xref: /openbmc/linux/fs/ceph/mds_client.h (revision a34a3ed7)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _FS_CEPH_MDS_CLIENT_H
3 #define _FS_CEPH_MDS_CLIENT_H
4 
5 #include <linux/completion.h>
6 #include <linux/kref.h>
7 #include <linux/list.h>
8 #include <linux/mutex.h>
9 #include <linux/rbtree.h>
10 #include <linux/spinlock.h>
11 #include <linux/refcount.h>
12 #include <linux/utsname.h>
13 
14 #include <linux/ceph/types.h>
15 #include <linux/ceph/messenger.h>
16 #include <linux/ceph/mdsmap.h>
17 #include <linux/ceph/auth.h>
18 
19 /*
20  * Some lock dependencies:
21  *
22  * session->s_mutex
23  *         mdsc->mutex
24  *
25  *         mdsc->snap_rwsem
26  *
27  *         ci->i_ceph_lock
28  *                 mdsc->snap_flush_lock
29  *                 mdsc->cap_delay_lock
30  *
31  */
32 
33 struct ceph_fs_client;
34 struct ceph_cap;
35 
36 /*
37  * parsed info about a single inode.  pointers are into the encoded
38  * on-wire structures within the mds reply message payload.
39  */
40 struct ceph_mds_reply_info_in {
41 	struct ceph_mds_reply_inode *in;
42 	struct ceph_dir_layout dir_layout;
43 	u32 symlink_len;
44 	char *symlink;
45 	u32 xattr_len;
46 	char *xattr_data;
47 	u64 inline_version;
48 	u32 inline_len;
49 	char *inline_data;
50 	u32 pool_ns_len;
51 	char *pool_ns_data;
52 };
53 
54 struct ceph_mds_reply_dir_entry {
55 	char                          *name;
56 	u32                           name_len;
57 	struct ceph_mds_reply_lease   *lease;
58 	struct ceph_mds_reply_info_in inode;
59 	loff_t			      offset;
60 };
61 
62 /*
63  * parsed info about an mds reply, including information about
64  * either: 1) the target inode and/or its parent directory and dentry,
65  * and directory contents (for readdir results), or
66  * 2) the file range lock info (for fcntl F_GETLK results).
67  */
68 struct ceph_mds_reply_info_parsed {
69 	struct ceph_mds_reply_head    *head;
70 
71 	/* trace */
72 	struct ceph_mds_reply_info_in diri, targeti;
73 	struct ceph_mds_reply_dirfrag *dirfrag;
74 	char                          *dname;
75 	u32                           dname_len;
76 	struct ceph_mds_reply_lease   *dlease;
77 
78 	/* extra */
79 	union {
80 		/* for fcntl F_GETLK results */
81 		struct ceph_filelock *filelock_reply;
82 
83 		/* for readdir results */
84 		struct {
85 			struct ceph_mds_reply_dirfrag *dir_dir;
86 			size_t			      dir_buf_size;
87 			int                           dir_nr;
88 			bool			      dir_end;
89 			bool			      dir_complete;
90 			bool			      hash_order;
91 			bool			      offset_hash;
92 			struct ceph_mds_reply_dir_entry  *dir_entries;
93 		};
94 
95 		/* for create results */
96 		struct {
97 			bool has_create_ino;
98 			u64 ino;
99 		};
100 	};
101 
102 	/* encoded blob describing snapshot contexts for certain
103 	   operations (e.g., open) */
104 	void *snapblob;
105 	int snapblob_len;
106 };
107 
108 
109 /*
110  * cap releases are batched and sent to the MDS en masse.
111  *
112  * Account for per-message overhead of mds_cap_release header
113  * and __le32 for osd epoch barrier trailing field.
