Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[muen/linux.git] / drivers / vhost / scsi.c
1 /*******************************************************************************
2  * Vhost kernel TCM fabric driver for virtio SCSI initiators
3  *
4  * (C) Copyright 2010-2013 Datera, Inc.
5  * (C) Copyright 2010-2012 IBM Corp.
6  *
7  * Licensed to the Linux Foundation under the General Public License (GPL) version 2.
8  *
9  * Authors: Nicholas A. Bellinger <nab@daterainc.com>
10  *          Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License as published by
14  * the Free Software Foundation; either version 2 of the License, or
15  * (at your option) any later version.
16  *
17  * This program is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20  * GNU General Public License for more details.
21  *
22  ****************************************************************************/
23
24 #include <linux/module.h>
25 #include <linux/moduleparam.h>
26 #include <generated/utsrelease.h>
27 #include <linux/utsname.h>
28 #include <linux/init.h>
29 #include <linux/slab.h>
30 #include <linux/kthread.h>
31 #include <linux/types.h>
32 #include <linux/string.h>
33 #include <linux/configfs.h>
34 #include <linux/ctype.h>
35 #include <linux/compat.h>
36 #include <linux/eventfd.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/miscdevice.h>
40 #include <asm/unaligned.h>
41 #include <scsi/scsi_common.h>
42 #include <scsi/scsi_proto.h>
43 #include <target/target_core_base.h>
44 #include <target/target_core_fabric.h>
45 #include <linux/vhost.h>
46 #include <linux/virtio_scsi.h>
47 #include <linux/llist.h>
48 #include <linux/bitmap.h>
49
50 #include "vhost.h"
51
52 #define VHOST_SCSI_VERSION  "v0.1"
53 #define VHOST_SCSI_NAMELEN 256
54 #define VHOST_SCSI_MAX_CDB_SIZE 32
55 #define VHOST_SCSI_DEFAULT_TAGS 256
56 #define VHOST_SCSI_PREALLOC_SGLS 2048
57 #define VHOST_SCSI_PREALLOC_UPAGES 2048
58 #define VHOST_SCSI_PREALLOC_PROT_SGLS 2048
59
60 struct vhost_scsi_inflight {
61         /* Wait for the flush operation to finish */
62         struct completion comp;
63         /* Refcount for the inflight reqs */
64         struct kref kref;
65 };
66
67 struct vhost_scsi_cmd {
68         /* Descriptor from vhost_get_vq_desc() for virt_queue segment */
69         int tvc_vq_desc;
70         /* virtio-scsi initiator task attribute */
71         int tvc_task_attr;
72         /* virtio-scsi response incoming iovecs */
73         int tvc_in_iovs;
74         /* virtio-scsi initiator data direction */
75         enum dma_data_direction tvc_data_direction;
76         /* Expected data transfer length from virtio-scsi header */
77         u32 tvc_exp_data_len;
78         /* The Tag from include/linux/virtio_scsi.h:struct virtio_scsi_cmd_req */
79         u64 tvc_tag;
80         /* The number of scatterlists associated with this cmd */
81         u32 tvc_sgl_count;
82         u32 tvc_prot_sgl_count;
83         /* Saved unpacked SCSI LUN for vhost_scsi_submission_work() */
84         u32 tvc_lun;
85         /* Pointer to the SGL formatted memory from virtio-scsi */
86         struct scatterlist *tvc_sgl;
87         struct scatterlist *tvc_prot_sgl;
88         struct page **tvc_upages;
89         /* Pointer to response header iovec */
90         struct iovec tvc_resp_iov;
91         /* Pointer to vhost_scsi for our device */
92         struct vhost_scsi *tvc_vhost;
93         /* Pointer to vhost_virtqueue for the cmd */
94         struct vhost_virtqueue *tvc_vq;
95         /* Pointer to vhost nexus memory */
96         struct vhost_scsi_nexus *tvc_nexus;
97         /* The TCM I/O descriptor that is accessed via container_of() */
98         struct se_cmd tvc_se_cmd;
99         /* work item used for cmwq dispatch to vhost_scsi_submission_work() */
100         struct work_struct work;
101         /* Copy of the incoming SCSI command descriptor block (CDB) */
102         unsigned char tvc_cdb[VHOST_SCSI_MAX_CDB_SIZE];
103         /* Sense buffer that will be mapped into outgoing status */
104         unsigned char tvc_sense_buf[TRANSPORT_SENSE_BUFFER];
105         /* Completed commands list, serviced from vhost worker thread */
106         struct llist_node tvc_completion_list;
107         /* Used to track inflight cmd */
108         struct vhost_scsi_inflight *inflight;
109 };
110
111 struct vhost_scsi_nexus {
112         /* Pointer to TCM session for I_T Nexus */
113         struct se_session *tvn_se_sess;
114 };
115
116 struct vhost_scsi_tpg {
117         /* Vhost port target portal group tag for TCM */
118         u16 tport_tpgt;
119         /* Used to track number of TPG Port/Lun Links wrt to explict I_T Nexus shutdown */
120         int tv_tpg_port_count;
121         /* Used for vhost_scsi device reference to tpg_nexus, protected by tv_tpg_mutex */
122         int tv_tpg_vhost_count;
123         /* Used for enabling T10-PI with legacy devices */
124         int tv_fabric_prot_type;
125         /* list for vhost_scsi_list */
126         struct list_head tv_tpg_list;
127         /* Used to protect access for tpg_nexus */
128         struct mutex tv_tpg_mutex;
129         /* Pointer to the TCM VHost I_T Nexus for this TPG endpoint */
130         struct vhost_scsi_nexus *tpg_nexus;
131         /* Pointer back to vhost_scsi_tport */
132         struct vhost_scsi_tport *tport;
133         /* Returned by vhost_scsi_make_tpg() */
134         struct se_portal_group se_tpg;
135         /* Pointer back to vhost_scsi, protected by tv_tpg_mutex */
136         struct vhost_scsi *vhost_scsi;
137 };
138
139 struct vhost_scsi_tport {
140         /* SCSI protocol the tport is providing */
141         u8 tport_proto_id;
142         /* Binary World Wide unique Port Name for Vhost Target port */
143         u64 tport_wwpn;
144         /* ASCII formatted WWPN for Vhost Target port */
145         char tport_name[VHOST_SCSI_NAMELEN];
146         /* Returned by vhost_scsi_make_tport() */
147         struct se_wwn tport_wwn;
148 };
149
150 struct vhost_scsi_evt {
151         /* event to be sent to guest */
152         struct virtio_scsi_event event;
153         /* event list, serviced from vhost worker thread */
154         struct llist_node list;
155 };
156
157 enum {
158         VHOST_SCSI_VQ_CTL = 0,
159         VHOST_SCSI_VQ_EVT = 1,
160         VHOST_SCSI_VQ_IO = 2,
161 };
162
163 /* Note: can't set VIRTIO_F_VERSION_1 yet, since that implies ANY_LAYOUT. */
164 enum {
165         VHOST_SCSI_FEATURES = VHOST_FEATURES | (1ULL << VIRTIO_SCSI_F_HOTPLUG) |
166                                                (1ULL << VIRTIO_SCSI_F_T10_PI)
167 };
168
169 #define VHOST_SCSI_MAX_TARGET   256
170 #define VHOST_SCSI_MAX_VQ       128
171 #define VHOST_SCSI_MAX_EVENT    128
172
173 struct vhost_scsi_virtqueue {
174         struct vhost_virtqueue vq;
175         /*
176          * Reference counting for inflight reqs, used for flush operation. At
177          * each time, one reference tracks new commands submitted, while we
178          * wait for another one to reach 0.
179          */
180         struct vhost_scsi_inflight inflights[2];
181         /*
182          * Indicate current inflight in use, protected by vq->mutex.
183          * Writers must also take dev mutex and flush under it.
184          */
185         int inflight_idx;
186 };
187
188 struct vhost_scsi {
189         /* Protected by vhost_scsi->dev.mutex */
190         struct vhost_scsi_tpg **vs_tpg;
191         char vs_vhost_wwpn[TRANSPORT_IQN_LEN];
192
193         struct vhost_dev dev;
194         struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
195
196         struct vhost_work vs_completion_work; /* cmd completion work item */
197         struct llist_head vs_completion_list; /* cmd completion queue */
198
199         struct vhost_work vs_event_work; /* evt injection work item */
200         struct llist_head vs_event_list; /* evt injection queue */
201
202         bool vs_events_missed; /* any missed events, protected by vq->mutex */
203         int vs_events_nr; /* num of pending events, protected by vq->mutex */
204 };
205
206 /*
207  * Context for processing request and control queue operations.
