2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/device.h>
24 #include <linux/blkdev.h>
25 #include <linux/major.h>
26 #include <linux/delay.h>
27 #include <linux/hdreg.h>
28 #include <linux/mutex.h>
29 #include <linux/slab.h>
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_dbg.h>
36 #include "version_info.h"
38 #include "storvsc_api.h"
41 #define BLKVSC_MINORS 64
43 enum blkvsc_device_type {
50 * This request ties the struct request and struct
51 * blkvsc_request/hv_storvsc_request together A struct request may be
52 * represented by 1 or more struct blkvsc_request
54 struct blkvsc_request_group {
57 struct list_head blkvsc_req_list; /* list of blkvsc_requests */
60 struct blkvsc_request {
61 /* blkvsc_request_group.blkvsc_req_list */
62 struct list_head req_entry;
64 /* block_device_context.pending_list */
65 struct list_head pend_entry;
67 /* This may be null if we generate a request internally */
70 struct block_device_context *dev;
72 /* The group this request is part of. Maybe null */
73 struct blkvsc_request_group *group;
75 wait_queue_head_t wevent;
79 sector_t sector_start;
80 unsigned long sector_count;
82 unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
83 unsigned char cmd_len;
84 unsigned char cmnd[MAX_COMMAND_SIZE];
86 struct hv_storvsc_request request;
88 * !!!DO NOT ADD ANYTHING BELOW HERE!!! Otherwise, memory can overlap,
89 * because - The extension buffer falls right here and is pointed to by
91 * Which sounds like a horrible idea, who designed this?
95 /* Per device structure */
96 struct block_device_context {
97 /* point back to our device context */
98 struct hv_device *device_ctx;
99 struct kmem_cache *request_pool;
102 enum blkvsc_device_type device_type;
103 struct list_head pending_list;
105 unsigned char device_id[64];
106 unsigned int device_id_len;
107 int num_outstanding_reqs;
109 int media_not_present;
110 unsigned int sector_size;
114 unsigned char target;
120 static DEFINE_MUTEX(blkvsc_mutex);
121 static int blkvsc_probe(struct device *dev);
122 static int blkvsc_remove(struct device *device);
123 static void blkvsc_shutdown(struct device *device);
125 static int blkvsc_open(struct block_device *bdev, fmode_t mode);
126 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
127 static int blkvsc_media_changed(struct gendisk *gd);
128 static int blkvsc_revalidate_disk(struct gendisk *gd);
129 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
130 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
131 unsigned cmd, unsigned long argument);
132 static void blkvsc_request(struct request_queue *queue);
133 static void blkvsc_request_completion(struct hv_storvsc_request *request);
134 static int blkvsc_do_request(struct block_device_context *blkdev,
135 struct request *req);
136 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
137 void (*request_completion)(struct hv_storvsc_request *));
138 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
139 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
140 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
141 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
142 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
143 static int blkvsc_do_flush(struct block_device_context *blkdev);
144 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
145 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
147 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
148 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
149 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
151 /* The one and only one */
152 static struct storvsc_driver_object g_blkvsc_drv;
154 static const struct block_device_operations block_ops = {
155 .owner = THIS_MODULE,
157 .release = blkvsc_release,
158 .media_changed = blkvsc_media_changed,
159 .revalidate_disk = blkvsc_revalidate_disk,
160 .getgeo = blkvsc_getgeo,
161 .ioctl = blkvsc_ioctl,
165 * blkvsc_drv_init - BlkVsc driver initialization.
