Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
40819f6f GF |
2 | |
3 | #include <linux/file.h> | |
4 | #include <linux/namei.h> | |
eb13e832 | 5 | #include <linux/random.h> |
40819f6f GF |
6 | |
7 | #include "super.h" | |
8 | #include "mds_client.h" | |
3d14c5d2 | 9 | #include <linux/ceph/pagelist.h> |
40819f6f | 10 | |
eb13e832 | 11 | static u64 lock_secret; |
9280be24 YZ |
12 | static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc, |
13 | struct ceph_mds_request *req); | |
eb13e832 YZ |
14 | |
15 | static inline u64 secure_addr(void *addr) | |
16 | { | |
17 | u64 v = lock_secret ^ (u64)(unsigned long)addr; | |
18 | /* | |
19 | * Set the most significant bit, so that MDS knows the 'owner' | |
20 | * is sufficient to identify the owner of lock. (old code uses | |
21 | * both 'owner' and 'pid') | |
22 | */ | |
23 | v |= (1ULL << 63); | |
24 | return v; | |
25 | } | |
26 | ||
27 | void __init ceph_flock_init(void) | |
28 | { | |
29 | get_random_bytes(&lock_secret, sizeof(lock_secret)); | |
30 | } | |
31 | ||
40819f6f GF |
32 | /** |
33 | * Implement fcntl and flock locking functions. | |
34 | */ | |
35 | static int ceph_lock_message(u8 lock_type, u16 operation, struct file *file, | |
637ae8d5 | 36 | int cmd, u8 wait, struct file_lock *fl) |
40819f6f | 37 | { |
496ad9aa | 38 | struct inode *inode = file_inode(file); |
eb13e832 | 39 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
40819f6f GF |
40 | struct ceph_mds_request *req; |
41 | int err; | |
637ae8d5 | 42 | u64 length = 0; |
eb13e832 | 43 | u64 owner; |
40819f6f | 44 | |
9280be24 YZ |
45 | if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK) |
46 | wait = 0; | |
47 | ||
40819f6f GF |
48 | req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS); |
49 | if (IS_ERR(req)) | |
50 | return PTR_ERR(req); | |
70b666c3 SW |
51 | req->r_inode = inode; |
52 | ihold(inode); | |
3bd58143 | 53 | req->r_num_caps = 1; |
40819f6f | 54 | |
637ae8d5 HS |
55 | /* mds requires start and length rather than start and end */ |
56 | if (LLONG_MAX == fl->fl_end) | |
57 | length = 0; | |
58 | else | |
59 | length = fl->fl_end - fl->fl_start + 1; | |
60 | ||
130d1f95 | 61 | owner = secure_addr(fl->fl_owner); |
eb13e832 YZ |
62 | |
63 | dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, " | |
64 | "start: %llu, length: %llu, wait: %d, type: %d", (int)lock_type, | |
65 | (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length, | |
66 | wait, fl->fl_type); | |
637ae8d5 | 67 | |
40819f6f GF |
68 | req->r_args.filelock_change.rule = lock_type; |
69 | req->r_args.filelock_change.type = cmd; | |
eb13e832 | 70 | req->r_args.filelock_change.owner = cpu_to_le64(owner); |
637ae8d5 | 71 | req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid); |
637ae8d5 | 72 | req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start); |
40819f6f GF |
73 | req->r_args.filelock_change.length = cpu_to_le64(length); |
74 | req->r_args.filelock_change.wait = wait; | |
75 | ||
9280be24 YZ |
76 | if (wait) |
77 | req->r_wait_for_completion = ceph_lock_wait_for_completion; | |
78 | ||
40819f6f | 79 | err = ceph_mdsc_do_request(mdsc, inode, req); |
a5b10629 | 80 | |
eb13e832 | 81 | if (operation == CEPH_MDS_OP_GETFILELOCK) { |
a5b10629 HS |
82 | fl->fl_pid = le64_to_cpu(req->r_reply_info.filelock_reply->pid); |
83 | if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type) | |
84 | fl->fl_type = F_RDLCK; | |
85 | else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type) | |
86 | fl->fl_type = F_WRLCK; | |
87 | else | |
88 | fl->fl_type = F_UNLCK; | |
