Loading...
1
2#include <linux/ceph/ceph_debug.h>
3
4#include <linux/backing-dev.h>
5#include <linux/ctype.h>
6#include <linux/fs.h>
7#include <linux/inet.h>
8#include <linux/in6.h>
9#include <linux/module.h>
10#include <linux/mount.h>
11#include <linux/parser.h>
12#include <linux/sched.h>
13#include <linux/seq_file.h>
14#include <linux/slab.h>
15#include <linux/statfs.h>
16#include <linux/string.h>
17
18#include "super.h"
19#include "mds_client.h"
20
21#include <linux/ceph/decode.h>
22#include <linux/ceph/mon_client.h>
23#include <linux/ceph/auth.h>
24#include <linux/ceph/debugfs.h>
25
26/*
27 * Ceph superblock operations
28 *
29 * Handle the basics of mounting, unmounting.
30 */
31
32/*
33 * super ops
34 */
35static void ceph_put_super(struct super_block *s)
36{
37 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
38
39 dout("put_super\n");
40 ceph_mdsc_close_sessions(fsc->mdsc);
41
42 /*
43 * ensure we release the bdi before put_anon_super releases
44 * the device name.
45 */
46 if (s->s_bdi == &fsc->backing_dev_info) {
47 bdi_unregister(&fsc->backing_dev_info);
48 s->s_bdi = NULL;
49 }
50
51 return;
52}
53
54static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
55{
56 struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
57 struct ceph_monmap *monmap = fsc->client->monc.monmap;
58 struct ceph_statfs st;
59 u64 fsid;
60 int err;
61
62 dout("statfs\n");
63 err = ceph_monc_do_statfs(&fsc->client->monc, &st);
64 if (err < 0)
65 return err;
66
67 /* fill in kstatfs */
68 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
69
70 /*
71 * express utilization in terms of large blocks to avoid
72 * overflow on 32-bit machines.
73 */
74 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
75 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
76 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
77 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
78
79 buf->f_files = le64_to_cpu(st.num_objects);
80 buf->f_ffree = -1;
81 buf->f_namelen = NAME_MAX;
82 buf->f_frsize = PAGE_CACHE_SIZE;
83
84 /* leave fsid little-endian, regardless of host endianness */
85 fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
86 buf->f_fsid.val[0] = fsid & 0xffffffff;
87 buf->f_fsid.val[1] = fsid >> 32;
88
89 return 0;
90}
91
92
93static int ceph_sync_fs(struct super_block *sb, int wait)
94{
95 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
96
97 if (!wait) {
98 dout("sync_fs (non-blocking)\n");
99 ceph_flush_dirty_caps(fsc->mdsc);
100 dout("sync_fs (non-blocking) done\n");
101 return 0;
102 }
103
104 dout("sync_fs (blocking)\n");
105 ceph_osdc_sync(&fsc->client->osdc);
106 ceph_mdsc_sync(fsc->mdsc);
107 dout("sync_fs (blocking) done\n");
108 return 0;
109}
110
111/*
112 * mount options
113 */
114enum {
115 Opt_wsize,
116 Opt_rsize,
117 Opt_caps_wanted_delay_min,
118 Opt_caps_wanted_delay_max,
119 Opt_cap_release_safety,
120 Opt_readdir_max_entries,
121 Opt_readdir_max_bytes,
122 Opt_congestion_kb,
123 Opt_last_int,
124 /* int args above */
125 Opt_snapdirname,
126 Opt_last_string,
127 /* string args above */
128 Opt_dirstat,
129 Opt_nodirstat,
130 Opt_rbytes,
131 Opt_norbytes,
132 Opt_noasyncreaddir,
133 Opt_ino32,
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 {Opt_ino32, "ino32"},
154 {-1, NULL}
155};
156
157static int parse_fsopt_token(char *c, void *private)
158{
159 struct ceph_mount_options *fsopt = private;
160 substring_t argstr[MAX_OPT_ARGS];
161 int token, intval, ret;
162
163 token = match_token((char *)c, fsopt_tokens, argstr);
164 if (token < 0)
165 return -EINVAL;
166
167 if (token < Opt_last_int) {
168 ret = match_int(&argstr[0], &intval);
169 if (ret < 0) {
170 pr_err("bad mount option arg (not int) "
171 "at '%s'\n", c);
172 return ret;
173 }
174 dout("got int token %d val %d\n", token, intval);
175 } else if (token > Opt_last_int && token < Opt_last_string) {
176 dout("got string token %d val %s\n", token,
177 argstr[0].from);
178 } else {
179 dout("got token %d\n", token);
180 }
181
182 switch (token) {
