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 CEPHFS_FEATURE_ALTERNATE_NAME,
33 CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
34 CEPHFS_FEATURE_OP_GETVXATTR,
35
36 CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_OP_GETVXATTR,
37 };
38
39 #define CEPHFS_FEATURES_CLIENT_SUPPORTED { \
40 0, 1, 2, 3, 4, 5, 6, 7, \
41 CEPHFS_FEATURE_MIMIC, \
42 CEPHFS_FEATURE_REPLY_ENCODING, \
43 CEPHFS_FEATURE_LAZY_CAP_WANTED, \
44 CEPHFS_FEATURE_MULTI_RECONNECT, \
45 CEPHFS_FEATURE_DELEG_INO, \
46 CEPHFS_FEATURE_METRIC_COLLECT, \
47 CEPHFS_FEATURE_NOTIFY_SESSION_STATE, \
48 CEPHFS_FEATURE_OP_GETVXATTR, \
49 }
50
51 /*
52 * Some lock dependencies:
53 *
54 * session->s_mutex
55 * mdsc->mutex
56 *
57 * mdsc->snap_rwsem
58 *
59 * ci->i_ceph_lock
60 * mdsc->snap_flush_lock
61 * mdsc->cap_delay_lock
62 *
63 */
64
65 struct ceph_fs_client;
66 struct ceph_cap;
67
68 /*
69 * parsed info about a single inode. pointers are into the encoded
70 * on-wire structures within the mds reply message payload.
71 */
72 struct ceph_mds_reply_info_in {
73 struct ceph_mds_reply_inode *in;
74 struct ceph_dir_layout dir_layout;
75 u32 symlink_len;
76 char *symlink;
77 u32 xattr_len;
78 char *xattr_data;
79 u64 inline_version;
80 u32 inline_len;
81 char *inline_data;
82 u32 pool_ns_len;
83 char *pool_ns_data;
84 u64 max_bytes;
85 u64 max_files;
86 s32 dir_pin;
87 struct ceph_timespec btime;
88 struct ceph_timespec snap_btime;
89 u64 rsnaps;
90 u64 change_attr;
91 };
92
93 struct ceph_mds_reply_dir_entry {
94 char *name;
95 u32 name_len;
96 struct ceph_mds_reply_lease *lease;
97 struct ceph_mds_reply_info_in inode;
98 loff_t offset;
99 };
100
101 struct ceph_mds_reply_xattr {
102 char *xattr_value;
103 size_t xattr_value_len;
104 };
105
106 /*
107 * parsed info about an mds reply, including information about
108 * either: 1) the target inode and/or its parent directory and dentry,
109 * and directory contents (for readdir results), or
110 * 2) the file range lock info (for fcntl F_GETLK results).
111 */
112 struct ceph_mds_reply_info_parsed {
113 struct ceph_mds_reply_head *head;
114
115 /* trace */
116 struct ceph_mds_reply_info_in diri, targeti;
117 struct ceph_mds_reply_dirfrag *dirfrag;
118 char *dname;
119 u32 dname_len;
120 struct ceph_mds_reply_lease *dlease;
121 struct ceph_mds_reply_xattr xattr_info;
122
123 /* extra */
124 union {
125 /* for fcntl F_GETLK results */
126 struct ceph_filelock *filelock_reply;
127
128 /* for readdir results */
129 struct {
130 struct ceph_mds_reply_dirfrag *dir_dir;
131 size_t dir_buf_size;
132 int dir_nr;
133 bool dir_end;
134 bool dir_complete;
135 bool hash_order;
136 bool offset_hash;
137 struct ceph_mds_reply_dir_entry *dir_entries;
138 };
139
140 /* for create results */
141 struct {
142 bool has_create_ino;
143 u64 ino;
144 };
145 };
146
147 /* encoded blob describing snapshot contexts for certain
148 operations (e.g., open) */
149 void *snapblob;
150 int snapblob_len;
151 };
152
153
154 /*
155 * cap releases are batched and sent to the MDS en masse.
156 *
157 * Account for per-message overhead of mds_cap_release header
158 * and __le32 for osd epoch barrier trailing field.
