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 (offset < 0)
 188                 return -EINVAL;
 189         if (inode->i_size < offset) {
 190                 unsigned long limit;
 191
 192                 limit = rlimit(RLIMIT_FSIZE);
 193                 if (limit != RLIM_INFINITY && offset > limit)
 194                         goto out_sig;
 195                 if (offset > inode->i_sb->s_maxbytes)
 196                         goto out_big;
 197         } else {
 198                 /*
 199                  * truncation of in-use swapfiles is disallowed - it would
 200                  * cause subsequent swapout to scribble on the now-freed
 201                  * blocks.
 202                  */
 203                 if (IS_SWAPFILE(inode))
 204                         return -ETXTBSY;
 205         }
 206
 207         return 0;
 208 out_sig:
 209         send_sig(SIGXFSZ, current, 0);
 210 out_big:
 211         return -EFBIG;
 212 }
 213 EXPORT_SYMBOL(inode_newsize_ok);
 214
 215 /**
 216  * setattr_copy - copy simple metadata updates into the generic inode
 217  * @mnt_userns: user namespace of the mount the inode was found from
 218  * @inode:      the inode to be updated
 219  * @attr:       the new attributes
 220  *
 221  * setattr_copy must be called with i_mutex held.
 222  *
 223  * setattr_copy updates the inode's metadata with that specified
 224  * in attr on idmapped mounts. If file ownership is changed setattr_copy
 225  * doesn't map ia_uid and ia_gid. It will asssume the caller has already
 226  * provided the intended values. Necessary permission checks to determine
 227  * whether or not the S_ISGID property needs to be removed are performed with
 228  * the correct idmapped mount permission helpers.
 229  * Noticeably missing is inode size update, which is more complex
 230  * as it requires pagecache updates.
 231  *
 232  * If the inode has been found through an idmapped mount the user namespace of
 233  * the vfsmount must be passed through @mnt_userns. This function will then
 234  * take care to map the inode according to @mnt_userns before checking
 235  * permissions. On non-idmapped mounts or if permission checking is to be
 236  * performed on the raw inode simply passs init_user_ns.
 237  *
 238  * The inode is not marked as dirty after this operation. The rationale is
 239  * that for "simple" filesystems, the struct inode is the inode storage.
 240  * The caller is free to mark the inode dirty afterwards if needed.
 241  */
 242 void setattr_copy(struct user_namespace *mnt_userns, struct inode *inode,
 243                   const struct iattr *attr)
 244 {
 245         unsigned int ia_valid = attr->ia_valid;
 246
 247         if (ia_valid & ATTR_UID)
 248                 inode->i_uid = attr->ia_uid;
 249         if (ia_valid & ATTR_GID)
 250                 inode->i_gid = attr->ia_gid;
 251         if (ia_valid & ATTR_ATIME)
 252                 inode->i_atime = attr->ia_atime;
 253         if (ia_valid & ATTR_MTIME)
 254                 inode->i_mtime = attr->ia_mtime;
 255         if (ia_valid & ATTR_CTIME)
 256                 inode->i_ctime = attr->ia_ctime;
 257         if (ia_valid & ATTR_MODE) {
 258                 umode_t mode = attr->ia_mode;
 259                 kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
 260                 if (!in_group_p(kgid) &&
 261                     !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FSETID))
 262                         mode &= ~S_ISGID;
 263                 inode->i_mode = mode;
 264         }
 265 }
 266 EXPORT_SYMBOL(setattr_copy);
 267
 268 int may_setattr(struct user_namespace *mnt_userns, struct inode *inode,
 269                 unsigned int ia_valid)
 270 {
 271         int error;
 272
 273         if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID | ATTR_TIMES_SET)) {
 274                 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 275                         return -EPERM;
 276         }
 277
 278         /*
 279          * If utimes(2) and friends are called with times == NULL (or both
 280          * times are UTIME_NOW), then we need to check for write permission
 281          */
 282         if (ia_valid & ATTR_TOUCH) {
 283                 if (IS_IMMUTABLE(inode))
 284                         return -EPERM;
 285
 286                 if (!inode_owner_or_capable(mnt_userns, inode)) {
 287                         error = inode_permission(mnt_userns, inode, MAY_WRITE);
 288                         if (error)
 289                                 return error;
 290                 }
 291         }
 292         return 0;
 293 }
 294 EXPORT_SYMBOL(may_setattr);
 295
 296 /**
 297  * notify_change - modify attributes of a filesytem object
 298  * @mnt_userns: user namespace of the mount the inode was found from
 299  * @dentry:     object affected
 300  * @attr:       new attributes
 301  * @delegated_inode: returns inode, if the inode is delegated
 302  *
 303  * The caller must hold the i_mutex on the affected object.
 304  *
 305  * If notify_change discovers a delegation in need of breaking,
 306  * it will return -EWOULDBLOCK and return a reference to the inode in
 307  * delegated_inode.  The caller should then break the delegation and
 308  * retry.  Because breaking a delegation may take a long time, the
 309  * caller should drop the i_mutex before doing so.
 310  *
 311  * If file ownership is changed notify_change() doesn't map ia_uid and
 312  * ia_gid. It will asssume the caller has already provided the intended values.
