iomap: fix a use after free in iomap_dio_rw
PATCH 1/1 iomap: fix a use after free in iomap_dio_rw
Introduce a local wait_for_completion variable to avoid an access to the potentially freed dio struture after dropping the last reference count.
Also use the chance to document the completion behavior to make the refcounting clear to the reader of the code.
--- fs/iomap.c | 28 +++++++++++++++++++++------- 1 file changed, 21 insertions(+), 7 deletions(-)
fs/iomap.c +21 -7
@@ -1813,6 +1813,7 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,loff_t pos = iocb->ki_pos, start = pos;loff_t end = iocb->ki_pos + count -1 , ret =0 ;unsigned int flags =IOMAP_DIRECT ; +bool wait_for_completion =is_sync_kiocb (iocb);struct blk_plug plug;struct iomap_dio *dio;
@@ -1832,7 +1833,6 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter, dio->end_io = end_io; dio->error =0 ; dio->flags =0 ; - dio->wait_for_completion =is_sync_kiocb (iocb); dio->submit.iter = iter; dio->submit.waiter = current;
@@ -1887,7 +1887,7 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,dio_warn_stale_pagecache (iocb->ki_filp); ret =0 ; -if (iov_iter_rw (iter) ==WRITE && !dio->wait_for_completion && +if (iov_iter_rw (iter) ==WRITE && !wait_for_completion && !inode->i_sb->s_dio_done_wq) { ret =sb_init_dio_done_wq (inode->i_sb);if (ret <0 )
@@ -1903,7 +1903,7 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,if (ret <=0 ) {/* magic error code to fall back to buffered I/O */ if (ret == -ENOTBLK ) { - dio->wait_for_completion = true; + wait_for_completion = true; ret =0 ; }break ;
@@ -1925,8 +1925,24 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,if (dio->flags &IOMAP_DIO_WRITE_FUA ) dio->flags &= ~IOMAP_DIO_NEED_SYNC ; +/* +* We are about to drop our additional submission reference, which +* might be the last reference to the dio. There are three three +* different ways we can progress here: +* +* (a) If this is the last reference we will always complete and free +* the dio ourselves. right here. +* (b) If this is not the last reference, and we serve an asynchronous +* iocb, we must never touch the dio after the decrement, the +* I/O completion handler will complete and free it. +* (c) If this is not the last reference, but we serve a synchronous +* iocb, the I/O completion handler will wake us up on the drop +* of the final reference, and we will complete and free it here +* after we got woken by the I/O completion handler. +*/ + dio->wait_for_completion = wait_for_completion;if (!atomic_dec_and_test (&dio->ref)) { -if (!dio->wait_for_completion) +if (!wait_for_completion)return -EIOCBQUEUED ;for (;;) {
@@ -1943,9 +1959,7 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,__set_current_state (TASK_RUNNING ); } - ret =iomap_dio_complete (dio); - -return ret; +return iomap_dio_complete (dio); out_free_dio:kfree (dio);
Signature · 1 line
2.20.1
Quoted 12 lines · Christoph Hellwig
Introduce a local
wait_for_completionvariable to avoid an access to the potentially freed dio struture after dropping the last reference count.Also use the chance to document the completion behavior to make the refcounting clear to the reader of the code.
Fixes ff6a9292e6 ("iomap: implement direct I/O")Reported-by Chandan Rajendra <[email protected]>Reported-by Darrick J. Wong <[email protected]>Signed-off-by Christoph Hellwig <[email protected]>Tested-by Chandan Rajendra <[email protected]>Tested-by Darrick J. Wong <[email protected]>
Quoted 3 lines
--- fs/iomap.c | 28 +++++++++++++++++++++------- 1 file changed, 21 insertions(+), 7 deletions(-)
Looks good, minor comment cleanup needed, though. maybe Darrick can do it on commit?
fs/iomap.c iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter, · line 1934
Quoted 11 lines
fs/iomap.c +7
-0@@ -1925,8 +1925,24 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,if (dio->flags &IOMAP_DIO_WRITE_FUA ) dio->flags &= ~IOMAP_DIO_NEED_SYNC ; +/* +* We are about to drop our additional submission reference, which +* might be the last reference to the dio. There are three three +* different ways we can progress here: +* +* (a) If this is the last reference we will always complete and free +* the dio ourselves. right here.
s/ right here.//
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Darrick J. Wong
Reviewed-byQuoted 26 lines · Dave Chinner
Quoted 12 lines · Christoph Hellwig
Introduce a local
wait_for_completionvariable to avoid an access to the potentially freed dio struture after dropping the last reference count.Also use the chance to document the completion behavior to make the refcounting clear to the reader of the code.
Fixes ff6a9292e6 ("iomap: implement direct I/O")Reported-by Chandan Rajendra <[email protected]>Reported-by Darrick J. Wong <[email protected]>Signed-off-by Christoph Hellwig <[email protected]>Tested-by Chandan Rajendra <[email protected]>Tested-by Darrick J. Wong <[email protected]>Quoted 3 lines
--- fs/iomap.c | 28 +++++++++++++++++++++------- 1 file changed, 21 insertions(+), 7 deletions(-)Looks good, minor comment cleanup needed, though. maybe Darrick can do it on commit?
Reviewed-by Dave Chinner <[email protected]>Quoted 4 lines
fs/iomap.c +0
-0@@ -1925,8 +1925,24 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,if (dio->flags &IOMAP_DIO_WRITE_FUA ) dio->flags &= ~IOMAP_DIO_NEED_SYNC ;
Quoted 9 lines
fs/iomap.c iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter, · line 1934
Quoted 7 lines
fs/iomap.c +7
-0+/* +* We are about to drop our additional submission reference, which +* might be the last reference to the dio. There are three three +* different ways we can progress here: +* +* (a) If this is the last reference we will always complete and free +* the dio ourselves. right here. s/ right here.//
Fixed.
--D
Quoted 4 lines
-Dave.
Signature · 2 lines
Dave Chinner [email protected]
Quoted 3 lines · Darrick J. Wong
Fixed.
Reviewed-by Darrick J. Wong <[email protected]>
Are you going to pick this one (and the migrate page fix) up for 5.0-rc?
Quoted 7 lines · Christoph Hellwig
Quoted 3 lines · Darrick J. Wong
Fixed.
Reviewed-by Darrick J. Wong <[email protected]>Are you going to pick this one (and the migrate page fix) up for 5.0-rc?
Yes. I've been mostly offline for a week or so now because last year's health problems came back, though I'm now patched up enough to push out the branch that was running endlessly (yay more fsx) while I was gone.
--D