>On Mon, 23 Aug 1999, Andrea Arcangeli wrote:
>>
>> Chih-Chung Chang pointed me out that in 2.3.x it's been added a test in
>> the CAN_UNUSE(inode) macro exactly to avoid freeing inodes with i_nrpages
>> != 0. I didn't noticed such change, sorry (so no FS corruption could
>> happen).
>>
>> But IMO such check is a dirty hack that can lead to a DoS. It seems to me
>> that it's possible to leak inodes by simply allocating lots of inodes with
>> only one page queued in them. The inode memory is not freeable.
>>
>> Now that truncate_inode_pages()/flushpage are capable of dealing with
>> dirty/protected data we don't need such hack anymore to avoid
>> fs-corruption. This patch goes on the top of my two previous
>> page-cache patches:
>
>You are wrong. This will cause "truncate_inode_pages()" to be called on
>such an inode, which will in turn flush the dirty state without ever
>writing it back to disk.
NOTENOTE: the patch was the third patch of a set of page-cache incremental
patches (you can find the other patches in thread with this subject that
started with the "ramdisk blocksize" subject). Sorry to post things
fragmented but as I said I noticed only now that the current 2.3.x code
was safe.
>In short, major filesystem corruption and general disaster.
That is _exactly_ what I thought it was just happening in 2.3.x because I
didn't noticed the CAN_UNUSE difference between 2.2.x and 2.3.x.
But before noticing CAN_UNUSE I fixed the problem properly (not with an
hack that can be exploited). Also because I need to be able to convert
page-cache-overlapped-buffers to regular buffers and the reverse to handle
the ramdisk device. A ramdisk page-cache will never be freed with
shrink_mmap so I need to be able to run flushpage even on dirty/protected
pages and let flushpage to convert them in regular buffers.
So basically now that truncate_inode_pages/flushpage are smart we can drop
the dirty-hack and apply my last patch that remove the i_nrpages check
from CAN_UNUSE.
Just to avoid further confusion with incremental patches, here it is the
whole patch against clean 2.3.15-pre2. This patch will fix also the
ramdisk driver.
diff -urN 2.3.15-pre2/fs/buffer.c 2.3.15-pre2-pagecache3/fs/buffer.c
--- 2.3.15-pre2/fs/buffer.c Thu Aug 12 02:53:23 1999
+++ 2.3.15-pre2-pagecache3/fs/buffer.c Mon Aug 23 15:15:44 1999
@@ -793,6 +793,10 @@
repeat:
bh = get_hash_table(dev, block, size);
if (bh) {
+ /* we may have found a orhpaned-hashed buffer, and so it
+ may have a end_io handler different than the io_sync one. */
+ bh->b_end_io = end_buffer_io_sync;
+ mb();
if (!buffer_dirty(bh)) {
bh->b_flushtime = 0;
}
@@ -1232,6 +1236,34 @@
return 0;
}
+static void hash_orphaned_buffers(struct page * page)
+{
+ struct buffer_head * bh = page->buffers, * first = bh;
+
+ do {
+ struct buffer_head ** head;
+ if (buffer_mapped(bh))
+ {
+ head = &hash(bh->b_dev, bh->b_blocknr);
+ write_lock(&hash_table_lock);
+ __hash_link(bh, head);
+ write_unlock(&hash_table_lock);
+ }
+ bh = bh->b_this_page;
+ } while (bh != first);
+}
+
+static void trash_buffer(struct buffer_head * bh)
+{
+ wait_on_buffer(bh);
+ mark_buffer_clean(bh);
+ clear_bit(BH_Uptodate, &bh->b_state);
+ clear_bit(BH_Mapped, &bh->b_state);
+ clear_bit(BH_Req, &bh->b_state);
+ clear_bit(BH_Protected, &bh->b_state);
+ bh->b_blocknr = -1;
+}
+
/*
* We don't have to release all buffers here, but
* we have to be sure that no dirty buffer is left
@@ -1239,7 +1271,7 @@
* we have truncated the file and are going to free the
* blocks on-disk..
