestion_kb:
221                 fsopt->congestion_kb = intval;
222                 break;
223         case Opt_dirstat:
224                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
225                 break;
226         case Opt_nodirstat:
227                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
228                 break;
229         case Opt_rbytes:
230                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
231                 break;
232         case Opt_norbytes:
233                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
234                 break;
235         case Opt_noasyncreaddir:
236                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
237                 break;
238         case Opt_dcache:
239                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
240                 break;
241         case Opt_nodcache:
242                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
243                 break;
244         case Opt_ino32:
245                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
246                 break;
247         default:
248                 BUG_ON(token);
249         }
250         return 0;
251 }
252
253 static void destroy_mount_options(struct ceph_mount_options *args)
254 {
255         dout("destroy_mount_options %p\n", args);
256         kfree(args->snapdir_name);
257         kfree(args);
258 }
259
260 static int strcmp_null(const char *s1, const char *s2)
261 {
262         if (!s1 && !s2)
263                 return 0;
264         if (s1 && !s2)
265                 return -1;
266         if (!s1 && s2)
267                 return 1;
268         return strcmp(s1, s2);
269 }
270
271 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
272                                  struct ceph_options *new_opt,
273                                  struct ceph_fs_client *fsc)
274 {
275         struct ceph_mount_options *fsopt1 = new_fsopt;
276         struct ceph_mount_options *fsopt2 = fsc->mount_options;
277         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
278         int ret;
279
280         ret = memcmp(fsopt1, fsopt2, ofs);
281         if (ret)
282                 return ret;
283
284         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
285         if (ret)
286                 return ret;
287
288         return ceph_compare_options(new_opt, fsc->client);
289 }
290
291 static int parse_mount_options(struct ceph_mount_options **pfsopt,
292                                struct ceph_options **popt,
293                                int flags, char *options,
294                                const char *dev_name,
295                                const char **path)
296 {
297         struct ceph_mount_options *fsopt;
298         const char *dev_name_end;
299         int err = -ENOMEM;
300
301         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
302         if (!fsopt)
303                 return -ENOMEM;
304
305         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
306
307         fsopt->sb_flags = flags;
308         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
309
310         fsopt->rsize = CEPH_RSIZE_DEFAULT;
311         fsopt->rasize = CEPH_RASIZE_DEFAULT;
312         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
313         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
314         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
315         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
316         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
317         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
318         fsopt->congestion_kb = default_congestion_kb();
319
320         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
321         err = -EINVAL;
322         if (!dev_name)
323                 goto out;
324         *path = strstr(dev_name, ":/");
325         if (*path == NULL) {
326                 pr_err("device name is missing path (no :/ in %s)\n",
327                                 dev_name);
328                 goto out;
329         }
330         dev_name_end = *path;
331         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
332
333         /* path on server */
334         *path += 2;
335         dout("server path '%s'\n", *path);
336
337         err = ceph_parse_options(popt, options, dev_name, dev_name_end,
338                                  parse_fsopt_token, (void *)fsopt);
339         if (err)
340                 goto out;
341
342         /* success */
343         *pfsopt = fsopt;
344         return 0;
345
346 out:
347         destroy_mount_options(fsopt);
348         return err;
349 }
350
351 /**
352  * ceph_show_options - Show mount options in /proc/mounts
353  * @m: seq_file to write to
354  * @root: root of that (sub)tree
355  */
356 static int ceph_show_options(struct seq_file *m, struct dentry *root)
357 {
358         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
359         struct ceph_mount_options *fsopt = fsc->mount_options;
360         struct ceph_options *opt = fsc->client->options;
361
362         if (opt->flags & CEPH_OPT_FSID)
363                 seq_printf(m, ",fsid=%pU", &opt->fsid);
364         if (opt->flags & CEPH_OPT_NOSHARE)
365                 seq_puts(m, ",noshare");
366         if (opt->flags & CEPH_OPT_NOCRC)
367                 seq_puts(m, ",nocrc");
368
369         if (opt->name)
370                 seq_printf(m, ",name=%s", opt->name);
371         if (opt->key)
372                 seq_puts(m, ",secret=<hidden>");
373
374         if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
375                 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
376         if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
377                 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
378         if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
379                 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
380         if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
381                 seq_printf(m, ",osdkeepalivetimeout=%d",
382                            opt->osd_keepalive_timeout);
383
384         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
385                 seq_puts(m, ",dirstat");
386         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
387                 seq_puts(m, ",norbytes");
388         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
389                 seq_puts(m, ",noasyncreaddir");
390         if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
391                 seq_puts(m, ",dcache");
392         else
393                 seq_puts(m, ",nodcache");
394
395         if (fsopt->wsize)
396                 seq_printf(m, ",wsize=%d", fsopt->wsize);
397         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
398                 seq_printf(m, ",rsize=%d", fsopt->rsize);
399         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
400                 seq_printf(m, ",rasize=%d", fsopt->rasize);
401         if (fsopt->congestion_kb != default_congestion_kb())
402                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
403         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
404                 seq_printf(m, ",caps_wanted_delay_min=%d",
405                          fsopt->caps_wanted_delay_min);
406         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
407                 seq_printf(m, ",caps_wanted_delay_max=%d",
408                            fsopt->caps_wanted_delay_max);
409         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
410                 seq_printf(m, ",cap_release_safety=%d",
411                            fsopt->cap_release_safety);
412         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
413                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
414         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
415                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
416         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
417                 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
418         return 0;
419 }
420
421 /*
422  * handle any mon messages the standard library doesn't understand.
