1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/ceph/ceph_debug.h>
5 #include <linux/backing-dev.h>
6 #include <linux/ctype.h>
8 #include <linux/inet.h>
10 #include <linux/module.h>
11 #include <linux/mount.h>
12 #include <linux/parser.h>
13 #include <linux/sched.h>
14 #include <linux/seq_file.h>
15 #include <linux/slab.h>
16 #include <linux/statfs.h>
17 #include <linux/string.h>
20 #include "mds_client.h"
23 #include <linux/ceph/ceph_features.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/mon_client.h>
26 #include <linux/ceph/auth.h>
27 #include <linux/ceph/debugfs.h>
30 * Ceph superblock operations
32 * Handle the basics of mounting, unmounting.
38 static void ceph_put_super(struct super_block *s)
40 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
43 ceph_mdsc_close_sessions(fsc->mdsc);
46 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
48 struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
49 struct ceph_mon_client *monc = &fsc->client->monc;
50 struct ceph_statfs st;
55 if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
56 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
58 data_pool = CEPH_NOPOOL;
62 err = ceph_monc_do_statfs(monc, data_pool, &st);
67 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
70 * express utilization in terms of large blocks to avoid
71 * overflow on 32-bit machines.
73 * NOTE: for the time being, we make bsize == frsize to humor
74 * not-yet-ancient versions of glibc that are broken.
75 * Someday, we will probably want to report a real block
76 * size... whatever that may mean for a network file system!
78 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
79 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
82 * By default use root quota for stats; fallback to overall filesystem
83 * usage if using 'noquotadf' mount option or if the root dir doesn't
84 * have max_bytes quota set.
86 if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
87 !ceph_quota_update_statfs(fsc, buf)) {
88 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
89 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
90 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
93 buf->f_files = le64_to_cpu(st.num_objects);
95 buf->f_namelen = NAME_MAX;
97 /* Must convert the fsid, for consistent values across arches */
98 mutex_lock(&monc->mutex);
99 fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
100 le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
101 mutex_unlock(&monc->mutex);
103 buf->f_fsid.val[0] = fsid & 0xffffffff;
104 buf->f_fsid.val[1] = fsid >> 32;
110 static int ceph_sync_fs(struct super_block *sb, int wait)
112 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
115 dout("sync_fs (non-blocking)\n");
116 ceph_flush_dirty_caps(fsc->mdsc);
117 dout("sync_fs (non-blocking) done\n");
121 dout("sync_fs (blocking)\n");
122 ceph_osdc_sync(&fsc->client->osdc);
123 ceph_mdsc_sync(fsc->mdsc);
124 dout("sync_fs (blocking) done\n");
135 Opt_caps_wanted_delay_min,
136 Opt_caps_wanted_delay_max,
138 Opt_readdir_max_entries,
139 Opt_readdir_max_bytes,
148 /* string args above */
163 Opt_require_active_mds,
164 Opt_norequire_active_mds,
165 #ifdef CONFIG_CEPH_FS_POSIX_ACL
175 static match_table_t fsopt_tokens = {
176 {Opt_wsize, "wsize=%d"},
177 {Opt_rsize, "rsize=%d"},
178 {Opt_rasize, "rasize=%d"},
179 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
180 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
181 {Opt_caps_max, "caps_max=%d"},
182 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
183 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
184 {Opt_congestion_kb, "write_congestion_kb=%d"},
186 {Opt_snapdirname, "snapdirname=%s"},
187 {Opt_mds_namespace, "mds_namespace=%s"},
188 {Opt_recover_session, "recover_session=%s"},
189 {Opt_fscache_uniq, "fsc=%s"},
190 /* string args above */
191 {Opt_dirstat, "dirstat"},
192 {Opt_nodirstat, "nodirstat"},
193 {Opt_rbytes, "rbytes"},
194 {Opt_norbytes, "norbytes"},
195 {Opt_asyncreaddir, "asyncreaddir"},
196 {Opt_noasyncreaddir, "noasyncreaddir"},
197 {Opt_dcache, "dcache"},
198 {Opt_nodcache, "nodcache"},
199 {Opt_ino32, "ino32"},
200 {Opt_noino32, "noino32"},
201 {Opt_fscache, "fsc"},
202 {Opt_nofscache, "nofsc"},
203 {Opt_poolperm, "poolperm"},
204 {Opt_nopoolperm, "nopoolperm"},
205 {Opt_require_active_mds, "require_active_mds"},
206 {Opt_norequire_active_mds, "norequire_active_mds"},
207 #ifdef CONFIG_CEPH_FS_POSIX_ACL
210 {Opt_noacl, "noacl"},
211 {Opt_quotadf, "quotadf"},
212 {Opt_noquotadf, "noquotadf"},
213 {Opt_copyfrom, "copyfrom"},
214 {Opt_nocopyfrom, "nocopyfrom"},
218 static int parse_fsopt_token(char *c, void *private)
220 struct ceph_mount_options *fsopt = private;
221 substring_t argstr[MAX_OPT_ARGS];
