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/security.h>
  18 #include <linux/evm.h>
  19 #include <linux/ima.h>
  20
  21 /**
  22  * chown_ok - verify permissions to chown inode
  23  * @mnt_userns: user namespace of the mount @inode was found from
  24  * @inode:      inode to check permissions on
  25  * @uid:        uid to chown @inode to
  26  *
  27  * If the inode has been found through an idmapped mount the user namespace of
  28  * the vfsmount must be passed through @mnt_userns. This function will then
  29  * take care to map the inode according to @mnt_userns before checking
  30  * permissions. On non-idmapped mounts or if permission checking is to be
  31  * performed on the raw inode simply passs init_user_ns.
  32  */
  33 static bool chown_ok(struct user_namespace *mnt_userns,
  34                      const struct inode *inode,
  35                      kuid_t uid)
  36 {
  37         kuid_t kuid = i_uid_into_mnt(mnt_userns, inode);
  38         if (uid_eq(current_fsuid(), kuid) && uid_eq(uid, inode->i_uid))
  39                 return true;
  40         if (capable_wrt_inode_uidgid(mnt_userns, inode, CAP_CHOWN))
  41                 return true;
  42         if (uid_eq(kuid, INVALID_UID) &&
  43             ns_capable(inode->i_sb->s_user_ns, CAP_CHOWN))
  44                 return true;
  45         return false;
  46 }
  47
  48 /**
  49  * chgrp_ok - verify permissions to chgrp inode
  50  * @mnt_userns: user namespace of the mount @inode was found from
  51  * @inode:      inode to check permissions on
  52  * @gid:        gid to chown @inode to
  53  *
  54  * If the inode has been found through an idmapped mount the user namespace of
  55  * the vfsmount must be passed through @mnt_userns. This function will then
  56  * take care to map the inode according to @mnt_userns before checking
  57  * permissions. On non-idmapped mounts or if permission checking is to be
  58  * performed on the raw inode simply passs init_user_ns.
  59  */
  60 static bool chgrp_ok(struct user_namespace *mnt_userns,
  61                      const struct inode *inode, kgid_t gid)
  62 {
  63         kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
  64         if (uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) {
  65                 kgid_t mapped_gid;
  66
  67                 if (gid_eq(gid, inode->i_gid))
  68                         return true;
  69                 mapped_gid = mapped_kgid_fs(mnt_userns, i_user_ns(inode), gid);
  70                 if (in_group_p(mapped_gid))
  71                         return true;
  72         }
  73         if (capable_wrt_inode_uidgid(mnt_userns, inode, CAP_CHOWN))
  74                 return true;
  75         if (gid_eq(kgid, INVALID_GID) &&
  76             ns_capable(inode->i_sb->s_user_ns, CAP_CHOWN))
  77                 return true;
  78         return false;
  79 }
  80
  81 /**
  82  * setattr_prepare - check if attribute changes to a dentry are allowed
  83  * @mnt_userns: user namespace of the mount the inode was found from
  84  * @dentry:     dentry to check
  85  * @attr:       attributes to change
  86  *
  87  * Check if we are allowed to change the attributes contained in @attr
  88  * in the given dentry.  This includes the normal unix access permission
  89  * checks, as well as checks for rlimits and others. The function also clears
  90  * SGID bit from mode if user is not allowed to set it. Also file capabilities
  91  * and IMA extended attributes are cleared if ATTR_KILL_PRIV is set.
  92  *
  93  * If the inode has been found through an idmapped mount the user namespace of
  94  * the vfsmount must be passed through @mnt_userns. This function will then
  95  * take care to map the inode according to @mnt_userns before checking
  96  * permissions. On non-idmapped mounts or if permission checking is to be
  97  * performed on the raw inode simply passs init_user_ns.
  98  *
  99  * Should be called as the first thing in ->setattr implementations,
 100  * possibly after taking additional locks.
 101  */
 102 int setattr_prepare(struct user_namespace *mnt_userns, struct dentry *dentry,
 103                     struct iattr *attr)
 104 {
 105         struct inode *inode = d_inode(dentry);
 106         unsigned int ia_valid = attr->ia_valid;
 107
 108         /*
 109          * First check size constraints.  These can't be overriden using
 110          * ATTR_FORCE.
