xref: /openbmc/linux/drivers/vhost/scsi.c (revision e53134fe)
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 
50 #include "vhost.h"
51 
52 #define VHOST_SCSI_VERSION  "v0.1"
53 #define VHOST_SCSI_NAMELEN 256
54 #define VHOST_SCSI_MAX_CDB_SIZE 32
55 #define VHOST_SCSI_DEFAULT_TAGS 256
56 #define VHOST_SCSI_PREALLOC_SGLS 2048
57 #define VHOST_SCSI_PREALLOC_UPAGES 2048
58 #define VHOST_SCSI_PREALLOC_PROT_SGLS 2048
59 
60 struct vhost_scsi_inflight {
61 	/* Wait for the flush operation to finish */
62 	struct completion comp;
63 	/* Refcount for the inflight reqs */
64 	struct kref kref;
65 };
66 
67 struct vhost_scsi_cmd {
68 	/* Descriptor from vhost_get_vq_desc() for virt_queue segment */
69 	int tvc_vq_desc;
70 	/* virtio-scsi initiator task attribute */
71 	int tvc_task_attr;
72 	/* virtio-scsi response incoming iovecs */
73 	int tvc_in_iovs;
74 	/* virtio-scsi initiator data direction */
75 	enum dma_data_direction tvc_data_direction;
76 	/* Expected data transfer length from virtio-scsi header */
77 	u32 tvc_exp_data_len;
78 	/* The Tag from include/linux/virtio_scsi.h:struct virtio_scsi_cmd_req */
79 	u64 tvc_tag;
80 	/* The number of scatterlists associated with this cmd */
81 	u32 tvc_sgl_count;
82 	u32 tvc_prot_sgl_count;
83 	/* Saved unpacked SCSI LUN for vhost_scsi_submission_work() */
84 	u32 tvc_lun;
85 	/* Pointer to the SGL formatted memory from virtio-scsi */
86 	struct scatterlist *tvc_sgl;
87 	struct scatterlist *tvc_prot_sgl;
88 	struct page **tvc_upages;
89 	/* Pointer to response header iovec */
90 	struct iovec tvc_resp_iov;
91 	/* Pointer to vhost_scsi for our device */
92 	struct vhost_scsi *tvc_vhost;
93 	/* Pointer to vhost_virtqueue for the cmd */
94 	struct vhost_virtqueue *tvc_vq;
95 	/* Pointer to vhost nexus memory */
96 	struct vhost_scsi_nexus *tvc_nexus;
97 	/* The TCM I/O descriptor that is accessed via container_of() */
98 	struct se_cmd tvc_se_cmd;
99 	/* work item used for cmwq dispatch to vhost_scsi_submission_work() */
100 	struct work_struct work;
101 	/* Copy of the incoming SCSI command descriptor block (CDB) */
102 	unsigned char tvc_cdb[VHOST_SCSI_MAX_CDB_SIZE];
103 	/* Sense buffer that will be mapped into outgoing status */
104 	unsigned char tvc_sense_buf[TRANSPORT_SENSE_BUFFER];
105 	/* Completed commands list, serviced from vhost worker thread */
106 	struct llist_node tvc_completion_list;
107 	/* Used to track inflight cmd */
108 	struct vhost_scsi_inflight *inflight;
109 };
110 
111 struct vhost_scsi_nexus {
112 	/* Pointer to TCM session for I_T Nexus */
113 	struct se_session *tvn_se_sess;
114 };
115 
116 struct vhost_scsi_tpg {
117 	/* Vhost port target portal group tag for TCM */
118 	u16 tport_tpgt;
119 	/* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
120 	int tv_tpg_port_count;
121 	/* Used for vhost_scsi device reference to tpg_nexus, protected by tv_tpg_mutex */
122 	int tv_tpg_vhost_count;
123 	/* Used for enabling T10-PI with legacy devices */
124 	int tv_fabric_prot_type;
125 	/* list for vhost_scsi_list */
126 	struct list_head tv_tpg_list;
127 	/* Used to protect access for tpg_nexus */
128 	struct mutex tv_tpg_mutex;
129 	/* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
130 	struct vhost_scsi_nexus *tpg_nexus;
131 	/* Pointer back to vhost_scsi_tport */
132 	struct vhost_scsi_tport *tport;
133 	/* Returned by vhost_scsi_make_tpg() */
134 	struct se_portal_group se_tpg;
135 	/* Pointer back to vhost_scsi, protected by tv_tpg_mutex */
136 	struct vhost_scsi *vhost_scsi;
137 };
138 
139 struct vhost_scsi_tport {
140 	/* SCSI protocol the tport is providing */
141 	u8 tport_proto_id;
142 	/* Binary World Wide unique Port Name for Vhost Target port */
143 	u64 tport_wwpn;
144 	/* ASCII formatted WWPN for Vhost Target port */
