16e790746SPaolo Bonzini /* 26e790746SPaolo Bonzini * Virtio Block Device 36e790746SPaolo Bonzini * 46e790746SPaolo Bonzini * Copyright IBM, Corp. 2007 56e790746SPaolo Bonzini * 66e790746SPaolo Bonzini * Authors: 76e790746SPaolo Bonzini * Anthony Liguori <aliguori@us.ibm.com> 86e790746SPaolo Bonzini * 96e790746SPaolo Bonzini * This work is licensed under the terms of the GNU GPL, version 2. See 106e790746SPaolo Bonzini * the COPYING file in the top-level directory. 116e790746SPaolo Bonzini * 126e790746SPaolo Bonzini */ 136e790746SPaolo Bonzini 1480c71a24SPeter Maydell #include "qemu/osdep.h" 15433fcea4SStefan Hajnoczi #include "qemu/defer-call.h" 16da34e65cSMarkus Armbruster #include "qapi/error.h" 17827805a2SFam Zheng #include "qemu/iov.h" 180b8fa32fSMarkus Armbruster #include "qemu/module.h" 196e790746SPaolo Bonzini #include "qemu/error-report.h" 209b92fbcfSSergio Lopez #include "qemu/main-loop.h" 214f736650SSam Li #include "block/block_int.h" 226e790746SPaolo Bonzini #include "trace.h" 236e790746SPaolo Bonzini #include "hw/block/block.h" 24a27bd6c7SMarkus Armbruster #include "hw/qdev-properties.h" 256e790746SPaolo Bonzini #include "sysemu/blockdev.h" 26baf42268SStefan Hajnoczi #include "sysemu/block-ram-registrar.h" 272f780b6aSMarkus Armbruster #include "sysemu/sysemu.h" 2854d31236SMarkus Armbruster #include "sysemu/runstate.h" 296e790746SPaolo Bonzini #include "hw/virtio/virtio-blk.h" 3008e2c9f1SPaolo Bonzini #include "scsi/constants.h" 316e790746SPaolo Bonzini #ifdef __linux__ 326e790746SPaolo Bonzini # include <scsi/sg.h> 336e790746SPaolo Bonzini #endif 346e790746SPaolo Bonzini #include "hw/virtio/virtio-bus.h" 35ca77ee28SMarkus Armbruster #include "migration/qemu-file-types.h" 36783d1897SRusty Russell #include "hw/virtio/virtio-access.h" 37d9cf55a8SDaniil Tatianin #include "hw/virtio/virtio-blk-common.h" 384c41c69eSHiroki Narukawa #include "qemu/coroutine.h" 396e790746SPaolo Bonzini 40d14dde5eSGreg Kurz static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq, 41edaffd9fSStefan Hajnoczi VirtIOBlockReq *req) 42671ec3f0SFam Zheng { 43671ec3f0SFam Zheng req->dev = s; 44edaffd9fSStefan Hajnoczi req->vq = vq; 45869d66afSStefan Hajnoczi req->qiov.size = 0; 462a6cdd6dSPaolo Bonzini req->in_len = 0; 47869d66afSStefan Hajnoczi req->next = NULL; 4895f7142aSPeter Lieven req->mr_next = NULL; 49671ec3f0SFam Zheng } 50671ec3f0SFam Zheng 51d14dde5eSGreg Kurz static void virtio_blk_free_request(VirtIOBlockReq *req) 52671ec3f0SFam Zheng { 53c84b3192SPaolo Bonzini g_free(req); 54671ec3f0SFam Zheng } 55671ec3f0SFam Zheng 5603de2f52SPaolo Bonzini static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status) 576e790746SPaolo Bonzini { 586e790746SPaolo Bonzini VirtIOBlock *s = req->dev; 596e790746SPaolo Bonzini VirtIODevice *vdev = VIRTIO_DEVICE(s); 606e790746SPaolo Bonzini 61a576ceacSStefan Hajnoczi trace_virtio_blk_req_complete(vdev, req, status); 626e790746SPaolo Bonzini 636e790746SPaolo Bonzini stb_p(&req->in->status, status); 647bd04a04SStefan Hajnoczi iov_discard_undo(&req->inhdr_undo); 657bd04a04SStefan Hajnoczi iov_discard_undo(&req->outhdr_undo); 66edaffd9fSStefan Hajnoczi virtqueue_push(req->vq, &req->elem, req->in_len); 673cdaf3ddSStefan Hajnoczi if (s->ioeventfd_started && !s->ioeventfd_disabled) { 683bcc17f0SStefan Hajnoczi virtio_notify_irqfd(vdev, req->vq); 6903de2f52SPaolo Bonzini } else { 70edaffd9fSStefan Hajnoczi virtio_notify(vdev, req->vq); 716e790746SPaolo Bonzini } 72bf4bd461SFam Zheng } 73bf4bd461SFam Zheng 746e790746SPaolo Bonzini static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error, 7500f639fbSStefano Garzarella bool is_read, bool acct_failed) 766e790746SPaolo Bonzini { 776e790746SPaolo Bonzini VirtIOBlock *s = req->dev; 789a6719d5SStefano Garzarella BlockErrorAction action = blk_get_error_action(s->blk, is_read, error); 796e790746SPaolo Bonzini 80a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_STOP) { 81466138dcSFam Zheng /* Break the link as the next request is going to be parsed from the 82466138dcSFam Zheng * ring again. Otherwise we may end up doing a double completion! */ 83466138dcSFam Zheng req->mr_next = NULL; 849c67f33fSStefan Hajnoczi 859c67f33fSStefan Hajnoczi WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 866e790746SPaolo Bonzini req->next = s->rq; 876e790746SPaolo Bonzini s->rq = req; 889c67f33fSStefan Hajnoczi } 89a589569fSWenchao Xia } else if (action == BLOCK_ERROR_ACTION_REPORT) { 906e790746SPaolo Bonzini virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); 9100f639fbSStefano Garzarella if (acct_failed) { 9201762e03SAlberto Garcia block_acct_failed(blk_get_stats(s->blk), &req->acct); 9300f639fbSStefano Garzarella } 94671ec3f0SFam Zheng virtio_blk_free_request(req); 956e790746SPaolo Bonzini } 966e790746SPaolo Bonzini 974be74634SMarkus Armbruster blk_error_action(s->blk, action, is_read, error); 98a589569fSWenchao Xia return action != BLOCK_ERROR_ACTION_IGNORE; 996e790746SPaolo Bonzini } 1006e790746SPaolo Bonzini 1016e790746SPaolo Bonzini static void virtio_blk_rw_complete(void *opaque, int ret) 1026e790746SPaolo Bonzini { 10395f7142aSPeter Lieven VirtIOBlockReq *next = opaque; 104b9e413ddSPaolo Bonzini VirtIOBlock *s = next->dev; 105a576ceacSStefan Hajnoczi VirtIODevice *vdev = VIRTIO_DEVICE(s); 1066e790746SPaolo Bonzini 10795f7142aSPeter Lieven while (next) { 10895f7142aSPeter Lieven VirtIOBlockReq *req = next; 10995f7142aSPeter Lieven next = req->mr_next; 110a576ceacSStefan Hajnoczi trace_virtio_blk_rw_complete(vdev, req, ret); 1116e790746SPaolo Bonzini 11295f7142aSPeter Lieven if (req->qiov.nalloc != -1) { 113e61809edSDongli Zhang /* If nalloc is != -1 req->qiov is a local copy of the original 1149bb192a4SYaowei Bai * external iovec. It was allocated in submit_requests to be 1159bb192a4SYaowei Bai * able to merge requests. */ 11695f7142aSPeter Lieven qemu_iovec_destroy(&req->qiov); 11795f7142aSPeter Lieven } 11895f7142aSPeter Lieven 1196e790746SPaolo Bonzini if (ret) { 120bf4069fbSAnastasiia Rusakova int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type); 121783d1897SRusty Russell bool is_read = !(p & VIRTIO_BLK_T_OUT); 1222a6cdd6dSPaolo Bonzini /* Note that memory may be dirtied on read failure. If the 1232a6cdd6dSPaolo Bonzini * virtio request is not completed here, as is the case for 1242a6cdd6dSPaolo Bonzini * BLOCK_ERROR_ACTION_STOP, the memory may not be copied 1252a6cdd6dSPaolo Bonzini * correctly during live migration. While this is ugly, 1262a6cdd6dSPaolo Bonzini * it is acceptable because the device is free to write to 1272a6cdd6dSPaolo Bonzini * the memory until the request is completed (which will 1282a6cdd6dSPaolo Bonzini * happen on the other side of the migration). 1292a6cdd6dSPaolo Bonzini */ 13000f639fbSStefano Garzarella if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) { 13195f7142aSPeter Lieven continue; 13295f7142aSPeter Lieven } 1336e790746SPaolo Bonzini } 1346e790746SPaolo Bonzini 1356e790746SPaolo Bonzini virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 136bf4069fbSAnastasiia Rusakova block_acct_done(blk_get_stats(s->blk), &req->acct); 137671ec3f0SFam Zheng virtio_blk_free_request(req); 1386e790746SPaolo Bonzini } 13995f7142aSPeter Lieven } 1406e790746SPaolo Bonzini 1416e790746SPaolo Bonzini static void virtio_blk_flush_complete(void *opaque, int ret) 1426e790746SPaolo Bonzini { 1436e790746SPaolo Bonzini VirtIOBlockReq *req = opaque; 144b9e413ddSPaolo Bonzini VirtIOBlock *s = req->dev; 1456e790746SPaolo Bonzini 146c1135913SStefan Hajnoczi if (ret && virtio_blk_handle_rw_error(req, -ret, 0, true)) { 147c1135913SStefan Hajnoczi return; 1486e790746SPaolo Bonzini } 1496e790746SPaolo Bonzini 1506e790746SPaolo Bonzini virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 1519a6719d5SStefano Garzarella block_acct_done(blk_get_stats(s->blk), &req->acct); 152671ec3f0SFam Zheng virtio_blk_free_request(req); 1536e790746SPaolo Bonzini } 1546e790746SPaolo Bonzini 15537b06f8dSStefano Garzarella static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret) 15637b06f8dSStefano Garzarella { 15737b06f8dSStefano Garzarella VirtIOBlockReq *req = opaque; 15837b06f8dSStefano Garzarella VirtIOBlock *s = req->dev; 15937b06f8dSStefano Garzarella bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) & 16037b06f8dSStefano Garzarella ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES; 16137b06f8dSStefano Garzarella 162c1135913SStefan Hajnoczi if (ret && virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) { 163c1135913SStefan Hajnoczi return; 16437b06f8dSStefano Garzarella } 16537b06f8dSStefano Garzarella 16637b06f8dSStefano Garzarella virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 16737b06f8dSStefano Garzarella if (is_write_zeroes) { 16837b06f8dSStefano Garzarella block_acct_done(blk_get_stats(s->blk), &req->acct); 16937b06f8dSStefano Garzarella } 17037b06f8dSStefano Garzarella virtio_blk_free_request(req); 17137b06f8dSStefano Garzarella } 17237b06f8dSStefano Garzarella 1731dc936aaSFam Zheng #ifdef __linux__ 1741dc936aaSFam Zheng 1751dc936aaSFam Zheng typedef struct { 1761dc936aaSFam Zheng VirtIOBlockReq *req; 1771dc936aaSFam Zheng struct sg_io_hdr hdr; 1781dc936aaSFam Zheng } VirtIOBlockIoctlReq; 1791dc936aaSFam Zheng 1801dc936aaSFam Zheng static void virtio_blk_ioctl_complete(void *opaque, int status) 1811dc936aaSFam Zheng { 1821dc936aaSFam Zheng VirtIOBlockIoctlReq *ioctl_req = opaque; 1831dc936aaSFam Zheng VirtIOBlockReq *req = ioctl_req->req; 1849d456654SPaolo Bonzini VirtIOBlock *s = req->dev; 1859d456654SPaolo Bonzini VirtIODevice *vdev = VIRTIO_DEVICE(s); 1861dc936aaSFam Zheng struct virtio_scsi_inhdr *scsi; 1871dc936aaSFam Zheng struct sg_io_hdr *hdr; 1881dc936aaSFam Zheng 1891dc936aaSFam Zheng scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base; 1901dc936aaSFam Zheng 1911dc936aaSFam Zheng if (status) { 1921dc936aaSFam Zheng status = VIRTIO_BLK_S_UNSUPP; 1931dc936aaSFam Zheng virtio_stl_p(vdev, &scsi->errors, 255); 1941dc936aaSFam Zheng goto out; 1951dc936aaSFam Zheng } 1961dc936aaSFam Zheng 1971dc936aaSFam Zheng hdr = &ioctl_req->hdr; 1981dc936aaSFam Zheng /* 1991dc936aaSFam Zheng * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi) 2001dc936aaSFam Zheng * clear the masked_status field [hence status gets cleared too, see 2011dc936aaSFam Zheng * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED 2021dc936aaSFam Zheng * status has occurred. However they do set DRIVER_SENSE in driver_status 2031dc936aaSFam Zheng * field. Also a (sb_len_wr > 0) indicates there is a sense buffer. 2041dc936aaSFam Zheng */ 2051dc936aaSFam Zheng if (hdr->status == 0 && hdr->sb_len_wr > 0) { 2061dc936aaSFam Zheng hdr->status = CHECK_CONDITION; 2071dc936aaSFam Zheng } 2081dc936aaSFam Zheng 2091dc936aaSFam Zheng virtio_stl_p(vdev, &scsi->errors, 2101dc936aaSFam Zheng hdr->status | (hdr->msg_status << 8) | 2111dc936aaSFam Zheng (hdr->host_status << 16) | (hdr->driver_status << 24)); 2121dc936aaSFam Zheng virtio_stl_p(vdev, &scsi->residual, hdr->resid); 2131dc936aaSFam Zheng virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr); 2141dc936aaSFam Zheng virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len); 2151dc936aaSFam Zheng 2161dc936aaSFam Zheng out: 2171dc936aaSFam Zheng virtio_blk_req_complete(req, status); 2181dc936aaSFam Zheng virtio_blk_free_request(req); 2191dc936aaSFam Zheng g_free(ioctl_req); 2201dc936aaSFam Zheng } 2211dc936aaSFam Zheng 2221dc936aaSFam Zheng #endif 2231dc936aaSFam Zheng 224edaffd9fSStefan Hajnoczi static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq) 2256e790746SPaolo Bonzini { 226edaffd9fSStefan Hajnoczi VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq)); 2276e790746SPaolo Bonzini 22851b19ebeSPaolo Bonzini if (req) { 229edaffd9fSStefan Hajnoczi virtio_blk_init_request(s, vq, req); 2306e790746SPaolo Bonzini } 2316e790746SPaolo Bonzini return req; 2326e790746SPaolo Bonzini } 2336e790746SPaolo Bonzini 23475344fa4SFam Zheng static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req) 2356e790746SPaolo Bonzini { 2366e790746SPaolo Bonzini int status = VIRTIO_BLK_S_OK; 2375a05cbeeSFam Zheng struct virtio_scsi_inhdr *scsi = NULL; 23875344fa4SFam Zheng VirtIOBlock *blk = req->dev; 239bf4069fbSAnastasiia Rusakova VirtIODevice *vdev = VIRTIO_DEVICE(blk); 240bf4069fbSAnastasiia Rusakova VirtQueueElement *elem = &req->elem; 241783d1897SRusty Russell 2425a05cbeeSFam Zheng #ifdef __linux__ 2435a05cbeeSFam Zheng int i; 2441dc936aaSFam Zheng VirtIOBlockIoctlReq *ioctl_req; 245a209f461SFam Zheng BlockAIOCB *acb; 2465a05cbeeSFam Zheng #endif 2476e790746SPaolo Bonzini 2486e790746SPaolo Bonzini /* 2496e790746SPaolo Bonzini * We require at least one output segment each for the virtio_blk_outhdr 2506e790746SPaolo Bonzini * and the SCSI command block. 2516e790746SPaolo Bonzini * 2526e790746SPaolo Bonzini * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr 2536e790746SPaolo Bonzini * and the sense buffer pointer in the input segments. 2546e790746SPaolo Bonzini */ 2555a05cbeeSFam Zheng if (elem->out_num < 2 || elem->in_num < 3) { 2565a05cbeeSFam Zheng status = VIRTIO_BLK_S_IOERR; 2575a05cbeeSFam Zheng goto fail; 2586e790746SPaolo Bonzini } 2596e790746SPaolo Bonzini 2606e790746SPaolo Bonzini /* 2616e790746SPaolo Bonzini * The scsi inhdr is placed in the second-to-last input segment, just 2626e790746SPaolo Bonzini * before the regular inhdr. 2636e790746SPaolo Bonzini */ 2645a05cbeeSFam Zheng scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base; 2656e790746SPaolo Bonzini 266bbe8bd4dSStefano Garzarella if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) { 2676e790746SPaolo Bonzini status = VIRTIO_BLK_S_UNSUPP; 2686e790746SPaolo Bonzini goto fail; 2696e790746SPaolo Bonzini } 2706e790746SPaolo Bonzini 2716e790746SPaolo Bonzini /* 2726e790746SPaolo Bonzini * No support for bidirection commands yet. 2736e790746SPaolo Bonzini */ 2745a05cbeeSFam Zheng if (elem->out_num > 2 && elem->in_num > 3) { 2756e790746SPaolo Bonzini status = VIRTIO_BLK_S_UNSUPP; 2766e790746SPaolo Bonzini goto fail; 2776e790746SPaolo Bonzini } 2786e790746SPaolo Bonzini 2796e790746SPaolo Bonzini #ifdef __linux__ 2801dc936aaSFam Zheng ioctl_req = g_new0(VirtIOBlockIoctlReq, 1); 2811dc936aaSFam Zheng ioctl_req->req = req; 2821dc936aaSFam Zheng ioctl_req->hdr.interface_id = 'S'; 2831dc936aaSFam Zheng ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len; 2841dc936aaSFam Zheng ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base; 2851dc936aaSFam Zheng ioctl_req->hdr.dxfer_len = 0; 2866e790746SPaolo Bonzini 2875a05cbeeSFam Zheng if (elem->out_num > 2) { 2886e790746SPaolo Bonzini /* 2896e790746SPaolo Bonzini * If there are more than the minimally required 2 output segments 2906e790746SPaolo Bonzini * there is write payload starting from the third iovec. 