Message ID | 1542963889-128825-1-git-send-email-bianpan2016@163.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | hfs: do not free node before using | expand |
On Fri, 2018-11-23 at 17:04 +0800, Pan Bian wrote: > The function hfs_bmap_free frees node via hfs_bnode_put(node). However, > it then reads node->this when dumping error message on an error path, > which may result in a use-after-free bug. This patch frees node only > when it is never used. > > Fixes: d614267329f("hfs/hfsplus: convert printks to pr_<level>") Hi. While this may indeed be a defect, and the "/* panic */" comment may be unwarranted, this isn't really a fix of a printk conversion. This dereference goes back to 2004, the printk(KERN_CRIT to pr_crit( conversion did not introduce it. So this patch is only a possible use after free fix. From a full history git tree: (similar to https://archive.org/details/git-history-of-linux) commit a1185ffa2fc491e23f3107a39f66ee703d102153 Author: Andrew Morton <akpm@osdl.org> Date: Wed Feb 25 16:17:36 2004 -0800 [PATCH] HFS rewrite From: Roman Zippel <zippel@linux-m68k.org> This is a complete rewrite of the HFS driver, it gets rid of a all the special conversion options, which belong in user space. The driver uses now a btree support very similiar to HFS+, so that both could be merged at some point. Thanks to Ethan Benson <erbenson@alaska.net> for a number of patches to make the driver more compliant with the spec and Christoph Hellwig <hch@lst.de> for fixing up the documentation. > diff --git a/fs/hfs/btree.c b/fs/hfs/btree.c [] > @@ -338,13 +338,14 @@ void hfs_bmap_free(struct hfs_bnode *node) > > nidx -= len * 8; > i = node->next; > - hfs_bnode_put(node); > if (!i) { > /* panic */; > pr_crit("unable to free bnode %u. bmap not found!\n", > node->this); > + hfs_bnode_put(node); > return; > } > + hfs_bnode_put(node); > node = hfs_bnode_find(tree, i); > if (IS_ERR(node)) > return;
On Fri, 2018-11-23 at 17:04 +0800, Pan Bian wrote: > The function hfs_bmap_free frees node via hfs_bnode_put(node). > However, > it then reads node->this when dumping error message on an error path, > which may result in a use-after-free bug. This patch frees node only > when it is never used. > > Fixes: d614267329f("hfs/hfsplus: convert printks to pr_<level>") > > Signed-off-by: Pan Bian <bianpan2016@163.com> > --- > fs/hfs/btree.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/fs/hfs/btree.c b/fs/hfs/btree.c > index 98b96ff..19017d2 100644 > --- a/fs/hfs/btree.c > +++ b/fs/hfs/btree.c > @@ -338,13 +338,14 @@ void hfs_bmap_free(struct hfs_bnode *node) > > nidx -= len * 8; > i = node->next; > - hfs_bnode_put(node); > if (!i) { > /* panic */; > pr_crit("unable to free bnode %u. bmap not > found!\n", > node->this); What's about simply to store node->this into the local variable in the beginning of the cycle? In this case, it is possible to use the node ID in the pr_crit() and to use the hfs_bnode_put(node) in the same place without to use it twice. What do you think? Thanks, Vyacheslav Dubeyko. > + hfs_bnode_put(node); > return; > } > + hfs_bnode_put(node); > node = hfs_bnode_find(tree, i); > if (IS_ERR(node)) > return;
On Fri, Nov 23, 2018 at 04:51:49PM -0800, Viacheslav Dubeyko wrote: > On Fri, 2018-11-23 at 17:04 +0800, Pan Bian wrote: > > The function hfs_bmap_free frees node via hfs_bnode_put(node). > > However, > > it then reads node->this when dumping error message on an error path, > > which may result in a use-after-free bug. This patch frees node only > > when it is never used. > > > > Fixes: d614267329f("hfs/hfsplus: convert printks to pr_<level>") > > > > Signed-off-by: Pan Bian <bianpan2016@163.com> > > --- > > fs/hfs/btree.c | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/fs/hfs/btree.c b/fs/hfs/btree.c > > index 98b96ff..19017d2 100644 > > --- a/fs/hfs/btree.c > > +++ b/fs/hfs/btree.c > > @@ -338,13 +338,14 @@ void hfs_bmap_free(struct hfs_bnode *node) > > > > nidx -= len * 8; > > i = node->next; > > - hfs_bnode_put(node); > > if (!i) { > > /* panic */; > > pr_crit("unable to free bnode %u. bmap not > > found!\n", > > node->this); > > > What's about simply to store node->this into the local variable in the > beginning of the cycle? In this case, it is possible to use the node ID > in the pr_crit() and to use the hfs_bnode_put(node) in the same place > without to use it twice. What do you think? Of course, it is a solution. But I think doing like that does not really simplify the code, as we have to declare an extra local variable, and assign node->this to it... Best, Pan > > Thanks, > Vyacheslav Dubeyko. > > > > + hfs_bnode_put(node); > > return; > > } > > + hfs_bnode_put(node); > > node = hfs_bnode_find(tree, i); > > if (IS_ERR(node)) > > return;
diff --git a/fs/hfs/btree.c b/fs/hfs/btree.c index 98b96ff..19017d2 100644 --- a/fs/hfs/btree.c +++ b/fs/hfs/btree.c @@ -338,13 +338,14 @@ void hfs_bmap_free(struct hfs_bnode *node) nidx -= len * 8; i = node->next; - hfs_bnode_put(node); if (!i) { /* panic */; pr_crit("unable to free bnode %u. bmap not found!\n", node->this); + hfs_bnode_put(node); return; } + hfs_bnode_put(node); node = hfs_bnode_find(tree, i); if (IS_ERR(node)) return;
The function hfs_bmap_free frees node via hfs_bnode_put(node). However, it then reads node->this when dumping error message on an error path, which may result in a use-after-free bug. This patch frees node only when it is never used. Fixes: d614267329f("hfs/hfsplus: convert printks to pr_<level>") Signed-off-by: Pan Bian <bianpan2016@163.com> --- fs/hfs/btree.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)