Root/fs/open.c

1/*
2 * linux/fs/open.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/string.h>
8#include <linux/mm.h>
9#include <linux/file.h>
10#include <linux/fdtable.h>
11#include <linux/fsnotify.h>
12#include <linux/module.h>
13#include <linux/tty.h>
14#include <linux/namei.h>
15#include <linux/backing-dev.h>
16#include <linux/capability.h>
17#include <linux/securebits.h>
18#include <linux/security.h>
19#include <linux/mount.h>
20#include <linux/vfs.h>
21#include <linux/fcntl.h>
22#include <linux/slab.h>
23#include <asm/uaccess.h>
24#include <linux/fs.h>
25#include <linux/personality.h>
26#include <linux/pagemap.h>
27#include <linux/syscalls.h>
28#include <linux/rcupdate.h>
29#include <linux/audit.h>
30#include <linux/falloc.h>
31#include <linux/fs_struct.h>
32#include <linux/ima.h>
33
34#include "internal.h"
35
36int vfs_statfs(struct dentry *dentry, struct kstatfs *buf)
37{
38    int retval = -ENODEV;
39
40    if (dentry) {
41        retval = -ENOSYS;
42        if (dentry->d_sb->s_op->statfs) {
43            memset(buf, 0, sizeof(*buf));
44            retval = security_sb_statfs(dentry);
45            if (retval)
46                return retval;
47            retval = dentry->d_sb->s_op->statfs(dentry, buf);
48            if (retval == 0 && buf->f_frsize == 0)
49                buf->f_frsize = buf->f_bsize;
50        }
51    }
52    return retval;
53}
54
55EXPORT_SYMBOL(vfs_statfs);
56
57static int vfs_statfs_native(struct dentry *dentry, struct statfs *buf)
58{
59    struct kstatfs st;
60    int retval;
61
62    retval = vfs_statfs(dentry, &st);
63    if (retval)
64        return retval;
65
66    if (sizeof(*buf) == sizeof(st))
67        memcpy(buf, &st, sizeof(st));
68    else {
69        if (sizeof buf->f_blocks == 4) {
70            if ((st.f_blocks | st.f_bfree | st.f_bavail |
71                 st.f_bsize | st.f_frsize) &
72                0xffffffff00000000ULL)
73                return -EOVERFLOW;
74            /*
75             * f_files and f_ffree may be -1; it's okay to stuff
76             * that into 32 bits
77             */
78            if (st.f_files != -1 &&
79                (st.f_files & 0xffffffff00000000ULL))
80                return -EOVERFLOW;
81            if (st.f_ffree != -1 &&
82                (st.f_ffree & 0xffffffff00000000ULL))
83                return -EOVERFLOW;
84        }
85
86        buf->f_type = st.f_type;
87        buf->f_bsize = st.f_bsize;
88        buf->f_blocks = st.f_blocks;
89        buf->f_bfree = st.f_bfree;
90        buf->f_bavail = st.f_bavail;
91        buf->f_files = st.f_files;
92        buf->f_ffree = st.f_ffree;
93        buf->f_fsid = st.f_fsid;
94        buf->f_namelen = st.f_namelen;
95        buf->f_frsize = st.f_frsize;
96        memset(buf->f_spare, 0, sizeof(buf->f_spare));
97    }
98    return 0;
99}
100
101static int vfs_statfs64(struct dentry *dentry, struct statfs64 *buf)
102{
103    struct kstatfs st;
104    int retval;
105
106    retval = vfs_statfs(dentry, &st);
107    if (retval)
108        return retval;
109
110    if (sizeof(*buf) == sizeof(st))
111        memcpy(buf, &st, sizeof(st));
112    else {
113        buf->f_type = st.f_type;
114        buf->f_bsize = st.f_bsize;
115        buf->f_blocks = st.f_blocks;
116        buf->f_bfree = st.f_bfree;
117        buf->f_bavail = st.f_bavail;
118        buf->f_files = st.f_files;
119        buf->f_ffree = st.f_ffree;
120        buf->f_fsid = st.f_fsid;
121        buf->f_namelen = st.f_namelen;
122        buf->f_frsize = st.f_frsize;
123        memset(buf->f_spare, 0, sizeof(buf->f_spare));
124    }
125    return 0;
126}
127
128SYSCALL_DEFINE2(statfs, const char __user *, pathname, struct statfs __user *, buf)
129{
130    struct path path;
131    int error;
132