2916e790746SPaolo Bonzini */ 2921dc936aaSFam Zheng ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV; 2931dc936aaSFam Zheng ioctl_req->hdr.iovec_count = elem->out_num - 2; 2946e790746SPaolo Bonzini 2951dc936aaSFam Zheng for (i = 0; i < ioctl_req->hdr.iovec_count; i++) { 2961dc936aaSFam Zheng ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len; 2971dc936aaSFam Zheng } 2986e790746SPaolo Bonzini 2991dc936aaSFam Zheng ioctl_req->hdr.dxferp = elem->out_sg + 2; 3006e790746SPaolo Bonzini 3015a05cbeeSFam Zheng } else if (elem->in_num > 3) { 3026e790746SPaolo Bonzini /* 3036e790746SPaolo Bonzini * If we have more than 3 input segments the guest wants to actually 3046e790746SPaolo Bonzini * read data. 3056e790746SPaolo Bonzini */ 3061dc936aaSFam Zheng ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV; 3071dc936aaSFam Zheng ioctl_req->hdr.iovec_count = elem->in_num - 3; 3081dc936aaSFam Zheng for (i = 0; i < ioctl_req->hdr.iovec_count; i++) { 3091dc936aaSFam Zheng ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len; 3101dc936aaSFam Zheng } 3116e790746SPaolo Bonzini 3121dc936aaSFam Zheng ioctl_req->hdr.dxferp = elem->in_sg; 3136e790746SPaolo Bonzini } else { 3146e790746SPaolo Bonzini /* 3156e790746SPaolo Bonzini * Some SCSI commands don't actually transfer any data. 3166e790746SPaolo Bonzini */ 3171dc936aaSFam Zheng ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE; 3186e790746SPaolo Bonzini } 3196e790746SPaolo Bonzini 3201dc936aaSFam Zheng ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base; 3211dc936aaSFam Zheng ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len; 3226e790746SPaolo Bonzini 323a209f461SFam Zheng acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr, 3241dc936aaSFam Zheng virtio_blk_ioctl_complete, ioctl_req); 325a209f461SFam Zheng if (!acb) { 326a209f461SFam Zheng g_free(ioctl_req); 327a209f461SFam Zheng status = VIRTIO_BLK_S_UNSUPP; 328a209f461SFam Zheng goto fail; 329a209f461SFam Zheng } 3301dc936aaSFam Zheng return -EINPROGRESS; 3316e790746SPaolo Bonzini #else 3326e790746SPaolo Bonzini abort(); 3336e790746SPaolo Bonzini #endif 3346e790746SPaolo Bonzini 3356e790746SPaolo Bonzini fail: 3366e790746SPaolo Bonzini /* Just put anything nonzero so that the ioctl fails in the guest. */ 3375a05cbeeSFam Zheng if (scsi) { 338783d1897SRusty Russell virtio_stl_p(vdev, &scsi->errors, 255); 3395a05cbeeSFam Zheng } 3405a05cbeeSFam Zheng return status; 3415a05cbeeSFam Zheng } 3425a05cbeeSFam Zheng 3435a05cbeeSFam Zheng static void virtio_blk_handle_scsi(VirtIOBlockReq *req) 3445a05cbeeSFam Zheng { 3455a05cbeeSFam Zheng int status; 3465a05cbeeSFam Zheng 34775344fa4SFam Zheng status = virtio_blk_handle_scsi_req(req); 3481dc936aaSFam Zheng if (status != -EINPROGRESS) { 3496e790746SPaolo Bonzini virtio_blk_req_complete(req, status); 350671ec3f0SFam Zheng virtio_blk_free_request(req); 3516e790746SPaolo Bonzini } 3521dc936aaSFam Zheng } 3536e790746SPaolo Bonzini 354baf42268SStefan Hajnoczi static inline void submit_requests(VirtIOBlock *s, MultiReqBuffer *mrb, 35595f7142aSPeter Lieven int start, int num_reqs, int niov) 3566e790746SPaolo Bonzini { 357baf42268SStefan Hajnoczi BlockBackend *blk = s->blk; 35895f7142aSPeter Lieven QEMUIOVector *qiov = &mrb->reqs[start]->qiov; 35995f7142aSPeter Lieven int64_t sector_num = mrb->reqs[start]->sector_num; 36095f7142aSPeter Lieven bool is_write = mrb->is_write; 361baf42268SStefan Hajnoczi BdrvRequestFlags flags = 0; 3626e790746SPaolo Bonzini 36395f7142aSPeter Lieven if (num_reqs > 1) { 36495f7142aSPeter Lieven int i; 36595f7142aSPeter Lieven struct iovec *tmp_iov = qiov->iov; 36695f7142aSPeter Lieven int tmp_niov = qiov->niov; 36795f7142aSPeter Lieven 36895f7142aSPeter Lieven /* mrb->reqs[start]->qiov was initialized from external so we can't 369b5772fddSEric Blake * modify it here. We need to initialize it locally and then add the 37095f7142aSPeter Lieven * external iovecs. */ 37195f7142aSPeter Lieven qemu_iovec_init(qiov, niov); 37295f7142aSPeter Lieven 37395f7142aSPeter Lieven for (i = 0; i < tmp_niov; i++) { 37495f7142aSPeter Lieven qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len); 37595f7142aSPeter Lieven } 37695f7142aSPeter Lieven 37795f7142aSPeter Lieven for (i = start + 1; i < start + num_reqs; i++) { 37895f7142aSPeter Lieven qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0, 37995f7142aSPeter Lieven mrb->reqs[i]->qiov.size); 38095f7142aSPeter Lieven mrb->reqs[i - 1]->mr_next = mrb->reqs[i]; 38195f7142aSPeter Lieven } 38295f7142aSPeter Lieven 383a576ceacSStefan Hajnoczi trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev), 384a576ceacSStefan Hajnoczi mrb, start, num_reqs, 385b5772fddSEric Blake sector_num << BDRV_SECTOR_BITS, 386b5772fddSEric Blake qiov->size, is_write); 38795f7142aSPeter Lieven block_acct_merge_done(blk_get_stats(blk), 38895f7142aSPeter Lieven is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ, 38995f7142aSPeter Lieven num_reqs - 1); 39095f7142aSPeter Lieven } 39195f7142aSPeter Lieven 392baf42268SStefan Hajnoczi if (blk_ram_registrar_ok(&s->blk_ram_registrar)) { 393baf42268SStefan Hajnoczi flags |= BDRV_REQ_REGISTERED_BUF; 394baf42268SStefan Hajnoczi } 395baf42268SStefan Hajnoczi 39695f7142aSPeter Lieven if (is_write) { 397baf42268SStefan Hajnoczi blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 398baf42268SStefan Hajnoczi flags, virtio_blk_rw_complete, 399baf42268SStefan Hajnoczi mrb->reqs[start]); 40095f7142aSPeter Lieven } else { 401baf42268SStefan Hajnoczi blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 402baf42268SStefan Hajnoczi flags, virtio_blk_rw_complete, 403baf42268SStefan Hajnoczi mrb->reqs[start]); 40495f7142aSPeter Lieven } 40595f7142aSPeter Lieven } 40695f7142aSPeter Lieven 40795f7142aSPeter Lieven static int multireq_compare(const void *a, const void *b) 40895f7142aSPeter Lieven { 40995f7142aSPeter Lieven const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a, 41095f7142aSPeter Lieven *req2 = *(VirtIOBlockReq **)b; 41195f7142aSPeter Lieven 41295f7142aSPeter Lieven /* 41395f7142aSPeter Lieven * Note that we can't simply subtract sector_num1 from sector_num2 41495f7142aSPeter Lieven * here as that could overflow the return value. 41595f7142aSPeter Lieven */ 41695f7142aSPeter Lieven if (req1->sector_num > req2->sector_num) { 41795f7142aSPeter Lieven return 1; 41895f7142aSPeter Lieven } else if (req1->sector_num < req2->sector_num) { 41995f7142aSPeter Lieven return -1; 42095f7142aSPeter Lieven } else { 42195f7142aSPeter Lieven return 0; 42295f7142aSPeter Lieven } 42395f7142aSPeter Lieven } 42495f7142aSPeter Lieven 425baf42268SStefan Hajnoczi static void virtio_blk_submit_multireq(VirtIOBlock *s, MultiReqBuffer *mrb) 42695f7142aSPeter Lieven { 42795f7142aSPeter Lieven int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0; 4285def6b80SEric Blake uint32_t max_transfer; 42995f7142aSPeter Lieven int64_t sector_num = 0; 43095f7142aSPeter Lieven 43195f7142aSPeter Lieven if (mrb->num_reqs == 1) { 432baf42268SStefan Hajnoczi submit_requests(s, mrb, 0, 1, -1); 43395f7142aSPeter Lieven mrb->num_reqs = 0; 4346e790746SPaolo Bonzini return; 4356e790746SPaolo Bonzini } 4366e790746SPaolo Bonzini 4375def6b80SEric Blake max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk); 43895f7142aSPeter Lieven 43995f7142aSPeter Lieven qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs), 44095f7142aSPeter Lieven &multireq_compare); 44195f7142aSPeter Lieven 44295f7142aSPeter Lieven for (i = 0; i < mrb->num_reqs; i++) { 44395f7142aSPeter Lieven VirtIOBlockReq *req = mrb->reqs[i]; 44495f7142aSPeter Lieven if (num_reqs > 0) { 44549cffbc6SGonglei /* 44649cffbc6SGonglei * NOTE: We cannot merge the requests in below situations: 44749cffbc6SGonglei * 1. requests are not sequential 44849cffbc6SGonglei * 2. merge would exceed maximum number of IOVs 44949cffbc6SGonglei * 3. merge would exceed maximum transfer length of backend device 45049cffbc6SGonglei */ 45149cffbc6SGonglei if (sector_num + nb_sectors != req->sector_num || 452baf42268SStefan Hajnoczi niov > blk_get_max_iov(s->blk) - req->qiov.niov || 4535def6b80SEric Blake req->qiov.size > max_transfer || 4545def6b80SEric Blake nb_sectors > (max_transfer - 4555def6b80SEric Blake req->qiov.size) / BDRV_SECTOR_SIZE) { 456baf42268SStefan Hajnoczi submit_requests(s, mrb, start, num_reqs, niov); 45795f7142aSPeter Lieven num_reqs = 0; 4586e790746SPaolo Bonzini } 4596e790746SPaolo Bonzini } 4606e790746SPaolo Bonzini 46195f7142aSPeter Lieven if (num_reqs == 0) { 46295f7142aSPeter Lieven sector_num = req->sector_num; 46395f7142aSPeter Lieven nb_sectors = niov = 0; 46495f7142aSPeter Lieven start = i; 46595f7142aSPeter Lieven } 46695f7142aSPeter Lieven 46795f7142aSPeter Lieven nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE; 46895f7142aSPeter Lieven niov += req->qiov.niov; 46995f7142aSPeter Lieven num_reqs++; 47095f7142aSPeter Lieven } 47195f7142aSPeter Lieven 472baf42268SStefan Hajnoczi submit_requests(s, mrb, start, num_reqs, niov); 47395f7142aSPeter Lieven mrb->num_reqs = 0; 4746e790746SPaolo Bonzini } 4756e790746SPaolo Bonzini 4766e790746SPaolo Bonzini static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb) 4776e790746SPaolo Bonzini { 478bf4069fbSAnastasiia Rusakova VirtIOBlock *s = req->dev; 479bf4069fbSAnastasiia Rusakova 480bf4069fbSAnastasiia Rusakova block_acct_start(blk_get_stats(s->blk), &req->acct, 0, 4815366d0c8SBenoît Canet BLOCK_ACCT_FLUSH); 4826e790746SPaolo Bonzini 4836e790746SPaolo Bonzini /* 4846e790746SPaolo Bonzini * Make sure all outstanding writes are posted to the backing device. 4856e790746SPaolo Bonzini */ 48695f7142aSPeter Lieven if (mrb->is_write && mrb->num_reqs > 0) { 487baf42268SStefan Hajnoczi virtio_blk_submit_multireq(s, mrb); 48895f7142aSPeter Lieven } 489bf4069fbSAnastasiia Rusakova blk_aio_flush(s->blk, virtio_blk_flush_complete, req); 4906e790746SPaolo Bonzini } 4916e790746SPaolo Bonzini 492d0e14376SMarkus Armbruster static bool virtio_blk_sect_range_ok(VirtIOBlock *dev, 493d0e14376SMarkus Armbruster uint64_t sector, size_t size) 494d0e14376SMarkus Armbruster { 4953c2daac0SMarkus Armbruster uint64_t nb_sectors = size >> BDRV_SECTOR_BITS; 4963c2daac0SMarkus Armbruster uint64_t total_sectors; 4973c2daac0SMarkus Armbruster 49875af1f34SPeter Lieven if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) { 49995f7142aSPeter Lieven return false; 50095f7142aSPeter Lieven } 501d0e14376SMarkus Armbruster if (sector & dev->sector_mask) { 502d0e14376SMarkus Armbruster return false; 503d0e14376SMarkus Armbruster } 5042a30307fSMarkus Armbruster if (size % dev->conf.conf.logical_block_size) { 505d0e14376SMarkus Armbruster return false; 506d0e14376SMarkus Armbruster } 5074be74634SMarkus Armbruster blk_get_geometry(dev->blk, &total_sectors); 5083c2daac0SMarkus Armbruster if (sector > total_sectors || nb_sectors > total_sectors - sector) { 5093c2daac0SMarkus Armbruster return false; 5103c2daac0SMarkus Armbruster } 511d0e14376SMarkus Armbruster return true; 512d0e14376SMarkus Armbruster } 513d0e14376SMarkus Armbruster 51437b06f8dSStefano Garzarella static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req, 51537b06f8dSStefano Garzarella struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes) 51637b06f8dSStefano Garzarella { 51737b06f8dSStefano Garzarella VirtIOBlock *s = req->dev; 51837b06f8dSStefano Garzarella VirtIODevice *vdev = VIRTIO_DEVICE(s); 51937b06f8dSStefano Garzarella uint64_t sector; 52037b06f8dSStefano Garzarella uint32_t num_sectors, flags, max_sectors; 52137b06f8dSStefano Garzarella uint8_t err_status; 52237b06f8dSStefano Garzarella int bytes; 52337b06f8dSStefano Garzarella 52437b06f8dSStefano Garzarella sector = virtio_ldq_p(vdev, &dwz_hdr->sector); 52537b06f8dSStefano Garzarella num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors); 52637b06f8dSStefano Garzarella flags = virtio_ldl_p(vdev, &dwz_hdr->flags); 52737b06f8dSStefano Garzarella max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors : 52837b06f8dSStefano Garzarella s->conf.max_discard_sectors; 52937b06f8dSStefano Garzarella 53037b06f8dSStefano Garzarella /* 53137b06f8dSStefano Garzarella * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check 53237b06f8dSStefano Garzarella * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in 53337b06f8dSStefano Garzarella * the integer variable. 