Merge tag 'selinux-pr-20190702' of git://git.kernel.org/pub/scm/linux/kernel/git...
[sfrench/cifs-2.6.git] / drivers / scsi / storvsc_drv.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2009, Microsoft Corporation.
4  *
5  * Authors:
6  *   Haiyang Zhang <haiyangz@microsoft.com>
7  *   Hank Janssen  <hjanssen@microsoft.com>
8  *   K. Y. Srinivasan <kys@microsoft.com>
9  */
10
11 #include <linux/kernel.h>
12 #include <linux/wait.h>
13 #include <linux/sched.h>
14 #include <linux/completion.h>
15 #include <linux/string.h>
16 #include <linux/mm.h>
17 #include <linux/delay.h>
18 #include <linux/init.h>
19 #include <linux/slab.h>
20 #include <linux/module.h>
21 #include <linux/device.h>
22 #include <linux/hyperv.h>
23 #include <linux/blkdev.h>
24 #include <scsi/scsi.h>
25 #include <scsi/scsi_cmnd.h>
26 #include <scsi/scsi_host.h>
27 #include <scsi/scsi_device.h>
28 #include <scsi/scsi_tcq.h>
29 #include <scsi/scsi_eh.h>
30 #include <scsi/scsi_devinfo.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_transport_fc.h>
33 #include <scsi/scsi_transport.h>
34
35 /*
36  * All wire protocol details (storage protocol between the guest and the host)
37  * are consolidated here.
38  *
39  * Begin protocol definitions.
40  */
41
42 /*
43  * Version history:
44  * V1 Beta: 0.1
45  * V1 RC < 2008/1/31: 1.0
46  * V1 RC > 2008/1/31:  2.0
47  * Win7: 4.2
48  * Win8: 5.1
49  * Win8.1: 6.0
50  * Win10: 6.2
51  */
52
53 #define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_)    ((((MAJOR_) & 0xff) << 8) | \
54                                                 (((MINOR_) & 0xff)))
55
56 #define VMSTOR_PROTO_VERSION_WIN6       VMSTOR_PROTO_VERSION(2, 0)
57 #define VMSTOR_PROTO_VERSION_WIN7       VMSTOR_PROTO_VERSION(4, 2)
58 #define VMSTOR_PROTO_VERSION_WIN8       VMSTOR_PROTO_VERSION(5, 1)
59 #define VMSTOR_PROTO_VERSION_WIN8_1     VMSTOR_PROTO_VERSION(6, 0)
60 #define VMSTOR_PROTO_VERSION_WIN10      VMSTOR_PROTO_VERSION(6, 2)
61
62 /*  Packet structure describing virtual storage requests. */
63 enum vstor_packet_operation {
64         VSTOR_OPERATION_COMPLETE_IO             = 1,
65         VSTOR_OPERATION_REMOVE_DEVICE           = 2,
66         VSTOR_OPERATION_EXECUTE_SRB             = 3,
67         VSTOR_OPERATION_RESET_LUN               = 4,
68         VSTOR_OPERATION_RESET_ADAPTER           = 5,
69         VSTOR_OPERATION_RESET_BUS               = 6,
70         VSTOR_OPERATION_BEGIN_INITIALIZATION    = 7,
71         VSTOR_OPERATION_END_INITIALIZATION      = 8,
72         VSTOR_OPERATION_QUERY_PROTOCOL_VERSION  = 9,
73         VSTOR_OPERATION_QUERY_PROPERTIES        = 10,
74         VSTOR_OPERATION_ENUMERATE_BUS           = 11,
75         VSTOR_OPERATION_FCHBA_DATA              = 12,
76         VSTOR_OPERATION_CREATE_SUB_CHANNELS     = 13,
77         VSTOR_OPERATION_MAXIMUM                 = 13
78 };
79
80 /*
81  * WWN packet for Fibre Channel HBA
82  */
83
84 struct hv_fc_wwn_packet {
85         u8      primary_active;
86         u8      reserved1[3];
87         u8      primary_port_wwn[8];
88         u8      primary_node_wwn[8];
89         u8      secondary_port_wwn[8];
90         u8      secondary_node_wwn[8];
91 };
92
93
94
95 /*
96  * SRB Flag Bits
97  */
98
99 #define SRB_FLAGS_QUEUE_ACTION_ENABLE           0x00000002
100 #define SRB_FLAGS_DISABLE_DISCONNECT            0x00000004
101 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER        0x00000008
102 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE           0x00000010
103 #define SRB_FLAGS_DISABLE_AUTOSENSE             0x00000020
104 #define SRB_FLAGS_DATA_IN                       0x00000040
105 #define SRB_FLAGS_DATA_OUT                      0x00000080
106 #define SRB_FLAGS_NO_DATA_TRANSFER              0x00000000
107 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT)
108 #define SRB_FLAGS_NO_QUEUE_FREEZE               0x00000100
109 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE          0x00000200
110 #define SRB_FLAGS_FREE_SENSE_BUFFER             0x00000400
111
112 /*
113  * This flag indicates the request is part of the workflow for processing a D3.
114  */
115 #define SRB_FLAGS_D3_PROCESSING                 0x00000800
116 #define SRB_FLAGS_IS_ACTIVE                     0x00010000
117 #define SRB_FLAGS_ALLOCATED_FROM_ZONE           0x00020000
118 #define SRB_FLAGS_SGLIST_FROM_POOL              0x00040000
119 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE           0x00080000
120 #define SRB_FLAGS_NO_KEEP_AWAKE                 0x00100000
121 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE        0x00200000
122 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT      0x00400000
123 #define SRB_FLAGS_DONT_START_NEXT_PACKET        0x00800000
124 #define SRB_FLAGS_PORT_DRIVER_RESERVED          0x0F000000
125 #define SRB_FLAGS_CLASS_DRIVER_RESERVED         0xF0000000
126
127 #define SP_UNTAGGED                     ((unsigned char) ~0)
128 #define SRB_SIMPLE_TAG_REQUEST          0x20
129
130 /*
131  * Platform neutral description of a scsi request -
132  * this remains the same across the write regardless of 32/64 bit
133  * note: it's patterned off the SCSI_PASS_THROUGH structure
134  */
135 #define STORVSC_MAX_CMD_LEN                     0x10
136
137 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE     0x14
138 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE      0x12
139
140 #define STORVSC_SENSE_BUFFER_SIZE               0x14
141 #define STORVSC_MAX_BUF_LEN_WITH_PADDING        0x14
142
143 /*
144  * Sense buffer size changed in win8; have a run-time
145  * variable to track the size we should use.  This value will
146  * likely change during protocol negotiation but it is valid
147  * to start by assuming pre-Win8.
148  */
149 static int sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE;
150
151 /*
152  * The storage protocol version is determined during the
153  * initial exchange with the host.  It will indicate which
154  * storage functionality is available in the host.
155 */
156 static int vmstor_proto_version;
157
158 #define STORVSC_LOGGING_NONE    0
159 #define STORVSC_LOGGING_ERROR   1
160 #define STORVSC_LOGGING_WARN    2
161
162 static int logging_level = STORVSC_LOGGING_ERROR;
163 module_param(logging_level, int, S_IRUGO|S_IWUSR);
164 MODULE_PARM_DESC(logging_level,
165         "Logging level, 0 - None, 1 - Error (default), 2 - Warning.");
166
167 static inline bool do_logging(int level)
168 {
169         return logging_level >= level;
170 }
171
172 #define storvsc_log(dev, level, fmt, ...)                       \
173 do {                                                            \
174         if (do_logging(level))                                  \
175                 dev_warn(&(dev)->device, fmt, ##__VA_ARGS__);   \
176 } while (0)
177
178 struct vmscsi_win8_extension {
179         /*
180          * The following were added in Windows 8
181          */
182         u16 reserve;
183         u8  queue_tag;
184         u8  queue_action;
185         u32 srb_flags;
186         u32 time_out_value;
187         u32 queue_sort_ey;
188 } __packed;
189
190 struct vmscsi_request {
191         u16 length;
192         u8 srb_status;
193         u8 scsi_status;
194
195         u8  port_number;
196         u8  path_id;
197         u8  target_id;
198         u8  lun;
199
200         u8  cdb_length;
201         u8  sense_info_length;
202         u8  data_in;
203         u8  reserved;
204
205         u32 data_transfer_length;
206
207         union {
208                 u8 cdb[STORVSC_MAX_CMD_LEN];
209                 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
210                 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
211         };
212         /*
213          * The following was added in win8.