133    error = user_path(pathname, &path);
134    if (!error) {
135        struct statfs tmp;
136        error = vfs_statfs_native(path.dentry, &tmp);
137        if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
138            error = -EFAULT;
139        path_put(&path);
140    }
141    return error;
142}
143
144SYSCALL_DEFINE3(statfs64, const char __user *, pathname, size_t, sz, struct statfs64 __user *, buf)
145{
146    struct path path;
147    long error;
148
149    if (sz != sizeof(*buf))
150        return -EINVAL;
151    error = user_path(pathname, &path);
152    if (!error) {
153        struct statfs64 tmp;
154        error = vfs_statfs64(path.dentry, &tmp);
155        if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
156            error = -EFAULT;
157        path_put(&path);
158    }
159    return error;
160}
161
162SYSCALL_DEFINE2(fstatfs, unsigned int, fd, struct statfs __user *, buf)
163{
164    struct file * file;
165    struct statfs tmp;
166    int error;
167
168    error = -EBADF;
169    file = fget(fd);
170    if (!file)
171        goto out;
172    error = vfs_statfs_native(file->f_path.dentry, &tmp);
173    if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
174        error = -EFAULT;
175    fput(file);
176out:
177    return error;
178}
179
180SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, size_t, sz, struct statfs64 __user *, buf)
181{
182    struct file * file;
183    struct statfs64 tmp;
184    int error;
185
186    if (sz != sizeof(*buf))
187        return -EINVAL;
188
189    error = -EBADF;
190    file = fget(fd);
191    if (!file)
192        goto out;
193    error = vfs_statfs64(file->f_path.dentry, &tmp);
194    if (!error && copy_to_user(buf, &tmp, sizeof(tmp)))
195        error = -EFAULT;
196    fput(file);
197out:
198    return error;
199}
200
201int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
202    struct file *filp)
203{
204    int ret;
205    struct iattr newattrs;
206
207    /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
208    if (length < 0)
209        return -EINVAL;
210
211    newattrs.ia_size = length;
212    newattrs.ia_valid = ATTR_SIZE | time_attrs;
213    if (filp) {
214        newattrs.ia_file = filp;
215        newattrs.ia_valid |= ATTR_FILE;
216    }
217
218    /* Remove suid/sgid on truncate too */
219    ret = should_remove_suid(dentry);
220    if (ret)
221        newattrs.ia_valid |= ret | ATTR_FORCE;
222
223    mutex_lock(&dentry->d_inode->i_mutex);
224    ret = notify_change(dentry, &newattrs);
225    mutex_unlock(&dentry->d_inode->i_mutex);
226    return ret;
227}
228
229static long do_sys_truncate(const char __user *pathname, loff_t length)
230{
231    struct path path;
232    struct inode *inode;
233    int error;
234
235    error = -EINVAL;
236    if (length < 0) /* sorry, but loff_t says... */
237        goto out;
238
239    error = user_path(pathname, &path);
240    if (error)
241        goto out;
242    inode = path.dentry->d_inode;
243
244    /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
245    error = -EISDIR;
246    if (S_ISDIR(inode->i_mode))
247        goto dput_and_out;
248
249    error = -EINVAL;
250    if (!S_ISREG(inode->i_mode))
251        goto dput_and_out;
252
253    error = mnt_want_write(path.mnt);
254    if (error)
255        goto dput_and_out;
256
257    error = inode_permission(inode, MAY_WRITE);
258    if (error)
259        goto mnt_drop_write_and_out;
260
261    error = -EPERM;
262    if (IS_APPEND(inode))
263        goto mnt_drop_write_and_out;
264
265    error = get_write_access(inode);
266    if (error)
267        goto mnt_drop_write_and_out;
268
269    /*
270     * Make sure that there are no leases. get_write_access() protects
271     * against the truncate racing with a lease-granting setlease().
