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 err = ceph_monc_do_statfs(&fsc->client->monc, &st);
59 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
62 * express utilization in terms of large blocks to avoid
63 * overflow on 32-bit machines.
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!
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);
76 buf->f_files = le64_to_cpu(st.num_objects);
78 buf->f_namelen = NAME_MAX;
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;
88 static int ceph_sync_fs(struct super_block *sb, int wait)
90 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
93 dout("sync_fs (non-blocking)\n");
94 ceph_flush_dirty_caps(fsc->mdsc);
95 dout("sync_fs (non-blocking) done\n");
99 dout("sync_fs (blocking)\n");
100 ceph_osdc_sync(&fsc->client->osdc);
101 ceph_mdsc_sync(fsc->mdsc);
102 dout("sync_fs (blocking) done\n");
113 Opt_caps_wanted_delay_min,
114 Opt_caps_wanted_delay_max,
115 Opt_cap_release_safety,
116 Opt_readdir_max_entries,
117 Opt_readdir_max_bytes,
124 /* string args above */
139 #ifdef CONFIG_CEPH_FS_POSIX_ACL
145 static match_table_t fsopt_tokens = {
146 {Opt_wsize, "wsize=%d"},
147 {Opt_rsize, "rsize=%d"},
148 {Opt_rasize, "rasize=%d"},
149 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
150 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
151 {Opt_cap_release_safety, "cap_release_safety=%d"},
152 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
153 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
154 {Opt_congestion_kb, "write_congestion_kb=%d"},
156 {Opt_snapdirname, "snapdirname=%s"},
157 {Opt_mds_namespace, "mds_namespace=%s"},
158 /* string args above */
159 {Opt_dirstat, "dirstat"},
160 {Opt_nodirstat, "nodirstat"},
161 {Opt_rbytes, "rbytes"},
162 {Opt_norbytes, "norbytes"},
163 {Opt_asyncreaddir, "asyncreaddir"},
164 {Opt_noasyncreaddir, "noasyncreaddir"},
165 {Opt_dcache, "dcache"},
166 {Opt_nodcache, "nodcache"},
167 {Opt_ino32, "ino32"},
168 {Opt_noino32, "noino32"},
169 {Opt_fscache, "fsc"},
170 {Opt_nofscache, "nofsc"},
171 {Opt_poolperm, "poolperm"},
172 {Opt_nopoolperm, "nopoolperm"},
173 #ifdef CONFIG_CEPH_FS_POSIX_ACL
176 {Opt_noacl, "noacl"},
181 * Remove adjacent slashes and then the trailing slash, unless it is
182 * the only remaining character.
184 * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
186 static void canonicalize_path(char *path)
190 for (i = 0; path[i] != '\0'; i++) {
191 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
195 if (j > 1 && path[j - 1] == '/')
200 static int parse_fsopt_token(char *c, void *private)
202 struct ceph_mount_options *fsopt = private;
203 substring_t argstr[MAX_OPT_ARGS];
204 int token, intval, ret;
206 token = match_token((char *)c, fsopt_tokens, argstr);
210 if (token < Opt_last_int) {
211 ret = match_int(&argstr[0], &intval);
213 pr_err("bad mount option arg (not int) "
217 dout("got int token %d val %d\n", token, intval);
218 } else if (token > Opt_last_int && token < Opt_last_string) {
219 dout("got string token %d val %s\n", token,
222 dout("got token %d\n", token);
226 case Opt_snapdirname:
227 kfree(fsopt->snapdir_name);
228 fsopt->snapdir_name = kstrndup(argstr[0].from,
229 argstr[0].to-argstr[0].from,
231 if (!fsopt->snapdir_name)
234 case Opt_mds_namespace:
235 fsopt->mds_namespace = kstrndup(argstr[0].from,
236 argstr[0].to-argstr[0].from,
238 if (!fsopt->mds_namespace)
243 fsopt->wsize = intval;
246 fsopt->rsize = intval;
249 fsopt->rasize = intval;
251 case Opt_caps_wanted_delay_min:
252 fsopt->caps_wanted_delay_min = intval;
254 case Opt_caps_wanted_delay_max:
255 fsopt->caps_wanted_delay_max = intval;
257 case Opt_readdir_max_entries:
258 fsopt->max_readdir = intval;
260 case Opt_readdir_max_bytes:
261 fsopt->max_readdir_bytes = intval;
263 case Opt_congestion_kb:
264 fsopt->congestion_kb = intval;
267 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
270 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
273 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
276 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
278 case Opt_asyncreaddir:
279 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
281 case Opt_noasyncreaddir:
282 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
285 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
288 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
291 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
294 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
297 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
300 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
303 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
304 printk ("pool perm");
307 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
309 #ifdef CONFIG_CEPH_FS_POSIX_ACL
311 fsopt->sb_flags |= MS_POSIXACL;
315 fsopt->sb_flags &= ~MS_POSIXACL;
323 static void destroy_mount_options(struct ceph_mount_options *args)
325 dout("destroy_mount_options %p\n", args);
326 kfree(args->snapdir_name);
327 kfree(args->mds_namespace);
328 kfree(args->server_path);
