kernel - refactor vm_page busy

[dragonfly.git] / sys / kern / vfs_bio.c

/*
 * Copyright (c) 1994,1997 John S. Dyson
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice immediately at the beginning of the file, without modification,
 *    this list of conditions, and the following disclaimer.
 * 2. Absolutely no warranty of function or purpose is made by the author
 *              John S. Dyson.
 *
 * $FreeBSD: src/sys/kern/vfs_bio.c,v 1.242.2.20 2003/05/28 18:38:10 alc Exp $
 */

/*
 * this file contains a new buffer I/O scheme implementing a coherent
 * VM object and buffer cache scheme.  Pains have been taken to make
 * sure that the performance degradation associated with schemes such
 * as this is not realized.
 *
 * Author:  John S. Dyson
 * Significant help during the development and debugging phases
 * had been provided by David Greenman, also of the FreeBSD core team.
 *
 * see man buf(9) for more info.  Note that man buf(9) doesn't reflect
 * the actual buf/bio implementation in DragonFly.
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/buf.h>
#include <sys/conf.h>
#include <sys/devicestat.h>
#include <sys/eventhandler.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mount.h>
#include <sys/kernel.h>
#include <sys/kthread.h>
#include <sys/proc.h>
#include <sys/reboot.h>
#include <sys/resourcevar.h>
#include <sys/sysctl.h>
#include <sys/vmmeter.h>
#include <sys/vnode.h>
#include <sys/dsched.h>
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/vm_kern.h>
#include <vm/vm_pageout.h>
#include <vm/vm_page.h>
#include <vm/vm_object.h>
#include <vm/vm_extern.h>
#include <vm/vm_map.h>
#include <vm/vm_pager.h>
#include <vm/swap_pager.h>

#include <sys/buf2.h>
#include <sys/thread2.h>
#include <sys/spinlock2.h>
#include <vm/vm_page2.h>

#include "opt_ddb.h"
#ifdef DDB
#include <ddb/ddb.h>
#endif

/*
 * Buffer queues.
 */
enum bufq_type {
        BQUEUE_NONE,            /* not on any queue */
        BQUEUE_LOCKED,          /* locked buffers */
        BQUEUE_CLEAN,           /* non-B_DELWRI buffers */
        BQUEUE_DIRTY,           /* B_DELWRI buffers */
        BQUEUE_DIRTY_HW,        /* B_DELWRI buffers - heavy weight */
        BQUEUE_EMPTY,           /* empty buffer headers */

        BUFFER_QUEUES           /* number of buffer queues */
};

typedef enum bufq_type bufq_type_t;

#define BD_WAKE_SIZE    16384
#define BD_WAKE_MASK    (BD_WAKE_SIZE - 1)

TAILQ_HEAD(bqueues, buf);

struct bufpcpu {
        struct spinlock spin;
        struct bqueues bufqueues[BUFFER_QUEUES];
} __cachealign;

struct bufpcpu bufpcpu[MAXCPU];

static MALLOC_DEFINE(M_BIOBUF, "BIO buffer", "BIO buffer");

struct buf *buf;                /* buffer header pool */

static void vfs_clean_pages(struct buf *bp);
static void vfs_clean_one_page(struct buf *bp, int pageno, vm_page_t m);
#if 0
static void vfs_dirty_one_page(struct buf *bp, int pageno, vm_page_t m);
#endif
static void vfs_vmio_release(struct buf *bp);
static int flushbufqueues(struct buf *marker, bufq_type_t q);
static vm_page_t bio_page_alloc(struct buf *bp, vm_object_t obj,
                                vm_pindex_t pg, int deficit);

static void bd_signal(long totalspace);
static void buf_daemon(void);
static void buf_daemon_hw(void);

/*
 * bogus page -- for I/O to/from partially complete buffers
 * this is a temporary solution to the problem, but it is not
 * really that bad.  it would be better to split the buffer
 * for input in the case of buffers partially already in memory,
 * but the code is intricate enough already.
 */
vm_page_t bogus_page;

/*
 * These are all static, but make the ones we export globals so we do
 * not need to use compiler magic.
 */
long bufspace;                  /* atomic ops */
long maxbufspace;
static long bufmallocspace;     /* atomic ops */
long maxbufmallocspace, lobufspace, hibufspace;
static long lorunningspace;
static long hirunningspace;
static long dirtykvaspace;              /* atomic */
long dirtybufspace;                     /* atomic (global for systat) */
static long dirtybufcount;              /* atomic */
static long dirtybufspacehw;            /* atomic */
static long dirtybufcounthw;            /* atomic */
static long runningbufspace;            /* atomic */
static long runningbufcount;            /* atomic */
long lodirtybufspace;
long hidirtybufspace;
static int getnewbufcalls;
static int recoverbufcalls;
static int needsbuffer;                 /* atomic */
static int runningbufreq;               /* atomic */
static int bd_request;                  /* atomic */
static int bd_request_hw;               /* atomic */
static u_int bd_wake_ary[BD_WAKE_SIZE];
static u_int bd_wake_index;
static u_int vm_cycle_point = 40; /* 23-36 will migrate more act->inact */
static int debug_commit;
static int debug_bufbio;
static long bufcache_bw = 200 * 1024 * 1024;

static struct thread *bufdaemon_td;
static struct thread *bufdaemonhw_td;
static u_int lowmempgallocs;
static u_int lowmempgfails;
static u_int flushperqueue = 1024;

/*
 * Sysctls for operational control of the buffer cache.
 */
SYSCTL_UINT(_vfs, OID_AUTO, flushperqueue, CTLFLAG_RW, &flushperqueue, 0,
        "Number of buffers to flush from each per-cpu queue");
SYSCTL_LONG(_vfs, OID_AUTO, lodirtybufspace, CTLFLAG_RW, &lodirtybufspace, 0,
        "Number of dirty buffers to flush before bufdaemon becomes inactive");
SYSCTL_LONG(_vfs, OID_AUTO, hidirtybufspace, CTLFLAG_RW, &hidirtybufspace, 0,
        "High watermark used to trigger explicit flushing of dirty buffers");
SYSCTL_LONG(_vfs, OID_AUTO, lorunningspace, CTLFLAG_RW, &lorunningspace, 0,
        "Minimum amount of buffer space required for active I/O");
SYSCTL_LONG(_vfs, OID_AUTO, hirunningspace, CTLFLAG_RW, &hirunningspace, 0,
        "Maximum amount of buffer space to usable for active I/O");
SYSCTL_LONG(_vfs, OID_AUTO, bufcache_bw, CTLFLAG_RW, &bufcache_bw, 0,
        "Buffer-cache -> VM page cache transfer bandwidth");
SYSCTL_UINT(_vfs, OID_AUTO, lowmempgallocs, CTLFLAG_RW, &lowmempgallocs, 0,
        "Page allocations done during periods of very low free memory");
SYSCTL_UINT(_vfs, OID_AUTO, lowmempgfails, CTLFLAG_RW, &lowmempgfails, 0,
        "Page allocations which failed during periods of very low free memory");
SYSCTL_UINT(_vfs, OID_AUTO, vm_cycle_point, CTLFLAG_RW, &vm_cycle_point, 0,
        "Recycle pages to active or inactive queue transition pt 0-64");
/*
 * Sysctls determining current state of the buffer cache.
 */
SYSCTL_LONG(_vfs, OID_AUTO, nbuf, CTLFLAG_RD, &nbuf, 0,
        "Total number of buffers in buffer cache");
SYSCTL_LONG(_vfs, OID_AUTO, dirtykvaspace, CTLFLAG_RD, &dirtykvaspace, 0,
        "KVA reserved by dirty buffers (all)");
SYSCTL_LONG(_vfs, OID_AUTO, dirtybufspace, CTLFLAG_RD, &dirtybufspace, 0,
        "Pending bytes of dirty buffers (all)");
SYSCTL_LONG(_vfs, OID_AUTO, dirtybufspacehw, CTLFLAG_RD, &dirtybufspacehw, 0,
        "Pending bytes of dirty buffers (heavy weight)");
SYSCTL_LONG(_vfs, OID_AUTO, dirtybufcount, CTLFLAG_RD, &dirtybufcount, 0,
        "Pending number of dirty buffers");
SYSCTL_LONG(_vfs, OID_AUTO, dirtybufcounthw, CTLFLAG_RD, &dirtybufcounthw, 0,
        "Pending number of dirty buffers (heavy weight)");
SYSCTL_LONG(_vfs, OID_AUTO, runningbufspace, CTLFLAG_RD, &runningbufspace, 0,
        "I/O bytes currently in progress due to asynchronous writes");
SYSCTL_LONG(_vfs, OID_AUTO, runningbufcount, CTLFLAG_RD, &runningbufcount, 0,
        "I/O buffers currently in progress due to asynchronous writes");
SYSCTL_LONG(_vfs, OID_AUTO, maxbufspace, CTLFLAG_RD, &maxbufspace, 0,
        "Hard limit on maximum amount of memory usable for buffer space");
SYSCTL_LONG(_vfs, OID_AUTO, hibufspace, CTLFLAG_RD, &hibufspace, 0,
        "Soft limit on maximum amount of memory usable for buffer space");
SYSCTL_LONG(_vfs, OID_AUTO, lobufspace, CTLFLAG_RD, &lobufspace, 0,
        "Minimum amount of memory to reserve for system buffer space");
SYSCTL_LONG(_vfs, OID_AUTO, bufspace, CTLFLAG_RD, &bufspace, 0,
        "Amount of memory available for buffers");
SYSCTL_LONG(_vfs, OID_AUTO, maxmallocbufspace, CTLFLAG_RD, &maxbufmallocspace,
        0, "Maximum amount of memory reserved for buffers using malloc");
SYSCTL_LONG(_vfs, OID_AUTO, bufmallocspace, CTLFLAG_RD, &bufmallocspace, 0,
        "Amount of memory left for buffers using malloc-scheme");
SYSCTL_INT(_vfs, OID_AUTO, getnewbufcalls, CTLFLAG_RD, &getnewbufcalls, 0,
        "New buffer header acquisition requests");
SYSCTL_INT(_vfs, OID_AUTO, recoverbufcalls, CTLFLAG_RD, &recoverbufcalls, 0,
        "Recover VM space in an emergency");
SYSCTL_INT(_vfs, OID_AUTO, debug_commit, CTLFLAG_RW, &debug_commit, 0, "");
SYSCTL_INT(_vfs, OID_AUTO, debug_bufbio, CTLFLAG_RW, &debug_bufbio, 0, "");
SYSCTL_INT(_debug_sizeof, OID_AUTO, buf, CTLFLAG_RD, 0, sizeof(struct buf),
        "sizeof(struct buf)");

char *buf_wmesg = BUF_WMESG;

#define VFS_BIO_NEED_ANY        0x01    /* any freeable buffer */
#define VFS_BIO_NEED_UNUSED02   0x02
#define VFS_BIO_NEED_UNUSED04   0x04
#define VFS_BIO_NEED_BUFSPACE   0x08    /* wait for buf space, lo hysteresis */

/*
 * Called when buffer space is potentially available for recovery.
 * getnewbuf() will block on this flag when it is unable to free
 * sufficient buffer space.  Buffer space becomes recoverable when
 * bp's get placed back in the queues.
 */
static __inline void
bufspacewakeup(void)
{
        /*
         * If someone is waiting for BUF space, wake them up.  Even
         * though we haven't freed the kva space yet, the waiting
         * process will be able to now.
         */
        for (;;) {
                int flags = needsbuffer;
                cpu_ccfence();
                if ((flags & VFS_BIO_NEED_BUFSPACE) == 0)
                        break;
                if (atomic_cmpset_int(&needsbuffer, flags,
                                      flags & ~VFS_BIO_NEED_BUFSPACE)) {
                        wakeup(&needsbuffer);
                        break;
                }
                /* retry */
        }
}

/*
 * runningbufwakeup:
 *
 *      Accounting for I/O in progress.
 *
 */
static __inline void
runningbufwakeup(struct buf *bp)
{
        long totalspace;
        long flags;

        if ((totalspace = bp->b_runningbufspace) != 0) {
                atomic_add_long(&runningbufspace, -totalspace);
                atomic_add_long(&runningbufcount, -1);
                bp->b_runningbufspace = 0;

                /*
                 * see waitrunningbufspace() for limit test.
                 */
                for (;;) {
                        flags = runningbufreq;
                        cpu_ccfence();
                        if (flags == 0)
                                break;
                        if (atomic_cmpset_int(&runningbufreq, flags, 0)) {
                                wakeup(&runningbufreq);
                                break;
                        }
                        /* retry */
                }
                bd_signal(totalspace);
        }
}

/*
 * bufcountwakeup:
 *
 *      Called when a buffer has been added to one of the free queues to
 *      account for the buffer and to wakeup anyone waiting for free buffers.
 *      This typically occurs when large amounts of metadata are being handled
 *      by the buffer cache ( else buffer space runs out first, usually ).
 */
static __inline void
bufcountwakeup(void) 
{
        long flags;

        for (;;) {
                flags = needsbuffer;
                if (flags == 0)
                        break;
                if (atomic_cmpset_int(&needsbuffer, flags,
                                      (flags & ~VFS_BIO_NEED_ANY))) {
                        wakeup(&needsbuffer);
                        break;
                }
                /* retry */
        }
}

/*
 * waitrunningbufspace()
 *
 * If runningbufspace exceeds 4/6 hirunningspace we block until
 * runningbufspace drops to 3/6 hirunningspace.  We also block if another
 * thread blocked here in order to be fair, even if runningbufspace
 * is now lower than the limit.
 *
 * The caller may be using this function to block in a tight loop, we
 * must block while runningbufspace is greater than at least
 * hirunningspace * 3 / 6.
 */
void
waitrunningbufspace(void)
{
        long limit = hirunningspace * 4 / 6;
        long flags;

        while (runningbufspace > limit || runningbufreq) {
                tsleep_interlock(&runningbufreq, 0);
                flags = atomic_fetchadd_int(&runningbufreq, 1);
                if (runningbufspace > limit || flags)
                        tsleep(&runningbufreq, PINTERLOCKED, "wdrn1", hz);
        }
}

/*
 * buf_dirty_count_severe:
 *
 *      Return true if we have too many dirty buffers.
 */
int
buf_dirty_count_severe(void)
{
        return (runningbufspace + dirtykvaspace >= hidirtybufspace ||
                dirtybufcount >= nbuf / 2);
}

/*
 * Return true if the amount of running I/O is severe and BIOQ should
 * start bursting.
 */
int
buf_runningbufspace_severe(void)
{
        return (runningbufspace >= hirunningspace * 4 / 6);
}

/*
 * vfs_buf_test_cache:
 *
 * Called when a buffer is extended.  This function clears the B_CACHE
 * bit if the newly extended portion of the buffer does not contain
 * valid data.
 *
 * NOTE! Dirty VM pages are not processed into dirty (B_DELWRI) buffer
 * cache buffers.  The VM pages remain dirty, as someone had mmap()'d
 * them while a clean buffer was present.
 */
static __inline__
void
vfs_buf_test_cache(struct buf *bp,
                  vm_ooffset_t foff, vm_offset_t off, vm_offset_t size,
                  vm_page_t m)
{
        if (bp->b_flags & B_CACHE) {
                int base = (foff + off) & PAGE_MASK;
                if (vm_page_is_valid(m, base, size) == 0)
                        bp->b_flags &= ~B_CACHE;
        }
}

/*
 * bd_speedup()
 *
 * Spank the buf_daemon[_hw] if the total dirty buffer space exceeds the
 * low water mark.
 */
static __inline__
void
bd_speedup(void)
{
        if (dirtykvaspace < lodirtybufspace && dirtybufcount < nbuf / 2)
                return;

        if (bd_request == 0 &&
            (dirtykvaspace > lodirtybufspace / 2 ||
             dirtybufcount - dirtybufcounthw >= nbuf / 2)) {
                if (atomic_fetchadd_int(&bd_request, 1) == 0)
                        wakeup(&bd_request);
        }
        if (bd_request_hw == 0 &&
            (dirtykvaspace > lodirtybufspace / 2 ||
             dirtybufcounthw >= nbuf / 2)) {
                if (atomic_fetchadd_int(&bd_request_hw, 1) == 0)
                        wakeup(&bd_request_hw);
        }
}

/*
 * bd_heatup()
 *
 *      Get the buf_daemon heated up when the number of running and dirty
 *      buffers exceeds the mid-point.
 *
 *      Return the total number of dirty bytes past the second mid point
 *      as a measure of how much excess dirty data there is in the system.
 */
long
bd_heatup(void)
{
        long mid1;
        long mid2;
        long totalspace;

        mid1 = lodirtybufspace + (hidirtybufspace - lodirtybufspace) / 2;

        totalspace = runningbufspace + dirtykvaspace;
        if (totalspace >= mid1 || dirtybufcount >= nbuf / 2) {
                bd_speedup();
                mid2 = mid1 + (hidirtybufspace - mid1) / 2;
                if (totalspace >= mid2)
                        return(totalspace - mid2);
        }
        return(0);
}

