xref: /linux/fs/ceph/mds_client.h (revision f788bb2fb915a79eae0a160580766ba52984d61f)
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/mdsmap.h>
18 #include <linux/ceph/auth.h>
19 
20 #include "metric.h"
21 
22 /* The first 8 bits are reserved for old ceph releases */
23 enum ceph_feature_type {
24 	CEPHFS_FEATURE_MIMIC = 8,
25 	CEPHFS_FEATURE_REPLY_ENCODING,
26 	CEPHFS_FEATURE_RECLAIM_CLIENT,
27 	CEPHFS_FEATURE_LAZY_CAP_WANTED,
28 	CEPHFS_FEATURE_MULTI_RECONNECT,
29 	CEPHFS_FEATURE_DELEG_INO,
30 
31 	CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_DELEG_INO,
32 };
33 
34 /*
35  * This will always have the highest feature bit value
36  * as the last element of the array.
37  */
38 #define CEPHFS_FEATURES_CLIENT_SUPPORTED {	\
39 	0, 1, 2, 3, 4, 5, 6, 7,			\
40 	CEPHFS_FEATURE_MIMIC,			\
41 	CEPHFS_FEATURE_REPLY_ENCODING,		\
42 	CEPHFS_FEATURE_LAZY_CAP_WANTED,		\
43 	CEPHFS_FEATURE_MULTI_RECONNECT,		\
44 	CEPHFS_FEATURE_DELEG_INO,		\
45 						\
46 	CEPHFS_FEATURE_MAX,			\
47 }
48 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
49 
50 /*
51  * Some lock dependencies:
52  *
53  * session->s_mutex
54  *         mdsc->mutex
55  *
56  *         mdsc->snap_rwsem
57  *
58  *         ci->i_ceph_lock
59  *                 mdsc->snap_flush_lock
60  *                 mdsc->cap_delay_lock
61  *
62  */
63 
64 struct ceph_fs_client;
65 struct ceph_cap;
66 
67 /*
68  * parsed info about a single inode.  pointers are into the encoded
69  * on-wire structures within the mds reply message payload.
70  */
71 struct ceph_mds_reply_info_in {
72 	struct ceph_mds_reply_inode *in;
73 	struct ceph_dir_layout dir_layout;
74 	u32 symlink_len;
75 	char *symlink;
76 	u32 xattr_len;
77 	char *xattr_data;
78 	u64 inline_version;
79 	u32 inline_len;
80 	char *inline_data;
81 	u32 pool_ns_len;
82 	char *pool_ns_data;
83 	u64 max_bytes;
84 	u64 max_files;
85 	s32 dir_pin;
86 	struct ceph_timespec btime;
87 	struct ceph_timespec snap_btime;
88 	u64 change_attr;
89 };
90 
91 struct ceph_mds_reply_dir_entry {
92 	char                          *name;
93 	u32                           name_len;
94 	struct ceph_mds_reply_lease   *lease;
95 	struct ceph_mds_reply_info_in inode;
96 	loff_t			      offset;
97 };
98 
99 /*
100  * parsed info about an mds reply, including information about
101  * either: 1) the target inode and/or its parent directory and dentry,
102  * and directory contents (for readdir results), or
103  * 2) the file range lock info (for fcntl F_GETLK results).
104  */
105 struct ceph_mds_reply_info_parsed {
106 	struct ceph_mds_reply_head    *head;
107 
108 	/* trace */
109 	struct ceph_mds_reply_info_in diri, targeti;
110 	struct ceph_mds_reply_dirfrag *dirfrag;
111 	char                          *dname;
112 	u32                           dname_len;
113 	struct ceph_mds_reply_lease   *dlease;
114 
115 	/* extra */
116 	union {
117 		/* for fcntl F_GETLK results */
118 		struct ceph_filelock *filelock_reply;
119 
120 		/* for readdir results */
121 		struct {
122 			struct ceph_mds_reply_dirfrag *dir_dir;
123 			size_t			      dir_buf_size;
124 			int                           dir_nr;
125 			bool			      dir_end;
126 			bool			      dir_complete;
127 			bool			      hash_order;
128 			bool			      offset_hash;
129 			struct ceph_mds_reply_dir_entry  *dir_entries;
130 		};
131 
132 		/* for create results */
133 		struct {
134 			bool has_create_ino;
135 			u64 ino;
136 		};
137 	};
138 
139 	/* encoded blob describing snapshot contexts for certain
140 	   operations (e.g., open) */
141 	void *snapblob;
142 	int snapblob_len;
143 };
144 
145 
146 /*
147  * cap releases are batched and sent to the MDS en masse.
