1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * fs/libfs.c 4 * Library for filesystems writers. 5 */ 6 7 #include <linux/blkdev.h> 8 #include <linux/export.h> 9 #include <linux/pagemap.h> 10 #include <linux/slab.h> 11 #include <linux/cred.h> 12 #include <linux/mount.h> 13 #include <linux/vfs.h> 14 #include <linux/quotaops.h> 15 #include <linux/mutex.h> 16 #include <linux/namei.h> 17 #include <linux/exportfs.h> 18 #include <linux/iversion.h> 19 #include <linux/writeback.h> 20 #include <linux/buffer_head.h> /* sync_mapping_buffers */ 21 #include <linux/fs_context.h> 22 #include <linux/pseudo_fs.h> 23 #include <linux/fsnotify.h> 24 #include <linux/unicode.h> 25 #include <linux/fscrypt.h> 26 27 #include <linux/uaccess.h> 28 29 #include "internal.h" 30 31 int simple_getattr(struct mnt_idmap *idmap, const struct path *path, 32 struct kstat *stat, u32 request_mask, 33 unsigned int query_flags) 34 { 35 struct inode *inode = d_inode(path->dentry); 36 generic_fillattr(&nop_mnt_idmap, inode, stat); 37 stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9); 38 return 0; 39 } 40 EXPORT_SYMBOL(simple_getattr); 41 42 int simple_statfs(struct dentry *dentry, struct kstatfs *buf) 43 { 44 buf->f_type = dentry->d_sb->s_magic; 45 buf->f_bsize = PAGE_SIZE; 46 buf->f_namelen = NAME_MAX; 47 return 0; 48 } 49 EXPORT_SYMBOL(simple_statfs); 50 51 /* 52 * Retaining negative dentries for an in-memory filesystem just wastes 53 * memory and lookup time: arrange for them to be deleted immediately. 54 */ 55 int always_delete_dentry(const struct dentry *dentry) 56 { 57 return 1; 58 } 59 EXPORT_SYMBOL(always_delete_dentry); 60 61 const struct dentry_operations simple_dentry_operations = { 62 .d_delete = always_delete_dentry, 63 }; 64 EXPORT_SYMBOL(simple_dentry_operations); 65 66 /* 67 * Lookup the data. This is trivial - if the dentry didn't already 68 * exist, we know it is negative. Set d_op to delete negative dentries. 69 */ 70 struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 71 { 72 if (dentry->d_name.len > NAME_MAX) 73 return ERR_PTR(-ENAMETOOLONG); 74 if (!dentry->d_sb->s_d_op) 75 d_set_d_op(dentry, &simple_dentry_operations); 76 d_add(dentry, NULL); 77 return NULL; 78 } 79 EXPORT_SYMBOL(simple_lookup); 80 81 int dcache_dir_open(struct inode *inode, struct file *file) 82 { 83 file->private_data = d_alloc_cursor(file->f_path.dentry); 84 85 return file->private_data ? 0 : -ENOMEM; 86 } 87 EXPORT_SYMBOL(dcache_dir_open); 88 89 int dcache_dir_close(struct inode *inode, struct file *file) 90 { 91 dput(file->private_data); 92 return 0; 93 } 94 EXPORT_SYMBOL(dcache_dir_close); 95 96 /* parent is locked at least shared */ 97 /* 98 * Returns an element of siblings' list. 99 * We are looking for <count>th positive after <p>; if 100 * found, dentry is grabbed and returned to caller. 101 * If no such element exists, NULL is returned. 102 */ 103 static struct dentry *scan_positives(struct dentry *cursor, 104 struct list_head *p, 105 loff_t count, 106 struct dentry *last) 107 { 108 struct dentry *dentry = cursor->d_parent, *found = NULL; 109 110 spin_lock(&dentry->d_lock); 111 while ((p = p->next) != &dentry->d_subdirs) { 112 struct dentry *d = list_entry(p, struct dentry, d_child); 113 // we must at least skip cursors, to avoid livelocks 114 if (d->d_flags & DCACHE_DENTRY_CURSOR) 115 continue; 116 if (simple_positive(d) && !--count) { 117 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED); 118 if (simple_positive(d)) 119 found = dget_dlock(d); 120 spin_unlock(&d->d_lock); 121 if (likely(found)) 122 break; 123 count = 1; 124 } 125 if (need_resched()) { 126 list_move(&cursor->d_child, p); 127 p = &cursor->d_child; 128 spin_unlock(&dentry->d_lock); 129 cond_resched(); 130 spin_lock(&dentry->d_lock); 131 } 132 } 133 spin_unlock(&dentry->d_lock); 134 dput(last); 135 return found; 136 } 137 138 loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence) 139 { 140 struct dentry *dentry = file->f_path.dentry; 141 switch (whence) { 142 case 1: 143 offset += file->f_pos; 144 fallthrough; 145 case 0: 146 if (offset >= 0) 147 break; 148 fallthrough; 149 default: 150 return -EINVAL; 151 } 152 if (offset != file->f_pos) { 153 struct dentry *cursor = file->private_data; 154 struct dentry *to = NULL; 155 156 inode_lock_shared(dentry->d_inode); 157 158 if (offset > 2) 159 to = scan_positives(cursor, &dentry->d_subdirs, 160 offset - 2, NULL); 161 spin_lock(&dentry->d_lock); 162 if (to) 163 list_move(&cursor->d_child, &to->d_child); 164 else 165 list_del_init(&cursor->d_child); 166 spin_unlock(&dentry->d_lock); 167 dput(to); 168 169 file->f_pos = offset; 170 171 inode_unlock_shared(dentry->d_inode); 172 } 173 return offset; 174 } 175 EXPORT_SYMBOL(dcache_dir_lseek); 176 177 /* 178 * Directory is locked and all positive dentries in it are safe, since 179 * for ramfs-type trees they can't go away without unlink() or rmdir(), 180 * both impossible due to the lock on directory. 181 */ 182 183 int dcache_readdir(struct file *file, struct dir_context *ctx) 184 { 185 struct dentry *dentry = file->f_path.dentry; 186 struct dentry *cursor = file->private_data; 187 struct list_head *anchor = &dentry->d_subdirs; 188 struct dentry *next = NULL; 189 struct list_head *p; 190 191 if (!dir_emit_dots(file, ctx)) 192 return 0; 193 194 if (ctx->pos == 2) 195 p = anchor; 196 else if (!list_empty(&cursor->d_child)) 197 p = &cursor->d_child; 198 else 199 return 0; 200 201 while ((next = scan_positives(cursor, p, 1, next)) != NULL) { 202 if (!dir_emit(ctx, next->d_name.name, next->d_name.len, 203 d_inode(next)->i_ino, 204 fs_umode_to_dtype(d_inode(next)->i_mode))) 205 break; 206 ctx->pos++; 207 p = &next->d_child; 208 } 209 spin_lock(&dentry->d_lock); 210 if (next) 211 list_move_tail(&cursor->d_child, &next->d_child); 212 else 213 list_del_init(&cursor->d_child); 214 spin_unlock(&dentry->d_lock); 215 dput(next); 216 217 return 0; 218 } 219 EXPORT_SYMBOL(dcache_readdir); 220 221 ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos) 222 { 223 return -EISDIR; 224 } 225 EXPORT_SYMBOL(generic_read_dir); 226 227 const struct file_operations simple_dir_operations = { 228 .open = dcache_dir_open, 229 .release = dcache_dir_close, 230 .llseek = dcache_dir_lseek, 231 .read = generic_read_dir, 232 .iterate_shared = dcache_readdir, 233 .fsync = noop_fsync, 234 }; 235 EXPORT_SYMBOL(simple_dir_operations); 236 237 const struct inode_operations simple_dir_inode_operations = { 238 .lookup = simple_lookup, 239 }; 240 EXPORT_SYMBOL(simple_dir_inode_operations); 241 242 static void offset_set(struct dentry *dentry, u32 offset) 243 { 244 dentry->d_fsdata = (void *)((uintptr_t)(offset)); 245 } 246 247 static u32 dentry2offset(struct dentry *dentry) 248 { 249 return (u32)((uintptr_t)(dentry->d_fsdata)); 250 } 251 252 /** 253 * simple_offset_init - initialize an offset_ctx 254 * @octx: directory offset map to be initialized 255 * 256 */ 257 void simple_offset_init(struct offset_ctx *octx) 258 { 259 xa_init_flags(&octx->xa, XA_FLAGS_ALLOC1); 260 261 /* 0 is '.', 1 is '..', so always start with offset 2 */ 262 octx->next_offset = 2; 263 } 264 265 /** 266 * simple_offset_add - Add an entry to a directory's offset map 267 * @octx: directory offset ctx to be updated 268 * @dentry: new dentry being added 269 * 270 * Returns zero on success. @so_ctx and the dentry offset are updated. 