/*
 * bd_wait()
 *
 *      Wait for the buffer cache to flush (totalspace) bytes worth of
 *      buffers, then return.
 *
 *      Regardless this function blocks while the number of dirty buffers
 *      exceeds hidirtybufspace.
 */
void
bd_wait(long totalspace)
{
        u_int i;
        u_int j;
        u_int mi;
        int count;

        if (curthread == bufdaemonhw_td || curthread == bufdaemon_td)
                return;

        while (totalspace > 0) {
                bd_heatup();

                /*
                 * Order is important.  Suppliers adjust bd_wake_index after
                 * updating runningbufspace/dirtykvaspace.  We want to fetch
                 * bd_wake_index before accessing.  Any error should thus
                 * be in our favor.
                 */
                i = atomic_fetchadd_int(&bd_wake_index, 0);
                if (totalspace > runningbufspace + dirtykvaspace)
                        totalspace = runningbufspace + dirtykvaspace;
                count = totalspace / MAXBSIZE;
                if (count >= BD_WAKE_SIZE / 2)
                        count = BD_WAKE_SIZE / 2;
                i = i + count;
                mi = i & BD_WAKE_MASK;

                /*
                 * This is not a strict interlock, so we play a bit loose
                 * with locking access to dirtybufspace*.  We have to re-check
                 * bd_wake_index to ensure that it hasn't passed us.
                 */
                tsleep_interlock(&bd_wake_ary[mi], 0);
                atomic_add_int(&bd_wake_ary[mi], 1);
                j = atomic_fetchadd_int(&bd_wake_index, 0);
                if ((int)(i - j) >= 0)
                        tsleep(&bd_wake_ary[mi], PINTERLOCKED, "flstik", hz);

                totalspace = runningbufspace + dirtykvaspace - hidirtybufspace;
        }
}

/*
 * bd_signal()
 * 
 *      This function is called whenever runningbufspace or dirtykvaspace
 *      is reduced.  Track threads waiting for run+dirty buffer I/O
 *      complete.
 */
static void
bd_signal(long totalspace)
{
        u_int i;

        if (totalspace > 0) {
                if (totalspace > MAXBSIZE * BD_WAKE_SIZE)
                        totalspace = MAXBSIZE * BD_WAKE_SIZE;
                while (totalspace > 0) {
                        i = atomic_fetchadd_int(&bd_wake_index, 1);
                        i &= BD_WAKE_MASK;
                        if (atomic_readandclear_int(&bd_wake_ary[i]))
                                wakeup(&bd_wake_ary[i]);
                        totalspace -= MAXBSIZE;
                }
        }
}

/*
 * BIO tracking support routines.
 *
 * Release a ref on a bio_track.  Wakeup requests are atomically released
 * along with the last reference so bk_active will never wind up set to
 * only 0x80000000.
 */
static
void
bio_track_rel(struct bio_track *track)
{
        int     active;
        int     desired;

        /*
         * Shortcut
         */
        active = track->bk_active;
        if (active == 1 && atomic_cmpset_int(&track->bk_active, 1, 0))
                return;

        /*
         * Full-on.  Note that the wait flag is only atomically released on
         * the 1->0 count transition.
         *
         * We check for a negative count transition using bit 30 since bit 31
         * has a different meaning.
         */
        for (;;) {
                desired = (active & 0x7FFFFFFF) - 1;
                if (desired)
                        desired |= active & 0x80000000;
                if (atomic_cmpset_int(&track->bk_active, active, desired)) {
                        if (desired & 0x40000000)
                                panic("bio_track_rel: bad count: %p", track);
                        if (active & 0x80000000)
                                wakeup(track);
                        break;
                }
                active = track->bk_active;
        }
}

/*
 * Wait for the tracking count to reach 0.
 *
 * Use atomic ops such that the wait flag is only set atomically when
 * bk_active is non-zero.
 */
int
bio_track_wait(struct bio_track *track, int slp_flags, int slp_timo)
{
        int     active;
        int     desired;
        int     error;

        /*
         * Shortcut
         */
        if (track->bk_active == 0)
                return(0);

        /*
         * Full-on.  Note that the wait flag may only be atomically set if
         * the active count is non-zero.
         *
         * NOTE: We cannot optimize active == desired since a wakeup could
         *       clear active prior to our tsleep_interlock().
         */
        error = 0;
        while ((active = track->bk_active) != 0) {
                cpu_ccfence();
                desired = active | 0x80000000;
                tsleep_interlock(track, slp_flags);
                if (atomic_cmpset_int(&track->bk_active, active, desired)) {
                        error = tsleep(track, slp_flags | PINTERLOCKED,
                                       "trwait", slp_timo);
                        if (error)
                                break;
                }
        }
        return (error);
}

/*
 * bufinit:
 *
 *      Load time initialisation of the buffer cache, called from machine
 *      dependant initialization code. 
 */
static
void
bufinit(void *dummy __unused)
{
        struct bufpcpu *pcpu;
        struct buf *bp;
        vm_offset_t bogus_offset;
        int i;
        int j;
        long n;

        /* next, make a null set of free lists */
        for (i = 0; i < ncpus; ++i) {
                pcpu = &bufpcpu[i];
                spin_init(&pcpu->spin, "bufinit");
                for (j = 0; j < BUFFER_QUEUES; j++)
                        TAILQ_INIT(&pcpu->bufqueues[j]);
        }

        /*
         * Finally, initialize each buffer header and stick on empty q.
         * Each buffer gets its own KVA reservation.
         */
        i = 0;
        pcpu = &bufpcpu[i];

        for (n = 0; n < nbuf; n++) {
                bp = &buf[n];
                bzero(bp, sizeof *bp);
                bp->b_flags = B_INVAL;  /* we're just an empty header */
                bp->b_cmd = BUF_CMD_DONE;
                bp->b_qindex = BQUEUE_EMPTY;
                bp->b_qcpu = i;
                bp->b_kvabase = (void *)(vm_map_min(&buffer_map) +
                                         MAXBSIZE * n);
                bp->b_kvasize = MAXBSIZE;
                initbufbio(bp);
                xio_init(&bp->b_xio);
                buf_dep_init(bp);
                TAILQ_INSERT_TAIL(&pcpu->bufqueues[bp->b_qindex],
                                  bp, b_freelist);

                i = (i + 1) % ncpus;
                pcpu = &bufpcpu[i];
        }

        /*
         * maxbufspace is the absolute maximum amount of buffer space we are 
         * allowed to reserve in KVM and in real terms.  The absolute maximum
         * is nominally used by buf_daemon.  hibufspace is the nominal maximum
         * used by most other processes.  The differential is required to 
         * ensure that buf_daemon is able to run when other processes might 
         * be blocked waiting for buffer space.
         *
         * Calculate hysteresis (lobufspace, hibufspace).  Don't make it
         * too large or we might lockup a cpu for too long a period of
         * time in our tight loop.
         */
        maxbufspace = nbuf * NBUFCALCSIZE;
        hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - MAXBSIZE * 10);
        lobufspace = hibufspace * 7 / 8;
        if (hibufspace - lobufspace > 64 * 1024 * 1024)
                lobufspace = hibufspace - 64 * 1024 * 1024;
        if (lobufspace > hibufspace - MAXBSIZE)
                lobufspace = hibufspace - MAXBSIZE;

        lorunningspace = 512 * 1024;
        /* hirunningspace -- see below */

        /*
         * Limit the amount of malloc memory since it is wired permanently
         * into the kernel space.  Even though this is accounted for in
         * the buffer allocation, we don't want the malloced region to grow
         * uncontrolled.  The malloc scheme improves memory utilization
         * significantly on average (small) directories.
         */
        maxbufmallocspace = hibufspace / 20;

        /*
         * Reduce the chance of a deadlock occuring by limiting the number
         * of delayed-write dirty buffers we allow to stack up.
         *
         * We don't want too much actually queued to the device at once
         * (XXX this needs to be per-mount!), because the buffers will
         * wind up locked for a very long period of time while the I/O
         * drains.
         */
        hidirtybufspace = hibufspace / 2;       /* dirty + running */
        hirunningspace = hibufspace / 16;       /* locked & queued to device */
        if (hirunningspace < 1024 * 1024)
                hirunningspace = 1024 * 1024;

        dirtykvaspace = 0;
        dirtybufspace = 0;
        dirtybufspacehw = 0;

        lodirtybufspace = hidirtybufspace / 2;

        /*
         * Maximum number of async ops initiated per buf_daemon loop.  This is
         * somewhat of a hack at the moment, we really need to limit ourselves
         * based on the number of bytes of I/O in-transit that were initiated
         * from buf_daemon.
         */

        bogus_offset = kmem_alloc_pageable(&kernel_map, PAGE_SIZE,
                                           VM_SUBSYS_BOGUS);
        vm_object_hold(&kernel_object);
        bogus_page = vm_page_alloc(&kernel_object,
                                   (bogus_offset >> PAGE_SHIFT),
                                   VM_ALLOC_NORMAL);
        vm_object_drop(&kernel_object);
        vmstats.v_wire_count++;

}

SYSINIT(do_bufinit, SI_BOOT2_MACHDEP, SI_ORDER_FIRST, bufinit, NULL);

/*
 * Initialize the embedded bio structures, typically used by
 * deprecated code which tries to allocate its own struct bufs.
 */
void
initbufbio(struct buf *bp)
{
        bp->b_bio1.bio_buf = bp;
        bp->b_bio1.bio_prev = NULL;
        bp->b_bio1.bio_offset = NOOFFSET;
        bp->b_bio1.bio_next = &bp->b_bio2;
        bp->b_bio1.bio_done = NULL;
        bp->b_bio1.bio_flags = 0;

        bp->b_bio2.bio_buf = bp;
        bp->b_bio2.bio_prev = &bp->b_bio1;
        bp->b_bio2.bio_offset = NOOFFSET;
        bp->b_bio2.bio_next = NULL;
        bp->b_bio2.bio_done = NULL;
        bp->b_bio2.bio_flags = 0;

        BUF_LOCKINIT(bp);
}

/*
 * Reinitialize the embedded bio structures as well as any additional
 * translation cache layers.
 */
void
reinitbufbio(struct buf *bp)
{
        struct bio *bio;

        for (bio = &bp->b_bio1; bio; bio = bio->bio_next) {
                bio->bio_done = NULL;
                bio->bio_offset = NOOFFSET;
        }
}

/*
 * Undo the effects of an initbufbio().
 */
void
uninitbufbio(struct buf *bp)
{
        dsched_buf_exit(bp);
        BUF_LOCKFREE(bp);
}

/*
 * Push another BIO layer onto an existing BIO and return it.  The new
 * BIO layer may already exist, holding cached translation data.
 */
struct bio *
push_bio(struct bio *bio)
{
        struct bio *nbio;

        if ((nbio = bio->bio_next) == NULL) {
                int index = bio - &bio->bio_buf->b_bio_array[0];
                if (index >= NBUF_BIO - 1) {
                        panic("push_bio: too many layers %d for bp %p",
                                index, bio->bio_buf);
                }
                nbio = &bio->bio_buf->b_bio_array[index + 1];
                bio->bio_next = nbio;
                nbio->bio_prev = bio;
                nbio->bio_buf = bio->bio_buf;
                nbio->bio_offset = NOOFFSET;
                nbio->bio_done = NULL;
                nbio->bio_next = NULL;
        }
        KKASSERT(nbio->bio_done == NULL);
        return(nbio);
}

/*
 * Pop a BIO translation layer, returning the previous layer.  The
 * must have been previously pushed.
 */
struct bio *
pop_bio(struct bio *bio)
{
        return(bio->bio_prev);
}

void
clearbiocache(struct bio *bio)
{
        while (bio) {
                bio->bio_offset = NOOFFSET;
                bio = bio->bio_next;
        }
}

/*
 * Remove the buffer from the appropriate free list.
 * (caller must be locked)
 */
static __inline void
_bremfree(struct buf *bp)
{
        struct bufpcpu *pcpu = &bufpcpu[bp->b_qcpu];

        if (bp->b_qindex != BQUEUE_NONE) {
                KASSERT(BUF_REFCNTNB(bp) == 1, 
                        ("bremfree: bp %p not locked",bp));
                TAILQ_REMOVE(&pcpu->bufqueues[bp->b_qindex], bp, b_freelist);
                bp->b_qindex = BQUEUE_NONE;
        } else {
                if (BUF_REFCNTNB(bp) <= 1)
                        panic("bremfree: removing a buffer not on a queue");
        }
}

/*
 * bremfree() - must be called with a locked buffer
 */
void
bremfree(struct buf *bp)
{
        struct bufpcpu *pcpu = &bufpcpu[bp->b_qcpu];

        spin_lock(&pcpu->spin);
        _bremfree(bp);
        spin_unlock(&pcpu->spin);
}

/*
 * bremfree_locked - must be called with pcpu->spin locked
 */
static void
bremfree_locked(struct buf *bp)
{
        _bremfree(bp);
}

/*
 * This version of bread issues any required I/O asyncnronously and
 * makes a callback on completion.
 *
 * The callback must check whether BIO_DONE is set in the bio and issue
 * the bpdone(bp, 0) if it isn't.  The callback is responsible for clearing
 * BIO_DONE and disposing of the I/O (bqrelse()ing it).
 */
void
breadcb(struct vnode *vp, off_t loffset, int size, int bflags,
        void (*func)(struct bio *), void *arg)
{
        struct buf *bp;

        bp = getblk(vp, loffset, size, 0, 0);

        /* if not found in cache, do some I/O */
        if ((bp->b_flags & B_CACHE) == 0) {
                bp->b_flags &= ~(B_ERROR | B_EINTR | B_INVAL | B_NOTMETA);
                bp->b_flags |= bflags;
                bp->b_cmd = BUF_CMD_READ;
                bp->b_bio1.bio_done = func;
                bp->b_bio1.bio_caller_info1.ptr = arg;
                vfs_busy_pages(vp, bp);
                BUF_KERNPROC(bp);
                vn_strategy(vp, &bp->b_bio1);
        } else if (func) {
                /*
                 * Since we are issuing the callback synchronously it cannot
                 * race the BIO_DONE, so no need for atomic ops here.
                 */
                /*bp->b_bio1.bio_done = func;*/
                bp->b_bio1.bio_caller_info1.ptr = arg;
                bp->b_bio1.bio_flags |= BIO_DONE;
                func(&bp->b_bio1);
        } else {
                bqrelse(bp);
        }
}

/*
 * breadnx() - Terminal function for bread() and breadn().
 *
 * This function will start asynchronous I/O on read-ahead blocks as well
 * as satisfy the primary request.
 *
 * We must clear B_ERROR and B_INVAL prior to initiating I/O.  If B_CACHE is
 * set, the buffer is valid and we do not have to do anything.
 */
int
breadnx(struct vnode *vp, off_t loffset, int size, int bflags,
        off_t *raoffset, int *rabsize,
        int cnt, struct buf **bpp)
{
        struct buf *bp, *rabp;
        int i;
        int rv = 0, readwait = 0;
        int blkflags = (bflags & B_KVABIO) ? GETBLK_KVABIO : 0;

        if (*bpp)
                bp = *bpp;
        else
                *bpp = bp = getblk(vp, loffset, size, blkflags, 0);

        /* if not found in cache, do some I/O */
        if ((bp->b_flags & B_CACHE) == 0) {
                bp->b_flags &= ~(B_ERROR | B_EINTR | B_INVAL | B_NOTMETA);
                bp->b_flags |= bflags;
                bp->b_cmd = BUF_CMD_READ;
                bp->b_bio1.bio_done = biodone_sync;
                bp->b_bio1.bio_flags |= BIO_SYNC;
                vfs_busy_pages(vp, bp);
                vn_strategy(vp, &bp->b_bio1);
                ++readwait;
        }

        for (i = 0; i < cnt; i++, raoffset++, rabsize++) {
                if (inmem(vp, *raoffset))
                        continue;
                rabp = getblk(vp, *raoffset, *rabsize, GETBLK_KVABIO, 0);

                if ((rabp->b_flags & B_CACHE) == 0) {
                        rabp->b_flags &= ~(B_ERROR | B_EINTR |
                                           B_INVAL | B_NOTMETA);
                        rabp->b_flags |= (bflags & ~B_KVABIO);
                        rabp->b_cmd = BUF_CMD_READ;
                        vfs_busy_pages(vp, rabp);
                        BUF_KERNPROC(rabp);
                        vn_strategy(vp, &rabp->b_bio1);
                } else {
                        brelse(rabp);
                }
        }
        if (readwait)
                rv = biowait(&bp->b_bio1, "biord");
        return (rv);
}

/*
 * bwrite:
 *
 *      Synchronous write, waits for completion.
 *
 *      Write, release buffer on completion.  (Done by iodone
 *      if async).  Do not bother writing anything if the buffer
 *      is invalid.
 *
 *      Note that we set B_CACHE here, indicating that buffer is
 *      fully valid and thus cacheable.  This is true even of NFS
 *      now so we set it generally.  This could be set either here 
 *      or in biodone() since the I/O is synchronous.  We put it
 *      here.
 */
int
bwrite(struct buf *bp)
{
        int error;

        if (bp->b_flags & B_INVAL) {
                brelse(bp);
                return (0);
        }
        if (BUF_REFCNTNB(bp) == 0)
                panic("bwrite: buffer is not busy???");

