xref: /openbmc/linux/drivers/vhost/scsi.c (revision 9b9c2cd4)
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 int vhost_scsi_shutdown_session(struct se_session *se_sess)
337 {
338 	return 0;
339 }
340 
341 static void vhost_scsi_close_session(struct se_session *se_sess)
342 {
343 	return;
344 }
345 
346 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
347 {
348 	return 0;
349 }
350 
351 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
352 {
353 	/* Go ahead and process the write immediately */
354 	target_execute_cmd(se_cmd);
355 	return 0;
356 }
357 
358 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
359 {
360 	return 0;
361 }
362 
363 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
364 {
365 	return;
366 }
367 
368 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
369 {
370 	return 0;
371 }
372 
373 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
374 {
375 	struct vhost_scsi *vs = cmd->tvc_vhost;
376 
377 	llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
378 
379 	vhost_work_queue(&vs->dev, &vs->vs_completion_work);
380 }
381 
382 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
383 {
384 	struct vhost_scsi_cmd *cmd = container_of(se_cmd,
385 				struct vhost_scsi_cmd, tvc_se_cmd);
386 	vhost_scsi_complete_cmd(cmd);
387 	return 0;
388 }
389 
390 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
391 {
392 	struct vhost_scsi_cmd *cmd = container_of(se_cmd,
393 				struct vhost_scsi_cmd, tvc_se_cmd);
394 	vhost_scsi_complete_cmd(cmd);
395 	return 0;
396 }
397 
398 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
399 {
400 	return;
401 }
402 
403 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
404 {
405 	return;
406 }
407 
408 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
409 {
410 	vs->vs_events_nr--;
411 	kfree(evt);
412 }
413 
414 static struct vhost_scsi_evt *
415 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
416 		       u32 event, u32 reason)
417 {
418 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
419 	struct vhost_scsi_evt *evt;
420 
421 	if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
422 		vs->vs_events_missed = true;
423 		return NULL;
424 	}
425 
426 	evt = kzalloc(sizeof(*evt), GFP_KERNEL);
427 	if (!evt) {
428 		vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
429 		vs->vs_events_missed = true;
430 		return NULL;
431 	}
432 
433 	evt->event.event = cpu_to_vhost32(vq, event);
434 	evt->event.reason = cpu_to_vhost32(vq, reason);
435 	vs->vs_events_nr++;
436 
437 	return evt;
438 }
439 
440 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
441 {
442 	struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
443 
444 	/* TODO locking against target/backend threads? */
445 	transport_generic_free_cmd(se_cmd, 0);
446 
447 }
448 
449 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
450 {
451 	return target_put_sess_cmd(se_cmd);
452 }
453 
454 static void
455 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
456 {
457 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
458 	struct virtio_scsi_event *event = &evt->event;
459 	struct virtio_scsi_event __user *eventp;
460 	unsigned out, in;
461 	int head, ret;
462 
463 	if (!vq->private_data) {
464 		vs->vs_events_missed = true;
465 		return;
466 	}
467 
468 again:
469 	vhost_disable_notify(&vs->dev, vq);
470 	head = vhost_get_vq_desc(vq, vq->iov,
471 			ARRAY_SIZE(vq->iov), &out, &in,
472 			NULL, NULL);
473 	if (head < 0) {
474 		vs->vs_events_missed = true;
475 		return;
476 	}
477 	if (head == vq->num) {
478 		if (vhost_enable_notify(&vs->dev, vq))
479 			goto again;
480 		vs->vs_events_missed = true;
481 		return;
482 	}
483 
484 	if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
485 		vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
486 				vq->iov[out].iov_len);
487 		vs->vs_events_missed = true;
488 		return;
489 	}
490 
491 	if (vs->vs_events_missed) {
492 		event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
493 		vs->vs_events_missed = false;
494 	}
495 
496 	eventp = vq->iov[out].iov_base;
497 	ret = __copy_to_user(eventp, event, sizeof(*event));
498 	if (!ret)
499 		vhost_add_used_and_signal(&vs->dev, vq, head, 0);
500 	else
501 		vq_err(vq, "Faulted on vhost_scsi_send_event\n");
502 }
503 
504 static void vhost_scsi_evt_work(struct vhost_work *work)
505 {
506 	struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
507 					vs_event_work);
508 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
509 	struct vhost_scsi_evt *evt;
510 	struct llist_node *llnode;
511 
512 	mutex_lock(&vq->mutex);
513 	llnode = llist_del_all(&vs->vs_event_list);
514 	while (llnode) {
515 		evt = llist_entry(llnode, struct vhost_scsi_evt, list);
516 		llnode = llist_next(llnode);
517 		vhost_scsi_do_evt_work(vs, evt);
518 		vhost_scsi_free_evt(vs, evt);
519 	}
520 	mutex_unlock(&vq->mutex);
521 }
522 
523 /* Fill in status and signal that we are done processing this command
524  *
525  * This is scheduled in the vhost work queue so we are called with the owner
526  * process mm and can access the vring.
527  */
528 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
529 {
530 	struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
531 					vs_completion_work);
532 	DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
533 	struct virtio_scsi_cmd_resp v_rsp;
534 	struct vhost_scsi_cmd *cmd;
535 	struct llist_node *llnode;
536 	struct se_cmd *se_cmd;
537 	struct iov_iter iov_iter;
538 	int ret, vq;
539 
540 	bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
541 	llnode = llist_del_all(&vs->vs_completion_list);
542 	while (llnode) {
543 		cmd = llist_entry(llnode, struct vhost_scsi_cmd,
544 				     tvc_completion_list);
545 		llnode = llist_next(llnode);
546 		se_cmd = &cmd->tvc_se_cmd;
547 
548 		pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
549 			cmd, se_cmd->residual_count, se_cmd->scsi_status);
550 
551 		memset(&v_rsp, 0, sizeof(v_rsp));
552 		v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
553 		/* TODO is status_qualifier field needed? */
554 		v_rsp.status = se_cmd->scsi_status;
555 		v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
556 						 se_cmd->scsi_sense_length);
557 		memcpy(v_rsp.sense, cmd->tvc_sense_buf,
558 		       se_cmd->scsi_sense_length);
559 
560 		iov_iter_init(&iov_iter, READ, cmd->tvc_resp_iov,
561 			      cmd->tvc_in_iovs, sizeof(v_rsp));
562 		ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
563 		if (likely(ret == sizeof(v_rsp))) {
564 			struct vhost_scsi_virtqueue *q;
565 			vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
566 			q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
567 			vq = q - vs->vqs;
568 			__set_bit(vq, signal);
569 		} else
570 			pr_err("Faulted on virtio_scsi_cmd_resp\n");
571 
572 		vhost_scsi_free_cmd(cmd);
573 	}
574 
575 	vq = -1;
576 	while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
577 		< VHOST_SCSI_MAX_VQ)
578 		vhost_signal(&vs->dev, &vs->vqs[vq].vq);
579 }
580 
581 static struct vhost_scsi_cmd *
582 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
583 		   unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
584 		   u32 exp_data_len, int data_direction)
585 {
586 	struct vhost_scsi_cmd *cmd;
587 	struct vhost_scsi_nexus *tv_nexus;
588 	struct se_session *se_sess;
589 	struct scatterlist *sg, *prot_sg;
590 	struct page **pages;
591 	int tag;
592 
593 	tv_nexus = tpg->tpg_nexus;
594 	if (!tv_nexus) {
595 		pr_err("Unable to locate active struct vhost_scsi_nexus\n");
596 		return ERR_PTR(-EIO);
597 	}
598 	se_sess = tv_nexus->tvn_se_sess;
599 
600 	tag = percpu_ida_alloc(&se_sess->sess_tag_pool, TASK_RUNNING);
601 	if (tag < 0) {
602 		pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
603 		return ERR_PTR(-ENOMEM);
604 	}
605 
606 	cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
607 	sg = cmd->tvc_sgl;
608 	prot_sg = cmd->tvc_prot_sgl;
609 	pages = cmd->tvc_upages;
610 	memset(cmd, 0, sizeof(struct vhost_scsi_cmd));
611 
612 	cmd->tvc_sgl = sg;
613 	cmd->tvc_prot_sgl = prot_sg;
614 	cmd->tvc_upages = pages;
615 	cmd->tvc_se_cmd.map_tag = tag;
616 	cmd->tvc_tag = scsi_tag;
617 	cmd->tvc_lun = lun;
618 	cmd->tvc_task_attr = task_attr;
619 	cmd->tvc_exp_data_len = exp_data_len;
620 	cmd->tvc_data_direction = data_direction;
621 	cmd->tvc_nexus = tv_nexus;
622 	cmd->inflight = vhost_scsi_get_inflight(vq);
623 
624 	memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
625 
626 	return cmd;
627 }
628 
629 /*
630  * Map a user memory range into a scatterlist
631  *
632  * Returns the number of scatterlist entries used or -errno on error.
