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_mon_client *monc = &fsc->client->monc;
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(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 mutex_lock(&monc->mutex);
98 fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
99 le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
100 mutex_unlock(&monc->mutex);
102 buf->f_fsid.val[0] = fsid & 0xffffffff;
103 buf->f_fsid.val[1] = fsid >> 32;
109 static int ceph_sync_fs(struct super_block *sb, int wait)
111 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
114 dout("sync_fs (non-blocking)\n");
115 ceph_flush_dirty_caps(fsc->mdsc);
116 dout("sync_fs (non-blocking) done\n");
120 dout("sync_fs (blocking)\n");
121 ceph_osdc_sync(&fsc->client->osdc);
122 ceph_mdsc_sync(fsc->mdsc);
123 dout("sync_fs (blocking) done\n");
134 Opt_caps_wanted_delay_min,
135 Opt_caps_wanted_delay_max,
136 Opt_readdir_max_entries,
137 Opt_readdir_max_bytes,
145 /* string args above */
160 Opt_require_active_mds,
161 Opt_norequire_active_mds,
162 #ifdef CONFIG_CEPH_FS_POSIX_ACL
170 static match_table_t fsopt_tokens = {
171 {Opt_wsize, "wsize=%d"},
172 {Opt_rsize, "rsize=%d"},
173 {Opt_rasize, "rasize=%d"},
174 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
175 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
176 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
177 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
178 {Opt_congestion_kb, "write_congestion_kb=%d"},
180 {Opt_snapdirname, "snapdirname=%s"},
181 {Opt_mds_namespace, "mds_namespace=%s"},
182 {Opt_fscache_uniq, "fsc=%s"},
183 /* string args above */
184 {Opt_dirstat, "dirstat"},
185 {Opt_nodirstat, "nodirstat"},
186 {Opt_rbytes, "rbytes"},
187 {Opt_norbytes, "norbytes"},
188 {Opt_asyncreaddir, "asyncreaddir"},
189 {Opt_noasyncreaddir, "noasyncreaddir"},
190 {Opt_dcache, "dcache"},
191 {Opt_nodcache, "nodcache"},
192 {Opt_ino32, "ino32"},
193 {Opt_noino32, "noino32"},
194 {Opt_fscache, "fsc"},
195 {Opt_nofscache, "nofsc"},
196 {Opt_poolperm, "poolperm"},
197 {Opt_nopoolperm, "nopoolperm"},
198 {Opt_require_active_mds, "require_active_mds"},
199 {Opt_norequire_active_mds, "norequire_active_mds"},
200 #ifdef CONFIG_CEPH_FS_POSIX_ACL
203 {Opt_noacl, "noacl"},
204 {Opt_quotadf, "quotadf"},
205 {Opt_noquotadf, "noquotadf"},
209 static int parse_fsopt_token(char *c, void *private)
211 struct ceph_mount_options *fsopt = private;
212 substring_t argstr[MAX_OPT_ARGS];
213 int token, intval, ret;
215 token = match_token((char *)c, fsopt_tokens, argstr);
219 if (token < Opt_last_int) {
220 ret = match_int(&argstr[0], &intval);
222 pr_err("bad option arg (not int) at '%s'\n", c);
225 dout("got int token %d val %d\n", token, intval);
226 } else if (token > Opt_last_int && token < Opt_last_string) {
227 dout("got string token %d val %s\n", token,
230 dout("got token %d\n", token);
234 case Opt_snapdirname:
235 kfree(fsopt->snapdir_name);
236 fsopt->snapdir_name = kstrndup(argstr[0].from,
237 argstr[0].to-argstr[0].from,
239 if (!fsopt->snapdir_name)
242 case Opt_mds_namespace:
243 kfree(fsopt->mds_namespace);
244 fsopt->mds_namespace = kstrndup(argstr[0].from,
245 argstr[0].to-argstr[0].from,
247 if (!fsopt->mds_namespace)
250 case Opt_fscache_uniq:
251 kfree(fsopt->fscache_uniq);
252 fsopt->fscache_uniq = kstrndup(argstr[0].from,
253 argstr[0].to-argstr[0].from,
255 if (!fsopt->fscache_uniq)
257 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
261 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
263 fsopt->wsize = ALIGN(intval, PAGE_SIZE);
266 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
268 fsopt->rsize = ALIGN(intval, PAGE_SIZE);
273 fsopt->rasize = ALIGN(intval, PAGE_SIZE);
275 case Opt_caps_wanted_delay_min:
278 fsopt->caps_wanted_delay_min = intval;
280 case Opt_caps_wanted_delay_max:
283 fsopt->caps_wanted_delay_max = intval;
285 case Opt_readdir_max_entries:
288 fsopt->max_readdir = intval;
290 case Opt_readdir_max_bytes:
291 if (intval < (int)PAGE_SIZE && intval != 0)
293 fsopt->max_readdir_bytes = intval;
295 case Opt_congestion_kb:
296 if (intval < 1024) /* at least 1M */
298 fsopt->congestion_kb = intval;
301 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
304 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
307 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
310 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
312 case Opt_asyncreaddir:
313 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
315 case Opt_noasyncreaddir:
316 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
319 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
322 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
325 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
328 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
331 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
332 kfree(fsopt->fscache_uniq);
333 fsopt->fscache_uniq = NULL;
336 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
337 kfree(fsopt->fscache_uniq);
338 fsopt->fscache_uniq = NULL;
341 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
344 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
346 case Opt_require_active_mds:
347 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
349 case Opt_norequire_active_mds:
350 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
353 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
356 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
358 #ifdef CONFIG_CEPH_FS_POSIX_ACL
