blob: 961cc6f65878df8bd536683a12911581f8cca036 [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
11#include "types.h"
12#include "messenger.h"
13#include "mdsmap.h"
14
15/*
16 * Some lock dependencies:
17 *
18 * session->s_mutex
19 * mdsc->mutex
20 *
21 * mdsc->snap_rwsem
22 *
23 * inode->i_lock
24 * mdsc->snap_flush_lock
25 * mdsc->cap_delay_lock
26 *
27 */
28
29struct ceph_client;
30struct ceph_cap;
31
32/*
33 * parsed info about a single inode. pointers are into the encoded
34 * on-wire structures within the mds reply message payload.
35 */
36struct ceph_mds_reply_info_in {
37 struct ceph_mds_reply_inode *in;
38 u32 symlink_len;
39 char *symlink;
40 u32 xattr_len;
41 char *xattr_data;
42};
43
44/*
45 * parsed info about an mds reply, including information about the
46 * target inode and/or its parent directory and dentry, and directory
47 * contents (for readdir results).
48 */
49struct ceph_mds_reply_info_parsed {
50 struct ceph_mds_reply_head *head;
51
52 struct ceph_mds_reply_info_in diri, targeti;
53 struct ceph_mds_reply_dirfrag *dirfrag;
54 char *dname;
55 u32 dname_len;
56 struct ceph_mds_reply_lease *dlease;
57
58 struct ceph_mds_reply_dirfrag *dir_dir;
59 int dir_nr;
60 char **dir_dname;
61 u32 *dir_dname_len;
62 struct ceph_mds_reply_lease **dir_dlease;
63 struct ceph_mds_reply_info_in *dir_in;
64 u8 dir_complete, dir_end;
65
66 /* encoded blob describing snapshot contexts for certain
67 operations (e.g., open) */
68 void *snapblob;
69 int snapblob_len;
70};
71
72
73/*
74 * cap releases are batched and sent to the MDS en masse.
75 */
76#define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE - \
77 sizeof(struct ceph_mds_cap_release)) / \
78 sizeof(struct ceph_mds_cap_item))
79
80
81/*
82 * state associated with each MDS<->client session
83 */
84enum {
85 CEPH_MDS_SESSION_NEW = 1,
86 CEPH_MDS_SESSION_OPENING = 2,
87 CEPH_MDS_SESSION_OPEN = 3,
88 CEPH_MDS_SESSION_HUNG = 4,
89 CEPH_MDS_SESSION_CLOSING = 5,
90 CEPH_MDS_SESSION_RESTARTING = 6,
91 CEPH_MDS_SESSION_RECONNECTING = 7,
92};
93
94struct ceph_mds_session {
95 struct ceph_mds_client *s_mdsc;
96 int s_mds;
97 int s_state;
98 unsigned long s_ttl; /* time until mds kills us */
99 u64 s_seq; /* incoming msg seq # */
100 struct mutex s_mutex; /* serialize session messages */
101
Sage Weil2f2dc052009-10-06 11:31:09 -0700102 struct ceph_connection s_con;
103
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800104 struct ceph_authorizer *s_authorizer;
105 void *s_authorizer_buf, *s_authorizer_reply_buf;
106 size_t s_authorizer_buf_len, s_authorizer_reply_buf_len;
107
Sage Weil2f2dc052009-10-06 11:31:09 -0700108 /* protected by s_cap_lock */
109 spinlock_t s_cap_lock;
110 u32 s_cap_gen; /* inc each time we get mds stale msg */
111 unsigned long s_cap_ttl; /* when session caps expire */
112 struct list_head s_caps; /* all caps issued by this session */
113 int s_nr_caps, s_trim_caps;
114 int s_num_cap_releases;
115 struct list_head s_cap_releases; /* waiting cap_release messages */
116 struct list_head s_cap_releases_done; /* ready to send */
Sage Weil7c1332b2010-02-16 11:39:45 -0800117 struct ceph_cap *s_cap_iterator;
Sage Weil2f2dc052009-10-06 11:31:09 -0700118
119 /* protected by mutex */
120 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
121 struct list_head s_cap_snaps_flushing;
122 unsigned long s_renew_requested; /* last time we sent a renew req */
123 u64 s_renew_seq;
124
125 atomic_t s_ref;
126 struct list_head s_waiting; /* waiting requests */
127 struct list_head s_unsafe; /* unsafe requests */
128};
129
130/*
131 * modes of choosing which MDS to send a request to
132 */
133enum {
134 USE_ANY_MDS,
135 USE_RANDOM_MDS,
