xref: /openbmc/linux/fs/ceph/mds_client.h (revision 4da722ca)
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 #include <linux/refcount.h>
11 
12 #include <linux/ceph/types.h>
13 #include <linux/ceph/messenger.h>
14 #include <linux/ceph/mdsmap.h>
15 #include <linux/ceph/auth.h>
16 
17 /*
18  * Some lock dependencies:
19  *
20  * session->s_mutex
21  *         mdsc->mutex
22  *
23  *         mdsc->snap_rwsem
24  *
25  *         ci->i_ceph_lock
26  *                 mdsc->snap_flush_lock
27  *                 mdsc->cap_delay_lock
28  *
29  */
30 
31 struct ceph_fs_client;
32 struct ceph_cap;
33 
34 /*
35  * parsed info about a single inode.  pointers are into the encoded
36  * on-wire structures within the mds reply message payload.
37  */
38 struct ceph_mds_reply_info_in {
39 	struct ceph_mds_reply_inode *in;
40 	struct ceph_dir_layout dir_layout;
41 	u32 symlink_len;
42 	char *symlink;
43 	u32 xattr_len;
44 	char *xattr_data;
45 	u64 inline_version;
46 	u32 inline_len;
47 	char *inline_data;
48 	u32 pool_ns_len;
49 	char *pool_ns_data;
50 };
51 
52 struct ceph_mds_reply_dir_entry {
53 	char                          *name;
54 	u32                           name_len;
55 	struct ceph_mds_reply_lease   *lease;
56 	struct ceph_mds_reply_info_in inode;
57 	loff_t			      offset;
58 };
59 
60 /*
61  * parsed info about an mds reply, including information about
62  * either: 1) the target inode and/or its parent directory and dentry,
63  * and directory contents (for readdir results), or
64  * 2) the file range lock info (for fcntl F_GETLK results).
65  */
66 struct ceph_mds_reply_info_parsed {
67 	struct ceph_mds_reply_head    *head;
68 
69 	/* trace */
70 	struct ceph_mds_reply_info_in diri, targeti;
71 	struct ceph_mds_reply_dirfrag *dirfrag;
72 	char                          *dname;
73 	u32                           dname_len;
74 	struct ceph_mds_reply_lease   *dlease;
75 
76 	/* extra */
77 	union {
78 		/* for fcntl F_GETLK results */
79 		struct ceph_filelock *filelock_reply;
80 
81 		/* for readdir results */
82 		struct {
83 			struct ceph_mds_reply_dirfrag *dir_dir;
84 			size_t			      dir_buf_size;
85 			int                           dir_nr;
86 			bool			      dir_end;
87 			bool			      dir_complete;
88 			bool			      hash_order;
89 			bool			      offset_hash;
90 			struct ceph_mds_reply_dir_entry  *dir_entries;
91 		};
92 
93 		/* for create results */
94 		struct {
95 			bool has_create_ino;
96 			u64 ino;
97 		};
98 	};
99 
100 	/* encoded blob describing snapshot contexts for certain
101 	   operations (e.g., open) */
102 	void *snapblob;
103 	int snapblob_len;
104 };
105 
106 
107 /*
108  * cap releases are batched and sent to the MDS en masse.
109  *
110  * Account for per-message overhead of mds_cap_release header
111  * and __le32 for osd epoch barrier trailing field.
