xref: /linux/fs/ceph/locks.c (revision 3a485c2be49dde899e506684bb5f06d5b3f73e82)
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 /**
34  * Implement fcntl and flock locking functions.
35  */
36 static int ceph_lock_message(u8 lock_type, u16 operation, struct file *file,
37 			     int cmd, u8 wait, struct file_lock *fl)
38 {
39 	struct inode *inode = file_inode(file);
40 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
41 	struct ceph_mds_request *req;
42 	int err;
43 	u64 length = 0;
44 	u64 owner;
45 
46 	if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK)
47 		wait = 0;
48 
49 	req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
50 	if (IS_ERR(req))
51 		return PTR_ERR(req);
52 	req->r_inode = inode;
53 	ihold(inode);
54 	req->r_num_caps = 1;
55 
56 	/* mds requires start and length rather than start and end */
57 	if (LLONG_MAX == fl->fl_end)
58 		length = 0;
59 	else
60 		length = fl->fl_end - fl->fl_start + 1;
61 
62 	owner = secure_addr(fl->fl_owner);
63 
64 	dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
65 	     "start: %llu, length: %llu, wait: %d, type: %d", (int)lock_type,
66 	     (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
67 	     wait, fl->fl_type);
68 
69 	req->r_args.filelock_change.rule = lock_type;
70 	req->r_args.filelock_change.type = cmd;
71 	req->r_args.filelock_change.owner = cpu_to_le64(owner);
72 	req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
73 	req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
74 	req->r_args.filelock_change.length = cpu_to_le64(length);
75 	req->r_args.filelock_change.wait = wait;
76 
77 	if (wait)
78 		req->r_wait_for_completion = ceph_lock_wait_for_completion;
79 
80 	err = ceph_mdsc_do_request(mdsc, inode, req);
81 
82 	if (operation == CEPH_MDS_OP_GETFILELOCK) {
83 		fl->fl_pid = -le64_to_cpu(req->r_reply_info.filelock_reply->pid);
84 		if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
85 			fl->fl_type = F_RDLCK;
86 		else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
87 			fl->fl_type = F_WRLCK;
88 		else
89 			fl->fl_type = F_UNLCK;
90 
91 		fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
92 		length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
93 						 le64_to_cpu(req->r_reply_info.filelock_reply->length);
94 		if (length >= 1)
95 			fl->fl_end = length -1;
96 		else
97 			fl->fl_end = 0;
98 
99 	}
100 	ceph_mdsc_put_request(req);
101 	dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
102 	     "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
103 	     (int)operation, (u64)fl->fl_pid, fl->fl_start,
104 	     length, wait, fl->fl_type, err);
105 	return err;
106 }
107 
108 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
109                                          struct ceph_mds_request *req)
110 {
111 	struct ceph_mds_request *intr_req;
112 	struct inode *inode = req->r_inode;
113 	int err, lock_type;
114 
115 	BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK);
116 	if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL)
117 		lock_type = CEPH_LOCK_FCNTL_INTR;
118 	else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK)
119 		lock_type = CEPH_LOCK_FLOCK_INTR;
120 	else
121 		BUG_ON(1);
122 	BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK);
123 
124 	err = wait_for_completion_interruptible(&req->r_completion);
125 	if (!err)
126 		return 0;
127 
128 	dout("ceph_lock_wait_for_completion: request %llu was interrupted\n",
129 	     req->r_tid);
130 
131 	mutex_lock(&mdsc->mutex);
132 	if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
133 		err = 0;
134 	} else {
135 		/*
136 		 * ensure we aren't running concurrently with
137 		 * ceph_fill_trace or ceph_readdir_prepopulate, which
138 		 * rely on locks (dir mutex) held by our caller.
