xref: /openbmc/linux/fs/ceph/mds_client.h (revision 2bc7d3e0)
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 struct ceph_mds_reply_xattr {
104 	char *xattr_value;
105 	size_t xattr_value_len;
106 };
107 
108 /*
109  * parsed info about an mds reply, including information about
110  * either: 1) the target inode and/or its parent directory and dentry,
111  * and directory contents (for readdir results), or
112  * 2) the file range lock info (for fcntl F_GETLK results).
113  */
114 struct ceph_mds_reply_info_parsed {
115 	struct ceph_mds_reply_head    *head;
116 
117 	/* trace */
118 	struct ceph_mds_reply_info_in diri, targeti;
119 	struct ceph_mds_reply_dirfrag *dirfrag;
120 	char                          *dname;
121 	u32                           dname_len;
122 	struct ceph_mds_reply_lease   *dlease;
123 	struct ceph_mds_reply_xattr   xattr_info;
124 
125 	/* extra */
126 	union {
127 		/* for fcntl F_GETLK results */
128 		struct ceph_filelock *filelock_reply;
129 
130 		/* for readdir results */
131 		struct {
132 			struct ceph_mds_reply_dirfrag *dir_dir;
133 			size_t			      dir_buf_size;
134 			int                           dir_nr;
135 			bool			      dir_end;
136 			bool			      dir_complete;
137 			bool			      hash_order;
138 			bool			      offset_hash;
139 			struct ceph_mds_reply_dir_entry  *dir_entries;
140 		};
141 
142 		/* for create results */
143 		struct {
144 			bool has_create_ino;
145 			u64 ino;
146 		};
147 	};
148 
149 	/* encoded blob describing snapshot contexts for certain
150 	   operations (e.g., open) */
151 	void *snapblob;
152 	int snapblob_len;
153 };
154 
155 
156 /*
157  * cap releases are batched and sent to the MDS en masse.
158  *
159  * Account for per-message overhead of mds_cap_release header
160  * and __le32 for osd epoch barrier trailing field.
161  */
162 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
163 				sizeof(struct ceph_mds_cap_release)) /	\
164 			        sizeof(struct ceph_mds_cap_item))
165 
166 
167 /*
168  * state associated with each MDS<->client session
169  */
170 enum {
171 	CEPH_MDS_SESSION_NEW = 1,
172 	CEPH_MDS_SESSION_OPENING = 2,
173 	CEPH_MDS_SESSION_OPEN = 3,
174 	CEPH_MDS_SESSION_HUNG = 4,
175 	CEPH_MDS_SESSION_RESTARTING = 5,
176 	CEPH_MDS_SESSION_RECONNECTING = 6,
177 	CEPH_MDS_SESSION_CLOSING = 7,
178 	CEPH_MDS_SESSION_CLOSED = 8,
179 	CEPH_MDS_SESSION_REJECTED = 9,
180 };
181 
182 struct ceph_mds_session {
183 	struct ceph_mds_client *s_mdsc;
184 	int               s_mds;
185 	int               s_state;
186 	unsigned long     s_ttl;      /* time until mds kills us */
187 	unsigned long	  s_features;
188 	u64               s_seq;      /* incoming msg seq # */
189 	struct mutex      s_mutex;    /* serialize session messages */
190 
191 	struct ceph_connection s_con;
192 
193 	struct ceph_auth_handshake s_auth;
194 
195 	atomic_t          s_cap_gen;  /* inc each time we get mds stale msg */
196 	unsigned long     s_cap_ttl;  /* when session caps expire. protected by s_mutex */
197 
198 	/* protected by s_cap_lock */
199 	spinlock_t        s_cap_lock;
200 	refcount_t        s_ref;
201 	struct list_head  s_caps;     /* all caps issued by this session */
202 	struct ceph_cap  *s_cap_iterator;
203 	int               s_nr_caps;
204 	int               s_num_cap_releases;
205 	int		  s_cap_reconnect;
206 	int		  s_readonly;
207 	struct list_head  s_cap_releases; /* waiting cap_release messages */
208 	struct work_struct s_cap_release_work;
209 
210 	/* See ceph_inode_info->i_dirty_item. */
211 	struct list_head  s_cap_dirty;	      /* inodes w/ dirty caps */
212 
213 	/* See ceph_inode_info->i_flushing_item. */
214 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
215 
216 	unsigned long     s_renew_requested; /* last time we sent a renew req */
217 	u64               s_renew_seq;
218 
219 	struct list_head  s_waiting;  /* waiting requests */
220 	struct list_head  s_unsafe;   /* unsafe requests */
221 	struct xarray	  s_delegated_inos;
222 };
223 
224 /*
225  * modes of choosing which MDS to send a request to
226  */
227 enum {
228 	USE_ANY_MDS,
229 	USE_RANDOM_MDS,
230 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
231 };
232 
233 struct ceph_mds_request;
234 struct ceph_mds_client;
235 
236 /*
237  * request completion callback
238  */
