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 int iov_num_pages(void __user *iov_base, size_t iov_len)
 214 {
 215         return (PAGE_ALIGN((unsigned long)iov_base + iov_len) -
 216                ((unsigned long)iov_base & PAGE_MASK)) >> PAGE_SHIFT;
 217 }
 218
 219 static void vhost_scsi_done_inflight(struct kref *kref)
 220 {
 221         struct vhost_scsi_inflight *inflight;
 222
 223         inflight = container_of(kref, struct vhost_scsi_inflight, kref);
 224         complete(&inflight->comp);
 225 }
 226
 227 static void vhost_scsi_init_inflight(struct vhost_scsi *vs,
 228                                     struct vhost_scsi_inflight *old_inflight[])
 229 {
 230         struct vhost_scsi_inflight *new_inflight;
 231         struct vhost_virtqueue *vq;
 232         int idx, i;
 233
 234         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
 235                 vq = &vs->vqs[i].vq;
 236
 237                 mutex_lock(&vq->mutex);
 238
 239                 /* store old infight */
 240                 idx = vs->vqs[i].inflight_idx;
 241                 if (old_inflight)
 242                         old_inflight[i] = &vs->vqs[i].inflights[idx];
 243
 244                 /* setup new infight */
 245                 vs->vqs[i].inflight_idx = idx ^ 1;
 246                 new_inflight = &vs->vqs[i].inflights[idx ^ 1];
 247                 kref_init(&new_inflight->kref);
 248                 init_completion(&new_inflight->comp);
 249
 250                 mutex_unlock(&vq->mutex);
 251         }
 252 }
 253
 254 static struct vhost_scsi_inflight *
 255 vhost_scsi_get_inflight(struct vhost_virtqueue *vq)
 256 {
 257         struct vhost_scsi_inflight *inflight;
 258         struct vhost_scsi_virtqueue *svq;
 259
 260         svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
 261         inflight = &svq->inflights[svq->inflight_idx];
 262         kref_get(&inflight->kref);
 263
 264         return inflight;
 265 }
 266
 267 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight *inflight)
 268 {
 269         kref_put(&inflight->kref, vhost_scsi_done_inflight);
 270 }
 271
 272 static int vhost_scsi_check_true(struct se_portal_group *se_tpg)
 273 {
 274         return 1;
 275 }
 276
 277 static int vhost_scsi_check_false(struct se_portal_group *se_tpg)
 278 {
 279         return 0;
 280 }
 281
 282 static char *vhost_scsi_get_fabric_name(void)
 283 {
 284         return "vhost";
 285 }
 286
 287 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group *se_tpg)
 288 {
 289         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 290                                 struct vhost_scsi_tpg, se_tpg);
 291         struct vhost_scsi_tport *tport = tpg->tport;
 292
 293         return &tport->tport_name[0];
 294 }
 295
 296 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
 297 {
 298         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 299                                 struct vhost_scsi_tpg, se_tpg);
 300         return tpg->tport_tpgt;
 301 }
 302
 303 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group *se_tpg)
 304 {
 305         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
 306                                 struct vhost_scsi_tpg, se_tpg);
 307
 308         return tpg->tv_fabric_prot_type;
 309 }
 310
 311 static u32 vhost_scsi_tpg_get_inst_index(struct se_portal_group *se_tpg)
 312 {
 313         return 1;
 314 }
 315
 316 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
 317 {
 318         struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
 319                                 struct vhost_scsi_cmd, tvc_se_cmd);
 320         struct se_session *se_sess = tv_cmd->tvc_nexus->tvn_se_sess;
 321         int i;
 322
 323         if (tv_cmd->tvc_sgl_count) {
 324                 for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
 325                         put_page(sg_page(&tv_cmd->tvc_sgl[i]));
 326         }
 327         if (tv_cmd->tvc_prot_sgl_count) {
 328                 for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
 329                         put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
 330         }
 331
 332         vhost_scsi_put_inflight(tv_cmd->inflight);
 333         percpu_ida_free(&se_sess->sess_tag_pool, se_cmd->map_tag);
 334 }
 335
 336 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
 337 {
 338         return 0;
 339 }
 340
 341 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
 342 {
 343         /* Go ahead and process the write immediately */
 344         target_execute_cmd(se_cmd);
 345         return 0;
 346 }
 347
 348 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
 349 {
 350         return 0;
 351 }
 352
 353 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
 354 {
 355         return;
 356 }
 357
 358 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
 359 {
 360         return 0;
 361 }
 362
 363 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
 364 {
 365         struct vhost_scsi *vs = cmd->tvc_vhost;
 366
 367         llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
 368
 369         vhost_work_queue(&vs->dev, &vs->vs_completion_work);
 370 }
 371
 372 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
 373 {
 374         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
 375                                 struct vhost_scsi_cmd, tvc_se_cmd);
 376         vhost_scsi_complete_cmd(cmd);
 377         return 0;
 378 }
 379
 380 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
 381 {
 382         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
 383                                 struct vhost_scsi_cmd, tvc_se_cmd);
 384         vhost_scsi_complete_cmd(cmd);
 385         return 0;
 386 }
 387
 388 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
 389 {
 390         return;
 391 }
 392
 393 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
 394 {
 395         return;
 396 }
 397
 398 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
 399 {
 400         vs->vs_events_nr--;
 401         kfree(evt);
 402 }
 403
 404 static struct vhost_scsi_evt *
 405 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
 406                        u32 event, u32 reason)
 407 {
 408         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 409         struct vhost_scsi_evt *evt;
 410
 411         if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
 412                 vs->vs_events_missed = true;
 413                 return NULL;
 414         }
 415
 416         evt = kzalloc(sizeof(*evt), GFP_KERNEL);
 417         if (!evt) {
 418                 vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
 419                 vs->vs_events_missed = true;
 420                 return NULL;
 421         }
 422
 423         evt->event.event = cpu_to_vhost32(vq, event);
 424         evt->event.reason = cpu_to_vhost32(vq, reason);
 425         vs->vs_events_nr++;
 426
 427         return evt;
 428 }
 429
 430 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
 431 {
 432         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
 433
 434         /* TODO locking against target/backend threads? */
 435         transport_generic_free_cmd(se_cmd, 0);
 436
 437 }
 438
 439 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
 440 {
 441         return target_put_sess_cmd(se_cmd);
 442 }
 443
 444 static void
 445 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
 446 {
 447         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 448         struct virtio_scsi_event *event = &evt->event;
 449         struct virtio_scsi_event __user *eventp;
 450         unsigned out, in;
 451         int head, ret;
 452
 453         if (!vq->private_data) {
 454                 vs->vs_events_missed = true;
 455                 return;
 456         }
 457
 458 again:
 459         vhost_disable_notify(&vs->dev, vq);
 460         head = vhost_get_vq_desc(vq, vq->iov,
 461                         ARRAY_SIZE(vq->iov), &out, &in,
 462                         NULL, NULL);
 463         if (head < 0) {
 464                 vs->vs_events_missed = true;
 465                 return;
 466         }
 467         if (head == vq->num) {
 468                 if (vhost_enable_notify(&vs->dev, vq))
 469                         goto again;
 470                 vs->vs_events_missed = true;
 471                 return;
 472         }
 473
 474         if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
 475                 vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
 476                                 vq->iov[out].iov_len);
 477                 vs->vs_events_missed = true;
 478                 return;
 479         }
 480
 481         if (vs->vs_events_missed) {
 482                 event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
 483                 vs->vs_events_missed = false;
 484         }
 485
 486         eventp = vq->iov[out].iov_base;
 487         ret = __copy_to_user(eventp, event, sizeof(*event));
 488         if (!ret)
 489                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
 490         else
 491                 vq_err(vq, "Faulted on vhost_scsi_send_event\n");
 492 }
 493
 494 static void vhost_scsi_evt_work(struct vhost_work *work)
 495 {
 496         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
 497                                         vs_event_work);
 498         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
 499         struct vhost_scsi_evt *evt, *t;
 500         struct llist_node *llnode;
 501
 502         mutex_lock(&vq->mutex);
 503         llnode = llist_del_all(&vs->vs_event_list);
 504         llist_for_each_entry_safe(evt, t, llnode, list) {
 505                 vhost_scsi_do_evt_work(vs, evt);
 506                 vhost_scsi_free_evt(vs, evt);
 507         }
 508         mutex_unlock(&vq->mutex);
 509 }
 510
 511 /* Fill in status and signal that we are done processing this command
 512  *
 513  * This is scheduled in the vhost work queue so we are called with the owner
 514  * process mm and can access the vring.
