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; | |
fb01d1f8 YZ |
44 | u64 inline_version; |
45 | u32 inline_len; | |
46 | char *inline_data; | |
5ea5c5e0 | 47 | u32 pool_ns_len; |
2f2dc053 SW |
48 | }; |
49 | ||
2a5beea3 YZ |
50 | struct ceph_mds_reply_dir_entry { |
51 | char *name; | |
52 | u32 name_len; | |
53 | struct ceph_mds_reply_lease *lease; | |
54 | struct ceph_mds_reply_info_in inode; | |
8974eebd | 55 | loff_t offset; |
2a5beea3 YZ |
56 | }; |
57 | ||
2f2dc053 | 58 | /* |
25933abd HS |
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). | |
2f2dc053 SW |
63 | */ |
64 | struct ceph_mds_reply_info_parsed { | |
65 | struct ceph_mds_reply_head *head; | |
66 | ||
25933abd | 67 | /* trace */ |
2f2dc053 SW |
68 | struct ceph_mds_reply_info_in diri, targeti; |
69 | struct ceph_mds_reply_dirfrag *dirfrag; | |
70 | char *dname; | |
71 | u32 dname_len; | |
72 | struct ceph_mds_reply_lease *dlease; | |
73 | ||
25933abd HS |
74 | /* extra */ |
75 | union { | |
76 | /* for fcntl F_GETLK results */ | |
77 | struct ceph_filelock *filelock_reply; | |
78 | ||
79 | /* for readdir results */ | |
80 | struct { | |
81 | struct ceph_mds_reply_dirfrag *dir_dir; | |
54008399 | 82 | size_t dir_buf_size; |
25933abd | 83 | int dir_nr; |
956d39d6 | 84 | bool dir_complete, dir_end; |
2a5beea3 | 85 | struct ceph_mds_reply_dir_entry *dir_entries; |
25933abd | 86 | }; |
6e8575fa SL |
87 | |
88 | /* for create results */ | |
89 | struct { | |
90 | bool has_create_ino; | |
91 | u64 ino; | |
92 | }; | |
25933abd | 93 | }; |
2f2dc053 SW |
94 | |
95 | /* encoded blob describing snapshot contexts for certain | |
96 | operations (e.g., open) */ | |
97 | void *snapblob; | |
98 | int snapblob_len; | |
99 | }; | |
100 | ||
101 | ||
102 | /* | |
103 | * cap releases are batched and sent to the MDS en masse. | |
104 | */ | |
09cbfeaf | 105 | #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - \ |
2f2dc053 SW |
106 | sizeof(struct ceph_mds_cap_release)) / \ |
107 | sizeof(struct ceph_mds_cap_item)) | |
108 | ||
109 | ||
110 | /* | |
111 | * state associated with each MDS<->client session | |
112 | */ | |
113 | enum { | |
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, | |
121 | }; | |
122 | ||
123 | struct ceph_mds_session { | |
124 | struct ceph_mds_client *s_mdsc; | |
125 | int s_mds; | |
126 | int s_state; | |
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 */ | |
130 | ||
131 | struct ceph_connection s_con; | |
132 | ||
6c4a1915 | 133 | struct ceph_auth_handshake s_auth; |
4e7a5dcd | 134 | |
d8fb02ab AE |
135 | /* protected by s_gen_ttl_lock */ |
136 | spinlock_t s_gen_ttl_lock; | |
2f2dc053 SW |
137 | u32 s_cap_gen; /* inc each time we get mds stale msg */ |
138 | unsigned long s_cap_ttl; /* when session caps expire */ | |
d8fb02ab AE |
139 | |
140 | /* protected by s_cap_lock */ | |
141 | spinlock_t s_cap_lock; | |
2f2dc053 SW |
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; | |
99a9c273 | 145 | int s_cap_reconnect; |
03f4fcb0 | 146 | int s_readonly; |
2f2dc053 | 147 | struct list_head s_cap_releases; /* waiting cap_release messages */ |
7c1332b8 | 148 | struct ceph_cap *s_cap_iterator; |
2f2dc053 SW |
149 | |
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 */ | |
154 | u64 s_renew_seq; | |
155 | ||
156 | atomic_t s_ref; | |
157 | struct list_head s_waiting; /* waiting requests */ | |
158 | struct list_head s_unsafe; /* unsafe requests */ | |
159 | }; | |
160 | ||
161 | /* | |
162 | * modes of choosing which MDS to send a request to | |
163 | */ | |
164 | enum { | |
165 | USE_ANY_MDS, | |
166 | USE_RANDOM_MDS, | |
167 | USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */ | |
168 | }; | |
169 | ||
170 | struct ceph_mds_request; | |
171 | struct ceph_mds_client; | |
172 | ||
173 | /* | |
174 | * request completion callback | |
175 | */ | |
176 | typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc, | |
177 | struct ceph_mds_request *req); | |
9280be24 YZ |
178 | /* |
179 | * wait for request completion callback | |
180 | */ | |
181 | typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc, | |
182 | struct ceph_mds_request *req); | |
2f2dc053 SW |
183 | |
184 | /* | |
185 | * an in-flight mds request | |
186 | */ | |
187 | struct ceph_mds_request { | |
188 | u64 r_tid; /* transaction id */ | |
