autofs4 - fix deal with autofs4_write races
[linux-2.6.git] / fs / ceph / super.c
1
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/backing-dev.h>
5 #include <linux/ctype.h>
6 #include <linux/fs.h>
7 #include <linux/inet.h>
8 #include <linux/in6.h>
9 #include <linux/module.h>
10 #include <linux/mount.h>
11 #include <linux/parser.h>
12 #include <linux/sched.h>
13 #include <linux/seq_file.h>
14 #include <linux/slab.h>
15 #include <linux/statfs.h>
16 #include <linux/string.h>
17
18 #include "super.h"
19 #include "mds_client.h"
20
21 #include <linux/ceph/decode.h>
22 #include <linux/ceph/mon_client.h>
23 #include <linux/ceph/auth.h>
24 #include <linux/ceph/debugfs.h>
25
26 /*
27  * Ceph superblock operations
28  *
29  * Handle the basics of mounting, unmounting.
30  */
31
32 /*
33  * super ops
34  */
35 static void ceph_put_super(struct super_block *s)
36 {
37         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
38
39         dout("put_super\n");
40         ceph_mdsc_close_sessions(fsc->mdsc);
41
42         /*
43          * ensure we release the bdi before put_anon_super releases
44          * the device name.
45          */
46         if (s->s_bdi == &fsc->backing_dev_info) {
47                 bdi_unregister(&fsc->backing_dev_info);
48                 s->s_bdi = NULL;
49         }
50
51         return;
52 }
53
54 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
55 {
56         struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
57         struct ceph_monmap *monmap = fsc->client->monc.monmap;
58         struct ceph_statfs st;
59         u64 fsid;
60         int err;
61
62         dout("statfs\n");
63         err = ceph_monc_do_statfs(&fsc->client->monc, &st);
64         if (err < 0)
65                 return err;
66
67         /* fill in kstatfs */
68         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
69
70         /*
71          * express utilization in terms of large blocks to avoid
72          * overflow on 32-bit machines.
73          */
74         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
75         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
76         buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
77         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
78
79         buf->f_files = le64_to_cpu(st.num_objects);
80         buf->f_ffree = -1;
81         buf->f_namelen = NAME_MAX;
82         buf->f_frsize = PAGE_CACHE_SIZE;
83
84         /* leave fsid little-endian, regardless of host endianness */
85         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
86         buf->f_fsid.val[0] = fsid & 0xffffffff;
87         buf->f_fsid.val[1] = fsid >> 32;
88
89         return 0;
90 }
91
92
93 static int ceph_sync_fs(struct super_block *sb, int wait)
94 {
95         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
96
97         if (!wait) {
98                 dout("sync_fs (non-blocking)\n");
99                 ceph_flush_dirty_caps(fsc->mdsc);
100                 dout("sync_fs (non-blocking) done\n");
101                 return 0;
102         }
103
104         dout("sync_fs (blocking)\n");
105         ceph_osdc_sync(&fsc->client->osdc);
106         ceph_mdsc_sync(fsc->mdsc);
107         dout("sync_fs (blocking) done\n");
108         return 0;
109 }
110
111 /*
112  * mount options
113  */
114 enum {
115         Opt_wsize,
116         Opt_rsize,
117         Opt_rasize,
118         Opt_caps_wanted_delay_min,
119         Opt_caps_wanted_delay_max,
120         Opt_cap_release_safety,
121         Opt_readdir_max_entries,
122         Opt_readdir_max_bytes,
123         Opt_congestion_kb,
124         Opt_last_int,
125         /* int args above */
126         Opt_snapdirname,
127         Opt_last_string,
128         /* string args above */
129         Opt_dirstat,
130         Opt_nodirstat,
131         Opt_rbytes,
132         Opt_norbytes,
133         Opt_noasyncreaddir,
134         Opt_ino32,
135 };
136
137 static match_table_t fsopt_tokens = {
138         {Opt_wsize, "wsize=%d"},
139         {Opt_rsize, "rsize=%d"},
140         {Opt_rasize, "rasize=%d"},
141         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
142         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
143         {Opt_cap_release_safety, "cap_release_safety=%d"},
144         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
145         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
146         {Opt_congestion_kb, "write_congestion_kb=%d"},
147         /* int args above */
148         {Opt_snapdirname, "snapdirname=%s"},
149         /* string args above */
150         {Opt_dirstat, "dirstat"},
151         {Opt_nodirstat, "nodirstat"},
152         {Opt_rbytes, "rbytes"},
153         {Opt_norbytes, "norbytes"},
154         {Opt_noasyncreaddir, "noasyncreaddir"},
155         {Opt_ino32, "ino32"},
156         {-1, NULL}
157 };
158
159 static int parse_fsopt_token(char *c, void *private)
160 {
161         struct ceph_mount_options *fsopt = private;
