ceph: move readahead default to fs/ceph from libceph
[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) - le64_to_cpu(st.kb_used)) >>
77                 (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_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_ino32,
135 };
136
137 static match_table_t fsopt_tokens = {
138         {Opt_wsize, "wsize=%d"},
139         {Opt_rsize, "rsize=%d"},
140         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
141         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
142         {Opt_cap_release_safety, "cap_release_safety=%d"},
143         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
144         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
145         {Opt_congestion_kb, "write_congestion_kb=%d"},
146         /* int args above */
147         {Opt_snapdirname, "snapdirname=%s"},
148         /* string args above */
149         {Opt_dirstat, "dirstat"},
150         {Opt_nodirstat, "nodirstat"},
151         {Opt_rbytes, "rbytes"},
152         {Opt_norbytes, "norbytes"},
153         {Opt_noasyncreaddir, "noasyncreaddir"},
154         {Opt_ino32, "ino32"},
155         {-1, NULL}
156 };
157
158 static int parse_fsopt_token(char *c, void *private)
159 {
160         struct ceph_mount_options *fsopt = private;
161         substring_t argstr[MAX_OPT_ARGS];
162         int token, intval, ret;
163
164         token = match_token((char *)c, fsopt_tokens, argstr);
165         if (token < 0)
166                 return -EINVAL;
167
168         if (token < Opt_last_int) {
169                 ret = match_int(&argstr[0], &intval);
170                 if (ret < 0) {
171                         pr_err("bad mount option arg (not int) "
172                                "at '%s'\n", c);
173                         return ret;
174                 }
175                 dout("got int token %d val %d\n", token, intval);
176         } else if (token > Opt_last_int && token < Opt_last_string) {
177                 dout("got string token %d val %s\n", token,
178                      argstr[0].from);
179         } else {
180                 dout("got token %d\n", token);
181         }
182
183         switch (token) {
184         case Opt_snapdirname:
185                 kfree(fsopt->snapdir_name);
186                 fsopt->snapdir_name = kstrndup(argstr[0].from,
187                                                argstr[0].to-argstr[0].from,
188                                                GFP_KERNEL);
189                 if (!fsopt->snapdir_name)
190                         return -ENOMEM;
191                 break;
192
193                 /* misc */
194         case Opt_wsize:
195                 fsopt->wsize = intval;
196                 break;
197         case Opt_rsize:
198                 fsopt->rsize = intval;
199                 break;
200         case Opt_caps_wanted_delay_min:
201                 fsopt->caps_wanted_delay_min = intval;
202                 break;
203         case Opt_caps_wanted_delay_max:
204                 fsopt->caps_wanted_delay_max = intval;
205                 break;
206         case Opt_readdir_max_entries:
207                 fsopt->max_readdir = intval;
208                 break;
209         case Opt_readdir_max_bytes:
210                 fsopt->max_readdir_bytes = intval;
211                 break;
212         case Opt_congestion_kb:
213                 fsopt->congestion_kb = intval;
214                 break;
215         case Opt_dirstat:
216                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
217                 break;
218         case Opt_nodirstat:
219                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
220                 break;
221         case Opt_rbytes:
222                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
223                 break;
224         case Opt_norbytes:
225                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
226                 break;
227         case Opt_noasyncreaddir:
228                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
229                 break;
230         case Opt_ino32:
231                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
232                 break;
233         default:
234                 BUG_ON(token);
235         }
236         return 0;
237 }
238
239 static void destroy_mount_options(struct ceph_mount_options *args)
240 {
241         dout("destroy_mount_options %p\n", args);
242         kfree(args->snapdir_name);
243         kfree(args);
244 }
245
246 static int strcmp_null(const char *s1, const char *s2)
247 {
248         if (!s1 && !s2)
249                 return 0;
250         if (s1 && !s2)
251                 return -1;
252         if (!s1 && s2)
253                 return 1;
254         return strcmp(s1, s2);
255 }
256
257 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
258                                  struct ceph_options *new_opt,
259                                  struct ceph_fs_client *fsc)
260 {
261         struct ceph_mount_options *fsopt1 = new_fsopt;
262         struct ceph_mount_options *fsopt2 = fsc->mount_options;
263         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
264         int ret;
265
266         ret = memcmp(fsopt1, fsopt2, ofs);
267         if (ret)
268                 return ret;
269
270         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
271         if (ret)
272                 return ret;
273
274         return ceph_compare_options(new_opt, fsc->client);
