aboutsummaryrefslogtreecommitdiff
path: root/lib/sg_pt_linux_nvme.c
diff options
context:
space:
mode:
Diffstat (limited to 'lib/sg_pt_linux_nvme.c')
-rw-r--r--lib/sg_pt_linux_nvme.c1935
1 files changed, 1935 insertions, 0 deletions
diff --git a/lib/sg_pt_linux_nvme.c b/lib/sg_pt_linux_nvme.c
new file mode 100644
index 00000000..5a710f9d
--- /dev/null
+++ b/lib/sg_pt_linux_nvme.c
@@ -0,0 +1,1935 @@
+/*
+ * Copyright (c) 2017-2021 Douglas Gilbert.
+ * All rights reserved.
+ * Use of this source code is governed by a BSD-style
+ * license that can be found in the BSD_LICENSE file.
+ *
+ * SPDX-License-Identifier: BSD-2-Clause
+ *
+ * The code to use the NVMe Management Interface (MI) SES pass-through
+ * was provided by WDC in November 2017.
+ */
+
+/*
+ * Copyright 2017, Western Digital Corporation
+ *
+ * Written by Berck Nash
+ *
+ * Use of this source code is governed by a BSD-style
+ * license that can be found in the BSD_LICENSE file.
+ *
+ * Based on the NVM-Express command line utility, which bore the following
+ * notice:
+ *
+ * Copyright (c) 2014-2015, Intel Corporation.
+ *
+ * Written by Keith Busch <keith.busch@intel.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
+ * MA 02110-1301, USA.
+ */
+
+/* sg_pt_linux_nvme version 1.18 20210601 */
+
+/* This file contains a small "SPC-only" SNTL to support the SES pass-through
+ * of SEND DIAGNOSTIC and RECEIVE DIAGNOSTIC RESULTS through NVME-MI
+ * SES Send and SES Receive. */
+
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdarg.h>
+#include <stdbool.h>
+#include <string.h>
+#include <ctype.h>
+#include <unistd.h>
+#include <errno.h>
+#include <fcntl.h>
+#define __STDC_FORMAT_MACROS 1
+#include <inttypes.h>
+#include <sys/ioctl.h>
+#include <sys/stat.h>
+#include <sys/sysmacros.h> /* to define 'major' */
+#ifndef major
+#include <sys/types.h>
+#endif
+
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <linux/major.h>
+
+#include "sg_pt.h"
+#include "sg_lib.h"
+#include "sg_linux_inc.h"
+#include "sg_pt_linux.h"
+#include "sg_unaligned.h"
+#include "sg_pr2serr.h"
+
+#define SCSI_INQUIRY_OPC 0x12
+#define SCSI_REPORT_LUNS_OPC 0xa0
+#define SCSI_TEST_UNIT_READY_OPC 0x0
+#define SCSI_REQUEST_SENSE_OPC 0x3
+#define SCSI_SEND_DIAGNOSTIC_OPC 0x1d
+#define SCSI_RECEIVE_DIAGNOSTIC_OPC 0x1c
+#define SCSI_MAINT_IN_OPC 0xa3
+#define SCSI_READ10_OPC 0x28
+#define SCSI_READ16_OPC 0x88
+#define SCSI_REP_SUP_OPCS_OPC 0xc
+#define SCSI_REP_SUP_TMFS_OPC 0xd
+#define SCSI_MODE_SENSE10_OPC 0x5a
+#define SCSI_MODE_SELECT10_OPC 0x55
+#define SCSI_READ_CAPACITY10_OPC 0x25
+#define SCSI_START_STOP_OPC 0x1b
+#define SCSI_SYNC_CACHE10_OPC 0x35
+#define SCSI_SYNC_CACHE16_OPC 0x91
+#define SCSI_VERIFY10_OPC 0x2f
+#define SCSI_VERIFY16_OPC 0x8f
+#define SCSI_WRITE10_OPC 0x2a
+#define SCSI_WRITE16_OPC 0x8a
+#define SCSI_WRITE_SAME10_OPC 0x41
+#define SCSI_WRITE_SAME16_OPC 0x93
+#define SCSI_SERVICE_ACT_IN_OPC 0x9e
+#define SCSI_READ_CAPACITY16_SA 0x10
+#define SCSI_SA_MSK 0x1f
+
+/* Additional Sense Code (ASC) */
+#define NO_ADDITIONAL_SENSE 0x0
+#define LOGICAL_UNIT_NOT_READY 0x4
+#define LOGICAL_UNIT_COMMUNICATION_FAILURE 0x8
+#define UNRECOVERED_READ_ERR 0x11
+#define PARAMETER_LIST_LENGTH_ERR 0x1a
+#define INVALID_OPCODE 0x20
+#define LBA_OUT_OF_RANGE 0x21
+#define INVALID_FIELD_IN_CDB 0x24
+#define INVALID_FIELD_IN_PARAM_LIST 0x26
+#define UA_RESET_ASC 0x29
+#define UA_CHANGED_ASC 0x2a
+#define TARGET_CHANGED_ASC 0x3f
+#define LUNS_CHANGED_ASCQ 0x0e
+#define INSUFF_RES_ASC 0x55
+#define INSUFF_RES_ASCQ 0x3
+#define LOW_POWER_COND_ON_ASC 0x5e /* ASCQ=0 */
+#define POWER_ON_RESET_ASCQ 0x0
+#define BUS_RESET_ASCQ 0x2 /* scsi bus reset occurred */
+#define MODE_CHANGED_ASCQ 0x1 /* mode parameters changed */
+#define CAPACITY_CHANGED_ASCQ 0x9
+#define SAVING_PARAMS_UNSUP 0x39
+#define TRANSPORT_PROBLEM 0x4b
+#define THRESHOLD_EXCEEDED 0x5d
+#define LOW_POWER_COND_ON 0x5e
+#define MISCOMPARE_VERIFY_ASC 0x1d
+#define MICROCODE_CHANGED_ASCQ 0x1 /* with TARGET_CHANGED_ASC */
+#define MICROCODE_CHANGED_WO_RESET_ASCQ 0x16
+#define PCIE_ERR_ASC 0x4b
+#define PCIE_UNSUPP_REQ_ASCQ 0x13
+
+/* NVMe Admin commands */
+#define SG_NVME_AD_GET_FEATURE 0xa
+#define SG_NVME_AD_SET_FEATURE 0x9
+#define SG_NVME_AD_IDENTIFY 0x6 /* similar to SCSI INQUIRY */
+#define SG_NVME_AD_DEV_SELT_TEST 0x14
+#define SG_NVME_AD_MI_RECEIVE 0x1e /* MI: Management Interface */
+#define SG_NVME_AD_MI_SEND 0x1d /* hmmm, same opcode as SEND DIAG */
+
+/* NVMe NVM (Non-Volatile Memory) commands */
+#define SG_NVME_NVM_FLUSH 0x0 /* SCSI SYNCHRONIZE CACHE */
+#define SG_NVME_NVM_COMPARE 0x5 /* SCSI VERIFY(BYTCHK=1) */
+#define SG_NVME_NVM_READ 0x2
+#define SG_NVME_NVM_VERIFY 0xc /* SCSI VERIFY(BYTCHK=0) */
+#define SG_NVME_NVM_WRITE 0x1
+#define SG_NVME_NVM_WRITE_ZEROES 0x8 /* SCSI WRITE SAME */
+
+#define SG_NVME_RW_CONTROL_FUA (1 << 14) /* Force Unit Access bit */
+
+
+#if (HAVE_NVME && (! IGNORE_NVME))
+
+/* This trims given NVMe block device name in Linux (e.g. /dev/nvme0n1p5)
+ * to the name of its associated char device (e.g. /dev/nvme0). If this
+ * occurs true is returned and the char device name is placed in 'b' (as
+ * long as b_len is sufficient). Otherwise false is returned. */
+bool
+sg_get_nvme_char_devname(const char * nvme_block_devname, uint32_t b_len,
+ char * b)
+{
+ uint32_t n, tlen;
+ const char * cp;
+ char buff[8];
+
+ if ((NULL == b) || (b_len < 5))
+ return false; /* degenerate cases */
+ cp = strstr(nvme_block_devname, "nvme");
+ if (NULL == cp)
+ return false; /* expected to find "nvme" in given name */
+ if (1 != sscanf(cp, "nvme%u", &n))
+ return false; /* didn't find valid "nvme<number>" */
+ snprintf(buff, sizeof(buff), "%u", n);
+ tlen = (cp - nvme_block_devname) + 4 + strlen(buff);
+ if ((tlen + 1) > b_len)
+ return false; /* b isn't long enough to fit output */
+ memcpy(b, nvme_block_devname, tlen);
+ b[tlen] = '\0';
+ return true;
+}
+
+static void
+mk_sense_asc_ascq(struct sg_pt_linux_scsi * ptp, int sk, int asc, int ascq,
+ int vb)
+{
+ bool dsense = !! ptp->dev_stat.scsi_dsense;
+ int n;
+ uint8_t * sbp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.response;
+
+ ptp->io_hdr.device_status = SAM_STAT_CHECK_CONDITION;
+ n = ptp->io_hdr.max_response_len;
+ if ((n < 8) || ((! dsense) && (n < 14))) {
+ if (vb)
+ pr2ws("%s: max_response_len=%d too short, want 14 or more\n",
+ __func__, n);
+ return;
+ } else
+ ptp->io_hdr.response_len = dsense ? n : ((n < 18) ? n : 18);
+ memset(sbp, 0, n);
+ sg_build_sense_buffer(dsense, sbp, sk, asc, ascq);
+ if (vb > 3)
+ pr2ws("%s: [sense_key,asc,ascq]: [0x%x,0x%x,0x%x]\n", __func__, sk,
+ asc, ascq);
+}
+
+static void
+mk_sense_from_nvme_status(struct sg_pt_linux_scsi * ptp, int vb)
+{
+ bool ok;
+ bool dsense = !! ptp->dev_stat.scsi_dsense;
+ int n;
+ uint8_t sstatus, sk, asc, ascq;
+ uint8_t * sbp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.response;
+
+ ok = sg_nvme_status2scsi(ptp->nvme_status, &sstatus, &sk, &asc, &ascq);
+ if (! ok) { /* can't find a mapping to a SCSI error, so ... */
+ sstatus = SAM_STAT_CHECK_CONDITION;
+ sk = SPC_SK_ILLEGAL_REQUEST;
+ asc = 0xb;
+ ascq = 0x0; /* asc: "WARNING" purposely vague */
+ }
+
+ ptp->io_hdr.device_status = sstatus;
