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 #define storvsc_log_ratelimited(dev, level, fmt, ...) \
175 do { \
176 if (do_logging(level)) \
177 dev_warn_ratelimited(&(dev)->device, fmt, ##__VA_ARGS__); \
178 } while (0)
179
180 struct vmscsi_request {
181 u16 length;
182 u8 srb_status;
183 u8 scsi_status;
184
185 u8 port_number;
186 u8 path_id;
187 u8 target_id;
188 u8 lun;
189
190 u8 cdb_length;
191 u8 sense_info_length;
192 u8 data_in;
193 u8 reserved;
194
195 u32 data_transfer_length;
196
197 union {
198 u8 cdb[STORVSC_MAX_CMD_LEN];
199 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
200 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
201 };
202 /*
203 * The following was added in win8.
204 */
205 u16 reserve;
206 u8 queue_tag;
207 u8 queue_action;
208 u32 srb_flags;
209 u32 time_out_value;
210 u32 queue_sort_ey;
211
212 } __attribute((packed));
213
214 /*
215 * The list of windows version in order of preference.
216 */
217
218 static const int protocol_version[] = {
219 VMSTOR_PROTO_VERSION_WIN10,
220 VMSTOR_PROTO_VERSION_WIN8_1,
221 VMSTOR_PROTO_VERSION_WIN8,
222 };
223
224
225 /*
226 * This structure is sent during the initialization phase to get the different
227 * properties of the channel.
228 */
229
230 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1
231
232 struct vmstorage_channel_properties {
233 u32 reserved;
234 u16 max_channel_cnt;
235 u16 reserved1;
236
237 u32 flags;
238 u32 max_transfer_bytes;
239
240 u64 reserved2;
241 } __packed;
242
243 /* This structure is sent during the storage protocol negotiations. */
244 struct vmstorage_protocol_version {
245 /* Major (MSW) and minor (LSW) version numbers. */
246 u16 major_minor;
247
248 /*
249 * Revision number is auto-incremented whenever this file is changed
250 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not
251 * definitely indicate incompatibility--but it does indicate mismatched
252 * builds.
253 * This is only used on the windows side. Just set it to 0.
254 */
255 u16 revision;
256 } __packed;
257
258 /* Channel Property Flags */
259 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
260 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
261
262 struct vstor_packet {
263 /* Requested operation type */
264 enum vstor_packet_operation operation;
265
266 /* Flags - see below for values */
267 u32 flags;
268
269 /* Status of the request returned from the server side. */
270 u32 status;
271
272 /* Data payload area */
273 union {
274 /*
275 * Structure used to forward SCSI commands from the
276 * client to the server.
277 */
278 struct vmscsi_request vm_srb;
279
280 /* Structure used to query channel properties. */
281 struct vmstorage_channel_properties storage_channel_properties;
282
283 /* Used during version negotiations. */
284 struct vmstorage_protocol_version version;
285
286 /* Fibre channel address packet */
287 struct hv_fc_wwn_packet wwn_packet;
288
289 /* Number of sub-channels to create */
290 u16 sub_channel_count;
291
292 /* This will be the maximum of the union members */
293 u8 buffer[0x34];
294 };
295 } __packed;
296
297 /*
298 * Packet Flags:
299 *
300 * This flag indicates that the server should send back a completion for this
301 * packet.
302 */
303
304 #define REQUEST_COMPLETION_FLAG 0x1
305
306 /* Matches Windows-end */
307 enum storvsc_request_type {
308 WRITE_TYPE = 0,
309 READ_TYPE,
310 UNKNOWN_TYPE,
311 };
312
313 /*
314 * SRB status codes and masks. In the 8-bit field, the two high order bits
315 * are flags, while the remaining 6 bits are an integer status code. The
316 * definitions here include only the subset of the integer status codes that
317 * are tested for in this driver.
318 */
319 #define SRB_STATUS_AUTOSENSE_VALID 0x80
320 #define SRB_STATUS_QUEUE_FROZEN 0x40
321
322 /* SRB status integer codes */
323 #define SRB_STATUS_SUCCESS 0x01
324 #define SRB_STATUS_ABORTED 0x02
325 #define SRB_STATUS_ERROR 0x04
326 #define SRB_STATUS_INVALID_REQUEST 0x06
327 #define SRB_STATUS_TIMEOUT 0x09
328 #define SRB_STATUS_SELECTION_TIMEOUT 0x0A
329 #define SRB_STATUS_BUS_RESET 0x0E
330 #define SRB_STATUS_DATA_OVERRUN 0x12
331 #define SRB_STATUS_INVALID_LUN 0x20
332 #define SRB_STATUS_INTERNAL_ERROR 0x30
333
334 #define SRB_STATUS(status) \
335 (status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
336 /*
337 * This is the end of Protocol specific defines.
338 */
339
340 static int storvsc_ringbuffer_size = (128 * 1024);
341 static int aligned_ringbuffer_size;
342 static u32 max_outstanding_req_per_channel;
343 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
344
345 static int storvsc_vcpus_per_sub_channel = 4;
346 static unsigned int storvsc_max_hw_queues;
347
348 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
349 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
350
351 module_param(storvsc_max_hw_queues, uint, 0644);
352 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
353
354 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
355 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
356
357 static int ring_avail_percent_lowater = 10;
358 module_param(ring_avail_percent_lowater, int, S_IRUGO);
359 MODULE_PARM_DESC(ring_avail_percent_lowater,
360 "Select a channel if available ring size > this in percent");
361
362 /*
363 * Timeout in seconds for all devices managed by this driver.
364 */
365 static int storvsc_timeout = 180;
366
367 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
368 static struct scsi_transport_template *fc_transport_template;
369 #endif
370
371 static struct scsi_host_template scsi_driver;
372 static void storvsc_on_channel_callback(void *context);
373
374 #define STORVSC_MAX_LUNS_PER_TARGET 255
375 #define STORVSC_MAX_TARGETS 2
376 #define STORVSC_MAX_CHANNELS 8
377
378 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255
379 #define STORVSC_FC_MAX_TARGETS 128
380 #define STORVSC_FC_MAX_CHANNELS 8
381 #define STORVSC_FC_MAX_XFER_SIZE ((u32)(512 * 1024))
382
383 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64
384 #define STORVSC_IDE_MAX_TARGETS 1
385 #define STORVSC_IDE_MAX_CHANNELS 1
386
387 /*
388 * Upper bound on the size of a storvsc packet.
389 */
390 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
391 sizeof(struct vstor_packet))
392
393 struct storvsc_cmd_request {
394 struct scsi_cmnd *cmd;
395
396 struct hv_device *device;
397
398 /* Synchronize the request/response if needed */
399 struct completion wait_event;
400
401 struct vmbus_channel_packet_multipage_buffer mpb;
402 struct vmbus_packet_mpb_array *payload;
403 u32 payload_sz;
404
405 struct vstor_packet vstor_packet;
406 };
407
408
409 /* A storvsc device is a device object that contains a vmbus channel */
410 struct storvsc_device {
411 struct hv_device *device;
412
413 bool destroy;
414 bool drain_notify;
415 atomic_t num_outstanding_req;
416 struct Scsi_Host *host;
417
418 wait_queue_head_t waiting_to_drain;
419
420 /*
421 * Each unique Port/Path/Target represents 1 channel ie scsi
422 * controller. In reality, the pathid, targetid is always 0
423 * and the port is set by us
424 */
425 unsigned int port_number;
426 unsigned char path_id;
427 unsigned char target_id;
428
429 /*
430 * Max I/O, the device can support.
431 */
432 u32 max_transfer_bytes;
433 /*
434 * Number of sub-channels we will open.
435 */
436 u16 num_sc;
437 struct vmbus_channel **stor_chns;
438 /*
439 * Mask of CPUs bound to subchannels.
440 */
441 struct cpumask alloced_cpus;
442 /*
443 * Serializes modifications of stor_chns[] from storvsc_do_io()
444 * and storvsc_change_target_cpu().
445 */
446 spinlock_t lock;
447 /* Used for vsc/vsp channel reset process */
448 struct storvsc_cmd_request init_request;
449 struct storvsc_cmd_request reset_request;
450 /*
451 * Currently active port and node names for FC devices.
