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;
79 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
80 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
81 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
83 buf->f_files = le64_to_cpu(st.num_objects);
85 buf->f_namelen = NAME_MAX;
87 /* leave fsid little-endian, regardless of host endianness */
88 fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
89 buf->f_fsid.val[0] = fsid & 0xffffffff;
90 buf->f_fsid.val[1] = fsid >> 32;
95 static int ceph_sync_fs(struct super_block *sb, int wait)
97 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
100 dout("sync_fs (non-blocking)\n");
101 ceph_flush_dirty_caps(fsc->mdsc);
102 dout("sync_fs (non-blocking) done\n");
106 dout("sync_fs (blocking)\n");
107 ceph_osdc_sync(&fsc->client->osdc);
108 ceph_mdsc_sync(fsc->mdsc);
109 dout("sync_fs (blocking) done\n");
120 Opt_caps_wanted_delay_min,
121 Opt_caps_wanted_delay_max,
122 Opt_readdir_max_entries,
123 Opt_readdir_max_bytes,
131 /* string args above */
146 Opt_require_active_mds,
147 Opt_norequire_active_mds,
148 #ifdef CONFIG_CEPH_FS_POSIX_ACL
154 static match_table_t fsopt_tokens = {
155 {Opt_wsize, "wsize=%d"},
156 {Opt_rsize, "rsize=%d"},
157 {Opt_rasize, "rasize=%d"},
158 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
159 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
160 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
161 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
162 {Opt_congestion_kb, "write_congestion_kb=%d"},
164 {Opt_snapdirname, "snapdirname=%s"},
165 {Opt_mds_namespace, "mds_namespace=%s"},
166 {Opt_fscache_uniq, "fsc=%s"},
167 /* string args above */
168 {Opt_dirstat, "dirstat"},
169 {Opt_nodirstat, "nodirstat"},
170 {Opt_rbytes, "rbytes"},
171 {Opt_norbytes, "norbytes"},
172 {Opt_asyncreaddir, "asyncreaddir"},
173 {Opt_noasyncreaddir, "noasyncreaddir"},
174 {Opt_dcache, "dcache"},
175 {Opt_nodcache, "nodcache"},
176 {Opt_ino32, "ino32"},
177 {Opt_noino32, "noino32"},
178 {Opt_fscache, "fsc"},
179 {Opt_nofscache, "nofsc"},
180 {Opt_poolperm, "poolperm"},
181 {Opt_nopoolperm, "nopoolperm"},
182 {Opt_require_active_mds, "require_active_mds"},
183 {Opt_norequire_active_mds, "norequire_active_mds"},
184 #ifdef CONFIG_CEPH_FS_POSIX_ACL
187 {Opt_noacl, "noacl"},
192 * Remove adjacent slashes and then the trailing slash, unless it is
193 * the only remaining character.
195 * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
197 static void canonicalize_path(char *path)
201 for (i = 0; path[i] != '\0'; i++) {
202 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
206 if (j > 1 && path[j - 1] == '/')
211 static int parse_fsopt_token(char *c, void *private)
213 struct ceph_mount_options *fsopt = private;
214 substring_t argstr[MAX_OPT_ARGS];
215 int token, intval, ret;
217 token = match_token((char *)c, fsopt_tokens, argstr);
221 if (token < Opt_last_int) {
222 ret = match_int(&argstr[0], &intval);
224 pr_err("bad mount option arg (not int) "
228 dout("got int token %d val %d\n", token, intval);
229 } else if (token > Opt_last_int && token < Opt_last_string) {
230 dout("got string token %d val %s\n", token,
233 dout("got token %d\n", token);
237 case Opt_snapdirname:
238 kfree(fsopt->snapdir_name);
239 fsopt->snapdir_name = kstrndup(argstr[0].from,
240 argstr[0].to-argstr[0].from,
242 if (!fsopt->snapdir_name)
245 case Opt_mds_namespace:
246 kfree(fsopt->mds_namespace);
247 fsopt->mds_namespace = kstrndup(argstr[0].from,
248 argstr[0].to-argstr[0].from,
250 if (!fsopt->mds_namespace)
253 case Opt_fscache_uniq:
254 #ifdef CONFIG_CEPH_FSCACHE
255 kfree(fsopt->fscache_uniq);
256 fsopt->fscache_uniq = kstrndup(argstr[0].from,
257 argstr[0].to-argstr[0].from,
259 if (!fsopt->fscache_uniq)
261 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
264 pr_err("fscache support is disabled\n");
268 if (intval < PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
270 fsopt->wsize = ALIGN(intval, PAGE_SIZE);
273 if (intval < PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
275 fsopt->rsize = ALIGN(intval, PAGE_SIZE);
