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>
30 static DEFINE_SPINLOCK(ceph_fsc_lock
);
31 static LIST_HEAD(ceph_fsc_list
);
34 * Ceph superblock operations
36 * Handle the basics of mounting, unmounting.
42 static void ceph_put_super(struct super_block
*s
)
44 struct ceph_fs_client
*fsc
= ceph_sb_to_client(s
);
47 ceph_mdsc_close_sessions(fsc
->mdsc
);
50 static int ceph_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
52 struct ceph_fs_client
*fsc
= ceph_inode_to_client(d_inode(dentry
));
53 struct ceph_mon_client
*monc
= &fsc
->client
->monc
;
54 struct ceph_statfs st
;
58 if (fsc
->mdsc
->mdsmap
->m_num_data_pg_pools
== 1) {
59 data_pool
= fsc
->mdsc
->mdsmap
->m_data_pg_pools
[0];
61 data_pool
= CEPH_NOPOOL
;
65 err
= ceph_monc_do_statfs(monc
, data_pool
, &st
);
70 buf
->f_type
= CEPH_SUPER_MAGIC
; /* ?? */
73 * express utilization in terms of large blocks to avoid
74 * overflow on 32-bit machines.
76 * NOTE: for the time being, we make bsize == frsize to humor
77 * not-yet-ancient versions of glibc that are broken.
78 * Someday, we will probably want to report a real block
79 * size... whatever that may mean for a network file system!
81 buf
->f_bsize
= 1 << CEPH_BLOCK_SHIFT
;
82 buf
->f_frsize
= 1 << CEPH_BLOCK_SHIFT
;
85 * By default use root quota for stats; fallback to overall filesystem
86 * usage if using 'noquotadf' mount option or if the root dir doesn't
87 * have max_bytes quota set.
89 if (ceph_test_mount_opt(fsc
, NOQUOTADF
) ||
90 !ceph_quota_update_statfs(fsc
, buf
)) {
91 buf
->f_blocks
= le64_to_cpu(st
.kb
) >> (CEPH_BLOCK_SHIFT
-10);
92 buf
->f_bfree
= le64_to_cpu(st
.kb_avail
) >> (CEPH_BLOCK_SHIFT
-10);
93 buf
->f_bavail
= le64_to_cpu(st
.kb_avail
) >> (CEPH_BLOCK_SHIFT
-10);
96 buf
->f_files
= le64_to_cpu(st
.num_objects
);
98 buf
->f_namelen
= NAME_MAX
;
100 /* Must convert the fsid, for consistent values across arches */
101 buf
->f_fsid
.val
[0] = 0;
102 mutex_lock(&monc
->mutex
);
103 for (i
= 0 ; i
< sizeof(monc
->monmap
->fsid
) / sizeof(__le32
) ; ++i
)
104 buf
->f_fsid
.val
[0] ^= le32_to_cpu(((__le32
*)&monc
->monmap
->fsid
)[i
]);
105 mutex_unlock(&monc
->mutex
);
107 /* fold the fs_cluster_id into the upper bits */
108 buf
->f_fsid
.val
[1] = monc
->fs_cluster_id
;
113 static int ceph_sync_fs(struct super_block
*sb
, int wait
)
115 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
118 dout("sync_fs (non-blocking)\n");
119 ceph_flush_dirty_caps(fsc
->mdsc
);
120 dout("sync_fs (non-blocking) done\n");
124 dout("sync_fs (blocking)\n");
125 ceph_osdc_sync(&fsc
->client
->osdc
);
126 ceph_mdsc_sync(fsc
->mdsc
);
127 dout("sync_fs (blocking) done\n");
138 Opt_caps_wanted_delay_min
,
139 Opt_caps_wanted_delay_max
,
141 Opt_readdir_max_entries
,
142 Opt_readdir_max_bytes
,
149 /* string args above */
157 Opt_require_active_mds
,
164 enum ceph_recover_session_mode
{
165 ceph_recover_session_no
,
166 ceph_recover_session_clean
169 static const struct constant_table ceph_param_recover
[] = {
170 { "no", ceph_recover_session_no
},
171 { "clean", ceph_recover_session_clean
},
175 static const struct fs_parameter_spec ceph_mount_parameters
[] = {
176 fsparam_flag_no ("acl", Opt_acl
),
177 fsparam_flag_no ("asyncreaddir", Opt_asyncreaddir
),
178 fsparam_s32 ("caps_max", Opt_caps_max
),
179 fsparam_u32 ("caps_wanted_delay_max", Opt_caps_wanted_delay_max
),
180 fsparam_u32 ("caps_wanted_delay_min", Opt_caps_wanted_delay_min
),
181 fsparam_u32 ("write_congestion_kb", Opt_congestion_kb
),
182 fsparam_flag_no ("copyfrom", Opt_copyfrom
),
183 fsparam_flag_no ("dcache", Opt_dcache
),
184 fsparam_flag_no ("dirstat", Opt_dirstat
),
185 fsparam_flag_no ("fsc", Opt_fscache
), // fsc|nofsc
186 fsparam_string ("fsc", Opt_fscache
), // fsc=...
