On 2024/5/6 10:55, Jingbo Xu wrote:
On 4/24/24 11:39 AM, libaokun@xxxxxxxxxxxxxxx wrote:
From: Baokun Li <libaokun1@xxxxxxxxxx>
The following concurrency may cause a read request to fail to be completed
and result in a hung:
t1 | t2
---------------------------------------------------------
cachefiles_ondemand_copen
req = xa_erase(&cache->reqs, id)
// Anon fd is maliciously closed.
cachefiles_ondemand_fd_release
xa_lock(&cache->reqs)
cachefiles_ondemand_set_object_close(object)
xa_unlock(&cache->reqs)
cachefiles_ondemand_set_object_open
// No one will ever close it again.
cachefiles_ondemand_daemon_read
cachefiles_ondemand_select_req
// Get a read req but its fd is already closed.
// The daemon can't issue a cread ioctl with an closed fd, then hung.
So add spin_lock for cachefiles_ondemand_info to protect ondemand_id and
state, thus we can avoid the above problem in cachefiles_ondemand_copen()
by using ondemand_id to determine if fd has been released.
Signed-off-by: Baokun Li <libaokun1@xxxxxxxxxx>
This indeed looks like a reasonable scenario where the kernel side
should fix, as a non-malicious daemon could also run into this.
How about reusing &cache->reqs spinlock rather than introducing a new
spinlock, as &cache->reqs spinlock is already held when setting object
to close state in cachefiles_ondemand_fd_release()?
We've considered reusing &cache->reqs spinlock before, but their
uses don't exactly overlap, and there are patches coming that will
use the new spin_lock,. In addition, this reduces competition for
&cache->reqs spinlock.
---
fs/cachefiles/internal.h | 1 +
fs/cachefiles/ondemand.c | 16 +++++++++++++++-
2 files changed, 16 insertions(+), 1 deletion(-)
diff --git a/fs/cachefiles/internal.h b/fs/cachefiles/internal.h
index 7745b8abc3aa..45c8bed60538 100644
--- a/fs/cachefiles/internal.h
+++ b/fs/cachefiles/internal.h
@@ -55,6 +55,7 @@ struct cachefiles_ondemand_info {
int ondemand_id;
enum cachefiles_object_state state;
struct cachefiles_object *object;
+ spinlock_t lock;
};
/*
diff --git a/fs/cachefiles/ondemand.c b/fs/cachefiles/ondemand.c
index 898fab68332b..b5e6a851ef04 100644
--- a/fs/cachefiles/ondemand.c
+++ b/fs/cachefiles/ondemand.c
@@ -16,13 +16,16 @@ static int cachefiles_ondemand_fd_release(struct inode *inode,
struct cachefiles_object *object = file->private_data;
struct cachefiles_cache *cache = object->volume->cache;
struct cachefiles_ondemand_info *info = object->ondemand;
- int object_id = info->ondemand_id;
+ int object_id;
struct cachefiles_req *req;
XA_STATE(xas, &cache->reqs, 0);
xa_lock(&cache->reqs);
+ spin_lock(&info->lock);
+ object_id = info->ondemand_id;
info->ondemand_id = CACHEFILES_ONDEMAND_ID_CLOSED;
cachefiles_ondemand_set_object_close(object);
+ spin_unlock(&info->lock);
/* Only flush CACHEFILES_REQ_NEW marked req to avoid race with daemon_read */
xas_for_each_marked(&xas, req, ULONG_MAX, CACHEFILES_REQ_NEW) {
@@ -127,6 +130,7 @@ int cachefiles_ondemand_copen(struct cachefiles_cache *cache, char *args)
{
struct cachefiles_req *req;
struct fscache_cookie *cookie;
+ struct cachefiles_ondemand_info *info;
char *pid, *psize;
unsigned long id;
long size;
@@ -185,6 +189,14 @@ int cachefiles_ondemand_copen(struct cachefiles_cache *cache, char *args)
goto out;
}
+ info = req->object->ondemand;
+ spin_lock(&info->lock);
+ /* The anonymous fd was closed before copen ? */
I would like describe more details in the comment, e.g. put the time
sequence described in the commit message here.
OK, thanks for your suggestion, I will describe it in more detail
in the next revision.
Thanks,
Baokun
+ if (info->ondemand_id == CACHEFILES_ONDEMAND_ID_CLOSED) {
+ spin_unlock(&info->lock);
+ req->error = -EBADFD;
+ goto out;
+ }
cookie = req->object->cookie;
cookie->object_size = size;
if (size)
@@ -194,6 +206,7 @@ int cachefiles_ondemand_copen(struct cachefiles_cache *cache, char *args)
trace_cachefiles_ondemand_copen(req->object, id, size);
cachefiles_ondemand_set_object_open(req->object);
+ spin_unlock(&info->lock);
wake_up_all(&cache->daemon_pollwq);
out:
@@ -596,6 +609,7 @@ int cachefiles_ondemand_init_obj_info(struct cachefiles_object *object,
return -ENOMEM;
object->ondemand->object = object;
+ spin_lock_init(&object->ondemand->lock);
INIT_WORK(&object->ondemand->ondemand_work, ondemand_object_worker);
return 0;
}
