xref: /openbmc/linux/fs/ceph/mds_client.h (revision e53134fe)
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 /* The first 8 bits are reserved for old ceph releases */
20 #define CEPHFS_FEATURE_MIMIC    8
21 
22 #define CEPHFS_FEATURES_ALL {           \
23   0, 1, 2, 3, 4, 5, 6, 7,		\
24   CEPHFS_FEATURE_MIMIC,                 \
25 }
26 
27 #define CEPHFS_FEATURES_CLIENT_SUPPORTED CEPHFS_FEATURES_ALL
28 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
29 
30 
31 /*
32  * Some lock dependencies:
33  *
34  * session->s_mutex
35  *         mdsc->mutex
36  *
37  *         mdsc->snap_rwsem
38  *
39  *         ci->i_ceph_lock
40  *                 mdsc->snap_flush_lock
41  *                 mdsc->cap_delay_lock
42  *
43  */
44 
45 struct ceph_fs_client;
46 struct ceph_cap;
47 
48 /*
49  * parsed info about a single inode.  pointers are into the encoded
50  * on-wire structures within the mds reply message payload.
51  */
52 struct ceph_mds_reply_info_in {
53 	struct ceph_mds_reply_inode *in;
54 	struct ceph_dir_layout dir_layout;
55 	u32 symlink_len;
56 	char *symlink;
57 	u32 xattr_len;
58 	char *xattr_data;
59 	u64 inline_version;
60 	u32 inline_len;
61 	char *inline_data;
62 	u32 pool_ns_len;
63 	char *pool_ns_data;
64 	u64 max_bytes;
65 	u64 max_files;
66 };
67 
68 struct ceph_mds_reply_dir_entry {
69 	char                          *name;
70 	u32                           name_len;
71 	struct ceph_mds_reply_lease   *lease;
72 	struct ceph_mds_reply_info_in inode;
73 	loff_t			      offset;
74 };
75 
76 /*
77  * parsed info about an mds reply, including information about
78  * either: 1) the target inode and/or its parent directory and dentry,
79  * and directory contents (for readdir results), or
80  * 2) the file range lock info (for fcntl F_GETLK results).
81  */
82 struct ceph_mds_reply_info_parsed {
83 	struct ceph_mds_reply_head    *head;
84 
85 	/* trace */
86 	struct ceph_mds_reply_info_in diri, targeti;
87 	struct ceph_mds_reply_dirfrag *dirfrag;
88 	char                          *dname;
89 	u32                           dname_len;
90 	struct ceph_mds_reply_lease   *dlease;
91 
92 	/* extra */
93 	union {
94 		/* for fcntl F_GETLK results */
95 		struct ceph_filelock *filelock_reply;
96 
97 		/* for readdir results */
98 		struct {
99 			struct ceph_mds_reply_dirfrag *dir_dir;
100 			size_t			      dir_buf_size;
101 			int                           dir_nr;
102 			bool			      dir_end;
103 			bool			      dir_complete;
104 			bool			      hash_order;
105 			bool			      offset_hash;
106 			struct ceph_mds_reply_dir_entry  *dir_entries;
107 		};
108 
109 		/* for create results */
110 		struct {
111 			bool has_create_ino;
112 			u64 ino;
113 		};
114 	};
115 
116 	/* encoded blob describing snapshot contexts for certain
117 	   operations (e.g., open) */
118 	void *snapblob;
119 	int snapblob_len;
120 };
121 
122 
123 /*
124  * cap releases are batched and sent to the MDS en masse.
125  *
126  * Account for per-message overhead of mds_cap_release header
127  * and __le32 for osd epoch barrier trailing field.
