| 1 | /* |
| 2 | * linux/fs/namei.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 5 | */ |
| 6 | |
| 7 | /* |
| 8 | * Some corrections by tytso. |
| 9 | */ |
| 10 | |
| 11 | /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname |
| 12 | * lookup logic. |
| 13 | */ |
| 14 | /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. |
| 15 | */ |
| 16 | |
| 17 | #include <linux/init.h> |
| 18 | #include <linux/module.h> |
| 19 | #include <linux/slab.h> |
| 20 | #include <linux/fs.h> |
| 21 | #include <linux/namei.h> |
| 22 | #include <linux/quotaops.h> |
| 23 | #include <linux/pagemap.h> |
| 24 | #include <linux/fsnotify.h> |
| 25 | #include <linux/personality.h> |
| 26 | #include <linux/security.h> |
| 27 | #include <linux/ima.h> |
| 28 | #include <linux/syscalls.h> |
| 29 | #include <linux/mount.h> |
| 30 | #include <linux/audit.h> |
| 31 | #include <linux/capability.h> |
| 32 | #include <linux/file.h> |
| 33 | #include <linux/fcntl.h> |
| 34 | #include <linux/device_cgroup.h> |
| 35 | #include <linux/fs_struct.h> |
| 36 | #include <asm/uaccess.h> |
| 37 | |
| 38 | #include "internal.h" |
| 39 | |
| 40 | /* [Feb-1997 T. Schoebel-Theuer] |
| 41 | * Fundamental changes in the pathname lookup mechanisms (namei) |
| 42 | * were necessary because of omirr. The reason is that omirr needs |
| 43 | * to know the _real_ pathname, not the user-supplied one, in case |
| 44 | * of symlinks (and also when transname replacements occur). |
| 45 | * |
| 46 | * The new code replaces the old recursive symlink resolution with |
| 47 | * an iterative one (in case of non-nested symlink chains). It does |
| 48 | * this with calls to <fs>_follow_link(). |
| 49 | * As a side effect, dir_namei(), _namei() and follow_link() are now |
| 50 | * replaced with a single function lookup_dentry() that can handle all |
| 51 | * the special cases of the former code. |
| 52 | * |
| 53 | * With the new dcache, the pathname is stored at each inode, at least as |
| 54 | * long as the refcount of the inode is positive. As a side effect, the |
| 55 | * size of the dcache depends on the inode cache and thus is dynamic. |
| 56 | * |
| 57 | * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink |
| 58 | * resolution to correspond with current state of the code. |
| 59 | * |
| 60 | * Note that the symlink resolution is not *completely* iterative. |
| 61 | * There is still a significant amount of tail- and mid- recursion in |
| 62 | * the algorithm. Also, note that <fs>_readlink() is not used in |
| 63 | * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() |
| 64 | * may return different results than <fs>_follow_link(). Many virtual |
| 65 | * filesystems (including /proc) exhibit this behavior. |
| 66 | */ |
| 67 | |
| 68 | /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: |
| 69 | * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL |
| 70 | * and the name already exists in form of a symlink, try to create the new |
| 71 | * name indicated by the symlink. The old code always complained that the |
| 72 | * name already exists, due to not following the symlink even if its target |
| 73 | * is nonexistent. The new semantics affects also mknod() and link() when |
| 74 | * the name is a symlink pointing to a non-existant name. |
| 75 | * |
| 76 | * I don't know which semantics is the right one, since I have no access |
| 77 | * to standards. But I found by trial that HP-UX 9.0 has the full "new" |
| 78 | * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the |
| 79 | * "old" one. Personally, I think the new semantics is much more logical. |
| 80 | * Note that "ln old new" where "new" is a symlink pointing to a non-existing |
| 81 | * file does succeed in both HP-UX and SunOs, but not in Solaris |
| 82 | * and in the old Linux semantics. |
| 83 | */ |
| 84 | |
| 85 | /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink |
| 86 | * semantics. See the comments in "open_namei" and "do_link" below. |
| 87 | * |
| 88 | * [10-Sep-98 Alan Modra] Another symlink change. |
| 89 | */ |
| 90 | |
| 91 | /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: |
| 92 | * inside the path - always follow. |
| 93 | * in the last component in creation/removal/renaming - never follow. |
| 94 | * if LOOKUP_FOLLOW passed - follow. |
| 95 | * if the pathname has trailing slashes - follow. |
| 96 | * otherwise - don't follow. |
| 97 | * (applied in that order). |
| 98 | * |
| 99 | * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT |
| 100 | * restored for 2.4. This is the last surviving part of old 4.2BSD bug. |
| 101 | * During the 2.4 we need to fix the userland stuff depending on it - |
| 102 | * hopefully we will be able to get rid of that wart in 2.5. So far only |
| 103 | * XEmacs seems to be relying on it... |
| 104 | */ |
| 105 | /* |
| 106 | * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) |
| 107 | * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives |
| 108 | * any extra contention... |
| 109 | */ |
| 110 | |
| 111 | /* In order to reduce some races, while at the same time doing additional |
| 112 | * checking and hopefully speeding things up, we copy filenames to the |
| 113 | * kernel data space before using them.. |
| 114 | * |
| 115 | * POSIX.1 2.4: an empty pathname is invalid (ENOENT). |
| 116 | * PATH_MAX includes the nul terminator --RR. |
| 117 | */ |
| 118 | static int do_getname(const char __user *filename, char *page) |
| 119 | { |
| 120 | int retval; |
| 121 | unsigned long len = PATH_MAX; |
| 122 | |
| 123 | if (!segment_eq(get_fs(), KERNEL_DS)) { |
| 124 | if ((unsigned long) filename >= TASK_SIZE) |
| 125 | return -EFAULT; |
| 126 | if (TASK_SIZE - (unsigned long) filename < PATH_MAX) |
| 127 | len = TASK_SIZE - (unsigned long) filename; |
| 128 | } |
| 129 | |
| 130 | retval = strncpy_from_user(page, filename, len); |
| 131 | if (retval > 0) { |
| 132 | if (retval < len) |
| 133 | return 0; |
| 134 | return -ENAMETOOLONG; |
| 135 | } else if (!retval) |
| 136 | retval = -ENOENT; |
| 137 | return retval; |
| 138 | } |
| 139 | |
| 140 | char * getname(const char __user * filename) |
| 141 | { |
| 142 | char *tmp, *result; |
| 143 | |
| 144 | result = ERR_PTR(-ENOMEM); |
| 145 | tmp = __getname(); |
| 146 | if (tmp) { |
| 147 | int retval = do_getname(filename, tmp); |
| 148 | |
| 149 | result = tmp; |
| 150 | if (retval < 0) { |
| 151 | __putname(tmp); |
| 152 | result = ERR_PTR(retval); |
| 153 | } |
| 154 | } |
| 155 | audit_getname(result); |
| 156 | return result; |
| 157 | } |
| 158 | |
| 159 | #ifdef CONFIG_AUDITSYSCALL |
| 160 | void putname(const char *name) |
| 161 | { |
| 162 | if (unlikely(!audit_dummy_context())) |
| 163 | audit_putname(name); |
| 164 | else |
| 165 | __putname(name); |
| 166 | } |
| 167 | EXPORT_SYMBOL(putname); |
| 168 | #endif |
| 169 | |
| 170 | /* |
| 171 | * This does basic POSIX ACL permission checking |
| 172 | */ |
| 173 | static int acl_permission_check(struct inode *inode, int mask, |
| 174 | int (*check_acl)(struct inode *inode, int mask)) |
| 175 | { |
| 176 | umode_t mode = inode->i_mode; |
| 177 | |
| 178 | mask &= MAY_READ | MAY_WRITE | MAY_EXEC; |
| 179 | |
| 180 | if (current_fsuid() == inode->i_uid) |
| 181 | mode >>= 6; |
| 182 | else { |
| 183 | if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { |
| 184 | int error = check_acl(inode, mask); |
| 185 | if (error != -EAGAIN) |
| 186 | return error; |
| 187 | } |
| 188 | |
| 189 | if (in_group_p(inode->i_gid)) |
| 190 | mode >>= 3; |
| 191 | } |
| 192 | |
| 193 | /* |
| 194 | * If the DACs are ok we don't need any capability check. |
| 195 | */ |
| 196 | if ((mask & ~mode) == 0) |
| 197 | return 0; |
| 198 | return -EACCES; |
| 199 | } |
| 200 | |
| 201 | /** |
| 202 | * generic_permission - check for access rights on a Posix-like filesystem |
| 203 | * @inode: inode to check access rights for |
| 204 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
| 205 | * @check_acl: optional callback to check for Posix ACLs |
| 206 | * |
| 207 | * Used to check for read/write/execute permissions on a file. |
| 208 | * We use "fsuid" for this, letting us set arbitrary permissions |
| 209 | * for filesystem access without changing the "normal" uids which |
| 210 | * are used for other things.. |
| 211 | */ |
| 212 | int generic_permission(struct inode *inode, int mask, |
| 213 | int (*check_acl)(struct inode *inode, int mask)) |
| 214 | { |
| 215 | int ret; |
| 216 | |
| 217 | /* |
| 218 | * Do the basic POSIX ACL permission checks. |
| 219 | */ |
| 220 | ret = acl_permission_check(inode, mask, check_acl); |
| 221 | if (ret != -EACCES) |
| 222 | return ret; |
| 223 | |
| 224 | /* |
| 225 | * Read/write DACs are always overridable. |
| 226 | * Executable DACs are overridable if at least one exec bit is set. |
| 227 | */ |
| 228 | if (!(mask & MAY_EXEC) || execute_ok(inode)) |
| 229 | if (capable(CAP_DAC_OVERRIDE)) |
| 230 | return 0; |
| 231 | |
| 232 | /* |
| 233 | * Searching includes executable on directories, else just read. |
| 234 | */ |
| 235 | if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) |
| 236 | if (capable(CAP_DAC_READ_SEARCH)) |
| 237 | return 0; |
| 238 | |
| 239 | return -EACCES; |
| 240 | } |
| 241 | |
| 242 | /** |
| 243 | * inode_permission - check for access rights to a given inode |
| 244 | * @inode: inode to check permission on |
| 245 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
| 246 | * |
| 247 | * Used to check for read/write/execute permissions on an inode. |
| 248 | * We use "fsuid" for this, letting us set arbitrary permissions |
| 249 | * for filesystem access without changing the "normal" uids which |
| 250 | * are used for other things. |
| 251 | */ |
| 252 | int inode_permission(struct inode *inode, int mask) |
| 253 | { |
| 254 | int retval; |
| 255 | |
| 256 | if (mask & MAY_WRITE) { |
| 257 | umode_t mode = inode->i_mode; |
| 258 | |
| 259 | /* |
| 260 | * Nobody gets write access to a read-only fs. |
| 261 | */ |
| 262 | if (IS_RDONLY(inode) && |
| 263 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) |
| 264 | return -EROFS; |
| 265 | |
| 266 | /* |
| 267 | * Nobody gets write access to an immutable file. |
| 268 | */ |
| 269 | if (IS_IMMUTABLE(inode)) |
| 270 | return -EACCES; |
| 271 | } |
| 272 | |
| 273 | if (inode->i_op->permission) |
| 274 | retval = inode->i_op->permission(inode, mask); |
| 275 | else |
| 276 | retval = generic_permission(inode, mask, inode->i_op->check_acl); |
| 277 | |
| 278 | if (retval) |
| 279 | return retval; |
| 280 | |
| 281 | retval = devcgroup_inode_permission(inode, mask); |
| 282 | if (retval) |
| 283 | return retval; |
| 284 | |
| 285 | return security_inode_permission(inode, |
| 286 | mask & (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND)); |
| 287 | } |
| 288 | |
| 289 | /** |
| 290 | * file_permission - check for additional access rights to a given file |
| 291 | * @file: file to check access rights for |
| 292 | * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) |
| 293 | * |
| 294 | * Used to check for read/write/execute permissions on an already opened |
| 295 | * file. |
| 296 | * |
| 297 | * Note: |
| 298 | * Do not use this function in new code. All access checks should |
| 299 | * be done using inode_permission(). |
| 300 | */ |
| 301 | int file_permission(struct file *file, int mask) |
| 302 | { |
| 303 | return inode_permission(file->f_path.dentry->d_inode, mask); |
| 304 | } |
| 305 | |
| 306 | /* |
| 307 | * get_write_access() gets write permission for a file. |
| 308 | * put_write_access() releases this write permission. |
| 309 | * This is used for regular files. |
| 310 | * We cannot support write (and maybe mmap read-write shared) accesses and |
| 311 | * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode |
| 312 | * can have the following values: |
| 313 | * 0: no writers, no VM_DENYWRITE mappings |
| 314 | * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist |
| 315 | * > 0: (i_writecount) users are writing to the file. |
| 316 | * |
| 317 | * Normally we operate on that counter with atomic_{inc,dec} and it's safe |
| 318 | * except for the cases where we don't hold i_writecount yet. Then we need to |
| 319 | * use {get,deny}_write_access() - these functions check the sign and refuse |
| 320 | * to do the change if sign is wrong. Exclusion between them is provided by |
