xref: /linux/fs/ceph/locks.c (revision 4ac6d90867a4de2e12117e755dbd76e08d88697f)
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/ceph/pagelist.h>
11 
12 static u64 lock_secret;
13 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
14                                          struct ceph_mds_request *req);
15 
16 static inline u64 secure_addr(void *addr)
17 {
18 	u64 v = lock_secret ^ (u64)(unsigned long)addr;
19 	/*
20 	 * Set the most significant bit, so that MDS knows the 'owner'
21 	 * is sufficient to identify the owner of lock. (old code uses
22 	 * both 'owner' and 'pid')
23 	 */
24 	v |= (1ULL << 63);
25 	return v;
26 }
27 
28 void __init ceph_flock_init(void)
29 {
30 	get_random_bytes(&lock_secret, sizeof(lock_secret));
31 }
32 
33 static void ceph_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
34 {
35 	struct ceph_file_info *fi = dst->fl_file->private_data;
36 	struct inode *inode = file_inode(dst->fl_file);
37 	atomic_inc(&ceph_inode(inode)->i_filelock_ref);
38 	atomic_inc(&fi->num_locks);
39 }
40 
41 static void ceph_fl_release_lock(struct file_lock *fl)
42 {
43 	struct ceph_file_info *fi = fl->fl_file->private_data;
44 	struct inode *inode = file_inode(fl->fl_file);
45 	struct ceph_inode_info *ci = ceph_inode(inode);
46 	atomic_dec(&fi->num_locks);
47 	if (atomic_dec_and_test(&ci->i_filelock_ref)) {
48 		/* clear error when all locks are released */
49 		spin_lock(&ci->i_ceph_lock);
50 		ci->i_ceph_flags &= ~CEPH_I_ERROR_FILELOCK;
51 		spin_unlock(&ci->i_ceph_lock);
52 	}
53 }
54 
55 static const struct file_lock_operations ceph_fl_lock_ops = {
56 	.fl_copy_lock = ceph_fl_copy_lock,
57 	.fl_release_private = ceph_fl_release_lock,
58 };
59 
60 /*
61  * Implement fcntl and flock locking functions.
62  */
63 static int ceph_lock_message(u8 lock_type, u16 operation, struct inode *inode,
64 			     int cmd, u8 wait, struct file_lock *fl)
65 {
66 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
67 	struct ceph_mds_request *req;
68 	int err;
69 	u64 length = 0;
70 	u64 owner;
71 
72 	if (operation == CEPH_MDS_OP_SETFILELOCK) {
73 		/*
74 		 * increasing i_filelock_ref closes race window between
75 		 * handling request reply and adding file_lock struct to
76 		 * inode. Otherwise, auth caps may get trimmed in the
77 		 * window. Caller function will decrease the counter.
78 		 */
79 		fl->fl_ops = &ceph_fl_lock_ops;
80 		fl->fl_ops->fl_copy_lock(fl, NULL);
81 	}
82 
83 	if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK)
84 		wait = 0;
85 
86 	req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
87 	if (IS_ERR(req))
88 		return PTR_ERR(req);
89 	req->r_inode = inode;
90 	ihold(inode);
91 	req->r_num_caps = 1;
92 
93 	/* mds requires start and length rather than start and end */
94 	if (LLONG_MAX == fl->fl_end)
95 		length = 0;
96 	else
97 		length = fl->fl_end - fl->fl_start + 1;
98 
99 	owner = secure_addr(fl->fl_owner);
100 
101 	dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
102 	     "start: %llu, length: %llu, wait: %d, type: %d\n", (int)lock_type,
103 	     (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
104 	     wait, fl->fl_type);
105 
106 	req->r_args.filelock_change.rule = lock_type;
107 	req->r_args.filelock_change.type = cmd;
108 	req->r_args.filelock_change.owner = cpu_to_le64(owner);
109 	req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
110 	req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
111 	req->r_args.filelock_change.length = cpu_to_le64(length);
112 	req->r_args.filelock_change.wait = wait;
113 
114 	if (wait)
115 		req->r_wait_for_completion = ceph_lock_wait_for_completion;
116 