*/
-int block_flushpage(struct inode *inode, struct page *page, unsigned long offset)
+int block_flushpage(struct inode *inode, struct page *page, unsigned long offset, int forget)
{
struct buffer_head *head, *bh, *next;
unsigned int curr_off = 0;
@@ -1260,16 +1292,14 @@
*/
if (offset <= curr_off) {
if (buffer_mapped(bh)) {
- atomic_inc(&bh->b_count);
- wait_on_buffer(bh);
if (bh->b_dev == B_FREE)
BUG();
- mark_buffer_clean(bh);
- clear_bit(BH_Uptodate, &bh->b_state);
- clear_bit(BH_Mapped, &bh->b_state);
- clear_bit(BH_Req, &bh->b_state);
- bh->b_blocknr = 0;
- atomic_dec(&bh->b_count);
+ if (forget)
+ /* the page can't be freed while we are
+ working on it since we hold the
+ page-lock and so shrink_mmap
+ won't race with us. */
+ trash_buffer(bh);
}
}
curr_off = next_off;
@@ -1288,7 +1318,13 @@
*/
if (!offset) {
if (!try_to_free_buffers(page))
+ {
atomic_add(PAGE_CACHE_SIZE, &buffermem);
+ if (!forget)
+ /* this must become a regular buffer so
+ we'll find it then. */
+ hash_orphaned_buffers(page);
+ }
}
return 0;
@@ -1455,11 +1491,39 @@
if (buffer_new(bh)) {
memset(bh->b_data, 0, bh->b_size);
} else {
- ll_rw_block(READ, 1, &bh);
- wait_on_buffer(bh);
- err = -EIO;
- if (!buffer_uptodate(bh))
- goto out;
+ struct buffer_head * bh_old;
+
+ bh_old = get_hash_table(bh->b_dev,
+ bh->b_blocknr,
+ bh->b_size);
+ if (!bh_old || !buffer_uptodate(bh_old))
+ {
+ if (bh_old)
+ {
+ trash_buffer(bh_old);
+ __remove_from_queues(bh_old);
+ atomic_dec(&bh_old->b_count);
+ }
+ ll_rw_block(READ, 1, &bh);
+ wait_on_buffer(bh);
+ err = -EIO;
+ if (!buffer_uptodate(bh))
+ goto out;
+ }
+ else
+ {
+ memcpy(bh->b_data, bh_old->b_data,
+ bh->b_size);
+ /* move the protected bit from
+ the old buffer to the new buffer. */
+ if (atomic_expose_buffer(bh_old))
+ atomic_protect_buffer(bh);
+
+ /* now trash the old regular buffer */
+ trash_buffer(bh_old);
+ __remove_from_queues(bh_old);
+ atomic_dec(&bh_old->b_count);
+ }
}
}
@@ -1617,11 +1681,39 @@
if (buffer_new(bh)) {
memset(bh->b_data, 0, bh->b_size);
} else {
- ll_rw_block(READ, 1, &bh);
- wait_on_buffer(bh);
- err = -EIO;
- if (!buffer_uptodate(bh))
- goto out;
+ struct buffer_head * bh_old;
+
+ bh_old = get_hash_table(bh->b_dev,
+ bh->b_blocknr,
+ bh->b_size);
+ if (!bh_old || !buffer_uptodate(bh_old))
+ {
+ if (bh_old)
+ {
+ trash_buffer(bh_old);
+ __remove_from_queues(bh_old);
+ atomic_dec(&bh_old->b_count);
+ }
+ ll_rw_block(READ, 1, &bh);
+ wait_on_buffer(bh);
+ err = -EIO;
+ if (!buffer_uptodate(bh))
+ goto out;
+ }
+ else
+ {
+ memcpy(bh->b_data, bh_old->b_data,
+ bh->b_size);
+ /* move the protected bit from
+ the old buffer to the new buffer. */
+ if (atomic_expose_buffer(bh_old))
+ atomic_protect_buffer(bh);
+
+ /* now trash the old regular buffer */
+ trash_buffer(bh_old);
+ __remove_from_queues(bh_old);
+ atomic_dec(&bh_old->b_count);
+ }
}
}
@@ -2036,7 +2128,54 @@
memset(bh->b_data, 0, blocksize);
set_bit(BH_Uptodate, &bh->b_state);
continue;
- }
+ } else {
+ struct buffer_head * bh_old;