183 case Opt_snapdirname:
184 kfree(fsopt->snapdir_name);
185 fsopt->snapdir_name = kstrndup(argstr[0].from,
186 argstr[0].to-argstr[0].from,
187 GFP_KERNEL);
188 if (!fsopt->snapdir_name)
189 return -ENOMEM;
190 break;
191
192 /* misc */
193 case Opt_wsize:
194 fsopt->wsize = intval;
195 break;
196 case Opt_rsize:
197 fsopt->rsize = intval;
198 break;
199 case Opt_caps_wanted_delay_min:
200 fsopt->caps_wanted_delay_min = intval;
201 break;
202 case Opt_caps_wanted_delay_max:
203 fsopt->caps_wanted_delay_max = intval;
204 break;
205 case Opt_readdir_max_entries:
206 fsopt->max_readdir = intval;
207 break;
208 case Opt_readdir_max_bytes:
209 fsopt->max_readdir_bytes = intval;
210 break;
211 case Opt_congestion_kb:
212 fsopt->congestion_kb = intval;
213 break;
214 case Opt_dirstat:
215 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
216 break;
217 case Opt_nodirstat:
218 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
219 break;
220 case Opt_rbytes:
221 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
222 break;
223 case Opt_norbytes:
224 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
225 break;
226 case Opt_noasyncreaddir:
227 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
228 break;
229 case Opt_ino32:
230 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
231 break;
232 default:
233 BUG_ON(token);
234 }
235 return 0;
236}
237
238static void destroy_mount_options(struct ceph_mount_options *args)
239{
240 dout("destroy_mount_options %p\n", args);
241 kfree(args->snapdir_name);
242 kfree(args);
243}
244
245static int strcmp_null(const char *s1, const char *s2)
246{
247 if (!s1 && !s2)
248 return 0;
249 if (s1 && !s2)
250 return -1;
251 if (!s1 && s2)
252 return 1;
253 return strcmp(s1, s2);
254}
255
256static int compare_mount_options(struct ceph_mount_options *new_fsopt,
257 struct ceph_options *new_opt,
258 struct ceph_fs_client *fsc)
259{
260 struct ceph_mount_options *fsopt1 = new_fsopt;
261 struct ceph_mount_options *fsopt2 = fsc->mount_options;
262 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
263 int ret;
264
265 ret = memcmp(fsopt1, fsopt2, ofs);
266 if (ret)
267 return ret;
268
269 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
270 if (ret)
271 return ret;
272
273 return ceph_compare_options(new_opt, fsc->client);
274}
275
276static int parse_mount_options(struct ceph_mount_options **pfsopt,
277 struct ceph_options **popt,
278 int flags, char *options,
279 const char *dev_name,
280 const char **path)
281{
282 struct ceph_mount_options *fsopt;
283 const char *dev_name_end;
284 int err = -ENOMEM;
285
286 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
287 if (!fsopt)
288 return -ENOMEM;
289
290 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
291
292 fsopt->sb_flags = flags;
293 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
294
295 fsopt->rsize = CEPH_RSIZE_DEFAULT;
296 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
297 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
298 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
299 fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
300 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
301 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
302 fsopt->congestion_kb = default_congestion_kb();
303
304 /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
305 err = -EINVAL;
306 if (!dev_name)
307 goto out;
308 *path = strstr(dev_name, ":/");
309 if (*path == NULL) {
310 pr_err("device name is missing path (no :/ in %s)\n",
311 dev_name);
312 goto out;
313 }
314 dev_name_end = *path;
315 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
316
317 /* path on server */
318 *path += 2;
319 dout("server path '%s'\n", *path);
320
321 err = ceph_parse_options(popt, options, dev_name, dev_name_end,
322 parse_fsopt_token, (void *)fsopt);
323 if (err)
324 goto out;
325
326 /* success */
327 *pfsopt = fsopt;
328 return 0;
329
330out:
331 destroy_mount_options(fsopt);
332 return err;
333}
334
335/**
