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>
11 #include <linux/utsname.h>
13 #include <linux/ceph/types.h>
14 #include <linux/ceph/messenger.h>
15 #include <linux/ceph/mdsmap.h>
16 #include <linux/ceph/auth.h>
19 * Some lock dependencies:
27 * mdsc->snap_flush_lock
28 * mdsc->cap_delay_lock
32 struct ceph_fs_client;
36 * parsed info about a single inode. pointers are into the encoded
37 * on-wire structures within the mds reply message payload.
39 struct ceph_mds_reply_info_in {
40 struct ceph_mds_reply_inode *in;
41 struct ceph_dir_layout dir_layout;
53 struct ceph_mds_reply_dir_entry {
56 struct ceph_mds_reply_lease *lease;
57 struct ceph_mds_reply_info_in inode;
62 * parsed info about an mds reply, including information about
63 * either: 1) the target inode and/or its parent directory and dentry,
64 * and directory contents (for readdir results), or
65 * 2) the file range lock info (for fcntl F_GETLK results).
67 struct ceph_mds_reply_info_parsed {
68 struct ceph_mds_reply_head *head;
71 struct ceph_mds_reply_info_in diri, targeti;
72 struct ceph_mds_reply_dirfrag *dirfrag;
75 struct ceph_mds_reply_lease *dlease;
79 /* for fcntl F_GETLK results */
80 struct ceph_filelock *filelock_reply;
82 /* for readdir results */
84 struct ceph_mds_reply_dirfrag *dir_dir;
91 struct ceph_mds_reply_dir_entry *dir_entries;
94 /* for create results */
101 /* encoded blob describing snapshot contexts for certain
102 operations (e.g., open) */
109 * cap releases are batched and sent to the MDS en masse.
111 * Account for per-message overhead of mds_cap_release header
112 * and __le32 for osd epoch barrier trailing field.
114 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
115 sizeof(struct ceph_mds_cap_release)) / \
116 sizeof(struct ceph_mds_cap_item))
120 * state associated with each MDS<->client session
123 CEPH_MDS_SESSION_NEW = 1,
124 CEPH_MDS_SESSION_OPENING = 2,
125 CEPH_MDS_SESSION_OPEN = 3,
126 CEPH_MDS_SESSION_HUNG = 4,
127 CEPH_MDS_SESSION_CLOSING = 5,
128 CEPH_MDS_SESSION_RESTARTING = 6,
129 CEPH_MDS_SESSION_RECONNECTING = 7,
130 CEPH_MDS_SESSION_REJECTED = 8,
133 struct ceph_mds_session {
134 struct ceph_mds_client *s_mdsc;
137 unsigned long s_ttl; /* time until mds kills us */
138 u64 s_seq; /* incoming msg seq # */
139 struct mutex s_mutex; /* serialize session messages */
141 struct ceph_connection s_con;
143 struct ceph_auth_handshake s_auth;
145 /* protected by s_gen_ttl_lock */
146 spinlock_t s_gen_ttl_lock;
147 u32 s_cap_gen; /* inc each time we get mds stale msg */
148 unsigned long s_cap_ttl; /* when session caps expire */
150 /* protected by s_cap_lock */
151 spinlock_t s_cap_lock;
152 struct list_head s_caps; /* all caps issued by this session */
153 int s_nr_caps, s_trim_caps;
154 int s_num_cap_releases;
157 struct list_head s_cap_releases; /* waiting cap_release messages */
158 struct ceph_cap *s_cap_iterator;
160 /* protected by mutex */
161 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
162 unsigned long s_renew_requested; /* last time we sent a renew req */
166 struct list_head s_waiting; /* waiting requests */
167 struct list_head s_unsafe; /* unsafe requests */
171 * modes of choosing which MDS to send a request to
176 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
179 struct ceph_mds_request;
180 struct ceph_mds_client;
183 * request completion callback
185 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
186 struct ceph_mds_request *req);
188 * wait for request completion callback
190 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
191 struct ceph_mds_request *req);
194 * an in-flight mds request
196 struct ceph_mds_request {
197 u64 r_tid; /* transaction id */
198 struct rb_node r_node;
199 struct ceph_mds_client *r_mdsc;
201 int r_op; /* mds op code */
203 /* operation on what? */
204 struct inode *r_inode; /* arg1 */
205 struct dentry *r_dentry; /* arg1 */
206 struct dentry *r_old_dentry; /* arg2: rename from or link from */
207 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
208 char *r_path1, *r_path2;
209 struct ceph_vino r_ino1, r_ino2;
211 struct inode *r_parent; /* parent dir inode */
212 struct inode *r_target_inode; /* resulting inode */
214 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
215 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
216 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
217 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
218 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
219 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
220 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
221 unsigned long r_req_flags;
223 struct mutex r_fill_mutex;
225 union ceph_mds_request_args r_args;
226 int r_fmode; /* file mode, if expecting cap */
229 struct timespec r_stamp;
231 /* for choosing which mds to send this request to */
233 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
235 /* data payload is used for xattr ops */
236 struct ceph_pagelist *r_pagelist;
238 /* what caps shall we drop? */
239 int r_inode_drop, r_inode_unless;
240 int r_dentry_drop, r_dentry_unless;
241 int r_old_dentry_drop, r_old_dentry_unless;
242 struct inode *r_old_inode;
243 int r_old_inode_drop, r_old_inode_unless;
245 struct ceph_msg *r_request; /* original request */
246 int r_request_release_offset;
247 struct ceph_msg *r_reply;
248 struct ceph_mds_reply_info_parsed r_reply_info;
249 struct page *r_locked_page;
252 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
253 unsigned long r_started; /* start time to measure timeout against */
254 unsigned long r_request_started; /* start time for mds request only,
255 used to measure lease durations */
257 /* link unsafe requests to parent directory, for fsync */
258 struct inode *r_unsafe_dir;