214          */
215         struct vmscsi_win8_extension win8_extension;
216
217 } __attribute((packed));
218
219
220 /*
221  * The size of the vmscsi_request has changed in win8. The
222  * additional size is because of new elements added to the
223  * structure. These elements are valid only when we are talking
224  * to a win8 host.
225  * Track the correction to size we need to apply. This value
226  * will likely change during protocol negotiation but it is
227  * valid to start by assuming pre-Win8.
228  */
229 static int vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
230
231 /*
232  * The list of storage protocols in order of preference.
233  */
234 struct vmstor_protocol {
235         int protocol_version;
236         int sense_buffer_size;
237         int vmscsi_size_delta;
238 };
239
240
241 static const struct vmstor_protocol vmstor_protocols[] = {
242         {
243                 VMSTOR_PROTO_VERSION_WIN10,
244                 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
245                 0
246         },
247         {
248                 VMSTOR_PROTO_VERSION_WIN8_1,
249                 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
250                 0
251         },
252         {
253                 VMSTOR_PROTO_VERSION_WIN8,
254                 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
255                 0
256         },
257         {
258                 VMSTOR_PROTO_VERSION_WIN7,
259                 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
260                 sizeof(struct vmscsi_win8_extension),
261         },
262         {
263                 VMSTOR_PROTO_VERSION_WIN6,
264                 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
265                 sizeof(struct vmscsi_win8_extension),
266         }
267 };
268
269
270 /*
271  * This structure is sent during the initialization phase to get the different
272  * properties of the channel.
273  */
274
275 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL          0x1
276
277 struct vmstorage_channel_properties {
278         u32 reserved;
279         u16 max_channel_cnt;
280         u16 reserved1;
281
282         u32 flags;
283         u32   max_transfer_bytes;
284
285         u64  reserved2;
286 } __packed;
287
288 /*  This structure is sent during the storage protocol negotiations. */
289 struct vmstorage_protocol_version {
290         /* Major (MSW) and minor (LSW) version numbers. */
291         u16 major_minor;
292
293         /*
294          * Revision number is auto-incremented whenever this file is changed
295          * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
296          * definitely indicate incompatibility--but it does indicate mismatched
297          * builds.
298          * This is only used on the windows side. Just set it to 0.
299          */
300         u16 revision;
301 } __packed;
302
303 /* Channel Property Flags */
304 #define STORAGE_CHANNEL_REMOVABLE_FLAG          0x1
305 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG       0x2
306
307 struct vstor_packet {
308         /* Requested operation type */
309         enum vstor_packet_operation operation;
310
311         /*  Flags - see below for values */
312         u32 flags;
313
314         /* Status of the request returned from the server side. */
315         u32 status;
316
317         /* Data payload area */
318         union {
319                 /*
320                  * Structure used to forward SCSI commands from the
321                  * client to the server.
322                  */
323                 struct vmscsi_request vm_srb;
324
325                 /* Structure used to query channel properties. */
326                 struct vmstorage_channel_properties storage_channel_properties;
327
328                 /* Used during version negotiations. */
329                 struct vmstorage_protocol_version version;
330
331                 /* Fibre channel address packet */
332                 struct hv_fc_wwn_packet wwn_packet;
333
334                 /* Number of sub-channels to create */
335                 u16 sub_channel_count;
336
337                 /* This will be the maximum of the union members */
338                 u8  buffer[0x34];
339         };
340 } __packed;
341
342 /*
343  * Packet Flags:
344  *
345  * This flag indicates that the server should send back a completion for this
346  * packet.
347  */
348
349 #define REQUEST_COMPLETION_FLAG 0x1
350
351 /* Matches Windows-end */
352 enum storvsc_request_type {
353         WRITE_TYPE = 0,
354         READ_TYPE,
355         UNKNOWN_TYPE,
356 };
357
358 /*
359  * SRB status codes and masks; a subset of the codes used here.
360  */
361
362 #define SRB_STATUS_AUTOSENSE_VALID      0x80
363 #define SRB_STATUS_QUEUE_FROZEN         0x40
364 #define SRB_STATUS_INVALID_LUN  0x20
365 #define SRB_STATUS_SUCCESS      0x01
366 #define SRB_STATUS_ABORTED      0x02
367 #define SRB_STATUS_ERROR        0x04
368 #define SRB_STATUS_DATA_OVERRUN 0x12
369
370 #define SRB_STATUS(status) \
371         (status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
372 /*
373  * This is the end of Protocol specific defines.
374  */
375
376 static int storvsc_ringbuffer_size = (128 * 1024);
377 static u32 max_outstanding_req_per_channel;
378
379 static int storvsc_vcpus_per_sub_channel = 4;
380
381 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
382 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
383
384 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
385 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
386
387 static int ring_avail_percent_lowater = 10;
388 module_param(ring_avail_percent_lowater, int, S_IRUGO);
389 MODULE_PARM_DESC(ring_avail_percent_lowater,
390                 "Select a channel if available ring size > this in percent");
391
392 /*
393  * Timeout in seconds for all devices managed by this driver.
394  */
395 static int storvsc_timeout = 180;
396
397 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
398 static struct scsi_transport_template *fc_transport_template;
399 #endif
400
401 static void storvsc_on_channel_callback(void *context);
402
403 #define STORVSC_MAX_LUNS_PER_TARGET                     255
404 #define STORVSC_MAX_TARGETS                             2
405 #define STORVSC_MAX_CHANNELS                            8
406
407 #define STORVSC_FC_MAX_LUNS_PER_TARGET                  255
408 #define STORVSC_FC_MAX_TARGETS                          128
409 #define STORVSC_FC_MAX_CHANNELS                         8
410
411 #define STORVSC_IDE_MAX_LUNS_PER_TARGET                 64
412 #define STORVSC_IDE_MAX_TARGETS                         1
413 #define STORVSC_IDE_MAX_CHANNELS                        1
414
415 struct storvsc_cmd_request {
416         struct scsi_cmnd *cmd;
417
418         struct hv_device *device;
419
420         /* Synchronize the request/response if needed */
421         struct completion wait_event;
422
423         struct vmbus_channel_packet_multipage_buffer mpb;
424         struct vmbus_packet_mpb_array *payload;
425         u32 payload_sz;
426
427         struct vstor_packet vstor_packet;
428 };
429
430
431 /* A storvsc device is a device object that contains a vmbus channel */
432 struct storvsc_device {
433         struct hv_device *device;
434
435         bool     destroy;
436         bool     drain_notify;
437         atomic_t num_outstanding_req;
438         struct Scsi_Host *host;
439
440         wait_queue_head_t waiting_to_drain;
441
442         /*
443          * Each unique Port/Path/Target represents 1 channel ie scsi
444          * controller. In reality, the pathid, targetid is always 0
445          * and the port is set by us
446          */
447         unsigned int port_number;
448         unsigned char path_id;
449         unsigned char target_id;
450
451         /*
452          * Max I/O, the device can support.
453          */
454         u32   max_transfer_bytes;
455         /*
456          * Number of sub-channels we will open.
457          */
458         u16 num_sc;
459         struct vmbus_channel **stor_chns;
460         /*
461          * Mask of CPUs bound to subchannels.
462          */
463         struct cpumask alloced_cpus;
464         /* Used for vsc/vsp channel reset process */
465         struct storvsc_cmd_request init_request;
466         struct storvsc_cmd_request reset_request;
467         /*
468          * Currently active port and node names for FC devices.