271 * Otherwise, a negative errno value is returned. 272 */ 273 int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry) 274 { 275 static const struct xa_limit limit = XA_LIMIT(2, U32_MAX); 276 u32 offset; 277 int ret; 278 279 if (dentry2offset(dentry) != 0) 280 return -EBUSY; 281 282 ret = xa_alloc_cyclic(&octx->xa, &offset, dentry, limit, 283 &octx->next_offset, GFP_KERNEL); 284 if (ret < 0) 285 return ret; 286 287 offset_set(dentry, offset); 288 return 0; 289 } 290 291 /** 292 * simple_offset_remove - Remove an entry to a directory's offset map 293 * @octx: directory offset ctx to be updated 294 * @dentry: dentry being removed 295 * 296 */ 297 void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry) 298 { 299 u32 offset; 300 301 offset = dentry2offset(dentry); 302 if (offset == 0) 303 return; 304 305 xa_erase(&octx->xa, offset); 306 offset_set(dentry, 0); 307 } 308 309 /** 310 * simple_offset_rename_exchange - exchange rename with directory offsets 311 * @old_dir: parent of dentry being moved 312 * @old_dentry: dentry being moved 313 * @new_dir: destination parent 314 * @new_dentry: destination dentry 315 * 316 * Returns zero on success. Otherwise a negative errno is returned and the 317 * rename is rolled back. 318 */ 319 int simple_offset_rename_exchange(struct inode *old_dir, 320 struct dentry *old_dentry, 321 struct inode *new_dir, 322 struct dentry *new_dentry) 323 { 324 struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir); 325 struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir); 326 u32 old_index = dentry2offset(old_dentry); 327 u32 new_index = dentry2offset(new_dentry); 328 int ret; 329 330 simple_offset_remove(old_ctx, old_dentry); 331 simple_offset_remove(new_ctx, new_dentry); 332 333 ret = simple_offset_add(new_ctx, old_dentry); 334 if (ret) 335 goto out_restore; 336 337 ret = simple_offset_add(old_ctx, new_dentry); 338 if (ret) { 339 simple_offset_remove(new_ctx, old_dentry); 340 goto out_restore; 341 } 342 343 ret = simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry); 344 if (ret) { 345 simple_offset_remove(new_ctx, old_dentry); 346 simple_offset_remove(old_ctx, new_dentry); 347 goto out_restore; 348 } 349 return 0; 350 351 out_restore: 352 offset_set(old_dentry, old_index); 353 xa_store(&old_ctx->xa, old_index, old_dentry, GFP_KERNEL); 354 offset_set(new_dentry, new_index); 355 xa_store(&new_ctx->xa, new_index, new_dentry, GFP_KERNEL); 356 return ret; 357 } 358 359 /** 360 * simple_offset_destroy - Release offset map 361 * @octx: directory offset ctx that is about to be destroyed 362 * 363 * During fs teardown (eg. umount), a directory's offset map might still 364 * contain entries. xa_destroy() cleans out anything that remains. 365 */ 366 void simple_offset_destroy(struct offset_ctx *octx) 367 { 368 xa_destroy(&octx->xa); 369 } 370 371 /** 372 * offset_dir_llseek - Advance the read position of a directory descriptor 373 * @file: an open directory whose position is to be updated 374 * @offset: a byte offset 375 * @whence: enumerator describing the starting position for this update 376 * 377 * SEEK_END, SEEK_DATA, and SEEK_HOLE are not supported for directories. 378 * 379 * Returns the updated read position if successful; otherwise a 380 * negative errno is returned and the read position remains unchanged. 381 */ 382 static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence) 383 { 384 switch (whence) { 385 case SEEK_CUR: 386 offset += file->f_pos; 387 fallthrough; 388 case SEEK_SET: 389 if (offset >= 0) 390 break; 391 fallthrough; 392 default: 393 return -EINVAL; 394 } 395 396 return vfs_setpos(file, offset, U32_MAX); 397 } 398 399 static struct dentry *offset_find_next(struct xa_state *xas) 400 { 401 struct dentry *child, *found = NULL; 402 403 rcu_read_lock(); 404 child = xas_next_entry(xas, U32_MAX); 405 if (!child) 406 goto out; 407 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED); 408 if (simple_positive(child)) 409 found = dget_dlock(child); 410 spin_unlock(&child->d_lock); 411 out: 412 rcu_read_unlock(); 413 return found; 414 } 415 416 static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry) 417 { 418 u32 offset = dentry2offset(dentry); 419 struct inode *inode = d_inode(dentry); 420 421 return ctx->actor(ctx, dentry->d_name.name, dentry->d_name.len, offset, 422 inode->i_ino, fs_umode_to_dtype(inode->i_mode)); 423 } 424 425 static void offset_iterate_dir(struct dentry *dir, struct dir_context *ctx) 426 { 427 struct inode *inode = d_inode(dir); 428 struct offset_ctx *so_ctx = inode->i_op->get_offset_ctx(inode); 429 XA_STATE(xas, &so_ctx->xa, ctx->pos); 430 struct dentry *dentry; 431 432 while (true) { 433 spin_lock(&dir->d_lock); 434 dentry = offset_find_next(&xas); 435 spin_unlock(&dir->d_lock); 436 if (!dentry) 437 break; 438 439 if (!offset_dir_emit(ctx, dentry)) { 440 dput(dentry); 441 break; 442 } 443 444 dput(dentry); 445 ctx->pos = xas.xa_index + 1; 446 } 447 } 448 449 /** 450 * offset_readdir - Emit entries starting at offset @ctx->pos 451 * @file: an open directory to iterate over 452 * @ctx: directory iteration context 453 * 454 * Caller must hold @file's i_rwsem to prevent insertion or removal of 455 * entries during this call. 456 * 457 * On entry, @ctx->pos contains an offset that represents the first entry 458 * to be read from the directory. 459 * 460 * The operation continues until there are no more entries to read, or 461 * until the ctx->actor indicates there is no more space in the caller's 462 * output buffer. 463 * 464 * On return, @ctx->pos contains an offset that will read the next entry 465 * in this directory when shmem_readdir() is called again with @ctx. 466 * 467 * Return values: 468 * %0 - Complete 469 */ 470 static int offset_readdir(struct file *file, struct dir_context *ctx) 471 { 472 struct dentry *dir = file->f_path.dentry; 473 474 lockdep_assert_held(&d_inode(dir)->i_rwsem); 475 476 if (!dir_emit_dots(file, ctx)) 477 return 0; 478 479 offset_iterate_dir(dir, ctx); 480 return 0; 481 } 482 483 const struct file_operations simple_offset_dir_operations = { 484 .llseek = offset_dir_llseek, 485 .iterate_shared = offset_readdir, 486 .read = generic_read_dir, 487 .fsync = noop_fsync, 488 }; 489 490 static struct dentry *find_next_child(struct dentry *parent, struct dentry *prev) 491 { 492 struct dentry *child = NULL; 493 struct list_head *p = prev ? &prev->d_child : &parent->d_subdirs; 494 495 spin_lock(&parent->d_lock); 496 while ((p = p->next) != &parent->d_subdirs) { 497 struct dentry *d = container_of(p, struct dentry, d_child); 498 if (simple_positive(d)) { 499 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED); 500 if (simple_positive(d)) 501 child = dget_dlock(d); 502 spin_unlock(&d->d_lock); 503 if (likely(child)) 504 break; 505 } 506 } 507 spin_unlock(&parent->d_lock); 508 dput(prev); 509 return child; 510 } 511 512 void