+
+ bh_old = get_hash_table(bh->b_dev,
+ bh->b_blocknr,
+ bh->b_size);
+ if (bh_old && buffer_uptodate(bh_old))
+ {
+ memcpy(bh->b_data, bh_old->b_data,
+ bh->b_size);
+ set_bit(BH_Uptodate, &bh->b_state);
+ /* move the protected bit from
+ the old buffer to the new buffer. */
+ if (atomic_expose_buffer(bh_old))
+ atomic_protect_buffer(bh);
+ if (atomic_set_buffer_clean(bh_old))
+ {
+ /* the buffer will do IO
+ so it must have a valid
+ end_io handler before
+ setting it dirty. */
+ bh->b_end_io = end_buffer_io_sync;
+ /* even if to refile the buffer
+ will run a spin_lock(), here
+ is more robust to enforce
+ a SMP memory barrier by hand
+ because a buffer is writable
+ only in function of the
+ dirty bit and not in
+ function of where it's
+ placed in the lru lists. */
+ mb();
+ mark_buffer_dirty(bh, 0);
+ }
+
+ /* now trash the old regular buffer */
+ trash_buffer(bh_old);
+ __remove_from_queues(bh_old);
+ atomic_dec(&bh_old->b_count);
+ continue;
+ }
+ if (bh_old)
+ {
+ trash_buffer(bh_old);
+ __remove_from_queues(bh_old);
+ atomic_dec(&bh_old->b_count);
+ }
+ }
}
init_buffer(bh, end_buffer_io_async, NULL);
diff -urN 2.3.15-pre2/fs/inode.c 2.3.15-pre2-pagecache3/fs/inode.c
--- 2.3.15-pre2/fs/inode.c Tue Jul 13 02:02:40 1999
+++ 2.3.15-pre2-pagecache3/fs/inode.c Mon Aug 23 15:44:45 1999
@@ -263,7 +263,7 @@
break;
inode = list_entry(tmp, struct inode, i_list);
if (inode->i_nrpages)
- truncate_inode_pages(inode, 0);
+ truncate_inode_pages(inode, 0, 0);
clear_inode(inode);
count++;
}
@@ -339,7 +339,7 @@
* dispose_list.
*/
#define CAN_UNUSE(inode) \
- (((inode)->i_count | (inode)->i_state | (inode)->i_nrpages) == 0)
+ (((inode)->i_count | (inode)->i_state) == 0)
#define INODE(entry) (list_entry(entry, struct inode, i_list))
static int free_inodes(void)
@@ -739,7 +739,7 @@
void (*delete)(struct inode *) = op->delete_inode;
spin_unlock(&inode_lock);
if (inode->i_nrpages)
- truncate_inode_pages(inode, 0);
+ truncate_inode_pages(inode, 0, 1);
delete(inode);
spin_lock(&inode_lock);
}
diff -urN 2.3.15-pre2/include/linux/fs.h 2.3.15-pre2-pagecache3/include/linux/fs.h
--- 2.3.15-pre2/include/linux/fs.h Sat Aug 21 16:45:36 1999
+++ 2.3.15-pre2-pagecache3/include/linux/fs.h Mon Aug 23 15:50:55 1999
@@ -627,7 +627,7 @@
int (*readpage) (struct file *, struct page *);
int (*writepage) (struct file *, struct page *);
- int (*flushpage) (struct inode *, struct page *, unsigned long);
+ int (*flushpage) (struct inode *, struct page *, unsigned long, int);
void (*truncate) (struct inode *);
int (*permission) (struct inode *, int);
@@ -780,6 +780,9 @@
clear_bit(BH_Uptodate, &bh->b_state);
}
+#define atomic_expose_buffer(bh) test_and_clear_bit(BH_Protected, &(bh)->b_state)
+#define atomic_protect_buffer(bh) test_and_set_bit(BH_Protected, &(bh)->b_state)
+
#define atomic_set_buffer_clean(bh) test_and_clear_bit(BH_Dirty, &(bh)->b_state)
extern inline void __mark_buffer_clean(struct buffer_head *bh)
@@ -899,7 +902,7 @@
extern int block_write_full_page (struct file *, struct page *);