360 fsopt->sb_flags |= SB_POSIXACL;
364 fsopt->sb_flags &= ~SB_POSIXACL;
372 static void destroy_mount_options(struct ceph_mount_options *args)
374 dout("destroy_mount_options %p\n", args);
375 kfree(args->snapdir_name);
376 kfree(args->mds_namespace);
377 kfree(args->server_path);
378 kfree(args->fscache_uniq);
382 static int strcmp_null(const char *s1, const char *s2)
390 return strcmp(s1, s2);
393 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
394 struct ceph_options *new_opt,
395 struct ceph_fs_client *fsc)
397 struct ceph_mount_options *fsopt1 = new_fsopt;
398 struct ceph_mount_options *fsopt2 = fsc->mount_options;
399 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
402 ret = memcmp(fsopt1, fsopt2, ofs);
406 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
409 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
412 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
415 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
419 return ceph_compare_options(new_opt, fsc->client);
422 static int parse_mount_options(struct ceph_mount_options **pfsopt,
423 struct ceph_options **popt,
424 int flags, char *options,
425 const char *dev_name)
427 struct ceph_mount_options *fsopt;
428 const char *dev_name_end;
431 if (!dev_name || !*dev_name)
434 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
438 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
440 fsopt->sb_flags = flags;
441 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
443 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
444 fsopt->rsize = CEPH_MAX_READ_SIZE;
445 fsopt->rasize = CEPH_RASIZE_DEFAULT;
446 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
447 if (!fsopt->snapdir_name) {
452 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
453 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
454 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
455 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
456 fsopt->congestion_kb = default_congestion_kb();
459 * Distinguish the server list from the path in "dev_name".
460 * Internally we do not include the leading '/' in the path.
462 * "dev_name" will look like:
463 * <server_spec>[,<server_spec>...]:[<path>]
465 * <server_spec> is <ip>[:<port>]
466 * <path> is optional, but if present must begin with '/'
468 dev_name_end = strchr(dev_name, '/');
470 if (strlen(dev_name_end) > 1) {
471 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
472 if (!fsopt->server_path) {
478 dev_name_end = dev_name + strlen(dev_name);
481 dev_name_end--; /* back up to ':' separator */
482 if (dev_name_end < dev_name || *dev_name_end != ':') {
483 pr_err("device name is missing path (no : separator in %s)\n",
487 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
488 if (fsopt->server_path)
489 dout("server path '%s'\n", fsopt->server_path);
491 *popt = ceph_parse_options(options, dev_name, dev_name_end,
492 parse_fsopt_token, (void *)fsopt);
494 err = PTR_ERR(*popt);
503 destroy_mount_options(fsopt);
508 * ceph_show_options - Show mount options in /proc/mounts
509 * @m: seq_file to write to
510 * @root: root of that (sub)tree
512 static int ceph_show_options(struct seq_file *m, struct dentry *root)
514 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
515 struct ceph_mount_options *fsopt = fsc->mount_options;
519 /* a comma between MNT/MS and client options */
523 ret = ceph_print_client_options(m, fsc->client);
527 /* retract our comma if no client options */
531 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
532 seq_puts(m, ",dirstat");
533 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
534 seq_puts(m, ",rbytes");
535 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
536 seq_puts(m, ",noasyncreaddir");
537 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
538 seq_puts(m, ",nodcache");
539 if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
540 seq_puts(m, ",ino32");
541 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
542 seq_show_option(m, "fsc", fsopt->fscache_uniq);
544 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
545 seq_puts(m, ",nopoolperm");
546 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
547 seq_puts(m, ",noquotadf");
549 #ifdef CONFIG_CEPH_FS_POSIX_ACL
550 if (fsopt->sb_flags & SB_POSIXACL)
553 seq_puts(m, ",noacl");
556 if (fsopt->mds_namespace)
557 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
558 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
559 seq_printf(m, ",wsize=%d", fsopt->wsize);
560 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
561 seq_printf(m, ",rsize=%d", fsopt->rsize);
562 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
563 seq_printf(m, ",rasize=%d", fsopt->rasize);
564 if (fsopt->congestion_kb != default_congestion_kb())
565 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
566 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
567 seq_printf(m, ",caps_wanted_delay_min=%d",
568 fsopt->caps_wanted_delay_min);
569 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
570 seq_printf(m, ",caps_wanted_delay_max=%d",
571 fsopt->caps_wanted_delay_max);
572 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
573 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
574 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
575 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
576 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
577 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
583 * handle any mon messages the standard library doesn't understand.
