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>
16 * Some lock dependencies:
24 * mdsc->snap_flush_lock
25 * mdsc->cap_delay_lock
29 struct ceph_fs_client;
33 * parsed info about a single inode. pointers are into the encoded
34 * on-wire structures within the mds reply message payload.
36 struct ceph_mds_reply_info_in {
37 struct ceph_mds_reply_inode *in;
45 * parsed info about an mds reply, including information about the
46 * target inode and/or its parent directory and dentry, and directory
47 * contents (for readdir results).
49 struct ceph_mds_reply_info_parsed {
50 struct ceph_mds_reply_head *head;
52 struct ceph_mds_reply_info_in diri, targeti;
53 struct ceph_mds_reply_dirfrag *dirfrag;
56 struct ceph_mds_reply_lease *dlease;
58 struct ceph_mds_reply_dirfrag *dir_dir;
62 struct ceph_mds_reply_lease **dir_dlease;
63 struct ceph_mds_reply_info_in *dir_in;
64 u8 dir_complete, dir_end;
66 /* encoded blob describing snapshot contexts for certain
67 operations (e.g., open) */
74 * cap releases are batched and sent to the MDS en masse.
76 #define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE - \
77 sizeof(struct ceph_mds_cap_release)) / \
78 sizeof(struct ceph_mds_cap_item))
82 * state associated with each MDS<->client session
85 CEPH_MDS_SESSION_NEW = 1,
86 CEPH_MDS_SESSION_OPENING = 2,
87 CEPH_MDS_SESSION_OPEN = 3,
88 CEPH_MDS_SESSION_HUNG = 4,
89 CEPH_MDS_SESSION_CLOSING = 5,
90 CEPH_MDS_SESSION_RESTARTING = 6,
91 CEPH_MDS_SESSION_RECONNECTING = 7,
94 struct ceph_mds_session {
95 struct ceph_mds_client *s_mdsc;
98 unsigned long s_ttl; /* time until mds kills us */
99 u64 s_seq; /* incoming msg seq # */
100 struct mutex s_mutex; /* serialize session messages */
102 struct ceph_connection s_con;
104 struct ceph_authorizer *s_authorizer;
105 void *s_authorizer_buf, *s_authorizer_reply_buf;
106 size_t s_authorizer_buf_len, s_authorizer_reply_buf_len;
108 /* protected by s_cap_lock */
109 spinlock_t s_cap_lock;
110 u32 s_cap_gen; /* inc each time we get mds stale msg */
111 unsigned long s_cap_ttl; /* when session caps expire */
112 struct list_head s_caps; /* all caps issued by this session */
113 int s_nr_caps, s_trim_caps;
114 int s_num_cap_releases;
115 struct list_head s_cap_releases; /* waiting cap_release messages */
116 struct list_head s_cap_releases_done; /* ready to send */
117 struct ceph_cap *s_cap_iterator;
119 /* protected by mutex */
120 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
121 struct list_head s_cap_snaps_flushing;
122 unsigned long s_renew_requested; /* last time we sent a renew req */
126 struct list_head s_waiting; /* waiting requests */
127 struct list_head s_unsafe; /* unsafe requests */
131 * modes of choosing which MDS to send a request to
136 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
139 struct ceph_mds_request;
140 struct ceph_mds_client;
143 * request completion callback
145 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
146 struct ceph_mds_request *req);
149 * an in-flight mds request
151 struct ceph_mds_request {
152 u64 r_tid; /* transaction id */
153 struct rb_node r_node;
154 struct ceph_mds_client *r_mdsc;
156 int r_op; /* mds op code */
159 /* operation on what? */
160 struct inode *r_inode; /* arg1 */
161 struct dentry *r_dentry; /* arg1 */
162 struct dentry *r_old_dentry; /* arg2: rename from or link from */
163 char *r_path1, *r_path2;
164 struct ceph_vino r_ino1, r_ino2;
166 struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
167 struct inode *r_target_inode; /* resulting inode */
169 struct mutex r_fill_mutex;
171 union ceph_mds_request_args r_args;
172 int r_fmode; /* file mode, if expecting cap */
176 /* for choosing which mds to send this request to */
178 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
179 bool r_direct_is_hash; /* true if r_direct_hash is valid */
181 /* data payload is used for xattr ops */
182 struct page **r_pages;
186 /* what caps shall we drop? */
187 int r_inode_drop, r_inode_unless;
188 int r_dentry_drop, r_dentry_unless;
189 int r_old_dentry_drop, r_old_dentry_unless;
190 struct inode *r_old_inode;
191 int r_old_inode_drop, r_old_inode_unless;
193 struct ceph_msg *r_request; /* original request */
194 int r_request_release_offset;
195 struct ceph_msg *r_reply;
196 struct ceph_mds_reply_info_parsed r_reply_info;
200 unsigned long r_timeout; /* optional. jiffies */
201 unsigned long r_started; /* start time to measure timeout against */
202 unsigned long r_request_started; /* start time for mds request only,
203 used to measure lease durations */
205 /* link unsafe requests to parent directory, for fsync */
206 struct inode *r_unsafe_dir;
207 struct list_head r_unsafe_dir_item;
209 struct ceph_mds_session *r_session;
211 int r_attempts; /* resend attempts */
212 int r_num_fwd; /* number of forward attempts */