        /*
         * NOTE: We no longer mark the buffer clear prior to the vn_strategy()
         *       call because it will remove the buffer from the vnode's
         *       dirty buffer list prematurely and possibly cause filesystem
         *       checks to race buffer flushes.  This is now handled in
         *       bpdone().
         *
         *       bundirty(bp); REMOVED
         */

        bp->b_flags &= ~(B_ERROR | B_EINTR);
        bp->b_flags |= B_CACHE;
        bp->b_cmd = BUF_CMD_WRITE;
        bp->b_bio1.bio_done = biodone_sync;
        bp->b_bio1.bio_flags |= BIO_SYNC;
        vfs_busy_pages(bp->b_vp, bp);

        /*
         * Normal bwrites pipeline writes.  NOTE: b_bufsize is only
         * valid for vnode-backed buffers.
         */
        bsetrunningbufspace(bp, bp->b_bufsize);
        vn_strategy(bp->b_vp, &bp->b_bio1);
        error = biowait(&bp->b_bio1, "biows");
        brelse(bp);

        return (error);
}

/*
 * bawrite:
 *
 *      Asynchronous write.  Start output on a buffer, but do not wait for
 *      it to complete.  The buffer is released when the output completes.
 *
 *      bwrite() ( or the VOP routine anyway ) is responsible for handling
 *      B_INVAL buffers.  Not us.
 */
void
bawrite(struct buf *bp)
{
        if (bp->b_flags & B_INVAL) {
                brelse(bp);
                return;
        }
        if (BUF_REFCNTNB(bp) == 0)
                panic("bawrite: buffer is not busy???");

        /*
         * NOTE: We no longer mark the buffer clear prior to the vn_strategy()
         *       call because it will remove the buffer from the vnode's
         *       dirty buffer list prematurely and possibly cause filesystem
         *       checks to race buffer flushes.  This is now handled in
         *       bpdone().
         *
         *       bundirty(bp); REMOVED
         */
        bp->b_flags &= ~(B_ERROR | B_EINTR);
        bp->b_flags |= B_CACHE;
        bp->b_cmd = BUF_CMD_WRITE;
        KKASSERT(bp->b_bio1.bio_done == NULL);
        vfs_busy_pages(bp->b_vp, bp);

        /*
         * Normal bwrites pipeline writes.  NOTE: b_bufsize is only
         * valid for vnode-backed buffers.
         */
        bsetrunningbufspace(bp, bp->b_bufsize);
        BUF_KERNPROC(bp);
        vn_strategy(bp->b_vp, &bp->b_bio1);
}

/*
 * bdwrite:
 *
 *      Delayed write. (Buffer is marked dirty).  Do not bother writing
 *      anything if the buffer is marked invalid.
 *
 *      Note that since the buffer must be completely valid, we can safely
 *      set B_CACHE.  In fact, we have to set B_CACHE here rather then in
 *      biodone() in order to prevent getblk from writing the buffer
 *      out synchronously.
 */
void
bdwrite(struct buf *bp)
{
        if (BUF_REFCNTNB(bp) == 0)
                panic("bdwrite: buffer is not busy");

        if (bp->b_flags & B_INVAL) {
                brelse(bp);
                return;
        }
        bdirty(bp);

        dsched_buf_enter(bp);   /* might stack */

        /*
         * Set B_CACHE, indicating that the buffer is fully valid.  This is
         * true even of NFS now.
         */
        bp->b_flags |= B_CACHE;

        /*
         * This bmap keeps the system from needing to do the bmap later,
         * perhaps when the system is attempting to do a sync.  Since it
         * is likely that the indirect block -- or whatever other datastructure
         * that the filesystem needs is still in memory now, it is a good
         * thing to do this.  Note also, that if the pageout daemon is
         * requesting a sync -- there might not be enough memory to do
         * the bmap then...  So, this is important to do.
         */
        if (bp->b_bio2.bio_offset == NOOFFSET) {
                VOP_BMAP(bp->b_vp, bp->b_loffset, &bp->b_bio2.bio_offset,
                         NULL, NULL, BUF_CMD_WRITE);
        }

        /*
         * Because the underlying pages may still be mapped and
         * writable trying to set the dirty buffer (b_dirtyoff/end)
         * range here will be inaccurate.
         *
         * However, we must still clean the pages to satisfy the
         * vnode_pager and pageout daemon, so they think the pages
         * have been "cleaned".  What has really occured is that
         * they've been earmarked for later writing by the buffer
         * cache.
         *
         * So we get the b_dirtyoff/end update but will not actually
         * depend on it (NFS that is) until the pages are busied for
         * writing later on.
         */
        vfs_clean_pages(bp);
        bqrelse(bp);

        /*
         * note: we cannot initiate I/O from a bdwrite even if we wanted to,
         * due to the softdep code.
         */
}

/*
 * Fake write - return pages to VM system as dirty, leave the buffer clean.
 * This is used by tmpfs.
 *
 * It is important for any VFS using this routine to NOT use it for
 * IO_SYNC or IO_ASYNC operations which occur when the system really
 * wants to flush VM pages to backing store.
 */
void
buwrite(struct buf *bp)
{
        vm_page_t m;
        int i;

        /*
         * Only works for VMIO buffers.  If the buffer is already
         * marked for delayed-write we can't avoid the bdwrite().
         */
        if ((bp->b_flags & B_VMIO) == 0 || (bp->b_flags & B_DELWRI)) {
                bdwrite(bp);
                return;
        }

        /*
         * Mark as needing a commit.
         */
        for (i = 0; i < bp->b_xio.xio_npages; i++) {
                m = bp->b_xio.xio_pages[i];
                vm_page_need_commit(m);
        }
        bqrelse(bp);
}

/*
 * bdirty:
 *
 *      Turn buffer into delayed write request by marking it B_DELWRI.
 *      B_RELBUF and B_NOCACHE must be cleared.
 *
 *      We reassign the buffer to itself to properly update it in the
 *      dirty/clean lists. 
 *
 *      Must be called from a critical section.
 *      The buffer must be on BQUEUE_NONE.
 */
void
bdirty(struct buf *bp)
{
        KASSERT(bp->b_qindex == BQUEUE_NONE,
                ("bdirty: buffer %p still on queue %d", bp, bp->b_qindex));
        if (bp->b_flags & B_NOCACHE) {
                kprintf("bdirty: clearing B_NOCACHE on buf %p\n", bp);
                bp->b_flags &= ~B_NOCACHE;
        }
        if (bp->b_flags & B_INVAL) {
                kprintf("bdirty: warning, dirtying invalid buffer %p\n", bp);
        }
        bp->b_flags &= ~B_RELBUF;

        if ((bp->b_flags & B_DELWRI) == 0) {
                lwkt_gettoken(&bp->b_vp->v_token);
                bp->b_flags |= B_DELWRI;
                reassignbuf(bp);
                lwkt_reltoken(&bp->b_vp->v_token);

                atomic_add_long(&dirtybufcount, 1);
                atomic_add_long(&dirtykvaspace, bp->b_kvasize);
                atomic_add_long(&dirtybufspace, bp->b_bufsize);
                if (bp->b_flags & B_HEAVY) {
                        atomic_add_long(&dirtybufcounthw, 1);
                        atomic_add_long(&dirtybufspacehw, bp->b_bufsize);
                }
                bd_heatup();
        }
}

/*
 * Set B_HEAVY, indicating that this is a heavy-weight buffer that
 * needs to be flushed with a different buf_daemon thread to avoid
 * deadlocks.  B_HEAVY also imposes restrictions in getnewbuf().
 */
void
bheavy(struct buf *bp)
{
        if ((bp->b_flags & B_HEAVY) == 0) {
                bp->b_flags |= B_HEAVY;
                if (bp->b_flags & B_DELWRI) {
                        atomic_add_long(&dirtybufcounthw, 1);
                        atomic_add_long(&dirtybufspacehw, bp->b_bufsize);
                }
        }
}

/*
 * bundirty:
 *
 *      Clear B_DELWRI for buffer.
 *
 *      Must be called from a critical section.
 *
 *      The buffer is typically on BQUEUE_NONE but there is one case in 
 *      brelse() that calls this function after placing the buffer on
 *      a different queue.
 */
void
bundirty(struct buf *bp)
{
        if (bp->b_flags & B_DELWRI) {
                lwkt_gettoken(&bp->b_vp->v_token);
                bp->b_flags &= ~B_DELWRI;
                reassignbuf(bp);
                lwkt_reltoken(&bp->b_vp->v_token);

                atomic_add_long(&dirtybufcount, -1);
                atomic_add_long(&dirtykvaspace, -bp->b_kvasize);
                atomic_add_long(&dirtybufspace, -bp->b_bufsize);
                if (bp->b_flags & B_HEAVY) {
                        atomic_add_long(&dirtybufcounthw, -1);
                        atomic_add_long(&dirtybufspacehw, -bp->b_bufsize);
                }
                bd_signal(bp->b_bufsize);
        }
        /*
         * Since it is now being written, we can clear its deferred write flag.
         */
        bp->b_flags &= ~B_DEFERRED;
}

/*
 * Set the b_runningbufspace field, used to track how much I/O is
 * in progress at any given moment.
 */
void
bsetrunningbufspace(struct buf *bp, int bytes)
{
        bp->b_runningbufspace = bytes;
        if (bytes) {
                atomic_add_long(&runningbufspace, bytes);
                atomic_add_long(&runningbufcount, 1);
        }
}

/*
 * brelse:
 *
 *      Release a busy buffer and, if requested, free its resources.  The
 *      buffer will be stashed in the appropriate bufqueue[] allowing it
 *      to be accessed later as a cache entity or reused for other purposes.
 */
void
brelse(struct buf *bp)
{
        struct bufpcpu *pcpu;
#ifdef INVARIANTS
        int saved_flags = bp->b_flags;
#endif

        KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)),
                ("brelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp));

        /*
         * If B_NOCACHE is set we are being asked to destroy the buffer and
         * its backing store.  Clear B_DELWRI.
         *
         * B_NOCACHE is set in two cases: (1) when the caller really wants
         * to destroy the buffer and backing store and (2) when the caller
         * wants to destroy the buffer and backing store after a write 
         * completes.
         */
        if ((bp->b_flags & (B_NOCACHE|B_DELWRI)) == (B_NOCACHE|B_DELWRI)) {
                bundirty(bp);
        }

        if ((bp->b_flags & (B_INVAL | B_DELWRI)) == B_DELWRI) {
                /*
                 * A re-dirtied buffer is only subject to destruction
                 * by B_INVAL.  B_ERROR and B_NOCACHE are ignored.
                 */
                /* leave buffer intact */
        } else if ((bp->b_flags & (B_NOCACHE | B_INVAL | B_ERROR)) ||
                   (bp->b_bufsize <= 0)) {
                /*
                 * Either a failed read or we were asked to free or not
                 * cache the buffer.  This path is reached with B_DELWRI
                 * set only if B_INVAL is already set.  B_NOCACHE governs
                 * backing store destruction.
                 *
                 * NOTE: HAMMER will set B_LOCKED in buf_deallocate if the
                 * buffer cannot be immediately freed.
                 */
                bp->b_flags |= B_INVAL;
                if (LIST_FIRST(&bp->b_dep) != NULL)
                        buf_deallocate(bp);
                if (bp->b_flags & B_DELWRI) {
                        atomic_add_long(&dirtybufcount, -1);
                        atomic_add_long(&dirtykvaspace, -bp->b_kvasize);
                        atomic_add_long(&dirtybufspace, -bp->b_bufsize);
                        if (bp->b_flags & B_HEAVY) {
                                atomic_add_long(&dirtybufcounthw, -1);
                                atomic_add_long(&dirtybufspacehw,
                                                -bp->b_bufsize);
                        }
                        bd_signal(bp->b_bufsize);
                }
                bp->b_flags &= ~(B_DELWRI | B_CACHE);
        }

        /*
         * We must clear B_RELBUF if B_DELWRI or B_LOCKED is set,
         * or if b_refs is non-zero.
         *
         * If vfs_vmio_release() is called with either bit set, the
         * underlying pages may wind up getting freed causing a previous
         * write (bdwrite()) to get 'lost' because pages associated with
         * a B_DELWRI bp are marked clean.  Pages associated with a
         * B_LOCKED buffer may be mapped by the filesystem.
         *
         * If we want to release the buffer ourselves (rather then the
         * originator asking us to release it), give the originator a
         * chance to countermand the release by setting B_LOCKED.
         * 
         * We still allow the B_INVAL case to call vfs_vmio_release(), even
         * if B_DELWRI is set.
         *
         * If B_DELWRI is not set we may have to set B_RELBUF if we are low
         * on pages to return pages to the VM page queues.
         */
        if ((bp->b_flags & (B_DELWRI | B_LOCKED)) || bp->b_refs) {
                bp->b_flags &= ~B_RELBUF;
        } else if (vm_page_count_min(0)) {
                if (LIST_FIRST(&bp->b_dep) != NULL)
                        buf_deallocate(bp);             /* can set B_LOCKED */
                if (bp->b_flags & (B_DELWRI | B_LOCKED))
                        bp->b_flags &= ~B_RELBUF;
                else
                        bp->b_flags |= B_RELBUF;
        }

        /*
         * Make sure b_cmd is clear.  It may have already been cleared by
         * biodone().
         *
         * At this point destroying the buffer is governed by the B_INVAL 
         * or B_RELBUF flags.
         */
        bp->b_cmd = BUF_CMD_DONE;
        dsched_buf_exit(bp);

        /*
         * VMIO buffer rundown.  Make sure the VM page array is restored
         * after an I/O may have replaces some of the pages with bogus pages
         * in order to not destroy dirty pages in a fill-in read.
         *
         * Note that due to the code above, if a buffer is marked B_DELWRI
         * then the B_RELBUF and B_NOCACHE bits will always be clear.
         * B_INVAL may still be set, however.
         *
         * For clean buffers, B_INVAL or B_RELBUF will destroy the buffer
         * but not the backing store.   B_NOCACHE will destroy the backing
         * store.
         *
         * Note that dirty NFS buffers contain byte-granular write ranges
         * and should not be destroyed w/ B_INVAL even if the backing store
         * is left intact.
         */
        if (bp->b_flags & B_VMIO) {
                /*
                 * Rundown for VMIO buffers which are not dirty NFS buffers.
                 */
                int i, j, resid;
                vm_page_t m;
                off_t foff;
                vm_pindex_t poff;
                vm_object_t obj;
                struct vnode *vp;

                vp = bp->b_vp;

                /*
                 * Get the base offset and length of the buffer.  Note that 
                 * in the VMIO case if the buffer block size is not
                 * page-aligned then b_data pointer may not be page-aligned.
                 * But our b_xio.xio_pages array *IS* page aligned.
                 *
                 * block sizes less then DEV_BSIZE (usually 512) are not 
                 * supported due to the page granularity bits (m->valid,
                 * m->dirty, etc...). 
                 *
                 * See man buf(9) for more information
                 */

                resid = bp->b_bufsize;
                foff = bp->b_loffset;

                for (i = 0; i < bp->b_xio.xio_npages; i++) {
                        m = bp->b_xio.xio_pages[i];

                        /*
                         * If we hit a bogus page, fixup *all* of them
                         * now.  Note that we left these pages wired
                         * when we removed them so they had better exist,
                         * and they cannot be ripped out from under us so
                         * no critical section protection is necessary.
                         */
                        if (m == bogus_page) {
                                obj = vp->v_object;
                                poff = OFF_TO_IDX(bp->b_loffset);

                                vm_object_hold(obj);
                                for (j = i; j < bp->b_xio.xio_npages; j++) {
                                        vm_page_t mtmp;

                                        mtmp = bp->b_xio.xio_pages[j];
                                        if (mtmp == bogus_page) {
                                                if ((bp->b_flags & B_HASBOGUS) == 0)
                                                        panic("brelse: bp %p corrupt bogus", bp);
                                                mtmp = vm_page_lookup(obj, poff + j);
                                                if (!mtmp)
                                                        panic("brelse: bp %p page %d missing", bp, j);
                                                bp->b_xio.xio_pages[j] = mtmp;
                                        }
                                }
                                vm_object_drop(obj);

                                if ((bp->b_flags & B_HASBOGUS) ||
                                    (bp->b_flags & B_INVAL) == 0) {
                                        pmap_qenter_noinval(
                                            trunc_page((vm_offset_t)bp->b_data),
                                            bp->b_xio.xio_pages,
                                            bp->b_xio.xio_npages);
                                        bp->b_flags &= ~B_HASBOGUS;
                                        bp->b_flags |= B_KVABIO;
                                        bkvareset(bp);
                                }
                                m = bp->b_xio.xio_pages[i];
                        }

                        /*
                         * Invalidate the backing store if B_NOCACHE is set
                         * (e.g. used with vinvalbuf()).  If this is NFS
                         * we impose a requirement that the block size be
                         * a multiple of PAGE_SIZE and create a temporary
                         * hack to basically invalidate the whole page.  The
                         * problem is that NFS uses really odd buffer sizes
                         * especially when tracking piecemeal writes and
                         * it also vinvalbuf()'s a lot, which would result
                         * in only partial page validation and invalidation
                         * here.  If the file page is mmap()'d, however,
                         * all the valid bits get set so after we invalidate
                         * here we would end up with weird m->valid values
                         * like 0xfc.  nfs_getpages() can't handle this so
                         * we clear all the valid bits for the NFS case
                         * instead of just some of them.
                         *
                         * The real bug is the VM system having to set m->valid
                         * to VM_PAGE_BITS_ALL for faulted-in pages, which
                         * itself is an artifact of the whole 512-byte
                         * granular mess that exists to support odd block 
                         * sizes and UFS meta-data block sizes (e.g. 6144).
                         * A complete rewrite is required.
                         *
                         * XXX
                         */
                        if (bp->b_flags & (B_NOCACHE|B_ERROR)) {
                                int poffset = foff & PAGE_MASK;
                                int presid;