| 321 | * the inode->i_lock spinlock. |
| 322 | */ |
| 323 | |
| 324 | int get_write_access(struct inode * inode) |
| 325 | { |
| 326 | spin_lock(&inode->i_lock); |
| 327 | if (atomic_read(&inode->i_writecount) < 0) { |
| 328 | spin_unlock(&inode->i_lock); |
| 329 | return -ETXTBSY; |
| 330 | } |
| 331 | atomic_inc(&inode->i_writecount); |
| 332 | spin_unlock(&inode->i_lock); |
| 333 | |
| 334 | return 0; |
| 335 | } |
| 336 | |
| 337 | int deny_write_access(struct file * file) |
| 338 | { |
| 339 | struct inode *inode = file->f_path.dentry->d_inode; |
| 340 | |
| 341 | spin_lock(&inode->i_lock); |
| 342 | if (atomic_read(&inode->i_writecount) > 0) { |
| 343 | spin_unlock(&inode->i_lock); |
| 344 | return -ETXTBSY; |
| 345 | } |
| 346 | atomic_dec(&inode->i_writecount); |
| 347 | spin_unlock(&inode->i_lock); |
| 348 | |
| 349 | return 0; |
| 350 | } |
| 351 | |
| 352 | /** |
| 353 | * path_get - get a reference to a path |
| 354 | * @path: path to get the reference to |
| 355 | * |
| 356 | * Given a path increment the reference count to the dentry and the vfsmount. |
| 357 | */ |
| 358 | void path_get(struct path *path) |
| 359 | { |
| 360 | mntget(path->mnt); |
| 361 | dget(path->dentry); |
| 362 | } |
| 363 | EXPORT_SYMBOL(path_get); |
| 364 | |
| 365 | /** |
| 366 | * path_put - put a reference to a path |
| 367 | * @path: path to put the reference to |
| 368 | * |
| 369 | * Given a path decrement the reference count to the dentry and the vfsmount. |
| 370 | */ |
| 371 | void path_put(struct path *path) |
| 372 | { |
| 373 | dput(path->dentry); |
| 374 | mntput(path->mnt); |
| 375 | } |
| 376 | EXPORT_SYMBOL(path_put); |
| 377 | |
| 378 | /** |
| 379 | * release_open_intent - free up open intent resources |
| 380 | * @nd: pointer to nameidata |
| 381 | */ |
| 382 | void release_open_intent(struct nameidata *nd) |
| 383 | { |
| 384 | if (nd->intent.open.file->f_path.dentry == NULL) |
| 385 | put_filp(nd->intent.open.file); |
| 386 | else |
| 387 | fput(nd->intent.open.file); |
| 388 | } |
| 389 | |
| 390 | static inline struct dentry * |
| 391 | do_revalidate(struct dentry *dentry, struct nameidata *nd) |
| 392 | { |
| 393 | int status = dentry->d_op->d_revalidate(dentry, nd); |
| 394 | if (unlikely(status <= 0)) { |
| 395 | /* |
| 396 | * The dentry failed validation. |
| 397 | * If d_revalidate returned 0 attempt to invalidate |
| 398 | * the dentry otherwise d_revalidate is asking us |
| 399 | * to return a fail status. |
| 400 | */ |
| 401 | if (!status) { |
| 402 | if (!d_invalidate(dentry)) { |
| 403 | dput(dentry); |
| 404 | dentry = NULL; |
| 405 | } |
| 406 | } else { |
| 407 | dput(dentry); |
| 408 | dentry = ERR_PTR(status); |
| 409 | } |
| 410 | } |
| 411 | return dentry; |
| 412 | } |
| 413 | |
| 414 | /* |
| 415 | * force_reval_path - force revalidation of a dentry |
| 416 | * |
| 417 | * In some situations the path walking code will trust dentries without |
| 418 | * revalidating them. This causes problems for filesystems that depend on |
| 419 | * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set |
| 420 | * (which indicates that it's possible for the dentry to go stale), force |
| 421 | * a d_revalidate call before proceeding. |
| 422 | * |
| 423 | * Returns 0 if the revalidation was successful. If the revalidation fails, |
| 424 | * either return the error returned by d_revalidate or -ESTALE if the |
| 425 | * revalidation it just returned 0. If d_revalidate returns 0, we attempt to |
| 426 | * invalidate the dentry. It's up to the caller to handle putting references |
| 427 | * to the path if necessary. |
| 428 | */ |
| 429 | static int |
| 430 | force_reval_path(struct path *path, struct nameidata *nd) |
| 431 | { |
| 432 | int status; |
| 433 | struct dentry *dentry = path->dentry; |
| 434 | |
| 435 | /* |
| 436 | * only check on filesystems where it's possible for the dentry to |
| 437 | * become stale. It's assumed that if this flag is set then the |
| 438 | * d_revalidate op will also be defined. |
| 439 | */ |
| 440 | if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) |
| 441 | return 0; |
| 442 | |
| 443 | status = dentry->d_op->d_revalidate(dentry, nd); |
| 444 | if (status > 0) |
| 445 | return 0; |
| 446 | |
| 447 | if (!status) { |
| 448 | d_invalidate(dentry); |
| 449 | status = -ESTALE; |
| 450 | } |
| 451 | return status; |
| 452 | } |
| 453 | |
| 454 | /* |
| 455 | * Short-cut version of permission(), for calling on directories |
| 456 | * during pathname resolution. Combines parts of permission() |
| 457 | * and generic_permission(), and tests ONLY for MAY_EXEC permission. |
| 458 | * |
| 459 | * If appropriate, check DAC only. If not appropriate, or |
| 460 | * short-cut DAC fails, then call ->permission() to do more |
| 461 | * complete permission check. |
| 462 | */ |
| 463 | static int exec_permission(struct inode *inode) |
| 464 | { |
| 465 | int ret; |
| 466 | |
| 467 | if (inode->i_op->permission) { |
| 468 | ret = inode->i_op->permission(inode, MAY_EXEC); |
| 469 | if (!ret) |
| 470 | goto ok; |
| 471 | return ret; |
| 472 | } |
| 473 | ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl); |
| 474 | if (!ret) |
| 475 | goto ok; |
| 476 | |
| 477 | if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) |
| 478 | goto ok; |
| 479 | |
| 480 | return ret; |
| 481 | ok: |
| 482 | return security_inode_permission(inode, MAY_EXEC); |
| 483 | } |
| 484 | |
| 485 | static __always_inline void set_root(struct nameidata *nd) |
| 486 | { |
| 487 | if (!nd->root.mnt) { |
| 488 | struct fs_struct *fs = current->fs; |
| 489 | read_lock(&fs->lock); |
| 490 | nd->root = fs->root; |
| 491 | path_get(&nd->root); |
| 492 | read_unlock(&fs->lock); |
| 493 | } |
| 494 | } |
| 495 | |
| 496 | static int link_path_walk(const char *, struct nameidata *); |
| 497 | |
| 498 | static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) |
| 499 | { |
| 500 | int res = 0; |
| 501 | char *name; |
| 502 | if (IS_ERR(link)) |
| 503 | goto fail; |
| 504 | |
| 505 | if (*link == '/') { |
| 506 | set_root(nd); |
| 507 | path_put(&nd->path); |
| 508 | nd->path = nd->root; |
| 509 | path_get(&nd->root); |
| 510 | } |
| 511 | |
| 512 | res = link_path_walk(link, nd); |
| 513 | if (nd->depth || res || nd->last_type!=LAST_NORM) |
| 514 | return res; |
| 515 | /* |
| 516 | * If it is an iterative symlinks resolution in open_namei() we |
| 517 | * have to copy the last component. And all that crap because of |
| 518 | * bloody create() on broken symlinks. Furrfu... |
| 519 | */ |
| 520 | name = __getname(); |
| 521 | if (unlikely(!name)) { |
| 522 | path_put(&nd->path); |
| 523 | return -ENOMEM; |
| 524 | } |
| 525 | strcpy(name, nd->last.name); |
| 526 | nd->last.name = name; |
| 527 | return 0; |
| 528 | fail: |
| 529 | path_put(&nd->path); |
| 530 | return PTR_ERR(link); |
| 531 | } |
| 532 | |
| 533 | static void path_put_conditional(struct path *path, struct nameidata *nd) |
| 534 | { |
| 535 | dput(path->dentry); |
| 536 | if (path->mnt != nd->path.mnt) |
| 537 | mntput(path->mnt); |
| 538 | } |
| 539 | |
| 540 | static inline void path_to_nameidata(struct path *path, struct nameidata *nd) |
| 541 | { |
| 542 | dput(nd->path.dentry); |
| 543 | if (nd->path.mnt != path->mnt) |
| 544 | mntput(nd->path.mnt); |
| 545 | nd->path.mnt = path->mnt; |
| 546 | nd->path.dentry = path->dentry; |
| 547 | } |
| 548 | |
| 549 | static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd) |
| 550 | { |
| 551 | int error; |
| 552 | void *cookie; |
| 553 | struct dentry *dentry = path->dentry; |
| 554 | |
| 555 | touch_atime(path->mnt, dentry); |
| 556 | nd_set_link(nd, NULL); |
| 557 | |
| 558 | if (path->mnt != nd->path.mnt) { |
| 559 | path_to_nameidata(path, nd); |
| 560 | dget(dentry); |
| 561 | } |
| 562 | mntget(path->mnt); |
| 563 | cookie = dentry->d_inode->i_op->follow_link(dentry, nd); |
| 564 | error = PTR_ERR(cookie); |
| 565 | if (!IS_ERR(cookie)) { |
| 566 | char *s = nd_get_link(nd); |
| 567 | error = 0; |
| 568 | if (s) |
| 569 | error = __vfs_follow_link(nd, s); |
| 570 | else if (nd->last_type == LAST_BIND) { |
| 571 | error = force_reval_path(&nd->path, nd); |
| 572 | if (error) |
| 573 | path_put(&nd->path); |
| 574 | } |
| 575 | if (dentry->d_inode->i_op->put_link) |
| 576 | dentry->d_inode->i_op->put_link(dentry, nd, cookie); |
| 577 | } |
| 578 | return error; |
| 579 | } |
| 580 | |
| 581 | /* |
| 582 | * This limits recursive symlink follows to 8, while |
| 583 | * limiting consecutive symlinks to 40. |
| 584 | * |
| 585 | * Without that kind of total limit, nasty chains of consecutive |
| 586 | * symlinks can cause almost arbitrarily long lookups. |
| 587 | */ |
| 588 | static inline int do_follow_link(struct path *path, struct nameidata *nd) |
| 589 | { |
| 590 | int err = -ELOOP; |
| 591 | if (current->link_count >= MAX_NESTED_LINKS) |
| 592 | goto loop; |
| 593 | if (current->total_link_count >= 40) |
| 594 | goto loop; |
| 595 | BUG_ON(nd->depth >= MAX_NESTED_LINKS); |
| 596 | cond_resched(); |
| 597 | err = security_inode_follow_link(path->dentry, nd); |
| 598 | if (err) |
| 599 | goto loop; |
| 600 | current->link_count++; |
| 601 | current->total_link_count++; |
| 602 | nd->depth++; |
| 603 | err = __do_follow_link(path, nd); |
| 604 | path_put(path); |
| 605 | current->link_count--; |
| 606 | nd->depth--; |
| 607 | return err; |
| 608 | loop: |
| 609 | path_put_conditional(path, nd); |
| 610 | path_put(&nd->path); |
| 611 | return err; |
| 612 | } |
| 613 | |
| 614 | int follow_up(struct path *path) |
| 615 | { |
| 616 | struct vfsmount *parent; |
| 617 | struct dentry *mountpoint; |
| 618 | spin_lock(&vfsmount_lock); |
| 619 | parent = path->mnt->mnt_parent; |
| 620 | if (parent == path->mnt) { |
| 621 | spin_unlock(&vfsmount_lock); |
| 622 | return 0; |
| 623 | } |
| 624 | mntget(parent); |
| 625 | mountpoint = dget(path->mnt->mnt_mountpoint); |
| 626 | spin_unlock(&vfsmount_lock); |
| 627 | dput(path->dentry); |
| 628 | path->dentry = mountpoint; |
| 629 | mntput(path->mnt); |
| 630 | path->mnt = parent; |
| 631 | return 1; |
| 632 | } |
| 633 | |
| 634 | /* no need for dcache_lock, as serialization is taken care in |
| 635 | * namespace.c |
| 636 | */ |
| 637 | static int __follow_mount(struct path *path) |
| 638 | { |
| 639 | int res = 0; |
| 640 | while (d_mountpoint(path->dentry)) { |
| 641 | struct vfsmount *mounted = lookup_mnt(path); |
| 642 | if (!mounted) |
| 643 | break; |
| 644 | dput(path->dentry); |
| 645 | if (res) |
| 646 | mntput(path->mnt); |
| 647 | path->mnt = mounted; |
| 648 | path->dentry = dget(mounted->mnt_root); |
| 649 | res = 1; |
| 650 | } |
| 651 | return res; |
| 652 | } |
| 653 | |
| 654 | static void follow_mount(struct path *path) |
| 655 | { |
| 656 | while (d_mountpoint(path->dentry)) { |
| 657 | struct vfsmount *mounted = lookup_mnt(path); |
| 658 | if (!mounted) |
| 659 | break; |
| 660 | dput(path->dentry); |
| 661 | mntput(path->mnt); |
| 662 | path->mnt = mounted; |
| 663 | path->dentry = dget(mounted->mnt_root); |
| 664 | } |
| 665 | } |
| 666 | |
| 667 | /* no need for dcache_lock, as serialization is taken care in |
| 668 | * namespace.c |
| 669 | */ |
| 670 | int follow_down(struct path *path) |
| 671 | { |
| 672 | struct vfsmount *mounted; |
| 673 | |
| 674 | mounted = lookup_mnt(path); |
| 675 | if (mounted) { |
| 676 | dput(path->dentry); |
| 677 | mntput(path->mnt); |
| 678 | path->mnt = mounted; |
| 679 | path->dentry = dget(mounted->mnt_root); |
| 680 | return 1; |
| 681 | } |
| 682 | return 0; |
| 683 | } |
| 684 | |
| 685 | static __always_inline void follow_dotdot(struct nameidata *nd) |
| 686 | { |
| 687 | set_root(nd); |
| 688 | |
| 689 | while(1) { |
| 690 | struct vfsmount *parent; |
| 691 | struct dentry *old = nd->path.dentry; |
| 692 | |
| 693 | if (nd->path.dentry == nd->root.dentry && |
| 694 | nd->path.mnt == nd->root.mnt) { |
| 695 | break; |
| 696 | } |
| 697 | spin_lock(&dcache_lock); |
| 698 | if (nd->path.dentry != nd->path.mnt->mnt_root) { |
| 699 | nd->path.dentry = dget(nd->path.dentry->d_parent); |
| 700 | spin_unlock(&dcache_lock); |
| 701 | dput(old); |
| 702 | break; |
| 703 | } |
| 704 | spin_unlock(&dcache_lock); |
| 705 | spin_lock(&vfsmount_lock); |
| 706 | parent = nd->path.mnt->mnt_parent; |
| 707 | if (parent == nd->path.mnt) { |
| 708 | spin_unlock(&vfsmount_lock); |
