1 #ifndef _FS_CEPH_MDS_CLIENT_H
2 #define _FS_CEPH_MDS_CLIENT_H
4 #include <linux/completion.h>
5 #include <linux/kref.h>
6 #include <linux/list.h>
7 #include <linux/mutex.h>
8 #include <linux/rbtree.h>
9 #include <linux/spinlock.h>
11 #include <linux/ceph/types.h>
12 #include <linux/ceph/messenger.h>
13 #include <linux/ceph/mdsmap.h>
14 #include <linux/ceph/auth.h>
17 * Some lock dependencies:
25 * mdsc->snap_flush_lock
26 * mdsc->cap_delay_lock
30 struct ceph_fs_client
;
34 * parsed info about a single inode. pointers are into the encoded
35 * on-wire structures within the mds reply message payload.
37 struct ceph_mds_reply_info_in
{
38 struct ceph_mds_reply_inode
*in
;
39 struct ceph_dir_layout dir_layout
;
50 * parsed info about an mds reply, including information about
51 * either: 1) the target inode and/or its parent directory and dentry,
52 * and directory contents (for readdir results), or
53 * 2) the file range lock info (for fcntl F_GETLK results).
55 struct ceph_mds_reply_info_parsed
{
56 struct ceph_mds_reply_head
*head
;
59 struct ceph_mds_reply_info_in diri
, targeti
;
60 struct ceph_mds_reply_dirfrag
*dirfrag
;
63 struct ceph_mds_reply_lease
*dlease
;
67 /* for fcntl F_GETLK results */
68 struct ceph_filelock
*filelock_reply
;
70 /* for readdir results */
72 struct ceph_mds_reply_dirfrag
*dir_dir
;
77 struct ceph_mds_reply_lease
**dir_dlease
;
78 struct ceph_mds_reply_info_in
*dir_in
;
79 u8 dir_complete
, dir_end
;
82 /* for create results */
89 /* encoded blob describing snapshot contexts for certain
90 operations (e.g., open) */
97 * cap releases are batched and sent to the MDS en masse.
99 #define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE - \
100 sizeof(struct ceph_mds_cap_release)) / \
101 sizeof(struct ceph_mds_cap_item))
105 * state associated with each MDS<->client session
108 CEPH_MDS_SESSION_NEW
= 1,
109 CEPH_MDS_SESSION_OPENING
= 2,
110 CEPH_MDS_SESSION_OPEN
= 3,
111 CEPH_MDS_SESSION_HUNG
= 4,
112 CEPH_MDS_SESSION_CLOSING
= 5,
113 CEPH_MDS_SESSION_RESTARTING
= 6,
114 CEPH_MDS_SESSION_RECONNECTING
= 7,
117 struct ceph_mds_session
{
118 struct ceph_mds_client
*s_mdsc
;
121 unsigned long s_ttl
; /* time until mds kills us */
122 u64 s_seq
; /* incoming msg seq # */
123 struct mutex s_mutex
; /* serialize session messages */
125 struct ceph_connection s_con
;
127 struct ceph_auth_handshake s_auth
;
129 /* protected by s_gen_ttl_lock */
130 spinlock_t s_gen_ttl_lock
;
131 u32 s_cap_gen
; /* inc each time we get mds stale msg */
132 unsigned long s_cap_ttl
; /* when session caps expire */
134 /* protected by s_cap_lock */
135 spinlock_t s_cap_lock
;
136 struct list_head s_caps
; /* all caps issued by this session */
137 int s_nr_caps
, s_trim_caps
;
138 int s_num_cap_releases
;
140 struct list_head s_cap_releases
; /* waiting cap_release messages */
141 struct list_head s_cap_releases_done
; /* ready to send */
142 struct ceph_cap
*s_cap_iterator
;
144 /* protected by mutex */
145 struct list_head s_cap_flushing
; /* inodes w/ flushing caps */
146 struct list_head s_cap_snaps_flushing
;
147 unsigned long s_renew_requested
; /* last time we sent a renew req */
151 struct list_head s_waiting
; /* waiting requests */
152 struct list_head s_unsafe
; /* unsafe requests */
156 * modes of choosing which MDS to send a request to
161 USE_AUTH_MDS
, /* prefer authoritative mds for this metadata item */
164 struct ceph_mds_request
;
165 struct ceph_mds_client
;
168 * request completion callback
170 typedef void (*ceph_mds_request_callback_t
