fs/remap_range.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <linux/slab.h>
   3 #include <linux/stat.h>
   4 #include <linux/sched/xacct.h>
   5 #include <linux/fcntl.h>
   6 #include <linux/file.h>
   7 #include <linux/uio.h>
   8 #include <linux/fsnotify.h>
   9 #include <linux/security.h>
  10 #include <linux/export.h>
  11 #include <linux/syscalls.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/splice.h>
  14 #include <linux/compat.h>
  15 #include <linux/mount.h>
  16 #include <linux/fs.h>
  17 #include "internal.h"
  18
  19 #include <linux/uaccess.h>
  20 #include <asm/unistd.h>
  21
  22 /*
  23  * Performs necessary checks before doing a clone.
  24  *
  25  * Can adjust amount of bytes to clone via @req_count argument.
  26  * Returns appropriate error code that caller should return or
  27  * zero in case the clone should be allowed.
  28  */
  29 static int generic_remap_checks(struct file *file_in, loff_t pos_in,
  30                                 struct file *file_out, loff_t pos_out,
  31                                 loff_t *req_count, unsigned int remap_flags)
  32 {
  33         struct inode *inode_in = file_in->f_mapping->host;
  34         struct inode *inode_out = file_out->f_mapping->host;
  35         uint64_t count = *req_count;
  36         uint64_t bcount;
  37         loff_t size_in, size_out;
  38         loff_t bs = inode_out->i_sb->s_blocksize;
  39         int ret;
  40
  41         /* The start of both ranges must be aligned to an fs block. */
  42         if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
  43                 return -EINVAL;
  44
  45         /* Ensure offsets don't wrap. */
  46         if (pos_in + count < pos_in || pos_out + count < pos_out)
  47                 return -EINVAL;
  48
  49         size_in = i_size_read(inode_in);
  50         size_out = i_size_read(inode_out);
  51
  52         /* Dedupe requires both ranges to be within EOF. */
  53         if ((remap_flags & REMAP_FILE_DEDUP) &&
  54             (pos_in >= size_in || pos_in + count > size_in ||
  55              pos_out >= size_out || pos_out + count > size_out))
  56                 return -EINVAL;
  57
  58         /* Ensure the infile range is within the infile. */
  59         if (pos_in >= size_in)
  60                 return -EINVAL;
  61         count = min(count, size_in - (uint64_t)pos_in);
  62
  63         ret = generic_write_check_limits(file_out, pos_out, &count);
  64         if (ret)
  65                 return ret;
  66
  67         /*
  68          * If the user wanted us to link to the infile's EOF, round up to the
  69          * next block boundary for this check.
  70          *
  71          * Otherwise, make sure the count is also block-aligned, having
  72          * already confirmed the starting offsets' block alignment.
  73          */
  74         if (pos_in + count == size_in &&
  75             (!(remap_flags & REMAP_FILE_DEDUP) || pos_out + count == size_out)) {
  76                 bcount = ALIGN(size_in, bs) - pos_in;
  77         } else {
  78                 if (!IS_ALIGNED(count, bs))
  79                         count = ALIGN_DOWN(count, bs);
  80                 bcount = count;
  81         }
  82
  83         /* Don't allow overlapped cloning within the same file. */
  84         if (inode_in == inode_out &&
  85             pos_out + bcount > pos_in &&
  86             pos_out < pos_in + bcount)
  87                 return -EINVAL;
  88
  89         /*
  90          * We shortened the request but the caller can't deal with that, so
  91          * bounce the request back to userspace.
  92          */
  93         if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
  94                 return -EINVAL;
  95
  96         *req_count = count;
  97         return 0;
  98 }
  99
 100 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
 101                              bool write)
 102 {
 103         if (unlikely(pos < 0 || len < 0))
 104                 return -EINVAL;
 105
 106         if (unlikely((loff_t) (pos + len) < 0))
 107                 return -EINVAL;
 108
 109         return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
 110 }
 111
 112 /*
 113  * Ensure that we don't remap a partial EOF block in the middle of something
 114  * else.  Assume that the offsets have already been checked for block
 115  * alignment.
 116  *
 117  * For clone we only link a partial EOF block above or at the destination file's
 118  * EOF.  For deduplication we accept a partial EOF block only if it ends at the
 119  * destination file's EOF (can not link it into the middle of a file).
 120  *
 121  * Shorten the request if possible.
