ceph: allow renewal of auth credentials
[linux-3.10.git] / fs / ceph / super.c
1
2 #include "ceph_debug.h"
3
4 #include <linux/backing-dev.h>
5 #include <linux/fs.h>
6 #include <linux/inet.h>
7 #include <linux/in6.h>
8 #include <linux/module.h>
9 #include <linux/mount.h>
10 #include <linux/parser.h>
11 #include <linux/rwsem.h>
12 #include <linux/sched.h>
13 #include <linux/seq_file.h>
14 #include <linux/statfs.h>
15 #include <linux/string.h>
16 #include <linux/version.h>
17 #include <linux/vmalloc.h>
18
19 #include "decode.h"
20 #include "super.h"
21 #include "mon_client.h"
22 #include "auth.h"
23
24 /*
25  * Ceph superblock operations
26  *
27  * Handle the basics of mounting, unmounting.
28  */
29
30
31 /*
32  * find filename portion of a path (/foo/bar/baz -> baz)
33  */
34 const char *ceph_file_part(const char *s, int len)
35 {
36         const char *e = s + len;
37
38         while (e != s && *(e-1) != '/')
39                 e--;
40         return e;
41 }
42
43
44 /*
45  * super ops
46  */
47 static void ceph_put_super(struct super_block *s)
48 {
49         struct ceph_client *cl = ceph_client(s);
50
51         dout("put_super\n");
52         ceph_mdsc_close_sessions(&cl->mdsc);
53         return;
54 }
55
56 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
57 {
58         struct ceph_client *client = ceph_inode_to_client(dentry->d_inode);
59         struct ceph_monmap *monmap = client->monc.monmap;
60         struct ceph_statfs st;
61         u64 fsid;
62         int err;
63
64         dout("statfs\n");
65         err = ceph_monc_do_statfs(&client->monc, &st);
66         if (err < 0)
67                 return err;
68
69         /* fill in kstatfs */
70         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
71
72         /*
73          * express utilization in terms of large blocks to avoid
74          * overflow on 32-bit machines.
75          */
76         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
77         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
78         buf->f_bfree = (le64_to_cpu(st.kb) - le64_to_cpu(st.kb_used)) >>
79                 (CEPH_BLOCK_SHIFT-10);
80         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
81
82         buf->f_files = le64_to_cpu(st.num_objects);
83         buf->f_ffree = -1;
84         buf->f_namelen = PATH_MAX;
85         buf->f_frsize = PAGE_CACHE_SIZE;
86
87         /* leave fsid little-endian, regardless of host endianness */
88         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
89         buf->f_fsid.val[0] = fsid & 0xffffffff;
90         buf->f_fsid.val[1] = fsid >> 32;
91
92         return 0;
93 }
94
95
96 static int ceph_syncfs(struct super_block *sb, int wait)
97 {
98         dout("sync_fs %d\n", wait);
99         ceph_osdc_sync(&ceph_client(sb)->osdc);
100         ceph_mdsc_sync(&ceph_client(sb)->mdsc);
101         dout("sync_fs %d done\n", wait);
102         return 0;
103 }
104
105
106 /**
107  * ceph_show_options - Show mount options in /proc/mounts
108  * @m: seq_file to write to
109  * @mnt: mount descriptor
110  */
111 static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
112 {
113         struct ceph_client *client = ceph_sb_to_client(mnt->mnt_sb);
114         struct ceph_mount_args *args = client->mount_args;
115
116         if (args->flags & CEPH_OPT_FSID)
117                 seq_printf(m, ",fsidmajor=%llu,fsidminor%llu",
118                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[0]),
119                            le64_to_cpu(*(__le64 *)&args->fsid.fsid[8]));
120         if (args->flags & CEPH_OPT_NOSHARE)
121                 seq_puts(m, ",noshare");
122         if (args->flags & CEPH_OPT_DIRSTAT)
123                 seq_puts(m, ",dirstat");
124         if ((args->flags & CEPH_OPT_RBYTES) == 0)
125                 seq_puts(m, ",norbytes");
126         if (args->flags & CEPH_OPT_NOCRC)
127                 seq_puts(m, ",nocrc");
128         if (args->flags & CEPH_OPT_NOASYNCREADDIR)
129                 seq_puts(m, ",noasyncreaddir");
130         if (strcmp(args->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
131                 seq_printf(m, ",snapdirname=%s", args->snapdir_name);
132         if (args->name)
133                 seq_printf(m, ",name=%s", args->name);
134         if (args->secret)
135                 seq_puts(m, ",secret=<hidden>");
136         return 0;
137 }
138
139 /*
140  * caches
141  */
142 struct kmem_cache *ceph_inode_cachep;
143 struct kmem_cache *ceph_cap_cachep;
144 struct kmem_cache *ceph_dentry_cachep;
145 struct kmem_cache *ceph_file_cachep;
146
147 static void ceph_inode_init_once(void *foo)
148 {
149         struct ceph_inode_info *ci = foo;
150         inode_init_once(&ci->vfs_inode);
151 }
152
153 static int default_congestion_kb(void)
154 {
155         int congestion_kb;
156
157         /*
158          * Copied from NFS
159          *
160          * congestion size, scale with available memory.