272     */
273    error = break_lease(inode, O_WRONLY);
274    if (error)
275        goto put_write_and_out;
276
277    error = locks_verify_truncate(inode, NULL, length);
278    if (!error)
279        error = security_path_truncate(&path, length, 0);
280    if (!error)
281        error = do_truncate(path.dentry, length, 0, NULL);
282
283put_write_and_out:
284    put_write_access(inode);
285mnt_drop_write_and_out:
286    mnt_drop_write(path.mnt);
287dput_and_out:
288    path_put(&path);
289out:
290    return error;
291}
292
293SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
294{
295    return do_sys_truncate(path, length);
296}
297
298static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
299{
300    struct inode * inode;
301    struct dentry *dentry;
302    struct file * file;
303    int error;
304
305    error = -EINVAL;
306    if (length < 0)
307        goto out;
308    error = -EBADF;
309    file = fget(fd);
310    if (!file)
311        goto out;
312
313    /* explicitly opened as large or we are on 64-bit box */
314    if (file->f_flags & O_LARGEFILE)
315        small = 0;
316
317    dentry = file->f_path.dentry;
318    inode = dentry->d_inode;
319    error = -EINVAL;
320    if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
321        goto out_putf;
322
323    error = -EINVAL;
324    /* Cannot ftruncate over 2^31 bytes without large file support */
325    if (small && length > MAX_NON_LFS)
326        goto out_putf;
327
328    error = -EPERM;
329    if (IS_APPEND(inode))
330        goto out_putf;
331
332    error = locks_verify_truncate(inode, file, length);
333    if (!error)
334        error = security_path_truncate(&file->f_path, length,
335                           ATTR_MTIME|ATTR_CTIME);
336    if (!error)
337        error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
338out_putf:
339    fput(file);
340out:
341    return error;
342}
343
344SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
345{
346    long ret = do_sys_ftruncate(fd, length, 1);
347    /* avoid REGPARM breakage on x86: */
348    asmlinkage_protect(2, ret, fd, length);
349    return ret;
350}
351
352/* LFS versions of truncate are only needed on 32 bit machines */
353#if BITS_PER_LONG == 32
354SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
355{
356    return do_sys_truncate(path, length);
357}
358#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
359asmlinkage long SyS_truncate64(long path, loff_t length)
360{
361    return SYSC_truncate64((const char __user *) path, length);
362}
363SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
364#endif
365
366SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
367{
368    long ret = do_sys_ftruncate(fd, length, 0);
369    /* avoid REGPARM breakage on x86: */
370    asmlinkage_protect(2, ret, fd, length);
371    return ret;
372}
373#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
374asmlinkage long SyS_ftruncate64(long fd, loff_t length)
375{
376    return SYSC_ftruncate64((unsigned int) fd, length);
377}
378SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
379#endif
380#endif /* BITS_PER_LONG == 32 */
381
382
383int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
384{
385    struct inode *inode = file->f_path.dentry->d_inode;
386    long ret;
387
388    if (offset < 0 || len <= 0)
389        return -EINVAL;
390
391    /* Return error if mode is not supported */
392    if (mode && !(mode & FALLOC_FL_KEEP_SIZE))
393        return -EOPNOTSUPP;
394
395    if (!(file->f_mode & FMODE_WRITE))
396        return -EBADF;
397    /*
398     * Revalidate the write permissions, in case security policy has
399     * changed since the files were opened.
400     */
401    ret = security_file_permission(file, MAY_WRITE);
402    if (ret)
403        return ret;
404
405    if (S_ISFIFO(inode->i_mode))
406        return -ESPIPE;
407
408    /*
409     * Let individual file system decide if it supports preallocation
410     * for directories or not.
411     */
412    if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
413        return -ENODEV;
414
415    /* Check for wrap through zero too */
416    if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
417        return -EFBIG;
418
419    if (!inode->i_op->fallocate)
420        return -EOPNOTSUPP;
421
422    return inode->i_op->fallocate(inode, mode, offset, len);
423}
424
425SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
426{
427    struct file *file;
428    int error = -EBADF;
429
430    file = fget(fd);
431    if (file) {
432        error = do_fallocate(file, mode, offset, len);
433        fput(file);
434    }
435
436    return error;
437}
438
439#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
440asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
441{
442    return SYSC_fallocate((int)fd, (int)mode, offset, len);
443}
444SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
445#endif
446
447/*
448 * access() needs to use the real uid/gid, not the effective uid/gid.
