]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2f2dc053 SW |
2 | #ifndef _FS_CEPH_MDS_CLIENT_H |
3 | #define _FS_CEPH_MDS_CLIENT_H | |
4 | ||
5 | #include <linux/completion.h> | |
153c8e6b | 6 | #include <linux/kref.h> |
2f2dc053 SW |
7 | #include <linux/list.h> |
8 | #include <linux/mutex.h> | |
44ca18f2 | 9 | #include <linux/rbtree.h> |
2f2dc053 | 10 | #include <linux/spinlock.h> |
3997c01d | 11 | #include <linux/refcount.h> |
717e6f28 | 12 | #include <linux/utsname.h> |
2f2dc053 | 13 | |
3d14c5d2 YS |
14 | #include <linux/ceph/types.h> |
15 | #include <linux/ceph/messenger.h> | |
16 | #include <linux/ceph/mdsmap.h> | |
6c4a1915 | 17 | #include <linux/ceph/auth.h> |
2f2dc053 SW |
18 | |
19 | /* | |
20 | * Some lock dependencies: | |
21 | * | |
22 | * session->s_mutex | |
23 | * mdsc->mutex | |
24 | * | |
25 | * mdsc->snap_rwsem | |
26 | * | |
be655596 | 27 | * ci->i_ceph_lock |
2f2dc053 SW |
28 | * mdsc->snap_flush_lock |
29 | * mdsc->cap_delay_lock | |
30 | * | |
31 | */ | |
32 | ||
3d14c5d2 | 33 | struct ceph_fs_client; |
2f2dc053 SW |
34 | struct ceph_cap; |
35 | ||
36 | /* | |
37 | * parsed info about a single inode. pointers are into the encoded | |
38 | * on-wire structures within the mds reply message payload. | |
39 | */ | |
40 | struct ceph_mds_reply_info_in { | |
41 | struct ceph_mds_reply_inode *in; | |
14303d20 | 42 | struct ceph_dir_layout dir_layout; |
2f2dc053 SW |
43 | u32 symlink_len; |
44 | char *symlink; | |
45 | u32 xattr_len; | |
46 | char *xattr_data; | |
fb01d1f8 YZ |
47 | u64 inline_version; |
48 | u32 inline_len; | |
49 | char *inline_data; | |
5ea5c5e0 | 50 | u32 pool_ns_len; |
779fe0fb | 51 | char *pool_ns_data; |
7370e8a4 LH |
52 | u64 max_bytes; |
53 | u64 max_files; | |
2f2dc053 SW |
54 | }; |
55 | ||
2a5beea3 YZ |
56 | struct ceph_mds_reply_dir_entry { |
57 | char *name; | |
58 | u32 name_len; | |
59 | struct ceph_mds_reply_lease *lease; | |
60 | struct ceph_mds_reply_info_in inode; | |
8974eebd | 61 | loff_t offset; |
2a5beea3 YZ |
62 | }; |
63 | ||
2f2dc053 | 64 | /* |
25933abd HS |
65 | * parsed info about an mds reply, including information about |
66 | * either: 1) the target inode and/or its parent directory and dentry, | |
67 | * and directory contents (for readdir results), or | |
68 | * 2) the file range lock info (for fcntl F_GETLK results). | |
2f2dc053 SW |
69 | */ |
70 | struct ceph_mds_reply_info_parsed { | |
71 | struct ceph_mds_reply_head *head; | |
72 | ||
25933abd | 73 | /* trace */ |
2f2dc053 SW |
74 | struct ceph_mds_reply_info_in diri, targeti; |
75 | struct ceph_mds_reply_dirfrag *dirfrag; | |
76 | char *dname; | |
77 | u32 dname_len; | |
78 | struct ceph_mds_reply_lease *dlease; | |
79 | ||
25933abd HS |
80 | /* extra */ |
81 | union { | |
82 | /* for fcntl F_GETLK results */ | |
83 | struct ceph_filelock *filelock_reply; | |
84 | ||
85 | /* for readdir results */ | |
86 | struct { | |
87 | struct ceph_mds_reply_dirfrag *dir_dir; | |
54008399 | 88 | size_t dir_buf_size; |
25933abd | 89 | int dir_nr; |
f3c4ebe6 | 90 | bool dir_end; |
79162547 | 91 | bool dir_complete; |
f3c4ebe6 | 92 | bool hash_order; |
79162547 | 93 | bool offset_hash; |
2a5beea3 | 94 | struct ceph_mds_reply_dir_entry *dir_entries; |
25933abd | 95 | }; |
6e8575fa SL |
96 | |
97 | /* for create results */ | |
98 | struct { | |
99 | bool has_create_ino; | |
100 | u64 ino; | |
101 | }; | |
25933abd | 102 | }; |
2f2dc053 SW |
103 | |
104 | /* encoded blob describing snapshot contexts for certain | |
105 | operations (e.g., open) */ | |
106 | void *snapblob; | |
107 | int snapblob_len; | |
108 | }; | |
109 | ||
110 | ||
111 | /* | |
112 | * cap releases are batched and sent to the MDS en masse. | |
92475f05 JL |
