Back to home page

OSCL-LXR
 
 

     

0001 /*
0002  * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
0003  * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
0004  * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved.
0005  *
0006  * This software is available to you under a choice of one of two
0007  * licenses.  You may choose to be licensed under the terms of the GNU
0008  * General Public License (GPL) Version 2, available from the file
0009  * COPYING in the main directory of this source tree, or the
0010  * OpenIB.org BSD license below:
0011  *
0012  *     Redistribution and use in source and binary forms, with or
0013  *     without modification, are permitted provided that the following
0014  *     conditions are met:
0015  *
0016  *  - Redistributions of source code must retain the above
0017  *    copyright notice, this list of conditions and the following
0018  *    disclaimer.
0019  *
0020  *  - Redistributions in binary form must reproduce the above
0021  *    copyright notice, this list of conditions and the following
0022  *    disclaimer in the documentation and/or other materials
0023  *    provided with the distribution.
0024  *
0025  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
0026  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
0027  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
0028  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
0029  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
0030  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
0031  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
0032  * SOFTWARE.
0033  */
0034 #include <linux/kernel.h>
0035 #include <linux/slab.h>
0036 #include <linux/delay.h>
0037 
0038 #include "iscsi_iser.h"
0039 
0040 #define ISCSI_ISER_MAX_CONN 8
0041 #define ISER_MAX_RX_LEN     (ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN)
0042 #define ISER_MAX_TX_LEN     (ISER_QP_MAX_REQ_DTOS  * ISCSI_ISER_MAX_CONN)
0043 #define ISER_MAX_CQ_LEN     (ISER_MAX_RX_LEN + ISER_MAX_TX_LEN + \
0044                  ISCSI_ISER_MAX_CONN)
0045 
0046 static void iser_qp_event_callback(struct ib_event *cause, void *context)
0047 {
0048     iser_err("qp event %s (%d)\n",
0049          ib_event_msg(cause->event), cause->event);
0050 }
0051 
0052 static void iser_event_handler(struct ib_event_handler *handler,
0053                 struct ib_event *event)
0054 {
0055     iser_err("async event %s (%d) on device %s port %d\n",
0056          ib_event_msg(event->event), event->event,
0057         dev_name(&event->device->dev), event->element.port_num);
0058 }
0059 
0060 /*
0061  * iser_create_device_ib_res - creates Protection Domain (PD), Completion
0062  * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
0063  * the adaptor.
0064  *
0065  * Return: 0 on success, -1 on failure
0066  */
0067 static int iser_create_device_ib_res(struct iser_device *device)
0068 {
0069     struct ib_device *ib_dev = device->ib_device;
0070 
0071     if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS)) {
0072         iser_err("IB device does not support memory registrations\n");
0073         return -1;
0074     }
0075 
0076     device->pd = ib_alloc_pd(ib_dev,
0077         iser_always_reg ? 0 : IB_PD_UNSAFE_GLOBAL_RKEY);
0078     if (IS_ERR(device->pd))
0079         goto pd_err;
0080 
0081     INIT_IB_EVENT_HANDLER(&device->event_handler, ib_dev,
0082                   iser_event_handler);
0083     ib_register_event_handler(&device->event_handler);
0084     return 0;
0085 
0086 pd_err:
0087     iser_err("failed to allocate an IB resource\n");
0088     return -1;
0089 }
0090 
0091 /*
0092  * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR,
0093  * CQ and PD created with the device associated with the adaptor.
