]>
Commit | Line | Data |
---|---|---|
2f2dc053 SW |
1 | #ifndef _FS_CEPH_MDS_CLIENT_H |
2 | #define _FS_CEPH_MDS_CLIENT_H | |
3 | ||
4 | #include <linux/completion.h> | |
153c8e6b | 5 | #include <linux/kref.h> |
2f2dc053 SW |
6 | #include <linux/list.h> |
7 | #include <linux/mutex.h> | |
44ca18f2 | 8 | #include <linux/rbtree.h> |
2f2dc053 SW |
9 | #include <linux/spinlock.h> |
10 | ||
3d14c5d2 YS |
11 | #include <linux/ceph/types.h> |
12 | #include <linux/ceph/messenger.h> | |
13 | #include <linux/ceph/mdsmap.h> | |
6c4a1915 | 14 | #include <linux/ceph/auth.h> |
2f2dc053 SW |
15 | |
16 | /* | |
17 | * Some lock dependencies: | |
18 | * | |
19 | * session->s_mutex | |
20 | * mdsc->mutex | |
21 | * | |
22 | * mdsc->snap_rwsem | |
23 | * | |
be655596 | 24 | * ci->i_ceph_lock |
2f2dc053 SW |
25 | * mdsc->snap_flush_lock |
26 | * mdsc->cap_delay_lock | |
27 | * | |
28 | */ | |
29 | ||
3d14c5d2 | 30 | struct ceph_fs_client; |
2f2dc053 SW |
31 | struct ceph_cap; |
32 | ||
33 | /* | |
34 | * parsed info about a single inode. pointers are into the encoded | |
35 | * on-wire structures within the mds reply message payload. | |
36 | */ | |
37 | struct ceph_mds_reply_info_in { | |
38 | struct ceph_mds_reply_inode *in; | |
14303d20 | 39 | struct ceph_dir_layout dir_layout; |
2f2dc053 SW |
40 | u32 symlink_len; |
41 | char *symlink; | |
42 | u32 xattr_len; | |
43 | char *xattr_data; | |
44 | }; | |
45 | ||
46 | /* | |
25933abd HS |
47 | * parsed info about an mds reply, including information about |
48 | * either: 1) the target inode and/or its parent directory and dentry, | |
49 | * and directory contents (for readdir results), or | |
50 | * 2) the file range lock info (for fcntl F_GETLK results). | |
2f2dc053 SW |
51 | */ |
52 | struct ceph_mds_reply_info_parsed { | |
53 | struct ceph_mds_reply_head *head; | |
54 | ||
25933abd | 55 | /* trace */ |
2f2dc053 SW |
56 | struct ceph_mds_reply_info_in diri, targeti; |
57 | struct ceph_mds_reply_dirfrag *dirfrag; | |
58 | char *dname; | |
59 | u32 dname_len; | |
60 | struct ceph_mds_reply_lease *dlease; | |
61 | ||
25933abd HS |
62 | /* extra */ |
63 | union { | |
64 | /* for fcntl F_GETLK results */ | |
65 | struct ceph_filelock *filelock_reply; | |
66 | ||
67 | /* for readdir results */ | |
68 | struct { | |
69 | struct ceph_mds_reply_dirfrag *dir_dir; | |
54008399 | 70 | size_t dir_buf_size; |
25933abd HS |
71 | int dir_nr; |
72 | char **dir_dname; | |
73 | u32 *dir_dname_len; | |
74 | struct ceph_mds_reply_lease **dir_dlease; | |
75 | struct ceph_mds_reply_info_in *dir_in; | |
76 | u8 dir_complete, dir_end; | |
77 | }; | |
6e8575fa SL |
78 | |
79 | /* for create results */ | |
80 | struct { | |
81 | bool has_create_ino; | |
82 | u64 ino; | |
83 | }; | |
25933abd | 84 | }; |
2f2dc053 SW |
85 | |
86 | /* encoded blob describing snapshot contexts for certain | |
87 | operations (e.g., open) */ | |
88 | void *snapblob; | |
89 | int snapblob_len; | |
90 | }; | |
91 | ||
92 | ||
93 | /* | |
94 | * cap releases are batched and sent to the MDS en masse. | |
95 | */ | |
96 | #define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE - \ | |
97 | sizeof(struct ceph_mds_cap_release)) / \ | |
98 | sizeof(struct ceph_mds_cap_item)) | |
99 | ||
100 | ||
101 | /* | |
102 | * state associated with each MDS<->client session | |
103 | */ | |
104 | enum { | |
105 | CEPH_MDS_SESSION_NEW = 1, | |
106 | CEPH_MDS_SESSION_OPENING = 2, | |
107 | CEPH_MDS_SESSION_OPEN = 3, | |
108 | CEPH_MDS_SESSION_HUNG = 4, | |
109 | CEPH_MDS_SESSION_CLOSING = 5, | |
110 | CEPH_MDS_SESSION_RESTARTING = 6, | |
111 | CEPH_MDS_SESSION_RECONNECTING = 7, | |
112 | }; | |
113 | ||
114 | struct ceph_mds_session { | |
115 | struct ceph_mds_client *s_mdsc; | |
116 | int s_mds; | |
117 | int s_state; | |
118 | unsigned long s_ttl; /* time until mds kills us */ | |
119 | u64 s_seq; /* incoming msg seq # */ | |
