Commit 7cd20dd3 authored by Shuhei Matsumoto's avatar Shuhei Matsumoto Committed by Tomasz Zawadzki
Browse files

lib/bdev: Add spdk_bdev_abort API 



Add spdk_bdev_abort function as a new public API.

This goes all the way down to the bdev driver module and attempts to
abort all I/Os which has bio_cb_arg as its callback argument.

We can separate when only a single I/O has bio_cb_arg and when multiple
I/Os have bio_cb_arg, but unify both by using parent - children I/O
relationship. To avoid confusion, return matched_ios by _bdev_abort() and
store it into split_outstanding by the caller.

Exclude any I/O submitted after this abort command because the same cb_arg
may be used by all I/Os and abort may never complete.

bdev_io needs to have both bio_cb_arg and bio_to_abort because bio_cb_arg
is used to continue abort processing when it is stopped due to the capacity
of bdev_io pool, and bio_to_abort is used to pass it to the underlying
bdev module at submission. Parent I/O is not submitted directly, and is
only used in the generic bdev layer, and parent I/O's bdev_io uses bio_cb_arg.
Hence add bio_cb_arg to bdev structure and add bio_to_abort to abort structure.

In the meantime of abort operation, target I/Os may be completed. Hence
check if the target I/O still exists at completion, and set the completion
status to false only if it still exists.

Upon completion of this, i.e., this returned zero, the status
SPDK_BDEV_IO_STATUS_SUCCESS indicates all I/Os were successfully aborted,
or the status SPDK_BDEV_IO_STATUS_FAILED indicates any I/O was failed to
abort by any reason.

spdk_bdev_abort() does not support aborting abort or reset request
due to the complexity for now.

Following patches will support I/O split case.

Add unit tests together to cover the basic paths.

Besides, ABI compatibility check required us to bump up SO version of
a few libraries or modules. Bump up SO version of blob bdev module simply
because it does not have any out-of-tree consumer, and suppress bumping
up SO version of lvol library because the affected struct spdk_lvol
is not part of public APIs.

Signed-off-by: default avatarShuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I515da688503557615d491bf0bfb36322ce37df08
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/2014


Tested-by: default avatarSPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI
Reviewed-by: default avatarJim Harris <james.r.harris@intel.com>
Reviewed-by: default avatarAleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: default avatarBen Walker <benjamin.walker@intel.com>
parent b62bfbf6

include/spdk/bdev.h

+32 −0
Original line number Diff line number Diff line
@@ -139,6 +139,7 @@ enum spdk_bdev_io_type { 
	SPDK_BDEV_IO_TYPE_ZONE_APPEND,

	SPDK_BDEV_IO_TYPE_COMPARE,

	SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE,

	SPDK_BDEV_IO_TYPE_ABORT,

	SPDK_BDEV_NUM_IO_TYPES /* Keep last */

};
@@ -1385,6 +1386,37 @@ int spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel * 
int spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,

		    spdk_bdev_io_completion_cb cb, void *cb_arg);



/**

 * Submit abort requests to abort all I/Os which has bio_cb_arg as its callback

 * context to the bdev on the given channel.

 *

 * This goes all the way down to the bdev driver module and attempts to abort all

 * I/Os which have bio_cb_arg as their callback context if they exist. This is a best

 * effort command. Upon completion of this, the status SPDK_BDEV_IO_STATUS_SUCCESS

 * indicates all the I/Os were successfully aborted, or the status

 * SPDK_BDEV_IO_STATUS_FAILED indicates any I/O was failed to abort for any reason

 * or no I/O which has bio_cb_arg as its callback context was found.

 *

 * \ingroup bdev_io_submit functions

 *

 * \param desc Block device descriptor.

 * \param ch The I/O channel which the I/Os to be aborted are associated with.

 * \param bio_cb_arg Callback argument for the outstanding requests which this

 * function attempts to abort.

 * \param cb Called when the abort request is completed.

 * \param cb_arg Argument passed to cb.

