fs/attr.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/attr.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  *  changes by Thomas Schoebel-Theuer
   7  */
   8
   9 #include <linux/export.h>
  10 #include <linux/time.h>
  11 #include <linux/mm.h>
  12 #include <linux/string.h>
  13 #include <linux/sched/signal.h>
  14 #include <linux/capability.h>
  15 #include <linux/fsnotify.h>
  16 #include <linux/fcntl.h>
  17 #include <linux/filelock.h>
  18 #include <linux/security.h>
  19
  20 /**
  21  * setattr_should_drop_sgid - determine whether the setgid bit needs to be
  22  *                            removed
  23  * @idmap:      idmap of the mount @inode was found from
  24  * @inode:      inode to check
  25  *
  26  * This function determines whether the setgid bit needs to be removed.
  27  * We retain backwards compatibility and require setgid bit to be removed
  28  * unconditionally if S_IXGRP is set. Otherwise we have the exact same
  29  * requirements as setattr_prepare() and setattr_copy().
  30  *
  31  * Return: ATTR_KILL_SGID if setgid bit needs to be removed, 0 otherwise.
  32  */
  33 int setattr_should_drop_sgid(struct mnt_idmap *idmap,
  34                              const struct inode *inode)
  35 {
  36         umode_t mode = inode->i_mode;
  37
  38         if (!(mode & S_ISGID))
  39                 return 0;
  40         if (mode & S_IXGRP)
  41                 return ATTR_KILL_SGID;
  42         if (!in_group_or_capable(idmap, inode, i_gid_into_vfsgid(idmap, inode)))
  43                 return ATTR_KILL_SGID;
  44         return 0;
  45 }
  46 EXPORT_SYMBOL(setattr_should_drop_sgid);
  47
  48 /**
  49  * setattr_should_drop_suidgid - determine whether the set{g,u}id bit needs to
  50  *                               be dropped
  51  * @idmap:      idmap of the mount @inode was found from
  52  * @inode:      inode to check
  53  *
  54  * This function determines whether the set{g,u}id bits need to be removed.
  55  * If the setuid bit needs to be removed ATTR_KILL_SUID is returned. If the
  56  * setgid bit needs to be removed ATTR_KILL_SGID is returned. If both
  57  * set{g,u}id bits need to be removed the corresponding mask of both flags is
  58  * returned.
  59  *
  60  * Return: A mask of ATTR_KILL_S{G,U}ID indicating which - if any - setid bits
  61  * to remove, 0 otherwise.
  62  */
  63 int setattr_should_drop_suidgid(struct mnt_idmap *idmap,
  64                                 struct inode *inode)
  65 {
  66         umode_t mode = inode->i_mode;
  67         int kill = 0;
  68
  69         /* suid always must be killed */
  70         if (unlikely(mode & S_ISUID))
  71                 kill = ATTR_KILL_SUID;
  72
  73         kill |= setattr_should_drop_sgid(idmap, inode);
  74
  75         if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
  76                 return kill;
  77
  78         return 0;
  79 }
  80 EXPORT_SYMBOL(setattr_should_drop_suidgid);
  81
  82 /**
  83  * chown_ok - verify permissions to chown inode
  84  * @idmap:      idmap of the mount @inode was found from
  85  * @inode:      inode to check permissions on
  86  * @ia_vfsuid:  uid to chown @inode to
  87  *
  88  * If the inode has been found through an idmapped mount the idmap of
  89  * the vfsmount must be passed through @idmap. This function will then
  90  * take care to map the inode according to @idmap before checking
  91  * permissions. On non-idmapped mounts or if permission checking is to be
  92  * performed on the raw inode simply pass @nop_mnt_idmap.