 515  */
 516 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
 517 {
 518         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
 519                                         vs_completion_work);
 520         DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
 521         struct virtio_scsi_cmd_resp v_rsp;
 522         struct vhost_scsi_cmd *cmd;
 523         struct llist_node *llnode;
 524         struct se_cmd *se_cmd;
 525         struct iov_iter iov_iter;
 526         int ret, vq;
 527
 528         bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
 529         llnode = llist_del_all(&vs->vs_completion_list);
 530         llist_for_each_entry(cmd, llnode, tvc_completion_list) {
 531                 se_cmd = &cmd->tvc_se_cmd;
 532
 533                 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
 534                         cmd, se_cmd->residual_count, se_cmd->scsi_status);
 535
 536                 memset(&v_rsp, 0, sizeof(v_rsp));
 537                 v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
 538                 /* TODO is status_qualifier field needed? */
 539                 v_rsp.status = se_cmd->scsi_status;
 540                 v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
 541                                                  se_cmd->scsi_sense_length);
 542                 memcpy(v_rsp.sense, cmd->tvc_sense_buf,
 543                        se_cmd->scsi_sense_length);
 544
 545                 iov_iter_init(&iov_iter, READ, &cmd->tvc_resp_iov,
 546                               cmd->tvc_in_iovs, sizeof(v_rsp));
 547                 ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
 548                 if (likely(ret == sizeof(v_rsp))) {
 549                         struct vhost_scsi_virtqueue *q;
 550                         vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
 551                         q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
 552                         vq = q - vs->vqs;
 553                         __set_bit(vq, signal);
 554                 } else
 555                         pr_err("Faulted on virtio_scsi_cmd_resp\n");
 556
 557                 vhost_scsi_free_cmd(cmd);
 558         }
 559
 560         vq = -1;
 561         while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
 562                 < VHOST_SCSI_MAX_VQ)
 563                 vhost_signal(&vs->dev, &vs->vqs[vq].vq);
 564 }
 565
 566 static struct vhost_scsi_cmd *
 567 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
 568                    unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
 569                    u32 exp_data_len, int data_direction)
 570 {
 571         struct vhost_scsi_cmd *cmd;
 572         struct vhost_scsi_nexus *tv_nexus;
 573         struct se_session *se_sess;
 574         struct scatterlist *sg, *prot_sg;
 575         struct page **pages;
 576         int tag;
 577
 578         tv_nexus = tpg->tpg_nexus;
 579         if (!tv_nexus) {
 580                 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
 581                 return ERR_PTR(-EIO);
 582         }
 583         se_sess = tv_nexus->tvn_se_sess;
 584
 585         tag = percpu_ida_alloc(&se_sess->sess_tag_pool, TASK_RUNNING);
 586         if (tag < 0) {
 587                 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
 588                 return ERR_PTR(-ENOMEM);
 589         }
 590
 591         cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
 592         sg = cmd->tvc_sgl;
 593         prot_sg = cmd->tvc_prot_sgl;
 594         pages = cmd->tvc_upages;
 595         memset(cmd, 0, sizeof(struct vhost_scsi_cmd));
 596
 597         cmd->tvc_sgl = sg;
 598         cmd->tvc_prot_sgl = prot_sg;
 599         cmd->tvc_upages = pages;
 600         cmd->tvc_se_cmd.map_tag = tag;
 601         cmd->tvc_tag = scsi_tag;
 602         cmd->tvc_lun = lun;
 603         cmd->tvc_task_attr = task_attr;
 604         cmd->tvc_exp_data_len = exp_data_len;
 605         cmd->tvc_data_direction = data_direction;
 606         cmd->tvc_nexus = tv_nexus;
 607         cmd->inflight = vhost_scsi_get_inflight(vq);
 608
 609         memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
 610
 611         return cmd;
 612 }
 613
 614 /*
 615  * Map a user memory range into a scatterlist
 616  *
 617  * Returns the number of scatterlist entries used or -errno on error.
 618  */
 619 static int
 620 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
 621                       void __user *ptr,
 622                       size_t len,
 623                       struct scatterlist *sgl,
 624                       bool write)
 625 {
 626         unsigned int npages = 0, offset, nbytes;
 627         unsigned int pages_nr = iov_num_pages(ptr, len);
 628         struct scatterlist *sg = sgl;
 629         struct page **pages = cmd->tvc_upages;
 630         int ret, i;
 631
 632         if (pages_nr > VHOST_SCSI_PREALLOC_UPAGES) {
 633                 pr_err("vhost_scsi_map_to_sgl() pages_nr: %u greater than"
 634                        " preallocated VHOST_SCSI_PREALLOC_UPAGES: %u\n",
 635                         pages_nr, VHOST_SCSI_PREALLOC_UPAGES);
 636                 return -ENOBUFS;
 637         }
 638
 639         ret = get_user_pages_fast((unsigned long)ptr, pages_nr, write, pages);
 640         /* No pages were pinned */
 641         if (ret < 0)
 642                 goto out;
 643         /* Less pages pinned than wanted */
 644         if (ret != pages_nr) {
 645                 for (i = 0; i < ret; i++)
 646                         put_page(pages[i]);
 647                 ret = -EFAULT;
 648                 goto out;
 649         }
 650
 651         while (len > 0) {
 652                 offset = (uintptr_t)ptr & ~PAGE_MASK;
 653                 nbytes = min_t(unsigned int, PAGE_SIZE - offset, len);
 654                 sg_set_page(sg, pages[npages], nbytes, offset);
 655                 ptr += nbytes;
 656                 len -= nbytes;
 657                 sg++;
 658                 npages++;
 659         }
 660
 661 out:
 662         return ret;
 663 }
 664
 665 static int
 666 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
 667 {
 668         int sgl_count = 0;
 669
 670         if (!iter || !iter->iov) {
 671                 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
 672                        " present\n", __func__, bytes);
 673                 return -EINVAL;
 674         }
 675
 676         sgl_count = iov_iter_npages(iter, 0xffff);
 677         if (sgl_count > max_sgls) {
 678                 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
 679                        " max_sgls: %d\n", __func__, sgl_count, max_sgls);
 680                 return -EINVAL;
 681         }
 682         return sgl_count;
 683 }
 684
 685 static int
 686 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
 687                       struct iov_iter *iter,
 688                       struct scatterlist *sg, int sg_count)
 689 {
 690         size_t off = iter->iov_offset;
 691         int i, ret;
 692
 693         for (i = 0; i < iter->nr_segs; i++) {
 694                 void __user *base = iter->iov[i].iov_base + off;
 695                 size_t len = iter->iov[i].iov_len - off;
 696
 697                 ret = vhost_scsi_map_to_sgl(cmd, base, len, sg, write);
 698                 if (ret < 0) {
 699                         for (i = 0; i < sg_count; i++) {
 700                                 struct page *page = sg_page(&sg[i]);
 701                                 if (page)
 702                                         put_page(page);
 703                         }
 704                         return ret;
 705                 }
 706                 sg += ret;
 707                 off = 0;
 708         }
 709         return 0;
 710 }
 711
 712 static int
 713 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
 714                  size_t prot_bytes, struct iov_iter *prot_iter,
 715                  size_t data_bytes, struct iov_iter *data_iter)
 716 {
 717         int sgl_count, ret;
 718         bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
 719
 720         if (prot_bytes) {
 721                 sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
 722                                                  VHOST_SCSI_PREALLOC_PROT_SGLS);
 723                 if (sgl_count < 0)
 724                         return sgl_count;
 725
 726                 sg_init_table(cmd->tvc_prot_sgl, sgl_count);
 727                 cmd->tvc_prot_sgl_count = sgl_count;
 728                 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
 729                          cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
 730
 731                 ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
 732                                             cmd->tvc_prot_sgl,
 733                                             cmd->tvc_prot_sgl_count);
 734                 if (ret < 0) {
 735                         cmd->tvc_prot_sgl_count = 0;
 736                         return ret;
 737                 }
 738         }
 739         sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
 740                                          VHOST_SCSI_PREALLOC_SGLS);
 741         if (sgl_count < 0)
 742                 return sgl_count;