                                presid = PAGE_SIZE - poffset;
                                if (bp->b_vp->v_tag == VT_NFS &&
                                    bp->b_vp->v_type == VREG) {
                                        ; /* entire page */
                                } else if (presid > resid) {
                                        presid = resid;
                                }
                                KASSERT(presid >= 0, ("brelse: extra page"));
                                vm_page_set_invalid(m, poffset, presid);

                                /*
                                 * Also make sure any swap cache is removed
                                 * as it is now stale (HAMMER in particular
                                 * uses B_NOCACHE to deal with buffer
                                 * aliasing).
                                 */
                                swap_pager_unswapped(m);
                        }
                        resid -= PAGE_SIZE - (foff & PAGE_MASK);
                        foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK;
                }
                if (bp->b_flags & (B_INVAL | B_RELBUF))
                        vfs_vmio_release(bp);
        } else {
                /*
                 * Rundown for non-VMIO buffers.
                 */
                if (bp->b_flags & (B_INVAL | B_RELBUF)) {
                        if (bp->b_bufsize)
                                allocbuf(bp, 0);
                        KKASSERT (LIST_FIRST(&bp->b_dep) == NULL);
                        if (bp->b_vp)
                                brelvp(bp);
                }
        }
                        
        if (bp->b_qindex != BQUEUE_NONE)
                panic("brelse: free buffer onto another queue???");
        if (BUF_REFCNTNB(bp) > 1) {
                /* Temporary panic to verify exclusive locking */
                /* This panic goes away when we allow shared refs */
                panic("brelse: multiple refs");
                /* NOT REACHED */
                return;
        }

        /*
         * Figure out the correct queue to place the cleaned up buffer on.
         * Buffers placed in the EMPTY or EMPTYKVA had better already be
         * disassociated from their vnode.
         *
         * Return the buffer to its original pcpu area
         */
        pcpu = &bufpcpu[bp->b_qcpu];
        spin_lock(&pcpu->spin);

        if (bp->b_flags & B_LOCKED) {
                /*
                 * Buffers that are locked are placed in the locked queue
                 * immediately, regardless of their state.
                 */
                bp->b_qindex = BQUEUE_LOCKED;
                TAILQ_INSERT_TAIL(&pcpu->bufqueues[bp->b_qindex],
                                  bp, b_freelist);
        } else if (bp->b_bufsize == 0) {
                /*
                 * Buffers with no memory.  Due to conditionals near the top
                 * of brelse() such buffers should probably already be
                 * marked B_INVAL and disassociated from their vnode.
                 */
                bp->b_flags |= B_INVAL;
                KASSERT(bp->b_vp == NULL,
                        ("bp1 %p flags %08x/%08x vnode %p "
                         "unexpectededly still associated!",
                        bp, saved_flags, bp->b_flags, bp->b_vp));
                KKASSERT((bp->b_flags & B_HASHED) == 0);
                bp->b_qindex = BQUEUE_EMPTY;
                TAILQ_INSERT_HEAD(&pcpu->bufqueues[bp->b_qindex],
                                  bp, b_freelist);
        } else if (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF)) {
                /*
                 * Buffers with junk contents.   Again these buffers had better
                 * already be disassociated from their vnode.
                 */
                KASSERT(bp->b_vp == NULL,
                        ("bp2 %p flags %08x/%08x vnode %p unexpectededly "
                         "still associated!",
                        bp, saved_flags, bp->b_flags, bp->b_vp));
                KKASSERT((bp->b_flags & B_HASHED) == 0);
                bp->b_flags |= B_INVAL;
                bp->b_qindex = BQUEUE_CLEAN;
                TAILQ_INSERT_HEAD(&pcpu->bufqueues[bp->b_qindex],
                                  bp, b_freelist);
        } else {
                /*
                 * Remaining buffers.  These buffers are still associated with
                 * their vnode.
                 */
                switch(bp->b_flags & (B_DELWRI|B_HEAVY)) {
                case B_DELWRI:
                        bp->b_qindex = BQUEUE_DIRTY;
                        TAILQ_INSERT_TAIL(&pcpu->bufqueues[bp->b_qindex],
                                          bp, b_freelist);
                        break;
                case B_DELWRI | B_HEAVY:
                        bp->b_qindex = BQUEUE_DIRTY_HW;
                        TAILQ_INSERT_TAIL(&pcpu->bufqueues[bp->b_qindex],
                                          bp, b_freelist);
                        break;
                default:
                        /*
                         * NOTE: Buffers are always placed at the end of the
                         * queue.  If B_AGE is not set the buffer will cycle
                         * through the queue twice.
                         */
                        bp->b_qindex = BQUEUE_CLEAN;
                        TAILQ_INSERT_TAIL(&pcpu->bufqueues[bp->b_qindex],
                                          bp, b_freelist);
                        break;
                }
        }
        spin_unlock(&pcpu->spin);

        /*
         * If B_INVAL, clear B_DELWRI.  We've already placed the buffer
         * on the correct queue but we have not yet unlocked it.
         */
        if ((bp->b_flags & (B_INVAL|B_DELWRI)) == (B_INVAL|B_DELWRI))
                bundirty(bp);

        /*
         * The bp is on an appropriate queue unless locked.  If it is not
         * locked or dirty we can wakeup threads waiting for buffer space.
         *
         * We've already handled the B_INVAL case ( B_DELWRI will be clear
         * if B_INVAL is set ).
         */
        if ((bp->b_flags & (B_LOCKED|B_DELWRI)) == 0)
                bufcountwakeup();

        /*
         * Something we can maybe free or reuse
         */
        if (bp->b_bufsize || bp->b_kvasize)
                bufspacewakeup();

        /*
         * Clean up temporary flags and unlock the buffer.
         */
        bp->b_flags &= ~(B_NOCACHE | B_RELBUF | B_DIRECT);
        BUF_UNLOCK(bp);
}

/*
 * bqrelse:
 *
 *      Release a buffer back to the appropriate queue but do not try to free
 *      it.  The buffer is expected to be used again soon.
 *
 *      bqrelse() is used by bdwrite() to requeue a delayed write, and used by
 *      biodone() to requeue an async I/O on completion.  It is also used when
 *      known good buffers need to be requeued but we think we may need the data
 *      again soon.
 *
 *      XXX we should be able to leave the B_RELBUF hint set on completion.
 */
void
bqrelse(struct buf *bp)
{
        struct bufpcpu *pcpu;

        KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)),
                ("bqrelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp));

        if (bp->b_qindex != BQUEUE_NONE)
                panic("bqrelse: free buffer onto another queue???");
        if (BUF_REFCNTNB(bp) > 1) {
                /* do not release to free list */
                panic("bqrelse: multiple refs");
                return;
        }

        buf_act_advance(bp);

        pcpu = &bufpcpu[bp->b_qcpu];
        spin_lock(&pcpu->spin);

        if (bp->b_flags & B_LOCKED) {
                /*
                 * Locked buffers are released to the locked queue.  However,
                 * if the buffer is dirty it will first go into the dirty
                 * queue and later on after the I/O completes successfully it
                 * will be released to the locked queue.
                 */
                bp->b_qindex = BQUEUE_LOCKED;
                TAILQ_INSERT_TAIL(&pcpu->bufqueues[bp->b_qindex],
                                  bp, b_freelist);
        } else if (bp->b_flags & B_DELWRI) {
                bp->b_qindex = (bp->b_flags & B_HEAVY) ?
                               BQUEUE_DIRTY_HW : BQUEUE_DIRTY;
                TAILQ_INSERT_TAIL(&pcpu->bufqueues[bp->b_qindex],
                                  bp, b_freelist);
        } else if (vm_page_count_min(0)) {
                /*
                 * We are too low on memory, we have to try to free the
                 * buffer (most importantly: the wired pages making up its
                 * backing store) *now*.
                 */
                spin_unlock(&pcpu->spin);
                brelse(bp);
                return;
        } else {
                bp->b_qindex = BQUEUE_CLEAN;
                TAILQ_INSERT_TAIL(&pcpu->bufqueues[bp->b_qindex],
                                  bp, b_freelist);
        }
        spin_unlock(&pcpu->spin);

        /*
         * We have now placed the buffer on the proper queue, but have yet
         * to unlock it.
         */
        if ((bp->b_flags & B_LOCKED) == 0 &&
            ((bp->b_flags & B_INVAL) || (bp->b_flags & B_DELWRI) == 0)) {
                bufcountwakeup();
        }

        /*
         * Something we can maybe free or reuse.
         */
        if (bp->b_bufsize && !(bp->b_flags & B_DELWRI))
                bufspacewakeup();

        /*
         * Final cleanup and unlock.  Clear bits that are only used while a
         * buffer is actively locked.
         */
        bp->b_flags &= ~(B_NOCACHE | B_RELBUF);
        dsched_buf_exit(bp);
        BUF_UNLOCK(bp);
}

/*
 * Hold a buffer, preventing it from being reused.  This will prevent
 * normal B_RELBUF operations on the buffer but will not prevent B_INVAL
 * operations.  If a B_INVAL operation occurs the buffer will remain held
 * but the underlying pages may get ripped out.
 *
 * These functions are typically used in VOP_READ/VOP_WRITE functions
 * to hold a buffer during a copyin or copyout, preventing deadlocks
 * or recursive lock panics when read()/write() is used over mmap()'d
 * space.
 *
 * NOTE: bqhold() requires that the buffer be locked at the time of the
 *       hold.  bqdrop() has no requirements other than the buffer having
 *       previously been held.
 */
void
bqhold(struct buf *bp)
{
        atomic_add_int(&bp->b_refs, 1);
}

void
bqdrop(struct buf *bp)
{
        KKASSERT(bp->b_refs > 0);
        atomic_add_int(&bp->b_refs, -1);
}

/*
 * Return backing pages held by the buffer 'bp' back to the VM system.
 * This routine is called when the bp is invalidated, released, or
 * reused.
 *
 * The KVA mapping (b_data) for the underlying pages is removed by
 * this function.
 *
 * WARNING! This routine is integral to the low memory critical path
 *          when a buffer is B_RELBUF'd.  If the system has a severe page
 *          deficit we need to get the page(s) onto the PQ_FREE or PQ_CACHE
 *          queues so they can be reused in the current pageout daemon
 *          pass.
 */
static void
vfs_vmio_release(struct buf *bp)
{
        int i;
        vm_page_t m;

        for (i = 0; i < bp->b_xio.xio_npages; i++) {
                m = bp->b_xio.xio_pages[i];
                bp->b_xio.xio_pages[i] = NULL;

                /*
                 * We need to own the page in order to safely unwire it.
                 */
                vm_page_busy_wait(m, FALSE, "vmiopg");

                /*
                 * The VFS is telling us this is not a meta-data buffer
                 * even if it is backed by a block device.
                 */
                if (bp->b_flags & B_NOTMETA)
                        vm_page_flag_set(m, PG_NOTMETA);

                /*
                 * This is a very important bit of code.  We try to track
                 * VM page use whether the pages are wired into the buffer
                 * cache or not.  While wired into the buffer cache the
                 * bp tracks the act_count.
                 *
                 * We can choose to place unwired pages on the inactive
                 * queue (0) or active queue (1).  If we place too many
                 * on the active queue the queue will cycle the act_count
                 * on pages we'd like to keep, just from single-use pages
                 * (such as when doing a tar-up or file scan).
                 */
                if (bp->b_act_count < vm_cycle_point)
                        vm_page_unwire(m, 0);
                else
                        vm_page_unwire(m, 1);

                /*
                 * If the wire_count has dropped to 0 we may need to take
                 * further action before unbusying the page.
                 *
                 * WARNING: vm_page_try_*() also checks PG_NEED_COMMIT for us.
                 */
                if (m->wire_count == 0) {
                        if (bp->b_flags & B_DIRECT) {
                                /*
                                 * Attempt to free the page if B_DIRECT is
                                 * set, the caller does not desire the page
                                 * to be cached.
                                 */
                                vm_page_wakeup(m);
                                vm_page_try_to_free(m);
                        } else if ((bp->b_flags & B_NOTMETA) ||
                                   vm_page_count_min(0)) {
                                /*
                                 * Attempt to move the page to PQ_CACHE
                                 * if B_NOTMETA is set.  This flag is set
                                 * by HAMMER to remove one of the two pages
                                 * present when double buffering is enabled.
                                 *
                                 * Attempt to move the page to PQ_CACHE
                                 * If we have a severe page deficit.  This
                                 * will cause buffer cache operations related
                                 * to pageouts to recycle the related pages
                                 * in order to avoid a low memory deadlock.
                                 */
                                m->act_count = bp->b_act_count;
                                vm_page_try_to_cache(m);
                        } else {
                                /*
                                 * Nominal case, leave the page on the
                                 * queue the original unwiring placed it on
                                 * (active or inactive).
                                 */
                                m->act_count = bp->b_act_count;
                                vm_page_wakeup(m);
                        }
                } else {
                        vm_page_wakeup(m);
                }
        }

        /*
         * Zero out the pmap pte's for the mapping, but don't bother
         * invalidating the TLB.  The range will be properly invalidating
         * when new pages are entered into the mapping.
         *
         * This in particular reduces tmpfs tear-down overhead and reduces
         * buffer cache re-use overhead (one invalidation sequence instead
         * of two per re-use).
         */
        pmap_qremove_noinval(trunc_page((vm_offset_t) bp->b_data),
                             bp->b_xio.xio_npages);
        CPUMASK_ASSZERO(bp->b_cpumask);
        if (bp->b_bufsize) {
                atomic_add_long(&bufspace, -bp->b_bufsize);
                bp->b_bufsize = 0;
                bufspacewakeup();
        }
        bp->b_xio.xio_npages = 0;
        bp->b_flags &= ~B_VMIO;
        KKASSERT (LIST_FIRST(&bp->b_dep) == NULL);
        if (bp->b_vp)
                brelvp(bp);
}

/*
 * Find and initialize a new buffer header, freeing up existing buffers
 * in the bufqueues as necessary.  The new buffer is returned locked.
 *
 * Important:  B_INVAL is not set.  If the caller wishes to throw the
 * buffer away, the caller must set B_INVAL prior to calling brelse().
 *
 * We block if:
 *      We have insufficient buffer headers
 *      We have insufficient buffer space
 *
 * To avoid VFS layer recursion we do not flush dirty buffers ourselves.
 * Instead we ask the buf daemon to do it for us.  We attempt to
 * avoid piecemeal wakeups of the pageout daemon.
 */
struct buf *
getnewbuf(int blkflags, int slptimeo, int size, int maxsize)
{
        struct bufpcpu *pcpu;
        struct buf *bp;
        struct buf *nbp;
        int nqindex;
        int nqcpu;
        int slpflags = (blkflags & GETBLK_PCATCH) ? PCATCH : 0;
        int maxloops = 200000;
        int restart_reason = 0;
        struct buf *restart_bp = NULL;
        static char flushingbufs[MAXCPU];
        char *flushingp;

        /*
         * We can't afford to block since we might be holding a vnode lock,
         * which may prevent system daemons from running.  We deal with
         * low-memory situations by proactively returning memory and running
         * async I/O rather then sync I/O.
         */
        
        ++getnewbufcalls;
        nqcpu = mycpu->gd_cpuid;
        flushingp = &flushingbufs[nqcpu];
restart:
        if (bufspace < lobufspace)
                *flushingp = 0;

        if (debug_bufbio && --maxloops == 0)
                panic("getnewbuf, excessive loops on cpu %d restart %d (%p)",
                        mycpu->gd_cpuid, restart_reason, restart_bp);

        /*
         * Setup for scan.  If we do not have enough free buffers,
         * we setup a degenerate case that immediately fails.  Note
         * that if we are specially marked process, we are allowed to
         * dip into our reserves.
         *
         * The scanning sequence is nominally:  EMPTY->CLEAN
         */
        pcpu = &bufpcpu[nqcpu];
        spin_lock(&pcpu->spin);

        /*
         * Prime the scan for this cpu.  Locate the first buffer to
         * check.  If we are flushing buffers we must skip the
         * EMPTY queue.
         */
        nqindex = BQUEUE_EMPTY;
        nbp = TAILQ_FIRST(&pcpu->bufqueues[BQUEUE_EMPTY]);
        if (nbp == NULL || *flushingp) {
                nqindex = BQUEUE_CLEAN;
                nbp = TAILQ_FIRST(&pcpu->bufqueues[BQUEUE_CLEAN]);
        }

        /*
         * Run scan, possibly freeing data and/or kva mappings on the fly,
         * depending.
         *
         * WARNING! spin is held!
         */
        while ((bp = nbp) != NULL) {
                int qindex = nqindex;

                nbp = TAILQ_NEXT(bp, b_freelist);

