Gitiles
Code Review Sign In
review.shift-gmbh.com / SHIFTPHONES / mainline / linux / 4e7a5dcd1bbab6560fbc8ada29a840e7a20ed7bc / . / fs / ceph / mds_client.h
blob: 9faa1b2f79a7c4d63738a4af870e7d5779512715 [file] [log] [blame]
Sage Weil2f2dc052009-10-06 11:31:09 -0700[diff] [blame]1#ifndef _FS_CEPH_MDS_CLIENT_H
2#define _FS_CEPH_MDS_CLIENT_H
3
4#include <linux/completion.h>
5#include <linux/list.h>
6#include <linux/mutex.h>
7#include <linux/radix-tree.h>
8#include <linux/spinlock.h>
9
10#include "types.h"
11#include "messenger.h"
12#include "mdsmap.h"
13
14/*
15 * Some lock dependencies:
16 *
17 * session->s_mutex
18 * mdsc->mutex
19 *
20 * mdsc->snap_rwsem
21 *
22 * inode->i_lock
23 * mdsc->snap_flush_lock
24 * mdsc->cap_delay_lock
25 *
26 */
27
28struct ceph_client;
29struct ceph_cap;
30
31/*
32 * parsed info about a single inode. pointers are into the encoded
33 * on-wire structures within the mds reply message payload.
34 */
35struct ceph_mds_reply_info_in {
36 struct ceph_mds_reply_inode *in;
37 u32 symlink_len;
38 char *symlink;
39 u32 xattr_len;
40 char *xattr_data;
41};
42
43/*
44 * parsed info about an mds reply, including information about the
45 * target inode and/or its parent directory and dentry, and directory
46 * contents (for readdir results).
47 */
48struct ceph_mds_reply_info_parsed {
49 struct ceph_mds_reply_head *head;
50
51 struct ceph_mds_reply_info_in diri, targeti;
52 struct ceph_mds_reply_dirfrag *dirfrag;
53 char *dname;
54 u32 dname_len;
55 struct ceph_mds_reply_lease *dlease;
56
57 struct ceph_mds_reply_dirfrag *dir_dir;
58 int dir_nr;
59 char **dir_dname;
60 u32 *dir_dname_len;
61 struct ceph_mds_reply_lease **dir_dlease;
62 struct ceph_mds_reply_info_in *dir_in;
63 u8 dir_complete, dir_end;
64
65 /* encoded blob describing snapshot contexts for certain
66 operations (e.g., open) */
67 void *snapblob;
68 int snapblob_len;
69};
70
71
72/*
73 * cap releases are batched and sent to the MDS en masse.
74 */
75#define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE - \
76 sizeof(struct ceph_mds_cap_release)) / \
77 sizeof(struct ceph_mds_cap_item))
78
79
80/*
81 * state associated with each MDS<->client session
82 */
83enum {
84 CEPH_MDS_SESSION_NEW = 1,
85 CEPH_MDS_SESSION_OPENING = 2,
86 CEPH_MDS_SESSION_OPEN = 3,
87 CEPH_MDS_SESSION_HUNG = 4,
88 CEPH_MDS_SESSION_CLOSING = 5,
89 CEPH_MDS_SESSION_RESTARTING = 6,
90 CEPH_MDS_SESSION_RECONNECTING = 7,
91};
92
93struct ceph_mds_session {
94 struct ceph_mds_client *s_mdsc;
95 int s_mds;
96 int s_state;
97 unsigned long s_ttl; /* time until mds kills us */
98 u64 s_seq; /* incoming msg seq # */
99 struct mutex s_mutex; /* serialize session messages */
100
Sage Weil2f2dc052009-10-06 11:31:09 -0700[diff] [blame]101 struct ceph_connection s_con;
102
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800[diff] [blame^]103 struct ceph_authorizer *s_authorizer;
104 void *s_authorizer_buf, *s_authorizer_reply_buf;
105 size_t s_authorizer_buf_len, s_authorizer_reply_buf_len;
106
Sage Weil2f2dc052009-10-06 11:31:09 -0700[diff] [blame]107 /* protected by s_cap_lock */
108 spinlock_t s_cap_lock;
109 u32 s_cap_gen; /* inc each time we get mds stale msg */
110 unsigned long s_cap_ttl; /* when session caps expire */
111 struct list_head s_caps; /* all caps issued by this session */
112 int s_nr_caps, s_trim_caps;
113 int s_num_cap_releases;
114 struct list_head s_cap_releases; /* waiting cap_release messages */
115 struct list_head s_cap_releases_done; /* ready to send */
116
117 /* protected by mutex */
118 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
119 struct list_head s_cap_snaps_flushing;
120 unsigned long s_renew_requested; /* last time we sent a renew req */
