xref: /linux/fs/ceph/mds_client.h (revision 0883c2c06fb5bcf5b9e008270827e63c09a88c1e)
1 #ifndef _FS_CEPH_MDS_CLIENT_H
2 #define _FS_CEPH_MDS_CLIENT_H
3 
4 #include <linux/completion.h>
5 #include <linux/kref.h>
6 #include <linux/list.h>
7 #include <linux/mutex.h>
8 #include <linux/rbtree.h>
9 #include <linux/spinlock.h>
10 
11 #include <linux/ceph/types.h>
12 #include <linux/ceph/messenger.h>
13 #include <linux/ceph/mdsmap.h>
14 #include <linux/ceph/auth.h>
15 
16 /*
17  * Some lock dependencies:
18  *
19  * session->s_mutex
20  *         mdsc->mutex
21  *
22  *         mdsc->snap_rwsem
23  *
24  *         ci->i_ceph_lock
25  *                 mdsc->snap_flush_lock
26  *                 mdsc->cap_delay_lock
27  *
28  */
29 
30 struct ceph_fs_client;
31 struct ceph_cap;
32 
33 /*
34  * parsed info about a single inode.  pointers are into the encoded
35  * on-wire structures within the mds reply message payload.
36  */
37 struct ceph_mds_reply_info_in {
38 	struct ceph_mds_reply_inode *in;
39 	struct ceph_dir_layout dir_layout;
40 	u32 symlink_len;
41 	char *symlink;
42 	u32 xattr_len;
43 	char *xattr_data;
44 	u64 inline_version;
45 	u32 inline_len;
46 	char *inline_data;
47 	u32 pool_ns_len;
48 };
49 
50 struct ceph_mds_reply_dir_entry {
51 	char                          *name;
52 	u32                           name_len;
53 	struct ceph_mds_reply_lease   *lease;
54 	struct ceph_mds_reply_info_in inode;
55 	loff_t			      offset;
56 };
57 
58 /*
59  * parsed info about an mds reply, including information about
60  * either: 1) the target inode and/or its parent directory and dentry,
61  * and directory contents (for readdir results), or
62  * 2) the file range lock info (for fcntl F_GETLK results).
63  */
64 struct ceph_mds_reply_info_parsed {
65 	struct ceph_mds_reply_head    *head;
66 
67 	/* trace */
68 	struct ceph_mds_reply_info_in diri, targeti;
69 	struct ceph_mds_reply_dirfrag *dirfrag;
70 	char                          *dname;
71 	u32                           dname_len;
72 	struct ceph_mds_reply_lease   *dlease;
73 
74 	/* extra */
75 	union {
76 		/* for fcntl F_GETLK results */
77 		struct ceph_filelock *filelock_reply;
78 
79 		/* for readdir results */
80 		struct {
81 			struct ceph_mds_reply_dirfrag *dir_dir;
82 			size_t			      dir_buf_size;
83 			int                           dir_nr;
84 			bool			      dir_complete;
85 			bool			      dir_end;
86 			bool			      hash_order;
87 			struct ceph_mds_reply_dir_entry  *dir_entries;
88 		};
89 
90 		/* for create results */
91 		struct {
92 			bool has_create_ino;
93 			u64 ino;
94 		};
95 	};
96 
97 	/* encoded blob describing snapshot contexts for certain
98 	   operations (e.g., open) */
99 	void *snapblob;
100 	int snapblob_len;
101 };
102 
103 
104 /*
105  * cap releases are batched and sent to the MDS en masse.
