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/fs_context.h>
13 #include <linux/fs_parser.h>
14 #include <linux/sched.h>
15 #include <linux/seq_file.h>
16 #include <linux/slab.h>
17 #include <linux/statfs.h>
18 #include <linux/string.h>
21 #include "mds_client.h"
24 #include <linux/ceph/ceph_features.h>
25 #include <linux/ceph/decode.h>
26 #include <linux/ceph/mon_client.h>
27 #include <linux/ceph/auth.h>
28 #include <linux/ceph/debugfs.h>
31 * Ceph superblock operations
33 * Handle the basics of mounting, unmounting.
39 static void ceph_put_super(struct super_block
*s
)
41 struct ceph_fs_client
*fsc
= ceph_sb_to_client(s
);
44 ceph_mdsc_close_sessions(fsc
->mdsc
);
47 static int ceph_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
49 struct ceph_fs_client
*fsc
= ceph_inode_to_client(d_inode(dentry
));
50 struct ceph_mon_client
*monc
= &fsc
->client
->monc
;
51 struct ceph_statfs st
;
56 if (fsc
->mdsc
->mdsmap
->m_num_data_pg_pools
== 1) {
57 data_pool
= fsc
->mdsc
->mdsmap
->m_data_pg_pools
[0];
59 data_pool
= CEPH_NOPOOL
;
63 err
= ceph_monc_do_statfs(monc
, data_pool
, &st
);
68 buf
->f_type
= CEPH_SUPER_MAGIC
; /* ?? */
71 * express utilization in terms of large blocks to avoid
72 * overflow on 32-bit machines.
74 * NOTE: for the time being, we make bsize == frsize to humor
75 * not-yet-ancient versions of glibc that are broken.
76 * Someday, we will probably want to report a real block
77 * size... whatever that may mean for a network file system!
79 buf
->f_bsize
= 1 << CEPH_BLOCK_SHIFT
;
80 buf
->f_frsize
= 1 << CEPH_BLOCK_SHIFT
;
83 * By default use root quota for stats; fallback to overall filesystem
84 * usage if using 'noquotadf' mount option or if the root dir doesn't
85 * have max_bytes quota set.
87 if (ceph_test_mount_opt(fsc
, NOQUOTADF
) ||
88 !ceph_quota_update_statfs(fsc
, buf
)) {
89 buf
->f_blocks
= le64_to_cpu(st
.kb
) >> (CEPH_BLOCK_SHIFT
-10);
90 buf
->f_bfree
= le64_to_cpu(st
.kb_avail
) >> (CEPH_BLOCK_SHIFT
-10);
91 buf
->f_bavail
= le64_to_cpu(st
.kb_avail
) >> (CEPH_BLOCK_SHIFT
-10);
94 buf
->f_files
= le64_to_cpu(st
.num_objects
);
96 buf
->f_namelen
= NAME_MAX
;
98 /* Must convert the fsid, for consistent values across arches */
99 mutex_lock(&monc
->mutex
);
100 fsid
= le64_to_cpu(*(__le64
*)(&monc
->monmap
->fsid
)) ^
101 le64_to_cpu(*((__le64
*)&monc
->monmap
->fsid
+ 1));
102 mutex_unlock(&monc
->mutex
);
104 buf
->f_fsid
.val
[0] = fsid
& 0xffffffff;
105 buf
->f_fsid
.val
[1] = fsid
>> 32;
111 static int ceph_sync_fs(struct super_block
*sb
, int wait
)
113 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
116 dout("sync_fs (non-blocking)\n");
117 ceph_flush_dirty_caps(fsc
->mdsc
);
118 dout("sync_fs (non-blocking) done\n");
122 dout("sync_fs (blocking)\n");
123 ceph_osdc_sync(&fsc
->client
->osdc
);
124 ceph_mdsc_sync(fsc
->mdsc
);
125 dout("sync_fs (blocking) done\n");
136 Opt_caps_wanted_delay_min
,
137 Opt_caps_wanted_delay_max
,
139 Opt_readdir_max_entries
,
140 Opt_readdir_max_bytes
,
147 /* string args above */
155 Opt_require_active_mds
,
161 enum ceph_recover_session_mode
{
162 ceph_recover_session_no
,
163 ceph_recover_session_clean
166 static const struct constant_table ceph_param_recover
[] = {
167 { "no", ceph_recover_session_no
},
168 { "clean", ceph_recover_session_clean
},
172 static const struct fs_parameter_spec ceph_mount_parameters
[] = {
173 fsparam_flag_no ("acl", Opt_acl
),
174 fsparam_flag_no ("asyncreaddir", Opt_asyncreaddir
),
