blob: 3288359353e9b84783a0d5ef65c06141398e625c [file] [log] [blame]
Sage Weil2f2dc052009-10-06 11:31:09 -07001#ifndef _FS_CEPH_MDS_CLIENT_H
2#define _FS_CEPH_MDS_CLIENT_H
3
4#include <linux/completion.h>
Sage Weil153c8e62009-12-07 12:31:09 -08005#include <linux/kref.h>
Sage Weil2f2dc052009-10-06 11:31:09 -07006#include <linux/list.h>
7#include <linux/mutex.h>
Sage Weil44ca18f2010-02-15 12:08:46 -08008#include <linux/rbtree.h>
Sage Weil2f2dc052009-10-06 11:31:09 -07009#include <linux/spinlock.h>
10
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070011#include <linux/ceph/types.h>
12#include <linux/ceph/messenger.h>
13#include <linux/ceph/mdsmap.h>
Alex Elder6c4a1912012-05-16 15:16:38 -050014#include <linux/ceph/auth.h>
Sage Weil2f2dc052009-10-06 11:31:09 -070015
16/*
17 * Some lock dependencies:
18 *
19 * session->s_mutex
20 * mdsc->mutex
21 *
22 * mdsc->snap_rwsem
23 *
Sage Weilbe655592011-11-30 09:47:09 -080024 * ci->i_ceph_lock
Sage Weil2f2dc052009-10-06 11:31:09 -070025 * mdsc->snap_flush_lock
26 * mdsc->cap_delay_lock
27 *
28 */
29
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070030struct ceph_fs_client;
Sage Weil2f2dc052009-10-06 11:31:09 -070031struct 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 */
37struct ceph_mds_reply_info_in {
38 struct ceph_mds_reply_inode *in;
Sage Weil14303d22010-12-14 17:37:52 -080039 struct ceph_dir_layout dir_layout;
Sage Weil2f2dc052009-10-06 11:31:09 -070040 u32 symlink_len;
41 char *symlink;
42 u32 xattr_len;
43 char *xattr_data;
44};
45
46/*
Herb Shiu25933ab2010-12-01 14:14:38 -080047 * parsed info about an mds reply, including information about
48 * either: 1) the target inode and/or its parent directory and dentry,
49 * and directory contents (for readdir results), or
50 * 2) the file range lock info (for fcntl F_GETLK results).
Sage Weil2f2dc052009-10-06 11:31:09 -070051 */
52struct ceph_mds_reply_info_parsed {
53 struct ceph_mds_reply_head *head;
54
Herb Shiu25933ab2010-12-01 14:14:38 -080055 /* trace */
Sage Weil2f2dc052009-10-06 11:31:09 -070056 struct ceph_mds_reply_info_in diri, targeti;
57 struct ceph_mds_reply_dirfrag *dirfrag;
58 char *dname;
59 u32 dname_len;
60 struct ceph_mds_reply_lease *dlease;
61
Herb Shiu25933ab2010-12-01 14:14:38 -080062 /* extra */
63 union {
64 /* for fcntl F_GETLK results */
65 struct ceph_filelock *filelock_reply;
66
67 /* for readdir results */
68 struct {
69 struct ceph_mds_reply_dirfrag *dir_dir;
Yan, Zheng54008392014-03-29 13:41:15 +080070 size_t dir_buf_size;
Herb Shiu25933ab2010-12-01 14:14:38 -080071 int dir_nr;
72 char **dir_dname;
73 u32 *dir_dname_len;
74 struct ceph_mds_reply_lease **dir_dlease;
75 struct ceph_mds_reply_info_in *dir_in;
76 u8 dir_complete, dir_end;
77 };
Sam Lang6e8575f2012-12-28 09:56:46 -080078
79 /* for create results */
80 struct {
81 bool has_create_ino;
82 u64 ino;
83 };
Herb Shiu25933ab2010-12-01 14:14:38 -080084 };
Sage Weil2f2dc052009-10-06 11:31:09 -070085
86 /* encoded blob describing snapshot contexts for certain
87 operations (e.g., open) */
88 void *snapblob;
89 int snapblob_len;
90};
91
92
93/*
94 * cap releases are batched and sent to the MDS en masse.