280 fsopt->rasize = ALIGN(intval, PAGE_SIZE);
282 case Opt_caps_wanted_delay_min:
285 fsopt->caps_wanted_delay_min = intval;
287 case Opt_caps_wanted_delay_max:
290 fsopt->caps_wanted_delay_max = intval;
292 case Opt_readdir_max_entries:
295 fsopt->max_readdir = intval;
297 case Opt_readdir_max_bytes:
298 if (intval < PAGE_SIZE && intval != 0)
300 fsopt->max_readdir_bytes = intval;
302 case Opt_congestion_kb:
303 if (intval < 1024) /* at least 1M */
305 fsopt->congestion_kb = intval;
308 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
311 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
314 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
317 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
319 case Opt_asyncreaddir:
320 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
322 case Opt_noasyncreaddir:
323 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
326 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
329 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
332 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
335 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
338 #ifdef CONFIG_CEPH_FSCACHE
339 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
342 pr_err("fscache support is disabled\n");
346 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
349 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
350 printk ("pool perm");
353 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
355 case Opt_require_active_mds:
356 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
358 case Opt_norequire_active_mds:
359 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
361 #ifdef CONFIG_CEPH_FS_POSIX_ACL
363 fsopt->sb_flags |= MS_POSIXACL;
367 fsopt->sb_flags &= ~MS_POSIXACL;
375 static void destroy_mount_options(struct ceph_mount_options *args)
377 dout("destroy_mount_options %p\n", args);
378 kfree(args->snapdir_name);
379 kfree(args->mds_namespace);
380 kfree(args->server_path);
381 kfree(args->fscache_uniq);
385 static int strcmp_null(const char *s1, const char *s2)
393 return strcmp(s1, s2);
396 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
397 struct ceph_options *new_opt,
398 struct ceph_fs_client *fsc)
400 struct ceph_mount_options *fsopt1 = new_fsopt;
401 struct ceph_mount_options *fsopt2 = fsc->mount_options;
402 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
405 ret = memcmp(fsopt1, fsopt2, ofs);
409 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
413 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
417 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
421 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
425 return ceph_compare_options(new_opt, fsc->client);
428 static int parse_mount_options(struct ceph_mount_options **pfsopt,
429 struct ceph_options **popt,
430 int flags, char *options,
431 const char *dev_name)
433 struct ceph_mount_options *fsopt;
434 const char *dev_name_end;
437 if (!dev_name || !*dev_name)
440 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
444 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
446 fsopt->sb_flags = flags;
447 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
449 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
450 fsopt->rsize = CEPH_MAX_READ_SIZE;
451 fsopt->rasize = CEPH_RASIZE_DEFAULT;
452 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
453 if (!fsopt->snapdir_name) {
458 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
459 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
460 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
461 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
462 fsopt->congestion_kb = default_congestion_kb();
465 * Distinguish the server list from the path in "dev_name".
466 * Internally we do not include the leading '/' in the path.
468 * "dev_name" will look like:
469 * <server_spec>[,<server_spec>...]:[<path>]
471 * <server_spec> is <ip>[:<port>]
472 * <path> is optional, but if present must begin with '/'
474 dev_name_end = strchr(dev_name, '/');
477 * The server_path will include the whole chars from userland
478 * including the leading '/'.