187 fsparam_flag_no ("ino32", Opt_ino32
),
188 fsparam_string ("mds_namespace", Opt_mds_namespace
),
189 fsparam_flag_no ("poolperm", Opt_poolperm
),
190 fsparam_flag_no ("quotadf", Opt_quotadf
),
191 fsparam_u32 ("rasize", Opt_rasize
),
192 fsparam_flag_no ("rbytes", Opt_rbytes
),
193 fsparam_u32 ("readdir_max_bytes", Opt_readdir_max_bytes
),
194 fsparam_u32 ("readdir_max_entries", Opt_readdir_max_entries
),
195 fsparam_enum ("recover_session", Opt_recover_session
, ceph_param_recover
),
196 fsparam_flag_no ("require_active_mds", Opt_require_active_mds
),
197 fsparam_u32 ("rsize", Opt_rsize
),
198 fsparam_string ("snapdirname", Opt_snapdirname
),
199 fsparam_string ("source", Opt_source
),
200 fsparam_u32 ("wsize", Opt_wsize
),
201 fsparam_flag_no ("wsync", Opt_wsync
),
205 struct ceph_parse_opts_ctx
{
206 struct ceph_options
*copts
;
207 struct ceph_mount_options
*opts
;
211 * Remove adjacent slashes and then the trailing slash, unless it is
212 * the only remaining character.
214 * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
216 static void canonicalize_path(char *path
)
220 for (i
= 0; path
[i
] != '\0'; i
++) {
221 if (path
[i
] != '/' || j
< 1 || path
[j
- 1] != '/')
225 if (j
> 1 && path
[j
- 1] == '/')
231 * Parse the source parameter. Distinguish the server list from the path.
233 * The source will look like:
234 * <server_spec>[,<server_spec>...]:[<path>]
236 * <server_spec> is <ip>[:<port>]
237 * <path> is optional, but if present must begin with '/'
239 static int ceph_parse_source(struct fs_parameter
*param
, struct fs_context
*fc
)
241 struct ceph_parse_opts_ctx
*pctx
= fc
->fs_private
;
242 struct ceph_mount_options
*fsopt
= pctx
->opts
;
243 char *dev_name
= param
->string
, *dev_name_end
;
246 dout("%s '%s'\n", __func__
, dev_name
);
247 if (!dev_name
|| !*dev_name
)
248 return invalfc(fc
, "Empty source");
250 dev_name_end
= strchr(dev_name
, '/');
253 * The server_path will include the whole chars from userland
254 * including the leading '/'.
256 kfree(fsopt
->server_path
);
257 fsopt
->server_path
= kstrdup(dev_name_end
, GFP_KERNEL
);
258 if (!fsopt
->server_path
)
261 canonicalize_path(fsopt
->server_path
);
263 dev_name_end
= dev_name
+ strlen(dev_name
);
266 dev_name_end
--; /* back up to ':' separator */
267 if (dev_name_end
< dev_name
|| *dev_name_end
!= ':')
268 return invalfc(fc
, "No path or : separator in source");
270 dout("device name '%.*s'\n", (int)(dev_name_end
- dev_name
), dev_name
);
271 if (fsopt
->server_path
)
272 dout("server path '%s'\n", fsopt
->server_path
);
274 ret
= ceph_parse_mon_ips(param
->string
, dev_name_end
- dev_name
,
275 pctx
->copts
, fc
->log
.log
);
279 fc
->source
= param
->string
;
280 param
->string
= NULL
;
284 static int ceph_parse_mount_param(struct fs_context
*fc
,
285 struct fs_parameter
*param
)
287 struct ceph_parse_opts_ctx
*pctx
= fc
->fs_private
;
288 struct ceph_mount_options
*fsopt
= pctx
->opts
;
289 struct fs_parse_result result
;
293 ret
= ceph_parse_param(param
, pctx
->copts
, fc
->log
.log
);
294 if (ret
!= -ENOPARAM
)
297 token
= fs_parse(fc
, ceph_mount_parameters
, param
, &result
);
298 dout("%s fs_parse '%s' token %d\n", __func__
, param
->key
, token
);
303 case Opt_snapdirname
:
304 kfree(fsopt
->snapdir_name
);
305 fsopt
->snapdir_name
= param
->string
;
306 param
->string
= NULL
;
308 case Opt_mds_namespace
:
309 kfree(fsopt
->mds_namespace
);
310 fsopt
->mds_namespace
= param
->string
;
311 param
->string
= NULL
;
313 case Opt_recover_session
:
314 mode
= result
.uint_32
;
315 if (mode
== ceph_recover_session_no
)
316 fsopt
->flags
&= ~CEPH_MOUNT_OPT_CLEANRECOVER
;
317 else if (mode