452 */
453 u64 node_name;
454 u64 port_name;
455 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
456 struct fc_rport *rport;
457 #endif
458 };
459
460 struct hv_host_device {
461 struct hv_device *dev;
462 unsigned int port;
463 unsigned char path;
464 unsigned char target;
465 struct workqueue_struct *handle_error_wq;
466 struct work_struct host_scan_work;
467 struct Scsi_Host *host;
468 };
469
470 struct storvsc_scan_work {
471 struct work_struct work;
472 struct Scsi_Host *host;
473 u8 lun;
474 u8 tgt_id;
475 };
476
storvsc_device_scan(struct work_struct * work)477 static void storvsc_device_scan(struct work_struct *work)
478 {
479 struct storvsc_scan_work *wrk;
480 struct scsi_device *sdev;
481
482 wrk = container_of(work, struct storvsc_scan_work, work);
483
484 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
485 if (!sdev)
486 goto done;
487 scsi_rescan_device(sdev);
488 scsi_device_put(sdev);
489
490 done:
491 kfree(wrk);
492 }
493
storvsc_host_scan(struct work_struct * work)494 static void storvsc_host_scan(struct work_struct *work)
495 {
496 struct Scsi_Host *host;
497 struct scsi_device *sdev;
498 struct hv_host_device *host_device =
499 container_of(work, struct hv_host_device, host_scan_work);
500
501 host = host_device->host;
502 /*
503 * Before scanning the host, first check to see if any of the
504 * currently known devices have been hot removed. We issue a
505 * "unit ready" command against all currently known devices.
506 * This I/O will result in an error for devices that have been
507 * removed. As part of handling the I/O error, we remove the device.
508 *
509 * When a LUN is added or removed, the host sends us a signal to
510 * scan the host. Thus we are forced to discover the LUNs that
511 * may have been removed this way.
512 */
513 mutex_lock(&host->scan_mutex);
514 shost_for_each_device(sdev, host)
515 scsi_test_unit_ready(sdev, 1, 1, NULL);
516 mutex_unlock(&host->scan_mutex);
517 /*
518 * Now scan the host to discover LUNs that may have been added.
519 */
520 scsi_scan_host(host);
521 }
522
storvsc_remove_lun(struct work_struct * work)523 static void storvsc_remove_lun(struct work_struct *work)
524 {
525 struct storvsc_scan_work *wrk;
526 struct scsi_device *sdev;
527
528 wrk = container_of(work, struct storvsc_scan_work, work);
529 if (!scsi_host_get(wrk->host))
530 goto done;
531
532 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
533
534 if (sdev) {
535 scsi_remove_device(sdev);
536 scsi_device_put(sdev);
537 }
538 scsi_host_put(wrk->host);
539
540 done:
541 kfree(wrk);
542 }
543
544
545 /*
546 * We can get incoming messages from the host that are not in response to
547 * messages that we have sent out. An example of this would be messages
548 * received by the guest to notify dynamic addition/removal of LUNs. To
549 * deal with potential race conditions where the driver may be in the
550 * midst of being unloaded when we might receive an unsolicited message
551 * from the host, we have implemented a mechanism to gurantee sequential
552 * consistency:
553 *
554 * 1) Once the device is marked as being destroyed, we will fail all
555 * outgoing messages.
556 * 2) We permit incoming messages when the device is being destroyed,
557 * only to properly account for messages already sent out.
558 */
559
get_out_stor_device(struct hv_device * device)560 static inline struct storvsc_device *get_out_stor_device(
561 struct hv_device *device)
562 {
563 struct storvsc_device *stor_device;
564
565 stor_device = hv_get_drvdata(device);
566
567 if (stor_device && stor_device->destroy)
568 stor_device = NULL;
569
570 return stor_device;
571 }
572
573
storvsc_wait_to_drain(struct storvsc_device * dev)574 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
575 {
576 dev->drain_notify = true;
577 wait_event(dev->waiting_to_drain,
578 atomic_read(&dev->num_outstanding_req) == 0);
579 dev->drain_notify = false;
580 }
581
get_in_stor_device(struct hv_device * device)582 static inline struct storvsc_device *get_in_stor_device(
583 struct hv_device *device)
584 {
585 struct storvsc_device *stor_device;
586
587 stor_device = hv_get_drvdata(device);
588
589 if (!stor_device)
590 goto get_in_err;
591
592 /*
593 * If the device is being destroyed; allow incoming
594 * traffic only to cleanup outstanding requests.
595 */
596
597 if (stor_device->destroy &&
598 (atomic_read(&stor_device->num_outstanding_req) == 0))
599 stor_device = NULL;
600
601 get_in_err:
602 return stor_device;
603
604 }
605
storvsc_change_target_cpu(struct vmbus_channel * channel,u32 old,u32 new)606 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
607 u32 new)
608 {
609 struct storvsc_device *stor_device;
610 struct vmbus_channel *cur_chn;
611 bool old_is_alloced = false;
612 struct hv_device *device;
613 unsigned long flags;
614 int cpu;
615
616 device = channel->primary_channel ?
617 channel->primary_channel->device_obj
618 : channel->device_obj;
619 stor_device = get_out_stor_device(device);
620 if (!stor_device)
621 return;
622
623 /* See storvsc_do_io() -> get_og_chn(). */
624 spin_lock_irqsave(&stor_device->lock, flags);
625
626 /*
627 * Determines if the storvsc device has other channels assigned to
628 * the "old" CPU to update the alloced_cpus mask and the stor_chns
629 * array.
630 */
631 if (device->channel != channel && device->channel->target_cpu == old) {
632 cur_chn = device->channel;
633 old_is_alloced = true;
634 goto old_is_alloced;
635 }
636 list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
637 if (cur_chn == channel)
638 continue;
639 if (cur_chn->target_cpu == old) {
640 old_is_alloced = true;
641 goto old_is_alloced;
642 }
643 }
644
645 old_is_alloced:
646 if (old_is_alloced)
647 WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
648 else
649 cpumask_clear_cpu(old, &stor_device->alloced_cpus);
650
651 /* "Flush" the stor_chns array. */
652 for_each_possible_cpu(cpu) {
653 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
654 cpu, &stor_device->alloced_cpus))
655 WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
656 }
657
658 WRITE_ONCE(stor_device->stor_chns[new], channel);
659 cpumask_set_cpu(new, &stor_device->alloced_cpus);
660
661 spin_unlock_irqrestore(&stor_device->lock, flags);
662 }
663
storvsc_next_request_id(struct vmbus_channel * channel,u64 rqst_addr)664 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
665 {
666 struct storvsc_cmd_request *request =
667 (struct storvsc_cmd_request *)(unsigned long)rqst_addr;
668
669 if (rqst_addr == VMBUS_RQST_INIT)
670 return VMBUS_RQST_INIT;
671 if (rqst_addr == VMBUS_RQST_RESET)
672 return VMBUS_RQST_RESET;
673
674 /*
675 * Cannot return an ID of 0, which is reserved for an unsolicited
676 * message from Hyper-V.
677 */
678 return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
679 }
680
handle_sc_creation(struct vmbus_channel * new_sc)681 static void handle_sc_creation(struct vmbus_channel *new_sc)
682 {
683 struct hv_device *device = new_sc->primary_channel->device_obj;
684 struct device *dev = &device->device;
685 struct storvsc_device *stor_device;
686 struct vmstorage_channel_properties props;
687 int ret;
688
689 stor_device = get_out_stor_device(device);
690 if (!stor_device)
691 return;
692
693 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
694 new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
695
696 new_sc->next_request_id_callback = storvsc_next_request_id;
697
698 ret = vmbus_open(new_sc,
699 aligned_ringbuffer_size,
700 aligned_ringbuffer_size,
701 (void *)&props,
702 sizeof(struct vmstorage_channel_properties),
703 storvsc_on_channel_callback, new_sc);
704
705 /* In case vmbus_open() fails, we don't use the sub-channel. */
706 if (ret != 0) {
707 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
708 return;
709 }
710
711 new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
712
713 /* Add the sub-channel to the array of available channels. */
714 stor_device->stor_chns[new_sc->target_cpu] = new_sc;
715 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
716 }
717
handle_multichannel_storage(struct hv_device * device,int max_chns)718 static void handle_multichannel_storage(struct hv_device *device, int max_chns)
719 {
720 struct device *dev = &device->device;
721 struct storvsc_device *stor_device;
722 int num_sc;
723 struct storvsc_cmd_request *request;
724 struct vstor_packet *vstor_packet;
725 int ret, t;
726
727 /*
728 * If the number of CPUs is artificially restricted, such as
729 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
730 * sub-channels >= the number of CPUs. These sub-channels
731 * should not be created. The primary channel is already created
732 * and assigned to one CPU, so check against # CPUs - 1.