0094  */
0095 static void iser_free_device_ib_res(struct iser_device *device)
0096 {
0097     ib_unregister_event_handler(&device->event_handler);
0098     ib_dealloc_pd(device->pd);
0099 
0100     device->pd = NULL;
0101 }
0102 
0103 static struct iser_fr_desc *
0104 iser_create_fastreg_desc(struct iser_device *device,
0105              struct ib_pd *pd,
0106              bool pi_enable,
0107              unsigned int size)
0108 {
0109     struct iser_fr_desc *desc;
0110     struct ib_device *ib_dev = device->ib_device;
0111     enum ib_mr_type mr_type;
0112     int ret;
0113 
0114     desc = kzalloc(sizeof(*desc), GFP_KERNEL);
0115     if (!desc)
0116         return ERR_PTR(-ENOMEM);
0117 
0118     if (ib_dev->attrs.kernel_cap_flags & IBK_SG_GAPS_REG)
0119         mr_type = IB_MR_TYPE_SG_GAPS;
0120     else
0121         mr_type = IB_MR_TYPE_MEM_REG;
0122 
0123     desc->rsc.mr = ib_alloc_mr(pd, mr_type, size);
0124     if (IS_ERR(desc->rsc.mr)) {
0125         ret = PTR_ERR(desc->rsc.mr);
0126         iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret);
0127         goto err_alloc_mr;
0128     }
0129 
0130     if (pi_enable) {
0131         desc->rsc.sig_mr = ib_alloc_mr_integrity(pd, size, size);
0132         if (IS_ERR(desc->rsc.sig_mr)) {
0133             ret = PTR_ERR(desc->rsc.sig_mr);
0134             iser_err("Failed to allocate sig_mr err=%d\n", ret);
0135             goto err_alloc_mr_integrity;
0136         }
0137     }
0138     desc->rsc.mr_valid = 0;
0139 
0140     return desc;
0141 
0142 err_alloc_mr_integrity:
0143     ib_dereg_mr(desc->rsc.mr);
0144 err_alloc_mr:
0145     kfree(desc);
0146 
0147     return ERR_PTR(ret);
0148 }
0149 
0150 static void iser_destroy_fastreg_desc(struct iser_fr_desc *desc)
0151 {
0152     struct iser_reg_resources *res = &desc->rsc;
0153 
0154     ib_dereg_mr(res->mr);
0155     if (res->sig_mr) {
0156         ib_dereg_mr(res->sig_mr);
0157         res->sig_mr = NULL;
0158     }
0159     kfree(desc);
0160 }
0161 
0162 /**
0163  * iser_alloc_fastreg_pool - Creates pool of fast_reg descriptors
0164  * for fast registration work requests.
0165  * @ib_conn: connection RDMA resources
0166  * @cmds_max: max number of SCSI commands for this connection
0167  * @size: max number of pages per map request
0168  *
0169  * Return: 0 on success, or errno code on failure
0170  */
0171 int iser_alloc_fastreg_pool(struct ib_conn *ib_conn,
0172                 unsigned cmds_max,
0173                 unsigned int size)
0174 {
0175     struct iser_device *device = ib_conn->device;
0176     struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
0177     struct iser_fr_desc *desc;
0178     int i, ret;
0179 
0180     INIT_LIST_HEAD(&fr_pool->list);
0181     INIT_LIST_HEAD(&fr_pool->all_list);
0182     spin_lock_init(&fr_pool->lock);
0183     fr_pool->size = 0;
0184     for (i = 0; i < cmds_max; i++) {
0185         desc = iser_create_fastreg_desc(device, device->pd,
0186                         ib_conn->pi_support, size);
0187         if (IS_ERR(desc)) {
0188             ret = PTR_ERR(desc);
0189             goto err;
0190         }
0191 
0192         list_add_tail(&desc->list, &fr_pool->list);
0193         list_add_tail(&desc->all_list, &fr_pool->all_list);
0194         fr_pool->size++;
0195     }
0196 
0197     return 0;
0198 
0199 err:
0200     iser_free_fastreg_pool(ib_conn);
0201     return ret;
0202 }
0203 
0204 /**
0205  * iser_free_fastreg_pool - releases the pool of fast_reg descriptors
0206  * @ib_conn: connection RDMA resources
0207  */
0208 void iser_free_fastreg_pool(struct ib_conn *ib_conn)
0209 {
0210     struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
0211     struct iser_fr_desc *desc, *tmp;
0212     int i = 0;
0213 
0214     if (list_empty(&fr_pool->all_list))
0215         return;
0216 
0217     iser_info("freeing conn %p fr pool\n", ib_conn);
0218 
0219     list_for_each_entry_safe(desc, tmp, &fr_pool->all_list, all_list) {
0220         list_del(&desc->all_list);
0221         iser_destroy_fastreg_desc(desc);
0222         ++i;
0223     }
0224 
0225     if (i < fr_pool->size)
0226         iser_warn("pool still has %d regions registered\n",
0227               fr_pool->size - i);
0228 }
0229 
0230 /*
0231  * iser_create_ib_conn_res - Queue-Pair (QP)
0232  *
0233  * Return: 0 on success, -1 on failure
0234  */
0235 static int iser_create_ib_conn_res(struct ib_conn *ib_conn)
0236 {
0237     struct iser_conn *iser_conn = to_iser_conn(ib_conn);
0238     struct iser_device  *device;
0239     struct ib_device    *ib_dev;
0240     struct ib_qp_init_attr  init_attr;
0241     int         ret = -ENOMEM;
0242     unsigned int max_send_wr, cq_size;
0243 
0244     BUG_ON(ib_conn->device == NULL);
0245 
0246     device = ib_conn->device;
0247     ib_dev = device->ib_device;
0248 
0249     /* +1 for drain */
0250     if (ib_conn->pi_support)
