xref: /openbmc/linux/fs/ceph/mds_client.h (revision 11788d9b)
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 change_attr;
92 };
93 
94 struct ceph_mds_reply_dir_entry {
95 	char                          *name;
96 	u32                           name_len;
97 	struct ceph_mds_reply_lease   *lease;
98 	struct ceph_mds_reply_info_in inode;
99 	loff_t			      offset;
100 };
101 
102 /*
103  * parsed info about an mds reply, including information about
104  * either: 1) the target inode and/or its parent directory and dentry,
105  * and directory contents (for readdir results), or
106  * 2) the file range lock info (for fcntl F_GETLK results).
107  */
108 struct ceph_mds_reply_info_parsed {
109 	struct ceph_mds_reply_head    *head;
110 
111 	/* trace */
112 	struct ceph_mds_reply_info_in diri, targeti;
113 	struct ceph_mds_reply_dirfrag *dirfrag;
114 	char                          *dname;
115 	u32                           dname_len;
116 	struct ceph_mds_reply_lease   *dlease;
117 
118 	/* extra */
119 	union {
120 		/* for fcntl F_GETLK results */
121 		struct ceph_filelock *filelock_reply;
122 
123 		/* for readdir results */
124 		struct {
125 			struct ceph_mds_reply_dirfrag *dir_dir;
126 			size_t			      dir_buf_size;
127 			int                           dir_nr;
128 			bool			      dir_end;
129 			bool			      dir_complete;
130 			bool			      hash_order;
131 			bool			      offset_hash;
132 			struct ceph_mds_reply_dir_entry  *dir_entries;
133 		};
134 
135 		/* for create results */
136 		struct {
137 			bool has_create_ino;
138 			u64 ino;
139 		};
140 	};
141 
142 	/* encoded blob describing snapshot contexts for certain
143 	   operations (e.g., open) */
144 	void *snapblob;
145 	int snapblob_len;
146 };
147 
148 
149 /*
150  * cap releases are batched and sent to the MDS en masse.
151  *
152  * Account for per-message overhead of mds_cap_release header
153  * and __le32 for osd epoch barrier trailing field.
154  */
155 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
156 				sizeof(struct ceph_mds_cap_release)) /	\
157 			        sizeof(struct ceph_mds_cap_item))
158 
159 
160 /*
161  * state associated with each MDS<->client session
162  */
163 enum {
164 	CEPH_MDS_SESSION_NEW = 1,
165 	CEPH_MDS_SESSION_OPENING = 2,
166 	CEPH_MDS_SESSION_OPEN = 3,
167 	CEPH_MDS_SESSION_HUNG = 4,
168 	CEPH_MDS_SESSION_RESTARTING = 5,
169 	CEPH_MDS_SESSION_RECONNECTING = 6,
170 	CEPH_MDS_SESSION_CLOSING = 7,
171 	CEPH_MDS_SESSION_CLOSED = 8,
172 	CEPH_MDS_SESSION_REJECTED = 9,
173 };
174 
175 struct ceph_mds_session {
176 	struct ceph_mds_client *s_mdsc;
177 	int               s_mds;
178 	int               s_state;
179 	unsigned long     s_ttl;      /* time until mds kills us */
180 	unsigned long	  s_features;
181 	u64               s_seq;      /* incoming msg seq # */
182 	struct mutex      s_mutex;    /* serialize session messages */
183 
184 	struct ceph_connection s_con;
185 
186 	struct ceph_auth_handshake s_auth;
187 
188 	/* protected by s_gen_ttl_lock */
189 	spinlock_t        s_gen_ttl_lock;
190 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
191 	unsigned long     s_cap_ttl;  /* when session caps expire */
192 
193 	/* protected by s_cap_lock */
194 	spinlock_t        s_cap_lock;
195 	refcount_t        s_ref;
196 	struct list_head  s_caps;     /* all caps issued by this session */
197 	struct ceph_cap  *s_cap_iterator;
198 	int               s_nr_caps;
199 	int               s_num_cap_releases;
200 	int		  s_cap_reconnect;
201 	int		  s_readonly;
202 	struct list_head  s_cap_releases; /* waiting cap_release messages */
203 	struct work_struct s_cap_release_work;
204 
205 	/* See ceph_inode_info->i_dirty_item. */
206 	struct list_head  s_cap_dirty;	      /* inodes w/ dirty caps */
207 
208 	/* See ceph_inode_info->i_flushing_item. */
209 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
210 
211 	unsigned long     s_renew_requested; /* last time we sent a renew req */
212 	u64               s_renew_seq;
213 
214 	struct list_head  s_waiting;  /* waiting requests */
