locks.c 12.6 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
2
#include <linux/ceph/ceph_debug.h>
G
Greg Farnum 已提交
3 4 5

#include <linux/file.h>
#include <linux/namei.h>
6
#include <linux/random.h>
G
Greg Farnum 已提交
7 8 9

#include "super.h"
#include "mds_client.h"
10
#include <linux/ceph/pagelist.h>
G
Greg Farnum 已提交
11

12
static u64 lock_secret;
Y
Yan, Zheng 已提交
13 14
static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
                                         struct ceph_mds_request *req);
15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32

static inline u64 secure_addr(void *addr)
{
	u64 v = lock_secret ^ (u64)(unsigned long)addr;
	/*
	 * Set the most significant bit, so that MDS knows the 'owner'
	 * is sufficient to identify the owner of lock. (old code uses
	 * both 'owner' and 'pid')
	 */
	v |= (1ULL << 63);
	return v;
}

void __init ceph_flock_init(void)
{
	get_random_bytes(&lock_secret, sizeof(lock_secret));
}

33 34 35 36 37 38 39 40 41
static void ceph_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
{
	struct inode *inode = file_inode(src->fl_file);
	atomic_inc(&ceph_inode(inode)->i_filelock_ref);
}

static void ceph_fl_release_lock(struct file_lock *fl)
{
	struct inode *inode = file_inode(fl->fl_file);
42 43 44 45 46 47 48
	struct ceph_inode_info *ci = ceph_inode(inode);
	if (atomic_dec_and_test(&ci->i_filelock_ref)) {
		/* clear error when all locks are released */
		spin_lock(&ci->i_ceph_lock);
		ci->i_ceph_flags &= ~CEPH_I_ERROR_FILELOCK;
		spin_unlock(&ci->i_ceph_lock);
	}
49 50 51 52 53 54 55
}

static const struct file_lock_operations ceph_fl_lock_ops = {
	.fl_copy_lock = ceph_fl_copy_lock,
	.fl_release_private = ceph_fl_release_lock,
};

G
Greg Farnum 已提交
56 57 58
/**
 * Implement fcntl and flock locking functions.
 */
59
static int ceph_lock_message(u8 lock_type, u16 operation, struct inode *inode,
60
			     int cmd, u8 wait, struct file_lock *fl)
G
Greg Farnum 已提交
61
{
62
	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
G
Greg Farnum 已提交
63 64
	struct ceph_mds_request *req;
	int err;
65
	u64 length = 0;
66
	u64 owner;
G
Greg Farnum 已提交
67

68 69 70 71 72 73 74 75 76 77 78
	if (operation == CEPH_MDS_OP_SETFILELOCK) {
		/*
		 * increasing i_filelock_ref closes race window between
		 * handling request reply and adding file_lock struct to
		 * inode. Otherwise, auth caps may get trimmed in the
		 * window. Caller function will decrease the counter.
		 */
		fl->fl_ops = &ceph_fl_lock_ops;
		atomic_inc(&ceph_inode(inode)->i_filelock_ref);
	}

Y
Yan, Zheng 已提交
79 80 81
	if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK)
		wait = 0;

G
Greg Farnum 已提交
82 83 84
	req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
	if (IS_ERR(req))
		return PTR_ERR(req);
85 86
	req->r_inode = inode;
	ihold(inode);
87
	req->r_num_caps = 1;
G
Greg Farnum 已提交
88

89 90 91 92 93 94
	/* mds requires start and length rather than start and end */
	if (LLONG_MAX == fl->fl_end)
		length = 0;
	else
		length = fl->fl_end - fl->fl_start + 1;

95
	owner = secure_addr(fl->fl_owner);
96 97 98 99 100

	dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
	     "start: %llu, length: %llu, wait: %d, type: %d", (int)lock_type,
	     (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
	     wait, fl->fl_type);
101

G
Greg Farnum 已提交
102 103
	req->r_args.filelock_change.rule = lock_type;
	req->r_args.filelock_change.type = cmd;
104
	req->r_args.filelock_change.owner = cpu_to_le64(owner);
105 106
	req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
	req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
G
Greg Farnum 已提交
107 108 109
	req->r_args.filelock_change.length = cpu_to_le64(length);
	req->r_args.filelock_change.wait = wait;

Y
Yan, Zheng 已提交
110 111 112
	if (wait)
		req->r_wait_for_completion = ceph_lock_wait_for_completion;