                /*
                 * BQUEUE_CLEAN - B_AGE special case.  If not set the bp
                 * cycles through the queue twice before being selected.
                 */
                if (qindex == BQUEUE_CLEAN && 
                    (bp->b_flags & B_AGE) == 0 && nbp) {
                        bp->b_flags |= B_AGE;
                        TAILQ_REMOVE(&pcpu->bufqueues[qindex],
                                     bp, b_freelist);
                        TAILQ_INSERT_TAIL(&pcpu->bufqueues[qindex],
                                          bp, b_freelist);
                        continue;
                }

                /*
                 * Calculate next bp ( we can only use it if we do not block
                 * or do other fancy things ).
                 */
                if (nbp == NULL) {
                        switch(qindex) {
                        case BQUEUE_EMPTY:
                                nqindex = BQUEUE_CLEAN;
                                if ((nbp = TAILQ_FIRST(&pcpu->bufqueues[BQUEUE_CLEAN])))
                                        break;
                                /* fall through */
                        case BQUEUE_CLEAN:
                                /*
                                 * nbp is NULL. 
                                 */
                                break;
                        }
                }

                /*
                 * Sanity Checks
                 */
                KASSERT(bp->b_qindex == qindex,
                        ("getnewbuf: inconsistent queue %d bp %p", qindex, bp));

                /*
                 * Note: we no longer distinguish between VMIO and non-VMIO
                 * buffers.
                 */
                KASSERT((bp->b_flags & B_DELWRI) == 0,
                        ("delwri buffer %p found in queue %d", bp, qindex));

                /*
                 * Do not try to reuse a buffer with a non-zero b_refs.
                 * This is an unsynchronized test.  A synchronized test
                 * is also performed after we lock the buffer.
                 */
                if (bp->b_refs)
                        continue;

                /*
                 * Start freeing the bp.  This is somewhat involved.  nbp
                 * remains valid only for BQUEUE_EMPTY bp's.  Buffers
                 * on the clean list must be disassociated from their 
                 * current vnode.  Buffers on the empty lists have
                 * already been disassociated.
                 *
                 * b_refs is checked after locking along with queue changes.
                 * We must check here to deal with zero->nonzero transitions
                 * made by the owner of the buffer lock, which is used by
                 * VFS's to hold the buffer while issuing an unlocked
                 * uiomove()s.  We cannot invalidate the buffer's pages
                 * for this case.  Once we successfully lock a buffer the
                 * only 0->1 transitions of b_refs will occur via findblk().
                 *
                 * We must also check for queue changes after successful
                 * locking as the current lock holder may dispose of the
                 * buffer and change its queue.
                 */
                if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
                        spin_unlock(&pcpu->spin);
                        tsleep(&bd_request, 0, "gnbxxx", (hz + 99) / 100);
                        restart_reason = 1;
                        restart_bp = bp;
                        goto restart;
                }
                if (bp->b_qindex != qindex || bp->b_refs) {
                        spin_unlock(&pcpu->spin);
                        BUF_UNLOCK(bp);
                        restart_reason = 2;
                        restart_bp = bp;
                        goto restart;
                }
                bremfree_locked(bp);
                spin_unlock(&pcpu->spin);

                /*
                 * Dependancies must be handled before we disassociate the
                 * vnode.
                 *
                 * NOTE: HAMMER will set B_LOCKED if the buffer cannot
                 * be immediately disassociated.  HAMMER then becomes
                 * responsible for releasing the buffer.
                 *
                 * NOTE: spin is UNLOCKED now.
                 */
                if (LIST_FIRST(&bp->b_dep) != NULL) {
                        buf_deallocate(bp);
                        if (bp->b_flags & B_LOCKED) {
                                bqrelse(bp);
                                restart_reason = 3;
                                restart_bp = bp;
                                goto restart;
                        }
                        KKASSERT(LIST_FIRST(&bp->b_dep) == NULL);
                }

                /*
                 * CLEAN buffers have content or associations that must be
                 * cleaned out if not repurposing.
                 */
                if (qindex == BQUEUE_CLEAN) {
                        if (bp->b_flags & B_VMIO)
                                vfs_vmio_release(bp);
                        if (bp->b_vp)
                                brelvp(bp);
                }

                /*
                 * NOTE:  nbp is now entirely invalid.  We can only restart
                 * the scan from this point on.
                 *
                 * Get the rest of the buffer freed up.  b_kva* is still
                 * valid after this operation.
                 */
                KASSERT(bp->b_vp == NULL,
                        ("bp3 %p flags %08x vnode %p qindex %d "
                         "unexpectededly still associated!",
                         bp, bp->b_flags, bp->b_vp, qindex));
                KKASSERT((bp->b_flags & B_HASHED) == 0);

                if (bp->b_bufsize)
                        allocbuf(bp, 0);

                if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN | B_HASHED)) {
                        kprintf("getnewbuf: caught bug vp queue "
                                "%p/%08x qidx %d\n",
                                bp, bp->b_flags, qindex);
                        brelvp(bp);
                }
                bp->b_flags = B_BNOCLIP;
                bp->b_cmd = BUF_CMD_DONE;
                bp->b_vp = NULL;
                bp->b_error = 0;
                bp->b_resid = 0;
                bp->b_bcount = 0;
                bp->b_xio.xio_npages = 0;
                bp->b_dirtyoff = bp->b_dirtyend = 0;
                bp->b_act_count = ACT_INIT;
                reinitbufbio(bp);
                KKASSERT(LIST_FIRST(&bp->b_dep) == NULL);
                buf_dep_init(bp);
                if (blkflags & GETBLK_BHEAVY)
                        bp->b_flags |= B_HEAVY;

                if (bufspace >= hibufspace)
                        *flushingp = 1;
                if (bufspace < lobufspace)
                        *flushingp = 0;
                if (*flushingp) {
                        bp->b_flags |= B_INVAL;
                        brelse(bp);
                        restart_reason = 5;
                        restart_bp = bp;
                        goto restart;
                }

                /*
                 * b_refs can transition to a non-zero value while we hold
                 * the buffer locked due to a findblk().  Our brelvp() above
                 * interlocked any future possible transitions due to
                 * findblk()s.
                 *
                 * If we find b_refs to be non-zero we can destroy the
                 * buffer's contents but we cannot yet reuse the buffer.
                 */
                if (bp->b_refs) {
                        bp->b_flags |= B_INVAL;
                        brelse(bp);
                        restart_reason = 6;
                        restart_bp = bp;

                        goto restart;
                }

                /*
                 * We found our buffer!
                 */
                break;
        }

        /*
         * If we exhausted our list, iterate other cpus.  If that fails,
         * sleep as appropriate.  We may have to wakeup various daemons
         * and write out some dirty buffers.
         *
         * Generally we are sleeping due to insufficient buffer space.
         *
         * NOTE: spin is held if bp is NULL, else it is not held.
         */
        if (bp == NULL) {
                int flags;
                char *waitmsg;

                spin_unlock(&pcpu->spin);

                nqcpu = (nqcpu + 1) % ncpus;
                if (nqcpu != mycpu->gd_cpuid) {
                        restart_reason = 7;
                        restart_bp = bp;
                        goto restart;
                }

                if (bufspace >= hibufspace) {
                        waitmsg = "bufspc";
                        flags = VFS_BIO_NEED_BUFSPACE;
                } else {
                        waitmsg = "newbuf";
                        flags = VFS_BIO_NEED_ANY;
                }

                bd_speedup();   /* heeeelp */
                atomic_set_int(&needsbuffer, flags);
                while (needsbuffer & flags) {
                        int value;

                        tsleep_interlock(&needsbuffer, 0);
                        value = atomic_fetchadd_int(&needsbuffer, 0);
                        if (value & flags) {
                                if (tsleep(&needsbuffer, PINTERLOCKED|slpflags,
                                           waitmsg, slptimeo)) {
                                        return (NULL);
                                }
                        }
                }
        } else {
                /*
                 * We finally have a valid bp.  Reset b_data.
                 *
                 * (spin is not held)
                 */
                bp->b_data = bp->b_kvabase;
        }
        return(bp);
}

/*
 * buf_daemon:
 *
 *      Buffer flushing daemon.  Buffers are normally flushed by the
 *      update daemon but if it cannot keep up this process starts to
 *      take the load in an attempt to prevent getnewbuf() from blocking.
 *
 *      Once a flush is initiated it does not stop until the number
 *      of buffers falls below lodirtybuffers, but we will wake up anyone
 *      waiting at the mid-point.
 */
static struct kproc_desc buf_kp = {
        "bufdaemon",
        buf_daemon,
        &bufdaemon_td
};
SYSINIT(bufdaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST,
        kproc_start, &buf_kp);

static struct kproc_desc bufhw_kp = {
        "bufdaemon_hw",
        buf_daemon_hw,
        &bufdaemonhw_td
};
SYSINIT(bufdaemon_hw, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST,
        kproc_start, &bufhw_kp);

static void
buf_daemon1(struct thread *td, int queue, int (*buf_limit_fn)(long), 
            int *bd_req)
{
        long limit;
        struct buf *marker;

        marker = kmalloc(sizeof(*marker), M_BIOBUF, M_WAITOK | M_ZERO);
        marker->b_flags |= B_MARKER;
        marker->b_qindex = BQUEUE_NONE;
        marker->b_qcpu = 0;

        /*
         * This process needs to be suspended prior to shutdown sync.
         */
        EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc,
                              td, SHUTDOWN_PRI_LAST);
        curthread->td_flags |= TDF_SYSTHREAD;

        /*
         * This process is allowed to take the buffer cache to the limit
         */
        for (;;) {
                kproc_suspend_loop();

                /*
                 * Do the flush as long as the number of dirty buffers
                 * (including those running) exceeds lodirtybufspace.
                 *
                 * When flushing limit running I/O to hirunningspace
                 * Do the flush.  Limit the amount of in-transit I/O we
                 * allow to build up, otherwise we would completely saturate
                 * the I/O system.  Wakeup any waiting processes before we
                 * normally would so they can run in parallel with our drain.
                 *
                 * Our aggregate normal+HW lo water mark is lodirtybufspace,
                 * but because we split the operation into two threads we
                 * have to cut it in half for each thread.
                 */
                waitrunningbufspace();
                limit = lodirtybufspace / 2;
                while (buf_limit_fn(limit)) {
                        if (flushbufqueues(marker, queue) == 0)
                                break;
                        if (runningbufspace < hirunningspace)
                                continue;
                        waitrunningbufspace();
                }

                /*
                 * We reached our low water mark, reset the
                 * request and sleep until we are needed again.
                 * The sleep is just so the suspend code works.
                 */
                tsleep_interlock(bd_req, 0);
                if (atomic_swap_int(bd_req, 0) == 0)
                        tsleep(bd_req, PINTERLOCKED, "psleep", hz);
        }
        /* NOT REACHED */
        /*kfree(marker, M_BIOBUF);*/
}

static int
buf_daemon_limit(long limit)
{
        return (runningbufspace + dirtykvaspace > limit ||
                dirtybufcount - dirtybufcounthw >= nbuf / 2);
}

static int
buf_daemon_hw_limit(long limit)
{
        return (runningbufspace + dirtykvaspace > limit ||
                dirtybufcounthw >= nbuf / 2);
}

static void
buf_daemon(void)
{
        buf_daemon1(bufdaemon_td, BQUEUE_DIRTY, buf_daemon_limit, 
                    &bd_request);
}

static void
buf_daemon_hw(void)
{
        buf_daemon1(bufdaemonhw_td, BQUEUE_DIRTY_HW, buf_daemon_hw_limit,
                    &bd_request_hw);
}

/*
 * Flush up to (flushperqueue) buffers in the dirty queue.  Each cpu has a
 * localized version of the queue.  Each call made to this function iterates
 * to another cpu.  It is desireable to flush several buffers from the same
 * cpu's queue at once, as these are likely going to be linear.
 *
 * We must be careful to free up B_INVAL buffers instead of write them, which
 * NFS is particularly sensitive to.
 *
 * B_RELBUF may only be set by VFSs.  We do set B_AGE to indicate that we
 * really want to try to get the buffer out and reuse it due to the write
 * load on the machine.
 *
 * We must lock the buffer in order to check its validity before we can mess
 * with its contents.  spin isn't enough.
 */
static int
flushbufqueues(struct buf *marker, bufq_type_t q)
{
        struct bufpcpu *pcpu;
        struct buf *bp;
        int r = 0;
        u_int loops = flushperqueue;
        int lcpu = marker->b_qcpu;

        KKASSERT(marker->b_qindex == BQUEUE_NONE);
        KKASSERT(marker->b_flags & B_MARKER);

again:
        /*
         * Spinlock needed to perform operations on the queue and may be
         * held through a non-blocking BUF_LOCK(), but cannot be held when
         * BUF_UNLOCK()ing or through any other major operation.
         */
        pcpu = &bufpcpu[marker->b_qcpu];
        spin_lock(&pcpu->spin);
        marker->b_qindex = q;
        TAILQ_INSERT_HEAD(&pcpu->bufqueues[q], marker, b_freelist);
        bp = marker;

        while ((bp = TAILQ_NEXT(bp, b_freelist)) != NULL) {
                /*
                 * NOTE: spinlock is always held at the top of the loop
                 */
                if (bp->b_flags & B_MARKER)
                        continue;
                if ((bp->b_flags & B_DELWRI) == 0) {
                        kprintf("Unexpected clean buffer %p\n", bp);
                        continue;
                }
                if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
                        continue;
                KKASSERT(bp->b_qcpu == marker->b_qcpu && bp->b_qindex == q);

                /*
                 * Once the buffer is locked we will have no choice but to
                 * unlock the spinlock around a later BUF_UNLOCK and re-set
                 * bp = marker when looping.  Move the marker now to make
                 * things easier.
                 */
                TAILQ_REMOVE(&pcpu->bufqueues[q], marker, b_freelist);
                TAILQ_INSERT_AFTER(&pcpu->bufqueues[q], bp, marker, b_freelist);

                /*
                 * Must recheck B_DELWRI after successfully locking
                 * the buffer.
                 */
                if ((bp->b_flags & B_DELWRI) == 0) {
                        spin_unlock(&pcpu->spin);
                        BUF_UNLOCK(bp);
                        spin_lock(&pcpu->spin);
                        bp = marker;
                        continue;
                }

                /*
                 * Remove the buffer from its queue.  We still own the
                 * spinlock here.
                 */
                _bremfree(bp);

                /*
                 * Disposing of an invalid buffer counts as a flush op
                 */
                if (bp->b_flags & B_INVAL) {
                        spin_unlock(&pcpu->spin);
                        brelse(bp);
                        goto doloop;
                }

                /*
                 * Release the spinlock for the more complex ops we
                 * are now going to do.
                 */
                spin_unlock(&pcpu->spin);
                lwkt_yield();

                /*
                 * This is a bit messy
                 */
                if (LIST_FIRST(&bp->b_dep) != NULL &&
                    (bp->b_flags & B_DEFERRED) == 0 &&
                    buf_countdeps(bp, 0)) {
                        spin_lock(&pcpu->spin);
                        TAILQ_INSERT_TAIL(&pcpu->bufqueues[q], bp, b_freelist);
                        bp->b_qindex = q;
                        bp->b_flags |= B_DEFERRED;
                        spin_unlock(&pcpu->spin);
                        BUF_UNLOCK(bp);
                        spin_lock(&pcpu->spin);
                        bp = marker;
                        continue;
                }

                /*
                 * spinlock not held here.
                 *
                 * If the buffer has a dependancy, buf_checkwrite() must
                 * also return 0 for us to be able to initate the write.
                 *
                 * If the buffer is flagged B_ERROR it may be requeued
                 * over and over again, we try to avoid a live lock.
                 */
                if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
                        brelse(bp);
                } else if (bp->b_flags & B_ERROR) {
                        tsleep(bp, 0, "bioer", 1);
                        bp->b_flags &= ~B_AGE;
                        cluster_awrite(bp);
                } else {
                        bp->b_flags |= B_AGE | B_KVABIO;
                        cluster_awrite(bp);
                }
                /* bp invalid but needs to be NULL-tested if we break out */
doloop:
                spin_lock(&pcpu->spin);
                ++r;
                if (--loops == 0)
                        break;
                bp = marker;
        }
        /* bp is invalid here but can be NULL-tested to advance */

        TAILQ_REMOVE(&pcpu->bufqueues[q], marker, b_freelist);
        marker->b_qindex = BQUEUE_NONE;
        spin_unlock(&pcpu->spin);

