fs/xattr.c

   1 /*
   2   File: fs/xattr.c
   3
   4   Extended attribute handling.
   5
   6   Copyright (C) 2001 by Andreas Gruenbacher <a.gruenbacher@computer.org>
   7   Copyright (C) 2001 SGI - Silicon Graphics, Inc <linux-xfs@oss.sgi.com>
   8   Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
   9  */
  10 #include <linux/fs.h>
  11 #include <linux/slab.h>
  12 #include <linux/file.h>
  13 #include <linux/xattr.h>
  14 #include <linux/mount.h>
  15 #include <linux/namei.h>
  16 #include <linux/security.h>
  17 #include <linux/evm.h>
  18 #include <linux/syscalls.h>
  19 #include <linux/export.h>
  20 #include <linux/fsnotify.h>
  21 #include <linux/audit.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/posix_acl_xattr.h>
  24
  25 #include <linux/uaccess.h>
  26
  27 static const char *
  28 strcmp_prefix(const char *a, const char *a_prefix)
  29 {
  30         while (*a_prefix && *a == *a_prefix) {
  31                 a++;
  32                 a_prefix++;
  33         }
  34         return *a_prefix ? NULL : a;
  35 }
  36
  37 /*
  38  * In order to implement different sets of xattr operations for each xattr
  39  * prefix, a filesystem should create a null-terminated array of struct
  40  * xattr_handler (one for each prefix) and hang a pointer to it off of the
  41  * s_xattr field of the superblock.
  42  */
  43 #define for_each_xattr_handler(handlers, handler)               \
  44         if (handlers)                                           \
  45                 for ((handler) = *(handlers)++;                 \
  46                         (handler) != NULL;                      \
  47                         (handler) = *(handlers)++)
  48
  49 /*
  50  * Find the xattr_handler with the matching prefix.
  51  */
  52 static const struct xattr_handler *
  53 xattr_resolve_name(struct inode *inode, const char **name)
  54 {
  55         const struct xattr_handler **handlers = inode->i_sb->s_xattr;
  56         const struct xattr_handler *handler;
  57
  58         if (!(inode->i_opflags & IOP_XATTR)) {
  59                 if (unlikely(is_bad_inode(inode)))
  60                         return ERR_PTR(-EIO);
  61                 return ERR_PTR(-EOPNOTSUPP);
  62         }
  63         for_each_xattr_handler(handlers, handler) {
  64                 const char *n;
  65
  66                 n = strcmp_prefix(*name, xattr_prefix(handler));
  67                 if (n) {
  68                         if (!handler->prefix ^ !*n) {
  69                                 if (*n)
  70                                         continue;
  71                                 return ERR_PTR(-EINVAL);
  72                         }
  73                         *name = n;
  74                         return handler;
  75                 }
  76         }
  77         return ERR_PTR(-EOPNOTSUPP);
  78 }
  79
  80 /*
  81  * Check permissions for extended attribute access.  This is a bit complicated
  82  * because different namespaces have very different rules.
  83  */
  84 static int
  85 xattr_permission(struct inode *inode, const char *name, int mask)
  86 {
  87         /*
  88          * We can never set or remove an extended attribute on a read-only
  89          * filesystem  or on an immutable / append-only inode.
  90          */
  91         if (mask & MAY_WRITE) {
  92                 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
  93                         return -EPERM;
  94                 /*
  95                  * Updating an xattr will likely cause i_uid and i_gid
  96                  * to be writen back improperly if their true value is
  97                  * unknown to the vfs.
  98                  */
  99                 if (HAS_UNMAPPED_ID(inode))
 100                         return -EPERM;
 101         }
 102
 103         /*
 104          * No restriction for security.* and system.* from the VFS.  Decision
 105          * on these is left to the underlying filesystem / security module.
 106          */
 107         if (!strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) ||
 108             !strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
 109                 return 0;
 110
 111         /*
 112          * The trusted.* namespace can only be accessed by privileged users.
 113          */
 114         if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN)) {
 115                 if (!capable(CAP_SYS_ADMIN))
 116                         return (mask & MAY_WRITE) ? -EPERM : -ENODATA;
 117                 return 0;
 118         }
 119
 120         /*
 121          * In the user.* namespace, only regular files and directories can have
 122          * extended attributes. For sticky directories, only the owner and
 123          * privileged users can write attributes.
