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 struct ceph_mds_reply_dir_entry {
53 struct ceph_mds_reply_lease *lease;
54 struct ceph_mds_reply_info_in inode;
59 * parsed info about an mds reply, including information about
60 * either: 1) the target inode and/or its parent directory and dentry,
61 * and directory contents (for readdir results), or
62 * 2) the file range lock info (for fcntl F_GETLK results).
64 struct ceph_mds_reply_info_parsed {
65 struct ceph_mds_reply_head *head;
68 struct ceph_mds_reply_info_in diri, targeti;
69 struct ceph_mds_reply_dirfrag *dirfrag;
72 struct ceph_mds_reply_lease *dlease;
76 /* for fcntl F_GETLK results */
77 struct ceph_filelock *filelock_reply;
79 /* for readdir results */
81 struct ceph_mds_reply_dirfrag *dir_dir;
84 bool dir_complete, dir_end;
85 struct ceph_mds_reply_dir_entry *dir_entries;
88 /* for create results */
95 /* encoded blob describing snapshot contexts for certain
96 operations (e.g., open) */
103 * cap releases are batched and sent to the MDS en masse.
105 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - \
106 sizeof(struct ceph_mds_cap_release)) / \
107 sizeof(struct ceph_mds_cap_item))
111 * state associated with each MDS<->client session
114 CEPH_MDS_SESSION_NEW = 1,
115 CEPH_MDS_SESSION_OPENING = 2,
116 CEPH_MDS_SESSION_OPEN = 3,
117 CEPH_MDS_SESSION_HUNG = 4,
118 CEPH_MDS_SESSION_CLOSING = 5,
119 CEPH_MDS_SESSION_RESTARTING = 6,
120 CEPH_MDS_SESSION_RECONNECTING = 7,
123 struct ceph_mds_session {
124 struct ceph_mds_client *s_mdsc;
127 unsigned long s_ttl; /* time until mds kills us */
128 u64 s_seq; /* incoming msg seq # */
129 struct mutex s_mutex; /* serialize session messages */
131 struct ceph_connection s_con;
133 struct ceph_auth_handshake s_auth;
135 /* protected by s_gen_ttl_lock */
136 spinlock_t s_gen_ttl_lock;
137 u32 s_cap_gen; /* inc each time we get mds stale msg */
138 unsigned long s_cap_ttl; /* when session caps expire */
140 /* protected by s_cap_lock */
141 spinlock_t s_cap_lock;
142 struct list_head s_caps; /* all caps issued by this session */
143 int s_nr_caps, s_trim_caps;
144 int s_num_cap_releases;
147 struct list_head s_cap_releases; /* waiting cap_release messages */
148 struct ceph_cap *s_cap_iterator;
150 /* protected by mutex */
151 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
152 struct list_head s_cap_snaps_flushing;
153 unsigned long s_renew_requested; /* last time we sent a renew req */
157 struct list_head s_waiting; /* waiting requests */
158 struct list_head s_unsafe; /* unsafe requests */
162 * modes of choosing which MDS to send a request to
167 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
170 struct ceph_mds_request;
171 struct ceph_mds_client;
174 * request completion callback
176 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
177 struct ceph_mds_request *req);
179 * wait for request completion callback
181 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
182 struct ceph_mds_request *req);
185 * an in-flight mds request
187 struct ceph_mds_request {
188 u64 r_tid; /* transaction id */
189 struct rb_node r_node;
190 struct ceph_mds_client *r_mdsc;
192 int r_op; /* mds op code */
194 /* operation on what? */
195 struct inode *r_inode; /* arg1 */
196 struct dentry *r_dentry; /* arg1 */
197 struct dentry *r_old_dentry; /* arg2: rename from or link from */
198 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
199 char *r_path1, *r_path2;
200 struct ceph_vino r_ino1, r_ino2;
202 struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
203 struct inode *r_target_inode; /* resulting inode */
205 struct mutex r_fill_mutex;
207 union ceph_mds_request_args r_args;
208 int r_fmode; /* file mode, if expecting cap */
211 struct timespec r_stamp;
213 /* for choosing which mds to send this request to */
215 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
216 bool r_direct_is_hash; /* true if r_direct_hash is valid */
218 /* data payload is used for xattr ops */
219 struct ceph_pagelist *r_pagelist;
221 /* what caps shall we drop? */
222 int r_inode_drop, r_inode_unless;
223 int r_dentry_drop, r_dentry_unless;
224 int r_old_dentry_drop, r_old_dentry_unless;
225 struct inode *r_old_inode;
226 int r_old_inode_drop, r_old_inode_unless;
228 struct ceph_msg *r_request; /* original request */
229 int r_request_release_offset;
230 struct ceph_msg *r_reply;
231 struct ceph_mds_reply_info_parsed r_reply_info;
232 struct page *r_locked_page;
236 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
237 unsigned long r_started; /* start time to measure timeout against */
238 unsigned long r_request_started; /* start time for mds request only,
239 used to measure lease durations */
241 /* link unsafe requests to parent directory, for fsync */
242 struct inode *r_unsafe_dir;
243 struct list_head r_unsafe_dir_item;
245 /* unsafe requests that modify the target inode */
246 struct list_head r_unsafe_target_item;
248 struct ceph_mds_session *r_session;
250 int r_attempts; /* resend attempts */
251 int r_num_fwd; /* number of forward attempts */
252 int r_resend_mds; /* mds to resend to next, if any*/
253 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
256 struct list_head r_wait;
257 struct completion r_completion;
258 struct completion r_safe_completion;