| 709 | break; |
| 710 | } |
| 711 | mntget(parent); |
| 712 | nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint); |
| 713 | spin_unlock(&vfsmount_lock); |
| 714 | dput(old); |
| 715 | mntput(nd->path.mnt); |
| 716 | nd->path.mnt = parent; |
| 717 | } |
| 718 | follow_mount(&nd->path); |
| 719 | } |
| 720 | |
| 721 | /* |
| 722 | * It's more convoluted than I'd like it to be, but... it's still fairly |
| 723 | * small and for now I'd prefer to have fast path as straight as possible. |
| 724 | * It _is_ time-critical. |
| 725 | */ |
| 726 | static int do_lookup(struct nameidata *nd, struct qstr *name, |
| 727 | struct path *path) |
| 728 | { |
| 729 | struct vfsmount *mnt = nd->path.mnt; |
| 730 | struct dentry *dentry, *parent; |
| 731 | struct inode *dir; |
| 732 | /* |
| 733 | * See if the low-level filesystem might want |
| 734 | * to use its own hash.. |
| 735 | */ |
| 736 | if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) { |
| 737 | int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name); |
| 738 | if (err < 0) |
| 739 | return err; |
| 740 | } |
| 741 | |
| 742 | dentry = __d_lookup(nd->path.dentry, name); |
| 743 | if (!dentry) |
| 744 | goto need_lookup; |
| 745 | if (dentry->d_op && dentry->d_op->d_revalidate) |
| 746 | goto need_revalidate; |
| 747 | done: |
| 748 | path->mnt = mnt; |
| 749 | path->dentry = dentry; |
| 750 | __follow_mount(path); |
| 751 | return 0; |
| 752 | |
| 753 | need_lookup: |
| 754 | parent = nd->path.dentry; |
| 755 | dir = parent->d_inode; |
| 756 | |
| 757 | mutex_lock(&dir->i_mutex); |
| 758 | /* |
| 759 | * First re-do the cached lookup just in case it was created |
| 760 | * while we waited for the directory semaphore.. |
| 761 | * |
| 762 | * FIXME! This could use version numbering or similar to |
| 763 | * avoid unnecessary cache lookups. |
| 764 | * |
| 765 | * The "dcache_lock" is purely to protect the RCU list walker |
| 766 | * from concurrent renames at this point (we mustn't get false |
| 767 | * negatives from the RCU list walk here, unlike the optimistic |
| 768 | * fast walk). |
| 769 | * |
| 770 | * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup |
| 771 | */ |
| 772 | dentry = d_lookup(parent, name); |
| 773 | if (!dentry) { |
| 774 | struct dentry *new; |
| 775 | |
| 776 | /* Don't create child dentry for a dead directory. */ |
| 777 | dentry = ERR_PTR(-ENOENT); |
| 778 | if (IS_DEADDIR(dir)) |
| 779 | goto out_unlock; |
| 780 | |
| 781 | new = d_alloc(parent, name); |
| 782 | dentry = ERR_PTR(-ENOMEM); |
| 783 | if (new) { |
| 784 | dentry = dir->i_op->lookup(dir, new, nd); |
| 785 | if (dentry) |
| 786 | dput(new); |
| 787 | else |
| 788 | dentry = new; |
| 789 | } |
| 790 | out_unlock: |
| 791 | mutex_unlock(&dir->i_mutex); |
| 792 | if (IS_ERR(dentry)) |
| 793 | goto fail; |
| 794 | goto done; |
| 795 | } |
| 796 | |
| 797 | /* |
| 798 | * Uhhuh! Nasty case: the cache was re-populated while |
| 799 | * we waited on the semaphore. Need to revalidate. |
| 800 | */ |
| 801 | mutex_unlock(&dir->i_mutex); |
| 802 | if (dentry->d_op && dentry->d_op->d_revalidate) { |
| 803 | dentry = do_revalidate(dentry, nd); |
| 804 | if (!dentry) |
| 805 | dentry = ERR_PTR(-ENOENT); |
| 806 | } |
| 807 | if (IS_ERR(dentry)) |
| 808 | goto fail; |
| 809 | goto done; |
| 810 | |
| 811 | need_revalidate: |
| 812 | dentry = do_revalidate(dentry, nd); |
| 813 | if (!dentry) |
| 814 | goto need_lookup; |
| 815 | if (IS_ERR(dentry)) |
| 816 | goto fail; |
| 817 | goto done; |
| 818 | |
| 819 | fail: |
| 820 | return PTR_ERR(dentry); |
| 821 | } |
| 822 | |
| 823 | /* |
| 824 | * Name resolution. |
| 825 | * This is the basic name resolution function, turning a pathname into |
| 826 | * the final dentry. We expect 'base' to be positive and a directory. |
| 827 | * |
| 828 | * Returns 0 and nd will have valid dentry and mnt on success. |
| 829 | * Returns error and drops reference to input namei data on failure. |
| 830 | */ |
| 831 | static int link_path_walk(const char *name, struct nameidata *nd) |
| 832 | { |
| 833 | struct path next; |
| 834 | struct inode *inode; |
| 835 | int err; |
| 836 | unsigned int lookup_flags = nd->flags; |
| 837 | |
| 838 | while (*name=='/') |
| 839 | name++; |
| 840 | if (!*name) |
| 841 | goto return_reval; |
| 842 | |
| 843 | inode = nd->path.dentry->d_inode; |
| 844 | if (nd->depth) |
| 845 | lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); |
| 846 | |
| 847 | /* At this point we know we have a real path component. */ |
| 848 | for(;;) { |
| 849 | unsigned long hash; |
| 850 | struct qstr this; |
| 851 | unsigned int c; |
| 852 | |
| 853 | nd->flags |= LOOKUP_CONTINUE; |
| 854 | err = exec_permission(inode); |
| 855 | if (err) |
| 856 | break; |
| 857 | |
| 858 | this.name = name; |
| 859 | c = *(const unsigned char *)name; |
| 860 | |
| 861 | hash = init_name_hash(); |
| 862 | do { |
| 863 | name++; |
| 864 | hash = partial_name_hash(c, hash); |
| 865 | c = *(const unsigned char *)name; |
| 866 | } while (c && (c != '/')); |
| 867 | this.len = name - (const char *) this.name; |
| 868 | this.hash = end_name_hash(hash); |
| 869 | |
| 870 | /* remove trailing slashes? */ |
| 871 | if (!c) |
| 872 | goto last_component; |
| 873 | while (*++name == '/'); |
| 874 | if (!*name) |
| 875 | goto last_with_slashes; |
| 876 | |
| 877 | /* |
| 878 | * "." and ".." are special - ".." especially so because it has |
| 879 | * to be able to know about the current root directory and |
| 880 | * parent relationships. |
| 881 | */ |
| 882 | if (this.name[0] == '.') switch (this.len) { |
| 883 | default: |
| 884 | break; |
| 885 | case 2: |
| 886 | if (this.name[1] != '.') |
| 887 | break; |
| 888 | follow_dotdot(nd); |
| 889 | inode = nd->path.dentry->d_inode; |
| 890 | /* fallthrough */ |
| 891 | case 1: |
| 892 | continue; |
| 893 | } |
| 894 | /* This does the actual lookups.. */ |
| 895 | err = do_lookup(nd, &this, &next); |
| 896 | if (err) |
| 897 | break; |
| 898 | |
| 899 | err = -ENOENT; |
| 900 | inode = next.dentry->d_inode; |
| 901 | if (!inode) |
| 902 | goto out_dput; |
| 903 | |
| 904 | if (inode->i_op->follow_link) { |
| 905 | err = do_follow_link(&next, nd); |
| 906 | if (err) |
| 907 | goto return_err; |
| 908 | err = -ENOENT; |
| 909 | inode = nd->path.dentry->d_inode; |
| 910 | if (!inode) |
| 911 | break; |
| 912 | } else |
| 913 | path_to_nameidata(&next, nd); |
| 914 | err = -ENOTDIR; |
| 915 | if (!inode->i_op->lookup) |
| 916 | break; |
| 917 | continue; |
| 918 | /* here ends the main loop */ |
| 919 | |
| 920 | last_with_slashes: |
| 921 | lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; |
| 922 | last_component: |
| 923 | /* Clear LOOKUP_CONTINUE iff it was previously unset */ |
| 924 | nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; |
| 925 | if (lookup_flags & LOOKUP_PARENT) |
| 926 | goto lookup_parent; |
| 927 | if (this.name[0] == '.') switch (this.len) { |
| 928 | default: |
| 929 | break; |
| 930 | case 2: |
| 931 | if (this.name[1] != '.') |
| 932 | break; |
| 933 | follow_dotdot(nd); |
| 934 | inode = nd->path.dentry->d_inode; |
| 935 | /* fallthrough */ |
| 936 | case 1: |
| 937 | goto return_reval; |
| 938 | } |
| 939 | err = do_lookup(nd, &this, &next); |
| 940 | if (err) |
| 941 | break; |
| 942 | inode = next.dentry->d_inode; |
| 943 | if ((lookup_flags & LOOKUP_FOLLOW) |
| 944 | && inode && inode->i_op->follow_link) { |
| 945 | err = do_follow_link(&next, nd); |
| 946 | if (err) |
| 947 | goto return_err; |
| 948 | inode = nd->path.dentry->d_inode; |
| 949 | } else |
| 950 | path_to_nameidata(&next, nd); |
| 951 | err = -ENOENT; |
| 952 | if (!inode) |
| 953 | break; |
| 954 | if (lookup_flags & LOOKUP_DIRECTORY) { |
| 955 | err = -ENOTDIR; |
| 956 | if (!inode->i_op->lookup) |
| 957 | break; |
| 958 | } |
| 959 | goto return_base; |
| 960 | lookup_parent: |
| 961 | nd->last = this; |
| 962 | nd->last_type = LAST_NORM; |
| 963 | if (this.name[0] != '.') |
| 964 | goto return_base; |
| 965 | if (this.len == 1) |
| 966 | nd->last_type = LAST_DOT; |
| 967 | else if (this.len == 2 && this.name[1] == '.') |
| 968 | nd->last_type = LAST_DOTDOT; |
| 969 | else |
| 970 | goto return_base; |
| 971 | return_reval: |
| 972 | /* |
| 973 | * We bypassed the ordinary revalidation routines. |
| 974 | * We may need to check the cached dentry for staleness. |
| 975 | */ |
| 976 | if (nd->path.dentry && nd->path.dentry->d_sb && |
| 977 | (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) { |
| 978 | err = -ESTALE; |
| 979 | /* Note: we do not d_invalidate() */ |
| 980 | if (!nd->path.dentry->d_op->d_revalidate( |
| 981 | nd->path.dentry, nd)) |
| 982 | break; |
| 983 | } |
| 984 | return_base: |
| 985 | return 0; |
| 986 | out_dput: |
| 987 | path_put_conditional(&next, nd); |
| 988 | break; |
| 989 | } |
| 990 | path_put(&nd->path); |
| 991 | return_err: |
| 992 | return err; |
| 993 | } |
| 994 | |
| 995 | static int path_walk(const char *name, struct nameidata *nd) |
| 996 | { |
| 997 | struct path save = nd->path; |
| 998 | int result; |
| 999 | |
| 1000 | current->total_link_count = 0; |
| 1001 | |
| 1002 | /* make sure the stuff we saved doesn't go away */ |
| 1003 | path_get(&save); |
| 1004 | |
| 1005 | result = link_path_walk(name, nd); |
| 1006 | if (result == -ESTALE) { |
| 1007 | /* nd->path had been dropped */ |
| 1008 | current->total_link_count = 0; |
| 1009 | nd->path = save; |
| 1010 | path_get(&nd->path); |
| 1011 | nd->flags |= LOOKUP_REVAL; |
| 1012 | result = link_path_walk(name, nd); |
| 1013 | } |
| 1014 | |
| 1015 | path_put(&save); |
| 1016 | |
| 1017 | return result; |
| 1018 | } |
| 1019 | |
| 1020 | static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd) |
| 1021 | { |
| 1022 | int retval = 0; |
| 1023 | int fput_needed; |
| 1024 | struct file *file; |
| 1025 | |
| 1026 | nd->last_type = LAST_ROOT; /* if there are only slashes... */ |
| 1027 | nd->flags = flags; |
| 1028 | nd->depth = 0; |
| 1029 | nd->root.mnt = NULL; |
| 1030 | |
| 1031 | if (*name=='/') { |
| 1032 | set_root(nd); |
| 1033 | nd->path = nd->root; |
| 1034 | path_get(&nd->root); |
| 1035 | } else if (dfd == AT_FDCWD) { |
| 1036 | struct fs_struct *fs = current->fs; |
| 1037 | read_lock(&fs->lock); |
| 1038 | nd->path = fs->pwd; |
| 1039 | path_get(&fs->pwd); |
| 1040 | read_unlock(&fs->lock); |
| 1041 | } else { |
| 1042 | struct dentry *dentry; |
| 1043 | |
| 1044 | file = fget_light(dfd, &fput_needed); |
| 1045 | retval = -EBADF; |
| 1046 | if (!file) |
| 1047 | goto out_fail; |
| 1048 | |
| 1049 | dentry = file->f_path.dentry; |
| 1050 | |
| 1051 | retval = -ENOTDIR; |
| 1052 | if (!S_ISDIR(dentry->d_inode->i_mode)) |
| 1053 | goto fput_fail; |
| 1054 | |
| 1055 | retval = file_permission(file, MAY_EXEC); |
| 1056 | if (retval) |
| 1057 | goto fput_fail; |
| 1058 | |
| 1059 | nd->path = file->f_path; |
| 1060 | path_get(&file->f_path); |
| 1061 | |
| 1062 | fput_light(file, fput_needed); |
| 1063 | } |
| 1064 | return 0; |
| 1065 | |
| 1066 | fput_fail: |
| 1067 | fput_light(file, fput_needed); |
| 1068 | out_fail: |
| 1069 | return retval; |
| 1070 | } |
| 1071 | |
| 1072 | /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ |
| 1073 | static int do_path_lookup(int dfd, const char *name, |
| 1074 | unsigned int flags, struct nameidata *nd) |
| 1075 | { |
| 1076 | int retval = path_init(dfd, name, flags, nd); |
| 1077 | if (!retval) |
| 1078 | retval = path_walk(name, nd); |
| 1079 | if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry && |
| 1080 | nd->path.dentry->d_inode)) |
| 1081 | audit_inode(name, nd->path.dentry); |
| 1082 | if (nd->root.mnt) { |
| 1083 | path_put(&nd->root); |
| 1084 | nd->root.mnt = NULL; |
| 1085 | } |
| 1086 | return retval; |
| 1087 | } |
| 1088 | |
| 1089 | int path_lookup(const char *name, unsigned int flags, |
| 1090 | struct nameidata *nd) |
| 1091 | { |
| 1092 | return do_path_lookup(AT_FDCWD, name, flags, nd); |
| 1093 | } |
| 1094 | |
| 1095 | int kern_path(const char *name, unsigned int flags, struct path *path) |
| 1096 | { |
| 1097 | struct nameidata nd; |
| 1098 | int res = do_path_lookup(AT_FDCWD, name, flags, &nd); |
| 1099 | if (!res) |
| 1100 | *path = nd.path; |
| 1101 | return res; |
| 1102 | } |
| 1103 | |
| 1104 | /** |
| 1105 | * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair |
| 1106 | * @dentry: pointer to dentry of the base directory |
| 1107 | * @mnt: pointer to vfs mount of the base directory |
| 1108 | * @name: pointer to file name |
| 1109 | * @flags: lookup flags |
| 1110 | * @nd: pointer to nameidata |
| 1111 | */ |
| 1112 | int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, |
| 1113 | const char *name, unsigned int flags, |
| 1114 | struct nameidata *nd) |
| 1115 | { |
| 1116 | int retval; |
| 1117 | |
| 1118 | /* same as do_path_lookup */ |
| 1119 | nd->last_type = LAST_ROOT; |
| 1120 | nd->flags = flags; |
| 1121 | nd->depth = 0; |
| 1122 | |
| 1123 | nd->path.dentry = dentry; |
| 1124 | nd->path.mnt = mnt; |
| 1125 | path_get(&nd->path); |
| 1126 | nd->root = nd->path; |
| 1127 | path_get(&nd->root); |
| 1128 | |
| 1129 | retval = path_walk(name, nd); |
| 1130 | if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry && |
| 1131 | nd->path.dentry->d_inode)) |
| 1132 | audit_inode(name, nd->path.dentry); |
| 1133 | |
| 1134 | path_put(&nd->root); |
| 1135 | nd->root.mnt = NULL; |
| 1136 | |
| 1137 | return retval; |
| 1138 | } |
| 1139 | |
| 1140 | static struct dentry *__lookup_hash(struct qstr *name, |
| 1141 | struct dentry *base, struct nameidata *nd) |
| 1142 | { |
| 1143 | struct dentry *dentry; |
| 1144 | struct inode *inode; |
| 1145 | int err; |
| 1146 | |
| 1147 | inode = base->d_inode; |
| 1148 | |
| 1149 | /* |
| 1150 | * See if the low-level filesystem might want |
| 1151 | * to use its own hash.. |
| 1152 | */ |
| 1153 | if (base->d_op && base->d_op->d_hash) { |
| 1154 | err = base->d_op->d_hash(base, name); |
| 1155 | dentry = ERR_PTR(err); |
| 1156 | if (err < 0) |
| 1157 | goto out; |
| 1158 | } |
| 1159 | |
| 1160 | dentry = __d_lookup(base, name); |
| 1161 | |
| 1162 | /* lockess __d_lookup may fail due to concurrent d_move() |
| 1163 | * in some unrelated directory, so try with d_lookup |
| 1164 | */ |
| 1165 | if (!dentry) |
| 1166 | dentry = d_lookup(base, name); |
| 1167 | |
| 1168 | if (dentry && dentry->d_op && dentry->d_op->d_revalidate) |
| 1169 | dentry = do_revalidate(dentry, nd); |
| 1170 | |
| 1171 | if (!dentry) { |
| 1172 | struct dentry *new; |
| 1173 | |
| 1174 | /* Don't create child dentry for a dead directory. */ |
| 1175 | dentry = ERR_PTR(-ENOENT); |
| 1176 | if (IS_DEADDIR(inode)) |
| 1177 | goto out; |
| 1178 | |
| 1179 | new = d_alloc(base, name); |
| 1180 | dentry = ERR_PTR(-ENOMEM); |
| 1181 | if (!new) |
| 1182 | goto out; |
| 1183 | dentry = inode->i_op->lookup(inode, new, nd); |
| 1184 | if (!dentry) |
| 1185 | dentry = new; |
| 1186 | else |
| 1187 | dput(new); |
| 1188 | } |
| 1189 | out: |
| 1190 | return dentry; |
| 1191 | } |
| 1192 | |
| 1193 | /* |
| 1194 | * Restricted form of lookup. Doesn't follow links, single-component only, |
| 1195 | * needs parent already locked. Doesn't follow mounts. |
| 1196 | * SMP-safe. |
| 1197 | */ |
| 1198 | static struct dentry *lookup_hash(struct nameidata *nd) |
| 1199 | { |
| 1200 | int err; |
| 1201 | |
| 1202 | err = exec_permission(nd->path.dentry->d_inode); |
| 1203 | if (err) |
| 1204 | return ERR_PTR(err); |
| 1205 | return __lookup_hash(&nd->last, nd->path.dentry, nd); |
| 1206 | } |
| 1207 | |
| 1208 | static int __lookup_one_len(const char *name, struct qstr *this, |
| 1209 | struct dentry *base, int len) |
| 1210 | { |
| 1211 | unsigned long hash; |
| 1212 | unsigned int c; |
| 1213 | |
| 1214 | this->name = name; |
| 1215 | this->len = len; |
| 1216 | if (!len) |
| 1217 | return -EACCES; |
| 1218 | |
| 1219 | hash = init_name_hash(); |
| 1220 | while (len--) { |
| 1221 | c = *(const unsigned char *)name++; |
| 1222 | if (c == '/' || c == '\0') |
| 1223 | return -EACCES; |
| 1224 | hash = partial_name_hash(c, hash); |
| 1225 | } |
| 1226 | this->hash = end_name_hash(hash); |
| 1227 | return 0; |
| 1228 | } |
| 1229 | |
| 1230 | /** |
| 1231 | * lookup_one_len - filesystem helper to lookup single pathname component |
| 1232 | * @name: pathname component to lookup |
| 1233 | * @base: base directory to lookup from |
| 1234 | * @len: maximum length @len should be interpreted to |
| 1235 | * |
| 1236 | * Note that this routine is purely a helper for filesystem usage and should |
| 1237 | * not be called by generic code. Also note that by using this function the |
| 1238 | * nameidata argument is passed to the filesystem methods and a filesystem |
| 1239 | * using this helper needs to be prepared for that. |
| 1240 | */ |
| 1241 | struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) |
| 1242 | { |
| 1243 | int err; |
| 1244 | struct qstr this; |
| 1245 | |
| 1246 | WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); |
| 1247 | |
| 1248 | err = __lookup_one_len(name, &this, base, len); |
| 1249 | if (err) |
| 1250 | return ERR_PTR(err); |
| 1251 | |
| 1252 | err = exec_permission(base->d_inode); |
| 1253 | if (err) |
| 1254 | return ERR_PTR(err); |
| 1255 | return __lookup_hash(&this, base, NULL); |
| 1256 | } |
| 1257 | |
| 1258 | int user_path_at(int dfd, const char __user *name, unsigned flags, |
| 1259 | struct path *path) |
| 1260 | { |
| 1261 | struct nameidata nd; |
| 1262 | char *tmp = getname(name); |
| 1263 | int err = PTR_ERR(tmp); |
| 1264 | if (!IS_ERR(tmp)) { |
| 1265 | |
| 1266 | BUG_ON(flags & LOOKUP_PARENT); |
| 1267 | |
| 1268 | err = do_path_lookup(dfd, tmp, flags, &nd); |
| 1269 | putname(tmp); |
| 1270 | if (!err) |
| 1271 | *path = nd.path; |
| 1272 | } |
| 1273 | return err; |
| 1274 | } |
| 1275 | |
| 1276 | static int user_path_parent(int dfd, const char __user *path, |
| 1277 | struct nameidata *nd, char **name) |
| 1278 | { |
| 1279 | char *s = getname(path); |
| 1280 | int error; |
| 1281 | |
| 1282 | if (IS_ERR(s)) |
| 1283 | return PTR_ERR(s); |
| 1284 | |
| 1285 | error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); |
| 1286 | if (error) |
| 1287 | putname(s); |
| 1288 | else |
| 1289 | *name = s; |
| 1290 | |
| 1291 | return error; |
| 1292 | } |
| 1293 | |
| 1294 | /* |
| 1295 | * It's inline, so penalty for filesystems that don't use sticky bit is |
| 1296 | * minimal. |
| 1297 | */ |
| 1298 | static inline int check_sticky(struct inode *dir, struct inode *inode) |
| 1299 | { |
| 1300 | uid_t fsuid = current_fsuid(); |
| 1301 | |
| 1302 | if (!(dir->i_mode & S_ISVTX)) |
| 1303 | return 0; |
| 1304 | if (inode->i_uid == fsuid) |
| 1305 | return 0; |
| 1306 | if (dir->i_uid == fsuid) |
| 1307 | return 0; |
| 1308 | return !capable(CAP_FOWNER); |
| 1309 | } |
| 1310 | |
| 1311 | /* |
| 1312 | * Check whether we can remove a link victim from directory dir, check |
| 1313 | * whether the type of victim is right. |
| 1314 | * 1. We can't do it if dir is read-only (done in permission()) |
| 1315 | * 2. We should have write and exec permissions on dir |
| 1316 | * 3. We can't remove anything from append-only dir |
| 1317 | * 4. We can't do anything with immutable dir (done in permission()) |
| 1318 | * 5. If the sticky bit on dir is set we should either |
| 1319 | * a. be owner of dir, or |
| 1320 | * b. be owner of victim, or |
| 1321 | * c. have CAP_FOWNER capability |
| 1322 | * 6. If the victim is append-only or immutable we can't do antyhing with |
| 1323 | * links pointing to it. |
| 1324 | * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. |
| 1325 | * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. |
| 1326 | * 9. We can't remove a root or mountpoint. |
| 1327 | * 10. We don't allow removal of NFS sillyrenamed files; it's handled by |
| 1328 | * nfs_async_unlink(). |
| 1329 | */ |
| 1330 | static int may_delete(struct inode *dir,struct dentry *victim,int isdir) |
| 1331 | { |
| 1332 | int error; |
| 1333 | |
| 1334 | if (!victim->d_inode) |
| 1335 | return -ENOENT; |
| 1336 | |
| 1337 | BUG_ON(victim->d_parent->d_inode != dir); |
| 1338 | audit_inode_child(victim->d_name.name, victim, dir); |
| 1339 | |
| 1340 | error = inode_permission(dir, MAY_WRITE | MAY_EXEC); |
| 1341 | if (error) |
| 1342 | return error; |
| 1343 | if (IS_APPEND(dir)) |
| 1344 | return -EPERM; |
| 1345 | if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| |
| 1346 | IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) |
| 1347 | return -EPERM; |
| 1348 | if (isdir) { |
| 1349 | if (!S_ISDIR(victim->d_inode->i_mode)) |
| 1350 | return -ENOTDIR; |
| 1351 | if (IS_ROOT(victim)) |
| 1352 | return -EBUSY; |
| 1353 | } else if (S_ISDIR(victim->d_inode->i_mode)) |
| 1354 | return -EISDIR; |
| 1355 | if (IS_DEADDIR(dir)) |
| 1356 | return -ENOENT; |
| 1357 | if (victim->d_flags & DCACHE_NFSFS_RENAMED) |
| 1358 | return -EBUSY; |
| 1359 | return 0; |
| 1360 | } |
| 1361 | |
| 1362 | /* Check whether we can create an object with dentry child in directory |
| 1363 | * dir. |
| 1364 | * 1. We can't do it if child already exists (open has special treatment for |
| 1365 | * this case, but since we are inlined it's OK) |
| 1366 | * 2. We can't do it if dir is read-only (done in permission()) |
| 1367 | * 3. We should have write and exec permissions on dir |
| 1368 | * 4. We can't do it if dir is immutable (done in permission()) |
| 1369 | */ |
| 1370 | static inline int may_create(struct inode *dir, struct dentry *child) |
| 1371 | { |
| 1372 | if (child->d_inode) |
| 1373 | return -EEXIST; |
| 1374 | if (IS_DEADDIR(dir)) |
| 1375 | return -ENOENT; |
| 1376 | return inode_permission(dir, MAY_WRITE | MAY_EXEC); |
| 1377 | } |
| 1378 | |
| 1379 | /* |
| 1380 | * O_DIRECTORY translates into forcing a directory lookup. |
| 1381 | */ |
| 1382 | static inline int lookup_flags(unsigned int f) |
| 1383 | { |
| 1384 | unsigned long retval = LOOKUP_FOLLOW; |
| 1385 | |
| 1386 | if (f & O_NOFOLLOW) |
| 1387 | retval &= ~LOOKUP_FOLLOW; |
| 1388 | |
| 1389 | if (f & O_DIRECTORY) |
| 1390 | retval |= LOOKUP_DIRECTORY; |
| 1391 | |
| 1392 | return retval; |
| 1393 | } |
| 1394 | |
| 1395 | /* |
| 1396 | * p1 and p2 should be directories on the same fs. |
| 1397 | */ |
| 1398 | struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) |
| 1399 | { |
| 1400 | struct dentry *p; |
| 1401 | |
| 1402 | if (p1 == p2) { |
| 1403 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
| 1404 | return NULL; |
| 1405 | } |
| 1406 | |
| 1407 | mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
| 1408 | |
| 1409 | p = d_ancestor(p2, p1); |
| 1410 | if (p) { |
| 1411 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); |
| 1412 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); |
| 1413 | return p; |
| 1414 | } |
| 1415 | |
| 1416 | p = d_ancestor(p1, p2); |
| 1417 | if (p) { |
| 1418 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
| 1419 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); |
| 1420 | return p; |
| 1421 | } |
| 1422 | |
| 1423 | mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); |
| 1424 | mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); |
| 1425 | return NULL; |
| 1426 | } |
| 1427 | |
| 1428 | void unlock_rename(struct dentry *p1, struct dentry *p2) |
| 1429 | { |
| 1430 | mutex_unlock(&p1->d_inode->i_mutex); |
| 1431 | if (p1 != p2) { |
| 1432 | mutex_unlock(&p2->d_inode->i_mutex); |
| 1433 | mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); |
| 1434 | } |
| 1435 | } |
| 1436 | |
| 1437 | int vfs_create(struct inode *dir, struct dentry *dentry, int mode, |
| 1438 | struct nameidata *nd) |
| 1439 | { |
| 1440 | int error = may_create(dir, dentry); |
| 1441 | |
| 1442 | if (error) |
| 1443 | return error; |
| 1444 | |
| 1445 | if (!dir->i_op->create) |
| 1446 | return -EACCES; /* shouldn't it be ENOSYS? */ |
| 1447 | mode &= S_IALLUGO; |
| 1448 | mode |= S_IFREG; |
| 1449 | error = security_inode_create(dir, dentry, mode); |
| 1450 | if (error) |
| 1451 | return error; |
| 1452 | vfs_dq_init(dir); |
| 1453 | error = dir->i_op->create(dir, dentry, mode, nd); |
| 1454 | if (!error) |
| 1455 | fsnotify_create(dir, dentry); |
| 1456 | return error; |
| 1457 | } |
| 1458 | |
| 1459 | int may_open(struct path *path, int acc_mode, int flag) |
| 1460 | { |
| 1461 | struct dentry *dentry = path->dentry; |
| 1462 | struct inode *inode = dentry->d_inode; |
| 1463 | int error; |
| 1464 | |
| 1465 | if (!inode) |
| 1466 | return -ENOENT; |
| 1467 | |
| 1468 | switch (inode->i_mode & S_IFMT) { |
| 1469 | case S_IFLNK: |
| 1470 | return -ELOOP; |
| 1471 | case S_IFDIR: |
| 1472 | if (acc_mode & MAY_WRITE) |