113 | * |
114 | * Account for per-message overhead of mds_cap_release header | |
115 | * and __le32 for osd epoch barrier trailing field. | |
2f2dc053 | 116 | */ |
92475f05 | 117 | #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \ |
2f2dc053 | 118 | sizeof(struct ceph_mds_cap_release)) / \ |
92475f05 | 119 | sizeof(struct ceph_mds_cap_item)) |
2f2dc053 SW |
120 | |
121 | ||
122 | /* | |
123 | * state associated with each MDS<->client session | |
124 | */ | |
125 | enum { | |
126 | CEPH_MDS_SESSION_NEW = 1, | |
127 | CEPH_MDS_SESSION_OPENING = 2, | |
128 | CEPH_MDS_SESSION_OPEN = 3, | |
129 | CEPH_MDS_SESSION_HUNG = 4, | |
130 | CEPH_MDS_SESSION_CLOSING = 5, | |
131 | CEPH_MDS_SESSION_RESTARTING = 6, | |
132 | CEPH_MDS_SESSION_RECONNECTING = 7, | |
fcff415c | 133 | CEPH_MDS_SESSION_REJECTED = 8, |
2f2dc053 SW |
134 | }; |
135 | ||
136 | struct ceph_mds_session { | |
137 | struct ceph_mds_client *s_mdsc; | |
138 | int s_mds; | |
139 | int s_state; | |
140 | unsigned long s_ttl; /* time until mds kills us */ | |
141 | u64 s_seq; /* incoming msg seq # */ | |
142 | struct mutex s_mutex; /* serialize session messages */ | |
143 | ||
144 | struct ceph_connection s_con; | |
145 | ||
6c4a1915 | 146 | struct ceph_auth_handshake s_auth; |
4e7a5dcd | 147 | |
d8fb02ab AE |
148 | /* protected by s_gen_ttl_lock */ |
149 | spinlock_t s_gen_ttl_lock; | |
2f2dc053 SW |
150 | u32 s_cap_gen; /* inc each time we get mds stale msg */ |
151 | unsigned long s_cap_ttl; /* when session caps expire */ | |
d8fb02ab AE |
152 | |
153 | /* protected by s_cap_lock */ | |
154 | spinlock_t s_cap_lock; | |
2f2dc053 | 155 | struct list_head s_caps; /* all caps issued by this session */ |
1de82206 | 156 | struct ceph_cap *s_cap_iterator; |
2f2dc053 SW |
157 | int s_nr_caps, s_trim_caps; |
158 | int s_num_cap_releases; | |
99a9c273 | 159 | int s_cap_reconnect; |
03f4fcb0 | 160 | int s_readonly; |
2f2dc053 | 161 | struct list_head s_cap_releases; /* waiting cap_release messages */ |
1de82206 | 162 | struct work_struct s_cap_release_work; |
2f2dc053 SW |
163 | |
164 | /* protected by mutex */ | |
165 | struct list_head s_cap_flushing; /* inodes w/ flushing caps */ | |
2f2dc053 SW |
166 | unsigned long s_renew_requested; /* last time we sent a renew req */ |
167 | u64 s_renew_seq; | |
168 | ||
3997c01d | 169 | refcount_t s_ref; |
2f2dc053 SW |
170 | struct list_head s_waiting; /* waiting requests */ |
171 | struct list_head s_unsafe; /* unsafe requests */ | |
172 | }; | |
173 | ||
174 | /* | |
175 | * modes of choosing which MDS to send a request to | |
176 | */ | |
177 | enum { | |
178 | USE_ANY_MDS, | |
179 | USE_RANDOM_MDS, | |
180 | USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */ | |
181 | }; | |
182 | ||
183 | struct ceph_mds_request; | |
184 | struct ceph_mds_client; | |
185 | ||
186 | /* | |
187 | * request completion callback | |
188 | */ | |
189 | typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc, | |
190 | struct ceph_mds_request *req); | |
9280be24 YZ |
191 | /* |
192 | * wait for request completion callback | |
193 | */ | |
194 | typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc, | |
195 | struct ceph_mds_request *req); | |
2f2dc053 SW |
196 | |
197 | /* | |
198 | * an in-flight mds request | |
199 | */ | |
200 | struct ceph_mds_request { | |
201 | u64 r_tid; /* transaction id */ | |
44ca18f2 | 202 | struct rb_node r_node; |
37151668 | 203 | struct ceph_mds_client *r_mdsc; |
2f2dc053 SW |
204 | |
205 | int r_op; /* mds op code */ | |
2f2dc053 SW |
206 | |
207 | /* operation on what? */ | |
208 | struct inode *r_inode; /* arg1 */ | |
209 | struct dentry *r_dentry; /* arg1 */ | |
210 | struct dentry *r_old_dentry; /* arg2: rename from or link from */ | |
