drivers/vhost/scsi.c

   1 /*******************************************************************************
   2  * Vhost kernel TCM fabric driver for virtio SCSI initiators
   3  *
   4  * (C) Copyright 2010-2013 Datera, Inc.
   5  * (C) Copyright 2010-2012 IBM Corp.
   6  *
   7  * Licensed to the Linux Foundation under the General Public License (GPL) version 2.
   8  *
   9  * Authors: Nicholas A. Bellinger <nab@daterainc.com>
  10  *          Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License as published by
  14  * the Free Software Foundation; either version 2 of the License, or
  15  * (at your option) any later version.
  16  *
  17  * This program is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  * GNU General Public License for more details.
  21  *
  22  ****************************************************************************/
  23
  24 #include <linux/module.h>
  25 #include <linux/moduleparam.h>
  26 #include <generated/utsrelease.h>
  27 #include <linux/utsname.h>
  28 #include <linux/init.h>
  29 #include <linux/slab.h>
  30 #include <linux/kthread.h>
  31 #include <linux/types.h>
  32 #include <linux/string.h>
  33 #include <linux/configfs.h>
  34 #include <linux/ctype.h>
  35 #include <linux/compat.h>
  36 #include <linux/eventfd.h>
  37 #include <linux/fs.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/miscdevice.h>
  40 #include <asm/unaligned.h>
  41 #include <scsi/scsi_common.h>
  42 #include <scsi/scsi_proto.h>
  43 #include <target/target_core_base.h>
  44 #include <target/target_core_fabric.h>
  45 #include <linux/vhost.h>
  46 #include <linux/virtio_scsi.h>
  47 #include <linux/llist.h>
  48 #include <linux/bitmap.h>
  49 #include <linux/percpu_ida.h>
  50
  51 #include "vhost.h"
  52
  53 #define VHOST_SCSI_VERSION  "v0.1"
  54 #define VHOST_SCSI_NAMELEN 256
  55 #define VHOST_SCSI_MAX_CDB_SIZE 32
  56 #define VHOST_SCSI_DEFAULT_TAGS 256
  57 #define VHOST_SCSI_PREALLOC_SGLS 2048
  58 #define VHOST_SCSI_PREALLOC_UPAGES 2048
  59 #define VHOST_SCSI_PREALLOC_PROT_SGLS 512
  60
  61 struct vhost_scsi_inflight {
  62         /* Wait for the flush operation to finish */
  63         struct completion comp;
  64         /* Refcount for the inflight reqs */
  65         struct kref kref;
  66 };
  67
  68 struct vhost_scsi_cmd {
  69         /* Descriptor from vhost_get_vq_desc() for virt_queue segment */
  70         int tvc_vq_desc;
  71         /* virtio-scsi initiator task attribute */
  72         int tvc_task_attr;
  73         /* virtio-scsi response incoming iovecs */
  74         int tvc_in_iovs;
  75         /* virtio-scsi initiator data direction */
  76         enum dma_data_direction tvc_data_direction;
  77         /* Expected data transfer length from virtio-scsi header */
  78         u32 tvc_exp_data_len;
  79         /* The Tag from include/linux/virtio_scsi.h:struct virtio_scsi_cmd_req */
  80         u64 tvc_tag;
  81         /* The number of scatterlists associated with this cmd */
  82         u32 tvc_sgl_count;
  83         u32 tvc_prot_sgl_count;
  84         /* Saved unpacked SCSI LUN for vhost_scsi_submission_work() */
  85         u32 tvc_lun;
  86         /* Pointer to the SGL formatted memory from virtio-scsi */
  87         struct scatterlist *tvc_sgl;
  88         struct scatterlist *tvc_prot_sgl;
  89         struct page **tvc_upages;
  90         /* Pointer to response header iovec */
  91         struct iovec tvc_resp_iov;
  92         /* Pointer to vhost_scsi for our device */
  93         struct vhost_scsi *tvc_vhost;
  94         /* Pointer to vhost_virtqueue for the cmd */
  95         struct vhost_virtqueue *tvc_vq;
  96         /* Pointer to vhost nexus memory */
  97         struct vhost_scsi_nexus *tvc_nexus;
  98         /* The TCM I/O descriptor that is accessed via container_of() */
  99         struct se_cmd tvc_se_cmd;
 100         /* work item used for cmwq dispatch to vhost_scsi_submission_work() */
 101         struct work_struct work;
 102         /* Copy of the incoming SCSI command descriptor block (CDB) */
 103         unsigned char tvc_cdb[VHOST_SCSI_MAX_CDB_SIZE];
 104         /* Sense buffer that will be mapped into outgoing status */
 105         unsigned char tvc_sense_buf[TRANSPORT_SENSE_BUFFER];
 106         /* Completed commands list, serviced from vhost worker thread */
 107         struct llist_node tvc_completion_list;
 108         /* Used to track inflight cmd */
 109         struct vhost_scsi_inflight *inflight;
 110 };
 111
 112 struct vhost_scsi_nexus {
 113         /* Pointer to TCM session for I_T Nexus */
 114         struct se_session *tvn_se_sess;
 115 };
 116
 117 struct vhost_scsi_tpg {
 118         /* Vhost port target portal group tag for TCM */
 119         u16 tport_tpgt;
 120         /* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
 121         int tv_tpg_port_count;
 122         /* Used for vhost_scsi device reference to tpg_nexus, protected by tv_tpg_mutex */
 123         int tv_tpg_vhost_count;
 124         /* Used for enabling T10-PI with legacy devices */
 125         int tv_fabric_prot_type;
 126         /* list for vhost_scsi_list */
 127         struct list_head tv_tpg_list;
 128         /* Used to protect access for tpg_nexus */
 129         struct mutex tv_tpg_mutex;
 130         /* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
 131         struct vhost_scsi_nexus *tpg_nexus;
 132         /* Pointer back to vhost_scsi_tport */
 133         struct vhost_scsi_tport *tport;
 134         /* Returned by vhost_scsi_make_tpg() */
 135         struct se_portal_group se_tpg;
 136         /* Pointer back to vhost_scsi, protected by tv_tpg_mutex */
 137         struct vhost_scsi *vhost_scsi;
 138 };
 139
 140 struct vhost_scsi_tport {
 141         /* SCSI protocol the tport is providing */
 142         u8 tport_proto_id;
 143         /* Binary World Wide unique Port Name for Vhost Target port */
 144         u64 tport_wwpn;
 145         /* ASCII formatted WWPN for Vhost Target port */
 146         char tport_name[VHOST_SCSI_NAMELEN];
 147         /* Returned by vhost_scsi_make_tport() */
 148         struct se_wwn tport_wwn;
 149 };
 150
 151 struct vhost_scsi_evt {
 152         /* event to be sent to guest */
 153         struct virtio_scsi_event event;
 154         /* event list, serviced from vhost worker thread */
 155         struct llist_node list;
 156 };
 157
 158 enum {
 159         VHOST_SCSI_VQ_CTL = 0,
 160         VHOST_SCSI_VQ_EVT = 1,
 161         VHOST_SCSI_VQ_IO = 2,
 162 };
 163
 164 /* Note: can't set VIRTIO_F_VERSION_1 yet, since that implies ANY_LAYOUT. */
 165 enum {
 166         VHOST_SCSI_FEATURES = VHOST_FEATURES | (1ULL << VIRTIO_SCSI_F_HOTPLUG) |
 167                                                (1ULL << VIRTIO_SCSI_F_T10_PI)
 168 };
 169
 170 #define VHOST_SCSI_MAX_TARGET   256
 171 #define VHOST_SCSI_MAX_VQ       128
 172 #define VHOST_SCSI_MAX_EVENT    128
 173
 174 struct vhost_scsi_virtqueue {
 175         struct vhost_virtqueue vq;
 176         /*
 177          * Reference counting for inflight reqs, used for flush operation. At
 178          * each time, one reference tracks new commands submitted, while we
 179          * wait for another one to reach 0.
 180          */
 181         struct vhost_scsi_inflight inflights[2];
 182         /*
 183          * Indicate current inflight in use, protected by vq->mutex.
 184          * Writers must also take dev mutex and flush under it.