| 1473 | return -EISDIR; |
| 1474 | break; |
| 1475 | case S_IFBLK: |
| 1476 | case S_IFCHR: |
| 1477 | if (path->mnt->mnt_flags & MNT_NODEV) |
| 1478 | return -EACCES; |
| 1479 | /*FALLTHRU*/ |
| 1480 | case S_IFIFO: |
| 1481 | case S_IFSOCK: |
| 1482 | flag &= ~O_TRUNC; |
| 1483 | break; |
| 1484 | } |
| 1485 | |
| 1486 | error = inode_permission(inode, acc_mode); |
| 1487 | if (error) |
| 1488 | return error; |
| 1489 | |
| 1490 | /* |
| 1491 | * An append-only file must be opened in append mode for writing. |
| 1492 | */ |
| 1493 | if (IS_APPEND(inode)) { |
| 1494 | if ((flag & FMODE_WRITE) && !(flag & O_APPEND)) |
| 1495 | return -EPERM; |
| 1496 | if (flag & O_TRUNC) |
| 1497 | return -EPERM; |
| 1498 | } |
| 1499 | |
| 1500 | /* O_NOATIME can only be set by the owner or superuser */ |
| 1501 | if (flag & O_NOATIME && !is_owner_or_cap(inode)) |
| 1502 | return -EPERM; |
| 1503 | |
| 1504 | /* |
| 1505 | * Ensure there are no outstanding leases on the file. |
| 1506 | */ |
| 1507 | return break_lease(inode, flag); |
| 1508 | } |
| 1509 | |
| 1510 | static int handle_truncate(struct path *path) |
| 1511 | { |
| 1512 | struct inode *inode = path->dentry->d_inode; |
| 1513 | int error = get_write_access(inode); |
| 1514 | if (error) |
| 1515 | return error; |
| 1516 | /* |
| 1517 | * Refuse to truncate files with mandatory locks held on them. |
| 1518 | */ |
| 1519 | error = locks_verify_locked(inode); |
| 1520 | if (!error) |
| 1521 | error = security_path_truncate(path, 0, |
| 1522 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN); |
| 1523 | if (!error) { |
| 1524 | error = do_truncate(path->dentry, 0, |
| 1525 | ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, |
| 1526 | NULL); |
| 1527 | } |
| 1528 | put_write_access(inode); |
| 1529 | return error; |
| 1530 | } |
| 1531 | |
| 1532 | /* |
| 1533 | * Be careful about ever adding any more callers of this |
| 1534 | * function. Its flags must be in the namei format, not |
| 1535 | * what get passed to sys_open(). |
| 1536 | */ |
| 1537 | static int __open_namei_create(struct nameidata *nd, struct path *path, |
| 1538 | int flag, int mode) |
| 1539 | { |
| 1540 | int error; |
| 1541 | struct dentry *dir = nd->path.dentry; |
| 1542 | |
| 1543 | if (!IS_POSIXACL(dir->d_inode)) |
| 1544 | mode &= ~current_umask(); |
| 1545 | error = security_path_mknod(&nd->path, path->dentry, mode, 0); |
| 1546 | if (error) |
| 1547 | goto out_unlock; |
| 1548 | error = vfs_create(dir->d_inode, path->dentry, mode, nd); |
| 1549 | out_unlock: |
| 1550 | mutex_unlock(&dir->d_inode->i_mutex); |
| 1551 | dput(nd->path.dentry); |
| 1552 | nd->path.dentry = path->dentry; |
| 1553 | if (error) |
| 1554 | return error; |
| 1555 | /* Don't check for write permission, don't truncate */ |
| 1556 | return may_open(&nd->path, 0, flag & ~O_TRUNC); |
| 1557 | } |
| 1558 | |
| 1559 | /* |
| 1560 | * Note that while the flag value (low two bits) for sys_open means: |
| 1561 | * 00 - read-only |
| 1562 | * 01 - write-only |
| 1563 | * 10 - read-write |
| 1564 | * 11 - special |
| 1565 | * it is changed into |
| 1566 | * 00 - no permissions needed |
| 1567 | * 01 - read-permission |
| 1568 | * 10 - write-permission |
| 1569 | * 11 - read-write |
| 1570 | * for the internal routines (ie open_namei()/follow_link() etc) |
| 1571 | * This is more logical, and also allows the 00 "no perm needed" |
| 1572 | * to be used for symlinks (where the permissions are checked |
| 1573 | * later). |
| 1574 | * |
| 1575 | */ |
| 1576 | static inline int open_to_namei_flags(int flag) |
| 1577 | { |
| 1578 | if ((flag+1) & O_ACCMODE) |
| 1579 | flag++; |
| 1580 | return flag; |
| 1581 | } |
| 1582 | |
| 1583 | static int open_will_truncate(int flag, struct inode *inode) |
| 1584 | { |
| 1585 | /* |
| 1586 | * We'll never write to the fs underlying |
| 1587 | * a device file. |
| 1588 | */ |
| 1589 | if (special_file(inode->i_mode)) |
| 1590 | return 0; |
| 1591 | return (flag & O_TRUNC); |
| 1592 | } |
| 1593 | |
| 1594 | /* |
| 1595 | * Note that the low bits of the passed in "open_flag" |
| 1596 | * are not the same as in the local variable "flag". See |
| 1597 | * open_to_namei_flags() for more details. |
| 1598 | */ |
| 1599 | struct file *do_filp_open(int dfd, const char *pathname, |
| 1600 | int open_flag, int mode, int acc_mode) |
| 1601 | { |
| 1602 | struct file *filp; |
| 1603 | struct nameidata nd; |
| 1604 | int error; |
| 1605 | struct path path, save; |
| 1606 | struct dentry *dir; |
| 1607 | int count = 0; |
| 1608 | int will_truncate; |
| 1609 | int flag = open_to_namei_flags(open_flag); |
| 1610 | |
| 1611 | /* |
| 1612 | * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only |
| 1613 | * check for O_DSYNC if the need any syncing at all we enforce it's |
| 1614 | * always set instead of having to deal with possibly weird behaviour |
| 1615 | * for malicious applications setting only __O_SYNC. |
| 1616 | */ |
| 1617 | if (open_flag & __O_SYNC) |
| 1618 | open_flag |= O_DSYNC; |
| 1619 | |
| 1620 | if (!acc_mode) |
| 1621 | acc_mode = MAY_OPEN | ACC_MODE(flag); |
| 1622 | |
| 1623 | /* O_TRUNC implies we need access checks for write permissions */ |
| 1624 | if (flag & O_TRUNC) |
| 1625 | acc_mode |= MAY_WRITE; |
| 1626 | |
| 1627 | /* Allow the LSM permission hook to distinguish append |
| 1628 | access from general write access. */ |
| 1629 | if (flag & O_APPEND) |
| 1630 | acc_mode |= MAY_APPEND; |
| 1631 | |
| 1632 | /* |
| 1633 | * The simplest case - just a plain lookup. |
| 1634 | */ |
| 1635 | if (!(flag & O_CREAT)) { |
| 1636 | filp = get_empty_filp(); |
| 1637 | |
| 1638 | if (filp == NULL) |
| 1639 | return ERR_PTR(-ENFILE); |
| 1640 | nd.intent.open.file = filp; |
| 1641 | filp->f_flags = open_flag; |
| 1642 | nd.intent.open.flags = flag; |
| 1643 | nd.intent.open.create_mode = 0; |
| 1644 | error = do_path_lookup(dfd, pathname, |
| 1645 | lookup_flags(flag)|LOOKUP_OPEN, &nd); |
| 1646 | if (IS_ERR(nd.intent.open.file)) { |
| 1647 | if (error == 0) { |
| 1648 | error = PTR_ERR(nd.intent.open.file); |
| 1649 | path_put(&nd.path); |
| 1650 | } |
| 1651 | } else if (error) |
| 1652 | release_open_intent(&nd); |
| 1653 | if (error) |
| 1654 | return ERR_PTR(error); |
| 1655 | goto ok; |
| 1656 | } |
| 1657 | |
| 1658 | /* |
| 1659 | * Create - we need to know the parent. |
| 1660 | */ |
| 1661 | error = path_init(dfd, pathname, LOOKUP_PARENT, &nd); |
| 1662 | if (error) |
| 1663 | return ERR_PTR(error); |
| 1664 | error = path_walk(pathname, &nd); |
| 1665 | if (error) { |
| 1666 | if (nd.root.mnt) |
| 1667 | path_put(&nd.root); |
| 1668 | return ERR_PTR(error); |
| 1669 | } |
| 1670 | if (unlikely(!audit_dummy_context())) |
| 1671 | audit_inode(pathname, nd.path.dentry); |
| 1672 | |
| 1673 | /* |
| 1674 | * We have the parent and last component. First of all, check |
| 1675 | * that we are not asked to creat(2) an obvious directory - that |
| 1676 | * will not do. |
| 1677 | */ |
| 1678 | error = -EISDIR; |
| 1679 | if (nd.last_type != LAST_NORM || nd.last.name[nd.last.len]) |
| 1680 | goto exit_parent; |
| 1681 | |
| 1682 | error = -ENFILE; |
| 1683 | filp = get_empty_filp(); |
| 1684 | if (filp == NULL) |
| 1685 | goto exit_parent; |
| 1686 | nd.intent.open.file = filp; |
| 1687 | filp->f_flags = open_flag; |
| 1688 | nd.intent.open.flags = flag; |
| 1689 | nd.intent.open.create_mode = mode; |
| 1690 | dir = nd.path.dentry; |
| 1691 | nd.flags &= ~LOOKUP_PARENT; |
| 1692 | nd.flags |= LOOKUP_CREATE | LOOKUP_OPEN; |
| 1693 | if (flag & O_EXCL) |
| 1694 | nd.flags |= LOOKUP_EXCL; |
| 1695 | mutex_lock(&dir->d_inode->i_mutex); |
| 1696 | path.dentry = lookup_hash(&nd); |
| 1697 | path.mnt = nd.path.mnt; |
| 1698 | |
| 1699 | do_last: |
| 1700 | error = PTR_ERR(path.dentry); |
| 1701 | if (IS_ERR(path.dentry)) { |
| 1702 | mutex_unlock(&dir->d_inode->i_mutex); |
| 1703 | goto exit; |
| 1704 | } |
| 1705 | |
| 1706 | if (IS_ERR(nd.intent.open.file)) { |
| 1707 | error = PTR_ERR(nd.intent.open.file); |
| 1708 | goto exit_mutex_unlock; |
| 1709 | } |
| 1710 | |
| 1711 | /* Negative dentry, just create the file */ |
| 1712 | if (!path.dentry->d_inode) { |
| 1713 | /* |
| 1714 | * This write is needed to ensure that a |
| 1715 | * ro->rw transition does not occur between |
| 1716 | * the time when the file is created and when |
| 1717 | * a permanent write count is taken through |
| 1718 | * the 'struct file' in nameidata_to_filp(). |
| 1719 | */ |
| 1720 | error = mnt_want_write(nd.path.mnt); |
| 1721 | if (error) |
| 1722 | goto exit_mutex_unlock; |
| 1723 | error = __open_namei_create(&nd, &path, flag, mode); |
| 1724 | if (error) { |
| 1725 | mnt_drop_write(nd.path.mnt); |
| 1726 | goto exit; |
| 1727 | } |
| 1728 | filp = nameidata_to_filp(&nd); |
| 1729 | mnt_drop_write(nd.path.mnt); |
| 1730 | if (nd.root.mnt) |
| 1731 | path_put(&nd.root); |
| 1732 | if (!IS_ERR(filp)) { |
| 1733 | error = ima_path_check(&filp->f_path, filp->f_mode & |
| 1734 | (MAY_READ | MAY_WRITE | MAY_EXEC)); |
| 1735 | if (error) { |
| 1736 | fput(filp); |
| 1737 | filp = ERR_PTR(error); |
| 1738 | } |
| 1739 | } |
| 1740 | return filp; |
| 1741 | } |
| 1742 | |
| 1743 | /* |
| 1744 | * It already exists. |
| 1745 | */ |
| 1746 | mutex_unlock(&dir->d_inode->i_mutex); |
| 1747 | audit_inode(pathname, path.dentry); |
| 1748 | |
| 1749 | error = -EEXIST; |
| 1750 | if (flag & O_EXCL) |
| 1751 | goto exit_dput; |
| 1752 | |
| 1753 | if (__follow_mount(&path)) { |
| 1754 | error = -ELOOP; |
| 1755 | if (flag & O_NOFOLLOW) |
| 1756 | goto exit_dput; |
| 1757 | } |
| 1758 | |
| 1759 | error = -ENOENT; |
| 1760 | if (!path.dentry->d_inode) |
| 1761 | goto exit_dput; |
| 1762 | if (path.dentry->d_inode->i_op->follow_link) |
| 1763 | goto do_link; |
| 1764 | |
| 1765 | path_to_nameidata(&path, &nd); |
| 1766 | error = -EISDIR; |
| 1767 | if (S_ISDIR(path.dentry->d_inode->i_mode)) |
| 1768 | goto exit; |
| 1769 | ok: |
| 1770 | /* |
| 1771 | * Consider: |
| 1772 | * 1. may_open() truncates a file |
| 1773 | * 2. a rw->ro mount transition occurs |
| 1774 | * 3. nameidata_to_filp() fails due to |
| 1775 | * the ro mount. |
| 1776 | * That would be inconsistent, and should |
| 1777 | * be avoided. Taking this mnt write here |
| 1778 | * ensures that (2) can not occur. |
| 1779 | */ |
| 1780 | will_truncate = open_will_truncate(flag, nd.path.dentry->d_inode); |
| 1781 | if (will_truncate) { |
| 1782 | error = mnt_want_write(nd.path.mnt); |
| 1783 | if (error) |
| 1784 | goto exit; |
| 1785 | } |
| 1786 | error = may_open(&nd.path, acc_mode, flag); |
| 1787 | if (error) { |
| 1788 | if (will_truncate) |
| 1789 | mnt_drop_write(nd.path.mnt); |
| 1790 | goto exit; |
| 1791 | } |
| 1792 | filp = nameidata_to_filp(&nd); |
| 1793 | if (!IS_ERR(filp)) { |
| 1794 | error = ima_path_check(&filp->f_path, filp->f_mode & |
| 1795 | (MAY_READ | MAY_WRITE | MAY_EXEC)); |
| 1796 | if (error) { |
| 1797 | fput(filp); |
| 1798 | filp = ERR_PTR(error); |
| 1799 | } |
| 1800 | } |
| 1801 | if (!IS_ERR(filp)) { |
| 1802 | if (acc_mode & MAY_WRITE) |
| 1803 | vfs_dq_init(nd.path.dentry->d_inode); |
| 1804 | |
| 1805 | if (will_truncate) { |
| 1806 | error = handle_truncate(&nd.path); |
| 1807 | if (error) { |
| 1808 | fput(filp); |
| 1809 | filp = ERR_PTR(error); |
| 1810 | } |
| 1811 | } |
| 1812 | } |
| 1813 | /* |
| 1814 | * It is now safe to drop the mnt write |
| 1815 | * because the filp has had a write taken |
| 1816 | * on its behalf. |
| 1817 | */ |
| 1818 | if (will_truncate) |
| 1819 | mnt_drop_write(nd.path.mnt); |
| 1820 | if (nd.root.mnt) |
| 1821 | path_put(&nd.root); |
| 1822 | return filp; |
| 1823 | |
| 1824 | exit_mutex_unlock: |
| 1825 | mutex_unlock(&dir->d_inode->i_mutex); |
| 1826 | exit_dput: |
| 1827 | path_put_conditional(&path, &nd); |
| 1828 | exit: |
| 1829 | if (!IS_ERR(nd.intent.open.file)) |
| 1830 | release_open_intent(&nd); |
| 1831 | exit_parent: |
| 1832 | if (nd.root.mnt) |
| 1833 | path_put(&nd.root); |
| 1834 | path_put(&nd.path); |
| 1835 | return ERR_PTR(error); |
| 1836 | |
| 1837 | do_link: |
| 1838 | error = -ELOOP; |
| 1839 | if (flag & O_NOFOLLOW) |
| 1840 | goto exit_dput; |
| 1841 | /* |
| 1842 | * This is subtle. Instead of calling do_follow_link() we do the |
| 1843 | * thing by hands. The reason is that this way we have zero link_count |