 122  */
 123 static int generic_remap_check_len(struct inode *inode_in,
 124                                    struct inode *inode_out,
 125                                    loff_t pos_out,
 126                                    loff_t *len,
 127                                    unsigned int remap_flags)
 128 {
 129         u64 blkmask = i_blocksize(inode_in) - 1;
 130         loff_t new_len = *len;
 131
 132         if ((*len & blkmask) == 0)
 133                 return 0;
 134
 135         if (pos_out + *len < i_size_read(inode_out))
 136                 new_len &= ~blkmask;
 137
 138         if (new_len == *len)
 139                 return 0;
 140
 141         if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
 142                 *len = new_len;
 143                 return 0;
 144         }
 145
 146         return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
 147 }
 148
 149 /* Read a page's worth of file data into the page cache. */
 150 static struct folio *vfs_dedupe_get_folio(struct file *file, loff_t pos)
 151 {
 152         struct folio *folio;
 153
 154         folio = read_mapping_folio(file->f_mapping, pos >> PAGE_SHIFT, file);
 155         if (IS_ERR(folio))
 156                 return folio;
 157         if (!folio_test_uptodate(folio)) {
 158                 folio_put(folio);
 159                 return ERR_PTR(-EIO);
 160         }
 161         return folio;
 162 }
 163
 164 /*
 165  * Lock two folios, ensuring that we lock in offset order if the folios
 166  * are from the same file.
 167  */
 168 static void vfs_lock_two_folios(struct folio *folio1, struct folio *folio2)
 169 {
 170         /* Always lock in order of increasing index. */
 171         if (folio1->index > folio2->index)
 172                 swap(folio1, folio2);
 173
 174         folio_lock(folio1);
 175         if (folio1 != folio2)
 176                 folio_lock(folio2);
 177 }
 178
 179 /* Unlock two folios, being careful not to unlock the same folio twice. */
 180 static void vfs_unlock_two_folios(struct folio *folio1, struct folio *folio2)
 181 {
 182         folio_unlock(folio1);
 183         if (folio1 != folio2)
 184                 folio_unlock(folio2);
 185 }
 186
 187 /*
 188  * Compare extents of two files to see if they are the same.
 189  * Caller must have locked both inodes to prevent write races.
 190  */
 191 static int vfs_dedupe_file_range_compare(struct file *src, loff_t srcoff,
 192                                          struct file *dest, loff_t dstoff,
 193                                          loff_t len, bool *is_same)
 194 {
 195         bool same = true;
 196         int error = -EINVAL;
 197
 198         while (len) {
 199                 struct folio *src_folio, *dst_folio;
 200                 void *src_addr, *dst_addr;
 201                 loff_t cmp_len = min(PAGE_SIZE - offset_in_page(srcoff),
 202                                      PAGE_SIZE - offset_in_page(dstoff));
 203
 204                 cmp_len = min(cmp_len, len);
 205                 if (cmp_len <= 0)
 206                         goto out_error;
 207
 208                 src_folio = vfs_dedupe_get_folio(src, srcoff);
 209                 if (IS_ERR(src_folio)) {
 210                         error = PTR_ERR(src_folio);
 211                         goto out_error;
 212                 }
 213                 dst_folio = vfs_dedupe_get_folio(dest, dstoff);
 214                 if (IS_ERR(dst_folio)) {
 215                         error = PTR_ERR(dst_folio);
 216                         folio_put(src_folio);
 217                         goto out_error;
 218                 }
 219
 220                 vfs_lock_two_folios(src_folio, dst_folio);
 221
 222                 /*
 223                  * Now that we've locked both folios, make sure they're still
 224                  * mapped to the file data we're interested in.  If not,
 225                  * someone is invalidating pages on us and we lose.
 226                  */
 227                 if (!folio_test_uptodate(src_folio) || !folio_test_uptodate(dst_folio) ||
 228                     src_folio->mapping != src->f_mapping ||
 229                     dst_folio->mapping != dest->f_mapping) {
 230                         same = false;
 231                         goto unlock;
 232                 }
 233
 234                 src_addr = kmap_local_folio(src_folio,
 235                                         offset_in_folio(src_folio, srcoff));
 236                 dst_addr = kmap_local_folio(dst_folio,
 237                                         offset_in_folio(dst_folio, dstoff));
 238
 239                 flush_dcache_folio(src_folio);
 240                 flush_dcache_folio(dst_folio);
 241
 242                 if (memcmp(src_addr, dst_addr, cmp_len))
 243                         same = false;
 244
 245                 kunmap_local(dst_addr);
 246                 kunmap_local(src_addr);
 247 unlock:
 248                 vfs_unlock_two_folios(src_folio, dst_folio);
 249                 folio_put(dst_folio);
 250                 folio_put(src_folio);
 251
 252                 if (!same)
 253                         break;
 254
 255                 srcoff += cmp_len;
 256                 dstoff += cmp_len;
 257                 len -= cmp_len;
 258         }
 259
 260         *is_same = same;
 261         return 0;
 262
 263 out_error:
 264         return error;
 265 }
 266
 267 /*
 268  * Check that the two inodes are eligible for cloning, the ranges make
 269  * sense, and then flush all dirty data.  Caller must ensure that the
 270  * inodes have been locked against any other modifications.
