blob: 08b460ae05395311f833d54d39c2bf95c3c89667 [file] [log] [blame]
Sage Weil16725b92009-10-06 11:31:07 -07001
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002#include <linux/ceph/ceph_debug.h>
Sage Weil16725b92009-10-06 11:31:07 -07003
4#include <linux/backing-dev.h>
Sage Weilc309f0a2010-06-30 21:34:01 -07005#include <linux/ctype.h>
Sage Weil16725b92009-10-06 11:31:07 -07006#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>
Sage Weil16725b92009-10-06 11:31:07 -070012#include <linux/sched.h>
13#include <linux/seq_file.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090014#include <linux/slab.h>
Sage Weil16725b92009-10-06 11:31:07 -070015#include <linux/statfs.h>
16#include <linux/string.h>
Sage Weil16725b92009-10-06 11:31:07 -070017
Sage Weil16725b92009-10-06 11:31:07 -070018#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070019#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>
Sage Weil16725b92009-10-06 11:31:07 -070025
26/*
27 * Ceph superblock operations
28 *
29 * Handle the basics of mounting, unmounting.
30 */
31
Sage Weil16725b92009-10-06 11:31:07 -070032/*
33 * super ops
34 */
35static void ceph_put_super(struct super_block *s)
36{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070037 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
Sage Weil16725b92009-10-06 11:31:07 -070038
39 dout("put_super\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070040 ceph_mdsc_close_sessions(fsc->mdsc);
Sage Weil5dfc5892010-05-04 16:14:46 -070041
42 /*
43 * ensure we release the bdi before put_anon_super releases
44 * the device name.
45 */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070046 if (s->s_bdi == &fsc->backing_dev_info) {
47 bdi_unregister(&fsc->backing_dev_info);
Sage Weil5dfc5892010-05-04 16:14:46 -070048 s->s_bdi = NULL;
49 }
50
Sage Weil16725b92009-10-06 11:31:07 -070051 return;
52}
53
54static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
55{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070056 struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
57 struct ceph_monmap *monmap = fsc->client->monc.monmap;
Sage Weil16725b92009-10-06 11:31:07 -070058 struct ceph_statfs st;
59 u64 fsid;
60 int err;
61
62 dout("statfs\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070063 err = ceph_monc_do_statfs(&fsc->client->monc, &st);
Sage Weil16725b92009-10-06 11:31:07 -070064 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) - le64_to_cpu(st.kb_used)) >>
77 (CEPH_BLOCK_SHIFT-10);
78 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
79
80 buf->f_files = le64_to_cpu(st.num_objects);
81 buf->f_ffree = -1;
Sage Weil558d3492010-06-01 12:51:12 -070082 buf->f_namelen = NAME_MAX;
Sage Weil16725b92009-10-06 11:31:07 -070083 buf->f_frsize = PAGE_CACHE_SIZE;
84
85 /* leave fsid little-endian, regardless of host endianness */
86 fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
87 buf->f_fsid.val[0] = fsid & 0xffffffff;
88 buf->f_fsid.val[1] = fsid >> 32;
89
90 return 0;
91}
92
93
Sage Weil2d9c98a2010-07-30 09:38:13 -070094static int ceph_sync_fs(struct super_block *sb, int wait)
Sage Weil16725b92009-10-06 11:31:07 -070095{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070096 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
Sage Weil2d9c98a2010-07-30 09:38:13 -070097
98 if (!wait) {
99 dout("sync_fs (non-blocking)\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700100 ceph_flush_dirty_caps(fsc->mdsc);
Sage Weil2d9c98a2010-07-30 09:38:13 -0700101 dout("sync_fs (non-blocking) done\n");
102 return 0;
103 }
104
105 dout("sync_fs (blocking)\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700106 ceph_osdc_sync(&fsc->client->osdc);
107 ceph_mdsc_sync(fsc->mdsc);
Sage Weil2d9c98a2010-07-30 09:38:13 -0700108 dout("sync_fs (blocking) done\n");
Sage Weil16725b92009-10-06 11:31:07 -0700109 return 0;
110}
111
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700112/*
113 * mount options
114 */
115enum {
116 Opt_wsize,
117 Opt_rsize,
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};
135
136static match_table_t fsopt_tokens = {
137 {Opt_wsize, "wsize=%d"},
138 {Opt_rsize, "rsize=%d"},
139 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
140 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
141 {Opt_cap_release_safety, "cap_release_safety=%d"},
142 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
143 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
