xref: /openbmc/linux/fs/ceph/mds_client.h (revision df687341)
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 #define CEPHFS_FEATURE_REPLY_ENCODING	9
22 #define CEPHFS_FEATURE_RECLAIM_CLIENT	10
23 #define CEPHFS_FEATURE_LAZY_CAP_WANTED	11
24 
25 #define CEPHFS_FEATURES_CLIENT_SUPPORTED { 	\
26 	0, 1, 2, 3, 4, 5, 6, 7,			\
27 	CEPHFS_FEATURE_MIMIC,			\
28 	CEPHFS_FEATURE_LAZY_CAP_WANTED,		\
29 }
30 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
31 
32 
33 /*
34  * Some lock dependencies:
35  *
36  * session->s_mutex
37  *         mdsc->mutex
38  *
39  *         mdsc->snap_rwsem
40  *
41  *         ci->i_ceph_lock
42  *                 mdsc->snap_flush_lock
43  *                 mdsc->cap_delay_lock
44  *
45  */
46 
47 struct ceph_fs_client;
48 struct ceph_cap;
49 
50 /*
51  * parsed info about a single inode.  pointers are into the encoded
52  * on-wire structures within the mds reply message payload.
53  */
54 struct ceph_mds_reply_info_in {
55 	struct ceph_mds_reply_inode *in;
56 	struct ceph_dir_layout dir_layout;
57 	u32 symlink_len;
58 	char *symlink;
59 	u32 xattr_len;
60 	char *xattr_data;
61 	u64 inline_version;
62 	u32 inline_len;
63 	char *inline_data;
64 	u32 pool_ns_len;
65 	char *pool_ns_data;
66 	u64 max_bytes;
67 	u64 max_files;
68 };
69 
70 struct ceph_mds_reply_dir_entry {
71 	char                          *name;
72 	u32                           name_len;
73 	struct ceph_mds_reply_lease   *lease;
74 	struct ceph_mds_reply_info_in inode;
75 	loff_t			      offset;
76 };
77 
78 /*
79  * parsed info about an mds reply, including information about
80  * either: 1) the target inode and/or its parent directory and dentry,
81  * and directory contents (for readdir results), or
82  * 2) the file range lock info (for fcntl F_GETLK results).
83  */
84 struct ceph_mds_reply_info_parsed {
85 	struct ceph_mds_reply_head    *head;
86 
87 	/* trace */
88 	struct ceph_mds_reply_info_in diri, targeti;
89 	struct ceph_mds_reply_dirfrag *dirfrag;
90 	char                          *dname;
91 	u32                           dname_len;
92 	struct ceph_mds_reply_lease   *dlease;
93 
94 	/* extra */
95 	union {
96 		/* for fcntl F_GETLK results */
97 		struct ceph_filelock *filelock_reply;
98 
99 		/* for readdir results */
100 		struct {
101 			struct ceph_mds_reply_dirfrag *dir_dir;
102 			size_t			      dir_buf_size;
103 			int                           dir_nr;
104 			bool			      dir_end;
105 			bool			      dir_complete;
106 			bool			      hash_order;
107 			bool			      offset_hash;
108 			struct ceph_mds_reply_dir_entry  *dir_entries;
109 		};
110 
111 		/* for create results */
112 		struct {
113 			bool has_create_ino;
114 			u64 ino;
115 		};
116 	};
117 
118 	/* encoded blob describing snapshot contexts for certain
119 	   operations (e.g., open) */
120 	void *snapblob;
121 	int snapblob_len;
122 };
123 
124 
125 /*
126  * cap releases are batched and sent to the MDS en masse.
127  *
128  * Account for per-message overhead of mds_cap_release header
129  * and __le32 for osd epoch barrier trailing field.
