xref: /openbmc/linux/fs/ceph/mds_client.h (revision f5b06569)
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 	char *pool_ns_data;
49 };
50 
51 struct ceph_mds_reply_dir_entry {
52 	char                          *name;
53 	u32                           name_len;
54 	struct ceph_mds_reply_lease   *lease;
55 	struct ceph_mds_reply_info_in inode;
56 	loff_t			      offset;
57 };
58 
59 /*
60  * parsed info about an mds reply, including information about
61  * either: 1) the target inode and/or its parent directory and dentry,
62  * and directory contents (for readdir results), or
63  * 2) the file range lock info (for fcntl F_GETLK results).
64  */
65 struct ceph_mds_reply_info_parsed {
66 	struct ceph_mds_reply_head    *head;
67 
68 	/* trace */
69 	struct ceph_mds_reply_info_in diri, targeti;
70 	struct ceph_mds_reply_dirfrag *dirfrag;
71 	char                          *dname;
72 	u32                           dname_len;
73 	struct ceph_mds_reply_lease   *dlease;
74 
75 	/* extra */
76 	union {
77 		/* for fcntl F_GETLK results */
78 		struct ceph_filelock *filelock_reply;
79 
80 		/* for readdir results */
81 		struct {
82 			struct ceph_mds_reply_dirfrag *dir_dir;
83 			size_t			      dir_buf_size;
84 			int                           dir_nr;
85 			bool			      dir_complete;
86 			bool			      dir_end;
87 			bool			      hash_order;
88 			struct ceph_mds_reply_dir_entry  *dir_entries;
89 		};
90 
91 		/* for create results */
92 		struct {
93 			bool has_create_ino;
94 			u64 ino;
95 		};
96 	};
97 
98 	/* encoded blob describing snapshot contexts for certain
99 	   operations (e.g., open) */
100 	void *snapblob;
101 	int snapblob_len;
102 };
103 
104 
105 /*
106  * cap releases are batched and sent to the MDS en masse.
107  */
108 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE -			\
109 				sizeof(struct ceph_mds_cap_release)) /	\
110 			       sizeof(struct ceph_mds_cap_item))
111 
112 
113 /*
114  * state associated with each MDS<->client session
115  */
116 enum {
117 	CEPH_MDS_SESSION_NEW = 1,
118 	CEPH_MDS_SESSION_OPENING = 2,
119 	CEPH_MDS_SESSION_OPEN = 3,
120 	CEPH_MDS_SESSION_HUNG = 4,
121 	CEPH_MDS_SESSION_CLOSING = 5,
122 	CEPH_MDS_SESSION_RESTARTING = 6,
123 	CEPH_MDS_SESSION_RECONNECTING = 7,
124 };
125 
126 struct ceph_mds_session {
127 	struct ceph_mds_client *s_mdsc;
128 	int               s_mds;
129 	int               s_state;
130 	unsigned long     s_ttl;      /* time until mds kills us */
131 	u64               s_seq;      /* incoming msg seq # */
132 	struct mutex      s_mutex;    /* serialize session messages */
133 
134 	struct ceph_connection s_con;
135 
136 	struct ceph_auth_handshake s_auth;
137 
138 	/* protected by s_gen_ttl_lock */
139 	spinlock_t        s_gen_ttl_lock;
140 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
141 	unsigned long     s_cap_ttl;  /* when session caps expire */
142 
143 	/* protected by s_cap_lock */
144 	spinlock_t        s_cap_lock;
145 	struct list_head  s_caps;     /* all caps issued by this session */
146 	int               s_nr_caps, s_trim_caps;
147 	int               s_num_cap_releases;
148 	int		  s_cap_reconnect;
149 	int		  s_readonly;
150 	struct list_head  s_cap_releases; /* waiting cap_release messages */
151 	struct ceph_cap  *s_cap_iterator;
152 
153 	/* protected by mutex */
154 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
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 	int perm;
279 	s64 pool;
280 	size_t pool_ns_len;
281 	char pool_ns[];
282 };
283 
284 /*
285  * mds client state
286  */
287 struct ceph_mds_client {
288 	struct ceph_fs_client  *fsc;
289 	struct mutex            mutex;         /* all nested structures */
290 
291 	struct ceph_mdsmap      *mdsmap;
292 	struct completion       safe_umount_waiters;
293 	wait_queue_head_t       session_close_wq;
294 	struct list_head        waiting_for_map;
295 	int 			mdsmap_err;
296 
297 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
298 	atomic_t		num_sessions;
299 	int                     max_sessions;  /* len of s_mds_sessions */
300 	int                     stopping;      /* true if shutting down */
301 
302 	/*
303 	 * snap_rwsem will cover cap linkage into snaprealms, and
304 	 * realm snap contexts.  (later, we can do per-realm snap
305 	 * contexts locks..)  the empty list contains realms with no
306 	 * references (implying they contain no inodes with caps) that
307 	 * should be destroyed.
