1// SPDX-License-Identifier: GPL-2.0 2/* 3 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 4 * All Rights Reserved. 5 */ 6#include "xfs.h" 7#include "xfs_fs.h" 8#include "xfs_shared.h" 9#include "xfs_format.h" 10#include "xfs_log_format.h" 11#include "xfs_trans_resv.h" 12#include "xfs_mount.h" 13#include "xfs_inode.h" 14#include "xfs_acl.h" 15#include "xfs_quota.h" 16#include "xfs_da_format.h" 17#include "xfs_da_btree.h" 18#include "xfs_attr.h" 19#include "xfs_trans.h" 20#include "xfs_trace.h" 21#include "xfs_icache.h" 22#include "xfs_symlink.h" 23#include "xfs_dir2.h" 24#include "xfs_iomap.h" 25#include "xfs_error.h" 26#include "xfs_ioctl.h" 27#include "xfs_xattr.h" 28 29#include <linux/posix_acl.h> 30#include <linux/security.h> 31#include <linux/iversion.h> 32#include <linux/fiemap.h> 33 34/* 35 * Directories have different lock order w.r.t. mmap_lock compared to regular 36 * files. This is due to readdir potentially triggering page faults on a user 37 * buffer inside filldir(), and this happens with the ilock on the directory 38 * held. For regular files, the lock order is the other way around - the 39 * mmap_lock is taken during the page fault, and then we lock the ilock to do 40 * block mapping. Hence we need a different class for the directory ilock so 41 * that lockdep can tell them apart. 42 */ 43static struct lock_class_key xfs_nondir_ilock_class; 44static struct lock_class_key xfs_dir_ilock_class; 45 46static int 47xfs_initxattrs( 48 struct inode *inode, 49 const struct xattr *xattr_array, 50 void *fs_info) 51{ 52 const struct xattr *xattr; 53 struct xfs_inode *ip = XFS_I(inode); 54 int error = 0; 55 56 for (xattr = xattr_array; xattr->name != NULL; xattr++) { 57 struct xfs_da_args args = { 58 .dp = ip, 59 .attr_filter = XFS_ATTR_SECURE, 60 .name = xattr->name, 61 .namelen = strlen(xattr->name), 62 .value = xattr->value, 63 .valuelen = xattr->value_len, 64 }; 65 error = xfs_attr_change(&args); 66 if (error < 0) 67 break; 68 } 69 return error; 70} 71 72/* 73 * Hook in SELinux. This is not quite correct yet, what we really need 74 * here (as we do for default ACLs) is a mechanism by which creation of 75 * these attrs can be journalled at inode creation time (along with the 76 * inode, of course, such that log replay can't cause these to be lost). 77 */ 78int 79xfs_inode_init_security( 80 struct inode *inode, 81 struct inode *dir, 82 const struct qstr *qstr) 83{ 84 return security_inode_init_security(inode, dir, qstr, 85 &xfs_initxattrs, NULL); 86} 87 88static void 89xfs_dentry_to_name( 90 struct xfs_name *namep, 91 struct dentry *dentry) 92{ 93 namep->name = dentry->d_name.name; 94 namep->len = dentry->d_name.len; 95 namep->type = XFS_DIR3_FT_UNKNOWN; 96} 97 98static int 99xfs_dentry_mode_to_name( 100 struct xfs_name *namep, 101 struct dentry *dentry, 102 int mode) 103{ 104 namep->name = dentry->d_name.name; 105 namep->len = dentry->d_name.len; 106 namep->type = xfs_mode_to_ftype(mode); 107 108 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN)) 109 return -EFSCORRUPTED; 110 111 return 0; 112} 113 114STATIC void 115xfs_cleanup_inode( 116 struct inode *dir, 117 struct inode *inode, 118 struct dentry *dentry) 119{ 120 struct xfs_name teardown; 121 122 /* Oh, the horror. 123 * If we can't add the ACL or we fail in 124 * xfs_inode_init_security we must back out. 125 * ENOSPC can hit here, among other things. 126 */ 127 xfs_dentry_to_name(&teardown, dentry); 128 129 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); 130} 131 132/* 133 * Check to see if we are likely to need an extended attribute to be added to 134 * the inode we are about to allocate. This allows the attribute fork to be 135 * created during the inode allocation, reducing the number of transactions we 136 * need to do in this fast path. 137 * 138 * The security checks are optimistic, but not guaranteed. The two LSMs that 139 * require xattrs to be added here (selinux and smack) are also the only two 140 * LSMs that add a sb->s_security structure to the superblock. Hence if security 141 * is enabled and sb->s_security is set, we have a pretty good idea that we are 142 * going to be asked to add a security xattr immediately after allocating the 143 * xfs inode and instantiating the VFS inode. 144 */ 145static inline bool 146xfs_create_need_xattr( 147 struct inode *dir, 148 struct posix_acl *default_acl, 149 struct posix_acl *acl) 150{ 151 if (acl) 152 return true; 153 if (default_acl) 154 return true; 155#if IS_ENABLED(CONFIG_SECURITY) 156 if (dir->i_sb->s_security) 157 return true; 158#endif 159 return false; 160} 161 162 163STATIC int 164xfs_generic_create( 165 struct mnt_idmap *idmap, 166 struct inode *dir, 167 struct dentry *dentry, 168 umode_t mode, 169 dev_t rdev, 170 struct file *tmpfile) /* unnamed file */ 171{ 172 struct inode *inode; 173 struct xfs_inode *ip = NULL; 174 struct posix_acl *default_acl, *acl; 175 struct xfs_name name; 176 int error; 177 178 /* 179 * Irix uses Missed'em'V split, but doesn't want to see 180 * the upper 5 bits of (14bit) major. 