blob: e9ba752f6bf378b0480c7e311ceed54ec53ddd59 [file] [log] [blame]
aliguori6e02c382008-12-04 19:52:44 +00001/*
2 * Virtio Block Device
3 *
4 * Copyright IBM, Corp. 2007
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 */
13
Peter Maydell80c71a22016-01-18 18:01:42 +000014#include "qemu/osdep.h"
Markus Armbrusterda34e652016-03-14 09:01:28 +010015#include "qapi/error.h"
Fam Zheng827805a2014-06-11 12:11:48 +080016#include "qemu/iov.h"
Markus Armbruster0b8fa322019-05-23 16:35:07 +020017#include "qemu/module.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010018#include "qemu/error-report.h"
Sergio Lopez9b92fbc2019-09-16 13:24:12 +020019#include "qemu/main-loop.h"
Stefan Hajnoczi6d519a52010-05-22 18:15:08 +010020#include "trace.h"
Paolo Bonzini0d09e412013-02-05 17:06:20 +010021#include "hw/block/block.h"
Markus Armbrustera27bd6c2019-08-12 07:23:51 +020022#include "hw/qdev-properties.h"
Paolo Bonzini9c17d612012-12-17 18:20:04 +010023#include "sysemu/blockdev.h"
Markus Armbruster2f780b62019-08-12 07:23:58 +020024#include "sysemu/sysemu.h"
Markus Armbruster54d31232019-08-12 07:23:59 +020025#include "sysemu/runstate.h"
Paolo Bonzini0d09e412013-02-05 17:06:20 +010026#include "hw/virtio/virtio-blk.h"
Fam Zheng52b53c02014-09-10 14:17:51 +080027#include "dataplane/virtio-blk.h"
Paolo Bonzini08e2c9f2017-08-22 09:23:55 +020028#include "scsi/constants.h"
Christoph Hellwig1063b8b2009-04-27 10:29:14 +020029#ifdef __linux__
30# include <scsi/sg.h>
31#endif
Paolo Bonzini0d09e412013-02-05 17:06:20 +010032#include "hw/virtio/virtio-bus.h"
Markus Armbrusterca77ee22019-08-12 07:23:39 +020033#include "migration/qemu-file-types.h"
Rusty Russell783d1892014-06-24 19:43:44 +020034#include "hw/virtio/virtio-access.h"
Hiroki Narukawa4c41c692022-02-14 20:53:02 +090035#include "qemu/coroutine.h"
aliguori6e02c382008-12-04 19:52:44 +000036
Stefano Garzarella20764be2019-02-21 11:33:09 +010037/* Config size before the discard support (hide associated config fields) */
Changpeng Liu42824b42019-02-13 09:48:57 +080038#define VIRTIO_BLK_CFG_SIZE offsetof(struct virtio_blk_config, \
39 max_discard_sectors)
Stefano Garzarella20764be2019-02-21 11:33:09 +010040/*
41 * Starting from the discard feature, we can use this array to properly
42 * set the config size depending on the features enabled.
43 */
Philippe Mathieu-Daudé01ce7722021-05-11 12:41:56 +020044static const VirtIOFeature feature_sizes[] = {
Stefano Garzarella20764be2019-02-21 11:33:09 +010045 {.flags = 1ULL << VIRTIO_BLK_F_DISCARD,
Max Reitz5d5b33c2019-10-11 17:27:59 +020046 .end = endof(struct virtio_blk_config, discard_sector_alignment)},
Stefano Garzarella20764be2019-02-21 11:33:09 +010047 {.flags = 1ULL << VIRTIO_BLK_F_WRITE_ZEROES,
Max Reitz5d5b33c2019-10-11 17:27:59 +020048 .end = endof(struct virtio_blk_config, write_zeroes_may_unmap)},
Stefano Garzarella20764be2019-02-21 11:33:09 +010049 {}
50};
51
52static void virtio_blk_set_config_size(VirtIOBlock *s, uint64_t host_features)
53{
54 s->config_size = MAX(VIRTIO_BLK_CFG_SIZE,
55 virtio_feature_get_config_size(feature_sizes, host_features));
56
57 assert(s->config_size <= sizeof(struct virtio_blk_config));
58}
Changpeng Liu42824b42019-02-13 09:48:57 +080059
Greg Kurzd14dde52016-09-30 17:12:50 +020060static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq,
61 VirtIOBlockReq *req)
Fam Zheng671ec3f2014-06-11 12:11:43 +080062{
Fam Zheng671ec3f2014-06-11 12:11:43 +080063 req->dev = s;
Stefan Hajnocziedaffd92016-06-21 13:13:13 +010064 req->vq = vq;
Stefan Hajnoczi869d66a2014-07-09 10:05:48 +020065 req->qiov.size = 0;
Paolo Bonzini2a6cdd62015-04-02 19:50:44 +020066 req->in_len = 0;
Stefan Hajnoczi869d66a2014-07-09 10:05:48 +020067 req->next = NULL;
Peter Lieven95f71422015-02-02 14:52:21 +010068 req->mr_next = NULL;
Fam Zheng671ec3f2014-06-11 12:11:43 +080069}
70
Greg Kurzd14dde52016-09-30 17:12:50 +020071static void virtio_blk_free_request(VirtIOBlockReq *req)
Fam Zheng671ec3f2014-06-11 12:11:43 +080072{
Fam Zheng1d29b5b2017-02-07 21:27:22 +080073 g_free(req);
Fam Zheng671ec3f2014-06-11 12:11:43 +080074}
75
Paolo Bonzini03de2f52016-02-14 18:17:09 +010076static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
aliguori869a5c62009-01-22 19:52:25 +000077{
78 VirtIOBlock *s = req->dev;
KONRAD Frederic1cc91b72013-03-18 17:37:27 +010079 VirtIODevice *vdev = VIRTIO_DEVICE(s);
aliguori869a5c62009-01-22 19:52:25 +000080
Stefan Hajnoczia576cea2017-06-14 10:29:30 +010081 trace_virtio_blk_req_complete(vdev, req, status);
Stefan Hajnoczi6d519a52010-05-22 18:15:08 +010082
Aurelien Jarno92e3c2a2011-01-25 11:55:14 +010083 stb_p(&req->in->status, status);
Stefan Hajnoczi7bd04a02020-09-17 10:44:54 +010084 iov_discard_undo(&req->inhdr_undo);
85 iov_discard_undo(&req->outhdr_undo);
Stefan Hajnocziedaffd92016-06-21 13:13:13 +010086 virtqueue_push(req->vq, &req->elem, req->in_len);
Paolo Bonzinieb41cf72016-04-06 12:16:23 +020087 if (s->dataplane_started && !s->dataplane_disabled) {
Stefan Hajnocziedaffd92016-06-21 13:13:13 +010088 virtio_blk_data_plane_notify(s->dataplane, req->vq);
Paolo Bonzini03de2f52016-02-14 18:17:09 +010089 } else {
Stefan Hajnocziedaffd92016-06-21 13:13:13 +010090 virtio_notify(vdev, req->vq);
Paolo Bonzini03de2f52016-02-14 18:17:09 +010091 }
Fam Zhengbf4bd462014-06-17 14:32:06 +080092}
93
Kevin Wolff35d68f2009-11-27 13:25:39 +010094static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
Stefano Garzarella00f639f2019-02-21 11:33:05 +010095 bool is_read, bool acct_failed)
aliguori869a5c62009-01-22 19:52:25 +000096{
aliguori869a5c62009-01-22 19:52:25 +000097 VirtIOBlock *s = req->dev;
Stefano Garzarella9a6719d2019-02-08 15:23:47 +010098 BlockErrorAction action = blk_get_error_action(s->blk, is_read, error);
aliguori869a5c62009-01-22 19:52:25 +000099
Wenchao Xiaa5895692014-06-18 08:43:30 +0200100 if (action == BLOCK_ERROR_ACTION_STOP) {
Fam Zheng466138d2015-11-23 08:41:20 +0800101 /* Break the link as the next request is going to be parsed from the
102 * ring again. Otherwise we may end up doing a double completion! */
103 req->mr_next = NULL;
aliguori869a5c62009-01-22 19:52:25 +0000104 req->next = s->rq;
105 s->rq = req;
Wenchao Xiaa5895692014-06-18 08:43:30 +0200106 } else if (action == BLOCK_ERROR_ACTION_REPORT) {
aliguori869a5c62009-01-22 19:52:25 +0000107 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
Stefano Garzarella00f639f2019-02-21 11:33:05 +0100108 if (acct_failed) {
109 block_acct_failed(blk_get_stats(s->blk), &req->acct);
110 }
Fam Zheng671ec3f2014-06-11 12:11:43 +0800111 virtio_blk_free_request(req);
aliguori869a5c62009-01-22 19:52:25 +0000112 }
113
Markus Armbruster4be74632014-10-07 13:59:18 +0200114 blk_error_action(s->blk, action, is_read, error);
Wenchao Xiaa5895692014-06-18 08:43:30 +0200115 return action != BLOCK_ERROR_ACTION_IGNORE;
aliguori869a5c62009-01-22 19:52:25 +0000116}
117
aliguori6e02c382008-12-04 19:52:44 +0000118static void virtio_blk_rw_complete(void *opaque, int ret)
119{
Peter Lieven95f71422015-02-02 14:52:21 +0100120 VirtIOBlockReq *next = opaque;
Paolo Bonzinib9e413d2017-02-13 14:52:32 +0100121 VirtIOBlock *s = next->dev;
Stefan Hajnoczia576cea2017-06-14 10:29:30 +0100122 VirtIODevice *vdev = VIRTIO_DEVICE(s);
aliguori6e02c382008-12-04 19:52:44 +0000123
Paolo Bonzinib9e413d2017-02-13 14:52:32 +0100124 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
Peter Lieven95f71422015-02-02 14:52:21 +0100125 while (next) {
126 VirtIOBlockReq *req = next;
127 next = req->mr_next;
Stefan Hajnoczia576cea2017-06-14 10:29:30 +0100128 trace_virtio_blk_rw_complete(vdev, req, ret);
Stefan Hajnoczi6d519a52010-05-22 18:15:08 +0100129
Peter Lieven95f71422015-02-02 14:52:21 +0100130 if (req->qiov.nalloc != -1) {
Dongli Zhange61809e2018-11-06 12:52:32 +0800131 /* If nalloc is != -1 req->qiov is a local copy of the original
Yaowei Bai9bb192a2018-07-28 13:18:44 +0800132 * external iovec. It was allocated in submit_requests to be
133 * able to merge requests. */
Peter Lieven95f71422015-02-02 14:52:21 +0100134 qemu_iovec_destroy(&req->qiov);
135 }
136
137 if (ret) {
Anastasiia Rusakovabf4069f2019-03-07 18:19:25 +0200138 int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type);
Peter Lieven95f71422015-02-02 14:52:21 +0100139 bool is_read = !(p & VIRTIO_BLK_T_OUT);
Paolo Bonzini2a6cdd62015-04-02 19:50:44 +0200140 /* Note that memory may be dirtied on read failure. If the
141 * virtio request is not completed here, as is the case for
142 * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
143 * correctly during live migration. While this is ugly,
144 * it is acceptable because the device is free to write to
145 * the memory until the request is completed (which will
146 * happen on the other side of the migration).