53437b06f8dSStefano Garzarella */ 53537b06f8dSStefano Garzarella if (unlikely(num_sectors > max_sectors)) { 53637b06f8dSStefano Garzarella err_status = VIRTIO_BLK_S_IOERR; 53737b06f8dSStefano Garzarella goto err; 53837b06f8dSStefano Garzarella } 53937b06f8dSStefano Garzarella 54037b06f8dSStefano Garzarella bytes = num_sectors << BDRV_SECTOR_BITS; 54137b06f8dSStefano Garzarella 54237b06f8dSStefano Garzarella if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) { 54337b06f8dSStefano Garzarella err_status = VIRTIO_BLK_S_IOERR; 54437b06f8dSStefano Garzarella goto err; 54537b06f8dSStefano Garzarella } 54637b06f8dSStefano Garzarella 54737b06f8dSStefano Garzarella /* 54837b06f8dSStefano Garzarella * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard 54937b06f8dSStefano Garzarella * and write zeroes commands if any unknown flag is set. 55037b06f8dSStefano Garzarella */ 55137b06f8dSStefano Garzarella if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { 55237b06f8dSStefano Garzarella err_status = VIRTIO_BLK_S_UNSUPP; 55337b06f8dSStefano Garzarella goto err; 55437b06f8dSStefano Garzarella } 55537b06f8dSStefano Garzarella 55637b06f8dSStefano Garzarella if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */ 55737b06f8dSStefano Garzarella int blk_aio_flags = 0; 55837b06f8dSStefano Garzarella 55937b06f8dSStefano Garzarella if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) { 56037b06f8dSStefano Garzarella blk_aio_flags |= BDRV_REQ_MAY_UNMAP; 56137b06f8dSStefano Garzarella } 56237b06f8dSStefano Garzarella 56337b06f8dSStefano Garzarella block_acct_start(blk_get_stats(s->blk), &req->acct, bytes, 56437b06f8dSStefano Garzarella BLOCK_ACCT_WRITE); 56537b06f8dSStefano Garzarella 56637b06f8dSStefano Garzarella blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS, 56737b06f8dSStefano Garzarella bytes, blk_aio_flags, 56837b06f8dSStefano Garzarella virtio_blk_discard_write_zeroes_complete, req); 56937b06f8dSStefano Garzarella } else { /* VIRTIO_BLK_T_DISCARD */ 57037b06f8dSStefano Garzarella /* 57137b06f8dSStefano Garzarella * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for 57237b06f8dSStefano Garzarella * discard commands if the unmap flag is set. 57337b06f8dSStefano Garzarella */ 57437b06f8dSStefano Garzarella if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { 57537b06f8dSStefano Garzarella err_status = VIRTIO_BLK_S_UNSUPP; 57637b06f8dSStefano Garzarella goto err; 57737b06f8dSStefano Garzarella } 57837b06f8dSStefano Garzarella 57937b06f8dSStefano Garzarella blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes, 58037b06f8dSStefano Garzarella virtio_blk_discard_write_zeroes_complete, req); 58137b06f8dSStefano Garzarella } 58237b06f8dSStefano Garzarella 58337b06f8dSStefano Garzarella return VIRTIO_BLK_S_OK; 58437b06f8dSStefano Garzarella 58537b06f8dSStefano Garzarella err: 58637b06f8dSStefano Garzarella if (is_write_zeroes) { 58737b06f8dSStefano Garzarella block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE); 58837b06f8dSStefano Garzarella } 58937b06f8dSStefano Garzarella return err_status; 59037b06f8dSStefano Garzarella } 59137b06f8dSStefano Garzarella 5924f736650SSam Li typedef struct ZoneCmdData { 5934f736650SSam Li VirtIOBlockReq *req; 5944f736650SSam Li struct iovec *in_iov; 5954f736650SSam Li unsigned in_num; 5964f736650SSam Li union { 5974f736650SSam Li struct { 5984f736650SSam Li unsigned int nr_zones; 5994f736650SSam Li BlockZoneDescriptor *zones; 6004f736650SSam Li } zone_report_data; 6014f736650SSam Li struct { 6024f736650SSam Li int64_t offset; 6034f736650SSam Li } zone_append_data; 6044f736650SSam Li }; 6054f736650SSam Li } ZoneCmdData; 6064f736650SSam Li 6074f736650SSam Li /* 6084f736650SSam Li * check zoned_request: error checking before issuing requests. If all checks 6094f736650SSam Li * passed, return true. 6104f736650SSam Li * append: true if only zone append requests issued. 6114f736650SSam Li */ 6124f736650SSam Li static bool check_zoned_request(VirtIOBlock *s, int64_t offset, int64_t len, 6134f736650SSam Li bool append, uint8_t *status) { 6144f736650SSam Li BlockDriverState *bs = blk_bs(s->blk); 6154f736650SSam Li int index; 6164f736650SSam Li 6174f736650SSam Li if (!virtio_has_feature(s->host_features, VIRTIO_BLK_F_ZONED)) { 6184f736650SSam Li *status = VIRTIO_BLK_S_UNSUPP; 6194f736650SSam Li return false; 6204f736650SSam Li } 6214f736650SSam Li 6224f736650SSam Li if (offset < 0 || len < 0 || len > (bs->total_sectors << BDRV_SECTOR_BITS) 6234f736650SSam Li || offset > (bs->total_sectors << BDRV_SECTOR_BITS) - len) { 6244f736650SSam Li *status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 6254f736650SSam Li return false; 6264f736650SSam Li } 6274f736650SSam Li 6284f736650SSam Li if (append) { 6294f736650SSam Li if (bs->bl.write_granularity) { 6304f736650SSam Li if ((offset % bs->bl.write_granularity) != 0) { 6314f736650SSam Li *status = VIRTIO_BLK_S_ZONE_UNALIGNED_WP; 6324f736650SSam Li return false; 6334f736650SSam Li } 6344f736650SSam Li } 6354f736650SSam Li 6364f736650SSam Li index = offset / bs->bl.zone_size; 6374f736650SSam Li if (BDRV_ZT_IS_CONV(bs->wps->wp[index])) { 6384f736650SSam Li *status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 6394f736650SSam Li return false; 6404f736650SSam Li } 6414f736650SSam Li 6424f736650SSam Li if (len / 512 > bs->bl.max_append_sectors) { 6434f736650SSam Li if (bs->bl.max_append_sectors == 0) { 6444f736650SSam Li *status = VIRTIO_BLK_S_UNSUPP; 6454f736650SSam Li } else { 6464f736650SSam Li *status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 6474f736650SSam Li } 6484f736650SSam Li return false; 6494f736650SSam Li } 6504f736650SSam Li } 6514f736650SSam Li return true; 6524f736650SSam Li } 6534f736650SSam Li 6544f736650SSam Li static void virtio_blk_zone_report_complete(void *opaque, int ret) 6554f736650SSam Li { 6564f736650SSam Li ZoneCmdData *data = opaque; 6574f736650SSam Li VirtIOBlockReq *req = data->req; 6584f736650SSam Li VirtIODevice *vdev = VIRTIO_DEVICE(req->dev); 6594f736650SSam Li struct iovec *in_iov = data->in_iov; 6604f736650SSam Li unsigned in_num = data->in_num; 6614f736650SSam Li int64_t zrp_size, n, j = 0; 6624f736650SSam Li int64_t nz = data->zone_report_data.nr_zones; 6634f736650SSam Li int8_t err_status = VIRTIO_BLK_S_OK; 6644f736650SSam Li 6654e92acf7SSam Li trace_virtio_blk_zone_report_complete(vdev, req, nz, ret); 6664f736650SSam Li if (ret) { 6674f736650SSam Li err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 6684f736650SSam Li goto out; 6694f736650SSam Li } 6704f736650SSam Li 6714f736650SSam Li struct virtio_blk_zone_report zrp_hdr = (struct virtio_blk_zone_report) { 6724f736650SSam Li .nr_zones = cpu_to_le64(nz), 6734f736650SSam Li }; 6744f736650SSam Li zrp_size = sizeof(struct virtio_blk_zone_report) 6754f736650SSam Li + sizeof(struct virtio_blk_zone_descriptor) * nz; 6764f736650SSam Li n = iov_from_buf(in_iov, in_num, 0, &zrp_hdr, sizeof(zrp_hdr)); 6774f736650SSam Li if (n != sizeof(zrp_hdr)) { 6784f736650SSam Li virtio_error(vdev, "Driver provided input buffer that is too small!"); 6794f736650SSam Li err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 6804f736650SSam Li goto out; 6814f736650SSam Li } 6824f736650SSam Li 6834f736650SSam Li for (size_t i = sizeof(zrp_hdr); i < zrp_size; 6844f736650SSam Li i += sizeof(struct virtio_blk_zone_descriptor), ++j) { 6854f736650SSam Li struct virtio_blk_zone_descriptor desc = 6864f736650SSam Li (struct virtio_blk_zone_descriptor) { 6874f736650SSam Li .z_start = cpu_to_le64(data->zone_report_data.zones[j].start 6884f736650SSam Li >> BDRV_SECTOR_BITS), 6894f736650SSam Li .z_cap = cpu_to_le64(data->zone_report_data.zones[j].cap 6904f736650SSam Li >> BDRV_SECTOR_BITS), 6914f736650SSam Li .z_wp = cpu_to_le64(data->zone_report_data.zones[j].wp 6924f736650SSam Li >> BDRV_SECTOR_BITS), 6934f736650SSam Li }; 6944f736650SSam Li 6954f736650SSam Li switch (data->zone_report_data.zones[j].type) { 6964f736650SSam Li case BLK_ZT_CONV: 6974f736650SSam Li desc.z_type = VIRTIO_BLK_ZT_CONV; 6984f736650SSam Li break; 6994f736650SSam Li case BLK_ZT_SWR: 7004f736650SSam Li desc.z_type = VIRTIO_BLK_ZT_SWR; 7014f736650SSam Li break; 7024f736650SSam Li case BLK_ZT_SWP: 7034f736650SSam Li desc.z_type = VIRTIO_BLK_ZT_SWP; 7044f736650SSam Li break; 7054f736650SSam Li default: 7064f736650SSam Li g_assert_not_reached(); 7074f736650SSam Li } 7084f736650SSam Li 7094f736650SSam Li switch (data->zone_report_data.zones[j].state) { 7104f736650SSam Li case BLK_ZS_RDONLY: 7114f736650SSam Li desc.z_state = VIRTIO_BLK_ZS_RDONLY; 7124f736650SSam Li break; 7134f736650SSam Li case BLK_ZS_OFFLINE: 7144f736650SSam Li desc.z_state = VIRTIO_BLK_ZS_OFFLINE; 7154f736650SSam Li break; 7164f736650SSam Li case BLK_ZS_EMPTY: 7174f736650SSam Li desc.z_state = VIRTIO_BLK_ZS_EMPTY; 7184f736650SSam Li break; 7194f736650SSam Li case BLK_ZS_CLOSED: 7204f736650SSam Li desc.z_state = VIRTIO_BLK_ZS_CLOSED; 7214f736650SSam Li break; 7224f736650SSam Li case BLK_ZS_FULL: 7234f736650SSam Li desc.z_state = VIRTIO_BLK_ZS_FULL; 7244f736650SSam Li break; 7254f736650SSam Li case BLK_ZS_EOPEN: 7264f736650SSam Li desc.z_state = VIRTIO_BLK_ZS_EOPEN; 7274f736650SSam Li break; 7284f736650SSam Li case BLK_ZS_IOPEN: 7294f736650SSam Li desc.z_state = VIRTIO_BLK_ZS_IOPEN; 7304f736650SSam Li break; 7314f736650SSam Li case BLK_ZS_NOT_WP: 7324f736650SSam Li desc.z_state = VIRTIO_BLK_ZS_NOT_WP; 7334f736650SSam Li break; 7344f736650SSam Li default: 7354f736650SSam Li g_assert_not_reached(); 7364f736650SSam Li } 7374f736650SSam Li 7384f736650SSam Li /* TODO: it takes O(n^2) time complexity. Optimizations required. */ 7394f736650SSam Li n = iov_from_buf(in_iov, in_num, i, &desc, sizeof(desc)); 7404f736650SSam Li if (n != sizeof(desc)) { 7414f736650SSam Li virtio_error(vdev, "Driver provided input buffer " 7424f736650SSam Li "for descriptors that is too small!"); 7434f736650SSam Li err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 7444f736650SSam Li } 7454f736650SSam Li } 7464f736650SSam Li 7474f736650SSam Li out: 7484f736650SSam Li virtio_blk_req_complete(req, err_status); 7494f736650SSam Li virtio_blk_free_request(req); 7504f736650SSam Li g_free(data->zone_report_data.zones); 7514f736650SSam Li g_free(data); 7524f736650SSam Li } 7534f736650SSam Li 7544f736650SSam Li static void virtio_blk_handle_zone_report(VirtIOBlockReq *req, 7554f736650SSam Li struct iovec *in_iov, 7564f736650SSam Li unsigned in_num) 7574f736650SSam Li { 7584f736650SSam Li VirtIOBlock *s = req->dev; 7594f736650SSam Li VirtIODevice *vdev = VIRTIO_DEVICE(s); 7604f736650SSam Li unsigned int nr_zones; 7614f736650SSam Li ZoneCmdData *data; 7624f736650SSam Li int64_t zone_size, offset; 7634f736650SSam Li uint8_t err_status; 7644f736650SSam Li 7654f736650SSam Li if (req->in_len < sizeof(struct virtio_blk_inhdr) + 7664f736650SSam Li sizeof(struct virtio_blk_zone_report) + 7674f736650SSam Li sizeof(struct virtio_blk_zone_descriptor)) { 7684f736650SSam Li virtio_error(vdev, "in buffer too small for zone report"); 7694f736650SSam Li return; 7704f736650SSam Li } 7714f736650SSam Li 7724f736650SSam Li /* start byte offset of the zone report */ 7734f736650SSam Li offset = virtio_ldq_p(vdev, &req->out.sector) << BDRV_SECTOR_BITS; 7744f736650SSam Li if (!check_zoned_request(s, offset, 0, false, &err_status)) { 7754f736650SSam Li goto out; 7764f736650SSam Li } 7774f736650SSam Li nr_zones = (req->in_len - sizeof(struct virtio_blk_inhdr) - 7784f736650SSam Li sizeof(struct virtio_blk_zone_report)) / 7794f736650SSam Li sizeof(struct virtio_blk_zone_descriptor); 7804e92acf7SSam Li trace_virtio_blk_handle_zone_report(vdev, req, 7814e92acf7SSam Li offset >> BDRV_SECTOR_BITS, nr_zones); 7824f736650SSam Li 7834f736650SSam Li zone_size = sizeof(BlockZoneDescriptor) * nr_zones; 7844f736650SSam Li data = g_malloc(sizeof(ZoneCmdData)); 7854f736650SSam Li data->req = req; 7864f736650SSam Li data->in_iov = in_iov; 7874f736650SSam Li data->in_num = in_num; 7884f736650SSam Li data->zone_report_data.nr_zones = nr_zones; 7894f736650SSam Li data->zone_report_data.zones = g_malloc(zone_size), 7904f736650SSam Li 7914f736650SSam Li blk_aio_zone_report(s->blk, offset, &data->zone_report_data.nr_zones, 7924f736650SSam Li data->zone_report_data.zones, 7934f736650SSam Li virtio_blk_zone_report_complete, data); 7944f736650SSam Li return; 7954f736650SSam Li out: 7964f736650SSam Li virtio_blk_req_complete(req, err_status); 7974f736650SSam Li virtio_blk_free_request(req); 7984f736650SSam Li } 7994f736650SSam Li 8004f736650SSam Li static void virtio_blk_zone_mgmt_complete(void *opaque, int ret) 8014f736650SSam Li { 8024f736650SSam Li VirtIOBlockReq *req = opaque; 8034f736650SSam Li VirtIOBlock *s = req->dev; 8044e92acf7SSam Li VirtIODevice *vdev = VIRTIO_DEVICE(s); 8054f736650SSam Li int8_t err_status = VIRTIO_BLK_S_OK; 8064e92acf7SSam Li trace_virtio_blk_zone_mgmt_complete(vdev, req,ret); 8074f736650SSam Li 8084f736650SSam Li if (ret) { 8094f736650SSam Li err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 8104f736650SSam Li } 8114f736650SSam Li 8124f736650SSam Li virtio_blk_req_complete(req, err_status); 8134f736650SSam Li virtio_blk_free_request(req); 8144f736650SSam Li } 8154f736650SSam Li 8164f736650SSam Li static int virtio_blk_handle_zone_mgmt(VirtIOBlockReq *req, BlockZoneOp op) 8174f736650SSam Li { 8184f736650SSam Li VirtIOBlock *s = req->dev; 8194f736650SSam Li VirtIODevice *vdev = VIRTIO_DEVICE(s); 8204f736650SSam Li BlockDriverState *bs = blk_bs(s->blk); 8214f736650SSam Li int64_t offset = virtio_ldq_p(vdev, &req->out.sector) << BDRV_SECTOR_BITS; 8224f736650SSam Li uint64_t len; 8234f736650SSam Li uint64_t capacity = bs->total_sectors << BDRV_SECTOR_BITS; 8244f736650SSam Li uint8_t err_status = VIRTIO_BLK_S_OK; 8254f736650SSam Li 8264f736650SSam Li uint32_t type = virtio_ldl_p(vdev, &req->out.type); 8274f736650SSam Li if (type == VIRTIO_BLK_T_ZONE_RESET_ALL) { 8284f736650SSam Li /* Entire drive capacity */ 8294f736650SSam Li offset = 0; 8304f736650SSam Li len = capacity; 8314e92acf7SSam Li trace_virtio_blk_handle_zone_reset_all(vdev, req, 0, 8324e92acf7SSam Li bs->total_sectors); 8334f736650SSam Li } else { 8344f736650SSam Li if (bs->bl.zone_size > capacity - offset) { 8354f736650SSam Li /* The zoned device allows the last smaller zone. */ 8364f736650SSam Li len = capacity - bs->bl.zone_size * (bs->bl.nr_zones - 1); 8374f736650SSam Li } else { 8384f736650SSam Li len = bs->bl.zone_size; 8394f736650SSam Li } 8404e92acf7SSam Li trace_virtio_blk_handle_zone_mgmt(vdev, req, op, 8414e92acf7SSam Li offset >> BDRV_SECTOR_BITS, 8424e92acf7SSam Li len >> BDRV_SECTOR_BITS); 8434f736650SSam Li } 8444f736650SSam Li 8454f736650SSam Li if (!check_zoned_request(s, offset, len, false, &err_status)) { 8464f736650SSam Li goto out; 8474f736650SSam Li } 8484f736650SSam Li 8494f736650SSam Li blk_aio_zone_mgmt(s->blk, op, offset, len, 8504f736650SSam Li virtio_blk_zone_mgmt_complete, req); 8514f736650SSam Li 8524f736650SSam Li return 0; 8534f736650SSam Li out: 8544f736650SSam Li virtio_blk_req_complete(req, err_status); 8554f736650SSam Li virtio_blk_free_request(req); 8564f736650SSam Li return err_status; 8574f736650SSam Li } 8584f736650SSam Li 8594f736650SSam Li static void virtio_blk_zone_append_complete(void *opaque, int ret) 8604f736650SSam