145 	char tport_name[VHOST_SCSI_NAMELEN];
146 	/* Returned by vhost_scsi_make_tport() */
147 	struct se_wwn tport_wwn;
148 };
149 
150 struct vhost_scsi_evt {
151 	/* event to be sent to guest */
152 	struct virtio_scsi_event event;
153 	/* event list, serviced from vhost worker thread */
154 	struct llist_node list;
155 };
156 
157 enum {
158 	VHOST_SCSI_VQ_CTL = 0,
159 	VHOST_SCSI_VQ_EVT = 1,
160 	VHOST_SCSI_VQ_IO = 2,
161 };
162 
163 /* Note: can't set VIRTIO_F_VERSION_1 yet, since that implies ANY_LAYOUT. */
164 enum {
165 	VHOST_SCSI_FEATURES = VHOST_FEATURES | (1ULL << VIRTIO_SCSI_F_HOTPLUG) |
166 					       (1ULL << VIRTIO_SCSI_F_T10_PI)
167 };
168 
169 #define VHOST_SCSI_MAX_TARGET	256
170 #define VHOST_SCSI_MAX_VQ	128
171 #define VHOST_SCSI_MAX_EVENT	128
172 
173 struct vhost_scsi_virtqueue {
174 	struct vhost_virtqueue vq;
175 	/*
176 	 * Reference counting for inflight reqs, used for flush operation. At
177 	 * each time, one reference tracks new commands submitted, while we
178 	 * wait for another one to reach 0.
179 	 */
180 	struct vhost_scsi_inflight inflights[2];
181 	/*
182 	 * Indicate current inflight in use, protected by vq->mutex.
183 	 * Writers must also take dev mutex and flush under it.
184 	 */
185 	int inflight_idx;
186 };
187 
188 struct vhost_scsi {
189 	/* Protected by vhost_scsi->dev.mutex */
190 	struct vhost_scsi_tpg **vs_tpg;
191 	char vs_vhost_wwpn[TRANSPORT_IQN_LEN];
192 
193 	struct vhost_dev dev;
194 	struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
195 
196 	struct vhost_work vs_completion_work; /* cmd completion work item */
197 	struct llist_head vs_completion_list; /* cmd completion queue */
198 
199 	struct vhost_work vs_event_work; /* evt injection work item */
200 	struct llist_head vs_event_list; /* evt injection queue */
201 
202 	bool vs_events_missed; /* any missed events, protected by vq->mutex */
203 	int vs_events_nr; /* num of pending events, protected by vq->mutex */
204 };
205 
206 static struct workqueue_struct *vhost_scsi_workqueue;
207 
208 /* Global spinlock to protect vhost_scsi TPG list for vhost IOCTL access */
209 static DEFINE_MUTEX(vhost_scsi_mutex);
210 static LIST_HEAD(vhost_scsi_list);
211 
212 static void vhost_scsi_done_inflight(struct kref *kref)
213 {
214 	struct vhost_scsi_inflight *inflight;
215 
216 	inflight = container_of(kref, struct vhost_scsi_inflight, kref);
217 	complete(&inflight->comp);
218 }
219 
220 static void vhost_scsi_init_inflight(struct vhost_scsi *vs,
221 				    struct vhost_scsi_inflight *old_inflight[])
222 {
223 	struct vhost_scsi_inflight *new_inflight;
224 	struct vhost_virtqueue *vq;
225 	int idx, i;
226 
227 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
228 		vq = &vs->vqs[i].vq;
229 
230 		mutex_lock(&vq->mutex);
231 
232 		/* store old infight */
233 		idx = vs->vqs[i].inflight_idx;
234 		if (old_inflight)
235 			old_inflight[i] = &vs->vqs[i].inflights[idx];
236 
237 		/* setup new infight */
238 		vs->vqs[i].inflight_idx = idx ^ 1;
239 		new_inflight = &vs->vqs[i].inflights[idx ^ 1];
240 		kref_init(&new_inflight->kref);
241 		init_completion(&new_inflight->comp);
242 
243 		mutex_unlock(&vq->mutex);
244 	}
245 }
246 
247 static struct vhost_scsi_inflight *
248 vhost_scsi_get_inflight(struct vhost_virtqueue *vq)
249 {
250 	struct vhost_scsi_inflight *inflight;
251 	struct vhost_scsi_virtqueue *svq;
252 
253 	svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
254 	inflight = &svq->inflights[svq->inflight_idx];
255 	kref_get(&inflight->kref);
256 
257 	return inflight;
258 }
259 
260 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight *inflight)
261 {
262 	kref_put(&inflight->kref, vhost_scsi_done_inflight);
263 }
264 