633  */
634 static int
635 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
636 		      void __user *ptr,
637 		      size_t len,
638 		      struct scatterlist *sgl,
639 		      bool write)
640 {
641 	unsigned int npages = 0, offset, nbytes;
642 	unsigned int pages_nr = iov_num_pages(ptr, len);
643 	struct scatterlist *sg = sgl;
644 	struct page **pages = cmd->tvc_upages;
645 	int ret, i;
646 
647 	if (pages_nr > VHOST_SCSI_PREALLOC_UPAGES) {
648 		pr_err("vhost_scsi_map_to_sgl() pages_nr: %u greater than"
649 		       " preallocated VHOST_SCSI_PREALLOC_UPAGES: %u\n",
650 			pages_nr, VHOST_SCSI_PREALLOC_UPAGES);
651 		return -ENOBUFS;
652 	}
653 
654 	ret = get_user_pages_fast((unsigned long)ptr, pages_nr, write, pages);
655 	/* No pages were pinned */
656 	if (ret < 0)
657 		goto out;
658 	/* Less pages pinned than wanted */
659 	if (ret != pages_nr) {
660 		for (i = 0; i < ret; i++)
661 			put_page(pages[i]);
662 		ret = -EFAULT;
663 		goto out;
664 	}
665 
666 	while (len > 0) {
667 		offset = (uintptr_t)ptr & ~PAGE_MASK;
668 		nbytes = min_t(unsigned int, PAGE_SIZE - offset, len);
669 		sg_set_page(sg, pages[npages], nbytes, offset);
670 		ptr += nbytes;
671 		len -= nbytes;
672 		sg++;
673 		npages++;
674 	}
675 
676 out:
677 	return ret;
678 }
679 
680 static int
681 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
682 {
683 	int sgl_count = 0;
684 
685 	if (!iter || !iter->iov) {
686 		pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
687 		       " present\n", __func__, bytes);
688 		return -EINVAL;
689 	}
690 
691 	sgl_count = iov_iter_npages(iter, 0xffff);
692 	if (sgl_count > max_sgls) {
693 		pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
694 		       " max_sgls: %d\n", __func__, sgl_count, max_sgls);
695 		return -EINVAL;
696 	}
697 	return sgl_count;
698 }
699 
700 static int
701 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
702 		      struct iov_iter *iter,
703 		      struct scatterlist *sg, int sg_count)
704 {
705 	size_t off = iter->iov_offset;
706 	int i, ret;
707 
708 	for (i = 0; i < iter->nr_segs; i++) {
709 		void __user *base = iter->iov[i].iov_base + off;
710 		size_t len = iter->iov[i].iov_len - off;
711 
712 		ret = vhost_scsi_map_to_sgl(cmd, base, len, sg, write);
713 		if (ret < 0) {
714 			for (i = 0; i < sg_count; i++) {
715 				struct page *page = sg_page(&sg[i]);
716 				if (page)
717 					put_page(page);
718 			}
719 			return ret;
720 		}
721 		sg += ret;
722 		off = 0;
723 	}
724 	return 0;
725 }
726 
727 static int
728 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
729 		 size_t prot_bytes, struct iov_iter *prot_iter,
730 		 size_t data_bytes, struct iov_iter *data_iter)
731 {
732 	int sgl_count, ret;
733 	bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
734 
735 	if (prot_bytes) {
736 		sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
737 						 VHOST_SCSI_PREALLOC_PROT_SGLS);
738 		if (sgl_count < 0)
739 			return sgl_count;
740 
741 		sg_init_table(cmd->tvc_prot_sgl, sgl_count);
742 		cmd->tvc_prot_sgl_count = sgl_count;
743 		pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
744 			 cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
745 
746 		ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
747 					    cmd->tvc_prot_sgl,
748 					    cmd->tvc_prot_sgl_count);
749 		if (ret < 0) {
750 			cmd->tvc_prot_sgl_count = 0;
751 			return ret;
752 		}
753 	}
754 	sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
755 					 VHOST_SCSI_PREALLOC_SGLS);
756 	if (sgl_count < 0)
757 		return sgl_count;
758 
759 	sg_init_table(cmd->tvc_sgl, sgl_count);
760 	cmd->tvc_sgl_count = sgl_count;
761 	pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
762 		  cmd->tvc_sgl, cmd->tvc_sgl_count);
763 
764 	ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
765 				    cmd->tvc_sgl, cmd->tvc_sgl_count);
766 	if (ret < 0) {
767 		cmd->tvc_sgl_count = 0;
768 		return ret;
769 	}
770 	return 0;
771 }
772 
773 static int vhost_scsi_to_tcm_attr(int attr)
774 {
775 	switch (attr) {
776 	case VIRTIO_SCSI_S_SIMPLE:
777 		return TCM_SIMPLE_TAG;
778 	case VIRTIO_SCSI_S_ORDERED:
779 		return TCM_ORDERED_TAG;
780 	case VIRTIO_SCSI_S_HEAD:
781 		return TCM_HEAD_TAG;
782 	case VIRTIO_SCSI_S_ACA:
783 		return TCM_ACA_TAG;
784 	default:
785 		break;
786 	}
787 	return TCM_SIMPLE_TAG;
788 }
789 
790 static void vhost_scsi_submission_work(struct work_struct *work)
791 {
792 	struct vhost_scsi_cmd *cmd =
793 		container_of(work, struct vhost_scsi_cmd, work);
794 	struct vhost_scsi_nexus *tv_nexus;
795 	struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
796 	struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
797 	int rc;
798 
799 	/* FIXME: BIDI operation */
800 	if (cmd->tvc_sgl_count) {
801 		sg_ptr = cmd->tvc_sgl;
802 
803 		if (cmd->tvc_prot_sgl_count)
804 			sg_prot_ptr = cmd->tvc_prot_sgl;
805 		else
806 			se_cmd->prot_pto = true;
807 	} else {
808 		sg_ptr = NULL;
809 	}
810 	tv_nexus = cmd->tvc_nexus;
811 
812 	se_cmd->tag = 0;
813 	rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
814 			cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
815 			cmd->tvc_lun, cmd->tvc_exp_data_len,
816 			vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
817 			cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
818 			sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
819 			cmd->tvc_prot_sgl_count);
820 	if (rc < 0) {
821 		transport_send_check_condition_and_sense(se_cmd,
822 				TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
823 		transport_generic_free_cmd(se_cmd, 0);
824 	}
825 }
826 
827 static void
828 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
829 			   struct vhost_virtqueue *vq,
830 			   int head, unsigned out)
831 {
832 	struct virtio_scsi_cmd_resp __user *resp;
833 	struct virtio_scsi_cmd_resp rsp;
834 	int ret;
835 
836 	memset(&rsp, 0, sizeof(rsp));
837 	rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
838 	resp = vq->iov[out].iov_base;
839 	ret = __copy_to_user(resp, &rsp, sizeof(rsp));
840 	if (!ret)
841 		vhost_add_used_and_signal(&vs->dev, vq, head, 0);
842 	else
843 		pr_err("Faulted on virtio_scsi_cmd_resp\n");
844 }
845 
846 static void
847 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
848 {
849 	struct vhost_scsi_tpg **vs_tpg, *tpg;
850 	struct virtio_scsi_cmd_req v_req;
851 	struct virtio_scsi_cmd_req_pi v_req_pi;
852 	struct vhost_scsi_cmd *cmd;
853 	struct iov_iter out_iter, in_iter, prot_iter, data_iter;
854 	u64 tag;
855 	u32 exp_data_len, data_direction;
856 	unsigned out, in;
857 	int head, ret, prot_bytes;
858 	size_t req_size, rsp_size = sizeof(struct virtio_scsi_cmd_resp);
859 	size_t out_size, in_size;
860 	u16 lun;
861 	u8 *target, *lunp, task_attr;
862 	bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
863 	void *req, *cdb;
864 
865 	mutex_lock(&vq->mutex);
866 	/*
867 	 * We can handle the vq only after the endpoint is setup by calling the
868 	 * VHOST_SCSI_SET_ENDPOINT ioctl.
