cec1d3343742d3725cd12c49f29810f0555bcb9b
[sfrench/cifs-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 #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_mon_client *monc = &fsc->client->monc;
49         struct ceph_statfs st;
50         u64 fsid;
51         int err;
52         u64 data_pool;
53
54         if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
55                 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
56         } else {
57                 data_pool = CEPH_NOPOOL;
58         }
59
60         dout("statfs\n");
61         err = ceph_monc_do_statfs(monc, data_pool, &st);
62         if (err < 0)
63                 return err;
64
65         /* fill in kstatfs */
66         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
67
68         /*
69          * express utilization in terms of large blocks to avoid
70          * overflow on 32-bit machines.
71          *
72          * NOTE: for the time being, we make bsize == frsize to humor
73          * not-yet-ancient versions of glibc that are broken.
74          * Someday, we will probably want to report a real block
75          * size...  whatever that may mean for a network file system!
76          */
77         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
78         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
79
80         /*
81          * By default use root quota for stats; fallback to overall filesystem
82          * usage if using 'noquotadf' mount option or if the root dir doesn't
83          * have max_bytes quota set.
84          */
85         if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
86             !ceph_quota_update_statfs(fsc, buf)) {
87                 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
88                 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
89                 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
90         }
91
92         buf->f_files = le64_to_cpu(st.num_objects);
93         buf->f_ffree = -1;
94         buf->f_namelen = NAME_MAX;
95
96         /* Must convert the fsid, for consistent values across arches */
97         mutex_lock(&monc->mutex);
98         fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
99                le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
100         mutex_unlock(&monc->mutex);
101
102         buf->f_fsid.val[0] = fsid & 0xffffffff;
103         buf->f_fsid.val[1] = fsid >> 32;
104
105         return 0;
106 }
107
108
109 static int ceph_sync_fs(struct super_block *sb, int wait)
110 {
111         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
112
113         if (!wait) {
114                 dout("sync_fs (non-blocking)\n");
115                 ceph_flush_dirty_caps(fsc->mdsc);
116                 dout("sync_fs (non-blocking) done\n");
117                 return 0;
118         }
119
120         dout("sync_fs (blocking)\n");
121         ceph_osdc_sync(&fsc->client->osdc);
122         ceph_mdsc_sync(fsc->mdsc);
123         dout("sync_fs (blocking) done\n");
124         return 0;
125 }
126
127 /*
128  * mount options
129  */
130 enum {
131         Opt_wsize,
132         Opt_rsize,
133         Opt_rasize,
134         Opt_caps_wanted_delay_min,
135         Opt_caps_wanted_delay_max,
136         Opt_readdir_max_entries,
137         Opt_readdir_max_bytes,
138         Opt_congestion_kb,
139         Opt_last_int,
140         /* int args above */
141         Opt_snapdirname,
142         Opt_mds_namespace,
143         Opt_fscache_uniq,
144         Opt_last_string,
145         /* string args above */
146         Opt_dirstat,
147         Opt_nodirstat,
148         Opt_rbytes,
149         Opt_norbytes,
150         Opt_asyncreaddir,
151         Opt_noasyncreaddir,
152         Opt_dcache,
153         Opt_nodcache,
154         Opt_ino32,
155         Opt_noino32,
156         Opt_fscache,
157         Opt_nofscache,
158         Opt_poolperm,
159         Opt_nopoolperm,
160         Opt_require_active_mds,
161         Opt_norequire_active_mds,
162 #ifdef CONFIG_CEPH_FS_POSIX_ACL
163         Opt_acl,
164 #endif
165         Opt_noacl,
166         Opt_quotadf,
167         Opt_noquotadf,
168 };
169
170 static match_table_t fsopt_tokens = {
171         {Opt_wsize, "wsize=%d"},
172         {Opt_rsize, "rsize=%d"},
173         {Opt_rasize, "rasize=%d"},
174         {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
175         {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
176         {Opt_readdir_max_entries, "readdir_max_entries=%d"},
177         {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
178         {Opt_congestion_kb, "write_congestion_kb=%d"},
179         /* int args above */
180         {Opt_snapdirname, "snapdirname=%s"},
181         {Opt_mds_namespace, "mds_namespace=%s"},
182         {Opt_fscache_uniq, "fsc=%s"},
183         /* string args above */
184         {Opt_dirstat, "dirstat"},
185         {Opt_nodirstat, "nodirstat"},
186         {Opt_rbytes, "rbytes"},
187         {Opt_norbytes, "norbytes"},
188         {Opt_asyncreaddir, "asyncreaddir"},
189         {Opt_noasyncreaddir, "noasyncreaddir"},
190         {Opt_dcache, "dcache"},
191         {Opt_nodcache, "nodcache"},
192         {Opt_ino32, "ino32"},
193         {Opt_noino32, "noino32"},
194         {Opt_fscache, "fsc"},
195         {Opt_nofscache, "nofsc"},