 185          */
 186         int inflight_idx;
 187 };
 188
 189 struct vhost_scsi {
 190         /* Protected by vhost_scsi->dev.mutex */
 191         struct vhost_scsi_tpg **vs_tpg;
 192         char vs_vhost_wwpn[TRANSPORT_IQN_LEN];
 193
 194         struct vhost_dev dev;
 195         struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
 196
 197         struct vhost_work vs_completion_work; /* cmd completion work item */
 198         struct llist_head vs_completion_list; /* cmd completion queue */
 199
 200         struct vhost_work vs_event_work; /* evt injection work item */
 201         struct llist_head vs_event_list; /* evt injection queue */
 202
 203         bool vs_events_missed; /* any missed events, protected by vq->mutex */
 204         int vs_events_nr; /* num of pending events, protected by vq->mutex */
 205 };
 206
 207 static struct workqueue_struct *vhost_scsi_workqueue;
 208
 209 /* Global spinlock to protect vhost_scsi TPG list for vhost IOCTL access */
 210 static DEFINE_MUTEX(vhost_scsi_mutex);
 211 static LIST_HEAD(vhost_scsi_list);
 212
 213 static void vhost_scsi_done_inflight(struct kref *kref)
 214 {
 215         struct vhost_scsi_inflight *inflight;
 216
 217         inflight = container_of(kref, struct vhost_scsi_inflight, kref);
 218         complete(&inflight->comp);
 219 }
 220
 221 static void vhost_scsi_init_inflight(struct vhost_scsi *vs,
 222                                     struct vhost_scsi_inflight *old_inflight[])
 223 {
 224         struct vhost_scsi_inflight *new_inflight;
 225         struct vhost_virtqueue *vq;
 226         int idx, i;
 227
 228         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
 229                 vq = &vs->vqs[i].vq;
 230
 231                 mutex_lock(&vq->mutex);
 232
 233                 /* store old infight */
 234                 idx = vs->vqs[i].inflight_idx;
 235                 if (old_inflight)
 236                         old_inflight[i] = &vs->vqs[i].inflights[idx];
 237
 238                 /* setup new infight */
 239                 vs->vqs[i].inflight_idx = idx ^ 1;
 240                 new_inflight = &vs->vqs[i].inflights[idx ^ 1];
 241                 kref_init(&new_inflight->kref);
 242                 init_completion(&new_inflight->comp);
 243
 244                 mutex_unlock(&vq->mutex);
 245         }
 246 }
 247
 248 static struct vhost_scsi_inflight *
 249 vhost_scsi_get_inflight(struct vhost_virtqueue *vq)
 250 {
 251         struct vhost_scsi_inflight *inflight;
 252         struct vhost_scsi_virtqueue *svq;
 253
 254         svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
 255         inflight = &svq->inflights[svq->inflight_idx];
 256         kref_get(&inflight->kref);
 257
 258         return inflight;
 259 }
 260
 261 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight *inflight)
 262 {
 263         kref_put(&inflight->kref, vhost_scsi_done_inflight);
 264 }
 265
 266 static int vhost_scsi_check_true(struct se_portal_group *se_tpg)
 267 {
 268         return 1;
 269 }
 270
 271 static int vhost_scsi_check_false(struct se_portal_group *se_tpg)
 272 {
 273         return 0;
 274 }
 275
 276 static char *vhost_scsi_get_fabric_name(void)
 277 {
 278         return "vhost";
 279 }
 280
 281 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group *se_tpg)
 282 {
 283         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 284                                 struct vhost_scsi_tpg, se_tpg);
 285         struct vhost_scsi_tport *tport = tpg->tport;
 286
 287         return &tport->tport_name[0];
 288 }
 289
 290 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
 291 {
 292         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 293                                 struct vhost_scsi_tpg, se_tpg);
 294         return tpg->tport_tpgt;
 295 }
 296
 297 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group *se_tpg)
 298 {
 299         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 300                                 struct vhost_scsi_tpg, se_tpg);
 301
 302         return tpg->tv_fabric_prot_type;
 303 }
 304
 305 static u32 vhost_scsi_tpg_get_inst_index(struct se_portal_group *se_tpg)
 306 {
 307         return 1;
 308 }
 309
 310 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
 311 {
 312         struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
 313                                 struct vhost_scsi_cmd, tvc_se_cmd);
 314         struct se_session *se_sess = tv_cmd->tvc_nexus->tvn_se_sess;
 315         int i;
 316
 317         if (tv_cmd->tvc_sgl_count) {
 318                 for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
 319                         put_page(sg_page(&tv_cmd->tvc_sgl[i]));
 320         }
 321         if (tv_cmd->tvc_prot_sgl_count) {
 322                 for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
 323                         put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
 324         }
 325
 326         vhost_scsi_put_inflight(tv_cmd->inflight);
 327         percpu_ida_free(&se_sess->sess_tag_pool, se_cmd->map_tag);
 328 }
 329
 330 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
 331 {
 332         return 0;
 333 }
 334
 335 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
 336 {
 337         /* Go ahead and process the write immediately */
 338         target_execute_cmd(se_cmd);
 339         return 0;
 340 }
 341
 342 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
 343 {
 344         return 0;
 345 }
 346
 347 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
 348 {
 349         return;
 350 }
 351
 352 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
 353 {
 354         return 0;
 355 }
 356
 357 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
 358 {
 359         struct vhost_scsi *vs = cmd->tvc_vhost;
 360
 361         llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
 362
 363         vhost_work_queue(&vs->dev, &vs->vs_completion_work);
 364 }
 365
 366 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
 367 {
 368         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
 369                                 struct vhost_scsi_cmd, tvc_se_cmd);
 370         vhost_scsi_complete_cmd(cmd);
 371         return 0;
 372 }
 373
 374 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
 375 {
 376         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
 377                                 struct vhost_scsi_cmd, tvc_se_cmd);
 378         vhost_scsi_complete_cmd(cmd);
 379         return 0;
 380 }
 381
 382 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
 383 {
 384         return;
 385 }
 386
 387 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
 388 {
 389         return;
 390 }
 391
 392 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
 393 {
 394         vs->vs_events_nr--;
 395         kfree(evt);
 396 }
 397
 398 static struct vhost_scsi_evt *
 399 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
 400                        u32 event, u32 reason)
 401 {
 402         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 403         struct vhost_scsi_evt *evt;
 404
 405         if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
 406                 vs->vs_events_missed = true;
 407                 return NULL;
 408         }
 409
 410         evt = kzalloc(sizeof(*evt), GFP_KERNEL);
 411         if (!evt) {
 412                 vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
 413                 vs->vs_events_missed = true;
 414                 return NULL;
 415         }
 416
 417         evt->event.event = cpu_to_vhost32(vq, event);
 418         evt->event.reason = cpu_to_vhost32(vq, reason);
 419         vs->vs_events_nr++;
 420
 421         return evt;
 422 }
 423
 424 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
 425 {
 426         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
 427
 428         /* TODO locking against target/backend threads? */
 429         transport_generic_free_cmd(se_cmd, 0);
 430
 431 }
 432
 433 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
 434 {
 435         return target_put_sess_cmd(se_cmd);
 436 }
 437
 438 static void
 439 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
 440 {
 441         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 442         struct virtio_scsi_event *event = &evt->event;
 443         struct virtio_scsi_event __user *eventp;
 444         unsigned out, in;
 445         int head, ret;
 446
 447         if (!vq->private_data) {
 448                 vs->vs_events_missed = true;
 449                 return;
 450         }
 451
 452 again:
 453         vhost_disable_notify(&vs->dev, vq);
 454         head = vhost_get_vq_desc(vq, vq->iov,
 455                         ARRAY_SIZE(vq->iov), &out, &in,
 456                         NULL, NULL);
 457         if (head < 0) {
 458                 vs->vs_events_missed = true;
 459                 return;
 460         }
 461         if (head == vq->num) {
 462                 if (vhost_enable_notify(&vs->dev, vq))
 463                         goto again;
 464                 vs->vs_events_missed = true;
 465                 return;
 466         }
 467
 468         if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
 469                 vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
 470                                 vq->iov[out].iov_len);
 471                 vs->vs_events_missed = true;
 472                 return;
 473         }
 474
 475         if (vs->vs_events_missed) {
 476                 event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
 477                 vs->vs_events_missed = false;
 478         }
 479
 480         eventp = vq->iov[out].iov_base;
 481         ret = __copy_to_user(eventp, event, sizeof(*event));
 482         if (!ret)
 483                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
 484         else
 485                 vq_err(vq, "Faulted on vhost_scsi_send_event\n");
 486 }
 487
 488 static void vhost_scsi_evt_work(struct vhost_work *work)
 489 {
 490         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
 491                                         vs_event_work);
 492         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 493         struct vhost_scsi_evt *evt, *t;
 494         struct llist_node *llnode;
 495
 496         mutex_lock(&vq->mutex);
 497         llnode = llist_del_all(&vs->vs_event_list);
 498         llist_for_each_entry_safe(evt, t, llnode, list) {
 499                 vhost_scsi_do_evt_work(vs, evt);
 500                 vhost_scsi_free_evt(vs, evt);
 501         }
 502         mutex_unlock(&vq->mutex);
 503 }
 504
 505 /* Fill in status and signal that we are done processing this command
 506  *
 507  * This is scheduled in the vhost work queue so we are called with the owner
 508  * process mm and can access the vring.
