xref: /openbmc/linux/fs/ceph/mds_client.h (revision 63705da3)
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 #include "super.h"
22 
23 /* The first 8 bits are reserved for old ceph releases */
24 enum ceph_feature_type {
25 	CEPHFS_FEATURE_MIMIC = 8,
26 	CEPHFS_FEATURE_REPLY_ENCODING,
27 	CEPHFS_FEATURE_RECLAIM_CLIENT,
28 	CEPHFS_FEATURE_LAZY_CAP_WANTED,
29 	CEPHFS_FEATURE_MULTI_RECONNECT,
30 	CEPHFS_FEATURE_DELEG_INO,
31 	CEPHFS_FEATURE_METRIC_COLLECT,
32 
33 	CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT,
34 };
35 
36 /*
37  * This will always have the highest feature bit value
38  * as the last element of the array.
39  */
40 #define CEPHFS_FEATURES_CLIENT_SUPPORTED {	\
41 	0, 1, 2, 3, 4, 5, 6, 7,			\
42 	CEPHFS_FEATURE_MIMIC,			\
43 	CEPHFS_FEATURE_REPLY_ENCODING,		\
44 	CEPHFS_FEATURE_LAZY_CAP_WANTED,		\
45 	CEPHFS_FEATURE_MULTI_RECONNECT,		\
46 	CEPHFS_FEATURE_DELEG_INO,		\
47 	CEPHFS_FEATURE_METRIC_COLLECT,		\
48 						\
49 	CEPHFS_FEATURE_MAX,			\
50 }
51 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
52 
53 /*
54  * Some lock dependencies:
55  *
56  * session->s_mutex
57  *         mdsc->mutex
58  *
59  *         mdsc->snap_rwsem
60  *
61  *         ci->i_ceph_lock
62  *                 mdsc->snap_flush_lock
63  *                 mdsc->cap_delay_lock
64  *
65  */
66 
67 struct ceph_fs_client;
68 struct ceph_cap;
69 
70 /*
71  * parsed info about a single inode.  pointers are into the encoded
72  * on-wire structures within the mds reply message payload.
73  */
74 struct ceph_mds_reply_info_in {
75 	struct ceph_mds_reply_inode *in;
76 	struct ceph_dir_layout dir_layout;
77 	u32 symlink_len;
78 	char *symlink;
79 	u32 xattr_len;
80 	char *xattr_data;
81 	u64 inline_version;
82 	u32 inline_len;
83 	char *inline_data;
84 	u32 pool_ns_len;
85 	char *pool_ns_data;
86 	u64 max_bytes;
87 	u64 max_files;
88 	s32 dir_pin;
89 	struct ceph_timespec btime;
90 	struct ceph_timespec snap_btime;
91 	u64 rsnaps;
92 	u64 change_attr;
93 };
94 
95 struct ceph_mds_reply_dir_entry {
96 	char                          *name;
97 	u32                           name_len;
98 	struct ceph_mds_reply_lease   *lease;
99 	struct ceph_mds_reply_info_in inode;
100 	loff_t			      offset;
101 };
102 
103 /*
104  * parsed info about an mds reply, including information about
105  * either: 1) the target inode and/or its parent directory and dentry,
106  * and directory contents (for readdir results), or
107  * 2) the file range lock info (for fcntl F_GETLK results).
108  */
109 struct ceph_mds_reply_info_parsed {
110 	struct ceph_mds_reply_head    *head;
111 
112 	/* trace */
113 	struct ceph_mds_reply_info_in diri, targeti;
114 	struct ceph_mds_reply_dirfrag *dirfrag;
115 	char                          *dname;
116 	u32                           dname_len;
117 	struct ceph_mds_reply_lease   *dlease;
118 
119 	/* extra */
120 	union {
121 		/* for fcntl F_GETLK results */
122 		struct ceph_filelock *filelock_reply;
123 
124 		/* for readdir results */
125 		struct {
126 			struct ceph_mds_reply_dirfrag *dir_dir;
127 			size_t			      dir_buf_size;
128 			int                           dir_nr;
129 			bool			      dir_end;
130 			bool			      dir_complete;
131 			bool			      hash_order;
132 			bool			      offset_hash;
133 			struct ceph_mds_reply_dir_entry  *dir_entries;
134 		};
135 
136 		/* for create results */
137 		struct {
138 			bool has_create_ino;
139 			u64 ino;
140 		};
141 	};
142 
143 	/* encoded blob describing snapshot contexts for certain
144 	   operations (e.g., open) */
145 	void *snapblob;
146 	int snapblob_len;
147 };
148 
149 
150 /*
151  * cap releases are batched and sent to the MDS en masse.
152  *
153  * Account for per-message overhead of mds_cap_release header
154  * and __le32 for osd epoch barrier trailing field.