 111          */
 112         if (ia_valid & ATTR_SIZE) {
 113                 int error = inode_newsize_ok(inode, attr->ia_size);
 114                 if (error)
 115                         return error;
 116         }
 117
 118         /* If force is set do it anyway. */
 119         if (ia_valid & ATTR_FORCE)
 120                 goto kill_priv;
 121
 122         /* Make sure a caller can chown. */
 123         if ((ia_valid & ATTR_UID) && !chown_ok(mnt_userns, inode, attr->ia_uid))
 124                 return -EPERM;
 125
 126         /* Make sure caller can chgrp. */
 127         if ((ia_valid & ATTR_GID) && !chgrp_ok(mnt_userns, inode, attr->ia_gid))
 128                 return -EPERM;
 129
 130         /* Make sure a caller can chmod. */
 131         if (ia_valid & ATTR_MODE) {
 132                 kgid_t mapped_gid;
 133
 134                 if (!inode_owner_or_capable(mnt_userns, inode))
 135                         return -EPERM;
 136
 137                 if (ia_valid & ATTR_GID)
 138                         mapped_gid = mapped_kgid_fs(mnt_userns,
 139                                                 i_user_ns(inode), attr->ia_gid);
 140                 else
 141                         mapped_gid = i_gid_into_mnt(mnt_userns, inode);
 142
 143                 /* Also check the setgid bit! */
 144                 if (!in_group_p(mapped_gid) &&
 145                     !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FSETID))
 146                         attr->ia_mode &= ~S_ISGID;
 147         }
 148
 149         /* Check for setting the inode time. */
 150         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)) {
 151                 if (!inode_owner_or_capable(mnt_userns, inode))
 152                         return -EPERM;
 153         }
 154
 155 kill_priv:
 156         /* User has permission for the change */
 157         if (ia_valid & ATTR_KILL_PRIV) {
 158                 int error;
 159
 160                 error = security_inode_killpriv(mnt_userns, dentry);
 161                 if (error)
 162                         return error;
 163         }
 164
 165         return 0;
 166 }
 167 EXPORT_SYMBOL(setattr_prepare);
 168
 169 /**
 170  * inode_newsize_ok - may this inode be truncated to a given size
 171  * @inode:      the inode to be truncated
 172  * @offset:     the new size to assign to the inode
 173  *
 174  * inode_newsize_ok must be called with i_mutex held.
 175  *
 176  * inode_newsize_ok will check filesystem limits and ulimits to check that the
 177  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
 178  * when necessary. Caller must not proceed with inode size change if failure is
 179  * returned. @inode must be a file (not directory), with appropriate
 180  * permissions to allow truncate (inode_newsize_ok does NOT check these
 181  * conditions).
 182  *
 183  * Return: 0 on success, -ve errno on failure
 184  */
 185 int inode_newsize_ok(const struct inode *inode, loff_t offset)
 186 {
 187         if (inode->i_size < offset) {
 188                 unsigned long limit;
 189
 190                 limit = rlimit(RLIMIT_FSIZE);
 191                 if (limit != RLIM_INFINITY && offset > limit)
 192                         goto out_sig;
 193                 if (offset > inode->i_sb->s_maxbytes)
 194                         goto out_big;
 195         } else {
 196                 /*
 197                  * truncation of in-use swapfiles is disallowed - it would
 198                  * cause subsequent swapout to scribble on the now-freed
 199                  * blocks.
 200                  */
 201                 if (IS_SWAPFILE(inode))
 202                         return -ETXTBSY;
 203         }
 204
 205         return 0;
 206 out_sig:
 207         send_sig(SIGXFSZ, current, 0);
 208 out_big:
 209         return -EFBIG;
 210 }
 211 EXPORT_SYMBOL(inode_newsize_ok);
 212
 213 /**
 214  * setattr_copy - copy simple metadata updates into the generic inode
 215  * @mnt_userns: user namespace of the mount the inode was found from
 216  * @inode:      the inode to be updated
 217  * @attr:       the new attributes
 218  *
 219  * setattr_copy must be called with i_mutex held.
 220  *
 221  * setattr_copy updates the inode's metadata with that specified
 222  * in attr on idmapped mounts. If file ownership is changed setattr_copy
 223  * doesn't map ia_uid and ia_gid. It will asssume the caller has already
 224  * provided the intended values. Necessary permission checks to determine
 225  * whether or not the S_ISGID property needs to be removed are performed with
 226  * the correct idmapped mount permission helpers.
 227  * Noticeably missing is inode size update, which is more complex
 228  * as it requires pagecache updates.
