xref: /openbmc/linux/drivers/vhost/scsi.c (revision 66de4859)
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/miscdevice.h>
39 #include <asm/unaligned.h>
40 #include <scsi/scsi.h>
41 #include <scsi/scsi_tcq.h>
42 #include <target/target_core_base.h>
43 #include <target/target_core_fabric.h>
44 #include <target/target_core_fabric_configfs.h>
45 #include <target/target_core_configfs.h>
46 #include <target/configfs_macros.h>
47 #include <linux/vhost.h>
48 #include <linux/virtio_scsi.h>
49 #include <linux/llist.h>
50 #include <linux/bitmap.h>
51 #include <linux/percpu_ida.h>
52 
53 #include "vhost.h"
54 
55 #define TCM_VHOST_VERSION  "v0.1"
56 #define TCM_VHOST_NAMELEN 256
57 #define TCM_VHOST_MAX_CDB_SIZE 32
58 #define TCM_VHOST_DEFAULT_TAGS 256
59 #define TCM_VHOST_PREALLOC_SGLS 2048
60 #define TCM_VHOST_PREALLOC_PAGES 2048
61 
62 struct vhost_scsi_inflight {
63 	/* Wait for the flush operation to finish */
64 	struct completion comp;
65 	/* Refcount for the inflight reqs */
66 	struct kref kref;
67 };
68 
69 struct tcm_vhost_cmd {
70 	/* Descriptor from vhost_get_vq_desc() for virt_queue segment */
71 	int tvc_vq_desc;
72 	/* virtio-scsi initiator task attribute */
73 	int tvc_task_attr;
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 	/* Saved unpacked SCSI LUN for tcm_vhost_submission_work() */
83 	u32 tvc_lun;
84 	/* Pointer to the SGL formatted memory from virtio-scsi */
85 	struct scatterlist *tvc_sgl;
86 	struct page **tvc_upages;
87 	/* Pointer to response */
88 	struct virtio_scsi_cmd_resp __user *tvc_resp;
89 	/* Pointer to vhost_scsi for our device */
90 	struct vhost_scsi *tvc_vhost;
91 	/* Pointer to vhost_virtqueue for the cmd */
92 	struct vhost_virtqueue *tvc_vq;
93 	/* Pointer to vhost nexus memory */
94 	struct tcm_vhost_nexus *tvc_nexus;
95 	/* The TCM I/O descriptor that is accessed via container_of() */
96 	struct se_cmd tvc_se_cmd;
97 	/* work item used for cmwq dispatch to tcm_vhost_submission_work() */
98 	struct work_struct work;
99 	/* Copy of the incoming SCSI command descriptor block (CDB) */
100 	unsigned char tvc_cdb[TCM_VHOST_MAX_CDB_SIZE];
101 	/* Sense buffer that will be mapped into outgoing status */
102 	unsigned char tvc_sense_buf[TRANSPORT_SENSE_BUFFER];
103 	/* Completed commands list, serviced from vhost worker thread */
104 	struct llist_node tvc_completion_list;
105 	/* Used to track inflight cmd */
106 	struct vhost_scsi_inflight *inflight;
107 };
108 
109 struct tcm_vhost_nexus {
110 	/* Pointer to TCM session for I_T Nexus */
111 	struct se_session *tvn_se_sess;
112 };
113 
114 struct tcm_vhost_nacl {
115 	/* Binary World Wide unique Port Name for Vhost Initiator port */
116 	u64 iport_wwpn;
117 	/* ASCII formatted WWPN for Sas Initiator port */
118 	char iport_name[TCM_VHOST_NAMELEN];
119 	/* Returned by tcm_vhost_make_nodeacl() */
120 	struct se_node_acl se_node_acl;
121 };
122 
123 struct tcm_vhost_tpg {
124 	/* Vhost port target portal group tag for TCM */
125 	u16 tport_tpgt;
126 	/* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
127 	int tv_tpg_port_count;
128 	/* Used for vhost_scsi device reference to tpg_nexus, protected by tv_tpg_mutex */
129 	int tv_tpg_vhost_count;
130 	/* list for tcm_vhost_list */
131 	struct list_head tv_tpg_list;
132 	/* Used to protect access for tpg_nexus */
133 	struct mutex tv_tpg_mutex;
134 	/* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
135 	struct tcm_vhost_nexus *tpg_nexus;
136 	/* Pointer back to tcm_vhost_tport */
137 	struct tcm_vhost_tport *tport;
138 	/* Returned by tcm_vhost_make_tpg() */
139 	struct se_portal_group se_tpg;
140 	/* Pointer back to vhost_scsi, protected by tv_tpg_mutex */
141 	struct vhost_scsi *vhost_scsi;
142 };
143 
144 struct tcm_vhost_tport {
145 	/* SCSI protocol the tport is providing */
146 	u8 tport_proto_id;
147 	/* Binary World Wide unique Port Name for Vhost Target port */
148 	u64 tport_wwpn;
149 	/* ASCII formatted WWPN for Vhost Target port */
150 	char tport_name[TCM_VHOST_NAMELEN];
151 	/* Returned by tcm_vhost_make_tport() */
152 	struct se_wwn tport_wwn;
153 };
154 
155 struct tcm_vhost_evt {
156 	/* event to be sent to guest */
157 	struct virtio_scsi_event event;
158 	/* event list, serviced from vhost worker thread */
159 	struct llist_node list;
160 };
161 
162 enum {
163 	VHOST_SCSI_VQ_CTL = 0,
164 	VHOST_SCSI_VQ_EVT = 1,
165 	VHOST_SCSI_VQ_IO = 2,
166 };
167 
168 enum {
169 	VHOST_SCSI_FEATURES = VHOST_FEATURES | (1ULL << VIRTIO_SCSI_F_HOTPLUG)
170 };
171 
172 #define VHOST_SCSI_MAX_TARGET	256
173 #define VHOST_SCSI_MAX_VQ	128
174 #define VHOST_SCSI_MAX_EVENT	128
175 
176 struct vhost_scsi_virtqueue {
177 	struct vhost_virtqueue vq;
178 	/*
179 	 * Reference counting for inflight reqs, used for flush operation. At
180 	 * each time, one reference tracks new commands submitted, while we
181 	 * wait for another one to reach 0.
182 	 */
183 	struct vhost_scsi_inflight inflights[2];
184 	/*
185 	 * Indicate current inflight in use, protected by vq->mutex.
186 	 * Writers must also take dev mutex and flush under it.
187 	 */
188 	int inflight_idx;
189 };
190 
191 struct vhost_scsi {
192 	/* Protected by vhost_scsi->dev.mutex */
193 	struct tcm_vhost_tpg **vs_tpg;
194 	char vs_vhost_wwpn[TRANSPORT_IQN_LEN];
195 
196 	struct vhost_dev dev;
197 	struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
198 
199 	struct vhost_work vs_completion_work; /* cmd completion work item */
200 	struct llist_head vs_completion_list; /* cmd completion queue */
201 
202 	struct vhost_work vs_event_work; /* evt injection work item */
203 	struct llist_head vs_event_list; /* evt injection queue */
204 
205 	bool vs_events_missed; /* any missed events, protected by vq->mutex */
206 	int vs_events_nr; /* num of pending events, protected by vq->mutex */
207 };
208 
209 /* Local pointer to allocated TCM configfs fabric module */
210 static struct target_fabric_configfs *tcm_vhost_fabric_configfs;
211 
212 static struct workqueue_struct *tcm_vhost_workqueue;
213 
214 /* Global spinlock to protect tcm_vhost TPG list for vhost IOCTL access */
215 static DEFINE_MUTEX(tcm_vhost_mutex);
216 static LIST_HEAD(tcm_vhost_list);
217 
218 static int iov_num_pages(struct iovec *iov)
219 {
220 	return (PAGE_ALIGN((unsigned long)iov->iov_base + iov->iov_len) -
221 	       ((unsigned long)iov->iov_base & PAGE_MASK)) >> PAGE_SHIFT;
222 }
223 
224 static void tcm_vhost_done_inflight(struct kref *kref)
225 {
226 	struct vhost_scsi_inflight *inflight;
227 
228 	inflight = container_of(kref, struct vhost_scsi_inflight, kref);
229 	complete(&inflight->comp);
230 }
231 
232 static void tcm_vhost_init_inflight(struct vhost_scsi *vs,
233 				    struct vhost_scsi_inflight *old_inflight[])
234 {
235 	struct vhost_scsi_inflight *new_inflight;
236 	struct vhost_virtqueue *vq;
237 	int idx, i;
238 
239 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
240 		vq = &vs->vqs[i].vq;
241 
242 		mutex_lock(&vq->mutex);
243 
244 		/* store old infight */
245 		idx = vs->vqs[i].inflight_idx;
246 		if (old_inflight)
247 			old_inflight[i] = &vs->vqs[i].inflights[idx];
248 
249 		/* setup new infight */
250 		vs->vqs[i].inflight_idx = idx ^ 1;
251 		new_inflight = &vs->vqs[i].inflights[idx ^ 1];
252 		kref_init(&new_inflight->kref);
253 		init_completion(&new_inflight->comp);
254 
255 		mutex_unlock(&vq->mutex);
256 	}
257 }
258 
259 static struct vhost_scsi_inflight *
260 tcm_vhost_get_inflight(struct vhost_virtqueue *vq)
261 {
262 	struct vhost_scsi_inflight *inflight;
263 	struct vhost_scsi_virtqueue *svq;
264 
265 	svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
266 	inflight = &svq->inflights[svq->inflight_idx];
267 	kref_get(&inflight->kref);
268 
269 	return inflight;
270 }
271 
272 static void tcm_vhost_put_inflight(struct vhost_scsi_inflight *inflight)
273 {
274 	kref_put(&inflight->kref, tcm_vhost_done_inflight);
275 }
276 
277 static int tcm_vhost_check_true(struct se_portal_group *se_tpg)
278 {
279 	return 1;
280 }
281 
282 static int tcm_vhost_check_false(struct se_portal_group *se_tpg)
283 {
284 	return 0;
285 }
286 
287 static char *tcm_vhost_get_fabric_name(void)
288 {
289 	return "vhost";
290 }
291 
292 static u8 tcm_vhost_get_fabric_proto_ident(struct se_portal_group *se_tpg)
293 {
294 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
295 				struct tcm_vhost_tpg, se_tpg);
296 	struct tcm_vhost_tport *tport = tpg->tport;
297 
298 	switch (tport->tport_proto_id) {
299 	case SCSI_PROTOCOL_SAS:
300 		return sas_get_fabric_proto_ident(se_tpg);
301 	case SCSI_PROTOCOL_FCP:
302 		return fc_get_fabric_proto_ident(se_tpg);
303 	case SCSI_PROTOCOL_ISCSI:
304 		return iscsi_get_fabric_proto_ident(se_tpg);
305 	default:
306 		pr_err("Unknown tport_proto_id: 0x%02x, using"
307 			" SAS emulation\n", tport->tport_proto_id);
308 		break;
309 	}
310 
311 	return sas_get_fabric_proto_ident(se_tpg);
312 }
313 
314 static char *tcm_vhost_get_fabric_wwn(struct se_portal_group *se_tpg)
315 {
316 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
317 				struct tcm_vhost_tpg, se_tpg);
318 	struct tcm_vhost_tport *tport = tpg->tport;
319 
320 	return &tport->tport_name[0];
321 }
322 
323 static u16 tcm_vhost_get_tag(struct se_portal_group *se_tpg)
324 {
325 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
326 				struct tcm_vhost_tpg, se_tpg);
327 	return tpg->tport_tpgt;
328 }
329 
330 static u32 tcm_vhost_get_default_depth(struct se_portal_group *se_tpg)