215 	struct list_head  s_unsafe;   /* unsafe requests */
216 	struct xarray	  s_delegated_inos;
217 };
218 
219 /*
220  * modes of choosing which MDS to send a request to
221  */
222 enum {
223 	USE_ANY_MDS,
224 	USE_RANDOM_MDS,
225 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
226 };
227 
228 struct ceph_mds_request;
229 struct ceph_mds_client;
230 
231 /*
232  * request completion callback
233  */
234 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
235 					     struct ceph_mds_request *req);
236 /*
237  * wait for request completion callback
238  */
239 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
240 						 struct ceph_mds_request *req);
241 
242 /*
243  * an in-flight mds request
244  */
245 struct ceph_mds_request {
246 	u64 r_tid;                   /* transaction id */
247 	struct rb_node r_node;
248 	struct ceph_mds_client *r_mdsc;
249 
250 	struct kref       r_kref;
251 	int r_op;                    /* mds op code */
252 
253 	/* operation on what? */
254 	struct inode *r_inode;              /* arg1 */
255 	struct dentry *r_dentry;            /* arg1 */
256 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
257 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
258 	char *r_path1, *r_path2;
259 	struct ceph_vino r_ino1, r_ino2;
260 
261 	struct inode *r_parent;		    /* parent dir inode */
262 	struct inode *r_target_inode;       /* resulting inode */
263 
264 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
265 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
266 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
267 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
268 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
269 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
270 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
271 #define CEPH_MDS_R_ASYNC		(8) /* async request */
272 	unsigned long	r_req_flags;
273 
274 	struct mutex r_fill_mutex;
275 
276 	union ceph_mds_request_args r_args;
277 	int r_fmode;        /* file mode, if expecting cap */
278 	kuid_t r_uid;
279 	kgid_t r_gid;
280 	int r_request_release_offset;
281 	struct timespec64 r_stamp;
282 
283 	/* for choosing which mds to send this request to */
284 	int r_direct_mode;
285 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
286 
287 	/* data payload is used for xattr ops */
288 	struct ceph_pagelist *r_pagelist;
289 
290 	/* what caps shall we drop? */
291 	int r_inode_drop, r_inode_unless;
292 	int r_dentry_drop, r_dentry_unless;
293 	int r_old_dentry_drop, r_old_dentry_unless;
294 	struct inode *r_old_inode;
295 	int r_old_inode_drop, r_old_inode_unless;
296 
297 	struct ceph_msg  *r_request;  /* original request */
298 	struct ceph_msg  *r_reply;
299 	struct ceph_mds_reply_info_parsed r_reply_info;
300 	int r_err;
301 
302 
303 	struct page *r_locked_page;
304 	int r_dir_caps;
305 	int r_num_caps;
306 	u32               r_readdir_offset;
307 
308 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
309 	unsigned long r_started;  /* start time to measure timeout against */
310 	unsigned long r_start_latency;  /* start time to measure latency */
311 	unsigned long r_end_latency;    /* finish time to measure latency */
312 	unsigned long r_request_started; /* start time for mds request only,
313 					    used to measure lease durations */
314 
315 	/* link unsafe requests to parent directory, for fsync */
316 	struct inode	*r_unsafe_dir;
317 	struct list_head r_unsafe_dir_item;
318 
319 	/* unsafe requests that modify the target inode */
320 	struct list_head r_unsafe_target_item;
321 
322 	struct ceph_mds_session *r_session;
323 
324 	int               r_attempts;   /* resend attempts */
325 	int               r_num_fwd;    /* number of forward attempts */
326 	int               r_resend_mds; /* mds to resend to next, if any*/
327 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
328 	u64		  r_deleg_ino;
329 
330 	struct list_head  r_wait;
331 	struct completion r_completion;
