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>
14 #include <linux/ceph/types.h>
15 #include <linux/ceph/messenger.h>
16 #include <linux/ceph/mdsmap.h>
17 #include <linux/ceph/auth.h>
19 /* The first 8 bits are reserved for old ceph releases */
20 #define CEPHFS_FEATURE_MIMIC 8
21 #define CEPHFS_FEATURE_REPLY_ENCODING 9
22 #define CEPHFS_FEATURE_RECLAIM_CLIENT 10
23 #define CEPHFS_FEATURE_LAZY_CAP_WANTED 11
24 #define CEPHFS_FEATURE_MULTI_RECONNECT 12
26 #define CEPHFS_FEATURES_CLIENT_SUPPORTED { \
27 0, 1, 2, 3, 4, 5, 6, 7, \
28 CEPHFS_FEATURE_MIMIC, \
29 CEPHFS_FEATURE_REPLY_ENCODING, \
30 CEPHFS_FEATURE_LAZY_CAP_WANTED, \
31 CEPHFS_FEATURE_MULTI_RECONNECT, \
33 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
37 * Some lock dependencies:
45 * mdsc->snap_flush_lock
46 * mdsc->cap_delay_lock
50 struct ceph_fs_client;
54 * parsed info about a single inode. pointers are into the encoded
55 * on-wire structures within the mds reply message payload.
57 struct ceph_mds_reply_info_in {
58 struct ceph_mds_reply_inode *in;
59 struct ceph_dir_layout dir_layout;
72 struct ceph_timespec btime;
73 struct ceph_timespec snap_btime;
77 struct ceph_mds_reply_dir_entry {
80 struct ceph_mds_reply_lease *lease;
81 struct ceph_mds_reply_info_in inode;
86 * parsed info about an mds reply, including information about
87 * either: 1) the target inode and/or its parent directory and dentry,
88 * and directory contents (for readdir results), or
89 * 2) the file range lock info (for fcntl F_GETLK results).
91 struct ceph_mds_reply_info_parsed {
92 struct ceph_mds_reply_head *head;
95 struct ceph_mds_reply_info_in diri, targeti;
96 struct ceph_mds_reply_dirfrag *dirfrag;
99 struct ceph_mds_reply_lease *dlease;
103 /* for fcntl F_GETLK results */
104 struct ceph_filelock *filelock_reply;
106 /* for readdir results */
108 struct ceph_mds_reply_dirfrag *dir_dir;
115 struct ceph_mds_reply_dir_entry *dir_entries;
118 /* for create results */
125 /* encoded blob describing snapshot contexts for certain
126 operations (e.g., open) */
133 * cap releases are batched and sent to the MDS en masse.
135 * Account for per-message overhead of mds_cap_release header
136 * and __le32 for osd epoch barrier trailing field.
138 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
139 sizeof(struct ceph_mds_cap_release)) / \
140 sizeof(struct ceph_mds_cap_item))
144 * state associated with each MDS<->client session
147 CEPH_MDS_SESSION_NEW = 1,
148 CEPH_MDS_SESSION_OPENING = 2,
149 CEPH_MDS_SESSION_OPEN = 3,
150 CEPH_MDS_SESSION_HUNG = 4,
151 CEPH_MDS_SESSION_RESTARTING = 5,
152 CEPH_MDS_SESSION_RECONNECTING = 6,
153 CEPH_MDS_SESSION_CLOSING = 7,
154 CEPH_MDS_SESSION_CLOSED = 8,
155 CEPH_MDS_SESSION_REJECTED = 9,
158 struct ceph_mds_session {
159 struct ceph_mds_client *s_mdsc;
162 unsigned long s_ttl; /* time until mds kills us */
163 unsigned long s_features;
164 u64 s_seq; /* incoming msg seq # */
165 struct mutex s_mutex; /* serialize session messages */
167 struct ceph_connection s_con;
169 struct ceph_auth_handshake s_auth;
171 /* protected by s_gen_ttl_lock */
172 spinlock_t s_gen_ttl_lock;
173 u32 s_cap_gen; /* inc each time we get mds stale msg */
174 unsigned long s_cap_ttl; /* when session caps expire */
176 /* protected by s_cap_lock */
177 spinlock_t s_cap_lock;
178 struct list_head s_caps; /* all caps issued by this session */
179 struct ceph_cap *s_cap_iterator;
181 int s_num_cap_releases;
184 struct list_head s_cap_releases; /* waiting cap_release messages */
185 struct work_struct s_cap_release_work;
187 /* protected by mutex */
188 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
189 unsigned long s_renew_requested; /* last time we sent a renew req */
193 struct list_head s_waiting; /* waiting requests */
194 struct list_head s_unsafe; /* unsafe requests */
198 * modes of choosing which MDS to send a request to
203 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
206 struct ceph_mds_request;
207 struct ceph_mds_client;
210 * request completion callback
212 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
213 struct ceph_mds_request *req);
215 * wait for request completion callback