159 */
160 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
161 sizeof(struct ceph_mds_cap_release)) / \
162 sizeof(struct ceph_mds_cap_item))
163
164
165 /*
166 * state associated with each MDS<->client session
167 */
168 enum {
169 CEPH_MDS_SESSION_NEW = 1,
170 CEPH_MDS_SESSION_OPENING = 2,
171 CEPH_MDS_SESSION_OPEN = 3,
172 CEPH_MDS_SESSION_HUNG = 4,
173 CEPH_MDS_SESSION_RESTARTING = 5,
174 CEPH_MDS_SESSION_RECONNECTING = 6,
175 CEPH_MDS_SESSION_CLOSING = 7,
176 CEPH_MDS_SESSION_CLOSED = 8,
177 CEPH_MDS_SESSION_REJECTED = 9,
178 };
179
180 struct ceph_mds_session {
181 struct ceph_mds_client *s_mdsc;
182 int s_mds;
183 int s_state;
184 unsigned long s_ttl; /* time until mds kills us */
185 unsigned long s_features;
186 u64 s_seq; /* incoming msg seq # */
187 struct mutex s_mutex; /* serialize session messages */
188
189 struct ceph_connection s_con;
190
191 struct ceph_auth_handshake s_auth;
192
193 atomic_t s_cap_gen; /* inc each time we get mds stale msg */
194 unsigned long s_cap_ttl; /* when session caps expire. protected by s_mutex */
195
196 /* protected by s_cap_lock */
197 spinlock_t s_cap_lock;
198 refcount_t s_ref;
199 struct list_head s_caps; /* all caps issued by this session */
200 struct ceph_cap *s_cap_iterator;
201 int s_nr_caps;
202 int s_num_cap_releases;
203 int s_cap_reconnect;
204 int s_readonly;
205 struct list_head s_cap_releases; /* waiting cap_release messages */
206 struct work_struct s_cap_release_work;
207
208 /* See ceph_inode_info->i_dirty_item. */
209 struct list_head s_cap_dirty; /* inodes w/ dirty caps */
210
211 /* See ceph_inode_info->i_flushing_item. */
212 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
213
214 unsigned long s_renew_requested; /* last time we sent a renew req */
215 u64 s_renew_seq;
216
217 struct list_head s_waiting; /* waiting requests */
218 struct list_head s_unsafe; /* unsafe requests */
219 struct xarray s_delegated_inos;
220 };
221
222 /*
223 * modes of choosing which MDS to send a request to
224 */
225 enum {
226 USE_ANY_MDS,
227 USE_RANDOM_MDS,
228 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
229 };
230
231 struct ceph_mds_request;
232 struct ceph_mds_client;
233
234 /*
235 * request completion callback
236 */
237 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
238 struct ceph_mds_request *req);
239 /*
240 * wait for request completion callback
241 */
242 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
243 struct ceph_mds_request *req);
244
245 /*
246 * an in-flight mds request
247 */
248 struct ceph_mds_request {
249 u64 r_tid; /* transaction id */
250 struct rb_node r_node;
251 struct ceph_mds_client *r_mdsc;
252
253 struct kref r_kref;
254 int r_op; /* mds op code */
255
256 /* operation on what? */
257 struct inode *r_inode; /* arg1 */
258 struct dentry *r_dentry; /* arg1 */
259 struct dentry *r_old_dentry; /* arg2: rename from or link from */
260 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
261 char *r_path1, *r_path2;
262 struct ceph_vino r_ino1, r_ino2;
263
264 struct inode *r_parent; /* parent dir inode */
265 struct inode *r_target_inode; /* resulting inode */
266
267 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
268 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
269 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
270 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
271 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
272 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
273 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
274 #define CEPH_MDS_R_ASYNC (8) /* async request */
275 unsigned long r_req_flags;
276
277 struct mutex r_fill_mutex;
278
279 union ceph_mds_request_args r_args;
280 int r_fmode; /* file mode, if expecting cap */
281 int r_request_release_offset;
282 const struct cred *r_cred;
283 struct timespec64 r_stamp;
284
285 /* for choosing which mds to send this request to */
286 int r_direct_mode;
287 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
288
289 /* data payload is used for xattr ops */
290 struct ceph_pagelist *r_pagelist;
291
292 /* what caps shall we drop? */
293 int r_inode_drop, r_inode_unless;
294 int r_dentry_drop, r_dentry_unless;
295 int r_old_dentry_drop, r_old_dentry_unless;
296 struct inode *r_old_inode;
297 int r_old_inode_drop, r_old_inode_unless;
298
299 struct ceph_msg *r_request; /* original request */
300 struct ceph_msg *r_reply;
301 struct ceph_mds_reply_info_parsed r_reply_info;
302 int r_err;
303 u32 r_readdir_offset;
304
305 struct page *r_locked_page;
306 int r_dir_caps;
307 int r_num_caps;
308
309 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
310 unsigned long r_started; /* start time to measure timeout against */
311 unsigned long r_start_latency; /* start time to measure latency */