0251         max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS + 1;
0252     else
0253         max_send_wr = ISER_QP_MAX_REQ_DTOS + 1;
0254     max_send_wr = min_t(unsigned int, max_send_wr,
0255                 (unsigned int)ib_dev->attrs.max_qp_wr);
0256 
0257     cq_size = max_send_wr + ISER_QP_MAX_RECV_DTOS;
0258     ib_conn->cq = ib_cq_pool_get(ib_dev, cq_size, -1, IB_POLL_SOFTIRQ);
0259     if (IS_ERR(ib_conn->cq)) {
0260         ret = PTR_ERR(ib_conn->cq);
0261         goto cq_err;
0262     }
0263     ib_conn->cq_size = cq_size;
0264 
0265     memset(&init_attr, 0, sizeof(init_attr));
0266 
0267     init_attr.event_handler = iser_qp_event_callback;
0268     init_attr.qp_context = (void *)ib_conn;
0269     init_attr.send_cq = ib_conn->cq;
0270     init_attr.recv_cq = ib_conn->cq;
0271     /* +1 for drain */
0272     init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS + 1;
0273     init_attr.cap.max_send_sge = 2;
0274     init_attr.cap.max_recv_sge = 1;
0275     init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
0276     init_attr.qp_type = IB_QPT_RC;
0277     init_attr.cap.max_send_wr = max_send_wr;
0278     if (ib_conn->pi_support)
0279         init_attr.create_flags |= IB_QP_CREATE_INTEGRITY_EN;
0280     iser_conn->max_cmds = ISER_GET_MAX_XMIT_CMDS(max_send_wr - 1);
0281 
0282     ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
0283     if (ret)
0284         goto out_err;
0285 
0286     ib_conn->qp = ib_conn->cma_id->qp;
0287     iser_info("setting conn %p cma_id %p qp %p max_send_wr %d\n", ib_conn,
0288           ib_conn->cma_id, ib_conn->cma_id->qp, max_send_wr);
0289     return ret;
0290 
0291 out_err:
0292     ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size);
0293 cq_err:
0294     iser_err("unable to alloc mem or create resource, err %d\n", ret);
0295 
0296     return ret;
0297 }
0298 
0299 /*
0300  * based on the resolved device node GUID see if there already allocated
0301  * device for this device. If there's no such, create one.
0302  */
0303 static
0304 struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
0305 {
0306     struct iser_device *device;
0307 
0308     mutex_lock(&ig.device_list_mutex);
0309 
0310     list_for_each_entry(device, &ig.device_list, ig_list)
0311         /* find if there's a match using the node GUID */
0312         if (device->ib_device->node_guid == cma_id->device->node_guid)
0313             goto inc_refcnt;
0314 
0315     device = kzalloc(sizeof *device, GFP_KERNEL);
0316     if (!device)
0317         goto out;
0318 
0319     /* assign this device to the device */
0320     device->ib_device = cma_id->device;
0321     /* init the device and link it into ig device list */
0322     if (iser_create_device_ib_res(device)) {
0323         kfree(device);
0324         device = NULL;
0325         goto out;
0326     }
0327     list_add(&device->ig_list, &ig.device_list);
0328 
0329 inc_refcnt:
0330     device->refcount++;
0331 out:
0332     mutex_unlock(&ig.device_list_mutex);
0333     return device;
0334 }
0335 
0336 /* if there's no demand for this device, release it */
0337 static void iser_device_try_release(struct iser_device *device)
0338 {
0339     mutex_lock(&ig.device_list_mutex);
0340     device->refcount--;
0341     iser_info("device %p refcount %d\n", device, device->refcount);
0342     if (!device->refcount) {
0343         iser_free_device_ib_res(device);
0344         list_del(&device->ig_list);
0345         kfree(device);
0346     }
0347     mutex_unlock(&ig.device_list_mutex);
0348 }
0349 
0350 /*
0351  * Called with state mutex held
0352  */
0353 static int iser_conn_state_comp_exch(struct iser_conn *iser_conn,
0354                      enum iser_conn_state comp,
0355                      enum iser_conn_state exch)
0356 {
0357     int ret;
0358 
0359     ret = (iser_conn->state == comp);
0360     if (ret)
0361         iser_conn->state = exch;
0362 
0363     return ret;
0364 }
0365 
0366 void iser_release_work(struct work_struct *work)
0367 {
0368     struct iser_conn *iser_conn;
0369 
0370     iser_conn = container_of(work, struct iser_conn, release_work);
0371 
0372     /* Wait for conn_stop to complete */
0373     wait_for_completion(&iser_conn->stop_completion);
0374     /* Wait for IB resouces cleanup to complete */
0375     wait_for_completion(&iser_conn->ib_completion);
0376 
0377     mutex_lock(&iser_conn->state_mutex);
0378     iser_conn->state = ISER_CONN_DOWN;
0379     mutex_unlock(&iser_conn->state_mutex);
0380 
0381     iser_conn_release(iser_conn);
0382 }
0383 
0384 /**
0385  * iser_free_ib_conn_res - release IB related resources
0386  * @iser_conn: iser connection struct
0387  * @destroy: indicator if we need to try to release the
0388  *     iser device and memory regoins pool (only iscsi
0389  *     shutdown and DEVICE_REMOVAL will use this).