208  */
209 struct vhost_scsi_ctx {
210         int head;
211         unsigned int out, in;
212         size_t req_size, rsp_size;
213         size_t out_size, in_size;
214         u8 *target, *lunp;
215         void *req;
216         struct iov_iter out_iter;
217 };
218
219 static struct workqueue_struct *vhost_scsi_workqueue;
220
221 /* Global spinlock to protect vhost_scsi TPG list for vhost IOCTL access */
222 static DEFINE_MUTEX(vhost_scsi_mutex);
223 static LIST_HEAD(vhost_scsi_list);
224
225 static void vhost_scsi_done_inflight(struct kref *kref)
226 {
227         struct vhost_scsi_inflight *inflight;
228
229         inflight = container_of(kref, struct vhost_scsi_inflight, kref);
230         complete(&inflight->comp);
231 }
232
233 static void vhost_scsi_init_inflight(struct vhost_scsi *vs,
234                                     struct vhost_scsi_inflight *old_inflight[])
235 {
236         struct vhost_scsi_inflight *new_inflight;
237         struct vhost_virtqueue *vq;
238         int idx, i;
239
240         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
241                 vq = &vs->vqs[i].vq;
242
243                 mutex_lock(&vq->mutex);
244
245                 /* store old infight */
246                 idx = vs->vqs[i].inflight_idx;
247                 if (old_inflight)
248                         old_inflight[i] = &vs->vqs[i].inflights[idx];
249
250                 /* setup new infight */
251                 vs->vqs[i].inflight_idx = idx ^ 1;
252                 new_inflight = &vs->vqs[i].inflights[idx ^ 1];
253                 kref_init(&new_inflight->kref);
254                 init_completion(&new_inflight->comp);
255
256                 mutex_unlock(&vq->mutex);
257         }
258 }
259
260 static struct vhost_scsi_inflight *
261 vhost_scsi_get_inflight(struct vhost_virtqueue *vq)
262 {
263         struct vhost_scsi_inflight *inflight;
264         struct vhost_scsi_virtqueue *svq;
265
266         svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
267         inflight = &svq->inflights[svq->inflight_idx];
268         kref_get(&inflight->kref);
269
270         return inflight;
271 }
272
273 static void vhost_scsi_put_inflight(struct vhost_scsi_inflight *inflight)
274 {
275         kref_put(&inflight->kref, vhost_scsi_done_inflight);
276 }
277
278 static int vhost_scsi_check_true(struct se_portal_group *se_tpg)
279 {
280         return 1;
281 }
282
283 static int vhost_scsi_check_false(struct se_portal_group *se_tpg)
284 {
285         return 0;
286 }
287
288 static char *vhost_scsi_get_fabric_name(void)
289 {
290         return "vhost";
291 }
292
293 static char *vhost_scsi_get_fabric_wwn(struct se_portal_group *se_tpg)
294 {
295         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
296                                 struct vhost_scsi_tpg, se_tpg);
297         struct vhost_scsi_tport *tport = tpg->tport;
298
299         return &tport->tport_name[0];
300 }
301
302 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
303 {
304         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
305                                 struct vhost_scsi_tpg, se_tpg);
306         return tpg->tport_tpgt;
307 }
308
309 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group *se_tpg)
310 {
311         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
312                                 struct vhost_scsi_tpg, se_tpg);
313
314         return tpg->tv_fabric_prot_type;
315 }
316
317 static u32 vhost_scsi_tpg_get_inst_index(struct se_portal_group *se_tpg)
318 {
319         return 1;
320 }
321
322 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
323 {
324         struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
325                                 struct vhost_scsi_cmd, tvc_se_cmd);
326         struct se_session *se_sess = tv_cmd->tvc_nexus->tvn_se_sess;
327         int i;
328
329         if (tv_cmd->tvc_sgl_count) {
330                 for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
331                         put_page(sg_page(&tv_cmd->tvc_sgl[i]));
332         }
333         if (tv_cmd->tvc_prot_sgl_count) {
334                 for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
335                         put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
336         }
337
338         vhost_scsi_put_inflight(tv_cmd->inflight);
339         target_free_tag(se_sess, se_cmd);
340 }
341
342 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
343 {
344         return 0;
345 }
346
347 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
348 {
349         /* Go ahead and process the write immediately */
350         target_execute_cmd(se_cmd);
351         return 0;
352 }
353
354 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
355 {
356         return 0;
357 }
358
359 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
360 {
361         return;
362 }
363
364 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
365 {
366         return 0;
367 }
368
369 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
370 {
371         struct vhost_scsi *vs = cmd->tvc_vhost;
372
373         llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
374
375         vhost_work_queue(&vs->dev, &vs->vs_completion_work);
376 }
377
378 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
379 {
380         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
381                                 struct vhost_scsi_cmd, tvc_se_cmd);
382         vhost_scsi_complete_cmd(cmd);
383         return 0;
384 }
385
386 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
387 {
388         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
389                                 struct vhost_scsi_cmd, tvc_se_cmd);
390         vhost_scsi_complete_cmd(cmd);
391         return 0;
392 }
393
394 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
395 {
396         return;
397 }
398
399 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
400 {
401         return;
402 }
403
404 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
405 {
406         vs->vs_events_nr--;
407         kfree(evt);
408 }
409
410 static struct vhost_scsi_evt *
411 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
412                        u32 event, u32 reason)
413 {
414         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
415         struct vhost_scsi_evt *evt;
416
417         if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
418                 vs->vs_events_missed = true;
419                 return NULL;
420         }
421
422         evt = kzalloc(sizeof(*evt), GFP_KERNEL);
423         if (!evt) {
424                 vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
425                 vs->vs_events_missed = true;
426                 return NULL;
427         }
428
429         evt->event.event = cpu_to_vhost32(vq, event);
430         evt->event.reason = cpu_to_vhost32(vq, reason);
431         vs->vs_events_nr++;
432
433         return evt;
434 }
435
436 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
437 {
438         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
439
440         /* TODO locking against target/backend threads? */
441         transport_generic_free_cmd(se_cmd, 0);
442
443 }
444
445 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
446 {
447         return target_put_sess_cmd(se_cmd);
448 }
449
450 static void
451 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
452 {
453         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
454         struct virtio_scsi_event *event = &evt->event;
455         struct virtio_scsi_event __user *eventp;
456         unsigned out, in;
457         int head, ret;
458
459         if (!vq->private_data) {
460                 vs->vs_events_missed = true;
461                 return;
462         }
463
464 again:
465         vhost_disable_notify(&vs->dev, vq);
466         head = vhost_get_vq_desc(vq, vq->iov,
467                         ARRAY_SIZE(vq->iov), &out, &in,
468                         NULL, NULL);
469         if (head < 0) {
470                 vs->vs_events_missed = true;
471                 return;
472         }
473         if (head == vq->num) {
474                 if (vhost_enable_notify(&vs->dev, vq))
475                         goto again;
476                 vs->vs_events_missed = true;
477                 return;
478         }
479
480         if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
481                 vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
482                                 vq->iov[out].iov_len);
483                 vs->vs_events_missed = true;
484                 return;
485         }
486
487         if (vs->vs_events_missed) {
488                 event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
489                 vs->vs_events_missed = false;
490         }
491
492         eventp = vq->iov[out].iov_base;
493         ret = __copy_to_user(eventp, event, sizeof(*event));
494         if (!ret)
495                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
496         else
497                 vq_err(vq, "Faulted on vhost_scsi_send_event\n");
498 }
499
500 static void vhost_scsi_evt_work(struct vhost_work *work)
501 {
502         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
503                                         vs_event_work);
504         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
505         struct vhost_scsi_evt *evt, *t;
506         struct llist_node *llnode;
507
508         mutex_lock(&vq->mutex);
509         llnode = llist_del_all(&vs->vs_event_list);
510         llist_for_each_entry_safe(evt, t, llnode, list) {
511                 vhost_scsi_do_evt_work(vs, evt);
512                 vhost_scsi_free_evt(vs, evt);
513         }
514         mutex_unlock(&vq->mutex);
515 }
516
517 /* Fill in status and signal that we are done processing this command
518  *
519  * This is scheduled in the vhost work queue so we are called with the owner
520  * process mm and can access the vring.
521  */
522 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
523 {
524         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
525                                         vs_completion_work);
526         DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
527         struct virtio_scsi_cmd_resp v_rsp;
528         struct vhost_scsi_cmd *cmd, *t;
529         struct llist_node *llnode;
530         struct se_cmd *se_cmd;
531         struct iov_iter iov_iter;
532         int ret, vq;
533
534         bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
535         llnode = llist_del_all(&vs->vs_completion_list);
536         llist_for_each_entry_safe(cmd, t, llnode, tvc_completion_list) {
537                 se_cmd = &cmd->tvc_se_cmd;
538
539                 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
540                         cmd, se_cmd->residual_count, se_cmd->scsi_status);
541
542                 memset(&v_rsp, 0, sizeof(v_rsp));
543                 v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
544                 /* TODO is status_qualifier field needed? */
545                 v_rsp.status = se_cmd->scsi_status;
546                 v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
547                                                  se_cmd->scsi_sense_length);
548                 memcpy(v_rsp.sense, cmd->tvc_sense_buf,
549                        se_cmd->scsi_sense_length);
550
551                 iov_iter_init(&iov_iter, READ, &cmd->tvc_resp_iov,
552                               cmd->tvc_in_iovs, sizeof(v_rsp));
553                 ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
554                 if (likely(ret == sizeof(v_rsp))) {
555                         struct vhost_scsi_virtqueue *q;
556                         vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
557                         q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
558                         vq = q - vs->vqs;
559                         __set_bit(vq, signal);
560                 } else
561                         pr_err("Faulted on virtio_scsi_cmd_resp\n");
562
563                 vhost_scsi_free_cmd(cmd);
564         }
565
566         vq = -1;
567         while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
568                 < VHOST_SCSI_MAX_VQ)
569                 vhost_signal(&vs->dev, &vs->vqs[vq].vq);
570 }
571
572 static struct vhost_scsi_cmd *
573 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
574                    unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
575                    u32 exp_data_len, int data_direction)
576 {
577         struct vhost_scsi_cmd *cmd;
578         struct vhost_scsi_nexus *tv_nexus;
579         struct se_session *se_sess;
580         struct scatterlist *sg, *prot_sg;
581         struct page **pages;
582         int tag, cpu;
583
584         tv_nexus = tpg->tpg_nexus;
585         if (!tv_nexus) {
586                 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
587                 return ERR_PTR(-EIO);
588         }
589         se_sess = tv_nexus->tvn_se_sess;
590
591         tag = sbitmap_queue_get(&se_sess->sess_tag_pool, &cpu);
592         if (tag < 0) {
593                 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
594                 return ERR_PTR(-ENOMEM);
595         }
596
597         cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
598         sg = cmd->tvc_sgl;
599         prot_sg = cmd->tvc_prot_sgl;
600         pages = cmd->tvc_upages;
601         memset(cmd, 0, sizeof(*cmd));
602         cmd->tvc_sgl = sg;
603         cmd->tvc_prot_sgl = prot_sg;
604         cmd->tvc_upages = pages;
605         cmd->tvc_se_cmd.map_tag = tag;
606         cmd->tvc_se_cmd.map_cpu = cpu;
607         cmd->tvc_tag = scsi_tag;
608         cmd->tvc_lun = lun;
609         cmd->tvc_task_attr = task_attr;
610         cmd->tvc_exp_data_len = exp_data_len;
611         cmd->tvc_data_direction = data_direction;
612         cmd->tvc_nexus = tv_nexus;
613         cmd->inflight = vhost_scsi_get_inflight(vq);
614
615         memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
616
617         return cmd;
618 }
619
620 /*
621  * Map a user memory range into a scatterlist
622  *
623  * Returns the number of scatterlist entries used or -errno on error.
