From a4c15cd957cbd728f685645de7a150df5912591a Mon Sep 17 00:00:00 2001 From: Jenny Derzhavetz Date: Sun, 6 Sep 2015 14:52:20 +0300 Subject: iser-target: remove command with state ISTATE_REMOVE As documented in iscsit_sequence_cmd: /* * Existing callers for iscsit_sequence_cmd() will silently * ignore commands with CMDSN_LOWER_THAN_EXP, so force this * return for CMDSN_MAXCMDSN_OVERRUN as well.. */ We need to silently finish a command when it's in ISTATE_REMOVE. This fixes an teardown hang we were seeing where a mis-behaved initiator (triggered by allocation error injections) sent us a cmdsn which was lower than expected. Signed-off-by: Jenny Derzhavetz Signed-off-by: Sagi Grimberg Cc: # v3.10+ Signed-off-by: Nicholas Bellinger --- drivers/infiniband/ulp/isert/ib_isert.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index 403bd29443b8..443ef33fbf30 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -2999,9 +2999,16 @@ isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery) static int isert_immediate_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state) { - int ret; + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); + int ret = 0; switch (state) { + case ISTATE_REMOVE: + spin_lock_bh(&conn->cmd_lock); + list_del_init(&cmd->i_conn_node); + spin_unlock_bh(&conn->cmd_lock); + isert_put_cmd(isert_cmd, true); + break; case ISTATE_SEND_NOPIN_WANT_RESPONSE: ret = isert_put_nopin(cmd, conn, false); break; -- cgit v1.2.3 From 3e03c4b01da3e6a5f3081eb0aa252490fe83e352 Mon Sep 17 00:00:00 2001 From: Jenny Derzhavetz Date: Sun, 6 Sep 2015 14:52:21 +0300 Subject: iser-target: Put the reference on commands waiting for unsol data The iscsi target core teardown sequence calls wait_conn for all active commands to finish gracefully by: - move the queue-pair to error state - drain all the completions - wait for the core to finish handling all session commands However, when tearing down a session while there are sequenced commands that are still waiting for unsolicited data outs, we can block forever as these are missing an extra reference put. We basically need the equivalent of iscsit_free_queue_reqs_for_conn() which is called after wait_conn has returned. Address this by an explicit walk on conn_cmd_list and put the extra reference. Signed-off-by: Jenny Derzhavetz Signed-off-by: Sagi Grimberg Cc: # v3.10+ Signed-off-by: Nicholas Bellinger --- drivers/infiniband/ulp/isert/ib_isert.c | 38 ++++++++++++++++++++++++++++++++- 1 file changed, 37 insertions(+), 1 deletion(-) diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index 443ef33fbf30..c8749a9e1ccd 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -3352,6 +3352,41 @@ isert_wait4flush(struct isert_conn *isert_conn) wait_for_completion(&isert_conn->wait_comp_err); } +/** + * isert_put_unsol_pending_cmds() - Drop commands waiting for + * unsolicitate dataout + * @conn: iscsi connection + * + * We might still have commands that are waiting for unsolicited + * dataouts messages. We must put the extra reference on those + * before blocking on the target_wait_for_session_cmds + */ +static void +isert_put_unsol_pending_cmds(struct iscsi_conn *conn) +{ + struct iscsi_cmd *cmd, *tmp; + static LIST_HEAD(drop_cmd_list); + + spin_lock_bh(&conn->cmd_lock); + list_for_each_entry_safe(cmd, tmp, &conn->conn_cmd_list, i_conn_node) { + if ((cmd->cmd_flags & ICF_NON_IMMEDIATE_UNSOLICITED_DATA) && + (cmd->write_data_done < conn->sess->sess_ops->FirstBurstLength) && + (cmd->write_data_done < cmd->se_cmd.data_length)) + list_move_tail(&cmd->i_conn_node, &drop_cmd_list); + } + spin_unlock_bh(&conn->cmd_lock); + + list_for_each_entry_safe(cmd, tmp, &drop_cmd_list, i_conn_node) { + list_del_init(&cmd->i_conn_node); + if (cmd->i_state != ISTATE_REMOVE) { + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); + + isert_info("conn %p dropping cmd %p\n", conn, cmd); + isert_put_cmd(isert_cmd, true); + } + } +} + static void isert_wait_conn(struct iscsi_conn *conn) { struct isert_conn *isert_conn = conn->context; @@ -3370,8 +3405,9 @@ static void isert_wait_conn(struct iscsi_conn *conn) isert_conn_terminate(isert_conn); mutex_unlock(&isert_conn->mutex); - isert_wait4cmds(conn); isert_wait4flush(isert_conn); + isert_put_unsol_pending_cmds(conn); + isert_wait4cmds(conn); isert_wait4logout(isert_conn); queue_work(isert_release_wq, &isert_conn->release_work); -- cgit v1.2.3 From f27dfa1f0eb91494577a395e6b8a9aac2832e9cf Mon Sep 17 00:00:00 2001 From: Jenny Derzhavetz Date: Sun, 6 Sep 2015 14:52:22 +0300 Subject: iser-target: Remove unused variables Signed-off-by: Jenny Derzhavetz Signed-off-by: Sagi Grimberg Signed-off-by: Nicholas Bellinger --- drivers/infiniband/ulp/isert/ib_isert.c | 1 - drivers/infiniband/ulp/isert/ib_isert.h | 5 ----- 2 files changed, 6 deletions(-) diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index c8749a9e1ccd..97475b2eecb7 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -3116,7 +3116,6 @@ isert_setup_np(struct iscsi_np *np, sema_init(&isert_np->np_sem, 0); mutex_init(&isert_np->np_accept_mutex); INIT_LIST_HEAD(&isert_np->np_accept_list); - init_completion(&isert_np->np_login_comp); isert_np->np = np; /* diff --git a/drivers/infiniband/ulp/isert/ib_isert.h b/drivers/infiniband/ulp/isert/ib_isert.h index 6a04ba3c0f72..0480ab89ea40 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.h +++ b/drivers/infiniband/ulp/isert/ib_isert.h @@ -113,7 +113,6 @@ enum { }; struct isert_rdma_wr { - struct list_head wr_list; struct isert_cmd *isert_cmd; enum iser_ib_op_code iser_ib_op; struct ib_sge *ib_sge; @@ -134,9 +133,6 @@ struct isert_cmd { uint64_t write_va; u64 pdu_buf_dma; u32 pdu_buf_len; - u32 read_va_off; - u32 write_va_off; - u32 rdma_wr_num; struct isert_conn *conn; struct iscsi_cmd *iscsi_cmd; struct