161          *
162          *  64MB:    8192k
163          * 128MB:   11585k
164          * 256MB:   16384k
165          * 512MB:   23170k
166          *   1GB:   32768k
167          *   2GB:   46340k
168          *   4GB:   65536k
169          *   8GB:   92681k
170          *  16GB:  131072k
171          *
172          * This allows larger machines to have larger/more transfers.
173          * Limit the default to 256M
174          */
175         congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
176         if (congestion_kb > 256*1024)
177                 congestion_kb = 256*1024;
178
179         return congestion_kb;
180 }
181
182 static int __init init_caches(void)
183 {
184         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
185                                       sizeof(struct ceph_inode_info),
186                                       __alignof__(struct ceph_inode_info),
187                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
188                                       ceph_inode_init_once);
189         if (ceph_inode_cachep == NULL)
190                 return -ENOMEM;
191
192         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
193                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
194         if (ceph_cap_cachep == NULL)
195                 goto bad_cap;
196
197         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
198                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
199         if (ceph_dentry_cachep == NULL)
200                 goto bad_dentry;
201
202         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
203                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
204         if (ceph_file_cachep == NULL)
205                 goto bad_file;
206
207         return 0;
208
209 bad_file:
210         kmem_cache_destroy(ceph_dentry_cachep);
211 bad_dentry:
212         kmem_cache_destroy(ceph_cap_cachep);
213 bad_cap:
214         kmem_cache_destroy(ceph_inode_cachep);
215         return -ENOMEM;
216 }
217
218 static void destroy_caches(void)
219 {
220         kmem_cache_destroy(ceph_inode_cachep);
221         kmem_cache_destroy(ceph_cap_cachep);
222         kmem_cache_destroy(ceph_dentry_cachep);
223         kmem_cache_destroy(ceph_file_cachep);
224 }
225
226
227 /*
228  * ceph_umount_begin - initiate forced umount.  Tear down down the
229  * mount, skipping steps that may hang while waiting for server(s).
230  */
231 static void ceph_umount_begin(struct super_block *sb)
232 {
233         struct ceph_client *client = ceph_sb_to_client(sb);
234
235         dout("ceph_umount_begin - starting forced umount\n");
236         if (!client)
237                 return;
238         client->mount_state = CEPH_MOUNT_SHUTDOWN;
239         return;
240 }
241
242 static const struct super_operations ceph_super_ops = {
243         .alloc_inode    = ceph_alloc_inode,
244         .destroy_inode  = ceph_destroy_inode,
245         .write_inode    = ceph_write_inode,
246         .sync_fs        = ceph_syncfs,
247         .put_super      = ceph_put_super,
248         .show_options   = ceph_show_options,
249         .statfs         = ceph_statfs,
250         .umount_begin   = ceph_umount_begin,
251 };
252
253
254 const char *ceph_msg_type_name(int type)
255 {
256         switch (type) {
257         case CEPH_MSG_SHUTDOWN: return "shutdown";
258         case CEPH_MSG_PING: return "ping";
259         case CEPH_MSG_AUTH: return "auth";
260         case CEPH_MSG_AUTH_REPLY: return "auth_reply";
261         case CEPH_MSG_MON_MAP: return "mon_map";
262         case CEPH_MSG_MON_GET_MAP: return "mon_get_map";
263         case CEPH_MSG_MON_SUBSCRIBE: return "mon_subscribe";
264         case CEPH_MSG_MON_SUBSCRIBE_ACK: return "mon_subscribe_ack";
265         case CEPH_MSG_STATFS: return "statfs";
266         case CEPH_MSG_STATFS_REPLY: return "statfs_reply";
267         case CEPH_MSG_MDS_MAP: return "mds_map";
268         case CEPH_MSG_CLIENT_SESSION: return "client_session";
269         case CEPH_MSG_CLIENT_RECONNECT: return "client_reconnect";
270         case CEPH_MSG_CLIENT_REQUEST: return "client_request";
271         case CEPH_MSG_CLIENT_REQUEST_FORWARD: return "client_request_forward";
272         case CEPH_MSG_CLIENT_REPLY: return "client_reply";
273         case CEPH_MSG_CLIENT_CAPS: return "client_caps";
274         case CEPH_MSG_CLIENT_CAPRELEASE: return "client_cap_release";
275         case CEPH_MSG_CLIENT_SNAP: return "client_snap";
