xref: /linux/fs/dlm/plock.c (revision 81c29435073355b8194986a2193d3e7b9d449225)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2005-2008 Red Hat, Inc.  All rights reserved.
4  */
5 
6 #include <linux/fs.h>
7 #include <linux/filelock.h>
8 #include <linux/miscdevice.h>
9 #include <linux/poll.h>
10 #include <linux/dlm.h>
11 #include <linux/dlm_plock.h>
12 #include <linux/slab.h>
13 
14 #include "dlm_internal.h"
15 #include "lockspace.h"
16 
17 static DEFINE_SPINLOCK(ops_lock);
18 static LIST_HEAD(send_list);
19 static LIST_HEAD(recv_list);
20 static DECLARE_WAIT_QUEUE_HEAD(send_wq);
21 static DECLARE_WAIT_QUEUE_HEAD(recv_wq);
22 
23 struct plock_async_data {
24 	void *fl;
25 	void *file;
26 	struct file_lock flc;
27 	int (*callback)(struct file_lock *fl, int result);
28 };
29 
30 struct plock_op {
31 	struct list_head list;
32 	int done;
33 	/* if lock op got interrupted while waiting dlm_controld reply */
34 	bool sigint;
35 	struct dlm_plock_info info;
36 	/* if set indicates async handling */
37 	struct plock_async_data *data;
38 };
39 
40 static inline void set_version(struct dlm_plock_info *info)
41 {
42 	info->version[0] = DLM_PLOCK_VERSION_MAJOR;
43 	info->version[1] = DLM_PLOCK_VERSION_MINOR;
44 	info->version[2] = DLM_PLOCK_VERSION_PATCH;
45 }
46 
47 static int check_version(struct dlm_plock_info *info)
48 {
49 	if ((DLM_PLOCK_VERSION_MAJOR != info->version[0]) ||
50 	    (DLM_PLOCK_VERSION_MINOR < info->version[1])) {
51 		log_print("plock device version mismatch: "
52 			  "kernel (%u.%u.%u), user (%u.%u.%u)",
53 			  DLM_PLOCK_VERSION_MAJOR,
54 			  DLM_PLOCK_VERSION_MINOR,
55 			  DLM_PLOCK_VERSION_PATCH,
56 			  info->version[0],
57 			  info->version[1],
58 			  info->version[2]);
59 		return -EINVAL;
60 	}
61 	return 0;
62 }
63 
64 static void dlm_release_plock_op(struct plock_op *op)
65 {
66 	kfree(op->data);
67 	kfree(op);
68 }
69 
70 static void send_op(struct plock_op *op)
71 {
72 	set_version(&op->info);
73 	spin_lock(&ops_lock);
74 	list_add_tail(&op->list, &send_list);
75 	spin_unlock(&ops_lock);
76 	wake_up(&send_wq);
77 }
78 
79 /* If a process was killed while waiting for the only plock on a file,
80    locks_remove_posix will not see any lock on the file so it won't
81    send an unlock-close to us to pass on to userspace to clean up the
82    abandoned waiter.  So, we have to insert the unlock-close when the
83    lock call is interrupted. */
84 
85 static void do_unlock_close(const struct dlm_plock_info *info)
86 {
87 	struct plock_op *op;
88 
89 	op = kzalloc(sizeof(*op), GFP_NOFS);
90 	if (!op)
91 		return;
92 
93 	op->info.optype		= DLM_PLOCK_OP_UNLOCK;
94 	op->info.pid		= info->pid;
95 	op->info.fsid		= info->fsid;
96 	op->info.number		= info->number;
97 	op->info.start		= 0;
98 	op->info.end		= OFFSET_MAX;
99 	op->info.owner		= info->owner;
100 
101 	op->info.flags |= DLM_PLOCK_FL_CLOSE;
102 	send_op(op);
103 }
104 
105 int dlm_posix_lock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
106 		   int cmd, struct file_lock *fl)
107 {
108 	struct plock_async_data *op_data;
109 	struct dlm_ls *ls;
110 	struct plock_op *op;
111 	int rv;
112 
113 	ls = dlm_find_lockspace_local(lockspace);
114 	if (!ls)
115 		return -EINVAL;
116 
117 	op = kzalloc(sizeof(*op), GFP_NOFS);