480 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
481 if (!fsopt->server_path) {
486 canonicalize_path(fsopt->server_path);
488 dev_name_end = dev_name + strlen(dev_name);
491 dev_name_end--; /* back up to ':' separator */
492 if (dev_name_end < dev_name || *dev_name_end != ':') {
493 pr_err("device name is missing path (no : separator in %s)\n",
497 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
498 if (fsopt->server_path)
499 dout("server path '%s'\n", fsopt->server_path);
501 *popt = ceph_parse_options(options, dev_name, dev_name_end,
502 parse_fsopt_token, (void *)fsopt);
504 err = PTR_ERR(*popt);
513 destroy_mount_options(fsopt);
518 * ceph_show_options - Show mount options in /proc/mounts
519 * @m: seq_file to write to
520 * @root: root of that (sub)tree
522 static int ceph_show_options(struct seq_file *m, struct dentry *root)
524 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
525 struct ceph_mount_options *fsopt = fsc->mount_options;
529 /* a comma between MNT/MS and client options */
533 ret = ceph_print_client_options(m, fsc->client);
537 /* retract our comma if no client options */
541 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
542 seq_puts(m, ",dirstat");
543 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
544 seq_puts(m, ",rbytes");
545 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
546 seq_puts(m, ",noasyncreaddir");
547 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
548 seq_puts(m, ",nodcache");
549 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
550 if (fsopt->fscache_uniq)
551 seq_printf(m, ",fsc=%s", fsopt->fscache_uniq);
555 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
556 seq_puts(m, ",nopoolperm");
558 #ifdef CONFIG_CEPH_FS_POSIX_ACL
559 if (fsopt->sb_flags & MS_POSIXACL)
562 seq_puts(m, ",noacl");
565 if (fsopt->mds_namespace)
566 seq_printf(m, ",mds_namespace=%s", fsopt->mds_namespace);
568 seq_printf(m, ",wsize=%d", fsopt->wsize);
569 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
570 seq_printf(m, ",rsize=%d", fsopt->rsize);
571 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
572 seq_printf(m, ",rasize=%d", fsopt->rasize);
573 if (fsopt->congestion_kb != default_congestion_kb())
574 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
575 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
576 seq_printf(m, ",caps_wanted_delay_min=%d",
577 fsopt->caps_wanted_delay_min);
578 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
579 seq_printf(m, ",caps_wanted_delay_max=%d",
580 fsopt->caps_wanted_delay_max);
581 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
582 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
583 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
584 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
585 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
586 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
592 * handle any mon messages the standard library doesn't understand.
593 * return error if we don't either.
595 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
597 struct ceph_fs_client *fsc = client->private;
598 int type = le16_to_cpu(msg->hdr.type);
601 case CEPH_MSG_MDS_MAP:
602 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
604 case CEPH_MSG_FS_MAP_USER:
605 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
613 * create a new fs client
615 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
616 struct ceph_options *opt)
618 struct ceph_fs_client *fsc;
623 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
625 return ERR_PTR(-ENOMEM);
627 fsc->client = ceph_create_client(opt, fsc);
628 if (IS_ERR(fsc->client)) {
629 err = PTR_ERR(fsc->client);
632 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
634 if (!fsopt->mds_namespace) {
635 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
638 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
642 fsc->mount_options = fsopt;
645 fsc->mount_state = CEPH_MOUNT_MOUNTING;
647 atomic_long_set(&fsc->writeback_count, 0);
651 * The number of concurrent works can be high but they don't need
652 * to be processed in parallel, limit concurrency.
654 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
657 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
660 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
664 /* set up mempools */
666 page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
667 size = sizeof (struct page *) * (page_count ? page_count : 1);
668 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
669 if (!fsc->wb_pagevec_pool)
673 fsc->min_caps = fsopt->max_readdir;
