xref: /linux/fs/ceph/mds_client.h (revision b52861897be8a7ba563077ae62cd48158d16a5d9)
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/auth.h>
18 
19 #include "mdsmap.h"
20 #include "metric.h"
21 #include "subvolume_metrics.h"
22 #include "super.h"
23 
24 /* The first 8 bits are reserved for old ceph releases */
25 enum ceph_feature_type {
26 	CEPHFS_FEATURE_MIMIC = 8,
27 	CEPHFS_FEATURE_REPLY_ENCODING,
28 	CEPHFS_FEATURE_RECLAIM_CLIENT,
29 	CEPHFS_FEATURE_LAZY_CAP_WANTED,
30 	CEPHFS_FEATURE_MULTI_RECONNECT,
31 	CEPHFS_FEATURE_DELEG_INO,
32 	CEPHFS_FEATURE_METRIC_COLLECT,
33 	CEPHFS_FEATURE_ALTERNATE_NAME,
34 	CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
35 	CEPHFS_FEATURE_OP_GETVXATTR,
36 	CEPHFS_FEATURE_32BITS_RETRY_FWD,
37 	CEPHFS_FEATURE_NEW_SNAPREALM_INFO,
38 	CEPHFS_FEATURE_HAS_OWNER_UIDGID,
39 	CEPHFS_FEATURE_MDS_AUTH_CAPS_CHECK,
40 	CEPHFS_FEATURE_SUBVOLUME_METRICS,
41 
42 	CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_SUBVOLUME_METRICS,
43 };
44 
45 #define CEPHFS_FEATURES_CLIENT_SUPPORTED {	\
46 	0, 1, 2, 3, 4, 5, 6, 7,			\
47 	CEPHFS_FEATURE_MIMIC,			\
48 	CEPHFS_FEATURE_REPLY_ENCODING,		\
49 	CEPHFS_FEATURE_LAZY_CAP_WANTED,		\
50 	CEPHFS_FEATURE_MULTI_RECONNECT,		\
51 	CEPHFS_FEATURE_DELEG_INO,		\
52 	CEPHFS_FEATURE_METRIC_COLLECT,		\
53 	CEPHFS_FEATURE_ALTERNATE_NAME,		\
54 	CEPHFS_FEATURE_NOTIFY_SESSION_STATE,	\
55 	CEPHFS_FEATURE_OP_GETVXATTR,		\
56 	CEPHFS_FEATURE_32BITS_RETRY_FWD,	\
57 	CEPHFS_FEATURE_HAS_OWNER_UIDGID,	\
58 	CEPHFS_FEATURE_MDS_AUTH_CAPS_CHECK,	\
59 	CEPHFS_FEATURE_SUBVOLUME_METRICS,	\
60 }
61 
62 /*
63  * Some lock dependencies:
64  *
65  * session->s_mutex
66  *         mdsc->mutex
67  *
68  *         mdsc->snap_rwsem
69  *
70  *         ci->i_ceph_lock
71  *                 mdsc->snap_flush_lock
72  *                 mdsc->cap_delay_lock
73  *
74  */
75 
76 struct ceph_fs_client;
77 struct ceph_cap;
78 
79 #define MDS_AUTH_UID_ANY -1
80 #define CEPH_CAP_FLUSH_WAIT_TIMEOUT_SEC 60
81 #define CEPH_CAP_FLUSH_MAX_DUMP_ENTRIES 5
82 #define CEPH_CAP_FLUSH_MAX_DUMP_ITERS 5
83 #define CEPH_CLIENT_RESET_REASON_LEN	64
84 #define CEPH_CLIENT_RESET_DRAIN_SEC	30
85 #define CEPH_CLIENT_RESET_CLOSE_GRACE_MS 100
86 #define CEPH_CLIENT_RESET_WAIT_TIMEOUT_SEC 120
87 
88 enum ceph_client_reset_phase {
89 	CEPH_CLIENT_RESET_IDLE = 0,
90 	/*
91 	 * QUIESCING is set synchronously by schedule_reset() before the
92 	 * workqueue item is dispatched.  It gates new requests (any
93 	 * phase != IDLE blocks callers) during the window between
94 	 * scheduling and the work function's transition to DRAINING.