275 }
276
277 static int parse_mount_options(struct ceph_mount_options **pfsopt,
278                                struct ceph_options **popt,
279                                int flags, char *options,
280                                const char *dev_name,
281                                const char **path)
282 {
283         struct ceph_mount_options *fsopt;
284         const char *dev_name_end;
285         int err = -ENOMEM;
286
287         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
288         if (!fsopt)
289                 return -ENOMEM;
290
291         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
292
293         fsopt->sb_flags = flags;
294         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
295
296         fsopt->rsize = CEPH_RSIZE_DEFAULT;
297         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
298         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
299         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
300         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
301         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
302         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
303         fsopt->congestion_kb = default_congestion_kb();
304         
305         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
306         err = -EINVAL;
307         if (!dev_name)
308                 goto out;
309         *path = strstr(dev_name, ":/");
310         if (*path == NULL) {
311                 pr_err("device name is missing path (no :/ in %s)\n",
312                        dev_name);
313                 goto out;
314         }
315         dev_name_end = *path;
316         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
317
318         /* path on server */
319         *path += 2;
320         dout("server path '%s'\n", *path);
321
322         err = ceph_parse_options(popt, options, dev_name, dev_name_end,
323                                  parse_fsopt_token, (void *)fsopt);
324         if (err)
325                 goto out;
326
327         /* success */
328         *pfsopt = fsopt;
329         return 0;
330
331 out:
332         destroy_mount_options(fsopt);
333         return err;
334 }
335
336 /**
337  * ceph_show_options - Show mount options in /proc/mounts
338  * @m: seq_file to write to
339  * @mnt: mount descriptor
340  */
341 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
342 {
343         struct ceph_fs_client *fsc = ceph_sb_to_client(mnt->mnt_sb);
344         struct ceph_mount_options *fsopt = fsc->mount_options;
345         struct ceph_options *opt = fsc->client->options;
346
347         if (opt->flags & CEPH_OPT_FSID)
348                 seq_printf(m, ",fsid=%pU", &opt->fsid);
349         if (opt->flags & CEPH_OPT_NOSHARE)
350                 seq_puts(m, ",noshare");
351         if (opt->flags & CEPH_OPT_NOCRC)
352                 seq_puts(m, ",nocrc");
353
354         if (opt->name)
355                 seq_printf(m, ",name=%s", opt->name);
356         if (opt->secret)
357                 seq_puts(m, ",secret=<hidden>");
358
359         if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
360                 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
361         if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
362                 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
363         if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
364                 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
365         if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
366                 seq_printf(m, ",osdkeepalivetimeout=%d",
367                            opt->osd_keepalive_timeout);
368
369         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
370                 seq_puts(m, ",dirstat");
371         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
372                 seq_puts(m, ",norbytes");
373         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
374                 seq_puts(m, ",noasyncreaddir");
375
376         if (fsopt->wsize)
377                 seq_printf(m, ",wsize=%d", fsopt->wsize);
378         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
379                 seq_printf(m, ",rsize=%d", fsopt->rsize);
380         if (fsopt->congestion_kb != default_congestion_kb())
381                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
382         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
383                 seq_printf(m, ",caps_wanted_delay_min=%d",
384                          fsopt->caps_wanted_delay_min);
385         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
386                 seq_printf(m, ",caps_wanted_delay_max=%d",
387                            fsopt->caps_wanted_delay_max);
388         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
389                 seq_printf(m, ",cap_release_safety=%d",
390                            fsopt->cap_release_safety);
391         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
392                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
393         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
394                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
395         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
396                 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
397         return 0;
398 }
399
400 /*
401  * handle any mon messages the standard library doesn't understand.