584 * return error if we don't either.
586 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
588 struct ceph_fs_client *fsc = client->private;
589 int type = le16_to_cpu(msg->hdr.type);
592 case CEPH_MSG_MDS_MAP:
593 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
595 case CEPH_MSG_FS_MAP_USER:
596 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
604 * create a new fs client
606 * Success or not, this function consumes @fsopt and @opt.
608 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
609 struct ceph_options *opt)
611 struct ceph_fs_client *fsc;
616 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
622 fsc->client = ceph_create_client(opt, fsc);
623 if (IS_ERR(fsc->client)) {
624 err = PTR_ERR(fsc->client);
627 opt = NULL; /* fsc->client now owns this */
629 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
630 fsc->client->osdc.abort_on_full = true;
632 if (!fsopt->mds_namespace) {
633 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
636 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
640 fsc->mount_options = fsopt;
643 fsc->mount_state = CEPH_MOUNT_MOUNTING;
645 atomic_long_set(&fsc->writeback_count, 0);
649 * The number of concurrent works can be high but they don't need
650 * to be processed in parallel, limit concurrency.
652 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
655 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
658 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
662 /* set up mempools */
664 page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
665 size = sizeof (struct page *) * (page_count ? page_count : 1);
666 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
667 if (!fsc->wb_pagevec_pool)
671 fsc->min_caps = fsopt->max_readdir;
676 destroy_workqueue(fsc->trunc_wq);
678 destroy_workqueue(fsc->pg_inv_wq);
680 destroy_workqueue(fsc->wb_wq);
682 ceph_destroy_client(fsc->client);
686 ceph_destroy_options(opt);
687 destroy_mount_options(fsopt);
691 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
693 flush_workqueue(fsc->wb_wq);
694 flush_workqueue(fsc->pg_inv_wq);
695 flush_workqueue(fsc->trunc_wq);
698 static void destroy_fs_client(struct ceph_fs_client *fsc)
700 dout("destroy_fs_client %p\n", fsc);
702 destroy_workqueue(fsc->wb_wq);
703 destroy_workqueue(fsc->pg_inv_wq);
704 destroy_workqueue(fsc->trunc_wq);
706 mempool_destroy(fsc->wb_pagevec_pool);
708 destroy_mount_options(fsc->mount_options);
710 ceph_destroy_client(fsc->client);
713 dout("destroy_fs_client %p done\n", fsc);
719 struct kmem_cache *ceph_inode_cachep;
720 struct kmem_cache *ceph_cap_cachep;
721 struct kmem_cache *ceph_cap_flush_cachep;
722 struct kmem_cache *ceph_dentry_cachep;
723 struct kmem_cache *ceph_file_cachep;
724 struct kmem_cache *ceph_dir_file_cachep;
726 static void ceph_inode_init_once(void *foo)
728 struct ceph_inode_info *ci = foo;
729 inode_init_once(&ci->vfs_inode);
732 static int __init init_caches(void)
736 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
737 sizeof(struct ceph_inode_info),
738 __alignof__(struct ceph_inode_info),
739 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
740 SLAB_ACCOUNT, ceph_inode_init_once);
741 if (!ceph_inode_cachep)