276         case CEPH_MSG_CLIENT_LEASE: return "client_lease";
277         case CEPH_MSG_OSD_MAP: return "osd_map";
278         case CEPH_MSG_OSD_OP: return "osd_op";
279         case CEPH_MSG_OSD_OPREPLY: return "osd_opreply";
280         default: return "unknown";
281         }
282 }
283
284
285 /*
286  * mount options
287  */
288 enum {
289         Opt_fsidmajor,
290         Opt_fsidminor,
291         Opt_monport,
292         Opt_wsize,
293         Opt_rsize,
294         Opt_osdtimeout,
295         Opt_mount_timeout,
296         Opt_caps_wanted_delay_min,
297         Opt_caps_wanted_delay_max,
298         Opt_readdir_max_entries,
299         Opt_congestion_kb,
300         Opt_last_int,
301         /* int args above */
302         Opt_snapdirname,
303         Opt_name,
304         Opt_secret,
305         Opt_last_string,
306         /* string args above */
307         Opt_ip,
308         Opt_noshare,
309         Opt_dirstat,
310         Opt_nodirstat,
311         Opt_rbytes,
312         Opt_norbytes,
313         Opt_nocrc,
314         Opt_noasyncreaddir,
315 };
316
317 static match_table_t arg_tokens = {
318         {Opt_fsidmajor, "fsidmajor=%ld"},
319         {Opt_fsidminor, "fsidminor=%ld"},
320         {Opt_monport, "monport=%d"},
321         {Opt_wsize, "wsize=%d"},
322         {Opt_rsize, "rsize=%d"},
323         {Opt_osdtimeout, "osdtimeout=%d"},
324         {Opt_mount_timeout, "mount_timeout=%d"},
325         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
326         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
327         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
328         {Opt_congestion_kb, "write_congestion_kb=%d"},
329         /* int args above */
330         {Opt_snapdirname, "snapdirname=%s"},
331         {Opt_name, "name=%s"},
332         {Opt_secret, "secret=%s"},
333         /* string args above */
334         {Opt_ip, "ip=%s"},
335         {Opt_noshare, "noshare"},
336         {Opt_dirstat, "dirstat"},
337         {Opt_nodirstat, "nodirstat"},
338         {Opt_rbytes, "rbytes"},
339         {Opt_norbytes, "norbytes"},
340         {Opt_nocrc, "nocrc"},
341         {Opt_noasyncreaddir, "noasyncreaddir"},
342         {-1, NULL}
343 };
344
345
346 static struct ceph_mount_args *parse_mount_args(int flags, char *options,
347                                                 const char *dev_name,
348                                                 const char **path)
349 {
350         struct ceph_mount_args *args;
351         const char *c;
352         int err = -ENOMEM;
353         substring_t argstr[MAX_OPT_ARGS];
354
355         args = kzalloc(sizeof(*args), GFP_KERNEL);
356         if (!args)
357                 return ERR_PTR(-ENOMEM);
358         args->mon_addr = kcalloc(CEPH_MAX_MON, sizeof(*args->mon_addr),
359                                  GFP_KERNEL);
360         if (!args->mon_addr)
361                 goto out;
362
363         dout("parse_mount_args %p, dev_name '%s'\n", args, dev_name);
364
365         /* start with defaults */
366         args->sb_flags = flags;
367         args->flags = CEPH_OPT_DEFAULT;
368         args->osd_timeout = 5;    /* seconds */
369         args->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT; /* seconds */
370         args->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
371         args->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
372         args->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
373         args->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
374         args->cap_release_safety = CEPH_CAPS_PER_RELEASE * 4;
375         args->max_readdir = 1024;
376         args->congestion_kb = default_congestion_kb();
377
378         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
379         err = -EINVAL;
380         if (!dev_name)
381                 goto out;
382         *path = strstr(dev_name, ":/");
383         if (*path == NULL) {
384                 pr_err("device name is missing path (no :/ in %s)\n",
385                        dev_name);
386                 goto out;
387         }
388
389         /* get mon ip(s) */
390         err = ceph_parse_ips(dev_name, *path, args->mon_addr,
391                              CEPH_MAX_MON, &args->num_mon);
392         if (err < 0)
393                 goto out;
394
395         /* path on server */
396         *path += 2;
397         dout("server path '%s'\n", *path);
398
399         /* parse mount options */
400         while ((c = strsep(&options, ",")) != NULL) {