196         {Opt_poolperm, "poolperm"},
197         {Opt_nopoolperm, "nopoolperm"},
198         {Opt_require_active_mds, "require_active_mds"},
199         {Opt_norequire_active_mds, "norequire_active_mds"},
200 #ifdef CONFIG_CEPH_FS_POSIX_ACL
201         {Opt_acl, "acl"},
202 #endif
203         {Opt_noacl, "noacl"},
204         {Opt_quotadf, "quotadf"},
205         {Opt_noquotadf, "noquotadf"},
206         {-1, NULL}
207 };
208
209 static int parse_fsopt_token(char *c, void *private)
210 {
211         struct ceph_mount_options *fsopt = private;
212         substring_t argstr[MAX_OPT_ARGS];
213         int token, intval, ret;
214
215         token = match_token((char *)c, fsopt_tokens, argstr);
216         if (token < 0)
217                 return -EINVAL;
218
219         if (token < Opt_last_int) {
220                 ret = match_int(&argstr[0], &intval);
221                 if (ret < 0) {
222                         pr_err("bad mount option arg (not int) "
223                                "at '%s'\n", c);
224                         return ret;
225                 }
226                 dout("got int token %d val %d\n", token, intval);
227         } else if (token > Opt_last_int && token < Opt_last_string) {
228                 dout("got string token %d val %s\n", token,
229                      argstr[0].from);
230         } else {
231                 dout("got token %d\n", token);
232         }
233
234         switch (token) {
235         case Opt_snapdirname:
236                 kfree(fsopt->snapdir_name);
237                 fsopt->snapdir_name = kstrndup(argstr[0].from,
238                                                argstr[0].to-argstr[0].from,
239                                                GFP_KERNEL);
240                 if (!fsopt->snapdir_name)
241                         return -ENOMEM;
242                 break;
243         case Opt_mds_namespace:
244                 kfree(fsopt->mds_namespace);
245                 fsopt->mds_namespace = kstrndup(argstr[0].from,
246                                                 argstr[0].to-argstr[0].from,
247                                                 GFP_KERNEL);
248                 if (!fsopt->mds_namespace)
249                         return -ENOMEM;
250                 break;
251         case Opt_fscache_uniq:
252                 kfree(fsopt->fscache_uniq);
253                 fsopt->fscache_uniq = kstrndup(argstr[0].from,
254                                                argstr[0].to-argstr[0].from,
255                                                GFP_KERNEL);
256                 if (!fsopt->fscache_uniq)
257                         return -ENOMEM;
258                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
259                 break;
260                 /* misc */
261         case Opt_wsize:
262                 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
263                         return -EINVAL;
264                 fsopt->wsize = ALIGN(intval, PAGE_SIZE);
265                 break;
266         case Opt_rsize:
267                 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
268                         return -EINVAL;
269                 fsopt->rsize = ALIGN(intval, PAGE_SIZE);
270                 break;
271         case Opt_rasize:
272                 if (intval < 0)
273                         return -EINVAL;
274                 fsopt->rasize = ALIGN(intval, PAGE_SIZE);
275                 break;
276         case Opt_caps_wanted_delay_min:
277                 if (intval < 1)
278                         return -EINVAL;
279                 fsopt->caps_wanted_delay_min = intval;
280                 break;
281         case Opt_caps_wanted_delay_max:
282                 if (intval < 1)
283                         return -EINVAL;
284                 fsopt->caps_wanted_delay_max = intval;
285                 break;
286         case Opt_readdir_max_entries:
287                 if (intval < 1)
288                         return -EINVAL;
289                 fsopt->max_readdir = intval;
290                 break;
291         case Opt_readdir_max_bytes:
292                 if (intval < (int)PAGE_SIZE && intval != 0)
293                         return -EINVAL;
294                 fsopt->max_readdir_bytes = intval;
295                 break;
296         case Opt_congestion_kb:
297                 if (intval < 1024) /* at least 1M */
298                         return -EINVAL;
299                 fsopt->congestion_kb = intval;
300                 break;
301         case Opt_dirstat:
302                 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
303                 break;
304         case Opt_nodirstat:
305                 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
306                 break;
307         case Opt_rbytes:
308                 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
309                 break;
310         case Opt_norbytes:
311                 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
312                 break;
313         case Opt_asyncreaddir:
314                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
315                 break;
316         case Opt_noasyncreaddir:
317                 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
318                 break;
