d4757c98ec1400f9a2328e8e17c1c8e2a8a8c836
[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/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_caps_wanted_delay_min,
118         Opt_caps_wanted_delay_max,
119         Opt_cap_release_safety,
120         Opt_readdir_max_entries,
121         Opt_readdir_max_bytes,
122         Opt_congestion_kb,
123         Opt_last_int,
124         /* int args above */
125         Opt_snapdirname,
126         Opt_last_string,
127         /* string args above */
128         Opt_dirstat,
129         Opt_nodirstat,
130         Opt_rbytes,
131         Opt_norbytes,
132         Opt_noasyncreaddir,
133         Opt_ino32,
134 };
135
136 static match_table_t fsopt_tokens = {
137         {Opt_wsize, "wsize=%d"},
138         {Opt_rsize, "rsize=%d"},
139         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
140         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
141         {Opt_cap_release_safety, "cap_release_safety=%d"},
142         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
143         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
144         {Opt_congestion_kb, "write_congestion_kb=%d"},
145         /* int args above */
146         {Opt_snapdirname, "snapdirname=%s"},
147         /* string args above */
148         {Opt_dirstat, "dirstat"},
149         {Opt_nodirstat, "nodirstat"},
150         {Opt_rbytes, "rbytes"},
151         {Opt_norbytes, "norbytes"},
152         {Opt_noasyncreaddir, "noasyncreaddir"},
153         {Opt_ino32, "ino32"},
154         {-1, NULL}
155 };
156
157 static int parse_fsopt_token(char *c, void *private)
158 {
159         struct ceph_mount_options *fsopt = private;
160         substring_t argstr[MAX_OPT_ARGS];
161         int token, intval, ret;
162
163         token = match_token((char *)c, fsopt_tokens, argstr);
164         if (token < 0)
165                 return -EINVAL;
166
167         if (token < Opt_last_int) {
168                 ret = match_int(&argstr[0], &intval);
169                 if (ret < 0) {
170                         pr_err("bad mount option arg (not int) "
171                                "at '%s'\n", c);
172                         return ret;
173                 }
174                 dout("got int token %d val %d\n", token, intval);
175         } else if (token > Opt_last_int && token < Opt_last_string) {
176                 dout("got string token %d val %s\n", token,
177                      argstr[0].from);
178         } else {
179                 dout("got token %d\n", token);
180         }
181
182         switch (token) {
183         case Opt_snapdirname:
184                 kfree(fsopt->snapdir_name);
185                 fsopt->snapdir_name = kstrndup(argstr[0].from,
186                                                argstr[0].to-argstr[0].from,
187                                                GFP_KERNEL);
188                 if (!fsopt->snapdir_name)
189                         return -ENOMEM;
190                 break;
191
192                 /* misc */
193         case Opt_wsize:
194                 fsopt->wsize = intval;
195                 break;
196         case Opt_rsize:
197                 fsopt->rsize = intval;
198                 break;
199         case Opt_caps_wanted_delay_min:
200                 fsopt->caps_wanted_delay_min = intval;
201                 break;
202         case Opt_caps_wanted_delay_max:
203                 fsopt->caps_wanted_delay_max = intval;
204                 break;
205         case Opt_readdir_max_entries:
206                 fsopt->max_readdir = intval;
207                 break;
208         case Opt_readdir_max_bytes:
209                 fsopt->max_readdir_bytes = intval;
210                 break;
211         case Opt_congestion_kb:
212                 fsopt->congestion_kb = intval;
213                 break;
214         case Opt_dirstat:
215                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
216                 break;
217         case Opt_nodirstat:
218                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
219                 break;
220         case Opt_rbytes:
221                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
222                 break;
223         case Opt_norbytes:
224                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
225                 break;
226         case Opt_noasyncreaddir:
227                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
228                 break;
229         case Opt_ino32:
230                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
231                 break;
232         default:
233                 BUG_ON(token);
234         }
235         return 0;
236 }
237
238 static void destroy_mount_options(struct ceph_mount_options *args)
239 {
240         dout("destroy_mount_options %p\n", args);
241         kfree(args->snapdir_name);
242         kfree(args);
243 }
244
245 static int strcmp_null(const char *s1, const char *s2)
246 {
247         if (!s1 && !s2)
248                 return 0;
249         if (s1 && !s2)
250                 return -1;
251         if (!s1 && s2)
252                 return 1;
253         return strcmp(s1, s2);
254 }
255
256 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
257                                  struct ceph_options *new_opt,
258                                  struct ceph_fs_client *fsc)
259 {
260         struct ceph_mount_options *fsopt1 = new_fsopt;
261         struct ceph_mount_options *fsopt2 = fsc->mount_options;
262         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
263         int ret;
264
265         ret = memcmp(fsopt1, fsopt2, ofs);
266         if (ret)