iser_tx_desc tx_desc; @@ -226,5 +222,4 @@ struct isert_np { struct rdma_cm_id *np_cm_id; struct mutex np_accept_mutex; struct list_head np_accept_list; - struct completion np_login_comp; }; -- cgit v1.2.3 From ed8cb0a4378c0e7035db047987fe0e7309020ab5 Mon Sep 17 00:00:00 2001 From: Jenny Derzhavetz Date: Sun, 6 Sep 2015 14:52:23 +0300 Subject: iser-target: Remove np_ prefix from isert_np members These are always referenced from np-> so no need for the prefix. Signed-off-by: Jenny Derzhavetz Signed-off-by: Sagi Grimberg Signed-off-by: Nicholas Bellinger --- drivers/infiniband/ulp/isert/ib_isert.c | 58 ++++++++++++++++----------------- drivers/infiniband/ulp/isert/ib_isert.h | 8 ++--- 2 files changed, 33 insertions(+), 33 deletions(-) diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index 97475b2eecb7..b163ec0ba2eb 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -778,12 +778,12 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) if (ret) goto out_conn_dev; - mutex_lock(&isert_np->np_accept_mutex); - list_add_tail(&isert_conn->accept_node, &isert_np->np_accept_list); - mutex_unlock(&isert_np->np_accept_mutex); + mutex_lock(&isert_np->mutex); + list_add_tail(&isert_conn->accept_node, &isert_np->accept_list); + mutex_unlock(&isert_np->mutex); isert_info("np %p: Allow accept_np to continue\n", np); - up(&isert_np->np_sem); + up(&isert_np->sem); return 0; out_conn_dev: @@ -903,14 +903,14 @@ isert_np_cma_handler(struct isert_np *isert_np, switch (event) { case RDMA_CM_EVENT_DEVICE_REMOVAL: - isert_np->np_cm_id = NULL; + isert_np->cm_id = NULL; break; case RDMA_CM_EVENT_ADDR_CHANGE: - isert_np->np_cm_id = isert_setup_id(isert_np); - if (IS_ERR(isert_np->np_cm_id)) { + isert_np->cm_id = isert_setup_id(isert_np); + if (IS_ERR(isert_np->cm_id)) { isert_err("isert np %p setup id failed: %ld\n", - isert_np, PTR_ERR(isert_np->np_cm_id)); - isert_np->np_cm_id = NULL; + isert_np, PTR_ERR(isert_np->cm_id)); + isert_np->cm_id = NULL; } break; default: @@ -929,7 +929,7 @@ isert_disconnected_handler(struct rdma_cm_id *cma_id, struct isert_conn *isert_conn; bool terminating = false; - if (isert_np->np_cm_id == cma_id) + if (isert_np->cm_id == cma_id) return isert_np_cma_handler(cma_id->context, event); isert_conn = cma_id->qp->qp_context; @@ -945,13 +945,13 @@ isert_disconnected_handler(struct rdma_cm_id *cma_id, if (terminating) goto out; - mutex_lock(&isert_np->np_accept_mutex); + mutex_lock(&isert_np->mutex); if (!list_empty(&isert_conn->accept_node)) { list_del_init(&isert_conn->accept_node); isert_put_conn(isert_conn); queue_work(isert_release_wq, &isert_conn->release_work); } - mutex_unlock(&isert_np->np_accept_mutex); + mutex_unlock(&isert_np->mutex); out: return 0; @@ -3113,9 +3113,9 @@ isert_setup_np(struct iscsi_np *np, isert_err("Unable to allocate struct isert_np\n"); return -ENOMEM; } - sema_init(&isert_np->np_sem, 0); - mutex_init(&isert_np->np_accept_mutex); - INIT_LIST_HEAD(&isert_np->np_accept_list); + sema_init(&isert_np->sem, 0); + mutex_init(&isert_np->mutex); + INIT_LIST_HEAD(&isert_np->accept_list); isert_np->np = np; /* @@ -3131,7 +3131,7 @@ isert_setup_np(struct iscsi_np *np, goto out; } - isert_np->np_cm_id = isert_lid; + isert_np->cm_id = isert_lid; np->np_context = isert_np; return 0; @@ -3220,7 +3220,7 @@ isert_accept_np(struct iscsi_np *np, struct iscsi_conn *conn) int ret; accept_wait: - ret = down_interruptible(&isert_np->np_sem); + ret = down_interruptible(&isert_np->sem); if (ret) return -ENODEV; @@ -3237,15 +3237,15 @@ accept_wait: } spin_unlock_bh(&np->np_thread_lock); - mutex_lock(&isert_np->np_accept_mutex); - if (list_empty(&isert_np->np_accept_list)) { - mutex_unlock(&isert_np->np_accept_mutex); + mutex_lock(&isert_np->mutex); + if (list_empty(&isert_np->accept_list)) { + mutex_unlock(&isert_np->mutex); goto accept_wait; } - isert_conn = list_first_entry(&isert_np->np_accept_list, + isert_conn = list_first_entry(&isert_np->accept_list, struct isert_conn, accept_node); list_del_init(&isert_conn->accept_node); - mutex_unlock(&isert_np->np_accept_mutex); + mutex_unlock(&isert_np->mutex); conn->context = isert_conn; isert_conn->conn = conn; @@ -3263,28 +3263,28 @@ isert_free_np(struct iscsi_np *np) struct isert_np *isert_np = np->np_context; struct isert_conn *isert_conn, *n; - if (isert_np->np_cm_id) - rdma_destroy_id(isert_np->np_cm_id); + if (isert_np->cm_id) + rdma_destroy_id(isert_np->cm_id); /* * FIXME: At this point we don't have a good way to insure * that at this point we don't have hanging connections that * completed RDMA establishment but didn't start iscsi login * process. So work-around this by cleaning up what ever piled - * up in np_accept_list. + * up in accept_list. */ - mutex_lock(&isert_np->np_accept_mutex); - if (!list_empty(&isert_np->np_accept_list)) { + mutex_lock(&isert_np->mutex); + if (!list_empty(&isert_np->accept_list)) { isert_info("Still have isert connections, cleaning up...