xref: /linux/fs/ceph/mds_client.h (revision b7019ac550eb3916f34d79db583e9b7ea2524afa)
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 #define CEPHFS_FEATURE_MIMIC		8
21 #define CEPHFS_FEATURE_REPLY_ENCODING	9
22 #define CEPHFS_FEATURE_RECLAIM_CLIENT	10
23 #define CEPHFS_FEATURE_LAZY_CAP_WANTED	11
24 #define CEPHFS_FEATURE_MULTI_RECONNECT  12
25 
26 #define CEPHFS_FEATURES_CLIENT_SUPPORTED { 	\
27 	0, 1, 2, 3, 4, 5, 6, 7,			\
28 	CEPHFS_FEATURE_MIMIC,			\
29 	CEPHFS_FEATURE_REPLY_ENCODING,		\
30 	CEPHFS_FEATURE_LAZY_CAP_WANTED,		\
31 	CEPHFS_FEATURE_MULTI_RECONNECT,		\
32 }
33 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
34 
35 
36 /*
37  * Some lock dependencies:
38  *
39  * session->s_mutex
40  *         mdsc->mutex
41  *
42  *         mdsc->snap_rwsem
43  *
44  *         ci->i_ceph_lock
45  *                 mdsc->snap_flush_lock
46  *                 mdsc->cap_delay_lock
47  *
48  */
49 
50 struct ceph_fs_client;
51 struct ceph_cap;
52 
53 /*
54  * parsed info about a single inode.  pointers are into the encoded
55  * on-wire structures within the mds reply message payload.
56  */
57 struct ceph_mds_reply_info_in {
58 	struct ceph_mds_reply_inode *in;
59 	struct ceph_dir_layout dir_layout;
60 	u32 symlink_len;
61 	char *symlink;
62 	u32 xattr_len;
63 	char *xattr_data;
64 	u64 inline_version;
65 	u32 inline_len;
66 	char *inline_data;
67 	u32 pool_ns_len;
68 	char *pool_ns_data;
69 	u64 max_bytes;
70 	u64 max_files;
71 	s32 dir_pin;
72 };
73 
74 struct ceph_mds_reply_dir_entry {
75 	char                          *name;
76 	u32                           name_len;
77 	struct ceph_mds_reply_lease   *lease;
78 	struct ceph_mds_reply_info_in inode;
79 	loff_t			      offset;
80 };
81 
82 /*
83  * parsed info about an mds reply, including information about
84  * either: 1) the target inode and/or its parent directory and dentry,
85  * and directory contents (for readdir results), or
86  * 2) the file range lock info (for fcntl F_GETLK results).
87  */
88 struct ceph_mds_reply_info_parsed {
89 	struct ceph_mds_reply_head    *head;
90 
91 	/* trace */
92 	struct ceph_mds_reply_info_in diri, targeti;
93 	struct ceph_mds_reply_dirfrag *dirfrag;
94 	char                          *dname;
95 	u32                           dname_len;
96 	struct ceph_mds_reply_lease   *dlease;
97 
98 	/* extra */
99 	union {
100 		/* for fcntl F_GETLK results */
101 		struct ceph_filelock *filelock_reply;
102 
103 		/* for readdir results */
104 		struct {
105 			struct ceph_mds_reply_dirfrag *dir_dir;
106 			size_t			      dir_buf_size;
107 			int                           dir_nr;
108 			bool			      dir_end;
109 			bool			      dir_complete;
110 			bool			      hash_order;
111 			bool			      offset_hash;
112 			struct ceph_mds_reply_dir_entry  *dir_entries;
113 		};
114 
115 		/* for create results */
116 		struct {
117 			bool has_create_ino;
118 			u64 ino;
119 		};
120 	};
121 
122 	/* encoded blob describing snapshot contexts for certain
123 	   operations (e.g., open) */
124 	void *snapblob;
125 	int snapblob_len;
126 };
127 
128 
129 /*
130  * cap releases are batched and sent to the MDS en masse.
131  *
132  * Account for per-message overhead of mds_cap_release header
133  * and __le32 for osd epoch barrier trailing field.