449 * We do this by temporarily clearing all FS-related capabilities and
450 * switching the fsuid/fsgid around to the real ones.
451 */
452SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
453{
454    const struct cred *old_cred;
455    struct cred *override_cred;
456    struct path path;
457    struct inode *inode;
458    int res;
459
460    if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
461        return -EINVAL;
462
463    override_cred = prepare_creds();
464    if (!override_cred)
465        return -ENOMEM;
466
467    override_cred->fsuid = override_cred->uid;
468    override_cred->fsgid = override_cred->gid;
469
470    if (!issecure(SECURE_NO_SETUID_FIXUP)) {
471        /* Clear the capabilities if we switch to a non-root user */
472        if (override_cred->uid)
473            cap_clear(override_cred->cap_effective);
474        else
475            override_cred->cap_effective =
476                override_cred->cap_permitted;
477    }
478
479    old_cred = override_creds(override_cred);
480
481    res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
482    if (res)
483        goto out;
484
485    inode = path.dentry->d_inode;
486
487    if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
488        /*
489         * MAY_EXEC on regular files is denied if the fs is mounted
490         * with the "noexec" flag.
491         */
492        res = -EACCES;
493        if (path.mnt->mnt_flags & MNT_NOEXEC)
494            goto out_path_release;
495    }
496
497    res = inode_permission(inode, mode | MAY_ACCESS);
498    /* SuS v2 requires we report a read only fs too */
499    if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
500        goto out_path_release;
501    /*
502     * This is a rare case where using __mnt_is_readonly()
503     * is OK without a mnt_want/drop_write() pair. Since
504     * no actual write to the fs is performed here, we do
505     * not need to telegraph to that to anyone.
506     *
507     * By doing this, we accept that this access is
508     * inherently racy and know that the fs may change
509     * state before we even see this result.
510     */
511    if (__mnt_is_readonly(path.mnt))
512        res = -EROFS;
513
514out_path_release:
515    path_put(&path);
516out:
517    revert_creds(old_cred);
518    put_cred(override_cred);
519    return res;
520}
521
522SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
523{
524    return sys_faccessat(AT_FDCWD, filename, mode);
525}
526
527SYSCALL_DEFINE1(chdir, const char __user *, filename)
528{
529    struct path path;
530    int error;
531
532    error = user_path_dir(filename, &path);
533    if (error)
534        goto out;
535
536    error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
537    if (error)
538        goto dput_and_out;
539
540    set_fs_pwd(current->fs, &path);
541
542dput_and_out:
543    path_put(&path);
544out:
545    return error;
546}
547
548SYSCALL_DEFINE1(fchdir, unsigned int, fd)
549{
550    struct file *file;
551    struct inode *inode;
552    int error;
553
554    error = -EBADF;
555    file = fget(fd);
556    if (!file)
557        goto out;
558
559    inode = file->f_path.dentry->d_inode;
560
561    error = -ENOTDIR;
562    if (!S_ISDIR(inode->i_mode))
563        goto out_putf;
564
565    error = inode_permission(inode, MAY_EXEC | MAY_ACCESS);
566    if (!error)
567        set_fs_pwd(current->fs, &file->f_path);
568out_putf:
569    fput(file);
570out:
571    return error;
572}
573
574SYSCALL_DEFINE1(chroot, const char __user *, filename)
575{
576    struct path path;
577    int error;
578
579    error = user_path_dir(filename, &path);
580    if (error)
581        goto out;
582
583    error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
584    if (error)
585        goto dput_and_out;
586
587    error = -EPERM;
588    if (!capable(CAP_SYS_CHROOT))
589        goto dput_and_out;
590    error = security_path_chroot(&path);
591    if (error)
592        goto dput_and_out;
593
594    set_fs_root(current->fs, &path);
595    error = 0;
596dput_and_out:
597    path_put(&path);
598out:
599    return error;
600}
601
602SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
603{
604    struct inode * inode;
605    struct dentry * dentry;
606    struct file * file;
607    int err = -EBADF;
608    struct iattr newattrs;
609
610    file = fget(fd);
611    if (!file)
612        goto out;
613
614    dentry = file->f_path.dentry;
615    inode = dentry->d_inode;
616
617    audit_inode(NULL, dentry);
618
619    err = mnt_want_write_file(file);
620    if (err)
621        goto out_putf;
622    mutex_lock(&inode->i_mutex);
623    err = security_path_chmod(dentry, file->f_vfsmnt, mode);
