Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[~shefty/rdma-dev.git] / fs / ceph / super.c
1
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/backing-dev.h>
5 #include <linux/ctype.h>
6 #include <linux/fs.h>
7 #include <linux/inet.h>
8 #include <linux/in6.h>
9 #include <linux/module.h>
10 #include <linux/mount.h>
11 #include <linux/parser.h>
12 #include <linux/sched.h>
13 #include <linux/seq_file.h>
14 #include <linux/slab.h>
15 #include <linux/statfs.h>
16 #include <linux/string.h>
17
18 #include "super.h"
19 #include "mds_client.h"
20
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/decode.h>
23 #include <linux/ceph/mon_client.h>
24 #include <linux/ceph/auth.h>
25 #include <linux/ceph/debugfs.h>
26
27 /*
28  * Ceph superblock operations
29  *
30  * Handle the basics of mounting, unmounting.
31  */
32
33 /*
34  * super ops
35  */
36 static void ceph_put_super(struct super_block *s)
37 {
38         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
39
40         dout("put_super\n");
41         ceph_mdsc_close_sessions(fsc->mdsc);
42
43         /*
44          * ensure we release the bdi before put_anon_super releases
45          * the device name.
46          */
47         if (s->s_bdi == &fsc->backing_dev_info) {
48                 bdi_unregister(&fsc->backing_dev_info);
49                 s->s_bdi = NULL;
50         }
51
52         return;
53 }
54
55 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
56 {
57         struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
58         struct ceph_monmap *monmap = fsc->client->monc.monmap;
59         struct ceph_statfs st;
60         u64 fsid;
61         int err;
62
63         dout("statfs\n");
64         err = ceph_monc_do_statfs(&fsc->client->monc, &st);
65         if (err < 0)
66                 return err;
67
68         /* fill in kstatfs */
69         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
70
71         /*
72          * express utilization in terms of large blocks to avoid
73          * overflow on 32-bit machines.
74          */
75         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
76         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
77         buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
78         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
79
80         buf->f_files = le64_to_cpu(st.num_objects);
81         buf->f_ffree = -1;
82         buf->f_namelen = NAME_MAX;
83         buf->f_frsize = PAGE_CACHE_SIZE;
84
85         /* leave fsid little-endian, regardless of host endianness */
86         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
87         buf->f_fsid.val[0] = fsid & 0xffffffff;
88         buf->f_fsid.val[1] = fsid >> 32;
89
90         return 0;
91 }
92
93
94 static int ceph_sync_fs(struct super_block *sb, int wait)
95 {
96         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
97
98         if (!wait) {
99                 dout("sync_fs (non-blocking)\n");
100                 ceph_flush_dirty_caps(fsc->mdsc);
101                 dout("sync_fs (non-blocking) done\n");
102                 return 0;
103         }
104
105         dout("sync_fs (blocking)\n");
106         ceph_osdc_sync(&fsc->client->osdc);
107         ceph_mdsc_sync(fsc->mdsc);
108         dout("sync_fs (blocking) done\n");
109         return 0;
110 }
111
112 /*
113  * mount options
114  */
115 enum {
116         Opt_wsize,
117         Opt_rsize,
118         Opt_rasize,
119         Opt_caps_wanted_delay_min,
120         Opt_caps_wanted_delay_max,
121         Opt_cap_release_safety,
122         Opt_readdir_max_entries,
123         Opt_readdir_max_bytes,
124         Opt_congestion_kb,
125         Opt_last_int,
126         /* int args above */
127         Opt_snapdirname,
128         Opt_last_string,
129         /* string args above */
130         Opt_dirstat,
131         Opt_nodirstat,
132         Opt_rbytes,
133         Opt_norbytes,
134         Opt_asyncreaddir,
135         Opt_noasyncreaddir,
136         Opt_dcache,
137         Opt_nodcache,
138         Opt_ino32,
139         Opt_noino32,
140 };
141
142 static match_table_t fsopt_tokens = {
143         {Opt_wsize, "wsize=%d"},
144         {Opt_rsize, "rsize=%d"},
145         {Opt_rasize, "rasize=%d"},
146         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
147         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
148         {Opt_cap_release_safety, "cap_release_safety=%d"},
149         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
150         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
151         {Opt_congestion_kb, "write_congestion_kb=%d"},
152         /* int args above */
153         {Opt_snapdirname, "snapdirname=%s"},
154         /* string args above */
155         {Opt_dirstat, "dirstat"},
156         {Opt_nodirstat, "nodirstat"},
157         {Opt_rbytes, "rbytes"},
158         {Opt_norbytes, "norbytes"},
159         {Opt_asyncreaddir, "asyncreaddir"},
160         {Opt_noasyncreaddir, "noasyncreaddir"},
161         {Opt_dcache, "dcache"},
162         {Opt_nodcache, "nodcache"},
163         {Opt_ino32, "ino32"},
164         {Opt_noino32, "noino32"},
165         {-1, NULL}
166 };
167
168 static int parse_fsopt_token(char *c, void *private)
169 {
170         struct ceph_mount_options *fsopt = private;
171         substring_t argstr[MAX_OPT_ARGS];