624  */
625 static int
626 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
627                       struct iov_iter *iter,
628                       struct scatterlist *sgl,
629                       bool write)
630 {
631         struct page **pages = cmd->tvc_upages;
632         struct scatterlist *sg = sgl;
633         ssize_t bytes;
634         size_t offset;
635         unsigned int npages = 0;
636
637         bytes = iov_iter_get_pages(iter, pages, LONG_MAX,
638                                 VHOST_SCSI_PREALLOC_UPAGES, &offset);
639         /* No pages were pinned */
640         if (bytes <= 0)
641                 return bytes < 0 ? bytes : -EFAULT;
642
643         iov_iter_advance(iter, bytes);
644
645         while (bytes) {
646                 unsigned n = min_t(unsigned, PAGE_SIZE - offset, bytes);
647                 sg_set_page(sg++, pages[npages++], n, offset);
648                 bytes -= n;
649                 offset = 0;
650         }
651         return npages;
652 }
653
654 static int
655 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
656 {
657         int sgl_count = 0;
658
659         if (!iter || !iter->iov) {
660                 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
661                        " present\n", __func__, bytes);
662                 return -EINVAL;
663         }
664
665         sgl_count = iov_iter_npages(iter, 0xffff);
666         if (sgl_count > max_sgls) {
667                 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
668                        " max_sgls: %d\n", __func__, sgl_count, max_sgls);
669                 return -EINVAL;
670         }
671         return sgl_count;
672 }
673
674 static int
675 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
676                       struct iov_iter *iter,
677                       struct scatterlist *sg, int sg_count)
678 {
679         struct scatterlist *p = sg;
680         int ret;
681
682         while (iov_iter_count(iter)) {
683                 ret = vhost_scsi_map_to_sgl(cmd, iter, sg, write);
684                 if (ret < 0) {
685                         while (p < sg) {
686                                 struct page *page = sg_page(p++);
687                                 if (page)
688                                         put_page(page);
689                         }
690                         return ret;
691                 }
692                 sg += ret;
693         }
694         return 0;
695 }
696
697 static int
698 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
699                  size_t prot_bytes, struct iov_iter *prot_iter,
700                  size_t data_bytes, struct iov_iter *data_iter)
701 {
702         int sgl_count, ret;
703         bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
704
705         if (prot_bytes) {
706                 sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
707                                                  VHOST_SCSI_PREALLOC_PROT_SGLS);
708                 if (sgl_count < 0)
709                         return sgl_count;
710
711                 sg_init_table(cmd->tvc_prot_sgl, sgl_count);
712                 cmd->tvc_prot_sgl_count = sgl_count;
713                 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
714                          cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
715
716                 ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
717                                             cmd->tvc_prot_sgl,
718                                             cmd->tvc_prot_sgl_count);
719                 if (ret < 0) {
720                         cmd->tvc_prot_sgl_count = 0;
721                         return ret;
722                 }
723         }
724         sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
725                                          VHOST_SCSI_PREALLOC_SGLS);
726         if (sgl_count < 0)
727                 return sgl_count;
728
729         sg_init_table(cmd->tvc_sgl, sgl_count);
730         cmd->tvc_sgl_count = sgl_count;
731         pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
732                   cmd->tvc_sgl, cmd->tvc_sgl_count);
733
734         ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
735                                     cmd->tvc_sgl, cmd->tvc_sgl_count);
736         if (ret < 0) {
737                 cmd->tvc_sgl_count = 0;
738                 return ret;
739         }
740         return 0;
741 }
742
743 static int vhost_scsi_to_tcm_attr(int attr)
744 {
745         switch (attr) {
746         case VIRTIO_SCSI_S_SIMPLE:
747                 return TCM_SIMPLE_TAG;
748         case VIRTIO_SCSI_S_ORDERED:
749                 return TCM_ORDERED_TAG;
750         case VIRTIO_SCSI_S_HEAD:
751                 return TCM_HEAD_TAG;
752         case VIRTIO_SCSI_S_ACA:
753                 return TCM_ACA_TAG;
754         default:
755                 break;
756         }
757         return TCM_SIMPLE_TAG;
758 }
759
760 static void vhost_scsi_submission_work(struct work_struct *work)
761 {
762         struct vhost_scsi_cmd *cmd =
763                 container_of(work, struct vhost_scsi_cmd, work);
764         struct vhost_scsi_nexus *tv_nexus;
765         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
766         struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
767         int rc;
768
769         /* FIXME: BIDI operation */
770         if (cmd->tvc_sgl_count) {
771                 sg_ptr = cmd->tvc_sgl;
772
773                 if (cmd->tvc_prot_sgl_count)
774                         sg_prot_ptr = cmd->tvc_prot_sgl;
775                 else
776                         se_cmd->prot_pto = true;
777         } else {
778                 sg_ptr = NULL;
779         }
780         tv_nexus = cmd->tvc_nexus;
781
782         se_cmd->tag = 0;
783         rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
784                         cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
785                         cmd->tvc_lun, cmd->tvc_exp_data_len,
786                         vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
787                         cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
788                         sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
789                         cmd->tvc_prot_sgl_count);
790         if (rc < 0) {
791                 transport_send_check_condition_and_sense(se_cmd,
792                                 TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
793                 transport_generic_free_cmd(se_cmd, 0);
794         }
795 }
796
797 static void
798 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
799                            struct vhost_virtqueue *vq,
800                            int head, unsigned out)
801 {
802         struct virtio_scsi_cmd_resp __user *resp;
803         struct virtio_scsi_cmd_resp rsp;
804         int ret;
805
806         memset(&rsp, 0, sizeof(rsp));
807         rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
808         resp = vq->iov[out].iov_base;
809         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
810         if (!ret)
811                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
812         else
813                 pr_err("Faulted on virtio_scsi_cmd_resp\n");
814 }
815
816 static int
817 vhost_scsi_get_desc(struct vhost_scsi *vs, struct vhost_virtqueue *vq,
818                     struct vhost_scsi_ctx *vc)
819 {
820         int ret = -ENXIO;
821
822         vc->head = vhost_get_vq_desc(vq, vq->iov,
823                                      ARRAY_SIZE(vq->iov), &vc->out, &vc->in,
824                                      NULL, NULL);
825
826         pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
827                  vc->head, vc->out, vc->in);
828
829         /* On error, stop handling until the next kick. */
830         if (unlikely(vc->head < 0))
831                 goto done;
832
833         /* Nothing new?  Wait for eventfd to tell us they refilled. */
834         if (vc->head == vq->num) {
835                 if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
836                         vhost_disable_notify(&vs->dev, vq);
837                         ret = -EAGAIN;
838                 }
839                 goto done;
840         }
841
842         /*
843          * Get the size of request and response buffers.
844          * FIXME: Not correct for BIDI operation
845          */
846         vc->out_size = iov_length(vq->iov, vc->out);
847         vc->in_size = iov_length(&vq->iov[vc->out], vc->in);
848
849         /*
850          * Copy over the virtio-scsi request header, which for a
851          * ANY_LAYOUT enabled guest may span multiple iovecs, or a
852          * single iovec may contain both the header + outgoing
853          * WRITE payloads.
854          *
855          * copy_from_iter() will advance out_iter, so that it will
856          * point at the start of the outgoing WRITE payload, if
857          * DMA_TO_DEVICE is set.
858          */
859         iov_iter_init(&vc->out_iter, WRITE, vq->iov, vc->out, vc->out_size);
860         ret = 0;
861
862 done:
863         return ret;
864 }
865
866 static int
867 vhost_scsi_chk_size(struct vhost_virtqueue *vq, struct vhost_scsi_ctx *vc)
868 {
869         if (unlikely(vc->in_size < vc->rsp_size)) {
870                 vq_err(vq,
871                        "Response buf too small, need min %zu bytes got %zu",
872                        vc->rsp_size, vc->in_size);
873                 return -EINVAL;
874         } else if (unlikely(vc->out_size < vc->req_size)) {
875                 vq_err(vq,
876                        "Request buf too small, need min %zu bytes got %zu",
877                        vc->req_size, vc->out_size);
878                 return -EIO;
879         }
880
881         return 0;
882 }
883
884 static int
885 vhost_scsi_get_req(struct vhost_virtqueue *vq, struct vhost_scsi_ctx *vc,
886                    struct vhost_scsi_tpg **tpgp)
887 {
888         int ret = -EIO;
889
890         if (unlikely(!copy_from_iter_full(vc->req, vc->req_size,
891                                           &vc->out_iter))) {
892                 vq_err(vq, "Faulted on copy_from_iter\n");
893         } else if (unlikely(*vc->lunp != 1)) {
894                 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
895                 vq_err(vq, "Illegal virtio-scsi lun: %u\n", *vc->lunp);
896         } else {
897                 struct vhost_scsi_tpg **vs_tpg, *tpg;
898
899                 vs_tpg = vq->private_data;      /* validated at handler entry */
900
901                 tpg = READ_ONCE(vs_tpg[*vc->target]);
902                 if (unlikely(!tpg)) {
903                         vq_err(vq, "Target 0x%x does not exist\n", *vc->target);
904                 } else {
905                         if (tpgp)
906                                 *tpgp = tpg;
907                         ret = 0;
908                 }
909         }
910
911         return ret;
912 }
913
914 static void
915 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
916 {
917         struct vhost_scsi_tpg **vs_tpg, *tpg;
918         struct virtio_scsi_cmd_req v_req;
919         struct virtio_scsi_cmd_req_pi v_req_pi;
920         struct vhost_scsi_ctx vc;
921         struct vhost_scsi_cmd *cmd;
922         struct iov_iter in_iter, prot_iter, data_iter;
923         u64 tag;
924         u32 exp_data_len, data_direction;
925         int ret, prot_bytes;
926         u16 lun;
927         u8 task_attr;
928         bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
929         void *cdb;
930
931         mutex_lock(&vq->mutex);
932         /*
933          * We can handle the vq only after the endpoint is setup by calling the
934          * VHOST_SCSI_SET_ENDPOINT ioctl.
