xref: /openbmc/linux/fs/ceph/mds_client.h (revision dc6a81c3)
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 
27 	CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_MULTI_RECONNECT,
28 };
29 
30 /*
31  * This will always have the highest feature bit value
32  * as the last element of the array.
33  */
34 #define CEPHFS_FEATURES_CLIENT_SUPPORTED {	\
35 	0, 1, 2, 3, 4, 5, 6, 7,			\
36 	CEPHFS_FEATURE_MIMIC,			\
37 	CEPHFS_FEATURE_REPLY_ENCODING,		\
38 	CEPHFS_FEATURE_LAZY_CAP_WANTED,		\
39 	CEPHFS_FEATURE_MULTI_RECONNECT,		\
40 						\
41 	CEPHFS_FEATURE_MAX,			\
42 }
43 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
44 
45 /*
46  * Some lock dependencies:
47  *
48  * session->s_mutex
49  *         mdsc->mutex
50  *
51  *         mdsc->snap_rwsem
52  *
53  *         ci->i_ceph_lock
54  *                 mdsc->snap_flush_lock
55  *                 mdsc->cap_delay_lock
56  *
57  */
58 
59 struct ceph_fs_client;
60 struct ceph_cap;
61 
62 /*
63  * parsed info about a single inode.  pointers are into the encoded
64  * on-wire structures within the mds reply message payload.
65  */
66 struct ceph_mds_reply_info_in {
67 	struct ceph_mds_reply_inode *in;
68 	struct ceph_dir_layout dir_layout;
69 	u32 symlink_len;
70 	char *symlink;
71 	u32 xattr_len;
72 	char *xattr_data;
73 	u64 inline_version;
74 	u32 inline_len;
75 	char *inline_data;
76 	u32 pool_ns_len;
77 	char *pool_ns_data;
78 	u64 max_bytes;
79 	u64 max_files;
80 	s32 dir_pin;
81 	struct ceph_timespec btime;
82 	struct ceph_timespec snap_btime;
83 	u64 change_attr;
84 };
85 
86 struct ceph_mds_reply_dir_entry {
87 	char                          *name;
88 	u32                           name_len;
89 	struct ceph_mds_reply_lease   *lease;
90 	struct ceph_mds_reply_info_in inode;
91 	loff_t			      offset;
92 };
93 
94 /*
95  * parsed info about an mds reply, including information about
96  * either: 1) the target inode and/or its parent directory and dentry,
97  * and directory contents (for readdir results), or
98  * 2) the file range lock info (for fcntl F_GETLK results).
99  */
100 struct ceph_mds_reply_info_parsed {
101 	struct ceph_mds_reply_head    *head;
102 
103 	/* trace */
104 	struct ceph_mds_reply_info_in diri, targeti;
105 	struct ceph_mds_reply_dirfrag *dirfrag;
106 	char                          *dname;
107 	u32                           dname_len;
108 	struct ceph_mds_reply_lease   *dlease;
109 
110 	/* extra */
111 	union {
112 		/* for fcntl F_GETLK results */
113 		struct ceph_filelock *filelock_reply;
114 
115 		/* for readdir results */
116 		struct {
117 			struct ceph_mds_reply_dirfrag *dir_dir;
118 			size_t			      dir_buf_size;
119 			int                           dir_nr;
120 			bool			      dir_end;
121 			bool			      dir_complete;
122 			bool			      hash_order;
123 			bool			      offset_hash;
124 			struct ceph_mds_reply_dir_entry  *dir_entries;
125 		};
126 
127 		/* for create results */
128 		struct {
129 			bool has_create_ino;
130 			u64 ino;
131 		};
132 	};
133 
134 	/* encoded blob describing snapshot contexts for certain
135 	   operations (e.g., open) */
136 	void *snapblob;
137 	int snapblob_len;
138 };
139 
140 
141 /*
142  * cap releases are batched and sent to the MDS en masse.
143  *
144  * Account for per-message overhead of mds_cap_release header
145  * and __le32 for osd epoch barrier trailing field.
