sys_ia32.c

linux-2.6.15.6

/*
 * sys_ia32.c: Conversion between 32bit and 64bit native syscalls. Derived from sys_sparc32.c.
 *
 * Copyright (C) 2000		VA Linux Co
 * Copyright (C) 2000		Don Dugger <[e-mail]>
 * Copyright (C) 1999		Arun Sharma <[e-mail]>
 * Copyright (C) 1997,1998	Jakub Jelinek ([e-mail])
 * Copyright (C) 1997		David S. Miller ([e-mail])
 * Copyright (C) 2000-2003, 2005 Hewlett-Packard Co
 *	David Mosberger-Tang <[e-mail]>
 * Copyright (C) 2004		Gordon Jin <[e-mail]>
 *
 * These routines maintain argument size conversion between 32bit and 64bit
 * environment.
 */

#include <linux/config.h>
#include <linux/kernel.h>
#include <linux/syscalls.h>
#include <linux/sysctl.h>
#include <linux/sched.h>
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/signal.h>
#include <linux/resource.h>
#include <linux/times.h>
#include <linux/utsname.h>
#include <linux/timex.h>
#include <linux/smp.h>
#include <linux/smp_lock.h>
#include <linux/sem.h>
#include <linux/msg.h>
#include <linux/mm.h>
#include <linux/shm.h>
#include <linux/slab.h>
#include <linux/uio.h>
#include <linux/nfs_fs.h>
#include <linux/quota.h>
#include <linux/syscalls.h>
#include <linux/sunrpc/svc.h>
#include <linux/nfsd/nfsd.h>
#include <linux/nfsd/cache.h>
#include <linux/nfsd/xdr.h>
#include <linux/nfsd/syscall.h>
#include <linux/poll.h>
#include <linux/eventpoll.h>
#include <linux/personality.h>
#include <linux/ptrace.h>
#include <linux/stat.h>
#include <linux/ipc.h>
#include <linux/compat.h>
#include <linux/vfs.h>
#include <linux/mman.h>

#include <asm/intrinsics.h>
#include <asm/semaphore.h>
#include <asm/types.h>
#include <asm/uaccess.h>
#include <asm/unistd.h>

#include "ia32priv.h"

#include <net/scm.h>
#include <net/sock.h>

#define DEBUG	0

#if DEBUG
# define DBG(fmt...)	printk(KERN_DEBUG fmt)
#else
# define DBG(fmt...)
#endif

#define ROUND_UP(x,a)	((__typeof__(x))(((unsigned long)(x) + ((a) - 1)) & ~((a) - 1)))

#define OFFSET4K(a)		((a) & 0xfff)
#define PAGE_START(addr)	((addr) & PAGE_MASK)
#define MINSIGSTKSZ_IA32	2048

#define high2lowuid(uid) ((uid) > 65535 ? 65534 : (uid))
#define high2lowgid(gid) ((gid) > 65535 ? 65534 : (gid))

/*
 * Anything that modifies or inspects ia32 user virtual memory must hold this semaphore
 * while doing so.
 */
/* XXX make per-mm: */
static DECLARE_MUTEX(ia32_mmap_sem);

asmlinkage long
sys32_execve (char __user *name, compat_uptr_t __user *argv, compat_uptr_t __user *envp,
	      struct pt_regs *regs)
{
	long error;
	char *filename;
	unsigned long old_map_base, old_task_size, tssd;

	filename = getname(name);
	error = PTR_ERR(filename);
	if (IS_ERR(filename))
		return error;

	old_map_base  = current->thread.map_base;
	old_task_size = current->thread.task_size;
	tssd = ia64_get_kr(IA64_KR_TSSD);

	/* we may be exec'ing a 64-bit process: reset map base, task-size, and io-base: */
	current->thread.map_base  = DEFAULT_MAP_BASE;
	current->thread.task_size = DEFAULT_TASK_SIZE;
	ia64_set_kr(IA64_KR_IO_BASE, current->thread.old_iob);
	ia64_set_kr(IA64_KR_TSSD, current->thread.old_k1);

	error = compat_do_execve(filename, argv, envp, regs);
	putname(filename);

	if (error < 0) {
		/* oops, execve failed, switch back to old values... */
		ia64_set_kr(IA64_KR_IO_BASE, IA32_IOBASE);
		ia64_set_kr(IA64_KR_TSSD, tssd);
		current->thread.map_base  = old_map_base;
		current->thread.task_size = old_task_size;
	}

	return error;
}

int cp_compat_stat(struct kstat *stat, struct compat_stat __user *ubuf)
{
	int err;