331 {
332 	return 1;
333 }
334 
335 static u32
336 tcm_vhost_get_pr_transport_id(struct se_portal_group *se_tpg,
337 			      struct se_node_acl *se_nacl,
338 			      struct t10_pr_registration *pr_reg,
339 			      int *format_code,
340 			      unsigned char *buf)
341 {
342 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
343 				struct tcm_vhost_tpg, se_tpg);
344 	struct tcm_vhost_tport *tport = tpg->tport;
345 
346 	switch (tport->tport_proto_id) {
347 	case SCSI_PROTOCOL_SAS:
348 		return sas_get_pr_transport_id(se_tpg, se_nacl, pr_reg,
349 					format_code, buf);
350 	case SCSI_PROTOCOL_FCP:
351 		return fc_get_pr_transport_id(se_tpg, se_nacl, pr_reg,
352 					format_code, buf);
353 	case SCSI_PROTOCOL_ISCSI:
354 		return iscsi_get_pr_transport_id(se_tpg, se_nacl, pr_reg,
355 					format_code, buf);
356 	default:
357 		pr_err("Unknown tport_proto_id: 0x%02x, using"
358 			" SAS emulation\n", tport->tport_proto_id);
359 		break;
360 	}
361 
362 	return sas_get_pr_transport_id(se_tpg, se_nacl, pr_reg,
363 			format_code, buf);
364 }
365 
366 static u32
367 tcm_vhost_get_pr_transport_id_len(struct se_portal_group *se_tpg,
368 				  struct se_node_acl *se_nacl,
369 				  struct t10_pr_registration *pr_reg,
370 				  int *format_code)
371 {
372 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
373 				struct tcm_vhost_tpg, se_tpg);
374 	struct tcm_vhost_tport *tport = tpg->tport;
375 
376 	switch (tport->tport_proto_id) {
377 	case SCSI_PROTOCOL_SAS:
378 		return sas_get_pr_transport_id_len(se_tpg, se_nacl, pr_reg,
379 					format_code);
380 	case SCSI_PROTOCOL_FCP:
381 		return fc_get_pr_transport_id_len(se_tpg, se_nacl, pr_reg,
382 					format_code);
383 	case SCSI_PROTOCOL_ISCSI:
384 		return iscsi_get_pr_transport_id_len(se_tpg, se_nacl, pr_reg,
385 					format_code);
386 	default:
387 		pr_err("Unknown tport_proto_id: 0x%02x, using"
388 			" SAS emulation\n", tport->tport_proto_id);
389 		break;
390 	}
391 
392 	return sas_get_pr_transport_id_len(se_tpg, se_nacl, pr_reg,
393 			format_code);
394 }
395 
396 static char *
397 tcm_vhost_parse_pr_out_transport_id(struct se_portal_group *se_tpg,
398 				    const char *buf,
399 				    u32 *out_tid_len,
400 				    char **port_nexus_ptr)
401 {
402 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
403 				struct tcm_vhost_tpg, se_tpg);
404 	struct tcm_vhost_tport *tport = tpg->tport;
405 
406 	switch (tport->tport_proto_id) {
407 	case SCSI_PROTOCOL_SAS:
408 		return sas_parse_pr_out_transport_id(se_tpg, buf, out_tid_len,
409 					port_nexus_ptr);
410 	case SCSI_PROTOCOL_FCP:
411 		return fc_parse_pr_out_transport_id(se_tpg, buf, out_tid_len,
412 					port_nexus_ptr);
413 	case SCSI_PROTOCOL_ISCSI:
414 		return iscsi_parse_pr_out_transport_id(se_tpg, buf, out_tid_len,
415 					port_nexus_ptr);
416 	default:
417 		pr_err("Unknown tport_proto_id: 0x%02x, using"
418 			" SAS emulation\n", tport->tport_proto_id);
419 		break;
420 	}
421 
422 	return sas_parse_pr_out_transport_id(se_tpg, buf, out_tid_len,
423 			port_nexus_ptr);
424 }
425 
426 static struct se_node_acl *
427 tcm_vhost_alloc_fabric_acl(struct se_portal_group *se_tpg)
428 {
429 	struct tcm_vhost_nacl *nacl;
430 
431 	nacl = kzalloc(sizeof(struct tcm_vhost_nacl), GFP_KERNEL);
432 	if (!nacl) {
433 		pr_err("Unable to allocate struct tcm_vhost_nacl\n");
434 		return NULL;
435 	}
436 
437 	return &nacl->se_node_acl;
438 }
439 
440 static void
441 tcm_vhost_release_fabric_acl(struct se_portal_group *se_tpg,
442 			     struct se_node_acl *se_nacl)
443 {
444 	struct tcm_vhost_nacl *nacl = container_of(se_nacl,
445 			struct tcm_vhost_nacl, se_node_acl);
446 	kfree(nacl);
447 }
448 
449 static u32 tcm_vhost_tpg_get_inst_index(struct se_portal_group *se_tpg)
450 {
451 	return 1;
452 }
453 
454 static void tcm_vhost_release_cmd(struct se_cmd *se_cmd)
455 {
456 	struct tcm_vhost_cmd *tv_cmd = container_of(se_cmd,
457 				struct tcm_vhost_cmd, tvc_se_cmd);
458 	struct se_session *se_sess = se_cmd->se_sess;
459 
460 	if (tv_cmd->tvc_sgl_count) {
461 		u32 i;
462 		for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
463 			put_page(sg_page(&tv_cmd->tvc_sgl[i]));
464 	}
465 
466 	tcm_vhost_put_inflight(tv_cmd->inflight);
467 	percpu_ida_free(&se_sess->sess_tag_pool, se_cmd->map_tag);
468 }
469 
470 static int tcm_vhost_shutdown_session(struct se_session *se_sess)
471 {
472 	return 0;
473 }
474 
475 static void tcm_vhost_close_session(struct se_session *se_sess)
476 {
477 	return;
478 }
479 
480 static u32 tcm_vhost_sess_get_index(struct se_session *se_sess)
481 {
482 	return 0;
483 }
484 
485 static int tcm_vhost_write_pending(struct se_cmd *se_cmd)
486 {
487 	/* Go ahead and process the write immediately */
488 	target_execute_cmd(se_cmd);
489 	return 0;
490 }
491 
492 static int tcm_vhost_write_pending_status(struct se_cmd *se_cmd)
493 {
494 	return 0;
495 }
496 
497 static void tcm_vhost_set_default_node_attrs(struct se_node_acl *nacl)
498 {
499 	return;
500 }
501 
502 static u32 tcm_vhost_get_task_tag(struct se_cmd *se_cmd)
503 {
504 	return 0;
505 }
506 
507 static int tcm_vhost_get_cmd_state(struct se_cmd *se_cmd)
508 {
509 	return 0;
510 }
511 
512 static void vhost_scsi_complete_cmd(struct tcm_vhost_cmd *cmd)
513 {
514 	struct vhost_scsi *vs = cmd->tvc_vhost;
515 
516 	llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
517 
518 	vhost_work_queue(&vs->dev, &vs->vs_completion_work);
519 }
520 
521 static int tcm_vhost_queue_data_in(struct se_cmd *se_cmd)
522 {
523 	struct tcm_vhost_cmd *cmd = container_of(se_cmd,
524 				struct tcm_vhost_cmd, tvc_se_cmd);
525 	vhost_scsi_complete_cmd(cmd);
526 	return 0;
527 }
528 
529 static int tcm_vhost_queue_status(struct se_cmd *se_cmd)
530 {
531 	struct tcm_vhost_cmd *cmd = container_of(se_cmd,
532 				struct tcm_vhost_cmd, tvc_se_cmd);
533 	vhost_scsi_complete_cmd(cmd);
534 	return 0;
535 }
536 
537 static void tcm_vhost_queue_tm_rsp(struct se_cmd *se_cmd)
538 {
539 	return;
540 }
541 
542 static void tcm_vhost_aborted_task(struct se_cmd *se_cmd)
543 {
544 	return;
545 }
546 
547 static void tcm_vhost_free_evt(struct vhost_scsi *vs, struct tcm_vhost_evt *evt)
548 {
549 	vs->vs_events_nr--;
550 	kfree(evt);
551 }
552 
553 static struct tcm_vhost_evt *
554 tcm_vhost_allocate_evt(struct vhost_scsi *vs,
555 		       u32 event, u32 reason)
556 {
557 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
558 	struct tcm_vhost_evt *evt;
559 
560 	if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
561 		vs->vs_events_missed = true;
562 		return NULL;
563 	}
564 
565 	evt = kzalloc(sizeof(*evt), GFP_KERNEL);
566 	if (!evt) {
567 		vq_err(vq, "Failed to allocate tcm_vhost_evt\n");
568 		vs->vs_events_missed = true;
569 		return NULL;
570 	}
571 
572 	evt->event.event = event;
573 	evt->event.reason = reason;
574 	vs->vs_events_nr++;
575 
576 	return evt;
577 }
578 
579 static void vhost_scsi_free_cmd(struct tcm_vhost_cmd *cmd)
580 {
581 	struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
582 
583 	/* TODO locking against target/backend threads? */
584 	transport_generic_free_cmd(se_cmd, 0);
585 
586 }
587 
588 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
589 {
590 	return target_put_sess_cmd(se_cmd->se_sess, se_cmd);
591 }
592 
593 static void
594 tcm_vhost_do_evt_work(struct vhost_scsi *vs, struct tcm_vhost_evt *evt)
595 {
596 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
597 	struct virtio_scsi_event *event = &evt->event;
598 	struct virtio_scsi_event __user *eventp;
599 	unsigned out, in;
600 	int head, ret;
601 
602 	if (!vq->private_data) {
603 		vs->vs_events_missed = true;
604 		return;
605 	}
606 
607 again:
608 	vhost_disable_notify(&vs->dev, vq);
609 	head = vhost_get_vq_desc(&vs->dev, vq, vq->iov,
610 			ARRAY_SIZE(vq->iov), &out, &in,
611 			NULL, NULL);
612 	if (head < 0) {
613 		vs->vs_events_missed = true;
614 		return;
615 	}
616 	if (head == vq->num) {
617 		if (vhost_enable_notify(&vs->dev, vq))
618 			goto again;
619 		vs->vs_events_missed = true;
620 		return;
621 	}
622 
623 	if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
624 		vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
625 				vq->iov[out].iov_len);
626 		vs->vs_events_missed = true;
627 		return;
628 	}
629 
630 	if (vs->vs_events_missed) {
631 		event->event |= VIRTIO_SCSI_T_EVENTS_MISSED;
632 		vs->vs_events_missed = false;
633 	}
634 
635 	eventp = vq->iov[out].iov_base;
636 	ret = __copy_to_user(eventp, event, sizeof(*event));
637 	if (!ret)
638 		vhost_add_used_and_signal(&vs->dev, vq, head, 0);
639 	else
640 		vq_err(vq, "Faulted on tcm_vhost_send_event\n");
641 }
642 
643 static void tcm_vhost_evt_work(struct vhost_work *work)
644 {
645 	struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
646 					vs_event_work);
647 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
648 	struct tcm_vhost_evt *evt;
649 	struct llist_node *llnode;
650 
651 	mutex_lock(&vq->mutex);
652 	llnode = llist_del_all(&vs->vs_event_list);
653 	while (llnode) {
654 		evt = llist_entry(llnode, struct tcm_vhost_evt, list);
655 		llnode = llist_next(llnode);
656 		tcm_vhost_do_evt_work(vs, evt);
657 		tcm_vhost_free_evt(vs, evt);
658 	}
659 	mutex_unlock(&vq->mutex);
660 }
661 
662 /* Fill in status and signal that we are done processing this command
663  *
664  * This is scheduled in the vhost work queue so we are called with the owner
665  * process mm and can access the vring.