simple_recursive_removal(struct dentry *dentry, 513 void (*callback)(struct dentry *)) 514 { 515 struct dentry *this = dget(dentry); 516 while (true) { 517 struct dentry *victim = NULL, *child; 518 struct inode *inode = this->d_inode; 519 520 inode_lock(inode); 521 if (d_is_dir(this)) 522 inode->i_flags |= S_DEAD; 523 while ((child = find_next_child(this, victim)) == NULL) { 524 // kill and ascend 525 // update metadata while it's still locked 526 inode->i_ctime = current_time(inode); 527 clear_nlink(inode); 528 inode_unlock(inode); 529 victim = this; 530 this = this->d_parent; 531 inode = this->d_inode; 532 inode_lock(inode); 533 if (simple_positive(victim)) { 534 d_invalidate(victim); // avoid lost mounts 535 if (d_is_dir(victim)) 536 fsnotify_rmdir(inode, victim); 537 else 538 fsnotify_unlink(inode, victim); 539 if (callback) 540 callback(victim); 541 dput(victim); // unpin it 542 } 543 if (victim == dentry) { 544 inode->i_ctime = inode->i_mtime = 545 current_time(inode); 546 if (d_is_dir(dentry)) 547 drop_nlink(inode); 548 inode_unlock(inode); 549 dput(dentry); 550 return; 551 } 552 } 553 inode_unlock(inode); 554 this = child; 555 } 556 } 557 EXPORT_SYMBOL(simple_recursive_removal); 558 559 static const struct super_operations simple_super_operations = { 560 .statfs = simple_statfs, 561 }; 562 563 static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc) 564 { 565 struct pseudo_fs_context *ctx = fc->fs_private; 566 struct inode *root; 567 568 s->s_maxbytes = MAX_LFS_FILESIZE; 569 s->s_blocksize = PAGE_SIZE; 570 s->s_blocksize_bits = PAGE_SHIFT; 571 s->s_magic = ctx->magic; 572 s->s_op = ctx->ops ?: &simple_super_operations; 573 s->s_xattr = ctx->xattr; 574 s->s_time_gran = 1; 575 root = new_inode(s); 576 if (!root) 577 return -ENOMEM; 578 579 /* 580 * since this is the first inode, make it number 1. New inodes created 581 * after this must take care not to collide with it (by passing 582 * max_reserved of 1 to iunique). 583 */ 584 root->i_ino = 1; 585 root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR; 586 root->i_atime = root->i_mtime = root->i_ctime = current_time(root); 587 s->s_root = d_make_root(root); 588 if (!s->s_root) 589 return -ENOMEM; 590 s->s_d_op = ctx->dops; 591 return 0; 592 } 593 594 static int pseudo_fs_get_tree(struct fs_context *fc) 595 { 596 return get_tree_nodev(fc, pseudo_fs_fill_super); 597 } 598 599 static void pseudo_fs_free(struct fs_context *fc) 600 { 601 kfree(fc->fs_private); 602 } 603 604 static const struct fs_context_operations pseudo_fs_context_ops = { 605 .free = pseudo_fs_free, 606 .get_tree = pseudo_fs_get_tree, 607 }; 608 609 /* 610 * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that 611 * will never be mountable) 612 */ 613 struct pseudo_fs_context *init_pseudo(struct fs_context *fc, 614 unsigned long magic) 615 { 616 struct pseudo_fs_context *ctx; 617 618 ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL); 619 if (likely(ctx)) { 620 ctx->magic = magic; 621 fc->fs_private = ctx; 622 fc->ops = &pseudo_fs_context_ops; 623 fc->sb_flags |= SB_NOUSER; 624 fc->global = true; 625 } 626 return ctx; 627 } 628 EXPORT_SYMBOL(init_pseudo); 629 630 int simple_open(struct inode *inode, struct file *file) 631 { 632 if (inode->i_private) 633 file->private_data = inode->i_private; 634 return 0; 635 } 636 EXPORT_SYMBOL(simple_open); 637 638 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 639 { 640 struct inode *inode = d_inode(old_dentry); 641 642 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); 643 inc_nlink(inode); 644 ihold(inode); 645 dget(dentry); 646 d_instantiate(dentry, inode); 647 return 0; 648 } 649 EXPORT_SYMBOL(simple_link); 650 651 int simple_empty(struct dentry *dentry) 652 { 653 struct dentry *child; 654 int ret = 0; 655 656 spin_lock(&dentry->d_lock); 657 list_for_each_entry(child, &dentry->d_subdirs, d_child) { 658 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED); 659 if (simple_positive(child)) { 660 spin_unlock(&child->d_lock); 661 goto out; 662 } 663 spin_unlock(&child->d_lock); 664 } 665 ret = 1; 666 out: 667 spin_unlock(&dentry->d_lock); 668 return ret; 669 } 670 EXPORT_SYMBOL(simple_empty); 671 672 int simple_unlink(struct inode *dir, struct dentry *dentry) 673 { 674 struct inode *inode = d_inode(dentry); 675 676 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); 677 drop_nlink(inode); 678 dput(dentry); 679 return 0; 680 } 681 EXPORT_SYMBOL(simple_unlink); 682 683 int simple_rmdir(struct inode *dir, struct dentry *dentry) 684 { 685 if (!simple_empty(dentry)) 686 return -ENOTEMPTY; 687 688 drop_nlink(d_inode(dentry)); 689 simple_unlink(dir, dentry); 690 drop_nlink(dir); 691 return 0; 692 } 693 EXPORT_SYMBOL(simple_rmdir); 694 695 int simple_rename_exchange(struct inode *old_dir, struct dentry *old_dentry, 696 struct inode *new_dir, struct dentry *new_dentry) 697 { 698 bool old_is_dir = d_is_dir(old_dentry); 699 bool new_is_dir = d_is_dir(new_dentry); 700 701 if (old_dir != new_dir && old_is_dir != new_is_dir) { 702 if (old_is_dir) { 703 drop_nlink(old_dir); 704 inc_nlink(new_dir); 705 } else { 706 drop_nlink(new_dir); 707 inc_nlink(old_dir); 708 } 709 } 710 old_dir->i_ctime = old_dir->i_mtime = 711 new_dir->i_ctime = new_dir->i_mtime = 712 d_inode(old_dentry)->i_ctime = 713 d_inode(new_dentry)->i_ctime = current_time(old_dir); 714 715 return 0; 716 } 717 EXPORT_SYMBOL_GPL(simple_rename_exchange); 718 719 int simple_rename(struct mnt_idmap *idmap, struct inode *old_dir, 720 struct dentry *old_dentry, struct inode *new_dir, 721 struct dentry *new_dentry, unsigned int flags) 722 { 723 struct inode *inode = d_inode(old_dentry); 724 int they_are_dirs = d_is_dir(old_dentry); 725 726 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE)) 727 return -EINVAL; 728 729 if (flags & RENAME_EXCHANGE) 730 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry); 731 732 if (!simple_empty(new_dentry)) 733 return -ENOTEMPTY; 734 735 if (d_really_is_positive(new_dentry)) { 736 simple_unlink(new_dir, new_dentry); 737 if (they_are_dirs) { 738 drop_nlink(d_inode(new_dentry)); 739 drop_nlink(old_dir); 740 } 741 } else if (they_are_dirs) { 742 drop_nlink(old_dir); 743 inc_nlink(new_dir); 744 } 745 746 old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime = 747 new_dir->i_mtime = inode->i_ctime = current_time(old_dir); 748 749 return 0; 750 } 751 EXPORT_SYMBOL(simple_rename); 752 753 /** 754 * simple_setattr - setattr for simple filesystem 755 * @idmap: idmap of the target mount 756 * @dentry: dentry 757 * @iattr: iattr structure 758 * 759 * Returns 0 on success, -error on failure. 760 * 761 * simple_setattr is a simple ->setattr implementation without a proper 762 * implementation of size changes. 763 * 764 * It can either be used for in-memory filesystems or special files 765 * on simple regular filesystems. Anything that needs to change on-disk 766 * or wire state on size changes needs its own setattr method. 