181 */ 182 if (S_ISCHR(mode) || S_ISBLK(mode)) { 183 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff)) 184 return -EINVAL; 185 } else { 186 rdev = 0; 187 } 188 189 error = posix_acl_create(dir, &mode, &default_acl, &acl); 190 if (error) 191 return error; 192 193 /* Verify mode is valid also for tmpfile case */ 194 error = xfs_dentry_mode_to_name(&name, dentry, mode); 195 if (unlikely(error)) 196 goto out_free_acl; 197 198 if (!tmpfile) { 199 error = xfs_create(idmap, XFS_I(dir), &name, mode, rdev, 200 xfs_create_need_xattr(dir, default_acl, acl), 201 &ip); 202 } else { 203 error = xfs_create_tmpfile(idmap, XFS_I(dir), mode, &ip); 204 } 205 if (unlikely(error)) 206 goto out_free_acl; 207 208 inode = VFS_I(ip); 209 210 error = xfs_inode_init_security(inode, dir, &dentry->d_name); 211 if (unlikely(error)) 212 goto out_cleanup_inode; 213 214 if (default_acl) { 215 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); 216 if (error) 217 goto out_cleanup_inode; 218 } 219 if (acl) { 220 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); 221 if (error) 222 goto out_cleanup_inode; 223 } 224 225 xfs_setup_iops(ip); 226 227 if (tmpfile) { 228 /* 229 * The VFS requires that any inode fed to d_tmpfile must have 230 * nlink == 1 so that it can decrement the nlink in d_tmpfile. 231 * However, we created the temp file with nlink == 0 because 232 * we're not allowed to put an inode with nlink > 0 on the 233 * unlinked list. Therefore we have to set nlink to 1 so that 234 * d_tmpfile can immediately set it back to zero. 235 */ 236 set_nlink(inode, 1); 237 d_tmpfile(tmpfile, inode); 238 } else 239 d_instantiate(dentry, inode); 240 241 xfs_finish_inode_setup(ip); 242 243 out_free_acl: 244 posix_acl_release(default_acl); 245 posix_acl_release(acl); 246 return error; 247 248 out_cleanup_inode: 249 xfs_finish_inode_setup(ip); 250 if (!tmpfile) 251 xfs_cleanup_inode(dir, inode, dentry); 252 xfs_irele(ip); 253 goto out_free_acl; 254} 255 256STATIC int 257xfs_vn_mknod( 258 struct mnt_idmap *idmap, 259 struct inode *dir, 260 struct dentry *dentry, 261 umode_t mode, 262 dev_t rdev) 263{ 264 return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL); 265} 266 267STATIC int 268xfs_vn_create( 269 struct mnt_idmap *idmap, 270 struct inode *dir, 271 struct dentry *dentry, 272 umode_t mode, 273 bool flags) 274{ 275 return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL); 276} 277 278STATIC int 279xfs_vn_mkdir( 280 struct mnt_idmap *idmap, 281 struct inode *dir, 282 struct dentry *dentry, 283 umode_t mode) 284{ 285 return xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL); 286} 287 288STATIC struct dentry * 289xfs_vn_lookup( 290 struct inode *dir, 291 struct dentry *dentry, 292 unsigned int flags) 293{ 294 struct inode *inode; 295 struct xfs_inode *cip; 296 struct xfs_name name; 297 int error; 298 299 if (dentry->d_name.len >= MAXNAMELEN) 300 return ERR_PTR(-ENAMETOOLONG); 301 302 xfs_dentry_to_name(&name, dentry); 303 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); 304 if (likely(!error)) 305 inode = VFS_I(cip); 306 else if (likely(error == -ENOENT)) 307 inode = NULL; 308 else 309 inode = ERR_PTR(error); 310 return d_splice_alias(inode, dentry); 311} 312 313STATIC struct dentry * 314xfs_vn_ci_lookup( 315 struct inode *dir, 316 struct dentry *dentry, 317 unsigned int flags) 318{ 319 struct xfs_inode *ip; 320 struct xfs_name xname; 321 struct xfs_name ci_name; 322 struct qstr dname; 323 int error; 324 325 if (dentry->d_name.len >= MAXNAMELEN) 326 return ERR_PTR(-ENAMETOOLONG); 327 328 xfs_dentry_to_name(&xname, dentry); 329 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); 330 if (unlikely(error)) { 331 if (unlikely(error != -ENOENT)) 332 return ERR_PTR(error); 333 /* 334 * call d_add(dentry, NULL) here when d_drop_negative_children 335 * is called in xfs_vn_mknod (ie. allow negative dentries 336 * with CI filesystems). 