xref: /openbmc/linux/fs/ceph/mds_client.h (revision 3d9b448b)
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 };
48 
49 /*
50  * parsed info about an mds reply, including information about
51  * either: 1) the target inode and/or its parent directory and dentry,
52  * and directory contents (for readdir results), or
53  * 2) the file range lock info (for fcntl F_GETLK results).
54  */
55 struct ceph_mds_reply_info_parsed {
56 	struct ceph_mds_reply_head    *head;
57 
58 	/* trace */
59 	struct ceph_mds_reply_info_in diri, targeti;
60 	struct ceph_mds_reply_dirfrag *dirfrag;
61 	char                          *dname;
62 	u32                           dname_len;
63 	struct ceph_mds_reply_lease   *dlease;
64 
65 	/* extra */
66 	union {
67 		/* for fcntl F_GETLK results */
68 		struct ceph_filelock *filelock_reply;
69 
70 		/* for readdir results */
71 		struct {
72 			struct ceph_mds_reply_dirfrag *dir_dir;
73 			size_t			      dir_buf_size;
74 			int                           dir_nr;
75 			char                          **dir_dname;
76 			u32                           *dir_dname_len;
77 			struct ceph_mds_reply_lease   **dir_dlease;
78 			struct ceph_mds_reply_info_in *dir_in;
79 			u8                            dir_complete, dir_end;
80 		};
81 
82 		/* for create results */
83 		struct {
84 			bool has_create_ino;
85 			u64 ino;
86 		};
87 	};
88 
89 	/* encoded blob describing snapshot contexts for certain
90 	   operations (e.g., open) */
91 	void *snapblob;
92 	int snapblob_len;
93 };
94 
95 
96 /*
97  * cap releases are batched and sent to the MDS en masse.
98  */
99 #define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE -			\
100 				sizeof(struct ceph_mds_cap_release)) /	\
101 			       sizeof(struct ceph_mds_cap_item))
102 
103 
104 /*
105  * state associated with each MDS<->client session
106  */
107 enum {
108 	CEPH_MDS_SESSION_NEW = 1,
109 	CEPH_MDS_SESSION_OPENING = 2,
110 	CEPH_MDS_SESSION_OPEN = 3,
111 	CEPH_MDS_SESSION_HUNG = 4,
112 	CEPH_MDS_SESSION_CLOSING = 5,
113 	CEPH_MDS_SESSION_RESTARTING = 6,
114 	CEPH_MDS_SESSION_RECONNECTING = 7,
115 };
116 
117 struct ceph_mds_session {
118 	struct ceph_mds_client *s_mdsc;
119 	int               s_mds;
120 	int               s_state;
121 	unsigned long     s_ttl;      /* time until mds kills us */
122 	u64               s_seq;      /* incoming msg seq # */
123 	struct mutex      s_mutex;    /* serialize session messages */
124 
125 	struct ceph_connection s_con;
126 
127 	struct ceph_auth_handshake s_auth;
128 
129 	/* protected by s_gen_ttl_lock */
130 	spinlock_t        s_gen_ttl_lock;
131 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
132 	unsigned long     s_cap_ttl;  /* when session caps expire */
133 
134 	/* protected by s_cap_lock */
135 	spinlock_t        s_cap_lock;
136 	struct list_head  s_caps;     /* all caps issued by this session */
137 	int               s_nr_caps, s_trim_caps;
138 	int               s_num_cap_releases;
139 	int		  s_cap_reconnect;
140 	int		  s_readonly;
141 	struct list_head  s_cap_releases; /* waiting cap_release messages */
142 	struct list_head  s_cap_releases_done; /* ready to send */
143 	struct ceph_cap  *s_cap_iterator;
144 
145 	/* protected by mutex */
146 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
147 	struct list_head  s_cap_snaps_flushing;
148 	unsigned long     s_renew_requested; /* last time we sent a renew req */
149 	u64               s_renew_seq;
150 
151 	atomic_t          s_ref;
152 	struct list_head  s_waiting;  /* waiting requests */
153 	struct list_head  s_unsafe;   /* unsafe requests */
154 };
155 
156 /*
157  * modes of choosing which MDS to send a request to
158  */
159 enum {
160 	USE_ANY_MDS,
161 	USE_RANDOM_MDS,
162 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
163 };
164 
165 struct ceph_mds_request;
166 struct ceph_mds_client;
167 
168 /*
169  * request completion callback
170  */
171 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
172 					     struct ceph_mds_request *req);
173 /*