+ n = ptp->io_hdr.max_response_len;
+ if ((n < 8) || ((! dsense) && (n < 14))) {
+ pr2ws("%s: sense_len=%d too short, want 14 or more\n", __func__, n);
+ return;
+ } else
+ ptp->io_hdr.response_len = dsense ? n : ((n < 18) ? n : 18);
+ memset(sbp, 0, n);
+ sg_build_sense_buffer(dsense, sbp, sk, asc, ascq);
+ if (dsense && (ptp->nvme_status > 0))
+ sg_nvme_desc2sense(sbp, ptp->nvme_stat_dnr, ptp->nvme_stat_more,
+ ptp->nvme_status);
+ if (vb > 3)
+ pr2ws("%s: [status, sense_key,asc,ascq]: [0x%x, 0x%x,0x%x,0x%x]\n",
+ __func__, sstatus, sk, asc, ascq);
+}
+
+/* Set in_bit to -1 to indicate no bit position of invalid field */
+static void
+mk_sense_invalid_fld(struct sg_pt_linux_scsi * ptp, bool in_cdb, int in_byte,
+ int in_bit, int vb)
+{
+ bool dsense = !! ptp->dev_stat.scsi_dsense;
+ int asc, n;
+ uint8_t * sbp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.response;
+ uint8_t sks[4];
+
+ ptp->io_hdr.device_status = SAM_STAT_CHECK_CONDITION;
+ asc = in_cdb ? INVALID_FIELD_IN_CDB : INVALID_FIELD_IN_PARAM_LIST;
+ n = ptp->io_hdr.max_response_len;
+ if ((n < 8) || ((! dsense) && (n < 14))) {
+ if (vb)
+ pr2ws("%s: max_response_len=%d too short, want 14 or more\n",
+ __func__, n);
+ return;
+ } else
+ ptp->io_hdr.response_len = dsense ? n : ((n < 18) ? n : 18);
+
+ memset(sbp, 0, n);
+ sg_build_sense_buffer(dsense, sbp, SPC_SK_ILLEGAL_REQUEST, asc, 0);
+ memset(sks, 0, sizeof(sks));
+ sks[0] = 0x80;
+ if (in_cdb)
+ sks[0] |= 0x40;
+ if (in_bit >= 0) {
+ sks[0] |= 0x8;
+ sks[0] |= (0x7 & in_bit);
+ }
+ sg_put_unaligned_be16(in_byte, sks + 1);
+ if (dsense) {
+ int sl = sbp[7] + 8;
+
+ sbp[7] = sl;
+ sbp[sl] = 0x2;
+ sbp[sl + 1] = 0x6;
+ memcpy(sbp + sl + 4, sks, 3);
+ } else
+ memcpy(sbp + 15, sks, 3);
+ if (vb > 3)
+ pr2ws("%s: [sense_key,asc,ascq]: [0x5,0x%x,0x0] %c byte=%d, bit=%d\n",
+ __func__, asc, in_cdb ? 'C' : 'D', in_byte,
+ ((in_bit > 0) ? (0x7 & in_bit) : 0));
+}
+
+/* Returns 0 for success. Returns SG_LIB_NVME_STATUS if there is non-zero
+ * NVMe status (from the completion queue) with the value placed in
+ * ptp->nvme_status. If Unix error from ioctl then return negated value
+ * (equivalent -errno from basic Unix system functions like open()).
+ * CDW0 from the completion queue is placed in ptp->nvme_result in the
+ * absence of a Unix error. If time_secs is negative it is treated as
+ * a timeout in milliseconds (of abs(time_secs) ). */
+static int
+sg_nvme_admin_cmd(struct sg_pt_linux_scsi * ptp,
+ struct sg_nvme_passthru_cmd *cmdp, void * dp, bool is_read,
+ int time_secs, int vb)
+{
+ const uint32_t cmd_len = sizeof(struct sg_nvme_passthru_cmd);
+ int res;
+ uint32_t n;
+ uint16_t sct_sc;
+ const uint8_t * up = ((const uint8_t *)cmdp) + SG_NVME_PT_OPCODE;
+ char nam[64];
+
+ if (vb)
+ sg_get_nvme_opcode_name(*up, true /* ADMIN */, sizeof(nam), nam);
+ else
+ nam[0] = '\0';
+ cmdp->timeout_ms = (time_secs < 0) ? (-time_secs) : (1000 * time_secs);
+ ptp->os_err = 0;
+ if (vb > 2) {
+ pr2ws("NVMe Admin command: %s\n", nam);
+ hex2stderr((const uint8_t *)cmdp, cmd_len, 1);
+ if ((vb > 4) && (! is_read) && dp) {
+ uint32_t len = sg_get_unaligned_le32(up + SG_NVME_PT_DATA_LEN);
+
+ if (len > 0) {
+ n = len;
+ if ((len < 512) || (vb > 5))
+ pr2ws("\nData-out buffer (%u bytes):\n", n);
+ else {
+ pr2ws("\nData-out buffer (first 512 of %u bytes):\n", n);
+ n = 512;
+ }
+ hex2stderr((const uint8_t *)dp, n, 0);
+ }
+ }
+ }
+ res = ioctl(ptp->dev_fd, NVME_IOCTL_ADMIN_CMD, cmdp);
+ if (res < 0) { /* OS error (errno negated) */
+ ptp->os_err = -res;
+ if (vb > 1) {
+ pr2ws("%s: ioctl for %s [0x%x] failed: %s "
+ "(errno=%d)\n", __func__, nam, *up, strerror(-res), -res);
+ }
+ return res;
+ }
+
+ /* Now res contains NVMe completion queue CDW3 31:17 (15 bits) */
+ ptp->nvme_result = cmdp->result;
+ if ((! ptp->nvme_our_sntl) && ptp->io_hdr.response &&
+ (ptp->io_hdr.max_response_len > 3)) {
+ /* build 32 byte "sense" buffer */
+ uint8_t * sbp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.response;
+ uint16_t st = (uint16_t)res;
+
+ n = ptp->io_hdr.max_response_len;
+ n = (n < 32) ? n : 32;
+ memset(sbp, 0 , n);
+ ptp->io_hdr.response_len = n;
+ sg_put_unaligned_le32(cmdp->result,
+ sbp + SG_NVME_PT_CQ_RESULT);
+ if (n > 15) /* LSBit will be 0 (Phase bit) after (st << 1) */
+ sg_put_unaligned_le16(st << 1, sbp + SG_NVME_PT_CQ_STATUS_P);
+ }
+ /* clear upper bits (DNR and More) leaving ((SCT << 8) | SC) */
+ sct_sc = 0x7ff & res; /* 11 bits */
+ ptp->nvme_status = sct_sc;
+ ptp->nvme_stat_dnr = !!(0x4000 & res);
+ ptp->nvme_stat_more = !!(0x2000 & res);
+ if (sct_sc) { /* when non-zero, treat as command error */
+ if (vb > 1) {
+ char b[80];
+
+ pr2ws("%s: ioctl for %s [0x%x] failed, status: %s [0x%x]\n",
+ __func__, nam, *up,
+ sg_get_nvme_cmd_status_str(sct_sc, sizeof(b), b), sct_sc);
+ }
+ return SG_LIB_NVME_STATUS; /* == SCSI_PT_DO_NVME_STATUS */
+ }
+ if ((vb > 4) && is_read && dp) {
+ uint32_t len = sg_get_unaligned_le32(up + SG_NVME_PT_DATA_LEN);
+
+ if (len > 0) {
+ n = len;
+ if ((len < 1024) || (vb > 5))
+ pr2ws("\nData-in buffer (%u bytes):\n", n);
+ else {
+ pr2ws("\nData-in buffer (first 1024 of %u bytes):\n", n);
+ n = 1024;
+ }
+ hex2stderr((const uint8_t *)dp, n, 0);
+ }
+ }
+ return 0;
+}
+
+/* see NVME MI document, NVMSR is NVM Subsystem Report */
+static void
+sntl_check_enclosure_override(struct sg_pt_linux_scsi * ptp, int vb)
+{
+ uint8_t * up = ptp->nvme_id_ctlp;
+ uint8_t nvmsr;
+
+ if (NULL == up)
+ return;
+ nvmsr = up[253];
+ if (vb > 5)
+ pr2ws("%s: enter, nvmsr=%u\n", __func__, nvmsr);
+ ptp->dev_stat.id_ctl253 = nvmsr;
+ switch (ptp->dev_stat.enclosure_override) {
+ case 0x0: /* no override */
+ if (0x3 == (0x3 & nvmsr)) {
+ ptp->dev_stat.pdt = PDT_DISK;
+ ptp->dev_stat.enc_serv = 1;
+ } else if (0x2 & nvmsr) {
+ ptp->dev_stat.pdt = PDT_SES;
+ ptp->dev_stat.enc_serv = 1;
+ } else if (0x1 & nvmsr) {
+ ptp->dev_stat.pdt = PDT_DISK;
+ ptp->dev_stat.enc_serv = 0;
+ } else {
+ uint32_t nn = sg_get_unaligned_le32(up + 516);
+
+ ptp->dev_stat.pdt = nn ? PDT_DISK : PDT_UNKNOWN;
+ ptp->dev_stat.enc_serv = 0;
+ }
+ break;
+ case 0x1: /* override to SES device */
+ ptp->dev_stat.pdt = PDT_SES;
+ ptp->dev_stat.enc_serv = 1;
+ break;
+ case 0x2: /* override to disk with attached SES device */
+ ptp->dev_stat.pdt = PDT_DISK;
+ ptp->dev_stat.enc_serv = 1;
+ break;
+ case 0x3: /* override to SAFTE device (PDT_PROCESSOR) */
+ ptp->dev_stat.pdt = PDT_PROCESSOR;
+ ptp->dev_stat.enc_serv = 1;
+ break;
+ case 0xff: /* override to normal disk */
+ ptp->dev_stat.pdt = PDT_DISK;
+ ptp->dev_stat.enc_serv = 0;
+ break;
+ default:
+ pr2ws("%s: unknown enclosure_override value: %d\n", __func__,
+ ptp->dev_stat.enclosure_override);
+ break;
+ }
+}
+
+static int
+sntl_do_identify(struct sg_pt_linux_scsi * ptp, int cns, int nsid,
+ int time_secs, int u_len, uint8_t * up, int vb)
+{
+ struct sg_nvme_passthru_cmd cmd;
+
+ memset(&cmd, 0, sizeof(cmd));
+ cmd.opcode = SG_NVME_AD_IDENTIFY;
+ cmd.nsid = nsid;
+ cmd.cdw10 = cns;
+ cmd.addr = (uint64_t)(sg_uintptr_t)up;
+ cmd.data_len = u_len;
+ return sg_nvme_admin_cmd(ptp, &cmd, up, true, time_secs, vb);
+}
+
+/* Currently only caches associated identify controller response (4096 bytes).