0390  *
0391  * This routine is called with the iser state mutex held
0392  * so the cm_id removal is out of here. It is Safe to
0393  * be invoked multiple times.
0394  */
0395 static void iser_free_ib_conn_res(struct iser_conn *iser_conn, bool destroy)
0396 {
0397     struct ib_conn *ib_conn = &iser_conn->ib_conn;
0398     struct iser_device *device = ib_conn->device;
0399 
0400     iser_info("freeing conn %p cma_id %p qp %p\n",
0401           iser_conn, ib_conn->cma_id, ib_conn->qp);
0402 
0403     if (ib_conn->qp) {
0404         rdma_destroy_qp(ib_conn->cma_id);
0405         ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size);
0406         ib_conn->qp = NULL;
0407     }
0408 
0409     if (destroy) {
0410         if (iser_conn->rx_descs)
0411             iser_free_rx_descriptors(iser_conn);
0412 
0413         if (device) {
0414             iser_device_try_release(device);
0415             ib_conn->device = NULL;
0416         }
0417     }
0418 }
0419 
0420 /**
0421  * iser_conn_release - Frees all conn objects and deallocs conn descriptor
0422  * @iser_conn: iSER connection context
0423  */
0424 void iser_conn_release(struct iser_conn *iser_conn)
0425 {
0426     struct ib_conn *ib_conn = &iser_conn->ib_conn;
0427 
0428     mutex_lock(&ig.connlist_mutex);
0429     list_del(&iser_conn->conn_list);
0430     mutex_unlock(&ig.connlist_mutex);
0431 
0432     mutex_lock(&iser_conn->state_mutex);
0433     /* In case we endup here without ep_disconnect being invoked. */
0434     if (iser_conn->state != ISER_CONN_DOWN) {
0435         iser_warn("iser conn %p state %d, expected state down.\n",
0436               iser_conn, iser_conn->state);
0437         iscsi_destroy_endpoint(iser_conn->ep);
0438         iser_conn->state = ISER_CONN_DOWN;
0439     }
0440     /*
0441      * In case we never got to bind stage, we still need to
0442      * release IB resources (which is safe to call more than once).
0443      */
0444     iser_free_ib_conn_res(iser_conn, true);
0445     mutex_unlock(&iser_conn->state_mutex);
0446 
0447     if (ib_conn->cma_id) {
0448         rdma_destroy_id(ib_conn->cma_id);
0449         ib_conn->cma_id = NULL;
0450     }
0451 
0452     kfree(iser_conn);
0453 }
0454 
0455 /**
0456  * iser_conn_terminate - triggers start of the disconnect procedures and
0457  * waits for them to be done
0458  * @iser_conn: iSER connection context
0459  *
0460  * Called with state mutex held
0461  */
0462 int iser_conn_terminate(struct iser_conn *iser_conn)
0463 {
0464     struct ib_conn *ib_conn = &iser_conn->ib_conn;
0465     int err = 0;
0466 
0467     /* terminate the iser conn only if the conn state is UP */
0468     if (!iser_conn_state_comp_exch(iser_conn, ISER_CONN_UP,
0469                        ISER_CONN_TERMINATING))
0470         return 0;
0471 
0472     iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state);
0473 
0474     /* suspend queuing of new iscsi commands */
0475     if (iser_conn->iscsi_conn)
0476         iscsi_suspend_queue(iser_conn->iscsi_conn);
0477 
0478     /*
0479      * In case we didn't already clean up the cma_id (peer initiated
0480      * a disconnection), we need to Cause the CMA to change the QP
0481      * state to ERROR.