239 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
240 					     struct ceph_mds_request *req);
241 /*
242  * wait for request completion callback
243  */
244 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
245 						 struct ceph_mds_request *req);
246 
247 /*
248  * an in-flight mds request
249  */
250 struct ceph_mds_request {
251 	u64 r_tid;                   /* transaction id */
252 	struct rb_node r_node;
253 	struct ceph_mds_client *r_mdsc;
254 
255 	struct kref       r_kref;
256 	int r_op;                    /* mds op code */
257 
258 	/* operation on what? */
259 	struct inode *r_inode;              /* arg1 */
260 	struct dentry *r_dentry;            /* arg1 */
261 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
262 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
263 	char *r_path1, *r_path2;
264 	struct ceph_vino r_ino1, r_ino2;
265 
266 	struct inode *r_parent;		    /* parent dir inode */
267 	struct inode *r_target_inode;       /* resulting inode */
268 
269 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
270 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
271 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
272 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
273 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
274 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
275 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
276 #define CEPH_MDS_R_ASYNC		(8) /* async request */
277 	unsigned long	r_req_flags;
278 
279 	struct mutex r_fill_mutex;
280 
281 	union ceph_mds_request_args r_args;
282 	int r_fmode;        /* file mode, if expecting cap */
283 	int r_request_release_offset;
284 	const struct cred *r_cred;
285 	struct timespec64 r_stamp;
286 
287 	/* for choosing which mds to send this request to */
288 	int r_direct_mode;
289 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
290 
291 	/* data payload is used for xattr ops */
292 	struct ceph_pagelist *r_pagelist;
293 
294 	/* what caps shall we drop? */
295 	int r_inode_drop, r_inode_unless;
296 	int r_dentry_drop, r_dentry_unless;
297 	int r_old_dentry_drop, r_old_dentry_unless;
298 	struct inode *r_old_inode;
299 	int r_old_inode_drop, r_old_inode_unless;
300 
301 	struct ceph_msg  *r_request;  /* original request */
302 	struct ceph_msg  *r_reply;
303 	struct ceph_mds_reply_info_parsed r_reply_info;
304 	int r_err;
305 	u32               r_readdir_offset;
306 
307 	struct page *r_locked_page;
308 	int r_dir_caps;
309 	int r_num_caps;
310 
311 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
312 	unsigned long r_started;  /* start time to measure timeout against */
313 	unsigned long r_start_latency;  /* start time to measure latency */
314 	unsigned long r_end_latency;    /* finish time to measure latency */
315 	unsigned long r_request_started; /* start time for mds request only,
316 					    used to measure lease durations */
317 
318 	/* link unsafe requests to parent directory, for fsync */
319 	struct inode	*r_unsafe_dir;
320 	struct list_head r_unsafe_dir_item;
321 
322 	/* unsafe requests that modify the target inode */
323 	struct list_head r_unsafe_target_item;
324 
325 	struct ceph_mds_session *r_session;
326 
327 	int               r_attempts;   /* resend attempts */
328 	int               r_num_fwd;    /* number of forward attempts */
329 	int               r_resend_mds; /* mds to resend to next, if any*/
330 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
331 	u64		  r_deleg_ino;
332 
333 	struct list_head  r_wait;
334 	struct completion r_completion;
335 	struct completion r_safe_completion;
336 	ceph_mds_request_callback_t r_callback;
337 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
338 
339 	long long	  r_dir_release_cnt;
340 	long long	  r_dir_ordered_cnt;
341 	int		  r_readdir_cache_idx;
342 
343 	struct ceph_cap_reservation r_caps_reservation;
344 };
345 
346 struct ceph_pool_perm {
347 	struct rb_node node;
348 	int perm;
349 	s64 pool;
350 	size_t pool_ns_len;
351 	char pool_ns[];
352 };
353 
354 struct ceph_snapid_map {
355 	struct rb_node node;
356 	struct list_head lru;
357 	atomic_t ref;
358 	u64 snap;
359 	dev_t dev;
360 	unsigned long last_used;
361 };
362 
363 /*
364  * node for list of quotarealm inodes that are not visible from the filesystem
365  * mountpoint, but required to handle, e.g. quotas.
