xref: /openbmc/linux/fs/ceph/mds_client.h (revision 78beef62)
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 	struct ceph_timespec btime;
73 	struct ceph_timespec snap_btime;
74 	u64 change_attr;
75 };
76 
77 struct ceph_mds_reply_dir_entry {
78 	char                          *name;
79 	u32                           name_len;
80 	struct ceph_mds_reply_lease   *lease;
81 	struct ceph_mds_reply_info_in inode;
82 	loff_t			      offset;
83 };
84 
85 /*
86  * parsed info about an mds reply, including information about
87  * either: 1) the target inode and/or its parent directory and dentry,
88  * and directory contents (for readdir results), or
89  * 2) the file range lock info (for fcntl F_GETLK results).
90  */
91 struct ceph_mds_reply_info_parsed {
92 	struct ceph_mds_reply_head    *head;
93 
94 	/* trace */
95 	struct ceph_mds_reply_info_in diri, targeti;
96 	struct ceph_mds_reply_dirfrag *dirfrag;
97 	char                          *dname;
98 	u32                           dname_len;
99 	struct ceph_mds_reply_lease   *dlease;
100 
101 	/* extra */
102 	union {
103 		/* for fcntl F_GETLK results */
104 		struct ceph_filelock *filelock_reply;
105 
106 		/* for readdir results */
107 		struct {
108 			struct ceph_mds_reply_dirfrag *dir_dir;
109 			size_t			      dir_buf_size;
110 			int                           dir_nr;
111 			bool			      dir_end;
112 			bool			      dir_complete;
113 			bool			      hash_order;
114 			bool			      offset_hash;
115 			struct ceph_mds_reply_dir_entry  *dir_entries;
116 		};
117 
118 		/* for create results */
119 		struct {
120 			bool has_create_ino;
121 			u64 ino;
122 		};
123 	};
124 
125 	/* encoded blob describing snapshot contexts for certain
126 	   operations (e.g., open) */
127 	void *snapblob;
128 	int snapblob_len;
129 };
130 
131 
132 /*
133  * cap releases are batched and sent to the MDS en masse.
134  *
135  * Account for per-message overhead of mds_cap_release header
136  * and __le32 for osd epoch barrier trailing field.
137  */
138 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
139 				sizeof(struct ceph_mds_cap_release)) /	\
140 			        sizeof(struct ceph_mds_cap_item))
141 
142 
143 /*
144  * state associated with each MDS<->client session
145  */
146 enum {
147 	CEPH_MDS_SESSION_NEW = 1,
148 	CEPH_MDS_SESSION_OPENING = 2,
149 	CEPH_MDS_SESSION_OPEN = 3,
150 	CEPH_MDS_SESSION_HUNG = 4,
151 	CEPH_MDS_SESSION_CLOSING = 5,
152 	CEPH_MDS_SESSION_RESTARTING = 6,
153 	CEPH_MDS_SESSION_RECONNECTING = 7,
154 	CEPH_MDS_SESSION_REJECTED = 8,
155 };
156 
157 struct ceph_mds_session {
158 	struct ceph_mds_client *s_mdsc;
159 	int               s_mds;
160 	int               s_state;
161 	unsigned long     s_ttl;      /* time until mds kills us */
162 	unsigned long	  s_features;
163 	u64               s_seq;      /* incoming msg seq # */
164 	struct mutex      s_mutex;    /* serialize session messages */
165 
166 	struct ceph_connection s_con;
167 
168 	struct ceph_auth_handshake s_auth;
169 
170 	/* protected by s_gen_ttl_lock */
171 	spinlock_t        s_gen_ttl_lock;
172 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
173 	unsigned long     s_cap_ttl;  /* when session caps expire */
174 
175 	/* protected by s_cap_lock */
176 	spinlock_t        s_cap_lock;
177 	struct list_head  s_caps;     /* all caps issued by this session */
178 	struct ceph_cap  *s_cap_iterator;
179 	int               s_nr_caps, s_trim_caps;
180 	int               s_num_cap_releases;
181 	int		  s_cap_reconnect;
182 	int		  s_readonly;
183 	struct list_head  s_cap_releases; /* waiting cap_release messages */
184 	struct work_struct s_cap_release_work;
185 
186 	/* protected by mutex */
187 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
188 	unsigned long     s_renew_requested; /* last time we sent a renew req */
189 	u64               s_renew_seq;
190 
191 	refcount_t        s_ref;
192 	struct list_head  s_waiting;  /* waiting requests */
193 	struct list_head  s_unsafe;   /* unsafe requests */
194 };
195 
196 /*
197  * modes of choosing which MDS to send a request to
198  */
199 enum {
200 	USE_ANY_MDS,
201 	USE_RANDOM_MDS,
202 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
203 };
204 
205 struct ceph_mds_request;
206 struct ceph_mds_client;
207 
208 /*
209  * request completion callback
210  */
211 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
212 					     struct ceph_mds_request *req);
213 /*
214  * wait for request completion callback
215  */
216 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
217 						 struct ceph_mds_request *req);
218 
219 /*
220  * an in-flight mds request
221  */
222 struct ceph_mds_request {
223 	u64 r_tid;                   /* transaction id */
224 	struct rb_node r_node;
