Skip to contentSkip to conversation

fix possible PAGE_CACHE_SHIFT overflows

PATCH In review [email protected] 3 participants 4 messages Jan 8, 2006

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 146/233 fix possible PAGE_CACHE_SHIFT overflows

From: Andrew Morton <[email protected]>

We've had two instances recently of overflows when doing

	64_bit_value = (32_bit_value << PAGE_CACHE_SHIFT)

I did a tree-wide grep of `<<.*PAGE_CACHE_SHIFT' and this is the result.

- afs_rxfs_fetch_descriptor.offset is of type off_t, which seems broken.

- jfs and jffs are limited to 4GB anyway.

- reiserfs map_block_for_writepage() takes an unsigned long for the block - it should take sector_t. (It'll fail for huge filesystems with blocksize<PAGE_CACHE_SIZE)

- cramfs_read() needs to use sector_t (I think cramsfs is busted on large filesystems anyway)

- affs is limited in file size anyway.

- I generally didn't fix 32-bit overflows in directory operations.

- arm's __flush_dcache_page() is peculiar. What if the page lies beyond 4G?

- gss_wrap_req_priv() needs checking (snd_buf->page_base)

Cc Oleg Drokin <[email protected]>
Cc David Howells <[email protected]>
Cc David Woodhouse <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Anton Altaparmakov <[email protected]>
Cc Jeff Dike <[email protected]>
Cc Paolo 'Blaisorblade' Giarrusso <[email protected]>
Cc Roman Zippel <[email protected]>
Cc Miklos Szeredi <[email protected]>
Cc Russell King <[email protected]>
Cc Trond Myklebust <[email protected]>
Cc Neil Brown <[email protected]>

---

 fs/afs/dir.c             |    2 +-
 fs/buffer.c              |    6 +++---
 fs/freevxfs/vxfs_immed.c |    4 ++--
 fs/jffs/inode-v23.c      |    4 ++--
 fs/mpage.c               |    4 ++--
 fs/romfs/inode.c         |    6 +++---
 fs/smbfs/file.c          |    4 ++--
 fs/sysv/dir.c            |    4 ++--
 8 files changed, 17 insertions(+), 17 deletions(-)
fs/afs/dir.c renamed from fs/afs/dir.c~fix-possible-page_cache_shift-overflows +1 -1
diff -puN fs/afs/dir.c~fix-possible-page_cache_shift-overflows fs/afs/dir.c
@@ -137,7 +137,7 @@ static inline void afs_dir_check_page(st
 #endif
 
 	/* determine how many magic numbers there should be in this page */
-	latter = dir->i_size - (page->index << PAGE_CACHE_SHIFT);
+	latter = dir->i_size - page_offset(page);
 	if (latter >= PAGE_SIZE)
 		qty = PAGE_SIZE;
 	else
fs/buffer.c renamed from fs/buffer.c~fix-possible-page_cache_shift-overflows +3 -3
diff -puN fs/buffer.c~fix-possible-page_cache_shift-overflows fs/buffer.c
@@ -1762,7 +1762,7 @@ static int __block_write_full_page(struc
 	 * handle that here by just cleaning them.
 	 */
 
-	block = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
+	block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 	head = page_buffers(page);
 	bh = head;
 
@@ -2635,7 +2635,7 @@ int block_truncate_page(struct address_s
 	pgoff_t index = from >> PAGE_CACHE_SHIFT;
 	unsigned offset = from & (PAGE_CACHE_SIZE-1);
 	unsigned blocksize;
-	pgoff_t iblock;
+	sector_t iblock;
 	unsigned length, pos;
 	struct inode *inode = mapping->host;
 	struct page *page;
@@ -2651,7 +2651,7 @@ int block_truncate_page(struct address_s
 		return 0;
 
