Merge tag 'for-4.21/block-20190102' of git://git.kernel.dk/linux-block
[sfrench/cifs-2.6.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_wwn(struct se_portal_group *se_tpg)
289 {
290         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
291                                 struct vhost_scsi_tpg, se_tpg);
292         struct vhost_scsi_tport *tport = tpg->tport;
293
294         return &tport->tport_name[0];
295 }
296
297 static u16 vhost_scsi_get_tpgt(struct se_portal_group *se_tpg)
298 {
299         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
300                                 struct vhost_scsi_tpg, se_tpg);
301         return tpg->tport_tpgt;
302 }
303
304 static int vhost_scsi_check_prot_fabric_only(struct se_portal_group *se_tpg)
305 {
306         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
307                                 struct vhost_scsi_tpg, se_tpg);
308
309         return tpg->tv_fabric_prot_type;
310 }
311
312 static u32 vhost_scsi_tpg_get_inst_index(struct se_portal_group *se_tpg)
313 {
314         return 1;
315 }
316
317 static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
318 {
319         struct vhost_scsi_cmd *tv_cmd = container_of(se_cmd,
320                                 struct vhost_scsi_cmd, tvc_se_cmd);
321         struct se_session *se_sess = tv_cmd->tvc_nexus->tvn_se_sess;
322         int i;
323
324         if (tv_cmd->tvc_sgl_count) {
325                 for (i = 0; i < tv_cmd->tvc_sgl_count; i++)
326                         put_page(sg_page(&tv_cmd->tvc_sgl[i]));
327         }
328         if (tv_cmd->tvc_prot_sgl_count) {
329                 for (i = 0; i < tv_cmd->tvc_prot_sgl_count; i++)
330                         put_page(sg_page(&tv_cmd->tvc_prot_sgl[i]));
331         }
332
333         vhost_scsi_put_inflight(tv_cmd->inflight);
334         target_free_tag(se_sess, se_cmd);
335 }
336
337 static u32 vhost_scsi_sess_get_index(struct se_session *se_sess)
338 {
339         return 0;
340 }
341
342 static int vhost_scsi_write_pending(struct se_cmd *se_cmd)
343 {
344         /* Go ahead and process the write immediately */
345         target_execute_cmd(se_cmd);
346         return 0;
347 }
348
349 static int vhost_scsi_write_pending_status(struct se_cmd *se_cmd)
350 {
351         return 0;
352 }
353
354 static void vhost_scsi_set_default_node_attrs(struct se_node_acl *nacl)
355 {
356         return;
357 }
358
359 static int vhost_scsi_get_cmd_state(struct se_cmd *se_cmd)
360 {
361         return 0;
362 }
363
364 static void vhost_scsi_complete_cmd(struct vhost_scsi_cmd *cmd)
365 {
366         struct vhost_scsi *vs = cmd->tvc_vhost;
367
368         llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
369
370         vhost_work_queue(&vs->dev, &vs->vs_completion_work);
371 }
372
373 static int vhost_scsi_queue_data_in(struct se_cmd *se_cmd)
374 {
375         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
376                                 struct vhost_scsi_cmd, tvc_se_cmd);
377         vhost_scsi_complete_cmd(cmd);
378         return 0;
379 }
380
381 static int vhost_scsi_queue_status(struct se_cmd *se_cmd)
382 {
383         struct vhost_scsi_cmd *cmd = container_of(se_cmd,
384                                 struct vhost_scsi_cmd, tvc_se_cmd);
385         vhost_scsi_complete_cmd(cmd);
386         return 0;
387 }
388
389 static void vhost_scsi_queue_tm_rsp(struct se_cmd *se_cmd)
390 {
391         return;
392 }
393
394 static void vhost_scsi_aborted_task(struct se_cmd *se_cmd)
395 {
396         return;
397 }
398
399 static void vhost_scsi_free_evt(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
400 {
401         vs->vs_events_nr--;
402         kfree(evt);
403 }
404
405 static struct vhost_scsi_evt *
406 vhost_scsi_allocate_evt(struct vhost_scsi *vs,
407                        u32 event, u32 reason)
408 {
409         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
410         struct vhost_scsi_evt *evt;
411
412         if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
413                 vs->vs_events_missed = true;
414                 return NULL;
415         }
416
417         evt = kzalloc(sizeof(*evt), GFP_KERNEL);
418         if (!evt) {
419                 vq_err(vq, "Failed to allocate vhost_scsi_evt\n");
420                 vs->vs_events_missed = true;
421                 return NULL;
422         }
423
424         evt->event.event = cpu_to_vhost32(vq, event);
425         evt->event.reason = cpu_to_vhost32(vq, reason);
426         vs->vs_events_nr++;
427
428         return evt;
429 }
430
431 static void vhost_scsi_free_cmd(struct vhost_scsi_cmd *cmd)
432 {
433         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
434
435         /* TODO locking against target/backend threads? */
436         transport_generic_free_cmd(se_cmd, 0);
437
438 }
439
440 static int vhost_scsi_check_stop_free(struct se_cmd *se_cmd)
441 {
442         return target_put_sess_cmd(se_cmd);
443 }
444
445 static void
446 vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt)
447 {
448         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
449         struct virtio_scsi_event *event = &evt->event;
450         struct virtio_scsi_event __user *eventp;
451         unsigned out, in;
452         int head, ret;
453
454         if (!vq->private_data) {
455                 vs->vs_events_missed = true;
456                 return;
457         }
458
459 again:
460         vhost_disable_notify(&vs->dev, vq);
461         head = vhost_get_vq_desc(vq, vq->iov,
462                         ARRAY_SIZE(vq->iov), &out, &in,
463                         NULL, NULL);
464         if (head < 0) {
465                 vs->vs_events_missed = true;
466                 return;
467         }
468         if (head == vq->num) {
469                 if (vhost_enable_notify(&vs->dev, vq))
470                         goto again;
471                 vs->vs_events_missed = true;
472                 return;
473         }
474
475         if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) {
476                 vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n",
477                                 vq->iov[out].iov_len);
478                 vs->vs_events_missed = true;
479                 return;
480         }
481
482         if (vs->vs_events_missed) {
483                 event->event |= cpu_to_vhost32(vq, VIRTIO_SCSI_T_EVENTS_MISSED);
484                 vs->vs_events_missed = false;
485         }
486
487         eventp = vq->iov[out].iov_base;
488         ret = __copy_to_user(eventp, event, sizeof(*event));
489         if (!ret)
490                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
491         else
492                 vq_err(vq, "Faulted on vhost_scsi_send_event\n");
493 }
494
495 static void vhost_scsi_evt_work(struct vhost_work *work)
496 {
497         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
498                                         vs_event_work);
499         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
500         struct vhost_scsi_evt *evt, *t;
501         struct llist_node *llnode;
502
503         mutex_lock(&vq->mutex);
504         llnode = llist_del_all(&vs->vs_event_list);
505         llist_for_each_entry_safe(evt, t, llnode, list) {
506                 vhost_scsi_do_evt_work(vs, evt);
507                 vhost_scsi_free_evt(vs, evt);
508         }
509         mutex_unlock(&vq->mutex);
510 }
511
512 /* Fill in status and signal that we are done processing this command
513  *
514  * This is scheduled in the vhost work queue so we are called with the owner
515  * process mm and can access the vring.
516  */
517 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
518 {
519         struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
520                                         vs_completion_work);
521         DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
522         struct virtio_scsi_cmd_resp v_rsp;
523         struct vhost_scsi_cmd *cmd, *t;
524         struct llist_node *llnode;
525         struct se_cmd *se_cmd;
526         struct iov_iter iov_iter;
527         int ret, vq;
528
529         bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
530         llnode = llist_del_all(&vs->vs_completion_list);
531         llist_for_each_entry_safe(cmd, t, llnode, tvc_completion_list) {
532                 se_cmd = &cmd->tvc_se_cmd;
533
534                 pr_debug("%s tv_cmd %p resid %u status %#02x\n", __func__,
535                         cmd, se_cmd->residual_count, se_cmd->scsi_status);
536
537                 memset(&v_rsp, 0, sizeof(v_rsp));
538                 v_rsp.resid = cpu_to_vhost32(cmd->tvc_vq, se_cmd->residual_count);
539                 /* TODO is status_qualifier field needed? */
540                 v_rsp.status = se_cmd->scsi_status;
541                 v_rsp.sense_len = cpu_to_vhost32(cmd->tvc_vq,
542                                                  se_cmd->scsi_sense_length);
543                 memcpy(v_rsp.sense, cmd->tvc_sense_buf,
544                        se_cmd->scsi_sense_length);
545
546                 iov_iter_init(&iov_iter, READ, &cmd->tvc_resp_iov,
547                               cmd->tvc_in_iovs, sizeof(v_rsp));
548                 ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
549                 if (likely(ret == sizeof(v_rsp))) {
550                         struct vhost_scsi_virtqueue *q;
551                         vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
552                         q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
553                         vq = q - vs->vqs;
554                         __set_bit(vq, signal);
555                 } else
556                         pr_err("Faulted on virtio_scsi_cmd_resp\n");
557
558                 vhost_scsi_free_cmd(cmd);
559         }
560
561         vq = -1;
562         while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
563                 < VHOST_SCSI_MAX_VQ)
564                 vhost_signal(&vs->dev, &vs->vqs[vq].vq);
565 }
566
567 static struct vhost_scsi_cmd *
568 vhost_scsi_get_tag(struct vhost_virtqueue *vq, struct vhost_scsi_tpg *tpg,
569                    unsigned char *cdb, u64 scsi_tag, u16 lun, u8 task_attr,
570                    u32 exp_data_len, int data_direction)
571 {
572         struct vhost_scsi_cmd *cmd;
573         struct vhost_scsi_nexus *tv_nexus;
574         struct se_session *se_sess;
575         struct scatterlist *sg, *prot_sg;
576         struct page **pages;
577         int tag, cpu;
578
579         tv_nexus = tpg->tpg_nexus;
580         if (!tv_nexus) {
581                 pr_err("Unable to locate active struct vhost_scsi_nexus\n");
582                 return ERR_PTR(-EIO);
583         }
584         se_sess = tv_nexus->tvn_se_sess;
585
586         tag = sbitmap_queue_get(&se_sess->sess_tag_pool, &cpu);
587         if (tag < 0) {
588                 pr_err("Unable to obtain tag for vhost_scsi_cmd\n");
589                 return ERR_PTR(-ENOMEM);
590         }
591
592         cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[tag];
593         sg = cmd->tvc_sgl;
594         prot_sg = cmd->tvc_prot_sgl;
595         pages = cmd->tvc_upages;
596         memset(cmd, 0, sizeof(*cmd));
597         cmd->tvc_sgl = sg;
598         cmd->tvc_prot_sgl = prot_sg;
599         cmd->tvc_upages = pages;
600         cmd->tvc_se_cmd.map_tag = tag;
601         cmd->tvc_se_cmd.map_cpu = cpu;
602         cmd->tvc_tag = scsi_tag;
603         cmd->tvc_lun = lun;
604         cmd->tvc_task_attr = task_attr;
605         cmd->tvc_exp_data_len = exp_data_len;
606         cmd->tvc_data_direction = data_direction;
607         cmd->tvc_nexus = tv_nexus;
608         cmd->inflight = vhost_scsi_get_inflight(vq);
609
610         memcpy(cmd->tvc_cdb, cdb, VHOST_SCSI_MAX_CDB_SIZE);
611
612         return cmd;
613 }
614
615 /*
616  * Map a user memory range into a scatterlist
617  *
618  * Returns the number of scatterlist entries used or -errno on error.