337 */ 338 return NULL; 339 } 340 341 /* if exact match, just splice and exit */ 342 if (!ci_name.name) 343 return d_splice_alias(VFS_I(ip), dentry); 344 345 /* else case-insensitive match... */ 346 dname.name = ci_name.name; 347 dname.len = ci_name.len; 348 dentry = d_add_ci(dentry, VFS_I(ip), &dname); 349 kfree(ci_name.name); 350 return dentry; 351} 352 353STATIC int 354xfs_vn_link( 355 struct dentry *old_dentry, 356 struct inode *dir, 357 struct dentry *dentry) 358{ 359 struct inode *inode = d_inode(old_dentry); 360 struct xfs_name name; 361 int error; 362 363 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode); 364 if (unlikely(error)) 365 return error; 366 367 error = xfs_link(XFS_I(dir), XFS_I(inode), &name); 368 if (unlikely(error)) 369 return error; 370 371 ihold(inode); 372 d_instantiate(dentry, inode); 373 return 0; 374} 375 376STATIC int 377xfs_vn_unlink( 378 struct inode *dir, 379 struct dentry *dentry) 380{ 381 struct xfs_name name; 382 int error; 383 384 xfs_dentry_to_name(&name, dentry); 385 386 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry))); 387 if (error) 388 return error; 389 390 /* 391 * With unlink, the VFS makes the dentry "negative": no inode, 392 * but still hashed. This is incompatible with case-insensitive 393 * mode, so invalidate (unhash) the dentry in CI-mode. 394 */ 395 if (xfs_has_asciici(XFS_M(dir->i_sb))) 396 d_invalidate(dentry); 397 return 0; 398} 399 400STATIC int 401xfs_vn_symlink( 402 struct mnt_idmap *idmap, 403 struct inode *dir, 404 struct dentry *dentry, 405 const char *symname) 406{ 407 struct inode *inode; 408 struct xfs_inode *cip = NULL; 409 struct xfs_name name; 410 int error; 411 umode_t mode; 412 413 mode = S_IFLNK | 414 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); 415 error = xfs_dentry_mode_to_name(&name, dentry, mode); 416 if (unlikely(error)) 417 goto out; 418 419 error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip); 420 if (unlikely(error)) 421 goto out; 422 423 inode = VFS_I(cip); 424 425 error = xfs_inode_init_security(inode, dir, &dentry->d_name); 426 if (unlikely(error)) 427 goto out_cleanup_inode; 428 429 xfs_setup_iops(cip); 430 431 d_instantiate(dentry, inode); 432 xfs_finish_inode_setup(cip); 433 return 0; 434 435 out_cleanup_inode: 436 xfs_finish_inode_setup(cip); 437 xfs_cleanup_inode(dir, inode, dentry); 438 xfs_irele(cip); 439 out: 440 return error; 441} 442 443STATIC int 444xfs_vn_rename( 445 struct mnt_idmap *idmap, 446 struct inode *odir, 447 struct dentry *odentry, 448 struct inode *ndir, 449 struct dentry *ndentry, 450 unsigned int flags) 451{ 452 struct inode *new_inode = d_inode(ndentry); 453 int omode = 0; 454 int error; 455 struct xfs_name oname; 456 struct xfs_name nname; 457 458 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 459 return -EINVAL; 460 461 /* if we are exchanging files, we need to set i_mode of both files */ 462 if (flags & RENAME_EXCHANGE) 463 omode = d_inode(ndentry)->i_mode; 464 465 error = xfs_dentry_mode_to_name(&oname, odentry, omode); 466 if (omode && unlikely(error)) 467 return error; 468 469 error = xfs_dentry_mode_to_name(&nname, ndentry, 470 d_inode(odentry)->i_mode); 471 if (unlikely(error)) 472 return error; 473 474 return xfs_rename(idmap, XFS_I(odir), &oname, 475 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname, 476 new_inode ? XFS_I(new_inode) : NULL, flags); 477} 478 479/* 480 * careful here - this function can get called recursively, so 481 * we need to be very careful about how much stack we use. 482 * uio is kmalloced for this reason... 483 */ 484STATIC const char * 485xfs_vn_get_link( 486 struct dentry *dentry, 487 struct inode *inode, 488 struct delayed_call *done) 489{ 490 char *link; 491 int error = -ENOMEM; 492 493 if (!dentry) 494 return ERR_PTR(-ECHILD); 495 496 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); 497 if (!link) 498 goto out_err; 499 500 error = xfs_readlink(XFS_I(d_inode(dentry)), link); 501 if (unlikely(error)) 502 goto out_kfree; 503 504 set_delayed_call(done, kfree_link, link); 505 return link; 506 507 out_kfree: 508 kfree(link); 509 out_err: 510 return ERR_PTR(error); 511} 512 513static uint32_t 514xfs_stat_blksize( 515 struct xfs_inode *ip) 516{ 517 struct xfs_mount *mp = ip->i_mount; 518 519 /* 520 * If the file blocks are being allocated from a realtime volume, then 521 * always return the realtime extent size. 522 */ 523 if (XFS_IS_REALTIME_INODE(ip)) 524 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip)); 525 526 /* 527 * Allow large block sizes to be reported to userspace programs if the 528 * "largeio" mount option is used. 529 * 530 * If compatibility mode is specified, simply return the basic unit of 531 * caching so that we don't get inefficient read/modify/write I/O from 532 * user apps. Otherwise.... 533 * 534 * If the underlying volume is a stripe, then return the stripe width in 535 * bytes as the recommended I/O size. It is not a stripe and we've set a 536 * default buffered I/O size, return that, otherwise return the compat 537 * default. 