106  */
107 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE -			\
108 				sizeof(struct ceph_mds_cap_release)) /	\
109 			       sizeof(struct ceph_mds_cap_item))
110 
111 
112 /*
113  * state associated with each MDS<->client session
114  */
115 enum {
116 	CEPH_MDS_SESSION_NEW = 1,
117 	CEPH_MDS_SESSION_OPENING = 2,
118 	CEPH_MDS_SESSION_OPEN = 3,
119 	CEPH_MDS_SESSION_HUNG = 4,
120 	CEPH_MDS_SESSION_CLOSING = 5,
121 	CEPH_MDS_SESSION_RESTARTING = 6,
122 	CEPH_MDS_SESSION_RECONNECTING = 7,
123 };
124 
125 struct ceph_mds_session {
126 	struct ceph_mds_client *s_mdsc;
127 	int               s_mds;
128 	int               s_state;
129 	unsigned long     s_ttl;      /* time until mds kills us */
130 	u64               s_seq;      /* incoming msg seq # */
131 	struct mutex      s_mutex;    /* serialize session messages */
132 
133 	struct ceph_connection s_con;
134 
135 	struct ceph_auth_handshake s_auth;
136 
137 	/* protected by s_gen_ttl_lock */
138 	spinlock_t        s_gen_ttl_lock;
139 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
140 	unsigned long     s_cap_ttl;  /* when session caps expire */
141 
142 	/* protected by s_cap_lock */
143 	spinlock_t        s_cap_lock;
144 	struct list_head  s_caps;     /* all caps issued by this session */
145 	int               s_nr_caps, s_trim_caps;
146 	int               s_num_cap_releases;
147 	int		  s_cap_reconnect;
148 	int		  s_readonly;
149 	struct list_head  s_cap_releases; /* waiting cap_release messages */
150 	struct ceph_cap  *s_cap_iterator;
151 
152 	/* protected by mutex */
153 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
154 	struct list_head  s_cap_snaps_flushing;
155 	unsigned long     s_renew_requested; /* last time we sent a renew req */
156 	u64               s_renew_seq;
157 
158 	atomic_t          s_ref;
159 	struct list_head  s_waiting;  /* waiting requests */
160 	struct list_head  s_unsafe;   /* unsafe requests */
161 };
162 
163 /*
164  * modes of choosing which MDS to send a request to
165  */
166 enum {
167 	USE_ANY_MDS,
168 	USE_RANDOM_MDS,
169 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
170 };
171 
172 struct ceph_mds_request;
173 struct ceph_mds_client;
174 
175 /*
176  * request completion callback
177  */
178 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
179 					     struct ceph_mds_request *req);
180 /*
181  * wait for request completion callback
182  */
183 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
184 						 struct ceph_mds_request *req);
185 
186 /*
187  * an in-flight mds request
188  */
189 struct ceph_mds_request {
190 	u64 r_tid;                   /* transaction id */
191 	struct rb_node r_node;
192 	struct ceph_mds_client *r_mdsc;
193 
194 	int r_op;                    /* mds op code */
195 
196 	/* operation on what? */
197 	struct inode *r_inode;              /* arg1 */
198 	struct dentry *r_dentry;            /* arg1 */
199 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
200 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
201 	char *r_path1, *r_path2;
202 	struct ceph_vino r_ino1, r_ino2;
203 
204 	struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
205 	struct inode *r_target_inode;       /* resulting inode */
206 
207 	struct mutex r_fill_mutex;
208 
209 	union ceph_mds_request_args r_args;
210 	int r_fmode;        /* file mode, if expecting cap */
211 	kuid_t r_uid;
212 	kgid_t r_gid;
213 	struct timespec r_stamp;
214 
215 	/* for choosing which mds to send this request to */
216 	int r_direct_mode;
217 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
218 	bool r_direct_is_hash;  /* true if r_direct_hash is valid */
219 
220 	/* data payload is used for xattr ops */
221 	struct ceph_pagelist *r_pagelist;
222 
223 	/* what caps shall we drop? */
224 	int r_inode_drop, r_inode_unless;
225 	int r_dentry_drop, r_dentry_unless;
226 	int r_old_dentry_drop, r_old_dentry_unless;
227 	struct inode *r_old_inode;
228 	int r_old_inode_drop, r_old_inode_unless;
229 
230 	struct ceph_msg  *r_request;  /* original request */
231 	int r_request_release_offset;
232 	struct ceph_msg  *r_reply;
233 	struct ceph_mds_reply_info_parsed r_reply_info;
234 	struct page *r_locked_page;
235 	int r_err;
236 	bool r_aborted;
237 
238 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
239 	unsigned long r_started;  /* start time to measure timeout against */
240 	unsigned long r_request_started; /* start time for mds request only,
241 					    used to measure lease durations */
242 
243 	/* link unsafe requests to parent directory, for fsync */
244 	struct inode	*r_unsafe_dir;
245 	struct list_head r_unsafe_dir_item;
246 
247 	/* unsafe requests that modify the target inode */
248 	struct list_head r_unsafe_target_item;
249 
250 	struct ceph_mds_session *r_session;
251 
252 	int               r_attempts;   /* resend attempts */
253 	int               r_num_fwd;    /* number of forward attempts */
254 	int               r_resend_mds; /* mds to resend to next, if any*/
255 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
256 
257 	struct kref       r_kref;
258 	struct list_head  r_wait;
259 	struct completion r_completion;
260 	struct completion r_safe_completion;
261 	ceph_mds_request_callback_t r_callback;
262 	ceph_mds_request_wait_callback_t r_wait_for_completion;
263 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
264 	bool		  r_got_unsafe, r_got_safe, r_got_result;
265 
266 	bool              r_did_prepopulate;
267 	long long	  r_dir_release_cnt;
268 	long long	  r_dir_ordered_cnt;
269 	int		  r_readdir_cache_idx;
270 	u32               r_readdir_offset;
271 
272 	struct ceph_cap_reservation r_caps_reservation;
273 	int r_num_caps;
274 };
275 
276 struct ceph_pool_perm {
277 	struct rb_node node;
278 	u32 pool;
279 	int perm;
280 };
281 
282 /*
283  * mds client state
284  */
285 struct ceph_mds_client {
286 	struct ceph_fs_client  *fsc;
287 	struct mutex            mutex;         /* all nested structures */
288 
289 	struct ceph_mdsmap      *mdsmap;
290 	struct completion       safe_umount_waiters;
291 	wait_queue_head_t       session_close_wq;
292 	struct list_head        waiting_for_map;
293 
294 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
295 	atomic_t		num_sessions;
296 	int                     max_sessions;  /* len of s_mds_sessions */
297 	int                     stopping;      /* true if shutting down */
298 
299 	/*
300 	 * snap_rwsem will cover cap linkage into snaprealms, and
301 	 * realm snap contexts.  (later, we can do per-realm snap
302 	 * contexts locks..)  the empty list contains realms with no
303 	 * references (implying they contain no inodes with caps) that
304 	 * should be destroyed.
