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 <linux/dma-mapping.h>
25
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_host.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_tcq.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_devinfo.h>
33 #include <scsi/scsi_dbg.h>
34 #include <scsi/scsi_transport_fc.h>
35 #include <scsi/scsi_transport.h>
36
37 /*
38 * All wire protocol details (storage protocol between the guest and the host)
39 * are consolidated here.
40 *
41 * Begin protocol definitions.
42 */
43
44 /*
45 * Version history:
46 * V1 Beta: 0.1
47 * V1 RC < 2008/1/31: 1.0
48 * V1 RC > 2008/1/31: 2.0
49 * Win7: 4.2
50 * Win8: 5.1
51 * Win8.1: 6.0
52 * Win10: 6.2
53 */
54
55 #define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \
56 (((MINOR_) & 0xff)))
57 #define VMSTOR_PROTO_VERSION_WIN6 VMSTOR_PROTO_VERSION(2, 0)
58 #define VMSTOR_PROTO_VERSION_WIN7 VMSTOR_PROTO_VERSION(4, 2)
59 #define VMSTOR_PROTO_VERSION_WIN8 VMSTOR_PROTO_VERSION(5, 1)
60 #define VMSTOR_PROTO_VERSION_WIN8_1 VMSTOR_PROTO_VERSION(6, 0)
61 #define VMSTOR_PROTO_VERSION_WIN10 VMSTOR_PROTO_VERSION(6, 2)
62
63 /* channel callback timeout in ms */
64 #define CALLBACK_TIMEOUT 2
65
66 /* Packet structure describing virtual storage requests. */
67 enum vstor_packet_operation {
68 VSTOR_OPERATION_COMPLETE_IO = 1,
69 VSTOR_OPERATION_REMOVE_DEVICE = 2,
70 VSTOR_OPERATION_EXECUTE_SRB = 3,
71 VSTOR_OPERATION_RESET_LUN = 4,
72 VSTOR_OPERATION_RESET_ADAPTER = 5,
73 VSTOR_OPERATION_RESET_BUS = 6,
74 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7,
75 VSTOR_OPERATION_END_INITIALIZATION = 8,
76 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9,
77 VSTOR_OPERATION_QUERY_PROPERTIES = 10,
78 VSTOR_OPERATION_ENUMERATE_BUS = 11,
79 VSTOR_OPERATION_FCHBA_DATA = 12,
80 VSTOR_OPERATION_CREATE_SUB_CHANNELS = 13,
81 VSTOR_OPERATION_MAXIMUM = 13
82 };
83
84 /*
85 * WWN packet for Fibre Channel HBA
86 */
87
88 struct hv_fc_wwn_packet {
89 u8 primary_active;
90 u8 reserved1[3];
91 u8 primary_port_wwn[8];
92 u8 primary_node_wwn[8];
93 u8 secondary_port_wwn[8];
94 u8 secondary_node_wwn[8];
95 };
96
97
98
99 /*
100 * SRB Flag Bits
101 */
102
103 #define SRB_FLAGS_QUEUE_ACTION_ENABLE 0x00000002
104 #define SRB_FLAGS_DISABLE_DISCONNECT 0x00000004
105 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER 0x00000008
106 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE 0x00000010
107 #define SRB_FLAGS_DISABLE_AUTOSENSE 0x00000020
108 #define SRB_FLAGS_DATA_IN 0x00000040
109 #define SRB_FLAGS_DATA_OUT 0x00000080
110 #define SRB_FLAGS_NO_DATA_TRANSFER 0x00000000
111 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT)
112 #define SRB_FLAGS_NO_QUEUE_FREEZE 0x00000100
113 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE 0x00000200
114 #define SRB_FLAGS_FREE_SENSE_BUFFER 0x00000400
115
116 /*
117 * This flag indicates the request is part of the workflow for processing a D3.
118 */
119 #define SRB_FLAGS_D3_PROCESSING 0x00000800
120 #define SRB_FLAGS_IS_ACTIVE 0x00010000
121 #define SRB_FLAGS_ALLOCATED_FROM_ZONE 0x00020000
122 #define SRB_FLAGS_SGLIST_FROM_POOL 0x00040000
123 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE 0x00080000
124 #define SRB_FLAGS_NO_KEEP_AWAKE 0x00100000
125 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE 0x00200000
126 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT 0x00400000
127 #define SRB_FLAGS_DONT_START_NEXT_PACKET 0x00800000
128 #define SRB_FLAGS_PORT_DRIVER_RESERVED 0x0F000000
129 #define SRB_FLAGS_CLASS_DRIVER_RESERVED 0xF0000000
130
131 #define SP_UNTAGGED ((unsigned char) ~0)
132 #define SRB_SIMPLE_TAG_REQUEST 0x20
133
134 /*
135 * Platform neutral description of a scsi request -
136 * this remains the same across the write regardless of 32/64 bit
137 * note: it's patterned off the SCSI_PASS_THROUGH structure
138 */
139 #define STORVSC_MAX_CMD_LEN 0x10
140
141 /* Sense buffer size is the same for all versions since Windows 8 */
142 #define STORVSC_SENSE_BUFFER_SIZE 0x14
143 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14
144
145 /*
146 * The storage protocol version is determined during the
147 * initial exchange with the host. It will indicate which
148 * storage functionality is available in the host.
149 */
150 static int vmstor_proto_version;
151
152 static bool hv_dev_is_fc(struct hv_device *hv_dev);
153
154 #define STORVSC_LOGGING_NONE 0
155 #define STORVSC_LOGGING_ERROR 1
156 #define STORVSC_LOGGING_WARN 2
157
158 static int logging_level = STORVSC_LOGGING_ERROR;
159 module_param(logging_level, int, S_IRUGO|S_IWUSR);
160 MODULE_PARM_DESC(logging_level,
161 "Logging level, 0 - None, 1 - Error (default), 2 - Warning.");
162
do_logging(int level)163 static inline bool do_logging(int level)
164 {
165 return logging_level >= level;
166 }
167
168 #define storvsc_log(dev, level, fmt, ...) \
169 do { \
170 if (do_logging(level)) \
171 dev_warn(&(dev)->device, fmt, ##__VA_ARGS__); \
172 } while (0)
173
174 struct vmscsi_request {
175 u16 length;
176 u8 srb_status;
177 u8 scsi_status;
178
179 u8 port_number;
180 u8 path_id;
181 u8 target_id;
182 u8 lun;
183
184 u8 cdb_length;
185 u8 sense_info_length;
186 u8 data_in;
187 u8 reserved;
188
189 u32 data_transfer_length;
190
191 union {
192 u8 cdb[STORVSC_MAX_CMD_LEN];
193 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
194 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
195 };
196 /*
197 * The following was added in win8.
198 */
199 u16 reserve;
200 u8 queue_tag;
201 u8 queue_action;
202 u32 srb_flags;
203 u32 time_out_value;
204 u32 queue_sort_ey;
205
206 } __attribute((packed));
207
208 /*
209 * The list of windows version in order of preference.
210 */
211
212 static const int protocol_version[] = {
213 VMSTOR_PROTO_VERSION_WIN10,
214 VMSTOR_PROTO_VERSION_WIN8_1,
215 VMSTOR_PROTO_VERSION_WIN8,
216 };
217
218
219 /*
220 * This structure is sent during the initialization phase to get the different
221 * properties of the channel.
222 */
223
224 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1
225
226 struct vmstorage_channel_properties {
227 u32 reserved;
228 u16 max_channel_cnt;
229 u16 reserved1;
230
231 u32 flags;
232 u32 max_transfer_bytes;
233
234 u64 reserved2;
235 } __packed;
236
237 /* This structure is sent during the storage protocol negotiations. */
238 struct vmstorage_protocol_version {
239 /* Major (MSW) and minor (LSW) version numbers. */
240 u16 major_minor;
241
242 /*
243 * Revision number is auto-incremented whenever this file is changed
244 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not
245 * definitely indicate incompatibility--but it does indicate mismatched
246 * builds.
247 * This is only used on the windows side. Just set it to 0.
248 */
249 u16 revision;
250 } __packed;
251
252 /* Channel Property Flags */
253 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
254 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
255
256 struct vstor_packet {
257 /* Requested operation type */
258 enum vstor_packet_operation operation;
259
260 /* Flags - see below for values */
261 u32 flags;
262
263 /* Status of the request returned from the server side. */
264 u32 status;
265
266 /* Data payload area */
267 union {
268 /*
269 * Structure used to forward SCSI commands from the
270 * client to the server.
271 */
272 struct vmscsi_request vm_srb;
273
274 /* Structure used to query channel properties. */
275 struct vmstorage_channel_properties storage_channel_properties;
276
277 /* Used during version negotiations. */
278 struct vmstorage_protocol_version version;
279
280 /* Fibre channel address packet */
281 struct hv_fc_wwn_packet wwn_packet;
282
283 /* Number of sub-channels to create */
284 u16 sub_channel_count;
285
286 /* This will be the maximum of the union members */
287 u8 buffer[0x34];
288 };
289 } __packed;
290
291 /*
292 * Packet Flags:
293 *
294 * This flag indicates that the server should send back a completion for this
295 * packet.