222 int token, intval, ret;
224 token = match_token((char *)c, fsopt_tokens, argstr);
228 if (token < Opt_last_int) {
229 ret = match_int(&argstr[0], &intval);
231 pr_err("bad option arg (not int) at '%s'\n", c);
234 dout("got int token %d val %d\n", token, intval);
235 } else if (token > Opt_last_int && token < Opt_last_string) {
236 dout("got string token %d val %s\n", token,
239 dout("got token %d\n", token);
243 case Opt_snapdirname:
244 kfree(fsopt->snapdir_name);
245 fsopt->snapdir_name = kstrndup(argstr[0].from,
246 argstr[0].to-argstr[0].from,
248 if (!fsopt->snapdir_name)
251 case Opt_mds_namespace:
252 kfree(fsopt->mds_namespace);
253 fsopt->mds_namespace = kstrndup(argstr[0].from,
254 argstr[0].to-argstr[0].from,
256 if (!fsopt->mds_namespace)
259 case Opt_recover_session:
260 if (!strncmp(argstr[0].from, "no",
261 argstr[0].to - argstr[0].from)) {
262 fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
263 } else if (!strncmp(argstr[0].from, "clean",
264 argstr[0].to - argstr[0].from)) {
265 fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
270 case Opt_fscache_uniq:
271 #ifdef CONFIG_CEPH_FSCACHE
272 kfree(fsopt->fscache_uniq);
273 fsopt->fscache_uniq = kstrndup(argstr[0].from,
274 argstr[0].to-argstr[0].from,
276 if (!fsopt->fscache_uniq)
278 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
281 pr_err("fscache support is disabled\n");
285 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
287 fsopt->wsize = ALIGN(intval, PAGE_SIZE);
290 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
292 fsopt->rsize = ALIGN(intval, PAGE_SIZE);
297 fsopt->rasize = ALIGN(intval, PAGE_SIZE);
299 case Opt_caps_wanted_delay_min:
302 fsopt->caps_wanted_delay_min = intval;
304 case Opt_caps_wanted_delay_max:
307 fsopt->caps_wanted_delay_max = intval;
312 fsopt->caps_max = intval;
314 case Opt_readdir_max_entries:
317 fsopt->max_readdir = intval;
319 case Opt_readdir_max_bytes:
320 if (intval < (int)PAGE_SIZE && intval != 0)
322 fsopt->max_readdir_bytes = intval;
324 case Opt_congestion_kb:
325 if (intval < 1024) /* at least 1M */
327 fsopt->congestion_kb = intval;
330 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
333 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
336 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
339 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
341 case Opt_asyncreaddir:
342 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
344 case Opt_noasyncreaddir:
345 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
348 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
351 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
354 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
357 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
360 #ifdef CONFIG_CEPH_FSCACHE
361 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
362 kfree(fsopt->fscache_uniq);
363 fsopt->fscache_uniq = NULL;
366 pr_err("fscache support is disabled\n");
370 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
371 kfree(fsopt->fscache_uniq);
372 fsopt->fscache_uniq = NULL;
375 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
378 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
380 case Opt_require_active_mds:
381 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
383 case Opt_norequire_active_mds:
384 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
387 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
390 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
393 fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
396 fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
398 #ifdef CONFIG_CEPH_FS_POSIX_ACL
400 fsopt->sb_flags |= SB_POSIXACL;
404 fsopt->sb_flags &= ~SB_POSIXACL;
412 static void destroy_mount_options(struct ceph_mount_options *args)
414 dout("destroy_mount_options %p\n", args);
415 kfree(args->snapdir_name);
416 kfree(args->mds_namespace);
417 kfree(args->server_path);
418 kfree(args->fscache_uniq);
422 static int strcmp_null(const char *s1, const char *s2)
430 return strcmp(s1, s2);
433 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
434 struct ceph_options *new_opt,
435 struct ceph_fs_client *fsc)
437 struct ceph_mount_options *fsopt1 = new_fsopt;
438 struct ceph_mount_options *fsopt2 = fsc->mount_options;
439 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
442 ret = memcmp(fsopt1, fsopt2, ofs);
446 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
449 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
452 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
455 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
459 return ceph_compare_options(new_opt, fsc->client);
462 static int parse_mount_options(struct ceph_mount_options **pfsopt,