401                 int token, intval, ret;
402                 if (!*c)
403                         continue;
404                 err = -EINVAL;
405                 token = match_token((char *)c, arg_tokens, argstr);
406                 if (token < 0) {
407                         pr_err("bad mount option at '%s'\n", c);
408                         goto out;
409                 }
410                 if (token < Opt_last_int) {
411                         ret = match_int(&argstr[0], &intval);
412                         if (ret < 0) {
413                                 pr_err("bad mount option arg (not int) "
414                                        "at '%s'\n", c);
415                                 continue;
416                         }
417                         dout("got int token %d val %d\n", token, intval);
418                 } else if (token > Opt_last_int && token < Opt_last_string) {
419                         dout("got string token %d val %s\n", token,
420                              argstr[0].from);
421                 } else {
422                         dout("got token %d\n", token);
423                 }
424                 switch (token) {
425                 case Opt_fsidmajor:
426                         *(__le64 *)&args->fsid.fsid[0] = cpu_to_le64(intval);
427                         break;
428                 case Opt_fsidminor:
429                         *(__le64 *)&args->fsid.fsid[8] = cpu_to_le64(intval);
430                         break;
431                 case Opt_ip:
432                         err = ceph_parse_ips(argstr[0].from,
433                                              argstr[0].to,
434                                              &args->my_addr,
435                                              1, NULL);
436                         if (err < 0)
437                                 goto out;
438                         args->flags |= CEPH_OPT_MYIP;
439                         break;
440
441                 case Opt_snapdirname:
442                         kfree(args->snapdir_name);
443                         args->snapdir_name = kstrndup(argstr[0].from,
444                                               argstr[0].to-argstr[0].from,
445                                               GFP_KERNEL);
446                         break;
447                 case Opt_name:
448                         args->name = kstrndup(argstr[0].from,
449                                               argstr[0].to-argstr[0].from,
450                                               GFP_KERNEL);
451                         break;
452                 case Opt_secret:
453                         args->secret = kstrndup(argstr[0].from,
454                                                 argstr[0].to-argstr[0].from,
455                                                 GFP_KERNEL);
456                         break;
457
458                         /* misc */
459                 case Opt_wsize:
460                         args->wsize = intval;
461                         break;
462                 case Opt_rsize:
463                         args->rsize = intval;
464                         break;
465                 case Opt_osdtimeout:
466                         args->osd_timeout = intval;
467                         break;
468                 case Opt_mount_timeout:
469                         args->mount_timeout = intval;
470                         break;
471                 case Opt_caps_wanted_delay_min:
472                         args->caps_wanted_delay_min = intval;
473                         break;
474                 case Opt_caps_wanted_delay_max:
475                         args->caps_wanted_delay_max = intval;
476                         break;
477                 case Opt_readdir_max_entries:
478                         args->max_readdir = intval;
479                         break;
480                 case Opt_congestion_kb:
481                         args->congestion_kb = intval;
482                         break;
483
484                 case Opt_noshare:
485                         args->flags |= CEPH_OPT_NOSHARE;
486                         break;
487
488                 case Opt_dirstat:
489                         args->flags |= CEPH_OPT_DIRSTAT;
490                         break;
491                 case Opt_nodirstat:
492                         args->flags &= ~CEPH_OPT_DIRSTAT;
493                         break;
494                 case Opt_rbytes:
