1 | // SPDX-License-Identifier: GPL-2.0 |
2 | #include <linux/ceph/ceph_debug.h> |
3 | |
4 | #include <linux/file.h> |
5 | #include <linux/namei.h> |
6 | #include <linux/random.h> |
7 | |
8 | #include "super.h" |
9 | #include "mds_client.h" |
10 | #include <linux/filelock.h> |
11 | #include <linux/ceph/pagelist.h> |
12 | |
13 | static u64 lock_secret; |
14 | static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc, |
15 | struct ceph_mds_request *req); |
16 | |
17 | static inline u64 secure_addr(void *addr) |
18 | { |
19 | u64 v = lock_secret ^ (u64)(unsigned long)addr; |
20 | /* |
21 | * Set the most significant bit, so that MDS knows the 'owner' |
22 | * is sufficient to identify the owner of lock. (old code uses |
23 | * both 'owner' and 'pid') |
24 | */ |
25 | v |= (1ULL << 63); |
26 | return v; |
27 | } |
28 | |
29 | void __init ceph_flock_init(void) |
30 | { |
31 | get_random_bytes(buf: &lock_secret, len: sizeof(lock_secret)); |
32 | } |
33 | |
34 | static void ceph_fl_copy_lock(struct file_lock *dst, struct file_lock *src) |
35 | { |
36 | struct inode *inode = file_inode(f: dst->c.flc_file); |
37 | atomic_inc(v: &ceph_inode(inode)->i_filelock_ref); |
38 | dst->fl_u.ceph.inode = igrab(inode); |
39 | } |
40 | |
41 | /* |
42 | * Do not use the 'fl->fl_file' in release function, which |
43 | * is possibly already released by another thread. |
44 | */ |
45 | static void ceph_fl_release_lock(struct file_lock *fl) |
46 | { |
47 | struct inode *inode = fl->fl_u.ceph.inode; |
48 | struct ceph_inode_info *ci; |
49 | |
50 | /* |
51 | * If inode is NULL it should be a request file_lock, |
52 | * nothing we can do. |
53 | */ |
54 | if (!inode) |
55 | return; |
56 | |
57 | ci = ceph_inode(inode); |
58 | if (atomic_dec_and_test(v: &ci->i_filelock_ref)) { |
59 | /* clear error when all locks are released */ |
60 | spin_lock(lock: &ci->i_ceph_lock); |
61 | ci->i_ceph_flags &= ~CEPH_I_ERROR_FILELOCK; |
62 | spin_unlock(lock: &ci->i_ceph_lock); |
63 | } |
64 | fl->fl_u.ceph.inode = NULL; |
65 | iput(inode); |
66 | } |
67 | |
68 | static const struct file_lock_operations ceph_fl_lock_ops = { |
69 | .fl_copy_lock = ceph_fl_copy_lock, |
70 | .fl_release_private = ceph_fl_release_lock, |
71 | }; |
72 | |
73 | /* |
74 | * Implement fcntl and flock locking functions. |
75 | */ |
76 | static int ceph_lock_message(u8 lock_type, u16 operation, struct inode *inode, |
77 | int cmd, u8 wait, struct file_lock *fl) |
78 | { |
79 | struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(sb: inode->i_sb); |
80 | struct ceph_client *cl = mdsc->fsc->client; |
81 | struct ceph_mds_request *req; |
82 | int err; |
83 | u64 length = 0; |
84 | u64 owner; |
85 | |
86 | if (operation == CEPH_MDS_OP_SETFILELOCK) { |
87 | /* |
88 | * increasing i_filelock_ref closes race window between |
89 | * handling request reply and adding file_lock struct to |
90 | * inode. Otherwise, auth caps may get trimmed in the |
91 | * window. Caller function will decrease the counter. |
92 | */ |
93 | fl->fl_ops = &ceph_fl_lock_ops; |
94 | fl->fl_ops->fl_copy_lock(fl, NULL); |
95 | } |
96 | |
97 | if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK) |
98 | wait = 0; |
99 | |
100 | req = ceph_mdsc_create_request(mdsc, op: operation, mode: USE_AUTH_MDS); |
101 | if (IS_ERR(ptr: req)) |
102 | return PTR_ERR(ptr: req); |
103 | req->r_inode = inode; |
104 | ihold(inode); |
105 | req->r_num_caps = 1; |
106 | |