733 */
734 num_sc = min((int)(num_online_cpus() - 1), max_chns);
735 if (!num_sc)
736 return;
737
738 stor_device = get_out_stor_device(device);
739 if (!stor_device)
740 return;
741
742 stor_device->num_sc = num_sc;
743 request = &stor_device->init_request;
744 vstor_packet = &request->vstor_packet;
745
746 /*
747 * Establish a handler for dealing with subchannels.
748 */
749 vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
750
751 /*
752 * Request the host to create sub-channels.
753 */
754 memset(request, 0, sizeof(struct storvsc_cmd_request));
755 init_completion(&request->wait_event);
756 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
757 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
758 vstor_packet->sub_channel_count = num_sc;
759
760 ret = vmbus_sendpacket(device->channel, vstor_packet,
761 sizeof(struct vstor_packet),
762 VMBUS_RQST_INIT,
763 VM_PKT_DATA_INBAND,
764 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
765
766 if (ret != 0) {
767 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
768 return;
769 }
770
771 t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
772 if (t == 0) {
773 dev_err(dev, "Failed to create sub-channel: timed out\n");
774 return;
775 }
776
777 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
778 vstor_packet->status != 0) {
779 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
780 vstor_packet->operation, vstor_packet->status);
781 return;
782 }
783
784 /*
785 * We need to do nothing here, because vmbus_process_offer()
786 * invokes channel->sc_creation_callback, which will open and use
787 * the sub-channel(s).
788 */
789 }
790
cache_wwn(struct storvsc_device * stor_device,struct vstor_packet * vstor_packet)791 static void cache_wwn(struct storvsc_device *stor_device,
792 struct vstor_packet *vstor_packet)
793 {
794 /*
795 * Cache the currently active port and node ww names.
796 */
797 if (vstor_packet->wwn_packet.primary_active) {
798 stor_device->node_name =
799 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
800 stor_device->port_name =
801 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
802 } else {
803 stor_device->node_name =
804 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
805 stor_device->port_name =
806 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
807 }
808 }
809
810
storvsc_execute_vstor_op(struct hv_device * device,struct storvsc_cmd_request * request,bool status_check)811 static int storvsc_execute_vstor_op(struct hv_device *device,
812 struct storvsc_cmd_request *request,
813 bool status_check)
814 {
815 struct storvsc_device *stor_device;
816 struct vstor_packet *vstor_packet;
817 int ret, t;
818
819 stor_device = get_out_stor_device(device);
820 if (!stor_device)
821 return -ENODEV;
822
823 vstor_packet = &request->vstor_packet;
824
825 init_completion(&request->wait_event);
826 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
827
828 ret = vmbus_sendpacket(device->channel, vstor_packet,
829 sizeof(struct vstor_packet),
830 VMBUS_RQST_INIT,
831 VM_PKT_DATA_INBAND,
832 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
833 if (ret != 0)
834 return ret;
835
836 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
837 if (t == 0)
838 return -ETIMEDOUT;
839
840 if (!status_check)
841 return ret;
842
843 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
844 vstor_packet->status != 0)
845 return -EINVAL;
846
847 return ret;
848 }
849
storvsc_channel_init(struct hv_device * device,bool is_fc)850 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
851 {
852 struct storvsc_device *stor_device;
853 struct storvsc_cmd_request *request;
854 struct vstor_packet *vstor_packet;
855 int ret, i;
856 int max_chns;
857 bool process_sub_channels = false;
858
859 stor_device = get_out_stor_device(device);
860 if (!stor_device)
861 return -ENODEV;
862
863 request = &stor_device->init_request;
864 vstor_packet = &request->vstor_packet;
865
866 /*
867 * Now, initiate the vsc/vsp initialization protocol on the open
868 * channel
869 */
870 memset(request, 0, sizeof(struct storvsc_cmd_request));
871 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
872 ret = storvsc_execute_vstor_op(device, request, true);
873 if (ret)
874 return ret;
875 /*
876 * Query host supported protocol version.
877 */
878
879 for (i = 0; i < ARRAY_SIZE(protocol_version); i++) {
880 /* reuse the packet for version range supported */
881 memset(vstor_packet, 0, sizeof(struct vstor_packet));
882 vstor_packet->operation =
883 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
884
885 vstor_packet->version.major_minor = protocol_version[i];
886
887 /*
888 * The revision number is only used in Windows; set it to 0.
889 */
890 vstor_packet->version.revision = 0;
891 ret = storvsc_execute_vstor_op(device, request, false);
892 if (ret != 0)
893 return ret;
894
895 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
896 return -EINVAL;
897
898 if (vstor_packet->status == 0) {
899 vmstor_proto_version = protocol_version[i];
900
901 break;
902 }
903 }
904
905 if (vstor_packet->status != 0) {
906 dev_err(&device->device, "Obsolete Hyper-V version\n");
907 return -EINVAL;
908 }
909
910
911 memset(vstor_packet, 0, sizeof(struct vstor_packet));
912 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
913 ret = storvsc_execute_vstor_op(device, request, true);
914 if (ret != 0)
915 return ret;
916
917 /*
918 * Check to see if multi-channel support is there.
919 * Hosts that implement protocol version of 5.1 and above
920 * support multi-channel.
921 */
922 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
923
924 /*
925 * Allocate state to manage the sub-channels.
926 * We allocate an array based on the numbers of possible CPUs
927 * (Hyper-V does not support cpu online/offline).
928 * This Array will be sparseley populated with unique
929 * channels - primary + sub-channels.
930 * We will however populate all the slots to evenly distribute
931 * the load.
932 */
933 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
934 GFP_KERNEL);
935 if (stor_device->stor_chns == NULL)
936 return -ENOMEM;
937
938 device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
939
940 stor_device->stor_chns[device->channel->target_cpu] = device->channel;
941 cpumask_set_cpu(device->channel->target_cpu,
942 &stor_device->alloced_cpus);
943
944 if (vstor_packet->storage_channel_properties.flags &
945 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
946 process_sub_channels = true;
947
948 stor_device->max_transfer_bytes =
949 vstor_packet->storage_channel_properties.max_transfer_bytes;
950
951 if (!is_fc)
952 goto done;
953
954 /*
955 * For FC devices retrieve FC HBA data.
956 */
957 memset(vstor_packet, 0, sizeof(struct vstor_packet));
958 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
959 ret = storvsc_execute_vstor_op(device, request, true);
960 if (ret != 0)
961 return ret;
962
963 /*
964 * Cache the currently active port and node ww names.
965 */
966 cache_wwn(stor_device, vstor_packet);
967
968 done:
969
970 memset(vstor_packet, 0, sizeof(struct vstor_packet));
971 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
972 ret = storvsc_execute_vstor_op(device, request, true);
973 if (ret != 0)
974 return ret;
975
976 if (process_sub_channels)
977 handle_multichannel_storage(device, max_chns);
978
979 return ret;
980 }
981
storvsc_handle_error(struct vmscsi_request * vm_srb,struct scsi_cmnd * scmnd,struct Scsi_Host * host,u8 asc,u8 ascq)982 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
983 struct scsi_cmnd *scmnd,
984 struct Scsi_Host *host,
985 u8 asc, u8 ascq)
986 {
987 struct storvsc_scan_work *wrk;
988 void (*process_err_fn)(struct work_struct *work);
989 struct hv_host_device *host_dev = shost_priv(host);
990
991 switch (SRB_STATUS(vm_srb->srb_status)) {
992 case SRB_STATUS_ERROR:
993 case SRB_STATUS_ABORTED:
994 case SRB_STATUS_INVALID_REQUEST:
995 case SRB_STATUS_INTERNAL_ERROR:
996 case SRB_STATUS_TIMEOUT:
997 case SRB_STATUS_SELECTION_TIMEOUT:
998 case SRB_STATUS_BUS_RESET:
999 case SRB_STATUS_DATA_OVERRUN:
1000 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) {
1001 /* Check for capacity change */
1002 if ((asc == 0x2a) && (ascq == 0x9)) {
1003 process_err_fn = storvsc_device_scan;
1004 /* Retry the I/O that triggered this. */
1005 set_host_byte(scmnd, DID_REQUEUE);
1006 goto do_work;
1007 }
1008
1009 /*
1010 * Check for "Operating parameters have changed"
1011 * due to Hyper-V changing the VHD/VHDX BlockSize
1012 * when adding/removing a differencing disk. This
1013 * causes discard_granularity to change, so do a
1014 * rescan to pick up the new granularity. We don't
1015 * want scsi_report_sense() to output a message
1016 * that a sysadmin wouldn't know what to do with.