95 	 */
96 	CEPH_CLIENT_RESET_QUIESCING,
97 	CEPH_CLIENT_RESET_DRAINING,
98 	CEPH_CLIENT_RESET_TEARDOWN,
99 };
100 
101 struct ceph_client_reset_state {
102 	spinlock_t lock;		/* protects all fields below */
103 	u64 trigger_count;		/* number of resets triggered */
104 	u64 success_count;		/* number of successful resets */
105 	u64 failure_count;		/* number of failed resets */
106 	unsigned long last_start;	/* jiffies when last reset started */
107 	unsigned long last_finish;	/* jiffies when last reset finished */
108 	int last_errno;			/* result of most recent reset */
109 	enum ceph_client_reset_phase phase; /* current reset phase */
110 	bool drain_timed_out;		/* drain exceeded timeout */
111 	bool shutdown;			/* destroy in progress */
112 	int sessions_reset;		/* sessions torn down in last reset */
113 	char last_reason[CEPH_CLIENT_RESET_REASON_LEN]; /* operator-supplied reason */
114 
115 	/* Request blocking during reset */
116 	wait_queue_head_t blocked_wq;	/* waitqueue for blocked callers */
117 	atomic_t blocked_requests;	/* count of blocked callers */
118 };
119 
120 static inline bool ceph_reset_is_idle(struct ceph_client_reset_state *st)
121 {
122 	return READ_ONCE(st->phase) == CEPH_CLIENT_RESET_IDLE;
123 }
124 
125 struct ceph_mds_cap_match {
126 	s64 uid;  /* default to MDS_AUTH_UID_ANY */
127 	u32 num_gids;
128 	u32 *gids;  /* use these GIDs */
129 	char *path;  /* require path to be child of this
130 			(may be "" or "/" for any) */
131 	char *fs_name;
132 	bool root_squash;  /* default to false */
133 };
134 
135 struct ceph_mds_cap_auth {
136 	struct ceph_mds_cap_match match;
137 	bool readable;
138 	bool writeable;
139 };
140 
141 /*
142  * parsed info about a single inode.  pointers are into the encoded
143  * on-wire structures within the mds reply message payload.
144  */
145 struct ceph_mds_reply_info_in {
146 	struct ceph_mds_reply_inode *in;
147 	struct ceph_dir_layout dir_layout;
148 	u32 symlink_len;
149 	char *symlink;
150 	u32 xattr_len;
151 	char *xattr_data;
152 	u64 inline_version;
153 	u32 inline_len;
154 	char *inline_data;
155 	u32 pool_ns_len;
156 	char *pool_ns_data;
157 	u64 max_bytes;
158 	u64 max_files;
159 	s32 dir_pin;
160 	struct ceph_timespec btime;
161 	struct ceph_timespec snap_btime;
162 	u8 *fscrypt_auth;
163 	u8 *fscrypt_file;
164 	u32 fscrypt_auth_len;
165 	u32 fscrypt_file_len;
166 	u64 rsnaps;
167 	u64 change_attr;
168 	u64 subvolume_id;
169 };
170 
171 struct ceph_mds_reply_dir_entry {
172 	bool			      is_nokey;
173 	char                          *name;
174 	u32                           name_len;
175 	u32			      raw_hash;
176 	struct ceph_mds_reply_lease   *lease;
177 	struct ceph_mds_reply_info_in inode;
178 	loff_t			      offset;
179 };
180 
181 struct ceph_mds_reply_xattr {
182 	char *xattr_value;
183 	size_t xattr_value_len;
184 };
185 
186 /*
187  * parsed info about an mds reply, including information about
188  * either: 1) the target inode and/or its parent directory and dentry,
189  * and directory contents (for readdir results), or
190  * 2) the file range lock info (for fcntl F_GETLK results).