146  */
147 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
148 				sizeof(struct ceph_mds_cap_release)) /	\
149 			        sizeof(struct ceph_mds_cap_item))
150 
151 
152 /*
153  * state associated with each MDS<->client session
154  */
155 enum {
156 	CEPH_MDS_SESSION_NEW = 1,
157 	CEPH_MDS_SESSION_OPENING = 2,
158 	CEPH_MDS_SESSION_OPEN = 3,
159 	CEPH_MDS_SESSION_HUNG = 4,
160 	CEPH_MDS_SESSION_RESTARTING = 5,
161 	CEPH_MDS_SESSION_RECONNECTING = 6,
162 	CEPH_MDS_SESSION_CLOSING = 7,
163 	CEPH_MDS_SESSION_CLOSED = 8,
164 	CEPH_MDS_SESSION_REJECTED = 9,
165 };
166 
167 struct ceph_mds_session {
168 	struct ceph_mds_client *s_mdsc;
169 	int               s_mds;
170 	int               s_state;
171 	unsigned long     s_ttl;      /* time until mds kills us */
172 	unsigned long	  s_features;
173 	u64               s_seq;      /* incoming msg seq # */
174 	struct mutex      s_mutex;    /* serialize session messages */
175 
176 	struct ceph_connection s_con;
177 
178 	struct ceph_auth_handshake s_auth;
179 
180 	/* protected by s_gen_ttl_lock */
181 	spinlock_t        s_gen_ttl_lock;
182 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
183 	unsigned long     s_cap_ttl;  /* when session caps expire */
184 
185 	/* protected by s_cap_lock */
186 	spinlock_t        s_cap_lock;
187 	refcount_t        s_ref;
188 	struct list_head  s_caps;     /* all caps issued by this session */
189 	struct ceph_cap  *s_cap_iterator;
190 	int               s_nr_caps;
191 	int               s_num_cap_releases;
192 	int		  s_cap_reconnect;
193 	int		  s_readonly;
194 	struct list_head  s_cap_releases; /* waiting cap_release messages */
195 	struct work_struct s_cap_release_work;
196 
197 	/* protected by mutex */
198 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
199 	unsigned long     s_renew_requested; /* last time we sent a renew req */
200 	u64               s_renew_seq;
201 
202 	struct list_head  s_waiting;  /* waiting requests */
203 	struct list_head  s_unsafe;   /* unsafe requests */
204 };
205 
206 /*
207  * modes of choosing which MDS to send a request to
208  */
209 enum {
210 	USE_ANY_MDS,
211 	USE_RANDOM_MDS,
212 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
213 };
214 
215 struct ceph_mds_request;
216 struct ceph_mds_client;
217 
218 /*
219  * request completion callback
220  */
221 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
222 					     struct ceph_mds_request *req);
223 /*
224  * wait for request completion callback
225  */
226 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
227 						 struct ceph_mds_request *req);
228 
229 /*
230  * an in-flight mds request
231  */
232 struct ceph_mds_request {
233 	u64 r_tid;                   /* transaction id */
234 	struct rb_node r_node;
235 	struct ceph_mds_client *r_mdsc;
236 
237 	struct kref       r_kref;
238 	int r_op;                    /* mds op code */
239 
240 	/* operation on what? */
241 	struct inode *r_inode;              /* arg1 */
242 	struct dentry *r_dentry;            /* arg1 */
243 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
244 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
245 	char *r_path1, *r_path2;
246 	struct ceph_vino r_ino1, r_ino2;
247 
248 	struct inode *r_parent;		    /* parent dir inode */
249 	struct inode *r_target_inode;       /* resulting inode */
250 
251 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
252 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
253 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
254 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
255 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
256 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
257 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
258 	unsigned long	r_req_flags;
259 
260 	struct mutex r_fill_mutex;
261 
262 	union ceph_mds_request_args r_args;
263 	int r_fmode;        /* file mode, if expecting cap */
264 	kuid_t r_uid;
265 	kgid_t r_gid;
266 	struct timespec64 r_stamp;
267 
268 	/* for choosing which mds to send this request to */