332 static int strcmp_null(const char *s1, const char *s2)
340 return strcmp(s1, s2);
343 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
344 struct ceph_options *new_opt,
345 struct ceph_fs_client *fsc)
347 struct ceph_mount_options *fsopt1 = new_fsopt;
348 struct ceph_mount_options *fsopt2 = fsc->mount_options;
349 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
352 ret = memcmp(fsopt1, fsopt2, ofs);
356 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
360 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
364 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
368 return ceph_compare_options(new_opt, fsc->client);
371 static int parse_mount_options(struct ceph_mount_options **pfsopt,
372 struct ceph_options **popt,
373 int flags, char *options,
374 const char *dev_name)
376 struct ceph_mount_options *fsopt;
377 const char *dev_name_end;
380 if (!dev_name || !*dev_name)
383 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
387 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
389 fsopt->sb_flags = flags;
390 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
392 fsopt->rsize = CEPH_RSIZE_DEFAULT;
393 fsopt->rasize = CEPH_RASIZE_DEFAULT;
394 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
395 if (!fsopt->snapdir_name) {
400 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
401 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
402 fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
403 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
404 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
405 fsopt->congestion_kb = default_congestion_kb();
408 * Distinguish the server list from the path in "dev_name".
409 * Internally we do not include the leading '/' in the path.
411 * "dev_name" will look like:
412 * <server_spec>[,<server_spec>...]:[<path>]
414 * <server_spec> is <ip>[:<port>]
415 * <path> is optional, but if present must begin with '/'
417 dev_name_end = strchr(dev_name, '/');
420 * The server_path will include the whole chars from userland
421 * including the leading '/'.
423 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
424 if (!fsopt->server_path) {
429 canonicalize_path(fsopt->server_path);
431 dev_name_end = dev_name + strlen(dev_name);
434 dev_name_end--; /* back up to ':' separator */
435 if (dev_name_end < dev_name || *dev_name_end != ':') {
436 pr_err("device name is missing path (no : separator in %s)\n",
440 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
441 if (fsopt->server_path)
442 dout("server path '%s'\n", fsopt->server_path);
444 *popt = ceph_parse_options(options, dev_name, dev_name_end,
445 parse_fsopt_token, (void *)fsopt);
447 err = PTR_ERR(*popt);
456 destroy_mount_options(fsopt);
461 * ceph_show_options - Show mount options in /proc/mounts
462 * @m: seq_file to write to
463 * @root: root of that (sub)tree
465 static int ceph_show_options(struct seq_file *m, struct dentry *root)
467 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
468 struct ceph_mount_options *fsopt = fsc->mount_options;
472 /* a comma between MNT/MS and client options */
476 ret = ceph_print_client_options(m, fsc->client);
480 /* retract our comma if no client options */
484 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
485 seq_puts(m, ",dirstat");
486 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
487 seq_puts(m, ",rbytes");
488 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
489 seq_puts(m, ",noasyncreaddir");
490 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
491 seq_puts(m, ",nodcache");
492 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE)
494 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
495 seq_puts(m, ",nopoolperm");
497 #ifdef CONFIG_CEPH_FS_POSIX_ACL
498 if (fsopt->sb_flags & MS_POSIXACL)
501 seq_puts(m, ",noacl");
504 if (fsopt->mds_namespace)
505 seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
507 seq_printf(m, ",wsize=%d", fsopt->wsize);
508 if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
509 seq_printf(m, ",rsize=%d", fsopt->rsize);
510 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
511 seq_printf(m, ",rasize=%d", fsopt->rasize);
512 if (fsopt->congestion_kb != default_congestion_kb())
513 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
514 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
515 seq_printf(m, ",caps_wanted_delay_min=%d",
516 fsopt->caps_wanted_delay_min);
517 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
518 seq_printf(m, ",caps_wanted_delay_max=%d",
519 fsopt->caps_wanted_delay_max);
520 if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
521 seq_printf(m, ",cap_release_safety=%d",
522 fsopt->cap_release_safety);
523 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
524 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
525 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
526 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
527 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
528 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
534 * handle any mon messages the standard library doesn't understand.
