1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Copyright (C) 2012-2013 Samsung Electronics Co., Ltd. 4 */ 5 6 #include <linux/slab.h> 7 #include <linux/compat.h> 8 #include <linux/cred.h> 9 #include <linux/buffer_head.h> 10 #include <linux/blkdev.h> 11 #include <linux/fsnotify.h> 12 #include <linux/security.h> 13 #include <linux/msdos_fs.h> 14 #include <linux/writeback.h> 15 #include <linux/filelock.h> 16 #include <linux/falloc.h> 17 #include <linux/fileattr.h> 18 #include <linux/iomap.h> 19 #include <linux/pagemap.h> 20 #include <linux/rmap.h> 21 22 #include "exfat_raw.h" 23 #include "exfat_fs.h" 24 #include "iomap.h" 25 26 /* 27 * Unlike pagecache_isize_extended(), this function updates i_size while 28 * holding the straddling folio lock, and does not skip the straddling 29 * folio when the filesystem block size is >= PAGE_SIZE or when 'from' 30 * and 'to' fall within the same filesystem block. 31 * 32 * This ensures that the folio is marked RO and the post-EOF range is 33 * zeroed before readers or writeback can observe the updated i_size. 34 */ 35 static void exfat_pagecache_isize_extended(struct inode *inode, loff_t from, 36 loff_t to) 37 { 38 struct folio *folio; 39 40 if (!(from & (PAGE_SIZE - 1))) { 41 i_size_write(inode, to); 42 return; 43 } 44 45 folio = filemap_lock_folio(inode->i_mapping, from >> PAGE_SHIFT); 46 47 /* Folio not cached? Nothing to do */ 48 if (IS_ERR(folio)) { 49 i_size_write(inode, to); 50 return; 51 } 52 53 /* 54 * See folio_clear_dirty_for_io() for details why folio_mark_dirty() 55 * is needed. 56 */ 57 if (folio_mkclean(folio)) 58 folio_mark_dirty(folio); 59 60 /* 61 * Zero the post-EOF range before publishing the new size. Writeback 62 * normally does this, but the extension would expose the tail to 63 * readers before writeback gets a chance to zero it. 64 */ 65 if (folio_test_dirty(folio)) { 66 loff_t offset, end; 67 68 offset = from - folio_pos(folio); 69 end = min_t(loff_t, to - folio_pos(folio), 70 folio_size(folio)); 71 folio_zero_segment(folio, offset, end); 72 } 73 74 /* 75 * Publish the new size after zeroing the tail. Uptodate folios can 76 * be read without taking the folio lock. 77 */ 78 i_size_write(inode, to); 79 80 folio_unlock(folio); 81 folio_put(folio); 82 } 83 84 static int exfat_cont_expand(struct inode *inode, loff_t size) 85 { 86 int ret; 87 unsigned int num_clusters, new_num_clusters, last_clu; 88 struct exfat_inode_info *ei = EXFAT_I(inode); 89 struct super_block *sb = inode->i_sb; 90 struct exfat_sb_info *sbi = EXFAT_SB(sb); 91 struct exfat_chain clu; 92 loff_t oldsize = i_size_read(inode); 93 94 ret = inode_newsize_ok(inode, size); 95 if (ret) 96 return ret; 97 98 num_clusters = exfat_bytes_to_cluster(sbi, exfat_ondisk_size(inode)); 99 /* integer overflow is already checked in inode_newsize_ok(). */ 100 new_num_clusters = exfat_bytes_to_cluster_round_up(sbi, size); 101 102 if (new_num_clusters == num_clusters) 103 goto out; 104 105 if (num_clusters) { 106 exfat_chain_set(&clu, ei->start_clu, num_clusters, ei->flags); 107 ret = exfat_find_last_cluster(sb, &clu, &last_clu); 108 if (ret) 109 return ret; 110 111 clu.dir = last_clu + 1; 112 } else { 113 last_clu = EXFAT_EOF_CLUSTER; 114 clu.dir = EXFAT_EOF_CLUSTER; 115 } 116 117 clu.size = 0; 118 clu.flags = ei->flags; 119 120 ret = exfat_alloc_cluster(inode, new_num_clusters - num_clusters, 121 &clu, inode_needs_sync(inode), false); 122 if (ret) 123 return ret; 124 125 /* Append new clusters to chain */ 126 if (num_clusters) { 127 if (clu.flags != ei->flags) 128 if (exfat_chain_cont_cluster(sb, ei->start_clu, num_clusters)) 129 goto free_clu; 130 131 if (clu.flags == ALLOC_FAT_CHAIN) 132 if (exfat_ent_set(sb, last_clu, clu.dir)) 133 goto free_clu; 134 } else 135 ei->start_clu = clu.dir; 136 137 ei->flags = clu.flags; 138 139 out: 140 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 141 /* 142 * When extending file size, call exfat_pagecache_isize_extended() 143 * to update i_size and ensure the straddling folio is properly 144 * marked RO so page_mkwrite() is called and post-EOF area is zeroed. 