336 * ceph_show_options - Show mount options in /proc/mounts
337 * @m: seq_file to write to
338 * @mnt: mount descriptor
339 */
340static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
341{
342 struct ceph_fs_client *fsc = ceph_sb_to_client(mnt->mnt_sb);
343 struct ceph_mount_options *fsopt = fsc->mount_options;
344 struct ceph_options *opt = fsc->client->options;
345
346 if (opt->flags & CEPH_OPT_FSID)
347 seq_printf(m, ",fsid=%pU", &opt->fsid);
348 if (opt->flags & CEPH_OPT_NOSHARE)
349 seq_puts(m, ",noshare");
350 if (opt->flags & CEPH_OPT_NOCRC)
351 seq_puts(m, ",nocrc");
352
353 if (opt->name)
354 seq_printf(m, ",name=%s", opt->name);
355 if (opt->key)
356 seq_puts(m, ",secret=<hidden>");
357
358 if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
359 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
360 if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
361 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
362 if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
363 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
364 if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
365 seq_printf(m, ",osdkeepalivetimeout=%d",
366 opt->osd_keepalive_timeout);
367
368 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
369 seq_puts(m, ",dirstat");
370 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
371 seq_puts(m, ",norbytes");
372 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
373 seq_puts(m, ",noasyncreaddir");
374
375 if (fsopt->wsize)
376 seq_printf(m, ",wsize=%d", fsopt->wsize);
377 if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
378 seq_printf(m, ",rsize=%d", fsopt->rsize);
379 if (fsopt->congestion_kb != default_congestion_kb())
380 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
381 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
382 seq_printf(m, ",caps_wanted_delay_min=%d",
383 fsopt->caps_wanted_delay_min);
384 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
385 seq_printf(m, ",caps_wanted_delay_max=%d",
386 fsopt->caps_wanted_delay_max);
387 if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
388 seq_printf(m, ",cap_release_safety=%d",
389 fsopt->cap_release_safety);
390 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
391 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
392 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
393 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
394 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
395 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
396 return 0;
397}
398
399/*
400 * handle any mon messages the standard library doesn't understand.
401 * return error if we don't either.
402 */
403static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
404{
405 struct ceph_fs_client *fsc = client->private;
406 int type = le16_to_cpu(msg->hdr.type);
407
408 switch (type) {
409 case CEPH_MSG_MDS_MAP:
410 ceph_mdsc_handle_map(fsc->mdsc, msg);
411 return 0;
412
413 default:
414 return -1;
415 }
416}
417
418/*
419 * create a new fs client
420 */
421struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
422 struct ceph_options *opt)
423{
424 struct ceph_fs_client *fsc;
425 int err = -ENOMEM;
426
427 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
428 if (!fsc)
429 return ERR_PTR(-ENOMEM);
430
431 fsc->client = ceph_create_client(opt, fsc);
432 if (IS_ERR(fsc->client)) {
433 err = PTR_ERR(fsc->client);
434 goto fail;
435 }
436 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
437 fsc->client->supported_features |= CEPH_FEATURE_FLOCK |
438 CEPH_FEATURE_DIRLAYOUTHASH;
439 fsc->client->monc.want_mdsmap = 1;
440
441 fsc->mount_options = fsopt;
442
443 fsc->sb = NULL;
444 fsc->mount_state = CEPH_MOUNT_MOUNTING;
445
446 atomic_long_set(&fsc->writeback_count, 0);
447
448 err = bdi_init(&fsc->backing_dev_info);
449 if (err < 0)
450 goto fail_client;
451
452 err = -ENOMEM;
453 /*
454 * The number of concurrent works can be high but they don't need
455 * to be processed in parallel, limit concurrency.
