xref: /linux/fs/ceph/mds_client.h (revision fab183d632628381b466a41479489541ac0e29a0)
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_GET_CAPS_WAIT_TIMEOUT (5 * HZ)
81 #define CEPH_CAP_FLUSH_WAIT_TIMEOUT_SEC 60
82 #define CEPH_CAP_FLUSH_MAX_DUMP_ENTRIES 5
83 #define CEPH_CAP_FLUSH_MAX_DUMP_ITERS 5
84 #define CEPH_CLIENT_RESET_REASON_LEN	64
85 #define CEPH_CLIENT_RESET_DRAIN_SEC	30
86 #define CEPH_CLIENT_RESET_CLOSE_GRACE_MS 100
87 #define CEPH_CLIENT_RESET_WAIT_TIMEOUT_SEC 120
88 
89 enum ceph_client_reset_phase {
90 	CEPH_CLIENT_RESET_IDLE = 0,
91 	/*
92 	 * QUIESCING is set synchronously by schedule_reset() before the
93 	 * workqueue item is dispatched.  It gates new requests (any
94 	 * phase != IDLE blocks callers) during the window between
95 	 * scheduling and the work function's transition to DRAINING.
96 	 */
97 	CEPH_CLIENT_RESET_QUIESCING,
98 	CEPH_CLIENT_RESET_DRAINING,
99 	CEPH_CLIENT_RESET_TEARDOWN,
100 };
101 
102 struct ceph_client_reset_state {
103 	spinlock_t lock;		/* protects all fields below */
104 	u64 trigger_count;		/* number of resets triggered */
105 	u64 success_count;		/* number of successful resets */
106 	u64 failure_count;		/* number of failed resets */
107 	unsigned long last_start;	/* jiffies when last reset started */
108 	unsigned long last_finish;	/* jiffies when last reset finished */
109 	int last_errno;			/* result of most recent reset */
110 	enum ceph_client_reset_phase phase; /* current reset phase */
111 	bool drain_timed_out;		/* drain exceeded timeout */
112 	bool shutdown;			/* destroy in progress */
113 	int sessions_reset;		/* sessions torn down in last reset */
114 	char last_reason[CEPH_CLIENT_RESET_REASON_LEN]; /* operator-supplied reason */
115 
116 	/* Request blocking during reset */
117 	wait_queue_head_t blocked_wq;	/* waitqueue for blocked callers */
118 	atomic_t blocked_requests;	/* count of blocked callers */
119 };
120 
ceph_reset_is_idle(struct ceph_client_reset_state * st)121 static inline bool ceph_reset_is_idle(struct ceph_client_reset_state *st)
122 {
123 	return READ_ONCE(st->phase) == CEPH_CLIENT_RESET_IDLE;
124 }
125 
126 struct ceph_mds_cap_match {
127 	s64 uid;  /* default to MDS_AUTH_UID_ANY */
128 	u32 num_gids;
129 	u32 *gids;  /* use these GIDs */
130 	char *path;  /* require path to be child of this
131 			(may be "" or "/" for any) */
132 	char *fs_name;
133 	bool root_squash;  /* default to false */
134 };
135 
136 struct ceph_mds_cap_auth {
137 	struct ceph_mds_cap_match match;
138 	bool readable;
139 	bool writeable;
140 };
141 
142 /*
143  * parsed info about a single inode.  pointers are into the encoded
144  * on-wire structures within the mds reply message payload.