162         substring_t argstr[MAX_OPT_ARGS];
163         int token, intval, ret;
164
165         token = match_token((char *)c, fsopt_tokens, argstr);
166         if (token < 0)
167                 return -EINVAL;
168
169         if (token < Opt_last_int) {
170                 ret = match_int(&argstr[0], &intval);
171                 if (ret < 0) {
172                         pr_err("bad mount option arg (not int) "
173                                "at '%s'\n", c);
174                         return ret;
175                 }
176                 dout("got int token %d val %d\n", token, intval);
177         } else if (token > Opt_last_int && token < Opt_last_string) {
178                 dout("got string token %d val %s\n", token,
179                      argstr[0].from);
180         } else {
181                 dout("got token %d\n", token);
182         }
183
184         switch (token) {
185         case Opt_snapdirname:
186                 kfree(fsopt->snapdir_name);
187                 fsopt->snapdir_name = kstrndup(argstr[0].from,
188                                                argstr[0].to-argstr[0].from,
189                                                GFP_KERNEL);
190                 if (!fsopt->snapdir_name)
191                         return -ENOMEM;
192                 break;
193
194                 /* misc */
195         case Opt_wsize:
196                 fsopt->wsize = intval;
197                 break;
198         case Opt_rsize:
199                 fsopt->rsize = intval;
200                 break;
201         case Opt_rasize:
202                 fsopt->rasize = intval;
203                 break;
204         case Opt_caps_wanted_delay_min:
205                 fsopt->caps_wanted_delay_min = intval;
206                 break;
207         case Opt_caps_wanted_delay_max:
208                 fsopt->caps_wanted_delay_max = intval;
209                 break;
210         case Opt_readdir_max_entries:
211                 fsopt->max_readdir = intval;
212                 break;
213         case Opt_readdir_max_bytes:
214                 fsopt->max_readdir_bytes = intval;
215                 break;
216         case Opt_congestion_kb:
217                 fsopt->congestion_kb = intval;
218                 break;
219         case Opt_dirstat:
220                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
221                 break;
222         case Opt_nodirstat:
223                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
224                 break;
225         case Opt_rbytes:
226                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
227                 break;
228         case Opt_norbytes:
229                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
230                 break;
231         case Opt_noasyncreaddir:
232                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
233                 break;
234         case Opt_ino32:
235                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
236                 break;
237         default:
238                 BUG_ON(token);
239         }
240         return 0;
241 }
242
243 static void destroy_mount_options(struct ceph_mount_options *args)
244 {
245         dout("destroy_mount_options %p\n", args);
246         kfree(args->snapdir_name);
247         kfree(args);
248 }
249
250 static int strcmp_null(const char *s1, const char *s2)
251 {
252         if (!s1 && !s2)
253                 return 0;
254         if (s1 && !s2)
255                 return -1;
256         if (!s1 && s2)
257                 return 1;
258         return strcmp(s1, s2);
259 }
260
261 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
262                                  struct ceph_options *new_opt,
263                                  struct ceph_fs_client *fsc)
264 {
265         struct ceph_mount_options *fsopt1 = new_fsopt;
266         struct ceph_mount_options *fsopt2 = fsc->mount_options;
267         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
268         int ret;
269
270         ret = memcmp(fsopt1, fsopt2, ofs);
271         if (ret)
272                 return ret;
273
274         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
275         if (ret)
276                 return ret;
277
278         return ceph_compare_options(new_opt, fsc->client);
279 }
280
281 static int parse_mount_options(struct ceph_mount_options **pfsopt,
282                                struct ceph_options **popt,
283                                int flags, char *options,
284                                const char *dev_name,
285                                const char **path)
286 {
287         struct ceph_mount_options *fsopt;
288         const char *dev_name_end;
289         int err = -ENOMEM;
290
291         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
292         if (!fsopt)
293                 return -ENOMEM;
294
295         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
296
297         fsopt->sb_flags = flags;