89 | ||
90 | fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start); | |
91 | length = le64_to_cpu(req->r_reply_info.filelock_reply->start) + | |
92 | le64_to_cpu(req->r_reply_info.filelock_reply->length); | |
93 | if (length >= 1) | |
94 | fl->fl_end = length -1; | |
95 | else | |
96 | fl->fl_end = 0; | |
97 | ||
98 | } | |
40819f6f GF |
99 | ceph_mdsc_put_request(req); |
100 | dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, " | |
0c1f91f2 | 101 | "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type, |
637ae8d5 HS |
102 | (int)operation, (u64)fl->fl_pid, fl->fl_start, |
103 | length, wait, fl->fl_type, err); | |
40819f6f GF |
104 | return err; |
105 | } | |
106 | ||
9280be24 YZ |
107 | static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc, |
108 | struct ceph_mds_request *req) | |
109 | { | |
110 | struct ceph_mds_request *intr_req; | |
111 | struct inode *inode = req->r_inode; | |
112 | int err, lock_type; | |
113 | ||
114 | BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK); | |
115 | if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL) | |
116 | lock_type = CEPH_LOCK_FCNTL_INTR; | |
117 | else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK) | |
118 | lock_type = CEPH_LOCK_FLOCK_INTR; | |
119 | else | |
120 | BUG_ON(1); | |
121 | BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK); | |
122 | ||
123 | err = wait_for_completion_interruptible(&req->r_completion); | |
124 | if (!err) | |
125 | return 0; | |
126 | ||
127 | dout("ceph_lock_wait_for_completion: request %llu was interrupted\n", | |
128 | req->r_tid); | |
129 | ||
130 | intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK, | |
131 | USE_AUTH_MDS); | |
132 | if (IS_ERR(intr_req)) | |
133 | return PTR_ERR(intr_req); | |
134 | ||
135 | intr_req->r_inode = inode; | |
136 | ihold(inode); | |
137 | intr_req->r_num_caps = 1; | |
138 | ||
139 | intr_req->r_args.filelock_change = req->r_args.filelock_change; | |
140 | intr_req->r_args.filelock_change.rule = lock_type; | |
141 | intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK; | |
142 | ||
143 | err = ceph_mdsc_do_request(mdsc, inode, intr_req); | |
144 | ceph_mdsc_put_request(intr_req); | |
145 | ||
146 | if (err && err != -ERESTARTSYS) | |
147 | return err; | |
148 | ||
149 | wait_for_completion(&req->r_completion); | |
150 | return 0; | |
151 | } | |
152 | ||
40819f6f GF |
153 | /** |
154 | * Attempt to set an fcntl lock. | |
155 | * For now, this just goes away to the server. Later it may be more awesome. | |
156 | */ | |
157 | int ceph_lock(struct file *file, int cmd, struct file_lock *fl) | |
158 | { | |
40819f6f GF |
159 | u8 lock_cmd; |
160 | int err; | |
161 | u8 wait = 0; | |
162 | u16 op = CEPH_MDS_OP_SETFILELOCK; | |
163 | ||
eb70c0ce YZ |
164 | if (!(fl->fl_flags & FL_POSIX)) |
165 | return -ENOLCK; | |
166 | /* No mandatory locks */ | |
167 | if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK) | |
168 | return -ENOLCK; | |
169 | ||
eb13e832 | 170 | dout("ceph_lock, fl_owner: %p", fl->fl_owner); |
40819f6f GF |
171 | |
172 | /* set wait bit as appropriate, then make command as Ceph expects it*/ | |
0e8e95d6 | 173 | if (IS_GETLK(cmd)) |
40819f6f | 174 | op = CEPH_MDS_OP_GETFILELOCK; |
0e8e95d6 YZ |
175 | else if (IS_SETLKW(cmd)) |
176 | wait = 1; | |
40819f6f GF |
177 | |
178 | if (F_RDLCK == fl->fl_type) | |
179 | lock_cmd = CEPH_LOCK_SHARED; | |
180 | else if (F_WRLCK == fl->fl_type) | |
181 | lock_cmd = CEPH_LOCK_EXCL; | |
182 | else | |
183 | lock_cmd = CEPH_LOCK_UNLOCK; | |