145  */
146 struct ceph_mds_reply_info_in {
147 	struct ceph_mds_reply_inode *in;
148 	struct ceph_dir_layout dir_layout;
149 	u32 symlink_len;
150 	char *symlink;
151 	u32 xattr_len;
152 	char *xattr_data;
153 	u64 inline_version;
154 	u32 inline_len;
155 	char *inline_data;
156 	u32 pool_ns_len;
157 	char *pool_ns_data;
158 	u64 max_bytes;
159 	u64 max_files;
160 	s32 dir_pin;
161 	struct ceph_timespec btime;
162 	struct ceph_timespec snap_btime;
163 	u8 *fscrypt_auth;
164 	u8 *fscrypt_file;
165 	u32 fscrypt_auth_len;
166 	u32 fscrypt_file_len;
167 	u64 rsnaps;
168 	u64 change_attr;
169 	u64 subvolume_id;
170 };
171 
172 struct ceph_mds_reply_dir_entry {
173 	bool			      is_nokey;
174 	char                          *name;
175 	u32                           name_len;
176 	u32			      raw_hash;
177 	struct ceph_mds_reply_lease   *lease;
178 	struct ceph_mds_reply_info_in inode;
179 	loff_t			      offset;
180 };
181 
182 struct ceph_mds_reply_xattr {
183 	char *xattr_value;
184 	size_t xattr_value_len;
185 };
186 
187 /*
188  * parsed info about an mds reply, including information about
189  * either: 1) the target inode and/or its parent directory and dentry,
190  * and directory contents (for readdir results), or
191  * 2) the file range lock info (for fcntl F_GETLK results).
192  */
193 struct ceph_mds_reply_info_parsed {
194 	struct ceph_mds_reply_head    *head;
195 
196 	/* trace */
197 	struct ceph_mds_reply_info_in diri, targeti;
198 	struct ceph_mds_reply_dirfrag *dirfrag;
199 	char                          *dname;
200 	u8			      *altname;
201 	u32                           dname_len;
202 	u32                           altname_len;
203 	struct ceph_mds_reply_lease   *dlease;
204 	struct ceph_mds_reply_xattr   xattr_info;
205 
206 	/* extra */
207 	union {
208 		/* for fcntl F_GETLK results */
209 		struct ceph_filelock *filelock_reply;
210 
211 		/* for readdir results */
212 		struct {
213 			struct ceph_mds_reply_dirfrag *dir_dir;
214 			size_t			      dir_buf_size;
215 			int                           dir_nr;
216 			bool			      dir_end;
217 			bool			      dir_complete;
218 			bool			      hash_order;
219 			bool			      offset_hash;
220 			struct ceph_mds_reply_dir_entry  *dir_entries;
221 		};
222 
223 		/* for create results */
224 		struct {
225 			bool has_create_ino;
226 			u64 ino;
227 		};
228 	};
229 
230 	/* encoded blob describing snapshot contexts for certain
231 	   operations (e.g., open) */
232 	void *snapblob;
233 	int snapblob_len;
234 };
235 
236 
237 /*
238  * cap releases are batched and sent to the MDS en masse.
239  *
240  * Account for per-message overhead of mds_cap_release header
241  * and __le32 for osd epoch barrier trailing field.
242  */
243 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
244 				sizeof(struct ceph_mds_cap_release)) /	\
245 			        sizeof(struct ceph_mds_cap_item))
246 
247 
248 /*
249  * state associated with each MDS<->client session
250  */
251 enum {
252 	CEPH_MDS_SESSION_NEW = 1,
253 	CEPH_MDS_SESSION_OPENING = 2,
254 	CEPH_MDS_SESSION_OPEN = 3,
255 	CEPH_MDS_SESSION_HUNG = 4,
256 	CEPH_MDS_SESSION_RESTARTING = 5,
257 	CEPH_MDS_SESSION_RECONNECTING = 6,
258 	CEPH_MDS_SESSION_CLOSING = 7,
259 	CEPH_MDS_SESSION_CLOSED = 8,
260 	CEPH_MDS_SESSION_REJECTED = 9,
261 };
262 
263 struct ceph_mds_session {
264 	struct ceph_mds_client *s_mdsc;
265 	int               s_mds;
266 	int               s_state;
267 	unsigned long     s_ttl;      /* time until mds kills us */
268 	unsigned long	  s_features;
269 	u64               s_seq;      /* incoming msg seq # */
270 	struct mutex      s_mutex;    /* serialize session messages */
271 
272 	struct ceph_connection s_con;
273 
274 	struct ceph_auth_handshake s_auth;
275 
276 	atomic_t          s_cap_gen;  /* inc each time we get mds stale msg */
277 	unsigned long     s_cap_ttl;  /* when session caps expire. protected by s_mutex */