538 */ 539 if (xfs_has_large_iosize(mp)) { 540 if (mp->m_swidth) 541 return XFS_FSB_TO_B(mp, mp->m_swidth); 542 if (xfs_has_allocsize(mp)) 543 return 1U << mp->m_allocsize_log; 544 } 545 546 return PAGE_SIZE; 547} 548 549STATIC int 550xfs_vn_getattr( 551 struct mnt_idmap *idmap, 552 const struct path *path, 553 struct kstat *stat, 554 u32 request_mask, 555 unsigned int query_flags) 556{ 557 struct inode *inode = d_inode(path->dentry); 558 struct xfs_inode *ip = XFS_I(inode); 559 struct xfs_mount *mp = ip->i_mount; 560 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode); 561 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode); 562 563 trace_xfs_getattr(ip); 564 565 if (xfs_is_shutdown(mp)) 566 return -EIO; 567 568 stat->size = XFS_ISIZE(ip); 569 stat->dev = inode->i_sb->s_dev; 570 stat->mode = inode->i_mode; 571 stat->nlink = inode->i_nlink; 572 stat->uid = vfsuid_into_kuid(vfsuid); 573 stat->gid = vfsgid_into_kgid(vfsgid); 574 stat->ino = ip->i_ino; 575 stat->atime = inode_get_atime(inode); 576 stat->mtime = inode_get_mtime(inode); 577 stat->ctime = inode_get_ctime(inode); 578 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); 579 580 if (xfs_has_v3inodes(mp)) { 581 if (request_mask & STATX_BTIME) { 582 stat->result_mask |= STATX_BTIME; 583 stat->btime = ip->i_crtime; 584 } 585 } 586 587 if ((request_mask & STATX_CHANGE_COOKIE) && IS_I_VERSION(inode)) { 588 stat->change_cookie = inode_query_iversion(inode); 589 stat->result_mask |= STATX_CHANGE_COOKIE; 590 } 591 592 /* 593 * Note: If you add another clause to set an attribute flag, please 594 * update attributes_mask below. 595 */ 596 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) 597 stat->attributes |= STATX_ATTR_IMMUTABLE; 598 if (ip->i_diflags & XFS_DIFLAG_APPEND) 599 stat->attributes |= STATX_ATTR_APPEND; 600 if (ip->i_diflags & XFS_DIFLAG_NODUMP) 601 stat->attributes |= STATX_ATTR_NODUMP; 602 603 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | 604 STATX_ATTR_APPEND | 605 STATX_ATTR_NODUMP); 606 607 switch (inode->i_mode & S_IFMT) { 608 case S_IFBLK: 609 case S_IFCHR: 610 stat->blksize = BLKDEV_IOSIZE; 611 stat->rdev = inode->i_rdev; 612 break; 613 case S_IFREG: 614 if (request_mask & STATX_DIOALIGN) { 615 struct xfs_buftarg *target = xfs_inode_buftarg(ip); 616 struct block_device *bdev = target->bt_bdev; 617 618 stat->result_mask |= STATX_DIOALIGN; 619 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; 620 stat->dio_offset_align = bdev_logical_block_size(bdev); 621 } 622 fallthrough; 623 default: 624 stat->blksize = xfs_stat_blksize(ip); 625 stat->rdev = 0; 626 break; 627 } 628 629 return 0; 630} 631 632static int 633xfs_vn_change_ok( 634 struct mnt_idmap *idmap, 635 struct dentry *dentry, 636 struct iattr *iattr) 637{ 638 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; 639 640 if (xfs_is_readonly(mp)) 641 return -EROFS; 642 643 if (xfs_is_shutdown(mp)) 644 return -EIO; 645 646 return setattr_prepare(idmap, dentry, iattr); 647} 648 649/* 650 * Set non-size attributes of an inode. 651 * 652 * Caution: The caller of this function is responsible for calling 653 * setattr_prepare() or otherwise verifying the change is fine. 654 */ 655static int 656xfs_setattr_nonsize( 657 struct mnt_idmap *idmap, 658 struct dentry *dentry, 659 struct xfs_inode *ip, 660 struct iattr *iattr) 661{ 662 xfs_mount_t *mp = ip->i_mount; 663 struct inode *inode = VFS_I(ip); 664 int mask = iattr->ia_valid; 665 xfs_trans_t *tp; 666 int error; 667 kuid_t uid = GLOBAL_ROOT_UID; 668 kgid_t gid = GLOBAL_ROOT_GID; 669 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 670 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL; 671 672 ASSERT((mask & ATTR_SIZE) == 0); 673 674 /* 675 * If disk quotas is on, we make sure that the dquots do exist on disk, 676 * before we start any other transactions. Trying to do this later 677 * is messy. We don't care to take a readlock to look at the ids 678 * in inode here, because we can't hold it across the trans_reserve. 679 * If the IDs do change before we take the ilock, we're covered 680 * because the i_*dquot fields will get updated anyway. 681 */ 682 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 683 uint qflags = 0; 684 685 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 686 uid = from_vfsuid(idmap, i_user_ns(inode), 687 iattr->ia_vfsuid); 688 qflags |= XFS_QMOPT_UQUOTA; 689 } else { 690 uid = inode->i_uid; 691 } 692 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 693 gid = from_vfsgid(idmap, i_user_ns(inode), 694 iattr->ia_vfsgid); 695 qflags |= XFS_QMOPT_GQUOTA; 696 } else { 697 gid = inode->i_gid; 698 } 699 700 /* 701 * We take a reference when we initialize udqp and gdqp, 702 * so it is important that we never blindly double trip on 703 * the same variable. See xfs_create() for an example. 