147 */
Stefano Garzarella00f639f2019-02-21 11:33:05 +0100148 if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) {
Peter Lieven95f71422015-02-02 14:52:21 +0100149 continue;
150 }
151 }
152
153 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
Anastasiia Rusakovabf4069f2019-03-07 18:19:25 +0200154 block_acct_done(blk_get_stats(s->blk), &req->acct);
Peter Lieven95f71422015-02-02 14:52:21 +0100155 virtio_blk_free_request(req);
aliguori6e02c382008-12-04 19:52:44 +0000156 }
Paolo Bonzinib9e413d2017-02-13 14:52:32 +0100157 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
aliguori869a5c62009-01-22 19:52:25 +0000158}
aliguori6e02c382008-12-04 19:52:44 +0000159
Christoph Hellwigaa659be2009-09-04 19:02:23 +0200160static void virtio_blk_flush_complete(void *opaque, int ret)
161{
162 VirtIOBlockReq *req = opaque;
Paolo Bonzinib9e413d2017-02-13 14:52:32 +0100163 VirtIOBlock *s = req->dev;
Christoph Hellwigaa659be2009-09-04 19:02:23 +0200164
Paolo Bonzinib9e413d2017-02-13 14:52:32 +0100165 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
Kevin Wolf8c269b52010-10-20 13:17:30 +0200166 if (ret) {
Stefano Garzarella00f639f2019-02-21 11:33:05 +0100167 if (virtio_blk_handle_rw_error(req, -ret, 0, true)) {
Paolo Bonzinib9e413d2017-02-13 14:52:32 +0100168 goto out;
Kevin Wolf8c269b52010-10-20 13:17:30 +0200169 }
170 }
171
172 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
Stefano Garzarella9a6719d2019-02-08 15:23:47 +0100173 block_acct_done(blk_get_stats(s->blk), &req->acct);
Fam Zheng671ec3f2014-06-11 12:11:43 +0800174 virtio_blk_free_request(req);
Paolo Bonzinib9e413d2017-02-13 14:52:32 +0100175
176out:
177 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
aliguori6e02c382008-12-04 19:52:44 +0000178}
179
Stefano Garzarella37b06f82019-02-21 11:33:10 +0100180static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret)
181{
182 VirtIOBlockReq *req = opaque;
183 VirtIOBlock *s = req->dev;
184 bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) &
185 ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES;
186
187 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
188 if (ret) {
189 if (virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) {
190 goto out;
191 }
192 }
193
194 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
195 if (is_write_zeroes) {
196 block_acct_done(blk_get_stats(s->blk), &req->acct);
197 }
198 virtio_blk_free_request(req);
199
200out:
201 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
202}
203
Fam Zheng1dc936a2015-01-20 11:28:47 +0800204#ifdef __linux__
205
206typedef struct {
207 VirtIOBlockReq *req;
208 struct sg_io_hdr hdr;
209} VirtIOBlockIoctlReq;
210
211static void virtio_blk_ioctl_complete(void *opaque, int status)
212{
213 VirtIOBlockIoctlReq *ioctl_req = opaque;
214 VirtIOBlockReq *req = ioctl_req->req;
Paolo Bonzini9d456652017-02-13 14:52:30 +0100215 VirtIOBlock *s = req->dev;
216 VirtIODevice *vdev = VIRTIO_DEVICE(s);
Fam Zheng1dc936a2015-01-20 11:28:47 +0800217 struct virtio_scsi_inhdr *scsi;
218 struct sg_io_hdr *hdr;
219
220 scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
221
222 if (status) {
223 status = VIRTIO_BLK_S_UNSUPP;
224 virtio_stl_p(vdev, &scsi->errors, 255);
225 goto out;
226 }
227
228 hdr = &ioctl_req->hdr;
229 /*
230 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
231 * clear the masked_status field [hence status gets cleared too, see
232 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
233 * status has occurred. However they do set DRIVER_SENSE in driver_status
234 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
235 */
236 if (hdr->status == 0 && hdr->sb_len_wr > 0) {
237 hdr->status = CHECK_CONDITION;
238 }
239
240 virtio_stl_p(vdev, &scsi->errors,
241 hdr->status | (hdr->msg_status << 8) |
242 (hdr->host_status << 16) | (hdr->driver_status << 24));
243 virtio_stl_p(vdev, &scsi->residual, hdr->resid);
244 virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
245 virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
246
247out:
Paolo Bonzinib9e413d2017-02-13 14:52:32 +0100248 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
Fam Zheng1dc936a2015-01-20 11:28:47 +0800249 virtio_blk_req_complete(req, status);
250 virtio_blk_free_request(req);
Paolo Bonzinib9e413d2017-02-13 14:52:32 +0100251 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
Fam Zheng1dc936a2015-01-20 11:28:47 +0800252 g_free(ioctl_req);
253}
254
255#endif
256
Stefan Hajnocziedaffd92016-06-21 13:13:13 +0100257static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
aliguori6e02c382008-12-04 19:52:44 +0000258{
Stefan Hajnocziedaffd92016-06-21 13:13:13 +0100259 VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
aliguori6e02c382008-12-04 19:52:44 +0000260
Paolo Bonzini51b19eb2016-02-04 16:26:51 +0200261 if (req) {
Stefan Hajnocziedaffd92016-06-21 13:13:13 +0100262 virtio_blk_init_request(s, vq, req);
aliguori6e02c382008-12-04 19:52:44 +0000263 }
aliguori6e02c382008-12-04 19:52:44 +0000264 return req;
265}
266
Fam Zheng75344fa2015-01-20 11:28:46 +0800267static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200268{
Stefan Weil47ce9ef2012-05-22 23:23:32 +0200269 int status = VIRTIO_BLK_S_OK;
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800270 struct virtio_scsi_inhdr *scsi = NULL;
Fam Zheng75344fa2015-01-20 11:28:46 +0800271 VirtIOBlock *blk = req->dev;
Anastasiia Rusakovabf4069f2019-03-07 18:19:25 +0200272 VirtIODevice *vdev = VIRTIO_DEVICE(blk);
273 VirtQueueElement *elem = &req->elem;
Rusty Russell783d1892014-06-24 19:43:44 +0200274
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800275#ifdef __linux__
276 int i;
Fam Zheng1dc936a2015-01-20 11:28:47 +0800277 VirtIOBlockIoctlReq *ioctl_req;
Fam Zhenga209f462015-02-17 17:55:53 +0800278 BlockAIOCB *acb;
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800279#endif
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200280
281 /*
282 * We require at least one output segment each for the virtio_blk_outhdr
283 * and the SCSI command block.
284 *
285 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
286 * and the sense buffer pointer in the input segments.
287 */
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800288 if (elem->out_num < 2 || elem->in_num < 3) {
289 status = VIRTIO_BLK_S_IOERR;
290 goto fail;
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200291 }
292
293 /*
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200294 * The scsi inhdr is placed in the second-to-last input segment, just
295 * before the regular inhdr.
296 */
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800297 scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200298
Stefano Garzarellabbe8bd42019-02-21 11:33:06 +0100299 if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) {
Paolo Bonzinif34e73c2012-05-16 12:54:03 +0200300 status = VIRTIO_BLK_S_UNSUPP;
301 goto fail;
302 }
303
304 /*
305 * No support for bidirection commands yet.
306 */
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800307 if (elem->out_num > 2 && elem->in_num > 3) {
Paolo Bonzinif34e73c2012-05-16 12:54:03 +0200308 status = VIRTIO_BLK_S_UNSUPP;
309 goto fail;
310 }
311
312#ifdef __linux__
Fam Zheng1dc936a2015-01-20 11:28:47 +0800313 ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
314 ioctl_req->req = req;
315 ioctl_req->hdr.interface_id = 'S';
316 ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
317 ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
318 ioctl_req->hdr.dxfer_len = 0;
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200319
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800320 if (elem->out_num > 2) {
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200321 /*
322 * If there are more than the minimally required 2 output segments
323 * there is write payload starting from the third iovec.
324 */
Fam Zheng1dc936a2015-01-20 11:28:47 +0800325 ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
326 ioctl_req->hdr.iovec_count = elem->out_num - 2;
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200327
Fam Zheng1dc936a2015-01-20 11:28:47 +0800328 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
329 ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
330 }
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200331
Fam Zheng1dc936a2015-01-20 11:28:47 +0800332 ioctl_req->hdr.dxferp = elem->out_sg + 2;
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200333
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800334 } else if (elem->in_num > 3) {
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200335 /*
336 * If we have more than 3 input segments the guest wants to actually
337 * read data.
338 */
Fam Zheng1dc936a2015-01-20 11:28:47 +0800339 ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
340 ioctl_req->hdr.iovec_count = elem->in_num - 3;
341 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
342 ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
343 }
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200344
Fam Zheng1dc936a2015-01-20 11:28:47 +0800345 ioctl_req->hdr.dxferp = elem->in_sg;
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200346 } else {
347 /*
348 * Some SCSI commands don't actually transfer any data.