 	length = blocksize - length;
-	iblock = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
+	iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 	
 	page = grab_cache_page(mapping, index);
 	err = -ENOMEM;
fs/freevxfs/vxfs_immed.c renamed from fs/freevxfs/vxfs_immed.c~fix-possible-page_cache_shift-overflows +2 -2
diff -puN fs/freevxfs/vxfs_immed.c~fix-possible-page_cache_shift-overflows fs/freevxfs/vxfs_immed.c
@@ -99,8 +99,8 @@ static int
 vxfs_immed_readpage(struct file *fp, struct page *pp)
 {
 	struct vxfs_inode_info	*vip = VXFS_INO(pp->mapping->host);
-	u_int64_t		offset = pp->index << PAGE_CACHE_SHIFT;
-	caddr_t			kaddr;
+	u_int64_t	offset = (u_int64_t)pp->index << PAGE_CACHE_SHIFT;
+	caddr_t		kaddr;
 
 	kaddr = kmap(pp);
 	memcpy(kaddr, vip->vii_immed.vi_immed + offset, PAGE_CACHE_SIZE);
fs/jffs/inode-v23.c renamed from fs/jffs/inode-v23.c~fix-possible-page_cache_shift-overflows +2 -2
diff -puN fs/jffs/inode-v23.c~fix-possible-page_cache_shift-overflows fs/jffs/inode-v23.c
@@ -757,7 +757,7 @@ jffs_do_readpage_nolock(struct file *fil
 
 	read_len = 0;
 	result = 0;
-	offset = page->index << PAGE_CACHE_SHIFT;
+	offset = page_offset(page);
 
 	kmap(page);
 	buf = page_address(page);
@@ -1545,7 +1545,7 @@ jffs_commit_write(struct file *filp, str
 {
        void *addr = page_address(page) + from;
        /* XXX: PAGE_CACHE_SHIFT or PAGE_SHIFT */
-       loff_t pos = (page->index<<PAGE_CACHE_SHIFT) + from;
+       loff_t pos = page_offset(page) + from;
 
        return jffs_file_write(filp, addr, to-from, &pos);
 } /* jffs_commit_write() */
fs/mpage.c renamed from fs/mpage.c~fix-possible-page_cache_shift-overflows +2 -2
diff -puN fs/mpage.c~fix-possible-page_cache_shift-overflows fs/mpage.c
@@ -184,7 +184,7 @@ do_mpage_readpage(struct bio *bio, struc
 	if (page_has_buffers(page))
 		goto confused;
 
-	block_in_file = page->index << (PAGE_CACHE_SHIFT - blkbits);
+	block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
 	last_block = (i_size_read(inode) + blocksize - 1) >> blkbits;
 
 	bh.b_page = page;
@@ -466,7 +466,7 @@ __mpage_writepage(struct bio *bio, struc
 	 * The page has no buffers: map it to disk
 	 */
 	BUG_ON(!PageUptodate(page));
-	block_in_file = page->index << (PAGE_CACHE_SHIFT - blkbits);
+	block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
 	last_block = (i_size - 1) >> blkbits;
 	map_bh.b_page = page;
 	for (page_block = 0; page_block < blocks_per_page; ) {
fs/romfs/inode.c renamed from fs/romfs/inode.c~fix-possible-page_cache_shift-overflows +3 -3
diff -puN fs/romfs/inode.c~fix-possible-page_cache_shift-overflows fs/romfs/inode.c
@@ -418,7 +418,7 @@ static int
 romfs_readpage(struct file *file, struct page * page)
 {
 	struct inode *inode = page->mapping->host;
-	unsigned long offset, avail, readlen;
+	loff_t offset, avail, readlen;
 	void *buf;
 	int result = -EIO;
 
@@ -429,8 +429,8 @@ romfs_readpage(struct file *file, struct
 		goto err_out;
 
