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>
10 #include <linux/refcount.h>
12 #include <linux/ceph/types.h>
13 #include <linux/ceph/messenger.h>
14 #include <linux/ceph/mdsmap.h>
15 #include <linux/ceph/auth.h>
18 * Some lock dependencies:
26 * mdsc->snap_flush_lock
27 * mdsc->cap_delay_lock
31 struct ceph_fs_client;
35 * parsed info about a single inode. pointers are into the encoded
36 * on-wire structures within the mds reply message payload.
38 struct ceph_mds_reply_info_in {
39 struct ceph_mds_reply_inode *in;
40 struct ceph_dir_layout dir_layout;
52 struct ceph_mds_reply_dir_entry {
55 struct ceph_mds_reply_lease *lease;
56 struct ceph_mds_reply_info_in inode;
61 * parsed info about an mds reply, including information about
62 * either: 1) the target inode and/or its parent directory and dentry,
63 * and directory contents (for readdir results), or
64 * 2) the file range lock info (for fcntl F_GETLK results).
66 struct ceph_mds_reply_info_parsed {
67 struct ceph_mds_reply_head *head;
70 struct ceph_mds_reply_info_in diri, targeti;
71 struct ceph_mds_reply_dirfrag *dirfrag;
74 struct ceph_mds_reply_lease *dlease;
78 /* for fcntl F_GETLK results */
79 struct ceph_filelock *filelock_reply;
81 /* for readdir results */
83 struct ceph_mds_reply_dirfrag *dir_dir;
89 struct ceph_mds_reply_dir_entry *dir_entries;
92 /* for create results */
99 /* encoded blob describing snapshot contexts for certain
100 operations (e.g., open) */
107 * cap releases are batched and sent to the MDS en masse.
109 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - \
110 sizeof(struct ceph_mds_cap_release)) / \
111 sizeof(struct ceph_mds_cap_item))
115 * state associated with each MDS<->client session
118 CEPH_MDS_SESSION_NEW = 1,
119 CEPH_MDS_SESSION_OPENING = 2,
120 CEPH_MDS_SESSION_OPEN = 3,
121 CEPH_MDS_SESSION_HUNG = 4,
122 CEPH_MDS_SESSION_CLOSING = 5,
123 CEPH_MDS_SESSION_RESTARTING = 6,
124 CEPH_MDS_SESSION_RECONNECTING = 7,
125 CEPH_MDS_SESSION_REJECTED = 8,
128 struct ceph_mds_session {
129 struct ceph_mds_client *s_mdsc;
132 unsigned long s_ttl; /* time until mds kills us */
133 u64 s_seq; /* incoming msg seq # */
134 struct mutex s_mutex; /* serialize session messages */
136 struct ceph_connection s_con;
138 struct ceph_auth_handshake s_auth;
140 /* protected by s_gen_ttl_lock */
141 spinlock_t s_gen_ttl_lock;
142 u32 s_cap_gen; /* inc each time we get mds stale msg */
143 unsigned long s_cap_ttl; /* when session caps expire */
145 /* protected by s_cap_lock */
146 spinlock_t s_cap_lock;
147 struct list_head s_caps; /* all caps issued by this session */
148 int s_nr_caps, s_trim_caps;
149 int s_num_cap_releases;
152 struct list_head s_cap_releases; /* waiting cap_release messages */
153 struct ceph_cap *s_cap_iterator;
155 /* protected by mutex */
156 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
157 unsigned long s_renew_requested; /* last time we sent a renew req */
161 struct list_head s_waiting; /* waiting requests */
162 struct list_head s_unsafe; /* unsafe requests */
166 * modes of choosing which MDS to send a request to
171 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
174 struct ceph_mds_request;
175 struct ceph_mds_client;
178 * request completion callback
180 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
181 struct ceph_mds_request *req);
183 * wait for request completion callback
185 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
186 struct ceph_mds_request *req);
189 * an in-flight mds request
191 struct ceph_mds_request {
192 u64 r_tid; /* transaction id */
193 struct rb_node r_node;
194 struct ceph_mds_client *r_mdsc;
196 int r_op; /* mds op code */
198 /* operation on what? */
199 struct inode *r_inode; /* arg1 */
200 struct dentry *r_dentry; /* arg1 */
201 struct dentry *r_old_dentry; /* arg2: rename from or link from */
202 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
203 char *r_path1, *r_path2;
204 struct ceph_vino r_ino1, r_ino2;
206 struct inode *r_parent; /* parent dir inode */
207 struct inode *r_target_inode; /* resulting inode */
209 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
210 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
211 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
212 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
213 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
214 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
215 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
216 unsigned long r_req_flags;
218 struct mutex r_fill_mutex;
220 union ceph_mds_request_args r_args;
221 int r_fmode; /* file mode, if expecting cap */
224 struct timespec r_stamp;
226 /* for choosing which mds to send this request to */
228 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
230 /* data payload is used for xattr ops */
231 struct ceph_pagelist *r_pagelist;
233 /* what caps shall we drop? */
234 int r_inode_drop, r_inode_unless;
235 int r_dentry_drop, r_dentry_unless;
236 int r_old_dentry_drop, r_old_dentry_unless;
237 struct inode *r_old_inode;
238 int r_old_inode_drop, r_old_inode_unless;
240 struct ceph_msg *r_request; /* original request */
241 int r_request_release_offset;
242 struct ceph_msg *r_reply;
243 struct ceph_mds_reply_info_parsed r_reply_info;
244 struct page *r_locked_page;
247 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
248 unsigned long r_started; /* start time to measure timeout against */
249 unsigned long r_request_started; /* start time for mds request only,
250 used to measure lease durations */
252 /* link unsafe requests to parent directory, for fsync */
253 struct inode *r_unsafe_dir;
254 struct list_head r_unsafe_dir_item;