191  */
192 struct ceph_mds_reply_info_parsed {
193 	struct ceph_mds_reply_head    *head;
194 
195 	/* trace */
196 	struct ceph_mds_reply_info_in diri, targeti;
197 	struct ceph_mds_reply_dirfrag *dirfrag;
198 	char                          *dname;
199 	u8			      *altname;
200 	u32                           dname_len;
201 	u32                           altname_len;
202 	struct ceph_mds_reply_lease   *dlease;
203 	struct ceph_mds_reply_xattr   xattr_info;
204 
205 	/* extra */
206 	union {
207 		/* for fcntl F_GETLK results */
208 		struct ceph_filelock *filelock_reply;
209 
210 		/* for readdir results */
211 		struct {
212 			struct ceph_mds_reply_dirfrag *dir_dir;
213 			size_t			      dir_buf_size;
214 			int                           dir_nr;
215 			bool			      dir_end;
216 			bool			      dir_complete;
217 			bool			      hash_order;
218 			bool			      offset_hash;
219 			struct ceph_mds_reply_dir_entry  *dir_entries;
220 		};
221 
222 		/* for create results */
223 		struct {
224 			bool has_create_ino;
225 			u64 ino;
226 		};
227 	};
228 
229 	/* encoded blob describing snapshot contexts for certain
230 	   operations (e.g., open) */
231 	void *snapblob;
232 	int snapblob_len;
233 };
234 
235 
236 /*
237  * cap releases are batched and sent to the MDS en masse.
238  *
239  * Account for per-message overhead of mds_cap_release header
240  * and __le32 for osd epoch barrier trailing field.
241  */
242 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
243 				sizeof(struct ceph_mds_cap_release)) /	\
244 			        sizeof(struct ceph_mds_cap_item))
245 
246 
247 /*
248  * state associated with each MDS<->client session
249  */
250 enum {
251 	CEPH_MDS_SESSION_NEW = 1,
252 	CEPH_MDS_SESSION_OPENING = 2,
253 	CEPH_MDS_SESSION_OPEN = 3,
254 	CEPH_MDS_SESSION_HUNG = 4,
255 	CEPH_MDS_SESSION_RESTARTING = 5,
256 	CEPH_MDS_SESSION_RECONNECTING = 6,
257 	CEPH_MDS_SESSION_CLOSING = 7,
258 	CEPH_MDS_SESSION_CLOSED = 8,
259 	CEPH_MDS_SESSION_REJECTED = 9,
260 };
261 
262 struct ceph_mds_session {
263 	struct ceph_mds_client *s_mdsc;
264 	int               s_mds;
265 	int               s_state;
266 	unsigned long     s_ttl;      /* time until mds kills us */
267 	unsigned long	  s_features;
268 	u64               s_seq;      /* incoming msg seq # */
269 	struct mutex      s_mutex;    /* serialize session messages */
270 
271 	struct ceph_connection s_con;
272 
273 	struct ceph_auth_handshake s_auth;
274 
275 	atomic_t          s_cap_gen;  /* inc each time we get mds stale msg */
276 	unsigned long     s_cap_ttl;  /* when session caps expire. protected by s_mutex */
277 
278 	/* protected by s_cap_lock */
279 	spinlock_t        s_cap_lock;
280 	refcount_t        s_ref;
281 	struct list_head  s_caps;     /* all caps issued by this session */
282 	struct ceph_cap  *s_cap_iterator;
283 	int               s_nr_caps;
284 	int               s_num_cap_releases;
285 	int		  s_cap_reconnect;
286 	int		  s_readonly;
287 	struct list_head  s_cap_releases; /* waiting cap_release messages */
288 	struct work_struct s_cap_release_work;
289 
290 	/* See ceph_inode_info->i_dirty_item. */
291 	struct list_head  s_cap_dirty;	      /* inodes w/ dirty caps */
292 
293 	/* See ceph_inode_info->i_flushing_item. */
294 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
295 
296 	unsigned long     s_renew_requested; /* last time we sent a renew req */
297 	u64               s_renew_seq;
298 
