1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _FS_CEPH_MDS_CLIENT_H
3 #define _FS_CEPH_MDS_CLIENT_H
4
5 #include <linux/completion.h>
6 #include <linux/kref.h>
7 #include <linux/list.h>
8 #include <linux/mutex.h>
9 #include <linux/rbtree.h>
10 #include <linux/spinlock.h>
11 #include <linux/refcount.h>
12 #include <linux/utsname.h>
13 #include <linux/ktime.h>
14
15 #include <linux/ceph/types.h>
16 #include <linux/ceph/messenger.h>
17 #include <linux/ceph/mdsmap.h>
18 #include <linux/ceph/auth.h>
19
20 #include "metric.h"
21 #include "super.h"
22
23 /* The first 8 bits are reserved for old ceph releases */
24 enum ceph_feature_type {
25 CEPHFS_FEATURE_MIMIC = 8,
26 CEPHFS_FEATURE_REPLY_ENCODING,
27 CEPHFS_FEATURE_RECLAIM_CLIENT,
28 CEPHFS_FEATURE_LAZY_CAP_WANTED,
29 CEPHFS_FEATURE_MULTI_RECONNECT,
30 CEPHFS_FEATURE_DELEG_INO,
31 CEPHFS_FEATURE_METRIC_COLLECT,
32
33 CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT,
34 };
35
36 #define CEPHFS_FEATURES_CLIENT_SUPPORTED { \
37 0, 1, 2, 3, 4, 5, 6, 7, \
38 CEPHFS_FEATURE_MIMIC, \
39 CEPHFS_FEATURE_REPLY_ENCODING, \
40 CEPHFS_FEATURE_LAZY_CAP_WANTED, \
41 CEPHFS_FEATURE_MULTI_RECONNECT, \
42 CEPHFS_FEATURE_DELEG_INO, \
43 CEPHFS_FEATURE_METRIC_COLLECT, \
44 }
45 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
46
47 /*
48 * Some lock dependencies:
49 *
50 * session->s_mutex
51 * mdsc->mutex
52 *
53 * mdsc->snap_rwsem
54 *
55 * ci->i_ceph_lock
56 * mdsc->snap_flush_lock
57 * mdsc->cap_delay_lock
58 *
59 */
60
61 struct ceph_fs_client;
62 struct ceph_cap;
63
64 /*
65 * parsed info about a single inode. pointers are into the encoded
66 * on-wire structures within the mds reply message payload.
67 */
68 struct ceph_mds_reply_info_in {
69 struct ceph_mds_reply_inode *in;
70 struct ceph_dir_layout dir_layout;
71 u32 symlink_len;
72 char *symlink;
73 u32 xattr_len;
74 char *xattr_data;
75 u64 inline_version;
76 u32 inline_len;
77 char *inline_data;
78 u32 pool_ns_len;
79 char *pool_ns_data;
80 u64 max_bytes;
81 u64 max_files;
82 s32 dir_pin;
83 struct ceph_timespec btime;
84 struct ceph_timespec snap_btime;
85 u64 rsnaps;
86 u64 change_attr;
87 };
88
89 struct ceph_mds_reply_dir_entry {
90 char *name;
91 u32 name_len;
92 struct ceph_mds_reply_lease *lease;
93 struct ceph_mds_reply_info_in inode;
94 loff_t offset;
95 };
96
97 /*
98 * parsed info about an mds reply, including information about
99 * either: 1) the target inode and/or its parent directory and dentry,
100 * and directory contents (for readdir results), or
101 * 2) the file range lock info (for fcntl F_GETLK results).
102 */
103 struct ceph_mds_reply_info_parsed {
104 struct ceph_mds_reply_head *head;
105
106 /* trace */
107 struct ceph_mds_reply_info_in diri, targeti;
108 struct ceph_mds_reply_dirfrag *dirfrag;
109 char *dname;
110 u32 dname_len;
111 struct ceph_mds_reply_lease *dlease;
112
113 /* extra */
114 union {
115 /* for fcntl F_GETLK results */
116 struct ceph_filelock *filelock_reply;
117
118 /* for readdir results */
119 struct {
120 struct ceph_mds_reply_dirfrag *dir_dir;
121 size_t dir_buf_size;
122 int dir_nr;
123 bool dir_end;
124 bool dir_complete;
125 bool hash_order;
126 bool offset_hash;
127 struct ceph_mds_reply_dir_entry *dir_entries;
128 };
129
130 /* for create results */
131 struct {
132 bool has_create_ino;
133 u64 ino;
134 };
135 };
136
137 /* encoded blob describing snapshot contexts for certain
138 operations (e.g., open) */
139 void *snapblob;
140 int snapblob_len;
141 };
142
143
144 /*
145 * cap releases are batched and sent to the MDS en masse.