423  * return error if we don't either.
424  */
425 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
426 {
427         struct ceph_fs_client *fsc = client->private;
428         int type = le16_to_cpu(msg->hdr.type);
429
430         switch (type) {
431         case CEPH_MSG_MDS_MAP:
432                 ceph_mdsc_handle_map(fsc->mdsc, msg);
433                 return 0;
434
435         default:
436                 return -1;
437         }
438 }
439
440 /*
441  * create a new fs client
442  */
443 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
444                                         struct ceph_options *opt)
445 {
446         struct ceph_fs_client *fsc;
447         const unsigned supported_features =
448                 CEPH_FEATURE_FLOCK |
449                 CEPH_FEATURE_DIRLAYOUTHASH;
450         const unsigned required_features = 0;
451         int err = -ENOMEM;
452
453         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
454         if (!fsc)
455                 return ERR_PTR(-ENOMEM);
456
457         fsc->client = ceph_create_client(opt, fsc, supported_features,
458                                          required_features);
459         if (IS_ERR(fsc->client)) {
460                 err = PTR_ERR(fsc->client);
461                 goto fail;
462         }
463         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
464         fsc->client->monc.want_mdsmap = 1;
465
466         fsc->mount_options = fsopt;
467
468         fsc->sb = NULL;
469         fsc->mount_state = CEPH_MOUNT_MOUNTING;
470
471         atomic_long_set(&fsc->writeback_count, 0);
472
473         err = bdi_init(&fsc->backing_dev_info);
474         if (err < 0)
475                 goto fail_client;
476
477         err = -ENOMEM;
478         /*
479          * The number of concurrent works can be high but they don't need
480          * to be processed in parallel, limit concurrency.
481          */
482         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
483         if (fsc->wb_wq == NULL)
484                 goto fail_bdi;
485         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
486         if (fsc->pg_inv_wq == NULL)
487                 goto fail_wb_wq;
488         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
489         if (fsc->trunc_wq == NULL)
490                 goto fail_pg_inv_wq;
491
492         /* set up mempools */
493         err = -ENOMEM;
494         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
495                               fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
496         if (!fsc->wb_pagevec_pool)
497                 goto fail_trunc_wq;
498
499         /* caps */
500         fsc->min_caps = fsopt->max_readdir;
501
502         return fsc;
503
504 fail_trunc_wq:
505         destroy_workqueue(fsc->trunc_wq);
506 fail_pg_inv_wq:
507         destroy_workqueue(fsc->pg_inv_wq);
508 fail_wb_wq:
509         destroy_workqueue(fsc->wb_wq);
510 fail_bdi:
511         bdi_destroy(&fsc->backing_dev_info);
512 fail_client:
513         ceph_destroy_client(fsc->client);
514 fail:
515         kfree(fsc);
516         return ERR_PTR(err);
517 }
518
519 static void destroy_fs_client(struct ceph_fs_client *fsc)
520 {
521         dout("destroy_fs_client %p\n", fsc);
522
523         destroy_workqueue(fsc->wb_wq);
524         destroy_workqueue(fsc->pg_inv_wq);
525         destroy_workqueue(fsc->trunc_wq);
526
527         bdi_destroy(&fsc->backing_dev_info);
528
529         mempool_destroy(fsc->wb_pagevec_pool);
530
531         destroy_mount_options(fsc->mount_options);
532
533         ceph_fs_debugfs_cleanup(fsc);
534
535         ceph_destroy_client(fsc->client);
536
537         kfree(fsc);
538         dout("destroy_fs_client %p done\n", fsc);
539 }
540
541 /*
542  * caches
543  */
544 struct kmem_cache *ceph_inode_cachep;
545 struct kmem_cache *ceph_cap_cachep;
546 struct kmem_cache *ceph_dentry_cachep;
547 struct kmem_cache *ceph_file_cachep;
548
549 static void ceph_inode_init_once(void *foo)
550 {
551         struct ceph_inode_info *ci = foo;
552         inode_init_once(&ci->vfs_inode);
553 }
554
555 static int __init init_caches(void)
556 {
557         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
558                                       sizeof(struct ceph_inode_info),
559                                       __alignof__(struct ceph_inode_info),
560                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
