Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[muen/linux.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 #include "cache.h"
21
22 #include <linux/ceph/ceph_features.h>
23 #include <linux/ceph/decode.h>
24 #include <linux/ceph/mon_client.h>
25 #include <linux/ceph/auth.h>
26 #include <linux/ceph/debugfs.h>
27
28 /*
29  * Ceph superblock operations
30  *
31  * Handle the basics of mounting, unmounting.
32  */
33
34 /*
35  * super ops
36  */
37 static void ceph_put_super(struct super_block *s)
38 {
39         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
40
41         dout("put_super\n");
42         ceph_mdsc_close_sessions(fsc->mdsc);
43 }
44
45 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
46 {
47         struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
48         struct ceph_monmap *monmap = fsc->client->monc.monmap;
49         struct ceph_statfs st;
50         u64 fsid;
51         int err;
52
53         dout("statfs\n");
54         err = ceph_monc_do_statfs(&fsc->client->monc, &st);
55         if (err < 0)
56                 return err;
57
58         /* fill in kstatfs */
59         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
60
61         /*
62          * express utilization in terms of large blocks to avoid
63          * overflow on 32-bit machines.
64          *
65          * NOTE: for the time being, we make bsize == frsize to humor
66          * not-yet-ancient versions of glibc that are broken.
67          * Someday, we will probably want to report a real block
68          * size...  whatever that may mean for a network file system!
69          */
70         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
71         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
72         buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
73         buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
74         buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
75
76         buf->f_files = le64_to_cpu(st.num_objects);
77         buf->f_ffree = -1;
78         buf->f_namelen = NAME_MAX;
79
80         /* leave fsid little-endian, regardless of host endianness */
81         fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
82         buf->f_fsid.val[0] = fsid & 0xffffffff;
83         buf->f_fsid.val[1] = fsid >> 32;
84
85         return 0;
86 }
87
88
89 static int ceph_sync_fs(struct super_block *sb, int wait)
90 {
91         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
92
93         if (!wait) {
94                 dout("sync_fs (non-blocking)\n");
95                 ceph_flush_dirty_caps(fsc->mdsc);
96                 dout("sync_fs (non-blocking) done\n");
97                 return 0;
98         }
99
100         dout("sync_fs (blocking)\n");
101         ceph_osdc_sync(&fsc->client->osdc);
102         ceph_mdsc_sync(fsc->mdsc);
103         dout("sync_fs (blocking) done\n");
104         return 0;
105 }
106
107 /*
108  * mount options
109  */
110 enum {
111         Opt_wsize,
112         Opt_rsize,
113         Opt_rasize,
114         Opt_caps_wanted_delay_min,
115         Opt_caps_wanted_delay_max,
116         Opt_cap_release_safety,
117         Opt_readdir_max_entries,
118         Opt_readdir_max_bytes,
119         Opt_congestion_kb,
120         Opt_last_int,
121         /* int args above */
122         Opt_snapdirname,
123         Opt_last_string,
124         /* string args above */
125         Opt_dirstat,
126         Opt_nodirstat,
127         Opt_rbytes,
128         Opt_norbytes,
129         Opt_asyncreaddir,
130         Opt_noasyncreaddir,
131         Opt_dcache,
132         Opt_nodcache,
133         Opt_ino32,
134         Opt_noino32,
135         Opt_fscache,
136         Opt_nofscache,
137         Opt_poolperm,
138         Opt_nopoolperm,
139 #ifdef CONFIG_CEPH_FS_POSIX_ACL
140         Opt_acl,
141 #endif
142         Opt_noacl,
143 };
144
145 static match_table_t fsopt_tokens = {
146         {Opt_wsize, "wsize=%d"},
147         {Opt_rsize, "rsize=%d"},
148         {Opt_rasize, "rasize=%d"},
149         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
150         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
151         {Opt_cap_release_safety, "cap_release_safety=%d"},
152         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
153         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
154         {Opt_congestion_kb, "write_congestion_kb=%d"},
155         /* int args above */
156         {Opt_snapdirname, "snapdirname=%s"},
157         /* string args above */
158         {Opt_dirstat, "dirstat"},
159         {Opt_nodirstat, "nodirstat"},
160         {Opt_rbytes, "rbytes"},
161         {Opt_norbytes, "norbytes"},
162         {Opt_asyncreaddir, "asyncreaddir"},
163         {Opt_noasyncreaddir, "noasyncreaddir"},
164         {Opt_dcache, "dcache"},
165         {Opt_nodcache, "nodcache"},