278 
279 	/* protected by s_cap_lock */
280 	spinlock_t        s_cap_lock;
281 	refcount_t        s_ref;
282 	struct list_head  s_caps;     /* all caps issued by this session */
283 	struct ceph_cap  *s_cap_iterator;
284 	int               s_nr_caps;
285 	int               s_num_cap_releases;
286 	int		  s_cap_reconnect;
287 	int		  s_readonly;
288 	struct list_head  s_cap_releases; /* waiting cap_release messages */
289 	struct work_struct s_cap_release_work;
290 
291 	/* See ceph_inode_info->i_dirty_item. */
292 	struct list_head  s_cap_dirty;	      /* inodes w/ dirty caps */
293 
294 	/* See ceph_inode_info->i_flushing_item. */
295 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
296 
297 	unsigned long     s_renew_requested; /* last time we sent a renew req */
298 	u64               s_renew_seq;
299 
300 	struct list_head  s_waiting;  /* waiting requests */
301 	struct list_head  s_unsafe;   /* unsafe requests */
302 	struct xarray	  s_delegated_inos;
303 };
304 
305 /*
306  * modes of choosing which MDS to send a request to
307  */
308 enum {
309 	USE_ANY_MDS,
310 	USE_RANDOM_MDS,
311 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
312 };
313 
314 struct ceph_mds_request;
315 struct ceph_mds_client;
316 
317 /*
318  * request completion callback
319  */
320 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
321 					     struct ceph_mds_request *req);
322 /*
323  * wait for request completion callback
324  */
325 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
326 						 struct ceph_mds_request *req);
327 
328 /*
329  * an in-flight mds request
330  */
331 struct ceph_mds_request {
332 	u64 r_tid;                   /* transaction id */
333 	struct rb_node r_node;
334 	struct ceph_mds_client *r_mdsc;
335 
336 	struct kref       r_kref;
337 	int r_op;                    /* mds op code */
338 
339 	/* operation on what? */
340 	struct inode *r_inode;              /* arg1 */
341 	struct dentry *r_dentry;            /* arg1 */
342 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
343 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
344 	char *r_path1, *r_path2;
345 	struct ceph_vino r_ino1, r_ino2;
346 
347 	struct inode *r_parent;		    /* parent dir inode */
348 	struct inode *r_target_inode;       /* resulting inode */
349 	struct inode *r_new_inode;	    /* new inode (for creates) */
350 
351 	const struct qstr *r_dname;	    /* stable name (for ->d_revalidate) */
352 
353 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
354 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
355 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
356 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
357 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
358 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
359 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
360 #define CEPH_MDS_R_ASYNC		(8) /* async request */
361 #define CEPH_MDS_R_FSCRYPT_FILE		(9) /* must marshal fscrypt_file field */
362 	unsigned long	r_req_flags;
363 
364 	struct mutex r_fill_mutex;
365 
366 	union ceph_mds_request_args r_args;
367 
368 	struct ceph_fscrypt_auth *r_fscrypt_auth;
369 	u64	r_fscrypt_file;
370 
371 	u8 *r_altname;		    /* fscrypt binary crypttext for long filenames */
372 	u32 r_altname_len;	    /* length of r_altname */
373 
374 	int r_fmode;        /* file mode, if expecting cap */
375 	int r_request_release_offset;
376 	const struct cred *r_cred;
377 	struct mnt_idmap *r_mnt_idmap;
378 	struct timespec64 r_stamp;
379 
380 	/* for choosing which mds to send this request to */
381 	int r_direct_mode;
382 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
383 
384 	/* data payload is used for xattr ops */
385 	struct ceph_pagelist *r_pagelist;
386 
387 	/* what caps shall we drop? */
388 	int r_inode_drop, r_inode_unless;
389 	int r_dentry_drop, r_dentry_unless;