456 */
457 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
458 if (fsc->wb_wq == NULL)
459 goto fail_bdi;
460 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
461 if (fsc->pg_inv_wq == NULL)
462 goto fail_wb_wq;
463 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
464 if (fsc->trunc_wq == NULL)
465 goto fail_pg_inv_wq;
466
467 /* set up mempools */
468 err = -ENOMEM;
469 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
470 fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
471 if (!fsc->wb_pagevec_pool)
472 goto fail_trunc_wq;
473
474 /* caps */
475 fsc->min_caps = fsopt->max_readdir;
476
477 return fsc;
478
479fail_trunc_wq:
480 destroy_workqueue(fsc->trunc_wq);
481fail_pg_inv_wq:
482 destroy_workqueue(fsc->pg_inv_wq);
483fail_wb_wq:
484 destroy_workqueue(fsc->wb_wq);
485fail_bdi:
486 bdi_destroy(&fsc->backing_dev_info);
487fail_client:
488 ceph_destroy_client(fsc->client);
489fail:
490 kfree(fsc);
491 return ERR_PTR(err);
492}
493
494void destroy_fs_client(struct ceph_fs_client *fsc)
495{
496 dout("destroy_fs_client %p\n", fsc);
497
498 destroy_workqueue(fsc->wb_wq);
499 destroy_workqueue(fsc->pg_inv_wq);
500 destroy_workqueue(fsc->trunc_wq);
501
502 bdi_destroy(&fsc->backing_dev_info);
503
504 mempool_destroy(fsc->wb_pagevec_pool);
505
506 destroy_mount_options(fsc->mount_options);
507
508 ceph_fs_debugfs_cleanup(fsc);
509
510 ceph_destroy_client(fsc->client);
511
512 kfree(fsc);
513 dout("destroy_fs_client %p done\n", fsc);
514}
515
516/*
517 * caches
518 */
519struct kmem_cache *ceph_inode_cachep;
520struct kmem_cache *ceph_cap_cachep;
521struct kmem_cache *ceph_dentry_cachep;
522struct kmem_cache *ceph_file_cachep;
523
524static void ceph_inode_init_once(void *foo)
525{
526 struct ceph_inode_info *ci = foo;
527 inode_init_once(&ci->vfs_inode);
528}
529
530static int __init init_caches(void)
531{
532 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
533 sizeof(struct ceph_inode_info),
534 __alignof__(struct ceph_inode_info),
535 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
536 ceph_inode_init_once);
537 if (ceph_inode_cachep == NULL)
538 return -ENOMEM;
539
540 ceph_cap_cachep = KMEM_CACHE(ceph_cap,
541 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
542 if (ceph_cap_cachep == NULL)
543 goto bad_cap;
544
545 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
546 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
547 if (ceph_dentry_cachep == NULL)
548 goto bad_dentry;
549
550 ceph_file_cachep = KMEM_CACHE(ceph_file_info,
551 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
552 if (ceph_file_cachep == NULL)
553 goto bad_file;
554
555 return 0;
556
557bad_file:
558 kmem_cache_destroy(ceph_dentry_cachep);
559bad_dentry:
560 kmem_cache_destroy(ceph_cap_cachep);
561bad_cap:
562 kmem_cache_destroy(ceph_inode_cachep);
563 return -ENOMEM;
564}
565
566static void destroy_caches(void)
567{
568 kmem_cache_destroy(ceph_inode_cachep);
569 kmem_cache_destroy(ceph_cap_cachep);
570 kmem_cache_destroy(ceph_dentry_cachep);
571 kmem_cache_destroy(ceph_file_cachep);
572}
573
574
575/*
576 * ceph_umount_begin - initiate forced umount. Tear down down the
577 * mount, skipping steps that may hang while waiting for server(s).
578 */
579static void ceph_umount_begin(struct super_block *sb)
580{
581 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
582
583 dout("ceph_umount_begin - starting forced umount\n");
584 if (!fsc)
585 return;
586 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
587 return;
588}
589
590static const struct super_operations ceph_super_ops = {
591 .alloc_inode = ceph_alloc_inode,
592 .destroy_inode = ceph_destroy_inode,
593 .write_inode = ceph_write_inode,
594 .sync_fs = ceph_sync_fs,
595 .put_super = ceph_put_super,
596 .show_options = ceph_show_options,
597 .statfs = ceph_statfs,
598 .umount_begin = ceph_umount_begin,
599};
600
601/*
602 * Bootstrap mount by opening the root directory. Note the mount
603 * @started time from caller, and time out if this takes too long.
604 */
605static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
606 const char *path,
607 unsigned long started)
608{
609 struct ceph_mds_client *mdsc = fsc->mdsc;
610 struct ceph_mds_request *req = NULL;
611 int err;
612 struct dentry *root;
613
614 /* open dir */
615 dout("open_root_inode opening '%s'\n", path);
616 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
617 if (IS_ERR(req))
618 return ERR_CAST(req);
619 req->r_path1 = kstrdup(path, GFP_NOFS);
620 req->r_ino1.ino = CEPH_INO_ROOT;
621 req->r_ino1.snap = CEPH_NOSNAP;
622 req->r_started = started;
623 req->r_timeout = fsc->client->options->mount_timeout * HZ;
624 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
625 req->r_num_caps = 2;
626 err = ceph_mdsc_do_request(mdsc, NULL, req);
627 if (err == 0) {
628 dout("open_root_inode success\n");
629 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
630 fsc->sb->s_root == NULL)
631 root = d_alloc_root(req->r_target_inode);
632 else
633 root = d_obtain_alias(req->r_target_inode);
634 req->r_target_inode = NULL;
635 dout("open_root_inode success, root dentry is %p\n", root);
636 } else {
637 root = ERR_PTR(err);
638 }
639 ceph_mdsc_put_request(req);
640 return root;
641}
642
643
644
645
646/*
647 * mount: join the ceph cluster, and open root directory.