130  */
131 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -		\
132 				sizeof(struct ceph_mds_cap_release)) /	\
133 			        sizeof(struct ceph_mds_cap_item))
134 
135 
136 /*
137  * state associated with each MDS<->client session
138  */
139 enum {
140 	CEPH_MDS_SESSION_NEW = 1,
141 	CEPH_MDS_SESSION_OPENING = 2,
142 	CEPH_MDS_SESSION_OPEN = 3,
143 	CEPH_MDS_SESSION_HUNG = 4,
144 	CEPH_MDS_SESSION_CLOSING = 5,
145 	CEPH_MDS_SESSION_RESTARTING = 6,
146 	CEPH_MDS_SESSION_RECONNECTING = 7,
147 	CEPH_MDS_SESSION_REJECTED = 8,
148 };
149 
150 struct ceph_mds_session {
151 	struct ceph_mds_client *s_mdsc;
152 	int               s_mds;
153 	int               s_state;
154 	unsigned long     s_ttl;      /* time until mds kills us */
155 	u64               s_seq;      /* incoming msg seq # */
156 	struct mutex      s_mutex;    /* serialize session messages */
157 
158 	struct ceph_connection s_con;
159 
160 	struct ceph_auth_handshake s_auth;
161 
162 	/* protected by s_gen_ttl_lock */
163 	spinlock_t        s_gen_ttl_lock;
164 	u32               s_cap_gen;  /* inc each time we get mds stale msg */
165 	unsigned long     s_cap_ttl;  /* when session caps expire */
166 
167 	/* protected by s_cap_lock */
168 	spinlock_t        s_cap_lock;
169 	struct list_head  s_caps;     /* all caps issued by this session */
170 	int               s_nr_caps, s_trim_caps;
171 	int               s_num_cap_releases;
172 	int		  s_cap_reconnect;
173 	int		  s_readonly;
174 	struct list_head  s_cap_releases; /* waiting cap_release messages */
175 	struct ceph_cap  *s_cap_iterator;
176 
177 	/* protected by mutex */
178 	struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
179 	unsigned long     s_renew_requested; /* last time we sent a renew req */
180 	u64               s_renew_seq;
181 
182 	refcount_t          s_ref;
183 	struct list_head  s_waiting;  /* waiting requests */
184 	struct list_head  s_unsafe;   /* unsafe requests */
185 };
186 
187 /*
188  * modes of choosing which MDS to send a request to
189  */
190 enum {
191 	USE_ANY_MDS,
192 	USE_RANDOM_MDS,
193 	USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
194 };
195 
196 struct ceph_mds_request;
197 struct ceph_mds_client;
198 
199 /*
200  * request completion callback
201  */
202 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
203 					     struct ceph_mds_request *req);
204 /*
205  * wait for request completion callback
206  */
207 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
208 						 struct ceph_mds_request *req);
209 
210 /*
211  * an in-flight mds request
212  */
213 struct ceph_mds_request {
214 	u64 r_tid;                   /* transaction id */
215 	struct rb_node r_node;
216 	struct ceph_mds_client *r_mdsc;
217 
218 	int r_op;                    /* mds op code */
219 
220 	/* operation on what? */
221 	struct inode *r_inode;              /* arg1 */
222 	struct dentry *r_dentry;            /* arg1 */
223 	struct dentry *r_old_dentry;        /* arg2: rename from or link from */
224 	struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
225 	char *r_path1, *r_path2;
226 	struct ceph_vino r_ino1, r_ino2;
227 
228 	struct inode *r_parent;		    /* parent dir inode */
229 	struct inode *r_target_inode;       /* resulting inode */
230 
231 #define CEPH_MDS_R_DIRECT_IS_HASH	(1) /* r_direct_hash is valid */
232 #define CEPH_MDS_R_ABORTED		(2) /* call was aborted */
233 #define CEPH_MDS_R_GOT_UNSAFE		(3) /* got an unsafe reply */
234 #define CEPH_MDS_R_GOT_SAFE		(4) /* got a safe reply */
235 #define CEPH_MDS_R_GOT_RESULT		(5) /* got a result */
236 #define CEPH_MDS_R_DID_PREPOPULATE	(6) /* prepopulated readdir */