296 */
297
298 #define REQUEST_COMPLETION_FLAG 0x1
299
300 /* Matches Windows-end */
301 enum storvsc_request_type {
302 WRITE_TYPE = 0,
303 READ_TYPE,
304 UNKNOWN_TYPE,
305 };
306
307 /*
308 * SRB status codes and masks. In the 8-bit field, the two high order bits
309 * are flags, while the remaining 6 bits are an integer status code. The
310 * definitions here include only the subset of the integer status codes that
311 * are tested for in this driver.
312 */
313 #define SRB_STATUS_AUTOSENSE_VALID 0x80
314 #define SRB_STATUS_QUEUE_FROZEN 0x40
315
316 /* SRB status integer codes */
317 #define SRB_STATUS_SUCCESS 0x01
318 #define SRB_STATUS_ABORTED 0x02
319 #define SRB_STATUS_ERROR 0x04
320 #define SRB_STATUS_INVALID_REQUEST 0x06
321 #define SRB_STATUS_TIMEOUT 0x09
322 #define SRB_STATUS_SELECTION_TIMEOUT 0x0A
323 #define SRB_STATUS_BUS_RESET 0x0E
324 #define SRB_STATUS_DATA_OVERRUN 0x12
325 #define SRB_STATUS_INVALID_LUN 0x20
326 #define SRB_STATUS_INTERNAL_ERROR 0x30
327
328 #define SRB_STATUS(status) \
329 (status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
330 /*
331 * This is the end of Protocol specific defines.
332 */
333
334 static int storvsc_ringbuffer_size = (128 * 1024);
335 static int aligned_ringbuffer_size;
336 static u32 max_outstanding_req_per_channel;
337 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
338
339 static int storvsc_vcpus_per_sub_channel = 4;
340 static unsigned int storvsc_max_hw_queues;
341
342 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
343 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
344
345 module_param(storvsc_max_hw_queues, uint, 0644);
346 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
347
348 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
349 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
350
351 static int ring_avail_percent_lowater = 10;
352 module_param(ring_avail_percent_lowater, int, S_IRUGO);
353 MODULE_PARM_DESC(ring_avail_percent_lowater,
354 "Select a channel if available ring size > this in percent");
355
356 /*
357 * Timeout in seconds for all devices managed by this driver.
358 */
359 static int storvsc_timeout = 180;
360
361 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
362 static struct scsi_transport_template *fc_transport_template;
363 #endif
364
365 static struct scsi_host_template scsi_driver;
366 static void storvsc_on_channel_callback(void *context);
367
368 #define STORVSC_MAX_LUNS_PER_TARGET 255
369 #define STORVSC_MAX_TARGETS 2
370 #define STORVSC_MAX_CHANNELS 8
371
372 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255
373 #define STORVSC_FC_MAX_TARGETS 128
374 #define STORVSC_FC_MAX_CHANNELS 8
375 #define STORVSC_FC_MAX_XFER_SIZE ((u32)(512 * 1024))
376
377 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64
378 #define STORVSC_IDE_MAX_TARGETS 1
379 #define STORVSC_IDE_MAX_CHANNELS 1
380
381 /*
382 * Upper bound on the size of a storvsc packet.
383 */
384 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
385 sizeof(struct vstor_packet))
386
387 struct storvsc_cmd_request {
388 struct scsi_cmnd *cmd;
389
390 struct hv_device *device;
391
392 /* Synchronize the request/response if needed */
393 struct completion wait_event;
394
395 struct vmbus_channel_packet_multipage_buffer mpb;
396 struct vmbus_packet_mpb_array *payload;
397 u32 payload_sz;
398
399 struct vstor_packet vstor_packet;
400 };
401
402
403 /* A storvsc device is a device object that contains a vmbus channel */
404 struct storvsc_device {
405 struct hv_device *device;
406
407 bool destroy;
408 bool drain_notify;
409 atomic_t num_outstanding_req;
410 struct Scsi_Host *host;
411
412 wait_queue_head_t waiting_to_drain;
413
414 /*
415 * Each unique Port/Path/Target represents 1 channel ie scsi
416 * controller. In reality, the pathid, targetid is always 0
417 * and the port is set by us
418 */
419 unsigned int port_number;
420 unsigned char path_id;
421 unsigned char target_id;
422
423 /*
424 * Max I/O, the device can support.
425 */
426 u32 max_transfer_bytes;
427 /*
428 * Number of sub-channels we will open.
429 */
430 u16 num_sc;
431 struct vmbus_channel **stor_chns;
432 /*
433 * Mask of CPUs bound to subchannels.
434 */
435 struct cpumask alloced_cpus;
436 /*
437 * Serializes modifications of stor_chns[] from storvsc_do_io()
438 * and storvsc_change_target_cpu().
439 */
440 spinlock_t lock;
441 /* Used for vsc/vsp channel reset process */
442 struct storvsc_cmd_request init_request;
443 struct storvsc_cmd_request reset_request;
444 /*
445 * Currently active port and node names for FC devices.
446 */
447 u64 node_name;
448 u64 port_name;
449 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
450 struct fc_rport *rport;
451 #endif
452 };
453
454 struct hv_host_device {
455 struct hv_device *dev;
456 unsigned int port;
457 unsigned char path;
458 unsigned char target;
459 struct workqueue_struct *handle_error_wq;
460 struct work_struct host_scan_work;
461 struct Scsi_Host *host;
462 };
463
464 struct storvsc_scan_work {
465 struct work_struct work;
466 struct Scsi_Host *host;
467 u8 lun;
468 u8 tgt_id;
469 };
470
storvsc_device_scan(struct work_struct * work)471 static void storvsc_device_scan(struct work_struct *work)
472 {
473 struct storvsc_scan_work *wrk;
474 struct scsi_device *sdev;
475
476 wrk = container_of(work, struct storvsc_scan_work, work);
477
478 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
479 if (!sdev)
480 goto done;
481 scsi_rescan_device(sdev);
482 scsi_device_put(sdev);
483
484 done:
485 kfree(wrk);
486 }
487
storvsc_host_scan(struct work_struct * work)488 static void storvsc_host_scan(struct work_struct *work)
489 {
490 struct Scsi_Host *host;
491 struct scsi_device *sdev;
492 struct hv_host_device *host_device =
493 container_of(work, struct hv_host_device, host_scan_work);
494
495 host = host_device->host;
496 /*
497 * Before scanning the host, first check to see if any of the
498 * currently known devices have been hot removed. We issue a
499 * "unit ready" command against all currently known devices.
500 * This I/O will result in an error for devices that have been
501 * removed. As part of handling the I/O error, we remove the device.
502 *
503 * When a LUN is added or removed, the host sends us a signal to
504 * scan the host. Thus we are forced to discover the LUNs that
505 * may have been removed this way.
506 */
507 mutex_lock(&host->scan_mutex);
508 shost_for_each_device(sdev, host)
509 scsi_test_unit_ready(sdev, 1, 1, NULL);
510 mutex_unlock(&host->scan_mutex);
511 /*
512 * Now scan the host to discover LUNs that may have been added.
513 */
514 scsi_scan_host(host);
515 }
516
storvsc_remove_lun(struct work_struct * work)517 static void storvsc_remove_lun(struct work_struct *work)
518 {
519 struct storvsc_scan_work *wrk;
520 struct scsi_device *sdev;
521
522 wrk = container_of(work, struct storvsc_scan_work, work);
523 if (!scsi_host_get(wrk->host))
524 goto done;
525
526 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
527
528 if (sdev) {
529 scsi_remove_device(sdev);
530 scsi_device_put(sdev);
531 }
532 scsi_host_put(wrk->host);
533
534 done:
535 kfree(wrk);
536 }
537
538
539 /*
540 * We can get incoming messages from the host that are not in response to
541 * messages that we have sent out. An example of this would be messages
542 * received by the guest to notify dynamic addition/removal of LUNs. To
543 * deal with potential race conditions where the driver may be in the
544 * midst of being unloaded when we might receive an unsolicited message
545 * from the host, we have implemented a mechanism to gurantee sequential
546 * consistency:
547 *
548 * 1) Once the device is marked as being destroyed, we will fail all
549 * outgoing messages.
550 * 2) We permit incoming messages when the device is being destroyed,
551 * only to properly account for messages already sent out.
552 */
553
get_out_stor_device(struct hv_device * device)554 static inline struct storvsc_device *get_out_stor_device(
555 struct hv_device *device)
556 {
557 struct storvsc_device *stor_device;
558
559 stor_device = hv_get_drvdata(device);
560
561 if (stor_device && stor_device->destroy)
562 stor_device = NULL;
563
564 return stor_device;
565 }
566
567
storvsc_wait_to_drain(struct storvsc_device * dev)568 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
569 {
570 dev->drain_notify = true;
571 wait_event(dev->waiting_to_drain,
572 atomic_read(&dev->num_outstanding_req) == 0);
573 dev->drain_notify = false;
574 }
575
get_in_stor_device(struct hv_device * device)576 static inline struct storvsc_device *get_in_stor_device(
577 struct hv_device *device)
578 {
579 struct storvsc_device *stor_device;
580
581 stor_device = hv_get_drvdata(device);
582
583 if (!stor_device)
584 goto get_in_err;
585
586 /*
587 * If the device is being destroyed; allow incoming
588 * traffic only to cleanup outstanding requests.