+ * Returns 0 on success; otherwise a positive value is returned */
+static int
+sntl_cache_identify(struct sg_pt_linux_scsi * ptp, int time_secs, int vb)
+{
+ int ret;
+ uint32_t pg_sz = sg_get_page_size();
+ uint8_t * up;
+
+ up = sg_memalign(pg_sz, pg_sz, &ptp->free_nvme_id_ctlp, false);
+ ptp->nvme_id_ctlp = up;
+ if (NULL == up) {
+ pr2ws("%s: sg_memalign() failed to get memory\n", __func__);
+ return sg_convert_errno(ENOMEM);
+ }
+ ret = sntl_do_identify(ptp, 0x1 /* CNS */, 0 /* nsid */, time_secs,
+ pg_sz, up, vb);
+ if (0 == ret)
+ sntl_check_enclosure_override(ptp, vb);
+ return (ret < 0) ? sg_convert_errno(-ret) : ret;
+}
+
+/* If nsid==0 then set cmdp->nsid to SG_NVME_BROADCAST_NSID. */
+static int
+sntl_get_features(struct sg_pt_linux_scsi * ptp, int feature_id, int select,
+ uint32_t nsid, uint64_t din_addr, int time_secs, int vb)
+{
+ int res;
+ struct sg_nvme_passthru_cmd cmd;
+ struct sg_nvme_passthru_cmd * cmdp = &cmd;
+
+ if (vb > 4)
+ pr2ws("%s: feature_id=0x%x, select=%d\n", __func__, feature_id,
+ select);
+ memset(cmdp, 0, sizeof(*cmdp));
+ cmdp->opcode = SG_NVME_AD_GET_FEATURE;
+ cmdp->nsid = nsid ? nsid : SG_NVME_BROADCAST_NSID;
+ select &= 0x7;
+ feature_id &= 0xff;
+ cmdp->cdw10 = (select << 8) | feature_id;
+ if (din_addr)
+ cmdp->addr = din_addr;
+ cmdp->timeout_ms = (time_secs < 0) ? 0 : (1000 * time_secs);
+ res = sg_nvme_admin_cmd(ptp, cmdp, NULL, false, time_secs, vb);
+ if (res)
+ return res;
+ ptp->os_err = 0;
+ ptp->nvme_status = 0;
+ return 0;
+}
+
+static const char * nvme_scsi_vendor_str = "NVMe ";
+static const uint16_t inq_resp_len = 36;
+
+static int
+sntl_inq(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp, int time_secs,
+ int vb)
+{
+ bool evpd;
+ int res;
+ uint16_t n, alloc_len, pg_cd;
+ uint32_t pg_sz = sg_get_page_size();
+ uint8_t * nvme_id_ns = NULL;
+ uint8_t * free_nvme_id_ns = NULL;
+ uint8_t inq_dout[256];
+
+ if (vb > 5)
+ pr2ws("%s: time_secs=%d\n", __func__, time_secs);
+
+ if (0x2 & cdbp[1]) { /* Reject CmdDt=1 */
+ mk_sense_invalid_fld(ptp, true, 1, 1, vb);
+ return 0;
+ }
+ if (NULL == ptp->nvme_id_ctlp) {
+ res = sntl_cache_identify(ptp, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else if (res) /* should be negative errno */
+ return res;
+ }
+ memset(inq_dout, 0, sizeof(inq_dout));
+ alloc_len = sg_get_unaligned_be16(cdbp + 3);
+ evpd = !!(0x1 & cdbp[1]);
+ pg_cd = cdbp[2];
+ if (evpd) { /* VPD page responses */
+ bool cp_id_ctl = false;
+
+ switch (pg_cd) {
+ case 0:
+ /* inq_dout[0] = (PQ=0)<<5 | (PDT=0); prefer pdt=0xd --> SES */
+ inq_dout[1] = pg_cd;
+ n = 11;
+ sg_put_unaligned_be16(n - 4, inq_dout + 2);
+ inq_dout[4] = 0x0;
+ inq_dout[5] = 0x80;
+ inq_dout[6] = 0x83;
+ inq_dout[7] = 0x86;
+ inq_dout[8] = 0x87;
+ inq_dout[9] = 0x92;
+ inq_dout[n - 1] = SG_NVME_VPD_NICR; /* last VPD number */
+ break;
+ case 0x80:
+ /* inq_dout[0] = (PQ=0)<<5 | (PDT=0); prefer pdt=0xd --> SES */
+ inq_dout[1] = pg_cd;
+ n = 24;
+ sg_put_unaligned_be16(n - 4, inq_dout + 2);
+ memcpy(inq_dout + 4, ptp->nvme_id_ctlp + 4, 20); /* SN */
+ break;
+ case 0x83:
+ if ((ptp->nvme_nsid > 0) &&
+ (ptp->nvme_nsid < SG_NVME_BROADCAST_NSID)) {
+ nvme_id_ns = sg_memalign(pg_sz, pg_sz, &free_nvme_id_ns,
+ false);
+ if (nvme_id_ns) {
+ /* CNS=0x0 Identify namespace */
+ res = sntl_do_identify(ptp, 0x0, ptp->nvme_nsid,
+ time_secs, pg_sz, nvme_id_ns, vb);
+ if (res) {
+ free(free_nvme_id_ns);
+ free_nvme_id_ns = NULL;
+ nvme_id_ns = NULL;
+ }
+ }
+ }
+ n = sg_make_vpd_devid_for_nvme(ptp->nvme_id_ctlp, nvme_id_ns,
+ 0 /* pdt */, -1 /*tproto */,
+ inq_dout, sizeof(inq_dout));
+ if (n > 3)
+ sg_put_unaligned_be16(n - 4, inq_dout + 2);
+ if (free_nvme_id_ns) {
+ free(free_nvme_id_ns);
+ free_nvme_id_ns = NULL;
+ nvme_id_ns = NULL;
+ }
+ break;
+ case 0x86: /* Extended INQUIRY (per SFS SPC Discovery 2016) */
+ inq_dout[1] = pg_cd;
+ n = 64;
+ sg_put_unaligned_be16(n - 4, inq_dout + 2);
+ inq_dout[5] = 0x1; /* SIMPSUP=1 */
+ inq_dout[7] = 0x1; /* LUICLR=1 */
+ inq_dout[13] = 0x40; /* max supported sense data length */
+ break;
+ case 0x87: /* Mode page policy (per SFS SPC Discovery 2016) */
+ inq_dout[1] = pg_cd;
+ n = 8;
+ sg_put_unaligned_be16(n - 4, inq_dout + 2);
+ inq_dout[4] = 0x3f; /* all mode pages */
+ inq_dout[5] = 0xff; /* and their sub-pages */
+ inq_dout[6] = 0x80; /* MLUS=1, policy=shared */
+ break;
+ case 0x92: /* SCSI Feature set: only SPC Discovery 2016 */
+ inq_dout[1] = pg_cd;
+ n = 10;
+ sg_put_unaligned_be16(n - 4, inq_dout + 2);
+ inq_dout[9] = 0x1; /* SFS SPC Discovery 2016 */
+ break;
+ case SG_NVME_VPD_NICR: /* 0xde (vendor (sg3_utils) specific) */
+ inq_dout[1] = pg_cd;
+ sg_put_unaligned_be16((16 + 4096) - 4, inq_dout + 2);
+ n = 16 + 4096;
+ cp_id_ctl = true;
+ break;
+ default: /* Point to page_code field in cdb */
+ mk_sense_invalid_fld(ptp, true, 2, 7, vb);
+ return 0;
+ }
+ if (alloc_len > 0) {
+ n = (alloc_len < n) ? alloc_len : n;
+ n = (n < ptp->io_hdr.din_xfer_len) ? n : ptp->io_hdr.din_xfer_len;
+ ptp->io_hdr.din_resid = ptp->io_hdr.din_xfer_len - n;
+ if (n > 0) {
+ uint8_t * dp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.din_xferp;
+
+ if (cp_id_ctl) {
+ memcpy(dp, inq_dout, (n < 16 ? n : 16));
+ if (n > 16)
+ memcpy(dp + 16, ptp->nvme_id_ctlp, n - 16);
+ } else
+ memcpy(dp, inq_dout, n);
+ }
+ }
+ } else { /* Standard INQUIRY response */
+ /* pdt=0 --> disk; pdt=0xd --> SES; pdt=3 --> processor (safte) */
+ inq_dout[0] = (0x1f & ptp->dev_stat.pdt); /* (PQ=0)<<5 */
+ /* inq_dout[1] = (RMD=0)<<7 | (LU_CONG=0)<<6 | (HOT_PLUG=0)<<4; */
+ inq_dout[2] = 6; /* version: SPC-4 */
+ inq_dout[3] = 2; /* NORMACA=0, HISUP=0, response data format: 2 */
+ inq_dout[4] = 31; /* so response length is (or could be) 36 bytes */
+ inq_dout[6] = ptp->dev_stat.enc_serv ? 0x40 : 0;
+ inq_dout[7] = 0x2; /* CMDQUE=1 */
+ memcpy(inq_dout + 8, nvme_scsi_vendor_str, 8); /* NVMe not Intel */
+ memcpy(inq_dout + 16, ptp->nvme_id_ctlp + 24, 16); /* Prod <-- MN */
+ memcpy(inq_dout + 32, ptp->nvme_id_ctlp + 64, 4); /* Rev <-- FR */
+ if (alloc_len > 0) {
+ n = (alloc_len < inq_resp_len) ? alloc_len : inq_resp_len;
+ n = (n < ptp->io_hdr.din_xfer_len) ? n : ptp->io_hdr.din_xfer_len;
+ ptp->io_hdr.din_resid = ptp->io_hdr.din_xfer_len - n;
+ if (n > 0)
+ memcpy((uint8_t *)(sg_uintptr_t)ptp->io_hdr.din_xferp,
+ inq_dout, n);
+ }
+ }
+ return 0;
+}
+
+static int
+sntl_rluns(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp, int time_secs,
+ int vb)
+{
+ int res;
+ uint16_t sel_report;
+ uint32_t alloc_len, k, n, num, max_nsid;
+ uint8_t * rl_doutp;
+ uint8_t * up;
+
+ if (vb > 5)
+ pr2ws("%s: time_secs=%d\n", __func__, time_secs);
+
+ sel_report = cdbp[2];
+ alloc_len = sg_get_unaligned_be32(cdbp + 6);
+ if (NULL == ptp->nvme_id_ctlp) {
+ res = sntl_cache_identify(ptp, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else if (res)
+ return res;
+ }
+ max_nsid = sg_get_unaligned_le32(ptp->nvme_id_ctlp + 516);
+ switch (sel_report) {
+ case 0:
+ case 2:
+ num = max_nsid;
+ break;
+ case 1:
+ case 0x10:
+ case 0x12:
+ num = 0;
+ break;
+ case 0x11:
+ num = (1 == ptp->nvme_nsid) ? max_nsid : 0;
+ break;