146 *
147 * Account for per-message overhead of mds_cap_release header
148 * and __le32 for osd epoch barrier trailing field.
149 */
150 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
151 sizeof(struct ceph_mds_cap_release)) / \
152 sizeof(struct ceph_mds_cap_item))
153
154
155 /*
156 * state associated with each MDS<->client session
157 */
158 enum {
159 CEPH_MDS_SESSION_NEW = 1,
160 CEPH_MDS_SESSION_OPENING = 2,
161 CEPH_MDS_SESSION_OPEN = 3,
162 CEPH_MDS_SESSION_HUNG = 4,
163 CEPH_MDS_SESSION_RESTARTING = 5,
164 CEPH_MDS_SESSION_RECONNECTING = 6,
165 CEPH_MDS_SESSION_CLOSING = 7,
166 CEPH_MDS_SESSION_CLOSED = 8,
167 CEPH_MDS_SESSION_REJECTED = 9,
168 };
169
170 struct ceph_mds_session {
171 struct ceph_mds_client *s_mdsc;
172 int s_mds;
173 int s_state;
174 unsigned long s_ttl; /* time until mds kills us */
175 unsigned long s_features;
176 u64 s_seq; /* incoming msg seq # */
177 struct mutex s_mutex; /* serialize session messages */
178
179 struct ceph_connection s_con;
180
181 struct ceph_auth_handshake s_auth;
182
183 atomic_t s_cap_gen; /* inc each time we get mds stale msg */
184 unsigned long s_cap_ttl; /* when session caps expire. protected by s_mutex */
185
186 /* protected by s_cap_lock */
187 spinlock_t s_cap_lock;
188 refcount_t s_ref;
189 struct list_head s_caps; /* all caps issued by this session */
190 struct ceph_cap *s_cap_iterator;
191 int s_nr_caps;
192 int s_num_cap_releases;
193 int s_cap_reconnect;
194 int s_readonly;
195 struct list_head s_cap_releases; /* waiting cap_release messages */
196 struct work_struct s_cap_release_work;
197
198 /* See ceph_inode_info->i_dirty_item. */
199 struct list_head s_cap_dirty; /* inodes w/ dirty caps */
200
201 /* See ceph_inode_info->i_flushing_item. */
202 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
203
204 unsigned long s_renew_requested; /* last time we sent a renew req */
205 u64 s_renew_seq;
206
207 struct list_head s_waiting; /* waiting requests */
208 struct list_head s_unsafe; /* unsafe requests */
209 struct xarray s_delegated_inos;
210 };
211
212 /*
213 * modes of choosing which MDS to send a request to
214 */
215 enum {
216 USE_ANY_MDS,
217 USE_RANDOM_MDS,
218 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
219 };
220
221 struct ceph_mds_request;
222 struct ceph_mds_client;
223
224 /*
225 * request completion callback
226 */
227 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
228 struct ceph_mds_request *req);
229 /*
230 * wait for request completion callback
231 */
232 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
233 struct ceph_mds_request *req);
234
235 /*
236 * an in-flight mds request
237 */
238 struct ceph_mds_request {
239 u64 r_tid; /* transaction id */
240 struct rb_node r_node;
241 struct ceph_mds_client *r_mdsc;
242
243 struct kref r_kref;
244 int r_op; /* mds op code */
245
246 /* operation on what? */
247 struct inode *r_inode; /* arg1 */
248 struct dentry *r_dentry; /* arg1 */
249 struct dentry *r_old_dentry; /* arg2: rename from or link from */
250 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
251 char *r_path1, *r_path2;
252 struct ceph_vino r_ino1, r_ino2;
253
254 struct inode *r_parent; /* parent dir inode */
255 struct inode *r_target_inode; /* resulting inode */
256
257 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
258 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
259 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
260 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
261 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
262 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
263 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
264 #define CEPH_MDS_R_ASYNC (8) /* async request */
265 unsigned long r_req_flags;
266
267 struct mutex r_fill_mutex;
268
269 union ceph_mds_request_args r_args;
270 int r_fmode; /* file mode, if expecting cap */
271 const struct cred *r_cred;
272 int r_request_release_offset;
273 struct timespec64 r_stamp;
274
275 /* for choosing which mds to send this request to */
