Embed a struct path into struct nameidata instead of nd->{dentry,mnt}
[safe/jmp/linux-2.6] / fs / compat.c
index a32e2ae..a8d6237 100644 (file)
 #include <linux/namei.h>
 #include <linux/file.h>
 #include <linux/vfs.h>
-#include <linux/ioctl32.h>
 #include <linux/ioctl.h>
 #include <linux/init.h>
-#include <linux/sockios.h>     /* for SIOCDEVPRIVATE */
 #include <linux/smb.h>
 #include <linux/smb_mount.h>
 #include <linux/ncp_mount.h>
 #include <linux/personality.h>
 #include <linux/rwsem.h>
 #include <linux/tsacct_kern.h>
+#include <linux/security.h>
 #include <linux/highmem.h>
+#include <linux/signal.h>
 #include <linux/poll.h>
 #include <linux/mm.h>
 #include <linux/eventpoll.h>
 
-#include <net/sock.h>          /* siocdevprivate_ioctl */
-
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
 #include <asm/ioctls.h>
@@ -80,30 +78,57 @@ int compat_printk(const char *fmt, ...)
  */
 asmlinkage long compat_sys_utime(char __user *filename, struct compat_utimbuf __user *t)
 {
-       struct timeval tv[2];
+       struct timespec tv[2];
 
        if (t) {
                if (get_user(tv[0].tv_sec, &t->actime) ||
                    get_user(tv[1].tv_sec, &t->modtime))
                        return -EFAULT;
-               tv[0].tv_usec = 0;
-               tv[1].tv_usec = 0;
+               tv[0].tv_nsec = 0;
+               tv[1].tv_nsec = 0;
+       }
+       return do_utimes(AT_FDCWD, filename, t ? tv : NULL, 0);
+}
+
+asmlinkage long compat_sys_utimensat(unsigned int dfd, char __user *filename, struct compat_timespec __user *t, int flags)
+{
+       struct timespec tv[2];
+
+       if  (t) {
+               if (get_compat_timespec(&tv[0], &t[0]) ||
+                   get_compat_timespec(&tv[1], &t[1]))
+                       return -EFAULT;
+
+               if ((tv[0].tv_nsec == UTIME_OMIT || tv[0].tv_nsec == UTIME_NOW)
+                   && tv[0].tv_sec != 0)
+                       return -EINVAL;
+               if ((tv[1].tv_nsec == UTIME_OMIT || tv[1].tv_nsec == UTIME_NOW)
+                   && tv[1].tv_sec != 0)
+                       return -EINVAL;
+
+               if (tv[0].tv_nsec == UTIME_OMIT && tv[1].tv_nsec == UTIME_OMIT)
+                       return 0;
        }
-       return do_utimes(AT_FDCWD, filename, t ? tv : NULL);
+       return do_utimes(dfd, filename, t ? tv : NULL, flags);
 }
 
 asmlinkage long compat_sys_futimesat(unsigned int dfd, char __user *filename, struct compat_timeval __user *t)
 {
-       struct timeval tv[2];
+       struct timespec tv[2];
 
        if (t) {
                if (get_user(tv[0].tv_sec, &t[0].tv_sec) ||
-                   get_user(tv[0].tv_usec, &t[0].tv_usec) ||
+                   get_user(tv[0].tv_nsec, &t[0].tv_usec) ||
                    get_user(tv[1].tv_sec, &t[1].tv_sec) ||
-                   get_user(tv[1].tv_usec, &t[1].tv_usec))
+                   get_user(tv[1].tv_nsec, &t[1].tv_usec))
                        return -EFAULT;
+               if (tv[0].tv_nsec >= 1000000 || tv[0].tv_nsec < 0 ||
+                   tv[1].tv_nsec >= 1000000 || tv[1].tv_nsec < 0)
+                       return -EINVAL;
+               tv[0].tv_nsec *= 1000;
+               tv[1].tv_nsec *= 1000;
        }
-       return do_utimes(dfd, filename, t ? tv : NULL);
+       return do_utimes(dfd, filename, t ? tv : NULL, 0);
 }
 