  93  */
  94 static bool chown_ok(struct mnt_idmap *idmap,
  95                      const struct inode *inode, vfsuid_t ia_vfsuid)
  96 {
  97         vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
  98         if (vfsuid_eq_kuid(vfsuid, current_fsuid()) &&
  99             vfsuid_eq(ia_vfsuid, vfsuid))
 100                 return true;
 101         if (capable_wrt_inode_uidgid(idmap, inode, CAP_CHOWN))
 102                 return true;
 103         if (!vfsuid_valid(vfsuid) &&
 104             ns_capable(inode->i_sb->s_user_ns, CAP_CHOWN))
 105                 return true;
 106         return false;
 107 }
 108
 109 /**
 110  * chgrp_ok - verify permissions to chgrp inode
 111  * @idmap:      idmap of the mount @inode was found from
 112  * @inode:      inode to check permissions on
 113  * @ia_vfsgid:  gid to chown @inode to
 114  *
 115  * If the inode has been found through an idmapped mount the idmap of
 116  * the vfsmount must be passed through @idmap. This function will then
 117  * take care to map the inode according to @idmap before checking
 118  * permissions. On non-idmapped mounts or if permission checking is to be
 119  * performed on the raw inode simply pass @nop_mnt_idmap.
 120  */
 121 static bool chgrp_ok(struct mnt_idmap *idmap,
 122                      const struct inode *inode, vfsgid_t ia_vfsgid)
 123 {
 124         vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
 125         vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
 126         if (vfsuid_eq_kuid(vfsuid, current_fsuid())) {
 127                 if (vfsgid_eq(ia_vfsgid, vfsgid))
 128                         return true;
 129                 if (vfsgid_in_group_p(ia_vfsgid))
 130                         return true;
 131         }
 132         if (capable_wrt_inode_uidgid(idmap, inode, CAP_CHOWN))
 133                 return true;
 134         if (!vfsgid_valid(vfsgid) &&
 135             ns_capable(inode->i_sb->s_user_ns, CAP_CHOWN))
 136                 return true;
 137         return false;
 138 }
 139
 140 /**
 141  * setattr_prepare - check if attribute changes to a dentry are allowed
 142  * @idmap:      idmap of the mount the inode was found from
 143  * @dentry:     dentry to check
 144  * @attr:       attributes to change
 145  *
 146  * Check if we are allowed to change the attributes contained in @attr
 147  * in the given dentry.  This includes the normal unix access permission
 148  * checks, as well as checks for rlimits and others. The function also clears
 149  * SGID bit from mode if user is not allowed to set it. Also file capabilities
 150  * and IMA extended attributes are cleared if ATTR_KILL_PRIV is set.
 151  *
 152  * If the inode has been found through an idmapped mount the idmap of
 153  * the vfsmount must be passed through @idmap. This function will then
 154  * take care to map the inode according to @idmap before checking
 155  * permissions. On non-idmapped mounts or if permission checking is to be
 156  * performed on the raw inode simply pass @nop_mnt_idmap.
 157  *
 158  * Should be called as the first thing in ->setattr implementations,
 159  * possibly after taking additional locks.
 160  */
 161 int setattr_prepare(struct mnt_idmap *idmap, struct dentry *dentry,
 162                     struct iattr *attr)
 163 {
 164         struct inode *inode = d_inode(dentry);
 165         unsigned int ia_valid = attr->ia_valid;
 166
 167         /*
 168          * First check size constraints.  These can't be overriden using
 169          * ATTR_FORCE.