259 struct list_head r_unsafe_dir_item;
261 /* unsafe requests that modify the target inode */
262 struct list_head r_unsafe_target_item;
264 struct ceph_mds_session *r_session;
266 int r_attempts; /* resend attempts */
267 int r_num_fwd; /* number of forward attempts */
268 int r_resend_mds; /* mds to resend to next, if any*/
269 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
272 struct list_head r_wait;
273 struct completion r_completion;
274 struct completion r_safe_completion;
275 ceph_mds_request_callback_t r_callback;
276 ceph_mds_request_wait_callback_t r_wait_for_completion;
277 struct list_head r_unsafe_item; /* per-session unsafe list item */
279 long long r_dir_release_cnt;
280 long long r_dir_ordered_cnt;
281 int r_readdir_cache_idx;
282 u32 r_readdir_offset;
284 struct ceph_cap_reservation r_caps_reservation;
288 struct ceph_pool_perm {
299 struct ceph_mds_client {
300 struct ceph_fs_client *fsc;
301 struct mutex mutex; /* all nested structures */
303 struct ceph_mdsmap *mdsmap;
304 struct completion safe_umount_waiters;
305 wait_queue_head_t session_close_wq;
306 struct list_head waiting_for_map;
309 struct ceph_mds_session **sessions; /* NULL for mds if no session */
310 atomic_t num_sessions;
311 int max_sessions; /* len of s_mds_sessions */
312 int stopping; /* true if shutting down */
315 * snap_rwsem will cover cap linkage into snaprealms, and
316 * realm snap contexts. (later, we can do per-realm snap
317 * contexts locks..) the empty list contains realms with no
318 * references (implying they contain no inodes with caps) that
319 * should be destroyed.
322 struct rw_semaphore snap_rwsem;
323 struct rb_root snap_realms;
324 struct list_head snap_empty;
325 spinlock_t snap_empty_lock; /* protect snap_empty */
327 u64 last_tid; /* most recent mds request */
328 u64 oldest_tid; /* oldest incomplete mds request,
329 excluding setfilelock requests */
330 struct rb_root request_tree; /* pending mds requests */
331 struct delayed_work delayed_work; /* delayed work */
332 unsigned long last_renew_caps; /* last time we renewed our caps */
333 struct list_head cap_delay_list; /* caps with delayed release */
334 spinlock_t cap_delay_lock; /* protects cap_delay_list */
335 struct list_head snap_flush_list; /* cap_snaps ready to flush */
336 spinlock_t snap_flush_lock;
338 u64 last_cap_flush_tid;
339 struct list_head cap_flush_list;
340 struct list_head cap_dirty; /* inodes with dirty caps */
341 struct list_head cap_dirty_migrating; /* ...that are migration... */
342 int num_cap_flushing; /* # caps we are flushing */
343 spinlock_t cap_dirty_lock; /* protects above items */
344 wait_queue_head_t cap_flushing_wq;
349 * Maintain a global pool of preallocated struct ceph_caps, referenced
350 * by struct ceph_caps_reservations. This ensures that we preallocate
351 * memory needed to successfully process an MDS response. (If an MDS
352 * sends us cap information and we fail to process it, we will have
353 * problems due to the client and MDS being out of sync.)
355 * Reservations are 'owned' by a ceph_cap_reservation context.
357 spinlock_t caps_list_lock;
358 struct list_head caps_list; /* unused (reserved or
360 int caps_total_count; /* total caps allocated */
361 int caps_use_count; /* in use */
362 int caps_reserve_count; /* unused, reserved */
363 int caps_avail_count; /* unused, unreserved */
364 int caps_min_count; /* keep at least this many
366 spinlock_t dentry_lru_lock;
367 struct list_head dentry_lru;
370 struct rw_semaphore pool_perm_rwsem;
371 struct rb_root pool_perm_tree;
373 char nodename[__NEW_UTS_LEN + 1];
376 extern const char *ceph_mds_op_name(int op);
378 extern struct ceph_mds_session *
379 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
381 static inline struct ceph_mds_session *
382 ceph_get_mds_session(struct ceph_mds_session *s)
384 refcount_inc(&s->s_ref);
388 extern const char *ceph_session_state_name(int s);
390 extern void ceph_put_mds_session(struct ceph_mds_session *s);
392 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
393 struct ceph_msg *msg, int mds);
395 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
396 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
397 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
398 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
400 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
402 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
403 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
405 extern struct ceph_mds_request *
406 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
407 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
408 struct ceph_mds_request *req);
409 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
411 struct ceph_mds_request *req);
412 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
414 kref_get(&req->r_kref);
416 extern void ceph_mdsc_release_request(struct kref *kref);
417 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
419 kref_put(&req->r_kref, ceph_mdsc_release_request);
422 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
423 struct ceph_mds_session *session);
425 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
427 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
430 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
431 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
433 struct dentry *dentry, char action,
436 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
437 struct ceph_msg *msg);
438 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
439 struct ceph_msg *msg);
441 extern struct ceph_mds_session *
442 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
443 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
444 struct ceph_mds_session *session);