95 */
96#define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE - \
97 sizeof(struct ceph_mds_cap_release)) / \
98 sizeof(struct ceph_mds_cap_item))
99
100
101/*
102 * state associated with each MDS<->client session
103 */
104enum {
105 CEPH_MDS_SESSION_NEW = 1,
106 CEPH_MDS_SESSION_OPENING = 2,
107 CEPH_MDS_SESSION_OPEN = 3,
108 CEPH_MDS_SESSION_HUNG = 4,
109 CEPH_MDS_SESSION_CLOSING = 5,
110 CEPH_MDS_SESSION_RESTARTING = 6,
111 CEPH_MDS_SESSION_RECONNECTING = 7,
112};
113
114struct ceph_mds_session {
115 struct ceph_mds_client *s_mdsc;
116 int s_mds;
117 int s_state;
118 unsigned long s_ttl; /* time until mds kills us */
119 u64 s_seq; /* incoming msg seq # */
120 struct mutex s_mutex; /* serialize session messages */
121
Sage Weil2f2dc052009-10-06 11:31:09 -0700122 struct ceph_connection s_con;
123
Alex Elder6c4a1912012-05-16 15:16:38 -0500124 struct ceph_auth_handshake s_auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800125
Alex Elderd8fb02a2012-01-12 17:48:10 -0800126 /* protected by s_gen_ttl_lock */
127 spinlock_t s_gen_ttl_lock;
Sage Weil2f2dc052009-10-06 11:31:09 -0700128 u32 s_cap_gen; /* inc each time we get mds stale msg */
129 unsigned long s_cap_ttl; /* when session caps expire */
Alex Elderd8fb02a2012-01-12 17:48:10 -0800130
131 /* protected by s_cap_lock */
132 spinlock_t s_cap_lock;
Sage Weil2f2dc052009-10-06 11:31:09 -0700133 struct list_head s_caps; /* all caps issued by this session */
134 int s_nr_caps, s_trim_caps;
135 int s_num_cap_releases;
Yan, Zheng99a9c272013-09-22 11:08:14 +0800136 int s_cap_reconnect;
Sage Weil2f2dc052009-10-06 11:31:09 -0700137 struct list_head s_cap_releases; /* waiting cap_release messages */
138 struct list_head s_cap_releases_done; /* ready to send */
Sage Weil7c1332b2010-02-16 11:39:45 -0800139 struct ceph_cap *s_cap_iterator;
Sage Weil2f2dc052009-10-06 11:31:09 -0700140
141 /* protected by mutex */
142 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
143 struct list_head s_cap_snaps_flushing;
144 unsigned long s_renew_requested; /* last time we sent a renew req */
145 u64 s_renew_seq;
146
147 atomic_t s_ref;
148 struct list_head s_waiting; /* waiting requests */
149 struct list_head s_unsafe; /* unsafe requests */
150};
151
152/*
153 * modes of choosing which MDS to send a request to
154 */
155enum {
156 USE_ANY_MDS,
157 USE_RANDOM_MDS,
158 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
159};
160
161struct ceph_mds_request;
162struct ceph_mds_client;
163
164/*
165 * request completion callback
166 */
167typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
168 struct ceph_mds_request *req);
169
170/*
171 * an in-flight mds request
172 */
173struct ceph_mds_request {
174 u64 r_tid; /* transaction id */
Sage Weil44ca18f2010-02-15 12:08:46 -0800175 struct rb_node r_node;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700176 struct ceph_mds_client *r_mdsc;
Sage Weil2f2dc052009-10-06 11:31:09 -0700177
178 int r_op; /* mds op code */
Sage Weil2f2dc052009-10-06 11:31:09 -0700179
180 /* operation on what? */
181 struct inode *r_inode; /* arg1 */
182 struct dentry *r_dentry; /* arg1 */
183 struct dentry *r_old_dentry; /* arg2: rename from or link from */
Sage Weil41b02e12011-07-26 11:31:14 -0700184 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
Sage Weil2f2dc052009-10-06 11:31:09 -0700185 char *r_path1, *r_path2;
186 struct ceph_vino r_ino1, r_ino2;
187
188 struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
189 struct inode *r_target_inode; /* resulting inode */
190
Sage Weilb4556392010-05-13 12:01:13 -0700191 struct mutex r_fill_mutex;
192