\n"); list_for_each_entry_safe(isert_conn, n, - &isert_np->np_accept_list, + &isert_np->accept_list, accept_node) { isert_info("cleaning isert_conn %p state (%d)\n", isert_conn, isert_conn->state); isert_connect_release(isert_conn); } } - mutex_unlock(&isert_np->np_accept_mutex); + mutex_unlock(&isert_np->mutex); np->np_context = NULL; kfree(isert_np); diff --git a/drivers/infiniband/ulp/isert/ib_isert.h b/drivers/infiniband/ulp/isert/ib_isert.h index 0480ab89ea40..6e85da3dd835 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.h +++ b/drivers/infiniband/ulp/isert/ib_isert.h @@ -218,8 +218,8 @@ struct isert_device { struct isert_np { struct iscsi_np *np; - struct semaphore np_sem; - struct rdma_cm_id *np_cm_id; - struct mutex np_accept_mutex; - struct list_head np_accept_list; + struct semaphore sem; + struct rdma_cm_id *cm_id; + struct mutex mutex; + struct list_head accept_list; }; -- cgit v1.2.3 From bd3792205aaeb79b994338af2e5499fa503d79c7 Mon Sep 17 00:00:00 2001 From: Jenny Derzhavetz Date: Sun, 6 Sep 2015 14:52:24 +0300 Subject: iser-target: Fix pending connections handling in target stack shutdown sequnce Instead of handing a connection to the iscsi stack for processing right after accepting (rdma_accept) we only hand the connection to the iscsi core after we reached to a connected state (ESTABLISHED CM event). This will prevent two error scenrios: 1. race between rdma connection teardown and iscsi login sequence reported by Nic in: (ce9a9fc20a78a "iser-target: Fix REJECT CM event use-after-free OOPs") 2. target stack shutdown sequence race with constant login attempts by multiple initiators. We address this by maintaining two queues at the isert_np level: - accepted: connections that were accepted but have not reached connected state (might get rejected, unreachable or error). - pending: connections in connected state, but have yet to handed to the iscsi core for login processing. iser connections are promoted to the pending queue only from the accepted queue. This way the iscsi core now will only handle functional iser connections and once we shutdown the target stack, we look for any stales that got left behind so we can safely release them. Signed-off-by: Jenny Derzhavetz Signed-off-by: Sagi Grimberg Cc: # v3.10+ Signed-off-by: Nicholas Bellinger --- drivers/infiniband/ulp/isert/ib_isert.c | 66 ++++++++++++++++++--------------- drivers/infiniband/ulp/isert/ib_isert.h | 5 ++- 2 files changed, 40 insertions(+), 31 deletions(-) diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index b163ec0ba2eb..f3f498f6ecc5 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -634,7 +634,7 @@ static void isert_init_conn(struct isert_conn *isert_conn) { isert_conn->state = ISER_CONN_INIT; - INIT_LIST_HEAD(&isert_conn->accept_node); + INIT_LIST_HEAD(&isert_conn->node); init_completion(&isert_conn->login_comp); init_completion(&isert_conn->login_req_comp); init_completion(&isert_conn->wait); @@ -762,28 +762,15 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) ret = isert_rdma_post_recvl(isert_conn); if (ret) goto out_conn_dev; - /* - * Obtain the second reference now before isert_rdma_accept() to - * ensure that any initiator generated REJECT CM event that occurs - * asynchronously won't drop the last reference until the error path - * in iscsi_target_login_sess_out() does it's ->iscsit_free_conn() -> - * isert_free_conn() -> isert_put_conn() -> kref_put(). - */ - if (!kref_get_unless_zero(&isert_conn->kref)) { - isert_warn("conn %p connect_release is running\n", isert_conn); - goto out_conn_dev; - } ret = isert_rdma_accept(isert_conn); if (ret) goto out_conn_dev; mutex_lock(&isert_np->mutex); - list_add_tail(&isert_conn->accept_node, &isert_np->accept_list); + list_add_tail(&isert_conn->node, &isert_np->accepted); mutex_unlock(&isert_np->mutex); - isert_info("np %p: Allow accept_np to continue\n", np); - up(&isert_np->sem); return 0; out_conn_dev: @@ -831,13 +818,21 @@ static void isert_connected_handler(struct rdma_cm_id *cma_id) { struct isert_conn *isert_conn = cma_id->qp->qp_context; + struct isert_np *isert_np = cma_id->context; isert_info("conn %p\n", isert_conn); mutex_lock(&isert_conn->mutex); - if (isert_conn->state != ISER_CONN_FULL_FEATURE) - isert_conn->state = ISER_CONN_UP; + isert_conn->state = ISER_CONN_UP; + kref_get(&isert_conn->kref); mutex_unlock(&isert_conn->mutex); + + mutex_lock(&isert_np->mutex); + list_move_tail(&isert_conn->node, &isert_np->pending); + mutex_unlock(&isert_np->mutex); + + isert_info("np %p: Allow accept_np to continue\n", isert_np); + up(&isert_np->sem); } static void @@ -946,8 +941,8 @@ isert_disconnected_handler(struct rdma_cm_id *cma_id, goto out; mutex_lock(&isert_np->mutex); - if (!list_empty(&isert_conn->accept_node)) { - list_del_init(&isert_conn->accept_node); + if (!list_empty(&isert_conn->node)) { + list_del_init(&isert_conn->node); isert_put_conn(isert_conn); queue_work(isert_release_wq, &isert_conn->release_work); } @@ -962,6 +957,7 @@ isert_connect_error(struct rdma_cm_id *cma_id) { struct isert_conn *isert_conn = cma_id->qp->qp_context; + list_del_init(&isert_conn->node); isert_conn->cm_id = NULL; isert_put_conn(isert_conn); @@ -3115,7 +3111,8 @@ isert_setup_np(struct iscsi_np *np, } sema_init(&isert_np->sem, 0); mutex_init(&isert_np->mutex); - INIT_LIST_HEAD(&isert_np->accept_list); + INIT_LIST_HEAD(&isert_np->accepted); + INIT_LIST_HEAD(&isert_np->pending); isert_np->np = np; /* @@ -3238,13 +3235,13 @@ accept_wait: spin_unlock_bh(&np->np_thread_lock); mutex_lock(&isert_np->mutex); - if (list_empty(&isert_np->accept_list)) { + if (list_empty(&isert_np->pending)) { mutex_unlock(&isert_np->mutex); goto accept_wait; } - isert_conn = list_first_entry(&isert_np->accept_list, - struct isert_conn, accept_node); - list_del_init(&isert_conn->accept_node); + isert_conn = list_first_entry(&isert_np->pending, + struct isert_conn, node); + list_del_init(&isert_conn->node); mutex_unlock(&isert_np->mutex); conn->context = isert_conn; @@ -3271,14 +3268,25 @@ isert_free_np(struct iscsi_np *np) * that at this point we don't have hanging connections that * completed RDMA establishment but didn't start iscsi login * process. So work-around this by cleaning up what ever piled - * up in accept_list. + * up in accepted and pending lists. */ mutex_lock(&isert_np->mutex); - if (!list_empty(&isert_np->accept_list)) { - isert_info("Still have isert connections, cleaning up...