469          */
470         u64 node_name;
471         u64 port_name;
472 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
473         struct fc_rport *rport;
474 #endif
475 };
476
477 struct hv_host_device {
478         struct hv_device *dev;
479         unsigned int port;
480         unsigned char path;
481         unsigned char target;
482         struct workqueue_struct *handle_error_wq;
483         struct work_struct host_scan_work;
484         struct Scsi_Host *host;
485 };
486
487 struct storvsc_scan_work {
488         struct work_struct work;
489         struct Scsi_Host *host;
490         u8 lun;
491         u8 tgt_id;
492 };
493
494 static void storvsc_device_scan(struct work_struct *work)
495 {
496         struct storvsc_scan_work *wrk;
497         struct scsi_device *sdev;
498
499         wrk = container_of(work, struct storvsc_scan_work, work);
500
501         sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
502         if (!sdev)
503                 goto done;
504         scsi_rescan_device(&sdev->sdev_gendev);
505         scsi_device_put(sdev);
506
507 done:
508         kfree(wrk);
509 }
510
511 static void storvsc_host_scan(struct work_struct *work)
512 {
513         struct Scsi_Host *host;
514         struct scsi_device *sdev;
515         struct hv_host_device *host_device =
516                 container_of(work, struct hv_host_device, host_scan_work);
517
518         host = host_device->host;
519         /*
520          * Before scanning the host, first check to see if any of the
521          * currrently known devices have been hot removed. We issue a
522          * "unit ready" command against all currently known devices.
523          * This I/O will result in an error for devices that have been
524          * removed. As part of handling the I/O error, we remove the device.
525          *
526          * When a LUN is added or removed, the host sends us a signal to
527          * scan the host. Thus we are forced to discover the LUNs that
528          * may have been removed this way.
529          */
530         mutex_lock(&host->scan_mutex);
531         shost_for_each_device(sdev, host)
532                 scsi_test_unit_ready(sdev, 1, 1, NULL);
533         mutex_unlock(&host->scan_mutex);
534         /*
535          * Now scan the host to discover LUNs that may have been added.
536          */
537         scsi_scan_host(host);
538 }
539
540 static void storvsc_remove_lun(struct work_struct *work)
541 {
542         struct storvsc_scan_work *wrk;
543         struct scsi_device *sdev;
544
545         wrk = container_of(work, struct storvsc_scan_work, work);
546         if (!scsi_host_get(wrk->host))
547                 goto done;
548
549         sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
550
551         if (sdev) {
552                 scsi_remove_device(sdev);
553                 scsi_device_put(sdev);
554         }
555         scsi_host_put(wrk->host);
556
557 done:
558         kfree(wrk);
559 }
560
561
562 /*
563  * We can get incoming messages from the host that are not in response to
564  * messages that we have sent out. An example of this would be messages
565  * received by the guest to notify dynamic addition/removal of LUNs. To
566  * deal with potential race conditions where the driver may be in the
567  * midst of being unloaded when we might receive an unsolicited message
568  * from the host, we have implemented a mechanism to gurantee sequential
569  * consistency:
570  *
571  * 1) Once the device is marked as being destroyed, we will fail all
572  *    outgoing messages.
573  * 2) We permit incoming messages when the device is being destroyed,
574  *    only to properly account for messages already sent out.
575  */
576
577 static inline struct storvsc_device *get_out_stor_device(
578                                         struct hv_device *device)
579 {
580         struct storvsc_device *stor_device;
581
582         stor_device = hv_get_drvdata(device);
583
584         if (stor_device && stor_device->destroy)
585                 stor_device = NULL;
586
587         return stor_device;
588 }
589
590
591 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
592 {
593         dev->drain_notify = true;
594         wait_event(dev->waiting_to_drain,
595                    atomic_read(&dev->num_outstanding_req) == 0);
596         dev->drain_notify = false;
597 }
598
599 static inline struct storvsc_device *get_in_stor_device(
600                                         struct hv_device *device)
601 {
602         struct storvsc_device *stor_device;
603
604         stor_device = hv_get_drvdata(device);
605
606         if (!stor_device)
607                 goto get_in_err;
608
609         /*
610          * If the device is being destroyed; allow incoming
611          * traffic only to cleanup outstanding requests.
612          */
613
614         if (stor_device->destroy  &&
615                 (atomic_read(&stor_device->num_outstanding_req) == 0))
616                 stor_device = NULL;
617
618 get_in_err:
619         return stor_device;
620
621 }
622
623 static void handle_sc_creation(struct vmbus_channel *new_sc)
624 {
625         struct hv_device *device = new_sc->primary_channel->device_obj;
626         struct device *dev = &device->device;
627         struct storvsc_device *stor_device;
628         struct vmstorage_channel_properties props;
629         int ret;
630
631         stor_device = get_out_stor_device(device);
632         if (!stor_device)
633                 return;
634
635         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
636
637         ret = vmbus_open(new_sc,
638                          storvsc_ringbuffer_size,
639                          storvsc_ringbuffer_size,
640                          (void *)&props,
641                          sizeof(struct vmstorage_channel_properties),
642                          storvsc_on_channel_callback, new_sc);
643
644         /* In case vmbus_open() fails, we don't use the sub-channel. */
645         if (ret != 0) {
646                 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
647                 return;
648         }
649
650         /* Add the sub-channel to the array of available channels. */
651         stor_device->stor_chns[new_sc->target_cpu] = new_sc;
652         cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
653 }
654
655 static void  handle_multichannel_storage(struct hv_device *device, int max_chns)
656 {
657         struct device *dev = &device->device;
658         struct storvsc_device *stor_device;
659         int num_sc;
660         struct storvsc_cmd_request *request;
661         struct vstor_packet *vstor_packet;
662         int ret, t;
663
664         /*
665          * If the number of CPUs is artificially restricted, such as
666          * with maxcpus=1 on the kernel boot line, Hyper-V could offer
667          * sub-channels >= the number of CPUs. These sub-channels
668          * should not be created. The primary channel is already created
669          * and assigned to one CPU, so check against # CPUs - 1.
670          */
671         num_sc = min((int)(num_online_cpus() - 1), max_chns);
672         if (!num_sc)
673                 return;
674
675         stor_device = get_out_stor_device(device);
676         if (!stor_device)
677                 return;
678
679         stor_device->num_sc = num_sc;
680         request = &stor_device->init_request;
681         vstor_packet = &request->vstor_packet;
682
683         /*
684          * Establish a handler for dealing with subchannels.
685          */
686         vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
687
688         /*
689          * Request the host to create sub-channels.
690          */
691         memset(request, 0, sizeof(struct storvsc_cmd_request));
692         init_completion(&request->wait_event);
693         vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
694         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
695         vstor_packet->sub_channel_count = num_sc;
696
697         ret = vmbus_sendpacket(device->channel, vstor_packet,
698                                (sizeof(struct vstor_packet) -
699                                vmscsi_size_delta),
700                                (unsigned long)request,
701                                VM_PKT_DATA_INBAND,
702                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
703
704         if (ret != 0) {
705                 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
706                 return;
707         }
708
709         t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
710         if (t == 0) {
711                 dev_err(dev, "Failed to create sub-channel: timed out\n");
712                 return;
713         }
714
715         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
716             vstor_packet->status != 0) {
717                 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
718                         vstor_packet->operation, vstor_packet->status);
719                 return;
720         }
721
722         /*
723          * We need to do nothing here, because vmbus_process_offer()
724          * invokes channel->sc_creation_callback, which will open and use
725          * the sub-channel(s).
726          */
727 }
728
729 static void cache_wwn(struct storvsc_device *stor_device,
730                       struct vstor_packet *vstor_packet)
731 {
732         /*
733          * Cache the currently active port and node ww names.