 743
 744         sg_init_table(cmd->tvc_sgl, sgl_count);
 745         cmd->tvc_sgl_count = sgl_count;
 746         pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
 747                   cmd->tvc_sgl, cmd->tvc_sgl_count);
 748
 749         ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
 750                                     cmd->tvc_sgl, cmd->tvc_sgl_count);
 751         if (ret < 0) {
 752                 cmd->tvc_sgl_count = 0;
 753                 return ret;
 754         }
 755         return 0;
 756 }
 757
 758 static int vhost_scsi_to_tcm_attr(int attr)
 759 {
 760         switch (attr) {
 761         case VIRTIO_SCSI_S_SIMPLE:
 762                 return TCM_SIMPLE_TAG;
 763         case VIRTIO_SCSI_S_ORDERED:
 764                 return TCM_ORDERED_TAG;
 765         case VIRTIO_SCSI_S_HEAD:
 766                 return TCM_HEAD_TAG;
 767         case VIRTIO_SCSI_S_ACA:
 768                 return TCM_ACA_TAG;
 769         default:
 770                 break;
 771         }
 772         return TCM_SIMPLE_TAG;
 773 }
 774
 775 static void vhost_scsi_submission_work(struct work_struct *work)
 776 {
 777         struct vhost_scsi_cmd *cmd =
 778                 container_of(work, struct vhost_scsi_cmd, work);
 779         struct vhost_scsi_nexus *tv_nexus;
 780         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
 781         struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
 782         int rc;
 783
 784         /* FIXME: BIDI operation */
 785         if (cmd->tvc_sgl_count) {
 786                 sg_ptr = cmd->tvc_sgl;
 787
 788                 if (cmd->tvc_prot_sgl_count)
 789                         sg_prot_ptr = cmd->tvc_prot_sgl;
 790                 else
 791                         se_cmd->prot_pto = true;
 792         } else {
 793                 sg_ptr = NULL;
 794         }
 795         tv_nexus = cmd->tvc_nexus;
 796
 797         se_cmd->tag = 0;
 798         rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
 799                         cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
 800                         cmd->tvc_lun, cmd->tvc_exp_data_len,
 801                         vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
 802                         cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
 803                         sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
 804                         cmd->tvc_prot_sgl_count);
 805         if (rc < 0) {
 806                 transport_send_check_condition_and_sense(se_cmd,
 807                                 TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
 808                 transport_generic_free_cmd(se_cmd, 0);
 809         }
 810 }
 811
 812 static void
 813 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
 814                            struct vhost_virtqueue *vq,
 815                            int head, unsigned out)
 816 {
 817         struct virtio_scsi_cmd_resp __user *resp;
 818         struct virtio_scsi_cmd_resp rsp;
 819         int ret;
 820
 821         memset(&rsp, 0, sizeof(rsp));
 822         rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
 823         resp = vq->iov[out].iov_base;
 824         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
 825         if (!ret)
 826                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
 827         else
 828                 pr_err("Faulted on virtio_scsi_cmd_resp\n");
 829 }
 830
 831 static void
 832 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
 833 {
 834         struct vhost_scsi_tpg **vs_tpg, *tpg;
 835         struct virtio_scsi_cmd_req v_req;
 836         struct virtio_scsi_cmd_req_pi v_req_pi;
 837         struct vhost_scsi_cmd *cmd;
 838         struct iov_iter out_iter, in_iter, prot_iter, data_iter;
 839         u64 tag;
 840         u32 exp_data_len, data_direction;
 841         unsigned int out = 0, in = 0;
 842         int head, ret, prot_bytes;
 843         size_t req_size, rsp_size = sizeof(struct virtio_scsi_cmd_resp);
 844         size_t out_size, in_size;
 845         u16 lun;
 846         u8 *target, *lunp, task_attr;
 847         bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
 848         void *req, *cdb;
 849
 850         mutex_lock(&vq->mutex);
 851         /*
 852          * We can handle the vq only after the endpoint is setup by calling the
 853          * VHOST_SCSI_SET_ENDPOINT ioctl.
 854          */
 855         vs_tpg = vq->private_data;
 856         if (!vs_tpg)
 857                 goto out;
 858
 859         vhost_disable_notify(&vs->dev, vq);
 860
 861         for (;;) {
 862                 head = vhost_get_vq_desc(vq, vq->iov,
 863                                          ARRAY_SIZE(vq->iov), &out, &in,
 864                                          NULL, NULL);
 865                 pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
 866                          head, out, in);
 867                 /* On error, stop handling until the next kick. */
 868                 if (unlikely(head < 0))
 869                         break;
 870                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
 871                 if (head == vq->num) {
 872                         if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
 873                                 vhost_disable_notify(&vs->dev, vq);
 874                                 continue;
 875                         }
 876                         break;
 877                 }
 878                 /*
 879                  * Check for a sane response buffer so we can report early
 880                  * errors back to the guest.
 881                  */
 882                 if (unlikely(vq->iov[out].iov_len < rsp_size)) {
 883                         vq_err(vq, "Expecting at least virtio_scsi_cmd_resp"
 884                                 " size, got %zu bytes\n", vq->iov[out].iov_len);
 885                         break;
 886                 }
 887                 /*
 888                  * Setup pointers and values based upon different virtio-scsi
 889                  * request header if T10_PI is enabled in KVM guest.
 890                  */
 891                 if (t10_pi) {
 892                         req = &v_req_pi;
 893                         req_size = sizeof(v_req_pi);
 894                         lunp = &v_req_pi.lun[0];
 895                         target = &v_req_pi.lun[1];
 896                 } else {
 897                         req = &v_req;
 898                         req_size = sizeof(v_req);
 899                         lunp = &v_req.lun[0];
 900                         target = &v_req.lun[1];
 901                 }
 902                 /*
 903                  * FIXME: Not correct for BIDI operation
 904                  */
 905                 out_size = iov_length(vq->iov, out);
 906                 in_size = iov_length(&vq->iov[out], in);
 907
 908                 /*
 909                  * Copy over the virtio-scsi request header, which for a
 910                  * ANY_LAYOUT enabled guest may span multiple iovecs, or a
 911                  * single iovec may contain both the header + outgoing
 912                  * WRITE payloads.
 913                  *
 914                  * copy_from_iter() will advance out_iter, so that it will
 915                  * point at the start of the outgoing WRITE payload, if
 916                  * DMA_TO_DEVICE is set.
 917                  */
 918                 iov_iter_init(&out_iter, WRITE, vq->iov, out, out_size);
 919
 920                 if (unlikely(!copy_from_iter_full(req, req_size, &out_iter))) {
 921                         vq_err(vq, "Faulted on copy_from_iter\n");
 922                         vhost_scsi_send_bad_target(vs, vq, head, out);
 923                         continue;
 924                 }
 925                 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
 926                 if (unlikely(*lunp != 1)) {
 927                         vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp);
 928                         vhost_scsi_send_bad_target(vs, vq, head, out);
 929                         continue;
 930                 }
 931
 932                 tpg = ACCESS_ONCE(vs_tpg[*target]);
 933                 if (unlikely(!tpg)) {
 934                         /* Target does not exist, fail the request */
 935                         vhost_scsi_send_bad_target(vs, vq, head, out);
 936                         continue;
 937                 }
 938                 /*
 939                  * Determine data_direction by calculating the total outgoing
 940                  * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
 941                  * response headers respectively.
 942                  *
 943                  * For DMA_TO_DEVICE this is out_iter, which is already pointing
 944                  * to the right place.
 945                  *
 946                  * For DMA_FROM_DEVICE, the iovec will be just past the end
 947                  * of the virtio-scsi response header in either the same
 948                  * or immediately following iovec.
 949                  *
 950                  * Any associated T10_PI bytes for the outgoing / incoming
 951                  * payloads are included in calculation of exp_data_len here.