1017 */
1018 if ((asc == 0x3f) && (ascq != 0x03) &&
1019 (ascq != 0x0e)) {
1020 process_err_fn = storvsc_device_scan;
1021 set_host_byte(scmnd, DID_REQUEUE);
1022 goto do_work;
1023 }
1024
1025 /*
1026 * Otherwise, let upper layer deal with the
1027 * error when sense message is present
1028 */
1029 return;
1030 }
1031
1032 /*
1033 * If there is an error; offline the device since all
1034 * error recovery strategies would have already been
1035 * deployed on the host side. However, if the command
1036 * were a pass-through command deal with it appropriately.
1037 */
1038 switch (scmnd->cmnd[0]) {
1039 case ATA_16:
1040 case ATA_12:
1041 set_host_byte(scmnd, DID_PASSTHROUGH);
1042 break;
1043 /*
1044 * On some Hyper-V hosts TEST_UNIT_READY command can
1045 * return SRB_STATUS_ERROR. Let the upper level code
1046 * deal with it based on the sense information.
1047 */
1048 case TEST_UNIT_READY:
1049 break;
1050 default:
1051 set_host_byte(scmnd, DID_ERROR);
1052 }
1053 return;
1054
1055 case SRB_STATUS_INVALID_LUN:
1056 set_host_byte(scmnd, DID_NO_CONNECT);
1057 process_err_fn = storvsc_remove_lun;
1058 goto do_work;
1059
1060 }
1061 return;
1062
1063 do_work:
1064 /*
1065 * We need to schedule work to process this error; schedule it.
1066 */
1067 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1068 if (!wrk) {
1069 set_host_byte(scmnd, DID_BAD_TARGET);
1070 return;
1071 }
1072
1073 wrk->host = host;
1074 wrk->lun = vm_srb->lun;
1075 wrk->tgt_id = vm_srb->target_id;
1076 INIT_WORK(&wrk->work, process_err_fn);
1077 queue_work(host_dev->handle_error_wq, &wrk->work);
1078 }
1079
1080
storvsc_command_completion(struct storvsc_cmd_request * cmd_request,struct storvsc_device * stor_dev)1081 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1082 struct storvsc_device *stor_dev)
1083 {
1084 struct scsi_cmnd *scmnd = cmd_request->cmd;
1085 struct scsi_sense_hdr sense_hdr;
1086 struct vmscsi_request *vm_srb;
1087 u32 data_transfer_length;
1088 struct Scsi_Host *host;
1089 u32 payload_sz = cmd_request->payload_sz;
1090 void *payload = cmd_request->payload;
1091 bool sense_ok;
1092
1093 host = stor_dev->host;
1094
1095 vm_srb = &cmd_request->vstor_packet.vm_srb;
1096 data_transfer_length = vm_srb->data_transfer_length;
1097
1098 scmnd->result = vm_srb->scsi_status;
1099
1100 if (scmnd->result) {
1101 sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1102 SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1103
1104 if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1105 scsi_print_sense_hdr(scmnd->device, "storvsc",
1106 &sense_hdr);
1107 }
1108
1109 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1110 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1111 sense_hdr.ascq);
1112 /*
1113 * The Windows driver set data_transfer_length on
1114 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1115 * is untouched. In these cases we set it to 0.
1116 */
1117 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1118 data_transfer_length = 0;
1119 }
1120
1121 /* Validate data_transfer_length (from Hyper-V) */
1122 if (data_transfer_length > cmd_request->payload->range.len)
1123 data_transfer_length = cmd_request->payload->range.len;
1124
1125 scsi_set_resid(scmnd,
1126 cmd_request->payload->range.len - data_transfer_length);
1127
1128 scsi_done(scmnd);
1129
1130 if (payload_sz >
1131 sizeof(struct vmbus_channel_packet_multipage_buffer))
1132 kfree(payload);
1133 }
1134
storvsc_on_io_completion(struct storvsc_device * stor_device,struct vstor_packet * vstor_packet,struct storvsc_cmd_request * request)1135 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1136 struct vstor_packet *vstor_packet,
1137 struct storvsc_cmd_request *request)
1138 {
1139 struct vstor_packet *stor_pkt;
1140 struct hv_device *device = stor_device->device;
1141
1142 stor_pkt = &request->vstor_packet;
1143
1144 /*
1145 * The current SCSI handling on the host side does
1146 * not correctly handle:
1147 * INQUIRY command with page code parameter set to 0x80
1148 * MODE_SENSE command with cmd[2] == 0x1c
1149 * MAINTENANCE_IN is not supported by HyperV FC passthrough
1150 *
1151 * Setup srb and scsi status so this won't be fatal.
1152 * We do this so we can distinguish truly fatal failues
1153 * (srb status == 0x4) and off-line the device in that case.
1154 */
1155
1156 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1157 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE) ||
1158 (stor_pkt->vm_srb.cdb[0] == MAINTENANCE_IN &&
1159 hv_dev_is_fc(device))) {
1160 vstor_packet->vm_srb.scsi_status = 0;
1161 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1162 }
1163
1164 /* Copy over the status...etc */
1165 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1166 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1167
1168 /*
1169 * Copy over the sense_info_length, but limit to the known max
1170 * size if Hyper-V returns a bad value.
1171 */
1172 stor_pkt->vm_srb.sense_info_length = min_t(u8, STORVSC_SENSE_BUFFER_SIZE,
1173 vstor_packet->vm_srb.sense_info_length);
1174
1175 if (vstor_packet->vm_srb.scsi_status != 0 ||
1176 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1177
1178 /*
1179 * Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1180 * return errors when detecting devices using TEST_UNIT_READY,
1181 * and logging these as errors produces unhelpful noise.
1182 */
1183 int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1184 STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1185
1186 storvsc_log_ratelimited(device, loglevel,
1187 "tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1188 scsi_cmd_to_rq(request->cmd)->tag,
1189 stor_pkt->vm_srb.cdb[0],
1190 vstor_packet->vm_srb.scsi_status,
1191 vstor_packet->vm_srb.srb_status,
1192 vstor_packet->status);
1193 }
1194
1195 if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1196 (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1197 memcpy(request->cmd->sense_buffer,
1198 vstor_packet->vm_srb.sense_data,
1199 stor_pkt->vm_srb.sense_info_length);
1200
1201 stor_pkt->vm_srb.data_transfer_length =
1202 vstor_packet->vm_srb.data_transfer_length;
1203
1204 storvsc_command_completion(request, stor_device);
1205
1206 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1207 stor_device->drain_notify)
1208 wake_up(&stor_device->waiting_to_drain);
1209 }
1210
storvsc_on_receive(struct storvsc_device * stor_device,struct vstor_packet * vstor_packet,struct storvsc_cmd_request * request)1211 static void storvsc_on_receive(struct storvsc_device *stor_device,
1212 struct vstor_packet *vstor_packet,
1213 struct storvsc_cmd_request *request)
1214 {
1215 struct hv_host_device *host_dev;
1216 switch (vstor_packet->operation) {
1217 case VSTOR_OPERATION_COMPLETE_IO:
1218 storvsc_on_io_completion(stor_device, vstor_packet, request);
1219 break;
1220
1221 case VSTOR_OPERATION_REMOVE_DEVICE:
1222 case VSTOR_OPERATION_ENUMERATE_BUS:
1223 host_dev = shost_priv(stor_device->host);
1224 queue_work(
1225 host_dev->handle_error_wq, &host_dev->host_scan_work);
1226 break;
1227
1228 case VSTOR_OPERATION_FCHBA_DATA:
1229 cache_wwn(stor_device, vstor_packet);
1230 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1231 fc_host_node_name(stor_device->host) = stor_device->node_name;
1232 fc_host_port_name(stor_device->host) = stor_device->port_name;
1233 #endif
1234 break;
1235 default:
1236 break;
1237 }
1238 }
1239
storvsc_on_channel_callback(void * context)1240 static void storvsc_on_channel_callback(void *context)
1241 {
1242 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1243 const struct vmpacket_descriptor *desc;
1244 struct hv_device *device;
1245 struct storvsc_device *stor_device;
1246 struct Scsi_Host *shost;
1247 unsigned long time_limit = jiffies + msecs_to_jiffies(CALLBACK_TIMEOUT);
1248
1249 if (channel->primary_channel != NULL)
1250 device = channel->primary_channel->device_obj;
1251 else
1252 device = channel->device_obj;
1253
1254 stor_device = get_in_stor_device(device);
1255 if (!stor_device)
1256 return;
1257
1258 shost = stor_device->host;
1259
1260 foreach_vmbus_pkt(desc, channel) {
1261 struct vstor_packet *packet = hv_pkt_data(desc);
1262 struct storvsc_cmd_request *request = NULL;
1263 u32 pktlen = hv_pkt_datalen(desc);
1264 u64 rqst_id = desc->trans_id;
1265 u32 minlen = rqst_id ? sizeof(struct vstor_packet) :
1266 sizeof(enum vstor_packet_operation);
1267
1268 if (unlikely(time_after(jiffies, time_limit))) {
1269 hv_pkt_iter_close(channel);
1270 return;
1271 }
1272
1273 if (pktlen < minlen) {
1274 dev_err(&device->device,
1275 "Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1276 rqst_id, pktlen, minlen);
1277 continue;
1278 }
1279
1280 if (rqst_id == VMBUS_RQST_INIT) {
1281 request = &stor_device->init_request;
1282 } else if (rqst_id == VMBUS_RQST_RESET) {
1283 request = &stor_device->reset_request;
1284 } else {
1285 /* Hyper-V can send an unsolicited message with ID of 0 */
1286 if (rqst_id == 0) {
1287 /*
1288 * storvsc_on_receive() looks at the vstor_packet in the message
1289 * from the ring buffer.