	if ((u64) stat->size > MAX_NON_LFS ||
	    !old_valid_dev(stat->dev) ||
	    !old_valid_dev(stat->rdev))
		return -EOVERFLOW;

	if (clear_user(ubuf, sizeof(*ubuf)))
		return -EFAULT;

	err  = __put_user(old_encode_dev(stat->dev), &ubuf->st_dev);
	err |= __put_user(stat->ino, &ubuf->st_ino);
	err |= __put_user(stat->mode, &ubuf->st_mode);
	err |= __put_user(stat->nlink, &ubuf->st_nlink);
	err |= __put_user(high2lowuid(stat->uid), &ubuf->st_uid);
	err |= __put_user(high2lowgid(stat->gid), &ubuf->st_gid);
	err |= __put_user(old_encode_dev(stat->rdev), &ubuf->st_rdev);
	err |= __put_user(stat->size, &ubuf->st_size);
	err |= __put_user(stat->atime.tv_sec, &ubuf->st_atime);
	err |= __put_user(stat->atime.tv_nsec, &ubuf->st_atime_nsec);
	err |= __put_user(stat->mtime.tv_sec, &ubuf->st_mtime);
	err |= __put_user(stat->mtime.tv_nsec, &ubuf->st_mtime_nsec);
	err |= __put_user(stat->ctime.tv_sec, &ubuf->st_ctime);
	err |= __put_user(stat->ctime.tv_nsec, &ubuf->st_ctime_nsec);
	err |= __put_user(stat->blksize, &ubuf->st_blksize);
	err |= __put_user(stat->blocks, &ubuf->st_blocks);
	return err;
}

#if PAGE_SHIFT > IA32_PAGE_SHIFT


static int
get_page_prot (struct vm_area_struct *vma, unsigned long addr)
{
	int prot = 0;

	if (!vma || vma->vm_start > addr)
		return 0;

	if (vma->vm_flags & VM_READ)
		prot |= PROT_READ;
	if (vma->vm_flags & VM_WRITE)
		prot |= PROT_WRITE;
	if (vma->vm_flags & VM_EXEC)
		prot |= PROT_EXEC;
	return prot;
}

/*
 * Map a subpage by creating an anonymous page that contains the union of the old page and
 * the subpage.
 */
static unsigned long
mmap_subpage (struct file *file, unsigned long start, unsigned long end, int prot, int flags,
	      loff_t off)
{
	void *page = NULL;
	struct inode *inode;
	unsigned long ret = 0;
	struct vm_area_struct *vma = find_vma(current->mm, start);
	int old_prot = get_page_prot(vma, start);

	DBG("mmap_subpage(file=%p,start=0x%lx,end=0x%lx,prot=%x,flags=%x,off=0x%llx)\n",
	    file, start, end, prot, flags, off);


	/* Optimize the case where the old mmap and the new mmap are both anonymous */
	if ((old_prot & PROT_WRITE) && (flags & MAP_ANONYMOUS) && !vma->vm_file) {
		if (clear_user((void __user *) start, end - start)) {
			ret = -EFAULT;
			goto out;
		}
		goto skip_mmap;
	}

	page = (void *) get_zeroed_page(GFP_KERNEL);
	if (!page)
		return -ENOMEM;

	if (old_prot)
		copy_from_user(page, (void __user *) PAGE_START(start), PAGE_SIZE);

	down_write(&current->mm->mmap_sem);
	{
		ret = do_mmap(NULL, PAGE_START(start), PAGE_SIZE, prot | PROT_WRITE,
			      flags | MAP_FIXED | MAP_ANONYMOUS, 0);
	}
	up_write(&current->mm->mmap_sem);

	if (IS_ERR((void *) ret))
		goto out;

	if (old_prot) {
		/* copy back the old page contents.  */
		if (offset_in_page(start))
			copy_to_user((void __user *) PAGE_START(start), page,
				     offset_in_page(start));
		if (offset_in_page(end))
			copy_to_user((void __user *) end, page + offset_in_page(end),
				     PAGE_SIZE - offset_in_page(end));
	}

	if (!(flags & MAP_ANONYMOUS)) {
		/* read the file contents */
		inode = file->f_dentry->d_inode;
		if (!inode->i_fop || !file->f_op->read
		    || ((*file->f_op->read)(file, (char __user *) start, end - start, &off) < 0))
		{
			ret = -EINVAL;
			goto out;
		}
	}

 skip_mmap:
	if (!(prot & PROT_WRITE))
		ret = sys_mprotect(PAGE_START(start), PAGE_SIZE, prot | old_prot);
  out:
	if (page)
		free_page((unsigned long) page);
	return ret;
}