0482      */
0483     if (ib_conn->cma_id) {
0484         err = rdma_disconnect(ib_conn->cma_id);
0485         if (err)
0486             iser_err("Failed to disconnect, conn: 0x%p err %d\n",
0487                  iser_conn, err);
0488 
0489         /* block until all flush errors are consumed */
0490         ib_drain_qp(ib_conn->qp);
0491     }
0492 
0493     return 1;
0494 }
0495 
0496 /*
0497  * Called with state mutex held
0498  */
0499 static void iser_connect_error(struct rdma_cm_id *cma_id)
0500 {
0501     struct iser_conn *iser_conn;
0502 
0503     iser_conn = cma_id->context;
0504     iser_conn->state = ISER_CONN_TERMINATING;
0505 }
0506 
0507 static void iser_calc_scsi_params(struct iser_conn *iser_conn,
0508                   unsigned int max_sectors)
0509 {
0510     struct iser_device *device = iser_conn->ib_conn.device;
0511     struct ib_device_attr *attr = &device->ib_device->attrs;
0512     unsigned short sg_tablesize, sup_sg_tablesize;
0513     unsigned short reserved_mr_pages;
0514     u32 max_num_sg;
0515 
0516     /*
0517      * FRs without SG_GAPS can only map up to a (device) page per entry,
0518      * but if the first entry is misaligned we'll end up using two entries
0519      * (head and tail) for a single page worth data, so one additional
0520      * entry is required.
0521      */
0522     if (attr->kernel_cap_flags & IBK_SG_GAPS_REG)
0523         reserved_mr_pages = 0;
0524     else
0525         reserved_mr_pages = 1;
0526 
0527     if (iser_conn->ib_conn.pi_support)
0528         max_num_sg = attr->max_pi_fast_reg_page_list_len;
0529     else
0530         max_num_sg = attr->max_fast_reg_page_list_len;
0531 
0532     sg_tablesize = DIV_ROUND_UP(max_sectors * SECTOR_SIZE, SZ_4K);
0533     sup_sg_tablesize = min_t(uint, ISCSI_ISER_MAX_SG_TABLESIZE,
0534                  max_num_sg - reserved_mr_pages);
0535     iser_conn->scsi_sg_tablesize = min(sg_tablesize, sup_sg_tablesize);
0536     iser_conn->pages_per_mr =
0537         iser_conn->scsi_sg_tablesize + reserved_mr_pages;
0538 }
0539 
0540 /*
0541  * Called with state mutex held
0542  */
0543 static void iser_addr_handler(struct rdma_cm_id *cma_id)
0544 {
0545     struct iser_device *device;
0546     struct iser_conn *iser_conn;
0547     struct ib_conn *ib_conn;
0548     int    ret;
0549 
0550     iser_conn = cma_id->context;
0551     if (iser_conn->state != ISER_CONN_PENDING)
0552         /* bailout */
0553         return;
0554 
0555     ib_conn = &iser_conn->ib_conn;
0556     device = iser_device_find_by_ib_device(cma_id);
0557     if (!device) {
0558         iser_err("device lookup/creation failed\n");
0559         iser_connect_error(cma_id);
0560         return;
0561     }
0562 
0563     ib_conn->device = device;
0564 
0565     /* connection T10-PI support */
0566     if (iser_pi_enable) {
0567         if (!(device->ib_device->attrs.kernel_cap_flags &
0568               IBK_INTEGRITY_HANDOVER)) {
0569             iser_warn("T10-PI requested but not supported on %s, "
0570                   "continue without T10-PI\n",
0571                   dev_name(&ib_conn->device->ib_device->dev));
0572             ib_conn->pi_support = false;
0573         } else {
0574             ib_conn->pi_support = true;
0575         }
0576     }
0577 
0578     iser_calc_scsi_params(iser_conn, iser_max_sectors);
0579 
0580     ret = rdma_resolve_route(cma_id, 1000);
0581     if (ret) {
0582         iser_err("resolve route failed: %d\n", ret);
0583         iser_connect_error(cma_id);
0584         return;
0585     }
0586 }
0587 
0588 /*
0589  * Called with state mutex held
0590  */
0591 static void iser_route_handler(struct rdma_cm_id *cma_id)
0592 {
0593     struct rdma_conn_param conn_param;
0594     int ret;
0595     struct iser_cm_hdr req_hdr;
0596     struct iser_conn *iser_conn = cma_id->context;
0597     struct ib_conn *ib_conn = &iser_conn->ib_conn;
0598     struct ib_device *ib_dev = ib_conn->device->ib_device;
0599 
0600     if (iser_conn->state != ISER_CONN_PENDING)
0601         /* bailout */
0602         return;
0603 
0604     ret = iser_create_ib_conn_res(ib_conn);
0605     if (ret)
0606         goto failure;
0607 
0608     memset(&conn_param, 0, sizeof conn_param);
0609     conn_param.responder_resources = ib_dev->attrs.max_qp_rd_atom;