734          */
735         if (vstor_packet->wwn_packet.primary_active) {
736                 stor_device->node_name =
737                         wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
738                 stor_device->port_name =
739                         wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
740         } else {
741                 stor_device->node_name =
742                         wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
743                 stor_device->port_name =
744                         wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
745         }
746 }
747
748
749 static int storvsc_execute_vstor_op(struct hv_device *device,
750                                     struct storvsc_cmd_request *request,
751                                     bool status_check)
752 {
753         struct vstor_packet *vstor_packet;
754         int ret, t;
755
756         vstor_packet = &request->vstor_packet;
757
758         init_completion(&request->wait_event);
759         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
760
761         ret = vmbus_sendpacket(device->channel, vstor_packet,
762                                (sizeof(struct vstor_packet) -
763                                vmscsi_size_delta),
764                                (unsigned long)request,
765                                VM_PKT_DATA_INBAND,
766                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
767         if (ret != 0)
768                 return ret;
769
770         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
771         if (t == 0)
772                 return -ETIMEDOUT;
773
774         if (!status_check)
775                 return ret;
776
777         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
778             vstor_packet->status != 0)
779                 return -EINVAL;
780
781         return ret;
782 }
783
784 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
785 {
786         struct storvsc_device *stor_device;
787         struct storvsc_cmd_request *request;
788         struct vstor_packet *vstor_packet;
789         int ret, i;
790         int max_chns;
791         bool process_sub_channels = false;
792
793         stor_device = get_out_stor_device(device);
794         if (!stor_device)
795                 return -ENODEV;
796
797         request = &stor_device->init_request;
798         vstor_packet = &request->vstor_packet;
799
800         /*
801          * Now, initiate the vsc/vsp initialization protocol on the open
802          * channel
803          */
804         memset(request, 0, sizeof(struct storvsc_cmd_request));
805         vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
806         ret = storvsc_execute_vstor_op(device, request, true);
807         if (ret)
808                 return ret;
809         /*
810          * Query host supported protocol version.
811          */
812
813         for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) {
814                 /* reuse the packet for version range supported */
815                 memset(vstor_packet, 0, sizeof(struct vstor_packet));
816                 vstor_packet->operation =
817                         VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
818
819                 vstor_packet->version.major_minor =
820                         vmstor_protocols[i].protocol_version;
821
822                 /*
823                  * The revision number is only used in Windows; set it to 0.
824                  */
825                 vstor_packet->version.revision = 0;
826                 ret = storvsc_execute_vstor_op(device, request, false);
827                 if (ret != 0)
828                         return ret;
829
830                 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
831                         return -EINVAL;
832
833                 if (vstor_packet->status == 0) {
834                         vmstor_proto_version =
835                                 vmstor_protocols[i].protocol_version;
836
837                         sense_buffer_size =
838                                 vmstor_protocols[i].sense_buffer_size;
839
840                         vmscsi_size_delta =
841                                 vmstor_protocols[i].vmscsi_size_delta;
842
843                         break;
844                 }
845         }
846
847         if (vstor_packet->status != 0)
848                 return -EINVAL;
849
850
851         memset(vstor_packet, 0, sizeof(struct vstor_packet));
852         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
853         ret = storvsc_execute_vstor_op(device, request, true);
854         if (ret != 0)
855                 return ret;
856
857         /*
858          * Check to see if multi-channel support is there.
859          * Hosts that implement protocol version of 5.1 and above
860          * support multi-channel.
861          */
862         max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
863
864         /*
865          * Allocate state to manage the sub-channels.
866          * We allocate an array based on the numbers of possible CPUs
867          * (Hyper-V does not support cpu online/offline).
868          * This Array will be sparseley populated with unique
869          * channels - primary + sub-channels.
870          * We will however populate all the slots to evenly distribute
871          * the load.
872          */
873         stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
874                                          GFP_KERNEL);
875         if (stor_device->stor_chns == NULL)
876                 return -ENOMEM;
877
878         stor_device->stor_chns[device->channel->target_cpu] = device->channel;
879         cpumask_set_cpu(device->channel->target_cpu,
880                         &stor_device->alloced_cpus);
881
882         if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) {
883                 if (vstor_packet->storage_channel_properties.flags &
884                     STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
885                         process_sub_channels = true;
886         }
887         stor_device->max_transfer_bytes =
888                 vstor_packet->storage_channel_properties.max_transfer_bytes;
889
890         if (!is_fc)
891                 goto done;
892
893         /*
894          * For FC devices retrieve FC HBA data.
895          */
896         memset(vstor_packet, 0, sizeof(struct vstor_packet));
897         vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
898         ret = storvsc_execute_vstor_op(device, request, true);
899         if (ret != 0)
900                 return ret;
901
902         /*
903          * Cache the currently active port and node ww names.
904          */
905         cache_wwn(stor_device, vstor_packet);
906
907 done:
908
909         memset(vstor_packet, 0, sizeof(struct vstor_packet));
910         vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
911         ret = storvsc_execute_vstor_op(device, request, true);
912         if (ret != 0)
913                 return ret;
914
915         if (process_sub_channels)
916                 handle_multichannel_storage(device, max_chns);
917
918         return ret;
919 }
920
921 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
922                                 struct scsi_cmnd *scmnd,
923                                 struct Scsi_Host *host,
924                                 u8 asc, u8 ascq)
925 {
926         struct storvsc_scan_work *wrk;
927         void (*process_err_fn)(struct work_struct *work);
928         struct hv_host_device *host_dev = shost_priv(host);
929         bool do_work = false;
930
931         switch (SRB_STATUS(vm_srb->srb_status)) {
932         case SRB_STATUS_ERROR:
933                 /*
934                  * Let upper layer deal with error when
935                  * sense message is present.
936                  */
937
938                 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)
939                         break;
940                 /*
941                  * If there is an error; offline the device since all
942                  * error recovery strategies would have already been
943                  * deployed on the host side. However, if the command
944                  * were a pass-through command deal with it appropriately.
945                  */
946                 switch (scmnd->cmnd[0]) {
947                 case ATA_16:
948                 case ATA_12:
949                         set_host_byte(scmnd, DID_PASSTHROUGH);
950                         break;
951                 /*
952                  * On Some Windows hosts TEST_UNIT_READY command can return
953                  * SRB_STATUS_ERROR, let the upper level code deal with it
954                  * based on the sense information.
955                  */
956                 case TEST_UNIT_READY:
957                         break;
958                 default:
959                         set_host_byte(scmnd, DID_ERROR);
960                 }
961                 break;
962         case SRB_STATUS_INVALID_LUN:
963                 set_host_byte(scmnd, DID_NO_CONNECT);
964                 do_work = true;
965                 process_err_fn = storvsc_remove_lun;
966                 break;
967         case SRB_STATUS_ABORTED:
968                 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID &&
969                     (asc == 0x2a) && (ascq == 0x9)) {
970                         do_work = true;
971                         process_err_fn = storvsc_device_scan;
972                         /*
973                          * Retry the I/O that trigerred this.
974                          */
975                         set_host_byte(scmnd, DID_REQUEUE);
976                 }
977                 break;
978         }
979
980         if (!do_work)
981                 return;
982
983         /*
984          * We need to schedule work to process this error; schedule it.
