From patchwork Thu Apr 29 15:50:24 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Martin Wilck X-Patchwork-Id: 12231527 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-18.7 required=3.0 tests=BAYES_00,DKIM_SIGNED, DKIM_VALID,DKIM_VALID_AU,HEADER_FROM_DIFFERENT_DOMAINS,INCLUDES_CR_TRAILER, INCLUDES_PATCH,MAILING_LIST_MULTI,SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED, USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id E59C2C43460 for ; Thu, 29 Apr 2021 15:50:50 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 9CEFE6145A for ; Thu, 29 Apr 2021 15:50:50 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232989AbhD2Pvg (ORCPT ); Thu, 29 Apr 2021 11:51:36 -0400 Received: from mx2.suse.de ([195.135.220.15]:53584 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S233132AbhD2Pvg (ORCPT ); Thu, 29 Apr 2021 11:51:36 -0400 X-Virus-Scanned: by amavisd-new at test-mx.suse.de DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=suse.com; s=susede1; t=1619711448; h=from:from:reply-to:date:date:message-id:message-id:to:to:cc:cc: mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=BKeSflXAip1a/Ym4ect3GjyGYWkw7IWehIwq8rQQzog=; b=rYOkpaSe6IFvnfm7SHhSAcShFqpgT34Lk8MlTvkxUcN43LHD9g4yEUifRv2J3+SgZSJz+m e4EZHHXTY5UeidFp1lWvJGWu4+6NgMkdV32UHgYHON+q8w5zrLoA4Xm6oYRjXvC7X4NABX lrdGmiJjk4YAmqKylSB3Znm6yvi4tds= Received: from relay2.suse.de (unknown [195.135.221.27]) by mx2.suse.de (Postfix) with ESMTP id 69DECAFCC; Thu, 29 Apr 2021 15:50:48 +0000 (UTC) From: mwilck@suse.com To: Mike Snitzer , Alasdair G Kergon , dm-devel@redhat.com, Hannes Reinecke Cc: Daniel Wagner , linux-block@vger.kernel.org, Paolo Bonzini , Christoph Hellwig , Benjamin Marzinski , Martin Wilck Subject: [RFC PATCH v2 2/2] dm: add CONFIG_DM_MULTIPATH_SG_IO - failover for SG_IO on dm-multipath Date: Thu, 29 Apr 2021 17:50:24 +0200 Message-Id: <20210429155024.4947-3-mwilck@suse.com> X-Mailer: git-send-email 2.31.1 In-Reply-To: <20210429155024.4947-1-mwilck@suse.com> References: <20210429155024.4947-1-mwilck@suse.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Martin Wilck In virtual deployments, SCSI passthrough over dm-multipath devices is a common setup. The qemu "pr-helper" was specifically invented for it. I believe that this is the most important real-world scenario for sending SG_IO ioctls to device-mapper devices. In this configuration, guests send SCSI IO to the hypervisor in the form of SG_IO ioctls issued by qemu. But on the device-mapper level, these SCSI ioctls aren't treated like regular IO. Until commit 2361ae595352 ("dm mpath: switch paths in dm_blk_ioctl() code path"), no path switching was done at all. Worse though, if an SG_IO call fails because of a path error, dm-multipath doesn't retry the IO on a another path; rather, the failure is passed back to the guest, and paths are not marked as faulty. This is in stark contrast with regular block IO of guests on dm-multipath devices, and certainly comes as a surprise to users who switch to SCSI passthrough in qemu. In general, users of dm-multipath devices would probably expect failover to work at least in a basic way. This patch fixes this by taking a special code path for SG_IO on request- based device mapper targets if CONFIG_DM_MULTIPATH_SG_IO is set. Rather then just choosing a single path, sending the IO to it, and failing to the caller if the IO on the path failed, it retries the same IO on another path for certain error codes, using blk_path_error() to determine if a retry would make sense for the given error code. Moreover, it sends a message to the multipath target to mark the path as failed. One problem remains open: if all paths in a multipath map are failed, normal multipath IO may switch to queueing mode (depending on configuration). This isn't possible for SG_IO, as SG_IO requests can't easily be queued like regular block I/O. Thus in the "no path" case, the guest will still see an error. Signed-off-by: Martin Wilck --- block/scsi_ioctl.c | 5 +- drivers/md/Kconfig | 11 ++++ drivers/md/Makefile | 4 ++ drivers/md/dm-core.h | 5 ++ drivers/md/dm-rq.h | 11 ++++ drivers/md/dm-scsi_ioctl.c | 127 +++++++++++++++++++++++++++++++++++++ drivers/md/dm.c | 20 +++++- include/linux/blkdev.h | 2 + 8 files changed, 180 insertions(+), 5 deletions(-) create mode 100644 drivers/md/dm-scsi_ioctl.c diff --git a/block/scsi_ioctl.c b/block/scsi_ioctl.c index 6599bac0a78c..bcc60552f7b1 100644 --- a/block/scsi_ioctl.c +++ b/block/scsi_ioctl.c @@ -279,8 +279,8 @@ static int blk_complete_sghdr_rq(struct request *rq, struct sg_io_hdr *hdr, return ret; } -static int sg_io(struct request_queue *q, struct gendisk *bd_disk, - struct sg_io_hdr *hdr, fmode_t mode) +int sg_io(struct request_queue *q, struct gendisk *bd_disk, + struct sg_io_hdr *hdr, fmode_t mode) { unsigned long start_time; ssize_t ret = 0; @@ -369,6 +369,7 @@ static int sg_io(struct request_queue *q, struct gendisk *bd_disk, blk_put_request(rq); return ret; } +EXPORT_SYMBOL_GPL(sg_io); /** * sg_scsi_ioctl -- handle deprecated SCSI_IOCTL_SEND_COMMAND ioctl diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig index f2014385d48b..f28f29e3bd11 100644 --- a/drivers/md/Kconfig +++ b/drivers/md/Kconfig @@ -473,6 +473,17 @@ config DM_MULTIPATH_IOA If unsure, say N. +config DM_MULTIPATH_SG_IO + bool "Retry SCSI generic I/O on multipath devices" + depends on DM_MULTIPATH && BLK_SCSI_REQUEST + help + With this option, SCSI generic (SG) requests issued on multipath + devices will behave similar to regular block I/O: upon failure, + they are repeated on a different path, and the erroring device + is marked as failed. + + If unsure, say N. + config DM_DELAY tristate "I/O delaying target" depends on BLK_DEV_DM diff --git a/drivers/md/Makefile b/drivers/md/Makefile index ef7ddc27685c..187ea469f64a 100644 --- a/drivers/md/Makefile +++ b/drivers/md/Makefile @@ -88,6 +88,10 @@ ifeq ($(CONFIG_DM_INIT),y) dm-mod-objs += dm-init.o endif +ifeq ($(CONFIG_DM_MULTIPATH_SG_IO),y) +dm-mod-objs += dm-scsi_ioctl.o +endif + ifeq ($(CONFIG_DM_UEVENT),y) dm-mod-objs += dm-uevent.o endif diff --git a/drivers/md/dm-core.h b/drivers/md/dm-core.h index 5953ff2bd260..8bd8a8e3916e 100644 --- a/drivers/md/dm-core.h +++ b/drivers/md/dm-core.h @@ -189,4 +189,9 @@ extern atomic_t dm_global_event_nr; extern wait_queue_head_t dm_global_eventq; void dm_issue_global_event(void); +int __dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, + struct block_device **bdev, + struct dm_target **target); +void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx); + #endif diff --git a/drivers/md/dm-rq.h b/drivers/md/dm-rq.h index 1eea0da641db..c6d2853e4d1d 100644 --- a/drivers/md/dm-rq.h +++ b/drivers/md/dm-rq.h @@ -44,4 +44,15 @@ ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, ch ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, const char *buf, size_t count); +#ifdef CONFIG_DM_MULTIPATH_SG_IO +int dm_sg_io_ioctl(struct block_device *bdev, fmode_t mode, + unsigned int cmd, unsigned long uarg); +#else +static inline int dm_sg_io_ioctl(struct block_device *bdev, fmode_t mode, + unsigned int cmd, unsigned long uarg) +{ + return -ENOTTY; +} +#endif + #endif diff --git a/drivers/md/dm-scsi_ioctl.c b/drivers/md/dm-scsi_ioctl.c new file mode 100644 index 000000000000..70c5eb763101 --- /dev/null +++ b/drivers/md/dm-scsi_ioctl.c @@ -0,0 +1,127 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Copyright (C) 2021 Martin Wilck, SUSE LLC + */ + +#include "dm-core.h" +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define DM_MSG_PREFIX "sg_io" + +int dm_sg_io_ioctl(struct block_device *bdev, fmode_t mode, + unsigned int cmd, unsigned long uarg) +{ + struct mapped_device *md = bdev->bd_disk->private_data; + struct sg_io_hdr hdr; + void __user *arg = (void __user *)uarg; + int rc, srcu_idx; + char path_name[BDEVNAME_SIZE]; + + if (cmd != SG_IO) + return -ENOTTY; + + if (copy_from_user(&hdr, arg, sizeof(hdr))) + return -EFAULT; + + if (hdr.interface_id != 'S') + return -EINVAL; + + if (hdr.dxfer_len > (queue_max_hw_sectors(bdev->bd_disk->queue) << 9)) + return -EIO; + + for (;;) { + struct dm_target *tgt; + struct sg_io_hdr rhdr; + + rc = __dm_prepare_ioctl(md, &srcu_idx, &bdev, &tgt); + if (rc < 0) { + DMERR("%s: failed to get path: %d", + __func__, rc); + goto out; + } + + rhdr = hdr; + + rc = sg_io(bdev->bd_disk->queue, bdev->bd_disk, &rhdr, mode); + + DMDEBUG("SG_IO via %s: rc = %d D%02xH%02xM%02xS%02x", + bdevname(bdev, path_name), rc, + rhdr.driver_status, rhdr.host_status, + rhdr.msg_status, rhdr.status); + + /* + * Errors resulting from invalid parameters shouldn't be retried + * on another path. + */ + switch (rc) { + case -ENOIOCTLCMD: + case -EFAULT: + case -EINVAL: + case -EPERM: + goto out; + default: + break; + } + + if (rhdr.info & SG_INFO_CHECK) { + int result; + blk_status_t sts; + + __set_status_byte(&result, rhdr.status); + __set_msg_byte(&result, rhdr.msg_status); + __set_host_byte(&result, rhdr.host_status); + __set_driver_byte(&result, rhdr.driver_status); + + sts = __scsi_result_to_blk_status(&result, result); + rhdr.host_status = host_byte(result); + + /* See if this is a target or path error. */ + if (sts == BLK_STS_OK) + rc = 0; + else if (blk_path_error(sts)) + rc = -EIO; + else { + rc = blk_status_to_errno(sts); + goto out; + } + } + + if (rc == 0) { + /* success */ + if (copy_to_user(arg, &rhdr, sizeof(rhdr))) + rc = -EFAULT; + goto out; + } + + /* Failure - fail path by sending a message to the target */ + if (!tgt->type->message) { + DMWARN("invalid target!"); + rc = -EIO; + goto out; + } else { + char bdbuf[BDEVT_SIZE]; + char *argv[2] = { "fail_path", bdbuf }; + + scnprintf(bdbuf, sizeof(bdbuf), "%u:%u", + MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev)); + + DMDEBUG("sending \"%s %s\" to target", argv[0], argv[1]); + rc = tgt->type->message(tgt, 2, argv, NULL, 0); + if (rc < 0) + goto out; + } + + dm_unprepare_ioctl(md, srcu_idx); + } +out: + dm_unprepare_ioctl(md, srcu_idx); + return rc; +} diff --git a/drivers/md/dm.c b/drivers/md/dm.c index 50b693d776d6..5c2c205bf20e 100644 --- a/drivers/md/dm.c +++ b/drivers/md/dm.c @@ -522,8 +522,9 @@ static int dm_blk_report_zones(struct gendisk *disk, sector_t sector, #define dm_blk_report_zones NULL #endif /* CONFIG_BLK_DEV_ZONED */ -static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, - struct block_device **bdev) +int __dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, + struct block_device **bdev, + struct dm_target **target) { struct dm_target *tgt; struct dm_table *map; @@ -553,10 +554,19 @@ static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, goto retry; } + if (r >= 0 && target) + *target = tgt; + return r; } -static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx) +static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, + struct block_device **bdev) +{ + return __dm_prepare_ioctl(md, srcu_idx, bdev, NULL); +} + +void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx) { dm_put_live_table(md, srcu_idx); } @@ -567,6 +577,10 @@ static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, struct mapped_device *md = bdev->bd_disk->private_data; int r, srcu_idx; + if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && + ((r = dm_sg_io_ioctl(bdev, mode, cmd, arg)) != -ENOTTY)) + return r; + r = dm_prepare_ioctl(md, &srcu_idx, &bdev); if (r < 0) goto out; diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h index c032cfe133c7..bded0e6546da 100644 --- a/include/linux/blkdev.h +++ b/include/linux/blkdev.h @@ -934,6 +934,8 @@ extern int scsi_cmd_ioctl(struct request_queue *, struct gendisk *, fmode_t, unsigned int, void __user *); extern int sg_scsi_ioctl(struct request_queue *, struct gendisk *, fmode_t, struct scsi_ioctl_command __user *); +extern int sg_io(struct request_queue *, struct gendisk *, + struct sg_io_hdr *, fmode_t); extern int get_sg_io_hdr(struct sg_io_hdr *hdr, const void __user *argp); extern int put_sg_io_hdr(const struct sg_io_hdr *hdr, void __user *argp);