167 static int blkvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
169 struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
170 struct hv_driver *drv = &g_blkvsc_drv.base;
173 storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
175 drv->priv = storvsc_drv_obj;
177 /* Callback to client driver to complete the initialization */
178 drv_init(&storvsc_drv_obj->base);
180 drv->driver.name = storvsc_drv_obj->base.name;
182 drv->driver.probe = blkvsc_probe;
183 drv->driver.remove = blkvsc_remove;
184 drv->driver.shutdown = blkvsc_shutdown;
186 /* The driver belongs to vmbus */
187 ret = vmbus_child_driver_register(&drv->driver);
192 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
194 struct device **curr = (struct device **)data;
196 return 1; /* stop iterating */
199 static void blkvsc_drv_exit(void)
201 struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
202 struct hv_driver *drv = &g_blkvsc_drv.base;
203 struct device *current_dev;
210 ret = driver_for_each_device(&drv->driver, NULL,
211 (void *) ¤t_dev,
215 DPRINT_WARN(BLKVSC_DRV,
216 "driver_for_each_device returned %d", ret);
219 if (current_dev == NULL)
222 /* Initiate removal from the top-down */
223 device_unregister(current_dev);
226 if (storvsc_drv_obj->base.cleanup)
227 storvsc_drv_obj->base.cleanup(&storvsc_drv_obj->base);
229 vmbus_child_driver_unregister(&drv->driver);
235 * blkvsc_probe - Add a new device for this driver
237 static int blkvsc_probe(struct device *device)
239 struct hv_driver *drv =
240 drv_to_hv_drv(device->driver);
241 struct storvsc_driver_object *storvsc_drv_obj =
243 struct hv_device *device_obj = device_to_hv_device(device);
245 struct block_device_context *blkdev = NULL;
246 struct storvsc_device_info device_info;
250 static int ide0_registered;
251 static int ide1_registered;
253 DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
255 if (!storvsc_drv_obj->base.dev_add) {
256 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
261 blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
267 INIT_LIST_HEAD(&blkdev->pending_list);
269 /* Initialize what we can here */
270 spin_lock_init(&blkdev->lock);
272 /* ASSERT(sizeof(struct blkvsc_request_group) <= */
273 /* sizeof(struct blkvsc_request)); */
275 blkdev->request_pool = kmem_cache_create(dev_name(&device_obj->device),
276 sizeof(struct blkvsc_request) +
277 storvsc_drv_obj->request_ext_size, 0,
278 SLAB_HWCACHE_ALIGN, NULL);
279 if (!blkdev->request_pool) {
285 /* Call to the vsc driver to add the device */
286 ret = storvsc_drv_obj->base.dev_add(device_obj, &device_info);
288 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
292 blkdev->device_ctx = device_obj;
293 /* this identified the device 0 or 1 */
294 blkdev->target = device_info.target_id;
295 /* this identified the ide ctrl 0 or 1 */
296 blkdev->path = device_info.path_id;
298 dev_set_drvdata(device, blkdev);
300 /* Calculate the major and device num */
301 if (blkdev->path == 0) {
303 devnum = blkdev->path + blkdev->target; /* 0 or 1 */
305 if (!ide0_registered) {
306 ret = register_blkdev(major, "ide");
308 DPRINT_ERR(BLKVSC_DRV,
309 "register_blkdev() failed! ret %d",
316 } else if (blkdev->path == 1) {
318 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
320 if (!ide1_registered) {
321 ret = register_blkdev(major, "ide");
323 DPRINT_ERR(BLKVSC_DRV,
324 "register_blkdev() failed! ret %d",
332 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
337 DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
339 blkdev->gd = alloc_disk(BLKVSC_MINORS);
341 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
346 blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
348 blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
349 blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
350 blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
351 blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
352 blk_queue_dma_alignment(blkdev->gd->queue, 511);
354 blkdev->gd->major = major;
355 if (devnum == 1 || devnum == 3)
356 blkdev->gd->first_minor = BLKVSC_MINORS;
358 blkdev->gd->first_minor = 0;
359 blkdev->gd->fops = &block_ops;
360 blkdev->gd->private_data = blkdev;
361 blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
362 sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
364 blkvsc_do_inquiry(blkdev);