136 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
137};
138
139struct ceph_mds_request;
140struct ceph_mds_client;
141
142/*
143 * request completion callback
144 */
145typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
146 struct ceph_mds_request *req);
147
148/*
149 * an in-flight mds request
150 */
151struct ceph_mds_request {
152 u64 r_tid; /* transaction id */
Sage Weil44ca18f2010-02-15 12:08:46 -0800153 struct rb_node r_node;
Sage Weil2f2dc052009-10-06 11:31:09 -0700154
155 int r_op; /* mds op code */
156 int r_mds;
157
158 /* operation on what? */
159 struct inode *r_inode; /* arg1 */
160 struct dentry *r_dentry; /* arg1 */
161 struct dentry *r_old_dentry; /* arg2: rename from or link from */
162 char *r_path1, *r_path2;
163 struct ceph_vino r_ino1, r_ino2;
164
165 struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
166 struct inode *r_target_inode; /* resulting inode */
167
168 union ceph_mds_request_args r_args;
169 int r_fmode; /* file mode, if expecting cap */
170
171 /* for choosing which mds to send this request to */
172 int r_direct_mode;
173 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
174 bool r_direct_is_hash; /* true if r_direct_hash is valid */
175
176 /* data payload is used for xattr ops */
177 struct page **r_pages;
178 int r_num_pages;
179 int r_data_len;
180
181 /* what caps shall we drop? */
182 int r_inode_drop, r_inode_unless;
183 int r_dentry_drop, r_dentry_unless;
184 int r_old_dentry_drop, r_old_dentry_unless;
185 struct inode *r_old_inode;
186 int r_old_inode_drop, r_old_inode_unless;
187
188 struct ceph_msg *r_request; /* original request */
189 struct ceph_msg *r_reply;
190 struct ceph_mds_reply_info_parsed r_reply_info;
191 int r_err;
Sage Weil5b1daec2010-01-25 11:33:08 -0800192 bool r_aborted;
Sage Weil2f2dc052009-10-06 11:31:09 -0700193
194 unsigned long r_timeout; /* optional. jiffies */
195 unsigned long r_started; /* start time to measure timeout against */
196 unsigned long r_request_started; /* start time for mds request only,
197 used to measure lease durations */
198
199 /* link unsafe requests to parent directory, for fsync */
200 struct inode *r_unsafe_dir;
201 struct list_head r_unsafe_dir_item;
202
203 struct ceph_mds_session *r_session;
204
205 int r_attempts; /* resend attempts */
206 int r_num_fwd; /* number of forward attempts */
207 int r_num_stale;
208 int r_resend_mds; /* mds to resend to next, if any*/
209
Sage Weil153c8e62009-12-07 12:31:09 -0800210 struct kref r_kref;
Sage Weil2f2dc052009-10-06 11:31:09 -0700211 struct list_head r_wait;
212 struct completion r_completion;
213 struct completion r_safe_completion;
214 ceph_mds_request_callback_t r_callback;
215 struct list_head r_unsafe_item; /* per-session unsafe list item */
216 bool r_got_unsafe, r_got_safe;
217
218 bool r_did_prepopulate;
219 u32 r_readdir_offset;
220
221 struct ceph_cap_reservation r_caps_reservation;
222 int r_num_caps;
223};
224
225/*
226 * mds client state
227 */
228struct ceph_mds_client {
229 struct ceph_client *client;
230 struct mutex mutex; /* all nested structures */
231
232 struct ceph_mdsmap *mdsmap;
233 struct completion safe_umount_waiters, session_close_waiters;
234 struct list_head waiting_for_map;
235
236 struct ceph_mds_session **sessions; /* NULL for mds if no session */
237 int max_sessions; /* len of s_mds_sessions */
238 int stopping; /* true if shutting down */
239
240 /*
241 * snap_rwsem will cover cap linkage into snaprealms, and
242 * realm snap contexts. (later, we can do per-realm snap
243 * contexts locks..) the empty list contains realms with no
244 * references (implying they contain no inodes with caps) that
245 * should be destroyed.
