xref: /openbmc/linux/fs/ceph/mds_client.h (revision 4f6cce39)
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 	CEPH_MDS_SESSION_REJECTED = 8,
125 };
126 
127 struct ceph_mds_session {
128 	struct ceph_mds_client *s_mdsc;
129 	int               s_mds;
130 	int               s_state;
131 	unsigned long     s_ttl;      /* time until mds kills us */
132 	u64               s_seq;      /* incoming msg seq # */
133 	struct mutex      s_mutex;    /* serialize session messages */
134 
135 	struct ceph_connection s_con;
136 
137 	struct ceph_auth_handshake s_auth;
138 
139 	/* protected by s_gen_ttl_lock */
140 	spinlock_t        s_gen_ttl_lock;
141 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
142 	unsigned long     s_cap_ttl;  /* when session caps expire */
143 
144 	/* protected by s_cap_lock */
145 	spinlock_t        s_cap_lock;
146 	struct list_head  s_caps;     /* all caps issued by this session */
147 	int               s_nr_caps, s_trim_caps;
148 	int               s_num_cap_releases;
149 	int		  s_cap_reconnect;
150 	int		  s_readonly;
151 	struct list_head  s_cap_releases; /* waiting cap_release messages */
152 	struct ceph_cap  *s_cap_iterator;
153 
154 	/* protected by mutex */
155 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
156 	unsigned long     s_renew_requested; /* last time we sent a renew req */
157 	u64               s_renew_seq;
158 
159 	atomic_t          s_ref;
160 	struct list_head  s_waiting;  /* waiting requests */
161 	struct list_head  s_unsafe;   /* unsafe requests */
162 };
163 
164 /*
165  * modes of choosing which MDS to send a request to
166  */
167 enum {
168 	USE_ANY_MDS,
169 	USE_RANDOM_MDS,
170 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
171 };
172 
173 struct ceph_mds_request;
174 struct ceph_mds_client;
175 
176 /*
177  * request completion callback
178  */
179 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
180 					     struct ceph_mds_request *req);
181 /*
182  * wait for request completion callback
183  */
184 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
185 						 struct ceph_mds_request *req);
186 
187 /*
188  * an in-flight mds request
189  */
190 struct ceph_mds_request {
191 	u64 r_tid;                   /* transaction id */
192 	struct rb_node r_node;
193 	struct ceph_mds_client *r_mdsc;
194 
195 	int r_op;                    /* mds op code */
196 
197 	/* operation on what? */
198 	struct inode *r_inode;              /* arg1 */
199 	struct dentry *r_dentry;            /* arg1 */
200 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
201 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
202 	char *r_path1, *r_path2;
203 	struct ceph_vino r_ino1, r_ino2;
204 
205 	struct inode *r_parent;		    /* parent dir inode */
206 	struct inode *r_target_inode;       /* resulting inode */
207 
208 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
209 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
210 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
211 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
212 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
213 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
214 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
215 	unsigned long	r_req_flags;
216 
217 	struct mutex r_fill_mutex;
218 
219 	union ceph_mds_request_args r_args;
220 	int r_fmode;        /* file mode, if expecting cap */
221 	kuid_t r_uid;
222 	kgid_t r_gid;
223 	struct timespec r_stamp;
224 
225 	/* for choosing which mds to send this request to */
226 	int r_direct_mode;
227 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
228 
229 	/* data payload is used for xattr ops */
230 	struct ceph_pagelist *r_pagelist;
231 
232 	/* what caps shall we drop? */
233 	int r_inode_drop, r_inode_unless;
234 	int r_dentry_drop, r_dentry_unless;
235 	int r_old_dentry_drop, r_old_dentry_unless;
236 	struct inode *r_old_inode;
237 	int r_old_inode_drop, r_old_inode_unless;
238 
239 	struct ceph_msg  *r_request;  /* original request */
240 	int r_request_release_offset;
241 	struct ceph_msg  *r_reply;
242 	struct ceph_mds_reply_info_parsed r_reply_info;
243 	struct page *r_locked_page;
244 	int r_err;
245 
246 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
247 	unsigned long r_started;  /* start time to measure timeout against */
248 	unsigned long r_request_started; /* start time for mds request only,
249 					    used to measure lease durations */
250 
251 	/* link unsafe requests to parent directory, for fsync */
252 	struct inode	*r_unsafe_dir;
253 	struct list_head r_unsafe_dir_item;
254 
255 	/* unsafe requests that modify the target inode */
256 	struct list_head r_unsafe_target_item;
257 
258 	struct ceph_mds_session *r_session;
259 
260 	int               r_attempts;   /* resend attempts */
261 	int               r_num_fwd;    /* number of forward attempts */
262 	int               r_resend_mds; /* mds to resend to next, if any*/
263 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
264 
265 	struct kref       r_kref;
266 	struct list_head  r_wait;
267 	struct completion r_completion;
268 	struct completion r_safe_completion;
269 	ceph_mds_request_callback_t r_callback;
270 	ceph_mds_request_wait_callback_t r_wait_for_completion;
271 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
272 
273 	long long	  r_dir_release_cnt;
274 	long long	  r_dir_ordered_cnt;
275 	int		  r_readdir_cache_idx;
276 	u32               r_readdir_offset;
277 
278 	struct ceph_cap_reservation r_caps_reservation;
279 	int r_num_caps;
280 };
281 
282 struct ceph_pool_perm {
283 	struct rb_node node;
284 	int perm;
285 	s64 pool;
286 	size_t pool_ns_len;
287 	char pool_ns[];
288 };
289 
290 /*
291  * mds client state
292  */
293 struct ceph_mds_client {
294 	struct ceph_fs_client  *fsc;
295 	struct mutex            mutex;         /* all nested structures */
296 
297 	struct ceph_mdsmap      *mdsmap;
298 	struct completion       safe_umount_waiters;
299 	wait_queue_head_t       session_close_wq;
300 	struct list_head        waiting_for_map;
301 	int 			mdsmap_err;
302 
303 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
304 	atomic_t		num_sessions;
305 	int                     max_sessions;  /* len of s_mds_sessions */
306 	int                     stopping;      /* true if shutting down */
307 
308 	/*
309 	 * snap_rwsem will cover cap linkage into snaprealms, and
310 	 * realm snap contexts.  (later, we can do per-realm snap
311 	 * contexts locks..)  the empty list contains realms with no
312 	 * references (implying they contain no inodes with caps) that
313 	 * should be destroyed.