935          */
936         vs_tpg = vq->private_data;
937         if (!vs_tpg)
938                 goto out;
939
940         memset(&vc, 0, sizeof(vc));
941         vc.rsp_size = sizeof(struct virtio_scsi_cmd_resp);
942
943         vhost_disable_notify(&vs->dev, vq);
944
945         for (;;) {
946                 ret = vhost_scsi_get_desc(vs, vq, &vc);
947                 if (ret)
948                         goto err;
949
950                 /*
951                  * Setup pointers and values based upon different virtio-scsi
952                  * request header if T10_PI is enabled in KVM guest.
953                  */
954                 if (t10_pi) {
955                         vc.req = &v_req_pi;
956                         vc.req_size = sizeof(v_req_pi);
957                         vc.lunp = &v_req_pi.lun[0];
958                         vc.target = &v_req_pi.lun[1];
959                 } else {
960                         vc.req = &v_req;
961                         vc.req_size = sizeof(v_req);
962                         vc.lunp = &v_req.lun[0];
963                         vc.target = &v_req.lun[1];
964                 }
965
966                 /*
967                  * Validate the size of request and response buffers.
968                  * Check for a sane response buffer so we can report
969                  * early errors back to the guest.
970                  */
971                 ret = vhost_scsi_chk_size(vq, &vc);
972                 if (ret)
973                         goto err;
974
975                 ret = vhost_scsi_get_req(vq, &vc, &tpg);
976                 if (ret)
977                         goto err;
978
979                 ret = -EIO;     /* bad target on any error from here on */
980
981                 /*
982                  * Determine data_direction by calculating the total outgoing
983                  * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
984                  * response headers respectively.
985                  *
986                  * For DMA_TO_DEVICE this is out_iter, which is already pointing
987                  * to the right place.
988                  *
989                  * For DMA_FROM_DEVICE, the iovec will be just past the end
990                  * of the virtio-scsi response header in either the same
991                  * or immediately following iovec.
992                  *
993                  * Any associated T10_PI bytes for the outgoing / incoming
994                  * payloads are included in calculation of exp_data_len here.
995                  */
996                 prot_bytes = 0;
997
998                 if (vc.out_size > vc.req_size) {
999                         data_direction = DMA_TO_DEVICE;
1000                         exp_data_len = vc.out_size - vc.req_size;
1001                         data_iter = vc.out_iter;
1002                 } else if (vc.in_size > vc.rsp_size) {
1003                         data_direction = DMA_FROM_DEVICE;
1004                         exp_data_len = vc.in_size - vc.rsp_size;
1005
1006                         iov_iter_init(&in_iter, READ, &vq->iov[vc.out], vc.in,
1007                                       vc.rsp_size + exp_data_len);
1008                         iov_iter_advance(&in_iter, vc.rsp_size);
1009                         data_iter = in_iter;
1010                 } else {
1011                         data_direction = DMA_NONE;
1012                         exp_data_len = 0;
1013                 }
1014                 /*
1015                  * If T10_PI header + payload is present, setup prot_iter values
1016                  * and recalculate data_iter for vhost_scsi_mapal() mapping to
1017                  * host scatterlists via get_user_pages_fast().
1018                  */
1019                 if (t10_pi) {
1020                         if (v_req_pi.pi_bytesout) {
1021                                 if (data_direction != DMA_TO_DEVICE) {
1022                                         vq_err(vq, "Received non zero pi_bytesout,"
1023                                                 " but wrong data_direction\n");
1024                                         goto err;
1025                                 }
1026                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
1027                         } else if (v_req_pi.pi_bytesin) {
1028                                 if (data_direction != DMA_FROM_DEVICE) {
1029                                         vq_err(vq, "Received non zero pi_bytesin,"
1030                                                 " but wrong data_direction\n");
1031                                         goto err;
1032                                 }
1033                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
1034                         }
1035                         /*
1036                          * Set prot_iter to data_iter and truncate it to
1037                          * prot_bytes, and advance data_iter past any
1038                          * preceeding prot_bytes that may be present.
1039                          *
1040                          * Also fix up the exp_data_len to reflect only the
1041                          * actual data payload length.
1042                          */
1043                         if (prot_bytes) {
1044                                 exp_data_len -= prot_bytes;
1045                                 prot_iter = data_iter;
1046                                 iov_iter_truncate(&prot_iter, prot_bytes);
1047                                 iov_iter_advance(&data_iter, prot_bytes);
1048                         }
1049                         tag = vhost64_to_cpu(vq, v_req_pi.tag);
1050                         task_attr = v_req_pi.task_attr;
1051                         cdb = &v_req_pi.cdb[0];
1052                         lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
1053                 } else {
1054                         tag = vhost64_to_cpu(vq, v_req.tag);
1055                         task_attr = v_req.task_attr;
1056                         cdb = &v_req.cdb[0];
1057                         lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1058                 }
1059                 /*
1060                  * Check that the received CDB size does not exceeded our
1061                  * hardcoded max for vhost-scsi, then get a pre-allocated
1062                  * cmd descriptor for the new virtio-scsi tag.
1063                  *
1064                  * TODO what if cdb was too small for varlen cdb header?
1065                  */
1066                 if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1067                         vq_err(vq, "Received SCSI CDB with command_size: %d that"
1068                                 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1069                                 scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1070                                 goto err;
1071                 }
1072                 cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1073                                          exp_data_len + prot_bytes,
1074                                          data_direction);
1075                 if (IS_ERR(cmd)) {
1076                         vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1077                                PTR_ERR(cmd));
1078                         goto err;
1079                 }
1080                 cmd->tvc_vhost = vs;
1081                 cmd->tvc_vq = vq;
1082                 cmd->tvc_resp_iov = vq->iov[vc.out];
1083                 cmd->tvc_in_iovs = vc.in;
1084
1085                 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1086                          cmd->tvc_cdb[0], cmd->tvc_lun);
1087                 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1088                          " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1089
1090                 if (data_direction != DMA_NONE) {
1091                         if (unlikely(vhost_scsi_mapal(cmd, prot_bytes,
1092                                                       &prot_iter, exp_data_len,
1093                                                       &data_iter))) {
1094                                 vq_err(vq, "Failed to map iov to sgl\n");
1095                                 vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1096                                 goto err;
1097                         }
1098                 }
1099                 /*
1100                  * Save the descriptor from vhost_get_vq_desc() to be used to
1101                  * complete the virtio-scsi request in TCM callback context via
1102                  * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1103                  */
1104                 cmd->tvc_vq_desc = vc.head;
1105                 /*
1106                  * Dispatch cmd descriptor for cmwq execution in process
1107                  * context provided by vhost_scsi_workqueue.  This also ensures
1108                  * cmd is executed on the same kworker CPU as this vhost
1109                  * thread to gain positive L2 cache locality effects.
1110                  */
1111                 INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1112                 queue_work(vhost_scsi_workqueue, &cmd->work);
1113                 ret = 0;
1114 err:
1115                 /*
1116                  * ENXIO:  No more requests, or read error, wait for next kick
1117                  * EINVAL: Invalid response buffer, drop the request
1118                  * EIO:    Respond with bad target
1119                  * EAGAIN: Pending request
1120                  */
1121                 if (ret == -ENXIO)
1122                         break;
1123                 else if (ret == -EIO)
1124                         vhost_scsi_send_bad_target(vs, vq, vc.head, vc.out);
1125         }
1126 out:
1127         mutex_unlock(&vq->mutex);
1128 }
1129
1130 static void
1131 vhost_scsi_send_tmf_reject(struct vhost_scsi *vs,
1132                            struct vhost_virtqueue *vq,
1133                            struct vhost_scsi_ctx *vc)
1134 {
1135         struct virtio_scsi_ctrl_tmf_resp __user *resp;
1136         struct virtio_scsi_ctrl_tmf_resp rsp;
1137         int ret;
1138
1139         pr_debug("%s\n", __func__);
1140         memset(&rsp, 0, sizeof(rsp));
1141         rsp.response = VIRTIO_SCSI_S_FUNCTION_REJECTED;
1142         resp = vq->iov[vc->out].iov_base;
1143         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
1144         if (!ret)
1145                 vhost_add_used_and_signal(&vs->dev, vq, vc->head, 0);
1146         else
1147                 pr_err("Faulted on virtio_scsi_ctrl_tmf_resp\n");
1148 }
1149
1150 static void
1151 vhost_scsi_send_an_resp(struct vhost_scsi *vs,
1152                         struct vhost_virtqueue *vq,
1153                         struct vhost_scsi_ctx *vc)
1154 {
1155         struct virtio_scsi_ctrl_an_resp __user *resp;
1156         struct virtio_scsi_ctrl_an_resp rsp;
1157         int ret;
1158
1159         pr_debug("%s\n", __func__);
1160         memset(&rsp, 0, sizeof(rsp));   /* event_actual = 0 */
1161         rsp.response = VIRTIO_SCSI_S_OK;
1162         resp = vq->iov[vc->out].iov_base;
1163         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
1164         if (!ret)
1165                 vhost_add_used_and_signal(&vs->dev, vq, vc->head, 0);
1166         else
1167                 pr_err("Faulted on virtio_scsi_ctrl_an_resp\n");
1168 }
1169
1170 static void
1171 vhost_scsi_ctl_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
1172 {
1173         union {
1174                 __virtio32 type;
1175                 struct virtio_scsi_ctrl_an_req an;
1176                 struct virtio_scsi_ctrl_tmf_req tmf;
1177         } v_req;
1178         struct vhost_scsi_ctx vc;
1179         size_t typ_size;
1180         int ret;
1181
1182         mutex_lock(&vq->mutex);
1183         /*
1184          * We can handle the vq only after the endpoint is setup by calling the
1185          * VHOST_SCSI_SET_ENDPOINT ioctl.