985          */
986         wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
987         if (!wrk) {
988                 set_host_byte(scmnd, DID_TARGET_FAILURE);
989                 return;
990         }
991
992         wrk->host = host;
993         wrk->lun = vm_srb->lun;
994         wrk->tgt_id = vm_srb->target_id;
995         INIT_WORK(&wrk->work, process_err_fn);
996         queue_work(host_dev->handle_error_wq, &wrk->work);
997 }
998
999
1000 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1001                                        struct storvsc_device *stor_dev)
1002 {
1003         struct scsi_cmnd *scmnd = cmd_request->cmd;
1004         struct scsi_sense_hdr sense_hdr;
1005         struct vmscsi_request *vm_srb;
1006         u32 data_transfer_length;
1007         struct Scsi_Host *host;
1008         u32 payload_sz = cmd_request->payload_sz;
1009         void *payload = cmd_request->payload;
1010
1011         host = stor_dev->host;
1012
1013         vm_srb = &cmd_request->vstor_packet.vm_srb;
1014         data_transfer_length = vm_srb->data_transfer_length;
1015
1016         scmnd->result = vm_srb->scsi_status;
1017
1018         if (scmnd->result) {
1019                 if (scsi_normalize_sense(scmnd->sense_buffer,
1020                                 SCSI_SENSE_BUFFERSIZE, &sense_hdr) &&
1021                     !(sense_hdr.sense_key == NOT_READY &&
1022                                  sense_hdr.asc == 0x03A) &&
1023                     do_logging(STORVSC_LOGGING_ERROR))
1024                         scsi_print_sense_hdr(scmnd->device, "storvsc",
1025                                              &sense_hdr);
1026         }
1027
1028         if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1029                 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1030                                          sense_hdr.ascq);
1031                 /*
1032                  * The Windows driver set data_transfer_length on
1033                  * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1034                  * is untouched.  In these cases we set it to 0.
1035                  */
1036                 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1037                         data_transfer_length = 0;
1038         }
1039
1040         scsi_set_resid(scmnd,
1041                 cmd_request->payload->range.len - data_transfer_length);
1042
1043         scmnd->scsi_done(scmnd);
1044
1045         if (payload_sz >
1046                 sizeof(struct vmbus_channel_packet_multipage_buffer))
1047                 kfree(payload);
1048 }
1049
1050 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1051                                   struct vstor_packet *vstor_packet,
1052                                   struct storvsc_cmd_request *request)
1053 {
1054         struct vstor_packet *stor_pkt;
1055         struct hv_device *device = stor_device->device;
1056
1057         stor_pkt = &request->vstor_packet;
1058
1059         /*
1060          * The current SCSI handling on the host side does
1061          * not correctly handle:
1062          * INQUIRY command with page code parameter set to 0x80
1063          * MODE_SENSE command with cmd[2] == 0x1c
1064          *
1065          * Setup srb and scsi status so this won't be fatal.
1066          * We do this so we can distinguish truly fatal failues
1067          * (srb status == 0x4) and off-line the device in that case.
1068          */
1069
1070         if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1071            (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1072                 vstor_packet->vm_srb.scsi_status = 0;
1073                 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1074         }
1075
1076
1077         /* Copy over the status...etc */
1078         stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1079         stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1080         stor_pkt->vm_srb.sense_info_length =
1081         vstor_packet->vm_srb.sense_info_length;
1082
1083         if (vstor_packet->vm_srb.scsi_status != 0 ||
1084             vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS)
1085                 storvsc_log(device, STORVSC_LOGGING_WARN,
1086                         "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
1087                         stor_pkt->vm_srb.cdb[0],
1088                         vstor_packet->vm_srb.scsi_status,
1089                         vstor_packet->vm_srb.srb_status);
1090
1091         if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
1092                 /* CHECK_CONDITION */
1093                 if (vstor_packet->vm_srb.srb_status &
1094                         SRB_STATUS_AUTOSENSE_VALID) {
1095                         /* autosense data available */
1096
1097                         storvsc_log(device, STORVSC_LOGGING_WARN,
1098                                 "stor pkt %p autosense data valid - len %d\n",
1099                                 request, vstor_packet->vm_srb.sense_info_length);
1100
1101                         memcpy(request->cmd->sense_buffer,
1102                                vstor_packet->vm_srb.sense_data,
1103                                vstor_packet->vm_srb.sense_info_length);
1104
1105                 }
1106         }
1107
1108         stor_pkt->vm_srb.data_transfer_length =
1109         vstor_packet->vm_srb.data_transfer_length;
1110
1111         storvsc_command_completion(request, stor_device);
1112
1113         if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1114                 stor_device->drain_notify)
1115                 wake_up(&stor_device->waiting_to_drain);
1116
1117
1118 }
1119
1120 static void storvsc_on_receive(struct storvsc_device *stor_device,
1121                              struct vstor_packet *vstor_packet,
1122                              struct storvsc_cmd_request *request)
1123 {
1124         struct hv_host_device *host_dev;
1125         switch (vstor_packet->operation) {
1126         case VSTOR_OPERATION_COMPLETE_IO:
1127                 storvsc_on_io_completion(stor_device, vstor_packet, request);
1128                 break;
1129
1130         case VSTOR_OPERATION_REMOVE_DEVICE:
1131         case VSTOR_OPERATION_ENUMERATE_BUS:
1132                 host_dev = shost_priv(stor_device->host);
1133                 queue_work(
1134                         host_dev->handle_error_wq, &host_dev->host_scan_work);
1135                 break;
1136
1137         case VSTOR_OPERATION_FCHBA_DATA:
1138                 cache_wwn(stor_device, vstor_packet);
1139 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1140                 fc_host_node_name(stor_device->host) = stor_device->node_name;
1141                 fc_host_port_name(stor_device->host) = stor_device->port_name;
1142 #endif
1143                 break;
1144         default:
1145                 break;
1146         }
1147 }
1148
1149 static void storvsc_on_channel_callback(void *context)
1150 {
1151         struct vmbus_channel *channel = (struct vmbus_channel *)context;
1152         const struct vmpacket_descriptor *desc;
1153         struct hv_device *device;
1154         struct storvsc_device *stor_device;
1155
1156         if (channel->primary_channel != NULL)
1157                 device = channel->primary_channel->device_obj;
1158         else
1159                 device = channel->device_obj;
1160
1161         stor_device = get_in_stor_device(device);
1162         if (!stor_device)
1163                 return;
1164
1165         foreach_vmbus_pkt(desc, channel) {
1166                 void *packet = hv_pkt_data(desc);
1167                 struct storvsc_cmd_request *request;
1168
1169                 request = (struct storvsc_cmd_request *)
1170                         ((unsigned long)desc->trans_id);
1171
1172                 if (request == &stor_device->init_request ||
1173                     request == &stor_device->reset_request) {
1174                         memcpy(&request->vstor_packet, packet,
1175                                (sizeof(struct vstor_packet) - vmscsi_size_delta));
1176                         complete(&request->wait_event);
1177                 } else {
1178                         storvsc_on_receive(stor_device, packet, request);
1179                 }
1180         }
1181 }
1182
1183 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1184                                   bool is_fc)
1185 {
1186         struct vmstorage_channel_properties props;
1187         int ret;
1188
1189         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1190
1191         ret = vmbus_open(device->channel,
1192                          ring_size,
1193                          ring_size,
1194                          (void *)&props,
1195                          sizeof(struct vmstorage_channel_properties),
1196                          storvsc_on_channel_callback, device->channel);
1197
1198         if (ret != 0)
1199                 return ret;
1200
1201         ret = storvsc_channel_init(device, is_fc);
1202
1203         return ret;
1204 }
1205
1206 static int storvsc_dev_remove(struct hv_device *device)
1207 {
1208         struct storvsc_device *stor_device;
1209
1210         stor_device = hv_get_drvdata(device);
1211
1212         stor_device->destroy = true;
1213
1214         /* Make sure flag is set before waiting */
1215         wmb();
1216
1217         /*
1218          * At this point, all outbound traffic should be disable. We
1219          * only allow inbound traffic (responses) to proceed so that
1220          * outstanding requests can be completed.
1221          */
1222
1223         storvsc_wait_to_drain(stor_device);
1224
1225         /*
1226          * Since we have already drained, we don't need to busy wait
1227          * as was done in final_release_stor_device()
1228          * Note that we cannot set the ext pointer to NULL until
1229          * we have drained - to drain the outgoing packets, we need to
1230          * allow incoming packets.