 170          */
 171         if (ia_valid & ATTR_SIZE) {
 172                 int error = inode_newsize_ok(inode, attr->ia_size);
 173                 if (error)
 174                         return error;
 175         }
 176
 177         /* If force is set do it anyway. */
 178         if (ia_valid & ATTR_FORCE)
 179                 goto kill_priv;
 180
 181         /* Make sure a caller can chown. */
 182         if ((ia_valid & ATTR_UID) &&
 183             !chown_ok(idmap, inode, attr->ia_vfsuid))
 184                 return -EPERM;
 185
 186         /* Make sure caller can chgrp. */
 187         if ((ia_valid & ATTR_GID) &&
 188             !chgrp_ok(idmap, inode, attr->ia_vfsgid))
 189                 return -EPERM;
 190
 191         /* Make sure a caller can chmod. */
 192         if (ia_valid & ATTR_MODE) {
 193                 vfsgid_t vfsgid;
 194
 195                 if (!inode_owner_or_capable(idmap, inode))
 196                         return -EPERM;
 197
 198                 if (ia_valid & ATTR_GID)
 199                         vfsgid = attr->ia_vfsgid;
 200                 else
 201                         vfsgid = i_gid_into_vfsgid(idmap, inode);
 202
 203                 /* Also check the setgid bit! */
 204                 if (!in_group_or_capable(idmap, inode, vfsgid))
 205                         attr->ia_mode &= ~S_ISGID;
 206         }
 207
 208         /* Check for setting the inode time. */
 209         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)) {
 210                 if (!inode_owner_or_capable(idmap, inode))
 211                         return -EPERM;
 212         }
 213
 214 kill_priv:
 215         /* User has permission for the change */
 216         if (ia_valid & ATTR_KILL_PRIV) {
 217                 int error;
 218
 219                 error = security_inode_killpriv(idmap, dentry);
 220                 if (error)
 221                         return error;
 222         }
 223
 224         return 0;
 225 }
 226 EXPORT_SYMBOL(setattr_prepare);
 227
 228 /**
 229  * inode_newsize_ok - may this inode be truncated to a given size
 230  * @inode:      the inode to be truncated
 231  * @offset:     the new size to assign to the inode
 232  *
 233  * inode_newsize_ok must be called with i_mutex held.
 234  *
 235  * inode_newsize_ok will check filesystem limits and ulimits to check that the
 236  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
 237  * when necessary. Caller must not proceed with inode size change if failure is
 238  * returned. @inode must be a file (not directory), with appropriate
 239  * permissions to allow truncate (inode_newsize_ok does NOT check these
 240  * conditions).
 241  *
 242  * Return: 0 on success, -ve errno on failure
 243  */
 244 int inode_newsize_ok(const struct inode *inode, loff_t offset)
 245 {
 246         if (offset < 0)
 247                 return -EINVAL;
 248         if (inode->i_size < offset) {
 249                 unsigned long limit;
 250
 251                 limit = rlimit(RLIMIT_FSIZE);
 252                 if (limit != RLIM_INFINITY && offset > limit)
 253                         goto out_sig;
 254                 if (offset > inode->i_sb->s_maxbytes)
 255                         goto out_big;
 256         } else {
 257                 /*
 258                  * truncation of in-use swapfiles is disallowed - it would
 259                  * cause subsequent swapout to scribble on the now-freed
 260                  * blocks.
 261                  */
 262                 if (IS_SWAPFILE(inode))
 263                         return -ETXTBSY;
 264         }
 265
 266         return 0;
 267 out_sig:
 268         send_sig(SIGXFSZ, current, 0);
 269 out_big:
 270         return -EFBIG;
 271 }
 272 EXPORT_SYMBOL(inode_newsize_ok);
 273
 274 /**
 275  * setattr_copy_mgtime - update timestamps for mgtime inodes
 276  * @inode: inode timestamps to be updated
 277  * @attr: attrs for the update
 278  *
 279  * With multigrain timestamps, take more care to prevent races when
 280  * updating the ctime. Always update the ctime to the very latest using
 281  * the standard mechanism, and use that to populate the atime and mtime
 282  * appropriately (unless those are being set to specific values).
 283  */
 284 static void setattr_copy_mgtime(struct inode *inode, const struct iattr *attr)
 285 {
 286         unsigned int ia_valid = attr->ia_valid;
 287         struct timespec64 now;
 288
 289         if (ia_valid & ATTR_CTIME) {
 290                 /*
 291                  * In the case of an update for a write delegation, we must respect
 292                  * the value in ia_ctime and not use the current time.
