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;
51 * parsed info about an mds reply, including information about
52 * either: 1) the target inode and/or its parent directory and dentry,
53 * and directory contents (for readdir results), or
54 * 2) the file range lock info (for fcntl F_GETLK results).
56 struct ceph_mds_reply_info_parsed {
57 struct ceph_mds_reply_head *head;
60 struct ceph_mds_reply_info_in diri, targeti;
61 struct ceph_mds_reply_dirfrag *dirfrag;
64 struct ceph_mds_reply_lease *dlease;
68 /* for fcntl F_GETLK results */
69 struct ceph_filelock *filelock_reply;
71 /* for readdir results */
73 struct ceph_mds_reply_dirfrag *dir_dir;
78 struct ceph_mds_reply_lease **dir_dlease;
79 struct ceph_mds_reply_info_in *dir_in;
80 u8 dir_complete, dir_end;
83 /* for create results */
90 /* encoded blob describing snapshot contexts for certain
91 operations (e.g., open) */
98 * cap releases are batched and sent to the MDS en masse.
100 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - \
101 sizeof(struct ceph_mds_cap_release)) / \
102 sizeof(struct ceph_mds_cap_item))
106 * state associated with each MDS<->client session
109 CEPH_MDS_SESSION_NEW = 1,
110 CEPH_MDS_SESSION_OPENING = 2,
111 CEPH_MDS_SESSION_OPEN = 3,
112 CEPH_MDS_SESSION_HUNG = 4,
113 CEPH_MDS_SESSION_CLOSING = 5,
114 CEPH_MDS_SESSION_RESTARTING = 6,
115 CEPH_MDS_SESSION_RECONNECTING = 7,
118 struct ceph_mds_session {
119 struct ceph_mds_client *s_mdsc;
122 unsigned long s_ttl; /* time until mds kills us */
123 u64 s_seq; /* incoming msg seq # */
124 struct mutex s_mutex; /* serialize session messages */
126 struct ceph_connection s_con;
128 struct ceph_auth_handshake s_auth;
130 /* protected by s_gen_ttl_lock */
131 spinlock_t s_gen_ttl_lock;
132 u32 s_cap_gen; /* inc each time we get mds stale msg */
133 unsigned long s_cap_ttl; /* when session caps expire */
135 /* protected by s_cap_lock */
136 spinlock_t s_cap_lock;
137 struct list_head s_caps; /* all caps issued by this session */
138 int s_nr_caps, s_trim_caps;
139 int s_num_cap_releases;
142 struct list_head s_cap_releases; /* waiting cap_release messages */
143 struct ceph_cap *s_cap_iterator;
145 /* protected by mutex */
146 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
147 struct list_head s_cap_snaps_flushing;
148 unsigned long s_renew_requested; /* last time we sent a renew req */
152 struct list_head s_waiting; /* waiting requests */
153 struct list_head s_unsafe; /* unsafe requests */
157 * modes of choosing which MDS to send a request to
162 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
165 struct ceph_mds_request;
166 struct ceph_mds_client;
169 * request completion callback
171 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
172 struct ceph_mds_request *req);
174 * wait for request completion callback
176 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
177 struct ceph_mds_request *req);
180 * an in-flight mds request
182 struct ceph_mds_request {
183 u64 r_tid; /* transaction id */
184 struct rb_node r_node;
185 struct ceph_mds_client *r_mdsc;
187 int r_op; /* mds op code */
189 /* operation on what? */
190 struct inode *r_inode; /* arg1 */
191 struct dentry *r_dentry; /* arg1 */
192 struct dentry *r_old_dentry; /* arg2: rename from or link from */
193 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
194 char *r_path1, *r_path2;
195 struct ceph_vino r_ino1, r_ino2;
197 struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
198 struct inode *r_target_inode; /* resulting inode */
200 struct mutex r_fill_mutex;
202 union ceph_mds_request_args r_args;
203 int r_fmode; /* file mode, if expecting cap */
206 struct timespec r_stamp;
208 /* for choosing which mds to send this request to */
210 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
211 bool r_direct_is_hash; /* true if r_direct_hash is valid */
213 /* data payload is used for xattr ops */
214 struct ceph_pagelist *r_pagelist;
216 /* what caps shall we drop? */
217 int r_inode_drop, r_inode_unless;
218 int r_dentry_drop, r_dentry_unless;
219 int r_old_dentry_drop, r_old_dentry_unless;
220 struct inode *r_old_inode;
221 int r_old_inode_drop, r_old_inode_unless;
223 struct ceph_msg *r_request; /* original request */
224 int r_request_release_offset;
225 struct ceph_msg *r_reply;
226 struct ceph_mds_reply_info_parsed r_reply_info;
227 struct page *r_locked_page;
231 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
232 unsigned long r_started; /* start time to measure timeout against */
233 unsigned long r_request_started; /* start time for mds request only,
234 used to measure lease durations */
236 /* link unsafe requests to parent directory, for fsync */
237 struct inode *r_unsafe_dir;
238 struct list_head r_unsafe_dir_item;
240 /* unsafe requests that modify the target inode */
241 struct list_head r_unsafe_target_item;
243 struct ceph_mds_session *r_session;
245 int r_attempts; /* resend attempts */
246 int r_num_fwd; /* number of forward attempts */
247 int r_resend_mds; /* mds to resend to next, if any*/
248 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
251 struct list_head r_wait;
252 struct completion r_completion;
253 struct completion r_safe_completion;
254 ceph_mds_request_callback_t r_callback;