== ceph_recover_session_clean
)
318 fsopt
->flags
|= CEPH_MOUNT_OPT_CLEANRECOVER
;
324 return invalfc(fc
, "Multiple sources specified");
325 return ceph_parse_source(param
, fc
);
327 if (result
.uint_32
< PAGE_SIZE
||
328 result
.uint_32
> CEPH_MAX_WRITE_SIZE
)
330 fsopt
->wsize
= ALIGN(result
.uint_32
, PAGE_SIZE
);
333 if (result
.uint_32
< PAGE_SIZE
||
334 result
.uint_32
> CEPH_MAX_READ_SIZE
)
336 fsopt
->rsize
= ALIGN(result
.uint_32
, PAGE_SIZE
);
339 fsopt
->rasize
= ALIGN(result
.uint_32
, PAGE_SIZE
);
341 case Opt_caps_wanted_delay_min
:
342 if (result
.uint_32
< 1)
344 fsopt
->caps_wanted_delay_min
= result
.uint_32
;
346 case Opt_caps_wanted_delay_max
:
347 if (result
.uint_32
< 1)
349 fsopt
->caps_wanted_delay_max
= result
.uint_32
;
352 if (result
.int_32
< 0)
354 fsopt
->caps_max
= result
.int_32
;
356 case Opt_readdir_max_entries
:
357 if (result
.uint_32
< 1)
359 fsopt
->max_readdir
= result
.uint_32
;
361 case Opt_readdir_max_bytes
:
362 if (result
.uint_32
< PAGE_SIZE
&& result
.uint_32
!= 0)
364 fsopt
->max_readdir_bytes
= result
.uint_32
;
366 case Opt_congestion_kb
:
367 if (result
.uint_32
< 1024) /* at least 1M */
369 fsopt
->congestion_kb
= result
.uint_32
;
373 fsopt
->flags
|= CEPH_MOUNT_OPT_DIRSTAT
;
375 fsopt
->flags
&= ~CEPH_MOUNT_OPT_DIRSTAT
;
379 fsopt
->flags
|= CEPH_MOUNT_OPT_RBYTES
;
381 fsopt
->flags
&= ~CEPH_MOUNT_OPT_RBYTES
;
383 case Opt_asyncreaddir
:
385 fsopt
->flags
&= ~CEPH_MOUNT_OPT_NOASYNCREADDIR
;
387 fsopt
->flags
|= CEPH_MOUNT_OPT_NOASYNCREADDIR
;
391 fsopt
->flags
|= CEPH_MOUNT_OPT_DCACHE
;
393 fsopt
->flags
&= ~CEPH_MOUNT_OPT_DCACHE
;
397 fsopt
->flags
|= CEPH_MOUNT_OPT_INO32
;
399 fsopt
->flags
&= ~CEPH_MOUNT_OPT_INO32
;
403 #ifdef CONFIG_CEPH_FSCACHE
404 kfree(fsopt
->fscache_uniq
);
405 fsopt
->fscache_uniq
= NULL
;
406 if (result
.negated
) {
407 fsopt
->flags
&= ~CEPH_MOUNT_OPT_FSCACHE
;
409 fsopt
->flags
|= CEPH_MOUNT_OPT_FSCACHE
;
410 fsopt
->fscache_uniq
= param
->string
;
411 param
->string
= NULL
;
415 return invalfc(fc
, "fscache support is disabled");
419 fsopt
->flags
&= ~CEPH_MOUNT_OPT_NOPOOLPERM
;
421 fsopt
->flags
|= CEPH_MOUNT_OPT_NOPOOLPERM
;
423 case Opt_require_active_mds
:
425 fsopt
->flags
&= ~CEPH_MOUNT_OPT_MOUNTWAIT
;
427 fsopt
->flags
|= CEPH_MOUNT_OPT_MOUNTWAIT
;
431 fsopt
->flags
&= ~CEPH_MOUNT_OPT_NOQUOTADF
;
433 fsopt
->flags
|= CEPH_MOUNT_OPT_NOQUOTADF
;
437 fsopt
->flags
&= ~CEPH_MOUNT_OPT_NOCOPYFROM
;
439 fsopt
->flags
|= CEPH_MOUNT_OPT_NOCOPYFROM
;
442 if (!result
.negated
) {
443 #ifdef CONFIG_CEPH_FS_POSIX_ACL
444 fc
->sb_flags
|= SB_POSIXACL
;
446 return invalfc(fc
, "POSIX ACL support is disabled");
449 fc
->sb_flags
&= ~SB_POSIXACL
;
454 fsopt
->flags
&= ~CEPH_MOUNT_OPT_ASYNC_DIROPS
;
456 fsopt
->flags
|= CEPH_MOUNT_OPT_ASYNC_DIROPS
;
464 return invalfc(fc
, "%s out of range", param
->key
);
467 static void destroy_mount_options(struct ceph_mount_options
*args
)
469 dout("destroy_mount_options %p\n", args
);
473 kfree(args
->snapdir_name
);
474 kfree(args
->mds_namespace
);
475 kfree(args
->server_path
);
476 kfree(args
->fscache_uniq
);
480 static int strcmp_null(const char *s1
, const char *s2
)
488 return strcmp(s1
, s2
);
491 static int compare_mount_options(struct ceph_mount_options
*new_fsopt
,
492 struct ceph_options
*new_opt
,
493 struct ceph_fs_client
*fsc
)
495 struct ceph_mount_options
*fsopt1
= new_fsopt
;
496 struct ceph_mount_options
*fsopt2
= fsc
->mount_options
;
497 int ofs
= offsetof(struct ceph_mount_options
, snapdir_name
);
500 ret
= memcmp(fsopt1
, fsopt2
, ofs
);
504 ret
= strcmp_null(fsopt1
->snapdir_name
, fsopt2
->snapdir_name
);
508 ret
= strcmp_null(fsopt1
->mds_namespace
, fsopt2