619  */
620 static int
621 vhost_scsi_map_to_sgl(struct vhost_scsi_cmd *cmd,
622                       struct iov_iter *iter,
623                       struct scatterlist *sgl,
624                       bool write)
625 {
626         struct page **pages = cmd->tvc_upages;
627         struct scatterlist *sg = sgl;
628         ssize_t bytes;
629         size_t offset;
630         unsigned int npages = 0;
631
632         bytes = iov_iter_get_pages(iter, pages, LONG_MAX,
633                                 VHOST_SCSI_PREALLOC_UPAGES, &offset);
634         /* No pages were pinned */
635         if (bytes <= 0)
636                 return bytes < 0 ? bytes : -EFAULT;
637
638         iov_iter_advance(iter, bytes);
639
640         while (bytes) {
641                 unsigned n = min_t(unsigned, PAGE_SIZE - offset, bytes);
642                 sg_set_page(sg++, pages[npages++], n, offset);
643                 bytes -= n;
644                 offset = 0;
645         }
646         return npages;
647 }
648
649 static int
650 vhost_scsi_calc_sgls(struct iov_iter *iter, size_t bytes, int max_sgls)
651 {
652         int sgl_count = 0;
653
654         if (!iter || !iter->iov) {
655                 pr_err("%s: iter->iov is NULL, but expected bytes: %zu"
656                        " present\n", __func__, bytes);
657                 return -EINVAL;
658         }
659
660         sgl_count = iov_iter_npages(iter, 0xffff);
661         if (sgl_count > max_sgls) {
662                 pr_err("%s: requested sgl_count: %d exceeds pre-allocated"
663                        " max_sgls: %d\n", __func__, sgl_count, max_sgls);
664                 return -EINVAL;
665         }
666         return sgl_count;
667 }
668
669 static int
670 vhost_scsi_iov_to_sgl(struct vhost_scsi_cmd *cmd, bool write,
671                       struct iov_iter *iter,
672                       struct scatterlist *sg, int sg_count)
673 {
674         struct scatterlist *p = sg;
675         int ret;
676
677         while (iov_iter_count(iter)) {
678                 ret = vhost_scsi_map_to_sgl(cmd, iter, sg, write);
679                 if (ret < 0) {
680                         while (p < sg) {
681                                 struct page *page = sg_page(p++);
682                                 if (page)
683                                         put_page(page);
684                         }
685                         return ret;
686                 }
687                 sg += ret;
688         }
689         return 0;
690 }
691
692 static int
693 vhost_scsi_mapal(struct vhost_scsi_cmd *cmd,
694                  size_t prot_bytes, struct iov_iter *prot_iter,
695                  size_t data_bytes, struct iov_iter *data_iter)
696 {
697         int sgl_count, ret;
698         bool write = (cmd->tvc_data_direction == DMA_FROM_DEVICE);
699
700         if (prot_bytes) {
701                 sgl_count = vhost_scsi_calc_sgls(prot_iter, prot_bytes,
702                                                  VHOST_SCSI_PREALLOC_PROT_SGLS);
703                 if (sgl_count < 0)
704                         return sgl_count;
705
706                 sg_init_table(cmd->tvc_prot_sgl, sgl_count);
707                 cmd->tvc_prot_sgl_count = sgl_count;
708                 pr_debug("%s prot_sg %p prot_sgl_count %u\n", __func__,
709                          cmd->tvc_prot_sgl, cmd->tvc_prot_sgl_count);
710
711                 ret = vhost_scsi_iov_to_sgl(cmd, write, prot_iter,
712                                             cmd->tvc_prot_sgl,
713                                             cmd->tvc_prot_sgl_count);
714                 if (ret < 0) {
715                         cmd->tvc_prot_sgl_count = 0;
716                         return ret;
717                 }
718         }
719         sgl_count = vhost_scsi_calc_sgls(data_iter, data_bytes,
720                                          VHOST_SCSI_PREALLOC_SGLS);
721         if (sgl_count < 0)
722                 return sgl_count;
723
724         sg_init_table(cmd->tvc_sgl, sgl_count);
725         cmd->tvc_sgl_count = sgl_count;
726         pr_debug("%s data_sg %p data_sgl_count %u\n", __func__,
727                   cmd->tvc_sgl, cmd->tvc_sgl_count);
728
729         ret = vhost_scsi_iov_to_sgl(cmd, write, data_iter,
730                                     cmd->tvc_sgl, cmd->tvc_sgl_count);
731         if (ret < 0) {
732                 cmd->tvc_sgl_count = 0;
733                 return ret;
734         }
735         return 0;
736 }
737
738 static int vhost_scsi_to_tcm_attr(int attr)
739 {
740         switch (attr) {
741         case VIRTIO_SCSI_S_SIMPLE:
742                 return TCM_SIMPLE_TAG;
743         case VIRTIO_SCSI_S_ORDERED:
744                 return TCM_ORDERED_TAG;
745         case VIRTIO_SCSI_S_HEAD:
746                 return TCM_HEAD_TAG;
747         case VIRTIO_SCSI_S_ACA:
748                 return TCM_ACA_TAG;
749         default:
750                 break;
751         }
752         return TCM_SIMPLE_TAG;
753 }
754
755 static void vhost_scsi_submission_work(struct work_struct *work)
756 {
757         struct vhost_scsi_cmd *cmd =
758                 container_of(work, struct vhost_scsi_cmd, work);
759         struct vhost_scsi_nexus *tv_nexus;
760         struct se_cmd *se_cmd = &cmd->tvc_se_cmd;
761         struct scatterlist *sg_ptr, *sg_prot_ptr = NULL;
762         int rc;
763
764         /* FIXME: BIDI operation */
765         if (cmd->tvc_sgl_count) {
766                 sg_ptr = cmd->tvc_sgl;
767
768                 if (cmd->tvc_prot_sgl_count)
769                         sg_prot_ptr = cmd->tvc_prot_sgl;
770                 else
771                         se_cmd->prot_pto = true;
772         } else {
773                 sg_ptr = NULL;
774         }
775         tv_nexus = cmd->tvc_nexus;
776
777         se_cmd->tag = 0;
778         rc = target_submit_cmd_map_sgls(se_cmd, tv_nexus->tvn_se_sess,
779                         cmd->tvc_cdb, &cmd->tvc_sense_buf[0],
780                         cmd->tvc_lun, cmd->tvc_exp_data_len,
781                         vhost_scsi_to_tcm_attr(cmd->tvc_task_attr),
782                         cmd->tvc_data_direction, TARGET_SCF_ACK_KREF,
783                         sg_ptr, cmd->tvc_sgl_count, NULL, 0, sg_prot_ptr,
784                         cmd->tvc_prot_sgl_count);
785         if (rc < 0) {
786                 transport_send_check_condition_and_sense(se_cmd,
787                                 TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE, 0);
788                 transport_generic_free_cmd(se_cmd, 0);
789         }
790 }
791
792 static void
793 vhost_scsi_send_bad_target(struct vhost_scsi *vs,
794                            struct vhost_virtqueue *vq,
795                            int head, unsigned out)
796 {
797         struct virtio_scsi_cmd_resp __user *resp;
798         struct virtio_scsi_cmd_resp rsp;
799         int ret;
800
801         memset(&rsp, 0, sizeof(rsp));
802         rsp.response = VIRTIO_SCSI_S_BAD_TARGET;
803         resp = vq->iov[out].iov_base;
804         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
805         if (!ret)
806                 vhost_add_used_and_signal(&vs->dev, vq, head, 0);
807         else
808                 pr_err("Faulted on virtio_scsi_cmd_resp\n");
809 }
810
811 static int
812 vhost_scsi_get_desc(struct vhost_scsi *vs, struct vhost_virtqueue *vq,
813                     struct vhost_scsi_ctx *vc)
814 {
815         int ret = -ENXIO;
816
817         vc->head = vhost_get_vq_desc(vq, vq->iov,
818                                      ARRAY_SIZE(vq->iov), &vc->out, &vc->in,
819                                      NULL, NULL);
820
821         pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n",
822                  vc->head, vc->out, vc->in);
823
824         /* On error, stop handling until the next kick. */
825         if (unlikely(vc->head < 0))
826                 goto done;
827
828         /* Nothing new?  Wait for eventfd to tell us they refilled. */
829         if (vc->head == vq->num) {
830                 if (unlikely(vhost_enable_notify(&vs->dev, vq))) {
831                         vhost_disable_notify(&vs->dev, vq);
832                         ret = -EAGAIN;
833                 }
834                 goto done;
835         }
836
837         /*
838          * Get the size of request and response buffers.
839          * FIXME: Not correct for BIDI operation
840          */
841         vc->out_size = iov_length(vq->iov, vc->out);
842         vc->in_size = iov_length(&vq->iov[vc->out], vc->in);
843
844         /*
845          * Copy over the virtio-scsi request header, which for a
846          * ANY_LAYOUT enabled guest may span multiple iovecs, or a
847          * single iovec may contain both the header + outgoing
848          * WRITE payloads.