390 	int r_old_dentry_drop, r_old_dentry_unless;
391 	struct inode *r_old_inode;
392 	int r_old_inode_drop, r_old_inode_unless;
393 
394 	struct ceph_msg  *r_request;  /* original request */
395 	struct ceph_msg  *r_reply;
396 	struct ceph_mds_reply_info_parsed r_reply_info;
397 	int r_err;
398 	u32               r_readdir_offset;
399 
400 	struct page *r_locked_page;
401 	int r_dir_caps;
402 	int r_num_caps;
403 
404 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
405 	unsigned long r_started;  /* start time to measure timeout against */
406 	unsigned long r_start_latency;  /* start time to measure latency */
407 	unsigned long r_end_latency;    /* finish time to measure latency */
408 	unsigned long r_request_started; /* start time for mds request only,
409 					    used to measure lease durations */
410 
411 	/* link unsafe requests to parent directory, for fsync */
412 	struct inode	*r_unsafe_dir;
413 	struct list_head r_unsafe_dir_item;
414 
415 	/* unsafe requests that modify the target inode */
416 	struct list_head r_unsafe_target_item;
417 
418 	struct ceph_mds_session *r_session;
419 
420 	int               r_attempts;   /* resend attempts */
421 	int               r_num_fwd;    /* number of forward attempts */
422 	int               r_resend_mds; /* mds to resend to next, if any*/
423 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
424 	u64		  r_deleg_ino;
425 
426 	struct list_head  r_wait;
427 	struct completion r_completion;
428 	struct completion r_safe_completion;
429 	ceph_mds_request_callback_t r_callback;
430 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
431 
432 	long long	  r_dir_release_cnt;
433 	long long	  r_dir_ordered_cnt;
434 	int		  r_readdir_cache_idx;
435 
436 	int		  r_feature_needed;
437 
438 	struct ceph_cap_reservation r_caps_reservation;
439 };
440 
441 struct ceph_pool_perm {
442 	struct rb_node node;
443 	int perm;
444 	s64 pool;
445 	size_t pool_ns_len;
446 	char pool_ns[];
447 };
448 
449 struct ceph_snapid_map {
450 	struct rb_node node;
451 	struct list_head lru;
452 	atomic_t ref;
453 	dev_t dev;
454 	u64 snap;
455 	unsigned long last_used;
456 };
457 
458 /*
459  * node for list of quotarealm inodes that are not visible from the filesystem
460  * mountpoint, but required to handle, e.g. quotas.
461  */
462 struct ceph_quotarealm_inode {
463 	struct rb_node node;
464 	u64 ino;
465 	unsigned long timeout; /* last time a lookup failed for this inode */
466 	struct mutex mutex;
467 	struct inode *inode;
468 };
469 
470 #ifdef CONFIG_DEBUG_FS
471 
472 struct cap_wait {
473 	struct list_head	list;
474 	u64			ino;
475 	pid_t			tgid;
476 	int			need;
477 	int			want;
478 };
479 
480 #endif
481 
482 enum {
483 	CEPH_MDSC_STOPPING_BEGIN = 1,
484 	CEPH_MDSC_STOPPING_FLUSHING = 2,
485 	CEPH_MDSC_STOPPING_FLUSHED = 3,
486 };
487 
488 /*
489  * mds client state
490  */
491 struct ceph_mds_client {
492 	struct ceph_fs_client  *fsc;
493 	struct mutex            mutex;         /* all nested structures */
494 
495 	struct ceph_mdsmap      *mdsmap;
496 	struct completion       safe_umount_waiters;
497 	wait_queue_head_t       session_close_wq;
498 	struct list_head        waiting_for_map;
499 	int 			mdsmap_err;
500 
501 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
502 	atomic_t		num_sessions;
503 	int                     max_sessions;  /* len of sessions array */
504 
505 	spinlock_t              stopping_lock;  /* protect snap_empty */
506 	int                     stopping;      /* the stage of shutting down */
507 	atomic_t                stopping_blockers;
508 	struct completion	stopping_waiter;
509 
510 	atomic64_t		dirty_folios;
511 	wait_queue_head_t	flush_end_wq;
512 
513 	atomic64_t		quotarealms_count; /* # realms with quota */
514 	/*
515 	 * We keep a list of inodes we don't see in the mountpoint but that we
516 	 * need to track quota realms.