366  */
367 struct ceph_quotarealm_inode {
368 	struct rb_node node;
369 	u64 ino;
370 	unsigned long timeout; /* last time a lookup failed for this inode */
371 	struct mutex mutex;
372 	struct inode *inode;
373 };
374 
375 struct cap_wait {
376 	struct list_head	list;
377 	u64			ino;
378 	pid_t			tgid;
379 	int			need;
380 	int			want;
381 };
382 
383 /*
384  * mds client state
385  */
386 struct ceph_mds_client {
387 	struct ceph_fs_client  *fsc;
388 	struct mutex            mutex;         /* all nested structures */
389 
390 	struct ceph_mdsmap      *mdsmap;
391 	struct completion       safe_umount_waiters;
392 	wait_queue_head_t       session_close_wq;
393 	struct list_head        waiting_for_map;
394 	int 			mdsmap_err;
395 
396 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
397 	atomic_t		num_sessions;
398 	int                     max_sessions;  /* len of sessions array */
399 	int                     stopping;      /* true if shutting down */
400 
401 	atomic64_t		quotarealms_count; /* # realms with quota */
402 	/*
403 	 * We keep a list of inodes we don't see in the mountpoint but that we
404 	 * need to track quota realms.
405 	 */
406 	struct rb_root		quotarealms_inodes;
407 	struct mutex		quotarealms_inodes_mutex;
408 
409 	/*
410 	 * snap_rwsem will cover cap linkage into snaprealms, and
411 	 * realm snap contexts.  (later, we can do per-realm snap
412 	 * contexts locks..)  the empty list contains realms with no
413 	 * references (implying they contain no inodes with caps) that
414 	 * should be destroyed.
415 	 */
416 	u64			last_snap_seq;
417 	struct rw_semaphore     snap_rwsem;
418 	struct rb_root          snap_realms;
419 	struct list_head        snap_empty;
420 	int			num_snap_realms;
421 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
422 
423 	u64                    last_tid;      /* most recent mds request */
424 	u64                    oldest_tid;    /* oldest incomplete mds request,
425 						 excluding setfilelock requests */
426 	struct rb_root         request_tree;  /* pending mds requests */
427 	struct delayed_work    delayed_work;  /* delayed work */
428 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
429 	struct list_head cap_delay_list;   /* caps with delayed release */
430 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
431 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
432 	spinlock_t       snap_flush_lock;
433 
434 	u64               last_cap_flush_tid;
435 	struct list_head  cap_flush_list;
436 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
437 	int               num_cap_flushing; /* # caps we are flushing */
438 	spinlock_t        cap_dirty_lock;   /* protects above items */
439 	wait_queue_head_t cap_flushing_wq;
440 
441 	struct work_struct cap_reclaim_work;
442 	atomic_t	   cap_reclaim_pending;
443 
444 	/*
445 	 * Cap reservations
446 	 *
447 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
448 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
449 	 * memory needed to successfully process an MDS response.  (If an MDS
450 	 * sends us cap information and we fail to process it, we will have
451 	 * problems due to the client and MDS being out of sync.)