849          *
850          * copy_from_iter() will advance out_iter, so that it will
851          * point at the start of the outgoing WRITE payload, if
852          * DMA_TO_DEVICE is set.
853          */
854         iov_iter_init(&vc->out_iter, WRITE, vq->iov, vc->out, vc->out_size);
855         ret = 0;
856
857 done:
858         return ret;
859 }
860
861 static int
862 vhost_scsi_chk_size(struct vhost_virtqueue *vq, struct vhost_scsi_ctx *vc)
863 {
864         if (unlikely(vc->in_size < vc->rsp_size)) {
865                 vq_err(vq,
866                        "Response buf too small, need min %zu bytes got %zu",
867                        vc->rsp_size, vc->in_size);
868                 return -EINVAL;
869         } else if (unlikely(vc->out_size < vc->req_size)) {
870                 vq_err(vq,
871                        "Request buf too small, need min %zu bytes got %zu",
872                        vc->req_size, vc->out_size);
873                 return -EIO;
874         }
875
876         return 0;
877 }
878
879 static int
880 vhost_scsi_get_req(struct vhost_virtqueue *vq, struct vhost_scsi_ctx *vc,
881                    struct vhost_scsi_tpg **tpgp)
882 {
883         int ret = -EIO;
884
885         if (unlikely(!copy_from_iter_full(vc->req, vc->req_size,
886                                           &vc->out_iter))) {
887                 vq_err(vq, "Faulted on copy_from_iter\n");
888         } else if (unlikely(*vc->lunp != 1)) {
889                 /* virtio-scsi spec requires byte 0 of the lun to be 1 */
890                 vq_err(vq, "Illegal virtio-scsi lun: %u\n", *vc->lunp);
891         } else {
892                 struct vhost_scsi_tpg **vs_tpg, *tpg;
893
894                 vs_tpg = vq->private_data;      /* validated at handler entry */
895
896                 tpg = READ_ONCE(vs_tpg[*vc->target]);
897                 if (unlikely(!tpg)) {
898                         vq_err(vq, "Target 0x%x does not exist\n", *vc->target);
899                 } else {
900                         if (tpgp)
901                                 *tpgp = tpg;
902                         ret = 0;
903                 }
904         }
905
906         return ret;
907 }
908
909 static void
910 vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
911 {
912         struct vhost_scsi_tpg **vs_tpg, *tpg;
913         struct virtio_scsi_cmd_req v_req;
914         struct virtio_scsi_cmd_req_pi v_req_pi;
915         struct vhost_scsi_ctx vc;
916         struct vhost_scsi_cmd *cmd;
917         struct iov_iter in_iter, prot_iter, data_iter;
918         u64 tag;
919         u32 exp_data_len, data_direction;
920         int ret, prot_bytes;
921         u16 lun;
922         u8 task_attr;
923         bool t10_pi = vhost_has_feature(vq, VIRTIO_SCSI_F_T10_PI);
924         void *cdb;
925
926         mutex_lock(&vq->mutex);
927         /*
928          * We can handle the vq only after the endpoint is setup by calling the
929          * VHOST_SCSI_SET_ENDPOINT ioctl.
930          */
931         vs_tpg = vq->private_data;
932         if (!vs_tpg)
933                 goto out;
934
935         memset(&vc, 0, sizeof(vc));
936         vc.rsp_size = sizeof(struct virtio_scsi_cmd_resp);
937
938         vhost_disable_notify(&vs->dev, vq);
939
940         for (;;) {
941                 ret = vhost_scsi_get_desc(vs, vq, &vc);
942                 if (ret)
943                         goto err;
944
945                 /*
946                  * Setup pointers and values based upon different virtio-scsi
947                  * request header if T10_PI is enabled in KVM guest.
948                  */
949                 if (t10_pi) {
950                         vc.req = &v_req_pi;
951                         vc.req_size = sizeof(v_req_pi);
952                         vc.lunp = &v_req_pi.lun[0];
953                         vc.target = &v_req_pi.lun[1];
954                 } else {
955                         vc.req = &v_req;
956                         vc.req_size = sizeof(v_req);
957                         vc.lunp = &v_req.lun[0];
958                         vc.target = &v_req.lun[1];
959                 }
960
961                 /*
962                  * Validate the size of request and response buffers.
963                  * Check for a sane response buffer so we can report
964                  * early errors back to the guest.
965                  */
966                 ret = vhost_scsi_chk_size(vq, &vc);
967                 if (ret)
968                         goto err;
969
970                 ret = vhost_scsi_get_req(vq, &vc, &tpg);
971                 if (ret)
972                         goto err;
973
974                 ret = -EIO;     /* bad target on any error from here on */
975
976                 /*
977                  * Determine data_direction by calculating the total outgoing
978                  * iovec sizes + incoming iovec sizes vs. virtio-scsi request +
979                  * response headers respectively.
980                  *
981                  * For DMA_TO_DEVICE this is out_iter, which is already pointing
982                  * to the right place.
983                  *
984                  * For DMA_FROM_DEVICE, the iovec will be just past the end
985                  * of the virtio-scsi response header in either the same
986                  * or immediately following iovec.
987                  *
988                  * Any associated T10_PI bytes for the outgoing / incoming
989                  * payloads are included in calculation of exp_data_len here.
990                  */
991                 prot_bytes = 0;
992
993                 if (vc.out_size > vc.req_size) {
994                         data_direction = DMA_TO_DEVICE;
995                         exp_data_len = vc.out_size - vc.req_size;
996                         data_iter = vc.out_iter;
997                 } else if (vc.in_size > vc.rsp_size) {
998                         data_direction = DMA_FROM_DEVICE;
999                         exp_data_len = vc.in_size - vc.rsp_size;
1000
1001                         iov_iter_init(&in_iter, READ, &vq->iov[vc.out], vc.in,
1002                                       vc.rsp_size + exp_data_len);
1003                         iov_iter_advance(&in_iter, vc.rsp_size);
1004                         data_iter = in_iter;
1005                 } else {
1006                         data_direction = DMA_NONE;
1007                         exp_data_len = 0;
1008                 }
1009                 /*
1010                  * If T10_PI header + payload is present, setup prot_iter values
1011                  * and recalculate data_iter for vhost_scsi_mapal() mapping to
1012                  * host scatterlists via get_user_pages_fast().
1013                  */
1014                 if (t10_pi) {
1015                         if (v_req_pi.pi_bytesout) {
1016                                 if (data_direction != DMA_TO_DEVICE) {
1017                                         vq_err(vq, "Received non zero pi_bytesout,"
1018                                                 " but wrong data_direction\n");
1019                                         goto err;
1020                                 }
1021                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout);
1022                         } else if (v_req_pi.pi_bytesin) {
1023                                 if (data_direction != DMA_FROM_DEVICE) {
1024                                         vq_err(vq, "Received non zero pi_bytesin,"
1025                                                 " but wrong data_direction\n");
1026                                         goto err;
1027                                 }
1028                                 prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin);
1029                         }
1030                         /*
1031                          * Set prot_iter to data_iter and truncate it to
1032                          * prot_bytes, and advance data_iter past any
1033                          * preceeding prot_bytes that may be present.
1034                          *
1035                          * Also fix up the exp_data_len to reflect only the
1036                          * actual data payload length.
1037                          */
1038                         if (prot_bytes) {
1039                                 exp_data_len -= prot_bytes;
1040                                 prot_iter = data_iter;
1041                                 iov_iter_truncate(&prot_iter, prot_bytes);
1042                                 iov_iter_advance(&data_iter, prot_bytes);
1043                         }
1044                         tag = vhost64_to_cpu(vq, v_req_pi.tag);
1045                         task_attr = v_req_pi.task_attr;
1046                         cdb = &v_req_pi.cdb[0];
1047                         lun = ((v_req_pi.lun[2] << 8) | v_req_pi.lun[3]) & 0x3FFF;
1048                 } else {
1049                         tag = vhost64_to_cpu(vq, v_req.tag);
1050                         task_attr = v_req.task_attr;
1051                         cdb = &v_req.cdb[0];
1052                         lun = ((v_req.lun[2] << 8) | v_req.lun[3]) & 0x3FFF;
1053                 }
1054                 /*
1055                  * Check that the received CDB size does not exceeded our
1056                  * hardcoded max for vhost-scsi, then get a pre-allocated
1057                  * cmd descriptor for the new virtio-scsi tag.
1058                  *
1059                  * TODO what if cdb was too small for varlen cdb header?
1060                  */
1061                 if (unlikely(scsi_command_size(cdb) > VHOST_SCSI_MAX_CDB_SIZE)) {
1062                         vq_err(vq, "Received SCSI CDB with command_size: %d that"
1063                                 " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n",
1064                                 scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE);
1065                                 goto err;
1066                 }
1067                 cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr,
1068                                          exp_data_len + prot_bytes,
1069                                          data_direction);
1070                 if (IS_ERR(cmd)) {
1071                         vq_err(vq, "vhost_scsi_get_tag failed %ld\n",
1072                                PTR_ERR(cmd));
1073                         goto err;
1074                 }
1075                 cmd->tvc_vhost = vs;
1076                 cmd->tvc_vq = vq;
1077                 cmd->tvc_resp_iov = vq->iov[vc.out];
1078                 cmd->tvc_in_iovs = vc.in;
1079
1080                 pr_debug("vhost_scsi got command opcode: %#02x, lun: %d\n",
1081                          cmd->tvc_cdb[0], cmd->tvc_lun);
1082                 pr_debug("cmd: %p exp_data_len: %d, prot_bytes: %d data_direction:"
1083                          " %d\n", cmd, exp_data_len, prot_bytes, data_direction);
1084
1085                 if (data_direction != DMA_NONE) {
1086                         if (unlikely(vhost_scsi_mapal(cmd, prot_bytes,
1087                                                       &prot_iter, exp_data_len,
1088                                                       &data_iter))) {
1089                                 vq_err(vq, "Failed to map iov to sgl\n");
1090                                 vhost_scsi_release_cmd(&cmd->tvc_se_cmd);
1091                                 goto err;
1092                         }
1093                 }
1094                 /*
1095                  * Save the descriptor from vhost_get_vq_desc() to be used to
1096                  * complete the virtio-scsi request in TCM callback context via
1097                  * vhost_scsi_queue_data_in() and vhost_scsi_queue_status()
1098                  */
1099                 cmd->tvc_vq_desc = vc.head;
1100                 /*
1101                  * Dispatch cmd descriptor for cmwq execution in process
1102                  * context provided by vhost_scsi_workqueue.  This also ensures
1103                  * cmd is executed on the same kworker CPU as this vhost
1104                  * thread to gain positive L2 cache locality effects.