166         {Opt_ino32, "ino32"},
167         {Opt_noino32, "noino32"},
168         {Opt_fscache, "fsc"},
169         {Opt_nofscache, "nofsc"},
170         {Opt_poolperm, "poolperm"},
171         {Opt_nopoolperm, "nopoolperm"},
172 #ifdef CONFIG_CEPH_FS_POSIX_ACL
173         {Opt_acl, "acl"},
174 #endif
175         {Opt_noacl, "noacl"},
176         {-1, NULL}
177 };
178
179 static int parse_fsopt_token(char *c, void *private)
180 {
181         struct ceph_mount_options *fsopt = private;
182         substring_t argstr[MAX_OPT_ARGS];
183         int token, intval, ret;
184
185         token = match_token((char *)c, fsopt_tokens, argstr);
186         if (token < 0)
187                 return -EINVAL;
188
189         if (token < Opt_last_int) {
190                 ret = match_int(&argstr[0], &intval);
191                 if (ret < 0) {
192                         pr_err("bad mount option arg (not int) "
193                                "at '%s'\n", c);
194                         return ret;
195                 }
196                 dout("got int token %d val %d\n", token, intval);
197         } else if (token > Opt_last_int && token < Opt_last_string) {
198                 dout("got string token %d val %s\n", token,
199                      argstr[0].from);
200         } else {
201                 dout("got token %d\n", token);
202         }
203
204         switch (token) {
205         case Opt_snapdirname:
206                 kfree(fsopt->snapdir_name);
207                 fsopt->snapdir_name = kstrndup(argstr[0].from,
208                                                argstr[0].to-argstr[0].from,
209                                                GFP_KERNEL);
210                 if (!fsopt->snapdir_name)
211                         return -ENOMEM;
212                 break;
213
214                 /* misc */
215         case Opt_wsize:
216                 fsopt->wsize = intval;
217                 break;
218         case Opt_rsize:
219                 fsopt->rsize = intval;
220                 break;
221         case Opt_rasize:
222                 fsopt->rasize = intval;
223                 break;
224         case Opt_caps_wanted_delay_min:
225                 fsopt->caps_wanted_delay_min = intval;
226                 break;
227         case Opt_caps_wanted_delay_max:
228                 fsopt->caps_wanted_delay_max = intval;
229                 break;
230         case Opt_readdir_max_entries:
231                 fsopt->max_readdir = intval;
232                 break;
233         case Opt_readdir_max_bytes:
234                 fsopt->max_readdir_bytes = intval;
235                 break;
236         case Opt_congestion_kb:
237                 fsopt->congestion_kb = intval;
238                 break;
239         case Opt_dirstat:
240                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
241                 break;
242         case Opt_nodirstat:
243                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
244                 break;
245         case Opt_rbytes:
246                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
247                 break;
248         case Opt_norbytes:
249                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
250                 break;
251         case Opt_asyncreaddir:
252                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
253                 break;
254         case Opt_noasyncreaddir:
255                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
256                 break;
257         case Opt_dcache:
258                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
259                 break;
260         case Opt_nodcache:
261                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
262                 break;
263         case Opt_ino32:
264                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
265                 break;
266         case Opt_noino32:
267                 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
268                 break;
269         case Opt_fscache:
270                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
271                 break;
272         case Opt_nofscache:
273                 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
274                 break;
275         case Opt_poolperm:
276                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
277                 printk ("pool perm");
278                 break;
279         case Opt_nopoolperm:
280                 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
281                 break;
282 #ifdef CONFIG_CEPH_FS_POSIX_ACL
283         case Opt_acl:
284                 fsopt->sb_flags |= MS_POSIXACL;
285                 break;
286 #endif
287         case Opt_noacl:
288                 fsopt->sb_flags &= ~MS_POSIXACL;