1290 *
1291 * - If the operation in the vstor_packet is COMPLETE_IO, then
1292 * we call storvsc_on_io_completion(), and dereference the
1293 * guest memory address. Make sure we don't call
1294 * storvsc_on_io_completion() with a guest memory address
1295 * that is zero if Hyper-V were to construct and send such
1296 * a bogus packet.
1297 *
1298 * - If the operation in the vstor_packet is FCHBA_DATA, then
1299 * we call cache_wwn(), and access the data payload area of
1300 * the packet (wwn_packet); however, there is no guarantee
1301 * that the packet is big enough to contain such area.
1302 * Future-proof the code by rejecting such a bogus packet.
1303 */
1304 if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1305 packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1306 dev_err(&device->device, "Invalid packet with ID of 0\n");
1307 continue;
1308 }
1309 } else {
1310 struct scsi_cmnd *scmnd;
1311
1312 /* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1313 scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1314 if (scmnd == NULL) {
1315 dev_err(&device->device, "Incorrect transaction ID\n");
1316 continue;
1317 }
1318 request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1319 scsi_dma_unmap(scmnd);
1320 }
1321
1322 storvsc_on_receive(stor_device, packet, request);
1323 continue;
1324 }
1325
1326 memcpy(&request->vstor_packet, packet,
1327 sizeof(struct vstor_packet));
1328 complete(&request->wait_event);
1329 }
1330 }
1331
storvsc_connect_to_vsp(struct hv_device * device,u32 ring_size,bool is_fc)1332 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1333 bool is_fc)
1334 {
1335 struct vmstorage_channel_properties props;
1336 int ret;
1337
1338 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1339
1340 device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1341 device->channel->next_request_id_callback = storvsc_next_request_id;
1342
1343 ret = vmbus_open(device->channel,
1344 ring_size,
1345 ring_size,
1346 (void *)&props,
1347 sizeof(struct vmstorage_channel_properties),
1348 storvsc_on_channel_callback, device->channel);
1349
1350 if (ret != 0)
1351 return ret;
1352
1353 ret = storvsc_channel_init(device, is_fc);
1354
1355 return ret;
1356 }
1357
storvsc_dev_remove(struct hv_device * device)1358 static int storvsc_dev_remove(struct hv_device *device)
1359 {
1360 struct storvsc_device *stor_device;
1361
1362 stor_device = hv_get_drvdata(device);
1363
1364 stor_device->destroy = true;
1365
1366 /* Make sure flag is set before waiting */
1367 wmb();
1368
1369 /*
1370 * At this point, all outbound traffic should be disable. We
1371 * only allow inbound traffic (responses) to proceed so that
1372 * outstanding requests can be completed.
1373 */
1374
1375 storvsc_wait_to_drain(stor_device);
1376
1377 /*
1378 * Since we have already drained, we don't need to busy wait
1379 * as was done in final_release_stor_device()
1380 * Note that we cannot set the ext pointer to NULL until
1381 * we have drained - to drain the outgoing packets, we need to
1382 * allow incoming packets.
1383 */
1384 hv_set_drvdata(device, NULL);
1385
1386 /* Close the channel */
1387 vmbus_close(device->channel);
1388
1389 kfree(stor_device->stor_chns);
1390 kfree(stor_device);
1391 return 0;
1392 }
1393
get_og_chn(struct storvsc_device * stor_device,u16 q_num)1394 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1395 u16 q_num)
1396 {
1397 u16 slot = 0;
1398 u16 hash_qnum;
1399 const struct cpumask *node_mask;
1400 int num_channels, tgt_cpu;
1401
1402 if (stor_device->num_sc == 0) {
1403 stor_device->stor_chns[q_num] = stor_device->device->channel;
1404 return stor_device->device->channel;
1405 }
1406
1407 /*
1408 * Our channel array is sparsley populated and we
1409 * initiated I/O on a processor/hw-q that does not
1410 * currently have a designated channel. Fix this.
1411 * The strategy is simple:
1412 * I. Ensure NUMA locality
1413 * II. Distribute evenly (best effort)
1414 */
1415
1416 node_mask = cpumask_of_node(cpu_to_node(q_num));
1417
1418 num_channels = 0;
1419 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1420 if (cpumask_test_cpu(tgt_cpu, node_mask))
1421 num_channels++;
1422 }
1423 if (num_channels == 0) {
1424 stor_device->stor_chns[q_num] = stor_device->device->channel;
1425 return stor_device->device->channel;
1426 }
1427
1428 hash_qnum = q_num;
1429 while (hash_qnum >= num_channels)
1430 hash_qnum -= num_channels;
1431
1432 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1433 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1434 continue;
1435 if (slot == hash_qnum)
1436 break;
1437 slot++;
1438 }
1439
1440 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1441
1442 return stor_device->stor_chns[q_num];
1443 }
1444
1445
storvsc_do_io(struct hv_device * device,struct storvsc_cmd_request * request,u16 q_num)1446 static int storvsc_do_io(struct hv_device *device,
1447 struct storvsc_cmd_request *request, u16 q_num)
1448 {
1449 struct storvsc_device *stor_device;
1450 struct vstor_packet *vstor_packet;
1451 struct vmbus_channel *outgoing_channel, *channel;
1452 unsigned long flags;
1453 int ret = 0;
1454 const struct cpumask *node_mask;
1455 int tgt_cpu;
1456
1457 vstor_packet = &request->vstor_packet;
1458 stor_device = get_out_stor_device(device);
1459
1460 if (!stor_device)
1461 return -ENODEV;
1462
1463
1464 request->device = device;
1465 /*
1466 * Select an appropriate channel to send the request out.
1467 */
1468 /* See storvsc_change_target_cpu(). */
1469 outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1470 if (outgoing_channel != NULL) {
1471 if (outgoing_channel->target_cpu == q_num) {
1472 /*
1473 * Ideally, we want to pick a different channel if
1474 * available on the same NUMA node.