1105                  */
1106                 INIT_WORK(&cmd->work, vhost_scsi_submission_work);
1107                 queue_work(vhost_scsi_workqueue, &cmd->work);
1108                 ret = 0;
1109 err:
1110                 /*
1111                  * ENXIO:  No more requests, or read error, wait for next kick
1112                  * EINVAL: Invalid response buffer, drop the request
1113                  * EIO:    Respond with bad target
1114                  * EAGAIN: Pending request
1115                  */
1116                 if (ret == -ENXIO)
1117                         break;
1118                 else if (ret == -EIO)
1119                         vhost_scsi_send_bad_target(vs, vq, vc.head, vc.out);
1120         }
1121 out:
1122         mutex_unlock(&vq->mutex);
1123 }
1124
1125 static void
1126 vhost_scsi_send_tmf_reject(struct vhost_scsi *vs,
1127                            struct vhost_virtqueue *vq,
1128                            struct vhost_scsi_ctx *vc)
1129 {
1130         struct virtio_scsi_ctrl_tmf_resp __user *resp;
1131         struct virtio_scsi_ctrl_tmf_resp rsp;
1132         int ret;
1133
1134         pr_debug("%s\n", __func__);
1135         memset(&rsp, 0, sizeof(rsp));
1136         rsp.response = VIRTIO_SCSI_S_FUNCTION_REJECTED;
1137         resp = vq->iov[vc->out].iov_base;
1138         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
1139         if (!ret)
1140                 vhost_add_used_and_signal(&vs->dev, vq, vc->head, 0);
1141         else
1142                 pr_err("Faulted on virtio_scsi_ctrl_tmf_resp\n");
1143 }
1144
1145 static void
1146 vhost_scsi_send_an_resp(struct vhost_scsi *vs,
1147                         struct vhost_virtqueue *vq,
1148                         struct vhost_scsi_ctx *vc)
1149 {
1150         struct virtio_scsi_ctrl_an_resp __user *resp;
1151         struct virtio_scsi_ctrl_an_resp rsp;
1152         int ret;
1153
1154         pr_debug("%s\n", __func__);
1155         memset(&rsp, 0, sizeof(rsp));   /* event_actual = 0 */
1156         rsp.response = VIRTIO_SCSI_S_OK;
1157         resp = vq->iov[vc->out].iov_base;
1158         ret = __copy_to_user(resp, &rsp, sizeof(rsp));
1159         if (!ret)
1160                 vhost_add_used_and_signal(&vs->dev, vq, vc->head, 0);
1161         else
1162                 pr_err("Faulted on virtio_scsi_ctrl_an_resp\n");
1163 }
1164
1165 static void
1166 vhost_scsi_ctl_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq)
1167 {
1168         union {
1169                 __virtio32 type;
1170                 struct virtio_scsi_ctrl_an_req an;
1171                 struct virtio_scsi_ctrl_tmf_req tmf;
1172         } v_req;
1173         struct vhost_scsi_ctx vc;
1174         size_t typ_size;
1175         int ret;
1176
1177         mutex_lock(&vq->mutex);
1178         /*
1179          * We can handle the vq only after the endpoint is setup by calling the
1180          * VHOST_SCSI_SET_ENDPOINT ioctl.
1181          */
1182         if (!vq->private_data)
1183                 goto out;
1184
1185         memset(&vc, 0, sizeof(vc));
1186
1187         vhost_disable_notify(&vs->dev, vq);
1188
1189         for (;;) {
1190                 ret = vhost_scsi_get_desc(vs, vq, &vc);
1191                 if (ret)
1192                         goto err;
1193
1194                 /*
1195                  * Get the request type first in order to setup
1196                  * other parameters dependent on the type.
1197                  */
1198                 vc.req = &v_req.type;
1199                 typ_size = sizeof(v_req.type);
1200
1201                 if (unlikely(!copy_from_iter_full(vc.req, typ_size,
1202                                                   &vc.out_iter))) {
1203                         vq_err(vq, "Faulted on copy_from_iter tmf type\n");
1204                         /*
1205                          * The size of the response buffer depends on the
1206                          * request type and must be validated against it.
1207                          * Since the request type is not known, don't send
1208                          * a response.
1209                          */
1210                         continue;
1211                 }
1212
1213                 switch (v_req.type) {
1214                 case VIRTIO_SCSI_T_TMF:
1215                         vc.req = &v_req.tmf;
1216                         vc.req_size = sizeof(struct virtio_scsi_ctrl_tmf_req);
1217                         vc.rsp_size = sizeof(struct virtio_scsi_ctrl_tmf_resp);
1218                         vc.lunp = &v_req.tmf.lun[0];
1219                         vc.target = &v_req.tmf.lun[1];
1220                         break;
1221                 case VIRTIO_SCSI_T_AN_QUERY:
1222                 case VIRTIO_SCSI_T_AN_SUBSCRIBE:
1223                         vc.req = &v_req.an;
1224                         vc.req_size = sizeof(struct virtio_scsi_ctrl_an_req);
1225                         vc.rsp_size = sizeof(struct virtio_scsi_ctrl_an_resp);
1226                         vc.lunp = &v_req.an.lun[0];
1227                         vc.target = NULL;
1228                         break;
1229                 default:
1230                         vq_err(vq, "Unknown control request %d", v_req.type);
1231                         continue;
1232                 }
1233
1234                 /*
1235                  * Validate the size of request and response buffers.
1236                  * Check for a sane response buffer so we can report
1237                  * early errors back to the guest.
1238                  */
1239                 ret = vhost_scsi_chk_size(vq, &vc);
1240                 if (ret)
1241                         goto err;
1242
1243                 /*
1244                  * Get the rest of the request now that its size is known.
1245                  */
1246                 vc.req += typ_size;
1247                 vc.req_size -= typ_size;
1248
1249                 ret = vhost_scsi_get_req(vq, &vc, NULL);
1250                 if (ret)
1251                         goto err;
1252
1253                 if (v_req.type == VIRTIO_SCSI_T_TMF)
1254                         vhost_scsi_send_tmf_reject(vs, vq, &vc);
1255                 else
1256                         vhost_scsi_send_an_resp(vs, vq, &vc);
1257 err:
1258                 /*
1259                  * ENXIO:  No more requests, or read error, wait for next kick
1260                  * EINVAL: Invalid response buffer, drop the request
1261                  * EIO:    Respond with bad target
1262                  * EAGAIN: Pending request
1263                  */
1264                 if (ret == -ENXIO)
1265                         break;
1266                 else if (ret == -EIO)
1267                         vhost_scsi_send_bad_target(vs, vq, vc.head, vc.out);
1268         }
1269 out:
1270         mutex_unlock(&vq->mutex);
1271 }
1272
1273 static void vhost_scsi_ctl_handle_kick(struct vhost_work *work)
1274 {
1275         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1276                                                 poll.work);
1277         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1278
1279         pr_debug("%s: The handling func for control queue.\n", __func__);
1280         vhost_scsi_ctl_handle_vq(vs, vq);
1281 }
1282
1283 static void
1284 vhost_scsi_send_evt(struct vhost_scsi *vs,
1285                    struct vhost_scsi_tpg *tpg,
1286                    struct se_lun *lun,
1287                    u32 event,
1288                    u32 reason)
1289 {
1290         struct vhost_scsi_evt *evt;
1291
1292         evt = vhost_scsi_allocate_evt(vs, event, reason);
1293         if (!evt)
1294                 return;
1295
1296         if (tpg && lun) {
1297                 /* TODO: share lun setup code with virtio-scsi.ko */
1298                 /*
1299                  * Note: evt->event is zeroed when we allocate it and
1300                  * lun[4-7] need to be zero according to virtio-scsi spec.
1301                  */
1302                 evt->event.lun[0] = 0x01;
1303                 evt->event.lun[1] = tpg->tport_tpgt;
1304                 if (lun->unpacked_lun >= 256)
1305                         evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ;
1306                 evt->event.lun[3] = lun->unpacked_lun & 0xFF;
1307         }
1308
1309         llist_add(&evt->list, &vs->vs_event_list);
1310         vhost_work_queue(&vs->dev, &vs->vs_event_work);
1311 }
1312
1313 static void vhost_scsi_evt_handle_kick(struct vhost_work *work)
1314 {
1315         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1316                                                 poll.work);
1317         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1318
1319         mutex_lock(&vq->mutex);
1320         if (!vq->private_data)
1321                 goto out;
1322
1323         if (vs->vs_events_missed)
1324                 vhost_scsi_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
1325 out:
1326         mutex_unlock(&vq->mutex);
1327 }
1328
1329 static void vhost_scsi_handle_kick(struct vhost_work *work)
1330 {
1331         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
1332                                                 poll.work);
1333         struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev);
1334
1335         vhost_scsi_handle_vq(vs, vq);
1336 }
1337
1338 static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
1339 {
1340         vhost_poll_flush(&vs->vqs[index].vq.poll);
1341 }
1342
1343 /* Callers must hold dev mutex */
1344 static void vhost_scsi_flush(struct vhost_scsi *vs)
1345 {
1346         struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
1347         int i;
1348
1349         /* Init new inflight and remember the old inflight */
1350         vhost_scsi_init_inflight(vs, old_inflight);
1351
1352         /*
1353          * The inflight->kref was initialized to 1. We decrement it here to
1354          * indicate the start of the flush operation so that it will reach 0
1355          * when all the reqs are finished.