->mds_namespace
);
512 ret
= strcmp_null(fsopt1
->server_path
, fsopt2
->server_path
);
516 ret
= strcmp_null(fsopt1
->fscache_uniq
, fsopt2
->fscache_uniq
);
520 return ceph_compare_options(new_opt
, fsc
->client
);
524 * ceph_show_options - Show mount options in /proc/mounts
525 * @m: seq_file to write to
526 * @root: root of that (sub)tree
528 static int ceph_show_options(struct seq_file
*m
, struct dentry
*root
)
530 struct ceph_fs_client
*fsc
= ceph_sb_to_client(root
->d_sb
);
531 struct ceph_mount_options
*fsopt
= fsc
->mount_options
;
535 /* a comma between MNT/MS and client options */
539 ret
= ceph_print_client_options(m
, fsc
->client
, false);
543 /* retract our comma if no client options */
547 if (fsopt
->flags
& CEPH_MOUNT_OPT_DIRSTAT
)
548 seq_puts(m
, ",dirstat");
549 if ((fsopt
->flags
& CEPH_MOUNT_OPT_RBYTES
))
550 seq_puts(m
, ",rbytes");
551 if (fsopt
->flags
& CEPH_MOUNT_OPT_NOASYNCREADDIR
)
552 seq_puts(m
, ",noasyncreaddir");
553 if ((fsopt
->flags
& CEPH_MOUNT_OPT_DCACHE
) == 0)
554 seq_puts(m
, ",nodcache");
555 if (fsopt
->flags
& CEPH_MOUNT_OPT_INO32
)
556 seq_puts(m
, ",ino32");
557 if (fsopt
->flags
& CEPH_MOUNT_OPT_FSCACHE
) {
558 seq_show_option(m
, "fsc", fsopt
->fscache_uniq
);
560 if (fsopt
->flags
& CEPH_MOUNT_OPT_NOPOOLPERM
)
561 seq_puts(m
, ",nopoolperm");
562 if (fsopt
->flags
& CEPH_MOUNT_OPT_NOQUOTADF
)
563 seq_puts(m
, ",noquotadf");
565 #ifdef CONFIG_CEPH_FS_POSIX_ACL
566 if (root
->d_sb
->s_flags
& SB_POSIXACL
)
569 seq_puts(m
, ",noacl");
572 if ((fsopt
->flags
& CEPH_MOUNT_OPT_NOCOPYFROM
) == 0)
573 seq_puts(m
, ",copyfrom");
575 if (fsopt
->mds_namespace
)
576 seq_show_option(m
, "mds_namespace", fsopt
->mds_namespace
);
578 if (fsopt
->flags
& CEPH_MOUNT_OPT_CLEANRECOVER
)
579 seq_show_option(m
, "recover_session", "clean");
581 if (fsopt
->flags
& CEPH_MOUNT_OPT_ASYNC_DIROPS
)
582 seq_puts(m
, ",nowsync");
584 if (fsopt
->wsize
!= CEPH_MAX_WRITE_SIZE
)
585 seq_printf(m
, ",wsize=%u", fsopt
->wsize
);
586 if (fsopt
->rsize
!= CEPH_MAX_READ_SIZE
)
587 seq_printf(m
, ",rsize=%u", fsopt
->rsize
);
588 if (fsopt
->rasize
!= CEPH_RASIZE_DEFAULT
)
589 seq_printf(m
, ",rasize=%u", fsopt
->rasize
);
590 if (fsopt
->congestion_kb
!= default_congestion_kb())
591 seq_printf(m
, ",write_congestion_kb=%u", fsopt
->congestion_kb
);
593 seq_printf(m
, ",caps_max=%d", fsopt
->caps_max
);
594 if (fsopt
->caps_wanted_delay_min
!= CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT
)
595 seq_printf(m
, ",caps_wanted_delay_min=%u",
596 fsopt
->caps_wanted_delay_min
);
597 if (fsopt
->caps_wanted_delay_max
!= CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT
)
598 seq_printf(m
, ",caps_wanted_delay_max=%u",
599 fsopt
->caps_wanted_delay_max
);
600 if (fsopt
->max_readdir
!= CEPH_MAX_READDIR_DEFAULT
)
601 seq_printf(m
, ",readdir_max_entries=%u", fsopt
->max_readdir
);
602 if (fsopt
->max_readdir_bytes
!= CEPH_MAX_READDIR_BYTES_DEFAULT
)
603 seq_printf(m
, ",readdir_max_bytes=%u", fsopt
->max_readdir_bytes
);
604 if (strcmp(fsopt
->snapdir_name
, CEPH_SNAPDIRNAME_DEFAULT
))
605 seq_show_option(m
, "snapdirname", fsopt
->snapdir_name
);
611 * handle any mon messages the standard library doesn't understand.
612 * return error if we don't either.
614 static int extra_mon_dispatch(struct ceph_client
*client
, struct ceph_msg
*msg
)
616 struct ceph_fs_client
*fsc
= client
->private;
617 int type
= le16_to_cpu(msg
->hdr
.type
);
620 case CEPH_MSG_MDS_MAP
:
621 ceph_mdsc_handle_mdsmap(fsc
->mdsc
, msg
);
623 case CEPH_MSG_FS_MAP_USER
:
624 ceph_mdsc_handle_fsmap(fsc
->mdsc
, msg
);
632 * create a new fs client
634 * Success or not, this function consumes @fsopt and @opt.