118 	if (!op) {
119 		rv = -ENOMEM;
120 		goto out;
121 	}
122 
123 	op->info.optype		= DLM_PLOCK_OP_LOCK;
124 	op->info.pid		= fl->fl_pid;
125 	op->info.ex		= (fl->fl_type == F_WRLCK);
126 	op->info.wait		= IS_SETLKW(cmd);
127 	op->info.fsid		= ls->ls_global_id;
128 	op->info.number		= number;
129 	op->info.start		= fl->fl_start;
130 	op->info.end		= fl->fl_end;
131 	/* async handling */
132 	if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
133 		op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
134 		if (!op_data) {
135 			dlm_release_plock_op(op);
136 			rv = -ENOMEM;
137 			goto out;
138 		}
139 
140 		/* fl_owner is lockd which doesn't distinguish
141 		   processes on the nfs client */
142 		op->info.owner	= (__u64) fl->fl_pid;
143 		op_data->callback = fl->fl_lmops->lm_grant;
144 		locks_init_lock(&op_data->flc);
145 		locks_copy_lock(&op_data->flc, fl);
146 		op_data->fl		= fl;
147 		op_data->file	= file;
148 
149 		op->data = op_data;
150 
151 		send_op(op);
152 		rv = FILE_LOCK_DEFERRED;
153 		goto out;
154 	} else {
155 		op->info.owner	= (__u64)(long) fl->fl_owner;
156 	}
157 
158 	send_op(op);
159 
160 	rv = wait_event_interruptible(recv_wq, (op->done != 0));
161 	if (rv == -ERESTARTSYS) {
162 		spin_lock(&ops_lock);
163 		/* recheck under ops_lock if we got a done != 0,
164 		 * if so this interrupt case should be ignored
165 		 */
166 		if (op->done != 0) {
167 			spin_unlock(&ops_lock);
168 			goto do_lock_wait;
169 		}
170 
171 		op->sigint = true;
172 		spin_unlock(&ops_lock);
173 		log_debug(ls, "%s: wait interrupted %x %llx pid %d",
174 			  __func__, ls->ls_global_id,
175 			  (unsigned long long)number, op->info.pid);
176 		goto out;
177 	}
178 
179 do_lock_wait:
180 
181 	WARN_ON(!list_empty(&op->list));
182 
183 	rv = op->info.rv;
184 
185 	if (!rv) {
186 		if (locks_lock_file_wait(file, fl) < 0)
187 			log_error(ls, "dlm_posix_lock: vfs lock error %llx",
188 				  (unsigned long long)number);
189 	}
190 
191 	dlm_release_plock_op(op);
192 out:
193 	dlm_put_lockspace(ls);
194 	return rv;
195 }
196 EXPORT_SYMBOL_GPL(dlm_posix_lock);
197 
198 /* Returns failure iff a successful lock operation should be canceled */
199 static int dlm_plock_callback(struct plock_op *op)
200 {
201 	struct plock_async_data *op_data = op->data;
202 	struct file *file;
203 	struct file_lock *fl;
204 	struct file_lock *flc;
205 	int (*notify)(struct file_lock *fl, int result) = NULL;
206 	int rv = 0;
207 
208 	WARN_ON(!list_empty(&op->list));
209 
210 	/* check if the following 2 are still valid or make a copy */
211 	file = op_data->file;
212 	flc = &op_data->flc;
213 	fl = op_data->fl;
214 	notify = op_data->callback;
215 
216 	if (op->info.rv) {
217 		notify(fl, op->info.rv);
218 		goto out;
219 	}
220 
221 	/* got fs lock; bookkeep locally as well: */
222 	flc->fl_flags &= ~FL_SLEEP;
223 	if (posix_lock_file(file, flc, NULL)) {
224 		/*
225 		 * This can only happen in the case of kmalloc() failure.
226 		 * The filesystem's own lock is the authoritative lock,
227 		 * so a failure to get the lock locally is not a disaster.
228 		 * As long as the fs cannot reliably cancel locks (especially
229 		 * in a low-memory situation), we're better off ignoring
230 		 * this failure than trying to recover.