365 if (blkdev->device_type == DVD_TYPE) {
366 set_disk_ro(blkdev->gd, 1);
367 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
368 blkvsc_do_read_capacity(blkdev);
370 blkvsc_do_read_capacity16(blkdev);
373 set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
374 blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
376 add_disk(blkdev->gd);
378 DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
379 blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
380 blkdev->sector_size);
385 storvsc_drv_obj->base.dev_rm(device_obj);
389 if (blkdev->request_pool) {
390 kmem_cache_destroy(blkdev->request_pool);
391 blkdev->request_pool = NULL;
400 static void blkvsc_shutdown(struct device *device)
402 struct block_device_context *blkdev = dev_get_drvdata(device);
408 DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
409 blkdev->users, blkdev->gd->disk_name);
411 spin_lock_irqsave(&blkdev->lock, flags);
413 blkdev->shutting_down = 1;
415 blk_stop_queue(blkdev->gd->queue);
417 spin_unlock_irqrestore(&blkdev->lock, flags);
419 while (blkdev->num_outstanding_reqs) {
420 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
421 blkdev->num_outstanding_reqs);
425 blkvsc_do_flush(blkdev);
427 spin_lock_irqsave(&blkdev->lock, flags);
429 blkvsc_cancel_pending_reqs(blkdev);
431 spin_unlock_irqrestore(&blkdev->lock, flags);
434 static int blkvsc_do_flush(struct block_device_context *blkdev)
436 struct blkvsc_request *blkvsc_req;
438 DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
440 if (blkdev->device_type != HARDDISK_TYPE)
443 blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
447 memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
448 init_waitqueue_head(&blkvsc_req->wevent);
449 blkvsc_req->dev = blkdev;
450 blkvsc_req->req = NULL;
451 blkvsc_req->write = 0;
453 blkvsc_req->request.data_buffer.pfn_array[0] = 0;
454 blkvsc_req->request.data_buffer.offset = 0;
455 blkvsc_req->request.data_buffer.len = 0;
457 blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
458 blkvsc_req->cmd_len = 10;
461 * Set this here since the completion routine may be invoked and
462 * completed before we return
464 blkvsc_req->cond = 0;
465 blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
467 wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
469 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
474 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
475 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
477 struct blkvsc_request *blkvsc_req;
478 struct page *page_buf;
480 unsigned char device_type;
482 DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
484 blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
488 memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
489 page_buf = alloc_page(GFP_KERNEL);
491 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
495 init_waitqueue_head(&blkvsc_req->wevent);
496 blkvsc_req->dev = blkdev;
497 blkvsc_req->req = NULL;
498 blkvsc_req->write = 0;
500 blkvsc_req->request.data_buffer.pfn_array[0] = page_to_pfn(page_buf);
501 blkvsc_req->request.data_buffer.offset = 0;
502 blkvsc_req->request.data_buffer.len = 64;
504 blkvsc_req->cmnd[0] = INQUIRY;
505 blkvsc_req->cmnd[1] = 0x1; /* Get product data */
506 blkvsc_req->cmnd[2] = 0x83; /* mode page 83 */
507 blkvsc_req->cmnd[4] = 64;
508 blkvsc_req->cmd_len = 6;
511 * Set this here since the completion routine may be invoked and
512 * completed before we return
514 blkvsc_req->cond = 0;
516 blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
518 DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
519 blkvsc_req, blkvsc_req->cond);
521 wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
523 buf = kmap(page_buf);
525 /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
527 device_type = buf[0] & 0x1F;
529 if (device_type == 0x0) {
530 blkdev->device_type = HARDDISK_TYPE;
531 } else if (device_type == 0x5) {
532 blkdev->device_type = DVD_TYPE;
534 /* TODO: this is currently unsupported device type */
535 blkdev->device_type = UNKNOWN_DEV_TYPE;
538 DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
540 blkdev->device_id_len = buf[7];
541 if (blkdev->device_id_len > 64)
542 blkdev->device_id_len = 64;
544 memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
545 /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