289                 break;
290         default:
291                 BUG_ON(token);
292         }
293         return 0;
294 }
295
296 static void destroy_mount_options(struct ceph_mount_options *args)
297 {
298         dout("destroy_mount_options %p\n", args);
299         kfree(args->snapdir_name);
300         kfree(args);
301 }
302
303 static int strcmp_null(const char *s1, const char *s2)
304 {
305         if (!s1 && !s2)
306                 return 0;
307         if (s1 && !s2)
308                 return -1;
309         if (!s1 && s2)
310                 return 1;
311         return strcmp(s1, s2);
312 }
313
314 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
315                                  struct ceph_options *new_opt,
316                                  struct ceph_fs_client *fsc)
317 {
318         struct ceph_mount_options *fsopt1 = new_fsopt;
319         struct ceph_mount_options *fsopt2 = fsc->mount_options;
320         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
321         int ret;
322
323         ret = memcmp(fsopt1, fsopt2, ofs);
324         if (ret)
325                 return ret;
326
327         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
328         if (ret)
329                 return ret;
330
331         return ceph_compare_options(new_opt, fsc->client);
332 }
333
334 static int parse_mount_options(struct ceph_mount_options **pfsopt,
335                                struct ceph_options **popt,
336                                int flags, char *options,
337                                const char *dev_name,
338                                const char **path)
339 {
340         struct ceph_mount_options *fsopt;
341         const char *dev_name_end;
342         int err;
343
344         if (!dev_name || !*dev_name)
345                 return -EINVAL;
346
347         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
348         if (!fsopt)
349                 return -ENOMEM;
350
351         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
352
353         fsopt->sb_flags = flags;
354         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
355
356         fsopt->rsize = CEPH_RSIZE_DEFAULT;
357         fsopt->rasize = CEPH_RASIZE_DEFAULT;
358         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
359         if (!fsopt->snapdir_name) {
360                 err = -ENOMEM;
361                 goto out;
362         }
363
364         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
365         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
366         fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
367         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
368         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
369         fsopt->congestion_kb = default_congestion_kb();
370
371         /*
372          * Distinguish the server list from the path in "dev_name".
373          * Internally we do not include the leading '/' in the path.
374          *
375          * "dev_name" will look like:
376          *     <server_spec>[,<server_spec>...]:[<path>]
377          * where
378          *     <server_spec> is <ip>[:<port>]
379          *     <path> is optional, but if present must begin with '/'
380          */
381         dev_name_end = strchr(dev_name, '/');
382         if (dev_name_end) {
383                 /* skip over leading '/' for path */
384                 *path = dev_name_end + 1;
385         } else {
386                 /* path is empty */
387                 dev_name_end = dev_name + strlen(dev_name);
388                 *path = dev_name_end;
389         }
390         err = -EINVAL;
391         dev_name_end--;         /* back up to ':' separator */
392         if (dev_name_end < dev_name || *dev_name_end != ':') {
393                 pr_err("device name is missing path (no : separator in %s)\n",
394                                 dev_name);
395                 goto out;
396         }
397         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
398         dout("server path '%s'\n", *path);
399
400         *popt = ceph_parse_options(options, dev_name, dev_name_end,
401                                  parse_fsopt_token, (void *)fsopt);
402         if (IS_ERR(*popt)) {
403                 err = PTR_ERR(*popt);
404                 goto out;
405         }
406
407         /* success */
408         *pfsopt = fsopt;
409         return 0;
410
411 out:
412         destroy_mount_options(fsopt);
413         return err;
414 }
415
416 /**
417  * ceph_show_options - Show mount options in /proc/mounts
418  * @m: seq_file to write to
419  * @root: root of that (sub)tree
420  */
421 static int ceph_show_options(struct seq_file *m, struct dentry *root)
422 {
423         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
424         struct ceph_mount_options *fsopt = fsc->mount_options;