155  */
156 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
157 				sizeof(struct ceph_mds_cap_release)) /	\
158 			        sizeof(struct ceph_mds_cap_item))
159 
160 
161 /*
162  * state associated with each MDS<->client session
163  */
164 enum {
165 	CEPH_MDS_SESSION_NEW = 1,
166 	CEPH_MDS_SESSION_OPENING = 2,
167 	CEPH_MDS_SESSION_OPEN = 3,
168 	CEPH_MDS_SESSION_HUNG = 4,
169 	CEPH_MDS_SESSION_RESTARTING = 5,
170 	CEPH_MDS_SESSION_RECONNECTING = 6,
171 	CEPH_MDS_SESSION_CLOSING = 7,
172 	CEPH_MDS_SESSION_CLOSED = 8,
173 	CEPH_MDS_SESSION_REJECTED = 9,
174 };
175 
176 struct ceph_mds_session {
177 	struct ceph_mds_client *s_mdsc;
178 	int               s_mds;
179 	int               s_state;
180 	unsigned long     s_ttl;      /* time until mds kills us */
181 	unsigned long	  s_features;
182 	u64               s_seq;      /* incoming msg seq # */
183 	struct mutex      s_mutex;    /* serialize session messages */
184 
185 	struct ceph_connection s_con;
186 
187 	struct ceph_auth_handshake s_auth;
188 
189 	atomic_t          s_cap_gen;  /* inc each time we get mds stale msg */
190 	unsigned long     s_cap_ttl;  /* when session caps expire. protected by s_mutex */
191 
192 	/* protected by s_cap_lock */
193 	spinlock_t        s_cap_lock;
194 	refcount_t        s_ref;
195 	struct list_head  s_caps;     /* all caps issued by this session */
196 	struct ceph_cap  *s_cap_iterator;
197 	int               s_nr_caps;
198 	int               s_num_cap_releases;
199 	int		  s_cap_reconnect;
200 	int		  s_readonly;
201 	struct list_head  s_cap_releases; /* waiting cap_release messages */
202 	struct work_struct s_cap_release_work;
203 
204 	/* See ceph_inode_info->i_dirty_item. */
205 	struct list_head  s_cap_dirty;	      /* inodes w/ dirty caps */
206 
207 	/* See ceph_inode_info->i_flushing_item. */
208 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
209 
210 	unsigned long     s_renew_requested; /* last time we sent a renew req */
211 	u64               s_renew_seq;
212 
213 	struct list_head  s_waiting;  /* waiting requests */
214 	struct list_head  s_unsafe;   /* unsafe requests */
215 	struct xarray	  s_delegated_inos;
216 };
217 
218 /*
219  * modes of choosing which MDS to send a request to
220  */
221 enum {
222 	USE_ANY_MDS,
223 	USE_RANDOM_MDS,
224 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
225 };
226 
227 struct ceph_mds_request;
228 struct ceph_mds_client;
229 
230 /*
231  * request completion callback
232  */
233 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
234 					     struct ceph_mds_request *req);
235 /*
236  * wait for request completion callback
237  */
238 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
239 						 struct ceph_mds_request *req);
240 
241 /*
242  * an in-flight mds request
243  */
244 struct ceph_mds_request {
245 	u64 r_tid;                   /* transaction id */
246 	struct rb_node r_node;
247 	struct ceph_mds_client *r_mdsc;
248 
249 	struct kref       r_kref;
250 	int r_op;                    /* mds op code */
251 
252 	/* operation on what? */
253 	struct inode *r_inode;              /* arg1 */
254 	struct dentry *r_dentry;            /* arg1 */
255 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
256 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
257 	char *r_path1, *r_path2;
258 	struct ceph_vino r_ino1, r_ino2;
259 
260 	struct inode *r_parent;		    /* parent dir inode */
261 	struct inode *r_target_inode;       /* resulting inode */
262 
263 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
264 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
265 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
266 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
267 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
268 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
269 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
270 #define CEPH_MDS_R_ASYNC		(8) /* async request */
271 	unsigned long	r_req_flags;
272 
273 	struct mutex r_fill_mutex;
274 
275 	union ceph_mds_request_args r_args;
276 	int r_fmode;        /* file mode, if expecting cap */
277 	const struct cred *r_cred;
278 	int r_request_release_offset;
279 	struct timespec64 r_stamp;
280 