| 1844 | * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT. |
| 1845 | * After that we have the parent and last component, i.e. |
| 1846 | * we are in the same situation as after the first path_walk(). |
| 1847 | * Well, almost - if the last component is normal we get its copy |
| 1848 | * stored in nd->last.name and we will have to putname() it when we |
| 1849 | * are done. Procfs-like symlinks just set LAST_BIND. |
| 1850 | */ |
| 1851 | nd.flags |= LOOKUP_PARENT; |
| 1852 | error = security_inode_follow_link(path.dentry, &nd); |
| 1853 | if (error) |
| 1854 | goto exit_dput; |
| 1855 | save = nd.path; |
| 1856 | path_get(&save); |
| 1857 | error = __do_follow_link(&path, &nd); |
| 1858 | if (error == -ESTALE) { |
| 1859 | /* nd.path had been dropped */ |
| 1860 | nd.path = save; |
| 1861 | path_get(&nd.path); |
| 1862 | nd.flags |= LOOKUP_REVAL; |
| 1863 | error = __do_follow_link(&path, &nd); |
| 1864 | } |
| 1865 | path_put(&save); |
| 1866 | path_put(&path); |
| 1867 | if (error) { |
| 1868 | /* Does someone understand code flow here? Or it is only |
| 1869 | * me so stupid? Anathema to whoever designed this non-sense |
| 1870 | * with "intent.open". |
| 1871 | */ |
| 1872 | release_open_intent(&nd); |
| 1873 | if (nd.root.mnt) |
| 1874 | path_put(&nd.root); |
| 1875 | return ERR_PTR(error); |
| 1876 | } |
| 1877 | nd.flags &= ~LOOKUP_PARENT; |
| 1878 | if (nd.last_type == LAST_BIND) |
| 1879 | goto ok; |
| 1880 | error = -EISDIR; |
| 1881 | if (nd.last_type != LAST_NORM) |
| 1882 | goto exit; |
| 1883 | if (nd.last.name[nd.last.len]) { |
| 1884 | __putname(nd.last.name); |
| 1885 | goto exit; |
| 1886 | } |
| 1887 | error = -ELOOP; |
| 1888 | if (count++==32) { |
| 1889 | __putname(nd.last.name); |
| 1890 | goto exit; |
| 1891 | } |
| 1892 | dir = nd.path.dentry; |
| 1893 | mutex_lock(&dir->d_inode->i_mutex); |
| 1894 | path.dentry = lookup_hash(&nd); |
| 1895 | path.mnt = nd.path.mnt; |
| 1896 | __putname(nd.last.name); |
| 1897 | goto do_last; |
| 1898 | } |
| 1899 | |
| 1900 | /** |
| 1901 | * filp_open - open file and return file pointer |
| 1902 | * |
| 1903 | * @filename: path to open |
| 1904 | * @flags: open flags as per the open(2) second argument |
| 1905 | * @mode: mode for the new file if O_CREAT is set, else ignored |
| 1906 | * |
| 1907 | * This is the helper to open a file from kernelspace if you really |
| 1908 | * have to. But in generally you should not do this, so please move |
| 1909 | * along, nothing to see here.. |
| 1910 | */ |
| 1911 | struct file *filp_open(const char *filename, int flags, int mode) |
| 1912 | { |
| 1913 | return do_filp_open(AT_FDCWD, filename, flags, mode, 0); |
| 1914 | } |
| 1915 | EXPORT_SYMBOL(filp_open); |
| 1916 | |
| 1917 | /** |
| 1918 | * lookup_create - lookup a dentry, creating it if it doesn't exist |
| 1919 | * @nd: nameidata info |
| 1920 | * @is_dir: directory flag |
| 1921 | * |
| 1922 | * Simple function to lookup and return a dentry and create it |
| 1923 | * if it doesn't exist. Is SMP-safe. |
| 1924 | * |
| 1925 | * Returns with nd->path.dentry->d_inode->i_mutex locked. |
| 1926 | */ |
| 1927 | struct dentry *lookup_create(struct nameidata *nd, int is_dir) |
| 1928 | { |
| 1929 | struct dentry *dentry = ERR_PTR(-EEXIST); |
| 1930 | |
| 1931 | mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
| 1932 | /* |
| 1933 | * Yucky last component or no last component at all? |
| 1934 | * (foo/., foo/.., /////) |
| 1935 | */ |
| 1936 | if (nd->last_type != LAST_NORM) |
| 1937 | goto fail; |
| 1938 | nd->flags &= ~LOOKUP_PARENT; |
| 1939 | nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; |
| 1940 | nd->intent.open.flags = O_EXCL; |
| 1941 | |
| 1942 | /* |
| 1943 | * Do the final lookup. |
| 1944 | */ |
| 1945 | dentry = lookup_hash(nd); |
| 1946 | if (IS_ERR(dentry)) |
| 1947 | goto fail; |
| 1948 | |
| 1949 | if (dentry->d_inode) |
| 1950 | goto eexist; |
| 1951 | /* |
| 1952 | * Special case - lookup gave negative, but... we had foo/bar/ |
| 1953 | * From the vfs_mknod() POV we just have a negative dentry - |
| 1954 | * all is fine. Let's be bastards - you had / on the end, you've |
| 1955 | * been asking for (non-existent) directory. -ENOENT for you. |
| 1956 | */ |
| 1957 | if (unlikely(!is_dir && nd->last.name[nd->last.len])) { |
| 1958 | dput(dentry); |
| 1959 | dentry = ERR_PTR(-ENOENT); |
| 1960 | } |
| 1961 | return dentry; |
| 1962 | eexist: |
| 1963 | dput(dentry); |
| 1964 | dentry = ERR_PTR(-EEXIST); |
| 1965 | fail: |
| 1966 | return dentry; |
| 1967 | } |
| 1968 | EXPORT_SYMBOL_GPL(lookup_create); |
| 1969 | |
| 1970 | int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) |
| 1971 | { |
| 1972 | int error = may_create(dir, dentry); |
| 1973 | |
| 1974 | if (error) |
| 1975 | return error; |
| 1976 | |
| 1977 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) |
| 1978 | return -EPERM; |
| 1979 | |
| 1980 | if (!dir->i_op->mknod) |
| 1981 | return -EPERM; |
| 1982 | |
| 1983 | error = devcgroup_inode_mknod(mode, dev); |
| 1984 | if (error) |
| 1985 | return error; |
| 1986 | |
| 1987 | error = security_inode_mknod(dir, dentry, mode, dev); |
| 1988 | if (error) |
| 1989 | return error; |
| 1990 | |
| 1991 | vfs_dq_init(dir); |
| 1992 | error = dir->i_op->mknod(dir, dentry, mode, dev); |
| 1993 | if (!error) |
| 1994 | fsnotify_create(dir, dentry); |
| 1995 | return error; |
| 1996 | } |
| 1997 | |
| 1998 | static int may_mknod(mode_t mode) |
| 1999 | { |
| 2000 | switch (mode & S_IFMT) { |
| 2001 | case S_IFREG: |
| 2002 | case S_IFCHR: |
| 2003 | case S_IFBLK: |
| 2004 | case S_IFIFO: |
| 2005 | case S_IFSOCK: |
| 2006 | case 0: /* zero mode translates to S_IFREG */ |
| 2007 | return 0; |
| 2008 | case S_IFDIR: |
| 2009 | return -EPERM; |
| 2010 | default: |
| 2011 | return -EINVAL; |
| 2012 | } |
| 2013 | } |
| 2014 | |
| 2015 | SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, |
| 2016 | unsigned, dev) |
| 2017 | { |
| 2018 | int error; |
| 2019 | char *tmp; |
| 2020 | struct dentry *dentry; |
| 2021 | struct nameidata nd; |
| 2022 | |
| 2023 | if (S_ISDIR(mode)) |
| 2024 | return -EPERM; |
| 2025 | |
| 2026 | error = user_path_parent(dfd, filename, &nd, &tmp); |
| 2027 | if (error) |
| 2028 | return error; |
| 2029 | |
| 2030 | dentry = lookup_create(&nd, 0); |
| 2031 | if (IS_ERR(dentry)) { |
| 2032 | error = PTR_ERR(dentry); |
| 2033 | goto out_unlock; |
| 2034 | } |
| 2035 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
| 2036 | mode &= ~current_umask(); |
| 2037 | error = may_mknod(mode); |
| 2038 | if (error) |
| 2039 | goto out_dput; |
| 2040 | error = mnt_want_write(nd.path.mnt); |
| 2041 | if (error) |
| 2042 | goto out_dput; |
| 2043 | error = security_path_mknod(&nd.path, dentry, mode, dev); |
| 2044 | if (error) |
| 2045 | goto out_drop_write; |
| 2046 | switch (mode & S_IFMT) { |
| 2047 | case 0: case S_IFREG: |
| 2048 | error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); |
| 2049 | break; |
| 2050 | case S_IFCHR: case S_IFBLK: |
| 2051 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, |
| 2052 | new_decode_dev(dev)); |
| 2053 | break; |
| 2054 | case S_IFIFO: case S_IFSOCK: |
| 2055 | error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); |
| 2056 | break; |
| 2057 | } |
| 2058 | out_drop_write: |
| 2059 | mnt_drop_write(nd.path.mnt); |
| 2060 | out_dput: |
| 2061 | dput(dentry); |
| 2062 | out_unlock: |
| 2063 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
| 2064 | path_put(&nd.path); |
| 2065 | putname(tmp); |
| 2066 | |
| 2067 | return error; |
| 2068 | } |
| 2069 | |
| 2070 | SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) |
| 2071 | { |
| 2072 | return sys_mknodat(AT_FDCWD, filename, mode, dev); |
| 2073 | } |
| 2074 | |
| 2075 | int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
| 2076 | { |
| 2077 | int error = may_create(dir, dentry); |
| 2078 | |
| 2079 | if (error) |
| 2080 | return error; |
| 2081 | |
| 2082 | if (!dir->i_op->mkdir) |
| 2083 | return -EPERM; |
| 2084 | |
| 2085 | mode &= (S_IRWXUGO|S_ISVTX); |
| 2086 | error = security_inode_mkdir(dir, dentry, mode); |
| 2087 | if (error) |
| 2088 | return error; |
| 2089 | |
| 2090 | vfs_dq_init(dir); |
| 2091 | error = dir->i_op->mkdir(dir, dentry, mode); |
| 2092 | if (!error) |
| 2093 | fsnotify_mkdir(dir, dentry); |
| 2094 | return error; |
| 2095 | } |
| 2096 | |
| 2097 | SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) |
| 2098 | { |
| 2099 | int error = 0; |
| 2100 | char * tmp; |
| 2101 | struct dentry *dentry; |
| 2102 | struct nameidata nd; |
| 2103 | |
| 2104 | error = user_path_parent(dfd, pathname, &nd, &tmp); |
| 2105 | if (error) |
| 2106 | goto out_err; |
| 2107 | |
| 2108 | dentry = lookup_create(&nd, 1); |
| 2109 | error = PTR_ERR(dentry); |
| 2110 | if (IS_ERR(dentry)) |
| 2111 | goto out_unlock; |
| 2112 | |
| 2113 | if (!IS_POSIXACL(nd.path.dentry->d_inode)) |
| 2114 | mode &= ~current_umask(); |
| 2115 | error = mnt_want_write(nd.path.mnt); |
| 2116 | if (error) |
| 2117 | goto out_dput; |
| 2118 | error = security_path_mkdir(&nd.path, dentry, mode); |
| 2119 | if (error) |
| 2120 | goto out_drop_write; |
| 2121 | error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); |
| 2122 | out_drop_write: |
| 2123 | mnt_drop_write(nd.path.mnt); |
| 2124 | out_dput: |
| 2125 | dput(dentry); |
| 2126 | out_unlock: |
| 2127 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
| 2128 | path_put(&nd.path); |
| 2129 | putname(tmp); |
| 2130 | out_err: |
| 2131 | return error; |
| 2132 | } |
| 2133 | |
| 2134 | SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) |
| 2135 | { |
| 2136 | return sys_mkdirat(AT_FDCWD, pathname, mode); |
| 2137 | } |
| 2138 | |
| 2139 | /* |
| 2140 | * We try to drop the dentry early: we should have |
| 2141 | * a usage count of 2 if we're the only user of this |
| 2142 | * dentry, and if that is true (possibly after pruning |
| 2143 | * the dcache), then we drop the dentry now. |
| 2144 | * |
| 2145 | * A low-level filesystem can, if it choses, legally |
| 2146 | * do a |
| 2147 | * |
| 2148 | * if (!d_unhashed(dentry)) |
| 2149 | * return -EBUSY; |
| 2150 | * |
| 2151 | * if it cannot handle the case of removing a directory |
| 2152 | * that is still in use by something else.. |
| 2153 | */ |
| 2154 | void dentry_unhash(struct dentry *dentry) |
| 2155 | { |
| 2156 | dget(dentry); |
| 2157 | shrink_dcache_parent(dentry); |
| 2158 | spin_lock(&dcache_lock); |
| 2159 | spin_lock(&dentry->d_lock); |
| 2160 | if (atomic_read(&dentry->d_count) == 2) |
| 2161 | __d_drop(dentry); |
| 2162 | spin_unlock(&dentry->d_lock); |
| 2163 | spin_unlock(&dcache_lock); |
| 2164 | } |
| 2165 | |
| 2166 | int vfs_rmdir(struct inode *dir, struct dentry *dentry) |
| 2167 | { |
| 2168 | int error = may_delete(dir, dentry, 1); |
| 2169 | |
| 2170 | if (error) |
| 2171 | return error; |
| 2172 | |
| 2173 | if (!dir->i_op->rmdir) |
| 2174 | return -EPERM; |
| 2175 | |
| 2176 | vfs_dq_init(dir); |
| 2177 | |
| 2178 | mutex_lock(&dentry->d_inode->i_mutex); |
| 2179 | dentry_unhash(dentry); |
| 2180 | if (d_mountpoint(dentry)) |
| 2181 | error = -EBUSY; |
| 2182 | else { |
| 2183 | error = security_inode_rmdir(dir, dentry); |
| 2184 | if (!error) { |
| 2185 | error = dir->i_op->rmdir(dir, dentry); |
| 2186 | if (!error) |
| 2187 | dentry->d_inode->i_flags |= S_DEAD; |
| 2188 | } |
| 2189 | } |
| 2190 | mutex_unlock(&dentry->d_inode->i_mutex); |
| 2191 | if (!error) { |
| 2192 | d_delete(dentry); |
| 2193 | } |
| 2194 | dput(dentry); |
| 2195 | |
| 2196 | return error; |
| 2197 | } |
| 2198 | |
| 2199 | static long do_rmdir(int dfd, const char __user *pathname) |
| 2200 | { |
| 2201 | int error = 0; |
| 2202 | char * name; |
| 2203 | struct dentry *dentry; |
| 2204 | struct nameidata nd; |
| 2205 | |
| 2206 | error = user_path_parent(dfd, pathname, &nd, &name); |
| 2207 | if (error) |
| 2208 | return error; |
| 2209 | |
| 2210 | switch(nd.last_type) { |
| 2211 | case LAST_DOTDOT: |
| 2212 | error = -ENOTEMPTY; |
| 2213 | goto exit1; |
| 2214 | case LAST_DOT: |
| 2215 | error = -EINVAL; |
| 2216 | goto exit1; |
| 2217 | case LAST_ROOT: |
| 2218 | error = -EBUSY; |
| 2219 | goto exit1; |
| 2220 | } |
| 2221 | |
| 2222 | nd.flags &= ~LOOKUP_PARENT; |
| 2223 | |
| 2224 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
| 2225 | dentry = lookup_hash(&nd); |