319         case Opt_dcache:
320                 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
321                 break;
322         case Opt_nodcache:
323                 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
324                 break;
325         case Opt_ino32:
326                 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
327                 break;
328         case Opt_noino32:
329                 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
330                 break;
331         case Opt_fscache:
332                 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
333                 kfree(fsopt->fscache_uniq);
334                 fsopt->fscache_uniq = NULL;
335                 break;
336         case Opt_nofscache:
337                 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
338                 kfree(fsopt->fscache_uniq);
339                 fsopt->fscache_uniq = NULL;
340                 break;
341         case Opt_poolperm:
342                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
343                 break;
344         case Opt_nopoolperm:
345                 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
346                 break;
347         case Opt_require_active_mds:
348                 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
349                 break;
350         case Opt_norequire_active_mds:
351                 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
352                 break;
353         case Opt_quotadf:
354                 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
355                 break;
356         case Opt_noquotadf:
357                 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
358                 break;
359 #ifdef CONFIG_CEPH_FS_POSIX_ACL
360         case Opt_acl:
361                 fsopt->sb_flags |= SB_POSIXACL;
362                 break;
363 #endif
364         case Opt_noacl:
365                 fsopt->sb_flags &= ~SB_POSIXACL;
366                 break;
367         default:
368                 BUG_ON(token);
369         }
370         return 0;
371 }
372
373 static void destroy_mount_options(struct ceph_mount_options *args)
374 {
375         dout("destroy_mount_options %p\n", args);
376         kfree(args->snapdir_name);
377         kfree(args->mds_namespace);
378         kfree(args->server_path);
379         kfree(args->fscache_uniq);
380         kfree(args);
381 }
382
383 static int strcmp_null(const char *s1, const char *s2)
384 {
385         if (!s1 && !s2)
386                 return 0;
387         if (s1 && !s2)
388                 return -1;
389         if (!s1 && s2)
390                 return 1;
391         return strcmp(s1, s2);
392 }
393
394 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
395                                  struct ceph_options *new_opt,
396                                  struct ceph_fs_client *fsc)
397 {
398         struct ceph_mount_options *fsopt1 = new_fsopt;
399         struct ceph_mount_options *fsopt2 = fsc->mount_options;
400         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
401         int ret;
402
403         ret = memcmp(fsopt1, fsopt2, ofs);
404         if (ret)
405                 return ret;
406
407         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
408         if (ret)
409                 return ret;
410         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
411         if (ret)
412                 return ret;
413         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
414         if (ret)
415                 return ret;
416         ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
417         if (ret)
418                 return ret;
419
420         return ceph_compare_options(new_opt, fsc->client);
421 }
422
423 static int parse_mount_options(struct ceph_mount_options **pfsopt,
424                                struct ceph_options **popt,
425                                int flags, char *options,
426                                const char *dev_name)
427 {
428         struct ceph_mount_options *fsopt;
429         const char *dev_name_end;
430         int err;
431
432         if (!dev_name || !*dev_name)
433                 return -EINVAL;
434
435         fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
436         if (!fsopt)
437                 return -ENOMEM;
438
439         dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
440
441         fsopt->sb_flags = flags;
442         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
443
444         fsopt->wsize = CEPH_MAX_WRITE_SIZE;
445         fsopt->rsize = CEPH_MAX_READ_SIZE;
446         fsopt->rasize = CEPH_RASIZE_DEFAULT;
447         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
448         if (!fsopt->snapdir_name) {
449                 err = -ENOMEM;
450                 goto out;
451         }
452
453         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
454         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
455         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
456         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
457         fsopt->congestion_kb = default_congestion_kb();
458
459         /*
460          * Distinguish the server list from the path in "dev_name".