535 * return error if we don't either.
537 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
539 struct ceph_fs_client *fsc = client->private;
540 int type = le16_to_cpu(msg->hdr.type);
543 case CEPH_MSG_MDS_MAP:
544 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
546 case CEPH_MSG_FS_MAP_USER:
547 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
555 * create a new fs client
557 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
558 struct ceph_options *opt)
560 struct ceph_fs_client *fsc;
561 const u64 supported_features =
562 CEPH_FEATURE_FLOCK | CEPH_FEATURE_DIRLAYOUTHASH |
563 CEPH_FEATURE_MDSENC | CEPH_FEATURE_MDS_INLINE_DATA;
564 const u64 required_features = 0;
569 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
571 return ERR_PTR(-ENOMEM);
573 fsc->client = ceph_create_client(opt, fsc, supported_features,
575 if (IS_ERR(fsc->client)) {
576 err = PTR_ERR(fsc->client);
579 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
581 if (fsopt->mds_namespace == NULL) {
582 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
585 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
589 fsc->mount_options = fsopt;
592 fsc->mount_state = CEPH_MOUNT_MOUNTING;
594 atomic_long_set(&fsc->writeback_count, 0);
596 err = bdi_init(&fsc->backing_dev_info);
602 * The number of concurrent works can be high but they don't need
603 * to be processed in parallel, limit concurrency.
605 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
606 if (fsc->wb_wq == NULL)
608 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
609 if (fsc->pg_inv_wq == NULL)
611 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
612 if (fsc->trunc_wq == NULL)
615 /* set up mempools */
617 page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
618 size = sizeof (struct page *) * (page_count ? page_count : 1);
619 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
620 if (!fsc->wb_pagevec_pool)
624 if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) &&
625 (ceph_fscache_register_fs(fsc) != 0))
629 fsc->min_caps = fsopt->max_readdir;
634 ceph_fscache_unregister_fs(fsc);
636 destroy_workqueue(fsc->trunc_wq);
638 destroy_workqueue(fsc->pg_inv_wq);
640 destroy_workqueue(fsc->wb_wq);
642 bdi_destroy(&fsc->backing_dev_info);
644 ceph_destroy_client(fsc->client);
650 static void destroy_fs_client(struct ceph_fs_client *fsc)
652 dout("destroy_fs_client %p\n", fsc);
654 ceph_fscache_unregister_fs(fsc);
656 destroy_workqueue(fsc->wb_wq);
657 destroy_workqueue(fsc->pg_inv_wq);
658 destroy_workqueue(fsc->trunc_wq);
660 bdi_destroy(&fsc->backing_dev_info);
662 mempool_destroy(fsc->wb_pagevec_pool);
664 destroy_mount_options(fsc->mount_options);
666 ceph_fs_debugfs_cleanup(fsc);
668 ceph_destroy_client(fsc->client);
671 dout("destroy_fs_client %p done\n", fsc);
677 struct kmem_cache *ceph_inode_cachep;
678 struct kmem_cache *ceph_cap_cachep;
679 struct kmem_cache *ceph_cap_flush_cachep;
680 struct kmem_cache *ceph_dentry_cachep;
681 struct kmem_cache *ceph_file_cachep;
683 static void ceph_inode_init_once(void *foo)
685 struct ceph_inode_info *ci = foo;
686 inode_init_once(&ci->vfs_inode);
689 static int __init init_caches(void)
693 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
694 sizeof(struct ceph_inode_info),
695 __alignof__(struct ceph_inode_info),
696 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
697 SLAB_ACCOUNT, ceph_inode_init_once);
698 if (ceph_inode_cachep == NULL)
701 ceph_cap_cachep = KMEM_CACHE(ceph_cap,
702 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
703 if (ceph_cap_cachep == NULL)
705 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
706 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
707 if (ceph_cap_flush_cachep == NULL)
710 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
711 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
712 if (ceph_dentry_cachep == NULL)
715 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
717 if (ceph_file_cachep == NULL)
720 if ((error = ceph_fscache_register()))
725 kmem_cache_destroy(ceph_dentry_cachep);
727 kmem_cache_destroy(ceph_cap_flush_cachep);
729 kmem_cache_destroy(ceph_cap_cachep);
731 kmem_cache_destroy(ceph_inode_cachep);
735 static void destroy_caches(void)
738 * Make sure all delayed rcu free inodes are flushed before we
743 kmem_cache_destroy(ceph_inode_cachep);
744 kmem_cache_destroy(ceph_cap_cachep);
745 kmem_cache_destroy(ceph_cap_flush_cachep);
746 kmem_cache_destroy(ceph_dentry_cachep);
747 kmem_cache_destroy(ceph_file_cachep);
749 ceph_fscache_unregister();
753 * ceph_umount_begin - initiate forced umount. Tear down down the
754 * mount, skipping steps that may hang while waiting for server(s).