869 	 */
870 	vs_tpg = vq->private_data;
871 	if (!vs_tpg)
872 		goto out;
873 
874 	vhost_disable_notify(&vs->dev, vq);
875 
876 	for (;;) {
877 		head = vhost_get_vq_desc(vq, vq->iov,
878 					 ARRAY_SIZE(vq->iov), &out, &in,
879 					 NULL, NULL);
880 		pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
881 			 head, out, in);
882 		/* On error, stop handling until the next kick. */
883 		if (unlikely(head < 0))
884 			break;
885 		/* Nothing new?  Wait for eventfd to tell us they refilled. */
886 		if (head == vq->num) {
887 			if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
888 				vhost_disable_notify(&vs->dev, vq);
889 				continue;
890 			}
891 			break;
892 		}
893 		/*
894 		 * Check for a sane response buffer so we can report early
895 		 * errors back to the guest.
896 		 */
897 		if (unlikely(vq->iov[out].iov_len < rsp_size)) {
898 			vq_err(vq, "Expecting at least virtio_scsi_cmd_resp"
899 				" size, got %zu bytes\n", vq->iov[out].iov_len);
900 			break;
901 		}
902 		/*
903 		 * Setup pointers and values based upon different virtio-scsi
904 		 * request header if T10_PI is enabled in KVM guest.
905 		 */
906 		if (t10_pi) {
907 			req = &v_req_pi;
908 			req_size = sizeof(v_req_pi);
909 			lunp = &v_req_pi.lun[0];
910 			target = &v_req_pi.lun[1];
911 		} else {
912 			req = &v_req;
913 			req_size = sizeof(v_req);
914 			lunp = &v_req.lun[0];
915 			target = &v_req.lun[1];
916 		}
917 		/*
918 		 * FIXME: Not correct for BIDI operation
919 		 */
920 		out_size = iov_length(vq->iov, out);
921 		in_size = iov_length(&vq->iov[out], in);
922 
923 		/*
924 		 * Copy over the virtio-scsi request header, which for a
925 		 * ANY_LAYOUT enabled guest may span multiple iovecs, or a
926 		 * single iovec may contain both the header + outgoing
927 		 * WRITE payloads.
928 		 *
929 		 * copy_from_iter() will advance out_iter, so that it will
930 		 * point at the start of the outgoing WRITE payload, if
931 		 * DMA_TO_DEVICE is set.
932 		 */
933 		iov_iter_init(&out_iter, WRITE, vq->iov, out, out_size);
934 
935 		ret = copy_from_iter(req, req_size, &out_iter);
936 		if (unlikely(ret != req_size)) {
937 			vq_err(vq, "Faulted on copy_from_iter\n");
938 			vhost_scsi_send_bad_target(vs, vq, head, out);
939 			continue;
940 		}
941 		/* virtio-scsi spec requires byte 0 of the lun to be 1 */
942 		if (unlikely(*lunp != 1)) {
943 			vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp);
944 			vhost_scsi_send_bad_target(vs, vq, head, out);
945 			continue;
946 		}
947 
948 		tpg = ACCESS_ONCE(vs_tpg[*target]);
949 		if (unlikely(!tpg)) {
950 			/* Target does not exist, fail the request */
951 			vhost_scsi_send_bad_target(vs, vq, head, out);
952 			continue;
953 		}
954 		/*
955 		 * Determine data_direction by calculating the total outgoing
956 		 * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
957 		 * response headers respectively.
958 		 *
959 		 * For DMA_TO_DEVICE this is out_iter, which is already pointing
960 		 * to the right place.
961 		 *
962 		 * For DMA_FROM_DEVICE, the iovec will be just past the end
963 		 * of the virtio-scsi response header in either the same
964 		 * or immediately following iovec.
965 		 *
966 		 * Any associated T10_PI bytes for the outgoing / incoming
967 		 * payloads are included in calculation of exp_data_len here.
968 		 */
969 		prot_bytes = 0;
970 
971 		if (out_size > req_size) {
972 			data_direction = DMA_TO_DEVICE;
973 			exp_data_len = out_size - req_size;
974 			data_iter = out_iter;
975 		} else if (in_size > rsp_size) {
976 			data_direction = DMA_FROM_DEVICE;
977 			exp_data_len = in_size - rsp_size;
978 
979 			iov_iter_init(&in_iter, READ, &vq->iov[out], in,
980 				      rsp_size + exp_data_len);
981 			iov_iter_advance(&in_iter, rsp_size);
982 			data_iter = in_iter;
983 		} else {
984 			data_direction = DMA_NONE;
985 			exp_data_len = 0;
986 		}
987 		/*
988 		 * If T10_PI header + payload is present, setup prot_iter values
989 		 * and recalculate data_iter for vhost_scsi_mapal() mapping to
990 		 * host scatterlists via get_user_pages_fast().
991 		 */
992 		if (t10_pi) {
993 			if (v_req_pi.pi_bytesout) {
994 				if (data_direction != DMA_TO_DEVICE) {
995 					vq_err(vq, "Received non zero pi_bytesout,"
996 						" but wrong data_direction\n");
997 					vhost_scsi_send_bad_target(vs, vq, head, out);
998 					continue;
999 				}
1000 				prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
1001 			} else if (v_req_pi.pi_bytesin) {
1002 				if (data_direction != DMA_FROM_DEVICE) {
1003 					vq_err(vq, "Received non zero pi_bytesin,"
1004 						" but wrong data_direction\n");
1005 					vhost_scsi_send_bad_target(vs, vq, head, out);
1006 					continue;
1007 				}
1008 				prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
1009 			}
1010 			/*
1011 			 * Set prot_iter to data_iter, and advance past any
1012 			 * preceeding prot_bytes that may be present.
1013 			 *
1014 			 * Also fix up the exp_data_len to reflect only the
1015 			 * actual data payload length.
1016 			 */
1017 			if (prot_bytes) {
1018 				exp_data_len -= prot_bytes;
1019 				prot_iter = data_iter;
1020 				iov_iter_advance(&data_iter, prot_bytes);
1021 			}
1022 			tag = vhost64_to_cpu(vq, v_req_pi.tag);
1023 			task_attr = v_req_pi.task_attr;
1024 			cdb = &v_req_pi.cdb[0];
1025 			lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
1026 		} else {
1027 			tag = vhost64_to_cpu(vq, v_req.tag);
1028 			task_attr = v_req.task_attr;
1029 			cdb = &v_req.cdb[0];
1030 			lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1031 		}
1032 		/*
1033 		 * Check that the received CDB size does not exceeded our
1034 		 * hardcoded max for vhost-scsi, then get a pre-allocated
1035 		 * cmd descriptor for the new virtio-scsi tag.
1036 		 *
1037 		 * TODO what if cdb was too small for varlen cdb header?