1186          */
1187         if (!vq->private_data)
1188                 goto out;
1189
1190         memset(&vc, 0, sizeof(vc));
1191
1192         vhost_disable_notify(&vs->dev, vq);
1193
1194         for (;;) {
1195                 ret = vhost_scsi_get_desc(vs, vq, &vc);
1196                 if (ret)
1197                         goto err;
1198
1199                 /*
1200                  * Get the request type first in order to setup
1201                  * other parameters dependent on the type.
1202                  */
1203                 vc.req = &v_req.type;
1204                 typ_size = sizeof(v_req.type);
1205
1206                 if (unlikely(!copy_from_iter_full(vc.req, typ_size,
1207                                                   &vc.out_iter))) {
1208                         vq_err(vq, "Faulted on copy_from_iter tmf type\n");
1209                         /*
1210                          * The size of the response buffer depends on the
1211                          * request type and must be validated against it.
1212                          * Since the request type is not known, don't send
1213                          * a response.
1214                          */
1215                         continue;
1216                 }
1217
1218                 switch (v_req.type) {
1219                 case VIRTIO_SCSI_T_TMF:
1220                         vc.req = &v_req.tmf;
1221                         vc.req_size = sizeof(struct virtio_scsi_ctrl_tmf_req);
1222                         vc.rsp_size = sizeof(struct virtio_scsi_ctrl_tmf_resp);
1223                         vc.lunp = &v_req.tmf.lun[0];
1224                         vc.target = &v_req.tmf.lun[1];
1225                         break;
1226                 case VIRTIO_SCSI_T_AN_QUERY:
1227                 case VIRTIO_SCSI_T_AN_SUBSCRIBE:
1228                         vc.req = &v_req.an;
1229                         vc.req_size = sizeof(struct virtio_scsi_ctrl_an_req);
1230                         vc.rsp_size = sizeof(struct virtio_scsi_ctrl_an_resp);
1231                         vc.lunp = &v_req.an.lun[0];
1232                         vc.target = NULL;
1233                         break;
1234                 default:
1235                         vq_err(vq, "Unknown control request %d", v_req.type);
1236                         continue;
1237                 }
1238
1239                 /*
1240                  * Validate the size of request and response buffers.
1241                  * Check for a sane response buffer so we can report
1242                  * early errors back to the guest.
1243                  */
1244                 ret = vhost_scsi_chk_size(vq, &vc);
1245                 if (ret)
1246                         goto err;
1247
1248                 /*
1249                  * Get the rest of the request now that its size is known.
1250                  */
1251                 vc.req += typ_size;
1252                 vc.req_size -= typ_size;
1253
1254                 ret = vhost_scsi_get_req(vq, &vc, NULL);
1255                 if (ret)
1256                         goto err;
1257
1258                 if (v_req.type == VIRTIO_SCSI_T_TMF)
1259                         vhost_scsi_send_tmf_reject(vs, vq, &vc);
1260                 else
1261                         vhost_scsi_send_an_resp(vs, vq, &vc);
1262 err:
1263                 /*
1264                  * ENXIO:  No more requests, or read error, wait for next kick
1265                  * EINVAL: Invalid response buffer, drop the request
1266                  * EIO:    Respond with bad target
1267                  * EAGAIN: Pending request
1268                  */
1269                 if (ret == -ENXIO)
1270                         break;
1271                 else if (ret == -EIO)
1272                         vhost_scsi_send_bad_target(vs, vq, vc.head, vc.out);
1273         }
1274 out:
1275         mutex_unlock(&vq->mutex);
1276 }
1277
1278 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1279 {
1280         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1281                                                 poll.work);
1282         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1283
1284         pr_debug("%s: The handling func for control queue.\n", __func__);
1285         vhost_scsi_ctl_handle_vq(vs, vq);
1286 }
1287
1288 static void
1289 vhost_scsi_send_evt(struct vhost_scsi *vs,
1290                    struct vhost_scsi_tpg *tpg,
1291                    struct se_lun *lun,
1292                    u32 event,
1293                    u32 reason)
1294 {
1295         struct vhost_scsi_evt *evt;
1296
1297         evt = vhost_scsi_allocate_evt(vs, event, reason);
1298         if (!evt)
1299                 return;
1300
1301         if (tpg && lun) {
1302                 /* TODO: share lun setup code with virtio-scsi.ko */
1303                 /*
1304                  * Note: evt->event is zeroed when we allocate it and
1305                  * lun[4-7] need to be zero according to virtio-scsi spec.
1306                  */
1307                 evt->event.lun[0] = 0x01;
1308                 evt->event.lun[1] = tpg->tport_tpgt;
1309                 if (lun->unpacked_lun >= 256)
1310                         evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1311                 evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1312         }
1313
1314         llist_add(&evt->list, &vs->vs_event_list);
1315         vhost_work_queue(&vs->dev, &vs->vs_event_work);
1316 }
1317
1318 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1319 {
1320         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1321                                                 poll.work);
1322         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1323
1324         mutex_lock(&vq->mutex);
1325         if (!vq->private_data)
1326                 goto out;
1327
1328         if (vs->vs_events_missed)
1329                 vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1330 out:
1331         mutex_unlock(&vq->mutex);
1332 }
1333
1334 static void vhost_scsi_handle_kick(struct vhost_work *work)
1335 {
1336         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1337                                                 poll.work);
1338         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1339
1340         vhost_scsi_handle_vq(vs, vq);
1341 }
1342
1343 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1344 {
1345         vhost_poll_flush(&vs->vqs[index].vq.poll);
1346 }
1347
1348 /* Callers must hold dev mutex */
1349 static void vhost_scsi_flush(struct vhost_scsi *vs)
1350 {
1351         struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1352         int i;
1353
1354         /* Init new inflight and remember the old inflight */
1355         vhost_scsi_init_inflight(vs, old_inflight);
1356
1357         /*
1358          * The inflight->kref was initialized to 1. We decrement it here to
1359          * indicate the start of the flush operation so that it will reach 0
1360          * when all the reqs are finished.
1361          */
1362         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1363                 kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1364
1365         /* Flush both the vhost poll and vhost work */
1366         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1367                 vhost_scsi_flush_vq(vs, i);
1368         vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1369         vhost_work_flush(&vs->dev, &vs->vs_event_work);
1370
1371         /* Wait for all reqs issued before the flush to be finished */
1372         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1373                 wait_for_completion(&old_inflight[i]->comp);
1374 }
1375
1376 /*
1377  * Called from vhost_scsi_ioctl() context to walk the list of available
1378  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1379  *
1380  *  The lock nesting rule is:
1381  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1382  */
1383 static int
1384 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1385                         struct vhost_scsi_target *t)
1386 {
1387         struct se_portal_group *se_tpg;
1388         struct vhost_scsi_tport *tv_tport;
1389         struct vhost_scsi_tpg *tpg;
1390         struct vhost_scsi_tpg **vs_tpg;
1391         struct vhost_virtqueue *vq;
1392         int index, ret, i, len;
1393         bool match = false;
1394
1395         mutex_lock(&vhost_scsi_mutex);
1396         mutex_lock(&vs->dev.mutex);
1397
1398         /* Verify that ring has been setup correctly. */
1399         for (index = 0; index < vs->dev.nvqs; ++index) {
1400                 /* Verify that ring has been setup correctly. */
1401                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1402                         ret = -EFAULT;
1403                         goto out;
1404                 }
1405         }
1406
1407         len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1408         vs_tpg = kzalloc(len, GFP_KERNEL);
1409         if (!vs_tpg) {
1410                 ret = -ENOMEM;
1411                 goto out;
1412         }
1413         if (vs->vs_tpg)
1414                 memcpy(vs_tpg, vs->vs_tpg, len);
1415
1416         list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1417                 mutex_lock(&tpg->tv_tpg_mutex);
1418                 if (!tpg->tpg_nexus) {
1419                         mutex_unlock(&tpg->tv_tpg_mutex);
1420                         continue;
1421                 }
1422                 if (tpg->tv_tpg_vhost_count != 0) {
1423                         mutex_unlock(&tpg->tv_tpg_mutex);
1424                         continue;
1425                 }
1426                 tv_tport = tpg->tport;
1427
1428                 if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1429                         if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1430                                 kfree(vs_tpg);
1431                                 mutex_unlock(&tpg->tv_tpg_mutex);
1432                                 ret = -EEXIST;
1433                                 goto out;
1434                         }
1435                         /*
1436                          * In order to ensure individual vhost-scsi configfs
1437                          * groups cannot be removed while in use by vhost ioctl,
1438                          * go ahead and take an explicit se_tpg->tpg_group.cg_item
1439                          * dependency now.
1440                          */
1441                         se_tpg = &tpg->se_tpg;
1442                         ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1443                         if (ret) {
1444                                 pr_warn("configfs_depend_item() failed: %d\n", ret);
1445                                 kfree(vs_tpg);
1446                                 mutex_unlock(&tpg->tv_tpg_mutex);
1447                                 goto out;
1448                         }
1449                         tpg->tv_tpg_vhost_count++;
1450                         tpg->vhost_scsi = vs;
1451                         vs_tpg[tpg->tport_tpgt] = tpg;
1452                         smp_mb__after_atomic();
1453                         match = true;
1454                 }
1455                 mutex_unlock(&tpg->tv_tpg_mutex);
1456         }
1457
1458         if (match) {
1459                 memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1460                        sizeof(vs->vs_vhost_wwpn));
1461                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1462                         vq = &vs->vqs[i].vq;
1463                         mutex_lock(&vq->mutex);
1464                         vq->private_data = vs_tpg;
1465                         vhost_vq_init_access(vq);
1466                         mutex_unlock(&vq->mutex);
1467                 }
1468                 ret = 0;
1469         } else {
1470                 ret = -EEXIST;
1471         }
1472
1473         /*
1474          * Act as synchronize_rcu to make sure access to
1475          * old vs->vs_tpg is finished.