1475 */
1476 node_mask = cpumask_of_node(cpu_to_node(q_num));
1477 for_each_cpu_wrap(tgt_cpu,
1478 &stor_device->alloced_cpus, q_num + 1) {
1479 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1480 continue;
1481 if (tgt_cpu == q_num)
1482 continue;
1483 channel = READ_ONCE(
1484 stor_device->stor_chns[tgt_cpu]);
1485 if (channel == NULL)
1486 continue;
1487 if (hv_get_avail_to_write_percent(
1488 &channel->outbound)
1489 > ring_avail_percent_lowater) {
1490 outgoing_channel = channel;
1491 goto found_channel;
1492 }
1493 }
1494
1495 /*
1496 * All the other channels on the same NUMA node are
1497 * busy. Try to use the channel on the current CPU
1498 */
1499 if (hv_get_avail_to_write_percent(
1500 &outgoing_channel->outbound)
1501 > ring_avail_percent_lowater)
1502 goto found_channel;
1503
1504 /*
1505 * If we reach here, all the channels on the current
1506 * NUMA node are busy. Try to find a channel in
1507 * other NUMA nodes
1508 */
1509 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1510 if (cpumask_test_cpu(tgt_cpu, node_mask))
1511 continue;
1512 channel = READ_ONCE(
1513 stor_device->stor_chns[tgt_cpu]);
1514 if (channel == NULL)
1515 continue;
1516 if (hv_get_avail_to_write_percent(
1517 &channel->outbound)
1518 > ring_avail_percent_lowater) {
1519 outgoing_channel = channel;
1520 goto found_channel;
1521 }
1522 }
1523 }
1524 } else {
1525 spin_lock_irqsave(&stor_device->lock, flags);
1526 outgoing_channel = stor_device->stor_chns[q_num];
1527 if (outgoing_channel != NULL) {
1528 spin_unlock_irqrestore(&stor_device->lock, flags);
1529 goto found_channel;
1530 }
1531 outgoing_channel = get_og_chn(stor_device, q_num);
1532 spin_unlock_irqrestore(&stor_device->lock, flags);
1533 }
1534
1535 found_channel:
1536 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1537
1538 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1539
1540
1541 vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1542
1543
1544 vstor_packet->vm_srb.data_transfer_length =
1545 request->payload->range.len;
1546
1547 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1548
1549 if (request->payload->range.len) {
1550
1551 ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1552 request->payload, request->payload_sz,
1553 vstor_packet,
1554 sizeof(struct vstor_packet),
1555 (unsigned long)request);
1556 } else {
1557 ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1558 sizeof(struct vstor_packet),
1559 (unsigned long)request,
1560 VM_PKT_DATA_INBAND,
1561 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1562 }
1563
1564 if (ret != 0)
1565 return ret;
1566
1567 atomic_inc(&stor_device->num_outstanding_req);
1568
1569 return ret;
1570 }
1571
storvsc_device_alloc(struct scsi_device * sdevice)1572 static int storvsc_device_alloc(struct scsi_device *sdevice)
1573 {
1574 /*
1575 * Set blist flag to permit the reading of the VPD pages even when
1576 * the target may claim SPC-2 compliance. MSFT targets currently
1577 * claim SPC-2 compliance while they implement post SPC-2 features.
1578 * With this flag we can correctly handle WRITE_SAME_16 issues.
1579 *
1580 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1581 * still supports REPORT LUN.
1582 */
1583 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1584
1585 return 0;
1586 }
1587
storvsc_device_configure(struct scsi_device * sdevice)1588 static int storvsc_device_configure(struct scsi_device *sdevice)
1589 {
1590 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1591
1592 /* storvsc devices don't support MAINTENANCE_IN SCSI cmd */
1593 sdevice->no_report_opcodes = 1;
1594 sdevice->no_write_same = 1;
1595
1596 /*
1597 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1598 * if the device is a MSFT virtual device. If the host is
1599 * WIN10 or newer, allow write_same.
1600 */
1601 if (!strncmp(sdevice->vendor, "Msft", 4)) {
1602 switch (vmstor_proto_version) {
1603 case VMSTOR_PROTO_VERSION_WIN8:
1604 case VMSTOR_PROTO_VERSION_WIN8_1:
1605 sdevice->scsi_level = SCSI_SPC_3;
1606 break;
1607 }
1608
1609 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1610 sdevice->no_write_same = 0;
1611 }
1612
1613 return 0;
1614 }
1615
storvsc_get_chs(struct scsi_device * sdev,struct block_device * bdev,sector_t capacity,int * info)1616 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1617 sector_t capacity, int *info)
1618 {
1619 sector_t nsect = capacity;
1620 sector_t cylinders = nsect;
1621 int heads, sectors_pt;
1622
1623 /*
1624 * We are making up these values; let us keep it simple.
1625 */
1626 heads = 0xff;
1627 sectors_pt = 0x3f; /* Sectors per track */
1628 sector_div(cylinders, heads * sectors_pt);
1629 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1630 cylinders = 0xffff;
1631
1632 info[0] = heads;
1633 info[1] = sectors_pt;
1634 info[2] = (int)cylinders;
1635
1636 return 0;
1637 }
1638
storvsc_host_reset_handler(struct scsi_cmnd * scmnd)1639 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1640 {
1641 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1642 struct hv_device *device = host_dev->dev;
1643
1644 struct storvsc_device *stor_device;
1645 struct storvsc_cmd_request *request;
1646 struct vstor_packet *vstor_packet;
1647 int ret, t;
1648
1649 stor_device = get_out_stor_device(device);
1650 if (!stor_device)
1651 return FAILED;
1652
1653 request = &stor_device->reset_request;
1654 vstor_packet = &request->vstor_packet;
1655 memset(vstor_packet, 0, sizeof(struct vstor_packet));
1656
1657 init_completion(&request->wait_event);
1658
1659 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1660 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1661 vstor_packet->vm_srb.path_id = stor_device->path_id;
1662
1663 ret = vmbus_sendpacket(device->channel, vstor_packet,
1664 sizeof(struct vstor_packet),
1665 VMBUS_RQST_RESET,
1666 VM_PKT_DATA_INBAND,
1667 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1668 if (ret != 0)
1669 return FAILED;
1670
1671 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1672 if (t == 0)
1673 return TIMEOUT_ERROR;
1674
1675
1676 /*
1677 * At this point, all outstanding requests in the adapter
1678 * should have been flushed out and return to us
1679 * There is a potential race here where the host may be in
1680 * the process of responding when we return from here.
1681 * Just wait for all in-transit packets to be accounted for
1682 * before we return from here.
1683 */
1684 storvsc_wait_to_drain(stor_device);
1685
1686 return SUCCESS;
1687 }
1688
1689 /*
1690 * The host guarantees to respond to each command, although I/O latencies might
1691 * be unbounded on Azure. Reset the timer unconditionally to give the host a
1692 * chance to perform EH.
1693 */
storvsc_eh_timed_out(struct scsi_cmnd * scmnd)1694 static enum scsi_timeout_action storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1695 {
1696 return SCSI_EH_RESET_TIMER;
1697 }
1698
storvsc_scsi_cmd_ok(struct scsi_cmnd * scmnd)1699 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1700 {
1701 bool allowed = true;
1702 u8 scsi_op = scmnd->cmnd[0];
1703
1704 switch (scsi_op) {
1705 /* the host does not handle WRITE_SAME, log accident usage */
1706 case WRITE_SAME:
1707 /*
1708 * smartd sends this command and the host does not handle
1709 * this. So, don't send it.
1710 */
1711 case SET_WINDOW:
1712 set_host_byte(scmnd, DID_ERROR);
1713 allowed = false;
1714 break;
1715 default:
1716 break;
1717 }
1718 return allowed;
1719 }
1720
storvsc_queuecommand(struct Scsi_Host * host,struct scsi_cmnd * scmnd)1721 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1722 {
1723 int ret;
1724 struct hv_host_device *host_dev = shost_priv(host);
1725 struct hv_device *dev = host_dev->dev;
1726 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1727 struct scatterlist *sgl;
1728 struct vmscsi_request *vm_srb;
1729 struct vmbus_packet_mpb_array *payload;
1730 u32 payload_sz;
1731 u32 length;
1732
1733 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1734 /*
1735 * On legacy hosts filter unimplemented commands.
1736 * Future hosts are expected to correctly handle
1737 * unsupported commands. Furthermore, it is
1738 * possible that some of the currently
1739 * unsupported commands maybe supported in
1740 * future versions of the host.
1741 */
1742 if (!storvsc_scsi_cmd_ok(scmnd)) {
1743 scsi_done(scmnd);
1744 return 0;
1745 }
1746 }
1747
1748 /* Setup the cmd request */
1749 cmd_request->cmd = scmnd;
1750
1751 memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1752 vm_srb = &cmd_request->vstor_packet.vm_srb;
1753 vm_srb->time_out_value = 60;
1754
1755 vm_srb->srb_flags |=
1756 SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1757
1758 if (scmnd->device->tagged_supported) {
1759 vm_srb->srb_flags |=
1760 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1761 vm_srb->queue_tag = SP_UNTAGGED;
1762 vm_srb->queue_action = SRB_SIMPLE_TAG_REQUEST;
1763 }
1764
1765 /* Build the SRB */
1766 switch (scmnd->sc_data_direction) {
1767 case DMA_TO_DEVICE:
1768 vm_srb->data_in = WRITE_TYPE;
1769 vm_srb->srb_flags |= SRB_FLAGS_DATA_OUT;
1770 break;
1771 case DMA_FROM_DEVICE:
1772 vm_srb->data_in = READ_TYPE;
1773 vm_srb->srb_flags |= SRB_FLAGS_DATA_IN;
1774 break;
1775 case DMA_NONE:
1776 vm_srb->data_in = UNKNOWN_TYPE;
1777 vm_srb->srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1778 break;
1779 default:
1780 /*
1781 * This is DMA_BIDIRECTIONAL or something else we are never
1782 * supposed to see here.