1038 		 */
1039 		if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1040 			vq_err(vq, "Received SCSI CDB with command_size: %d that"
1041 				" exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1042 				scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1043 			vhost_scsi_send_bad_target(vs, vq, head, out);
1044 			continue;
1045 		}
1046 		cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1047 					 exp_data_len + prot_bytes,
1048 					 data_direction);
1049 		if (IS_ERR(cmd)) {
1050 			vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1051 			       PTR_ERR(cmd));
1052 			vhost_scsi_send_bad_target(vs, vq, head, out);
1053 			continue;
1054 		}
1055 		cmd->tvc_vhost = vs;
1056 		cmd->tvc_vq = vq;
1057 		cmd->tvc_resp_iov = &vq->iov[out];
1058 		cmd->tvc_in_iovs = in;
1059 
1060 		pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1061 			 cmd->tvc_cdb[0], cmd->tvc_lun);
1062 		pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1063 			 " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1064 
1065 		if (data_direction != DMA_NONE) {
1066 			ret = vhost_scsi_mapal(cmd,
1067 					       prot_bytes, &prot_iter,
1068 					       exp_data_len, &data_iter);
1069 			if (unlikely(ret)) {
1070 				vq_err(vq, "Failed to map iov to sgl\n");
1071 				vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1072 				vhost_scsi_send_bad_target(vs, vq, head, out);
1073 				continue;
1074 			}
1075 		}
1076 		/*
1077 		 * Save the descriptor from vhost_get_vq_desc() to be used to
1078 		 * complete the virtio-scsi request in TCM callback context via
1079 		 * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1080 		 */
1081 		cmd->tvc_vq_desc = head;
1082 		/*
1083 		 * Dispatch cmd descriptor for cmwq execution in process
1084 		 * context provided by vhost_scsi_workqueue.  This also ensures
1085 		 * cmd is executed on the same kworker CPU as this vhost
1086 		 * thread to gain positive L2 cache locality effects.
1087 		 */
1088 		INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1089 		queue_work(vhost_scsi_workqueue, &cmd->work);
1090 	}
1091 out:
1092 	mutex_unlock(&vq->mutex);
1093 }
1094 
1095 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1096 {
1097 	pr_debug("%s: The handling func for control queue.\n", __func__);
1098 }
1099 
1100 static void
1101 vhost_scsi_send_evt(struct vhost_scsi *vs,
1102 		   struct vhost_scsi_tpg *tpg,
1103 		   struct se_lun *lun,
1104 		   u32 event,
1105 		   u32 reason)
1106 {
1107 	struct vhost_scsi_evt *evt;
1108 
1109 	evt = vhost_scsi_allocate_evt(vs, event, reason);
1110 	if (!evt)
1111 		return;
1112 
1113 	if (tpg && lun) {
1114 		/* TODO: share lun setup code with virtio-scsi.ko */
1115 		/*
1116 		 * Note: evt->event is zeroed when we allocate it and
1117 		 * lun[4-7] need to be zero according to virtio-scsi spec.
1118 		 */
1119 		evt->event.lun[0] = 0x01;
1120 		evt->event.lun[1] = tpg->tport_tpgt;
1121 		if (lun->unpacked_lun >= 256)
1122 			evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1123 		evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1124 	}
1125 
1126 	llist_add(&evt->list, &vs->vs_event_list);
1127 	vhost_work_queue(&vs->dev, &vs->vs_event_work);
1128 }
1129 
1130 static void vhost_scsi_evt_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 	mutex_lock(&vq->mutex);
1137 	if (!vq->private_data)
1138 		goto out;
1139 
1140 	if (vs->vs_events_missed)
1141 		vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1142 out:
1143 	mutex_unlock(&vq->mutex);
1144 }
1145 
1146 static void vhost_scsi_handle_kick(struct vhost_work *work)
1147 {
1148 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1149 						poll.work);
1150 	struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1151 
1152 	vhost_scsi_handle_vq(vs, vq);
1153 }
1154 
1155 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1156 {
1157 	vhost_poll_flush(&vs->vqs[index].vq.poll);
1158 }
1159 
1160 /* Callers must hold dev mutex */
1161 static void vhost_scsi_flush(struct vhost_scsi *vs)
1162 {
1163 	struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1164 	int i;
1165 
1166 	/* Init new inflight and remember the old inflight */
1167 	vhost_scsi_init_inflight(vs, old_inflight);
1168 
1169 	/*
1170 	 * The inflight->kref was initialized to 1. We decrement it here to
1171 	 * indicate the start of the flush operation so that it will reach 0
1172 	 * when all the reqs are finished.
1173 	 */
1174 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1175 		kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1176 
1177 	/* Flush both the vhost poll and vhost work */
1178 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1179 		vhost_scsi_flush_vq(vs, i);
1180 	vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1181 	vhost_work_flush(&vs->dev, &vs->vs_event_work);
1182 
1183 	/* Wait for all reqs issued before the flush to be finished */
1184 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1185 		wait_for_completion(&old_inflight[i]->comp);
1186 }
1187 
1188 /*
1189  * Called from vhost_scsi_ioctl() context to walk the list of available
1190  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1191  *
1192  *  The lock nesting rule is:
1193  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1194  */
1195 static int
1196 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1197 			struct vhost_scsi_target *t)
1198 {
1199 	struct se_portal_group *se_tpg;
1200 	struct vhost_scsi_tport *tv_tport;
1201 	struct vhost_scsi_tpg *tpg;
1202 	struct vhost_scsi_tpg **vs_tpg;
1203 	struct vhost_virtqueue *vq;
1204 	int index, ret, i, len;
1205 	bool match = false;
1206 
1207 	mutex_lock(&vhost_scsi_mutex);
1208 	mutex_lock(&vs->dev.mutex);
1209 
1210 	/* Verify that ring has been setup correctly. */
1211 	for (index = 0; index < vs->dev.nvqs; ++index) {
1212 		/* Verify that ring has been setup correctly. */
1213 		if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1214 			ret = -EFAULT;
1215 			goto out;
1216 		}
1217 	}
1218 
1219 	len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1220 	vs_tpg = kzalloc(len, GFP_KERNEL);
1221 	if (!vs_tpg) {
1222 		ret = -ENOMEM;
1223 		goto out;
1224 	}
1225 	if (vs->vs_tpg)
1226 		memcpy(vs_tpg, vs->vs_tpg, len);
1227 
1228 	list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1229 		mutex_lock(&tpg->tv_tpg_mutex);
1230 		if (!tpg->tpg_nexus) {
1231 			mutex_unlock(&tpg->tv_tpg_mutex);
1232 			continue;
1233 		}
1234 		if (tpg->tv_tpg_vhost_count != 0) {
1235 			mutex_unlock(&tpg->tv_tpg_mutex);
1236 			continue;
1237 		}
1238 		tv_tport = tpg->tport;
1239 
1240 		if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1241 			if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1242 				kfree(vs_tpg);
1243 				mutex_unlock(&tpg->tv_tpg_mutex);
1244 				ret = -EEXIST;
1245 				goto out;
1246 			}
1247 			/*
1248 			 * In order to ensure individual vhost-scsi configfs
1249 			 * groups cannot be removed while in use by vhost ioctl,
1250 			 * go ahead and take an explicit se_tpg->tpg_group.cg_item
1251 			 * dependency now.
1252 			 */
1253 			se_tpg = &tpg->se_tpg;
1254 			ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1255 			if (ret) {
1256 				pr_warn("configfs_depend_item() failed: %d\n", ret);
1257 				kfree(vs_tpg);
1258 				mutex_unlock(&tpg->tv_tpg_mutex);
1259 				goto out;
1260 			}
1261 			tpg->tv_tpg_vhost_count++;
1262 			tpg->vhost_scsi = vs;
1263 			vs_tpg[tpg->tport_tpgt] = tpg;
1264 			smp_mb__after_atomic();
1265 			match = true;
1266 		}
1267 		mutex_unlock(&tpg->tv_tpg_mutex);
1268 	}
1269 
1270 	if (match) {
1271 		memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1272 		       sizeof(vs->vs_vhost_wwpn));
1273 		for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1274 			vq = &vs->vqs[i].vq;
1275 			mutex_lock(&vq->mutex);
1276 			vq->private_data = vs_tpg;
1277 			vhost_init_used(vq);
1278 			mutex_unlock(&vq->mutex);
1279 		}
1280 		ret = 0;
1281 	} else {
1282 		ret = -EEXIST;
1283 	}
1284 
1285 	/*
1286 	 * Act as synchronize_rcu to make sure access to
1287 	 * old vs->vs_tpg is finished.