217 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
218 struct ceph_mds_request *req);
221 * an in-flight mds request
223 struct ceph_mds_request {
224 u64 r_tid; /* transaction id */
225 struct rb_node r_node;
226 struct ceph_mds_client *r_mdsc;
228 int r_op; /* mds op code */
230 /* operation on what? */
231 struct inode *r_inode; /* arg1 */
232 struct dentry *r_dentry; /* arg1 */
233 struct dentry *r_old_dentry; /* arg2: rename from or link from */
234 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
235 char *r_path1, *r_path2;
236 struct ceph_vino r_ino1, r_ino2;
238 struct inode *r_parent; /* parent dir inode */
239 struct inode *r_target_inode; /* resulting inode */
241 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
242 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
243 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
244 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
245 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
246 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
247 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
248 unsigned long r_req_flags;
250 struct mutex r_fill_mutex;
252 union ceph_mds_request_args r_args;
253 int r_fmode; /* file mode, if expecting cap */
256 struct timespec64 r_stamp;
258 /* for choosing which mds to send this request to */
260 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
262 /* data payload is used for xattr ops */
263 struct ceph_pagelist *r_pagelist;
265 /* what caps shall we drop? */
266 int r_inode_drop, r_inode_unless;
267 int r_dentry_drop, r_dentry_unless;
268 int r_old_dentry_drop, r_old_dentry_unless;
269 struct inode *r_old_inode;
270 int r_old_inode_drop, r_old_inode_unless;
272 struct ceph_msg *r_request; /* original request */
273 int r_request_release_offset;
274 struct ceph_msg *r_reply;
275 struct ceph_mds_reply_info_parsed r_reply_info;
276 struct page *r_locked_page;
279 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
280 unsigned long r_started; /* start time to measure timeout against */
281 unsigned long r_request_started; /* start time for mds request only,
282 used to measure lease durations */
284 /* link unsafe requests to parent directory, for fsync */
285 struct inode *r_unsafe_dir;
286 struct list_head r_unsafe_dir_item;
288 /* unsafe requests that modify the target inode */
289 struct list_head r_unsafe_target_item;
291 struct ceph_mds_session *r_session;
293 int r_attempts; /* resend attempts */
294 int r_num_fwd; /* number of forward attempts */
295 int r_resend_mds; /* mds to resend to next, if any*/
296 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
299 struct list_head r_wait;
300 struct completion r_completion;
301 struct completion r_safe_completion;
302 ceph_mds_request_callback_t r_callback;
303 ceph_mds_request_wait_callback_t r_wait_for_completion;
304 struct list_head r_unsafe_item; /* per-session unsafe list item */
306 long long r_dir_release_cnt;
307 long long r_dir_ordered_cnt;
308 int r_readdir_cache_idx;
309 u32 r_readdir_offset;
311 struct ceph_cap_reservation r_caps_reservation;
315 struct ceph_pool_perm {
323 struct ceph_snapid_map {
325 struct list_head lru;
329 unsigned long last_used;
333 * node for list of quotarealm inodes that are not visible from the filesystem
334 * mountpoint, but required to handle, e.g. quotas.
336 struct ceph_quotarealm_inode {
339 unsigned long timeout; /* last time a lookup failed for this inode */
345 struct list_head list;
355 struct ceph_mds_client {
356 struct ceph_fs_client *fsc;
357 struct mutex mutex; /* all nested structures */
359 struct ceph_mdsmap *mdsmap;
360 struct completion safe_umount_waiters;
361 wait_queue_head_t session_close_wq;
362 struct list_head waiting_for_map;
365 struct ceph_mds_session **sessions; /* NULL for mds if no session */
366 atomic_t num_sessions;
367 int max_sessions; /* len of s_mds_sessions */
368 int stopping; /* true if shutting down */
370 atomic64_t quotarealms_count; /* # realms with quota */
372 * We keep a list of inodes we don't see in the mountpoint but that we
373 * need to track quota realms.