666  */
667 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
668 {
669 	struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
670 					vs_completion_work);
671 	DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
672 	struct virtio_scsi_cmd_resp v_rsp;
673 	struct tcm_vhost_cmd *cmd;
674 	struct llist_node *llnode;
675 	struct se_cmd *se_cmd;
676 	int ret, vq;
677 
678 	bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
679 	llnode = llist_del_all(&vs->vs_completion_list);
680 	while (llnode) {
681 		cmd = llist_entry(llnode, struct tcm_vhost_cmd,
682 				     tvc_completion_list);
683 		llnode = llist_next(llnode);
684 		se_cmd = &cmd->tvc_se_cmd;
685 
686 		pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
687 			cmd, se_cmd->residual_count, se_cmd->scsi_status);
688 
689 		memset(&v_rsp, 0, sizeof(v_rsp));
690 		v_rsp.resid = se_cmd->residual_count;
691 		/* TODO is status_qualifier field needed? */
692 		v_rsp.status = se_cmd->scsi_status;
693 		v_rsp.sense_len = se_cmd->scsi_sense_length;
694 		memcpy(v_rsp.sense, cmd->tvc_sense_buf,
695 		       v_rsp.sense_len);
696 		ret = copy_to_user(cmd->tvc_resp, &v_rsp, sizeof(v_rsp));
697 		if (likely(ret == 0)) {
698 			struct vhost_scsi_virtqueue *q;
699 			vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
700 			q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
701 			vq = q - vs->vqs;
702 			__set_bit(vq, signal);
703 		} else
704 			pr_err("Faulted on virtio_scsi_cmd_resp\n");
705 
706 		vhost_scsi_free_cmd(cmd);
707 	}
708 
709 	vq = -1;
710 	while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
711 		< VHOST_SCSI_MAX_VQ)
712 		vhost_signal(&vs->dev, &vs->vqs[vq].vq);
713 }
714 
715 static struct tcm_vhost_cmd *
716 vhost_scsi_get_tag(struct vhost_virtqueue *vq,
717 			struct tcm_vhost_tpg *tpg,
718 			struct virtio_scsi_cmd_req *v_req,
719 			u32 exp_data_len,
720 			int data_direction)
721 {
722 	struct tcm_vhost_cmd *cmd;
723 	struct tcm_vhost_nexus *tv_nexus;
724 	struct se_session *se_sess;
725 	struct scatterlist *sg;
726 	struct page **pages;
727 	int tag;
728 
729 	tv_nexus = tpg->tpg_nexus;
730 	if (!tv_nexus) {
731 		pr_err("Unable to locate active struct tcm_vhost_nexus\n");
732 		return ERR_PTR(-EIO);
733 	}
734 	se_sess = tv_nexus->tvn_se_sess;
735 
736 	tag = percpu_ida_alloc(&se_sess->sess_tag_pool, TASK_RUNNING);
737 	if (tag < 0) {
738 		pr_err("Unable to obtain tag for tcm_vhost_cmd\n");
739 		return ERR_PTR(-ENOMEM);
740 	}
741 
742 	cmd = &((struct tcm_vhost_cmd *)se_sess->sess_cmd_map)[tag];
743 	sg = cmd->tvc_sgl;
744 	pages = cmd->tvc_upages;
745 	memset(cmd, 0, sizeof(struct tcm_vhost_cmd));
746 
747 	cmd->tvc_sgl = sg;
748 	cmd->tvc_upages = pages;
749 	cmd->tvc_se_cmd.map_tag = tag;
750 	cmd->tvc_tag = v_req->tag;
751 	cmd->tvc_task_attr = v_req->task_attr;
752 	cmd->tvc_exp_data_len = exp_data_len;
753 	cmd->tvc_data_direction = data_direction;
754 	cmd->tvc_nexus = tv_nexus;
755 	cmd->inflight = tcm_vhost_get_inflight(vq);
756 
757 	return cmd;
758 }
759 
760 /*
761  * Map a user memory range into a scatterlist
762  *
763  * Returns the number of scatterlist entries used or -errno on error.
764  */
765 static int
766 vhost_scsi_map_to_sgl(struct tcm_vhost_cmd *tv_cmd,
767 		      struct scatterlist *sgl,
768 		      unsigned int sgl_count,
769 		      struct iovec *iov,
770 		      int write)
771 {
772 	unsigned int npages = 0, pages_nr, offset, nbytes;
773 	struct scatterlist *sg = sgl;
774 	void __user *ptr = iov->iov_base;
775 	size_t len = iov->iov_len;
776 	struct page **pages;
777 	int ret, i;
778 
779 	if (sgl_count > TCM_VHOST_PREALLOC_SGLS) {
780 		pr_err("vhost_scsi_map_to_sgl() psgl_count: %u greater than"
781 		       " preallocated TCM_VHOST_PREALLOC_SGLS: %u\n",
782 			sgl_count, TCM_VHOST_PREALLOC_SGLS);
783 		return -ENOBUFS;
784 	}
785 
786 	pages_nr = iov_num_pages(iov);
787 	if (pages_nr > sgl_count)
788 		return -ENOBUFS;
789 
790 	if (pages_nr > TCM_VHOST_PREALLOC_PAGES) {
791 		pr_err("vhost_scsi_map_to_sgl() pages_nr: %u greater than"
792 		       " preallocated TCM_VHOST_PREALLOC_PAGES: %u\n",
793 			pages_nr, TCM_VHOST_PREALLOC_PAGES);
794 		return -ENOBUFS;
795 	}
796 
797 	pages = tv_cmd->tvc_upages;
798 
799 	ret = get_user_pages_fast((unsigned long)ptr, pages_nr, write, pages);
800 	/* No pages were pinned */
801 	if (ret < 0)
802 		goto out;
803 	/* Less pages pinned than wanted */
804 	if (ret != pages_nr) {
805 		for (i = 0; i < ret; i++)
806 			put_page(pages[i]);
807 		ret = -EFAULT;
808 		goto out;
809 	}
810 
811 	while (len > 0) {
812 		offset = (uintptr_t)ptr & ~PAGE_MASK;
813 		nbytes = min_t(unsigned int, PAGE_SIZE - offset, len);
814 		sg_set_page(sg, pages[npages], nbytes, offset);
815 		ptr += nbytes;
816 		len -= nbytes;
817 		sg++;
818 		npages++;
819 	}
820 
821 out:
822 	return ret;
823 }
824 
825 static int
826 vhost_scsi_map_iov_to_sgl(struct tcm_vhost_cmd *cmd,
827 			  struct iovec *iov,
828 			  unsigned int niov,
829 			  int write)
830 {
831 	int ret;
832 	unsigned int i;
833 	u32 sgl_count;
834 	struct scatterlist *sg;
835 
836 	/*
837 	 * Find out how long sglist needs to be
838 	 */
839 	sgl_count = 0;
840 	for (i = 0; i < niov; i++)
841 		sgl_count += iov_num_pages(&iov[i]);
842 
843 	/* TODO overflow checking */
844 
845 	sg = cmd->tvc_sgl;
846 	pr_debug("%s sg %p sgl_count %u\n", __func__, sg, sgl_count);
847 	sg_init_table(sg, sgl_count);
848 
849 	cmd->tvc_sgl_count = sgl_count;
850 
851 	pr_debug("Mapping %u iovecs for %u pages\n", niov, sgl_count);
852 	for (i = 0; i < niov; i++) {
853 		ret = vhost_scsi_map_to_sgl(cmd, sg, sgl_count, &iov[i],
854 					    write);
855 		if (ret < 0) {
856 			for (i = 0; i < cmd->tvc_sgl_count; i++)
857 				put_page(sg_page(&cmd->tvc_sgl[i]));
858 
859 			cmd->tvc_sgl_count = 0;
860 			return ret;
861 		}
862 
863 		sg += ret;
864 		sgl_count -= ret;
865 	}
866 	return 0;
867 }
868 
869 static void tcm_vhost_submission_work(struct work_struct *work)
870 {
871 	struct tcm_vhost_cmd *cmd =
872 		container_of(work, struct tcm_vhost_cmd, work);
873 	struct tcm_vhost_nexus *tv_nexus;
874 	struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
875 	struct scatterlist *sg_ptr, *sg_bidi_ptr = NULL;
876 	int rc, sg_no_bidi = 0;
877 
878 	if (cmd->tvc_sgl_count) {
879 		sg_ptr = cmd->tvc_sgl;
880 /* FIXME: Fix BIDI operation in tcm_vhost_submission_work() */
881 #if 0
882 		if (se_cmd->se_cmd_flags & SCF_BIDI) {
883 			sg_bidi_ptr = NULL;
884 			sg_no_bidi = 0;
885 		}
886 #endif
887 	} else {
888 		sg_ptr = NULL;
889 	}
890 	tv_nexus = cmd->tvc_nexus;
891 
892 	rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
893 			cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
894 			cmd->tvc_lun, cmd->tvc_exp_data_len,
895 			cmd->tvc_task_attr, cmd->tvc_data_direction,
896 			TARGET_SCF_ACK_KREF, sg_ptr, cmd->tvc_sgl_count,
897 			sg_bidi_ptr, sg_no_bidi, NULL, 0);
898 	if (rc < 0) {
899 		transport_send_check_condition_and_sense(se_cmd,
900 				TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
901 		transport_generic_free_cmd(se_cmd, 0);
902 	}
903 }
904 
905 static void
906 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
907 			   struct vhost_virtqueue *vq,
908 			   int head, unsigned out)
909 {
910 	struct virtio_scsi_cmd_resp __user *resp;
911 	struct virtio_scsi_cmd_resp rsp;
912 	int ret;
913 
914 	memset(&rsp, 0, sizeof(rsp));
915 	rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
916 	resp = vq->iov[out].iov_base;
917 	ret = __copy_to_user(resp, &rsp, sizeof(rsp));
918 	if (!ret)
919 		vhost_add_used_and_signal(&vs->dev, vq, head, 0);
920 	else
921 		pr_err("Faulted on virtio_scsi_cmd_resp\n");
922 }
923 
924 static void
925 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
926 {
927 	struct tcm_vhost_tpg **vs_tpg;
928 	struct virtio_scsi_cmd_req v_req;
929 	struct tcm_vhost_tpg *tpg;
930 	struct tcm_vhost_cmd *cmd;
931 	u32 exp_data_len, data_first, data_num, data_direction;
932 	unsigned out, in, i;
933 	int head, ret;
934 	u8 target;
935 
936 	mutex_lock(&vq->mutex);
937 	/*
938 	 * We can handle the vq only after the endpoint is setup by calling the
939 	 * VHOST_SCSI_SET_ENDPOINT ioctl.