0610     conn_param.initiator_depth = 1;
0611     conn_param.retry_count = 7;
0612     conn_param.rnr_retry_count = 6;
0613 
0614     memset(&req_hdr, 0, sizeof(req_hdr));
0615     req_hdr.flags = ISER_ZBVA_NOT_SUP;
0616     if (!iser_always_reg)
0617         req_hdr.flags |= ISER_SEND_W_INV_NOT_SUP;
0618     conn_param.private_data = (void *)&req_hdr;
0619     conn_param.private_data_len = sizeof(struct iser_cm_hdr);
0620 
0621     ret = rdma_connect_locked(cma_id, &conn_param);
0622     if (ret) {
0623         iser_err("failure connecting: %d\n", ret);
0624         goto failure;
0625     }
0626 
0627     return;
0628 failure:
0629     iser_connect_error(cma_id);
0630 }
0631 
0632 static void iser_connected_handler(struct rdma_cm_id *cma_id,
0633                    const void *private_data)
0634 {
0635     struct iser_conn *iser_conn;
0636     struct ib_qp_attr attr;
0637     struct ib_qp_init_attr init_attr;
0638 
0639     iser_conn = cma_id->context;
0640     if (iser_conn->state != ISER_CONN_PENDING)
0641         /* bailout */
0642         return;
0643 
0644     (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr);
0645     iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num);
0646 
0647     if (private_data) {
0648         u8 flags = *(u8 *)private_data;
0649 
0650         iser_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP);
0651     }
0652 
0653     iser_info("conn %p: negotiated %s invalidation\n",
0654           iser_conn, iser_conn->snd_w_inv ? "remote" : "local");
0655 
0656     iser_conn->state = ISER_CONN_UP;
0657     complete(&iser_conn->up_completion);
0658 }
0659 
0660 static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
0661 {
0662     struct iser_conn *iser_conn = cma_id->context;
0663 
0664     if (iser_conn_terminate(iser_conn)) {
0665         if (iser_conn->iscsi_conn)
0666             iscsi_conn_failure(iser_conn->iscsi_conn,
0667                        ISCSI_ERR_CONN_FAILED);
0668         else
0669             iser_err("iscsi_iser connection isn't bound\n");
0670     }
0671 }
0672 
0673 static void iser_cleanup_handler(struct rdma_cm_id *cma_id,
0674                  bool destroy)
0675 {
0676     struct iser_conn *iser_conn = cma_id->context;
0677 
0678     /*
0679      * We are not guaranteed that we visited disconnected_handler
0680      * by now, call it here to be safe that we handle CM drep
0681      * and flush errors.
0682      */
0683     iser_disconnected_handler(cma_id);
0684     iser_free_ib_conn_res(iser_conn, destroy);
0685     complete(&iser_conn->ib_completion);
0686 }
0687 
0688 static int iser_cma_handler(struct rdma_cm_id *cma_id,
0689                 struct rdma_cm_event *event)
0690 {
0691     struct iser_conn *iser_conn;
0692     int ret = 0;
0693 
0694     iser_conn = cma_id->context;
0695     iser_info("%s (%d): status %d conn %p id %p\n",
0696           rdma_event_msg(event->event), event->event,
0697           event->status, cma_id->context, cma_id);
0698 
0699     mutex_lock(&iser_conn->state_mutex);
0700     switch (event->event) {
0701     case RDMA_CM_EVENT_ADDR_RESOLVED:
0702         iser_addr_handler(cma_id);
0703         break;
0704     case RDMA_CM_EVENT_ROUTE_RESOLVED:
0705         iser_route_handler(cma_id);
0706         break;
0707     case RDMA_CM_EVENT_ESTABLISHED:
0708         iser_connected_handler(cma_id, event->param.conn.private_data);
0709         break;
0710     case RDMA_CM_EVENT_REJECTED:
0711         iser_info("Connection rejected: %s\n",
0712              rdma_reject_msg(cma_id, event->status));
0713         fallthrough;
0714     case RDMA_CM_EVENT_ADDR_ERROR:
0715     case RDMA_CM_EVENT_ROUTE_ERROR:
0716     case RDMA_CM_EVENT_CONNECT_ERROR:
0717     case RDMA_CM_EVENT_UNREACHABLE:
0718         iser_connect_error(cma_id);
0719         break;
0720     case RDMA_CM_EVENT_DISCONNECTED:
0721     case RDMA_CM_EVENT_ADDR_CHANGE:
0722     case RDMA_CM_EVENT_TIMEWAIT_EXIT:
0723         iser_cleanup_handler(cma_id, false);
0724         break;
0725     case RDMA_CM_EVENT_DEVICE_REMOVAL:
0726         /*
0727          * we *must* destroy the device as we cannot rely
0728          * on iscsid to be around to initiate error handling.
0729          * also if we are not in state DOWN implicitly destroy
0730          * the cma_id.