121 u64 s_renew_seq;
122
123 atomic_t s_ref;
124 struct list_head s_waiting; /* waiting requests */
125 struct list_head s_unsafe; /* unsafe requests */
126};
127
128/*
129 * modes of choosing which MDS to send a request to
130 */
131enum {
132 USE_ANY_MDS,
133 USE_RANDOM_MDS,
134 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
135};
136
137struct ceph_mds_request;
138struct ceph_mds_client;
139
140/*
141 * request completion callback
142 */
143typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
144 struct ceph_mds_request *req);
145
146/*
147 * an in-flight mds request
148 */
149struct ceph_mds_request {
150 u64 r_tid; /* transaction id */
151
152 int r_op; /* mds op code */
153 int r_mds;
154
155 /* operation on what? */
156 struct inode *r_inode; /* arg1 */
157 struct dentry *r_dentry; /* arg1 */
158 struct dentry *r_old_dentry; /* arg2: rename from or link from */
159 char *r_path1, *r_path2;
160 struct ceph_vino r_ino1, r_ino2;
161
162 struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
163 struct inode *r_target_inode; /* resulting inode */
164
165 union ceph_mds_request_args r_args;
166 int r_fmode; /* file mode, if expecting cap */
167
168 /* for choosing which mds to send this request to */
169 int r_direct_mode;
170 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
171 bool r_direct_is_hash; /* true if r_direct_hash is valid */
172
173 /* data payload is used for xattr ops */
174 struct page **r_pages;
175 int r_num_pages;
176 int r_data_len;
177
178 /* what caps shall we drop? */
179 int r_inode_drop, r_inode_unless;
180 int r_dentry_drop, r_dentry_unless;
181 int r_old_dentry_drop, r_old_dentry_unless;
182 struct inode *r_old_inode;
183 int r_old_inode_drop, r_old_inode_unless;
184
185 struct ceph_msg *r_request; /* original request */
186 struct ceph_msg *r_reply;
187 struct ceph_mds_reply_info_parsed r_reply_info;
188 int r_err;
189
190 unsigned long r_timeout; /* optional. jiffies */
191 unsigned long r_started; /* start time to measure timeout against */
192 unsigned long r_request_started; /* start time for mds request only,
193 used to measure lease durations */
194
195 /* link unsafe requests to parent directory, for fsync */
196 struct inode *r_unsafe_dir;
197 struct list_head r_unsafe_dir_item;
198
199 struct ceph_mds_session *r_session;
200
201 int r_attempts; /* resend attempts */
202 int r_num_fwd; /* number of forward attempts */
203 int r_num_stale;
204 int r_resend_mds; /* mds to resend to next, if any*/
205
206 atomic_t r_ref;
207 struct list_head r_wait;
208 struct completion r_completion;
209 struct completion r_safe_completion;
210 ceph_mds_request_callback_t r_callback;
211 struct list_head r_unsafe_item; /* per-session unsafe list item */
212 bool r_got_unsafe, r_got_safe;
213
214 bool r_did_prepopulate;
215 u32 r_readdir_offset;
216
217 struct ceph_cap_reservation r_caps_reservation;
218 int r_num_caps;
219};
220
221/*
222 * mds client state
223 */
224struct ceph_mds_client {
225 struct ceph_client *client;
226 struct mutex mutex; /* all nested structures */
227
228 struct ceph_mdsmap *mdsmap;
229 struct completion safe_umount_waiters, session_close_waiters;
230 struct list_head waiting_for_map;
231
232 struct ceph_mds_session **sessions; /* NULL for mds if no session */
233 int max_sessions; /* len of s_mds_sessions */
234 int stopping; /* true if shutting down */
235
236 /*
237 * snap_rwsem will cover cap linkage into snaprealms, and
238 * realm snap contexts. (later, we can do per-realm snap
239 * contexts locks..) the empty list contains realms with no
240 * references (implying they contain no inodes with caps) that
241 * should be destroyed.