 124          */
 125         if (!strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN)) {
 126                 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
 127                         return (mask & MAY_WRITE) ? -EPERM : -ENODATA;
 128                 if (S_ISDIR(inode->i_mode) && (inode->i_mode & S_ISVTX) &&
 129                     (mask & MAY_WRITE) && !inode_owner_or_capable(inode))
 130                         return -EPERM;
 131         }
 132
 133         return inode_permission(inode, mask);
 134 }
 135
 136 int
 137 __vfs_setxattr(struct dentry *dentry, struct inode *inode, const char *name,
 138                const void *value, size_t size, int flags)
 139 {
 140         const struct xattr_handler *handler;
 141
 142         handler = xattr_resolve_name(inode, &name);
 143         if (IS_ERR(handler))
 144                 return PTR_ERR(handler);
 145         if (!handler->set)
 146                 return -EOPNOTSUPP;
 147         if (size == 0)
 148                 value = "";  /* empty EA, do not remove */
 149         return handler->set(handler, dentry, inode, name, value, size, flags);
 150 }
 151 EXPORT_SYMBOL(__vfs_setxattr);
 152
 153 /**
 154  *  __vfs_setxattr_noperm - perform setxattr operation without performing
 155  *  permission checks.
 156  *
 157  *  @dentry - object to perform setxattr on
 158  *  @name - xattr name to set
 159  *  @value - value to set @name to
 160  *  @size - size of @value
 161  *  @flags - flags to pass into filesystem operations
 162  *
 163  *  returns the result of the internal setxattr or setsecurity operations.
 164  *
 165  *  This function requires the caller to lock the inode's i_mutex before it
 166  *  is executed. It also assumes that the caller will make the appropriate
 167  *  permission checks.
 168  */
 169 int __vfs_setxattr_noperm(struct dentry *dentry, const char *name,
 170                 const void *value, size_t size, int flags)
 171 {
 172         struct inode *inode = dentry->d_inode;
 173         int error = -EAGAIN;
 174         int issec = !strncmp(name, XATTR_SECURITY_PREFIX,
 175                                    XATTR_SECURITY_PREFIX_LEN);
 176
 177         if (issec)
 178                 inode->i_flags &= ~S_NOSEC;
 179         if (inode->i_opflags & IOP_XATTR) {
 180                 error = __vfs_setxattr(dentry, inode, name, value, size, flags);
 181                 if (!error) {
 182                         fsnotify_xattr(dentry);
 183                         security_inode_post_setxattr(dentry, name, value,
 184                                                      size, flags);
 185                 }
 186         } else {
 187                 if (unlikely(is_bad_inode(inode)))
 188                         return -EIO;
 189         }
 190         if (error == -EAGAIN) {
 191                 error = -EOPNOTSUPP;
 192
 193                 if (issec) {
 194                         const char *suffix = name + XATTR_SECURITY_PREFIX_LEN;
 195
 196                         error = security_inode_setsecurity(inode, suffix, value,
 197                                                            size, flags);
 198                         if (!error)
 199                                 fsnotify_xattr(dentry);
 200                 }
 201         }
 202
 203         return error;
 204 }
 205
 206
 207 int
 208 vfs_setxattr(struct dentry *dentry, const char *name, const void *value,
 209                 size_t size, int flags)
 210 {
 211         struct inode *inode = dentry->d_inode;
 212         int error;
 213
 214         error = xattr_permission(inode, name, MAY_WRITE);
 215         if (error)
 216                 return error;
 217
 218         inode_lock(inode);
 219         error = security_inode_setxattr(dentry, name, value, size, flags);
 220         if (error)
 221                 goto out;
 222
 223         error = __vfs_setxattr_noperm(dentry, name, value, size, flags);
 224
 225 out:
 226         inode_unlock(inode);
 227         return error;
 228 }
 229 EXPORT_SYMBOL_GPL(vfs_setxattr);
 230
 231 ssize_t
 232 xattr_getsecurity(struct inode *inode, const char *name, void *value,
 233                         size_t size)
 234 {
 235         void *buffer = NULL;
 236         ssize_t len;
 237
 238         if (!value || !size) {
 239                 len = security_inode_getsecurity(inode, name, &buffer, false);
 240                 goto out_noalloc;
 241         }
 242
 243         len = security_inode_getsecurity(inode, name, &buffer, true);
 244         if (len < 0)
 245                 return len;
 246         if (size < len) {
 247                 len = -ERANGE;
 248                 goto out;
 249         }
 250         memcpy(value, buffer, len);
 251 out:
 252         security_release_secctx(buffer, len);
 253 out_noalloc:
 254         return len;
 255 }
 256 EXPORT_SYMBOL_GPL(xattr_getsecurity);
 257
 258 /*
 259  * vfs_getxattr_alloc - allocate memory, if necessary, before calling getxattr
 260  *
 261  * Allocate memory, if not already allocated, or re-allocate correct size,
 262  * before retrieving the extended attribute.