G
Greg Farnum 已提交
113
	err = ceph_mdsc_do_request(mdsc, inode, req);
114

115
	if (operation == CEPH_MDS_OP_GETFILELOCK) {
116
		fl->fl_pid = -le64_to_cpu(req->r_reply_info.filelock_reply->pid);
117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132
		if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
			fl->fl_type = F_RDLCK;
		else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
			fl->fl_type = F_WRLCK;
		else
			fl->fl_type = F_UNLCK;

		fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
		length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
						 le64_to_cpu(req->r_reply_info.filelock_reply->length);
		if (length >= 1)
			fl->fl_end = length -1;
		else
			fl->fl_end = 0;

	}
G
Greg Farnum 已提交
133 134
	ceph_mdsc_put_request(req);
	dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
S
Sage Weil 已提交
135
	     "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
136 137
	     (int)operation, (u64)fl->fl_pid, fl->fl_start,
	     length, wait, fl->fl_type, err);
G
Greg Farnum 已提交
138 139 140
	return err;
}

Y
Yan, Zheng 已提交
141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163
static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
                                         struct ceph_mds_request *req)
{
	struct ceph_mds_request *intr_req;
	struct inode *inode = req->r_inode;
	int err, lock_type;

	BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK);
	if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL)
		lock_type = CEPH_LOCK_FCNTL_INTR;
	else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK)
		lock_type = CEPH_LOCK_FLOCK_INTR;
	else
		BUG_ON(1);
	BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK);

	err = wait_for_completion_interruptible(&req->r_completion);
	if (!err)
		return 0;

	dout("ceph_lock_wait_for_completion: request %llu was interrupted\n",
	     req->r_tid);

164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186
	mutex_lock(&mdsc->mutex);
	if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
		err = 0;
	} else {
		/*
		 * ensure we aren't running concurrently with
		 * ceph_fill_trace or ceph_readdir_prepopulate, which
		 * rely on locks (dir mutex) held by our caller.
		 */
		mutex_lock(&req->r_fill_mutex);
		req->r_err = err;
		set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
		mutex_unlock(&req->r_fill_mutex);

		if (!req->r_session) {
			// haven't sent the request
			err = 0;
		}
	}
	mutex_unlock(&mdsc->mutex);
	if (!err)
		return 0;

Y
Yan, Zheng 已提交
187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205
	intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK,
					    USE_AUTH_MDS);
	if (IS_ERR(intr_req))
		return PTR_ERR(intr_req);

	intr_req->r_inode = inode;
	ihold(inode);
	intr_req->r_num_caps = 1;

	intr_req->r_args.filelock_change = req->r_args.filelock_change;
	intr_req->r_args.filelock_change.rule = lock_type;
	intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK;

	err = ceph_mdsc_do_request(mdsc, inode, intr_req);
	ceph_mdsc_put_request(intr_req);

	if (err && err != -ERESTARTSYS)
		return err;

206
	wait_for_completion_killable(&req->r_safe_completion);
Y
Yan, Zheng 已提交
207 208 209
	return 0;
}

G
Greg Farnum 已提交
210 211 212 213 214 215
/**
 * Attempt to set an fcntl lock.
 * For now, this just goes away to the server. Later it may be more awesome.
 */
int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
{
216
	struct inode *inode = file_inode(file);
217 218
	struct ceph_inode_info *ci = ceph_inode(inode);
	int err = 0;
G
Greg Farnum 已提交
219
	u16 op = CEPH_MDS_OP_SETFILELOCK;
220
	u8 wait = 0;
221
	u8 lock_cmd;
G
Greg Farnum 已提交
222

Y
Yan, Zheng 已提交
223 224 225 226 227 228
	if (!(fl->fl_flags & FL_POSIX))
		return -ENOLCK;
	/* No mandatory locks */
	if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
		return -ENOLCK;

229
	dout("ceph_lock, fl_owner: %p", fl->fl_owner);
G
Greg Farnum 已提交
230 231

	/* set wait bit as appropriate, then make command as Ceph expects it*/
232
	if (IS_GETLK(cmd))
G
Greg Farnum 已提交
233
		op = CEPH_MDS_OP_GETFILELOCK;
234 235
	else if (IS_SETLKW(cmd))
		wait = 1;
G
Greg Farnum 已提交
236