 952                  */
 953                 prot_bytes = 0;
 954
 955                 if (out_size > req_size) {
 956                         data_direction = DMA_TO_DEVICE;
 957                         exp_data_len = out_size - req_size;
 958                         data_iter = out_iter;
 959                 } else if (in_size > rsp_size) {
 960                         data_direction = DMA_FROM_DEVICE;
 961                         exp_data_len = in_size - rsp_size;
 962
 963                         iov_iter_init(&in_iter, READ, &vq->iov[out], in,
 964                                       rsp_size + exp_data_len);
 965                         iov_iter_advance(&in_iter, rsp_size);
 966                         data_iter = in_iter;
 967                 } else {
 968                         data_direction = DMA_NONE;
 969                         exp_data_len = 0;
 970                 }
 971                 /*
 972                  * If T10_PI header + payload is present, setup prot_iter values
 973                  * and recalculate data_iter for vhost_scsi_mapal() mapping to
 974                  * host scatterlists via get_user_pages_fast().
 975                  */
 976                 if (t10_pi) {
 977                         if (v_req_pi.pi_bytesout) {
 978                                 if (data_direction != DMA_TO_DEVICE) {
 979                                         vq_err(vq, "Received non zero pi_bytesout,"
 980                                                 " but wrong data_direction\n");
 981                                         vhost_scsi_send_bad_target(vs, vq, head, out);
 982                                         continue;
 983                                 }
 984                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
 985                         } else if (v_req_pi.pi_bytesin) {
 986                                 if (data_direction != DMA_FROM_DEVICE) {
 987                                         vq_err(vq, "Received non zero pi_bytesin,"
 988                                                 " but wrong data_direction\n");
 989                                         vhost_scsi_send_bad_target(vs, vq, head, out);
 990                                         continue;
 991                                 }
 992                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
 993                         }
 994                         /*
 995                          * Set prot_iter to data_iter, and advance past any
 996                          * preceeding prot_bytes that may be present.
 997                          *
 998                          * Also fix up the exp_data_len to reflect only the
 999                          * actual data payload length.
1000                          */
1001                         if (prot_bytes) {
1002                                 exp_data_len -= prot_bytes;
1003                                 prot_iter = data_iter;
1004                                 iov_iter_advance(&data_iter, prot_bytes);
1005                         }
1006                         tag = vhost64_to_cpu(vq, v_req_pi.tag);
1007                         task_attr = v_req_pi.task_attr;
1008                         cdb = &v_req_pi.cdb[0];
1009                         lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
1010                 } else {
1011                         tag = vhost64_to_cpu(vq, v_req.tag);
1012                         task_attr = v_req.task_attr;
1013                         cdb = &v_req.cdb[0];
1014                         lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1015                 }
1016                 /*
1017                  * Check that the received CDB size does not exceeded our
1018                  * hardcoded max for vhost-scsi, then get a pre-allocated
1019                  * cmd descriptor for the new virtio-scsi tag.
1020                  *
1021                  * TODO what if cdb was too small for varlen cdb header?
1022                  */
1023                 if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1024                         vq_err(vq, "Received SCSI CDB with command_size: %d that"
1025                                 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1026                                 scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1027                         vhost_scsi_send_bad_target(vs, vq, head, out);
1028                         continue;
1029                 }
1030                 cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1031                                          exp_data_len + prot_bytes,
1032                                          data_direction);
1033                 if (IS_ERR(cmd)) {
1034                         vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1035                                PTR_ERR(cmd));
1036                         vhost_scsi_send_bad_target(vs, vq, head, out);
1037                         continue;
1038                 }
1039                 cmd->tvc_vhost = vs;
1040                 cmd->tvc_vq = vq;
1041                 cmd->tvc_resp_iov = vq->iov[out];
1042                 cmd->tvc_in_iovs = in;
1043
1044                 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1045                          cmd->tvc_cdb[0], cmd->tvc_lun);
1046                 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1047                          " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1048
1049                 if (data_direction != DMA_NONE) {
1050                         ret = vhost_scsi_mapal(cmd,
1051                                                prot_bytes, &prot_iter,
1052                                                exp_data_len, &data_iter);
1053                         if (unlikely(ret)) {
1054                                 vq_err(vq, "Failed to map iov to sgl\n");
1055                                 vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1056                                 vhost_scsi_send_bad_target(vs, vq, head, out);
1057                                 continue;
1058                         }
1059                 }
1060                 /*
1061                  * Save the descriptor from vhost_get_vq_desc() to be used to
1062                  * complete the virtio-scsi request in TCM callback context via
1063                  * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1064                  */
1065                 cmd->tvc_vq_desc = head;
1066                 /*
1067                  * Dispatch cmd descriptor for cmwq execution in process
1068                  * context provided by vhost_scsi_workqueue.  This also ensures
1069                  * cmd is executed on the same kworker CPU as this vhost
1070                  * thread to gain positive L2 cache locality effects.
1071                  */
1072                 INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1073                 queue_work(vhost_scsi_workqueue, &cmd->work);
1074         }
1075 out:
1076         mutex_unlock(&vq->mutex);
1077 }
1078
1079 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1080 {
1081         pr_debug("%s: The handling func for control queue.\n", __func__);
1082 }
1083
1084 static void
1085 vhost_scsi_send_evt(struct vhost_scsi *vs,
1086                    struct vhost_scsi_tpg *tpg,
1087                    struct se_lun *lun,
1088                    u32 event,
1089                    u32 reason)
1090 {
1091         struct vhost_scsi_evt *evt;
1092
1093         evt = vhost_scsi_allocate_evt(vs, event, reason);
1094         if (!evt)
1095                 return;
1096
1097         if (tpg && lun) {
1098                 /* TODO: share lun setup code with virtio-scsi.ko */
1099                 /*
1100                  * Note: evt->event is zeroed when we allocate it and
1101                  * lun[4-7] need to be zero according to virtio-scsi spec.
1102                  */
1103                 evt->event.lun[0] = 0x01;
1104                 evt->event.lun[1] = tpg->tport_tpgt;
1105                 if (lun->unpacked_lun >= 256)
1106                         evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1107                 evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1108         }
1109
1110         llist_add(&evt->list, &vs->vs_event_list);
1111         vhost_work_queue(&vs->dev, &vs->vs_event_work);
1112 }
1113
1114 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1115 {
1116         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1117                                                 poll.work);
1118         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1119
1120         mutex_lock(&vq->mutex);
1121         if (!vq->private_data)
1122                 goto out;
1123
1124         if (vs->vs_events_missed)
1125                 vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1126 out:
1127         mutex_unlock(&vq->mutex);
1128 }
1129
1130 static void vhost_scsi_handle_kick(struct vhost_work *work)
1131 {
1132         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1133                                                 poll.work);
1134         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1135
1136         vhost_scsi_handle_vq(vs, vq);
1137 }
1138
1139 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1140 {
1141         vhost_poll_flush(&vs->vqs[index].vq.poll);
1142 }
1143
1144 /* Callers must hold dev mutex */
1145 static void vhost_scsi_flush(struct vhost_scsi *vs)
1146 {
1147         struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1148         int i;
1149
1150         /* Init new inflight and remember the old inflight */
1151         vhost_scsi_init_inflight(vs, old_inflight);
1152
1153         /*
1154          * The inflight->kref was initialized to 1. We decrement it here to
1155          * indicate the start of the flush operation so that it will reach 0
1156          * when all the reqs are finished.
1157          */
1158         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1159                 kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1160
1161         /* Flush both the vhost poll and vhost work */
1162         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1163                 vhost_scsi_flush_vq(vs, i);
1164         vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1165         vhost_work_flush(&vs->dev, &vs->vs_event_work);
1166
1167         /* Wait for all reqs issued before the flush to be finished */
1168         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1169                 wait_for_completion(&old_inflight[i]->comp);
1170 }
1171
1172 /*
1173  * Called from vhost_scsi_ioctl() context to walk the list of available
1174  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1175  *
1176  *  The lock nesting rule is:
1177  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1178  */
1179 static int
1180 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1181                         struct vhost_scsi_target *t)
1182 {
1183         struct se_portal_group *se_tpg;
1184         struct vhost_scsi_tport *tv_tport;
1185         struct vhost_scsi_tpg *tpg;
1186         struct vhost_scsi_tpg **vs_tpg;
1187         struct vhost_virtqueue *vq;
1188         int index, ret, i, len;
1189         bool match = false;
1190
1191         mutex_lock(&vhost_scsi_mutex);
1192         mutex_lock(&vs->dev.mutex);
1193
1194         /* Verify that ring has been setup correctly. */
1195         for (index = 0; index < vs->dev.nvqs; ++index) {
1196                 /* Verify that ring has been setup correctly. */
1197                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1198                         ret = -EFAULT;
1199                         goto out;
1200                 }
1201         }
1202
1203         len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1204         vs_tpg = kzalloc(len, GFP_KERNEL);
1205         if (!vs_tpg) {
1206                 ret = -ENOMEM;
1207                 goto out;
1208         }
1209         if (vs->vs_tpg)
1210                 memcpy(vs_tpg, vs->vs_tpg, len);
1211
1212         list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1213                 mutex_lock(&tpg->tv_tpg_mutex);
1214                 if (!tpg->tpg_nexus) {
1215                         mutex_unlock(&tpg->tv_tpg_mutex);
1216                         continue;
1217                 }
1218                 if (tpg->tv_tpg_vhost_count != 0) {
1219                         mutex_unlock(&tpg->tv_tpg_mutex);
1220                         continue;
1221                 }
1222                 tv_tport = tpg->tport;
1223
1224                 if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1225                         if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1226                                 kfree(vs_tpg);
1227                                 mutex_unlock(&tpg->tv_tpg_mutex);
1228                                 ret = -EEXIST;
1229                                 goto out;
1230                         }
1231                         /*
1232                          * In order to ensure individual vhost-scsi configfs
1233                          * groups cannot be removed while in use by vhost ioctl,
1234                          * go ahead and take an explicit se_tpg->tpg_group.cg_item
1235                          * dependency now.