 *

 * \return 0 on success. On success, the callback will always be called (even if the

 * request ultimately failed). Return negated errno on failure, in which case the

 * callback will not be called.

 *   * -EINVAL - bio_cb_arg was not specified.

 *   * -ENOMEM - spdk_bdev_io buffer cannot be allocated.

 *   * -ENOTSUP - the bdev does not support abort.

 */

int spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,

		    void *bio_cb_arg,

		    spdk_bdev_io_completion_cb cb, void *cb_arg);



/**

 * Submit an NVMe Admin command to the bdev. This passes directly through

 * the block layer to the device. Support for NVMe passthru is optional,

include/spdk/bdev_module.h

+11 −0
Original line number Diff line number Diff line
@@ -533,11 +533,22 @@ struct spdk_bdev_io { 
				/** True if this request is in the 'start' phase of zcopy. False if in 'end'. */

				uint8_t start : 1;

			} zcopy;



			struct {

				/** The callback argument for the outstanding request which this abort

				 *  attempts to cancel.

				 */

				void *bio_cb_arg;

			} abort;

		} bdev;

		struct {

			/** Channel reference held while messages for this reset are in progress. */

			struct spdk_io_channel *ch_ref;

		} reset;

		struct {

			/** The outstanding request matching bio_cb_arg which this abort attempts to cancel. */

			struct spdk_bdev_io *bio_to_abort;

		} abort;

		struct {

			/* The NVMe command to execute */

			struct spdk_nvme_cmd cmd;

lib/bdev/bdev.c

+211 −0
Original line number Diff line number Diff line
@@ -370,6 +370,8 @@ bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch, 
		      uint64_t offset, uint64_t length,

		      lock_range_cb cb_fn, void *cb_arg);



static inline void bdev_io_complete(void *ctx);



void

spdk_bdev_get_opts(struct spdk_bdev_opts *opts)

{
@@ -4410,6 +4412,215 @@ spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channe 
	return 0;

}



static void bdev_abort_retry(void *ctx);



static void

bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)

{

	struct spdk_bdev_channel *channel = bdev_io->internal.ch;

	struct spdk_bdev_io *parent_io = cb_arg;

	struct spdk_bdev_io *bio_to_abort, *tmp_io;



	bio_to_abort = bdev_io->u.abort.bio_to_abort;



	spdk_bdev_free_io(bdev_io);



	if (!success) {

		/* Check if the target I/O completed in the meantime. */

		TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {

			if (tmp_io == bio_to_abort) {

				break;

			}

		}



		/* If the target I/O still exists, set the parent to failed. */

		if (tmp_io != NULL) {

			parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;

		}

	}



	parent_io->u.bdev.split_outstanding--;

	if (parent_io->u.bdev.split_outstanding == 0) {

		if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {

			bdev_abort_retry(parent_io);

		} else {

			bdev_io_complete(parent_io);

		}

	}

}



static int

bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,

	      struct spdk_bdev_io *bio_to_abort,

	      spdk_bdev_io_completion_cb cb, void *cb_arg)

{

	struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);

	struct spdk_bdev_io *bdev_io;



	if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||

	    bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {

		/* TODO: Abort reset or abort request. */

		return -ENOTSUP;

	}



	if (bdev->split_on_optimal_io_boundary && bdev_io_should_split(bio_to_abort)) {

		return -ENOTSUP;

	}



	bdev_io = bdev_channel_get_io(channel);

	if (bdev_io == NULL) {

		return -ENOMEM;

	}



	bdev_io->internal.ch = channel;

	bdev_io->internal.desc = desc;

	bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;

	bdev_io_init(bdev_io, bdev, cb_arg, cb);



	bdev_io->u.abort.bio_to_abort = bio_to_abort;



	/* Submit the abort request to the underlying bdev module. */

	bdev_io_submit(bdev_io);



	return 0;

}



static uint32_t

_bdev_abort(struct spdk_bdev_io *parent_io)

{

	struct spdk_bdev_desc *desc = parent_io->internal.desc;

	struct spdk_bdev_channel *channel = parent_io->internal.ch;

	void *bio_cb_arg;

	struct spdk_bdev_io *bio_to_abort;

	uint32_t matched_ios;

	int rc;



	bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;



	/* matched_ios is returned and will be kept by the caller.