| 2226 | error = PTR_ERR(dentry); |
| 2227 | if (IS_ERR(dentry)) |
| 2228 | goto exit2; |
| 2229 | error = mnt_want_write(nd.path.mnt); |
| 2230 | if (error) |
| 2231 | goto exit3; |
| 2232 | error = security_path_rmdir(&nd.path, dentry); |
| 2233 | if (error) |
| 2234 | goto exit4; |
| 2235 | error = vfs_rmdir(nd.path.dentry->d_inode, dentry); |
| 2236 | exit4: |
| 2237 | mnt_drop_write(nd.path.mnt); |
| 2238 | exit3: |
| 2239 | dput(dentry); |
| 2240 | exit2: |
| 2241 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
| 2242 | exit1: |
| 2243 | path_put(&nd.path); |
| 2244 | putname(name); |
| 2245 | return error; |
| 2246 | } |
| 2247 | |
| 2248 | SYSCALL_DEFINE1(rmdir, const char __user *, pathname) |
| 2249 | { |
| 2250 | return do_rmdir(AT_FDCWD, pathname); |
| 2251 | } |
| 2252 | |
| 2253 | int vfs_unlink(struct inode *dir, struct dentry *dentry) |
| 2254 | { |
| 2255 | int error = may_delete(dir, dentry, 0); |
| 2256 | |
| 2257 | if (error) |
| 2258 | return error; |
| 2259 | |
| 2260 | if (!dir->i_op->unlink) |
| 2261 | return -EPERM; |
| 2262 | |
| 2263 | vfs_dq_init(dir); |
| 2264 | |
| 2265 | mutex_lock(&dentry->d_inode->i_mutex); |
| 2266 | if (d_mountpoint(dentry)) |
| 2267 | error = -EBUSY; |
| 2268 | else { |
| 2269 | error = security_inode_unlink(dir, dentry); |
| 2270 | if (!error) |
| 2271 | error = dir->i_op->unlink(dir, dentry); |
| 2272 | } |
| 2273 | mutex_unlock(&dentry->d_inode->i_mutex); |
| 2274 | |
| 2275 | /* We don't d_delete() NFS sillyrenamed files--they still exist. */ |
| 2276 | if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { |
| 2277 | fsnotify_link_count(dentry->d_inode); |
| 2278 | d_delete(dentry); |
| 2279 | } |
| 2280 | |
| 2281 | return error; |
| 2282 | } |
| 2283 | |
| 2284 | /* |
| 2285 | * Make sure that the actual truncation of the file will occur outside its |
| 2286 | * directory's i_mutex. Truncate can take a long time if there is a lot of |
| 2287 | * writeout happening, and we don't want to prevent access to the directory |
| 2288 | * while waiting on the I/O. |
| 2289 | */ |
| 2290 | static long do_unlinkat(int dfd, const char __user *pathname) |
| 2291 | { |
| 2292 | int error; |
| 2293 | char *name; |
| 2294 | struct dentry *dentry; |
| 2295 | struct nameidata nd; |
| 2296 | struct inode *inode = NULL; |
| 2297 | |
| 2298 | error = user_path_parent(dfd, pathname, &nd, &name); |
| 2299 | if (error) |
| 2300 | return error; |
| 2301 | |
| 2302 | error = -EISDIR; |
| 2303 | if (nd.last_type != LAST_NORM) |
| 2304 | goto exit1; |
| 2305 | |
| 2306 | nd.flags &= ~LOOKUP_PARENT; |
| 2307 | |
| 2308 | mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); |
| 2309 | dentry = lookup_hash(&nd); |
| 2310 | error = PTR_ERR(dentry); |
| 2311 | if (!IS_ERR(dentry)) { |
| 2312 | /* Why not before? Because we want correct error value */ |
| 2313 | if (nd.last.name[nd.last.len]) |
| 2314 | goto slashes; |
| 2315 | inode = dentry->d_inode; |
| 2316 | if (inode) |
| 2317 | atomic_inc(&inode->i_count); |
| 2318 | error = mnt_want_write(nd.path.mnt); |
| 2319 | if (error) |
| 2320 | goto exit2; |
| 2321 | error = security_path_unlink(&nd.path, dentry); |
| 2322 | if (error) |
| 2323 | goto exit3; |
| 2324 | error = vfs_unlink(nd.path.dentry->d_inode, dentry); |
| 2325 | exit3: |
| 2326 | mnt_drop_write(nd.path.mnt); |
| 2327 | exit2: |
| 2328 | dput(dentry); |
| 2329 | } |
| 2330 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
| 2331 | if (inode) |
| 2332 | iput(inode); /* truncate the inode here */ |
| 2333 | exit1: |
| 2334 | path_put(&nd.path); |
| 2335 | putname(name); |
| 2336 | return error; |
| 2337 | |
| 2338 | slashes: |
| 2339 | error = !dentry->d_inode ? -ENOENT : |
| 2340 | S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; |
| 2341 | goto exit2; |
| 2342 | } |
| 2343 | |
| 2344 | SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) |
| 2345 | { |
| 2346 | if ((flag & ~AT_REMOVEDIR) != 0) |
| 2347 | return -EINVAL; |
| 2348 | |
| 2349 | if (flag & AT_REMOVEDIR) |
| 2350 | return do_rmdir(dfd, pathname); |
| 2351 | |
| 2352 | return do_unlinkat(dfd, pathname); |
| 2353 | } |
| 2354 | |
| 2355 | SYSCALL_DEFINE1(unlink, const char __user *, pathname) |
| 2356 | { |
| 2357 | return do_unlinkat(AT_FDCWD, pathname); |
| 2358 | } |
| 2359 | |
| 2360 | int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) |
| 2361 | { |
| 2362 | int error = may_create(dir, dentry); |
| 2363 | |
| 2364 | if (error) |
| 2365 | return error; |
| 2366 | |
| 2367 | if (!dir->i_op->symlink) |
| 2368 | return -EPERM; |
| 2369 | |
| 2370 | error = security_inode_symlink(dir, dentry, oldname); |
| 2371 | if (error) |
| 2372 | return error; |
| 2373 | |
| 2374 | vfs_dq_init(dir); |
| 2375 | error = dir->i_op->symlink(dir, dentry, oldname); |
| 2376 | if (!error) |
| 2377 | fsnotify_create(dir, dentry); |
| 2378 | return error; |
| 2379 | } |
| 2380 | |
| 2381 | SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, |
| 2382 | int, newdfd, const char __user *, newname) |
| 2383 | { |
| 2384 | int error; |
| 2385 | char *from; |
| 2386 | char *to; |
| 2387 | struct dentry *dentry; |
| 2388 | struct nameidata nd; |
| 2389 | |
| 2390 | from = getname(oldname); |
| 2391 | if (IS_ERR(from)) |
| 2392 | return PTR_ERR(from); |
| 2393 | |
| 2394 | error = user_path_parent(newdfd, newname, &nd, &to); |
| 2395 | if (error) |
| 2396 | goto out_putname; |
| 2397 | |
| 2398 | dentry = lookup_create(&nd, 0); |
| 2399 | error = PTR_ERR(dentry); |
| 2400 | if (IS_ERR(dentry)) |
| 2401 | goto out_unlock; |
| 2402 | |
| 2403 | error = mnt_want_write(nd.path.mnt); |
| 2404 | if (error) |
| 2405 | goto out_dput; |
| 2406 | error = security_path_symlink(&nd.path, dentry, from); |
| 2407 | if (error) |
| 2408 | goto out_drop_write; |
| 2409 | error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); |
| 2410 | out_drop_write: |
| 2411 | mnt_drop_write(nd.path.mnt); |
| 2412 | out_dput: |
| 2413 | dput(dentry); |
| 2414 | out_unlock: |
| 2415 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
| 2416 | path_put(&nd.path); |
| 2417 | putname(to); |
| 2418 | out_putname: |
| 2419 | putname(from); |
| 2420 | return error; |
| 2421 | } |
| 2422 | |
| 2423 | SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) |
| 2424 | { |
| 2425 | return sys_symlinkat(oldname, AT_FDCWD, newname); |
| 2426 | } |
| 2427 | |
| 2428 | int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) |
| 2429 | { |
| 2430 | struct inode *inode = old_dentry->d_inode; |
| 2431 | int error; |
| 2432 | |
| 2433 | if (!inode) |
| 2434 | return -ENOENT; |
| 2435 | |
| 2436 | error = may_create(dir, new_dentry); |
| 2437 | if (error) |
| 2438 | return error; |
| 2439 | |
| 2440 | if (dir->i_sb != inode->i_sb) |
| 2441 | return -EXDEV; |
| 2442 | |
| 2443 | /* |
| 2444 | * A link to an append-only or immutable file cannot be created. |
| 2445 | */ |
| 2446 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) |
| 2447 | return -EPERM; |
| 2448 | if (!dir->i_op->link) |
| 2449 | return -EPERM; |
| 2450 | if (S_ISDIR(inode->i_mode)) |
| 2451 | return -EPERM; |
| 2452 | |
| 2453 | error = security_inode_link(old_dentry, dir, new_dentry); |
| 2454 | if (error) |
| 2455 | return error; |
| 2456 | |
| 2457 | mutex_lock(&inode->i_mutex); |
| 2458 | vfs_dq_init(dir); |
| 2459 | error = dir->i_op->link(old_dentry, dir, new_dentry); |
| 2460 | mutex_unlock(&inode->i_mutex); |
| 2461 | if (!error) |
| 2462 | fsnotify_link(dir, inode, new_dentry); |
| 2463 | return error; |
| 2464 | } |
| 2465 | |
| 2466 | /* |
| 2467 | * Hardlinks are often used in delicate situations. We avoid |
| 2468 | * security-related surprises by not following symlinks on the |
| 2469 | * newname. --KAB |
| 2470 | * |
| 2471 | * We don't follow them on the oldname either to be compatible |
| 2472 | * with linux 2.0, and to avoid hard-linking to directories |
| 2473 | * and other special files. --ADM |
| 2474 | */ |
| 2475 | SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, |
| 2476 | int, newdfd, const char __user *, newname, int, flags) |
| 2477 | { |
| 2478 | struct dentry *new_dentry; |
| 2479 | struct nameidata nd; |
| 2480 | struct path old_path; |
| 2481 | int error; |
| 2482 | char *to; |
| 2483 | |
| 2484 | if ((flags & ~AT_SYMLINK_FOLLOW) != 0) |
| 2485 | return -EINVAL; |
| 2486 | |
| 2487 | error = user_path_at(olddfd, oldname, |
| 2488 | flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, |
| 2489 | &old_path); |
| 2490 | if (error) |
| 2491 | return error; |
| 2492 | |
| 2493 | error = user_path_parent(newdfd, newname, &nd, &to); |
| 2494 | if (error) |
| 2495 | goto out; |
| 2496 | error = -EXDEV; |
| 2497 | if (old_path.mnt != nd.path.mnt) |
| 2498 | goto out_release; |
| 2499 | new_dentry = lookup_create(&nd, 0); |
| 2500 | error = PTR_ERR(new_dentry); |
| 2501 | if (IS_ERR(new_dentry)) |
| 2502 | goto out_unlock; |
| 2503 | error = mnt_want_write(nd.path.mnt); |
| 2504 | if (error) |
| 2505 | goto out_dput; |
| 2506 | error = security_path_link(old_path.dentry, &nd.path, new_dentry); |
| 2507 | if (error) |
| 2508 | goto out_drop_write; |
| 2509 | error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); |
| 2510 | out_drop_write: |
| 2511 | mnt_drop_write(nd.path.mnt); |
| 2512 | out_dput: |
| 2513 | dput(new_dentry); |
| 2514 | out_unlock: |
| 2515 | mutex_unlock(&nd.path.dentry->d_inode->i_mutex); |
| 2516 | out_release: |
| 2517 | path_put(&nd.path); |
| 2518 | putname(to); |
| 2519 | out: |
| 2520 | path_put(&old_path); |
| 2521 | |
| 2522 | return error; |
| 2523 | } |
| 2524 | |
| 2525 | SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) |
| 2526 | { |
| 2527 | return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); |
| 2528 | } |
| 2529 | |
| 2530 | /* |
| 2531 | * The worst of all namespace operations - renaming directory. "Perverted" |
| 2532 | * doesn't even start to describe it. Somebody in UCB had a heck of a trip... |
| 2533 | * Problems: |
| 2534 | * a) we can get into loop creation. Check is done in is_subdir(). |
| 2535 | * b) race potential - two innocent renames can create a loop together. |
| 2536 | * That's where 4.4 screws up. Current fix: serialization on |
| 2537 | * sb->s_vfs_rename_mutex. We might be more accurate, but that's another |
| 2538 | * story. |
| 2539 | * c) we have to lock _three_ objects - parents and victim (if it exists). |
| 2540 | * And that - after we got ->i_mutex on parents (until then we don't know |
| 2541 | * whether the target exists). Solution: try to be smart with locking |
| 2542 | * order for inodes. We rely on the fact that tree topology may change |
| 2543 | * only under ->s_vfs_rename_mutex _and_ that parent of the object we |
| 2544 | * move will be locked. Thus we can rank directories by the tree |
| 2545 | * (ancestors first) and rank all non-directories after them. |
| 2546 | * That works since everybody except rename does "lock parent, lookup, |
| 2547 | * lock child" and rename is under ->s_vfs_rename_mutex. |
| 2548 | * HOWEVER, it relies on the assumption that any object with ->lookup() |
| 2549 | * has no more than 1 dentry. If "hybrid" objects will ever appear, |
| 2550 | * we'd better make sure that there's no link(2) for them. |
| 2551 | * d) some filesystems don't support opened-but-unlinked directories, |
| 2552 | * either because of layout or because they are not ready to deal with |
| 2553 | * all cases correctly. The latter will be fixed (taking this sort of |
| 2554 | * stuff into VFS), but the former is not going away. Solution: the same |
| 2555 | * trick as in rmdir(). |
| 2556 | * e) conversion from fhandle to dentry may come in the wrong moment - when |
| 2557 | * we are removing the target. Solution: we will have to grab ->i_mutex |
| 2558 | * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on |
| 2559 | * ->i_mutex on parents, which works but leads to some truely excessive |
| 2560 | * locking]. |
| 2561 | */ |
| 2562 | static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, |
| 2563 | struct inode *new_dir, struct dentry *new_dentry) |
| 2564 | { |
| 2565 | int error = 0; |
| 2566 | struct inode *target; |
| 2567 | |
| 2568 | /* |
| 2569 | * If we are going to change the parent - check write permissions, |
| 2570 | * we'll need to flip '..'. |
| 2571 | */ |
| 2572 | if (new_dir != old_dir) { |