461          * Internally we do not include the leading '/' in the path.
462          *
463          * "dev_name" will look like:
464          *     <server_spec>[,<server_spec>...]:[<path>]
465          * where
466          *     <server_spec> is <ip>[:<port>]
467          *     <path> is optional, but if present must begin with '/'
468          */
469         dev_name_end = strchr(dev_name, '/');
470         if (dev_name_end) {
471                 if (strlen(dev_name_end) > 1) {
472                         fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
473                         if (!fsopt->server_path) {
474                                 err = -ENOMEM;
475                                 goto out;
476                         }
477                 }
478         } else {
479                 dev_name_end = dev_name + strlen(dev_name);
480         }
481         err = -EINVAL;
482         dev_name_end--;         /* back up to ':' separator */
483         if (dev_name_end < dev_name || *dev_name_end != ':') {
484                 pr_err("device name is missing path (no : separator in %s)\n",
485                                 dev_name);
486                 goto out;
487         }
488         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
489         if (fsopt->server_path)
490                 dout("server path '%s'\n", fsopt->server_path);
491
492         *popt = ceph_parse_options(options, dev_name, dev_name_end,
493                                  parse_fsopt_token, (void *)fsopt);
494         if (IS_ERR(*popt)) {
495                 err = PTR_ERR(*popt);
496                 goto out;
497         }
498
499         /* success */
500         *pfsopt = fsopt;
501         return 0;
502
503 out:
504         destroy_mount_options(fsopt);
505         return err;
506 }
507
508 /**
509  * ceph_show_options - Show mount options in /proc/mounts
510  * @m: seq_file to write to
511  * @root: root of that (sub)tree
512  */
513 static int ceph_show_options(struct seq_file *m, struct dentry *root)
514 {
515         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
516         struct ceph_mount_options *fsopt = fsc->mount_options;
517         size_t pos;
518         int ret;
519
520         /* a comma between MNT/MS and client options */
521         seq_putc(m, ',');
522         pos = m->count;
523
524         ret = ceph_print_client_options(m, fsc->client);
525         if (ret)
526                 return ret;
527
528         /* retract our comma if no client options */
529         if (m->count == pos)
530                 m->count--;
531
532         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
533                 seq_puts(m, ",dirstat");
534         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
535                 seq_puts(m, ",rbytes");
536         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
537                 seq_puts(m, ",noasyncreaddir");
538         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
539                 seq_puts(m, ",nodcache");
540         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
541                 seq_show_option(m, "fsc", fsopt->fscache_uniq);
542         }
543         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
544                 seq_puts(m, ",nopoolperm");
545         if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
546                 seq_puts(m, ",noquotadf");
547
548 #ifdef CONFIG_CEPH_FS_POSIX_ACL
549         if (fsopt->sb_flags & SB_POSIXACL)
550                 seq_puts(m, ",acl");
551         else
552                 seq_puts(m, ",noacl");
553 #endif
554
555         if (fsopt->mds_namespace)
556                 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
557         if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
558                 seq_printf(m, ",wsize=%d", fsopt->wsize);
559         if (fsopt->rsize != CEPH_MAX_READ_SIZE)
560                 seq_printf(m, ",rsize=%d", fsopt->rsize);
561         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
562                 seq_printf(m, ",rasize=%d", fsopt->rasize);
563         if (fsopt->congestion_kb != default_congestion_kb())
564                 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
565         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
566                 seq_printf(m, ",caps_wanted_delay_min=%d",
567                          fsopt->caps_wanted_delay_min);
568         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
569                 seq_printf(m, ",caps_wanted_delay_max=%d",
570                            fsopt->caps_wanted_delay_max);
571         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
572                 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
573         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
574                 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
575         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
576                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
577
578         return 0;
579 }
580
581 /*
582  * handle any mon messages the standard library doesn't understand.
583  * return error if we don't either.
