xref: /openbmc/linux/fs/ceph/mds_client.h (revision c6fddb28)
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 
14 #include <linux/ceph/types.h>
15 #include <linux/ceph/messenger.h>
16 #include <linux/ceph/mdsmap.h>
17 #include <linux/ceph/auth.h>
18 
19 /* The first 8 bits are reserved for old ceph releases */
20 enum ceph_feature_type {
21 	CEPHFS_FEATURE_MIMIC = 8,
22 	CEPHFS_FEATURE_REPLY_ENCODING,
23 	CEPHFS_FEATURE_RECLAIM_CLIENT,
24 	CEPHFS_FEATURE_LAZY_CAP_WANTED,
25 	CEPHFS_FEATURE_MULTI_RECONNECT,
26 	CEPHFS_FEATURE_DELEG_INO,
27 
28 	CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_DELEG_INO,
29 };
30 
31 /*
32  * This will always have the highest feature bit value
33  * as the last element of the array.
34  */
35 #define CEPHFS_FEATURES_CLIENT_SUPPORTED {	\
36 	0, 1, 2, 3, 4, 5, 6, 7,			\
37 	CEPHFS_FEATURE_MIMIC,			\
38 	CEPHFS_FEATURE_REPLY_ENCODING,		\
39 	CEPHFS_FEATURE_LAZY_CAP_WANTED,		\
40 	CEPHFS_FEATURE_MULTI_RECONNECT,		\
41 	CEPHFS_FEATURE_DELEG_INO,		\
42 						\
43 	CEPHFS_FEATURE_MAX,			\
44 }
45 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
46 
47 /*
48  * Some lock dependencies:
49  *
50  * session->s_mutex
51  *         mdsc->mutex
52  *
53  *         mdsc->snap_rwsem
54  *
55  *         ci->i_ceph_lock
56  *                 mdsc->snap_flush_lock
57  *                 mdsc->cap_delay_lock
58  *
59  */
60 
61 struct ceph_fs_client;
62 struct ceph_cap;
63 
64 /*
65  * parsed info about a single inode.  pointers are into the encoded
66  * on-wire structures within the mds reply message payload.
67  */
68 struct ceph_mds_reply_info_in {
69 	struct ceph_mds_reply_inode *in;
70 	struct ceph_dir_layout dir_layout;
71 	u32 symlink_len;
72 	char *symlink;
73 	u32 xattr_len;
74 	char *xattr_data;
75 	u64 inline_version;
76 	u32 inline_len;
77 	char *inline_data;
78 	u32 pool_ns_len;
79 	char *pool_ns_data;
80 	u64 max_bytes;
81 	u64 max_files;
82 	s32 dir_pin;
83 	struct ceph_timespec btime;
84 	struct ceph_timespec snap_btime;
85 	u64 change_attr;
86 };
87 
88 struct ceph_mds_reply_dir_entry {
89 	char                          *name;
90 	u32                           name_len;
91 	struct ceph_mds_reply_lease   *lease;
92 	struct ceph_mds_reply_info_in inode;
93 	loff_t			      offset;
94 };
95 
96 /*
97  * parsed info about an mds reply, including information about
98  * either: 1) the target inode and/or its parent directory and dentry,
99  * and directory contents (for readdir results), or
100  * 2) the file range lock info (for fcntl F_GETLK results).
101  */
102 struct ceph_mds_reply_info_parsed {
103 	struct ceph_mds_reply_head    *head;
104 
105 	/* trace */
106 	struct ceph_mds_reply_info_in diri, targeti;
107 	struct ceph_mds_reply_dirfrag *dirfrag;
108 	char                          *dname;
109 	u32                           dname_len;
110 	struct ceph_mds_reply_lease   *dlease;
111 
112 	/* extra */
113 	union {
114 		/* for fcntl F_GETLK results */
115 		struct ceph_filelock *filelock_reply;
116 
117 		/* for readdir results */
118 		struct {
119 			struct ceph_mds_reply_dirfrag *dir_dir;
120 			size_t			      dir_buf_size;
121 			int                           dir_nr;
122 			bool			      dir_end;
123 			bool			      dir_complete;
124 			bool			      hash_order;
125 			bool			      offset_hash;
126 			struct ceph_mds_reply_dir_entry  *dir_entries;
127 		};
128 
129 		/* for create results */
130 		struct {
131 			bool has_create_ino;
132 			u64 ino;
133 		};
134 	};
135 
136 	/* encoded blob describing snapshot contexts for certain
137 	   operations (e.g., open) */
138 	void *snapblob;
139 	int snapblob_len;
140 };
141 
142 
143 /*
144  * cap releases are batched and sent to the MDS en masse.