144 {Opt_congestion_kb, "write_congestion_kb=%d"},
145 /* int args above */
146 {Opt_snapdirname, "snapdirname=%s"},
147 /* string args above */
148 {Opt_dirstat, "dirstat"},
149 {Opt_nodirstat, "nodirstat"},
150 {Opt_rbytes, "rbytes"},
151 {Opt_norbytes, "norbytes"},
152 {Opt_noasyncreaddir, "noasyncreaddir"},
153 {-1, NULL}
154};
155
156static int parse_fsopt_token(char *c, void *private)
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800157{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700158 struct ceph_mount_options *fsopt = private;
159 substring_t argstr[MAX_OPT_ARGS];
160 int token, intval, ret;
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800161
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700162 token = match_token((char *)c, fsopt_tokens, argstr);
163 if (token < 0)
164 return -EINVAL;
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800165
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700166 if (token < Opt_last_int) {
167 ret = match_int(&argstr[0], &intval);
168 if (ret < 0) {
169 pr_err("bad mount option arg (not int) "
170 "at '%s'\n", c);
171 return ret;
172 }
173 dout("got int token %d val %d\n", token, intval);
174 } else if (token > Opt_last_int && token < Opt_last_string) {
175 dout("got string token %d val %s\n", token,
176 argstr[0].from);
177 } else {
178 dout("got token %d\n", token);
179 }
180
181 switch (token) {
182 case Opt_snapdirname:
183 kfree(fsopt->snapdir_name);
184 fsopt->snapdir_name = kstrndup(argstr[0].from,
185 argstr[0].to-argstr[0].from,
186 GFP_KERNEL);
187 if (!fsopt->snapdir_name)
188 return -ENOMEM;
189 break;
190
191 /* misc */
192 case Opt_wsize:
193 fsopt->wsize = intval;
194 break;
195 case Opt_rsize:
196 fsopt->rsize = intval;
197 break;
198 case Opt_caps_wanted_delay_min:
199 fsopt->caps_wanted_delay_min = intval;
200 break;
201 case Opt_caps_wanted_delay_max:
202 fsopt->caps_wanted_delay_max = intval;
203 break;
204 case Opt_readdir_max_entries:
205 fsopt->max_readdir = intval;
206 break;
207 case Opt_readdir_max_bytes:
208 fsopt->max_readdir_bytes = intval;
209 break;
210 case Opt_congestion_kb:
211 fsopt->congestion_kb = intval;
212 break;
213 case Opt_dirstat:
214 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
215 break;
216 case Opt_nodirstat:
217 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
218 break;
219 case Opt_rbytes:
220 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
221 break;
222 case Opt_norbytes:
223 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
224 break;
225 case Opt_noasyncreaddir:
226 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
227 break;
228 default:
229 BUG_ON(token);
230 }
231 return 0;
232}
233
234static void destroy_mount_options(struct ceph_mount_options *args)
235{
236 dout("destroy_mount_options %p\n", args);
237 kfree(args->snapdir_name);
238 kfree(args);
239}
240
241static int strcmp_null(const char *s1, const char *s2)
242{
243 if (!s1 && !s2)
244 return 0;
245 if (s1 && !s2)
246 return -1;
247 if (!s1 && s2)
248 return 1;
249 return strcmp(s1, s2);
250}
251
252static int compare_mount_options(struct ceph_mount_options *new_fsopt,
253 struct ceph_options *new_opt,
254 struct ceph_fs_client *fsc)
255{
256 struct ceph_mount_options *fsopt1 = new_fsopt;
257 struct ceph_mount_options *fsopt2 = fsc->mount_options;
258 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
259 int ret;
260
261 ret = memcmp(fsopt1, fsopt2, ofs);
262 if (ret)
263 return ret;
264
265 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
266 if (ret)
267 return ret;
268
269 return ceph_compare_options(new_opt, fsc->client);
270}
271
272static int parse_mount_options(struct ceph_mount_options **pfsopt,
273 struct ceph_options **popt,
274 int flags, char *options,
275 const char *dev_name,
276 const char **path)
277{
278 struct ceph_mount_options *fsopt;
279 const char *dev_name_end;
280 int err = -ENOMEM;
281
282 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
283 if (!fsopt)
284 return -ENOMEM;
285
286 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
287
288 fsopt->sb_flags = flags;
289 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
290
291 fsopt->rsize = CEPH_MOUNT_RSIZE_DEFAULT;
292 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
293 fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
294 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