546 * blkdev->device_id_len); */
550 __free_page(page_buf);
552 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
557 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
558 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
560 struct blkvsc_request *blkvsc_req;
561 struct page *page_buf;
563 struct scsi_sense_hdr sense_hdr;
565 DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
567 blkdev->sector_size = 0;
568 blkdev->capacity = 0;
569 blkdev->media_not_present = 0; /* assume a disk is present */
571 blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
575 memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
576 page_buf = alloc_page(GFP_KERNEL);
578 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
582 init_waitqueue_head(&blkvsc_req->wevent);
583 blkvsc_req->dev = blkdev;
584 blkvsc_req->req = NULL;
585 blkvsc_req->write = 0;
587 blkvsc_req->request.data_buffer.pfn_array[0] = page_to_pfn(page_buf);
588 blkvsc_req->request.data_buffer.offset = 0;
589 blkvsc_req->request.data_buffer.len = 8;
591 blkvsc_req->cmnd[0] = READ_CAPACITY;
592 blkvsc_req->cmd_len = 16;
595 * Set this here since the completion routine may be invoked
596 * and completed before we return
598 blkvsc_req->cond = 0;
600 blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
602 DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
603 blkvsc_req, blkvsc_req->cond);
605 wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
608 if (blkvsc_req->request.status) {
609 scsi_normalize_sense(blkvsc_req->sense_buffer,
610 SCSI_SENSE_BUFFERSIZE, &sense_hdr);
612 if (sense_hdr.asc == 0x3A) {
613 /* Medium not present */
614 blkdev->media_not_present = 1;
618 buf = kmap(page_buf);
621 blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
622 (buf[2] << 8) | buf[3]) + 1;
623 blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
624 (buf[6] << 8) | buf[7];
628 __free_page(page_buf);
630 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
635 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
637 struct blkvsc_request *blkvsc_req;
638 struct page *page_buf;
640 struct scsi_sense_hdr sense_hdr;
642 DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
644 blkdev->sector_size = 0;
645 blkdev->capacity = 0;
646 blkdev->media_not_present = 0; /* assume a disk is present */
648 blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
652 memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
653 page_buf = alloc_page(GFP_KERNEL);
655 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
659 init_waitqueue_head(&blkvsc_req->wevent);
660 blkvsc_req->dev = blkdev;
661 blkvsc_req->req = NULL;
662 blkvsc_req->write = 0;
664 blkvsc_req->request.data_buffer.pfn_array[0] = page_to_pfn(page_buf);
665 blkvsc_req->request.data_buffer.offset = 0;
666 blkvsc_req->request.data_buffer.len = 12;
668 blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
669 blkvsc_req->cmd_len = 16;
672 * Set this here since the completion routine may be invoked
673 * and completed before we return
675 blkvsc_req->cond = 0;
677 blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
679 DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
680 blkvsc_req, blkvsc_req->cond);
682 wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
685 if (blkvsc_req->request.status) {
686 scsi_normalize_sense(blkvsc_req->sense_buffer,
687 SCSI_SENSE_BUFFERSIZE, &sense_hdr);
688 if (sense_hdr.asc == 0x3A) {
689 /* Medium not present */
690 blkdev->media_not_present = 1;
694 buf = kmap(page_buf);
697 blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
698 blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
701 blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
702 (buf[2] << 8) | buf[3]) + 1;
703 blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
704 (buf[6] << 8) | buf[7];
709 __free_page(page_buf);
711 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
717 * blkvsc_remove() - Callback when our device is removed
719 static int blkvsc_remove(struct device *device)
721 struct hv_driver *drv =
722 drv_to_hv_drv(device->driver);
723 struct storvsc_driver_object *storvsc_drv_obj =
725 struct hv_device *device_obj = device_to_hv_device(device);
726 struct block_device_context *blkdev = dev_get_drvdata(device);
730 DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
732 if (!storvsc_drv_obj->base.dev_rm)
