blob: ae6d14e82b0f439153b5e62adfd4d0c2ebed73a1 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Greg Farnum40819f62010-08-02 15:34:23 -07002
3#include <linux/file.h>
4#include <linux/namei.h>
5
6#include "super.h"
7#include "mds_client.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07008#include <linux/ceph/pagelist.h>
Greg Farnum40819f62010-08-02 15:34:23 -07009
10/**
11 * Implement fcntl and flock locking functions.
12 */
13static int ceph_lock_message(u8 lock_type, u16 operation, struct file *file,
Herb Shiu637ae8d2010-11-23 13:42:23 -080014 int cmd, u8 wait, struct file_lock *fl)
Greg Farnum40819f62010-08-02 15:34:23 -070015{
Al Viro496ad9a2013-01-23 17:07:38 -050016 struct inode *inode = file_inode(file);
Greg Farnum40819f62010-08-02 15:34:23 -070017 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070018 ceph_sb_to_client(inode->i_sb)->mdsc;
Greg Farnum40819f62010-08-02 15:34:23 -070019 struct ceph_mds_request *req;
20 int err;
Herb Shiu637ae8d2010-11-23 13:42:23 -080021 u64 length = 0;
Greg Farnum40819f62010-08-02 15:34:23 -070022
23 req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
24 if (IS_ERR(req))
25 return PTR_ERR(req);
Sage Weil70b666c2011-05-27 09:24:26 -070026 req->r_inode = inode;
27 ihold(inode);
Greg Farnum40819f62010-08-02 15:34:23 -070028
Herb Shiu637ae8d2010-11-23 13:42:23 -080029 /* 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
Greg Farnum40819f62010-08-02 15:34:23 -070035 dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
Sage Weil0c1f91f2011-05-25 14:56:12 -070036 "length: %llu, wait: %d, type: %d", (int)lock_type,
Herb Shiu637ae8d2010-11-23 13:42:23 -080037 (int)operation, (u64)fl->fl_pid, fl->fl_start,
38 length, wait, fl->fl_type);
39
Greg Farnum40819f62010-08-02 15:34:23 -070040 req->r_args.filelock_change.rule = lock_type;
41 req->r_args.filelock_change.type = cmd;
Herb Shiu637ae8d2010-11-23 13:42:23 -080042 req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
Greg Farnum40819f62010-08-02 15:34:23 -070043 /* This should be adjusted, but I'm not sure if
44 namespaces actually get id numbers*/
45 req->r_args.filelock_change.pid_namespace =
Herb Shiu637ae8d2010-11-23 13:42:23 -080046 cpu_to_le64((u64)(unsigned long)fl->fl_nspid);
47 req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
Greg Farnum40819f62010-08-02 15:34:23 -070048 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);
Herb Shiua5b10622010-11-23 13:58:29 -080052
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 }
Greg Farnum40819f62010-08-02 15:34:23 -070071 ceph_mdsc_put_request(req);
72 dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
Sage Weil0c1f91f2011-05-25 14:56:12 -070073 "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
Herb Shiu637ae8d2010-11-23 13:42:23 -080074 (int)operation, (u64)fl->fl_pid, fl->fl_start,
75 length, wait, fl->fl_type, err);
Greg Farnum40819f62010-08-02 15:34:23 -070076 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 */
83int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
84{
Greg Farnum40819f62010-08-02 15:34:23 -070085 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
Herb Shiu637ae8d2010-11-23 13:42:23 -0800106 err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
Greg Farnum40819f62010-08-02 15:34:23 -0700107 if (!err) {
Herb Shiua5b10622010-11-23 13:58:29 -0800108 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)) {
Sage Weil0c1f91f2011-05-25 14:56:12 -0700112 /* undo! This should only happen if
113 * the kernel detects local
114 * deadlock. */
Herb Shiua5b10622010-11-23 13:58:29 -0800115 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
116 CEPH_LOCK_UNLOCK, 0, fl);
Sage Weil0c1f91f2011-05-25 14:56:12 -0700117 dout("got %d on posix_lock_file, undid lock",
118 err);
Herb Shiua5b10622010-11-23 13:58:29 -0800119 }
Greg Farnum40819f62010-08-02 15:34:23 -0700120 }
Herb Shiua5b10622010-11-23 13:58:29 -0800121
Sage Weil0c1f91f2011-05-25 14:56:12 -0700122 } else if (err == -ERESTARTSYS) {
123 dout("undoing lock\n");
124 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
125 CEPH_LOCK_UNLOCK, 0, fl);
Greg Farnum40819f62010-08-02 15:34:23 -0700126 }
127 return err;
128}
129
130int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
131{
Greg Farnum40819f62010-08-02 15:34:23 -0700132 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;
Greg Farnum40819f62010-08-02 15:34:23 -0700151
152 err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
Herb Shiu637ae8d2010-11-23 13:42:23 -0800153 file, lock_cmd, wait, fl);
Greg Farnum40819f62010-08-02 15:34:23 -0700154 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,
Herb Shiu637ae8d2010-11-23 13:42:23 -0800159 file, CEPH_LOCK_UNLOCK, 0, fl);
Greg Farnum40819f62010-08-02 15:34:23 -0700160 dout("got %d on flock_lock_file_wait, undid lock", err);
161 }
Sage Weil0c1f91f2011-05-25 14:56:12 -0700162 } 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);
Greg Farnum40819f62010-08-02 15:34:23 -0700167 }
168 return err;
169}
170
171/**
Jim Schutt4d1bf792013-05-15 10:38:12 -0600172 * Must be called with lock_flocks() already held. Fills in the passed
Greg Farnum40819f62010-08-02 15:34:23 -0700173 * counter variables, so you can prepare pagelist metadata before calling
174 * ceph_encode_locks.