145 */ 146 exfat_pagecache_isize_extended(inode, oldsize, size); 147 148 inode->i_blocks = round_up(size, sbi->cluster_size) >> 9; 149 mark_inode_dirty(inode); 150 151 if (IS_SYNC(inode)) 152 return write_inode_now(inode, 1); 153 154 return 0; 155 156 free_clu: 157 exfat_free_cluster(inode, &clu); 158 return -EIO; 159 } 160 161 /* 162 * Preallocate space for a file. This implements exfat's fallocate file 163 * operation, which gets called from sys_fallocate system call. User space 164 * requests len bytes at offset. In contrary to fat, we only support 165 * FALLOC_FL_ALLOCATE_RANGE because by leaving the valid data length(VDL) 166 * field, it is unnecessary to zero out the newly allocated clusters. 167 */ 168 static long exfat_fallocate(struct file *file, int mode, 169 loff_t offset, loff_t len) 170 { 171 struct inode *inode = file->f_mapping->host; 172 loff_t newsize = offset + len; 173 int err = 0; 174 175 /* No support for other modes */ 176 if (mode != FALLOC_FL_ALLOCATE_RANGE) 177 return -EOPNOTSUPP; 178 179 /* No support for dir */ 180 if (!S_ISREG(inode->i_mode)) 181 return -EOPNOTSUPP; 182 183 if (unlikely(exfat_forced_shutdown(inode->i_sb))) 184 return -EIO; 185 186 inode_lock(inode); 187 188 if (newsize <= i_size_read(inode)) 189 goto error; 190 191 /* This is just an expanding truncate */ 192 err = exfat_cont_expand(inode, newsize); 193 194 error: 195 inode_unlock(inode); 196 197 return err; 198 } 199 200 static bool exfat_allow_set_time(const struct mnt_idmap *idmap, 201 struct exfat_sb_info *sbi, struct inode *inode) 202 { 203 mode_t allow_utime = sbi->options.allow_utime; 204 205 if (!vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), 206 current_fsuid())) { 207 if (vfsgid_in_group_p(i_gid_into_vfsgid(idmap, inode))) 208 allow_utime >>= 3; 209 if (allow_utime & MAY_WRITE) 210 return true; 211 } 212 213 /* use a default check */ 214 return false; 215 } 216 217 static int exfat_sanitize_mode(const struct exfat_sb_info *sbi, 218 struct inode *inode, umode_t *mode_ptr) 219 { 220 mode_t i_mode, mask, perm; 221 222 i_mode = inode->i_mode; 223 224 mask = (S_ISREG(i_mode) || S_ISLNK(i_mode)) ? 225 sbi->options.fs_fmask : sbi->options.fs_dmask; 226 perm = *mode_ptr & ~(S_IFMT | mask); 227 228 /* Of the r and x bits, all (subject to umask) must be present.*/ 229 if ((perm & 0555) != (i_mode & 0555)) 230 return -EPERM; 231 232 if (exfat_mode_can_hold_ro(inode)) { 233 /* 234 * Of the w bits, either all (subject to umask) or none must 235 * be present. 236 */ 237 if ((perm & 0222) && ((perm & 0222) != (0222 & ~mask))) 238 return -EPERM; 239 } else { 240 /* 241 * If exfat_mode_can_hold_ro(inode) is false, can't change 242 * w bits. 243 */ 244 if ((perm & 0222) != (0222 & ~mask)) 245 return -EPERM; 246 } 247 248 *mode_ptr &= S_IFMT | perm; 249 250 return 0; 251 } 252 253 /* resize the file length */ 254 int __exfat_truncate(struct inode *inode) 255 { 256 unsigned int num_clusters_new, num_clusters_phys; 257 unsigned int last_clu = EXFAT_FREE_CLUSTER; 258 struct exfat_chain clu; 259 struct super_block *sb = inode->i_sb; 260 struct exfat_sb_info *sbi = EXFAT_SB(sb); 261 struct exfat_inode_info *ei = EXFAT_I(inode); 262 263 /* check if the given file ID is opened */ 264 if (ei->type != TYPE_FILE && ei->type != TYPE_DIR) 265 return -EPERM; 266 267 exfat_set_volume_dirty(sb); 268 269 num_clusters_new = exfat_bytes_to_cluster_round_up(sbi, i_size_read(inode)); 270 num_clusters_phys = exfat_bytes_to_cluster(sbi, exfat_ondisk_size(inode)); 271 272 exfat_chain_set(&clu, ei->start_clu, num_clusters_phys, ei->flags); 273 274 if (i_size_read(inode) > 