736 * Call to the vsc driver to let it know that the device is being
739 ret = storvsc_drv_obj->base.dev_rm(device_obj);
742 DPRINT_ERR(BLKVSC_DRV,
743 "unable to remove blkvsc device (ret %d)", ret);
746 /* Get to a known state */
747 spin_lock_irqsave(&blkdev->lock, flags);
749 blkdev->shutting_down = 1;
751 blk_stop_queue(blkdev->gd->queue);
753 spin_unlock_irqrestore(&blkdev->lock, flags);
755 while (blkdev->num_outstanding_reqs) {
756 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
757 blkdev->num_outstanding_reqs);
761 blkvsc_do_flush(blkdev);
763 spin_lock_irqsave(&blkdev->lock, flags);
765 blkvsc_cancel_pending_reqs(blkdev);
767 spin_unlock_irqrestore(&blkdev->lock, flags);
769 blk_cleanup_queue(blkdev->gd->queue);
771 del_gendisk(blkdev->gd);
773 kmem_cache_destroy(blkdev->request_pool);
780 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
782 /* ASSERT(blkvsc_req->req); */
783 /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
785 blkvsc_req->cmd_len = 16;
787 if (blkvsc_req->sector_start > 0xffffffff) {
788 if (rq_data_dir(blkvsc_req->req)) {
789 blkvsc_req->write = 1;
790 blkvsc_req->cmnd[0] = WRITE_16;
792 blkvsc_req->write = 0;
793 blkvsc_req->cmnd[0] = READ_16;
796 blkvsc_req->cmnd[1] |=
797 (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
799 *(unsigned long long *)&blkvsc_req->cmnd[2] =
800 cpu_to_be64(blkvsc_req->sector_start);
801 *(unsigned int *)&blkvsc_req->cmnd[10] =
802 cpu_to_be32(blkvsc_req->sector_count);
803 } else if ((blkvsc_req->sector_count > 0xff) ||
804 (blkvsc_req->sector_start > 0x1fffff)) {
805 if (rq_data_dir(blkvsc_req->req)) {
806 blkvsc_req->write = 1;
807 blkvsc_req->cmnd[0] = WRITE_10;
809 blkvsc_req->write = 0;
810 blkvsc_req->cmnd[0] = READ_10;
813 blkvsc_req->cmnd[1] |=
814 (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
816 *(unsigned int *)&blkvsc_req->cmnd[2] =
817 cpu_to_be32(blkvsc_req->sector_start);
818 *(unsigned short *)&blkvsc_req->cmnd[7] =
819 cpu_to_be16(blkvsc_req->sector_count);
821 if (rq_data_dir(blkvsc_req->req)) {
822 blkvsc_req->write = 1;
823 blkvsc_req->cmnd[0] = WRITE_6;
825 blkvsc_req->write = 0;
826 blkvsc_req->cmnd[0] = READ_6;
829 *(unsigned int *)&blkvsc_req->cmnd[1] =
830 cpu_to_be32(blkvsc_req->sector_start) >> 8;
831 blkvsc_req->cmnd[1] &= 0x1f;
832 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
836 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
837 void (*request_completion)(struct hv_storvsc_request *))
839 struct block_device_context *blkdev = blkvsc_req->dev;
840 struct hv_device *device_ctx = blkdev->device_ctx;
841 struct hv_driver *drv =
842 drv_to_hv_drv(device_ctx->device.driver);
843 struct storvsc_driver_object *storvsc_drv_obj =
845 struct hv_storvsc_request *storvsc_req;
848 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
849 "req %p type %s start_sector %lu count %ld offset %d "
850 "len %d\n", blkvsc_req,
851 (blkvsc_req->write) ? "WRITE" : "READ",
852 (unsigned long) blkvsc_req->sector_start,
853 blkvsc_req->sector_count,
854 blkvsc_req->request.data_buffer.offset,
855 blkvsc_req->request.data_buffer.len);
857 for (i = 0; i < (blkvsc_req->request.data_buffer.len >> 12); i++) {
858 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
859 "req %p pfn[%d] %llx\n",
861 blkvsc_req->request.data_buffer.pfn_array[i]);
865 storvsc_req = &blkvsc_req->request;
866 storvsc_req->extension = (void *)((unsigned long)blkvsc_req +
867 sizeof(struct blkvsc_request));
869 storvsc_req->type = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
871 storvsc_req->on_io_completion = request_completion;
872 storvsc_req->context = blkvsc_req;
874 storvsc_req->host = blkdev->port;
875 storvsc_req->bus = blkdev->path;
876 storvsc_req->target_id = blkdev->target;
877 storvsc_req->lun_id = 0; /* this is not really used at all */
879 storvsc_req->cdb_len = blkvsc_req->cmd_len;
880 storvsc_req->cdb = blkvsc_req->cmnd;
882 storvsc_req->sense_buffer = blkvsc_req->sense_buffer;
883 storvsc_req->sense_buffer_size = SCSI_SENSE_BUFFERSIZE;
885 ret = storvsc_drv_obj->on_io_request(blkdev->device_ctx,
886 &blkvsc_req->request);
888 blkdev->num_outstanding_reqs++;
894 * We break the request into 1 or more blkvsc_requests and submit
895 * them. If we cant submit them all, we put them on the
896 * pending_list. The blkvsc_request() will work on the pending_list.