1231          */
1232         hv_set_drvdata(device, NULL);
1233
1234         /* Close the channel */
1235         vmbus_close(device->channel);
1236
1237         kfree(stor_device->stor_chns);
1238         kfree(stor_device);
1239         return 0;
1240 }
1241
1242 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1243                                         u16 q_num)
1244 {
1245         u16 slot = 0;
1246         u16 hash_qnum;
1247         const struct cpumask *node_mask;
1248         int num_channels, tgt_cpu;
1249
1250         if (stor_device->num_sc == 0)
1251                 return stor_device->device->channel;
1252
1253         /*
1254          * Our channel array is sparsley populated and we
1255          * initiated I/O on a processor/hw-q that does not
1256          * currently have a designated channel. Fix this.
1257          * The strategy is simple:
1258          * I. Ensure NUMA locality
1259          * II. Distribute evenly (best effort)
1260          * III. Mapping is persistent.
1261          */
1262
1263         node_mask = cpumask_of_node(cpu_to_node(q_num));
1264
1265         num_channels = 0;
1266         for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1267                 if (cpumask_test_cpu(tgt_cpu, node_mask))
1268                         num_channels++;
1269         }
1270         if (num_channels == 0)
1271                 return stor_device->device->channel;
1272
1273         hash_qnum = q_num;
1274         while (hash_qnum >= num_channels)
1275                 hash_qnum -= num_channels;
1276
1277         for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1278                 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1279                         continue;
1280                 if (slot == hash_qnum)
1281                         break;
1282                 slot++;
1283         }
1284
1285         stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1286
1287         return stor_device->stor_chns[q_num];
1288 }
1289
1290
1291 static int storvsc_do_io(struct hv_device *device,
1292                          struct storvsc_cmd_request *request, u16 q_num)
1293 {
1294         struct storvsc_device *stor_device;
1295         struct vstor_packet *vstor_packet;
1296         struct vmbus_channel *outgoing_channel, *channel;
1297         int ret = 0;
1298         const struct cpumask *node_mask;
1299         int tgt_cpu;
1300
1301         vstor_packet = &request->vstor_packet;
1302         stor_device = get_out_stor_device(device);
1303
1304         if (!stor_device)
1305                 return -ENODEV;
1306
1307
1308         request->device  = device;
1309         /*
1310          * Select an an appropriate channel to send the request out.
1311          */
1312         if (stor_device->stor_chns[q_num] != NULL) {
1313                 outgoing_channel = stor_device->stor_chns[q_num];
1314                 if (outgoing_channel->target_cpu == q_num) {
1315                         /*
1316                          * Ideally, we want to pick a different channel if
1317                          * available on the same NUMA node.
1318                          */
1319                         node_mask = cpumask_of_node(cpu_to_node(q_num));
1320                         for_each_cpu_wrap(tgt_cpu,
1321                                  &stor_device->alloced_cpus, q_num + 1) {
1322                                 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1323                                         continue;
1324                                 if (tgt_cpu == q_num)
1325                                         continue;
1326                                 channel = stor_device->stor_chns[tgt_cpu];
1327                                 if (hv_get_avail_to_write_percent(
1328                                                         &channel->outbound)
1329                                                 > ring_avail_percent_lowater) {
1330                                         outgoing_channel = channel;
1331                                         goto found_channel;
1332                                 }
1333                         }
1334
1335                         /*
1336                          * All the other channels on the same NUMA node are
1337                          * busy. Try to use the channel on the current CPU
1338                          */
1339                         if (hv_get_avail_to_write_percent(
1340                                                 &outgoing_channel->outbound)
1341                                         > ring_avail_percent_lowater)
1342                                 goto found_channel;
1343
1344                         /*
1345                          * If we reach here, all the channels on the current
1346                          * NUMA node are busy. Try to find a channel in
1347                          * other NUMA nodes
1348                          */
1349                         for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1350                                 if (cpumask_test_cpu(tgt_cpu, node_mask))
1351                                         continue;
1352                                 channel = stor_device->stor_chns[tgt_cpu];
1353                                 if (hv_get_avail_to_write_percent(
1354                                                         &channel->outbound)
1355                                                 > ring_avail_percent_lowater) {
1356                                         outgoing_channel = channel;
1357                                         goto found_channel;
1358                                 }
1359                         }
1360                 }
1361         } else {
1362                 outgoing_channel = get_og_chn(stor_device, q_num);
1363         }
1364
1365 found_channel:
1366         vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1367
1368         vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1369                                         vmscsi_size_delta);
1370
1371
1372         vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1373
1374
1375         vstor_packet->vm_srb.data_transfer_length =
1376         request->payload->range.len;
1377
1378         vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1379
1380         if (request->payload->range.len) {
1381
1382                 ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1383                                 request->payload, request->payload_sz,
1384                                 vstor_packet,
1385                                 (sizeof(struct vstor_packet) -
1386                                 vmscsi_size_delta),
1387                                 (unsigned long)request);
1388         } else {
1389                 ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1390                                (sizeof(struct vstor_packet) -
1391                                 vmscsi_size_delta),
1392                                (unsigned long)request,
1393                                VM_PKT_DATA_INBAND,
1394                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1395         }
1396
1397         if (ret != 0)
1398                 return ret;
1399
1400         atomic_inc(&stor_device->num_outstanding_req);
1401
1402         return ret;
1403 }
1404
1405 static int storvsc_device_alloc(struct scsi_device *sdevice)
1406 {
1407         /*
1408          * Set blist flag to permit the reading of the VPD pages even when
1409          * the target may claim SPC-2 compliance. MSFT targets currently
1410          * claim SPC-2 compliance while they implement post SPC-2 features.
1411          * With this flag we can correctly handle WRITE_SAME_16 issues.
1412          *
1413          * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1414          * still supports REPORT LUN.
1415          */
1416         sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1417
1418         return 0;
1419 }
1420
1421 static int storvsc_device_configure(struct scsi_device *sdevice)
1422 {
1423         blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1424
1425         /* Ensure there are no gaps in presented sgls */
1426         blk_queue_virt_boundary(sdevice->request_queue, PAGE_SIZE - 1);
1427
1428         sdevice->no_write_same = 1;
1429
1430         /*
1431          * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1432          * if the device is a MSFT virtual device.  If the host is
1433          * WIN10 or newer, allow write_same.
1434          */
1435         if (!strncmp(sdevice->vendor, "Msft", 4)) {
1436                 switch (vmstor_proto_version) {
1437                 case VMSTOR_PROTO_VERSION_WIN8:
1438                 case VMSTOR_PROTO_VERSION_WIN8_1:
1439                         sdevice->scsi_level = SCSI_SPC_3;
1440                         break;
1441                 }
1442
1443                 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1444                         sdevice->no_write_same = 0;
1445         }
1446
1447         return 0;
1448 }
1449
1450 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1451                            sector_t capacity, int *info)
1452 {
1453         sector_t nsect = capacity;
1454         sector_t cylinders = nsect;
1455         int heads, sectors_pt;
1456
1457         /*
1458          * We are making up these values; let us keep it simple.
1459          */
1460         heads = 0xff;
1461         sectors_pt = 0x3f;      /* Sectors per track */
1462         sector_div(cylinders, heads * sectors_pt);
1463         if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1464                 cylinders = 0xffff;
1465
1466         info[0] = heads;
1467         info[1] = sectors_pt;
1468         info[2] = (int)cylinders;
1469
1470         return 0;
1471 }
1472
1473 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1474 {
1475         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1476         struct hv_device *device = host_dev->dev;
1477
1478         struct storvsc_device *stor_device;
1479         struct storvsc_cmd_request *request;
1480         struct vstor_packet *vstor_packet;
1481         int ret, t;
1482
1483
1484         stor_device = get_out_stor_device(device);
1485         if (!stor_device)
1486                 return FAILED;
1487
1488         request = &stor_device->reset_request;
1489         vstor_packet = &request->vstor_packet;
1490
1491         init_completion(&request->wait_event);
1492
1493         vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1494         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1495         vstor_packet->vm_srb.path_id = stor_device->path_id;
1496
1497         ret = vmbus_sendpacket(device->channel, vstor_packet,
1498                                (sizeof(struct vstor_packet) -
1499                                 vmscsi_size_delta),
1500                                (unsigned long)&stor_device->reset_request,
1501                                VM_PKT_DATA_INBAND,
1502                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1503         if (ret != 0)
1504                 return FAILED;
1505
1506         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1507         if (t == 0)
1508                 return TIMEOUT_ERROR;
1509
1510
1511         /*
1512          * At this point, all outstanding requests in the adapter
1513          * should have been flushed out and return to us
1514          * There is a potential race here where the host may be in
1515          * the process of responding when we return from here.