314 	 */
315 	u64			last_snap_seq;
316 	struct rw_semaphore     snap_rwsem;
317 	struct rb_root          snap_realms;
318 	struct list_head        snap_empty;
319 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
320 
321 	u64                    last_tid;      /* most recent mds request */
322 	u64                    oldest_tid;    /* oldest incomplete mds request,
323 						 excluding setfilelock requests */
324 	struct rb_root         request_tree;  /* pending mds requests */
325 	struct delayed_work    delayed_work;  /* delayed work */
326 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
327 	struct list_head cap_delay_list;   /* caps with delayed release */
328 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
329 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
330 	spinlock_t       snap_flush_lock;
331 
332 	u64               last_cap_flush_tid;
333 	struct list_head  cap_flush_list;
334 	struct list_head  cap_dirty;        /* inodes with dirty caps */
335 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
336 	int               num_cap_flushing; /* # caps we are flushing */
337 	spinlock_t        cap_dirty_lock;   /* protects above items */
338 	wait_queue_head_t cap_flushing_wq;
339 
340 	/*
341 	 * Cap reservations
342 	 *
343 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
344 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
345 	 * memory needed to successfully process an MDS response.  (If an MDS
346 	 * sends us cap information and we fail to process it, we will have
347 	 * problems due to the client and MDS being out of sync.)
348 	 *
349 	 * Reservations are 'owned' by a ceph_cap_reservation context.
350 	 */
351 	spinlock_t	caps_list_lock;
352 	struct		list_head caps_list; /* unused (reserved or
353 						unreserved) */
354 	int		caps_total_count;    /* total caps allocated */
355 	int		caps_use_count;      /* in use */
356 	int		caps_reserve_count;  /* unused, reserved */
357 	int		caps_avail_count;    /* unused, unreserved */
358 	int		caps_min_count;      /* keep at least this many
359 						(unreserved) */
360 	spinlock_t	  dentry_lru_lock;
361 	struct list_head  dentry_lru;
362 	int		  num_dentry;
363 
364 	struct rw_semaphore     pool_perm_rwsem;
365 	struct rb_root		pool_perm_tree;
366 };
367 
368 extern const char *ceph_mds_op_name(int op);
369 
370 extern struct ceph_mds_session *
371 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
372 
373 static inline struct ceph_mds_session *
374 ceph_get_mds_session(struct ceph_mds_session *s)
375 {
376 	atomic_inc(&s->s_ref);
377 	return s;
378 }
379 
380 extern const char *ceph_session_state_name(int s);
381 
382 extern void ceph_put_mds_session(struct ceph_mds_session *s);
383 
384 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
385 			     struct ceph_msg *msg, int mds);
386 
387 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
388 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
389 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
390 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
391 
392 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
393 
394 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
395 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
396 					   struct inode *dir);
397 extern struct ceph_mds_request *
398 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
399 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
400 				     struct ceph_mds_request *req);
401 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
402 				struct inode *dir,
403 				struct ceph_mds_request *req);
404 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
405 {
406 	kref_get(&req->r_kref);
407 }
408 extern void ceph_mdsc_release_request(struct kref *kref);
409 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
410 {
411 	kref_put(&req->r_kref, ceph_mdsc_release_request);
412 }
413 
414 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
415 				   struct ceph_mds_session *session);
416 
417 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
418 
419 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
420 				  int stop_on_nosnap);
421 
422 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
423 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
424 				     struct inode *inode,
425 				     struct dentry *dentry, char action,
426 				     u32 seq);
427 
428 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
429 				    struct ceph_msg *msg);
430 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
431 				   struct ceph_msg *msg);
432 
433 extern struct ceph_mds_session *
434 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
435 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
436 					  struct ceph_mds_session *session);
437 
438 #endif
439