402  * return error if we don't either.
403  */
404 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
405 {
406         struct ceph_fs_client *fsc = client->private;
407         int type = le16_to_cpu(msg->hdr.type);
408
409         switch (type) {
410         case CEPH_MSG_MDS_MAP:
411                 ceph_mdsc_handle_map(fsc->mdsc, msg);
412                 return 0;
413
414         default:
415                 return -1;
416         }
417 }
418
419 /*
420  * create a new fs client
421  */
422 struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
423                                         struct ceph_options *opt)
424 {
425         struct ceph_fs_client *fsc;
426         int err = -ENOMEM;
427
428         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
429         if (!fsc)
430                 return ERR_PTR(-ENOMEM);
431
432         fsc->client = ceph_create_client(opt, fsc);
433         if (IS_ERR(fsc->client)) {
434                 err = PTR_ERR(fsc->client);
435                 goto fail;
436         }
437         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
438         fsc->client->supported_features |= CEPH_FEATURE_FLOCK |
439                 CEPH_FEATURE_DIRLAYOUTHASH;
440         fsc->client->monc.want_mdsmap = 1;
441
442         fsc->mount_options = fsopt;
443
444         fsc->sb = NULL;
445         fsc->mount_state = CEPH_MOUNT_MOUNTING;
446
447         atomic_long_set(&fsc->writeback_count, 0);
448
449         err = bdi_init(&fsc->backing_dev_info);
450         if (err < 0)
451                 goto fail_client;
452
453         err = -ENOMEM;
454         /*
455          * The number of concurrent works can be high but they don't need
456          * to be processed in parallel, limit concurrency.
457          */
458         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
459         if (fsc->wb_wq == NULL)
460                 goto fail_bdi;
461         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
462         if (fsc->pg_inv_wq == NULL)
463                 goto fail_wb_wq;
464         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
465         if (fsc->trunc_wq == NULL)
466                 goto fail_pg_inv_wq;
467
468         /* set up mempools */
469         err = -ENOMEM;
470         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
471                               fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
472         if (!fsc->wb_pagevec_pool)
473                 goto fail_trunc_wq;
474
475         /* caps */
476         fsc->min_caps = fsopt->max_readdir;
477
478         return fsc;
479
480 fail_trunc_wq:
481         destroy_workqueue(fsc->trunc_wq);
482 fail_pg_inv_wq:
483         destroy_workqueue(fsc->pg_inv_wq);
484 fail_wb_wq:
485         destroy_workqueue(fsc->wb_wq);
486 fail_bdi:
487         bdi_destroy(&fsc->backing_dev_info);
488 fail_client:
489         ceph_destroy_client(fsc->client);
490 fail:
491         kfree(fsc);
492         return ERR_PTR(err);
493 }
494
495 void destroy_fs_client(struct ceph_fs_client *fsc)
496 {
497         dout("destroy_fs_client %p\n", fsc);
498
499         destroy_workqueue(fsc->wb_wq);
500         destroy_workqueue(fsc->pg_inv_wq);
501         destroy_workqueue(fsc->trunc_wq);
502
503         bdi_destroy(&fsc->backing_dev_info);
504
505         mempool_destroy(fsc->wb_pagevec_pool);
506
507         destroy_mount_options(fsc->mount_options);
508
509         ceph_fs_debugfs_cleanup(fsc);
510
511         ceph_destroy_client(fsc->client);
512
513         kfree(fsc);
514         dout("destroy_fs_client %p done\n", fsc);
515 }
516
517 /*
518  * caches
519  */
520 struct kmem_cache *ceph_inode_cachep;
521 struct kmem_cache *ceph_cap_cachep;
522 struct kmem_cache *ceph_dentry_cachep;
523 struct kmem_cache *ceph_file_cachep;
524
525 static void ceph_inode_init_once(void *foo)
526 {
527         struct ceph_inode_info *ci = foo;
528         inode_init_once(&ci->vfs_inode);
529 }
530
531 static int __init init_caches(void)
532 {
533         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
534                                       sizeof(struct ceph_inode_info),
535                                       __alignof__(struct ceph_inode_info),