175 fsparam_s32 ("caps_max", Opt_caps_max
),
176 fsparam_u32 ("caps_wanted_delay_max", Opt_caps_wanted_delay_max
),
177 fsparam_u32 ("caps_wanted_delay_min", Opt_caps_wanted_delay_min
),
178 fsparam_u32 ("write_congestion_kb", Opt_congestion_kb
),
179 fsparam_flag_no ("copyfrom", Opt_copyfrom
),
180 fsparam_flag_no ("dcache", Opt_dcache
),
181 fsparam_flag_no ("dirstat", Opt_dirstat
),
182 __fsparam (fs_param_is_string
, "fsc", Opt_fscache
,
183 fs_param_neg_with_no
| fs_param_v_optional
, NULL
),
184 fsparam_flag_no ("ino32", Opt_ino32
),
185 fsparam_string ("mds_namespace", Opt_mds_namespace
),
186 fsparam_flag_no ("poolperm", Opt_poolperm
),
187 fsparam_flag_no ("quotadf", Opt_quotadf
),
188 fsparam_u32 ("rasize", Opt_rasize
),
189 fsparam_flag_no ("rbytes", Opt_rbytes
),
190 fsparam_u32 ("readdir_max_bytes", Opt_readdir_max_bytes
),
191 fsparam_u32 ("readdir_max_entries", Opt_readdir_max_entries
),
192 fsparam_enum ("recover_session", Opt_recover_session
, ceph_param_recover
),
193 fsparam_flag_no ("require_active_mds", Opt_require_active_mds
),
194 fsparam_u32 ("rsize", Opt_rsize
),
195 fsparam_string ("snapdirname", Opt_snapdirname
),
196 fsparam_string ("source", Opt_source
),
197 fsparam_u32 ("wsize", Opt_wsize
),
201 struct ceph_parse_opts_ctx
{
202 struct ceph_options
*copts
;
203 struct ceph_mount_options
*opts
;
207 * Parse the source parameter. Distinguish the server list from the path.
208 * Internally we do not include the leading '/' in the path.
210 * The source will look like:
211 * <server_spec>[,<server_spec>...]:[<path>]
213 * <server_spec> is <ip>[:<port>]
214 * <path> is optional, but if present must begin with '/'
216 static int ceph_parse_source(struct fs_parameter
*param
, struct fs_context
*fc
)
218 struct ceph_parse_opts_ctx
*pctx
= fc
->fs_private
;
219 struct ceph_mount_options
*fsopt
= pctx
->opts
;
220 char *dev_name
= param
->string
, *dev_name_end
;
223 dout("%s '%s'\n", __func__
, dev_name
);
224 if (!dev_name
|| !*dev_name
)
225 return invalf(fc
, "ceph: Empty source");
227 dev_name_end
= strchr(dev_name
, '/');
229 if (strlen(dev_name_end
) > 1) {
230 kfree(fsopt
->server_path
);
231 fsopt
->server_path
= kstrdup(dev_name_end
, GFP_KERNEL
);
232 if (!fsopt
->server_path
)
236 dev_name_end
= dev_name
+ strlen(dev_name
);
239 dev_name_end
--; /* back up to ':' separator */
240 if (dev_name_end
< dev_name
|| *dev_name_end
!= ':')
241 return invalf(fc
, "ceph: No path or : separator in source");
243 dout("device name '%.*s'\n", (int)(dev_name_end
- dev_name
), dev_name
);
244 if (fsopt
->server_path
)
245 dout("server path '%s'\n", fsopt
->server_path
);
247 ret
= ceph_parse_mon_ips(param
->string
, dev_name_end
- dev_name
,
248 pctx
->copts
, fc
->log
.log
);
252 fc
->source
= param
->string
;
253 param
->string
= NULL
;
257 static int ceph_parse_mount_param(struct fs_context
*fc
,
258 struct fs_parameter
*param
)
260 struct ceph_parse_opts_ctx
*pctx
= fc
->fs_private
;
261 struct ceph_mount_options
*fsopt
= pctx
->opts
;
262 struct fs_parse_result result
;
266 ret
= ceph_parse_param(param
, pctx
->copts
, fc
->log
.log
);
267 if (ret
!= -ENOPARAM
)
270 token
= fs_parse(fc
, ceph_mount_parameters
, param
, &result
);
271 dout("%s fs_parse '%s' token %d\n", __func__
, param
->key
, token
);
276 case Opt_snapdirname
:
277 kfree(fsopt
->snapdir_name
);
278 fsopt
->snapdir_name
= param
->string
;
279 param
->string
= NULL
;
281 case Opt_mds_namespace
:
282 kfree(fsopt
->mds_namespace