117 	err = ceph_mdsc_do_request(mdsc, inode, req);
118 	if (!err && operation == CEPH_MDS_OP_GETFILELOCK) {
119 		fl->fl_pid = -le64_to_cpu(req->r_reply_info.filelock_reply->pid);
120 		if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
121 			fl->fl_type = F_RDLCK;
122 		else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
123 			fl->fl_type = F_WRLCK;
124 		else
125 			fl->fl_type = F_UNLCK;
126 
127 		fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
128 		length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
129 						 le64_to_cpu(req->r_reply_info.filelock_reply->length);
130 		if (length >= 1)
131 			fl->fl_end = length -1;
132 		else
133 			fl->fl_end = 0;
134 
135 	}
136 	ceph_mdsc_put_request(req);
137 	dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
138 	     "length: %llu, wait: %d, type: %d, err code %d\n", (int)lock_type,
139 	     (int)operation, (u64)fl->fl_pid, fl->fl_start,
140 	     length, wait, fl->fl_type, err);
141 	return err;
142 }
143 
144 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
145                                          struct ceph_mds_request *req)
146 {
147 	struct ceph_mds_request *intr_req;
148 	struct inode *inode = req->r_inode;
149 	int err, lock_type;
150 
151 	BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK);
152 	if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL)
153 		lock_type = CEPH_LOCK_FCNTL_INTR;
154 	else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK)
155 		lock_type = CEPH_LOCK_FLOCK_INTR;
156 	else
157 		BUG_ON(1);
158 	BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK);
159 
160 	err = wait_for_completion_interruptible(&req->r_completion);
161 	if (!err)
162 		return 0;
163 
164 	dout("ceph_lock_wait_for_completion: request %llu was interrupted\n",
165 	     req->r_tid);
166 
167 	mutex_lock(&mdsc->mutex);
168 	if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
169 		err = 0;
170 	} else {
171 		/*
172 		 * ensure we aren't running concurrently with
173 		 * ceph_fill_trace or ceph_readdir_prepopulate, which
174 		 * rely on locks (dir mutex) held by our caller.
175 		 */
176 		mutex_lock(&req->r_fill_mutex);
177 		req->r_err = err;
178 		set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
179 		mutex_unlock(&req->r_fill_mutex);
180 
181 		if (!req->r_session) {
182 			// haven't sent the request
183 			err = 0;
184 		}
185 	}
186 	mutex_unlock(&mdsc->mutex);
187 	if (!err)
188 		return 0;
189 
190 	intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK,
191 					    USE_AUTH_MDS);
192 	if (IS_ERR(intr_req))
193 		return PTR_ERR(intr_req);
194 
195 	intr_req->r_inode = inode;
196 	ihold(inode);
197 	intr_req->r_num_caps = 1;
198 
199 	intr_req->r_args.filelock_change = req->r_args.filelock_change;
200 	intr_req->r_args.filelock_change.rule = lock_type;
201 	intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK;
202 
203 	err = ceph_mdsc_do_request(mdsc, inode, intr_req);
204 	ceph_mdsc_put_request(intr_req);
205 
206 	if (err && err != -ERESTARTSYS)
207 		return err;
208 
209 	wait_for_completion_killable(&req->r_safe_completion);
210 	return 0;
211 }
212 
213 static int try_unlock_file(struct file *file, struct file_lock *fl)
214 {
215 	int err;
216 	unsigned int orig_flags = fl->fl_flags;
217 	fl->fl_flags |= FL_EXISTS;
218 	err = locks_lock_file_wait(file, fl);
219 	fl->fl_flags = orig_flags;
220 	if (err == -ENOENT) {
221 		if (!(orig_flags & FL_EXISTS))
222 			err = 0;
223 		return err;
224 	}
225 	return 1;
226 }
227 
228 /*
229  * Attempt to set an fcntl lock.
230  * For now, this just goes away to the server. Later it may be more awesome.