 313  *
 314  * Alternatively, a caller may pass NULL for delegated_inode.  This may
 315  * be appropriate for callers that expect the underlying filesystem not
 316  * to be NFS exported.  Also, passing NULL is fine for callers holding
 317  * the file open for write, as there can be no conflicting delegation in
 318  * that case.
 319  *
 320  * If the inode has been found through an idmapped mount the user namespace of
 321  * the vfsmount must be passed through @mnt_userns. This function will then
 322  * take care to map the inode according to @mnt_userns before checking
 323  * permissions. On non-idmapped mounts or if permission checking is to be
 324  * performed on the raw inode simply passs init_user_ns.
 325  */
 326 int notify_change(struct user_namespace *mnt_userns, struct dentry *dentry,
 327                   struct iattr *attr, struct inode **delegated_inode)
 328 {
 329         struct inode *inode = dentry->d_inode;
 330         umode_t mode = inode->i_mode;
 331         int error;
 332         struct timespec64 now;
 333         unsigned int ia_valid = attr->ia_valid;
 334
 335         WARN_ON_ONCE(!inode_is_locked(inode));
 336
 337         error = may_setattr(mnt_userns, inode, ia_valid);
 338         if (error)
 339                 return error;
 340
 341         if ((ia_valid & ATTR_MODE)) {
 342                 umode_t amode = attr->ia_mode;
 343                 /* Flag setting protected by i_mutex */
 344                 if (is_sxid(amode))
 345                         inode->i_flags &= ~S_NOSEC;
 346         }
 347
 348         now = current_time(inode);
 349
 350         attr->ia_ctime = now;
 351         if (!(ia_valid & ATTR_ATIME_SET))
 352                 attr->ia_atime = now;
 353         else
 354                 attr->ia_atime = timestamp_truncate(attr->ia_atime, inode);
 355         if (!(ia_valid & ATTR_MTIME_SET))
 356                 attr->ia_mtime = now;
 357         else
 358                 attr->ia_mtime = timestamp_truncate(attr->ia_mtime, inode);
 359
 360         if (ia_valid & ATTR_KILL_PRIV) {
 361                 error = security_inode_need_killpriv(dentry);
 362                 if (error < 0)
 363                         return error;
 364                 if (error == 0)
 365                         ia_valid = attr->ia_valid &= ~ATTR_KILL_PRIV;
 366         }
 367
 368         /*
 369          * We now pass ATTR_KILL_S*ID to the lower level setattr function so
 370          * that the function has the ability to reinterpret a mode change
 371          * that's due to these bits. This adds an implicit restriction that
 372          * no function will ever call notify_change with both ATTR_MODE and
 373          * ATTR_KILL_S*ID set.
 374          */
 375         if ((ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) &&
 376             (ia_valid & ATTR_MODE))
 377                 BUG();
 378
 379         if (ia_valid & ATTR_KILL_SUID) {
 380                 if (mode & S_ISUID) {
 381                         ia_valid = attr->ia_valid |= ATTR_MODE;
 382                         attr->ia_mode = (inode->i_mode & ~S_ISUID);
 383                 }
 384         }
 385         if (ia_valid & ATTR_KILL_SGID) {
 386                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 387                         if (!(ia_valid & ATTR_MODE)) {
 388                                 ia_valid = attr->ia_valid |= ATTR_MODE;
 389                                 attr->ia_mode = inode->i_mode;
 390                         }
 391                         attr->ia_mode &= ~S_ISGID;
 392                 }
 393         }
 394         if (!(attr->ia_valid & ~(ATTR_KILL_SUID | ATTR_KILL_SGID)))
 395                 return 0;
 396
 397         /*
 398          * Verify that uid/gid changes are valid in the target
 399          * namespace of the superblock.
 400          */
 401         if (ia_valid & ATTR_UID &&
 402             !kuid_has_mapping(inode->i_sb->s_user_ns, attr->ia_uid))
 403                 return -EOVERFLOW;
 404         if (ia_valid & ATTR_GID &&
 405             !kgid_has_mapping(inode->i_sb->s_user_ns, attr->ia_gid))
 406                 return -EOVERFLOW;
 407
 408         /* Don't allow modifications of files with invalid uids or
 409          * gids unless those uids & gids are being made valid.
 410          */
 411         if (!(ia_valid & ATTR_UID) &&
 412             !uid_valid(i_uid_into_mnt(mnt_userns, inode)))
 413                 return -EOVERFLOW;
 414         if (!(ia_valid & ATTR_GID) &&
 415             !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
 416                 return -EOVERFLOW;
 417
 418         error = security_inode_setattr(dentry, attr);
 419         if (error)
 420                 return error;
 421         error = try_break_deleg(inode, delegated_inode);
 422         if (error)
 423                 return error;
 424
 425         if (inode->i_op->setattr)
 426                 error = inode->i_op->setattr(mnt_userns, dentry, attr);
 427         else
 428                 error = simple_setattr(mnt_userns, dentry, attr);
 429
 430         if (!error) {
 431                 fsnotify_change(dentry, ia_valid);
 432                 ima_inode_post_setattr(mnt_userns, dentry);
 433                 evm_inode_post_setattr(dentry, ia_valid);
 434         }
 435
 436         return error;
 437 }
 438 EXPORT_SYMBOL(notify_change);