1236                          */
1237                         se_tpg = &tpg->se_tpg;
1238                         ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1239                         if (ret) {
1240                                 pr_warn("configfs_depend_item() failed: %d\n", ret);
1241                                 kfree(vs_tpg);
1242                                 mutex_unlock(&tpg->tv_tpg_mutex);
1243                                 goto out;
1244                         }
1245                         tpg->tv_tpg_vhost_count++;
1246                         tpg->vhost_scsi = vs;
1247                         vs_tpg[tpg->tport_tpgt] = tpg;
1248                         smp_mb__after_atomic();
1249                         match = true;
1250                 }
1251                 mutex_unlock(&tpg->tv_tpg_mutex);
1252         }
1253
1254         if (match) {
1255                 memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1256                        sizeof(vs->vs_vhost_wwpn));
1257                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1258                         vq = &vs->vqs[i].vq;
1259                         mutex_lock(&vq->mutex);
1260                         vq->private_data = vs_tpg;
1261                         vhost_vq_init_access(vq);
1262                         mutex_unlock(&vq->mutex);
1263                 }
1264                 ret = 0;
1265         } else {
1266                 ret = -EEXIST;
1267         }
1268
1269         /*
1270          * Act as synchronize_rcu to make sure access to
1271          * old vs->vs_tpg is finished.
1272          */
1273         vhost_scsi_flush(vs);
1274         kfree(vs->vs_tpg);
1275         vs->vs_tpg = vs_tpg;
1276
1277 out:
1278         mutex_unlock(&vs->dev.mutex);
1279         mutex_unlock(&vhost_scsi_mutex);
1280         return ret;
1281 }
1282
1283 static int
1284 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1285                           struct vhost_scsi_target *t)
1286 {
1287         struct se_portal_group *se_tpg;
1288         struct vhost_scsi_tport *tv_tport;
1289         struct vhost_scsi_tpg *tpg;
1290         struct vhost_virtqueue *vq;
1291         bool match = false;
1292         int index, ret, i;
1293         u8 target;
1294
1295         mutex_lock(&vhost_scsi_mutex);
1296         mutex_lock(&vs->dev.mutex);
1297         /* Verify that ring has been setup correctly. */
1298         for (index = 0; index < vs->dev.nvqs; ++index) {
1299                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1300                         ret = -EFAULT;
1301                         goto err_dev;
1302                 }
1303         }
1304
1305         if (!vs->vs_tpg) {
1306                 ret = 0;
1307                 goto err_dev;
1308         }
1309
1310         for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1311                 target = i;
1312                 tpg = vs->vs_tpg[target];
1313                 if (!tpg)
1314                         continue;
1315
1316                 mutex_lock(&tpg->tv_tpg_mutex);
1317                 tv_tport = tpg->tport;
1318                 if (!tv_tport) {
1319                         ret = -ENODEV;
1320                         goto err_tpg;
1321                 }
1322
1323                 if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1324                         pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1325                                 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1326                                 tv_tport->tport_name, tpg->tport_tpgt,
1327                                 t->vhost_wwpn, t->vhost_tpgt);
1328                         ret = -EINVAL;
1329                         goto err_tpg;
1330                 }
1331                 tpg->tv_tpg_vhost_count--;
1332                 tpg->vhost_scsi = NULL;
1333                 vs->vs_tpg[target] = NULL;
1334                 match = true;
1335                 mutex_unlock(&tpg->tv_tpg_mutex);
1336                 /*
1337                  * Release se_tpg->tpg_group.cg_item configfs dependency now
1338                  * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1339                  */
1340                 se_tpg = &tpg->se_tpg;
1341                 target_undepend_item(&se_tpg->tpg_group.cg_item);
1342         }
1343         if (match) {
1344                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1345                         vq = &vs->vqs[i].vq;
1346                         mutex_lock(&vq->mutex);
1347                         vq->private_data = NULL;
1348                         mutex_unlock(&vq->mutex);
1349                 }
1350         }
1351         /*
1352          * Act as synchronize_rcu to make sure access to
1353          * old vs->vs_tpg is finished.
1354          */
1355         vhost_scsi_flush(vs);
1356         kfree(vs->vs_tpg);
1357         vs->vs_tpg = NULL;
1358         WARN_ON(vs->vs_events_nr);
1359         mutex_unlock(&vs->dev.mutex);
1360         mutex_unlock(&vhost_scsi_mutex);
1361         return 0;
1362
1363 err_tpg:
1364         mutex_unlock(&tpg->tv_tpg_mutex);
1365 err_dev:
1366         mutex_unlock(&vs->dev.mutex);
1367         mutex_unlock(&vhost_scsi_mutex);
1368         return ret;
1369 }
1370
1371 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1372 {
1373         struct vhost_virtqueue *vq;
1374         int i;
1375
1376         if (features & ~VHOST_SCSI_FEATURES)
1377                 return -EOPNOTSUPP;
1378
1379         mutex_lock(&vs->dev.mutex);
1380         if ((features & (1 << VHOST_F_LOG_ALL)) &&
1381             !vhost_log_access_ok(&vs->dev)) {
1382                 mutex_unlock(&vs->dev.mutex);
1383                 return -EFAULT;
1384         }
1385
1386         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1387                 vq = &vs->vqs[i].vq;
1388                 mutex_lock(&vq->mutex);
1389                 vq->acked_features = features;
1390                 mutex_unlock(&vq->mutex);
1391         }
1392         mutex_unlock(&vs->dev.mutex);
1393         return 0;
1394 }
1395
1396 static int vhost_scsi_open(struct inode *inode, struct file *f)
1397 {
1398         struct vhost_scsi *vs;
1399         struct vhost_virtqueue **vqs;
1400         int r = -ENOMEM, i;
1401
1402         vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_RETRY_MAYFAIL);
1403         if (!vs) {
1404                 vs = vzalloc(sizeof(*vs));
1405                 if (!vs)
1406                         goto err_vs;
1407         }
1408
1409         vqs = kmalloc(VHOST_SCSI_MAX_VQ * sizeof(*vqs), GFP_KERNEL);
1410         if (!vqs)
1411                 goto err_vqs;
1412
1413         vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1414         vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1415
1416         vs->vs_events_nr = 0;
1417         vs->vs_events_missed = false;
1418
1419         vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1420         vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1421         vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1422         vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1423         for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1424                 vqs[i] = &vs->vqs[i].vq;
1425                 vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1426         }
1427         vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ);
1428
1429         vhost_scsi_init_inflight(vs, NULL);
1430
1431         f->private_data = vs;
1432         return 0;
1433
1434 err_vqs:
1435         kvfree(vs);
1436 err_vs:
1437         return r;
1438 }
1439
1440 static int vhost_scsi_release(struct inode *inode, struct file *f)
1441 {
1442         struct vhost_scsi *vs = f->private_data;
1443         struct vhost_scsi_target t;
1444
1445         mutex_lock(&vs->dev.mutex);
1446         memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1447         mutex_unlock(&vs->dev.mutex);
1448         vhost_scsi_clear_endpoint(vs, &t);
1449         vhost_dev_stop(&vs->dev);
1450         vhost_dev_cleanup(&vs->dev, false);
1451         /* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1452         vhost_scsi_flush(vs);
1453         kfree(vs->dev.vqs);
1454         kvfree(vs);
1455         return 0;
1456 }
1457
1458 static long
1459 vhost_scsi_ioctl(struct file *f,
1460                  unsigned int ioctl,
1461                  unsigned long arg)
1462 {
1463         struct vhost_scsi *vs = f->private_data;
1464         struct vhost_scsi_target backend;
1465         void __user *argp = (void __user *)arg;
1466         u64 __user *featurep = argp;
1467         u32 __user *eventsp = argp;
1468         u32 events_missed;
1469         u64 features;
1470         int r, abi_version = VHOST_SCSI_ABI_VERSION;
1471         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1472
1473         switch (ioctl) {
1474         case VHOST_SCSI_SET_ENDPOINT:
1475                 if (copy_from_user(&backend, argp, sizeof backend))
1476                         return -EFAULT;
1477                 if (backend.reserved != 0)
1478                         return -EOPNOTSUPP;
1479
1480                 return vhost_scsi_set_endpoint(vs, &backend);
1481         case VHOST_SCSI_CLEAR_ENDPOINT:
1482                 if (copy_from_user(&backend, argp, sizeof backend))
1483                         return -EFAULT;
1484                 if (backend.reserved != 0)
1485                         return -EOPNOTSUPP;
1486
1487                 return vhost_scsi_clear_endpoint(vs, &backend);
1488         case VHOST_SCSI_GET_ABI_VERSION:
1489                 if (copy_to_user(argp, &abi_version, sizeof abi_version))
1490                         return -EFAULT;
1491                 return 0;
1492         case VHOST_SCSI_SET_EVENTS_MISSED:
1493                 if (get_user(events_missed, eventsp))