898 static int blkvsc_do_request(struct block_device_context *blkdev,
901 struct bio *bio = NULL;
902 struct bio_vec *bvec = NULL;
903 struct bio_vec *prev_bvec = NULL;
904 struct blkvsc_request *blkvsc_req = NULL;
905 struct blkvsc_request *tmp;
908 sector_t start_sector;
909 unsigned long num_sectors = 0;
912 struct blkvsc_request_group *group = NULL;
914 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
915 (unsigned long)blk_rq_pos(req));
917 /* Create a group to tie req to list of blkvsc_reqs */
918 group = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
922 INIT_LIST_HEAD(&group->blkvsc_req_list);
923 group->outstanding = group->status = 0;
925 start_sector = blk_rq_pos(req);
927 /* foreach bio in the request */
929 for (bio = req->bio; bio; bio = bio->bi_next) {
931 * Map this bio into an existing or new storvsc request
933 bio_for_each_segment(bvec, bio, seg_idx) {
934 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
935 "- req %p bio %p bvec %p seg_idx %d "
936 "databuf_idx %d\n", req, bio, bvec,
937 seg_idx, databuf_idx);
939 /* Get a new storvsc request */
942 (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
943 /* hole at the begin of page */
944 || (bvec->bv_offset != 0) ||
945 /* hold at the end of page */
947 (prev_bvec->bv_len != PAGE_SIZE))) {
948 /* submit the prev one */
950 blkvsc_req->sector_start = start_sector;
951 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
953 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
954 blkvsc_init_rw(blkvsc_req);
958 * Create new blkvsc_req to represent
961 blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
963 /* free up everything */
964 list_for_each_entry_safe(
966 &group->blkvsc_req_list,
968 list_del(&blkvsc_req->req_entry);
969 kmem_cache_free(blkdev->request_pool, blkvsc_req);
972 kmem_cache_free(blkdev->request_pool, group);
976 memset(blkvsc_req, 0,
977 sizeof(struct blkvsc_request));
979 blkvsc_req->dev = blkdev;
980 blkvsc_req->req = req;
981 blkvsc_req->request.data_buffer.offset
983 blkvsc_req->request.data_buffer.len
986 /* Add to the group */
987 blkvsc_req->group = group;
988 blkvsc_req->group->outstanding++;
989 list_add_tail(&blkvsc_req->req_entry,
990 &blkvsc_req->group->blkvsc_req_list);
992 start_sector += num_sectors;
997 /* Add the curr bvec/segment to the curr blkvsc_req */
998 blkvsc_req->request.data_buffer.
999 pfn_array[databuf_idx]
1000 = page_to_pfn(bvec->bv_page);
1001 blkvsc_req->request.data_buffer.len
1007 num_sectors += bvec->bv_len >> 9;
1009 } /* bio_for_each_segment */
1011 } /* rq_for_each_bio */
1014 /* Handle the last one */
1016 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1017 blkdev, req, blkvsc_req->group,
1018 blkvsc_req->group->outstanding);
1020 blkvsc_req->sector_start = start_sector;
1021 sector_div(blkvsc_req->sector_start,
1022 (blkdev->sector_size >> 9));
1024 blkvsc_req->sector_count = num_sectors /
1025 (blkdev->sector_size >> 9);
1027 blkvsc_init_rw(blkvsc_req);
1030 list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1032 DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1033 "pending_list - blkvsc_req %p start_sect %lu"
1034 " sect_count %ld (%lu %ld)\n", blkvsc_req,
1035 (unsigned long)blkvsc_req->sector_start,
1036 blkvsc_req->sector_count,
1037 (unsigned long)start_sector,
1038 (unsigned long)num_sectors);
1040 list_add_tail(&blkvsc_req->pend_entry,
1041 &blkdev->pending_list);
1043 ret = blkvsc_submit_request(blkvsc_req,
1044 blkvsc_request_completion);
1047 list_add_tail(&blkvsc_req->pend_entry,
1048 &blkdev->pending_list);
1051 DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1052 "start_sect %lu sect_count %ld (%lu %ld) "
1053 "ret %d\n", blkvsc_req,
1054 (unsigned long)blkvsc_req->sector_start,
1055 blkvsc_req->sector_count,
1056 (unsigned long)start_sector,
1064 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1066 struct blkvsc_request *blkvsc_req =
1067 (struct blkvsc_request *)request->context;
1068 struct block_device_context *blkdev =
1069 (struct block_device_context *)blkvsc_req->dev;
1070 struct scsi_sense_hdr sense_hdr;
1072 DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1075 blkdev->num_outstanding_reqs--;
1077 if (blkvsc_req->request.status)
1078 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1079 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1080 scsi_print_sense_hdr("blkvsc", &sense_hdr);
1082 blkvsc_req->cond = 1;