112  */
113 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
114 				sizeof(struct ceph_mds_cap_release)) /	\
115 			        sizeof(struct ceph_mds_cap_item))
116 
117 
118 /*
119  * state associated with each MDS<->client session
120  */
121 enum {
122 	CEPH_MDS_SESSION_NEW = 1,
123 	CEPH_MDS_SESSION_OPENING = 2,
124 	CEPH_MDS_SESSION_OPEN = 3,
125 	CEPH_MDS_SESSION_HUNG = 4,
126 	CEPH_MDS_SESSION_CLOSING = 5,
127 	CEPH_MDS_SESSION_RESTARTING = 6,
128 	CEPH_MDS_SESSION_RECONNECTING = 7,
129 	CEPH_MDS_SESSION_REJECTED = 8,
130 };
131 
132 struct ceph_mds_session {
133 	struct ceph_mds_client *s_mdsc;
134 	int               s_mds;
135 	int               s_state;
136 	unsigned long     s_ttl;      /* time until mds kills us */
137 	u64               s_seq;      /* incoming msg seq # */
138 	struct mutex      s_mutex;    /* serialize session messages */
139 
140 	struct ceph_connection s_con;
141 
142 	struct ceph_auth_handshake s_auth;
143 
144 	/* protected by s_gen_ttl_lock */
145 	spinlock_t        s_gen_ttl_lock;
146 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
147 	unsigned long     s_cap_ttl;  /* when session caps expire */
148 
149 	/* protected by s_cap_lock */
150 	spinlock_t        s_cap_lock;
151 	struct list_head  s_caps;     /* all caps issued by this session */
152 	int               s_nr_caps, s_trim_caps;
153 	int               s_num_cap_releases;
154 	int		  s_cap_reconnect;
155 	int		  s_readonly;
156 	struct list_head  s_cap_releases; /* waiting cap_release messages */
157 	struct ceph_cap  *s_cap_iterator;
158 
159 	/* protected by mutex */
160 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
161 	unsigned long     s_renew_requested; /* last time we sent a renew req */
162 	u64               s_renew_seq;
163 
164 	refcount_t          s_ref;
165 	struct list_head  s_waiting;  /* waiting requests */
166 	struct list_head  s_unsafe;   /* unsafe requests */
167 };
168 
169 /*
170  * modes of choosing which MDS to send a request to
171  */
172 enum {
173 	USE_ANY_MDS,
174 	USE_RANDOM_MDS,
175 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
176 };
177 
178 struct ceph_mds_request;
179 struct ceph_mds_client;
180 
181 /*
182  * request completion callback
183  */
184 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
185 					     struct ceph_mds_request *req);
186 /*
187  * wait for request completion callback
188  */
189 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
190 						 struct ceph_mds_request *req);
191 
192 /*
193  * an in-flight mds request
194  */
195 struct ceph_mds_request {
196 	u64 r_tid;                   /* transaction id */
197 	struct rb_node r_node;
198 	struct ceph_mds_client *r_mdsc;
199 
200 	int r_op;                    /* mds op code */
201 
202 	/* operation on what? */
203 	struct inode *r_inode;              /* arg1 */
204 	struct dentry *r_dentry;            /* arg1 */
205 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
206 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
207 	char *r_path1, *r_path2;
208 	struct ceph_vino r_ino1, r_ino2;
209 
210 	struct inode *r_parent;		    /* parent dir inode */
211 	struct inode *r_target_inode;       /* resulting inode */
212 
213 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
214 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
215 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
216 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
217 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
218 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
219 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
220 	unsigned long	r_req_flags;
221 
222 	struct mutex r_fill_mutex;
223 
224 	union ceph_mds_request_args r_args;
225 	int r_fmode;        /* file mode, if expecting cap */
226 	kuid_t r_uid;
227 	kgid_t r_gid;
228 	struct timespec r_stamp;
229 
230 	/* for choosing which mds to send this request to */
231 	int r_direct_mode;
232 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
233 
234 	/* data payload is used for xattr ops */
235 	struct ceph_pagelist *r_pagelist;
236 
237 	/* what caps shall we drop? */
238 	int r_inode_drop, r_inode_unless;
239 	int r_dentry_drop, r_dentry_unless;
240 	int r_old_dentry_drop, r_old_dentry_unless;
241 	struct inode *r_old_inode;
242 	int r_old_inode_drop, r_old_inode_unless;
243 
244 	struct ceph_msg  *r_request;  /* original request */
245 	int r_request_release_offset;
246 	struct ceph_msg  *r_reply;
247 	struct ceph_mds_reply_info_parsed r_reply_info;
248 	struct page *r_locked_page;
249 	int r_err;
250 
251 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
252 	unsigned long r_started;  /* start time to measure timeout against */
253 	unsigned long r_request_started; /* start time for mds request only,
254 					    used to measure lease durations */
255 
256 	/* link unsafe requests to parent directory, for fsync */
257 	struct inode	*r_unsafe_dir;
258 	struct list_head r_unsafe_dir_item;
259 
260 	/* unsafe requests that modify the target inode */
261 	struct list_head r_unsafe_target_item;
262 
263 	struct ceph_mds_session *r_session;
264 
265 	int               r_attempts;   /* resend attempts */
266 	int               r_num_fwd;    /* number of forward attempts */
267 	int               r_resend_mds; /* mds to resend to next, if any*/
268 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
269 
270 	struct kref       r_kref;
271 	struct list_head  r_wait;
272 	struct completion r_completion;
273 	struct completion r_safe_completion;
274 	ceph_mds_request_callback_t r_callback;
275 	ceph_mds_request_wait_callback_t r_wait_for_completion;
276 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
277 
278 	long long	  r_dir_release_cnt;
279 	long long	  r_dir_ordered_cnt;
280 	int		  r_readdir_cache_idx;
281 	u32               r_readdir_offset;
282 
283 	struct ceph_cap_reservation r_caps_reservation;
284 	int r_num_caps;
285 };
286 
287 struct ceph_pool_perm {
288 	struct rb_node node;
289 	int perm;
290 	s64 pool;
291 	size_t pool_ns_len;
292 	char pool_ns[];
293 };
294 
295 /*
296  * mds client state
297  */
298 struct ceph_mds_client {
299 	struct ceph_fs_client  *fsc;
300 	struct mutex            mutex;         /* all nested structures */
301 
302 	struct ceph_mdsmap      *mdsmap;
303 	struct completion       safe_umount_waiters;
304 	wait_queue_head_t       session_close_wq;
305 	struct list_head        waiting_for_map;
306 	int 			mdsmap_err;
307 
308 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
309 	atomic_t		num_sessions;
310 	int                     max_sessions;  /* len of s_mds_sessions */
311 	int                     stopping;      /* true if shutting down */
312 
313 	/*
314 	 * snap_rwsem will cover cap linkage into snaprealms, and
315 	 * realm snap contexts.  (later, we can do per-realm snap
316 	 * contexts locks..)  the empty list contains realms with no
317 	 * references (implying they contain no inodes with caps) that
318 	 * should be destroyed.
