xref: /openbmc/linux/fs/ceph/mds_client.h (revision 37185b33)
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 
79 	/* encoded blob describing snapshot contexts for certain
80 	   operations (e.g., open) */
81 	void *snapblob;
82 	int snapblob_len;
83 };
84 
85 
86 /*
87  * cap releases are batched and sent to the MDS en masse.
88  */
89 #define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE -			\
90 				sizeof(struct ceph_mds_cap_release)) /	\
91 			       sizeof(struct ceph_mds_cap_item))
92 
93 
94 /*
95  * state associated with each MDS<->client session
96  */
97 enum {
98 	CEPH_MDS_SESSION_NEW = 1,
99 	CEPH_MDS_SESSION_OPENING = 2,
100 	CEPH_MDS_SESSION_OPEN = 3,
101 	CEPH_MDS_SESSION_HUNG = 4,
102 	CEPH_MDS_SESSION_CLOSING = 5,
103 	CEPH_MDS_SESSION_RESTARTING = 6,
104 	CEPH_MDS_SESSION_RECONNECTING = 7,
105 };
106 
107 struct ceph_mds_session {
108 	struct ceph_mds_client *s_mdsc;
109 	int               s_mds;
110 	int               s_state;
111 	unsigned long     s_ttl;      /* time until mds kills us */
112 	u64               s_seq;      /* incoming msg seq # */
113 	struct mutex      s_mutex;    /* serialize session messages */
114 
115 	struct ceph_connection s_con;
116 
117 	struct ceph_auth_handshake s_auth;
118 
119 	/* protected by s_gen_ttl_lock */
120 	spinlock_t        s_gen_ttl_lock;
121 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
122 	unsigned long     s_cap_ttl;  /* when session caps expire */
123 
124 	/* protected by s_cap_lock */
125 	spinlock_t        s_cap_lock;
126 	struct list_head  s_caps;     /* all caps issued by this session */
127 	int               s_nr_caps, s_trim_caps;
128 	int               s_num_cap_releases;
129 	struct list_head  s_cap_releases; /* waiting cap_release messages */
130 	struct list_head  s_cap_releases_done; /* ready to send */
131 	struct ceph_cap  *s_cap_iterator;
132 
133 	/* protected by mutex */
134 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
135 	struct list_head  s_cap_snaps_flushing;
136 	unsigned long     s_renew_requested; /* last time we sent a renew req */
137 	u64               s_renew_seq;
138 
139 	atomic_t          s_ref;
140 	struct list_head  s_waiting;  /* waiting requests */
141 	struct list_head  s_unsafe;   /* unsafe requests */
142 };
143 
144 /*
145  * modes of choosing which MDS to send a request to
146  */
147 enum {
148 	USE_ANY_MDS,
149 	USE_RANDOM_MDS,
150 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
151 };
152 
153 struct ceph_mds_request;
154 struct ceph_mds_client;
155 
156 /*
157  * request completion callback
158  */
159 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
160 					     struct ceph_mds_request *req);
161 
162 /*
163  * an in-flight mds request
164  */
165 struct ceph_mds_request {
166 	u64 r_tid;                   /* transaction id */
167 	struct rb_node r_node;
168 	struct ceph_mds_client *r_mdsc;
169 
170 	int r_op;                    /* mds op code */
171 
172 	/* operation on what? */
173 	struct inode *r_inode;              /* arg1 */
174 	struct dentry *r_dentry;            /* arg1 */
175 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
176 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
177 	char *r_path1, *r_path2;
178 	struct ceph_vino r_ino1, r_ino2;
179 
180 	struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
181 	struct inode *r_target_inode;       /* resulting inode */
182 
183 	struct mutex r_fill_mutex;
184 
185 	union ceph_mds_request_args r_args;
186 	int r_fmode;        /* file mode, if expecting cap */
187 	uid_t r_uid;
188 	gid_t r_gid;
189 
190 	/* for choosing which mds to send this request to */
191 	int r_direct_mode;
192 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
193 	bool r_direct_is_hash;  /* true if r_direct_hash is valid */
194 
195 	/* data payload is used for xattr ops */
196 	struct page **r_pages;
197 	int r_num_pages;
198 	int r_data_len;
199 
200 	/* what caps shall we drop? */
201 	int r_inode_drop, r_inode_unless;
202 	int r_dentry_drop, r_dentry_unless;
203 	int r_old_dentry_drop, r_old_dentry_unless;
204 	struct inode *r_old_inode;
205 	int r_old_inode_drop, r_old_inode_unless;
206 
207 	struct ceph_msg  *r_request;  /* original request */
208 	int r_request_release_offset;
209 	struct ceph_msg  *r_reply;
210 	struct ceph_mds_reply_info_parsed r_reply_info;
211 	int r_err;
212 	bool r_aborted;
213 
214 	unsigned long r_timeout;  /* optional.  jiffies */
215 	unsigned long r_started;  /* start time to measure timeout against */
216 	unsigned long r_request_started; /* start time for mds request only,
217 					    used to measure lease durations */
218 
219 	/* link unsafe requests to parent directory, for fsync */
220 	struct inode	*r_unsafe_dir;
221 	struct list_head r_unsafe_dir_item;
222 
223 	struct ceph_mds_session *r_session;
224 
225 	int               r_attempts;   /* resend attempts */
226 	int               r_num_fwd;    /* number of forward attempts */
227 	int               r_resend_mds; /* mds to resend to next, if any*/
228 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
229 
230 	struct kref       r_kref;
231 	struct list_head  r_wait;
232 	struct completion r_completion;
233 	struct completion r_safe_completion;
234 	ceph_mds_request_callback_t r_callback;
235 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
236 	bool		  r_got_unsafe, r_got_safe, r_got_result;
237 