767 */ 768 int simple_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 769 struct iattr *iattr) 770 { 771 struct inode *inode = d_inode(dentry); 772 int error; 773 774 error = setattr_prepare(idmap, dentry, iattr); 775 if (error) 776 return error; 777 778 if (iattr->ia_valid & ATTR_SIZE) 779 truncate_setsize(inode, iattr->ia_size); 780 setattr_copy(idmap, inode, iattr); 781 mark_inode_dirty(inode); 782 return 0; 783 } 784 EXPORT_SYMBOL(simple_setattr); 785 786 static int simple_read_folio(struct file *file, struct folio *folio) 787 { 788 folio_zero_range(folio, 0, folio_size(folio)); 789 flush_dcache_folio(folio); 790 folio_mark_uptodate(folio); 791 folio_unlock(folio); 792 return 0; 793 } 794 795 int simple_write_begin(struct file *file, struct address_space *mapping, 796 loff_t pos, unsigned len, 797 struct page **pagep, void **fsdata) 798 { 799 struct page *page; 800 pgoff_t index; 801 802 index = pos >> PAGE_SHIFT; 803 804 page = grab_cache_page_write_begin(mapping, index); 805 if (!page) 806 return -ENOMEM; 807 808 *pagep = page; 809 810 if (!PageUptodate(page) && (len != PAGE_SIZE)) { 811 unsigned from = pos & (PAGE_SIZE - 1); 812 813 zero_user_segments(page, 0, from, from + len, PAGE_SIZE); 814 } 815 return 0; 816 } 817 EXPORT_SYMBOL(simple_write_begin); 818 819 /** 820 * simple_write_end - .write_end helper for non-block-device FSes 821 * @file: See .write_end of address_space_operations 822 * @mapping: " 823 * @pos: " 824 * @len: " 825 * @copied: " 826 * @page: " 827 * @fsdata: " 828 * 829 * simple_write_end does the minimum needed for updating a page after writing is 830 * done. It has the same API signature as the .write_end of 831 * address_space_operations vector. So it can just be set onto .write_end for 832 * FSes that don't need any other processing. i_mutex is assumed to be held. 833 * Block based filesystems should use generic_write_end(). 834 * NOTE: Even though i_size might get updated by this function, mark_inode_dirty 835 * is not called, so a filesystem that actually does store data in .write_inode 836 * should extend on what's done here with a call to mark_inode_dirty() in the 837 * case that i_size has changed. 838 * 839 * Use *ONLY* with simple_read_folio() 840 */ 841 static int simple_write_end(struct file *file, struct address_space *mapping, 842 loff_t pos, unsigned len, unsigned copied, 843 struct page *page, void *fsdata) 844 { 845 struct inode *inode = page->mapping->host; 846 loff_t last_pos = pos + copied; 847 848 /* zero the stale part of the page if we did a short copy */ 849 if (!PageUptodate(page)) { 850 if (copied < len) { 851 unsigned from = pos & (PAGE_SIZE - 1); 852 853 zero_user(page, from + copied, len - copied); 854 } 855 SetPageUptodate(page); 856 } 857 /* 858 * No need to use i_size_read() here, the i_size 859 * cannot change under us because we hold the i_mutex. 860 */ 861 if (last_pos > inode->i_size) 862 i_size_write(inode, last_pos); 863 864 set_page_dirty(page); 865 unlock_page(page); 866 put_page(page); 867 868 return copied; 869 } 870 871 /* 872 * Provides ramfs-style behavior: data in the pagecache, but no writeback. 873 */ 874 const struct address_space_operations ram_aops = { 875 .read_folio = simple_read_folio, 876 .write_begin = simple_write_begin, 877 .write_end = simple_write_end, 878 .dirty_folio = noop_dirty_folio, 879 }; 880 EXPORT_SYMBOL(ram_aops); 881 882 /* 883 * the inodes created here are not hashed. If you use iunique to generate 884 * unique inode values later for this filesystem, then you must take care 885 * to pass it an appropriate max_reserved value to avoid collisions. 886 */ 887 int simple_fill_super(struct super_block *s, unsigned long magic, 888 const struct tree_descr *files) 889 { 890 struct inode *inode; 891 struct dentry *root; 892 struct dentry *dentry; 893 int i; 894 895 s->s_blocksize = PAGE_SIZE; 896 s->s_blocksize_bits = PAGE_SHIFT; 897 s->s_magic = magic; 898 s->s_op = &simple_super_operations; 899 s->s_time_gran = 1; 900 901 inode = new_inode(s); 902 if (!inode) 903 return -ENOMEM; 904 /* 905 * because the root inode is 1, the files array must not contain an 906 * entry at index 1 907 */ 908 inode->i_ino = 1; 909 inode->i_mode = S_IFDIR | 0755; 910 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 911 inode->i_op = &simple_dir_inode_operations; 912 inode->i_fop = &simple_dir_operations; 913 set_nlink(inode, 2); 914 root = d_make_root(inode); 915 if (!root) 916 return -ENOMEM; 917 for (i = 0; !files->name || files->name[0]; i++, files++) { 918 if (!files->name) 919 continue; 920 921 /* warn if it tries to conflict with the root inode */ 922 if (unlikely(i == 1)) 923 printk(KERN_WARNING "%s: %s passed in a files array" 924 "with an index of 1!\n", __func__, 925 s->s_type->name); 926 927 dentry = d_alloc_name(root, files->name); 928 if (!dentry) 929 goto out; 930 inode = new_inode(s); 931 if (!inode) { 932 dput(dentry); 933 goto out; 934 } 935 inode->i_mode = S_IFREG | files->mode; 936 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 937 inode->i_fop = files->ops; 938 inode->i_ino = i; 939 d_add(dentry, inode); 940 } 941 s->s_root = root; 942 return 0; 943 out: 944 d_genocide(root); 945 shrink_dcache_parent(root); 946 dput(root); 947 return -ENOMEM; 948 } 949 EXPORT_SYMBOL(simple_fill_super); 950 951 static DEFINE_SPINLOCK(pin_fs_lock); 952 953 int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count) 954 { 955 struct vfsmount *mnt = NULL; 956 spin_lock(&pin_fs_lock); 957 if (unlikely(!*mount)) { 958 spin_unlock(&pin_fs_lock); 959 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 960 if (IS_ERR(mnt)) 961 return PTR_ERR(mnt); 962 spin_lock(&pin_fs_lock); 963 if (!*mount) 964 *mount = mnt; 965 } 966 mntget(*mount); 967 ++*count; 968 spin_unlock(&pin_fs_lock); 969 mntput(mnt); 970 return 0; 971 } 972 EXPORT_SYMBOL(simple_pin_fs); 973 974 void simple_release_fs(struct vfsmount **mount, int *count) 975 { 976 struct vfsmount *mnt; 977 spin_lock(&pin_fs_lock); 978 mnt = *mount; 979 if (!--*count) 980 *mount = NULL; 981 spin_unlock(&pin_fs_lock); 982 mntput(mnt); 983 } 984 EXPORT_SYMBOL(simple_release_fs); 985 986 /** 987 * simple_read_from_buffer - copy data from the buffer to user space 988 * @to: the user space buffer to read to 989 * @count: the maximum number of bytes to read 990 * @ppos: the current position in the buffer 991 * @from: the buffer to read from 992 * @available: the size of the buffer 993 * 994 * The simple_read_from_buffer() function reads up to @count bytes from the 995 * buffer @from at offset @ppos into the user space address starting at @to. 996 * 997 * On success, the number of bytes read is returned and the offset @ppos is 998 * advanced by this number, or negative value is returned on error. 