blob: aabe563b54db308af80ee9fff18dd81610a9cf94 [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>
Sage Weil2f2dc052009-10-06 11:31:09 -070014
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
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070029struct ceph_fs_client;
Sage Weil2f2dc052009-10-06 11:31:09 -070030struct 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/*
Herb Shiu25933ab2010-12-01 14:14:38 -080045 * parsed info about an mds reply, including information about
46 * either: 1) the target inode and/or its parent directory and dentry,
47 * and directory contents (for readdir results), or
48 * 2) the file range lock info (for fcntl F_GETLK results).
Sage Weil2f2dc052009-10-06 11:31:09 -070049 */
50struct ceph_mds_reply_info_parsed {
51 struct ceph_mds_reply_head *head;
52
Herb Shiu25933ab2010-12-01 14:14:38 -080053 /* trace */
Sage Weil2f2dc052009-10-06 11:31:09 -070054 struct ceph_mds_reply_info_in diri, targeti;
55 struct ceph_mds_reply_dirfrag *dirfrag;
56 char *dname;
57 u32 dname_len;
58 struct ceph_mds_reply_lease *dlease;
59
Herb Shiu25933ab2010-12-01 14:14:38 -080060 /* extra */
61 union {
62 /* for fcntl F_GETLK results */
63 struct ceph_filelock *filelock_reply;
64
65 /* for readdir results */
66 struct {
67 struct ceph_mds_reply_dirfrag *dir_dir;
68 int dir_nr;
69 char **dir_dname;
70 u32 *dir_dname_len;
71 struct ceph_mds_reply_lease **dir_dlease;
72 struct ceph_mds_reply_info_in *dir_in;
73 u8 dir_complete, dir_end;
74 };
75 };
Sage Weil2f2dc052009-10-06 11:31:09 -070076
77 /* encoded blob describing snapshot contexts for certain
78 operations (e.g., open) */
79 void *snapblob;
80 int snapblob_len;
81};
82
83
84/*
85 * cap releases are batched and sent to the MDS en masse.
86 */
87#define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE - \
88 sizeof(struct ceph_mds_cap_release)) / \
89 sizeof(struct ceph_mds_cap_item))
90
91
92/*
93 * state associated with each MDS<->client session
94 */
95enum {
96 CEPH_MDS_SESSION_NEW = 1,
97 CEPH_MDS_SESSION_OPENING = 2,
98 CEPH_MDS_SESSION_OPEN = 3,
99 CEPH_MDS_SESSION_HUNG = 4,
100 CEPH_MDS_SESSION_CLOSING = 5,
101 CEPH_MDS_SESSION_RESTARTING = 6,
102 CEPH_MDS_SESSION_RECONNECTING = 7,
103};
104
105struct ceph_mds_session {
106 struct ceph_mds_client *s_mdsc;
107 int s_mds;
108 int s_state;
109 unsigned long s_ttl; /* time until mds kills us */
110 u64 s_seq; /* incoming msg seq # */
111 struct mutex s_mutex; /* serialize session messages */
112
Sage Weil2f2dc052009-10-06 11:31:09 -0700113 struct ceph_connection s_con;
114
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800115 struct ceph_authorizer *s_authorizer;
116 void *s_authorizer_buf, *s_authorizer_reply_buf;
117 size_t s_authorizer_buf_len, s_authorizer_reply_buf_len;
118
Sage Weil2f2dc052009-10-06 11:31:09 -0700119 /* protected by s_cap_lock */
120 spinlock_t s_cap_lock;
121 u32 s_cap_gen; /* inc each time we get mds stale msg */
122 unsigned long s_cap_ttl; /* when session caps expire */
123 struct list_head s_caps; /* all caps issued by this session */
124 int s_nr_caps, s_trim_caps;
125 int s_num_cap_releases;
126 struct list_head s_cap_releases; /* waiting cap_release messages */
127 struct list_head s_cap_releases_done; /* ready to send */
Sage Weil7c1332b2010-02-16 11:39:45 -0800128 struct ceph_cap *s_cap_iterator;
Sage Weil2f2dc052009-10-06 11:31:09 -0700129
130 /* protected by mutex */
131 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
132 struct list_head s_cap_snaps_flushing;
133 unsigned long s_renew_requested; /* last time we sent a renew req */
134 u64 s_renew_seq;
135
136 atomic_t s_ref;
137 struct list_head s_waiting; /* waiting requests */
138 struct list_head s_unsafe; /* unsafe requests */
139};
140
141/*
142 * modes of choosing which MDS to send a request to
143 */
144enum {
145 USE_ANY_MDS,
146 USE_RANDOM_MDS,
147 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
148};
149
150struct ceph_mds_request;
151struct ceph_mds_client;
152
153/*
154 * request completion callback
155 */
156typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
157 struct ceph_mds_request *req);
158
159/*
160 * an in-flight mds request
161 */
162struct ceph_mds_request {