1356          */
1357         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1358                 kref_put(&old_inflight[i]->kref, vhost_scsi_done_inflight);
1359
1360         /* Flush both the vhost poll and vhost work */
1361         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1362                 vhost_scsi_flush_vq(vs, i);
1363         vhost_work_flush(&vs->dev, &vs->vs_completion_work);
1364         vhost_work_flush(&vs->dev, &vs->vs_event_work);
1365
1366         /* Wait for all reqs issued before the flush to be finished */
1367         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
1368                 wait_for_completion(&old_inflight[i]->comp);
1369 }
1370
1371 /*
1372  * Called from vhost_scsi_ioctl() context to walk the list of available
1373  * vhost_scsi_tpg with an active struct vhost_scsi_nexus
1374  *
1375  *  The lock nesting rule is:
1376  *    vhost_scsi_mutex -> vs->dev.mutex -> tpg->tv_tpg_mutex -> vq->mutex
1377  */
1378 static int
1379 vhost_scsi_set_endpoint(struct vhost_scsi *vs,
1380                         struct vhost_scsi_target *t)
1381 {
1382         struct se_portal_group *se_tpg;
1383         struct vhost_scsi_tport *tv_tport;
1384         struct vhost_scsi_tpg *tpg;
1385         struct vhost_scsi_tpg **vs_tpg;
1386         struct vhost_virtqueue *vq;
1387         int index, ret, i, len;
1388         bool match = false;
1389
1390         mutex_lock(&vhost_scsi_mutex);
1391         mutex_lock(&vs->dev.mutex);
1392
1393         /* Verify that ring has been setup correctly. */
1394         for (index = 0; index < vs->dev.nvqs; ++index) {
1395                 /* Verify that ring has been setup correctly. */
1396                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1397                         ret = -EFAULT;
1398                         goto out;
1399                 }
1400         }
1401
1402         len = sizeof(vs_tpg[0]) * VHOST_SCSI_MAX_TARGET;
1403         vs_tpg = kzalloc(len, GFP_KERNEL);
1404         if (!vs_tpg) {
1405                 ret = -ENOMEM;
1406                 goto out;
1407         }
1408         if (vs->vs_tpg)
1409                 memcpy(vs_tpg, vs->vs_tpg, len);
1410
1411         list_for_each_entry(tpg, &vhost_scsi_list, tv_tpg_list) {
1412                 mutex_lock(&tpg->tv_tpg_mutex);
1413                 if (!tpg->tpg_nexus) {
1414                         mutex_unlock(&tpg->tv_tpg_mutex);
1415                         continue;
1416                 }
1417                 if (tpg->tv_tpg_vhost_count != 0) {
1418                         mutex_unlock(&tpg->tv_tpg_mutex);
1419                         continue;
1420                 }
1421                 tv_tport = tpg->tport;
1422
1423                 if (!strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1424                         if (vs->vs_tpg && vs->vs_tpg[tpg->tport_tpgt]) {
1425                                 kfree(vs_tpg);
1426                                 mutex_unlock(&tpg->tv_tpg_mutex);
1427                                 ret = -EEXIST;
1428                                 goto out;
1429                         }
1430                         /*
1431                          * In order to ensure individual vhost-scsi configfs
1432                          * groups cannot be removed while in use by vhost ioctl,
1433                          * go ahead and take an explicit se_tpg->tpg_group.cg_item
1434                          * dependency now.
1435                          */
1436                         se_tpg = &tpg->se_tpg;
1437                         ret = target_depend_item(&se_tpg->tpg_group.cg_item);
1438                         if (ret) {
1439                                 pr_warn("configfs_depend_item() failed: %d\n", ret);
1440                                 kfree(vs_tpg);
1441                                 mutex_unlock(&tpg->tv_tpg_mutex);
1442                                 goto out;
1443                         }
1444                         tpg->tv_tpg_vhost_count++;
1445                         tpg->vhost_scsi = vs;
1446                         vs_tpg[tpg->tport_tpgt] = tpg;
1447                         smp_mb__after_atomic();
1448                         match = true;
1449                 }
1450                 mutex_unlock(&tpg->tv_tpg_mutex);
1451         }
1452
1453         if (match) {
1454                 memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
1455                        sizeof(vs->vs_vhost_wwpn));
1456                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1457                         vq = &vs->vqs[i].vq;
1458                         mutex_lock(&vq->mutex);
1459                         vq->private_data = vs_tpg;
1460                         vhost_vq_init_access(vq);
1461                         mutex_unlock(&vq->mutex);
1462                 }
1463                 ret = 0;
1464         } else {
1465                 ret = -EEXIST;
1466         }
1467
1468         /*
1469          * Act as synchronize_rcu to make sure access to
1470          * old vs->vs_tpg is finished.
1471          */
1472         vhost_scsi_flush(vs);
1473         kfree(vs->vs_tpg);
1474         vs->vs_tpg = vs_tpg;
1475
1476 out:
1477         mutex_unlock(&vs->dev.mutex);
1478         mutex_unlock(&vhost_scsi_mutex);
1479         return ret;
1480 }
1481
1482 static int
1483 vhost_scsi_clear_endpoint(struct vhost_scsi *vs,
1484                           struct vhost_scsi_target *t)
1485 {
1486         struct se_portal_group *se_tpg;
1487         struct vhost_scsi_tport *tv_tport;
1488         struct vhost_scsi_tpg *tpg;
1489         struct vhost_virtqueue *vq;
1490         bool match = false;
1491         int index, ret, i;
1492         u8 target;
1493
1494         mutex_lock(&vhost_scsi_mutex);
1495         mutex_lock(&vs->dev.mutex);
1496         /* Verify that ring has been setup correctly. */
1497         for (index = 0; index < vs->dev.nvqs; ++index) {
1498                 if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
1499                         ret = -EFAULT;
1500                         goto err_dev;
1501                 }
1502         }
1503
1504         if (!vs->vs_tpg) {
1505                 ret = 0;
1506                 goto err_dev;
1507         }
1508
1509         for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
1510                 target = i;
1511                 tpg = vs->vs_tpg[target];
1512                 if (!tpg)
1513                         continue;
1514
1515                 mutex_lock(&tpg->tv_tpg_mutex);
1516                 tv_tport = tpg->tport;
1517                 if (!tv_tport) {
1518                         ret = -ENODEV;
1519                         goto err_tpg;
1520                 }
1521
1522                 if (strcmp(tv_tport->tport_name, t->vhost_wwpn)) {
1523                         pr_warn("tv_tport->tport_name: %s, tpg->tport_tpgt: %hu"
1524                                 " does not match t->vhost_wwpn: %s, t->vhost_tpgt: %hu\n",
1525                                 tv_tport->tport_name, tpg->tport_tpgt,
1526                                 t->vhost_wwpn, t->vhost_tpgt);
1527                         ret = -EINVAL;
1528                         goto err_tpg;
1529                 }
1530                 tpg->tv_tpg_vhost_count--;
1531                 tpg->vhost_scsi = NULL;
1532                 vs->vs_tpg[target] = NULL;
1533                 match = true;
1534                 mutex_unlock(&tpg->tv_tpg_mutex);
1535                 /*
1536                  * Release se_tpg->tpg_group.cg_item configfs dependency now
1537                  * to allow vhost-scsi WWPN se_tpg->tpg_group shutdown to occur.
1538                  */
1539                 se_tpg = &tpg->se_tpg;
1540                 target_undepend_item(&se_tpg->tpg_group.cg_item);
1541         }
1542         if (match) {
1543                 for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1544                         vq = &vs->vqs[i].vq;
1545                         mutex_lock(&vq->mutex);
1546                         vq->private_data = NULL;
1547                         mutex_unlock(&vq->mutex);
1548                 }
1549         }
1550         /*
1551          * Act as synchronize_rcu to make sure access to
1552          * old vs->vs_tpg is finished.