312 unsigned long r_end_latency; /* finish time to measure latency */
313 unsigned long r_request_started; /* start time for mds request only,
314 used to measure lease durations */
315
316 /* link unsafe requests to parent directory, for fsync */
317 struct inode *r_unsafe_dir;
318 struct list_head r_unsafe_dir_item;
319
320 /* unsafe requests that modify the target inode */
321 struct list_head r_unsafe_target_item;
322
323 struct ceph_mds_session *r_session;
324
325 int r_attempts; /* resend attempts */
326 int r_num_fwd; /* number of forward attempts */
327 int r_resend_mds; /* mds to resend to next, if any*/
328 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
329 u64 r_deleg_ino;
330
331 struct list_head r_wait;
332 struct completion r_completion;
333 struct completion r_safe_completion;
334 ceph_mds_request_callback_t r_callback;
335 struct list_head r_unsafe_item; /* per-session unsafe list item */
336
337 long long r_dir_release_cnt;
338 long long r_dir_ordered_cnt;
339 int r_readdir_cache_idx;
340
341 int r_feature_needed;
342
343 struct ceph_cap_reservation r_caps_reservation;
344 };
345
346 struct ceph_pool_perm {
347 struct rb_node node;
348 int perm;
349 s64 pool;
350 size_t pool_ns_len;
351 char pool_ns[];
352 };
353
354 struct ceph_snapid_map {
355 struct rb_node node;
356 struct list_head lru;
357 atomic_t ref;
358 u64 snap;
359 dev_t dev;
360 unsigned long last_used;
361 };
362
363 /*
364 * node for list of quotarealm inodes that are not visible from the filesystem
365 * mountpoint, but required to handle, e.g. quotas.
366 */
367 struct ceph_quotarealm_inode {
368 struct rb_node node;
369 u64 ino;
370 unsigned long timeout; /* last time a lookup failed for this inode */
371 struct mutex mutex;
372 struct inode *inode;
373 };
374
375 struct cap_wait {
376 struct list_head list;
377 u64 ino;
378 pid_t tgid;
379 int need;
380 int want;
381 };
382
383 enum {
384 CEPH_MDSC_STOPPING_BEGIN = 1,
385 CEPH_MDSC_STOPPING_FLUSHING = 2,
386 CEPH_MDSC_STOPPING_FLUSHED = 3,
387 };
388
389 /*
390 * mds client state
391 */
392 struct ceph_mds_client {
393 struct ceph_fs_client *fsc;
394 struct mutex mutex; /* all nested structures */
395
396 struct ceph_mdsmap *mdsmap;
397 struct completion safe_umount_waiters;
398 wait_queue_head_t session_close_wq;
399 struct list_head waiting_for_map;
400 int mdsmap_err;
401
402 struct ceph_mds_session **sessions; /* NULL for mds if no session */
403 atomic_t num_sessions;
404 int max_sessions; /* len of sessions array */
405
406 spinlock_t stopping_lock; /* protect snap_empty */
407 int stopping; /* the stage of shutting down */
408 atomic_t stopping_blockers;
409 struct completion stopping_waiter;
410
411 atomic64_t quotarealms_count; /* # realms with quota */
412 /*
413 * We keep a list of inodes we don't see in the mountpoint but that we
414 * need to track quota realms.
415 */
416 struct rb_root quotarealms_inodes;
417 struct mutex quotarealms_inodes_mutex;
418
419 /*
420 * snap_rwsem will cover cap linkage into snaprealms, and
421 * realm snap contexts. (later, we can do per-realm snap
422 * contexts locks..) the empty list contains realms with no
423 * references (implying they contain no inodes with caps) that
424 * should be destroyed.
425 */
426 u64 last_snap_seq;
427 struct rw_semaphore snap_rwsem;
428 struct rb_root snap_realms;
429 struct list_head snap_empty;
430 int num_snap_realms;
431 spinlock_t snap_empty_lock; /* protect snap_empty */
432
433 u64 last_tid; /* most recent mds request */
434 u64 oldest_tid; /* oldest incomplete mds request,
435 excluding setfilelock requests */
436 struct rb_root request_tree; /* pending mds requests */
437 struct delayed_work delayed_work; /* delayed work */
438 unsigned long last_renew_caps; /* last time we renewed our caps */
439 struct list_head cap_delay_list; /* caps with delayed release */
440 spinlock_t cap_delay_lock; /* protects cap_delay_list */
441 struct list_head snap_flush_list; /* cap_snaps ready to flush */
442 spinlock_t snap_flush_lock;
443
444 u64 last_cap_flush_tid;
445 struct list_head cap_flush_list;
446 struct list_head cap_dirty_migrating; /* ...that are migration... */
447 int num_cap_flushing; /* # caps we are flushing */
448 spinlock_t cap_dirty_lock; /* protects above items */
449 wait_queue_head_t cap_flushing_wq;
450
451 struct work_struct cap_reclaim_work;
452 atomic_t cap_reclaim_pending;
453
454 /*
455 * Cap reservations
456 *
457 * Maintain a global pool of preallocated struct ceph_caps, referenced
458 * by struct ceph_caps_reservations. This ensures that we preallocate
459 * memory needed to successfully process an MDS response. (If an MDS
460 * sends us cap information and we fail to process it, we will have
461 * problems due to the client and MDS being out of sync.)