704 */ 705 ASSERT(udqp == NULL); 706 ASSERT(gdqp == NULL); 707 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, 708 qflags, &udqp, &gdqp, NULL); 709 if (error) 710 return error; 711 } 712 713 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, 714 has_capability_noaudit(current, CAP_FOWNER), &tp); 715 if (error) 716 goto out_dqrele; 717 718 /* 719 * Register quota modifications in the transaction. Must be the owner 720 * or privileged. These IDs could have changed since we last looked at 721 * them. But, we're assured that if the ownership did change while we 722 * didn't have the inode locked, inode's dquot(s) would have changed 723 * also. 724 */ 725 if (XFS_IS_UQUOTA_ON(mp) && 726 i_uid_needs_update(idmap, iattr, inode)) { 727 ASSERT(udqp); 728 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp); 729 } 730 if (XFS_IS_GQUOTA_ON(mp) && 731 i_gid_needs_update(idmap, iattr, inode)) { 732 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp)); 733 ASSERT(gdqp); 734 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp); 735 } 736 737 setattr_copy(idmap, inode, iattr); 738 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 739 740 XFS_STATS_INC(mp, xs_ig_attrchg); 741 742 if (xfs_has_wsync(mp)) 743 xfs_trans_set_sync(tp); 744 error = xfs_trans_commit(tp); 745 746 /* 747 * Release any dquot(s) the inode had kept before chown. 748 */ 749 xfs_qm_dqrele(old_udqp); 750 xfs_qm_dqrele(old_gdqp); 751 xfs_qm_dqrele(udqp); 752 xfs_qm_dqrele(gdqp); 753 754 if (error) 755 return error; 756 757 /* 758 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 759 * update. We could avoid this with linked transactions 760 * and passing down the transaction pointer all the way 761 * to attr_set. No previous user of the generic 762 * Posix ACL code seems to care about this issue either. 763 */ 764 if (mask & ATTR_MODE) { 765 error = posix_acl_chmod(idmap, dentry, inode->i_mode); 766 if (error) 767 return error; 768 } 769 770 return 0; 771 772out_dqrele: 773 xfs_qm_dqrele(udqp); 774 xfs_qm_dqrele(gdqp); 775 return error; 776} 777 778/* 779 * Truncate file. Must have write permission and not be a directory. 780 * 781 * Caution: The caller of this function is responsible for calling 782 * setattr_prepare() or otherwise verifying the change is fine. 783 */ 784STATIC int 785xfs_setattr_size( 786 struct mnt_idmap *idmap, 787 struct dentry *dentry, 788 struct xfs_inode *ip, 789 struct iattr *iattr) 790{ 791 struct xfs_mount *mp = ip->i_mount; 792 struct inode *inode = VFS_I(ip); 793 xfs_off_t oldsize, newsize; 794 struct xfs_trans *tp; 795 int error; 796 uint lock_flags = 0; 797 bool did_zeroing = false; 798 799 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL); 800 ASSERT(S_ISREG(inode->i_mode)); 801 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 802 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); 803 804 oldsize = inode->i_size; 805 newsize = iattr->ia_size; 806 807 /* 808 * Short circuit the truncate case for zero length files. 809 */ 810 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { 811 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 812 return 0; 813 814 /* 815 * Use the regular setattr path to update the timestamps. 816 */ 817 iattr->ia_valid &= ~ATTR_SIZE; 818 return xfs_setattr_nonsize(idmap, dentry, ip, iattr); 819 } 820 821 /* 822 * Make sure that the dquots are attached to the inode. 823 */ 824 error = xfs_qm_dqattach(ip); 825 if (error) 826 return error; 827 828 /* 829 * Wait for all direct I/O to complete. 830 */ 831 inode_dio_wait(inode); 832 833 /* 834 * File data changes must be complete before we start the transaction to 835 * modify the inode. This needs to be done before joining the inode to 836 * the transaction because the inode cannot be unlocked once it is a 837 * part of the transaction. 838 * 839 * Start with zeroing any data beyond EOF that we may expose on file 840 * extension, or zeroing out the rest of the block on a downward 841 * truncate. 842 */ 843 if (newsize > oldsize) { 844 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); 845 error = xfs_zero_range(ip, oldsize, newsize - oldsize, 846 &did_zeroing); 847 } else { 848 /* 849 * iomap won't detect a dirty page over an unwritten block (or a 850 * cow block over a hole) and subsequently skips zeroing the 851 * newly post-EOF portion of the page. Flush the new EOF to 852 * convert the block before the pagecache truncate. 853 */ 854 error = filemap_write_and_wait_range(inode->i_mapping, newsize, 855 newsize); 856 if (error) 857 return error; 858 error = xfs_truncate_page(ip, newsize, &did_zeroing); 859 } 860 861 if (error) 862 return error; 863 864 /* 865 * We've already locked out new page faults, so now we can safely remove 866 * pages from the page cache knowing they won't get refaulted until we 867 * drop the XFS_MMAP_EXCL lock after the extent manipulations are 868 * complete. The truncate_setsize() call also cleans partial EOF page 869 * PTEs on extending truncates and hence ensures sub-page block size 870 * filesystems are correctly handled, too. 871 * 872 * We have to do all the page cache truncate work outside the 873 * transaction context as the "lock" order is page lock->log space 874 * reservation as defined by extent allocation in the writeback path. 875 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but 876 * having already truncated the in-memory version of the file (i.e. made 877 * user visible changes). There's not much we can do about this, except 878 * to hope that the caller sees ENOMEM and retries the truncate 879 * operation. 