495                         args->flags |= CEPH_OPT_RBYTES;
496                         break;
497                 case Opt_norbytes:
498                         args->flags &= ~CEPH_OPT_RBYTES;
499                         break;
500                 case Opt_nocrc:
501                         args->flags |= CEPH_OPT_NOCRC;
502                         break;
503                 case Opt_noasyncreaddir:
504                         args->flags |= CEPH_OPT_NOASYNCREADDIR;
505                         break;
506
507                 default:
508                         BUG_ON(token);
509                 }
510         }
511         return args;
512
513 out:
514         kfree(args->mon_addr);
515         kfree(args);
516         return ERR_PTR(err);
517 }
518
519 static void destroy_mount_args(struct ceph_mount_args *args)
520 {
521         dout("destroy_mount_args %p\n", args);
522         kfree(args->snapdir_name);
523         args->snapdir_name = NULL;
524         kfree(args->name);
525         args->name = NULL;
526         kfree(args->secret);
527         args->secret = NULL;
528         kfree(args);
529 }
530
531 /*
532  * create a fresh client instance
533  */
534 static struct ceph_client *ceph_create_client(struct ceph_mount_args *args)
535 {
536         struct ceph_client *client;
537         int err = -ENOMEM;
538
539         client = kzalloc(sizeof(*client), GFP_KERNEL);
540         if (client == NULL)
541                 return ERR_PTR(-ENOMEM);
542
543         mutex_init(&client->mount_mutex);
544
545         init_waitqueue_head(&client->auth_wq);
546
547         client->sb = NULL;
548         client->mount_state = CEPH_MOUNT_MOUNTING;
549         client->mount_args = args;
550
551         client->msgr = NULL;
552
553         client->auth_err = 0;
554         atomic_long_set(&client->writeback_count, 0);
555
556         err = bdi_init(&client->backing_dev_info);
557         if (err < 0)
558                 goto fail;
559
560         err = -ENOMEM;
561         client->wb_wq = create_workqueue("ceph-writeback");
562         if (client->wb_wq == NULL)
563                 goto fail_bdi;
564         client->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
565         if (client->pg_inv_wq == NULL)
566                 goto fail_wb_wq;
567         client->trunc_wq = create_singlethread_workqueue("ceph-trunc");
568         if (client->trunc_wq == NULL)
569                 goto fail_pg_inv_wq;
570
571         /* set up mempools */
572         err = -ENOMEM;
573         client->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
574                               client->mount_args->wsize >> PAGE_CACHE_SHIFT);
575         if (!client->wb_pagevec_pool)
576                 goto fail_trunc_wq;
577
578
579         /* subsystems */
580         err = ceph_monc_init(&client->monc, client);
581         if (err < 0)
582                 goto fail_mempool;
583         err = ceph_osdc_init(&client->osdc, client);
584         if (err < 0)
585                 goto fail_monc;
586         err = ceph_mdsc_init(&client->mdsc, client);
587         if (err < 0)
588                 goto fail_osdc;
589         return client;
590
591 fail_osdc:
592         ceph_osdc_stop(&client->osdc);
593 fail_monc:
594         ceph_monc_stop(&client->monc);
595 fail_mempool:
596         mempool_destroy(client->wb_pagevec_pool);
597 fail_trunc_wq:
598         destroy_workqueue(client->trunc_wq);
599 fail_pg_inv_wq:
600         destroy_workqueue(client->pg_inv_wq);
601 fail_wb_wq:
602         destroy_workqueue(client->wb_wq);
603 fail_bdi:
604         bdi_destroy(&client->backing_dev_info);
605 fail:
606         kfree(client);
607         return ERR_PTR(err);
608 }
609
610 static void ceph_destroy_client(struct ceph_client *client)
611 {
612         dout("destroy_client %p\n", client);
613
614         /* unmount */
615         ceph_mdsc_stop(&client->mdsc);
616         ceph_monc_stop(&client->monc);
617         ceph_osdc_stop(&client->osdc);
618
619         ceph_debugfs_client_cleanup(client);
620         destroy_workqueue(client->wb_wq);
621         destroy_workqueue(client->pg_inv_wq);
622         destroy_workqueue(client->trunc_wq);
623
624         if (client->msgr)
625                 ceph_messenger_destroy(client->msgr);
626         mempool_destroy(client->wb_pagevec_pool);
627
628         destroy_mount_args(client->mount_args);
629
630         kfree(client);
631         dout("destroy_client %p done\n", client);
632 }
633
634 /*
635  * Initially learn our fsid, or verify an fsid matches.