\n"); + if (!list_empty(&isert_np->pending)) { + isert_info("Still have isert pending connections\n"); + list_for_each_entry_safe(isert_conn, n, + &isert_np->pending, + node) { + isert_info("cleaning isert_conn %p state (%d)\n", + isert_conn, isert_conn->state); + isert_connect_release(isert_conn); + } + } + + if (!list_empty(&isert_np->accepted)) { + isert_info("Still have isert accepted connections\n"); list_for_each_entry_safe(isert_conn, n, - &isert_np->accept_list, - accept_node) { + &isert_np->accepted, + node) { isert_info("cleaning isert_conn %p state (%d)\n", isert_conn, isert_conn->state); isert_connect_release(isert_conn); diff --git a/drivers/infiniband/ulp/isert/ib_isert.h b/drivers/infiniband/ulp/isert/ib_isert.h index 6e85da3dd835..b81dfe07ce62 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.h +++ b/drivers/infiniband/ulp/isert/ib_isert.h @@ -159,7 +159,7 @@ struct isert_conn { struct iser_rx_desc *rx_descs; struct ib_recv_wr rx_wr[ISERT_MIN_POSTED_RX]; struct iscsi_conn *conn; - struct list_head accept_node; + struct list_head node; struct completion login_comp; struct completion login_req_comp; struct iser_tx_desc login_tx_desc; @@ -221,5 +221,6 @@ struct isert_np { struct semaphore sem; struct rdma_cm_id *cm_id; struct mutex mutex; - struct list_head accept_list; + struct list_head accepted; + struct list_head pending; }; -- cgit v1.2.3 From 4366b19ca5eb15e63d6640565ade135cf06be91a Mon Sep 17 00:00:00 2001 From: Jenny Derzhavetz Date: Sun, 6 Sep 2015 14:52:25 +0300 Subject: iser-target: Change the recv buffers posting logic iser target batches post recv operations to avoid the overhead of acquiring the recv queue lock and posting a HW doorbell for each command. We change it to be per command in order to support zcopy immediate data for IOs that fits in the 8K transfer boundary (in the next patch). (Fix minor patch fuzz due to ib_mr removal - nab) Signed-off-by: Jenny Derzhavetz Signed-off-by: Sagi Grimberg Signed-off-by: Nicholas Bellinger --- drivers/infiniband/ulp/isert/ib_isert.c | 111 +++++++++++++++++++------------- drivers/infiniband/ulp/isert/ib_isert.h | 4 +- 2 files changed, 67 insertions(+), 48 deletions(-) diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index f3f498f6ecc5..233b8c7a78df 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -238,8 +238,6 @@ isert_alloc_rx_descriptors(struct isert_conn *isert_conn) rx_sg->lkey = device->pd->local_dma_lkey; } - isert_conn->rx_desc_head = 0; - return 0; dma_map_fail: @@ -1002,35 +1000,51 @@ isert_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) } static int -isert_post_recv(struct isert_conn *isert_conn, u32 count) +isert_post_recvm(struct isert_conn *isert_conn, u32 count) { struct ib_recv_wr *rx_wr, *rx_wr_failed; int i, ret; - unsigned int rx_head = isert_conn->rx_desc_head; struct iser_rx_desc *rx_desc; for (rx_wr = isert_conn->rx_wr, i = 0; i < count; i++, rx_wr++) { - rx_desc = &isert_conn->rx_descs[rx_head]; - rx_wr->wr_id = (uintptr_t)rx_desc; - rx_wr->sg_list = &rx_desc->rx_sg; - rx_wr->num_sge = 1; - rx_wr->next = rx_wr + 1; - rx_head = (rx_head + 1) & (ISERT_QP_MAX_RECV_DTOS - 1); + rx_desc = &isert_conn->rx_descs[i]; + rx_wr->wr_id = (uintptr_t)rx_desc; + rx_wr->sg_list = &rx_desc->rx_sg; + rx_wr->num_sge = 1; + rx_wr->next = rx_wr + 1; } - rx_wr--; rx_wr->next = NULL; /* mark end of work requests list */ isert_conn->post_recv_buf_count += count; ret = ib_post_recv(isert_conn->qp, isert_conn->rx_wr, - &rx_wr_failed); + &rx_wr_failed); if (ret) { isert_err("ib_post_recv() failed with ret: %d\n", ret); isert_conn->post_recv_buf_count -= count; - } else { - isert_dbg("Posted %d RX buffers\n", count); - isert_conn->rx_desc_head = rx_head; } + + return ret; +} + +static int +isert_post_recv(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc) +{ + struct ib_recv_wr *rx_wr_failed, rx_wr; + int ret; + + rx_wr.wr_id = (uintptr_t)rx_desc; + rx_wr.sg_list = &rx_desc->rx_sg; + rx_wr.num_sge = 1; + rx_wr.next = NULL; + + isert_conn->post_recv_buf_count++; + ret = ib_post_recv(isert_conn->qp, &rx_wr, &rx_wr_failed); + if (ret) { + isert_err("ib_post_recv() failed with ret: %d\n", ret); + isert_conn->post_recv_buf_count--; + } + return ret; } @@ -1201,7 +1215,8 @@ isert_put_login_tx(struct iscsi_conn *conn, struct iscsi_login *login, if (ret) return ret; - ret = isert_post_recv(isert_conn, ISERT_MIN_POSTED_RX); + ret = isert_post_recvm(isert_conn, + ISERT_QP_MAX_RECV_DTOS); if (ret) return ret; @@ -1274,7 +1289,7 @@ isert_rx_login_req(struct isert_conn *isert_conn) } static struct iscsi_cmd -*isert_allocate_cmd(struct iscsi_conn *conn) +*isert_allocate_cmd(struct iscsi_conn *conn, struct iser_rx_desc *rx_desc) { struct isert_conn *isert_conn = conn->context; struct isert_cmd *isert_cmd; @@ -1288,6 +1303,7 @@ static struct iscsi_cmd isert_cmd = iscsit_priv_cmd(cmd); isert_cmd->conn = isert_conn; isert_cmd->iscsi_cmd = cmd; + isert_cmd->rx_desc = rx_desc; return cmd; } @@ -1403,6 +1419,15 @@ isert_handle_iscsi_dataout(struct isert_conn *isert_conn, if (rc < 0) return rc; + /* + * multiple data-outs on the same command can arrive - + * so post the buffer before hand + */ + rc = isert_post_recv(isert_conn, rx_desc); + if (rc) { + isert_err("ib_post_recv failed with %d\n", rc); + return rc; + } return 0; } @@ -1475,7 +1500,7 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, switch (opcode) { case ISCSI_OP_SCSI_CMD: - cmd = isert_allocate_cmd(conn); + cmd = isert_allocate_cmd(conn, rx_desc); if (!cmd) break; @@ -1489,7 +1514,7 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, rx_desc, (unsigned char *)hdr); break; case ISCSI_OP_NOOP_OUT: - cmd = isert_allocate_cmd(conn); + cmd = isert_allocate_cmd(conn, rx_desc); if (!cmd) break; @@ -1502,7 +1527,7 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, (unsigned char *)hdr); break; case ISCSI_OP_SCSI_TMFUNC: - cmd = isert_allocate_cmd(conn); + cmd = isert_allocate_cmd(conn, rx_desc); if (!cmd) break; @@ -1510,22 +1535,20 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, (unsigned char *)hdr); break; case ISCSI_OP_LOGOUT: - cmd = isert_allocate_cmd(conn); + cmd = isert_allocate_cmd(conn, rx_desc); if (!cmd) break; ret = iscsit_handle_logout_cmd(conn, cmd, (unsigned char *)hdr); break; case ISCSI_OP_TEXT: - if (be32_to_cpu(hdr->ttt) != 0xFFFFFFFF) { + if (be32_to_cpu(hdr->ttt) != 0xFFFFFFFF) cmd = iscsit_find_cmd_from_itt(conn, hdr->itt); - if (!cmd) - break; - } else { - cmd = isert_allocate_cmd(conn); - if (!cmd) - break; - } + else + cmd = isert_allocate_cmd(conn, rx_desc); + + if (!cmd) + break; isert_cmd = iscsit_priv_cmd(cmd); ret = isert_handle_text_cmd(isert_conn, isert_cmd, cmd, @@ -1585,7 +1608,7 @@ isert_rcv_completion(struct iser_rx_desc *desc, struct ib_device *ib_dev = isert_conn->cm_id->device; struct iscsi_hdr *hdr; u64 rx_dma; - int rx_buflen, outstanding; + int rx_buflen; if ((char *)desc == isert_conn->login_req_buf) { rx_dma = isert_conn->login_req_dma; @@ -1625,22 +1648,6 @@ isert_rcv_completion(struct iser_rx_desc *desc, DMA_FROM_DEVICE); isert_conn->post_recv_buf_count--; - isert_dbg("Decremented post_recv_buf_count: %d\n", - isert_conn->post_recv_buf_count); - - if ((char *)desc == isert_conn->login_req_buf) - return; - - outstanding = isert_conn->post_recv_buf_count; - if (outstanding + ISERT_MIN_POSTED_RX <= ISERT_QP_MAX_RECV_DTOS) { - int err, count = min(ISERT_QP_MAX_RECV_DTOS - outstanding, - ISERT_MIN_POSTED_RX); - err = isert_post_recv(isert_conn, count); - if (err) { - isert_err("isert_post_recv() count: %d failed, %d\n", - count, err); - } - } } static int @@ -2152,6 +2159,12 @@ isert_post_response(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd) struct ib_send_wr *wr_failed; int ret; + ret = isert_post_recv(isert_conn, isert_cmd->rx_desc); + if (ret) { + isert_err("ib_post_recv failed with %d\n", ret); + return ret; + } + ret = ib_post_send(isert_conn->qp, &isert_cmd->tx_desc.send_wr, &wr_failed); if (ret) { @@ -2946,6 +2959,12 @@ isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd) &isert_cmd->tx_desc.send_wr); isert_cmd->rdma_wr.s_send_wr.next = &isert_cmd->tx_desc.send_wr; wr->send_wr_num += 1; + + rc = isert_post_recv(isert_conn, isert_cmd->rx_desc); + if (rc) { + isert_err("ib_post_recv failed with %d\n", rc); + return rc; + } } rc = ib_post_send(isert_conn->qp, wr->send_wr, &wr_failed); diff --git a/drivers/infiniband/ulp/isert/ib_isert.h b/drivers/infiniband/ulp/isert/ib_isert.h index b81dfe07ce62..e586ee1691e9 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.h +++ b/drivers/infiniband/ulp/isert/ib_isert.h @@ -136,6 +136,7 @@ struct isert_cmd { struct isert_conn *conn; struct iscsi_cmd *iscsi_cmd; struct iser_tx_desc tx_desc; + struct iser_rx_desc *rx_desc; struct isert_rdma_wr rdma_wr; struct work_struct comp_work; }; @@ -155,9 +156,8 @@ struct isert_conn { u64 login_req_dma; int login_req_len; u64 login_rsp_dma; - unsigned int rx_desc_head; struct iser_rx_desc *rx_descs; - struct ib_recv_wr rx_wr[ISERT_MIN_POSTED_RX]; + struct ib_recv_wr rx_wr[ISERT_QP_MAX_RECV_DTOS]; struct iscsi_conn *conn; struct list_head node; struct completion login_comp; -- cgit v1.2.3 From 9fd60088ffed7573c2d409ddc63a2150a5edd5d8 Mon Sep 17 00:00:00 2001 From: Jenny Derzhavetz Date: Sun, 6 Sep 2015 14:52:26 +0300 Subject: iser-target: Skip data copy if all the command data comes as immediate Given that supporting zcopy immediate data for all IOs requires iser driver to use its own buffer allocations, we settle with avoiding data copy for IOs with data length of up to 8K (which is more latency sensitive anyway). This trims IO write latency by up to 3us and increase IOPs by up to 40% by saving CPU time doing sg_copy_from_buffer (8K IO size is the obvious winner here). Signed-off-by: Jenny Derzhavetz Signed-off-by: Sagi Grimberg Signed-off-by: Nicholas Bellinger --- drivers/infiniband/ulp/isert/ib_isert.c | 26 ++++++++++++++++++-------- drivers/infiniband/ulp/isert/ib_isert.h | 1 + 2 files changed, 19 insertions(+), 8 deletions(-) diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index 233b8c7a78df..aa59037d7504 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -1315,9 +1315,9 @@ isert_handle_scsi_cmd(struct isert_conn *isert_conn, { struct iscsi_conn *conn = isert_conn->conn; struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)buf; - struct scatterlist *sg; int imm_data, imm_data_len, unsol_data, sg_nents, rc; bool dump_payload = false; + unsigned int data_len; rc = iscsit_setup_scsi_cmd(conn, cmd, buf); if (rc < 0) @@ -1326,7 +1326,10 @@ isert_handle_scsi_cmd(struct isert_conn *isert_conn, imm_data = cmd->immediate_data; imm_data_len = cmd->first_burst_len; unsol_data = cmd->unsolicited_data; + data_len = cmd->se_cmd.data_length; + if (imm_data && imm_data_len == data_len) + cmd->se_cmd.se_cmd_flags |= SCF_PASSTHROUGH_SG_TO_MEM_NOALLOC; rc = iscsit_process_scsi_cmd(conn, cmd, hdr); if (rc < 0) { return 0; @@ -1338,13 +1341,20 @@ isert_handle_scsi_cmd(struct isert_conn *isert_conn, if (!imm_data) return 0; - sg = &cmd->se_cmd.t_data_sg[0]; - sg_nents = max(1UL, DIV_ROUND_UP(imm_data_len, PAGE_SIZE)); - - isert_dbg("Copying Immediate SG: %p sg_nents: %u from %p imm_data_len: %d\n", - sg, sg_nents, &rx_desc->data[0], imm_data_len); - - sg_copy_from_buffer(sg, sg_nents, &rx_desc->data[0], imm_data_len); + if (imm_data_len != data_len) { + sg_nents = max(1UL, DIV_ROUND_UP(imm_data_len, PAGE_SIZE)); + sg_copy_from_buffer(cmd->se_cmd.t_data_sg, sg_nents, + &rx_desc->data[0], imm_data_len); + isert_dbg("Copy Immediate sg_nents: %u imm_data_len: %d\n", + sg_nents, imm_data_len); + } else { + sg_init_table(&isert_cmd->sg, 1); + cmd->se_cmd.t_data_sg = &isert_cmd->sg; + cmd->se_cmd.t_data_nents = 