463 struct ceph_options **popt,
464 int flags, char *options,
465 const char *dev_name)
467 struct ceph_mount_options *fsopt;
468 const char *dev_name_end;
471 if (!dev_name || !*dev_name)
474 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
478 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
480 fsopt->sb_flags = flags;
481 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
483 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
484 fsopt->rsize = CEPH_MAX_READ_SIZE;
485 fsopt->rasize = CEPH_RASIZE_DEFAULT;
486 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
487 if (!fsopt->snapdir_name) {
492 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
493 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
494 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
495 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
496 fsopt->congestion_kb = default_congestion_kb();
499 * Distinguish the server list from the path in "dev_name".
500 * Internally we do not include the leading '/' in the path.
502 * "dev_name" will look like:
503 * <server_spec>[,<server_spec>...]:[<path>]
505 * <server_spec> is <ip>[:<port>]
506 * <path> is optional, but if present must begin with '/'
508 dev_name_end = strchr(dev_name, '/');
510 if (strlen(dev_name_end) > 1) {
511 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
512 if (!fsopt->server_path) {
518 dev_name_end = dev_name + strlen(dev_name);
521 dev_name_end--; /* back up to ':' separator */
522 if (dev_name_end < dev_name || *dev_name_end != ':') {
523 pr_err("device name is missing path (no : separator in %s)\n",
527 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
528 if (fsopt->server_path)
529 dout("server path '%s'\n", fsopt->server_path);
531 *popt = ceph_parse_options(options, dev_name, dev_name_end,
532 parse_fsopt_token, (void *)fsopt);
534 err = PTR_ERR(*popt);
543 destroy_mount_options(fsopt);
548 * ceph_show_options - Show mount options in /proc/mounts
549 * @m: seq_file to write to
550 * @root: root of that (sub)tree
552 static int ceph_show_options(struct seq_file *m, struct dentry *root)
554 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
555 struct ceph_mount_options *fsopt = fsc->mount_options;
559 /* a comma between MNT/MS and client options */
563 ret = ceph_print_client_options(m, fsc->client, false);
567 /* retract our comma if no client options */
571 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
572 seq_puts(m, ",dirstat");
573 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
574 seq_puts(m, ",rbytes");
575 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
576 seq_puts(m, ",noasyncreaddir");
577 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
578 seq_puts(m, ",nodcache");
579 if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
580 seq_puts(m, ",ino32");
581 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
582 seq_show_option(m, "fsc", fsopt->fscache_uniq);
584 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
585 seq_puts(m, ",nopoolperm");
586 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
587 seq_puts(m, ",noquotadf");
589 #ifdef CONFIG_CEPH_FS_POSIX_ACL
590 if (fsopt->sb_flags & SB_POSIXACL)
593 seq_puts(m, ",noacl");
596 if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
597 seq_puts(m, ",copyfrom");
599 if (fsopt->mds_namespace)
600 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
602 if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
603 seq_show_option(m, "recover_session", "clean");
605 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
606 seq_printf(m, ",wsize=%d", fsopt->wsize);
607 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
608 seq_printf(m, ",rsize=%d", fsopt->rsize);
609 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
610 seq_printf(m, ",rasize=%d", fsopt->rasize);
611 if (fsopt->congestion_kb != default_congestion_kb())
612 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
614 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
615 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
616 seq_printf(m, ",caps_wanted_delay_min=%d",
617 fsopt->caps_wanted_delay_min);
618 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
619 seq_printf(m, ",caps_wanted_delay_max=%d",
620 fsopt->caps_wanted_delay_max);
621 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
622 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
623 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
624 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
625 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
626 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
632 * handle any mon messages the standard library doesn't understand.