589 */
590
591 if (stor_device->destroy &&
592 (atomic_read(&stor_device->num_outstanding_req) == 0))
593 stor_device = NULL;
594
595 get_in_err:
596 return stor_device;
597
598 }
599
storvsc_change_target_cpu(struct vmbus_channel * channel,u32 old,u32 new)600 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
601 u32 new)
602 {
603 struct storvsc_device *stor_device;
604 struct vmbus_channel *cur_chn;
605 bool old_is_alloced = false;
606 struct hv_device *device;
607 unsigned long flags;
608 int cpu;
609
610 device = channel->primary_channel ?
611 channel->primary_channel->device_obj
612 : channel->device_obj;
613 stor_device = get_out_stor_device(device);
614 if (!stor_device)
615 return;
616
617 /* See storvsc_do_io() -> get_og_chn(). */
618 spin_lock_irqsave(&stor_device->lock, flags);
619
620 /*
621 * Determines if the storvsc device has other channels assigned to
622 * the "old" CPU to update the alloced_cpus mask and the stor_chns
623 * array.
624 */
625 if (device->channel != channel && device->channel->target_cpu == old) {
626 cur_chn = device->channel;
627 old_is_alloced = true;
628 goto old_is_alloced;
629 }
630 list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
631 if (cur_chn == channel)
632 continue;
633 if (cur_chn->target_cpu == old) {
634 old_is_alloced = true;
635 goto old_is_alloced;
636 }
637 }
638
639 old_is_alloced:
640 if (old_is_alloced)
641 WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
642 else
643 cpumask_clear_cpu(old, &stor_device->alloced_cpus);
644
645 /* "Flush" the stor_chns array. */
646 for_each_possible_cpu(cpu) {
647 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
648 cpu, &stor_device->alloced_cpus))
649 WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
650 }
651
652 WRITE_ONCE(stor_device->stor_chns[new], channel);
653 cpumask_set_cpu(new, &stor_device->alloced_cpus);
654
655 spin_unlock_irqrestore(&stor_device->lock, flags);
656 }
657
storvsc_next_request_id(struct vmbus_channel * channel,u64 rqst_addr)658 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
659 {
660 struct storvsc_cmd_request *request =
661 (struct storvsc_cmd_request *)(unsigned long)rqst_addr;
662
663 if (rqst_addr == VMBUS_RQST_INIT)
664 return VMBUS_RQST_INIT;
665 if (rqst_addr == VMBUS_RQST_RESET)
666 return VMBUS_RQST_RESET;
667
668 /*
669 * Cannot return an ID of 0, which is reserved for an unsolicited
670 * message from Hyper-V.
671 */
672 return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
673 }
674
handle_sc_creation(struct vmbus_channel * new_sc)675 static void handle_sc_creation(struct vmbus_channel *new_sc)
676 {
677 struct hv_device *device = new_sc->primary_channel->device_obj;
678 struct device *dev = &device->device;
679 struct storvsc_device *stor_device;
680 struct vmstorage_channel_properties props;
681 int ret;
682
683 stor_device = get_out_stor_device(device);
684 if (!stor_device)
685 return;
686
687 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
688 new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
689
690 new_sc->next_request_id_callback = storvsc_next_request_id;
691
692 ret = vmbus_open(new_sc,
693 aligned_ringbuffer_size,
694 aligned_ringbuffer_size,
695 (void *)&props,
696 sizeof(struct vmstorage_channel_properties),
697 storvsc_on_channel_callback, new_sc);
698
699 /* In case vmbus_open() fails, we don't use the sub-channel. */
700 if (ret != 0) {
701 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
702 return;
703 }
704
705 new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
706
707 /* Add the sub-channel to the array of available channels. */
708 stor_device->stor_chns[new_sc->target_cpu] = new_sc;
709 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
710 }
711
handle_multichannel_storage(struct hv_device * device,int max_chns)712 static void handle_multichannel_storage(struct hv_device *device, int max_chns)
713 {
714 struct device *dev = &device->device;
715 struct storvsc_device *stor_device;
716 int num_sc;
717 struct storvsc_cmd_request *request;
718 struct vstor_packet *vstor_packet;
719 int ret, t;
720
721 /*
722 * If the number of CPUs is artificially restricted, such as
723 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
724 * sub-channels >= the number of CPUs. These sub-channels
725 * should not be created. The primary channel is already created
726 * and assigned to one CPU, so check against # CPUs - 1.
727 */
728 num_sc = min((int)(num_online_cpus() - 1), max_chns);
729 if (!num_sc)
730 return;
731
732 stor_device = get_out_stor_device(device);
733 if (!stor_device)
734 return;
735
736 stor_device->num_sc = num_sc;
737 request = &stor_device->init_request;
738 vstor_packet = &request->vstor_packet;
739
740 /*
741 * Establish a handler for dealing with subchannels.
742 */
743 vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
744
745 /*
746 * Request the host to create sub-channels.
747 */
748 memset(request, 0, sizeof(struct storvsc_cmd_request));
749 init_completion(&request->wait_event);
750 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
751 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
752 vstor_packet->sub_channel_count = num_sc;
753
754 ret = vmbus_sendpacket(device->channel, vstor_packet,
755 sizeof(struct vstor_packet),
756 VMBUS_RQST_INIT,
757 VM_PKT_DATA_INBAND,
758 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
759
760 if (ret != 0) {
761 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
762 return;
763 }
764
765 t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
766 if (t == 0) {
767 dev_err(dev, "Failed to create sub-channel: timed out\n");
768 return;
769 }
770
771 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
772 vstor_packet->status != 0) {
773 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
774 vstor_packet->operation, vstor_packet->status);
775 return;
776 }
777
778 /*
779 * We need to do nothing here, because vmbus_process_offer()
780 * invokes channel->sc_creation_callback, which will open and use
781 * the sub-channel(s).
782 */
783 }
784
cache_wwn(struct storvsc_device * stor_device,struct vstor_packet * vstor_packet)785 static void cache_wwn(struct storvsc_device *stor_device,
786 struct vstor_packet *vstor_packet)
787 {
788 /*
789 * Cache the currently active port and node ww names.
790 */
791 if (vstor_packet->wwn_packet.primary_active) {
792 stor_device->node_name =
793 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
794 stor_device->port_name =
795 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
796 } else {
797 stor_device->node_name =
798 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
799 stor_device->port_name =
800 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
801 }
802 }
803
804
storvsc_execute_vstor_op(struct hv_device * device,struct storvsc_cmd_request * request,bool status_check)805 static int storvsc_execute_vstor_op(struct hv_device *device,
806 struct storvsc_cmd_request *request,
807 bool status_check)
808 {
809 struct storvsc_device *stor_device;
810 struct vstor_packet *vstor_packet;
811 int ret, t;
812
813 stor_device = get_out_stor_device(device);
814 if (!stor_device)
815 return -ENODEV;
816
817 vstor_packet = &request->vstor_packet;
818
819 init_completion(&request->wait_event);
820 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
821
822 ret = vmbus_sendpacket(device->channel, vstor_packet,
823 sizeof(struct vstor_packet),
824 VMBUS_RQST_INIT,
825 VM_PKT_DATA_INBAND,
826 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
827 if (ret != 0)
828 return ret;
829
830 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
831 if (t == 0)
832 return -ETIMEDOUT;
833
834 if (!status_check)
835 return ret;
836
837 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
838 vstor_packet->status != 0)
839 return -EINVAL;
840
841 return ret;
842 }
843
storvsc_channel_init(struct hv_device * device,bool is_fc)844 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
845 {
846 struct storvsc_device *stor_device;
847 struct storvsc_cmd_request *request;
848 struct vstor_packet *vstor_packet;
849 int ret, i;
850 int max_chns;
851 bool process_sub_channels = false;
852
853 stor_device = get_out_stor_device(device);
854 if (!stor_device)
855 return -ENODEV;
856
857 request = &stor_device->init_request;
858 vstor_packet = &request->vstor_packet;
859
860 /*
861 * Now, initiate the vsc/vsp initialization protocol on the open
862 * channel
863 */
864 memset(request, 0, sizeof(struct storvsc_cmd_request));
865 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
866 ret = storvsc_execute_vstor_op(device, request, true);
867 if (ret)
868 return ret;
869 /*
870 * Query host supported protocol version.
871 */
872
873 for (i = 0; i < ARRAY_SIZE(protocol_version); i++) {
874 /* reuse the packet for version range supported */
875 memset(vstor_packet, 0, sizeof(struct vstor_packet));
876 vstor_packet->operation =
877 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
878
879 vstor_packet->version.major_minor = protocol_version[i];
880
881 /*
882 * The revision number is only used in Windows; set it to 0.