756 static void ceph_umount_begin(struct super_block *sb)
758 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
760 dout("ceph_umount_begin - starting forced umount\n");
763 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
764 ceph_mdsc_force_umount(fsc->mdsc);
768 static int ceph_remount(struct super_block *sb, int *flags, char *data)
774 static const struct super_operations ceph_super_ops = {
775 .alloc_inode = ceph_alloc_inode,
776 .destroy_inode = ceph_destroy_inode,
777 .write_inode = ceph_write_inode,
778 .drop_inode = ceph_drop_inode,
779 .evict_inode = ceph_evict_inode,
780 .sync_fs = ceph_sync_fs,
781 .put_super = ceph_put_super,
782 .remount_fs = ceph_remount,
783 .show_options = ceph_show_options,
784 .statfs = ceph_statfs,
785 .umount_begin = ceph_umount_begin,
789 * Bootstrap mount by opening the root directory. Note the mount
790 * @started time from caller, and time out if this takes too long.
792 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
794 unsigned long started)
796 struct ceph_mds_client *mdsc = fsc->mdsc;
797 struct ceph_mds_request *req = NULL;
802 dout("open_root_inode opening '%s'\n", path);
803 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
805 return ERR_CAST(req);
806 req->r_path1 = kstrdup(path, GFP_NOFS);
808 root = ERR_PTR(-ENOMEM);
812 req->r_ino1.ino = CEPH_INO_ROOT;
813 req->r_ino1.snap = CEPH_NOSNAP;
814 req->r_started = started;
815 req->r_timeout = fsc->client->options->mount_timeout;
816 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
818 err = ceph_mdsc_do_request(mdsc, NULL, req);
820 struct inode *inode = req->r_target_inode;
821 req->r_target_inode = NULL;
822 dout("open_root_inode success\n");
823 root = d_make_root(inode);
825 root = ERR_PTR(-ENOMEM);
828 ceph_init_dentry(root);
829 dout("open_root_inode success, root dentry is %p\n", root);
834 ceph_mdsc_put_request(req);
839 * mount: join the ceph cluster, and open root directory.
841 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
844 unsigned long started = jiffies; /* note the start time */
847 dout("mount start %p\n", fsc);
848 mutex_lock(&fsc->client->mount_mutex);
850 if (!fsc->sb->s_root) {
851 const char *path = fsc->mount_options->server_path ?
852 fsc->mount_options->server_path + 1 : "";
854 err = __ceph_open_session(fsc->client, started);
858 dout("mount opening path '%s'\n", path);
860 err = ceph_fs_debugfs_init(fsc);
864 root = open_root_dentry(fsc, path, started);
869 fsc->sb->s_root = dget(root);
871 root = dget(fsc->sb->s_root);
874 fsc->mount_state = CEPH_MOUNT_MOUNTED;
875 dout("mount success\n");
876 mutex_unlock(&fsc->client->mount_mutex);
880 mutex_unlock(&fsc->client->mount_mutex);
884 static int ceph_set_super(struct super_block *s, void *data)
886 struct ceph_fs_client *fsc = data;
889 dout("set_super %p data %p\n", s, data);
891 s->s_flags = fsc->mount_options->sb_flags;
892 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
894 s->s_xattr = ceph_xattr_handlers;
898 s->s_op = &ceph_super_ops;
899 s->s_export_op = &ceph_export_ops;
901 s->s_time_gran = 1000; /* 1000 ns == 1 us */
903 ret = set_anon_super(s, NULL); /* what is that second arg for? */
916 * share superblock if same fs AND options
918 static int ceph_compare_super(struct super_block *sb, void *data)
920 struct ceph_fs_client *new = data;
921 struct ceph_mount_options *fsopt = new->mount_options;
922 struct ceph_options *opt = new->client->options;
923 struct ceph_fs_client *other = ceph_sb_to_client(sb);
925 dout("ceph_compare_super %p\n", sb);
927 if (compare_mount_options(fsopt, opt, other)) {
928 dout("monitor(s)/mount options don't match\n");
931 if ((opt->flags & CEPH_OPT_FSID) &&
932 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
933 dout("fsid doesn't match\n");
936 if (fsopt->sb_flags != other->mount_options->sb_flags) {
937 dout("flags differ\n");
944 * construct our own bdi so we can control readahead, etc.