	 *

	 * This funcion will be used for two cases, 1) the same cb_arg is used for

	 * multiple I/Os, 2) a single large I/O is split into smaller ones.

	 * Incrementing split_outstanding directly here may confuse readers especially

	 * for the 1st case.

	 *

	 * Completion of I/O abort is processed after stack unwinding. Hence this trick

	 * works as expected.

	 */

	matched_ios = 0;

	parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;



	TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {

		if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {

			continue;

		}



		if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {

			/* Any I/O which was submitted after this abort command should be excluded. */

			continue;

		}



		rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);

		if (rc != 0) {

			if (rc == -ENOMEM) {

				parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;

			} else {

				parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;

			}

			break;

		}

		matched_ios++;

	}



	return matched_ios;

}



static void

bdev_abort_retry(void *ctx)

{

	struct spdk_bdev_io *parent_io = ctx;

	uint32_t matched_ios;



	matched_ios = _bdev_abort(parent_io);



	if (matched_ios == 0) {

		if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {

			bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);

		} else {

			/* For retry, the case that no target I/O was found is success

			 * because it means target I/Os completed in the meantime.

			 */

			bdev_io_complete(parent_io);

		}

		return;

	}



	/* Use split_outstanding to manage the progress of aborting I/Os. */

	parent_io->u.bdev.split_outstanding = matched_ios;

}



static void

bdev_abort(struct spdk_bdev_io *parent_io)

{

	uint32_t matched_ios;



	matched_ios = _bdev_abort(parent_io);



	if (matched_ios == 0) {

		if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {

			bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);

		} else {

			/* The case the no target I/O was found is failure. */

			parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;

			bdev_io_complete(parent_io);

		}

		return;

	}



	/* Use split_outstanding to manage the progress of aborting I/Os. */

	parent_io->u.bdev.split_outstanding = matched_ios;

}



int

spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,

		void *bio_cb_arg,

		spdk_bdev_io_completion_cb cb, void *cb_arg)

{

	struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);

	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	struct spdk_bdev_io *bdev_io;



	if (bio_cb_arg == NULL) {

		return -EINVAL;

	}



	if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {

		return -ENOTSUP;

	}



	bdev_io = bdev_channel_get_io(channel);

	if (bdev_io == NULL) {

		return -ENOMEM;

	}



	bdev_io->internal.ch = channel;

	bdev_io->internal.desc = desc;

	bdev_io->internal.submit_tsc = spdk_get_ticks();

	bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;

	bdev_io_init(bdev_io, bdev, cb_arg, cb);



	bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;



	/* Parent abort request is not submitted directly, but to manage its execution,

	 * add it to the submitted list here.

	 */

	TAILQ_INSERT_TAIL(&channel->io_submitted, bdev_io, internal.ch_link);



	bdev_abort(bdev_io);



	return 0;

}



int

spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,

			struct spdk_bdev_io_wait_entry *entry)

lib/bdev/spdk_bdev.map

+1 −0
Original line number Diff line number Diff line
@@ -73,6 +73,7 @@ 
	spdk_bdev_flush;

	spdk_bdev_flush_blocks;

	spdk_bdev_reset;

	spdk_bdev_abort;

	spdk_bdev_nvme_admin_passthru;

	spdk_bdev_nvme_io_passthru;

	spdk_bdev_nvme_io_passthru_md;

module/blob/bdev/Makefile

+1 −1
Original line number Diff line number Diff line
@@ -34,7 +34,7 @@ 
SPDK_ROOT_DIR := $(abspath $(CURDIR)/../../..)

include $(SPDK_ROOT_DIR)/mk/spdk.common.mk



SO_VER := 2

SO_VER := 3

SO_MINOR := 0



C_SRCS = blob_bdev.c