Sage Weil2f2dc052009-10-06 11:31:09 -0700193 union ceph_mds_request_args r_args;
194 int r_fmode; /* file mode, if expecting cap */
Eric W. Biedermanff3d0042013-01-31 04:01:53 -0800195 kuid_t r_uid;
196 kgid_t r_gid;
Sage Weilb8e69062014-05-21 17:41:08 -0700197 struct timespec r_stamp;
Sage Weil2f2dc052009-10-06 11:31:09 -0700198
199 /* for choosing which mds to send this request to */
200 int r_direct_mode;
201 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
202 bool r_direct_is_hash; /* true if r_direct_hash is valid */
203
204 /* data payload is used for xattr ops */
Yan, Zheng25e6bae2014-09-16 19:15:28 +0800205 struct ceph_pagelist *r_pagelist;
Sage Weil2f2dc052009-10-06 11:31:09 -0700206
207 /* what caps shall we drop? */
208 int r_inode_drop, r_inode_unless;
209 int r_dentry_drop, r_dentry_unless;
210 int r_old_dentry_drop, r_old_dentry_unless;
211 struct inode *r_old_inode;
212 int r_old_inode_drop, r_old_inode_unless;
213
214 struct ceph_msg *r_request; /* original request */
Sage Weile979cf52010-07-15 14:58:39 -0700215 int r_request_release_offset;
Sage Weil2f2dc052009-10-06 11:31:09 -0700216 struct ceph_msg *r_reply;
217 struct ceph_mds_reply_info_parsed r_reply_info;
218 int r_err;
Sage Weil5b1daec2010-01-25 11:33:08 -0800219 bool r_aborted;
Sage Weil2f2dc052009-10-06 11:31:09 -0700220
221 unsigned long r_timeout; /* optional. jiffies */
222 unsigned long r_started; /* start time to measure timeout against */
223 unsigned long r_request_started; /* start time for mds request only,
224 used to measure lease durations */
225
226 /* link unsafe requests to parent directory, for fsync */
227 struct inode *r_unsafe_dir;
228 struct list_head r_unsafe_dir_item;
229
230 struct ceph_mds_session *r_session;
231
232 int r_attempts; /* resend attempts */
233 int r_num_fwd; /* number of forward attempts */
Sage Weil2f2dc052009-10-06 11:31:09 -0700234 int r_resend_mds; /* mds to resend to next, if any*/
Greg Farnume55b71f2010-06-22 15:58:01 -0700235 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
Sage Weil2f2dc052009-10-06 11:31:09 -0700236
Sage Weil153c8e62009-12-07 12:31:09 -0800237 struct kref r_kref;
Sage Weil2f2dc052009-10-06 11:31:09 -0700238 struct list_head r_wait;
239 struct completion r_completion;
240 struct completion r_safe_completion;
241 ceph_mds_request_callback_t r_callback;
242 struct list_head r_unsafe_item; /* per-session unsafe list item */
Sage Weile1518c72010-05-13 11:19:06 -0700243 bool r_got_unsafe, r_got_safe, r_got_result;
Sage Weil2f2dc052009-10-06 11:31:09 -0700244
245 bool r_did_prepopulate;
246 u32 r_readdir_offset;
247
248 struct ceph_cap_reservation r_caps_reservation;
249 int r_num_caps;
250};
251
252/*
253 * mds client state
254 */
255struct ceph_mds_client {
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700256 struct ceph_fs_client *fsc;
Sage Weil2f2dc052009-10-06 11:31:09 -0700257 struct mutex mutex; /* all nested structures */
258
259 struct ceph_mdsmap *mdsmap;
Sage Weilf3c60c52010-08-11 14:51:23 -0700260 struct completion safe_umount_waiters;
261 wait_queue_head_t session_close_wq;
Sage Weil2f2dc052009-10-06 11:31:09 -0700262 struct list_head waiting_for_map;
263
264 struct ceph_mds_session **sessions; /* NULL for mds if no session */
265 int max_sessions; /* len of s_mds_sessions */
266 int stopping; /* true if shutting down */
267
268 /*
269 * snap_rwsem will cover cap linkage into snaprealms, and
270 * realm snap contexts. (later, we can do per-realm snap
271 * contexts locks..) the empty list contains realms with no
272 * references (implying they contain no inodes with caps) that
273 * should be destroyed.