298         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
299
300         fsopt->rsize = CEPH_RSIZE_DEFAULT;
301         fsopt->rasize = CEPH_RASIZE_DEFAULT;
302         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
303         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
304         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
305         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
306         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
307         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
308         fsopt->congestion_kb = default_congestion_kb();
309
310         /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
311         err = -EINVAL;
312         if (!dev_name)
313                 goto out;
314         *path = strstr(dev_name, ":/");
315         if (*path == NULL) {
316                 pr_err("device name is missing path (no :/ in %s)\n",
317                                 dev_name);
318                 goto out;
319         }
320         dev_name_end = *path;
321         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
322
323         /* path on server */
324         *path += 2;
325         dout("server path '%s'\n", *path);
326
327         err = ceph_parse_options(popt, options, dev_name, dev_name_end,
328                                  parse_fsopt_token, (void *)fsopt);
329         if (err)
330                 goto out;
331
332         /* success */
333         *pfsopt = fsopt;
334         return 0;
335
336 out:
337         destroy_mount_options(fsopt);
338         return err;
339 }
340
341 /**
342  * ceph_show_options - Show mount options in /proc/mounts
343  * @m: seq_file to write to
344  * @root: root of that (sub)tree
345  */
346 static int ceph_show_options(struct seq_file *m, struct dentry *root)
347 {
348         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
349         struct ceph_mount_options *fsopt = fsc->mount_options;
350         struct ceph_options *opt = fsc->client->options;
351
352         if (opt->flags & CEPH_OPT_FSID)
353                 seq_printf(m, ",fsid=%pU", &opt->fsid);
354         if (opt->flags & CEPH_OPT_NOSHARE)
355                 seq_puts(m, ",noshare");
356         if (opt->flags & CEPH_OPT_NOCRC)
357                 seq_puts(m, ",nocrc");
358
359         if (opt->name)
360                 seq_printf(m, ",name=%s", opt->name);
361         if (opt->key)
362                 seq_puts(m, ",secret=<hidden>");
363
364         if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
365                 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
366         if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
367                 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
368         if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
369                 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
370         if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
371                 seq_printf(m, ",osdkeepalivetimeout=%d",
372                            opt->osd_keepalive_timeout);
373
374         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
375                 seq_puts(m, ",dirstat");
376         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
377                 seq_puts(m, ",norbytes");
378         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
379                 seq_puts(m, ",noasyncreaddir");
380
381         if (fsopt->wsize)
382                 seq_printf(m, ",wsize=%d", fsopt->wsize);
383         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
384                 seq_printf(m, ",rsize=%d", fsopt->rsize);
385         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
386                 seq_printf(m, ",rasize=%d", fsopt->rasize);
387         if (fsopt->congestion_kb != default_congestion_kb())
388                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
389         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
390                 seq_printf(m, ",caps_wanted_delay_min=%d",
391                          fsopt->caps_wanted_delay_min);
392         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
393                 seq_printf(m, ",caps_wanted_delay_max=%d",
394                            fsopt->caps_wanted_delay_max);
395         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
396                 seq_printf(m, ",cap_release_safety=%d",
397                            fsopt->cap_release_safety);
398         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
399                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
400         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
401                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
402         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
403                 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
404         return 0;
405 }
406
407 /*
408  * handle any mon messages the standard library doesn't understand.