319 	 */
320 	u64			last_snap_seq;
321 	struct rw_semaphore     snap_rwsem;
322 	struct rb_root          snap_realms;
323 	struct list_head        snap_empty;
324 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
325 
326 	u64                    last_tid;      /* most recent mds request */
327 	u64                    oldest_tid;    /* oldest incomplete mds request,
328 						 excluding setfilelock requests */
329 	struct rb_root         request_tree;  /* pending mds requests */
330 	struct delayed_work    delayed_work;  /* delayed work */
331 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
332 	struct list_head cap_delay_list;   /* caps with delayed release */
333 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
334 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
335 	spinlock_t       snap_flush_lock;
336 
337 	u64               last_cap_flush_tid;
338 	struct list_head  cap_flush_list;
339 	struct list_head  cap_dirty;        /* inodes with dirty caps */
340 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
341 	int               num_cap_flushing; /* # caps we are flushing */
342 	spinlock_t        cap_dirty_lock;   /* protects above items */
343 	wait_queue_head_t cap_flushing_wq;
344 
345 	/*
346 	 * Cap reservations
347 	 *
348 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
349 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
350 	 * memory needed to successfully process an MDS response.  (If an MDS
351 	 * sends us cap information and we fail to process it, we will have
352 	 * problems due to the client and MDS being out of sync.)
353 	 *
354 	 * Reservations are 'owned' by a ceph_cap_reservation context.
355 	 */
356 	spinlock_t	caps_list_lock;
357 	struct		list_head caps_list; /* unused (reserved or
358 						unreserved) */
359 	int		caps_total_count;    /* total caps allocated */
360 	int		caps_use_count;      /* in use */
361 	int		caps_reserve_count;  /* unused, reserved */
362 	int		caps_avail_count;    /* unused, unreserved */
363 	int		caps_min_count;      /* keep at least this many
364 						(unreserved) */
365 	spinlock_t	  dentry_lru_lock;
366 	struct list_head  dentry_lru;
367 	int		  num_dentry;
368 
369 	struct rw_semaphore     pool_perm_rwsem;
370 	struct rb_root		pool_perm_tree;
371 };
372 
373 extern const char *ceph_mds_op_name(int op);
374 
375 extern struct ceph_mds_session *
376 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
377 
378 static inline struct ceph_mds_session *
379 ceph_get_mds_session(struct ceph_mds_session *s)
380 {
381 	refcount_inc(&s->s_ref);
382 	return s;
383 }
384 
385 extern const char *ceph_session_state_name(int s);
386 
387 extern void ceph_put_mds_session(struct ceph_mds_session *s);
388 
389 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
390 			     struct ceph_msg *msg, int mds);
391 
392 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
393 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
394 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
395 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
396 
397 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
398 
399 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
400 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
401 					   struct inode *dir);
402 extern struct ceph_mds_request *
403 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
404 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
405 				     struct ceph_mds_request *req);
406 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
407 				struct inode *dir,
408 				struct ceph_mds_request *req);
409 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
410 {
411 	kref_get(&req->r_kref);
412 }
413 extern void ceph_mdsc_release_request(struct kref *kref);
414 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
415 {
416 	kref_put(&req->r_kref, ceph_mdsc_release_request);
417 }
418 
419 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
420 				   struct ceph_mds_session *session);
421 
422 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
423 
424 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
425 				  int stop_on_nosnap);
426 
427 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
428 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
429 				     struct inode *inode,
430 				     struct dentry *dentry, char action,
431 				     u32 seq);
432 
433 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
434 				    struct ceph_msg *msg);
435 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
436 				   struct ceph_msg *msg);
437 
438 extern struct ceph_mds_session *
439 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
440 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
441 					  struct ceph_mds_session *session);
442 
443 #endif
444