584  */
585 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
586 {
587         struct ceph_fs_client *fsc = client->private;
588         int type = le16_to_cpu(msg->hdr.type);
589
590         switch (type) {
591         case CEPH_MSG_MDS_MAP:
592                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
593                 return 0;
594         case CEPH_MSG_FS_MAP_USER:
595                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
596                 return 0;
597         default:
598                 return -1;
599         }
600 }
601
602 /*
603  * create a new fs client
604  */
605 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
606                                         struct ceph_options *opt)
607 {
608         struct ceph_fs_client *fsc;
609         int page_count;
610         size_t size;
611         int err = -ENOMEM;
612
613         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
614         if (!fsc)
615                 return ERR_PTR(-ENOMEM);
616
617         fsc->client = ceph_create_client(opt, fsc);
618         if (IS_ERR(fsc->client)) {
619                 err = PTR_ERR(fsc->client);
620                 goto fail;
621         }
622
623         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
624         fsc->client->osdc.abort_on_full = true;
625
626         if (!fsopt->mds_namespace) {
627                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
628                                    0, true);
629         } else {
630                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
631                                    0, false);
632         }
633
634         fsc->mount_options = fsopt;
635
636         fsc->sb = NULL;
637         fsc->mount_state = CEPH_MOUNT_MOUNTING;
638
639         atomic_long_set(&fsc->writeback_count, 0);
640
641         err = -ENOMEM;
642         /*
643          * The number of concurrent works can be high but they don't need
644          * to be processed in parallel, limit concurrency.
645          */
646         fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
647         if (!fsc->wb_wq)
648                 goto fail_client;
649         fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
650         if (!fsc->pg_inv_wq)
651                 goto fail_wb_wq;
652         fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
653         if (!fsc->trunc_wq)
654                 goto fail_pg_inv_wq;
655
656         /* set up mempools */
657         err = -ENOMEM;
658         page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
659         size = sizeof (struct page *) * (page_count ? page_count : 1);
660         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
661         if (!fsc->wb_pagevec_pool)
662                 goto fail_trunc_wq;
663
664         /* caps */
665         fsc->min_caps = fsopt->max_readdir;
666
667         return fsc;
668
669 fail_trunc_wq:
670         destroy_workqueue(fsc->trunc_wq);
671 fail_pg_inv_wq:
672         destroy_workqueue(fsc->pg_inv_wq);
673 fail_wb_wq:
674         destroy_workqueue(fsc->wb_wq);
675 fail_client:
676         ceph_destroy_client(fsc->client);
677 fail:
678         kfree(fsc);
679         return ERR_PTR(err);
680 }
681
682 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
683 {
684         flush_workqueue(fsc->wb_wq);
685         flush_workqueue(fsc->pg_inv_wq);
686         flush_workqueue(fsc->trunc_wq);
687 }
688
689 static void destroy_fs_client(struct ceph_fs_client *fsc)
690 {
691         dout("destroy_fs_client %p\n", fsc);
692
693         destroy_workqueue(fsc->wb_wq);
694         destroy_workqueue(fsc->pg_inv_wq);
695         destroy_workqueue(fsc->trunc_wq);
696
697         mempool_destroy(fsc->wb_pagevec_pool);
698
699         destroy_mount_options(fsc->mount_options);
700
701         ceph_destroy_client(fsc->client);
702
703         kfree(fsc);
704         dout("destroy_fs_client %p done\n", fsc);
705 }
706
707 /*
708  * caches
709  */
710 struct kmem_cache *ceph_inode_cachep;
711 struct kmem_cache *ceph_cap_cachep;
712 struct kmem_cache *ceph_cap_flush_cachep;
713 struct kmem_cache *ceph_dentry_cachep;
714 struct kmem_cache *ceph_file_cachep;
715 struct kmem_cache *ceph_dir_file_cachep;
716
717 static void ceph_inode_init_once(void *foo)
718 {
719         struct ceph_inode_info *ci = foo;
720         inode_init_once(&ci->vfs_inode);
721 }
722
723 static int __init init_caches(void)
724 {
725         int error = -ENOMEM;
726
727         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
728                                       sizeof(struct ceph_inode_info),
729                                       __alignof__(struct ceph_inode_info),
730                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
731                                       SLAB_ACCOUNT, ceph_inode_init_once);
732         if (!ceph_inode_cachep)
733                 return -ENOMEM;
734
735         ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
736         if (!ceph_cap_cachep)
737                 goto bad_cap;
738         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
739                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
740         if (!ceph_cap_flush_cachep)
741                 goto bad_cap_flush;
742
743         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
744                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
745         if (!ceph_dentry_cachep)
746                 goto bad_dentry;
747
748         ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
749         if (!ceph_file_cachep)
750                 goto bad_file;
751
752         ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
753         if (!ceph_dir_file_cachep)
754                 goto bad_dir_file;
755
756         error = ceph_fscache_register();
757         if (error)
758                 goto bad_fscache;
759
760         return 0;
761
762 bad_fscache:
763         kmem_cache_destroy(ceph_dir_file_cachep);
764 bad_dir_file:
765         kmem_cache_destroy(ceph_file_cachep);
766 bad_file:
767         kmem_cache_destroy(ceph_dentry_cachep);
768 bad_dentry:
769         kmem_cache_destroy(ceph_cap_flush_cachep);
770 bad_cap_flush:
771         kmem_cache_destroy(ceph_cap_cachep);
772 bad_cap:
773         kmem_cache_destroy(ceph_inode_cachep);
774         return error;
775 }
776
777 static void destroy_caches(void)
778 {
779         /*
780          * Make sure all delayed rcu free inodes are flushed before we
781          * destroy cache.