1553          */
1554         vhost_scsi_flush(vs);
1555         kfree(vs->vs_tpg);
1556         vs->vs_tpg = NULL;
1557         WARN_ON(vs->vs_events_nr);
1558         mutex_unlock(&vs->dev.mutex);
1559         mutex_unlock(&vhost_scsi_mutex);
1560         return 0;
1561
1562 err_tpg:
1563         mutex_unlock(&tpg->tv_tpg_mutex);
1564 err_dev:
1565         mutex_unlock(&vs->dev.mutex);
1566         mutex_unlock(&vhost_scsi_mutex);
1567         return ret;
1568 }
1569
1570 static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
1571 {
1572         struct vhost_virtqueue *vq;
1573         int i;
1574
1575         if (features & ~VHOST_SCSI_FEATURES)
1576                 return -EOPNOTSUPP;
1577
1578         mutex_lock(&vs->dev.mutex);
1579         if ((features & (1 << VHOST_F_LOG_ALL)) &&
1580             !vhost_log_access_ok(&vs->dev)) {
1581                 mutex_unlock(&vs->dev.mutex);
1582                 return -EFAULT;
1583         }
1584
1585         for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
1586                 vq = &vs->vqs[i].vq;
1587                 mutex_lock(&vq->mutex);
1588                 vq->acked_features = features;
1589                 mutex_unlock(&vq->mutex);
1590         }
1591         mutex_unlock(&vs->dev.mutex);
1592         return 0;
1593 }
1594
1595 static int vhost_scsi_open(struct inode *inode, struct file *f)
1596 {
1597         struct vhost_scsi *vs;
1598         struct vhost_virtqueue **vqs;
1599         int r = -ENOMEM, i;
1600
1601         vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN | __GFP_RETRY_MAYFAIL);
1602         if (!vs) {
1603                 vs = vzalloc(sizeof(*vs));
1604                 if (!vs)
1605                         goto err_vs;
1606         }
1607
1608         vqs = kmalloc_array(VHOST_SCSI_MAX_VQ, sizeof(*vqs), GFP_KERNEL);
1609         if (!vqs)
1610                 goto err_vqs;
1611
1612         vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
1613         vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
1614
1615         vs->vs_events_nr = 0;
1616         vs->vs_events_missed = false;
1617
1618         vqs[VHOST_SCSI_VQ_CTL] = &vs->vqs[VHOST_SCSI_VQ_CTL].vq;
1619         vqs[VHOST_SCSI_VQ_EVT] = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1620         vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
1621         vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
1622         for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
1623                 vqs[i] = &vs->vqs[i].vq;
1624                 vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
1625         }
1626         vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ);
1627
1628         vhost_scsi_init_inflight(vs, NULL);
1629
1630         f->private_data = vs;
1631         return 0;
1632
1633 err_vqs:
1634         kvfree(vs);
1635 err_vs:
1636         return r;
1637 }
1638
1639 static int vhost_scsi_release(struct inode *inode, struct file *f)
1640 {
1641         struct vhost_scsi *vs = f->private_data;
1642         struct vhost_scsi_target t;
1643
1644         mutex_lock(&vs->dev.mutex);
1645         memcpy(t.vhost_wwpn, vs->vs_vhost_wwpn, sizeof(t.vhost_wwpn));
1646         mutex_unlock(&vs->dev.mutex);
1647         vhost_scsi_clear_endpoint(vs, &t);
1648         vhost_dev_stop(&vs->dev);
1649         vhost_dev_cleanup(&vs->dev);
1650         /* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
1651         vhost_scsi_flush(vs);
1652         kfree(vs->dev.vqs);
1653         kvfree(vs);
1654         return 0;
1655 }
1656
1657 static long
1658 vhost_scsi_ioctl(struct file *f,
1659                  unsigned int ioctl,
1660                  unsigned long arg)
1661 {
1662         struct vhost_scsi *vs = f->private_data;
1663         struct vhost_scsi_target backend;
1664         void __user *argp = (void __user *)arg;
1665         u64 __user *featurep = argp;
1666         u32 __user *eventsp = argp;
1667         u32 events_missed;
1668         u64 features;
1669         int r, abi_version = VHOST_SCSI_ABI_VERSION;
1670         struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1671
1672         switch (ioctl) {
1673         case VHOST_SCSI_SET_ENDPOINT:
1674                 if (copy_from_user(&backend, argp, sizeof backend))
1675                         return -EFAULT;
1676                 if (backend.reserved != 0)
1677                         return -EOPNOTSUPP;
1678
1679                 return vhost_scsi_set_endpoint(vs, &backend);
1680         case VHOST_SCSI_CLEAR_ENDPOINT:
1681                 if (copy_from_user(&backend, argp, sizeof backend))
1682                         return -EFAULT;
1683                 if (backend.reserved != 0)
1684                         return -EOPNOTSUPP;
1685
1686                 return vhost_scsi_clear_endpoint(vs, &backend);
1687         case VHOST_SCSI_GET_ABI_VERSION:
1688                 if (copy_to_user(argp, &abi_version, sizeof abi_version))
1689                         return -EFAULT;
1690                 return 0;
1691         case VHOST_SCSI_SET_EVENTS_MISSED:
1692                 if (get_user(events_missed, eventsp))
1693                         return -EFAULT;
1694                 mutex_lock(&vq->mutex);
1695                 vs->vs_events_missed = events_missed;
1696                 mutex_unlock(&vq->mutex);
1697                 return 0;
1698         case VHOST_SCSI_GET_EVENTS_MISSED:
1699                 mutex_lock(&vq->mutex);
1700                 events_missed = vs->vs_events_missed;
1701                 mutex_unlock(&vq->mutex);
1702                 if (put_user(events_missed, eventsp))
1703                         return -EFAULT;
1704                 return 0;
1705         case VHOST_GET_FEATURES:
1706                 features = VHOST_SCSI_FEATURES;
1707                 if (copy_to_user(featurep, &features, sizeof features))
1708                         return -EFAULT;
1709                 return 0;
1710         case VHOST_SET_FEATURES:
1711                 if (copy_from_user(&features, featurep, sizeof features))
1712                         return -EFAULT;
1713                 return vhost_scsi_set_features(vs, features);
1714         default:
1715                 mutex_lock(&vs->dev.mutex);
1716                 r = vhost_dev_ioctl(&vs->dev, ioctl, argp);
1717                 /* TODO: flush backend after dev ioctl. */
1718                 if (r == -ENOIOCTLCMD)
1719                         r = vhost_vring_ioctl(&vs->dev, ioctl, argp);
1720                 mutex_unlock(&vs->dev.mutex);
1721                 return r;
1722         }
1723 }
1724
1725 #ifdef CONFIG_COMPAT
1726 static long vhost_scsi_compat_ioctl(struct file *f, unsigned int ioctl,
1727                                 unsigned long arg)
1728 {
1729         return vhost_scsi_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
1730 }
1731 #endif
1732
1733 static const struct file_operations vhost_scsi_fops = {
1734         .owner          = THIS_MODULE,
1735         .release        = vhost_scsi_release,
1736         .unlocked_ioctl = vhost_scsi_ioctl,
1737 #ifdef CONFIG_COMPAT
1738         .compat_ioctl   = vhost_scsi_compat_ioctl,
1739 #endif
1740         .open           = vhost_scsi_open,
1741         .llseek         = noop_llseek,
1742 };
1743
1744 static struct miscdevice vhost_scsi_misc = {
1745         MISC_DYNAMIC_MINOR,
1746         "vhost-scsi",
1747         &vhost_scsi_fops,
1748 };
1749
1750 static int __init vhost_scsi_register(void)
1751 {
1752         return misc_register(&vhost_scsi_misc);
1753 }
1754
1755 static void vhost_scsi_deregister(void)
1756 {
1757         misc_deregister(&vhost_scsi_misc);
1758 }
1759
1760 static char *vhost_scsi_dump_proto_id(struct vhost_scsi_tport *tport)
1761 {
1762         switch (tport->tport_proto_id) {
1763         case SCSI_PROTOCOL_SAS:
1764                 return "SAS";
1765         case SCSI_PROTOCOL_FCP:
1766                 return "FCP";
1767         case SCSI_PROTOCOL_ISCSI:
1768                 return "iSCSI";
1769         default:
1770                 break;
1771         }
1772
1773         return "Unknown";
1774 }
1775
1776 static void
1777 vhost_scsi_do_plug(struct vhost_scsi_tpg *tpg,
1778                   struct se_lun *lun, bool plug)
1779 {
1780
1781         struct vhost_scsi *vs = tpg->vhost_scsi;
1782         struct vhost_virtqueue *vq;
1783         u32 reason;
1784
1785         if (!vs)
1786                 return;
1787
1788         mutex_lock(&vs->dev.mutex);
1789
1790         if (plug)
1791                 reason = VIRTIO_SCSI_EVT_RESET_RESCAN;
1792         else
1793                 reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
1794
1795         vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
1796         mutex_lock(&vq->mutex);
1797         if (vhost_has_feature(vq, VIRTIO_SCSI_F_HOTPLUG))
1798                 vhost_scsi_send_evt(vs, tpg, lun,
1799                                    VIRTIO_SCSI_T_TRANSPORT_RESET, reason);
1800         mutex_unlock(&vq->mutex);
1801         mutex_unlock(&vs->dev.mutex);
1802 }
1803
1804 static void vhost_scsi_hotplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1805 {
1806         vhost_scsi_do_plug(tpg, lun, true);
1807 }
1808
1809 static void vhost_scsi_hotunplug(struct vhost_scsi_tpg *tpg, struct se_lun *lun)
1810 {
1811         vhost_scsi_do_plug(tpg, lun, false);
1812 }
1813
1814 static int vhost_scsi_port_link(struct se_portal_group *se_tpg,
1815                                struct se_lun *lun)
1816 {
1817         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1818                                 struct vhost_scsi_tpg, se_tpg);
1819
1820         mutex_lock(&vhost_scsi_mutex);
1821
1822         mutex_lock(&tpg->tv_tpg_mutex);
1823         tpg->tv_tpg_port_count++;
1824         mutex_unlock(&tpg->tv_tpg_mutex);
1825
1826         vhost_scsi_hotplug(tpg, lun);
1827
1828         mutex_unlock(&vhost_scsi_mutex);
1829
1830         return 0;
1831 }
1832
1833 static void vhost_scsi_port_unlink(struct se_portal_group *se_tpg,
1834                                   struct se_lun *lun)
1835 {
1836         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1837                                 struct vhost_scsi_tpg, se_tpg);
1838
1839         mutex_lock(&vhost_scsi_mutex);
1840
1841         mutex_lock(&tpg->tv_tpg_mutex);
1842         tpg->tv_tpg_port_count--;
1843         mutex_unlock(&tpg->tv_tpg_mutex);
1844
1845         vhost_scsi_hotunplug(tpg, lun);
1846
1847         mutex_unlock(&vhost_scsi_mutex);
1848 }
1849
1850 static void vhost_scsi_free_cmd_map_res(struct se_session *se_sess)
1851 {
1852         struct vhost_scsi_cmd *tv_cmd;
1853         unsigned int i;
1854
1855         if (!se_sess->sess_cmd_map)
1856                 return;
1857
1858         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1859                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1860
1861                 kfree(tv_cmd->tvc_sgl);
1862                 kfree(tv_cmd->tvc_prot_sgl);
1863                 kfree(tv_cmd->tvc_upages);
1864         }
1865 }
1866
1867 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_store(
1868                 struct config_item *item, const char *page, size_t count)
1869 {
1870         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1871         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1872                                 struct vhost_scsi_tpg, se_tpg);
1873         unsigned long val;
1874         int ret = kstrtoul(page, 0, &val);
1875
1876         if (ret) {
1877                 pr_err("kstrtoul() returned %d for fabric_prot_type\n", ret);
1878                 return ret;
1879         }
1880         if (val != 0 && val != 1 && val != 3) {
1881                 pr_err("Invalid vhost_scsi fabric_prot_type: %lu\n", val);
1882                 return -EINVAL;
1883         }
1884         tpg->tv_fabric_prot_type = val;
1885
1886         return count;
1887 }
1888
1889 static ssize_t vhost_scsi_tpg_attrib_fabric_prot_type_show(
1890                 struct config_item *item, char *page)
1891 {
1892         struct se_portal_group *se_tpg = attrib_to_tpg(item);
1893         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
1894                                 struct vhost_scsi_tpg, se_tpg);
1895
1896         return sprintf(page, "%d\n", tpg->tv_fabric_prot_type);
1897 }
1898
1899 CONFIGFS_ATTR(vhost_scsi_tpg_attrib_, fabric_prot_type);
1900
1901 static struct configfs_attribute *vhost_scsi_tpg_attrib_attrs[] = {
1902         &vhost_scsi_tpg_attrib_attr_fabric_prot_type,
1903         NULL,
1904 };
1905
1906 static int vhost_scsi_nexus_cb(struct se_portal_group *se_tpg,
1907                                struct se_session *se_sess, void *p)
1908 {
1909         struct vhost_scsi_cmd *tv_cmd;
1910         unsigned int i;
1911
1912         for (i = 0; i < VHOST_SCSI_DEFAULT_TAGS; i++) {
1913                 tv_cmd = &((struct vhost_scsi_cmd *)se_sess->sess_cmd_map)[i];
1914
1915                 tv_cmd->tvc_sgl = kcalloc(VHOST_SCSI_PREALLOC_SGLS,
1916                                           sizeof(struct scatterlist),
1917                                           GFP_KERNEL);
1918                 if (!tv_cmd->tvc_sgl) {
1919                         pr_err("Unable to allocate tv_cmd->tvc_sgl\n");
1920                         goto out;
1921                 }
1922
1923                 tv_cmd->tvc_upages = kcalloc(VHOST_SCSI_PREALLOC_UPAGES,
1924                                              sizeof(struct page *),
1925                                              GFP_KERNEL);
1926                 if (!tv_cmd->tvc_upages) {
1927                         pr_err("Unable to allocate tv_cmd->tvc_upages\n");
1928                         goto out;
1929                 }
1930
1931                 tv_cmd->tvc_prot_sgl = kcalloc(VHOST_SCSI_PREALLOC_PROT_SGLS,
1932                                                sizeof(struct scatterlist),
1933                                                GFP_KERNEL);
1934                 if (!tv_cmd->tvc_prot_sgl) {
1935                         pr_err("Unable to allocate tv_cmd->tvc_prot_sgl\n");
1936                         goto out;
1937                 }
1938         }
1939         return 0;
1940 out:
1941         vhost_scsi_free_cmd_map_res(se_sess);
1942         return -ENOMEM;
1943 }
1944
1945 static int vhost_scsi_make_nexus(struct vhost_scsi_tpg *tpg,
1946                                 const char *name)
1947 {
1948         struct vhost_scsi_nexus *tv_nexus;
1949
1950         mutex_lock(&tpg->tv_tpg_mutex);
1951         if (tpg->tpg_nexus) {
1952                 mutex_unlock(&tpg->tv_tpg_mutex);
1953                 pr_debug("tpg->tpg_nexus already exists\n");
1954                 return -EEXIST;
1955         }
1956
1957         tv_nexus = kzalloc(sizeof(*tv_nexus), GFP_KERNEL);
1958         if (!tv_nexus) {
1959                 mutex_unlock(&tpg->tv_tpg_mutex);
1960                 pr_err("Unable to allocate struct vhost_scsi_nexus\n");
1961                 return -ENOMEM;
1962         }
1963         /*
1964          * Since we are running in 'demo mode' this call with generate a
1965          * struct se_node_acl for the vhost_scsi struct se_portal_group with
1966          * the SCSI Initiator port name of the passed configfs group 'name'.