 229  *
 230  * If the inode has been found through an idmapped mount the user namespace of
 231  * the vfsmount must be passed through @mnt_userns. This function will then
 232  * take care to map the inode according to @mnt_userns before checking
 233  * permissions. On non-idmapped mounts or if permission checking is to be
 234  * performed on the raw inode simply passs init_user_ns.
 235  *
 236  * The inode is not marked as dirty after this operation. The rationale is
 237  * that for "simple" filesystems, the struct inode is the inode storage.
 238  * The caller is free to mark the inode dirty afterwards if needed.
 239  */
 240 void setattr_copy(struct user_namespace *mnt_userns, struct inode *inode,
 241                   const struct iattr *attr)
 242 {
 243         unsigned int ia_valid = attr->ia_valid;
 244
 245         if (ia_valid & ATTR_UID)
 246                 inode->i_uid = attr->ia_uid;
 247         if (ia_valid & ATTR_GID)
 248                 inode->i_gid = attr->ia_gid;
 249         if (ia_valid & ATTR_ATIME)
 250                 inode->i_atime = attr->ia_atime;
 251         if (ia_valid & ATTR_MTIME)
 252                 inode->i_mtime = attr->ia_mtime;
 253         if (ia_valid & ATTR_CTIME)
 254                 inode->i_ctime = attr->ia_ctime;
 255         if (ia_valid & ATTR_MODE) {
 256                 umode_t mode = attr->ia_mode;
 257                 kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
 258                 if (!in_group_p(kgid) &&
 259                     !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FSETID))
 260                         mode &= ~S_ISGID;
 261                 inode->i_mode = mode;
 262         }
 263 }
 264 EXPORT_SYMBOL(setattr_copy);
 265
 266 int may_setattr(struct user_namespace *mnt_userns, struct inode *inode,
 267                 unsigned int ia_valid)
 268 {
 269         int error;
 270
 271         if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID | ATTR_TIMES_SET)) {
 272                 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 273                         return -EPERM;
 274         }
 275
 276         /*
 277          * If utimes(2) and friends are called with times == NULL (or both
 278          * times are UTIME_NOW), then we need to check for write permission
 279          */
 280         if (ia_valid & ATTR_TOUCH) {
 281                 if (IS_IMMUTABLE(inode))
 282                         return -EPERM;
 283
 284                 if (!inode_owner_or_capable(mnt_userns, inode)) {
 285                         error = inode_permission(mnt_userns, inode, MAY_WRITE);
 286                         if (error)
 287                                 return error;
 288                 }
 289         }
 290         return 0;
 291 }
 292 EXPORT_SYMBOL(may_setattr);
 293
 294 /**
 295  * notify_change - modify attributes of a filesytem object
 296  * @mnt_userns: user namespace of the mount the inode was found from
 297  * @dentry:     object affected
 298  * @attr:       new attributes
 299  * @delegated_inode: returns inode, if the inode is delegated
 300  *
 301  * The caller must hold the i_mutex on the affected object.
 302  *
 303  * If notify_change discovers a delegation in need of breaking,
 304  * it will return -EWOULDBLOCK and return a reference to the inode in
 305  * delegated_inode.  The caller should then break the delegation and
 306  * retry.  Because breaking a delegation may take a long time, the
 307  * caller should drop the i_mutex before doing so.
 308  *
 309  * If file ownership is changed notify_change() doesn't map ia_uid and
 310  * ia_gid. It will asssume the caller has already provided the intended values.
 311  *
 312  * Alternatively, a caller may pass NULL for delegated_inode.  This may
 313  * be appropriate for callers that expect the underlying filesystem not
 314  * to be NFS exported.  Also, passing NULL is fine for callers holding
 315  * the file open for write, as there can be no conflicting delegation in
 316  * that case.
 317  *
 318  * If the inode has been found through an idmapped mount the user namespace of
 319  * the vfsmount must be passed through @mnt_userns. This function will then
 320  * take care to map the inode according to @mnt_userns before checking
 321  * permissions. On non-idmapped mounts or if permission checking is to be
 322  * performed on the raw inode simply passs init_user_ns.