 asmlinkage long compat_sys_utimes(char __user *filename, struct compat_timeval __user *t)
@@ -216,7 +241,7 @@ asmlinkage long compat_sys_statfs(const char __user *path, struct compat_statfs
        error = user_path_walk(path, &nd);
        if (!error) {
                struct kstatfs tmp;
-               error = vfs_statfs(nd.dentry, &tmp);
+               error = vfs_statfs(nd.path.dentry, &tmp);
                if (!error)
                        error = put_compat_statfs(buf, &tmp);
                path_release(&nd);
@@ -284,7 +309,7 @@ asmlinkage long compat_sys_statfs64(const char __user *path, compat_size_t sz, s
        error = user_path_walk(path, &nd);
        if (!error) {
                struct kstatfs tmp;
-               error = vfs_statfs(nd.dentry, &tmp);
+               error = vfs_statfs(nd.path.dentry, &tmp);
                if (!error)
                        error = put_compat_statfs64(buf, &tmp);
                path_release(&nd);
@@ -313,163 +338,6 @@ out:
        return error;
 }
 
-/* ioctl32 stuff, used by sparc64, parisc, s390x, ppc64, x86_64, MIPS */
-
-#define IOCTL_HASHSIZE 256
-static struct ioctl_trans *ioctl32_hash_table[IOCTL_HASHSIZE];
-
-static inline unsigned long ioctl32_hash(unsigned long cmd)
-{
-       return (((cmd >> 6) ^ (cmd >> 4) ^ cmd)) % IOCTL_HASHSIZE;
-}
-
-static void ioctl32_insert_translation(struct ioctl_trans *trans)
-{
-       unsigned long hash;
-       struct ioctl_trans *t;
-
-       hash = ioctl32_hash (trans->cmd);
-       if (!ioctl32_hash_table[hash])
-               ioctl32_hash_table[hash] = trans;
-       else {
-               t = ioctl32_hash_table[hash];
-               while (t->next)
-                       t = t->next;
-               trans->next = NULL;
-               t->next = trans;
-       }
-}
-
-static int __init init_sys32_ioctl(void)
-{
-       int i;
-
-       for (i = 0; i < ioctl_table_size; i++) {
-               if (ioctl_start[i].next != 0) { 
-                       printk("ioctl translation %d bad\n",i); 
-                       return -1;
-               }
-
-               ioctl32_insert_translation(&ioctl_start[i]);
-       }
-       return 0;
-}
-
-__initcall(init_sys32_ioctl);
-
-static void compat_ioctl_error(struct file *filp, unsigned int fd,
-               unsigned int cmd, unsigned long arg)
-{
-       char buf[10];
-       char *fn = "?";
-       char *path;
-
-       /* find the name of the device. */
-       path = (char *)__get_free_page(GFP_KERNEL);
-       if (path) {
-               fn = d_path(filp->f_path.dentry, filp->f_path.mnt, path, PAGE_SIZE);
-               if (IS_ERR(fn))
-                       fn = "?";
-       }
-
-       sprintf(buf,"'%c'", (cmd>>_IOC_TYPESHIFT) & _IOC_TYPEMASK);
-       if (!isprint(buf[1]))
-               sprintf(buf, "%02x", buf[1]);
-       compat_printk("ioctl32(%s:%d): Unknown cmd fd(%d) "
-                       "cmd(%08x){t:%s;sz:%u} arg(%08x) on %s\n",
-                       current->comm, current->pid,
-                       (int)fd, (unsigned int)cmd, buf,
-                       (cmd >> _IOC_SIZESHIFT) & _IOC_SIZEMASK,
-                       (unsigned int)arg, fn);
-
-       if (path)
-               free_page((unsigned long)path);
-}
-
-asmlinkage long compat_sys_ioctl(unsigned int fd, unsigned int cmd,
-                               unsigned long arg)
-{
-       struct file *filp;
-       int error = -EBADF;
-       struct ioctl_trans *t;
-       int fput_needed;
-
-       filp = fget_light(fd, &fput_needed);
-       if (!filp)
-               goto out;
-
-       /* RED-PEN how should LSM module know it's handling 32bit? */
-       error = security_file_ioctl(filp, cmd, arg);
-       if (error)
-               goto out_fput;
-
-       /*
-        * To allow the compat_ioctl handlers to be self contained
-        * we need to check the common ioctls here first.
-        * Just handle them with the standard handlers below.
-        */
-       switch (cmd) {
-       case FIOCLEX:
-       case FIONCLEX:
-       case FIONBIO:
-       case FIOASYNC:
-       case FIOQSIZE:
-               break;
-
-       case FIBMAP:
-       case FIGETBSZ:
-       case FIONREAD:
-               if (S_ISREG(filp->f_path.dentry->d_inode->i_mode))
-                       break;
-               /*FALL THROUGH*/
-
-       default:
-               if (filp->f_op && filp->f_op->compat_ioctl) {
-                       error = filp->f_op->compat_ioctl(filp, cmd, arg);
-                       if (error != -ENOIOCTLCMD)
-                               goto out_fput;
-               }
-
-               if (!filp->f_op ||
-                   (!filp->f_op->ioctl && !filp->f_op->unlocked_ioctl))
-                       goto do_ioctl;
-               break;
-       }
-
-       for (t = ioctl32_hash_table[ioctl32_hash(cmd)]; t; t = t->next) {
-               if (t->cmd == cmd)
-                       goto found_handler;
-       }
-
-       if (S_ISSOCK(filp->f_path.dentry->d_inode->i_mode) &&
-           cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15)) {
-               error = siocdevprivate_ioctl(fd, cmd, arg);
-       } else {
-               static int count;
-
-               if (++count <= 50)
-                       compat_ioctl_error(filp, fd, cmd, arg);
-               error = -EINVAL;
-       }
-
-       goto out_fput;
-
- found_handler:
-       if (t->handler) {
-               lock_kernel();
-               error = t->handler(fd, cmd, arg, filp);
-               unlock_kernel();
-               goto out_fput;
-       }
-
- do_ioctl:
-       error = vfs_ioctl(filp, fd, cmd, arg);
- out_fput:
-       fput_light(filp, fput_needed);
- out:
-       return error;
-}
-
 static int get_compat_flock(struct flock *kfl, struct compat_flock __user *ufl)
 {
        if (!access_ok(VERIFY_READ, ufl, sizeof(*ufl)) ||
@@ -1236,10 +1104,6 @@ static ssize_t compat_do_readv_writev(int type, struct file *file,
        if (ret < 0)
                goto out;
 