940 	 */
941 	vs_tpg = vq->private_data;
942 	if (!vs_tpg)
943 		goto out;
944 
945 	vhost_disable_notify(&vs->dev, vq);
946 
947 	for (;;) {
948 		head = vhost_get_vq_desc(&vs->dev, vq, vq->iov,
949 					ARRAY_SIZE(vq->iov), &out, &in,
950 					NULL, NULL);
951 		pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
952 					head, out, in);
953 		/* On error, stop handling until the next kick. */
954 		if (unlikely(head < 0))
955 			break;
956 		/* Nothing new?  Wait for eventfd to tell us they refilled. */
957 		if (head == vq->num) {
958 			if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
959 				vhost_disable_notify(&vs->dev, vq);
960 				continue;
961 			}
962 			break;
963 		}
964 
965 /* FIXME: BIDI operation */
966 		if (out == 1 && in == 1) {
967 			data_direction = DMA_NONE;
968 			data_first = 0;
969 			data_num = 0;
970 		} else if (out == 1 && in > 1) {
971 			data_direction = DMA_FROM_DEVICE;
972 			data_first = out + 1;
973 			data_num = in - 1;
974 		} else if (out > 1 && in == 1) {
975 			data_direction = DMA_TO_DEVICE;
976 			data_first = 1;
977 			data_num = out - 1;
978 		} else {
979 			vq_err(vq, "Invalid buffer layout out: %u in: %u\n",
980 					out, in);
981 			break;
982 		}
983 
984 		/*
985 		 * Check for a sane resp buffer so we can report errors to
986 		 * the guest.
987 		 */
988 		if (unlikely(vq->iov[out].iov_len !=
989 					sizeof(struct virtio_scsi_cmd_resp))) {
990 			vq_err(vq, "Expecting virtio_scsi_cmd_resp, got %zu"
991 				" bytes\n", vq->iov[out].iov_len);
992 			break;
993 		}
994 
995 		if (unlikely(vq->iov[0].iov_len != sizeof(v_req))) {
996 			vq_err(vq, "Expecting virtio_scsi_cmd_req, got %zu"
997 				" bytes\n", vq->iov[0].iov_len);
998 			break;
999 		}
1000 		pr_debug("Calling __copy_from_user: vq->iov[0].iov_base: %p,"
1001 			" len: %zu\n", vq->iov[0].iov_base, sizeof(v_req));
1002 		ret = __copy_from_user(&v_req, vq->iov[0].iov_base,
1003 				sizeof(v_req));
1004 		if (unlikely(ret)) {
1005 			vq_err(vq, "Faulted on virtio_scsi_cmd_req\n");
1006 			break;
1007 		}
1008 
1009 		/* virtio-scsi spec requires byte 0 of the lun to be 1 */
1010 		if (unlikely(v_req.lun[0] != 1)) {
1011 			vhost_scsi_send_bad_target(vs, vq, head, out);
1012 			continue;
1013 		}
1014 
1015 		/* Extract the tpgt */
1016 		target = v_req.lun[1];
1017 		tpg = ACCESS_ONCE(vs_tpg[target]);
1018 
1019 		/* Target does not exist, fail the request */
1020 		if (unlikely(!tpg)) {
1021 			vhost_scsi_send_bad_target(vs, vq, head, out);
1022 			continue;
1023 		}
1024 
1025 		exp_data_len = 0;
1026 		for (i = 0; i < data_num; i++)
1027 			exp_data_len += vq->iov[data_first + i].iov_len;
1028 
1029 		cmd = vhost_scsi_get_tag(vq, tpg, &v_req,
1030 					 exp_data_len, data_direction);
1031 		if (IS_ERR(cmd)) {
1032 			vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1033 					PTR_ERR(cmd));
1034 			goto err_cmd;
1035 		}
1036 		pr_debug("Allocated tv_cmd: %p exp_data_len: %d, data_direction"
1037 			": %d\n", cmd, exp_data_len, data_direction);
1038 
1039 		cmd->tvc_vhost = vs;
1040 		cmd->tvc_vq = vq;
1041 		cmd->tvc_resp = vq->iov[out].iov_base;
1042 
1043 		/*
1044 		 * Copy in the recieved CDB descriptor into cmd->tvc_cdb
1045 		 * that will be used by tcm_vhost_new_cmd_map() and down into
1046 		 * target_setup_cmd_from_cdb()
1047 		 */
1048 		memcpy(cmd->tvc_cdb, v_req.cdb, TCM_VHOST_MAX_CDB_SIZE);
1049 		/*
1050 		 * Check that the recieved CDB size does not exceeded our
1051 		 * hardcoded max for tcm_vhost
1052 		 */
1053 		/* TODO what if cdb was too small for varlen cdb header? */
1054 		if (unlikely(scsi_command_size(cmd->tvc_cdb) >
1055 					TCM_VHOST_MAX_CDB_SIZE)) {
1056 			vq_err(vq, "Received SCSI CDB with command_size: %d that"
1057 				" exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1058 				scsi_command_size(cmd->tvc_cdb),
1059 				TCM_VHOST_MAX_CDB_SIZE);
1060 			goto err_free;
1061 		}
1062 		cmd->tvc_lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1063 
1064 		pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1065 			cmd->tvc_cdb[0], cmd->tvc_lun);
1066 
1067 		if (data_direction != DMA_NONE) {
1068 			ret = vhost_scsi_map_iov_to_sgl(cmd,
1069 					&vq->iov[data_first], data_num,
1070 					data_direction == DMA_FROM_DEVICE);
1071 			if (unlikely(ret)) {
1072 				vq_err(vq, "Failed to map iov to sgl\n");
1073 				goto err_free;
1074 			}
1075 		}
1076 
1077 		/*
1078 		 * Save the descriptor from vhost_get_vq_desc() to be used to
1079 		 * complete the virtio-scsi request in TCM callback context via
1080 		 * tcm_vhost_queue_data_in() and tcm_vhost_queue_status()
1081 		 */
1082 		cmd->tvc_vq_desc = head;
1083 		/*
1084 		 * Dispatch tv_cmd descriptor for cmwq execution in process
1085 		 * context provided by tcm_vhost_workqueue.  This also ensures
1086 		 * tv_cmd is executed on the same kworker CPU as this vhost
1087 		 * thread to gain positive L2 cache locality effects..
1088 		 */
1089 		INIT_WORK(&cmd->work, tcm_vhost_submission_work);
1090 		queue_work(tcm_vhost_workqueue, &cmd->work);
1091 	}
1092 
1093 	mutex_unlock(&vq->mutex);
1094 	return;
1095 
1096 err_free:
1097 	vhost_scsi_free_cmd(cmd);
1098 err_cmd:
1099 	vhost_scsi_send_bad_target(vs, vq, head, out);
1100 out:
1101 	mutex_unlock(&vq->mutex);
1102 }
1103 
1104 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1105 {
1106 	pr_debug("%s: The handling func for control queue.\n", __func__);
1107 }
1108 
1109 static void
1110 tcm_vhost_send_evt(struct vhost_scsi *vs,
1111 		   struct tcm_vhost_tpg *tpg,
1112 		   struct se_lun *lun,
1113 		   u32 event,
1114 		   u32 reason)
1115 {
1116 	struct tcm_vhost_evt *evt;
1117 
1118 	evt = tcm_vhost_allocate_evt(vs, event, reason);
1119 	if (!evt)
1120 		return;
1121 
1122 	if (tpg && lun) {
1123 		/* TODO: share lun setup code with virtio-scsi.ko */
1124 		/*
1125 		 * Note: evt->event is zeroed when we allocate it and
1126 		 * lun[4-7] need to be zero according to virtio-scsi spec.
1127 		 */
1128 		evt->event.lun[0] = 0x01;
1129 		evt->event.lun[1] = tpg->tport_tpgt & 0xFF;
1130 		if (lun->unpacked_lun >= 256)
1131 			evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1132 		evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1133 	}
1134 
1135 	llist_add(&evt->list, &vs->vs_event_list);
1136 	vhost_work_queue(&vs->dev, &vs->vs_event_work);
1137 }
1138 
1139 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1140 {
1141 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1142 						poll.work);
1143 	struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1144 
1145 	mutex_lock(&vq->mutex);
1146 	if (!vq->private_data)
1147 		goto out;
1148 
1149 	if (vs->vs_events_missed)
1150 		tcm_vhost_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1151 out:
1152 	mutex_unlock(&vq->mutex);
1153 }
1154 
1155 static void vhost_scsi_handle_kick(struct vhost_work *work)
1156 {
1157 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1158 						poll.work);
1159 	struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1160 
1161 	vhost_scsi_handle_vq(vs, vq);
1162 }
1163 
1164 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1165 {
1166 	vhost_poll_flush(&vs->vqs[index].vq.poll);
1167 }
1168 
1169 /* Callers must hold dev mutex */
1170 static void vhost_scsi_flush(struct vhost_scsi *vs)
1171 {
1172 	struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1173 	int i;
1174 
1175 	/* Init new inflight and remember the old inflight */
1176 	tcm_vhost_init_inflight(vs, old_inflight);
1177 
1178 	/*
1179 	 * The inflight->kref was initialized to 1. We decrement it here to
1180 	 * indicate the start of the flush operation so that it will reach 0
1181 	 * when all the reqs are finished.