633 * return error if we don't either.
635 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
637 struct ceph_fs_client *fsc = client->private;
638 int type = le16_to_cpu(msg->hdr.type);
641 case CEPH_MSG_MDS_MAP:
642 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
644 case CEPH_MSG_FS_MAP_USER:
645 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
653 * create a new fs client
655 * Success or not, this function consumes @fsopt and @opt.
657 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
658 struct ceph_options *opt)
660 struct ceph_fs_client *fsc;
665 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
671 fsc->client = ceph_create_client(opt, fsc);
672 if (IS_ERR(fsc->client)) {
673 err = PTR_ERR(fsc->client);
676 opt = NULL; /* fsc->client now owns this */
678 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
679 ceph_set_opt(fsc->client, ABORT_ON_FULL);
681 if (!fsopt->mds_namespace) {
682 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
685 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
689 fsc->mount_options = fsopt;
692 fsc->mount_state = CEPH_MOUNT_MOUNTING;
695 atomic_long_set(&fsc->writeback_count, 0);
699 * The number of concurrent works can be high but they don't need
700 * to be processed in parallel, limit concurrency.
702 fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
705 fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
709 /* set up mempools */
711 page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
712 size = sizeof (struct page *) * (page_count ? page_count : 1);
713 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
714 if (!fsc->wb_pagevec_pool)
720 destroy_workqueue(fsc->cap_wq);
722 destroy_workqueue(fsc->inode_wq);
724 ceph_destroy_client(fsc->client);
728 ceph_destroy_options(opt);
729 destroy_mount_options(fsopt);
733 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
735 flush_workqueue(fsc->inode_wq);
736 flush_workqueue(fsc->cap_wq);
739 static void destroy_fs_client(struct ceph_fs_client *fsc)
741 dout("destroy_fs_client %p\n", fsc);
743 ceph_mdsc_destroy(fsc);
744 destroy_workqueue(fsc->inode_wq);
745 destroy_workqueue(fsc->cap_wq);
747 mempool_destroy(fsc->wb_pagevec_pool);
749 destroy_mount_options(fsc->mount_options);
751 ceph_destroy_client(fsc->client);
754 dout("destroy_fs_client %p done\n", fsc);
760 struct kmem_cache *ceph_inode_cachep;
761 struct kmem_cache *ceph_cap_cachep;
762 struct kmem_cache *ceph_cap_flush_cachep;
763 struct kmem_cache *ceph_dentry_cachep;
764 struct kmem_cache *ceph_file_cachep;
765 struct kmem_cache *ceph_dir_file_cachep;
767 static void ceph_inode_init_once(void *foo)
769 struct ceph_inode_info *ci = foo;
770 inode_init_once(&ci->vfs_inode);
773 static int __init init_caches(void)
777 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
778 sizeof(struct ceph_inode_info),
779 __alignof__(struct ceph_inode_info),
780 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
781 SLAB_ACCOUNT, ceph_inode_init_once);
782 if (!ceph_inode_cachep)
785 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
786 if (!ceph_cap_cachep)
788 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
789 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
790 if (!ceph_cap_flush_cachep)
793 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
794 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
795 if (!ceph_dentry_cachep)
798 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
799 if (!ceph_file_cachep)
802 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
803 if (!ceph_dir_file_cachep)
806 error = ceph_fscache_register();
813 kmem_cache_destroy(ceph_dir_file_cachep);
815 kmem_cache_destroy(ceph_file_cachep);
817 kmem_cache_destroy(ceph_dentry_cachep);
819 kmem_cache_destroy(ceph_cap_flush_cachep);
821 kmem_cache_destroy(ceph_cap_cachep);
823 kmem_cache_destroy(ceph_inode_cachep);
827 static void destroy_caches(void)
830 * Make sure all delayed rcu free inodes are flushed before we
835 kmem_cache_destroy(ceph_inode_cachep);
836 kmem_cache_destroy(ceph_cap_cachep);
837 kmem_cache_destroy(ceph_cap_flush_cachep);
838 kmem_cache_destroy(ceph_dentry_cachep);
839 kmem_cache_destroy(ceph_file_cachep);
840 kmem_cache_destroy(ceph_dir_file_cachep);
842 ceph_fscache_unregister();
847 * ceph_umount_begin - initiate forced umount. Tear down down the
848 * mount, skipping steps that may hang while waiting for server(s).