265 static int vhost_scsi_check_true(struct se_portal_group *se_tpg)
266 {
267 	return 1;
268 }
269 
270 static int vhost_scsi_check_false(struct se_portal_group *se_tpg)
271 {
272 	return 0;
273 }
274 
275 static char *vhost_scsi_get_fabric_name(void)
276 {
277 	return "vhost";
278 }
279 
280 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group *se_tpg)
281 {
282 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
283 				struct vhost_scsi_tpg, se_tpg);
284 	struct vhost_scsi_tport *tport = tpg->tport;
285 
286 	return &tport->tport_name[0];
287 }
288 
289 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
290 {
291 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
292 				struct vhost_scsi_tpg, se_tpg);
293 	return tpg->tport_tpgt;
294 }
295 
296 static int vhost_scsi_check_prot_fabric_only(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 
301 	return tpg->tv_fabric_prot_type;
302 }
303 
304 static u32 vhost_scsi_tpg_get_inst_index(struct se_portal_group *se_tpg)
305 {
306 	return 1;
307 }
308 
309 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
310 {
311 	struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
312 				struct vhost_scsi_cmd, tvc_se_cmd);
313 	struct se_session *se_sess = tv_cmd->tvc_nexus->tvn_se_sess;
314 	int i;
315 
316 	if (tv_cmd->tvc_sgl_count) {
317 		for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
318 			put_page(sg_page(&tv_cmd->tvc_sgl[i]));
319 	}
320 	if (tv_cmd->tvc_prot_sgl_count) {
321 		for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
322 			put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
323 	}
324 
325 	vhost_scsi_put_inflight(tv_cmd->inflight);
326 	target_free_tag(se_sess, se_cmd);
327 }
328 
329 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
330 {
331 	return 0;
332 }
333 
334 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
335 {
336 	/* Go ahead and process the write immediately */
337 	target_execute_cmd(se_cmd);
338 	return 0;
339 }
340 
341 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
342 {
343 	return 0;
344 }
345 
346 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
347 {
348 	return;
349 }
350 
351 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
352 {
353 	return 0;
354 }
355 
356 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
357 {
358 	struct vhost_scsi *vs = cmd->tvc_vhost;
359 
360 	llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
361 
362 	vhost_work_queue(&vs->dev, &vs->vs_completion_work);
363 }
364 
365 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
366 {
367 	struct vhost_scsi_cmd *cmd = container_of(se_cmd,
368 				struct vhost_scsi_cmd, tvc_se_cmd);
369 	vhost_scsi_complete_cmd(cmd);
370 	return 0;
371 }
372 
373 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
374 {
375 	struct vhost_scsi_cmd *cmd = container_of(se_cmd,
376 				struct vhost_scsi_cmd, tvc_se_cmd);
377 	vhost_scsi_complete_cmd(cmd);
378 	return 0;
379 }
380 
381 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
382 {
383 	return;
384 }
385 
386 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
387 {
388 	return;
389 }
390 
391 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
392 {
393 	vs->vs_events_nr--;
394 	kfree(evt);
395 }
396 
397 static struct vhost_scsi_evt *
398 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
399 		       u32 event, u32 reason)
400 {
401 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
402 	struct vhost_scsi_evt *evt;
403 
404 	if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
405 		vs->vs_events_missed = true;
406 		return NULL;
407 	}
408 
409 	evt = kzalloc(sizeof(*evt), GFP_KERNEL);
410 	if (!evt) {
411 		vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
412 		vs->vs_events_missed = true;
413 		return NULL;