184 | ||
637ae8d5 | 185 | err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl); |
40819f6f | 186 | if (!err) { |
eb13e832 | 187 | if (op != CEPH_MDS_OP_GETFILELOCK) { |
a5b10629 HS |
188 | dout("mds locked, locking locally"); |
189 | err = posix_lock_file(file, fl, NULL); | |
190 | if (err && (CEPH_MDS_OP_SETFILELOCK == op)) { | |
0c1f91f2 SW |
191 | /* undo! This should only happen if |
192 | * the kernel detects local | |
193 | * deadlock. */ | |
a5b10629 HS |
194 | ceph_lock_message(CEPH_LOCK_FCNTL, op, file, |
195 | CEPH_LOCK_UNLOCK, 0, fl); | |
0c1f91f2 SW |
196 | dout("got %d on posix_lock_file, undid lock", |
197 | err); | |
a5b10629 | 198 | } |
40819f6f | 199 | } |
40819f6f GF |
200 | } |
201 | return err; | |
202 | } | |
203 | ||
204 | int ceph_flock(struct file *file, int cmd, struct file_lock *fl) | |
205 | { | |
40819f6f GF |
206 | u8 lock_cmd; |
207 | int err; | |
0e8e95d6 | 208 | u8 wait = 0; |
40819f6f | 209 | |
eb70c0ce YZ |
210 | if (!(fl->fl_flags & FL_FLOCK)) |
211 | return -ENOLCK; | |
212 | /* No mandatory locks */ | |
213 | if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK) | |
214 | return -ENOLCK; | |
215 | ||
eb13e832 | 216 | dout("ceph_flock, fl_file: %p", fl->fl_file); |
40819f6f | 217 | |
0e8e95d6 YZ |
218 | if (IS_SETLKW(cmd)) |
219 | wait = 1; | |
220 | ||
221 | if (F_RDLCK == fl->fl_type) | |
40819f6f | 222 | lock_cmd = CEPH_LOCK_SHARED; |
0e8e95d6 | 223 | else if (F_WRLCK == fl->fl_type) |
40819f6f GF |
224 | lock_cmd = CEPH_LOCK_EXCL; |
225 | else | |
226 | lock_cmd = CEPH_LOCK_UNLOCK; | |
40819f6f GF |
227 | |
228 | err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK, | |
637ae8d5 | 229 | file, lock_cmd, wait, fl); |
40819f6f | 230 | if (!err) { |
4f656367 | 231 | err = locks_lock_file_wait(file, fl); |
40819f6f GF |
232 | if (err) { |
233 | ceph_lock_message(CEPH_LOCK_FLOCK, | |
234 | CEPH_MDS_OP_SETFILELOCK, | |
637ae8d5 | 235 | file, CEPH_LOCK_UNLOCK, 0, fl); |
4f656367 | 236 | dout("got %d on locks_lock_file_wait, undid lock", err); |
40819f6f | 237 | } |
40819f6f GF |
238 | } |
239 | return err; | |
240 | } | |
241 | ||
5263e31e JL |
242 | /* |
243 | * Fills in the passed counter variables, so you can prepare pagelist metadata | |
244 | * before calling ceph_encode_locks. | |
40819f6f GF |
245 | */ |
246 | void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count) | |
247 | { | |
e084c1bd | 248 | struct file_lock *lock; |
5263e31e | 249 | struct file_lock_context *ctx; |
40819f6f GF |
250 | |
251 | *fcntl_count = 0; | |
252 | *flock_count = 0; | |
253 | ||
5263e31e JL |
254 | ctx = inode->i_flctx; |
255 | if (ctx) { | |
e084c1bd JL |
256 | spin_lock(&ctx->flc_lock); |
257 | list_for_each_entry(lock, &ctx->flc_posix, fl_list) | |
258 | ++(*fcntl_count); | |
259 | list_for_each_entry(lock, &ctx->flc_flock, fl_list) | |
260 | ++(*flock_count); | |
261 | spin_unlock(&ctx->flc_lock); | |
40819f6f GF |
262 | } |
263 | dout("counted %d flock locks and %d fcntl locks", | |
264 | *flock_count, *fcntl_count); | |
265 | } | |
266 | ||
267 | /** | |
39be95e9 | 268 | * Encode the flock and fcntl locks for the given inode into the ceph_filelock |
1c8c601a | 269 | * array. Must be called with inode->i_lock already held. |
39be95e9 | 270 | * If we encounter more of a specific lock type than expected, return -ENOSPC. |
40819f6f | 271 | */ |
39be95e9 JS |
272 | int ceph_encode_locks_to_buffer(struct inode *inode, |
273 | struct ceph_filelock *flocks, | |
274 | int num_fcntl_locks, int num_flock_locks) | |
40819f6f GF |
275 | { |
