2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/backing-dev.h>
5 #include <linux/ctype.h>
7 #include <linux/inet.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>
19 #include "mds_client.h"
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>
29 * Ceph superblock operations
31 * Handle the basics of mounting, unmounting.
37 static void ceph_put_super(struct super_block *s)
39 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
42 ceph_mdsc_close_sessions(fsc->mdsc);
45 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
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;
54 if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
55 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
57 data_pool = CEPH_NOPOOL;
61 err = ceph_monc_do_statfs(&fsc->client->monc, data_pool, &st);
66 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
69 * express utilization in terms of large blocks to avoid
70 * overflow on 32-bit machines.
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!
77 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
78 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
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.
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);
92 buf->f_files = le64_to_cpu(st.num_objects);
94 buf->f_namelen = NAME_MAX;
96 /* Must convert the fsid, for consistent values across arches */
97 fsid = le64_to_cpu(*(__le64 *)(&monmap->fsid)) ^
98 le64_to_cpu(*((__le64 *)&monmap->fsid + 1));
99 buf->f_fsid.val[0] = fsid & 0xffffffff;
100 buf->f_fsid.val[1] = fsid >> 32;
106 static int ceph_sync_fs(struct super_block *sb, int wait)
108 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
111 dout("sync_fs (non-blocking)\n");
112 ceph_flush_dirty_caps(fsc->mdsc);
113 dout("sync_fs (non-blocking) done\n");
117 dout("sync_fs (blocking)\n");
118 ceph_osdc_sync(&fsc->client->osdc);
119 ceph_mdsc_sync(fsc->mdsc);
120 dout("sync_fs (blocking) done\n");
131 Opt_caps_wanted_delay_min,
132 Opt_caps_wanted_delay_max,
133 Opt_readdir_max_entries,
134 Opt_readdir_max_bytes,
142 /* string args above */
157 Opt_require_active_mds,
158 Opt_norequire_active_mds,
159 #ifdef CONFIG_CEPH_FS_POSIX_ACL
167 static match_table_t fsopt_tokens = {
168 {Opt_wsize, "wsize=%d"},
169 {Opt_rsize, "rsize=%d"},
170 {Opt_rasize, "rasize=%d"},
171 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
172 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
173 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
174 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
175 {Opt_congestion_kb, "write_congestion_kb=%d"},
177 {Opt_snapdirname, "snapdirname=%s"},
178 {Opt_mds_namespace, "mds_namespace=%s"},
179 {Opt_fscache_uniq, "fsc=%s"},
180 /* string args above */
181 {Opt_dirstat, "dirstat"},
182 {Opt_nodirstat, "nodirstat"},
183 {Opt_rbytes, "rbytes"},
184 {Opt_norbytes, "norbytes"},
185 {Opt_asyncreaddir, "asyncreaddir"},
186 {Opt_noasyncreaddir, "noasyncreaddir"},
187 {Opt_dcache, "dcache"},
188 {Opt_nodcache, "nodcache"},
189 {Opt_ino32, "ino32"},
190 {Opt_noino32, "noino32"},
191 {Opt_fscache, "fsc"},
192 {Opt_nofscache, "nofsc"},
193 {Opt_poolperm, "poolperm"},
194 {Opt_nopoolperm, "nopoolperm"},
195 {Opt_require_active_mds, "require_active_mds"},
196 {Opt_norequire_active_mds, "norequire_active_mds"},
197 #ifdef CONFIG_CEPH_FS_POSIX_ACL
200 {Opt_noacl, "noacl"},
201 {Opt_quotadf, "quotadf"},
202 {Opt_noquotadf, "noquotadf"},
206 static int parse_fsopt_token(char *c, void *private)
208 struct ceph_mount_options *fsopt = private;
209 substring_t argstr[MAX_OPT_ARGS];
210 int token, intval, ret;
212 token = match_token((char *)c, fsopt_tokens, argstr);
216 if (token < Opt_last_int) {
217 ret = match_int(&argstr[0], &intval);
219 pr_err("bad mount option arg (not int) "
223 dout("got int token %d val %d\n", token, intval);
224 } else if (token > Opt_last_int && token < Opt_last_string) {
225 dout("got string token %d val %s\n", token,
228 dout("got token %d\n", token);
232 case Opt_snapdirname:
233 kfree(fsopt->snapdir_name);
234 fsopt->snapdir_name = kstrndup(argstr[0].from,
235 argstr[0].to-argstr[0].from,
237 if (!fsopt->snapdir_name)
240 case Opt_mds_namespace:
241 kfree(fsopt->mds_namespace);
242 fsopt->mds_namespace = kstrndup(argstr[0].from,
243 argstr[0].to-argstr[0].from,
245 if (!fsopt->mds_namespace)
248 case Opt_fscache_uniq:
249 kfree(fsopt->fscache_uniq);