139 		 */
140 		mutex_lock(&req->r_fill_mutex);
141 		req->r_err = err;
142 		set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
143 		mutex_unlock(&req->r_fill_mutex);
144 
145 		if (!req->r_session) {
146 			// haven't sent the request
147 			err = 0;
148 		}
149 	}
150 	mutex_unlock(&mdsc->mutex);
151 	if (!err)
152 		return 0;
153 
154 	intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK,
155 					    USE_AUTH_MDS);
156 	if (IS_ERR(intr_req))
157 		return PTR_ERR(intr_req);
158 
159 	intr_req->r_inode = inode;
160 	ihold(inode);
161 	intr_req->r_num_caps = 1;
162 
163 	intr_req->r_args.filelock_change = req->r_args.filelock_change;
164 	intr_req->r_args.filelock_change.rule = lock_type;
165 	intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK;
166 
167 	err = ceph_mdsc_do_request(mdsc, inode, intr_req);
168 	ceph_mdsc_put_request(intr_req);
169 
170 	if (err && err != -ERESTARTSYS)
171 		return err;
172 
173 	wait_for_completion_killable(&req->r_safe_completion);
174 	return 0;
175 }
176 
177 /**
178  * Attempt to set an fcntl lock.
179  * For now, this just goes away to the server. Later it may be more awesome.
180  */
181 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
182 {
183 	u8 lock_cmd;
184 	int err;
185 	u8 wait = 0;
186 	u16 op = CEPH_MDS_OP_SETFILELOCK;
187 
188 	if (!(fl->fl_flags & FL_POSIX))
189 		return -ENOLCK;
190 	/* No mandatory locks */
191 	if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
192 		return -ENOLCK;
193 
194 	dout("ceph_lock, fl_owner: %p", fl->fl_owner);
195 
196 	/* set wait bit as appropriate, then make command as Ceph expects it*/
197 	if (IS_GETLK(cmd))
198 		op = CEPH_MDS_OP_GETFILELOCK;
199 	else if (IS_SETLKW(cmd))
200 		wait = 1;
201 
202 	if (F_RDLCK == fl->fl_type)
203 		lock_cmd = CEPH_LOCK_SHARED;
204 	else if (F_WRLCK == fl->fl_type)
205 		lock_cmd = CEPH_LOCK_EXCL;
206 	else
207 		lock_cmd = CEPH_LOCK_UNLOCK;
208 
209 	err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
210 	if (!err) {
211 		if (op != CEPH_MDS_OP_GETFILELOCK) {
212 			dout("mds locked, locking locally");
213 			err = posix_lock_file(file, fl, NULL);
214 			if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
215 				/* undo! This should only happen if
216 				 * the kernel detects local
217 				 * deadlock. */
218 				ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
219 						  CEPH_LOCK_UNLOCK, 0, fl);
220 				dout("got %d on posix_lock_file, undid lock",
221 				     err);
222 			}
223 		}
224 	}
225 	return err;
226 }
227 
228 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
229 {
230 	u8 lock_cmd;
231 	int err;
232 	u8 wait = 0;
233 
234 	if (!(fl->fl_flags & FL_FLOCK))
235 		return -ENOLCK;
236 	/* No mandatory locks */
237 	if (fl->fl_type & LOCK_MAND)
238 		return -EOPNOTSUPP;
239 
240 	dout("ceph_flock, fl_file: %p", fl->fl_file);
241 
242 	if (IS_SETLKW(cmd))
243 		wait = 1;
244 
245 	if (F_RDLCK == fl->fl_type)
246 		lock_cmd = CEPH_LOCK_SHARED;
247 	else if (F_WRLCK == fl->fl_type)
248 		lock_cmd = CEPH_LOCK_EXCL;
249 	else
250 		lock_cmd = CEPH_LOCK_UNLOCK;
251 
252 	err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
253 				file, lock_cmd, wait, fl);
254 	if (!err) {
255 		err = locks_lock_file_wait(file, fl);
256 		if (err) {
257 			ceph_lock_message(CEPH_LOCK_FLOCK,
258 					  CEPH_MDS_OP_SETFILELOCK,
259 					  file, CEPH_LOCK_UNLOCK, 0, fl);
260 			dout("got %d on locks_lock_file_wait, undid lock", err);
261 		}
262 	}
263 	return err;
264 }
265 
266 /*
267  * Fills in the passed counter variables, so you can prepare pagelist metadata
268  * before calling ceph_encode_locks.