414 	}
415 
416 	evt->event.event = cpu_to_vhost32(vq, event);
417 	evt->event.reason = cpu_to_vhost32(vq, reason);
418 	vs->vs_events_nr++;
419 
420 	return evt;
421 }
422 
423 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
424 {
425 	struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
426 
427 	/* TODO locking against target/backend threads? */
428 	transport_generic_free_cmd(se_cmd, 0);
429 
430 }
431 
432 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
433 {
434 	return target_put_sess_cmd(se_cmd);
435 }
436 
437 static void
438 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
439 {
440 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
441 	struct virtio_scsi_event *event = &evt->event;
442 	struct virtio_scsi_event __user *eventp;
443 	unsigned out, in;
444 	int head, ret;
445 
446 	if (!vq->private_data) {
447 		vs->vs_events_missed = true;
448 		return;
449 	}
450 
451 again:
452 	vhost_disable_notify(&vs->dev, vq);
453 	head = vhost_get_vq_desc(vq, vq->iov,
454 			ARRAY_SIZE(vq->iov), &out, &in,
455 			NULL, NULL);
456 	if (head < 0) {
457 		vs->vs_events_missed = true;
458 		return;
459 	}
460 	if (head == vq->num) {
461 		if (vhost_enable_notify(&vs->dev, vq))
462 			goto again;
463 		vs->vs_events_missed = true;
464 		return;
465 	}
466 
467 	if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
468 		vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
469 				vq->iov[out].iov_len);
470 		vs->vs_events_missed = true;
471 		return;
472 	}
473 
474 	if (vs->vs_events_missed) {
475 		event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
476 		vs->vs_events_missed = false;
477 	}
478 
479 	eventp = vq->iov[out].iov_base;
480 	ret = __copy_to_user(eventp, event, sizeof(*event));
481 	if (!ret)
482 		vhost_add_used_and_signal(&vs->dev, vq, head, 0);
483 	else
484 		vq_err(vq, "Faulted on vhost_scsi_send_event\n");
485 }
486 
487 static void vhost_scsi_evt_work(struct vhost_work *work)
488 {
489 	struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
490 					vs_event_work);
491 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
492 	struct vhost_scsi_evt *evt, *t;
493 	struct llist_node *llnode;
494 
495 	mutex_lock(&vq->mutex);
496 	llnode = llist_del_all(&vs->vs_event_list);
497 	llist_for_each_entry_safe(evt, t, llnode, list) {
498 		vhost_scsi_do_evt_work(vs, evt);
499 		vhost_scsi_free_evt(vs, evt);
500 	}
501 	mutex_unlock(&vq->mutex);
502 }
503 
504 /* Fill in status and signal that we are done processing this command
505  *
506  * This is scheduled in the vhost work queue so we are called with the owner
507  * process mm and can access the vring.
508  */
509 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
510 {
511 	struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
512 					vs_completion_work);
513 	DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
514 	struct virtio_scsi_cmd_resp v_rsp;
515 	struct vhost_scsi_cmd *cmd, *t;
516 	struct llist_node *llnode;
517 	struct se_cmd *se_cmd;
518 	struct iov_iter iov_iter;
519 	int ret, vq;
520 
521 	bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
522 	llnode = llist_del_all(&vs->vs_completion_list);
523 	llist_for_each_entry_safe(cmd, t, llnode, tvc_completion_list) {
524 		se_cmd = &cmd->tvc_se_cmd;
525 
526 		pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
527 			cmd, se_cmd->residual_count, se_cmd->scsi_status);
528 
529 		memset(&v_rsp, 0, sizeof(v_rsp));
530 		v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
531 		/* TODO is status_qualifier field needed? */
532 		v_rsp.status = se_cmd->scsi_status;
533 		v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
534 						 se_cmd->scsi_sense_length);
535 		memcpy(v_rsp.sense, cmd->tvc_sense_buf,