44ca18f2 | 189 | struct rb_node r_node; |
37151668 | 190 | struct ceph_mds_client *r_mdsc; |
2f2dc053 SW |
191 | |
192 | int r_op; /* mds op code */ | |
2f2dc053 SW |
193 | |
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 */ | |
41b02e1f | 198 | struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */ |
2f2dc053 SW |
199 | char *r_path1, *r_path2; |
200 | struct ceph_vino r_ino1, r_ino2; | |
201 | ||
202 | struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */ | |
203 | struct inode *r_target_inode; /* resulting inode */ | |
204 | ||
b4556396 SW |
205 | struct mutex r_fill_mutex; |
206 | ||
2f2dc053 SW |
207 | union ceph_mds_request_args r_args; |
208 | int r_fmode; /* file mode, if expecting cap */ | |
ff3d0046 EB |
209 | kuid_t r_uid; |
210 | kgid_t r_gid; | |
b8e69066 | 211 | struct timespec r_stamp; |
2f2dc053 SW |
212 | |
213 | /* for choosing which mds to send this request to */ | |
214 | int r_direct_mode; | |
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 */ | |
217 | ||
218 | /* data payload is used for xattr ops */ | |
25e6bae3 | 219 | struct ceph_pagelist *r_pagelist; |
2f2dc053 SW |
220 | |
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; | |
227 | ||
228 | struct ceph_msg *r_request; /* original request */ | |
e979cf50 | 229 | int r_request_release_offset; |
2f2dc053 SW |
230 | struct ceph_msg *r_reply; |
231 | struct ceph_mds_reply_info_parsed r_reply_info; | |
01deead0 | 232 | struct page *r_locked_page; |
2f2dc053 | 233 | int r_err; |
5b1daecd | 234 | bool r_aborted; |
2f2dc053 | 235 | |
a319bf56 | 236 | unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */ |
2f2dc053 SW |
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 */ | |
240 | ||
241 | /* link unsafe requests to parent directory, for fsync */ | |
242 | struct inode *r_unsafe_dir; | |
243 | struct list_head r_unsafe_dir_item; | |
244 | ||
68cd5b4b YZ |
245 | /* unsafe requests that modify the target inode */ |
246 | struct list_head r_unsafe_target_item; | |
247 | ||
2f2dc053 SW |
248 | struct ceph_mds_session *r_session; |
249 | ||
250 | int r_attempts; /* resend attempts */ | |
251 | int r_num_fwd; /* number of forward attempts */ | |
2f2dc053 | 252 | int r_resend_mds; /* mds to resend to next, if any*/ |
e55b71f8 | 253 | u32 r_sent_on_mseq; /* cap mseq request was sent at*/ |
2f2dc053 | 254 | |
153c8e6b | 255 | struct kref r_kref; |
2f2dc053 SW |
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; | |
9280be24 | 260 | ceph_mds_request_wait_callback_t r_wait_for_completion; |
2f2dc053 | 261 | struct list_head r_unsafe_item; /* per-session unsafe list item */ |
e1518c7c | 262 | bool r_got_unsafe, r_got_safe, r_got_result; |
2f2dc053 SW |
263 | |
264 | bool r_did_prepopulate; | |
fdd4e158 YZ |
265 | long long r_dir_release_cnt; |
266 | long long r_dir_ordered_cnt; | |
267 | int r_readdir_cache_idx; | |
2f2dc053 SW |
268 | u32 r_readdir_offset; |
269 | ||
270 | struct ceph_cap_reservation r_caps_reservation; | |
271 | int r_num_caps; | |
272 | }; | |
273 | ||
10183a69 YZ |
274 | struct ceph_pool_perm { |
275 | struct rb_node node; | |
276 | u32 pool; | |
277 | int perm; | |
278 | }; | |
279 | ||
2f2dc053 SW |
280 | /* |
281 | * mds client state | |
282 | */ | |
283 | struct ceph_mds_client { | |
3d14c5d2 | 284 | struct ceph_fs_client *fsc; |
2f2dc053 SW |
285 | struct mutex mutex; /* all nested structures */ |
286 | ||
287 | struct ceph_mdsmap *mdsmap; | |
f3c60c59 SW |
288 | struct completion safe_umount_waiters; |
289 | wait_queue_head_t session_close_wq; | |
2f2dc053 SW |
290 | struct list_head waiting_for_map; |
291 | ||
292 | struct ceph_mds_session **sessions; /* NULL for mds if no session */ | |
86d8f67b | 293 | atomic_t num_sessions; |
2f2dc053 SW |
294 | int max_sessions; /* len of s_mds_sessions */ |
295 | int stopping; /* true if shutting down */ | |
296 | ||
297 | /* | |
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. | |
303 | */ | |
affbc19a | 304 | u64 last_snap_seq; |
2f2dc053 | 305 | struct rw_semaphore snap_rwsem; |
a105f00c | 306 | struct rb_root snap_realms; |
2f2dc053 SW |
307 | struct list_head snap_empty; |
308 | spinlock_t snap_empty_lock; /* protect snap_empty */ | |
309 | ||
310 | u64 last_tid; /* most recent mds request */ | |
e8a7b8b1 YZ |