134  */
135 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
136 				sizeof(struct ceph_mds_cap_release)) /	\
137 			        sizeof(struct ceph_mds_cap_item))
138 
139 
140 /*
141  * state associated with each MDS<->client session
142  */
143 enum {
144 	CEPH_MDS_SESSION_NEW = 1,
145 	CEPH_MDS_SESSION_OPENING = 2,
146 	CEPH_MDS_SESSION_OPEN = 3,
147 	CEPH_MDS_SESSION_HUNG = 4,
148 	CEPH_MDS_SESSION_CLOSING = 5,
149 	CEPH_MDS_SESSION_RESTARTING = 6,
150 	CEPH_MDS_SESSION_RECONNECTING = 7,
151 	CEPH_MDS_SESSION_REJECTED = 8,
152 };
153 
154 struct ceph_mds_session {
155 	struct ceph_mds_client *s_mdsc;
156 	int               s_mds;
157 	int               s_state;
158 	unsigned long     s_ttl;      /* time until mds kills us */
159 	unsigned long	  s_features;
160 	u64               s_seq;      /* incoming msg seq # */
161 	struct mutex      s_mutex;    /* serialize session messages */
162 
163 	struct ceph_connection s_con;
164 
165 	struct ceph_auth_handshake s_auth;
166 
167 	/* protected by s_gen_ttl_lock */
168 	spinlock_t        s_gen_ttl_lock;
169 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
170 	unsigned long     s_cap_ttl;  /* when session caps expire */
171 
172 	/* protected by s_cap_lock */
173 	spinlock_t        s_cap_lock;
174 	struct list_head  s_caps;     /* all caps issued by this session */
175 	struct ceph_cap  *s_cap_iterator;
176 	int               s_nr_caps, s_trim_caps;
177 	int               s_num_cap_releases;
178 	int		  s_cap_reconnect;
179 	int		  s_readonly;
180 	struct list_head  s_cap_releases; /* waiting cap_release messages */
181 	struct work_struct s_cap_release_work;
182 
183 	/* protected by mutex */
184 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
185 	unsigned long     s_renew_requested; /* last time we sent a renew req */
186 	u64               s_renew_seq;
187 
188 	refcount_t        s_ref;
189 	struct list_head  s_waiting;  /* waiting requests */
190 	struct list_head  s_unsafe;   /* unsafe requests */
191 };
192 
193 /*
194  * modes of choosing which MDS to send a request to
195  */
196 enum {
197 	USE_ANY_MDS,
198 	USE_RANDOM_MDS,
199 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
200 };
201 
202 struct ceph_mds_request;
203 struct ceph_mds_client;
204 
205 /*
206  * request completion callback
207  */
208 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
209 					     struct ceph_mds_request *req);
210 /*
211  * wait for request completion callback
212  */
213 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
214 						 struct ceph_mds_request *req);
215 
216 /*
217  * an in-flight mds request
218  */
219 struct ceph_mds_request {
220 	u64 r_tid;                   /* transaction id */
221 	struct rb_node r_node;
222 	struct ceph_mds_client *r_mdsc;
223 
224 	int r_op;                    /* mds op code */
225 
226 	/* operation on what? */
227 	struct inode *r_inode;              /* arg1 */
228 	struct dentry *r_dentry;            /* arg1 */
229 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
230 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
231 	char *r_path1, *r_path2;
232 	struct ceph_vino r_ino1, r_ino2;
233 
234 	struct inode *r_parent;		    /* parent dir inode */
235 	struct inode *r_target_inode;       /* resulting inode */
236 
237 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
238 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
239 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
240 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
241 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
242 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
243 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
244 	unsigned long	r_req_flags;
245 
246 	struct mutex r_fill_mutex;
247 
248 	union ceph_mds_request_args r_args;
249 	int r_fmode;        /* file mode, if expecting cap */
250 	kuid_t r_uid;
251 	kgid_t r_gid;
252 	struct timespec64 r_stamp;
253 