 293                  */
 294                 if (ia_valid & ATTR_DELEG)
 295                         now = inode_set_ctime_deleg(inode, attr->ia_ctime);
 296                 else
 297                         now = inode_set_ctime_current(inode);
 298         } else {
 299                 /* If ATTR_CTIME isn't set, then ATTR_MTIME shouldn't be either. */
 300                 WARN_ON_ONCE(ia_valid & ATTR_MTIME);
 301                 now = current_time(inode);
 302         }
 303
 304         if (ia_valid & ATTR_ATIME_SET)
 305                 inode_set_atime_to_ts(inode, attr->ia_atime);
 306         else if (ia_valid & ATTR_ATIME)
 307                 inode_set_atime_to_ts(inode, now);
 308
 309         if (ia_valid & ATTR_MTIME_SET)
 310                 inode_set_mtime_to_ts(inode, attr->ia_mtime);
 311         else if (ia_valid & ATTR_MTIME)
 312                 inode_set_mtime_to_ts(inode, now);
 313 }
 314
 315 /**
 316  * setattr_copy - copy simple metadata updates into the generic inode
 317  * @idmap:      idmap of the mount the inode was found from
 318  * @inode:      the inode to be updated
 319  * @attr:       the new attributes
 320  *
 321  * setattr_copy must be called with i_mutex held.
 322  *
 323  * setattr_copy updates the inode's metadata with that specified
 324  * in attr on idmapped mounts. Necessary permission checks to determine
 325  * whether or not the S_ISGID property needs to be removed are performed with
 326  * the correct idmapped mount permission helpers.
 327  * Noticeably missing is inode size update, which is more complex
 328  * as it requires pagecache updates.
 329  *
 330  * If the inode has been found through an idmapped mount the idmap of
 331  * the vfsmount must be passed through @idmap. This function will then
 332  * take care to map the inode according to @idmap before checking
 333  * permissions. On non-idmapped mounts or if permission checking is to be
 334  * performed on the raw inode simply pass @nop_mnt_idmap.
 335  *
 336  * The inode is not marked as dirty after this operation. The rationale is
 337  * that for "simple" filesystems, the struct inode is the inode storage.
 338  * The caller is free to mark the inode dirty afterwards if needed.
 339  */
 340 void setattr_copy(struct mnt_idmap *idmap, struct inode *inode,
 341                   const struct iattr *attr)
 342 {
 343         unsigned int ia_valid = attr->ia_valid;
 344
 345         i_uid_update(idmap, attr, inode);
 346         i_gid_update(idmap, attr, inode);
 347         if (ia_valid & ATTR_MODE) {
 348                 umode_t mode = attr->ia_mode;
 349                 if (!in_group_or_capable(idmap, inode,
 350                                          i_gid_into_vfsgid(idmap, inode)))
 351                         mode &= ~S_ISGID;
 352                 inode->i_mode = mode;
 353         }
 354
 355         if (is_mgtime(inode))
 356                 return setattr_copy_mgtime(inode, attr);
 357
 358         if (ia_valid & ATTR_ATIME)
 359                 inode_set_atime_to_ts(inode, attr->ia_atime);
 360         if (ia_valid & ATTR_MTIME)
 361                 inode_set_mtime_to_ts(inode, attr->ia_mtime);
 362         if (ia_valid & ATTR_CTIME) {
 363                 if (ia_valid & ATTR_DELEG)
 364                         inode_set_ctime_deleg(inode, attr->ia_ctime);
 365                 else
 366                         inode_set_ctime_to_ts(inode, attr->ia_ctime);
 367         }
 368 }
 369 EXPORT_SYMBOL(setattr_copy);
 370
 371 int may_setattr(struct mnt_idmap *idmap, struct inode *inode,
 372                 unsigned int ia_valid)
 373 {
 374         int error;
 375
 376         if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID | ATTR_TIMES_SET)) {
 377                 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 378                         return -EPERM;
 379         }
 380
 381         /*
 382          * If utimes(2) and friends are called with times == NULL (or both
 383          * times are UTIME_NOW), then we need to check for write permission
 384          */
 385         if (ia_valid & ATTR_TOUCH) {
 386                 if (IS_IMMUTABLE(inode))
 387                         return -EPERM;
 388
 389                 if (!inode_owner_or_capable(idmap, inode)) {
 390                         error = inode_permission(idmap, inode, MAY_WRITE);
 391                         if (error)
 392                                 return error;
 393                 }
 394         }
 395         return 0;
 396 }
 397 EXPORT_SYMBOL(may_setattr);
 398
 399 /**
 400  * notify_change - modify attributes of a filesystem object
 401  * @idmap:      idmap of the mount the inode was found from
 402  * @dentry:     object affected
 403  * @attr:       new attributes
 404  * @delegated_inode: returns inode, if the inode is delegated
 405  *
 406  * The caller must hold the i_mutex on the affected object.