536 		       se_cmd->scsi_sense_length);
537 
538 		iov_iter_init(&iov_iter, READ, &cmd->tvc_resp_iov,
539 			      cmd->tvc_in_iovs, sizeof(v_rsp));
540 		ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
541 		if (likely(ret == sizeof(v_rsp))) {
542 			struct vhost_scsi_virtqueue *q;
543 			vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
544 			q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
545 			vq = q - vs->vqs;
546 			__set_bit(vq, signal);
547 		} else
548 			pr_err("Faulted on virtio_scsi_cmd_resp\n");
549 
550 		vhost_scsi_free_cmd(cmd);
551 	}
552 
553 	vq = -1;
554 	while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
555 		< VHOST_SCSI_MAX_VQ)
556 		vhost_signal(&vs->dev, &vs->vqs[vq].vq);
557 }
558 
559 static struct vhost_scsi_cmd *
560 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
561 		   unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
562 		   u32 exp_data_len, int data_direction)
563 {
564 	struct vhost_scsi_cmd *cmd;
565 	struct vhost_scsi_nexus *tv_nexus;
566 	struct se_session *se_sess;
567 	struct scatterlist *sg, *prot_sg;
568 	struct page **pages;
569 	int tag, cpu;
570 
571 	tv_nexus = tpg->tpg_nexus;
572 	if (!tv_nexus) {
573 		pr_err("Unable to locate active struct vhost_scsi_nexus\n");
574 		return ERR_PTR(-EIO);
575 	}
576 	se_sess = tv_nexus->tvn_se_sess;
577 
578 	tag = sbitmap_queue_get(&se_sess->sess_tag_pool, &cpu);
579 	if (tag < 0) {
580 		pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
581 		return ERR_PTR(-ENOMEM);
582 	}
583 
584 	cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
585 	sg = cmd->tvc_sgl;
586 	prot_sg = cmd->tvc_prot_sgl;
587 	pages = cmd->tvc_upages;
588 	memset(cmd, 0, sizeof(*cmd));
589 	cmd->tvc_sgl = sg;
590 	cmd->tvc_prot_sgl = prot_sg;
591 	cmd->tvc_upages = pages;
592 	cmd->tvc_se_cmd.map_tag = tag;
593 	cmd->tvc_se_cmd.map_cpu = cpu;
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 = kcalloc(VHOST_SCSI_PREALLOC_SGLS,
1689 					  sizeof(struct scatterlist),
1690 					  GFP_KERNEL);
1691 		if (!tv_cmd->tvc_sgl) {
1692 			pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1693 			goto out;
1694 		}
1695 
1696 		tv_cmd->tvc_upages = kcalloc(VHOST_SCSI_PREALLOC_UPAGES,
1697 					     sizeof(struct page *),
1698 					     GFP_KERNEL);
1699 		if (!tv_cmd->tvc_upages) {
1700 			pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1701 			goto out;
1702 		}
1703 
1704 		tv_cmd->tvc_prot_sgl = kcalloc(VHOST_SCSI_PREALLOC_PROT_SGLS,
1705 					       sizeof(struct scatterlist),
1706 					       GFP_KERNEL);
1707 		if (!tv_cmd->tvc_prot_sgl) {
1708 			pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1709 			goto out;
1710 		}
1711 	}
1712 	return 0;
1713 out:
1714 	vhost_scsi_free_cmd_map_res(se_sess);
1715 	return -ENOMEM;
1716 }
1717 
1718 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1719 				const char *name)
1720 {
1721 	struct vhost_scsi_nexus *tv_nexus;
1722 
1723 	mutex_lock(&tpg->tv_tpg_mutex);
1724 	if (tpg->tpg_nexus) {
1725 		mutex_unlock(&tpg->tv_tpg_mutex);
1726 		pr_debug("tpg->tpg_nexus already exists\n");
1727 		return -EEXIST;
1728 	}
1729 
1730 	tv_nexus = kzalloc(sizeof(*tv_nexus), GFP_KERNEL);
1731 	if (!tv_nexus) {
1732 		mutex_unlock(&tpg->tv_tpg_mutex);
1733 		pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1734 		return -ENOMEM;
1735 	}
1736 	/*
1737 	 * Since we are running in 'demo mode' this call with generate a
1738 	 * struct se_node_acl for the vhost_scsi struct se_portal_group with
1739 	 * the SCSI Initiator port name of the passed configfs group 'name'.
1740 	 */
1741 	tv_nexus->tvn_se_sess = target_setup_session(&tpg->se_tpg,
1742 					VHOST_SCSI_DEFAULT_TAGS,
1743 					sizeof(struct vhost_scsi_cmd),