);
283 fsopt
->mds_namespace
= param
->string
;
284 param
->string
= NULL
;
286 case Opt_recover_session
:
287 mode
= result
.uint_32
;
288 if (mode
== ceph_recover_session_no
)
289 fsopt
->flags
&= ~CEPH_MOUNT_OPT_CLEANRECOVER
;
290 else if (mode
== ceph_recover_session_clean
)
291 fsopt
->flags
|= CEPH_MOUNT_OPT_CLEANRECOVER
;
297 return invalf(fc
, "ceph: Multiple sources specified");
298 return ceph_parse_source(param
, fc
);
300 if (result
.uint_32
< PAGE_SIZE
||
301 result
.uint_32
> CEPH_MAX_WRITE_SIZE
)
303 fsopt
->wsize
= ALIGN(result
.uint_32
, PAGE_SIZE
);
306 if (result
.uint_32
< PAGE_SIZE
||
307 result
.uint_32
> CEPH_MAX_READ_SIZE
)
309 fsopt
->rsize
= ALIGN(result
.uint_32
, PAGE_SIZE
);
312 fsopt
->rasize
= ALIGN(result
.uint_32
, PAGE_SIZE
);
314 case Opt_caps_wanted_delay_min
:
315 if (result
.uint_32
< 1)
317 fsopt
->caps_wanted_delay_min
= result
.uint_32
;
319 case Opt_caps_wanted_delay_max
:
320 if (result
.uint_32
< 1)
322 fsopt
->caps_wanted_delay_max
= result
.uint_32
;
325 if (result
.int_32
< 0)
327 fsopt
->caps_max
= result
.int_32
;
329 case Opt_readdir_max_entries
:
330 if (result
.uint_32
< 1)
332 fsopt
->max_readdir
= result
.uint_32
;
334 case Opt_readdir_max_bytes
:
335 if (result
.uint_32
< PAGE_SIZE
&& result
.uint_32
!= 0)
337 fsopt
->max_readdir_bytes
= result
.uint_32
;
339 case Opt_congestion_kb
:
340 if (result
.uint_32
< 1024) /* at least 1M */
342 fsopt
->congestion_kb
= result
.uint_32
;
346 fsopt
->flags
|= CEPH_MOUNT_OPT_DIRSTAT
;
348 fsopt
->flags
&= ~CEPH_MOUNT_OPT_DIRSTAT
;
352 fsopt
->flags
|= CEPH_MOUNT_OPT_RBYTES
;
354 fsopt
->flags
&= ~CEPH_MOUNT_OPT_RBYTES
;
356 case Opt_asyncreaddir
:
358 fsopt
->flags
&= ~CEPH_MOUNT_OPT_NOASYNCREADDIR
;
360 fsopt
->flags
|= CEPH_MOUNT_OPT_NOASYNCREADDIR
;
364 fsopt
->flags
|= CEPH_MOUNT_OPT_DCACHE
;
366 fsopt
->flags
&= ~CEPH_MOUNT_OPT_DCACHE
;
370 fsopt
->flags
|= CEPH_MOUNT_OPT_INO32
;
372 fsopt
->flags
&= ~CEPH_MOUNT_OPT_INO32
;
376 #ifdef CONFIG_CEPH_FSCACHE
377 kfree(fsopt
->fscache_uniq
);
378 fsopt
->fscache_uniq
= NULL
;
379 if (result
.negated
) {
380 fsopt
->flags
&= ~CEPH_MOUNT_OPT_FSCACHE
;
382 fsopt
->flags
|= CEPH_MOUNT_OPT_FSCACHE
;
383 fsopt
->fscache_uniq
= param
->string
;
384 param
->string
= NULL
;
388 return invalf(fc
, "ceph: fscache support is disabled");
392 fsopt
->flags
&= ~CEPH_MOUNT_OPT_NOPOOLPERM
;
394 fsopt
->flags
|= CEPH_MOUNT_OPT_NOPOOLPERM
;
396 case Opt_require_active_mds
:
398 fsopt
->flags
&= ~CEPH_MOUNT_OPT_MOUNTWAIT
;
400 fsopt
->flags
|= CEPH_MOUNT_OPT_MOUNTWAIT
;
404 fsopt
->flags
&= ~CEPH_MOUNT_OPT_NOQUOTADF
;
406 fsopt
->flags
|= CEPH_MOUNT_OPT_NOQUOTADF
;
410 fsopt
->flags
&= ~CEPH_MOUNT_OPT_NOCOPYFROM
;
412 fsopt
->flags
|= CEPH_MOUNT_OPT_NOCOPYFROM
;
415 if (!result
.negated
) {
416 #ifdef CONFIG_CEPH_FS_POSIX_ACL
417 fc
->sb_flags
|= SB_POSIXACL
;
419 return invalf(fc
, "ceph: POSIX ACL support is disabled");
422 fc
->sb_flags
&= ~SB_POSIXACL
;
431 return invalf(fc
, "ceph: %s out of range", param
->key
);
434 static void destroy_mount_options(struct ceph_mount_options
*args
)
436 dout("destroy_mount_options %p\n", args
);
440 kfree(args
->snapdir_name
);
441 kfree(args
->mds_namespace
);
442 kfree(args
->server_path
);
443 kfree(args
->fscache_uniq
);
447 static int strcmp_null(const char *s1
, const char *s2
)
455 return strcmp(s1
, s2
);
458 static int compare_mount_options(struct ceph_mount_options
*new_fsopt
,
459 struct ceph_options
*new_opt
,
460 struct ceph_fs_client