107 | /* mds requires start and length rather than start and end */ |
108 | if (LLONG_MAX == fl->fl_end) |
109 | length = 0; |
110 | else |
111 | length = fl->fl_end - fl->fl_start + 1; |
112 | |
113 | owner = secure_addr(addr: fl->c.flc_owner); |
114 | |
115 | doutc(cl, "rule: %d, op: %d, owner: %llx, pid: %llu, " |
116 | "start: %llu, length: %llu, wait: %d, type: %d\n" , |
117 | (int)lock_type, (int)operation, owner, |
118 | (u64) fl->c.flc_pid, |
119 | fl->fl_start, length, wait, fl->c.flc_type); |
120 | |
121 | req->r_args.filelock_change.rule = lock_type; |
122 | req->r_args.filelock_change.type = cmd; |
123 | req->r_args.filelock_change.owner = cpu_to_le64(owner); |
124 | req->r_args.filelock_change.pid = cpu_to_le64((u64) fl->c.flc_pid); |
125 | req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start); |
126 | req->r_args.filelock_change.length = cpu_to_le64(length); |
127 | req->r_args.filelock_change.wait = wait; |
128 | |
129 | err = ceph_mdsc_submit_request(mdsc, dir: inode, req); |
130 | if (!err) |
131 | err = ceph_mdsc_wait_request(mdsc, req, wait_func: wait ? |
132 | ceph_lock_wait_for_completion : NULL); |
133 | if (!err && operation == CEPH_MDS_OP_GETFILELOCK) { |
134 | fl->c.flc_pid = -le64_to_cpu(req->r_reply_info.filelock_reply->pid); |
135 | if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type) |
136 | fl->c.flc_type = F_RDLCK; |
137 | else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type) |
138 | fl->c.flc_type = F_WRLCK; |
139 | else |
140 | fl->c.flc_type = F_UNLCK; |
141 | |
142 | fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start); |
143 | length = le64_to_cpu(req->r_reply_info.filelock_reply->start) + |
144 | le64_to_cpu(req->r_reply_info.filelock_reply->length); |
145 | if (length >= 1) |
146 | fl->fl_end = length -1; |
147 | else |
148 | fl->fl_end = 0; |
149 | |
150 | } |
151 | ceph_mdsc_put_request(req); |
152 | doutc(cl, "rule: %d, op: %d, pid: %llu, start: %llu, " |
153 | "length: %llu, wait: %d, type: %d, err code %d\n" , |
154 | (int)lock_type, (int)operation, (u64) fl->c.flc_pid, |
155 | fl->fl_start, length, wait, fl->c.flc_type, err); |
156 | return err; |
157 | } |
158 | |
159 | static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc, |
160 | struct ceph_mds_request *req) |
161 | { |
162 | struct ceph_client *cl = mdsc->fsc->client; |
163 | struct ceph_mds_request *intr_req; |
164 | struct inode *inode = req->r_inode; |
165 | int err, lock_type; |
166 | |
167 | BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK); |
168 | if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL) |
169 | lock_type = CEPH_LOCK_FCNTL_INTR; |
170 | else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK) |
171 | lock_type = CEPH_LOCK_FLOCK_INTR; |
172 | else |
173 | BUG_ON(1); |
174 | BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK); |
175 | |
176 | err = wait_for_completion_interruptible(x: &req->r_completion); |
177 | if (!err) |
178 | return 0; |
179 | |
180 | doutc(cl, "request %llu was interrupted\n" , req->r_tid); |
181 | |
182 | mutex_lock(&mdsc->mutex); |
183 | if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) { |
184 | err = 0; |
185 | } else { |
186 | /* |
187 | * ensure we aren't running concurrently with |
188 | * ceph_fill_trace or ceph_readdir_prepopulate, which |
189 | * rely on locks (dir mutex) held by our caller. |
190 | */ |
191 | mutex_lock(&req->r_fill_mutex); |
192 | req->r_err = err; |