0) { 275 /* 276 * Truncate FAT chain num_clusters after the first cluster 277 * num_clusters = min(new, phys); 278 */ 279 unsigned int num_clusters = 280 min(num_clusters_new, num_clusters_phys); 281 282 /* 283 * Follow FAT chain 284 * (defensive coding - works fine even with corrupted FAT table 285 */ 286 if (clu.flags == ALLOC_NO_FAT_CHAIN) { 287 clu.dir += num_clusters; 288 clu.size -= num_clusters; 289 } else { 290 while (num_clusters > 0) { 291 last_clu = clu.dir; 292 if (exfat_get_next_cluster(sb, &(clu.dir))) 293 return -EIO; 294 295 num_clusters--; 296 clu.size--; 297 } 298 } 299 } else { 300 ei->flags = ALLOC_NO_FAT_CHAIN; 301 ei->start_clu = EXFAT_EOF_CLUSTER; 302 } 303 304 if (i_size_read(inode) < exfat_get_valid_size(ei)) { 305 exfat_set_valid_size(ei, i_size_read(inode)); 306 exfat_set_zeroed_size(ei, i_size_read(inode)); 307 } 308 309 if (ei->type == TYPE_FILE) 310 ei->attr |= EXFAT_ATTR_ARCHIVE; 311 312 /* 313 * update the directory entry 314 * 315 * If the directory entry is updated by mark_inode_dirty(), the 316 * directory entry will be written after a writeback cycle of 317 * updating the bitmap/FAT, which may result in clusters being 318 * freed but referenced by the directory entry in the event of a 319 * sudden power failure. 320 * __exfat_write_inode() is called for directory entry, bitmap 321 * and FAT to be written in a same writeback. 322 */ 323 if (__exfat_write_inode(inode, inode_needs_sync(inode))) 324 return -EIO; 325 326 /* cut off from the FAT chain */ 327 if (ei->flags == ALLOC_FAT_CHAIN && last_clu != EXFAT_FREE_CLUSTER && 328 last_clu != EXFAT_EOF_CLUSTER) { 329 if (exfat_ent_set(sb, last_clu, EXFAT_EOF_CLUSTER)) 330 return -EIO; 331 } 332 333 /* invalidate cache and free the clusters */ 334 /* clear exfat cache */ 335 exfat_cache_inval_inode(inode); 336 337 /* hint information */ 338 ei->hint_bmap.off = EXFAT_EOF_CLUSTER; 339 ei->hint_bmap.clu = EXFAT_EOF_CLUSTER; 340 341 /* hint_stat will be used if this is directory. */ 342 ei->hint_stat.eidx = 0; 343 ei->hint_stat.clu = ei->start_clu; 344 ei->hint_femp.eidx = EXFAT_HINT_NONE; 345 346 /* free the clusters */ 347 if (exfat_free_cluster(inode, &clu)) 348 return -EIO; 349 350 return 0; 351 } 352 353 static void exfat_truncate(struct inode *inode) 354 { 355 struct super_block *sb = inode->i_sb; 356 struct exfat_sb_info *sbi = EXFAT_SB(sb); 357 struct exfat_inode_info *ei = EXFAT_I(inode); 358 int err; 359 360 mutex_lock(&sbi->s_lock); 361 if (ei->start_clu == 0) { 362 /* 363 * Empty start_clu != ~0 (not allocated) 364 */ 365 exfat_fs_error(sb, "tried to truncate zeroed cluster."); 366 goto write_size; 367 } 368 369 err = __exfat_truncate(inode); 370 if (err) 371 goto write_size; 372 373 inode->i_blocks = round_up(i_size_read(inode), sbi->cluster_size) >> 9; 374 write_size: 375 mutex_unlock(&sbi->s_lock); 376 } 377 378 int exfat_getattr(const struct mnt_idmap *idmap, const struct path *path, 379 struct kstat *stat, unsigned int request_mask, 380 unsigned int query_flags) 381 { 382 struct inode *inode = d_backing_inode(path->dentry); 383 struct exfat_inode_info *ei = EXFAT_I(inode); 384 385 generic_fillattr(idmap, request_mask, inode, stat); 386 exfat_truncate_atime(&stat->atime); 387 stat->result_mask |= STATX_BTIME; 388 stat->btime.tv_sec = ei->i_crtime.tv_sec; 389 stat->btime.tv_nsec = ei->i_crtime.tv_nsec; 390 stat->blksize = EXFAT_SB(inode->i_sb)->cluster_size; 391 return 0; 392 } 393 394 int exfat_fileattr_get(struct dentry *dentry, struct file_kattr *fa) 395 { 396 /* 397 * exFAT compares filenames through an upcase table, so lookup 398 * is always case-insensitive. Long names are stored in UTF-16 399 * with case intact; CASENONPRESERVING stays clear. 