295 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
296 fsopt->congestion_kb = default_congestion_kb();
297
298 /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
299 err = -EINVAL;
300 if (!dev_name)
301 goto out;
302 *path = strstr(dev_name, ":/");
303 if (*path == NULL) {
304 pr_err("device name is missing path (no :/ in %s)\n",
305 dev_name);
306 goto out;
307 }
308 dev_name_end = *path;
309 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
310
311 /* path on server */
312 *path += 2;
313 dout("server path '%s'\n", *path);
314
315 err = ceph_parse_options(popt, options, dev_name, dev_name_end,
316 parse_fsopt_token, (void *)fsopt);
317 if (err)
318 goto out;
319
320 /* success */
321 *pfsopt = fsopt;
322 return 0;
323
324out:
325 destroy_mount_options(fsopt);
326 return err;
Yehuda Sadeh2baba252009-12-18 13:51:57 -0800327}
328
Sage Weil6e19a162010-04-29 16:38:32 -0700329/**
330 * ceph_show_options - Show mount options in /proc/mounts
331 * @m: seq_file to write to
332 * @mnt: mount descriptor
333 */
334static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
335{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700336 struct ceph_fs_client *fsc = ceph_sb_to_client(mnt->mnt_sb);
337 struct ceph_mount_options *fsopt = fsc->mount_options;
338 struct ceph_options *opt = fsc->client->options;
Sage Weil6e19a162010-04-29 16:38:32 -0700339
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700340 if (opt->flags & CEPH_OPT_FSID)
341 seq_printf(m, ",fsid=%pU", &opt->fsid);
342 if (opt->flags & CEPH_OPT_NOSHARE)
Sage Weil6e19a162010-04-29 16:38:32 -0700343 seq_puts(m, ",noshare");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700344 if (opt->flags & CEPH_OPT_NOCRC)
Sage Weil6e19a162010-04-29 16:38:32 -0700345 seq_puts(m, ",nocrc");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700346
347 if (opt->name)
348 seq_printf(m, ",name=%s", opt->name);
349 if (opt->secret)
350 seq_puts(m, ",secret=<hidden>");
351
352 if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
353 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
354 if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
355 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
356 if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
357 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
358 if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
359 seq_printf(m, ",osdkeepalivetimeout=%d",
360 opt->osd_keepalive_timeout);
361
362 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
363 seq_puts(m, ",dirstat");
364 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
365 seq_puts(m, ",norbytes");
366 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
Sage Weil6e19a162010-04-29 16:38:32 -0700367 seq_puts(m, ",noasyncreaddir");
368
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700369 if (fsopt->wsize)
370 seq_printf(m, ",wsize=%d", fsopt->wsize);
371 if (fsopt->rsize != CEPH_MOUNT_RSIZE_DEFAULT)
372 seq_printf(m, ",rsize=%d", fsopt->rsize);
373 if (fsopt->congestion_kb != default_congestion_kb())
374 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
375 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
Sage Weil6e19a162010-04-29 16:38:32 -0700376 seq_printf(m, ",caps_wanted_delay_min=%d",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700377 fsopt->caps_wanted_delay_min);
378 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
Sage Weil6e19a162010-04-29 16:38:32 -0700379 seq_printf(m, ",caps_wanted_delay_max=%d",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700380 fsopt->caps_wanted_delay_max);
381 if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
Sage Weil6e19a162010-04-29 16:38:32 -0700382 seq_printf(m, ",cap_release_safety=%d",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700383 fsopt->cap_release_safety);
384 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
385 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
386 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
387 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
388 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
389 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
Sage Weil6e19a162010-04-29 16:38:32 -0700390 return 0;
391}
392
393/*
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700394 * handle any mon messages the standard library doesn't understand.
395 * return error if we don't either.