678 destroy_workqueue(fsc->trunc_wq);
680 destroy_workqueue(fsc->pg_inv_wq);
682 destroy_workqueue(fsc->wb_wq);
684 ceph_destroy_client(fsc->client);
690 static void destroy_fs_client(struct ceph_fs_client *fsc)
692 dout("destroy_fs_client %p\n", fsc);
694 destroy_workqueue(fsc->wb_wq);
695 destroy_workqueue(fsc->pg_inv_wq);
696 destroy_workqueue(fsc->trunc_wq);
698 mempool_destroy(fsc->wb_pagevec_pool);
700 destroy_mount_options(fsc->mount_options);
702 ceph_destroy_client(fsc->client);
705 dout("destroy_fs_client %p done\n", fsc);
711 struct kmem_cache *ceph_inode_cachep;
712 struct kmem_cache *ceph_cap_cachep;
713 struct kmem_cache *ceph_cap_flush_cachep;
714 struct kmem_cache *ceph_dentry_cachep;
715 struct kmem_cache *ceph_file_cachep;
717 static void ceph_inode_init_once(void *foo)
719 struct ceph_inode_info *ci = foo;
720 inode_init_once(&ci->vfs_inode);
723 static int __init init_caches(void)
727 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
728 sizeof(struct ceph_inode_info),
729 __alignof__(struct ceph_inode_info),
730 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
731 SLAB_ACCOUNT, ceph_inode_init_once);
732 if (!ceph_inode_cachep)
735 ceph_cap_cachep = KMEM_CACHE(ceph_cap,
736 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
737 if (!ceph_cap_cachep)
739 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
740 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
741 if (!ceph_cap_flush_cachep)
744 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
745 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
746 if (!ceph_dentry_cachep)
749 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
750 if (!ceph_file_cachep)
753 error = ceph_fscache_register();
760 kmem_cache_destroy(ceph_file_cachep);
762 kmem_cache_destroy(ceph_dentry_cachep);
764 kmem_cache_destroy(ceph_cap_flush_cachep);
766 kmem_cache_destroy(ceph_cap_cachep);
768 kmem_cache_destroy(ceph_inode_cachep);
772 static void destroy_caches(void)
775 * Make sure all delayed rcu free inodes are flushed before we
780 kmem_cache_destroy(ceph_inode_cachep);
781 kmem_cache_destroy(ceph_cap_cachep);
782 kmem_cache_destroy(ceph_cap_flush_cachep);
783 kmem_cache_destroy(ceph_dentry_cachep);
784 kmem_cache_destroy(ceph_file_cachep);
786 ceph_fscache_unregister();
790 * ceph_umount_begin - initiate forced umount. Tear down down the
791 * mount, skipping steps that may hang while waiting for server(s).
793 static void ceph_umount_begin(struct super_block *sb)
795 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
797 dout("ceph_umount_begin - starting forced umount\n");
800 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
801 ceph_mdsc_force_umount(fsc->mdsc);
805 static int ceph_remount(struct super_block *sb, int *flags, char *data)
811 static const struct super_operations ceph_super_ops = {
812 .alloc_inode = ceph_alloc_inode,
813 .destroy_inode = ceph_destroy_inode,
814 .write_inode = ceph_write_inode,
815 .drop_inode = ceph_drop_inode,
816 .sync_fs = ceph_sync_fs,
817 .put_super = ceph_put_super,
818 .remount_fs = ceph_remount,
819 .show_options = ceph_show_options,
820 .statfs = ceph_statfs,
821 .umount_begin = ceph_umount_begin,
825 * Bootstrap mount by opening the root directory. Note the mount
826 * @started time from caller, and time out if this takes too long.
828 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
830 unsigned long started)
832 struct ceph_mds_client *mdsc = fsc->mdsc;
833 struct ceph_mds_request *req = NULL;
838 dout("open_root_inode opening '%s'\n", path);
839 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
841 return ERR_CAST(req);
842 req->r_path1 = kstrdup(path, GFP_NOFS);
844 root = ERR_PTR(-ENOMEM);
848 req->r_ino1.ino = CEPH_INO_ROOT;
849 req->r_ino1.snap = CEPH_NOSNAP;
850 req->r_started = started;
851 req->r_timeout = fsc->client->options->mount_timeout;
852 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
854 err = ceph_mdsc_do_request(mdsc, NULL, req);
856 struct inode *inode = req->r_target_inode;
857 req->r_target_inode = NULL;
858 dout("open_root_inode success\n");
859 root = d_make_root(inode);
861 root = ERR_PTR(-ENOMEM);
864 dout("open_root_inode success, root dentry is %p\n", root);
869 ceph_mdsc_put_request(req);
874 * mount: join the ceph cluster, and open root directory.
876 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
879 unsigned long started = jiffies; /* note the start time */
882 dout("mount start %p\n", fsc);
883 mutex_lock(&fsc->client->mount_mutex);
885 if (!fsc->sb->s_root) {
886 const char *path = fsc->mount_options->server_path ?