1516          * Just wait for all in-transit packets to be accounted for
1517          * before we return from here.
1518          */
1519         storvsc_wait_to_drain(stor_device);
1520
1521         return SUCCESS;
1522 }
1523
1524 /*
1525  * The host guarantees to respond to each command, although I/O latencies might
1526  * be unbounded on Azure.  Reset the timer unconditionally to give the host a
1527  * chance to perform EH.
1528  */
1529 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1530 {
1531 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1532         if (scmnd->device->host->transportt == fc_transport_template)
1533                 return fc_eh_timed_out(scmnd);
1534 #endif
1535         return BLK_EH_RESET_TIMER;
1536 }
1537
1538 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1539 {
1540         bool allowed = true;
1541         u8 scsi_op = scmnd->cmnd[0];
1542
1543         switch (scsi_op) {
1544         /* the host does not handle WRITE_SAME, log accident usage */
1545         case WRITE_SAME:
1546         /*
1547          * smartd sends this command and the host does not handle
1548          * this. So, don't send it.
1549          */
1550         case SET_WINDOW:
1551                 scmnd->result = ILLEGAL_REQUEST << 16;
1552                 allowed = false;
1553                 break;
1554         default:
1555                 break;
1556         }
1557         return allowed;
1558 }
1559
1560 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1561 {
1562         int ret;
1563         struct hv_host_device *host_dev = shost_priv(host);
1564         struct hv_device *dev = host_dev->dev;
1565         struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1566         int i;
1567         struct scatterlist *sgl;
1568         unsigned int sg_count = 0;
1569         struct vmscsi_request *vm_srb;
1570         struct scatterlist *cur_sgl;
1571         struct vmbus_packet_mpb_array  *payload;
1572         u32 payload_sz;
1573         u32 length;
1574
1575         if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1576                 /*
1577                  * On legacy hosts filter unimplemented commands.
1578                  * Future hosts are expected to correctly handle
1579                  * unsupported commands. Furthermore, it is
1580                  * possible that some of the currently
1581                  * unsupported commands maybe supported in
1582                  * future versions of the host.
1583                  */
1584                 if (!storvsc_scsi_cmd_ok(scmnd)) {
1585                         scmnd->scsi_done(scmnd);
1586                         return 0;
1587                 }
1588         }
1589
1590         /* Setup the cmd request */
1591         cmd_request->cmd = scmnd;
1592
1593         vm_srb = &cmd_request->vstor_packet.vm_srb;
1594         vm_srb->win8_extension.time_out_value = 60;
1595
1596         vm_srb->win8_extension.srb_flags |=
1597                 SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1598
1599         if (scmnd->device->tagged_supported) {
1600                 vm_srb->win8_extension.srb_flags |=
1601                 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1602                 vm_srb->win8_extension.queue_tag = SP_UNTAGGED;
1603                 vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST;
1604         }
1605
1606         /* Build the SRB */
1607         switch (scmnd->sc_data_direction) {
1608         case DMA_TO_DEVICE:
1609                 vm_srb->data_in = WRITE_TYPE;
1610                 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1611                 break;
1612         case DMA_FROM_DEVICE:
1613                 vm_srb->data_in = READ_TYPE;
1614                 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1615                 break;
1616         case DMA_NONE:
1617                 vm_srb->data_in = UNKNOWN_TYPE;
1618                 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1619                 break;
1620         default:
1621                 /*
1622                  * This is DMA_BIDIRECTIONAL or something else we are never
1623                  * supposed to see here.
1624                  */
1625                 WARN(1, "Unexpected data direction: %d\n",
1626                      scmnd->sc_data_direction);
1627                 return -EINVAL;
1628         }
1629
1630
1631         vm_srb->port_number = host_dev->port;
1632         vm_srb->path_id = scmnd->device->channel;
1633         vm_srb->target_id = scmnd->device->id;
1634         vm_srb->lun = scmnd->device->lun;
1635
1636         vm_srb->cdb_length = scmnd->cmd_len;
1637
1638         memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1639
1640         sgl = (struct scatterlist *)scsi_sglist(scmnd);
1641         sg_count = scsi_sg_count(scmnd);
1642
1643         length = scsi_bufflen(scmnd);
1644         payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1645         payload_sz = sizeof(cmd_request->mpb);
1646
1647         if (sg_count) {
1648                 if (sg_count > MAX_PAGE_BUFFER_COUNT) {
1649
1650                         payload_sz = (sg_count * sizeof(u64) +
1651                                       sizeof(struct vmbus_packet_mpb_array));
1652                         payload = kzalloc(payload_sz, GFP_ATOMIC);
1653                         if (!payload)
1654                                 return SCSI_MLQUEUE_DEVICE_BUSY;
1655                 }
1656
1657                 payload->range.len = length;
1658                 payload->range.offset = sgl[0].offset;
1659
1660                 cur_sgl = sgl;
1661                 for (i = 0; i < sg_count; i++) {
1662                         payload->range.pfn_array[i] =
1663                                 page_to_pfn(sg_page((cur_sgl)));
1664                         cur_sgl = sg_next(cur_sgl);
1665                 }
1666         }
1667
1668         cmd_request->payload = payload;
1669         cmd_request->payload_sz = payload_sz;
1670
1671         /* Invokes the vsc to start an IO */
1672         ret = storvsc_do_io(dev, cmd_request, get_cpu());
1673         put_cpu();
1674
1675         if (ret == -EAGAIN) {
1676                 if (payload_sz > sizeof(cmd_request->mpb))
1677                         kfree(payload);
1678                 /* no more space */
1679                 return SCSI_MLQUEUE_DEVICE_BUSY;
1680         }
1681
1682         return 0;
1683 }
1684
1685 static struct scsi_host_template scsi_driver = {
1686         .module =               THIS_MODULE,
1687         .name =                 "storvsc_host_t",
1688         .cmd_size =             sizeof(struct storvsc_cmd_request),
1689         .bios_param =           storvsc_get_chs,
1690         .queuecommand =         storvsc_queuecommand,
1691         .eh_host_reset_handler =        storvsc_host_reset_handler,
1692         .proc_name =            "storvsc_host",
1693         .eh_timed_out =         storvsc_eh_timed_out,
1694         .slave_alloc =          storvsc_device_alloc,
1695         .slave_configure =      storvsc_device_configure,
1696         .cmd_per_lun =          2048,
1697         .this_id =              -1,
1698         /* Make sure we dont get a sg segment crosses a page boundary */
1699         .dma_boundary =         PAGE_SIZE-1,
1700         .no_write_same =        1,
1701         .track_queue_depth =    1,
1702 };
1703
1704 enum {
1705         SCSI_GUID,
1706         IDE_GUID,
1707         SFC_GUID,
1708 };
1709
1710 static const struct hv_vmbus_device_id id_table[] = {
1711         /* SCSI guid */
1712         { HV_SCSI_GUID,
1713           .driver_data = SCSI_GUID
1714         },
1715         /* IDE guid */
1716         { HV_IDE_GUID,
1717           .driver_data = IDE_GUID
1718         },
1719         /* Fibre Channel GUID */
1720         {
1721           HV_SYNTHFC_GUID,
1722           .driver_data = SFC_GUID
1723         },
1724         { },
1725 };
1726
1727 MODULE_DEVICE_TABLE(vmbus, id_table);
1728
1729 static int storvsc_probe(struct hv_device *device,
1730                         const struct hv_vmbus_device_id *dev_id)
1731 {
1732         int ret;
1733         int num_cpus = num_online_cpus();
1734         struct Scsi_Host *host;
1735         struct hv_host_device *host_dev;
1736         bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1737         bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1738         int target = 0;
1739         struct storvsc_device *stor_device;
1740         int max_luns_per_target;
1741         int max_targets;
1742         int max_channels;
1743         int max_sub_channels = 0;
1744
1745         /*
1746          * Based on the windows host we are running on,
1747          * set state to properly communicate with the host.
