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