311 | u64 oldest_tid; /* oldest incomplete mds request, |
312 | excluding setfilelock requests */ | |
44ca18f2 | 313 | struct rb_root request_tree; /* pending mds requests */ |
2f2dc053 SW |
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; | |
320 | ||
553adfd9 | 321 | u64 last_cap_flush_tid; |
8310b089 | 322 | struct rb_root cap_flush_tree; |
2f2dc053 | 323 | struct list_head cap_dirty; /* inodes with dirty caps */ |
db354052 | 324 | struct list_head cap_dirty_migrating; /* ...that are migration... */ |
2f2dc053 SW |
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; | |
328 | ||
37151668 YS |
329 | /* |
330 | * Cap reservations | |
331 | * | |
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.) | |
337 | * | |
338 | * Reservations are 'owned' by a ceph_cap_reservation context. | |
339 | */ | |
340 | spinlock_t caps_list_lock; | |
341 | struct list_head caps_list; /* unused (reserved or | |
342 | unreserved) */ | |
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 | |
348 | (unreserved) */ | |
2f2dc053 SW |
349 | spinlock_t dentry_lru_lock; |
350 | struct list_head dentry_lru; | |
351 | int num_dentry; | |
10183a69 YZ |
352 | |
353 | struct rw_semaphore pool_perm_rwsem; | |
354 | struct rb_root pool_perm_tree; | |
2f2dc053 SW |
355 | }; |
356 | ||
357 | extern const char *ceph_mds_op_name(int op); | |
358 | ||
359 | extern struct ceph_mds_session * | |
360 | __ceph_lookup_mds_session(struct ceph_mds_client *, int mds); | |
361 | ||
362 | static inline struct ceph_mds_session * | |
363 | ceph_get_mds_session(struct ceph_mds_session *s) | |
364 | { | |
365 | atomic_inc(&s->s_ref); | |
366 | return s; | |
367 | } | |
368 | ||
a687ecaf JS |
369 | extern const char *ceph_session_state_name(int s); |
370 | ||
2f2dc053 SW |
371 | extern void ceph_put_mds_session(struct ceph_mds_session *s); |
372 | ||
373 | extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc, | |
374 | struct ceph_msg *msg, int mds); | |
375 | ||
3d14c5d2 | 376 | extern int ceph_mdsc_init(struct ceph_fs_client *fsc); |
2f2dc053 | 377 | extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc); |
48fec5d0 | 378 | extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc); |
3d14c5d2 | 379 | extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc); |
2f2dc053 SW |
380 | |
381 | extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc); | |
382 | ||
383 | extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, | |
384 | struct inode *inode, | |
2f90b852 | 385 | struct dentry *dn); |
2f2dc053 | 386 | |
167c9e35 | 387 | extern void ceph_invalidate_dir_request(struct ceph_mds_request *req); |
54008399 YZ |
388 | extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req, |
389 | struct inode *dir); | |
2f2dc053 SW |
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, | |
395 | struct inode *dir, | |
396 | struct ceph_mds_request *req); | |
397 | static inline void ceph_mdsc_get_request(struct ceph_mds_request *req) | |
398 | { | |
153c8e6b SW |
399 | kref_get(&req->r_kref); |
400 | } | |
401 | extern void ceph_mdsc_release_request(struct kref *kref); | |
402 | static inline void ceph_mdsc_put_request(struct ceph_mds_request *req) | |
403 | { | |
404 | kref_put(&req->r_kref, ceph_mdsc_release_request); | |
2f2dc053 | 405 | } |
2f2dc053 | 406 | |
3d7ded4d SW |
407 | extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc, |
408 | struct ceph_mds_session *session); | |
409 | ||
2f2dc053 SW |
410 | extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc); |
411 | ||
412 | extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base, | |
413 | int stop_on_nosnap); | |
414 | ||
415 | extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry); | |
416 | extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session, | |
417 | struct inode *inode, | |
418 | struct dentry *dentry, char action, | |
419 | u32 seq); | |
420 | ||
421 | extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, | |
422 | struct ceph_msg *msg); | |
423 | ||
5d72d13c YZ |
424 | extern struct ceph_mds_session * |
425 | ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target); | |
154f42c2 SW |
426 | extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc, |
427 | struct ceph_mds_session *session); | |
428 | ||
2f2dc053 | 429 | #endif |