237 #define CEPH_MDS_R_PARENT_LOCKED	(7) /* is r_parent->i_rwsem wlocked? */
238 	unsigned long	r_req_flags;
239 
240 	struct mutex r_fill_mutex;
241 
242 	union ceph_mds_request_args r_args;
243 	int r_fmode;        /* file mode, if expecting cap */
244 	kuid_t r_uid;
245 	kgid_t r_gid;
246 	struct timespec64 r_stamp;
247 
248 	/* for choosing which mds to send this request to */
249 	int r_direct_mode;
250 	u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
251 
252 	/* data payload is used for xattr ops */
253 	struct ceph_pagelist *r_pagelist;
254 
255 	/* what caps shall we drop? */
256 	int r_inode_drop, r_inode_unless;
257 	int r_dentry_drop, r_dentry_unless;
258 	int r_old_dentry_drop, r_old_dentry_unless;
259 	struct inode *r_old_inode;
260 	int r_old_inode_drop, r_old_inode_unless;
261 
262 	struct ceph_msg  *r_request;  /* original request */
263 	int r_request_release_offset;
264 	struct ceph_msg  *r_reply;
265 	struct ceph_mds_reply_info_parsed r_reply_info;
266 	struct page *r_locked_page;
267 	int r_err;
268 
269 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
270 	unsigned long r_started;  /* start time to measure timeout against */
271 	unsigned long r_request_started; /* start time for mds request only,
272 					    used to measure lease durations */
273 
274 	/* link unsafe requests to parent directory, for fsync */
275 	struct inode	*r_unsafe_dir;
276 	struct list_head r_unsafe_dir_item;
277 
278 	/* unsafe requests that modify the target inode */
279 	struct list_head r_unsafe_target_item;
280 
281 	struct ceph_mds_session *r_session;
282 
283 	int               r_attempts;   /* resend attempts */
284 	int               r_num_fwd;    /* number of forward attempts */
285 	int               r_resend_mds; /* mds to resend to next, if any*/
286 	u32               r_sent_on_mseq; /* cap mseq request was sent at*/
287 
288 	struct kref       r_kref;
289 	struct list_head  r_wait;
290 	struct completion r_completion;
291 	struct completion r_safe_completion;
292 	ceph_mds_request_callback_t r_callback;
293 	ceph_mds_request_wait_callback_t r_wait_for_completion;
294 	struct list_head  r_unsafe_item;  /* per-session unsafe list item */
295 
296 	long long	  r_dir_release_cnt;
297 	long long	  r_dir_ordered_cnt;
298 	int		  r_readdir_cache_idx;
299 	u32               r_readdir_offset;
300 
301 	struct ceph_cap_reservation r_caps_reservation;
302 	int r_num_caps;
303 };
304 
305 struct ceph_pool_perm {
306 	struct rb_node node;
307 	int perm;
308 	s64 pool;
309 	size_t pool_ns_len;
310 	char pool_ns[];
311 };
312 
313 /*
314  * mds client state
315  */
316 struct ceph_mds_client {
317 	struct ceph_fs_client  *fsc;
318 	struct mutex            mutex;         /* all nested structures */
319 
320 	struct ceph_mdsmap      *mdsmap;
321 	struct completion       safe_umount_waiters;
322 	wait_queue_head_t       session_close_wq;
323 	struct list_head        waiting_for_map;
324 	int 			mdsmap_err;
325 
326 	struct ceph_mds_session **sessions;    /* NULL for mds if no session */
327 	atomic_t		num_sessions;
328 	int                     max_sessions;  /* len of s_mds_sessions */
329 	int                     stopping;      /* true if shutting down */
330 
331 	atomic64_t		quotarealms_count; /* # realms with quota */
332 
333 	/*
334 	 * snap_rwsem will cover cap linkage into snaprealms, and
335 	 * realm snap contexts.  (later, we can do per-realm snap
336 	 * contexts locks..)  the empty list contains realms with no
337 	 * references (implying they contain no inodes with caps) that
338 	 * should be destroyed.