231 		 */
232 		log_print("dlm_plock_callback: vfs lock error %llx file %p fl %p",
233 			  (unsigned long long)op->info.number, file, fl);
234 	}
235 
236 	rv = notify(fl, 0);
237 	if (rv) {
238 		/* XXX: We need to cancel the fs lock here: */
239 		log_print("dlm_plock_callback: lock granted after lock request "
240 			  "failed; dangling lock!\n");
241 		goto out;
242 	}
243 
244 out:
245 	dlm_release_plock_op(op);
246 	return rv;
247 }
248 
249 int dlm_posix_unlock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
250 		     struct file_lock *fl)
251 {
252 	struct dlm_ls *ls;
253 	struct plock_op *op;
254 	int rv;
255 	unsigned char fl_flags = fl->fl_flags;
256 
257 	ls = dlm_find_lockspace_local(lockspace);
258 	if (!ls)
259 		return -EINVAL;
260 
261 	op = kzalloc(sizeof(*op), GFP_NOFS);
262 	if (!op) {
263 		rv = -ENOMEM;
264 		goto out;
265 	}
266 
267 	/* cause the vfs unlock to return ENOENT if lock is not found */
268 	fl->fl_flags |= FL_EXISTS;
269 
270 	rv = locks_lock_file_wait(file, fl);
271 	if (rv == -ENOENT) {
272 		rv = 0;
273 		goto out_free;
274 	}
275 	if (rv < 0) {
276 		log_error(ls, "dlm_posix_unlock: vfs unlock error %d %llx",
277 			  rv, (unsigned long long)number);
278 	}
279 
280 	op->info.optype		= DLM_PLOCK_OP_UNLOCK;
281 	op->info.pid		= fl->fl_pid;
282 	op->info.fsid		= ls->ls_global_id;
283 	op->info.number		= number;
284 	op->info.start		= fl->fl_start;
285 	op->info.end		= fl->fl_end;
286 	if (fl->fl_lmops && fl->fl_lmops->lm_grant)
287 		op->info.owner	= (__u64) fl->fl_pid;
288 	else
289 		op->info.owner	= (__u64)(long) fl->fl_owner;
290 
291 	if (fl->fl_flags & FL_CLOSE) {
292 		op->info.flags |= DLM_PLOCK_FL_CLOSE;
293 		send_op(op);
294 		rv = 0;
295 		goto out;
296 	}
297 
298 	send_op(op);
299 	wait_event(recv_wq, (op->done != 0));
300 
301 	WARN_ON(!list_empty(&op->list));
302 
303 	rv = op->info.rv;
304 
305 	if (rv == -ENOENT)
306 		rv = 0;
307 
308 out_free:
309 	dlm_release_plock_op(op);
310 out:
311 	dlm_put_lockspace(ls);
312 	fl->fl_flags = fl_flags;
313 	return rv;
314 }
315 EXPORT_SYMBOL_GPL(dlm_posix_unlock);
316 
317 int dlm_posix_get(dlm_lockspace_t *lockspace, u64 number, struct file *file,
318 		  struct file_lock *fl)
319 {
320 	struct dlm_ls *ls;
321 	struct plock_op *op;
322 	int rv;
323 
324 	ls = dlm_find_lockspace_local(lockspace);
325 	if (!ls)
326 		return -EINVAL;
327 
328 	op = kzalloc(sizeof(*op), GFP_NOFS);
329 	if (!op) {
330 		rv = -ENOMEM;
331 		goto out;
332 	}
333 
334 	op->info.optype		= DLM_PLOCK_OP_GET;
335 	op->info.pid		= fl->fl_pid;
336 	op->info.ex		= (fl->fl_type == F_WRLCK);
337 	op->info.fsid		= ls->ls_global_id;
338 	op->info.number		= number;
339 	op->info.start		= fl->fl_start;
340 	op->info.end		= fl->fl_end;
341 	if (fl->fl_lmops && fl->fl_lmops->lm_grant)
342 		op->info.owner	= (__u64) fl->fl_pid;
343 	else
344 		op->info.owner	= (__u64)(long) fl->fl_owner;
345 
346 	send_op(op);
347 	wait_event(recv_wq, (op->done != 0));
348 
349 	WARN_ON(!list_empty(&op->list));
350 
351 	/* info.rv from userspace is 1 for conflict, 0 for no-conflict,
352 	   -ENOENT if there are no locks on the file */
353 
354 	rv = op->info.rv;
355 
356 	fl->fl_type = F_UNLCK;
357 	if (rv == -ENOENT)
358 		rv = 0;
359 	else if (rv > 0) {
360 		locks_init_lock(fl);
361 		fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
362 		fl->fl_flags = FL_POSIX;
363 		fl->fl_pid = -op->info.pid;
364 		fl->fl_start = op->info.start;
365 		fl->fl_end = op->info.end;
366 		rv = 0;
367 	}
368 
369 	dlm_release_plock_op(op);
370 out:
371 	dlm_put_lockspace(ls);
372 	return rv;
373 }
374 EXPORT_SYMBOL_GPL(dlm_posix_get);