41b02e1f | 211 | struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */ |
2f2dc053 SW |
212 | char *r_path1, *r_path2; |
213 | struct ceph_vino r_ino1, r_ino2; | |
214 | ||
3dd69aab | 215 | struct inode *r_parent; /* parent dir inode */ |
2f2dc053 SW |
216 | struct inode *r_target_inode; /* resulting inode */ |
217 | ||
bc2de10d JL |
218 | #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */ |
219 | #define CEPH_MDS_R_ABORTED (2) /* call was aborted */ | |
220 | #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */ | |
221 | #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */ | |
222 | #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */ | |
223 | #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */ | |
3dd69aab | 224 | #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */ |
bc2de10d JL |
225 | unsigned long r_req_flags; |
226 | ||
b4556396 SW |
227 | struct mutex r_fill_mutex; |
228 | ||
2f2dc053 SW |
229 | union ceph_mds_request_args r_args; |
230 | int r_fmode; /* file mode, if expecting cap */ | |
ff3d0046 EB |
231 | kuid_t r_uid; |
232 | kgid_t r_gid; | |
b8e69066 | 233 | struct timespec r_stamp; |
2f2dc053 SW |
234 | |
235 | /* for choosing which mds to send this request to */ | |
236 | int r_direct_mode; | |
237 | u32 r_direct_hash; /* choose dir frag based on this dentry hash */ | |
2f2dc053 SW |
238 | |
239 | /* data payload is used for xattr ops */ | |
25e6bae3 | 240 | struct ceph_pagelist *r_pagelist; |
2f2dc053 SW |
241 | |
242 | /* what caps shall we drop? */ | |
243 | int r_inode_drop, r_inode_unless; | |
244 | int r_dentry_drop, r_dentry_unless; | |
245 | int r_old_dentry_drop, r_old_dentry_unless; | |
246 | struct inode *r_old_inode; | |
247 | int r_old_inode_drop, r_old_inode_unless; | |
248 | ||
249 | struct ceph_msg *r_request; /* original request */ | |
e979cf50 | 250 | int r_request_release_offset; |
2f2dc053 SW |
251 | struct ceph_msg *r_reply; |
252 | struct ceph_mds_reply_info_parsed r_reply_info; | |
01deead0 | 253 | struct page *r_locked_page; |
2f2dc053 SW |
254 | int r_err; |
255 | ||
a319bf56 | 256 | unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */ |
2f2dc053 SW |
257 | unsigned long r_started; /* start time to measure timeout against */ |
258 | unsigned long r_request_started; /* start time for mds request only, | |
259 | used to measure lease durations */ | |
260 | ||
261 | /* link unsafe requests to parent directory, for fsync */ | |
262 | struct inode *r_unsafe_dir; | |
263 | struct list_head r_unsafe_dir_item; | |
264 | ||
68cd5b4b YZ |
265 | /* unsafe requests that modify the target inode */ |
266 | struct list_head r_unsafe_target_item; | |
267 | ||
2f2dc053 SW |
268 | struct ceph_mds_session *r_session; |
269 | ||
270 | int r_attempts; /* resend attempts */ | |
271 | int r_num_fwd; /* number of forward attempts */ | |
2f2dc053 | 272 | int r_resend_mds; /* mds to resend to next, if any*/ |
e55b71f8 | 273 | u32 r_sent_on_mseq; /* cap mseq request was sent at*/ |
2f2dc053 | 274 | |
153c8e6b | 275 | struct kref r_kref; |
2f2dc053 SW |
276 | struct list_head r_wait; |
277 | struct completion r_completion; | |
278 | struct completion r_safe_completion; | |
279 | ceph_mds_request_callback_t r_callback; | |
9280be24 | 280 | ceph_mds_request_wait_callback_t r_wait_for_completion; |
2f2dc053 | 281 | struct list_head r_unsafe_item; /* per-session unsafe list item */ |
2f2dc053 | 282 | |
fdd4e158 YZ |
283 | long long r_dir_release_cnt; |
284 | long long r_dir_ordered_cnt; | |
285 | int r_readdir_cache_idx; | |
2f2dc053 SW |
286 | u32 r_readdir_offset; |
287 | ||
288 | struct ceph_cap_reservation r_caps_reservation; | |
289 | int r_num_caps; | |
