Deal with kernels pre-6.12 kernel that relocated unaligned.h

[systemtap.git] / tapset / linux / ioblock.stp

// Block I/O tapset
// Copyright (C) 2006 Intel Corp.
// Copyright (C) 2006 IBM Corp.
// Copyright (C) 2014-2017 Red Hat Inc.
//
// This file is part of systemtap, and is free software.  You can
// redistribute it and/or modify it under the terms of the GNU General
// Public License (GPL); either version 2, or (at your option) any
// later version.

%{
#include <linux/bio.h>
#ifdef STAPCONF_GENHD_H
#include <linux/genhd.h>
#else
#include <linux/blkdev.h>
#endif
#ifdef STAPCONF_BLK_TYPES
#include <linux/blk_types.h>
#endif
%}

private function disk_get_part_start_sect:long(disk:long, partno:long)
%{ /* pure */ /* unprivileged */
#if defined(STAPCONF_DISK_GET_PART)
    struct gendisk *disk = (struct gendisk *)(uintptr_t)STAP_ARG_disk;
    int partno = (int)STAP_ARG_partno;
    struct hd_struct *part;

    STAP_RETVALUE = 0;
    if (disk) {
	/* Before calling disk_get_part() on 'disk', we need to make
	 * sure the memory is readable. */
	(void)kread(&(disk->part_tbl));
	part = disk_get_part(disk, partno);

	if (part) {
	    /* Let's be doubly paranoid and make sure this memory is
	     * safe for reading. */
	    (void)kread(&(part->start_sect));
	    STAP_RETVALUE = part->start_sect;
	    disk_put_part(part);
	}
    }
    CATCH_DEREF_FAULT();
#endif
%}

/* Returns the REQ_OP_* bits for a bio structure. */
function bio_op:long(bio:long)
%{ /* pure */ /* unprivileged */
#if defined(STAPCONF_BIO_BI_OPF)
    struct bio *bio = (struct bio *)(uintptr_t)STAP_ARG_bio;

    STAP_RETVALUE = 0;
    if (bio) {
	/* Before calling bio_op() on bio, we need to make sure the
	 * memory is readable. */
	(void)kread(&(bio->bi_opf));
	STAP_RETVALUE = bio_op(bio);
    }
#else
    struct bio *bio = (struct bio *)(uintptr_t)STAP_ARG_bio;

    STAP_RETVALUE = 0;
    if (bio) {
	STAP_RETVALUE = kread(&(bio->bi_rw));
    }
#endif
    CATCH_DEREF_FAULT();
%}

/* get i-node number of mapped file */
@__private30 function __bio_ino:long(bio:long)
%{ /* pure */
    struct bio *bio = (struct bio *)(uintptr_t)STAP_ARG_bio;
    struct page *bv_page = ((bio && kread(&(bio->bi_vcnt)))
			    ? kread(&(bio->bi_io_vec[0].bv_page)) : NULL);

    STAP_RETVALUE = -1;
    if (bv_page) {
	/* Before calling PageSlab() and PageSwapCache() on bv_page, we
	 * need to make sure the bv_page struct is valid. */
	(void)kderef_buffer(NULL, bv_page, sizeof(struct page));

	if (!PageSlab(bv_page) && !PageSwapCache(bv_page)) {
	    struct address_space *mapping = kread(&(bv_page->mapping));
	    if (mapping && ((unsigned long)mapping & PAGE_MAPPING_ANON) == 0) {
		struct inode *host = kread(&(mapping->host));
		if (host)
		    STAP_RETVALUE = kread(&(host->i_ino));
	    }
	}
    }
    CATCH_DEREF_FAULT();
%}

/* returns 0 for read, 1 for write */
function bio_rw_num:long(rw:long)
{
    return (rw & BIO_WRITE)
}

/* returns R for read, W for write */
function bio_rw_str(rw:long)
{
    return bio_rw_num(rw) == BIO_READ ? "R" : "W"
}

/* returns start sector */
@__private30 function __bio_start_sect:long(bio:long)
{
    try {
	if (@defined(@cast(bio, "bio", "kernel")->bi_dev)) {
	    return @cast(bio, "bio", "kernel")->bi_bdev->bd_part->start_sect
	}
	else if (@defined(@cast(bio, "bio", "kernel")->bi_disk)) {
	    return disk_get_part_start_sect(@cast(bio, "bio", "kernel")->bi_disk,
					    @cast(bio, "bio", "kernel")->bi_partno)
	}
    } catch {
        return -1
    }
}