120 | struct mutex s_mutex; /* serialize session messages */ | |
121 | ||
122 | struct ceph_connection s_con; | |
123 | ||
6c4a1915 | 124 | struct ceph_auth_handshake s_auth; |
4e7a5dcd | 125 | |
d8fb02ab AE |
126 | /* protected by s_gen_ttl_lock */ |
127 | spinlock_t s_gen_ttl_lock; | |
2f2dc053 SW |
128 | u32 s_cap_gen; /* inc each time we get mds stale msg */ |
129 | unsigned long s_cap_ttl; /* when session caps expire */ | |
d8fb02ab AE |
130 | |
131 | /* protected by s_cap_lock */ | |
132 | spinlock_t s_cap_lock; | |
2f2dc053 SW |
133 | struct list_head s_caps; /* all caps issued by this session */ |
134 | int s_nr_caps, s_trim_caps; | |
135 | int s_num_cap_releases; | |
99a9c273 | 136 | int s_cap_reconnect; |
2f2dc053 SW |
137 | struct list_head s_cap_releases; /* waiting cap_release messages */ |
138 | struct list_head s_cap_releases_done; /* ready to send */ | |
7c1332b8 | 139 | struct ceph_cap *s_cap_iterator; |
2f2dc053 SW |
140 | |
141 | /* protected by mutex */ | |
142 | struct list_head s_cap_flushing; /* inodes w/ flushing caps */ | |
143 | struct list_head s_cap_snaps_flushing; | |
144 | unsigned long s_renew_requested; /* last time we sent a renew req */ | |
145 | u64 s_renew_seq; | |
146 | ||
147 | atomic_t s_ref; | |
148 | struct list_head s_waiting; /* waiting requests */ | |
149 | struct list_head s_unsafe; /* unsafe requests */ | |
150 | }; | |
151 | ||
152 | /* | |
153 | * modes of choosing which MDS to send a request to | |
154 | */ | |
155 | enum { | |
156 | USE_ANY_MDS, | |
157 | USE_RANDOM_MDS, | |
158 | USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */ | |
159 | }; | |
160 | ||
161 | struct ceph_mds_request; | |
162 | struct ceph_mds_client; | |
163 | ||
164 | /* | |
165 | * request completion callback | |
166 | */ | |
167 | typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc, | |
168 | struct ceph_mds_request *req); | |
169 | ||
170 | /* | |
171 | * an in-flight mds request | |
172 | */ | |
173 | struct ceph_mds_request { | |
174 | u64 r_tid; /* transaction id */ | |
44ca18f2 | 175 | struct rb_node r_node; |
37151668 | 176 | struct ceph_mds_client *r_mdsc; |
2f2dc053 SW |
177 | |
178 | int r_op; /* mds op code */ | |
2f2dc053 SW |
179 | |
180 | /* operation on what? */ | |
181 | struct inode *r_inode; /* arg1 */ | |
182 | struct dentry *r_dentry; /* arg1 */ | |
183 | struct dentry *r_old_dentry; /* arg2: rename from or link from */ | |
41b02e1f | 184 | struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */ |
2f2dc053 SW |
185 | char *r_path1, *r_path2; |
186 | struct ceph_vino r_ino1, r_ino2; | |
187 | ||
188 | struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */ | |
189 | struct inode *r_target_inode; /* resulting inode */ | |
190 | ||
b4556396 SW |
191 | struct mutex r_fill_mutex; |
192 | ||
2f2dc053 SW |
193 | union ceph_mds_request_args r_args; |
194 | int r_fmode; /* file mode, if expecting cap */ | |
ff3d0046 EB |
195 | kuid_t r_uid; |
196 | kgid_t r_gid; | |
2f2dc053 SW |
197 | |
198 | /* for choosing which mds to send this request to */ | |
199 | int r_direct_mode; | |
200 | u32 r_direct_hash; /* choose dir frag based on this dentry hash */ | |
201 | bool r_direct_is_hash; /* true if r_direct_hash is valid */ | |
202 | ||
203 | /* data payload is used for xattr ops */ | |
204 | struct page **r_pages; | |
205 | int r_num_pages; | |
206 | int r_data_len; | |
207 | ||
208 | /* what caps shall we drop? */ | |
209 | int r_inode_drop, r_inode_unless; | |
210 | int r_dentry_drop, r_dentry_unless; | |
211 | int r_old_dentry_drop, r_old_dentry_unless; | |
212 | struct inode *r_old_inode; | |
213 | int r_old_inode_drop, r_old_inode_unless; | |
214 | ||
215 | struct ceph_msg *r_request; /* original request */ | |
e979cf50 | 216 | int r_request_release_offset; |
2f2dc053 SW |
217 | struct ceph_msg *r_reply; |
218 | struct ceph_mds_reply_info_parsed r_reply_info; | |
219 | int r_err; | |