193 | set_bit(CEPH_MDS_R_ABORTED, addr: &req->r_req_flags); |
194 | mutex_unlock(lock: &req->r_fill_mutex); |
195 | |
196 | if (!req->r_session) { |
197 | // haven't sent the request |
198 | err = 0; |
199 | } |
200 | } |
201 | mutex_unlock(lock: &mdsc->mutex); |
202 | if (!err) |
203 | return 0; |
204 | |
205 | intr_req = ceph_mdsc_create_request(mdsc, op: CEPH_MDS_OP_SETFILELOCK, |
206 | mode: USE_AUTH_MDS); |
207 | if (IS_ERR(ptr: intr_req)) |
208 | return PTR_ERR(ptr: intr_req); |
209 | |
210 | intr_req->r_inode = inode; |
211 | ihold(inode); |
212 | intr_req->r_num_caps = 1; |
213 | |
214 | intr_req->r_args.filelock_change = req->r_args.filelock_change; |
215 | intr_req->r_args.filelock_change.rule = lock_type; |
216 | intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK; |
217 | |
218 | err = ceph_mdsc_do_request(mdsc, dir: inode, req: intr_req); |
219 | ceph_mdsc_put_request(req: intr_req); |
220 | |
221 | if (err && err != -ERESTARTSYS) |
222 | return err; |
223 | |
224 | wait_for_completion_killable(x: &req->r_safe_completion); |
225 | return 0; |
226 | } |
227 | |
228 | static int try_unlock_file(struct file *file, struct file_lock *fl) |
229 | { |
230 | int err; |
231 | unsigned int orig_flags = fl->c.flc_flags; |
232 | fl->c.flc_flags |= FL_EXISTS; |
233 | err = locks_lock_file_wait(filp: file, fl); |
234 | fl->c.flc_flags = orig_flags; |
235 | if (err == -ENOENT) { |
236 | if (!(orig_flags & FL_EXISTS)) |
237 | err = 0; |
238 | return err; |
239 | } |
240 | return 1; |
241 | } |
242 | |
243 | /* |
244 | * Attempt to set an fcntl lock. |
245 | * For now, this just goes away to the server. Later it may be more awesome. |
246 | */ |
247 | int ceph_lock(struct file *file, int cmd, struct file_lock *fl) |
248 | { |
249 | struct inode *inode = file_inode(f: file); |
250 | struct ceph_inode_info *ci = ceph_inode(inode); |
251 | struct ceph_client *cl = ceph_inode_to_client(inode); |
252 | int err = 0; |
253 | u16 op = CEPH_MDS_OP_SETFILELOCK; |
254 | u8 wait = 0; |
255 | u8 lock_cmd; |
256 | |
257 | if (!(fl->c.flc_flags & FL_POSIX)) |
258 | return -ENOLCK; |
259 | |
260 | if (ceph_inode_is_shutdown(inode)) |
261 | return -ESTALE; |
262 | |
263 | doutc(cl, "fl_owner: %p\n" , fl->c.flc_owner); |
264 | |
265 | /* set wait bit as appropriate, then make command as Ceph expects it*/ |
266 | if (IS_GETLK(cmd)) |
267 | op = CEPH_MDS_OP_GETFILELOCK; |
268 | else if (IS_SETLKW(cmd)) |
269 | wait = 1; |
270 | |
271 | spin_lock(lock: &ci->i_ceph_lock); |
272 | if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) { |
273 | err = -EIO; |
274 | } |
275 | spin_unlock(lock: &ci->i_ceph_lock); |
276 | if (err < 0) { |
277 | if (op == CEPH_MDS_OP_SETFILELOCK && lock_is_unlock(fl)) |
278 | posix_lock_file(file, fl, NULL); |
279 | return err; |
280 | } |
281 | |
282 | if (lock_is_read(fl)) |
283 | lock_cmd = CEPH_LOCK_SHARED; |
284 | else if (lock_is_write(fl)) |
285 | lock_cmd = CEPH_LOCK_EXCL; |
286 | else |
287 | lock_cmd = CEPH_LOCK_UNLOCK; |
288 | |
289 | if (op == CEPH_MDS_OP_SETFILELOCK && lock_is_unlock(fl)) { |
290 | err = try_unlock_file(file, fl); |
291 | if (err <= 0) |
292 | return err; |
293 | } |
294 | |
295 | err = ceph_lock_message(CEPH_LOCK_FCNTL, operation: op, inode, cmd: lock_cmd, wait, fl); |
296 | if (!err) { |
297 | if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK != fl->c.flc_type) { |
298 | doutc(cl, "locking locally\n" ); |
299 | err = posix_lock_file(file, fl, NULL); |
