1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _FS_CEPH_MDS_CLIENT_H
3 #define _FS_CEPH_MDS_CLIENT_H
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>
15 #include <linux/ceph/types.h>
16 #include <linux/ceph/messenger.h>
17 #include <linux/ceph/auth.h>
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 CEPHFS_FEATURE_32BITS_RETRY_FWD,
36 CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
37 CEPHFS_FEATURE_HAS_OWNER_UIDGID,
38 CEPHFS_FEATURE_MDS_AUTH_CAPS_CHECK,
40 CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_MDS_AUTH_CAPS_CHECK,
43 #define CEPHFS_FEATURES_CLIENT_SUPPORTED { \
44 0, 1, 2, 3, 4, 5, 6, 7, \
45 CEPHFS_FEATURE_MIMIC, \
46 CEPHFS_FEATURE_REPLY_ENCODING, \
47 CEPHFS_FEATURE_LAZY_CAP_WANTED, \
48 CEPHFS_FEATURE_MULTI_RECONNECT, \
49 CEPHFS_FEATURE_DELEG_INO, \
50 CEPHFS_FEATURE_METRIC_COLLECT, \
51 CEPHFS_FEATURE_ALTERNATE_NAME, \
52 CEPHFS_FEATURE_NOTIFY_SESSION_STATE, \
53 CEPHFS_FEATURE_OP_GETVXATTR, \
54 CEPHFS_FEATURE_32BITS_RETRY_FWD, \
55 CEPHFS_FEATURE_HAS_OWNER_UIDGID, \
56 CEPHFS_FEATURE_MDS_AUTH_CAPS_CHECK, \
60 * Some lock dependencies:
68 * mdsc->snap_flush_lock
69 * mdsc->cap_delay_lock
73 struct ceph_fs_client;
76 #define MDS_AUTH_UID_ANY -1
78 struct ceph_mds_cap_match {
79 s64 uid; /* default to MDS_AUTH_UID_ANY */
81 u32 *gids; /* use these GIDs */
82 char *path; /* require path to be child of this
83 (may be "" or "/" for any) */
85 bool root_squash; /* default to false */
88 struct ceph_mds_cap_auth {
89 struct ceph_mds_cap_match match;
95 * parsed info about a single inode. pointers are into the encoded
96 * on-wire structures within the mds reply message payload.
98 struct ceph_mds_reply_info_in {
99 struct ceph_mds_reply_inode *in;
100 struct ceph_dir_layout dir_layout;
113 struct ceph_timespec btime;
114 struct ceph_timespec snap_btime;
117 u32 fscrypt_auth_len;
118 u32 fscrypt_file_len;
123 struct ceph_mds_reply_dir_entry {
128 struct ceph_mds_reply_lease *lease;
129 struct ceph_mds_reply_info_in inode;
133 struct ceph_mds_reply_xattr {
135 size_t xattr_value_len;
139 * parsed info about an mds reply, including information about
140 * either: 1) the target inode and/or its parent directory and dentry,
141 * and directory contents (for readdir results), or
142 * 2) the file range lock info (for fcntl F_GETLK results).
144 struct ceph_mds_reply_info_parsed {
145 struct ceph_mds_reply_head *head;
148 struct ceph_mds_reply_info_in diri, targeti;
149 struct ceph_mds_reply_dirfrag *dirfrag;
154 struct ceph_mds_reply_lease *dlease;
155 struct ceph_mds_reply_xattr xattr_info;
159 /* for fcntl F_GETLK results */
160 struct ceph_filelock *filelock_reply;
162 /* for readdir results */
164 struct ceph_mds_reply_dirfrag *dir_dir;
171 struct ceph_mds_reply_dir_entry *dir_entries;
174 /* for create results */
181 /* encoded blob describing snapshot contexts for certain
182 operations (e.g., open) */
189 * cap releases are batched and sent to the MDS en masse.
191 * Account for per-message overhead of mds_cap_release header
192 * and __le32 for osd epoch barrier trailing field.