Li { 8614f736650SSam Li ZoneCmdData *data = opaque; 8624f736650SSam Li VirtIOBlockReq *req = data->req; 8634f736650SSam Li VirtIODevice *vdev = VIRTIO_DEVICE(req->dev); 8644f736650SSam Li int64_t append_sector, n; 8654f736650SSam Li uint8_t err_status = VIRTIO_BLK_S_OK; 8664f736650SSam Li 8674f736650SSam Li if (ret) { 8684f736650SSam Li err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 8694f736650SSam Li goto out; 8704f736650SSam Li } 8714f736650SSam Li 8724f736650SSam Li virtio_stq_p(vdev, &append_sector, 8734f736650SSam Li data->zone_append_data.offset >> BDRV_SECTOR_BITS); 8744f736650SSam Li n = iov_from_buf(data->in_iov, data->in_num, 0, &append_sector, 8754f736650SSam Li sizeof(append_sector)); 8764f736650SSam Li if (n != sizeof(append_sector)) { 8774f736650SSam Li virtio_error(vdev, "Driver provided input buffer less than size of " 8784f736650SSam Li "append_sector"); 8794f736650SSam Li err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 8804f736650SSam Li goto out; 8814f736650SSam Li } 8824e92acf7SSam Li trace_virtio_blk_zone_append_complete(vdev, req, append_sector, ret); 8834f736650SSam Li 8844f736650SSam Li out: 8854f736650SSam Li virtio_blk_req_complete(req, err_status); 8864f736650SSam Li virtio_blk_free_request(req); 8874f736650SSam Li g_free(data); 8884f736650SSam Li } 8894f736650SSam Li 8904f736650SSam Li static int virtio_blk_handle_zone_append(VirtIOBlockReq *req, 8914f736650SSam Li struct iovec *out_iov, 8924f736650SSam Li struct iovec *in_iov, 8934f736650SSam Li uint64_t out_num, 8944f736650SSam Li unsigned in_num) { 8954f736650SSam Li VirtIOBlock *s = req->dev; 8964f736650SSam Li VirtIODevice *vdev = VIRTIO_DEVICE(s); 8974f736650SSam Li uint8_t err_status = VIRTIO_BLK_S_OK; 8984f736650SSam Li 8994f736650SSam Li int64_t offset = virtio_ldq_p(vdev, &req->out.sector) << BDRV_SECTOR_BITS; 9004f736650SSam Li int64_t len = iov_size(out_iov, out_num); 9014f736650SSam Li 9024e92acf7SSam Li trace_virtio_blk_handle_zone_append(vdev, req, offset >> BDRV_SECTOR_BITS); 9034f736650SSam Li if (!check_zoned_request(s, offset, len, true, &err_status)) { 9044f736650SSam Li goto out; 9054f736650SSam Li } 9064f736650SSam Li 9074f736650SSam Li ZoneCmdData *data = g_malloc(sizeof(ZoneCmdData)); 9084f736650SSam Li data->req = req; 9094f736650SSam Li data->in_iov = in_iov; 9104f736650SSam Li data->in_num = in_num; 9114f736650SSam Li data->zone_append_data.offset = offset; 9124f736650SSam Li qemu_iovec_init_external(&req->qiov, out_iov, out_num); 91352eb76f4SSam Li 91452eb76f4SSam Li block_acct_start(blk_get_stats(s->blk), &req->acct, len, 91552eb76f4SSam Li BLOCK_ACCT_ZONE_APPEND); 91652eb76f4SSam Li 9174f736650SSam Li blk_aio_zone_append(s->blk, &data->zone_append_data.offset, &req->qiov, 0, 9184f736650SSam Li virtio_blk_zone_append_complete, data); 9194f736650SSam Li return 0; 9204f736650SSam Li 9214f736650SSam Li out: 9224f736650SSam Li virtio_blk_req_complete(req, err_status); 9234f736650SSam Li virtio_blk_free_request(req); 9244f736650SSam Li return err_status; 9254f736650SSam Li } 9264f736650SSam Li 92720ea686aSGreg Kurz static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) 9286e790746SPaolo Bonzini { 9296e790746SPaolo Bonzini uint32_t type; 930f897bf75SStefan Hajnoczi struct iovec *in_iov = req->elem.in_sg; 9315636da76SDongli Zhang struct iovec *out_iov = req->elem.out_sg; 932f897bf75SStefan Hajnoczi unsigned in_num = req->elem.in_num; 933f897bf75SStefan Hajnoczi unsigned out_num = req->elem.out_num; 93420ea686aSGreg Kurz VirtIOBlock *s = req->dev; 93520ea686aSGreg Kurz VirtIODevice *vdev = VIRTIO_DEVICE(s); 9366e790746SPaolo Bonzini 937f897bf75SStefan Hajnoczi if (req->elem.out_num < 1 || req->elem.in_num < 1) { 93820ea686aSGreg Kurz virtio_error(vdev, "virtio-blk missing headers"); 93920ea686aSGreg Kurz return -1; 9406e790746SPaolo Bonzini } 9416e790746SPaolo Bonzini 9425636da76SDongli Zhang if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out, 943827805a2SFam Zheng sizeof(req->out)) != sizeof(req->out))) { 94420ea686aSGreg Kurz virtio_error(vdev, "virtio-blk request outhdr too short"); 94520ea686aSGreg Kurz return -1; 946827805a2SFam Zheng } 947ee17e848SFam Zheng 9487bd04a04SStefan Hajnoczi iov_discard_front_undoable(&out_iov, &out_num, sizeof(req->out), 9497bd04a04SStefan Hajnoczi &req->outhdr_undo); 950ee17e848SFam Zheng 95112048545SGonglei if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) { 95220ea686aSGreg Kurz virtio_error(vdev, "virtio-blk request inhdr too short"); 9537bd04a04SStefan Hajnoczi iov_discard_undo(&req->outhdr_undo); 95420ea686aSGreg Kurz return -1; 955ee17e848SFam Zheng } 956ee17e848SFam Zheng 9572a6cdd6dSPaolo Bonzini /* We always touch the last byte, so just see how big in_iov is. */ 9582a6cdd6dSPaolo Bonzini req->in_len = iov_size(in_iov, in_num); 959ee17e848SFam Zheng req->in = (void *)in_iov[in_num - 1].iov_base 960ee17e848SFam Zheng + in_iov[in_num - 1].iov_len 961ee17e848SFam Zheng - sizeof(struct virtio_blk_inhdr); 9627bd04a04SStefan Hajnoczi iov_discard_back_undoable(in_iov, &in_num, sizeof(struct virtio_blk_inhdr), 9637bd04a04SStefan Hajnoczi &req->inhdr_undo); 9646e790746SPaolo Bonzini 9659a6719d5SStefano Garzarella type = virtio_ldl_p(vdev, &req->out.type); 9666e790746SPaolo Bonzini 96795f7142aSPeter Lieven /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER 968631b22eaSStefan Weil * is an optional flag. Although a guest should not send this flag if 96995f7142aSPeter Lieven * not negotiated we ignored it in the past. So keep ignoring it. */ 97095f7142aSPeter Lieven switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) { 97195f7142aSPeter Lieven case VIRTIO_BLK_T_IN: 97295f7142aSPeter Lieven { 97395f7142aSPeter Lieven bool is_write = type & VIRTIO_BLK_T_OUT; 9749a6719d5SStefano Garzarella req->sector_num = virtio_ldq_p(vdev, &req->out.sector); 97595f7142aSPeter Lieven 97695f7142aSPeter Lieven if (is_write) { 9775636da76SDongli Zhang qemu_iovec_init_external(&req->qiov, out_iov, out_num); 978a576ceacSStefan Hajnoczi trace_virtio_blk_handle_write(vdev, req, req->sector_num, 97995f7142aSPeter Lieven req->qiov.size / BDRV_SECTOR_SIZE); 98095f7142aSPeter Lieven } else { 98195f7142aSPeter Lieven qemu_iovec_init_external(&req->qiov, in_iov, in_num); 982a576ceacSStefan Hajnoczi trace_virtio_blk_handle_read(vdev, req, req->sector_num, 98395f7142aSPeter Lieven req->qiov.size / BDRV_SECTOR_SIZE); 98495f7142aSPeter Lieven } 98595f7142aSPeter Lieven 9869a6719d5SStefano Garzarella if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) { 98795f7142aSPeter Lieven virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); 9889a6719d5SStefano Garzarella block_acct_invalid(blk_get_stats(s->blk), 98901762e03SAlberto Garcia is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); 99095f7142aSPeter Lieven virtio_blk_free_request(req); 99120ea686aSGreg Kurz return 0; 99295f7142aSPeter Lieven } 99395f7142aSPeter Lieven 9949a6719d5SStefano Garzarella block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size, 99595f7142aSPeter Lieven is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); 99695f7142aSPeter Lieven 99795f7142aSPeter Lieven /* merge would exceed maximum number of requests or IO direction 99895f7142aSPeter Lieven * changes */ 99995f7142aSPeter Lieven if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS || 1000c99495acSPeter Lieven is_write != mrb->is_write || 10019a6719d5SStefano Garzarella !s->conf.request_merging)) { 1002baf42268SStefan Hajnoczi virtio_blk_submit_multireq(s, mrb); 100395f7142aSPeter Lieven } 100495f7142aSPeter Lieven 100595f7142aSPeter Lieven assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS); 100695f7142aSPeter Lieven mrb->reqs[mrb->num_reqs++] = req; 100795f7142aSPeter Lieven mrb->is_write = is_write; 100895f7142aSPeter Lieven break; 100995f7142aSPeter Lieven } 101095f7142aSPeter Lieven case VIRTIO_BLK_T_FLUSH: 10116e790746SPaolo Bonzini virtio_blk_handle_flush(req, mrb); 101295f7142aSPeter Lieven break; 10134f736650SSam Li case VIRTIO_BLK_T_ZONE_REPORT: 10144f736650SSam Li virtio_blk_handle_zone_report(req, in_iov, in_num); 10154f736650SSam Li break; 10164f736650SSam Li case VIRTIO_BLK_T_ZONE_OPEN: 10174f736650SSam Li virtio_blk_handle_zone_mgmt(req, BLK_ZO_OPEN); 10184f736650SSam Li break; 10194f736650SSam Li case VIRTIO_BLK_T_ZONE_CLOSE: 10204f736650SSam Li virtio_blk_handle_zone_mgmt(req, BLK_ZO_CLOSE); 10214f736650SSam Li break; 10224f736650SSam Li case VIRTIO_BLK_T_ZONE_FINISH: 10234f736650SSam Li virtio_blk_handle_zone_mgmt(req, BLK_ZO_FINISH); 10244f736650SSam Li break; 10254f736650SSam Li case VIRTIO_BLK_T_ZONE_RESET: 10264f736650SSam Li virtio_blk_handle_zone_mgmt(req, BLK_ZO_RESET); 10274f736650SSam Li break; 10284f736650SSam Li case VIRTIO_BLK_T_ZONE_RESET_ALL: 10294f736650SSam Li virtio_blk_handle_zone_mgmt(req, BLK_ZO_RESET); 10304f736650SSam Li break; 103195f7142aSPeter Lieven case VIRTIO_BLK_T_SCSI_CMD: 10326e790746SPaolo Bonzini virtio_blk_handle_scsi(req); 103395f7142aSPeter Lieven break; 103495f7142aSPeter Lieven case VIRTIO_BLK_T_GET_ID: 103595f7142aSPeter Lieven { 10366e790746SPaolo Bonzini /* 10376e790746SPaolo Bonzini * NB: per existing s/n string convention the string is 10386e790746SPaolo Bonzini * terminated by '\0' only when shorter than buffer. 10396e790746SPaolo Bonzini */ 10402a30307fSMarkus Armbruster const char *serial = s->conf.serial ? s->conf.serial : ""; 1041a83ceea8SMarc Marí size_t size = MIN(strlen(serial) + 1, 1042a83ceea8SMarc Marí MIN(iov_size(in_iov, in_num), 1043a83ceea8SMarc Marí VIRTIO_BLK_ID_BYTES)); 1044a83ceea8SMarc Marí iov_from_buf(in_iov, in_num, 0, serial, size); 10456e790746SPaolo Bonzini virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 1046671ec3f0SFam Zheng virtio_blk_free_request(req); 104795f7142aSPeter Lieven break; 104895f7142aSPeter Lieven } 10494f736650SSam Li case VIRTIO_BLK_T_ZONE_APPEND & ~VIRTIO_BLK_T_OUT: 10504f736650SSam Li /* 10514f736650SSam Li * Passing out_iov/out_num and in_iov/in_num is not safe 10524f736650SSam Li * to access req->elem.out_sg directly because it may be 10534f736650SSam Li * modified by virtio_blk_handle_request(). 10544f736650SSam Li */ 10554f736650SSam Li virtio_blk_handle_zone_append(req, out_iov, in_iov, out_num, in_num); 10564f736650SSam Li break; 105737b06f8dSStefano Garzarella /* 105837b06f8dSStefano Garzarella * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with 105937b06f8dSStefano Garzarella * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement, 106037b06f8dSStefano Garzarella * so we must mask it for these requests, then we will check if it is set. 106137b06f8dSStefano Garzarella */ 106237b06f8dSStefano Garzarella case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT: 106337b06f8dSStefano Garzarella case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT: 106437b06f8dSStefano Garzarella { 106537b06f8dSStefano Garzarella struct virtio_blk_discard_write_zeroes dwz_hdr; 106637b06f8dSStefano Garzarella size_t out_len = iov_size(out_iov, out_num); 106737b06f8dSStefano Garzarella bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) == 106837b06f8dSStefano Garzarella VIRTIO_BLK_T_WRITE_ZEROES; 106937b06f8dSStefano Garzarella uint8_t err_status; 107037b06f8dSStefano Garzarella 107137b06f8dSStefano Garzarella /* 107237b06f8dSStefano Garzarella * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains 107337b06f8dSStefano Garzarella * more than one segment. 107437b06f8dSStefano Garzarella */ 107537b06f8dSStefano Garzarella if (unlikely(!(type & VIRTIO_BLK_T_OUT) || 107637b06f8dSStefano Garzarella out_len > sizeof(dwz_hdr))) { 107737b06f8dSStefano Garzarella virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); 107837b06f8dSStefano Garzarella virtio_blk_free_request(req); 107937b06f8dSStefano Garzarella return 0; 108037b06f8dSStefano Garzarella } 108137b06f8dSStefano Garzarella 108237b06f8dSStefano Garzarella if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr, 108337b06f8dSStefano Garzarella sizeof(dwz_hdr)) != sizeof(dwz_hdr))) { 10847bd04a04SStefan Hajnoczi iov_discard_undo(&req->inhdr_undo); 10857bd04a04SStefan Hajnoczi iov_discard_undo(&req->outhdr_undo); 108637b06f8dSStefano Garzarella virtio_error(vdev, "virtio-blk discard/write_zeroes header" 108737b06f8dSStefano Garzarella " too short"); 108837b06f8dSStefano Garzarella return -1; 108937b06f8dSStefano Garzarella } 109037b06f8dSStefano Garzarella 109137b06f8dSStefano Garzarella err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr, 109237b06f8dSStefano Garzarella is_write_zeroes); 109337b06f8dSStefano Garzarella if (err_status != VIRTIO_BLK_S_OK) { 109437b06f8dSStefano Garzarella virtio_blk_req_complete(req, err_status); 109537b06f8dSStefano Garzarella virtio_blk_free_request(req); 109637b06f8dSStefano Garzarella } 109737b06f8dSStefano Garzarella 109837b06f8dSStefano Garzarella break; 109937b06f8dSStefano Garzarella } 110095f7142aSPeter Lieven default: 11016e790746SPaolo Bonzini virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); 1102671ec3f0SFam Zheng virtio_blk_free_request(req); 11036e790746SPaolo Bonzini } 110420ea686aSGreg Kurz return 0; 11056e790746SPaolo Bonzini } 11066e790746SPaolo Bonzini 1107186b9691SStefan Hajnoczi void virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq) 11086e790746SPaolo Bonzini { 11096e790746SPaolo Bonzini VirtIOBlockReq *req; 111095f7142aSPeter Lieven MultiReqBuffer mrb = {}; 1111d0435bc5SStefan Hajnoczi bool suppress_notifications = virtio_queue_get_notification(vq); 11126e790746SPaolo Bonzini 1113ccee48aaSStefan Hajnoczi defer_call_begin(); 1114fc73548eSStefan Hajnoczi 11159ef9d402SStefan Hajnoczi do { 1116d0435bc5SStefan Hajnoczi if (suppress_notifications) { 11179ef9d402SStefan Hajnoczi virtio_queue_set_notification(vq, 0); 1118d0435bc5SStefan Hajnoczi } 11199ef9d402SStefan Hajnoczi 1120edaffd9fSStefan Hajnoczi while ((req = virtio_blk_get_request(s, vq))) { 112120ea686aSGreg Kurz if (virtio_blk_handle_request(req, &mrb)) { 112220ea686aSGreg Kurz virtqueue_detach_element(req->vq, &req->elem, 0); 112320ea686aSGreg Kurz virtio_blk_free_request(req); 112420ea686aSGreg Kurz break; 112520ea686aSGreg Kurz } 11266e790746SPaolo Bonzini } 11276e790746SPaolo Bonzini 1128d0435bc5SStefan Hajnoczi if (suppress_notifications) { 11299ef9d402SStefan Hajnoczi virtio_queue_set_notification(vq, 1); 1130d0435bc5SStefan Hajnoczi } 11319ef9d402SStefan Hajnoczi } while (!virtio_queue_empty(vq)); 11329ef9d402SStefan Hajnoczi 113395f7142aSPeter Lieven if (mrb.num_reqs) { 1134baf42268SStefan Hajnoczi virtio_blk_submit_multireq(s, &mrb); 113595f7142aSPeter Lieven } 1136fc73548eSStefan Hajnoczi 1137ccee48aaSStefan Hajnoczi defer_call_end(); 11386e790746SPaolo Bonzini } 11396e790746SPaolo Bonzini 11408a2fad57SMichael S. Tsirkin static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq) 11418a2fad57SMichael S. Tsirkin { 11428a2fad57SMichael S. Tsirkin VirtIOBlock *s = (VirtIOBlock *)vdev; 11438a2fad57SMichael S. Tsirkin 11443cdaf3ddSStefan Hajnoczi if (!s->ioeventfd_disabled && !s->ioeventfd_started) { 11458a2fad57SMichael S. Tsirkin /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start 11463cdaf3ddSStefan Hajnoczi * ioeventfd here instead of waiting for .set_status(). 