| 2573 | error = inode_permission(old_dentry->d_inode, MAY_WRITE); |
| 2574 | if (error) |
| 2575 | return error; |
| 2576 | } |
| 2577 | |
| 2578 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); |
| 2579 | if (error) |
| 2580 | return error; |
| 2581 | |
| 2582 | target = new_dentry->d_inode; |
| 2583 | if (target) { |
| 2584 | mutex_lock(&target->i_mutex); |
| 2585 | dentry_unhash(new_dentry); |
| 2586 | } |
| 2587 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
| 2588 | error = -EBUSY; |
| 2589 | else |
| 2590 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
| 2591 | if (target) { |
| 2592 | if (!error) |
| 2593 | target->i_flags |= S_DEAD; |
| 2594 | mutex_unlock(&target->i_mutex); |
| 2595 | if (d_unhashed(new_dentry)) |
| 2596 | d_rehash(new_dentry); |
| 2597 | dput(new_dentry); |
| 2598 | } |
| 2599 | if (!error) |
| 2600 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
| 2601 | d_move(old_dentry,new_dentry); |
| 2602 | return error; |
| 2603 | } |
| 2604 | |
| 2605 | static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, |
| 2606 | struct inode *new_dir, struct dentry *new_dentry) |
| 2607 | { |
| 2608 | struct inode *target; |
| 2609 | int error; |
| 2610 | |
| 2611 | error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); |
| 2612 | if (error) |
| 2613 | return error; |
| 2614 | |
| 2615 | dget(new_dentry); |
| 2616 | target = new_dentry->d_inode; |
| 2617 | if (target) |
| 2618 | mutex_lock(&target->i_mutex); |
| 2619 | if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) |
| 2620 | error = -EBUSY; |
| 2621 | else |
| 2622 | error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); |
| 2623 | if (!error) { |
| 2624 | if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) |
| 2625 | d_move(old_dentry, new_dentry); |
| 2626 | } |
| 2627 | if (target) |
| 2628 | mutex_unlock(&target->i_mutex); |
| 2629 | dput(new_dentry); |
| 2630 | return error; |
| 2631 | } |
| 2632 | |
| 2633 | int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
| 2634 | struct inode *new_dir, struct dentry *new_dentry) |
| 2635 | { |
| 2636 | int error; |
| 2637 | int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); |
| 2638 | const char *old_name; |
| 2639 | |
| 2640 | if (old_dentry->d_inode == new_dentry->d_inode) |
| 2641 | return 0; |
| 2642 | |
| 2643 | error = may_delete(old_dir, old_dentry, is_dir); |
| 2644 | if (error) |
| 2645 | return error; |
| 2646 | |
| 2647 | if (!new_dentry->d_inode) |
| 2648 | error = may_create(new_dir, new_dentry); |
| 2649 | else |
| 2650 | error = may_delete(new_dir, new_dentry, is_dir); |
| 2651 | if (error) |
| 2652 | return error; |
| 2653 | |
| 2654 | if (!old_dir->i_op->rename) |
| 2655 | return -EPERM; |
| 2656 | |
| 2657 | vfs_dq_init(old_dir); |
| 2658 | vfs_dq_init(new_dir); |
| 2659 | |
| 2660 | old_name = fsnotify_oldname_init(old_dentry->d_name.name); |
| 2661 | |
| 2662 | if (is_dir) |
| 2663 | error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); |
| 2664 | else |
| 2665 | error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); |
| 2666 | if (!error) { |
| 2667 | const char *new_name = old_dentry->d_name.name; |
| 2668 | fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir, |
| 2669 | new_dentry->d_inode, old_dentry); |
| 2670 | } |
| 2671 | fsnotify_oldname_free(old_name); |
| 2672 | |
| 2673 | return error; |
| 2674 | } |
| 2675 | |
| 2676 | SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, |
| 2677 | int, newdfd, const char __user *, newname) |
| 2678 | { |
| 2679 | struct dentry *old_dir, *new_dir; |
| 2680 | struct dentry *old_dentry, *new_dentry; |
| 2681 | struct dentry *trap; |
| 2682 | struct nameidata oldnd, newnd; |
| 2683 | char *from; |
| 2684 | char *to; |
| 2685 | int error; |
| 2686 | |
| 2687 | error = user_path_parent(olddfd, oldname, &oldnd, &from); |
| 2688 | if (error) |
| 2689 | goto exit; |
| 2690 | |
| 2691 | error = user_path_parent(newdfd, newname, &newnd, &to); |
| 2692 | if (error) |
| 2693 | goto exit1; |
| 2694 | |
| 2695 | error = -EXDEV; |
| 2696 | if (oldnd.path.mnt != newnd.path.mnt) |
| 2697 | goto exit2; |
| 2698 | |
| 2699 | old_dir = oldnd.path.dentry; |
| 2700 | error = -EBUSY; |
| 2701 | if (oldnd.last_type != LAST_NORM) |
| 2702 | goto exit2; |
| 2703 | |
| 2704 | new_dir = newnd.path.dentry; |
| 2705 | if (newnd.last_type != LAST_NORM) |
| 2706 | goto exit2; |
| 2707 | |
| 2708 | oldnd.flags &= ~LOOKUP_PARENT; |
| 2709 | newnd.flags &= ~LOOKUP_PARENT; |
| 2710 | newnd.flags |= LOOKUP_RENAME_TARGET; |
| 2711 | |
| 2712 | trap = lock_rename(new_dir, old_dir); |
| 2713 | |
| 2714 | old_dentry = lookup_hash(&oldnd); |
| 2715 | error = PTR_ERR(old_dentry); |
| 2716 | if (IS_ERR(old_dentry)) |
| 2717 | goto exit3; |
| 2718 | /* source must exist */ |
| 2719 | error = -ENOENT; |
| 2720 | if (!old_dentry->d_inode) |
| 2721 | goto exit4; |
| 2722 | /* unless the source is a directory trailing slashes give -ENOTDIR */ |
| 2723 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) { |
| 2724 | error = -ENOTDIR; |
| 2725 | if (oldnd.last.name[oldnd.last.len]) |
| 2726 | goto exit4; |
| 2727 | if (newnd.last.name[newnd.last.len]) |
| 2728 | goto exit4; |
| 2729 | } |
| 2730 | /* source should not be ancestor of target */ |
| 2731 | error = -EINVAL; |
| 2732 | if (old_dentry == trap) |
| 2733 | goto exit4; |
| 2734 | new_dentry = lookup_hash(&newnd); |
| 2735 | error = PTR_ERR(new_dentry); |
| 2736 | if (IS_ERR(new_dentry)) |
| 2737 | goto exit4; |
| 2738 | /* target should not be an ancestor of source */ |
| 2739 | error = -ENOTEMPTY; |
| 2740 | if (new_dentry == trap) |
| 2741 | goto exit5; |
| 2742 | |
| 2743 | error = mnt_want_write(oldnd.path.mnt); |
| 2744 | if (error) |
| 2745 | goto exit5; |
| 2746 | error = security_path_rename(&oldnd.path, old_dentry, |
| 2747 | &newnd.path, new_dentry); |
| 2748 | if (error) |
| 2749 | goto exit6; |
| 2750 | error = vfs_rename(old_dir->d_inode, old_dentry, |
| 2751 | new_dir->d_inode, new_dentry); |
| 2752 | exit6: |
| 2753 | mnt_drop_write(oldnd.path.mnt); |
| 2754 | exit5: |
| 2755 | dput(new_dentry); |
| 2756 | exit4: |
| 2757 | dput(old_dentry); |
| 2758 | exit3: |
| 2759 | unlock_rename(new_dir, old_dir); |
| 2760 | exit2: |
| 2761 | path_put(&newnd.path); |
| 2762 | putname(to); |
| 2763 | exit1: |
| 2764 | path_put(&oldnd.path); |
| 2765 | putname(from); |
| 2766 | exit: |
| 2767 | return error; |
| 2768 | } |
| 2769 | |
| 2770 | SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) |
| 2771 | { |
| 2772 | return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); |
| 2773 | } |
| 2774 | |
| 2775 | int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) |
| 2776 | { |
| 2777 | int len; |
| 2778 | |
| 2779 | len = PTR_ERR(link); |
| 2780 | if (IS_ERR(link)) |
| 2781 | goto out; |
| 2782 | |
| 2783 | len = strlen(link); |
| 2784 | if (len > (unsigned) buflen) |
| 2785 | len = buflen; |
| 2786 | if (copy_to_user(buffer, link, len)) |
| 2787 | len = -EFAULT; |
| 2788 | out: |
| 2789 | return len; |
| 2790 | } |
| 2791 | |
| 2792 | /* |
| 2793 | * A helper for ->readlink(). This should be used *ONLY* for symlinks that |
| 2794 | * have ->follow_link() touching nd only in nd_set_link(). Using (or not |
| 2795 | * using) it for any given inode is up to filesystem. |
| 2796 | */ |
| 2797 | int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
| 2798 | { |
| 2799 | struct nameidata nd; |
| 2800 | void *cookie; |
| 2801 | int res; |
| 2802 | |
| 2803 | nd.depth = 0; |
| 2804 | cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); |
| 2805 | if (IS_ERR(cookie)) |
| 2806 | return PTR_ERR(cookie); |
| 2807 | |
| 2808 | res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); |
| 2809 | if (dentry->d_inode->i_op->put_link) |
| 2810 | dentry->d_inode->i_op->put_link(dentry, &nd, cookie); |
| 2811 | return res; |
| 2812 | } |
| 2813 | |
| 2814 | int vfs_follow_link(struct nameidata *nd, const char *link) |
| 2815 | { |
| 2816 | return __vfs_follow_link(nd, link); |
| 2817 | } |
| 2818 | |
| 2819 | /* get the link contents into pagecache */ |
| 2820 | static char *page_getlink(struct dentry * dentry, struct page **ppage) |
| 2821 | { |
| 2822 | char *kaddr; |
| 2823 | struct page *page; |
| 2824 | struct address_space *mapping = dentry->d_inode->i_mapping; |
| 2825 | page = read_mapping_page(mapping, 0, NULL); |
| 2826 | if (IS_ERR(page)) |
| 2827 | return (char*)page; |
| 2828 | *ppage = page; |
| 2829 | kaddr = kmap(page); |
| 2830 | nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); |
| 2831 | return kaddr; |
| 2832 | } |
| 2833 | |
| 2834 | int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) |
| 2835 | { |
| 2836 | struct page *page = NULL; |
| 2837 | char *s = page_getlink(dentry, &page); |
| 2838 | int res = vfs_readlink(dentry,buffer,buflen,s); |
| 2839 | if (page) { |
| 2840 | kunmap(page); |
| 2841 | page_cache_release(page); |
| 2842 | } |
| 2843 | return res; |
| 2844 | } |
| 2845 | |
| 2846 | void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) |
| 2847 | { |
| 2848 | struct page *page = NULL; |
| 2849 | nd_set_link(nd, page_getlink(dentry, &page)); |
| 2850 | return page; |
| 2851 | } |
| 2852 | |
| 2853 | void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) |
| 2854 | { |
| 2855 | struct page *page = cookie; |
| 2856 | |
| 2857 | if (page) { |
| 2858 | kunmap(page); |
| 2859 | page_cache_release(page); |
| 2860 | } |
| 2861 | } |
| 2862 | |
| 2863 | /* |
| 2864 | * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS |
| 2865 | */ |
| 2866 | int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) |
| 2867 | { |
| 2868 | struct address_space *mapping = inode->i_mapping; |
| 2869 | struct page *page; |
| 2870 | void *fsdata; |
| 2871 | int err; |
| 2872 | char *kaddr; |
| 2873 | unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; |
| 2874 | if (nofs) |
| 2875 | flags |= AOP_FLAG_NOFS; |
| 2876 | |
| 2877 | retry: |
| 2878 | err = pagecache_write_begin(NULL, mapping, 0, len-1, |
| 2879 | flags, &page, &fsdata); |
| 2880 | if (err) |
| 2881 | goto fail; |
| 2882 | |
| 2883 | kaddr = kmap_atomic(page, KM_USER0); |
| 2884 | memcpy(kaddr, symname, len-1); |
| 2885 | kunmap_atomic(kaddr, KM_USER0); |
| 2886 | |
| 2887 | err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, |
| 2888 | page, fsdata); |
| 2889 | if (err < 0) |
| 2890 | goto fail; |
| 2891 | if (err < len-1) |
| 2892 | goto retry; |
| 2893 | |
| 2894 | mark_inode_dirty(inode); |
| 2895 | return 0; |
| 2896 | fail: |
| 2897 | return err; |
| 2898 | } |
| 2899 | |
| 2900 | int page_symlink(struct inode *inode, const char *symname, int len) |
| 2901 | { |
| 2902 | return __page_symlink(inode, symname, len, |
| 2903 | !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); |
| 2904 | } |
| 2905 | |
| 2906 | const struct inode_operations page_symlink_inode_operations = { |
| 2907 | .readlink = generic_readlink, |
| 2908 | .follow_link = page_follow_link_light, |
| 2909 | .put_link = page_put_link, |
| 2910 | }; |
| 2911 | |
| 2912 | EXPORT_SYMBOL(user_path_at); |
| 2913 | EXPORT_SYMBOL(follow_down); |
| 2914 | EXPORT_SYMBOL(follow_up); |
| 2915 | EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ |
| 2916 | EXPORT_SYMBOL(getname); |
| 2917 | EXPORT_SYMBOL(lock_rename); |
| 2918 | EXPORT_SYMBOL(lookup_one_len); |
| 2919 | EXPORT_SYMBOL(page_follow_link_light); |
| 2920 | EXPORT_SYMBOL(page_put_link); |
| 2921 | EXPORT_SYMBOL(page_readlink); |
| 2922 | EXPORT_SYMBOL(__page_symlink); |
| 2923 | EXPORT_SYMBOL(page_symlink); |
| 2924 | EXPORT_SYMBOL(page_symlink_inode_operations); |
| 2925 | EXPORT_SYMBOL(path_lookup); |
| 2926 | EXPORT_SYMBOL(kern_path); |
| 2927 | EXPORT_SYMBOL(vfs_path_lookup); |
| 2928 | EXPORT_SYMBOL(inode_permission); |
| 2929 | EXPORT_SYMBOL(file_permission); |
| 2930 | EXPORT_SYMBOL(unlock_rename); |
| 2931 | EXPORT_SYMBOL(vfs_create); |
| 2932 | EXPORT_SYMBOL(vfs_follow_link); |
| 2933 | EXPORT_SYMBOL(vfs_link); |
| 2934 | EXPORT_SYMBOL(vfs_mkdir); |
| 2935 | EXPORT_SYMBOL(vfs_mknod); |
| 2936 | EXPORT_SYMBOL(generic_permission); |
| 2937 | EXPORT_SYMBOL(vfs_readlink); |
| 2938 | EXPORT_SYMBOL(vfs_rename); |
| 2939 | EXPORT_SYMBOL(vfs_rmdir); |
| 2940 | EXPORT_SYMBOL(vfs_symlink); |
| 2941 | EXPORT_SYMBOL(vfs_unlink); |
| 2942 | EXPORT_SYMBOL(dentry_unhash); |
| 2943 | EXPORT_SYMBOL(generic_readlink); |