172         int token, intval, ret;
173
174         token = match_token((char *)c, fsopt_tokens, argstr);
175         if (token < 0)
176                 return -EINVAL;
177
178         if (token < Opt_last_int) {
179                 ret = match_int(&argstr[0], &intval);
180                 if (ret < 0) {
181                         pr_err("bad mount option arg (not int) "
182                                "at '%s'\n", c);
183                         return ret;
184                 }
185                 dout("got int token %d val %d\n", token, intval);
186         } else if (token > Opt_last_int && token < Opt_last_string) {
187                 dout("got string token %d val %s\n", token,
188                      argstr[0].from);
189         } else {
190                 dout("got token %d\n", token);
191         }
192
193         switch (token) {
194         case Opt_snapdirname:
195                 kfree(fsopt->snapdir_name);
196                 fsopt->snapdir_name = kstrndup(argstr[0].from,
197                                                argstr[0].to-argstr[0].from,
198                                                GFP_KERNEL);
199                 if (!fsopt->snapdir_name)
200                         return -ENOMEM;
201                 break;
202
203                 /* misc */
204         case Opt_wsize:
205                 fsopt->wsize = intval;
206                 break;
207         case Opt_rsize:
208                 fsopt->rsize = intval;
209                 break;
210         case Opt_rasize:
211                 fsopt->rasize = intval;
212                 break;
213         case Opt_caps_wanted_delay_min:
214                 fsopt->caps_wanted_delay_min = intval;
215                 break;
216         case Opt_caps_wanted_delay_max:
217                 fsopt->caps_wanted_delay_max = intval;
218                 break;
219         case Opt_readdir_max_entries:
220                 fsopt->max_readdir = intval;
221                 break;
222         case Opt_readdir_max_bytes:
223                 fsopt->max_readdir_bytes = intval;
224                 break;
225         case Opt_congestion_kb:
226                 fsopt->congestion_kb = intval;
227                 break;
228         case Opt_dirstat:
229                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
230                 break;
231         case Opt_nodirstat:
232                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
233                 break;
234         case Opt_rbytes:
235                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
236                 break;
237         case Opt_norbytes:
238                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
239                 break;
240         case Opt_asyncreaddir:
241                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
242                 break;
243         case Opt_noasyncreaddir:
244                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
245                 break;
246         case Opt_dcache:
247                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
248                 break;
249         case Opt_nodcache:
250                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
251                 break;
252         case Opt_ino32:
253                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
254                 break;
255         case Opt_noino32:
256                 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
257                 break;
258         default:
259                 BUG_ON(token);
260         }
261         return 0;
262 }
263
264 static void destroy_mount_options(struct ceph_mount_options *args)
265 {
266         dout("destroy_mount_options %p\n", args);
267         kfree(args->snapdir_name);
268         kfree(args);
269 }
270
271 static int strcmp_null(const char *s1, const char *s2)
272 {
273         if (!s1 && !s2)
274                 return 0;
275         if (s1 && !s2)
276                 return -1;
277         if (!s1 && s2)
278                 return 1;
279         return strcmp(s1, s2);
280 }
281
282 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
283                                  struct ceph_options *new_opt,
284                                  struct ceph_fs_client *fsc)
285 {
286         struct ceph_mount_options *fsopt1 = new_fsopt;
287         struct ceph_mount_options *fsopt2 = fsc->mount_options;
288         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
289         int ret;
290
291         ret = memcmp(fsopt1, fsopt2, ofs);
292         if (ret)
293                 return ret;
294
295         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
296         if (ret)
297                 return ret;
298
299         return ceph_compare_options(new_opt, fsc->client);
300 }
301
302 static int parse_mount_options(struct ceph_mount_options **pfsopt,
303                                struct ceph_options **popt,
304                                int flags, char *options,
305                                const char *dev_name,
306                                const char **path)
307 {
308         struct ceph_mount_options *fsopt;
309         const char *dev_name_end;
310         int err = -ENOMEM;
311
312         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
313         if (!fsopt)
314                 return -ENOMEM;