 509  */
 510 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
 511 {
 512         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
 513                                         vs_completion_work);
 514         DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
 515         struct virtio_scsi_cmd_resp v_rsp;
 516         struct vhost_scsi_cmd *cmd, *t;
 517         struct llist_node *llnode;
 518         struct se_cmd *se_cmd;
 519         struct iov_iter iov_iter;
 520         int ret, vq;
 521
 522         bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
 523         llnode = llist_del_all(&vs->vs_completion_list);
 524         llist_for_each_entry_safe(cmd, t, llnode, tvc_completion_list) {
 525                 se_cmd = &cmd->tvc_se_cmd;
 526
 527                 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
 528                         cmd, se_cmd->residual_count, se_cmd->scsi_status);
 529
 530                 memset(&v_rsp, 0, sizeof(v_rsp));
 531                 v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
 532                 /* TODO is status_qualifier field needed? */
 533                 v_rsp.status = se_cmd->scsi_status;
 534                 v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
 535                                                  se_cmd->scsi_sense_length);
 536                 memcpy(v_rsp.sense, cmd->tvc_sense_buf,
 537                        se_cmd->scsi_sense_length);
 538
 539                 iov_iter_init(&iov_iter, READ, &cmd->tvc_resp_iov,
 540                               cmd->tvc_in_iovs, sizeof(v_rsp));
 541                 ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
 542                 if (likely(ret == sizeof(v_rsp))) {
 543                         struct vhost_scsi_virtqueue *q;
 544                         vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
 545                         q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
 546                         vq = q - vs->vqs;
 547                         __set_bit(vq, signal);
 548                 } else
 549                         pr_err("Faulted on virtio_scsi_cmd_resp\n");
 550
 551                 vhost_scsi_free_cmd(cmd);
 552         }
 553
 554         vq = -1;
 555         while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
 556                 < VHOST_SCSI_MAX_VQ)
 557                 vhost_signal(&vs->dev, &vs->vqs[vq].vq);
 558 }
 559
 560 static struct vhost_scsi_cmd *
 561 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
 562                    unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
 563                    u32 exp_data_len, int data_direction)
 564 {
 565         struct vhost_scsi_cmd *cmd;
 566         struct vhost_scsi_nexus *tv_nexus;
 567         struct se_session *se_sess;
 568         struct scatterlist *sg, *prot_sg;
 569         struct page **pages;
 570         int tag;
 571
 572         tv_nexus = tpg->tpg_nexus;
 573         if (!tv_nexus) {
 574                 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
 575                 return ERR_PTR(-EIO);
 576         }
 577         se_sess = tv_nexus->tvn_se_sess;
 578
 579         tag = percpu_ida_alloc(&se_sess->sess_tag_pool, TASK_RUNNING);
 580         if (tag < 0) {
 581                 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
 582                 return ERR_PTR(-ENOMEM);
 583         }
 584
 585         cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
 586         sg = cmd->tvc_sgl;
 587         prot_sg = cmd->tvc_prot_sgl;
 588         pages = cmd->tvc_upages;
 589         memset(cmd, 0, sizeof(*cmd));
 590         cmd->tvc_sgl = sg;
 591         cmd->tvc_prot_sgl = prot_sg;
 592         cmd->tvc_upages = pages;
 593         cmd->tvc_se_cmd.map_tag = tag;
 594         cmd->tvc_tag = scsi_tag;
 595         cmd->tvc_lun = lun;
 596         cmd->tvc_task_attr = task_attr;
 597         cmd->tvc_exp_data_len = exp_data_len;
 598         cmd->tvc_data_direction = data_direction;
 599         cmd->tvc_nexus = tv_nexus;
 600         cmd->inflight = vhost_scsi_get_inflight(vq);
 601
 602         memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
 603
 604         return cmd;
 605 }
 606
 607 /*
 608  * Map a user memory range into a scatterlist
 609  *
 610  * Returns the number of scatterlist entries used or -errno on error.
 611  */
 612 static int
 613 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
 614                       struct iov_iter *iter,
 615                       struct scatterlist *sgl,
 616                       bool write)
 617 {
 618         struct page **pages = cmd->tvc_upages;
 619         struct scatterlist *sg = sgl;
 620         ssize_t bytes;
 621         size_t offset;
 622         unsigned int npages = 0;
 623
 624         bytes = iov_iter_get_pages(iter, pages, LONG_MAX,
 625                                 VHOST_SCSI_PREALLOC_UPAGES, &offset);
 626         /* No pages were pinned */
 627         if (bytes <= 0)
 628                 return bytes < 0 ? bytes : -EFAULT;
 629
 630         iov_iter_advance(iter, bytes);
 631
 632         while (bytes) {
 633                 unsigned n = min_t(unsigned, PAGE_SIZE - offset, bytes);
 634                 sg_set_page(sg++, pages[npages++], n, offset);
 635                 bytes -= n;
 636                 offset = 0;
 637         }
 638         return npages;
 639 }
 640
 641 static int
 642 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
 643 {
 644         int sgl_count = 0;
 645
 646         if (!iter || !iter->iov) {
 647                 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
 648                        " present\n", __func__, bytes);
 649                 return -EINVAL;
 650         }
 651
 652         sgl_count = iov_iter_npages(iter, 0xffff);
 653         if (sgl_count > max_sgls) {
 654                 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
 655                        " max_sgls: %d\n", __func__, sgl_count, max_sgls);
 656                 return -EINVAL;
 657         }
 658         return sgl_count;
 659 }
 660
 661 static int
 662 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
 663                       struct iov_iter *iter,
 664                       struct scatterlist *sg, int sg_count)
 665 {
 666         struct scatterlist *p = sg;
 667         int ret;
 668
 669         while (iov_iter_count(iter)) {
 670                 ret = vhost_scsi_map_to_sgl(cmd, iter, sg, write);
 671                 if (ret < 0) {
 672                         while (p < sg) {
 673                                 struct page *page = sg_page(p++);
 674                                 if (page)
 675                                         put_page(page);
 676                         }
 677                         return ret;
 678                 }
 679                 sg += ret;
 680         }
 681         return 0;
 682 }
 683
 684 static int
 685 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
 686                  size_t prot_bytes, struct iov_iter *prot_iter,
 687                  size_t data_bytes, struct iov_iter *data_iter)
 688 {
 689         int sgl_count, ret;
 690         bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
 691
 692         if (prot_bytes) {
 693                 sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
 694                                                  VHOST_SCSI_PREALLOC_PROT_SGLS);
 695                 if (sgl_count < 0)
 696                         return sgl_count;
 697
 698                 sg_init_table(cmd->tvc_prot_sgl, sgl_count);
 699                 cmd->tvc_prot_sgl_count = sgl_count;
 700                 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
 701                          cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
 702
 703                 ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
 704                                             cmd->tvc_prot_sgl,
 705                                             cmd->tvc_prot_sgl_count);
 706                 if (ret < 0) {
 707                         cmd->tvc_prot_sgl_count = 0;
 708                         return ret;
 709                 }
 710         }
 711         sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
 712                                          VHOST_SCSI_PREALLOC_SGLS);
 713         if (sgl_count < 0)
 714                 return sgl_count;
 715
 716         sg_init_table(cmd->tvc_sgl, sgl_count);
 717         cmd->tvc_sgl_count = sgl_count;
 718         pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
 719                   cmd->tvc_sgl, cmd->tvc_sgl_count);
 720
 721         ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
 722                                     cmd->tvc_sgl, cmd->tvc_sgl_count);
 723         if (ret < 0) {
 724                 cmd->tvc_sgl_count = 0;
 725                 return ret;
 726         }
 727         return 0;
 728 }
 729
 730 static int vhost_scsi_to_tcm_attr(int attr)
 731 {
 732         switch (attr) {
 733         case VIRTIO_SCSI_S_SIMPLE:
 734                 return TCM_SIMPLE_TAG;
 735         case VIRTIO_SCSI_S_ORDERED:
 736                 return TCM_ORDERED_TAG;
 737         case VIRTIO_SCSI_S_HEAD:
 738                 return TCM_HEAD_TAG;
 739         case VIRTIO_SCSI_S_ACA:
 740                 return TCM_ACA_TAG;
 741         default:
 742                 break;
 743         }
 744         return TCM_SIMPLE_TAG;
 745 }
 746
 747 static void vhost_scsi_submission_work(struct work_struct *work)
 748 {
 749         struct vhost_scsi_cmd *cmd =
 750                 container_of(work, struct vhost_scsi_cmd, work);
 751         struct vhost_scsi_nexus *tv_nexus;
 752         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
 753         struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
 754         int rc;
 755
 756         /* FIXME: BIDI operation */
 757         if (cmd->tvc_sgl_count) {
 758                 sg_ptr = cmd->tvc_sgl;
 759
 760                 if (cmd->tvc_prot_sgl_count)
 761                         sg_prot_ptr = cmd->tvc_prot_sgl;
 762                 else
 763                         se_cmd->prot_pto = true;
 764         } else {
 765                 sg_ptr = NULL;
 766         }
 767         tv_nexus = cmd->tvc_nexus;
 768
 769         se_cmd->tag = 0;
 770         rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
 771                         cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
 772                         cmd->tvc_lun, cmd->tvc_exp_data_len,
 773                         vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
 774                         cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
 775                         sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
 776                         cmd->tvc_prot_sgl_count);
 777         if (rc < 0) {
 778                 transport_send_check_condition_and_sense(se_cmd,
 779                                 TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
 780                 transport_generic_free_cmd(se_cmd, 0);
 781         }
 782 }
 783
 784 static void
 785 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
 786                            struct vhost_virtqueue *vq,
 787                            int head, unsigned out)
 788 {
 789         struct virtio_scsi_cmd_resp __user *resp;
 790         struct virtio_scsi_cmd_resp rsp;
 791         int ret;
 792
 793         memset(&rsp, 0, sizeof(rsp));
 794         rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
 795         resp = vq->iov[out].iov_base;
 796         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
 797         if (!ret)
 798                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
 799         else
 800                 pr_err("Faulted on virtio_scsi_cmd_resp\n");
 801 }
 802
 803 static void
 804 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
 805 {
 806         struct vhost_scsi_tpg **vs_tpg, *tpg;
 807         struct virtio_scsi_cmd_req v_req;
 808         struct virtio_scsi_cmd_req_pi v_req_pi;
 809         struct vhost_scsi_cmd *cmd;
 810         struct iov_iter out_iter, in_iter, prot_iter, data_iter;
 811         u64 tag;
 812         u32 exp_data_len, data_direction;
 813         unsigned int out = 0, in = 0;
 814         int head, ret, prot_bytes;
 815         size_t req_size, rsp_size = sizeof(struct virtio_scsi_cmd_resp);
 816         size_t out_size, in_size;
 817         u16 lun;
 818         u8 *target, *lunp, task_attr;
 819         bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
 820         void *req, *cdb;
 821
 822         mutex_lock(&vq->mutex);
 823         /*
 824          * We can handle the vq only after the endpoint is setup by calling the
 825          * VHOST_SCSI_SET_ENDPOINT ioctl.