883 */
884 vstor_packet->version.revision = 0;
885 ret = storvsc_execute_vstor_op(device, request, false);
886 if (ret != 0)
887 return ret;
888
889 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
890 return -EINVAL;
891
892 if (vstor_packet->status == 0) {
893 vmstor_proto_version = protocol_version[i];
894
895 break;
896 }
897 }
898
899 if (vstor_packet->status != 0) {
900 dev_err(&device->device, "Obsolete Hyper-V version\n");
901 return -EINVAL;
902 }
903
904
905 memset(vstor_packet, 0, sizeof(struct vstor_packet));
906 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
907 ret = storvsc_execute_vstor_op(device, request, true);
908 if (ret != 0)
909 return ret;
910
911 /*
912 * Check to see if multi-channel support is there.
913 * Hosts that implement protocol version of 5.1 and above
914 * support multi-channel.
915 */
916 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
917
918 /*
919 * Allocate state to manage the sub-channels.
920 * We allocate an array based on the numbers of possible CPUs
921 * (Hyper-V does not support cpu online/offline).
922 * This Array will be sparseley populated with unique
923 * channels - primary + sub-channels.
924 * We will however populate all the slots to evenly distribute
925 * the load.
926 */
927 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
928 GFP_KERNEL);
929 if (stor_device->stor_chns == NULL)
930 return -ENOMEM;
931
932 device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
933
934 stor_device->stor_chns[device->channel->target_cpu] = device->channel;
935 cpumask_set_cpu(device->channel->target_cpu,
936 &stor_device->alloced_cpus);
937
938 if (vstor_packet->storage_channel_properties.flags &
939 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
940 process_sub_channels = true;
941
942 stor_device->max_transfer_bytes =
943 vstor_packet->storage_channel_properties.max_transfer_bytes;
944
945 if (!is_fc)
946 goto done;
947
948 /*
949 * For FC devices retrieve FC HBA data.
950 */
951 memset(vstor_packet, 0, sizeof(struct vstor_packet));
952 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
953 ret = storvsc_execute_vstor_op(device, request, true);
954 if (ret != 0)
955 return ret;
956
957 /*
958 * Cache the currently active port and node ww names.
959 */
960 cache_wwn(stor_device, vstor_packet);
961
962 done:
963
964 memset(vstor_packet, 0, sizeof(struct vstor_packet));
965 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
966 ret = storvsc_execute_vstor_op(device, request, true);
967 if (ret != 0)
968 return ret;
969
970 if (process_sub_channels)
971 handle_multichannel_storage(device, max_chns);
972
973 return ret;
974 }
975
storvsc_handle_error(struct vmscsi_request * vm_srb,struct scsi_cmnd * scmnd,struct Scsi_Host * host,u8 asc,u8 ascq)976 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
977 struct scsi_cmnd *scmnd,
978 struct Scsi_Host *host,
979 u8 asc, u8 ascq)
980 {
981 struct storvsc_scan_work *wrk;
982 void (*process_err_fn)(struct work_struct *work);
983 struct hv_host_device *host_dev = shost_priv(host);
984
985 switch (SRB_STATUS(vm_srb->srb_status)) {
986 case SRB_STATUS_ERROR:
987 case SRB_STATUS_ABORTED:
988 case SRB_STATUS_INVALID_REQUEST:
989 case SRB_STATUS_INTERNAL_ERROR:
990 case SRB_STATUS_TIMEOUT:
991 case SRB_STATUS_SELECTION_TIMEOUT:
992 case SRB_STATUS_BUS_RESET:
993 case SRB_STATUS_DATA_OVERRUN:
994 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) {
995 /* Check for capacity change */
996 if ((asc == 0x2a) && (ascq == 0x9)) {
997 process_err_fn = storvsc_device_scan;
998 /* Retry the I/O that triggered this. */
999 set_host_byte(scmnd, DID_REQUEUE);
1000 goto do_work;
1001 }
1002
1003 /*
1004 * Check for "Operating parameters have changed"
1005 * due to Hyper-V changing the VHD/VHDX BlockSize
1006 * when adding/removing a differencing disk. This
1007 * causes discard_granularity to change, so do a
1008 * rescan to pick up the new granularity. We don't
1009 * want scsi_report_sense() to output a message
1010 * that a sysadmin wouldn't know what to do with.
1011 */
1012 if ((asc == 0x3f) && (ascq != 0x03) &&
1013 (ascq != 0x0e)) {
1014 process_err_fn = storvsc_device_scan;
1015 set_host_byte(scmnd, DID_REQUEUE);
1016 goto do_work;
1017 }
1018
1019 /*
1020 * Otherwise, let upper layer deal with the
1021 * error when sense message is present
1022 */
1023 return;
1024 }
1025
1026 /*
1027 * If there is an error; offline the device since all
1028 * error recovery strategies would have already been
1029 * deployed on the host side. However, if the command
1030 * were a pass-through command deal with it appropriately.
1031 */
1032 switch (scmnd->cmnd[0]) {
1033 case ATA_16:
1034 case ATA_12:
1035 set_host_byte(scmnd, DID_PASSTHROUGH);
1036 break;
1037 /*
1038 * On some Hyper-V hosts TEST_UNIT_READY command can
1039 * return SRB_STATUS_ERROR. Let the upper level code
1040 * deal with it based on the sense information.
1041 */
1042 case TEST_UNIT_READY:
1043 break;
1044 default:
1045 set_host_byte(scmnd, DID_ERROR);
1046 }
1047 return;
1048
1049 case SRB_STATUS_INVALID_LUN:
1050 set_host_byte(scmnd, DID_NO_CONNECT);
1051 process_err_fn = storvsc_remove_lun;
1052 goto do_work;
1053
1054 }
1055 return;
1056
1057 do_work:
1058 /*
1059 * We need to schedule work to process this error; schedule it.
1060 */
1061 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1062 if (!wrk) {
1063 set_host_byte(scmnd, DID_BAD_TARGET);
1064 return;
1065 }
1066
1067 wrk->host = host;
1068 wrk->lun = vm_srb->lun;
1069 wrk->tgt_id = vm_srb->target_id;
1070 INIT_WORK(&wrk->work, process_err_fn);
1071 queue_work(host_dev->handle_error_wq, &wrk->work);
1072 }
1073
1074
storvsc_command_completion(struct storvsc_cmd_request * cmd_request,struct storvsc_device * stor_dev)1075 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1076 struct storvsc_device *stor_dev)
1077 {
1078 struct scsi_cmnd *scmnd = cmd_request->cmd;
1079 struct scsi_sense_hdr sense_hdr;
1080 struct vmscsi_request *vm_srb;
1081 u32 data_transfer_length;
1082 struct Scsi_Host *host;
1083 u32 payload_sz = cmd_request->payload_sz;
1084 void *payload = cmd_request->payload;
1085 bool sense_ok;
1086
1087 host = stor_dev->host;
1088
1089 vm_srb = &cmd_request->vstor_packet.vm_srb;
1090 data_transfer_length = vm_srb->data_transfer_length;
1091
1092 scmnd->result = vm_srb->scsi_status;
1093
1094 if (scmnd->result) {
1095 sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1096 SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1097
1098 if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1099 scsi_print_sense_hdr(scmnd->device, "storvsc",
1100 &sense_hdr);
1101 }
1102
1103 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1104 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1105 sense_hdr.ascq);
1106 /*
1107 * The Windows driver set data_transfer_length on
1108 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1109 * is untouched. In these cases we set it to 0.
1110 */
1111 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1112 data_transfer_length = 0;
1113 }
1114
1115 /* Validate data_transfer_length (from Hyper-V) */
1116 if (data_transfer_length > cmd_request->payload->range.len)
1117 data_transfer_length = cmd_request->payload->range.len;
1118
1119 scsi_set_resid(scmnd,
1120 cmd_request->payload->range.len - data_transfer_length);
1121
1122 scsi_done(scmnd);
1123
1124 if (payload_sz >
1125 sizeof(struct vmbus_channel_packet_multipage_buffer))
1126 kfree(payload);
1127 }
1128
storvsc_on_io_completion(struct storvsc_device * stor_device,struct vstor_packet * vstor_packet,struct storvsc_cmd_request * request)1129 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1130 struct vstor_packet *vstor_packet,
1131 struct storvsc_cmd_request *request)
1132 {
1133 struct vstor_packet *stor_pkt;
1134 struct hv_device *device = stor_device->device;
1135
1136 stor_pkt = &request->vstor_packet;
1137
1138 /*
1139 * The current SCSI handling on the host side does
1140 * not correctly handle:
1141 * INQUIRY command with page code parameter set to 0x80
1142 * MODE_SENSE command with cmd[2] == 0x1c
1143 * MAINTENANCE_IN is not supported by HyperV FC passthrough
1144 *
1145 * Setup srb and scsi status so this won't be fatal.
1146 * We do this so we can distinguish truly fatal failues
1147 * (srb status == 0x4) and off-line the device in that case.
1148 */
1149
1150 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1151 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE) ||
1152 (stor_pkt->vm_srb.cdb[0] == MAINTENANCE_IN &&
1153 hv_dev_is_fc(device))) {
1154 vstor_packet->vm_srb.scsi_status = 0;
1155 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1156 }
1157
1158 /* Copy over the status...etc */
1159 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1160 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1161
1162 /*
1163 * Copy over the sense_info_length, but limit to the known max
1164 * size if Hyper-V returns a bad value.