276 | struct file_lock *lock; | |
bd61e0a9 | 277 | struct file_lock_context *ctx = inode->i_flctx; |
40819f6f | 278 | int err = 0; |
fca4451a GF |
279 | int seen_fcntl = 0; |
280 | int seen_flock = 0; | |
39be95e9 | 281 | int l = 0; |
40819f6f GF |
282 | |
283 | dout("encoding %d flock and %d fcntl locks", num_flock_locks, | |
284 | num_fcntl_locks); | |
39be95e9 | 285 | |
bd61e0a9 JL |
286 | if (!ctx) |
287 | return 0; | |
288 | ||
6109c850 | 289 | spin_lock(&ctx->flc_lock); |
f6762cb2 | 290 | list_for_each_entry(lock, &ctx->flc_posix, fl_list) { |
bd61e0a9 JL |
291 | ++seen_fcntl; |
292 | if (seen_fcntl > num_fcntl_locks) { | |
293 | err = -ENOSPC; | |
294 | goto fail; | |
40819f6f | 295 | } |
bd61e0a9 JL |
296 | err = lock_to_ceph_filelock(lock, &flocks[l]); |
297 | if (err) | |
298 | goto fail; | |
299 | ++l; | |
40819f6f | 300 | } |
bd61e0a9 JL |
301 | list_for_each_entry(lock, &ctx->flc_flock, fl_list) { |
302 | ++seen_flock; | |
303 | if (seen_flock > num_flock_locks) { | |
304 | err = -ENOSPC; | |
305 | goto fail; | |
40819f6f | 306 | } |
bd61e0a9 JL |
307 | err = lock_to_ceph_filelock(lock, &flocks[l]); |
308 | if (err) | |
309 | goto fail; | |
310 | ++l; | |
40819f6f GF |
311 | } |
312 | fail: | |
6109c850 | 313 | spin_unlock(&ctx->flc_lock); |
40819f6f GF |
314 | return err; |
315 | } | |
316 | ||
39be95e9 JS |
317 | /** |
318 | * Copy the encoded flock and fcntl locks into the pagelist. | |
319 | * Format is: #fcntl locks, sequential fcntl locks, #flock locks, | |
320 | * sequential flock locks. | |
321 | * Returns zero on success. | |
322 | */ | |
323 | int ceph_locks_to_pagelist(struct ceph_filelock *flocks, | |
324 | struct ceph_pagelist *pagelist, | |
325 | int num_fcntl_locks, int num_flock_locks) | |
326 | { | |
327 | int err = 0; | |
328 | __le32 nlocks; | |
329 | ||
330 | nlocks = cpu_to_le32(num_fcntl_locks); | |
331 | err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks)); | |
332 | if (err) | |
333 | goto out_fail; | |
334 | ||
335 | err = ceph_pagelist_append(pagelist, flocks, | |
336 | num_fcntl_locks * sizeof(*flocks)); | |
337 | if (err) | |
338 | goto out_fail; | |
339 | ||
340 | nlocks = cpu_to_le32(num_flock_locks); | |
341 | err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks)); | |
342 | if (err) | |
343 | goto out_fail; | |
344 | ||
345 | err = ceph_pagelist_append(pagelist, | |
346 | &flocks[num_fcntl_locks], | |
347 | num_flock_locks * sizeof(*flocks)); | |
348 | out_fail: | |
349 | return err; | |
350 | } | |
351 | ||
40819f6f GF |
352 | /* |
353 | * Given a pointer to a lock, convert it to a ceph filelock | |
354 | */ | |
355 | int lock_to_ceph_filelock(struct file_lock *lock, | |
356 | struct ceph_filelock *cephlock) | |
357 | { | |
358 | int err = 0; | |
40819f6f GF |
359 | cephlock->start = cpu_to_le64(lock->fl_start); |
360 | cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1); | |
361 | cephlock->client = cpu_to_le64(0); | |
eb13e832 | 362 | cephlock->pid = cpu_to_le64((u64)lock->fl_pid); |
130d1f95 | 363 | cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner)); |
40819f6f GF |
364 | |
365 | switch (lock->fl_type) { | |
366 | case F_RDLCK: | |
367 | cephlock->type = CEPH_LOCK_SHARED; | |
368 | break; | |
369 | case F_WRLCK: | |
370 | cephlock->type = CEPH_LOCK_EXCL; | |
371 | break; | |
372 | case F_UNLCK: | |
373 | cephlock->type = CEPH_LOCK_UNLOCK; | |
374 | break; | |
375 | default: | |
376 | dout("Have unknown lock type %d", lock->fl_type); | |
377 | err = -EINVAL; | |
378 | } | |
379 | ||
380 | return err; | |
381 | } |