332 	struct completion r_safe_completion;
333 	ceph_mds_request_callback_t r_callback;
334 	ceph_mds_request_wait_callback_t r_wait_for_completion;
335 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
336 
337 	long long	  r_dir_release_cnt;
338 	long long	  r_dir_ordered_cnt;
339 	int		  r_readdir_cache_idx;
340 
341 	struct ceph_cap_reservation r_caps_reservation;
342 };
343 
344 struct ceph_pool_perm {
345 	struct rb_node node;
346 	int perm;
347 	s64 pool;
348 	size_t pool_ns_len;
349 	char pool_ns[];
350 };
351 
352 struct ceph_snapid_map {
353 	struct rb_node node;
354 	struct list_head lru;
355 	atomic_t ref;
356 	u64 snap;
357 	dev_t dev;
358 	unsigned long last_used;
359 };
360 
361 /*
362  * node for list of quotarealm inodes that are not visible from the filesystem
363  * mountpoint, but required to handle, e.g. quotas.
364  */
365 struct ceph_quotarealm_inode {
366 	struct rb_node node;
367 	u64 ino;
368 	unsigned long timeout; /* last time a lookup failed for this inode */
369 	struct mutex mutex;
370 	struct inode *inode;
371 };
372 
373 struct cap_wait {
374 	struct list_head	list;
375 	unsigned long		ino;
376 	pid_t			tgid;
377 	int			need;
378 	int			want;
379 };
380 
381 /*
382  * mds client state
383  */
384 struct ceph_mds_client {
385 	struct ceph_fs_client  *fsc;
386 	struct mutex            mutex;         /* all nested structures */
387 
388 	struct ceph_mdsmap      *mdsmap;
389 	struct completion       safe_umount_waiters;
390 	wait_queue_head_t       session_close_wq;
391 	struct list_head        waiting_for_map;
392 	int 			mdsmap_err;
393 
394 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
395 	atomic_t		num_sessions;
396 	int                     max_sessions;  /* len of s_mds_sessions */
397 	int                     stopping;      /* true if shutting down */
398 
399 	atomic64_t		quotarealms_count; /* # realms with quota */
400 	/*
401 	 * We keep a list of inodes we don't see in the mountpoint but that we
402 	 * need to track quota realms.
403 	 */
404 	struct rb_root		quotarealms_inodes;
405 	struct mutex		quotarealms_inodes_mutex;
406 
407 	/*
408 	 * snap_rwsem will cover cap linkage into snaprealms, and
409 	 * realm snap contexts.  (later, we can do per-realm snap
410 	 * contexts locks..)  the empty list contains realms with no
411 	 * references (implying they contain no inodes with caps) that
412 	 * should be destroyed.
413 	 */
414 	u64			last_snap_seq;
415 	struct rw_semaphore     snap_rwsem;
416 	struct rb_root          snap_realms;
417 	struct list_head        snap_empty;
418 	int			num_snap_realms;
419 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
420 
421 	u64                    last_tid;      /* most recent mds request */
422 	u64                    oldest_tid;    /* oldest incomplete mds request,
423 						 excluding setfilelock requests */
424 	struct rb_root         request_tree;  /* pending mds requests */
425 	struct delayed_work    delayed_work;  /* delayed work */
426 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
427 	struct list_head cap_delay_list;   /* caps with delayed release */
428 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
429 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
430 	spinlock_t       snap_flush_lock;
431 
432 	u64               last_cap_flush_tid;
433 	struct list_head  cap_flush_list;
434 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
435 	int               num_cap_flushing; /* # caps we are flushing */
436 	spinlock_t        cap_dirty_lock;   /* protects above items */
437 	wait_queue_head_t cap_flushing_wq;
438 
439 	struct work_struct cap_reclaim_work;
440 	atomic_t	   cap_reclaim_pending;
441 
442 	/*
443 	 * Cap reservations
444 	 *
445 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
446 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
447 	 * memory needed to successfully process an MDS response.  (If an MDS
448 	 * sends us cap information and we fail to process it, we will have
449 	 * problems due to the client and MDS being out of sync.)