536                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
537                                       ceph_inode_init_once);
538         if (ceph_inode_cachep == NULL)
539                 return -ENOMEM;
540
541         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
542                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
543         if (ceph_cap_cachep == NULL)
544                 goto bad_cap;
545
546         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
547                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
548         if (ceph_dentry_cachep == NULL)
549                 goto bad_dentry;
550
551         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
552                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
553         if (ceph_file_cachep == NULL)
554                 goto bad_file;
555
556         return 0;
557
558 bad_file:
559         kmem_cache_destroy(ceph_dentry_cachep);
560 bad_dentry:
561         kmem_cache_destroy(ceph_cap_cachep);
562 bad_cap:
563         kmem_cache_destroy(ceph_inode_cachep);
564         return -ENOMEM;
565 }
566
567 static void destroy_caches(void)
568 {
569         kmem_cache_destroy(ceph_inode_cachep);
570         kmem_cache_destroy(ceph_cap_cachep);
571         kmem_cache_destroy(ceph_dentry_cachep);
572         kmem_cache_destroy(ceph_file_cachep);
573 }
574
575
576 /*
577  * ceph_umount_begin - initiate forced umount.  Tear down down the
578  * mount, skipping steps that may hang while waiting for server(s).
579  */
580 static void ceph_umount_begin(struct super_block *sb)
581 {
582         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
583
584         dout("ceph_umount_begin - starting forced umount\n");
585         if (!fsc)
586                 return;
587         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
588         return;
589 }
590
591 static const struct super_operations ceph_super_ops = {
592         .alloc_inode    = ceph_alloc_inode,
593         .destroy_inode  = ceph_destroy_inode,
594         .write_inode    = ceph_write_inode,
595         .sync_fs        = ceph_sync_fs,
596         .put_super      = ceph_put_super,
597         .show_options   = ceph_show_options,
598         .statfs         = ceph_statfs,
599         .umount_begin   = ceph_umount_begin,
600 };
601
602 /*
603  * Bootstrap mount by opening the root directory.  Note the mount
604  * @started time from caller, and time out if this takes too long.
605  */
606 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
607                                        const char *path,
608                                        unsigned long started)
609 {
610         struct ceph_mds_client *mdsc = fsc->mdsc;
611         struct ceph_mds_request *req = NULL;
612         int err;
613         struct dentry *root;
614
615         /* open dir */
616         dout("open_root_inode opening '%s'\n", path);
617         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
618         if (IS_ERR(req))
619                 return ERR_CAST(req);
620         req->r_path1 = kstrdup(path, GFP_NOFS);
621         req->r_ino1.ino = CEPH_INO_ROOT;
622         req->r_ino1.snap = CEPH_NOSNAP;
623         req->r_started = started;
624         req->r_timeout = fsc->client->options->mount_timeout * HZ;
625         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
626         req->r_num_caps = 2;
627         err = ceph_mdsc_do_request(mdsc, NULL, req);
628         if (err == 0) {
629                 dout("open_root_inode success\n");
630                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
631                     fsc->sb->s_root == NULL)
632                         root = d_alloc_root(req->r_target_inode);
633                 else
634                         root = d_obtain_alias(req->r_target_inode);
635                 req->r_target_inode = NULL;
636                 dout("open_root_inode success, root dentry is %p\n", root);
637         } else {
638                 root = ERR_PTR(err);
639         }
640         ceph_mdsc_put_request(req);
641         return root;
642 }
643
644
645
646
647 /*
648  * mount: join the ceph cluster, and open root directory.