+ default:
+ if (vb > 1)
+ pr2ws("%s: bad select_report value: 0x%x\n", __func__,
+ sel_report);
+ mk_sense_invalid_fld(ptp, true, 2, 7, vb);
+ return 0;
+ }
+ rl_doutp = (uint8_t *)calloc(num + 1, 8);
+ if (NULL == rl_doutp) {
+ pr2ws("%s: calloc() failed to get memory\n", __func__);
+ return sg_convert_errno(ENOMEM);
+ }
+ for (k = 0, up = rl_doutp + 8; k < num; ++k, up += 8)
+ sg_put_unaligned_be16(k, up);
+ n = num * 8;
+ sg_put_unaligned_be32(n, rl_doutp);
+ n+= 8;
+ if (alloc_len > 0) {
+ n = (alloc_len < n) ? alloc_len : n;
+ n = (n < ptp->io_hdr.din_xfer_len) ? n : ptp->io_hdr.din_xfer_len;
+ ptp->io_hdr.din_resid = ptp->io_hdr.din_xfer_len - n;
+ if (n > 0)
+ memcpy((uint8_t *)(sg_uintptr_t)ptp->io_hdr.din_xferp, rl_doutp,
+ n);
+ }
+ res = 0;
+ free(rl_doutp);
+ return res;
+}
+
+static int
+sntl_tur(struct sg_pt_linux_scsi * ptp, int time_secs, int vb)
+{
+ int res;
+ uint32_t pow_state;
+
+ if (vb > 5)
+ pr2ws("%s: start\n", __func__);
+ if (NULL == ptp->nvme_id_ctlp) {
+ res = sntl_cache_identify(ptp, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else if (res)
+ return res;
+ }
+ res = sntl_get_features(ptp, 2 /* Power Management */, 0 /* current */,
+ 0, 0, time_secs, vb);
+ if (0 != res) {
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else
+ return res;
+ }
+ pow_state = (0x1f & ptp->nvme_result);
+ if (vb > 5)
+ pr2ws("%s: pow_state=%u\n", __func__, pow_state);
+#if 0 /* pow_state bounces around too much on laptop */
+ if (pow_state)
+ mk_sense_asc_ascq(ptp, SPC_SK_NOT_READY, LOW_POWER_COND_ON_ASC, 0,
+ vb);
+#endif
+ return 0;
+}
+
+static int
+sntl_req_sense(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool desc;
+ int res;
+ uint32_t pow_state, alloc_len, n;
+ uint8_t rs_dout[64];
+
+ if (vb > 5)
+ pr2ws("%s: time_secs=%d\n", __func__, time_secs);
+ if (NULL == ptp->nvme_id_ctlp) {
+ res = sntl_cache_identify(ptp, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else if (res)
+ return res;
+ }
+ desc = !!(0x1 & cdbp[1]);
+ alloc_len = cdbp[4];
+ res = sntl_get_features(ptp, 0x2 /* Power Management */, 0 /* current */,
+ 0, 0, time_secs, vb);
+ if (0 != res) {
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else
+ return res;
+ }
+ ptp->io_hdr.response_len = 0;
+ pow_state = (0x1f & ptp->nvme_result);
+ if (vb > 5)
+ pr2ws("%s: pow_state=%u\n", __func__, pow_state);
+ memset(rs_dout, 0, sizeof(rs_dout));
+ if (pow_state)
+ sg_build_sense_buffer(desc, rs_dout, SPC_SK_NO_SENSE,
+ LOW_POWER_COND_ON_ASC, 0);
+ else
+ sg_build_sense_buffer(desc, rs_dout, SPC_SK_NO_SENSE,
+ NO_ADDITIONAL_SENSE, 0);
+ n = desc ? 8 : 18;
+ n = (n < alloc_len) ? n : alloc_len;
+ n = (n < ptp->io_hdr.din_xfer_len) ? n : ptp->io_hdr.din_xfer_len;
+ ptp->io_hdr.din_resid = ptp->io_hdr.din_xfer_len - n;
+ if (n > 0)
+ memcpy((uint8_t *)(sg_uintptr_t)ptp->io_hdr.din_xferp, rs_dout, n);
+ return 0;
+}
+
+static uint8_t pc_t10_2_select[] = {0, 3, 1, 2};
+
+/* For MODE SENSE(10) and MODE SELECT(10). 6 byte variants not supported */
+static int
+sntl_mode_ss(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool is_msense = (SCSI_MODE_SENSE10_OPC == cdbp[0]);
+ int res, n, len;
+ uint8_t * bp;
+ struct sg_sntl_result_t sntl_result;
+
+ if (vb > 5)
+ pr2ws("%s: mode se%s\n", __func__, (is_msense ? "nse" : "lect"));
+ if (NULL == ptp->nvme_id_ctlp) {
+ res = sntl_cache_identify(ptp, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else if (res)
+ return res;
+ }
+ if (is_msense) { /* MODE SENSE(10) */
+ uint8_t pc_t10 = (cdbp[2] >> 6) & 0x3;
+ int mp_t10 = (cdbp[2] & 0x3f);
+
+ if ((0x3f == mp_t10) || (0x8 /* caching mpage */ == mp_t10)) {
+ /* 0x6 is "Volatile write cache" feature id */
+ res = sntl_get_features(ptp, 0x6, pc_t10_2_select[pc_t10], 0,
+ 0, time_secs, vb);
+ if (0 != res) {
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else
+ return res;
+ }
+ ptp->dev_stat.wce = !!(0x1 & ptp->nvme_result);
+ }
+ len = ptp->io_hdr.din_xfer_len;
+ bp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.din_xferp;
+ n = sntl_resp_mode_sense10(&ptp->dev_stat, cdbp, bp, len,
+ &sntl_result);
+ ptp->io_hdr.din_resid = (n >= 0) ? len - n : len;
+ } else { /* MODE SELECT(10) */
+ bool sp = !!(0x1 & cdbp[1]); /* Save Page indication */
+ uint8_t pre_enc_ov = ptp->dev_stat.enclosure_override;
+
+ len = ptp->io_hdr.dout_xfer_len;
+ bp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.dout_xferp;
+ ptp->dev_stat.wce_changed = false;
+ n = sntl_resp_mode_select10(&ptp->dev_stat, cdbp, bp, len,
+ &sntl_result);
+ if (ptp->dev_stat.wce_changed) {
+ uint32_t nsid = ptp->nvme_nsid;
+ struct sg_nvme_passthru_cmd cmd;
+ struct sg_nvme_passthru_cmd * cmdp = &cmd;
+
+ ptp->dev_stat.wce_changed = false;
+ memset(cmdp, 0, sizeof(*cmdp));
+ cmdp->opcode = SG_NVME_AD_SET_FEATURE;
+ cmdp->nsid = nsid ? nsid : SG_NVME_BROADCAST_NSID;
+ cmdp->cdw10 = 0x6; /* "Volatile write cache" feature id */
+ if (sp)
+ cmdp->cdw10 |= (1U << 31);
+ cmdp->cdw11 = (uint32_t)ptp->dev_stat.wce;
+ cmdp->timeout_ms = (time_secs < 0) ? 0 : (1000 * time_secs);
+ res = sg_nvme_admin_cmd(ptp, cmdp, NULL, false, time_secs, vb);
+ if (0 != res) {
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else
+ return res;
+ }
+ ptp->os_err = 0;
+ ptp->nvme_status = 0;
+ }
+ if (pre_enc_ov != ptp->dev_stat.enclosure_override)
+ sntl_check_enclosure_override(ptp, vb); /* ENC_OV has changed */
+ }
+ if (n < 0) {
+ int in_bit = (255 == sntl_result.in_bit) ? (int)sntl_result.in_bit :
+ -1;
+ if ((SAM_STAT_CHECK_CONDITION == sntl_result.sstatus) &&
+ (SPC_SK_ILLEGAL_REQUEST == sntl_result.sk)) {
+ if (INVALID_FIELD_IN_CDB == sntl_result.asc)
+ mk_sense_invalid_fld(ptp, true, sntl_result.in_byte, in_bit,
+ vb);
+ else if (INVALID_FIELD_IN_PARAM_LIST == sntl_result.asc)
+ mk_sense_invalid_fld(ptp, false, sntl_result.in_byte, in_bit,
+ vb);
+ else
+ mk_sense_asc_ascq(ptp, sntl_result.sk, sntl_result.asc,
+ sntl_result.ascq, vb);
+ } else
+ pr2ws("%s: error but no sense?? n=%d\n", __func__, n);
+ }
+ return 0;
+}
+
+/* This is not really a SNTL. For SCSI SEND DIAGNOSTIC(PF=1) NVMe-MI
+ * has a special command (SES Send) to tunnel through pages to an
+ * enclosure. The NVMe enclosure is meant to understand the SES
+ * (SCSI Enclosure Services) use of diagnostics pages that are
+ * related to SES. */
+static int
+sntl_senddiag(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool pf, self_test;
+ int res;
+ uint8_t st_cd, dpg_cd;
+ uint32_t alloc_len, n, dout_len, dpg_len;
+ const uint32_t pg_sz = sg_get_page_size();
+ uint8_t * dop;
+ struct sg_nvme_passthru_cmd cmd;
+ uint8_t * cmd_up = (uint8_t *)&cmd;
+
+ st_cd = 0x7 & (cdbp[1] >> 5);
+ self_test = !! (0x4 & cdbp[1]);
+ pf = !! (0x10 & cdbp[1]);
+ if (vb > 5)
+ pr2ws("%s: pf=%d, self_test=%d (st_code=%d)\n", __func__, (int)pf,
+ (int)self_test, (int)st_cd);
+ if (self_test || st_cd) {
+ uint32_t nvme_dst;
+
+ memset(cmd_up, 0, sizeof(cmd));
+ cmd_up[SG_NVME_PT_OPCODE] = SG_NVME_AD_DEV_SELT_TEST;
+ /* just this namespace (if there is one) and controller */
+ sg_put_unaligned_le32(ptp->nvme_nsid, cmd_up + SG_NVME_PT_NSID);
+ switch (st_cd) {