281 	/* for choosing which mds to send this request to */
282 	int r_direct_mode;
283 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
284 
285 	/* data payload is used for xattr ops */
286 	struct ceph_pagelist *r_pagelist;
287 
288 	/* what caps shall we drop? */
289 	int r_inode_drop, r_inode_unless;
290 	int r_dentry_drop, r_dentry_unless;
291 	int r_old_dentry_drop, r_old_dentry_unless;
292 	struct inode *r_old_inode;
293 	int r_old_inode_drop, r_old_inode_unless;
294 
295 	struct ceph_msg  *r_request;  /* original request */
296 	struct ceph_msg  *r_reply;
297 	struct ceph_mds_reply_info_parsed r_reply_info;
298 	int r_err;
299 
300 
301 	struct page *r_locked_page;
302 	int r_dir_caps;
303 	int r_num_caps;
304 	u32               r_readdir_offset;
305 
306 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
307 	unsigned long r_started;  /* start time to measure timeout against */
308 	unsigned long r_start_latency;  /* start time to measure latency */
309 	unsigned long r_end_latency;    /* finish time to measure latency */
310 	unsigned long r_request_started; /* start time for mds request only,
311 					    used to measure lease durations */
312 
313 	/* link unsafe requests to parent directory, for fsync */
314 	struct inode	*r_unsafe_dir;
315 	struct list_head r_unsafe_dir_item;
316 
317 	/* unsafe requests that modify the target inode */
318 	struct list_head r_unsafe_target_item;
319 
320 	struct ceph_mds_session *r_session;
321 
322 	int               r_attempts;   /* resend attempts */
323 	int               r_num_fwd;    /* number of forward attempts */
324 	int               r_resend_mds; /* mds to resend to next, if any*/
325 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
326 	u64		  r_deleg_ino;
327 
328 	struct list_head  r_wait;
329 	struct completion r_completion;
330 	struct completion r_safe_completion;
331 	ceph_mds_request_callback_t r_callback;
332 	ceph_mds_request_wait_callback_t r_wait_for_completion;
333 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
334 
335 	long long	  r_dir_release_cnt;
336 	long long	  r_dir_ordered_cnt;
337 	int		  r_readdir_cache_idx;
338 
339 	struct ceph_cap_reservation r_caps_reservation;
340 };
341 
342 struct ceph_pool_perm {
343 	struct rb_node node;
344 	int perm;
345 	s64 pool;
346 	size_t pool_ns_len;
347 	char pool_ns[];
348 };
349 
350 struct ceph_snapid_map {
351 	struct rb_node node;
352 	struct list_head lru;
353 	atomic_t ref;
354 	u64 snap;
355 	dev_t dev;
356 	unsigned long last_used;
357 };
358 
359 /*
360  * node for list of quotarealm inodes that are not visible from the filesystem
361  * mountpoint, but required to handle, e.g. quotas.
362  */
363 struct ceph_quotarealm_inode {
364 	struct rb_node node;
365 	u64 ino;
366 	unsigned long timeout; /* last time a lookup failed for this inode */
367 	struct mutex mutex;
368 	struct inode *inode;
369 };
370 
371 struct cap_wait {
372 	struct list_head	list;
373 	u64			ino;
374 	pid_t			tgid;
375 	int			need;
376 	int			want;
377 };
378 
379 /*
380  * mds client state
381  */
382 struct ceph_mds_client {
383 	struct ceph_fs_client  *fsc;
384 	struct mutex            mutex;         /* all nested structures */
385 
386 	struct ceph_mdsmap      *mdsmap;
387 	struct completion       safe_umount_waiters;
388 	wait_queue_head_t       session_close_wq;
389 	struct list_head        waiting_for_map;
390 	int 			mdsmap_err;
391 
392 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
393 	atomic_t		num_sessions;
394 	int                     max_sessions;  /* len of sessions array */
395 	int                     stopping;      /* true if shutting down */
396 
397 	atomic64_t		quotarealms_count; /* # realms with quota */
398 	/*
399 	 * We keep a list of inodes we don't see in the mountpoint but that we
400 	 * need to track quota realms.
401 	 */
402 	struct rb_root		quotarealms_inodes;
403 	struct mutex		quotarealms_inodes_mutex;
404 
405 	/*
406 	 * snap_rwsem will cover cap linkage into snaprealms, and
407 	 * realm snap contexts.  (later, we can do per-realm snap
408 	 * contexts locks..)  the empty list contains realms with no
409 	 * references (implying they contain no inodes with caps) that
410 	 * should be destroyed.