) (struct ceph_mds_client
*mdsc
,
171 struct ceph_mds_request
*req
);
173 * wait for request completion callback
175 typedef int (*ceph_mds_request_wait_callback_t
) (struct ceph_mds_client
*mdsc
,
176 struct ceph_mds_request
*req
);
179 * an in-flight mds request
181 struct ceph_mds_request
{
182 u64 r_tid
; /* transaction id */
183 struct rb_node r_node
;
184 struct ceph_mds_client
*r_mdsc
;
186 int r_op
; /* mds op code */
188 /* operation on what? */
189 struct inode
*r_inode
; /* arg1 */
190 struct dentry
*r_dentry
; /* arg1 */
191 struct dentry
*r_old_dentry
; /* arg2: rename from or link from */
192 struct inode
*r_old_dentry_dir
; /* arg2: old dentry's parent dir */
193 char *r_path1
, *r_path2
;
194 struct ceph_vino r_ino1
, r_ino2
;
196 struct inode
*r_locked_dir
; /* dir (if any) i_mutex locked by vfs */
197 struct inode
*r_target_inode
; /* resulting inode */
199 struct mutex r_fill_mutex
;
201 union ceph_mds_request_args r_args
;
202 int r_fmode
; /* file mode, if expecting cap */
205 struct timespec r_stamp
;
207 /* for choosing which mds to send this request to */
209 u32 r_direct_hash
; /* choose dir frag based on this dentry hash */
210 bool r_direct_is_hash
; /* true if r_direct_hash is valid */
212 /* data payload is used for xattr ops */
213 struct ceph_pagelist
*r_pagelist
;
215 /* what caps shall we drop? */
216 int r_inode_drop
, r_inode_unless
;
217 int r_dentry_drop
, r_dentry_unless
;
218 int r_old_dentry_drop
, r_old_dentry_unless
;
219 struct inode
*r_old_inode
;
220 int r_old_inode_drop
, r_old_inode_unless
;
222 struct ceph_msg
*r_request
; /* original request */
223 int r_request_release_offset
;
224 struct ceph_msg
*r_reply
;
225 struct ceph_mds_reply_info_parsed r_reply_info
;
229 unsigned long r_timeout
; /* optional. jiffies */
230 unsigned long r_started
; /* start time to measure timeout against */
231 unsigned long r_request_started
; /* start time for mds request only,
232 used to measure lease durations */
234 /* link unsafe requests to parent directory, for fsync */
235 struct inode
*r_unsafe_dir
;
236 struct list_head r_unsafe_dir_item
;
238 struct ceph_mds_session
*r_session
;
240 int r_attempts
; /* resend attempts */
241 int r_num_fwd
; /* number of forward attempts */
242 int r_resend_mds
; /* mds to resend to next, if any*/
243 u32 r_sent_on_mseq
; /* cap mseq request was sent at*/
246 struct list_head r_wait
;
247 struct completion r_completion
;
248 struct completion r_safe_completion
;
249 ceph_mds_request_callback_t r_callback
;
250 ceph_mds_request_wait_callback_t r_wait_for_completion
;
251 struct list_head r_unsafe_item
; /* per-session unsafe list item */
252 bool r_got_unsafe
, r_got_safe
, r_got_result
;
254 bool r_did_prepopulate
;
255 u32 r_readdir_offset
;
257 struct ceph_cap_reservation r_caps_reservation
;
264 struct ceph_mds_client
{
265 struct ceph_fs_client
*fsc
;
266 struct mutex mutex
; /* all nested structures */
268 struct ceph_mdsmap
*mdsmap
;
269 struct completion safe_umount_waiters
;
270 wait_queue_head_t session_close_wq
;
271 struct list_head waiting_for_map
;
273 struct ceph_mds_session
**sessions
; /* NULL for mds if no session */
274 int max_sessions
; /* len of s_mds_sessions */
275 int stopping
; /* true if shutting down */
278 * snap_rwsem will cover cap linkage into snaprealms, and
279 * realm snap contexts. (later, we can do per-realm snap
280 * contexts locks..) the empty list contains realms with no
281 * references (implying they contain no inodes with caps) that
282 * should be destroyed.