246 */
247 struct rw_semaphore snap_rwsem;
Sage Weila105f002010-02-15 14:37:55 -0800248 struct rb_root snap_realms;
Sage Weil2f2dc052009-10-06 11:31:09 -0700249 struct list_head snap_empty;
250 spinlock_t snap_empty_lock; /* protect snap_empty */
251
252 u64 last_tid; /* most recent mds request */
Sage Weil44ca18f2010-02-15 12:08:46 -0800253 struct rb_root request_tree; /* pending mds requests */
Sage Weil2f2dc052009-10-06 11:31:09 -0700254 struct delayed_work delayed_work; /* delayed work */
255 unsigned long last_renew_caps; /* last time we renewed our caps */
256 struct list_head cap_delay_list; /* caps with delayed release */
257 spinlock_t cap_delay_lock; /* protects cap_delay_list */
258 struct list_head snap_flush_list; /* cap_snaps ready to flush */
259 spinlock_t snap_flush_lock;
260
261 u64 cap_flush_seq;
262 struct list_head cap_dirty; /* inodes with dirty caps */
263 int num_cap_flushing; /* # caps we are flushing */
264 spinlock_t cap_dirty_lock; /* protects above items */
265 wait_queue_head_t cap_flushing_wq;
266
Sage Weil039934b2009-11-12 15:05:52 -0800267#ifdef CONFIG_DEBUG_FS
Sage Weil2f2dc052009-10-06 11:31:09 -0700268 struct dentry *debugfs_file;
Sage Weil039934b2009-11-12 15:05:52 -0800269#endif
Sage Weil2f2dc052009-10-06 11:31:09 -0700270
271 spinlock_t dentry_lru_lock;
272 struct list_head dentry_lru;
273 int num_dentry;
274};
275
276extern const char *ceph_mds_op_name(int op);
277
278extern struct ceph_mds_session *
279__ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
280
281static inline struct ceph_mds_session *
282ceph_get_mds_session(struct ceph_mds_session *s)
283{
284 atomic_inc(&s->s_ref);
285 return s;
286}
287
288extern void ceph_put_mds_session(struct ceph_mds_session *s);
289
290extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
291 struct ceph_msg *msg, int mds);
292
Sage Weil5f44f142009-11-18 14:52:18 -0800293extern int ceph_mdsc_init(struct ceph_mds_client *mdsc,
Sage Weil2f2dc052009-10-06 11:31:09 -0700294 struct ceph_client *client);
295extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
296extern void ceph_mdsc_stop(struct ceph_mds_client *mdsc);
297
298extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
299
300extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
301 struct inode *inode,
302 struct dentry *dn, int mask);
303
304extern struct ceph_mds_request *
305ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
306extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
307 struct ceph_mds_request *req);
308extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
309 struct inode *dir,
310 struct ceph_mds_request *req);
311static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
312{
Sage Weil153c8e62009-12-07 12:31:09 -0800313 kref_get(&req->r_kref);
Sage Weil2f2dc052009-10-06 11:31:09 -0700314}
Sage Weil153c8e62009-12-07 12:31:09 -0800315extern void ceph_mdsc_release_request(struct kref *kref);
316static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
317{
318 kref_put(&req->r_kref, ceph_mdsc_release_request);
319}
Sage Weil2f2dc052009-10-06 11:31:09 -0700320
321extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
322
323extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
324 int stop_on_nosnap);
325
326extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
327extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
328 struct inode *inode,
329 struct dentry *dentry, char action,
330 u32 seq);
331
332extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
333 struct ceph_msg *msg);
334
335#endif