999 **/ 1000 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos, 1001 const void *from, size_t available) 1002 { 1003 loff_t pos = *ppos; 1004 size_t ret; 1005 1006 if (pos < 0) 1007 return -EINVAL; 1008 if (pos >= available || !count) 1009 return 0; 1010 if (count > available - pos) 1011 count = available - pos; 1012 ret = copy_to_user(to, from + pos, count); 1013 if (ret == count) 1014 return -EFAULT; 1015 count -= ret; 1016 *ppos = pos + count; 1017 return count; 1018 } 1019 EXPORT_SYMBOL(simple_read_from_buffer); 1020 1021 /** 1022 * simple_write_to_buffer - copy data from user space to the buffer 1023 * @to: the buffer to write to 1024 * @available: the size of the buffer 1025 * @ppos: the current position in the buffer 1026 * @from: the user space buffer to read from 1027 * @count: the maximum number of bytes to read 1028 * 1029 * The simple_write_to_buffer() function reads up to @count bytes from the user 1030 * space address starting at @from into the buffer @to at offset @ppos. 1031 * 1032 * On success, the number of bytes written is returned and the offset @ppos is 1033 * advanced by this number, or negative value is returned on error. 1034 **/ 1035 ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos, 1036 const void __user *from, size_t count) 1037 { 1038 loff_t pos = *ppos; 1039 size_t res; 1040 1041 if (pos < 0) 1042 return -EINVAL; 1043 if (pos >= available || !count) 1044 return 0; 1045 if (count > available - pos) 1046 count = available - pos; 1047 res = copy_from_user(to + pos, from, count); 1048 if (res == count) 1049 return -EFAULT; 1050 count -= res; 1051 *ppos = pos + count; 1052 return count; 1053 } 1054 EXPORT_SYMBOL(simple_write_to_buffer); 1055 1056 /** 1057 * memory_read_from_buffer - copy data from the buffer 1058 * @to: the kernel space buffer to read to 1059 * @count: the maximum number of bytes to read 1060 * @ppos: the current position in the buffer 1061 * @from: the buffer to read from 1062 * @available: the size of the buffer 1063 * 1064 * The memory_read_from_buffer() function reads up to @count bytes from the 1065 * buffer @from at offset @ppos into the kernel space address starting at @to. 1066 * 1067 * On success, the number of bytes read is returned and the offset @ppos is 1068 * advanced by this number, or negative value is returned on error. 1069 **/ 1070 ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos, 1071 const void *from, size_t available) 1072 { 1073 loff_t pos = *ppos; 1074 1075 if (pos < 0) 1076 return -EINVAL; 1077 if (pos >= available) 1078 return 0; 1079 if (count > available - pos) 1080 count = available - pos; 1081 memcpy(to, from + pos, count); 1082 *ppos = pos + count; 1083 1084 return count; 1085 } 1086 EXPORT_SYMBOL(memory_read_from_buffer); 1087 1088 /* 1089 * Transaction based IO. 1090 * The file expects a single write which triggers the transaction, and then 1091 * possibly a read which collects the result - which is stored in a 1092 * file-local buffer. 1093 */ 1094 1095 void simple_transaction_set(struct file *file, size_t n) 1096 { 1097 struct simple_transaction_argresp *ar = file->private_data; 1098 1099 BUG_ON(n > SIMPLE_TRANSACTION_LIMIT); 1100 1101 /* 1102 * The barrier ensures that ar->size will really remain zero until 1103 * ar->data is ready for reading. 1104 */ 1105 smp_mb(); 1106 ar->size = n; 1107 } 1108 EXPORT_SYMBOL(simple_transaction_set); 1109 1110 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size) 1111 { 1112 struct simple_transaction_argresp *ar; 1113 static DEFINE_SPINLOCK(simple_transaction_lock); 1114 1115 if (size > SIMPLE_TRANSACTION_LIMIT - 1) 1116 return ERR_PTR(-EFBIG); 1117 1118 ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL); 1119 if (!ar) 1120 return ERR_PTR(-ENOMEM); 1121 1122 spin_lock(&simple_transaction_lock); 1123 1124 /* only one write allowed per open */ 1125 if (file->private_data) { 1126 spin_unlock(&simple_transaction_lock); 1127 free_page((unsigned long)ar); 1128 return ERR_PTR(-EBUSY); 1129 } 1130 1131 file->private_data = ar; 1132 1133 spin_unlock(&simple_transaction_lock); 1134 1135 if (copy_from_user(ar->data, buf, size)) 1136 return ERR_PTR(-EFAULT); 1137 1138 return ar->data; 1139 } 1140 EXPORT_SYMBOL(simple_transaction_get); 1141 1142 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos) 1143 { 1144 struct simple_transaction_argresp *ar = file->private_data; 1145 1146 if (!ar) 1147 return 0; 1148 return simple_read_from_buffer(buf, size, pos, ar->data, ar->size); 1149 } 1150 EXPORT_SYMBOL(simple_transaction_read); 1151 1152 int simple_transaction_release(struct inode *inode, struct file *file) 1153 { 1154 free_page((unsigned long)file->private_data); 1155 return 0; 1156 } 1157 EXPORT_SYMBOL(simple_transaction_release); 1158 1159 /* Simple attribute files */ 1160 1161 struct simple_attr { 1162 int (*get)(void *, u64 *); 1163 int (*set)(void *, u64); 1164 char get_buf[24]; /* enough to store a u64 and "\n\0" */ 1165 char set_buf[24]; 1166 void *data; 1167 const char *fmt; /* format for read operation */ 1168 struct mutex mutex; /* protects access to these buffers */ 1169 }; 1170 1171 /* simple_attr_open is called by an actual attribute open file operation 1172 * to set the attribute specific access operations. */ 1173 int simple_attr_open(struct inode *inode, struct file *file, 1174 int (*get)(void *, u64 *), int (*set)(void *, u64), 1175 const char *fmt) 1176 { 1177 struct simple_attr *attr; 1178 1179 attr = kzalloc(sizeof(*attr), GFP_KERNEL); 1180 if (!attr) 1181 return -ENOMEM; 1182 1183 attr->get = get; 1184 attr->set = set; 1185 attr->data = inode->i_private; 1186 attr->fmt = fmt; 1187 mutex_init(&attr->mutex); 1188 1189 file->private_data = attr; 1190 1191 return nonseekable_open(inode, file); 1192 } 1193 EXPORT_SYMBOL_GPL(simple_attr_open); 1194 1195 int simple_attr_release(struct inode *inode, struct file *file) 1196 { 1197 kfree(file->private_data); 1198 return 0; 1199 } 1200 EXPORT_SYMBOL_GPL(simple_attr_release); /* GPL-only? This? Really? */ 1201 1202 /* read from the buffer that is filled with the get function */ 1203 ssize_t simple_attr_read(struct file *file, char __user *buf, 1204 size_t len, loff_t *ppos) 1205 { 1206 struct simple_attr *attr; 1207 size_t size; 1208 ssize_t ret; 1209 1210 attr = file->private_data; 1211 1212 if (!attr->get) 1213 return -EACCES; 1214 1215 ret = mutex_lock_interruptible(&attr->mutex); 1216 if (ret) 1217 return ret; 1218 1219 if (*ppos && attr->get_buf[0]) { 1220 /* continued read */ 1221 size = strlen(attr->get_buf); 1222 } else { 1223 /* first read */ 1224 u64 val; 1225 ret = attr->get(attr->data, &val); 1226 if (ret) 1227 goto out; 1228 1229 size = scnprintf(attr->get_buf, sizeof(attr->get_buf), 1230 attr->fmt, (unsigned long long)val); 1231 } 1232 1233 ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size); 1234 out: 1235 mutex_unlock(&attr->mutex); 1236 return ret; 1237 } 1238 EXPORT_SYMBOL_GPL(simple_attr_read); 1239 1240 /* interpret the buffer as a number to call the set function with */ 1241 static ssize_t simple_attr_write_xsigned(struct file *file, const char __user *buf, 1242 size_t len, loff_t *ppos, bool is_signed) 1243 { 1244 struct simple_attr *attr; 1245 unsigned long long val; 1246 size_t size; 1247 ssize_t ret; 1248 1249 attr = file->private_data; 1250 if (!attr->set) 1251 return -EACCES; 1252 1253 ret = mutex_lock_interruptible(&attr->mutex); 1254 if (ret) 1255 return ret; 1256 1257 ret = -EFAULT; 1258 size = min(sizeof(attr->set_buf) - 1, len); 1259 if (copy_from_user(attr->set_buf, buf, size)) 1260 goto out; 1261 1262 attr->set_buf[size] = '\0'; 1263 if (is_signed) 1264 ret = kstrtoll(attr->set_buf, 