648 */
649static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
650 const char *path)
651{
652 int err;
653 unsigned long started = jiffies; /* note the start time */
654 struct dentry *root;
655 int first = 0; /* first vfsmount for this super_block */
656
657 dout("mount start\n");
658 mutex_lock(&fsc->client->mount_mutex);
659
660 err = __ceph_open_session(fsc->client, started);
661 if (err < 0)
662 goto out;
663
664 dout("mount opening root\n");
665 root = open_root_dentry(fsc, "", started);
666 if (IS_ERR(root)) {
667 err = PTR_ERR(root);
668 goto out;
669 }
670 if (fsc->sb->s_root) {
671 dput(root);
672 } else {
673 fsc->sb->s_root = root;
674 first = 1;
675
676 err = ceph_fs_debugfs_init(fsc);
677 if (err < 0)
678 goto fail;
679 }
680
681 if (path[0] == 0) {
682 dget(root);
683 } else {
684 dout("mount opening base mountpoint\n");
685 root = open_root_dentry(fsc, path, started);
686 if (IS_ERR(root)) {
687 err = PTR_ERR(root);
688 goto fail;
689 }
690 }
691
692 fsc->mount_state = CEPH_MOUNT_MOUNTED;
693 dout("mount success\n");
694 mutex_unlock(&fsc->client->mount_mutex);
695 return root;
696
697out:
698 mutex_unlock(&fsc->client->mount_mutex);
699 return ERR_PTR(err);
700
701fail:
702 if (first) {
703 dput(fsc->sb->s_root);
704 fsc->sb->s_root = NULL;
705 }
706 goto out;
707}
708
709static int ceph_set_super(struct super_block *s, void *data)
710{
711 struct ceph_fs_client *fsc = data;
712 int ret;
713
714 dout("set_super %p data %p\n", s, data);
715
716 s->s_flags = fsc->mount_options->sb_flags;
717 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
718
719 s->s_fs_info = fsc;
720 fsc->sb = s;
721
722 s->s_op = &ceph_super_ops;
723 s->s_export_op = &ceph_export_ops;
724
725 s->s_time_gran = 1000; /* 1000 ns == 1 us */
726
727 ret = set_anon_super(s, NULL); /* what is that second arg for? */
728 if (ret != 0)
729 goto fail;
730
731 return ret;
732
733fail:
734 s->s_fs_info = NULL;
735 fsc->sb = NULL;
736 return ret;
737}
738
739/*
740 * share superblock if same fs AND options
741 */
742static int ceph_compare_super(struct super_block *sb, void *data)
743{
744 struct ceph_fs_client *new = data;
745 struct ceph_mount_options *fsopt = new->mount_options;
746 struct ceph_options *opt = new->client->options;
747 struct ceph_fs_client *other = ceph_sb_to_client(sb);
748
749 dout("ceph_compare_super %p\n", sb);
750
751 if (compare_mount_options(fsopt, opt, other)) {
752 dout("monitor(s)/mount options don't match\n");
753 return 0;
754 }
755 if ((opt->flags & CEPH_OPT_FSID) &&
756 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
757 dout("fsid doesn't match\n");
758 return 0;
759 }
760 if (fsopt->sb_flags != other->mount_options->sb_flags) {
761 dout("flags differ\n");
762 return 0;
763 }
764 return 1;
765}
766
767/*
768 * construct our own bdi so we can control readahead, etc.