114  */
115 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
116 				sizeof(struct ceph_mds_cap_release)) /	\
117 			        sizeof(struct ceph_mds_cap_item))
118 
119 
120 /*
121  * state associated with each MDS<->client session
122  */
123 enum {
124 	CEPH_MDS_SESSION_NEW = 1,
125 	CEPH_MDS_SESSION_OPENING = 2,
126 	CEPH_MDS_SESSION_OPEN = 3,
127 	CEPH_MDS_SESSION_HUNG = 4,
128 	CEPH_MDS_SESSION_CLOSING = 5,
129 	CEPH_MDS_SESSION_RESTARTING = 6,
130 	CEPH_MDS_SESSION_RECONNECTING = 7,
131 	CEPH_MDS_SESSION_REJECTED = 8,
132 };
133 
134 struct ceph_mds_session {
135 	struct ceph_mds_client *s_mdsc;
136 	int               s_mds;
137 	int               s_state;
138 	unsigned long     s_ttl;      /* time until mds kills us */
139 	u64               s_seq;      /* incoming msg seq # */
140 	struct mutex      s_mutex;    /* serialize session messages */
141 
142 	struct ceph_connection s_con;
143 
144 	struct ceph_auth_handshake s_auth;
145 
146 	/* protected by s_gen_ttl_lock */
147 	spinlock_t        s_gen_ttl_lock;
148 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
149 	unsigned long     s_cap_ttl;  /* when session caps expire */
150 
151 	/* protected by s_cap_lock */
152 	spinlock_t        s_cap_lock;
153 	struct list_head  s_caps;     /* all caps issued by this session */
154 	int               s_nr_caps, s_trim_caps;
155 	int               s_num_cap_releases;
156 	int		  s_cap_reconnect;
157 	int		  s_readonly;
158 	struct list_head  s_cap_releases; /* waiting cap_release messages */
159 	struct ceph_cap  *s_cap_iterator;
160 
161 	/* protected by mutex */
162 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
163 	unsigned long     s_renew_requested; /* last time we sent a renew req */
164 	u64               s_renew_seq;
165 
166 	refcount_t          s_ref;
167 	struct list_head  s_waiting;  /* waiting requests */
168 	struct list_head  s_unsafe;   /* unsafe requests */
169 };
170 
171 /*
172  * modes of choosing which MDS to send a request to
173  */
174 enum {
175 	USE_ANY_MDS,
176 	USE_RANDOM_MDS,
177 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
178 };
179 
180 struct ceph_mds_request;
181 struct ceph_mds_client;
182 
183 /*
184  * request completion callback
185  */
186 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
187 					     struct ceph_mds_request *req);
188 /*
189  * wait for request completion callback
190  */
191 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
192 						 struct ceph_mds_request *req);
193 
194 /*
195  * an in-flight mds request
196  */
197 struct ceph_mds_request {
198 	u64 r_tid;                   /* transaction id */
199 	struct rb_node r_node;
200 	struct ceph_mds_client *r_mdsc;
201 
202 	int r_op;                    /* mds op code */
203 
204 	/* operation on what? */
205 	struct inode *r_inode;              /* arg1 */
206 	struct dentry *r_dentry;            /* arg1 */
207 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
208 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
209 	char *r_path1, *r_path2;
210 	struct ceph_vino r_ino1, r_ino2;
211 
212 	struct inode *r_parent;		    /* parent dir inode */
213 	struct inode *r_target_inode;       /* resulting inode */
214 
215 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
216 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
217 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
218 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
219 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
220 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
221 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
222 	unsigned long	r_req_flags;
223 
224 	struct mutex r_fill_mutex;
225 
226 	union ceph_mds_request_args r_args;
227 	int r_fmode;        /* file mode, if expecting cap */
228 	kuid_t r_uid;
229 	kgid_t r_gid;
230 	struct timespec r_stamp;
231 
232 	/* for choosing which mds to send this request to */
233 	int r_direct_mode;
234 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
235 
236 	/* data payload is used for xattr ops */
237 	struct ceph_pagelist *r_pagelist;
238 
239 	/* what caps shall we drop? */
240 	int r_inode_drop, r_inode_unless;
241 	int r_dentry_drop, r_dentry_unless;
242 	int r_old_dentry_drop, r_old_dentry_unless;
243 	struct inode *r_old_inode;
244 	int r_old_inode_drop, r_old_inode_unless;
245 
246 	struct ceph_msg  *r_request;  /* original request */
247 	int r_request_release_offset;
248 	struct ceph_msg  *r_reply;
249 	struct ceph_mds_reply_info_parsed r_reply_info;
250 	struct page *r_locked_page;
251 	int r_err;
252 
253 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
254 	unsigned long r_started;  /* start time to measure timeout against */
255 	unsigned long r_request_started; /* start time for mds request only,
256 					    used to measure lease durations */
257 
258 	/* link unsafe requests to parent directory, for fsync */
259 	struct inode	*r_unsafe_dir;
260 	struct list_head r_unsafe_dir_item;
261 
262 	/* unsafe requests that modify the target inode */
263 	struct list_head r_unsafe_target_item;
264 
265 	struct ceph_mds_session *r_session;
266 
267 	int               r_attempts;   /* resend attempts */
268 	int               r_num_fwd;    /* number of forward attempts */
269 	int               r_resend_mds; /* mds to resend to next, if any*/
270 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
271 
272 	struct kref       r_kref;
273 	struct list_head  r_wait;
274 	struct completion r_completion;
275 	struct completion r_safe_completion;
276 	ceph_mds_request_callback_t r_callback;
277 	ceph_mds_request_wait_callback_t r_wait_for_completion;
278 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
279 
280 	long long	  r_dir_release_cnt;
281 	long long	  r_dir_ordered_cnt;
282 	int		  r_readdir_cache_idx;
283 	u32               r_readdir_offset;
284 
285 	struct ceph_cap_reservation r_caps_reservation;
286 	int r_num_caps;
287 };
288 
289 struct ceph_pool_perm {
290 	struct rb_node node;
291 	int perm;
292 	s64 pool;
293 	size_t pool_ns_len;
294 	char pool_ns[];
295 };
296 
297 /*
298  * mds client state
299  */
300 struct ceph_mds_client {
301 	struct ceph_fs_client  *fsc;
302 	struct mutex            mutex;         /* all nested structures */
303 
304 	struct ceph_mdsmap      *mdsmap;
305 	struct completion       safe_umount_waiters;
306 	wait_queue_head_t       session_close_wq;
307 	struct list_head        waiting_for_map;
308 	int 			mdsmap_err;
309 
310 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
311 	atomic_t		num_sessions;
312 	int                     max_sessions;  /* len of s_mds_sessions */
313 	int                     stopping;      /* true if shutting down */
314 
315 	/*
316 	 * snap_rwsem will cover cap linkage into snaprealms, and
317 	 * realm snap contexts.  (later, we can do per-realm snap
318 	 * contexts locks..)  the empty list contains realms with no
319 	 * references (implying they contain no inodes with caps) that
320 	 * should be destroyed.