946 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
948 static int ceph_register_bdi(struct super_block *sb,
949 struct ceph_fs_client *fsc)
953 /* set ra_pages based on rasize mount option? */
954 if (fsc->mount_options->rasize >= PAGE_SIZE)
955 fsc->backing_dev_info.ra_pages =
956 (fsc->mount_options->rasize + PAGE_SIZE - 1)
959 fsc->backing_dev_info.ra_pages =
960 VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
962 err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld",
963 atomic_long_inc_return(&bdi_seq));
965 sb->s_bdi = &fsc->backing_dev_info;
969 static struct dentry *ceph_mount(struct file_system_type *fs_type,
970 int flags, const char *dev_name, void *data)
972 struct super_block *sb;
973 struct ceph_fs_client *fsc;
976 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
977 struct ceph_mount_options *fsopt = NULL;
978 struct ceph_options *opt = NULL;
980 dout("ceph_mount\n");
982 #ifdef CONFIG_CEPH_FS_POSIX_ACL
983 flags |= MS_POSIXACL;
985 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
991 /* create client (which we may/may not use) */
992 fsc = create_fs_client(fsopt, opt);
995 destroy_mount_options(fsopt);
996 ceph_destroy_options(opt);
1000 err = ceph_mdsc_init(fsc);
1006 if (ceph_test_opt(fsc->client, NOSHARE))
1007 compare_super = NULL;
1008 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1014 if (ceph_sb_to_client(sb) != fsc) {
1015 ceph_mdsc_destroy(fsc);
1016 destroy_fs_client(fsc);
1017 fsc = ceph_sb_to_client(sb);
1018 dout("get_sb got existing client %p\n", fsc);
1020 dout("get_sb using new client %p\n", fsc);
1021 err = ceph_register_bdi(sb, fsc);
1028 res = ceph_real_mount(fsc);
1031 dout("root %p inode %p ino %llx.%llx\n", res,
1032 d_inode(res), ceph_vinop(d_inode(res)));
1036 ceph_mdsc_close_sessions(fsc->mdsc);
1037 deactivate_locked_super(sb);
1041 ceph_mdsc_destroy(fsc);
1042 destroy_fs_client(fsc);
1044 dout("ceph_mount fail %ld\n", PTR_ERR(res));
1048 static void ceph_kill_sb(struct super_block *s)
1050 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1051 dev_t dev = s->s_dev;
1053 dout("kill_sb %p\n", s);
1055 ceph_mdsc_pre_umount(fsc->mdsc);
1056 generic_shutdown_super(s);
1057 ceph_mdsc_destroy(fsc);
1059 destroy_fs_client(fsc);
1060 free_anon_bdev(dev);
1063 static struct file_system_type ceph_fs_type = {
1064 .owner = THIS_MODULE,
1066 .mount = ceph_mount,
1067 .kill_sb = ceph_kill_sb,
1068 .fs_flags = FS_RENAME_DOES_D_MOVE,
1070 MODULE_ALIAS_FS("ceph");
1072 static int __init init_ceph(void)
1074 int ret = init_caches();
1080 ret = register_filesystem(&ceph_fs_type);
1084 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1095 static void __exit exit_ceph(void)
1097 dout("exit_ceph\n");
1098 unregister_filesystem(&ceph_fs_type);
1103 module_init(init_ceph);
1104 module_exit(exit_ceph);
1106 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1107 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1108 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1109 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1110 MODULE_LICENSE("GPL");