269 	int r_direct_mode;
270 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
271 
272 	/* data payload is used for xattr ops */
273 	struct ceph_pagelist *r_pagelist;
274 
275 	/* what caps shall we drop? */
276 	int r_inode_drop, r_inode_unless;
277 	int r_dentry_drop, r_dentry_unless;
278 	int r_old_dentry_drop, r_old_dentry_unless;
279 	struct inode *r_old_inode;
280 	int r_old_inode_drop, r_old_inode_unless;
281 
282 	struct ceph_msg  *r_request;  /* original request */
283 	int r_request_release_offset;
284 	struct ceph_msg  *r_reply;
285 	struct ceph_mds_reply_info_parsed r_reply_info;
286 	struct page *r_locked_page;
287 	int r_err;
288 
289 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
290 	unsigned long r_started;  /* start time to measure timeout against */
291 	unsigned long r_request_started; /* start time for mds request only,
292 					    used to measure lease durations */
293 
294 	/* link unsafe requests to parent directory, for fsync */
295 	struct inode	*r_unsafe_dir;
296 	struct list_head r_unsafe_dir_item;
297 
298 	/* unsafe requests that modify the target inode */
299 	struct list_head r_unsafe_target_item;
300 
301 	struct ceph_mds_session *r_session;
302 
303 	int               r_attempts;   /* resend attempts */
304 	int               r_num_fwd;    /* number of forward attempts */
305 	int               r_resend_mds; /* mds to resend to next, if any*/
306 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
307 
308 	struct list_head  r_wait;
309 	struct completion r_completion;
310 	struct completion r_safe_completion;
311 	ceph_mds_request_callback_t r_callback;
312 	ceph_mds_request_wait_callback_t r_wait_for_completion;
313 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
314 
315 	long long	  r_dir_release_cnt;
316 	long long	  r_dir_ordered_cnt;
317 	int		  r_readdir_cache_idx;
318 	u32               r_readdir_offset;
319 
320 	struct ceph_cap_reservation r_caps_reservation;
321 	int r_num_caps;
322 };
323 
324 struct ceph_pool_perm {
325 	struct rb_node node;
326 	int perm;
327 	s64 pool;
328 	size_t pool_ns_len;
329 	char pool_ns[];
330 };
331 
332 struct ceph_snapid_map {
333 	struct rb_node node;
334 	struct list_head lru;
335 	atomic_t ref;
336 	u64 snap;
337 	dev_t dev;
338 	unsigned long last_used;
339 };
340 
341 /*
342  * node for list of quotarealm inodes that are not visible from the filesystem
343  * mountpoint, but required to handle, e.g. quotas.
344  */
345 struct ceph_quotarealm_inode {
346 	struct rb_node node;
347 	u64 ino;
348 	unsigned long timeout; /* last time a lookup failed for this inode */
349 	struct mutex mutex;
350 	struct inode *inode;
351 };
352 
353 struct cap_wait {
354 	struct list_head	list;
355 	unsigned long		ino;
356 	pid_t			tgid;
357 	int			need;
358 	int			want;
359 };
360 
361 /*
362  * mds client state
363  */
364 struct ceph_mds_client {
365 	struct ceph_fs_client  *fsc;
366 	struct mutex            mutex;         /* all nested structures */
367 
368 	struct ceph_mdsmap      *mdsmap;
369 	struct completion       safe_umount_waiters;
370 	wait_queue_head_t       session_close_wq;
371 	struct list_head        waiting_for_map;
372 	int 			mdsmap_err;
373 
374 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
375 	atomic_t		num_sessions;
376 	int                     max_sessions;  /* len of s_mds_sessions */
377 	int                     stopping;      /* true if shutting down */
378 
379 	atomic64_t		quotarealms_count; /* # realms with quota */
380 	/*
381 	 * We keep a list of inodes we don't see in the mountpoint but that we
382 	 * need to track quota realms.
383 	 */
384 	struct rb_root		quotarealms_inodes;
385 	struct mutex		quotarealms_inodes_mutex;
386 
387 	/*
388 	 * snap_rwsem will cover cap linkage into snaprealms, and
389 	 * realm snap contexts.  (later, we can do per-realm snap
390 	 * contexts locks..)  the empty list contains realms with no
391 	 * references (implying they contain no inodes with caps) that
392 	 * should be destroyed.