321 	 */
322 	u64			last_snap_seq;
323 	struct rw_semaphore     snap_rwsem;
324 	struct rb_root          snap_realms;
325 	struct list_head        snap_empty;
326 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
327 
328 	u64                    last_tid;      /* most recent mds request */
329 	u64                    oldest_tid;    /* oldest incomplete mds request,
330 						 excluding setfilelock requests */
331 	struct rb_root         request_tree;  /* pending mds requests */
332 	struct delayed_work    delayed_work;  /* delayed work */
333 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
334 	struct list_head cap_delay_list;   /* caps with delayed release */
335 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
336 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
337 	spinlock_t       snap_flush_lock;
338 
339 	u64               last_cap_flush_tid;
340 	struct list_head  cap_flush_list;
341 	struct list_head  cap_dirty;        /* inodes with dirty caps */
342 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
343 	int               num_cap_flushing; /* # caps we are flushing */
344 	spinlock_t        cap_dirty_lock;   /* protects above items */
345 	wait_queue_head_t cap_flushing_wq;
346 
347 	/*
348 	 * Cap reservations
349 	 *
350 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
351 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
352 	 * memory needed to successfully process an MDS response.  (If an MDS
353 	 * sends us cap information and we fail to process it, we will have
354 	 * problems due to the client and MDS being out of sync.)
355 	 *
356 	 * Reservations are 'owned' by a ceph_cap_reservation context.
357 	 */
358 	spinlock_t	caps_list_lock;
359 	struct		list_head caps_list; /* unused (reserved or
360 						unreserved) */
361 	int		caps_total_count;    /* total caps allocated */
362 	int		caps_use_count;      /* in use */
363 	int		caps_reserve_count;  /* unused, reserved */
364 	int		caps_avail_count;    /* unused, unreserved */
365 	int		caps_min_count;      /* keep at least this many
366 						(unreserved) */
367 	spinlock_t	  dentry_lru_lock;
368 	struct list_head  dentry_lru;
369 	int		  num_dentry;
370 
371 	struct rw_semaphore     pool_perm_rwsem;
372 	struct rb_root		pool_perm_tree;
373 
374 	char nodename[__NEW_UTS_LEN + 1];
375 };
376 
377 extern const char *ceph_mds_op_name(int op);
378 
379 extern struct ceph_mds_session *
380 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
381 
382 static inline struct ceph_mds_session *
383 ceph_get_mds_session(struct ceph_mds_session *s)
384 {
385 	refcount_inc(&s->s_ref);
386 	return s;
387 }
388 
389 extern const char *ceph_session_state_name(int s);
390 
391 extern void ceph_put_mds_session(struct ceph_mds_session *s);
392 
393 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
394 			     struct ceph_msg *msg, int mds);
395 
396 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
397 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
398 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
399 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
400 
401 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
402 
403 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
404 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
405 					   struct inode *dir);
406 extern struct ceph_mds_request *
407 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
408 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
409 				     struct ceph_mds_request *req);
410 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
411 				struct inode *dir,
412 				struct ceph_mds_request *req);
413 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
414 {
415 	kref_get(&req->r_kref);
416 }
417 extern void ceph_mdsc_release_request(struct kref *kref);
418 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
419 {
420 	kref_put(&req->r_kref, ceph_mdsc_release_request);
421 }
422 
423 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
424 				   struct ceph_mds_session *session);
425 
426 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
427 
428 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
429 				  int stop_on_nosnap);
430 
431 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
432 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
433 				     struct inode *inode,
434 				     struct dentry *dentry, char action,
435 				     u32 seq);
436 
437 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
438 				    struct ceph_msg *msg);
439 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
440 				   struct ceph_msg *msg);
441 
442 extern struct ceph_mds_session *
443 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
444 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
445 					  struct ceph_mds_session *session);
446 
447 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
448 			  struct ceph_mds_session *session,
449 			  int max_caps);
450 #endif
451