259 ceph_mds_request_callback_t r_callback;
260 ceph_mds_request_wait_callback_t r_wait_for_completion;
261 struct list_head r_unsafe_item; /* per-session unsafe list item */
262 bool r_got_unsafe, r_got_safe, r_got_result;
264 bool r_did_prepopulate;
265 long long r_dir_release_cnt;
266 long long r_dir_ordered_cnt;
267 int r_readdir_cache_idx;
268 u32 r_readdir_offset;
270 struct ceph_cap_reservation r_caps_reservation;
274 struct ceph_pool_perm {
283 struct ceph_mds_client {
284 struct ceph_fs_client *fsc;
285 struct mutex mutex; /* all nested structures */
287 struct ceph_mdsmap *mdsmap;
288 struct completion safe_umount_waiters;
289 wait_queue_head_t session_close_wq;
290 struct list_head waiting_for_map;
292 struct ceph_mds_session **sessions; /* NULL for mds if no session */
293 atomic_t num_sessions;
294 int max_sessions; /* len of s_mds_sessions */
295 int stopping; /* true if shutting down */
298 * snap_rwsem will cover cap linkage into snaprealms, and
299 * realm snap contexts. (later, we can do per-realm snap
300 * contexts locks..) the empty list contains realms with no
301 * references (implying they contain no inodes with caps) that
302 * should be destroyed.
305 struct rw_semaphore snap_rwsem;
306 struct rb_root snap_realms;
307 struct list_head snap_empty;
308 spinlock_t snap_empty_lock; /* protect snap_empty */
310 u64 last_tid; /* most recent mds request */
311 u64 oldest_tid; /* oldest incomplete mds request,
312 excluding setfilelock requests */
313 struct rb_root request_tree; /* pending mds requests */
314 struct delayed_work delayed_work; /* delayed work */
315 unsigned long last_renew_caps; /* last time we renewed our caps */
316 struct list_head cap_delay_list; /* caps with delayed release */
317 spinlock_t cap_delay_lock; /* protects cap_delay_list */
318 struct list_head snap_flush_list; /* cap_snaps ready to flush */
319 spinlock_t snap_flush_lock;
321 u64 last_cap_flush_tid;
322 struct rb_root cap_flush_tree;
323 struct list_head cap_dirty; /* inodes with dirty caps */
324 struct list_head cap_dirty_migrating; /* ...that are migration... */
325 int num_cap_flushing; /* # caps we are flushing */
326 spinlock_t cap_dirty_lock; /* protects above items */
327 wait_queue_head_t cap_flushing_wq;
332 * Maintain a global pool of preallocated struct ceph_caps, referenced
333 * by struct ceph_caps_reservations. This ensures that we preallocate
334 * memory needed to successfully process an MDS response. (If an MDS
335 * sends us cap information and we fail to process it, we will have
336 * problems due to the client and MDS being out of sync.)
338 * Reservations are 'owned' by a ceph_cap_reservation context.
340 spinlock_t caps_list_lock;
341 struct list_head caps_list; /* unused (reserved or
343 int caps_total_count; /* total caps allocated */
344 int caps_use_count; /* in use */
345 int caps_reserve_count; /* unused, reserved */
346 int caps_avail_count; /* unused, unreserved */
347 int caps_min_count; /* keep at least this many
349 spinlock_t dentry_lru_lock;
350 struct list_head dentry_lru;
353 struct rw_semaphore pool_perm_rwsem;
354 struct rb_root pool_perm_tree;
357 extern const char *ceph_mds_op_name(int op);
359 extern struct ceph_mds_session *
360 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
362 static inline struct ceph_mds_session *
363 ceph_get_mds_session(struct ceph_mds_session *s)
365 atomic_inc(&s->s_ref);
369 extern const char *ceph_session_state_name(int s);
371 extern void ceph_put_mds_session(struct ceph_mds_session *s);
373 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
374 struct ceph_msg *msg, int mds);
376 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
377 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
378 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
379 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
381 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
383 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
387 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
388 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
390 extern struct ceph_mds_request *
391 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
392 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
393 struct ceph_mds_request *req);
394 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
396 struct ceph_mds_request *req);
397 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
399 kref_get(&req->r_kref);
401 extern void ceph_mdsc_release_request(struct kref *kref);
402 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
404 kref_put(&req->r_kref, ceph_mdsc_release_request);
407 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
408 struct ceph_mds_session *session);
410 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
412 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
415 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
416 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
418 struct dentry *dentry, char action,
421 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
422 struct ceph_msg *msg);
424 extern struct ceph_mds_session *
425 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
426 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
427 struct ceph_mds_session *session);