636 static struct ceph_fs_client
*create_fs_client(struct ceph_mount_options
*fsopt
,
637 struct ceph_options
*opt
)
639 struct ceph_fs_client
*fsc
;
642 fsc
= kzalloc(sizeof(*fsc
), GFP_KERNEL
);
648 fsc
->client
= ceph_create_client(opt
, fsc
);
649 if (IS_ERR(fsc
->client
)) {
650 err
= PTR_ERR(fsc
->client
);
653 opt
= NULL
; /* fsc->client now owns this */
655 fsc
->client
->extra_mon_dispatch
= extra_mon_dispatch
;
656 ceph_set_opt(fsc
->client
, ABORT_ON_FULL
);
658 if (!fsopt
->mds_namespace
) {
659 ceph_monc_want_map(&fsc
->client
->monc
, CEPH_SUB_MDSMAP
,
662 ceph_monc_want_map(&fsc
->client
->monc
, CEPH_SUB_FSMAP
,
666 fsc
->mount_options
= fsopt
;
669 fsc
->mount_state
= CEPH_MOUNT_MOUNTING
;
671 fsc
->have_copy_from2
= true;
673 atomic_long_set(&fsc
->writeback_count
, 0);
677 * The number of concurrent works can be high but they don't need
678 * to be processed in parallel, limit concurrency.
680 fsc
->inode_wq
= alloc_workqueue("ceph-inode", WQ_UNBOUND
, 0);
683 fsc
->cap_wq
= alloc_workqueue("ceph-cap", 0, 1);
687 spin_lock(&ceph_fsc_lock
);
688 list_add_tail(&fsc
->metric_wakeup
, &ceph_fsc_list
);
689 spin_unlock(&ceph_fsc_lock
);
694 destroy_workqueue(fsc
->inode_wq
);
696 ceph_destroy_client(fsc
->client
);
700 ceph_destroy_options(opt
);
701 destroy_mount_options(fsopt
);
705 static void flush_fs_workqueues(struct ceph_fs_client
*fsc
)
707 flush_workqueue(fsc
->inode_wq
);
708 flush_workqueue(fsc
->cap_wq
);
711 static void destroy_fs_client(struct ceph_fs_client
*fsc
)
713 dout("destroy_fs_client %p\n", fsc
);
715 spin_lock(&ceph_fsc_lock
);
716 list_del(&fsc
->metric_wakeup
);
717 spin_unlock(&ceph_fsc_lock
);
719 ceph_mdsc_destroy(fsc
);
720 destroy_workqueue(fsc
->inode_wq
);
721 destroy_workqueue(fsc
->cap_wq
);
723 destroy_mount_options(fsc
->mount_options
);
725 ceph_destroy_client(fsc
->client
);
728 dout("destroy_fs_client %p done\n", fsc
);
734 struct kmem_cache
*ceph_inode_cachep
;
735 struct kmem_cache
*ceph_cap_cachep
;
736 struct kmem_cache
*ceph_cap_flush_cachep
;
737 struct kmem_cache
*ceph_dentry_cachep
;
738 struct kmem_cache
*ceph_file_cachep
;
739 struct kmem_cache
*ceph_dir_file_cachep
;
740 struct kmem_cache
*ceph_mds_request_cachep
;
741 mempool_t
*ceph_wb_pagevec_pool
;
743 static void ceph_inode_init_once(void *foo
)
745 struct ceph_inode_info
*ci
= foo
;
746 inode_init_once(&ci
->vfs_inode
);
749 static int __init
init_caches(void)
753 ceph_inode_cachep
= kmem_cache_create("ceph_inode_info",
754 sizeof(struct ceph_inode_info
),
755 __alignof__(struct ceph_inode_info
),
756 SLAB_RECLAIM_ACCOUNT
|SLAB_MEM_SPREAD
|
757 SLAB_ACCOUNT
, ceph_inode_init_once
);
758 if (!ceph_inode_cachep
)
761 ceph_cap_cachep
= KMEM_CACHE(ceph_cap
, SLAB_MEM_SPREAD
);
762 if (!ceph_cap_cachep
)
764 ceph_cap_flush_cachep
= KMEM_CACHE(ceph_cap_flush
,
765 SLAB_RECLAIM_ACCOUNT
|SLAB_MEM_SPREAD
);
766 if (!ceph_cap_flush_cachep
)
769 ceph_dentry_cachep
= KMEM_CACHE(ceph_dentry_info
,
770 SLAB_RECLAIM_ACCOUNT
|SLAB_MEM_SPREAD
);
771 if (!ceph_dentry_cachep
)
774 ceph_file_cachep
= KMEM_CACHE(ceph_file_info
, SLAB_MEM_SPREAD
);
775 if (!ceph_file_cachep
)
778 ceph_dir_file_cachep
= KMEM_CACHE(ceph_dir_file_info
, SLAB_MEM_SPREAD
);
779 if (!ceph_dir_file_cachep
)
782 ceph_mds_request_cachep
= KMEM_CACHE(ceph_mds_request
, SLAB_MEM_SPREAD
);
783 if (!ceph_mds_request_cachep
)
786 ceph_wb_pagevec_pool
= mempool_create_kmalloc_pool(10, CEPH_MAX_WRITE_SIZE
>> PAGE_SHIFT
);
787 if (!ceph_wb_pagevec_pool
)
788 goto bad_pagevec_pool
;
790 error
= ceph_fscache_register();
797 kmem_cache_destroy(ceph_mds_request_cachep
);
799 mempool_destroy(ceph_wb_pagevec_pool
);
801 kmem_cache_destroy(ceph_dir_file_cachep
);
803 kmem_cache_destroy(ceph_file_cachep
);
805 kmem_cache_destroy(ceph_dentry_cachep
);
807 kmem_cache_destroy(ceph_cap_flush_cachep
);
809 kmem_cache_destroy(ceph_cap_cachep
);
811 kmem_cache_destroy(ceph_inode_cachep
);
815 static void destroy_caches(void)
818 * Make sure all delayed rcu free inodes are flushed before we
823 kmem_cache_destroy(ceph_inode_cachep
);
824 kmem_cache_destroy(ceph_cap_cachep
);
825 kmem_cache_destroy(ceph_cap_flush_cachep
);
826 kmem_cache_destroy(ceph_dentry_cachep
);
827 kmem_cache_destroy(ceph_file_cachep
);
828 kmem_cache_destroy(ceph_dir_file_cachep
);
829 kmem_cache_destroy(ceph_mds_request_cachep
);
830 mempool_destroy(ceph_wb_pagevec_pool
);
832 ceph_fscache_unregister();
835 static void __ceph_umount_begin(struct ceph_fs_client
*fsc
)
837 ceph_osdc_abort_requests(&fsc
->client
->osdc
, -EIO
);
838 ceph_mdsc_force_umount(fsc
->mdsc
);
839 fsc
->filp_gen
++; // invalidate open files
843 * ceph_umount_begin - initiate forced umount. Tear down the
844 * mount, skipping steps that may hang while waiting for server(s).