517 	 */
518 	struct rb_root		quotarealms_inodes;
519 	struct mutex		quotarealms_inodes_mutex;
520 
521 	/*
522 	 * snap_rwsem will cover cap linkage into snaprealms, and
523 	 * realm snap contexts.  (later, we can do per-realm snap
524 	 * contexts locks..)  the empty list contains realms with no
525 	 * references (implying they contain no inodes with caps) that
526 	 * should be destroyed.
527 	 */
528 	u64			last_snap_seq;
529 	struct rw_semaphore     snap_rwsem;
530 	struct rb_root          snap_realms;
531 	struct list_head        snap_empty;
532 	int			num_snap_realms;
533 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
534 
535 	u64                    last_tid;      /* most recent mds request */
536 	u64                    oldest_tid;    /* oldest incomplete mds request,
537 						 excluding setfilelock requests */
538 	struct rb_root         request_tree;  /* pending mds requests */
539 	struct delayed_work    delayed_work;  /* delayed work */
540 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
541 	struct list_head cap_delay_list;   /* caps with delayed release */
542 	struct list_head cap_unlink_delay_list;  /* caps with delayed release for unlink */
543 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list and cap_unlink_delay_list */
544 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
545 	spinlock_t       snap_flush_lock;
546 
547 	u64               last_cap_flush_tid;
548 	struct list_head  cap_flush_list;
549 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
550 	int               num_cap_flushing; /* # caps we are flushing */
551 	spinlock_t        cap_dirty_lock;   /* protects above items */
552 	wait_queue_head_t cap_flushing_wq;
553 
554 	struct work_struct cap_reclaim_work;
555 	atomic_t	   cap_reclaim_pending;
556 
557 	struct work_struct cap_unlink_work;
558 
559 	/*
560 	 * Cap reservations
561 	 *
562 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
563 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
564 	 * memory needed to successfully process an MDS response.  (If an MDS
565 	 * sends us cap information and we fail to process it, we will have
566 	 * problems due to the client and MDS being out of sync.)
567 	 *
568 	 * Reservations are 'owned' by a ceph_cap_reservation context.
569 	 */
570 	spinlock_t	caps_list_lock;
571 	struct		list_head caps_list; /* unused (reserved or
572 						unreserved) */
573 #ifdef CONFIG_DEBUG_FS
574 	struct		list_head cap_wait_list;
575 #endif
576 	int		caps_total_count;    /* total caps allocated */
577 	int		caps_use_count;      /* in use */
578 	int		caps_use_max;	     /* max used caps */
579 	int		caps_reserve_count;  /* unused, reserved */
580 	int		caps_avail_count;    /* unused, unreserved */
581 	int		caps_min_count;      /* keep at least this many
582 						(unreserved) */
583 	spinlock_t	  dentry_list_lock;
584 	struct list_head  dentry_leases;     /* fifo list */
585 	struct list_head  dentry_dir_leases; /* lru list */
586 
587 	struct ceph_client_metric metric;
588 	struct work_struct	reset_work;
589 	struct ceph_client_reset_state reset_state;
590 	struct ceph_subvolume_metrics_tracker subvol_metrics;
591 
592 	/* Subvolume metrics send tracking */
593 	struct mutex		subvol_metrics_last_mutex;
594 	struct ceph_subvol_metric_snapshot *subvol_metrics_last;
595 	u32			subvol_metrics_last_nr;
596 	u64			subvol_metrics_sent;
597 	u64			subvol_metrics_nonzero_sends;
598 
599 	spinlock_t		snapid_map_lock;
600 	struct rb_root		snapid_map_tree;
601 	struct list_head	snapid_map_lru;
602 
603 	struct rw_semaphore     pool_perm_rwsem;
604 	struct rb_root		pool_perm_tree;
605 
606 	u32			 s_cap_auths_num;
607 	struct ceph_mds_cap_auth *s_cap_auths;
608 
609 	char nodename[__NEW_UTS_LEN + 1];
610 };
611 
612 extern const char *ceph_mds_op_name(int op);
613 
614 extern bool check_session_state(struct ceph_mds_session *s);
615 void inc_session_sequence(struct ceph_mds_session *s);
616 
617 extern struct ceph_mds_session *
618 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
619 
620 extern const char *ceph_session_state_name(int s);
621 extern const char *ceph_reset_phase_name(enum ceph_client_reset_phase phase);
622 
623 extern struct ceph_mds_session *
624 ceph_get_mds_session(struct ceph_mds_session *s);
625 extern void ceph_put_mds_session(struct ceph_mds_session *s);
626 int ceph_mdsc_schedule_reset(struct ceph_mds_client *mdsc,