425         size_t pos;
426         int ret;
427
428         /* a comma between MNT/MS and client options */
429         seq_putc(m, ',');
430         pos = m->count;
431
432         ret = ceph_print_client_options(m, fsc->client);
433         if (ret)
434                 return ret;
435
436         /* retract our comma if no client options */
437         if (m->count == pos)
438                 m->count--;
439
440         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
441                 seq_puts(m, ",dirstat");
442         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
443                 seq_puts(m, ",norbytes");
444         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
445                 seq_puts(m, ",noasyncreaddir");
446         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
447                 seq_puts(m, ",nodcache");
448         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
449                 seq_puts(m, ",fsc");
450         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
451                 seq_puts(m, ",nopoolperm");
452
453 #ifdef CONFIG_CEPH_FS_POSIX_ACL
454         if (fsopt->sb_flags & MS_POSIXACL)
455                 seq_puts(m, ",acl");
456         else
457                 seq_puts(m, ",noacl");
458 #endif
459
460         if (fsopt->wsize)
461                 seq_printf(m, ",wsize=%d", fsopt->wsize);
462         if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
463                 seq_printf(m, ",rsize=%d", fsopt->rsize);
464         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
465                 seq_printf(m, ",rasize=%d", fsopt->rasize);
466         if (fsopt->congestion_kb != default_congestion_kb())
467                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
468         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
469                 seq_printf(m, ",caps_wanted_delay_min=%d",
470                          fsopt->caps_wanted_delay_min);
471         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
472                 seq_printf(m, ",caps_wanted_delay_max=%d",
473                            fsopt->caps_wanted_delay_max);
474         if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
475                 seq_printf(m, ",cap_release_safety=%d",
476                            fsopt->cap_release_safety);
477         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
478                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
479         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
480                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
481         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
482                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
483
484         return 0;
485 }
486
487 /*
488  * handle any mon messages the standard library doesn't understand.
489  * return error if we don't either.
490  */
491 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
492 {
493         struct ceph_fs_client *fsc = client->private;
494         int type = le16_to_cpu(msg->hdr.type);
495
496         switch (type) {
497         case CEPH_MSG_MDS_MAP:
498                 ceph_mdsc_handle_map(fsc->mdsc, msg);
499                 return 0;
500
501         default:
502                 return -1;
503         }
504 }
505
506 /*
507  * create a new fs client
508  */
509 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
510                                         struct ceph_options *opt)
511 {
512         struct ceph_fs_client *fsc;
513         const u64 supported_features =
514                 CEPH_FEATURE_FLOCK |
515                 CEPH_FEATURE_DIRLAYOUTHASH |
516                 CEPH_FEATURE_MDS_INLINE_DATA;
517         const u64 required_features = 0;
518         int page_count;
519         size_t size;
520         int err = -ENOMEM;
521
522         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
523         if (!fsc)
524                 return ERR_PTR(-ENOMEM);
525
526         fsc->client = ceph_create_client(opt, fsc, supported_features,
527                                          required_features);
528         if (IS_ERR(fsc->client)) {
529                 err = PTR_ERR(fsc->client);
530                 goto fail;
531         }
532         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
533         fsc->client->monc.want_mdsmap = 1;
534
535         fsc->mount_options = fsopt;
536
537         fsc->sb = NULL;
538         fsc->mount_state = CEPH_MOUNT_MOUNTING;
539
540         atomic_long_set(&fsc->writeback_count, 0);
541
542         err = bdi_init(&fsc->backing_dev_info);
543         if (err < 0)
544                 goto fail_client;
545
546         err = -ENOMEM;
547         /*
548          * The number of concurrent works can be high but they don't need
549          * to be processed in parallel, limit concurrency.
