Return-Path: krmurthy@cisco.com Received: from sj-iport-1.cisco.com (sj-iport-1-in.cisco.com [171.71.176.70]) by mcfeely.cisco.com (8.8.6 (PHNE_17190)/CISCO.SERVER.1.2) with ESMTP id GAA20735; Thu, 3 Jun 2004 06:36:07 -0400 (EDT) Received: from sj-core-2.cisco.com (171.71.177.254) by sj-iport-1.cisco.com with ESMTP; 03 Jun 2004 03:36:11 -0700 X-BrightmailFiltered: true Received: from krmurthy1-lnx.cisco.com (krmurthy1-lnx.cisco.com [10.77.7.60]) by sj-core-2.cisco.com (8.12.10/8.12.6) with ESMTP id i53Aa2Xl005288; Thu, 3 Jun 2004 03:36:03 -0700 (PDT) Received: from localhost (localhost [[UNIX: localhost]]) by krmurthy1-lnx.cisco.com (8.11.6/8.11.6) id i53AZoY29918; Thu, 3 Jun 2004 16:05:50 +0530 From: "N.C.Krishna Murthy" Subject: Patch for OOPs seen when using crypto APIs Date: Thu, 3 Jun 2004 16:05:49 +0530 User-Agent: KMail/1.4.3 To: chaapala@cisco.com Cc: Dave Myers , mbhojasi@cisco.com Message-Id: <200406031605.49913.krmurthy@cisco.com> MIME-Version: 1.0 Content-Type: multipart/mixed; boundary="===-=-=" --===-=-= Content-Type: text/plain; charset=iso-8859-1 Hi Clay, I was testing iSCSI driver code from 4.0 branch against 2.6.6 kernel. I had the in kernel crypto CRC32 feature enabled. Initial testing resulted in CRC errors. This was due to bugs in iSCSI driver and it has been reported on sourceforge([965626 ] CRC errors seen when crypto APIs are used). I fixed the bugs in iSCSI driver and ran some file I/O. This resulted in an OOPs. ----------------------------------------------------------------------------- ---------- Process iscsi-tx (pid: 3344, threadinfo=f63e4000 task=f5c9f2a0) Stack: f64d82b0 00000000 c0215af5 ffffffff fffe2000 00002000 fffe2000 c0214c32 f64d82b0 fffe2000 00002000 f64d82b0 f63e4000 f63e4000 00002000 f63e5e94 f7f609a0 f5c74000 f8b0dc80 f64d8284 f63e5e94 00000001 f63e5e78 f63e5e7c Call Trace: [] chksum_update+0x25/0x30 [] update+0x92/0x100 [] iscsi_xmit_data+0x460/0xb60 [iscsi_sfnet] [] scheduler_tick+0x43/0x630 [] schedule+0x36f/0x6a0 [] iscsi_xmit_r2t_data+0xca/0x1b0 [iscsi_sfnet] [] process_tx_requests+0x1f5/0x320 [iscsi_sfnet] [] default_wake_function+0x0/0x20 [] iscsi_tx_thread+0x195/0x1d0 [iscsi_sfnet] [] default_wake_function+0x0/0x20 [] iscsi_tx_thread+0x0/0x1d0 [iscsi_sfnet] [] kernel_thread_helper+0x5/0x10 ----------------------------------------------------------------------------- ----------- Upon looking into the crypto/digest.c, I found that "update()" function does crypto_kmap() only the first page in the sg. A "struct scatterlist " can point to a single page or physically contiguous pages. The oops seen is a result of "update()" not handling contiguous pages. The patch I have attached solves the bug. I was able to test it and did not see any oops after that. File I/O went through fine without any digest errors. Could you please forward the patch to the appropriate forum? Thanx N.C.Krishna Murthy ------------------------------------------------------- --===-=-= Content-Type: text/x-diff; charset=iso-8859-1; name=digest.patch Content-Disposition: attachment; filename=digest.patch --- /usr/src/linux-2.6.6/crypto/digest.c 2004-06-03 14:36:09.000000000 +0530 +++ /usr/src/linux-2.6.6/crypto.new/digest.c 2004-06-03 15:12:10.683261696 +0530 @@ -27,13 +27,28 @@ struct scatterlist *sg, unsigned int nsg) { unsigned int i; - + for (i = 0; i < nsg; i++) { - char *p = crypto_kmap(sg[i].page, 0) + sg[i].offset; - tfm->__crt_alg->cra_digest.dia_update(crypto_tfm_ctx(tfm), - p, sg[i].length); - crypto_kunmap(p, 0); - crypto_yield(tfm); + + struct page *pg = sg[i].page; + unsigned int offset = sg[i].offset; + unsigned int l = sg[i].length; + + do { + unsigned int bytes_from_page = min(l, ((unsigned int) + (PAGE_SIZE)) - + offset); + char *p = crypto_kmap(pg, 0) + offset; + + tfm->__crt_alg->cra_digest.dia_update + (crypto_tfm_ctx(tfm), p, + bytes_from_page); + crypto_kunmap(p, 0); + crypto_yield(tfm); + offset = 0; + pg++; + l -= bytes_from_page; + } while (l > 0); } } --===-=-=--