1494                         return -EFAULT;
1495                 mutex_lock(&vq->mutex);
1496                 vs->vs_events_missed = events_missed;
1497                 mutex_unlock(&vq->mutex);
1498                 return 0;
1499         case VHOST_SCSI_GET_EVENTS_MISSED:
1500                 mutex_lock(&vq->mutex);
1501                 events_missed = vs->vs_events_missed;
1502                 mutex_unlock(&vq->mutex);
1503                 if (put_user(events_missed, eventsp))
1504                         return -EFAULT;
1505                 return 0;
1506         case VHOST_GET_FEATURES:
1507                 features = VHOST_SCSI_FEATURES;
1508                 if (copy_to_user(featurep, &features, sizeof features))
1509                         return -EFAULT;
1510                 return 0;
1511         case VHOST_SET_FEATURES:
1512                 if (copy_from_user(&features, featurep, sizeof features))
1513                         return -EFAULT;
1514                 return vhost_scsi_set_features(vs, features);
1515         default:
1516                 mutex_lock(&vs->dev.mutex);
1517                 r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1518                 /* TODO: flush backend after dev ioctl. */
1519                 if (r == -ENOIOCTLCMD)
1520                         r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1521                 mutex_unlock(&vs->dev.mutex);
1522                 return r;
1523         }
1524 }
1525
1526 #ifdef CONFIG_COMPAT
1527 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1528                                 unsigned long arg)
1529 {
1530         return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1531 }
1532 #endif
1533
1534 static const struct file_operations vhost_scsi_fops = {
1535         .owner          = THIS_MODULE,
1536         .release        = vhost_scsi_release,
1537         .unlocked_ioctl = vhost_scsi_ioctl,
1538 #ifdef CONFIG_COMPAT
1539         .compat_ioctl   = vhost_scsi_compat_ioctl,
1540 #endif
1541         .open           = vhost_scsi_open,
1542         .llseek         = noop_llseek,
1543 };
1544
1545 static struct miscdevice vhost_scsi_misc = {
1546         MISC_DYNAMIC_MINOR,
1547         "vhost-scsi",
1548         &vhost_scsi_fops,
1549 };
1550
1551 static int __init vhost_scsi_register(void)
1552 {
1553         return misc_register(&vhost_scsi_misc);
1554 }
1555
1556 static void vhost_scsi_deregister(void)
1557 {
1558         misc_deregister(&vhost_scsi_misc);
1559 }
1560
1561 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1562 {
1563         switch (tport->tport_proto_id) {
1564         case SCSI_PROTOCOL_SAS:
1565                 return "SAS";
1566         case SCSI_PROTOCOL_FCP:
1567                 return "FCP";
1568         case SCSI_PROTOCOL_ISCSI:
1569                 return "iSCSI";
1570         default:
1571                 break;
1572         }
1573
1574         return "Unknown";
1575 }
1576
1577 static void
1578 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1579                   struct se_lun *lun, bool plug)
1580 {
1581
1582         struct vhost_scsi *vs = tpg->vhost_scsi;
1583         struct vhost_virtqueue *vq;
1584         u32 reason;
1585
1586         if (!vs)
1587                 return;
1588
1589         mutex_lock(&vs->dev.mutex);
1590
1591         if (plug)
1592                 reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1593         else
1594                 reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1595
1596         vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1597         mutex_lock(&vq->mutex);
1598         if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1599                 vhost_scsi_send_evt(vs, tpg, lun,
1600                                    VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1601         mutex_unlock(&vq->mutex);
1602         mutex_unlock(&vs->dev.mutex);
1603 }
1604
1605 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1606 {
1607         vhost_scsi_do_plug(tpg, lun, true);
1608 }
1609
1610 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1611 {
1612         vhost_scsi_do_plug(tpg, lun, false);
1613 }
1614
1615 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1616                                struct se_lun *lun)
1617 {
1618         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1619                                 struct vhost_scsi_tpg, se_tpg);
1620
1621         mutex_lock(&vhost_scsi_mutex);
1622
1623         mutex_lock(&tpg->tv_tpg_mutex);
1624         tpg->tv_tpg_port_count++;
1625         mutex_unlock(&tpg->tv_tpg_mutex);
1626
1627         vhost_scsi_hotplug(tpg, lun);
1628
1629         mutex_unlock(&vhost_scsi_mutex);
1630
1631         return 0;
1632 }
1633
1634 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1635                                   struct se_lun *lun)
1636 {
1637         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1638                                 struct vhost_scsi_tpg, se_tpg);
1639
1640         mutex_lock(&vhost_scsi_mutex);
1641
1642         mutex_lock(&tpg->tv_tpg_mutex);
1643         tpg->tv_tpg_port_count--;
1644         mutex_unlock(&tpg->tv_tpg_mutex);
1645
1646         vhost_scsi_hotunplug(tpg, lun);
1647
1648         mutex_unlock(&vhost_scsi_mutex);
1649 }
1650
1651 static void vhost_scsi_free_cmd_map_res(struct se_session *se_sess)
1652 {
1653         struct vhost_scsi_cmd *tv_cmd;
1654         unsigned int i;
1655
1656         if (!se_sess->sess_cmd_map)
1657                 return;
1658
1659         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1660                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1661
1662                 kfree(tv_cmd->tvc_sgl);
1663                 kfree(tv_cmd->tvc_prot_sgl);
1664                 kfree(tv_cmd->tvc_upages);
1665         }
1666 }
1667
1668 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1669                 struct config_item *item, const char *page, size_t count)
1670 {
1671         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1672         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1673                                 struct vhost_scsi_tpg, se_tpg);
1674         unsigned long val;
1675         int ret = kstrtoul(page, 0, &val);
1676
1677         if (ret) {
1678                 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1679                 return ret;
1680         }
1681         if (val != 0 && val != 1 && val != 3) {
1682                 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1683                 return -EINVAL;
1684         }
1685         tpg->tv_fabric_prot_type = val;
1686
1687         return count;
1688 }
1689
1690 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1691                 struct config_item *item, char *page)
1692 {
1693         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1694         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1695                                 struct vhost_scsi_tpg, se_tpg);
1696
1697         return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1698 }
1699
1700 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1701
1702 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1703         &vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1704         NULL,
1705 };
1706
1707 static int vhost_scsi_nexus_cb(struct se_portal_group *se_tpg,
1708                                struct se_session *se_sess, void *p)
1709 {
1710         struct vhost_scsi_cmd *tv_cmd;
1711         unsigned int i;
1712
1713         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1714                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1715
1716                 tv_cmd->tvc_sgl = kzalloc(sizeof(struct scatterlist) *
1717                                         VHOST_SCSI_PREALLOC_SGLS, GFP_KERNEL);
1718                 if (!tv_cmd->tvc_sgl) {
1719                         pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1720                         goto out;
1721                 }
1722
1723                 tv_cmd->tvc_upages = kzalloc(sizeof(struct page *) *
1724                                 VHOST_SCSI_PREALLOC_UPAGES, GFP_KERNEL);
1725                 if (!tv_cmd->tvc_upages) {
1726                         pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1727                         goto out;
1728                 }
1729
1730                 tv_cmd->tvc_prot_sgl = kzalloc(sizeof(struct scatterlist) *
1731                                 VHOST_SCSI_PREALLOC_PROT_SGLS, GFP_KERNEL);
1732                 if (!tv_cmd->tvc_prot_sgl) {
1733                         pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1734                         goto out;
1735                 }
1736         }
1737         return 0;
1738 out:
1739         vhost_scsi_free_cmd_map_res(se_sess);
1740         return -ENOMEM;
1741 }
1742
1743 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1744                                 const char *name)
1745 {
1746         struct vhost_scsi_nexus *tv_nexus;
1747
1748         mutex_lock(&tpg->tv_tpg_mutex);
1749         if (tpg->tpg_nexus) {
1750                 mutex_unlock(&tpg->tv_tpg_mutex);
1751                 pr_debug("tpg->tpg_nexus already exists\n");
1752                 return -EEXIST;
1753         }
1754
1755         tv_nexus = kzalloc(sizeof(struct vhost_scsi_nexus), GFP_KERNEL);
1756         if (!tv_nexus) {
1757                 mutex_unlock(&tpg->tv_tpg_mutex);
1758                 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1759                 return -ENOMEM;
1760         }
1761         /*
1762          * Since we are running in 'demo mode' this call with generate a
1763          * struct se_node_acl for the vhost_scsi struct se_portal_group with
1764          * the SCSI Initiator port name of the passed configfs group 'name'.