0, &val); 1265 else 1266 ret = kstrtoull(attr->set_buf, 0, &val); 1267 if (ret) 1268 goto out; 1269 ret = attr->set(attr->data, val); 1270 if (ret == 0) 1271 ret = len; /* on success, claim we got the whole input */ 1272 out: 1273 mutex_unlock(&attr->mutex); 1274 return ret; 1275 } 1276 1277 ssize_t simple_attr_write(struct file *file, const char __user *buf, 1278 size_t len, loff_t *ppos) 1279 { 1280 return simple_attr_write_xsigned(file, buf, len, ppos, false); 1281 } 1282 EXPORT_SYMBOL_GPL(simple_attr_write); 1283 1284 ssize_t simple_attr_write_signed(struct file *file, const char __user *buf, 1285 size_t len, loff_t *ppos) 1286 { 1287 return simple_attr_write_xsigned(file, buf, len, ppos, true); 1288 } 1289 EXPORT_SYMBOL_GPL(simple_attr_write_signed); 1290 1291 /** 1292 * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation 1293 * @sb: filesystem to do the file handle conversion on 1294 * @fid: file handle to convert 1295 * @fh_len: length of the file handle in bytes 1296 * @fh_type: type of file handle 1297 * @get_inode: filesystem callback to retrieve inode 1298 * 1299 * This function decodes @fid as long as it has one of the well-known 1300 * Linux filehandle types and calls @get_inode on it to retrieve the 1301 * inode for the object specified in the file handle. 1302 */ 1303 struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid, 1304 int fh_len, int fh_type, struct inode *(*get_inode) 1305 (struct super_block *sb, u64 ino, u32 gen)) 1306 { 1307 struct inode *inode = NULL; 1308 1309 if (fh_len < 2) 1310 return NULL; 1311 1312 switch (fh_type) { 1313 case FILEID_INO32_GEN: 1314 case FILEID_INO32_GEN_PARENT: 1315 inode = get_inode(sb, fid->i32.ino, fid->i32.gen); 1316 break; 1317 } 1318 1319 return d_obtain_alias(inode); 1320 } 1321 EXPORT_SYMBOL_GPL(generic_fh_to_dentry); 1322 1323 /** 1324 * generic_fh_to_parent - generic helper for the fh_to_parent export operation 1325 * @sb: filesystem to do the file handle conversion on 1326 * @fid: file handle to convert 1327 * @fh_len: length of the file handle in bytes 1328 * @fh_type: type of file handle 1329 * @get_inode: filesystem callback to retrieve inode 1330 * 1331 * This function decodes @fid as long as it has one of the well-known 1332 * Linux filehandle types and calls @get_inode on it to retrieve the 1333 * inode for the _parent_ object specified in the file handle if it 1334 * is specified in the file handle, or NULL otherwise. 1335 */ 1336 struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid, 1337 int fh_len, int fh_type, struct inode *(*get_inode) 1338 (struct super_block *sb, u64 ino, u32 gen)) 1339 { 1340 struct inode *inode = NULL; 1341 1342 if (fh_len <= 2) 1343 return NULL; 1344 1345 switch (fh_type) { 1346 case FILEID_INO32_GEN_PARENT: 1347 inode = get_inode(sb, fid->i32.parent_ino, 1348 (fh_len > 3 ? fid->i32.parent_gen : 0)); 1349 break; 1350 } 1351 1352 return d_obtain_alias(inode); 1353 } 1354 EXPORT_SYMBOL_GPL(generic_fh_to_parent); 1355 1356 /** 1357 * __generic_file_fsync - generic fsync implementation for simple filesystems 1358 * 1359 * @file: file to synchronize 1360 * @start: start offset in bytes 1361 * @end: end offset in bytes (inclusive) 1362 * @datasync: only synchronize essential metadata if true 1363 * 1364 * This is a generic implementation of the fsync method for simple 1365 * filesystems which track all non-inode metadata in the buffers list 1366 * hanging off the address_space structure. 1367 */ 1368 int __generic_file_fsync(struct file *file, loff_t start, loff_t end, 1369 int datasync) 1370 { 1371 struct inode *inode = file->f_mapping->host; 1372 int err; 1373 int ret; 1374 1375 err = file_write_and_wait_range(file, start, end); 1376 if (err) 1377 return err; 1378 1379 inode_lock(inode); 1380 ret = sync_mapping_buffers(inode->i_mapping); 1381 if (!(inode->i_state & I_DIRTY_ALL)) 1382 goto out; 1383 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) 1384 goto out; 1385 1386 err = sync_inode_metadata(inode, 1); 1387 if (ret == 0) 1388 ret = err; 1389 1390 out: 1391 inode_unlock(inode); 1392 /* check and advance again to catch errors after syncing out buffers */ 1393 err = file_check_and_advance_wb_err(file); 1394 if (ret == 0) 1395 ret = err; 1396 return ret; 1397 } 1398 EXPORT_SYMBOL(__generic_file_fsync); 1399 1400 /** 1401 * generic_file_fsync - generic fsync implementation for simple filesystems 1402 * with flush 1403 * @file: file to synchronize 1404 * @start: start offset in bytes 1405 * @end: end offset in bytes (inclusive) 1406 * @datasync: only synchronize essential metadata if true 1407 * 1408 */ 1409 1410 int generic_file_fsync(struct file *file, loff_t start, loff_t end, 1411 int datasync) 1412 { 1413 struct inode *inode = file->f_mapping->host; 1414 int err; 1415 1416 err = __generic_file_fsync(file, start, end, datasync); 1417 if (err) 1418 return err; 1419 return blkdev_issue_flush(inode->i_sb->s_bdev); 1420 } 1421 EXPORT_SYMBOL(generic_file_fsync); 1422 1423 /** 1424 * generic_check_addressable - Check addressability of file system 1425 * @blocksize_bits: log of file system block size 1426 * @num_blocks: number of blocks in file system 1427 * 1428 * Determine whether a file system with @num_blocks blocks (and a 1429 * block size of 2**@blocksize_bits) is addressable by the sector_t 1430 * and page cache of the system. Return 0 if so and -EFBIG otherwise. 1431 */ 1432 int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks) 1433 { 1434 u64 last_fs_block = num_blocks - 1; 1435 u64 last_fs_page = 1436 last_fs_block >> (PAGE_SHIFT - blocksize_bits); 1437 1438 if (unlikely(num_blocks == 0)) 1439 return 0; 1440 1441 if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT)) 1442 return -EINVAL; 1443 1444 if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) || 1445 (last_fs_page > (pgoff_t)(~0ULL))) { 1446 return -EFBIG; 1447 } 1448 return 0; 1449 } 1450 EXPORT_SYMBOL(generic_check_addressable); 1451 1452 /* 1453 * No-op implementation of ->fsync for in-memory filesystems. 1454 */ 1455 int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync) 1456 { 1457 return 0; 1458 } 1459 EXPORT_SYMBOL(noop_fsync); 1460 1461 ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 1462 { 1463 /* 1464 * iomap based filesystems support direct I/O without need for 1465 * this callback. However, it still needs to be set in 1466 * inode->a_ops so that open/fcntl know that direct I/O is 1467 * generally supported. 