1476          */
1477         vhost_scsi_flush(vs);
1478         kfree(vs->vs_tpg);
1479         vs->vs_tpg = vs_tpg;
1480
1481 out:
1482         mutex_unlock(&vs->dev.mutex);
1483         mutex_unlock(&vhost_scsi_mutex);
1484         return ret;
1485 }
1486
1487 static int
1488 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1489                           struct vhost_scsi_target *t)
1490 {
1491         struct se_portal_group *se_tpg;
1492         struct vhost_scsi_tport *tv_tport;
1493         struct vhost_scsi_tpg *tpg;
1494         struct vhost_virtqueue *vq;
1495         bool match = false;
1496         int index, ret, i;
1497         u8 target;
1498
1499         mutex_lock(&vhost_scsi_mutex);
1500         mutex_lock(&vs->dev.mutex);
1501         /* Verify that ring has been setup correctly. */
1502         for (index = 0; index < vs->dev.nvqs; ++index) {
1503                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1504                         ret = -EFAULT;
1505                         goto err_dev;
1506                 }
1507         }
1508
1509         if (!vs->vs_tpg) {
1510                 ret = 0;
1511                 goto err_dev;
1512         }
1513
1514         for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1515                 target = i;
1516                 tpg = vs->vs_tpg[target];
1517                 if (!tpg)
1518                         continue;
1519
1520                 mutex_lock(&tpg->tv_tpg_mutex);
1521                 tv_tport = tpg->tport;
1522                 if (!tv_tport) {
1523                         ret = -ENODEV;
1524                         goto err_tpg;
1525                 }
1526
1527                 if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1528                         pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1529                                 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1530                                 tv_tport->tport_name, tpg->tport_tpgt,
1531                                 t->vhost_wwpn, t->vhost_tpgt);
1532                         ret = -EINVAL;
1533                         goto err_tpg;
1534                 }
1535                 tpg->tv_tpg_vhost_count--;
1536                 tpg->vhost_scsi = NULL;
1537                 vs->vs_tpg[target] = NULL;
1538                 match = true;
1539                 mutex_unlock(&tpg->tv_tpg_mutex);
1540                 /*
1541                  * Release se_tpg->tpg_group.cg_item configfs dependency now
1542                  * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1543                  */
1544                 se_tpg = &tpg->se_tpg;
1545                 target_undepend_item(&se_tpg->tpg_group.cg_item);
1546         }
1547         if (match) {
1548                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1549                         vq = &vs->vqs[i].vq;
1550                         mutex_lock(&vq->mutex);
1551                         vq->private_data = NULL;
1552                         mutex_unlock(&vq->mutex);
1553                 }
1554         }
1555         /*
1556          * Act as synchronize_rcu to make sure access to
1557          * old vs->vs_tpg is finished.
1558          */
1559         vhost_scsi_flush(vs);
1560         kfree(vs->vs_tpg);
1561         vs->vs_tpg = NULL;
1562         WARN_ON(vs->vs_events_nr);
1563         mutex_unlock(&vs->dev.mutex);
1564         mutex_unlock(&vhost_scsi_mutex);
1565         return 0;
1566
1567 err_tpg:
1568         mutex_unlock(&tpg->tv_tpg_mutex);
1569 err_dev:
1570         mutex_unlock(&vs->dev.mutex);
1571         mutex_unlock(&vhost_scsi_mutex);
1572         return ret;
1573 }
1574
1575 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1576 {
1577         struct vhost_virtqueue *vq;
1578         int i;
1579
1580         if (features & ~VHOST_SCSI_FEATURES)
1581                 return -EOPNOTSUPP;
1582
1583         mutex_lock(&vs->dev.mutex);
1584         if ((features & (1 << VHOST_F_LOG_ALL)) &&
1585             !vhost_log_access_ok(&vs->dev)) {
1586                 mutex_unlock(&vs->dev.mutex);
1587                 return -EFAULT;
1588         }
1589
1590         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1591                 vq = &vs->vqs[i].vq;
1592                 mutex_lock(&vq->mutex);
1593                 vq->acked_features = features;
1594                 mutex_unlock(&vq->mutex);
1595         }
1596         mutex_unlock(&vs->dev.mutex);
1597         return 0;
1598 }
1599
1600 static int vhost_scsi_open(struct inode *inode, struct file *f)
1601 {
1602         struct vhost_scsi *vs;
1603         struct vhost_virtqueue **vqs;
1604         int r = -ENOMEM, i;
1605
1606         vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_RETRY_MAYFAIL);
1607         if (!vs) {
1608                 vs = vzalloc(sizeof(*vs));
1609                 if (!vs)
1610                         goto err_vs;
1611         }
1612
1613         vqs = kmalloc_array(VHOST_SCSI_MAX_VQ, sizeof(*vqs), GFP_KERNEL);
1614         if (!vqs)
1615                 goto err_vqs;
1616
1617         vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1618         vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1619
1620         vs->vs_events_nr = 0;
1621         vs->vs_events_missed = false;
1622
1623         vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1624         vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1625         vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1626         vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1627         for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1628                 vqs[i] = &vs->vqs[i].vq;
1629                 vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1630         }
1631         vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ);
1632
1633         vhost_scsi_init_inflight(vs, NULL);
1634
1635         f->private_data = vs;
1636         return 0;
1637
1638 err_vqs:
1639         kvfree(vs);
1640 err_vs:
1641         return r;
1642 }
1643
1644 static int vhost_scsi_release(struct inode *inode, struct file *f)
1645 {
1646         struct vhost_scsi *vs = f->private_data;
1647         struct vhost_scsi_target t;
1648
1649         mutex_lock(&vs->dev.mutex);
1650         memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1651         mutex_unlock(&vs->dev.mutex);
1652         vhost_scsi_clear_endpoint(vs, &t);
1653         vhost_dev_stop(&vs->dev);
1654         vhost_dev_cleanup(&vs->dev);
1655         /* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1656         vhost_scsi_flush(vs);
1657         kfree(vs->dev.vqs);
1658         kvfree(vs);
1659         return 0;
1660 }
1661
1662 static long
1663 vhost_scsi_ioctl(struct file *f,
1664                  unsigned int ioctl,
1665                  unsigned long arg)
1666 {
1667         struct vhost_scsi *vs = f->private_data;
1668         struct vhost_scsi_target backend;
1669         void __user *argp = (void __user *)arg;
1670         u64 __user *featurep = argp;
1671         u32 __user *eventsp = argp;
1672         u32 events_missed;
1673         u64 features;
1674         int r, abi_version = VHOST_SCSI_ABI_VERSION;
1675         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1676
1677         switch (ioctl) {
1678         case VHOST_SCSI_SET_ENDPOINT:
1679                 if (copy_from_user(&backend, argp, sizeof backend))
1680                         return -EFAULT;
1681                 if (backend.reserved != 0)
1682                         return -EOPNOTSUPP;
1683
1684                 return vhost_scsi_set_endpoint(vs, &backend);
1685         case VHOST_SCSI_CLEAR_ENDPOINT:
1686                 if (copy_from_user(&backend, argp, sizeof backend))
1687                         return -EFAULT;
1688                 if (backend.reserved != 0)
1689                         return -EOPNOTSUPP;
1690
1691                 return vhost_scsi_clear_endpoint(vs, &backend);
1692         case VHOST_SCSI_GET_ABI_VERSION:
1693                 if (copy_to_user(argp, &abi_version, sizeof abi_version))
1694                         return -EFAULT;
1695                 return 0;
1696         case VHOST_SCSI_SET_EVENTS_MISSED:
1697                 if (get_user(events_missed, eventsp))
1698                         return -EFAULT;
1699                 mutex_lock(&vq->mutex);
1700                 vs->vs_events_missed = events_missed;
1701                 mutex_unlock(&vq->mutex);
1702                 return 0;
1703         case VHOST_SCSI_GET_EVENTS_MISSED:
1704                 mutex_lock(&vq->mutex);
1705                 events_missed = vs->vs_events_missed;
1706                 mutex_unlock(&vq->mutex);
1707                 if (put_user(events_missed, eventsp))
1708                         return -EFAULT;
1709                 return 0;
1710         case VHOST_GET_FEATURES:
1711                 features = VHOST_SCSI_FEATURES;
1712                 if (copy_to_user(featurep, &features, sizeof features))
1713                         return -EFAULT;
1714                 return 0;
1715         case VHOST_SET_FEATURES:
1716                 if (copy_from_user(&features, featurep, sizeof features))
1717                         return -EFAULT;
1718                 return vhost_scsi_set_features(vs, features);
1719         default:
1720                 mutex_lock(&vs->dev.mutex);
1721                 r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1722                 /* TODO: flush backend after dev ioctl. */
1723                 if (r == -ENOIOCTLCMD)
1724                         r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1725                 mutex_unlock(&vs->dev.mutex);
1726                 return r;
1727         }
1728 }
1729
1730 #ifdef CONFIG_COMPAT
1731 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1732                                 unsigned long arg)
1733 {
1734         return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1735 }
1736 #endif
1737
1738 static const struct file_operations vhost_scsi_fops = {
1739         .owner          = THIS_MODULE,
1740         .release        = vhost_scsi_release,
1741         .unlocked_ioctl = vhost_scsi_ioctl,
1742 #ifdef CONFIG_COMPAT
1743         .compat_ioctl   = vhost_scsi_compat_ioctl,
1744 #endif
1745         .open           = vhost_scsi_open,
1746         .llseek         = noop_llseek,
1747 };
1748
1749 static struct miscdevice vhost_scsi_misc = {
1750         MISC_DYNAMIC_MINOR,
1751         "vhost-scsi",
1752         &vhost_scsi_fops,
1753 };
1754
1755 static int __init vhost_scsi_register(void)
1756 {
1757         return misc_register(&vhost_scsi_misc);
1758 }
1759
1760 static void vhost_scsi_deregister(void)
1761 {
1762         misc_deregister(&vhost_scsi_misc);
1763 }
1764
1765 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1766 {
1767         switch (tport->tport_proto_id) {
1768         case SCSI_PROTOCOL_SAS:
1769                 return "SAS";
1770         case SCSI_PROTOCOL_FCP:
1771                 return "FCP";
1772         case SCSI_PROTOCOL_ISCSI:
1773                 return "iSCSI";
1774         default:
1775                 break;
1776         }
1777
1778         return "Unknown";
1779 }
1780
1781 static void
1782 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1783                   struct se_lun *lun, bool plug)