*fsc
)
462 struct ceph_mount_options
*fsopt1
= new_fsopt
;
463 struct ceph_mount_options
*fsopt2
= fsc
->mount_options
;
464 int ofs
= offsetof(struct ceph_mount_options
, snapdir_name
);
467 ret
= memcmp(fsopt1
, fsopt2
, ofs
);
471 ret
= strcmp_null(fsopt1
->snapdir_name
, fsopt2
->snapdir_name
);
474 ret
= strcmp_null(fsopt1
->mds_namespace
, fsopt2
->mds_namespace
);
477 ret
= strcmp_null(fsopt1
->server_path
, fsopt2
->server_path
);
480 ret
= strcmp_null(fsopt1
->fscache_uniq
, fsopt2
->fscache_uniq
);
484 return ceph_compare_options(new_opt
, fsc
->client
);
488 * ceph_show_options - Show mount options in /proc/mounts
489 * @m: seq_file to write to
490 * @root: root of that (sub)tree
492 static int ceph_show_options(struct seq_file
*m
, struct dentry
*root
)
494 struct ceph_fs_client
*fsc
= ceph_sb_to_client(root
->d_sb
);
495 struct ceph_mount_options
*fsopt
= fsc
->mount_options
;
499 /* a comma between MNT/MS and client options */
503 ret
= ceph_print_client_options(m
, fsc
->client
, false);
507 /* retract our comma if no client options */
511 if (fsopt
->flags
& CEPH_MOUNT_OPT_DIRSTAT
)
512 seq_puts(m
, ",dirstat");
513 if ((fsopt
->flags
& CEPH_MOUNT_OPT_RBYTES
))
514 seq_puts(m
, ",rbytes");
515 if (fsopt
->flags
& CEPH_MOUNT_OPT_NOASYNCREADDIR
)
516 seq_puts(m
, ",noasyncreaddir");
517 if ((fsopt
->flags
& CEPH_MOUNT_OPT_DCACHE
) == 0)
518 seq_puts(m
, ",nodcache");
519 if (fsopt
->flags
& CEPH_MOUNT_OPT_INO32
)
520 seq_puts(m
, ",ino32");
521 if (fsopt
->flags
& CEPH_MOUNT_OPT_FSCACHE
) {
522 seq_show_option(m
, "fsc", fsopt
->fscache_uniq
);
524 if (fsopt
->flags
& CEPH_MOUNT_OPT_NOPOOLPERM
)
525 seq_puts(m
, ",nopoolperm");
526 if (fsopt
->flags
& CEPH_MOUNT_OPT_NOQUOTADF
)
527 seq_puts(m
, ",noquotadf");
529 #ifdef CONFIG_CEPH_FS_POSIX_ACL
530 if (root
->d_sb
->s_flags
& SB_POSIXACL
)
533 seq_puts(m
, ",noacl");
536 if ((fsopt
->flags
& CEPH_MOUNT_OPT_NOCOPYFROM
) == 0)
537 seq_puts(m
, ",copyfrom");
539 if (fsopt
->mds_namespace
)
540 seq_show_option(m
, "mds_namespace", fsopt
->mds_namespace
);
542 if (fsopt
->flags
& CEPH_MOUNT_OPT_CLEANRECOVER
)
543 seq_show_option(m
, "recover_session", "clean");
545 if (fsopt
->wsize
!= CEPH_MAX_WRITE_SIZE
)
546 seq_printf(m
, ",wsize=%u", fsopt
->wsize
);
547 if (fsopt
->rsize
!= CEPH_MAX_READ_SIZE
)
548 seq_printf(m
, ",rsize=%u", fsopt
->rsize
);
549 if (fsopt
->rasize
!= CEPH_RASIZE_DEFAULT
)
550 seq_printf(m
, ",rasize=%u", fsopt
->rasize
);
551 if (fsopt
->congestion_kb
!= default_congestion_kb())
552 seq_printf(m
, ",write_congestion_kb=%u", fsopt
->congestion_kb
);
554 seq_printf(m
, ",caps_max=%d", fsopt
->caps_max
);
555 if (fsopt
->caps_wanted_delay_min
!= CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT
)
556 seq_printf(m
, ",caps_wanted_delay_min=%u",
557 fsopt
->caps_wanted_delay_min
);
558 if (fsopt
->caps_wanted_delay_max
!= CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT
)
559 seq_printf(m
, ",caps_wanted_delay_max=%u",
560 fsopt
->caps_wanted_delay_max
);
561 if (fsopt
->max_readdir
!= CEPH_MAX_READDIR_DEFAULT
)
562 seq_printf(m
, ",readdir_max_entries=%u", fsopt
->max_readdir
);
563 if (fsopt
->max_readdir_bytes
!= CEPH_MAX_READDIR_BYTES_DEFAULT
)
564 seq_printf(m
, ",readdir_max_bytes=%u", fsopt
->max_readdir_bytes
);
565 if (strcmp(fsopt
->snapdir_name
, CEPH_SNAPDIRNAME_DEFAULT
))
566 seq_show_option(m
, "snapdirname", fsopt
->snapdir_name
);
572 * handle any mon messages the standard library doesn't understand.
