| Message ID | 20230427143142.3013020-1-tudor.cretu@arm.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | io_uring/kbuf: Fix size for shared buffer ring | expand |
On 4/27/23 8:31?AM, Tudor Cretu wrote: > The size of the ring is the product of ring_entries and the size of > struct io_uring_buf. Using struct_size is equivalent to > (ring_entries + 1) * sizeof(struct io_uring_buf) > and generates an off-by-one error. Fix it by using size_mul directly. Looks right, so we're doing one entry too much. This is probably a remnant of when the initial version had it done a bit differently, but we overlaid the tail with the first entry reserved field.
On 4/27/23 8:31 AM, Tudor Cretu wrote: > The size of the ring is the product of ring_entries and the size of > struct io_uring_buf. Using struct_size is equivalent to > (ring_entries + 1) * sizeof(struct io_uring_buf) > and generates an off-by-one error. Fix it by using size_mul directly. > > Signed-off-by: Tudor Cretu <tudor.cretu@arm.com> > --- > io_uring/kbuf.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/io_uring/kbuf.c b/io_uring/kbuf.c > index 4a6401080c1f..9770757c89a0 100644 > --- a/io_uring/kbuf.c > +++ b/io_uring/kbuf.c > @@ -505,7 +505,7 @@ int io_register_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg) > } > > pages = io_pin_pages(reg.ring_addr, > - struct_size(br, bufs, reg.ring_entries), > + size_mul(sizeof(struct io_uring_buf), reg.ring_entries), > &nr_pages); > if (IS_ERR(pages)) { > kfree(free_bl); Looking into this again, and some bells ringing in the back of my head, we do have: commit 48ba08374e779421ca34bd14b4834aae19fc3e6a Author: Wojciech Lukowicz <wlukowicz01@gmail.com> Date: Sat Feb 18 18:41:41 2023 +0000 io_uring: fix size calculation when registering buf ring which should have fixed that issue. What kernel version are you looking at?
On 27-04-2023 19:42, Jens Axboe wrote: > On 4/27/23 8:31 AM, Tudor Cretu wrote: >> The size of the ring is the product of ring_entries and the size of >> struct io_uring_buf. Using struct_size is equivalent to >> (ring_entries + 1) * sizeof(struct io_uring_buf) >> and generates an off-by-one error. Fix it by using size_mul directly. >> >> Signed-off-by: Tudor Cretu <tudor.cretu@arm.com> >> --- >> io_uring/kbuf.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/io_uring/kbuf.c b/io_uring/kbuf.c >> index 4a6401080c1f..9770757c89a0 100644 >> --- a/io_uring/kbuf.c >> +++ b/io_uring/kbuf.c >> @@ -505,7 +505,7 @@ int io_register_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg) >> } >> >> pages = io_pin_pages(reg.ring_addr, >> - struct_size(br, bufs, reg.ring_entries), >> + size_mul(sizeof(struct io_uring_buf), reg.ring_entries), >> &nr_pages); >> if (IS_ERR(pages)) { >> kfree(free_bl); > > Looking into this again, and some bells ringing in the back of my head, > we do have: > > commit 48ba08374e779421ca34bd14b4834aae19fc3e6a > Author: Wojciech Lukowicz <wlukowicz01@gmail.com> > Date: Sat Feb 18 18:41:41 2023 +0000 > > io_uring: fix size calculation when registering buf ring > > which should have fixed that issue. What kernel version are you looking at? Hi Jens, Thank you for your message. Indeed I was looking at a slightly older version of the kernel. Apologies for the noise! Kind regards, Tudor >
diff --git a/io_uring/kbuf.c b/io_uring/kbuf.c index 4a6401080c1f..9770757c89a0 100644 --- a/io_uring/kbuf.c +++ b/io_uring/kbuf.c @@ -505,7 +505,7 @@ int io_register_pbuf_ring(struct io_ring_ctx *ctx, void __user *arg) } pages = io_pin_pages(reg.ring_addr, - struct_size(br, bufs, reg.ring_entries), + size_mul(sizeof(struct io_uring_buf), reg.ring_entries), &nr_pages); if (IS_ERR(pages)) { kfree(free_bl);
The size of the ring is the product of ring_entries and the size of struct io_uring_buf. Using struct_size is equivalent to (ring_entries + 1) * sizeof(struct io_uring_buf) and generates an off-by-one error. Fix it by using size_mul directly. Signed-off-by: Tudor Cretu <tudor.cretu@arm.com> --- io_uring/kbuf.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)