ceph: use new D_COMPLETE dentry flag
[linux-2.6.git] / fs / ceph / locks.c
1 #include <linux/ceph/ceph_debug.h>
2
3 #include <linux/file.h>
4 #include <linux/namei.h>
5
6 #include "super.h"
7 #include "mds_client.h"
8 #include <linux/ceph/pagelist.h>
9
10 /**
11  * Implement fcntl and flock locking functions.
12  */
13 static int ceph_lock_message(u8 lock_type, u16 operation, struct file *file,
14                              int cmd, u8 wait, struct file_lock *fl)
15 {
16         struct inode *inode = file->f_dentry->d_inode;
17         struct ceph_mds_client *mdsc =
18                 ceph_sb_to_client(inode->i_sb)->mdsc;
19         struct ceph_mds_request *req;
20         int err;
21         u64 length = 0;
22
23         req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
24         if (IS_ERR(req))
25                 return PTR_ERR(req);
26         req->r_inode = inode;
27         ihold(inode);
28
29         /* mds requires start and length rather than start and end */
30         if (LLONG_MAX == fl->fl_end)
31                 length = 0;
32         else
33                 length = fl->fl_end - fl->fl_start + 1;
34
35         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
36              "length: %llu, wait: %d, type: %d", (int)lock_type,
37              (int)operation, (u64)fl->fl_pid, fl->fl_start,
38              length, wait, fl->fl_type);
39
40         req->r_args.filelock_change.rule = lock_type;
41         req->r_args.filelock_change.type = cmd;
42         req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
43         /* This should be adjusted, but I'm not sure if
44            namespaces actually get id numbers*/
45         req->r_args.filelock_change.pid_namespace =
46                 cpu_to_le64((u64)(unsigned long)fl->fl_nspid);
47         req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
48         req->r_args.filelock_change.length = cpu_to_le64(length);
49         req->r_args.filelock_change.wait = wait;
50
51         err = ceph_mdsc_do_request(mdsc, inode, req);
52
53         if ( operation == CEPH_MDS_OP_GETFILELOCK){
54                 fl->fl_pid = le64_to_cpu(req->r_reply_info.filelock_reply->pid);
55                 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
56                         fl->fl_type = F_RDLCK;
57                 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
58                         fl->fl_type = F_WRLCK;
59                 else
60                         fl->fl_type = F_UNLCK;
61
62                 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
63                 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
64                                                  le64_to_cpu(req->r_reply_info.filelock_reply->length);
65                 if (length >= 1)
66                         fl->fl_end = length -1;
67                 else
68                         fl->fl_end = 0;
69
70         }
71         ceph_mdsc_put_request(req);
72         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
73              "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
74              (int)operation, (u64)fl->fl_pid, fl->fl_start,
75              length, wait, fl->fl_type, err);
76         return err;
77 }
78
79 /**
80  * Attempt to set an fcntl lock.
81  * For now, this just goes away to the server. Later it may be more awesome.
82  */
83 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
84 {
85         u8 lock_cmd;
86         int err;
87         u8 wait = 0;
88         u16 op = CEPH_MDS_OP_SETFILELOCK;
89
90         fl->fl_nspid = get_pid(task_tgid(current));
91         dout("ceph_lock, fl_pid:%d", fl->fl_pid);
92
93         /* set wait bit as appropriate, then make command as Ceph expects it*/
94         if (F_SETLKW == cmd)
95                 wait = 1;
96         if (F_GETLK == cmd)
97                 op = CEPH_MDS_OP_GETFILELOCK;
98
99         if (F_RDLCK == fl->fl_type)
100                 lock_cmd = CEPH_LOCK_SHARED;
101         else if (F_WRLCK == fl->fl_type)
102                 lock_cmd = CEPH_LOCK_EXCL;
103         else
104                 lock_cmd = CEPH_LOCK_UNLOCK;
105
106         err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
107         if (!err) {
108                 if ( op != CEPH_MDS_OP_GETFILELOCK ){
109                         dout("mds locked, locking locally");
110                         err = posix_lock_file(file, fl, NULL);
111                         if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
112                                 /* undo! This should only happen if
113                                  * the kernel detects local
114                                  * deadlock. */
115                                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
116                                                   CEPH_LOCK_UNLOCK, 0, fl);
117                                 dout("got %d on posix_lock_file, undid lock",
118                                      err);
119                         }
120                 }
121
122         } else if (err == -ERESTARTSYS) {
123                 dout("undoing lock\n");
124                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
125                                   CEPH_LOCK_UNLOCK, 0, fl);
126         }
127         return err;
128 }
129
130 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
131 {
132         u8 lock_cmd;
133         int err;
134         u8 wait = 1;
135
136         fl->fl_nspid = get_pid(task_tgid(current));
137         dout("ceph_flock, fl_pid:%d", fl->fl_pid);
138
139         /* set wait bit, then clear it out of cmd*/
140         if (cmd & LOCK_NB)
141                 wait = 0;
142         cmd = cmd & (LOCK_SH | LOCK_EX | LOCK_UN);
143         /* set command sequence that Ceph wants to see:
144            shared lock, exclusive lock, or unlock */
145         if (LOCK_SH == cmd)
146                 lock_cmd = CEPH_LOCK_SHARED;
147         else if (LOCK_EX == cmd)
148                 lock_cmd = CEPH_LOCK_EXCL;
149         else
150                 lock_cmd = CEPH_LOCK_UNLOCK;
151
152         err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
153                                 file, lock_cmd, wait, fl);
154         if (!err) {
155                 err = flock_lock_file_wait(file, fl);
156                 if (err) {
157                         ceph_lock_message(CEPH_LOCK_FLOCK,
158                                           CEPH_MDS_OP_SETFILELOCK,
159                                           file, CEPH_LOCK_UNLOCK, 0, fl);
160                         dout("got %d on flock_lock_file_wait, undid lock", err);
161                 }
162         } else if (err == -ERESTARTSYS) {
163                 dout("undoing lock\n");
164                 ceph_lock_message(CEPH_LOCK_FLOCK,
165                                   CEPH_MDS_OP_SETFILELOCK,
166                                   file, CEPH_LOCK_UNLOCK, 0, fl);
167         }
168         return err;
169 }
170
171 /**
172  * Must be called with BKL already held. Fills in the passed
173  * counter variables, so you can prepare pagelist metadata before calling
174  * ceph_encode_locks.