396 */
397static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
398{
399 struct ceph_fs_client *fsc = client->private;
400 int type = le16_to_cpu(msg->hdr.type);
401
402 switch (type) {
403 case CEPH_MSG_MDS_MAP:
404 ceph_mdsc_handle_map(fsc->mdsc, msg);
405 return 0;
406
407 default:
408 return -1;
409 }
410}
411
412/*
413 * create a new fs client
414 */
415struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
416 struct ceph_options *opt)
417{
418 struct ceph_fs_client *fsc;
419 int err = -ENOMEM;
420
421 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
422 if (!fsc)
423 return ERR_PTR(-ENOMEM);
424
425 fsc->client = ceph_create_client(opt, fsc);
426 if (IS_ERR(fsc->client)) {
427 err = PTR_ERR(fsc->client);
428 goto fail;
429 }
430 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
431 fsc->client->supported_features |= CEPH_FEATURE_FLOCK;
432 fsc->client->monc.want_mdsmap = 1;
433
434 fsc->mount_options = fsopt;
435
436 fsc->sb = NULL;
437 fsc->mount_state = CEPH_MOUNT_MOUNTING;
438
439 atomic_long_set(&fsc->writeback_count, 0);
440
441 err = bdi_init(&fsc->backing_dev_info);
442 if (err < 0)
443 goto fail_client;
444
445 err = -ENOMEM;
446 fsc->wb_wq = create_workqueue("ceph-writeback");
447 if (fsc->wb_wq == NULL)
448 goto fail_bdi;
449 fsc->pg_inv_wq = create_singlethread_workqueue("ceph-pg-invalid");
450 if (fsc->pg_inv_wq == NULL)
451 goto fail_wb_wq;
452 fsc->trunc_wq = create_singlethread_workqueue("ceph-trunc");
453 if (fsc->trunc_wq == NULL)
454 goto fail_pg_inv_wq;
455
456 /* set up mempools */
457 err = -ENOMEM;
458 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
459 fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
460 if (!fsc->wb_pagevec_pool)
461 goto fail_trunc_wq;
462
463 /* caps */
464 fsc->min_caps = fsopt->max_readdir;
465
466 return fsc;
467
468fail_trunc_wq:
469 destroy_workqueue(fsc->trunc_wq);
470fail_pg_inv_wq:
471 destroy_workqueue(fsc->pg_inv_wq);
472fail_wb_wq:
473 destroy_workqueue(fsc->wb_wq);
474fail_bdi:
475 bdi_destroy(&fsc->backing_dev_info);
476fail_client:
477 ceph_destroy_client(fsc->client);
478fail:
479 kfree(fsc);
480 return ERR_PTR(err);
481}
482
483void destroy_fs_client(struct ceph_fs_client *fsc)
484{
485 dout("destroy_fs_client %p\n", fsc);
486
487 destroy_workqueue(fsc->wb_wq);
488 destroy_workqueue(fsc->pg_inv_wq);
489 destroy_workqueue(fsc->trunc_wq);
490
491 bdi_destroy(&fsc->backing_dev_info);
492
493 mempool_destroy(fsc->wb_pagevec_pool);
494
495 destroy_mount_options(fsc->mount_options);
496
497 ceph_fs_debugfs_cleanup(fsc);
498
499 ceph_destroy_client(fsc->client);
500
501 kfree(fsc);
502 dout("destroy_fs_client %p done\n", fsc);
503}
504
505/*
Sage Weil6e19a162010-04-29 16:38:32 -0700506 * caches
507 */
508struct kmem_cache *ceph_inode_cachep;
509struct kmem_cache *ceph_cap_cachep;
510struct kmem_cache *ceph_dentry_cachep;
511struct kmem_cache *ceph_file_cachep;
512
513static void ceph_inode_init_once(void *foo)
514{
515 struct ceph_inode_info *ci = foo;
516 inode_init_once(&ci->vfs_inode);
517}
518
Sage Weil16725b92009-10-06 11:31:07 -0700519static int __init init_caches(void)
520{
521 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
522 sizeof(struct ceph_inode_info),
523 __alignof__(struct ceph_inode_info),
524 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
525 ceph_inode_init_once);
526 if (ceph_inode_cachep == NULL)
527 return -ENOMEM;
528
529 ceph_cap_cachep = KMEM_CACHE(ceph_cap,
530 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
531 if (ceph_cap_cachep == NULL)
532 goto bad_cap;
533
534 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
535 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
536 if (ceph_dentry_cachep == NULL)
537 goto bad_dentry;
538
539 ceph_file_cachep = KMEM_CACHE(ceph_file_info,
540 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
541 if (ceph_file_cachep == NULL)
542 goto bad_file;
543
544 return 0;
545
546bad_file:
547 kmem_cache_destroy(ceph_dentry_cachep);
548bad_dentry:
549 kmem_cache_destroy(ceph_cap_cachep);
550bad_cap:
551 kmem_cache_destroy(ceph_inode_cachep);
552 return -ENOMEM;
553}
554
555static void destroy_caches(void)
556{
557 kmem_cache_destroy(ceph_inode_cachep);
558 kmem_cache_destroy(ceph_cap_cachep);
559 kmem_cache_destroy(ceph_dentry_cachep);
560 kmem_cache_destroy(ceph_file_cachep);
561}
562
563
564/*
565 * ceph_umount_begin - initiate forced umount. Tear down down the
566 * mount, skipping steps that may hang while waiting for server(s).