/* SLAB cache for partial_page structures */
kmem_cache_t *partial_page_cachep;

/*
 * init partial_page_list.
 * return 0 means kmalloc fail.
 */
struct partial_page_list*
ia32_init_pp_list(void)
{
	struct partial_page_list *p;

	if ((p = kmalloc(sizeof(*p), GFP_KERNEL)) == NULL)
		return p;
	p->pp_head = NULL;
	p->ppl_rb = RB_ROOT;
	p->pp_hint = NULL;
	atomic_set(&p->pp_count, 1);
	return p;
}

/*
 * Search for the partial page with @start in partial page list @ppl.
 * If finds the partial page, return the found partial page.
 * Else, return 0 and provide @pprev, @rb_link, @rb_parent to
 * be used by later __ia32_insert_pp().
 */
static struct partial_page *
__ia32_find_pp(struct partial_page_list *ppl, unsigned int start,
	struct partial_page **pprev, struct rb_node ***rb_link,
	struct rb_node **rb_parent)
{
	struct partial_page *pp;
	struct rb_node **__rb_link, *__rb_parent, *rb_prev;

	pp = ppl->pp_hint;
	if (pp && pp->base == start)
		return pp;

	__rb_link = &ppl->ppl_rb.rb_node;
	rb_prev = __rb_parent = NULL;

	while (*__rb_link) {
		__rb_parent = *__rb_link;
		pp = rb_entry(__rb_parent, struct partial_page, pp_rb);

		if (pp->base == start) {
			ppl->pp_hint = pp;
			return pp;
		} else if (pp->base < start) {
			rb_prev = __rb_parent;
			__rb_link = &__rb_parent->rb_right;
		} else {
			__rb_link = &__rb_parent->rb_left;
		}
	}

	*rb_link = __rb_link;
	*rb_parent = __rb_parent;
	*pprev = NULL;
	if (rb_prev)
		*pprev = rb_entry(rb_prev, struct partial_page, pp_rb);
	return NULL;
}

/*
 * insert @pp into @ppl.
 */
static void
__ia32_insert_pp(struct partial_page_list *ppl, struct partial_page *pp,
	 struct partial_page *prev, struct rb_node **rb_link,
	struct rb_node *rb_parent)
{
	/* link list */
	if (prev) {
		pp->next = prev->next;
		prev->next = pp;
	} else {
		ppl->pp_head = pp;
		if (rb_parent)
			pp->next = rb_entry(rb_parent,
				struct partial_page, pp_rb);
		else
			pp->next = NULL;
	}

	/* link rb */
	rb_link_node(&pp->pp_rb, rb_parent, rb_link);
	rb_insert_color(&pp->pp_rb, &ppl->ppl_rb);

	ppl->pp_hint = pp;
}

/*
 * delete @pp from partial page list @ppl.
 */
static void
__ia32_delete_pp(struct partial_page_list *ppl, struct partial_page *pp,
	struct partial_page *prev)
{
	if (prev) {
		prev->next = pp->next;
		if (ppl->pp_hint == pp)
			ppl->pp_hint = prev;
	} else {
		ppl->pp_head = pp->next;
		if (ppl->pp_hint == pp)
			ppl->pp_hint = pp->next;
	}
	rb_erase(&pp->pp_rb, &ppl->ppl_rb);
	kmem_cache_free(partial_page_cachep, pp);
}

static struct partial_page *
__pp_prev(struct partial_page *pp)
{
	struct rb_node *prev = rb_prev(&pp->pp_rb);
	if (prev)
		return rb_entry(prev, struct partial_page, pp_rb);
	else
		return NULL;
}

/*
 * Delete partial pages with address between @start and @end.
 * @start and @end are page aligned.
 */
static void
__ia32_delete_pp_range(unsigned int start, unsigned int end)
{
	struct partial_page *pp, *prev;
	struct rb_node **rb_link, *rb_parent;

	if (start >= end)
		return;

	pp = __ia32_find_pp(current->thread.ppl, start, &prev,
					&rb_link, &rb_parent);
	if (pp)
		prev = __pp_prev(pp);
	else {
		if (prev)
			pp = prev->next;
		else
			pp = current->thread.ppl->pp_head;
	}

	while (pp && pp->base < end) {
		struct partial_page *tmp = pp->next;
		__ia32_delete_pp(current->thread.ppl, pp, prev);
		pp = tmp;
	}
}