1765          */
1766         tv_nexus->tvn_se_sess = target_alloc_session(&tpg->se_tpg,
1767                                         VHOST_SCSI_DEFAULT_TAGS,
1768                                         sizeof(struct vhost_scsi_cmd),
1769                                         TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1770                                         (unsigned char *)name, tv_nexus,
1771                                         vhost_scsi_nexus_cb);
1772         if (IS_ERR(tv_nexus->tvn_se_sess)) {
1773                 mutex_unlock(&tpg->tv_tpg_mutex);
1774                 kfree(tv_nexus);
1775                 return -ENOMEM;
1776         }
1777         tpg->tpg_nexus = tv_nexus;
1778
1779         mutex_unlock(&tpg->tv_tpg_mutex);
1780         return 0;
1781 }
1782
1783 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1784 {
1785         struct se_session *se_sess;
1786         struct vhost_scsi_nexus *tv_nexus;
1787
1788         mutex_lock(&tpg->tv_tpg_mutex);
1789         tv_nexus = tpg->tpg_nexus;
1790         if (!tv_nexus) {
1791                 mutex_unlock(&tpg->tv_tpg_mutex);
1792                 return -ENODEV;
1793         }
1794
1795         se_sess = tv_nexus->tvn_se_sess;
1796         if (!se_sess) {
1797                 mutex_unlock(&tpg->tv_tpg_mutex);
1798                 return -ENODEV;
1799         }
1800
1801         if (tpg->tv_tpg_port_count != 0) {
1802                 mutex_unlock(&tpg->tv_tpg_mutex);
1803                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1804                         " active TPG port count: %d\n",
1805                         tpg->tv_tpg_port_count);
1806                 return -EBUSY;
1807         }
1808
1809         if (tpg->tv_tpg_vhost_count != 0) {
1810                 mutex_unlock(&tpg->tv_tpg_mutex);
1811                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
1812                         " active TPG vhost count: %d\n",
1813                         tpg->tv_tpg_vhost_count);
1814                 return -EBUSY;
1815         }
1816
1817         pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1818                 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1819                 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1820
1821         vhost_scsi_free_cmd_map_res(se_sess);
1822         /*
1823          * Release the SCSI I_T Nexus to the emulated vhost Target Port
1824          */
1825         transport_deregister_session(tv_nexus->tvn_se_sess);
1826         tpg->tpg_nexus = NULL;
1827         mutex_unlock(&tpg->tv_tpg_mutex);
1828
1829         kfree(tv_nexus);
1830         return 0;
1831 }
1832
1833 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
1834 {
1835         struct se_portal_group *se_tpg = to_tpg(item);
1836         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1837                                 struct vhost_scsi_tpg, se_tpg);
1838         struct vhost_scsi_nexus *tv_nexus;
1839         ssize_t ret;
1840
1841         mutex_lock(&tpg->tv_tpg_mutex);
1842         tv_nexus = tpg->tpg_nexus;
1843         if (!tv_nexus) {
1844                 mutex_unlock(&tpg->tv_tpg_mutex);
1845                 return -ENODEV;
1846         }
1847         ret = snprintf(page, PAGE_SIZE, "%s\n",
1848                         tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1849         mutex_unlock(&tpg->tv_tpg_mutex);
1850
1851         return ret;
1852 }
1853
1854 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
1855                 const char *page, size_t count)
1856 {
1857         struct se_portal_group *se_tpg = to_tpg(item);
1858         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1859                                 struct vhost_scsi_tpg, se_tpg);
1860         struct vhost_scsi_tport *tport_wwn = tpg->tport;
1861         unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1862         int ret;
1863         /*
1864          * Shutdown the active I_T nexus if 'NULL' is passed..
1865          */
1866         if (!strncmp(page, "NULL", 4)) {
1867                 ret = vhost_scsi_drop_nexus(tpg);
1868                 return (!ret) ? count : ret;
1869         }
1870         /*
1871          * Otherwise make sure the passed virtual Initiator port WWN matches
1872          * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1873          * vhost_scsi_make_nexus().
1874          */
1875         if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1876                 pr_err("Emulated NAA Sas Address: %s, exceeds"
1877                                 " max: %d\n", page, VHOST_SCSI_NAMELEN);
1878                 return -EINVAL;
1879         }
1880         snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1881
1882         ptr = strstr(i_port, "naa.");
1883         if (ptr) {
1884                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1885                         pr_err("Passed SAS Initiator Port %s does not"
1886                                 " match target port protoid: %s\n", i_port,
1887                                 vhost_scsi_dump_proto_id(tport_wwn));
1888                         return -EINVAL;
1889                 }
1890                 port_ptr = &i_port[0];
1891                 goto check_newline;
1892         }
1893         ptr = strstr(i_port, "fc.");
1894         if (ptr) {
1895                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1896                         pr_err("Passed FCP Initiator Port %s does not"
1897                                 " match target port protoid: %s\n", i_port,
1898                                 vhost_scsi_dump_proto_id(tport_wwn));
1899                         return -EINVAL;
1900                 }
1901                 port_ptr = &i_port[3]; /* Skip over "fc." */
1902                 goto check_newline;
1903         }
1904         ptr = strstr(i_port, "iqn.");
1905         if (ptr) {
1906                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1907                         pr_err("Passed iSCSI Initiator Port %s does not"
1908                                 " match target port protoid: %s\n", i_port,
1909                                 vhost_scsi_dump_proto_id(tport_wwn));
1910                         return -EINVAL;
1911                 }
1912                 port_ptr = &i_port[0];
1913                 goto check_newline;
1914         }
1915         pr_err("Unable to locate prefix for emulated Initiator Port:"
1916                         " %s\n", i_port);
1917         return -EINVAL;
1918         /*
1919          * Clear any trailing newline for the NAA WWN
1920          */
1921 check_newline:
1922         if (i_port[strlen(i_port)-1] == '\n')
1923                 i_port[strlen(i_port)-1] = '\0';
1924
1925         ret = vhost_scsi_make_nexus(tpg, port_ptr);
1926         if (ret < 0)
1927                 return ret;
1928
1929         return count;
1930 }
1931
1932 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
1933
1934 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1935         &vhost_scsi_tpg_attr_nexus,
1936         NULL,
1937 };
1938
1939 static struct se_portal_group *
1940 vhost_scsi_make_tpg(struct se_wwn *wwn,
1941                    struct config_group *group,
1942                    const char *name)
1943 {
1944         struct vhost_scsi_tport *tport = container_of(wwn,
1945                         struct vhost_scsi_tport, tport_wwn);
1946
1947         struct vhost_scsi_tpg *tpg;
1948         u16 tpgt;
1949         int ret;
1950
1951         if (strstr(name, "tpgt_") != name)
1952                 return ERR_PTR(-EINVAL);
1953         if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1954                 return ERR_PTR(-EINVAL);
1955
1956         tpg = kzalloc(sizeof(struct vhost_scsi_tpg), GFP_KERNEL);
1957         if (!tpg) {
1958                 pr_err("Unable to allocate struct vhost_scsi_tpg");
1959                 return ERR_PTR(-ENOMEM);
1960         }
1961         mutex_init(&tpg->tv_tpg_mutex);
1962         INIT_LIST_HEAD(&tpg->tv_tpg_list);
1963         tpg->tport = tport;
1964         tpg->tport_tpgt = tpgt;
1965
1966         ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
1967         if (ret < 0) {
1968                 kfree(tpg);
1969                 return NULL;
1970         }
1971         mutex_lock(&vhost_scsi_mutex);
1972         list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
1973         mutex_unlock(&vhost_scsi_mutex);
1974
1975         return &tpg->se_tpg;
1976 }
1977
1978 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
1979 {
1980         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1981                                 struct vhost_scsi_tpg, se_tpg);
1982
1983         mutex_lock(&vhost_scsi_mutex);
1984         list_del(&tpg->tv_tpg_list);
1985         mutex_unlock(&vhost_scsi_mutex);
1986         /*
1987          * Release the virtual I_T Nexus for this vhost TPG
1988          */
1989         vhost_scsi_drop_nexus(tpg);
1990         /*
1991          * Deregister the se_tpg from TCM..