11478a2fad57SMichael S. Tsirkin */ 11489ffe337cSPaolo Bonzini virtio_device_start_ioeventfd(vdev); 11493cdaf3ddSStefan Hajnoczi if (!s->ioeventfd_disabled) { 11508a2fad57SMichael S. Tsirkin return; 11518a2fad57SMichael S. Tsirkin } 11528a2fad57SMichael S. Tsirkin } 1153b6948ab0SStefan Hajnoczi 1154186b9691SStefan Hajnoczi virtio_blk_handle_vq(s, vq); 11558a2fad57SMichael S. Tsirkin } 11568a2fad57SMichael S. Tsirkin 1157a937f8e8SStefan Hajnoczi static void virtio_blk_dma_restart_bh(void *opaque) 11586e790746SPaolo Bonzini { 1159a937f8e8SStefan Hajnoczi VirtIOBlock *s = opaque; 1160a937f8e8SStefan Hajnoczi 11619c67f33fSStefan Hajnoczi VirtIOBlockReq *req; 116295f7142aSPeter Lieven MultiReqBuffer mrb = {}; 11636e790746SPaolo Bonzini 11649c67f33fSStefan Hajnoczi WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 11659c67f33fSStefan Hajnoczi req = s->rq; 11666e790746SPaolo Bonzini s->rq = NULL; 11679c67f33fSStefan Hajnoczi } 11686e790746SPaolo Bonzini 11696e790746SPaolo Bonzini while (req) { 11701bdb176aSzhanghailiang VirtIOBlockReq *next = req->next; 117120ea686aSGreg Kurz if (virtio_blk_handle_request(req, &mrb)) { 117220ea686aSGreg Kurz /* Device is now broken and won't do any processing until it gets 117320ea686aSGreg Kurz * reset. Already queued requests will be lost: let's purge them. 117420ea686aSGreg Kurz */ 117520ea686aSGreg Kurz while (req) { 117620ea686aSGreg Kurz next = req->next; 117720ea686aSGreg Kurz virtqueue_detach_element(req->vq, &req->elem, 0); 117820ea686aSGreg Kurz virtio_blk_free_request(req); 117920ea686aSGreg Kurz req = next; 118020ea686aSGreg Kurz } 118120ea686aSGreg Kurz break; 118220ea686aSGreg Kurz } 11831bdb176aSzhanghailiang req = next; 11846e790746SPaolo Bonzini } 11856e790746SPaolo Bonzini 118695f7142aSPeter Lieven if (mrb.num_reqs) { 1187baf42268SStefan Hajnoczi virtio_blk_submit_multireq(s, &mrb); 118895f7142aSPeter Lieven } 1189a937f8e8SStefan Hajnoczi 1190a937f8e8SStefan Hajnoczi /* Paired with inc in virtio_blk_dma_restart_cb() */ 1191680f2002SKevin Wolf blk_dec_in_flight(s->conf.conf.blk); 11926e790746SPaolo Bonzini } 11936e790746SPaolo Bonzini 1194538f0497SPhilippe Mathieu-Daudé static void virtio_blk_dma_restart_cb(void *opaque, bool running, 11956e790746SPaolo Bonzini RunState state) 11966e790746SPaolo Bonzini { 11976e790746SPaolo Bonzini VirtIOBlock *s = opaque; 11986e790746SPaolo Bonzini 11996e790746SPaolo Bonzini if (!running) { 12006e790746SPaolo Bonzini return; 12016e790746SPaolo Bonzini } 12026e790746SPaolo Bonzini 1203a937f8e8SStefan Hajnoczi /* Paired with dec in virtio_blk_dma_restart_bh() */ 1204680f2002SKevin Wolf blk_inc_in_flight(s->conf.conf.blk); 1205a937f8e8SStefan Hajnoczi 1206a937f8e8SStefan Hajnoczi aio_bh_schedule_oneshot(blk_get_aio_context(s->conf.conf.blk), 1207a937f8e8SStefan Hajnoczi virtio_blk_dma_restart_bh, s); 12086e790746SPaolo Bonzini } 12096e790746SPaolo Bonzini 12106e790746SPaolo Bonzini static void virtio_blk_reset(VirtIODevice *vdev) 12116e790746SPaolo Bonzini { 12126e790746SPaolo Bonzini VirtIOBlock *s = VIRTIO_BLK(vdev); 121326307f6aSFam Zheng VirtIOBlockReq *req; 12146e790746SPaolo Bonzini 12159c67f33fSStefan Hajnoczi /* Dataplane has stopped... */ 12163cdaf3ddSStefan Hajnoczi assert(!s->ioeventfd_started); 12179c67f33fSStefan Hajnoczi 12189c67f33fSStefan Hajnoczi /* ...but requests may still be in flight. */ 12196e40b3bfSAlexander Yarygin blk_drain(s->blk); 12206e40b3bfSAlexander Yarygin 122126307f6aSFam Zheng /* We drop queued requests after blk_drain() because blk_drain() itself can 122226307f6aSFam Zheng * produce them. */ 12239c67f33fSStefan Hajnoczi WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 122426307f6aSFam Zheng while (s->rq) { 122526307f6aSFam Zheng req = s->rq; 122626307f6aSFam Zheng s->rq = req->next; 12279c67f33fSStefan Hajnoczi 12289c67f33fSStefan Hajnoczi /* No other threads can access req->vq here */ 122997b93c8aSStefan Hajnoczi virtqueue_detach_element(req->vq, &req->elem, 0); 12309c67f33fSStefan Hajnoczi 123126307f6aSFam Zheng virtio_blk_free_request(req); 123226307f6aSFam Zheng } 12339c67f33fSStefan Hajnoczi } 123426307f6aSFam Zheng 12354be74634SMarkus Armbruster blk_set_enable_write_cache(s->blk, s->original_wce); 12366e790746SPaolo Bonzini } 12376e790746SPaolo Bonzini 12386e790746SPaolo Bonzini /* coalesce internal state, copy to pci i/o region 0 12396e790746SPaolo Bonzini */ 12406e790746SPaolo Bonzini static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config) 12416e790746SPaolo Bonzini { 12426e790746SPaolo Bonzini VirtIOBlock *s = VIRTIO_BLK(vdev); 12432a30307fSMarkus Armbruster BlockConf *conf = &s->conf.conf; 12444f736650SSam Li BlockDriverState *bs = blk_bs(s->blk); 12456e790746SPaolo Bonzini struct virtio_blk_config blkcfg; 12466e790746SPaolo Bonzini uint64_t capacity; 124717d0bc01SStefan Hajnoczi int64_t length; 1248f7516731SMarkus Armbruster int blk_size = conf->logical_block_size; 12496e790746SPaolo Bonzini 12504be74634SMarkus Armbruster blk_get_geometry(s->blk, &capacity); 12516e790746SPaolo Bonzini memset(&blkcfg, 0, sizeof(blkcfg)); 1252783d1897SRusty Russell virtio_stq_p(vdev, &blkcfg.capacity, capacity); 12531bf8a989SDenis Plotnikov virtio_stl_p(vdev, &blkcfg.seg_max, 12541bf8a989SDenis Plotnikov s->conf.seg_max_adjust ? s->conf.queue_size - 2 : 128 - 2); 1255907eb3e5SMichael S. Tsirkin virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls); 1256783d1897SRusty Russell virtio_stl_p(vdev, &blkcfg.blk_size, blk_size); 1257f7516731SMarkus Armbruster virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size); 12586abee260SRoman Kagan virtio_stl_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size); 1259907eb3e5SMichael S. Tsirkin blkcfg.geometry.heads = conf->heads; 12606e790746SPaolo Bonzini /* 12616e790746SPaolo Bonzini * We must ensure that the block device capacity is a multiple of 1262e03ba136SPeter Maydell * the logical block size. If that is not the case, let's use 12636e790746SPaolo Bonzini * sector_mask to adopt the geometry to have a correct picture. 12646e790746SPaolo Bonzini * For those devices where the capacity is ok for the given geometry 1265e03ba136SPeter Maydell * we don't touch the sector value of the geometry, since some devices 12666e790746SPaolo Bonzini * (like s390 dasd) need a specific value. Here the capacity is already 12676e790746SPaolo Bonzini * cyls*heads*secs*blk_size and the sector value is not block size 12686e790746SPaolo Bonzini * divided by 512 - instead it is the amount of blk_size blocks 12696e790746SPaolo Bonzini * per track (cylinder). 12706e790746SPaolo Bonzini */ 127117d0bc01SStefan Hajnoczi length = blk_getlength(s->blk); 127217d0bc01SStefan Hajnoczi if (length > 0 && length / conf->heads / conf->secs % blk_size) { 1273907eb3e5SMichael S. Tsirkin blkcfg.geometry.sectors = conf->secs & ~s->sector_mask; 12746e790746SPaolo Bonzini } else { 1275907eb3e5SMichael S. Tsirkin blkcfg.geometry.sectors = conf->secs; 12766e790746SPaolo Bonzini } 12776e790746SPaolo Bonzini blkcfg.size_max = 0; 1278f7516731SMarkus Armbruster blkcfg.physical_block_exp = get_physical_block_exp(conf); 12796e790746SPaolo Bonzini blkcfg.alignment_offset = 0; 12804be74634SMarkus Armbruster blkcfg.wce = blk_enable_write_cache(s->blk); 12812f270590SStefan Hajnoczi virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues); 128237b06f8dSStefano Garzarella if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) { 1283fb0b154cSAkihiko Odaki uint32_t discard_granularity = conf->discard_granularity; 1284fb0b154cSAkihiko Odaki if (discard_granularity == -1 || !s->conf.report_discard_granularity) { 1285fb0b154cSAkihiko Odaki discard_granularity = blk_size; 1286fb0b154cSAkihiko Odaki } 128737b06f8dSStefano Garzarella virtio_stl_p(vdev, &blkcfg.max_discard_sectors, 128837b06f8dSStefano Garzarella s->conf.max_discard_sectors); 128937b06f8dSStefano Garzarella virtio_stl_p(vdev, &blkcfg.discard_sector_alignment, 1290fb0b154cSAkihiko Odaki discard_granularity >> BDRV_SECTOR_BITS); 129137b06f8dSStefano Garzarella /* 129237b06f8dSStefano Garzarella * We support only one segment per request since multiple segments 129337b06f8dSStefano Garzarella * are not widely used and there are no userspace APIs that allow 129437b06f8dSStefano Garzarella * applications to submit multiple segments in a single call. 129537b06f8dSStefano Garzarella */ 129637b06f8dSStefano Garzarella virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1); 129737b06f8dSStefano Garzarella } 129837b06f8dSStefano Garzarella if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) { 129937b06f8dSStefano Garzarella virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors, 130037b06f8dSStefano Garzarella s->conf.max_write_zeroes_sectors); 130137b06f8dSStefano Garzarella blkcfg.write_zeroes_may_unmap = 1; 130237b06f8dSStefano Garzarella virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1); 130337b06f8dSStefano Garzarella } 13044f736650SSam Li if (bs->bl.zoned != BLK_Z_NONE) { 13054f736650SSam Li switch (bs->bl.zoned) { 13064f736650SSam Li case BLK_Z_HM: 13074f736650SSam Li blkcfg.zoned.model = VIRTIO_BLK_Z_HM; 13084f736650SSam Li break; 13094f736650SSam Li case BLK_Z_HA: 13104f736650SSam Li blkcfg.zoned.model = VIRTIO_BLK_Z_HA; 13114f736650SSam Li break; 13124f736650SSam Li default: 13134f736650SSam Li g_assert_not_reached(); 13144f736650SSam Li } 13154f736650SSam Li 13164f736650SSam Li virtio_stl_p(vdev, &blkcfg.zoned.zone_sectors, 13174f736650SSam Li bs->bl.zone_size / 512); 13184f736650SSam Li virtio_stl_p(vdev, &blkcfg.zoned.max_active_zones, 13194f736650SSam Li bs->bl.max_active_zones); 13204f736650SSam Li virtio_stl_p(vdev, &blkcfg.zoned.max_open_zones, 13214f736650SSam Li bs->bl.max_open_zones); 13224f736650SSam Li virtio_stl_p(vdev, &blkcfg.zoned.write_granularity, blk_size); 13234f736650SSam Li virtio_stl_p(vdev, &blkcfg.zoned.max_append_sectors, 13244f736650SSam Li bs->bl.max_append_sectors); 13254f736650SSam Li } else { 13264f736650SSam Li blkcfg.zoned.model = VIRTIO_BLK_Z_NONE; 13274f736650SSam Li } 132820764be0SStefano Garzarella memcpy(config, &blkcfg, s->config_size); 13296e790746SPaolo Bonzini } 13306e790746SPaolo Bonzini 13316e790746SPaolo Bonzini static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config) 13326e790746SPaolo Bonzini { 13336e790746SPaolo Bonzini VirtIOBlock *s = VIRTIO_BLK(vdev); 13346e790746SPaolo Bonzini struct virtio_blk_config blkcfg; 13356e790746SPaolo Bonzini 133620764be0SStefano Garzarella memcpy(&blkcfg, config, s->config_size); 13376d7e73d6SFam Zheng 13384be74634SMarkus Armbruster blk_set_enable_write_cache(s->blk, blkcfg.wce != 0); 13396e790746SPaolo Bonzini } 13406e790746SPaolo Bonzini 13419d5b731dSJason Wang static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features, 13429d5b731dSJason Wang Error **errp) 13436e790746SPaolo Bonzini { 13446e790746SPaolo Bonzini VirtIOBlock *s = VIRTIO_BLK(vdev); 13456e790746SPaolo Bonzini 1346bbe8bd4dSStefano Garzarella /* Firstly sync all virtio-blk possible supported features */ 1347bbe8bd4dSStefano Garzarella features |= s->host_features; 1348bbe8bd4dSStefano Garzarella 13490cd09c3aSCornelia Huck virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX); 13500cd09c3aSCornelia Huck virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY); 13510cd09c3aSCornelia Huck virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY); 13520cd09c3aSCornelia Huck virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE); 135395129d6fSCornelia Huck if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) { 1354bbe8bd4dSStefano Garzarella if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) { 1355efb8206cSJason Wang error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0"); 1356efb8206cSJason Wang return 0; 1357efb8206cSJason Wang } 1358efb8206cSJason Wang } else { 1359c9b11f97SJason Wang virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT); 1360efb8206cSJason Wang virtio_add_feature(&features, VIRTIO_BLK_F_SCSI); 1361efb8206cSJason Wang } 13626e790746SPaolo Bonzini 13635f258577SEvgeny Yakovlev if (blk_enable_write_cache(s->blk) || 13645f258577SEvgeny Yakovlev (s->conf.x_enable_wce_if_config_wce && 13655f258577SEvgeny Yakovlev virtio_has_feature(features, VIRTIO_BLK_F_CONFIG_WCE))) { 13660cd09c3aSCornelia Huck virtio_add_feature(&features, VIRTIO_BLK_F_WCE); 13674be74634SMarkus Armbruster } 136886b1cf32SKevin Wolf if (!blk_is_writable(s->blk)) { 13690cd09c3aSCornelia Huck virtio_add_feature(&features, VIRTIO_BLK_F_RO); 13704be74634SMarkus Armbruster } 13712f270590SStefan Hajnoczi if (s->conf.num_queues > 1) { 13722f270590SStefan Hajnoczi virtio_add_feature(&features, VIRTIO_BLK_F_MQ); 13732f270590SStefan Hajnoczi } 13746e790746SPaolo Bonzini 13756e790746SPaolo Bonzini return features; 13766e790746SPaolo Bonzini } 13776e790746SPaolo Bonzini 13786e790746SPaolo Bonzini static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status) 13796e790746SPaolo Bonzini { 13806e790746SPaolo Bonzini VirtIOBlock *s = VIRTIO_BLK(vdev); 13816e790746SPaolo Bonzini 13829ffe337cSPaolo Bonzini if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) { 13833cdaf3ddSStefan Hajnoczi assert(!s->ioeventfd_started); 13846e790746SPaolo Bonzini } 13856e790746SPaolo Bonzini 13866e790746SPaolo Bonzini if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) { 13876e790746SPaolo Bonzini return; 13886e790746SPaolo Bonzini } 13896e790746SPaolo Bonzini 1390ef5bc962SPaolo Bonzini /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send 1391ef5bc962SPaolo Bonzini * cache flushes. Thus, the "auto writethrough" behavior is never 1392ef5bc962SPaolo Bonzini * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature. 