561                                       ceph_inode_init_once);
562         if (ceph_inode_cachep == NULL)
563                 return -ENOMEM;
564
565         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
566                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
567         if (ceph_cap_cachep == NULL)
568                 goto bad_cap;
569
570         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
571                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
572         if (ceph_dentry_cachep == NULL)
573                 goto bad_dentry;
574
575         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
576                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
577         if (ceph_file_cachep == NULL)
578                 goto bad_file;
579
580         return 0;
581
582 bad_file:
583         kmem_cache_destroy(ceph_dentry_cachep);
584 bad_dentry:
585         kmem_cache_destroy(ceph_cap_cachep);
586 bad_cap:
587         kmem_cache_destroy(ceph_inode_cachep);
588         return -ENOMEM;
589 }
590
591 static void destroy_caches(void)
592 {
593         kmem_cache_destroy(ceph_inode_cachep);
594         kmem_cache_destroy(ceph_cap_cachep);
595         kmem_cache_destroy(ceph_dentry_cachep);
596         kmem_cache_destroy(ceph_file_cachep);
597 }
598
599
600 /*
601  * ceph_umount_begin - initiate forced umount.  Tear down down the
602  * mount, skipping steps that may hang while waiting for server(s).
603  */
604 static void ceph_umount_begin(struct super_block *sb)
605 {
606         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
607
608         dout("ceph_umount_begin - starting forced umount\n");
609         if (!fsc)
610                 return;
611         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
612         return;
613 }
614
615 static const struct super_operations ceph_super_ops = {
616         .alloc_inode    = ceph_alloc_inode,
617         .destroy_inode  = ceph_destroy_inode,
618         .write_inode    = ceph_write_inode,
619         .sync_fs        = ceph_sync_fs,
620         .put_super      = ceph_put_super,
621         .show_options   = ceph_show_options,
622         .statfs         = ceph_statfs,
623         .umount_begin   = ceph_umount_begin,
624 };
625
626 /*
627  * Bootstrap mount by opening the root directory.  Note the mount
628  * @started time from caller, and time out if this takes too long.
629  */
630 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
631                                        const char *path,
632                                        unsigned long started)
633 {
634         struct ceph_mds_client *mdsc = fsc->mdsc;
635         struct ceph_mds_request *req = NULL;
636         int err;
637         struct dentry *root;
638
639         /* open dir */
640         dout("open_root_inode opening '%s'\n", path);
641         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
642         if (IS_ERR(req))
643                 return ERR_CAST(req);
644         req->r_path1 = kstrdup(path, GFP_NOFS);
645         req->r_ino1.ino = CEPH_INO_ROOT;
646         req->r_ino1.snap = CEPH_NOSNAP;
647         req->r_started = started;
648         req->r_timeout = fsc->client->options->mount_timeout * HZ;
649         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
650         req->r_num_caps = 2;
651         err = ceph_mdsc_do_request(mdsc, NULL, req);
652         if (err == 0) {
653                 struct inode *inode = req->r_target_inode;
654                 req->r_target_inode = NULL;
655                 dout("open_root_inode success\n");
656                 if (ceph_ino(inode) == CEPH_INO_ROOT &&
657                     fsc->sb->s_root == NULL) {
658                         root = d_alloc_root(inode);
659                         if (!root) {
660                                 iput(inode);
661                                 root = ERR_PTR(-ENOMEM);
662                                 goto out;
663                         }
664                 } else {
665                         root = d_obtain_alias(inode);
666                 }
667                 ceph_init_dentry(root);
668                 dout("open_root_inode success, root dentry is %p\n", root);
669         } else {
670                 root = ERR_PTR(err);
671         }
672 out:
673         ceph_mdsc_put_request(req);
674         return root;
675 }
676
677
678
679
680 /*
681  * mount: join the ceph cluster, and open root directory.