400 */ 401 fa->fsx_xflags |= FS_XFLAG_CASEFOLD; 402 fa->flags |= FS_CASEFOLD_FL; 403 return 0; 404 } 405 406 int exfat_setattr(const struct mnt_idmap *idmap, struct dentry *dentry, 407 struct iattr *attr) 408 { 409 struct exfat_sb_info *sbi = EXFAT_SB(dentry->d_sb); 410 struct inode *inode = dentry->d_inode; 411 unsigned int ia_valid; 412 int error; 413 414 if (unlikely(exfat_forced_shutdown(inode->i_sb))) 415 return -EIO; 416 417 if ((attr->ia_valid & ATTR_SIZE) && 418 attr->ia_size > i_size_read(inode)) { 419 error = exfat_cont_expand(inode, attr->ia_size); 420 if (error || attr->ia_valid == ATTR_SIZE) 421 return error; 422 attr->ia_valid &= ~ATTR_SIZE; 423 } 424 425 /* Check for setting the inode time. */ 426 ia_valid = attr->ia_valid; 427 if ((ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)) && 428 exfat_allow_set_time(idmap, sbi, inode)) { 429 attr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET | 430 ATTR_TIMES_SET); 431 } 432 433 error = setattr_prepare(idmap, dentry, attr); 434 attr->ia_valid = ia_valid; 435 if (error) 436 goto out; 437 438 if (((attr->ia_valid & ATTR_UID) && 439 (!uid_eq(from_vfsuid(idmap, i_user_ns(inode), attr->ia_vfsuid), 440 sbi->options.fs_uid))) || 441 ((attr->ia_valid & ATTR_GID) && 442 (!gid_eq(from_vfsgid(idmap, i_user_ns(inode), attr->ia_vfsgid), 443 sbi->options.fs_gid))) || 444 ((attr->ia_valid & ATTR_MODE) && 445 (attr->ia_mode & ~(S_IFREG | S_IFLNK | S_IFDIR | 0777)))) { 446 error = -EPERM; 447 goto out; 448 } 449 450 /* 451 * We don't return -EPERM here. Yes, strange, but this is too 452 * old behavior. 453 */ 454 if (attr->ia_valid & ATTR_MODE) { 455 if (exfat_sanitize_mode(sbi, inode, &attr->ia_mode) < 0) 456 attr->ia_valid &= ~ATTR_MODE; 457 } 458 459 if (attr->ia_valid & ATTR_SIZE) 460 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 461 462 setattr_copy(idmap, inode, attr); 463 exfat_truncate_inode_atime(inode); 464 465 if (attr->ia_valid & ATTR_SIZE) { 466 /* 467 * Wait for any in-flight DIO to finish before truncating to 468 * prevent a concurrent DIO from writing to clusters that are 469 * about to be freed. 470 */ 471 inode_dio_wait(inode); 472 filemap_invalidate_lock(inode->i_mapping); 473 truncate_setsize(inode, attr->ia_size); 474 475 /* 476 * __exfat_write_inode() is called from exfat_truncate(), inode 477 * is already written by it, so mark_inode_dirty() is unneeded. 478 */ 479 exfat_truncate(inode); 480 filemap_invalidate_unlock(inode->i_mapping); 481 } else 482 mark_inode_dirty(inode); 483 484 out: 485 return error; 486 } 487 488 /* 489 * modified ioctls from fat/file.c by Welmer Almesberger 490 */ 491 static int exfat_ioctl_get_attributes(struct inode *inode, u32 __user *user_attr) 492 { 493 u32 attr; 494 495 inode_lock_shared(inode); 496 attr = exfat_make_attr(inode); 497 inode_unlock_shared(inode); 498 499 return put_user(attr, user_attr); 500 } 501 502 static int exfat_ioctl_set_attributes(struct file *file, u32 __user *user_attr) 503 { 504 struct inode *inode = file_inode(file); 505 struct exfat_sb_info *sbi = EXFAT_SB(inode->i_sb); 506 int is_dir = S_ISDIR(inode->i_mode); 507 u32 attr, oldattr; 508 struct iattr ia; 509 int err; 510 511 err = get_user(attr, user_attr); 512 if (err) 513 goto out; 514 515 err = mnt_want_write_file(file); 516 if (err) 517 goto out; 518 inode_lock(inode); 519 520 oldattr = exfat_make_attr(inode); 521 522 /* 523 * Mask attributes so we don't set reserved fields. 