1967          */
1968         tv_nexus->tvn_se_sess = target_setup_session(&tpg->se_tpg,
1969                                         VHOST_SCSI_DEFAULT_TAGS,
1970                                         sizeof(struct vhost_scsi_cmd),
1971                                         TARGET_PROT_DIN_PASS | TARGET_PROT_DOUT_PASS,
1972                                         (unsigned char *)name, tv_nexus,
1973                                         vhost_scsi_nexus_cb);
1974         if (IS_ERR(tv_nexus->tvn_se_sess)) {
1975                 mutex_unlock(&tpg->tv_tpg_mutex);
1976                 kfree(tv_nexus);
1977                 return -ENOMEM;
1978         }
1979         tpg->tpg_nexus = tv_nexus;
1980
1981         mutex_unlock(&tpg->tv_tpg_mutex);
1982         return 0;
1983 }
1984
1985 static int vhost_scsi_drop_nexus(struct vhost_scsi_tpg *tpg)
1986 {
1987         struct se_session *se_sess;
1988         struct vhost_scsi_nexus *tv_nexus;
1989
1990         mutex_lock(&tpg->tv_tpg_mutex);
1991         tv_nexus = tpg->tpg_nexus;
1992         if (!tv_nexus) {
1993                 mutex_unlock(&tpg->tv_tpg_mutex);
1994                 return -ENODEV;
1995         }
1996
1997         se_sess = tv_nexus->tvn_se_sess;
1998         if (!se_sess) {
1999                 mutex_unlock(&tpg->tv_tpg_mutex);
2000                 return -ENODEV;
2001         }
2002
2003         if (tpg->tv_tpg_port_count != 0) {
2004                 mutex_unlock(&tpg->tv_tpg_mutex);
2005                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
2006                         " active TPG port count: %d\n",
2007                         tpg->tv_tpg_port_count);
2008                 return -EBUSY;
2009         }
2010
2011         if (tpg->tv_tpg_vhost_count != 0) {
2012                 mutex_unlock(&tpg->tv_tpg_mutex);
2013                 pr_err("Unable to remove TCM_vhost I_T Nexus with"
2014                         " active TPG vhost count: %d\n",
2015                         tpg->tv_tpg_vhost_count);
2016                 return -EBUSY;
2017         }
2018
2019         pr_debug("TCM_vhost_ConfigFS: Removing I_T Nexus to emulated"
2020                 " %s Initiator Port: %s\n", vhost_scsi_dump_proto_id(tpg->tport),
2021                 tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
2022
2023         vhost_scsi_free_cmd_map_res(se_sess);
2024         /*
2025          * Release the SCSI I_T Nexus to the emulated vhost Target Port
2026          */
2027         target_remove_session(se_sess);
2028         tpg->tpg_nexus = NULL;
2029         mutex_unlock(&tpg->tv_tpg_mutex);
2030
2031         kfree(tv_nexus);
2032         return 0;
2033 }
2034
2035 static ssize_t vhost_scsi_tpg_nexus_show(struct config_item *item, char *page)
2036 {
2037         struct se_portal_group *se_tpg = to_tpg(item);
2038         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2039                                 struct vhost_scsi_tpg, se_tpg);
2040         struct vhost_scsi_nexus *tv_nexus;
2041         ssize_t ret;
2042
2043         mutex_lock(&tpg->tv_tpg_mutex);
2044         tv_nexus = tpg->tpg_nexus;
2045         if (!tv_nexus) {
2046                 mutex_unlock(&tpg->tv_tpg_mutex);
2047                 return -ENODEV;
2048         }
2049         ret = snprintf(page, PAGE_SIZE, "%s\n",
2050                         tv_nexus->tvn_se_sess->se_node_acl->initiatorname);
2051         mutex_unlock(&tpg->tv_tpg_mutex);
2052
2053         return ret;
2054 }
2055
2056 static ssize_t vhost_scsi_tpg_nexus_store(struct config_item *item,
2057                 const char *page, size_t count)
2058 {
2059         struct se_portal_group *se_tpg = to_tpg(item);
2060         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2061                                 struct vhost_scsi_tpg, se_tpg);
2062         struct vhost_scsi_tport *tport_wwn = tpg->tport;
2063         unsigned char i_port[VHOST_SCSI_NAMELEN], *ptr, *port_ptr;
2064         int ret;
2065         /*
2066          * Shutdown the active I_T nexus if 'NULL' is passed..
2067          */
2068         if (!strncmp(page, "NULL", 4)) {
2069                 ret = vhost_scsi_drop_nexus(tpg);
2070                 return (!ret) ? count : ret;
2071         }
2072         /*
2073          * Otherwise make sure the passed virtual Initiator port WWN matches
2074          * the fabric protocol_id set in vhost_scsi_make_tport(), and call
2075          * vhost_scsi_make_nexus().
2076          */
2077         if (strlen(page) >= VHOST_SCSI_NAMELEN) {
2078                 pr_err("Emulated NAA Sas Address: %s, exceeds"
2079                                 " max: %d\n", page, VHOST_SCSI_NAMELEN);
2080                 return -EINVAL;
2081         }
2082         snprintf(&i_port[0], VHOST_SCSI_NAMELEN, "%s", page);
2083
2084         ptr = strstr(i_port, "naa.");
2085         if (ptr) {
2086                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_SAS) {
2087                         pr_err("Passed SAS Initiator Port %s does not"
2088                                 " match target port protoid: %s\n", i_port,
2089                                 vhost_scsi_dump_proto_id(tport_wwn));
2090                         return -EINVAL;
2091                 }
2092                 port_ptr = &i_port[0];
2093                 goto check_newline;
2094         }
2095         ptr = strstr(i_port, "fc.");
2096         if (ptr) {
2097                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_FCP) {
2098                         pr_err("Passed FCP Initiator Port %s does not"
2099                                 " match target port protoid: %s\n", i_port,
2100                                 vhost_scsi_dump_proto_id(tport_wwn));
2101                         return -EINVAL;
2102                 }
2103                 port_ptr = &i_port[3]; /* Skip over "fc." */
2104                 goto check_newline;
2105         }
2106         ptr = strstr(i_port, "iqn.");
2107         if (ptr) {
2108                 if (tport_wwn->tport_proto_id != SCSI_PROTOCOL_ISCSI) {
2109                         pr_err("Passed iSCSI Initiator Port %s does not"
2110                                 " match target port protoid: %s\n", i_port,
2111                                 vhost_scsi_dump_proto_id(tport_wwn));
2112                         return -EINVAL;
2113                 }
2114                 port_ptr = &i_port[0];
2115                 goto check_newline;
2116         }
2117         pr_err("Unable to locate prefix for emulated Initiator Port:"
2118                         " %s\n", i_port);
2119         return -EINVAL;
2120         /*
2121          * Clear any trailing newline for the NAA WWN
2122          */
2123 check_newline:
2124         if (i_port[strlen(i_port)-1] == '\n')
2125                 i_port[strlen(i_port)-1] = '\0';
2126
2127         ret = vhost_scsi_make_nexus(tpg, port_ptr);
2128         if (ret < 0)
2129                 return ret;
2130
2131         return count;
2132 }
2133
2134 CONFIGFS_ATTR(vhost_scsi_tpg_, nexus);
2135
2136 static struct configfs_attribute *vhost_scsi_tpg_attrs[] = {
2137         &vhost_scsi_tpg_attr_nexus,
2138         NULL,
2139 };
2140
2141 static struct se_portal_group *
2142 vhost_scsi_make_tpg(struct se_wwn *wwn, const char *name)
2143 {
2144         struct vhost_scsi_tport *tport = container_of(wwn,
2145                         struct vhost_scsi_tport, tport_wwn);
2146
2147         struct vhost_scsi_tpg *tpg;
2148         u16 tpgt;
2149         int ret;
2150
2151         if (strstr(name, "tpgt_") != name)
2152                 return ERR_PTR(-EINVAL);
2153         if (kstrtou16(name + 5, 10, &tpgt) || tpgt >= VHOST_SCSI_MAX_TARGET)
2154                 return ERR_PTR(-EINVAL);
2155
2156         tpg = kzalloc(sizeof(*tpg), GFP_KERNEL);
2157         if (!tpg) {
2158                 pr_err("Unable to allocate struct vhost_scsi_tpg");
2159                 return ERR_PTR(-ENOMEM);
2160         }
2161         mutex_init(&tpg->tv_tpg_mutex);
2162         INIT_LIST_HEAD(&tpg->tv_tpg_list);
2163         tpg->tport = tport;
2164         tpg->tport_tpgt = tpgt;
2165
2166         ret = core_tpg_register(wwn, &tpg->se_tpg, tport->tport_proto_id);
2167         if (ret < 0) {
2168                 kfree(tpg);
2169                 return NULL;
2170         }
2171         mutex_lock(&vhost_scsi_mutex);
2172         list_add_tail(&tpg->tv_tpg_list, &vhost_scsi_list);
2173         mutex_unlock(&vhost_scsi_mutex);
2174
2175         return &tpg->se_tpg;
2176 }
2177
2178 static void vhost_scsi_drop_tpg(struct se_portal_group *se_tpg)
2179 {
2180         struct vhost_scsi_tpg *tpg = container_of(se_tpg,
2181                                 struct vhost_scsi_tpg, se_tpg);
2182
2183         mutex_lock(&vhost_scsi_mutex);
2184         list_del(&tpg->tv_tpg_list);
2185         mutex_unlock(&vhost_scsi_mutex);
2186         /*
2187          * Release the virtual I_T Nexus for this vhost TPG
2188          */
2189         vhost_scsi_drop_nexus(tpg);
2190         /*
2191          * Deregister the se_tpg from TCM..