 826          */
 827         vs_tpg = vq->private_data;
 828         if (!vs_tpg)
 829                 goto out;
 830
 831         vhost_disable_notify(&vs->dev, vq);
 832
 833         for (;;) {
 834                 head = vhost_get_vq_desc(vq, vq->iov,
 835                                          ARRAY_SIZE(vq->iov), &out, &in,
 836                                          NULL, NULL);
 837                 pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
 838                          head, out, in);
 839                 /* On error, stop handling until the next kick. */
 840                 if (unlikely(head < 0))
 841                         break;
 842                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
 843                 if (head == vq->num) {
 844                         if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
 845                                 vhost_disable_notify(&vs->dev, vq);
 846                                 continue;
 847                         }
 848                         break;
 849                 }
 850                 /*
 851                  * Check for a sane response buffer so we can report early
 852                  * errors back to the guest.
 853                  */
 854                 if (unlikely(vq->iov[out].iov_len < rsp_size)) {
 855                         vq_err(vq, "Expecting at least virtio_scsi_cmd_resp"
 856                                 " size, got %zu bytes\n", vq->iov[out].iov_len);
 857                         break;
 858                 }
 859                 /*
 860                  * Setup pointers and values based upon different virtio-scsi
 861                  * request header if T10_PI is enabled in KVM guest.
 862                  */
 863                 if (t10_pi) {
 864                         req = &v_req_pi;
 865                         req_size = sizeof(v_req_pi);
 866                         lunp = &v_req_pi.lun[0];
 867                         target = &v_req_pi.lun[1];
 868                 } else {
 869                         req = &v_req;
 870                         req_size = sizeof(v_req);
 871                         lunp = &v_req.lun[0];
 872                         target = &v_req.lun[1];
 873                 }
 874                 /*
 875                  * FIXME: Not correct for BIDI operation
 876                  */
 877                 out_size = iov_length(vq->iov, out);
 878                 in_size = iov_length(&vq->iov[out], in);
 879
 880                 /*
 881                  * Copy over the virtio-scsi request header, which for a
 882                  * ANY_LAYOUT enabled guest may span multiple iovecs, or a
 883                  * single iovec may contain both the header + outgoing
 884                  * WRITE payloads.
 885                  *
 886                  * copy_from_iter() will advance out_iter, so that it will
 887                  * point at the start of the outgoing WRITE payload, if
 888                  * DMA_TO_DEVICE is set.
 889                  */
 890                 iov_iter_init(&out_iter, WRITE, vq->iov, out, out_size);
 891
 892                 if (unlikely(!copy_from_iter_full(req, req_size, &out_iter))) {
 893                         vq_err(vq, "Faulted on copy_from_iter\n");
 894                         vhost_scsi_send_bad_target(vs, vq, head, out);
 895                         continue;
 896                 }
 897                 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
 898                 if (unlikely(*lunp != 1)) {
 899                         vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp);
 900                         vhost_scsi_send_bad_target(vs, vq, head, out);
 901                         continue;
 902                 }
 903
 904                 tpg = READ_ONCE(vs_tpg[*target]);
 905                 if (unlikely(!tpg)) {
 906                         /* Target does not exist, fail the request */
 907                         vhost_scsi_send_bad_target(vs, vq, head, out);
 908                         continue;
 909                 }
 910                 /*
 911                  * Determine data_direction by calculating the total outgoing
 912                  * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
 913                  * response headers respectively.
 914                  *
 915                  * For DMA_TO_DEVICE this is out_iter, which is already pointing
 916                  * to the right place.
 917                  *
 918                  * For DMA_FROM_DEVICE, the iovec will be just past the end
 919                  * of the virtio-scsi response header in either the same
 920                  * or immediately following iovec.
 921                  *
 922                  * Any associated T10_PI bytes for the outgoing / incoming
 923                  * payloads are included in calculation of exp_data_len here.
 924                  */
 925                 prot_bytes = 0;
 926
 927                 if (out_size > req_size) {
 928                         data_direction = DMA_TO_DEVICE;
 929                         exp_data_len = out_size - req_size;
 930                         data_iter = out_iter;
 931                 } else if (in_size > rsp_size) {
 932                         data_direction = DMA_FROM_DEVICE;
 933                         exp_data_len = in_size - rsp_size;
 934
 935                         iov_iter_init(&in_iter, READ, &vq->iov[out], in,
 936                                       rsp_size + exp_data_len);
 937                         iov_iter_advance(&in_iter, rsp_size);
 938                         data_iter = in_iter;
 939                 } else {
 940                         data_direction = DMA_NONE;
 941                         exp_data_len = 0;
 942                 }
 943                 /*
 944                  * If T10_PI header + payload is present, setup prot_iter values
 945                  * and recalculate data_iter for vhost_scsi_mapal() mapping to
 946                  * host scatterlists via get_user_pages_fast().
 947                  */
 948                 if (t10_pi) {
 949                         if (v_req_pi.pi_bytesout) {
 950                                 if (data_direction != DMA_TO_DEVICE) {
 951                                         vq_err(vq, "Received non zero pi_bytesout,"
 952                                                 " but wrong data_direction\n");
 953                                         vhost_scsi_send_bad_target(vs, vq, head, out);
 954                                         continue;
 955                                 }
 956                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
 957                         } else if (v_req_pi.pi_bytesin) {
 958                                 if (data_direction != DMA_FROM_DEVICE) {
 959                                         vq_err(vq, "Received non zero pi_bytesin,"
 960                                                 " but wrong data_direction\n");
 961                                         vhost_scsi_send_bad_target(vs, vq, head, out);
 962                                         continue;
 963                                 }
 964                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
 965                         }
 966                         /*
 967                          * Set prot_iter to data_iter, and advance past any
 968                          * preceeding prot_bytes that may be present.
 969                          *
 970                          * Also fix up the exp_data_len to reflect only the
 971                          * actual data payload length.
 972                          */
 973                         if (prot_bytes) {
 974                                 exp_data_len -= prot_bytes;
 975                                 prot_iter = data_iter;
 976                                 iov_iter_advance(&data_iter, prot_bytes);
 977                         }
 978                         tag = vhost64_to_cpu(vq, v_req_pi.tag);
 979                         task_attr = v_req_pi.task_attr;
 980                         cdb = &v_req_pi.cdb[0];
 981                         lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
 982                 } else {
 983                         tag = vhost64_to_cpu(vq, v_req.tag);
 984                         task_attr = v_req.task_attr;
 985                         cdb = &v_req.cdb[0];
 986                         lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
 987                 }
 988                 /*
 989                  * Check that the received CDB size does not exceeded our
 990                  * hardcoded max for vhost-scsi, then get a pre-allocated
 991                  * cmd descriptor for the new virtio-scsi tag.
 992                  *
 993                  * TODO what if cdb was too small for varlen cdb header?
 994                  */
 995                 if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
 996                         vq_err(vq, "Received SCSI CDB with command_size: %d that"
 997                                 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
 998                                 scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
 999                         vhost_scsi_send_bad_target(vs, vq, head, out);
1000                         continue;
1001                 }
1002                 cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1003                                          exp_data_len + prot_bytes,
1004                                          data_direction);
1005                 if (IS_ERR(cmd)) {
1006                         vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1007                                PTR_ERR(cmd));
1008                         vhost_scsi_send_bad_target(vs, vq, head, out);
1009                         continue;
1010                 }
1011                 cmd->tvc_vhost = vs;
1012                 cmd->tvc_vq = vq;
1013                 cmd->tvc_resp_iov = vq->iov[out];
1014                 cmd->tvc_in_iovs = in;
1015
1016                 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1017                          cmd->tvc_cdb[0], cmd->tvc_lun);
1018                 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1019                          " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1020
1021                 if (data_direction != DMA_NONE) {
1022                         ret = vhost_scsi_mapal(cmd,
1023                                                prot_bytes, &prot_iter,
1024                                                exp_data_len, &data_iter);
1025                         if (unlikely(ret)) {
1026                                 vq_err(vq, "Failed to map iov to sgl\n");
1027                                 vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1028                                 vhost_scsi_send_bad_target(vs, vq, head, out);
1029                                 continue;
1030                         }
1031                 }
1032                 /*
1033                  * Save the descriptor from vhost_get_vq_desc() to be used to
1034                  * complete the virtio-scsi request in TCM callback context via
1035                  * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1036                  */
1037                 cmd->tvc_vq_desc = head;
1038                 /*
1039                  * Dispatch cmd descriptor for cmwq execution in process
1040                  * context provided by vhost_scsi_workqueue.  This also ensures
1041                  * cmd is executed on the same kworker CPU as this vhost
1042                  * thread to gain positive L2 cache locality effects.