256 /* unsafe requests that modify the target inode */
257 struct list_head r_unsafe_target_item;
259 struct ceph_mds_session *r_session;
261 int r_attempts; /* resend attempts */
262 int r_num_fwd; /* number of forward attempts */
263 int r_resend_mds; /* mds to resend to next, if any*/
264 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
267 struct list_head r_wait;
268 struct completion r_completion;
269 struct completion r_safe_completion;
270 ceph_mds_request_callback_t r_callback;
271 ceph_mds_request_wait_callback_t r_wait_for_completion;
272 struct list_head r_unsafe_item; /* per-session unsafe list item */
274 long long r_dir_release_cnt;
275 long long r_dir_ordered_cnt;
276 int r_readdir_cache_idx;
277 u32 r_readdir_offset;
279 struct ceph_cap_reservation r_caps_reservation;
283 struct ceph_pool_perm {
294 struct ceph_mds_client {
295 struct ceph_fs_client *fsc;
296 struct mutex mutex; /* all nested structures */
298 struct ceph_mdsmap *mdsmap;
299 struct completion safe_umount_waiters;
300 wait_queue_head_t session_close_wq;
301 struct list_head waiting_for_map;
304 struct ceph_mds_session **sessions; /* NULL for mds if no session */
305 atomic_t num_sessions;
306 int max_sessions; /* len of s_mds_sessions */
307 int stopping; /* true if shutting down */
310 * snap_rwsem will cover cap linkage into snaprealms, and
311 * realm snap contexts. (later, we can do per-realm snap
312 * contexts locks..) the empty list contains realms with no
313 * references (implying they contain no inodes with caps) that
314 * should be destroyed.
317 struct rw_semaphore snap_rwsem;
318 struct rb_root snap_realms;
319 struct list_head snap_empty;
320 spinlock_t snap_empty_lock; /* protect snap_empty */
322 u64 last_tid; /* most recent mds request */
323 u64 oldest_tid; /* oldest incomplete mds request,
324 excluding setfilelock requests */
325 struct rb_root request_tree; /* pending mds requests */
326 struct delayed_work delayed_work; /* delayed work */
327 unsigned long last_renew_caps; /* last time we renewed our caps */
328 struct list_head cap_delay_list; /* caps with delayed release */
329 spinlock_t cap_delay_lock; /* protects cap_delay_list */
330 struct list_head snap_flush_list; /* cap_snaps ready to flush */
331 spinlock_t snap_flush_lock;
333 u64 last_cap_flush_tid;
334 struct list_head cap_flush_list;
335 struct list_head cap_dirty; /* inodes with dirty caps */
336 struct list_head cap_dirty_migrating; /* ...that are migration... */
337 int num_cap_flushing; /* # caps we are flushing */
338 spinlock_t cap_dirty_lock; /* protects above items */
339 wait_queue_head_t cap_flushing_wq;
344 * Maintain a global pool of preallocated struct ceph_caps, referenced
345 * by struct ceph_caps_reservations. This ensures that we preallocate
346 * memory needed to successfully process an MDS response. (If an MDS
347 * sends us cap information and we fail to process it, we will have
348 * problems due to the client and MDS being out of sync.)
350 * Reservations are 'owned' by a ceph_cap_reservation context.
352 spinlock_t caps_list_lock;
353 struct list_head caps_list; /* unused (reserved or
355 int caps_total_count; /* total caps allocated */
356 int caps_use_count; /* in use */
357 int caps_reserve_count; /* unused, reserved */
358 int caps_avail_count; /* unused, unreserved */
359 int caps_min_count; /* keep at least this many
361 spinlock_t dentry_lru_lock;
362 struct list_head dentry_lru;
365 struct rw_semaphore pool_perm_rwsem;
366 struct rb_root pool_perm_tree;
369 extern const char *ceph_mds_op_name(int op);
371 extern struct ceph_mds_session *
372 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
374 static inline struct ceph_mds_session *
375 ceph_get_mds_session(struct ceph_mds_session *s)
377 refcount_inc(&s->s_ref);
381 extern const char *ceph_session_state_name(int s);
383 extern void ceph_put_mds_session(struct ceph_mds_session *s);
385 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
386 struct ceph_msg *msg, int mds);
388 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
389 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
390 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
391 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
393 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
395 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
396 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
398 extern struct ceph_mds_request *
399 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
400 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
401 struct ceph_mds_request *req);
402 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
404 struct ceph_mds_request *req);
405 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
407 kref_get(&req->r_kref);
409 extern void ceph_mdsc_release_request(struct kref *kref);
410 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
412 kref_put(&req->r_kref, ceph_mdsc_release_request);
415 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
416 struct ceph_mds_session *session);
418 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
420 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
423 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
424 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
426 struct dentry *dentry, char action,
429 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
430 struct ceph_msg *msg);
431 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
432 struct ceph_msg *msg);
434 extern struct ceph_mds_session *
435 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
436 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
437 struct ceph_mds_session *session);