 271  *
 272  * If there's an error, then the usual negative error code is returned.
 273  * Otherwise returns 0 with *len set to the request length.
 274  */
 275 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
 276                                   struct file *file_out, loff_t pos_out,
 277                                   loff_t *len, unsigned int remap_flags)
 278 {
 279         struct inode *inode_in = file_inode(file_in);
 280         struct inode *inode_out = file_inode(file_out);
 281         bool same_inode = (inode_in == inode_out);
 282         int ret;
 283
 284         /* Don't touch certain kinds of inodes */
 285         if (IS_IMMUTABLE(inode_out))
 286                 return -EPERM;
 287
 288         if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
 289                 return -ETXTBSY;
 290
 291         /* Don't reflink dirs, pipes, sockets... */
 292         if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
 293                 return -EISDIR;
 294         if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
 295                 return -EINVAL;
 296
 297         /* Zero length dedupe exits immediately; reflink goes to EOF. */
 298         if (*len == 0) {
 299                 loff_t isize = i_size_read(inode_in);
 300
 301                 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
 302                         return 0;
 303                 if (pos_in > isize)
 304                         return -EINVAL;
 305                 *len = isize - pos_in;
 306                 if (*len == 0)
 307                         return 0;
 308         }
 309
 310         /* Check that we don't violate system file offset limits. */
 311         ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
 312                         remap_flags);
 313         if (ret)
 314                 return ret;
 315
 316         /* Wait for the completion of any pending IOs on both files */
 317         inode_dio_wait(inode_in);
 318         if (!same_inode)
 319                 inode_dio_wait(inode_out);
 320
 321         ret = filemap_write_and_wait_range(inode_in->i_mapping,
 322                         pos_in, pos_in + *len - 1);
 323         if (ret)
 324                 return ret;
 325
 326         ret = filemap_write_and_wait_range(inode_out->i_mapping,
 327                         pos_out, pos_out + *len - 1);
 328         if (ret)
 329                 return ret;
 330
 331         /*
 332          * Check that the extents are the same.
 333          */
 334         if (remap_flags & REMAP_FILE_DEDUP) {
 335                 bool            is_same = false;
 336
 337                 ret = vfs_dedupe_file_range_compare(file_in, pos_in,
 338                                 file_out, pos_out, *len, &is_same);
 339                 if (ret)
 340                         return ret;
 341                 if (!is_same)
 342                         return -EBADE;
 343         }
 344
 345         ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
 346                         remap_flags);
 347         if (ret)
 348                 return ret;
 349
 350         /* If can't alter the file contents, we're done. */
 351         if (!(remap_flags & REMAP_FILE_DEDUP))
 352                 ret = file_modified(file_out);
 353
 354         return ret;
 355 }
 356 EXPORT_SYMBOL(generic_remap_file_range_prep);
 357
 358 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
 359                            struct file *file_out, loff_t pos_out,
 360                            loff_t len, unsigned int remap_flags)
 361 {
 362         loff_t ret;
 363
 364         WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
 365
 366         if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
 367                 return -EXDEV;
 368
 369         ret = generic_file_rw_checks(file_in, file_out);
 370         if (ret < 0)
 371                 return ret;
 372
 373         if (!file_in->f_op->remap_file_range)
 374                 return -EOPNOTSUPP;
 375
 376         ret = remap_verify_area(file_in, pos_in, len, false);
 377         if (ret)
 378                 return ret;
 379
 380         ret = remap_verify_area(file_out, pos_out, len, true);
 381         if (ret)
 382                 return ret;
 383
 384         ret = file_in->f_op->remap_file_range(file_in, pos_in,
 385                         file_out, pos_out, len, remap_flags);
 386         if (ret < 0)
 387                 return ret;
 388
 389         fsnotify_access(file_in);
 390         fsnotify_modify(file_out);
 391         return ret;
 392 }
 393 EXPORT_SYMBOL(do_clone_file_range);
 394
 395 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
 396                             struct file *file_out, loff_t pos_out,
 397                             loff_t len, unsigned int remap_flags)
 398 {
 399         loff_t ret;
 400
 401         file_start_write(file_out);
 402         ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
 403                                   remap_flags);
 404         file_end_write(file_out);
 405
 406         return ret;
 407 }
 408 EXPORT_SYMBOL(vfs_clone_file_range);
 409
 410 /* Check whether we are allowed to dedupe the destination file */
 411 static bool allow_file_dedupe(struct file *file)
 412 {
 413         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
 414         struct inode *inode = file_inode(file);
 415
 416         if (capable(CAP_SYS_ADMIN))
 417                 return true;
 418         if (file->f_mode & FMODE_WRITE)
 419                 return true;
 420         if (uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
 421                 return true;
 422         if (!inode_permission(mnt_userns, inode, MAY_WRITE))
 423                 return true;
 424         return false;
 425 }
 426
 427 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
 428                                  struct file *dst_file, loff_t dst_pos,
 429                                  loff_t len, unsigned int remap_flags)
 430 {
 431         loff_t ret;
 432
 433         WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
 434                                      REMAP_FILE_CAN_SHORTEN));
 435
 436         ret = mnt_want_write_file(dst_file);
 437         if (ret)
 438                 return ret;
 439
 440         /*
 441          * This is redundant if called from vfs_dedupe_file_range(), but other
 442          * callers need it and it's not performance sesitive...