2192          */
2193         core_tpg_deregister(se_tpg);
2194         kfree(tpg);
2195 }
2196
2197 static struct se_wwn *
2198 vhost_scsi_make_tport(struct target_fabric_configfs *tf,
2199                      struct config_group *group,
2200                      const char *name)
2201 {
2202         struct vhost_scsi_tport *tport;
2203         char *ptr;
2204         u64 wwpn = 0;
2205         int off = 0;
2206
2207         /* if (vhost_scsi_parse_wwn(name, &wwpn, 1) < 0)
2208                 return ERR_PTR(-EINVAL); */
2209
2210         tport = kzalloc(sizeof(*tport), GFP_KERNEL);
2211         if (!tport) {
2212                 pr_err("Unable to allocate struct vhost_scsi_tport");
2213                 return ERR_PTR(-ENOMEM);
2214         }
2215         tport->tport_wwpn = wwpn;
2216         /*
2217          * Determine the emulated Protocol Identifier and Target Port Name
2218          * based on the incoming configfs directory name.
2219          */
2220         ptr = strstr(name, "naa.");
2221         if (ptr) {
2222                 tport->tport_proto_id = SCSI_PROTOCOL_SAS;
2223                 goto check_len;
2224         }
2225         ptr = strstr(name, "fc.");
2226         if (ptr) {
2227                 tport->tport_proto_id = SCSI_PROTOCOL_FCP;
2228                 off = 3; /* Skip over "fc." */
2229                 goto check_len;
2230         }
2231         ptr = strstr(name, "iqn.");
2232         if (ptr) {
2233                 tport->tport_proto_id = SCSI_PROTOCOL_ISCSI;
2234                 goto check_len;
2235         }
2236
2237         pr_err("Unable to locate prefix for emulated Target Port:"
2238                         " %s\n", name);
2239         kfree(tport);
2240         return ERR_PTR(-EINVAL);
2241
2242 check_len:
2243         if (strlen(name) >= VHOST_SCSI_NAMELEN) {
2244                 pr_err("Emulated %s Address: %s, exceeds"
2245                         " max: %d\n", name, vhost_scsi_dump_proto_id(tport),
2246                         VHOST_SCSI_NAMELEN);
2247                 kfree(tport);
2248                 return ERR_PTR(-EINVAL);
2249         }
2250         snprintf(&tport->tport_name[0], VHOST_SCSI_NAMELEN, "%s", &name[off]);
2251
2252         pr_debug("TCM_VHost_ConfigFS: Allocated emulated Target"
2253                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport), name);
2254
2255         return &tport->tport_wwn;
2256 }
2257
2258 static void vhost_scsi_drop_tport(struct se_wwn *wwn)
2259 {
2260         struct vhost_scsi_tport *tport = container_of(wwn,
2261                                 struct vhost_scsi_tport, tport_wwn);
2262
2263         pr_debug("TCM_VHost_ConfigFS: Deallocating emulated Target"
2264                 " %s Address: %s\n", vhost_scsi_dump_proto_id(tport),
2265                 tport->tport_name);
2266
2267         kfree(tport);
2268 }
2269
2270 static ssize_t
2271 vhost_scsi_wwn_version_show(struct config_item *item, char *page)
2272 {
2273         return sprintf(page, "TCM_VHOST fabric module %s on %s/%s"
2274                 "on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2275                 utsname()->machine);
2276 }
2277
2278 CONFIGFS_ATTR_RO(vhost_scsi_wwn_, version);
2279
2280 static struct configfs_attribute *vhost_scsi_wwn_attrs[] = {
2281         &vhost_scsi_wwn_attr_version,
2282         NULL,
2283 };
2284
2285 static const struct target_core_fabric_ops vhost_scsi_ops = {
2286         .module                         = THIS_MODULE,
2287         .fabric_name                    = "vhost",
2288         .tpg_get_wwn                    = vhost_scsi_get_fabric_wwn,
2289         .tpg_get_tag                    = vhost_scsi_get_tpgt,
2290         .tpg_check_demo_mode            = vhost_scsi_check_true,
2291         .tpg_check_demo_mode_cache      = vhost_scsi_check_true,
2292         .tpg_check_demo_mode_write_protect = vhost_scsi_check_false,
2293         .tpg_check_prod_mode_write_protect = vhost_scsi_check_false,
2294         .tpg_check_prot_fabric_only     = vhost_scsi_check_prot_fabric_only,
2295         .tpg_get_inst_index             = vhost_scsi_tpg_get_inst_index,
2296         .release_cmd                    = vhost_scsi_release_cmd,
2297         .check_stop_free                = vhost_scsi_check_stop_free,
2298         .sess_get_index                 = vhost_scsi_sess_get_index,
2299         .sess_get_initiator_sid         = NULL,
2300         .write_pending                  = vhost_scsi_write_pending,
2301         .write_pending_status           = vhost_scsi_write_pending_status,
2302         .set_default_node_attributes    = vhost_scsi_set_default_node_attrs,
2303         .get_cmd_state                  = vhost_scsi_get_cmd_state,
2304         .queue_data_in                  = vhost_scsi_queue_data_in,
2305         .queue_status                   = vhost_scsi_queue_status,
2306         .queue_tm_rsp                   = vhost_scsi_queue_tm_rsp,
2307         .aborted_task                   = vhost_scsi_aborted_task,
2308         /*
2309          * Setup callers for generic logic in target_core_fabric_configfs.c
2310          */
2311         .fabric_make_wwn                = vhost_scsi_make_tport,
2312         .fabric_drop_wwn                = vhost_scsi_drop_tport,
2313         .fabric_make_tpg                = vhost_scsi_make_tpg,
2314         .fabric_drop_tpg                = vhost_scsi_drop_tpg,
2315         .fabric_post_link               = vhost_scsi_port_link,
2316         .fabric_pre_unlink              = vhost_scsi_port_unlink,
2317
2318         .tfc_wwn_attrs                  = vhost_scsi_wwn_attrs,
2319         .tfc_tpg_base_attrs             = vhost_scsi_tpg_attrs,
2320         .tfc_tpg_attrib_attrs           = vhost_scsi_tpg_attrib_attrs,
2321 };
2322
2323 static int __init vhost_scsi_init(void)
2324 {
2325         int ret = -ENOMEM;
2326
2327         pr_debug("TCM_VHOST fabric module %s on %s/%s"
2328                 " on "UTS_RELEASE"\n", VHOST_SCSI_VERSION, utsname()->sysname,
2329                 utsname()->machine);
2330
2331         /*
2332          * Use our own dedicated workqueue for submitting I/O into
2333          * target core to avoid contention within system_wq.
2334          */
2335         vhost_scsi_workqueue = alloc_workqueue("vhost_scsi", 0, 0);
2336         if (!vhost_scsi_workqueue)
2337                 goto out;
2338
2339         ret = vhost_scsi_register();
2340         if (ret < 0)
2341                 goto out_destroy_workqueue;
2342
2343         ret = target_register_template(&vhost_scsi_ops);
2344         if (ret < 0)
2345                 goto out_vhost_scsi_deregister;
2346
2347         return 0;
2348
2349 out_vhost_scsi_deregister:
2350         vhost_scsi_deregister();
2351 out_destroy_workqueue:
2352         destroy_workqueue(vhost_scsi_workqueue);
2353 out:
2354         return ret;
2355 };
2356
2357 static void vhost_scsi_exit(void)
2358 {
2359         target_unregister_template(&vhost_scsi_ops);
2360         vhost_scsi_deregister();
2361         destroy_workqueue(vhost_scsi_workqueue);
2362 };
2363
2364 MODULE_DESCRIPTION("VHOST_SCSI series fabric driver");
2365 MODULE_ALIAS("tcm_vhost");
2366 MODULE_LICENSE("GPL");
2367 module_init(vhost_scsi_init);
2368 module_exit(vhost_scsi_exit);