1992          */
1993         core_tpg_deregister(se_tpg);
1994         kfree(tpg);
1995 }
1996
1997 static struct se_wwn *
1998 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
1999                      struct config_group *group,
2000                      const char *name)
2001 {
2002         struct vhost_scsi_tport *tport;
2003         char *ptr;
2004         u64 wwpn = 0;
2005         int off = 0;
2006
2007         /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2008                 return ERR_PTR(-EINVAL); */
2009
2010         tport = kzalloc(sizeof(struct vhost_scsi_tport), GFP_KERNEL);
2011         if (!tport) {
2012                 pr_err("Unable to allocate struct vhost_scsi_tport");
2013                 return ERR_PTR(-ENOMEM);
2014         }
2015         tport->tport_wwpn = wwpn;
2016         /*
2017          * Determine the emulated Protocol Identifier and Target Port Name
2018          * based on the incoming configfs directory name.
2019          */
2020         ptr = strstr(name, "naa.");
2021         if (ptr) {
2022                 tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2023                 goto check_len;
2024         }
2025         ptr = strstr(name, "fc.");
2026         if (ptr) {
2027                 tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2028                 off = 3; /* Skip over "fc." */
2029                 goto check_len;
2030         }
2031         ptr = strstr(name, "iqn.");
2032         if (ptr) {
2033                 tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2034                 goto check_len;
2035         }
2036
2037         pr_err("Unable to locate prefix for emulated Target Port:"
2038                         " %s\n", name);
2039         kfree(tport);
2040         return ERR_PTR(-EINVAL);
2041
2042 check_len:
2043         if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2044                 pr_err("Emulated %s Address: %s, exceeds"
2045                         " max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2046                         VHOST_SCSI_NAMELEN);
2047                 kfree(tport);
2048                 return ERR_PTR(-EINVAL);
2049         }
2050         snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2051
2052         pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2053                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2054
2055         return &tport->tport_wwn;
2056 }
2057
2058 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2059 {
2060         struct vhost_scsi_tport *tport = container_of(wwn,
2061                                 struct vhost_scsi_tport, tport_wwn);
2062
2063         pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2064                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2065                 tport->tport_name);
2066
2067         kfree(tport);
2068 }
2069
2070 static ssize_t
2071 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2072 {
2073         return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2074                 "on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2075                 utsname()->machine);
2076 }
2077
2078 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2079
2080 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2081         &vhost_scsi_wwn_attr_version,
2082         NULL,
2083 };
2084
2085 static const struct target_core_fabric_ops vhost_scsi_ops = {
2086         .module                         = THIS_MODULE,
2087         .name                           = "vhost",
2088         .get_fabric_name                = vhost_scsi_get_fabric_name,
2089         .tpg_get_wwn                    = vhost_scsi_get_fabric_wwn,
2090         .tpg_get_tag                    = vhost_scsi_get_tpgt,
2091         .tpg_check_demo_mode            = vhost_scsi_check_true,
2092         .tpg_check_demo_mode_cache      = vhost_scsi_check_true,
2093         .tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2094         .tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2095         .tpg_check_prot_fabric_only     = vhost_scsi_check_prot_fabric_only,
2096         .tpg_get_inst_index             = vhost_scsi_tpg_get_inst_index,
2097         .release_cmd                    = vhost_scsi_release_cmd,
2098         .check_stop_free                = vhost_scsi_check_stop_free,
2099         .sess_get_index                 = vhost_scsi_sess_get_index,
2100         .sess_get_initiator_sid         = NULL,
2101         .write_pending                  = vhost_scsi_write_pending,
2102         .write_pending_status           = vhost_scsi_write_pending_status,
2103         .set_default_node_attributes    = vhost_scsi_set_default_node_attrs,
2104         .get_cmd_state                  = vhost_scsi_get_cmd_state,
2105         .queue_data_in                  = vhost_scsi_queue_data_in,
2106         .queue_status                   = vhost_scsi_queue_status,
2107         .queue_tm_rsp                   = vhost_scsi_queue_tm_rsp,
2108         .aborted_task                   = vhost_scsi_aborted_task,
2109         /*
2110          * Setup callers for generic logic in target_core_fabric_configfs.c
2111          */
2112         .fabric_make_wwn                = vhost_scsi_make_tport,
2113         .fabric_drop_wwn                = vhost_scsi_drop_tport,
2114         .fabric_make_tpg                = vhost_scsi_make_tpg,
2115         .fabric_drop_tpg                = vhost_scsi_drop_tpg,
2116         .fabric_post_link               = vhost_scsi_port_link,
2117         .fabric_pre_unlink              = vhost_scsi_port_unlink,
2118
2119         .tfc_wwn_attrs                  = vhost_scsi_wwn_attrs,
2120         .tfc_tpg_base_attrs             = vhost_scsi_tpg_attrs,
2121         .tfc_tpg_attrib_attrs           = vhost_scsi_tpg_attrib_attrs,
2122 };
2123
2124 static int __init vhost_scsi_init(void)
2125 {
2126         int ret = -ENOMEM;
2127
2128         pr_debug("TCM_VHOST fabric module %s on %s/%s"
2129                 " on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2130                 utsname()->machine);
2131
2132         /*
2133          * Use our own dedicated workqueue for submitting I/O into
2134          * target core to avoid contention within system_wq.
2135          */
2136         vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2137         if (!vhost_scsi_workqueue)
2138                 goto out;
2139
2140         ret = vhost_scsi_register();
2141         if (ret < 0)
2142                 goto out_destroy_workqueue;
2143
2144         ret = target_register_template(&vhost_scsi_ops);
2145         if (ret < 0)
2146                 goto out_vhost_scsi_deregister;
2147
2148         return 0;
2149
2150 out_vhost_scsi_deregister:
2151         vhost_scsi_deregister();
2152 out_destroy_workqueue:
2153         destroy_workqueue(vhost_scsi_workqueue);
2154 out:
2155         return ret;
2156 };
2157
2158 static void vhost_scsi_exit(void)
2159 {
2160         target_unregister_template(&vhost_scsi_ops);
2161         vhost_scsi_deregister();
2162         destroy_workqueue(vhost_scsi_workqueue);
2163 };
2164
2165 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2166 MODULE_ALIAS("tcm_vhost");
2167 MODULE_LICENSE("GPL");
2168 module_init(vhost_scsi_init);
2169 module_exit(vhost_scsi_exit);