175  */
176 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
177 {
178         struct file_lock *lock;
179
180         *fcntl_count = 0;
181         *flock_count = 0;
182
183         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
184                 if (lock->fl_flags & FL_POSIX)
185                         ++(*fcntl_count);
186                 else if (lock->fl_flags & FL_FLOCK)
187                         ++(*flock_count);
188         }
189         dout("counted %d flock locks and %d fcntl locks",
190              *flock_count, *fcntl_count);
191 }
192
193 /**
194  * Encode the flock and fcntl locks for the given inode into the pagelist.
195  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
196  * sequential flock locks.
197  * Must be called with lock_flocks() already held.
198  * If we encounter more of a specific lock type than expected,
199  * we return the value 1.
200  */
201 int ceph_encode_locks(struct inode *inode, struct ceph_pagelist *pagelist,
202                       int num_fcntl_locks, int num_flock_locks)
203 {
204         struct file_lock *lock;
205         struct ceph_filelock cephlock;
206         int err = 0;
207         int seen_fcntl = 0;
208         int seen_flock = 0;
209
210         dout("encoding %d flock and %d fcntl locks", num_flock_locks,
211              num_fcntl_locks);
212         err = ceph_pagelist_append(pagelist, &num_fcntl_locks, sizeof(u32));
213         if (err)
214                 goto fail;
215         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
216                 if (lock->fl_flags & FL_POSIX) {
217                         ++seen_fcntl;
218                         if (seen_fcntl > num_fcntl_locks) {
219                                 err = -ENOSPC;
220                                 goto fail;
221                         }
222                         err = lock_to_ceph_filelock(lock, &cephlock);
223                         if (err)
224                                 goto fail;
225                         err = ceph_pagelist_append(pagelist, &cephlock,
226                                            sizeof(struct ceph_filelock));
227                 }
228                 if (err)
229                         goto fail;
230         }
231
232         err = ceph_pagelist_append(pagelist, &num_flock_locks, sizeof(u32));
233         if (err)
234                 goto fail;
235         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
236                 if (lock->fl_flags & FL_FLOCK) {
237                         ++seen_flock;
238                         if (seen_flock > num_flock_locks) {
239                                 err = -ENOSPC;
240                                 goto fail;
241                         }
242                         err = lock_to_ceph_filelock(lock, &cephlock);
243                         if (err)
244                                 goto fail;
245                         err = ceph_pagelist_append(pagelist, &cephlock,
246                                            sizeof(struct ceph_filelock));
247                 }
248                 if (err)
249                         goto fail;
250         }
251 fail:
252         return err;
253 }
254
255 /*
256  * Given a pointer to a lock, convert it to a ceph filelock
257  */
258 int lock_to_ceph_filelock(struct file_lock *lock,
259                           struct ceph_filelock *cephlock)
260 {
261         int err = 0;
262
263         cephlock->start = cpu_to_le64(lock->fl_start);
264         cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
265         cephlock->client = cpu_to_le64(0);
266         cephlock->pid = cpu_to_le64(lock->fl_pid);
267         cephlock->pid_namespace =
268                 cpu_to_le64((u64)(unsigned long)lock->fl_nspid);
269
270         switch (lock->fl_type) {
271         case F_RDLCK:
272                 cephlock->type = CEPH_LOCK_SHARED;
273                 break;
274         case F_WRLCK:
275                 cephlock->type = CEPH_LOCK_EXCL;
276                 break;
277         case F_UNLCK:
278                 cephlock->type = CEPH_LOCK_UNLOCK;
279                 break;
280         default:
281                 dout("Have unknown lock type %d", lock->fl_type);
282                 err = -EINVAL;
283         }
284
285         return err;
286 }