276 int r_direct_mode;
277 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
278
279 /* data payload is used for xattr ops */
280 struct ceph_pagelist *r_pagelist;
281
282 /* what caps shall we drop? */
283 int r_inode_drop, r_inode_unless;
284 int r_dentry_drop, r_dentry_unless;
285 int r_old_dentry_drop, r_old_dentry_unless;
286 struct inode *r_old_inode;
287 int r_old_inode_drop, r_old_inode_unless;
288
289 struct ceph_msg *r_request; /* original request */
290 struct ceph_msg *r_reply;
291 struct ceph_mds_reply_info_parsed r_reply_info;
292 int r_err;
293
294
295 struct page *r_locked_page;
296 int r_dir_caps;
297 int r_num_caps;
298 u32 r_readdir_offset;
299
300 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
301 unsigned long r_started; /* start time to measure timeout against */
302 unsigned long r_start_latency; /* start time to measure latency */
303 unsigned long r_end_latency; /* finish time to measure latency */
304 unsigned long r_request_started; /* start time for mds request only,
305 used to measure lease durations */
306
307 /* link unsafe requests to parent directory, for fsync */
308 struct inode *r_unsafe_dir;
309 struct list_head r_unsafe_dir_item;
310
311 /* unsafe requests that modify the target inode */
312 struct list_head r_unsafe_target_item;
313
314 struct ceph_mds_session *r_session;
315
316 int r_attempts; /* resend attempts */
317 int r_num_fwd; /* number of forward attempts */
318 int r_resend_mds; /* mds to resend to next, if any*/
319 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
320 u64 r_deleg_ino;
321
322 struct list_head r_wait;
323 struct completion r_completion;
324 struct completion r_safe_completion;
325 ceph_mds_request_callback_t r_callback;
326 ceph_mds_request_wait_callback_t r_wait_for_completion;
327 struct list_head r_unsafe_item; /* per-session unsafe list item */
328
329 long long r_dir_release_cnt;
330 long long r_dir_ordered_cnt;
331 int r_readdir_cache_idx;
332
333 struct ceph_cap_reservation r_caps_reservation;
334 };
335
336 struct ceph_pool_perm {
337 struct rb_node node;
338 int perm;
339 s64 pool;
340 size_t pool_ns_len;
341 char pool_ns[];
342 };
343
344 struct ceph_snapid_map {
345 struct rb_node node;
346 struct list_head lru;
347 atomic_t ref;
348 u64 snap;
349 dev_t dev;
350 unsigned long last_used;
351 };
352
353 /*
354 * node for list of quotarealm inodes that are not visible from the filesystem
355 * mountpoint, but required to handle, e.g. quotas.
356 */
357 struct ceph_quotarealm_inode {
358 struct rb_node node;
359 u64 ino;
360 unsigned long timeout; /* last time a lookup failed for this inode */
361 struct mutex mutex;
362 struct inode *inode;
363 };
364
365 struct cap_wait {
366 struct list_head list;
367 u64 ino;
368 pid_t tgid;
369 int need;
370 int want;
371 };
372
373 enum {
374 CEPH_MDSC_STOPPING_BEGIN = 1,
375 CEPH_MDSC_STOPPING_FLUSHED = 2,
376 };
377
378 /*
379 * mds client state
380 */
381 struct ceph_mds_client {
382 struct ceph_fs_client *fsc;
383 struct mutex mutex; /* all nested structures */
384
385 struct ceph_mdsmap *mdsmap;
386 struct completion safe_umount_waiters;
387 wait_queue_head_t session_close_wq;
388 struct list_head waiting_for_map;
389 int mdsmap_err;
390
391 struct ceph_mds_session **sessions; /* NULL for mds if no session */
392 atomic_t num_sessions;
393 int max_sessions; /* len of sessions array */
394 int stopping; /* true if shutting down */
395
396 atomic64_t quotarealms_count; /* # realms with quota */
397 /*
398 * We keep a list of inodes we don't see in the mountpoint but that we
399 * need to track quota realms.
400 */
401 struct rb_root quotarealms_inodes;
402 struct mutex quotarealms_inodes_mutex;
403
404 /*
405 * snap_rwsem will cover cap linkage into snaprealms, and
406 * realm snap contexts. (later, we can do per-realm snap
407 * contexts locks..) the empty list contains realms with no
408 * references (implying they contain no inodes with caps) that
409 * should be destroyed.