524 */ 525 attr &= (EXFAT_ATTR_READONLY | EXFAT_ATTR_HIDDEN | EXFAT_ATTR_SYSTEM | 526 EXFAT_ATTR_ARCHIVE); 527 attr |= (is_dir ? EXFAT_ATTR_SUBDIR : 0); 528 529 /* Equivalent to a chmod() */ 530 ia.ia_valid = ATTR_MODE | ATTR_CTIME; 531 ia.ia_ctime = current_time(inode); 532 if (is_dir) 533 ia.ia_mode = exfat_make_mode(sbi, attr, 0777); 534 else 535 ia.ia_mode = exfat_make_mode(sbi, attr, 0666 | (inode->i_mode & 0111)); 536 537 /* The root directory has no attributes */ 538 if (inode->i_ino == EXFAT_ROOT_INO && attr != EXFAT_ATTR_SUBDIR) { 539 err = -EINVAL; 540 goto out_unlock_inode; 541 } 542 543 if (((attr | oldattr) & EXFAT_ATTR_SYSTEM) && 544 !capable(CAP_LINUX_IMMUTABLE)) { 545 err = -EPERM; 546 goto out_unlock_inode; 547 } 548 549 /* 550 * The security check is questionable... We single 551 * out the RO attribute for checking by the security 552 * module, just because it maps to a file mode. 553 */ 554 err = security_inode_setattr(file_mnt_idmap(file), 555 file->f_path.dentry, &ia); 556 if (err) 557 goto out_unlock_inode; 558 559 /* This MUST be done before doing anything irreversible... */ 560 err = exfat_setattr(file_mnt_idmap(file), file->f_path.dentry, &ia); 561 if (err) 562 goto out_unlock_inode; 563 564 fsnotify_change(file->f_path.dentry, ia.ia_valid); 565 566 exfat_save_attr(inode, attr); 567 mark_inode_dirty(inode); 568 out_unlock_inode: 569 inode_unlock(inode); 570 mnt_drop_write_file(file); 571 out: 572 return err; 573 } 574 575 static int exfat_ioctl_fitrim(struct inode *inode, unsigned long arg) 576 { 577 struct fstrim_range range; 578 int ret = 0; 579 580 if (!capable(CAP_SYS_ADMIN)) 581 return -EPERM; 582 583 if (!bdev_max_discard_sectors(inode->i_sb->s_bdev)) 584 return -EOPNOTSUPP; 585 586 if (copy_from_user(&range, (struct fstrim_range __user *)arg, sizeof(range))) 587 return -EFAULT; 588 589 range.minlen = max_t(unsigned int, range.minlen, 590 bdev_discard_granularity(inode->i_sb->s_bdev)); 591 592 ret = exfat_trim_fs(inode, &range); 593 if (ret < 0) 594 return ret; 595 596 if (copy_to_user((struct fstrim_range __user *)arg, &range, sizeof(range))) 597 return -EFAULT; 598 599 return 0; 600 } 601 602 static int exfat_ioctl_shutdown(struct super_block *sb, unsigned long arg) 603 { 604 u32 flags; 605 606 if (!capable(CAP_SYS_ADMIN)) 607 return -EPERM; 608 609 if (get_user(flags, (__u32 __user *)arg)) 610 return -EFAULT; 611 612 return exfat_force_shutdown(sb, flags); 613 } 614 615 static int exfat_ioctl_get_volume_label(struct super_block *sb, unsigned long arg) 616 { 617 int ret; 618 char label[FSLABEL_MAX] = {0}; 619 struct exfat_uni_name uniname; 620 621 ret = exfat_read_volume_label(sb, &uniname); 622 if (ret < 0) 623 return ret; 624 625 ret = exfat_utf16_to_nls(sb, &uniname, label, sizeof(label)); 626 if (ret < 0) 627 return ret; 628 629 if (copy_to_user((char __user *)arg, label, ret + 1)) 630 return -EFAULT; 631 632 return 0; 633 } 634 635 static int exfat_ioctl_set_volume_label(struct super_block *sb, 636 unsigned long arg) 637 { 638 int ret = 0, lossy, label_len; 639 char label[FSLABEL_MAX] = {0}; 640 struct exfat_uni_name uniname; 641 642 if (!capable(CAP_SYS_ADMIN)) 643 return -EPERM; 644 645 if (copy_from_user(label, (char __user *)arg, FSLABEL_MAX)) 646 return -EFAULT; 647 648 memset(&uniname, 0, sizeof(uniname)); 649 label_len = strnlen(label, FSLABEL_MAX - 1); 650 if (label[0]) { 651 ret = exfat_nls_to_utf16(sb, label, label_len, 652 &uniname, &lossy); 653 if (ret < 0) 654 return ret; 655 else if (lossy & NLS_NAME_LOSSY) 656 return -EINVAL; 657 } 658 659 uniname.name_len = ret; 660 661 return exfat_write_volume_label(sb, &uniname); 662 } 663 664 long exfat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 665 { 666 struct inode *inode = file_inode(filp); 667 u32 __user *user_attr = (u32 __user *)arg; 668 669 switch (cmd) { 670 case FAT_IOCTL_GET_ATTRIBUTES: 671 return exfat_ioctl_get_attributes(inode, user_attr); 672 case FAT_IOCTL_SET_ATTRIBUTES: 673 return exfat_ioctl_set_attributes(filp, user_attr); 674 case EXFAT_IOC_SHUTDOWN: 675 return exfat_ioctl_shutdown(inode->i_sb, arg); 676 case FITRIM: 677 return exfat_ioctl_fitrim(inode, arg); 678 case FS_IOC_GETFSLABEL: 679 return exfat_ioctl_get_volume_label(inode->i_sb, arg); 680 case FS_IOC_SETFSLABEL: 681 return exfat_ioctl_set_volume_label(inode->i_sb, arg); 682 default: 683 return -ENOTTY; 684 } 685 } 686 687 #ifdef CONFIG_COMPAT 688 