extern int block_write_partial_page (struct file *, struct page *, unsigned long, unsigned long, const char *);
extern int block_write_cont_page (struct file *, struct page *, unsigned long, unsigned long, const char *);
-extern int block_flushpage(struct inode *, struct page *, unsigned long);
+extern int block_flushpage(struct inode *, struct page *, unsigned long, int);
extern int generic_file_mmap(struct file *, struct vm_area_struct *);
extern ssize_t generic_file_read(struct file *, char *, size_t, loff_t *);
diff -urN 2.3.15-pre2/include/linux/mm.h 2.3.15-pre2-pagecache3/include/linux/mm.h
--- 2.3.15-pre2/include/linux/mm.h Sat Aug 21 16:45:36 1999
+++ 2.3.15-pre2-pagecache3/include/linux/mm.h Mon Aug 23 15:50:55 1999
@@ -344,7 +344,7 @@
extern void remove_inode_page(struct page *);
extern unsigned long page_unuse(struct page *);
extern int shrink_mmap(int, int);
-extern void truncate_inode_pages(struct inode *, unsigned long);
+extern void truncate_inode_pages(struct inode *, unsigned long, int);
extern unsigned long get_cached_page(struct inode *, unsigned long, int);
extern void put_cached_page(unsigned long);
diff -urN 2.3.15-pre2/mm/filemap.c 2.3.15-pre2-pagecache3/mm/filemap.c
--- 2.3.15-pre2/mm/filemap.c Sun Aug 22 14:33:47 1999
+++ 2.3.15-pre2-pagecache3/mm/filemap.c Mon Aug 23 15:15:44 1999
@@ -132,7 +132,8 @@
* Truncate the page cache at a set offset, removing the pages
* that are beyond that offset (and zeroing out partial pages).
*/
-void truncate_inode_pages(struct inode * inode, unsigned long start)
+void truncate_inode_pages(struct inode * inode, unsigned long start,
+ int forget)
{
struct page ** p;
struct page * page;
@@ -152,7 +153,7 @@
lock_page(page);
if (inode->i_op->flushpage)
- inode->i_op->flushpage(inode, page, 0);
+ inode->i_op->flushpage(inode, page, 0, forget);
/*
* We remove the page from the page cache
@@ -199,7 +200,7 @@
flush_page_to_ram(address);
if (inode->i_op->flushpage)
- inode->i_op->flushpage(inode, page, offset);
+ inode->i_op->flushpage(inode, page, offset, forget);
/*
* we have dropped the spinlock so we have to
* restart.
diff -urN 2.3.15-pre2/mm/memory.c 2.3.15-pre2-pagecache3/mm/memory.c
--- 2.3.15-pre2/mm/memory.c Sun Aug 22 14:33:47 1999
+++ 2.3.15-pre2-pagecache3/mm/memory.c Mon Aug 23 15:15:44 1999
@@ -890,7 +890,7 @@
{
struct vm_area_struct * mpnt;
- truncate_inode_pages(inode, offset);
+ truncate_inode_pages(inode, offset, 1);
spin_lock(&inode->i_shared_lock);
if (!inode->i_mmap)
goto out_unlock;
diff -urN 2.3.15-pre2/mm/swap_state.c 2.3.15-pre2-pagecache3/mm/swap_state.c
--- 2.3.15-pre2/mm/swap_state.c Thu Aug 12 02:53:25 1999
+++ 2.3.15-pre2-pagecache3/mm/swap_state.c Mon Aug 23 15:15:44 1999
@@ -215,7 +215,7 @@
#endif
PageClearSwapCache(page);
if (inode->i_op->flushpage)
- inode->i_op->flushpage(inode, page, 0);
+ inode->i_op->flushpage(inode, page, 0, 1);
remove_inode_page(page);
}
Andrea
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/