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