1043                  */
1044                 INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1045                 queue_work(vhost_scsi_workqueue, &cmd->work);
1046         }
1047 out:
1048         mutex_unlock(&vq->mutex);
1049 }
1050
1051 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1052 {
1053         pr_debug("%s: The handling func for control queue.\n", __func__);
1054 }
1055
1056 static void
1057 vhost_scsi_send_evt(struct vhost_scsi *vs,
1058                    struct vhost_scsi_tpg *tpg,
1059                    struct se_lun *lun,
1060                    u32 event,
1061                    u32 reason)
1062 {
1063         struct vhost_scsi_evt *evt;
1064
1065         evt = vhost_scsi_allocate_evt(vs, event, reason);
1066         if (!evt)
1067                 return;
1068
1069         if (tpg && lun) {
1070                 /* TODO: share lun setup code with virtio-scsi.ko */
1071                 /*
1072                  * Note: evt->event is zeroed when we allocate it and
1073                  * lun[4-7] need to be zero according to virtio-scsi spec.
1074                  */
1075                 evt->event.lun[0] = 0x01;
1076                 evt->event.lun[1] = tpg->tport_tpgt;
1077                 if (lun->unpacked_lun >= 256)
1078                         evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1079                 evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1080         }
1081
1082         llist_add(&evt->list, &vs->vs_event_list);
1083         vhost_work_queue(&vs->dev, &vs->vs_event_work);
1084 }
1085
1086 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1087 {
1088         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1089                                                 poll.work);
1090         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1091
1092         mutex_lock(&vq->mutex);
1093         if (!vq->private_data)
1094                 goto out;
1095
1096         if (vs->vs_events_missed)
1097                 vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1098 out:
1099         mutex_unlock(&vq->mutex);
1100 }
1101
1102 static void vhost_scsi_handle_kick(struct vhost_work *work)
1103 {
1104         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1105                                                 poll.work);
1106         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1107
1108         vhost_scsi_handle_vq(vs, vq);
1109 }
1110
1111 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1112 {
1113         vhost_poll_flush(&vs->vqs[index].vq.poll);
1114 }
1115
1116 /* Callers must hold dev mutex */
1117 static void vhost_scsi_flush(struct vhost_scsi *vs)
1118 {
1119         struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1120         int i;
1121
1122         /* Init new inflight and remember the old inflight */
1123         vhost_scsi_init_inflight(vs, old_inflight);
1124
1125         /*
1126          * The inflight->kref was initialized to 1. We decrement it here to
1127          * indicate the start of the flush operation so that it will reach 0
1128          * when all the reqs are finished.
1129          */
1130         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1131                 kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1132
1133         /* Flush both the vhost poll and vhost work */
1134         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1135                 vhost_scsi_flush_vq(vs, i);
1136         vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1137         vhost_work_flush(&vs->dev, &vs->vs_event_work);
1138
1139         /* Wait for all reqs issued before the flush to be finished */
1140         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1141                 wait_for_completion(&old_inflight[i]->comp);
1142 }
1143
1144 /*
1145  * Called from vhost_scsi_ioctl() context to walk the list of available
1146  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1147  *
1148  *  The lock nesting rule is:
1149  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1150  */
1151 static int
1152 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1153                         struct vhost_scsi_target *t)
1154 {
1155         struct se_portal_group *se_tpg;
1156         struct vhost_scsi_tport *tv_tport;
1157         struct vhost_scsi_tpg *tpg;
1158         struct vhost_scsi_tpg **vs_tpg;
1159         struct vhost_virtqueue *vq;
1160         int index, ret, i, len;
1161         bool match = false;
1162
1163         mutex_lock(&vhost_scsi_mutex);
1164         mutex_lock(&vs->dev.mutex);
1165
1166         /* Verify that ring has been setup correctly. */
1167         for (index = 0; index < vs->dev.nvqs; ++index) {
1168                 /* Verify that ring has been setup correctly. */
1169                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1170                         ret = -EFAULT;
1171                         goto out;
1172                 }
1173         }
1174
1175         len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1176         vs_tpg = kzalloc(len, GFP_KERNEL);
1177         if (!vs_tpg) {
1178                 ret = -ENOMEM;
1179                 goto out;
1180         }
1181         if (vs->vs_tpg)
1182                 memcpy(vs_tpg, vs->vs_tpg, len);
1183
1184         list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1185                 mutex_lock(&tpg->tv_tpg_mutex);
1186                 if (!tpg->tpg_nexus) {
1187                         mutex_unlock(&tpg->tv_tpg_mutex);
1188                         continue;
1189                 }
1190                 if (tpg->tv_tpg_vhost_count != 0) {
1191                         mutex_unlock(&tpg->tv_tpg_mutex);
1192                         continue;
1193                 }
1194                 tv_tport = tpg->tport;
1195
1196                 if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1197                         if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1198                                 kfree(vs_tpg);
1199                                 mutex_unlock(&tpg->tv_tpg_mutex);
1200                                 ret = -EEXIST;
1201                                 goto out;
1202                         }
1203                         /*
1204                          * In order to ensure individual vhost-scsi configfs
1205                          * groups cannot be removed while in use by vhost ioctl,
1206                          * go ahead and take an explicit se_tpg->tpg_group.cg_item
1207                          * dependency now.
1208                          */
1209                         se_tpg = &tpg->se_tpg;
1210                         ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1211                         if (ret) {
1212                                 pr_warn("configfs_depend_item() failed: %d\n", ret);
1213                                 kfree(vs_tpg);
1214                                 mutex_unlock(&tpg->tv_tpg_mutex);
1215                                 goto out;
1216                         }
1217                         tpg->tv_tpg_vhost_count++;
1218                         tpg->vhost_scsi = vs;
1219                         vs_tpg[tpg->tport_tpgt] = tpg;
1220                         smp_mb__after_atomic();
1221                         match = true;
1222                 }
1223                 mutex_unlock(&tpg->tv_tpg_mutex);
1224         }
1225
1226         if (match) {
1227                 memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1228                        sizeof(vs->vs_vhost_wwpn));
1229                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1230                         vq = &vs->vqs[i].vq;
1231                         mutex_lock(&vq->mutex);
1232                         vq->private_data = vs_tpg;
1233                         vhost_vq_init_access(vq);
1234                         mutex_unlock(&vq->mutex);
1235                 }
1236                 ret = 0;
1237         } else {
1238                 ret = -EEXIST;
1239         }
1240
1241         /*
1242          * Act as synchronize_rcu to make sure access to
1243          * old vs->vs_tpg is finished.
1244          */
1245         vhost_scsi_flush(vs);
1246         kfree(vs->vs_tpg);
1247         vs->vs_tpg = vs_tpg;
1248
1249 out:
1250         mutex_unlock(&vs->dev.mutex);
1251         mutex_unlock(&vhost_scsi_mutex);
1252         return ret;
1253 }
1254
1255 static int
1256 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1257                           struct vhost_scsi_target *t)
1258 {
1259         struct se_portal_group *se_tpg;
1260         struct vhost_scsi_tport *tv_tport;
1261         struct vhost_scsi_tpg *tpg;
1262         struct vhost_virtqueue *vq;
1263         bool match = false;
1264         int index, ret, i;
1265         u8 target;
1266
1267         mutex_lock(&vhost_scsi_mutex);
1268         mutex_lock(&vs->dev.mutex);
1269         /* Verify that ring has been setup correctly. */
1270         for (index = 0; index < vs->dev.nvqs; ++index) {
1271                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1272                         ret = -EFAULT;
1273                         goto err_dev;
1274                 }
1275         }
1276
1277         if (!vs->vs_tpg) {
1278                 ret = 0;
1279                 goto err_dev;
1280         }
1281
1282         for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1283                 target = i;
1284                 tpg = vs->vs_tpg[target];
1285                 if (!tpg)
1286                         continue;
1287
1288                 mutex_lock(&tpg->tv_tpg_mutex);
1289                 tv_tport = tpg->tport;
1290                 if (!tv_tport) {
1291                         ret = -ENODEV;
1292                         goto err_tpg;
1293                 }
1294
1295                 if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1296                         pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1297                                 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1298                                 tv_tport->tport_name, tpg->tport_tpgt,
1299                                 t->vhost_wwpn, t->vhost_tpgt);
1300                         ret = -EINVAL;
1301                         goto err_tpg;
1302                 }
1303                 tpg->tv_tpg_vhost_count--;
1304                 tpg->vhost_scsi = NULL;
1305                 vs->vs_tpg[target] = NULL;
1306                 match = true;
1307                 mutex_unlock(&tpg->tv_tpg_mutex);
1308                 /*
1309                  * Release se_tpg->tpg_group.cg_item configfs dependency now
1310                  * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1311                  */
1312                 se_tpg = &tpg->se_tpg;
1313                 target_undepend_item(&se_tpg->tpg_group.cg_item);
1314         }
1315         if (match) {
1316                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1317                         vq = &vs->vqs[i].vq;
1318                         mutex_lock(&vq->mutex);
1319                         vq->private_data = NULL;
1320                         mutex_unlock(&vq->mutex);
1321                 }
1322         }
1323         /*
1324          * Act as synchronize_rcu to make sure access to
1325          * old vs->vs_tpg is finished.