550          */
551         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
552         if (fsc->wb_wq == NULL)
553                 goto fail_bdi;
554         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
555         if (fsc->pg_inv_wq == NULL)
556                 goto fail_wb_wq;
557         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
558         if (fsc->trunc_wq == NULL)
559                 goto fail_pg_inv_wq;
560
561         /* set up mempools */
562         err = -ENOMEM;
563         page_count = fsc->mount_options->wsize >> PAGE_CACHE_SHIFT;
564         size = sizeof (struct page *) * (page_count ? page_count : 1);
565         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
566         if (!fsc->wb_pagevec_pool)
567                 goto fail_trunc_wq;
568
569         /* setup fscache */
570         if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
571             (ceph_fscache_register_fs(fsc) != 0))
572                 goto fail_fscache;
573
574         /* caps */
575         fsc->min_caps = fsopt->max_readdir;
576
577         return fsc;
578
579 fail_fscache:
580         ceph_fscache_unregister_fs(fsc);
581 fail_trunc_wq:
582         destroy_workqueue(fsc->trunc_wq);
583 fail_pg_inv_wq:
584         destroy_workqueue(fsc->pg_inv_wq);
585 fail_wb_wq:
586         destroy_workqueue(fsc->wb_wq);
587 fail_bdi:
588         bdi_destroy(&fsc->backing_dev_info);
589 fail_client:
590         ceph_destroy_client(fsc->client);
591 fail:
592         kfree(fsc);
593         return ERR_PTR(err);
594 }
595
596 static void destroy_fs_client(struct ceph_fs_client *fsc)
597 {
598         dout("destroy_fs_client %p\n", fsc);
599
600         ceph_fscache_unregister_fs(fsc);
601
602         destroy_workqueue(fsc->wb_wq);
603         destroy_workqueue(fsc->pg_inv_wq);
604         destroy_workqueue(fsc->trunc_wq);
605
606         bdi_destroy(&fsc->backing_dev_info);
607
608         mempool_destroy(fsc->wb_pagevec_pool);
609
610         destroy_mount_options(fsc->mount_options);
611
612         ceph_fs_debugfs_cleanup(fsc);
613
614         ceph_destroy_client(fsc->client);
615
616         kfree(fsc);
617         dout("destroy_fs_client %p done\n", fsc);
618 }
619
620 /*
621  * caches
622  */
623 struct kmem_cache *ceph_inode_cachep;
624 struct kmem_cache *ceph_cap_cachep;
625 struct kmem_cache *ceph_cap_flush_cachep;
626 struct kmem_cache *ceph_dentry_cachep;
627 struct kmem_cache *ceph_file_cachep;
628
629 static void ceph_inode_init_once(void *foo)
630 {
631         struct ceph_inode_info *ci = foo;
632         inode_init_once(&ci->vfs_inode);
633 }
634
635 static int __init init_caches(void)
636 {
637         int error = -ENOMEM;
638
639         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
640                                       sizeof(struct ceph_inode_info),
641                                       __alignof__(struct ceph_inode_info),
642                                       (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
643                                       ceph_inode_init_once);
644         if (ceph_inode_cachep == NULL)
645                 return -ENOMEM;
646
647         ceph_cap_cachep = KMEM_CACHE(ceph_cap,
648                                      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
649         if (ceph_cap_cachep == NULL)
650                 goto bad_cap;
651         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
652                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
653         if (ceph_cap_flush_cachep == NULL)
654                 goto bad_cap_flush;
655
656         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
657                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
658         if (ceph_dentry_cachep == NULL)
659                 goto bad_dentry;
660
661         ceph_file_cachep = KMEM_CACHE(ceph_file_info,
662                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
663         if (ceph_file_cachep == NULL)
664                 goto bad_file;
665
666         if ((error = ceph_fscache_register()))
667                 goto bad_file;
668
669         return 0;
670 bad_file:
671         kmem_cache_destroy(ceph_dentry_cachep);
672 bad_dentry:
673         kmem_cache_destroy(ceph_cap_flush_cachep);
674 bad_cap_flush:
675         kmem_cache_destroy(ceph_cap_cachep);
676 bad_cap:
677         kmem_cache_destroy(ceph_inode_cachep);
678         return error;
679 }
680
681 static void destroy_caches(void)
682 {
683         /*
684          * Make sure all delayed rcu free inodes are flushed before we
685          * destroy cache.