255 ceph_mds_request_wait_callback_t r_wait_for_completion;
256 struct list_head r_unsafe_item; /* per-session unsafe list item */
257 bool r_got_unsafe, r_got_safe, r_got_result;
259 bool r_did_prepopulate;
260 long long r_dir_release_cnt;
261 long long r_dir_ordered_cnt;
262 int r_readdir_cache_idx;
263 u32 r_readdir_offset;
265 struct ceph_cap_reservation r_caps_reservation;
269 struct ceph_pool_perm {
278 struct ceph_mds_client {
279 struct ceph_fs_client *fsc;
280 struct mutex mutex; /* all nested structures */
282 struct ceph_mdsmap *mdsmap;
283 struct completion safe_umount_waiters;
284 wait_queue_head_t session_close_wq;
285 struct list_head waiting_for_map;
287 struct ceph_mds_session **sessions; /* NULL for mds if no session */
288 atomic_t num_sessions;
289 int max_sessions; /* len of s_mds_sessions */
290 int stopping; /* true if shutting down */
293 * snap_rwsem will cover cap linkage into snaprealms, and
294 * realm snap contexts. (later, we can do per-realm snap
295 * contexts locks..) the empty list contains realms with no
296 * references (implying they contain no inodes with caps) that
297 * should be destroyed.
300 struct rw_semaphore snap_rwsem;
301 struct rb_root snap_realms;
302 struct list_head snap_empty;
303 spinlock_t snap_empty_lock; /* protect snap_empty */
305 u64 last_tid; /* most recent mds request */
306 u64 oldest_tid; /* oldest incomplete mds request,
307 excluding setfilelock requests */
308 struct rb_root request_tree; /* pending mds requests */
309 struct delayed_work delayed_work; /* delayed work */
310 unsigned long last_renew_caps; /* last time we renewed our caps */
311 struct list_head cap_delay_list; /* caps with delayed release */
312 spinlock_t cap_delay_lock; /* protects cap_delay_list */
313 struct list_head snap_flush_list; /* cap_snaps ready to flush */
314 spinlock_t snap_flush_lock;
316 u64 last_cap_flush_tid;
317 struct rb_root cap_flush_tree;
318 struct list_head cap_dirty; /* inodes with dirty caps */
319 struct list_head cap_dirty_migrating; /* ...that are migration... */
320 int num_cap_flushing; /* # caps we are flushing */
321 spinlock_t cap_dirty_lock; /* protects above items */
322 wait_queue_head_t cap_flushing_wq;
327 * Maintain a global pool of preallocated struct ceph_caps, referenced
328 * by struct ceph_caps_reservations. This ensures that we preallocate
329 * memory needed to successfully process an MDS response. (If an MDS
330 * sends us cap information and we fail to process it, we will have
331 * problems due to the client and MDS being out of sync.)
333 * Reservations are 'owned' by a ceph_cap_reservation context.
335 spinlock_t caps_list_lock;
336 struct list_head caps_list; /* unused (reserved or
338 int caps_total_count; /* total caps allocated */
339 int caps_use_count; /* in use */
340 int caps_reserve_count; /* unused, reserved */
341 int caps_avail_count; /* unused, unreserved */
342 int caps_min_count; /* keep at least this many
344 spinlock_t dentry_lru_lock;
345 struct list_head dentry_lru;
348 struct rw_semaphore pool_perm_rwsem;
349 struct rb_root pool_perm_tree;
352 extern const char *ceph_mds_op_name(int op);
354 extern struct ceph_mds_session *
355 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
357 static inline struct ceph_mds_session *
358 ceph_get_mds_session(struct ceph_mds_session *s)
360 atomic_inc(&s->s_ref);
364 extern const char *ceph_session_state_name(int s);
366 extern void ceph_put_mds_session(struct ceph_mds_session *s);
368 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
369 struct ceph_msg *msg, int mds);
371 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
372 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
373 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
374 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
376 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
378 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
382 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
383 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
385 extern struct ceph_mds_request *
386 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
387 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
388 struct ceph_mds_request *req);
389 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
391 struct ceph_mds_request *req);
392 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
394 kref_get(&req->r_kref);
396 extern void ceph_mdsc_release_request(struct kref *kref);
397 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
399 kref_put(&req->r_kref, ceph_mdsc_release_request);
402 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
403 struct ceph_mds_session *session);
405 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
407 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
410 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
411 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
413 struct dentry *dentry, char action,
416 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
417 struct ceph_msg *msg);
419 extern struct ceph_mds_session *
420 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
421 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
422 struct ceph_mds_session *session);