682  */
683 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
684                       const char *path)
685 {
686         int err;
687         unsigned long started = jiffies;  /* note the start time */
688         struct dentry *root;
689         int first = 0;   /* first vfsmount for this super_block */
690
691         dout("mount start\n");
692         mutex_lock(&fsc->client->mount_mutex);
693
694         err = __ceph_open_session(fsc->client, started);
695         if (err < 0)
696                 goto out;
697
698         dout("mount opening root\n");
699         root = open_root_dentry(fsc, "", started);
700         if (IS_ERR(root)) {
701                 err = PTR_ERR(root);
702                 goto out;
703         }
704         if (fsc->sb->s_root) {
705                 dput(root);
706         } else {
707                 fsc->sb->s_root = root;
708                 first = 1;
709
710                 err = ceph_fs_debugfs_init(fsc);
711                 if (err < 0)
712                         goto fail;
713         }
714
715         if (path[0] == 0) {
716                 dget(root);
717         } else {
718                 dout("mount opening base mountpoint\n");
719                 root = open_root_dentry(fsc, path, started);
720                 if (IS_ERR(root)) {
721                         err = PTR_ERR(root);
722                         goto fail;
723                 }
724         }
725
726         fsc->mount_state = CEPH_MOUNT_MOUNTED;
727         dout("mount success\n");
728         mutex_unlock(&fsc->client->mount_mutex);
729         return root;
730
731 out:
732         mutex_unlock(&fsc->client->mount_mutex);
733         return ERR_PTR(err);
734
735 fail:
736         if (first) {
737                 dput(fsc->sb->s_root);
738                 fsc->sb->s_root = NULL;
739         }
740         goto out;
741 }
742
743 static int ceph_set_super(struct super_block *s, void *data)
744 {
745         struct ceph_fs_client *fsc = data;
746         int ret;
747
748         dout("set_super %p data %p\n", s, data);
749
750         s->s_flags = fsc->mount_options->sb_flags;
751         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
752
753         s->s_fs_info = fsc;
754         fsc->sb = s;
755
756         s->s_op = &ceph_super_ops;
757         s->s_export_op = &ceph_export_ops;
758
759         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
760
761         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
762         if (ret != 0)
763                 goto fail;
764
765         return ret;
766
767 fail:
768         s->s_fs_info = NULL;
769         fsc->sb = NULL;
770         return ret;
771 }
772
773 /*
774  * share superblock if same fs AND options
775  */
776 static int ceph_compare_super(struct super_block *sb, void *data)
777 {
778         struct ceph_fs_client *new = data;
779         struct ceph_mount_options *fsopt = new->mount_options;
780         struct ceph_options *opt = new->client->options;
781         struct ceph_fs_client *other = ceph_sb_to_client(sb);
782
783         dout("ceph_compare_super %p\n", sb);
784
785         if (compare_mount_options(fsopt, opt, other)) {
786                 dout("monitor(s)/mount options don't match\n");
787                 return 0;
788         }
789         if ((opt->flags & CEPH_OPT_FSID) &&
790             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
791                 dout("fsid doesn't match\n");
792                 return 0;
793         }
794         if (fsopt->sb_flags != other->mount_options->sb_flags) {
795                 dout("flags differ\n");
796                 return 0;
797         }
798         return 1;
799 }
800
801 /*
802  * construct our own bdi so we can control readahead, etc.