782          */
783         rcu_barrier();
784
785         kmem_cache_destroy(ceph_inode_cachep);
786         kmem_cache_destroy(ceph_cap_cachep);
787         kmem_cache_destroy(ceph_cap_flush_cachep);
788         kmem_cache_destroy(ceph_dentry_cachep);
789         kmem_cache_destroy(ceph_file_cachep);
790         kmem_cache_destroy(ceph_dir_file_cachep);
791
792         ceph_fscache_unregister();
793 }
794
795
796 /*
797  * ceph_umount_begin - initiate forced umount.  Tear down down the
798  * mount, skipping steps that may hang while waiting for server(s).
799  */
800 static void ceph_umount_begin(struct super_block *sb)
801 {
802         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
803
804         dout("ceph_umount_begin - starting forced umount\n");
805         if (!fsc)
806                 return;
807         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
808         ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
809         ceph_mdsc_force_umount(fsc->mdsc);
810         return;
811 }
812
813 static const struct super_operations ceph_super_ops = {
814         .alloc_inode    = ceph_alloc_inode,
815         .destroy_inode  = ceph_destroy_inode,
816         .write_inode    = ceph_write_inode,
817         .drop_inode     = ceph_drop_inode,
818         .sync_fs        = ceph_sync_fs,
819         .put_super      = ceph_put_super,
820         .show_options   = ceph_show_options,
821         .statfs         = ceph_statfs,
822         .umount_begin   = ceph_umount_begin,
823 };
824
825 /*
826  * Bootstrap mount by opening the root directory.  Note the mount
827  * @started time from caller, and time out if this takes too long.
828  */
829 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
830                                        const char *path,
831                                        unsigned long started)
832 {
833         struct ceph_mds_client *mdsc = fsc->mdsc;
834         struct ceph_mds_request *req = NULL;
835         int err;
836         struct dentry *root;
837
838         /* open dir */
839         dout("open_root_inode opening '%s'\n", path);
840         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
841         if (IS_ERR(req))
842                 return ERR_CAST(req);
843         req->r_path1 = kstrdup(path, GFP_NOFS);
844         if (!req->r_path1) {
845                 root = ERR_PTR(-ENOMEM);
846                 goto out;
847         }
848
849         req->r_ino1.ino = CEPH_INO_ROOT;
850         req->r_ino1.snap = CEPH_NOSNAP;
851         req->r_started = started;
852         req->r_timeout = fsc->client->options->mount_timeout;
853         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
854         req->r_num_caps = 2;
855         err = ceph_mdsc_do_request(mdsc, NULL, req);
856         if (err == 0) {
857                 struct inode *inode = req->r_target_inode;
858                 req->r_target_inode = NULL;
859                 dout("open_root_inode success\n");
860                 root = d_make_root(inode);
861                 if (!root) {
862                         root = ERR_PTR(-ENOMEM);
863                         goto out;
864                 }
865                 dout("open_root_inode success, root dentry is %p\n", root);
866         } else {
867                 root = ERR_PTR(err);
868         }
869 out:
870         ceph_mdsc_put_request(req);
871         return root;
872 }
873
874
875
876
877 /*
878  * mount: join the ceph cluster, and open root directory.