231  */
232 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
233 {
234 	struct inode *inode = file_inode(file);
235 	struct ceph_inode_info *ci = ceph_inode(inode);
236 	int err = 0;
237 	u16 op = CEPH_MDS_OP_SETFILELOCK;
238 	u8 wait = 0;
239 	u8 lock_cmd;
240 
241 	if (!(fl->fl_flags & FL_POSIX))
242 		return -ENOLCK;
243 
244 	dout("ceph_lock, fl_owner: %p\n", fl->fl_owner);
245 
246 	/* set wait bit as appropriate, then make command as Ceph expects it*/
247 	if (IS_GETLK(cmd))
248 		op = CEPH_MDS_OP_GETFILELOCK;
249 	else if (IS_SETLKW(cmd))
250 		wait = 1;
251 
252 	spin_lock(&ci->i_ceph_lock);
253 	if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
254 		err = -EIO;
255 	}
256 	spin_unlock(&ci->i_ceph_lock);
257 	if (err < 0) {
258 		if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK == fl->fl_type)
259 			posix_lock_file(file, fl, NULL);
260 		return err;
261 	}
262 
263 	if (F_RDLCK == fl->fl_type)
264 		lock_cmd = CEPH_LOCK_SHARED;
265 	else if (F_WRLCK == fl->fl_type)
266 		lock_cmd = CEPH_LOCK_EXCL;
267 	else
268 		lock_cmd = CEPH_LOCK_UNLOCK;
269 
270 	if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK == fl->fl_type) {
271 		err = try_unlock_file(file, fl);
272 		if (err <= 0)
273 			return err;
274 	}
275 
276 	err = ceph_lock_message(CEPH_LOCK_FCNTL, op, inode, lock_cmd, wait, fl);
277 	if (!err) {
278 		if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK != fl->fl_type) {
279 			dout("mds locked, locking locally\n");
280 			err = posix_lock_file(file, fl, NULL);
281 			if (err) {
282 				/* undo! This should only happen if
283 				 * the kernel detects local
284 				 * deadlock. */
285 				ceph_lock_message(CEPH_LOCK_FCNTL, op, inode,
286 						  CEPH_LOCK_UNLOCK, 0, fl);
287 				dout("got %d on posix_lock_file, undid lock\n",
288 				     err);
289 			}
290 		}
291 	}
292 	return err;
293 }
294 
295 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
296 {
297 	struct inode *inode = file_inode(file);
298 	struct ceph_inode_info *ci = ceph_inode(inode);
299 	int err = 0;
300 	u8 wait = 0;
301 	u8 lock_cmd;
302 
303 	if (!(fl->fl_flags & FL_FLOCK))
304 		return -ENOLCK;
305 	/* No mandatory locks */
306 	if (fl->fl_type & LOCK_MAND)
307 		return -EOPNOTSUPP;
308 
309 	dout("ceph_flock, fl_file: %p\n", fl->fl_file);
310 
311 	spin_lock(&ci->i_ceph_lock);
312 	if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
313 		err = -EIO;
314 	}
315 	spin_unlock(&ci->i_ceph_lock);
316 	if (err < 0) {
317 		if (F_UNLCK == fl->fl_type)
318 			locks_lock_file_wait(file, fl);
319 		return err;
320 	}
321 
322 	if (IS_SETLKW(cmd))
323 		wait = 1;
324 
325 	if (F_RDLCK == fl->fl_type)
326 		lock_cmd = CEPH_LOCK_SHARED;
327 	else if (F_WRLCK == fl->fl_type)
328 		lock_cmd = CEPH_LOCK_EXCL;
329 	else
330 		lock_cmd = CEPH_LOCK_UNLOCK;
331 
332 	if (F_UNLCK == fl->fl_type) {
333 		err = try_unlock_file(file, fl);
334 		if (err <= 0)
335 			return err;
336 	}
337 
338 	err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
339 				inode, lock_cmd, wait, fl);
340 	if (!err && F_UNLCK != fl->fl_type) {
341 		err = locks_lock_file_wait(file, fl);
342 		if (err) {
343 			ceph_lock_message(CEPH_LOCK_FLOCK,
344 					  CEPH_MDS_OP_SETFILELOCK,
345 					  inode, CEPH_LOCK_UNLOCK, 0, fl);
346 			dout("got %d on locks_lock_file_wait, undid lock\n", err);
347 		}
348 	}
349 	return err;
350 }
351 
352 /*
353  * Fills in the passed counter variables, so you can prepare pagelist metadata
354  * before calling ceph_encode_locks.
