xref: /linux/fs/ceph/mds_client.h (revision f79e4d5f92a129a1159c973735007d4ddc8541f3)
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 /*
20  * Some lock dependencies:
21  *
22  * session->s_mutex
23  *         mdsc->mutex
24  *
25  *         mdsc->snap_rwsem
26  *
27  *         ci->i_ceph_lock
28  *                 mdsc->snap_flush_lock
29  *                 mdsc->cap_delay_lock
30  *
31  */
32 
33 struct ceph_fs_client;
34 struct ceph_cap;
35 
36 /*
37  * parsed info about a single inode.  pointers are into the encoded
38  * on-wire structures within the mds reply message payload.
39  */
40 struct ceph_mds_reply_info_in {
41 	struct ceph_mds_reply_inode *in;
42 	struct ceph_dir_layout dir_layout;
43 	u32 symlink_len;
44 	char *symlink;
45 	u32 xattr_len;
46 	char *xattr_data;
47 	u64 inline_version;
48 	u32 inline_len;
49 	char *inline_data;
50 	u32 pool_ns_len;
51 	char *pool_ns_data;
52 	u64 max_bytes;
53 	u64 max_files;
54 };
55 
56 struct ceph_mds_reply_dir_entry {
57 	char                          *name;
58 	u32                           name_len;
59 	struct ceph_mds_reply_lease   *lease;
60 	struct ceph_mds_reply_info_in inode;
61 	loff_t			      offset;
62 };
63 
64 /*
65  * parsed info about an mds reply, including information about
66  * either: 1) the target inode and/or its parent directory and dentry,
67  * and directory contents (for readdir results), or
68  * 2) the file range lock info (for fcntl F_GETLK results).
69  */
70 struct ceph_mds_reply_info_parsed {
71 	struct ceph_mds_reply_head    *head;
72 
73 	/* trace */
74 	struct ceph_mds_reply_info_in diri, targeti;
75 	struct ceph_mds_reply_dirfrag *dirfrag;
76 	char                          *dname;
77 	u32                           dname_len;
78 	struct ceph_mds_reply_lease   *dlease;
79 
80 	/* extra */
81 	union {
82 		/* for fcntl F_GETLK results */
83 		struct ceph_filelock *filelock_reply;
84 
85 		/* for readdir results */
86 		struct {
87 			struct ceph_mds_reply_dirfrag *dir_dir;
88 			size_t			      dir_buf_size;
89 			int                           dir_nr;
90 			bool			      dir_end;
91 			bool			      dir_complete;
92 			bool			      hash_order;
93 			bool			      offset_hash;
94 			struct ceph_mds_reply_dir_entry  *dir_entries;
95 		};
96 
97 		/* for create results */
98 		struct {
99 			bool has_create_ino;
100 			u64 ino;
101 		};
102 	};
103 
104 	/* encoded blob describing snapshot contexts for certain
105 	   operations (e.g., open) */
106 	void *snapblob;
107 	int snapblob_len;
108 };
109 
110 
111 /*
112  * cap releases are batched and sent to the MDS en masse.
113  *
114  * Account for per-message overhead of mds_cap_release header
115  * and __le32 for osd epoch barrier trailing field.
116  */
117 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
118 				sizeof(struct ceph_mds_cap_release)) /	\
119 			        sizeof(struct ceph_mds_cap_item))
120 
121 
122 /*
123  * state associated with each MDS<->client session
124  */
125 enum {
126 	CEPH_MDS_SESSION_NEW = 1,
127 	CEPH_MDS_SESSION_OPENING = 2,
128 	CEPH_MDS_SESSION_OPEN = 3,
129 	CEPH_MDS_SESSION_HUNG = 4,
130 	CEPH_MDS_SESSION_CLOSING = 5,
131 	CEPH_MDS_SESSION_RESTARTING = 6,
132 	CEPH_MDS_SESSION_RECONNECTING = 7,
133 	CEPH_MDS_SESSION_REJECTED = 8,
134 };
135 
136 struct ceph_mds_session {
137 	struct ceph_mds_client *s_mdsc;
138 	int               s_mds;
139 	int               s_state;
140 	unsigned long     s_ttl;      /* time until mds kills us */
141 	u64               s_seq;      /* incoming msg seq # */
142 	struct mutex      s_mutex;    /* serialize session messages */
143 
144 	struct ceph_connection s_con;
145 
146 	struct ceph_auth_handshake s_auth;
147 
148 	/* protected by s_gen_ttl_lock */
149 	spinlock_t        s_gen_ttl_lock;
150 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
151 	unsigned long     s_cap_ttl;  /* when session caps expire */
152 
153 	/* protected by s_cap_lock */
154 	spinlock_t        s_cap_lock;
155 	struct list_head  s_caps;     /* all caps issued by this session */
156 	int               s_nr_caps, s_trim_caps;
157 	int               s_num_cap_releases;
158 	int		  s_cap_reconnect;
159 	int		  s_readonly;
160 	struct list_head  s_cap_releases; /* waiting cap_release messages */
161 	struct ceph_cap  *s_cap_iterator;
162 
163 	/* protected by mutex */
164 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
165 	unsigned long     s_renew_requested; /* last time we sent a renew req */
166 	u64               s_renew_seq;
167 
168 	refcount_t          s_ref;
169 	struct list_head  s_waiting;  /* waiting requests */
170 	struct list_head  s_unsafe;   /* unsafe requests */
171 };
172 
173 /*
174  * modes of choosing which MDS to send a request to
175  */
176 enum {
177 	USE_ANY_MDS,
178 	USE_RANDOM_MDS,
179 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
180 };
181 
182 struct ceph_mds_request;
183 struct ceph_mds_client;