254 	/* for choosing which mds to send this request to */
255 	int r_direct_mode;
256 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
257 
258 	/* data payload is used for xattr ops */
259 	struct ceph_pagelist *r_pagelist;
260 
261 	/* what caps shall we drop? */
262 	int r_inode_drop, r_inode_unless;
263 	int r_dentry_drop, r_dentry_unless;
264 	int r_old_dentry_drop, r_old_dentry_unless;
265 	struct inode *r_old_inode;
266 	int r_old_inode_drop, r_old_inode_unless;
267 
268 	struct ceph_msg  *r_request;  /* original request */
269 	int r_request_release_offset;
270 	struct ceph_msg  *r_reply;
271 	struct ceph_mds_reply_info_parsed r_reply_info;
272 	struct page *r_locked_page;
273 	int r_err;
274 
275 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
276 	unsigned long r_started;  /* start time to measure timeout against */
277 	unsigned long r_request_started; /* start time for mds request only,
278 					    used to measure lease durations */
279 
280 	/* link unsafe requests to parent directory, for fsync */
281 	struct inode	*r_unsafe_dir;
282 	struct list_head r_unsafe_dir_item;
283 
284 	/* unsafe requests that modify the target inode */
285 	struct list_head r_unsafe_target_item;
286 
287 	struct ceph_mds_session *r_session;
288 
289 	int               r_attempts;   /* resend attempts */
290 	int               r_num_fwd;    /* number of forward attempts */
291 	int               r_resend_mds; /* mds to resend to next, if any*/
292 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
293 
294 	struct kref       r_kref;
295 	struct list_head  r_wait;
296 	struct completion r_completion;
297 	struct completion r_safe_completion;
298 	ceph_mds_request_callback_t r_callback;
299 	ceph_mds_request_wait_callback_t r_wait_for_completion;
300 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
301 
302 	long long	  r_dir_release_cnt;
303 	long long	  r_dir_ordered_cnt;
304 	int		  r_readdir_cache_idx;
305 	u32               r_readdir_offset;
306 
307 	struct ceph_cap_reservation r_caps_reservation;
308 	int r_num_caps;
309 };
310 
311 struct ceph_pool_perm {
312 	struct rb_node node;
313 	int perm;
314 	s64 pool;
315 	size_t pool_ns_len;
316 	char pool_ns[];
317 };
318 
319 struct ceph_snapid_map {
320 	struct rb_node node;
321 	struct list_head lru;
322 	atomic_t ref;
323 	u64 snap;
324 	dev_t dev;
325 	unsigned long last_used;
326 };
327 
328 /*
329  * node for list of quotarealm inodes that are not visible from the filesystem
330  * mountpoint, but required to handle, e.g. quotas.
331  */
332 struct ceph_quotarealm_inode {
333 	struct rb_node node;
334 	u64 ino;
335 	unsigned long timeout; /* last time a lookup failed for this inode */
336 	struct mutex mutex;
337 	struct inode *inode;
338 };
339 
340 /*
341  * mds client state
342  */
343 struct ceph_mds_client {
344 	struct ceph_fs_client  *fsc;
345 	struct mutex            mutex;         /* all nested structures */
346 
347 	struct ceph_mdsmap      *mdsmap;
348 	struct completion       safe_umount_waiters;
349 	wait_queue_head_t       session_close_wq;
350 	struct list_head        waiting_for_map;
351 	int 			mdsmap_err;
352 
353 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
354 	atomic_t		num_sessions;
355 	int                     max_sessions;  /* len of s_mds_sessions */
356 	int                     stopping;      /* true if shutting down */
357 
358 	atomic64_t		quotarealms_count; /* # realms with quota */
359 	/*
360 	 * We keep a list of inodes we don't see in the mountpoint but that we
361 	 * need to track quota realms.
362 	 */
363 	struct rb_root		quotarealms_inodes;
364 	struct mutex		quotarealms_inodes_mutex;
365 
366 	/*
367 	 * snap_rwsem will cover cap linkage into snaprealms, and
368 	 * realm snap contexts.  (later, we can do per-realm snap
369 	 * contexts locks..)  the empty list contains realms with no
370 	 * references (implying they contain no inodes with caps) that
371 	 * should be destroyed.