1288 	 */
1289 	vhost_scsi_flush(vs);
1290 	kfree(vs->vs_tpg);
1291 	vs->vs_tpg = vs_tpg;
1292 
1293 out:
1294 	mutex_unlock(&vs->dev.mutex);
1295 	mutex_unlock(&vhost_scsi_mutex);
1296 	return ret;
1297 }
1298 
1299 static int
1300 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1301 			  struct vhost_scsi_target *t)
1302 {
1303 	struct se_portal_group *se_tpg;
1304 	struct vhost_scsi_tport *tv_tport;
1305 	struct vhost_scsi_tpg *tpg;
1306 	struct vhost_virtqueue *vq;
1307 	bool match = false;
1308 	int index, ret, i;
1309 	u8 target;
1310 
1311 	mutex_lock(&vhost_scsi_mutex);
1312 	mutex_lock(&vs->dev.mutex);
1313 	/* Verify that ring has been setup correctly. */
1314 	for (index = 0; index < vs->dev.nvqs; ++index) {
1315 		if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1316 			ret = -EFAULT;
1317 			goto err_dev;
1318 		}
1319 	}
1320 
1321 	if (!vs->vs_tpg) {
1322 		ret = 0;
1323 		goto err_dev;
1324 	}
1325 
1326 	for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1327 		target = i;
1328 		tpg = vs->vs_tpg[target];
1329 		if (!tpg)
1330 			continue;
1331 
1332 		mutex_lock(&tpg->tv_tpg_mutex);
1333 		tv_tport = tpg->tport;
1334 		if (!tv_tport) {
1335 			ret = -ENODEV;
1336 			goto err_tpg;
1337 		}
1338 
1339 		if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1340 			pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1341 				" does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1342 				tv_tport->tport_name, tpg->tport_tpgt,
1343 				t->vhost_wwpn, t->vhost_tpgt);
1344 			ret = -EINVAL;
1345 			goto err_tpg;
1346 		}
1347 		tpg->tv_tpg_vhost_count--;
1348 		tpg->vhost_scsi = NULL;
1349 		vs->vs_tpg[target] = NULL;
1350 		match = true;
1351 		mutex_unlock(&tpg->tv_tpg_mutex);
1352 		/*
1353 		 * Release se_tpg->tpg_group.cg_item configfs dependency now
1354 		 * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1355 		 */
1356 		se_tpg = &tpg->se_tpg;
1357 		target_undepend_item(&se_tpg->tpg_group.cg_item);
1358 	}
1359 	if (match) {
1360 		for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1361 			vq = &vs->vqs[i].vq;
1362 			mutex_lock(&vq->mutex);
1363 			vq->private_data = NULL;
1364 			mutex_unlock(&vq->mutex);
1365 		}
1366 	}
1367 	/*
1368 	 * Act as synchronize_rcu to make sure access to
1369 	 * old vs->vs_tpg is finished.
1370 	 */
1371 	vhost_scsi_flush(vs);
1372 	kfree(vs->vs_tpg);
1373 	vs->vs_tpg = NULL;
1374 	WARN_ON(vs->vs_events_nr);
1375 	mutex_unlock(&vs->dev.mutex);
1376 	mutex_unlock(&vhost_scsi_mutex);
1377 	return 0;
1378 
1379 err_tpg:
1380 	mutex_unlock(&tpg->tv_tpg_mutex);
1381 err_dev:
1382 	mutex_unlock(&vs->dev.mutex);
1383 	mutex_unlock(&vhost_scsi_mutex);
1384 	return ret;
1385 }
1386 
1387 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1388 {
1389 	struct vhost_virtqueue *vq;
1390 	int i;
1391 
1392 	if (features & ~VHOST_SCSI_FEATURES)
1393 		return -EOPNOTSUPP;
1394 
1395 	mutex_lock(&vs->dev.mutex);
1396 	if ((features & (1 << VHOST_F_LOG_ALL)) &&
1397 	    !vhost_log_access_ok(&vs->dev)) {
1398 		mutex_unlock(&vs->dev.mutex);
1399 		return -EFAULT;
1400 	}
1401 
1402 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1403 		vq = &vs->vqs[i].vq;
1404 		mutex_lock(&vq->mutex);
1405 		vq->acked_features = features;
1406 		mutex_unlock(&vq->mutex);
1407 	}
1408 	mutex_unlock(&vs->dev.mutex);
1409 	return 0;
1410 }
1411 
1412 static int vhost_scsi_open(struct inode *inode, struct file *f)
1413 {
1414 	struct vhost_scsi *vs;
1415 	struct vhost_virtqueue **vqs;
1416 	int r = -ENOMEM, i;
1417 
1418 	vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
1419 	if (!vs) {
1420 		vs = vzalloc(sizeof(*vs));
1421 		if (!vs)
1422 			goto err_vs;
1423 	}
1424 
1425 	vqs = kmalloc(VHOST_SCSI_MAX_VQ * sizeof(*vqs), GFP_KERNEL);
1426 	if (!vqs)
1427 		goto err_vqs;
1428 
1429 	vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1430 	vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1431 
1432 	vs->vs_events_nr = 0;
1433 	vs->vs_events_missed = false;
1434 
1435 	vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1436 	vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1437 	vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1438 	vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1439 	for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1440 		vqs[i] = &vs->vqs[i].vq;
1441 		vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1442 	}
1443 	vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ);
1444 
1445 	vhost_scsi_init_inflight(vs, NULL);
1446 
1447 	f->private_data = vs;
1448 	return 0;
1449 
1450 err_vqs:
1451 	kvfree(vs);
1452 err_vs:
1453 	return r;
1454 }
1455 
1456 static int vhost_scsi_release(struct inode *inode, struct file *f)
1457 {
1458 	struct vhost_scsi *vs = f->private_data;
1459 	struct vhost_scsi_target t;
1460 
1461 	mutex_lock(&vs->dev.mutex);
1462 	memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1463 	mutex_unlock(&vs->dev.mutex);
1464 	vhost_scsi_clear_endpoint(vs, &t);
1465 	vhost_dev_stop(&vs->dev);
1466 	vhost_dev_cleanup(&vs->dev, false);
1467 	/* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1468 	vhost_scsi_flush(vs);
1469 	kfree(vs->dev.vqs);
1470 	kvfree(vs);
1471 	return 0;
1472 }
1473 
1474 static long
1475 vhost_scsi_ioctl(struct file *f,
1476 		 unsigned int ioctl,
1477 		 unsigned long arg)
1478 {
1479 	struct vhost_scsi *vs = f->private_data;
1480 	struct vhost_scsi_target backend;
1481 	void __user *argp = (void __user *)arg;
1482 	u64 __user *featurep = argp;
1483 	u32 __user *eventsp = argp;
1484 	u32 events_missed;
1485 	u64 features;
1486 	int r, abi_version = VHOST_SCSI_ABI_VERSION;
1487 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1488 
1489 	switch (ioctl) {
1490 	case VHOST_SCSI_SET_ENDPOINT:
1491 		if (copy_from_user(&backend, argp, sizeof backend))
1492 			return -EFAULT;
1493 		if (backend.reserved != 0)
1494 			return -EOPNOTSUPP;
1495 
1496 		return vhost_scsi_set_endpoint(vs, &backend);
1497 	case VHOST_SCSI_CLEAR_ENDPOINT:
1498 		if (copy_from_user(&backend, argp, sizeof backend))
1499 			return -EFAULT;
1500 		if (backend.reserved != 0)
1501 			return -EOPNOTSUPP;
1502 
1503 		return vhost_scsi_clear_endpoint(vs, &backend);
1504 	case VHOST_SCSI_GET_ABI_VERSION:
1505 		if (copy_to_user(argp, &abi_version, sizeof abi_version))
1506 			return -EFAULT;
1507 		return 0;
1508 	case VHOST_SCSI_SET_EVENTS_MISSED:
1509 		if (get_user(events_missed, eventsp))
1510 			return -EFAULT;
1511 		mutex_lock(&vq->mutex);
1512 		vs->vs_events_missed = events_missed;
1513 		mutex_unlock(&vq->mutex);
1514 		return 0;
1515 	case VHOST_SCSI_GET_EVENTS_MISSED:
1516 		mutex_lock(&vq->mutex);
1517 		events_missed = vs->vs_events_missed;
1518 		mutex_unlock(&vq->mutex);
1519 		if (put_user(events_missed, eventsp))
1520 			return -EFAULT;
1521 		return 0;
1522 	case VHOST_GET_FEATURES:
1523 		features = VHOST_SCSI_FEATURES;
1524 		if (copy_to_user(featurep, &features, sizeof features))
1525 			return -EFAULT;
1526 		return 0;
1527 	case VHOST_SET_FEATURES:
1528 		if (copy_from_user(&features, featurep, sizeof features))
1529 			return -EFAULT;
1530 		return vhost_scsi_set_features(vs, features);
1531 	default:
1532 		mutex_lock(&vs->dev.mutex);
1533 		r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1534 		/* TODO: flush backend after dev ioctl. */
1535 		if (r == -ENOIOCTLCMD)
1536 			r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1537 		mutex_unlock(&vs->dev.mutex);