148  *
149  * Account for per-message overhead of mds_cap_release header
150  * and __le32 for osd epoch barrier trailing field.
151  */
152 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
153 				sizeof(struct ceph_mds_cap_release)) /	\
154 			        sizeof(struct ceph_mds_cap_item))
155 
156 
157 /*
158  * state associated with each MDS<->client session
159  */
160 enum {
161 	CEPH_MDS_SESSION_NEW = 1,
162 	CEPH_MDS_SESSION_OPENING = 2,
163 	CEPH_MDS_SESSION_OPEN = 3,
164 	CEPH_MDS_SESSION_HUNG = 4,
165 	CEPH_MDS_SESSION_RESTARTING = 5,
166 	CEPH_MDS_SESSION_RECONNECTING = 6,
167 	CEPH_MDS_SESSION_CLOSING = 7,
168 	CEPH_MDS_SESSION_CLOSED = 8,
169 	CEPH_MDS_SESSION_REJECTED = 9,
170 };
171 
172 struct ceph_mds_session {
173 	struct ceph_mds_client *s_mdsc;
174 	int               s_mds;
175 	int               s_state;
176 	unsigned long     s_ttl;      /* time until mds kills us */
177 	unsigned long	  s_features;
178 	u64               s_seq;      /* incoming msg seq # */
179 	struct mutex      s_mutex;    /* serialize session messages */
180 
181 	struct ceph_connection s_con;
182 
183 	struct ceph_auth_handshake s_auth;
184 
185 	/* protected by s_gen_ttl_lock */
186 	spinlock_t        s_gen_ttl_lock;
187 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
188 	unsigned long     s_cap_ttl;  /* when session caps expire */
189 
190 	/* protected by s_cap_lock */
191 	spinlock_t        s_cap_lock;
192 	refcount_t        s_ref;
193 	struct list_head  s_caps;     /* all caps issued by this session */
194 	struct ceph_cap  *s_cap_iterator;
195 	int               s_nr_caps;
196 	int               s_num_cap_releases;
197 	int		  s_cap_reconnect;
198 	int		  s_readonly;
199 	struct list_head  s_cap_releases; /* waiting cap_release messages */
200 	struct work_struct s_cap_release_work;
201 
202 	/* See ceph_inode_info->i_dirty_item. */
203 	struct list_head  s_cap_dirty;	      /* inodes w/ dirty caps */
204 
205 	/* See ceph_inode_info->i_flushing_item. */
206 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
207 
208 	unsigned long     s_renew_requested; /* last time we sent a renew req */
209 	u64               s_renew_seq;
210 
211 	struct list_head  s_waiting;  /* waiting requests */
212 	struct list_head  s_unsafe;   /* unsafe requests */
213 	struct xarray	  s_delegated_inos;
214 };
215 
216 /*
217  * modes of choosing which MDS to send a request to
218  */
219 enum {
220 	USE_ANY_MDS,
221 	USE_RANDOM_MDS,
222 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
223 };
224 
225 struct ceph_mds_request;
226 struct ceph_mds_client;
227 
228 /*
229  * request completion callback
230  */
231 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
232 					     struct ceph_mds_request *req);
233 /*
234  * wait for request completion callback
235  */
236 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
237 						 struct ceph_mds_request *req);
238 
239 /*
240  * an in-flight mds request
241  */
242 struct ceph_mds_request {
243 	u64 r_tid;                   /* transaction id */
244 	struct rb_node r_node;
245 	struct ceph_mds_client *r_mdsc;
246 
247 	struct kref       r_kref;
248 	int r_op;                    /* mds op code */
249 
250 	/* operation on what? */
251 	struct inode *r_inode;              /* arg1 */
252 	struct dentry *r_dentry;            /* arg1 */
253 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
254 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
255 	char *r_path1, *r_path2;
256 	struct ceph_vino r_ino1, r_ino2;
257 
258 	struct inode *r_parent;		    /* parent dir inode */
259 	struct inode *r_target_inode;       /* resulting inode */
260 
261 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
262 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
263 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
264 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
265 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
266 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
267 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
268 #define CEPH_MDS_R_ASYNC		(8) /* async request */
269 	unsigned long	r_req_flags;
270 
271 	struct mutex r_fill_mutex;
272 
273 	union ceph_mds_request_args r_args;
274 	int r_fmode;        /* file mode, if expecting cap */
275 	kuid_t r_uid;
276 	kgid_t r_gid;