649  */
650 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
651                       const char *path)
652 {
653         int err;
654         unsigned long started = jiffies;  /* note the start time */
655         struct dentry *root;
656         int first = 0;   /* first vfsmount for this super_block */
657
658         dout("mount start\n");
659         mutex_lock(&fsc->client->mount_mutex);
660
661         err = __ceph_open_session(fsc->client, started);
662         if (err < 0)
663                 goto out;
664
665         dout("mount opening root\n");
666         root = open_root_dentry(fsc, "", started);
667         if (IS_ERR(root)) {
668                 err = PTR_ERR(root);
669                 goto out;
670         }
671         if (fsc->sb->s_root) {
672                 dput(root);
673         } else {
674                 fsc->sb->s_root = root;
675                 first = 1;
676
677                 err = ceph_fs_debugfs_init(fsc);
678                 if (err < 0)
679                         goto fail;
680         }
681
682         if (path[0] == 0) {
683                 dget(root);
684         } else {
685                 dout("mount opening base mountpoint\n");
686                 root = open_root_dentry(fsc, path, started);
687                 if (IS_ERR(root)) {
688                         err = PTR_ERR(root);
689                         goto fail;
690                 }
691         }
692
693         fsc->mount_state = CEPH_MOUNT_MOUNTED;
694         dout("mount success\n");
695         mutex_unlock(&fsc->client->mount_mutex);
696         return root;
697
698 out:
699         mutex_unlock(&fsc->client->mount_mutex);
700         return ERR_PTR(err);
701
702 fail:
703         if (first) {
704                 dput(fsc->sb->s_root);
705                 fsc->sb->s_root = NULL;
706         }
707         goto out;
708 }
709
710 static int ceph_set_super(struct super_block *s, void *data)
711 {
712         struct ceph_fs_client *fsc = data;
713         int ret;
714
715         dout("set_super %p data %p\n", s, data);
716
717         s->s_flags = fsc->mount_options->sb_flags;
718         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
719
720         s->s_fs_info = fsc;
721         fsc->sb = s;
722
723         s->s_op = &ceph_super_ops;
724         s->s_export_op = &ceph_export_ops;
725
726         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
727
728         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
729         if (ret != 0)
730                 goto fail;
731
732         return ret;
733
734 fail:
735         s->s_fs_info = NULL;
736         fsc->sb = NULL;
737         return ret;
738 }
739
740 /*
741  * share superblock if same fs AND options
742  */
743 static int ceph_compare_super(struct super_block *sb, void *data)
744 {
745         struct ceph_fs_client *new = data;
746         struct ceph_mount_options *fsopt = new->mount_options;
747         struct ceph_options *opt = new->client->options;
748         struct ceph_fs_client *other = ceph_sb_to_client(sb);
749
750         dout("ceph_compare_super %p\n", sb);
751
752         if (compare_mount_options(fsopt, opt, other)) {
753                 dout("monitor(s)/mount options don't match\n");
754                 return 0;
755         }
756         if ((opt->flags & CEPH_OPT_FSID) &&
757             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
758                 dout("fsid doesn't match\n");
759                 return 0;
760         }
761         if (fsopt->sb_flags != other->mount_options->sb_flags) {
762                 dout("flags differ\n");
763                 return 0;
764         }
765         return 1;
766 }
767
768 /*
769  * construct our own bdi so we can control readahead, etc.