1784 {
1785
1786         struct vhost_scsi *vs = tpg->vhost_scsi;
1787         struct vhost_virtqueue *vq;
1788         u32 reason;
1789
1790         if (!vs)
1791                 return;
1792
1793         mutex_lock(&vs->dev.mutex);
1794
1795         if (plug)
1796                 reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1797         else
1798                 reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1799
1800         vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1801         mutex_lock(&vq->mutex);
1802         if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1803                 vhost_scsi_send_evt(vs, tpg, lun,
1804                                    VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1805         mutex_unlock(&vq->mutex);
1806         mutex_unlock(&vs->dev.mutex);
1807 }
1808
1809 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1810 {
1811         vhost_scsi_do_plug(tpg, lun, true);
1812 }
1813
1814 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1815 {
1816         vhost_scsi_do_plug(tpg, lun, false);
1817 }
1818
1819 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1820                                struct se_lun *lun)
1821 {
1822         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1823                                 struct vhost_scsi_tpg, se_tpg);
1824
1825         mutex_lock(&vhost_scsi_mutex);
1826
1827         mutex_lock(&tpg->tv_tpg_mutex);
1828         tpg->tv_tpg_port_count++;
1829         mutex_unlock(&tpg->tv_tpg_mutex);
1830
1831         vhost_scsi_hotplug(tpg, lun);
1832
1833         mutex_unlock(&vhost_scsi_mutex);
1834
1835         return 0;
1836 }
1837
1838 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1839                                   struct se_lun *lun)
1840 {
1841         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1842                                 struct vhost_scsi_tpg, se_tpg);
1843
1844         mutex_lock(&vhost_scsi_mutex);
1845
1846         mutex_lock(&tpg->tv_tpg_mutex);
1847         tpg->tv_tpg_port_count--;
1848         mutex_unlock(&tpg->tv_tpg_mutex);
1849
1850         vhost_scsi_hotunplug(tpg, lun);
1851
1852         mutex_unlock(&vhost_scsi_mutex);
1853 }
1854
1855 static void vhost_scsi_free_cmd_map_res(struct se_session *se_sess)
1856 {
1857         struct vhost_scsi_cmd *tv_cmd;
1858         unsigned int i;
1859
1860         if (!se_sess->sess_cmd_map)
1861                 return;
1862
1863         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1864                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1865
1866                 kfree(tv_cmd->tvc_sgl);
1867                 kfree(tv_cmd->tvc_prot_sgl);
1868                 kfree(tv_cmd->tvc_upages);
1869         }
1870 }
1871
1872 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1873                 struct config_item *item, const char *page, size_t count)
1874 {
1875         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1876         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1877                                 struct vhost_scsi_tpg, se_tpg);
1878         unsigned long val;
1879         int ret = kstrtoul(page, 0, &val);
1880
1881         if (ret) {
1882                 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1883                 return ret;
1884         }
1885         if (val != 0 && val != 1 && val != 3) {
1886                 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1887                 return -EINVAL;
1888         }
1889         tpg->tv_fabric_prot_type = val;
1890
1891         return count;
1892 }
1893
1894 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1895                 struct config_item *item, char *page)
1896 {
1897         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1898         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1899                                 struct vhost_scsi_tpg, se_tpg);
1900
1901         return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1902 }
1903
1904 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1905
1906 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1907         &vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1908         NULL,
1909 };
1910
1911 static int vhost_scsi_nexus_cb(struct se_portal_group *se_tpg,
1912                                struct se_session *se_sess, void *p)
1913 {
1914         struct vhost_scsi_cmd *tv_cmd;
1915         unsigned int i;
1916
1917         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1918                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1919
1920                 tv_cmd->tvc_sgl = kcalloc(VHOST_SCSI_PREALLOC_SGLS,
1921                                           sizeof(struct scatterlist),
1922                                           GFP_KERNEL);
1923                 if (!tv_cmd->tvc_sgl) {
1924                         pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1925                         goto out;
1926                 }
1927
1928                 tv_cmd->tvc_upages = kcalloc(VHOST_SCSI_PREALLOC_UPAGES,
1929                                              sizeof(struct page *),
1930                                              GFP_KERNEL);
1931                 if (!tv_cmd->tvc_upages) {
1932                         pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1933                         goto out;
1934                 }
1935
1936                 tv_cmd->tvc_prot_sgl = kcalloc(VHOST_SCSI_PREALLOC_PROT_SGLS,
1937                                                sizeof(struct scatterlist),
1938                                                GFP_KERNEL);
1939                 if (!tv_cmd->tvc_prot_sgl) {
1940                         pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1941                         goto out;
1942                 }
1943         }
1944         return 0;
1945 out:
1946         vhost_scsi_free_cmd_map_res(se_sess);
1947         return -ENOMEM;
1948 }
1949
1950 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1951                                 const char *name)
1952 {
1953         struct vhost_scsi_nexus *tv_nexus;
1954
1955         mutex_lock(&tpg->tv_tpg_mutex);
1956         if (tpg->tpg_nexus) {
1957                 mutex_unlock(&tpg->tv_tpg_mutex);
1958                 pr_debug("tpg->tpg_nexus already exists\n");
1959                 return -EEXIST;
1960         }
1961
1962         tv_nexus = kzalloc(sizeof(*tv_nexus), GFP_KERNEL);
1963         if (!tv_nexus) {
1964                 mutex_unlock(&tpg->tv_tpg_mutex);
1965                 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1966                 return -ENOMEM;
1967         }
1968         /*
1969          * Since we are running in 'demo mode' this call with generate a
1970          * struct se_node_acl for the vhost_scsi struct se_portal_group with
1971          * the SCSI Initiator port name of the passed configfs group 'name'.
1972          */
1973         tv_nexus->tvn_se_sess = target_setup_session(&tpg->se_tpg,
1974                                         VHOST_SCSI_DEFAULT_TAGS,
1975                                         sizeof(struct vhost_scsi_cmd),
1976                                         TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1977                                         (unsigned char *)name, tv_nexus,
1978                                         vhost_scsi_nexus_cb);
1979         if (IS_ERR(tv_nexus->tvn_se_sess)) {
1980                 mutex_unlock(&tpg->tv_tpg_mutex);
1981                 kfree(tv_nexus);
1982                 return -ENOMEM;
1983         }
1984         tpg->tpg_nexus = tv_nexus;
1985
1986         mutex_unlock(&tpg->tv_tpg_mutex);
1987         return 0;
1988 }
1989
1990 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1991 {
1992         struct se_session *se_sess;
1993         struct vhost_scsi_nexus *tv_nexus;
1994
1995         mutex_lock(&tpg->tv_tpg_mutex);
1996         tv_nexus = tpg->tpg_nexus;
1997         if (!tv_nexus) {
1998                 mutex_unlock(&tpg->tv_tpg_mutex);
1999                 return -ENODEV;
2000         }
2001
2002         se_sess = tv_nexus->tvn_se_sess;
2003         if (!se_sess) {
2004                 mutex_unlock(&tpg->tv_tpg_mutex);
2005                 return -ENODEV;
2006         }
2007
2008         if (tpg->tv_tpg_port_count != 0) {
2009                 mutex_unlock(&tpg->tv_tpg_mutex);
2010                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
2011                         " active TPG port count: %d\n",
2012                         tpg->tv_tpg_port_count);
2013                 return -EBUSY;
2014         }
2015
2016         if (tpg->tv_tpg_vhost_count != 0) {
2017                 mutex_unlock(&tpg->tv_tpg_mutex);
2018                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
2019                         " active TPG vhost count: %d\n",
2020                         tpg->tv_tpg_vhost_count);
2021                 return -EBUSY;
2022         }
2023
2024         pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
2025                 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
2026                 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
2027
2028         vhost_scsi_free_cmd_map_res(se_sess);
2029         /*
2030          * Release the SCSI I_T Nexus to the emulated vhost Target Port
2031          */
2032         target_remove_session(se_sess);
2033         tpg->tpg_nexus = NULL;
2034         mutex_unlock(&tpg->tv_tpg_mutex);
2035
2036         kfree(tv_nexus);
2037         return 0;
2038 }
2039
2040 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
2041 {
2042         struct se_portal_group *se_tpg = to_tpg(item);
2043         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2044                                 struct vhost_scsi_tpg, se_tpg);
2045         struct vhost_scsi_nexus *tv_nexus;
2046         ssize_t ret;
2047
2048         mutex_lock(&tpg->tv_tpg_mutex);
2049         tv_nexus = tpg->tpg_nexus;
2050         if (!tv_nexus) {
2051                 mutex_unlock(&tpg->tv_tpg_mutex);
2052                 return -ENODEV;
2053         }
2054         ret = snprintf(page, PAGE_SIZE, "%s\n",
2055                         tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
2056         mutex_unlock(&tpg->tv_tpg_mutex);
2057
2058         return ret;
2059 }
2060
2061 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
2062                 const char *page, size_t count)
2063 {
2064         struct se_portal_group *se_tpg = to_tpg(item);
2065         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2066                                 struct vhost_scsi_tpg, se_tpg);
2067         struct vhost_scsi_tport *tport_wwn = tpg->tport;
2068         unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
2069         int ret;
2070         /*
2071          * Shutdown the active I_T nexus if 'NULL' is passed..
2072          */
2073         if (!strncmp(page, "NULL", 4)) {
2074                 ret = vhost_scsi_drop_nexus(tpg);
2075                 return (!ret) ? count : ret;
2076         }
2077         /*
2078          * Otherwise make sure the passed virtual Initiator port WWN matches
2079          * the fabric protocol_id set in vhost_scsi_make_tport(), and call
2080          * vhost_scsi_make_nexus().