1165 */
1166 stor_pkt->vm_srb.sense_info_length = min_t(u8, STORVSC_SENSE_BUFFER_SIZE,
1167 vstor_packet->vm_srb.sense_info_length);
1168
1169 if (vstor_packet->vm_srb.scsi_status != 0 ||
1170 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1171
1172 /*
1173 * Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1174 * return errors when detecting devices using TEST_UNIT_READY,
1175 * and logging these as errors produces unhelpful noise.
1176 */
1177 int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1178 STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1179
1180 storvsc_log(device, loglevel,
1181 "tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1182 scsi_cmd_to_rq(request->cmd)->tag,
1183 stor_pkt->vm_srb.cdb[0],
1184 vstor_packet->vm_srb.scsi_status,
1185 vstor_packet->vm_srb.srb_status,
1186 vstor_packet->status);
1187 }
1188
1189 if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1190 (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1191 memcpy(request->cmd->sense_buffer,
1192 vstor_packet->vm_srb.sense_data,
1193 stor_pkt->vm_srb.sense_info_length);
1194
1195 stor_pkt->vm_srb.data_transfer_length =
1196 vstor_packet->vm_srb.data_transfer_length;
1197
1198 storvsc_command_completion(request, stor_device);
1199
1200 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1201 stor_device->drain_notify)
1202 wake_up(&stor_device->waiting_to_drain);
1203 }
1204
storvsc_on_receive(struct storvsc_device * stor_device,struct vstor_packet * vstor_packet,struct storvsc_cmd_request * request)1205 static void storvsc_on_receive(struct storvsc_device *stor_device,
1206 struct vstor_packet *vstor_packet,
1207 struct storvsc_cmd_request *request)
1208 {
1209 struct hv_host_device *host_dev;
1210 switch (vstor_packet->operation) {
1211 case VSTOR_OPERATION_COMPLETE_IO:
1212 storvsc_on_io_completion(stor_device, vstor_packet, request);
1213 break;
1214
1215 case VSTOR_OPERATION_REMOVE_DEVICE:
1216 case VSTOR_OPERATION_ENUMERATE_BUS:
1217 host_dev = shost_priv(stor_device->host);
1218 queue_work(
1219 host_dev->handle_error_wq, &host_dev->host_scan_work);
1220 break;
1221
1222 case VSTOR_OPERATION_FCHBA_DATA:
1223 cache_wwn(stor_device, vstor_packet);
1224 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1225 fc_host_node_name(stor_device->host) = stor_device->node_name;
1226 fc_host_port_name(stor_device->host) = stor_device->port_name;
1227 #endif
1228 break;
1229 default:
1230 break;
1231 }
1232 }
1233
storvsc_on_channel_callback(void * context)1234 static void storvsc_on_channel_callback(void *context)
1235 {
1236 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1237 const struct vmpacket_descriptor *desc;
1238 struct hv_device *device;
1239 struct storvsc_device *stor_device;
1240 struct Scsi_Host *shost;
1241 unsigned long time_limit = jiffies + msecs_to_jiffies(CALLBACK_TIMEOUT);
1242
1243 if (channel->primary_channel != NULL)
1244 device = channel->primary_channel->device_obj;
1245 else
1246 device = channel->device_obj;
1247
1248 stor_device = get_in_stor_device(device);
1249 if (!stor_device)
1250 return;
1251
1252 shost = stor_device->host;
1253
1254 foreach_vmbus_pkt(desc, channel) {
1255 struct vstor_packet *packet = hv_pkt_data(desc);
1256 struct storvsc_cmd_request *request = NULL;
1257 u32 pktlen = hv_pkt_datalen(desc);
1258 u64 rqst_id = desc->trans_id;
1259 u32 minlen = rqst_id ? sizeof(struct vstor_packet) :
1260 sizeof(enum vstor_packet_operation);
1261
1262 if (unlikely(time_after(jiffies, time_limit))) {
1263 hv_pkt_iter_close(channel);
1264 return;
1265 }
1266
1267 if (pktlen < minlen) {
1268 dev_err(&device->device,
1269 "Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1270 rqst_id, pktlen, minlen);
1271 continue;
1272 }
1273
1274 if (rqst_id == VMBUS_RQST_INIT) {
1275 request = &stor_device->init_request;
1276 } else if (rqst_id == VMBUS_RQST_RESET) {
1277 request = &stor_device->reset_request;
1278 } else {
1279 /* Hyper-V can send an unsolicited message with ID of 0 */
1280 if (rqst_id == 0) {
1281 /*
1282 * storvsc_on_receive() looks at the vstor_packet in the message
1283 * from the ring buffer.
1284 *
1285 * - If the operation in the vstor_packet is COMPLETE_IO, then
1286 * we call storvsc_on_io_completion(), and dereference the
1287 * guest memory address. Make sure we don't call
1288 * storvsc_on_io_completion() with a guest memory address
1289 * that is zero if Hyper-V were to construct and send such
1290 * a bogus packet.
1291 *
1292 * - If the operation in the vstor_packet is FCHBA_DATA, then
1293 * we call cache_wwn(), and access the data payload area of
1294 * the packet (wwn_packet); however, there is no guarantee
1295 * that the packet is big enough to contain such area.
1296 * Future-proof the code by rejecting such a bogus packet.
1297 */
1298 if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1299 packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1300 dev_err(&device->device, "Invalid packet with ID of 0\n");
1301 continue;
1302 }
1303 } else {
1304 struct scsi_cmnd *scmnd;
1305
1306 /* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1307 scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1308 if (scmnd == NULL) {
1309 dev_err(&device->device, "Incorrect transaction ID\n");
1310 continue;
1311 }
1312 request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1313 scsi_dma_unmap(scmnd);
1314 }
1315
1316 storvsc_on_receive(stor_device, packet, request);
1317 continue;
1318 }
1319
1320 memcpy(&request->vstor_packet, packet,
1321 sizeof(struct vstor_packet));
1322 complete(&request->wait_event);
1323 }
1324 }
1325
storvsc_connect_to_vsp(struct hv_device * device,u32 ring_size,bool is_fc)1326 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1327 bool is_fc)
1328 {
1329 struct vmstorage_channel_properties props;
1330 int ret;
1331
1332 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1333
1334 device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1335 device->channel->next_request_id_callback = storvsc_next_request_id;
1336
1337 ret = vmbus_open(device->channel,
1338 ring_size,
1339 ring_size,
1340 (void *)&props,
1341 sizeof(struct vmstorage_channel_properties),
1342 storvsc_on_channel_callback, device->channel);
1343
1344 if (ret != 0)
1345 return ret;
1346
1347 ret = storvsc_channel_init(device, is_fc);
1348
1349 return ret;
1350 }
1351
storvsc_dev_remove(struct hv_device * device)1352 static int storvsc_dev_remove(struct hv_device *device)
1353 {
1354 struct storvsc_device *stor_device;
1355
1356 stor_device = hv_get_drvdata(device);
1357
1358 stor_device->destroy = true;
1359
1360 /* Make sure flag is set before waiting */
1361 wmb();
1362
1363 /*
1364 * At this point, all outbound traffic should be disable. We
1365 * only allow inbound traffic (responses) to proceed so that
1366 * outstanding requests can be completed.
1367 */
1368
1369 storvsc_wait_to_drain(stor_device);
1370
1371 /*
1372 * Since we have already drained, we don't need to busy wait
1373 * as was done in final_release_stor_device()
1374 * Note that we cannot set the ext pointer to NULL until
1375 * we have drained - to drain the outgoing packets, we need to
1376 * allow incoming packets.
1377 */
1378 hv_set_drvdata(device, NULL);
1379
1380 /* Close the channel */
1381 vmbus_close(device->channel);
1382
1383 kfree(stor_device->stor_chns);
1384 kfree(stor_device);
1385 return 0;
1386 }
1387
get_og_chn(struct storvsc_device * stor_device,u16 q_num)1388 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1389 u16 q_num)
1390 {
1391 u16 slot = 0;
1392 u16 hash_qnum;
1393 const struct cpumask *node_mask;
1394 int num_channels, tgt_cpu;
1395
1396 if (stor_device->num_sc == 0) {
1397 stor_device->stor_chns[q_num] = stor_device->device->channel;
1398 return stor_device->device->channel;
1399 }
1400
1401 /*
1402 * Our channel array is sparsley populated and we
1403 * initiated I/O on a processor/hw-q that does not
1404 * currently have a designated channel. Fix this.