1538 		return r;
1539 	}
1540 }
1541 
1542 #ifdef CONFIG_COMPAT
1543 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1544 				unsigned long arg)
1545 {
1546 	return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1547 }
1548 #endif
1549 
1550 static const struct file_operations vhost_scsi_fops = {
1551 	.owner          = THIS_MODULE,
1552 	.release        = vhost_scsi_release,
1553 	.unlocked_ioctl = vhost_scsi_ioctl,
1554 #ifdef CONFIG_COMPAT
1555 	.compat_ioctl	= vhost_scsi_compat_ioctl,
1556 #endif
1557 	.open           = vhost_scsi_open,
1558 	.llseek		= noop_llseek,
1559 };
1560 
1561 static struct miscdevice vhost_scsi_misc = {
1562 	MISC_DYNAMIC_MINOR,
1563 	"vhost-scsi",
1564 	&vhost_scsi_fops,
1565 };
1566 
1567 static int __init vhost_scsi_register(void)
1568 {
1569 	return misc_register(&vhost_scsi_misc);
1570 }
1571 
1572 static void vhost_scsi_deregister(void)
1573 {
1574 	misc_deregister(&vhost_scsi_misc);
1575 }
1576 
1577 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1578 {
1579 	switch (tport->tport_proto_id) {
1580 	case SCSI_PROTOCOL_SAS:
1581 		return "SAS";
1582 	case SCSI_PROTOCOL_FCP:
1583 		return "FCP";
1584 	case SCSI_PROTOCOL_ISCSI:
1585 		return "iSCSI";
1586 	default:
1587 		break;
1588 	}
1589 
1590 	return "Unknown";
1591 }
1592 
1593 static void
1594 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1595 		  struct se_lun *lun, bool plug)
1596 {
1597 
1598 	struct vhost_scsi *vs = tpg->vhost_scsi;
1599 	struct vhost_virtqueue *vq;
1600 	u32 reason;
1601 
1602 	if (!vs)
1603 		return;
1604 
1605 	mutex_lock(&vs->dev.mutex);
1606 
1607 	if (plug)
1608 		reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1609 	else
1610 		reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1611 
1612 	vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1613 	mutex_lock(&vq->mutex);
1614 	if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1615 		vhost_scsi_send_evt(vs, tpg, lun,
1616 				   VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1617 	mutex_unlock(&vq->mutex);
1618 	mutex_unlock(&vs->dev.mutex);
1619 }
1620 
1621 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1622 {
1623 	vhost_scsi_do_plug(tpg, lun, true);
1624 }
1625 
1626 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1627 {
1628 	vhost_scsi_do_plug(tpg, lun, false);
1629 }
1630 
1631 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1632 			       struct se_lun *lun)
1633 {
1634 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1635 				struct vhost_scsi_tpg, se_tpg);
1636 
1637 	mutex_lock(&vhost_scsi_mutex);
1638 
1639 	mutex_lock(&tpg->tv_tpg_mutex);
1640 	tpg->tv_tpg_port_count++;
1641 	mutex_unlock(&tpg->tv_tpg_mutex);
1642 
1643 	vhost_scsi_hotplug(tpg, lun);
1644 
1645 	mutex_unlock(&vhost_scsi_mutex);
1646 
1647 	return 0;
1648 }
1649 
1650 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1651 				  struct se_lun *lun)
1652 {
1653 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1654 				struct vhost_scsi_tpg, se_tpg);
1655 
1656 	mutex_lock(&vhost_scsi_mutex);
1657 
1658 	mutex_lock(&tpg->tv_tpg_mutex);
1659 	tpg->tv_tpg_port_count--;
1660 	mutex_unlock(&tpg->tv_tpg_mutex);
1661 
1662 	vhost_scsi_hotunplug(tpg, lun);
1663 
1664 	mutex_unlock(&vhost_scsi_mutex);
1665 }
1666 
1667 static void vhost_scsi_free_cmd_map_res(struct vhost_scsi_nexus *nexus,
1668 				       struct se_session *se_sess)
1669 {
1670 	struct vhost_scsi_cmd *tv_cmd;
1671 	unsigned int i;
1672 
1673 	if (!se_sess->sess_cmd_map)
1674 		return;
1675 
1676 	for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1677 		tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1678 
1679 		kfree(tv_cmd->tvc_sgl);
1680 		kfree(tv_cmd->tvc_prot_sgl);
1681 		kfree(tv_cmd->tvc_upages);
1682 	}
1683 }
1684 
1685 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1686 		struct config_item *item, const char *page, size_t count)
1687 {
1688 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
1689 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1690 				struct vhost_scsi_tpg, se_tpg);
1691 	unsigned long val;
1692 	int ret = kstrtoul(page, 0, &val);
1693 
1694 	if (ret) {
1695 		pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1696 		return ret;
1697 	}
1698 	if (val != 0 && val != 1 && val != 3) {
1699 		pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1700 		return -EINVAL;
1701 	}
1702 	tpg->tv_fabric_prot_type = val;
1703 
1704 	return count;
1705 }
1706 
1707 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1708 		struct config_item *item, char *page)
1709 {
1710 	struct se_portal_group *se_tpg = attrib_to_tpg(item);
1711 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1712 				struct vhost_scsi_tpg, se_tpg);
1713 
1714 	return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1715 }
1716 
1717 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1718 
1719 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1720 	&vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1721 	NULL,
1722 };
1723 
1724 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1725 				const char *name)
1726 {
1727 	struct se_portal_group *se_tpg;
1728 	struct se_session *se_sess;
1729 	struct vhost_scsi_nexus *tv_nexus;
1730 	struct vhost_scsi_cmd *tv_cmd;
1731 	unsigned int i;
1732 
1733 	mutex_lock(&tpg->tv_tpg_mutex);
1734 	if (tpg->tpg_nexus) {
1735 		mutex_unlock(&tpg->tv_tpg_mutex);
1736 		pr_debug("tpg->tpg_nexus already exists\n");
1737 		return -EEXIST;
1738 	}
1739 	se_tpg = &tpg->se_tpg;
1740 
1741 	tv_nexus = kzalloc(sizeof(struct vhost_scsi_nexus), GFP_KERNEL);
1742 	if (!tv_nexus) {
1743 		mutex_unlock(&tpg->tv_tpg_mutex);
1744 		pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1745 		return -ENOMEM;
1746 	}
1747 	/*
1748 	 *  Initialize the struct se_session pointer and setup tagpool
1749 	 *  for struct vhost_scsi_cmd descriptors
1750 	 */
1751 	tv_nexus->tvn_se_sess = transport_init_session_tags(
1752 					VHOST_SCSI_DEFAULT_TAGS,
1753 					sizeof(struct vhost_scsi_cmd),
1754 					TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS);
1755 	if (IS_ERR(tv_nexus->tvn_se_sess)) {
1756 		mutex_unlock(&tpg->tv_tpg_mutex);
1757 		kfree(tv_nexus);
1758 		return -ENOMEM;
1759 	}
1760 	se_sess = tv_nexus->tvn_se_sess;
1761 	for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1762 		tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1763 
1764 		tv_cmd->tvc_sgl = kzalloc(sizeof(struct scatterlist) *
1765 					VHOST_SCSI_PREALLOC_SGLS, GFP_KERNEL);
1766 		if (!tv_cmd->tvc_sgl) {
1767 			mutex_unlock(&tpg->tv_tpg_mutex);
1768 			pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1769 			goto out;
1770 		}
1771 
1772 		tv_cmd->tvc_upages = kzalloc(sizeof(struct page *) *
1773 					VHOST_SCSI_PREALLOC_UPAGES, GFP_KERNEL);
1774 		if (!tv_cmd->tvc_upages) {
1775 			mutex_unlock(&tpg->tv_tpg_mutex);
1776 			pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1777 			goto out;
1778 		}
1779 
1780 		tv_cmd->tvc_prot_sgl = kzalloc(sizeof(struct scatterlist) *
1781 					VHOST_SCSI_PREALLOC_PROT_SGLS, GFP_KERNEL);
1782 		if (!tv_cmd->tvc_prot_sgl) {
1783 			mutex_unlock(&tpg->tv_tpg_mutex);
1784 			pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1785 			goto out;
1786 		}
1787 	}
1788 	/*
1789 	 * Since we are running in 'demo mode' this call with generate a
1790 	 * struct se_node_acl for the vhost_scsi struct se_portal_group with
1791 	 * the SCSI Initiator port name of the passed configfs group 'name'.