803  */
804 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
805
806 static int ceph_register_bdi(struct super_block *sb,
807                              struct ceph_fs_client *fsc)
808 {
809         int err;
810
811         /* set ra_pages based on rasize mount option? */
812         if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
813                 fsc->backing_dev_info.ra_pages =
814                         (fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
815                         >> PAGE_SHIFT;
816         else
817                 fsc->backing_dev_info.ra_pages =
818                         default_backing_dev_info.ra_pages;
819
820         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
821                            atomic_long_inc_return(&bdi_seq));
822         if (!err)
823                 sb->s_bdi = &fsc->backing_dev_info;
824         return err;
825 }
826
827 static struct dentry *ceph_mount(struct file_system_type *fs_type,
828                        int flags, const char *dev_name, void *data)
829 {
830         struct super_block *sb;
831         struct ceph_fs_client *fsc;
832         struct dentry *res;
833         int err;
834         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
835         const char *path = NULL;
836         struct ceph_mount_options *fsopt = NULL;
837         struct ceph_options *opt = NULL;
838
839         dout("ceph_mount\n");
840         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
841         if (err < 0) {
842                 res = ERR_PTR(err);
843                 goto out_final;
844         }
845
846         /* create client (which we may/may not use) */
847         fsc = create_fs_client(fsopt, opt);
848         if (IS_ERR(fsc)) {
849                 res = ERR_CAST(fsc);
850                 destroy_mount_options(fsopt);
851                 ceph_destroy_options(opt);
852                 goto out_final;
853         }
854
855         err = ceph_mdsc_init(fsc);
856         if (err < 0) {
857                 res = ERR_PTR(err);
858                 goto out;
859         }
860
861         if (ceph_test_opt(fsc->client, NOSHARE))
862                 compare_super = NULL;
863         sb = sget(fs_type, compare_super, ceph_set_super, fsc);
864         if (IS_ERR(sb)) {
865                 res = ERR_CAST(sb);
866                 goto out;
867         }
868
869         if (ceph_sb_to_client(sb) != fsc) {
870                 ceph_mdsc_destroy(fsc);
871                 destroy_fs_client(fsc);
872                 fsc = ceph_sb_to_client(sb);
873                 dout("get_sb got existing client %p\n", fsc);
874         } else {
875                 dout("get_sb using new client %p\n", fsc);
876                 err = ceph_register_bdi(sb, fsc);
877                 if (err < 0) {
878                         res = ERR_PTR(err);
879                         goto out_splat;
880                 }
881         }
882
883         res = ceph_real_mount(fsc, path);
884         if (IS_ERR(res))
885                 goto out_splat;
886         dout("root %p inode %p ino %llx.%llx\n", res,
887              res->d_inode, ceph_vinop(res->d_inode));
888         return res;
889
890 out_splat:
891         ceph_mdsc_close_sessions(fsc->mdsc);
892         deactivate_locked_super(sb);
893         goto out_final;
894
895 out:
896         ceph_mdsc_destroy(fsc);
897         destroy_fs_client(fsc);
898 out_final:
899         dout("ceph_mount fail %ld\n", PTR_ERR(res));
900         return res;
901 }
902
903 static void ceph_kill_sb(struct super_block *s)
904 {
905         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
906         dout("kill_sb %p\n", s);
907         ceph_mdsc_pre_umount(fsc->mdsc);
908         kill_anon_super(s);    /* will call put_super after sb is r/o */
909         ceph_mdsc_destroy(fsc);
910         destroy_fs_client(fsc);
911 }
912
913 static struct file_system_type ceph_fs_type = {
914         .owner          = THIS_MODULE,
915         .name           = "ceph",
916         .mount          = ceph_mount,
917         .kill_sb        = ceph_kill_sb,
918         .fs_flags       = FS_RENAME_DOES_D_MOVE,
919 };
920
921 #define _STRINGIFY(x) #x
922 #define STRINGIFY(x) _STRINGIFY(x)
923
924 static int __init init_ceph(void)
925 {
926         int ret = init_caches();
927         if (ret)
928                 goto out;
929
930         ret = register_filesystem(&ceph_fs_type);
931         if (ret)
932                 goto out_icache;
933
934         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
935
936         return 0;
937
938 out_icache:
939         destroy_caches();
940 out:
941         return ret;
942 }
943
944 static void __exit exit_ceph(void)
945 {
946         dout("exit_ceph\n");
947         unregister_filesystem(&ceph_fs_type);
948         destroy_caches();
949 }
950
951 module_init(init_ceph);
952 module_exit(exit_ceph);
953
954 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
955 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
956 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
957 MODULE_DESCRIPTION("Ceph filesystem for Linux");
958 MODULE_LICENSE("GPL");