770  */
771 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
772
773 static int ceph_register_bdi(struct super_block *sb,
774                              struct ceph_fs_client *fsc)
775 {
776         int err;
777
778         /* set ra_pages based on rsize mount option? */
779         if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
780                 fsc->backing_dev_info.ra_pages =
781                         (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
782                         >> PAGE_SHIFT;
783         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
784                            atomic_long_inc_return(&bdi_seq));
785         if (!err)
786                 sb->s_bdi = &fsc->backing_dev_info;
787         return err;
788 }
789
790 static struct dentry *ceph_mount(struct file_system_type *fs_type,
791                        int flags, const char *dev_name, void *data)
792 {
793         struct super_block *sb;
794         struct ceph_fs_client *fsc;
795         struct dentry *res;
796         int err;
797         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
798         const char *path = NULL;
799         struct ceph_mount_options *fsopt = NULL;
800         struct ceph_options *opt = NULL;
801
802         dout("ceph_mount\n");
803         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
804         if (err < 0) {
805                 res = ERR_PTR(err);
806                 goto out_final;
807         }
808
809         /* create client (which we may/may not use) */
810         fsc = create_fs_client(fsopt, opt);
811         if (IS_ERR(fsc)) {
812                 res = ERR_CAST(fsc);
813                 kfree(fsopt);
814                 kfree(opt);
815                 goto out_final;
816         }
817
818         err = ceph_mdsc_init(fsc);
819         if (err < 0) {
820                 res = ERR_PTR(err);
821                 goto out;
822         }
823
824         if (ceph_test_opt(fsc->client, NOSHARE))
825                 compare_super = NULL;
826         sb = sget(fs_type, compare_super, ceph_set_super, fsc);
827         if (IS_ERR(sb)) {
828                 res = ERR_CAST(sb);
829                 goto out;
830         }
831
832         if (ceph_sb_to_client(sb) != fsc) {
833                 ceph_mdsc_destroy(fsc);
834                 destroy_fs_client(fsc);
835                 fsc = ceph_sb_to_client(sb);
836                 dout("get_sb got existing client %p\n", fsc);
837         } else {
838                 dout("get_sb using new client %p\n", fsc);
839                 err = ceph_register_bdi(sb, fsc);
840                 if (err < 0) {
841                         res = ERR_PTR(err);
842                         goto out_splat;
843                 }
844         }
845
846         res = ceph_real_mount(fsc, path);
847         if (IS_ERR(res))
848                 goto out_splat;
849         dout("root %p inode %p ino %llx.%llx\n", res,
850              res->d_inode, ceph_vinop(res->d_inode));
851         return res;
852
853 out_splat:
854         ceph_mdsc_close_sessions(fsc->mdsc);
855         deactivate_locked_super(sb);
856         goto out_final;
857
858 out:
859         ceph_mdsc_destroy(fsc);
860         destroy_fs_client(fsc);
861 out_final:
862         dout("ceph_mount fail %ld\n", PTR_ERR(res));
863         return res;
864 }
865
866 static void ceph_kill_sb(struct super_block *s)
867 {
868         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
869         dout("kill_sb %p\n", s);
870         ceph_mdsc_pre_umount(fsc->mdsc);
871         kill_anon_super(s);    /* will call put_super after sb is r/o */
872         ceph_mdsc_destroy(fsc);
873         destroy_fs_client(fsc);
874 }
875
876 static struct file_system_type ceph_fs_type = {
877         .owner          = THIS_MODULE,
878         .name           = "ceph",
879         .mount          = ceph_mount,
880         .kill_sb        = ceph_kill_sb,
881         .fs_flags       = FS_RENAME_DOES_D_MOVE,
882 };
883
884 #define _STRINGIFY(x) #x
885 #define STRINGIFY(x) _STRINGIFY(x)
886
887 static int __init init_ceph(void)
888 {
889         int ret = init_caches();
890         if (ret)
891                 goto out;
892
893         ret = register_filesystem(&ceph_fs_type);
894         if (ret)
895                 goto out_icache;
896
897         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
898
899         return 0;
900
901 out_icache:
902         destroy_caches();
903 out:
904         return ret;
905 }
906
907 static void __exit exit_ceph(void)
908 {
909         dout("exit_ceph\n");
910         unregister_filesystem(&ceph_fs_type);
911         destroy_caches();
912 }
913
914 module_init(init_ceph);
915 module_exit(exit_ceph);
916
917 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
918 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
919 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
920 MODULE_DESCRIPTION("Ceph filesystem for Linux");
921 MODULE_LICENSE("GPL");