315
316         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
317
318         fsopt->sb_flags = flags;
319         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
320
321         fsopt->rsize = CEPH_RSIZE_DEFAULT;
322         fsopt->rasize = CEPH_RASIZE_DEFAULT;
323         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
324         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
325         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
326         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
327         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
328         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
329         fsopt->congestion_kb = default_congestion_kb();
330
331         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
332         err = -EINVAL;
333         if (!dev_name)
334                 goto out;
335         *path = strstr(dev_name, ":/");
336         if (*path == NULL) {
337                 pr_err("device name is missing path (no :/ in %s)\n",
338                                 dev_name);
339                 goto out;
340         }
341         dev_name_end = *path;
342         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
343
344         /* path on server */
345         *path += 2;
346         dout("server path '%s'\n", *path);
347
348         *popt = ceph_parse_options(options, dev_name, dev_name_end,
349                                  parse_fsopt_token, (void *)fsopt);
350         if (IS_ERR(*popt)) {
351                 err = PTR_ERR(*popt);
352                 goto out;
353         }
354
355         /* success */
356         *pfsopt = fsopt;
357         return 0;
358
359 out:
360         destroy_mount_options(fsopt);
361         return err;
362 }
363
364 /**
365  * ceph_show_options - Show mount options in /proc/mounts
366  * @m: seq_file to write to
367  * @root: root of that (sub)tree
368  */
369 static int ceph_show_options(struct seq_file *m, struct dentry *root)
370 {
371         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
372         struct ceph_mount_options *fsopt = fsc->mount_options;
373         struct ceph_options *opt = fsc->client->options;
374
375         if (opt->flags & CEPH_OPT_FSID)
376                 seq_printf(m, ",fsid=%pU", &opt->fsid);
377         if (opt->flags & CEPH_OPT_NOSHARE)
378                 seq_puts(m, ",noshare");
379         if (opt->flags & CEPH_OPT_NOCRC)
380                 seq_puts(m, ",nocrc");
381
382         if (opt->name)
383                 seq_printf(m, ",name=%s", opt->name);
384         if (opt->key)
385                 seq_puts(m, ",secret=<hidden>");
386
387         if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
388                 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
389         if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
390                 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
391         if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
392                 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
393         if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
394                 seq_printf(m, ",osdkeepalivetimeout=%d",
395                            opt->osd_keepalive_timeout);
396
397         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
398                 seq_puts(m, ",dirstat");
399         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
400                 seq_puts(m, ",norbytes");
401         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
402                 seq_puts(m, ",noasyncreaddir");
403         if (fsopt->flags & CEPH_MOUNT_OPT_DCACHE)
404                 seq_puts(m, ",dcache");
405         else
406                 seq_puts(m, ",nodcache");
407
408         if (fsopt->wsize)
409                 seq_printf(m, ",wsize=%d", fsopt->wsize);
410         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
411                 seq_printf(m, ",rsize=%d", fsopt->rsize);
412         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
413                 seq_printf(m, ",rasize=%d", fsopt->rasize);
414         if (fsopt->congestion_kb != default_congestion_kb())
415                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
416         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
417                 seq_printf(m, ",caps_wanted_delay_min=%d",
418                          fsopt->caps_wanted_delay_min);
419         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
420                 seq_printf(m, ",caps_wanted_delay_max=%d",
421                            fsopt->caps_wanted_delay_max);
422         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
423                 seq_printf(m, ",cap_release_safety=%d",
424                            fsopt->cap_release_safety);
425         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
426                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
427         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
428                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
429         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
430                 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
431         return 0;
432 }
433
434 /*
435  * handle any mon messages the standard library doesn't understand.