/* returns the block device name */
@__private30 function __bio_devname:string(bio:long)
{
    if (@defined(@cast(bio, "bio", "kernel")->bi_bdev)) {
	return bdevname(@cast(bio, "bio", "kernel")->bi_bdev)
    }
    else {
	return disk_name(@cast(bio, "bio", "kernel")->bi_disk,
			 @cast(bio, "bio", "kernel")->bi_partno)
    }
}

global BIO_READ = 0, BIO_WRITE = 1


/**
 * probe ioblock.request - Fires whenever making a generic block I/O request.
 * @name      : name of the probe point
 * @devname   : block device name
 * @ino       : i-node number of the mapped file
 * @sector    : beginning sector for the entire bio
 * @flags     : see below
 * 	    BIO_UPTODATE    0       ok after I/O completion
 * 	    BIO_RW_BLOCK    1       RW_AHEAD set, and read/write would block
 * 	    BIO_EOF         2       out-out-bounds error
 * 	    BIO_SEG_VALID   3       nr_hw_seg valid 
 * 	    BIO_CLONED      4       doesn't own data
 * 	    BIO_BOUNCED     5       bio is a bounce bio
 * 	    BIO_USER_MAPPED 6       contains user pages
 * 	    BIO_EOPNOTSUPP  7       not supported
 * @rw        : binary trace for read/write request
 * @opf       : operations and flags
 * @vcnt      : bio vector count which represents number of array element (page, offset, length) which make up this I/O request
 * @idx       : offset into the bio vector array
 * @phys_segments : number of segments in this bio after physical address coalescing is performed
 * @hw_segments :   number of segments after physical and DMA remapping hardware coalescing is performed
 * @size      : total size in bytes
 * @bdev      : target block device
 * @bdev_contains : points to the device object which contains the partition (when bio structure represents a partition)
 * @p_start_sect :  points to the start sector of the partition structure of the device
 *
 * Context:
 *  The process makes block I/O request
 */
probe ioblock.request = kernel.function ("submit_bio_noacct")!,
    kernel.function ("generic_make_request")
{
	name = "ioblock.request"
        devname = __bio_devname($bio)
        ino = __bio_ino($bio)

        sector = @choose_defined($bio->bi_iter->bi_sector, $bio->bi_sector)
        flags = $bio->bi_flags
	if (@defined($bio->bi_opf)) {
		rw = bio_op($bio)
		opf = $bio->bi_opf
	}
	else {
		rw = $bio->bi_rw
		opf = $bio->bi_rw
	}
        vcnt = $bio->bi_vcnt
        idx = @choose_defined($bio->bi_iter->bi_idx, $bio->bi_idx)
	/* use bi_vcnt as a proxy as it is close enough */
        phys_segments = @choose_defined($bio->bi_phys_segments, $bio->bi_vcnt)
	hw_segments = @choose_defined($bio->bi_hw_segments, 0)
        size = @choose_defined($bio->bi_iter->bi_size, $bio->bi_size)

        bdev = @choose_defined($bio->bi_bdev, 0)
        bdev_contains = @choose_defined($bio->bi_bdev->bd_contains, 0)
        p_start_sect = __bio_start_sect($bio)
}

/**
 * probe ioblock.end - Fires whenever a block I/O transfer is complete.
 * @name      : name of the probe point
 * @devname   : block device name
 * @ino       : i-node number of the mapped file
 * @bytes_done : number of bytes transferred
 * @sector    : beginning sector for the entire bio
 * @flags     : see below
 *      BIO_UPTODATE    0       ok after I/O completion
 *      BIO_RW_BLOCK    1       RW_AHEAD set, and read/write would block
 *      BIO_EOF         2       out-out-bounds error
 *      BIO_SEG_VALID   3       nr_hw_seg valid
 *      BIO_CLONED      4       doesn't own data
 *      BIO_BOUNCED     5       bio is a bounce bio
 *      BIO_USER_MAPPED 6       contains user pages
 *      BIO_EOPNOTSUPP  7       not supported
 * @error     : 0 on success
 * @rw        : binary trace for read/write request
 * @opf       : operations and flags
 * @vcnt      : bio vector count which represents number of array element (page, offset, length) which makes up this I/O request
 * @idx       : offset into the bio vector array
 * @phys_segments : number of segments in this bio after physical address coalescing is performed.
 * @hw_segments :   number of segments after physical and DMA remapping hardware coalescing is performed
 * @size      : total size in bytes
 *
 * Context:
 *  The process signals the transfer is done.
 */
probe ioblock.end = kernel.function("bio_endio")
{
	name = "ioblock.end"
        devname = __bio_devname($bio)
        ino = __bio_ino($bio)

