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;
74 struct ceph_mds_reply_dir_entry {
77 struct ceph_mds_reply_lease *lease;
78 struct ceph_mds_reply_info_in inode;
83 * parsed info about an mds reply, including information about
84 * either: 1) the target inode and/or its parent directory and dentry,
85 * and directory contents (for readdir results), or
86 * 2) the file range lock info (for fcntl F_GETLK results).
88 struct ceph_mds_reply_info_parsed {
89 struct ceph_mds_reply_head *head;
92 struct ceph_mds_reply_info_in diri, targeti;
93 struct ceph_mds_reply_dirfrag *dirfrag;
96 struct ceph_mds_reply_lease *dlease;
100 /* for fcntl F_GETLK results */
101 struct ceph_filelock *filelock_reply;
103 /* for readdir results */
105 struct ceph_mds_reply_dirfrag *dir_dir;
112 struct ceph_mds_reply_dir_entry *dir_entries;
115 /* for create results */
122 /* encoded blob describing snapshot contexts for certain
123 operations (e.g., open) */
130 * cap releases are batched and sent to the MDS en masse.
132 * Account for per-message overhead of mds_cap_release header
133 * and __le32 for osd epoch barrier trailing field.
135 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
136 sizeof(struct ceph_mds_cap_release)) / \
137 sizeof(struct ceph_mds_cap_item))
141 * state associated with each MDS<->client session
144 CEPH_MDS_SESSION_NEW = 1,
145 CEPH_MDS_SESSION_OPENING = 2,
146 CEPH_MDS_SESSION_OPEN = 3,
147 CEPH_MDS_SESSION_HUNG = 4,
148 CEPH_MDS_SESSION_CLOSING = 5,
149 CEPH_MDS_SESSION_RESTARTING = 6,
150 CEPH_MDS_SESSION_RECONNECTING = 7,
151 CEPH_MDS_SESSION_REJECTED = 8,
154 struct ceph_mds_session {
155 struct ceph_mds_client *s_mdsc;
158 unsigned long s_ttl; /* time until mds kills us */
159 unsigned long s_features;
160 u64 s_seq; /* incoming msg seq # */
161 struct mutex s_mutex; /* serialize session messages */
163 struct ceph_connection s_con;
165 struct ceph_auth_handshake s_auth;
167 /* protected by s_gen_ttl_lock */
168 spinlock_t s_gen_ttl_lock;
169 u32 s_cap_gen; /* inc each time we get mds stale msg */
170 unsigned long s_cap_ttl; /* when session caps expire */
172 /* protected by s_cap_lock */
173 spinlock_t s_cap_lock;
174 struct list_head s_caps; /* all caps issued by this session */
175 int s_nr_caps, s_trim_caps;
176 int s_num_cap_releases;
179 struct list_head s_cap_releases; /* waiting cap_release messages */
180 struct ceph_cap *s_cap_iterator;
182 /* protected by mutex */
183 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
184 unsigned long s_renew_requested; /* last time we sent a renew req */
188 struct list_head s_waiting; /* waiting requests */
189 struct list_head s_unsafe; /* unsafe requests */
193 * modes of choosing which MDS to send a request to
198 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
201 struct ceph_mds_request;
202 struct ceph_mds_client;
205 * request completion callback
207 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
208 struct ceph_mds_request *req);
210 * wait for request completion callback
212 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
213 struct ceph_mds_request *req);
216 * an in-flight mds request
218 struct ceph_mds_request {
219 u64 r_tid; /* transaction id */
220 struct rb_node r_node;
221 struct ceph_mds_client *r_mdsc;
223 int r_op; /* mds op code */
225 /* operation on what? */
226 struct inode *r_inode; /* arg1 */
227 struct dentry *r_dentry; /* arg1 */
228 struct dentry *r_old_dentry; /* arg2: rename from or link from */
229 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
230 char *r_path1, *r_path2;
231 struct ceph_vino r_ino1, r_ino2;
233 struct inode *r_parent; /* parent dir inode */
234 struct inode *r_target_inode; /* resulting inode */
236 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
237 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
238 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
239 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
240 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
241 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
242 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
243 unsigned long r_req_flags;
245 struct mutex r_fill_mutex;
247 union ceph_mds_request_args r_args;
248 int r_fmode; /* file mode, if expecting cap */
251 struct timespec64 r_stamp;
253 /* for choosing which mds to send this request to */
255 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
257 /* data payload is used for xattr ops */
258 struct ceph_pagelist *r_pagelist;
260 /* what caps shall we drop? */
261 int r_inode_drop, r_inode_unless;
262 int r_dentry_drop, r_dentry_unless;
263 int r_old_dentry_drop, r_old_dentry_unless;
264 struct inode *r_old_inode;
265 int r_old_inode_drop, r_old_inode_unless;
267 struct ceph_msg *r_request; /* original request */
268 int r_request_release_offset;
269 struct ceph_msg *r_reply;
270 struct ceph_mds_reply_info_parsed r_reply_info;
271 struct page *r_locked_page;
274 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
275 unsigned long r_started; /* start time to measure timeout against */
276 unsigned long r_request_started; /* start time for mds request only,
277 used to measure lease durations */
279 /* link unsafe requests to parent directory, for fsync */
280 struct inode *r_unsafe_dir;
281 struct list_head r_unsafe_dir_item;
283 /* unsafe requests that modify the target inode */
284 struct list_head r_unsafe_target_item;
286 struct ceph_mds_session *r_session;
288 int r_attempts; /* resend attempts */
289 int r_num_fwd; /* number of forward attempts */
290 int r_resend_mds; /* mds to resend to next, if any*/
291 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
294 struct list_head r_wait;