237 238 239 240
	spin_lock(&ci->i_ceph_lock);
	if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
		err = -EIO;
	} else if (op == CEPH_MDS_OP_SETFILELOCK) {
241 242 243 244 245 246 247
		/*
		 * increasing i_filelock_ref closes race window between
		 * handling request reply and adding file_lock struct to
		 * inode. Otherwise, i_auth_cap may get trimmed in the
		 * window. Caller function will decrease the counter.
		 */
		fl->fl_ops = &ceph_fl_lock_ops;
248 249 250 251 252 253 254
		atomic_inc(&ci->i_filelock_ref);
	}
	spin_unlock(&ci->i_ceph_lock);
	if (err < 0) {
		if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK == fl->fl_type)
			posix_lock_file(file, fl, NULL);
		return err;
255 256
	}

G
Greg Farnum 已提交
257 258 259 260 261 262 263
	if (F_RDLCK == fl->fl_type)
		lock_cmd = CEPH_LOCK_SHARED;
	else if (F_WRLCK == fl->fl_type)
		lock_cmd = CEPH_LOCK_EXCL;
	else
		lock_cmd = CEPH_LOCK_UNLOCK;

264
	err = ceph_lock_message(CEPH_LOCK_FCNTL, op, inode, lock_cmd, wait, fl);
G
Greg Farnum 已提交
265
	if (!err) {
266
		if (op == CEPH_MDS_OP_SETFILELOCK) {
267 268
			dout("mds locked, locking locally");
			err = posix_lock_file(file, fl, NULL);
269
			if (err) {
S
Sage Weil 已提交
270 271 272
				/* undo! This should only happen if
				 * the kernel detects local
				 * deadlock. */
273
				ceph_lock_message(CEPH_LOCK_FCNTL, op, inode,
274
						  CEPH_LOCK_UNLOCK, 0, fl);
S
Sage Weil 已提交
275 276
				dout("got %d on posix_lock_file, undid lock",
				     err);
277
			}
G
Greg Farnum 已提交
278 279 280 281 282 283 284
		}
	}
	return err;
}

int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
{
285
	struct inode *inode = file_inode(file);
286 287
	struct ceph_inode_info *ci = ceph_inode(inode);
	int err = 0;
288
	u8 wait = 0;
289
	u8 lock_cmd;
G
Greg Farnum 已提交
290

Y
Yan, Zheng 已提交
291 292 293
	if (!(fl->fl_flags & FL_FLOCK))
		return -ENOLCK;
	/* No mandatory locks */
Y
Yan, Zheng 已提交
294 295
	if (fl->fl_type & LOCK_MAND)
		return -EOPNOTSUPP;
Y
Yan, Zheng 已提交
296

297
	dout("ceph_flock, fl_file: %p", fl->fl_file);
G
Greg Farnum 已提交
298

299 300 301 302 303 304 305 306 307 308 309 310 311 312
	spin_lock(&ci->i_ceph_lock);
	if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
		err = -EIO;
	} else {
		/* see comment in ceph_lock */
		fl->fl_ops = &ceph_fl_lock_ops;
		atomic_inc(&ci->i_filelock_ref);
	}
	spin_unlock(&ci->i_ceph_lock);
	if (err < 0) {
		if (F_UNLCK == fl->fl_type)
			locks_lock_file_wait(file, fl);
		return err;
	}
313

314 315 316 317
	if (IS_SETLKW(cmd))
		wait = 1;

	if (F_RDLCK == fl->fl_type)
G
Greg Farnum 已提交
318
		lock_cmd = CEPH_LOCK_SHARED;
319
	else if (F_WRLCK == fl->fl_type)
G
Greg Farnum 已提交
320 321 322 323 324
		lock_cmd = CEPH_LOCK_EXCL;
	else
		lock_cmd = CEPH_LOCK_UNLOCK;

	err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
325
				inode, lock_cmd, wait, fl);
G
Greg Farnum 已提交
326
	if (!err) {
327
		err = locks_lock_file_wait(file, fl);
G
Greg Farnum 已提交
328 329 330
		if (err) {
			ceph_lock_message(CEPH_LOCK_FLOCK,
					  CEPH_MDS_OP_SETFILELOCK,
331
					  inode, CEPH_LOCK_UNLOCK, 0, fl);
332
			dout("got %d on locks_lock_file_wait, undid lock", err);
G
Greg Farnum 已提交
333 334 335 336 337
		}
	}
	return err;
}

338 339 340
/*
 * Fills in the passed counter variables, so you can prepare pagelist metadata
 * before calling ceph_encode_locks.
G
Greg Farnum 已提交
341 342 343
 */
void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
{
344
	struct file_lock *lock;
345
	struct file_lock_context *ctx;
G
Greg Farnum 已提交
346 347 348 349

	*fcntl_count = 0;
	*flock_count = 0;