        /*
         * Advance the marker to be fair.
         */
        marker->b_qcpu = (marker->b_qcpu + 1) % ncpus;
        if (bp == NULL) {
                if (marker->b_qcpu != lcpu)
                        goto again;
        }

        return (r);
}

/*
 * inmem:
 *
 *      Returns true if no I/O is needed to access the associated VM object.
 *      This is like findblk except it also hunts around in the VM system for
 *      the data.
 *
 *      Note that we ignore vm_page_free() races from interrupts against our
 *      lookup, since if the caller is not protected our return value will not
 *      be any more valid then otherwise once we exit the critical section.
 */
int
inmem(struct vnode *vp, off_t loffset)
{
        vm_object_t obj;
        vm_offset_t toff, tinc, size;
        vm_page_t m;
        int res = 1;

        if (findblk(vp, loffset, FINDBLK_TEST))
                return 1;
        if (vp->v_mount == NULL)
                return 0;
        if ((obj = vp->v_object) == NULL)
                return 0;

        size = PAGE_SIZE;
        if (size > vp->v_mount->mnt_stat.f_iosize)
                size = vp->v_mount->mnt_stat.f_iosize;

        vm_object_hold(obj);
        for (toff = 0; toff < vp->v_mount->mnt_stat.f_iosize; toff += tinc) {
                m = vm_page_lookup(obj, OFF_TO_IDX(loffset + toff));
                if (m == NULL) {
                        res = 0;
                        break;
                }
                tinc = size;
                if (tinc > PAGE_SIZE - ((toff + loffset) & PAGE_MASK))
                        tinc = PAGE_SIZE - ((toff + loffset) & PAGE_MASK);
                if (vm_page_is_valid(m,
                    (vm_offset_t) ((toff + loffset) & PAGE_MASK), tinc) == 0) {
                        res = 0;
                        break;
                }
        }
        vm_object_drop(obj);
        return (res);
}

/*
 * findblk:
 *
 *      Locate and return the specified buffer.  Unless flagged otherwise,
 *      a locked buffer will be returned if it exists or NULL if it does not.
 *
 *      findblk()'d buffers are still on the bufqueues and if you intend
 *      to use your (locked NON-TEST) buffer you need to bremfree(bp)
 *      and possibly do other stuff to it.
 *
 *      FINDBLK_TEST    - Do not lock the buffer.  The caller is responsible
 *                        for locking the buffer and ensuring that it remains
 *                        the desired buffer after locking.
 *
 *      FINDBLK_NBLOCK  - Lock the buffer non-blocking.  If we are unable
 *                        to acquire the lock we return NULL, even if the
 *                        buffer exists.
 *
 *      FINDBLK_REF     - Returns the buffer ref'd, which prevents normal
 *                        reuse by getnewbuf() but does not prevent
 *                        disassociation (B_INVAL).  Used to avoid deadlocks
 *                        against random (vp,loffset)s due to reassignment.
 *
 *      FINDBLK_KVABIO  - Only applicable when returning a locked buffer.
 *                        Indicates that the caller supports B_KVABIO.
 *
 *      (0)             - Lock the buffer blocking.
 */
struct buf *
findblk(struct vnode *vp, off_t loffset, int flags)
{
        struct buf *bp;
        int lkflags;

        lkflags = LK_EXCLUSIVE;
        if (flags & FINDBLK_NBLOCK)
                lkflags |= LK_NOWAIT;

        for (;;) {
                /*
                 * Lookup.  Ref the buf while holding v_token to prevent
                 * reuse (but does not prevent diassociation).
                 */
                lwkt_gettoken_shared(&vp->v_token);
                bp = buf_rb_hash_RB_LOOKUP(&vp->v_rbhash_tree, loffset);
                if (bp == NULL) {
                        lwkt_reltoken(&vp->v_token);
                        return(NULL);
                }
                bqhold(bp);
                lwkt_reltoken(&vp->v_token);

                /*
                 * If testing only break and return bp, do not lock.
                 */
                if (flags & FINDBLK_TEST)
                        break;

                /*
                 * Lock the buffer, return an error if the lock fails.
                 * (only FINDBLK_NBLOCK can cause the lock to fail).
                 */
                if (BUF_LOCK(bp, lkflags)) {
                        atomic_subtract_int(&bp->b_refs, 1);
                        /* bp = NULL; not needed */
                        return(NULL);
                }

                /*
                 * Revalidate the locked buf before allowing it to be
                 * returned.
                 *
                 * B_KVABIO is only set/cleared when locking.  When
                 * clearing B_KVABIO, we must ensure that the buffer
                 * is synchronized to all cpus.
                 */
                if (bp->b_vp == vp && bp->b_loffset == loffset) {
                        if (flags & FINDBLK_KVABIO)
                                bp->b_flags |= B_KVABIO;
                        else
                                bkvasync_all(bp);
                        break;
                }
                atomic_subtract_int(&bp->b_refs, 1);
                BUF_UNLOCK(bp);
        }

        /*
         * Success
         */
        if ((flags & FINDBLK_REF) == 0)
                atomic_subtract_int(&bp->b_refs, 1);
        return(bp);
}

/*
 * getcacheblk:
 *
 *      Similar to getblk() except only returns the buffer if it is
 *      B_CACHE and requires no other manipulation.  Otherwise NULL
 *      is returned.  NULL is also returned if GETBLK_NOWAIT is set
 *      and the getblk() would block.
 *
 *      If B_RAM is set the buffer might be just fine, but we return
 *      NULL anyway because we want the code to fall through to the
 *      cluster read to issue more read-aheads.  Otherwise read-ahead breaks.
 *
 *      If blksize is 0 the buffer cache buffer must already be fully
 *      cached.
 *
 *      If blksize is non-zero getblk() will be used, allowing a buffer
 *      to be reinstantiated from its VM backing store.  The buffer must
 *      still be fully cached after reinstantiation to be returned.
 */
struct buf *
getcacheblk(struct vnode *vp, off_t loffset, int blksize, int blkflags)
{
        struct buf *bp;
        int fndflags = 0;

        if (blkflags & GETBLK_NOWAIT)
                fndflags |= FINDBLK_NBLOCK;
        if (blkflags & GETBLK_KVABIO)
                fndflags |= FINDBLK_KVABIO;

        if (blksize) {
                bp = getblk(vp, loffset, blksize, blkflags, 0);
                if (bp) {
                        if ((bp->b_flags & (B_INVAL | B_CACHE)) == B_CACHE) {
                                bp->b_flags &= ~B_AGE;
                                if (bp->b_flags & B_RAM) {
                                        bqrelse(bp);
                                        bp = NULL;
                                }
                        } else {
                                brelse(bp);
                                bp = NULL;
                        }
                }
        } else {
                bp = findblk(vp, loffset, fndflags);
                if (bp) {
                        if ((bp->b_flags & (B_INVAL | B_CACHE | B_RAM)) ==
                            B_CACHE) {
                                bp->b_flags &= ~B_AGE;
                                bremfree(bp);
                        } else {
                                BUF_UNLOCK(bp);
                                bp = NULL;
                        }
                }
        }
        return (bp);
}

/*
 * getblk:
 *
 *      Get a block given a specified block and offset into a file/device.
 *      B_INVAL may or may not be set on return.  The caller should clear
 *      B_INVAL prior to initiating a READ.
 *
 *      IT IS IMPORTANT TO UNDERSTAND THAT IF YOU CALL GETBLK() AND B_CACHE
 *      IS NOT SET, YOU MUST INITIALIZE THE RETURNED BUFFER, ISSUE A READ,
 *      OR SET B_INVAL BEFORE RETIRING IT.  If you retire a getblk'd buffer
 *      without doing any of those things the system will likely believe
 *      the buffer to be valid (especially if it is not B_VMIO), and the
 *      next getblk() will return the buffer with B_CACHE set.
 *
 *      For a non-VMIO buffer, B_CACHE is set to the opposite of B_INVAL for
 *      an existing buffer.
 *
 *      For a VMIO buffer, B_CACHE is modified according to the backing VM.
 *      If getblk()ing a previously 0-sized invalid buffer, B_CACHE is set
 *      and then cleared based on the backing VM.  If the previous buffer is
 *      non-0-sized but invalid, B_CACHE will be cleared.
 *
 *      If getblk() must create a new buffer, the new buffer is returned with
 *      both B_INVAL and B_CACHE clear unless it is a VMIO buffer, in which
 *      case it is returned with B_INVAL clear and B_CACHE set based on the
 *      backing VM.
 *
 *      getblk() also forces a bwrite() for any B_DELWRI buffer whos
 *      B_CACHE bit is clear.
 *      
 *      What this means, basically, is that the caller should use B_CACHE to
 *      determine whether the buffer is fully valid or not and should clear
 *      B_INVAL prior to issuing a read.  If the caller intends to validate
 *      the buffer by loading its data area with something, the caller needs
 *      to clear B_INVAL.  If the caller does this without issuing an I/O, 
 *      the caller should set B_CACHE ( as an optimization ), else the caller
 *      should issue the I/O and biodone() will set B_CACHE if the I/O was
 *      a write attempt or if it was a successfull read.  If the caller 
 *      intends to issue a READ, the caller must clear B_INVAL and B_ERROR
 *      prior to issuing the READ.  biodone() will *not* clear B_INVAL.
 *
 *      getblk flags:
 *
 *      GETBLK_PCATCH - catch signal if blocked, can cause NULL return
 *      GETBLK_BHEAVY - heavy-weight buffer cache buffer
 */
struct buf *
getblk(struct vnode *vp, off_t loffset, int size, int blkflags, int slptimeo)
{
        struct buf *bp;
        int slpflags = (blkflags & GETBLK_PCATCH) ? PCATCH : 0;
        int error;
        int lkflags;

        if (size > MAXBSIZE)
                panic("getblk: size(%d) > MAXBSIZE(%d)", size, MAXBSIZE);
        if (vp->v_object == NULL)
                panic("getblk: vnode %p has no object!", vp);

        /*
         * NOTE: findblk does not try to resolve KVABIO in REF-only mode.
         *       we still have to handle that ourselves.
         */
loop:
        if ((bp = findblk(vp, loffset, FINDBLK_REF | FINDBLK_TEST)) != NULL) {
                /*
                 * The buffer was found in the cache, but we need to lock it.
                 * We must acquire a ref on the bp to prevent reuse, but
                 * this will not prevent disassociation (brelvp()) so we
                 * must recheck (vp,loffset) after acquiring the lock.
                 *
                 * Without the ref the buffer could potentially be reused
                 * before we acquire the lock and create a deadlock
                 * situation between the thread trying to reuse the buffer
                 * and us due to the fact that we would wind up blocking
                 * on a random (vp,loffset).
                 */
                if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
                        if (blkflags & GETBLK_NOWAIT) {
                                bqdrop(bp);
                                return(NULL);
                        }
                        lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
                        if (blkflags & GETBLK_PCATCH)
                                lkflags |= LK_PCATCH;
                        error = BUF_TIMELOCK(bp, lkflags, "getblk", slptimeo);
                        if (error) {
                                bqdrop(bp);
                                if (error == ENOLCK)
                                        goto loop;
                                return (NULL);
                        }
                        /* buffer may have changed on us */
                }
                bqdrop(bp);

                /*
                 * Once the buffer has been locked, make sure we didn't race
                 * a buffer recyclement.  Buffers that are no longer hashed
                 * will have b_vp == NULL, so this takes care of that check
                 * as well.
                 */
                if (bp->b_vp != vp || bp->b_loffset != loffset) {
#if 0
                        kprintf("Warning buffer %p (vp %p loffset %lld) "
                                "was recycled\n",
                                bp, vp, (long long)loffset);
#endif
                        BUF_UNLOCK(bp);
                        goto loop;
                }

                /*
                 * If SZMATCH any pre-existing buffer must be of the requested
                 * size or NULL is returned.  The caller absolutely does not
                 * want getblk() to bwrite() the buffer on a size mismatch.
                 */
                if ((blkflags & GETBLK_SZMATCH) && size != bp->b_bcount) {
                        BUF_UNLOCK(bp);
                        return(NULL);
                }

                /*
                 * All vnode-based buffers must be backed by a VM object.
                 *
                 * Set B_KVABIO for any incidental work, we will fix it
                 * up later.
                 */
                KKASSERT(bp->b_flags & B_VMIO);
                KKASSERT(bp->b_cmd == BUF_CMD_DONE);
                bp->b_flags &= ~B_AGE;
                bp->b_flags |= B_KVABIO;

                /*
                 * Make sure that B_INVAL buffers do not have a cached
                 * block number translation.
                 */
                if ((bp->b_flags & B_INVAL) &&
                    (bp->b_bio2.bio_offset != NOOFFSET)) {
                        kprintf("Warning invalid buffer %p (vp %p loffset %lld)"
                                " did not have cleared bio_offset cache\n",
                                bp, vp, (long long)loffset);
                        clearbiocache(&bp->b_bio2);
                }

                /*
                 * The buffer is locked.  B_CACHE is cleared if the buffer is 
                 * invalid.
                 *
                 * After the bremfree(), disposals must use b[q]relse().
                 */
                if (bp->b_flags & B_INVAL)
                        bp->b_flags &= ~B_CACHE;
                bremfree(bp);

                /*
                 * Any size inconsistancy with a dirty buffer or a buffer
                 * with a softupdates dependancy must be resolved.  Resizing
                 * the buffer in such circumstances can lead to problems.
                 *
                 * Dirty or dependant buffers are written synchronously.
                 * Other types of buffers are simply released and
                 * reconstituted as they may be backed by valid, dirty VM
                 * pages (but not marked B_DELWRI).
                 *
                 * NFS NOTE: NFS buffers which straddle EOF are oddly-sized
                 * and may be left over from a prior truncation (and thus
                 * no longer represent the actual EOF point), so we
                 * definitely do not want to B_NOCACHE the backing store.
                 */
                if (size != bp->b_bcount) {
                        if (bp->b_flags & B_DELWRI) {
                                bp->b_flags |= B_RELBUF;
                                bwrite(bp);
                        } else if (LIST_FIRST(&bp->b_dep)) {
                                bp->b_flags |= B_RELBUF;
                                bwrite(bp);
                        } else {
                                bp->b_flags |= B_RELBUF;
                                brelse(bp);
                        }
                        goto loop;
                }
                KKASSERT(size <= bp->b_kvasize);
                KASSERT(bp->b_loffset != NOOFFSET, 
                        ("getblk: no buffer offset"));

                /*
                 * A buffer with B_DELWRI set and B_CACHE clear must
                 * be committed before we can return the buffer in
                 * order to prevent the caller from issuing a read
                 * ( due to B_CACHE not being set ) and overwriting
                 * it.
                 *
                 * Most callers, including NFS and FFS, need this to
                 * operate properly either because they assume they
                 * can issue a read if B_CACHE is not set, or because
                 * ( for example ) an uncached B_DELWRI might loop due 
                 * to softupdates re-dirtying the buffer.  In the latter
                 * case, B_CACHE is set after the first write completes,
                 * preventing further loops.
                 *
                 * NOTE!  b*write() sets B_CACHE.  If we cleared B_CACHE
                 * above while extending the buffer, we cannot allow the
                 * buffer to remain with B_CACHE set after the write
                 * completes or it will represent a corrupt state.  To
                 * deal with this we set B_NOCACHE to scrap the buffer
                 * after the write.
                 *
                 * XXX Should this be B_RELBUF instead of B_NOCACHE?
                 *     I'm not even sure this state is still possible
                 *     now that getblk() writes out any dirty buffers
                 *     on size changes.
                 *
                 * We might be able to do something fancy, like setting
                 * B_CACHE in bwrite() except if B_DELWRI is already set,
                 * so the below call doesn't set B_CACHE, but that gets real
                 * confusing.  This is much easier.
                 */
                if ((bp->b_flags & (B_CACHE|B_DELWRI)) == B_DELWRI) {
                        kprintf("getblk: Warning, bp %p loff=%jx DELWRI set "
                                "and CACHE clear, b_flags %08x\n",
                                bp, (uintmax_t)bp->b_loffset, bp->b_flags);
                        bp->b_flags |= B_NOCACHE;
                        bwrite(bp);
                        goto loop;
                }
        } else {
                /*
                 * Buffer is not in-core, create new buffer.  The buffer
                 * returned by getnewbuf() is locked.  Note that the returned
                 * buffer is also considered valid (not marked B_INVAL).
                 *
                 * Calculating the offset for the I/O requires figuring out
                 * the block size.  We use DEV_BSIZE for VBLK or VCHR and
                 * the mount's f_iosize otherwise.  If the vnode does not
                 * have an associated mount we assume that the passed size is 
                 * the block size.  
                 *
                 * Note that vn_isdisk() cannot be used here since it may
                 * return a failure for numerous reasons.   Note that the
                 * buffer size may be larger then the block size (the caller
                 * will use block numbers with the proper multiple).  Beware
                 * of using any v_* fields which are part of unions.  In
                 * particular, in DragonFly the mount point overloading 
                 * mechanism uses the namecache only and the underlying
                 * directory vnode is not a special case.
                 */
                int bsize, maxsize;

                if (vp->v_type == VBLK || vp->v_type == VCHR)
                        bsize = DEV_BSIZE;
                else if (vp->v_mount)
                        bsize = vp->v_mount->mnt_stat.f_iosize;
                else
                        bsize = size;

                maxsize = size + (loffset & PAGE_MASK);
                maxsize = imax(maxsize, bsize);

                bp = getnewbuf(blkflags, slptimeo, size, maxsize);
                if (bp == NULL) {
                        if (slpflags || slptimeo)
                                return NULL;
                        goto loop;
                }

                /*
                 * Atomically insert the buffer into the hash, so that it can
                 * be found by findblk().
                 *
                 * If bgetvp() returns non-zero a collision occured, and the
                 * bp will not be associated with the vnode.
                 *
                 * Make sure the translation layer has been cleared.
                 */
                bp->b_loffset = loffset;
                bp->b_bio2.bio_offset = NOOFFSET;
                /* bp->b_bio2.bio_next = NULL; */

                if (bgetvp(vp, bp, size)) {
                        bp->b_flags |= B_INVAL;
                        brelse(bp);
                        goto loop;
                }