567 */
568static void ceph_umount_begin(struct super_block *sb)
569{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700570 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
Sage Weil16725b92009-10-06 11:31:07 -0700571
572 dout("ceph_umount_begin - starting forced umount\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700573 if (!fsc)
Sage Weil16725b92009-10-06 11:31:07 -0700574 return;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700575 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
Sage Weil16725b92009-10-06 11:31:07 -0700576 return;
577}
578
579static const struct super_operations ceph_super_ops = {
580 .alloc_inode = ceph_alloc_inode,
581 .destroy_inode = ceph_destroy_inode,
582 .write_inode = ceph_write_inode,
Sage Weil2d9c98a2010-07-30 09:38:13 -0700583 .sync_fs = ceph_sync_fs,
Sage Weil16725b92009-10-06 11:31:07 -0700584 .put_super = ceph_put_super,
585 .show_options = ceph_show_options,
586 .statfs = ceph_statfs,
587 .umount_begin = ceph_umount_begin,
588};
589
Sage Weil16725b92009-10-06 11:31:07 -0700590/*
591 * Bootstrap mount by opening the root directory. Note the mount
592 * @started time from caller, and time out if this takes too long.
593 */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700594static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
Sage Weil16725b92009-10-06 11:31:07 -0700595 const char *path,
596 unsigned long started)
597{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700598 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil16725b92009-10-06 11:31:07 -0700599 struct ceph_mds_request *req = NULL;
600 int err;
601 struct dentry *root;
602
603 /* open dir */
604 dout("open_root_inode opening '%s'\n", path);
605 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
606 if (IS_ERR(req))
Julia Lawall7e34bc52010-05-22 12:01:14 +0200607 return ERR_CAST(req);
Sage Weil16725b92009-10-06 11:31:07 -0700608 req->r_path1 = kstrdup(path, GFP_NOFS);
609 req->r_ino1.ino = CEPH_INO_ROOT;
610 req->r_ino1.snap = CEPH_NOSNAP;
611 req->r_started = started;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700612 req->r_timeout = fsc->client->options->mount_timeout * HZ;
Sage Weil16725b92009-10-06 11:31:07 -0700613 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
614 req->r_num_caps = 2;
615 err = ceph_mdsc_do_request(mdsc, NULL, req);
616 if (err == 0) {
617 dout("open_root_inode success\n");
618 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700619 fsc->sb->s_root == NULL)
Sage Weil16725b92009-10-06 11:31:07 -0700620 root = d_alloc_root(req->r_target_inode);
621 else
622 root = d_obtain_alias(req->r_target_inode);
623 req->r_target_inode = NULL;
624 dout("open_root_inode success, root dentry is %p\n", root);
625 } else {
626 root = ERR_PTR(err);
627 }
628 ceph_mdsc_put_request(req);
629 return root;
630}
631
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700632
633
634
Sage Weil16725b92009-10-06 11:31:07 -0700635/*
636 * mount: join the ceph cluster, and open root directory.