349 */
Fam Zheng1dc936a2015-01-20 11:28:47 +0800350 ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200351 }
352
Fam Zheng1dc936a2015-01-20 11:28:47 +0800353 ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
354 ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200355
Fam Zhenga209f462015-02-17 17:55:53 +0800356 acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
357 virtio_blk_ioctl_complete, ioctl_req);
358 if (!acb) {
359 g_free(ioctl_req);
360 status = VIRTIO_BLK_S_UNSUPP;
361 goto fail;
362 }
Fam Zheng1dc936a2015-01-20 11:28:47 +0800363 return -EINPROGRESS;
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200364#else
Paolo Bonzinif34e73c2012-05-16 12:54:03 +0200365 abort();
366#endif
367
368fail:
369 /* Just put anything nonzero so that the ioctl fails in the guest. */
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800370 if (scsi) {
Rusty Russell783d1892014-06-24 19:43:44 +0200371 virtio_stl_p(vdev, &scsi->errors, 255);
Fam Zheng5a05cbe2014-05-22 16:22:42 +0800372 }
373 return status;
374}
375
376static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
377{
378 int status;
379
Fam Zheng75344fa2015-01-20 11:28:46 +0800380 status = virtio_blk_handle_scsi_req(req);
Fam Zheng1dc936a2015-01-20 11:28:47 +0800381 if (status != -EINPROGRESS) {
382 virtio_blk_req_complete(req, status);
383 virtio_blk_free_request(req);
384 }
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200385}
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200386
Peter Lieven95f71422015-02-02 14:52:21 +0100387static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb,
388 int start, int num_reqs, int niov)
aliguori869a5c62009-01-22 19:52:25 +0000389{
Peter Lieven95f71422015-02-02 14:52:21 +0100390 QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
391 int64_t sector_num = mrb->reqs[start]->sector_num;
Peter Lieven95f71422015-02-02 14:52:21 +0100392 bool is_write = mrb->is_write;
Christoph Hellwig87b245d2009-08-13 16:49:56 +0200393
Peter Lieven95f71422015-02-02 14:52:21 +0100394 if (num_reqs > 1) {
395 int i;
396 struct iovec *tmp_iov = qiov->iov;
397 int tmp_niov = qiov->niov;
398
399 /* mrb->reqs[start]->qiov was initialized from external so we can't
Eric Blakeb5772fd2016-05-06 10:26:33 -0600400 * modify it here. We need to initialize it locally and then add the
Peter Lieven95f71422015-02-02 14:52:21 +0100401 * external iovecs. */
402 qemu_iovec_init(qiov, niov);
403
404 for (i = 0; i < tmp_niov; i++) {
405 qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
406 }
407
408 for (i = start + 1; i < start + num_reqs; i++) {
409 qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
410 mrb->reqs[i]->qiov.size);
411 mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
Peter Lieven95f71422015-02-02 14:52:21 +0100412 }
Peter Lieven95f71422015-02-02 14:52:21 +0100413
Stefan Hajnoczia576cea2017-06-14 10:29:30 +0100414 trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev),
415 mrb, start, num_reqs,
Eric Blakeb5772fd2016-05-06 10:26:33 -0600416 sector_num << BDRV_SECTOR_BITS,
417 qiov->size, is_write);
Peter Lieven95f71422015-02-02 14:52:21 +0100418 block_acct_merge_done(blk_get_stats(blk),
419 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
420 num_reqs - 1);
421 }
422
423 if (is_write) {
Eric Blakeb5772fd2016-05-06 10:26:33 -0600424 blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
425 virtio_blk_rw_complete, mrb->reqs[start]);
Peter Lieven95f71422015-02-02 14:52:21 +0100426 } else {
Eric Blakeb5772fd2016-05-06 10:26:33 -0600427 blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
428 virtio_blk_rw_complete, mrb->reqs[start]);
Peter Lieven95f71422015-02-02 14:52:21 +0100429 }
430}
431
432static int multireq_compare(const void *a, const void *b)
433{
434 const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
435 *req2 = *(VirtIOBlockReq **)b;
436
437 /*
438 * Note that we can't simply subtract sector_num1 from sector_num2
439 * here as that could overflow the return value.
440 */
441 if (req1->sector_num > req2->sector_num) {
442 return 1;
443 } else if (req1->sector_num < req2->sector_num) {
444 return -1;
445 } else {
446 return 0;
447 }
448}
449
Greg Kurzd14dde52016-09-30 17:12:50 +0200450static void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb)
Peter Lieven95f71422015-02-02 14:52:21 +0100451{
452 int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
Eric Blake5def6b82016-06-23 16:37:19 -0600453 uint32_t max_transfer;
Peter Lieven95f71422015-02-02 14:52:21 +0100454 int64_t sector_num = 0;
455
456 if (mrb->num_reqs == 1) {
457 submit_requests(blk, mrb, 0, 1, -1);
458 mrb->num_reqs = 0;
Christoph Hellwigc20fd872010-06-08 18:26:07 +0200459 return;
460 }
461
Eric Blake5def6b82016-06-23 16:37:19 -0600462 max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk);
Peter Lieven95f71422015-02-02 14:52:21 +0100463
464 qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
465 &multireq_compare);
466
467 for (i = 0; i < mrb->num_reqs; i++) {
468 VirtIOBlockReq *req = mrb->reqs[i];
469 if (num_reqs > 0) {
Gonglei49cffbc2015-11-11 09:59:26 +0800470 /*
471 * NOTE: We cannot merge the requests in below situations:
472 * 1. requests are not sequential
473 * 2. merge would exceed maximum number of IOVs
474 * 3. merge would exceed maximum transfer length of backend device
475 */
476 if (sector_num + nb_sectors != req->sector_num ||
Stefan Hajnoczi222565f2015-07-09 10:56:46 +0100477 niov > blk_get_max_iov(blk) - req->qiov.niov ||
Eric Blake5def6b82016-06-23 16:37:19 -0600478 req->qiov.size > max_transfer ||
479 nb_sectors > (max_transfer -
480 req->qiov.size) / BDRV_SECTOR_SIZE) {
Peter Lieven95f71422015-02-02 14:52:21 +0100481 submit_requests(blk, mrb, start, num_reqs, niov);
482 num_reqs = 0;
Kevin Wolf91553dc2009-09-09 17:53:38 +0200483 }
484 }
Peter Lieven95f71422015-02-02 14:52:21 +0100485
486 if (num_reqs == 0) {
487 sector_num = req->sector_num;
488 nb_sectors = niov = 0;
489 start = i;
490 }
491
492 nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
493 niov += req->qiov.niov;
494 num_reqs++;
Christoph Hellwig87b245d2009-08-13 16:49:56 +0200495 }
Christoph Hellwigc20fd872010-06-08 18:26:07 +0200496
Peter Lieven95f71422015-02-02 14:52:21 +0100497 submit_requests(blk, mrb, start, num_reqs, niov);
498 mrb->num_reqs = 0;
aliguorid28a1b62009-03-28 17:46:14 +0000499}
aliguori869a5c62009-01-22 19:52:25 +0000500
Christoph Hellwigc20fd872010-06-08 18:26:07 +0200501static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
Christoph Hellwigaa659be2009-09-04 19:02:23 +0200502{
Anastasiia Rusakovabf4069f2019-03-07 18:19:25 +0200503 VirtIOBlock *s = req->dev;
504
505 block_acct_start(blk_get_stats(s->blk), &req->acct, 0,
Benoît Canet5366d0c2014-09-05 15:46:18 +0200506 BLOCK_ACCT_FLUSH);
Christoph Hellwiga597e792011-08-25 08:26:01 +0200507
Christoph Hellwig618fbb82010-05-19 12:40:09 +0200508 /*
509 * Make sure all outstanding writes are posted to the backing device.
510 */
Peter Lieven95f71422015-02-02 14:52:21 +0100511 if (mrb->is_write && mrb->num_reqs > 0) {
Anastasiia Rusakovabf4069f2019-03-07 18:19:25 +0200512 virtio_blk_submit_multireq(s->blk, mrb);
Peter Lieven95f71422015-02-02 14:52:21 +0100513 }
Anastasiia Rusakovabf4069f2019-03-07 18:19:25 +0200514 blk_aio_flush(s->blk, virtio_blk_flush_complete, req);
Christoph Hellwigaa659be2009-09-04 19:02:23 +0200515}
516
Markus Armbrusterd0e14372014-07-09 19:07:29 +0200517static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
518 uint64_t sector, size_t size)
519{
Markus Armbruster3c2daac2014-07-09 19:07:31 +0200520 uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
521 uint64_t total_sectors;
522
Peter Lieven75af1f32015-02-06 11:54:11 +0100523 if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
Peter Lieven95f71422015-02-02 14:52:21 +0100524 return false;
525 }
Markus Armbrusterd0e14372014-07-09 19:07:29 +0200526 if (sector & dev->sector_mask) {
527 return false;
528 }
Markus Armbruster2a303072014-10-07 13:59:17 +0200529 if (size % dev->conf.conf.logical_block_size) {
Markus Armbrusterd0e14372014-07-09 19:07:29 +0200530 return false;
531 }
Markus Armbruster4be74632014-10-07 13:59:18 +0200532 blk_get_geometry(dev->blk, &total_sectors);
Markus Armbruster3c2daac2014-07-09 19:07:31 +0200533 if (sector > total_sectors || nb_sectors > total_sectors - sector) {
534 return false;
535 }
Markus Armbrusterd0e14372014-07-09 19:07:29 +0200536 return true;
537}
538
Stefano Garzarella37b06f82019-02-21 11:33:10 +0100539static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req,
540 struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes)
541{
542 VirtIOBlock *s = req->dev;
543 VirtIODevice *vdev = VIRTIO_DEVICE(s);
544 uint64_t sector;
545 uint32_t num_sectors, flags, max_sectors;
546 uint8_t err_status;
547 int bytes;
548
549 sector = virtio_ldq_p(vdev, &dwz_hdr->sector);
550 num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors);
551 flags = virtio_ldl_p(vdev, &dwz_hdr->flags);
552 max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors :
553 s->conf.max_discard_sectors;
554
555 /*
556 * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check
557 * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in
558 * the integer variable.