372 	 */
373 	u64			last_snap_seq;
374 	struct rw_semaphore     snap_rwsem;
375 	struct rb_root          snap_realms;
376 	struct list_head        snap_empty;
377 	int			num_snap_realms;
378 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
379 
380 	u64                    last_tid;      /* most recent mds request */
381 	u64                    oldest_tid;    /* oldest incomplete mds request,
382 						 excluding setfilelock requests */
383 	struct rb_root         request_tree;  /* pending mds requests */
384 	struct delayed_work    delayed_work;  /* delayed work */
385 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
386 	struct list_head cap_delay_list;   /* caps with delayed release */
387 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
388 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
389 	spinlock_t       snap_flush_lock;
390 
391 	u64               last_cap_flush_tid;
392 	struct list_head  cap_flush_list;
393 	struct list_head  cap_dirty;        /* inodes with dirty caps */
394 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
395 	int               num_cap_flushing; /* # caps we are flushing */
396 	spinlock_t        cap_dirty_lock;   /* protects above items */
397 	wait_queue_head_t cap_flushing_wq;
398 
399 	struct work_struct cap_reclaim_work;
400 	atomic_t	   cap_reclaim_pending;
401 
402 	/*
403 	 * Cap reservations
404 	 *
405 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
406 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
407 	 * memory needed to successfully process an MDS response.  (If an MDS
408 	 * sends us cap information and we fail to process it, we will have
409 	 * problems due to the client and MDS being out of sync.)
410 	 *
411 	 * Reservations are 'owned' by a ceph_cap_reservation context.
412 	 */
413 	spinlock_t	caps_list_lock;
414 	struct		list_head caps_list; /* unused (reserved or
415 						unreserved) */
416 	int		caps_total_count;    /* total caps allocated */
417 	int		caps_use_count;      /* in use */
418 	int		caps_use_max;	     /* max used caps */
419 	int		caps_reserve_count;  /* unused, reserved */
420 	int		caps_avail_count;    /* unused, unreserved */
421 	int		caps_min_count;      /* keep at least this many
422 						(unreserved) */
423 	spinlock_t	  dentry_list_lock;
424 	struct list_head  dentry_leases;     /* fifo list */
425 	struct list_head  dentry_dir_leases; /* lru list */
426 
427 	spinlock_t		snapid_map_lock;
428 	struct rb_root		snapid_map_tree;
429 	struct list_head	snapid_map_lru;
430 
431 	struct rw_semaphore     pool_perm_rwsem;
432 	struct rb_root		pool_perm_tree;
433 
434 	char nodename[__NEW_UTS_LEN + 1];
435 };
436 
437 extern const char *ceph_mds_op_name(int op);
438 
439 extern struct ceph_mds_session *
440 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
441 
442 static inline struct ceph_mds_session *
443 ceph_get_mds_session(struct ceph_mds_session *s)
444 {
445 	refcount_inc(&s->s_ref);
446 	return s;
447 }
448 
449 extern const char *ceph_session_state_name(int s);
450 
451 extern void ceph_put_mds_session(struct ceph_mds_session *s);
452 
453 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
454 			     struct ceph_msg *msg, int mds);
455 
456 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
457 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
458 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
459 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
460 
461 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
462 
463 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
464 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
465 					   struct inode *dir);
466 extern struct ceph_mds_request *
467 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
468 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
469 				    struct inode *dir,
470 				    struct ceph_mds_request *req);
471 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
472 				struct inode *dir,
473 				struct ceph_mds_request *req);
474 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
475 {
476 	kref_get(&req->r_kref);
477 }
478 extern void ceph_mdsc_release_request(struct kref *kref);
479 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
480 {
481 	kref_put(&req->r_kref, ceph_mdsc_release_request);
482 }
483 
484 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
485 				    struct ceph_cap *cap);
486 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
487 				    struct ceph_mds_session *session);
488 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
489 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
490 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
491 				     int (*cb)(struct inode *,
492 					       struct ceph_cap *, void *),
493 				     void *arg);
494 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
495 
496 static inline void ceph_mdsc_free_path(char *path, int len)
497 {
498 	if (path)
499 		__putname(path - (PATH_MAX - 1 - len));
500 }
501 
502 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
503 				  int stop_on_nosnap);
504 
505 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
506 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
507 				     struct inode *inode,
508 				     struct dentry *dentry, char action,
509 				     u32 seq);
510 
511 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
512 				    struct ceph_msg *msg);
513 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
514 				   struct ceph_msg *msg);
515 
516 extern struct ceph_mds_session *
517 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
518 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
519 					  struct ceph_mds_session *session);
520 
521 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
522 			  struct ceph_mds_session *session,
523 			  int max_caps);
524 #endif
525