637 */
Al Viroa7f9fb22010-07-26 16:17:55 +0400638static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
Sage Weil16725b92009-10-06 11:31:07 -0700639 const char *path)
640{
Sage Weil16725b92009-10-06 11:31:07 -0700641 int err;
Sage Weil16725b92009-10-06 11:31:07 -0700642 unsigned long started = jiffies; /* note the start time */
643 struct dentry *root;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700644 int first = 0; /* first vfsmount for this super_block */
Sage Weil16725b92009-10-06 11:31:07 -0700645
646 dout("mount start\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700647 mutex_lock(&fsc->client->mount_mutex);
Sage Weil16725b92009-10-06 11:31:07 -0700648
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700649 err = __ceph_open_session(fsc->client, started);
Sage Weil16725b92009-10-06 11:31:07 -0700650 if (err < 0)
651 goto out;
652
Sage Weil16725b92009-10-06 11:31:07 -0700653 dout("mount opening root\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700654 root = open_root_dentry(fsc, "", started);
Sage Weil16725b92009-10-06 11:31:07 -0700655 if (IS_ERR(root)) {
656 err = PTR_ERR(root);
657 goto out;
658 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700659 if (fsc->sb->s_root) {
Sage Weil16725b92009-10-06 11:31:07 -0700660 dput(root);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700661 } else {
662 fsc->sb->s_root = root;
663 first = 1;
664
665 err = ceph_fs_debugfs_init(fsc);
666 if (err < 0)
667 goto fail;
668 }
Sage Weil16725b92009-10-06 11:31:07 -0700669
670 if (path[0] == 0) {
671 dget(root);
672 } else {
673 dout("mount opening base mountpoint\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700674 root = open_root_dentry(fsc, path, started);
Sage Weil16725b92009-10-06 11:31:07 -0700675 if (IS_ERR(root)) {
676 err = PTR_ERR(root);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700677 goto fail;
Sage Weil16725b92009-10-06 11:31:07 -0700678 }
679 }
680
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700681 fsc->mount_state = CEPH_MOUNT_MOUNTED;
Sage Weil16725b92009-10-06 11:31:07 -0700682 dout("mount success\n");
Al Viroa7f9fb22010-07-26 16:17:55 +0400683 mutex_unlock(&fsc->client->mount_mutex);
684 return root;
Sage Weil16725b92009-10-06 11:31:07 -0700685
686out:
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700687 mutex_unlock(&fsc->client->mount_mutex);
Al Viroa7f9fb22010-07-26 16:17:55 +0400688 return ERR_PTR(err);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700689
690fail:
691 if (first) {
692 dput(fsc->sb->s_root);
693 fsc->sb->s_root = NULL;
694 }
695 goto out;
Sage Weil16725b92009-10-06 11:31:07 -0700696}
697
698static int ceph_set_super(struct super_block *s, void *data)
699{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700700 struct ceph_fs_client *fsc = data;
Sage Weil16725b92009-10-06 11:31:07 -0700701 int ret;
702
703 dout("set_super %p data %p\n", s, data);
704
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700705 s->s_flags = fsc->mount_options->sb_flags;
Sage Weil16725b92009-10-06 11:31:07 -0700706 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
707
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700708 s->s_fs_info = fsc;
709 fsc->sb = s;
Sage Weil16725b92009-10-06 11:31:07 -0700710
711 s->s_op = &ceph_super_ops;
712 s->s_export_op = &ceph_export_ops;
713
714 s->s_time_gran = 1000; /* 1000 ns == 1 us */
715
716 ret = set_anon_super(s, NULL); /* what is that second arg for? */
717 if (ret != 0)
718 goto fail;
719
720 return ret;
721
722fail:
723 s->s_fs_info = NULL;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700724 fsc->sb = NULL;
Sage Weil16725b92009-10-06 11:31:07 -0700725 return ret;
726}
727
728/*
729 * share superblock if same fs AND options
730 */
731static int ceph_compare_super(struct super_block *sb, void *data)
732{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700733 struct ceph_fs_client *new = data;
734 struct ceph_mount_options *fsopt = new->mount_options;
735 struct ceph_options *opt = new->client->options;
736 struct ceph_fs_client *other = ceph_sb_to_client(sb);
Sage Weil16725b92009-10-06 11:31:07 -0700737
738 dout("ceph_compare_super %p\n", sb);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700739
740 if (compare_mount_options(fsopt, opt, other)) {
741 dout("monitor(s)/mount options don't match\n");
742 return 0;
Sage Weil16725b92009-10-06 11:31:07 -0700743 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700744 if ((opt->flags & CEPH_OPT_FSID) &&
745 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
746 dout("fsid doesn't match\n");
747 return 0;
748 }
749 if (fsopt->sb_flags != other->mount_options->sb_flags) {
Sage Weil16725b92009-10-06 11:31:07 -0700750 dout("flags differ\n");
751 return 0;
752 }
753 return 1;
754}
755
756/*
757 * construct our own bdi so we can control readahead, etc.