450 	 *
451 	 * Reservations are 'owned' by a ceph_cap_reservation context.
452 	 */
453 	spinlock_t	caps_list_lock;
454 	struct		list_head caps_list; /* unused (reserved or
455 						unreserved) */
456 	struct		list_head cap_wait_list;
457 	int		caps_total_count;    /* total caps allocated */
458 	int		caps_use_count;      /* in use */
459 	int		caps_use_max;	     /* max used caps */
460 	int		caps_reserve_count;  /* unused, reserved */
461 	int		caps_avail_count;    /* unused, unreserved */
462 	int		caps_min_count;      /* keep at least this many
463 						(unreserved) */
464 	spinlock_t	  dentry_list_lock;
465 	struct list_head  dentry_leases;     /* fifo list */
466 	struct list_head  dentry_dir_leases; /* lru list */
467 
468 	struct ceph_client_metric metric;
469 
470 	spinlock_t		snapid_map_lock;
471 	struct rb_root		snapid_map_tree;
472 	struct list_head	snapid_map_lru;
473 
474 	struct rw_semaphore     pool_perm_rwsem;
475 	struct rb_root		pool_perm_tree;
476 
477 	char nodename[__NEW_UTS_LEN + 1];
478 };
479 
480 extern const char *ceph_mds_op_name(int op);
481 
482 extern bool check_session_state(struct ceph_mds_session *s);
483 
484 extern struct ceph_mds_session *
485 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
486 
487 extern const char *ceph_session_state_name(int s);
488 
489 extern struct ceph_mds_session *
490 ceph_get_mds_session(struct ceph_mds_session *s);
491 extern void ceph_put_mds_session(struct ceph_mds_session *s);
492 
493 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
494 			     struct ceph_msg *msg, int mds);
495 
496 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
497 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
498 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
499 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
500 
501 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
502 
503 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
504 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
505 					   struct inode *dir);
506 extern struct ceph_mds_request *
507 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
508 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
509 				    struct inode *dir,
510 				    struct ceph_mds_request *req);
511 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
512 				struct inode *dir,
513 				struct ceph_mds_request *req);
514 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
515 extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
516 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
517 {
518 	kref_get(&req->r_kref);
519 }
520 extern void ceph_mdsc_release_request(struct kref *kref);
521 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
522 {
523 	kref_put(&req->r_kref, ceph_mdsc_release_request);
524 }
525 
526 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
527 				    struct ceph_cap *cap);
528 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
529 				    struct ceph_mds_session *session);
530 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
531 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
532 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
533 				     int (*cb)(struct inode *,
534 					       struct ceph_cap *, void *),
535 				     void *arg);
536 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
537 
538 static inline void ceph_mdsc_free_path(char *path, int len)
539 {
540 	if (!IS_ERR_OR_NULL(path))
541 		__putname(path - (PATH_MAX - 1 - len));
542 }
543 
544 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
545 				  int stop_on_nosnap);
546 
547 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
548 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
549 				     struct dentry *dentry, char action,
550 				     u32 seq);
551 
552 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
553 				    struct ceph_msg *msg);
554 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
555 				   struct ceph_msg *msg);
556 
557 extern struct ceph_mds_session *
558 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
559 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
560 					  struct ceph_mds_session *session);
561 
562 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
563 			  struct ceph_mds_session *session,
564 			  int max_caps);
565 
566 static inline int ceph_wait_on_async_create(struct inode *inode)
567 {
568 	struct ceph_inode_info *ci = ceph_inode(inode);
569 
570 	return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
571 			   TASK_INTERRUPTIBLE);
572 }
573 
574 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
575 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
576 #endif
577