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