1182 	 */
1183 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1184 		kref_put(&old_inflight[i]->kref, tcm_vhost_done_inflight);
1185 
1186 	/* Flush both the vhost poll and vhost work */
1187 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1188 		vhost_scsi_flush_vq(vs, i);
1189 	vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1190 	vhost_work_flush(&vs->dev, &vs->vs_event_work);
1191 
1192 	/* Wait for all reqs issued before the flush to be finished */
1193 	for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1194 		wait_for_completion(&old_inflight[i]->comp);
1195 }
1196 
1197 /*
1198  * Called from vhost_scsi_ioctl() context to walk the list of available
1199  * tcm_vhost_tpg with an active struct tcm_vhost_nexus
1200  *
1201  *  The lock nesting rule is:
1202  *    tcm_vhost_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1203  */
1204 static int
1205 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1206 			struct vhost_scsi_target *t)
1207 {
1208 	struct tcm_vhost_tport *tv_tport;
1209 	struct tcm_vhost_tpg *tpg;
1210 	struct tcm_vhost_tpg **vs_tpg;
1211 	struct vhost_virtqueue *vq;
1212 	int index, ret, i, len;
1213 	bool match = false;
1214 
1215 	mutex_lock(&tcm_vhost_mutex);
1216 	mutex_lock(&vs->dev.mutex);
1217 
1218 	/* Verify that ring has been setup correctly. */
1219 	for (index = 0; index < vs->dev.nvqs; ++index) {
1220 		/* Verify that ring has been setup correctly. */
1221 		if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1222 			ret = -EFAULT;
1223 			goto out;
1224 		}
1225 	}
1226 
1227 	len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1228 	vs_tpg = kzalloc(len, GFP_KERNEL);
1229 	if (!vs_tpg) {
1230 		ret = -ENOMEM;
1231 		goto out;
1232 	}
1233 	if (vs->vs_tpg)
1234 		memcpy(vs_tpg, vs->vs_tpg, len);
1235 
1236 	list_for_each_entry(tpg, &tcm_vhost_list, tv_tpg_list) {
1237 		mutex_lock(&tpg->tv_tpg_mutex);
1238 		if (!tpg->tpg_nexus) {
1239 			mutex_unlock(&tpg->tv_tpg_mutex);
1240 			continue;
1241 		}
1242 		if (tpg->tv_tpg_vhost_count != 0) {
1243 			mutex_unlock(&tpg->tv_tpg_mutex);
1244 			continue;
1245 		}
1246 		tv_tport = tpg->tport;
1247 
1248 		if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1249 			if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1250 				kfree(vs_tpg);
1251 				mutex_unlock(&tpg->tv_tpg_mutex);
1252 				ret = -EEXIST;
1253 				goto out;
1254 			}
1255 			tpg->tv_tpg_vhost_count++;
1256 			tpg->vhost_scsi = vs;
1257 			vs_tpg[tpg->tport_tpgt] = tpg;
1258 			smp_mb__after_atomic_inc();
1259 			match = true;
1260 		}
1261 		mutex_unlock(&tpg->tv_tpg_mutex);
1262 	}
1263 
1264 	if (match) {
1265 		memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1266 		       sizeof(vs->vs_vhost_wwpn));
1267 		for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1268 			vq = &vs->vqs[i].vq;
1269 			mutex_lock(&vq->mutex);
1270 			vq->private_data = vs_tpg;
1271 			vhost_init_used(vq);
1272 			mutex_unlock(&vq->mutex);
1273 		}
1274 		ret = 0;
1275 	} else {
1276 		ret = -EEXIST;
1277 	}
1278 
1279 	/*
1280 	 * Act as synchronize_rcu to make sure access to
1281 	 * old vs->vs_tpg is finished.
1282 	 */
1283 	vhost_scsi_flush(vs);
1284 	kfree(vs->vs_tpg);
1285 	vs->vs_tpg = vs_tpg;
1286 
1287 out:
1288 	mutex_unlock(&vs->dev.mutex);
1289 	mutex_unlock(&tcm_vhost_mutex);
1290 	return ret;
1291 }
1292 
1293 static int
1294 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1295 			  struct vhost_scsi_target *t)
1296 {
1297 	struct tcm_vhost_tport *tv_tport;
1298 	struct tcm_vhost_tpg *tpg;
1299 	struct vhost_virtqueue *vq;
1300 	bool match = false;
1301 	int index, ret, i;
1302 	u8 target;
1303 
1304 	mutex_lock(&tcm_vhost_mutex);
1305 	mutex_lock(&vs->dev.mutex);
1306 	/* Verify that ring has been setup correctly. */
1307 	for (index = 0; index < vs->dev.nvqs; ++index) {
1308 		if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1309 			ret = -EFAULT;
1310 			goto err_dev;
1311 		}
1312 	}
1313 
1314 	if (!vs->vs_tpg) {
1315 		ret = 0;
1316 		goto err_dev;
1317 	}
1318 
1319 	for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1320 		target = i;
1321 		tpg = vs->vs_tpg[target];
1322 		if (!tpg)
1323 			continue;
1324 
1325 		mutex_lock(&tpg->tv_tpg_mutex);
1326 		tv_tport = tpg->tport;
1327 		if (!tv_tport) {
1328 			ret = -ENODEV;
1329 			goto err_tpg;
1330 		}
1331 
1332 		if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1333 			pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1334 				" does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1335 				tv_tport->tport_name, tpg->tport_tpgt,
1336 				t->vhost_wwpn, t->vhost_tpgt);
1337 			ret = -EINVAL;
1338 			goto err_tpg;
1339 		}
1340 		tpg->tv_tpg_vhost_count--;
1341 		tpg->vhost_scsi = NULL;
1342 		vs->vs_tpg[target] = NULL;
1343 		match = true;
1344 		mutex_unlock(&tpg->tv_tpg_mutex);
1345 	}
1346 	if (match) {
1347 		for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1348 			vq = &vs->vqs[i].vq;
1349 			mutex_lock(&vq->mutex);
1350 			vq->private_data = NULL;
1351 			mutex_unlock(&vq->mutex);
1352 		}
1353 	}
1354 	/*
1355 	 * Act as synchronize_rcu to make sure access to
1356 	 * old vs->vs_tpg is finished.
1357 	 */
1358 	vhost_scsi_flush(vs);
1359 	kfree(vs->vs_tpg);
1360 	vs->vs_tpg = NULL;
1361 	WARN_ON(vs->vs_events_nr);
1362 	mutex_unlock(&vs->dev.mutex);
1363 	mutex_unlock(&tcm_vhost_mutex);
1364 	return 0;
1365 
1366 err_tpg:
1367 	mutex_unlock(&tpg->tv_tpg_mutex);
1368 err_dev:
1369 	mutex_unlock(&vs->dev.mutex);
1370 	mutex_unlock(&tcm_vhost_mutex);
1371 	return ret;
1372 }
1373 
1374 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1375 {
1376 	if (features & ~VHOST_SCSI_FEATURES)
1377 		return -EOPNOTSUPP;
1378 
1379 	mutex_lock(&vs->dev.mutex);
1380 	if ((features & (1 << VHOST_F_LOG_ALL)) &&
1381 	    !vhost_log_access_ok(&vs->dev)) {
1382 		mutex_unlock(&vs->dev.mutex);
1383 		return -EFAULT;
1384 	}
1385 	vs->dev.acked_features = features;
1386 	smp_wmb();
1387 	vhost_scsi_flush(vs);
1388 	mutex_unlock(&vs->dev.mutex);
1389 	return 0;
1390 }
1391 
1392 static void vhost_scsi_free(struct vhost_scsi *vs)
1393 {
1394 	if (is_vmalloc_addr(vs))
1395 		vfree(vs);
1396 	else
1397 		kfree(vs);
1398 }
1399 
1400 static int vhost_scsi_open(struct inode *inode, struct file *f)
1401 {
1402 	struct vhost_scsi *vs;
1403 	struct vhost_virtqueue **vqs;
1404 	int r = -ENOMEM, i;
1405 
1406 	vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
1407 	if (!vs) {
1408 		vs = vzalloc(sizeof(*vs));
1409 		if (!vs)
1410 			goto err_vs;
1411 	}
1412 
1413 	vqs = kmalloc(VHOST_SCSI_MAX_VQ * sizeof(*vqs), GFP_KERNEL);
1414 	if (!vqs)
1415 		goto err_vqs;
1416 
1417 	vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1418 	vhost_work_init(&vs->vs_event_work, tcm_vhost_evt_work);
1419 
1420 	vs->vs_events_nr = 0;
1421 	vs->vs_events_missed = false;
1422 
1423 	vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1424 	vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1425 	vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1426 	vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1427 	for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1428 		vqs[i] = &vs->vqs[i].vq;
1429 		vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1430 	}
1431 	vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ);
1432 
1433 	tcm_vhost_init_inflight(vs, NULL);
1434 
1435 	f->private_data = vs;
1436 	return 0;
1437 
1438 err_vqs:
1439 	vhost_scsi_free(vs);
1440 err_vs:
1441 	return r;
1442 }
1443 
1444 static int vhost_scsi_release(struct inode *inode, struct file *f)
1445 {
1446 	struct vhost_scsi *vs = f->private_data;
1447 	struct vhost_scsi_target t;
1448 
1449 	mutex_lock(&vs->dev.mutex);
1450 	memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1451 	mutex_unlock(&vs->dev.mutex);
1452 	vhost_scsi_clear_endpoint(vs, &t);
1453 	vhost_dev_stop(&vs->dev);
1454 	vhost_dev_cleanup(&vs->dev, false);
1455 	/* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1456 	vhost_scsi_flush(vs);
1457 	kfree(vs->dev.vqs);
1458 	vhost_scsi_free(vs);
1459 	return 0;
1460 }
1461 
1462 static long
1463 vhost_scsi_ioctl(struct file *f,
1464 		 unsigned int ioctl,
1465 		 unsigned long arg)
1466 {
1467 	struct vhost_scsi *vs = f->private_data;
1468 	struct vhost_scsi_target backend;
1469 	void __user *argp = (void __user *)arg;
1470 	u64 __user *featurep = argp;
1471 	u32 __user *eventsp = argp;
1472 	u32 events_missed;
1473 	u64 features;
1474 	int r, abi_version = VHOST_SCSI_ABI_VERSION;
1475 	struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1476 
1477 	switch (ioctl) {
1478 	case VHOST_SCSI_SET_ENDPOINT:
1479 		if (copy_from_user(&backend, argp, sizeof backend))
1480 			return -EFAULT;
1481 		if (backend.reserved != 0)
1482 			return -EOPNOTSUPP;
1483 
1484 		return vhost_scsi_set_endpoint(vs, &backend);
1485 	case VHOST_SCSI_CLEAR_ENDPOINT:
1486 		if (copy_from_user(&backend, argp, sizeof backend))
1487 			return -EFAULT;
1488 		if (backend.reserved != 0)
1489 			return -EOPNOTSUPP;
1490 
1491 		return vhost_scsi_clear_endpoint(vs, &backend);
1492 	case VHOST_SCSI_GET_ABI_VERSION:
1493 		if (copy_to_user(argp, &abi_version, sizeof abi_version))
1494 			return -EFAULT;
1495 		return 0;
1496 	case VHOST_SCSI_SET_EVENTS_MISSED:
1497 		if (get_user(events_missed, eventsp))
1498 			return -EFAULT;
1499 		mutex_lock(&vq->mutex);
1500 		vs->vs_events_missed = events_missed;
1501 		mutex_unlock(&vq->mutex);
1502 		return 0;
1503 	case VHOST_SCSI_GET_EVENTS_MISSED:
1504 		mutex_lock(&vq->mutex);
1505 		events_missed = vs->vs_events_missed;
1506 		mutex_unlock(&vq->mutex);
1507 		if (put_user(events_missed, eventsp))
1508 			return -EFAULT;
1509 		return 0;
1510 	case VHOST_GET_FEATURES:
1511 		features = VHOST_SCSI_FEATURES;
1512 		if (copy_to_user(featurep, &features, sizeof features))
1513 			return -EFAULT;
1514 		return 0;
1515 	case VHOST_SET_FEATURES:
1516 		if (copy_from_user(&features, featurep, sizeof features))
1517 			return -EFAULT;
1518 		return vhost_scsi_set_features(vs, features);
1519 	default:
1520 		mutex_lock(&vs->dev.mutex);
1521 		r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1522 		/* TODO: flush backend after dev ioctl. */
1523 		if (r == -ENOIOCTLCMD)
1524 			r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1525 		mutex_unlock(&vs->dev.mutex);
1526 		return r;
1527 	}
1528 }
1529 
1530 #ifdef CONFIG_COMPAT
1531 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1532 				unsigned long arg)
1533 {
1534 	return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1535 }
1536 #endif
1537 
1538 static const struct file_operations vhost_scsi_fops = {
1539 	.owner          = THIS_MODULE,
1540 	.release        = vhost_scsi_release,
1541 	.unlocked_ioctl = vhost_scsi_ioctl,
1542 #ifdef CONFIG_COMPAT
1543 	.compat_ioctl	= vhost_scsi_compat_ioctl,
1544 #endif
1545 	.open           = vhost_scsi_open,
1546 	.llseek		= noop_llseek,
1547 };
1548 
1549 static struct miscdevice vhost_scsi_misc = {
1550 	MISC_DYNAMIC_MINOR,
1551 	"vhost-scsi",
1552 	&vhost_scsi_fops,
1553 };
1554 
1555 static int __init vhost_scsi_register(void)
1556 {
1557 	return misc_register(&vhost_scsi_misc);
1558 }
1559 
1560 static int vhost_scsi_deregister(void)
1561 {
1562 	return misc_deregister(&vhost_scsi_misc);
1563 }
1564 
1565 static char *tcm_vhost_dump_proto_id(struct tcm_vhost_tport *tport)
1566 {
1567 	switch (tport->tport_proto_id) {
1568 	case SCSI_PROTOCOL_SAS:
1569 		return "SAS";
1570 	case SCSI_PROTOCOL_FCP:
1571 		return "FCP";
1572 	case SCSI_PROTOCOL_ISCSI:
1573 		return "iSCSI";
1574 	default:
1575 		break;
1576 	}
1577 
1578 	return "Unknown";
1579 }
1580 
1581 static void
1582 tcm_vhost_do_plug(struct tcm_vhost_tpg *tpg,
1583 		  struct se_lun *lun, bool plug)
1584 {
1585 
1586 	struct vhost_scsi *vs = tpg->vhost_scsi;
1587 	struct vhost_virtqueue *vq;
1588 	u32 reason;
1589 
1590 	if (!vs)
1591 		return;
1592 
1593 	mutex_lock(&vs->dev.mutex);
1594 	if (!vhost_has_feature(&vs->dev, VIRTIO_SCSI_F_HOTPLUG)) {
1595 		mutex_unlock(&vs->dev.mutex);
1596 		return;
1597 	}
1598 
1599 	if (plug)
1600 		reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1601 	else
1602 		reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1603 
1604 	vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1605 	mutex_lock(&vq->mutex);
1606 	tcm_vhost_send_evt(vs, tpg, lun,
1607 			VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1608 	mutex_unlock(&vq->mutex);
1609 	mutex_unlock(&vs->dev.mutex);
1610 }
1611 
1612 static void tcm_vhost_hotplug(struct tcm_vhost_tpg *tpg, struct se_lun *lun)
1613 {
1614 	tcm_vhost_do_plug(tpg, lun, true);
1615 }
1616 
1617 static void tcm_vhost_hotunplug(struct tcm_vhost_tpg *tpg, struct se_lun *lun)
1618 {
1619 	tcm_vhost_do_plug(tpg, lun, false);
1620 }
1621 
1622 static int tcm_vhost_port_link(struct se_portal_group *se_tpg,
1623 			       struct se_lun *lun)
1624 {
1625 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
1626 				struct tcm_vhost_tpg, se_tpg);
1627 
1628 	mutex_lock(&tcm_vhost_mutex);
1629 
1630 	mutex_lock(&tpg->tv_tpg_mutex);
1631 	tpg->tv_tpg_port_count++;
1632 	mutex_unlock(&tpg->tv_tpg_mutex);
1633 
1634 	tcm_vhost_hotplug(tpg, lun);
1635 
1636 	mutex_unlock(&tcm_vhost_mutex);
1637 
1638 	return 0;
1639 }
1640 
1641 static void tcm_vhost_port_unlink(struct se_portal_group *se_tpg,
1642 				  struct se_lun *lun)
1643 {
1644 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
1645 				struct tcm_vhost_tpg, se_tpg);
1646 
1647 	mutex_lock(&tcm_vhost_mutex);
1648 
1649 	mutex_lock(&tpg->tv_tpg_mutex);
1650 	tpg->tv_tpg_port_count--;
1651 	mutex_unlock(&tpg->tv_tpg_mutex);
1652 
1653 	tcm_vhost_hotunplug(tpg, lun);
1654 
1655 	mutex_unlock(&tcm_vhost_mutex);
1656 }
1657 
1658 static struct se_node_acl *
1659 tcm_vhost_make_nodeacl(struct se_portal_group *se_tpg,
1660 		       struct config_group *group,
1661 		       const char *name)
1662 {
1663 	struct se_node_acl *se_nacl, *se_nacl_new;
1664 	struct tcm_vhost_nacl *nacl;
1665 	u64 wwpn = 0;
1666 	u32 nexus_depth;
1667 
1668 	/* tcm_vhost_parse_wwn(name, &wwpn, 1) < 0)
1669 		return ERR_PTR(-EINVAL); */
1670 	se_nacl_new = tcm_vhost_alloc_fabric_acl(se_tpg);
1671 	if (!se_nacl_new)
1672 		return ERR_PTR(-ENOMEM);
1673 
1674 	nexus_depth = 1;
1675 	/*
1676 	 * se_nacl_new may be released by core_tpg_add_initiator_node_acl()
1677 	 * when converting a NodeACL from demo mode -> explict
1678 	 */
1679 	se_nacl = core_tpg_add_initiator_node_acl(se_tpg, se_nacl_new,
1680 				name, nexus_depth);
1681 	if (IS_ERR(se_nacl)) {
1682 		tcm_vhost_release_fabric_acl(se_tpg, se_nacl_new);
1683 		return se_nacl;
1684 	}
1685 	/*
1686 	 * Locate our struct tcm_vhost_nacl and set the FC Nport WWPN
1687 	 */
1688 	nacl = container_of(se_nacl, struct tcm_vhost_nacl, se_node_acl);
1689 	nacl->iport_wwpn = wwpn;
1690 
1691 	return se_nacl;
1692 }
1693 
1694 static void tcm_vhost_drop_nodeacl(struct se_node_acl *se_acl)
1695 {
1696 	struct tcm_vhost_nacl *nacl = container_of(se_acl,
1697 				struct tcm_vhost_nacl, se_node_acl);
1698 	core_tpg_del_initiator_node_acl(se_acl->se_tpg, se_acl, 1);
1699 	kfree(nacl);
1700 }
1701 
1702 static void tcm_vhost_free_cmd_map_res(struct tcm_vhost_nexus *nexus,
1703 				       struct se_session *se_sess)
1704 {
1705 	struct tcm_vhost_cmd *tv_cmd;
1706 	unsigned int i;
1707 
1708 	if (!se_sess->sess_cmd_map)
1709 		return;
1710 
1711 	for (i = 0; i < TCM_VHOST_DEFAULT_TAGS; i++) {
1712 		tv_cmd = &((struct tcm_vhost_cmd *)se_sess->sess_cmd_map)[i];
1713 
1714 		kfree(tv_cmd->tvc_sgl);
1715 		kfree(tv_cmd->tvc_upages);
1716 	}
1717 }
1718 
1719 static int tcm_vhost_make_nexus(struct tcm_vhost_tpg *tpg,
1720 				const char *name)
1721 {
1722 	struct se_portal_group *se_tpg;
1723 	struct se_session *se_sess;
1724 	struct tcm_vhost_nexus *tv_nexus;
1725 	struct tcm_vhost_cmd *tv_cmd;
1726 	unsigned int i;
1727 
1728 	mutex_lock(&tpg->tv_tpg_mutex);
1729 	if (tpg->tpg_nexus) {
1730 		mutex_unlock(&tpg->tv_tpg_mutex);
1731 		pr_debug("tpg->tpg_nexus already exists\n");
1732 		return -EEXIST;
1733 	}
1734 	se_tpg = &tpg->se_tpg;
1735 
1736 	tv_nexus = kzalloc(sizeof(struct tcm_vhost_nexus), GFP_KERNEL);
1737 	if (!tv_nexus) {
1738 		mutex_unlock(&tpg->tv_tpg_mutex);
1739 		pr_err("Unable to allocate struct tcm_vhost_nexus\n");
1740 		return -ENOMEM;
1741 	}
1742 	/*
1743 	 *  Initialize the struct se_session pointer and setup tagpool
1744 	 *  for struct tcm_vhost_cmd descriptors
1745 	 */
1746 	tv_nexus->tvn_se_sess = transport_init_session_tags(
1747 					TCM_VHOST_DEFAULT_TAGS,
1748 					sizeof(struct tcm_vhost_cmd),
1749 					TARGET_PROT_NORMAL);
1750 	if (IS_ERR(tv_nexus->tvn_se_sess)) {
1751 		mutex_unlock(&tpg->tv_tpg_mutex);
1752 		kfree(tv_nexus);
1753 		return -ENOMEM;
1754 	}
1755 	se_sess = tv_nexus->tvn_se_sess;
1756 	for (i = 0; i < TCM_VHOST_DEFAULT_TAGS; i++) {
1757 		tv_cmd = &((struct tcm_vhost_cmd *)se_sess->sess_cmd_map)[i];
1758 
1759 		tv_cmd->tvc_sgl = kzalloc(sizeof(struct scatterlist) *
1760 					TCM_VHOST_PREALLOC_SGLS, GFP_KERNEL);
1761 		if (!tv_cmd->tvc_sgl) {
1762 			mutex_unlock(&tpg->tv_tpg_mutex);
1763 			pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1764 			goto out;
1765 		}
1766 
1767 		tv_cmd->tvc_upages = kzalloc(sizeof(struct page *) *
1768 					TCM_VHOST_PREALLOC_PAGES, GFP_KERNEL);
1769 		if (!tv_cmd->tvc_upages) {
1770 			mutex_unlock(&tpg->tv_tpg_mutex);
1771 			pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1772 			goto out;
1773 		}
1774 	}
1775 	/*
1776 	 * Since we are running in 'demo mode' this call with generate a
1777 	 * struct se_node_acl for the tcm_vhost struct se_portal_group with
1778 	 * the SCSI Initiator port name of the passed configfs group 'name'.