300 | if (err) { |
301 | /* undo! This should only happen if |
302 | * the kernel detects local |
303 | * deadlock. */ |
304 | ceph_lock_message(CEPH_LOCK_FCNTL, operation: op, inode, |
305 | CEPH_LOCK_UNLOCK, wait: 0, fl); |
306 | doutc(cl, "got %d on posix_lock_file, undid lock\n" , |
307 | err); |
308 | } |
309 | } |
310 | } |
311 | return err; |
312 | } |
313 | |
314 | int ceph_flock(struct file *file, int cmd, struct file_lock *fl) |
315 | { |
316 | struct inode *inode = file_inode(f: file); |
317 | struct ceph_inode_info *ci = ceph_inode(inode); |
318 | struct ceph_client *cl = ceph_inode_to_client(inode); |
319 | int err = 0; |
320 | u8 wait = 0; |
321 | u8 lock_cmd; |
322 | |
323 | if (!(fl->c.flc_flags & FL_FLOCK)) |
324 | return -ENOLCK; |
325 | |
326 | if (ceph_inode_is_shutdown(inode)) |
327 | return -ESTALE; |
328 | |
329 | doutc(cl, "fl_file: %p\n" , fl->c.flc_file); |
330 | |
331 | spin_lock(lock: &ci->i_ceph_lock); |
332 | if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) { |
333 | err = -EIO; |
334 | } |
335 | spin_unlock(lock: &ci->i_ceph_lock); |
336 | if (err < 0) { |
337 | if (lock_is_unlock(fl)) |
338 | locks_lock_file_wait(filp: file, fl); |
339 | return err; |
340 | } |
341 | |
342 | if (IS_SETLKW(cmd)) |
343 | wait = 1; |
344 | |
345 | if (lock_is_read(fl)) |
346 | lock_cmd = CEPH_LOCK_SHARED; |
347 | else if (lock_is_write(fl)) |
348 | lock_cmd = CEPH_LOCK_EXCL; |
349 | else |
350 | lock_cmd = CEPH_LOCK_UNLOCK; |
351 | |
352 | if (lock_is_unlock(fl)) { |
353 | err = try_unlock_file(file, fl); |
354 | if (err <= 0) |
355 | return err; |
356 | } |
357 | |
358 | err = ceph_lock_message(CEPH_LOCK_FLOCK, operation: CEPH_MDS_OP_SETFILELOCK, |
359 | inode, cmd: lock_cmd, wait, fl); |
360 | if (!err && F_UNLCK != fl->c.flc_type) { |
361 | err = locks_lock_file_wait(filp: file, fl); |
362 | if (err) { |
363 | ceph_lock_message(CEPH_LOCK_FLOCK, |
364 | operation: CEPH_MDS_OP_SETFILELOCK, |
365 | inode, CEPH_LOCK_UNLOCK, wait: 0, fl); |
366 | doutc(cl, "got %d on locks_lock_file_wait, undid lock\n" , |
367 | err); |
368 | } |
369 | } |
370 | return err; |
371 | } |
372 | |
373 | /* |
374 | * Fills in the passed counter variables, so you can prepare pagelist metadata |
375 | * before calling ceph_encode_locks. |
376 | */ |
377 | void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count) |
378 | { |
379 | struct ceph_client *cl = ceph_inode_to_client(inode); |
380 | struct file_lock *lock; |
381 | struct file_lock_context *ctx; |
382 | |
383 | *fcntl_count = 0; |
384 | *flock_count = 0; |
385 | |
386 | ctx = locks_inode_context(inode); |
387 | if (ctx) { |
388 | spin_lock(lock: &ctx->flc_lock); |
389 | for_each_file_lock(lock, &ctx->flc_posix) |
390 | ++(*fcntl_count); |
391 | for_each_file_lock(lock, &ctx->flc_flock) |
392 | ++(*flock_count); |
393 | spin_unlock(lock: &ctx->flc_lock); |
394 | } |
395 | doutc(cl, "counted %d flock locks and %d fcntl locks\n" , |
396 | *flock_count, *fcntl_count); |
397 | } |
398 | |
399 | /* |
400 | * Given a pointer to a lock, convert it to a ceph filelock |
401 | */ |
402 | static int lock_to_ceph_filelock(struct inode *inode, |
403 | struct file_lock *lock, |
404 | struct ceph_filelock *cephlock) |
405 | { |
406 | struct ceph_client *cl = ceph_inode_to_client(inode); |
407 | int err = 0; |
408 | |
409 | cephlock->start = cpu_to_le64(lock->fl_start); |
410 | cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1); |