1405 * The strategy is simple:
1406 * I. Ensure NUMA locality
1407 * II. Distribute evenly (best effort)
1408 */
1409
1410 node_mask = cpumask_of_node(cpu_to_node(q_num));
1411
1412 num_channels = 0;
1413 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1414 if (cpumask_test_cpu(tgt_cpu, node_mask))
1415 num_channels++;
1416 }
1417 if (num_channels == 0) {
1418 stor_device->stor_chns[q_num] = stor_device->device->channel;
1419 return stor_device->device->channel;
1420 }
1421
1422 hash_qnum = q_num;
1423 while (hash_qnum >= num_channels)
1424 hash_qnum -= num_channels;
1425
1426 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1427 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1428 continue;
1429 if (slot == hash_qnum)
1430 break;
1431 slot++;
1432 }
1433
1434 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1435
1436 return stor_device->stor_chns[q_num];
1437 }
1438
1439
storvsc_do_io(struct hv_device * device,struct storvsc_cmd_request * request,u16 q_num)1440 static int storvsc_do_io(struct hv_device *device,
1441 struct storvsc_cmd_request *request, u16 q_num)
1442 {
1443 struct storvsc_device *stor_device;
1444 struct vstor_packet *vstor_packet;
1445 struct vmbus_channel *outgoing_channel, *channel;
1446 unsigned long flags;
1447 int ret = 0;
1448 const struct cpumask *node_mask;
1449 int tgt_cpu;
1450
1451 vstor_packet = &request->vstor_packet;
1452 stor_device = get_out_stor_device(device);
1453
1454 if (!stor_device)
1455 return -ENODEV;
1456
1457
1458 request->device = device;
1459 /*
1460 * Select an appropriate channel to send the request out.
1461 */
1462 /* See storvsc_change_target_cpu(). */
1463 outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1464 if (outgoing_channel != NULL) {
1465 if (outgoing_channel->target_cpu == q_num) {
1466 /*
1467 * Ideally, we want to pick a different channel if
1468 * available on the same NUMA node.
1469 */
1470 node_mask = cpumask_of_node(cpu_to_node(q_num));
1471 for_each_cpu_wrap(tgt_cpu,
1472 &stor_device->alloced_cpus, q_num + 1) {
1473 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1474 continue;
1475 if (tgt_cpu == q_num)
1476 continue;
1477 channel = READ_ONCE(
1478 stor_device->stor_chns[tgt_cpu]);
1479 if (channel == NULL)
1480 continue;
1481 if (hv_get_avail_to_write_percent(
1482 &channel->outbound)
1483 > ring_avail_percent_lowater) {
1484 outgoing_channel = channel;
1485 goto found_channel;
1486 }
1487 }
1488
1489 /*
1490 * All the other channels on the same NUMA node are
1491 * busy. Try to use the channel on the current CPU
1492 */
1493 if (hv_get_avail_to_write_percent(
1494 &outgoing_channel->outbound)
1495 > ring_avail_percent_lowater)
1496 goto found_channel;
1497
1498 /*
1499 * If we reach here, all the channels on the current
1500 * NUMA node are busy. Try to find a channel in
1501 * other NUMA nodes
1502 */
1503 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1504 if (cpumask_test_cpu(tgt_cpu, node_mask))
1505 continue;
1506 channel = READ_ONCE(
1507 stor_device->stor_chns[tgt_cpu]);
1508 if (channel == NULL)
1509 continue;
1510 if (hv_get_avail_to_write_percent(
1511 &channel->outbound)
1512 > ring_avail_percent_lowater) {
1513 outgoing_channel = channel;
1514 goto found_channel;
1515 }
1516 }
1517 }
1518 } else {
1519 spin_lock_irqsave(&stor_device->lock, flags);
1520 outgoing_channel = stor_device->stor_chns[q_num];
1521 if (outgoing_channel != NULL) {
1522 spin_unlock_irqrestore(&stor_device->lock, flags);
1523 goto found_channel;
1524 }
1525 outgoing_channel = get_og_chn(stor_device, q_num);
1526 spin_unlock_irqrestore(&stor_device->lock, flags);
1527 }
1528
1529 found_channel:
1530 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1531
1532 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1533
1534
1535 vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1536
1537
1538 vstor_packet->vm_srb.data_transfer_length =
1539 request->payload->range.len;
1540
1541 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1542
1543 if (request->payload->range.len) {
1544
1545 ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1546 request->payload, request->payload_sz,
1547 vstor_packet,
1548 sizeof(struct vstor_packet),
1549 (unsigned long)request);
1550 } else {
1551 ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1552 sizeof(struct vstor_packet),
1553 (unsigned long)request,
1554 VM_PKT_DATA_INBAND,
1555 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1556 }
1557
1558 if (ret != 0)
1559 return ret;
1560
1561 atomic_inc(&stor_device->num_outstanding_req);
1562
1563 return ret;
1564 }
1565
storvsc_device_alloc(struct scsi_device * sdevice)1566 static int storvsc_device_alloc(struct scsi_device *sdevice)
1567 {
1568 /*
1569 * Set blist flag to permit the reading of the VPD pages even when
1570 * the target may claim SPC-2 compliance. MSFT targets currently
1571 * claim SPC-2 compliance while they implement post SPC-2 features.
1572 * With this flag we can correctly handle WRITE_SAME_16 issues.
1573 *
1574 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1575 * still supports REPORT LUN.
1576 */
1577 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1578
1579 return 0;
1580 }
1581
storvsc_device_configure(struct scsi_device * sdevice)1582 static int storvsc_device_configure(struct scsi_device *sdevice)
1583 {
1584 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1585
1586 /* storvsc devices don't support MAINTENANCE_IN SCSI cmd */
1587 sdevice->no_report_opcodes = 1;
1588 sdevice->no_write_same = 1;
1589
1590 /*
1591 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1592 * if the device is a MSFT virtual device. If the host is
1593 * WIN10 or newer, allow write_same.
1594 */
1595 if (!strncmp(sdevice->vendor, "Msft", 4)) {
1596 switch (vmstor_proto_version) {
1597 case VMSTOR_PROTO_VERSION_WIN8:
1598 case VMSTOR_PROTO_VERSION_WIN8_1:
1599 sdevice->scsi_level = SCSI_SPC_3;
1600 break;
1601 }
1602
1603 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1604 sdevice->no_write_same = 0;
1605 }
1606
1607 return 0;
1608 }
1609
storvsc_get_chs(struct scsi_device * sdev,struct block_device * bdev,sector_t capacity,int * info)1610 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1611 sector_t capacity, int *info)
1612 {
1613 sector_t nsect = capacity;
1614 sector_t cylinders = nsect;
1615 int heads, sectors_pt;
1616
1617 /*
1618 * We are making up these values; let us keep it simple.
1619 */
1620 heads = 0xff;
1621 sectors_pt = 0x3f; /* Sectors per track */
1622 sector_div(cylinders, heads * sectors_pt);
1623 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1624 cylinders = 0xffff;
1625
1626 info[0] = heads;
1627 info[1] = sectors_pt;
1628 info[2] = (int)cylinders;
1629
1630 return 0;
1631 }
1632
storvsc_host_reset_handler(struct scsi_cmnd * scmnd)1633 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1634 {
1635 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1636 struct hv_device *device = host_dev->dev;
1637
1638 struct storvsc_device *stor_device;
1639 struct storvsc_cmd_request *request;
1640 struct vstor_packet *vstor_packet;
1641 int ret, t;
1642
1643 stor_device = get_out_stor_device(device);
1644 if (!stor_device)
1645 return FAILED;
1646
1647 request = &stor_device->reset_request;
1648 vstor_packet = &request->vstor_packet;
1649 memset(vstor_packet, 0, sizeof(struct vstor_packet));
1650
1651 init_completion(&request->wait_event);
1652
1653 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1654 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1655 vstor_packet->vm_srb.path_id = stor_device->path_id;
1656
1657 ret = vmbus_sendpacket(device->channel, vstor_packet,
1658 sizeof(struct vstor_packet),
1659 VMBUS_RQST_RESET,
1660 VM_PKT_DATA_INBAND,
1661 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1662 if (ret != 0)
1663 return FAILED;
1664
1665 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1666 if (t == 0)
1667 return TIMEOUT_ERROR;
1668
1669
1670 /*
1671 * At this point, all outstanding requests in the adapter
1672 * should have been flushed out and return to us
1673 * There is a potential race here where the host may be in
1674 * the process of responding when we return from here.
1675 * Just wait for all in-transit packets to be accounted for
1676 * before we return from here.
1677 */
1678 storvsc_wait_to_drain(stor_device);
1679
1680 return SUCCESS;
1681 }
1682
1683 /*
1684 * The host guarantees to respond to each command, although I/O latencies might
1685 * be unbounded on Azure. Reset the timer unconditionally to give the host a
1686 * chance to perform EH.
1687 */
storvsc_eh_timed_out(struct scsi_cmnd * scmnd)1688 static enum scsi_timeout_action storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1689 {
1690 return SCSI_EH_RESET_TIMER;
1691 }
1692
storvsc_scsi_cmd_ok(struct scsi_cmnd * scmnd)1693 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1694 {
1695 bool allowed = true;
1696 u8 scsi_op = scmnd->cmnd[0];
1697
1698 switch (scsi_op) {
1699 /* the host does not handle WRITE_SAME, log accident usage */
1700 case WRITE_SAME:
1701 /*
1702 * smartd sends this command and the host does not handle
1703 * this. So, don't send it.