375 struct rb_root quotarealms_inodes;
376 struct mutex quotarealms_inodes_mutex;
379 * snap_rwsem will cover cap linkage into snaprealms, and
380 * realm snap contexts. (later, we can do per-realm snap
381 * contexts locks..) the empty list contains realms with no
382 * references (implying they contain no inodes with caps) that
383 * should be destroyed.
386 struct rw_semaphore snap_rwsem;
387 struct rb_root snap_realms;
388 struct list_head snap_empty;
390 spinlock_t snap_empty_lock; /* protect snap_empty */
392 u64 last_tid; /* most recent mds request */
393 u64 oldest_tid; /* oldest incomplete mds request,
394 excluding setfilelock requests */
395 struct rb_root request_tree; /* pending mds requests */
396 struct delayed_work delayed_work; /* delayed work */
397 unsigned long last_renew_caps; /* last time we renewed our caps */
398 struct list_head cap_delay_list; /* caps with delayed release */
399 spinlock_t cap_delay_lock; /* protects cap_delay_list */
400 struct list_head snap_flush_list; /* cap_snaps ready to flush */
401 spinlock_t snap_flush_lock;
403 u64 last_cap_flush_tid;
404 struct list_head cap_flush_list;
405 struct list_head cap_dirty; /* inodes with dirty caps */
406 struct list_head cap_dirty_migrating; /* ...that are migration... */
407 int num_cap_flushing; /* # caps we are flushing */
408 spinlock_t cap_dirty_lock; /* protects above items */
409 wait_queue_head_t cap_flushing_wq;
411 struct work_struct cap_reclaim_work;
412 atomic_t cap_reclaim_pending;
417 * Maintain a global pool of preallocated struct ceph_caps, referenced
418 * by struct ceph_caps_reservations. This ensures that we preallocate
419 * memory needed to successfully process an MDS response. (If an MDS
420 * sends us cap information and we fail to process it, we will have
421 * problems due to the client and MDS being out of sync.)
423 * Reservations are 'owned' by a ceph_cap_reservation context.
425 spinlock_t caps_list_lock;
426 struct list_head caps_list; /* unused (reserved or
428 struct list_head cap_wait_list;
429 int caps_total_count; /* total caps allocated */
430 int caps_use_count; /* in use */
431 int caps_use_max; /* max used caps */
432 int caps_reserve_count; /* unused, reserved */
433 int caps_avail_count; /* unused, unreserved */
434 int caps_min_count; /* keep at least this many
436 spinlock_t dentry_list_lock;
437 struct list_head dentry_leases; /* fifo list */
438 struct list_head dentry_dir_leases; /* lru list */
440 spinlock_t snapid_map_lock;
441 struct rb_root snapid_map_tree;
442 struct list_head snapid_map_lru;
444 struct rw_semaphore pool_perm_rwsem;
445 struct rb_root pool_perm_tree;
447 char nodename[__NEW_UTS_LEN + 1];
450 extern const char *ceph_mds_op_name(int op);
452 extern struct ceph_mds_session *
453 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
455 extern const char *ceph_session_state_name(int s);
457 extern struct ceph_mds_session *
458 ceph_get_mds_session(struct ceph_mds_session *s);
459 extern void ceph_put_mds_session(struct ceph_mds_session *s);
461 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
462 struct ceph_msg *msg, int mds);
464 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
465 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
466 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
467 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
469 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
471 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
472 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
474 extern struct ceph_mds_request *
475 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
476 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
478 struct ceph_mds_request *req);
479 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
481 struct ceph_mds_request *req);
482 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
484 kref_get(&req->r_kref);
486 extern void ceph_mdsc_release_request(struct kref *kref);
487 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
489 kref_put(&req->r_kref, ceph_mdsc_release_request);
492 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
493 struct ceph_cap *cap);
494 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
495 struct ceph_mds_session *session);
496 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
497 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
498 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
499 int (*cb)(struct inode *,
500 struct ceph_cap *, void *),
502 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
504 static inline void ceph_mdsc_free_path(char *path, int len)
507 __putname(path - (PATH_MAX - 1 - len));
510 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
513 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
514 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
515 struct dentry *dentry, char action,
518 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
519 struct ceph_msg *msg);
520 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
521 struct ceph_msg *msg);
523 extern struct ceph_mds_session *
524 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
525 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
526 struct ceph_mds_session *session);
528 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
529 struct ceph_mds_session *session,