411 | cephlock->client = cpu_to_le64(0); |
412 | cephlock->pid = cpu_to_le64((u64) lock->c.flc_pid); |
413 | cephlock->owner = cpu_to_le64(secure_addr(lock->c.flc_owner)); |
414 | |
415 | switch (lock->c.flc_type) { |
416 | case F_RDLCK: |
417 | cephlock->type = CEPH_LOCK_SHARED; |
418 | break; |
419 | case F_WRLCK: |
420 | cephlock->type = CEPH_LOCK_EXCL; |
421 | break; |
422 | case F_UNLCK: |
423 | cephlock->type = CEPH_LOCK_UNLOCK; |
424 | break; |
425 | default: |
426 | doutc(cl, "Have unknown lock type %d\n" , |
427 | lock->c.flc_type); |
428 | err = -EINVAL; |
429 | } |
430 | |
431 | return err; |
432 | } |
433 | |
434 | /* |
435 | * Encode the flock and fcntl locks for the given inode into the ceph_filelock |
436 | * array. Must be called with inode->i_lock already held. |
437 | * If we encounter more of a specific lock type than expected, return -ENOSPC. |
438 | */ |
439 | int ceph_encode_locks_to_buffer(struct inode *inode, |
440 | struct ceph_filelock *flocks, |
441 | int num_fcntl_locks, int num_flock_locks) |
442 | { |
443 | struct file_lock *lock; |
444 | struct file_lock_context *ctx = locks_inode_context(inode); |
445 | struct ceph_client *cl = ceph_inode_to_client(inode); |
446 | int err = 0; |
447 | int seen_fcntl = 0; |
448 | int seen_flock = 0; |
449 | int l = 0; |
450 | |
451 | doutc(cl, "encoding %d flock and %d fcntl locks\n" , num_flock_locks, |
452 | num_fcntl_locks); |
453 | |
454 | if (!ctx) |
455 | return 0; |
456 | |
457 | spin_lock(lock: &ctx->flc_lock); |
458 | for_each_file_lock(lock, &ctx->flc_posix) { |
459 | ++seen_fcntl; |
460 | if (seen_fcntl > num_fcntl_locks) { |
461 | err = -ENOSPC; |
462 | goto fail; |
463 | } |
464 | err = lock_to_ceph_filelock(inode, lock, cephlock: &flocks[l]); |
465 | if (err) |
466 | goto fail; |
467 | ++l; |
468 | } |
469 | for_each_file_lock(lock, &ctx->flc_flock) { |
470 | ++seen_flock; |
471 | if (seen_flock > num_flock_locks) { |
472 | err = -ENOSPC; |
473 | goto fail; |
474 | } |
475 | err = lock_to_ceph_filelock(inode, lock, cephlock: &flocks[l]); |
476 | if (err) |
477 | goto fail; |
478 | ++l; |
479 | } |
480 | fail: |
481 | spin_unlock(lock: &ctx->flc_lock); |
482 | return err; |
483 | } |
484 | |
485 | /* |
486 | * Copy the encoded flock and fcntl locks into the pagelist. |
487 | * Format is: #fcntl locks, sequential fcntl locks, #flock locks, |
488 | * sequential flock locks. |
489 | * Returns zero on success. |
490 | */ |
491 | int ceph_locks_to_pagelist(struct ceph_filelock *flocks, |
492 | struct ceph_pagelist *pagelist, |
493 | int num_fcntl_locks, int num_flock_locks) |
494 | { |
495 | int err = 0; |
496 | __le32 nlocks; |
497 | |
498 | nlocks = cpu_to_le32(num_fcntl_locks); |
499 | err = ceph_pagelist_append(pl: pagelist, d: &nlocks, l: sizeof(nlocks)); |
500 | if (err) |
501 | goto out_fail; |
502 | |
503 | if (num_fcntl_locks > 0) { |
504 | err = ceph_pagelist_append(pl: pagelist, d: flocks, |
505 | l: num_fcntl_locks * sizeof(*flocks)); |
506 | if (err) |
507 | goto out_fail; |
508 | } |
509 | |
510 | nlocks = cpu_to_le32(num_flock_locks); |
511 | err = ceph_pagelist_append(pl: pagelist, d: &nlocks, l: sizeof(nlocks)); |
512 | if (err) |
513 | goto out_fail; |
514 | |
515 | if (num_flock_locks > 0) { |
516 | err = ceph_pagelist_append(pl: pagelist, d: &flocks[num_fcntl_locks], |
517 | l: num_flock_locks * sizeof(*flocks)); |
518 | } |
519 | out_fail: |
520 | return err; |
521 | } |
522 | |