1; + sg_set_buf(&isert_cmd->sg, &rx_desc->data[0], imm_data_len); + isert_dbg("Transfer Immediate imm_data_len: %d\n", + imm_data_len); + } cmd->write_data_done += imm_data_len; diff --git a/drivers/infiniband/ulp/isert/ib_isert.h b/drivers/infiniband/ulp/isert/ib_isert.h index e586ee1691e9..c5b99bcecbcf 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.h +++ b/drivers/infiniband/ulp/isert/ib_isert.h @@ -139,6 +139,7 @@ struct isert_cmd { struct iser_rx_desc *rx_desc; struct isert_rdma_wr rdma_wr; struct work_struct comp_work; + struct scatterlist sg; }; struct isert_device; -- cgit v1.2.3 From 3ccd6e83df8a0d4a664edeecc453c4fa046395fb Mon Sep 17 00:00:00 2001 From: Nicholas Bellinger Date: Sun, 13 Sep 2015 02:30:46 -0700 Subject: target: Fix PR registration + APTPL RCU conversion regression This patch fixes a v4.2+ regression introduced by commit 79dc9c9e86 where lookup of t10_pr_registration->pr_reg_deve and associated ->pr_kref get was missing from __core_scsi3_do_alloc_registration(), which is responsible for setting DEF_PR_REG_ACTIVE. This would result in REGISTER operations completing successfully, but subsequent core_scsi3_pr_seq_non_holder() checking would fail with !DEF_PR_REG_ACTIVE -> RESERVATION CONFLICT status. Update __core_scsi3_add_registration() to drop ->pr_kref reference after registration and any optional ALL_TG_PT=1 processing has completed. Update core_scsi3_decode_spec_i_port() to release the new parent local_pr_reg->pr_kref as well. Also, update __core_scsi3_check_aptpl_registration() to perform the same target_nacl_find_deve() lookup + ->pr_kref get, now that __core_scsi3_add_registration() expects to drop the reference. Finally, since there are cases when se_dev_entry->se_lun_acl can still be dereferenced in core_scsi3_lunacl_undepend_item() while holding ->pr_kref, go ahead and move explicit rcu_assign_pointer() NULL assignments within core_disable_device_list_for_node() until after orig->pr_comp finishes. Reported-by: Scott L. Lykens Tested-by: Scott L. Lykens Cc: Christoph Hellwig Cc: Hannes Reinecke Cc: Lee Duncan Cc: # v4.2+ Signed-off-by: Nicholas Bellinger --- drivers/target/target_core_device.c | 5 +- drivers/target/target_core_pr.c | 91 +++++++++++++++++++++++++++---------- 2 files changed, 70 insertions(+), 26 deletions(-) diff --git a/drivers/target/target_core_device.c b/drivers/target/target_core_device.c index dcc424ac35d4..abf20763b0b6 100644 --- a/drivers/target/target_core_device.c +++ b/drivers/target/target_core_device.c @@ -427,8 +427,6 @@ void core_disable_device_list_for_node( hlist_del_rcu(&orig->link); clear_bit(DEF_PR_REG_ACTIVE, &orig->deve_flags); - rcu_assign_pointer(orig->se_lun, NULL); - rcu_assign_pointer(orig->se_lun_acl, NULL); orig->lun_flags = 0; orig->creation_time = 0; orig->attach_count--; @@ -439,6 +437,9 @@ void core_disable_device_list_for_node( kref_put(&orig->pr_kref, target_pr_kref_release); wait_for_completion(&orig->pr_comp); + rcu_assign_pointer(orig->se_lun, NULL); + rcu_assign_pointer(orig->se_lun_acl, NULL); + kfree_rcu(orig, rcu_head); core_scsi3_free_pr_reg_from_nacl(dev, nacl); diff --git a/drivers/target/target_core_pr.c b/drivers/target/target_core_pr.c index 5ab7100de17e..e7933115087a 100644 --- a/drivers/target/target_core_pr.c +++ b/drivers/target/target_core_pr.c @@ -618,7 +618,7 @@ static struct t10_pr_registration *__core_scsi3_do_alloc_registration( struct se_device *dev, struct se_node_acl *nacl, struct se_lun *lun, - struct se_dev_entry *deve, + struct se_dev_entry *dest_deve, u64 mapped_lun, unsigned char *isid, u64 sa_res_key, @@ -640,7 +640,29 @@ static struct t10_pr_registration *__core_scsi3_do_alloc_registration( INIT_LIST_HEAD(&pr_reg->pr_reg_atp_mem_list); atomic_set(&pr_reg->pr_res_holders, 0); pr_reg->pr_reg_nacl = nacl; - pr_reg->pr_reg_deve = deve; + /* + * For destination registrations for ALL_TG_PT=1 and SPEC_I_PT=1, + * the se_dev_entry->pr_ref will have been already obtained by + * core_get_se_deve_from_rtpi() or __core_scsi3_alloc_registration(). + * + * Otherwise, locate se_dev_entry now and obtain a reference until + * registration completes in __core_scsi3_add_registration(). + */ + if (dest_deve) { + pr_reg->pr_reg_deve = dest_deve; + } else { + rcu_read_lock(); + pr_reg->pr_reg_deve = target_nacl_find_deve(nacl, mapped_lun); + if (!pr_reg->pr_reg_deve) { + rcu_read_unlock(); + pr_err("Unable to locate PR deve %s mapped_lun: %llu\n", + nacl->initiatorname, mapped_lun); + kmem_cache_free(t10_pr_reg_cache, pr_reg); + return NULL; + } + kref_get(&pr_reg->pr_reg_deve->pr_kref); + rcu_read_unlock(); + } pr_reg->pr_res_mapped_lun = mapped_lun; pr_reg->pr_aptpl_target_lun = lun->unpacked_lun; pr_reg->tg_pt_sep_rtpi = lun->lun_rtpi; @@ -936,17 +958,29 @@ static int __core_scsi3_check_aptpl_registration( !(strcmp(pr_reg->pr_tport, t_port)) && (pr_reg->pr_reg_tpgt == tpgt) && (pr_reg->pr_aptpl_target_lun == target_lun)) { + /* + * Obtain the ->pr_reg_deve pointer + reference, that + * is released by __core_scsi3_add_registration() below. + */ + rcu_read_lock(); + pr_reg->pr_reg_deve = target_nacl_find_deve(nacl, mapped_lun); + if (!pr_reg->pr_reg_deve) { + pr_err("Unable to locate PR APTPL %s mapped_lun:" + " %llu\n", nacl->initiatorname, mapped_lun); + rcu_read_unlock(); + continue; + } + kref_get(&pr_reg->pr_reg_deve->pr_kref); + rcu_read_unlock(); pr_reg->pr_reg_nacl = nacl; pr_reg->tg_pt_sep_rtpi = lun->lun_rtpi; - list_del(&pr_reg->pr_reg_aptpl_list); spin_unlock(&pr_tmpl->aptpl_reg_lock); /* * At this point all of the pointers in *pr_reg will * be setup, so go ahead and add the registration. */ - __core_scsi3_add_registration(dev, nacl, pr_reg, 0, 0); /* * If this registration is the reservation holder, @@ -1044,18 +1078,11 @@ static void __core_scsi3_add_registration( __core_scsi3_dump_registration(tfo, dev, nacl, pr_reg, register_type); spin_unlock(&pr_tmpl->registration_lock); - - rcu_read_lock(); - deve = pr_reg->pr_reg_deve; - if (deve) - set_bit(DEF_PR_REG_ACTIVE, &deve->deve_flags); - rcu_read_unlock(); - /* * Skip extra processing for ALL_TG_PT=0 or REGISTER_AND_MOVE. */ if (!pr_reg->pr_reg_all_tg_pt || register_move) - return; + goto out; /* * Walk pr_reg->pr_reg_atp_list and add registrations for ALL_TG_PT=1 * allocated in __core_scsi3_alloc_registration() @@ -1075,19 +1102,31 @@ static void __core_scsi3_add_registration( __core_scsi3_dump_registration(tfo, dev, nacl_tmp, pr_reg_tmp, register_type); spin_unlock(&pr_tmpl->registration_lock); - + /* + * Drop configfs group dependency reference and deve->pr_kref + * obtained from __core_scsi3_alloc_registration() code. + */ rcu_read_lock(); deve = pr_reg_tmp->pr_reg_deve; - if (deve) + if (deve) { set_bit(DEF_PR_REG_ACTIVE, &deve->deve_flags); + core_scsi3_lunacl_undepend_item(deve); + pr_reg_tmp->pr_reg_deve = NULL; + } rcu_read_unlock(); - - /* - * Drop configfs group dependency reference from - * __core_scsi3_alloc_registration() - */ - core_scsi3_lunacl_undepend_item(pr_reg_tmp->pr_reg_deve); } +out: + /* + * Drop deve->pr_kref obtained in __core_scsi3_do_alloc_registration() + */ + rcu_read_lock(); + deve = pr_reg->pr_reg_deve; + if (deve) { + set_bit(DEF_PR_REG_ACTIVE, &deve->deve_flags); + kref_put(&deve->pr_kref, target_pr_kref_release); + pr_reg->pr_reg_deve = NULL; + } + rcu_read_unlock(); } static int core_scsi3_alloc_registration( @@ -1785,9 +1824,11 @@ core_scsi3_decode_spec_i_port( dest_node_acl->initiatorname, i_buf, (dest_se_deve) ? dest_se_deve->mapped_lun : 0); - if (!dest_se_deve) + if (!dest_se_deve) { + kref_put(&local_pr_reg->pr_reg_deve->pr_kref, + target_pr_kref_release); continue; - + } core_scsi3_lunacl_undepend_item(dest_se_deve); core_scsi3_nodeacl_undepend_item(dest_node_acl); core_scsi3_tpg_undepend_item(dest_tpg); @@ -1823,9 +1864,11 @@ out: kmem_cache_free(t10_pr_reg_cache, dest_pr_reg); - if (!dest_se_deve) + if (!dest_se_deve) { + kref_put(&local_pr_reg->pr_reg_deve->pr_kref, + target_pr_kref_release); continue; - + } core_scsi3_lunacl_undepend_item(dest_se_deve); core_scsi3_nodeacl_undepend_item(dest_node_acl); core_scsi3_tpg_undepend_item(dest_tpg); -- cgit v1.2.3 From eeeb9522231118138be418ff527dc8c9050f4707 Mon Sep 17 00:00:00 2001 From: Nicholas Bellinger Date: Tue, 15 Sep 2015 17:27:35 -0700 Subject: target: Propigate backend read-only to core_tpg_add_lun This patch adds a DF_READ_ONLY flag that is used by IBLOCK to signal when a backend has been set to read-only mode, in order to propigate read-only status up to core_tpg_add_lun() for all future LUN fabric exports. With this is place, existing emulation for reporting read-only in spc_emulate_modesense() and normal transport_lookup_cmd_lun() TCM_WRITE_PROTECTED status checking just works as expected. Reported-by: Joeue Deng Reported-by: Andy Grover Signed-off-by: Nicholas Bellinger --- drivers/target/target_core_iblock.c | 2 ++ drivers/target/target_core_tpg.c | 5 ++++- include/target/target_core_base.h | 1 + 3 files changed, 7 insertions(+), 1 deletion(-) diff --git a/drivers/target/target_core_iblock.c b/drivers/target/target_core_iblock.c index 5a9982f5d5d6..0f19e11acac2 100644 --- a/drivers/target/target_core_iblock.c +++ b/drivers/target/target_core_iblock.c @@ -105,6 +105,8 @@ static int iblock_configure_device(struct se_device *dev) mode = FMODE_READ|FMODE_EXCL; if (!ib_dev->ibd_readonly) mode |= FMODE_WRITE; + else + dev->dev_flags |= DF_READ_ONLY; bd = blkdev_get_by_path(ib_dev->ibd_udev_path, mode, ib_dev); if (IS_ERR(bd)) { diff --git a/drivers/target/target_core_tpg.c b/drivers/target/target_core_tpg.c index 2d0381dd105c..5fb9dd7f08bb 100644 --- a/drivers/target/target_core_tpg.c +++ b/drivers/target/target_core_tpg.c @@ -668,7 +668,10 @@ int core_tpg_add_lun( list_add_tail(&lun->lun_dev_link, &dev->dev_sep_list); spin_unlock(&dev->se_port_lock); - lun->lun_access = lun_access; + if (dev->dev_flags & DF_READ_ONLY) + lun->lun_access = TRANSPORT_LUNFLAGS_READ_ONLY; + else + lun->lun_access = lun_access; if (!(dev->se_hba->hba_flags & HBA_FLAGS_INTERNAL_USE)) hlist_add_head_rcu(&lun->link, &tpg->tpg_lun_hlist); mutex_unlock(&tpg->tpg_lun_mutex); diff --git a/include/target/target_core_base.h b/include/target/target_core_base.h index ac9bf1c0e42d..5f48754dc36a 100644 --- a/include/target/target_core_base.h +++ b/include/target/target_core_base.h @@ -730,6 +730,7 @@ struct se_device { #define DF_EMULATED_VPD_UNIT_SERIAL 0x00000004 #define DF_USING_UDEV_PATH 0x00000008 #define DF_USING_ALIAS 0x00000010 +#define DF_READ_ONLY 0x00000020 /* Physical device queue depth */ u32 queue_depth; /* Used for SPC-2 reservations enforce of ISIDs */ -- cgit v1.2.3 From f3bb467ff64b2598d023b3a07592748fa7768b6a Mon Sep 17 00:00:00 2001 From: Nicholas Bellinger Date: Tue, 15 Sep 2015 23:07:45 -0700 Subject: target: Fix target_sense_desc_format NULL pointer dereference This patch allows target_sense_desc_format() to be called without a valid se_device pointer, which can occur during an early exception ahead of transport_lookup_cmd_lun() setting up se_cmd->se_device. This addresses a v4.3-rc1 specific NULL pointer dereference regression introduced by commit 4e4937e8. Cc: Sagi Grimberg Cc: Christoph Hellwig Cc: Hannes Reinecke Signed-off-by: Nicholas Bellinger --- drivers/target/target_core_hba.