 263  *
 264  * Returns the result of alloc, if failed, or the getxattr operation.
 265  */
 266 ssize_t
 267 vfs_getxattr_alloc(struct dentry *dentry, const char *name, char **xattr_value,
 268                    size_t xattr_size, gfp_t flags)
 269 {
 270         const struct xattr_handler *handler;
 271         struct inode *inode = dentry->d_inode;
 272         char *value = *xattr_value;
 273         int error;
 274
 275         error = xattr_permission(inode, name, MAY_READ);
 276         if (error)
 277                 return error;
 278
 279         handler = xattr_resolve_name(inode, &name);
 280         if (IS_ERR(handler))
 281                 return PTR_ERR(handler);
 282         if (!handler->get)
 283                 return -EOPNOTSUPP;
 284         error = handler->get(handler, dentry, inode, name, NULL, 0);
 285         if (error < 0)
 286                 return error;
 287
 288         if (!value || (error > xattr_size)) {
 289                 value = krealloc(*xattr_value, error + 1, flags);
 290                 if (!value)
 291                         return -ENOMEM;
 292                 memset(value, 0, error + 1);
 293         }
 294
 295         error = handler->get(handler, dentry, inode, name, value, error);
 296         *xattr_value = value;
 297         return error;
 298 }
 299
 300 ssize_t
 301 __vfs_getxattr(struct dentry *dentry, struct inode *inode, const char *name,
 302                void *value, size_t size)
 303 {
 304         const struct xattr_handler *handler;
 305
 306         handler = xattr_resolve_name(inode, &name);
 307         if (IS_ERR(handler))
 308                 return PTR_ERR(handler);
 309         if (!handler->get)
 310                 return -EOPNOTSUPP;
 311         return handler->get(handler, dentry, inode, name, value, size);
 312 }
 313 EXPORT_SYMBOL(__vfs_getxattr);
 314
 315 ssize_t
 316 vfs_getxattr(struct dentry *dentry, const char *name, void *value, size_t size)
 317 {
 318         struct inode *inode = dentry->d_inode;
 319         int error;
 320
 321         error = xattr_permission(inode, name, MAY_READ);
 322         if (error)
 323                 return error;
 324
 325         error = security_inode_getxattr(dentry, name);
 326         if (error)
 327                 return error;
 328
 329         if (!strncmp(name, XATTR_SECURITY_PREFIX,
 330                                 XATTR_SECURITY_PREFIX_LEN)) {
 331                 const char *suffix = name + XATTR_SECURITY_PREFIX_LEN;
 332                 int ret = xattr_getsecurity(inode, suffix, value, size);
 333                 /*
 334                  * Only overwrite the return value if a security module
 335                  * is actually active.