184 
185 /*
186  * request completion callback
187  */
188 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
189 					     struct ceph_mds_request *req);
190 /*
191  * wait for request completion callback
192  */
193 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
194 						 struct ceph_mds_request *req);
195 
196 /*
197  * an in-flight mds request
198  */
199 struct ceph_mds_request {
200 	u64 r_tid;                   /* transaction id */
201 	struct rb_node r_node;
202 	struct ceph_mds_client *r_mdsc;
203 
204 	int r_op;                    /* mds op code */
205 
206 	/* operation on what? */
207 	struct inode *r_inode;              /* arg1 */
208 	struct dentry *r_dentry;            /* arg1 */
209 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
210 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
211 	char *r_path1, *r_path2;
212 	struct ceph_vino r_ino1, r_ino2;
213 
214 	struct inode *r_parent;		    /* parent dir inode */
215 	struct inode *r_target_inode;       /* resulting inode */
216 
217 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
218 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
219 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
220 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
221 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
222 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
223 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
224 	unsigned long	r_req_flags;
225 
226 	struct mutex r_fill_mutex;
227 
228 	union ceph_mds_request_args r_args;
229 	int r_fmode;        /* file mode, if expecting cap */
230 	kuid_t r_uid;
231 	kgid_t r_gid;
232 	struct timespec r_stamp;
233 
234 	/* for choosing which mds to send this request to */
235 	int r_direct_mode;
236 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
237 
238 	/* data payload is used for xattr ops */
239 	struct ceph_pagelist *r_pagelist;
240 
241 	/* what caps shall we drop? */
242 	int r_inode_drop, r_inode_unless;
243 	int r_dentry_drop, r_dentry_unless;
244 	int r_old_dentry_drop, r_old_dentry_unless;
245 	struct inode *r_old_inode;
246 	int r_old_inode_drop, r_old_inode_unless;
247 
248 	struct ceph_msg  *r_request;  /* original request */
249 	int r_request_release_offset;
250 	struct ceph_msg  *r_reply;
251 	struct ceph_mds_reply_info_parsed r_reply_info;
252 	struct page *r_locked_page;
253 	int r_err;
254 
255 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
256 	unsigned long r_started;  /* start time to measure timeout against */
257 	unsigned long r_request_started; /* start time for mds request only,
258 					    used to measure lease durations */
259 
260 	/* link unsafe requests to parent directory, for fsync */
261 	struct inode	*r_unsafe_dir;
262 	struct list_head r_unsafe_dir_item;
263 
264 	/* unsafe requests that modify the target inode */
265 	struct list_head r_unsafe_target_item;
266 
267 	struct ceph_mds_session *r_session;
268 
269 	int               r_attempts;   /* resend attempts */
270 	int               r_num_fwd;    /* number of forward attempts */
271 	int               r_resend_mds; /* mds to resend to next, if any*/
272 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
273 
274 	struct kref       r_kref;
275 	struct list_head  r_wait;
276 	struct completion r_completion;
277 	struct completion r_safe_completion;
278 	ceph_mds_request_callback_t r_callback;
279 	ceph_mds_request_wait_callback_t r_wait_for_completion;
280 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
281 
282 	long long	  r_dir_release_cnt;
283 	long long	  r_dir_ordered_cnt;
284 	int		  r_readdir_cache_idx;
285 	u32               r_readdir_offset;
286 
287 	struct ceph_cap_reservation r_caps_reservation;
288 	int r_num_caps;
289 };
290 
291 struct ceph_pool_perm {
292 	struct rb_node node;
293 	int perm;
294 	s64 pool;
295 	size_t pool_ns_len;
296 	char pool_ns[];
297 };
298 
299 /*
300  * mds client state
301  */
302 struct ceph_mds_client {
303 	struct ceph_fs_client  *fsc;
304 	struct mutex            mutex;         /* all nested structures */
305 
306 	struct ceph_mdsmap      *mdsmap;
307 	struct completion       safe_umount_waiters;
308 	wait_queue_head_t       session_close_wq;
309 	struct list_head        waiting_for_map;
310 	int 			mdsmap_err;
311 
312 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
313 	atomic_t		num_sessions;
314 	int                     max_sessions;  /* len of s_mds_sessions */
315 	int                     stopping;      /* true if shutting down */
316 
317 	atomic64_t		quotarealms_count; /* # realms with quota */
318 
319 	/*
320 	 * snap_rwsem will cover cap linkage into snaprealms, and
321 	 * realm snap contexts.  (later, we can do per-realm snap
322 	 * contexts locks..)  the empty list contains realms with no
323 	 * references (implying they contain no inodes with caps) that
324 	 * should be destroyed.