128  */
129 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
130 				sizeof(struct ceph_mds_cap_release)) /	\
131 			        sizeof(struct ceph_mds_cap_item))
132 
133 
134 /*
135  * state associated with each MDS<->client session
136  */
137 enum {
138 	CEPH_MDS_SESSION_NEW = 1,
139 	CEPH_MDS_SESSION_OPENING = 2,
140 	CEPH_MDS_SESSION_OPEN = 3,
141 	CEPH_MDS_SESSION_HUNG = 4,
142 	CEPH_MDS_SESSION_CLOSING = 5,
143 	CEPH_MDS_SESSION_RESTARTING = 6,
144 	CEPH_MDS_SESSION_RECONNECTING = 7,
145 	CEPH_MDS_SESSION_REJECTED = 8,
146 };
147 
148 struct ceph_mds_session {
149 	struct ceph_mds_client *s_mdsc;
150 	int               s_mds;
151 	int               s_state;
152 	unsigned long     s_ttl;      /* time until mds kills us */
153 	u64               s_seq;      /* incoming msg seq # */
154 	struct mutex      s_mutex;    /* serialize session messages */
155 
156 	struct ceph_connection s_con;
157 
158 	struct ceph_auth_handshake s_auth;
159 
160 	/* protected by s_gen_ttl_lock */
161 	spinlock_t        s_gen_ttl_lock;
162 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
163 	unsigned long     s_cap_ttl;  /* when session caps expire */
164 
165 	/* protected by s_cap_lock */
166 	spinlock_t        s_cap_lock;
167 	struct list_head  s_caps;     /* all caps issued by this session */
168 	int               s_nr_caps, s_trim_caps;
169 	int               s_num_cap_releases;
170 	int		  s_cap_reconnect;
171 	int		  s_readonly;
172 	struct list_head  s_cap_releases; /* waiting cap_release messages */
173 	struct ceph_cap  *s_cap_iterator;
174 
175 	/* protected by mutex */
176 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
177 	unsigned long     s_renew_requested; /* last time we sent a renew req */
178 	u64               s_renew_seq;
179 
180 	refcount_t          s_ref;
181 	struct list_head  s_waiting;  /* waiting requests */
182 	struct list_head  s_unsafe;   /* unsafe requests */
183 };
184 
185 /*
186  * modes of choosing which MDS to send a request to
187  */
188 enum {
189 	USE_ANY_MDS,
190 	USE_RANDOM_MDS,
191 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
192 };
193 
194 struct ceph_mds_request;
195 struct ceph_mds_client;
196 
197 /*
198  * request completion callback
199  */
200 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
201 					     struct ceph_mds_request *req);
202 /*
203  * wait for request completion callback
204  */
205 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
206 						 struct ceph_mds_request *req);
207 
208 /*
209  * an in-flight mds request
210  */
211 struct ceph_mds_request {
212 	u64 r_tid;                   /* transaction id */
213 	struct rb_node r_node;
214 	struct ceph_mds_client *r_mdsc;
215 
216 	int r_op;                    /* mds op code */
217 
218 	/* operation on what? */
219 	struct inode *r_inode;              /* arg1 */
220 	struct dentry *r_dentry;            /* arg1 */
221 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
222 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
223 	char *r_path1, *r_path2;
224 	struct ceph_vino r_ino1, r_ino2;
225 
226 	struct inode *r_parent;		    /* parent dir inode */
227 	struct inode *r_target_inode;       /* resulting inode */
228 
229 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
230 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
231 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
232 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
233 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
234 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
235 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
236 	unsigned long	r_req_flags;
237 
238 	struct mutex r_fill_mutex;
239 
240 	union ceph_mds_request_args r_args;
241 	int r_fmode;        /* file mode, if expecting cap */
242 	kuid_t r_uid;
243 	kgid_t r_gid;
244 	struct timespec64 r_stamp;
245 
246 	/* for choosing which mds to send this request to */
247 	int r_direct_mode;
248 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
249 
250 	/* data payload is used for xattr ops */
251 	struct ceph_pagelist *r_pagelist;
252 
253 	/* what caps shall we drop? */
254 	int r_inode_drop, r_inode_unless;
255 	int r_dentry_drop, r_dentry_unless;
256 	int r_old_dentry_drop, r_old_dentry_unless;
257 	struct inode *r_old_inode;
258 	int r_old_inode_drop, r_old_inode_unless;
259 
260 	struct ceph_msg  *r_request;  /* original request */
261 	int r_request_release_offset;
262 	struct ceph_msg  *r_reply;
263 	struct ceph_mds_reply_info_parsed r_reply_info;
264 	struct page *r_locked_page;
265 	int r_err;
266 
267 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
268 	unsigned long r_started;  /* start time to measure timeout against */
269 	unsigned long r_request_started; /* start time for mds request only,
270 					    used to measure lease durations */
271 
272 	/* link unsafe requests to parent directory, for fsync */
273 	struct inode	*r_unsafe_dir;
274 	struct list_head r_unsafe_dir_item;
275 
276 	/* unsafe requests that modify the target inode */
277 	struct list_head r_unsafe_target_item;
278 
279 	struct ceph_mds_session *r_session;
280 
281 	int               r_attempts;   /* resend attempts */
282 	int               r_num_fwd;    /* number of forward attempts */
283 	int               r_resend_mds; /* mds to resend to next, if any*/
284 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
285 
286 	struct kref       r_kref;
287 	struct list_head  r_wait;
288 	struct completion r_completion;
289 	struct completion r_safe_completion;
290 	ceph_mds_request_callback_t r_callback;
291 	ceph_mds_request_wait_callback_t r_wait_for_completion;
292 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
293 
294 	long long	  r_dir_release_cnt;
295 	long long	  r_dir_ordered_cnt;
296 	int		  r_readdir_cache_idx;
297 	u32               r_readdir_offset;
298 
299 	struct ceph_cap_reservation r_caps_reservation;
300 	int r_num_caps;
301 };
302 
303 struct ceph_pool_perm {
304 	struct rb_node node;
305 	int perm;
306 	s64 pool;
307 	size_t pool_ns_len;
308 	char pool_ns[];
309 };
310 
311 /*
312  * mds client state
313  */
314 struct ceph_mds_client {
315 	struct ceph_fs_client  *fsc;
316 	struct mutex            mutex;         /* all nested structures */
317 
318 	struct ceph_mdsmap      *mdsmap;
319 	struct completion       safe_umount_waiters;
320 	wait_queue_head_t       session_close_wq;
321 	struct list_head        waiting_for_map;
322 	int 			mdsmap_err;
323 
324 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
325 	atomic_t		num_sessions;
326 	int                     max_sessions;  /* len of s_mds_sessions */
327 	int                     stopping;      /* true if shutting down */
328 
329 	atomic64_t		quotarealms_count; /* # realms with quota */
330 
331 	/*
332 	 * snap_rwsem will cover cap linkage into snaprealms, and
333 	 * realm snap contexts.  (later, we can do per-realm snap
334 	 * contexts locks..)  the empty list contains realms with no
335 	 * references (implying they contain no inodes with caps) that
336 	 * should be destroyed.