194 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
195 sizeof(struct ceph_mds_cap_release)) / \
196 sizeof(struct ceph_mds_cap_item))
200 * state associated with each MDS<->client session
203 CEPH_MDS_SESSION_NEW = 1,
204 CEPH_MDS_SESSION_OPENING = 2,
205 CEPH_MDS_SESSION_OPEN = 3,
206 CEPH_MDS_SESSION_HUNG = 4,
207 CEPH_MDS_SESSION_RESTARTING = 5,
208 CEPH_MDS_SESSION_RECONNECTING = 6,
209 CEPH_MDS_SESSION_CLOSING = 7,
210 CEPH_MDS_SESSION_CLOSED = 8,
211 CEPH_MDS_SESSION_REJECTED = 9,
214 struct ceph_mds_session {
215 struct ceph_mds_client *s_mdsc;
218 unsigned long s_ttl; /* time until mds kills us */
219 unsigned long s_features;
220 u64 s_seq; /* incoming msg seq # */
221 struct mutex s_mutex; /* serialize session messages */
223 struct ceph_connection s_con;
225 struct ceph_auth_handshake s_auth;
227 atomic_t s_cap_gen; /* inc each time we get mds stale msg */
228 unsigned long s_cap_ttl; /* when session caps expire. protected by s_mutex */
230 /* protected by s_cap_lock */
231 spinlock_t s_cap_lock;
233 struct list_head s_caps; /* all caps issued by this session */
234 struct ceph_cap *s_cap_iterator;
236 int s_num_cap_releases;
239 struct list_head s_cap_releases; /* waiting cap_release messages */
240 struct work_struct s_cap_release_work;
242 /* See ceph_inode_info->i_dirty_item. */
243 struct list_head s_cap_dirty; /* inodes w/ dirty caps */
245 /* See ceph_inode_info->i_flushing_item. */
246 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
248 unsigned long s_renew_requested; /* last time we sent a renew req */
251 struct list_head s_waiting; /* waiting requests */
252 struct list_head s_unsafe; /* unsafe requests */
253 struct xarray s_delegated_inos;
257 * modes of choosing which MDS to send a request to
262 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
265 struct ceph_mds_request;
266 struct ceph_mds_client;
269 * request completion callback
271 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
272 struct ceph_mds_request *req);
274 * wait for request completion callback
276 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
277 struct ceph_mds_request *req);
280 * an in-flight mds request
282 struct ceph_mds_request {
283 u64 r_tid; /* transaction id */
284 struct rb_node r_node;
285 struct ceph_mds_client *r_mdsc;
288 int r_op; /* mds op code */
290 /* operation on what? */
291 struct inode *r_inode; /* arg1 */
292 struct dentry *r_dentry; /* arg1 */
293 struct dentry *r_old_dentry; /* arg2: rename from or link from */
294 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
295 char *r_path1, *r_path2;
296 struct ceph_vino r_ino1, r_ino2;
298 struct inode *r_parent; /* parent dir inode */
299 struct inode *r_target_inode; /* resulting inode */
300 struct inode *r_new_inode; /* new inode (for creates) */
302 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
303 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
304 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
305 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
306 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
307 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
308 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
309 #define CEPH_MDS_R_ASYNC (8) /* async request */
310 #define CEPH_MDS_R_FSCRYPT_FILE (9) /* must marshal fscrypt_file field */
311 unsigned long r_req_flags;
313 struct mutex r_fill_mutex;
315 union ceph_mds_request_args r_args;
317 struct ceph_fscrypt_auth *r_fscrypt_auth;
320 u8 *r_altname; /* fscrypt binary crypttext for long filenames */
321 u32 r_altname_len; /* length of r_altname */
323 int r_fmode; /* file mode, if expecting cap */
324 int r_request_release_offset;
325 const struct cred *r_cred;
326 struct mnt_idmap *r_mnt_idmap;
327 struct timespec64 r_stamp;
329 /* for choosing which mds to send this request to */
331 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
333 /* data payload is used for xattr ops */
334 struct ceph_pagelist *r_pagelist;
336 /* what caps shall we drop? */
337 int r_inode_drop, r_inode_unless;
338 int r_dentry_drop, r_dentry_unless;
339 int r_old_dentry_drop, r_old_dentry_unless;
340 struct inode *r_old_inode;
341 int r_old_inode_drop, r_old_inode_unless;
343 struct ceph_msg *r_request; /* original request */
344 struct ceph_msg *r_reply;
345 struct ceph_mds_reply_info_parsed r_reply_info;
347 u32 r_readdir_offset;
349 struct page *r_locked_page;