308 	 */
309 	u64			last_snap_seq;
310 	struct rw_semaphore     snap_rwsem;
311 	struct rb_root          snap_realms;
312 	struct list_head        snap_empty;
313 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
314 
315 	u64                    last_tid;      /* most recent mds request */
316 	u64                    oldest_tid;    /* oldest incomplete mds request,
317 						 excluding setfilelock requests */
318 	struct rb_root         request_tree;  /* pending mds requests */
319 	struct delayed_work    delayed_work;  /* delayed work */
320 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
321 	struct list_head cap_delay_list;   /* caps with delayed release */
322 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
323 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
324 	spinlock_t       snap_flush_lock;
325 
326 	u64               last_cap_flush_tid;
327 	struct list_head  cap_flush_list;
328 	struct list_head  cap_dirty;        /* inodes with dirty caps */
329 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
330 	int               num_cap_flushing; /* # caps we are flushing */
331 	spinlock_t        cap_dirty_lock;   /* protects above items */
332 	wait_queue_head_t cap_flushing_wq;
333 
334 	/*
335 	 * Cap reservations
336 	 *
337 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
338 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
339 	 * memory needed to successfully process an MDS response.  (If an MDS
340 	 * sends us cap information and we fail to process it, we will have
341 	 * problems due to the client and MDS being out of sync.)
342 	 *
343 	 * Reservations are 'owned' by a ceph_cap_reservation context.
344 	 */
345 	spinlock_t	caps_list_lock;
346 	struct		list_head caps_list; /* unused (reserved or
347 						unreserved) */
348 	int		caps_total_count;    /* total caps allocated */
349 	int		caps_use_count;      /* in use */
350 	int		caps_reserve_count;  /* unused, reserved */
351 	int		caps_avail_count;    /* unused, unreserved */
352 	int		caps_min_count;      /* keep at least this many
353 						(unreserved) */
354 	spinlock_t	  dentry_lru_lock;
355 	struct list_head  dentry_lru;
356 	int		  num_dentry;
357 
358 	struct rw_semaphore     pool_perm_rwsem;
359 	struct rb_root		pool_perm_tree;
360 };
361 
362 extern const char *ceph_mds_op_name(int op);
363 
364 extern struct ceph_mds_session *
365 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
366 
367 static inline struct ceph_mds_session *
368 ceph_get_mds_session(struct ceph_mds_session *s)
369 {
370 	atomic_inc(&s->s_ref);
371 	return s;
372 }
373 
374 extern const char *ceph_session_state_name(int s);
375 
376 extern void ceph_put_mds_session(struct ceph_mds_session *s);
377 
378 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
379 			     struct ceph_msg *msg, int mds);
380 
381 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
382 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
383 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
384 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
385 
386 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
387 
388 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
389 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
390 					   struct inode *dir);
391 extern struct ceph_mds_request *
392 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
393 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
394 				     struct ceph_mds_request *req);
395 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
396 				struct inode *dir,
397 				struct ceph_mds_request *req);
398 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
399 {
400 	kref_get(&req->r_kref);
401 }
402 extern void ceph_mdsc_release_request(struct kref *kref);
403 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
404 {
405 	kref_put(&req->r_kref, ceph_mdsc_release_request);
406 }
407 
408 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
409 				   struct ceph_mds_session *session);
410 
411 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
412 
413 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
414 				  int stop_on_nosnap);
415 
416 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
417 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
418 				     struct inode *inode,
419 				     struct dentry *dentry, char action,
420 				     u32 seq);
421 
422 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
423 				    struct ceph_msg *msg);
424 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
425 				   struct ceph_msg *msg);
426 
427 extern struct ceph_mds_session *
428 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
429 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
430 					  struct ceph_mds_session *session);
431 
432 #endif
433