269  */
270 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
271 {
272 	struct file_lock *lock;
273 	struct file_lock_context *ctx;
274 
275 	*fcntl_count = 0;
276 	*flock_count = 0;
277 
278 	ctx = inode->i_flctx;
279 	if (ctx) {
280 		spin_lock(&ctx->flc_lock);
281 		list_for_each_entry(lock, &ctx->flc_posix, fl_list)
282 			++(*fcntl_count);
283 		list_for_each_entry(lock, &ctx->flc_flock, fl_list)
284 			++(*flock_count);
285 		spin_unlock(&ctx->flc_lock);
286 	}
287 	dout("counted %d flock locks and %d fcntl locks",
288 	     *flock_count, *fcntl_count);
289 }
290 
291 /**
292  * Encode the flock and fcntl locks for the given inode into the ceph_filelock
293  * array. Must be called with inode->i_lock already held.
294  * If we encounter more of a specific lock type than expected, return -ENOSPC.
295  */
296 int ceph_encode_locks_to_buffer(struct inode *inode,
297 				struct ceph_filelock *flocks,
298 				int num_fcntl_locks, int num_flock_locks)
299 {
300 	struct file_lock *lock;
301 	struct file_lock_context *ctx = inode->i_flctx;
302 	int err = 0;
303 	int seen_fcntl = 0;
304 	int seen_flock = 0;
305 	int l = 0;
306 
307 	dout("encoding %d flock and %d fcntl locks", num_flock_locks,
308 	     num_fcntl_locks);
309 
310 	if (!ctx)
311 		return 0;
312 
313 	spin_lock(&ctx->flc_lock);
314 	list_for_each_entry(lock, &ctx->flc_posix, fl_list) {
315 		++seen_fcntl;
316 		if (seen_fcntl > num_fcntl_locks) {
317 			err = -ENOSPC;
318 			goto fail;
319 		}
320 		err = lock_to_ceph_filelock(lock, &flocks[l]);
321 		if (err)
322 			goto fail;
323 		++l;
324 	}
325 	list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
326 		++seen_flock;
327 		if (seen_flock > num_flock_locks) {
328 			err = -ENOSPC;
329 			goto fail;
330 		}
331 		err = lock_to_ceph_filelock(lock, &flocks[l]);
332 		if (err)
333 			goto fail;
334 		++l;
335 	}
336 fail:
337 	spin_unlock(&ctx->flc_lock);
338 	return err;
339 }
340 
341 /**
342  * Copy the encoded flock and fcntl locks into the pagelist.
343  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
344  * sequential flock locks.
345  * Returns zero on success.
346  */
347 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
348 			   struct ceph_pagelist *pagelist,
349 			   int num_fcntl_locks, int num_flock_locks)
350 {
351 	int err = 0;
352 	__le32 nlocks;
353 
354 	nlocks = cpu_to_le32(num_fcntl_locks);
355 	err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
356 	if (err)
357 		goto out_fail;
358 
359 	err = ceph_pagelist_append(pagelist, flocks,
360 				   num_fcntl_locks * sizeof(*flocks));
361 	if (err)
362 		goto out_fail;
363 
364 	nlocks = cpu_to_le32(num_flock_locks);
365 	err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
366 	if (err)
367 		goto out_fail;
368 
369 	err = ceph_pagelist_append(pagelist,
370 				   &flocks[num_fcntl_locks],
371 				   num_flock_locks * sizeof(*flocks));
372 out_fail:
373 	return err;
374 }
375 
376 /*
377  * Given a pointer to a lock, convert it to a ceph filelock
378  */
379 int lock_to_ceph_filelock(struct file_lock *lock,
380 			  struct ceph_filelock *cephlock)
381 {
382 	int err = 0;
383 	cephlock->start = cpu_to_le64(lock->fl_start);
384 	cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
385 	cephlock->client = cpu_to_le64(0);
386 	cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
387 	cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
388 
389 	switch (lock->fl_type) {
390 	case F_RDLCK:
391 		cephlock->type = CEPH_LOCK_SHARED;
392 		break;
393 	case F_WRLCK:
394 		cephlock->type = CEPH_LOCK_EXCL;
395 		break;
396 	case F_UNLCK:
397 		cephlock->type = CEPH_LOCK_UNLOCK;
398 		break;
399 	default:
400 		dout("Have unknown lock type %d", lock->fl_type);
401 		err = -EINVAL;
402 	}
403 
404 	return err;
405 }
406