573 * return error if we don't either.
575 static int extra_mon_dispatch(struct ceph_client
*client
, struct ceph_msg
*msg
)
577 struct ceph_fs_client
*fsc
= client
->private;
578 int type
= le16_to_cpu(msg
->hdr
.type
);
581 case CEPH_MSG_MDS_MAP
:
582 ceph_mdsc_handle_mdsmap(fsc
->mdsc
, msg
);
584 case CEPH_MSG_FS_MAP_USER
:
585 ceph_mdsc_handle_fsmap(fsc
->mdsc
, msg
);
593 * create a new fs client
595 * Success or not, this function consumes @fsopt and @opt.
597 static struct ceph_fs_client
*create_fs_client(struct ceph_mount_options
*fsopt
,
598 struct ceph_options
*opt
)
600 struct ceph_fs_client
*fsc
;
605 fsc
= kzalloc(sizeof(*fsc
), GFP_KERNEL
);
611 fsc
->client
= ceph_create_client(opt
, fsc
);
612 if (IS_ERR(fsc
->client
)) {
613 err
= PTR_ERR(fsc
->client
);
616 opt
= NULL
; /* fsc->client now owns this */
618 fsc
->client
->extra_mon_dispatch
= extra_mon_dispatch
;
619 ceph_set_opt(fsc
->client
, ABORT_ON_FULL
);
621 if (!fsopt
->mds_namespace
) {
622 ceph_monc_want_map(&fsc
->client
->monc
, CEPH_SUB_MDSMAP
,
625 ceph_monc_want_map(&fsc
->client
->monc
, CEPH_SUB_FSMAP
,
629 fsc
->mount_options
= fsopt
;
632 fsc
->mount_state
= CEPH_MOUNT_MOUNTING
;
635 atomic_long_set(&fsc
->writeback_count
, 0);
639 * The number of concurrent works can be high but they don't need
640 * to be processed in parallel, limit concurrency.
642 fsc
->inode_wq
= alloc_workqueue("ceph-inode", WQ_UNBOUND
, 0);
645 fsc
->cap_wq
= alloc_workqueue("ceph-cap", 0, 1);
649 /* set up mempools */
651 page_count
= fsc
->mount_options
->wsize
>> PAGE_SHIFT
;
652 size
= sizeof (struct page
*) * (page_count
? page_count
: 1);
653 fsc
->wb_pagevec_pool
= mempool_create_kmalloc_pool(10, size
);
654 if (!fsc
->wb_pagevec_pool
)
660 destroy_workqueue(fsc
->cap_wq
);
662 destroy_workqueue(fsc
->inode_wq
);
664 ceph_destroy_client(fsc
->client
);
668 ceph_destroy_options(opt
);
669 destroy_mount_options(fsopt
);
673 static void flush_fs_workqueues(struct ceph_fs_client
*fsc
)
675 flush_workqueue(fsc
->inode_wq
);
676 flush_workqueue(fsc
->cap_wq
);
679 static void destroy_fs_client(struct ceph_fs_client
*fsc
)
681 dout("destroy_fs_client %p\n", fsc
);
683 ceph_mdsc_destroy(fsc
);
684 destroy_workqueue(fsc
->inode_wq
);
685 destroy_workqueue(fsc
->cap_wq
);
687 mempool_destroy(fsc
->wb_pagevec_pool
);
689 destroy_mount_options(fsc
->mount_options
);
691 ceph_destroy_client(fsc
->client
);
694 dout("destroy_fs_client %p done\n", fsc
);
700 struct kmem_cache
*ceph_inode_cachep
;
701 struct kmem_cache
*ceph_cap_cachep
;
702 struct kmem_cache
*ceph_cap_flush_cachep
;
703 struct kmem_cache
*ceph_dentry_cachep
;
704 struct kmem_cache
*ceph_file_cachep
;
705 struct kmem_cache
*ceph_dir_file_cachep
;
707 static void ceph_inode_init_once(void *foo
)
709 struct ceph_inode_info
*ci
= foo
;
710 inode_init_once(&ci
->vfs_inode
);
713 static int __init
init_caches(void)
717 ceph_inode_cachep
= kmem_cache_create("ceph_inode_info",
718 sizeof(struct ceph_inode_info
),
719 __alignof__(struct ceph_inode_info
),
720 SLAB_RECLAIM_ACCOUNT
|SLAB_MEM_SPREAD
|
721 SLAB_ACCOUNT
, ceph_inode_init_once
);
722 if (!ceph_inode_cachep
)
725 ceph_cap_cachep
= KMEM_CACHE(ceph_cap
, SLAB_MEM_SPREAD
);
726 if (!ceph_cap_cachep
)
728 ceph_cap_flush_cachep
= KMEM_CACHE(ceph_cap_flush