305 	 */
306 	u64			last_snap_seq;
307 	struct rw_semaphore     snap_rwsem;
308 	struct rb_root          snap_realms;
309 	struct list_head        snap_empty;
310 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
311 
312 	u64                    last_tid;      /* most recent mds request */
313 	u64                    oldest_tid;    /* oldest incomplete mds request,
314 						 excluding setfilelock requests */
315 	struct rb_root         request_tree;  /* pending mds requests */
316 	struct delayed_work    delayed_work;  /* delayed work */
317 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
318 	struct list_head cap_delay_list;   /* caps with delayed release */
319 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
320 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
321 	spinlock_t       snap_flush_lock;
322 
323 	u64               last_cap_flush_tid;
324 	struct rb_root    cap_flush_tree;
325 	struct list_head  cap_dirty;        /* inodes with dirty caps */
326 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
327 	int               num_cap_flushing; /* # caps we are flushing */
328 	spinlock_t        cap_dirty_lock;   /* protects above items */
329 	wait_queue_head_t cap_flushing_wq;
330 
331 	/*
332 	 * Cap reservations
333 	 *
334 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
335 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
336 	 * memory needed to successfully process an MDS response.  (If an MDS
337 	 * sends us cap information and we fail to process it, we will have
338 	 * problems due to the client and MDS being out of sync.)
339 	 *
340 	 * Reservations are 'owned' by a ceph_cap_reservation context.
341 	 */
342 	spinlock_t	caps_list_lock;
343 	struct		list_head caps_list; /* unused (reserved or
344 						unreserved) */
345 	int		caps_total_count;    /* total caps allocated */
346 	int		caps_use_count;      /* in use */
347 	int		caps_reserve_count;  /* unused, reserved */
348 	int		caps_avail_count;    /* unused, unreserved */
349 	int		caps_min_count;      /* keep at least this many
350 						(unreserved) */
351 	spinlock_t	  dentry_lru_lock;
352 	struct list_head  dentry_lru;
353 	int		  num_dentry;
354 
355 	struct rw_semaphore     pool_perm_rwsem;
356 	struct rb_root		pool_perm_tree;
357 };
358 
359 extern const char *ceph_mds_op_name(int op);
360 
361 extern struct ceph_mds_session *
362 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
363 
364 static inline struct ceph_mds_session *
365 ceph_get_mds_session(struct ceph_mds_session *s)
366 {
367 	atomic_inc(&s->s_ref);
368 	return s;
369 }
370 
371 extern const char *ceph_session_state_name(int s);
372 
373 extern void ceph_put_mds_session(struct ceph_mds_session *s);
374 
375 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
376 			     struct ceph_msg *msg, int mds);
377 
378 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
379 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
380 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
381 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
382 
383 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
384 
385 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
386 				    struct inode *inode,
387 				    struct dentry *dn);
388 
389 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
390 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
391 					   struct inode *dir);
392 extern struct ceph_mds_request *
393 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
394 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
395 				     struct ceph_mds_request *req);
396 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
397 				struct inode *dir,
398 				struct ceph_mds_request *req);
399 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
400 {
401 	kref_get(&req->r_kref);
402 }
403 extern void ceph_mdsc_release_request(struct kref *kref);
404 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
405 {
406 	kref_put(&req->r_kref, ceph_mdsc_release_request);
407 }
408 
409 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
410 				   struct ceph_mds_session *session);
411 
412 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
413 
414 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
415 				  int stop_on_nosnap);
416 
417 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
418 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
419 				     struct inode *inode,
420 				     struct dentry *dentry, char action,
421 				     u32 seq);
422 
423 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
424 				 struct ceph_msg *msg);
425 
426 extern struct ceph_mds_session *
427 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
428 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
429 					  struct ceph_mds_session *session);
430 
431 #endif
432