1704 */
1705 case SET_WINDOW:
1706 set_host_byte(scmnd, DID_ERROR);
1707 allowed = false;
1708 break;
1709 default:
1710 break;
1711 }
1712 return allowed;
1713 }
1714
storvsc_queuecommand(struct Scsi_Host * host,struct scsi_cmnd * scmnd)1715 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1716 {
1717 int ret;
1718 struct hv_host_device *host_dev = shost_priv(host);
1719 struct hv_device *dev = host_dev->dev;
1720 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1721 struct scatterlist *sgl;
1722 struct vmscsi_request *vm_srb;
1723 struct vmbus_packet_mpb_array *payload;
1724 u32 payload_sz;
1725 u32 length;
1726
1727 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1728 /*
1729 * On legacy hosts filter unimplemented commands.
1730 * Future hosts are expected to correctly handle
1731 * unsupported commands. Furthermore, it is
1732 * possible that some of the currently
1733 * unsupported commands maybe supported in
1734 * future versions of the host.
1735 */
1736 if (!storvsc_scsi_cmd_ok(scmnd)) {
1737 scsi_done(scmnd);
1738 return 0;
1739 }
1740 }
1741
1742 /* Setup the cmd request */
1743 cmd_request->cmd = scmnd;
1744
1745 memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1746 vm_srb = &cmd_request->vstor_packet.vm_srb;
1747 vm_srb->time_out_value = 60;
1748
1749 vm_srb->srb_flags |=
1750 SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1751
1752 if (scmnd->device->tagged_supported) {
1753 vm_srb->srb_flags |=
1754 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1755 vm_srb->queue_tag = SP_UNTAGGED;
1756 vm_srb->queue_action = SRB_SIMPLE_TAG_REQUEST;
1757 }
1758
1759 /* Build the SRB */
1760 switch (scmnd->sc_data_direction) {
1761 case DMA_TO_DEVICE:
1762 vm_srb->data_in = WRITE_TYPE;
1763 vm_srb->srb_flags |= SRB_FLAGS_DATA_OUT;
1764 break;
1765 case DMA_FROM_DEVICE:
1766 vm_srb->data_in = READ_TYPE;
1767 vm_srb->srb_flags |= SRB_FLAGS_DATA_IN;
1768 break;
1769 case DMA_NONE:
1770 vm_srb->data_in = UNKNOWN_TYPE;
1771 vm_srb->srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1772 break;
1773 default:
1774 /*
1775 * This is DMA_BIDIRECTIONAL or something else we are never
1776 * supposed to see here.
1777 */
1778 WARN(1, "Unexpected data direction: %d\n",
1779 scmnd->sc_data_direction);
1780 return -EINVAL;
1781 }
1782
1783
1784 vm_srb->port_number = host_dev->port;
1785 vm_srb->path_id = scmnd->device->channel;
1786 vm_srb->target_id = scmnd->device->id;
1787 vm_srb->lun = scmnd->device->lun;
1788
1789 vm_srb->cdb_length = scmnd->cmd_len;
1790
1791 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1792
1793 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1794
1795 length = scsi_bufflen(scmnd);
1796 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1797 payload_sz = 0;
1798
1799 if (scsi_sg_count(scmnd)) {
1800 unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1801 unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1802 struct scatterlist *sg;
1803 unsigned long hvpfn, hvpfns_to_add;
1804 int j, i = 0, sg_count;
1805
1806 payload_sz = (hvpg_count * sizeof(u64) +
1807 sizeof(struct vmbus_packet_mpb_array));
1808
1809 if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1810 payload = kzalloc(payload_sz, GFP_ATOMIC);
1811 if (!payload)
1812 return SCSI_MLQUEUE_DEVICE_BUSY;
1813 }
1814
1815 payload->range.len = length;
1816 payload->range.offset = offset_in_hvpg;
1817
1818 sg_count = scsi_dma_map(scmnd);
1819 if (sg_count < 0) {
1820 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1821 goto err_free_payload;
1822 }
1823
1824 for_each_sg(sgl, sg, sg_count, j) {
1825 /*
1826 * Init values for the current sgl entry. hvpfns_to_add
1827 * is in units of Hyper-V size pages. Handling the
1828 * PAGE_SIZE != HV_HYP_PAGE_SIZE case also handles
1829 * values of sgl->offset that are larger than PAGE_SIZE.
1830 * Such offsets are handled even on other than the first
1831 * sgl entry, provided they are a multiple of PAGE_SIZE.
1832 */
1833 hvpfn = HVPFN_DOWN(sg_dma_address(sg));
1834 hvpfns_to_add = HVPFN_UP(sg_dma_address(sg) +
1835 sg_dma_len(sg)) - hvpfn;
1836
1837 /*
1838 * Fill the next portion of the PFN array with
1839 * sequential Hyper-V PFNs for the continguous physical
1840 * memory described by the sgl entry. The end of the
1841 * last sgl should be reached at the same time that
1842 * the PFN array is filled.
1843 */
1844 while (hvpfns_to_add--)
1845 payload->range.pfn_array[i++] = hvpfn++;
1846 }
1847 }
1848
1849 cmd_request->payload = payload;
1850 cmd_request->payload_sz = payload_sz;
1851
1852 /* Invokes the vsc to start an IO */
1853 ret = storvsc_do_io(dev, cmd_request, get_cpu());
1854 put_cpu();
1855
1856 if (ret)
1857 scsi_dma_unmap(scmnd);
1858
1859 if (ret == -EAGAIN) {
1860 /* no more space */
1861 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1862 goto err_free_payload;
1863 }
1864
1865 return 0;
1866
1867 err_free_payload:
1868 if (payload_sz > sizeof(cmd_request->mpb))
1869 kfree(payload);
1870
1871 return ret;
1872 }
1873
1874 static struct scsi_host_template scsi_driver = {
1875 .module = THIS_MODULE,
1876 .name = "storvsc_host_t",
1877 .cmd_size = sizeof(struct storvsc_cmd_request),
1878 .bios_param = storvsc_get_chs,
1879 .queuecommand = storvsc_queuecommand,
1880 .eh_host_reset_handler = storvsc_host_reset_handler,
1881 .proc_name = "storvsc_host",
1882 .eh_timed_out = storvsc_eh_timed_out,
1883 .slave_alloc = storvsc_device_alloc,
1884 .slave_configure = storvsc_device_configure,
1885 .cmd_per_lun = 2048,
1886 .this_id = -1,
1887 /* Ensure there are no gaps in presented sgls */
1888 .virt_boundary_mask = HV_HYP_PAGE_SIZE - 1,
1889 .no_write_same = 1,
1890 .track_queue_depth = 1,
1891 .change_queue_depth = storvsc_change_queue_depth,
1892 };
1893
1894 enum {
1895 SCSI_GUID,
1896 IDE_GUID,
1897 SFC_GUID,
1898 };
1899
1900 static const struct hv_vmbus_device_id id_table[] = {
1901 /* SCSI guid */
1902 { HV_SCSI_GUID,
1903 .driver_data = SCSI_GUID
1904 },
1905 /* IDE guid */
1906 { HV_IDE_GUID,
1907 .driver_data = IDE_GUID
1908 },
1909 /* Fibre Channel GUID */
1910 {
1911 HV_SYNTHFC_GUID,
1912 .driver_data = SFC_GUID
1913 },
1914 { },
1915 };
1916
1917 MODULE_DEVICE_TABLE(vmbus, id_table);
1918
1919 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1920
hv_dev_is_fc(struct hv_device * hv_dev)1921 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1922 {
1923 return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1924 }
1925
storvsc_probe(struct hv_device * device,const struct hv_vmbus_device_id * dev_id)1926 static int storvsc_probe(struct hv_device *device,
1927 const struct hv_vmbus_device_id *dev_id)
1928 {
1929 int ret;
1930 int num_cpus = num_online_cpus();
1931 int num_present_cpus = num_present_cpus();
1932 struct Scsi_Host *host;
1933 struct hv_host_device *host_dev;
1934 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1935 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1936 int target = 0;
1937 struct storvsc_device *stor_device;
1938 int max_sub_channels = 0;
1939 u32 max_xfer_bytes;
1940
1941 /*
1942 * We support sub-channels for storage on SCSI and FC controllers.
1943 * The number of sub-channels offerred is based on the number of
1944 * VCPUs in the guest.
1945 */
1946 if (!dev_is_ide)
1947 max_sub_channels =
1948 (num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1949
1950 scsi_driver.can_queue = max_outstanding_req_per_channel *
1951 (max_sub_channels + 1) *
1952 (100 - ring_avail_percent_lowater) / 100;
1953
1954 host = scsi_host_alloc(&scsi_driver,
1955 sizeof(struct hv_host_device));
1956 if (!host)
1957 return -ENOMEM;
1958
1959 host_dev = shost_priv(host);
1960 memset(host_dev, 0, sizeof(struct hv_host_device));
1961
1962 host_dev->port = host->host_no;
1963 host_dev->dev = device;
1964 host_dev->host = host;
1965
1966
1967 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1968 if (!stor_device) {
1969 ret = -ENOMEM;
1970 goto err_out0;
1971 }
1972
1973 stor_device->destroy = false;
1974 init_waitqueue_head(&stor_device->waiting_to_drain);
1975 stor_device->device = device;
1976 stor_device->host = host;
1977 spin_lock_init(&stor_device->lock);
1978 hv_set_drvdata(device, stor_device);
1979 dma_set_min_align_mask(&device->device, HV_HYP_PAGE_SIZE - 1);
1980
1981 stor_device->port_number = host->host_no;
1982 ret = storvsc_connect_to_vsp(device, aligned_ringbuffer_size, is_fc);
1983 if (ret)
1984 goto err_out1;
1985
1986 host_dev->path = stor_device->path_id;
1987 host_dev->target = stor_device->target_id;
1988
1989 switch (dev_id->driver_data) {
1990 case SFC_GUID:
1991 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1992 host->max_id = STORVSC_FC_MAX_TARGETS;
1993 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1994 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1995 host->transportt = fc_transport_template;
1996 #endif
1997 break;
1998
1999 case SCSI_GUID:
2000 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
2001 host->max_id = STORVSC_MAX_TARGETS;
2002 host->max_channel = STORVSC_MAX_CHANNELS - 1;
2003 break;
2004
2005 default:
2006 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
2007 host->max_id = STORVSC_IDE_MAX_TARGETS;
2008 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2009 break;
2010 }
2011 /* max cmd length */
2012 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2013 /*
2014 * Any reasonable Hyper-V configuration should provide
2015 * max_transfer_bytes value aligning to HV_HYP_PAGE_SIZE,
2016 * protecting it from any weird value.