353 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
354 unsigned long r_started; /* start time to measure timeout against */
355 unsigned long r_start_latency; /* start time to measure latency */
356 unsigned long r_end_latency; /* finish time to measure latency */
357 unsigned long r_request_started; /* start time for mds request only,
358 used to measure lease durations */
360 /* link unsafe requests to parent directory, for fsync */
361 struct inode *r_unsafe_dir;
362 struct list_head r_unsafe_dir_item;
364 /* unsafe requests that modify the target inode */
365 struct list_head r_unsafe_target_item;
367 struct ceph_mds_session *r_session;
369 int r_attempts; /* resend attempts */
370 int r_num_fwd; /* number of forward attempts */
371 int r_resend_mds; /* mds to resend to next, if any*/
372 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
375 struct list_head r_wait;
376 struct completion r_completion;
377 struct completion r_safe_completion;
378 ceph_mds_request_callback_t r_callback;
379 struct list_head r_unsafe_item; /* per-session unsafe list item */
381 long long r_dir_release_cnt;
382 long long r_dir_ordered_cnt;
383 int r_readdir_cache_idx;
385 int r_feature_needed;
387 struct ceph_cap_reservation r_caps_reservation;
390 struct ceph_pool_perm {
398 struct ceph_snapid_map {
400 struct list_head lru;
404 unsigned long last_used;
408 * node for list of quotarealm inodes that are not visible from the filesystem
409 * mountpoint, but required to handle, e.g. quotas.
411 struct ceph_quotarealm_inode {
414 unsigned long timeout; /* last time a lookup failed for this inode */
420 struct list_head list;
428 CEPH_MDSC_STOPPING_BEGIN = 1,
429 CEPH_MDSC_STOPPING_FLUSHING = 2,
430 CEPH_MDSC_STOPPING_FLUSHED = 3,
436 struct ceph_mds_client {
437 struct ceph_fs_client *fsc;
438 struct mutex mutex; /* all nested structures */
440 struct ceph_mdsmap *mdsmap;
441 struct completion safe_umount_waiters;
442 wait_queue_head_t session_close_wq;
443 struct list_head waiting_for_map;
446 struct ceph_mds_session **sessions; /* NULL for mds if no session */
447 atomic_t num_sessions;
448 int max_sessions; /* len of sessions array */
450 spinlock_t stopping_lock; /* protect snap_empty */
451 int stopping; /* the stage of shutting down */
452 atomic_t stopping_blockers;
453 struct completion stopping_waiter;
455 atomic64_t quotarealms_count; /* # realms with quota */
457 * We keep a list of inodes we don't see in the mountpoint but that we
458 * need to track quota realms.
460 struct rb_root quotarealms_inodes;
461 struct mutex quotarealms_inodes_mutex;
464 * snap_rwsem will cover cap linkage into snaprealms, and
465 * realm snap contexts. (later, we can do per-realm snap
466 * contexts locks..) the empty list contains realms with no
467 * references (implying they contain no inodes with caps) that
468 * should be destroyed.
471 struct rw_semaphore snap_rwsem;
472 struct rb_root snap_realms;
473 struct list_head snap_empty;
475 spinlock_t snap_empty_lock; /* protect snap_empty */
477 u64 last_tid; /* most recent mds request */
478 u64 oldest_tid; /* oldest incomplete mds request,
479 excluding setfilelock requests */
480 struct rb_root request_tree; /* pending mds requests */
481 struct delayed_work delayed_work; /* delayed work */
482 unsigned long last_renew_caps; /* last time we renewed our caps */
483 struct list_head cap_delay_list; /* caps with delayed release */
484 struct list_head cap_unlink_delay_list; /* caps with delayed release for unlink */
485 spinlock_t cap_delay_lock; /* protects cap_delay_list and cap_unlink_delay_list */
486 struct list_head snap_flush_list; /* cap_snaps ready to flush */
487 spinlock_t snap_flush_lock;
489 u64 last_cap_flush_tid;
490 struct list_head cap_flush_list;
491 struct list_head cap_dirty_migrating; /* ...that are migration... */
492 int num_cap_flushing; /* # caps we are flushing */
493 spinlock_t cap_dirty_lock; /* protects above items */
494 wait_queue_head_t cap_flushing_wq;
496 struct work_struct cap_reclaim_work;
497 atomic_t cap_reclaim_pending;
499 struct work_struct cap_unlink_work;
504 * Maintain a global pool of preallocated struct ceph_caps, referenced
505 * by struct ceph_caps_reservations. This ensures that we preallocate
506 * memory needed to successfully process an MDS response. (If an MDS
507 * sends us cap information and we fail to process it, we will have
508 * problems due to the client and MDS being out of sync.)