213 int r_resend_mds; /* mds to resend to next, if any*/
214 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
217 struct list_head r_wait;
218 struct completion r_completion;
219 struct completion r_safe_completion;
220 ceph_mds_request_callback_t r_callback;
221 struct list_head r_unsafe_item; /* per-session unsafe list item */
222 bool r_got_unsafe, r_got_safe, r_got_result;
224 bool r_did_prepopulate;
225 u32 r_readdir_offset;
227 struct ceph_cap_reservation r_caps_reservation;
234 struct ceph_mds_client {
235 struct ceph_fs_client *fsc;
236 struct mutex mutex; /* all nested structures */
238 struct ceph_mdsmap *mdsmap;
239 struct completion safe_umount_waiters;
240 wait_queue_head_t session_close_wq;
241 struct list_head waiting_for_map;
243 struct ceph_mds_session **sessions; /* NULL for mds if no session */
244 int max_sessions; /* len of s_mds_sessions */
245 int stopping; /* true if shutting down */
248 * snap_rwsem will cover cap linkage into snaprealms, and
249 * realm snap contexts. (later, we can do per-realm snap
250 * contexts locks..) the empty list contains realms with no
251 * references (implying they contain no inodes with caps) that
252 * should be destroyed.
254 struct rw_semaphore snap_rwsem;
255 struct rb_root snap_realms;
256 struct list_head snap_empty;
257 spinlock_t snap_empty_lock; /* protect snap_empty */
259 u64 last_tid; /* most recent mds request */
260 struct rb_root request_tree; /* pending mds requests */
261 struct delayed_work delayed_work; /* delayed work */
262 unsigned long last_renew_caps; /* last time we renewed our caps */
263 struct list_head cap_delay_list; /* caps with delayed release */
264 spinlock_t cap_delay_lock; /* protects cap_delay_list */
265 struct list_head snap_flush_list; /* cap_snaps ready to flush */
266 spinlock_t snap_flush_lock;
269 struct list_head cap_dirty; /* inodes with dirty caps */
270 int num_cap_flushing; /* # caps we are flushing */
271 spinlock_t cap_dirty_lock; /* protects above items */
272 wait_queue_head_t cap_flushing_wq;
277 * Maintain a global pool of preallocated struct ceph_caps, referenced
278 * by struct ceph_caps_reservations. This ensures that we preallocate
279 * memory needed to successfully process an MDS response. (If an MDS
280 * sends us cap information and we fail to process it, we will have
281 * problems due to the client and MDS being out of sync.)
283 * Reservations are 'owned' by a ceph_cap_reservation context.
285 spinlock_t caps_list_lock;
286 struct list_head caps_list; /* unused (reserved or
288 int caps_total_count; /* total caps allocated */
289 int caps_use_count; /* in use */
290 int caps_reserve_count; /* unused, reserved */
291 int caps_avail_count; /* unused, unreserved */
292 int caps_min_count; /* keep at least this many
294 spinlock_t dentry_lru_lock;
295 struct list_head dentry_lru;
299 extern const char *ceph_mds_op_name(int op);
301 extern struct ceph_mds_session *
302 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
304 static inline struct ceph_mds_session *
305 ceph_get_mds_session(struct ceph_mds_session *s)
307 atomic_inc(&s->s_ref);
311 extern void ceph_put_mds_session(struct ceph_mds_session *s);
313 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
314 struct ceph_msg *msg, int mds);
316 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
317 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
318 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
320 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
322 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
324 struct dentry *dn, int mask);
326 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
328 extern struct ceph_mds_request *
329 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
330 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
331 struct ceph_mds_request *req);
332 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
334 struct ceph_mds_request *req);
335 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
337 kref_get(&req->r_kref);
339 extern void ceph_mdsc_release_request(struct kref *kref);
340 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
342 kref_put(&req->r_kref, ceph_mdsc_release_request);
345 extern int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
346 struct ceph_mds_session *session);
347 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
348 struct ceph_mds_session *session);
350 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
352 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
355 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
356 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
358 struct dentry *dentry, char action,
361 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
362 struct ceph_msg *msg);
364 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
365 struct ceph_mds_session *session);