Hello, I managed to reproduce something similiar with small buffers...
Does attached patch help in your case? -Jussi Quoting Jorgen Lundman <[email protected]>:
I have a situation where I setup scatterlists as: input scatterlist of 1, address ffffc90003627000 len 0x20000. output scatterlist of 2, address 0 ffffc90002d45000 len 0x20000 address 1 ffff88003b079d98 len 0x000c When I call crypto_aead_encrypt(req); it will die with: kernel: [ 925.151113] BUG: unable to handle kernel paging request at ffffeb04000b5140 kernel: [ 925.151253] IP: [<ffffffff812f4880>] scatterwalk_done+0x50/0x60 kernel: [ 925.151325] PGD 0 kernel: [ 925.151381] Oops: 0000 [#1] SMP kernel: [ 925.151442] CPU 1 kernel: [ 925.154255] [<ffffffff812f7640>] blkcipher_walk_done+0xb0/0x230kernel: [ 925.154255] [<ffffffffa02e9169>] crypto_ctr_crypt+0x129/0x2b0 [ctr]kernel: [ 925.154255] [<ffffffff812fe580>] ? crypto_aes_set_key+0x40/0x40 kernel: [ 925.154255] [<ffffffff812f6cbd>] async_encrypt+0x3d/0x40 kernel: [ 925.154255] [<ffffffffa0149326>] crypto_ccm_encrypt+0x246/0x290 [ccm] kernel: [ 925.154255] [<ffffffffa01633bd>] crypto_encrypt+0x26d/0x2d0 What is interesting about that is, if I allocate a linear buffer instead: dst = kmalloc(cryptlen, GFP_KERNEL); // 0x20000 + 0x000c sg_init_table(sg, 1 ); sg_set_buf(&sg[0], dst, cryptlen); crypto_aead_encrypt(req); will no longer panic. However, when I try to copy the linear buffer back to scatterlist; scatterwalk_map_and_copy(dst, sg, 0, cryptlen, 1); then it will panic there instead. However, if I replace it with the call: sg_copy_from_buffer(sg, sg_nents(sg), dst, cryptlen); everything works! <- So, what am I doing wrong that makes scatterwalk_map_and_copy() fail, and sg_copy_from_buffer() work fine? It would be nice if I could fix it, so I did not need to copy to a temporary buffer. Lund -- Jorgen Lundman | <[email protected]> Unix Administrator | +81 (0)3 -5456-2687 ext 1017 (work) Shibuya-ku, Tokyo | +81 (0)90-5578-8500 (cell) Japan | +81 (0)3 -3375-1767 (home) -- To unsubscribe from this list: send the line "unsubscribe linux-crypto" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
crypto: scatterwalk - fix broken scatterlist manipulation From: Jussi Kivilinna <[email protected]> scatterlist_sg_chain() manipulates scatterlist structures directly in wrong way, chaining without marking 'chain' bit 0x01. This can in some cases lead to problems, such as triggering BUG_ON(!sg->length) in scatterwalk_start(). So instead of reinventing wheel, change scatterwalk to use existing functions from scatterlist API. --- include/crypto/scatterwalk.h | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-) diff --git a/include/crypto/scatterwalk.h b/include/crypto/scatterwalk.h index 3744d2a..d31870c 100644 --- a/include/crypto/scatterwalk.h +++ b/include/crypto/scatterwalk.h @@ -34,16 +34,12 @@ static inline void crypto_yield(u32 flags) static inline void scatterwalk_sg_chain(struct scatterlist *sg1, int num, struct scatterlist *sg2) { - sg_set_page(&sg1[num - 1], (void *)sg2, 0, 0); - sg1[num - 1].page_link &= ~0x02; + sg_chain(sg1, num, sg2); } static inline struct scatterlist *scatterwalk_sg_next(struct scatterlist *sg) { - if (sg_is_last(sg)) - return NULL; - - return (++sg)->length ? sg : (void *)sg_page(sg); + return sg_next(sg); } static inline void scatterwalk_crypto_chain(struct scatterlist *head,