+ case 0: /* Here if self_test is set, do short self-test */
+ case 1: /* Background short */
+ case 5: /* Foreground short */
+ nvme_dst = 1;
+ break;
+ case 2: /* Background extended */
+ case 6: /* Foreground extended */
+ nvme_dst = 2;
+ break;
+ case 4: /* Abort self-test */
+ nvme_dst = 0xf;
+ break;
+ default:
+ pr2ws("%s: bad self-test code [0x%x]\n", __func__, st_cd);
+ mk_sense_invalid_fld(ptp, true, 1, 7, vb);
+ return 0;
+ }
+ sg_put_unaligned_le32(nvme_dst, cmd_up + SG_NVME_PT_CDW10);
+ res = sg_nvme_admin_cmd(ptp, &cmd, NULL, false, time_secs, vb);
+ if (0 != res) {
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else
+ return res;
+ }
+ }
+ alloc_len = sg_get_unaligned_be16(cdbp + 3); /* parameter list length */
+ dout_len = ptp->io_hdr.dout_xfer_len;
+ if (pf) {
+ if (0 == alloc_len) {
+ mk_sense_invalid_fld(ptp, true, 3, 7, vb);
+ if (vb)
+ pr2ws("%s: PF bit set bit param_list_len=0\n", __func__);
+ return 0;
+ }
+ } else { /* PF bit clear */
+ if (alloc_len) {
+ mk_sense_invalid_fld(ptp, true, 3, 7, vb);
+ if (vb)
+ pr2ws("%s: param_list_len>0 but PF clear\n", __func__);
+ return 0;
+ } else
+ return 0; /* nothing to do */
+ }
+ if (dout_len < 4) {
+ if (vb)
+ pr2ws("%s: dout length (%u bytes) too short\n", __func__,
+ dout_len);
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+ n = dout_len;
+ n = (n < alloc_len) ? n : alloc_len;
+ dop = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.dout_xferp;
+ if (! sg_is_aligned(dop, pg_sz)) { /* is dop page aligned ? */
+ if (vb)
+ pr2ws("%s: dout [0x%" PRIx64 "] not page aligned\n", __func__,
+ (uint64_t)ptp->io_hdr.dout_xferp);
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+ dpg_cd = dop[0];
+ dpg_len = sg_get_unaligned_be16(dop + 2) + 4;
+ /* should we allow for more than one D_PG is dout ?? */
+ n = (n < dpg_len) ? n : dpg_len; /* not yet ... */
+
+ if (vb)
+ pr2ws("%s: passing through d_pg=0x%x, len=%u to NVME_MI SES send\n",
+ __func__, dpg_cd, dpg_len);
+ memset(&cmd, 0, sizeof(cmd));
+ cmd.opcode = SG_NVME_AD_MI_SEND;
+ cmd.addr = (uint64_t)(sg_uintptr_t)dop;
+ cmd.data_len = 0x1000; /* NVMe 4k page size. Maybe determine this? */
+ /* dout_len > 0x1000, is this a problem?? */
+ cmd.cdw10 = 0x0804; /* NVMe Message Header */
+ cmd.cdw11 = 0x9; /* nvme_mi_ses_send; (0x8 -> mi_ses_recv) */
+ cmd.cdw13 = n;
+ res = sg_nvme_admin_cmd(ptp, &cmd, dop, false, time_secs, vb);
+ if (0 != res) {
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ }
+ }
+ return res;
+}
+
+/* This is not really a SNTL. For SCSI RECEIVE DIAGNOSTIC RESULTS(PCV=1)
+ * NVMe-MI has a special command (SES Receive) to read pages through a
+ * tunnel from an enclosure. The NVMe enclosure is meant to understand the
+ * SES (SCSI Enclosure Services) use of diagnostics pages that are
+ * related to SES. */
+static int
+sntl_recvdiag(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool pcv;
+ int res;
+ uint8_t dpg_cd;
+ uint32_t alloc_len, n, din_len;
+ uint32_t pg_sz = sg_get_page_size();
+ uint8_t * dip;
+ struct sg_nvme_passthru_cmd cmd;
+
+ pcv = !! (0x1 & cdbp[1]);
+ dpg_cd = cdbp[2];
+ alloc_len = sg_get_unaligned_be16(cdbp + 3); /* parameter list length */
+ if (vb > 5)
+ pr2ws("%s: dpg_cd=0x%x, pcv=%d, alloc_len=0x%x\n", __func__,
+ dpg_cd, (int)pcv, alloc_len);
+ din_len = ptp->io_hdr.din_xfer_len;
+ n = din_len;
+ n = (n < alloc_len) ? n : alloc_len;
+ dip = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.din_xferp;
+ if (! sg_is_aligned(dip, pg_sz)) {
+ if (vb)
+ pr2ws("%s: din [0x%" PRIx64 "] not page aligned\n", __func__,
+ (uint64_t)ptp->io_hdr.din_xferp);
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+
+ if (vb)
+ pr2ws("%s: expecting d_pg=0x%x from NVME_MI SES receive\n", __func__,
+ dpg_cd);
+ memset(&cmd, 0, sizeof(cmd));
+ cmd.opcode = SG_NVME_AD_MI_RECEIVE;
+ cmd.addr = (uint64_t)(sg_uintptr_t)dip;
+ cmd.data_len = 0x1000; /* NVMe 4k page size. Maybe determine this? */
+ /* din_len > 0x1000, is this a problem?? */
+ cmd.cdw10 = 0x0804; /* NVMe Message Header */
+ cmd.cdw11 = 0x8; /* nvme_mi_ses_receive */
+ cmd.cdw12 = dpg_cd;
+ cmd.cdw13 = n;
+ res = sg_nvme_admin_cmd(ptp, &cmd, dip, true, time_secs, vb);
+ if (0 != res) {
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else
+ return res;
+ }
+ ptp->io_hdr.din_resid = din_len - n;
+ return res;
+}
+
+#define F_SA_LOW 0x80 /* cdb byte 1, bits 4 to 0 */
+#define F_SA_HIGH 0x100 /* as used by variable length cdbs */
+#define FF_SA (F_SA_HIGH | F_SA_LOW)
+#define F_INV_OP 0x200
+
+static int
+sntl_rep_opcodes(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool rctd;
+ uint8_t reporting_opts, req_opcode, supp;
+ uint16_t req_sa;
+ uint32_t alloc_len, offset, a_len;
+ uint32_t pg_sz = sg_get_page_size();
+ int len, count, bump;
+ const struct sg_opcode_info_t *oip;
+ uint8_t *arr;
+ uint8_t *free_arr;
+
+ if (vb > 5)
+ pr2ws("%s: time_secs=%d\n", __func__, time_secs);
+ rctd = !!(cdbp[2] & 0x80); /* report command timeout desc. */
+ reporting_opts = cdbp[2] & 0x7;
+ req_opcode = cdbp[3];
+ req_sa = sg_get_unaligned_be16(cdbp + 4);
+ alloc_len = sg_get_unaligned_be32(cdbp + 6);
+ if (alloc_len < 4 || alloc_len > 0xffff) {
+ mk_sense_invalid_fld(ptp, true, 6, -1, vb);
+ return 0;
+ }
+ a_len = pg_sz - 72;
+ arr = sg_memalign(pg_sz, pg_sz, &free_arr, false);
+ if (NULL == arr) {
+ pr2ws("%s: calloc() failed to get memory\n", __func__);
+ return sg_convert_errno(ENOMEM);
+ }
+ switch (reporting_opts) {
+ case 0: /* all commands */
+ count = 0;
+ bump = rctd ? 20 : 8;
+ for (offset = 4, oip = sg_get_opcode_translation();
+ (oip->flags != 0xffff) && (offset < a_len); ++oip) {
+ if (F_INV_OP & oip->flags)
+ continue;
+ ++count;
+ arr[offset] = oip->opcode;
+ sg_put_unaligned_be16(oip->sa, arr + offset + 2);
+ if (rctd)
+ arr[offset + 5] |= 0x2;
+ if (FF_SA & oip->flags)
+ arr[offset + 5] |= 0x1;
+ sg_put_unaligned_be16(oip->len_mask[0], arr + offset + 6);
+ if (rctd)
+ sg_put_unaligned_be16(0xa, arr + offset + 8);
+ offset += bump;
+ }
+ sg_put_unaligned_be32(count * bump, arr + 0);
+ break;
+ case 1: /* one command: opcode only */
+ case 2: /* one command: opcode plus service action */
+ case 3: /* one command: if sa==0 then opcode only else opcode+sa */
+ for (oip = sg_get_opcode_translation(); oip->flags != 0xffff; ++oip) {
+ if ((req_opcode == oip->opcode) && (req_sa == oip->sa))
+ break;
+ }
+ if ((0xffff == oip->flags) || (F_INV_OP & oip->flags)) {
+ supp = 1;
+ offset = 4;
+ } else {
+ if (1 == reporting_opts) {
+ if (FF_SA & oip->flags) {
+ mk_sense_invalid_fld(ptp, true, 2, 2, vb);
+ free(free_arr);
+ return 0;
+ }
+ req_sa = 0;
+ } else if ((2 == reporting_opts) && 0 == (FF_SA & oip->flags)) {
+ mk_sense_invalid_fld(ptp, true, 4, -1, vb);
+ free(free_arr);
+ return 0;
+ }
+ if ((0 == (FF_SA & oip->flags)) && (req_opcode == oip->opcode))
+ supp = 3;
+ else if (0 == (FF_SA & oip->flags))
+ supp = 1;
+ else if (req_sa != oip->sa)
+ supp = 1;
+ else
+ supp = 3;
+ if (3 == supp) {
+ uint16_t u;
+ int k;
+
+ u = oip->len_mask[0];
+ sg_put_unaligned_be16(u, arr + 2);
+ arr[4] = oip->opcode;
+ for (k = 1; k < u; ++k)
+ arr[4 + k] = (k < 16) ?