559 */
560 if (unlikely(num_sectors > max_sectors)) {
561 err_status = VIRTIO_BLK_S_IOERR;
562 goto err;
563 }
564
565 bytes = num_sectors << BDRV_SECTOR_BITS;
566
567 if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) {
568 err_status = VIRTIO_BLK_S_IOERR;
569 goto err;
570 }
571
572 /*
573 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard
574 * and write zeroes commands if any unknown flag is set.
575 */
576 if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
577 err_status = VIRTIO_BLK_S_UNSUPP;
578 goto err;
579 }
580
581 if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */
582 int blk_aio_flags = 0;
583
584 if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) {
585 blk_aio_flags |= BDRV_REQ_MAY_UNMAP;
586 }
587
588 block_acct_start(blk_get_stats(s->blk), &req->acct, bytes,
589 BLOCK_ACCT_WRITE);
590
591 blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS,
592 bytes, blk_aio_flags,
593 virtio_blk_discard_write_zeroes_complete, req);
594 } else { /* VIRTIO_BLK_T_DISCARD */
595 /*
596 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for
597 * discard commands if the unmap flag is set.
598 */
599 if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
600 err_status = VIRTIO_BLK_S_UNSUPP;
601 goto err;
602 }
603
604 blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes,
605 virtio_blk_discard_write_zeroes_complete, req);
606 }
607
608 return VIRTIO_BLK_S_OK;
609
610err:
611 if (is_write_zeroes) {
612 block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE);
613 }
614 return err_status;
615}
616
Greg Kurz20ea6862016-09-30 17:13:07 +0200617static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
Kevin Wolfbc6694d2010-01-27 13:12:34 +0100618{
Aurelien Jarno92e3c2a2011-01-25 11:55:14 +0100619 uint32_t type;
Stefan Hajnoczif897bf72014-07-09 10:05:49 +0200620 struct iovec *in_iov = req->elem.in_sg;
Dongli Zhang5636da72018-11-07 00:09:16 +0800621 struct iovec *out_iov = req->elem.out_sg;
Stefan Hajnoczif897bf72014-07-09 10:05:49 +0200622 unsigned in_num = req->elem.in_num;
623 unsigned out_num = req->elem.out_num;
Greg Kurz20ea6862016-09-30 17:13:07 +0200624 VirtIOBlock *s = req->dev;
625 VirtIODevice *vdev = VIRTIO_DEVICE(s);
Aurelien Jarno92e3c2a2011-01-25 11:55:14 +0100626
Stefan Hajnoczif897bf72014-07-09 10:05:49 +0200627 if (req->elem.out_num < 1 || req->elem.in_num < 1) {
Greg Kurz20ea6862016-09-30 17:13:07 +0200628 virtio_error(vdev, "virtio-blk missing headers");
629 return -1;
Kevin Wolfbc6694d2010-01-27 13:12:34 +0100630 }
631
Dongli Zhang5636da72018-11-07 00:09:16 +0800632 if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out,
Fam Zheng827805a2014-06-11 12:11:48 +0800633 sizeof(req->out)) != sizeof(req->out))) {
Greg Kurz20ea6862016-09-30 17:13:07 +0200634 virtio_error(vdev, "virtio-blk request outhdr too short");
635 return -1;
Fam Zheng827805a2014-06-11 12:11:48 +0800636 }
Fam Zhengee17e842014-06-11 12:11:50 +0800637
Stefan Hajnoczi7bd04a02020-09-17 10:44:54 +0100638 iov_discard_front_undoable(&out_iov, &out_num, sizeof(req->out),
639 &req->outhdr_undo);
Fam Zhengee17e842014-06-11 12:11:50 +0800640
Gonglei12048542015-06-24 17:29:24 +0800641 if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
Greg Kurz20ea6862016-09-30 17:13:07 +0200642 virtio_error(vdev, "virtio-blk request inhdr too short");
Stefan Hajnoczi7bd04a02020-09-17 10:44:54 +0100643 iov_discard_undo(&req->outhdr_undo);
Greg Kurz20ea6862016-09-30 17:13:07 +0200644 return -1;
Fam Zhengee17e842014-06-11 12:11:50 +0800645 }
646
Paolo Bonzini2a6cdd62015-04-02 19:50:44 +0200647 /* We always touch the last byte, so just see how big in_iov is. */
648 req->in_len = iov_size(in_iov, in_num);
Fam Zhengee17e842014-06-11 12:11:50 +0800649 req->in = (void *)in_iov[in_num - 1].iov_base
650 + in_iov[in_num - 1].iov_len
651 - sizeof(struct virtio_blk_inhdr);
Stefan Hajnoczi7bd04a02020-09-17 10:44:54 +0100652 iov_discard_back_undoable(in_iov, &in_num, sizeof(struct virtio_blk_inhdr),
653 &req->inhdr_undo);
Kevin Wolfbc6694d2010-01-27 13:12:34 +0100654
Stefano Garzarella9a6719d2019-02-08 15:23:47 +0100655 type = virtio_ldl_p(vdev, &req->out.type);
Aurelien Jarno92e3c2a2011-01-25 11:55:14 +0100656
Peter Lieven95f71422015-02-02 14:52:21 +0100657 /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
Stefan Weil631b22e2015-04-09 20:32:39 +0200658 * is an optional flag. Although a guest should not send this flag if
Peter Lieven95f71422015-02-02 14:52:21 +0100659 * not negotiated we ignored it in the past. So keep ignoring it. */
660 switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
661 case VIRTIO_BLK_T_IN:
662 {
663 bool is_write = type & VIRTIO_BLK_T_OUT;
Stefano Garzarella9a6719d2019-02-08 15:23:47 +0100664 req->sector_num = virtio_ldq_p(vdev, &req->out.sector);
Peter Lieven95f71422015-02-02 14:52:21 +0100665
666 if (is_write) {
Dongli Zhang5636da72018-11-07 00:09:16 +0800667 qemu_iovec_init_external(&req->qiov, out_iov, out_num);
Stefan Hajnoczia576cea2017-06-14 10:29:30 +0100668 trace_virtio_blk_handle_write(vdev, req, req->sector_num,
Peter Lieven95f71422015-02-02 14:52:21 +0100669 req->qiov.size / BDRV_SECTOR_SIZE);
670 } else {
671 qemu_iovec_init_external(&req->qiov, in_iov, in_num);
Stefan Hajnoczia576cea2017-06-14 10:29:30 +0100672 trace_virtio_blk_handle_read(vdev, req, req->sector_num,
Peter Lieven95f71422015-02-02 14:52:21 +0100673 req->qiov.size / BDRV_SECTOR_SIZE);
674 }
675
Stefano Garzarella9a6719d2019-02-08 15:23:47 +0100676 if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) {
Peter Lieven95f71422015-02-02 14:52:21 +0100677 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
Stefano Garzarella9a6719d2019-02-08 15:23:47 +0100678 block_acct_invalid(blk_get_stats(s->blk),
Alberto Garcia01762e02015-10-28 17:33:12 +0200679 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
Peter Lieven95f71422015-02-02 14:52:21 +0100680 virtio_blk_free_request(req);
Greg Kurz20ea6862016-09-30 17:13:07 +0200681 return 0;
Peter Lieven95f71422015-02-02 14:52:21 +0100682 }
683
Stefano Garzarella9a6719d2019-02-08 15:23:47 +0100684 block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size,
Peter Lieven95f71422015-02-02 14:52:21 +0100685 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
686
687 /* merge would exceed maximum number of requests or IO direction
688 * changes */
689 if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS ||
Peter Lievenc99495a2015-02-02 14:52:22 +0100690 is_write != mrb->is_write ||
Stefano Garzarella9a6719d2019-02-08 15:23:47 +0100691 !s->conf.request_merging)) {
692 virtio_blk_submit_multireq(s->blk, mrb);
Peter Lieven95f71422015-02-02 14:52:21 +0100693 }
694
695 assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS);
696 mrb->reqs[mrb->num_reqs++] = req;
697 mrb->is_write = is_write;
698 break;
699 }
700 case VIRTIO_BLK_T_FLUSH:
Christoph Hellwigc20fd872010-06-08 18:26:07 +0200701 virtio_blk_handle_flush(req, mrb);
Peter Lieven95f71422015-02-02 14:52:21 +0100702 break;
703 case VIRTIO_BLK_T_SCSI_CMD:
Kevin Wolfbc6694d2010-01-27 13:12:34 +0100704 virtio_blk_handle_scsi(req);
Peter Lieven95f71422015-02-02 14:52:21 +0100705 break;
706 case VIRTIO_BLK_T_GET_ID:
707 {
Markus Armbrustera8686a92011-06-20 11:35:18 +0200708 /*
709 * NB: per existing s/n string convention the string is
710 * terminated by '\0' only when shorter than buffer.
711 */
Markus Armbruster2a303072014-10-07 13:59:17 +0200712 const char *serial = s->conf.serial ? s->conf.serial : "";
Marc María83ceea2014-08-12 13:41:51 +0200713 size_t size = MIN(strlen(serial) + 1,
714 MIN(iov_size(in_iov, in_num),
715 VIRTIO_BLK_ID_BYTES));
716 iov_from_buf(in_iov, in_num, 0, serial, size);
john cooper2930b312010-07-02 13:44:25 -0400717 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
Fam Zheng671ec3f2014-06-11 12:11:43 +0800718 virtio_blk_free_request(req);
Peter Lieven95f71422015-02-02 14:52:21 +0100719 break;
720 }
Stefano Garzarella37b06f82019-02-21 11:33:10 +0100721 /*
722 * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with
723 * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement,
724 * so we must mask it for these requests, then we will check if it is set.