744 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
745 if (!ceph_cap_cachep)
747 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
748 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
749 if (!ceph_cap_flush_cachep)
752 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
753 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
754 if (!ceph_dentry_cachep)
757 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
758 if (!ceph_file_cachep)
761 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
762 if (!ceph_dir_file_cachep)
765 error = ceph_fscache_register();
772 kmem_cache_destroy(ceph_dir_file_cachep);
774 kmem_cache_destroy(ceph_file_cachep);
776 kmem_cache_destroy(ceph_dentry_cachep);
778 kmem_cache_destroy(ceph_cap_flush_cachep);
780 kmem_cache_destroy(ceph_cap_cachep);
782 kmem_cache_destroy(ceph_inode_cachep);
786 static void destroy_caches(void)
789 * Make sure all delayed rcu free inodes are flushed before we
794 kmem_cache_destroy(ceph_inode_cachep);
795 kmem_cache_destroy(ceph_cap_cachep);
796 kmem_cache_destroy(ceph_cap_flush_cachep);
797 kmem_cache_destroy(ceph_dentry_cachep);
798 kmem_cache_destroy(ceph_file_cachep);
799 kmem_cache_destroy(ceph_dir_file_cachep);
801 ceph_fscache_unregister();
806 * ceph_umount_begin - initiate forced umount. Tear down down the
807 * mount, skipping steps that may hang while waiting for server(s).
809 static void ceph_umount_begin(struct super_block *sb)
811 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
813 dout("ceph_umount_begin - starting forced umount\n");
816 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
817 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
818 ceph_mdsc_force_umount(fsc->mdsc);
822 static const struct super_operations ceph_super_ops = {
823 .alloc_inode = ceph_alloc_inode,
824 .destroy_inode = ceph_destroy_inode,
825 .write_inode = ceph_write_inode,
826 .drop_inode = ceph_drop_inode,
827 .sync_fs = ceph_sync_fs,
828 .put_super = ceph_put_super,
829 .show_options = ceph_show_options,
830 .statfs = ceph_statfs,
831 .umount_begin = ceph_umount_begin,
835 * Bootstrap mount by opening the root directory. Note the mount
836 * @started time from caller, and time out if this takes too long.
838 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
840 unsigned long started)
842 struct ceph_mds_client *mdsc = fsc->mdsc;
843 struct ceph_mds_request *req = NULL;
848 dout("open_root_inode opening '%s'\n", path);
849 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
851 return ERR_CAST(req);
852 req->r_path1 = kstrdup(path, GFP_NOFS);
854 root = ERR_PTR(-ENOMEM);
858 req->r_ino1.ino = CEPH_INO_ROOT;
859 req->r_ino1.snap = CEPH_NOSNAP;
860 req->r_started = started;
861 req->r_timeout = fsc->client->options->mount_timeout;
862 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
864 err = ceph_mdsc_do_request(mdsc, NULL, req);
866 struct inode *inode = req->r_target_inode;
867 req->r_target_inode = NULL;
868 dout("open_root_inode success\n");
869 root = d_make_root(inode);
871 root = ERR_PTR(-ENOMEM);
874 dout("open_root_inode success, root dentry is %p\n", root);
879 ceph_mdsc_put_request(req);
887 * mount: join the ceph cluster, and open root directory.