636  */
637 int ceph_check_fsid(struct ceph_client *client, struct ceph_fsid *fsid)
638 {
639         if (client->have_fsid) {
640                 if (ceph_fsid_compare(&client->fsid, fsid)) {
641                         pr_err("bad fsid, had " FSID_FORMAT " got " FSID_FORMAT,
642                                PR_FSID(&client->fsid), PR_FSID(fsid));
643                         return -1;
644                 }
645         } else {
646                 pr_info("client%lld fsid " FSID_FORMAT "\n",
647                         client->monc.auth->global_id, PR_FSID(fsid));
648                 memcpy(&client->fsid, fsid, sizeof(*fsid));
649                 ceph_debugfs_client_init(client);
650                 client->have_fsid = true;
651         }
652         return 0;
653 }
654
655 /*
656  * true if we have the mon map (and have thus joined the cluster)
657  */
658 static int have_mon_map(struct ceph_client *client)
659 {
660         return client->monc.monmap && client->monc.monmap->epoch;
661 }
662
663 /*
664  * Bootstrap mount by opening the root directory.  Note the mount
665  * @started time from caller, and time out if this takes too long.
666  */
667 static struct dentry *open_root_dentry(struct ceph_client *client,
668                                        const char *path,
669                                        unsigned long started)
670 {
671         struct ceph_mds_client *mdsc = &client->mdsc;
672         struct ceph_mds_request *req = NULL;
673         int err;
674         struct dentry *root;
675
676         /* open dir */
677         dout("open_root_inode opening '%s'\n", path);
678         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
679         if (IS_ERR(req))
680                 return ERR_PTR(PTR_ERR(req));
681         req->r_path1 = kstrdup(path, GFP_NOFS);
682         req->r_ino1.ino = CEPH_INO_ROOT;
683         req->r_ino1.snap = CEPH_NOSNAP;
684         req->r_started = started;
685         req->r_timeout = client->mount_args->mount_timeout * HZ;
686         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
687         req->r_num_caps = 2;
688         err = ceph_mdsc_do_request(mdsc, NULL, req);
689         if (err == 0) {
690                 dout("open_root_inode success\n");
691                 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
692                     client->sb->s_root == NULL)
693                         root = d_alloc_root(req->r_target_inode);
694                 else
695                         root = d_obtain_alias(req->r_target_inode);
696                 req->r_target_inode = NULL;
697                 dout("open_root_inode success, root dentry is %p\n", root);
698         } else {
699                 root = ERR_PTR(err);
700         }
701         ceph_mdsc_put_request(req);
702         return root;
703 }
704
705 /*
706  * mount: join the ceph cluster, and open root directory.
707  */
708 static int ceph_mount(struct ceph_client *client, struct vfsmount *mnt,
709                       const char *path)
710 {
711         struct ceph_entity_addr *myaddr = NULL;
712         int err;
713         unsigned long timeout = client->mount_args->mount_timeout * HZ;
714         unsigned long started = jiffies;  /* note the start time */
715         struct dentry *root;
716
717         dout("mount start\n");
718         mutex_lock(&client->mount_mutex);
719
720         /* initialize the messenger */
721         if (client->msgr == NULL) {
722                 if (ceph_test_opt(client, MYIP))
723                         myaddr = &client->mount_args->my_addr;
724                 client->msgr = ceph_messenger_create(myaddr);
725                 if (IS_ERR(client->msgr)) {
726                         err = PTR_ERR(client->msgr);
727                         client->msgr = NULL;
728                         goto out;
729                 }
730                 client->msgr->nocrc = ceph_test_opt(client, NOCRC);
731         }
732
733         /* open session, and wait for mon, mds, and osd maps */
734         err = ceph_monc_open_session(&client->monc);
735         if (err < 0)
736                 goto out;
737
738         while (!have_mon_map(client)) {