299 	struct list_head  s_waiting;  /* waiting requests */
300 	struct list_head  s_unsafe;   /* unsafe requests */
301 	struct xarray	  s_delegated_inos;
302 };
303 
304 /*
305  * modes of choosing which MDS to send a request to
306  */
307 enum {
308 	USE_ANY_MDS,
309 	USE_RANDOM_MDS,
310 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
311 };
312 
313 struct ceph_mds_request;
314 struct ceph_mds_client;
315 
316 /*
317  * request completion callback
318  */
319 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
320 					     struct ceph_mds_request *req);
321 /*
322  * wait for request completion callback
323  */
324 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
325 						 struct ceph_mds_request *req);
326 
327 /*
328  * an in-flight mds request
329  */
330 struct ceph_mds_request {
331 	u64 r_tid;                   /* transaction id */
332 	struct rb_node r_node;
333 	struct ceph_mds_client *r_mdsc;
334 
335 	struct kref       r_kref;
336 	int r_op;                    /* mds op code */
337 
338 	/* operation on what? */
339 	struct inode *r_inode;              /* arg1 */
340 	struct dentry *r_dentry;            /* arg1 */
341 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
342 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
343 	char *r_path1, *r_path2;
344 	struct ceph_vino r_ino1, r_ino2;
345 
346 	struct inode *r_parent;		    /* parent dir inode */
347 	struct inode *r_target_inode;       /* resulting inode */
348 	struct inode *r_new_inode;	    /* new inode (for creates) */
349 
350 	const struct qstr *r_dname;	    /* stable name (for ->d_revalidate) */
351 
352 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
353 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
354 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
355 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
356 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
357 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
358 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
359 #define CEPH_MDS_R_ASYNC		(8) /* async request */
360 #define CEPH_MDS_R_FSCRYPT_FILE		(9) /* must marshal fscrypt_file field */
361 	unsigned long	r_req_flags;
362 
363 	struct mutex r_fill_mutex;
364 
365 	union ceph_mds_request_args r_args;
366 
367 	struct ceph_fscrypt_auth *r_fscrypt_auth;
368 	u64	r_fscrypt_file;
369 
370 	u8 *r_altname;		    /* fscrypt binary crypttext for long filenames */
371 	u32 r_altname_len;	    /* length of r_altname */
372 
373 	int r_fmode;        /* file mode, if expecting cap */
374 	int r_request_release_offset;
375 	const struct cred *r_cred;
376 	struct mnt_idmap *r_mnt_idmap;
377 	struct timespec64 r_stamp;
378 
379 	/* for choosing which mds to send this request to */
380 	int r_direct_mode;
381 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
382 
383 	/* data payload is used for xattr ops */
384 	struct ceph_pagelist *r_pagelist;
385 
386 	/* what caps shall we drop? */
387 	int r_inode_drop, r_inode_unless;
388 	int r_dentry_drop, r_dentry_unless;
389 	int r_old_dentry_drop, r_old_dentry_unless;
390 	struct inode *r_old_inode;
391 	int r_old_inode_drop, r_old_inode_unless;
392 
393 	struct ceph_msg  *r_request;  /* original request */
394 	struct ceph_msg  *r_reply;
395 	struct ceph_mds_reply_info_parsed r_reply_info;
396 	int r_err;
397 	u32               r_readdir_offset;
398 
399 	struct page *r_locked_page;
400 	int r_dir_caps;
401 	int r_num_caps;
402 