1083 wake_up_interruptible(&blkvsc_req->wevent);
1086 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1088 struct blkvsc_request *blkvsc_req =
1089 (struct blkvsc_request *)request->context;
1090 struct block_device_context *blkdev =
1091 (struct block_device_context *)blkvsc_req->dev;
1092 unsigned long flags;
1093 struct blkvsc_request *comp_req, *tmp;
1095 /* ASSERT(blkvsc_req->group); */
1097 DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1098 "sect_start %lu sect_count %ld len %d group outstd %d "
1099 "total outstd %d\n",
1100 blkdev, blkvsc_req, blkvsc_req->group,
1101 (blkvsc_req->write) ? "WRITE" : "READ",
1102 (unsigned long)blkvsc_req->sector_start,
1103 blkvsc_req->sector_count,
1104 blkvsc_req->request.data_buffer.len,
1105 blkvsc_req->group->outstanding,
1106 blkdev->num_outstanding_reqs);
1108 spin_lock_irqsave(&blkdev->lock, flags);
1110 blkdev->num_outstanding_reqs--;
1111 blkvsc_req->group->outstanding--;
1114 * Only start processing when all the blkvsc_reqs are
1115 * completed. This guarantees no out-of-order blkvsc_req
1116 * completion when calling end_that_request_first()
1118 if (blkvsc_req->group->outstanding == 0) {
1119 list_for_each_entry_safe(comp_req, tmp,
1120 &blkvsc_req->group->blkvsc_req_list,
1122 DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1123 "sect_start %lu sect_count %ld\n",
1125 (unsigned long)comp_req->sector_start,
1126 comp_req->sector_count);
1128 list_del(&comp_req->req_entry);
1130 if (!__blk_end_request(comp_req->req,
1131 (!comp_req->request.status ? 0 : -EIO),
1132 comp_req->sector_count * blkdev->sector_size)) {
1134 * All the sectors have been xferred ie the
1137 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1139 kmem_cache_free(blkdev->request_pool,
1143 kmem_cache_free(blkdev->request_pool, comp_req);
1146 if (!blkdev->shutting_down) {
1147 blkvsc_do_pending_reqs(blkdev);
1148 blk_start_queue(blkdev->gd->queue);
1149 blkvsc_request(blkdev->gd->queue);
1153 spin_unlock_irqrestore(&blkdev->lock, flags);
1156 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1158 struct blkvsc_request *pend_req, *tmp;
1159 struct blkvsc_request *comp_req, *tmp2;
1163 DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1165 /* Flush the pending list first */
1166 list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1169 * The pend_req could be part of a partially completed
1170 * request. If so, complete those req first until we
1173 list_for_each_entry_safe(comp_req, tmp2,
1174 &pend_req->group->blkvsc_req_list,
1176 DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1177 "sect_start %lu sect_count %ld\n",
1179 (unsigned long) comp_req->sector_start,
1180 comp_req->sector_count);
1182 if (comp_req == pend_req)
1185 list_del(&comp_req->req_entry);
1187 if (comp_req->req) {
1188 ret = __blk_end_request(comp_req->req,
1189 (!comp_req->request.status ? 0 : -EIO),
1190 comp_req->sector_count *
1191 blkdev->sector_size);
1193 /* FIXME: shouldn't this do more than return? */
1198 kmem_cache_free(blkdev->request_pool, comp_req);
1201 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1204 list_del(&pend_req->pend_entry);
1206 list_del(&pend_req->req_entry);
1208 if (comp_req->req) {
1209 if (!__blk_end_request(pend_req->req, -EIO,
1210 pend_req->sector_count *
1211 blkdev->sector_size)) {
1213 * All the sectors have been xferred ie the
1216 DPRINT_DBG(BLKVSC_DRV,
1217 "blkvsc_cancel_pending_reqs() - "
1218 "req %p COMPLETED\n", pend_req->req);
1219 kmem_cache_free(blkdev->request_pool,
1224 kmem_cache_free(blkdev->request_pool, pend_req);
1231 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1233 struct blkvsc_request *pend_req, *tmp;
1236 /* Flush the pending list first */
1237 list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1239 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1242 ret = blkvsc_submit_request(pend_req,
1243 blkvsc_request_completion);
1247 list_del(&pend_req->pend_entry);
1253 static void blkvsc_request(struct request_queue *queue)
1255 struct block_device_context *blkdev = NULL;
1256 struct request *req;
1259 DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1260 while ((req = blk_peek_request(queue)) != NULL) {
1261 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1263 blkdev = req->rq_disk->private_data;
1264 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1265 blkdev->media_not_present) {
1266 __blk_end_request_cur(req, 0);