long exfat_compat_ioctl(struct file *filp, unsigned int cmd, 689 unsigned long arg) 690 { 691 return exfat_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)); 692 } 693 #endif 694 695 int exfat_file_fsync(struct file *filp, loff_t start, loff_t end, int datasync) 696 { 697 struct inode *inode = filp->f_mapping->host; 698 int err; 699 700 if (unlikely(exfat_forced_shutdown(inode->i_sb))) 701 return -EIO; 702 703 err = simple_fsync_noflush(filp, start, end, datasync); 704 if (err) 705 return err; 706 707 err = sync_blockdev(inode->i_sb->s_bdev); 708 if (err) 709 return err; 710 711 return blkdev_issue_flush(inode->i_sb->s_bdev); 712 } 713 714 /* 715 * exfat_prepare_valid_range - populate and dirty folios covering [start, end) 716 * 717 * Populate missing blocks via the read path, which zero-fills data 718 * beyond the current valid_size and skips uptodate blocks. 719 * Mark entire folios dirty so the newly valid range is written back. 720 * 721 * Call before advancing valid_size. 722 */ 723 static int exfat_prepare_valid_range(struct inode *inode, loff_t start, 724 loff_t end) 725 { 726 struct address_space *mapping = inode->i_mapping; 727 loff_t next, pos = start; 728 struct folio *folio; 729 pgoff_t index; 730 731 if (pos >= end) 732 return 0; 733 734 while (pos < end) { 735 index = pos >> PAGE_SHIFT; 736 next = min(((loff_t)index + 1) << PAGE_SHIFT, end); 737 738 balance_dirty_pages_ratelimited(mapping); 739 740 folio = read_mapping_folio(mapping, index, NULL); 741 if (IS_ERR(folio)) 742 return PTR_ERR(folio); 743 744 folio_lock(folio); 745 if (folio->mapping == mapping) { 746 folio_mark_dirty(folio); 747 pos = next; 748 } 749 folio_unlock(folio); 750 folio_put(folio); 751 } 752 753 return 0; 754 } 755 756 static int exfat_extend_valid_size(struct inode *inode, loff_t new_valid_size) 757 { 758 struct exfat_inode_info *ei = EXFAT_I(inode); 759 loff_t old_valid_size = exfat_get_valid_size(ei); 760 int ret = 0; 761 762 if (old_valid_size < new_valid_size) { 763 inode_dio_wait(inode); 764 765 /* Do not re-zero blocks already covered by zeroed_size. */ 766 loff_t gap_start = max(old_valid_size, exfat_get_zeroed_size(ei)); 767 768 if (i_size_read(inode) < new_valid_size) { 769 /* 770 * Allocate clusters before increasing i_size. The gap 771 * may already be zeroed, so the subsequent zeroing 772 * can be skipped. 773 */ 774 ret = exfat_cont_expand(inode, new_valid_size); 775 if (ret) 776 return ret; 777 } 778 779 /* 780 * Revoke writable PTEs while zeroing the gap. A racing mmap 781 * store re-faults through exfat_page_mkwrite() after valid_size 782 * is updated. 783 */ 784 filemap_invalidate_lock(inode->i_mapping); 785 if (gap_start < new_valid_size) 786 unmap_mapping_range(inode->i_mapping, gap_start, 787 new_valid_size - gap_start, 0); 788 ret = exfat_prepare_valid_range(inode, gap_start, new_valid_size); 789 filemap_invalidate_unlock(inode->i_mapping); 790 if (ret) { 791 filemap_invalidate_lock(inode->i_mapping); 792 truncate_setsize(inode, old_valid_size); 793 exfat_truncate(inode); 794 filemap_invalidate_unlock(inode->i_mapping); 795 return ret; 796 } 797 798 exfat_set_valid_size(ei, new_valid_size); 799 exfat_advance_zeroed_size(ei, 800 round_up(new_valid_size, i_blocksize(inode))); 801 mark_inode_dirty(inode); 802 } 803 804 return ret; 805 } 806 807 /* 808 * The clusters for a write are allocated before the data is copied. Free the 809 * ones that a failed or short write has left beyond the end of the file. 810 */ 811 static void exfat_write_failed(struct inode *inode) 812 { 813 inode_dio_wait(inode); 814 filemap_invalidate_lock(inode->i_mapping); 815 truncate_pagecache(inode, i_size_read(inode)); 816 exfat_truncate(inode); 817 filemap_invalidate_unlock(inode->i_mapping); 818 } 819 820 static ssize_t exfat_fallback_buffered_write(struct kiocb *iocb, 821 struct iov_iter *from) 822 { 823 loff_t offset = iocb->ki_pos, end; 824 ssize_t written; 825 int ret; 826 827 iocb->ki_flags &= ~IOCB_DIRECT; 828 829 inode_dio_wait(file_inode(iocb->ki_filp)); 830 831 written = iomap_file_buffered_write(iocb, from, exfat_write_iomap_next, 832 NULL, NULL); 833 if (written < 0) 834 return written; 835 836 end = iocb->ki_pos + written - 1; 837 ret = filemap_write_and_wait_range(iocb->ki_filp->f_mapping, 838 offset, end); 839 if (ret) 840 return -EIO; 841 842 invalidate_mapping_pages(iocb->ki_filp->f_mapping, 843 offset >> PAGE_SHIFT, 844 end >> PAGE_SHIFT); 845 846 return written; 847 } 848 849 static ssize_t exfat_dio_write_iter(struct kiocb *iocb, struct iov_iter *from) 850 { 851 ssize_t ret; 852 853 ret = iomap_dio_rw(iocb, from, exfat_write_iomap_next, 854 &exfat_write_dio_ops, 0, NULL, 0); 855 if (ret == -ENOTBLK) 856 ret = 0; 857 else if (ret < 0) 858 return ret; 859 860 if (iov_iter_count(from)) { 861 ssize_t written; 862 863 written = exfat_fallback_buffered_write(iocb, from); 864 if (written < 0) 865 return written; 866 ret += written; 867 } 868 869 return ret; 870 } 871 872 static ssize_t exfat_file_write_iter(struct kiocb *iocb, struct iov_iter *iter) 873 { 874 ssize_t ret; 875 struct file *file = iocb->ki_filp; 876 struct inode *inode = file_inode(file); 877 struct exfat_inode_info *ei = EXFAT_I(inode); 878 loff_t pos, valid_size, end; 879 int err; 880 881 if (unlikely(exfat_forced_shutdown(inode->i_sb))) 882 return -EIO; 883 884 inode_lock(inode); 885 886 ret = generic_write_checks(iocb, iter); 887 if (ret <= 0) 888 goto unlock; 889 890 err = file_modified(iocb->ki_filp); 891 if (err) { 892 ret = err; 893 goto unlock; 894 } 895 896 pos = iocb->ki_pos; 897 end = pos + iov_iter_count(iter); 898 if (pos > i_size_read(inode)) 899 truncate_pagecache(inode, i_size_read(inode)); 900 901 valid_size = exfat_get_valid_size(ei); 902 if (pos > valid_size) { 903 ret = exfat_extend_valid_size(inode, pos); 904 if (ret < 0 && ret != -ENOSPC) { 905 exfat_err(inode->i_sb, 906 "write: fail to zero from %llu to %llu(%zd)", 907 valid_size, pos, ret); 908 } 909 if (ret < 0) 910 goto unlock; 911 912 pos = valid_size; 913 } 914 915 if (iocb->ki_flags & IOCB_DIRECT) { 916 ret = exfat_dio_write_iter(iocb, iter); 917 } else { 918 /* 919 * Prevent concurrent direct I/O and buffered I/O to the same file 920 * range. Wait for in-flight DIO to finish before dirtying pages. 921 */ 922 inode_dio_wait(inode); 923 ret = iomap_file_buffered_write(iocb, iter, 924 exfat_write_iomap_next, NULL, NULL); 925 } 926 if (ret != -EIOCBQUEUED && end > i_size_read(inode)) 927 exfat_write_failed(inode); 928 if (ret < 0) 929 goto unlock; 930 931 inode_unlock(inode); 932 933 if (iocb->ki_pos > pos) { 934 ssize_t err = generic_write_sync(iocb, iocb->ki_pos - pos); 935 936 if (err < 0) 937 return err; 938 } 939 940 return ret; 941 942 unlock: 943 inode_unlock(inode); 944 945 return ret; 946 } 947 948 static ssize_t exfat_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) 949 { 950 struct inode *inode = file_inode(iocb->ki_filp); 951 ssize_t ret; 952 953 if (unlikely(exfat_forced_shutdown(inode->i_sb))) 954 return -EIO; 955 956 inode_lock_shared(inode); 957 958 if (iocb->ki_flags & IOCB_DIRECT) { 959 file_accessed(iocb->ki_filp); 960 ret = iomap_dio_rw(iocb, iter, exfat_iomap_next, NULL, 0, 961 NULL, 0); 962 } else { 963 ret = generic_file_read_iter(iocb, iter); 964 } 965 966 inode_unlock_shared(inode); 967 968 return ret; 969 } 970 971 static vm_fault_t exfat_page_mkwrite(struct vm_fault *vmf) 972 { 973 struct inode *inode = file_inode(vmf->vma->vm_file); 974 struct exfat_inode_info *ei = EXFAT_I(inode); 975 vm_fault_t ret; 976 loff_t new_valid_size, mmap_valid_size, fault_page_start; 977 978 if (!inode_trylock(inode)) 979 return VM_FAULT_RETRY; 980 981 mmap_valid_size = ((loff_t)vmf->pgoff + 