 443          */
 444         ret = remap_verify_area(src_file, src_pos, len, false);
 445         if (ret)
 446                 goto out_drop_write;
 447
 448         ret = remap_verify_area(dst_file, dst_pos, len, true);
 449         if (ret)
 450                 goto out_drop_write;
 451
 452         ret = -EPERM;
 453         if (!allow_file_dedupe(dst_file))
 454                 goto out_drop_write;
 455
 456         ret = -EXDEV;
 457         if (file_inode(src_file)->i_sb != file_inode(dst_file)->i_sb)
 458                 goto out_drop_write;
 459
 460         ret = -EISDIR;
 461         if (S_ISDIR(file_inode(dst_file)->i_mode))
 462                 goto out_drop_write;
 463
 464         ret = -EINVAL;
 465         if (!dst_file->f_op->remap_file_range)
 466                 goto out_drop_write;
 467
 468         if (len == 0) {
 469                 ret = 0;
 470                 goto out_drop_write;
 471         }
 472
 473         ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
 474                         dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
 475 out_drop_write:
 476         mnt_drop_write_file(dst_file);
 477
 478         return ret;
 479 }
 480 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
 481
 482 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
 483 {
 484         struct file_dedupe_range_info *info;
 485         struct inode *src = file_inode(file);
 486         u64 off;
 487         u64 len;
 488         int i;
 489         int ret;
 490         u16 count = same->dest_count;
 491         loff_t deduped;
 492
 493         if (!(file->f_mode & FMODE_READ))
 494                 return -EINVAL;
 495
 496         if (same->reserved1 || same->reserved2)
 497                 return -EINVAL;
 498
 499         off = same->src_offset;
 500         len = same->src_length;
 501
 502         if (S_ISDIR(src->i_mode))
 503                 return -EISDIR;
 504
 505         if (!S_ISREG(src->i_mode))
 506                 return -EINVAL;
 507
 508         if (!file->f_op->remap_file_range)
 509                 return -EOPNOTSUPP;
 510
 511         ret = remap_verify_area(file, off, len, false);
 512         if (ret < 0)
 513                 return ret;
 514         ret = 0;
 515
 516         if (off + len > i_size_read(src))
 517                 return -EINVAL;
 518
 519         /* Arbitrary 1G limit on a single dedupe request, can be raised. */
 520         len = min_t(u64, len, 1 << 30);
 521
 522         /* pre-format output fields to sane values */
 523         for (i = 0; i < count; i++) {
 524                 same->info[i].bytes_deduped = 0ULL;
 525                 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
 526         }
 527
 528         for (i = 0, info = same->info; i < count; i++, info++) {
 529                 struct fd dst_fd = fdget(info->dest_fd);
 530                 struct file *dst_file = dst_fd.file;
 531
 532                 if (!dst_file) {
 533                         info->status = -EBADF;
 534                         goto next_loop;
 535                 }
 536
 537                 if (info->reserved) {
 538                         info->status = -EINVAL;
 539                         goto next_fdput;
 540                 }
 541
 542                 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
 543                                                     info->dest_offset, len,
 544                                                     REMAP_FILE_CAN_SHORTEN);
 545                 if (deduped == -EBADE)
 546                         info->status = FILE_DEDUPE_RANGE_DIFFERS;
 547                 else if (deduped < 0)
 548                         info->status = deduped;
 549                 else
 550                         info->bytes_deduped = len;
 551
 552 next_fdput:
 553                 fdput(dst_fd);
 554 next_loop:
 555                 if (fatal_signal_pending(current))
 556                         break;
 557         }
 558         return ret;
 559 }
 560 EXPORT_SYMBOL(vfs_dedupe_file_range);