-       ret = security_file_permission(file, type == READ ? MAY_READ:MAY_WRITE);
-       if (ret)
-               goto out;
-
        fnv = NULL;
        if (type == READ) {
                fn = file->f_op->read;
@@ -1389,6 +1253,7 @@ static int compat_copy_strings(int argc, compat_uptr_t __user *argv,
 {
        struct page *kmapped_page = NULL;
        char *kaddr = NULL;
+       unsigned long kpos = 0;
        int ret;
 
        while (argc-- > 0) {
@@ -1397,92 +1262,84 @@ static int compat_copy_strings(int argc, compat_uptr_t __user *argv,
                unsigned long pos;
 
                if (get_user(str, argv+argc) ||
-                       !(len = strnlen_user(compat_ptr(str), bprm->p))) {
+                   !(len = strnlen_user(compat_ptr(str), MAX_ARG_STRLEN))) {
                        ret = -EFAULT;
                        goto out;
                }
 
-               if (bprm->p < len)  {
+               if (len > MAX_ARG_STRLEN) {
                        ret = -E2BIG;
                        goto out;
                }
 
-               bprm->p -= len;
-               /* XXX: add architecture specific overflow check here. */
+               /* We're going to work our way backwords. */
                pos = bprm->p;
+               str += len;
+               bprm->p -= len;
 
                while (len > 0) {
-                       int i, new, err;
                        int offset, bytes_to_copy;
-                       struct page *page;
 
