xref: /linux/fs/ceph/mds_client.h (revision c4ee0af3fa0dc65f690fc908f02b8355f9576ea0)
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 };
45 
46 /*
47  * parsed info about an mds reply, including information about
48  * either: 1) the target inode and/or its parent directory and dentry,
49  * and directory contents (for readdir results), or
50  * 2) the file range lock info (for fcntl F_GETLK results).
51  */
52 struct ceph_mds_reply_info_parsed {
53 	struct ceph_mds_reply_head    *head;
54 
55 	/* trace */
56 	struct ceph_mds_reply_info_in diri, targeti;
57 	struct ceph_mds_reply_dirfrag *dirfrag;
58 	char                          *dname;
59 	u32                           dname_len;
60 	struct ceph_mds_reply_lease   *dlease;
61 
62 	/* extra */
63 	union {
64 		/* for fcntl F_GETLK results */
65 		struct ceph_filelock *filelock_reply;
66 
67 		/* for readdir results */
68 		struct {
69 			struct ceph_mds_reply_dirfrag *dir_dir;
70 			int                           dir_nr;
71 			char                          **dir_dname;
72 			u32                           *dir_dname_len;
73 			struct ceph_mds_reply_lease   **dir_dlease;
74 			struct ceph_mds_reply_info_in *dir_in;
75 			u8                            dir_complete, dir_end;
76 		};
77 
78 		/* for create results */
79 		struct {
80 			bool has_create_ino;
81 			u64 ino;
82 		};
83 	};
84 
85 	/* encoded blob describing snapshot contexts for certain
86 	   operations (e.g., open) */
87 	void *snapblob;
88 	int snapblob_len;
89 };
90 
91 
92 /*
93  * cap releases are batched and sent to the MDS en masse.
94  */
95 #define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE -			\
96 				sizeof(struct ceph_mds_cap_release)) /	\
97 			       sizeof(struct ceph_mds_cap_item))
98 
99 
100 /*
101  * state associated with each MDS<->client session
102  */
103 enum {
104 	CEPH_MDS_SESSION_NEW = 1,
105 	CEPH_MDS_SESSION_OPENING = 2,
106 	CEPH_MDS_SESSION_OPEN = 3,
107 	CEPH_MDS_SESSION_HUNG = 4,
108 	CEPH_MDS_SESSION_CLOSING = 5,
109 	CEPH_MDS_SESSION_RESTARTING = 6,
110 	CEPH_MDS_SESSION_RECONNECTING = 7,
111 };
112 
113 struct ceph_mds_session {
114 	struct ceph_mds_client *s_mdsc;
115 	int               s_mds;
116 	int               s_state;
117 	unsigned long     s_ttl;      /* time until mds kills us */
118 	u64               s_seq;      /* incoming msg seq # */
119 	struct mutex      s_mutex;    /* serialize session messages */
120 
121 	struct ceph_connection s_con;
122 
123 	struct ceph_auth_handshake s_auth;
124 
125 	/* protected by s_gen_ttl_lock */
126 	spinlock_t        s_gen_ttl_lock;
127 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
128 	unsigned long     s_cap_ttl;  /* when session caps expire */
129 
130 	/* protected by s_cap_lock */
131 	spinlock_t        s_cap_lock;
132 	struct list_head  s_caps;     /* all caps issued by this session */
133 	int               s_nr_caps, s_trim_caps;
134 	int               s_num_cap_releases;
135 	int		  s_cap_reconnect;
136 	struct list_head  s_cap_releases; /* waiting cap_release messages */
137 	struct list_head  s_cap_releases_done; /* ready to send */
138 	struct ceph_cap  *s_cap_iterator;
139 
140 	/* protected by mutex */
141 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
142 	struct list_head  s_cap_snaps_flushing;
143 	unsigned long     s_renew_requested; /* last time we sent a renew req */
144 	u64               s_renew_seq;
145 
146 	atomic_t          s_ref;
147 	struct list_head  s_waiting;  /* waiting requests */
148 	struct list_head  s_unsafe;   /* unsafe requests */
149 };
150 
151 /*
152  * modes of choosing which MDS to send a request to
153  */
154 enum {
155 	USE_ANY_MDS,
156 	USE_RANDOM_MDS,
157 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
158 };
159 
160 struct ceph_mds_request;
161 struct ceph_mds_client;
162 
163 /*
164  * request completion callback
