diff --git a/sys/cam/ctl/ctl_frontend_iscsi.c b/sys/cam/ctl/ctl_frontend_iscsi.c index b8fafcea69ed..0cbe3bcefc73 100644 --- a/sys/cam/ctl/ctl_frontend_iscsi.c +++ b/sys/cam/ctl/ctl_frontend_iscsi.c @@ -1,3058 +1,3058 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 The FreeBSD Foundation * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ /* * CTL frontend for the iSCSI protocol. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef ICL_KERNEL_PROXY #include #endif #ifdef ICL_KERNEL_PROXY FEATURE(cfiscsi_kernel_proxy, "iSCSI target built with ICL_KERNEL_PROXY"); #endif static MALLOC_DEFINE(M_CFISCSI, "cfiscsi", "Memory used for CTL iSCSI frontend"); static uma_zone_t cfiscsi_data_wait_zone; SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, iscsi, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "CAM Target Layer iSCSI Frontend"); static int debug = 1; SYSCTL_INT(_kern_cam_ctl_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN, &debug, 1, "Enable debug messages"); static int ping_timeout = 5; SYSCTL_INT(_kern_cam_ctl_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout, 5, "Interval between ping (NOP-Out) requests, in seconds"); static int login_timeout = 60; SYSCTL_INT(_kern_cam_ctl_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout, 60, "Time to wait for ctld(8) to finish Login Phase, in seconds"); static int maxtags = 256; SYSCTL_INT(_kern_cam_ctl_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags, 0, "Max number of requests queued by initiator"); #define CFISCSI_DEBUG(X, ...) \ do { \ if (debug > 1) { \ printf("%s: " X "\n", \ __func__, ## __VA_ARGS__); \ } \ } while (0) #define CFISCSI_WARN(X, ...) \ do { \ if (debug > 0) { \ printf("WARNING: %s: " X "\n", \ __func__, ## __VA_ARGS__); \ } \ } while (0) #define CFISCSI_SESSION_DEBUG(S, X, ...) \ do { \ if (debug > 1) { \ printf("%s: %s (%s): " X "\n", \ __func__, S->cs_initiator_addr, \ S->cs_initiator_name, ## __VA_ARGS__); \ } \ } while (0) #define CFISCSI_SESSION_WARN(S, X, ...) \ do { \ if (debug > 0) { \ printf("WARNING: %s (%s): " X "\n", \ S->cs_initiator_addr, \ S->cs_initiator_name, ## __VA_ARGS__); \ } \ } while (0) #define CFISCSI_SESSION_LOCK(X) mtx_lock(&X->cs_lock) #define CFISCSI_SESSION_UNLOCK(X) mtx_unlock(&X->cs_lock) #define CFISCSI_SESSION_LOCK_ASSERT(X) mtx_assert(&X->cs_lock, MA_OWNED) #define CONN_SESSION(X) ((struct cfiscsi_session *)(X)->ic_prv0) #define PDU_SESSION(X) CONN_SESSION((X)->ip_conn) struct cfiscsi_priv { void *request; uint32_t expdatasn; uint32_t r2tsn; }; #define PRIV(io) \ ((struct cfiscsi_priv *)&(io)->io_hdr.ctl_private[CTL_PRIV_FRONTEND]) #define PRIV_REQUEST(io) PRIV(io)->request #define PRIV_EXPDATASN(io) PRIV(io)->expdatasn #define PRIV_R2TSN(io) PRIV(io)->r2tsn static int cfiscsi_init(void); static int cfiscsi_shutdown(void); static void cfiscsi_online(void *arg); static void cfiscsi_offline(void *arg); static int cfiscsi_info(void *arg, struct sbuf *sb); static int cfiscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td); static void cfiscsi_datamove(union ctl_io *io); static void cfiscsi_datamove_in(union ctl_io *io); static void cfiscsi_datamove_out(union ctl_io *io); static void cfiscsi_done(union ctl_io *io); static bool cfiscsi_pdu_update_cmdsn(const struct icl_pdu *request); static void cfiscsi_pdu_handle_nop_out(struct icl_pdu *request); static void cfiscsi_pdu_handle_scsi_command(struct icl_pdu *request); static void cfiscsi_pdu_handle_task_request(struct icl_pdu *request); static void cfiscsi_pdu_handle_data_out(struct icl_pdu *request); static void cfiscsi_pdu_handle_logout_request(struct icl_pdu *request); static void cfiscsi_session_terminate(struct cfiscsi_session *cs); static struct cfiscsi_data_wait *cfiscsi_data_wait_new( struct cfiscsi_session *cs, union ctl_io *io, uint32_t initiator_task_tag, uint32_t *target_transfer_tagp); static void cfiscsi_data_wait_free(struct cfiscsi_session *cs, struct cfiscsi_data_wait *cdw); static struct cfiscsi_target *cfiscsi_target_find(struct cfiscsi_softc *softc, const char *name, uint16_t tag); static struct cfiscsi_target *cfiscsi_target_find_or_create( struct cfiscsi_softc *softc, const char *name, const char *alias, uint16_t tag); static void cfiscsi_target_release(struct cfiscsi_target *ct); static void cfiscsi_session_delete(struct cfiscsi_session *cs); static struct cfiscsi_softc cfiscsi_softc; static struct ctl_frontend cfiscsi_frontend = { .name = "iscsi", .init = cfiscsi_init, .ioctl = cfiscsi_ioctl, .shutdown = cfiscsi_shutdown, }; CTL_FRONTEND_DECLARE(cfiscsi, cfiscsi_frontend); MODULE_DEPEND(cfiscsi, icl, 1, 1, 1); static struct icl_pdu * cfiscsi_pdu_new_response(struct icl_pdu *request, int flags) { return (icl_pdu_new(request->ip_conn, flags)); } static bool cfiscsi_pdu_update_cmdsn(const struct icl_pdu *request) { const struct iscsi_bhs_scsi_command *bhssc; struct cfiscsi_session *cs; uint32_t cmdsn, curcmdsn; cs = PDU_SESSION(request); /* * Every incoming PDU - not just NOP-Out - resets the ping timer. * The purpose of the timeout is to reset the connection when it stalls; * we don't want this to happen when NOP-In or NOP-Out ends up delayed * in some queue. */ cs->cs_timeout = 0; /* * Immediate commands carry cmdsn, but it is neither incremented nor * verified. */ if (request->ip_bhs->bhs_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) return (false); /* * Data-Out PDUs don't contain CmdSN. */ if (request->ip_bhs->bhs_opcode == ISCSI_BHS_OPCODE_SCSI_DATA_OUT) return (false); /* * We're only using fields common for all the request * (initiator -> target) PDUs. */ bhssc = (const struct iscsi_bhs_scsi_command *)request->ip_bhs; curcmdsn = cmdsn = ntohl(bhssc->bhssc_cmdsn); /* * Increment session cmdsn and exit if we received the expected value. */ do { if (atomic_fcmpset_32(&cs->cs_cmdsn, &curcmdsn, cmdsn + 1)) return (false); } while (curcmdsn == cmdsn); /* * The target MUST silently ignore any non-immediate command outside * of this range. */ if (ISCSI_SNLT(cmdsn, curcmdsn) || ISCSI_SNGT(cmdsn, curcmdsn - 1 + maxtags)) { CFISCSI_SESSION_WARN(cs, "received PDU with CmdSN %u, " "while expected %u", cmdsn, curcmdsn); return (true); } /* * We don't support multiple connections now, so any discontinuity in * CmdSN means lost PDUs. Since we don't support PDU retransmission -- * terminate the connection. */ CFISCSI_SESSION_WARN(cs, "received PDU with CmdSN %u, " "while expected %u; dropping connection", cmdsn, curcmdsn); cfiscsi_session_terminate(cs); return (true); } static void cfiscsi_pdu_handle(struct icl_pdu *request) { struct cfiscsi_session *cs; bool ignore; cs = PDU_SESSION(request); ignore = cfiscsi_pdu_update_cmdsn(request); if (ignore) { icl_pdu_free(request); return; } /* * Handle the PDU; this includes e.g. receiving the remaining * part of PDU and submitting the SCSI command to CTL * or queueing a reply. The handling routine is responsible * for freeing the PDU when it's no longer needed. */ switch (request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) { case ISCSI_BHS_OPCODE_NOP_OUT: cfiscsi_pdu_handle_nop_out(request); break; case ISCSI_BHS_OPCODE_SCSI_COMMAND: cfiscsi_pdu_handle_scsi_command(request); break; case ISCSI_BHS_OPCODE_TASK_REQUEST: cfiscsi_pdu_handle_task_request(request); break; case ISCSI_BHS_OPCODE_SCSI_DATA_OUT: cfiscsi_pdu_handle_data_out(request); break; case ISCSI_BHS_OPCODE_LOGOUT_REQUEST: cfiscsi_pdu_handle_logout_request(request); break; default: CFISCSI_SESSION_WARN(cs, "received PDU with unsupported " "opcode 0x%x; dropping connection", request->ip_bhs->bhs_opcode); icl_pdu_free(request); cfiscsi_session_terminate(cs); } } static void cfiscsi_receive_callback(struct icl_pdu *request) { #ifdef ICL_KERNEL_PROXY struct cfiscsi_session *cs; cs = PDU_SESSION(request); if (cs->cs_waiting_for_ctld || cs->cs_login_phase) { if (cs->cs_login_pdu == NULL) cs->cs_login_pdu = request; else icl_pdu_free(request); cv_signal(&cs->cs_login_cv); return; } #endif cfiscsi_pdu_handle(request); } static void cfiscsi_error_callback(struct icl_conn *ic) { struct cfiscsi_session *cs; cs = CONN_SESSION(ic); CFISCSI_SESSION_WARN(cs, "connection error; dropping connection"); cfiscsi_session_terminate(cs); } static int cfiscsi_pdu_prepare(struct icl_pdu *response) { struct cfiscsi_session *cs; struct iscsi_bhs_scsi_response *bhssr; bool advance_statsn = true; uint32_t cmdsn; cs = PDU_SESSION(response); CFISCSI_SESSION_LOCK_ASSERT(cs); /* * We're only using fields common for all the response * (target -> initiator) PDUs. */ bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs; /* * 10.8.3: "The StatSN for this connection is not advanced * after this PDU is sent." */ if (bhssr->bhssr_opcode == ISCSI_BHS_OPCODE_R2T) advance_statsn = false; /* * 10.19.2: "However, when the Initiator Task Tag is set to 0xffffffff, * StatSN for the connection is not advanced after this PDU is sent." */ if (bhssr->bhssr_opcode == ISCSI_BHS_OPCODE_NOP_IN && bhssr->bhssr_initiator_task_tag == 0xffffffff) advance_statsn = false; /* * See the comment below - StatSN is not meaningful and must * not be advanced. */ if (bhssr->bhssr_opcode == ISCSI_BHS_OPCODE_SCSI_DATA_IN && (bhssr->bhssr_flags & BHSDI_FLAGS_S) == 0) advance_statsn = false; /* * 10.7.3: "The fields StatSN, Status, and Residual Count * only have meaningful content if the S bit is set to 1." */ if (bhssr->bhssr_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN || (bhssr->bhssr_flags & BHSDI_FLAGS_S)) bhssr->bhssr_statsn = htonl(cs->cs_statsn); cmdsn = cs->cs_cmdsn; bhssr->bhssr_expcmdsn = htonl(cmdsn); bhssr->bhssr_maxcmdsn = htonl(cmdsn - 1 + imax(0, maxtags - cs->cs_outstanding_ctl_pdus)); if (advance_statsn) cs->cs_statsn++; return (0); } static void cfiscsi_pdu_queue(struct icl_pdu *response) { struct cfiscsi_session *cs; cs = PDU_SESSION(response); CFISCSI_SESSION_LOCK(cs); cfiscsi_pdu_prepare(response); icl_pdu_queue(response); CFISCSI_SESSION_UNLOCK(cs); } static void cfiscsi_pdu_queue_cb(struct icl_pdu *response, icl_pdu_cb cb) { struct cfiscsi_session *cs = PDU_SESSION(response); CFISCSI_SESSION_LOCK(cs); cfiscsi_pdu_prepare(response); icl_pdu_queue_cb(response, cb); CFISCSI_SESSION_UNLOCK(cs); } static void cfiscsi_pdu_handle_nop_out(struct icl_pdu *request) { struct cfiscsi_session *cs; struct iscsi_bhs_nop_out *bhsno; struct iscsi_bhs_nop_in *bhsni; struct icl_pdu *response; void *data = NULL; size_t datasize; int error; cs = PDU_SESSION(request); bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs; if (bhsno->bhsno_initiator_task_tag == 0xffffffff) { /* * Nothing to do, iscsi_pdu_update_statsn() already * zeroed the timeout. */ icl_pdu_free(request); return; } datasize = icl_pdu_data_segment_length(request); if (datasize > 0) { data = malloc(datasize, M_CFISCSI, M_NOWAIT | M_ZERO); if (data == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; " "dropping connection"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } icl_pdu_get_data(request, 0, data, datasize); } response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; " "droppping connection"); free(data, M_CFISCSI); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs; bhsni->bhsni_opcode = ISCSI_BHS_OPCODE_NOP_IN; bhsni->bhsni_flags = 0x80; bhsni->bhsni_initiator_task_tag = bhsno->bhsno_initiator_task_tag; bhsni->bhsni_target_transfer_tag = 0xffffffff; if (datasize > 0) { error = icl_pdu_append_data(response, data, datasize, M_NOWAIT); if (error != 0) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; " "dropping connection"); free(data, M_CFISCSI); icl_pdu_free(request); icl_pdu_free(response); cfiscsi_session_terminate(cs); return; } free(data, M_CFISCSI); } icl_pdu_free(request); cfiscsi_pdu_queue(response); } static void cfiscsi_pdu_handle_scsi_command(struct icl_pdu *request) { struct iscsi_bhs_scsi_command *bhssc; struct cfiscsi_session *cs; union ctl_io *io; int error; cs = PDU_SESSION(request); bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs; //CFISCSI_SESSION_DEBUG(cs, "initiator task tag 0x%x", // bhssc->bhssc_initiator_task_tag); if (request->ip_data_len > 0 && cs->cs_immediate_data == false) { CFISCSI_SESSION_WARN(cs, "unsolicited data with " "ImmediateData=No; dropping connection"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } io = ctl_alloc_io(cs->cs_target->ct_port.ctl_pool_ref); ctl_zero_io(io); PRIV_REQUEST(io) = request; io->io_hdr.io_type = CTL_IO_SCSI; io->io_hdr.nexus.initid = cs->cs_ctl_initid; io->io_hdr.nexus.targ_port = cs->cs_target->ct_port.targ_port; io->io_hdr.nexus.targ_lun = ctl_decode_lun(be64toh(bhssc->bhssc_lun)); io->scsiio.priority = (bhssc->bhssc_pri & BHSSC_PRI_MASK) >> BHSSC_PRI_SHIFT; io->scsiio.tag_num = bhssc->bhssc_initiator_task_tag; switch ((bhssc->bhssc_flags & BHSSC_FLAGS_ATTR)) { case BHSSC_FLAGS_ATTR_UNTAGGED: io->scsiio.tag_type = CTL_TAG_UNTAGGED; break; case BHSSC_FLAGS_ATTR_SIMPLE: io->scsiio.tag_type = CTL_TAG_SIMPLE; break; case BHSSC_FLAGS_ATTR_ORDERED: io->scsiio.tag_type = CTL_TAG_ORDERED; break; case BHSSC_FLAGS_ATTR_HOQ: io->scsiio.tag_type = CTL_TAG_HEAD_OF_QUEUE; break; case BHSSC_FLAGS_ATTR_ACA: io->scsiio.tag_type = CTL_TAG_ACA; break; default: io->scsiio.tag_type = CTL_TAG_UNTAGGED; CFISCSI_SESSION_WARN(cs, "unhandled tag type %d", bhssc->bhssc_flags & BHSSC_FLAGS_ATTR); break; } io->scsiio.cdb_len = sizeof(bhssc->bhssc_cdb); /* Which is 16. */ memcpy(io->scsiio.cdb, bhssc->bhssc_cdb, sizeof(bhssc->bhssc_cdb)); refcount_acquire(&cs->cs_outstanding_ctl_pdus); error = ctl_run(io); if (error != CTL_RETVAL_COMPLETE) { CFISCSI_SESSION_WARN(cs, "ctl_run() failed; error %d; " "dropping connection", error); ctl_free_io(io); refcount_release(&cs->cs_outstanding_ctl_pdus); icl_pdu_free(request); cfiscsi_session_terminate(cs); } } static void cfiscsi_pdu_handle_task_request(struct icl_pdu *request) { struct iscsi_bhs_task_management_request *bhstmr; struct iscsi_bhs_task_management_response *bhstmr2; struct icl_pdu *response; struct cfiscsi_session *cs; union ctl_io *io; int error; cs = PDU_SESSION(request); bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs; io = ctl_alloc_io(cs->cs_target->ct_port.ctl_pool_ref); ctl_zero_io(io); PRIV_REQUEST(io) = request; io->io_hdr.io_type = CTL_IO_TASK; io->io_hdr.nexus.initid = cs->cs_ctl_initid; io->io_hdr.nexus.targ_port = cs->cs_target->ct_port.targ_port; io->io_hdr.nexus.targ_lun = ctl_decode_lun(be64toh(bhstmr->bhstmr_lun)); io->taskio.tag_type = CTL_TAG_SIMPLE; /* XXX */ switch (bhstmr->bhstmr_function & ~0x80) { case BHSTMR_FUNCTION_ABORT_TASK: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_ABORT_TASK"); #endif io->taskio.task_action = CTL_TASK_ABORT_TASK; io->taskio.tag_num = bhstmr->bhstmr_referenced_task_tag; break; case BHSTMR_FUNCTION_ABORT_TASK_SET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_ABORT_TASK_SET"); #endif io->taskio.task_action = CTL_TASK_ABORT_TASK_SET; break; case BHSTMR_FUNCTION_CLEAR_TASK_SET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_CLEAR_TASK_SET"); #endif io->taskio.task_action = CTL_TASK_CLEAR_TASK_SET; break; case BHSTMR_FUNCTION_LOGICAL_UNIT_RESET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_LOGICAL_UNIT_RESET"); #endif io->taskio.task_action = CTL_TASK_LUN_RESET; break; case BHSTMR_FUNCTION_TARGET_WARM_RESET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_TARGET_WARM_RESET"); #endif io->taskio.task_action = CTL_TASK_TARGET_RESET; break; case BHSTMR_FUNCTION_TARGET_COLD_RESET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_TARGET_COLD_RESET"); #endif io->taskio.task_action = CTL_TASK_TARGET_RESET; break; case BHSTMR_FUNCTION_QUERY_TASK: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_QUERY_TASK"); #endif io->taskio.task_action = CTL_TASK_QUERY_TASK; io->taskio.tag_num = bhstmr->bhstmr_referenced_task_tag; break; case BHSTMR_FUNCTION_QUERY_TASK_SET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_QUERY_TASK_SET"); #endif io->taskio.task_action = CTL_TASK_QUERY_TASK_SET; break; case BHSTMR_FUNCTION_I_T_NEXUS_RESET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_I_T_NEXUS_RESET"); #endif io->taskio.task_action = CTL_TASK_I_T_NEXUS_RESET; break; case BHSTMR_FUNCTION_QUERY_ASYNC_EVENT: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_QUERY_ASYNC_EVENT"); #endif io->taskio.task_action = CTL_TASK_QUERY_ASYNC_EVENT; break; default: CFISCSI_SESSION_DEBUG(cs, "unsupported function 0x%x", bhstmr->bhstmr_function & ~0x80); ctl_free_io(io); response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; " "dropping connection"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } bhstmr2 = (struct iscsi_bhs_task_management_response *) response->ip_bhs; bhstmr2->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_RESPONSE; bhstmr2->bhstmr_flags = 0x80; bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_NOT_SUPPORTED; bhstmr2->bhstmr_initiator_task_tag = bhstmr->bhstmr_initiator_task_tag; icl_pdu_free(request); cfiscsi_pdu_queue(response); return; } refcount_acquire(&cs->cs_outstanding_ctl_pdus); error = ctl_run(io); if (error != CTL_RETVAL_COMPLETE) { CFISCSI_SESSION_WARN(cs, "ctl_run() failed; error %d; " "dropping connection", error); ctl_free_io(io); refcount_release(&cs->cs_outstanding_ctl_pdus); icl_pdu_free(request); cfiscsi_session_terminate(cs); } } static bool cfiscsi_handle_data_segment(struct icl_pdu *request, struct cfiscsi_data_wait *cdw) { struct iscsi_bhs_data_out *bhsdo; struct cfiscsi_session *cs; struct ctl_sg_entry ctl_sg_entry, *ctl_sglist; size_t copy_len, len, off, buffer_offset; int ctl_sg_count; union ctl_io *io; cs = PDU_SESSION(request); KASSERT((request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_DATA_OUT || (request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND, ("bad opcode 0x%x", request->ip_bhs->bhs_opcode)); /* * We're only using fields common for Data-Out and SCSI Command PDUs. */ bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs; io = cdw->cdw_ctl_io; KASSERT((io->io_hdr.flags & CTL_FLAG_DATA_MASK) != CTL_FLAG_DATA_IN, ("CTL_FLAG_DATA_IN")); #if 0 CFISCSI_SESSION_DEBUG(cs, "received %zd bytes out of %d", request->ip_data_len, io->scsiio.kern_total_len); #endif if (io->scsiio.kern_sg_entries > 0) { ctl_sglist = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; ctl_sg_count = io->scsiio.kern_sg_entries; } else { ctl_sglist = &ctl_sg_entry; ctl_sglist->addr = io->scsiio.kern_data_ptr; ctl_sglist->len = io->scsiio.kern_data_len; ctl_sg_count = 1; } if ((request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_DATA_OUT) buffer_offset = ntohl(bhsdo->bhsdo_buffer_offset); else buffer_offset = 0; len = icl_pdu_data_segment_length(request); /* * Make sure the offset, as sent by the initiator, matches the offset * we're supposed to be at in the scatter-gather list. */ if (buffer_offset > io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled || buffer_offset + len <= io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled) { CFISCSI_SESSION_WARN(cs, "received bad buffer offset %zd, " "expected %zd; dropping connection", buffer_offset, (size_t)io->scsiio.kern_rel_offset + (size_t)io->scsiio.ext_data_filled); ctl_set_data_phase_error(&io->scsiio); cfiscsi_session_terminate(cs); return (true); } /* * This is the offset within the PDU data segment, as opposed * to buffer_offset, which is the offset within the task (SCSI * command). */ off = io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled - buffer_offset; /* * Iterate over the scatter/gather segments, filling them with data * from the PDU data segment. Note that this can get called multiple * times for one SCSI command; the cdw structure holds state for the * scatter/gather list. */ for (;;) { KASSERT(cdw->cdw_sg_index < ctl_sg_count, ("cdw->cdw_sg_index >= ctl_sg_count")); if (cdw->cdw_sg_len == 0) { cdw->cdw_sg_addr = ctl_sglist[cdw->cdw_sg_index].addr; cdw->cdw_sg_len = ctl_sglist[cdw->cdw_sg_index].len; } KASSERT(off <= len, ("len > off")); copy_len = len - off; if (copy_len > cdw->cdw_sg_len) copy_len = cdw->cdw_sg_len; icl_pdu_get_data(request, off, cdw->cdw_sg_addr, copy_len); cdw->cdw_sg_addr += copy_len; cdw->cdw_sg_len -= copy_len; off += copy_len; io->scsiio.ext_data_filled += copy_len; io->scsiio.kern_data_resid -= copy_len; if (cdw->cdw_sg_len == 0) { /* * End of current segment. */ if (cdw->cdw_sg_index == ctl_sg_count - 1) { /* * Last segment in scatter/gather list. */ break; } cdw->cdw_sg_index++; } if (off == len) { /* * End of PDU payload. */ break; } } if (len > off) { /* * In case of unsolicited data, it's possible that the buffer * provided by CTL is smaller than negotiated FirstBurstLength. * Just ignore the superfluous data; will ask for them with R2T * on next call to cfiscsi_datamove(). * * This obviously can only happen with SCSI Command PDU. */ if ((request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND) return (true); CFISCSI_SESSION_WARN(cs, "received too much data: got %zd bytes, " "expected %zd; dropping connection", icl_pdu_data_segment_length(request), off); ctl_set_data_phase_error(&io->scsiio); cfiscsi_session_terminate(cs); return (true); } if (io->scsiio.ext_data_filled == cdw->cdw_r2t_end && (bhsdo->bhsdo_flags & BHSDO_FLAGS_F) == 0) { CFISCSI_SESSION_WARN(cs, "got the final packet without " "the F flag; flags = 0x%x; dropping connection", bhsdo->bhsdo_flags); ctl_set_data_phase_error(&io->scsiio); cfiscsi_session_terminate(cs); return (true); } if (io->scsiio.ext_data_filled != cdw->cdw_r2t_end && (bhsdo->bhsdo_flags & BHSDO_FLAGS_F) != 0) { if ((request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_DATA_OUT) { CFISCSI_SESSION_WARN(cs, "got the final packet, but the " "transmitted size was %zd bytes instead of %d; " "dropping connection", (size_t)io->scsiio.ext_data_filled, cdw->cdw_r2t_end); ctl_set_data_phase_error(&io->scsiio); cfiscsi_session_terminate(cs); return (true); } else { /* * For SCSI Command PDU, this just means we need to * solicit more data by sending R2T. */ return (false); } } if (io->scsiio.ext_data_filled == cdw->cdw_r2t_end) { #if 0 CFISCSI_SESSION_DEBUG(cs, "no longer expecting Data-Out with target " "transfer tag 0x%x", cdw->cdw_target_transfer_tag); #endif return (true); } return (false); } static void cfiscsi_pdu_handle_data_out(struct icl_pdu *request) { struct iscsi_bhs_data_out *bhsdo; struct cfiscsi_session *cs; struct cfiscsi_data_wait *cdw = NULL; union ctl_io *io; bool done; cs = PDU_SESSION(request); bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs; CFISCSI_SESSION_LOCK(cs); TAILQ_FOREACH(cdw, &cs->cs_waiting_for_data_out, cdw_next) { #if 0 CFISCSI_SESSION_DEBUG(cs, "have ttt 0x%x, itt 0x%x; looking for " "ttt 0x%x, itt 0x%x", bhsdo->bhsdo_target_transfer_tag, bhsdo->bhsdo_initiator_task_tag, cdw->cdw_target_transfer_tag, cdw->cdw_initiator_task_tag)); #endif if (bhsdo->bhsdo_target_transfer_tag == cdw->cdw_target_transfer_tag) break; } CFISCSI_SESSION_UNLOCK(cs); if (cdw == NULL) { CFISCSI_SESSION_WARN(cs, "data transfer tag 0x%x, initiator task tag " "0x%x, not found; dropping connection", bhsdo->bhsdo_target_transfer_tag, bhsdo->bhsdo_initiator_task_tag); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } if (cdw->cdw_datasn != ntohl(bhsdo->bhsdo_datasn)) { CFISCSI_SESSION_WARN(cs, "received Data-Out PDU with " "DataSN %u, while expected %u; dropping connection", ntohl(bhsdo->bhsdo_datasn), cdw->cdw_datasn); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } - cdw->cdw_datasn++; + cdw->cdw_datasn += request->ip_additional_pdus + 1; io = cdw->cdw_ctl_io; KASSERT((io->io_hdr.flags & CTL_FLAG_DATA_MASK) != CTL_FLAG_DATA_IN, ("CTL_FLAG_DATA_IN")); done = cfiscsi_handle_data_segment(request, cdw); if (done) { CFISCSI_SESSION_LOCK(cs); TAILQ_REMOVE(&cs->cs_waiting_for_data_out, cdw, cdw_next); CFISCSI_SESSION_UNLOCK(cs); done = (io->scsiio.ext_data_filled != cdw->cdw_r2t_end || io->scsiio.ext_data_filled == io->scsiio.kern_data_len); cfiscsi_data_wait_free(cs, cdw); io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG; if (done) ctl_datamove_done(io, false); else cfiscsi_datamove_out(io); } icl_pdu_free(request); } static void cfiscsi_pdu_handle_logout_request(struct icl_pdu *request) { struct iscsi_bhs_logout_request *bhslr; struct iscsi_bhs_logout_response *bhslr2; struct icl_pdu *response; struct cfiscsi_session *cs; cs = PDU_SESSION(request); bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs; switch (bhslr->bhslr_reason & 0x7f) { case BHSLR_REASON_CLOSE_SESSION: case BHSLR_REASON_CLOSE_CONNECTION: response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_DEBUG(cs, "failed to allocate memory"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } bhslr2 = (struct iscsi_bhs_logout_response *)response->ip_bhs; bhslr2->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_RESPONSE; bhslr2->bhslr_flags = 0x80; bhslr2->bhslr_response = BHSLR_RESPONSE_CLOSED_SUCCESSFULLY; bhslr2->bhslr_initiator_task_tag = bhslr->bhslr_initiator_task_tag; icl_pdu_free(request); cfiscsi_pdu_queue(response); cfiscsi_session_terminate(cs); break; case BHSLR_REASON_REMOVE_FOR_RECOVERY: response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; dropping connection"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } bhslr2 = (struct iscsi_bhs_logout_response *)response->ip_bhs; bhslr2->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_RESPONSE; bhslr2->bhslr_flags = 0x80; bhslr2->bhslr_response = BHSLR_RESPONSE_RECOVERY_NOT_SUPPORTED; bhslr2->bhslr_initiator_task_tag = bhslr->bhslr_initiator_task_tag; icl_pdu_free(request); cfiscsi_pdu_queue(response); break; default: CFISCSI_SESSION_WARN(cs, "invalid reason 0%x; dropping connection", bhslr->bhslr_reason); icl_pdu_free(request); cfiscsi_session_terminate(cs); break; } } static void cfiscsi_callout(void *context) { struct icl_pdu *cp; struct iscsi_bhs_nop_in *bhsni; struct cfiscsi_session *cs; cs = context; if (cs->cs_terminating) return; callout_schedule(&cs->cs_callout, 1 * hz); atomic_add_int(&cs->cs_timeout, 1); #ifdef ICL_KERNEL_PROXY if (cs->cs_waiting_for_ctld || cs->cs_login_phase) { if (login_timeout > 0 && cs->cs_timeout > login_timeout) { CFISCSI_SESSION_WARN(cs, "login timed out after " "%d seconds; dropping connection", cs->cs_timeout); cfiscsi_session_terminate(cs); } return; } #endif if (ping_timeout <= 0) { /* * Pings are disabled. Don't send NOP-In in this case; * user might have disabled pings to work around problems * with certain initiators that can't properly handle * NOP-In, such as iPXE. Reset the timeout, to avoid * triggering reconnection, should the user decide to * reenable them. */ cs->cs_timeout = 0; return; } if (cs->cs_timeout >= ping_timeout) { CFISCSI_SESSION_WARN(cs, "no ping reply (NOP-Out) after %d seconds; " "dropping connection", ping_timeout); cfiscsi_session_terminate(cs); return; } /* * If the ping was reset less than one second ago - which means * that we've received some PDU during the last second - assume * the traffic flows correctly and don't bother sending a NOP-Out. * * (It's 2 - one for one second, and one for incrementing is_timeout * earlier in this routine.) */ if (cs->cs_timeout < 2) return; cp = icl_pdu_new(cs->cs_conn, M_NOWAIT); if (cp == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory"); return; } bhsni = (struct iscsi_bhs_nop_in *)cp->ip_bhs; bhsni->bhsni_opcode = ISCSI_BHS_OPCODE_NOP_IN; bhsni->bhsni_flags = 0x80; bhsni->bhsni_initiator_task_tag = 0xffffffff; cfiscsi_pdu_queue(cp); } static struct cfiscsi_data_wait * cfiscsi_data_wait_new(struct cfiscsi_session *cs, union ctl_io *io, uint32_t initiator_task_tag, uint32_t *target_transfer_tagp) { struct cfiscsi_data_wait *cdw; int error; cdw = uma_zalloc(cfiscsi_data_wait_zone, M_NOWAIT | M_ZERO); if (cdw == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate %zd bytes", sizeof(*cdw)); return (NULL); } error = icl_conn_transfer_setup(cs->cs_conn, io, target_transfer_tagp, &cdw->cdw_icl_prv); if (error != 0) { CFISCSI_SESSION_WARN(cs, "icl_conn_transfer_setup() failed with error %d", error); uma_zfree(cfiscsi_data_wait_zone, cdw); return (NULL); } cdw->cdw_ctl_io = io; cdw->cdw_target_transfer_tag = *target_transfer_tagp; cdw->cdw_initiator_task_tag = initiator_task_tag; return (cdw); } static void cfiscsi_data_wait_free(struct cfiscsi_session *cs, struct cfiscsi_data_wait *cdw) { icl_conn_transfer_done(cs->cs_conn, cdw->cdw_icl_prv); uma_zfree(cfiscsi_data_wait_zone, cdw); } static void cfiscsi_session_terminate_tasks(struct cfiscsi_session *cs) { struct cfiscsi_data_wait *cdw; union ctl_io *io, *cdw_io; int error, last, wait; if (cs->cs_target == NULL) return; /* No target yet, so nothing to do. */ io = ctl_alloc_io(cs->cs_target->ct_port.ctl_pool_ref); ctl_zero_io(io); PRIV_REQUEST(io) = cs; io->io_hdr.io_type = CTL_IO_TASK; io->io_hdr.nexus.initid = cs->cs_ctl_initid; io->io_hdr.nexus.targ_port = cs->cs_target->ct_port.targ_port; io->io_hdr.nexus.targ_lun = 0; io->taskio.tag_type = CTL_TAG_SIMPLE; /* XXX */ io->taskio.task_action = CTL_TASK_I_T_NEXUS_RESET; wait = cs->cs_outstanding_ctl_pdus; refcount_acquire(&cs->cs_outstanding_ctl_pdus); error = ctl_run(io); if (error != CTL_RETVAL_COMPLETE) { CFISCSI_SESSION_WARN(cs, "ctl_run() failed; error %d", error); refcount_release(&cs->cs_outstanding_ctl_pdus); ctl_free_io(io); } CFISCSI_SESSION_LOCK(cs); while ((cdw = TAILQ_FIRST(&cs->cs_waiting_for_data_out)) != NULL) { TAILQ_REMOVE(&cs->cs_waiting_for_data_out, cdw, cdw_next); CFISCSI_SESSION_UNLOCK(cs); /* * Set nonzero port status; this prevents backends from * assuming that the data transfer actually succeeded * and writing uninitialized data to disk. */ cdw_io = cdw->cdw_ctl_io; cdw_io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG; cdw_io->scsiio.io_hdr.port_status = 42; cfiscsi_data_wait_free(cs, cdw); ctl_datamove_done(cdw_io, false); CFISCSI_SESSION_LOCK(cs); } CFISCSI_SESSION_UNLOCK(cs); /* * Wait for CTL to terminate all the tasks. */ if (wait > 0) CFISCSI_SESSION_WARN(cs, "waiting for CTL to terminate %d tasks", wait); for (;;) { refcount_acquire(&cs->cs_outstanding_ctl_pdus); last = refcount_release(&cs->cs_outstanding_ctl_pdus); if (last != 0) break; tsleep(__DEVOLATILE(void *, &cs->cs_outstanding_ctl_pdus), 0, "cfiscsi_terminate", hz / 100); } if (wait > 0) CFISCSI_SESSION_WARN(cs, "tasks terminated"); } static void cfiscsi_maintenance_thread(void *arg) { struct cfiscsi_session *cs; cs = arg; for (;;) { CFISCSI_SESSION_LOCK(cs); if (cs->cs_terminating == false || cs->cs_handoff_in_progress) cv_wait(&cs->cs_maintenance_cv, &cs->cs_lock); CFISCSI_SESSION_UNLOCK(cs); if (cs->cs_terminating && cs->cs_handoff_in_progress == false) { /* * We used to wait up to 30 seconds to deliver queued * PDUs to the initiator. We also tried hard to deliver * SCSI Responses for the aborted PDUs. We don't do * that anymore. We might need to revisit that. */ callout_drain(&cs->cs_callout); icl_conn_close(cs->cs_conn); /* * At this point ICL receive thread is no longer * running; no new tasks can be queued. */ cfiscsi_session_terminate_tasks(cs); cfiscsi_session_delete(cs); kthread_exit(); return; } CFISCSI_SESSION_DEBUG(cs, "nothing to do"); } } static void cfiscsi_session_terminate(struct cfiscsi_session *cs) { cs->cs_terminating = true; cv_signal(&cs->cs_maintenance_cv); #ifdef ICL_KERNEL_PROXY cv_signal(&cs->cs_login_cv); #endif } static int cfiscsi_session_register_initiator(struct cfiscsi_session *cs) { struct cfiscsi_target *ct; char *name; int i; KASSERT(cs->cs_ctl_initid == -1, ("already registered")); ct = cs->cs_target; name = strdup(cs->cs_initiator_id, M_CTL); i = ctl_add_initiator(&ct->ct_port, -1, 0, name); if (i < 0) { CFISCSI_SESSION_WARN(cs, "ctl_add_initiator failed with error %d", i); cs->cs_ctl_initid = -1; return (1); } cs->cs_ctl_initid = i; #if 0 CFISCSI_SESSION_DEBUG(cs, "added initiator id %d", i); #endif return (0); } static void cfiscsi_session_unregister_initiator(struct cfiscsi_session *cs) { int error; if (cs->cs_ctl_initid == -1) return; error = ctl_remove_initiator(&cs->cs_target->ct_port, cs->cs_ctl_initid); if (error != 0) { CFISCSI_SESSION_WARN(cs, "ctl_remove_initiator failed with error %d", error); } cs->cs_ctl_initid = -1; } static struct cfiscsi_session * cfiscsi_session_new(struct cfiscsi_softc *softc, const char *offload) { struct cfiscsi_session *cs; int error; cs = malloc(sizeof(*cs), M_CFISCSI, M_NOWAIT | M_ZERO); if (cs == NULL) { CFISCSI_WARN("malloc failed"); return (NULL); } cs->cs_ctl_initid = -1; refcount_init(&cs->cs_outstanding_ctl_pdus, 0); TAILQ_INIT(&cs->cs_waiting_for_data_out); mtx_init(&cs->cs_lock, "cfiscsi_lock", NULL, MTX_DEF); cv_init(&cs->cs_maintenance_cv, "cfiscsi_mt"); #ifdef ICL_KERNEL_PROXY cv_init(&cs->cs_login_cv, "cfiscsi_login"); #endif /* * The purpose of this is to avoid racing with session shutdown. * Otherwise we could have the maintenance thread call icl_conn_close() * before we call icl_conn_handoff(). */ cs->cs_handoff_in_progress = true; cs->cs_conn = icl_new_conn(offload, false, "cfiscsi", &cs->cs_lock); if (cs->cs_conn == NULL) { free(cs, M_CFISCSI); return (NULL); } cs->cs_conn->ic_receive = cfiscsi_receive_callback; cs->cs_conn->ic_error = cfiscsi_error_callback; cs->cs_conn->ic_prv0 = cs; error = kthread_add(cfiscsi_maintenance_thread, cs, NULL, NULL, 0, 0, "cfiscsimt"); if (error != 0) { CFISCSI_SESSION_WARN(cs, "kthread_add(9) failed with error %d", error); free(cs, M_CFISCSI); return (NULL); } mtx_lock(&softc->lock); cs->cs_id = ++softc->last_session_id; TAILQ_INSERT_TAIL(&softc->sessions, cs, cs_next); mtx_unlock(&softc->lock); /* * Start pinging the initiator. */ callout_init(&cs->cs_callout, 1); callout_reset(&cs->cs_callout, 1 * hz, cfiscsi_callout, cs); return (cs); } static void cfiscsi_session_delete(struct cfiscsi_session *cs) { struct cfiscsi_softc *softc; softc = &cfiscsi_softc; KASSERT(cs->cs_outstanding_ctl_pdus == 0, ("destroying session with outstanding CTL pdus")); KASSERT(TAILQ_EMPTY(&cs->cs_waiting_for_data_out), ("destroying session with non-empty queue")); mtx_lock(&softc->lock); TAILQ_REMOVE(&softc->sessions, cs, cs_next); mtx_unlock(&softc->lock); cfiscsi_session_unregister_initiator(cs); if (cs->cs_target != NULL) cfiscsi_target_release(cs->cs_target); icl_conn_close(cs->cs_conn); icl_conn_free(cs->cs_conn); free(cs, M_CFISCSI); cv_signal(&softc->sessions_cv); } static int cfiscsi_init(void) { struct cfiscsi_softc *softc; softc = &cfiscsi_softc; bzero(softc, sizeof(*softc)); mtx_init(&softc->lock, "cfiscsi", NULL, MTX_DEF); cv_init(&softc->sessions_cv, "cfiscsi_sessions"); #ifdef ICL_KERNEL_PROXY cv_init(&softc->accept_cv, "cfiscsi_accept"); #endif TAILQ_INIT(&softc->sessions); TAILQ_INIT(&softc->targets); cfiscsi_data_wait_zone = uma_zcreate("cfiscsi_data_wait", sizeof(struct cfiscsi_data_wait), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); return (0); } static int cfiscsi_shutdown(void) { struct cfiscsi_softc *softc = &cfiscsi_softc; if (!TAILQ_EMPTY(&softc->sessions) || !TAILQ_EMPTY(&softc->targets)) return (EBUSY); uma_zdestroy(cfiscsi_data_wait_zone); #ifdef ICL_KERNEL_PROXY cv_destroy(&softc->accept_cv); #endif