225 	struct ceph_mds_client *r_mdsc;
226 
227 	int r_op;                    /* mds op code */
228 
229 	/* operation on what? */
230 	struct inode *r_inode;              /* arg1 */
231 	struct dentry *r_dentry;            /* arg1 */
232 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
233 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
234 	char *r_path1, *r_path2;
235 	struct ceph_vino r_ino1, r_ino2;
236 
237 	struct inode *r_parent;		    /* parent dir inode */
238 	struct inode *r_target_inode;       /* resulting inode */
239 
240 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
241 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
242 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
243 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
244 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
245 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
246 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
247 	unsigned long	r_req_flags;
248 
249 	struct mutex r_fill_mutex;
250 
251 	union ceph_mds_request_args r_args;
252 	int r_fmode;        /* file mode, if expecting cap */
253 	kuid_t r_uid;
254 	kgid_t r_gid;
255 	struct timespec64 r_stamp;
256 
257 	/* for choosing which mds to send this request to */
258 	int r_direct_mode;
259 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
260 
261 	/* data payload is used for xattr ops */
262 	struct ceph_pagelist *r_pagelist;
263 
264 	/* what caps shall we drop? */
265 	int r_inode_drop, r_inode_unless;
266 	int r_dentry_drop, r_dentry_unless;
267 	int r_old_dentry_drop, r_old_dentry_unless;
268 	struct inode *r_old_inode;
269 	int r_old_inode_drop, r_old_inode_unless;
270 
271 	struct ceph_msg  *r_request;  /* original request */
272 	int r_request_release_offset;
273 	struct ceph_msg  *r_reply;
274 	struct ceph_mds_reply_info_parsed r_reply_info;
275 	struct page *r_locked_page;
276 	int r_err;
277 
278 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
279 	unsigned long r_started;  /* start time to measure timeout against */
280 	unsigned long r_request_started; /* start time for mds request only,
281 					    used to measure lease durations */
282 
283 	/* link unsafe requests to parent directory, for fsync */
284 	struct inode	*r_unsafe_dir;
285 	struct list_head r_unsafe_dir_item;
286 
287 	/* unsafe requests that modify the target inode */
288 	struct list_head r_unsafe_target_item;
289 
290 	struct ceph_mds_session *r_session;
291 
292 	int               r_attempts;   /* resend attempts */
293 	int               r_num_fwd;    /* number of forward attempts */
294 	int               r_resend_mds; /* mds to resend to next, if any*/
295 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
296 
297 	struct kref       r_kref;
298 	struct list_head  r_wait;
299 	struct completion r_completion;
300 	struct completion r_safe_completion;
301 	ceph_mds_request_callback_t r_callback;
302 	ceph_mds_request_wait_callback_t r_wait_for_completion;
303 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
304 
305 	long long	  r_dir_release_cnt;
306 	long long	  r_dir_ordered_cnt;
307 	int		  r_readdir_cache_idx;
308 	u32               r_readdir_offset;
309 
310 	struct ceph_cap_reservation r_caps_reservation;
311 	int r_num_caps;
312 };
313 
314 struct ceph_pool_perm {
315 	struct rb_node node;
316 	int perm;
317 	s64 pool;
318 	size_t pool_ns_len;
319 	char pool_ns[];
320 };
321 
322 struct ceph_snapid_map {
323 	struct rb_node node;
324 	struct list_head lru;
325 	atomic_t ref;
326 	u64 snap;
327 	dev_t dev;
328 	unsigned long last_used;
329 };
330 
331 /*
332  * node for list of quotarealm inodes that are not visible from the filesystem
333  * mountpoint, but required to handle, e.g. quotas.
334  */
335 struct ceph_quotarealm_inode {
336 	struct rb_node node;
337 	u64 ino;
338 	unsigned long timeout; /* last time a lookup failed for this inode */
339 	struct mutex mutex;
340 	struct inode *inode;
341 };
342 
343 /*
344  * mds client state
345  */
346 struct ceph_mds_client {
347 	struct ceph_fs_client  *fsc;
348 	struct mutex            mutex;         /* all nested structures */
349 
350 	struct ceph_mdsmap      *mdsmap;
351 	struct completion       safe_umount_waiters;
352 	wait_queue_head_t       session_close_wq;
353 	struct list_head        waiting_for_map;
354 	int 			mdsmap_err;
355 
356 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
357 	atomic_t		num_sessions;
358 	int                     max_sessions;  /* len of s_mds_sessions */
359 	int                     stopping;      /* true if shutting down */
360 
361 	atomic64_t		quotarealms_count; /* # realms with quota */
362 	/*
363 	 * We keep a list of inodes we don't see in the mountpoint but that we
364 	 * need to track quota realms.