337 	 */
338 	u64			last_snap_seq;
339 	struct rw_semaphore     snap_rwsem;
340 	struct rb_root          snap_realms;
341 	struct list_head        snap_empty;
342 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
343 
344 	u64                    last_tid;      /* most recent mds request */
345 	u64                    oldest_tid;    /* oldest incomplete mds request,
346 						 excluding setfilelock requests */
347 	struct rb_root         request_tree;  /* pending mds requests */
348 	struct delayed_work    delayed_work;  /* delayed work */
349 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
350 	struct list_head cap_delay_list;   /* caps with delayed release */
351 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
352 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
353 	spinlock_t       snap_flush_lock;
354 
355 	u64               last_cap_flush_tid;
356 	struct list_head  cap_flush_list;
357 	struct list_head  cap_dirty;        /* inodes with dirty caps */
358 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
359 	int               num_cap_flushing; /* # caps we are flushing */
360 	spinlock_t        cap_dirty_lock;   /* protects above items */
361 	wait_queue_head_t cap_flushing_wq;
362 
363 	/*
364 	 * Cap reservations
365 	 *
366 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
367 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
368 	 * memory needed to successfully process an MDS response.  (If an MDS
369 	 * sends us cap information and we fail to process it, we will have
370 	 * problems due to the client and MDS being out of sync.)
371 	 *
372 	 * Reservations are 'owned' by a ceph_cap_reservation context.
373 	 */
374 	spinlock_t	caps_list_lock;
375 	struct		list_head caps_list; /* unused (reserved or
376 						unreserved) */
377 	int		caps_total_count;    /* total caps allocated */
378 	int		caps_use_count;      /* in use */
379 	int		caps_reserve_count;  /* unused, reserved */
380 	int		caps_avail_count;    /* unused, unreserved */
381 	int		caps_min_count;      /* keep at least this many
382 						(unreserved) */
383 	spinlock_t	  dentry_lru_lock;
384 	struct list_head  dentry_lru;
385 	int		  num_dentry;
386 
387 	struct rw_semaphore     pool_perm_rwsem;
388 	struct rb_root		pool_perm_tree;
389 
390 	char nodename[__NEW_UTS_LEN + 1];
391 };
392 
393 extern const char *ceph_mds_op_name(int op);
394 
395 extern struct ceph_mds_session *
396 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
397 
398 static inline struct ceph_mds_session *
399 ceph_get_mds_session(struct ceph_mds_session *s)
400 {
401 	refcount_inc(&s->s_ref);
402 	return s;
403 }
404 
405 extern const char *ceph_session_state_name(int s);
406 
407 extern void ceph_put_mds_session(struct ceph_mds_session *s);
408 
409 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
410 			     struct ceph_msg *msg, int mds);
411 
412 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
413 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
414 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
415 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
416 
417 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
418 
419 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
420 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
421 					   struct inode *dir);
422 extern struct ceph_mds_request *
423 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
424 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
425 				     struct ceph_mds_request *req);
426 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
427 				struct inode *dir,
428 				struct ceph_mds_request *req);
429 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
430 {
431 	kref_get(&req->r_kref);
432 }
433 extern void ceph_mdsc_release_request(struct kref *kref);
434 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
435 {
436 	kref_put(&req->r_kref, ceph_mdsc_release_request);
437 }
438 
439 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
440 				   struct ceph_mds_session *session);
441 
442 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
443 
444 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
445 				  int stop_on_nosnap);
446 
447 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
448 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
449 				     struct inode *inode,
450 				     struct dentry *dentry, char action,
451 				     u32 seq);
452 
453 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
454 				    struct ceph_msg *msg);
455 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
456 				   struct ceph_msg *msg);
457 
458 extern struct ceph_mds_session *
459 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
460 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
461 					  struct ceph_mds_session *session);
462 
463 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
464 			  struct ceph_mds_session *session,
465 			  int max_caps);
466 #endif
467