1792 	 */
1793 	tv_nexus->tvn_se_sess->se_node_acl = core_tpg_check_initiator_node_acl(
1794 				se_tpg, (unsigned char *)name);
1795 	if (!tv_nexus->tvn_se_sess->se_node_acl) {
1796 		mutex_unlock(&tpg->tv_tpg_mutex);
1797 		pr_debug("core_tpg_check_initiator_node_acl() failed"
1798 				" for %s\n", name);
1799 		goto out;
1800 	}
1801 	/*
1802 	 * Now register the TCM vhost virtual I_T Nexus as active.
1803 	 */
1804 	transport_register_session(se_tpg, tv_nexus->tvn_se_sess->se_node_acl,
1805 			tv_nexus->tvn_se_sess, tv_nexus);
1806 	tpg->tpg_nexus = tv_nexus;
1807 
1808 	mutex_unlock(&tpg->tv_tpg_mutex);
1809 	return 0;
1810 
1811 out:
1812 	vhost_scsi_free_cmd_map_res(tv_nexus, se_sess);
1813 	transport_free_session(se_sess);
1814 	kfree(tv_nexus);
1815 	return -ENOMEM;
1816 }
1817 
1818 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1819 {
1820 	struct se_session *se_sess;
1821 	struct vhost_scsi_nexus *tv_nexus;
1822 
1823 	mutex_lock(&tpg->tv_tpg_mutex);
1824 	tv_nexus = tpg->tpg_nexus;
1825 	if (!tv_nexus) {
1826 		mutex_unlock(&tpg->tv_tpg_mutex);
1827 		return -ENODEV;
1828 	}
1829 
1830 	se_sess = tv_nexus->tvn_se_sess;
1831 	if (!se_sess) {
1832 		mutex_unlock(&tpg->tv_tpg_mutex);
1833 		return -ENODEV;
1834 	}
1835 
1836 	if (tpg->tv_tpg_port_count != 0) {
1837 		mutex_unlock(&tpg->tv_tpg_mutex);
1838 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1839 			" active TPG port count: %d\n",
1840 			tpg->tv_tpg_port_count);
1841 		return -EBUSY;
1842 	}
1843 
1844 	if (tpg->tv_tpg_vhost_count != 0) {
1845 		mutex_unlock(&tpg->tv_tpg_mutex);
1846 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1847 			" active TPG vhost count: %d\n",
1848 			tpg->tv_tpg_vhost_count);
1849 		return -EBUSY;
1850 	}
1851 
1852 	pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1853 		" %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
1854 		tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1855 
1856 	vhost_scsi_free_cmd_map_res(tv_nexus, se_sess);
1857 	/*
1858 	 * Release the SCSI I_T Nexus to the emulated vhost Target Port
1859 	 */
1860 	transport_deregister_session(tv_nexus->tvn_se_sess);
1861 	tpg->tpg_nexus = NULL;
1862 	mutex_unlock(&tpg->tv_tpg_mutex);
1863 
1864 	kfree(tv_nexus);
1865 	return 0;
1866 }
1867 
1868 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
1869 {
1870 	struct se_portal_group *se_tpg = to_tpg(item);
1871 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1872 				struct vhost_scsi_tpg, se_tpg);
1873 	struct vhost_scsi_nexus *tv_nexus;
1874 	ssize_t ret;
1875 
1876 	mutex_lock(&tpg->tv_tpg_mutex);
1877 	tv_nexus = tpg->tpg_nexus;
1878 	if (!tv_nexus) {
1879 		mutex_unlock(&tpg->tv_tpg_mutex);
1880 		return -ENODEV;
1881 	}
1882 	ret = snprintf(page, PAGE_SIZE, "%s\n",
1883 			tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1884 	mutex_unlock(&tpg->tv_tpg_mutex);
1885 
1886 	return ret;
1887 }
1888 
1889 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
1890 		const char *page, size_t count)
1891 {
1892 	struct se_portal_group *se_tpg = to_tpg(item);
1893 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1894 				struct vhost_scsi_tpg, se_tpg);
1895 	struct vhost_scsi_tport *tport_wwn = tpg->tport;
1896 	unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
1897 	int ret;
1898 	/*
1899 	 * Shutdown the active I_T nexus if 'NULL' is passed..
1900 	 */
1901 	if (!strncmp(page, "NULL", 4)) {
1902 		ret = vhost_scsi_drop_nexus(tpg);
1903 		return (!ret) ? count : ret;
1904 	}
1905 	/*
1906 	 * Otherwise make sure the passed virtual Initiator port WWN matches
1907 	 * the fabric protocol_id set in vhost_scsi_make_tport(), and call
1908 	 * vhost_scsi_make_nexus().
1909 	 */
1910 	if (strlen(page) >= VHOST_SCSI_NAMELEN) {
1911 		pr_err("Emulated NAA Sas Address: %s, exceeds"
1912 				" max: %d\n", page, VHOST_SCSI_NAMELEN);
1913 		return -EINVAL;
1914 	}
1915 	snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
1916 
1917 	ptr = strstr(i_port, "naa.");
1918 	if (ptr) {
1919 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1920 			pr_err("Passed SAS Initiator Port %s does not"
1921 				" match target port protoid: %s\n", i_port,
1922 				vhost_scsi_dump_proto_id(tport_wwn));
1923 			return -EINVAL;
1924 		}
1925 		port_ptr = &i_port[0];
1926 		goto check_newline;
1927 	}
1928 	ptr = strstr(i_port, "fc.");
1929 	if (ptr) {
1930 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1931 			pr_err("Passed FCP Initiator Port %s does not"
1932 				" match target port protoid: %s\n", i_port,
1933 				vhost_scsi_dump_proto_id(tport_wwn));
1934 			return -EINVAL;
1935 		}
1936 		port_ptr = &i_port[3]; /* Skip over "fc." */
1937 		goto check_newline;
1938 	}
1939 	ptr = strstr(i_port, "iqn.");
1940 	if (ptr) {
1941 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1942 			pr_err("Passed iSCSI Initiator Port %s does not"
1943 				" match target port protoid: %s\n", i_port,
1944 				vhost_scsi_dump_proto_id(tport_wwn));
1945 			return -EINVAL;
1946 		}
1947 		port_ptr = &i_port[0];
1948 		goto check_newline;
1949 	}
1950 	pr_err("Unable to locate prefix for emulated Initiator Port:"
1951 			" %s\n", i_port);
1952 	return -EINVAL;
1953 	/*
1954 	 * Clear any trailing newline for the NAA WWN
1955 	 */
1956 check_newline:
1957 	if (i_port[strlen(i_port)-1] == '\n')
1958 		i_port[strlen(i_port)-1] = '\0';
1959 
1960 	ret = vhost_scsi_make_nexus(tpg, port_ptr);
1961 	if (ret < 0)
1962 		return ret;
1963 
1964 	return count;
1965 }
1966 
1967 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
1968 
1969 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
1970 	&vhost_scsi_tpg_attr_nexus,
1971 	NULL,
1972 };
1973 
1974 static struct se_portal_group *
1975 vhost_scsi_make_tpg(struct se_wwn *wwn,
1976 		   struct config_group *group,
1977 		   const char *name)
1978 {
1979 	struct vhost_scsi_tport *tport = container_of(wwn,
1980 			struct vhost_scsi_tport, tport_wwn);
1981 
1982 	struct vhost_scsi_tpg *tpg;
1983 	u16 tpgt;
1984 	int ret;
1985 
1986 	if (strstr(name, "tpgt_") != name)
1987 		return ERR_PTR(-EINVAL);
1988 	if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
1989 		return ERR_PTR(-EINVAL);
1990 
1991 	tpg = kzalloc(sizeof(struct vhost_scsi_tpg), GFP_KERNEL);
1992 	if (!tpg) {
1993 		pr_err("Unable to allocate struct vhost_scsi_tpg");
1994 		return ERR_PTR(-ENOMEM);
1995 	}
1996 	mutex_init(&tpg->tv_tpg_mutex);
1997 	INIT_LIST_HEAD(&tpg->tv_tpg_list);
1998 	tpg->tport = tport;
1999 	tpg->tport_tpgt = tpgt;
2000 
2001 	ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
2002 	if (ret < 0) {
2003 		kfree(tpg);
2004 		return NULL;
2005 	}
2006 	mutex_lock(&vhost_scsi_mutex);
2007 	list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
2008 	mutex_unlock(&vhost_scsi_mutex);
2009 
2010 	return &tpg->se_tpg;
2011 }
2012 
2013 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
2014 {
2015 	struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2016 				struct vhost_scsi_tpg, se_tpg);
2017 
2018 	mutex_lock(&vhost_scsi_mutex);
2019 	list_del(&tpg->tv_tpg_list);
2020 	mutex_unlock(&vhost_scsi_mutex);
2021 	/*
2022 	 * Release the virtual I_T Nexus for this vhost TPG
2023 	 */
2024 	vhost_scsi_drop_nexus(tpg);
2025 	/*
2026 	 * Deregister the se_tpg from TCM..