879  */
880 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
881 {
882         int err;
883         unsigned long started = jiffies;  /* note the start time */
884         struct dentry *root;
885
886         dout("mount start %p\n", fsc);
887         mutex_lock(&fsc->client->mount_mutex);
888
889         if (!fsc->sb->s_root) {
890                 const char *path;
891                 err = __ceph_open_session(fsc->client, started);
892                 if (err < 0)
893                         goto out;
894
895                 /* setup fscache */
896                 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
897                         err = ceph_fscache_register_fs(fsc);
898                         if (err < 0)
899                                 goto out;
900                 }
901
902                 if (!fsc->mount_options->server_path) {
903                         path = "";
904                         dout("mount opening path \\t\n");
905                 } else {
906                         path = fsc->mount_options->server_path + 1;
907                         dout("mount opening path %s\n", path);
908                 }
909
910                 err = ceph_fs_debugfs_init(fsc);
911                 if (err < 0)
912                         goto out;
913
914                 root = open_root_dentry(fsc, path, started);
915                 if (IS_ERR(root)) {
916                         err = PTR_ERR(root);
917                         goto out;
918                 }
919                 fsc->sb->s_root = dget(root);
920         } else {
921                 root = dget(fsc->sb->s_root);
922         }
923
924         fsc->mount_state = CEPH_MOUNT_MOUNTED;
925         dout("mount success\n");
926         mutex_unlock(&fsc->client->mount_mutex);
927         return root;
928
929 out:
930         mutex_unlock(&fsc->client->mount_mutex);
931         return ERR_PTR(err);
932 }
933
934 static int ceph_set_super(struct super_block *s, void *data)
935 {
936         struct ceph_fs_client *fsc = data;
937         int ret;
938
939         dout("set_super %p data %p\n", s, data);
940
941         s->s_flags = fsc->mount_options->sb_flags;
942         s->s_maxbytes = 1ULL << 40;  /* temp value until we get mdsmap */
943
944         s->s_xattr = ceph_xattr_handlers;
945         s->s_fs_info = fsc;
946         fsc->sb = s;
947
948         s->s_op = &ceph_super_ops;
949         s->s_d_op = &ceph_dentry_ops;
950         s->s_export_op = &ceph_export_ops;
951
952         s->s_time_gran = 1000;  /* 1000 ns == 1 us */
953
954         ret = set_anon_super(s, NULL);  /* what is that second arg for? */
955         if (ret != 0)
956                 goto fail;
957
958         return ret;
959
960 fail:
961         s->s_fs_info = NULL;
962         fsc->sb = NULL;
963         return ret;
964 }
965
966 /*
967  * share superblock if same fs AND options
968  */
969 static int ceph_compare_super(struct super_block *sb, void *data)
970 {
971         struct ceph_fs_client *new = data;
972         struct ceph_mount_options *fsopt = new->mount_options;
973         struct ceph_options *opt = new->client->options;
974         struct ceph_fs_client *other = ceph_sb_to_client(sb);
975
976         dout("ceph_compare_super %p\n", sb);
977
978         if (compare_mount_options(fsopt, opt, other)) {
979                 dout("monitor(s)/mount options don't match\n");
980                 return 0;
981         }
982         if ((opt->flags & CEPH_OPT_FSID) &&
983             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
984                 dout("fsid doesn't match\n");
985                 return 0;
986         }
987         if (fsopt->sb_flags != other->mount_options->sb_flags) {
988                 dout("flags differ\n");
989                 return 0;
990         }
991         return 1;
992 }
993
994 /*
995  * construct our own bdi so we can control readahead, etc.