277 	int r_request_release_offset;
278 	struct timespec64 r_stamp;
279 
280 	/* for choosing which mds to send this request to */
281 	int r_direct_mode;
282 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
283 
284 	/* data payload is used for xattr ops */
285 	struct ceph_pagelist *r_pagelist;
286 
287 	/* what caps shall we drop? */
288 	int r_inode_drop, r_inode_unless;
289 	int r_dentry_drop, r_dentry_unless;
290 	int r_old_dentry_drop, r_old_dentry_unless;
291 	struct inode *r_old_inode;
292 	int r_old_inode_drop, r_old_inode_unless;
293 
294 	struct ceph_msg  *r_request;  /* original request */
295 	struct ceph_msg  *r_reply;
296 	struct ceph_mds_reply_info_parsed r_reply_info;
297 	int r_err;
298 
299 
300 	struct page *r_locked_page;
301 	int r_dir_caps;
302 	int r_num_caps;
303 	u32               r_readdir_offset;
304 
305 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
306 	unsigned long r_started;  /* start time to measure timeout against */
307 	unsigned long r_start_latency;  /* start time to measure latency */
308 	unsigned long r_end_latency;    /* finish time to measure latency */
309 	unsigned long r_request_started; /* start time for mds request only,
310 					    used to measure lease durations */
311 
312 	/* link unsafe requests to parent directory, for fsync */
313 	struct inode	*r_unsafe_dir;
314 	struct list_head r_unsafe_dir_item;
315 
316 	/* unsafe requests that modify the target inode */
317 	struct list_head r_unsafe_target_item;
318 
319 	struct ceph_mds_session *r_session;
320 
321 	int               r_attempts;   /* resend attempts */
322 	int               r_num_fwd;    /* number of forward attempts */
323 	int               r_resend_mds; /* mds to resend to next, if any*/
324 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
325 	u64		  r_deleg_ino;
326 
327 	struct list_head  r_wait;
328 	struct completion r_completion;
329 	struct completion r_safe_completion;
330 	ceph_mds_request_callback_t r_callback;
331 	ceph_mds_request_wait_callback_t r_wait_for_completion;
332 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
333 
334 	long long	  r_dir_release_cnt;
335 	long long	  r_dir_ordered_cnt;
336 	int		  r_readdir_cache_idx;
337 
338 	struct ceph_cap_reservation r_caps_reservation;
339 };
340 
341 struct ceph_pool_perm {
342 	struct rb_node node;
343 	int perm;
344 	s64 pool;
345 	size_t pool_ns_len;
346 	char pool_ns[];
347 };
348 
349 struct ceph_snapid_map {
350 	struct rb_node node;
351 	struct list_head lru;
352 	atomic_t ref;
353 	u64 snap;
354 	dev_t dev;
355 	unsigned long last_used;
356 };
357 
358 /*
359  * node for list of quotarealm inodes that are not visible from the filesystem
360  * mountpoint, but required to handle, e.g. quotas.
361  */
362 struct ceph_quotarealm_inode {
363 	struct rb_node node;
364 	u64 ino;
365 	unsigned long timeout; /* last time a lookup failed for this inode */
366 	struct mutex mutex;
367 	struct inode *inode;
368 };
369 
370 struct cap_wait {
371 	struct list_head	list;
372 	unsigned long		ino;
373 	pid_t			tgid;
374 	int			need;
375 	int			want;
376 };
377 
378 /*
379  * mds client state
380  */
381 struct ceph_mds_client {
382 	struct ceph_fs_client  *fsc;
383 	struct mutex            mutex;         /* all nested structures */
384 
385 	struct ceph_mdsmap      *mdsmap;
386 	struct completion       safe_umount_waiters;
387 	wait_queue_head_t       session_close_wq;
388 	struct list_head        waiting_for_map;
389 	int 			mdsmap_err;
390 
391 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
392 	atomic_t		num_sessions;
393 	int                     max_sessions;  /* len of s_mds_sessions */
394 	int                     stopping;      /* true if shutting down */
395 
396 	atomic64_t		quotarealms_count; /* # realms with quota */
397 	/*
398 	 * We keep a list of inodes we don't see in the mountpoint but that we
399 	 * need to track quota realms.
400 	 */
401 	struct rb_root		quotarealms_inodes;
402 	struct mutex		quotarealms_inodes_mutex;
403 
404 	/*
405 	 * snap_rwsem will cover cap linkage into snaprealms, and
406 	 * realm snap contexts.  (later, we can do per-realm snap
407 	 * contexts locks..)  the empty list contains realms with no
408 	 * references (implying they contain no inodes with caps) that
409 	 * should be destroyed.