2081          */
2082         if (strlen(page) >= VHOST_SCSI_NAMELEN) {
2083                 pr_err("Emulated NAA Sas Address: %s, exceeds"
2084                                 " max: %d\n", page, VHOST_SCSI_NAMELEN);
2085                 return -EINVAL;
2086         }
2087         snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
2088
2089         ptr = strstr(i_port, "naa.");
2090         if (ptr) {
2091                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
2092                         pr_err("Passed SAS Initiator Port %s does not"
2093                                 " match target port protoid: %s\n", i_port,
2094                                 vhost_scsi_dump_proto_id(tport_wwn));
2095                         return -EINVAL;
2096                 }
2097                 port_ptr = &i_port[0];
2098                 goto check_newline;
2099         }
2100         ptr = strstr(i_port, "fc.");
2101         if (ptr) {
2102                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
2103                         pr_err("Passed FCP Initiator Port %s does not"
2104                                 " match target port protoid: %s\n", i_port,
2105                                 vhost_scsi_dump_proto_id(tport_wwn));
2106                         return -EINVAL;
2107                 }
2108                 port_ptr = &i_port[3]; /* Skip over "fc." */
2109                 goto check_newline;
2110         }
2111         ptr = strstr(i_port, "iqn.");
2112         if (ptr) {
2113                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
2114                         pr_err("Passed iSCSI Initiator Port %s does not"
2115                                 " match target port protoid: %s\n", i_port,
2116                                 vhost_scsi_dump_proto_id(tport_wwn));
2117                         return -EINVAL;
2118                 }
2119                 port_ptr = &i_port[0];
2120                 goto check_newline;
2121         }
2122         pr_err("Unable to locate prefix for emulated Initiator Port:"
2123                         " %s\n", i_port);
2124         return -EINVAL;
2125         /*
2126          * Clear any trailing newline for the NAA WWN
2127          */
2128 check_newline:
2129         if (i_port[strlen(i_port)-1] == '\n')
2130                 i_port[strlen(i_port)-1] = '\0';
2131
2132         ret = vhost_scsi_make_nexus(tpg, port_ptr);
2133         if (ret < 0)
2134                 return ret;
2135
2136         return count;
2137 }
2138
2139 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
2140
2141 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
2142         &vhost_scsi_tpg_attr_nexus,
2143         NULL,
2144 };
2145
2146 static struct se_portal_group *
2147 vhost_scsi_make_tpg(struct se_wwn *wwn, const char *name)
2148 {
2149         struct vhost_scsi_tport *tport = container_of(wwn,
2150                         struct vhost_scsi_tport, tport_wwn);
2151
2152         struct vhost_scsi_tpg *tpg;
2153         u16 tpgt;
2154         int ret;
2155
2156         if (strstr(name, "tpgt_") != name)
2157                 return ERR_PTR(-EINVAL);
2158         if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
2159                 return ERR_PTR(-EINVAL);
2160
2161         tpg = kzalloc(sizeof(*tpg), GFP_KERNEL);
2162         if (!tpg) {
2163                 pr_err("Unable to allocate struct vhost_scsi_tpg");
2164                 return ERR_PTR(-ENOMEM);
2165         }
2166         mutex_init(&tpg->tv_tpg_mutex);
2167         INIT_LIST_HEAD(&tpg->tv_tpg_list);
2168         tpg->tport = tport;
2169         tpg->tport_tpgt = tpgt;
2170
2171         ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
2172         if (ret < 0) {
2173                 kfree(tpg);
2174                 return NULL;
2175         }
2176         mutex_lock(&vhost_scsi_mutex);
2177         list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
2178         mutex_unlock(&vhost_scsi_mutex);
2179
2180         return &tpg->se_tpg;
2181 }
2182
2183 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
2184 {
2185         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2186                                 struct vhost_scsi_tpg, se_tpg);
2187
2188         mutex_lock(&vhost_scsi_mutex);
2189         list_del(&tpg->tv_tpg_list);
2190         mutex_unlock(&vhost_scsi_mutex);
2191         /*
2192          * Release the virtual I_T Nexus for this vhost TPG
2193          */
2194         vhost_scsi_drop_nexus(tpg);
2195         /*
2196          * Deregister the se_tpg from TCM..
2197          */
2198         core_tpg_deregister(se_tpg);
2199         kfree(tpg);
2200 }
2201
2202 static struct se_wwn *
2203 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
2204                      struct config_group *group,
2205                      const char *name)
2206 {
2207         struct vhost_scsi_tport *tport;
2208         char *ptr;
2209         u64 wwpn = 0;
2210         int off = 0;
2211
2212         /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2213                 return ERR_PTR(-EINVAL); */
2214
2215         tport = kzalloc(sizeof(*tport), GFP_KERNEL);
2216         if (!tport) {
2217                 pr_err("Unable to allocate struct vhost_scsi_tport");
2218                 return ERR_PTR(-ENOMEM);
2219         }
2220         tport->tport_wwpn = wwpn;
2221         /*
2222          * Determine the emulated Protocol Identifier and Target Port Name
2223          * based on the incoming configfs directory name.
2224          */
2225         ptr = strstr(name, "naa.");
2226         if (ptr) {
2227                 tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2228                 goto check_len;
2229         }
2230         ptr = strstr(name, "fc.");
2231         if (ptr) {
2232                 tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2233                 off = 3; /* Skip over "fc." */
2234                 goto check_len;
2235         }
2236         ptr = strstr(name, "iqn.");
2237         if (ptr) {
2238                 tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2239                 goto check_len;
2240         }
2241
2242         pr_err("Unable to locate prefix for emulated Target Port:"
2243                         " %s\n", name);
2244         kfree(tport);
2245         return ERR_PTR(-EINVAL);
2246
2247 check_len:
2248         if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2249                 pr_err("Emulated %s Address: %s, exceeds"
2250                         " max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2251                         VHOST_SCSI_NAMELEN);
2252                 kfree(tport);
2253                 return ERR_PTR(-EINVAL);
2254         }
2255         snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2256
2257         pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2258                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2259
2260         return &tport->tport_wwn;
2261 }
2262
2263 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2264 {
2265         struct vhost_scsi_tport *tport = container_of(wwn,
2266                                 struct vhost_scsi_tport, tport_wwn);
2267
2268         pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2269                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2270                 tport->tport_name);
2271
2272         kfree(tport);
2273 }
2274
2275 static ssize_t
2276 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2277 {
2278         return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2279                 "on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2280                 utsname()->machine);
2281 }
2282
2283 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2284
2285 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2286         &vhost_scsi_wwn_attr_version,
2287         NULL,
2288 };
2289
2290 static const struct target_core_fabric_ops vhost_scsi_ops = {
2291         .module                         = THIS_MODULE,
2292         .name                           = "vhost",
2293         .get_fabric_name                = vhost_scsi_get_fabric_name,
2294         .tpg_get_wwn                    = vhost_scsi_get_fabric_wwn,
2295         .tpg_get_tag                    = vhost_scsi_get_tpgt,
2296         .tpg_check_demo_mode            = vhost_scsi_check_true,
2297         .tpg_check_demo_mode_cache      = vhost_scsi_check_true,
2298         .tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2299         .tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2300         .tpg_check_prot_fabric_only     = vhost_scsi_check_prot_fabric_only,
2301         .tpg_get_inst_index             = vhost_scsi_tpg_get_inst_index,
2302         .release_cmd                    = vhost_scsi_release_cmd,
2303         .check_stop_free                = vhost_scsi_check_stop_free,
2304         .sess_get_index                 = vhost_scsi_sess_get_index,
2305         .sess_get_initiator_sid         = NULL,
2306         .write_pending                  = vhost_scsi_write_pending,
2307         .write_pending_status           = vhost_scsi_write_pending_status,
2308         .set_default_node_attributes    = vhost_scsi_set_default_node_attrs,
2309         .get_cmd_state                  = vhost_scsi_get_cmd_state,
2310         .queue_data_in                  = vhost_scsi_queue_data_in,
2311         .queue_status                   = vhost_scsi_queue_status,
2312         .queue_tm_rsp                   = vhost_scsi_queue_tm_rsp,
2313         .aborted_task                   = vhost_scsi_aborted_task,
2314         /*
2315          * Setup callers for generic logic in target_core_fabric_configfs.c
2316          */
2317         .fabric_make_wwn                = vhost_scsi_make_tport,
2318         .fabric_drop_wwn                = vhost_scsi_drop_tport,
2319         .fabric_make_tpg                = vhost_scsi_make_tpg,
2320         .fabric_drop_tpg                = vhost_scsi_drop_tpg,
2321         .fabric_post_link               = vhost_scsi_port_link,
2322         .fabric_pre_unlink              = vhost_scsi_port_unlink,
2323
2324         .tfc_wwn_attrs                  = vhost_scsi_wwn_attrs,
2325         .tfc_tpg_base_attrs             = vhost_scsi_tpg_attrs,
2326         .tfc_tpg_attrib_attrs           = vhost_scsi_tpg_attrib_attrs,
2327 };
2328
2329 static int __init vhost_scsi_init(void)
2330 {
2331         int ret = -ENOMEM;
2332
2333         pr_debug("TCM_VHOST fabric module %s on %s/%s"
2334                 " on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2335                 utsname()->machine);
2336
2337         /*
2338          * Use our own dedicated workqueue for submitting I/O into
2339          * target core to avoid contention within system_wq.
2340          */
2341         vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2342         if (!vhost_scsi_workqueue)
2343                 goto out;
2344
2345         ret = vhost_scsi_register();
2346         if (ret < 0)
2347                 goto out_destroy_workqueue;
2348
2349         ret = target_register_template(&vhost_scsi_ops);
2350         if (ret < 0)
2351                 goto out_vhost_scsi_deregister;
2352
2353         return 0;
2354
2355 out_vhost_scsi_deregister:
2356         vhost_scsi_deregister();
2357 out_destroy_workqueue:
2358         destroy_workqueue(vhost_scsi_workqueue);
2359 out:
2360         return ret;
2361 };
2362
2363 static void vhost_scsi_exit(void)
2364 {
2365         target_unregister_template(&vhost_scsi_ops);
2366         vhost_scsi_deregister();
2367         destroy_workqueue(vhost_scsi_workqueue);
2368 };
2369
2370 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2371 MODULE_ALIAS("tcm_vhost");
2372 MODULE_LICENSE("GPL");
2373 module_init(vhost_scsi_init);
2374 module_exit(vhost_scsi_exit);