1393ef5bc962SPaolo Bonzini * Leaving it enabled would break the following sequence: 1394ef5bc962SPaolo Bonzini * 1395ef5bc962SPaolo Bonzini * Guest started with "-drive cache=writethrough" 1396ef5bc962SPaolo Bonzini * Guest sets status to 0 1397ef5bc962SPaolo Bonzini * Guest sets DRIVER bit in status field 1398ef5bc962SPaolo Bonzini * Guest reads host features (WCE=0, CONFIG_WCE=1) 1399ef5bc962SPaolo Bonzini * Guest writes guest features (WCE=0, CONFIG_WCE=1) 1400ef5bc962SPaolo Bonzini * Guest writes 1 to the WCE configuration field (writeback mode) 1401ef5bc962SPaolo Bonzini * Guest sets DRIVER_OK bit in status field 1402ef5bc962SPaolo Bonzini * 14034be74634SMarkus Armbruster * s->blk would erroneously be placed in writethrough mode. 1404ef5bc962SPaolo Bonzini */ 140595129d6fSCornelia Huck if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) { 14064be74634SMarkus Armbruster blk_set_enable_write_cache(s->blk, 140795129d6fSCornelia Huck virtio_vdev_has_feature(vdev, 140895129d6fSCornelia Huck VIRTIO_BLK_F_WCE)); 14096e790746SPaolo Bonzini } 1410ef5bc962SPaolo Bonzini } 14116e790746SPaolo Bonzini 1412b2b295a7SGreg Kurz static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f) 1413b2b295a7SGreg Kurz { 1414b2b295a7SGreg Kurz VirtIOBlock *s = VIRTIO_BLK(vdev); 14159c67f33fSStefan Hajnoczi 14169c67f33fSStefan Hajnoczi WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 1417b2b295a7SGreg Kurz VirtIOBlockReq *req = s->rq; 14186e790746SPaolo Bonzini 14196e790746SPaolo Bonzini while (req) { 14206e790746SPaolo Bonzini qemu_put_sbyte(f, 1); 142130d8bf6dSStefan Hajnoczi 142230d8bf6dSStefan Hajnoczi if (s->conf.num_queues > 1) { 142330d8bf6dSStefan Hajnoczi qemu_put_be32(f, virtio_get_queue_index(req->vq)); 142430d8bf6dSStefan Hajnoczi } 142530d8bf6dSStefan Hajnoczi 142686044b24SJason Wang qemu_put_virtqueue_element(vdev, f, &req->elem); 14276e790746SPaolo Bonzini req = req->next; 14286e790746SPaolo Bonzini } 14299c67f33fSStefan Hajnoczi } 14309c67f33fSStefan Hajnoczi 14316e790746SPaolo Bonzini qemu_put_sbyte(f, 0); 14326e790746SPaolo Bonzini } 14336e790746SPaolo Bonzini 1434b2b295a7SGreg Kurz static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f, 1435b2b295a7SGreg Kurz int version_id) 1436b2b295a7SGreg Kurz { 1437b2b295a7SGreg Kurz VirtIOBlock *s = VIRTIO_BLK(vdev); 1438b2b295a7SGreg Kurz 14396e790746SPaolo Bonzini while (qemu_get_sbyte(f)) { 144030d8bf6dSStefan Hajnoczi unsigned nvqs = s->conf.num_queues; 144130d8bf6dSStefan Hajnoczi unsigned vq_idx = 0; 1442ab281c17SPaolo Bonzini VirtIOBlockReq *req; 144330d8bf6dSStefan Hajnoczi 144430d8bf6dSStefan Hajnoczi if (nvqs > 1) { 144530d8bf6dSStefan Hajnoczi vq_idx = qemu_get_be32(f); 144630d8bf6dSStefan Hajnoczi 144730d8bf6dSStefan Hajnoczi if (vq_idx >= nvqs) { 144830d8bf6dSStefan Hajnoczi error_report("Invalid virtqueue index in request list: %#x", 144930d8bf6dSStefan Hajnoczi vq_idx); 145030d8bf6dSStefan Hajnoczi return -EINVAL; 145130d8bf6dSStefan Hajnoczi } 145230d8bf6dSStefan Hajnoczi } 145330d8bf6dSStefan Hajnoczi 14548607f5c3SJason Wang req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq)); 145530d8bf6dSStefan Hajnoczi virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req); 14569c67f33fSStefan Hajnoczi 14579c67f33fSStefan Hajnoczi WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 14586e790746SPaolo Bonzini req->next = s->rq; 14596e790746SPaolo Bonzini s->rq = req; 14606e790746SPaolo Bonzini } 14619c67f33fSStefan Hajnoczi } 14626e790746SPaolo Bonzini 14636e790746SPaolo Bonzini return 0; 14646e790746SPaolo Bonzini } 14656e790746SPaolo Bonzini 1466b6948ab0SStefan Hajnoczi static bool 1467b6948ab0SStefan Hajnoczi validate_iothread_vq_mapping_list(IOThreadVirtQueueMappingList *list, 1468b6948ab0SStefan Hajnoczi uint16_t num_queues, Error **errp) 1469b6948ab0SStefan Hajnoczi { 1470b6948ab0SStefan Hajnoczi g_autofree unsigned long *vqs = bitmap_new(num_queues); 1471b6948ab0SStefan Hajnoczi g_autoptr(GHashTable) iothreads = 1472b6948ab0SStefan Hajnoczi g_hash_table_new(g_str_hash, g_str_equal); 1473b6948ab0SStefan Hajnoczi 1474b6948ab0SStefan Hajnoczi for (IOThreadVirtQueueMappingList *node = list; node; node = node->next) { 1475b6948ab0SStefan Hajnoczi const char *name = node->value->iothread; 1476b6948ab0SStefan Hajnoczi uint16List *vq; 1477b6948ab0SStefan Hajnoczi 1478b6948ab0SStefan Hajnoczi if (!iothread_by_id(name)) { 1479b6948ab0SStefan Hajnoczi error_setg(errp, "IOThread \"%s\" object does not exist", name); 1480b6948ab0SStefan Hajnoczi return false; 1481b6948ab0SStefan Hajnoczi } 1482b6948ab0SStefan Hajnoczi 1483b6948ab0SStefan Hajnoczi if (!g_hash_table_add(iothreads, (gpointer)name)) { 1484b6948ab0SStefan Hajnoczi error_setg(errp, 1485b6948ab0SStefan Hajnoczi "duplicate IOThread name \"%s\" in iothread-vq-mapping", 1486b6948ab0SStefan Hajnoczi name); 1487b6948ab0SStefan Hajnoczi return false; 1488b6948ab0SStefan Hajnoczi } 1489b6948ab0SStefan Hajnoczi 1490b6948ab0SStefan Hajnoczi if (node != list) { 1491b6948ab0SStefan Hajnoczi if (!!node->value->vqs != !!list->value->vqs) { 1492b6948ab0SStefan Hajnoczi error_setg(errp, "either all items in iothread-vq-mapping " 1493b6948ab0SStefan Hajnoczi "must have vqs or none of them must have it"); 1494b6948ab0SStefan Hajnoczi return false; 1495b6948ab0SStefan Hajnoczi } 1496b6948ab0SStefan Hajnoczi } 1497b6948ab0SStefan Hajnoczi 1498b6948ab0SStefan Hajnoczi for (vq = node->value->vqs; vq; vq = vq->next) { 1499b6948ab0SStefan Hajnoczi if (vq->value >= num_queues) { 1500b6948ab0SStefan Hajnoczi error_setg(errp, "vq index %u for IOThread \"%s\" must be " 1501b6948ab0SStefan Hajnoczi "less than num_queues %u in iothread-vq-mapping", 1502b6948ab0SStefan Hajnoczi vq->value, name, num_queues); 1503b6948ab0SStefan Hajnoczi return false; 1504b6948ab0SStefan Hajnoczi } 1505b6948ab0SStefan Hajnoczi 1506b6948ab0SStefan Hajnoczi if (test_and_set_bit(vq->value, vqs)) { 1507b6948ab0SStefan Hajnoczi error_setg(errp, "cannot assign vq %u to IOThread \"%s\" " 1508b6948ab0SStefan Hajnoczi "because it is already assigned", vq->value, name); 1509b6948ab0SStefan Hajnoczi return false; 1510b6948ab0SStefan Hajnoczi } 1511b6948ab0SStefan Hajnoczi } 1512b6948ab0SStefan Hajnoczi } 1513b6948ab0SStefan Hajnoczi 1514b6948ab0SStefan Hajnoczi if (list->value->vqs) { 1515b6948ab0SStefan Hajnoczi for (uint16_t i = 0; i < num_queues; i++) { 1516b6948ab0SStefan Hajnoczi if (!test_bit(i, vqs)) { 1517b6948ab0SStefan Hajnoczi error_setg(errp, 1518b6948ab0SStefan Hajnoczi "missing vq %u IOThread assignment in iothread-vq-mapping", 1519b6948ab0SStefan Hajnoczi i); 1520b6948ab0SStefan Hajnoczi return false; 1521b6948ab0SStefan Hajnoczi } 1522b6948ab0SStefan Hajnoczi } 1523b6948ab0SStefan Hajnoczi } 1524b6948ab0SStefan Hajnoczi 1525b6948ab0SStefan Hajnoczi return true; 1526b6948ab0SStefan Hajnoczi } 1527b6948ab0SStefan Hajnoczi 15289b92fbcfSSergio Lopez static void virtio_resize_cb(void *opaque) 15299b92fbcfSSergio Lopez { 15309b92fbcfSSergio Lopez VirtIODevice *vdev = opaque; 15319b92fbcfSSergio Lopez 15329b92fbcfSSergio Lopez assert(qemu_get_current_aio_context() == qemu_get_aio_context()); 15339b92fbcfSSergio Lopez virtio_notify_config(vdev); 15349b92fbcfSSergio Lopez } 15359b92fbcfSSergio Lopez 15366e790746SPaolo Bonzini static void virtio_blk_resize(void *opaque) 15376e790746SPaolo Bonzini { 15386e790746SPaolo Bonzini VirtIODevice *vdev = VIRTIO_DEVICE(opaque); 15396e790746SPaolo Bonzini 15409b92fbcfSSergio Lopez /* 15410b2675c4SStefan Hajnoczi * virtio_notify_config() needs to acquire the BQL, 15429b92fbcfSSergio Lopez * so it can't be called from an iothread. Instead, schedule 15439b92fbcfSSergio Lopez * it to be run in the main context BH. 15449b92fbcfSSergio Lopez */ 15459b92fbcfSSergio Lopez aio_bh_schedule_oneshot(qemu_get_aio_context(), virtio_resize_cb, vdev); 15466e790746SPaolo Bonzini } 15476e790746SPaolo Bonzini 15483cdaf3ddSStefan Hajnoczi static void virtio_blk_ioeventfd_detach(VirtIOBlock *s) 15493bcc17f0SStefan Hajnoczi { 15503bcc17f0SStefan Hajnoczi VirtIODevice *vdev = VIRTIO_DEVICE(s); 15513bcc17f0SStefan Hajnoczi 15523bcc17f0SStefan Hajnoczi for (uint16_t i = 0; i < s->conf.num_queues; i++) { 15533bcc17f0SStefan Hajnoczi VirtQueue *vq = virtio_get_queue(vdev, i); 15543bcc17f0SStefan Hajnoczi virtio_queue_aio_detach_host_notifier(vq, s->vq_aio_context[i]); 15553bcc17f0SStefan Hajnoczi } 15563bcc17f0SStefan Hajnoczi } 15573bcc17f0SStefan Hajnoczi 15583cdaf3ddSStefan Hajnoczi static void virtio_blk_ioeventfd_attach(VirtIOBlock *s) 15593bcc17f0SStefan Hajnoczi { 15603bcc17f0SStefan Hajnoczi VirtIODevice *vdev = VIRTIO_DEVICE(s); 15613bcc17f0SStefan Hajnoczi 15623bcc17f0SStefan Hajnoczi for (uint16_t i = 0; i < s->conf.num_queues; i++) { 15633bcc17f0SStefan Hajnoczi VirtQueue *vq = virtio_get_queue(vdev, i); 15643bcc17f0SStefan Hajnoczi virtio_queue_aio_attach_host_notifier(vq, s->vq_aio_context[i]); 15653bcc17f0SStefan Hajnoczi } 15663bcc17f0SStefan Hajnoczi } 15673bcc17f0SStefan Hajnoczi 15681665d932SStefan Hajnoczi /* Suspend virtqueue ioeventfd processing during drain */ 15691665d932SStefan Hajnoczi static void virtio_blk_drained_begin(void *opaque) 15701665d932SStefan Hajnoczi { 15711665d932SStefan Hajnoczi VirtIOBlock *s = opaque; 15721665d932SStefan Hajnoczi 15733cdaf3ddSStefan Hajnoczi if (s->ioeventfd_started) { 15743cdaf3ddSStefan Hajnoczi virtio_blk_ioeventfd_detach(s); 15751665d932SStefan Hajnoczi } 15761665d932SStefan Hajnoczi } 15771665d932SStefan Hajnoczi 15781665d932SStefan Hajnoczi /* Resume virtqueue ioeventfd processing after drain */ 15791665d932SStefan Hajnoczi static void virtio_blk_drained_end(void *opaque) 15801665d932SStefan Hajnoczi { 15811665d932SStefan Hajnoczi VirtIOBlock *s = opaque; 15821665d932SStefan Hajnoczi 15833cdaf3ddSStefan Hajnoczi if (s->ioeventfd_started) { 15843cdaf3ddSStefan Hajnoczi virtio_blk_ioeventfd_attach(s); 15851665d932SStefan Hajnoczi } 15861665d932SStefan Hajnoczi } 15871665d932SStefan Hajnoczi 15886e790746SPaolo Bonzini static const BlockDevOps virtio_block_ops = { 15896e790746SPaolo Bonzini .resize_cb = virtio_blk_resize, 15901665d932SStefan Hajnoczi .drained_begin = virtio_blk_drained_begin, 15911665d932SStefan Hajnoczi .drained_end = virtio_blk_drained_end, 15926e790746SPaolo Bonzini }; 15936e790746SPaolo Bonzini 15943bcc17f0SStefan Hajnoczi /* Generate vq:AioContext mappings from a validated iothread-vq-mapping list */ 15953bcc17f0SStefan Hajnoczi static void 15963bcc17f0SStefan Hajnoczi apply_vq_mapping(IOThreadVirtQueueMappingList *iothread_vq_mapping_list, 15973bcc17f0SStefan Hajnoczi AioContext **vq_aio_context, uint16_t num_queues) 15983bcc17f0SStefan Hajnoczi { 15993bcc17f0SStefan Hajnoczi IOThreadVirtQueueMappingList *node; 16003bcc17f0SStefan Hajnoczi size_t num_iothreads = 0; 16013bcc17f0SStefan Hajnoczi size_t cur_iothread = 0; 16023bcc17f0SStefan Hajnoczi 16033bcc17f0SStefan Hajnoczi for (node = iothread_vq_mapping_list; node; node = node->next) { 16043bcc17f0SStefan Hajnoczi num_iothreads++; 16053bcc17f0SStefan Hajnoczi } 16063bcc17f0SStefan Hajnoczi 16073bcc17f0SStefan Hajnoczi for (node = iothread_vq_mapping_list; node; node = node->next) { 16083bcc17f0SStefan Hajnoczi IOThread *iothread = iothread_by_id(node->value->iothread); 16093bcc17f0SStefan Hajnoczi AioContext *ctx = iothread_get_aio_context(iothread); 16103bcc17f0SStefan Hajnoczi 161157bc2658SStefan Hajnoczi /* Released in virtio_blk_vq_aio_context_cleanup() */ 16123bcc17f0SStefan Hajnoczi object_ref(OBJECT(iothread)); 16133bcc17f0SStefan Hajnoczi 16143bcc17f0SStefan Hajnoczi if (node->value->vqs) { 16153bcc17f0SStefan Hajnoczi uint16List *vq; 16163bcc17f0SStefan Hajnoczi 16173bcc17f0SStefan Hajnoczi /* Explicit vq:IOThread assignment */ 16183bcc17f0SStefan Hajnoczi for (vq = node->value->vqs; vq; vq = vq->next) { 16193bcc17f0SStefan Hajnoczi vq_aio_context[vq->value] = ctx; 16203bcc17f0SStefan Hajnoczi } 16213bcc17f0SStefan Hajnoczi } else { 16223bcc17f0SStefan Hajnoczi /* Round-robin vq:IOThread assignment */ 16233bcc17f0SStefan Hajnoczi for (unsigned i = cur_iothread; i < num_queues; 16243bcc17f0SStefan Hajnoczi i += num_iothreads) { 16253bcc17f0SStefan Hajnoczi vq_aio_context[i] = ctx; 16263bcc17f0SStefan Hajnoczi } 16273bcc17f0SStefan Hajnoczi } 16283bcc17f0SStefan Hajnoczi 16293bcc17f0SStefan Hajnoczi cur_iothread++; 16303bcc17f0SStefan Hajnoczi } 16313bcc17f0SStefan Hajnoczi } 16323bcc17f0SStefan Hajnoczi 16333bcc17f0SStefan Hajnoczi /* Context: BQL held */ 163457bc2658SStefan Hajnoczi static bool virtio_blk_vq_aio_context_init(VirtIOBlock *s, Error **errp) 16353bcc17f0SStefan Hajnoczi { 16363bcc17f0SStefan Hajnoczi VirtIODevice *vdev = VIRTIO_DEVICE(s); 16373bcc17f0SStefan Hajnoczi VirtIOBlkConf *conf = &s->conf; 16383bcc17f0SStefan Hajnoczi BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev))); 16393bcc17f0SStefan Hajnoczi VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 16403bcc17f0SStefan Hajnoczi 16413bcc17f0SStefan Hajnoczi if (conf->iothread || conf->iothread_vq_mapping_list) { 16423bcc17f0SStefan Hajnoczi if (!k->set_guest_notifiers || !k->ioeventfd_assign) { 16433bcc17f0SStefan Hajnoczi error_setg(errp, 16443bcc17f0SStefan Hajnoczi "device is incompatible with iothread " 16453bcc17f0SStefan Hajnoczi "(transport does not support notifiers)"); 16463bcc17f0SStefan Hajnoczi return false; 16473bcc17f0SStefan Hajnoczi } 16483bcc17f0SStefan Hajnoczi if (!virtio_device_ioeventfd_enabled(vdev)) { 16493bcc17f0SStefan Hajnoczi error_setg(errp, "ioeventfd is required for iothread"); 16503bcc17f0SStefan Hajnoczi return false; 16513bcc17f0SStefan Hajnoczi } 16523bcc17f0SStefan Hajnoczi 16533bcc17f0SStefan Hajnoczi /* 16543cdaf3ddSStefan Hajnoczi * If ioeventfd is (re-)enabled while the guest is running there could 16553bcc17f0SStefan Hajnoczi * be block jobs that can conflict. 