350 351
	ctx = inode->i_flctx;
	if (ctx) {
352 353 354 355 356 357
		spin_lock(&ctx->flc_lock);
		list_for_each_entry(lock, &ctx->flc_posix, fl_list)
			++(*fcntl_count);
		list_for_each_entry(lock, &ctx->flc_flock, fl_list)
			++(*flock_count);
		spin_unlock(&ctx->flc_lock);
G
Greg Farnum 已提交
358 359 360 361 362
	}
	dout("counted %d flock locks and %d fcntl locks",
	     *flock_count, *fcntl_count);
}

363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393
/*
 * Given a pointer to a lock, convert it to a ceph filelock
 */
static int lock_to_ceph_filelock(struct file_lock *lock,
				 struct ceph_filelock *cephlock)
{
	int err = 0;
	cephlock->start = cpu_to_le64(lock->fl_start);
	cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
	cephlock->client = cpu_to_le64(0);
	cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
	cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));

	switch (lock->fl_type) {
	case F_RDLCK:
		cephlock->type = CEPH_LOCK_SHARED;
		break;
	case F_WRLCK:
		cephlock->type = CEPH_LOCK_EXCL;
		break;
	case F_UNLCK:
		cephlock->type = CEPH_LOCK_UNLOCK;
		break;
	default:
		dout("Have unknown lock type %d", lock->fl_type);
		err = -EINVAL;
	}

	return err;
}

G
Greg Farnum 已提交
394
/**
395
 * Encode the flock and fcntl locks for the given inode into the ceph_filelock
396
 * array. Must be called with inode->i_lock already held.
397
 * If we encounter more of a specific lock type than expected, return -ENOSPC.
G
Greg Farnum 已提交
398
 */
399 400 401
int ceph_encode_locks_to_buffer(struct inode *inode,
				struct ceph_filelock *flocks,
				int num_fcntl_locks, int num_flock_locks)
G
Greg Farnum 已提交
402 403
{
	struct file_lock *lock;
404
	struct file_lock_context *ctx = inode->i_flctx;
G
Greg Farnum 已提交
405
	int err = 0;
406 407
	int seen_fcntl = 0;
	int seen_flock = 0;
408
	int l = 0;
G
Greg Farnum 已提交
409 410 411

	dout("encoding %d flock and %d fcntl locks", num_flock_locks,
	     num_fcntl_locks);
412

413 414 415
	if (!ctx)
		return 0;

416
	spin_lock(&ctx->flc_lock);
417
	list_for_each_entry(lock, &ctx->flc_posix, fl_list) {
418 419 420 421
		++seen_fcntl;
		if (seen_fcntl > num_fcntl_locks) {
			err = -ENOSPC;
			goto fail;
G
Greg Farnum 已提交
422
		}
423 424 425 426
		err = lock_to_ceph_filelock(lock, &flocks[l]);
		if (err)
			goto fail;
		++l;
G
Greg Farnum 已提交
427
	}
428 429 430 431 432
	list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
		++seen_flock;
		if (seen_flock > num_flock_locks) {
			err = -ENOSPC;
			goto fail;
G
Greg Farnum 已提交
433
		}
434 435 436 437
		err = lock_to_ceph_filelock(lock, &flocks[l]);
		if (err)
			goto fail;
		++l;
G
Greg Farnum 已提交
438 439
	}
fail:
440
	spin_unlock(&ctx->flc_lock);
G
Greg Farnum 已提交
441 442 443
	return err;
}

444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461
/**
 * Copy the encoded flock and fcntl locks into the pagelist.
 * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
 * sequential flock locks.
 * Returns zero on success.
 */
int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
			   struct ceph_pagelist *pagelist,
			   int num_fcntl_locks, int num_flock_locks)
{
	int err = 0;
	__le32 nlocks;

	nlocks = cpu_to_le32(num_fcntl_locks);
	err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
	if (err)
		goto out_fail;

462 463 464 465 466 467
	if (num_fcntl_locks > 0) {
		err = ceph_pagelist_append(pagelist, flocks,
					   num_fcntl_locks * sizeof(*flocks));
		if (err)
			goto out_fail;
	}
468 469 470 471 472 473

	nlocks = cpu_to_le32(num_flock_locks);
	err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
	if (err)
		goto out_fail;

474 475 476 477
	if (num_flock_locks > 0) {
		err = ceph_pagelist_append(pagelist, &flocks[num_fcntl_locks],
					   num_flock_locks * sizeof(*flocks));
	}
478 479 480
out_fail:
	return err;
}