2017 */
2018 max_xfer_bytes = round_down(stor_device->max_transfer_bytes, HV_HYP_PAGE_SIZE);
2019 if (is_fc)
2020 max_xfer_bytes = min(max_xfer_bytes, STORVSC_FC_MAX_XFER_SIZE);
2021
2022 /* max_hw_sectors_kb */
2023 host->max_sectors = max_xfer_bytes >> 9;
2024 /*
2025 * There are 2 requirements for Hyper-V storvsc sgl segments,
2026 * based on which the below calculation for max segments is
2027 * done:
2028 *
2029 * 1. Except for the first and last sgl segment, all sgl segments
2030 * should be align to HV_HYP_PAGE_SIZE, that also means the
2031 * maximum number of segments in a sgl can be calculated by
2032 * dividing the total max transfer length by HV_HYP_PAGE_SIZE.
2033 *
2034 * 2. Except for the first and last, each entry in the SGL must
2035 * have an offset that is a multiple of HV_HYP_PAGE_SIZE.
2036 */
2037 host->sg_tablesize = (max_xfer_bytes >> HV_HYP_PAGE_SHIFT) + 1;
2038 /*
2039 * For non-IDE disks, the host supports multiple channels.
2040 * Set the number of HW queues we are supporting.
2041 */
2042 if (!dev_is_ide) {
2043 if (storvsc_max_hw_queues > num_present_cpus) {
2044 storvsc_max_hw_queues = 0;
2045 storvsc_log(device, STORVSC_LOGGING_WARN,
2046 "Resetting invalid storvsc_max_hw_queues value to default.\n");
2047 }
2048 if (storvsc_max_hw_queues)
2049 host->nr_hw_queues = storvsc_max_hw_queues;
2050 else
2051 host->nr_hw_queues = num_present_cpus;
2052 }
2053
2054 /*
2055 * Set the error handler work queue.
2056 */
2057 host_dev->handle_error_wq =
2058 alloc_ordered_workqueue("storvsc_error_wq_%d",
2059 0,
2060 host->host_no);
2061 if (!host_dev->handle_error_wq) {
2062 ret = -ENOMEM;
2063 goto err_out2;
2064 }
2065 INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2066 /* Register the HBA and start the scsi bus scan */
2067 ret = scsi_add_host(host, &device->device);
2068 if (ret != 0)
2069 goto err_out3;
2070
2071 if (!dev_is_ide) {
2072 scsi_scan_host(host);
2073 } else {
2074 target = (device->dev_instance.b[5] << 8 |
2075 device->dev_instance.b[4]);
2076 ret = scsi_add_device(host, 0, target, 0);
2077 if (ret)
2078 goto err_out4;
2079 }
2080 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2081 if (host->transportt == fc_transport_template) {
2082 struct fc_rport_identifiers ids = {
2083 .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2084 };
2085
2086 fc_host_node_name(host) = stor_device->node_name;
2087 fc_host_port_name(host) = stor_device->port_name;
2088 stor_device->rport = fc_remote_port_add(host, 0, &ids);
2089 if (!stor_device->rport) {
2090 ret = -ENOMEM;
2091 goto err_out4;
2092 }
2093 }
2094 #endif
2095 return 0;
2096
2097 err_out4:
2098 scsi_remove_host(host);
2099
2100 err_out3:
2101 destroy_workqueue(host_dev->handle_error_wq);
2102
2103 err_out2:
2104 /*
2105 * Once we have connected with the host, we would need to
2106 * invoke storvsc_dev_remove() to rollback this state and
2107 * this call also frees up the stor_device; hence the jump around
2108 * err_out1 label.
2109 */
2110 storvsc_dev_remove(device);
2111 goto err_out0;
2112
2113 err_out1:
2114 kfree(stor_device->stor_chns);
2115 kfree(stor_device);
2116
2117 err_out0:
2118 scsi_host_put(host);
2119 return ret;
2120 }
2121
2122 /* Change a scsi target's queue depth */
storvsc_change_queue_depth(struct scsi_device * sdev,int queue_depth)2123 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2124 {
2125 if (queue_depth > scsi_driver.can_queue)
2126 queue_depth = scsi_driver.can_queue;
2127
2128 return scsi_change_queue_depth(sdev, queue_depth);
2129 }
2130
storvsc_remove(struct hv_device * dev)2131 static void storvsc_remove(struct hv_device *dev)
2132 {
2133 struct storvsc_device *stor_device = hv_get_drvdata(dev);
2134 struct Scsi_Host *host = stor_device->host;
2135 struct hv_host_device *host_dev = shost_priv(host);
2136
2137 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2138 if (host->transportt == fc_transport_template) {
2139 fc_remote_port_delete(stor_device->rport);
2140 fc_remove_host(host);
2141 }
2142 #endif
2143 destroy_workqueue(host_dev->handle_error_wq);
2144 scsi_remove_host(host);
2145 storvsc_dev_remove(dev);
2146 scsi_host_put(host);
2147 }
2148
storvsc_suspend(struct hv_device * hv_dev)2149 static int storvsc_suspend(struct hv_device *hv_dev)
2150 {
2151 struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2152 struct Scsi_Host *host = stor_device->host;
2153 struct hv_host_device *host_dev = shost_priv(host);
2154
2155 storvsc_wait_to_drain(stor_device);
2156
2157 drain_workqueue(host_dev->handle_error_wq);
2158
2159 vmbus_close(hv_dev->channel);
2160
2161 kfree(stor_device->stor_chns);
2162 stor_device->stor_chns = NULL;
2163
2164 cpumask_clear(&stor_device->alloced_cpus);
2165
2166 return 0;
2167 }
2168
storvsc_resume(struct hv_device * hv_dev)2169 static int storvsc_resume(struct hv_device *hv_dev)
2170 {
2171 int ret;
2172
2173 ret = storvsc_connect_to_vsp(hv_dev, aligned_ringbuffer_size,
2174 hv_dev_is_fc(hv_dev));
2175 return ret;
2176 }
2177
2178 static struct hv_driver storvsc_drv = {
2179 .name = KBUILD_MODNAME,
2180 .id_table = id_table,
2181 .probe = storvsc_probe,
2182 .remove = storvsc_remove,
2183 .suspend = storvsc_suspend,
2184 .resume = storvsc_resume,
2185 .driver = {
2186 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2187 },
2188 };
2189
2190 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2191 static struct fc_function_template fc_transport_functions = {
2192 .show_host_node_name = 1,
2193 .show_host_port_name = 1,
2194 };
2195 #endif
2196
storvsc_drv_init(void)2197 static int __init storvsc_drv_init(void)
2198 {
2199 int ret;
2200
2201 /*
2202 * Divide the ring buffer data size (which is 1 page less
2203 * than the ring buffer size since that page is reserved for
2204 * the ring buffer indices) by the max request size (which is
2205 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2206 */
2207 aligned_ringbuffer_size = VMBUS_RING_SIZE(storvsc_ringbuffer_size);
2208 max_outstanding_req_per_channel =
2209 ((aligned_ringbuffer_size - PAGE_SIZE) /
2210 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2211 sizeof(struct vstor_packet) + sizeof(u64),
2212 sizeof(u64)));
2213
2214 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2215 fc_transport_template = fc_attach_transport(&fc_transport_functions);
2216 if (!fc_transport_template)
2217 return -ENODEV;
2218 #endif
2219
2220 ret = vmbus_driver_register(&storvsc_drv);
2221
2222 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2223 if (ret)
2224 fc_release_transport(fc_transport_template);
2225 #endif
2226
2227 return ret;
2228 }
2229
storvsc_drv_exit(void)2230 static void __exit storvsc_drv_exit(void)
2231 {
2232 vmbus_driver_unregister(&storvsc_drv);
2233 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2234 fc_release_transport(fc_transport_template);
2235 #endif
2236 }
2237
2238 MODULE_LICENSE("GPL");
2239 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2240 module_init(storvsc_drv_init);
2241 module_exit(storvsc_drv_exit);
2242