 407  *
 408  * If notify_change discovers a delegation in need of breaking,
 409  * it will return -EWOULDBLOCK and return a reference to the inode in
 410  * delegated_inode.  The caller should then break the delegation and
 411  * retry.  Because breaking a delegation may take a long time, the
 412  * caller should drop the i_mutex before doing so.
 413  *
 414  * Alternatively, a caller may pass NULL for delegated_inode.  This may
 415  * be appropriate for callers that expect the underlying filesystem not
 416  * to be NFS exported.  Also, passing NULL is fine for callers holding
 417  * the file open for write, as there can be no conflicting delegation in
 418  * that case.
 419  *
 420  * If the inode has been found through an idmapped mount the idmap of
 421  * the vfsmount must be passed through @idmap. This function will then
 422  * take care to map the inode according to @idmap before checking
 423  * permissions. On non-idmapped mounts or if permission checking is to be
 424  * performed on the raw inode simply pass @nop_mnt_idmap.
 425  */
 426 int notify_change(struct mnt_idmap *idmap, struct dentry *dentry,
 427                   struct iattr *attr, struct inode **delegated_inode)
 428 {
 429         struct inode *inode = dentry->d_inode;
 430         umode_t mode = inode->i_mode;
 431         int error;
 432         struct timespec64 now;
 433         unsigned int ia_valid = attr->ia_valid;
 434
 435         WARN_ON_ONCE(!inode_is_locked(inode));
 436
 437         error = may_setattr(idmap, inode, ia_valid);
 438         if (error)
 439                 return error;
 440
 441         if ((ia_valid & ATTR_MODE)) {
 442                 /*
 443                  * Don't allow changing the mode of symlinks:
 444                  *
 445                  * (1) The vfs doesn't take the mode of symlinks into account
 446                  *     during permission checking.
 447                  * (2) This has never worked correctly. Most major filesystems
 448                  *     did return EOPNOTSUPP due to interactions with POSIX ACLs
 449                  *     but did still updated the mode of the symlink.
 450                  *     This inconsistency led system call wrapper providers such
 451                  *     as libc to block changing the mode of symlinks with
 452                  *     EOPNOTSUPP already.
 453                  * (3) To even do this in the first place one would have to use
 454                  *     specific file descriptors and quite some effort.