403 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
404 	unsigned long r_started;  /* start time to measure timeout against */
405 	unsigned long r_start_latency;  /* start time to measure latency */
406 	unsigned long r_end_latency;    /* finish time to measure latency */
407 	unsigned long r_request_started; /* start time for mds request only,
408 					    used to measure lease durations */
409 
410 	/* link unsafe requests to parent directory, for fsync */
411 	struct inode	*r_unsafe_dir;
412 	struct list_head r_unsafe_dir_item;
413 
414 	/* unsafe requests that modify the target inode */
415 	struct list_head r_unsafe_target_item;
416 
417 	struct ceph_mds_session *r_session;
418 
419 	int               r_attempts;   /* resend attempts */
420 	int               r_num_fwd;    /* number of forward attempts */
421 	int               r_resend_mds; /* mds to resend to next, if any*/
422 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
423 	u64		  r_deleg_ino;
424 
425 	struct list_head  r_wait;
426 	struct completion r_completion;
427 	struct completion r_safe_completion;
428 	ceph_mds_request_callback_t r_callback;
429 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
430 
431 	long long	  r_dir_release_cnt;
432 	long long	  r_dir_ordered_cnt;
433 	int		  r_readdir_cache_idx;
434 
435 	int		  r_feature_needed;
436 
437 	struct ceph_cap_reservation r_caps_reservation;
438 };
439 
440 struct ceph_pool_perm {
441 	struct rb_node node;
442 	int perm;
443 	s64 pool;
444 	size_t pool_ns_len;
445 	char pool_ns[];
446 };
447 
448 struct ceph_snapid_map {
449 	struct rb_node node;
450 	struct list_head lru;
451 	atomic_t ref;
452 	dev_t dev;
453 	u64 snap;
454 	unsigned long last_used;
455 };
456 
457 /*
458  * node for list of quotarealm inodes that are not visible from the filesystem
459  * mountpoint, but required to handle, e.g. quotas.
460  */
461 struct ceph_quotarealm_inode {
462 	struct rb_node node;
463 	u64 ino;
464 	unsigned long timeout; /* last time a lookup failed for this inode */
465 	struct mutex mutex;
466 	struct inode *inode;
467 };
468 
469 #ifdef CONFIG_DEBUG_FS
470 
471 struct cap_wait {
472 	struct list_head	list;
473 	u64			ino;
474 	pid_t			tgid;
475 	int			need;
476 	int			want;
477 };
478 
479 #endif
480 
481 enum {
482 	CEPH_MDSC_STOPPING_BEGIN = 1,
483 	CEPH_MDSC_STOPPING_FLUSHING = 2,
484 	CEPH_MDSC_STOPPING_FLUSHED = 3,
485 };
486 
487 /*
488  * mds client state
489  */
490 struct ceph_mds_client {
491 	struct ceph_fs_client  *fsc;
492 	struct mutex            mutex;         /* all nested structures */
493 
494 	struct ceph_mdsmap      *mdsmap;
495 	struct completion       safe_umount_waiters;
496 	wait_queue_head_t       session_close_wq;
497 	struct list_head        waiting_for_map;
498 	int 			mdsmap_err;
499 
500 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
501 	atomic_t		num_sessions;
502 	int                     max_sessions;  /* len of sessions array */
503 
504 	spinlock_t              stopping_lock;  /* protect snap_empty */
505 	int                     stopping;      /* the stage of shutting down */
506 	atomic_t                stopping_blockers;
507 	struct completion	stopping_waiter;
508 
509 	atomic64_t		dirty_folios;
510 	wait_queue_head_t	flush_end_wq;
511 
512 	atomic64_t		quotarealms_count; /* # realms with quota */
513 	/*
514 	 * We keep a list of inodes we don't see in the mountpoint but that we
515 	 * need to track quota realms.