145  *
146  * Account for per-message overhead of mds_cap_release header
147  * and __le32 for osd epoch barrier trailing field.
148  */
149 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
150 				sizeof(struct ceph_mds_cap_release)) /	\
151 			        sizeof(struct ceph_mds_cap_item))
152 
153 
154 /*
155  * state associated with each MDS<->client session
156  */
157 enum {
158 	CEPH_MDS_SESSION_NEW = 1,
159 	CEPH_MDS_SESSION_OPENING = 2,
160 	CEPH_MDS_SESSION_OPEN = 3,
161 	CEPH_MDS_SESSION_HUNG = 4,
162 	CEPH_MDS_SESSION_RESTARTING = 5,
163 	CEPH_MDS_SESSION_RECONNECTING = 6,
164 	CEPH_MDS_SESSION_CLOSING = 7,
165 	CEPH_MDS_SESSION_CLOSED = 8,
166 	CEPH_MDS_SESSION_REJECTED = 9,
167 };
168 
169 struct ceph_mds_session {
170 	struct ceph_mds_client *s_mdsc;
171 	int               s_mds;
172 	int               s_state;
173 	unsigned long     s_ttl;      /* time until mds kills us */
174 	unsigned long	  s_features;
175 	u64               s_seq;      /* incoming msg seq # */
176 	struct mutex      s_mutex;    /* serialize session messages */
177 
178 	struct ceph_connection s_con;
179 
180 	struct ceph_auth_handshake s_auth;
181 
182 	/* protected by s_gen_ttl_lock */
183 	spinlock_t        s_gen_ttl_lock;
184 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
185 	unsigned long     s_cap_ttl;  /* when session caps expire */
186 
187 	/* protected by s_cap_lock */
188 	spinlock_t        s_cap_lock;
189 	refcount_t        s_ref;
190 	struct list_head  s_caps;     /* all caps issued by this session */
191 	struct ceph_cap  *s_cap_iterator;
192 	int               s_nr_caps;
193 	int               s_num_cap_releases;
194 	int		  s_cap_reconnect;
195 	int		  s_readonly;
196 	struct list_head  s_cap_releases; /* waiting cap_release messages */
197 	struct work_struct s_cap_release_work;
198 
199 	/* protected by mutex */
200 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
201 	unsigned long     s_renew_requested; /* last time we sent a renew req */
202 	u64               s_renew_seq;
203 
204 	struct list_head  s_waiting;  /* waiting requests */
205 	struct list_head  s_unsafe;   /* unsafe requests */
206 	struct xarray	  s_delegated_inos;
207 };
208 
209 /*
210  * modes of choosing which MDS to send a request to
211  */
212 enum {
213 	USE_ANY_MDS,
214 	USE_RANDOM_MDS,
215 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
216 };
217 
218 struct ceph_mds_request;
219 struct ceph_mds_client;
220 
221 /*
222  * request completion callback
223  */
224 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
225 					     struct ceph_mds_request *req);
226 /*
227  * wait for request completion callback
228  */
229 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
230 						 struct ceph_mds_request *req);
231 
232 /*
233  * an in-flight mds request
234  */
235 struct ceph_mds_request {
236 	u64 r_tid;                   /* transaction id */
237 	struct rb_node r_node;
238 	struct ceph_mds_client *r_mdsc;
239 
240 	struct kref       r_kref;
241 	int r_op;                    /* mds op code */
242 
243 	/* operation on what? */
244 	struct inode *r_inode;              /* arg1 */
245 	struct dentry *r_dentry;            /* arg1 */
246 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
247 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
248 	char *r_path1, *r_path2;
249 	struct ceph_vino r_ino1, r_ino2;
250 
251 	struct inode *r_parent;		    /* parent dir inode */
252 	struct inode *r_target_inode;       /* resulting inode */
253 
254 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
255 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
256 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
257 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
258 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
259 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
260 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
261 #define CEPH_MDS_R_ASYNC		(8) /* async request */
262 	unsigned long	r_req_flags;
263 
264 	struct mutex r_fill_mutex;
265 
266 	union ceph_mds_request_args r_args;
267 	int r_fmode;        /* file mode, if expecting cap */
268 	kuid_t r_uid;
269 	kgid_t r_gid;