,
729 SLAB_RECLAIM_ACCOUNT
|SLAB_MEM_SPREAD
);
730 if (!ceph_cap_flush_cachep
)
733 ceph_dentry_cachep
= KMEM_CACHE(ceph_dentry_info
,
734 SLAB_RECLAIM_ACCOUNT
|SLAB_MEM_SPREAD
);
735 if (!ceph_dentry_cachep
)
738 ceph_file_cachep
= KMEM_CACHE(ceph_file_info
, SLAB_MEM_SPREAD
);
739 if (!ceph_file_cachep
)
742 ceph_dir_file_cachep
= KMEM_CACHE(ceph_dir_file_info
, SLAB_MEM_SPREAD
);
743 if (!ceph_dir_file_cachep
)
746 error
= ceph_fscache_register();
753 kmem_cache_destroy(ceph_dir_file_cachep
);
755 kmem_cache_destroy(ceph_file_cachep
);
757 kmem_cache_destroy(ceph_dentry_cachep
);
759 kmem_cache_destroy(ceph_cap_flush_cachep
);
761 kmem_cache_destroy(ceph_cap_cachep
);
763 kmem_cache_destroy(ceph_inode_cachep
);
767 static void destroy_caches(void)
770 * Make sure all delayed rcu free inodes are flushed before we
775 kmem_cache_destroy(ceph_inode_cachep
);
776 kmem_cache_destroy(ceph_cap_cachep
);
777 kmem_cache_destroy(ceph_cap_flush_cachep
);
778 kmem_cache_destroy(ceph_dentry_cachep
);
779 kmem_cache_destroy(ceph_file_cachep
);
780 kmem_cache_destroy(ceph_dir_file_cachep
);
782 ceph_fscache_unregister();
787 * ceph_umount_begin - initiate forced umount. Tear down down the
788 * mount, skipping steps that may hang while waiting for server(s).
790 static void ceph_umount_begin(struct super_block
*sb
)
792 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
794 dout("ceph_umount_begin - starting forced umount\n");
797 fsc
->mount_state
= CEPH_MOUNT_SHUTDOWN
;
798 ceph_osdc_abort_requests(&fsc
->client
->osdc
, -EIO
);
799 ceph_mdsc_force_umount(fsc
->mdsc
);
800 fsc
->filp_gen
++; // invalidate open files
803 static const struct super_operations ceph_super_ops
= {
804 .alloc_inode
= ceph_alloc_inode
,
805 .free_inode
= ceph_free_inode
,
806 .write_inode
= ceph_write_inode
,
807 .drop_inode
= generic_delete_inode
,
808 .evict_inode
= ceph_evict_inode
,
809 .sync_fs
= ceph_sync_fs
,
810 .put_super
= ceph_put_super
,
811 .show_options
= ceph_show_options
,
812 .statfs
= ceph_statfs
,
813 .umount_begin
= ceph_umount_begin
,
817 * Bootstrap mount by opening the root directory. Note the mount
818 * @started time from caller, and time out if this takes too long.
820 static struct dentry
*open_root_dentry(struct ceph_fs_client
*fsc
,
822 unsigned long started
)
824 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
825 struct ceph_mds_request
*req
= NULL
;
830 dout("open_root_inode opening '%s'\n", path
);
831 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
833 return ERR_CAST(req
);
834 req
->r_path1
= kstrdup(path
, GFP_NOFS
);
836 root
= ERR_PTR(-ENOMEM
);
840 req
->r_ino1
.ino
= CEPH_INO_ROOT
;
841 req
->r_ino1
.snap
= CEPH_NOSNAP
;
842 req
->r_started
= started
;
843 req
->r_timeout
= fsc
->client
->options
->mount_timeout
;
844 req
->r_args
.getattr
.mask
= cpu_to_le32(CEPH_STAT_CAP_INODE
);
846 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
848 struct inode
*inode
= req
->r_target_inode
;
849 req
->r_target_inode
= NULL
;
850 dout("open_root_inode success\n");
851 root
= d_make_root(inode
);
853 root
= ERR_PTR(-ENOMEM
);
856 dout("open_root_inode success, root dentry is %p\n", root
);
861 ceph_mdsc_put_request(req
);
869 * mount: join the ceph cluster, and open root directory.