1783 */
1784 WARN(1, "Unexpected data direction: %d\n",
1785 scmnd->sc_data_direction);
1786 return -EINVAL;
1787 }
1788
1789
1790 vm_srb->port_number = host_dev->port;
1791 vm_srb->path_id = scmnd->device->channel;
1792 vm_srb->target_id = scmnd->device->id;
1793 vm_srb->lun = scmnd->device->lun;
1794
1795 vm_srb->cdb_length = scmnd->cmd_len;
1796
1797 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1798
1799 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1800
1801 length = scsi_bufflen(scmnd);
1802 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1803 payload_sz = 0;
1804
1805 if (scsi_sg_count(scmnd)) {
1806 unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1807 unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1808 struct scatterlist *sg;
1809 unsigned long hvpfn, hvpfns_to_add;
1810 int j, i = 0, sg_count;
1811
1812 payload_sz = (hvpg_count * sizeof(u64) +
1813 sizeof(struct vmbus_packet_mpb_array));
1814
1815 if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1816 payload = kzalloc(payload_sz, GFP_ATOMIC);
1817 if (!payload)
1818 return SCSI_MLQUEUE_DEVICE_BUSY;
1819 }
1820
1821 payload->range.len = length;
1822 payload->range.offset = offset_in_hvpg;
1823
1824 sg_count = scsi_dma_map(scmnd);
1825 if (sg_count < 0) {
1826 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1827 goto err_free_payload;
1828 }
1829
1830 for_each_sg(sgl, sg, sg_count, j) {
1831 /*
1832 * Init values for the current sgl entry. hvpfns_to_add
1833 * is in units of Hyper-V size pages. Handling the
1834 * PAGE_SIZE != HV_HYP_PAGE_SIZE case also handles
1835 * values of sgl->offset that are larger than PAGE_SIZE.
1836 * Such offsets are handled even on other than the first
1837 * sgl entry, provided they are a multiple of PAGE_SIZE.
1838 */
1839 hvpfn = HVPFN_DOWN(sg_dma_address(sg));
1840 hvpfns_to_add = HVPFN_UP(sg_dma_address(sg) +
1841 sg_dma_len(sg)) - hvpfn;
1842
1843 /*
1844 * Fill the next portion of the PFN array with
1845 * sequential Hyper-V PFNs for the continguous physical
1846 * memory described by the sgl entry. The end of the
1847 * last sgl should be reached at the same time that
1848 * the PFN array is filled.
1849 */
1850 while (hvpfns_to_add--)
1851 payload->range.pfn_array[i++] = hvpfn++;
1852 }
1853 }
1854
1855 cmd_request->payload = payload;
1856 cmd_request->payload_sz = payload_sz;
1857
1858 /* Invokes the vsc to start an IO */
1859 ret = storvsc_do_io(dev, cmd_request, get_cpu());
1860 put_cpu();
1861
1862 if (ret)
1863 scsi_dma_unmap(scmnd);
1864
1865 if (ret == -EAGAIN) {
1866 /* no more space */
1867 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1868 goto err_free_payload;
1869 }
1870
1871 return 0;
1872
1873 err_free_payload:
1874 if (payload_sz > sizeof(cmd_request->mpb))
1875 kfree(payload);
1876
1877 return ret;
1878 }
1879
1880 static struct scsi_host_template scsi_driver = {
1881 .module = THIS_MODULE,
1882 .name = "storvsc_host_t",
1883 .cmd_size = sizeof(struct storvsc_cmd_request),
1884 .bios_param = storvsc_get_chs,
1885 .queuecommand = storvsc_queuecommand,
1886 .eh_host_reset_handler = storvsc_host_reset_handler,
1887 .proc_name = "storvsc_host",
1888 .eh_timed_out = storvsc_eh_timed_out,
1889 .slave_alloc = storvsc_device_alloc,
1890 .slave_configure = storvsc_device_configure,
1891 .cmd_per_lun = 2048,
1892 .this_id = -1,
1893 /* Ensure there are no gaps in presented sgls */
1894 .virt_boundary_mask = HV_HYP_PAGE_SIZE - 1,
1895 .no_write_same = 1,
1896 .track_queue_depth = 1,
1897 .change_queue_depth = storvsc_change_queue_depth,
1898 };
1899
1900 enum {
1901 SCSI_GUID,
1902 IDE_GUID,
1903 SFC_GUID,
1904 };
1905
1906 static const struct hv_vmbus_device_id id_table[] = {
1907 /* SCSI guid */
1908 { HV_SCSI_GUID,
1909 .driver_data = SCSI_GUID
1910 },
1911 /* IDE guid */
1912 { HV_IDE_GUID,
1913 .driver_data = IDE_GUID
1914 },
1915 /* Fibre Channel GUID */
1916 {
1917 HV_SYNTHFC_GUID,
1918 .driver_data = SFC_GUID
1919 },
1920 { },
1921 };
1922
1923 MODULE_DEVICE_TABLE(vmbus, id_table);
1924
1925 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1926
hv_dev_is_fc(struct hv_device * hv_dev)1927 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1928 {
1929 return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1930 }
1931
storvsc_probe(struct hv_device * device,const struct hv_vmbus_device_id * dev_id)1932 static int storvsc_probe(struct hv_device *device,
1933 const struct hv_vmbus_device_id *dev_id)
1934 {
1935 int ret;
1936 int num_cpus = num_online_cpus();
1937 int num_present_cpus = num_present_cpus();
1938 struct Scsi_Host *host;
1939 struct hv_host_device *host_dev;
1940 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1941 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1942 int target = 0;
1943 struct storvsc_device *stor_device;
1944 int max_sub_channels = 0;
1945 u32 max_xfer_bytes;
1946
1947 /*
1948 * We support sub-channels for storage on SCSI and FC controllers.
1949 * The number of sub-channels offerred is based on the number of
1950 * VCPUs in the guest.
1951 */
1952 if (!dev_is_ide)
1953 max_sub_channels =
1954 (num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1955
1956 scsi_driver.can_queue = max_outstanding_req_per_channel *
1957 (max_sub_channels + 1) *
1958 (100 - ring_avail_percent_lowater) / 100;
1959
1960 host = scsi_host_alloc(&scsi_driver,
1961 sizeof(struct hv_host_device));
1962 if (!host)
1963 return -ENOMEM;
1964
1965 host_dev = shost_priv(host);
1966 memset(host_dev, 0, sizeof(struct hv_host_device));
1967
1968 host_dev->port = host->host_no;
1969 host_dev->dev = device;
1970 host_dev->host = host;
1971
1972
1973 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1974 if (!stor_device) {
1975 ret = -ENOMEM;
1976 goto err_out0;
1977 }
1978
1979 stor_device->destroy = false;
1980 init_waitqueue_head(&stor_device->waiting_to_drain);
1981 stor_device->device = device;
1982 stor_device->host = host;
1983 spin_lock_init(&stor_device->lock);
1984 hv_set_drvdata(device, stor_device);
1985 dma_set_min_align_mask(&device->device, HV_HYP_PAGE_SIZE - 1);
1986
1987 stor_device->port_number = host->host_no;
1988 ret = storvsc_connect_to_vsp(device, aligned_ringbuffer_size, is_fc);
1989 if (ret)
1990 goto err_out1;
1991
1992 host_dev->path = stor_device->path_id;
1993 host_dev->target = stor_device->target_id;
1994
1995 switch (dev_id->driver_data) {
1996 case SFC_GUID:
1997 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1998 host->max_id = STORVSC_FC_MAX_TARGETS;
1999 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
2000 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2001 host->transportt = fc_transport_template;
2002 #endif
2003 break;
2004
2005 case SCSI_GUID:
2006 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
2007 host->max_id = STORVSC_MAX_TARGETS;
2008 host->max_channel = STORVSC_MAX_CHANNELS - 1;
2009 break;
2010
2011 default:
2012 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
2013 host->max_id = STORVSC_IDE_MAX_TARGETS;
2014 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2015 break;
2016 }
2017 /* max cmd length */
2018 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2019 /*
2020 * Any reasonable Hyper-V configuration should provide
2021 * max_transfer_bytes value aligning to HV_HYP_PAGE_SIZE,
2022 * protecting it from any weird value.