758 */
Jeff Mahoney00d56432010-06-10 11:13:58 -0400759static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
Sage Weil31e0cf82010-05-04 16:39:35 -0700760
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700761static int ceph_register_bdi(struct super_block *sb,
762 struct ceph_fs_client *fsc)
Sage Weil16725b92009-10-06 11:31:07 -0700763{
764 int err;
765
Sage Weil16725b92009-10-06 11:31:07 -0700766 /* set ra_pages based on rsize mount option? */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700767 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
768 fsc->backing_dev_info.ra_pages =
769 (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
Sage Weil16725b92009-10-06 11:31:07 -0700770 >> PAGE_SHIFT;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700771 err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
Sage Weil31e0cf82010-05-04 16:39:35 -0700772 atomic_long_inc_return(&bdi_seq));
Sage Weil5dfc5892010-05-04 16:14:46 -0700773 if (!err)
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700774 sb->s_bdi = &fsc->backing_dev_info;
Sage Weil16725b92009-10-06 11:31:07 -0700775 return err;
776}
777
Al Viroa7f9fb22010-07-26 16:17:55 +0400778static struct dentry *ceph_mount(struct file_system_type *fs_type,
779 int flags, const char *dev_name, void *data)
Sage Weil16725b92009-10-06 11:31:07 -0700780{
781 struct super_block *sb;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700782 struct ceph_fs_client *fsc;
Al Viroa7f9fb22010-07-26 16:17:55 +0400783 struct dentry *res;
Sage Weil16725b92009-10-06 11:31:07 -0700784 int err;
785 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
Sage Weil6a18be12009-11-04 11:40:05 -0800786 const char *path = NULL;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700787 struct ceph_mount_options *fsopt = NULL;
788 struct ceph_options *opt = NULL;
Sage Weil16725b92009-10-06 11:31:07 -0700789
Al Viroa7f9fb22010-07-26 16:17:55 +0400790 dout("ceph_mount\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700791 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
Al Viroa7f9fb22010-07-26 16:17:55 +0400792 if (err < 0) {
793 res = ERR_PTR(err);
Sage Weil6b805182009-10-27 11:50:50 -0700794 goto out_final;
Al Viroa7f9fb22010-07-26 16:17:55 +0400795 }
Sage Weil16725b92009-10-06 11:31:07 -0700796
797 /* create client (which we may/may not use) */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700798 fsc = create_fs_client(fsopt, opt);
799 if (IS_ERR(fsc)) {
Al Viroa7f9fb22010-07-26 16:17:55 +0400800 res = ERR_CAST(fsc);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700801 kfree(fsopt);
802 kfree(opt);
Sage Weil6b805182009-10-27 11:50:50 -0700803 goto out_final;
804 }
Sage Weil16725b92009-10-06 11:31:07 -0700805
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700806 err = ceph_mdsc_init(fsc);
Al Viroa7f9fb22010-07-26 16:17:55 +0400807 if (err < 0) {
808 res = ERR_PTR(err);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700809 goto out;
Al Viroa7f9fb22010-07-26 16:17:55 +0400810 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700811
812 if (ceph_test_opt(fsc->client, NOSHARE))
Sage Weil16725b92009-10-06 11:31:07 -0700813 compare_super = NULL;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700814 sb = sget(fs_type, compare_super, ceph_set_super, fsc);
Sage Weil16725b92009-10-06 11:31:07 -0700815 if (IS_ERR(sb)) {
Al Viroa7f9fb22010-07-26 16:17:55 +0400816 res = ERR_CAST(sb);
Sage Weil16725b92009-10-06 11:31:07 -0700817 goto out;
818 }
819
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700820 if (ceph_sb_to_client(sb) != fsc) {
821 ceph_mdsc_destroy(fsc);
822 destroy_fs_client(fsc);