871 static struct dentry
*ceph_real_mount(struct ceph_fs_client
*fsc
,
872 struct fs_context
*fc
)
875 unsigned long started
= jiffies
; /* note the start time */
878 dout("mount start %p\n", fsc
);
879 mutex_lock(&fsc
->client
->mount_mutex
);
881 if (!fsc
->sb
->s_root
) {
883 err
= __ceph_open_session(fsc
->client
, started
);
888 if (fsc
->mount_options
->flags
& CEPH_MOUNT_OPT_FSCACHE
) {
889 err
= ceph_fscache_register_fs(fsc
, fc
);
894 if (!fsc
->mount_options
->server_path
) {
896 dout("mount opening path \\t\n");
898 path
= fsc
->mount_options
->server_path
+ 1;
899 dout("mount opening path %s\n", path
);
902 ceph_fs_debugfs_init(fsc
);
904 root
= open_root_dentry(fsc
, path
, started
);
909 fsc
->sb
->s_root
= dget(root
);
911 root
= dget(fsc
->sb
->s_root
);
914 fsc
->mount_state
= CEPH_MOUNT_MOUNTED
;
915 dout("mount success\n");
916 mutex_unlock(&fsc
->client
->mount_mutex
);
920 mutex_unlock(&fsc
->client
->mount_mutex
);
924 static int ceph_set_super(struct super_block
*s
, struct fs_context
*fc
)
926 struct ceph_fs_client
*fsc
= s
->s_fs_info
;
929 dout("set_super %p\n", s
);
931 s
->s_maxbytes
= MAX_LFS_FILESIZE
;
933 s
->s_xattr
= ceph_xattr_handlers
;
935 fsc
->max_file_size
= 1ULL << 40; /* temp value until we get mdsmap */
937 s
->s_op
= &ceph_super_ops
;
938 s
->s_d_op
= &ceph_dentry_ops
;
939 s
->s_export_op
= &ceph_export_ops
;
943 s
->s_time_max
= U32_MAX
;
945 ret
= set_anon_super_fc(s
, fc
);
952 * share superblock if same fs AND options
954 static int ceph_compare_super(struct super_block
*sb
, struct fs_context
*fc
)
956 struct ceph_fs_client
*new = fc
->s_fs_info
;
957 struct ceph_mount_options
*fsopt
= new->mount_options
;
958 struct ceph_options
*opt
= new->client
->options
;
959 struct ceph_fs_client
*other
= ceph_sb_to_client(sb
);
961 dout("ceph_compare_super %p\n", sb
);
963 if (compare_mount_options(fsopt
, opt
, other
)) {
964 dout("monitor(s)/mount options don't match\n");
967 if ((opt
->flags
& CEPH_OPT_FSID
) &&
968 ceph_fsid_compare(&opt
->fsid
, &other
->client
->fsid
)) {
969 dout("fsid doesn't match\n");
972 if (fc
->sb_flags
!= (sb
->s_flags
& ~SB_BORN
)) {
973 dout("flags differ\n");
980 * construct our own bdi so we can control readahead, etc.
982 static atomic_long_t bdi_seq
= ATOMIC_LONG_INIT(0);
984 static int ceph_setup_bdi(struct super_block
*sb
, struct ceph_fs_client
*fsc
)
988 err
= super_setup_bdi_name(sb
, "ceph-%ld",
989 atomic_long_inc_return(&bdi_seq
));
993 /* set ra_pages based on rasize mount option? */
994 sb
->s_bdi
->ra_pages
= fsc
->mount_options
->rasize
>> PAGE_SHIFT
;
996 /* set io_pages based on max osd read size */
997 sb
->s_bdi
->io_pages
= fsc
->mount_options
->rsize
>> PAGE_SHIFT
;
1002 static int ceph_get_tree(struct fs_context
*fc
)
1004 struct ceph_parse_opts_ctx
*pctx
= fc
->fs_private
;
1005 struct super_block
*sb
;
1006 struct ceph_fs_client
*fsc
;
1008 int (*compare_super
)(struct super_block
*, struct fs_context
*) =
1012 dout("ceph_get_tree\n");
1015 return invalf(fc
, "ceph: No source");
1017 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1018 fc
->sb_flags
|= SB_POSIXACL
;
1021 /* create client (which we may/may not use) */
1022 fsc
= create_fs_client(pctx
->opts
, pctx
->copts
);
1030 err
= ceph_mdsc_init(fsc
);
1034 if (ceph_test_opt(fsc
->client
, NOSHARE
))
1035 compare_super
= NULL
;
1037 fc
->s_fs_info
= fsc
;
1038 sb
= sget_fc(fc
, compare_super
, ceph_set_super
);
1039 fc
->s_fs_info
= NULL
;
1045 if (ceph_sb_to_client(sb
) != fsc
) {
1046 destroy_fs_client(fsc
);
1047 fsc
= ceph_sb_to_client(sb
);
1048 dout("get_sb got existing client %p\n", fsc
);
1050 dout("get_sb using new client %p\n", fsc
);
1051 err
= ceph_setup_bdi(sb
, fsc
);
1056 res
= ceph_real_mount(fsc
, fc
);
1061 dout("root %p inode %p ino %llx.%llx\n", res
,
1062 d_inode(res
), ceph_vinop(d_inode(res
)));
1063 fc
->root
= fsc
->sb
->s_root
;
1067 ceph_mdsc_close_sessions(fsc
->mdsc
);
1068 deactivate_locked_super(sb
);
1072 destroy_fs_client(fsc
);
1074 dout("ceph_get_tree fail %d\n", err
);
1078 static void ceph_free_fc(struct fs_context
*fc
)
1080 struct ceph_parse_opts_ctx
*pctx
= fc
->fs_private
;
1083 destroy_mount_options(pctx
->opts
);
1084 ceph_destroy_options(pctx
->copts
);
1089 static int ceph_reconfigure_fc(struct fs_context
*fc
)
1091 sync_filesystem(fc
->root
->d_sb
);
1095 static const struct fs_context_operations ceph_context_ops
= {
1096 .free
= ceph_free_fc
,
1097 .parse_param
= ceph_parse_mount_param
,
1098 .get_tree
= ceph_get_tree
,
1099 .reconfigure
= ceph_reconfigure_fc
,
1103 * Set up the filesystem mount context.