355  */
356 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
357 {
358 	struct file_lock *lock;
359 	struct file_lock_context *ctx;
360 
361 	*fcntl_count = 0;
362 	*flock_count = 0;
363 
364 	ctx = inode->i_flctx;
365 	if (ctx) {
366 		spin_lock(&ctx->flc_lock);
367 		list_for_each_entry(lock, &ctx->flc_posix, fl_list)
368 			++(*fcntl_count);
369 		list_for_each_entry(lock, &ctx->flc_flock, fl_list)
370 			++(*flock_count);
371 		spin_unlock(&ctx->flc_lock);
372 	}
373 	dout("counted %d flock locks and %d fcntl locks\n",
374 	     *flock_count, *fcntl_count);
375 }
376 
377 /*
378  * Given a pointer to a lock, convert it to a ceph filelock
379  */
380 static int lock_to_ceph_filelock(struct file_lock *lock,
381 				 struct ceph_filelock *cephlock)
382 {
383 	int err = 0;
384 	cephlock->start = cpu_to_le64(lock->fl_start);
385 	cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
386 	cephlock->client = cpu_to_le64(0);
387 	cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
388 	cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
389 
390 	switch (lock->fl_type) {
391 	case F_RDLCK:
392 		cephlock->type = CEPH_LOCK_SHARED;
393 		break;
394 	case F_WRLCK:
395 		cephlock->type = CEPH_LOCK_EXCL;
396 		break;
397 	case F_UNLCK:
398 		cephlock->type = CEPH_LOCK_UNLOCK;
399 		break;
400 	default:
401 		dout("Have unknown lock type %d\n", lock->fl_type);
402 		err = -EINVAL;
403 	}
404 
405 	return err;
406 }
407 
408 /*
409  * Encode the flock and fcntl locks for the given inode into the ceph_filelock
410  * array. Must be called with inode->i_lock already held.
411  * If we encounter more of a specific lock type than expected, return -ENOSPC.
412  */
413 int ceph_encode_locks_to_buffer(struct inode *inode,
414 				struct ceph_filelock *flocks,
415 				int num_fcntl_locks, int num_flock_locks)
416 {
417 	struct file_lock *lock;
418 	struct file_lock_context *ctx = inode->i_flctx;
419 	int err = 0;
420 	int seen_fcntl = 0;
421 	int seen_flock = 0;
422 	int l = 0;
423 
424 	dout("encoding %d flock and %d fcntl locks\n", num_flock_locks,
425 	     num_fcntl_locks);
426 
427 	if (!ctx)
428 		return 0;
429 
430 	spin_lock(&ctx->flc_lock);
431 	list_for_each_entry(lock, &ctx->flc_posix, fl_list) {
432 		++seen_fcntl;
433 		if (seen_fcntl > num_fcntl_locks) {
434 			err = -ENOSPC;
435 			goto fail;
436 		}
437 		err = lock_to_ceph_filelock(lock, &flocks[l]);
438 		if (err)
439 			goto fail;
440 		++l;
441 	}
442 	list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
443 		++seen_flock;
444 		if (seen_flock > num_flock_locks) {
445 			err = -ENOSPC;
446 			goto fail;
447 		}
448 		err = lock_to_ceph_filelock(lock, &flocks[l]);
449 		if (err)
450 			goto fail;
451 		++l;
452 	}
453 fail:
454 	spin_unlock(&ctx->flc_lock);
455 	return err;
456 }
457 
458 /*
459  * Copy the encoded flock and fcntl locks into the pagelist.
460  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
461  * sequential flock locks.
462  * Returns zero on success.
463  */
464 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
465 			   struct ceph_pagelist *pagelist,
466 			   int num_fcntl_locks, int num_flock_locks)
467 {
468 	int err = 0;
469 	__le32 nlocks;
470 
471 	nlocks = cpu_to_le32(num_fcntl_locks);
472 	err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
473 	if (err)
474 		goto out_fail;
475 
476 	if (num_fcntl_locks > 0) {
477 		err = ceph_pagelist_append(pagelist, flocks,
478 					   num_fcntl_locks * sizeof(*flocks));
479 		if (err)
480 			goto out_fail;
481 	}
482 
483 	nlocks = cpu_to_le32(num_flock_locks);
484 	err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
485 	if (err)
486 		goto out_fail;
487 
488 	if (num_flock_locks > 0) {
489 		err = ceph_pagelist_append(pagelist, &flocks[num_fcntl_locks],
490 					   num_flock_locks * sizeof(*flocks));
491 	}
492 out_fail:
493 	return err;
494 }
495