996  */
997 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
998
999 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1000 {
1001         int err;
1002
1003         err = super_setup_bdi_name(sb, "ceph-%ld",
1004                                    atomic_long_inc_return(&bdi_seq));
1005         if (err)
1006                 return err;
1007
1008         /* set ra_pages based on rasize mount option? */
1009         sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1010
1011         /* set io_pages based on max osd read size */
1012         sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1013
1014         return 0;
1015 }
1016
1017 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1018                        int flags, const char *dev_name, void *data)
1019 {
1020         struct super_block *sb;
1021         struct ceph_fs_client *fsc;
1022         struct dentry *res;
1023         int err;
1024         int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1025         struct ceph_mount_options *fsopt = NULL;
1026         struct ceph_options *opt = NULL;
1027
1028         dout("ceph_mount\n");
1029
1030 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1031         flags |= SB_POSIXACL;
1032 #endif
1033         err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1034         if (err < 0) {
1035                 res = ERR_PTR(err);
1036                 goto out_final;
1037         }
1038
1039         /* create client (which we may/may not use) */
1040         fsc = create_fs_client(fsopt, opt);
1041         if (IS_ERR(fsc)) {
1042                 res = ERR_CAST(fsc);
1043                 destroy_mount_options(fsopt);
1044                 ceph_destroy_options(opt);
1045                 goto out_final;
1046         }
1047
1048         err = ceph_mdsc_init(fsc);
1049         if (err < 0) {
1050                 res = ERR_PTR(err);
1051                 goto out;
1052         }
1053
1054         if (ceph_test_opt(fsc->client, NOSHARE))
1055                 compare_super = NULL;
1056         sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1057         if (IS_ERR(sb)) {
1058                 res = ERR_CAST(sb);
1059                 goto out;
1060         }
1061
1062         if (ceph_sb_to_client(sb) != fsc) {
1063                 ceph_mdsc_destroy(fsc);
1064                 destroy_fs_client(fsc);
1065                 fsc = ceph_sb_to_client(sb);
1066                 dout("get_sb got existing client %p\n", fsc);
1067         } else {
1068                 dout("get_sb using new client %p\n", fsc);
1069                 err = ceph_setup_bdi(sb, fsc);
1070                 if (err < 0) {
1071                         res = ERR_PTR(err);
1072                         goto out_splat;
1073                 }
1074         }
1075
1076         res = ceph_real_mount(fsc);
1077         if (IS_ERR(res))
1078                 goto out_splat;
1079         dout("root %p inode %p ino %llx.%llx\n", res,
1080              d_inode(res), ceph_vinop(d_inode(res)));
1081         return res;
1082
1083 out_splat:
1084         ceph_mdsc_close_sessions(fsc->mdsc);
1085         deactivate_locked_super(sb);
1086         goto out_final;
1087
1088 out:
1089         ceph_mdsc_destroy(fsc);
1090         destroy_fs_client(fsc);
1091 out_final:
1092         dout("ceph_mount fail %ld\n", PTR_ERR(res));
1093         return res;
1094 }
1095
1096 static void ceph_kill_sb(struct super_block *s)
1097 {
1098         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1099         dev_t dev = s->s_dev;
1100
1101         dout("kill_sb %p\n", s);
1102
1103         ceph_mdsc_pre_umount(fsc->mdsc);
1104         flush_fs_workqueues(fsc);
1105
1106         generic_shutdown_super(s);
1107
1108         fsc->client->extra_mon_dispatch = NULL;
1109         ceph_fs_debugfs_cleanup(fsc);
1110
1111         ceph_fscache_unregister_fs(fsc);
1112
1113         ceph_mdsc_destroy(fsc);
1114
1115         destroy_fs_client(fsc);
1116         free_anon_bdev(dev);
1117 }
1118
1119 static struct file_system_type ceph_fs_type = {
1120         .owner          = THIS_MODULE,
1121         .name           = "ceph",
1122         .mount          = ceph_mount,
1123         .kill_sb        = ceph_kill_sb,
1124         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1125 };
1126 MODULE_ALIAS_FS("ceph");
1127
1128 static int __init init_ceph(void)
1129 {
1130         int ret = init_caches();
1131         if (ret)
1132                 goto out;
1133
1134         ceph_flock_init();
1135         ceph_xattr_init();
1136         ret = register_filesystem(&ceph_fs_type);
1137         if (ret)
1138                 goto out_xattr;
1139
1140         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1141
1142         return 0;
1143
1144 out_xattr:
1145         ceph_xattr_exit();
1146         destroy_caches();
1147 out:
1148         return ret;
1149 }
1150
1151 static void __exit exit_ceph(void)
1152 {
1153         dout("exit_ceph\n");
1154         unregister_filesystem(&ceph_fs_type);
1155         ceph_xattr_exit();
1156         destroy_caches();
1157 }
1158
1159 module_init(init_ceph);
1160 module_exit(exit_ceph);
1161
1162 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1163 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1164 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1165 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1166 MODULE_LICENSE("GPL");