1748          */
1749
1750         if (vmbus_proto_version < VERSION_WIN8) {
1751                 max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1752                 max_targets = STORVSC_IDE_MAX_TARGETS;
1753                 max_channels = STORVSC_IDE_MAX_CHANNELS;
1754         } else {
1755                 max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET;
1756                 max_targets = STORVSC_MAX_TARGETS;
1757                 max_channels = STORVSC_MAX_CHANNELS;
1758                 /*
1759                  * On Windows8 and above, we support sub-channels for storage
1760                  * on SCSI and FC controllers.
1761                  * The number of sub-channels offerred is based on the number of
1762                  * VCPUs in the guest.
1763                  */
1764                 if (!dev_is_ide)
1765                         max_sub_channels =
1766                                 (num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1767         }
1768
1769         scsi_driver.can_queue = max_outstanding_req_per_channel *
1770                                 (max_sub_channels + 1) *
1771                                 (100 - ring_avail_percent_lowater) / 100;
1772
1773         host = scsi_host_alloc(&scsi_driver,
1774                                sizeof(struct hv_host_device));
1775         if (!host)
1776                 return -ENOMEM;
1777
1778         host_dev = shost_priv(host);
1779         memset(host_dev, 0, sizeof(struct hv_host_device));
1780
1781         host_dev->port = host->host_no;
1782         host_dev->dev = device;
1783         host_dev->host = host;
1784
1785
1786         stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1787         if (!stor_device) {
1788                 ret = -ENOMEM;
1789                 goto err_out0;
1790         }
1791
1792         stor_device->destroy = false;
1793         init_waitqueue_head(&stor_device->waiting_to_drain);
1794         stor_device->device = device;
1795         stor_device->host = host;
1796         hv_set_drvdata(device, stor_device);
1797
1798         stor_device->port_number = host->host_no;
1799         ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1800         if (ret)
1801                 goto err_out1;
1802
1803         host_dev->path = stor_device->path_id;
1804         host_dev->target = stor_device->target_id;
1805
1806         switch (dev_id->driver_data) {
1807         case SFC_GUID:
1808                 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1809                 host->max_id = STORVSC_FC_MAX_TARGETS;
1810                 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1811 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1812                 host->transportt = fc_transport_template;
1813 #endif
1814                 break;
1815
1816         case SCSI_GUID:
1817                 host->max_lun = max_luns_per_target;
1818                 host->max_id = max_targets;
1819                 host->max_channel = max_channels - 1;
1820                 break;
1821
1822         default:
1823                 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1824                 host->max_id = STORVSC_IDE_MAX_TARGETS;
1825                 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
1826                 break;
1827         }
1828         /* max cmd length */
1829         host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1830
1831         /*
1832          * set the table size based on the info we got
1833          * from the host.
1834          */
1835         host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT);
1836         /*
1837          * Set the number of HW queues we are supporting.
1838          */
1839         if (stor_device->num_sc != 0)
1840                 host->nr_hw_queues = stor_device->num_sc + 1;
1841
1842         /*
1843          * Set the error handler work queue.
1844          */
1845         host_dev->handle_error_wq =
1846                         alloc_ordered_workqueue("storvsc_error_wq_%d",
1847                                                 WQ_MEM_RECLAIM,
1848                                                 host->host_no);
1849         if (!host_dev->handle_error_wq)
1850                 goto err_out2;
1851         INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
1852         /* Register the HBA and start the scsi bus scan */
1853         ret = scsi_add_host(host, &device->device);
1854         if (ret != 0)
1855                 goto err_out3;
1856
1857         if (!dev_is_ide) {
1858                 scsi_scan_host(host);
1859         } else {
1860                 target = (device->dev_instance.b[5] << 8 |
1861                          device->dev_instance.b[4]);
1862                 ret = scsi_add_device(host, 0, target, 0);
1863                 if (ret)
1864                         goto err_out4;
1865         }
1866 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1867         if (host->transportt == fc_transport_template) {
1868                 struct fc_rport_identifiers ids = {
1869                         .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
1870                 };
1871
1872                 fc_host_node_name(host) = stor_device->node_name;
1873                 fc_host_port_name(host) = stor_device->port_name;
1874                 stor_device->rport = fc_remote_port_add(host, 0, &ids);
1875                 if (!stor_device->rport) {
1876                         ret = -ENOMEM;
1877                         goto err_out4;
1878                 }
1879         }
1880 #endif
1881         return 0;
1882
1883 err_out4:
1884         scsi_remove_host(host);
1885
1886 err_out3:
1887         destroy_workqueue(host_dev->handle_error_wq);
1888
1889 err_out2:
1890         /*
1891          * Once we have connected with the host, we would need to
1892          * to invoke storvsc_dev_remove() to rollback this state and
1893          * this call also frees up the stor_device; hence the jump around
1894          * err_out1 label.
1895          */
1896         storvsc_dev_remove(device);
1897         goto err_out0;
1898
1899 err_out1:
1900         kfree(stor_device->stor_chns);
1901         kfree(stor_device);
1902
1903 err_out0:
1904         scsi_host_put(host);
1905         return ret;
1906 }
1907
1908 static int storvsc_remove(struct hv_device *dev)
1909 {
1910         struct storvsc_device *stor_device = hv_get_drvdata(dev);
1911         struct Scsi_Host *host = stor_device->host;
1912         struct hv_host_device *host_dev = shost_priv(host);
1913
1914 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1915         if (host->transportt == fc_transport_template) {
1916                 fc_remote_port_delete(stor_device->rport);
1917                 fc_remove_host(host);
1918         }
1919 #endif
1920         destroy_workqueue(host_dev->handle_error_wq);
1921         scsi_remove_host(host);
1922         storvsc_dev_remove(dev);
1923         scsi_host_put(host);
1924
1925         return 0;
1926 }
1927
1928 static struct hv_driver storvsc_drv = {
1929         .name = KBUILD_MODNAME,
1930         .id_table = id_table,
1931         .probe = storvsc_probe,
1932         .remove = storvsc_remove,
1933         .driver = {
1934                 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
1935         },
1936 };
1937
1938 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1939 static struct fc_function_template fc_transport_functions = {
1940         .show_host_node_name = 1,
1941         .show_host_port_name = 1,
1942 };
1943 #endif
1944
1945 static int __init storvsc_drv_init(void)
1946 {
1947         int ret;
1948
1949         /*
1950          * Divide the ring buffer data size (which is 1 page less
1951          * than the ring buffer size since that page is reserved for
1952          * the ring buffer indices) by the max request size (which is
1953          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1954          */
1955         max_outstanding_req_per_channel =
1956                 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1957                 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1958                 sizeof(struct vstor_packet) + sizeof(u64) -
1959                 vmscsi_size_delta,
1960                 sizeof(u64)));
1961
1962 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1963         fc_transport_template = fc_attach_transport(&fc_transport_functions);
1964         if (!fc_transport_template)
1965                 return -ENODEV;
1966 #endif
1967
1968         ret = vmbus_driver_register(&storvsc_drv);
1969
1970 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1971         if (ret)
1972                 fc_release_transport(fc_transport_template);
1973 #endif
1974
1975         return ret;
1976 }
1977
1978 static void __exit storvsc_drv_exit(void)
1979 {
1980         vmbus_driver_unregister(&storvsc_drv);
1981 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1982         fc_release_transport(fc_transport_template);
1983 #endif
1984 }
1985
1986 MODULE_LICENSE("GPL");
1987 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1988 module_init(storvsc_drv_init);
1989 module_exit(storvsc_drv_exit);