                /*
                 * All vnode-based buffers must be backed by a VM object.
                 *
                 * Set B_KVABIO for incidental work
                 */
                KKASSERT(vp->v_object != NULL);
                bp->b_flags |= B_VMIO | B_KVABIO;
                KKASSERT(bp->b_cmd == BUF_CMD_DONE);

                allocbuf(bp, size);
        }

        /*
         * Do the nasty smp broadcast (if the buffer needs it) when KVABIO
         * is not supported.
         */
        if (bp && (blkflags & GETBLK_KVABIO) == 0) {
                bkvasync_all(bp);
        }
        return (bp);
}

/*
 * regetblk(bp)
 *
 * Reacquire a buffer that was previously released to the locked queue,
 * or reacquire a buffer which is interlocked by having bioops->io_deallocate
 * set B_LOCKED (which handles the acquisition race).
 *
 * To this end, either B_LOCKED must be set or the dependancy list must be
 * non-empty.
 */
void
regetblk(struct buf *bp)
{
        KKASSERT((bp->b_flags & B_LOCKED) || LIST_FIRST(&bp->b_dep) != NULL);
        BUF_LOCK(bp, LK_EXCLUSIVE | LK_RETRY);
        bremfree(bp);
}

/*
 * allocbuf:
 *
 *      This code constitutes the buffer memory from either anonymous system
 *      memory (in the case of non-VMIO operations) or from an associated
 *      VM object (in the case of VMIO operations).  This code is able to
 *      resize a buffer up or down.
 *
 *      Note that this code is tricky, and has many complications to resolve
 *      deadlock or inconsistant data situations.  Tread lightly!!! 
 *      There are B_CACHE and B_DELWRI interactions that must be dealt with by 
 *      the caller.  Calling this code willy nilly can result in the loss of
 *      data.
 *
 *      allocbuf() only adjusts B_CACHE for VMIO buffers.  getblk() deals with
 *      B_CACHE for the non-VMIO case.
 *
 *      This routine does not need to be called from a critical section but you
 *      must own the buffer.
 */
void
allocbuf(struct buf *bp, int size)
{
        vm_page_t m;
        int newbsize;
        int desiredpages;
        int i;

        if (BUF_REFCNT(bp) == 0)
                panic("allocbuf: buffer not busy");

        if (bp->b_kvasize < size)
                panic("allocbuf: buffer too small");

        KKASSERT(bp->b_flags & B_VMIO);

        newbsize = roundup2(size, DEV_BSIZE);
        desiredpages = ((int)(bp->b_loffset & PAGE_MASK) +
                        newbsize + PAGE_MASK) >> PAGE_SHIFT;
        KKASSERT(desiredpages <= XIO_INTERNAL_PAGES);

        /*
         * Set B_CACHE initially if buffer is 0 length or will become
         * 0-length.
         */
        if (size == 0 || bp->b_bufsize == 0)
                bp->b_flags |= B_CACHE;

        if (newbsize < bp->b_bufsize) {
                /*
                 * DEV_BSIZE aligned new buffer size is less then the
                 * DEV_BSIZE aligned existing buffer size.  Figure out
                 * if we have to remove any pages.
                 */
                if (desiredpages < bp->b_xio.xio_npages) {
                        for (i = desiredpages; i < bp->b_xio.xio_npages; i++) {
                                /*
                                 * the page is not freed here -- it
                                 * is the responsibility of
                                 * vnode_pager_setsize
                                 */
                                m = bp->b_xio.xio_pages[i];
                                KASSERT(m != bogus_page,
                                    ("allocbuf: bogus page found"));
                                vm_page_busy_wait(m, TRUE, "biodep");
                                bp->b_xio.xio_pages[i] = NULL;
                                vm_page_unwire(m, 0);
                                vm_page_wakeup(m);
                        }
                        pmap_qremove_noinval((vm_offset_t)
                                      trunc_page((vm_offset_t)bp->b_data) +
                                      (desiredpages << PAGE_SHIFT),
                                     (bp->b_xio.xio_npages - desiredpages));
                        bp->b_xio.xio_npages = desiredpages;

                        /*
                         * Don't bother invalidating the pmap changes
                         * (which wastes global SMP invalidation IPIs)
                         * when setting the size to 0.  This case occurs
                         * when called via getnewbuf() during buffer
                         * recyclement.
                         */
                        if (desiredpages == 0) {
                                CPUMASK_ASSZERO(bp->b_cpumask);
                        } else {
                                bkvareset(bp);
                        }
                }
        } else if (size > bp->b_bcount) {
                /*
                 * We are growing the buffer, possibly in a
                 * byte-granular fashion.
                 */
                struct vnode *vp;
                vm_object_t obj;
                vm_offset_t toff;
                vm_offset_t tinc;

                /*
                 * Step 1, bring in the VM pages from the object,
                 * allocating them if necessary.  We must clear
                 * B_CACHE if these pages are not valid for the
                 * range covered by the buffer.
                 */
                vp = bp->b_vp;
                obj = vp->v_object;

                vm_object_hold(obj);
                while (bp->b_xio.xio_npages < desiredpages) {
                        vm_page_t m;
                        vm_pindex_t pi;
                        int error;

                        pi = OFF_TO_IDX(bp->b_loffset) +
                             bp->b_xio.xio_npages;

                        /*
                         * Blocking on m->busy_count might lead to a
                         * deadlock:
                         *
                         *  vm_fault->getpages->cluster_read->allocbuf
                         */
                        m = vm_page_lookup_busy_try(obj, pi, FALSE,
                                                    &error);
                        if (error) {
                                vm_page_sleep_busy(m, FALSE, "pgtblk");
                                continue;
                        }
                        if (m == NULL) {
                                /*
                                 * note: must allocate system pages
                                 * since blocking here could intefere
                                 * with paging I/O, no matter which
                                 * process we are.
                                 */
                                m = bio_page_alloc(bp, obj, pi,
                                                   desiredpages -
                                                    bp->b_xio.xio_npages);
                                if (m) {
                                        vm_page_wire(m);
                                        vm_page_wakeup(m);
                                        bp->b_flags &= ~B_CACHE;
                                        bp->b_xio.xio_pages[bp->b_xio.xio_npages] = m;
                                        ++bp->b_xio.xio_npages;
                                }
                                continue;
                        }

                        /*
                         * We found a page and were able to busy it.
                         */
                        vm_page_wire(m);
                        vm_page_wakeup(m);
                        bp->b_xio.xio_pages[bp->b_xio.xio_npages] = m;
                        ++bp->b_xio.xio_npages;
                        if (bp->b_act_count < m->act_count)
                                bp->b_act_count = m->act_count;
                }
                vm_object_drop(obj);

                /*
                 * Step 2.  We've loaded the pages into the buffer,
                 * we have to figure out if we can still have B_CACHE
                 * set.  Note that B_CACHE is set according to the
                 * byte-granular range ( bcount and size ), not the
                 * aligned range ( newbsize ).
                 *
                 * The VM test is against m->valid, which is DEV_BSIZE
                 * aligned.  Needless to say, the validity of the data
                 * needs to also be DEV_BSIZE aligned.  Note that this
                 * fails with NFS if the server or some other client
                 * extends the file's EOF.  If our buffer is resized,
                 * B_CACHE may remain set! XXX
                 */

                toff = bp->b_bcount;
                tinc = PAGE_SIZE - ((bp->b_loffset + toff) & PAGE_MASK);

                while ((bp->b_flags & B_CACHE) && toff < size) {
                        vm_pindex_t pi;

                        if (tinc > (size - toff))
                                tinc = size - toff;

                        pi = ((bp->b_loffset & PAGE_MASK) + toff) >>
                            PAGE_SHIFT;

                        vfs_buf_test_cache(
                            bp,
                            bp->b_loffset,
                            toff,
                            tinc,
                            bp->b_xio.xio_pages[pi]
                        );
                        toff += tinc;
                        tinc = PAGE_SIZE;
                }

                /*
                 * Step 3, fixup the KVM pmap.  Remember that
                 * bp->b_data is relative to bp->b_loffset, but
                 * bp->b_loffset may be offset into the first page.
                 */
                bp->b_data = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
                pmap_qenter_noinval((vm_offset_t)bp->b_data,
                            bp->b_xio.xio_pages, bp->b_xio.xio_npages);
                bp->b_data = (caddr_t)((vm_offset_t)bp->b_data |
                                      (vm_offset_t)(bp->b_loffset & PAGE_MASK));
                bkvareset(bp);
        }
        atomic_add_long(&bufspace, newbsize - bp->b_bufsize);

        /* adjust space use on already-dirty buffer */
        if (bp->b_flags & B_DELWRI) {
                /* dirtykvaspace unchanged */
                atomic_add_long(&dirtybufspace, newbsize - bp->b_bufsize);
                if (bp->b_flags & B_HEAVY) {
                        atomic_add_long(&dirtybufspacehw,
                                        newbsize - bp->b_bufsize);
                }
        }
        bp->b_bufsize = newbsize;       /* actual buffer allocation     */
        bp->b_bcount = size;            /* requested buffer size        */
        bufspacewakeup();
}

/*
 * biowait:
 *
 *      Wait for buffer I/O completion, returning error status. B_EINTR
 *      is converted into an EINTR error but not cleared (since a chain
 *      of biowait() calls may occur).
 *
 *      On return bpdone() will have been called but the buffer will remain
 *      locked and will not have been brelse()'d.
 *
 *      NOTE!  If a timeout is specified and ETIMEDOUT occurs the I/O is
 *      likely still in progress on return.
 *
 *      NOTE!  This operation is on a BIO, not a BUF.
 *
 *      NOTE!  BIO_DONE is cleared by vn_strategy()
 */
static __inline int
_biowait(struct bio *bio, const char *wmesg, int to)
{
        struct buf *bp = bio->bio_buf;
        u_int32_t flags;
        u_int32_t nflags;
        int error;

        KKASSERT(bio == &bp->b_bio1);
        for (;;) {
                flags = bio->bio_flags;
                if (flags & BIO_DONE)
                        break;
                nflags = flags | BIO_WANT;
                tsleep_interlock(bio, 0);
                if (atomic_cmpset_int(&bio->bio_flags, flags, nflags)) {
                        if (wmesg)
                                error = tsleep(bio, PINTERLOCKED, wmesg, to);
                        else if (bp->b_cmd == BUF_CMD_READ)
                                error = tsleep(bio, PINTERLOCKED, "biord", to);
                        else
                                error = tsleep(bio, PINTERLOCKED, "biowr", to);
                        if (error) {
                                kprintf("tsleep error biowait %d\n", error);
                                return (error);
                        }
                }
        }

        /*
         * Finish up.
         */
        KKASSERT(bp->b_cmd == BUF_CMD_DONE);
        bio->bio_flags &= ~(BIO_DONE | BIO_SYNC);
        if (bp->b_flags & B_EINTR)
                return (EINTR);
        if (bp->b_flags & B_ERROR)
                return (bp->b_error ? bp->b_error : EIO);
        return (0);
}

int
biowait(struct bio *bio, const char *wmesg)
{
        return(_biowait(bio, wmesg, 0));
}

int
biowait_timeout(struct bio *bio, const char *wmesg, int to)
{
        return(_biowait(bio, wmesg, to));
}

/*
 * This associates a tracking count with an I/O.  vn_strategy() and
 * dev_dstrategy() do this automatically but there are a few cases
 * where a vnode or device layer is bypassed when a block translation
 * is cached.  In such cases bio_start_transaction() may be called on
 * the bypassed layers so the system gets an I/O in progress indication 
 * for those higher layers.
 */
void
bio_start_transaction(struct bio *bio, struct bio_track *track)
{
        bio->bio_track = track;
        bio_track_ref(track);
        dsched_buf_enter(bio->bio_buf); /* might stack */
}

/*
 * Initiate I/O on a vnode.
 *
 * SWAPCACHE OPERATION:
 *
 *      Real buffer cache buffers have a non-NULL bp->b_vp.  Unfortunately
 *      devfs also uses b_vp for fake buffers so we also have to check
 *      that B_PAGING is 0.  In this case the passed 'vp' is probably the
 *      underlying block device.  The swap assignments are related to the
 *      buffer cache buffer's b_vp, not the passed vp.
 *
 *      The passed vp == bp->b_vp only in the case where the strategy call
 *      is made on the vp itself for its own buffers (a regular file or
 *      block device vp).  The filesystem usually then re-calls vn_strategy()
 *      after translating the request to an underlying device.
 *
 *      Cluster buffers set B_CLUSTER and the passed vp is the vp of the
 *      underlying buffer cache buffers.
 *
 *      We can only deal with page-aligned buffers at the moment, because
 *      we can't tell what the real dirty state for pages straddling a buffer
 *      are.
 *
 *      In order to call swap_pager_strategy() we must provide the VM object
 *      and base offset for the underlying buffer cache pages so it can find
 *      the swap blocks.
 */
void
vn_strategy(struct vnode *vp, struct bio *bio)
{
        struct bio_track *track;
        struct buf *bp = bio->bio_buf;

        KKASSERT(bp->b_cmd != BUF_CMD_DONE);

        /*
         * Set when an I/O is issued on the bp.  Cleared by consumers
         * (aka HAMMER), allowing the consumer to determine if I/O had
         * actually occurred.
         */
        bp->b_flags |= B_IOISSUED;

        /*
         * Handle the swapcache intercept.
         *
         * NOTE: The swapcache itself always supports KVABIO and will
         *       do the right thing if its underlying devices do not.
         */
        if (vn_cache_strategy(vp, bio))
                return;

        /*
         * If the vnode does not support KVABIO and the buffer is using
         * KVABIO, we must synchronize b_data to all cpus before dispatching.
         */
        if ((vp->v_flag & VKVABIO) == 0 && (bp->b_flags & B_KVABIO))
                bkvasync_all(bp);

        /*
         * Otherwise do the operation through the filesystem
         */
        if (bp->b_cmd == BUF_CMD_READ)
                track = &vp->v_track_read;
        else
                track = &vp->v_track_write;
        KKASSERT((bio->bio_flags & BIO_DONE) == 0);
        bio->bio_track = track;
        bio_track_ref(track);
        dsched_buf_enter(bp);   /* might stack */
        vop_strategy(*vp->v_ops, vp, bio);
}

/*
 * vn_cache_strategy()
 *
 * NOTE: This function supports the KVABIO API wherein b_data might not
 *       be synchronized to the current cpu.
 */
static void vn_cache_strategy_callback(struct bio *bio);

int
vn_cache_strategy(struct vnode *vp, struct bio *bio)
{
        struct buf *bp = bio->bio_buf;
        struct bio *nbio;
        vm_object_t object;
        vm_page_t m;
        int i;

        /*
         * Stop using swapcache if paniced, dumping, or dumped
         */
        if (panicstr || dumping)
                return(0);

        /*
         * Is this buffer cache buffer suitable for reading from
         * the swap cache?
         */
        if (vm_swapcache_read_enable == 0 ||
            bp->b_cmd != BUF_CMD_READ ||
            ((bp->b_flags & B_CLUSTER) == 0 &&
             (bp->b_vp == NULL || (bp->b_flags & B_PAGING))) ||
            ((int)bp->b_loffset & PAGE_MASK) != 0 ||
            (bp->b_bcount & PAGE_MASK) != 0) {
                return(0);
        }

        /*
         * Figure out the original VM object (it will match the underlying
         * VM pages).  Note that swap cached data uses page indices relative
         * to that object, not relative to bio->bio_offset.
         */
        if (bp->b_flags & B_CLUSTER)
                object = vp->v_object;
        else
                object = bp->b_vp->v_object;

        /*
         * In order to be able to use the swap cache all underlying VM
         * pages must be marked as such, and we can't have any bogus pages.
         */
        for (i = 0; i < bp->b_xio.xio_npages; ++i) {
                m = bp->b_xio.xio_pages[i];
                if ((m->flags & PG_SWAPPED) == 0)
                        break;
                if (m == bogus_page)
                        break;
        }

        /*
         * If we are good then issue the I/O using swap_pager_strategy().
         *
         * We can only do this if the buffer actually supports object-backed
         * I/O.  If it doesn't npages will be 0.
         */
        if (i && i == bp->b_xio.xio_npages) {
                m = bp->b_xio.xio_pages[0];
                nbio = push_bio(bio);
                nbio->bio_done = vn_cache_strategy_callback;
                nbio->bio_offset = ptoa(m->pindex);
                KKASSERT(m->object == object);
                swap_pager_strategy(object, nbio);
                return(1);
        }
        return(0);
}

/*
 * This is a bit of a hack but since the vn_cache_strategy() function can
 * override a VFS's strategy function we must make sure that the bio, which
 * is probably bio2, doesn't leak an unexpected offset value back to the
 * filesystem.  The filesystem (e.g. UFS) might otherwise assume that the
 * bio went through its own file strategy function and the the bio2 offset
 * is a cached disk offset when, in fact, it isn't.
 */
static void
vn_cache_strategy_callback(struct bio *bio)
{
        bio->bio_offset = NOOFFSET;
        biodone(pop_bio(bio));
}