823 fsc = ceph_sb_to_client(sb);
824 dout("get_sb got existing client %p\n", fsc);
Sage Weil16725b92009-10-06 11:31:07 -0700825 } else {
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700826 dout("get_sb using new client %p\n", fsc);
827 err = ceph_register_bdi(sb, fsc);
Al Viroa7f9fb22010-07-26 16:17:55 +0400828 if (err < 0) {
829 res = ERR_PTR(err);
Sage Weil16725b92009-10-06 11:31:07 -0700830 goto out_splat;
Al Viroa7f9fb22010-07-26 16:17:55 +0400831 }
Sage Weil16725b92009-10-06 11:31:07 -0700832 }
833
Al Viroa7f9fb22010-07-26 16:17:55 +0400834 res = ceph_real_mount(fsc, path);
835 if (IS_ERR(res))
Sage Weil16725b92009-10-06 11:31:07 -0700836 goto out_splat;
Al Viroa7f9fb22010-07-26 16:17:55 +0400837 dout("root %p inode %p ino %llx.%llx\n", res,
838 res->d_inode, ceph_vinop(res->d_inode));
839 return res;
Sage Weil16725b92009-10-06 11:31:07 -0700840
841out_splat:
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700842 ceph_mdsc_close_sessions(fsc->mdsc);
Al Viro3981f2e2010-03-21 19:22:29 -0400843 deactivate_locked_super(sb);
Sage Weil16725b92009-10-06 11:31:07 -0700844 goto out_final;
845
846out:
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700847 ceph_mdsc_destroy(fsc);
848 destroy_fs_client(fsc);
Sage Weil16725b92009-10-06 11:31:07 -0700849out_final:
Al Viroa7f9fb22010-07-26 16:17:55 +0400850 dout("ceph_mount fail %ld\n", PTR_ERR(res));
851 return res;
Sage Weil16725b92009-10-06 11:31:07 -0700852}
853
854static void ceph_kill_sb(struct super_block *s)
855{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700856 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
Sage Weil16725b92009-10-06 11:31:07 -0700857 dout("kill_sb %p\n", s);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700858 ceph_mdsc_pre_umount(fsc->mdsc);
Sage Weil16725b92009-10-06 11:31:07 -0700859 kill_anon_super(s); /* will call put_super after sb is r/o */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700860 ceph_mdsc_destroy(fsc);
861 destroy_fs_client(fsc);
Sage Weil16725b92009-10-06 11:31:07 -0700862}
863
864static struct file_system_type ceph_fs_type = {
865 .owner = THIS_MODULE,
866 .name = "ceph",
Al Viroa7f9fb22010-07-26 16:17:55 +0400867 .mount = ceph_mount,
Sage Weil16725b92009-10-06 11:31:07 -0700868 .kill_sb = ceph_kill_sb,
869 .fs_flags = FS_RENAME_DOES_D_MOVE,
870};
871
872#define _STRINGIFY(x) #x
873#define STRINGIFY(x) _STRINGIFY(x)
874
875static int __init init_ceph(void)
876{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700877 int ret = init_caches();
Sage Weil16725b92009-10-06 11:31:07 -0700878 if (ret)
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700879 goto out;
Sage Weil16725b92009-10-06 11:31:07 -0700880
Sage Weil16725b92009-10-06 11:31:07 -0700881 ret = register_filesystem(&ceph_fs_type);
882 if (ret)
883 goto out_icache;
884
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700885 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
886
Sage Weil16725b92009-10-06 11:31:07 -0700887 return 0;
888
889out_icache:
890 destroy_caches();
Sage Weil16725b92009-10-06 11:31:07 -0700891out:
892 return ret;
893}
894
895static void __exit exit_ceph(void)
896{
897 dout("exit_ceph\n");
898 unregister_filesystem(&ceph_fs_type);
Sage Weil16725b92009-10-06 11:31:07 -0700899 destroy_caches();
Sage Weil16725b92009-10-06 11:31:07 -0700900}
901
902module_init(init_ceph);
903module_exit(exit_ceph);
904
905MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
906MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
907MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
908MODULE_DESCRIPTION("Ceph filesystem for Linux");
909MODULE_LICENSE("GPL");