850 static void ceph_umount_begin(struct super_block *sb)
852 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
854 dout("ceph_umount_begin - starting forced umount\n");
857 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
858 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
859 ceph_mdsc_force_umount(fsc->mdsc);
860 fsc->filp_gen++; // invalidate open files
863 static int ceph_remount(struct super_block *sb, int *flags, char *data)
869 static const struct super_operations ceph_super_ops = {
870 .alloc_inode = ceph_alloc_inode,
871 .free_inode = ceph_free_inode,
872 .write_inode = ceph_write_inode,
873 .drop_inode = generic_delete_inode,
874 .evict_inode = ceph_evict_inode,
875 .sync_fs = ceph_sync_fs,
876 .put_super = ceph_put_super,
877 .remount_fs = ceph_remount,
878 .show_options = ceph_show_options,
879 .statfs = ceph_statfs,
880 .umount_begin = ceph_umount_begin,
884 * Bootstrap mount by opening the root directory. Note the mount
885 * @started time from caller, and time out if this takes too long.
887 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
889 unsigned long started)
891 struct ceph_mds_client *mdsc = fsc->mdsc;
892 struct ceph_mds_request *req = NULL;
897 dout("open_root_inode opening '%s'\n", path);
898 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
900 return ERR_CAST(req);
901 req->r_path1 = kstrdup(path, GFP_NOFS);
903 root = ERR_PTR(-ENOMEM);
907 req->r_ino1.ino = CEPH_INO_ROOT;
908 req->r_ino1.snap = CEPH_NOSNAP;
909 req->r_started = started;
910 req->r_timeout = fsc->client->options->mount_timeout;
911 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
913 err = ceph_mdsc_do_request(mdsc, NULL, req);
915 struct inode *inode = req->r_target_inode;
916 req->r_target_inode = NULL;
917 dout("open_root_inode success\n");
918 root = d_make_root(inode);
920 root = ERR_PTR(-ENOMEM);
923 dout("open_root_inode success, root dentry is %p\n", root);
928 ceph_mdsc_put_request(req);
936 * mount: join the ceph cluster, and open root directory.
938 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
941 unsigned long started = jiffies; /* note the start time */
944 dout("mount start %p\n", fsc);
945 mutex_lock(&fsc->client->mount_mutex);
947 if (!fsc->sb->s_root) {
949 err = __ceph_open_session(fsc->client, started);
954 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
955 err = ceph_fscache_register_fs(fsc);
960 if (!fsc->mount_options->server_path) {
962 dout("mount opening path \\t\n");
964 path = fsc->mount_options->server_path + 1;
965 dout("mount opening path %s\n", path);
968 ceph_fs_debugfs_init(fsc);
970 root = open_root_dentry(fsc, path, started);
975 fsc->sb->s_root = dget(root);
977 root = dget(fsc->sb->s_root);
980 fsc->mount_state = CEPH_MOUNT_MOUNTED;
981 dout("mount success\n");
982 mutex_unlock(&fsc->client->mount_mutex);
986 mutex_unlock(&fsc->client->mount_mutex);
990 static int ceph_set_super(struct super_block *s, void *data)
992 struct ceph_fs_client *fsc = data;
995 dout("set_super %p data %p\n", s, data);
997 s->s_flags = fsc->mount_options->sb_flags;
998 s->s_maxbytes = MAX_LFS_FILESIZE;
1000 s->s_xattr = ceph_xattr_handlers;
1003 fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1005 s->s_op = &ceph_super_ops;
1006 s->s_d_op = &ceph_dentry_ops;
1007 s->s_export_op = &ceph_export_ops;
1011 s->s_time_max = U32_MAX;
1013 ret = set_anon_super(s, NULL); /* what is that second arg for? */
1020 s->s_fs_info = NULL;
1026 * share superblock if same fs AND options
1028 static int ceph_compare_super(struct super_block *sb, void *data)
1030 struct ceph_fs_client *new = data;
1031 struct ceph_mount_options *fsopt = new->mount_options;
1032 struct ceph_options *opt = new->client->options;
1033 struct ceph_fs_client *other = ceph_sb_to_client(sb);
1035 dout("ceph_compare_super %p\n", sb);
1037 if (compare_mount_options(fsopt, opt, other)) {
1038 dout("monitor(s)/mount options don't match\n");
1041 if ((opt->flags & CEPH_OPT_FSID) &&
1042 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
1043 dout("fsid doesn't match\n");
1046 if (fsopt->sb_flags != other->mount_options->sb_flags) {
1047 dout("flags differ\n");
1054 * construct our own bdi so we can control readahead, etc.