175 */
176void 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/**
Jim Schutt39be95e2013-05-15 13:03:35 -0500194 * Encode the flock and fcntl locks for the given inode into the ceph_filelock
Jeff Layton1c8c6012013-06-21 08:58:15 -0400195 * array. Must be called with inode->i_lock already held.
Jim Schutt39be95e2013-05-15 13:03:35 -0500196 * If we encounter more of a specific lock type than expected, return -ENOSPC.
Greg Farnum40819f62010-08-02 15:34:23 -0700197 */
Jim Schutt39be95e2013-05-15 13:03:35 -0500198int ceph_encode_locks_to_buffer(struct inode *inode,
199 struct ceph_filelock *flocks,
200 int num_fcntl_locks, int num_flock_locks)
Greg Farnum40819f62010-08-02 15:34:23 -0700201{
202 struct file_lock *lock;
Greg Farnum40819f62010-08-02 15:34:23 -0700203 int err = 0;
Greg Farnumfca44512010-09-17 10:24:02 -0700204 int seen_fcntl = 0;
205 int seen_flock = 0;
Jim Schutt39be95e2013-05-15 13:03:35 -0500206 int l = 0;
Greg Farnum40819f62010-08-02 15:34:23 -0700207
208 dout("encoding %d flock and %d fcntl locks", num_flock_locks,
209 num_fcntl_locks);
Jim Schutt39be95e2013-05-15 13:03:35 -0500210
Greg Farnum40819f62010-08-02 15:34:23 -0700211 for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
212 if (lock->fl_flags & FL_POSIX) {
Greg Farnumfca44512010-09-17 10:24:02 -0700213 ++seen_fcntl;
214 if (seen_fcntl > num_fcntl_locks) {
215 err = -ENOSPC;
216 goto fail;
217 }
Jim Schutt39be95e2013-05-15 13:03:35 -0500218 err = lock_to_ceph_filelock(lock, &flocks[l]);
Greg Farnum40819f62010-08-02 15:34:23 -0700219 if (err)
220 goto fail;
Jim Schutt39be95e2013-05-15 13:03:35 -0500221 ++l;
Greg Farnum40819f62010-08-02 15:34:23 -0700222 }
Greg Farnum40819f62010-08-02 15:34:23 -0700223 }
Greg Farnum40819f62010-08-02 15:34:23 -0700224 for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
225 if (lock->fl_flags & FL_FLOCK) {
Greg Farnumfca44512010-09-17 10:24:02 -0700226 ++seen_flock;
227 if (seen_flock > num_flock_locks) {
228 err = -ENOSPC;
229 goto fail;
230 }
Jim Schutt39be95e2013-05-15 13:03:35 -0500231 err = lock_to_ceph_filelock(lock, &flocks[l]);
Greg Farnum40819f62010-08-02 15:34:23 -0700232 if (err)
233 goto fail;
Jim Schutt39be95e2013-05-15 13:03:35 -0500234 ++l;
Greg Farnum40819f62010-08-02 15:34:23 -0700235 }
Greg Farnum40819f62010-08-02 15:34:23 -0700236 }
237fail:
238 return err;
239}
240
Jim Schutt39be95e2013-05-15 13:03:35 -0500241/**
242 * Copy the encoded flock and fcntl locks into the pagelist.
243 * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
244 * sequential flock locks.
245 * Returns zero on success.
246 */
247int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
248 struct ceph_pagelist *pagelist,
249 int num_fcntl_locks, int num_flock_locks)
250{
251 int err = 0;
252 __le32 nlocks;
253
254 nlocks = cpu_to_le32(num_fcntl_locks);
255 err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
256 if (err)
257 goto out_fail;
258
259 err = ceph_pagelist_append(pagelist, flocks,
260 num_fcntl_locks * sizeof(*flocks));
261 if (err)
262 goto out_fail;
263
264 nlocks = cpu_to_le32(num_flock_locks);
265 err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
266 if (err)
267 goto out_fail;
268
269 err = ceph_pagelist_append(pagelist,
270 &flocks[num_fcntl_locks],
271 num_flock_locks * sizeof(*flocks));
272out_fail:
273 return err;
274}
275
Greg Farnum40819f62010-08-02 15:34:23 -0700276/*
277 * Given a pointer to a lock, convert it to a ceph filelock
278 */
279int lock_to_ceph_filelock(struct file_lock *lock,
280 struct ceph_filelock *cephlock)
281{
282 int err = 0;
283
284 cephlock->start = cpu_to_le64(lock->fl_start);
285 cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
286 cephlock->client = cpu_to_le64(0);
287 cephlock->pid = cpu_to_le64(lock->fl_pid);
Alan Coxad8453a2010-08-25 13:26:32 +0100288 cephlock->pid_namespace =
289 cpu_to_le64((u64)(unsigned long)lock->fl_nspid);
Greg Farnum40819f62010-08-02 15:34:23 -0700290
291 switch (lock->fl_type) {
292 case F_RDLCK:
293 cephlock->type = CEPH_LOCK_SHARED;
294 break;
295 case F_WRLCK:
296 cephlock->type = CEPH_LOCK_EXCL;
297 break;
298 case F_UNLCK:
299 cephlock->type = CEPH_LOCK_UNLOCK;
300 break;
301 default:
302 dout("Have unknown lock type %d", lock->fl_type);
303 err = -EINVAL;
304 }
305
306 return err;
307}