624    if (err)
625        goto out_unlock;
626    if (mode == (mode_t) -1)
627        mode = inode->i_mode;
628    newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
629    newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
630    err = notify_change(dentry, &newattrs);
631out_unlock:
632    mutex_unlock(&inode->i_mutex);
633    mnt_drop_write(file->f_path.mnt);
634out_putf:
635    fput(file);
636out:
637    return err;
638}
639
640SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
641{
642    struct path path;
643    struct inode *inode;
644    int error;
645    struct iattr newattrs;
646
647    error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
648    if (error)
649        goto out;
650    inode = path.dentry->d_inode;
651
652    error = mnt_want_write(path.mnt);
653    if (error)
654        goto dput_and_out;
655    mutex_lock(&inode->i_mutex);
656    error = security_path_chmod(path.dentry, path.mnt, mode);
657    if (error)
658        goto out_unlock;
659    if (mode == (mode_t) -1)
660        mode = inode->i_mode;
661    newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
662    newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
663    error = notify_change(path.dentry, &newattrs);
664out_unlock:
665    mutex_unlock(&inode->i_mutex);
666    mnt_drop_write(path.mnt);
667dput_and_out:
668    path_put(&path);
669out:
670    return error;
671}
672
673SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
674{
675    return sys_fchmodat(AT_FDCWD, filename, mode);
676}
677
678static int chown_common(struct path *path, uid_t user, gid_t group)
679{
680    struct inode *inode = path->dentry->d_inode;
681    int error;
682    struct iattr newattrs;
683
684    newattrs.ia_valid = ATTR_CTIME;
685    if (user != (uid_t) -1) {
686        newattrs.ia_valid |= ATTR_UID;
687        newattrs.ia_uid = user;
688    }
689    if (group != (gid_t) -1) {
690        newattrs.ia_valid |= ATTR_GID;
691        newattrs.ia_gid = group;
692    }
693    if (!S_ISDIR(inode->i_mode))
694        newattrs.ia_valid |=
695            ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
696    mutex_lock(&inode->i_mutex);
697    error = security_path_chown(path, user, group);
698    if (!error)
699        error = notify_change(path->dentry, &newattrs);
700    mutex_unlock(&inode->i_mutex);
701
702    return error;
703}
704
705SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
706{
707    struct path path;
708    int error;
709
710    error = user_path(filename, &path);
711    if (error)
712        goto out;
713    error = mnt_want_write(path.mnt);
714    if (error)
715        goto out_release;
716    error = chown_common(&path, user, group);
717    mnt_drop_write(path.mnt);
718out_release:
719    path_put(&path);
720out:
721    return error;
722}
723
724SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
725        gid_t, group, int, flag)
726{
727    struct path path;
728    int error = -EINVAL;
729    int follow;
730
731    if ((flag & ~AT_SYMLINK_NOFOLLOW) != 0)
732        goto out;
733
734    follow = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
735    error = user_path_at(dfd, filename, follow, &path);
736    if (error)
737        goto out;
738    error = mnt_want_write(path.mnt);
739    if (error)
740        goto out_release;
741    error = chown_common(&path, user, group);
742    mnt_drop_write(path.mnt);
743out_release:
744    path_put(&path);
745out:
746    return error;
747}
748
749SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
750{
751    struct path path;
752    int error;
753
754    error = user_lpath(filename, &path);
755    if (error)
756        goto out;
757    error = mnt_want_write(path.mnt);
758    if (error)
759        goto out_release;
760    error = chown_common(&path, user, group);
761    mnt_drop_write(path.mnt);
762out_release:
763    path_put(&path);
764out:
765    return error;
766}
767
768SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
769{
770    struct file * file;
771    int error = -EBADF;
772    struct dentry * dentry;
773
774    file = fget(fd);
775    if (!file)
776        goto out;
777
778    error = mnt_want_write_file(file);
779    if (error)
780        goto out_fput;
781    dentry = file->f_path.dentry;
782    audit_inode(NULL, dentry);
783    error = chown_common(&file->f_path, user, group);
784    mnt_drop_write(file->f_path.mnt);
785out_fput:
786    fput(file);
787out:
788    return error;
789}
790
791/*
792 * You have to be very careful that these write
793 * counts get cleaned up in error cases and
794 * upon __fput(). This should probably never
795 * be called outside of __dentry_open().