739                 err = -EIO;
740                 if (timeout && time_after_eq(jiffies, started + timeout))
741                         goto out;
742
743                 /* wait */
744                 dout("mount waiting for mon_map\n");
745                 err = wait_event_interruptible_timeout(client->auth_wq,
746                                have_mon_map(client) || (client->auth_err < 0),
747                                timeout);
748                 if (err == -EINTR || err == -ERESTARTSYS)
749                         goto out;
750                 if (client->auth_err < 0) {
751                         err = client->auth_err;
752                         goto out;
753                 }
754         }
755
756         dout("mount opening root\n");
757         root = open_root_dentry(client, "", started);
758         if (IS_ERR(root)) {
759                 err = PTR_ERR(root);
760                 goto out;
761         }
762         if (client->sb->s_root)
763                 dput(root);
764         else
765                 client->sb->s_root = root;
766
767         if (path[0] == 0) {
768                 dget(root);
769         } else {
770                 dout("mount opening base mountpoint\n");
771                 root = open_root_dentry(client, path, started);
772                 if (IS_ERR(root)) {
773                         err = PTR_ERR(root);
774                         dput(client->sb->s_root);
775                         client->sb->s_root = NULL;
776                         goto out;
777                 }
778         }
779
780         mnt->mnt_root = root;
781         mnt->mnt_sb = client->sb;
782
783         client->mount_state = CEPH_MOUNT_MOUNTED;
784         dout("mount success\n");
785         err = 0;
786
787 out:
788         mutex_unlock(&client->mount_mutex);
789         return err;
790 }
791
792 static int ceph_set_super(struct super_block *s, void *data)
793 {
794         struct ceph_client *client = data;
795         int ret;
796
797         dout("set_super %p data %p\n", s, data);
798
799         s->s_flags = client->mount_args->sb_flags;
800         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
801
802         s->s_fs_info = client;
803         client->sb = s;
804
805         s->s_op = &ceph_super_ops;
806         s->s_export_op = &ceph_export_ops;
807
808         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
809
810         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
811         if (ret != 0)
812                 goto fail;
813
814         return ret;
815
816 fail:
817         s->s_fs_info = NULL;
818         client->sb = NULL;
819         return ret;
820 }
821
822 /*
823  * share superblock if same fs AND options
824  */
825 static int ceph_compare_super(struct super_block *sb, void *data)
826 {
827         struct ceph_client *new = data;
828         struct ceph_mount_args *args = new->mount_args;
829         struct ceph_client *other = ceph_sb_to_client(sb);
830         int i;
831
832         dout("ceph_compare_super %p\n", sb);
833         if (args->flags & CEPH_OPT_FSID) {
834                 if (ceph_fsid_compare(&args->fsid, &other->fsid)) {
835                         dout("fsid doesn't match\n");
836                         return 0;
837                 }
838         } else {
839                 /* do we share (a) monitor? */
840                 for (i = 0; i < new->monc.monmap->num_mon; i++)
841                         if (ceph_monmap_contains(other->monc.monmap,
842                                          &new->monc.monmap->mon_inst[i].addr))
843                                 break;
844                 if (i == new->monc.monmap->num_mon) {
845                         dout("mon ip not part of monmap\n");
846                         return 0;
847                 }
848                 dout("mon ip matches existing sb %p\n", sb);
849         }
850         if (args->sb_flags != other->mount_args->sb_flags) {
851                 dout("flags differ\n");
852                 return 0;
853         }
854         return 1;
855 }
856
857 /*
858  * construct our own bdi so we can control readahead, etc.