1468 */ 1469 return -EINVAL; 1470 } 1471 EXPORT_SYMBOL_GPL(noop_direct_IO); 1472 1473 /* Because kfree isn't assignment-compatible with void(void*) ;-/ */ 1474 void kfree_link(void *p) 1475 { 1476 kfree(p); 1477 } 1478 EXPORT_SYMBOL(kfree_link); 1479 1480 struct inode *alloc_anon_inode(struct super_block *s) 1481 { 1482 static const struct address_space_operations anon_aops = { 1483 .dirty_folio = noop_dirty_folio, 1484 }; 1485 struct inode *inode = new_inode_pseudo(s); 1486 1487 if (!inode) 1488 return ERR_PTR(-ENOMEM); 1489 1490 inode->i_ino = get_next_ino(); 1491 inode->i_mapping->a_ops = &anon_aops; 1492 1493 /* 1494 * Mark the inode dirty from the very beginning, 1495 * that way it will never be moved to the dirty 1496 * list because mark_inode_dirty() will think 1497 * that it already _is_ on the dirty list. 1498 */ 1499 inode->i_state = I_DIRTY; 1500 inode->i_mode = S_IRUSR | S_IWUSR; 1501 inode->i_uid = current_fsuid(); 1502 inode->i_gid = current_fsgid(); 1503 inode->i_flags |= S_PRIVATE; 1504 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 1505 return inode; 1506 } 1507 EXPORT_SYMBOL(alloc_anon_inode); 1508 1509 /** 1510 * simple_nosetlease - generic helper for prohibiting leases 1511 * @filp: file pointer 1512 * @arg: type of lease to obtain 1513 * @flp: new lease supplied for insertion 1514 * @priv: private data for lm_setup operation 1515 * 1516 * Generic helper for filesystems that do not wish to allow leases to be set. 1517 * All arguments are ignored and it just returns -EINVAL. 1518 */ 1519 int 1520 simple_nosetlease(struct file *filp, long arg, struct file_lock **flp, 1521 void **priv) 1522 { 1523 return -EINVAL; 1524 } 1525 EXPORT_SYMBOL(simple_nosetlease); 1526 1527 /** 1528 * simple_get_link - generic helper to get the target of "fast" symlinks 1529 * @dentry: not used here 1530 * @inode: the symlink inode 1531 * @done: not used here 1532 * 1533 * Generic helper for filesystems to use for symlink inodes where a pointer to 1534 * the symlink target is stored in ->i_link. NOTE: this isn't normally called, 1535 * since as an optimization the path lookup code uses any non-NULL ->i_link 1536 * directly, without calling ->get_link(). But ->get_link() still must be set, 1537 * to mark the inode_operations as being for a symlink. 1538 * 1539 * Return: the symlink target 1540 */ 1541 const char *simple_get_link(struct dentry *dentry, struct inode *inode, 1542 struct delayed_call *done) 1543 { 1544 return inode->i_link; 1545 } 1546 EXPORT_SYMBOL(simple_get_link); 1547 1548 const struct inode_operations simple_symlink_inode_operations = { 1549 .get_link = simple_get_link, 1550 }; 1551 EXPORT_SYMBOL(simple_symlink_inode_operations); 1552 1553 /* 1554 * Operations for a permanently empty directory. 1555 */ 1556 static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 1557 { 1558 return ERR_PTR(-ENOENT); 1559 } 1560 1561 static int empty_dir_getattr(struct mnt_idmap *idmap, 1562 const struct path *path, struct kstat *stat, 1563 u32 request_mask, unsigned int query_flags) 1564 { 1565 struct inode *inode = d_inode(path->dentry); 1566 generic_fillattr(&nop_mnt_idmap, inode, stat); 1567 return 0; 1568 } 1569 1570 static int empty_dir_setattr(struct mnt_idmap *idmap, 1571 struct dentry *dentry, struct iattr *attr) 1572 { 1573 return -EPERM; 1574 } 1575 1576 static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size) 1577 { 1578 return -EOPNOTSUPP; 1579 } 1580 1581 static const struct inode_operations empty_dir_inode_operations = { 1582 .lookup = empty_dir_lookup, 1583 .permission = generic_permission, 1584 .setattr = empty_dir_setattr, 1585 .getattr = empty_dir_getattr, 1586 .listxattr = empty_dir_listxattr, 1587 }; 1588 1589 static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence) 1590 { 1591 /* An empty directory has two entries . and .. at offsets 0 and 1 */ 1592 return generic_file_llseek_size(file, offset, whence, 2, 2); 1593 } 1594 1595 static int empty_dir_readdir(struct file *file, struct dir_context *ctx) 1596 { 1597 dir_emit_dots(file, ctx); 1598 return 0; 1599 } 1600 1601 static const struct file_operations empty_dir_operations = { 1602 .llseek = empty_dir_llseek, 1603 .read = generic_read_dir, 1604 .iterate_shared = empty_dir_readdir, 1605 .fsync = noop_fsync, 1606 }; 1607 1608 1609 void make_empty_dir_inode(struct inode *inode) 1610 { 1611 set_nlink(inode, 2); 1612 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO; 1613 inode->i_uid = GLOBAL_ROOT_UID; 1614 inode->i_gid = GLOBAL_ROOT_GID; 1615 inode->i_rdev = 0; 1616 inode->i_size = 0; 1617 inode->i_blkbits = PAGE_SHIFT; 1618 inode->i_blocks = 0; 1619 1620 inode->i_op = &empty_dir_inode_operations; 1621 inode->i_opflags &= ~IOP_XATTR; 1622 inode->i_fop = &empty_dir_operations; 1623 } 1624 1625 bool is_empty_dir_inode(struct inode *inode) 1626 { 1627 return (inode->i_fop == &empty_dir_operations) && 1628 (inode->i_op == &empty_dir_inode_operations); 1629 } 1630 1631 #if IS_ENABLED(CONFIG_UNICODE) 1632 /* 1633 * Determine if the name of a dentry should be casefolded. 1634 * 1635 * Return: if names will need casefolding 1636 */ 1637 static bool needs_casefold(const struct inode *dir) 1638 { 1639 return IS_CASEFOLDED(dir) && dir->i_sb->s_encoding; 1640 } 1641 1642 /** 1643 * generic_ci_d_compare - generic d_compare implementation for casefolding filesystems 1644 * @dentry: dentry whose name we are checking against 1645 * @len: len of name of dentry 1646 * @str: str pointer to name of dentry 1647 * @name: Name to compare against 1648 * 1649 * Return: 0 if names match, 1 if mismatch, or -ERRNO 1650 */ 1651 static int generic_ci_d_compare(const struct dentry *dentry, unsigned int len, 1652 const char *str, const struct qstr *name) 1653 { 1654 const struct dentry *parent = READ_ONCE(dentry->d_parent); 1655 const struct inode *dir = READ_ONCE(parent->d_inode); 1656 const struct super_block *sb = dentry->d_sb; 1657 const struct unicode_map *um = sb->s_encoding; 1658 struct qstr qstr = QSTR_INIT(str, len); 1659 char strbuf[DNAME_INLINE_LEN]; 1660 int ret; 1661 1662 if (!dir || !needs_casefold(dir)) 1663 goto fallback; 1664 /* 1665 * If the dentry name is stored in-line, then it may be concurrently 1666 * modified by a rename. If this happens, the VFS will eventually retry 1667 * the lookup, so it doesn't matter what ->d_compare() returns. 1668 * However, it's unsafe to call utf8_strncasecmp() with an unstable 1669 * string. Therefore, we have to copy the name into a temporary buffer. 1670 */ 1671 if (len <= DNAME_INLINE_LEN - 1) { 1672 memcpy(strbuf, str, len); 1673 strbuf[len] = 0; 1674 qstr.name = strbuf; 1675 /* prevent compiler from optimizing out the temporary buffer */ 1676 barrier(); 1677 } 1678 ret = utf8_strncasecmp(um, name, &qstr); 1679 if (ret >= 0) 1680 return ret; 1681 1682 if (sb_has_strict_encoding(sb)) 1683 return -EINVAL; 1684 fallback: 1685 if (len != name->len) 1686 return 1; 1687 return !!memcmp(str, name->name, len); 1688 } 1689 1690 /** 1691 * generic_ci_d_hash - generic d_hash implementation for casefolding filesystems 1692 * @dentry: dentry of the parent directory 1693 * @str: qstr of name whose hash we should fill in 1694 * 1695 * Return: 0 if hash was successful or unchanged, and -EINVAL on error 1696 */ 1697 static int generic_ci_d_hash(const struct dentry *dentry, struct qstr *str) 1698 { 1699 const struct inode *dir = READ_ONCE(dentry->d_inode); 1700 struct super_block *sb = dentry->d_sb; 1701 const struct unicode_map *um = sb->s_encoding; 1702 int ret = 0; 1703 1704 if (!dir || !needs_casefold(dir)) 1705 return 0; 1706 1707 ret = utf8_casefold_hash(um, dentry, str); 1708 if (ret < 0 && sb_has_strict_encoding(sb)) 1709 return -EINVAL; 1710 return 0; 1711 } 1712 1713 static const struct dentry_operations