284 struct rw_semaphore snap_rwsem
;
285 struct rb_root snap_realms
;
286 struct list_head snap_empty
;
287 spinlock_t snap_empty_lock
; /* protect snap_empty */
289 u64 last_tid
; /* most recent mds request */
290 struct rb_root request_tree
; /* pending mds requests */
291 struct delayed_work delayed_work
; /* delayed work */
292 unsigned long last_renew_caps
; /* last time we renewed our caps */
293 struct list_head cap_delay_list
; /* caps with delayed release */
294 spinlock_t cap_delay_lock
; /* protects cap_delay_list */
295 struct list_head snap_flush_list
; /* cap_snaps ready to flush */
296 spinlock_t snap_flush_lock
;
299 struct list_head cap_dirty
; /* inodes with dirty caps */
300 struct list_head cap_dirty_migrating
; /* ...that are migration... */
301 int num_cap_flushing
; /* # caps we are flushing */
302 spinlock_t cap_dirty_lock
; /* protects above items */
303 wait_queue_head_t cap_flushing_wq
;
308 * Maintain a global pool of preallocated struct ceph_caps, referenced
309 * by struct ceph_caps_reservations. This ensures that we preallocate
310 * memory needed to successfully process an MDS response. (If an MDS
311 * sends us cap information and we fail to process it, we will have
312 * problems due to the client and MDS being out of sync.)
314 * Reservations are 'owned' by a ceph_cap_reservation context.
316 spinlock_t caps_list_lock
;
317 struct list_head caps_list
; /* unused (reserved or
319 int caps_total_count
; /* total caps allocated */
320 int caps_use_count
; /* in use */
321 int caps_reserve_count
; /* unused, reserved */
322 int caps_avail_count
; /* unused, unreserved */
323 int caps_min_count
; /* keep at least this many
325 spinlock_t dentry_lru_lock
;
326 struct list_head dentry_lru
;
330 extern const char *ceph_mds_op_name(int op
);
332 extern struct ceph_mds_session
*
333 __ceph_lookup_mds_session(struct ceph_mds_client
*, int mds
);
335 static inline struct ceph_mds_session
*
336 ceph_get_mds_session(struct ceph_mds_session
*s
)
338 atomic_inc(&s
->s_ref
);
342 extern const char *ceph_session_state_name(int s
);
344 extern void ceph_put_mds_session(struct ceph_mds_session
*s
);
346 extern int ceph_send_msg_mds(struct ceph_mds_client
*mdsc
,
347 struct ceph_msg
*msg
, int mds
);
349 extern int ceph_mdsc_init(struct ceph_fs_client
*fsc
);
350 extern void ceph_mdsc_close_sessions(struct ceph_mds_client
*mdsc
);
351 extern void ceph_mdsc_destroy(struct ceph_fs_client
*fsc
);
353 extern void ceph_mdsc_sync(struct ceph_mds_client
*mdsc
);
355 extern void ceph_mdsc_lease_release(struct ceph_mds_client
*mdsc
,
359 extern void ceph_invalidate_dir_request(struct ceph_mds_request
*req
);
360 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request
*req
,
362 extern struct ceph_mds_request
*
363 ceph_mdsc_create_request(struct ceph_mds_client
*mdsc
, int op
, int mode
);
364 extern void ceph_mdsc_submit_request(struct ceph_mds_client
*mdsc
,
365 struct ceph_mds_request
*req
);
366 extern int ceph_mdsc_do_request(struct ceph_mds_client
*mdsc
,
368 struct ceph_mds_request
*req
);
369 static inline void ceph_mdsc_get_request(struct ceph_mds_request
*req
)
371 kref_get(&req
->r_kref
);
373 extern void ceph_mdsc_release_request(struct kref
*kref
);
374 static inline void ceph_mdsc_put_request(struct ceph_mds_request
*req
)
376 kref_put(&req
->r_kref
, ceph_mdsc_release_request
);
379 extern int ceph_add_cap_releases(struct ceph_mds_client
*mdsc
,
380 struct ceph_mds_session
*session
);
381 extern void ceph_send_cap_releases(struct ceph_mds_client
*mdsc
,
382 struct ceph_mds_session
*session
);
384 extern void ceph_mdsc_pre_umount(struct ceph_mds_client
*mdsc
);
386 extern char *ceph_mdsc_build_path(struct dentry
*dentry
, int *plen
, u64
*base
,
389 extern void __ceph_mdsc_drop_dentry_lease(struct dentry
*dentry
);
390 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session
*session
,
392 struct dentry
*dentry
, char action
,
395 extern void ceph_mdsc_handle_map(struct ceph_mds_client
*mdsc
,
396 struct ceph_msg
*msg
);
398 extern struct ceph_mds_session
*
399 ceph_mdsc_open_export_target_session(struct ceph_mds_client
*mdsc
, int target
);
400 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client
*mdsc
,
401 struct ceph_mds_session
*session
);