1326          */
1327         vhost_scsi_flush(vs);
1328         kfree(vs->vs_tpg);
1329         vs->vs_tpg = NULL;
1330         WARN_ON(vs->vs_events_nr);
1331         mutex_unlock(&vs->dev.mutex);
1332         mutex_unlock(&vhost_scsi_mutex);
1333         return 0;
1334
1335 err_tpg:
1336         mutex_unlock(&tpg->tv_tpg_mutex);
1337 err_dev:
1338         mutex_unlock(&vs->dev.mutex);
1339         mutex_unlock(&vhost_scsi_mutex);
1340         return ret;
1341 }
1342
1343 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1344 {
1345         struct vhost_virtqueue *vq;
1346         int i;
1347
1348         if (features & ~VHOST_SCSI_FEATURES)
1349                 return -EOPNOTSUPP;
1350
1351         mutex_lock(&vs->dev.mutex);
1352         if ((features & (1 << VHOST_F_LOG_ALL)) &&
1353             !vhost_log_access_ok(&vs->dev)) {
1354                 mutex_unlock(&vs->dev.mutex);
1355                 return -EFAULT;
1356         }
1357
1358         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1359                 vq = &vs->vqs[i].vq;
1360                 mutex_lock(&vq->mutex);
1361                 vq->acked_features = features;
1362                 mutex_unlock(&vq->mutex);
1363         }
1364         mutex_unlock(&vs->dev.mutex);
1365         return 0;
1366 }
1367
1368 static int vhost_scsi_open(struct inode *inode, struct file *f)
1369 {
1370         struct vhost_scsi *vs;
1371         struct vhost_virtqueue **vqs;
1372         int r = -ENOMEM, i;
1373
1374         vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_RETRY_MAYFAIL);
1375         if (!vs) {
1376                 vs = vzalloc(sizeof(*vs));
1377                 if (!vs)
1378                         goto err_vs;
1379         }
1380
1381         vqs = kmalloc_array(VHOST_SCSI_MAX_VQ, sizeof(*vqs), GFP_KERNEL);
1382         if (!vqs)
1383                 goto err_vqs;
1384
1385         vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1386         vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1387
1388         vs->vs_events_nr = 0;
1389         vs->vs_events_missed = false;
1390
1391         vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1392         vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1393         vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1394         vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1395         for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1396                 vqs[i] = &vs->vqs[i].vq;
1397                 vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1398         }
1399         vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ);
1400
1401         vhost_scsi_init_inflight(vs, NULL);
1402
1403         f->private_data = vs;
1404         return 0;
1405
1406 err_vqs:
1407         kvfree(vs);
1408 err_vs:
1409         return r;
1410 }
1411
1412 static int vhost_scsi_release(struct inode *inode, struct file *f)
1413 {
1414         struct vhost_scsi *vs = f->private_data;
1415         struct vhost_scsi_target t;
1416
1417         mutex_lock(&vs->dev.mutex);
1418         memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1419         mutex_unlock(&vs->dev.mutex);
1420         vhost_scsi_clear_endpoint(vs, &t);
1421         vhost_dev_stop(&vs->dev);
1422         vhost_dev_cleanup(&vs->dev);
1423         /* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1424         vhost_scsi_flush(vs);
1425         kfree(vs->dev.vqs);
1426         kvfree(vs);
1427         return 0;
1428 }
1429
1430 static long
1431 vhost_scsi_ioctl(struct file *f,
1432                  unsigned int ioctl,
1433                  unsigned long arg)
1434 {
1435         struct vhost_scsi *vs = f->private_data;
1436         struct vhost_scsi_target backend;
1437         void __user *argp = (void __user *)arg;
1438         u64 __user *featurep = argp;
1439         u32 __user *eventsp = argp;
1440         u32 events_missed;
1441         u64 features;
1442         int r, abi_version = VHOST_SCSI_ABI_VERSION;
1443         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1444
1445         switch (ioctl) {
1446         case VHOST_SCSI_SET_ENDPOINT:
1447                 if (copy_from_user(&backend, argp, sizeof backend))
1448                         return -EFAULT;
1449                 if (backend.reserved != 0)
1450                         return -EOPNOTSUPP;
1451
1452                 return vhost_scsi_set_endpoint(vs, &backend);
1453         case VHOST_SCSI_CLEAR_ENDPOINT:
1454                 if (copy_from_user(&backend, argp, sizeof backend))
1455                         return -EFAULT;
1456                 if (backend.reserved != 0)
1457                         return -EOPNOTSUPP;
1458
1459                 return vhost_scsi_clear_endpoint(vs, &backend);
1460         case VHOST_SCSI_GET_ABI_VERSION:
1461                 if (copy_to_user(argp, &abi_version, sizeof abi_version))
1462                         return -EFAULT;
1463                 return 0;
1464         case VHOST_SCSI_SET_EVENTS_MISSED:
1465                 if (get_user(events_missed, eventsp))
1466                         return -EFAULT;
1467                 mutex_lock(&vq->mutex);
1468                 vs->vs_events_missed = events_missed;
1469                 mutex_unlock(&vq->mutex);
1470                 return 0;
1471         case VHOST_SCSI_GET_EVENTS_MISSED:
1472                 mutex_lock(&vq->mutex);
1473                 events_missed = vs->vs_events_missed;
1474                 mutex_unlock(&vq->mutex);
1475                 if (put_user(events_missed, eventsp))
1476                         return -EFAULT;
1477                 return 0;
1478         case VHOST_GET_FEATURES:
1479                 features = VHOST_SCSI_FEATURES;
1480                 if (copy_to_user(featurep, &features, sizeof features))
1481                         return -EFAULT;
1482                 return 0;
1483         case VHOST_SET_FEATURES:
1484                 if (copy_from_user(&features, featurep, sizeof features))
1485                         return -EFAULT;
1486                 return vhost_scsi_set_features(vs, features);
1487         default:
1488                 mutex_lock(&vs->dev.mutex);
1489                 r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1490                 /* TODO: flush backend after dev ioctl. */
1491                 if (r == -ENOIOCTLCMD)
1492                         r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1493                 mutex_unlock(&vs->dev.mutex);
1494                 return r;
1495         }
1496 }
1497
1498 #ifdef CONFIG_COMPAT
1499 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1500                                 unsigned long arg)
1501 {
1502         return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1503 }
1504 #endif
1505
1506 static const struct file_operations vhost_scsi_fops = {
1507         .owner          = THIS_MODULE,
1508         .release        = vhost_scsi_release,
1509         .unlocked_ioctl = vhost_scsi_ioctl,
1510 #ifdef CONFIG_COMPAT
1511         .compat_ioctl   = vhost_scsi_compat_ioctl,
1512 #endif
1513         .open           = vhost_scsi_open,
1514         .llseek         = noop_llseek,
1515 };
1516
1517 static struct miscdevice vhost_scsi_misc = {
1518         MISC_DYNAMIC_MINOR,
1519         "vhost-scsi",
1520         &vhost_scsi_fops,
1521 };
1522
1523 static int __init vhost_scsi_register(void)
1524 {
1525         return misc_register(&vhost_scsi_misc);
1526 }
1527
1528 static void vhost_scsi_deregister(void)
1529 {
1530         misc_deregister(&vhost_scsi_misc);
1531 }
1532
1533 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1534 {
1535         switch (tport->tport_proto_id) {
1536         case SCSI_PROTOCOL_SAS:
1537                 return "SAS";
1538         case SCSI_PROTOCOL_FCP:
1539                 return "FCP";
1540         case SCSI_PROTOCOL_ISCSI:
1541                 return "iSCSI";
1542         default:
1543                 break;
1544         }
1545
1546         return "Unknown";
1547 }
1548
1549 static void
1550 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1551                   struct se_lun *lun, bool plug)
1552 {
1553
1554         struct vhost_scsi *vs = tpg->vhost_scsi;
1555         struct vhost_virtqueue *vq;
1556         u32 reason;
1557
1558         if (!vs)
1559                 return;
1560
1561         mutex_lock(&vs->dev.mutex);
1562
1563         if (plug)
1564                 reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1565         else
1566                 reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1567
1568         vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1569         mutex_lock(&vq->mutex);
1570         if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1571                 vhost_scsi_send_evt(vs, tpg, lun,
1572                                    VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1573         mutex_unlock(&vq->mutex);
1574         mutex_unlock(&vs->dev.mutex);
1575 }
1576
1577 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1578 {
1579         vhost_scsi_do_plug(tpg, lun, true);
1580 }
1581
1582 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1583 {
1584         vhost_scsi_do_plug(tpg, lun, false);
1585 }
1586
1587 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1588                                struct se_lun *lun)
1589 {
1590         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1591                                 struct vhost_scsi_tpg, se_tpg);
1592
1593         mutex_lock(&vhost_scsi_mutex);
1594
1595         mutex_lock(&tpg->tv_tpg_mutex);
1596         tpg->tv_tpg_port_count++;