725 */
726 case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT:
727 case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT:
728 {
729 struct virtio_blk_discard_write_zeroes dwz_hdr;
730 size_t out_len = iov_size(out_iov, out_num);
731 bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) ==
732 VIRTIO_BLK_T_WRITE_ZEROES;
733 uint8_t err_status;
734
735 /*
736 * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains
737 * more than one segment.
738 */
739 if (unlikely(!(type & VIRTIO_BLK_T_OUT) ||
740 out_len > sizeof(dwz_hdr))) {
741 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
742 virtio_blk_free_request(req);
743 return 0;
744 }
745
746 if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr,
747 sizeof(dwz_hdr)) != sizeof(dwz_hdr))) {
Stefan Hajnoczi7bd04a02020-09-17 10:44:54 +0100748 iov_discard_undo(&req->inhdr_undo);
749 iov_discard_undo(&req->outhdr_undo);
Stefano Garzarella37b06f82019-02-21 11:33:10 +0100750 virtio_error(vdev, "virtio-blk discard/write_zeroes header"
751 " too short");
752 return -1;
753 }
754
755 err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr,
756 is_write_zeroes);
757 if (err_status != VIRTIO_BLK_S_OK) {
758 virtio_blk_req_complete(req, err_status);
759 virtio_blk_free_request(req);
760 }
761
762 break;
763 }
Peter Lieven95f71422015-02-02 14:52:21 +0100764 default:
Alexey Zaytsev9e72c452012-12-13 09:03:43 +0200765 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
Fam Zheng671ec3f2014-06-11 12:11:43 +0800766 virtio_blk_free_request(req);
Kevin Wolfbc6694d2010-01-27 13:12:34 +0100767 }
Greg Kurz20ea6862016-09-30 17:13:07 +0200768 return 0;
Kevin Wolfbc6694d2010-01-27 13:12:34 +0100769}
770
Stefan Hajnoczi186b9692021-12-07 13:23:33 +0000771void virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
aliguori6e02c382008-12-04 19:52:44 +0000772{
aliguori6e02c382008-12-04 19:52:44 +0000773 VirtIOBlockReq *req;
Peter Lieven95f71422015-02-02 14:52:21 +0100774 MultiReqBuffer mrb = {};
Stefan Hajnoczid0435bc2019-12-09 21:09:57 +0000775 bool suppress_notifications = virtio_queue_get_notification(vq);
aliguori6e02c382008-12-04 19:52:44 +0000776
Paolo Bonzini9d456652017-02-13 14:52:30 +0100777 aio_context_acquire(blk_get_aio_context(s->blk));
Stefan Hajnoczifc735482015-07-20 16:54:16 +0100778 blk_io_plug(s->blk);
779
Stefan Hajnoczi9ef9d402016-12-01 19:26:46 +0000780 do {
Stefan Hajnoczid0435bc2019-12-09 21:09:57 +0000781 if (suppress_notifications) {
782 virtio_queue_set_notification(vq, 0);
783 }
Stefan Hajnoczi9ef9d402016-12-01 19:26:46 +0000784
785 while ((req = virtio_blk_get_request(s, vq))) {
786 if (virtio_blk_handle_request(req, &mrb)) {
787 virtqueue_detach_element(req->vq, &req->elem, 0);
788 virtio_blk_free_request(req);
789 break;
790 }
Greg Kurz20ea6862016-09-30 17:13:07 +0200791 }
Stefan Hajnoczi9ef9d402016-12-01 19:26:46 +0000792
Stefan Hajnoczid0435bc2019-12-09 21:09:57 +0000793 if (suppress_notifications) {
794 virtio_queue_set_notification(vq, 1);
795 }
Stefan Hajnoczi9ef9d402016-12-01 19:26:46 +0000796 } while (!virtio_queue_empty(vq));
Kevin Wolf91553dc2009-09-09 17:53:38 +0200797
Peter Lieven95f71422015-02-02 14:52:21 +0100798 if (mrb.num_reqs) {
799 virtio_blk_submit_multireq(s->blk, &mrb);
800 }
Stefan Hajnoczifc735482015-07-20 16:54:16 +0100801
802 blk_io_unplug(s->blk);
Paolo Bonzini9d456652017-02-13 14:52:30 +0100803 aio_context_release(blk_get_aio_context(s->blk));
aliguori6e02c382008-12-04 19:52:44 +0000804}
805
Michael S. Tsirkin8a2fad52016-04-06 12:16:26 +0200806static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
807{
808 VirtIOBlock *s = (VirtIOBlock *)vdev;
809
Stefan Hajnoczi186b9692021-12-07 13:23:33 +0000810 if (s->dataplane && !s->dataplane_started) {
Michael S. Tsirkin8a2fad52016-04-06 12:16:26 +0200811 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
812 * dataplane here instead of waiting for .set_status().
813 */
Paolo Bonzini9ffe3372016-10-21 22:48:09 +0200814 virtio_device_start_ioeventfd(vdev);
Michael S. Tsirkin8a2fad52016-04-06 12:16:26 +0200815 if (!s->dataplane_disabled) {
816 return;
817 }
818 }
Stefan Hajnoczi186b9692021-12-07 13:23:33 +0000819 virtio_blk_handle_vq(s, vq);
Michael S. Tsirkin8a2fad52016-04-06 12:16:26 +0200820}
821
Sergio Lopez49b44542020-06-03 11:32:40 +0200822void virtio_blk_process_queued_requests(VirtIOBlock *s, bool is_bh)
aliguori869a5c62009-01-22 19:52:25 +0000823{
aliguori869a5c62009-01-22 19:52:25 +0000824 VirtIOBlockReq *req = s->rq;
Peter Lieven95f71422015-02-02 14:52:21 +0100825 MultiReqBuffer mrb = {};
aliguori869a5c62009-01-22 19:52:25 +0000826
aliguori869a5c62009-01-22 19:52:25 +0000827 s->rq = NULL;
828
Paolo Bonzini19196312017-02-13 14:52:31 +0100829 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
aliguori869a5c62009-01-22 19:52:25 +0000830 while (req) {
zhanghailiang1bdb1762014-08-18 15:42:50 +0800831 VirtIOBlockReq *next = req->next;
Greg Kurz20ea6862016-09-30 17:13:07 +0200832 if (virtio_blk_handle_request(req, &mrb)) {
833 /* Device is now broken and won't do any processing until it gets
834 * reset. Already queued requests will be lost: let's purge them.
835 */
836 while (req) {
837 next = req->next;
838 virtqueue_detach_element(req->vq, &req->elem, 0);
839 virtio_blk_free_request(req);
840 req = next;
841 }
842 break;
843 }
zhanghailiang1bdb1762014-08-18 15:42:50 +0800844 req = next;
aliguori869a5c62009-01-22 19:52:25 +0000845 }
Kevin Wolff1b52862010-01-27 13:12:35 +0100846
Peter Lieven95f71422015-02-02 14:52:21 +0100847 if (mrb.num_reqs) {
848 virtio_blk_submit_multireq(s->blk, &mrb);
849 }
Sergio Lopez49b44542020-06-03 11:32:40 +0200850 if (is_bh) {
851 blk_dec_in_flight(s->conf.conf.blk);
852 }
Paolo Bonzini19196312017-02-13 14:52:31 +0100853 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
aliguori869a5c62009-01-22 19:52:25 +0000854}
855
Sergio Lopez7aa1c242020-06-03 11:32:39 +0200856static void virtio_blk_dma_restart_bh(void *opaque)
857{
858 VirtIOBlock *s = opaque;
859
860 qemu_bh_delete(s->bh);
861 s->bh = NULL;
862
Sergio Lopez49b44542020-06-03 11:32:40 +0200863 virtio_blk_process_queued_requests(s, true);
Sergio Lopez7aa1c242020-06-03 11:32:39 +0200864}
865
Philippe Mathieu-Daudé538f0492021-01-11 16:20:20 +0100866static void virtio_blk_dma_restart_cb(void *opaque, bool running,
Luiz Capitulino1dfb4dd2011-07-29 14:26:33 -0300867 RunState state)
Markus Armbruster213189a2009-07-28 14:33:41 -0400868{
869 VirtIOBlock *s = opaque;
Sergio Lopez49b44542020-06-03 11:32:40 +0200870 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(s)));
871 VirtioBusState *bus = VIRTIO_BUS(qbus);
Markus Armbruster213189a2009-07-28 14:33:41 -0400872
Stefan Hajnoczi392808b2012-11-14 15:45:38 +0100873 if (!running) {
Markus Armbruster213189a2009-07-28 14:33:41 -0400874 return;
Stefan Hajnoczi392808b2012-11-14 15:45:38 +0100875 }
Markus Armbruster213189a2009-07-28 14:33:41 -0400876
Sergio Lopez49b44542020-06-03 11:32:40 +0200877 /*
878 * If ioeventfd is enabled, don't schedule the BH here as queued
879 * requests will be processed while starting the data plane.