436  * return error if we don't either.
437  */
438 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
439 {
440         struct ceph_fs_client *fsc = client->private;
441         int type = le16_to_cpu(msg->hdr.type);
442
443         switch (type) {
444         case CEPH_MSG_MDS_MAP:
445                 ceph_mdsc_handle_map(fsc->mdsc, msg);
446                 return 0;
447
448         default:
449                 return -1;
450         }
451 }
452
453 /*
454  * create a new fs client
455  */
456 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
457                                         struct ceph_options *opt)
458 {
459         struct ceph_fs_client *fsc;
460         const unsigned supported_features =
461                 CEPH_FEATURE_FLOCK |
462                 CEPH_FEATURE_DIRLAYOUTHASH;
463         const unsigned required_features = 0;
464         int err = -ENOMEM;
465
466         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
467         if (!fsc)
468                 return ERR_PTR(-ENOMEM);
469
470         fsc->client = ceph_create_client(opt, fsc, supported_features,
471                                          required_features);
472         if (IS_ERR(fsc->client)) {
473                 err = PTR_ERR(fsc->client);
474                 goto fail;
475         }
476         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
477         fsc->client->monc.want_mdsmap = 1;
478
479         fsc->mount_options = fsopt;
480
481         fsc->sb = NULL;
482         fsc->mount_state = CEPH_MOUNT_MOUNTING;
483
484         atomic_long_set(&fsc->writeback_count, 0);
485
486         err = bdi_init(&fsc->backing_dev_info);
487         if (err < 0)
488                 goto fail_client;
489
490         err = -ENOMEM;
491         /*
492          * The number of concurrent works can be high but they don't need
493          * to be processed in parallel, limit concurrency.
494          */
495         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
496         if (fsc->wb_wq == NULL)
497                 goto fail_bdi;
498         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
499         if (fsc->pg_inv_wq == NULL)
500                 goto fail_wb_wq;
501         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
502         if (fsc->trunc_wq == NULL)
503                 goto fail_pg_inv_wq;
504
505         /* set up mempools */
506         err = -ENOMEM;
507         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
508                               fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
509         if (!fsc->wb_pagevec_pool)
510                 goto fail_trunc_wq;
511
512         /* caps */
513         fsc->min_caps = fsopt->max_readdir;
514
515         return fsc;
516
517 fail_trunc_wq:
518         destroy_workqueue(fsc->trunc_wq);
519 fail_pg_inv_wq:
520         destroy_workqueue(fsc->pg_inv_wq);
521 fail_wb_wq:
522         destroy_workqueue(fsc->wb_wq);
523 fail_bdi:
524         bdi_destroy(&fsc->backing_dev_info);
525 fail_client:
526         ceph_destroy_client(fsc->client);
527 fail:
528         kfree(fsc);
529         return ERR_PTR(err);
530 }
531
532 static void destroy_fs_client(struct ceph_fs_client *fsc)
533 {
534         dout("destroy_fs_client %p\n", fsc);
535
536         destroy_workqueue(fsc->wb_wq);
537         destroy_workqueue(fsc->pg_inv_wq);
538         destroy_workqueue(fsc->trunc_wq);
539
540         bdi_destroy(&fsc->backing_dev_info);
541
542         mempool_destroy(fsc->wb_pagevec_pool);
543
544         destroy_mount_options(fsc->mount_options);
545
546         ceph_fs_debugfs_cleanup(fsc);
547
548         ceph_destroy_client(fsc->client);
549
550         kfree(fsc);
551         dout("destroy_fs_client %p done\n", fsc);
552 }
553
554 /*
555  * caches
556  */
557 struct kmem_cache *ceph_inode_cachep;
558 struct kmem_cache *ceph_cap_cachep;
559 struct kmem_cache *ceph_dentry_cachep;
560 struct kmem_cache *ceph_file_cachep;
561
562 static void ceph_inode_init_once(void *foo)
563 {
564         struct ceph_inode_info *ci = foo;
565         inode_init_once(&ci->vfs_inode);
566 }
567
568 static int __init init_caches(void)
569 {
570         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
571                                       sizeof(struct ceph_inode_info),
572                                       __alignof__(struct ceph_inode_info),
573                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
