fs/ceph/locks.c

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