339 	 */
340 	u64			last_snap_seq;
341 	struct rw_semaphore     snap_rwsem;
342 	struct rb_root          snap_realms;
343 	struct list_head        snap_empty;
344 	spinlock_t              snap_empty_lock;  /* protect snap_empty */
345 
346 	u64                    last_tid;      /* most recent mds request */
347 	u64                    oldest_tid;    /* oldest incomplete mds request,
348 						 excluding setfilelock requests */
349 	struct rb_root         request_tree;  /* pending mds requests */
350 	struct delayed_work    delayed_work;  /* delayed work */
351 	unsigned long    last_renew_caps;  /* last time we renewed our caps */
352 	struct list_head cap_delay_list;   /* caps with delayed release */
353 	spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
354 	struct list_head snap_flush_list;  /* cap_snaps ready to flush */
355 	spinlock_t       snap_flush_lock;
356 
357 	u64               last_cap_flush_tid;
358 	struct list_head  cap_flush_list;
359 	struct list_head  cap_dirty;        /* inodes with dirty caps */
360 	struct list_head  cap_dirty_migrating; /* ...that are migration... */
361 	int               num_cap_flushing; /* # caps we are flushing */
362 	spinlock_t        cap_dirty_lock;   /* protects above items */
363 	wait_queue_head_t cap_flushing_wq;
364 
365 	/*
366 	 * Cap reservations
367 	 *
368 	 * Maintain a global pool of preallocated struct ceph_caps, referenced
369 	 * by struct ceph_caps_reservations.  This ensures that we preallocate
370 	 * memory needed to successfully process an MDS response.  (If an MDS
371 	 * sends us cap information and we fail to process it, we will have
372 	 * problems due to the client and MDS being out of sync.)
373 	 *
374 	 * Reservations are 'owned' by a ceph_cap_reservation context.
375 	 */
376 	spinlock_t	caps_list_lock;
377 	struct		list_head caps_list; /* unused (reserved or
378 						unreserved) */
379 	int		caps_total_count;    /* total caps allocated */
380 	int		caps_use_count;      /* in use */
381 	int		caps_reserve_count;  /* unused, reserved */
382 	int		caps_avail_count;    /* unused, unreserved */
383 	int		caps_min_count;      /* keep at least this many
384 						(unreserved) */
385 	spinlock_t	  dentry_lru_lock;
386 	struct list_head  dentry_lru;
387 	int		  num_dentry;
388 
389 	struct rw_semaphore     pool_perm_rwsem;
390 	struct rb_root		pool_perm_tree;
391 
392 	char nodename[__NEW_UTS_LEN + 1];
393 };
394 
395 extern const char *ceph_mds_op_name(int op);
396 
397 extern struct ceph_mds_session *
398 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
399 
400 static inline struct ceph_mds_session *
401 ceph_get_mds_session(struct ceph_mds_session *s)
402 {
403 	refcount_inc(&s->s_ref);
404 	return s;
405 }
406 
407 extern const char *ceph_session_state_name(int s);
408 
409 extern void ceph_put_mds_session(struct ceph_mds_session *s);
410 
411 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
412 			     struct ceph_msg *msg, int mds);
413 
414 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
415 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
416 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
417 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
418 
419 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
420 
421 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
422 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
423 					   struct inode *dir);
424 extern struct ceph_mds_request *
425 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
426 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
427 				     struct ceph_mds_request *req);
428 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
429 				struct inode *dir,
430 				struct ceph_mds_request *req);
431 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
432 {
433 	kref_get(&req->r_kref);
434 }
435 extern void ceph_mdsc_release_request(struct kref *kref);
436 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
437 {
438 	kref_put(&req->r_kref, ceph_mdsc_release_request);
439 }
440 
441 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
442 				   struct ceph_mds_session *session);
443 
444 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
445 
446 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
447 				  int stop_on_nosnap);
448 
449 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
450 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
451 				     struct inode *inode,
452 				     struct dentry *dentry, char action,
453 				     u32 seq);
454 
455 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
456 				    struct ceph_msg *msg);
457 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
458 				   struct ceph_msg *msg);
459 
460 extern struct ceph_mds_session *
461 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
462 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
463 					  struct ceph_mds_session *session);
464 
465 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
466 			  struct ceph_mds_session *session,
467 			  int max_caps);
468 #endif
469