238 	bool              r_did_prepopulate;
239 	u32               r_readdir_offset;
240 
241 	struct ceph_cap_reservation r_caps_reservation;
242 	int r_num_caps;
243 };
244 
245 /*
246  * mds client state
247  */
248 struct ceph_mds_client {
249 	struct ceph_fs_client  *fsc;
250 	struct mutex            mutex;         /* all nested structures */
251 
252 	struct ceph_mdsmap      *mdsmap;
253 	struct completion       safe_umount_waiters;
254 	wait_queue_head_t       session_close_wq;
255 	struct list_head        waiting_for_map;
256 
257 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
258 	int                     max_sessions;  /* len of s_mds_sessions */
259 	int                     stopping;      /* true if shutting down */
260 
261 	/*
262 	 * snap_rwsem will cover cap linkage into snaprealms, and
263 	 * realm snap contexts.  (later, we can do per-realm snap
264 	 * contexts locks..)  the empty list contains realms with no
265 	 * references (implying they contain no inodes with caps) that
266 	 * should be destroyed.
267 	 */
268 	struct rw_semaphore     snap_rwsem;
269 	struct rb_root          snap_realms;
270 	struct list_head        snap_empty;
271 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
272 
273 	u64                    last_tid;      /* most recent mds request */
274 	struct rb_root         request_tree;  /* pending mds requests */
275 	struct delayed_work    delayed_work;  /* delayed work */
276 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
277 	struct list_head cap_delay_list;   /* caps with delayed release */
278 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
279 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
280 	spinlock_t       snap_flush_lock;
281 
282 	u64               cap_flush_seq;
283 	struct list_head  cap_dirty;        /* inodes with dirty caps */
284 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
285 	int               num_cap_flushing; /* # caps we are flushing */
286 	spinlock_t        cap_dirty_lock;   /* protects above items */
287 	wait_queue_head_t cap_flushing_wq;
288 
289 	/*
290 	 * Cap reservations
291 	 *
292 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
293 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
294 	 * memory needed to successfully process an MDS response.  (If an MDS
295 	 * sends us cap information and we fail to process it, we will have
296 	 * problems due to the client and MDS being out of sync.)
297 	 *
298 	 * Reservations are 'owned' by a ceph_cap_reservation context.
299 	 */
300 	spinlock_t	caps_list_lock;
301 	struct		list_head caps_list; /* unused (reserved or
302 						unreserved) */
303 	int		caps_total_count;    /* total caps allocated */
304 	int		caps_use_count;      /* in use */
305 	int		caps_reserve_count;  /* unused, reserved */
306 	int		caps_avail_count;    /* unused, unreserved */
307 	int		caps_min_count;      /* keep at least this many
308 						(unreserved) */
309 	spinlock_t	  dentry_lru_lock;
310 	struct list_head  dentry_lru;
311 	int		  num_dentry;
312 };
313 
314 extern const char *ceph_mds_op_name(int op);
315 
316 extern struct ceph_mds_session *
317 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
318 
319 static inline struct ceph_mds_session *
320 ceph_get_mds_session(struct ceph_mds_session *s)
321 {
322 	atomic_inc(&s->s_ref);
323 	return s;
324 }
325 
326 extern void ceph_put_mds_session(struct ceph_mds_session *s);
327 
328 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
329 			     struct ceph_msg *msg, int mds);
330 
331 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
332 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
333 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
334 
335 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
336 
337 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
338 				    struct inode *inode,
339 				    struct dentry *dn);
340 
341 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
342 
343 extern struct ceph_mds_request *
344 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
345 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
346 				     struct ceph_mds_request *req);
347 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
348 				struct inode *dir,
349 				struct ceph_mds_request *req);
350 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
351 {
352 	kref_get(&req->r_kref);
353 }
354 extern void ceph_mdsc_release_request(struct kref *kref);
355 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
356 {
357 	kref_put(&req->r_kref, ceph_mdsc_release_request);
358 }
359 
360 extern int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
361 				 struct ceph_mds_session *session);
362 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
363 				   struct ceph_mds_session *session);
364 
365 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
366 
367 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
368 				  int stop_on_nosnap);
369 
370 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
371 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
372 				     struct inode *inode,
373 				     struct dentry *dentry, char action,
374 				     u32 seq);
375 
376 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
377 				 struct ceph_msg *msg);
378 
379 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
380 					  struct ceph_mds_session *session);
381 
382 #endif
383