+ oip->len_mask[k] : 0xff;
+ offset = 4 + u;
+ } else
+ offset = 4;
+ }
+ arr[1] = (rctd ? 0x80 : 0) | supp;
+ if (rctd) {
+ sg_put_unaligned_be16(0xa, arr + offset);
+ offset += 12;
+ }
+ break;
+ default:
+ mk_sense_invalid_fld(ptp, true, 2, 2, vb);
+ free(free_arr);
+ return 0;
+ }
+ offset = (offset < a_len) ? offset : a_len;
+ len = (offset < alloc_len) ? offset : alloc_len;
+ ptp->io_hdr.din_resid = ptp->io_hdr.din_xfer_len - len;
+ if (len > 0)
+ memcpy((uint8_t *)(sg_uintptr_t)ptp->io_hdr.din_xferp, arr, len);
+ free(free_arr);
+ return 0;
+}
+
+static int
+sntl_rep_tmfs(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool repd;
+ uint32_t alloc_len, len;
+ uint8_t arr[16];
+
+ if (vb > 5)
+ pr2ws("%s: time_secs=%d\n", __func__, time_secs);
+ memset(arr, 0, sizeof(arr));
+ repd = !!(cdbp[2] & 0x80);
+ alloc_len = sg_get_unaligned_be32(cdbp + 6);
+ if (alloc_len < 4) {
+ mk_sense_invalid_fld(ptp, true, 6, -1, vb);
+ return 0;
+ }
+ arr[0] = 0xc8; /* ATS | ATSS | LURS */
+ arr[1] = 0x1; /* ITNRS */
+ if (repd) {
+ arr[3] = 0xc;
+ len = 16;
+ } else
+ len = 4;
+
+ len = (len < alloc_len) ? len : alloc_len;
+ ptp->io_hdr.din_resid = ptp->io_hdr.din_xfer_len - len;
+ if (len > 0)
+ memcpy((uint8_t *)(sg_uintptr_t)ptp->io_hdr.din_xferp, arr, len);
+ return 0;
+}
+
+/* Note that the "Returned logical block address" (RLBA) field in the SCSI
+ * READ CAPACITY (10+16) command's response provides the address of the _last_
+ * LBA (counting origin 0) which will be one less that the "size" in the
+ * NVMe Identify command response's NSZE field. One problem is that in
+ * some situations NSZE can be zero: temporarily set RLBA field to 0
+ * (implying a 1 LB logical units size) pending further research. The LBLIB
+ * is the "Logical Block Length In Bytes" field in the RCAP response. */
+static int
+sntl_readcap(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool is_rcap10 = (SCSI_READ_CAPACITY10_OPC == cdbp[0]);
+ int res, n, len, alloc_len, dps;
+ uint8_t flbas, index, lbads; /* NVMe: 2**LBADS --> Logical Block size */
+ uint32_t lbafx; /* NVME: LBAF0...LBAF15, each 16 bytes */
+ uint32_t pg_sz = sg_get_page_size();
+ uint64_t nsze;
+ uint8_t * bp;
+ uint8_t * up;
+ uint8_t * free_up = NULL;
+ uint8_t resp[32];
+
+ if (vb > 5)
+ pr2ws("%s: RCAP%d, time_secs=%d\n", __func__,
+ (is_rcap10 ? 10 : 16), time_secs);
+ up = sg_memalign(pg_sz, pg_sz, &free_up, false);
+ if (NULL == up) {
+ pr2ws("%s: sg_memalign() failed to get memory\n", __func__);
+ return sg_convert_errno(ENOMEM);
+ }
+ res = sntl_do_identify(ptp, 0x0 /* CNS */, ptp->nvme_nsid, time_secs,
+ pg_sz, up, vb);
+ if (res < 0) {
+ res = sg_convert_errno(-res);
+ goto fini;
+ }
+ memset(resp, 0, sizeof(resp));
+ nsze = sg_get_unaligned_le64(up + 0);
+ flbas = up[26]; /* NVME FLBAS field from Identify, want LBAF[flbas] */
+ index = 128 + (4 * (flbas & 0xf));
+ lbafx = sg_get_unaligned_le32(up + index);
+ lbads = (lbafx >> 16) & 0xff; /* bits 16 to 23 inclusive, pow2 */
+ if (is_rcap10) {
+ alloc_len = 8; /* implicit, not in cdb */
+ if (nsze > 0xffffffff)
+ sg_put_unaligned_be32(0xffffffff, resp + 0);
+ else if (0 == nsze) /* no good answer here */
+ sg_put_unaligned_be32(0, resp + 0); /* SCSI RLBA field */
+ else
+ sg_put_unaligned_be32((uint32_t)(nsze - 1), resp + 0);
+ sg_put_unaligned_be32(1 << lbads, resp + 4); /* SCSI LBLIB field */
+ } else {
+ alloc_len = sg_get_unaligned_be32(cdbp + 10);
+ dps = up[29];
+ if (0x7 & dps) {
+ resp[12] = 0x1;
+ n = (0x7 & dps) - 1;
+ if (n > 0)
+ resp[12] |= (n + n);
+ }
+ if (0 == nsze) /* no good answer here */
+ sg_put_unaligned_be64(0, resp + 0);
+ else
+ sg_put_unaligned_be64(nsze - 1, resp + 0);
+ sg_put_unaligned_be32(1 << lbads, resp + 8); /* SCSI LBLIB field */
+ }
+ len = ptp->io_hdr.din_xfer_len;
+ bp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.din_xferp;
+ n = 32;
+ n = (n < alloc_len) ? n : alloc_len;
+ n = (n < len) ? n : len;
+ ptp->io_hdr.din_resid = len - n;
+ if (n > 0)
+ memcpy(bp, resp, n);
+fini:
+ if (free_up)
+ free(free_up);
+ return res;
+}
+
+static int
+do_nvm_pt_low(struct sg_pt_linux_scsi * ptp,
+ struct sg_nvme_passthru_cmd *cmdp, void * dp, int dlen,
+ bool is_read, int time_secs, int vb)
+{
+ const uint32_t cmd_len = sizeof(struct sg_nvme_passthru_cmd);
+ int res;
+ uint32_t n;
+ uint16_t sct_sc;
+ const uint8_t * up = ((const uint8_t *)cmdp) + SG_NVME_PT_OPCODE;
+ char nam[64];
+
+ if (vb)
+ sg_get_nvme_opcode_name(*up, false /* NVM */ , sizeof(nam), nam);
+ else
+ nam[0] = '\0';
+ cmdp->timeout_ms = (time_secs < 0) ? (-time_secs) : (1000 * time_secs);
+ ptp->os_err = 0;
+ if (vb > 2) {
+ pr2ws("NVMe NVM command: %s\n", nam);
+ hex2stderr((const uint8_t *)cmdp, cmd_len, 1);
+ if ((vb > 4) && (! is_read) && dp) {
+ if (dlen > 0) {
+ n = dlen;
+ if ((dlen < 512) || (vb > 5))
+ pr2ws("\nData-out buffer (%u bytes):\n", n);
+ else {
+ pr2ws("\nData-out buffer (first 512 of %u bytes):\n", n);
+ n = 512;
+ }
+ hex2stderr((const uint8_t *)dp, n, 0);
+ }
+ }
+ }
+ res = ioctl(ptp->dev_fd, NVME_IOCTL_IO_CMD, cmdp);
+ if (res < 0) { /* OS error (errno negated) */
+ ptp->os_err = -res;
+ if (vb > 1) {
+ pr2ws("%s: ioctl for %s [0x%x] failed: %s "
+ "(errno=%d)\n", __func__, nam, *up, strerror(-res), -res);
+ }
+ return res;
+ }
+
+ /* Now res contains NVMe completion queue CDW3 31:17 (15 bits) */
+ ptp->nvme_result = cmdp->result;
+ if ((! ptp->nvme_our_sntl) && ptp->io_hdr.response &&
+ (ptp->io_hdr.max_response_len > 3)) {
+ /* build 32 byte "sense" buffer */
+ uint8_t * sbp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.response;
+ uint16_t st = (uint16_t)res;
+
+ n = ptp->io_hdr.max_response_len;
+ n = (n < 32) ? n : 32;
+ memset(sbp, 0 , n);
+ ptp->io_hdr.response_len = n;
+ sg_put_unaligned_le32(cmdp->result,
+ sbp + SG_NVME_PT_CQ_RESULT);
+ if (n > 15) /* LSBit will be 0 (Phase bit) after (st << 1) */
+ sg_put_unaligned_le16(st << 1, sbp + SG_NVME_PT_CQ_STATUS_P);
+ }
+ /* clear upper bits (DNR and More) leaving ((SCT << 8) | SC) */
+ sct_sc = 0x7ff & res; /* 11 bits */
+ ptp->nvme_status = sct_sc;
+ ptp->nvme_stat_dnr = !!(0x4000 & res);
+ ptp->nvme_stat_more = !!(0x2000 & res);
+ if (sct_sc) { /* when non-zero, treat as command error */
+ if (vb > 1) {
+ char b[80];
+
+ pr2ws("%s: ioctl for %s [0x%x] failed, status: %s [0x%x]\n",
+ __func__, nam, *up,
+ sg_get_nvme_cmd_status_str(sct_sc, sizeof(b), b), sct_sc);
+ }
+ return SG_LIB_NVME_STATUS; /* == SCSI_PT_DO_NVME_STATUS */
+ }
+ if ((vb > 4) && is_read && dp) {
+ if (dlen > 0) {
+ n = dlen;
+ if ((dlen < 1024) || (vb > 5))
+ pr2ws("\nData-in buffer (%u bytes):\n", n);
+ else {
+ pr2ws("\nData-in buffer (first 1024 of %u bytes):\n", n);
+ n = 1024;
+ }
+ hex2stderr((const uint8_t *)dp, n, 0);
+ }
+ }
+ return 0;
+}
+
+/* Since ptp can be a char device (e.g. /dev/nvme0) or a blocks device
+ * (e.g. /dev/nvme0n1 or /dev/nvme0n1p3) use NVME_IOCTL_IO_CMD which is
+ * common to both (and takes a timeout). The difficult is that
+ * NVME_IOCTL_IO_CMD takes a nvme_passthru_cmd object point. */
+static int
+sntl_do_nvm_cmd(struct sg_pt_linux_scsi * ptp, struct sg_nvme_user_io * iop,
+ uint32_t dlen, bool is_read, int time_secs, int vb)