2027 	 */
2028 	core_tpg_deregister(se_tpg);
2029 	kfree(tpg);
2030 }
2031 
2032 static struct se_wwn *
2033 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
2034 		     struct config_group *group,
2035 		     const char *name)
2036 {
2037 	struct vhost_scsi_tport *tport;
2038 	char *ptr;
2039 	u64 wwpn = 0;
2040 	int off = 0;
2041 
2042 	/* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2043 		return ERR_PTR(-EINVAL); */
2044 
2045 	tport = kzalloc(sizeof(struct vhost_scsi_tport), GFP_KERNEL);
2046 	if (!tport) {
2047 		pr_err("Unable to allocate struct vhost_scsi_tport");
2048 		return ERR_PTR(-ENOMEM);
2049 	}
2050 	tport->tport_wwpn = wwpn;
2051 	/*
2052 	 * Determine the emulated Protocol Identifier and Target Port Name
2053 	 * based on the incoming configfs directory name.
2054 	 */
2055 	ptr = strstr(name, "naa.");
2056 	if (ptr) {
2057 		tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2058 		goto check_len;
2059 	}
2060 	ptr = strstr(name, "fc.");
2061 	if (ptr) {
2062 		tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2063 		off = 3; /* Skip over "fc." */
2064 		goto check_len;
2065 	}
2066 	ptr = strstr(name, "iqn.");
2067 	if (ptr) {
2068 		tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2069 		goto check_len;
2070 	}
2071 
2072 	pr_err("Unable to locate prefix for emulated Target Port:"
2073 			" %s\n", name);
2074 	kfree(tport);
2075 	return ERR_PTR(-EINVAL);
2076 
2077 check_len:
2078 	if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2079 		pr_err("Emulated %s Address: %s, exceeds"
2080 			" max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2081 			VHOST_SCSI_NAMELEN);
2082 		kfree(tport);
2083 		return ERR_PTR(-EINVAL);
2084 	}
2085 	snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2086 
2087 	pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2088 		" %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2089 
2090 	return &tport->tport_wwn;
2091 }
2092 
2093 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2094 {
2095 	struct vhost_scsi_tport *tport = container_of(wwn,
2096 				struct vhost_scsi_tport, tport_wwn);
2097 
2098 	pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2099 		" %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2100 		tport->tport_name);
2101 
2102 	kfree(tport);
2103 }
2104 
2105 static ssize_t
2106 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2107 {
2108 	return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2109 		"on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2110 		utsname()->machine);
2111 }
2112 
2113 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2114 
2115 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2116 	&vhost_scsi_wwn_attr_version,
2117 	NULL,
2118 };
2119 
2120 static struct target_core_fabric_ops vhost_scsi_ops = {
2121 	.module				= THIS_MODULE,
2122 	.name				= "vhost",
2123 	.get_fabric_name		= vhost_scsi_get_fabric_name,
2124 	.tpg_get_wwn			= vhost_scsi_get_fabric_wwn,
2125 	.tpg_get_tag			= vhost_scsi_get_tpgt,
2126 	.tpg_check_demo_mode		= vhost_scsi_check_true,
2127 	.tpg_check_demo_mode_cache	= vhost_scsi_check_true,
2128 	.tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2129 	.tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2130 	.tpg_check_prot_fabric_only	= vhost_scsi_check_prot_fabric_only,
2131 	.tpg_get_inst_index		= vhost_scsi_tpg_get_inst_index,
2132 	.release_cmd			= vhost_scsi_release_cmd,
2133 	.check_stop_free		= vhost_scsi_check_stop_free,
2134 	.shutdown_session		= vhost_scsi_shutdown_session,
2135 	.close_session			= vhost_scsi_close_session,
2136 	.sess_get_index			= vhost_scsi_sess_get_index,
2137 	.sess_get_initiator_sid		= NULL,
2138 	.write_pending			= vhost_scsi_write_pending,
2139 	.write_pending_status		= vhost_scsi_write_pending_status,
2140 	.set_default_node_attributes	= vhost_scsi_set_default_node_attrs,
2141 	.get_cmd_state			= vhost_scsi_get_cmd_state,
2142 	.queue_data_in			= vhost_scsi_queue_data_in,
2143 	.queue_status			= vhost_scsi_queue_status,
2144 	.queue_tm_rsp			= vhost_scsi_queue_tm_rsp,
2145 	.aborted_task			= vhost_scsi_aborted_task,
2146 	/*
2147 	 * Setup callers for generic logic in target_core_fabric_configfs.c
2148 	 */
2149 	.fabric_make_wwn		= vhost_scsi_make_tport,
2150 	.fabric_drop_wwn		= vhost_scsi_drop_tport,
2151 	.fabric_make_tpg		= vhost_scsi_make_tpg,
2152 	.fabric_drop_tpg		= vhost_scsi_drop_tpg,
2153 	.fabric_post_link		= vhost_scsi_port_link,
2154 	.fabric_pre_unlink		= vhost_scsi_port_unlink,
2155 
2156 	.tfc_wwn_attrs			= vhost_scsi_wwn_attrs,
2157 	.tfc_tpg_base_attrs		= vhost_scsi_tpg_attrs,
2158 	.tfc_tpg_attrib_attrs		= vhost_scsi_tpg_attrib_attrs,
2159 };
2160 
2161 static int __init vhost_scsi_init(void)
2162 {
2163 	int ret = -ENOMEM;
2164 
2165 	pr_debug("TCM_VHOST fabric module %s on %s/%s"
2166 		" on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2167 		utsname()->machine);
2168 
2169 	/*
2170 	 * Use our own dedicated workqueue for submitting I/O into
2171 	 * target core to avoid contention within system_wq.
2172 	 */
2173 	vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2174 	if (!vhost_scsi_workqueue)
2175 		goto out;
2176 
2177 	ret = vhost_scsi_register();
2178 	if (ret < 0)
2179 		goto out_destroy_workqueue;
2180 
2181 	ret = target_register_template(&vhost_scsi_ops);
2182 	if (ret < 0)
2183 		goto out_vhost_scsi_deregister;
2184 
2185 	return 0;
2186 
2187 out_vhost_scsi_deregister:
2188 	vhost_scsi_deregister();
2189 out_destroy_workqueue:
2190 	destroy_workqueue(vhost_scsi_workqueue);
2191 out:
2192 	return ret;
2193 };
2194 
2195 static void vhost_scsi_exit(void)
2196 {
2197 	target_unregister_template(&vhost_scsi_ops);
2198 	vhost_scsi_deregister();
2199 	destroy_workqueue(vhost_scsi_workqueue);
2200 };
2201 
2202 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2203 MODULE_ALIAS("tcm_vhost");
2204 MODULE_LICENSE("GPL");
2205 module_init(vhost_scsi_init);
2206 module_exit(vhost_scsi_exit);
2207