410 */
411 u64 last_snap_seq;
412 struct rw_semaphore snap_rwsem;
413 struct rb_root snap_realms;
414 struct list_head snap_empty;
415 int num_snap_realms;
416 spinlock_t snap_empty_lock; /* protect snap_empty */
417
418 u64 last_tid; /* most recent mds request */
419 u64 oldest_tid; /* oldest incomplete mds request,
420 excluding setfilelock requests */
421 struct rb_root request_tree; /* pending mds requests */
422 struct delayed_work delayed_work; /* delayed work */
423 unsigned long last_renew_caps; /* last time we renewed our caps */
424 struct list_head cap_delay_list; /* caps with delayed release */
425 spinlock_t cap_delay_lock; /* protects cap_delay_list */
426 struct list_head snap_flush_list; /* cap_snaps ready to flush */
427 spinlock_t snap_flush_lock;
428
429 u64 last_cap_flush_tid;
430 struct list_head cap_flush_list;
431 struct list_head cap_dirty_migrating; /* ...that are migration... */
432 int num_cap_flushing; /* # caps we are flushing */
433 spinlock_t cap_dirty_lock; /* protects above items */
434 wait_queue_head_t cap_flushing_wq;
435
436 struct work_struct cap_reclaim_work;
437 atomic_t cap_reclaim_pending;
438
439 /*
440 * Cap reservations
441 *
442 * Maintain a global pool of preallocated struct ceph_caps, referenced
443 * by struct ceph_caps_reservations. This ensures that we preallocate
444 * memory needed to successfully process an MDS response. (If an MDS
445 * sends us cap information and we fail to process it, we will have
446 * problems due to the client and MDS being out of sync.)
447 *
448 * Reservations are 'owned' by a ceph_cap_reservation context.
449 */
450 spinlock_t caps_list_lock;
451 struct list_head caps_list; /* unused (reserved or
452 unreserved) */
453 struct list_head cap_wait_list;
454 int caps_total_count; /* total caps allocated */
455 int caps_use_count; /* in use */
456 int caps_use_max; /* max used caps */
457 int caps_reserve_count; /* unused, reserved */
458 int caps_avail_count; /* unused, unreserved */
459 int caps_min_count; /* keep at least this many
460 (unreserved) */
461 spinlock_t dentry_list_lock;
462 struct list_head dentry_leases; /* fifo list */
463 struct list_head dentry_dir_leases; /* lru list */
464
465 struct ceph_client_metric metric;
466
467 spinlock_t snapid_map_lock;
468 struct rb_root snapid_map_tree;
469 struct list_head snapid_map_lru;
470
471 struct rw_semaphore pool_perm_rwsem;
472 struct rb_root pool_perm_tree;
473
474 char nodename[__NEW_UTS_LEN + 1];
475 };
476
477 extern const char *ceph_mds_op_name(int op);
478
479 extern bool check_session_state(struct ceph_mds_session *s);
480 void inc_session_sequence(struct ceph_mds_session *s);
481
482 extern struct ceph_mds_session *
483 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
484
485 extern const char *ceph_session_state_name(int s);
486
487 extern struct ceph_mds_session *
488 ceph_get_mds_session(struct ceph_mds_session *s);
489 extern void ceph_put_mds_session(struct ceph_mds_session *s);
490
491 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
492 struct ceph_msg *msg, int mds);
493
494 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
495 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
496 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
497 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
498
499 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
500
501 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
502 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
503 struct inode *dir);
504 extern struct ceph_mds_request *
505 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
506 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
507 struct inode *dir,
508 struct ceph_mds_request *req);
509 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
510 struct inode *dir,
511 struct ceph_mds_request *req);
512 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
513 extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
ceph_mdsc_get_request(struct ceph_mds_request * req)514 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
515 {
516 kref_get(&req->r_kref);
517 }
518 extern void ceph_mdsc_release_request(struct kref *kref);
ceph_mdsc_put_request(struct ceph_mds_request * req)519 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
520 {
521 kref_put(&req->r_kref, ceph_mdsc_release_request);
522 }
523
524 extern void send_flush_mdlog(struct ceph_mds_session *s);
525 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
526 void (*cb)(struct ceph_mds_session *),
527 bool check_state);
528 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
529 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
530 struct ceph_cap *cap);
531 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
532 struct ceph_mds_session *session);
533 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
534 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
535 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
536 int (*cb)(struct inode *,
537 struct ceph_cap *, void *),
538 void *arg);
539 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
540
ceph_mdsc_free_path(char * path,int len)541 static inline void ceph_mdsc_free_path(char *path, int len)
542 {
543 if (!IS_ERR_OR_NULL(path))
544 __putname(path - (PATH_MAX - 1 - len));
545 }
546
547 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
548 int stop_on_nosnap);
549
550 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
551 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
552 struct dentry *dentry, char action,
553 u32 seq);
554
555 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
556 struct ceph_msg *msg);
557 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
558 struct ceph_msg *msg);
559
560 extern struct ceph_mds_session *
561 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
562 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
563 struct ceph_mds_session *session);
564
565 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
566 struct ceph_mds_session *session,
567 int max_caps);
568
ceph_wait_on_async_create(struct inode * inode)569 static inline int ceph_wait_on_async_create(struct inode *inode)
570 {
571 struct ceph_inode_info *ci = ceph_inode(inode);
572
573 return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
574 TASK_INTERRUPTIBLE);
575 }
576
577 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
578 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
579 #endif
580