                        offset = pos % PAGE_SIZE;
-                       i = pos/PAGE_SIZE;
-                       page = bprm->page[i];
-                       new = 0;
-                       if (!page) {
-                               page = alloc_page(GFP_HIGHUSER);
-                               bprm->page[i] = page;
-                               if (!page) {
-                                       ret = -ENOMEM;
+                       if (offset == 0)
+                               offset = PAGE_SIZE;
+
+                       bytes_to_copy = offset;
+                       if (bytes_to_copy > len)
+                               bytes_to_copy = len;
+
+                       offset -= bytes_to_copy;
+                       pos -= bytes_to_copy;
+                       str -= bytes_to_copy;
+                       len -= bytes_to_copy;
+
+                       if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
+                               struct page *page;
+
+#ifdef CONFIG_STACK_GROWSUP
+                               ret = expand_stack_downwards(bprm->vma, pos);
+                               if (ret < 0) {
+                                       /* We've exceed the stack rlimit. */
+                                       ret = -E2BIG;
+                                       goto out;
+                               }
+#endif
+                               ret = get_user_pages(current, bprm->mm, pos,
+                                                    1, 1, 1, &page, NULL);
+                               if (ret <= 0) {
+                                       /* We've exceed the stack rlimit. */
+                                       ret = -E2BIG;
                                        goto out;
                                }
-                               new = 1;
-                       }
 
-                       if (page != kmapped_page) {
-                               if (kmapped_page)
+                               if (kmapped_page) {
+                                       flush_kernel_dcache_page(kmapped_page);
                                        kunmap(kmapped_page);
+                                       put_page(kmapped_page);
+                               }
                                kmapped_page = page;
                                kaddr = kmap(kmapped_page);
+                               kpos = pos & PAGE_MASK;
+                               flush_cache_page(bprm->vma, kpos,
+                                                page_to_pfn(kmapped_page));
                        }
-                       if (new && offset)
-                               memset(kaddr, 0, offset);
-                       bytes_to_copy = PAGE_SIZE - offset;
-                       if (bytes_to_copy > len) {
-                               bytes_to_copy = len;
-                               if (new)
-                                       memset(kaddr+offset+len, 0,
-                                               PAGE_SIZE-offset-len);
-                       }
-                       err = copy_from_user(kaddr+offset, compat_ptr(str),
-                                               bytes_to_copy);
-                       if (err) {
+                       if (copy_from_user(kaddr+offset, compat_ptr(str),
+                                               bytes_to_copy)) {
                                ret = -EFAULT;
                                goto out;
                        }
-
-                       pos += bytes_to_copy;
-                       str += bytes_to_copy;
-                       len -= bytes_to_copy;
                }
        }
        ret = 0;
 out:
-       if (kmapped_page)
+       if (kmapped_page) {
+               flush_kernel_dcache_page(kmapped_page);
                kunmap(kmapped_page);
-       return ret;
-}
-
-#ifdef CONFIG_MMU
-
-#define free_arg_pages(bprm) do { } while (0)
-
-#else
-
-static inline void free_arg_pages(struct linux_binprm *bprm)
-{
-       int i;
-
-       for (i = 0; i < MAX_ARG_PAGES; i++) {
-               if (bprm->page[i])
-                       __free_page(bprm->page[i]);
-               bprm->page[i] = NULL;
+               put_page(kmapped_page);
        }
+       return ret;
 }
 
-#endif /* CONFIG_MMU */
-
 /*
  * compat_do_execve() is mostly a copy of do_execve(), with the exception
  * that it processes 32 bit argv and envp pointers.
@@ -1495,7 +1352,6 @@ int compat_do_execve(char * filename,
        struct linux_binprm *bprm;
        struct file *file;
        int retval;
-       int i;
 
        retval = -ENOMEM;
        bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
@@ -1509,24 +1365,19 @@ int compat_do_execve(char * filename,
 
        sched_exec();
 
-       bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
        bprm->file = file;
        bprm->filename = filename;
        bprm->interp = filename;
-       bprm->mm = mm_alloc();
-       retval = -ENOMEM;
-       if (!bprm->mm)
-               goto out_file;
 
-       retval = init_new_context(current, bprm->mm);
-       if (retval < 0)
-               goto out_mm;
+       retval = bprm_mm_init(bprm);
+       if (retval)
+               goto out_file;
 