769 */
770static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
771
772static int ceph_register_bdi(struct super_block *sb,
773 struct ceph_fs_client *fsc)
774{
775 int err;
776
777 /* set ra_pages based on rsize mount option? */
778 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
779 fsc->backing_dev_info.ra_pages =
780 (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
781 >> PAGE_SHIFT;
782 else
783 fsc->backing_dev_info.ra_pages =
784 default_backing_dev_info.ra_pages;
785
786 err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
787 atomic_long_inc_return(&bdi_seq));
788 if (!err)
789 sb->s_bdi = &fsc->backing_dev_info;
790 return err;
791}
792
793static struct dentry *ceph_mount(struct file_system_type *fs_type,
794 int flags, const char *dev_name, void *data)
795{
796 struct super_block *sb;
797 struct ceph_fs_client *fsc;
798 struct dentry *res;
799 int err;
800 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
801 const char *path = NULL;
802 struct ceph_mount_options *fsopt = NULL;
803 struct ceph_options *opt = NULL;
804
805 dout("ceph_mount\n");
806 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
807 if (err < 0) {
808 res = ERR_PTR(err);
809 goto out_final;
810 }
811
812 /* create client (which we may/may not use) */
813 fsc = create_fs_client(fsopt, opt);
814 if (IS_ERR(fsc)) {
815 res = ERR_CAST(fsc);
816 destroy_mount_options(fsopt);
817 ceph_destroy_options(opt);
818 goto out_final;
819 }
820
821 err = ceph_mdsc_init(fsc);
822 if (err < 0) {
823 res = ERR_PTR(err);
824 goto out;
825 }
826
827 if (ceph_test_opt(fsc->client, NOSHARE))
828 compare_super = NULL;
829 sb = sget(fs_type, compare_super, ceph_set_super, fsc);
830 if (IS_ERR(sb)) {
831 res = ERR_CAST(sb);
832 goto out;
833 }
834
835 if (ceph_sb_to_client(sb) != fsc) {
836 ceph_mdsc_destroy(fsc);
837 destroy_fs_client(fsc);
838 fsc = ceph_sb_to_client(sb);
839 dout("get_sb got existing client %p\n", fsc);
840 } else {
841 dout("get_sb using new client %p\n", fsc);
842 err = ceph_register_bdi(sb, fsc);
843 if (err < 0) {
844 res = ERR_PTR(err);
845 goto out_splat;
846 }
847 }
848
849 res = ceph_real_mount(fsc, path);
850 if (IS_ERR(res))
851 goto out_splat;
852 dout("root %p inode %p ino %llx.%llx\n", res,
853 res->d_inode, ceph_vinop(res->d_inode));
854 return res;
855
856out_splat:
857 ceph_mdsc_close_sessions(fsc->mdsc);
858 deactivate_locked_super(sb);
859 goto out_final;
860
861out:
862 ceph_mdsc_destroy(fsc);
863 destroy_fs_client(fsc);
864out_final:
865 dout("ceph_mount fail %ld\n", PTR_ERR(res));
866 return res;
867}
868
869static void ceph_kill_sb(struct super_block *s)
870{
871 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
872 dout("kill_sb %p\n", s);
873 ceph_mdsc_pre_umount(fsc->mdsc);
874 kill_anon_super(s); /* will call put_super after sb is r/o */
875 ceph_mdsc_destroy(fsc);
876 destroy_fs_client(fsc);
877}
878
879static struct file_system_type ceph_fs_type = {
880 .owner = THIS_MODULE,
881 .name = "ceph",
882 .mount = ceph_mount,
883 .kill_sb = ceph_kill_sb,
884 .fs_flags = FS_RENAME_DOES_D_MOVE,
885};
886
887#define _STRINGIFY(x) #x
888#define STRINGIFY(x) _STRINGIFY(x)
889
890static int __init init_ceph(void)
891{
892 int ret = init_caches();
893 if (ret)
894 goto out;
895
896 ret = register_filesystem(&ceph_fs_type);
897 if (ret)
898 goto out_icache;
899
900 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
901
902 return 0;
903
904out_icache:
905 destroy_caches();
906out:
907 return ret;
908}
909
910static void __exit exit_ceph(void)
911{
912 dout("exit_ceph\n");
913 unregister_filesystem(&ceph_fs_type);
914 destroy_caches();
915}
916
917module_init(init_ceph);
918module_exit(exit_ceph);
919
920MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
921MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
922MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
923MODULE_DESCRIPTION("Ceph filesystem for Linux");
924MODULE_LICENSE("GPL");
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 */
37static 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
45static 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
89static 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 */
110enum {
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
145static 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
179static 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
296static 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
303static 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
314static 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
334static 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
411out:
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 */
421static 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))
443 seq_puts(m, ",rbytes");
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 */
491static 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 */
509static 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 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP, 0, true);
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_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
579fail_fscache:
580 ceph_fscache_unregister_fs(fsc);
581fail_trunc_wq:
582 destroy_workqueue(fsc->trunc_wq);
583fail_pg_inv_wq:
584 destroy_workqueue(fsc->pg_inv_wq);
585fail_wb_wq:
586 destroy_workqueue(fsc->wb_wq);
587fail_bdi:
588 bdi_destroy(&fsc->backing_dev_info);
589fail_client:
590 ceph_destroy_client(fsc->client);
591fail:
592 kfree(fsc);
593 return ERR_PTR(err);
594}
595
596static 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 */
623struct kmem_cache *ceph_inode_cachep;
624struct kmem_cache *ceph_cap_cachep;
625struct kmem_cache *ceph_cap_flush_cachep;
626struct kmem_cache *ceph_dentry_cachep;
627struct kmem_cache *ceph_file_cachep;
628
629static void ceph_inode_init_once(void *foo)
630{
631 struct ceph_inode_info *ci = foo;
632 inode_init_once(&ci->vfs_inode);
633}
634
635static 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 SLAB_ACCOUNT, 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;
670bad_file:
671 kmem_cache_destroy(ceph_dentry_cachep);
672bad_dentry:
673 kmem_cache_destroy(ceph_cap_flush_cachep);
674bad_cap_flush:
675 kmem_cache_destroy(ceph_cap_cachep);
676bad_cap:
677 kmem_cache_destroy(ceph_inode_cachep);
678 return error;
679}
680
681static 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 */
703static 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
715static 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 */
731static 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 }
777out:
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 */
788static 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 %p\n", fsc);
797 mutex_lock(&fsc->client->mount_mutex);
798
799 if (!fsc->sb->s_root) {
800 err = __ceph_open_session(fsc->client, started);
801 if (err < 0)
802 goto out;
803
804 dout("mount opening root\n");
805 root = open_root_dentry(fsc, "", started);
806 if (IS_ERR(root)) {
807 err = PTR_ERR(root);
808 goto out;
809 }
810 fsc->sb->s_root = root;
811 first = 1;
812
813 err = ceph_fs_debugfs_init(fsc);
814 if (err < 0)
815 goto fail;
816 }
817
818 if (path[0] == 0) {
819 root = fsc->sb->s_root;
820 dget(root);
821 } else {
822 dout("mount opening base mountpoint\n");
823 root = open_root_dentry(fsc, path, started);
824 if (IS_ERR(root)) {
825 err = PTR_ERR(root);
826 goto fail;
827 }
828 }
829
830 fsc->mount_state = CEPH_MOUNT_MOUNTED;
831 dout("mount success\n");
832 mutex_unlock(&fsc->client->mount_mutex);
833 return root;
834
835fail:
836 if (first) {
837 dput(fsc->sb->s_root);
838 fsc->sb->s_root = NULL;
839 }
840out:
841 mutex_unlock(&fsc->client->mount_mutex);
842 return ERR_PTR(err);
843}
844
845static int ceph_set_super(struct super_block *s, void *data)
846{
847 struct ceph_fs_client *fsc = data;
848 int ret;
849
850 dout("set_super %p data %p\n", s, data);
851
852 s->s_flags = fsc->mount_options->sb_flags;
853 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
854
855 s->s_xattr = ceph_xattr_handlers;
856 s->s_fs_info = fsc;
857 fsc->sb = s;
858
859 s->s_op = &ceph_super_ops;
860 s->s_export_op = &ceph_export_ops;
861
862 s->s_time_gran = 1000; /* 1000 ns == 1 us */
863
864 ret = set_anon_super(s, NULL); /* what is that second arg for? */
865 if (ret != 0)
866 goto fail;
867
868 return ret;
869
870fail:
871 s->s_fs_info = NULL;
872 fsc->sb = NULL;
873 return ret;
874}
875
876/*
877 * share superblock if same fs AND options
878 */
879static int ceph_compare_super(struct super_block *sb, void *data)
880{
881 struct ceph_fs_client *new = data;
882 struct ceph_mount_options *fsopt = new->mount_options;
883 struct ceph_options *opt = new->client->options;
884 struct ceph_fs_client *other = ceph_sb_to_client(sb);
885
886 dout("ceph_compare_super %p\n", sb);
887
888 if (compare_mount_options(fsopt, opt, other)) {
889 dout("monitor(s)/mount options don't match\n");
890 return 0;
891 }
892 if ((opt->flags & CEPH_OPT_FSID) &&
893 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
894 dout("fsid doesn't match\n");
895 return 0;
896 }
897 if (fsopt->sb_flags != other->mount_options->sb_flags) {
898 dout("flags differ\n");
899 return 0;
900 }
901 return 1;
902}
903
904/*
905 * construct our own bdi so we can control readahead, etc.