250 fsopt->fscache_uniq = kstrndup(argstr[0].from,
251 argstr[0].to-argstr[0].from,
253 if (!fsopt->fscache_uniq)
255 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
259 if (intval < PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
261 fsopt->wsize = ALIGN(intval, PAGE_SIZE);
264 if (intval < PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
266 fsopt->rsize = ALIGN(intval, PAGE_SIZE);
271 fsopt->rasize = ALIGN(intval + PAGE_SIZE - 1, PAGE_SIZE);
273 case Opt_caps_wanted_delay_min:
276 fsopt->caps_wanted_delay_min = intval;
278 case Opt_caps_wanted_delay_max:
281 fsopt->caps_wanted_delay_max = intval;
283 case Opt_readdir_max_entries:
286 fsopt->max_readdir = intval;
288 case Opt_readdir_max_bytes:
289 if (intval < PAGE_SIZE && intval != 0)
291 fsopt->max_readdir_bytes = intval;
293 case Opt_congestion_kb:
294 if (intval < 1024) /* at least 1M */
296 fsopt->congestion_kb = intval;
299 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
302 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
305 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
308 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
310 case Opt_asyncreaddir:
311 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
313 case Opt_noasyncreaddir:
314 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
317 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
320 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
323 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
326 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
329 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
330 kfree(fsopt->fscache_uniq);
331 fsopt->fscache_uniq = NULL;
334 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
335 kfree(fsopt->fscache_uniq);
336 fsopt->fscache_uniq = NULL;
339 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
342 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
344 case Opt_require_active_mds:
345 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
347 case Opt_norequire_active_mds:
348 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
351 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
354 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
356 #ifdef CONFIG_CEPH_FS_POSIX_ACL
358 fsopt->sb_flags |= SB_POSIXACL;
362 fsopt->sb_flags &= ~SB_POSIXACL;
370 static void destroy_mount_options(struct ceph_mount_options *args)
372 dout("destroy_mount_options %p\n", args);
373 kfree(args->snapdir_name);
374 kfree(args->mds_namespace);
375 kfree(args->server_path);
376 kfree(args->fscache_uniq);
380 static int strcmp_null(const char *s1, const char *s2)
388 return strcmp(s1, s2);
391 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
392 struct ceph_options *new_opt,
393 struct ceph_fs_client *fsc)
395 struct ceph_mount_options *fsopt1 = new_fsopt;
396 struct ceph_mount_options *fsopt2 = fsc->mount_options;
397 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
400 ret = memcmp(fsopt1, fsopt2, ofs);
404 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
407 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
410 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
413 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
417 return ceph_compare_options(new_opt, fsc->client);
420 static int parse_mount_options(struct ceph_mount_options **pfsopt,
421 struct ceph_options **popt,
422 int flags, char *options,
423 const char *dev_name)
425 struct ceph_mount_options *fsopt;
426 const char *dev_name_end;
429 if (!dev_name || !*dev_name)
432 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
436 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
438 fsopt->sb_flags = flags;
439 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
441 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
442 fsopt->rsize = CEPH_MAX_READ_SIZE;
443 fsopt->rasize = CEPH_RASIZE_DEFAULT;
444 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
445 if (!fsopt->snapdir_name) {
450 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
451 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
452 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
453 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
454 fsopt->congestion_kb = default_congestion_kb();
457 * Distinguish the server list from the path in "dev_name".
458 * Internally we do not include the leading '/' in the path.