5b1daecd | 220 | bool r_aborted; |
2f2dc053 SW |
221 | |
222 | unsigned long r_timeout; /* optional. jiffies */ | |
223 | unsigned long r_started; /* start time to measure timeout against */ | |
224 | unsigned long r_request_started; /* start time for mds request only, | |
225 | used to measure lease durations */ | |
226 | ||
227 | /* link unsafe requests to parent directory, for fsync */ | |
228 | struct inode *r_unsafe_dir; | |
229 | struct list_head r_unsafe_dir_item; | |
230 | ||
231 | struct ceph_mds_session *r_session; | |
232 | ||
233 | int r_attempts; /* resend attempts */ | |
234 | int r_num_fwd; /* number of forward attempts */ | |
2f2dc053 | 235 | int r_resend_mds; /* mds to resend to next, if any*/ |
e55b71f8 | 236 | u32 r_sent_on_mseq; /* cap mseq request was sent at*/ |
2f2dc053 | 237 | |
153c8e6b | 238 | struct kref r_kref; |
2f2dc053 SW |
239 | struct list_head r_wait; |
240 | struct completion r_completion; | |
241 | struct completion r_safe_completion; | |
242 | ceph_mds_request_callback_t r_callback; | |
243 | struct list_head r_unsafe_item; /* per-session unsafe list item */ | |
e1518c7c | 244 | bool r_got_unsafe, r_got_safe, r_got_result; |
2f2dc053 SW |
245 | |
246 | bool r_did_prepopulate; | |
247 | u32 r_readdir_offset; | |
248 | ||
249 | struct ceph_cap_reservation r_caps_reservation; | |
250 | int r_num_caps; | |
251 | }; | |
252 | ||
253 | /* | |
254 | * mds client state | |
255 | */ | |
256 | struct ceph_mds_client { | |
3d14c5d2 | 257 | struct ceph_fs_client *fsc; |
2f2dc053 SW |
258 | struct mutex mutex; /* all nested structures */ |
259 | ||
260 | struct ceph_mdsmap *mdsmap; | |
f3c60c59 SW |
261 | struct completion safe_umount_waiters; |
262 | wait_queue_head_t session_close_wq; | |
2f2dc053 SW |
263 | struct list_head waiting_for_map; |
264 | ||
265 | struct ceph_mds_session **sessions; /* NULL for mds if no session */ | |
266 | int max_sessions; /* len of s_mds_sessions */ | |
267 | int stopping; /* true if shutting down */ | |
268 | ||
269 | /* | |
270 | * snap_rwsem will cover cap linkage into snaprealms, and | |
271 | * realm snap contexts. (later, we can do per-realm snap | |
272 | * contexts locks..) the empty list contains realms with no | |
273 | * references (implying they contain no inodes with caps) that | |
274 | * should be destroyed. | |
275 | */ | |
276 | struct rw_semaphore snap_rwsem; | |
a105f00c | 277 | struct rb_root snap_realms; |
2f2dc053 SW |
278 | struct list_head snap_empty; |
279 | spinlock_t snap_empty_lock; /* protect snap_empty */ | |
280 | ||
281 | u64 last_tid; /* most recent mds request */ | |
44ca18f2 | 282 | struct rb_root request_tree; /* pending mds requests */ |
2f2dc053 SW |
283 | struct delayed_work delayed_work; /* delayed work */ |
284 | unsigned long last_renew_caps; /* last time we renewed our caps */ | |
285 | struct list_head cap_delay_list; /* caps with delayed release */ | |
286 | spinlock_t cap_delay_lock; /* protects cap_delay_list */ | |
287 | struct list_head snap_flush_list; /* cap_snaps ready to flush */ | |
288 | spinlock_t snap_flush_lock; | |
289 | ||
290 | u64 cap_flush_seq; | |
291 | struct list_head cap_dirty; /* inodes with dirty caps */ | |
db354052 | 292 | struct list_head cap_dirty_migrating; /* ...that are migration... */ |
2f2dc053 SW |
293 | int num_cap_flushing; /* # caps we are flushing */ |
294 | spinlock_t cap_dirty_lock; /* protects above items */ | |
295 | wait_queue_head_t cap_flushing_wq; | |
296 | ||
37151668 YS |
297 | /* |
298 | * Cap reservations | |
299 | * | |
300 | * Maintain a global pool of preallocated struct ceph_caps, referenced | |
301 | * by struct ceph_caps_reservations. This ensures that we preallocate | |
302 | * memory needed to successfully process an MDS response. (If an MDS | |
303 | * sends us cap information and we fail to process it, we will have | |