267                 return ret;
268
269         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
270         if (ret)
271                 return ret;
272
273         return ceph_compare_options(new_opt, fsc->client);
274 }
275
276 static int parse_mount_options(struct ceph_mount_options **pfsopt,
277                                struct ceph_options **popt,
278                                int flags, char *options,
279                                const char *dev_name,
280                                const char **path)
281 {
282         struct ceph_mount_options *fsopt;
283         const char *dev_name_end;
284         int err = -ENOMEM;
285
286         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
287         if (!fsopt)
288                 return -ENOMEM;
289
290         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
291
292         fsopt->sb_flags = flags;
293         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
294
295         fsopt->rsize = CEPH_RSIZE_DEFAULT;
296         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
297         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
298         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
299         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
300         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
301         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
302         fsopt->congestion_kb = default_congestion_kb();
303         
304         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
305         err = -EINVAL;
306         if (!dev_name)
307                 goto out;
308         *path = strstr(dev_name, ":/");
309         if (*path == NULL) {
310                 pr_err("device name is missing path (no :/ in %s)\n",
311                        dev_name);
312                 goto out;
313         }
314         dev_name_end = *path;
315         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
316
317         /* path on server */
318         *path += 2;
319         dout("server path '%s'\n", *path);
320
321         err = ceph_parse_options(popt, options, dev_name, dev_name_end,
322                                  parse_fsopt_token, (void *)fsopt);
323         if (err)
324                 goto out;
325
326         /* success */
327         *pfsopt = fsopt;
328         return 0;
329
330 out:
331         destroy_mount_options(fsopt);
332         return err;
333 }
334
335 /**
336  * ceph_show_options - Show mount options in /proc/mounts
337  * @m: seq_file to write to
338  * @mnt: mount descriptor
339  */
340 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
341 {
342         struct ceph_fs_client *fsc = ceph_sb_to_client(mnt->mnt_sb);
343         struct ceph_mount_options *fsopt = fsc->mount_options;
344         struct ceph_options *opt = fsc->client->options;
345
346         if (opt->flags & CEPH_OPT_FSID)
347                 seq_printf(m, ",fsid=%pU", &opt->fsid);
348         if (opt->flags & CEPH_OPT_NOSHARE)
349                 seq_puts(m, ",noshare");
350         if (opt->flags & CEPH_OPT_NOCRC)
351                 seq_puts(m, ",nocrc");
352
353         if (opt->name)
354                 seq_printf(m, ",name=%s", opt->name);
355         if (opt->key)
356                 seq_puts(m, ",secret=<hidden>");
357
358         if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
359                 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
360         if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
361                 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
362         if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
363                 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
364         if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
365                 seq_printf(m, ",osdkeepalivetimeout=%d",
366                            opt->osd_keepalive_timeout);
367
368         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
369                 seq_puts(m, ",dirstat");
370         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
371                 seq_puts(m, ",norbytes");
372         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
373                 seq_puts(m, ",noasyncreaddir");
374
375         if (fsopt->wsize)
376                 seq_printf(m, ",wsize=%d", fsopt->wsize);
377         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
378                 seq_printf(m, ",rsize=%d", fsopt->rsize);
379         if (fsopt->congestion_kb != default_congestion_kb())
380                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
381         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
382                 seq_printf(m, ",caps_wanted_delay_min=%d",
383                          fsopt->caps_wanted_delay_min);
384         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
385                 seq_printf(m, ",caps_wanted_delay_max=%d",
386                            fsopt->caps_wanted_delay_max);
387         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
388                 seq_printf(m, ",cap_release_safety=%d",
389                            fsopt->cap_release_safety);
390         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
391                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
392         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
393                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
394         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
395                 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
396         return 0;
397 }
398
399 /*
400  * handle any mon messages the standard library doesn't understand.