510 * Reservations are 'owned' by a ceph_cap_reservation context.
512 spinlock_t caps_list_lock;
513 struct list_head caps_list; /* unused (reserved or
515 struct list_head cap_wait_list;
516 int caps_total_count; /* total caps allocated */
517 int caps_use_count; /* in use */
518 int caps_use_max; /* max used caps */
519 int caps_reserve_count; /* unused, reserved */
520 int caps_avail_count; /* unused, unreserved */
521 int caps_min_count; /* keep at least this many
523 spinlock_t dentry_list_lock;
524 struct list_head dentry_leases; /* fifo list */
525 struct list_head dentry_dir_leases; /* lru list */
527 struct ceph_client_metric metric;
529 spinlock_t snapid_map_lock;
530 struct rb_root snapid_map_tree;
531 struct list_head snapid_map_lru;
533 struct rw_semaphore pool_perm_rwsem;
534 struct rb_root pool_perm_tree;
537 struct ceph_mds_cap_auth *s_cap_auths;
539 char nodename[__NEW_UTS_LEN + 1];
542 extern const char *ceph_mds_op_name(int op);
544 extern bool check_session_state(struct ceph_mds_session *s);
545 void inc_session_sequence(struct ceph_mds_session *s);
547 extern struct ceph_mds_session *
548 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
550 extern const char *ceph_session_state_name(int s);
552 extern struct ceph_mds_session *
553 ceph_get_mds_session(struct ceph_mds_session *s);
554 extern void ceph_put_mds_session(struct ceph_mds_session *s);
556 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
557 struct ceph_msg *msg, int mds);
559 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
560 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
561 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
562 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
564 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
566 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
567 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
569 extern struct ceph_mds_request *
570 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
571 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
573 struct ceph_mds_request *req);
574 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
575 struct ceph_mds_request *req,
576 ceph_mds_request_wait_callback_t wait_func);
577 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
579 struct ceph_mds_request *req);
580 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
581 extern void ceph_mdsc_release_dir_caps_async(struct ceph_mds_request *req);
582 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
584 kref_get(&req->r_kref);
586 extern void ceph_mdsc_release_request(struct kref *kref);
587 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
589 kref_put(&req->r_kref, ceph_mdsc_release_request);
592 extern void send_flush_mdlog(struct ceph_mds_session *s);
593 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
594 void (*cb)(struct ceph_mds_session *),
596 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
597 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
598 struct ceph_cap *cap);
599 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
600 struct ceph_mds_session *session);
601 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
602 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
603 extern void ceph_queue_cap_unlink_work(struct ceph_mds_client *mdsc);
604 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
605 int (*cb)(struct inode *, int mds, void *),
607 extern int ceph_mds_check_access(struct ceph_mds_client *mdsc, char *tpath,
610 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
612 static inline void ceph_mdsc_free_path(char *path, int len)
614 if (!IS_ERR_OR_NULL(path))
615 __putname(path - (PATH_MAX - 1 - len));
618 extern char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc,
619 struct dentry *dentry, int *plen, u64 *base,
622 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
623 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
624 struct dentry *dentry, char action,
627 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
628 struct ceph_msg *msg);
629 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
630 struct ceph_msg *msg);
632 extern struct ceph_mds_session *
633 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
634 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
635 struct ceph_mds_session *session);
637 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
638 struct ceph_mds_session *session,
641 static inline int ceph_wait_on_async_create(struct inode *inode)
643 struct ceph_inode_info *ci = ceph_inode(inode);
645 return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
649 extern int ceph_wait_on_conflict_unlink(struct dentry *dentry);
650 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
651 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
653 extern bool enable_unsafe_idmap;