generic_ci_dentry_ops = { 1714 .d_hash = generic_ci_d_hash, 1715 .d_compare = generic_ci_d_compare, 1716 }; 1717 #endif 1718 1719 #ifdef CONFIG_FS_ENCRYPTION 1720 static const struct dentry_operations generic_encrypted_dentry_ops = { 1721 .d_revalidate = fscrypt_d_revalidate, 1722 }; 1723 #endif 1724 1725 #if defined(CONFIG_FS_ENCRYPTION) && IS_ENABLED(CONFIG_UNICODE) 1726 static const struct dentry_operations generic_encrypted_ci_dentry_ops = { 1727 .d_hash = generic_ci_d_hash, 1728 .d_compare = generic_ci_d_compare, 1729 .d_revalidate = fscrypt_d_revalidate, 1730 }; 1731 #endif 1732 1733 /** 1734 * generic_set_encrypted_ci_d_ops - helper for setting d_ops for given dentry 1735 * @dentry: dentry to set ops on 1736 * 1737 * Casefolded directories need d_hash and d_compare set, so that the dentries 1738 * contained in them are handled case-insensitively. Note that these operations 1739 * are needed on the parent directory rather than on the dentries in it, and 1740 * while the casefolding flag can be toggled on and off on an empty directory, 1741 * dentry_operations can't be changed later. As a result, if the filesystem has 1742 * casefolding support enabled at all, we have to give all dentries the 1743 * casefolding operations even if their inode doesn't have the casefolding flag 1744 * currently (and thus the casefolding ops would be no-ops for now). 1745 * 1746 * Encryption works differently in that the only dentry operation it needs is 1747 * d_revalidate, which it only needs on dentries that have the no-key name flag. 1748 * The no-key flag can't be set "later", so we don't have to worry about that. 1749 * 1750 * Finally, to maximize compatibility with overlayfs (which isn't compatible 1751 * with certain dentry operations) and to avoid taking an unnecessary 1752 * performance hit, we use custom dentry_operations for each possible 1753 * combination rather than always installing all operations. 1754 */ 1755 void generic_set_encrypted_ci_d_ops(struct dentry *dentry) 1756 { 1757 #ifdef CONFIG_FS_ENCRYPTION 1758 bool needs_encrypt_ops = dentry->d_flags & DCACHE_NOKEY_NAME; 1759 #endif 1760 #if IS_ENABLED(CONFIG_UNICODE) 1761 bool needs_ci_ops = dentry->d_sb->s_encoding; 1762 #endif 1763 #if defined(CONFIG_FS_ENCRYPTION) && IS_ENABLED(CONFIG_UNICODE) 1764 if (needs_encrypt_ops && needs_ci_ops) { 1765 d_set_d_op(dentry, &generic_encrypted_ci_dentry_ops); 1766 return; 1767 } 1768 #endif 1769 #ifdef CONFIG_FS_ENCRYPTION 1770 if (needs_encrypt_ops) { 1771 d_set_d_op(dentry, &generic_encrypted_dentry_ops); 1772 return; 1773 } 1774 #endif 1775 #if IS_ENABLED(CONFIG_UNICODE) 1776 if (needs_ci_ops) { 1777 d_set_d_op(dentry, &generic_ci_dentry_ops); 1778 return; 1779 } 1780 #endif 1781 } 1782 EXPORT_SYMBOL(generic_set_encrypted_ci_d_ops); 1783 1784 /** 1785 * inode_maybe_inc_iversion - increments i_version 1786 * @inode: inode with the i_version that should be updated 1787 * @force: increment the counter even if it's not necessary? 1788 * 1789 * Every time the inode is modified, the i_version field must be seen to have 1790 * changed by any observer. 1791 * 1792 * If "force" is set or the QUERIED flag is set, then ensure that we increment 1793 * the value, and clear the queried flag. 1794 * 1795 * In the common case where neither is set, then we can return "false" without 1796 * updating i_version. 1797 * 1798 * If this function returns false, and no other metadata has changed, then we 1799 * can avoid logging the metadata. 1800 */ 1801 bool inode_maybe_inc_iversion(struct inode *inode, bool force) 1802 { 1803 u64 cur, new; 1804 1805 /* 1806 * The i_version field is not strictly ordered with any other inode 1807 * information, but the legacy inode_inc_iversion code used a spinlock 1808 * to serialize increments. 1809 * 1810 * Here, we add full memory barriers to ensure that any de-facto 1811 * ordering with other info is preserved. 1812 * 1813 * This barrier pairs with the barrier in inode_query_iversion() 1814 */ 1815 smp_mb(); 1816 cur = inode_peek_iversion_raw(inode); 1817 do { 1818 /* If flag is clear then we needn't do anything */ 1819 if (!force && !(cur & I_VERSION_QUERIED)) 1820 return false; 1821 1822 /* Since lowest bit is flag, add 2 to avoid it */ 1823 new = (cur & ~I_VERSION_QUERIED) + I_VERSION_INCREMENT; 1824 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new)); 1825 return true; 1826 } 1827 EXPORT_SYMBOL(inode_maybe_inc_iversion); 1828 1829 /** 1830 * inode_query_iversion - read i_version for later use 1831 * @inode: inode from which i_version should be read 1832 * 1833 * Read the inode i_version counter. This should be used by callers that wish 1834 * to store the returned i_version for later comparison. This will guarantee 1835 * that a later query of the i_version will result in a different value if 1836 * anything has changed. 1837 * 1838 * In this implementation, we fetch the current value, set the QUERIED flag and 1839 * then try to swap it into place with a cmpxchg, if it wasn't already set. If 1840 * that fails, we try again with the newly fetched value from the cmpxchg. 1841 */ 1842 u64 inode_query_iversion(struct inode *inode) 1843 { 1844 u64 cur, new; 1845 1846 cur = inode_peek_iversion_raw(inode); 1847 do { 1848 /* If flag is already set, then no need to swap */ 1849 if (cur & I_VERSION_QUERIED) { 1850 /* 1851 * This barrier (and the implicit barrier in the 1852 * cmpxchg below) pairs with the barrier in 1853 * inode_maybe_inc_iversion(). 1854 */ 1855 smp_mb(); 1856 break; 1857 } 1858 1859 new = cur | I_VERSION_QUERIED; 1860 } while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new)); 1861 return cur >> I_VERSION_QUERIED_SHIFT; 1862 } 1863 EXPORT_SYMBOL(inode_query_iversion); 1864 1865 ssize_t direct_write_fallback(struct kiocb *iocb, struct iov_iter *iter, 1866 ssize_t direct_written, ssize_t buffered_written) 1867 { 1868 struct address_space *mapping = iocb->ki_filp->f_mapping; 1869 loff_t pos = iocb->ki_pos - buffered_written; 1870 loff_t end = iocb->ki_pos - 1; 1871 int err; 1872 1873 /* 1874 * If the buffered write fallback returned an error, we want to return 1875 * the number of bytes which were written by direct I/O, or the error 1876 * code if that was zero. 1877 * 1878 * Note that this differs from normal direct-io semantics, which will 1879 * return -EFOO even if some bytes were written. 1880 */ 1881 if (unlikely(buffered_written < 0)) { 1882 if (direct_written) 1883 return direct_written; 1884 return buffered_written; 1885 } 1886 1887 /* 1888 * We need to ensure that the page cache pages are written to disk and 1889 * invalidated to preserve the expected O_DIRECT semantics. 1890 */ 1891 err = filemap_write_and_wait_range(mapping, pos, end); 1892 if (err < 0) { 1893 /* 1894 * We don't know how much we wrote, so just return the number of 1895 * bytes which were direct-written 1896 */ 1897 if (direct_written) 1898 return direct_written; 1899 return err; 1900 } 1901 invalidate_mapping_pages(mapping, pos >> PAGE_SHIFT, end >> PAGE_SHIFT); 1902 return direct_written + buffered_written; 1903 } 1904 EXPORT_SYMBOL_GPL(direct_write_fallback); 1905