410 	 */
411 	u64			last_snap_seq;
412 	struct rw_semaphore     snap_rwsem;
413 	struct rb_root          snap_realms;
414 	struct list_head        snap_empty;
415 	int			num_snap_realms;
416 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
417 
418 	u64                    last_tid;      /* most recent mds request */
419 	u64                    oldest_tid;    /* oldest incomplete mds request,
420 						 excluding setfilelock requests */
421 	struct rb_root         request_tree;  /* pending mds requests */
422 	struct delayed_work    delayed_work;  /* delayed work */
423 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
424 	struct list_head cap_delay_list;   /* caps with delayed release */
425 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
426 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
427 	spinlock_t       snap_flush_lock;
428 
429 	u64               last_cap_flush_tid;
430 	struct list_head  cap_flush_list;
431 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
432 	int               num_cap_flushing; /* # caps we are flushing */
433 	spinlock_t        cap_dirty_lock;   /* protects above items */
434 	wait_queue_head_t cap_flushing_wq;
435 
436 	struct work_struct cap_reclaim_work;
437 	atomic_t	   cap_reclaim_pending;
438 
439 	/*
440 	 * Cap reservations
441 	 *
442 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
443 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
444 	 * memory needed to successfully process an MDS response.  (If an MDS
445 	 * sends us cap information and we fail to process it, we will have
446 	 * problems due to the client and MDS being out of sync.)
447 	 *
448 	 * Reservations are 'owned' by a ceph_cap_reservation context.
449 	 */
450 	spinlock_t	caps_list_lock;
451 	struct		list_head caps_list; /* unused (reserved or
452 						unreserved) */
453 	struct		list_head cap_wait_list;
454 	int		caps_total_count;    /* total caps allocated */
455 	int		caps_use_count;      /* in use */
456 	int		caps_use_max;	     /* max used caps */
457 	int		caps_reserve_count;  /* unused, reserved */
458 	int		caps_avail_count;    /* unused, unreserved */
459 	int		caps_min_count;      /* keep at least this many
460 						(unreserved) */
461 	spinlock_t	  dentry_list_lock;
462 	struct list_head  dentry_leases;     /* fifo list */
463 	struct list_head  dentry_dir_leases; /* lru list */
464 
465 	struct ceph_client_metric metric;
466 
467 	spinlock_t		snapid_map_lock;
468 	struct rb_root		snapid_map_tree;
469 	struct list_head	snapid_map_lru;
470 
471 	struct rw_semaphore     pool_perm_rwsem;
472 	struct rb_root		pool_perm_tree;
473 
474 	char nodename[__NEW_UTS_LEN + 1];
475 };
476 
477 extern const char *ceph_mds_op_name(int op);
478 
479 extern struct ceph_mds_session *
480 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
481 
482 extern const char *ceph_session_state_name(int s);
483 
484 extern struct ceph_mds_session *
485 ceph_get_mds_session(struct ceph_mds_session *s);
486 extern void ceph_put_mds_session(struct ceph_mds_session *s);
487 
488 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
489 			     struct ceph_msg *msg, int mds);
490 
491 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
492 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
493 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
494 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
495 
496 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
497 
498 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
499 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
500 					   struct inode *dir);
501 extern struct ceph_mds_request *
502 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
503 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
504 				    struct inode *dir,
505 				    struct ceph_mds_request *req);
506 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
507 				struct inode *dir,
508 				struct ceph_mds_request *req);
509 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
510 extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
511 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
512 {
513 	kref_get(&req->r_kref);
514 }
515 extern void ceph_mdsc_release_request(struct kref *kref);
516 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
517 {
518 	kref_put(&req->r_kref, ceph_mdsc_release_request);
519 }
520 
521 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
522 				    struct ceph_cap *cap);
523 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
524 				    struct ceph_mds_session *session);
525 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
526 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
527 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
528 				     int (*cb)(struct inode *,
529 					       struct ceph_cap *, void *),
530 				     void *arg);
531 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
532 
533 static inline void ceph_mdsc_free_path(char *path, int len)
534 {
535 	if (!IS_ERR_OR_NULL(path))
536 		__putname(path - (PATH_MAX - 1 - len));
537 }
538 
539 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
540 				  int stop_on_nosnap);
541 
542 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
543 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
544 				     struct dentry *dentry, char action,
545 				     u32 seq);
546 
547 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
548 				    struct ceph_msg *msg);
549 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
550 				   struct ceph_msg *msg);
551 
552 extern struct ceph_mds_session *
553 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
554 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
555 					  struct ceph_mds_session *session);
556 
557 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
558 			  struct ceph_mds_session *session,
559 			  int max_caps);
560 
561 static inline int ceph_wait_on_async_create(struct inode *inode)
562 {
563 	struct ceph_inode_info *ci = ceph_inode(inode);
564 
565 	return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
566 			   TASK_INTERRUPTIBLE);
567 }
568 
569 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
570 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
571 #endif
572