 	/* 32 bit warning -- but not for us :) */
-	offset = page->index << PAGE_CACHE_SHIFT;
-	if (offset < inode->i_size) {
+	offset = page_offset(page);
+	if (offset < i_size_read(inode)) {
 		avail = inode->i_size-offset;
 		readlen = min_t(unsigned long, avail, PAGE_SIZE);
 		if (romfs_copyfrom(inode, buf, ROMFS_I(inode)->i_dataoffset+offset, readlen) == readlen) {
fs/smbfs/file.c renamed from fs/smbfs/file.c~fix-possible-page_cache_shift-overflows +2 -2
diff -puN fs/smbfs/file.c~fix-possible-page_cache_shift-overflows fs/smbfs/file.c
@@ -209,8 +209,8 @@ smb_updatepage(struct file *file, struct
 {
 	struct dentry *dentry = file->f_dentry;
 
-	DEBUG1("(%s/%s %d@%ld)\n", DENTRY_PATH(dentry), 
-	       count, (page->index << PAGE_CACHE_SHIFT)+offset);
+	DEBUG1("(%s/%s %d@%lld)\n", DENTRY_PATH(dentry), count,
+		((unsigned long long)page->index << PAGE_CACHE_SHIFT) + offset);
 
 	return smb_writepage_sync(dentry->d_inode, page, offset, count);
 }
fs/sysv/dir.c renamed from fs/sysv/dir.c~fix-possible-page_cache_shift-overflows +2 -2
diff -puN fs/sysv/dir.c~fix-possible-page_cache_shift-overflows fs/sysv/dir.c
@@ -103,7 +103,7 @@ static int sysv_readdir(struct file * fi
 			offset = (char *)de - kaddr;
 
 			over = filldir(dirent, name, strnlen(name,SYSV_NAMELEN),
-					(n<<PAGE_CACHE_SHIFT) | offset,
+					((loff_t)n<<PAGE_CACHE_SHIFT) | offset,
 					fs16_to_cpu(SYSV_SB(sb), de->inode),
 					DT_UNKNOWN);
 			if (over) {
@@ -115,7 +115,7 @@ static int sysv_readdir(struct file * fi
 	}
 
 done:
-	filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
+	filp->f_pos = ((loff_t)n << PAGE_CACHE_SHIFT) | offset;
 	unlock_kernel();
 	return 0;
 }

_

#1 lore raw

First time I've seen this patch...

Quoted 1 line · On Sun, Jan 08, 2006 at 01:03:05AM -0800, [email protected] wrote:

- arm's __flush_dcache_page() is peculiar. What if the page lies beyond 4G?

Then it continues working as it always has done. It only uses two specific bits from page->index to determine the colour of the page.

IOW: the truncation isn't a problem.

Signature · 3 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core
#2 lore raw
In reply to [email protected]
Quoted 11 lines · On Sun, 2006-01-08 at 01:03 -0800, [email protected] wrote:

From: Andrew Morton <[email protected]>

We've had two instances recently of overflows when doing

	64_bit_value = (32_bit_value << PAGE_CACHE_SHIFT)

I did a tree-wide grep of `<<.*PAGE_CACHE_SHIFT' and this is the result.

- afs_rxfs_fetch_descriptor.offset is of type off_t, which seems broken.

- jfs and jffs are limited to 4GB anyway.

What makes you say jfs is limited to 4GB? No matter. There shouldn't be any problem in jfs anyway.

Signature · 2 lines
David Kleikamp
IBM Linux Technology Center
#3 lore raw
Quoted 8 lines · Dave Kleikamp
Quoted 6 lines · On Sun, 2006-01-08 at 01:03 -0800, [email protected] wrote:

From: Andrew Morton <[email protected]>

We've had two instances recently of overflows when doing

	64_bit_value = (32_bit_value << PAGE_CACHE_SHIFT)
Quoted 9 lines
Quoted 5 lines

I did a tree-wide grep of `<<.*PAGE_CACHE_SHIFT' and this is the result.

- afs_rxfs_fetch_descriptor.offset is of type off_t, which seems broken.

- jfs and jffs are limited to 4GB anyway.

What makes you say jfs is limited to 4GB? No matter. There shouldn't be any problem in jfs anyway.

I think I meant jffs and jffs2. Off-by-one ;)