889 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
892 unsigned long started = jiffies; /* note the start time */
895 dout("mount start %p\n", fsc);
896 mutex_lock(&fsc->client->mount_mutex);
898 if (!fsc->sb->s_root) {
900 err = __ceph_open_session(fsc->client, started);
905 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
906 err = ceph_fscache_register_fs(fsc);
911 if (!fsc->mount_options->server_path) {
913 dout("mount opening path \\t\n");
915 path = fsc->mount_options->server_path + 1;
916 dout("mount opening path %s\n", path);
919 err = ceph_fs_debugfs_init(fsc);
923 root = open_root_dentry(fsc, path, started);
928 fsc->sb->s_root = dget(root);
930 root = dget(fsc->sb->s_root);
933 fsc->mount_state = CEPH_MOUNT_MOUNTED;
934 dout("mount success\n");
935 mutex_unlock(&fsc->client->mount_mutex);
939 mutex_unlock(&fsc->client->mount_mutex);
943 static int ceph_set_super(struct super_block *s, void *data)
945 struct ceph_fs_client *fsc = data;
948 dout("set_super %p data %p\n", s, data);
950 s->s_flags = fsc->mount_options->sb_flags;
951 s->s_maxbytes = MAX_LFS_FILESIZE;
953 s->s_xattr = ceph_xattr_handlers;
956 fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
958 s->s_op = &ceph_super_ops;
959 s->s_d_op = &ceph_dentry_ops;
960 s->s_export_op = &ceph_export_ops;
962 s->s_time_gran = 1000; /* 1000 ns == 1 us */
964 ret = set_anon_super(s, NULL); /* what is that second arg for? */
977 * share superblock if same fs AND options
979 static int ceph_compare_super(struct super_block *sb, void *data)
981 struct ceph_fs_client *new = data;
982 struct ceph_mount_options *fsopt = new->mount_options;
983 struct ceph_options *opt = new->client->options;
984 struct ceph_fs_client *other = ceph_sb_to_client(sb);
986 dout("ceph_compare_super %p\n", sb);
988 if (compare_mount_options(fsopt, opt, other)) {
989 dout("monitor(s)/mount options don't match\n");
992 if ((opt->flags & CEPH_OPT_FSID) &&
993 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
994 dout("fsid doesn't match\n");
997 if (fsopt->sb_flags != other->mount_options->sb_flags) {
998 dout("flags differ\n");
1005 * construct our own bdi so we can control readahead, etc.
1007 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1009 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1013 err = super_setup_bdi_name(sb, "ceph-%ld",
1014 atomic_long_inc_return(&bdi_seq));
1018 /* set ra_pages based on rasize mount option? */
1019 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1021 /* set io_pages based on max osd read size */
1022 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1027 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1028 int flags, const char *dev_name, void *data)
1030 struct super_block *sb;
1031 struct ceph_fs_client *fsc;
1034 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1035 struct ceph_mount_options *fsopt = NULL;
1036 struct ceph_options *opt = NULL;
1038 dout("ceph_mount\n");
1040 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1041 flags |= SB_POSIXACL;
1043 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1049 /* create client (which we may/may not use) */
1050 fsc = create_fs_client(fsopt, opt);
1052 res = ERR_CAST(fsc);
1056 err = ceph_mdsc_init(fsc);
1062 if (ceph_test_opt(fsc->client, NOSHARE))
1063 compare_super = NULL;
1064 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1070 if (ceph_sb_to_client(sb) != fsc) {
1071 ceph_mdsc_destroy(fsc);
1072 destroy_fs_client(fsc);
1073 fsc = ceph_sb_to_client(sb);
1074 dout("get_sb got existing client %p\n", fsc);
1076 dout("get_sb using new client %p\n", fsc);
1077 err = ceph_setup_bdi(sb, fsc);
1084 res = ceph_real_mount(fsc);
1087 dout("root %p inode %p ino %llx.%llx\n", res,
1088 d_inode(res), ceph_vinop(d_inode(res)));
1092 ceph_mdsc_close_sessions(fsc->mdsc);
1093 deactivate_locked_super(sb);
1097 ceph_mdsc_destroy(fsc);
1098 destroy_fs_client(fsc);
1100 dout("ceph_mount fail %ld\n", PTR_ERR(res));
1104 static void ceph_kill_sb(struct super_block *s)
1106 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1107 dev_t dev = s->s_dev;
1109 dout("kill_sb %p\n", s);
1111 ceph_mdsc_pre_umount(fsc->mdsc);
1112 flush_fs_workqueues(fsc);
1114 generic_shutdown_super(s);
1116 fsc->client->extra_mon_dispatch = NULL;
1117 ceph_fs_debugfs_cleanup(fsc);
1119 ceph_fscache_unregister_fs(fsc);
1121 ceph_mdsc_destroy(fsc);
1123 destroy_fs_client(fsc);
1124 free_anon_bdev(dev);
1127 static struct file_system_type ceph_fs_type = {
1128 .owner = THIS_MODULE,
1130 .mount = ceph_mount,
1131 .kill_sb = ceph_kill_sb,
1132 .fs_flags = FS_RENAME_DOES_D_MOVE,
1134 MODULE_ALIAS_FS("ceph");
1136 static int __init init_ceph(void)
1138 int ret = init_caches();
1144 ret = register_filesystem(&ceph_fs_type);
1148 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1159 static void __exit exit_ceph(void)
1161 dout("exit_ceph\n");
1162 unregister_filesystem(&ceph_fs_type);
1167 module_init(init_ceph);
1168 module_exit(exit_ceph);
1170 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1171 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1172 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1173 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1174 MODULE_LICENSE("GPL");