270 	int r_request_release_offset;
271 	struct timespec64 r_stamp;
272 
273 	/* for choosing which mds to send this request to */
274 	int r_direct_mode;
275 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
276 
277 	/* data payload is used for xattr ops */
278 	struct ceph_pagelist *r_pagelist;
279 
280 	/* what caps shall we drop? */
281 	int r_inode_drop, r_inode_unless;
282 	int r_dentry_drop, r_dentry_unless;
283 	int r_old_dentry_drop, r_old_dentry_unless;
284 	struct inode *r_old_inode;
285 	int r_old_inode_drop, r_old_inode_unless;
286 
287 	struct ceph_msg  *r_request;  /* original request */
288 	struct ceph_msg  *r_reply;
289 	struct ceph_mds_reply_info_parsed r_reply_info;
290 	int r_err;
291 
292 
293 	struct page *r_locked_page;
294 	int r_dir_caps;
295 	int r_num_caps;
296 	u32               r_readdir_offset;
297 
298 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
299 	unsigned long r_started;  /* start time to measure timeout against */
300 	unsigned long r_request_started; /* start time for mds request only,
301 					    used to measure lease durations */
302 
303 	/* link unsafe requests to parent directory, for fsync */
304 	struct inode	*r_unsafe_dir;
305 	struct list_head r_unsafe_dir_item;
306 
307 	/* unsafe requests that modify the target inode */
308 	struct list_head r_unsafe_target_item;
309 
310 	struct ceph_mds_session *r_session;
311 
312 	int               r_attempts;   /* resend attempts */
313 	int               r_num_fwd;    /* number of forward attempts */
314 	int               r_resend_mds; /* mds to resend to next, if any*/
315 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
316 	u64		  r_deleg_ino;
317 
318 	struct list_head  r_wait;
319 	struct completion r_completion;
320 	struct completion r_safe_completion;
321 	ceph_mds_request_callback_t r_callback;
322 	ceph_mds_request_wait_callback_t r_wait_for_completion;
323 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
324 
325 	long long	  r_dir_release_cnt;
326 	long long	  r_dir_ordered_cnt;
327 	int		  r_readdir_cache_idx;
328 
329 	struct ceph_cap_reservation r_caps_reservation;
330 };
331 
332 struct ceph_pool_perm {
333 	struct rb_node node;
334 	int perm;
335 	s64 pool;
336 	size_t pool_ns_len;
337 	char pool_ns[];
338 };
339 
340 struct ceph_snapid_map {
341 	struct rb_node node;
342 	struct list_head lru;
343 	atomic_t ref;
344 	u64 snap;
345 	dev_t dev;
346 	unsigned long last_used;
347 };
348 
349 /*
350  * node for list of quotarealm inodes that are not visible from the filesystem
351  * mountpoint, but required to handle, e.g. quotas.
352  */
353 struct ceph_quotarealm_inode {
354 	struct rb_node node;
355 	u64 ino;
356 	unsigned long timeout; /* last time a lookup failed for this inode */
357 	struct mutex mutex;
358 	struct inode *inode;
359 };
360 
361 struct cap_wait {
362 	struct list_head	list;
363 	unsigned long		ino;
364 	pid_t			tgid;
365 	int			need;
366 	int			want;
367 };
368 
369 /*
370  * mds client state
371  */
372 struct ceph_mds_client {
373 	struct ceph_fs_client  *fsc;
374 	struct mutex            mutex;         /* all nested structures */
375 
376 	struct ceph_mdsmap      *mdsmap;
377 	struct completion       safe_umount_waiters;
378 	wait_queue_head_t       session_close_wq;
379 	struct list_head        waiting_for_map;
380 	int 			mdsmap_err;
381 
382 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
383 	atomic_t		num_sessions;
384 	int                     max_sessions;  /* len of s_mds_sessions */
385 	int                     stopping;      /* true if shutting down */
386 
387 	atomic64_t		quotarealms_count; /* # realms with quota */
388 	/*
389 	 * We keep a list of inodes we don't see in the mountpoint but that we
390 	 * need to track quota realms.
391 	 */
392 	struct rb_root		quotarealms_inodes;
393 	struct mutex		quotarealms_inodes_mutex;
394 
395 	/*
396 	 * snap_rwsem will cover cap linkage into snaprealms, and
397 	 * realm snap contexts.  (later, we can do per-realm snap
398 	 * contexts locks..)  the empty list contains realms with no
399 	 * references (implying they contain no inodes with caps) that
400 	 * should be destroyed.