880 */
881 if (!s->bh && !virtio_bus_ioeventfd_enabled(bus)) {
Markus Armbruster4be74632014-10-07 13:59:18 +0200882 s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk),
Fam Zheng4407c1c2014-06-17 14:32:08 +0800883 virtio_blk_dma_restart_bh, s);
Kevin Wolf680f2002019-02-14 18:51:03 +0100884 blk_inc_in_flight(s->conf.conf.blk);
Markus Armbruster213189a2009-07-28 14:33:41 -0400885 qemu_bh_schedule(s->bh);
886 }
887}
888
aliguori6e02c382008-12-04 19:52:44 +0000889static void virtio_blk_reset(VirtIODevice *vdev)
890{
KONRAD Frederic1cc91b72013-03-18 17:37:27 +0100891 VirtIOBlock *s = VIRTIO_BLK(vdev);
Alexander Yarygin6e40b3b2015-06-17 13:37:20 +0300892 AioContext *ctx;
Fam Zheng26307f62016-08-04 10:44:13 +0800893 VirtIOBlockReq *req;
Stefan Hajnoczi392808b2012-11-14 15:45:38 +0100894
Alexander Yarygin6e40b3b2015-06-17 13:37:20 +0300895 ctx = blk_get_aio_context(s->blk);
896 aio_context_acquire(ctx);
897 blk_drain(s->blk);
898
Fam Zheng26307f62016-08-04 10:44:13 +0800899 /* We drop queued requests after blk_drain() because blk_drain() itself can
900 * produce them. */
901 while (s->rq) {
902 req = s->rq;
903 s->rq = req->next;
Stefan Hajnoczi97b93c82016-09-19 14:28:04 +0100904 virtqueue_detach_element(req->vq, &req->elem, 0);
Fam Zheng26307f62016-08-04 10:44:13 +0800905 virtio_blk_free_request(req);
906 }
907
Alexander Yarygin6e40b3b2015-06-17 13:37:20 +0300908 aio_context_release(ctx);
909
Paolo Bonzini9ffe3372016-10-21 22:48:09 +0200910 assert(!s->dataplane_started);
Markus Armbruster4be74632014-10-07 13:59:18 +0200911 blk_set_enable_write_cache(s->blk, s->original_wce);
aliguori6e02c382008-12-04 19:52:44 +0000912}
913
john cooperbf011292009-06-22 14:26:51 -0400914/* coalesce internal state, copy to pci i/o region 0
915 */
aliguori6e02c382008-12-04 19:52:44 +0000916static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
917{
KONRAD Frederic1cc91b72013-03-18 17:37:27 +0100918 VirtIOBlock *s = VIRTIO_BLK(vdev);
Markus Armbruster2a303072014-10-07 13:59:17 +0200919 BlockConf *conf = &s->conf.conf;
aliguori6e02c382008-12-04 19:52:44 +0000920 struct virtio_blk_config blkcfg;
921 uint64_t capacity;
Stefan Hajnoczi17d0bc02017-08-08 13:22:51 +0100922 int64_t length;
Markus Armbrusterf7516732014-10-07 13:59:16 +0200923 int blk_size = conf->logical_block_size;
aliguori6e02c382008-12-04 19:52:44 +0000924
Markus Armbruster4be74632014-10-07 13:59:18 +0200925 blk_get_geometry(s->blk, &capacity);
Gerd Hoffmann5c5dafd2009-06-12 09:50:18 +0200926 memset(&blkcfg, 0, sizeof(blkcfg));
Rusty Russell783d1892014-06-24 19:43:44 +0200927 virtio_stq_p(vdev, &blkcfg.capacity, capacity);
Denis Plotnikov1bf8a982019-12-20 17:09:04 +0300928 virtio_stl_p(vdev, &blkcfg.seg_max,
929 s->conf.seg_max_adjust ? s->conf.queue_size - 2 : 128 - 2);
Michael S. Tsirkin907eb3e2015-02-16 22:36:03 +0100930 virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls);
Rusty Russell783d1892014-06-24 19:43:44 +0200931 virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
Markus Armbrusterf7516732014-10-07 13:59:16 +0200932 virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size);
Roman Kagan6abee262020-05-29 01:55:09 +0300933 virtio_stl_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size);
Michael S. Tsirkin907eb3e2015-02-16 22:36:03 +0100934 blkcfg.geometry.heads = conf->heads;
Christian Borntraeger136be992012-05-24 13:22:55 +0200935 /*
936 * We must ensure that the block device capacity is a multiple of
Peter Maydelle03ba132013-04-09 12:48:19 +0100937 * the logical block size. If that is not the case, let's use
Christian Borntraeger136be992012-05-24 13:22:55 +0200938 * sector_mask to adopt the geometry to have a correct picture.
939 * For those devices where the capacity is ok for the given geometry
Peter Maydelle03ba132013-04-09 12:48:19 +0100940 * we don't touch the sector value of the geometry, since some devices
Christian Borntraeger136be992012-05-24 13:22:55 +0200941 * (like s390 dasd) need a specific value. Here the capacity is already
942 * cyls*heads*secs*blk_size and the sector value is not block size
943 * divided by 512 - instead it is the amount of blk_size blocks
944 * per track (cylinder).
945 */
Stefan Hajnoczi17d0bc02017-08-08 13:22:51 +0100946 length = blk_getlength(s->blk);
947 if (length > 0 && length / conf->heads / conf->secs % blk_size) {
Michael S. Tsirkin907eb3e2015-02-16 22:36:03 +0100948 blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
Christian Borntraeger136be992012-05-24 13:22:55 +0200949 } else {
Michael S. Tsirkin907eb3e2015-02-16 22:36:03 +0100950 blkcfg.geometry.sectors = conf->secs;
Christian Borntraeger136be992012-05-24 13:22:55 +0200951 }
Blue Swirlc7085da2009-06-13 13:20:25 +0000952 blkcfg.size_max = 0;
Markus Armbrusterf7516732014-10-07 13:59:16 +0200953 blkcfg.physical_block_exp = get_physical_block_exp(conf);
Christoph Hellwig9752c372010-02-10 23:37:25 +0100954 blkcfg.alignment_offset = 0;
Markus Armbruster4be74632014-10-07 13:59:18 +0200955 blkcfg.wce = blk_enable_write_cache(s->blk);
Stefan Hajnoczi2f270592016-06-21 13:13:16 +0100956 virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
Stefano Garzarella37b06f82019-02-21 11:33:10 +0100957 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) {
Akihiko Odakifb0b1542021-02-25 09:12:39 +0900958 uint32_t discard_granularity = conf->discard_granularity;
959 if (discard_granularity == -1 || !s->conf.report_discard_granularity) {
960 discard_granularity = blk_size;
961 }
Stefano Garzarella37b06f82019-02-21 11:33:10 +0100962 virtio_stl_p(vdev, &blkcfg.max_discard_sectors,
963 s->conf.max_discard_sectors);
964 virtio_stl_p(vdev, &blkcfg.discard_sector_alignment,
Akihiko Odakifb0b1542021-02-25 09:12:39 +0900965 discard_granularity >> BDRV_SECTOR_BITS);
Stefano Garzarella37b06f82019-02-21 11:33:10 +0100966 /*
967 * We support only one segment per request since multiple segments
968 * are not widely used and there are no userspace APIs that allow
969 * applications to submit multiple segments in a single call.
970 */
971 virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1);
972 }
973 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) {
974 virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors,
975 s->conf.max_write_zeroes_sectors);
976 blkcfg.write_zeroes_may_unmap = 1;
977 virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1);
978 }
Stefano Garzarella20764be2019-02-21 11:33:09 +0100979 memcpy(config, &blkcfg, s->config_size);
aliguori6e02c382008-12-04 19:52:44 +0000980}
981
Paolo Bonzini13e3dce2012-08-09 16:07:19 +0200982static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
983{
KONRAD Frederic1cc91b72013-03-18 17:37:27 +0100984 VirtIOBlock *s = VIRTIO_BLK(vdev);
Paolo Bonzini13e3dce2012-08-09 16:07:19 +0200985 struct virtio_blk_config blkcfg;
986
Stefano Garzarella20764be2019-02-21 11:33:09 +0100987 memcpy(&blkcfg, config, s->config_size);
Fam Zheng6d7e73d2014-05-15 19:22:06 +0800988
Markus Armbruster4be74632014-10-07 13:59:18 +0200989 aio_context_acquire(blk_get_aio_context(s->blk));
990 blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
991 aio_context_release(blk_get_aio_context(s->blk));
Paolo Bonzini13e3dce2012-08-09 16:07:19 +0200992}
993
Jason Wang9d5b7312015-07-27 17:49:19 +0800994static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
995 Error **errp)
aliguori6e02c382008-12-04 19:52:44 +0000996{
KONRAD Frederic1cc91b72013-03-18 17:37:27 +0100997 VirtIOBlock *s = VIRTIO_BLK(vdev);
Christoph Hellwig1063b8b2009-04-27 10:29:14 +0200998
Stefano Garzarellabbe8bd42019-02-21 11:33:06 +0100999 /* Firstly sync all virtio-blk possible supported features */
1000 features |= s->host_features;
1001
Cornelia Huck0cd09c32014-12-11 14:25:05 +01001002 virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
1003 virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
1004 virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
1005 virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
Cornelia Huck95129d62015-08-17 11:48:29 +02001006 if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
Stefano Garzarellabbe8bd42019-02-21 11:33:06 +01001007 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) {
Jason Wangefb82062015-07-27 17:49:20 +08001008 error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
1009 return 0;
1010 }
Jason Wangefb82062015-07-27 17:49:20 +08001011 } else {
Jason Wangc9b11f92015-07-27 17:49:21 +08001012 virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
Jason Wangefb82062015-07-27 17:49:20 +08001013 virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
1014 }
Christoph Hellwigaa659be2009-09-04 19:02:23 +02001015
Evgeny Yakovlev5f258572019-11-05 21:22:17 +03001016 if (blk_enable_write_cache(s->blk) ||
1017 (s->conf.x_enable_wce_if_config_wce &&
1018 virtio_has_feature(features, VIRTIO_BLK_F_CONFIG_WCE))) {
Cornelia Huck0cd09c32014-12-11 14:25:05 +01001019 virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
Markus Armbruster4be74632014-10-07 13:59:18 +02001020 }
Kevin Wolf86b1cf32021-01-18 13:34:47 +01001021 if (!blk_is_writable(s->blk)) {
Cornelia Huck0cd09c32014-12-11 14:25:05 +01001022 virtio_add_feature(&features, VIRTIO_BLK_F_RO);
Markus Armbruster4be74632014-10-07 13:59:18 +02001023 }
Stefan Hajnoczi2f270592016-06-21 13:13:16 +01001024 if (s->conf.num_queues > 1) {
1025 virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
1026 }
Christoph Hellwig1063b8b2009-04-27 10:29:14 +02001027
1028 return features;
aliguori6e02c382008-12-04 19:52:44 +00001029}
1030
Paolo Bonzini9315cbf2012-08-09 16:07:20 +02001031static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
1032{
KONRAD Frederic1cc91b72013-03-18 17:37:27 +01001033 VirtIOBlock *s = VIRTIO_BLK(vdev);
Paolo Bonzini9315cbf2012-08-09 16:07:20 +02001034
Paolo Bonzini9ffe3372016-10-21 22:48:09 +02001035 if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) {
1036 assert(!s->dataplane_started);
Stefan Hajnoczi392808b2012-11-14 15:45:38 +01001037 }
Stefan Hajnoczi392808b2012-11-14 15:45:38 +01001038
Paolo Bonzini9315cbf2012-08-09 16:07:20 +02001039 if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1040 return;
1041 }
1042
Paolo Bonzinief5bc962013-09-20 17:31:55 +02001043 /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
1044 * cache flushes. Thus, the "auto writethrough" behavior is never
1045 * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
1046 * Leaving it enabled would break the following sequence:
1047 *
1048 * Guest started with "-drive cache=writethrough"
1049 * Guest sets status to 0
1050 * Guest sets DRIVER bit in status field
1051 * Guest reads host features (WCE=0, CONFIG_WCE=1)
1052 * Guest writes guest features (WCE=0, CONFIG_WCE=1)
1053 * Guest writes 1 to the WCE configuration field (writeback mode)
1054 * Guest sets DRIVER_OK bit in status field
1055 *
Markus Armbruster4be74632014-10-07 13:59:18 +02001056 * s->blk would erroneously be placed in writethrough mode.