401  * return error if we don't either.
402  */
403 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
404 {
405         struct ceph_fs_client *fsc = client->private;
406         int type = le16_to_cpu(msg->hdr.type);
407
408         switch (type) {
409         case CEPH_MSG_MDS_MAP:
410                 ceph_mdsc_handle_map(fsc->mdsc, msg);
411                 return 0;
412
413         default:
414                 return -1;
415         }
416 }
417
418 /*
419  * create a new fs client
420  */
421 struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
422                                         struct ceph_options *opt)
423 {
424         struct ceph_fs_client *fsc;
425         int err = -ENOMEM;
426
427         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
428         if (!fsc)
429                 return ERR_PTR(-ENOMEM);
430
431         fsc->client = ceph_create_client(opt, fsc);
432         if (IS_ERR(fsc->client)) {
433                 err = PTR_ERR(fsc->client);
434                 goto fail;
435         }
436         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
437         fsc->client->supported_features |= CEPH_FEATURE_FLOCK |
438                 CEPH_FEATURE_DIRLAYOUTHASH;
439         fsc->client->monc.want_mdsmap = 1;
440
441         fsc->mount_options = fsopt;
442
443         fsc->sb = NULL;
444         fsc->mount_state = CEPH_MOUNT_MOUNTING;
445
446         atomic_long_set(&fsc->writeback_count, 0);
447
448         err = bdi_init(&fsc->backing_dev_info);
449         if (err < 0)
450                 goto fail_client;
451
452         err = -ENOMEM;
453         /*
454          * The number of concurrent works can be high but they don't need
455          * to be processed in parallel, limit concurrency.
456          */
457         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
458         if (fsc->wb_wq == NULL)
459                 goto fail_bdi;
460         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
461         if (fsc->pg_inv_wq == NULL)
462                 goto fail_wb_wq;
463         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
464         if (fsc->trunc_wq == NULL)
465                 goto fail_pg_inv_wq;
466
467         /* set up mempools */
468         err = -ENOMEM;
469         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
470                               fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
471         if (!fsc->wb_pagevec_pool)
472                 goto fail_trunc_wq;
473
474         /* caps */
475         fsc->min_caps = fsopt->max_readdir;
476
477         return fsc;
478
479 fail_trunc_wq:
480         destroy_workqueue(fsc->trunc_wq);
481 fail_pg_inv_wq:
482         destroy_workqueue(fsc->pg_inv_wq);
483 fail_wb_wq:
484         destroy_workqueue(fsc->wb_wq);
485 fail_bdi:
486         bdi_destroy(&fsc->backing_dev_info);
487 fail_client:
488         ceph_destroy_client(fsc->client);
489 fail:
490         kfree(fsc);
491         return ERR_PTR(err);
492 }
493
494 void destroy_fs_client(struct ceph_fs_client *fsc)
495 {
496         dout("destroy_fs_client %p\n", fsc);
497
498         destroy_workqueue(fsc->wb_wq);
499         destroy_workqueue(fsc->pg_inv_wq);
500         destroy_workqueue(fsc->trunc_wq);
501
502         bdi_destroy(&fsc->backing_dev_info);
503
504         mempool_destroy(fsc->wb_pagevec_pool);
505
506         destroy_mount_options(fsc->mount_options);
507
508         ceph_fs_debugfs_cleanup(fsc);
509
510         ceph_destroy_client(fsc->client);
511
512         kfree(fsc);
513         dout("destroy_fs_client %p done\n", fsc);
514 }
515
516 /*
517  * caches
518  */
519 struct kmem_cache *ceph_inode_cachep;
520 struct kmem_cache *ceph_cap_cachep;
521 struct kmem_cache *ceph_dentry_cachep;
522 struct kmem_cache *ceph_file_cachep;
523
524 static void ceph_inode_init_once(void *foo)
525 {
526         struct ceph_inode_info *ci = foo;
527         inode_init_once(&ci->vfs_inode);
528 }
529
530 static int __init init_caches(void)
531 {
532         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
533                                       sizeof(struct ceph_inode_info),
534                                       __alignof__(struct ceph_inode_info),