846 static void ceph_umount_begin(struct super_block
*sb
)
848 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
850 dout("ceph_umount_begin - starting forced umount\n");
853 fsc
->mount_state
= CEPH_MOUNT_SHUTDOWN
;
854 __ceph_umount_begin(fsc
);
857 static const struct super_operations ceph_super_ops
= {
858 .alloc_inode
= ceph_alloc_inode
,
859 .free_inode
= ceph_free_inode
,
860 .write_inode
= ceph_write_inode
,
861 .drop_inode
= generic_delete_inode
,
862 .evict_inode
= ceph_evict_inode
,
863 .sync_fs
= ceph_sync_fs
,
864 .put_super
= ceph_put_super
,
865 .show_options
= ceph_show_options
,
866 .statfs
= ceph_statfs
,
867 .umount_begin
= ceph_umount_begin
,
871 * Bootstrap mount by opening the root directory. Note the mount
872 * @started time from caller, and time out if this takes too long.
874 static struct dentry
*open_root_dentry(struct ceph_fs_client
*fsc
,
876 unsigned long started
)
878 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
879 struct ceph_mds_request
*req
= NULL
;
884 dout("open_root_inode opening '%s'\n", path
);
885 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
887 return ERR_CAST(req
);
888 req
->r_path1
= kstrdup(path
, GFP_NOFS
);
890 root
= ERR_PTR(-ENOMEM
);
894 req
->r_ino1
.ino
= CEPH_INO_ROOT
;
895 req
->r_ino1
.snap
= CEPH_NOSNAP
;
896 req
->r_started
= started
;
897 req
->r_timeout
= fsc
->client
->options
->mount_timeout
;
898 req
->r_args
.getattr
.mask
= cpu_to_le32(CEPH_STAT_CAP_INODE
);
900 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
902 struct inode
*inode
= req
->r_target_inode
;
903 req
->r_target_inode
= NULL
;
904 dout("open_root_inode success\n");
905 root
= d_make_root(inode
);
907 root
= ERR_PTR(-ENOMEM
);
910 dout("open_root_inode success, root dentry is %p\n", root
);
915 ceph_mdsc_put_request(req
);
920 * mount: join the ceph cluster, and open root directory.
922 static struct dentry
*ceph_real_mount(struct ceph_fs_client
*fsc
,
923 struct fs_context
*fc
)
926 unsigned long started
= jiffies
; /* note the start time */
929 dout("mount start %p\n", fsc
);
930 mutex_lock(&fsc
->client
->mount_mutex
);
932 if (!fsc
->sb
->s_root
) {
933 const char *path
= fsc
->mount_options
->server_path
?