-       bprm->argc = compat_count(argv, bprm->p / sizeof(compat_uptr_t));
+       bprm->argc = compat_count(argv, MAX_ARG_STRINGS);
        if ((retval = bprm->argc) < 0)
                goto out_mm;
 
-       bprm->envc = compat_count(envp, bprm->p / sizeof(compat_uptr_t));
+       bprm->envc = compat_count(envp, MAX_ARG_STRINGS);
        if ((retval = bprm->envc) < 0)
                goto out_mm;
 
@@ -1553,8 +1404,6 @@ int compat_do_execve(char * filename,
 
        retval = search_binary_handler(bprm, regs);
        if (retval >= 0) {
-               free_arg_pages(bprm);
-
                /* execve success */
                security_bprm_free(bprm);
                acct_update_integrals(current);
@@ -1563,19 +1412,12 @@ int compat_do_execve(char * filename,
        }
 
 out:
-       /* Something went wrong, return the inode and free the argument pages*/
-       for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
-               struct page * page = bprm->page[i];
-               if (page)
-                       __free_page(page);
-       }
-
        if (bprm->security)
                security_bprm_free(bprm);
 
 out_mm:
        if (bprm->mm)
-               mmdrop(bprm->mm);
+               mmput(bprm->mm);
 
 out_file:
        if (bprm->file) {
@@ -1676,9 +1518,10 @@ int compat_core_sys_select(int n, compat_ulong_t __user *inp,
        compat_ulong_t __user *outp, compat_ulong_t __user *exp, s64 *timeout)
 {
        fd_set_bits fds;
-       char *bits;
+       void *bits;
        int size, max_fds, ret = -EINVAL;
        struct fdtable *fdt;
+       long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
 
        if (n < 0)
                goto out_nofds;
@@ -1696,11 +1539,14 @@ int compat_core_sys_select(int n, compat_ulong_t __user *inp,
         * since we used fdset we need to allocate memory in units of
         * long-words.
         */
-       ret = -ENOMEM;
        size = FDS_BYTES(n);
-       bits = kmalloc(6 * size, GFP_KERNEL);
-       if (!bits)
-               goto out_nofds;
+       bits = stack_fds;
+       if (size > sizeof(stack_fds) / 6) {
+               bits = kmalloc(6 * size, GFP_KERNEL);
+               ret = -ENOMEM;
+               if (!bits)
+                       goto out_nofds;
+       }
        fds.in      = (unsigned long *)  bits;
        fds.out     = (unsigned long *) (bits +   size);
        fds.ex      = (unsigned long *) (bits + 2*size);
@@ -1732,7 +1578,8 @@ int compat_core_sys_select(int n, compat_ulong_t __user *inp,
            compat_set_fd_set(n, exp, fds.res_ex))
                ret = -EFAULT;
 out:
-       kfree(bits);
+       if (bits != stack_fds)
+               kfree(bits);
 out_nofds:
        return ret;
 }
@@ -2236,51 +2083,6 @@ long asmlinkage compat_sys_nfsservctl(int cmd, void *notused, void *notused2)
 