516 	 */
517 	struct rb_root		quotarealms_inodes;
518 	struct mutex		quotarealms_inodes_mutex;
519 
520 	/*
521 	 * snap_rwsem will cover cap linkage into snaprealms, and
522 	 * realm snap contexts.  (later, we can do per-realm snap
523 	 * contexts locks..)  the empty list contains realms with no
524 	 * references (implying they contain no inodes with caps) that
525 	 * should be destroyed.
526 	 */
527 	u64			last_snap_seq;
528 	struct rw_semaphore     snap_rwsem;
529 	struct rb_root          snap_realms;
530 	struct list_head        snap_empty;
531 	int			num_snap_realms;
532 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
533 
534 	u64                    last_tid;      /* most recent mds request */
535 	u64                    oldest_tid;    /* oldest incomplete mds request,
536 						 excluding setfilelock requests */
537 	struct rb_root         request_tree;  /* pending mds requests */
538 	struct delayed_work    delayed_work;  /* delayed work */
539 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
540 	struct list_head cap_delay_list;   /* caps with delayed release */
541 	struct list_head cap_unlink_delay_list;  /* caps with delayed release for unlink */
542 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list and cap_unlink_delay_list */
543 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
544 	spinlock_t       snap_flush_lock;
545 
546 	u64               last_cap_flush_tid;
547 	struct list_head  cap_flush_list;
548 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
549 	int               num_cap_flushing; /* # caps we are flushing */
550 	spinlock_t        cap_dirty_lock;   /* protects above items */
551 	wait_queue_head_t cap_flushing_wq;
552 
553 	struct work_struct cap_reclaim_work;
554 	atomic_t	   cap_reclaim_pending;
555 
556 	struct work_struct cap_unlink_work;
557 
558 	/*
559 	 * Cap reservations
560 	 *
561 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
562 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
563 	 * memory needed to successfully process an MDS response.  (If an MDS
564 	 * sends us cap information and we fail to process it, we will have
565 	 * problems due to the client and MDS being out of sync.)
566 	 *
567 	 * Reservations are 'owned' by a ceph_cap_reservation context.
568 	 */
569 	spinlock_t	caps_list_lock;
570 	struct		list_head caps_list; /* unused (reserved or
571 						unreserved) */
572 #ifdef CONFIG_DEBUG_FS
573 	struct		list_head cap_wait_list;
574 #endif
575 	int		caps_total_count;    /* total caps allocated */
576 	int		caps_use_count;      /* in use */
577 	int		caps_use_max;	     /* max used caps */
578 	int		caps_reserve_count;  /* unused, reserved */
579 	int		caps_avail_count;    /* unused, unreserved */
580 	int		caps_min_count;      /* keep at least this many
581 						(unreserved) */
582 	spinlock_t	  dentry_list_lock;
583 	struct list_head  dentry_leases;     /* fifo list */
584 	struct list_head  dentry_dir_leases; /* lru list */
585 
586 	struct ceph_client_metric metric;
587 	struct work_struct	reset_work;
588 	struct ceph_client_reset_state reset_state;
589 	struct ceph_subvolume_metrics_tracker subvol_metrics;
590 
591 	/* Subvolume metrics send tracking */
592 	struct mutex		subvol_metrics_last_mutex;
593 	struct ceph_subvol_metric_snapshot *subvol_metrics_last;
594 	u32			subvol_metrics_last_nr;
595 	u64			subvol_metrics_sent;
596 	u64			subvol_metrics_nonzero_sends;
597 
598 	spinlock_t		snapid_map_lock;
599 	struct rb_root		snapid_map_tree;
600 	struct list_head	snapid_map_lru;
601 
602 	struct rw_semaphore     pool_perm_rwsem;
603 	struct rb_root		pool_perm_tree;
604 
605 	u32			 s_cap_auths_num;
606 	struct ceph_mds_cap_auth *s_cap_auths;
607 
608 	char nodename[__NEW_UTS_LEN + 1];
609 };
610 
611 extern const char *ceph_mds_op_name(int op);
612 
613 extern bool check_session_state(struct ceph_mds_session *s);
614 void inc_session_sequence(struct ceph_mds_session *s);
615 
616 extern struct ceph_mds_session *
617 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
618 
619 extern const char *ceph_session_state_name(int s);
620 extern const char *ceph_reset_phase_name(enum ceph_client_reset_phase phase);
621 
622 extern struct ceph_mds_session *
623 ceph_get_mds_session(struct ceph_mds_session *s);