574                                       ceph_inode_init_once);
575         if (ceph_inode_cachep == NULL)
576                 return -ENOMEM;
577
578         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
579                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
580         if (ceph_cap_cachep == NULL)
581                 goto bad_cap;
582
583         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
584                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
585         if (ceph_dentry_cachep == NULL)
586                 goto bad_dentry;
587
588         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
589                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
590         if (ceph_file_cachep == NULL)
591                 goto bad_file;
592
593         return 0;
594
595 bad_file:
596         kmem_cache_destroy(ceph_dentry_cachep);
597 bad_dentry:
598         kmem_cache_destroy(ceph_cap_cachep);
599 bad_cap:
600         kmem_cache_destroy(ceph_inode_cachep);
601         return -ENOMEM;
602 }
603
604 static void destroy_caches(void)
605 {
606         kmem_cache_destroy(ceph_inode_cachep);
607         kmem_cache_destroy(ceph_cap_cachep);
608         kmem_cache_destroy(ceph_dentry_cachep);
609         kmem_cache_destroy(ceph_file_cachep);
610 }
611
612
613 /*
614  * ceph_umount_begin - initiate forced umount.  Tear down down the
615  * mount, skipping steps that may hang while waiting for server(s).
616  */
617 static void ceph_umount_begin(struct super_block *sb)
618 {
619         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
620
621         dout("ceph_umount_begin - starting forced umount\n");
622         if (!fsc)
623                 return;
624         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
625         return;
626 }
627
628 static const struct super_operations ceph_super_ops = {
629         .alloc_inode    = ceph_alloc_inode,
630         .destroy_inode  = ceph_destroy_inode,
631         .write_inode    = ceph_write_inode,
632         .sync_fs        = ceph_sync_fs,
633         .put_super      = ceph_put_super,
634         .show_options   = ceph_show_options,
635         .statfs         = ceph_statfs,
636         .umount_begin   = ceph_umount_begin,
637 };
638
639 /*
640  * Bootstrap mount by opening the root directory.  Note the mount
641  * @started time from caller, and time out if this takes too long.
642  */
643 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
644                                        const char *path,
645                                        unsigned long started)
646 {
647         struct ceph_mds_client *mdsc = fsc->mdsc;
648         struct ceph_mds_request *req = NULL;
649         int err;
650         struct dentry *root;
651
652         /* open dir */
653         dout("open_root_inode opening '%s'\n", path);
654         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
655         if (IS_ERR(req))
656                 return ERR_CAST(req);
657         req->r_path1 = kstrdup(path, GFP_NOFS);
658         req->r_ino1.ino = CEPH_INO_ROOT;
659         req->r_ino1.snap = CEPH_NOSNAP;
660         req->r_started = started;
661         req->r_timeout = fsc->client->options->mount_timeout * HZ;
662         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
663         req->r_num_caps = 2;
664         err = ceph_mdsc_do_request(mdsc, NULL, req);
665         if (err == 0) {
666                 struct inode *inode = req->r_target_inode;
667                 req->r_target_inode = NULL;
668                 dout("open_root_inode success\n");
669                 if (ceph_ino(inode) == CEPH_INO_ROOT &&
670                     fsc->sb->s_root == NULL) {
671                         root = d_make_root(inode);
672                         if (!root) {
673                                 root = ERR_PTR(-ENOMEM);
674                                 goto out;
675                         }
676                 } else {
677                         root = d_obtain_alias(inode);
678                 }
679                 ceph_init_dentry(root);
680                 dout("open_root_inode success, root dentry is %p\n", root);
681         } else {
682                 root = ERR_PTR(err);
683         }
684 out:
685         ceph_mdsc_put_request(req);
686         return root;
687 }
688
689
690
691
692 /*
693  * mount: join the ceph cluster, and open root directory.