 323  */
 324 int notify_change(struct user_namespace *mnt_userns, struct dentry *dentry,
 325                   struct iattr *attr, struct inode **delegated_inode)
 326 {
 327         struct inode *inode = dentry->d_inode;
 328         umode_t mode = inode->i_mode;
 329         int error;
 330         struct timespec64 now;
 331         unsigned int ia_valid = attr->ia_valid;
 332
 333         WARN_ON_ONCE(!inode_is_locked(inode));
 334
 335         error = may_setattr(mnt_userns, inode, ia_valid);
 336         if (error)
 337                 return error;
 338
 339         if ((ia_valid & ATTR_MODE)) {
 340                 umode_t amode = attr->ia_mode;
 341                 /* Flag setting protected by i_mutex */
 342                 if (is_sxid(amode))
 343                         inode->i_flags &= ~S_NOSEC;
 344         }
 345
 346         now = current_time(inode);
 347
 348         attr->ia_ctime = now;
 349         if (!(ia_valid & ATTR_ATIME_SET))
 350                 attr->ia_atime = now;
 351         else
 352                 attr->ia_atime = timestamp_truncate(attr->ia_atime, inode);
 353         if (!(ia_valid & ATTR_MTIME_SET))
 354                 attr->ia_mtime = now;
 355         else
 356                 attr->ia_mtime = timestamp_truncate(attr->ia_mtime, inode);
 357
 358         if (ia_valid & ATTR_KILL_PRIV) {
 359                 error = security_inode_need_killpriv(dentry);
 360                 if (error < 0)
 361                         return error;
 362                 if (error == 0)
 363                         ia_valid = attr->ia_valid &= ~ATTR_KILL_PRIV;
 364         }
 365
 366         /*
 367          * We now pass ATTR_KILL_S*ID to the lower level setattr function so
 368          * that the function has the ability to reinterpret a mode change
 369          * that's due to these bits. This adds an implicit restriction that
 370          * no function will ever call notify_change with both ATTR_MODE and
 371          * ATTR_KILL_S*ID set.
 372          */
 373         if ((ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) &&
 374             (ia_valid & ATTR_MODE))
 375                 BUG();
 376
 377         if (ia_valid & ATTR_KILL_SUID) {
 378                 if (mode & S_ISUID) {
 379                         ia_valid = attr->ia_valid |= ATTR_MODE;
 380                         attr->ia_mode = (inode->i_mode & ~S_ISUID);
 381                 }
 382         }
 383         if (ia_valid & ATTR_KILL_SGID) {
 384                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 385                         if (!(ia_valid & ATTR_MODE)) {
 386                                 ia_valid = attr->ia_valid |= ATTR_MODE;
 387                                 attr->ia_mode = inode->i_mode;
 388                         }
 389                         attr->ia_mode &= ~S_ISGID;
 390                 }
 391         }
 392         if (!(attr->ia_valid & ~(ATTR_KILL_SUID | ATTR_KILL_SGID)))
 393                 return 0;
 394
 395         /*
 396          * Verify that uid/gid changes are valid in the target
 397          * namespace of the superblock.
 398          */
 399         if (ia_valid & ATTR_UID &&
 400             !kuid_has_mapping(inode->i_sb->s_user_ns, attr->ia_uid))
 401                 return -EOVERFLOW;
 402         if (ia_valid & ATTR_GID &&
 403             !kgid_has_mapping(inode->i_sb->s_user_ns, attr->ia_gid))
 404                 return -EOVERFLOW;
 405
 406         /* Don't allow modifications of files with invalid uids or
 407          * gids unless those uids & gids are being made valid.
 408          */
 409         if (!(ia_valid & ATTR_UID) &&
 410             !uid_valid(i_uid_into_mnt(mnt_userns, inode)))
 411                 return -EOVERFLOW;
 412         if (!(ia_valid & ATTR_GID) &&
 413             !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
 414                 return -EOVERFLOW;
 415
 416         error = security_inode_setattr(dentry, attr);
 417         if (error)
 418                 return error;
 419         error = try_break_deleg(inode, delegated_inode);
 420         if (error)
 421                 return error;
 422
 423         if (inode->i_op->setattr)
 424                 error = inode->i_op->setattr(mnt_userns, dentry, attr);
 425         else
 426                 error = simple_setattr(mnt_userns, dentry, attr);
 427
 428         if (!error) {
 429                 fsnotify_change(dentry, ia_valid);
 430                 ima_inode_post_setattr(mnt_userns, dentry);
 431                 evm_inode_post_setattr(dentry, ia_valid);
 432         }
 433
 434         return error;
 435 }
 436 EXPORT_SYMBOL(notify_change);