411 	 */
412 	u64			last_snap_seq;
413 	struct rw_semaphore     snap_rwsem;
414 	struct rb_root          snap_realms;
415 	struct list_head        snap_empty;
416 	int			num_snap_realms;
417 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
418 
419 	u64                    last_tid;      /* most recent mds request */
420 	u64                    oldest_tid;    /* oldest incomplete mds request,
421 						 excluding setfilelock requests */
422 	struct rb_root         request_tree;  /* pending mds requests */
423 	struct delayed_work    delayed_work;  /* delayed work */
424 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
425 	struct list_head cap_delay_list;   /* caps with delayed release */
426 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
427 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
428 	spinlock_t       snap_flush_lock;
429 
430 	u64               last_cap_flush_tid;
431 	struct list_head  cap_flush_list;
432 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
433 	int               num_cap_flushing; /* # caps we are flushing */
434 	spinlock_t        cap_dirty_lock;   /* protects above items */
435 	wait_queue_head_t cap_flushing_wq;
436 
437 	struct work_struct cap_reclaim_work;
438 	atomic_t	   cap_reclaim_pending;
439 
440 	/*
441 	 * Cap reservations
442 	 *
443 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
444 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
445 	 * memory needed to successfully process an MDS response.  (If an MDS
446 	 * sends us cap information and we fail to process it, we will have
447 	 * problems due to the client and MDS being out of sync.)
448 	 *
449 	 * Reservations are 'owned' by a ceph_cap_reservation context.
450 	 */
451 	spinlock_t	caps_list_lock;
452 	struct		list_head caps_list; /* unused (reserved or
453 						unreserved) */
454 	struct		list_head cap_wait_list;
455 	int		caps_total_count;    /* total caps allocated */
456 	int		caps_use_count;      /* in use */
457 	int		caps_use_max;	     /* max used caps */
458 	int		caps_reserve_count;  /* unused, reserved */
459 	int		caps_avail_count;    /* unused, unreserved */
460 	int		caps_min_count;      /* keep at least this many
461 						(unreserved) */
462 	spinlock_t	  dentry_list_lock;
463 	struct list_head  dentry_leases;     /* fifo list */
464 	struct list_head  dentry_dir_leases; /* lru list */
465 
466 	struct ceph_client_metric metric;
467 
468 	spinlock_t		snapid_map_lock;
469 	struct rb_root		snapid_map_tree;
470 	struct list_head	snapid_map_lru;
471 
472 	struct rw_semaphore     pool_perm_rwsem;
473 	struct rb_root		pool_perm_tree;
474 
475 	char nodename[__NEW_UTS_LEN + 1];
476 };
477 
478 extern const char *ceph_mds_op_name(int op);
479 
480 extern bool check_session_state(struct ceph_mds_session *s);
481 void inc_session_sequence(struct ceph_mds_session *s);
482 
483 extern struct ceph_mds_session *
484 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
485 
486 extern const char *ceph_session_state_name(int s);
487 
488 extern struct ceph_mds_session *
489 ceph_get_mds_session(struct ceph_mds_session *s);
490 extern void ceph_put_mds_session(struct ceph_mds_session *s);
491 
492 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
493 			     struct ceph_msg *msg, int mds);
494 
495 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
496 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
497 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
498 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
499 
500 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
501 
502 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
503 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
504 					   struct inode *dir);
505 extern struct ceph_mds_request *
506 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
507 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
508 				    struct inode *dir,
509 				    struct ceph_mds_request *req);
510 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
511 				struct inode *dir,
512 				struct ceph_mds_request *req);
513 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
514 extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
515 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
516 {
517 	kref_get(&req->r_kref);
518 }
519 extern void ceph_mdsc_release_request(struct kref *kref);
520 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
521 {
522 	kref_put(&req->r_kref, ceph_mdsc_release_request);
523 }
524 
525 extern void send_flush_mdlog(struct ceph_mds_session *s);
526 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
527 				       void (*cb)(struct ceph_mds_session *),
528 				       bool check_state);
529 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
530 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
531 				    struct ceph_cap *cap);
532 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
533 				    struct ceph_mds_session *session);
534 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
535 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
536 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
537 				     int (*cb)(struct inode *,
538 					       struct ceph_cap *, void *),
539 				     void *arg);
540 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
541 
542 static inline void ceph_mdsc_free_path(char *path, int len)
543 {
544 	if (!IS_ERR_OR_NULL(path))
545 		__putname(path - (PATH_MAX - 1 - len));
546 }
547 
548 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
549 				  int stop_on_nosnap);
550 
551 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
552 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
553 				     struct dentry *dentry, char action,
554 				     u32 seq);
555 
556 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
557 				    struct ceph_msg *msg);
558 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
559 				   struct ceph_msg *msg);
560 
561 extern struct ceph_mds_session *
562 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
563 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
564 					  struct ceph_mds_session *session);
565 
566 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
567 			  struct ceph_mds_session *session,
568 			  int max_caps);
569 
570 static inline int ceph_wait_on_async_create(struct inode *inode)
571 {
572 	struct ceph_inode_info *ci = ceph_inode(inode);
573 
574 	return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
575 			   TASK_INTERRUPTIBLE);
576 }
577 
578 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
579 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
580 #endif
581