694  */
695 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
696                       const char *path)
697 {
698         int err;
699         unsigned long started = jiffies;  /* note the start time */
700         struct dentry *root;
701         int first = 0;   /* first vfsmount for this super_block */
702
703         dout("mount start\n");
704         mutex_lock(&fsc->client->mount_mutex);
705
706         err = __ceph_open_session(fsc->client, started);
707         if (err < 0)
708                 goto out;
709
710         dout("mount opening root\n");
711         root = open_root_dentry(fsc, "", started);
712         if (IS_ERR(root)) {
713                 err = PTR_ERR(root);
714                 goto out;
715         }
716         if (fsc->sb->s_root) {
717                 dput(root);
718         } else {
719                 fsc->sb->s_root = root;
720                 first = 1;
721
722                 err = ceph_fs_debugfs_init(fsc);
723                 if (err < 0)
724                         goto fail;
725         }
726
727         if (path[0] == 0) {
728                 dget(root);
729         } else {
730                 dout("mount opening base mountpoint\n");
731                 root = open_root_dentry(fsc, path, started);
732                 if (IS_ERR(root)) {
733                         err = PTR_ERR(root);
734                         goto fail;
735                 }
736         }
737
738         fsc->mount_state = CEPH_MOUNT_MOUNTED;
739         dout("mount success\n");
740         mutex_unlock(&fsc->client->mount_mutex);
741         return root;
742
743 out:
744         mutex_unlock(&fsc->client->mount_mutex);
745         return ERR_PTR(err);
746
747 fail:
748         if (first) {
749                 dput(fsc->sb->s_root);
750                 fsc->sb->s_root = NULL;
751         }
752         goto out;
753 }
754
755 static int ceph_set_super(struct super_block *s, void *data)
756 {
757         struct ceph_fs_client *fsc = data;
758         int ret;
759
760         dout("set_super %p data %p\n", s, data);
761
762         s->s_flags = fsc->mount_options->sb_flags;
763         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
764
765         s->s_fs_info = fsc;
766         fsc->sb = s;
767
768         s->s_op = &ceph_super_ops;
769         s->s_export_op = &ceph_export_ops;
770
771         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
772
773         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
774         if (ret != 0)
775                 goto fail;
776
777         return ret;
778
779 fail:
780         s->s_fs_info = NULL;
781         fsc->sb = NULL;
782         return ret;
783 }
784
785 /*
786  * share superblock if same fs AND options
787  */
788 static int ceph_compare_super(struct super_block *sb, void *data)
789 {
790         struct ceph_fs_client *new = data;
791         struct ceph_mount_options *fsopt = new->mount_options;
792         struct ceph_options *opt = new->client->options;
793         struct ceph_fs_client *other = ceph_sb_to_client(sb);
794
795         dout("ceph_compare_super %p\n", sb);
796
797         if (compare_mount_options(fsopt, opt, other)) {
798                 dout("monitor(s)/mount options don't match\n");
799                 return 0;
800         }
801         if ((opt->flags & CEPH_OPT_FSID) &&
802             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
803                 dout("fsid doesn't match\n");
804                 return 0;
805         }
806         if (fsopt->sb_flags != other->mount_options->sb_flags) {
807                 dout("flags differ\n");
808                 return 0;
809         }
810         return 1;
811 }
812
813 /*
814  * construct our own bdi so we can control readahead, etc.
