ceph: fix xattr rbtree search
[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 = igrab(inode);
27
28         /* mds requires start and length rather than start and end */
29         if (LLONG_MAX == fl->fl_end)
30                 length = 0;
31         else
32                 length = fl->fl_end - fl->fl_start + 1;
33
34         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
35              "length: %llu, wait: %d, type`: %d", (int)lock_type,
36              (int)operation, (u64)fl->fl_pid, fl->fl_start,
37              length, wait, fl->fl_type);
38
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 the kernel detects
113                                  * local deadlock. */
114                                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
115                                                   CEPH_LOCK_UNLOCK, 0, fl);
116                                 dout("got %d on posix_lock_file, undid lock", err);
117                         }
118                 }
119
120         } else {
121                 dout("mds returned error code %d", err);
122         }
123         return err;
124 }
125
126 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
127 {
128         u8 lock_cmd;
129         int err;
130         u8 wait = 1;
131
132         fl->fl_nspid = get_pid(task_tgid(current));
133         dout("ceph_flock, fl_pid:%d", fl->fl_pid);
134
135         /* set wait bit, then clear it out of cmd*/
136         if (cmd & LOCK_NB)
137                 wait = 0;
138         cmd = cmd & (LOCK_SH | LOCK_EX | LOCK_UN);
139         /* set command sequence that Ceph wants to see:
140            shared lock, exclusive lock, or unlock */
141         if (LOCK_SH == cmd)
142                 lock_cmd = CEPH_LOCK_SHARED;
143         else if (LOCK_EX == cmd)
144                 lock_cmd = CEPH_LOCK_EXCL;
145         else
146                 lock_cmd = CEPH_LOCK_UNLOCK;
147
148         err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
149                                 file, lock_cmd, wait, fl);
150         if (!err) {
151                 err = flock_lock_file_wait(file, fl);
152                 if (err) {
153                         ceph_lock_message(CEPH_LOCK_FLOCK,
154                                           CEPH_MDS_OP_SETFILELOCK,
155                                           file, CEPH_LOCK_UNLOCK, 0, fl);
156                         dout("got %d on flock_lock_file_wait, undid lock", err);
157                 }
158         } else {
159                 dout("mds error code %d", err);
160         }
161         return err;
162 }
163
164 /**
165  * Must be called with BKL already held. Fills in the passed
166  * counter variables, so you can prepare pagelist metadata before calling
167  * ceph_encode_locks.
168  */
169 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
170 {
171         struct file_lock *lock;
172
173         *fcntl_count = 0;
174         *flock_count = 0;
175
176         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
177                 if (lock->fl_flags & FL_POSIX)
178                         ++(*fcntl_count);
179                 else if (lock->fl_flags & FL_FLOCK)
180                         ++(*flock_count);
181         }
182         dout("counted %d flock locks and %d fcntl locks",
183              *flock_count, *fcntl_count);
184 }
185
186 /**
187  * Encode the flock and fcntl locks for the given inode into the pagelist.
188  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
189  * sequential flock locks.
190  * Must be called with lock_flocks() already held.
191  * If we encounter more of a specific lock type than expected,
192  * we return the value 1.
193  */
194 int ceph_encode_locks(struct inode *inode, struct ceph_pagelist *pagelist,
195                       int num_fcntl_locks, int num_flock_locks)
196 {
197         struct file_lock *lock;
198         struct ceph_filelock cephlock;
199         int err = 0;
200         int seen_fcntl = 0;
201         int seen_flock = 0;
202
203         dout("encoding %d flock and %d fcntl locks", num_flock_locks,
204              num_fcntl_locks);
205         err = ceph_pagelist_append(pagelist, &num_fcntl_locks, sizeof(u32));
206         if (err)
207                 goto fail;
208         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
209                 if (lock->fl_flags & FL_POSIX) {
210                         ++seen_fcntl;
211                         if (seen_fcntl > num_fcntl_locks) {
212                                 err = -ENOSPC;
213                                 goto fail;
214                         }
215                         err = lock_to_ceph_filelock(lock, &cephlock);
216                         if (err)
217                                 goto fail;
218                         err = ceph_pagelist_append(pagelist, &cephlock,
219                                            sizeof(struct ceph_filelock));
220                 }
221                 if (err)
222                         goto fail;
223         }
224
225         err = ceph_pagelist_append(pagelist, &num_flock_locks, sizeof(u32));
226         if (err)
227                 goto fail;
228         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
229                 if (lock->fl_flags & FL_FLOCK) {
230                         ++seen_flock;
231                         if (seen_flock > num_flock_locks) {
232                                 err = -ENOSPC;
233                                 goto fail;
234                         }
235                         err = lock_to_ceph_filelock(lock, &cephlock);
236                         if (err)
237                                 goto fail;
238                         err = ceph_pagelist_append(pagelist, &cephlock,
239                                            sizeof(struct ceph_filelock));
240                 }
241                 if (err)
242                         goto fail;
243         }
244 fail:
245         return err;
246 }
247
248 /*
249  * Given a pointer to a lock, convert it to a ceph filelock
250  */
251 int lock_to_ceph_filelock(struct file_lock *lock,
252                           struct ceph_filelock *cephlock)
253 {
254         int err = 0;
255
256         cephlock->start = cpu_to_le64(lock->fl_start);
257         cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
258         cephlock->client = cpu_to_le64(0);
259         cephlock->pid = cpu_to_le64(lock->fl_pid);
260         cephlock->pid_namespace =
261                 cpu_to_le64((u64)(unsigned long)lock->fl_nspid);
262
263         switch (lock->fl_type) {
264         case F_RDLCK:
265                 cephlock->type = CEPH_LOCK_SHARED;
266                 break;
267         case F_WRLCK:
268                 cephlock->type = CEPH_LOCK_EXCL;
269                 break;
270         case F_UNLCK:
271                 cephlock->type = CEPH_LOCK_UNLOCK;
272                 break;
273         default:
274                 dout("Have unknown lock type %d", lock->fl_type);
275                 err = -EINVAL;
276         }
277
278         return err;
279 }