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