 336                  */
 337                 if (ret == -EOPNOTSUPP)
 338                         goto nolsm;
 339                 return ret;
 340         }
 341 nolsm:
 342         return __vfs_getxattr(dentry, inode, name, value, size);
 343 }
 344 EXPORT_SYMBOL_GPL(vfs_getxattr);
 345
 346 ssize_t
 347 vfs_listxattr(struct dentry *dentry, char *list, size_t size)
 348 {
 349         struct inode *inode = d_inode(dentry);
 350         ssize_t error;
 351
 352         error = security_inode_listxattr(dentry);
 353         if (error)
 354                 return error;
 355         if (inode->i_op->listxattr && (inode->i_opflags & IOP_XATTR)) {
 356                 error = -EOPNOTSUPP;
 357                 error = inode->i_op->listxattr(dentry, list, size);
 358         } else {
 359                 error = security_inode_listsecurity(inode, list, size);
 360                 if (size && error > size)
 361                         error = -ERANGE;
 362         }
 363         return error;
 364 }
 365 EXPORT_SYMBOL_GPL(vfs_listxattr);
 366
 367 int
 368 __vfs_removexattr(struct dentry *dentry, const char *name)
 369 {
 370         struct inode *inode = d_inode(dentry);
 371         const struct xattr_handler *handler;
 372
 373         handler = xattr_resolve_name(inode, &name);
 374         if (IS_ERR(handler))
 375                 return PTR_ERR(handler);
 376         if (!handler->set)
 377                 return -EOPNOTSUPP;
 378         return handler->set(handler, dentry, inode, name, NULL, 0, XATTR_REPLACE);
 379 }
 380 EXPORT_SYMBOL(__vfs_removexattr);
 381
 382 int
 383 vfs_removexattr(struct dentry *dentry, const char *name)
 384 {
 385         struct inode *inode = dentry->d_inode;
 386         int error;
 387
 388         error = xattr_permission(inode, name, MAY_WRITE);
 389         if (error)
 390                 return error;
 391
 392         inode_lock(inode);
 393         error = security_inode_removexattr(dentry, name);
 394         if (error)
 395                 goto out;
 396
 397         error = __vfs_removexattr(dentry, name);
 398
 399         if (!error) {
 400                 fsnotify_xattr(dentry);
 401                 evm_inode_post_removexattr(dentry, name);
 402         }
 403
 404 out:
 405         inode_unlock(inode);
 406         return error;
 407 }
 408 EXPORT_SYMBOL_GPL(vfs_removexattr);
 409
 410
 411 /*
 412  * Extended attribute SET operations
 413  */
 414 static long
 415 setxattr(struct dentry *d, const char __user *name, const void __user *value,
 416          size_t size, int flags)
 417 {
 418         int error;
 419         void *kvalue = NULL;
 420         char kname[XATTR_NAME_MAX + 1];
 421
 422         if (flags & ~(XATTR_CREATE|XATTR_REPLACE))
 423                 return -EINVAL;
 424
 425         error = strncpy_from_user(kname, name, sizeof(kname));
 426         if (error == 0 || error == sizeof(kname))
 427                 error = -ERANGE;
 428         if (error < 0)
 429                 return error;
 430
 431         if (size) {
 432                 if (size > XATTR_SIZE_MAX)
 433                         return -E2BIG;
 434                 kvalue = kvmalloc(size, GFP_KERNEL);
 435                 if (!kvalue)
 436                         return -ENOMEM;
 437                 if (copy_from_user(kvalue, value, size)) {
 438                         error = -EFAULT;
 439                         goto out;
 440                 }
 441                 if ((strcmp(kname, XATTR_NAME_POSIX_ACL_ACCESS) == 0) ||
 442                     (strcmp(kname, XATTR_NAME_POSIX_ACL_DEFAULT) == 0))
 443                         posix_acl_fix_xattr_from_user(kvalue, size);
 444         }
 445
 446         error = vfs_setxattr(d, kname, kvalue, size, flags);
 447 out:
 448         kvfree(kvalue);
 449
 450         return error;
 451 }
 452
 453 static int path_setxattr(const char __user *pathname,
 454                          const char __user *name, const void __user *value,
 455                          size_t size, int flags, unsigned int lookup_flags)
 456 {
 457         struct path path;
 458         int error;
 459 retry:
 460         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 461         if (error)
 462                 return error;
 463         error = mnt_want_write(path.mnt);
 464         if (!error) {
 465                 error = setxattr(path.dentry, name, value, size, flags);
 466                 mnt_drop_write(path.mnt);
 467         }
 468         path_put(&path);
 469         if (retry_estale(error, lookup_flags)) {
 470                 lookup_flags |= LOOKUP_REVAL;