Paolo Bonzinief5bc962013-09-20 17:31:55 +02001057 */
Cornelia Huck95129d62015-08-17 11:48:29 +02001058 if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
Markus Armbruster4be74632014-10-07 13:59:18 +02001059 aio_context_acquire(blk_get_aio_context(s->blk));
1060 blk_set_enable_write_cache(s->blk,
Cornelia Huck95129d62015-08-17 11:48:29 +02001061 virtio_vdev_has_feature(vdev,
1062 VIRTIO_BLK_F_WCE));
Markus Armbruster4be74632014-10-07 13:59:18 +02001063 aio_context_release(blk_get_aio_context(s->blk));
Paolo Bonzinief5bc962013-09-20 17:31:55 +02001064 }
Paolo Bonzini9315cbf2012-08-09 16:07:20 +02001065}
1066
Greg Kurzb2b295a2014-06-24 19:19:23 +02001067static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
1068{
1069 VirtIOBlock *s = VIRTIO_BLK(vdev);
1070 VirtIOBlockReq *req = s->rq;
1071
aliguori869a5c62009-01-22 19:52:25 +00001072 while (req) {
1073 qemu_put_sbyte(f, 1);
Stefan Hajnoczi30d8bf62016-06-21 13:13:14 +01001074
1075 if (s->conf.num_queues > 1) {
1076 qemu_put_be32(f, virtio_get_queue_index(req->vq));
1077 }
1078
Jason Wang86044b22019-10-25 10:35:24 +02001079 qemu_put_virtqueue_element(vdev, f, &req->elem);
aliguori869a5c62009-01-22 19:52:25 +00001080 req = req->next;
1081 }
1082 qemu_put_sbyte(f, 0);
aliguori6e02c382008-12-04 19:52:44 +00001083}
1084
Greg Kurzb2b295a2014-06-24 19:19:23 +02001085static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
1086 int version_id)
1087{
1088 VirtIOBlock *s = VIRTIO_BLK(vdev);
Orit Wassermann2a633c42012-05-16 12:21:35 +02001089
aliguori869a5c62009-01-22 19:52:25 +00001090 while (qemu_get_sbyte(f)) {
Stefan Hajnoczi30d8bf62016-06-21 13:13:14 +01001091 unsigned nvqs = s->conf.num_queues;
1092 unsigned vq_idx = 0;
Paolo Bonziniab281c12016-01-31 11:28:59 +01001093 VirtIOBlockReq *req;
Stefan Hajnoczi30d8bf62016-06-21 13:13:14 +01001094
1095 if (nvqs > 1) {
1096 vq_idx = qemu_get_be32(f);
1097
1098 if (vq_idx >= nvqs) {
1099 error_report("Invalid virtqueue index in request list: %#x",
1100 vq_idx);
1101 return -EINVAL;
1102 }
1103 }
1104
Jason Wang8607f5c2016-12-30 18:09:10 +08001105 req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq));
Stefan Hajnoczi30d8bf62016-06-21 13:13:14 +01001106 virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
aliguori869a5c62009-01-22 19:52:25 +00001107 req->next = s->rq;
Yoshiaki Tamura20a81e42010-06-21 17:50:01 +09001108 s->rq = req;
aliguori869a5c62009-01-22 19:52:25 +00001109 }
aliguori6e02c382008-12-04 19:52:44 +00001110
1111 return 0;
1112}
1113
Sergio Lopez9b92fbc2019-09-16 13:24:12 +02001114static void virtio_resize_cb(void *opaque)
1115{
1116 VirtIODevice *vdev = opaque;
1117
1118 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1119 virtio_notify_config(vdev);
1120}
1121
Markus Armbruster145feb12011-08-03 15:07:42 +02001122static void virtio_blk_resize(void *opaque)
Christoph Hellwige5051fc2011-01-24 13:32:51 +01001123{
KONRAD Frederic1cc91b72013-03-18 17:37:27 +01001124 VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
Christoph Hellwige5051fc2011-01-24 13:32:51 +01001125
Sergio Lopez9b92fbc2019-09-16 13:24:12 +02001126 /*
1127 * virtio_notify_config() needs to acquire the global mutex,
1128 * so it can't be called from an iothread. Instead, schedule
1129 * it to be run in the main context BH.
1130 */
1131 aio_bh_schedule_oneshot(qemu_get_aio_context(), virtio_resize_cb, vdev);
Christoph Hellwige5051fc2011-01-24 13:32:51 +01001132}
1133
Markus Armbruster0e49de52011-08-03 15:07:41 +02001134static const BlockDevOps virtio_block_ops = {
Markus Armbruster145feb12011-08-03 15:07:42 +02001135 .resize_cb = virtio_blk_resize,
Markus Armbruster0e49de52011-08-03 15:07:41 +02001136};
1137
Andreas Färber75884af2013-07-30 01:35:08 +02001138static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001139{
Andreas Färber75884af2013-07-30 01:35:08 +02001140 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
Andreas Färber179b4172013-07-30 04:48:15 +02001141 VirtIOBlock *s = VIRTIO_BLK(dev);
Markus Armbruster2a303072014-10-07 13:59:17 +02001142 VirtIOBlkConf *conf = &s->conf;
Andreas Färber3ffeeef2013-06-07 16:18:50 +02001143 Error *err = NULL;
Stefan Hajnoczi2f270592016-06-21 13:13:16 +01001144 unsigned i;
aliguori6e02c382008-12-04 19:52:44 +00001145
Markus Armbruster4be74632014-10-07 13:59:18 +02001146 if (!conf->conf.blk) {
Andreas Färber75884af2013-07-30 01:35:08 +02001147 error_setg(errp, "drive property not set");
1148 return;
Markus Armbrusterd75d25e2010-07-06 14:37:43 +02001149 }
Markus Armbruster4be74632014-10-07 13:59:18 +02001150 if (!blk_is_inserted(conf->conf.blk)) {
Andreas Färber75884af2013-07-30 01:35:08 +02001151 error_setg(errp, "Device needs media, but drive is empty");
1152 return;
Markus Armbruster98f28ad2010-07-06 14:37:44 +02001153 }
Stefan Hajnoczi9445e1e2020-08-18 15:33:47 +01001154 if (conf->num_queues == VIRTIO_BLK_AUTO_NUM_QUEUES) {
1155 conf->num_queues = 1;
1156 }
Stefan Hajnoczi2f270592016-06-21 13:13:16 +01001157 if (!conf->num_queues) {
1158 error_setg(errp, "num-queues property must be larger than 0");
1159 return;
1160 }
Denis Plotnikov1bf8a982019-12-20 17:09:04 +03001161 if (conf->queue_size <= 2) {
1162 error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1163 "must be > 2", conf->queue_size);
1164 return;
1165 }
Mark Kanda6040aed2017-12-11 09:16:24 -06001166 if (!is_power_of_2(conf->queue_size) ||
1167 conf->queue_size > VIRTQUEUE_MAX_SIZE) {
1168 error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1169 "must be a power of 2 (max %d)",
1170 conf->queue_size, VIRTQUEUE_MAX_SIZE);
1171 return;
1172 }
Markus Armbrusterd75d25e2010-07-06 14:37:43 +02001173
Mao Zhongyiceff3e12017-11-22 11:08:45 +08001174 if (!blkconf_apply_backend_options(&conf->conf,
Kevin Wolf86b1cf32021-01-18 13:34:47 +01001175 !blk_supports_write_perm(conf->conf.blk),
1176 true, errp)) {
Kevin Wolfa17c17a2017-01-24 13:43:31 +01001177 return;
1178 }
Markus Armbruster4be74632014-10-07 13:59:18 +02001179 s->original_wce = blk_enable_write_cache(conf->conf.blk);
Mao Zhongyiceff3e12017-11-22 11:08:45 +08001180 if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) {
Andreas Färber75884af2013-07-30 01:35:08 +02001181 return;
Markus Armbrusterb7eb0c92012-07-11 15:08:39 +02001182 }
Mao Zhongyiceff3e12017-11-22 11:08:45 +08001183
Roman Kaganc56ee922020-05-29 01:55:10 +03001184 if (!blkconf_blocksizes(&conf->conf, errp)) {
Mark Kanda0a75b602017-12-11 09:16:25 -06001185 return;
1186 }
1187
Stefano Garzarella37b06f82019-02-21 11:33:10 +01001188 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) &&
1189 (!conf->max_discard_sectors ||
1190 conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1191 error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")"
1192 ", must be between 1 and %d",
1193 conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS);
1194 return;
1195 }
1196
1197 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) &&
1198 (!conf->max_write_zeroes_sectors ||
1199 conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1200 error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32
1201 "), must be between 1 and %d",
1202 conf->max_write_zeroes_sectors,
1203 (int)BDRV_REQUEST_MAX_SECTORS);
1204 return;
1205 }
1206
Stefano Garzarella20764be2019-02-21 11:33:09 +01001207 virtio_blk_set_config_size(s, s->host_features);
1208
Jonah Palmer3857cd52022-04-01 09:23:18 -04001209 virtio_init(vdev, VIRTIO_ID_BLOCK, s->config_size);
aliguori6e02c382008-12-04 19:52:44 +00001210
Markus Armbruster4be74632014-10-07 13:59:18 +02001211 s->blk = conf->conf.blk;
aliguori869a5c62009-01-22 19:52:25 +00001212 s->rq = NULL;
Markus Armbruster2a303072014-10-07 13:59:17 +02001213 s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
Markus Armbrustere63e7fd2012-07-10 11:12:43 +02001214
Stefan Hajnoczi2f270592016-06-21 13:13:16 +01001215 for (i = 0; i < conf->num_queues; i++) {
Mark Kanda6040aed2017-12-11 09:16:24 -06001216 virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output);
Stefan Hajnoczi2f270592016-06-21 13:13:16 +01001217 }