934 fsc
->mount_options
->server_path
+ 1 : "";
936 err
= __ceph_open_session(fsc
->client
, started
);
941 if (fsc
->mount_options
->flags
& CEPH_MOUNT_OPT_FSCACHE
) {
942 err
= ceph_fscache_register_fs(fsc
, fc
);
947 dout("mount opening path '%s'\n", path
);
949 ceph_fs_debugfs_init(fsc
);
951 root
= open_root_dentry(fsc
, path
, started
);
956 fsc
->sb
->s_root
= dget(root
);
958 root
= dget(fsc
->sb
->s_root
);
961 fsc
->mount_state
= CEPH_MOUNT_MOUNTED
;
962 dout("mount success\n");
963 mutex_unlock(&fsc
->client
->mount_mutex
);
967 mutex_unlock(&fsc
->client
->mount_mutex
);
971 static int ceph_set_super(struct super_block
*s
, struct fs_context
*fc
)
973 struct ceph_fs_client
*fsc
= s
->s_fs_info
;
976 dout("set_super %p\n", s
);
978 s
->s_maxbytes
= MAX_LFS_FILESIZE
;
980 s
->s_xattr
= ceph_xattr_handlers
;
982 fsc
->max_file_size
= 1ULL << 40; /* temp value until we get mdsmap */
984 s
->s_op
= &ceph_super_ops
;
985 s
->s_d_op
= &ceph_dentry_ops
;
986 s
->s_export_op
= &ceph_export_ops
;
990 s
->s_time_max
= U32_MAX
;
992 ret
= set_anon_super_fc(s
, fc
);
999 * share superblock if same fs AND options
1001 static int ceph_compare_super(struct super_block
*sb
, struct fs_context
*fc
)
1003 struct ceph_fs_client
*new = fc
->s_fs_info
;
1004 struct ceph_mount_options
*fsopt
= new->mount_options
;
1005 struct ceph_options
*opt
= new->client
->options
;
1006 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1008 dout("ceph_compare_super %p\n", sb
);
1010 if (compare_mount_options(fsopt
, opt
, fsc
)) {
1011 dout("monitor(s)/mount options don't match\n");
1014 if ((opt
->flags
& CEPH_OPT_FSID
) &&
1015 ceph_fsid_compare(&opt
->fsid
, &fsc
->client
->fsid
)) {
1016 dout("fsid doesn't match\n");
1019 if (fc
->sb_flags
!= (sb
->s_flags
& ~SB_BORN
)) {
1020 dout("flags differ\n");
1024 if (fsc
->blocklisted
&& !ceph_test_mount_opt(fsc
, CLEANRECOVER
)) {
1025 dout("client is blocklisted (and CLEANRECOVER is not set)\n");
1029 if (fsc
->mount_state
== CEPH_MOUNT_SHUTDOWN
) {
1030 dout("client has been forcibly unmounted\n");
1038 * construct our own bdi so we can control readahead, etc.
1040 static atomic_long_t bdi_seq
= ATOMIC_LONG_INIT(0);
1042 static int ceph_setup_bdi(struct super_block
*sb
, struct ceph_fs_client
*fsc
)
1046 err
= super_setup_bdi_name(sb
, "ceph-%ld",
1047 atomic_long_inc_return(&bdi_seq
));
1051 /* set ra_pages based on rasize mount option? */
1052 sb
->s_bdi
->ra_pages
= fsc
->mount_options
->rasize
>> PAGE_SHIFT
;
1054 /* set io_pages based on max osd read size */
1055 sb
->s_bdi
->io_pages
= fsc
->mount_options
->rsize
>> PAGE_SHIFT
;
1060 static int ceph_get_tree(struct fs_context
*fc
)
1062 struct ceph_parse_opts_ctx
*pctx
= fc
->fs_private
;
1063 struct super_block
*sb
;
1064 struct ceph_fs_client
*fsc
;
1066 int (*compare_super
)(struct super_block
*, struct fs_context
*) =
1070 dout("ceph_get_tree\n");
1073 return invalfc(fc
, "No source");
1075 /* create client (which we may/may not use) */
1076 fsc
= create_fs_client(pctx
->opts
, pctx
->copts
);
1084 err
= ceph_mdsc_init(fsc
);
1088 if (ceph_test_opt(fsc
->client
, NOSHARE
))
1089 compare_super
= NULL
;
1091 fc
->s_fs_info
= fsc
;
1092 sb
= sget_fc(fc
, compare_super
, ceph_set_super
);
1093 fc
->s_fs_info
= NULL
;
1099 if (ceph_sb_to_client(sb
) != fsc
) {
1100 destroy_fs_client(fsc
);
1101 fsc
= ceph_sb_to_client(sb
);
1102 dout("get_sb got existing client %p\n", fsc
);
1104 dout("get_sb using new client %p\n", fsc
);
1105 err
= ceph_setup_bdi(sb
, fsc
);
1110 res
= ceph_real_mount(fsc
, fc
);
1115 dout("root %p inode %p ino %llx.%llx\n", res
,
1116 d_inode(res
), ceph_vinop(d_inode(res
)));
1117 fc
->root
= fsc
->sb
->s_root
;
1121 if (!ceph_mdsmap_is_cluster_available(fsc
->mdsc
->mdsmap
)) {
1122 pr_info("No mds server is up or the cluster is laggy\n");
1123 err
= -EHOSTUNREACH
;
1126 ceph_mdsc_close_sessions(fsc
->mdsc
);
1127 deactivate_locked_super(sb
);
1131 destroy_fs_client(fsc
);
1133 dout("ceph_get_tree fail %d\n", err
);
1137 static void ceph_free_fc(struct fs_context
*fc
)
1139 struct ceph_parse_opts_ctx
*pctx
= fc
->fs_private
;
1142 destroy_mount_options(pctx
->opts
);
1143 ceph_destroy_options(pctx
->copts
);
1148 static int ceph_reconfigure_fc(struct fs_context
*fc
)
1150 struct ceph_parse_opts_ctx
*pctx
= fc
->fs_private
;
1151 struct ceph_mount_options
*fsopt
= pctx
->opts
;
1152 struct ceph_fs_client
*fsc
= ceph_sb_to_client(fc
->root
->d_sb
);
1154 if (fsopt
->flags
& CEPH_MOUNT_OPT_ASYNC_DIROPS
)
1155 ceph_set_mount_opt(fsc
, ASYNC_DIROPS
);
1157 ceph_clear_mount_opt(fsc
, ASYNC_DIROPS
);
1159 sync_filesystem(fc
->root
->d_sb
);
1163 static const struct fs_context_operations ceph_context_ops
= {
1164 .free
= ceph_free_fc
,
1165 .parse_param
= ceph_parse_mount_param
,
1166 .get_tree
= ceph_get_tree
,
1167 .reconfigure
= ceph_reconfigure_fc
,
1171 * Set up the filesystem mount context.