1779 	 */
1780 	tv_nexus->tvn_se_sess->se_node_acl = core_tpg_check_initiator_node_acl(
1781 				se_tpg, (unsigned char *)name);
1782 	if (!tv_nexus->tvn_se_sess->se_node_acl) {
1783 		mutex_unlock(&tpg->tv_tpg_mutex);
1784 		pr_debug("core_tpg_check_initiator_node_acl() failed"
1785 				" for %s\n", name);
1786 		goto out;
1787 	}
1788 	/*
1789 	 * Now register the TCM vhost virtual I_T Nexus as active with the
1790 	 * call to __transport_register_session()
1791 	 */
1792 	__transport_register_session(se_tpg, tv_nexus->tvn_se_sess->se_node_acl,
1793 			tv_nexus->tvn_se_sess, tv_nexus);
1794 	tpg->tpg_nexus = tv_nexus;
1795 
1796 	mutex_unlock(&tpg->tv_tpg_mutex);
1797 	return 0;
1798 
1799 out:
1800 	tcm_vhost_free_cmd_map_res(tv_nexus, se_sess);
1801 	transport_free_session(se_sess);
1802 	kfree(tv_nexus);
1803 	return -ENOMEM;
1804 }
1805 
1806 static int tcm_vhost_drop_nexus(struct tcm_vhost_tpg *tpg)
1807 {
1808 	struct se_session *se_sess;
1809 	struct tcm_vhost_nexus *tv_nexus;
1810 
1811 	mutex_lock(&tpg->tv_tpg_mutex);
1812 	tv_nexus = tpg->tpg_nexus;
1813 	if (!tv_nexus) {
1814 		mutex_unlock(&tpg->tv_tpg_mutex);
1815 		return -ENODEV;
1816 	}
1817 
1818 	se_sess = tv_nexus->tvn_se_sess;
1819 	if (!se_sess) {
1820 		mutex_unlock(&tpg->tv_tpg_mutex);
1821 		return -ENODEV;
1822 	}
1823 
1824 	if (tpg->tv_tpg_port_count != 0) {
1825 		mutex_unlock(&tpg->tv_tpg_mutex);
1826 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1827 			" active TPG port count: %d\n",
1828 			tpg->tv_tpg_port_count);
1829 		return -EBUSY;
1830 	}
1831 
1832 	if (tpg->tv_tpg_vhost_count != 0) {
1833 		mutex_unlock(&tpg->tv_tpg_mutex);
1834 		pr_err("Unable to remove TCM_vhost I_T Nexus with"
1835 			" active TPG vhost count: %d\n",
1836 			tpg->tv_tpg_vhost_count);
1837 		return -EBUSY;
1838 	}
1839 
1840 	pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
1841 		" %s Initiator Port: %s\n", tcm_vhost_dump_proto_id(tpg->tport),
1842 		tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1843 
1844 	tcm_vhost_free_cmd_map_res(tv_nexus, se_sess);
1845 	/*
1846 	 * Release the SCSI I_T Nexus to the emulated vhost Target Port
1847 	 */
1848 	transport_deregister_session(tv_nexus->tvn_se_sess);
1849 	tpg->tpg_nexus = NULL;
1850 	mutex_unlock(&tpg->tv_tpg_mutex);
1851 
1852 	kfree(tv_nexus);
1853 	return 0;
1854 }
1855 
1856 static ssize_t tcm_vhost_tpg_show_nexus(struct se_portal_group *se_tpg,
1857 					char *page)
1858 {
1859 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
1860 				struct tcm_vhost_tpg, se_tpg);
1861 	struct tcm_vhost_nexus *tv_nexus;
1862 	ssize_t ret;
1863 
1864 	mutex_lock(&tpg->tv_tpg_mutex);
1865 	tv_nexus = tpg->tpg_nexus;
1866 	if (!tv_nexus) {
1867 		mutex_unlock(&tpg->tv_tpg_mutex);
1868 		return -ENODEV;
1869 	}
1870 	ret = snprintf(page, PAGE_SIZE, "%s\n",
1871 			tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
1872 	mutex_unlock(&tpg->tv_tpg_mutex);
1873 
1874 	return ret;
1875 }
1876 
1877 static ssize_t tcm_vhost_tpg_store_nexus(struct se_portal_group *se_tpg,
1878 					 const char *page,
1879 					 size_t count)
1880 {
1881 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
1882 				struct tcm_vhost_tpg, se_tpg);
1883 	struct tcm_vhost_tport *tport_wwn = tpg->tport;
1884 	unsigned char i_port[TCM_VHOST_NAMELEN], *ptr, *port_ptr;
1885 	int ret;
1886 	/*
1887 	 * Shutdown the active I_T nexus if 'NULL' is passed..
1888 	 */
1889 	if (!strncmp(page, "NULL", 4)) {
1890 		ret = tcm_vhost_drop_nexus(tpg);
1891 		return (!ret) ? count : ret;
1892 	}
1893 	/*
1894 	 * Otherwise make sure the passed virtual Initiator port WWN matches
1895 	 * the fabric protocol_id set in tcm_vhost_make_tport(), and call
1896 	 * tcm_vhost_make_nexus().
1897 	 */
1898 	if (strlen(page) >= TCM_VHOST_NAMELEN) {
1899 		pr_err("Emulated NAA Sas Address: %s, exceeds"
1900 				" max: %d\n", page, TCM_VHOST_NAMELEN);
1901 		return -EINVAL;
1902 	}
1903 	snprintf(&i_port[0], TCM_VHOST_NAMELEN, "%s", page);
1904 
1905 	ptr = strstr(i_port, "naa.");
1906 	if (ptr) {
1907 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
1908 			pr_err("Passed SAS Initiator Port %s does not"
1909 				" match target port protoid: %s\n", i_port,
1910 				tcm_vhost_dump_proto_id(tport_wwn));
1911 			return -EINVAL;
1912 		}
1913 		port_ptr = &i_port[0];
1914 		goto check_newline;
1915 	}
1916 	ptr = strstr(i_port, "fc.");
1917 	if (ptr) {
1918 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
1919 			pr_err("Passed FCP Initiator Port %s does not"
1920 				" match target port protoid: %s\n", i_port,
1921 				tcm_vhost_dump_proto_id(tport_wwn));
1922 			return -EINVAL;
1923 		}
1924 		port_ptr = &i_port[3]; /* Skip over "fc." */
1925 		goto check_newline;
1926 	}
1927 	ptr = strstr(i_port, "iqn.");
1928 	if (ptr) {
1929 		if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
1930 			pr_err("Passed iSCSI Initiator Port %s does not"
1931 				" match target port protoid: %s\n", i_port,
1932 				tcm_vhost_dump_proto_id(tport_wwn));
1933 			return -EINVAL;
1934 		}
1935 		port_ptr = &i_port[0];
1936 		goto check_newline;
1937 	}
1938 	pr_err("Unable to locate prefix for emulated Initiator Port:"
1939 			" %s\n", i_port);
1940 	return -EINVAL;
1941 	/*
1942 	 * Clear any trailing newline for the NAA WWN
1943 	 */
1944 check_newline:
1945 	if (i_port[strlen(i_port)-1] == '\n')
1946 		i_port[strlen(i_port)-1] = '\0';
1947 
1948 	ret = tcm_vhost_make_nexus(tpg, port_ptr);
1949 	if (ret < 0)
1950 		return ret;
1951 
1952 	return count;
1953 }
1954 
1955 TF_TPG_BASE_ATTR(tcm_vhost, nexus, S_IRUGO | S_IWUSR);
1956 
1957 static struct configfs_attribute *tcm_vhost_tpg_attrs[] = {
1958 	&tcm_vhost_tpg_nexus.attr,
1959 	NULL,
1960 };
1961 
1962 static struct se_portal_group *
1963 tcm_vhost_make_tpg(struct se_wwn *wwn,
1964 		   struct config_group *group,
1965 		   const char *name)
1966 {
1967 	struct tcm_vhost_tport *tport = container_of(wwn,
1968 			struct tcm_vhost_tport, tport_wwn);
1969 
1970 	struct tcm_vhost_tpg *tpg;
1971 	unsigned long tpgt;
1972 	int ret;
1973 
1974 	if (strstr(name, "tpgt_") != name)
1975 		return ERR_PTR(-EINVAL);
1976 	if (kstrtoul(name + 5, 10, &tpgt) || tpgt > UINT_MAX)
1977 		return ERR_PTR(-EINVAL);
1978 
1979 	tpg = kzalloc(sizeof(struct tcm_vhost_tpg), GFP_KERNEL);
1980 	if (!tpg) {
1981 		pr_err("Unable to allocate struct tcm_vhost_tpg");
1982 		return ERR_PTR(-ENOMEM);
1983 	}
1984 	mutex_init(&tpg->tv_tpg_mutex);
1985 	INIT_LIST_HEAD(&tpg->tv_tpg_list);
1986 	tpg->tport = tport;
1987 	tpg->tport_tpgt = tpgt;
1988 
1989 	ret = core_tpg_register(&tcm_vhost_fabric_configfs->tf_ops, wwn,
1990 				&tpg->se_tpg, tpg, TRANSPORT_TPG_TYPE_NORMAL);
1991 	if (ret < 0) {
1992 		kfree(tpg);
1993 		return NULL;
1994 	}
1995 	mutex_lock(&tcm_vhost_mutex);
1996 	list_add_tail(&tpg->tv_tpg_list, &tcm_vhost_list);
1997 	mutex_unlock(&tcm_vhost_mutex);
1998 
1999 	return &tpg->se_tpg;
2000 }
2001 
2002 static void tcm_vhost_drop_tpg(struct se_portal_group *se_tpg)
2003 {
2004 	struct tcm_vhost_tpg *tpg = container_of(se_tpg,
2005 				struct tcm_vhost_tpg, se_tpg);
2006 
2007 	mutex_lock(&tcm_vhost_mutex);
2008 	list_del(&tpg->tv_tpg_list);
2009 	mutex_unlock(&tcm_vhost_mutex);
2010 	/*
2011 	 * Release the virtual I_T Nexus for this vhost TPG
2012 	 */
2013 	tcm_vhost_drop_nexus(tpg);
2014 	/*
2015 	 * Deregister the se_tpg from TCM..