        bytes_done = @choose_defined($bytes_done,
				     @choose_defined($bio->bi_iter->bi_size,
						     $bio->bi_size))
        error = @choose_defined($error, @choose_defined($bio->bi_error,
							$bio->bi_status))

        sector = @choose_defined($bio->bi_iter->bi_sector, $bio->bi_sector)
        flags = $bio->bi_flags
	if (@defined($bio->bi_opf)) {
		rw = bio_op($bio)
		opf = $bio->bi_opf
	}
	else {
		rw = $bio->bi_rw
		opf = $bio->bi_rw
	}
        vcnt = $bio->bi_vcnt
        idx = @choose_defined($bio->bi_iter->bi_idx, $bio->bi_idx)
	/* use bi_vcnt as a proxy as it is close enough */
        phys_segments = @choose_defined($bio->bi_phys_segments, $bio->bi_vcnt)
	hw_segments = @choose_defined($bio->bi_hw_segments, 0)
        size = @choose_defined($bio->bi_iter->bi_size, $bio->bi_size)
}

/**
 * probe ioblock_trace.bounce - Fires whenever a buffer bounce is needed for at least one page of a block IO request.
 * @name      : name of the probe point
 * @q         : request queue on which this bio was queued.
 * @devname   : device for which a buffer bounce was needed.
 * @ino       : i-node number of the mapped file
 * @bytes_done : number of bytes transferred
 * @sector    : beginning sector for the entire bio
 * @flags     : see below
 *     BIO_UPTODATE    0       ok after I/O completion
 *     BIO_RW_BLOCK    1       RW_AHEAD set, and read/write would block
 *     BIO_EOF         2       out-out-bounds error
 *     BIO_SEG_VALID   3       nr_hw_seg valid
 *     BIO_CLONED      4       doesn't own data
 *     BIO_BOUNCED     5       bio is a bounce bio
 *     BIO_USER_MAPPED 6       contains user pages
 *     BIO_EOPNOTSUPP  7       not supported
 * @rw        : binary trace for read/write request
 * @opf       : operations and flags
 * @vcnt      : bio vector count which represents number of array element (page, offset, length) which makes up this I/O request
 * @idx       : offset into the bio vector array
 * @phys_segments - number of segments in this bio after physical address coalescing is performed.
 * @size      : total size in bytes
 * @bdev      : target block device
 * @bdev_contains : points to the device object which contains the partition (when bio structure represents a partition)
 * @p_start_sect :  points to the start sector of the partition structure of the device
 *
 * Context :
 *	The process creating a block IO request.
 */
probe ioblock_trace.bounce = kernel.trace("block_bio_bounce")
{
	name = "ioblock_trace.bounce"
	q = @choose_defined($q, $bio->bi_bdev->bd_queue)
        devname = __bio_devname($bio)
        ino = __bio_ino($bio)

        bytes_done = @choose_defined($bio->bi_iter->bi_size, $bio->bi_size)
        sector = @choose_defined($bio->bi_iter->bi_sector, $bio->bi_sector)
        flags = $bio->bi_flags
	if (@defined($bio->bi_opf)) {
		rw = bio_op($bio)
		opf = $bio->bi_opf
	}
	else {
		rw = $bio->bi_rw
		opf = $bio->bi_rw
	}
        vcnt = $bio->bi_vcnt
        idx = @choose_defined($bio->bi_iter->bi_idx, $bio->bi_idx)
	/* use bi_vcnt as a proxy as it is close enough */
        phys_segments = @choose_defined($bio->bi_phys_segments, $bio->bi_vcnt)
        size = @choose_defined($bio->bi_iter->bi_size, $bio->bi_size)
        bdev_contains = @choose_defined($bio->bi_bdev->bd_contains, 0)
        bdev = @choose_defined($bio->bi_bdev, 0)
        p_start_sect = __bio_start_sect($bio)
}

/**
 * probe ioblock_trace.request - Fires just as a generic block I/O request is created for a bio.
 * @name      : name of the probe point
 * @q         : request queue on which this bio was queued.
 * @devname   : block device name
 * @ino       : i-node number of the mapped file
 * @bytes_done : number of bytes transferred
 * @sector    : beginning sector for the entire bio
 * @flags     : see below
 *	    BIO_UPTODATE    0       ok after I/O completion
 *	    BIO_RW_BLOCK    1       RW_AHEAD set, and read/write would block
 *	    BIO_EOF         2       out-out-bounds error
 *	    BIO_SEG_VALID   3       nr_hw_seg valid
 *	    BIO_CLONED      4       doesn't own data
 *	    BIO_BOUNCED     5       bio is a bounce bio
 *	    BIO_USER_MAPPED 6       contains user pages
 *	    BIO_EOPNOTSUPP  7       not supported
 * @rw        : binary trace for read/write request
 * @opf       : operations and flags
 * @vcnt      : bio vector count which represents number of array element (page, offset, length) which make up this I/O request
 * @idx       : offset into the bio vector array
 * @phys_segments - number of segments in this bio after physical address coalescing is performed.
 * @size      : total size in bytes
 * @bdev      : target block device
 * @bdev_contains : points to the device object which contains the partition (when bio structure represents a partition)
 * @p_start_sect :  points to the start sector of the partition structure of the device
 *
 * Context:
 *  The process makes block I/O request
 */
probe ioblock_trace.request = kernel.trace("block_bio_queue")
{
	name = "ioblock_trace.request"
	q = @choose_defined($q, $bio->bi_bdev->bd_queue)
        devname = __bio_devname($bio)
        ino = __bio_ino($bio)