1270 ret = blkvsc_do_pending_reqs(blkdev);
1273 DPRINT_DBG(BLKVSC_DRV,
1274 "- stop queue - pending_list not empty\n");
1275 blk_stop_queue(queue);
1279 blk_start_request(req);
1281 ret = blkvsc_do_request(blkdev, req);
1283 DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1284 blk_stop_queue(queue);
1286 } else if (ret < 0) {
1287 DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1288 blk_requeue_request(queue, req);
1289 blk_stop_queue(queue);
1295 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1297 struct block_device_context *blkdev = bdev->bd_disk->private_data;
1299 DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1300 blkdev->gd->disk_name);
1302 mutex_lock(&blkvsc_mutex);
1303 spin_lock(&blkdev->lock);
1305 if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1306 spin_unlock(&blkdev->lock);
1307 check_disk_change(bdev);
1308 spin_lock(&blkdev->lock);
1313 spin_unlock(&blkdev->lock);
1314 mutex_unlock(&blkvsc_mutex);
1318 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1320 struct block_device_context *blkdev = disk->private_data;
1322 DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1323 blkdev->gd->disk_name);
1325 mutex_lock(&blkvsc_mutex);
1326 spin_lock(&blkdev->lock);
1327 if (blkdev->users == 1) {
1328 spin_unlock(&blkdev->lock);
1329 blkvsc_do_flush(blkdev);
1330 spin_lock(&blkdev->lock);
1335 spin_unlock(&blkdev->lock);
1336 mutex_unlock(&blkvsc_mutex);
1340 static int blkvsc_media_changed(struct gendisk *gd)
1342 DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1346 static int blkvsc_revalidate_disk(struct gendisk *gd)
1348 struct block_device_context *blkdev = gd->private_data;
1350 DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1352 if (blkdev->device_type == DVD_TYPE) {
1353 blkvsc_do_read_capacity(blkdev);
1354 set_capacity(blkdev->gd, blkdev->capacity *
1355 (blkdev->sector_size/512));
1356 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1361 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1363 sector_t total_sectors = get_capacity(bd->bd_disk);
1364 sector_t cylinder_times_heads = 0;
1367 int sectors_per_track = 0;
1372 if (total_sectors > (65535 * 16 * 255))
1373 total_sectors = (65535 * 16 * 255);
1375 if (total_sectors >= (65535 * 16 * 63)) {
1376 sectors_per_track = 255;
1379 cylinder_times_heads = total_sectors;
1380 /* sector_div stores the quotient in cylinder_times_heads */
1381 rem = sector_div(cylinder_times_heads, sectors_per_track);
1383 sectors_per_track = 17;
1385 cylinder_times_heads = total_sectors;
1386 /* sector_div stores the quotient in cylinder_times_heads */
1387 rem = sector_div(cylinder_times_heads, sectors_per_track);
1389 temp = cylinder_times_heads + 1023;
1390 /* sector_div stores the quotient in temp */
1391 rem = sector_div(temp, 1024);
1399 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1400 sectors_per_track = 31;
1403 cylinder_times_heads = total_sectors;
1405 * sector_div stores the quotient in
1406 * cylinder_times_heads
1408 rem = sector_div(cylinder_times_heads,
1412 if (cylinder_times_heads >= (heads * 1024)) {
1413 sectors_per_track = 63;
1416 cylinder_times_heads = total_sectors;
1418 * sector_div stores the quotient in
1419 * cylinder_times_heads
1421 rem = sector_div(cylinder_times_heads,
1426 temp = cylinder_times_heads;
1427 /* sector_div stores the quotient in temp */
1428 rem = sector_div(temp, heads);
1432 hg->sectors = sectors_per_track;
1433 hg->cylinders = cylinders;
1435 DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1441 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1442 unsigned cmd, unsigned long argument)
1444 /* struct block_device_context *blkdev = bd->bd_disk->private_data; */
1449 * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1450 * than just a GUID. Commented it out for now.
1453 case HDIO_GET_IDENTITY:
1454 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1455 if (copy_to_user((void __user *)arg, blkdev->device_id,
1456 blkdev->device_id_len))
1468 static int __init blkvsc_init(void)
1472 BUILD_BUG_ON(sizeof(sector_t) != 8);
1474 DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1476 ret = blkvsc_drv_init(blk_vsc_initialize);
1481 static void __exit blkvsc_exit(void)
1486 MODULE_LICENSE("GPL");
1487 MODULE_VERSION(HV_DRV_VERSION);
1488 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1489 module_init(blkvsc_init);
1490 module_exit(blkvsc_exit);