/*
 * Set the range between @start and @end in bitmap.
 * @start and @end should be IA32 page aligned and in the same IA64 page.
 */
static int
__ia32_set_pp(unsigned int start, unsigned int end, int flags)
{
	struct partial_page *pp, *prev;
	struct rb_node ** rb_link, *rb_parent;
	unsigned int pstart, start_bit, end_bit, i;

	pstart = PAGE_START(start);
	start_bit = (start % PAGE_SIZE) / IA32_PAGE_SIZE;
	end_bit = (end % PAGE_SIZE) / IA32_PAGE_SIZE;
	if (end_bit == 0)
		end_bit = PAGE_SIZE / IA32_PAGE_SIZE;
	pp = __ia32_find_pp(current->thread.ppl, pstart, &prev,
					&rb_link, &rb_parent);
	if (pp) {
		for (i = start_bit; i < end_bit; i++)
			set_bit(i, &pp->bitmap);
		/*
		 * Check: if this partial page has been set to a full page,
		 * then delete it.
		 */
		if (find_first_zero_bit(&pp->bitmap, sizeof(pp->bitmap)*8) >=
				PAGE_SIZE/IA32_PAGE_SIZE) {
			__ia32_delete_pp(current->thread.ppl, pp, __pp_prev(pp));
		}
		return 0;
	}

	/*
	 * MAP_FIXED may lead to overlapping mmap.
	 * In this case, the requested mmap area may already mmaped as a full
	 * page. So check vma before adding a new partial page.
	 */
	if (flags & MAP_FIXED) {
		struct vm_area_struct *vma = find_vma(current->mm, pstart);
		if (vma && vma->vm_start <= pstart)
			return 0;
	}

	/* new a partial_page */
	pp = kmem_cache_alloc(partial_page_cachep, GFP_KERNEL);
	if (!pp)
		return -ENOMEM;
	pp->base = pstart;
	pp->bitmap = 0;
	for (i=start_bit; i<end_bit; i++)
		set_bit(i, &(pp->bitmap));
	pp->next = NULL;
	__ia32_insert_pp(current->thread.ppl, pp, prev, rb_link, rb_parent);
	return 0;
}

/*
 * @start and @end should be IA32 page aligned, but don't need to be in the
 * same IA64 page. Split @start and @end to make sure they're in the same IA64
 * page, then call __ia32_set_pp().
 */
static void
ia32_set_pp(unsigned int start, unsigned int end, int flags)
{
	down_write(&current->mm->mmap_sem);
	if (flags & MAP_FIXED) {
		/*
		 * MAP_FIXED may lead to overlapping mmap. When this happens,
		 * a series of complete IA64 pages results in deletion of
		 * old partial pages in that range.
		 */
		__ia32_delete_pp_range(PAGE_ALIGN(start), PAGE_START(end));
	}

	if (end < PAGE_ALIGN(start)) {
		__ia32_set_pp(start, end, flags);
	} else {
		if (offset_in_page(start))
			__ia32_set_pp(start, PAGE_ALIGN(start), flags);
		if (offset_in_page(end))
			__ia32_set_pp(PAGE_START(end), end, flags);
	}
	up_write(&current->mm->mmap_sem);
}

/*
 * Unset the range between @start and @end in bitmap.
 * @start and @end should be IA32 page aligned and in the same IA64 page.
 * After doing that, if the bitmap is 0, then free the page and return 1,
 * 	else return 0;
 * If not find the partial page in the list, then
 * 	If the vma exists, then the full page is set to a partial page;
 *	Else return -ENOMEM.
 */
static int
__ia32_unset_pp(unsigned int start, unsigned int end)
{
	struct partial_page *pp, *prev;
	struct rb_node ** rb_link, *rb_parent;
	unsigned int pstart, start_bit, end_bit, i;
	struct vm_area_struct *vma;

	pstart = PAGE_START(start);
	start_bit = (start % PAGE_SIZE) / IA32_PAGE_SIZE;
	end_bit = (end % PAGE_SIZE) / IA32_PAGE_SIZE;
	if (end_bit == 0)
		end_bit = PAGE_SIZE / IA32_PAGE_SIZE;

	pp = __ia32_find_pp(current->thread.ppl, pstart, &prev,
					&rb_link, &rb_parent);
	if (pp) {
		for (i = start_bit; i < end_bit; i++)
			clear_bit(i, &pp->bitmap);
		if (pp->bitmap == 0) {
			__ia32_delete_pp(current->thread.ppl, pp, __pp_prev(pp));