535                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
536                                       ceph_inode_init_once);
537         if (ceph_inode_cachep == NULL)
538                 return -ENOMEM;
539
540         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
541                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
542         if (ceph_cap_cachep == NULL)
543                 goto bad_cap;
544
545         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
546                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
547         if (ceph_dentry_cachep == NULL)
548                 goto bad_dentry;
549
550         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
551                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
552         if (ceph_file_cachep == NULL)
553                 goto bad_file;
554
555         return 0;
556
557 bad_file:
558         kmem_cache_destroy(ceph_dentry_cachep);
559 bad_dentry:
560         kmem_cache_destroy(ceph_cap_cachep);
561 bad_cap:
562         kmem_cache_destroy(ceph_inode_cachep);
563         return -ENOMEM;
564 }
565
566 static void destroy_caches(void)
567 {
568         kmem_cache_destroy(ceph_inode_cachep);
569         kmem_cache_destroy(ceph_cap_cachep);
570         kmem_cache_destroy(ceph_dentry_cachep);
571         kmem_cache_destroy(ceph_file_cachep);
572 }
573
574
575 /*
576  * ceph_umount_begin - initiate forced umount.  Tear down down the
577  * mount, skipping steps that may hang while waiting for server(s).
578  */
579 static void ceph_umount_begin(struct super_block *sb)
580 {
581         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
582
583         dout("ceph_umount_begin - starting forced umount\n");
584         if (!fsc)
585                 return;
586         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
587         return;
588 }
589
590 static const struct super_operations ceph_super_ops = {
591         .alloc_inode    = ceph_alloc_inode,
592         .destroy_inode  = ceph_destroy_inode,
593         .write_inode    = ceph_write_inode,
594         .sync_fs        = ceph_sync_fs,
595         .put_super      = ceph_put_super,
596         .show_options   = ceph_show_options,
597         .statfs         = ceph_statfs,
598         .umount_begin   = ceph_umount_begin,
599 };
600
601 /*
602  * Bootstrap mount by opening the root directory.  Note the mount
603  * @started time from caller, and time out if this takes too long.
604  */
605 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
606                                        const char *path,
607                                        unsigned long started)
608 {
609         struct ceph_mds_client *mdsc = fsc->mdsc;
610         struct ceph_mds_request *req = NULL;
611         int err;
612         struct dentry *root;
613
614         /* open dir */
615         dout("open_root_inode opening '%s'\n", path);
616         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
617         if (IS_ERR(req))
618                 return ERR_CAST(req);
619         req->r_path1 = kstrdup(path, GFP_NOFS);
620         req->r_ino1.ino = CEPH_INO_ROOT;
621         req->r_ino1.snap = CEPH_NOSNAP;
622         req->r_started = started;
623         req->r_timeout = fsc->client->options->mount_timeout * HZ;
624         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
625         req->r_num_caps = 2;
626         err = ceph_mdsc_do_request(mdsc, NULL, req);
627         if (err == 0) {
628                 dout("open_root_inode success\n");
629                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
630                     fsc->sb->s_root == NULL)
631                         root = d_alloc_root(req->r_target_inode);
632                 else
633                         root = d_obtain_alias(req->r_target_inode);
634                 req->r_target_inode = NULL;
635                 dout("open_root_inode success, root dentry is %p\n", root);
636         } else {
637                 root = ERR_PTR(err);
638         }
639         ceph_mdsc_put_request(req);
640         return root;
641 }
642
643
644
645
646 /*
647  * mount: join the ceph cluster, and open root directory.