 #ifdef CONFIG_EPOLL
 
-#ifdef CONFIG_HAS_COMPAT_EPOLL_EVENT
-asmlinkage long compat_sys_epoll_ctl(int epfd, int op, int fd,
-                       struct compat_epoll_event __user *event)
-{
-       long err = 0;
-       struct compat_epoll_event user;
-       struct epoll_event __user *kernel = NULL;
-
-       if (event) {
-               if (copy_from_user(&user, event, sizeof(user)))
-                       return -EFAULT;
-               kernel = compat_alloc_user_space(sizeof(struct epoll_event));
-               err |= __put_user(user.events, &kernel->events);
-               err |= __put_user(user.data, &kernel->data);
-       }
-
-       return err ? err : sys_epoll_ctl(epfd, op, fd, kernel);
-}
-
-
-asmlinkage long compat_sys_epoll_wait(int epfd,
-                       struct compat_epoll_event __user *events,
-                       int maxevents, int timeout)
-{
-       long i, ret, err = 0;
-       struct epoll_event __user *kbuf;
-       struct epoll_event ev;
-
-       if ((maxevents <= 0) ||
-                       (maxevents > (INT_MAX / sizeof(struct epoll_event))))
-               return -EINVAL;
-       kbuf = compat_alloc_user_space(sizeof(struct epoll_event) * maxevents);
-       ret = sys_epoll_wait(epfd, kbuf, maxevents, timeout);
-       for (i = 0; i < ret; i++) {
-               err |= __get_user(ev.events, &kbuf[i].events);
-               err |= __get_user(ev.data, &kbuf[i].data);
-               err |= __put_user(ev.events, &events->events);
-               err |= __put_user_unaligned(ev.data, &events->data);
-               events++;
-       }
-
-       return err ? -EFAULT: ret;
-}
-#endif /* CONFIG_HAS_COMPAT_EPOLL_EVENT */
-
 #ifdef TIF_RESTORE_SIGMASK
 asmlinkage long compat_sys_epoll_pwait(int epfd,
                        struct compat_epoll_event __user *events,
@@ -2306,11 +2108,7 @@ asmlinkage long compat_sys_epoll_pwait(int epfd,
                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
        }
 
-#ifdef CONFIG_HAS_COMPAT_EPOLL_EVENT
-       err = compat_sys_epoll_wait(epfd, events, maxevents, timeout);
-#else
        err = sys_epoll_wait(epfd, events, maxevents, timeout);
-#endif
 
        /*
         * If we changed the signal mask, we need to restore the original one.
@@ -2332,3 +2130,68 @@ asmlinkage long compat_sys_epoll_pwait(int epfd,
 #endif /* TIF_RESTORE_SIGMASK */
 
 #endif /* CONFIG_EPOLL */
+
+#ifdef CONFIG_SIGNALFD
+
+asmlinkage long compat_sys_signalfd(int ufd,
+                                   const compat_sigset_t __user *sigmask,
+                                   compat_size_t sigsetsize)
+{
+       compat_sigset_t ss32;
+       sigset_t tmp;
+       sigset_t __user *ksigmask;
+
+       if (sigsetsize != sizeof(compat_sigset_t))
+               return -EINVAL;
+       if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
+               return -EFAULT;
+       sigset_from_compat(&tmp, &ss32);
+       ksigmask = compat_alloc_user_space(sizeof(sigset_t));
+       if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
+               return -EFAULT;
+
+       return sys_signalfd(ufd, ksigmask, sizeof(sigset_t));
+}
+
+#endif /* CONFIG_SIGNALFD */
+
+#ifdef CONFIG_TIMERFD
+
+asmlinkage long compat_sys_timerfd_settime(int ufd, int flags,
+                                  const struct compat_itimerspec __user *utmr,
+                                  struct compat_itimerspec __user *otmr)
+{
+       int error;
+       struct itimerspec t;
+       struct itimerspec __user *ut;
+
+       if (get_compat_itimerspec(&t, utmr))
+               return -EFAULT;
+       ut = compat_alloc_user_space(2 * sizeof(struct itimerspec));
+       if (copy_to_user(&ut[0], &t, sizeof(t)))
+               return -EFAULT;
+       error = sys_timerfd_settime(ufd, flags, &ut[0], &ut[1]);
+       if (!error && otmr)
+               error = (copy_from_user(&t, &ut[1], sizeof(struct itimerspec)) ||
+                        put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
+
+       return error;
+}
+
+asmlinkage long compat_sys_timerfd_gettime(int ufd,
+                                  struct compat_itimerspec __user *otmr)
+{
+       int error;
+       struct itimerspec t;
+       struct itimerspec __user *ut;
+
+       ut = compat_alloc_user_space(sizeof(struct itimerspec));
+       error = sys_timerfd_gettime(ufd, ut);
+       if (!error)
+               error = (copy_from_user(&t, ut, sizeof(struct itimerspec)) ||
+                        put_compat_itimerspec(otmr, &t)) ? -EFAULT: 0;
+
+       return error;
+}
+
+#endif /* CONFIG_TIMERFD */