174  * wait for request completion callback
175  */
176 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
177 						 struct ceph_mds_request *req);
178 
179 /*
180  * an in-flight mds request
181  */
182 struct ceph_mds_request {
183 	u64 r_tid;                   /* transaction id */
184 	struct rb_node r_node;
185 	struct ceph_mds_client *r_mdsc;
186 
187 	int r_op;                    /* mds op code */
188 
189 	/* operation on what? */
190 	struct inode *r_inode;              /* arg1 */
191 	struct dentry *r_dentry;            /* arg1 */
192 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
193 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
194 	char *r_path1, *r_path2;
195 	struct ceph_vino r_ino1, r_ino2;
196 
197 	struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
198 	struct inode *r_target_inode;       /* resulting inode */
199 
200 	struct mutex r_fill_mutex;
201 
202 	union ceph_mds_request_args r_args;
203 	int r_fmode;        /* file mode, if expecting cap */
204 	kuid_t r_uid;
205 	kgid_t r_gid;
206 	struct timespec r_stamp;
207 
208 	/* for choosing which mds to send this request to */
209 	int r_direct_mode;
210 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
211 	bool r_direct_is_hash;  /* true if r_direct_hash is valid */
212 
213 	/* data payload is used for xattr ops */
214 	struct ceph_pagelist *r_pagelist;
215 
216 	/* what caps shall we drop? */
217 	int r_inode_drop, r_inode_unless;
218 	int r_dentry_drop, r_dentry_unless;
219 	int r_old_dentry_drop, r_old_dentry_unless;
220 	struct inode *r_old_inode;
221 	int r_old_inode_drop, r_old_inode_unless;
222 
223 	struct ceph_msg  *r_request;  /* original request */
224 	int r_request_release_offset;
225 	struct ceph_msg  *r_reply;
226 	struct ceph_mds_reply_info_parsed r_reply_info;
227 	struct page *r_locked_page;
228 	int r_err;
229 	bool r_aborted;
230 
231 	unsigned long r_timeout;  /* optional.  jiffies */
232 	unsigned long r_started;  /* start time to measure timeout against */
233 	unsigned long r_request_started; /* start time for mds request only,
234 					    used to measure lease durations */
235 
236 	/* link unsafe requests to parent directory, for fsync */
237 	struct inode	*r_unsafe_dir;
238 	struct list_head r_unsafe_dir_item;
239 
240 	struct ceph_mds_session *r_session;
241 
242 	int               r_attempts;   /* resend attempts */
243 	int               r_num_fwd;    /* number of forward attempts */
244 	int               r_resend_mds; /* mds to resend to next, if any*/
245 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
246 
247 	struct kref       r_kref;
248 	struct list_head  r_wait;
249 	struct completion r_completion;
250 	struct completion r_safe_completion;
251 	ceph_mds_request_callback_t r_callback;
252 	ceph_mds_request_wait_callback_t r_wait_for_completion;
253 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
254 	bool		  r_got_unsafe, r_got_safe, r_got_result;
255 
256 	bool              r_did_prepopulate;
257 	u32               r_readdir_offset;
258 
259 	struct ceph_cap_reservation r_caps_reservation;
260 	int r_num_caps;
261 };
262 
263 /*
264  * mds client state
265  */
266 struct ceph_mds_client {
267 	struct ceph_fs_client  *fsc;
268 	struct mutex            mutex;         /* all nested structures */
269 
270 	struct ceph_mdsmap      *mdsmap;
271 	struct completion       safe_umount_waiters;
272 	wait_queue_head_t       session_close_wq;
273 	struct list_head        waiting_for_map;
274 
275 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
276 	atomic_t		num_sessions;
277 	int                     max_sessions;  /* len of s_mds_sessions */
278 	int                     stopping;      /* true if shutting down */
279 
280 	/*
281 	 * snap_rwsem will cover cap linkage into snaprealms, and
282 	 * realm snap contexts.  (later, we can do per-realm snap
283 	 * contexts locks..)  the empty list contains realms with no
284 	 * references (implying they contain no inodes with caps) that
285 	 * should be destroyed.