 471                 goto retry;
 472         }
 473         return error;
 474 }
 475
 476 SYSCALL_DEFINE5(setxattr, const char __user *, pathname,
 477                 const char __user *, name, const void __user *, value,
 478                 size_t, size, int, flags)
 479 {
 480         return path_setxattr(pathname, name, value, size, flags, LOOKUP_FOLLOW);
 481 }
 482
 483 SYSCALL_DEFINE5(lsetxattr, const char __user *, pathname,
 484                 const char __user *, name, const void __user *, value,
 485                 size_t, size, int, flags)
 486 {
 487         return path_setxattr(pathname, name, value, size, flags, 0);
 488 }
 489
 490 SYSCALL_DEFINE5(fsetxattr, int, fd, const char __user *, name,
 491                 const void __user *,value, size_t, size, int, flags)
 492 {
 493         struct fd f = fdget(fd);
 494         int error = -EBADF;
 495
 496         if (!f.file)
 497                 return error;
 498         audit_file(f.file);
 499         error = mnt_want_write_file(f.file);
 500         if (!error) {
 501                 error = setxattr(f.file->f_path.dentry, name, value, size, flags);
 502                 mnt_drop_write_file(f.file);
 503         }
 504         fdput(f);
 505         return error;
 506 }
 507
 508 /*
 509  * Extended attribute GET operations
 510  */
 511 static ssize_t
 512 getxattr(struct dentry *d, const char __user *name, void __user *value,
 513          size_t size)
 514 {
 515         ssize_t error;
 516         void *kvalue = NULL;
 517         char kname[XATTR_NAME_MAX + 1];
 518
 519         error = strncpy_from_user(kname, name, sizeof(kname));
 520         if (error == 0 || error == sizeof(kname))
 521                 error = -ERANGE;
 522         if (error < 0)
 523                 return error;
 524
 525         if (size) {
 526                 if (size > XATTR_SIZE_MAX)
 527                         size = XATTR_SIZE_MAX;
 528                 kvalue = kvzalloc(size, GFP_KERNEL);
 529                 if (!kvalue)
 530                         return -ENOMEM;
 531         }
 532
 533         error = vfs_getxattr(d, kname, kvalue, size);
 534         if (error > 0) {
 535                 if ((strcmp(kname, XATTR_NAME_POSIX_ACL_ACCESS) == 0) ||
 536                     (strcmp(kname, XATTR_NAME_POSIX_ACL_DEFAULT) == 0))
 537                         posix_acl_fix_xattr_to_user(kvalue, size);
 538                 if (size && copy_to_user(value, kvalue, error))
 539                         error = -EFAULT;
 540         } else if (error == -ERANGE && size >= XATTR_SIZE_MAX) {
 541                 /* The file system tried to returned a value bigger
 542                    than XATTR_SIZE_MAX bytes. Not possible. */
 543                 error = -E2BIG;
 544         }
 545
 546         kvfree(kvalue);
 547
 548         return error;
 549 }
 550
 551 static ssize_t path_getxattr(const char __user *pathname,
 552                              const char __user *name, void __user *value,
 553                              size_t size, unsigned int lookup_flags)
 554 {
 555         struct path path;
 556         ssize_t error;
 557 retry:
 558         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 559         if (error)
 560                 return error;
 561         error = getxattr(path.dentry, name, value, size);
 562         path_put(&path);
 563         if (retry_estale(error, lookup_flags)) {
 564                 lookup_flags |= LOOKUP_REVAL;
 565                 goto retry;
 566         }
 567         return error;
 568 }
 569
 570 SYSCALL_DEFINE4(getxattr, const char __user *, pathname,
 571                 const char __user *, name, void __user *, value, size_t, size)
 572 {
 573         return path_getxattr(pathname, name, value, size, LOOKUP_FOLLOW);
 574 }
 575
 576 SYSCALL_DEFINE4(lgetxattr, const char __user *, pathname,
 577                 const char __user *, name, void __user *, value, size_t, size)
 578 {
 579         return path_getxattr(pathname, name, value, size, 0);
 580 }
 581
 582 SYSCALL_DEFINE4(fgetxattr, int, fd, const char __user *, name,
 583                 void __user *, value, size_t, size)
 584 {
 585         struct fd f = fdget(fd);
 586         ssize_t error = -EBADF;
 587
 588         if (!f.file)
 589                 return error;
 590         audit_file(f.file);
 591         error = getxattr(f.file->f_path.dentry, name, value, size);
 592         fdput(f);
 593         return error;
 594 }
 595
 596 /*
 597  * Extended attribute LIST operations
 598  */
 599 static ssize_t
 600 listxattr(struct dentry *d, char __user *list, size_t size)
 601 {