409  * return error if we don't either.
410  */
411 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
412 {
413         struct ceph_fs_client *fsc = client->private;
414         int type = le16_to_cpu(msg->hdr.type);
415
416         switch (type) {
417         case CEPH_MSG_MDS_MAP:
418                 ceph_mdsc_handle_map(fsc->mdsc, msg);
419                 return 0;
420
421         default:
422                 return -1;
423         }
424 }
425
426 /*
427  * create a new fs client
428  */
429 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
430                                         struct ceph_options *opt)
431 {
432         struct ceph_fs_client *fsc;
433         const unsigned supported_features =
434                 CEPH_FEATURE_FLOCK |
435                 CEPH_FEATURE_DIRLAYOUTHASH;
436         const unsigned required_features = 0;
437         int err = -ENOMEM;
438
439         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
440         if (!fsc)
441                 return ERR_PTR(-ENOMEM);
442
443         fsc->client = ceph_create_client(opt, fsc, supported_features,
444                                          required_features);
445         if (IS_ERR(fsc->client)) {
446                 err = PTR_ERR(fsc->client);
447                 goto fail;
448         }
449         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
450         fsc->client->monc.want_mdsmap = 1;
451
452         fsc->mount_options = fsopt;
453
454         fsc->sb = NULL;
455         fsc->mount_state = CEPH_MOUNT_MOUNTING;
456
457         atomic_long_set(&fsc->writeback_count, 0);
458
459         err = bdi_init(&fsc->backing_dev_info);
460         if (err < 0)
461                 goto fail_client;
462
463         err = -ENOMEM;
464         /*
465          * The number of concurrent works can be high but they don't need
466          * to be processed in parallel, limit concurrency.
467          */
468         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
469         if (fsc->wb_wq == NULL)
470                 goto fail_bdi;
471         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
472         if (fsc->pg_inv_wq == NULL)
473                 goto fail_wb_wq;
474         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
475         if (fsc->trunc_wq == NULL)
476                 goto fail_pg_inv_wq;
477
478         /* set up mempools */
479         err = -ENOMEM;
480         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
481                               fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
482         if (!fsc->wb_pagevec_pool)
483                 goto fail_trunc_wq;
484
485         /* caps */
486         fsc->min_caps = fsopt->max_readdir;
487
488         return fsc;
489
490 fail_trunc_wq:
491         destroy_workqueue(fsc->trunc_wq);
492 fail_pg_inv_wq:
493         destroy_workqueue(fsc->pg_inv_wq);
494 fail_wb_wq:
495         destroy_workqueue(fsc->wb_wq);
496 fail_bdi:
497         bdi_destroy(&fsc->backing_dev_info);
498 fail_client:
499         ceph_destroy_client(fsc->client);
500 fail:
501         kfree(fsc);
502         return ERR_PTR(err);
503 }
504
505 static void destroy_fs_client(struct ceph_fs_client *fsc)
506 {
507         dout("destroy_fs_client %p\n", fsc);
508
509         destroy_workqueue(fsc->wb_wq);
510         destroy_workqueue(fsc->pg_inv_wq);
511         destroy_workqueue(fsc->trunc_wq);
512
513         bdi_destroy(&fsc->backing_dev_info);
514
515         mempool_destroy(fsc->wb_pagevec_pool);
516
517         destroy_mount_options(fsc->mount_options);
518
519         ceph_fs_debugfs_cleanup(fsc);
520
521         ceph_destroy_client(fsc->client);
522
523         kfree(fsc);
524         dout("destroy_fs_client %p done\n", fsc);
525 }
526
527 /*
528  * caches
529  */
530 struct kmem_cache *ceph_inode_cachep;
531 struct kmem_cache *ceph_cap_cachep;
532 struct kmem_cache *ceph_dentry_cachep;
533 struct kmem_cache *ceph_file_cachep;
534
535 static void ceph_inode_init_once(void *foo)
536 {
537         struct ceph_inode_info *ci = foo;
538         inode_init_once(&ci->vfs_inode);
539 }
540
541 static int __init init_caches(void)
542 {
543         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
544                                       sizeof(struct ceph_inode_info),
545                                       __alignof__(struct ceph_inode_info),
546                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