286 	 */
287 	struct rw_semaphore     snap_rwsem;
288 	struct rb_root          snap_realms;
289 	struct list_head        snap_empty;
290 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
291 
292 	u64                    last_tid;      /* most recent mds request */
293 	struct rb_root         request_tree;  /* pending mds requests */
294 	struct delayed_work    delayed_work;  /* delayed work */
295 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
296 	struct list_head cap_delay_list;   /* caps with delayed release */
297 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
298 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
299 	spinlock_t       snap_flush_lock;
300 
301 	u64               cap_flush_seq;
302 	struct list_head  cap_dirty;        /* inodes with dirty caps */
303 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
304 	int               num_cap_flushing; /* # caps we are flushing */
305 	spinlock_t        cap_dirty_lock;   /* protects above items */
306 	wait_queue_head_t cap_flushing_wq;
307 
308 	/*
309 	 * Cap reservations
310 	 *
311 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
312 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
313 	 * memory needed to successfully process an MDS response.  (If an MDS
314 	 * sends us cap information and we fail to process it, we will have
315 	 * problems due to the client and MDS being out of sync.)
316 	 *
317 	 * Reservations are 'owned' by a ceph_cap_reservation context.
318 	 */
319 	spinlock_t	caps_list_lock;
320 	struct		list_head caps_list; /* unused (reserved or
321 						unreserved) */
322 	int		caps_total_count;    /* total caps allocated */
323 	int		caps_use_count;      /* in use */
324 	int		caps_reserve_count;  /* unused, reserved */
325 	int		caps_avail_count;    /* unused, unreserved */
326 	int		caps_min_count;      /* keep at least this many
327 						(unreserved) */
328 	spinlock_t	  dentry_lru_lock;
329 	struct list_head  dentry_lru;
330 	int		  num_dentry;
331 };
332 
333 extern const char *ceph_mds_op_name(int op);
334 
335 extern struct ceph_mds_session *
336 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
337 
338 static inline struct ceph_mds_session *
339 ceph_get_mds_session(struct ceph_mds_session *s)
340 {
341 	atomic_inc(&s->s_ref);
342 	return s;
343 }
344 
345 extern const char *ceph_session_state_name(int s);
346 
347 extern void ceph_put_mds_session(struct ceph_mds_session *s);
348 
349 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
350 			     struct ceph_msg *msg, int mds);
351 
352 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
353 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
354 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
355 
356 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
357 
358 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
359 				    struct inode *inode,
360 				    struct dentry *dn);
361 
362 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
363 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
364 					   struct inode *dir);
365 extern struct ceph_mds_request *
366 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
367 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
368 				     struct ceph_mds_request *req);
369 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
370 				struct inode *dir,
371 				struct ceph_mds_request *req);
372 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
373 {
374 	kref_get(&req->r_kref);
375 }
376 extern void ceph_mdsc_release_request(struct kref *kref);
377 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
378 {
379 	kref_put(&req->r_kref, ceph_mdsc_release_request);
380 }
381 
382 extern int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
383 				 struct ceph_mds_session *session);
384 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
385 				   struct ceph_mds_session *session);
386 
387 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
388 
389 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
390 				  int stop_on_nosnap);
391 
392 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
393 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
394 				     struct inode *inode,
395 				     struct dentry *dentry, char action,
396 				     u32 seq);
397 
398 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
399 				 struct ceph_msg *msg);
400 
401 extern struct ceph_mds_session *
402 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
403 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
404 					  struct ceph_mds_session *session);
405 
406 #endif
407