1056 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1058 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1062 err = super_setup_bdi_name(sb, "ceph-%ld",
1063 atomic_long_inc_return(&bdi_seq));
1067 /* set ra_pages based on rasize mount option? */
1068 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1070 /* set io_pages based on max osd read size */
1071 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1076 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1077 int flags, const char *dev_name, void *data)
1079 struct super_block *sb;
1080 struct ceph_fs_client *fsc;
1083 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1084 struct ceph_mount_options *fsopt = NULL;
1085 struct ceph_options *opt = NULL;
1087 dout("ceph_mount\n");
1089 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1090 flags |= SB_POSIXACL;
1092 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1098 /* create client (which we may/may not use) */
1099 fsc = create_fs_client(fsopt, opt);
1101 res = ERR_CAST(fsc);
1105 err = ceph_mdsc_init(fsc);
1111 if (ceph_test_opt(fsc->client, NOSHARE))
1112 compare_super = NULL;
1113 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1119 if (ceph_sb_to_client(sb) != fsc) {
1120 destroy_fs_client(fsc);
1121 fsc = ceph_sb_to_client(sb);
1122 dout("get_sb got existing client %p\n", fsc);
1124 dout("get_sb using new client %p\n", fsc);
1125 err = ceph_setup_bdi(sb, fsc);
1132 res = ceph_real_mount(fsc);
1135 dout("root %p inode %p ino %llx.%llx\n", res,
1136 d_inode(res), ceph_vinop(d_inode(res)));
1140 ceph_mdsc_close_sessions(fsc->mdsc);
1141 deactivate_locked_super(sb);
1145 destroy_fs_client(fsc);
1147 dout("ceph_mount fail %ld\n", PTR_ERR(res));
1151 static void ceph_kill_sb(struct super_block *s)
1153 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1154 dev_t dev = s->s_dev;
1156 dout("kill_sb %p\n", s);
1158 ceph_mdsc_pre_umount(fsc->mdsc);
1159 flush_fs_workqueues(fsc);
1161 generic_shutdown_super(s);
1163 fsc->client->extra_mon_dispatch = NULL;
1164 ceph_fs_debugfs_cleanup(fsc);
1166 ceph_fscache_unregister_fs(fsc);
1168 destroy_fs_client(fsc);
1169 free_anon_bdev(dev);
1172 static struct file_system_type ceph_fs_type = {
1173 .owner = THIS_MODULE,
1175 .mount = ceph_mount,
1176 .kill_sb = ceph_kill_sb,
1177 .fs_flags = FS_RENAME_DOES_D_MOVE,
1179 MODULE_ALIAS_FS("ceph");
1181 int ceph_force_reconnect(struct super_block *sb)
1183 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1186 ceph_umount_begin(sb);
1188 /* Make sure all page caches get invalidated.
1189 * see remove_session_caps_cb() */
1190 flush_workqueue(fsc->inode_wq);
1192 /* In case that we were blacklisted. This also reset
1193 * all mon/osd connections */
1194 ceph_reset_client_addr(fsc->client);
1196 ceph_osdc_clear_abort_err(&fsc->client->osdc);
1198 fsc->blacklisted = false;
1199 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1202 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1203 CEPH_STAT_CAP_INODE, true);
1208 static int __init init_ceph(void)
1210 int ret = init_caches();
1215 ret = register_filesystem(&ceph_fs_type);
1219 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1229 static void __exit exit_ceph(void)
1231 dout("exit_ceph\n");
1232 unregister_filesystem(&ceph_fs_type);
1236 module_init(init_ceph);
1237 module_exit(exit_ceph);
1239 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1240 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1241 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1242 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1243 MODULE_LICENSE("GPL");