 602         ssize_t error;
 603         char *klist = NULL;
 604
 605         if (size) {
 606                 if (size > XATTR_LIST_MAX)
 607                         size = XATTR_LIST_MAX;
 608                 klist = kvmalloc(size, GFP_KERNEL);
 609                 if (!klist)
 610                         return -ENOMEM;
 611         }
 612
 613         error = vfs_listxattr(d, klist, size);
 614         if (error > 0) {
 615                 if (size && copy_to_user(list, klist, error))
 616                         error = -EFAULT;
 617         } else if (error == -ERANGE && size >= XATTR_LIST_MAX) {
 618                 /* The file system tried to returned a list bigger
 619                    than XATTR_LIST_MAX bytes. Not possible. */
 620                 error = -E2BIG;
 621         }
 622
 623         kvfree(klist);
 624
 625         return error;
 626 }
 627
 628 static ssize_t path_listxattr(const char __user *pathname, char __user *list,
 629                               size_t size, unsigned int lookup_flags)
 630 {
 631         struct path path;
 632         ssize_t error;
 633 retry:
 634         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 635         if (error)
 636                 return error;
 637         error = listxattr(path.dentry, list, size);
 638         path_put(&path);
 639         if (retry_estale(error, lookup_flags)) {
 640                 lookup_flags |= LOOKUP_REVAL;
 641                 goto retry;
 642         }
 643         return error;
 644 }
 645
 646 SYSCALL_DEFINE3(listxattr, const char __user *, pathname, char __user *, list,
 647                 size_t, size)
 648 {
 649         return path_listxattr(pathname, list, size, LOOKUP_FOLLOW);
 650 }
 651
 652 SYSCALL_DEFINE3(llistxattr, const char __user *, pathname, char __user *, list,
 653                 size_t, size)
 654 {
 655         return path_listxattr(pathname, list, size, 0);
 656 }
 657
 658 SYSCALL_DEFINE3(flistxattr, int, fd, char __user *, list, size_t, size)
 659 {
 660         struct fd f = fdget(fd);
 661         ssize_t error = -EBADF;
 662
 663         if (!f.file)
 664                 return error;
 665         audit_file(f.file);
 666         error = listxattr(f.file->f_path.dentry, list, size);
 667         fdput(f);
 668         return error;
 669 }
 670
 671 /*
 672  * Extended attribute REMOVE operations
 673  */
 674 static long
 675 removexattr(struct dentry *d, const char __user *name)
 676 {
 677         int error;
 678         char kname[XATTR_NAME_MAX + 1];
 679
 680         error = strncpy_from_user(kname, name, sizeof(kname));
 681         if (error == 0 || error == sizeof(kname))
 682                 error = -ERANGE;
 683         if (error < 0)
 684                 return error;
 685
 686         return vfs_removexattr(d, kname);
 687 }
 688
 689 static int path_removexattr(const char __user *pathname,
 690                             const char __user *name, unsigned int lookup_flags)
 691 {
 692         struct path path;
 693         int error;
 694 retry:
 695         error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
 696         if (error)
 697                 return error;
 698         error = mnt_want_write(path.mnt);
 699         if (!error) {
 700                 error = removexattr(path.dentry, name);
 701                 mnt_drop_write(path.mnt);
 702         }
 703         path_put(&path);
 704         if (retry_estale(error, lookup_flags)) {
 705                 lookup_flags |= LOOKUP_REVAL;
 706                 goto retry;
 707         }
 708         return error;
 709 }
 710
 711 SYSCALL_DEFINE2(removexattr, const char __user *, pathname,
 712                 const char __user *, name)
 713 {
 714         return path_removexattr(pathname, name, LOOKUP_FOLLOW);
 715 }
 716
 717 SYSCALL_DEFINE2(lremovexattr, const char __user *, pathname,
 718                 const char __user *, name)
 719 {
 720         return path_removexattr(pathname, name, 0);
 721 }
 722
 723 SYSCALL_DEFINE2(fremovexattr, int, fd, const char __user *, name)
 724 {
 725         struct fd f = fdget(fd);
 726         int error = -EBADF;
 727
 728         if (!f.file)
 729                 return error;
 730         audit_file(f.file);
 731         error = mnt_want_write_file(f.file);
 732         if (!error) {
 733                 error = removexattr(f.file->f_path.dentry, name);
 734                 mnt_drop_write_file(f.file);
 735         }
 736         fdput(f);
 737         return error;
 738 }
 739
 740 /*
 741  * Combine the results of the list() operation from every xattr_handler in the
 742  * list.