304 | * problems due to the client and MDS being out of sync.) | |
305 | * | |
306 | * Reservations are 'owned' by a ceph_cap_reservation context. | |
307 | */ | |
308 | spinlock_t caps_list_lock; | |
309 | struct list_head caps_list; /* unused (reserved or | |
310 | unreserved) */ | |
311 | int caps_total_count; /* total caps allocated */ | |
312 | int caps_use_count; /* in use */ | |
313 | int caps_reserve_count; /* unused, reserved */ | |
314 | int caps_avail_count; /* unused, unreserved */ | |
315 | int caps_min_count; /* keep at least this many | |
316 | (unreserved) */ | |
2f2dc053 SW |
317 | spinlock_t dentry_lru_lock; |
318 | struct list_head dentry_lru; | |
319 | int num_dentry; | |
320 | }; | |
321 | ||
322 | extern const char *ceph_mds_op_name(int op); | |
323 | ||
324 | extern struct ceph_mds_session * | |
325 | __ceph_lookup_mds_session(struct ceph_mds_client *, int mds); | |
326 | ||
327 | static inline struct ceph_mds_session * | |
328 | ceph_get_mds_session(struct ceph_mds_session *s) | |
329 | { | |
330 | atomic_inc(&s->s_ref); | |
331 | return s; | |
332 | } | |
333 | ||
334 | extern void ceph_put_mds_session(struct ceph_mds_session *s); | |
335 | ||
336 | extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc, | |
337 | struct ceph_msg *msg, int mds); | |
338 | ||
3d14c5d2 | 339 | extern int ceph_mdsc_init(struct ceph_fs_client *fsc); |
2f2dc053 | 340 | extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc); |
3d14c5d2 | 341 | extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc); |
2f2dc053 SW |
342 | |
343 | extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc); | |
344 | ||
345 | extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, | |
346 | struct inode *inode, | |
2f90b852 | 347 | struct dentry *dn); |
2f2dc053 | 348 | |
167c9e35 | 349 | extern void ceph_invalidate_dir_request(struct ceph_mds_request *req); |
54008399 YZ |
350 | extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req, |
351 | struct inode *dir); | |
2f2dc053 SW |
352 | extern struct ceph_mds_request * |
353 | ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode); | |
354 | extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, | |
355 | struct ceph_mds_request *req); | |
356 | extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc, | |
357 | struct inode *dir, | |
358 | struct ceph_mds_request *req); | |
359 | static inline void ceph_mdsc_get_request(struct ceph_mds_request *req) | |
360 | { | |
153c8e6b SW |
361 | kref_get(&req->r_kref); |
362 | } | |
363 | extern void ceph_mdsc_release_request(struct kref *kref); | |
364 | static inline void ceph_mdsc_put_request(struct ceph_mds_request *req) | |
365 | { | |
366 | kref_put(&req->r_kref, ceph_mdsc_release_request); | |
2f2dc053 | 367 | } |
2f2dc053 | 368 | |
2b2300d6 | 369 | extern int ceph_add_cap_releases(struct ceph_mds_client *mdsc, |
ee6b272b | 370 | struct ceph_mds_session *session); |
3d7ded4d SW |
371 | extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc, |
372 | struct ceph_mds_session *session); | |
373 | ||
2f2dc053 SW |
374 | extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc); |
375 | ||
376 | extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base, | |
377 | int stop_on_nosnap); | |
378 | ||
379 | extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry); | |
380 | extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session, | |
381 | struct inode *inode, | |
382 | struct dentry *dentry, char action, | |
383 | u32 seq); | |
384 | ||
385 | extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, | |
386 | struct ceph_msg *msg); | |
387 | ||
5d72d13c YZ |
388 | extern struct ceph_mds_session * |
389 | ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target); | |
154f42c2 SW |
390 | extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc, |
391 | struct ceph_mds_session *session); | |
392 | ||
2f2dc053 | 393 | #endif |