1744 					TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1745 					(unsigned char *)name, tv_nexus,
1746 					vhost_scsi_nexus_cb);
1747 	if (IS_ERR(tv_nexus->tvn_se_sess)) {
1748 		mutex_unlock(&tpg->tv_tpg_mutex);
1749 		kfree(tv_nexus);
1750 		return -ENOMEM;
1751 	}
1752 	tpg->tpg_nexus = tv_nexus;
1753 
1754 	mutex_unlock(&tpg->tv_tpg_mutex);
1755 	return 0;
1756 }
1757 
1758 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1759 {
1760 	struct se_session *se_sess;
1761 	struct vhost_scsi_nexus *tv_nexus;
1762 
1763 	mutex_lock(&tpg->tv_tpg_mutex);
1764 	tv_nexus = tpg->tpg_nexus;
1765 	if (!tv_nexus) {
1766 		mutex_unlock(&tpg->tv_tpg_mutex);
1767 		return -ENODEV;
1768 	}
1769 
1770 	se_sess = tv_nexus->tvn_se_sess;
1771 	if (!se_sess) {
1772 		mutex_unlock(&tpg->tv_tpg_mutex);
1773 		return -ENODEV;
1774 	}
1775 
1776 	if (tpg->tv_tpg_port_count != 0) {
1777 		mutex_unlock(&tpg->tv_tpg_mutex);
1778 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1779 			" active TPG port count: %d\n",
1780 			tpg->tv_tpg_port_count);
1781 		return -EBUSY;
1782 	}
1783 
1784 	if (tpg->tv_tpg_vhost_count != 0) {
1785 		mutex_unlock(&tpg->tv_tpg_mutex);
1786 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1787 			" active TPG vhost count: %d\n",
1788 			tpg->tv_tpg_vhost_count);
1789 		return -EBUSY;
1790 	}
1791 
1792 	pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1793 		" %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1794 		tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1795 
1796 	vhost_scsi_free_cmd_map_res(se_sess);
1797 	/*
1798 	 * Release the SCSI I_T Nexus to the emulated vhost Target Port
1799 	 */
1800 	target_remove_session(se_sess);
1801 	tpg->tpg_nexus = NULL;
1802 	mutex_unlock(&tpg->tv_tpg_mutex);
1803 
1804 	kfree(tv_nexus);
1805 	return 0;
1806 }
1807 
1808 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
1809 {
1810 	struct se_portal_group *se_tpg = to_tpg(item);
1811 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1812 				struct vhost_scsi_tpg, se_tpg);
1813 	struct vhost_scsi_nexus *tv_nexus;
1814 	ssize_t ret;
1815 
1816 	mutex_lock(&tpg->tv_tpg_mutex);
1817 	tv_nexus = tpg->tpg_nexus;
1818 	if (!tv_nexus) {
1819 		mutex_unlock(&tpg->tv_tpg_mutex);
1820 		return -ENODEV;
1821 	}
1822 	ret = snprintf(page, PAGE_SIZE, "%s\n",
1823 			tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1824 	mutex_unlock(&tpg->tv_tpg_mutex);
1825 
1826 	return ret;
1827 }
1828 
1829 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
1830 		const char *page, size_t count)
1831 {
1832 	struct se_portal_group *se_tpg = to_tpg(item);
1833 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1834 				struct vhost_scsi_tpg, se_tpg);
1835 	struct vhost_scsi_tport *tport_wwn = tpg->tport;
1836 	unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1837 	int ret;
1838 	/*
1839 	 * Shutdown the active I_T nexus if 'NULL' is passed..
1840 	 */
1841 	if (!strncmp(page, "NULL", 4)) {
1842 		ret = vhost_scsi_drop_nexus(tpg);
1843 		return (!ret) ? count : ret;
1844 	}
1845 	/*
1846 	 * Otherwise make sure the passed virtual Initiator port WWN matches
1847 	 * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1848 	 * vhost_scsi_make_nexus().
1849 	 */
1850 	if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1851 		pr_err("Emulated NAA Sas Address: %s, exceeds"
1852 				" max: %d\n", page, VHOST_SCSI_NAMELEN);
1853 		return -EINVAL;
1854 	}
1855 	snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1856 
1857 	ptr = strstr(i_port, "naa.");
1858 	if (ptr) {
1859 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1860 			pr_err("Passed SAS Initiator Port %s does not"
1861 				" match target port protoid: %s\n", i_port,
1862 				vhost_scsi_dump_proto_id(tport_wwn));