 743  */
 744 ssize_t
 745 generic_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 746 {
 747         const struct xattr_handler *handler, **handlers = dentry->d_sb->s_xattr;
 748         unsigned int size = 0;
 749
 750         if (!buffer) {
 751                 for_each_xattr_handler(handlers, handler) {
 752                         if (!handler->name ||
 753                             (handler->list && !handler->list(dentry)))
 754                                 continue;
 755                         size += strlen(handler->name) + 1;
 756                 }
 757         } else {
 758                 char *buf = buffer;
 759                 size_t len;
 760
 761                 for_each_xattr_handler(handlers, handler) {
 762                         if (!handler->name ||
 763                             (handler->list && !handler->list(dentry)))
 764                                 continue;
 765                         len = strlen(handler->name);
 766                         if (len + 1 > buffer_size)
 767                                 return -ERANGE;
 768                         memcpy(buf, handler->name, len + 1);
 769                         buf += len + 1;
 770                         buffer_size -= len + 1;
 771                 }
 772                 size = buf - buffer;
 773         }
 774         return size;
 775 }
 776 EXPORT_SYMBOL(generic_listxattr);
 777
 778 /**
 779  * xattr_full_name  -  Compute full attribute name from suffix
 780  *
 781  * @handler:    handler of the xattr_handler operation
 782  * @name:       name passed to the xattr_handler operation
 783  *
 784  * The get and set xattr handler operations are called with the remainder of
 785  * the attribute name after skipping the handler's prefix: for example, "foo"
 786  * is passed to the get operation of a handler with prefix "user." to get
 787  * attribute "user.foo".  The full name is still "there" in the name though.
 788  *
 789  * Note: the list xattr handler operation when called from the vfs is passed a
 790  * NULL name; some file systems use this operation internally, with varying
 791  * semantics.
 792  */
 793 const char *xattr_full_name(const struct xattr_handler *handler,
 794                             const char *name)
 795 {
 796         size_t prefix_len = strlen(xattr_prefix(handler));
 797
 798         return name - prefix_len;
 799 }
 800 EXPORT_SYMBOL(xattr_full_name);
 801
 802 /*
 803  * Allocate new xattr and copy in the value; but leave the name to callers.
 804  */
 805 struct simple_xattr *simple_xattr_alloc(const void *value, size_t size)
 806 {
 807         struct simple_xattr *new_xattr;
 808         size_t len;
 809
 810         /* wrap around? */
 811         len = sizeof(*new_xattr) + size;
 812         if (len < sizeof(*new_xattr))
 813                 return NULL;
 814
 815         new_xattr = kmalloc(len, GFP_KERNEL);
 816         if (!new_xattr)
 817                 return NULL;
 818
 819         new_xattr->size = size;
 820         memcpy(new_xattr->value, value, size);
 821         return new_xattr;
 822 }
 823
 824 /*
 825  * xattr GET operation for in-memory/pseudo filesystems
 826  */
 827 int simple_xattr_get(struct simple_xattrs *xattrs, const char *name,
 828                      void *buffer, size_t size)
 829 {
 830         struct simple_xattr *xattr;
 831         int ret = -ENODATA;
 832
 833         spin_lock(&xattrs->lock);
 834         list_for_each_entry(xattr, &xattrs->head, list) {
 835                 if (strcmp(name, xattr->name))
 836                         continue;
 837
 838                 ret = xattr->size;
 839                 if (buffer) {
 840                         if (size < xattr->size)
 841                                 ret = -ERANGE;
 842                         else
 843                                 memcpy(buffer, xattr->value, xattr->size);
 844                 }
 845                 break;
 846         }
 847         spin_unlock(&xattrs->lock);
 848         return ret;
 849 }
 850
 851 /**
 852  * simple_xattr_set - xattr SET operation for in-memory/pseudo filesystems
 853  * @xattrs: target simple_xattr list
 854  * @name: name of the extended attribute
 855  * @value: value of the xattr. If %NULL, will remove the attribute.
 856  * @size: size of the new xattr
 857  * @flags: %XATTR_{CREATE|REPLACE}
 858  *
 859  * %XATTR_CREATE is set, the xattr shouldn't exist already; otherwise fails
 860  * with -EEXIST.  If %XATTR_REPLACE is set, the xattr should exist;
 861  * otherwise, fails with -ENODATA.