0731          */
0732         iser_cleanup_handler(cma_id, true);
0733         if (iser_conn->state != ISER_CONN_DOWN) {
0734             iser_conn->ib_conn.cma_id = NULL;
0735             ret = 1;
0736         }
0737         break;
0738     default:
0739         iser_err("Unexpected RDMA CM event: %s (%d)\n",
0740              rdma_event_msg(event->event), event->event);
0741         break;
0742     }
0743     mutex_unlock(&iser_conn->state_mutex);
0744 
0745     return ret;
0746 }
0747 
0748 void iser_conn_init(struct iser_conn *iser_conn)
0749 {
0750     struct ib_conn *ib_conn = &iser_conn->ib_conn;
0751 
0752     iser_conn->state = ISER_CONN_INIT;
0753     init_completion(&iser_conn->stop_completion);
0754     init_completion(&iser_conn->ib_completion);
0755     init_completion(&iser_conn->up_completion);
0756     INIT_LIST_HEAD(&iser_conn->conn_list);
0757     mutex_init(&iser_conn->state_mutex);
0758 
0759     ib_conn->reg_cqe.done = iser_reg_comp;
0760 }
0761 
0762 /*
0763  * starts the process of connecting to the target
0764  * sleeps until the connection is established or rejected
0765  */
0766 int iser_connect(struct iser_conn *iser_conn, struct sockaddr *src_addr,
0767          struct sockaddr *dst_addr, int non_blocking)
0768 {
0769     struct ib_conn *ib_conn = &iser_conn->ib_conn;
0770     int err = 0;
0771 
0772     mutex_lock(&iser_conn->state_mutex);
0773 
0774     sprintf(iser_conn->name, "%pISp", dst_addr);
0775 
0776     iser_info("connecting to: %s\n", iser_conn->name);
0777 
0778     /* the device is known only --after-- address resolution */
0779     ib_conn->device = NULL;
0780 
0781     iser_conn->state = ISER_CONN_PENDING;
0782 
0783     ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler,
0784                      iser_conn, RDMA_PS_TCP, IB_QPT_RC);
0785     if (IS_ERR(ib_conn->cma_id)) {
0786         err = PTR_ERR(ib_conn->cma_id);
0787         iser_err("rdma_create_id failed: %d\n", err);
0788         goto id_failure;
0789     }
0790 
0791     err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000);
0792     if (err) {
0793         iser_err("rdma_resolve_addr failed: %d\n", err);
0794         goto addr_failure;
0795     }
0796 
0797     if (!non_blocking) {
0798         wait_for_completion_interruptible(&iser_conn->up_completion);
0799 
0800         if (iser_conn->state != ISER_CONN_UP) {
0801             err =  -EIO;
0802             goto connect_failure;
0803         }
0804     }
0805     mutex_unlock(&iser_conn->state_mutex);
0806 
0807     mutex_lock(&ig.connlist_mutex);
0808     list_add(&iser_conn->conn_list, &ig.connlist);
0809     mutex_unlock(&ig.connlist_mutex);
0810     return 0;
0811 
0812 id_failure:
0813     ib_conn->cma_id = NULL;
0814 addr_failure:
0815     iser_conn->state = ISER_CONN_DOWN;
0816 connect_failure:
0817     mutex_unlock(&iser_conn->state_mutex);
0818     iser_conn_release(iser_conn);
0819     return err;
0820 }
0821 
0822 int iser_post_recvl(struct iser_conn *iser_conn)
0823 {
0824     struct ib_conn *ib_conn = &iser_conn->ib_conn;
0825     struct iser_login_desc *desc = &iser_conn->login_desc;
0826     struct ib_recv_wr wr;
0827     int ret;
0828 
0829     desc->sge.addr = desc->rsp_dma;
0830     desc->sge.length = ISER_RX_LOGIN_SIZE;
0831     desc->sge.lkey = ib_conn->device->pd->local_dma_lkey;
0832 
0833     desc->cqe.done = iser_login_rsp;
0834     wr.wr_cqe = &desc->cqe;
0835     wr.sg_list = &desc->sge;
0836     wr.num_sge = 1;
0837     wr.next = NULL;
0838 
0839     ret = ib_post_recv(ib_conn->qp, &wr, NULL);
0840     if (unlikely(ret))
0841         iser_err("ib_post_recv login failed ret=%d\n", ret);
0842 
0843     return ret;
0844 }
0845 
0846 int iser_post_recvm(struct iser_conn *iser_conn, struct iser_rx_desc *rx_desc)
0847 {
0848     struct ib_conn *ib_conn = &iser_conn->ib_conn;
0849     struct ib_recv_wr wr;
0850     int ret;
0851 
0852     rx_desc->cqe.done = iser_task_rsp;
0853     wr.wr_cqe = &rx_desc->cqe;
0854     wr.sg_list = &rx_desc->rx_sg;