624 extern void ceph_put_mds_session(struct ceph_mds_session *s);
625 int ceph_mdsc_schedule_reset(struct ceph_mds_client *mdsc,
626 			     const char *reason);
627 int ceph_mdsc_wait_for_reset(struct ceph_mds_client *mdsc);
628 
629 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
630 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
631 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
632 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
633 
634 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
635 
636 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
637 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
638 					   struct inode *dir);
639 extern struct ceph_mds_request *
640 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
641 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
642 				    struct inode *dir,
643 				    struct ceph_mds_request *req);
644 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
645 			struct ceph_mds_request *req,
646 			ceph_mds_request_wait_callback_t wait_func);
647 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
648 				struct inode *dir,
649 				struct ceph_mds_request *req);
650 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
651 extern void ceph_mdsc_release_dir_caps_async(struct ceph_mds_request *req);
652 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
653 {
654 	kref_get(&req->r_kref);
655 }
656 extern void ceph_mdsc_release_request(struct kref *kref);
657 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
658 {
659 	kref_put(&req->r_kref, ceph_mdsc_release_request);
660 }
661 
662 extern void send_flush_mdlog(struct ceph_mds_session *s);
663 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
664 				       void (*cb)(struct ceph_mds_session *),
665 				       bool check_state);
666 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
667 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
668 				    struct ceph_cap *cap);
669 extern void ceph_flush_session_cap_releases(struct ceph_mds_client *mdsc,
670 					    struct ceph_mds_session *session);
671 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
672 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
673 extern void ceph_queue_cap_unlink_work(struct ceph_mds_client *mdsc);
674 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
675 				     int (*cb)(struct inode *, int mds, void *),
676 				     void *arg);
677 extern int ceph_mds_check_access(struct ceph_mds_client *mdsc, char *tpath,
678 				 int mask);
679 
680 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
681 
682 /*
683  * Structure to group path-related output parameters for build_*_path functions
684  */
685 struct ceph_path_info {
686 	const char *path;
687 	int pathlen;
688 	struct ceph_vino vino;
689 	bool freepath;
690 };
691 
692 static inline void ceph_mdsc_free_path_info(const struct ceph_path_info *path_info)
693 {
694 	if (path_info && path_info->freepath && !IS_ERR_OR_NULL(path_info->path))
695 		__putname((char *)path_info->path - (PATH_MAX - 1 - path_info->pathlen));
696 }
697 
698 extern char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc,
699 				  struct dentry *dentry, struct ceph_path_info *path_info,
700 				  int for_wire);
701 
702 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
703 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
704 				     struct dentry *dentry, char action,
705 				     u32 seq);
706 
707 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
708 				    struct ceph_msg *msg);
709 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
710 				   struct ceph_msg *msg);
711 
712 extern struct ceph_mds_session *
713 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
714 
715 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
716 			  struct ceph_mds_session *session,
717 			  int max_caps);
718 
719 static inline int ceph_wait_on_async_create(struct inode *inode)
720 {
721 	struct ceph_inode_info *ci = ceph_inode(inode);
722 
723 	return wait_on_bit(&ci->i_ceph_flags, CEPH_I_ASYNC_CREATE_BIT,
724 			   TASK_KILLABLE);
725 }
726 
727 extern int ceph_wait_on_conflict_unlink(struct dentry *dentry);
728 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
729 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
730 
731 extern bool enable_unsafe_idmap;
732 #endif
733