2023 */
2024 max_xfer_bytes = round_down(stor_device->max_transfer_bytes, HV_HYP_PAGE_SIZE);
2025 if (is_fc)
2026 max_xfer_bytes = min(max_xfer_bytes, STORVSC_FC_MAX_XFER_SIZE);
2027
2028 /* max_hw_sectors_kb */
2029 host->max_sectors = max_xfer_bytes >> 9;
2030 /*
2031 * There are 2 requirements for Hyper-V storvsc sgl segments,
2032 * based on which the below calculation for max segments is
2033 * done:
2034 *
2035 * 1. Except for the first and last sgl segment, all sgl segments
2036 * should be align to HV_HYP_PAGE_SIZE, that also means the
2037 * maximum number of segments in a sgl can be calculated by
2038 * dividing the total max transfer length by HV_HYP_PAGE_SIZE.
2039 *
2040 * 2. Except for the first and last, each entry in the SGL must
2041 * have an offset that is a multiple of HV_HYP_PAGE_SIZE.
2042 */
2043 host->sg_tablesize = (max_xfer_bytes >> HV_HYP_PAGE_SHIFT) + 1;
2044 /*
2045 * For non-IDE disks, the host supports multiple channels.
2046 * Set the number of HW queues we are supporting.
2047 */
2048 if (!dev_is_ide) {
2049 if (storvsc_max_hw_queues > num_present_cpus) {
2050 storvsc_max_hw_queues = 0;
2051 storvsc_log(device, STORVSC_LOGGING_WARN,
2052 "Resetting invalid storvsc_max_hw_queues value to default.\n");
2053 }
2054 if (storvsc_max_hw_queues)
2055 host->nr_hw_queues = storvsc_max_hw_queues;
2056 else
2057 host->nr_hw_queues = num_present_cpus;
2058 }
2059
2060 /*
2061 * Set the error handler work queue.
2062 */
2063 host_dev->handle_error_wq =
2064 alloc_ordered_workqueue("storvsc_error_wq_%d",
2065 0,
2066 host->host_no);
2067 if (!host_dev->handle_error_wq) {
2068 ret = -ENOMEM;
2069 goto err_out2;
2070 }
2071 INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2072 /* Register the HBA and start the scsi bus scan */
2073 ret = scsi_add_host(host, &device->device);
2074 if (ret != 0)
2075 goto err_out3;
2076
2077 if (!dev_is_ide) {
2078 scsi_scan_host(host);
2079 } else {
2080 target = (device->dev_instance.b[5] << 8 |
2081 device->dev_instance.b[4]);
2082 ret = scsi_add_device(host, 0, target, 0);
2083 if (ret)
2084 goto err_out4;
2085 }
2086 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2087 if (host->transportt == fc_transport_template) {
2088 struct fc_rport_identifiers ids = {
2089 .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2090 };
2091
2092 fc_host_node_name(host) = stor_device->node_name;
2093 fc_host_port_name(host) = stor_device->port_name;
2094 stor_device->rport = fc_remote_port_add(host, 0, &ids);
2095 if (!stor_device->rport) {
2096 ret = -ENOMEM;
2097 goto err_out4;
2098 }
2099 }
2100 #endif
2101 return 0;
2102
2103 err_out4:
2104 scsi_remove_host(host);
2105
2106 err_out3:
2107 destroy_workqueue(host_dev->handle_error_wq);
2108
2109 err_out2:
2110 /*
2111 * Once we have connected with the host, we would need to
2112 * invoke storvsc_dev_remove() to rollback this state and
2113 * this call also frees up the stor_device; hence the jump around
2114 * err_out1 label.
2115 */
2116 storvsc_dev_remove(device);
2117 goto err_out0;
2118
2119 err_out1:
2120 kfree(stor_device->stor_chns);
2121 kfree(stor_device);
2122
2123 err_out0:
2124 scsi_host_put(host);
2125 return ret;
2126 }
2127
2128 /* Change a scsi target's queue depth */
storvsc_change_queue_depth(struct scsi_device * sdev,int queue_depth)2129 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2130 {
2131 if (queue_depth > scsi_driver.can_queue)
2132 queue_depth = scsi_driver.can_queue;
2133
2134 return scsi_change_queue_depth(sdev, queue_depth);
2135 }
2136
storvsc_remove(struct hv_device * dev)2137 static void storvsc_remove(struct hv_device *dev)
2138 {
2139 struct storvsc_device *stor_device = hv_get_drvdata(dev);
2140 struct Scsi_Host *host = stor_device->host;
2141 struct hv_host_device *host_dev = shost_priv(host);
2142
2143 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2144 if (host->transportt == fc_transport_template) {
2145 fc_remote_port_delete(stor_device->rport);
2146 fc_remove_host(host);
2147 }
2148 #endif
2149 destroy_workqueue(host_dev->handle_error_wq);
2150 scsi_remove_host(host);
2151 storvsc_dev_remove(dev);
2152 scsi_host_put(host);
2153 }
2154
storvsc_suspend(struct hv_device * hv_dev)2155 static int storvsc_suspend(struct hv_device *hv_dev)
2156 {
2157 struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2158 struct Scsi_Host *host = stor_device->host;
2159 struct hv_host_device *host_dev = shost_priv(host);
2160
2161 storvsc_wait_to_drain(stor_device);
2162
2163 drain_workqueue(host_dev->handle_error_wq);
2164
2165 vmbus_close(hv_dev->channel);
2166
2167 kfree(stor_device->stor_chns);
2168 stor_device->stor_chns = NULL;
2169
2170 cpumask_clear(&stor_device->alloced_cpus);
2171
2172 return 0;
2173 }
2174
storvsc_resume(struct hv_device * hv_dev)2175 static int storvsc_resume(struct hv_device *hv_dev)
2176 {
2177 int ret;
2178
2179 ret = storvsc_connect_to_vsp(hv_dev, aligned_ringbuffer_size,
2180 hv_dev_is_fc(hv_dev));
2181 return ret;
2182 }
2183
2184 static struct hv_driver storvsc_drv = {
2185 .name = KBUILD_MODNAME,
2186 .id_table = id_table,
2187 .probe = storvsc_probe,
2188 .remove = storvsc_remove,
2189 .suspend = storvsc_suspend,
2190 .resume = storvsc_resume,
2191 .driver = {
2192 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2193 },
2194 };
2195
2196 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2197 static struct fc_function_template fc_transport_functions = {
2198 .show_host_node_name = 1,
2199 .show_host_port_name = 1,
2200 };
2201 #endif
2202
storvsc_drv_init(void)2203 static int __init storvsc_drv_init(void)
2204 {
2205 int ret;
2206
2207 /*
2208 * Divide the ring buffer data size (which is 1 page less
2209 * than the ring buffer size since that page is reserved for
2210 * the ring buffer indices) by the max request size (which is
2211 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2212 */
2213 aligned_ringbuffer_size = VMBUS_RING_SIZE(storvsc_ringbuffer_size);
2214 max_outstanding_req_per_channel =
2215 ((aligned_ringbuffer_size - PAGE_SIZE) /
2216 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2217 sizeof(struct vstor_packet) + sizeof(u64),
2218 sizeof(u64)));
2219
2220 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2221 fc_transport_template = fc_attach_transport(&fc_transport_functions);
2222 if (!fc_transport_template)
2223 return -ENODEV;
2224 #endif
2225
2226 ret = vmbus_driver_register(&storvsc_drv);
2227
2228 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2229 if (ret)
2230 fc_release_transport(fc_transport_template);
2231 #endif
2232
2233 return ret;
2234 }
2235
storvsc_drv_exit(void)2236 static void __exit storvsc_drv_exit(void)
2237 {
2238 vmbus_driver_unregister(&storvsc_drv);
2239 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2240 fc_release_transport(fc_transport_template);
2241 #endif
2242 }
2243
2244 MODULE_LICENSE("GPL");
2245 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2246 module_init(storvsc_drv_init);
2247 module_exit(storvsc_drv_exit);
2248