 862  *
 863  * Returns 0 on success, -errno on failure.
 864  */
 865 int simple_xattr_set(struct simple_xattrs *xattrs, const char *name,
 866                      const void *value, size_t size, int flags)
 867 {
 868         struct simple_xattr *xattr;
 869         struct simple_xattr *new_xattr = NULL;
 870         int err = 0;
 871
 872         /* value == NULL means remove */
 873         if (value) {
 874                 new_xattr = simple_xattr_alloc(value, size);
 875                 if (!new_xattr)
 876                         return -ENOMEM;
 877
 878                 new_xattr->name = kstrdup(name, GFP_KERNEL);
 879                 if (!new_xattr->name) {
 880                         kfree(new_xattr);
 881                         return -ENOMEM;
 882                 }
 883         }
 884
 885         spin_lock(&xattrs->lock);
 886         list_for_each_entry(xattr, &xattrs->head, list) {
 887                 if (!strcmp(name, xattr->name)) {
 888                         if (flags & XATTR_CREATE) {
 889                                 xattr = new_xattr;
 890                                 err = -EEXIST;
 891                         } else if (new_xattr) {
 892                                 list_replace(&xattr->list, &new_xattr->list);
 893                         } else {
 894                                 list_del(&xattr->list);
 895                         }
 896                         goto out;
 897                 }
 898         }
 899         if (flags & XATTR_REPLACE) {
 900                 xattr = new_xattr;
 901                 err = -ENODATA;
 902         } else {
 903                 list_add(&new_xattr->list, &xattrs->head);
 904                 xattr = NULL;
 905         }
 906 out:
 907         spin_unlock(&xattrs->lock);
 908         if (xattr) {
 909                 kfree(xattr->name);
 910                 kfree(xattr);
 911         }
 912         return err;
 913
 914 }
 915
 916 static bool xattr_is_trusted(const char *name)
 917 {
 918         return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
 919 }
 920
 921 static int xattr_list_one(char **buffer, ssize_t *remaining_size,
 922                           const char *name)
 923 {
 924         size_t len = strlen(name) + 1;
 925         if (*buffer) {
 926                 if (*remaining_size < len)
 927                         return -ERANGE;
 928                 memcpy(*buffer, name, len);
 929                 *buffer += len;
 930         }
 931         *remaining_size -= len;
 932         return 0;
 933 }
 934
 935 /*
 936  * xattr LIST operation for in-memory/pseudo filesystems
 937  */
 938 ssize_t simple_xattr_list(struct inode *inode, struct simple_xattrs *xattrs,
 939                           char *buffer, size_t size)
 940 {
 941         bool trusted = capable(CAP_SYS_ADMIN);
 942         struct simple_xattr *xattr;
 943         ssize_t remaining_size = size;
 944         int err = 0;
 945
 946 #ifdef CONFIG_FS_POSIX_ACL
 947         if (inode->i_acl) {
 948                 err = xattr_list_one(&buffer, &remaining_size,
 949                                      XATTR_NAME_POSIX_ACL_ACCESS);
 950                 if (err)
 951                         return err;
 952         }
 953         if (inode->i_default_acl) {
 954                 err = xattr_list_one(&buffer, &remaining_size,
 955                                      XATTR_NAME_POSIX_ACL_DEFAULT);
 956                 if (err)
 957                         return err;
 958         }
 959 #endif
 960
 961         spin_lock(&xattrs->lock);
 962         list_for_each_entry(xattr, &xattrs->head, list) {
 963                 /* skip "trusted." attributes for unprivileged callers */
 964                 if (!trusted && xattr_is_trusted(xattr->name))
 965                         continue;
 966
 967                 err = xattr_list_one(&buffer, &remaining_size, xattr->name);
 968                 if (err)
 969                         break;
 970         }
 971         spin_unlock(&xattrs->lock);
 972
 973         return err ? err : size - remaining_size;
 974 }
 975
 976 /*
 977  * Adds an extended attribute to the list
 978  */
 979 void simple_xattr_list_add(struct simple_xattrs *xattrs,
 980                            struct simple_xattr *new_xattr)
 981 {
 982         spin_lock(&xattrs->lock);
 983         list_add(&new_xattr->list, &xattrs->head);
 984         spin_unlock(&xattrs->lock);
 985 }