0855     wr.num_sge = 1;
0856     wr.next = NULL;
0857 
0858     ret = ib_post_recv(ib_conn->qp, &wr, NULL);
0859     if (unlikely(ret))
0860         iser_err("ib_post_recv failed ret=%d\n", ret);
0861 
0862     return ret;
0863 }
0864 
0865 
0866 /**
0867  * iser_post_send - Initiate a Send DTO operation
0868  * @ib_conn: connection RDMA resources
0869  * @tx_desc: iSER TX descriptor
0870  *
0871  * Return: 0 on success, -1 on failure
0872  */
0873 int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc)
0874 {
0875     struct ib_send_wr *wr = &tx_desc->send_wr;
0876     struct ib_send_wr *first_wr;
0877     int ret;
0878 
0879     ib_dma_sync_single_for_device(ib_conn->device->ib_device,
0880                       tx_desc->dma_addr, ISER_HEADERS_LEN,
0881                       DMA_TO_DEVICE);
0882 
0883     wr->next = NULL;
0884     wr->wr_cqe = &tx_desc->cqe;
0885     wr->sg_list = tx_desc->tx_sg;
0886     wr->num_sge = tx_desc->num_sge;
0887     wr->opcode = IB_WR_SEND;
0888     wr->send_flags = IB_SEND_SIGNALED;
0889 
0890     if (tx_desc->inv_wr.next)
0891         first_wr = &tx_desc->inv_wr;
0892     else if (tx_desc->reg_wr.wr.next)
0893         first_wr = &tx_desc->reg_wr.wr;
0894     else
0895         first_wr = wr;
0896 
0897     ret = ib_post_send(ib_conn->qp, first_wr, NULL);
0898     if (unlikely(ret))
0899         iser_err("ib_post_send failed, ret:%d opcode:%d\n",
0900              ret, wr->opcode);
0901 
0902     return ret;
0903 }
0904 
0905 u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task,
0906                  enum iser_data_dir cmd_dir, sector_t *sector)
0907 {
0908     struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
0909     struct iser_fr_desc *desc = reg->desc;
0910     unsigned long sector_size = iser_task->sc->device->sector_size;
0911     struct ib_mr_status mr_status;
0912     int ret;
0913 
0914     if (desc && desc->sig_protected) {
0915         desc->sig_protected = false;
0916         ret = ib_check_mr_status(desc->rsc.sig_mr,
0917                      IB_MR_CHECK_SIG_STATUS, &mr_status);
0918         if (ret) {
0919             iser_err("ib_check_mr_status failed, ret %d\n", ret);
0920             /* Not a lot we can do, return ambiguous guard error */
0921             *sector = 0;
0922             return 0x1;
0923         }
0924 
0925         if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
0926             sector_t sector_off = mr_status.sig_err.sig_err_offset;
0927 
0928             sector_div(sector_off, sector_size + 8);
0929             *sector = scsi_get_sector(iser_task->sc) + sector_off;
0930 
0931             iser_err("PI error found type %d at sector %llx "
0932                    "expected %x vs actual %x\n",
0933                    mr_status.sig_err.err_type,
0934                    (unsigned long long)*sector,
0935                    mr_status.sig_err.expected,
0936                    mr_status.sig_err.actual);
0937 
0938             switch (mr_status.sig_err.err_type) {
0939             case IB_SIG_BAD_GUARD:
0940                 return 0x1;
0941             case IB_SIG_BAD_REFTAG:
0942                 return 0x3;
0943             case IB_SIG_BAD_APPTAG:
0944                 return 0x2;
0945             }
0946         }
0947     }
0948 
0949     return 0;
0950 }
0951 
0952 void iser_err_comp(struct ib_wc *wc, const char *type)
0953 {
0954     if (wc->status != IB_WC_WR_FLUSH_ERR) {
0955         struct iser_conn *iser_conn = to_iser_conn(wc->qp->qp_context);
0956 
0957         iser_err("%s failure: %s (%d) vend_err %#x\n", type,
0958              ib_wc_status_msg(wc->status), wc->status,
0959              wc->vendor_err);
0960 
0961         if (iser_conn->iscsi_conn)
0962             iscsi_conn_failure(iser_conn->iscsi_conn,
0963                        ISCSI_ERR_CONN_FAILED);
0964     } else {
0965         iser_dbg("%s failure: %s (%d)\n", type,
0966              ib_wc_status_msg(wc->status), wc->status);
0967     }
0968 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!