1863 			return -EINVAL;
1864 		}
1865 		port_ptr = &i_port[0];
1866 		goto check_newline;
1867 	}
1868 	ptr = strstr(i_port, "fc.");
1869 	if (ptr) {
1870 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1871 			pr_err("Passed FCP Initiator Port %s does not"
1872 				" match target port protoid: %s\n", i_port,
1873 				vhost_scsi_dump_proto_id(tport_wwn));
1874 			return -EINVAL;
1875 		}
1876 		port_ptr = &i_port[3]; /* Skip over "fc." */
1877 		goto check_newline;
1878 	}
1879 	ptr = strstr(i_port, "iqn.");
1880 	if (ptr) {
1881 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1882 			pr_err("Passed iSCSI Initiator Port %s does not"
1883 				" match target port protoid: %s\n", i_port,
1884 				vhost_scsi_dump_proto_id(tport_wwn));
1885 			return -EINVAL;
1886 		}
1887 		port_ptr = &i_port[0];
1888 		goto check_newline;
1889 	}
1890 	pr_err("Unable to locate prefix for emulated Initiator Port:"
1891 			" %s\n", i_port);
1892 	return -EINVAL;
1893 	/*
1894 	 * Clear any trailing newline for the NAA WWN
1895 	 */
1896 check_newline:
1897 	if (i_port[strlen(i_port)-1] == '\n')
1898 		i_port[strlen(i_port)-1] = '\0';
1899 
1900 	ret = vhost_scsi_make_nexus(tpg, port_ptr);
1901 	if (ret < 0)
1902 		return ret;
1903 
1904 	return count;
1905 }
1906 
1907 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
1908 
1909 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1910 	&vhost_scsi_tpg_attr_nexus,
1911 	NULL,
1912 };
1913 
1914 static struct se_portal_group *
1915 vhost_scsi_make_tpg(struct se_wwn *wwn, const char *name)
1916 {
1917 	struct vhost_scsi_tport *tport = container_of(wwn,
1918 			struct vhost_scsi_tport, tport_wwn);
1919 
1920 	struct vhost_scsi_tpg *tpg;
1921 	u16 tpgt;
1922 	int ret;
1923 
1924 	if (strstr(name, "tpgt_") != name)
1925 		return ERR_PTR(-EINVAL);
1926 	if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1927 		return ERR_PTR(-EINVAL);
1928 
1929 	tpg = kzalloc(sizeof(*tpg), GFP_KERNEL);
1930 	if (!tpg) {
1931 		pr_err("Unable to allocate struct vhost_scsi_tpg");
1932 		return ERR_PTR(-ENOMEM);
1933 	}
1934 	mutex_init(&tpg->tv_tpg_mutex);
1935 	INIT_LIST_HEAD(&tpg->tv_tpg_list);
1936 	tpg->tport = tport;
1937 	tpg->tport_tpgt = tpgt;
1938 
1939 	ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
1940 	if (ret < 0) {
1941 		kfree(tpg);
1942 		return NULL;
1943 	}
1944 	mutex_lock(&vhost_scsi_mutex);
1945 	list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
1946 	mutex_unlock(&vhost_scsi_mutex);
1947 
1948 	return &tpg->se_tpg;
1949 }
1950 
1951 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
1952 {
1953 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1954 				struct vhost_scsi_tpg, se_tpg);
1955 
1956 	mutex_lock(&vhost_scsi_mutex);
1957 	list_del(&tpg->tv_tpg_list);
1958 	mutex_unlock(&vhost_scsi_mutex);
1959 	/*
1960 	 * Release the virtual I_T Nexus for this vhost TPG
1961 	 */
1962 	vhost_scsi_drop_nexus(tpg);
1963 	/*
1964 	 * Deregister the se_tpg from TCM..
1965 	 */
1966 	core_tpg_deregister(se_tpg);
1967 	kfree(tpg);
1968 }
1969 
1970 static struct se_wwn *
1971 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
1972 		     struct config_group *group,
1973 		     const char *name)
1974 {
1975 	struct vhost_scsi_tport *tport;
1976 	char *ptr;
1977 	u64 wwpn = 0;
1978 	int off = 0;
1979 
1980 	/* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
1981 		return ERR_PTR(-EINVAL); */
1982 
1983 	tport = kzalloc(sizeof(*tport), GFP_KERNEL);
1984 	if (!tport) {
1985 		pr_err("Unable to allocate struct vhost_scsi_tport");
1986 		return ERR_PTR(-ENOMEM);
1987 	}
1988 	tport->tport_wwpn = wwpn;
1989 	/*
1990 	 * Determine the emulated Protocol Identifier and Target Port Name
1991 	 * based on the incoming configfs directory name.