859  */
860 static int ceph_register_bdi(struct super_block *sb, struct ceph_client *client)
861 {
862         int err;
863
864         sb->s_bdi = &client->backing_dev_info;
865
866         /* set ra_pages based on rsize mount option? */
867         if (client->mount_args->rsize >= PAGE_CACHE_SIZE)
868                 client->backing_dev_info.ra_pages =
869                         (client->mount_args->rsize + PAGE_CACHE_SIZE - 1)
870                         >> PAGE_SHIFT;
871         err = bdi_register_dev(&client->backing_dev_info, sb->s_dev);
872         return err;
873 }
874
875 static int ceph_get_sb(struct file_system_type *fs_type,
876                        int flags, const char *dev_name, void *data,
877                        struct vfsmount *mnt)
878 {
879         struct super_block *sb;
880         struct ceph_client *client;
881         int err;
882         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
883         const char *path = NULL;
884         struct ceph_mount_args *args;
885
886         dout("ceph_get_sb\n");
887         args = parse_mount_args(flags, data, dev_name, &path);
888         if (IS_ERR(args)) {
889                 err = PTR_ERR(args);
890                 goto out_final;
891         }
892
893         /* create client (which we may/may not use) */
894         client = ceph_create_client(args);
895         if (IS_ERR(client)) {
896                 err = PTR_ERR(client);
897                 goto out_final;
898         }
899
900         if (client->mount_args->flags & CEPH_OPT_NOSHARE)
901                 compare_super = NULL;
902         sb = sget(fs_type, compare_super, ceph_set_super, client);
903         if (IS_ERR(sb)) {
904                 err = PTR_ERR(sb);
905                 goto out;
906         }
907
908         if (ceph_client(sb) != client) {
909                 ceph_destroy_client(client);
910                 client = ceph_client(sb);
911                 dout("get_sb got existing client %p\n", client);
912         } else {
913                 dout("get_sb using new client %p\n", client);
914                 err = ceph_register_bdi(sb, client);
915                 if (err < 0)
916                         goto out_splat;
917         }
918
919         err = ceph_mount(client, mnt, path);
920         if (err < 0)
921                 goto out_splat;
922         dout("root %p inode %p ino %llx.%llx\n", mnt->mnt_root,
923              mnt->mnt_root->d_inode, ceph_vinop(mnt->mnt_root->d_inode));
924         return 0;
925
926 out_splat:
927         ceph_mdsc_close_sessions(&client->mdsc);
928         up_write(&sb->s_umount);
929         deactivate_super(sb);
930         goto out_final;
931
932 out:
933         ceph_destroy_client(client);
934 out_final:
935         dout("ceph_get_sb fail %d\n", err);
936         return err;
937 }
938
939 static void ceph_kill_sb(struct super_block *s)
940 {
941         struct ceph_client *client = ceph_sb_to_client(s);
942         dout("kill_sb %p\n", s);
943         ceph_mdsc_pre_umount(&client->mdsc);
944         kill_anon_super(s);    /* will call put_super after sb is r/o */
945         if (s->s_bdi == &client->backing_dev_info)
946                 bdi_unregister(&client->backing_dev_info);
947         bdi_destroy(&client->backing_dev_info);
948         ceph_destroy_client(client);
949 }
950
951 static struct file_system_type ceph_fs_type = {
952         .owner          = THIS_MODULE,
953         .name           = "ceph",
954         .get_sb         = ceph_get_sb,
955         .kill_sb        = ceph_kill_sb,
956         .fs_flags       = FS_RENAME_DOES_D_MOVE,
957 };
958
959 #define _STRINGIFY(x) #x
960 #define STRINGIFY(x) _STRINGIFY(x)
961
962 static int __init init_ceph(void)
963 {
964         int ret = 0;
965
966         ret = ceph_debugfs_init();
967         if (ret < 0)
968                 goto out;
969
970         ret = ceph_msgr_init();
971         if (ret < 0)
972                 goto out_debugfs;
973
974         ret = init_caches();
975         if (ret)
976                 goto out_msgr;
977
978         ceph_caps_init();
979
980         ret = register_filesystem(&ceph_fs_type);
981         if (ret)
982                 goto out_icache;
983
984         pr_info("loaded %d.%d.%d (mon/mds/osd proto %d/%d/%d)\n",
985                 CEPH_VERSION_MAJOR, CEPH_VERSION_MINOR, CEPH_VERSION_PATCH,
986                 CEPH_MONC_PROTOCOL, CEPH_MDSC_PROTOCOL, CEPH_OSDC_PROTOCOL);
987         return 0;
988
989 out_icache:
990         destroy_caches();
991 out_msgr:
992         ceph_msgr_exit();
993 out_debugfs:
994         ceph_debugfs_cleanup();
995 out:
996         return ret;
997 }
998
999 static void __exit exit_ceph(void)
1000 {
1001         dout("exit_ceph\n");
1002         unregister_filesystem(&ceph_fs_type);
1003         ceph_caps_finalize();
1004         destroy_caches();
1005         ceph_msgr_exit();
1006         ceph_debugfs_cleanup();
1007 }
1008
1009 module_init(init_ceph);
1010 module_exit(exit_ceph);
1011
1012 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1013 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1014 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1015 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1016 MODULE_LICENSE("GPL");