1597         mutex_unlock(&tpg->tv_tpg_mutex);
1598
1599         vhost_scsi_hotplug(tpg, lun);
1600
1601         mutex_unlock(&vhost_scsi_mutex);
1602
1603         return 0;
1604 }
1605
1606 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1607                                   struct se_lun *lun)
1608 {
1609         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1610                                 struct vhost_scsi_tpg, se_tpg);
1611
1612         mutex_lock(&vhost_scsi_mutex);
1613
1614         mutex_lock(&tpg->tv_tpg_mutex);
1615         tpg->tv_tpg_port_count--;
1616         mutex_unlock(&tpg->tv_tpg_mutex);
1617
1618         vhost_scsi_hotunplug(tpg, lun);
1619
1620         mutex_unlock(&vhost_scsi_mutex);
1621 }
1622
1623 static void vhost_scsi_free_cmd_map_res(struct se_session *se_sess)
1624 {
1625         struct vhost_scsi_cmd *tv_cmd;
1626         unsigned int i;
1627
1628         if (!se_sess->sess_cmd_map)
1629                 return;
1630
1631         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1632                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1633
1634                 kfree(tv_cmd->tvc_sgl);
1635                 kfree(tv_cmd->tvc_prot_sgl);
1636                 kfree(tv_cmd->tvc_upages);
1637         }
1638 }
1639
1640 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1641                 struct config_item *item, const char *page, size_t count)
1642 {
1643         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1644         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1645                                 struct vhost_scsi_tpg, se_tpg);
1646         unsigned long val;
1647         int ret = kstrtoul(page, 0, &val);
1648
1649         if (ret) {
1650                 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1651                 return ret;
1652         }
1653         if (val != 0 && val != 1 && val != 3) {
1654                 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1655                 return -EINVAL;
1656         }
1657         tpg->tv_fabric_prot_type = val;
1658
1659         return count;
1660 }
1661
1662 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1663                 struct config_item *item, char *page)
1664 {
1665         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1666         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1667                                 struct vhost_scsi_tpg, se_tpg);
1668
1669         return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1670 }
1671
1672 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1673
1674 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1675         &vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1676         NULL,
1677 };
1678
1679 static int vhost_scsi_nexus_cb(struct se_portal_group *se_tpg,
1680                                struct se_session *se_sess, void *p)
1681 {
1682         struct vhost_scsi_cmd *tv_cmd;
1683         unsigned int i;
1684
1685         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1686                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1687
1688                 tv_cmd->tvc_sgl = kzalloc(sizeof(struct scatterlist) *
1689                                         VHOST_SCSI_PREALLOC_SGLS, GFP_KERNEL);
1690                 if (!tv_cmd->tvc_sgl) {
1691                         pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1692                         goto out;
1693                 }
1694
1695                 tv_cmd->tvc_upages = kzalloc(sizeof(struct page *) *
1696                                 VHOST_SCSI_PREALLOC_UPAGES, GFP_KERNEL);
1697                 if (!tv_cmd->tvc_upages) {
1698                         pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1699                         goto out;
1700                 }
1701
1702                 tv_cmd->tvc_prot_sgl = kzalloc(sizeof(struct scatterlist) *
1703                                 VHOST_SCSI_PREALLOC_PROT_SGLS, GFP_KERNEL);
1704                 if (!tv_cmd->tvc_prot_sgl) {
1705                         pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1706                         goto out;
1707                 }
1708         }
1709         return 0;
1710 out:
1711         vhost_scsi_free_cmd_map_res(se_sess);
1712         return -ENOMEM;
1713 }
1714
1715 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1716                                 const char *name)
1717 {
1718         struct vhost_scsi_nexus *tv_nexus;
1719
1720         mutex_lock(&tpg->tv_tpg_mutex);
1721         if (tpg->tpg_nexus) {
1722                 mutex_unlock(&tpg->tv_tpg_mutex);
1723                 pr_debug("tpg->tpg_nexus already exists\n");
1724                 return -EEXIST;
1725         }
1726
1727         tv_nexus = kzalloc(sizeof(*tv_nexus), GFP_KERNEL);
1728         if (!tv_nexus) {
1729                 mutex_unlock(&tpg->tv_tpg_mutex);
1730                 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1731                 return -ENOMEM;
1732         }
1733         /*
1734          * Since we are running in 'demo mode' this call with generate a
1735          * struct se_node_acl for the vhost_scsi struct se_portal_group with
1736          * the SCSI Initiator port name of the passed configfs group 'name'.
1737          */
1738         tv_nexus->tvn_se_sess = target_alloc_session(&tpg->se_tpg,
1739                                         VHOST_SCSI_DEFAULT_TAGS,
1740                                         sizeof(struct vhost_scsi_cmd),
1741                                         TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1742                                         (unsigned char *)name, tv_nexus,
1743                                         vhost_scsi_nexus_cb);
1744         if (IS_ERR(tv_nexus->tvn_se_sess)) {
1745                 mutex_unlock(&tpg->tv_tpg_mutex);
1746                 kfree(tv_nexus);
1747                 return -ENOMEM;
1748         }
1749         tpg->tpg_nexus = tv_nexus;
1750
1751         mutex_unlock(&tpg->tv_tpg_mutex);
1752         return 0;
1753 }
1754
1755 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1756 {
1757         struct se_session *se_sess;
1758         struct vhost_scsi_nexus *tv_nexus;
1759
1760         mutex_lock(&tpg->tv_tpg_mutex);
1761         tv_nexus = tpg->tpg_nexus;
1762         if (!tv_nexus) {
1763                 mutex_unlock(&tpg->tv_tpg_mutex);
1764                 return -ENODEV;
1765         }
1766
1767         se_sess = tv_nexus->tvn_se_sess;
1768         if (!se_sess) {
1769                 mutex_unlock(&tpg->tv_tpg_mutex);
1770                 return -ENODEV;
1771         }
1772
1773         if (tpg->tv_tpg_port_count != 0) {
1774                 mutex_unlock(&tpg->tv_tpg_mutex);
1775                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1776                         " active TPG port count: %d\n",
1777                         tpg->tv_tpg_port_count);
1778                 return -EBUSY;
1779         }
1780
1781         if (tpg->tv_tpg_vhost_count != 0) {
1782                 mutex_unlock(&tpg->tv_tpg_mutex);
1783                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1784                         " active TPG vhost count: %d\n",
1785                         tpg->tv_tpg_vhost_count);
1786                 return -EBUSY;
1787         }
1788
1789         pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1790                 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1791                 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1792
1793         vhost_scsi_free_cmd_map_res(se_sess);
1794         /*
1795          * Release the SCSI I_T Nexus to the emulated vhost Target Port
1796          */
1797         transport_deregister_session(tv_nexus->tvn_se_sess);
1798         tpg->tpg_nexus = NULL;
1799         mutex_unlock(&tpg->tv_tpg_mutex);
1800
1801         kfree(tv_nexus);
1802         return 0;
1803 }
1804
1805 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
1806 {
1807         struct se_portal_group *se_tpg = to_tpg(item);
1808         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1809                                 struct vhost_scsi_tpg, se_tpg);
1810         struct vhost_scsi_nexus *tv_nexus;
1811         ssize_t ret;
1812
1813         mutex_lock(&tpg->tv_tpg_mutex);
1814         tv_nexus = tpg->tpg_nexus;
1815         if (!tv_nexus) {
1816                 mutex_unlock(&tpg->tv_tpg_mutex);
1817                 return -ENODEV;
1818         }
1819         ret = snprintf(page, PAGE_SIZE, "%s\n",
1820                         tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1821         mutex_unlock(&tpg->tv_tpg_mutex);
1822
1823         return ret;
1824 }
1825
1826 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
1827                 const char *page, size_t count)
1828 {
1829         struct se_portal_group *se_tpg = to_tpg(item);
1830         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1831                                 struct vhost_scsi_tpg, se_tpg);
1832         struct vhost_scsi_tport *tport_wwn = tpg->tport;
1833         unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1834         int ret;
1835         /*
1836          * Shutdown the active I_T nexus if 'NULL' is passed..
1837          */
1838         if (!strncmp(page, "NULL", 4)) {
1839                 ret = vhost_scsi_drop_nexus(tpg);
1840                 return (!ret) ? count : ret;
1841         }
1842         /*
1843          * Otherwise make sure the passed virtual Initiator port WWN matches
1844          * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1845          * vhost_scsi_make_nexus().