815  */
816 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
817
818 static int ceph_register_bdi(struct super_block *sb,
819                              struct ceph_fs_client *fsc)
820 {
821         int err;
822
823         /* set ra_pages based on rasize mount option? */
824         if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
825                 fsc->backing_dev_info.ra_pages =
826                         (fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
827                         >> PAGE_SHIFT;
828         else
829                 fsc->backing_dev_info.ra_pages =
830                         default_backing_dev_info.ra_pages;
831
832         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
833                            atomic_long_inc_return(&bdi_seq));
834         if (!err)
835                 sb->s_bdi = &fsc->backing_dev_info;
836         return err;
837 }
838
839 static struct dentry *ceph_mount(struct file_system_type *fs_type,
840                        int flags, const char *dev_name, void *data)
841 {
842         struct super_block *sb;
843         struct ceph_fs_client *fsc;
844         struct dentry *res;
845         int err;
846         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
847         const char *path = NULL;
848         struct ceph_mount_options *fsopt = NULL;
849         struct ceph_options *opt = NULL;
850
851         dout("ceph_mount\n");
852         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
853         if (err < 0) {
854                 res = ERR_PTR(err);
855                 goto out_final;
856         }
857
858         /* create client (which we may/may not use) */
859         fsc = create_fs_client(fsopt, opt);
860         if (IS_ERR(fsc)) {
861                 res = ERR_CAST(fsc);
862                 destroy_mount_options(fsopt);
863                 ceph_destroy_options(opt);
864                 goto out_final;
865         }
866
867         err = ceph_mdsc_init(fsc);
868         if (err < 0) {
869                 res = ERR_PTR(err);
870                 goto out;
871         }
872
873         if (ceph_test_opt(fsc->client, NOSHARE))
874                 compare_super = NULL;
875         sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
876         if (IS_ERR(sb)) {
877                 res = ERR_CAST(sb);
878                 goto out;
879         }
880
881         if (ceph_sb_to_client(sb) != fsc) {
882                 ceph_mdsc_destroy(fsc);
883                 destroy_fs_client(fsc);
884                 fsc = ceph_sb_to_client(sb);
885                 dout("get_sb got existing client %p\n", fsc);
886         } else {
887                 dout("get_sb using new client %p\n", fsc);
888                 err = ceph_register_bdi(sb, fsc);
889                 if (err < 0) {
890                         res = ERR_PTR(err);
891                         goto out_splat;
892                 }
893         }
894
895         res = ceph_real_mount(fsc, path);
896         if (IS_ERR(res))
897                 goto out_splat;
898         dout("root %p inode %p ino %llx.%llx\n", res,
899              res->d_inode, ceph_vinop(res->d_inode));
900         return res;
901
902 out_splat:
903         ceph_mdsc_close_sessions(fsc->mdsc);
904         deactivate_locked_super(sb);
905         goto out_final;
906
907 out:
908         ceph_mdsc_destroy(fsc);
909         destroy_fs_client(fsc);
910 out_final:
911         dout("ceph_mount fail %ld\n", PTR_ERR(res));
912         return res;
913 }
914
915 static void ceph_kill_sb(struct super_block *s)
916 {
917         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
918         dout("kill_sb %p\n", s);
919         ceph_mdsc_pre_umount(fsc->mdsc);
920         kill_anon_super(s);    /* will call put_super after sb is r/o */
921         ceph_mdsc_destroy(fsc);
922         destroy_fs_client(fsc);
923 }
924
925 static struct file_system_type ceph_fs_type = {
926         .owner          = THIS_MODULE,
927         .name           = "ceph",
928         .mount          = ceph_mount,
929         .kill_sb        = ceph_kill_sb,
930         .fs_flags       = FS_RENAME_DOES_D_MOVE,
931 };
932
933 #define _STRINGIFY(x) #x
934 #define STRINGIFY(x) _STRINGIFY(x)
935
936 static int __init init_ceph(void)
937 {
938         int ret = init_caches();
939         if (ret)
940                 goto out;
941
942         ceph_xattr_init();
943         ret = register_filesystem(&ceph_fs_type);
944         if (ret)
945                 goto out_icache;
946
947         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
948
949         return 0;
950
951 out_icache:
952         ceph_xattr_exit();
953         destroy_caches();
954 out:
955         return ret;
956 }
957
958 static void __exit exit_ceph(void)
959 {
960         dout("exit_ceph\n");
961         unregister_filesystem(&ceph_fs_type);
962         ceph_xattr_exit();
963         destroy_caches();
964 }
965
966 module_init(init_ceph);
967 module_exit(exit_ceph);
968
969 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
970 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
971 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
972 MODULE_DESCRIPTION("Ceph filesystem for Linux");
973 MODULE_LICENSE("GPL");