686          */
687         rcu_barrier();
688
689         kmem_cache_destroy(ceph_inode_cachep);
690         kmem_cache_destroy(ceph_cap_cachep);
691         kmem_cache_destroy(ceph_cap_flush_cachep);
692         kmem_cache_destroy(ceph_dentry_cachep);
693         kmem_cache_destroy(ceph_file_cachep);
694
695         ceph_fscache_unregister();
696 }
697
698
699 /*
700  * ceph_umount_begin - initiate forced umount.  Tear down down the
701  * mount, skipping steps that may hang while waiting for server(s).
702  */
703 static void ceph_umount_begin(struct super_block *sb)
704 {
705         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
706
707         dout("ceph_umount_begin - starting forced umount\n");
708         if (!fsc)
709                 return;
710         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
711         ceph_mdsc_force_umount(fsc->mdsc);
712         return;
713 }
714
715 static const struct super_operations ceph_super_ops = {
716         .alloc_inode    = ceph_alloc_inode,
717         .destroy_inode  = ceph_destroy_inode,
718         .write_inode    = ceph_write_inode,
719         .drop_inode     = ceph_drop_inode,
720         .sync_fs        = ceph_sync_fs,
721         .put_super      = ceph_put_super,
722         .show_options   = ceph_show_options,
723         .statfs         = ceph_statfs,
724         .umount_begin   = ceph_umount_begin,
725 };
726
727 /*
728  * Bootstrap mount by opening the root directory.  Note the mount
729  * @started time from caller, and time out if this takes too long.
730  */
731 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
732                                        const char *path,
733                                        unsigned long started)
734 {
735         struct ceph_mds_client *mdsc = fsc->mdsc;
736         struct ceph_mds_request *req = NULL;
737         int err;
738         struct dentry *root;
739
740         /* open dir */
741         dout("open_root_inode opening '%s'\n", path);
742         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
743         if (IS_ERR(req))
744                 return ERR_CAST(req);
745         req->r_path1 = kstrdup(path, GFP_NOFS);
746         if (!req->r_path1) {
747                 root = ERR_PTR(-ENOMEM);
748                 goto out;
749         }
750
751         req->r_ino1.ino = CEPH_INO_ROOT;
752         req->r_ino1.snap = CEPH_NOSNAP;
753         req->r_started = started;
754         req->r_timeout = fsc->client->options->mount_timeout;
755         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
756         req->r_num_caps = 2;
757         err = ceph_mdsc_do_request(mdsc, NULL, req);
758         if (err == 0) {
759                 struct inode *inode = req->r_target_inode;
760                 req->r_target_inode = NULL;
761                 dout("open_root_inode success\n");
762                 if (ceph_ino(inode) == CEPH_INO_ROOT &&
763                     fsc->sb->s_root == NULL) {
764                         root = d_make_root(inode);
765                         if (!root) {
766                                 root = ERR_PTR(-ENOMEM);
767                                 goto out;
768                         }
769                 } else {
770                         root = d_obtain_root(inode);
771                 }
772                 ceph_init_dentry(root);
773                 dout("open_root_inode success, root dentry is %p\n", root);
774         } else {
775                 root = ERR_PTR(err);
776         }
777 out:
778         ceph_mdsc_put_request(req);
779         return root;
780 }
781
782
783
784
785 /*
786  * mount: join the ceph cluster, and open root directory.