796 */
797static inline int __get_file_write_access(struct inode *inode,
798                      struct vfsmount *mnt)
799{
800    int error;
801    error = get_write_access(inode);
802    if (error)
803        return error;
804    /*
805     * Do not take mount writer counts on
806     * special files since no writes to
807     * the mount itself will occur.
808     */
809    if (!special_file(inode->i_mode)) {
810        /*
811         * Balanced in __fput()
812         */
813        error = mnt_want_write(mnt);
814        if (error)
815            put_write_access(inode);
816    }
817    return error;
818}
819
820static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
821                    struct file *f,
822                    int (*open)(struct inode *, struct file *),
823                    const struct cred *cred)
824{
825    struct inode *inode;
826    int error;
827
828    f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
829                FMODE_PREAD | FMODE_PWRITE;
830    inode = dentry->d_inode;
831    if (f->f_mode & FMODE_WRITE) {
832        error = __get_file_write_access(inode, mnt);
833        if (error)
834            goto cleanup_file;
835        if (!special_file(inode->i_mode))
836            file_take_write(f);
837    }
838
839    f->f_mapping = inode->i_mapping;
840    f->f_path.dentry = dentry;
841    f->f_path.mnt = mnt;
842    f->f_pos = 0;
843    f->f_op = fops_get(inode->i_fop);
844    file_move(f, &inode->i_sb->s_files);
845
846    error = security_dentry_open(f, cred);
847    if (error)
848        goto cleanup_all;
849
850    if (!open && f->f_op)
851        open = f->f_op->open;
852    if (open) {
853        error = open(inode, f);
854        if (error)
855            goto cleanup_all;
856    }
857    ima_counts_get(f);
858
859    f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
860
861    file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
862
863    /* NB: we're sure to have correct a_ops only after f_op->open */
864    if (f->f_flags & O_DIRECT) {
865        if (!f->f_mapping->a_ops ||
866            ((!f->f_mapping->a_ops->direct_IO) &&
867            (!f->f_mapping->a_ops->get_xip_mem))) {
868            fput(f);
869            f = ERR_PTR(-EINVAL);
870        }
871    }
872
873    return f;
874
875cleanup_all:
876    fops_put(f->f_op);
877    if (f->f_mode & FMODE_WRITE) {
878        put_write_access(inode);
879        if (!special_file(inode->i_mode)) {
880            /*
881             * We don't consider this a real
882             * mnt_want/drop_write() pair
883             * because it all happenend right
884             * here, so just reset the state.
885             */
886            file_reset_write(f);
887            mnt_drop_write(mnt);
888        }
889    }
890    file_kill(f);
891    f->f_path.dentry = NULL;
892    f->f_path.mnt = NULL;
893cleanup_file:
894    put_filp(f);
895    dput(dentry);
896    mntput(mnt);
897    return ERR_PTR(error);
898}
899
900/**
901 * lookup_instantiate_filp - instantiates the open intent filp
902 * @nd: pointer to nameidata
903 * @dentry: pointer to dentry
904 * @open: open callback
905 *
906 * Helper for filesystems that want to use lookup open intents and pass back
907 * a fully instantiated struct file to the caller.
908 * This function is meant to be called from within a filesystem's
909 * lookup method.
910 * Beware of calling it for non-regular files! Those ->open methods might block
911 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
912 * leading to a deadlock, as nobody can open that fifo anymore, because
913 * another process to open fifo will block on locked parent when doing lookup).
914 * Note that in case of error, nd->intent.open.file is destroyed, but the
915 * path information remains valid.
916 * If the open callback is set to NULL, then the standard f_op->open()
917 * filesystem callback is substituted.