547                                       ceph_inode_init_once);
548         if (ceph_inode_cachep == NULL)
549                 return -ENOMEM;
550
551         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
552                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
553         if (ceph_cap_cachep == NULL)
554                 goto bad_cap;
555
556         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
557                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
558         if (ceph_dentry_cachep == NULL)
559                 goto bad_dentry;
560
561         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
562                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
563         if (ceph_file_cachep == NULL)
564                 goto bad_file;
565
566         return 0;
567
568 bad_file:
569         kmem_cache_destroy(ceph_dentry_cachep);
570 bad_dentry:
571         kmem_cache_destroy(ceph_cap_cachep);
572 bad_cap:
573         kmem_cache_destroy(ceph_inode_cachep);
574         return -ENOMEM;
575 }
576
577 static void destroy_caches(void)
578 {
579         kmem_cache_destroy(ceph_inode_cachep);
580         kmem_cache_destroy(ceph_cap_cachep);
581         kmem_cache_destroy(ceph_dentry_cachep);
582         kmem_cache_destroy(ceph_file_cachep);
583 }
584
585
586 /*
587  * ceph_umount_begin - initiate forced umount.  Tear down down the
588  * mount, skipping steps that may hang while waiting for server(s).
589  */
590 static void ceph_umount_begin(struct super_block *sb)
591 {
592         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
593
594         dout("ceph_umount_begin - starting forced umount\n");
595         if (!fsc)
596                 return;
597         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
598         return;
599 }
600
601 static const struct super_operations ceph_super_ops = {
602         .alloc_inode    = ceph_alloc_inode,
603         .destroy_inode  = ceph_destroy_inode,
604         .write_inode    = ceph_write_inode,
605         .sync_fs        = ceph_sync_fs,
606         .put_super      = ceph_put_super,
607         .show_options   = ceph_show_options,
608         .statfs         = ceph_statfs,
609         .umount_begin   = ceph_umount_begin,
610 };
611
612 /*
613  * Bootstrap mount by opening the root directory.  Note the mount
614  * @started time from caller, and time out if this takes too long.
615  */
616 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
617                                        const char *path,
618                                        unsigned long started)
619 {
620         struct ceph_mds_client *mdsc = fsc->mdsc;
621         struct ceph_mds_request *req = NULL;
622         int err;
623         struct dentry *root;
624
625         /* open dir */
626         dout("open_root_inode opening '%s'\n", path);
627         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
628         if (IS_ERR(req))
629                 return ERR_CAST(req);
630         req->r_path1 = kstrdup(path, GFP_NOFS);
631         req->r_ino1.ino = CEPH_INO_ROOT;
632         req->r_ino1.snap = CEPH_NOSNAP;
633         req->r_started = started;
634         req->r_timeout = fsc->client->options->mount_timeout * HZ;
635         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
636         req->r_num_caps = 2;
637         err = ceph_mdsc_do_request(mdsc, NULL, req);
638         if (err == 0) {
639                 struct inode *inode = req->r_target_inode;
640                 req->r_target_inode = NULL;
641                 dout("open_root_inode success\n");
642                 if (ceph_ino(inode) == CEPH_INO_ROOT &&
643                     fsc->sb->s_root == NULL) {
644                         root = d_alloc_root(inode);
645                         if (!root) {
646                                 iput(inode);
647                                 root = ERR_PTR(-ENOMEM);
648                                 goto out;
649                         }
650                         ceph_init_dentry(root);
651                 } else {
652                         root = d_obtain_alias(inode);
653                 }
654                 dout("open_root_inode success, root dentry is %p\n", root);
655         } else {
656                 root = ERR_PTR(err);
657         }
658 out:
659         ceph_mdsc_put_request(req);
660         return root;
661 }
662
663
664
665
666 /*
667  * mount: join the ceph cluster, and open root directory.