+{
+
+ struct sg_nvme_passthru_cmd nvme_pt_cmd;
+ struct sg_nvme_passthru_cmd *cmdp = &nvme_pt_cmd;
+ void * dp = (void *)(sg_uintptr_t)iop->addr;
+
+ memset(cmdp, 0, sizeof(*cmdp));
+ cmdp->opcode = iop->opcode;
+ cmdp->flags = iop->flags;
+ cmdp->nsid = ptp->nvme_nsid;
+ cmdp->addr = iop->addr;
+ cmdp->data_len = dlen;
+ cmdp->cdw10 = iop->slba & 0xffffffff;
+ cmdp->cdw11 = (iop->slba >> 32) & 0xffffffff;
+ cmdp->cdw12 = iop->nblocks; /* lower 16 bits already "0's based" count */
+
+ return do_nvm_pt_low(ptp, cmdp, dp, dlen, is_read, time_secs, vb);
+}
+
+static int
+sntl_rread(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool is_read10 = (SCSI_READ10_OPC == cdbp[0]);
+ bool have_fua = !!(cdbp[1] & 0x8);
+ int res;
+ uint32_t nblks_t10 = 0;
+ struct sg_nvme_user_io io;
+ struct sg_nvme_user_io * iop = &io;
+
+ if (vb > 5)
+ pr2ws("%s: fua=%d, time_secs=%d\n", __func__, (int)have_fua,
+ time_secs);
+ memset(iop, 0, sizeof(*iop));
+ iop->opcode = SG_NVME_NVM_READ;
+ if (is_read10) {
+ iop->slba = sg_get_unaligned_be32(cdbp + 2);
+ nblks_t10 = sg_get_unaligned_be16(cdbp + 7);
+ } else {
+ iop->slba = sg_get_unaligned_be64(cdbp + 2);
+ nblks_t10 = sg_get_unaligned_be32(cdbp + 10);
+ if (nblks_t10 > (UINT16_MAX + 1)) {
+ mk_sense_invalid_fld(ptp, true, 11, -1, vb);
+ return 0;
+ }
+ }
+ if (0 == nblks_t10) { /* NOP in SCSI */
+ if (vb > 4)
+ pr2ws("%s: nblks_t10 is 0, a NOP in SCSI, can't map to NVMe\n",
+ __func__);
+ return 0;
+ }
+ iop->nblocks = nblks_t10 - 1; /* crazy "0's based" */
+ if (have_fua)
+ iop->control |= SG_NVME_RW_CONTROL_FUA;
+ iop->addr = (uint64_t)ptp->io_hdr.din_xferp;
+ res = sntl_do_nvm_cmd(ptp, iop, ptp->io_hdr.din_xfer_len,
+ true /* is_read */, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ }
+ return res;
+}
+
+static int
+sntl_write(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool is_write10 = (SCSI_WRITE10_OPC == cdbp[0]);
+ bool have_fua = !!(cdbp[1] & 0x8);
+ int res;
+ uint32_t nblks_t10 = 0;
+ struct sg_nvme_user_io io;
+ struct sg_nvme_user_io * iop = &io;
+
+ if (vb > 5)
+ pr2ws("%s: fua=%d, time_secs=%d\n", __func__, (int)have_fua,
+ time_secs);
+ memset(iop, 0, sizeof(*iop));
+ iop->opcode = SG_NVME_NVM_WRITE;
+ if (is_write10) {
+ iop->slba = sg_get_unaligned_be32(cdbp + 2);
+ nblks_t10 = sg_get_unaligned_be16(cdbp + 7);
+ } else {
+ iop->slba = sg_get_unaligned_be64(cdbp + 2);
+ nblks_t10 = sg_get_unaligned_be32(cdbp + 10);
+ if (nblks_t10 > (UINT16_MAX + 1)) {
+ mk_sense_invalid_fld(ptp, true, 11, -1, vb);
+ return 0;
+ }
+ }
+ if (0 == nblks_t10) { /* NOP in SCSI */
+ if (vb > 4)
+ pr2ws("%s: nblks_t10 is 0, a NOP in SCSI, can't map to NVMe\n",
+ __func__);
+ return 0;
+ }
+ iop->nblocks = nblks_t10 - 1;
+ if (have_fua)
+ iop->control |= SG_NVME_RW_CONTROL_FUA;
+ iop->addr = (uint64_t)ptp->io_hdr.dout_xferp;
+ res = sntl_do_nvm_cmd(ptp, iop, ptp->io_hdr.dout_xfer_len, false,
+ time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ }
+ return res;
+}
+
+static int
+sntl_verify(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool is_verify10 = (SCSI_VERIFY10_OPC == cdbp[0]);
+ uint8_t bytchk = (cdbp[1] >> 1) & 0x3;
+ uint32_t dlen = 0;
+ int res;
+ uint32_t nblks_t10 = 0;
+ struct sg_nvme_user_io io;
+ struct sg_nvme_user_io * iop = &io;
+
+ if (vb > 5)
+ pr2ws("%s: bytchk=%d, time_secs=%d\n", __func__, bytchk, time_secs);
+ if (bytchk > 1) {
+ mk_sense_invalid_fld(ptp, true, 1, 2, vb);
+ return 0;
+ }
+ memset(iop, 0, sizeof(*iop));
+ iop->opcode = bytchk ? SG_NVME_NVM_COMPARE : SG_NVME_NVM_VERIFY;
+ if (is_verify10) {
+ iop->slba = sg_get_unaligned_be32(cdbp + 2);
+ nblks_t10 = sg_get_unaligned_be16(cdbp + 7);
+ } else {
+ iop->slba = sg_get_unaligned_be64(cdbp + 2);
+ nblks_t10 = sg_get_unaligned_be32(cdbp + 10);
+ if (nblks_t10 > (UINT16_MAX + 1)) {
+ mk_sense_invalid_fld(ptp, true, 11, -1, vb);
+ return 0;
+ }
+ }
+ if (0 == nblks_t10) { /* NOP in SCSI */
+ if (vb > 4)
+ pr2ws("%s: nblks_t10 is 0, a NOP in SCSI, can't map to NVMe\n",
+ __func__);
+ return 0;
+ }
+ iop->nblocks = nblks_t10 - 1;
+ if (bytchk) {
+ iop->addr = (uint64_t)ptp->io_hdr.dout_xferp;
+ dlen = ptp->io_hdr.dout_xfer_len;
+ }
+ res = sntl_do_nvm_cmd(ptp, iop, dlen, false, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ }
+ return res;
+}
+
+static int
+sntl_write_same(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool is_ws10 = (SCSI_WRITE_SAME10_OPC == cdbp[0]);
+ bool ndob = is_ws10 ? false : !!(0x1 & cdbp[1]);
+ int res;
+ int nblks_t10 = 0;
+ struct sg_nvme_user_io io;
+ struct sg_nvme_user_io * iop = &io;
+
+ if (vb > 5)
+ pr2ws("%s: ndob=%d, time_secs=%d\n", __func__, (int)ndob, time_secs);
+ if (! ndob) {
+ int flbas, index, lbafx, lbads, lbsize;
+ uint8_t * up;
+ uint8_t * dp;
+
+ dp = (uint8_t *)(sg_uintptr_t)ptp->io_hdr.dout_xferp;
+ if (dp == NULL)
+ return sg_convert_errno(ENOMEM);
+ if (NULL == ptp->nvme_id_ctlp) {
+ res = sntl_cache_identify(ptp, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ } else if (res)
+ return res;
+ }
+ up = ptp->nvme_id_ctlp;
+ flbas = up[26]; /* NVME FLBAS field from Identify */
+ index = 128 + (4 * (flbas & 0xf));
+ lbafx = sg_get_unaligned_le32(up + index);
+ lbads = (lbafx >> 16) & 0xff; /* bits 16 to 23 inclusive, pow2 */
+ lbsize = 1 << lbads;
+ if (! sg_all_zeros(dp, lbsize)) {
+ mk_sense_asc_ascq(ptp, SPC_SK_ILLEGAL_REQUEST, PCIE_ERR_ASC,
+ PCIE_UNSUPP_REQ_ASCQ, vb);
+ return 0;
+ }
+ /* so given single LB full of zeros, can translate .... */
+ }
+ memset(iop, 0, sizeof(*iop));
+ iop->opcode = SG_NVME_NVM_WRITE_ZEROES;
+ if (is_ws10) {
+ iop->slba = sg_get_unaligned_be32(cdbp + 2);
+ nblks_t10 = sg_get_unaligned_be16(cdbp + 7);
+ } else {
+ uint32_t num = sg_get_unaligned_be32(cdbp + 10);
+
+ iop->slba = sg_get_unaligned_be64(cdbp + 2);
+ if (num > (UINT16_MAX + 1)) {
+ mk_sense_invalid_fld(ptp, true, 11, -1, vb);
+ return 0;
+ } else
+ nblks_t10 = num;
+ }
+ if (0 == nblks_t10) { /* NOP in SCSI */
+ if (vb > 4)
+ pr2ws("%s: nblks_t10 is 0, a NOP in SCSI, can't map to NVMe\n",
+ __func__);
+ return 0;
+ }
+ iop->nblocks = nblks_t10 - 1;
+ res = sntl_do_nvm_cmd(ptp, iop, 0, false, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ }
+ return res;
+}
+
+static int
+sntl_sync_cache(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool immed = !!(0x2 & cdbp[1]);
+ struct sg_nvme_user_io io;
+ struct sg_nvme_user_io * iop = &io;
+ int res;
+
+ if (vb > 5)
+ pr2ws("%s: immed=%d, time_secs=%d\n", __func__, (int)immed,
+ time_secs);
+ memset(iop, 0, sizeof(*iop));
+ iop->opcode = SG_NVME_NVM_FLUSH;
+ if (vb > 4)
+ pr2ws("%s: immed bit, lba and num_lbs fields ignored\n", __func__);
+ res = sntl_do_nvm_cmd(ptp, iop, 0, false, time_secs, vb);
+ if (SG_LIB_NVME_STATUS == res) {
+ mk_sense_from_nvme_status(ptp, vb);
+ return 0;
+ }
+ return res;
+}
+
+static int
+sntl_start_stop(struct sg_pt_linux_scsi * ptp, const uint8_t * cdbp,
+ int time_secs, int vb)
+{
+ bool immed = !!(0x1 & cdbp[1]);
+
+ if (vb > 5)
+ pr2ws("%s: immed=%d, time_secs=%d, ignore\n", __func__, (int)immed,
+ time_secs);
+ if (ptp) { } /* suppress warning */
+ return 0;
+}
+
+/* Executes NVMe Admin command (or at least forwards it to lower layers).