401 	 */
402 	u64			last_snap_seq;
403 	struct rw_semaphore     snap_rwsem;
404 	struct rb_root          snap_realms;
405 	struct list_head        snap_empty;
406 	int			num_snap_realms;
407 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
408 
409 	u64                    last_tid;      /* most recent mds request */
410 	u64                    oldest_tid;    /* oldest incomplete mds request,
411 						 excluding setfilelock requests */
412 	struct rb_root         request_tree;  /* pending mds requests */
413 	struct delayed_work    delayed_work;  /* delayed work */
414 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
415 	struct list_head cap_delay_list;   /* caps with delayed release */
416 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
417 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
418 	spinlock_t       snap_flush_lock;
419 
420 	u64               last_cap_flush_tid;
421 	struct list_head  cap_flush_list;
422 	struct list_head  cap_dirty;        /* inodes with dirty caps */
423 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
424 	int               num_cap_flushing; /* # caps we are flushing */
425 	spinlock_t        cap_dirty_lock;   /* protects above items */
426 	wait_queue_head_t cap_flushing_wq;
427 
428 	struct work_struct cap_reclaim_work;
429 	atomic_t	   cap_reclaim_pending;
430 
431 	/*
432 	 * Cap reservations
433 	 *
434 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
435 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
436 	 * memory needed to successfully process an MDS response.  (If an MDS
437 	 * sends us cap information and we fail to process it, we will have
438 	 * problems due to the client and MDS being out of sync.)
439 	 *
440 	 * Reservations are 'owned' by a ceph_cap_reservation context.
441 	 */
442 	spinlock_t	caps_list_lock;
443 	struct		list_head caps_list; /* unused (reserved or
444 						unreserved) */
445 	struct		list_head cap_wait_list;
446 	int		caps_total_count;    /* total caps allocated */
447 	int		caps_use_count;      /* in use */
448 	int		caps_use_max;	     /* max used caps */
449 	int		caps_reserve_count;  /* unused, reserved */
450 	int		caps_avail_count;    /* unused, unreserved */
451 	int		caps_min_count;      /* keep at least this many
452 						(unreserved) */
453 	spinlock_t	  dentry_list_lock;
454 	struct list_head  dentry_leases;     /* fifo list */
455 	struct list_head  dentry_dir_leases; /* lru list */
456 
457 	spinlock_t		snapid_map_lock;
458 	struct rb_root		snapid_map_tree;
459 	struct list_head	snapid_map_lru;
460 
461 	struct rw_semaphore     pool_perm_rwsem;
462 	struct rb_root		pool_perm_tree;
463 
464 	char nodename[__NEW_UTS_LEN + 1];
465 };
466 
467 extern const char *ceph_mds_op_name(int op);
468 
469 extern struct ceph_mds_session *
470 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
471 
472 extern const char *ceph_session_state_name(int s);
473 
474 extern struct ceph_mds_session *
475 ceph_get_mds_session(struct ceph_mds_session *s);
476 extern void ceph_put_mds_session(struct ceph_mds_session *s);
477 
478 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
479 			     struct ceph_msg *msg, int mds);
480 
481 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
482 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
483 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
484 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
485 
486 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
487 
488 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
489 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
490 					   struct inode *dir);
491 extern struct ceph_mds_request *
492 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
493 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
494 				    struct inode *dir,
495 				    struct ceph_mds_request *req);
496 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
497 				struct inode *dir,
498 				struct ceph_mds_request *req);
499 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
500 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
501 {
502 	kref_get(&req->r_kref);
503 }
504 extern void ceph_mdsc_release_request(struct kref *kref);
505 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
506 {
507 	kref_put(&req->r_kref, ceph_mdsc_release_request);
508 }
509 
510 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
511 				    struct ceph_cap *cap);
512 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
513 				    struct ceph_mds_session *session);
514 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
515 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
516 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
517 				     int (*cb)(struct inode *,
518 					       struct ceph_cap *, void *),
519 				     void *arg);
520 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
521 
522 static inline void ceph_mdsc_free_path(char *path, int len)
523 {
524 	if (path)
525 		__putname(path - (PATH_MAX - 1 - len));
526 }
527 
528 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
529 				  int stop_on_nosnap);
530 
531 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
532 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
533 				     struct dentry *dentry, char action,
534 				     u32 seq);
535 
536 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
537 				    struct ceph_msg *msg);
538 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
539 				   struct ceph_msg *msg);
540 
541 extern struct ceph_mds_session *
542 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
543 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
544 					  struct ceph_mds_session *session);
545 
546 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
547 			  struct ceph_mds_session *session,
548 			  int max_caps);
549 
550 static inline int ceph_wait_on_async_create(struct inode *inode)
551 {
552 	struct ceph_inode_info *ci = ceph_inode(inode);
553 
554 	return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
555 			   TASK_INTERRUPTIBLE);
556 }
557 
558 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
559 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
560 #endif
561