1992 	 */
1993 	ptr = strstr(name, "naa.");
1994 	if (ptr) {
1995 		tport->tport_proto_id = SCSI_PROTOCOL_SAS;
1996 		goto check_len;
1997 	}
1998 	ptr = strstr(name, "fc.");
1999 	if (ptr) {
2000 		tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2001 		off = 3; /* Skip over "fc." */
2002 		goto check_len;
2003 	}
2004 	ptr = strstr(name, "iqn.");
2005 	if (ptr) {
2006 		tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2007 		goto check_len;
2008 	}
2009 
2010 	pr_err("Unable to locate prefix for emulated Target Port:"
2011 			" %s\n", name);
2012 	kfree(tport);
2013 	return ERR_PTR(-EINVAL);
2014 
2015 check_len:
2016 	if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2017 		pr_err("Emulated %s Address: %s, exceeds"
2018 			" max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2019 			VHOST_SCSI_NAMELEN);
2020 		kfree(tport);
2021 		return ERR_PTR(-EINVAL);
2022 	}
2023 	snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2024 
2025 	pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2026 		" %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2027 
2028 	return &tport->tport_wwn;
2029 }
2030 
2031 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2032 {
2033 	struct vhost_scsi_tport *tport = container_of(wwn,
2034 				struct vhost_scsi_tport, tport_wwn);
2035 
2036 	pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2037 		" %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2038 		tport->tport_name);
2039 
2040 	kfree(tport);
2041 }
2042 
2043 static ssize_t
2044 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2045 {
2046 	return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2047 		"on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2048 		utsname()->machine);
2049 }
2050 
2051 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2052 
2053 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2054 	&vhost_scsi_wwn_attr_version,
2055 	NULL,
2056 };
2057 
2058 static const struct target_core_fabric_ops vhost_scsi_ops = {
2059 	.module				= THIS_MODULE,
2060 	.name				= "vhost",
2061 	.get_fabric_name		= vhost_scsi_get_fabric_name,
2062 	.tpg_get_wwn			= vhost_scsi_get_fabric_wwn,
2063 	.tpg_get_tag			= vhost_scsi_get_tpgt,
2064 	.tpg_check_demo_mode		= vhost_scsi_check_true,
2065 	.tpg_check_demo_mode_cache	= vhost_scsi_check_true,
2066 	.tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2067 	.tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2068 	.tpg_check_prot_fabric_only	= vhost_scsi_check_prot_fabric_only,
2069 	.tpg_get_inst_index		= vhost_scsi_tpg_get_inst_index,
2070 	.release_cmd			= vhost_scsi_release_cmd,
2071 	.check_stop_free		= vhost_scsi_check_stop_free,
2072 	.sess_get_index			= vhost_scsi_sess_get_index,
2073 	.sess_get_initiator_sid		= NULL,
2074 	.write_pending			= vhost_scsi_write_pending,
2075 	.write_pending_status		= vhost_scsi_write_pending_status,
2076 	.set_default_node_attributes	= vhost_scsi_set_default_node_attrs,
2077 	.get_cmd_state			= vhost_scsi_get_cmd_state,
2078 	.queue_data_in			= vhost_scsi_queue_data_in,
2079 	.queue_status			= vhost_scsi_queue_status,
2080 	.queue_tm_rsp			= vhost_scsi_queue_tm_rsp,
2081 	.aborted_task			= vhost_scsi_aborted_task,
2082 	/*
2083 	 * Setup callers for generic logic in target_core_fabric_configfs.c
2084 	 */
2085 	.fabric_make_wwn		= vhost_scsi_make_tport,
2086 	.fabric_drop_wwn		= vhost_scsi_drop_tport,
2087 	.fabric_make_tpg		= vhost_scsi_make_tpg,
2088 	.fabric_drop_tpg		= vhost_scsi_drop_tpg,
2089 	.fabric_post_link		= vhost_scsi_port_link,
2090 	.fabric_pre_unlink		= vhost_scsi_port_unlink,
2091 
2092 	.tfc_wwn_attrs			= vhost_scsi_wwn_attrs,
2093 	.tfc_tpg_base_attrs		= vhost_scsi_tpg_attrs,
2094 	.tfc_tpg_attrib_attrs		= vhost_scsi_tpg_attrib_attrs,
2095 };
2096 
2097 static int __init vhost_scsi_init(void)
2098 {
2099 	int ret = -ENOMEM;
2100 
2101 	pr_debug("TCM_VHOST fabric module %s on %s/%s"
2102 		" on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2103 		utsname()->machine);
2104 
2105 	/*
2106 	 * Use our own dedicated workqueue for submitting I/O into
2107 	 * target core to avoid contention within system_wq.
2108 	 */
2109 	vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2110 	if (!vhost_scsi_workqueue)
2111 		goto out;
2112 
2113 	ret = vhost_scsi_register();
2114 	if (ret < 0)
2115 		goto out_destroy_workqueue;
2116 
2117 	ret = target_register_template(&vhost_scsi_ops);
2118 	if (ret < 0)
2119 		goto out_vhost_scsi_deregister;
2120 
2121 	return 0;
2122 
2123 out_vhost_scsi_deregister:
2124 	vhost_scsi_deregister();
2125 out_destroy_workqueue:
2126 	destroy_workqueue(vhost_scsi_workqueue);
2127 out:
2128 	return ret;
2129 };
2130 
2131 static void vhost_scsi_exit(void)
2132 {
2133 	target_unregister_template(&vhost_scsi_ops);
2134 	vhost_scsi_deregister();
2135 	destroy_workqueue(vhost_scsi_workqueue);
2136 };
2137 
2138 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2139 MODULE_ALIAS("tcm_vhost");
2140 MODULE_LICENSE("GPL");
2141 module_init(vhost_scsi_init);
2142 module_exit(vhost_scsi_exit);
2143