668  */
669 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
670                       const char *path)
671 {
672         int err;
673         unsigned long started = jiffies;  /* note the start time */
674         struct dentry *root;
675         int first = 0;   /* first vfsmount for this super_block */
676
677         dout("mount start\n");
678         mutex_lock(&fsc->client->mount_mutex);
679
680         err = __ceph_open_session(fsc->client, started);
681         if (err < 0)
682                 goto out;
683
684         dout("mount opening root\n");
685         root = open_root_dentry(fsc, "", started);
686         if (IS_ERR(root)) {
687                 err = PTR_ERR(root);
688                 goto out;
689         }
690         if (fsc->sb->s_root) {
691                 dput(root);
692         } else {
693                 fsc->sb->s_root = root;
694                 first = 1;
695
696                 err = ceph_fs_debugfs_init(fsc);
697                 if (err < 0)
698                         goto fail;
699         }
700
701         if (path[0] == 0) {
702                 dget(root);
703         } else {
704                 dout("mount opening base mountpoint\n");
705                 root = open_root_dentry(fsc, path, started);
706                 if (IS_ERR(root)) {
707                         err = PTR_ERR(root);
708                         goto fail;
709                 }
710         }
711
712         fsc->mount_state = CEPH_MOUNT_MOUNTED;
713         dout("mount success\n");
714         mutex_unlock(&fsc->client->mount_mutex);
715         return root;
716
717 out:
718         mutex_unlock(&fsc->client->mount_mutex);
719         return ERR_PTR(err);
720
721 fail:
722         if (first) {
723                 dput(fsc->sb->s_root);
724                 fsc->sb->s_root = NULL;
725         }
726         goto out;
727 }
728
729 static int ceph_set_super(struct super_block *s, void *data)
730 {
731         struct ceph_fs_client *fsc = data;
732         int ret;
733
734         dout("set_super %p data %p\n", s, data);
735
736         s->s_flags = fsc->mount_options->sb_flags;
737         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
738
739         s->s_fs_info = fsc;
740         fsc->sb = s;
741
742         s->s_op = &ceph_super_ops;
743         s->s_export_op = &ceph_export_ops;
744
745         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
746
747         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
748         if (ret != 0)
749                 goto fail;
750
751         return ret;
752
753 fail:
754         s->s_fs_info = NULL;
755         fsc->sb = NULL;
756         return ret;
757 }
758
759 /*
760  * share superblock if same fs AND options
761  */
762 static int ceph_compare_super(struct super_block *sb, void *data)
763 {
764         struct ceph_fs_client *new = data;
765         struct ceph_mount_options *fsopt = new->mount_options;
766         struct ceph_options *opt = new->client->options;
767         struct ceph_fs_client *other = ceph_sb_to_client(sb);
768
769         dout("ceph_compare_super %p\n", sb);
770
771         if (compare_mount_options(fsopt, opt, other)) {
772                 dout("monitor(s)/mount options don't match\n");
773                 return 0;
774         }
775         if ((opt->flags & CEPH_OPT_FSID) &&
776             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
777                 dout("fsid doesn't match\n");
778                 return 0;
779         }
780         if (fsopt->sb_flags != other->mount_options->sb_flags) {
781                 dout("flags differ\n");
782                 return 0;
783         }
784         return 1;
785 }
786
787 /*
788  * construct our own bdi so we can control readahead, etc.