295 struct completion r_completion;
296 struct completion r_safe_completion;
297 ceph_mds_request_callback_t r_callback;
298 ceph_mds_request_wait_callback_t r_wait_for_completion;
299 struct list_head r_unsafe_item; /* per-session unsafe list item */
301 long long r_dir_release_cnt;
302 long long r_dir_ordered_cnt;
303 int r_readdir_cache_idx;
304 u32 r_readdir_offset;
306 struct ceph_cap_reservation r_caps_reservation;
310 struct ceph_pool_perm {
318 struct ceph_snapid_map {
320 struct list_head lru;
324 unsigned long last_used;
330 struct ceph_mds_client {
331 struct ceph_fs_client *fsc;
332 struct mutex mutex; /* all nested structures */
334 struct ceph_mdsmap *mdsmap;
335 struct completion safe_umount_waiters;
336 wait_queue_head_t session_close_wq;
337 struct list_head waiting_for_map;
340 struct ceph_mds_session **sessions; /* NULL for mds if no session */
341 atomic_t num_sessions;
342 int max_sessions; /* len of s_mds_sessions */
343 int stopping; /* true if shutting down */
345 atomic64_t quotarealms_count; /* # realms with quota */
348 * snap_rwsem will cover cap linkage into snaprealms, and
349 * realm snap contexts. (later, we can do per-realm snap
350 * contexts locks..) the empty list contains realms with no
351 * references (implying they contain no inodes with caps) that
352 * should be destroyed.
355 struct rw_semaphore snap_rwsem;
356 struct rb_root snap_realms;
357 struct list_head snap_empty;
359 spinlock_t snap_empty_lock; /* protect snap_empty */
361 u64 last_tid; /* most recent mds request */
362 u64 oldest_tid; /* oldest incomplete mds request,
363 excluding setfilelock requests */
364 struct rb_root request_tree; /* pending mds requests */
365 struct delayed_work delayed_work; /* delayed work */
366 unsigned long last_renew_caps; /* last time we renewed our caps */
367 struct list_head cap_delay_list; /* caps with delayed release */
368 spinlock_t cap_delay_lock; /* protects cap_delay_list */
369 struct list_head snap_flush_list; /* cap_snaps ready to flush */
370 spinlock_t snap_flush_lock;
372 u64 last_cap_flush_tid;
373 struct list_head cap_flush_list;
374 struct list_head cap_dirty; /* inodes with dirty caps */
375 struct list_head cap_dirty_migrating; /* ...that are migration... */
376 int num_cap_flushing; /* # caps we are flushing */
377 spinlock_t cap_dirty_lock; /* protects above items */
378 wait_queue_head_t cap_flushing_wq;
383 * Maintain a global pool of preallocated struct ceph_caps, referenced
384 * by struct ceph_caps_reservations. This ensures that we preallocate
385 * memory needed to successfully process an MDS response. (If an MDS
386 * sends us cap information and we fail to process it, we will have
387 * problems due to the client and MDS being out of sync.)
389 * Reservations are 'owned' by a ceph_cap_reservation context.
391 spinlock_t caps_list_lock;
392 struct list_head caps_list; /* unused (reserved or
394 int caps_total_count; /* total caps allocated */
395 int caps_use_count; /* in use */
396 int caps_reserve_count; /* unused, reserved */
397 int caps_avail_count; /* unused, unreserved */
398 int caps_min_count; /* keep at least this many
400 spinlock_t dentry_lru_lock;
401 struct list_head dentry_lru;
404 spinlock_t snapid_map_lock;
405 struct rb_root snapid_map_tree;
406 struct list_head snapid_map_lru;
408 struct rw_semaphore pool_perm_rwsem;
409 struct rb_root pool_perm_tree;
411 char nodename[__NEW_UTS_LEN + 1];
414 extern const char *ceph_mds_op_name(int op);
416 extern struct ceph_mds_session *
417 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
419 static inline struct ceph_mds_session *
420 ceph_get_mds_session(struct ceph_mds_session *s)
422 refcount_inc(&s->s_ref);
426 extern const char *ceph_session_state_name(int s);
428 extern void ceph_put_mds_session(struct ceph_mds_session *s);
430 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
431 struct ceph_msg *msg, int mds);
433 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
434 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
435 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
436 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
438 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
440 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
441 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
443 extern struct ceph_mds_request *
444 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
445 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
446 struct ceph_mds_request *req);
447 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
449 struct ceph_mds_request *req);
450 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
452 kref_get(&req->r_kref);
454 extern void ceph_mdsc_release_request(struct kref *kref);
455 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
457 kref_put(&req->r_kref, ceph_mdsc_release_request);
460 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
461 struct ceph_mds_session *session);
463 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
465 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
468 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
469 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
471 struct dentry *dentry, char action,
474 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
475 struct ceph_msg *msg);
476 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
477 struct ceph_msg *msg);
479 extern struct ceph_mds_session *
480 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
481 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
482 struct ceph_mds_session *session);
484 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
485 struct ceph_mds_session *session,