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;
90 struct ceph_mds_reply_dir_entry *dir_entries;
93 /* for create results */
100 /* encoded blob describing snapshot contexts for certain
101 operations (e.g., open) */
108 * cap releases are batched and sent to the MDS en masse.
110 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - \
111 sizeof(struct ceph_mds_cap_release)) / \
112 sizeof(struct ceph_mds_cap_item))
116 * state associated with each MDS<->client session
119 CEPH_MDS_SESSION_NEW = 1,
120 CEPH_MDS_SESSION_OPENING = 2,
121 CEPH_MDS_SESSION_OPEN = 3,
122 CEPH_MDS_SESSION_HUNG = 4,
123 CEPH_MDS_SESSION_CLOSING = 5,
124 CEPH_MDS_SESSION_RESTARTING = 6,
125 CEPH_MDS_SESSION_RECONNECTING = 7,
126 CEPH_MDS_SESSION_REJECTED = 8,
129 struct ceph_mds_session {
130 struct ceph_mds_client *s_mdsc;
133 unsigned long s_ttl; /* time until mds kills us */
134 u64 s_seq; /* incoming msg seq # */
135 struct mutex s_mutex; /* serialize session messages */
137 struct ceph_connection s_con;
139 struct ceph_auth_handshake s_auth;
141 /* protected by s_gen_ttl_lock */
142 spinlock_t s_gen_ttl_lock;
143 u32 s_cap_gen; /* inc each time we get mds stale msg */
144 unsigned long s_cap_ttl; /* when session caps expire */
146 /* protected by s_cap_lock */
147 spinlock_t s_cap_lock;
148 struct list_head s_caps; /* all caps issued by this session */
149 int s_nr_caps, s_trim_caps;
150 int s_num_cap_releases;
153 struct list_head s_cap_releases; /* waiting cap_release messages */
154 struct ceph_cap *s_cap_iterator;
156 /* protected by mutex */
157 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
158 unsigned long s_renew_requested; /* last time we sent a renew req */
162 struct list_head s_waiting; /* waiting requests */
163 struct list_head s_unsafe; /* unsafe requests */
167 * modes of choosing which MDS to send a request to
172 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
175 struct ceph_mds_request;
176 struct ceph_mds_client;
179 * request completion callback
181 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
182 struct ceph_mds_request *req);
184 * wait for request completion callback
186 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
187 struct ceph_mds_request *req);
190 * an in-flight mds request
192 struct ceph_mds_request {
193 u64 r_tid; /* transaction id */
194 struct rb_node r_node;
195 struct ceph_mds_client *r_mdsc;
197 int r_op; /* mds op code */
199 /* operation on what? */
200 struct inode *r_inode; /* arg1 */
201 struct dentry *r_dentry; /* arg1 */
202 struct dentry *r_old_dentry; /* arg2: rename from or link from */
203 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
204 char *r_path1, *r_path2;
205 struct ceph_vino r_ino1, r_ino2;
207 struct inode *r_parent; /* parent dir inode */
208 struct inode *r_target_inode; /* resulting inode */
210 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
211 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
212 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
213 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
214 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
215 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
216 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
217 unsigned long r_req_flags;
219 struct mutex r_fill_mutex;
221 union ceph_mds_request_args r_args;
222 int r_fmode; /* file mode, if expecting cap */
225 struct timespec r_stamp;
227 /* for choosing which mds to send this request to */
229 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
231 /* data payload is used for xattr ops */
232 struct ceph_pagelist *r_pagelist;
234 /* what caps shall we drop? */
235 int r_inode_drop, r_inode_unless;
236 int r_dentry_drop, r_dentry_unless;
237 int r_old_dentry_drop, r_old_dentry_unless;
238 struct inode *r_old_inode;
239 int r_old_inode_drop, r_old_inode_unless;
241 struct ceph_msg *r_request; /* original request */
242 int r_request_release_offset;
243 struct ceph_msg *r_reply;
244 struct ceph_mds_reply_info_parsed r_reply_info;
245 struct page *r_locked_page;
248 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
249 unsigned long r_started; /* start time to measure timeout against */
250 unsigned long r_request_started; /* start time for mds request only,
251 used to measure lease durations */
253 /* link unsafe requests to parent directory, for fsync */
254 struct inode *r_unsafe_dir;
255 struct list_head r_unsafe_dir_item;
257 /* unsafe requests that modify the target inode */
258 struct list_head r_unsafe_target_item;
260 struct ceph_mds_session *r_session;
262 int r_attempts; /* resend attempts */
263 int r_num_fwd; /* number of forward attempts */
264 int r_resend_mds; /* mds to resend to next, if any*/
265 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
268 struct list_head r_wait;
269 struct completion r_completion;
270 struct completion r_safe_completion;
271 ceph_mds_request_callback_t r_callback;
272 ceph_mds_request_wait_callback_t r_wait_for_completion;
273 struct list_head r_unsafe_item; /* per-session unsafe list item */
275 long long r_dir_release_cnt;
276 long long r_dir_ordered_cnt;
277 int r_readdir_cache_idx;
278 u32 r_readdir_offset;
280 struct ceph_cap_reservation r_caps_reservation;
284 struct ceph_pool_perm {
295 struct ceph_mds_client {
296 struct ceph_fs_client *fsc;
297 struct mutex mutex; /* all nested structures */
299 struct ceph_mdsmap *mdsmap;
300 struct completion safe_umount_waiters;
301 wait_queue_head_t session_close_wq;
302 struct list_head waiting_for_map;
305 struct ceph_mds_session **sessions; /* NULL for mds if no session */
306 atomic_t num_sessions;
307 int max_sessions; /* len of s_mds_sessions */
308 int stopping; /* true if shutting down */
311 * snap_rwsem will cover cap linkage into snaprealms, and
312 * realm snap contexts. (later, we can do per-realm snap
313 * contexts locks..) the empty list contains realms with no
314 * references (implying they contain no inodes with caps) that
315 * should be destroyed.