1846          */
1847         if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1848                 pr_err("Emulated NAA Sas Address: %s, exceeds"
1849                                 " max: %d\n", page, VHOST_SCSI_NAMELEN);
1850                 return -EINVAL;
1851         }
1852         snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1853
1854         ptr = strstr(i_port, "naa.");
1855         if (ptr) {
1856                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1857                         pr_err("Passed SAS Initiator Port %s does not"
1858                                 " match target port protoid: %s\n", i_port,
1859                                 vhost_scsi_dump_proto_id(tport_wwn));
1860                         return -EINVAL;
1861                 }
1862                 port_ptr = &i_port[0];
1863                 goto check_newline;
1864         }
1865         ptr = strstr(i_port, "fc.");
1866         if (ptr) {
1867                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1868                         pr_err("Passed FCP Initiator Port %s does not"
1869                                 " match target port protoid: %s\n", i_port,
1870                                 vhost_scsi_dump_proto_id(tport_wwn));
1871                         return -EINVAL;
1872                 }
1873                 port_ptr = &i_port[3]; /* Skip over "fc." */
1874                 goto check_newline;
1875         }
1876         ptr = strstr(i_port, "iqn.");
1877         if (ptr) {
1878                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1879                         pr_err("Passed iSCSI Initiator Port %s does not"
1880                                 " match target port protoid: %s\n", i_port,
1881                                 vhost_scsi_dump_proto_id(tport_wwn));
1882                         return -EINVAL;
1883                 }
1884                 port_ptr = &i_port[0];
1885                 goto check_newline;
1886         }
1887         pr_err("Unable to locate prefix for emulated Initiator Port:"
1888                         " %s\n", i_port);
1889         return -EINVAL;
1890         /*
1891          * Clear any trailing newline for the NAA WWN
1892          */
1893 check_newline:
1894         if (i_port[strlen(i_port)-1] == '\n')
1895                 i_port[strlen(i_port)-1] = '\0';
1896
1897         ret = vhost_scsi_make_nexus(tpg, port_ptr);
1898         if (ret < 0)
1899                 return ret;
1900
1901         return count;
1902 }
1903
1904 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
1905
1906 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1907         &vhost_scsi_tpg_attr_nexus,
1908         NULL,
1909 };
1910
1911 static struct se_portal_group *
1912 vhost_scsi_make_tpg(struct se_wwn *wwn,
1913                    struct config_group *group,
1914                    const char *name)
1915 {
1916         struct vhost_scsi_tport *tport = container_of(wwn,
1917                         struct vhost_scsi_tport, tport_wwn);
1918
1919         struct vhost_scsi_tpg *tpg;
1920         u16 tpgt;
1921         int ret;
1922
1923         if (strstr(name, "tpgt_") != name)
1924                 return ERR_PTR(-EINVAL);
1925         if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1926                 return ERR_PTR(-EINVAL);
1927
1928         tpg = kzalloc(sizeof(*tpg), GFP_KERNEL);
1929         if (!tpg) {
1930                 pr_err("Unable to allocate struct vhost_scsi_tpg");
1931                 return ERR_PTR(-ENOMEM);
1932         }
1933         mutex_init(&tpg->tv_tpg_mutex);
1934         INIT_LIST_HEAD(&tpg->tv_tpg_list);
1935         tpg->tport = tport;
1936         tpg->tport_tpgt = tpgt;
1937
1938         ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
1939         if (ret < 0) {
1940                 kfree(tpg);
1941                 return NULL;
1942         }
1943         mutex_lock(&vhost_scsi_mutex);
1944         list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
1945         mutex_unlock(&vhost_scsi_mutex);
1946
1947         return &tpg->se_tpg;
1948 }
1949
1950 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
1951 {
1952         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1953                                 struct vhost_scsi_tpg, se_tpg);
1954
1955         mutex_lock(&vhost_scsi_mutex);
1956         list_del(&tpg->tv_tpg_list);
1957         mutex_unlock(&vhost_scsi_mutex);
1958         /*
1959          * Release the virtual I_T Nexus for this vhost TPG
1960          */
1961         vhost_scsi_drop_nexus(tpg);
1962         /*
1963          * Deregister the se_tpg from TCM..
1964          */
1965         core_tpg_deregister(se_tpg);
1966         kfree(tpg);
1967 }
1968
1969 static struct se_wwn *
1970 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
1971                      struct config_group *group,
1972                      const char *name)
1973 {
1974         struct vhost_scsi_tport *tport;
1975         char *ptr;
1976         u64 wwpn = 0;
1977         int off = 0;
1978
1979         /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
1980                 return ERR_PTR(-EINVAL); */
1981
1982         tport = kzalloc(sizeof(*tport), GFP_KERNEL);
1983         if (!tport) {
1984                 pr_err("Unable to allocate struct vhost_scsi_tport");
1985                 return ERR_PTR(-ENOMEM);
1986         }
1987         tport->tport_wwpn = wwpn;
1988         /*
1989          * Determine the emulated Protocol Identifier and Target Port Name
1990          * based on the incoming configfs directory name.
1991          */
1992         ptr = strstr(name, "naa.");
1993         if (ptr) {
1994                 tport->tport_proto_id = SCSI_PROTOCOL_SAS;
1995                 goto check_len;
1996         }
1997         ptr = strstr(name, "fc.");
1998         if (ptr) {
1999                 tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2000                 off = 3; /* Skip over "fc." */
2001                 goto check_len;
2002         }
2003         ptr = strstr(name, "iqn.");
2004         if (ptr) {
2005                 tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2006                 goto check_len;
2007         }
2008
2009         pr_err("Unable to locate prefix for emulated Target Port:"
2010                         " %s\n", name);
2011         kfree(tport);
2012         return ERR_PTR(-EINVAL);
2013
2014 check_len:
2015         if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2016                 pr_err("Emulated %s Address: %s, exceeds"
2017                         " max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2018                         VHOST_SCSI_NAMELEN);
2019                 kfree(tport);
2020                 return ERR_PTR(-EINVAL);
2021         }
2022         snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2023
2024         pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2025                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2026
2027         return &tport->tport_wwn;
2028 }
2029
2030 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2031 {
2032         struct vhost_scsi_tport *tport = container_of(wwn,
2033                                 struct vhost_scsi_tport, tport_wwn);
2034
2035         pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2036                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2037                 tport->tport_name);
2038
2039         kfree(tport);
2040 }
2041
2042 static ssize_t
2043 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2044 {
2045         return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2046                 "on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2047                 utsname()->machine);
2048 }
2049
2050 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2051
2052 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2053         &vhost_scsi_wwn_attr_version,
2054         NULL,
2055 };
2056
2057 static const struct target_core_fabric_ops vhost_scsi_ops = {
2058         .module                         = THIS_MODULE,
2059         .name                           = "vhost",
2060         .get_fabric_name                = vhost_scsi_get_fabric_name,
2061         .tpg_get_wwn                    = vhost_scsi_get_fabric_wwn,
2062         .tpg_get_tag                    = vhost_scsi_get_tpgt,
2063         .tpg_check_demo_mode            = vhost_scsi_check_true,
2064         .tpg_check_demo_mode_cache      = vhost_scsi_check_true,
2065         .tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2066         .tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2067         .tpg_check_prot_fabric_only     = vhost_scsi_check_prot_fabric_only,
2068         .tpg_get_inst_index             = vhost_scsi_tpg_get_inst_index,
2069         .release_cmd                    = vhost_scsi_release_cmd,
2070         .check_stop_free                = vhost_scsi_check_stop_free,
2071         .sess_get_index                 = vhost_scsi_sess_get_index,
2072         .sess_get_initiator_sid         = NULL,
2073         .write_pending                  = vhost_scsi_write_pending,
2074         .write_pending_status           = vhost_scsi_write_pending_status,
2075         .set_default_node_attributes    = vhost_scsi_set_default_node_attrs,
2076         .get_cmd_state                  = vhost_scsi_get_cmd_state,
2077         .queue_data_in                  = vhost_scsi_queue_data_in,
2078         .queue_status                   = vhost_scsi_queue_status,
2079         .queue_tm_rsp                   = vhost_scsi_queue_tm_rsp,
2080         .aborted_task                   = vhost_scsi_aborted_task,
2081         /*
2082          * Setup callers for generic logic in target_core_fabric_configfs.c
2083          */
2084         .fabric_make_wwn                = vhost_scsi_make_tport,
2085         .fabric_drop_wwn                = vhost_scsi_drop_tport,
2086         .fabric_make_tpg                = vhost_scsi_make_tpg,
2087         .fabric_drop_tpg                = vhost_scsi_drop_tpg,
2088         .fabric_post_link               = vhost_scsi_port_link,
2089         .fabric_pre_unlink              = vhost_scsi_port_unlink,
2090
2091         .tfc_wwn_attrs                  = vhost_scsi_wwn_attrs,
2092         .tfc_tpg_base_attrs             = vhost_scsi_tpg_attrs,
2093         .tfc_tpg_attrib_attrs           = vhost_scsi_tpg_attrib_attrs,
2094 };
2095
2096 static int __init vhost_scsi_init(void)
2097 {
2098         int ret = -ENOMEM;
2099
2100         pr_debug("TCM_VHOST fabric module %s on %s/%s"
2101                 " on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2102                 utsname()->machine);
2103
2104         /*
2105          * Use our own dedicated workqueue for submitting I/O into
2106          * target core to avoid contention within system_wq.
2107          */
2108         vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2109         if (!vhost_scsi_workqueue)
2110                 goto out;
2111
2112         ret = vhost_scsi_register();
2113         if (ret < 0)
2114                 goto out_destroy_workqueue;
2115
2116         ret = target_register_template(&vhost_scsi_ops);
2117         if (ret < 0)
2118                 goto out_vhost_scsi_deregister;
2119
2120         return 0;
2121
2122 out_vhost_scsi_deregister:
2123         vhost_scsi_deregister();
2124 out_destroy_workqueue:
2125         destroy_workqueue(vhost_scsi_workqueue);
2126 out:
2127         return ret;
2128 };
2129
2130 static void vhost_scsi_exit(void)
2131 {
2132         target_unregister_template(&vhost_scsi_ops);
2133         vhost_scsi_deregister();
2134         destroy_workqueue(vhost_scsi_workqueue);
2135 };
2136
2137 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2138 MODULE_ALIAS("tcm_vhost");
2139 MODULE_LICENSE("GPL");
2140 module_init(vhost_scsi_init);
2141 module_exit(vhost_scsi_exit);