16563bcc17f0SStefan Hajnoczi */ 16573bcc17f0SStefan Hajnoczi if (blk_op_is_blocked(conf->conf.blk, BLOCK_OP_TYPE_DATAPLANE, errp)) { 16583cdaf3ddSStefan Hajnoczi error_prepend(errp, "cannot start virtio-blk ioeventfd: "); 16593bcc17f0SStefan Hajnoczi return false; 16603bcc17f0SStefan Hajnoczi } 16613bcc17f0SStefan Hajnoczi } 16623bcc17f0SStefan Hajnoczi 16633bcc17f0SStefan Hajnoczi s->vq_aio_context = g_new(AioContext *, conf->num_queues); 16643bcc17f0SStefan Hajnoczi 16653bcc17f0SStefan Hajnoczi if (conf->iothread_vq_mapping_list) { 16663bcc17f0SStefan Hajnoczi apply_vq_mapping(conf->iothread_vq_mapping_list, s->vq_aio_context, 16673bcc17f0SStefan Hajnoczi conf->num_queues); 16683bcc17f0SStefan Hajnoczi } else if (conf->iothread) { 16693bcc17f0SStefan Hajnoczi AioContext *ctx = iothread_get_aio_context(conf->iothread); 16703bcc17f0SStefan Hajnoczi for (unsigned i = 0; i < conf->num_queues; i++) { 16713bcc17f0SStefan Hajnoczi s->vq_aio_context[i] = ctx; 16723bcc17f0SStefan Hajnoczi } 16733bcc17f0SStefan Hajnoczi 167457bc2658SStefan Hajnoczi /* Released in virtio_blk_vq_aio_context_cleanup() */ 16753bcc17f0SStefan Hajnoczi object_ref(OBJECT(conf->iothread)); 16763bcc17f0SStefan Hajnoczi } else { 16773bcc17f0SStefan Hajnoczi AioContext *ctx = qemu_get_aio_context(); 16783bcc17f0SStefan Hajnoczi for (unsigned i = 0; i < conf->num_queues; i++) { 16793bcc17f0SStefan Hajnoczi s->vq_aio_context[i] = ctx; 16803bcc17f0SStefan Hajnoczi } 16813bcc17f0SStefan Hajnoczi } 16823bcc17f0SStefan Hajnoczi 16833bcc17f0SStefan Hajnoczi return true; 16843bcc17f0SStefan Hajnoczi } 16853bcc17f0SStefan Hajnoczi 16863bcc17f0SStefan Hajnoczi /* Context: BQL held */ 168757bc2658SStefan Hajnoczi static void virtio_blk_vq_aio_context_cleanup(VirtIOBlock *s) 16883bcc17f0SStefan Hajnoczi { 16893bcc17f0SStefan Hajnoczi VirtIOBlkConf *conf = &s->conf; 16903bcc17f0SStefan Hajnoczi 16913cdaf3ddSStefan Hajnoczi assert(!s->ioeventfd_started); 16923bcc17f0SStefan Hajnoczi 16933bcc17f0SStefan Hajnoczi if (conf->iothread_vq_mapping_list) { 16943bcc17f0SStefan Hajnoczi IOThreadVirtQueueMappingList *node; 16953bcc17f0SStefan Hajnoczi 16963bcc17f0SStefan Hajnoczi for (node = conf->iothread_vq_mapping_list; node; node = node->next) { 16973bcc17f0SStefan Hajnoczi IOThread *iothread = iothread_by_id(node->value->iothread); 16983bcc17f0SStefan Hajnoczi object_unref(OBJECT(iothread)); 16993bcc17f0SStefan Hajnoczi } 17003bcc17f0SStefan Hajnoczi } 17013bcc17f0SStefan Hajnoczi 17023bcc17f0SStefan Hajnoczi if (conf->iothread) { 17033bcc17f0SStefan Hajnoczi object_unref(OBJECT(conf->iothread)); 17043bcc17f0SStefan Hajnoczi } 17053bcc17f0SStefan Hajnoczi 17063bcc17f0SStefan Hajnoczi g_free(s->vq_aio_context); 17073bcc17f0SStefan Hajnoczi s->vq_aio_context = NULL; 17083bcc17f0SStefan Hajnoczi } 17093bcc17f0SStefan Hajnoczi 17103bcc17f0SStefan Hajnoczi /* Context: BQL held */ 17113cdaf3ddSStefan Hajnoczi static int virtio_blk_start_ioeventfd(VirtIODevice *vdev) 17123bcc17f0SStefan Hajnoczi { 17133bcc17f0SStefan Hajnoczi VirtIOBlock *s = VIRTIO_BLK(vdev); 17143bcc17f0SStefan Hajnoczi BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(s))); 17153bcc17f0SStefan Hajnoczi VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 17163bcc17f0SStefan Hajnoczi unsigned i; 17173bcc17f0SStefan Hajnoczi unsigned nvqs = s->conf.num_queues; 17183bcc17f0SStefan Hajnoczi Error *local_err = NULL; 17193bcc17f0SStefan Hajnoczi int r; 17203bcc17f0SStefan Hajnoczi 17213cdaf3ddSStefan Hajnoczi if (s->ioeventfd_started || s->ioeventfd_starting) { 17223bcc17f0SStefan Hajnoczi return 0; 17233bcc17f0SStefan Hajnoczi } 17243bcc17f0SStefan Hajnoczi 17253cdaf3ddSStefan Hajnoczi s->ioeventfd_starting = true; 17263bcc17f0SStefan Hajnoczi 17273bcc17f0SStefan Hajnoczi /* Set up guest notifier (irq) */ 17283bcc17f0SStefan Hajnoczi r = k->set_guest_notifiers(qbus->parent, nvqs, true); 17293bcc17f0SStefan Hajnoczi if (r != 0) { 17303bcc17f0SStefan Hajnoczi error_report("virtio-blk failed to set guest notifier (%d), " 17313bcc17f0SStefan Hajnoczi "ensure -accel kvm is set.", r); 17323bcc17f0SStefan Hajnoczi goto fail_guest_notifiers; 17333bcc17f0SStefan Hajnoczi } 17343bcc17f0SStefan Hajnoczi 17353bcc17f0SStefan Hajnoczi /* 17363bcc17f0SStefan Hajnoczi * Batch all the host notifiers in a single transaction to avoid 17373bcc17f0SStefan Hajnoczi * quadratic time complexity in address_space_update_ioeventfds(). 17383bcc17f0SStefan Hajnoczi */ 17393bcc17f0SStefan Hajnoczi memory_region_transaction_begin(); 17403bcc17f0SStefan Hajnoczi 17413bcc17f0SStefan Hajnoczi /* Set up virtqueue notify */ 17423bcc17f0SStefan Hajnoczi for (i = 0; i < nvqs; i++) { 17433bcc17f0SStefan Hajnoczi r = virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), i, true); 17443bcc17f0SStefan Hajnoczi if (r != 0) { 17453bcc17f0SStefan Hajnoczi int j = i; 17463bcc17f0SStefan Hajnoczi 17473bcc17f0SStefan Hajnoczi fprintf(stderr, "virtio-blk failed to set host notifier (%d)\n", r); 17483bcc17f0SStefan Hajnoczi while (i--) { 17493bcc17f0SStefan Hajnoczi virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), i, false); 17503bcc17f0SStefan Hajnoczi } 17513bcc17f0SStefan Hajnoczi 17523bcc17f0SStefan Hajnoczi /* 17533bcc17f0SStefan Hajnoczi * The transaction expects the ioeventfds to be open when it 17543bcc17f0SStefan Hajnoczi * commits. Do it now, before the cleanup loop. 17553bcc17f0SStefan Hajnoczi */ 17563bcc17f0SStefan Hajnoczi memory_region_transaction_commit(); 17573bcc17f0SStefan Hajnoczi 17583bcc17f0SStefan Hajnoczi while (j--) { 17593bcc17f0SStefan Hajnoczi virtio_bus_cleanup_host_notifier(VIRTIO_BUS(qbus), j); 17603bcc17f0SStefan Hajnoczi } 17613bcc17f0SStefan Hajnoczi goto fail_host_notifiers; 17623bcc17f0SStefan Hajnoczi } 17633bcc17f0SStefan Hajnoczi } 17643bcc17f0SStefan Hajnoczi 17653bcc17f0SStefan Hajnoczi memory_region_transaction_commit(); 17663bcc17f0SStefan Hajnoczi 17673bcc17f0SStefan Hajnoczi r = blk_set_aio_context(s->conf.conf.blk, s->vq_aio_context[0], 17683bcc17f0SStefan Hajnoczi &local_err); 17693bcc17f0SStefan Hajnoczi if (r < 0) { 17703bcc17f0SStefan Hajnoczi error_report_err(local_err); 17713bcc17f0SStefan Hajnoczi goto fail_aio_context; 17723bcc17f0SStefan Hajnoczi } 17733bcc17f0SStefan Hajnoczi 17743bcc17f0SStefan Hajnoczi /* 17753bcc17f0SStefan Hajnoczi * These fields must be visible to the IOThread when it processes the 17763cdaf3ddSStefan Hajnoczi * virtqueue, otherwise it will think ioeventfd has not started yet. 17773bcc17f0SStefan Hajnoczi * 17783cdaf3ddSStefan Hajnoczi * Make sure ->ioeventfd_started is false when blk_set_aio_context() is 17793bcc17f0SStefan Hajnoczi * called above so that draining does not cause the host notifier to be 17803bcc17f0SStefan Hajnoczi * detached/attached prematurely. 17813bcc17f0SStefan Hajnoczi */ 17823cdaf3ddSStefan Hajnoczi s->ioeventfd_starting = false; 17833cdaf3ddSStefan Hajnoczi s->ioeventfd_started = true; 17843bcc17f0SStefan Hajnoczi smp_wmb(); /* paired with aio_notify_accept() on the read side */ 17853bcc17f0SStefan Hajnoczi 17863bcc17f0SStefan Hajnoczi /* Get this show started by hooking up our callbacks */ 17873bcc17f0SStefan Hajnoczi if (!blk_in_drain(s->conf.conf.blk)) { 17883bcc17f0SStefan Hajnoczi for (i = 0; i < nvqs; i++) { 17893bcc17f0SStefan Hajnoczi VirtQueue *vq = virtio_get_queue(vdev, i); 17903bcc17f0SStefan Hajnoczi AioContext *ctx = s->vq_aio_context[i]; 17913bcc17f0SStefan Hajnoczi 17923bcc17f0SStefan Hajnoczi /* Kick right away to begin processing requests already in vring */ 17933bcc17f0SStefan Hajnoczi event_notifier_set(virtio_queue_get_host_notifier(vq)); 17943bcc17f0SStefan Hajnoczi 17953bcc17f0SStefan Hajnoczi virtio_queue_aio_attach_host_notifier(vq, ctx); 17963bcc17f0SStefan Hajnoczi } 17973bcc17f0SStefan Hajnoczi } 17983bcc17f0SStefan Hajnoczi return 0; 17993bcc17f0SStefan Hajnoczi 18003bcc17f0SStefan Hajnoczi fail_aio_context: 18013bcc17f0SStefan Hajnoczi memory_region_transaction_begin(); 18023bcc17f0SStefan Hajnoczi 18033bcc17f0SStefan Hajnoczi for (i = 0; i < nvqs; i++) { 18043bcc17f0SStefan Hajnoczi virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), i, false); 18053bcc17f0SStefan Hajnoczi } 18063bcc17f0SStefan Hajnoczi 18073bcc17f0SStefan Hajnoczi memory_region_transaction_commit(); 18083bcc17f0SStefan Hajnoczi 18093bcc17f0SStefan Hajnoczi for (i = 0; i < nvqs; i++) { 18103bcc17f0SStefan Hajnoczi virtio_bus_cleanup_host_notifier(VIRTIO_BUS(qbus), i); 18113bcc17f0SStefan Hajnoczi } 18123bcc17f0SStefan Hajnoczi fail_host_notifiers: 18133bcc17f0SStefan Hajnoczi k->set_guest_notifiers(qbus->parent, nvqs, false); 18143bcc17f0SStefan Hajnoczi fail_guest_notifiers: 18153cdaf3ddSStefan Hajnoczi s->ioeventfd_disabled = true; 18163cdaf3ddSStefan Hajnoczi s->ioeventfd_starting = false; 18173bcc17f0SStefan Hajnoczi return -ENOSYS; 18183bcc17f0SStefan Hajnoczi } 18193bcc17f0SStefan Hajnoczi 18203bcc17f0SStefan Hajnoczi /* Stop notifications for new requests from guest. 18213bcc17f0SStefan Hajnoczi * 18223bcc17f0SStefan Hajnoczi * Context: BH in IOThread 18233bcc17f0SStefan Hajnoczi */ 18243cdaf3ddSStefan Hajnoczi static void virtio_blk_ioeventfd_stop_vq_bh(void *opaque) 18253bcc17f0SStefan Hajnoczi { 18263bcc17f0SStefan Hajnoczi VirtQueue *vq = opaque; 18273bcc17f0SStefan Hajnoczi EventNotifier *host_notifier = virtio_queue_get_host_notifier(vq); 18283bcc17f0SStefan Hajnoczi 18293bcc17f0SStefan Hajnoczi virtio_queue_aio_detach_host_notifier(vq, qemu_get_current_aio_context()); 18303bcc17f0SStefan Hajnoczi 18313bcc17f0SStefan Hajnoczi /* 18323bcc17f0SStefan Hajnoczi * Test and clear notifier after disabling event, in case poll callback 18333bcc17f0SStefan Hajnoczi * didn't have time to run. 18343bcc17f0SStefan Hajnoczi */ 18353bcc17f0SStefan Hajnoczi virtio_queue_host_notifier_read(host_notifier); 18363bcc17f0SStefan Hajnoczi } 18373bcc17f0SStefan Hajnoczi 18383bcc17f0SStefan Hajnoczi /* Context: BQL held */ 18393cdaf3ddSStefan Hajnoczi static void virtio_blk_stop_ioeventfd(VirtIODevice *vdev) 18403bcc17f0SStefan Hajnoczi { 18413bcc17f0SStefan Hajnoczi VirtIOBlock *s = VIRTIO_BLK(vdev); 18423bcc17f0SStefan Hajnoczi BusState *qbus = qdev_get_parent_bus(DEVICE(s)); 18433bcc17f0SStefan Hajnoczi VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 18443bcc17f0SStefan Hajnoczi unsigned i; 18453bcc17f0SStefan Hajnoczi unsigned nvqs = s->conf.num_queues; 18463bcc17f0SStefan Hajnoczi 18473cdaf3ddSStefan Hajnoczi if (!s->ioeventfd_started || s->ioeventfd_stopping) { 18483bcc17f0SStefan Hajnoczi return; 18493bcc17f0SStefan Hajnoczi } 18503bcc17f0SStefan Hajnoczi 18513bcc17f0SStefan Hajnoczi /* Better luck next time. */ 18523cdaf3ddSStefan Hajnoczi if (s->ioeventfd_disabled) { 18533cdaf3ddSStefan Hajnoczi s->ioeventfd_disabled = false; 18543cdaf3ddSStefan Hajnoczi s->ioeventfd_started = false; 18553bcc17f0SStefan Hajnoczi return; 18563bcc17f0SStefan Hajnoczi } 18573cdaf3ddSStefan Hajnoczi s->ioeventfd_stopping = true; 18583bcc17f0SStefan Hajnoczi 18593bcc17f0SStefan Hajnoczi if (!blk_in_drain(s->conf.conf.blk)) { 18603bcc17f0SStefan Hajnoczi for (i = 0; i < nvqs; i++) { 18613bcc17f0SStefan Hajnoczi VirtQueue *vq = virtio_get_queue(vdev, i); 18623bcc17f0SStefan Hajnoczi AioContext *ctx = s->vq_aio_context[i]; 18633bcc17f0SStefan Hajnoczi 18643cdaf3ddSStefan Hajnoczi aio_wait_bh_oneshot(ctx, virtio_blk_ioeventfd_stop_vq_bh, vq); 18653bcc17f0SStefan Hajnoczi } 18663bcc17f0SStefan Hajnoczi } 18673bcc17f0SStefan Hajnoczi 18683bcc17f0SStefan Hajnoczi /* 18693bcc17f0SStefan Hajnoczi * Batch all the host notifiers in a single transaction to avoid 18703bcc17f0SStefan Hajnoczi * quadratic time complexity in address_space_update_ioeventfds(). 18713bcc17f0SStefan Hajnoczi */ 18723bcc17f0SStefan Hajnoczi memory_region_transaction_begin(); 18733bcc17f0SStefan Hajnoczi 18743bcc17f0SStefan Hajnoczi for (i = 0; i < nvqs; i++) { 18753bcc17f0SStefan Hajnoczi virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), i, false); 18763bcc17f0SStefan Hajnoczi } 18773bcc17f0SStefan Hajnoczi 18783bcc17f0SStefan Hajnoczi /* 18793bcc17f0SStefan Hajnoczi * The transaction expects the ioeventfds to be open when it 18803bcc17f0SStefan Hajnoczi * commits. Do it now, before the cleanup loop. 18813bcc17f0SStefan Hajnoczi */ 18823bcc17f0SStefan Hajnoczi memory_region_transaction_commit(); 18833bcc17f0SStefan Hajnoczi 18843bcc17f0SStefan Hajnoczi for (i = 0; i < nvqs; i++) { 18853bcc17f0SStefan Hajnoczi virtio_bus_cleanup_host_notifier(VIRTIO_BUS(qbus), i); 18863bcc17f0SStefan Hajnoczi } 18873bcc17f0SStefan Hajnoczi 18883bcc17f0SStefan Hajnoczi /* 18893cdaf3ddSStefan Hajnoczi * Set ->ioeventfd_started to false before draining so that host notifiers 18903bcc17f0SStefan Hajnoczi * are not detached/attached anymore. 