Kevin Wolf98e3ab32022-05-10 17:10:19 +02001218 qemu_coroutine_inc_pool_size(conf->num_queues * conf->queue_size / 2);
Markus Armbruster2a303072014-10-07 13:59:17 +02001219 virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err);
Andreas Färber3ffeeef2013-06-07 16:18:50 +02001220 if (err != NULL) {
Andreas Färber75884af2013-07-30 01:35:08 +02001221 error_propagate(errp, err);
Pan Nengyuancfaf7572020-03-28 08:57:04 +08001222 for (i = 0; i < conf->num_queues; i++) {
1223 virtio_del_queue(vdev, i);
1224 }
KONRAD Frederic6a1a8cc2013-04-24 10:21:22 +02001225 virtio_cleanup(vdev);
Andreas Färber75884af2013-07-30 01:35:08 +02001226 return;
Stefan Hajnoczi392808b2012-11-14 15:45:38 +01001227 }
aliguori6e02c382008-12-04 19:52:44 +00001228
Christian Borntraeger69b302b2013-02-22 14:37:10 +01001229 s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
Markus Armbruster4be74632014-10-07 13:59:18 +02001230 blk_set_dev_ops(s->blk, &virtio_block_ops, s);
Paul Brook07e3af92009-05-14 22:35:08 +01001231
Markus Armbruster4be74632014-10-07 13:59:18 +02001232 blk_iostatus_enable(s->blk);
Sam Eiderman71f571a2019-10-16 19:41:42 +03001233
1234 add_boot_device_lchs(dev, "/disk@0,0",
1235 conf->conf.lcyls,
1236 conf->conf.lheads,
1237 conf->conf.lsecs);
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001238}
1239
Markus Armbrusterb69c3c22020-05-05 17:29:24 +02001240static void virtio_blk_device_unrealize(DeviceState *dev)
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001241{
Andreas Färber306ec6c2013-07-30 03:50:44 +02001242 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1243 VirtIOBlock *s = VIRTIO_BLK(dev);
Eugenio Pérez4a0117c2019-10-25 10:35:22 +02001244 VirtIOBlkConf *conf = &s->conf;
1245 unsigned i;
Andreas Färber306ec6c2013-07-30 03:50:44 +02001246
Julia Suvorova7bfde682019-10-18 16:28:56 +02001247 blk_drain(s->blk);
Sam Eiderman71f571a2019-10-16 19:41:42 +03001248 del_boot_device_lchs(dev, "/disk@0,0");
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001249 virtio_blk_data_plane_destroy(s->dataplane);
1250 s->dataplane = NULL;
Eugenio Pérez4a0117c2019-10-25 10:35:22 +02001251 for (i = 0; i < conf->num_queues; i++) {
1252 virtio_del_queue(vdev, i);
1253 }
Kevin Wolf98e3ab32022-05-10 17:10:19 +02001254 qemu_coroutine_dec_pool_size(conf->num_queues * conf->queue_size / 2);
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001255 qemu_del_vm_change_state_handler(s->change);
Markus Armbruster4be74632014-10-07 13:59:18 +02001256 blockdev_mark_auto_del(s->blk);
KONRAD Frederic6a1a8cc2013-04-24 10:21:22 +02001257 virtio_cleanup(vdev);
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001258}
1259
Stefan Hajnoczi467b3f32014-06-10 09:03:20 +02001260static void virtio_blk_instance_init(Object *obj)
1261{
1262 VirtIOBlock *s = VIRTIO_BLK(obj);
1263
Markus Armbruster2a303072014-10-07 13:59:17 +02001264 device_add_bootindex_property(obj, &s->conf.conf.bootindex,
Gonglei3342ec32014-10-07 16:00:30 +08001265 "bootindex", "/disk@0,0",
Markus Armbruster40c22812020-05-05 17:29:23 +02001266 DEVICE(obj));
Stefan Hajnoczi467b3f32014-06-10 09:03:20 +02001267}
1268
Halil Pasic977a1172016-10-06 14:55:40 +02001269static const VMStateDescription vmstate_virtio_blk = {
1270 .name = "virtio-blk",
1271 .minimum_version_id = 2,
1272 .version_id = 2,
1273 .fields = (VMStateField[]) {
1274 VMSTATE_VIRTIO_DEVICE,
1275 VMSTATE_END_OF_LIST()
1276 },
1277};
Dr. David Alan Gilbertbbded322016-07-14 18:22:47 +01001278
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001279static Property virtio_blk_properties[] = {
Markus Armbruster2a303072014-10-07 13:59:17 +02001280 DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
Kevin Wolf8c398252016-06-29 17:41:35 +02001281 DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf),
Markus Armbruster2a303072014-10-07 13:59:17 +02001282 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
1283 DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
Stefano Garzarellabbe8bd42019-02-21 11:33:06 +01001284 DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features,
1285 VIRTIO_BLK_F_CONFIG_WCE, true),
Stefan Hajnoczi32a877e2014-06-18 17:58:36 +08001286#ifdef __linux__
Stefano Garzarellabbe8bd42019-02-21 11:33:06 +01001287 DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features,
1288 VIRTIO_BLK_F_SCSI, false),
Stefan Hajnoczi32a877e2014-06-18 17:58:36 +08001289#endif
Peter Lievenc99495a2015-02-02 14:52:22 +01001290 DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
1291 true),
Stefan Hajnoczi9445e1e2020-08-18 15:33:47 +01001292 DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues,
1293 VIRTIO_BLK_AUTO_NUM_QUEUES),
Denis Plotnikovc9b7d9e2020-02-14 10:46:48 +03001294 DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 256),
Denis Plotnikov1bf8a982019-12-20 17:09:04 +03001295 DEFINE_PROP_BOOL("seg-max-adjust", VirtIOBlock, conf.seg_max_adjust, true),
Fam Zhengd679ac02017-07-14 10:14:55 +08001296 DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD,
1297 IOThread *),
Stefano Garzarella5c811612019-02-21 11:33:07 +01001298 DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features,
1299 VIRTIO_BLK_F_DISCARD, true),
Akihiko Odakifb0b1542021-02-25 09:12:39 +09001300 DEFINE_PROP_BOOL("report-discard-granularity", VirtIOBlock,
1301 conf.report_discard_granularity, true),
Stefano Garzarella5c811612019-02-21 11:33:07 +01001302 DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features,
1303 VIRTIO_BLK_F_WRITE_ZEROES, true),
Stefano Garzarella37b06f82019-02-21 11:33:10 +01001304 DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock,
1305 conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS),
1306 DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock,
1307 conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS),
Evgeny Yakovlev5f258572019-11-05 21:22:17 +03001308 DEFINE_PROP_BOOL("x-enable-wce-if-config-wce", VirtIOBlock,
1309 conf.x_enable_wce_if_config_wce, true),
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001310 DEFINE_PROP_END_OF_LIST(),
1311};
1312
1313static void virtio_blk_class_init(ObjectClass *klass, void *data)
1314{
1315 DeviceClass *dc = DEVICE_CLASS(klass);
1316 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
Andreas Färber75884af2013-07-30 01:35:08 +02001317
Marc-André Lureau4f67d302020-01-10 19:30:32 +04001318 device_class_set_props(dc, virtio_blk_properties);
Dr. David Alan Gilbertbbded322016-07-14 18:22:47 +01001319 dc->vmsd = &vmstate_virtio_blk;
Marcel Apfelbaum125ee0e2013-07-29 17:17:45 +03001320 set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
Andreas Färber75884af2013-07-30 01:35:08 +02001321 vdc->realize = virtio_blk_device_realize;
Andreas Färber306ec6c2013-07-30 03:50:44 +02001322 vdc->unrealize = virtio_blk_device_unrealize;
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001323 vdc->get_config = virtio_blk_update_config;
1324 vdc->set_config = virtio_blk_set_config;
1325 vdc->get_features = virtio_blk_get_features;
1326 vdc->set_status = virtio_blk_set_status;
1327 vdc->reset = virtio_blk_reset;
Greg Kurzb2b295a2014-06-24 19:19:23 +02001328 vdc->save = virtio_blk_save_device;
1329 vdc->load = virtio_blk_load_device;
Paolo Bonzini9ffe3372016-10-21 22:48:09 +02001330 vdc->start_ioeventfd = virtio_blk_data_plane_start;
1331 vdc->stop_ioeventfd = virtio_blk_data_plane_stop;
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001332}
1333
Changlong Xieb5c7cea2016-08-03 16:49:07 +08001334static const TypeInfo virtio_blk_info = {
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001335 .name = TYPE_VIRTIO_BLK,
1336 .parent = TYPE_VIRTIO_DEVICE,
1337 .instance_size = sizeof(VirtIOBlock),
Stefan Hajnoczi467b3f32014-06-10 09:03:20 +02001338 .instance_init = virtio_blk_instance_init,
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001339 .class_init = virtio_blk_class_init,
1340};
1341
1342static void virtio_register_types(void)
1343{
Changlong Xieb5c7cea2016-08-03 16:49:07 +08001344 type_register_static(&virtio_blk_info);
KONRAD Frederic1c028dd2013-03-18 17:37:22 +01001345}
1346
1347type_init(virtio_register_types)