290 | }; | |
291 | ||
10183a69 YZ |
292 | struct ceph_pool_perm { |
293 | struct rb_node node; | |
10183a69 | 294 | int perm; |
7627151e | 295 | s64 pool; |
779fe0fb YZ |
296 | size_t pool_ns_len; |
297 | char pool_ns[]; | |
10183a69 YZ |
298 | }; |
299 | ||
2f2dc053 SW |
300 | /* |
301 | * mds client state | |
302 | */ | |
303 | struct ceph_mds_client { | |
3d14c5d2 | 304 | struct ceph_fs_client *fsc; |
2f2dc053 SW |
305 | struct mutex mutex; /* all nested structures */ |
306 | ||
307 | struct ceph_mdsmap *mdsmap; | |
f3c60c59 SW |
308 | struct completion safe_umount_waiters; |
309 | wait_queue_head_t session_close_wq; | |
2f2dc053 | 310 | struct list_head waiting_for_map; |
430afbad | 311 | int mdsmap_err; |
2f2dc053 SW |
312 | |
313 | struct ceph_mds_session **sessions; /* NULL for mds if no session */ | |
86d8f67b | 314 | atomic_t num_sessions; |
2f2dc053 SW |
315 | int max_sessions; /* len of s_mds_sessions */ |
316 | int stopping; /* true if shutting down */ | |
317 | ||
60147660 LH |
318 | atomic64_t quotarealms_count; /* # realms with quota */ |
319 | ||
2f2dc053 SW |
320 | /* |
321 | * snap_rwsem will cover cap linkage into snaprealms, and | |
322 | * realm snap contexts. (later, we can do per-realm snap | |
323 | * contexts locks..) the empty list contains realms with no | |
324 | * references (implying they contain no inodes with caps) that | |
325 | * should be destroyed. | |
326 | */ | |
affbc19a | 327 | u64 last_snap_seq; |
2f2dc053 | 328 | struct rw_semaphore snap_rwsem; |
a105f00c | 329 | struct rb_root snap_realms; |
2f2dc053 SW |
330 | struct list_head snap_empty; |
331 | spinlock_t snap_empty_lock; /* protect snap_empty */ | |
332 | ||
333 | u64 last_tid; /* most recent mds request */ | |
e8a7b8b1 YZ |
334 | u64 oldest_tid; /* oldest incomplete mds request, |
335 | excluding setfilelock requests */ | |
44ca18f2 | 336 | struct rb_root request_tree; /* pending mds requests */ |
2f2dc053 SW |
337 | struct delayed_work delayed_work; /* delayed work */ |
338 | unsigned long last_renew_caps; /* last time we renewed our caps */ | |
339 | struct list_head cap_delay_list; /* caps with delayed release */ | |
340 | spinlock_t cap_delay_lock; /* protects cap_delay_list */ | |
341 | struct list_head snap_flush_list; /* cap_snaps ready to flush */ | |
342 | spinlock_t snap_flush_lock; | |
343 | ||
553adfd9 | 344 | u64 last_cap_flush_tid; |
e4500b5e | 345 | struct list_head cap_flush_list; |
2f2dc053 | 346 | struct list_head cap_dirty; /* inodes with dirty caps */ |
db354052 | 347 | struct list_head cap_dirty_migrating; /* ...that are migration... */ |
2f2dc053 SW |
348 | int num_cap_flushing; /* # caps we are flushing */ |
349 | spinlock_t cap_dirty_lock; /* protects above items */ | |
350 | wait_queue_head_t cap_flushing_wq; | |
351 | ||
37151668 YS |
352 | /* |
353 | * Cap reservations | |
354 | * | |
355 | * Maintain a global pool of preallocated struct ceph_caps, referenced | |
356 | * by struct ceph_caps_reservations. This ensures that we preallocate | |
357 | * memory needed to successfully process an MDS response. (If an MDS | |
358 | * sends us cap information and we fail to process it, we will have | |
359 | * problems due to the client and MDS being out of sync.) | |
360 | * | |
361 | * Reservations are 'owned' by a ceph_cap_reservation context. | |
362 | */ | |
363 | spinlock_t caps_list_lock; | |
364 | struct list_head caps_list; /* unused (reserved or | |
365 | unreserved) */ | |
366 | int caps_total_count; /* total caps allocated */ | |
367 | int caps_use_count; /* in use */ | |
368 | int caps_reserve_count; /* unused, reserved */ | |
369 | int caps_avail_count; /* unused, unreserved */ | |