648  */
649 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
650                       const char *path)
651 {
652         int err;
653         unsigned long started = jiffies;  /* note the start time */
654         struct dentry *root;
655         int first = 0;   /* first vfsmount for this super_block */
656
657         dout("mount start\n");
658         mutex_lock(&fsc->client->mount_mutex);
659
660         err = __ceph_open_session(fsc->client, started);
661         if (err < 0)
662                 goto out;
663
664         dout("mount opening root\n");
665         root = open_root_dentry(fsc, "", started);
666         if (IS_ERR(root)) {
667                 err = PTR_ERR(root);
668                 goto out;
669         }
670         if (fsc->sb->s_root) {
671                 dput(root);
672         } else {
673                 fsc->sb->s_root = root;
674                 first = 1;
675
676                 err = ceph_fs_debugfs_init(fsc);
677                 if (err < 0)
678                         goto fail;
679         }
680
681         if (path[0] == 0) {
682                 dget(root);
683         } else {
684                 dout("mount opening base mountpoint\n");
685                 root = open_root_dentry(fsc, path, started);
686                 if (IS_ERR(root)) {
687                         err = PTR_ERR(root);
688                         goto fail;
689                 }
690         }
691
692         fsc->mount_state = CEPH_MOUNT_MOUNTED;
693         dout("mount success\n");
694         mutex_unlock(&fsc->client->mount_mutex);
695         return root;
696
697 out:
698         mutex_unlock(&fsc->client->mount_mutex);
699         return ERR_PTR(err);
700
701 fail:
702         if (first) {
703                 dput(fsc->sb->s_root);
704                 fsc->sb->s_root = NULL;
705         }
706         goto out;
707 }
708
709 static int ceph_set_super(struct super_block *s, void *data)
710 {
711         struct ceph_fs_client *fsc = data;
712         int ret;
713
714         dout("set_super %p data %p\n", s, data);
715
716         s->s_flags = fsc->mount_options->sb_flags;
717         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
718
719         s->s_fs_info = fsc;
720         fsc->sb = s;
721
722         s->s_op = &ceph_super_ops;
723         s->s_export_op = &ceph_export_ops;
724
725         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
726
727         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
728         if (ret != 0)
729                 goto fail;
730
731         return ret;
732
733 fail:
734         s->s_fs_info = NULL;
735         fsc->sb = NULL;
736         return ret;
737 }
738
739 /*
740  * share superblock if same fs AND options
741  */
742 static int ceph_compare_super(struct super_block *sb, void *data)
743 {
744         struct ceph_fs_client *new = data;
745         struct ceph_mount_options *fsopt = new->mount_options;
746         struct ceph_options *opt = new->client->options;
747         struct ceph_fs_client *other = ceph_sb_to_client(sb);
748
749         dout("ceph_compare_super %p\n", sb);
750
751         if (compare_mount_options(fsopt, opt, other)) {
752                 dout("monitor(s)/mount options don't match\n");
753                 return 0;
754         }
755         if ((opt->flags & CEPH_OPT_FSID) &&
756             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
757                 dout("fsid doesn't match\n");
758                 return 0;
759         }
760         if (fsopt->sb_flags != other->mount_options->sb_flags) {
761                 dout("flags differ\n");
762                 return 0;
763         }
764         return 1;
765 }
766
767 /*
768  * construct our own bdi so we can control readahead, etc.