789  */
790 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
791
792 static int ceph_register_bdi(struct super_block *sb,
793                              struct ceph_fs_client *fsc)
794 {
795         int err;
796
797         /* set ra_pages based on rasize mount option? */
798         if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
799                 fsc->backing_dev_info.ra_pages =
800                         (fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
801                         >> PAGE_SHIFT;
802         else
803                 fsc->backing_dev_info.ra_pages =
804                         default_backing_dev_info.ra_pages;
805
806         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
807                            atomic_long_inc_return(&bdi_seq));
808         if (!err)
809                 sb->s_bdi = &fsc->backing_dev_info;
810         return err;
811 }
812
813 static struct dentry *ceph_mount(struct file_system_type *fs_type,
814                        int flags, const char *dev_name, void *data)
815 {
816         struct super_block *sb;
817         struct ceph_fs_client *fsc;
818         struct dentry *res;
819         int err;
820         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
821         const char *path = NULL;
822         struct ceph_mount_options *fsopt = NULL;
823         struct ceph_options *opt = NULL;
824
825         dout("ceph_mount\n");
826         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
827         if (err < 0) {
828                 res = ERR_PTR(err);
829                 goto out_final;
830         }
831
832         /* create client (which we may/may not use) */
833         fsc = create_fs_client(fsopt, opt);
834         if (IS_ERR(fsc)) {
835                 res = ERR_CAST(fsc);
836                 destroy_mount_options(fsopt);
837                 ceph_destroy_options(opt);
838                 goto out_final;
839         }
840
841         err = ceph_mdsc_init(fsc);
842         if (err < 0) {
843                 res = ERR_PTR(err);
844                 goto out;
845         }
846
847         if (ceph_test_opt(fsc->client, NOSHARE))
848                 compare_super = NULL;
849         sb = sget(fs_type, compare_super, ceph_set_super, fsc);
850         if (IS_ERR(sb)) {
851                 res = ERR_CAST(sb);
852                 goto out;
853         }
854
855         if (ceph_sb_to_client(sb) != fsc) {
856                 ceph_mdsc_destroy(fsc);
857                 destroy_fs_client(fsc);
858                 fsc = ceph_sb_to_client(sb);
859                 dout("get_sb got existing client %p\n", fsc);
860         } else {
861                 dout("get_sb using new client %p\n", fsc);
862                 err = ceph_register_bdi(sb, fsc);
863                 if (err < 0) {
864                         res = ERR_PTR(err);
865                         goto out_splat;
866                 }
867         }
868
869         res = ceph_real_mount(fsc, path);
870         if (IS_ERR(res))
871                 goto out_splat;
872         dout("root %p inode %p ino %llx.%llx\n", res,
873              res->d_inode, ceph_vinop(res->d_inode));
874         return res;
875
876 out_splat:
877         ceph_mdsc_close_sessions(fsc->mdsc);
878         deactivate_locked_super(sb);
879         goto out_final;
880
881 out:
882         ceph_mdsc_destroy(fsc);
883         destroy_fs_client(fsc);
884 out_final:
885         dout("ceph_mount fail %ld\n", PTR_ERR(res));
886         return res;
887 }
888
889 static void ceph_kill_sb(struct super_block *s)
890 {
891         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
892         dout("kill_sb %p\n", s);
893         ceph_mdsc_pre_umount(fsc->mdsc);
894         kill_anon_super(s);    /* will call put_super after sb is r/o */
895         ceph_mdsc_destroy(fsc);
896         destroy_fs_client(fsc);
897 }
898
899 static struct file_system_type ceph_fs_type = {
900         .owner          = THIS_MODULE,
901         .name           = "ceph",
902         .mount          = ceph_mount,
903         .kill_sb        = ceph_kill_sb,
904         .fs_flags       = FS_RENAME_DOES_D_MOVE,
905 };
906
907 #define _STRINGIFY(x) #x
908 #define STRINGIFY(x) _STRINGIFY(x)
909
910 static int __init init_ceph(void)
911 {
912         int ret = init_caches();
913         if (ret)
914                 goto out;
915
916         ret = register_filesystem(&ceph_fs_type);
917         if (ret)
918                 goto out_icache;
919
920         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
921
922         return 0;
923
924 out_icache:
925         destroy_caches();
926 out:
927         return ret;
928 }
929
930 static void __exit exit_ceph(void)
931 {
932         dout("exit_ceph\n");
933         unregister_filesystem(&ceph_fs_type);
934         destroy_caches();
935 }
936
937 module_init(init_ceph);
938 module_exit(exit_ceph);
939
940 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
941 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
942 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
943 MODULE_DESCRIPTION("Ceph filesystem for Linux");
944 MODULE_LICENSE("GPL");