370 | int caps_min_count; /* keep at least this many | |
371 | (unreserved) */ | |
2f2dc053 SW |
372 | spinlock_t dentry_lru_lock; |
373 | struct list_head dentry_lru; | |
374 | int num_dentry; | |
10183a69 YZ |
375 | |
376 | struct rw_semaphore pool_perm_rwsem; | |
377 | struct rb_root pool_perm_tree; | |
717e6f28 YZ |
378 | |
379 | char nodename[__NEW_UTS_LEN + 1]; | |
2f2dc053 SW |
380 | }; |
381 | ||
382 | extern const char *ceph_mds_op_name(int op); | |
383 | ||
384 | extern struct ceph_mds_session * | |
385 | __ceph_lookup_mds_session(struct ceph_mds_client *, int mds); | |
386 | ||
387 | static inline struct ceph_mds_session * | |
388 | ceph_get_mds_session(struct ceph_mds_session *s) | |
389 | { | |
3997c01d | 390 | refcount_inc(&s->s_ref); |
2f2dc053 SW |
391 | return s; |
392 | } | |
393 | ||
a687ecaf JS |
394 | extern const char *ceph_session_state_name(int s); |
395 | ||
2f2dc053 SW |
396 | extern void ceph_put_mds_session(struct ceph_mds_session *s); |
397 | ||
398 | extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc, | |
399 | struct ceph_msg *msg, int mds); | |
400 | ||
3d14c5d2 | 401 | extern int ceph_mdsc_init(struct ceph_fs_client *fsc); |
2f2dc053 | 402 | extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc); |
48fec5d0 | 403 | extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc); |
3d14c5d2 | 404 | extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc); |
2f2dc053 SW |
405 | |
406 | extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc); | |
407 | ||
167c9e35 | 408 | extern void ceph_invalidate_dir_request(struct ceph_mds_request *req); |
54008399 YZ |
409 | extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req, |
410 | struct inode *dir); | |
2f2dc053 SW |
411 | extern struct ceph_mds_request * |
412 | ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode); | |
413 | extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, | |
414 | struct ceph_mds_request *req); | |
415 | extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc, | |
416 | struct inode *dir, | |
417 | struct ceph_mds_request *req); | |
418 | static inline void ceph_mdsc_get_request(struct ceph_mds_request *req) | |
419 | { | |
153c8e6b SW |
420 | kref_get(&req->r_kref); |
421 | } | |
422 | extern void ceph_mdsc_release_request(struct kref *kref); | |
423 | static inline void ceph_mdsc_put_request(struct ceph_mds_request *req) | |
424 | { | |
425 | kref_put(&req->r_kref, ceph_mdsc_release_request); | |
2f2dc053 | 426 | } |
2f2dc053 | 427 | |
1de82206 YZ |
428 | extern void __ceph_queue_cap_release(struct ceph_mds_session *session, |
429 | struct ceph_cap *cap); | |
430 | extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc, | |
431 | struct ceph_mds_session *session); | |
2f2dc053 SW |
432 | extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc); |
433 | ||
434 | extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base, | |
435 | int stop_on_nosnap); | |
436 | ||
437 | extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry); | |
438 | extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session, | |
439 | struct inode *inode, | |
440 | struct dentry *dentry, char action, | |
441 | u32 seq); | |
442 | ||
430afbad YZ |
443 | extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, |
444 | struct ceph_msg *msg); | |
445 | extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, | |
446 | struct ceph_msg *msg); | |
2f2dc053 | 447 | |
5d72d13c YZ |
448 | extern struct ceph_mds_session * |
449 | ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target); | |
154f42c2 SW |
450 | extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc, |
451 | struct ceph_mds_session *session); | |
452 | ||
2f2dc053 | 453 | #endif |