163 u64 r_tid; /* transaction id */
Sage Weil44ca18f2010-02-15 12:08:46 -0800164 struct rb_node r_node;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700165 struct ceph_mds_client *r_mdsc;
Sage Weil2f2dc052009-10-06 11:31:09 -0700166
167 int r_op; /* mds op code */
168 int r_mds;
169
170 /* operation on what? */
171 struct inode *r_inode; /* arg1 */
172 struct dentry *r_dentry; /* arg1 */
173 struct dentry *r_old_dentry; /* arg2: rename from or link from */
174 char *r_path1, *r_path2;
175 struct ceph_vino r_ino1, r_ino2;
176
177 struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
178 struct inode *r_target_inode; /* resulting inode */
179
Sage Weilb4556392010-05-13 12:01:13 -0700180 struct mutex r_fill_mutex;
181
Sage Weil2f2dc052009-10-06 11:31:09 -0700182 union ceph_mds_request_args r_args;
183 int r_fmode; /* file mode, if expecting cap */
Sage Weilcb4276c2010-11-08 07:28:52 -0800184 uid_t r_uid;
185 gid_t r_gid;
Sage Weil2f2dc052009-10-06 11:31:09 -0700186
187 /* for choosing which mds to send this request to */
188 int r_direct_mode;
189 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
190 bool r_direct_is_hash; /* true if r_direct_hash is valid */
191
192 /* data payload is used for xattr ops */
193 struct page **r_pages;
194 int r_num_pages;
195 int r_data_len;
196
197 /* what caps shall we drop? */
198 int r_inode_drop, r_inode_unless;
199 int r_dentry_drop, r_dentry_unless;
200 int r_old_dentry_drop, r_old_dentry_unless;
201 struct inode *r_old_inode;
202 int r_old_inode_drop, r_old_inode_unless;
203
204 struct ceph_msg *r_request; /* original request */
Sage Weile979cf52010-07-15 14:58:39 -0700205 int r_request_release_offset;
Sage Weil2f2dc052009-10-06 11:31:09 -0700206 struct ceph_msg *r_reply;
207 struct ceph_mds_reply_info_parsed r_reply_info;
208 int r_err;
Sage Weil5b1daec2010-01-25 11:33:08 -0800209 bool r_aborted;
Sage Weil2f2dc052009-10-06 11:31:09 -0700210
211 unsigned long r_timeout; /* optional. jiffies */
212 unsigned long r_started; /* start time to measure timeout against */
213 unsigned long r_request_started; /* start time for mds request only,
214 used to measure lease durations */
215
216 /* link unsafe requests to parent directory, for fsync */
217 struct inode *r_unsafe_dir;
218 struct list_head r_unsafe_dir_item;
219
220 struct ceph_mds_session *r_session;
221
222 int r_attempts; /* resend attempts */
223 int r_num_fwd; /* number of forward attempts */
Sage Weil2f2dc052009-10-06 11:31:09 -0700224 int r_resend_mds; /* mds to resend to next, if any*/
Greg Farnume55b71f2010-06-22 15:58:01 -0700225 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
Sage Weil2f2dc052009-10-06 11:31:09 -0700226
Sage Weil153c8e62009-12-07 12:31:09 -0800227 struct kref r_kref;
Sage Weil2f2dc052009-10-06 11:31:09 -0700228 struct list_head r_wait;
229 struct completion r_completion;
230 struct completion r_safe_completion;
231 ceph_mds_request_callback_t r_callback;
232 struct list_head r_unsafe_item; /* per-session unsafe list item */
Sage Weile1518c72010-05-13 11:19:06 -0700233 bool r_got_unsafe, r_got_safe, r_got_result;
Sage Weil2f2dc052009-10-06 11:31:09 -0700234
235 bool r_did_prepopulate;
236 u32 r_readdir_offset;
237
238 struct ceph_cap_reservation r_caps_reservation;
239 int r_num_caps;
240};
241
242/*
243 * mds client state
244 */
245struct ceph_mds_client {
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700246 struct ceph_fs_client *fsc;
Sage Weil2f2dc052009-10-06 11:31:09 -0700247 struct mutex mutex; /* all nested structures */
248
249 struct ceph_mdsmap *mdsmap;
Sage Weilf3c60c52010-08-11 14:51:23 -0700250 struct completion safe_umount_waiters;
251 wait_queue_head_t session_close_wq;
Sage Weil2f2dc052009-10-06 11:31:09 -0700252 struct list_head waiting_for_map;
253
254 struct ceph_mds_session **sessions; /* NULL for mds if no session */
255 int max_sessions; /* len of s_mds_sessions */
256 int stopping; /* true if shutting down */
257
258 /*
259 * snap_rwsem will cover cap linkage into snaprealms, and
260 * realm snap contexts. (later, we can do per-realm snap
261 * contexts locks..) the empty list contains realms with no
262 * references (implying they contain no inodes with caps) that
263 * should be destroyed.