460 * "dev_name" will look like:
461 * <server_spec>[,<server_spec>...]:[<path>]
463 * <server_spec> is <ip>[:<port>]
464 * <path> is optional, but if present must begin with '/'
466 dev_name_end = strchr(dev_name, '/');
468 if (strlen(dev_name_end) > 1) {
469 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
470 if (!fsopt->server_path) {
476 dev_name_end = dev_name + strlen(dev_name);
479 dev_name_end--; /* back up to ':' separator */
480 if (dev_name_end < dev_name || *dev_name_end != ':') {
481 pr_err("device name is missing path (no : separator in %s)\n",
485 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
486 if (fsopt->server_path)
487 dout("server path '%s'\n", fsopt->server_path);
489 *popt = ceph_parse_options(options, dev_name, dev_name_end,
490 parse_fsopt_token, (void *)fsopt);
492 err = PTR_ERR(*popt);
501 destroy_mount_options(fsopt);
506 * ceph_show_options - Show mount options in /proc/mounts
507 * @m: seq_file to write to
508 * @root: root of that (sub)tree
510 static int ceph_show_options(struct seq_file *m, struct dentry *root)
512 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
513 struct ceph_mount_options *fsopt = fsc->mount_options;
517 /* a comma between MNT/MS and client options */
521 ret = ceph_print_client_options(m, fsc->client);
525 /* retract our comma if no client options */
529 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
530 seq_puts(m, ",dirstat");
531 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
532 seq_puts(m, ",rbytes");
533 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
534 seq_puts(m, ",noasyncreaddir");
535 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
536 seq_puts(m, ",nodcache");
537 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
538 seq_show_option(m, "fsc", fsopt->fscache_uniq);
540 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
541 seq_puts(m, ",nopoolperm");
542 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
543 seq_puts(m, ",noquotadf");
545 #ifdef CONFIG_CEPH_FS_POSIX_ACL
546 if (fsopt->sb_flags & SB_POSIXACL)
549 seq_puts(m, ",noacl");
552 if (fsopt->mds_namespace)
553 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
554 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
555 seq_printf(m, ",wsize=%d", fsopt->wsize);
556 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
557 seq_printf(m, ",rsize=%d", fsopt->rsize);
558 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
559 seq_printf(m, ",rasize=%d", fsopt->rasize);
560 if (fsopt->congestion_kb != default_congestion_kb())
561 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
562 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
563 seq_printf(m, ",caps_wanted_delay_min=%d",
564 fsopt->caps_wanted_delay_min);
565 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
566 seq_printf(m, ",caps_wanted_delay_max=%d",
567 fsopt->caps_wanted_delay_max);
568 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
569 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
570 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
571 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
572 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
573 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
579 * handle any mon messages the standard library doesn't understand.
580 * return error if we don't either.
582 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
584 struct ceph_fs_client *fsc = client->private;
585 int type = le16_to_cpu(msg->hdr.type);
588 case CEPH_MSG_MDS_MAP:
589 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
591 case CEPH_MSG_FS_MAP_USER:
592 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
600 * create a new fs client
602 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
603 struct ceph_options *opt)
605 struct ceph_fs_client *fsc;
610 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
612 return ERR_PTR(-ENOMEM);
614 fsc->client = ceph_create_client(opt, fsc);
615 if (IS_ERR(fsc->client)) {
616 err = PTR_ERR(fsc->client);
620 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
621 fsc->client->osdc.abort_on_full = true;
623 if (!fsopt->mds_namespace) {
624 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
627 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
631 fsc->mount_options = fsopt;
634 fsc->mount_state = CEPH_MOUNT_MOUNTING;
636 atomic_long_set(&fsc->writeback_count, 0);
640 * The number of concurrent works can be high but they don't need
641 * to be processed in parallel, limit concurrency.
643 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
646 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
649 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
653 /* set up mempools */
655 page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