cv_destroy(&softc->sessions_cv); mtx_destroy(&softc->lock); return (0); } #ifdef ICL_KERNEL_PROXY static void cfiscsi_accept(struct socket *so, struct sockaddr *sa, int portal_id) { struct cfiscsi_session *cs; cs = cfiscsi_session_new(&cfiscsi_softc, NULL); if (cs == NULL) { CFISCSI_WARN("failed to create session"); return; } icl_conn_handoff_sock(cs->cs_conn, so); cs->cs_initiator_sa = sa; cs->cs_portal_id = portal_id; cs->cs_handoff_in_progress = false; cs->cs_waiting_for_ctld = true; cv_signal(&cfiscsi_softc.accept_cv); CFISCSI_SESSION_LOCK(cs); /* * Wake up the maintenance thread if we got scheduled for termination * somewhere between cfiscsi_session_new() and icl_conn_handoff_sock(). */ if (cs->cs_terminating) cfiscsi_session_terminate(cs); CFISCSI_SESSION_UNLOCK(cs); } #endif static void cfiscsi_online(void *arg) { struct cfiscsi_softc *softc; struct cfiscsi_target *ct; int online; ct = (struct cfiscsi_target *)arg; softc = ct->ct_softc; mtx_lock(&softc->lock); if (ct->ct_online) { mtx_unlock(&softc->lock); return; } ct->ct_online = 1; online = softc->online++; mtx_unlock(&softc->lock); if (online > 0) return; #ifdef ICL_KERNEL_PROXY if (softc->listener != NULL) icl_listen_free(softc->listener); softc->listener = icl_listen_new(cfiscsi_accept); #endif } static void cfiscsi_offline(void *arg) { struct cfiscsi_softc *softc; struct cfiscsi_target *ct; struct cfiscsi_session *cs; int error, online; ct = (struct cfiscsi_target *)arg; softc = ct->ct_softc; mtx_lock(&softc->lock); if (!ct->ct_online) { mtx_unlock(&softc->lock); return; } ct->ct_online = 0; online = --softc->online; do { TAILQ_FOREACH(cs, &softc->sessions, cs_next) { if (cs->cs_target == ct) cfiscsi_session_terminate(cs); } TAILQ_FOREACH(cs, &softc->sessions, cs_next) { if (cs->cs_target == ct) break; } if (cs != NULL) { error = cv_wait_sig(&softc->sessions_cv, &softc->lock); if (error != 0) { CFISCSI_SESSION_DEBUG(cs, "cv_wait failed with error %d\n", error); break; } } } while (cs != NULL && ct->ct_online == 0); mtx_unlock(&softc->lock); if (online > 0) return; #ifdef ICL_KERNEL_PROXY icl_listen_free(softc->listener); softc->listener = NULL; #endif } static int cfiscsi_info(void *arg, struct sbuf *sb) { struct cfiscsi_target *ct = (struct cfiscsi_target *)arg; int retval; retval = sbuf_printf(sb, "\t%d\n", ct->ct_state); return (retval); } static void cfiscsi_ioctl_handoff(struct ctl_iscsi *ci) { struct cfiscsi_softc *softc; struct cfiscsi_session *cs, *cs2; struct cfiscsi_target *ct; struct ctl_iscsi_handoff_params *cihp; int error; cihp = (struct ctl_iscsi_handoff_params *)&(ci->data); softc = &cfiscsi_softc; CFISCSI_DEBUG("new connection from %s (%s) to %s", cihp->initiator_name, cihp->initiator_addr, cihp->target_name); ct = cfiscsi_target_find(softc, cihp->target_name, cihp->portal_group_tag); if (ct == NULL) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: target not found", __func__); return; } #ifdef ICL_KERNEL_PROXY if (cihp->socket > 0 && cihp->connection_id > 0) { snprintf(ci->error_str, sizeof(ci->error_str), "both socket and connection_id set"); ci->status = CTL_ISCSI_ERROR; cfiscsi_target_release(ct); return; } if (cihp->socket == 0) { mtx_lock(&cfiscsi_softc.lock); TAILQ_FOREACH(cs, &cfiscsi_softc.sessions, cs_next) { if (cs->cs_id == cihp->connection_id) break; } if (cs == NULL) { mtx_unlock(&cfiscsi_softc.lock); snprintf(ci->error_str, sizeof(ci->error_str), "connection not found"); ci->status = CTL_ISCSI_ERROR; cfiscsi_target_release(ct); return; } mtx_unlock(&cfiscsi_softc.lock); } else { #endif cs = cfiscsi_session_new(softc, cihp->offload); if (cs == NULL) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: cfiscsi_session_new failed", __func__); cfiscsi_target_release(ct); return; } #ifdef ICL_KERNEL_PROXY } #endif /* * First PDU of Full Feature phase has the same CmdSN as the last * PDU from the Login Phase received from the initiator. Thus, * the -1 below. */ cs->cs_cmdsn = cihp->cmdsn; cs->cs_statsn = cihp->statsn; cs->cs_conn->ic_max_recv_data_segment_length = cihp->max_recv_data_segment_length; cs->cs_conn->ic_max_send_data_segment_length = cihp->max_send_data_segment_length; cs->cs_max_burst_length = cihp->max_burst_length; cs->cs_first_burst_length = cihp->first_burst_length; cs->cs_immediate_data = !!cihp->immediate_data; if (cihp->header_digest == CTL_ISCSI_DIGEST_CRC32C) cs->cs_conn->ic_header_crc32c = true; if (cihp->data_digest == CTL_ISCSI_DIGEST_CRC32C) cs->cs_conn->ic_data_crc32c = true; strlcpy(cs->cs_initiator_name, cihp->initiator_name, sizeof(cs->cs_initiator_name)); strlcpy(cs->cs_initiator_addr, cihp->initiator_addr, sizeof(cs->cs_initiator_addr)); strlcpy(cs->cs_initiator_alias, cihp->initiator_alias, sizeof(cs->cs_initiator_alias)); memcpy(cs->cs_initiator_isid, cihp->initiator_isid, sizeof(cs->cs_initiator_isid)); snprintf(cs->cs_initiator_id, sizeof(cs->cs_initiator_id), "%s,i,0x%02x%02x%02x%02x%02x%02x", cs->cs_initiator_name, cihp->initiator_isid[0], cihp->initiator_isid[1], cihp->initiator_isid[2], cihp->initiator_isid[3], cihp->initiator_isid[4], cihp->initiator_isid[5]); mtx_lock(&softc->lock); if (ct->ct_online == 0) { mtx_unlock(&softc->lock); CFISCSI_SESSION_LOCK(cs); cs->cs_handoff_in_progress = false; cfiscsi_session_terminate(cs); CFISCSI_SESSION_UNLOCK(cs); cfiscsi_target_release(ct); ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: port offline", __func__); return; } cs->cs_target = ct; mtx_unlock(&softc->lock); restart: if (!cs->cs_terminating) { mtx_lock(&softc->lock); TAILQ_FOREACH(cs2, &softc->sessions, cs_next) { if (cs2 != cs && cs2->cs_tasks_aborted == false && cs->cs_target == cs2->cs_target && strcmp(cs->cs_initiator_id, cs2->cs_initiator_id) == 0) { if (strcmp(cs->cs_initiator_addr, cs2->cs_initiator_addr) != 0) { CFISCSI_SESSION_WARN(cs2, "session reinstatement from " "different address %s", cs->cs_initiator_addr); } else { CFISCSI_SESSION_DEBUG(cs2, "session reinstatement"); } cfiscsi_session_terminate(cs2); mtx_unlock(&softc->lock); pause("cfiscsi_reinstate", 1); goto restart; } } mtx_unlock(&softc->lock); } /* * Register initiator with CTL. */ cfiscsi_session_register_initiator(cs); #ifdef ICL_KERNEL_PROXY if (cihp->socket > 0) { #endif error = icl_conn_handoff(cs->cs_conn, cihp->socket); if (error != 0) { CFISCSI_SESSION_LOCK(cs); cs->cs_handoff_in_progress = false; cfiscsi_session_terminate(cs); CFISCSI_SESSION_UNLOCK(cs); ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: icl_conn_handoff failed with error %d", __func__, error); return; } #ifdef ICL_KERNEL_PROXY } #endif #ifdef ICL_KERNEL_PROXY cs->cs_login_phase = false; /* * First PDU of the Full Feature phase has likely already arrived. * We have to pick it up and execute properly. */ if (cs->cs_login_pdu != NULL) { CFISCSI_SESSION_DEBUG(cs, "picking up first PDU"); cfiscsi_pdu_handle(cs->cs_login_pdu); cs->cs_login_pdu = NULL; } #endif CFISCSI_SESSION_LOCK(cs); cs->cs_handoff_in_progress = false; /* * Wake up the maintenance thread if we got scheduled for termination. */ if (cs->cs_terminating) cfiscsi_session_terminate(cs); CFISCSI_SESSION_UNLOCK(cs); ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_list(struct ctl_iscsi *ci) { struct ctl_iscsi_list_params *cilp; struct cfiscsi_session *cs; struct cfiscsi_softc *softc; struct sbuf *sb; int error; cilp = (struct ctl_iscsi_list_params *)&(ci->data); softc = &cfiscsi_softc; sb = sbuf_new(NULL, NULL, cilp->alloc_len, SBUF_FIXEDLEN); if (sb == NULL) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "Unable to allocate %d bytes for iSCSI session list", cilp->alloc_len); return; } sbuf_printf(sb, "\n"); mtx_lock(&softc->lock); TAILQ_FOREACH(cs, &softc->sessions, cs_next) { if (cs->cs_target == NULL) continue; error = sbuf_printf(sb, "" "%s" "%s" "%s" "%s" "%s" "%u" "%s" "%s" "%d" "%d" "%d" "%d" "%d" "%d" "%s" "\n", cs->cs_id, cs->cs_initiator_name, cs->cs_initiator_addr, cs->cs_initiator_alias, cs->cs_target->ct_name, cs->cs_target->ct_alias, cs->cs_target->ct_tag, cs->cs_conn->ic_header_crc32c ? "CRC32C" : "None", cs->cs_conn->ic_data_crc32c ? "CRC32C" : "None", cs->cs_conn->ic_max_recv_data_segment_length, cs->cs_conn->ic_max_send_data_segment_length, cs->cs_max_burst_length, cs->cs_first_burst_length, cs->cs_immediate_data, cs->cs_conn->ic_iser, cs->cs_conn->ic_offload); if (error != 0) break; } mtx_unlock(&softc->lock); error = sbuf_printf(sb, "\n"); if (error != 0) { sbuf_delete(sb); ci->status = CTL_ISCSI_LIST_NEED_MORE_SPACE; snprintf(ci->error_str, sizeof(ci->error_str), "Out of space, %d bytes is too small", cilp->alloc_len); return; } sbuf_finish(sb); error = copyout(sbuf_data(sb), cilp->conn_xml, sbuf_len(sb) + 1); if (error != 0) { sbuf_delete(sb); snprintf(ci->error_str, sizeof(ci->error_str), "copyout failed with error %d", error); ci->status = CTL_ISCSI_ERROR; return; } cilp->fill_len = sbuf_len(sb) + 1; ci->status = CTL_ISCSI_OK; sbuf_delete(sb); } static void cfiscsi_ioctl_logout(struct ctl_iscsi *ci) { struct icl_pdu *response; struct iscsi_bhs_asynchronous_message *bhsam; struct ctl_iscsi_logout_params *cilp; struct cfiscsi_session *cs; struct cfiscsi_softc *softc; int found = 0; cilp = (struct ctl_iscsi_logout_params *)&(ci->data); softc = &cfiscsi_softc; mtx_lock(&softc->lock); TAILQ_FOREACH(cs, &softc->sessions, cs_next) { if (cilp->all == 0 && cs->cs_id != cilp->connection_id && strcmp(cs->cs_initiator_name, cilp->initiator_name) != 0 && strcmp(cs->cs_initiator_addr, cilp->initiator_addr) != 0) continue; response = icl_pdu_new(cs->cs_conn, M_NOWAIT); if (response == NULL) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "Unable to allocate memory"); mtx_unlock(&softc->lock); return; } bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs; bhsam->bhsam_opcode = ISCSI_BHS_OPCODE_ASYNC_MESSAGE; bhsam->bhsam_flags = 0x80; bhsam->bhsam_async_event = BHSAM_EVENT_TARGET_REQUESTS_LOGOUT; bhsam->bhsam_parameter3 = htons(10); cfiscsi_pdu_queue(response); found++; } mtx_unlock(&softc->lock); if (found == 0) { ci->status = CTL_ISCSI_SESSION_NOT_FOUND; snprintf(ci->error_str, sizeof(ci->error_str), "No matching connections found"); return; } ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_terminate(struct ctl_iscsi *ci) { struct icl_pdu *response; struct iscsi_bhs_asynchronous_message *bhsam; struct ctl_iscsi_terminate_params *citp; struct cfiscsi_session *cs; struct cfiscsi_softc *softc; int found = 0; citp = (struct ctl_iscsi_terminate_params *)&(ci->data); softc = &cfiscsi_softc; mtx_lock(&softc->lock); TAILQ_FOREACH(cs, &softc->sessions, cs_next) { if (citp->all == 0 && cs->cs_id != citp->connection_id && strcmp(cs->cs_initiator_name, citp->initiator_name) != 0 && strcmp(cs->cs_initiator_addr, citp->initiator_addr) != 0) continue; response = icl_pdu_new(cs->cs_conn, M_NOWAIT); if (response == NULL) { /* * Oh well. Just terminate the connection. */ } else { bhsam = (struct iscsi_bhs_asynchronous_message *) response->ip_bhs; bhsam->bhsam_opcode = ISCSI_BHS_OPCODE_ASYNC_MESSAGE; bhsam->bhsam_flags = 0x80; bhsam->bhsam_0xffffffff = 0xffffffff; bhsam->bhsam_async_event = BHSAM_EVENT_TARGET_TERMINATES_SESSION; cfiscsi_pdu_queue(response); } cfiscsi_session_terminate(cs); found++; } mtx_unlock(&softc->lock); if (found == 0) { ci->status = CTL_ISCSI_SESSION_NOT_FOUND; snprintf(ci->error_str, sizeof(ci->error_str), "No matching connections found"); return; } ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_limits(struct ctl_iscsi *ci) { struct ctl_iscsi_limits_params *cilp; struct icl_drv_limits idl; int error; cilp = (struct ctl_iscsi_limits_params *)&(ci->data); error = icl_limits(cilp->offload, false, &idl); if (error != 0) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: icl_limits failed with error %d", __func__, error); return; } cilp->max_recv_data_segment_length = idl.idl_max_recv_data_segment_length; cilp->max_send_data_segment_length = idl.idl_max_send_data_segment_length; cilp->max_burst_length = idl.idl_max_burst_length; cilp->first_burst_length = idl.idl_first_burst_length; ci->status = CTL_ISCSI_OK; } #ifdef ICL_KERNEL_PROXY static void cfiscsi_ioctl_listen(struct ctl_iscsi *ci) { struct ctl_iscsi_listen_params *cilp; struct sockaddr *sa; int error; cilp = (struct ctl_iscsi_listen_params *)&(ci->data); if (cfiscsi_softc.listener == NULL) { CFISCSI_DEBUG("no listener"); snprintf(ci->error_str, sizeof(ci->error_str), "no listener"); ci->status = CTL_ISCSI_ERROR; return; } error = getsockaddr(&sa, (void *)cilp->addr, cilp->addrlen); if (error != 0) { CFISCSI_DEBUG("getsockaddr, error %d", error); snprintf(ci->error_str, sizeof(ci->error_str), "getsockaddr failed"); ci->status = CTL_ISCSI_ERROR; return; } error = icl_listen_add(cfiscsi_softc.listener, cilp->iser, cilp->domain, cilp->socktype, cilp->protocol, sa, cilp->portal_id); if (error != 0) { free(sa, M_SONAME); CFISCSI_DEBUG("icl_listen_add, error %d", error); snprintf(ci->error_str, sizeof(ci->error_str), "icl_listen_add failed, error %d", error); ci->status = CTL_ISCSI_ERROR; return; } ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_accept(struct ctl_iscsi *ci) { struct ctl_iscsi_accept_params *ciap; struct cfiscsi_session *cs; int error; ciap = (struct ctl_iscsi_accept_params *)&(ci->data); mtx_lock(&cfiscsi_softc.lock); for (;;) { TAILQ_FOREACH(cs, &cfiscsi_softc.sessions, cs_next) { if (cs->cs_waiting_for_ctld) break; } if (cs != NULL) break; error = cv_wait_sig(&cfiscsi_softc.accept_cv, &cfiscsi_softc.lock); if (error != 0) { mtx_unlock(&cfiscsi_softc.lock); snprintf(ci->error_str, sizeof(ci->error_str), "interrupted"); ci->status = CTL_ISCSI_ERROR; return; } } mtx_unlock(&cfiscsi_softc.lock); cs->cs_waiting_for_ctld = false; cs->cs_login_phase = true; ciap->connection_id = cs->cs_id; ciap->portal_id = cs->cs_portal_id; ciap->initiator_addrlen = cs->cs_initiator_sa->sa_len; error = copyout(cs->cs_initiator_sa, ciap->initiator_addr, cs->cs_initiator_sa->sa_len); if (error != 0) { snprintf(ci->error_str, sizeof(ci->error_str), "copyout failed with error %d", error); ci->status = CTL_ISCSI_ERROR; return; } ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_send(struct ctl_iscsi *ci) { struct ctl_iscsi_send_params *cisp; struct cfiscsi_session *cs; struct icl_pdu *ip; size_t datalen; void *data; int error; cisp = (struct ctl_iscsi_send_params *)&(ci->data); mtx_lock(&cfiscsi_softc.lock); TAILQ_FOREACH(cs, &cfiscsi_softc.sessions, cs_next) { if (cs->cs_id == cisp->connection_id) break; } if (cs == NULL) { mtx_unlock(&cfiscsi_softc.lock); snprintf(ci->error_str, sizeof(ci->error_str), "connection not found"); ci->status = CTL_ISCSI_ERROR; return; } mtx_unlock(&cfiscsi_softc.lock); #if 0 if (cs->cs_login_phase == false) return (EBUSY); #endif if (cs->cs_terminating) { snprintf(ci->error_str, sizeof(ci->error_str), "connection is terminating"); ci->status = CTL_ISCSI_ERROR; return; } datalen = cisp->data_segment_len; /* * XXX */ //if (datalen > CFISCSI_MAX_DATA_SEGMENT_LENGTH) { if (datalen > 65535) { snprintf(ci->error_str, sizeof(ci->error_str), "data segment too big"); ci->status = CTL_ISCSI_ERROR; return; } if (datalen > 0) { data = malloc(datalen, M_CFISCSI, M_WAITOK); error = copyin(cisp->data_segment, data, datalen); if (error != 0) { free(data, M_CFISCSI); snprintf(ci->error_str, sizeof(ci->error_str), "copyin error %d", error); ci->status = CTL_ISCSI_ERROR; return; } } ip = icl_pdu_new(cs->cs_conn, M_WAITOK); memcpy(ip->ip_bhs, cisp->bhs, sizeof(*ip->ip_bhs)); if (datalen > 0) { icl_pdu_append_data(ip, data, datalen, M_WAITOK); free(data, M_CFISCSI); } CFISCSI_SESSION_LOCK(cs); icl_pdu_queue(ip); CFISCSI_SESSION_UNLOCK(cs); ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_receive(struct ctl_iscsi *ci) { struct ctl_iscsi_receive_params *cirp; struct cfiscsi_session *cs; struct icl_pdu *ip; void *data; int error; cirp = (struct ctl_iscsi_receive_params *)&(ci->data); mtx_lock(&cfiscsi_softc.lock); TAILQ_FOREACH(cs, &cfiscsi_softc.sessions, cs_next) { if (cs->cs_id == cirp->connection_id) break; } if (cs == NULL) { mtx_unlock(&cfiscsi_softc.lock); snprintf(ci->error_str, sizeof(ci->error_str), "connection not found"); ci->status = CTL_ISCSI_ERROR; return; } mtx_unlock(&cfiscsi_softc.lock); #if 0 if (is->is_login_phase == false) return (EBUSY); #endif CFISCSI_SESSION_LOCK(cs); while (cs->cs_login_pdu == NULL && cs->cs_terminating == false) { error = cv_wait_sig(&cs->cs_login_cv, &cs->cs_lock); if (error != 0) { CFISCSI_SESSION_UNLOCK(cs); snprintf(ci->error_str, sizeof(ci->error_str), "interrupted by signal"); ci->status = CTL_ISCSI_ERROR; return; } } if (cs->cs_terminating) { CFISCSI_SESSION_UNLOCK(cs); snprintf(ci->error_str, sizeof(ci->error_str), "connection terminating"); ci->status = CTL_ISCSI_ERROR; return; } ip = cs->cs_login_pdu; cs->cs_login_pdu = NULL; CFISCSI_SESSION_UNLOCK(cs); if (ip->ip_data_len > cirp->data_segment_len) { icl_pdu_free(ip); snprintf(ci->error_str, sizeof(ci->error_str), "data segment too big"); ci->status = CTL_ISCSI_ERROR; return; } copyout(ip->ip_bhs, cirp->bhs, sizeof(*ip->ip_bhs)); if (ip->ip_data_len > 0) { data = malloc(ip->ip_data_len, M_CFISCSI, M_WAITOK); icl_pdu_get_data(ip, 0, data, ip->ip_data_len); copyout(data, cirp->data_segment, ip->ip_data_len); free(data, M_CFISCSI); } icl_pdu_free(ip); ci->status = CTL_ISCSI_OK; } #endif /* !ICL_KERNEL_PROXY */ static void cfiscsi_ioctl_port_create(struct ctl_req *req) { struct cfiscsi_target *ct; struct ctl_port *port; const char *target, *alias, *val; struct scsi_vpd_id_descriptor *desc; int retval, len, idlen; uint16_t tag; target = dnvlist_get_string(req->args_nvl, "cfiscsi_target", NULL); alias = dnvlist_get_string(req->args_nvl, "cfiscsi_target_alias", NULL); val = dnvlist_get_string(req->args_nvl, "cfiscsi_portal_group_tag", NULL); if (target == NULL || val == NULL) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "Missing required argument"); return; } tag = strtoul(val, NULL, 0); ct = cfiscsi_target_find_or_create(&cfiscsi_softc, target, alias, tag); if (ct == NULL) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "failed to create target \"%s\"", target); return; } if (ct->ct_state == CFISCSI_TARGET_STATE_ACTIVE) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "target \"%s\" for portal group tag %u already exists", target, tag); cfiscsi_target_release(ct); return; } port = &ct->ct_port; // WAT if (ct->ct_state == CFISCSI_TARGET_STATE_DYING) goto done; port->frontend = &cfiscsi_frontend; port->port_type = CTL_PORT_ISCSI; /* XXX KDM what should the real number be here? */ port->num_requested_ctl_io = 4096; port->port_name = "iscsi"; port->physical_port = (int)tag; port->virtual_port = ct->ct_target_id; port->port_online = cfiscsi_online; port->port_offline = cfiscsi_offline; port->port_info = cfiscsi_info; port->onoff_arg = ct; port->fe_datamove = cfiscsi_datamove; port->fe_done = cfiscsi_done; port->targ_port = -1; port->options = nvlist_clone(req->args_nvl); /* Generate Port ID. */ idlen = strlen(target) + strlen(",t,0x0001") + 1; idlen = roundup2(idlen, 4); len = sizeof(struct scsi_vpd_device_id) + idlen; port->port_devid = malloc(sizeof(struct ctl_devid) + len, M_CTL, M_WAITOK | M_ZERO); port->port_devid->len = len; desc = (struct scsi_vpd_id_descriptor *)port->port_devid->data; desc->proto_codeset = (SCSI_PROTO_ISCSI << 4) | SVPD_ID_CODESET_UTF8; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_SCSI_NAME; desc->length = idlen; snprintf(desc->identifier, idlen, "%s,t,0x%4.4x", target, tag); /* Generate Target ID. */ idlen = strlen(target) + 1; idlen = roundup2(idlen, 4); len = sizeof(struct scsi_vpd_device_id) + idlen; port->target_devid = malloc(sizeof(struct ctl_devid) + len, M_CTL, M_WAITOK | M_ZERO); port->target_devid->len = len; desc = (struct scsi_vpd_id_descriptor *)port->target_devid->data; desc->proto_codeset = (SCSI_PROTO_ISCSI << 4) | SVPD_ID_CODESET_UTF8; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_TARGET | SVPD_ID_TYPE_SCSI_NAME; desc->length = idlen; strlcpy(desc->identifier, target, idlen); retval = ctl_port_register(port); if (retval != 0) { free(port->port_devid, M_CFISCSI); free(port->target_devid, M_CFISCSI); cfiscsi_target_release(ct); req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "ctl_port_register() failed with error %d", retval); return; } done: ct->ct_state = CFISCSI_TARGET_STATE_ACTIVE; req->status = CTL_LUN_OK; req->result_nvl = nvlist_create(0); nvlist_add_number(req->result_nvl, "port_id", port->targ_port); } static void cfiscsi_ioctl_port_remove(struct ctl_req *req) { struct cfiscsi_target *ct; const char *target, *val; uint16_t tag; target = dnvlist_get_string(req->args_nvl, "cfiscsi_target", NULL); val = dnvlist_get_string(req->args_nvl, "cfiscsi_portal_group_tag", NULL); if (target == NULL || val == NULL) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "Missing required argument"); return; } tag = strtoul(val, NULL, 0); ct = cfiscsi_target_find(&cfiscsi_softc, target, tag); if (ct == NULL) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "can't find target \"%s\"", target); return; } ct->ct_state = CFISCSI_TARGET_STATE_DYING; ctl_port_offline(&ct->ct_port); cfiscsi_target_release(ct); cfiscsi_target_release(ct); req->status = CTL_LUN_OK; } static int cfiscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td) { struct ctl_iscsi *ci; struct ctl_req *req; if (cmd == CTL_PORT_REQ) { req = (struct ctl_req *)addr; switch (req->reqtype) { case CTL_REQ_CREATE: cfiscsi_ioctl_port_create(req); break; case CTL_REQ_REMOVE: cfiscsi_ioctl_port_remove(req); break; default: req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "Unsupported request type %d", req->reqtype); } return (0); } if (cmd != CTL_ISCSI) return (ENOTTY); ci = (struct ctl_iscsi *)addr; switch (ci->type) { case CTL_ISCSI_HANDOFF: cfiscsi_ioctl_handoff(ci); break; case CTL_ISCSI_LIST: cfiscsi_ioctl_list(ci); break; case CTL_ISCSI_LOGOUT: cfiscsi_ioctl_logout(ci); break; case CTL_ISCSI_TERMINATE: cfiscsi_ioctl_terminate(ci); break; case CTL_ISCSI_LIMITS: cfiscsi_ioctl_limits(ci); break; #ifdef ICL_KERNEL_PROXY case CTL_ISCSI_LISTEN: cfiscsi_ioctl_listen(ci); break; case CTL_ISCSI_ACCEPT: cfiscsi_ioctl_accept(ci); break; case CTL_ISCSI_SEND: cfiscsi_ioctl_send(ci); break; case CTL_ISCSI_RECEIVE: cfiscsi_ioctl_receive(ci); break; #else case CTL_ISCSI_LISTEN: case CTL_ISCSI_ACCEPT: case CTL_ISCSI_SEND: case CTL_ISCSI_RECEIVE: ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: CTL compiled without ICL_KERNEL_PROXY", __func__); break; #endif /* !ICL_KERNEL_PROXY */ default: ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: invalid iSCSI request type %d", __func__, ci->type); break; } return (0); } static void cfiscsi_target_hold(struct cfiscsi_target *ct) { refcount_acquire(&ct->ct_refcount); } static void cfiscsi_target_release(struct cfiscsi_target *ct) { struct cfiscsi_softc *softc; softc = ct->ct_softc; mtx_lock(&softc->lock); if (refcount_release(&ct->ct_refcount)) { TAILQ_REMOVE(&softc->targets, ct, ct_next); mtx_unlock(&softc->lock); if (ct->ct_state != CFISCSI_TARGET_STATE_INVALID) { ct->ct_state = CFISCSI_TARGET_STATE_INVALID; if (ctl_port_deregister(&ct->ct_port) != 0) printf("%s: ctl_port_deregister() failed\n", __func__); } free(ct, M_CFISCSI); return; } mtx_unlock(&softc->lock); } static struct cfiscsi_target * cfiscsi_target_find(struct cfiscsi_softc *softc, const char *name, uint16_t tag) { struct cfiscsi_target *ct; mtx_lock(&softc->lock); TAILQ_FOREACH(ct, &softc->targets, ct_next) { if (ct->ct_tag != tag || strcmp(name, ct->ct_name) != 0 || ct->ct_state != CFISCSI_TARGET_STATE_ACTIVE) continue; cfiscsi_target_hold(ct); mtx_unlock(&softc->lock); return (ct); } mtx_unlock(&softc->lock); return (NULL); } static struct cfiscsi_target * cfiscsi_target_find_or_create(struct cfiscsi_softc *softc, const char *name, const char *alias, uint16_t tag) { struct cfiscsi_target *ct, *newct; if (name[0] == '\0' || strlen(name) >= CTL_ISCSI_NAME_LEN) return (NULL); newct = malloc(sizeof(*newct), M_CFISCSI, M_WAITOK | M_ZERO); mtx_lock(&softc->lock); TAILQ_FOREACH(ct, &softc->targets, ct_next) { if (ct->ct_tag != tag || strcmp(name, ct->ct_name) != 0 || ct->ct_state == CFISCSI_TARGET_STATE_INVALID) continue; cfiscsi_target_hold(ct); mtx_unlock(&softc->lock); free(newct, M_CFISCSI); return (ct); } strlcpy(newct->ct_name, name, sizeof(newct->ct_name)); if (alias != NULL) strlcpy(newct->ct_alias, alias, sizeof(newct->ct_alias)); newct->ct_tag = tag; refcount_init(&newct->ct_refcount, 1); newct->ct_softc = softc; if (TAILQ_EMPTY(&softc->targets)) softc->last_target_id = 0; newct->ct_target_id = ++softc->last_target_id; TAILQ_INSERT_TAIL(&softc->targets, newct, ct_next); mtx_unlock(&softc->lock); return (newct); } static void cfiscsi_pdu_done(struct icl_pdu *ip, int error) { if (error != 0) ; // XXX: Do something on error? ((ctl_ref)ip->ip_prv0)(ip->ip_prv1, -1); } static void cfiscsi_datamove_in(union ctl_io *io) { struct cfiscsi_session *cs; struct icl_pdu *request, *response; const struct iscsi_bhs_scsi_command *bhssc; struct iscsi_bhs_data_in *bhsdi; struct ctl_sg_entry ctl_sg_entry, *ctl_sglist; size_t len, expected_len, sg_len, buffer_offset; size_t max_send_data_segment_length; const char *sg_addr; icl_pdu_cb cb; int ctl_sg_count, error, i; request = PRIV_REQUEST(io); cs = PDU_SESSION(request); bhssc = (const struct iscsi_bhs_scsi_command *)request->ip_bhs; KASSERT((bhssc->bhssc_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND, ("bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_COMMAND")); if (io->scsiio.kern_sg_entries > 0) { ctl_sglist = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; ctl_sg_count = io->scsiio.kern_sg_entries; } else { ctl_sglist = &ctl_sg_entry; ctl_sglist->addr = io->scsiio.kern_data_ptr; ctl_sglist->len = io->scsiio.kern_data_len; ctl_sg_count = 1; } /* * This is the offset within the current SCSI command; for the first * call to cfiscsi_datamove() it will be 0, and for subsequent ones * it will be the sum of lengths of previous ones. */ buffer_offset = io->scsiio.kern_rel_offset; /* * This is the transfer length expected by the initiator. It can be * different from the amount of data from the SCSI point of view. */ expected_len = ntohl(bhssc->bhssc_expected_data_transfer_length); /* * If the transfer is outside of expected length -- we are done. */ if (buffer_offset >= expected_len) { #if 0 CFISCSI_SESSION_DEBUG(cs, "buffer_offset = %zd, " "already sent the expected len", buffer_offset); #endif ctl_datamove_done(io, true); return; } if (io->scsiio.kern_data_ref != NULL) cb = cfiscsi_pdu_done; else cb = NULL; i = 0; sg_addr = NULL; sg_len = 0; response = NULL; bhsdi = NULL; if (cs->cs_conn->ic_hw_isomax != 0) max_send_data_segment_length = cs->cs_conn->ic_hw_isomax; else max_send_data_segment_length = cs->cs_conn->ic_max_send_data_segment_length; for (;;) { if (response == NULL) { response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to " "allocate memory; dropping connection"); ctl_set_busy(&io->scsiio); ctl_datamove_done(io, true); cfiscsi_session_terminate(cs); return; } bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs; bhsdi->bhsdi_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_IN; bhsdi->bhsdi_initiator_task_tag = bhssc->bhssc_initiator_task_tag; bhsdi->bhsdi_target_transfer_tag = 0xffffffff; bhsdi->bhsdi_datasn = htonl(PRIV_EXPDATASN(io)); bhsdi->bhsdi_buffer_offset = htonl(buffer_offset); } KASSERT(i < ctl_sg_count, ("i >= ctl_sg_count")); if (sg_len == 0) { sg_addr = ctl_sglist[i].addr; sg_len = ctl_sglist[i].len; KASSERT(sg_len > 0, ("sg_len <= 0")); } len = sg_len; /* * Truncate to maximum data segment length. */ KASSERT(response->ip_data_len < max_send_data_segment_length, ("ip_data_len %zd >= max_send_data_segment_length %zd", response->ip_data_len, max_send_data_segment_length)); if (response->ip_data_len + len > max_send_data_segment_length) { len = max_send_data_segment_length - response->ip_data_len; KASSERT(len <= sg_len, ("len %zd > sg_len %zd", len, sg_len)); } /* * Truncate to expected data transfer length. */ KASSERT(buffer_offset + response->ip_data_len < expected_len, ("buffer_offset %zd + ip_data_len %zd >= expected_len %zd", buffer_offset, response->ip_data_len, expected_len)); if (buffer_offset + response->ip_data_len + len > expected_len) { CFISCSI_SESSION_DEBUG(cs, "truncating from %zd " "to expected data transfer length %zd", buffer_offset + response->ip_data_len + len, expected_len); len = expected_len - (buffer_offset + response->ip_data_len); KASSERT(len <= sg_len, ("len %zd > sg_len %zd", len, sg_len)); } error = icl_pdu_append_data(response, sg_addr, len, M_NOWAIT | (cb ? ICL_NOCOPY : 0)); if (error != 0) { CFISCSI_SESSION_WARN(cs, "failed to " "allocate memory; dropping connection"); icl_pdu_free(response); ctl_set_busy(&io->scsiio); ctl_datamove_done(io, true); cfiscsi_session_terminate(cs); return; } sg_addr += len; sg_len -= len; io->scsiio.kern_data_resid -= len; KASSERT(buffer_offset + response->ip_data_len <= expected_len, ("buffer_offset %zd + ip_data_len %zd > expected_len %zd", buffer_offset, response->ip_data_len, expected_len)); if (buffer_offset + response->ip_data_len == expected_len) { /* * Already have the amount of data the initiator wanted. */ break; } if (sg_len == 0) { /* * End of scatter-gather segment; * proceed to the next one... */ if (i == ctl_sg_count - 1) { /* * ... unless this was the last one. */ break; } i++; } if (response->ip_data_len == max_send_data_segment_length) { /* * Can't stuff more data into the current PDU; * queue it. Note that's not enough to check * for kern_data_resid == 0 instead; there * may be several Data-In PDUs for the final * call to cfiscsi_datamove(), and we want * to set the F flag only on the last of them. */ buffer_offset += response->ip_data_len; if (buffer_offset == io->scsiio.kern_total_len || buffer_offset == expected_len) { buffer_offset -= response->ip_data_len; break; } PRIV_EXPDATASN(io) += howmany(response->ip_data_len, cs->cs_conn->ic_max_send_data_segment_length); if (cb != NULL) { response->ip_prv0 = io->scsiio.kern_data_ref; response->ip_prv1 = io->scsiio.kern_data_arg; io->scsiio.kern_data_ref(io->scsiio.kern_data_arg, 1); } cfiscsi_pdu_queue_cb(response, cb); response = NULL; bhsdi = NULL; } } if (response != NULL) { buffer_offset += response->ip_data_len; if (buffer_offset == io->scsiio.kern_total_len || buffer_offset == expected_len) { bhsdi->bhsdi_flags |= BHSDI_FLAGS_F; if (io->io_hdr.status == CTL_SUCCESS) { bhsdi->bhsdi_flags |= BHSDI_FLAGS_S; if (io->scsiio.kern_total_len < ntohl(bhssc->bhssc_expected_data_transfer_length)) { bhsdi->bhsdi_flags |= BHSSR_FLAGS_RESIDUAL_UNDERFLOW; bhsdi->bhsdi_residual_count = htonl(ntohl(bhssc->bhssc_expected_data_transfer_length) - io->scsiio.kern_total_len); } else if (io->scsiio.kern_total_len > ntohl(bhssc->bhssc_expected_data_transfer_length)) { bhsdi->bhsdi_flags |= BHSSR_FLAGS_RESIDUAL_OVERFLOW; bhsdi->bhsdi_residual_count = htonl(io->scsiio.kern_total_len - ntohl(bhssc->bhssc_expected_data_transfer_length)); } bhsdi->bhsdi_status = io->scsiio.scsi_status; io->io_hdr.flags |= CTL_FLAG_STATUS_SENT; } } KASSERT(response->ip_data_len > 0, ("sending empty Data-In")); PRIV_EXPDATASN(io) += howmany(response->ip_data_len, cs->cs_conn->ic_max_send_data_segment_length); if (cb != NULL) { response->ip_prv0 = io->scsiio.kern_data_ref; response->ip_prv1 = io->scsiio.kern_data_arg; io->scsiio.kern_data_ref(io->scsiio.kern_data_arg, 1); } cfiscsi_pdu_queue_cb(response, cb); } ctl_datamove_done(io, true); } static void cfiscsi_datamove_out(union ctl_io *io) { struct cfiscsi_session *cs; struct icl_pdu *request, *response; const struct iscsi_bhs_scsi_command *bhssc; struct iscsi_bhs_r2t *bhsr2t; struct cfiscsi_data_wait *cdw; struct ctl_sg_entry ctl_sg_entry, *ctl_sglist; uint32_t expected_len, datamove_len, r2t_off, r2t_len; uint32_t target_transfer_tag; bool done; request = PRIV_REQUEST(io); cs = PDU_SESSION(request); bhssc = (const struct iscsi_bhs_scsi_command *)request->ip_bhs; KASSERT((bhssc->bhssc_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND, ("bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_COMMAND")); /* * Complete write underflow. Not a single byte to read. Return. */ expected_len = ntohl(bhssc->bhssc_expected_data_transfer_length); if (io->scsiio.kern_rel_offset >= expected_len) { ctl_datamove_done(io, true); return; } datamove_len = MIN(io->scsiio.kern_data_len, expected_len - io->scsiio.kern_rel_offset); target_transfer_tag = atomic_fetchadd_32(&cs->cs_target_transfer_tag, 1); if (target_transfer_tag == 0xffffffff) { target_transfer_tag = atomic_fetchadd_32(&cs->cs_target_transfer_tag, 1); } cdw = cfiscsi_data_wait_new(cs, io, bhssc->bhssc_initiator_task_tag, &target_transfer_tag); if (cdw == NULL) { CFISCSI_SESSION_WARN(cs, "failed to " "allocate memory; dropping connection"); ctl_set_busy(&io->scsiio); ctl_datamove_done(io, true); cfiscsi_session_terminate(cs); return; } #if 0 CFISCSI_SESSION_DEBUG(cs, "expecting Data-Out with initiator " "task tag 0x%x, target transfer tag 0x%x", bhssc->bhssc_initiator_task_tag, target_transfer_tag); #endif cdw->cdw_ctl_io = io; cdw->cdw_target_transfer_tag = target_transfer_tag; cdw->cdw_initiator_task_tag = bhssc->bhssc_initiator_task_tag; cdw->cdw_r2t_end = datamove_len; cdw->cdw_datasn = 0; /* Set initial data pointer for the CDW respecting ext_data_filled. */ if (io->scsiio.kern_sg_entries > 0) { ctl_sglist = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; } else { ctl_sglist = &ctl_sg_entry; ctl_sglist->addr = io->scsiio.kern_data_ptr; ctl_sglist->len = datamove_len; } cdw->cdw_sg_index = 0; cdw->cdw_sg_addr = ctl_sglist[cdw->cdw_sg_index].addr; cdw->cdw_sg_len = ctl_sglist[cdw->cdw_sg_index].len; r2t_off = io->scsiio.ext_data_filled; while (r2t_off > 0) { if (r2t_off >= cdw->cdw_sg_len) { r2t_off -= cdw->cdw_sg_len; cdw->cdw_sg_index++; cdw->cdw_sg_addr = ctl_sglist[cdw->cdw_sg_index].addr; cdw->cdw_sg_len = ctl_sglist[cdw->cdw_sg_index].len; continue; } cdw->cdw_sg_addr += r2t_off; cdw->cdw_sg_len -= r2t_off; r2t_off = 0; } if (cs->cs_immediate_data && io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled < icl_pdu_data_segment_length(request)) { done = cfiscsi_handle_data_segment(request, cdw); if (done) { cfiscsi_data_wait_free(cs, cdw); ctl_datamove_done(io, true); return; } } r2t_off = io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled; r2t_len = MIN(datamove_len - io->scsiio.ext_data_filled, cs->cs_max_burst_length); cdw->cdw_r2t_end = io->scsiio.ext_data_filled + r2t_len; CFISCSI_SESSION_LOCK(cs); TAILQ_INSERT_TAIL(&cs->cs_waiting_for_data_out, cdw, cdw_next); CFISCSI_SESSION_UNLOCK(cs); /* * XXX: We should limit the number of outstanding R2T PDUs * per task to MaxOutstandingR2T. */ response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to " "allocate memory; dropping connection"); ctl_set_busy(&io->scsiio); ctl_datamove_done(io, true); cfiscsi_session_terminate(cs); return; } io->io_hdr.flags |= CTL_FLAG_DMA_INPROG; bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs; bhsr2t->bhsr2t_opcode = ISCSI_BHS_OPCODE_R2T; bhsr2t->bhsr2t_flags = 0x80; bhsr2t->bhsr2t_lun = bhssc->bhssc_lun; bhsr2t->bhsr2t_initiator_task_tag = bhssc->bhssc_initiator_task_tag; bhsr2t->bhsr2t_target_transfer_tag = target_transfer_tag; /* * XXX: Here we assume that cfiscsi_datamove() won't ever * be running concurrently on several CPUs for a given * command. */ bhsr2t->bhsr2t_r2tsn = htonl(PRIV_R2TSN(io)++); /* * This is the offset within the current SCSI command; * i.e. for the first call of datamove(), it will be 0, * and for subsequent ones it will be the sum of lengths * of previous ones. * * The ext_data_filled is to account for unsolicited * (immediate) data that might have already arrived. */ bhsr2t->bhsr2t_buffer_offset = htonl(r2t_off); /* * This is the total length (sum of S/G lengths) this call * to cfiscsi_datamove() is supposed to handle, limited by * MaxBurstLength. */ bhsr2t->bhsr2t_desired_data_transfer_length = htonl(r2t_len); cfiscsi_pdu_queue(response); } static void cfiscsi_datamove(union ctl_io *io) { if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) cfiscsi_datamove_in(io); else { /* We hadn't received anything during this datamove yet. */ io->scsiio.ext_data_filled = 0; cfiscsi_datamove_out(io); } } static void cfiscsi_scsi_command_done(union ctl_io *io) { struct icl_pdu *request, *response; struct iscsi_bhs_scsi_command *bhssc; struct iscsi_bhs_scsi_response *bhssr; #ifdef DIAGNOSTIC struct cfiscsi_data_wait *cdw; #endif struct cfiscsi_session *cs; uint16_t sense_length; request = PRIV_REQUEST(io); cs = PDU_SESSION(request); bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs; KASSERT((bhssc->bhssc_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND, ("replying to wrong opcode 0x%x", bhssc->bhssc_opcode)); //CFISCSI_SESSION_DEBUG(cs, "initiator task tag 0x%x", // bhssc->bhssc_initiator_task_tag); #ifdef DIAGNOSTIC CFISCSI_SESSION_LOCK(cs); TAILQ_FOREACH(cdw, &cs->cs_waiting_for_data_out, cdw_next) KASSERT(bhssc->bhssc_initiator_task_tag != cdw->cdw_initiator_task_tag, ("dangling cdw")); CFISCSI_SESSION_UNLOCK(cs); #endif /* * Do not return status for aborted commands. * There are exceptions, but none supported by CTL yet. */ if (((io->io_hdr.flags & CTL_FLAG_ABORT) && (io->io_hdr.flags & CTL_FLAG_ABORT_STATUS) == 0) || (io->io_hdr.flags & CTL_FLAG_STATUS_SENT)) { ctl_free_io(io); icl_pdu_free(request); return; } response = cfiscsi_pdu_new_response(request, M_WAITOK); bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs; bhssr->bhssr_opcode = ISCSI_BHS_OPCODE_SCSI_RESPONSE; bhssr->bhssr_flags = 0x80; /* * XXX: We don't deal with bidirectional under/overflows; * does anything actually support those? */ if (io->scsiio.kern_total_len < ntohl(bhssc->bhssc_expected_data_transfer_length)) { bhssr->bhssr_flags |= BHSSR_FLAGS_RESIDUAL_UNDERFLOW; bhssr->bhssr_residual_count = htonl(ntohl(bhssc->bhssc_expected_data_transfer_length) - io->scsiio.kern_total_len); //CFISCSI_SESSION_DEBUG(cs, "underflow; residual count %d", // ntohl(bhssr->bhssr_residual_count)); } else if (io->scsiio.kern_total_len > ntohl(bhssc->bhssc_expected_data_transfer_length)) { bhssr->bhssr_flags |= BHSSR_FLAGS_RESIDUAL_OVERFLOW; bhssr->bhssr_residual_count = htonl(io->scsiio.kern_total_len - ntohl(bhssc->bhssc_expected_data_transfer_length)); //CFISCSI_SESSION_DEBUG(cs, "overflow; residual count %d", // ntohl(bhssr->bhssr_residual_count)); } bhssr->bhssr_response = BHSSR_RESPONSE_COMMAND_COMPLETED; bhssr->bhssr_status = io->scsiio.scsi_status; bhssr->bhssr_initiator_task_tag = bhssc->bhssc_initiator_task_tag; bhssr->bhssr_expdatasn = htonl(PRIV_EXPDATASN(io)); if (io->scsiio.sense_len > 0) { #if 0 CFISCSI_SESSION_DEBUG(cs, "returning %d bytes of sense data", io->scsiio.sense_len); #endif sense_length = htons(io->scsiio.sense_len); icl_pdu_append_data(response, &sense_length, sizeof(sense_length), M_WAITOK); icl_pdu_append_data(response, &io->scsiio.sense_data, io->scsiio.sense_len, M_WAITOK); } ctl_free_io(io); icl_pdu_free(request); cfiscsi_pdu_queue(response); } static void cfiscsi_task_management_done(union ctl_io *io) { struct icl_pdu *request, *response; struct iscsi_bhs_task_management_request *bhstmr; struct iscsi_bhs_task_management_response *bhstmr2; struct cfiscsi_data_wait *cdw, *tmpcdw; struct cfiscsi_session *cs, *tcs; struct cfiscsi_softc *softc; union ctl_io *cdw_io; int cold_reset = 0; request = PRIV_REQUEST(io); cs = PDU_SESSION(request); bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs; KASSERT((bhstmr->bhstmr_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_TASK_REQUEST, ("replying to wrong opcode 0x%x", bhstmr->bhstmr_opcode)); #if 0 CFISCSI_SESSION_DEBUG(cs, "initiator task tag 0x%x; referenced task tag 0x%x", bhstmr->bhstmr_initiator_task_tag, bhstmr->bhstmr_referenced_task_tag); #endif if ((bhstmr->bhstmr_function & ~0x80) == BHSTMR_FUNCTION_ABORT_TASK) { /* * Make sure we no longer wait for Data-Out for this command. */ CFISCSI_SESSION_LOCK(cs); TAILQ_FOREACH_SAFE(cdw, &cs->cs_waiting_for_data_out, cdw_next, tmpcdw) { if (bhstmr->bhstmr_referenced_task_tag != cdw->cdw_initiator_task_tag) continue; #if 0 CFISCSI_SESSION_DEBUG(cs, "removing csw for initiator task " "tag 0x%x", bhstmr->bhstmr_initiator_task_tag); #endif TAILQ_REMOVE(&cs->cs_waiting_for_data_out, cdw, cdw_next); cdw_io = cdw->cdw_ctl_io; cdw_io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG; cdw_io->scsiio.io_hdr.port_status = 43; cfiscsi_data_wait_free(cs, cdw); ctl_datamove_done(cdw_io, false); } CFISCSI_SESSION_UNLOCK(cs); } if ((bhstmr->bhstmr_function & ~0x80) == BHSTMR_FUNCTION_TARGET_COLD_RESET && io->io_hdr.status == CTL_SUCCESS) cold_reset = 1; response = cfiscsi_pdu_new_response(request, M_WAITOK); bhstmr2 = (struct iscsi_bhs_task_management_response *) response->ip_bhs; bhstmr2->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_RESPONSE; bhstmr2->bhstmr_flags = 0x80; switch (io->taskio.task_status) { case CTL_TASK_FUNCTION_COMPLETE: bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_COMPLETE; break; case CTL_TASK_FUNCTION_SUCCEEDED: bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_SUCCEEDED; break; case CTL_TASK_LUN_DOES_NOT_EXIST: bhstmr2->bhstmr_response = BHSTMR_RESPONSE_LUN_DOES_NOT_EXIST; break; case CTL_TASK_FUNCTION_NOT_SUPPORTED: default: bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_NOT_SUPPORTED; break; } memcpy(bhstmr2->bhstmr_additional_reponse_information, io->taskio.task_resp, sizeof(io->taskio.task_resp)); bhstmr2->bhstmr_initiator_task_tag = bhstmr->bhstmr_initiator_task_tag; ctl_free_io(io); icl_pdu_free(request); cfiscsi_pdu_queue(response); if (cold_reset) { softc = cs->cs_target->ct_softc; mtx_lock(&softc->lock); TAILQ_FOREACH(tcs, &softc->sessions, cs_next) { if (tcs->cs_target == cs->cs_target) cfiscsi_session_terminate(tcs); } mtx_unlock(&softc->lock); } } static void cfiscsi_done(union ctl_io *io) { struct icl_pdu *request; struct cfiscsi_session *cs; KASSERT(((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE), ("invalid CTL status %#x", io->io_hdr.status)); if (io->io_hdr.io_type == CTL_IO_TASK && io->taskio.task_action == CTL_TASK_I_T_NEXUS_RESET) { /* * Implicit task termination has just completed; nothing to do. */ cs = PRIV_REQUEST(io); cs->cs_tasks_aborted = true; refcount_release(&cs->cs_outstanding_ctl_pdus); wakeup(__DEVOLATILE(void *, &cs->cs_outstanding_ctl_pdus)); ctl_free_io(io); return; } request = PRIV_REQUEST(io); cs = PDU_SESSION(request); switch (request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) { case ISCSI_BHS_OPCODE_SCSI_COMMAND: cfiscsi_scsi_command_done(io); break; case ISCSI_BHS_OPCODE_TASK_REQUEST: cfiscsi_task_management_done(io); break; default: panic("cfiscsi_done called with wrong opcode 0x%x", request->ip_bhs->bhs_opcode); } refcount_release(&cs->cs_outstanding_ctl_pdus); } diff --git a/sys/dev/cxgbe/cxgbei/cxgbei.c b/sys/dev/cxgbe/cxgbei/cxgbei.c index 8dc580d65770..6af7043460fb 100644 --- a/sys/dev/cxgbe/cxgbei/cxgbei.c +++ b/sys/dev/cxgbe/cxgbei/cxgbei.c @@ -1,1015 +1,1019 @@ /*- * Copyright (c) 2012 Chelsio Communications, Inc. * All rights reserved. * * Chelsio T5xx iSCSI driver * * Written by: Sreenivasa Honnur * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #ifdef TCP_OFFLOAD #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "common/common.h" #include "common/t4_msg.h" #include "common/t4_regs.h" /* for PCIE_MEM_ACCESS */ #include "tom/t4_tom.h" #include "cxgbei.h" static int worker_thread_count; static struct cxgbei_worker_thread_softc *cwt_softc; static struct proc *cxgbei_proc; static void read_pdu_limits(struct adapter *sc, uint32_t *max_tx_data_len, uint32_t *max_rx_data_len, struct ppod_region *pr) { uint32_t tx_len, rx_len, r, v; rx_len = t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE); tx_len = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE); r = t4_read_reg(sc, A_TP_PARA_REG2); rx_len = min(rx_len, G_MAXRXDATA(r)); tx_len = min(tx_len, G_MAXRXDATA(r)); r = t4_read_reg(sc, A_TP_PARA_REG7); v = min(G_PMMAXXFERLEN0(r), G_PMMAXXFERLEN1(r)); rx_len = min(rx_len, v); tx_len = min(tx_len, v); /* * AHS is not supported by the kernel so we'll not account for * it either in our PDU len -> data segment len conversions. */ rx_len -= ISCSI_BHS_SIZE + ISCSI_HEADER_DIGEST_SIZE + ISCSI_DATA_DIGEST_SIZE; tx_len -= ISCSI_BHS_SIZE + ISCSI_HEADER_DIGEST_SIZE + ISCSI_DATA_DIGEST_SIZE; /* * DDP can place only 4 pages for a single PDU. A single * request might use larger pages than the smallest page size, * but that cannot be guaranteed. Assume the smallest DDP * page size for this limit. */ rx_len = min(rx_len, 4 * (1U << pr->pr_page_shift[0])); if (chip_id(sc) == CHELSIO_T5) { rx_len = rounddown2(rx_len, 512); tx_len = rounddown2(tx_len, 512); } *max_tx_data_len = tx_len; *max_rx_data_len = rx_len; } /* * Initialize the software state of the iSCSI ULP driver. * * ENXIO means firmware didn't set up something that it was supposed to. */ static int cxgbei_init(struct adapter *sc, struct cxgbei_data *ci) { struct sysctl_oid *oid; struct sysctl_oid_list *children; struct ppod_region *pr; uint32_t r; int rc; MPASS(sc->vres.iscsi.size > 0); MPASS(ci != NULL); pr = &ci->pr; r = t4_read_reg(sc, A_ULP_RX_ISCSI_PSZ); rc = t4_init_ppod_region(pr, &sc->vres.iscsi, r, "iSCSI page pods"); if (rc != 0) { device_printf(sc->dev, "%s: failed to initialize the iSCSI page pod region: %u.\n", __func__, rc); return (rc); } r = t4_read_reg(sc, A_ULP_RX_ISCSI_TAGMASK); r &= V_ISCSITAGMASK(M_ISCSITAGMASK); if (r != pr->pr_tag_mask) { /* * Recent firmwares are supposed to set up the iSCSI tagmask * but we'll do it ourselves it the computed value doesn't match * what's in the register. */ device_printf(sc->dev, "tagmask 0x%08x does not match computed mask 0x%08x.\n", r, pr->pr_tag_mask); t4_set_reg_field(sc, A_ULP_RX_ISCSI_TAGMASK, V_ISCSITAGMASK(M_ISCSITAGMASK), pr->pr_tag_mask); } read_pdu_limits(sc, &ci->max_tx_data_len, &ci->max_rx_data_len, pr); sysctl_ctx_init(&ci->ctx); oid = device_get_sysctl_tree(sc->dev); /* dev.t5nex.X */ children = SYSCTL_CHILDREN(oid); oid = SYSCTL_ADD_NODE(&ci->ctx, children, OID_AUTO, "iscsi", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "iSCSI ULP settings"); children = SYSCTL_CHILDREN(oid); ci->ddp_threshold = 2048; SYSCTL_ADD_UINT(&ci->ctx, children, OID_AUTO, "ddp_threshold", CTLFLAG_RW, &ci->ddp_threshold, 0, "Rx zero copy threshold"); return (0); } static int do_rx_iscsi_hdr(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; struct cpl_iscsi_hdr *cpl = mtod(m, struct cpl_iscsi_hdr *); u_int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct icl_pdu *ip; struct icl_cxgbei_pdu *icp; uint16_t len_ddp = be16toh(cpl->pdu_len_ddp); uint16_t len = be16toh(cpl->len); M_ASSERTPKTHDR(m); MPASS(m->m_pkthdr.len == len + sizeof(*cpl)); ip = icl_cxgbei_new_pdu(M_NOWAIT); if (ip == NULL) CXGBE_UNIMPLEMENTED("PDU allocation failure"); m_copydata(m, sizeof(*cpl), ISCSI_BHS_SIZE, (caddr_t)ip->ip_bhs); ip->ip_data_len = G_ISCSI_PDU_LEN(len_ddp) - len; icp = ip_to_icp(ip); icp->icp_seq = ntohl(cpl->seq); icp->icp_flags = ICPF_RX_HDR; /* This is the start of a new PDU. There should be no old state. */ MPASS(toep->ulpcb2 == NULL); toep->ulpcb2 = icp; #if 0 CTR5(KTR_CXGBE, "%s: tid %u, cpl->len %u, pdu_len_ddp 0x%04x, icp %p", __func__, tid, len, len_ddp, icp); #endif m_freem(m); return (0); } static int do_rx_iscsi_data(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; struct cpl_iscsi_data *cpl = mtod(m, struct cpl_iscsi_data *); u_int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct icl_cxgbei_pdu *icp = toep->ulpcb2; struct icl_pdu *ip; M_ASSERTPKTHDR(m); MPASS(m->m_pkthdr.len == be16toh(cpl->len) + sizeof(*cpl)); if (icp == NULL) { /* * T6 completion enabled, start of a new pdu. Header * will come in completion CPL. */ ip = icl_cxgbei_new_pdu(M_NOWAIT); if (ip == NULL) CXGBE_UNIMPLEMENTED("PDU allocation failure"); icp = ip_to_icp(ip); } else { /* T5 mode, header is already received. */ MPASS(icp->icp_flags == ICPF_RX_HDR); MPASS(icp->ip.ip_data_mbuf == NULL); MPASS(icp->ip.ip_data_len == m->m_pkthdr.len - sizeof(*cpl)); } /* Trim the cpl header from mbuf. */ m_adj(m, sizeof(*cpl)); icp->icp_flags |= ICPF_RX_FLBUF; icp->ip.ip_data_mbuf = m; toep->ofld_rxq->rx_iscsi_fl_pdus++; toep->ofld_rxq->rx_iscsi_fl_octets += m->m_pkthdr.len; /* * For T6, save the icp for further processing in the * completion handler. */ if (icp->icp_flags == ICPF_RX_FLBUF) { MPASS(toep->ulpcb2 == NULL); toep->ulpcb2 = icp; } #if 0 CTR4(KTR_CXGBE, "%s: tid %u, cpl->len %u, icp %p", __func__, tid, be16toh(cpl->len), icp); #endif return (0); } static int do_rx_iscsi_ddp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_rx_data_ddp *cpl = (const void *)(rss + 1); u_int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct inpcb *inp = toep->inp; struct socket *so; struct sockbuf *sb; struct tcpcb *tp; struct icl_cxgbei_conn *icc; struct icl_conn *ic; struct icl_cxgbei_pdu *icp = toep->ulpcb2; struct icl_pdu *ip; u_int pdu_len, val; struct epoch_tracker et; MPASS(m == NULL); /* Must already be assembling a PDU. */ MPASS(icp != NULL); MPASS(icp->icp_flags & ICPF_RX_HDR); /* Data is optional. */ MPASS((icp->icp_flags & ICPF_RX_STATUS) == 0); pdu_len = be16toh(cpl->len); /* includes everything. */ val = be32toh(cpl->ddpvld); #if 0 CTR5(KTR_CXGBE, "%s: tid %u, cpl->len %u, ddpvld 0x%08x, icp_flags 0x%08x", __func__, tid, pdu_len, val, icp->icp_flags); #endif icp->icp_flags |= ICPF_RX_STATUS; ip = &icp->ip; if (val & F_DDP_PADDING_ERR) icp->icp_flags |= ICPF_PAD_ERR; if (val & F_DDP_HDRCRC_ERR) icp->icp_flags |= ICPF_HCRC_ERR; if (val & F_DDP_DATACRC_ERR) icp->icp_flags |= ICPF_DCRC_ERR; if (val & F_DDP_PDU && ip->ip_data_mbuf == NULL) { MPASS((icp->icp_flags & ICPF_RX_FLBUF) == 0); MPASS(ip->ip_data_len > 0); icp->icp_flags |= ICPF_RX_DDP; toep->ofld_rxq->rx_iscsi_ddp_pdus++; toep->ofld_rxq->rx_iscsi_ddp_octets += ip->ip_data_len; } INP_WLOCK(inp); if (__predict_false(inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT))) { CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), inp_flags 0x%x", __func__, tid, pdu_len, inp->inp_flags); INP_WUNLOCK(inp); icl_cxgbei_conn_pdu_free(NULL, ip); toep->ulpcb2 = NULL; return (0); } /* * T6+ does not report data PDUs received via DDP without F * set. This can result in gaps in the TCP sequence space. */ tp = intotcpcb(inp); MPASS(chip_id(sc) >= CHELSIO_T6 || icp->icp_seq == tp->rcv_nxt); tp->rcv_nxt = icp->icp_seq + pdu_len; tp->t_rcvtime = ticks; /* * Don't update the window size or return credits since RX * flow control is disabled. */ so = inp->inp_socket; sb = &so->so_rcv; SOCKBUF_LOCK(sb); icc = toep->ulpcb; if (__predict_false(icc == NULL || sb->sb_state & SBS_CANTRCVMORE)) { CTR5(KTR_CXGBE, "%s: tid %u, excess rx (%d bytes), icc %p, sb_state 0x%x", __func__, tid, pdu_len, icc, sb->sb_state); SOCKBUF_UNLOCK(sb); INP_WUNLOCK(inp); CURVNET_SET(so->so_vnet); NET_EPOCH_ENTER(et); INP_WLOCK(inp); tp = tcp_drop(tp, ECONNRESET); if (tp) INP_WUNLOCK(inp); NET_EPOCH_EXIT(et); CURVNET_RESTORE(); icl_cxgbei_conn_pdu_free(NULL, ip); toep->ulpcb2 = NULL; return (0); } MPASS(icc->icc_signature == CXGBEI_CONN_SIGNATURE); ic = &icc->ic; icl_cxgbei_new_pdu_set_conn(ip, ic); MPASS(m == NULL); /* was unused, we'll use it now. */ m = sbcut_locked(sb, sbused(sb)); /* XXXNP: toep->sb_cc accounting? */ if (__predict_false(m != NULL)) { int len = m_length(m, NULL); /* * PDUs were received before the tid transitioned to ULP mode. * Convert them to icl_cxgbei_pdus and send them to ICL before * the PDU in icp/ip. */ CTR3(KTR_CXGBE, "%s: tid %u, %u bytes in so_rcv", __func__, tid, len); /* XXXNP: needs to be rewritten. */ if (len == sizeof(struct iscsi_bhs) || len == 4 + sizeof(struct iscsi_bhs)) { struct icl_cxgbei_pdu *icp0; struct icl_pdu *ip0; ip0 = icl_cxgbei_new_pdu(M_NOWAIT); if (ip0 == NULL) CXGBE_UNIMPLEMENTED("PDU allocation failure"); icl_cxgbei_new_pdu_set_conn(ip0, ic); icp0 = ip_to_icp(ip0); icp0->icp_seq = 0; /* XXX */ icp0->icp_flags = ICPF_RX_HDR | ICPF_RX_STATUS; m_copydata(m, 0, sizeof(struct iscsi_bhs), (void *)ip0->ip_bhs); STAILQ_INSERT_TAIL(&icc->rcvd_pdus, ip0, ip_next); } m_freem(m); } STAILQ_INSERT_TAIL(&icc->rcvd_pdus, ip, ip_next); if ((icc->rx_flags & RXF_ACTIVE) == 0) { struct cxgbei_worker_thread_softc *cwt = &cwt_softc[icc->cwt]; mtx_lock(&cwt->cwt_lock); icc->rx_flags |= RXF_ACTIVE; TAILQ_INSERT_TAIL(&cwt->rx_head, icc, rx_link); if (cwt->cwt_state == CWT_SLEEPING) { cwt->cwt_state = CWT_RUNNING; cv_signal(&cwt->cwt_cv); } mtx_unlock(&cwt->cwt_lock); } SOCKBUF_UNLOCK(sb); INP_WUNLOCK(inp); toep->ulpcb2 = NULL; return (0); } static int do_rx_iscsi_cmp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct epoch_tracker et; struct adapter *sc = iq->adapter; struct cpl_rx_iscsi_cmp *cpl = mtod(m, struct cpl_rx_iscsi_cmp *); u_int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct icl_cxgbei_pdu *icp = toep->ulpcb2; struct icl_pdu *ip; struct cxgbei_cmp *cmp; struct inpcb *inp = toep->inp; #ifdef INVARIANTS uint16_t len = be16toh(cpl->len); #endif struct socket *so; struct sockbuf *sb; struct tcpcb *tp; struct icl_cxgbei_conn *icc; struct icl_conn *ic; struct iscsi_bhs_data_out *bhsdo; u_int val = be32toh(cpl->ddpvld); u_int npdus, pdu_len, data_digest_len, hdr_digest_len; uint32_t prev_seg_len; M_ASSERTPKTHDR(m); MPASS(m->m_pkthdr.len == len + sizeof(*cpl)); if ((val & F_DDP_PDU) == 0) { MPASS(icp != NULL); MPASS((icp->icp_flags & ICPF_RX_STATUS) == 0); ip = &icp->ip; } if (icp == NULL) { /* T6 completion enabled, start of a new PDU. */ ip = icl_cxgbei_new_pdu(M_NOWAIT); if (ip == NULL) CXGBE_UNIMPLEMENTED("PDU allocation failure"); icp = ip_to_icp(ip); } pdu_len = G_ISCSI_PDU_LEN(be16toh(cpl->pdu_len_ddp)); #if 0 CTR5(KTR_CXGBE, "%s: tid %u, cpl->len %u, ddpvld 0x%08x, icp %p", __func__, tid, pdu_len, val, icp); #endif /* Copy header */ m_copydata(m, sizeof(*cpl), ISCSI_BHS_SIZE, (caddr_t)ip->ip_bhs); bhsdo = (struct iscsi_bhs_data_out *)ip->ip_bhs; ip->ip_data_len = bhsdo->bhsdo_data_segment_len[0] << 16 | bhsdo->bhsdo_data_segment_len[1] << 8 | bhsdo->bhsdo_data_segment_len[2]; icp->icp_seq = ntohl(cpl->seq); icp->icp_flags |= ICPF_RX_HDR; icp->icp_flags |= ICPF_RX_STATUS; if (val & F_DDP_PADDING_ERR) icp->icp_flags |= ICPF_PAD_ERR; if (val & F_DDP_HDRCRC_ERR) icp->icp_flags |= ICPF_HCRC_ERR; if (val & F_DDP_DATACRC_ERR) icp->icp_flags |= ICPF_DCRC_ERR; INP_WLOCK(inp); if (__predict_false(inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT))) { CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), inp_flags 0x%x", __func__, tid, pdu_len, inp->inp_flags); INP_WUNLOCK(inp); icl_cxgbei_conn_pdu_free(NULL, ip); toep->ulpcb2 = NULL; m_freem(m); return (0); } tp = intotcpcb(inp); /* * If icc is NULL, the connection is being closed in * icl_cxgbei_conn_close(), just drop this data. */ icc = toep->ulpcb; if (__predict_false(icc == NULL)) { CTR4(KTR_CXGBE, "%s: tid %u, excess rx (%d bytes), icc %p", __func__, tid, pdu_len, icc); /* * Update rcv_nxt so the sequence number of the FIN * doesn't appear wrong. */ tp->rcv_nxt = icp->icp_seq + pdu_len; tp->t_rcvtime = ticks; INP_WUNLOCK(inp); icl_cxgbei_conn_pdu_free(NULL, ip); toep->ulpcb2 = NULL; m_freem(m); return (0); } data_digest_len = (icc->ulp_submode & ULP_CRC_DATA) ? ISCSI_DATA_DIGEST_SIZE : 0; hdr_digest_len = (icc->ulp_submode & ULP_CRC_HEADER) ? ISCSI_HEADER_DIGEST_SIZE : 0; MPASS(roundup2(ip->ip_data_len, 4) == pdu_len - len - data_digest_len); if (val & F_DDP_PDU && ip->ip_data_mbuf == NULL) { MPASS((icp->icp_flags & ICPF_RX_FLBUF) == 0); MPASS(ip->ip_data_len > 0); icp->icp_flags |= ICPF_RX_DDP; bhsdo = (struct iscsi_bhs_data_out *)ip->ip_bhs; switch (ip->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) { case ISCSI_BHS_OPCODE_SCSI_DATA_IN: cmp = cxgbei_find_cmp(icc, be32toh(bhsdo->bhsdo_initiator_task_tag)); break; case ISCSI_BHS_OPCODE_SCSI_DATA_OUT: cmp = cxgbei_find_cmp(icc, be32toh(bhsdo->bhsdo_target_transfer_tag)); break; default: __assert_unreachable(); } MPASS(cmp != NULL); /* * The difference between the end of the last burst * and the offset of the last PDU in this burst is * the additional data received via DDP. */ prev_seg_len = be32toh(bhsdo->bhsdo_buffer_offset) - cmp->next_buffer_offset; if (prev_seg_len != 0) { + uint32_t orig_datasn; + /* - * Since cfiscsi doesn't know about previous - * headers, pretend that the entire r2t data - * length was received in this single segment. + * Return a "large" PDU representing the burst + * of PDUs. Adjust the offset and length of + * this PDU to represent the entire burst. */ ip->ip_data_len += prev_seg_len; bhsdo->bhsdo_data_segment_len[2] = ip->ip_data_len; bhsdo->bhsdo_data_segment_len[1] = ip->ip_data_len >> 8; bhsdo->bhsdo_data_segment_len[0] = ip->ip_data_len >> 16; bhsdo->bhsdo_buffer_offset = htobe32(cmp->next_buffer_offset); - npdus = htobe32(bhsdo->bhsdo_datasn) - cmp->last_datasn; + orig_datasn = htobe32(bhsdo->bhsdo_datasn); + npdus = orig_datasn - cmp->last_datasn; + bhsdo->bhsdo_datasn = htobe32(cmp->last_datasn + 1); + cmp->last_datasn = orig_datasn; + ip->ip_additional_pdus = npdus - 1; } else { MPASS(htobe32(bhsdo->bhsdo_datasn) == cmp->last_datasn + 1); npdus = 1; + cmp->last_datasn = htobe32(bhsdo->bhsdo_datasn); } cmp->next_buffer_offset += ip->ip_data_len; - cmp->last_datasn = htobe32(bhsdo->bhsdo_datasn); - bhsdo->bhsdo_datasn = htobe32(cmp->next_datasn); - cmp->next_datasn++; toep->ofld_rxq->rx_iscsi_ddp_pdus += npdus; toep->ofld_rxq->rx_iscsi_ddp_octets += ip->ip_data_len; } else { MPASS(icp->icp_flags & (ICPF_RX_FLBUF)); MPASS(ip->ip_data_len == ip->ip_data_mbuf->m_pkthdr.len); } tp->rcv_nxt = icp->icp_seq + pdu_len; tp->t_rcvtime = ticks; /* * Don't update the window size or return credits since RX * flow control is disabled. */ so = inp->inp_socket; sb = &so->so_rcv; SOCKBUF_LOCK(sb); if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) { CTR5(KTR_CXGBE, "%s: tid %u, excess rx (%d bytes), icc %p, sb_state 0x%x", __func__, tid, pdu_len, icc, sb->sb_state); SOCKBUF_UNLOCK(sb); INP_WUNLOCK(inp); CURVNET_SET(so->so_vnet); NET_EPOCH_ENTER(et); INP_WLOCK(inp); tp = tcp_drop(tp, ECONNRESET); if (tp != NULL) INP_WUNLOCK(inp); NET_EPOCH_EXIT(et); CURVNET_RESTORE(); icl_cxgbei_conn_pdu_free(NULL, ip); toep->ulpcb2 = NULL; m_freem(m); return (0); } MPASS(icc->icc_signature == CXGBEI_CONN_SIGNATURE); ic = &icc->ic; icl_cxgbei_new_pdu_set_conn(ip, ic); /* Enqueue the PDU to the received pdus queue. */ STAILQ_INSERT_TAIL(&icc->rcvd_pdus, ip, ip_next); if ((icc->rx_flags & RXF_ACTIVE) == 0) { struct cxgbei_worker_thread_softc *cwt = &cwt_softc[icc->cwt]; mtx_lock(&cwt->cwt_lock); icc->rx_flags |= RXF_ACTIVE; TAILQ_INSERT_TAIL(&cwt->rx_head, icc, rx_link); if (cwt->cwt_state == CWT_SLEEPING) { cwt->cwt_state = CWT_RUNNING; cv_signal(&cwt->cwt_cv); } mtx_unlock(&cwt->cwt_lock); } SOCKBUF_UNLOCK(sb); INP_WUNLOCK(inp); toep->ulpcb2 = NULL; m_freem(m); return (0); } static int cxgbei_activate(struct adapter *sc) { struct cxgbei_data *ci; int rc; ASSERT_SYNCHRONIZED_OP(sc); if (uld_active(sc, ULD_ISCSI)) { KASSERT(0, ("%s: iSCSI offload already enabled on adapter %p", __func__, sc)); return (0); } if (sc->iscsicaps == 0 || sc->vres.iscsi.size == 0) { device_printf(sc->dev, "not iSCSI offload capable, or capability disabled.\n"); return (ENOSYS); } /* per-adapter softc for iSCSI */ ci = malloc(sizeof(*ci), M_CXGBE, M_ZERO | M_WAITOK); if (ci == NULL) return (ENOMEM); rc = cxgbei_init(sc, ci); if (rc != 0) { free(ci, M_CXGBE); return (rc); } sc->iscsi_ulp_softc = ci; return (0); } static int cxgbei_deactivate(struct adapter *sc) { struct cxgbei_data *ci = sc->iscsi_ulp_softc; ASSERT_SYNCHRONIZED_OP(sc); if (ci != NULL) { sysctl_ctx_free(&ci->ctx); t4_free_ppod_region(&ci->pr); free(ci, M_CXGBE); sc->iscsi_ulp_softc = NULL; } return (0); } static void cxgbei_activate_all(struct adapter *sc, void *arg __unused) { if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4isact") != 0) return; /* Activate iSCSI if any port on this adapter has IFCAP_TOE enabled. */ if (sc->offload_map && !uld_active(sc, ULD_ISCSI)) (void) t4_activate_uld(sc, ULD_ISCSI); end_synchronized_op(sc, 0); } static void cxgbei_deactivate_all(struct adapter *sc, void *arg __unused) { if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4isdea") != 0) return; if (uld_active(sc, ULD_ISCSI)) (void) t4_deactivate_uld(sc, ULD_ISCSI); end_synchronized_op(sc, 0); } static struct uld_info cxgbei_uld_info = { .uld_id = ULD_ISCSI, .activate = cxgbei_activate, .deactivate = cxgbei_deactivate, }; static void cwt_main(void *arg) { struct cxgbei_worker_thread_softc *cwt = arg; struct icl_cxgbei_conn *icc = NULL; struct icl_conn *ic; struct icl_pdu *ip; struct sockbuf *sb; STAILQ_HEAD(, icl_pdu) rx_pdus = STAILQ_HEAD_INITIALIZER(rx_pdus); MPASS(cwt != NULL); mtx_lock(&cwt->cwt_lock); MPASS(cwt->cwt_state == 0); cwt->cwt_state = CWT_RUNNING; cv_signal(&cwt->cwt_cv); while (__predict_true(cwt->cwt_state != CWT_STOP)) { cwt->cwt_state = CWT_RUNNING; while ((icc = TAILQ_FIRST(&cwt->rx_head)) != NULL) { TAILQ_REMOVE(&cwt->rx_head, icc, rx_link); mtx_unlock(&cwt->cwt_lock); ic = &icc->ic; sb = &ic->ic_socket->so_rcv; SOCKBUF_LOCK(sb); MPASS(icc->rx_flags & RXF_ACTIVE); if (__predict_true(!(sb->sb_state & SBS_CANTRCVMORE))) { MPASS(STAILQ_EMPTY(&rx_pdus)); STAILQ_SWAP(&icc->rcvd_pdus, &rx_pdus, icl_pdu); SOCKBUF_UNLOCK(sb); /* Hand over PDUs to ICL. */ while ((ip = STAILQ_FIRST(&rx_pdus)) != NULL) { STAILQ_REMOVE_HEAD(&rx_pdus, ip_next); ic->ic_receive(ip); } SOCKBUF_LOCK(sb); MPASS(STAILQ_EMPTY(&rx_pdus)); } MPASS(icc->rx_flags & RXF_ACTIVE); if (STAILQ_EMPTY(&icc->rcvd_pdus) || __predict_false(sb->sb_state & SBS_CANTRCVMORE)) { icc->rx_flags &= ~RXF_ACTIVE; } else { /* * More PDUs were received while we were busy * handing over the previous batch to ICL. * Re-add this connection to the end of the * queue. */ mtx_lock(&cwt->cwt_lock); TAILQ_INSERT_TAIL(&cwt->rx_head, icc, rx_link); mtx_unlock(&cwt->cwt_lock); } SOCKBUF_UNLOCK(sb); mtx_lock(&cwt->cwt_lock); } /* Inner loop doesn't check for CWT_STOP, do that first. */ if (__predict_false(cwt->cwt_state == CWT_STOP)) break; cwt->cwt_state = CWT_SLEEPING; cv_wait(&cwt->cwt_cv, &cwt->cwt_lock); } MPASS(TAILQ_FIRST(&cwt->rx_head) == NULL); mtx_assert(&cwt->cwt_lock, MA_OWNED); cwt->cwt_state = CWT_STOPPED; cv_signal(&cwt->cwt_cv); mtx_unlock(&cwt->cwt_lock); kthread_exit(); } static int start_worker_threads(void) { int i, rc; struct cxgbei_worker_thread_softc *cwt; worker_thread_count = min(mp_ncpus, 32); cwt_softc = malloc(worker_thread_count * sizeof(*cwt), M_CXGBE, M_WAITOK | M_ZERO); MPASS(cxgbei_proc == NULL); for (i = 0, cwt = &cwt_softc[0]; i < worker_thread_count; i++, cwt++) { mtx_init(&cwt->cwt_lock, "cwt lock", NULL, MTX_DEF); cv_init(&cwt->cwt_cv, "cwt cv"); TAILQ_INIT(&cwt->rx_head); rc = kproc_kthread_add(cwt_main, cwt, &cxgbei_proc, NULL, 0, 0, "cxgbei", "%d", i); if (rc != 0) { printf("cxgbei: failed to start thread #%d/%d (%d)\n", i + 1, worker_thread_count, rc); mtx_destroy(&cwt->cwt_lock); cv_destroy(&cwt->cwt_cv); bzero(cwt, sizeof(*cwt)); if (i == 0) { free(cwt_softc, M_CXGBE); worker_thread_count = 0; return (rc); } /* Not fatal, carry on with fewer threads. */ worker_thread_count = i; rc = 0; break; } /* Wait for thread to start before moving on to the next one. */ mtx_lock(&cwt->cwt_lock); while (cwt->cwt_state == 0) cv_wait(&cwt->cwt_cv, &cwt->cwt_lock); mtx_unlock(&cwt->cwt_lock); } MPASS(cwt_softc != NULL); MPASS(worker_thread_count > 0); return (0); } static void stop_worker_threads(void) { int i; struct cxgbei_worker_thread_softc *cwt = &cwt_softc[0]; MPASS(worker_thread_count >= 0); for (i = 0, cwt = &cwt_softc[0]; i < worker_thread_count; i++, cwt++) { mtx_lock(&cwt->cwt_lock); MPASS(cwt->cwt_state == CWT_RUNNING || cwt->cwt_state == CWT_SLEEPING); cwt->cwt_state = CWT_STOP; cv_signal(&cwt->cwt_cv); do { cv_wait(&cwt->cwt_cv, &cwt->cwt_lock); } while (cwt->cwt_state != CWT_STOPPED); mtx_unlock(&cwt->cwt_lock); mtx_destroy(&cwt->cwt_lock); cv_destroy(&cwt->cwt_cv); } free(cwt_softc, M_CXGBE); } /* Select a worker thread for a connection. */ u_int cxgbei_select_worker_thread(struct icl_cxgbei_conn *icc) { struct adapter *sc = icc->sc; struct toepcb *toep = icc->toep; u_int i, n; n = worker_thread_count / sc->sge.nofldrxq; if (n > 0) i = toep->vi->pi->port_id * n + arc4random() % n; else i = arc4random() % worker_thread_count; CTR3(KTR_CXGBE, "%s: tid %u, cwt %u", __func__, toep->tid, i); return (i); } static int cxgbei_mod_load(void) { int rc; t4_register_cpl_handler(CPL_ISCSI_HDR, do_rx_iscsi_hdr); t4_register_cpl_handler(CPL_ISCSI_DATA, do_rx_iscsi_data); t4_register_cpl_handler(CPL_RX_ISCSI_DDP, do_rx_iscsi_ddp); t4_register_cpl_handler(CPL_RX_ISCSI_CMP, do_rx_iscsi_cmp); rc = start_worker_threads(); if (rc != 0) return (rc); rc = t4_register_uld(&cxgbei_uld_info); if (rc != 0) { stop_worker_threads(); return (rc); } t4_iterate(cxgbei_activate_all, NULL); return (rc); } static int cxgbei_mod_unload(void) { t4_iterate(cxgbei_deactivate_all, NULL); if (t4_unregister_uld(&cxgbei_uld_info) == EBUSY) return (EBUSY); stop_worker_threads(); t4_register_cpl_handler(CPL_ISCSI_HDR, NULL); t4_register_cpl_handler(CPL_ISCSI_DATA, NULL); t4_register_cpl_handler(CPL_RX_ISCSI_DDP, NULL); t4_register_cpl_handler(CPL_RX_ISCSI_CMP, NULL); return (0); } #endif static int cxgbei_modevent(module_t mod, int cmd, void *arg) { int rc = 0; #ifdef TCP_OFFLOAD switch (cmd) { case MOD_LOAD: rc = cxgbei_mod_load(); if (rc == 0) rc = icl_cxgbei_mod_load(); break; case MOD_UNLOAD: rc = icl_cxgbei_mod_unload(); if (rc == 0) rc = cxgbei_mod_unload(); break; default: rc = EINVAL; } #else printf("cxgbei: compiled without TCP_OFFLOAD support.\n"); rc = EOPNOTSUPP; #endif return (rc); } static moduledata_t cxgbei_mod = { "cxgbei", cxgbei_modevent, NULL, }; MODULE_VERSION(cxgbei, 1); DECLARE_MODULE(cxgbei, cxgbei_mod, SI_SUB_EXEC, SI_ORDER_ANY); MODULE_DEPEND(cxgbei, t4_tom, 1, 1, 1); MODULE_DEPEND(cxgbei, cxgbe, 1, 1, 1); MODULE_DEPEND(cxgbei, icl, 1, 1, 1); diff --git a/sys/dev/cxgbe/cxgbei/cxgbei.h b/sys/dev/cxgbe/cxgbei/cxgbei.h index 20754fa893a8..09d556988091 100644 --- a/sys/dev/cxgbe/cxgbei/cxgbei.h +++ b/sys/dev/cxgbe/cxgbei/cxgbei.h @@ -1,150 +1,149 @@ /*- * Copyright (c) 2012, 2015 Chelsio Communications, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ * */ #ifndef __CXGBEI_OFLD_H__ #define __CXGBEI_OFLD_H__ #include enum { CWT_SLEEPING = 1, CWT_RUNNING = 2, CWT_STOP = 3, CWT_STOPPED = 4, }; struct cxgbei_worker_thread_softc { struct mtx cwt_lock; struct cv cwt_cv; volatile int cwt_state; TAILQ_HEAD(, icl_cxgbei_conn) rx_head; } __aligned(CACHE_LINE_SIZE); #define CXGBEI_CONN_SIGNATURE 0x56788765 enum { RXF_ACTIVE = 1 << 0, /* In the worker thread's queue */ }; struct cxgbei_cmp { LIST_ENTRY(cxgbei_cmp) link; uint32_t tt; /* Transfer tag. */ - uint32_t next_datasn; uint32_t next_buffer_offset; uint32_t last_datasn; }; LIST_HEAD(cxgbei_cmp_head, cxgbei_cmp); struct icl_cxgbei_conn { struct icl_conn ic; /* cxgbei specific stuff goes here. */ uint32_t icc_signature; int ulp_submode; struct adapter *sc; struct toepcb *toep; /* Receive related. */ u_int rx_flags; /* protected by so_rcv lock */ u_int cwt; STAILQ_HEAD(, icl_pdu) rcvd_pdus; /* protected by so_rcv lock */ TAILQ_ENTRY(icl_cxgbei_conn) rx_link; /* protected by cwt lock */ struct cxgbei_cmp_head *cmp_table; /* protected by cmp_lock */ struct mtx cmp_lock; unsigned long cmp_hash_mask; }; static inline struct icl_cxgbei_conn * ic_to_icc(struct icl_conn *ic) { return (__containerof(ic, struct icl_cxgbei_conn, ic)); } /* PDU flags and signature. */ enum { ICPF_RX_HDR = 1 << 0, /* PDU header received. */ ICPF_RX_FLBUF = 1 << 1, /* PDU payload received in a freelist. */ ICPF_RX_DDP = 1 << 2, /* PDU payload DDP'd. */ ICPF_RX_STATUS = 1 << 3, /* Rx status received. */ ICPF_HCRC_ERR = 1 << 4, /* Header digest error. */ ICPF_DCRC_ERR = 1 << 5, /* Data digest error. */ ICPF_PAD_ERR = 1 << 6, /* Padding error. */ CXGBEI_PDU_SIGNATURE = 0x12344321 }; struct icl_cxgbei_pdu { struct icl_pdu ip; /* cxgbei specific stuff goes here. */ uint32_t icp_signature; uint32_t icp_seq; /* For debug only */ u_int icp_flags; u_int ref_cnt; icl_pdu_cb cb; int error; }; static inline struct icl_cxgbei_pdu * ip_to_icp(struct icl_pdu *ip) { return (__containerof(ip, struct icl_cxgbei_pdu, ip)); } struct cxgbei_data { u_int max_tx_data_len; u_int max_rx_data_len; u_int ddp_threshold; struct ppod_region pr; struct sysctl_ctx_list ctx; /* from uld_activate to deactivate */ }; #define CXGBEI_MAX_ISO_PAYLOAD 65535 /* cxgbei.c */ u_int cxgbei_select_worker_thread(struct icl_cxgbei_conn *); /* icl_cxgbei.c */ int icl_cxgbei_mod_load(void); int icl_cxgbei_mod_unload(void); struct icl_pdu *icl_cxgbei_new_pdu(int); void icl_cxgbei_new_pdu_set_conn(struct icl_pdu *, struct icl_conn *); void icl_cxgbei_conn_pdu_free(struct icl_conn *, struct icl_pdu *); struct cxgbei_cmp *cxgbei_find_cmp(struct icl_cxgbei_conn *, uint32_t); #endif diff --git a/sys/dev/cxgbe/cxgbei/icl_cxgbei.c b/sys/dev/cxgbe/cxgbei/icl_cxgbei.c index a6e7f8b95815..687fc545cebd 100644 --- a/sys/dev/cxgbe/cxgbei/icl_cxgbei.c +++ b/sys/dev/cxgbe/cxgbei/icl_cxgbei.c @@ -1,1447 +1,1446 @@ /*- * Copyright (c) 2012 The FreeBSD Foundation * Copyright (c) 2015 Chelsio Communications, Inc. * All rights reserved. * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ /* * cxgbei implementation of iSCSI Common Layer kobj(9) interface. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #ifdef TCP_OFFLOAD #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "common/common.h" #include "common/t4_tcb.h" #include "tom/t4_tom.h" #include "cxgbei.h" /* * Use the page pod tag for the TT hash. */ #define TT_HASH(icc, tt) (G_PPOD_TAG(tt) & (icc)->cmp_hash_mask) struct cxgbei_ddp_state { struct ppod_reservation prsv; struct cxgbei_cmp cmp; }; static MALLOC_DEFINE(M_CXGBEI, "cxgbei", "cxgbei(4)"); SYSCTL_NODE(_kern_icl, OID_AUTO, cxgbei, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Chelsio iSCSI offload"); static int first_burst_length = 8192; SYSCTL_INT(_kern_icl_cxgbei, OID_AUTO, first_burst_length, CTLFLAG_RWTUN, &first_burst_length, 0, "First burst length"); static int max_burst_length = 2 * 1024 * 1024; SYSCTL_INT(_kern_icl_cxgbei, OID_AUTO, max_burst_length, CTLFLAG_RWTUN, &max_burst_length, 0, "Maximum burst length"); static int sendspace = 1048576; SYSCTL_INT(_kern_icl_cxgbei, OID_AUTO, sendspace, CTLFLAG_RWTUN, &sendspace, 0, "Default send socket buffer size"); static int recvspace = 1048576; SYSCTL_INT(_kern_icl_cxgbei, OID_AUTO, recvspace, CTLFLAG_RWTUN, &recvspace, 0, "Default receive socket buffer size"); static volatile u_int icl_cxgbei_ncons; #define ICL_CONN_LOCK(X) mtx_lock(X->ic_lock) #define ICL_CONN_UNLOCK(X) mtx_unlock(X->ic_lock) #define ICL_CONN_LOCK_ASSERT(X) mtx_assert(X->ic_lock, MA_OWNED) #define ICL_CONN_LOCK_ASSERT_NOT(X) mtx_assert(X->ic_lock, MA_NOTOWNED) static icl_conn_new_pdu_t icl_cxgbei_conn_new_pdu; static icl_conn_pdu_data_segment_length_t icl_cxgbei_conn_pdu_data_segment_length; static icl_conn_pdu_append_data_t icl_cxgbei_conn_pdu_append_data; static icl_conn_pdu_get_data_t icl_cxgbei_conn_pdu_get_data; static icl_conn_pdu_queue_t icl_cxgbei_conn_pdu_queue; static icl_conn_pdu_queue_cb_t icl_cxgbei_conn_pdu_queue_cb; static icl_conn_handoff_t icl_cxgbei_conn_handoff; static icl_conn_free_t icl_cxgbei_conn_free; static icl_conn_close_t icl_cxgbei_conn_close; static icl_conn_task_setup_t icl_cxgbei_conn_task_setup; static icl_conn_task_done_t icl_cxgbei_conn_task_done; static icl_conn_transfer_setup_t icl_cxgbei_conn_transfer_setup; static icl_conn_transfer_done_t icl_cxgbei_conn_transfer_done; static kobj_method_t icl_cxgbei_methods[] = { KOBJMETHOD(icl_conn_new_pdu, icl_cxgbei_conn_new_pdu), KOBJMETHOD(icl_conn_pdu_free, icl_cxgbei_conn_pdu_free), KOBJMETHOD(icl_conn_pdu_data_segment_length, icl_cxgbei_conn_pdu_data_segment_length), KOBJMETHOD(icl_conn_pdu_append_data, icl_cxgbei_conn_pdu_append_data), KOBJMETHOD(icl_conn_pdu_get_data, icl_cxgbei_conn_pdu_get_data), KOBJMETHOD(icl_conn_pdu_queue, icl_cxgbei_conn_pdu_queue), KOBJMETHOD(icl_conn_pdu_queue_cb, icl_cxgbei_conn_pdu_queue_cb), KOBJMETHOD(icl_conn_handoff, icl_cxgbei_conn_handoff), KOBJMETHOD(icl_conn_free, icl_cxgbei_conn_free), KOBJMETHOD(icl_conn_close, icl_cxgbei_conn_close), KOBJMETHOD(icl_conn_task_setup, icl_cxgbei_conn_task_setup), KOBJMETHOD(icl_conn_task_done, icl_cxgbei_conn_task_done), KOBJMETHOD(icl_conn_transfer_setup, icl_cxgbei_conn_transfer_setup), KOBJMETHOD(icl_conn_transfer_done, icl_cxgbei_conn_transfer_done), { 0, 0 } }; DEFINE_CLASS(icl_cxgbei, icl_cxgbei_methods, sizeof(struct icl_cxgbei_conn)); void icl_cxgbei_conn_pdu_free(struct icl_conn *ic, struct icl_pdu *ip) { struct icl_cxgbei_pdu *icp = ip_to_icp(ip); KASSERT(icp->ref_cnt != 0, ("freeing deleted PDU")); MPASS(icp->icp_signature == CXGBEI_PDU_SIGNATURE); MPASS(ic == ip->ip_conn); m_freem(ip->ip_ahs_mbuf); m_freem(ip->ip_data_mbuf); m_freem(ip->ip_bhs_mbuf); KASSERT(ic != NULL || icp->ref_cnt == 1, ("orphaned PDU has oustanding references")); if (atomic_fetchadd_int(&icp->ref_cnt, -1) != 1) return; free(icp, M_CXGBEI); #ifdef DIAGNOSTIC if (__predict_true(ic != NULL)) refcount_release(&ic->ic_outstanding_pdus); #endif } static void icl_cxgbei_pdu_call_cb(struct icl_pdu *ip) { struct icl_cxgbei_pdu *icp = ip_to_icp(ip); MPASS(icp->icp_signature == CXGBEI_PDU_SIGNATURE); if (icp->cb != NULL) icp->cb(ip, icp->error); #ifdef DIAGNOSTIC if (__predict_true(ip->ip_conn != NULL)) refcount_release(&ip->ip_conn->ic_outstanding_pdus); #endif free(icp, M_CXGBEI); } static void icl_cxgbei_pdu_done(struct icl_pdu *ip, int error) { struct icl_cxgbei_pdu *icp = ip_to_icp(ip); if (error != 0) icp->error = error; m_freem(ip->ip_ahs_mbuf); ip->ip_ahs_mbuf = NULL; m_freem(ip->ip_data_mbuf); ip->ip_data_mbuf = NULL; m_freem(ip->ip_bhs_mbuf); ip->ip_bhs_mbuf = NULL; /* * All other references to this PDU should have been dropped * by the m_freem() of ip_data_mbuf. */ if (atomic_fetchadd_int(&icp->ref_cnt, -1) == 1) icl_cxgbei_pdu_call_cb(ip); else __assert_unreachable(); } static void icl_cxgbei_mbuf_done(struct mbuf *mb) { struct icl_cxgbei_pdu *icp = (struct icl_cxgbei_pdu *)mb->m_ext.ext_arg1; /* * NB: mb_free_mext() might leave ref_cnt as 1 without * decrementing it if it hits the fast path in the ref_cnt * check. */ icl_cxgbei_pdu_call_cb(&icp->ip); } struct icl_pdu * icl_cxgbei_new_pdu(int flags) { struct icl_cxgbei_pdu *icp; struct icl_pdu *ip; struct mbuf *m; icp = malloc(sizeof(*icp), M_CXGBEI, flags | M_ZERO); if (__predict_false(icp == NULL)) return (NULL); icp->icp_signature = CXGBEI_PDU_SIGNATURE; icp->ref_cnt = 1; ip = &icp->ip; m = m_gethdr(flags, MT_DATA); if (__predict_false(m == NULL)) { free(icp, M_CXGBEI); return (NULL); } ip->ip_bhs_mbuf = m; ip->ip_bhs = mtod(m, struct iscsi_bhs *); memset(ip->ip_bhs, 0, sizeof(*ip->ip_bhs)); m->m_len = sizeof(struct iscsi_bhs); m->m_pkthdr.len = m->m_len; return (ip); } void icl_cxgbei_new_pdu_set_conn(struct icl_pdu *ip, struct icl_conn *ic) { ip->ip_conn = ic; #ifdef DIAGNOSTIC refcount_acquire(&ic->ic_outstanding_pdus); #endif } /* * Allocate icl_pdu with empty BHS to fill up by the caller. */ static struct icl_pdu * icl_cxgbei_conn_new_pdu(struct icl_conn *ic, int flags) { struct icl_pdu *ip; ip = icl_cxgbei_new_pdu(flags); if (__predict_false(ip == NULL)) return (NULL); icl_cxgbei_new_pdu_set_conn(ip, ic); return (ip); } static size_t icl_pdu_data_segment_length(const struct icl_pdu *request) { uint32_t len = 0; len += request->ip_bhs->bhs_data_segment_len[0]; len <<= 8; len += request->ip_bhs->bhs_data_segment_len[1]; len <<= 8; len += request->ip_bhs->bhs_data_segment_len[2]; return (len); } size_t icl_cxgbei_conn_pdu_data_segment_length(struct icl_conn *ic, const struct icl_pdu *request) { return (icl_pdu_data_segment_length(request)); } static struct mbuf * finalize_pdu(struct icl_cxgbei_conn *icc, struct icl_cxgbei_pdu *icp) { struct icl_pdu *ip = &icp->ip; uint8_t ulp_submode, padding; struct mbuf *m, *last; struct iscsi_bhs *bhs; int data_len; /* * Fix up the data segment mbuf first. */ m = ip->ip_data_mbuf; ulp_submode = icc->ulp_submode; if (m != NULL) { last = m_last(m); /* * Round up the data segment to a 4B boundary. Pad with 0 if * necessary. There will definitely be room in the mbuf. */ padding = roundup2(ip->ip_data_len, 4) - ip->ip_data_len; if (padding != 0) { MPASS(padding <= M_TRAILINGSPACE(last)); bzero(mtod(last, uint8_t *) + last->m_len, padding); last->m_len += padding; } } else { MPASS(ip->ip_data_len == 0); ulp_submode &= ~ULP_CRC_DATA; padding = 0; } /* * Now the header mbuf that has the BHS. */ m = ip->ip_bhs_mbuf; MPASS(m->m_pkthdr.len == sizeof(struct iscsi_bhs)); MPASS(m->m_len == sizeof(struct iscsi_bhs)); bhs = ip->ip_bhs; data_len = ip->ip_data_len; if (data_len > icc->ic.ic_max_send_data_segment_length) { struct iscsi_bhs_data_in *bhsdi; int flags; KASSERT(padding == 0, ("%s: ISO with padding %d for icp %p", __func__, padding, icp)); switch (bhs->bhs_opcode) { case ISCSI_BHS_OPCODE_SCSI_DATA_OUT: flags = 1; break; case ISCSI_BHS_OPCODE_SCSI_DATA_IN: flags = 2; break; default: panic("invalid opcode %#x for ISO", bhs->bhs_opcode); } data_len = icc->ic.ic_max_send_data_segment_length; bhsdi = (struct iscsi_bhs_data_in *)bhs; if (bhsdi->bhsdi_flags & BHSDI_FLAGS_F) { /* * Firmware will set F on the final PDU in the * burst. */ flags |= CXGBE_ISO_F; bhsdi->bhsdi_flags &= ~BHSDI_FLAGS_F; } set_mbuf_iscsi_iso(m, true); set_mbuf_iscsi_iso_flags(m, flags); set_mbuf_iscsi_iso_mss(m, data_len); } bhs->bhs_data_segment_len[2] = data_len; bhs->bhs_data_segment_len[1] = data_len >> 8; bhs->bhs_data_segment_len[0] = data_len >> 16; /* * Extract mbuf chain from PDU. */ m->m_pkthdr.len += ip->ip_data_len + padding; m->m_next = ip->ip_data_mbuf; set_mbuf_ulp_submode(m, ulp_submode); ip->ip_bhs_mbuf = NULL; ip->ip_data_mbuf = NULL; ip->ip_bhs = NULL; /* * Drop PDU reference on icp. Additional references might * still be held by zero-copy PDU buffers (ICL_NOCOPY). */ if (atomic_fetchadd_int(&icp->ref_cnt, -1) == 1) icl_cxgbei_pdu_call_cb(ip); return (m); } int icl_cxgbei_conn_pdu_append_data(struct icl_conn *ic, struct icl_pdu *ip, const void *addr, size_t len, int flags) { struct icl_cxgbei_pdu *icp = ip_to_icp(ip); struct mbuf *m, *m_tail; const char *src; MPASS(icp->icp_signature == CXGBEI_PDU_SIGNATURE); MPASS(ic == ip->ip_conn); KASSERT(len > 0, ("%s: len is %jd", __func__, (intmax_t)len)); m_tail = ip->ip_data_mbuf; if (m_tail != NULL) for (; m_tail->m_next != NULL; m_tail = m_tail->m_next) ; if (flags & ICL_NOCOPY) { m = m_get(flags & ~ICL_NOCOPY, MT_DATA); if (m == NULL) { ICL_WARN("failed to allocate mbuf"); return (ENOMEM); } m->m_flags |= M_RDONLY; m_extaddref(m, __DECONST(char *, addr), len, &icp->ref_cnt, icl_cxgbei_mbuf_done, icp, NULL); m->m_len = len; if (ip->ip_data_mbuf == NULL) { ip->ip_data_mbuf = m; ip->ip_data_len = len; } else { m_tail->m_next = m; m_tail = m_tail->m_next; ip->ip_data_len += len; } return (0); } src = (const char *)addr; /* Allocate as jumbo mbufs of size MJUM16BYTES. */ while (len >= MJUM16BYTES) { m = m_getjcl(M_NOWAIT, MT_DATA, 0, MJUM16BYTES); if (__predict_false(m == NULL)) { if ((flags & M_WAITOK) != 0) { /* Fall back to non-jumbo mbufs. */ break; } return (ENOMEM); } memcpy(mtod(m, void *), src, MJUM16BYTES); m->m_len = MJUM16BYTES; if (ip->ip_data_mbuf == NULL) { ip->ip_data_mbuf = m_tail = m; ip->ip_data_len = MJUM16BYTES; } else { m_tail->m_next = m; m_tail = m_tail->m_next; ip->ip_data_len += MJUM16BYTES; } src += MJUM16BYTES; len -= MJUM16BYTES; } /* Allocate mbuf chain for the remaining data. */ if (len != 0) { m = m_getm2(NULL, len, flags, MT_DATA, 0); if (__predict_false(m == NULL)) return (ENOMEM); if (ip->ip_data_mbuf == NULL) { ip->ip_data_mbuf = m; ip->ip_data_len = len; } else { m_tail->m_next = m; ip->ip_data_len += len; } for (; m != NULL; m = m->m_next) { m->m_len = min(len, M_SIZE(m)); memcpy(mtod(m, void *), src, m->m_len); src += m->m_len; len -= m->m_len; } MPASS(len == 0); } MPASS(ip->ip_data_len <= max(ic->ic_max_send_data_segment_length, ic->ic_hw_isomax)); return (0); } void icl_cxgbei_conn_pdu_get_data(struct icl_conn *ic, struct icl_pdu *ip, size_t off, void *addr, size_t len) { struct icl_cxgbei_pdu *icp = ip_to_icp(ip); if (icp->icp_flags & ICPF_RX_DDP) return; /* data is DDP'ed, no need to copy */ m_copydata(ip->ip_data_mbuf, off, len, addr); } void icl_cxgbei_conn_pdu_queue(struct icl_conn *ic, struct icl_pdu *ip) { icl_cxgbei_conn_pdu_queue_cb(ic, ip, NULL); } void icl_cxgbei_conn_pdu_queue_cb(struct icl_conn *ic, struct icl_pdu *ip, icl_pdu_cb cb) { struct epoch_tracker et; struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct icl_cxgbei_pdu *icp = ip_to_icp(ip); struct socket *so = ic->ic_socket; struct toepcb *toep = icc->toep; struct inpcb *inp; struct mbuf *m; MPASS(ic == ip->ip_conn); MPASS(ip->ip_bhs_mbuf != NULL); /* The kernel doesn't generate PDUs with AHS. */ MPASS(ip->ip_ahs_mbuf == NULL && ip->ip_ahs_len == 0); ICL_CONN_LOCK_ASSERT(ic); icp->cb = cb; /* NOTE: sowriteable without so_snd lock is a mostly harmless race. */ if (ic->ic_disconnecting || so == NULL || !sowriteable(so)) { icl_cxgbei_pdu_done(ip, ENOTCONN); return; } m = finalize_pdu(icc, icp); M_ASSERTPKTHDR(m); MPASS((m->m_pkthdr.len & 3) == 0); /* * Do not get inp from toep->inp as the toepcb might have detached * already. */ inp = sotoinpcb(so); CURVNET_SET(toep->vnet); NET_EPOCH_ENTER(et); INP_WLOCK(inp); if (__predict_false(inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) || __predict_false((toep->flags & TPF_ATTACHED) == 0)) m_freem(m); else { mbufq_enqueue(&toep->ulp_pduq, m); t4_push_pdus(icc->sc, toep, 0); } INP_WUNLOCK(inp); NET_EPOCH_EXIT(et); CURVNET_RESTORE(); } static struct icl_conn * icl_cxgbei_new_conn(const char *name, struct mtx *lock) { struct icl_cxgbei_conn *icc; struct icl_conn *ic; refcount_acquire(&icl_cxgbei_ncons); icc = (struct icl_cxgbei_conn *)kobj_create(&icl_cxgbei_class, M_CXGBE, M_WAITOK | M_ZERO); icc->icc_signature = CXGBEI_CONN_SIGNATURE; STAILQ_INIT(&icc->rcvd_pdus); icc->cmp_table = hashinit(64, M_CXGBEI, &icc->cmp_hash_mask); mtx_init(&icc->cmp_lock, "cxgbei_cmp", NULL, MTX_DEF); ic = &icc->ic; ic->ic_lock = lock; #ifdef DIAGNOSTIC refcount_init(&ic->ic_outstanding_pdus, 0); #endif ic->ic_name = name; ic->ic_offload = "cxgbei"; ic->ic_unmapped = false; CTR2(KTR_CXGBE, "%s: icc %p", __func__, icc); return (ic); } void icl_cxgbei_conn_free(struct icl_conn *ic) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); MPASS(icc->icc_signature == CXGBEI_CONN_SIGNATURE); CTR2(KTR_CXGBE, "%s: icc %p", __func__, icc); mtx_destroy(&icc->cmp_lock); hashdestroy(icc->cmp_table, M_CXGBEI, icc->cmp_hash_mask); kobj_delete((struct kobj *)icc, M_CXGBE); refcount_release(&icl_cxgbei_ncons); } static int icl_cxgbei_setsockopt(struct icl_conn *ic, struct socket *so, int sspace, int rspace) { struct sockopt opt; int error, one = 1, ss, rs; ss = max(sendspace, sspace); rs = max(recvspace, rspace); error = soreserve(so, ss, rs); if (error != 0) { icl_cxgbei_conn_close(ic); return (error); } SOCKBUF_LOCK(&so->so_snd); so->so_snd.sb_flags |= SB_AUTOSIZE; SOCKBUF_UNLOCK(&so->so_snd); SOCKBUF_LOCK(&so->so_rcv); so->so_rcv.sb_flags |= SB_AUTOSIZE; SOCKBUF_UNLOCK(&so->so_rcv); /* * Disable Nagle. */ bzero(&opt, sizeof(opt)); opt.sopt_dir = SOPT_SET; opt.sopt_level = IPPROTO_TCP; opt.sopt_name = TCP_NODELAY; opt.sopt_val = &one; opt.sopt_valsize = sizeof(one); error = sosetopt(so, &opt); if (error != 0) { icl_cxgbei_conn_close(ic); return (error); } return (0); } /* * Request/response structure used to find out the adapter offloading a socket. */ struct find_ofld_adapter_rr { struct socket *so; struct adapter *sc; /* result */ }; static void find_offload_adapter(struct adapter *sc, void *arg) { struct find_ofld_adapter_rr *fa = arg; struct socket *so = fa->so; struct tom_data *td = sc->tom_softc; struct tcpcb *tp; struct inpcb *inp; /* Non-TCP were filtered out earlier. */ MPASS(so->so_proto->pr_protocol == IPPROTO_TCP); if (fa->sc != NULL) return; /* Found already. */ if (td == NULL) return; /* TOE not enabled on this adapter. */ inp = sotoinpcb(so); INP_WLOCK(inp); if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) == 0) { tp = intotcpcb(inp); if (tp->t_flags & TF_TOE && tp->tod == &td->tod) fa->sc = sc; /* Found. */ } INP_WUNLOCK(inp); } /* XXXNP: move this to t4_tom. */ static void send_iscsi_flowc_wr(struct adapter *sc, struct toepcb *toep, int maxlen) { struct wrqe *wr; struct fw_flowc_wr *flowc; const u_int nparams = 1; u_int flowclen; struct ofld_tx_sdesc *txsd = &toep->txsd[toep->txsd_pidx]; flowclen = sizeof(*flowc) + nparams * sizeof(struct fw_flowc_mnemval); wr = alloc_wrqe(roundup2(flowclen, 16), &toep->ofld_txq->wrq); if (wr == NULL) { /* XXX */ panic("%s: allocation failure.", __func__); } flowc = wrtod(wr); memset(flowc, 0, wr->wr_len); flowc->op_to_nparams = htobe32(V_FW_WR_OP(FW_FLOWC_WR) | V_FW_FLOWC_WR_NPARAMS(nparams)); flowc->flowid_len16 = htonl(V_FW_WR_LEN16(howmany(flowclen, 16)) | V_FW_WR_FLOWID(toep->tid)); flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_TXDATAPLEN_MAX; flowc->mnemval[0].val = htobe32(maxlen); txsd->tx_credits = howmany(flowclen, 16); txsd->plen = 0; KASSERT(toep->tx_credits >= txsd->tx_credits && toep->txsd_avail > 0, ("%s: not enough credits (%d)", __func__, toep->tx_credits)); toep->tx_credits -= txsd->tx_credits; if (__predict_false(++toep->txsd_pidx == toep->txsd_total)) toep->txsd_pidx = 0; toep->txsd_avail--; t4_wrq_tx(sc, wr); } static void set_ulp_mode_iscsi(struct adapter *sc, struct toepcb *toep, u_int ulp_submode) { uint64_t val; CTR3(KTR_CXGBE, "%s: tid %u, ULP_MODE_ISCSI, submode=%#x", __func__, toep->tid, ulp_submode); val = V_TCB_ULP_TYPE(ULP_MODE_ISCSI) | V_TCB_ULP_RAW(ulp_submode); t4_set_tcb_field(sc, toep->ctrlq, toep, W_TCB_ULP_TYPE, V_TCB_ULP_TYPE(M_TCB_ULP_TYPE) | V_TCB_ULP_RAW(M_TCB_ULP_RAW), val, 0, 0); val = V_TF_RX_FLOW_CONTROL_DISABLE(1ULL); t4_set_tcb_field(sc, toep->ctrlq, toep, W_TCB_T_FLAGS, val, val, 0, 0); } /* * XXXNP: Who is responsible for cleaning up the socket if this returns with an * error? Review all error paths. * * XXXNP: What happens to the socket's fd reference if the operation is * successful, and how does that affect the socket's life cycle? */ int icl_cxgbei_conn_handoff(struct icl_conn *ic, int fd) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct cxgbei_data *ci; struct find_ofld_adapter_rr fa; struct file *fp; struct socket *so; struct inpcb *inp; struct tcpcb *tp; struct toepcb *toep; cap_rights_t rights; u_int max_rx_pdu_len, max_tx_pdu_len; int error, max_iso_pdus; MPASS(icc->icc_signature == CXGBEI_CONN_SIGNATURE); ICL_CONN_LOCK_ASSERT_NOT(ic); /* * Steal the socket from userland. */ error = fget(curthread, fd, cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp); if (error != 0) return (error); if (fp->f_type != DTYPE_SOCKET) { fdrop(fp, curthread); return (EINVAL); } so = fp->f_data; if (so->so_type != SOCK_STREAM || so->so_proto->pr_protocol != IPPROTO_TCP) { fdrop(fp, curthread); return (EINVAL); } ICL_CONN_LOCK(ic); if (ic->ic_socket != NULL) { ICL_CONN_UNLOCK(ic); fdrop(fp, curthread); return (EBUSY); } ic->ic_disconnecting = false; ic->ic_socket = so; fp->f_ops = &badfileops; fp->f_data = NULL; fdrop(fp, curthread); ICL_CONN_UNLOCK(ic); /* Find the adapter offloading this socket. */ fa.sc = NULL; fa.so = so; t4_iterate(find_offload_adapter, &fa); if (fa.sc == NULL) return (EINVAL); icc->sc = fa.sc; ci = icc->sc->iscsi_ulp_softc; max_rx_pdu_len = ISCSI_BHS_SIZE + ic->ic_max_recv_data_segment_length; max_tx_pdu_len = ISCSI_BHS_SIZE + ic->ic_max_send_data_segment_length; if (ic->ic_header_crc32c) { max_rx_pdu_len += ISCSI_HEADER_DIGEST_SIZE; max_tx_pdu_len += ISCSI_HEADER_DIGEST_SIZE; } if (ic->ic_data_crc32c) { max_rx_pdu_len += ISCSI_DATA_DIGEST_SIZE; max_tx_pdu_len += ISCSI_DATA_DIGEST_SIZE; } inp = sotoinpcb(so); INP_WLOCK(inp); tp = intotcpcb(inp); if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) error = EBUSY; else { /* * socket could not have been "unoffloaded" if here. */ MPASS(tp->t_flags & TF_TOE); MPASS(tp->tod != NULL); MPASS(tp->t_toe != NULL); toep = tp->t_toe; MPASS(toep->vi->adapter == icc->sc); icc->toep = toep; icc->cwt = cxgbei_select_worker_thread(icc); icc->ulp_submode = 0; if (ic->ic_header_crc32c) icc->ulp_submode |= ULP_CRC_HEADER; if (ic->ic_data_crc32c) icc->ulp_submode |= ULP_CRC_DATA; if (icc->sc->tt.iso && chip_id(icc->sc) >= CHELSIO_T5) { max_iso_pdus = CXGBEI_MAX_ISO_PAYLOAD / max_tx_pdu_len; ic->ic_hw_isomax = max_iso_pdus * ic->ic_max_send_data_segment_length; } else max_iso_pdus = 1; so->so_options |= SO_NO_DDP; toep->params.ulp_mode = ULP_MODE_ISCSI; toep->ulpcb = icc; send_iscsi_flowc_wr(icc->sc, toep, roundup(max_iso_pdus * max_tx_pdu_len, tp->t_maxseg)); set_ulp_mode_iscsi(icc->sc, toep, icc->ulp_submode); error = 0; } INP_WUNLOCK(inp); if (error == 0) { error = icl_cxgbei_setsockopt(ic, so, max_tx_pdu_len, max_rx_pdu_len); } return (error); } void icl_cxgbei_conn_close(struct icl_conn *ic) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct icl_pdu *ip; struct socket *so; struct sockbuf *sb; struct inpcb *inp; struct toepcb *toep = icc->toep; MPASS(icc->icc_signature == CXGBEI_CONN_SIGNATURE); ICL_CONN_LOCK_ASSERT_NOT(ic); ICL_CONN_LOCK(ic); so = ic->ic_socket; if (ic->ic_disconnecting || so == NULL) { CTR4(KTR_CXGBE, "%s: icc %p (disconnecting = %d), so %p", __func__, icc, ic->ic_disconnecting, so); ICL_CONN_UNLOCK(ic); return; } ic->ic_disconnecting = true; #ifdef DIAGNOSTIC KASSERT(ic->ic_outstanding_pdus == 0, ("destroying session with %d outstanding PDUs", ic->ic_outstanding_pdus)); #endif ICL_CONN_UNLOCK(ic); CTR3(KTR_CXGBE, "%s: tid %d, icc %p", __func__, toep ? toep->tid : -1, icc); inp = sotoinpcb(so); sb = &so->so_rcv; INP_WLOCK(inp); if (toep != NULL) { /* NULL if connection was never offloaded. */ toep->ulpcb = NULL; /* Discard PDUs queued for TX. */ mbufq_drain(&toep->ulp_pduq); /* * Wait for the cwt threads to stop processing this * connection. */ SOCKBUF_LOCK(sb); if (icc->rx_flags & RXF_ACTIVE) { volatile u_int *p = &icc->rx_flags; SOCKBUF_UNLOCK(sb); INP_WUNLOCK(inp); while (*p & RXF_ACTIVE) pause("conclo", 1); INP_WLOCK(inp); SOCKBUF_LOCK(sb); } /* * Discard received PDUs not passed to the iSCSI * layer. */ while (!STAILQ_EMPTY(&icc->rcvd_pdus)) { ip = STAILQ_FIRST(&icc->rcvd_pdus); STAILQ_REMOVE_HEAD(&icc->rcvd_pdus, ip_next); icl_cxgbei_pdu_done(ip, ENOTCONN); } SOCKBUF_UNLOCK(sb); /* * Grab a reference to use when waiting for the final * CPL to be received. If toep->inp is NULL, then * final_cpl_received() has already been called (e.g. * due to the peer sending a RST). */ if (toep->inp != NULL) { toep = hold_toepcb(toep); toep->flags |= TPF_WAITING_FOR_FINAL; } else toep = NULL; } INP_WUNLOCK(inp); ICL_CONN_LOCK(ic); ic->ic_socket = NULL; ICL_CONN_UNLOCK(ic); /* * XXXNP: we should send RST instead of FIN when PDUs held in various * queues were purged instead of delivered reliably but soabort isn't * really general purpose and wouldn't do the right thing here. */ soclose(so); /* * Wait for the socket to fully close. This ensures any * pending received data has been received (and in particular, * any data that would be received by DDP has been handled). * Callers assume that it is safe to free buffers for tasks * and transfers after this function returns. */ if (toep != NULL) { struct mtx *lock = mtx_pool_find(mtxpool_sleep, toep); mtx_lock(lock); while ((toep->flags & TPF_WAITING_FOR_FINAL) != 0) mtx_sleep(toep, lock, PSOCK, "conclo2", 0); mtx_unlock(lock); free_toepcb(toep); } } static void cxgbei_insert_cmp(struct icl_cxgbei_conn *icc, struct cxgbei_cmp *cmp, uint32_t tt) { #ifdef INVARIANTS struct cxgbei_cmp *cmp2; #endif cmp->tt = tt; mtx_lock(&icc->cmp_lock); #ifdef INVARIANTS LIST_FOREACH(cmp2, &icc->cmp_table[TT_HASH(icc, tt)], link) { KASSERT(cmp2->tt != tt, ("%s: duplicate cmp", __func__)); } #endif LIST_INSERT_HEAD(&icc->cmp_table[TT_HASH(icc, tt)], cmp, link); mtx_unlock(&icc->cmp_lock); } struct cxgbei_cmp * cxgbei_find_cmp(struct icl_cxgbei_conn *icc, uint32_t tt) { struct cxgbei_cmp *cmp; mtx_lock(&icc->cmp_lock); LIST_FOREACH(cmp, &icc->cmp_table[TT_HASH(icc, tt)], link) { if (cmp->tt == tt) break; } mtx_unlock(&icc->cmp_lock); return (cmp); } static void cxgbei_rm_cmp(struct icl_cxgbei_conn *icc, struct cxgbei_cmp *cmp) { #ifdef INVARIANTS struct cxgbei_cmp *cmp2; #endif mtx_lock(&icc->cmp_lock); #ifdef INVARIANTS LIST_FOREACH(cmp2, &icc->cmp_table[TT_HASH(icc, cmp->tt)], link) { if (cmp2 == cmp) goto found; } panic("%s: could not find cmp", __func__); found: #endif LIST_REMOVE(cmp, link); mtx_unlock(&icc->cmp_lock); } int icl_cxgbei_conn_task_setup(struct icl_conn *ic, struct icl_pdu *ip, struct ccb_scsiio *csio, uint32_t *ittp, void **arg) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct toepcb *toep = icc->toep; struct adapter *sc = icc->sc; struct cxgbei_data *ci = sc->iscsi_ulp_softc; struct ppod_region *pr = &ci->pr; struct cxgbei_ddp_state *ddp; struct ppod_reservation *prsv; struct inpcb *inp; struct mbufq mq; uint32_t itt; int rc = 0; ICL_CONN_LOCK_ASSERT(ic); /* This is for the offload driver's state. Must not be set already. */ MPASS(arg != NULL); MPASS(*arg == NULL); if (ic->ic_disconnecting || ic->ic_socket == NULL) return (ECONNRESET); if ((csio->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_IN || csio->dxfer_len < ci->ddp_threshold) { no_ddp: /* * No DDP for this I/O. Allocate an ITT (based on the one * passed in) that cannot be a valid hardware DDP tag in the * iSCSI region. */ itt = *ittp & M_PPOD_TAG; itt = V_PPOD_TAG(itt) | pr->pr_invalid_bit; *ittp = htobe32(itt); MPASS(*arg == NULL); /* State is maintained for DDP only. */ if (rc != 0) counter_u64_add( toep->ofld_rxq->rx_iscsi_ddp_setup_error, 1); return (0); } /* * Reserve resources for DDP, update the itt that should be used in the * PDU, and save DDP specific state for this I/O in *arg. */ ddp = malloc(sizeof(*ddp), M_CXGBEI, M_NOWAIT | M_ZERO); if (ddp == NULL) { rc = ENOMEM; goto no_ddp; } prsv = &ddp->prsv; /* XXX add support for all CAM_DATA_ types */ MPASS((csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR); rc = t4_alloc_page_pods_for_buf(pr, (vm_offset_t)csio->data_ptr, csio->dxfer_len, prsv); if (rc != 0) { free(ddp, M_CXGBEI); goto no_ddp; } mbufq_init(&mq, INT_MAX); rc = t4_write_page_pods_for_buf(sc, toep, prsv, (vm_offset_t)csio->data_ptr, csio->dxfer_len, &mq); if (__predict_false(rc != 0)) { mbufq_drain(&mq); t4_free_page_pods(prsv); free(ddp, M_CXGBEI); goto no_ddp; } /* * Do not get inp from toep->inp as the toepcb might have * detached already. */ inp = sotoinpcb(ic->ic_socket); INP_WLOCK(inp); if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) != 0) { INP_WUNLOCK(inp); mbufq_drain(&mq); t4_free_page_pods(prsv); free(ddp, M_CXGBEI); return (ECONNRESET); } mbufq_concat(&toep->ulp_pduq, &mq); INP_WUNLOCK(inp); ddp->cmp.last_datasn = -1; cxgbei_insert_cmp(icc, &ddp->cmp, prsv->prsv_tag); *ittp = htobe32(prsv->prsv_tag); *arg = prsv; counter_u64_add(toep->ofld_rxq->rx_iscsi_ddp_setup_ok, 1); return (0); } void icl_cxgbei_conn_task_done(struct icl_conn *ic, void *arg) { if (arg != NULL) { struct cxgbei_ddp_state *ddp = arg; cxgbei_rm_cmp(ic_to_icc(ic), &ddp->cmp); t4_free_page_pods(&ddp->prsv); free(ddp, M_CXGBEI); } } static inline bool ddp_sgl_check(struct ctl_sg_entry *sg, int entries, int xferlen) { int total_len = 0; MPASS(entries > 0); if (((vm_offset_t)sg[--entries].addr & 3U) != 0) return (false); total_len += sg[entries].len; while (--entries >= 0) { if (((vm_offset_t)sg[entries].addr & PAGE_MASK) != 0 || (sg[entries].len % PAGE_SIZE) != 0) return (false); total_len += sg[entries].len; } MPASS(total_len == xferlen); return (true); } /* XXXNP: PDU should be passed in as parameter, like on the initiator. */ #define io_to_request_pdu(io) ((io)->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr) #define io_to_ddp_state(io) ((io)->io_hdr.ctl_private[CTL_PRIV_FRONTEND2].ptr) int icl_cxgbei_conn_transfer_setup(struct icl_conn *ic, union ctl_io *io, uint32_t *tttp, void **arg) { struct icl_cxgbei_conn *icc = ic_to_icc(ic); struct toepcb *toep = icc->toep; struct ctl_scsiio *ctsio = &io->scsiio; struct adapter *sc = icc->sc; struct cxgbei_data *ci = sc->iscsi_ulp_softc; struct ppod_region *pr = &ci->pr; struct cxgbei_ddp_state *ddp; struct ppod_reservation *prsv; struct ctl_sg_entry *sgl, sg_entry; struct inpcb *inp; struct mbufq mq; int sg_entries = ctsio->kern_sg_entries; uint32_t ttt; int xferlen, rc = 0, alias; /* This is for the offload driver's state. Must not be set already. */ MPASS(arg != NULL); MPASS(*arg == NULL); if (ctsio->ext_data_filled == 0) { int first_burst; struct icl_pdu *ip = io_to_request_pdu(io); #ifdef INVARIANTS struct icl_cxgbei_pdu *icp = ip_to_icp(ip); MPASS(icp->icp_signature == CXGBEI_PDU_SIGNATURE); MPASS(ic == ip->ip_conn); MPASS(ip->ip_bhs_mbuf != NULL); #endif first_burst = icl_pdu_data_segment_length(ip); /* * Note that ICL calls conn_transfer_setup even if the first * burst had everything and there's nothing left to transfer. * * NB: The CTL frontend might have provided a buffer * whose length (kern_data_len) is smaller than the * FirstBurstLength of unsolicited data. Treat those * as an empty transfer. */ xferlen = ctsio->kern_data_len; if (xferlen < first_burst || xferlen - first_burst < ci->ddp_threshold) { no_ddp: /* * No DDP for this transfer. Allocate a TTT (based on * the one passed in) that cannot be a valid hardware * DDP tag in the iSCSI region. */ ttt = *tttp & M_PPOD_TAG; ttt = V_PPOD_TAG(ttt) | pr->pr_invalid_bit; *tttp = htobe32(ttt); MPASS(io_to_ddp_state(io) == NULL); if (rc != 0) counter_u64_add( toep->ofld_rxq->rx_iscsi_ddp_setup_error, 1); return (0); } if (sg_entries == 0) { sgl = &sg_entry; sgl->len = xferlen; sgl->addr = (void *)ctsio->kern_data_ptr; sg_entries = 1; } else sgl = (void *)ctsio->kern_data_ptr; if (!ddp_sgl_check(sgl, sg_entries, xferlen)) goto no_ddp; /* * Reserve resources for DDP, update the ttt that should be used * in the PDU, and save DDP specific state for this I/O. */ MPASS(io_to_ddp_state(io) == NULL); ddp = malloc(sizeof(*ddp), M_CXGBEI, M_NOWAIT | M_ZERO); if (ddp == NULL) { rc = ENOMEM; goto no_ddp; } prsv = &ddp->prsv; rc = t4_alloc_page_pods_for_sgl(pr, sgl, sg_entries, prsv); if (rc != 0) { free(ddp, M_CXGBEI); goto no_ddp; } mbufq_init(&mq, INT_MAX); rc = t4_write_page_pods_for_sgl(sc, toep, prsv, sgl, sg_entries, xferlen, &mq); if (__predict_false(rc != 0)) { mbufq_drain(&mq); t4_free_page_pods(prsv); free(ddp, M_CXGBEI); goto no_ddp; } /* * Do not get inp from toep->inp as the toepcb might * have detached already. */ ICL_CONN_LOCK(ic); if (ic->ic_disconnecting || ic->ic_socket == NULL) { ICL_CONN_UNLOCK(ic); mbufq_drain(&mq); t4_free_page_pods(prsv); free(ddp, M_CXGBEI); return (ECONNRESET); } inp = sotoinpcb(ic->ic_socket); INP_WLOCK(inp); ICL_CONN_UNLOCK(ic); if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) != 0) { INP_WUNLOCK(inp); mbufq_drain(&mq); t4_free_page_pods(prsv); free(ddp, M_CXGBEI); return (ECONNRESET); } mbufq_concat(&toep->ulp_pduq, &mq); INP_WUNLOCK(inp); ddp->cmp.next_buffer_offset = ctsio->kern_rel_offset + first_burst; ddp->cmp.last_datasn = -1; cxgbei_insert_cmp(icc, &ddp->cmp, prsv->prsv_tag); *tttp = htobe32(prsv->prsv_tag); io_to_ddp_state(io) = ddp; *arg = ctsio; counter_u64_add(toep->ofld_rxq->rx_iscsi_ddp_setup_ok, 1); return (0); } /* * In the middle of an I/O. A non-NULL page pod reservation indicates * that a DDP buffer is being used for the I/O. */ ddp = io_to_ddp_state(ctsio); if (ddp == NULL) goto no_ddp; prsv = &ddp->prsv; alias = (prsv->prsv_tag & pr->pr_alias_mask) >> pr->pr_alias_shift; alias++; prsv->prsv_tag &= ~pr->pr_alias_mask; prsv->prsv_tag |= alias << pr->pr_alias_shift & pr->pr_alias_mask; - ddp->cmp.next_datasn = 0; ddp->cmp.last_datasn = -1; cxgbei_insert_cmp(icc, &ddp->cmp, prsv->prsv_tag); *tttp = htobe32(prsv->prsv_tag); *arg = ctsio; return (0); } void icl_cxgbei_conn_transfer_done(struct icl_conn *ic, void *arg) { struct ctl_scsiio *ctsio = arg; if (ctsio != NULL) { struct cxgbei_ddp_state *ddp; ddp = io_to_ddp_state(ctsio); MPASS(ddp != NULL); cxgbei_rm_cmp(ic_to_icc(ic), &ddp->cmp); if (ctsio->kern_data_len == ctsio->ext_data_filled || ic->ic_disconnecting) { t4_free_page_pods(&ddp->prsv); free(ddp, M_CXGBEI); io_to_ddp_state(ctsio) = NULL; } } } static void cxgbei_limits(struct adapter *sc, void *arg) { struct icl_drv_limits *idl = arg; struct cxgbei_data *ci; int max_dsl; if (begin_synchronized_op(sc, NULL, HOLD_LOCK, "t4lims") != 0) return; if (uld_active(sc, ULD_ISCSI)) { ci = sc->iscsi_ulp_softc; MPASS(ci != NULL); max_dsl = ci->max_rx_data_len; if (idl->idl_max_recv_data_segment_length > max_dsl) idl->idl_max_recv_data_segment_length = max_dsl; max_dsl = ci->max_tx_data_len; if (idl->idl_max_send_data_segment_length > max_dsl) idl->idl_max_send_data_segment_length = max_dsl; } end_synchronized_op(sc, LOCK_HELD); } static int icl_cxgbei_limits(struct icl_drv_limits *idl) { /* Maximum allowed by the RFC. cxgbei_limits will clip them. */ idl->idl_max_recv_data_segment_length = (1 << 24) - 1; idl->idl_max_send_data_segment_length = (1 << 24) - 1; /* These are somewhat arbitrary. */ idl->idl_max_burst_length = max_burst_length; idl->idl_first_burst_length = first_burst_length; t4_iterate(cxgbei_limits, idl); return (0); } int icl_cxgbei_mod_load(void) { int rc; refcount_init(&icl_cxgbei_ncons, 0); rc = icl_register("cxgbei", false, -100, icl_cxgbei_limits, icl_cxgbei_new_conn); return (rc); } int icl_cxgbei_mod_unload(void) { if (icl_cxgbei_ncons != 0) return (EBUSY); icl_unregister("cxgbei", false); return (0); } #endif diff --git a/sys/dev/iscsi/icl.h b/sys/dev/iscsi/icl.h index bd12569a8957..07dcbbf2a0b5 100644 --- a/sys/dev/iscsi/icl.h +++ b/sys/dev/iscsi/icl.h @@ -1,159 +1,167 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 The FreeBSD Foundation * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef ICL_H #define ICL_H /* * iSCSI Common Layer. It's used by both the initiator and target to send * and receive iSCSI PDUs. */ #include #include #include #include SYSCTL_DECL(_kern_icl); extern int icl_debug; #define ICL_DEBUG(X, ...) \ do { \ if (icl_debug > 1) \ printf("%s: " X "\n", __func__, ## __VA_ARGS__);\ } while (0) #define ICL_WARN(X, ...) \ do { \ if (icl_debug > 0) { \ printf("WARNING: %s: " X "\n", \ __func__, ## __VA_ARGS__); \ } \ } while (0) struct icl_conn; struct ccb_scsiio; union ctl_io; struct icl_pdu { STAILQ_ENTRY(icl_pdu) ip_next; struct icl_conn *ip_conn; struct iscsi_bhs *ip_bhs; struct mbuf *ip_bhs_mbuf; size_t ip_ahs_len; struct mbuf *ip_ahs_mbuf; size_t ip_data_len; struct mbuf *ip_data_mbuf; + /* + * When a "large" received PDU represents multiple on-the-wire + * PDUs, this is the count of additional on-the-wire PDUs. + * For PDUs that match on-the-wire PDUs, this should be set to + * zero. + */ + u_int ip_additional_pdus; + /* * User (initiator or provider) private fields. */ void *ip_prv0; void *ip_prv1; }; #define ICL_NOCOPY (1 << 30) struct icl_conn { KOBJ_FIELDS; struct mtx *ic_lock; struct socket *ic_socket; #ifdef DIAGNOSTIC volatile u_int ic_outstanding_pdus; #endif uint32_t ic_max_recv_data_segment_length; uint32_t ic_max_send_data_segment_length; size_t ic_hw_isomax; size_t ic_maxtags; bool ic_header_crc32c; bool ic_data_crc32c; bool ic_disconnecting; bool ic_iser; bool ic_unmapped; const char *ic_name; const char *ic_offload; void (*ic_receive)(struct icl_pdu *); void (*ic_error)(struct icl_conn *); /* * User (initiator or provider) private fields. */ void *ic_prv0; }; struct icl_drv_limits { int idl_max_recv_data_segment_length; int idl_max_send_data_segment_length; int idl_max_burst_length; int idl_first_burst_length; int spare[4]; }; typedef void (*icl_pdu_cb)(struct icl_pdu *, int error); struct icl_conn *icl_new_conn(const char *offload, bool iser, const char *name, struct mtx *lock); int icl_limits(const char *offload, bool iser, struct icl_drv_limits *idl); int icl_register(const char *offload, bool iser, int priority, int (*limits)(struct icl_drv_limits *), struct icl_conn *(*new_conn)(const char *, struct mtx *)); int icl_unregister(const char *offload, bool rdma); #ifdef ICL_KERNEL_PROXY struct sockaddr; struct icl_listen; /* * Target part. */ struct icl_listen *icl_listen_new(void (*accept_cb)(struct socket *, struct sockaddr *, int)); void icl_listen_free(struct icl_listen *il); int icl_listen_add(struct icl_listen *il, bool rdma, int domain, int socktype, int protocol, struct sockaddr *sa, int portal_id); int icl_listen_remove(struct icl_listen *il, struct sockaddr *sa); /* * Those two are not a public API; only to be used between icl_soft.c * and icl_soft_proxy.c. */ int icl_soft_handoff_sock(struct icl_conn *ic, struct socket *so); int icl_soft_proxy_connect(struct icl_conn *ic, int domain, int socktype, int protocol, struct sockaddr *from_sa, struct sockaddr *to_sa); #endif /* ICL_KERNEL_PROXY */ #endif /* !ICL_H */