452 	 *
453 	 * Reservations are 'owned' by a ceph_cap_reservation context.
454 	 */
455 	spinlock_t	caps_list_lock;
456 	struct		list_head caps_list; /* unused (reserved or
457 						unreserved) */
458 	struct		list_head cap_wait_list;
459 	int		caps_total_count;    /* total caps allocated */
460 	int		caps_use_count;      /* in use */
461 	int		caps_use_max;	     /* max used caps */
462 	int		caps_reserve_count;  /* unused, reserved */
463 	int		caps_avail_count;    /* unused, unreserved */
464 	int		caps_min_count;      /* keep at least this many
465 						(unreserved) */
466 	spinlock_t	  dentry_list_lock;
467 	struct list_head  dentry_leases;     /* fifo list */
468 	struct list_head  dentry_dir_leases; /* lru list */
469 
470 	struct ceph_client_metric metric;
471 
472 	spinlock_t		snapid_map_lock;
473 	struct rb_root		snapid_map_tree;
474 	struct list_head	snapid_map_lru;
475 
476 	struct rw_semaphore     pool_perm_rwsem;
477 	struct rb_root		pool_perm_tree;
478 
479 	char nodename[__NEW_UTS_LEN + 1];
480 };
481 
482 extern const char *ceph_mds_op_name(int op);
483 
484 extern bool check_session_state(struct ceph_mds_session *s);
485 void inc_session_sequence(struct ceph_mds_session *s);
486 
487 extern struct ceph_mds_session *
488 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
489 
490 extern const char *ceph_session_state_name(int s);
491 
492 extern struct ceph_mds_session *
493 ceph_get_mds_session(struct ceph_mds_session *s);
494 extern void ceph_put_mds_session(struct ceph_mds_session *s);
495 
496 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
497 			     struct ceph_msg *msg, int mds);
498 
499 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
500 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
501 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
502 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
503 
504 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
505 
506 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
507 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
508 					   struct inode *dir);
509 extern struct ceph_mds_request *
510 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
511 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
512 				    struct inode *dir,
513 				    struct ceph_mds_request *req);
514 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
515 			struct ceph_mds_request *req,
516 			ceph_mds_request_wait_callback_t wait_func);
517 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
518 				struct inode *dir,
519 				struct ceph_mds_request *req);
520 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
521 extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
522 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
523 {
524 	kref_get(&req->r_kref);
525 }
526 extern void ceph_mdsc_release_request(struct kref *kref);
527 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
528 {
529 	kref_put(&req->r_kref, ceph_mdsc_release_request);
530 }
531 
532 extern void send_flush_mdlog(struct ceph_mds_session *s);
533 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
534 				       void (*cb)(struct ceph_mds_session *),
535 				       bool check_state);
536 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
537 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
538 				    struct ceph_cap *cap);
539 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
540 				    struct ceph_mds_session *session);
541 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
542 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
543 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
544 				     int (*cb)(struct inode *,
545 					       struct ceph_cap *, void *),
546 				     void *arg);
547 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
548 
549 static inline void ceph_mdsc_free_path(char *path, int len)
550 {
551 	if (!IS_ERR_OR_NULL(path))
552 		__putname(path - (PATH_MAX - 1 - len));
553 }
554 
555 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
556 				  int stop_on_nosnap);
557 
558 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
559 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
560 				     struct dentry *dentry, char action,
561 				     u32 seq);
562 
563 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
564 				    struct ceph_msg *msg);
565 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
566 				   struct ceph_msg *msg);
567 
568 extern struct ceph_mds_session *
569 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
570 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
571 					  struct ceph_mds_session *session);
572 
573 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
574 			  struct ceph_mds_session *session,
575 			  int max_caps);
576 
577 static inline int ceph_wait_on_async_create(struct inode *inode)
578 {
579 	struct ceph_inode_info *ci = ceph_inode(inode);
580 
581 	return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
582 			   TASK_KILLABLE);
583 }
584 
585 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
586 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
587 #endif
588