880 * 881 * And we update in-core i_size and truncate page cache beyond newsize 882 * before writeback the [i_disk_size, newsize] range, so we're 883 * guaranteed not to write stale data past the new EOF on truncate down. 884 */ 885 truncate_setsize(inode, newsize); 886 887 /* 888 * We are going to log the inode size change in this transaction so 889 * any previous writes that are beyond the on disk EOF and the new 890 * EOF that have not been written out need to be written here. If we 891 * do not write the data out, we expose ourselves to the null files 892 * problem. Note that this includes any block zeroing we did above; 893 * otherwise those blocks may not be zeroed after a crash. 894 */ 895 if (did_zeroing || 896 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { 897 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 898 ip->i_disk_size, newsize - 1); 899 if (error) 900 return error; 901 } 902 903 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 904 if (error) 905 return error; 906 907 lock_flags |= XFS_ILOCK_EXCL; 908 xfs_ilock(ip, XFS_ILOCK_EXCL); 909 xfs_trans_ijoin(tp, ip, 0); 910 911 /* 912 * Only change the c/mtime if we are changing the size or we are 913 * explicitly asked to change it. This handles the semantic difference 914 * between truncate() and ftruncate() as implemented in the VFS. 915 * 916 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 917 * special case where we need to update the times despite not having 918 * these flags set. For all other operations the VFS set these flags 919 * explicitly if it wants a timestamp update. 920 */ 921 if (newsize != oldsize && 922 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 923 iattr->ia_ctime = iattr->ia_mtime = 924 current_time(inode); 925 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 926 } 927 928 /* 929 * The first thing we do is set the size to new_size permanently on 930 * disk. This way we don't have to worry about anyone ever being able 931 * to look at the data being freed even in the face of a crash. 932 * What we're getting around here is the case where we free a block, it 933 * is allocated to another file, it is written to, and then we crash. 934 * If the new data gets written to the file but the log buffers 935 * containing the free and reallocation don't, then we'd end up with 936 * garbage in the blocks being freed. As long as we make the new size 937 * permanent before actually freeing any blocks it doesn't matter if 938 * they get written to. 939 */ 940 ip->i_disk_size = newsize; 941 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 942 943 if (newsize <= oldsize) { 944 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 945 if (error) 946 goto out_trans_cancel; 947 948 /* 949 * Truncated "down", so we're removing references to old data 950 * here - if we delay flushing for a long time, we expose 951 * ourselves unduly to the notorious NULL files problem. So, 952 * we mark this inode and flush it when the file is closed, 953 * and do not wait the usual (long) time for writeout. 954 */ 955 xfs_iflags_set(ip, XFS_ITRUNCATED); 956 957 /* A truncate down always removes post-EOF blocks. */ 958 xfs_inode_clear_eofblocks_tag(ip); 959 } 960 961 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID))); 962 setattr_copy(idmap, inode, iattr); 963 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 964 965 XFS_STATS_INC(mp, xs_ig_attrchg); 966 967 if (xfs_has_wsync(mp)) 968 xfs_trans_set_sync(tp); 969 970 error = xfs_trans_commit(tp); 971out_unlock: 972 if (lock_flags) 973 xfs_iunlock(ip, lock_flags); 974 return error; 975 976out_trans_cancel: 977 xfs_trans_cancel(tp); 978 goto out_unlock; 979} 980 981int 982xfs_vn_setattr_size( 983 struct mnt_idmap *idmap, 984 struct dentry *dentry, 985 struct iattr *iattr) 986{ 987 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 988 int error; 989 990 trace_xfs_setattr(ip); 991 992 error = xfs_vn_change_ok(idmap, dentry, iattr); 993 if (error) 994 return error; 995 return xfs_setattr_size(idmap, dentry, ip, iattr); 996} 997 998STATIC int 999xfs_vn_setattr( 1000 struct mnt_idmap *idmap, 1001 struct dentry *dentry, 1002 struct iattr *iattr) 1003{ 1004 struct inode *inode = d_inode(dentry); 1005 struct xfs_inode *ip = XFS_I(inode); 1006 int error; 1007 1008 if (iattr->ia_valid & ATTR_SIZE) { 1009 uint iolock; 1010 1011 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 1012 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; 1013 1014 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); 