627 			     const char *reason);
628 int ceph_mdsc_wait_for_reset(struct ceph_mds_client *mdsc);
629 
630 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
631 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
632 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
633 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
634 
635 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
636 
637 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
638 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
639 					   struct inode *dir);
640 extern struct ceph_mds_request *
641 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
642 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
643 				    struct inode *dir,
644 				    struct ceph_mds_request *req);
645 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
646 			struct ceph_mds_request *req,
647 			ceph_mds_request_wait_callback_t wait_func);
648 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
649 				struct inode *dir,
650 				struct ceph_mds_request *req);
651 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
652 extern void ceph_mdsc_release_dir_caps_async(struct ceph_mds_request *req);
ceph_mdsc_get_request(struct ceph_mds_request * req)653 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
654 {
655 	kref_get(&req->r_kref);
656 }
657 extern void ceph_mdsc_release_request(struct kref *kref);
ceph_mdsc_put_request(struct ceph_mds_request * req)658 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
659 {
660 	kref_put(&req->r_kref, ceph_mdsc_release_request);
661 }
662 
663 extern void send_flush_mdlog(struct ceph_mds_session *s);
664 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
665 				       void (*cb)(struct ceph_mds_session *),
666 				       bool check_state);
667 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
668 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
669 				    struct ceph_cap *cap);
670 extern void ceph_flush_session_cap_releases(struct ceph_mds_client *mdsc,
671 					    struct ceph_mds_session *session);
672 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
673 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
674 extern void ceph_queue_cap_unlink_work(struct ceph_mds_client *mdsc);
675 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
676 				     int (*cb)(struct inode *, int mds, void *),
677 				     void *arg);
678 extern int ceph_mds_check_access(struct ceph_mds_client *mdsc, char *tpath,
679 				 int mask);
680 
681 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
682 
683 /*
684  * Structure to group path-related output parameters for build_*_path functions
685  */
686 struct ceph_path_info {
687 	const char *path;
688 	int pathlen;
689 	struct ceph_vino vino;
690 	bool freepath;
691 };
692 
ceph_mdsc_free_path_info(const struct ceph_path_info * path_info)693 static inline void ceph_mdsc_free_path_info(const struct ceph_path_info *path_info)
694 {
695 	if (path_info && path_info->freepath && !IS_ERR_OR_NULL(path_info->path))
696 		__putname((char *)path_info->path - (PATH_MAX - 1 - path_info->pathlen));
697 }
698 
699 extern char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc,
700 				  struct dentry *dentry, struct ceph_path_info *path_info,
701 				  int for_wire);
702 
703 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
704 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
705 				     struct dentry *dentry, char action,
706 				     u32 seq);
707 
708 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
709 				    struct ceph_msg *msg);
710 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
711 				   struct ceph_msg *msg);
712 
713 extern struct ceph_mds_session *
714 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
715 
716 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
717 			  struct ceph_mds_session *session,
718 			  int max_caps);
719 
ceph_wait_on_async_create(struct inode * inode)720 static inline int ceph_wait_on_async_create(struct inode *inode)
721 {
722 	struct ceph_inode_info *ci = ceph_inode(inode);
723 
724 	return wait_on_bit(&ci->i_ceph_flags, CEPH_I_ASYNC_CREATE_BIT,
725 			   TASK_KILLABLE);
726 }
727 
728 extern int ceph_wait_on_conflict_unlink(struct dentry *dentry);
729 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
730 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
731 
732 extern bool enable_unsafe_idmap;
733 #endif
734