264 */
265 struct rw_semaphore snap_rwsem;
Sage Weila105f002010-02-15 14:37:55 -0800266 struct rb_root snap_realms;
Sage Weil2f2dc052009-10-06 11:31:09 -0700267 struct list_head snap_empty;
268 spinlock_t snap_empty_lock; /* protect snap_empty */
269
270 u64 last_tid; /* most recent mds request */
Sage Weil44ca18f2010-02-15 12:08:46 -0800271 struct rb_root request_tree; /* pending mds requests */
Sage Weil2f2dc052009-10-06 11:31:09 -0700272 struct delayed_work delayed_work; /* delayed work */
273 unsigned long last_renew_caps; /* last time we renewed our caps */
274 struct list_head cap_delay_list; /* caps with delayed release */
275 spinlock_t cap_delay_lock; /* protects cap_delay_list */
276 struct list_head snap_flush_list; /* cap_snaps ready to flush */
277 spinlock_t snap_flush_lock;
278
279 u64 cap_flush_seq;
280 struct list_head cap_dirty; /* inodes with dirty caps */
281 int num_cap_flushing; /* # caps we are flushing */
282 spinlock_t cap_dirty_lock; /* protects above items */
283 wait_queue_head_t cap_flushing_wq;
284
Yehuda Sadeh37151662010-06-17 16:16:12 -0700285 /*
286 * Cap reservations
287 *
288 * Maintain a global pool of preallocated struct ceph_caps, referenced
289 * by struct ceph_caps_reservations. This ensures that we preallocate
290 * memory needed to successfully process an MDS response. (If an MDS
291 * sends us cap information and we fail to process it, we will have
292 * problems due to the client and MDS being out of sync.)
293 *
294 * Reservations are 'owned' by a ceph_cap_reservation context.
295 */
296 spinlock_t caps_list_lock;
297 struct list_head caps_list; /* unused (reserved or
298 unreserved) */
299 int caps_total_count; /* total caps allocated */
300 int caps_use_count; /* in use */
301 int caps_reserve_count; /* unused, reserved */
302 int caps_avail_count; /* unused, unreserved */
303 int caps_min_count; /* keep at least this many
304 (unreserved) */
Sage Weil2f2dc052009-10-06 11:31:09 -0700305 spinlock_t dentry_lru_lock;
306 struct list_head dentry_lru;
307 int num_dentry;
308};
309
310extern const char *ceph_mds_op_name(int op);
311
312extern struct ceph_mds_session *
313__ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
314
315static inline struct ceph_mds_session *
316ceph_get_mds_session(struct ceph_mds_session *s)
317{
318 atomic_inc(&s->s_ref);
319 return s;
320}
321
322extern void ceph_put_mds_session(struct ceph_mds_session *s);
323
324extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
325 struct ceph_msg *msg, int mds);
326
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700327extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
Sage Weil2f2dc052009-10-06 11:31:09 -0700328extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700329extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
Sage Weil2f2dc052009-10-06 11:31:09 -0700330
331extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
332
333extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
334 struct inode *inode,
335 struct dentry *dn, int mask);
336
Sage Weil167c9e32010-05-14 10:02:57 -0700337extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
338
Sage Weil2f2dc052009-10-06 11:31:09 -0700339extern struct ceph_mds_request *
340ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
341extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
342 struct ceph_mds_request *req);
343extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
344 struct inode *dir,
345 struct ceph_mds_request *req);
346static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
347{
Sage Weil153c8e62009-12-07 12:31:09 -0800348 kref_get(&req->r_kref);
Sage Weil2f2dc052009-10-06 11:31:09 -0700349}
Sage Weil153c8e62009-12-07 12:31:09 -0800350extern void ceph_mdsc_release_request(struct kref *kref);
351static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
352{
353 kref_put(&req->r_kref, ceph_mdsc_release_request);
354}
Sage Weil2f2dc052009-10-06 11:31:09 -0700355
Sage Weil2b2300d2010-06-09 16:52:04 -0700356extern int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
Sage Weilee6b2722010-06-10 12:55:52 -0700357 struct ceph_mds_session *session);
Sage Weil3d7ded42010-06-09 16:47:10 -0700358extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
359 struct ceph_mds_session *session);
360
Sage Weil2f2dc052009-10-06 11:31:09 -0700361extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
362
363extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
364 int stop_on_nosnap);
365
366extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
367extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
368 struct inode *inode,
369 struct dentry *dentry, char action,
370 u32 seq);
371
372extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
373 struct ceph_msg *msg);
374
Sage Weil154f42c2010-06-21 13:45:04 -0700375extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
376 struct ceph_mds_session *session);
377
Sage Weil2f2dc052009-10-06 11:31:09 -0700378#endif