1015 if (error) { 1016 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1017 return error; 1018 } 1019 1020 error = xfs_vn_setattr_size(idmap, dentry, iattr); 1021 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1022 } else { 1023 trace_xfs_setattr(ip); 1024 1025 error = xfs_vn_change_ok(idmap, dentry, iattr); 1026 if (!error) 1027 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr); 1028 } 1029 1030 return error; 1031} 1032 1033STATIC int 1034xfs_vn_update_time( 1035 struct inode *inode, 1036 int flags) 1037{ 1038 struct xfs_inode *ip = XFS_I(inode); 1039 struct xfs_mount *mp = ip->i_mount; 1040 int log_flags = XFS_ILOG_TIMESTAMP; 1041 struct xfs_trans *tp; 1042 int error; 1043 struct timespec64 now; 1044 1045 trace_xfs_update_time(ip); 1046 1047 if (inode->i_sb->s_flags & SB_LAZYTIME) { 1048 if (!((flags & S_VERSION) && 1049 inode_maybe_inc_iversion(inode, false))) { 1050 generic_update_time(inode, flags); 1051 return 0; 1052 } 1053 1054 /* Capture the iversion update that just occurred */ 1055 log_flags |= XFS_ILOG_CORE; 1056 } 1057 1058 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); 1059 if (error) 1060 return error; 1061 1062 xfs_ilock(ip, XFS_ILOCK_EXCL); 1063 if (flags & (S_CTIME|S_MTIME)) 1064 now = inode_set_ctime_current(inode); 1065 else 1066 now = current_time(inode); 1067 1068 if (flags & S_MTIME) 1069 inode_set_mtime_to_ts(inode, now); 1070 if (flags & S_ATIME) 1071 inode_set_atime_to_ts(inode, now); 1072 1073 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1074 xfs_trans_log_inode(tp, ip, log_flags); 1075 return xfs_trans_commit(tp); 1076} 1077 1078STATIC int 1079xfs_vn_fiemap( 1080 struct inode *inode, 1081 struct fiemap_extent_info *fieinfo, 1082 u64 start, 1083 u64 length) 1084{ 1085 int error; 1086 1087 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); 1088 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1089 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; 1090 error = iomap_fiemap(inode, fieinfo, start, length, 1091 &xfs_xattr_iomap_ops); 1092 } else { 1093 error = iomap_fiemap(inode, fieinfo, start, length, 1094 &xfs_read_iomap_ops); 1095 } 1096 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); 1097 1098 return error; 1099} 1100 1101STATIC int 1102xfs_vn_tmpfile( 1103 struct mnt_idmap *idmap, 1104 struct inode *dir, 1105 struct file *file, 1106 umode_t mode) 1107{ 1108 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file); 1109 1110 return finish_open_simple(file, err); 1111} 1112 1113static const struct inode_operations xfs_inode_operations = { 1114 .get_inode_acl = xfs_get_acl, 1115 .set_acl = xfs_set_acl, 1116 .getattr = xfs_vn_getattr, 1117 .setattr = xfs_vn_setattr, 1118 .listxattr = xfs_vn_listxattr, 1119 .fiemap = xfs_vn_fiemap, 1120 .update_time = xfs_vn_update_time, 1121 .fileattr_get = xfs_fileattr_get, 1122 .fileattr_set = xfs_fileattr_set, 1123}; 1124 1125static const struct inode_operations xfs_dir_inode_operations = { 1126 .create = xfs_vn_create, 1127 .lookup = xfs_vn_lookup, 1128 .link = xfs_vn_link, 1129 .unlink = xfs_vn_unlink, 1130 .symlink = xfs_vn_symlink, 1131 .mkdir = xfs_vn_mkdir, 1132 /* 1133 * Yes, XFS uses the same method for rmdir and unlink. 1134 * 1135 * There are some subtile differences deeper in the code, 1136 * but we use S_ISDIR to check for those. 1137 */ 1138 .rmdir = xfs_vn_unlink, 1139 .mknod = xfs_vn_mknod, 1140 .rename = xfs_vn_rename, 1141 .get_inode_acl = xfs_get_acl, 1142 .set_acl = xfs_set_acl, 1143 .getattr = xfs_vn_getattr, 1144 .setattr = xfs_vn_setattr, 1145 .listxattr = xfs_vn_listxattr, 1146 .update_time = xfs_vn_update_time, 1147 .tmpfile = xfs_vn_tmpfile, 1148 .fileattr_get = xfs_fileattr_get, 1149 .fileattr_set = xfs_fileattr_set, 1150}; 1151 1152static const struct inode_operations xfs_dir_ci_inode_operations = { 1153 .create = xfs_vn_create, 1154 .lookup = xfs_vn_ci_lookup, 1155 .link = xfs_vn_link, 1156 .unlink = xfs_vn_unlink, 1157 .symlink = xfs_vn_symlink, 1158 .mkdir = xfs_vn_mkdir, 1159 /* 1160 * Yes, XFS uses the same method for rmdir and unlink. 1161 * 1162 * There are some subtile differences deeper in the code, 1163 * but we use S_ISDIR to check for those. 