887 fsc->mount_options->server_path + 1 : "";
889 err = __ceph_open_session(fsc->client, started);
894 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
895 err = ceph_fscache_register_fs(fsc);
900 dout("mount opening path '%s'\n", path);
902 err = ceph_fs_debugfs_init(fsc);
906 root = open_root_dentry(fsc, path, started);
911 fsc->sb->s_root = dget(root);
913 root = dget(fsc->sb->s_root);
916 fsc->mount_state = CEPH_MOUNT_MOUNTED;
917 dout("mount success\n");
918 mutex_unlock(&fsc->client->mount_mutex);
922 mutex_unlock(&fsc->client->mount_mutex);
926 static int ceph_set_super(struct super_block *s, void *data)
928 struct ceph_fs_client *fsc = data;
931 dout("set_super %p data %p\n", s, data);
933 s->s_flags = fsc->mount_options->sb_flags;
934 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
936 s->s_xattr = ceph_xattr_handlers;
940 s->s_op = &ceph_super_ops;
941 s->s_d_op = &ceph_dentry_ops;
942 s->s_export_op = &ceph_export_ops;
944 s->s_time_gran = 1000; /* 1000 ns == 1 us */
946 ret = set_anon_super(s, NULL); /* what is that second arg for? */
959 * share superblock if same fs AND options
961 static int ceph_compare_super(struct super_block *sb, void *data)
963 struct ceph_fs_client *new = data;
964 struct ceph_mount_options *fsopt = new->mount_options;
965 struct ceph_options *opt = new->client->options;
966 struct ceph_fs_client *other = ceph_sb_to_client(sb);
968 dout("ceph_compare_super %p\n", sb);
970 if (compare_mount_options(fsopt, opt, other)) {
971 dout("monitor(s)/mount options don't match\n");
974 if ((opt->flags & CEPH_OPT_FSID) &&
975 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
976 dout("fsid doesn't match\n");
979 if (fsopt->sb_flags != other->mount_options->sb_flags) {
980 dout("flags differ\n");
987 * construct our own bdi so we can control readahead, etc.
989 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
991 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
995 err = super_setup_bdi_name(sb, "ceph-%ld",
996 atomic_long_inc_return(&bdi_seq));
1000 /* set ra_pages based on rasize mount option? */
1001 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1003 /* set io_pages based on max osd read size */
1004 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1009 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1010 int flags, const char *dev_name, void *data)
1012 struct super_block *sb;
1013 struct ceph_fs_client *fsc;
1016 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1017 struct ceph_mount_options *fsopt = NULL;
1018 struct ceph_options *opt = NULL;
1020 dout("ceph_mount\n");
1022 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1023 flags |= MS_POSIXACL;
1025 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1031 /* create client (which we may/may not use) */
1032 fsc = create_fs_client(fsopt, opt);
1034 res = ERR_CAST(fsc);
1035 destroy_mount_options(fsopt);
1036 ceph_destroy_options(opt);
1040 err = ceph_mdsc_init(fsc);
1046 if (ceph_test_opt(fsc->client, NOSHARE))
1047 compare_super = NULL;
1048 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1054 if (ceph_sb_to_client(sb) != fsc) {
1055 ceph_mdsc_destroy(fsc);
1056 destroy_fs_client(fsc);
1057 fsc = ceph_sb_to_client(sb);
1058 dout("get_sb got existing client %p\n", fsc);
1060 dout("get_sb using new client %p\n", fsc);
1061 err = ceph_setup_bdi(sb, fsc);
1068 res = ceph_real_mount(fsc);
1071 dout("root %p inode %p ino %llx.%llx\n", res,
1072 d_inode(res), ceph_vinop(d_inode(res)));
1076 if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1077 pr_info("No mds server is up or the cluster is laggy\n");
1078 err = -EHOSTUNREACH;
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 generic_shutdown_super(s);
1103 fsc->client->extra_mon_dispatch = NULL;
1104 ceph_fs_debugfs_cleanup(fsc);
1106 ceph_fscache_unregister_fs(fsc);
1108 ceph_mdsc_destroy(fsc);
1110 destroy_fs_client(fsc);
1111 free_anon_bdev(dev);
1114 static struct file_system_type ceph_fs_type = {
1115 .owner = THIS_MODULE,
1117 .mount = ceph_mount,
1118 .kill_sb = ceph_kill_sb,
1119 .fs_flags = FS_RENAME_DOES_D_MOVE,
1121 MODULE_ALIAS_FS("ceph");
1123 static int __init init_ceph(void)
1125 int ret = init_caches();
1131 ret = register_filesystem(&ceph_fs_type);
1135 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1146 static void __exit exit_ceph(void)
1148 dout("exit_ceph\n");
1149 unregister_filesystem(&ceph_fs_type);
1154 module_init(init_ceph);
1155 module_exit(exit_ceph);
1157 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1158 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1159 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1160 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1161 MODULE_LICENSE("GPL");