918 */
919struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
920        int (*open)(struct inode *, struct file *))
921{
922    const struct cred *cred = current_cred();
923
924    if (IS_ERR(nd->intent.open.file))
925        goto out;
926    if (IS_ERR(dentry))
927        goto out_err;
928    nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
929                         nd->intent.open.file,
930                         open, cred);
931out:
932    return nd->intent.open.file;
933out_err:
934    release_open_intent(nd);
935    nd->intent.open.file = (struct file *)dentry;
936    goto out;
937}
938EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
939
940/**
941 * nameidata_to_filp - convert a nameidata to an open filp.
942 * @nd: pointer to nameidata
943 * @flags: open flags
944 *
945 * Note that this function destroys the original nameidata
946 */
947struct file *nameidata_to_filp(struct nameidata *nd)
948{
949    const struct cred *cred = current_cred();
950    struct file *filp;
951
952    /* Pick up the filp from the open intent */
953    filp = nd->intent.open.file;
954    /* Has the filesystem initialised the file for us? */
955    if (filp->f_path.dentry == NULL)
956        filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
957                     NULL, cred);
958    else
959        path_put(&nd->path);
960    return filp;
961}
962
963/*
964 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
965 * error.
966 */
967struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
968             const struct cred *cred)
969{
970    int error;
971    struct file *f;
972
973    validate_creds(cred);
974
975    /*
976     * We must always pass in a valid mount pointer. Historically
977     * callers got away with not passing it, but we must enforce this at
978     * the earliest possible point now to avoid strange problems deep in the
979     * filesystem stack.
980     */
981    if (!mnt) {
982        printk(KERN_WARNING "%s called with NULL vfsmount\n", __func__);
983        dump_stack();
984        return ERR_PTR(-EINVAL);
985    }
986
987    error = -ENFILE;
988    f = get_empty_filp();
989    if (f == NULL) {
990        dput(dentry);
991        mntput(mnt);
992        return ERR_PTR(error);
993    }
994
995    f->f_flags = flags;
996    return __dentry_open(dentry, mnt, f, NULL, cred);
997}
998EXPORT_SYMBOL(dentry_open);
999
1000static void __put_unused_fd(struct files_struct *files, unsigned int fd)
1001{
1002    struct fdtable *fdt = files_fdtable(files);
1003    __FD_CLR(fd, fdt->open_fds);
1004    if (fd < files->next_fd)
1005        files->next_fd = fd;
1006}
1007
1008void put_unused_fd(unsigned int fd)
1009{
1010    struct files_struct *files = current->files;
1011    spin_lock(&files->file_lock);
1012    __put_unused_fd(files, fd);
1013    spin_unlock(&files->file_lock);
1014}
1015
1016EXPORT_SYMBOL(put_unused_fd);
1017
1018/*
1019 * Install a file pointer in the fd array.
1020 *
1021 * The VFS is full of places where we drop the files lock between
1022 * setting the open_fds bitmap and installing the file in the file
1023 * array. At any such point, we are vulnerable to a dup2() race
1024 * installing a file in the array before us. We need to detect this and
1025 * fput() the struct file we are about to overwrite in this case.
1026 *
1027 * It should never happen - if we allow dup2() do it, _really_ bad things
1028 * will follow.
1029 */
1030
1031void fd_install(unsigned int fd, struct file *file)
1032{
1033    struct files_struct *files = current->files;
1034    struct fdtable *fdt;
1035    spin_lock(&files->file_lock);
1036    fdt = files_fdtable(files);
1037    BUG_ON(fdt->fd[fd] != NULL);
1038    rcu_assign_pointer(fdt->fd[fd], file);
1039    spin_unlock(&files->file_lock);
1040}
1041
1042EXPORT_SYMBOL(fd_install);
1043
1044long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
1045{
1046    char *tmp = getname(filename);
1047    int fd = PTR_ERR(tmp);
1048
1049    if (!IS_ERR(tmp)) {
1050        fd = get_unused_fd_flags(flags);
1051        if (fd >= 0) {
1052            struct file *f = do_filp_open(dfd, tmp, flags, mode, 0);
1053            if (IS_ERR(f)) {
1054                put_unused_fd(fd);
1055                fd = PTR_ERR(f);
1056            } else {
1057                fsnotify_open(f->f_path.dentry);
1058                fd_install(fd, f);
1059            }
1060        }
1061        putname(tmp);
1062    }
1063    return fd;
1064}
1065
1066SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
1067{
1068    long ret;
1069
1070    if (force_o_largefile())
1071        flags |= O_LARGEFILE;
1072
1073    ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1074    /* avoid REGPARM breakage on x86: */
1075    asmlinkage_protect(3, ret, filename, flags, mode);
1076    return ret;
1077}
1078
1079SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1080        int, mode)
1081{
1082    long ret;
1083
1084    if (force_o_largefile())
1085        flags |= O_LARGEFILE;
1086
1087    ret = do_sys_open(dfd, filename, flags, mode);
1088    /* avoid REGPARM breakage on x86: */
1089    asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1090    return ret;
1091}
1092
1093#ifndef __alpha__
1094
1095/*
1096 * For backward compatibility? Maybe this should be moved
1097 * into arch/i386 instead?