1) << PAGE_SHIFT; 982 fault_page_start = ((loff_t)vmf->pgoff) << PAGE_SHIFT; 983 new_valid_size = min(mmap_valid_size, i_size_read(inode)); 984 985 if (exfat_get_valid_size(ei) < new_valid_size) { 986 loff_t zeroed_size = exfat_get_zeroed_size(ei); 987 988 if (zeroed_size < fault_page_start) { 989 int err; 990 991 /* 992 * Zero only the gap below the faulting page. The read 993 * fault populated its folio and iomap_page_mkwrite() 994 * will dirty it. 995 */ 996 err = exfat_prepare_valid_range(inode, zeroed_size, 997 fault_page_start); 998 if (err < 0) { 999 inode_unlock(inode); 1000 return vmf_fs_error(err); 1001 } 1002 } 1003 1004 /* 1005 * Track zeroed_size by block, not page, because writeback stops 1006 * at i_size recording blocks wholly beyond it could skip a 1007 * later required zeroing. 1008 */ 1009 exfat_advance_zeroed_size(ei, 1010 round_up(new_valid_size, i_blocksize(inode))); 1011 exfat_set_valid_size(ei, new_valid_size); 1012 mark_inode_dirty(inode); 1013 } 1014 1015 sb_start_pagefault(inode->i_sb); 1016 file_update_time(vmf->vma->vm_file); 1017 1018 filemap_invalidate_lock_shared(inode->i_mapping); 1019 ret = iomap_page_mkwrite(vmf, exfat_iomap_next, NULL); 1020 filemap_invalidate_unlock_shared(inode->i_mapping); 1021 sb_end_pagefault(inode->i_sb); 1022 inode_unlock(inode); 1023 1024 return ret; 1025 } 1026 1027 static const struct vm_operations_struct exfat_file_vm_ops = { 1028 .fault = filemap_fault, 1029 .page_mkwrite = exfat_page_mkwrite, 1030 }; 1031 1032 static int exfat_file_mmap_prepare(struct vm_area_desc *desc) 1033 { 1034 struct file *file = desc->file; 1035 1036 if (unlikely(exfat_forced_shutdown(file_inode(desc->file)->i_sb))) 1037 return -EIO; 1038 1039 file_accessed(file); 1040 desc->vm_ops = &exfat_file_vm_ops; 1041 return 0; 1042 } 1043 1044 static ssize_t exfat_splice_read(struct file *in, loff_t *ppos, 1045 struct pipe_inode_info *pipe, size_t len, unsigned int flags) 1046 { 1047 if (unlikely(exfat_forced_shutdown(file_inode(in)->i_sb))) 1048 return -EIO; 1049 1050 return filemap_splice_read(in, ppos, pipe, len, flags); 1051 } 1052 1053 static int exfat_file_open(struct inode *inode, struct file *filp) 1054 { 1055 int err; 1056 1057 if (unlikely(exfat_forced_shutdown(inode->i_sb))) 1058 return -EIO; 1059 1060 err = generic_file_open(inode, filp); 1061 if (err) 1062 return err; 1063 1064 filp->f_mode |= FMODE_CAN_ODIRECT; 1065 1066 return 0; 1067 } 1068 1069 static loff_t exfat_file_llseek(struct file *file, loff_t offset, int whence) 1070 { 1071 struct inode *inode = file->f_mapping->host; 1072 1073 switch (whence) { 1074 case SEEK_HOLE: 1075 inode_lock_shared(inode); 1076 offset = iomap_seek_hole(inode, offset, exfat_iomap_next); 1077 inode_unlock_shared(inode); 1078 break; 1079 case SEEK_DATA: 1080 inode_lock_shared(inode); 1081 offset = iomap_seek_data(inode, offset, exfat_iomap_next); 1082 inode_unlock_shared(inode); 1083 break; 1084 default: 1085 return generic_file_llseek(file, offset, whence); 1086 } 1087 if (offset < 0) 1088 return offset; 1089 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); 1090 } 1091 1092 const struct file_operations exfat_file_operations = { 1093 .open = exfat_file_open, 1094 .llseek = exfat_file_llseek, 1095 .read_iter = exfat_file_read_iter, 1096 .write_iter = exfat_file_write_iter, 1097 .unlocked_ioctl = exfat_ioctl, 1098 #ifdef CONFIG_COMPAT 1099 .compat_ioctl = exfat_compat_ioctl, 1100 #endif 1101 .mmap_prepare = exfat_file_mmap_prepare, 1102 .fsync = exfat_file_fsync, 1103 .splice_read = exfat_splice_read, 1104 .splice_write = iter_file_splice_write, 1105 .fallocate = exfat_fallocate, 1106 .setlease = generic_setlease, 1107 }; 1108 1109 const struct inode_operations exfat_file_inode_operations = { 1110 .setattr = exfat_setattr, 1111 .getattr = exfat_getattr, 1112 .fileattr_get = exfat_fileattr_get, 1113 }; 1114