462 *
463 * Reservations are 'owned' by a ceph_cap_reservation context.
464 */
465 spinlock_t caps_list_lock;
466 struct list_head caps_list; /* unused (reserved or
467 unreserved) */
468 struct list_head cap_wait_list;
469 int caps_total_count; /* total caps allocated */
470 int caps_use_count; /* in use */
471 int caps_use_max; /* max used caps */
472 int caps_reserve_count; /* unused, reserved */
473 int caps_avail_count; /* unused, unreserved */
474 int caps_min_count; /* keep at least this many
475 (unreserved) */
476 spinlock_t dentry_list_lock;
477 struct list_head dentry_leases; /* fifo list */
478 struct list_head dentry_dir_leases; /* lru list */
479
480 struct ceph_client_metric metric;
481
482 spinlock_t snapid_map_lock;
483 struct rb_root snapid_map_tree;
484 struct list_head snapid_map_lru;
485
486 struct rw_semaphore pool_perm_rwsem;
487 struct rb_root pool_perm_tree;
488
489 char nodename[__NEW_UTS_LEN + 1];
490 };
491
492 extern const char *ceph_mds_op_name(int op);
493
494 extern bool check_session_state(struct ceph_mds_session *s);
495 void inc_session_sequence(struct ceph_mds_session *s);
496
497 extern struct ceph_mds_session *
498 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
499
500 extern const char *ceph_session_state_name(int s);
501
502 extern struct ceph_mds_session *
503 ceph_get_mds_session(struct ceph_mds_session *s);
504 extern void ceph_put_mds_session(struct ceph_mds_session *s);
505
506 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
507 struct ceph_msg *msg, int mds);
508
509 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
510 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
511 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
512 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
513
514 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
515
516 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
517 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
518 struct inode *dir);
519 extern struct ceph_mds_request *
520 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
521 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
522 struct inode *dir,
523 struct ceph_mds_request *req);
524 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
525 struct ceph_mds_request *req,
526 ceph_mds_request_wait_callback_t wait_func);
527 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
528 struct inode *dir,
529 struct ceph_mds_request *req);
530 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
531 extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
ceph_mdsc_get_request(struct ceph_mds_request * req)532 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
533 {
534 kref_get(&req->r_kref);
535 }
536 extern void ceph_mdsc_release_request(struct kref *kref);
ceph_mdsc_put_request(struct ceph_mds_request * req)537 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
538 {
539 kref_put(&req->r_kref, ceph_mdsc_release_request);
540 }
541
542 extern void send_flush_mdlog(struct ceph_mds_session *s);
543 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
544 void (*cb)(struct ceph_mds_session *),
545 bool check_state);
546 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
547 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
548 struct ceph_cap *cap);
549 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
550 struct ceph_mds_session *session);
551 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
552 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
553 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
554 int (*cb)(struct inode *, int mds, void *),
555 void *arg);
556 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
557
ceph_mdsc_free_path(char * path,int len)558 static inline void ceph_mdsc_free_path(char *path, int len)
559 {
560 if (!IS_ERR_OR_NULL(path))
561 __putname(path - (PATH_MAX - 1 - len));
562 }
563
564 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
565 int stop_on_nosnap);
566
567 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
568 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
569 struct dentry *dentry, char action,
570 u32 seq);
571
572 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
573 struct ceph_msg *msg);
574 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
575 struct ceph_msg *msg);
576
577 extern struct ceph_mds_session *
578 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
579 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
580 struct ceph_mds_session *session);
581
582 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
583 struct ceph_mds_session *session,
584 int max_caps);
585
ceph_wait_on_async_create(struct inode * inode)586 static inline int ceph_wait_on_async_create(struct inode *inode)
587 {
588 struct ceph_inode_info *ci = ceph_inode(inode);
589
590 return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
591 TASK_KILLABLE);
592 }
593
594 extern int ceph_wait_on_conflict_unlink(struct dentry *dentry);
595 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
596 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
597 #endif
598