1098 */
1099SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1100{
1101    return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1102}
1103
1104#endif
1105
1106/*
1107 * "id" is the POSIX thread ID. We use the
1108 * files pointer for this..
1109 */
1110int filp_close(struct file *filp, fl_owner_t id)
1111{
1112    int retval = 0;
1113
1114    if (!file_count(filp)) {
1115        printk(KERN_ERR "VFS: Close: file count is 0\n");
1116        return 0;
1117    }
1118
1119    if (filp->f_op && filp->f_op->flush)
1120        retval = filp->f_op->flush(filp, id);
1121
1122    dnotify_flush(filp, id);
1123    locks_remove_posix(filp, id);
1124    fput(filp);
1125    return retval;
1126}
1127
1128EXPORT_SYMBOL(filp_close);
1129
1130/*
1131 * Careful here! We test whether the file pointer is NULL before
1132 * releasing the fd. This ensures that one clone task can't release
1133 * an fd while another clone is opening it.
1134 */
1135SYSCALL_DEFINE1(close, unsigned int, fd)
1136{
1137    struct file * filp;
1138    struct files_struct *files = current->files;
1139    struct fdtable *fdt;
1140    int retval;
1141
1142    spin_lock(&files->file_lock);
1143    fdt = files_fdtable(files);
1144    if (fd >= fdt->max_fds)
1145        goto out_unlock;
1146    filp = fdt->fd[fd];
1147    if (!filp)
1148        goto out_unlock;
1149    rcu_assign_pointer(fdt->fd[fd], NULL);
1150    FD_CLR(fd, fdt->close_on_exec);
1151    __put_unused_fd(files, fd);
1152    spin_unlock(&files->file_lock);
1153    retval = filp_close(filp, files);
1154
1155    /* can't restart close syscall because file table entry was cleared */
1156    if (unlikely(retval == -ERESTARTSYS ||
1157             retval == -ERESTARTNOINTR ||
1158             retval == -ERESTARTNOHAND ||
1159             retval == -ERESTART_RESTARTBLOCK))
1160        retval = -EINTR;
1161
1162    return retval;
1163
1164out_unlock:
1165    spin_unlock(&files->file_lock);
1166    return -EBADF;
1167}
1168EXPORT_SYMBOL(sys_close);
1169
1170/*
1171 * This routine simulates a hangup on the tty, to arrange that users
1172 * are given clean terminals at login time.
1173 */
1174SYSCALL_DEFINE0(vhangup)
1175{
1176    if (capable(CAP_SYS_TTY_CONFIG)) {
1177        tty_vhangup_self();
1178        return 0;
1179    }
1180    return -EPERM;
1181}
1182
1183/*
1184 * Called when an inode is about to be open.
1185 * We use this to disallow opening large files on 32bit systems if
1186 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1187 * on this flag in sys_open.
1188 */
1189int generic_file_open(struct inode * inode, struct file * filp)
1190{
1191    if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1192        return -EOVERFLOW;
1193    return 0;
1194}
1195
1196EXPORT_SYMBOL(generic_file_open);
1197
1198/*
1199 * This is used by subsystems that don't want seekable
1200 * file descriptors
1201 */
1202int nonseekable_open(struct inode *inode, struct file *filp)
1203{
1204    filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1205    return 0;
1206}
1207
1208EXPORT_SYMBOL(nonseekable_open);
1209

Archive Download this file



interactive