769  */
770 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
771
772 static int ceph_register_bdi(struct super_block *sb,
773                              struct ceph_fs_client *fsc)
774 {
775         int err;
776
777         /* set ra_pages based on rsize mount option? */
778         if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
779                 fsc->backing_dev_info.ra_pages =
780                         (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
781                         >> PAGE_SHIFT;
782         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
783                            atomic_long_inc_return(&bdi_seq));
784         if (!err)
785                 sb->s_bdi = &fsc->backing_dev_info;
786         return err;
787 }
788
789 static struct dentry *ceph_mount(struct file_system_type *fs_type,
790                        int flags, const char *dev_name, void *data)
791 {
792         struct super_block *sb;
793         struct ceph_fs_client *fsc;
794         struct dentry *res;
795         int err;
796         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
797         const char *path = NULL;
798         struct ceph_mount_options *fsopt = NULL;
799         struct ceph_options *opt = NULL;
800
801         dout("ceph_mount\n");
802         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
803         if (err < 0) {
804                 res = ERR_PTR(err);
805                 goto out_final;
806         }
807
808         /* create client (which we may/may not use) */
809         fsc = create_fs_client(fsopt, opt);
810         if (IS_ERR(fsc)) {
811                 res = ERR_CAST(fsc);
812                 kfree(fsopt);
813                 kfree(opt);
814                 goto out_final;
815         }
816
817         err = ceph_mdsc_init(fsc);
818         if (err < 0) {
819                 res = ERR_PTR(err);
820                 goto out;
821         }
822
823         if (ceph_test_opt(fsc->client, NOSHARE))
824                 compare_super = NULL;
825         sb = sget(fs_type, compare_super, ceph_set_super, fsc);
826         if (IS_ERR(sb)) {
827                 res = ERR_CAST(sb);
828                 goto out;
829         }
830
831         if (ceph_sb_to_client(sb) != fsc) {
832                 ceph_mdsc_destroy(fsc);
833                 destroy_fs_client(fsc);
834                 fsc = ceph_sb_to_client(sb);
835                 dout("get_sb got existing client %p\n", fsc);
836         } else {
837                 dout("get_sb using new client %p\n", fsc);
838                 err = ceph_register_bdi(sb, fsc);
839                 if (err < 0) {
840                         res = ERR_PTR(err);
841                         goto out_splat;
842                 }
843         }
844
845         res = ceph_real_mount(fsc, path);
846         if (IS_ERR(res))
847                 goto out_splat;
848         dout("root %p inode %p ino %llx.%llx\n", res,
849              res->d_inode, ceph_vinop(res->d_inode));
850         return res;
851
852 out_splat:
853         ceph_mdsc_close_sessions(fsc->mdsc);
854         deactivate_locked_super(sb);
855         goto out_final;
856
857 out:
858         ceph_mdsc_destroy(fsc);
859         destroy_fs_client(fsc);
860 out_final:
861         dout("ceph_mount fail %ld\n", PTR_ERR(res));
862         return res;
863 }
864
865 static void ceph_kill_sb(struct super_block *s)
866 {
867         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
868         dout("kill_sb %p\n", s);
869         ceph_mdsc_pre_umount(fsc->mdsc);
870         kill_anon_super(s);    /* will call put_super after sb is r/o */
871         ceph_mdsc_destroy(fsc);
872         destroy_fs_client(fsc);
873 }
874
875 static struct file_system_type ceph_fs_type = {
876         .owner          = THIS_MODULE,
877         .name           = "ceph",
878         .mount          = ceph_mount,
879         .kill_sb        = ceph_kill_sb,
880         .fs_flags       = FS_RENAME_DOES_D_MOVE,
881 };
882
883 #define _STRINGIFY(x) #x
884 #define STRINGIFY(x) _STRINGIFY(x)
885
886 static int __init init_ceph(void)
887 {
888         int ret = init_caches();
889         if (ret)
890                 goto out;
891
892         ret = register_filesystem(&ceph_fs_type);
893         if (ret)
894                 goto out_icache;
895
896         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
897
898         return 0;
899
900 out_icache:
901         destroy_caches();
902 out:
903         return ret;
904 }
905
906 static void __exit exit_ceph(void)
907 {
908         dout("exit_ceph\n");
909         unregister_filesystem(&ceph_fs_type);
910         destroy_caches();
911 }
912
913 module_init(init_ceph);
914 module_exit(exit_ceph);
915
916 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
917 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
918 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
919 MODULE_DESCRIPTION("Ceph filesystem for Linux");
920 MODULE_LICENSE("GPL");