165  */
166 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
167 					     struct ceph_mds_request *req);
168 
169 /*
170  * an in-flight mds request
171  */
172 struct ceph_mds_request {
173 	u64 r_tid;                   /* transaction id */
174 	struct rb_node r_node;
175 	struct ceph_mds_client *r_mdsc;
176 
177 	int r_op;                    /* mds op code */
178 
179 	/* operation on what? */
180 	struct inode *r_inode;              /* arg1 */
181 	struct dentry *r_dentry;            /* arg1 */
182 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
183 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
184 	char *r_path1, *r_path2;
185 	struct ceph_vino r_ino1, r_ino2;
186 
187 	struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
188 	struct inode *r_target_inode;       /* resulting inode */
189 
190 	struct mutex r_fill_mutex;
191 
192 	union ceph_mds_request_args r_args;
193 	int r_fmode;        /* file mode, if expecting cap */
194 	kuid_t r_uid;
195 	kgid_t r_gid;
196 
197 	/* for choosing which mds to send this request to */
198 	int r_direct_mode;
199 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
200 	bool r_direct_is_hash;  /* true if r_direct_hash is valid */
201 
202 	/* data payload is used for xattr ops */
203 	struct page **r_pages;
204 	int r_num_pages;
205 	int r_data_len;
206 
207 	/* what caps shall we drop? */
208 	int r_inode_drop, r_inode_unless;
209 	int r_dentry_drop, r_dentry_unless;
210 	int r_old_dentry_drop, r_old_dentry_unless;
211 	struct inode *r_old_inode;
212 	int r_old_inode_drop, r_old_inode_unless;
213 
214 	struct ceph_msg  *r_request;  /* original request */
215 	int r_request_release_offset;
216 	struct ceph_msg  *r_reply;
217 	struct ceph_mds_reply_info_parsed r_reply_info;
218 	int r_err;
219 	bool r_aborted;
220 
221 	unsigned long r_timeout;  /* optional.  jiffies */
222 	unsigned long r_started;  /* start time to measure timeout against */
223 	unsigned long r_request_started; /* start time for mds request only,
224 					    used to measure lease durations */
225 
226 	/* link unsafe requests to parent directory, for fsync */
227 	struct inode	*r_unsafe_dir;
228 	struct list_head r_unsafe_dir_item;
229 
230 	struct ceph_mds_session *r_session;
231 
232 	int               r_attempts;   /* resend attempts */
233 	int               r_num_fwd;    /* number of forward attempts */
234 	int               r_resend_mds; /* mds to resend to next, if any*/
235 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
236 
237 	struct kref       r_kref;
238 	struct list_head  r_wait;
239 	struct completion r_completion;
240 	struct completion r_safe_completion;
241 	ceph_mds_request_callback_t r_callback;
242 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
243 	bool		  r_got_unsafe, r_got_safe, r_got_result;
244 
245 	bool              r_did_prepopulate;
246 	u32               r_readdir_offset;
247 
248 	struct ceph_cap_reservation r_caps_reservation;
249 	int r_num_caps;
250 };
251 
252 /*
253  * mds client state
254  */
255 struct ceph_mds_client {
256 	struct ceph_fs_client  *fsc;
257 	struct mutex            mutex;         /* all nested structures */
258 
259 	struct ceph_mdsmap      *mdsmap;
260 	struct completion       safe_umount_waiters;
261 	wait_queue_head_t       session_close_wq;
262 	struct list_head        waiting_for_map;
263 
264 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
265 	int                     max_sessions;  /* len of s_mds_sessions */
266 	int                     stopping;      /* true if shutting down */
267 
268 	/*
269 	 * snap_rwsem will cover cap linkage into snaprealms, and
270 	 * realm snap contexts.  (later, we can do per-realm snap
271 	 * contexts locks..)  the empty list contains realms with no
272 	 * references (implying they contain no inodes with caps) that
273 	 * should be destroyed.