656 size = sizeof (struct page *) * (page_count ? page_count : 1);
657 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
658 if (!fsc->wb_pagevec_pool)
662 fsc->min_caps = fsopt->max_readdir;
667 destroy_workqueue(fsc->trunc_wq);
669 destroy_workqueue(fsc->pg_inv_wq);
671 destroy_workqueue(fsc->wb_wq);
673 ceph_destroy_client(fsc->client);
679 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
681 flush_workqueue(fsc->wb_wq);
682 flush_workqueue(fsc->pg_inv_wq);
683 flush_workqueue(fsc->trunc_wq);
686 static void destroy_fs_client(struct ceph_fs_client *fsc)
688 dout("destroy_fs_client %p\n", fsc);
690 destroy_workqueue(fsc->wb_wq);
691 destroy_workqueue(fsc->pg_inv_wq);
692 destroy_workqueue(fsc->trunc_wq);
694 mempool_destroy(fsc->wb_pagevec_pool);
696 destroy_mount_options(fsc->mount_options);
698 ceph_destroy_client(fsc->client);
701 dout("destroy_fs_client %p done\n", fsc);
707 struct kmem_cache *ceph_inode_cachep;
708 struct kmem_cache *ceph_cap_cachep;
709 struct kmem_cache *ceph_cap_flush_cachep;
710 struct kmem_cache *ceph_dentry_cachep;
711 struct kmem_cache *ceph_file_cachep;
712 struct kmem_cache *ceph_dir_file_cachep;
714 static void ceph_inode_init_once(void *foo)
716 struct ceph_inode_info *ci = foo;
717 inode_init_once(&ci->vfs_inode);
720 static int __init init_caches(void)
724 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
725 sizeof(struct ceph_inode_info),
726 __alignof__(struct ceph_inode_info),
727 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
728 SLAB_ACCOUNT, ceph_inode_init_once);
729 if (!ceph_inode_cachep)
732 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
733 if (!ceph_cap_cachep)
735 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
736 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
737 if (!ceph_cap_flush_cachep)
740 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
741 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
742 if (!ceph_dentry_cachep)
745 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
746 if (!ceph_file_cachep)
749 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
750 if (!ceph_dir_file_cachep)
753 error = ceph_fscache_register();
760 kmem_cache_destroy(ceph_dir_file_cachep);
762 kmem_cache_destroy(ceph_file_cachep);
764 kmem_cache_destroy(ceph_dentry_cachep);
766 kmem_cache_destroy(ceph_cap_flush_cachep);
768 kmem_cache_destroy(ceph_cap_cachep);
770 kmem_cache_destroy(ceph_inode_cachep);
774 static void destroy_caches(void)
777 * Make sure all delayed rcu free inodes are flushed before we
782 kmem_cache_destroy(ceph_inode_cachep);
783 kmem_cache_destroy(ceph_cap_cachep);
784 kmem_cache_destroy(ceph_cap_flush_cachep);
785 kmem_cache_destroy(ceph_dentry_cachep);
786 kmem_cache_destroy(ceph_file_cachep);
787 kmem_cache_destroy(ceph_dir_file_cachep);
789 ceph_fscache_unregister();
794 * ceph_umount_begin - initiate forced umount. Tear down down the
795 * mount, skipping steps that may hang while waiting for server(s).
797 static void ceph_umount_begin(struct super_block *sb)
799 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
801 dout("ceph_umount_begin - starting forced umount\n");
804 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
805 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
806 ceph_mdsc_force_umount(fsc->mdsc);
810 static const struct super_operations ceph_super_ops = {
811 .alloc_inode = ceph_alloc_inode,
812 .destroy_inode = ceph_destroy_inode,
813 .write_inode = ceph_write_inode,
814 .drop_inode = ceph_drop_inode,
815 .sync_fs = ceph_sync_fs,
816 .put_super = ceph_put_super,
817 .show_options = ceph_show_options,
818 .statfs = ceph_statfs,
819 .umount_begin = ceph_umount_begin,
823 * Bootstrap mount by opening the root directory. Note the mount
824 * @started time from caller, and time out if this takes too long.
826 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
828 unsigned long started)
830 struct ceph_mds_client *mdsc = fsc->mdsc;
831 struct ceph_mds_request *req = NULL;
836 dout("open_root_inode opening '%s'\n", path);
837 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
839 return ERR_CAST(req);
840 req->r_path1 = kstrdup(path, GFP_NOFS);
842 root = ERR_PTR(-ENOMEM);
846 req->r_ino1.ino = CEPH_INO_ROOT;
847 req->r_ino1.snap = CEPH_NOSNAP;
848 req->r_started = started;
849 req->r_timeout = fsc->client->options->mount_timeout;
850 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
852 err = ceph_mdsc_do_request(mdsc, NULL, req);
854 struct inode *inode = req->r_target_inode;
855 req->r_target_inode = NULL;
856 dout("open_root_inode success\n");
857 root = d_make_root(inode);
859 root = ERR_PTR(-ENOMEM);
862 dout("open_root_inode success, root dentry is %p\n", root);
867 ceph_mdsc_put_request(req);
875 * mount: join the ceph cluster, and open root directory.