1105 static int ceph_init_fs_context(struct fs_context
*fc
)
1107 struct ceph_parse_opts_ctx
*pctx
;
1108 struct ceph_mount_options
*fsopt
;
1110 pctx
= kzalloc(sizeof(*pctx
), GFP_KERNEL
);
1114 pctx
->copts
= ceph_alloc_options();
1118 pctx
->opts
= kzalloc(sizeof(*pctx
->opts
), GFP_KERNEL
);
1123 fsopt
->flags
= CEPH_MOUNT_OPT_DEFAULT
;
1125 fsopt
->wsize
= CEPH_MAX_WRITE_SIZE
;
1126 fsopt
->rsize
= CEPH_MAX_READ_SIZE
;
1127 fsopt
->rasize
= CEPH_RASIZE_DEFAULT
;
1128 fsopt
->snapdir_name
= kstrdup(CEPH_SNAPDIRNAME_DEFAULT
, GFP_KERNEL
);
1129 if (!fsopt
->snapdir_name
)
1132 fsopt
->caps_wanted_delay_min
= CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT
;
1133 fsopt
->caps_wanted_delay_max
= CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT
;
1134 fsopt
->max_readdir
= CEPH_MAX_READDIR_DEFAULT
;
1135 fsopt
->max_readdir_bytes
= CEPH_MAX_READDIR_BYTES_DEFAULT
;
1136 fsopt
->congestion_kb
= default_congestion_kb();
1138 fc
->fs_private
= pctx
;
1139 fc
->ops
= &ceph_context_ops
;
1143 destroy_mount_options(pctx
->opts
);
1144 ceph_destroy_options(pctx
->copts
);
1149 static void ceph_kill_sb(struct super_block
*s
)
1151 struct ceph_fs_client
*fsc
= ceph_sb_to_client(s
);
1152 dev_t dev
= s
->s_dev
;
1154 dout("kill_sb %p\n", s
);
1156 ceph_mdsc_pre_umount(fsc
->mdsc
);
1157 flush_fs_workqueues(fsc
);
1159 generic_shutdown_super(s
);
1161 fsc
->client
->extra_mon_dispatch
= NULL
;
1162 ceph_fs_debugfs_cleanup(fsc
);
1164 ceph_fscache_unregister_fs(fsc
);
1166 destroy_fs_client(fsc
);
1167 free_anon_bdev(dev
);
1170 static struct file_system_type ceph_fs_type
= {
1171 .owner
= THIS_MODULE
,
1173 .init_fs_context
= ceph_init_fs_context
,
1174 .kill_sb
= ceph_kill_sb
,
1175 .fs_flags
= FS_RENAME_DOES_D_MOVE
,
1177 MODULE_ALIAS_FS("ceph");
1179 int ceph_force_reconnect(struct super_block
*sb
)
1181 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1184 ceph_umount_begin(sb
);
1186 /* Make sure all page caches get invalidated.
1187 * see remove_session_caps_cb() */
1188 flush_workqueue(fsc
->inode_wq
);
1190 /* In case that we were blacklisted. This also reset
1191 * all mon/osd connections */
1192 ceph_reset_client_addr(fsc
->client
);
1194 ceph_osdc_clear_abort_err(&fsc
->client
->osdc
);
1196 fsc
->blacklisted
= false;
1197 fsc
->mount_state
= CEPH_MOUNT_MOUNTED
;
1200 err
= __ceph_do_getattr(d_inode(sb
->s_root
), NULL
,
1201 CEPH_STAT_CAP_INODE
, true);
1206 static int __init
init_ceph(void)
1208 int ret
= init_caches();
1213 ret
= register_filesystem(&ceph_fs_type
);
1217 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL
);
1227 static void __exit
exit_ceph(void)
1229 dout("exit_ceph\n");
1230 unregister_filesystem(&ceph_fs_type
);
1234 module_init(init_ceph
);
1235 module_exit(exit_ceph
);
1237 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1238 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1239 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1240 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1241 MODULE_LICENSE("GPL");