274 	 */
275 	struct rw_semaphore     snap_rwsem;
276 	struct rb_root          snap_realms;
277 	struct list_head        snap_empty;
278 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
279 
280 	u64                    last_tid;      /* most recent mds request */
281 	struct rb_root         request_tree;  /* pending mds requests */
282 	struct delayed_work    delayed_work;  /* delayed work */
283 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
284 	struct list_head cap_delay_list;   /* caps with delayed release */
285 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
286 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
287 	spinlock_t       snap_flush_lock;
288 
289 	u64               cap_flush_seq;
290 	struct list_head  cap_dirty;        /* inodes with dirty caps */
291 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
292 	int               num_cap_flushing; /* # caps we are flushing */
293 	spinlock_t        cap_dirty_lock;   /* protects above items */
294 	wait_queue_head_t cap_flushing_wq;
295 
296 	/*
297 	 * Cap reservations
298 	 *
299 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
300 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
301 	 * memory needed to successfully process an MDS response.  (If an MDS
302 	 * sends us cap information and we fail to process it, we will have
303 	 * problems due to the client and MDS being out of sync.)
304 	 *
305 	 * Reservations are 'owned' by a ceph_cap_reservation context.
306 	 */
307 	spinlock_t	caps_list_lock;
308 	struct		list_head caps_list; /* unused (reserved or
309 						unreserved) */
310 	int		caps_total_count;    /* total caps allocated */
311 	int		caps_use_count;      /* in use */
312 	int		caps_reserve_count;  /* unused, reserved */
313 	int		caps_avail_count;    /* unused, unreserved */
314 	int		caps_min_count;      /* keep at least this many
315 						(unreserved) */
316 	spinlock_t	  dentry_lru_lock;
317 	struct list_head  dentry_lru;
318 	int		  num_dentry;
319 };
320 
321 extern const char *ceph_mds_op_name(int op);
322 
323 extern struct ceph_mds_session *
324 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
325 
326 static inline struct ceph_mds_session *
327 ceph_get_mds_session(struct ceph_mds_session *s)
328 {
329 	atomic_inc(&s->s_ref);
330 	return s;
331 }
332 
333 extern void ceph_put_mds_session(struct ceph_mds_session *s);
334 
335 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
336 			     struct ceph_msg *msg, int mds);
337 
338 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
339 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
340 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
341 
342 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
343 
344 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
345 				    struct inode *inode,
346 				    struct dentry *dn);
347 
348 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
349 
350 extern struct ceph_mds_request *
351 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
352 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
353 				     struct ceph_mds_request *req);
354 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
355 				struct inode *dir,
356 				struct ceph_mds_request *req);
357 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
358 {
359 	kref_get(&req->r_kref);
360 }
361 extern void ceph_mdsc_release_request(struct kref *kref);
362 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
363 {
364 	kref_put(&req->r_kref, ceph_mdsc_release_request);
365 }
366 
367 extern int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
368 				 struct ceph_mds_session *session);
369 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
370 				   struct ceph_mds_session *session);
371 
372 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
373 
374 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
375 				  int stop_on_nosnap);
376 
377 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
378 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
379 				     struct inode *inode,
380 				     struct dentry *dentry, char action,
381 				     u32 seq);
382 
383 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
384 				 struct ceph_msg *msg);
385 
386 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
387 					  struct ceph_mds_session *session);
388 
389 #endif
390