274 */
275 struct rw_semaphore snap_rwsem;
Sage Weila105f002010-02-15 14:37:55 -0800276 struct rb_root snap_realms;
Sage Weil2f2dc052009-10-06 11:31:09 -0700277 struct list_head snap_empty;
278 spinlock_t snap_empty_lock; /* protect snap_empty */
279
280 u64 last_tid; /* most recent mds request */
Sage Weil44ca18f2010-02-15 12:08:46 -0800281 struct rb_root request_tree; /* pending mds requests */
Sage Weil2f2dc052009-10-06 11:31:09 -0700282 struct delayed_work delayed_work; /* delayed work */
283 unsigned long last_renew_caps; /* last time we renewed our caps */
284 struct list_head cap_delay_list; /* caps with delayed release */
285 spinlock_t cap_delay_lock; /* protects cap_delay_list */
286 struct list_head snap_flush_list; /* cap_snaps ready to flush */
287 spinlock_t snap_flush_lock;
288
289 u64 cap_flush_seq;
290 struct list_head cap_dirty; /* inodes with dirty caps */
Sage Weildb354052011-05-24 11:46:31 -0700291 struct list_head cap_dirty_migrating; /* ...that are migration... */
Sage Weil2f2dc052009-10-06 11:31:09 -0700292 int num_cap_flushing; /* # caps we are flushing */
293 spinlock_t cap_dirty_lock; /* protects above items */
294 wait_queue_head_t cap_flushing_wq;
295
Yehuda Sadeh37151662010-06-17 16:16:12 -0700296 /*
297 * Cap reservations
298 *
299 * Maintain a global pool of preallocated struct ceph_caps, referenced
300 * by struct ceph_caps_reservations. This ensures that we preallocate
301 * memory needed to successfully process an MDS response. (If an MDS
302 * sends us cap information and we fail to process it, we will have
303 * problems due to the client and MDS being out of sync.)
304 *
305 * Reservations are 'owned' by a ceph_cap_reservation context.
306 */
307 spinlock_t caps_list_lock;
308 struct list_head caps_list; /* unused (reserved or
309 unreserved) */
310 int caps_total_count; /* total caps allocated */
311 int caps_use_count; /* in use */
312 int caps_reserve_count; /* unused, reserved */
313 int caps_avail_count; /* unused, unreserved */
314 int caps_min_count; /* keep at least this many
315 (unreserved) */
Sage Weil2f2dc052009-10-06 11:31:09 -0700316 spinlock_t dentry_lru_lock;
317 struct list_head dentry_lru;
318 int num_dentry;
319};
320
321extern const char *ceph_mds_op_name(int op);
322
323extern struct ceph_mds_session *
324__ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
325
326static inline struct ceph_mds_session *
327ceph_get_mds_session(struct ceph_mds_session *s)
328{
329 atomic_inc(&s->s_ref);
330 return s;
331}
332
John Spraya687eca2014-09-19 13:51:08 +0100333extern const char *ceph_session_state_name(int s);
334
Sage Weil2f2dc052009-10-06 11:31:09 -0700335extern void ceph_put_mds_session(struct ceph_mds_session *s);
336
337extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
338 struct ceph_msg *msg, int mds);
339
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700340extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
Sage Weil2f2dc052009-10-06 11:31:09 -0700341extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700342extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
Sage Weil2f2dc052009-10-06 11:31:09 -0700343
344extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
345
346extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
347 struct inode *inode,
Sage Weil2f90b852011-07-26 11:28:25 -0700348 struct dentry *dn);
Sage Weil2f2dc052009-10-06 11:31:09 -0700349
Sage Weil167c9e32010-05-14 10:02:57 -0700350extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
Yan, Zheng54008392014-03-29 13:41:15 +0800351extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
352 struct inode *dir);
Sage Weil2f2dc052009-10-06 11:31:09 -0700353extern struct ceph_mds_request *
354ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
355extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
356 struct ceph_mds_request *req);
357extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
358 struct inode *dir,
359 struct ceph_mds_request *req);
360static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
361{
Sage Weil153c8e62009-12-07 12:31:09 -0800362 kref_get(&req->r_kref);
Sage Weil2f2dc052009-10-06 11:31:09 -0700363}
Sage Weil153c8e62009-12-07 12:31:09 -0800364extern void ceph_mdsc_release_request(struct kref *kref);
365static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
366{
367 kref_put(&req->r_kref, ceph_mdsc_release_request);
368}
Sage Weil2f2dc052009-10-06 11:31:09 -0700369
Sage Weil2b2300d2010-06-09 16:52:04 -0700370extern int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
Sage Weilee6b2722010-06-10 12:55:52 -0700371 struct ceph_mds_session *session);
Sage Weil3d7ded42010-06-09 16:47:10 -0700372extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
373 struct ceph_mds_session *session);
374
Sage Weil2f2dc052009-10-06 11:31:09 -0700375extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
376
377extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
378 int stop_on_nosnap);
379
380extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
381extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
382 struct inode *inode,
383 struct dentry *dentry, char action,
384 u32 seq);
385
386extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
387 struct ceph_msg *msg);
388
Yan, Zheng5d72d132013-11-24 14:33:01 +0800389extern struct ceph_mds_session *
390ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
Sage Weil154f42c2010-06-21 13:45:04 -0700391extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
392 struct ceph_mds_session *session);
393
Sage Weil2f2dc052009-10-06 11:31:09 -0700394#endif