906 */
907static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
908
909static int ceph_register_bdi(struct super_block *sb,
910 struct ceph_fs_client *fsc)
911{
912 int err;
913
914 /* set ra_pages based on rasize mount option? */
915 if (fsc->mount_options->rasize >= PAGE_SIZE)
916 fsc->backing_dev_info.ra_pages =
917 (fsc->mount_options->rasize + PAGE_SIZE - 1)
918 >> PAGE_SHIFT;
919 else
920 fsc->backing_dev_info.ra_pages =
921 VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
922
923 err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
924 atomic_long_inc_return(&bdi_seq));
925 if (!err)
926 sb->s_bdi = &fsc->backing_dev_info;
927 return err;
928}
929
930static struct dentry *ceph_mount(struct file_system_type *fs_type,
931 int flags, const char *dev_name, void *data)
932{
933 struct super_block *sb;
934 struct ceph_fs_client *fsc;
935 struct dentry *res;
936 int err;
937 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
938 const char *path = NULL;
939 struct ceph_mount_options *fsopt = NULL;
940 struct ceph_options *opt = NULL;
941
942 dout("ceph_mount\n");
943
944#ifdef CONFIG_CEPH_FS_POSIX_ACL
945 flags |= MS_POSIXACL;
946#endif
947 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
948 if (err < 0) {
949 res = ERR_PTR(err);
950 goto out_final;
951 }
952
953 /* create client (which we may/may not use) */
954 fsc = create_fs_client(fsopt, opt);
955 if (IS_ERR(fsc)) {
956 res = ERR_CAST(fsc);
957 destroy_mount_options(fsopt);
958 ceph_destroy_options(opt);
959 goto out_final;
960 }
961
962 err = ceph_mdsc_init(fsc);
963 if (err < 0) {
964 res = ERR_PTR(err);
965 goto out;
966 }
967
968 if (ceph_test_opt(fsc->client, NOSHARE))
969 compare_super = NULL;
970 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
971 if (IS_ERR(sb)) {
972 res = ERR_CAST(sb);
973 goto out;
974 }
975
976 if (ceph_sb_to_client(sb) != fsc) {
977 ceph_mdsc_destroy(fsc);
978 destroy_fs_client(fsc);
979 fsc = ceph_sb_to_client(sb);
980 dout("get_sb got existing client %p\n", fsc);
981 } else {
982 dout("get_sb using new client %p\n", fsc);
983 err = ceph_register_bdi(sb, fsc);
984 if (err < 0) {
985 res = ERR_PTR(err);
986 goto out_splat;
987 }
988 }
989
990 res = ceph_real_mount(fsc, path);
991 if (IS_ERR(res))
992 goto out_splat;
993 dout("root %p inode %p ino %llx.%llx\n", res,
994 d_inode(res), ceph_vinop(d_inode(res)));
995 return res;
996
997out_splat:
998 ceph_mdsc_close_sessions(fsc->mdsc);
999 deactivate_locked_super(sb);
1000 goto out_final;
1001
1002out:
1003 ceph_mdsc_destroy(fsc);
1004 destroy_fs_client(fsc);
1005out_final:
1006 dout("ceph_mount fail %ld\n", PTR_ERR(res));
1007 return res;
1008}
1009
1010static void ceph_kill_sb(struct super_block *s)
1011{
1012 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1013 dev_t dev = s->s_dev;
1014
1015 dout("kill_sb %p\n", s);
1016
1017 ceph_mdsc_pre_umount(fsc->mdsc);
1018 generic_shutdown_super(s);
1019 ceph_mdsc_destroy(fsc);
1020
1021 destroy_fs_client(fsc);
1022 free_anon_bdev(dev);
1023}
1024
1025static struct file_system_type ceph_fs_type = {
1026 .owner = THIS_MODULE,
1027 .name = "ceph",
1028 .mount = ceph_mount,
1029 .kill_sb = ceph_kill_sb,
1030 .fs_flags = FS_RENAME_DOES_D_MOVE,
1031};
1032MODULE_ALIAS_FS("ceph");
1033
1034static int __init init_ceph(void)
1035{
1036 int ret = init_caches();
1037 if (ret)
1038 goto out;
1039
1040 ceph_flock_init();
1041 ceph_xattr_init();
1042 ret = register_filesystem(&ceph_fs_type);
1043 if (ret)
1044 goto out_xattr;
1045
1046 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1047
1048 return 0;
1049
1050out_xattr:
1051 ceph_xattr_exit();
1052 destroy_caches();
1053out:
1054 return ret;
1055}
1056
1057static void __exit exit_ceph(void)
1058{
1059 dout("exit_ceph\n");
1060 unregister_filesystem(&ceph_fs_type);
1061 ceph_xattr_exit();
1062 destroy_caches();
1063}
1064
1065module_init(init_ceph);
1066module_exit(exit_ceph);
1067
1068MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1069MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1070MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1071MODULE_DESCRIPTION("Ceph filesystem for Linux");
1072MODULE_LICENSE("GPL");