242 */
243 struct rw_semaphore snap_rwsem;
244 struct radix_tree_root snap_realms;
245 struct list_head snap_empty;
246 spinlock_t snap_empty_lock; /* protect snap_empty */
247
248 u64 last_tid; /* most recent mds request */
249 struct radix_tree_root request_tree; /* pending mds requests */
250 struct delayed_work delayed_work; /* delayed work */
251 unsigned long last_renew_caps; /* last time we renewed our caps */
252 struct list_head cap_delay_list; /* caps with delayed release */
253 spinlock_t cap_delay_lock; /* protects cap_delay_list */
254 struct list_head snap_flush_list; /* cap_snaps ready to flush */
255 spinlock_t snap_flush_lock;
256
257 u64 cap_flush_seq;
258 struct list_head cap_dirty; /* inodes with dirty caps */
259 int num_cap_flushing; /* # caps we are flushing */
260 spinlock_t cap_dirty_lock; /* protects above items */
261 wait_queue_head_t cap_flushing_wq;
262
Sage Weil039934b2009-11-12 15:05:52 -0800[diff] [blame]263#ifdef CONFIG_DEBUG_FS
Sage Weil2f2dc052009-10-06 11:31:09 -0700[diff] [blame]264 struct dentry *debugfs_file;
Sage Weil039934b2009-11-12 15:05:52 -0800[diff] [blame]265#endif
Sage Weil2f2dc052009-10-06 11:31:09 -0700[diff] [blame]266
267 spinlock_t dentry_lru_lock;
268 struct list_head dentry_lru;
269 int num_dentry;
270};
271
272extern const char *ceph_mds_op_name(int op);
273
274extern struct ceph_mds_session *
275__ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
276
277static inline struct ceph_mds_session *
278ceph_get_mds_session(struct ceph_mds_session *s)
279{
280 atomic_inc(&s->s_ref);
281 return s;
282}
283
284extern void ceph_put_mds_session(struct ceph_mds_session *s);
285
286extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
287 struct ceph_msg *msg, int mds);
288
Sage Weil5f44f142009-11-18 14:52:18 -0800[diff] [blame]289extern int ceph_mdsc_init(struct ceph_mds_client *mdsc,
Sage Weil2f2dc052009-10-06 11:31:09 -0700[diff] [blame]290 struct ceph_client *client);
291extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
292extern void ceph_mdsc_stop(struct ceph_mds_client *mdsc);
293
294extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
295
296extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
297 struct inode *inode,
298 struct dentry *dn, int mask);
299
300extern struct ceph_mds_request *
301ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
302extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
303 struct ceph_mds_request *req);
304extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
305 struct inode *dir,
306 struct ceph_mds_request *req);
307static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
308{
309 atomic_inc(&req->r_ref);
310}
311extern void ceph_mdsc_put_request(struct ceph_mds_request *req);
312
313extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
314
315extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
316 int stop_on_nosnap);
317
318extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
319extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
320 struct inode *inode,
321 struct dentry *dentry, char action,
322 u32 seq);
323
324extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
325 struct ceph_msg *msg);
326
327#endif