365 	 */
366 	struct rb_root		quotarealms_inodes;
367 	struct mutex		quotarealms_inodes_mutex;
368 
369 	/*
370 	 * snap_rwsem will cover cap linkage into snaprealms, and
371 	 * realm snap contexts.  (later, we can do per-realm snap
372 	 * contexts locks..)  the empty list contains realms with no
373 	 * references (implying they contain no inodes with caps) that
374 	 * should be destroyed.
375 	 */
376 	u64			last_snap_seq;
377 	struct rw_semaphore     snap_rwsem;
378 	struct rb_root          snap_realms;
379 	struct list_head        snap_empty;
380 	int			num_snap_realms;
381 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
382 
383 	u64                    last_tid;      /* most recent mds request */
384 	u64                    oldest_tid;    /* oldest incomplete mds request,
385 						 excluding setfilelock requests */
386 	struct rb_root         request_tree;  /* pending mds requests */
387 	struct delayed_work    delayed_work;  /* delayed work */
388 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
389 	struct list_head cap_delay_list;   /* caps with delayed release */
390 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
391 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
392 	spinlock_t       snap_flush_lock;
393 
394 	u64               last_cap_flush_tid;
395 	struct list_head  cap_flush_list;
396 	struct list_head  cap_dirty;        /* inodes with dirty caps */
397 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
398 	int               num_cap_flushing; /* # caps we are flushing */
399 	spinlock_t        cap_dirty_lock;   /* protects above items */
400 	wait_queue_head_t cap_flushing_wq;
401 
402 	struct work_struct cap_reclaim_work;
403 	atomic_t	   cap_reclaim_pending;
404 
405 	/*
406 	 * Cap reservations
407 	 *
408 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
409 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
410 	 * memory needed to successfully process an MDS response.  (If an MDS
411 	 * sends us cap information and we fail to process it, we will have
412 	 * problems due to the client and MDS being out of sync.)
413 	 *
414 	 * Reservations are 'owned' by a ceph_cap_reservation context.
415 	 */
416 	spinlock_t	caps_list_lock;
417 	struct		list_head caps_list; /* unused (reserved or
418 						unreserved) */
419 	int		caps_total_count;    /* total caps allocated */
420 	int		caps_use_count;      /* in use */
421 	int		caps_use_max;	     /* max used caps */
422 	int		caps_reserve_count;  /* unused, reserved */
423 	int		caps_avail_count;    /* unused, unreserved */
424 	int		caps_min_count;      /* keep at least this many
425 						(unreserved) */
426 	spinlock_t	  dentry_list_lock;
427 	struct list_head  dentry_leases;     /* fifo list */
428 	struct list_head  dentry_dir_leases; /* lru list */
429 
430 	spinlock_t		snapid_map_lock;
431 	struct rb_root		snapid_map_tree;
432 	struct list_head	snapid_map_lru;
433 
434 	struct rw_semaphore     pool_perm_rwsem;
435 	struct rb_root		pool_perm_tree;
436 
437 	char nodename[__NEW_UTS_LEN + 1];
438 };
439 
440 extern const char *ceph_mds_op_name(int op);
441 
442 extern struct ceph_mds_session *
443 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
444 
445 static inline struct ceph_mds_session *
446 ceph_get_mds_session(struct ceph_mds_session *s)
447 {
448 	refcount_inc(&s->s_ref);
449 	return s;
450 }
451 
452 extern const char *ceph_session_state_name(int s);
453 
454 extern void ceph_put_mds_session(struct ceph_mds_session *s);
455 
456 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
457 			     struct ceph_msg *msg, int mds);
458 
459 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
460 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
461 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
462 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
463 
464 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
465 
466 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
467 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
468 					   struct inode *dir);
469 extern struct ceph_mds_request *
470 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
471 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
472 				    struct inode *dir,
473 				    struct ceph_mds_request *req);
474 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
475 				struct inode *dir,
476 				struct ceph_mds_request *req);
477 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
478 {
479 	kref_get(&req->r_kref);
480 }
481 extern void ceph_mdsc_release_request(struct kref *kref);
482 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
483 {
484 	kref_put(&req->r_kref, ceph_mdsc_release_request);
485 }
486 
487 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
488 				    struct ceph_cap *cap);
489 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
490 				    struct ceph_mds_session *session);
491 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
492 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
493 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
494 				     int (*cb)(struct inode *,
495 					       struct ceph_cap *, void *),
496 				     void *arg);
497 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
498 
499 static inline void ceph_mdsc_free_path(char *path, int len)
500 {
501 	if (path)
502 		__putname(path - (PATH_MAX - 1 - len));
503 }
504 
505 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
506 				  int stop_on_nosnap);
507 
508 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
509 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
510 				     struct dentry *dentry, char action,
511 				     u32 seq);
512 
513 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
514 				    struct ceph_msg *msg);
515 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
516 				   struct ceph_msg *msg);
517 
518 extern struct ceph_mds_session *
519 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
520 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
521 					  struct ceph_mds_session *session);
522 
523 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
524 			  struct ceph_mds_session *session,
525 			  int max_caps);
526 #endif
527