        bytes_done = @choose_defined($bio->bi_iter->bi_size, $bio->bi_size)
        sector = @choose_defined($bio->bi_iter->bi_sector, $bio->bi_sector)
        flags = $bio->bi_flags
	if (@defined($bio->bi_opf)) {
		rw = bio_op($bio)
		opf = $bio->bi_opf
	}
	else {
		rw = $bio->bi_rw
		opf = $bio->bi_rw
	}
        vcnt = $bio->bi_vcnt
        idx = @choose_defined($bio->bi_iter->bi_idx, $bio->bi_idx)
	/* use bi_vcnt as a proxy as it is close enough */
        phys_segments = @choose_defined($bio->bi_phys_segments, $bio->bi_vcnt)
        size = @choose_defined($bio->bi_iter->bi_size, $bio->bi_size)
        bdev_contains = @choose_defined($bio->bi_bdev->bd_contains, 0)
        bdev = @choose_defined($bio->bi_bdev, 0)
        p_start_sect = __bio_start_sect($bio)
}

/**
 * probe ioblock_trace.end - Fires whenever a block I/O transfer is complete.
 * @name      : name of the probe point
 * @q         : request queue on which this bio was queued.
 * @devname   : block device name
 * @ino       : i-node number of the mapped file
 * @bytes_done : number of bytes transferred
 * @sector    : beginning sector for the entire bio
 * @flags     : see below
 *      BIO_UPTODATE    0       ok after I/O completion
 *      BIO_RW_BLOCK    1       RW_AHEAD set, and read/write would block
 *      BIO_EOF         2       out-out-bounds error
 *      BIO_SEG_VALID   3       nr_hw_seg valid
 *      BIO_CLONED      4       doesn't own data
 *      BIO_BOUNCED     5       bio is a bounce bio
 *      BIO_USER_MAPPED 6       contains user pages
 *      BIO_EOPNOTSUPP  7       not supported
 * @rw        : binary trace for read/write request
 * @opf       : operations and flags
 * @vcnt      : bio vector count which represents number of array element (page, offset, length) which makes up this I/O request
 * @idx       : offset into the bio vector array
 * @phys_segments - number of segments in this bio after physical address coalescing is performed.
 * @size      : total size in bytes
 * @bdev      : target block device
 * @bdev_contains : points to the device object which contains the partition (when bio structure represents a partition)
 * @p_start_sect :  points to the start sector of the partition structure of the device
 *
 * Context:
 *  The process signals the transfer is done.
 */
probe ioblock_trace.end = kernel.trace("block_bio_complete")
{
	name = "ioblock_trace.end"
	q = @choose_defined($q, $bio->bi_bdev->bd_queue)
        devname = __bio_devname($bio)
        ino = __bio_ino($bio)

        bytes_done = @choose_defined($bio->bi_iter->bi_size, $bio->bi_size)

        sector = @choose_defined($bio->bi_iter->bi_sector, $bio->bi_sector)
        flags = $bio->bi_flags
	if (@defined($bio->bi_opf)) {
		rw = bio_op($bio)
		opf = $bio->bi_opf
	}
	else {
		rw = $bio->bi_rw
		opf = $bio->bi_rw
	}
        vcnt = $bio->bi_vcnt
        idx = @choose_defined($bio->bi_iter->bi_idx, $bio->bi_idx)
	/* use bi_vcnt as a proxy as it is close enough */
        phys_segments = @choose_defined($bio->bi_phys_segments, $bio->bi_vcnt)
        size = @choose_defined($bio->bi_iter->bi_size, $bio->bi_size)
        bdev_contains = @choose_defined($bio->bi_bdev->bd_contains, 0)
        bdev = @choose_defined($bio->bi_bdev, 0)
        p_start_sect = __bio_start_sect($bio)
}