+ * Returns 0 for success, negative numbers are negated 'errno' values from
+ * OS system calls. Positive return values are errors from this package.
+ * When time_secs is 0 the Linux NVMe Admin command default of 60 seconds
+ * is used. */
+int
+sg_do_nvme_pt(struct sg_pt_base * vp, int fd, int time_secs, int vb)
+{
+ bool scsi_cdb;
+ bool is_read = false;
+ int n, len, hold_dev_fd;
+ uint16_t sa;
+ struct sg_pt_linux_scsi * ptp = &vp->impl;
+ struct sg_nvme_passthru_cmd cmd;
+ const uint8_t * cdbp;
+ void * dp = NULL;
+
+ if (! ptp->io_hdr.request) {
+ if (vb)
+ pr2ws("No NVMe command given (set_scsi_pt_cdb())\n");
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+ hold_dev_fd = ptp->dev_fd;
+ if (fd >= 0) {
+ if ((ptp->dev_fd >= 0) && (fd != ptp->dev_fd)) {
+ if (vb)
+ pr2ws("%s: file descriptor given to create() and here "
+ "differ\n", __func__);
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+ ptp->dev_fd = fd;
+ } else if (ptp->dev_fd < 0) {
+ if (vb)
+ pr2ws("%s: invalid file descriptors\n", __func__);
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+ n = ptp->io_hdr.request_len;
+ cdbp = (const uint8_t *)(sg_uintptr_t)ptp->io_hdr.request;
+ if (vb > 4)
+ pr2ws("%s: opcode=0x%x, fd=%d (dev_fd=%d), time_secs=%d\n", __func__,
+ cdbp[0], fd, hold_dev_fd, time_secs);
+ scsi_cdb = sg_is_scsi_cdb(cdbp, n);
+ /* direct NVMe command (i.e. 64 bytes long) or SNTL */
+ ptp->nvme_our_sntl = scsi_cdb;
+ if (scsi_cdb) {
+ switch (cdbp[0]) {
+ case SCSI_INQUIRY_OPC:
+ return sntl_inq(ptp, cdbp, time_secs, vb);
+ case SCSI_REPORT_LUNS_OPC:
+ return sntl_rluns(ptp, cdbp, time_secs, vb);
+ case SCSI_TEST_UNIT_READY_OPC:
+ return sntl_tur(ptp, time_secs, vb);
+ case SCSI_REQUEST_SENSE_OPC:
+ return sntl_req_sense(ptp, cdbp, time_secs, vb);
+ case SCSI_READ10_OPC:
+ case SCSI_READ16_OPC:
+ return sntl_rread(ptp, cdbp, time_secs, vb);
+ case SCSI_WRITE10_OPC:
+ case SCSI_WRITE16_OPC:
+ return sntl_write(ptp, cdbp, time_secs, vb);
+ case SCSI_START_STOP_OPC:
+ return sntl_start_stop(ptp, cdbp, time_secs, vb);
+ case SCSI_SEND_DIAGNOSTIC_OPC:
+ return sntl_senddiag(ptp, cdbp, time_secs, vb);
+ case SCSI_RECEIVE_DIAGNOSTIC_OPC:
+ return sntl_recvdiag(ptp, cdbp, time_secs, vb);
+ case SCSI_MODE_SENSE10_OPC:
+ case SCSI_MODE_SELECT10_OPC:
+ return sntl_mode_ss(ptp, cdbp, time_secs, vb);
+ case SCSI_READ_CAPACITY10_OPC:
+ return sntl_readcap(ptp, cdbp, time_secs, vb);
+ case SCSI_VERIFY10_OPC:
+ case SCSI_VERIFY16_OPC:
+ return sntl_verify(ptp, cdbp, time_secs, vb);
+ case SCSI_WRITE_SAME10_OPC:
+ case SCSI_WRITE_SAME16_OPC:
+ return sntl_write_same(ptp, cdbp, time_secs, vb);
+ case SCSI_SYNC_CACHE10_OPC:
+ case SCSI_SYNC_CACHE16_OPC:
+ return sntl_sync_cache(ptp, cdbp, time_secs, vb);
+ case SCSI_SERVICE_ACT_IN_OPC:
+ if (SCSI_READ_CAPACITY16_SA == (cdbp[1] & SCSI_SA_MSK))
+ return sntl_readcap(ptp, cdbp, time_secs, vb);
+ goto fini;
+ case SCSI_MAINT_IN_OPC:
+ sa = SCSI_SA_MSK & cdbp[1]; /* service action */
+ if (SCSI_REP_SUP_OPCS_OPC == sa)
+ return sntl_rep_opcodes(ptp, cdbp, time_secs, vb);
+ else if (SCSI_REP_SUP_TMFS_OPC == sa)
+ return sntl_rep_tmfs(ptp, cdbp, time_secs, vb);
+ /* fall through */
+ default:
+fini:
+ if (vb > 2) {
+ char b[64];
+
+ sg_get_command_name(cdbp, -1, sizeof(b), b);
+ pr2ws("%s: no translation to NVMe for SCSI %s command\n",
+ __func__, b);
+ }
+ mk_sense_asc_ascq(ptp, SPC_SK_ILLEGAL_REQUEST, INVALID_OPCODE,
+ 0, vb);
+ return 0;
+ }
+ }
+ len = (int)sizeof(cmd);
+ n = (n < len) ? n : len;
+ if (n < 64) {
+ if (vb)
+ pr2ws("%s: command length of %d bytes is too short\n", __func__,
+ n);
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+ memcpy(&cmd, (const uint8_t *)(sg_uintptr_t)ptp->io_hdr.request, n);
+ if (n < len) /* zero out rest of 'cmd' */
+ memset((uint8_t *)&cmd + n, 0, len - n);
+ if (ptp->io_hdr.din_xfer_len > 0) {
+ cmd.data_len = ptp->io_hdr.din_xfer_len;
+ dp = (void *)(sg_uintptr_t)ptp->io_hdr.din_xferp;
+ cmd.addr = (uint64_t)(sg_uintptr_t)ptp->io_hdr.din_xferp;
+ is_read = true;
+ } else if (ptp->io_hdr.dout_xfer_len > 0) {
+ cmd.data_len = ptp->io_hdr.dout_xfer_len;
+ dp = (void *)(sg_uintptr_t)ptp->io_hdr.dout_xferp;
+ cmd.addr = (uint64_t)(sg_uintptr_t)ptp->io_hdr.dout_xferp;
+ is_read = false;
+ }
+ return sg_nvme_admin_cmd(ptp, &cmd, dp, is_read, time_secs, vb);
+}
+
+#else /* (HAVE_NVME && (! IGNORE_NVME)) [around line 140] */
+
+int
+sg_do_nvme_pt(struct sg_pt_base * vp, int fd, int time_secs, int vb)
+{
+ if (vb) {
+ pr2ws("%s: not supported, ", __func__);
+#ifdef HAVE_NVME
+ pr2ws("HAVE_NVME, ");
+#else
+ pr2ws("don't HAVE_NVME, ");
+#endif
+
+#ifdef IGNORE_NVME
+ pr2ws("IGNORE_NVME");
+#else
+ pr2ws("don't IGNORE_NVME");
+#endif
+ pr2ws("\n");
+ }
+ if (vp) { ; } /* suppress warning */
+ if (fd) { ; } /* suppress warning */
+ if (time_secs) { ; } /* suppress warning */
+ return -ENOTTY; /* inappropriate ioctl error */
+}
+
+#endif /* (HAVE_NVME && (! IGNORE_NVME)) */
+
+#if (HAVE_NVME && (! IGNORE_NVME))
+
+int
+do_nvm_pt(struct sg_pt_base * vp, int submq, int timeout_secs, int vb)
+{
+ bool is_read = false;
+ int dlen;
+ struct sg_pt_linux_scsi * ptp = &vp->impl;
+ struct sg_nvme_passthru_cmd cmd;
+ uint8_t * cmdp = (uint8_t *)&cmd;
+ void * dp = NULL;
+
+ if (vb && (submq != 0))
+ pr2ws("%s: warning, uses submit queue 0\n", __func__);
+ if (ptp->dev_fd < 0) {
+ if (vb > 1)
+ pr2ws("%s: no NVMe file descriptor given\n", __func__);
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+ if (! ptp->is_nvme) {
+ if (vb > 1)
+ pr2ws("%s: file descriptor is not NVMe device\n", __func__);
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+ if ((! ptp->io_hdr.request) || (64 != ptp->io_hdr.request_len)) {
+ if (vb > 1)
+ pr2ws("%s: no NVMe 64 byte command present\n", __func__);
+ return SCSI_PT_DO_BAD_PARAMS;
+ }
+ if (sizeof(cmd) > 64)
+ memset(cmdp + 64, 0, sizeof(cmd) - 64);
+ memcpy(cmdp, (uint8_t *)(sg_uintptr_t)ptp->io_hdr.request, 64);
+ ptp->nvme_our_sntl = false;
+
+ dlen = ptp->io_hdr.din_xfer_len;
+ if (dlen > 0) {
+ is_read = true;
+ dp = (void *)(sg_uintptr_t)ptp->io_hdr.din_xferp;
+ } else {
+ dlen = ptp->io_hdr.dout_xfer_len;
+ if (dlen > 0)
+ dp = (void *)(sg_uintptr_t)ptp->io_hdr.dout_xferp;
+ }
+ return do_nvm_pt_low(ptp, &cmd, dp, dlen, is_read, timeout_secs, vb);
+}
+
+#else /* (HAVE_NVME && (! IGNORE_NVME)) */
+
+int
+do_nvm_pt(struct sg_pt_base * vp, int submq, int timeout_secs, int vb)
+{
+ if (vb) {
+ pr2ws("%s: not supported, ", __func__);
+#ifdef HAVE_NVME
+ pr2ws("HAVE_NVME, ");
+#else
+ pr2ws("don't HAVE_NVME, ");
+#endif
+
+#ifdef IGNORE_NVME
+ pr2ws("IGNORE_NVME");
+#else
+ pr2ws("don't IGNORE_NVME");
+#endif
+ }
+ if (vp) { }
+ if (submq) { }
+ if (timeout_secs) { }
+ return SCSI_PT_DO_NOT_SUPPORTED;
+}
+
+#endif /* (HAVE_NVME && (! IGNORE_NVME)) */