318 struct rw_semaphore snap_rwsem;
319 struct rb_root snap_realms;
320 struct list_head snap_empty;
321 spinlock_t snap_empty_lock; /* protect snap_empty */
323 u64 last_tid; /* most recent mds request */
324 u64 oldest_tid; /* oldest incomplete mds request,
325 excluding setfilelock requests */
326 struct rb_root request_tree; /* pending mds requests */
327 struct delayed_work delayed_work; /* delayed work */
328 unsigned long last_renew_caps; /* last time we renewed our caps */
329 struct list_head cap_delay_list; /* caps with delayed release */
330 spinlock_t cap_delay_lock; /* protects cap_delay_list */
331 struct list_head snap_flush_list; /* cap_snaps ready to flush */
332 spinlock_t snap_flush_lock;
334 u64 last_cap_flush_tid;
335 struct list_head cap_flush_list;
336 struct list_head cap_dirty; /* inodes with dirty caps */
337 struct list_head cap_dirty_migrating; /* ...that are migration... */
338 int num_cap_flushing; /* # caps we are flushing */
339 spinlock_t cap_dirty_lock; /* protects above items */
340 wait_queue_head_t cap_flushing_wq;
345 * Maintain a global pool of preallocated struct ceph_caps, referenced
346 * by struct ceph_caps_reservations. This ensures that we preallocate
347 * memory needed to successfully process an MDS response. (If an MDS
348 * sends us cap information and we fail to process it, we will have
349 * problems due to the client and MDS being out of sync.)
351 * Reservations are 'owned' by a ceph_cap_reservation context.
353 spinlock_t caps_list_lock;
354 struct list_head caps_list; /* unused (reserved or
356 int caps_total_count; /* total caps allocated */
357 int caps_use_count; /* in use */
358 int caps_reserve_count; /* unused, reserved */
359 int caps_avail_count; /* unused, unreserved */
360 int caps_min_count; /* keep at least this many
362 spinlock_t dentry_lru_lock;
363 struct list_head dentry_lru;
366 struct rw_semaphore pool_perm_rwsem;
367 struct rb_root pool_perm_tree;
370 extern const char *ceph_mds_op_name(int op);
372 extern struct ceph_mds_session *
373 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
375 static inline struct ceph_mds_session *
376 ceph_get_mds_session(struct ceph_mds_session *s)
378 refcount_inc(&s->s_ref);
382 extern const char *ceph_session_state_name(int s);
384 extern void ceph_put_mds_session(struct ceph_mds_session *s);
386 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
387 struct ceph_msg *msg, int mds);
389 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
390 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
391 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
392 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
394 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
396 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
397 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
399 extern struct ceph_mds_request *
400 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
401 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
402 struct ceph_mds_request *req);
403 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
405 struct ceph_mds_request *req);
406 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
408 kref_get(&req->r_kref);
410 extern void ceph_mdsc_release_request(struct kref *kref);
411 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
413 kref_put(&req->r_kref, ceph_mdsc_release_request);
416 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
417 struct ceph_mds_session *session);
419 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
421 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
424 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
425 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
427 struct dentry *dentry, char action,
430 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
431 struct ceph_msg *msg);
432 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
433 struct ceph_msg *msg);
435 extern struct ceph_mds_session *
436 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
437 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
438 struct ceph_mds_session *session);