1173 static int ceph_init_fs_context(struct fs_context
*fc
)
1175 struct ceph_parse_opts_ctx
*pctx
;
1176 struct ceph_mount_options
*fsopt
;
1178 pctx
= kzalloc(sizeof(*pctx
), GFP_KERNEL
);
1182 pctx
->copts
= ceph_alloc_options();
1186 pctx
->opts
= kzalloc(sizeof(*pctx
->opts
), GFP_KERNEL
);
1191 fsopt
->flags
= CEPH_MOUNT_OPT_DEFAULT
;
1193 fsopt
->wsize
= CEPH_MAX_WRITE_SIZE
;
1194 fsopt
->rsize
= CEPH_MAX_READ_SIZE
;
1195 fsopt
->rasize
= CEPH_RASIZE_DEFAULT
;
1196 fsopt
->snapdir_name
= kstrdup(CEPH_SNAPDIRNAME_DEFAULT
, GFP_KERNEL
);
1197 if (!fsopt
->snapdir_name
)
1200 fsopt
->caps_wanted_delay_min
= CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT
;
1201 fsopt
->caps_wanted_delay_max
= CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT
;
1202 fsopt
->max_readdir
= CEPH_MAX_READDIR_DEFAULT
;
1203 fsopt
->max_readdir_bytes
= CEPH_MAX_READDIR_BYTES_DEFAULT
;
1204 fsopt
->congestion_kb
= default_congestion_kb();
1206 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1207 fc
->sb_flags
|= SB_POSIXACL
;
1210 fc
->fs_private
= pctx
;
1211 fc
->ops
= &ceph_context_ops
;
1215 destroy_mount_options(pctx
->opts
);
1216 ceph_destroy_options(pctx
->copts
);
1221 static void ceph_kill_sb(struct super_block
*s
)
1223 struct ceph_fs_client
*fsc
= ceph_sb_to_client(s
);
1225 dout("kill_sb %p\n", s
);
1227 ceph_mdsc_pre_umount(fsc
->mdsc
);
1228 flush_fs_workqueues(fsc
);
1232 fsc
->client
->extra_mon_dispatch
= NULL
;
1233 ceph_fs_debugfs_cleanup(fsc
);
1235 ceph_fscache_unregister_fs(fsc
);
1237 destroy_fs_client(fsc
);
1240 static struct file_system_type ceph_fs_type
= {
1241 .owner
= THIS_MODULE
,
1243 .init_fs_context
= ceph_init_fs_context
,
1244 .kill_sb
= ceph_kill_sb
,
1245 .fs_flags
= FS_RENAME_DOES_D_MOVE
,
1247 MODULE_ALIAS_FS("ceph");
1249 int ceph_force_reconnect(struct super_block
*sb
)
1251 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1254 fsc
->mount_state
= CEPH_MOUNT_RECOVER
;
1255 __ceph_umount_begin(fsc
);
1257 /* Make sure all page caches get invalidated.
1258 * see remove_session_caps_cb() */
1259 flush_workqueue(fsc
->inode_wq
);
1261 /* In case that we were blocklisted. This also reset
1262 * all mon/osd connections */
1263 ceph_reset_client_addr(fsc
->client
);
1265 ceph_osdc_clear_abort_err(&fsc
->client
->osdc
);
1267 fsc
->blocklisted
= false;
1268 fsc
->mount_state
= CEPH_MOUNT_MOUNTED
;
1271 err
= __ceph_do_getattr(d_inode(sb
->s_root
), NULL
,
1272 CEPH_STAT_CAP_INODE
, true);
1277 static int __init
init_ceph(void)
1279 int ret
= init_caches();
1284 ret
= register_filesystem(&ceph_fs_type
);
1288 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL
);
1298 static void __exit
exit_ceph(void)
1300 dout("exit_ceph\n");
1301 unregister_filesystem(&ceph_fs_type
);
1305 static int param_set_metrics(const char *val
, const struct kernel_param
*kp
)
1307 struct ceph_fs_client
*fsc
;
1310 ret
= param_set_bool(val
, kp
);
1312 pr_err("Failed to parse sending metrics switch value '%s'\n",
1315 } else if (!disable_send_metrics
) {
1316 // wake up all the mds clients
1317 spin_lock(&ceph_fsc_lock
);
1318 list_for_each_entry(fsc
, &ceph_fsc_list
, metric_wakeup
) {
1319 metric_schedule_delayed(&fsc
->mdsc
->metric
);
1321 spin_unlock(&ceph_fsc_lock
);
1327 static const struct kernel_param_ops param_ops_metrics
= {
1328 .set
= param_set_metrics
,
1329 .get
= param_get_bool
,
1332 bool disable_send_metrics
= false;
1333 module_param_cb(disable_send_metrics
, ¶m_ops_metrics
, &disable_send_metrics
, 0644);
1334 MODULE_PARM_DESC(disable_send_metrics
, "Enable sending perf metrics to ceph cluster (default: on)");
1336 module_init(init_ceph
);
1337 module_exit(exit_ceph
);
1339 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1340 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1341 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1342 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1343 MODULE_LICENSE("GPL");