325 	 */
326 	u64			last_snap_seq;
327 	struct rw_semaphore     snap_rwsem;
328 	struct rb_root          snap_realms;
329 	struct list_head        snap_empty;
330 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
331 
332 	u64                    last_tid;      /* most recent mds request */
333 	u64                    oldest_tid;    /* oldest incomplete mds request,
334 						 excluding setfilelock requests */
335 	struct rb_root         request_tree;  /* pending mds requests */
336 	struct delayed_work    delayed_work;  /* delayed work */
337 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
338 	struct list_head cap_delay_list;   /* caps with delayed release */
339 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
340 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
341 	spinlock_t       snap_flush_lock;
342 
343 	u64               last_cap_flush_tid;
344 	struct list_head  cap_flush_list;
345 	struct list_head  cap_dirty;        /* inodes with dirty caps */
346 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
347 	int               num_cap_flushing; /* # caps we are flushing */
348 	spinlock_t        cap_dirty_lock;   /* protects above items */
349 	wait_queue_head_t cap_flushing_wq;
350 
351 	/*
352 	 * Cap reservations
353 	 *
354 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
355 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
356 	 * memory needed to successfully process an MDS response.  (If an MDS
357 	 * sends us cap information and we fail to process it, we will have
358 	 * problems due to the client and MDS being out of sync.)
359 	 *
360 	 * Reservations are 'owned' by a ceph_cap_reservation context.
361 	 */
362 	spinlock_t	caps_list_lock;
363 	struct		list_head caps_list; /* unused (reserved or
364 						unreserved) */
365 	int		caps_total_count;    /* total caps allocated */
366 	int		caps_use_count;      /* in use */
367 	int		caps_reserve_count;  /* unused, reserved */
368 	int		caps_avail_count;    /* unused, unreserved */
369 	int		caps_min_count;      /* keep at least this many
370 						(unreserved) */
371 	spinlock_t	  dentry_lru_lock;
372 	struct list_head  dentry_lru;
373 	int		  num_dentry;
374 
375 	struct rw_semaphore     pool_perm_rwsem;
376 	struct rb_root		pool_perm_tree;
377 
378 	char nodename[__NEW_UTS_LEN + 1];
379 };
380 
381 extern const char *ceph_mds_op_name(int op);
382 
383 extern struct ceph_mds_session *
384 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
385 
386 static inline struct ceph_mds_session *
387 ceph_get_mds_session(struct ceph_mds_session *s)
388 {
389 	refcount_inc(&s->s_ref);
390 	return s;
391 }
392 
393 extern const char *ceph_session_state_name(int s);
394 
395 extern void ceph_put_mds_session(struct ceph_mds_session *s);
396 
397 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
398 			     struct ceph_msg *msg, int mds);
399 
400 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
401 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
402 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
403 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
404 
405 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
406 
407 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
408 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
409 					   struct inode *dir);
410 extern struct ceph_mds_request *
411 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
412 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
413 				     struct ceph_mds_request *req);
414 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
415 				struct inode *dir,
416 				struct ceph_mds_request *req);
417 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
418 {
419 	kref_get(&req->r_kref);
420 }
421 extern void ceph_mdsc_release_request(struct kref *kref);
422 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
423 {
424 	kref_put(&req->r_kref, ceph_mdsc_release_request);
425 }
426 
427 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
428 				   struct ceph_mds_session *session);
429 
430 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
431 
432 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
433 				  int stop_on_nosnap);
434 
435 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
436 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
437 				     struct inode *inode,
438 				     struct dentry *dentry, char action,
439 				     u32 seq);
440 
441 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
442 				    struct ceph_msg *msg);
443 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
444 				   struct ceph_msg *msg);
445 
446 extern struct ceph_mds_session *
447 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
448 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
449 					  struct ceph_mds_session *session);
450 
451 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
452 			  struct ceph_mds_session *session,
453 			  int max_caps);
454 #endif
455