877 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
880 unsigned long started = jiffies; /* note the start time */
883 dout("mount start %p\n", fsc);
884 mutex_lock(&fsc->client->mount_mutex);
886 if (!fsc->sb->s_root) {
888 err = __ceph_open_session(fsc->client, started);
893 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
894 err = ceph_fscache_register_fs(fsc);
899 if (!fsc->mount_options->server_path) {
901 dout("mount opening path \\t\n");
903 path = fsc->mount_options->server_path + 1;
904 dout("mount opening path %s\n", path);
907 err = ceph_fs_debugfs_init(fsc);
911 root = open_root_dentry(fsc, path, started);
916 fsc->sb->s_root = dget(root);
918 root = dget(fsc->sb->s_root);
921 fsc->mount_state = CEPH_MOUNT_MOUNTED;
922 dout("mount success\n");
923 mutex_unlock(&fsc->client->mount_mutex);
927 mutex_unlock(&fsc->client->mount_mutex);
931 static int ceph_set_super(struct super_block *s, void *data)
933 struct ceph_fs_client *fsc = data;
936 dout("set_super %p data %p\n", s, data);
938 s->s_flags = fsc->mount_options->sb_flags;
939 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
941 s->s_xattr = ceph_xattr_handlers;
945 s->s_op = &ceph_super_ops;
946 s->s_d_op = &ceph_dentry_ops;
947 s->s_export_op = &ceph_export_ops;
949 s->s_time_gran = 1000; /* 1000 ns == 1 us */
951 ret = set_anon_super(s, NULL); /* what is that second arg for? */
964 * share superblock if same fs AND options
966 static int ceph_compare_super(struct super_block *sb, void *data)
968 struct ceph_fs_client *new = data;
969 struct ceph_mount_options *fsopt = new->mount_options;
970 struct ceph_options *opt = new->client->options;
971 struct ceph_fs_client *other = ceph_sb_to_client(sb);
973 dout("ceph_compare_super %p\n", sb);
975 if (compare_mount_options(fsopt, opt, other)) {
976 dout("monitor(s)/mount options don't match\n");
979 if ((opt->flags & CEPH_OPT_FSID) &&
980 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
981 dout("fsid doesn't match\n");
984 if (fsopt->sb_flags != other->mount_options->sb_flags) {
985 dout("flags differ\n");
992 * construct our own bdi so we can control readahead, etc.
994 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
996 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1000 err = super_setup_bdi_name(sb, "ceph-%ld",
1001 atomic_long_inc_return(&bdi_seq));
1005 /* set ra_pages based on rasize mount option? */
1006 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1008 /* set io_pages based on max osd read size */
1009 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1014 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1015 int flags, const char *dev_name, void *data)
1017 struct super_block *sb;
1018 struct ceph_fs_client *fsc;
1021 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1022 struct ceph_mount_options *fsopt = NULL;
1023 struct ceph_options *opt = NULL;
1025 dout("ceph_mount\n");
1027 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1028 flags |= SB_POSIXACL;
1030 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1036 /* create client (which we may/may not use) */
1037 fsc = create_fs_client(fsopt, opt);
1039 res = ERR_CAST(fsc);
1040 destroy_mount_options(fsopt);
1041 ceph_destroy_options(opt);
1045 err = ceph_mdsc_init(fsc);
1051 if (ceph_test_opt(fsc->client, NOSHARE))
1052 compare_super = NULL;
1053 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1059 if (ceph_sb_to_client(sb) != fsc) {
1060 ceph_mdsc_destroy(fsc);
1061 destroy_fs_client(fsc);
1062 fsc = ceph_sb_to_client(sb);
1063 dout("get_sb got existing client %p\n", fsc);
1065 dout("get_sb using new client %p\n", fsc);
1066 err = ceph_setup_bdi(sb, fsc);
1073 res = ceph_real_mount(fsc);
1076 dout("root %p inode %p ino %llx.%llx\n", res,
1077 d_inode(res), ceph_vinop(d_inode(res)));
1081 ceph_mdsc_close_sessions(fsc->mdsc);
1082 deactivate_locked_super(sb);
1086 ceph_mdsc_destroy(fsc);
1087 destroy_fs_client(fsc);
1089 dout("ceph_mount fail %ld\n", PTR_ERR(res));
1093 static void ceph_kill_sb(struct super_block *s)
1095 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1096 dev_t dev = s->s_dev;
1098 dout("kill_sb %p\n", s);
1100 ceph_mdsc_pre_umount(fsc->mdsc);
1101 flush_fs_workqueues(fsc);
1103 generic_shutdown_super(s);
1105 fsc->client->extra_mon_dispatch = NULL;
1106 ceph_fs_debugfs_cleanup(fsc);
1108 ceph_fscache_unregister_fs(fsc);
1110 ceph_mdsc_destroy(fsc);
1112 destroy_fs_client(fsc);
1113 free_anon_bdev(dev);
1116 static struct file_system_type ceph_fs_type = {
1117 .owner = THIS_MODULE,
1119 .mount = ceph_mount,
1120 .kill_sb = ceph_kill_sb,
1121 .fs_flags = FS_RENAME_DOES_D_MOVE,
1123 MODULE_ALIAS_FS("ceph");
1125 static int __init init_ceph(void)
1127 int ret = init_caches();
1133 ret = register_filesystem(&ceph_fs_type);
1137 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1148 static void __exit exit_ceph(void)
1150 dout("exit_ceph\n");
1151 unregister_filesystem(&ceph_fs_type);
1156 module_init(init_ceph);
1157 module_exit(exit_ceph);
1159 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1160 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1161 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1162 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1163 MODULE_LICENSE("GPL");