2016 	 */
2017 	core_tpg_deregister(se_tpg);
2018 	kfree(tpg);
2019 }
2020 
2021 static struct se_wwn *
2022 tcm_vhost_make_tport(struct target_fabric_configfs *tf,
2023 		     struct config_group *group,
2024 		     const char *name)
2025 {
2026 	struct tcm_vhost_tport *tport;
2027 	char *ptr;
2028 	u64 wwpn = 0;
2029 	int off = 0;
2030 
2031 	/* if (tcm_vhost_parse_wwn(name, &wwpn, 1) < 0)
2032 		return ERR_PTR(-EINVAL); */
2033 
2034 	tport = kzalloc(sizeof(struct tcm_vhost_tport), GFP_KERNEL);
2035 	if (!tport) {
2036 		pr_err("Unable to allocate struct tcm_vhost_tport");
2037 		return ERR_PTR(-ENOMEM);
2038 	}
2039 	tport->tport_wwpn = wwpn;
2040 	/*
2041 	 * Determine the emulated Protocol Identifier and Target Port Name
2042 	 * based on the incoming configfs directory name.
2043 	 */
2044 	ptr = strstr(name, "naa.");
2045 	if (ptr) {
2046 		tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2047 		goto check_len;
2048 	}
2049 	ptr = strstr(name, "fc.");
2050 	if (ptr) {
2051 		tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2052 		off = 3; /* Skip over "fc." */
2053 		goto check_len;
2054 	}
2055 	ptr = strstr(name, "iqn.");
2056 	if (ptr) {
2057 		tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2058 		goto check_len;
2059 	}
2060 
2061 	pr_err("Unable to locate prefix for emulated Target Port:"
2062 			" %s\n", name);
2063 	kfree(tport);
2064 	return ERR_PTR(-EINVAL);
2065 
2066 check_len:
2067 	if (strlen(name) >= TCM_VHOST_NAMELEN) {
2068 		pr_err("Emulated %s Address: %s, exceeds"
2069 			" max: %d\n", name, tcm_vhost_dump_proto_id(tport),
2070 			TCM_VHOST_NAMELEN);
2071 		kfree(tport);
2072 		return ERR_PTR(-EINVAL);
2073 	}
2074 	snprintf(&tport->tport_name[0], TCM_VHOST_NAMELEN, "%s", &name[off]);
2075 
2076 	pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2077 		" %s Address: %s\n", tcm_vhost_dump_proto_id(tport), name);
2078 
2079 	return &tport->tport_wwn;
2080 }
2081 
2082 static void tcm_vhost_drop_tport(struct se_wwn *wwn)
2083 {
2084 	struct tcm_vhost_tport *tport = container_of(wwn,
2085 				struct tcm_vhost_tport, tport_wwn);
2086 
2087 	pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2088 		" %s Address: %s\n", tcm_vhost_dump_proto_id(tport),
2089 		tport->tport_name);
2090 
2091 	kfree(tport);
2092 }
2093 
2094 static ssize_t
2095 tcm_vhost_wwn_show_attr_version(struct target_fabric_configfs *tf,
2096 				char *page)
2097 {
2098 	return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2099 		"on "UTS_RELEASE"\n", TCM_VHOST_VERSION, utsname()->sysname,
2100 		utsname()->machine);
2101 }
2102 
2103 TF_WWN_ATTR_RO(tcm_vhost, version);
2104 
2105 static struct configfs_attribute *tcm_vhost_wwn_attrs[] = {
2106 	&tcm_vhost_wwn_version.attr,
2107 	NULL,
2108 };
2109 
2110 static struct target_core_fabric_ops tcm_vhost_ops = {
2111 	.get_fabric_name		= tcm_vhost_get_fabric_name,
2112 	.get_fabric_proto_ident		= tcm_vhost_get_fabric_proto_ident,
2113 	.tpg_get_wwn			= tcm_vhost_get_fabric_wwn,
2114 	.tpg_get_tag			= tcm_vhost_get_tag,
2115 	.tpg_get_default_depth		= tcm_vhost_get_default_depth,
2116 	.tpg_get_pr_transport_id	= tcm_vhost_get_pr_transport_id,
2117 	.tpg_get_pr_transport_id_len	= tcm_vhost_get_pr_transport_id_len,
2118 	.tpg_parse_pr_out_transport_id	= tcm_vhost_parse_pr_out_transport_id,
2119 	.tpg_check_demo_mode		= tcm_vhost_check_true,
2120 	.tpg_check_demo_mode_cache	= tcm_vhost_check_true,
2121 	.tpg_check_demo_mode_write_protect = tcm_vhost_check_false,
2122 	.tpg_check_prod_mode_write_protect = tcm_vhost_check_false,
2123 	.tpg_alloc_fabric_acl		= tcm_vhost_alloc_fabric_acl,
2124 	.tpg_release_fabric_acl		= tcm_vhost_release_fabric_acl,
2125 	.tpg_get_inst_index		= tcm_vhost_tpg_get_inst_index,
2126 	.release_cmd			= tcm_vhost_release_cmd,
2127 	.check_stop_free		= vhost_scsi_check_stop_free,
2128 	.shutdown_session		= tcm_vhost_shutdown_session,
2129 	.close_session			= tcm_vhost_close_session,
2130 	.sess_get_index			= tcm_vhost_sess_get_index,
2131 	.sess_get_initiator_sid		= NULL,
2132 	.write_pending			= tcm_vhost_write_pending,
2133 	.write_pending_status		= tcm_vhost_write_pending_status,
2134 	.set_default_node_attributes	= tcm_vhost_set_default_node_attrs,
2135 	.get_task_tag			= tcm_vhost_get_task_tag,
2136 	.get_cmd_state			= tcm_vhost_get_cmd_state,
2137 	.queue_data_in			= tcm_vhost_queue_data_in,
2138 	.queue_status			= tcm_vhost_queue_status,
2139 	.queue_tm_rsp			= tcm_vhost_queue_tm_rsp,
2140 	.aborted_task			= tcm_vhost_aborted_task,
2141 	/*
2142 	 * Setup callers for generic logic in target_core_fabric_configfs.c
2143 	 */
2144 	.fabric_make_wwn		= tcm_vhost_make_tport,
2145 	.fabric_drop_wwn		= tcm_vhost_drop_tport,
2146 	.fabric_make_tpg		= tcm_vhost_make_tpg,
2147 	.fabric_drop_tpg		= tcm_vhost_drop_tpg,
2148 	.fabric_post_link		= tcm_vhost_port_link,
2149 	.fabric_pre_unlink		= tcm_vhost_port_unlink,
2150 	.fabric_make_np			= NULL,
2151 	.fabric_drop_np			= NULL,
2152 	.fabric_make_nodeacl		= tcm_vhost_make_nodeacl,
2153 	.fabric_drop_nodeacl		= tcm_vhost_drop_nodeacl,
2154 };
2155 
2156 static int tcm_vhost_register_configfs(void)
2157 {
2158 	struct target_fabric_configfs *fabric;
2159 	int ret;
2160 
2161 	pr_debug("TCM_VHOST fabric module %s on %s/%s"
2162 		" on "UTS_RELEASE"\n", TCM_VHOST_VERSION, utsname()->sysname,
2163 		utsname()->machine);
2164 	/*
2165 	 * Register the top level struct config_item_type with TCM core
2166 	 */
2167 	fabric = target_fabric_configfs_init(THIS_MODULE, "vhost");
2168 	if (IS_ERR(fabric)) {
2169 		pr_err("target_fabric_configfs_init() failed\n");
2170 		return PTR_ERR(fabric);
2171 	}
2172 	/*
2173 	 * Setup fabric->tf_ops from our local tcm_vhost_ops
2174 	 */
2175 	fabric->tf_ops = tcm_vhost_ops;
2176 	/*
2177 	 * Setup default attribute lists for various fabric->tf_cit_tmpl
2178 	 */
2179 	fabric->tf_cit_tmpl.tfc_wwn_cit.ct_attrs = tcm_vhost_wwn_attrs;
2180 	fabric->tf_cit_tmpl.tfc_tpg_base_cit.ct_attrs = tcm_vhost_tpg_attrs;
2181 	fabric->tf_cit_tmpl.tfc_tpg_attrib_cit.ct_attrs = NULL;
2182 	fabric->tf_cit_tmpl.tfc_tpg_param_cit.ct_attrs = NULL;
2183 	fabric->tf_cit_tmpl.tfc_tpg_np_base_cit.ct_attrs = NULL;
2184 	fabric->tf_cit_tmpl.tfc_tpg_nacl_base_cit.ct_attrs = NULL;
2185 	fabric->tf_cit_tmpl.tfc_tpg_nacl_attrib_cit.ct_attrs = NULL;
2186 	fabric->tf_cit_tmpl.tfc_tpg_nacl_auth_cit.ct_attrs = NULL;
2187 	fabric->tf_cit_tmpl.tfc_tpg_nacl_param_cit.ct_attrs = NULL;
2188 	/*
2189 	 * Register the fabric for use within TCM
2190 	 */
2191 	ret = target_fabric_configfs_register(fabric);
2192 	if (ret < 0) {
2193 		pr_err("target_fabric_configfs_register() failed"
2194 				" for TCM_VHOST\n");
2195 		return ret;
2196 	}
2197 	/*
2198 	 * Setup our local pointer to *fabric
2199 	 */
2200 	tcm_vhost_fabric_configfs = fabric;
2201 	pr_debug("TCM_VHOST[0] - Set fabric -> tcm_vhost_fabric_configfs\n");
2202 	return 0;
2203 };
2204 
2205 static void tcm_vhost_deregister_configfs(void)
2206 {
2207 	if (!tcm_vhost_fabric_configfs)
2208 		return;
2209 
2210 	target_fabric_configfs_deregister(tcm_vhost_fabric_configfs);
2211 	tcm_vhost_fabric_configfs = NULL;
2212 	pr_debug("TCM_VHOST[0] - Cleared tcm_vhost_fabric_configfs\n");
2213 };
2214 
2215 static int __init tcm_vhost_init(void)
2216 {
2217 	int ret = -ENOMEM;
2218 	/*
2219 	 * Use our own dedicated workqueue for submitting I/O into
2220 	 * target core to avoid contention within system_wq.
2221 	 */
2222 	tcm_vhost_workqueue = alloc_workqueue("tcm_vhost", 0, 0);
2223 	if (!tcm_vhost_workqueue)
2224 		goto out;
2225 
2226 	ret = vhost_scsi_register();
2227 	if (ret < 0)
2228 		goto out_destroy_workqueue;
2229 
2230 	ret = tcm_vhost_register_configfs();
2231 	if (ret < 0)
2232 		goto out_vhost_scsi_deregister;
2233 
2234 	return 0;
2235 
2236 out_vhost_scsi_deregister:
2237 	vhost_scsi_deregister();
2238 out_destroy_workqueue:
2239 	destroy_workqueue(tcm_vhost_workqueue);
2240 out:
2241 	return ret;
2242 };
2243 
2244 static void tcm_vhost_exit(void)
2245 {
2246 	tcm_vhost_deregister_configfs();
2247 	vhost_scsi_deregister();
2248 	destroy_workqueue(tcm_vhost_workqueue);
2249 };
2250 
2251 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2252 MODULE_ALIAS("tcm_vhost");
2253 MODULE_LICENSE("GPL");
2254 module_init(tcm_vhost_init);
2255 module_exit(tcm_vhost_exit);
2256