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/target/target_core_hba.c b/drivers/target/target_core_hba.c index 9522960c7fdd..22390e0e046c 100644 --- a/drivers/target/target_core_hba.c +++ b/drivers/target/target_core_hba.c @@ -187,5 +187,5 @@ core_delete_hba(struct se_hba *hba) bool target_sense_desc_format(struct se_device *dev) { - return dev->transport->get_blocks(dev) > U32_MAX; + return (dev) ? dev->transport->get_blocks(dev) > U32_MAX : false; } -- cgit v1.2.3 From 8fa3a867486f85df66eba8c4df85804d3309c6ad Mon Sep 17 00:00:00 2001 From: Nicholas Bellinger Date: Wed, 16 Sep 2015 20:23:53 -0700 Subject: target: Make TCM_WRITE_PROTECT failure honor D_SENSE bit This patch changes transport_lookup_cmd_lun() to obtain se_lun->lun_ref + se_cmd->se_device rcu_dereference during TCM_WRITE_PROTECT -> CHECK_CONDITION failure status. Do this to ensure the active control D_SENSE mode page bit is being honored. Reported-by: Sagi Grimberg Reviewed-by: Sagi Grimberg Signed-off-by: Nicholas Bellinger --- drivers/target/target_core_device.c | 40 +++++++++++++++++++++---------------- 1 file changed, 23 insertions(+), 17 deletions(-) diff --git a/drivers/target/target_core_device.c b/drivers/target/target_core_device.c index abf20763b0b6..88ea4e4f124b 100644 --- a/drivers/target/target_core_device.c +++ b/drivers/target/target_core_device.c @@ -62,22 +62,13 @@ transport_lookup_cmd_lun(struct se_cmd *se_cmd, u64 unpacked_lun) struct se_session *se_sess = se_cmd->se_sess; struct se_node_acl *nacl = se_sess->se_node_acl; struct se_dev_entry *deve; + sense_reason_t ret = TCM_NO_SENSE; rcu_read_lock(); deve = target_nacl_find_deve(nacl, unpacked_lun); if (deve) { atomic_long_inc(&deve->total_cmds); - if ((se_cmd->data_direction == DMA_TO_DEVICE) && - (deve->lun_flags & TRANSPORT_LUNFLAGS_READ_ONLY)) { - pr_err("TARGET_CORE[%s]: Detected WRITE_PROTECTED LUN" - " Access for 0x%08llx\n", - se_cmd->se_tfo->get_fabric_name(), - unpacked_lun); - rcu_read_unlock(); - return TCM_WRITE_PROTECTED; - } - if (se_cmd->data_direction == DMA_TO_DEVICE) atomic_long_add(se_cmd->data_length, &deve->write_bytes); @@ -93,6 +84,17 @@ transport_lookup_cmd_lun(struct se_cmd *se_cmd, u64 unpacked_lun) percpu_ref_get(&se_lun->lun_ref); se_cmd->lun_ref_active = true; + + if ((se_cmd->data_direction == DMA_TO_DEVICE) && + (deve->lun_flags & TRANSPORT_LUNFLAGS_READ_ONLY)) { + pr_err("TARGET_CORE[%s]: Detected WRITE_PROTECTED LUN" + " Access for 0x%08llx\n", + se_cmd->se_tfo->get_fabric_name(), + unpacked_lun); + rcu_read_unlock(); + ret = TCM_WRITE_PROTECTED; + goto ref_dev; + } } rcu_read_unlock(); @@ -109,12 +111,6 @@ transport_lookup_cmd_lun(struct se_cmd *se_cmd, u64 unpacked_lun) unpacked_lun); return TCM_NON_EXISTENT_LUN; } - /* - * Force WRITE PROTECT for virtual LUN 0 - */ - if ((se_cmd->data_direction != DMA_FROM_DEVICE) && - (se_cmd->data_direction != DMA_NONE)) - return TCM_WRITE_PROTECTED; se_lun = se_sess->se_tpg->tpg_virt_lun0; se_cmd->se_lun = se_sess->se_tpg->tpg_virt_lun0; @@ -123,6 +119,15 @@ transport_lookup_cmd_lun(struct se_cmd *se_cmd, u64 unpacked_lun) percpu_ref_get(&se_lun->lun_ref); se_cmd->lun_ref_active = true; + + /* + * Force WRITE PROTECT for virtual LUN 0 + */ + if ((se_cmd->data_direction != DMA_FROM_DEVICE) && + (se_cmd->data_direction != DMA_NONE)) { + ret = TCM_WRITE_PROTECTED; + goto ref_dev; + } } /* * RCU reference protected by percpu se_lun->lun_ref taken above that @@ -130,6 +135,7 @@ transport_lookup_cmd_lun(struct se_cmd *se_cmd, u64 unpacked_lun) * pointer can be kfree_rcu() by the final se_lun->lun_group put via * target_core_fabric_configfs.c:target_fabric_port_release */ +ref_dev: se_cmd->se_dev = rcu_dereference_raw(se_lun->lun_se_dev); atomic_long_inc(&se_cmd->se_dev->num_cmds); @@ -140,7 +146,7 @@ transport_lookup_cmd_lun(struct se_cmd *se_cmd, u64 unpacked_lun) atomic_long_add(se_cmd->data_length, &se_cmd->se_dev->read_bytes); - return 0; + return ret; } EXPORT_SYMBOL(transport_lookup_cmd_lun); -- cgit v1.2.3 From 673681cafa99776e334c3e61cafa2cf115950c32 Mon Sep 17 00:00:00 2001 From: Nicholas Bellinger Date: Tue, 22 Sep 2015 22:32:14 -0700 Subject: iscsi-target: Avoid OFMarker + IFMarker negotiation MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This patch fixes a v4.2+ regression introduced by commit c04a6091 that removed support for obsolete sync-and-steering markers usage as originally defined in RFC-3720. The regression would involve attempting to send OFMarker=No + IFMarker=No keys during opertional negotiation login phase, including when initiators did not actually propose these keys. The result for MSFT iSCSI initiators would be random junk in TCP stream after the last successful login request was been sent signaling the move to full feature phase (FFP) operation. To address this bug, go ahead and avoid negotiating these keys by default unless the initiator explicitly proposes them, but still respond to them with 'No' if they are proposed. Reported-by: Dragan Milivojević Bisected-by: Christophe Vu-Brugier Tested-by: Christophe Vu-Brugier Signed-off-by: Nicholas Bellinger --- drivers/target/iscsi/iscsi_target_parameters.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/drivers/target/iscsi/iscsi_target_parameters.c b/drivers/target/iscsi/iscsi_target_parameters.c index e8a52f7d6204..51d1734d5390 100644 --- a/drivers/target/iscsi/iscsi_target_parameters.c +++ b/drivers/target/iscsi/iscsi_target_parameters.c @@ -407,6 +407,7 @@ int iscsi_create_default_params(struct iscsi_param_list **param_list_ptr) TYPERANGE_UTF8, USE_INITIAL_ONLY); if (!param) goto out; + /* * Extra parameters for ISER from RFC-5046 */ @@ -496,9 +497,9 @@ int iscsi_set_keys_to_negotiate( } else if (!strcmp(param->name, SESSIONTYPE)) { SET_PSTATE_NEGOTIATE(param); } else if (!strcmp(param->name, IFMARKER)) { - SET_PSTATE_NEGOTIATE(param); + SET_PSTATE_REJECT(param); } else if (!strcmp(param->name, OFMARKER)) { - SET_PSTATE_NEGOTIATE(param); + SET_PSTATE_REJECT(param); } else if (!strcmp(param->name, IFMARKINT)) { SET_PSTATE_REJECT(param); } else if (!strcmp(param->name, OFMARKINT)) { -- cgit v1.2.3