1164 */ 1165 .rmdir = xfs_vn_unlink, 1166 .mknod = xfs_vn_mknod, 1167 .rename = xfs_vn_rename, 1168 .get_inode_acl = xfs_get_acl, 1169 .set_acl = xfs_set_acl, 1170 .getattr = xfs_vn_getattr, 1171 .setattr = xfs_vn_setattr, 1172 .listxattr = xfs_vn_listxattr, 1173 .update_time = xfs_vn_update_time, 1174 .tmpfile = xfs_vn_tmpfile, 1175 .fileattr_get = xfs_fileattr_get, 1176 .fileattr_set = xfs_fileattr_set, 1177}; 1178 1179static const struct inode_operations xfs_symlink_inode_operations = { 1180 .get_link = xfs_vn_get_link, 1181 .getattr = xfs_vn_getattr, 1182 .setattr = xfs_vn_setattr, 1183 .listxattr = xfs_vn_listxattr, 1184 .update_time = xfs_vn_update_time, 1185}; 1186 1187/* Figure out if this file actually supports DAX. */ 1188static bool 1189xfs_inode_supports_dax( 1190 struct xfs_inode *ip) 1191{ 1192 struct xfs_mount *mp = ip->i_mount; 1193 1194 /* Only supported on regular files. */ 1195 if (!S_ISREG(VFS_I(ip)->i_mode)) 1196 return false; 1197 1198 /* Block size must match page size */ 1199 if (mp->m_sb.sb_blocksize != PAGE_SIZE) 1200 return false; 1201 1202 /* Device has to support DAX too. */ 1203 return xfs_inode_buftarg(ip)->bt_daxdev != NULL; 1204} 1205 1206static bool 1207xfs_inode_should_enable_dax( 1208 struct xfs_inode *ip) 1209{ 1210 if (!IS_ENABLED(CONFIG_FS_DAX)) 1211 return false; 1212 if (xfs_has_dax_never(ip->i_mount)) 1213 return false; 1214 if (!xfs_inode_supports_dax(ip)) 1215 return false; 1216 if (xfs_has_dax_always(ip->i_mount)) 1217 return true; 1218 if (ip->i_diflags2 & XFS_DIFLAG2_DAX) 1219 return true; 1220 return false; 1221} 1222 1223void 1224xfs_diflags_to_iflags( 1225 struct xfs_inode *ip, 1226 bool init) 1227{ 1228 struct inode *inode = VFS_I(ip); 1229 unsigned int xflags = xfs_ip2xflags(ip); 1230 unsigned int flags = 0; 1231 1232 ASSERT(!(IS_DAX(inode) && init)); 1233 1234 if (xflags & FS_XFLAG_IMMUTABLE) 1235 flags |= S_IMMUTABLE; 1236 if (xflags & FS_XFLAG_APPEND) 1237 flags |= S_APPEND; 1238 if (xflags & FS_XFLAG_SYNC) 1239 flags |= S_SYNC; 1240 if (xflags & FS_XFLAG_NOATIME) 1241 flags |= S_NOATIME; 1242 if (init && xfs_inode_should_enable_dax(ip)) 1243 flags |= S_DAX; 1244 1245 /* 1246 * S_DAX can only be set during inode initialization and is never set by 1247 * the VFS, so we cannot mask off S_DAX in i_flags. 1248 */ 1249 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME); 1250 inode->i_flags |= flags; 1251} 1252 1253/* 1254 * Initialize the Linux inode. 1255 * 1256 * When reading existing inodes from disk this is called directly from xfs_iget, 1257 * when creating a new inode it is called from xfs_init_new_inode after setting 1258 * up the inode. These callers have different criteria for clearing XFS_INEW, so 1259 * leave it up to the caller to deal with unlocking the inode appropriately. 1260 */ 1261void 1262xfs_setup_inode( 1263 struct xfs_inode *ip) 1264{ 1265 struct inode *inode = &ip->i_vnode; 1266 gfp_t gfp_mask; 1267 1268 inode->i_ino = ip->i_ino; 1269 inode->i_state |= I_NEW; 1270 1271 inode_sb_list_add(inode); 1272 /* make the inode look hashed for the writeback code */ 1273 inode_fake_hash(inode); 1274 1275 i_size_write(inode, ip->i_disk_size); 1276 xfs_diflags_to_iflags(ip, true); 1277 1278 if (S_ISDIR(inode->i_mode)) { 1279 /* 1280 * We set the i_rwsem class here to avoid potential races with 1281 * lockdep_annotate_inode_mutex_key() reinitialising the lock 1282 * after a filehandle lookup has already found the inode in 1283 * cache before it has been unlocked via unlock_new_inode(). 1284 */ 1285 lockdep_set_class(&inode->i_rwsem, 1286 &inode->i_sb->s_type->i_mutex_dir_key); 1287 lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class); 1288 } else { 1289 lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class); 1290 } 1291 1292 /* 1293 * Ensure all page cache allocations are done from GFP_NOFS context to 1294 * prevent direct reclaim recursion back into the filesystem and blowing 1295 * stacks or deadlocking. 1296 */ 1297 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1298 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1299 1300 /* 1301 * For real-time inodes update the stable write flags to that of the RT 1302 * device instead of the data device. 1303 */ 1304 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip)) 1305 xfs_update_stable_writes(ip); 1306 1307 /* 1308 * If there is no attribute fork no ACL can exist on this inode, 1309 * and it can't have any file capabilities attached to it either. 1310 */ 1311 if (!xfs_inode_has_attr_fork(ip)) { 1312 inode_has_no_xattr(inode); 1313 cache_no_acl(inode); 1314 } 1315} 1316 1317void 1318xfs_setup_iops( 1319 struct xfs_inode *ip) 1320{ 1321 struct inode *inode = &ip->i_vnode; 1322 1323 switch (inode->i_mode & S_IFMT) { 1324 case S_IFREG: 1325 inode->i_op = &xfs_inode_operations; 1326 inode->i_fop = &xfs_file_operations; 1327 if (IS_DAX(inode)) 1328 inode->i_mapping->a_ops = &xfs_dax_aops; 1329 else 1330 inode->i_mapping->a_ops = &xfs_address_space_operations; 1331 break; 1332 case S_IFDIR: 1333 if (xfs_has_asciici(XFS_M(inode->i_sb))) 1334 inode->i_op = &xfs_dir_ci_inode_operations; 1335 else 1336 inode->i_op = &xfs_dir_inode_operations; 1337 inode->i_fop = &xfs_dir_file_operations; 1338 break; 1339 case S_IFLNK: 1340 inode->i_op = &xfs_symlink_inode_operations; 1341 break; 1342 default: 1343 inode->i_op = &xfs_inode_operations; 1344 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1345 break; 1346 } 1347} 1348