18913bcc17f0SStefan Hajnoczi */ 18923cdaf3ddSStefan Hajnoczi s->ioeventfd_started = false; 18933bcc17f0SStefan Hajnoczi 18943bcc17f0SStefan Hajnoczi /* Wait for virtio_blk_dma_restart_bh() and in flight I/O to complete */ 18953bcc17f0SStefan Hajnoczi blk_drain(s->conf.conf.blk); 18963bcc17f0SStefan Hajnoczi 18973bcc17f0SStefan Hajnoczi /* 18983bcc17f0SStefan Hajnoczi * Try to switch bs back to the QEMU main loop. If other users keep the 18993bcc17f0SStefan Hajnoczi * BlockBackend in the iothread, that's ok 19003bcc17f0SStefan Hajnoczi */ 19013bcc17f0SStefan Hajnoczi blk_set_aio_context(s->conf.conf.blk, qemu_get_aio_context(), NULL); 19023bcc17f0SStefan Hajnoczi 19033bcc17f0SStefan Hajnoczi /* Clean up guest notifier (irq) */ 19043bcc17f0SStefan Hajnoczi k->set_guest_notifiers(qbus->parent, nvqs, false); 19053bcc17f0SStefan Hajnoczi 19063cdaf3ddSStefan Hajnoczi s->ioeventfd_stopping = false; 19073bcc17f0SStefan Hajnoczi } 19083bcc17f0SStefan Hajnoczi 190975884afdSAndreas Färber static void virtio_blk_device_realize(DeviceState *dev, Error **errp) 19106e790746SPaolo Bonzini { 191175884afdSAndreas Färber VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1912179b417eSAndreas Färber VirtIOBlock *s = VIRTIO_BLK(dev); 19132a30307fSMarkus Armbruster VirtIOBlkConf *conf = &s->conf; 19143ffeeef7SAndreas Färber Error *err = NULL; 19152f270590SStefan Hajnoczi unsigned i; 19166e790746SPaolo Bonzini 19174be74634SMarkus Armbruster if (!conf->conf.blk) { 191875884afdSAndreas Färber error_setg(errp, "drive property not set"); 191975884afdSAndreas Färber return; 19206e790746SPaolo Bonzini } 19214be74634SMarkus Armbruster if (!blk_is_inserted(conf->conf.blk)) { 192275884afdSAndreas Färber error_setg(errp, "Device needs media, but drive is empty"); 192375884afdSAndreas Färber return; 19246e790746SPaolo Bonzini } 19259445e1e1SStefan Hajnoczi if (conf->num_queues == VIRTIO_BLK_AUTO_NUM_QUEUES) { 19269445e1e1SStefan Hajnoczi conf->num_queues = 1; 19279445e1e1SStefan Hajnoczi } 19282f270590SStefan Hajnoczi if (!conf->num_queues) { 19292f270590SStefan Hajnoczi error_setg(errp, "num-queues property must be larger than 0"); 19302f270590SStefan Hajnoczi return; 19312f270590SStefan Hajnoczi } 19321bf8a989SDenis Plotnikov if (conf->queue_size <= 2) { 19331bf8a989SDenis Plotnikov error_setg(errp, "invalid queue-size property (%" PRIu16 "), " 19341bf8a989SDenis Plotnikov "must be > 2", conf->queue_size); 19351bf8a989SDenis Plotnikov return; 19361bf8a989SDenis Plotnikov } 19376040aeddSMark Kanda if (!is_power_of_2(conf->queue_size) || 19386040aeddSMark Kanda conf->queue_size > VIRTQUEUE_MAX_SIZE) { 19396040aeddSMark Kanda error_setg(errp, "invalid queue-size property (%" PRIu16 "), " 19406040aeddSMark Kanda "must be a power of 2 (max %d)", 19416040aeddSMark Kanda conf->queue_size, VIRTQUEUE_MAX_SIZE); 19426040aeddSMark Kanda return; 19436040aeddSMark Kanda } 19446e790746SPaolo Bonzini 1945ceff3e1fSMao Zhongyi if (!blkconf_apply_backend_options(&conf->conf, 194686b1cf32SKevin Wolf !blk_supports_write_perm(conf->conf.blk), 194786b1cf32SKevin Wolf true, errp)) { 1948a17c17a2SKevin Wolf return; 1949a17c17a2SKevin Wolf } 19504be74634SMarkus Armbruster s->original_wce = blk_enable_write_cache(conf->conf.blk); 1951ceff3e1fSMao Zhongyi if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) { 195275884afdSAndreas Färber return; 19536e790746SPaolo Bonzini } 1954ceff3e1fSMao Zhongyi 1955c56ee92fSRoman Kagan if (!blkconf_blocksizes(&conf->conf, errp)) { 19560a75b60cSMark Kanda return; 19570a75b60cSMark Kanda } 19580a75b60cSMark Kanda 19594f736650SSam Li BlockDriverState *bs = blk_bs(conf->conf.blk); 19604f736650SSam Li if (bs->bl.zoned != BLK_Z_NONE) { 19614f736650SSam Li virtio_add_feature(&s->host_features, VIRTIO_BLK_F_ZONED); 19624f736650SSam Li if (bs->bl.zoned == BLK_Z_HM) { 19634f736650SSam Li virtio_clear_feature(&s->host_features, VIRTIO_BLK_F_DISCARD); 19644f736650SSam Li } 19654f736650SSam Li } 19664f736650SSam Li 196737b06f8dSStefano Garzarella if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) && 196837b06f8dSStefano Garzarella (!conf->max_discard_sectors || 196937b06f8dSStefano Garzarella conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) { 197037b06f8dSStefano Garzarella error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")" 197137b06f8dSStefano Garzarella ", must be between 1 and %d", 197237b06f8dSStefano Garzarella conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS); 197337b06f8dSStefano Garzarella return; 197437b06f8dSStefano Garzarella } 197537b06f8dSStefano Garzarella 197637b06f8dSStefano Garzarella if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) && 197737b06f8dSStefano Garzarella (!conf->max_write_zeroes_sectors || 197837b06f8dSStefano Garzarella conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) { 197937b06f8dSStefano Garzarella error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32 198037b06f8dSStefano Garzarella "), must be between 1 and %d", 198137b06f8dSStefano Garzarella conf->max_write_zeroes_sectors, 198237b06f8dSStefano Garzarella (int)BDRV_REQUEST_MAX_SECTORS); 198337b06f8dSStefano Garzarella return; 198437b06f8dSStefano Garzarella } 198537b06f8dSStefano Garzarella 1986b6948ab0SStefan Hajnoczi if (conf->iothread_vq_mapping_list) { 1987b6948ab0SStefan Hajnoczi if (conf->iothread) { 1988b6948ab0SStefan Hajnoczi error_setg(errp, "iothread and iothread-vq-mapping properties " 1989b6948ab0SStefan Hajnoczi "cannot be set at the same time"); 1990b6948ab0SStefan Hajnoczi return; 1991b6948ab0SStefan Hajnoczi } 1992b6948ab0SStefan Hajnoczi 1993b6948ab0SStefan Hajnoczi if (!validate_iothread_vq_mapping_list(conf->iothread_vq_mapping_list, 1994b6948ab0SStefan Hajnoczi conf->num_queues, errp)) { 1995b6948ab0SStefan Hajnoczi return; 1996b6948ab0SStefan Hajnoczi } 1997b6948ab0SStefan Hajnoczi } 1998b6948ab0SStefan Hajnoczi 1999d9cf55a8SDaniil Tatianin s->config_size = virtio_get_config_size(&virtio_blk_cfg_size_params, 2000d74c30c8SDaniil Tatianin s->host_features); 20013857cd5cSJonah Palmer virtio_init(vdev, VIRTIO_ID_BLOCK, s->config_size); 20026e790746SPaolo Bonzini 20039c67f33fSStefan Hajnoczi qemu_mutex_init(&s->rq_lock); 20049c67f33fSStefan Hajnoczi 20054be74634SMarkus Armbruster s->blk = conf->conf.blk; 20066e790746SPaolo Bonzini s->rq = NULL; 20072a30307fSMarkus Armbruster s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1; 20086e790746SPaolo Bonzini 20092f270590SStefan Hajnoczi for (i = 0; i < conf->num_queues; i++) { 20106040aeddSMark Kanda virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output); 20112f270590SStefan Hajnoczi } 201298e3ab35SKevin Wolf qemu_coroutine_inc_pool_size(conf->num_queues * conf->queue_size / 2); 201357bc2658SStefan Hajnoczi 20143cdaf3ddSStefan Hajnoczi /* Don't start ioeventfd if transport does not support notifiers. */ 201557bc2658SStefan Hajnoczi if (!virtio_device_ioeventfd_enabled(vdev)) { 20163cdaf3ddSStefan Hajnoczi s->ioeventfd_disabled = true; 201757bc2658SStefan Hajnoczi } 201857bc2658SStefan Hajnoczi 201957bc2658SStefan Hajnoczi virtio_blk_vq_aio_context_init(s, &err); 20203ffeeef7SAndreas Färber if (err != NULL) { 202175884afdSAndreas Färber error_propagate(errp, err); 2022cfaf757eSPan Nengyuan for (i = 0; i < conf->num_queues; i++) { 2023cfaf757eSPan Nengyuan virtio_del_queue(vdev, i); 2024cfaf757eSPan Nengyuan } 20256a1a8cc7SKONRAD Frederic virtio_cleanup(vdev); 202675884afdSAndreas Färber return; 20276e790746SPaolo Bonzini } 20286e790746SPaolo Bonzini 2029a937f8e8SStefan Hajnoczi /* 2030a937f8e8SStefan Hajnoczi * This must be after virtio_init() so virtio_blk_dma_restart_cb() gets 2031a937f8e8SStefan Hajnoczi * called after ->start_ioeventfd() has already set blk's AioContext. 2032a937f8e8SStefan Hajnoczi */ 2033a937f8e8SStefan Hajnoczi s->change = 2034a937f8e8SStefan Hajnoczi qdev_add_vm_change_state_handler(dev, virtio_blk_dma_restart_cb, s); 2035a937f8e8SStefan Hajnoczi 2036baf42268SStefan Hajnoczi blk_ram_registrar_init(&s->blk_ram_registrar, s->blk); 20374be74634SMarkus Armbruster blk_set_dev_ops(s->blk, &virtio_block_ops, s); 20386e790746SPaolo Bonzini 20394be74634SMarkus Armbruster blk_iostatus_enable(s->blk); 204071f571a2SSam Eiderman 204171f571a2SSam Eiderman add_boot_device_lchs(dev, "/disk@0,0", 204271f571a2SSam Eiderman conf->conf.lcyls, 204371f571a2SSam Eiderman conf->conf.lheads, 204471f571a2SSam Eiderman conf->conf.lsecs); 20456e790746SPaolo Bonzini } 20466e790746SPaolo Bonzini 2047b69c3c21SMarkus Armbruster static void virtio_blk_device_unrealize(DeviceState *dev) 20486e790746SPaolo Bonzini { 2049306ec6c3SAndreas Färber VirtIODevice *vdev = VIRTIO_DEVICE(dev); 2050306ec6c3SAndreas Färber VirtIOBlock *s = VIRTIO_BLK(dev); 20514a0117cfSEugenio Pérez VirtIOBlkConf *conf = &s->conf; 20524a0117cfSEugenio Pérez unsigned i; 2053306ec6c3SAndreas Färber 20547bfde688SJulia Suvorova blk_drain(s->blk); 205571f571a2SSam Eiderman del_boot_device_lchs(dev, "/disk@0,0"); 205657bc2658SStefan Hajnoczi virtio_blk_vq_aio_context_cleanup(s); 20574a0117cfSEugenio Pérez for (i = 0; i < conf->num_queues; i++) { 20584a0117cfSEugenio Pérez virtio_del_queue(vdev, i); 20594a0117cfSEugenio Pérez } 206098e3ab35SKevin Wolf qemu_coroutine_dec_pool_size(conf->num_queues * conf->queue_size / 2); 20619c67f33fSStefan Hajnoczi qemu_mutex_destroy(&s->rq_lock); 2062baf42268SStefan Hajnoczi blk_ram_registrar_destroy(&s->blk_ram_registrar); 20636e790746SPaolo Bonzini qemu_del_vm_change_state_handler(s->change); 20644be74634SMarkus Armbruster blockdev_mark_auto_del(s->blk); 20656a1a8cc7SKONRAD Frederic virtio_cleanup(vdev); 20666e790746SPaolo Bonzini } 20676e790746SPaolo Bonzini 2068467b3f33SStefan Hajnoczi static void virtio_blk_instance_init(Object *obj) 2069467b3f33SStefan Hajnoczi { 2070467b3f33SStefan Hajnoczi VirtIOBlock *s = VIRTIO_BLK(obj); 2071467b3f33SStefan Hajnoczi 20722a30307fSMarkus Armbruster device_add_bootindex_property(obj, &s->conf.conf.bootindex, 20733342ec32SGonglei "bootindex", "/disk@0,0", 207440c2281cSMarkus Armbruster DEVICE(obj)); 2075467b3f33SStefan Hajnoczi } 2076467b3f33SStefan Hajnoczi 2077977a117fSHalil Pasic static const VMStateDescription vmstate_virtio_blk = { 2078977a117fSHalil Pasic .name = "virtio-blk", 2079977a117fSHalil Pasic .minimum_version_id = 2, 2080977a117fSHalil Pasic .version_id = 2, 20817d5dc0a3SRichard Henderson .fields = (const VMStateField[]) { 2082977a117fSHalil Pasic VMSTATE_VIRTIO_DEVICE, 2083977a117fSHalil Pasic VMSTATE_END_OF_LIST() 2084977a117fSHalil Pasic }, 2085977a117fSHalil Pasic }; 2086bbded32cSDr. David Alan Gilbert 20876e790746SPaolo Bonzini static Property virtio_blk_properties[] = { 20882a30307fSMarkus Armbruster DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf), 20898c398252SKevin Wolf DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf), 20902a30307fSMarkus Armbruster DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf), 20912a30307fSMarkus Armbruster DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial), 2092bbe8bd4dSStefano Garzarella DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features, 2093bbe8bd4dSStefano Garzarella VIRTIO_BLK_F_CONFIG_WCE, true), 209432a877e4SStefan Hajnoczi #ifdef __linux__ 2095bbe8bd4dSStefano Garzarella DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features, 2096bbe8bd4dSStefano Garzarella VIRTIO_BLK_F_SCSI, false), 209732a877e4SStefan Hajnoczi #endif 2098c99495acSPeter Lieven DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0, 2099c99495acSPeter Lieven true), 21009445e1e1SStefan Hajnoczi DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 21019445e1e1SStefan Hajnoczi VIRTIO_BLK_AUTO_NUM_QUEUES), 2102c9b7d9ecSDenis Plotnikov DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 256), 21031bf8a989SDenis Plotnikov DEFINE_PROP_BOOL("seg-max-adjust", VirtIOBlock, conf.seg_max_adjust, true), 2104d679ac09SFam Zheng DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD, 2105d679ac09SFam Zheng IOThread *), 2106b6948ab0SStefan Hajnoczi DEFINE_PROP_IOTHREAD_VQ_MAPPING_LIST("iothread-vq-mapping", VirtIOBlock, 2107b6948ab0SStefan Hajnoczi conf.iothread_vq_mapping_list), 21085c81161fSStefano Garzarella DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features, 21095c81161fSStefano Garzarella VIRTIO_BLK_F_DISCARD, true), 2110fb0b154cSAkihiko Odaki DEFINE_PROP_BOOL("report-discard-granularity", VirtIOBlock, 2111fb0b154cSAkihiko Odaki conf.report_discard_granularity, true), 21125c81161fSStefano Garzarella DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features, 21135c81161fSStefano Garzarella VIRTIO_BLK_F_WRITE_ZEROES, true), 211437b06f8dSStefano Garzarella DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock, 211537b06f8dSStefano Garzarella conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS), 211637b06f8dSStefano Garzarella DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock, 211737b06f8dSStefano Garzarella conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS), 21185f258577SEvgeny Yakovlev DEFINE_PROP_BOOL("x-enable-wce-if-config-wce", VirtIOBlock, 21195f258577SEvgeny Yakovlev conf.x_enable_wce_if_config_wce, true), 21206e790746SPaolo Bonzini DEFINE_PROP_END_OF_LIST(), 21216e790746SPaolo Bonzini }; 21226e790746SPaolo Bonzini 21236e790746SPaolo Bonzini static void virtio_blk_class_init(ObjectClass *klass, void *data) 21246e790746SPaolo Bonzini { 21256e790746SPaolo Bonzini DeviceClass *dc = DEVICE_CLASS(klass); 21266e790746SPaolo Bonzini VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 212775884afdSAndreas Färber 21284f67d30bSMarc-André Lureau device_class_set_props(dc, virtio_blk_properties); 2129bbded32cSDr. David Alan Gilbert dc->vmsd = &vmstate_virtio_blk; 2130125ee0edSMarcel Apfelbaum set_bit(DEVICE_CATEGORY_STORAGE, dc->categories); 213175884afdSAndreas Färber vdc->realize = virtio_blk_device_realize; 2132306ec6c3SAndreas Färber vdc->unrealize = virtio_blk_device_unrealize; 21336e790746SPaolo Bonzini vdc->get_config = virtio_blk_update_config; 21346e790746SPaolo Bonzini vdc->set_config = virtio_blk_set_config; 21356e790746SPaolo Bonzini vdc->get_features = virtio_blk_get_features; 21366e790746SPaolo Bonzini vdc->set_status = virtio_blk_set_status; 21376e790746SPaolo Bonzini vdc->reset = virtio_blk_reset; 2138b2b295a7SGreg Kurz vdc->save = virtio_blk_save_device; 2139b2b295a7SGreg Kurz vdc->load = virtio_blk_load_device; 21403cdaf3ddSStefan Hajnoczi vdc->start_ioeventfd = virtio_blk_start_ioeventfd; 21413cdaf3ddSStefan Hajnoczi vdc->stop_ioeventfd = virtio_blk_stop_ioeventfd; 21426e790746SPaolo Bonzini } 21436e790746SPaolo Bonzini 2144b5c7ceafSChanglong Xie static const TypeInfo virtio_blk_info = { 21456e790746SPaolo Bonzini .name = TYPE_VIRTIO_BLK, 21466e790746SPaolo Bonzini .parent = TYPE_VIRTIO_DEVICE, 21476e790746SPaolo Bonzini .instance_size = sizeof(VirtIOBlock), 2148467b3f33SStefan Hajnoczi .instance_init = virtio_blk_instance_init, 21496e790746SPaolo Bonzini .class_init = virtio_blk_class_init, 21506e790746SPaolo Bonzini }; 21516e790746SPaolo Bonzini 21526e790746SPaolo Bonzini static void virtio_register_types(void) 21536e790746SPaolo Bonzini { 2154b5c7ceafSChanglong Xie type_register_static(&virtio_blk_info); 21556e790746SPaolo Bonzini } 21566e790746SPaolo Bonzini 21576e790746SPaolo Bonzini type_init(virtio_register_types) 2158