787  */
788 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
789                       const char *path)
790 {
791         int err;
792         unsigned long started = jiffies;  /* note the start time */
793         struct dentry *root;
794         int first = 0;   /* first vfsmount for this super_block */
795
796         dout("mount start\n");
797         mutex_lock(&fsc->client->mount_mutex);
798
799         err = __ceph_open_session(fsc->client, started);
800         if (err < 0)
801                 goto out;
802
803         dout("mount opening root\n");
804         root = open_root_dentry(fsc, "", started);
805         if (IS_ERR(root)) {
806                 err = PTR_ERR(root);
807                 goto out;
808         }
809         if (fsc->sb->s_root) {
810                 dput(root);
811         } else {
812                 fsc->sb->s_root = root;
813                 first = 1;
814
815                 err = ceph_fs_debugfs_init(fsc);
816                 if (err < 0)
817                         goto fail;
818         }
819
820         if (path[0] == 0) {
821                 dget(root);
822         } else {
823                 dout("mount opening base mountpoint\n");
824                 root = open_root_dentry(fsc, path, started);
825                 if (IS_ERR(root)) {
826                         err = PTR_ERR(root);
827                         goto fail;
828                 }
829         }
830
831         fsc->mount_state = CEPH_MOUNT_MOUNTED;
832         dout("mount success\n");
833         mutex_unlock(&fsc->client->mount_mutex);
834         return root;
835
836 out:
837         mutex_unlock(&fsc->client->mount_mutex);
838         return ERR_PTR(err);
839
840 fail:
841         if (first) {
842                 dput(fsc->sb->s_root);
843                 fsc->sb->s_root = NULL;
844         }
845         goto out;
846 }
847
848 static int ceph_set_super(struct super_block *s, void *data)
849 {
850         struct ceph_fs_client *fsc = data;
851         int ret;
852
853         dout("set_super %p data %p\n", s, data);
854
855         s->s_flags = fsc->mount_options->sb_flags;
856         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
857
858         s->s_xattr = ceph_xattr_handlers;
859         s->s_fs_info = fsc;
860         fsc->sb = s;
861
862         s->s_op = &ceph_super_ops;
863         s->s_export_op = &ceph_export_ops;
864
865         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
866
867         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
868         if (ret != 0)
869                 goto fail;
870
871         return ret;
872
873 fail:
874         s->s_fs_info = NULL;
875         fsc->sb = NULL;
876         return ret;
877 }
878
879 /*
880  * share superblock if same fs AND options
881  */
882 static int ceph_compare_super(struct super_block *sb, void *data)
883 {
884         struct ceph_fs_client *new = data;
885         struct ceph_mount_options *fsopt = new->mount_options;
886         struct ceph_options *opt = new->client->options;
887         struct ceph_fs_client *other = ceph_sb_to_client(sb);
888
889         dout("ceph_compare_super %p\n", sb);
890
891         if (compare_mount_options(fsopt, opt, other)) {
892                 dout("monitor(s)/mount options don't match\n");
893                 return 0;
894         }
895         if ((opt->flags & CEPH_OPT_FSID) &&
896             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
897                 dout("fsid doesn't match\n");
898                 return 0;
899         }
900         if (fsopt->sb_flags != other->mount_options->sb_flags) {
901                 dout("flags differ\n");
902                 return 0;
903         }
904         return 1;
905 }
906
907 /*
908  * construct our own bdi so we can control readahead, etc.
909  */
910 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
911
912 static int ceph_register_bdi(struct super_block *sb,
913                              struct ceph_fs_client *fsc)
914 {
915         int err;
916
917         /* set ra_pages based on rasize mount option? */
918         if (fsc->mount_options->rasize >= PAGE_CACHE_SIZE)
919                 fsc->backing_dev_info.ra_pages =
920                         (fsc->mount_options->rasize + PAGE_CACHE_SIZE - 1)
921                         >> PAGE_SHIFT;
922         else
923                 fsc->backing_dev_info.ra_pages =
924                         VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
925
926         err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
927                            atomic_long_inc_return(&bdi_seq));
928         if (!err)
929                 sb->s_bdi = &fsc->backing_dev_info;
930         return err;