 455                  */
 456                 if (S_ISLNK(inode->i_mode))
 457                         return -EOPNOTSUPP;
 458
 459                 /* Flag setting protected by i_mutex */
 460                 if (is_sxid(attr->ia_mode))
 461                         inode->i_flags &= ~S_NOSEC;
 462         }
 463
 464         now = current_time(inode);
 465
 466         attr->ia_ctime = now;
 467         if (!(ia_valid & ATTR_ATIME_SET))
 468                 attr->ia_atime = now;
 469         else
 470                 attr->ia_atime = timestamp_truncate(attr->ia_atime, inode);
 471         if (!(ia_valid & ATTR_MTIME_SET))
 472                 attr->ia_mtime = now;
 473         else
 474                 attr->ia_mtime = timestamp_truncate(attr->ia_mtime, inode);
 475
 476         if (ia_valid & ATTR_KILL_PRIV) {
 477                 error = security_inode_need_killpriv(dentry);
 478                 if (error < 0)
 479                         return error;
 480                 if (error == 0)
 481                         ia_valid = attr->ia_valid &= ~ATTR_KILL_PRIV;
 482         }
 483
 484         /*
 485          * We now pass ATTR_KILL_S*ID to the lower level setattr function so
 486          * that the function has the ability to reinterpret a mode change
 487          * that's due to these bits. This adds an implicit restriction that
 488          * no function will ever call notify_change with both ATTR_MODE and
 489          * ATTR_KILL_S*ID set.
 490          */
 491         if ((ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) &&
 492             (ia_valid & ATTR_MODE))
 493                 BUG();
 494
 495         if (ia_valid & ATTR_KILL_SUID) {
 496                 if (mode & S_ISUID) {
 497                         ia_valid = attr->ia_valid |= ATTR_MODE;
 498                         attr->ia_mode = (inode->i_mode & ~S_ISUID);
 499                 }
 500         }
 501         if (ia_valid & ATTR_KILL_SGID) {
 502                 if (mode & S_ISGID) {
 503                         if (!(ia_valid & ATTR_MODE)) {
 504                                 ia_valid = attr->ia_valid |= ATTR_MODE;
 505                                 attr->ia_mode = inode->i_mode;
 506                         }
 507                         attr->ia_mode &= ~S_ISGID;
 508                 }
 509         }
 510         if (!(attr->ia_valid & ~(ATTR_KILL_SUID | ATTR_KILL_SGID)))
 511                 return 0;
 512
 513         /*
 514          * Verify that uid/gid changes are valid in the target
 515          * namespace of the superblock.
 516          */
 517         if (ia_valid & ATTR_UID &&
 518             !vfsuid_has_fsmapping(idmap, inode->i_sb->s_user_ns,
 519                                   attr->ia_vfsuid))
 520                 return -EOVERFLOW;
 521         if (ia_valid & ATTR_GID &&
 522             !vfsgid_has_fsmapping(idmap, inode->i_sb->s_user_ns,
 523                                   attr->ia_vfsgid))
 524                 return -EOVERFLOW;
 525
 526         /* Don't allow modifications of files with invalid uids or
 527          * gids unless those uids & gids are being made valid.
 528          */
 529         if (!(ia_valid & ATTR_UID) &&
 530             !vfsuid_valid(i_uid_into_vfsuid(idmap, inode)))
 531                 return -EOVERFLOW;
 532         if (!(ia_valid & ATTR_GID) &&
 533             !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
 534                 return -EOVERFLOW;
 535
 536         error = security_inode_setattr(idmap, dentry, attr);
 537         if (error)
 538                 return error;
 539
 540         /*
 541          * If ATTR_DELEG is set, then these attributes are being set on
 542          * behalf of the holder of a write delegation. We want to avoid
 543          * breaking the delegation in this case.
 544          */
 545         if (!(ia_valid & ATTR_DELEG)) {
 546                 error = try_break_deleg(inode, delegated_inode);
 547                 if (error)
 548                         return error;
 549         }
 550
 551         if (inode->i_op->setattr)
 552                 error = inode->i_op->setattr(idmap, dentry, attr);
 553         else
 554                 error = simple_setattr(idmap, dentry, attr);
 555
 556         if (!error) {
 557                 fsnotify_change(dentry, ia_valid);
 558                 security_inode_post_setattr(idmap, dentry, ia_valid);
 559         }
 560
 561         return error;
 562 }
 563 EXPORT_SYMBOL(notify_change);