393 	 */
394 	u64			last_snap_seq;
395 	struct rw_semaphore     snap_rwsem;
396 	struct rb_root          snap_realms;
397 	struct list_head        snap_empty;
398 	int			num_snap_realms;
399 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
400 
401 	u64                    last_tid;      /* most recent mds request */
402 	u64                    oldest_tid;    /* oldest incomplete mds request,
403 						 excluding setfilelock requests */
404 	struct rb_root         request_tree;  /* pending mds requests */
405 	struct delayed_work    delayed_work;  /* delayed work */
406 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
407 	struct list_head cap_delay_list;   /* caps with delayed release */
408 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
409 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
410 	spinlock_t       snap_flush_lock;
411 
412 	u64               last_cap_flush_tid;
413 	struct list_head  cap_flush_list;
414 	struct list_head  cap_dirty;        /* inodes with dirty caps */
415 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
416 	int               num_cap_flushing; /* # caps we are flushing */
417 	spinlock_t        cap_dirty_lock;   /* protects above items */
418 	wait_queue_head_t cap_flushing_wq;
419 
420 	struct work_struct cap_reclaim_work;
421 	atomic_t	   cap_reclaim_pending;
422 
423 	/*
424 	 * Cap reservations
425 	 *
426 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
427 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
428 	 * memory needed to successfully process an MDS response.  (If an MDS
429 	 * sends us cap information and we fail to process it, we will have
430 	 * problems due to the client and MDS being out of sync.)
431 	 *
432 	 * Reservations are 'owned' by a ceph_cap_reservation context.
433 	 */
434 	spinlock_t	caps_list_lock;
435 	struct		list_head caps_list; /* unused (reserved or
436 						unreserved) */
437 	struct		list_head cap_wait_list;
438 	int		caps_total_count;    /* total caps allocated */
439 	int		caps_use_count;      /* in use */
440 	int		caps_use_max;	     /* max used caps */
441 	int		caps_reserve_count;  /* unused, reserved */
442 	int		caps_avail_count;    /* unused, unreserved */
443 	int		caps_min_count;      /* keep at least this many
444 						(unreserved) */
445 	spinlock_t	  dentry_list_lock;
446 	struct list_head  dentry_leases;     /* fifo list */
447 	struct list_head  dentry_dir_leases; /* lru list */
448 
449 	spinlock_t		snapid_map_lock;
450 	struct rb_root		snapid_map_tree;
451 	struct list_head	snapid_map_lru;
452 
453 	struct rw_semaphore     pool_perm_rwsem;
454 	struct rb_root		pool_perm_tree;
455 
456 	char nodename[__NEW_UTS_LEN + 1];
457 };
458 
459 extern const char *ceph_mds_op_name(int op);
460 
461 extern struct ceph_mds_session *
462 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
463 
464 extern const char *ceph_session_state_name(int s);
465 
466 extern struct ceph_mds_session *
467 ceph_get_mds_session(struct ceph_mds_session *s);
468 extern void ceph_put_mds_session(struct ceph_mds_session *s);
469 
470 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
471 			     struct ceph_msg *msg, int mds);
472 
473 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
474 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
475 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
476 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
477 
478 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
479 
480 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
481 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
482 					   struct inode *dir);
483 extern struct ceph_mds_request *
484 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
485 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
486 				    struct inode *dir,
487 				    struct ceph_mds_request *req);
488 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
489 				struct inode *dir,
490 				struct ceph_mds_request *req);
491 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
492 {
493 	kref_get(&req->r_kref);
494 }
495 extern void ceph_mdsc_release_request(struct kref *kref);
496 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
497 {
498 	kref_put(&req->r_kref, ceph_mdsc_release_request);
499 }
500 
501 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
502 				    struct ceph_cap *cap);
503 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
504 				    struct ceph_mds_session *session);
505 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
506 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
507 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
508 				     int (*cb)(struct inode *,
509 					       struct ceph_cap *, void *),
510 				     void *arg);
511 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
512 
513 static inline void ceph_mdsc_free_path(char *path, int len)
514 {
515 	if (path)
516 		__putname(path - (PATH_MAX - 1 - len));
517 }
518 
519 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
520 				  int stop_on_nosnap);
521 
522 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
523 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
524 				     struct dentry *dentry, char action,
525 				     u32 seq);
526 
527 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
528 				    struct ceph_msg *msg);
529 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
530 				   struct ceph_msg *msg);
531 
532 extern struct ceph_mds_session *
533 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
534 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
535 					  struct ceph_mds_session *session);
536 
537 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
538 			  struct ceph_mds_session *session,
539 			  int max_caps);
540 #endif
541