| Message ID | 20220124104450.152481-1-ross.lagerwall@citrix.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] xen-mapcache: Avoid entry->lock overflow | expand |
On Mon, 24 Jan 2022, Ross Lagerwall wrote: > In some cases, a particular mapcache entry may be mapped 256 times > causing the lock field to wrap to 0. For example, this may happen when > using emulated NVME and the guest submits a large scatter-gather write. > At this point, the entry map be remapped causing QEMU to write the wrong > data or crash (since remap is not atomic). > > Avoid this overflow by increasing the lock field to a uint32_t and also > detect it and abort rather than continuing regardless. > > Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com> Reviewed-by: Stefano Stabellini <sstabellini@kernel.org> > --- > Changes in v2: Change type to uint32_t since there is a hole there > anyway. The struct size remains at 48 bytes on x86_64. > > hw/i386/xen/xen-mapcache.c | 8 +++++++- > 1 file changed, 7 insertions(+), 1 deletion(-) > > diff --git a/hw/i386/xen/xen-mapcache.c b/hw/i386/xen/xen-mapcache.c > index bd47c3d672..f2ef977963 100644 > --- a/hw/i386/xen/xen-mapcache.c > +++ b/hw/i386/xen/xen-mapcache.c > @@ -52,7 +52,7 @@ typedef struct MapCacheEntry { > hwaddr paddr_index; > uint8_t *vaddr_base; > unsigned long *valid_mapping; > - uint8_t lock; > + uint32_t lock; > #define XEN_MAPCACHE_ENTRY_DUMMY (1 << 0) > uint8_t flags; > hwaddr size; > @@ -355,6 +355,12 @@ tryagain: > if (lock) { > MapCacheRev *reventry = g_malloc0(sizeof(MapCacheRev)); > entry->lock++; > + if (entry->lock == 0) { > + fprintf(stderr, > + "mapcache entry lock overflow: "TARGET_FMT_plx" -> %p\n", > + entry->paddr_index, entry->vaddr_base); > + abort(); > + } > reventry->dma = dma; > reventry->vaddr_req = mapcache->last_entry->vaddr_base + address_offset; > reventry->paddr_index = mapcache->last_entry->paddr_index; > -- > 2.27.0 >
On 24/01/2022 10:44, Ross Lagerwall wrote: > In some cases, a particular mapcache entry may be mapped 256 times > causing the lock field to wrap to 0. For example, this may happen when > using emulated NVME and the guest submits a large scatter-gather write. > At this point, the entry map be remapped causing QEMU to write the wrong > data or crash (since remap is not atomic). > > Avoid this overflow by increasing the lock field to a uint32_t and also > detect it and abort rather than continuing regardless. > > Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com> Reviewed-by: Paul Durrant <paul@xen.org> > --- > Changes in v2: Change type to uint32_t since there is a hole there > anyway. The struct size remains at 48 bytes on x86_64. > > hw/i386/xen/xen-mapcache.c | 8 +++++++- > 1 file changed, 7 insertions(+), 1 deletion(-) > > diff --git a/hw/i386/xen/xen-mapcache.c b/hw/i386/xen/xen-mapcache.c > index bd47c3d672..f2ef977963 100644 > --- a/hw/i386/xen/xen-mapcache.c > +++ b/hw/i386/xen/xen-mapcache.c > @@ -52,7 +52,7 @@ typedef struct MapCacheEntry { > hwaddr paddr_index; > uint8_t *vaddr_base; > unsigned long *valid_mapping; > - uint8_t lock; > + uint32_t lock; > #define XEN_MAPCACHE_ENTRY_DUMMY (1 << 0) > uint8_t flags; > hwaddr size; > @@ -355,6 +355,12 @@ tryagain: > if (lock) { > MapCacheRev *reventry = g_malloc0(sizeof(MapCacheRev)); > entry->lock++; > + if (entry->lock == 0) { > + fprintf(stderr, > + "mapcache entry lock overflow: "TARGET_FMT_plx" -> %p\n", > + entry->paddr_index, entry->vaddr_base); > + abort(); > + } > reventry->dma = dma; > reventry->vaddr_req = mapcache->last_entry->vaddr_base + address_offset; > reventry->paddr_index = mapcache->last_entry->paddr_index;
diff --git a/hw/i386/xen/xen-mapcache.c b/hw/i386/xen/xen-mapcache.c index bd47c3d672..f2ef977963 100644 --- a/hw/i386/xen/xen-mapcache.c +++ b/hw/i386/xen/xen-mapcache.c @@ -52,7 +52,7 @@ typedef struct MapCacheEntry { hwaddr paddr_index; uint8_t *vaddr_base; unsigned long *valid_mapping; - uint8_t lock; + uint32_t lock; #define XEN_MAPCACHE_ENTRY_DUMMY (1 << 0) uint8_t flags; hwaddr size; @@ -355,6 +355,12 @@ tryagain: if (lock) { MapCacheRev *reventry = g_malloc0(sizeof(MapCacheRev)); entry->lock++; + if (entry->lock == 0) { + fprintf(stderr, + "mapcache entry lock overflow: "TARGET_FMT_plx" -> %p\n", + entry->paddr_index, entry->vaddr_base); + abort(); + } reventry->dma = dma; reventry->vaddr_req = mapcache->last_entry->vaddr_base + address_offset; reventry->paddr_index = mapcache->last_entry->paddr_index;
In some cases, a particular mapcache entry may be mapped 256 times causing the lock field to wrap to 0. For example, this may happen when using emulated NVME and the guest submits a large scatter-gather write. At this point, the entry map be remapped causing QEMU to write the wrong data or crash (since remap is not atomic). Avoid this overflow by increasing the lock field to a uint32_t and also detect it and abort rather than continuing regardless. Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com> --- Changes in v2: Change type to uint32_t since there is a hole there anyway. The struct size remains at 48 bytes on x86_64. hw/i386/xen/xen-mapcache.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)