931 }
932
933 static struct dentry *ceph_mount(struct file_system_type *fs_type,
934                        int flags, const char *dev_name, void *data)
935 {
936         struct super_block *sb;
937         struct ceph_fs_client *fsc;
938         struct dentry *res;
939         int err;
940         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
941         const char *path = NULL;
942         struct ceph_mount_options *fsopt = NULL;
943         struct ceph_options *opt = NULL;
944
945         dout("ceph_mount\n");
946
947 #ifdef CONFIG_CEPH_FS_POSIX_ACL
948         flags |= MS_POSIXACL;
949 #endif
950         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
951         if (err < 0) {
952                 res = ERR_PTR(err);
953                 goto out_final;
954         }
955
956         /* create client (which we may/may not use) */
957         fsc = create_fs_client(fsopt, opt);
958         if (IS_ERR(fsc)) {
959                 res = ERR_CAST(fsc);
960                 destroy_mount_options(fsopt);
961                 ceph_destroy_options(opt);
962                 goto out_final;
963         }
964
965         err = ceph_mdsc_init(fsc);
966         if (err < 0) {
967                 res = ERR_PTR(err);
968                 goto out;
969         }
970
971         if (ceph_test_opt(fsc->client, NOSHARE))
972                 compare_super = NULL;
973         sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
974         if (IS_ERR(sb)) {
975                 res = ERR_CAST(sb);
976                 goto out;
977         }
978
979         if (ceph_sb_to_client(sb) != fsc) {
980                 ceph_mdsc_destroy(fsc);
981                 destroy_fs_client(fsc);
982                 fsc = ceph_sb_to_client(sb);
983                 dout("get_sb got existing client %p\n", fsc);
984         } else {
985                 dout("get_sb using new client %p\n", fsc);
986                 err = ceph_register_bdi(sb, fsc);
987                 if (err < 0) {
988                         res = ERR_PTR(err);
989                         goto out_splat;
990                 }
991         }
992
993         res = ceph_real_mount(fsc, path);
994         if (IS_ERR(res))
995                 goto out_splat;
996         dout("root %p inode %p ino %llx.%llx\n", res,
997              d_inode(res), ceph_vinop(d_inode(res)));
998         return res;
999
1000 out_splat:
1001         ceph_mdsc_close_sessions(fsc->mdsc);
1002         deactivate_locked_super(sb);
1003         goto out_final;
1004
1005 out:
1006         ceph_mdsc_destroy(fsc);
1007         destroy_fs_client(fsc);
1008 out_final:
1009         dout("ceph_mount fail %ld\n", PTR_ERR(res));
1010         return res;
1011 }
1012
1013 static void ceph_kill_sb(struct super_block *s)
1014 {
1015         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1016         dev_t dev = s->s_dev;
1017
1018         dout("kill_sb %p\n", s);
1019
1020         ceph_mdsc_pre_umount(fsc->mdsc);
1021         generic_shutdown_super(s);
1022         ceph_mdsc_destroy(fsc);
1023
1024         destroy_fs_client(fsc);
1025         free_anon_bdev(dev);
1026 }
1027
1028 static struct file_system_type ceph_fs_type = {
1029         .owner          = THIS_MODULE,
1030         .name           = "ceph",
1031         .mount          = ceph_mount,
1032         .kill_sb        = ceph_kill_sb,
1033         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1034 };
1035 MODULE_ALIAS_FS("ceph");
1036
1037 static int __init init_ceph(void)
1038 {
1039         int ret = init_caches();
1040         if (ret)
1041                 goto out;
1042
1043         ceph_flock_init();
1044         ceph_xattr_init();
1045         ret = ceph_snap_init();
1046         if (ret)
1047                 goto out_xattr;
1048         ret = register_filesystem(&ceph_fs_type);
1049         if (ret)
1050                 goto out_snap;
1051
1052         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1053
1054         return 0;
1055
1056 out_snap:
1057         ceph_snap_exit();
1058 out_xattr:
1059         ceph_xattr_exit();
1060         destroy_caches();
1061 out:
1062         return ret;
1063 }
1064
1065 static void __exit exit_ceph(void)
1066 {
1067         dout("exit_ceph\n");
1068         unregister_filesystem(&ceph_fs_type);
1069         ceph_snap_exit();
1070         ceph_xattr_exit();
1071         destroy_caches();
1072 }
1073
1074 module_init(init_ceph);
1075 module_exit(exit_ceph);
1076
1077 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1078 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1079 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1080 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1081 MODULE_LICENSE("GPL");