375 
376 /* a read copies out one plock request from the send list */
377 static ssize_t dev_read(struct file *file, char __user *u, size_t count,
378 			loff_t *ppos)
379 {
380 	struct dlm_plock_info info;
381 	struct plock_op *op = NULL;
382 
383 	if (count < sizeof(info))
384 		return -EINVAL;
385 
386 	spin_lock(&ops_lock);
387 	if (!list_empty(&send_list)) {
388 		op = list_first_entry(&send_list, struct plock_op, list);
389 		if (op->info.flags & DLM_PLOCK_FL_CLOSE)
390 			list_del(&op->list);
391 		else
392 			list_move(&op->list, &recv_list);
393 		memcpy(&info, &op->info, sizeof(info));
394 	}
395 	spin_unlock(&ops_lock);
396 
397 	if (!op)
398 		return -EAGAIN;
399 
400 	/* there is no need to get a reply from userspace for unlocks
401 	   that were generated by the vfs cleaning up for a close
402 	   (the process did not make an unlock call). */
403 
404 	if (op->info.flags & DLM_PLOCK_FL_CLOSE)
405 		dlm_release_plock_op(op);
406 
407 	if (copy_to_user(u, &info, sizeof(info)))
408 		return -EFAULT;
409 	return sizeof(info);
410 }
411 
412 /* a write copies in one plock result that should match a plock_op
413    on the recv list */
414 static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
415 			 loff_t *ppos)
416 {
417 	struct plock_op *op = NULL, *iter;
418 	struct dlm_plock_info info;
419 	int do_callback = 0;
420 
421 	if (count != sizeof(info))
422 		return -EINVAL;
423 
424 	if (copy_from_user(&info, u, sizeof(info)))
425 		return -EFAULT;
426 
427 	if (check_version(&info))
428 		return -EINVAL;
429 
430 	spin_lock(&ops_lock);
431 	list_for_each_entry(iter, &recv_list, list) {
432 		if (iter->info.fsid == info.fsid &&
433 		    iter->info.number == info.number &&
434 		    iter->info.owner == info.owner) {
435 			if (iter->sigint) {
436 				list_del(&iter->list);
437 				spin_unlock(&ops_lock);
438 
439 				pr_debug("%s: sigint cleanup %x %llx pid %d",
440 					  __func__, iter->info.fsid,
441 					  (unsigned long long)iter->info.number,
442 					  iter->info.pid);
443 				do_unlock_close(&iter->info);
444 				memcpy(&iter->info, &info, sizeof(info));
445 				dlm_release_plock_op(iter);
446 				return count;
447 			}
448 			list_del_init(&iter->list);
449 			memcpy(&iter->info, &info, sizeof(info));
450 			if (iter->data)
451 				do_callback = 1;
452 			else
453 				iter->done = 1;
454 			op = iter;
455 			break;
456 		}
457 	}
458 	spin_unlock(&ops_lock);
459 
460 	if (op) {
461 		if (do_callback)
462 			dlm_plock_callback(op);
463 		else
464 			wake_up(&recv_wq);
465 	} else
466 		log_print("%s: no op %x %llx", __func__,
467 			  info.fsid, (unsigned long long)info.number);
468 	return count;
469 }
470 
471 static __poll_t dev_poll(struct file *file, poll_table *wait)
472 {
473 	__poll_t mask = 0;
474 
475 	poll_wait(file, &send_wq, wait);
476 
477 	spin_lock(&ops_lock);
478 	if (!list_empty(&send_list))
479 		mask = EPOLLIN | EPOLLRDNORM;
480 	spin_unlock(&ops_lock);
481 
482 	return mask;
483 }
484 
485 static const struct file_operations dev_fops = {
486 	.read    = dev_read,
487 	.write   = dev_write,
488 	.poll    = dev_poll,
489 	.owner   = THIS_MODULE,
490 	.llseek  = noop_llseek,
491 };
492 
493 static struct miscdevice plock_dev_misc = {
494 	.minor = MISC_DYNAMIC_MINOR,
495 	.name = DLM_PLOCK_MISC_NAME,
496 	.fops = &dev_fops
497 };
498 
499 int dlm_plock_init(void)
500 {
501 	int rv;
502 
503 	rv = misc_register(&plock_dev_misc);
504 	if (rv)
505 		log_print("dlm_plock_init: misc_register failed %d", rv);
506 	return rv;
507 }
508 
509 void dlm_plock_exit(void)
510 {
511 	misc_deregister(&plock_dev_misc);
512 }
513 
514