/*
 * bpdone:
 *
 *      Finish I/O on a buffer after all BIOs have been processed.
 *      Called when the bio chain is exhausted or by biowait.  If called
 *      by biowait, elseit is typically 0.
 *
 *      bpdone is also responsible for setting B_CACHE in a B_VMIO bp.
 *      In a non-VMIO bp, B_CACHE will be set on the next getblk() 
 *      assuming B_INVAL is clear.
 *
 *      For the VMIO case, we set B_CACHE if the op was a read and no
 *      read error occured, or if the op was a write.  B_CACHE is never
 *      set if the buffer is invalid or otherwise uncacheable.
 *
 *      bpdone does not mess with B_INVAL, allowing the I/O routine or the
 *      initiator to leave B_INVAL set to brelse the buffer out of existance
 *      in the biodone routine.
 *
 *      bpdone is responsible for calling bundirty() on the buffer after a
 *      successful write.  We previously did this prior to initiating the
 *      write under the assumption that the buffer might be dirtied again
 *      while the write was in progress, however doing it before-hand creates
 *      a race condition prior to the call to vn_strategy() where the
 *      filesystem may not be aware that a dirty buffer is present.
 *      It should not be possible for the buffer or its underlying pages to
 *      be redirtied prior to bpdone()'s unbusying of the underlying VM
 *      pages.
 */
void
bpdone(struct buf *bp, int elseit)
{
        buf_cmd_t cmd;

        KASSERT(BUF_REFCNTNB(bp) > 0, 
                ("bpdone: bp %p not busy %d", bp, BUF_REFCNTNB(bp)));
        KASSERT(bp->b_cmd != BUF_CMD_DONE, 
                ("bpdone: bp %p already done!", bp));

        /*
         * No more BIOs are left.  All completion functions have been dealt
         * with, now we clean up the buffer.
         */
        cmd = bp->b_cmd;
        bp->b_cmd = BUF_CMD_DONE;

        /*
         * Only reads and writes are processed past this point.
         */
        if (cmd != BUF_CMD_READ && cmd != BUF_CMD_WRITE) {
                if (cmd == BUF_CMD_FREEBLKS)
                        bp->b_flags |= B_NOCACHE;
                if (elseit)
                        brelse(bp);
                return;
        }

        /*
         * A failed write must re-dirty the buffer unless B_INVAL
         * was set.
         *
         * A successful write must clear the dirty flag.  This is done after
         * the write to ensure that the buffer remains on the vnode's dirty
         * list for filesystem interlocks / checks until the write is actually
         * complete.  HAMMER2 is sensitive to this issue.
         *
         * Only applicable to normal buffers (with VPs).  vinum buffers may
         * not have a vp.
         *
         * Must be done prior to calling buf_complete() as the callback might
         * re-dirty the buffer.
         */
        if (cmd == BUF_CMD_WRITE) {
                if ((bp->b_flags & (B_ERROR | B_INVAL)) == B_ERROR) {
                        bp->b_flags &= ~B_NOCACHE;
                        if (bp->b_vp)
                                bdirty(bp);
                } else {
                        if (bp->b_vp)
                                bundirty(bp);
                }
        }

        /*
         * Warning: softupdates may re-dirty the buffer, and HAMMER can do
         * a lot worse.  XXX - move this above the clearing of b_cmd
         */
        if (LIST_FIRST(&bp->b_dep) != NULL)
                buf_complete(bp);

        if (bp->b_flags & B_VMIO) {
                int i;
                vm_ooffset_t foff;
                vm_page_t m;
                vm_object_t obj;
                int iosize;
                struct vnode *vp = bp->b_vp;

                obj = vp->v_object;

#if defined(VFS_BIO_DEBUG)
                if (vp->v_auxrefs == 0)
                        panic("bpdone: zero vnode hold count");
                if ((vp->v_flag & VOBJBUF) == 0)
                        panic("bpdone: vnode is not setup for merged cache");
#endif

                foff = bp->b_loffset;
                KASSERT(foff != NOOFFSET, ("bpdone: no buffer offset"));
                KASSERT(obj != NULL, ("bpdone: missing VM object"));

#if defined(VFS_BIO_DEBUG)
                if (obj->paging_in_progress < bp->b_xio.xio_npages) {
                        kprintf("bpdone: paging in progress(%d) < "
                                "bp->b_xio.xio_npages(%d)\n",
                                obj->paging_in_progress,
                                bp->b_xio.xio_npages);
                }
#endif

                /*
                 * Set B_CACHE if the op was a normal read and no error
                 * occured.  B_CACHE is set for writes in the b*write()
                 * routines.
                 */
                iosize = bp->b_bcount - bp->b_resid;
                if (cmd == BUF_CMD_READ &&
                    (bp->b_flags & (B_INVAL|B_NOCACHE|B_ERROR)) == 0) {
                        bp->b_flags |= B_CACHE;
                }

                vm_object_hold(obj);
                for (i = 0; i < bp->b_xio.xio_npages; i++) {
                        int resid;
                        int isbogus;

                        resid = ((foff + PAGE_SIZE) & ~(off_t)PAGE_MASK) - foff;
                        if (resid > iosize)
                                resid = iosize;

                        /*
                         * cleanup bogus pages, restoring the originals.  Since
                         * the originals should still be wired, we don't have
                         * to worry about interrupt/freeing races destroying
                         * the VM object association.
                         */
                        m = bp->b_xio.xio_pages[i];
                        if (m == bogus_page) {
                                if ((bp->b_flags & B_HASBOGUS) == 0)
                                        panic("bpdone: bp %p corrupt bogus", bp);
                                m = vm_page_lookup(obj, OFF_TO_IDX(foff));
                                if (m == NULL)
                                        panic("bpdone: page disappeared");
                                bp->b_xio.xio_pages[i] = m;
                                isbogus = 1;
                        } else {
                                isbogus = 0;
                        }
#if defined(VFS_BIO_DEBUG)
                        if (OFF_TO_IDX(foff) != m->pindex) {
                                kprintf("bpdone: foff(%lu)/m->pindex(%ld) "
                                        "mismatch\n",
                                        (unsigned long)foff, (long)m->pindex);
                        }
#endif

                        /*
                         * In the write case, the valid and clean bits are
                         * already changed correctly (see bdwrite()), so we
                         * only need to do this here in the read case.
                         */
                        vm_page_busy_wait(m, FALSE, "bpdpgw");
                        if (cmd == BUF_CMD_READ && isbogus == 0 && resid > 0)
                                vfs_clean_one_page(bp, i, m);

                        /*
                         * when debugging new filesystems or buffer I/O
                         * methods, this is the most common error that pops
                         * up.  if you see this, you have not set the page
                         * busy flag correctly!!!
                         */
                        if ((m->busy_count & PBUSY_MASK) == 0) {
                                kprintf("bpdone: page busy < 0, "
                                    "pindex: %d, foff: 0x(%x,%x), "
                                    "resid: %d, index: %d\n",
                                    (int) m->pindex, (int)(foff >> 32),
                                                (int) foff & 0xffffffff, resid, i);
                                if (!vn_isdisk(vp, NULL))
                                        kprintf(" iosize: %ld, loffset: %lld, "
                                                "flags: 0x%08x, npages: %d\n",
                                            bp->b_vp->v_mount->mnt_stat.f_iosize,
                                            (long long)bp->b_loffset,
                                            bp->b_flags, bp->b_xio.xio_npages);
                                else
                                        kprintf(" VDEV, loffset: %lld, flags: 0x%08x, npages: %d\n",
                                            (long long)bp->b_loffset,
                                            bp->b_flags, bp->b_xio.xio_npages);
                                kprintf(" valid: 0x%x, dirty: 0x%x, "
                                        "wired: %d\n",
                                        m->valid, m->dirty,
                                        m->wire_count);
                                panic("bpdone: page busy < 0");
                        }
                        vm_page_io_finish(m);
                        vm_page_wakeup(m);
                        vm_object_pip_wakeup(obj);
                        foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK;
                        iosize -= resid;
                }
                if (bp->b_flags & B_HASBOGUS) {
                        pmap_qenter_noinval(trunc_page((vm_offset_t)bp->b_data),
                                            bp->b_xio.xio_pages,
                                            bp->b_xio.xio_npages);
                        bp->b_flags &= ~B_HASBOGUS;
                        bkvareset(bp);
                }
                vm_object_drop(obj);
        }

        /*
         * Finish up by releasing the buffer.  There are no more synchronous
         * or asynchronous completions, those were handled by bio_done
         * callbacks.
         */
        if (elseit) {
                if (bp->b_flags & (B_NOCACHE|B_INVAL|B_ERROR|B_RELBUF))
                        brelse(bp);
                else
                        bqrelse(bp);
        }
}

/*
 * Normal biodone.
 */
void
biodone(struct bio *bio)
{
        struct buf *bp = bio->bio_buf;

        runningbufwakeup(bp);

        /*
         * Run up the chain of BIO's.   Leave b_cmd intact for the duration.
         */
        while (bio) {
                biodone_t *done_func;
                struct bio_track *track;

                /*
                 * BIO tracking.  Most but not all BIOs are tracked.
                 */
                if ((track = bio->bio_track) != NULL) {
                        bio_track_rel(track);
                        bio->bio_track = NULL;
                }

                /*
                 * A bio_done function terminates the loop.  The function
                 * will be responsible for any further chaining and/or
                 * buffer management.
                 *
                 * WARNING!  The done function can deallocate the buffer!
                 */
                if ((done_func = bio->bio_done) != NULL) {
                        bio->bio_done = NULL;
                        done_func(bio);
                        return;
                }
                bio = bio->bio_prev;
        }

        /*
         * If we've run out of bio's do normal [a]synchronous completion.
         */
        bpdone(bp, 1);
}

/*
 * Synchronous biodone - this terminates a synchronous BIO.
 *
 * bpdone() is called with elseit=FALSE, leaving the buffer completed
 * but still locked.  The caller must brelse() the buffer after waiting
 * for completion.
 */
void
biodone_sync(struct bio *bio)
{
        struct buf *bp = bio->bio_buf;
        int flags;
        int nflags;

        KKASSERT(bio == &bp->b_bio1);
        bpdone(bp, 0);

        for (;;) {
                flags = bio->bio_flags;
                nflags = (flags | BIO_DONE) & ~BIO_WANT;

                if (atomic_cmpset_int(&bio->bio_flags, flags, nflags)) {
                        if (flags & BIO_WANT)
                                wakeup(bio);
                        break;
                }
        }
}

/*
 * vfs_unbusy_pages:
 *
 *      This routine is called in lieu of iodone in the case of
 *      incomplete I/O.  This keeps the busy status for pages
 *      consistant.
 */
void
vfs_unbusy_pages(struct buf *bp)
{
        int i;

        runningbufwakeup(bp);

        if (bp->b_flags & B_VMIO) {
                struct vnode *vp = bp->b_vp;
                vm_object_t obj;

                obj = vp->v_object;
                vm_object_hold(obj);

                for (i = 0; i < bp->b_xio.xio_npages; i++) {
                        vm_page_t m = bp->b_xio.xio_pages[i];

                        /*
                         * When restoring bogus changes the original pages
                         * should still be wired, so we are in no danger of
                         * losing the object association and do not need
                         * critical section protection particularly.
                         */
                        if (m == bogus_page) {
                                m = vm_page_lookup(obj, OFF_TO_IDX(bp->b_loffset) + i);
                                if (!m) {
                                        panic("vfs_unbusy_pages: page missing");
                                }
                                bp->b_xio.xio_pages[i] = m;
                        }
                        vm_page_busy_wait(m, FALSE, "bpdpgw");
                        vm_page_io_finish(m);
                        vm_page_wakeup(m);
                        vm_object_pip_wakeup(obj);
                }
                if (bp->b_flags & B_HASBOGUS) {
                        pmap_qenter_noinval(trunc_page((vm_offset_t)bp->b_data),
                                            bp->b_xio.xio_pages,
                                            bp->b_xio.xio_npages);
                        bp->b_flags &= ~B_HASBOGUS;
                        bkvareset(bp);
                }
                vm_object_drop(obj);
        }
}

/*
 * vfs_busy_pages:
 *
 *      This routine is called before a device strategy routine.
 *      It is used to tell the VM system that paging I/O is in
 *      progress, and treat the pages associated with the buffer
 *      almost as being PBUSY_LOCKED.  Also the object 'paging_in_progress'
 *      flag is handled to make sure that the object doesn't become
 *      inconsistant.
 *
 *      Since I/O has not been initiated yet, certain buffer flags
 *      such as B_ERROR or B_INVAL may be in an inconsistant state
 *      and should be ignored.
 */
void
vfs_busy_pages(struct vnode *vp, struct buf *bp)
{
        int i, bogus;
        struct lwp *lp = curthread->td_lwp;

        /*
         * The buffer's I/O command must already be set.  If reading,
         * B_CACHE must be 0 (double check against callers only doing
         * I/O when B_CACHE is 0).
         */
        KKASSERT(bp->b_cmd != BUF_CMD_DONE);
        KKASSERT(bp->b_cmd == BUF_CMD_WRITE || (bp->b_flags & B_CACHE) == 0);

        if (bp->b_flags & B_VMIO) {
                vm_object_t obj;

                obj = vp->v_object;
                KASSERT(bp->b_loffset != NOOFFSET,
                        ("vfs_busy_pages: no buffer offset"));

                /*
                 * Busy all the pages.  We have to busy them all at once
                 * to avoid deadlocks.
                 */
retry:
                for (i = 0; i < bp->b_xio.xio_npages; i++) {
                        vm_page_t m = bp->b_xio.xio_pages[i];

                        if (vm_page_busy_try(m, FALSE)) {
                                vm_page_sleep_busy(m, FALSE, "vbpage");
                                while (--i >= 0)
                                        vm_page_wakeup(bp->b_xio.xio_pages[i]);
                                goto retry;
                        }
                }

                /*
                 * Setup for I/O, soft-busy the page right now because
                 * the next loop may block.
                 */
                for (i = 0; i < bp->b_xio.xio_npages; i++) {
                        vm_page_t m = bp->b_xio.xio_pages[i];

                        if ((bp->b_flags & B_CLUSTER) == 0) {
                                vm_object_pip_add(obj, 1);
                                vm_page_io_start(m);
                        }
                }

                /*
                 * Adjust protections for I/O and do bogus-page mapping.
                 * Assume that vm_page_protect() can block (it can block
                 * if VM_PROT_NONE, don't take any chances regardless).
                 *
                 * In particular note that for writes we must incorporate
                 * page dirtyness from the VM system into the buffer's
                 * dirty range.
                 *
                 * For reads we theoretically must incorporate page dirtyness
                 * from the VM system to determine if the page needs bogus
                 * replacement, but we shortcut the test by simply checking
                 * that all m->valid bits are set, indicating that the page
                 * is fully valid and does not need to be re-read.  For any
                 * VM system dirtyness the page will also be fully valid
                 * since it was mapped at one point.
                 */
                bogus = 0;
                for (i = 0; i < bp->b_xio.xio_npages; i++) {
                        vm_page_t m = bp->b_xio.xio_pages[i];

                        if (bp->b_cmd == BUF_CMD_WRITE) {
                                /*
                                 * When readying a vnode-backed buffer for
                                 * a write we must zero-fill any invalid
                                 * portions of the backing VM pages, mark
                                 * it valid and clear related dirty bits.
                                 *
                                 * vfs_clean_one_page() incorporates any
                                 * VM dirtyness and updates the b_dirtyoff
                                 * range (after we've made the page RO).
                                 *
                                 * It is also expected that the pmap modified
                                 * bit has already been cleared by the
                                 * vm_page_protect().  We may not be able
                                 * to clear all dirty bits for a page if it
                                 * was also memory mapped (NFS).
                                 *
                                 * Finally be sure to unassign any swap-cache
                                 * backing store as it is now stale.
                                 */
                                vm_page_protect(m, VM_PROT_READ);
                                vfs_clean_one_page(bp, i, m);
                                swap_pager_unswapped(m);
                        } else if (m->valid == VM_PAGE_BITS_ALL) {
                                /*
                                 * When readying a vnode-backed buffer for
                                 * read we must replace any dirty pages with
                                 * a bogus page so dirty data is not destroyed
                                 * when filling gaps.
                                 *
                                 * To avoid testing whether the page is
                                 * dirty we instead test that the page was
                                 * at some point mapped (m->valid fully
                                 * valid) with the understanding that
                                 * this also covers the dirty case.
                                 */
                                bp->b_xio.xio_pages[i] = bogus_page;
                                bp->b_flags |= B_HASBOGUS;
                                bogus++;
                        } else if (m->valid & m->dirty) {
                                /*
                                 * This case should not occur as partial
                                 * dirtyment can only happen if the buffer
                                 * is B_CACHE, and this code is not entered
                                 * if the buffer is B_CACHE.
                                 */
                                kprintf("Warning: vfs_busy_pages - page not "
                                        "fully valid! loff=%jx bpf=%08x "
                                        "idx=%d val=%02x dir=%02x\n",
                                        (uintmax_t)bp->b_loffset, bp->b_flags,
                                        i, m->valid, m->dirty);
                                vm_page_protect(m, VM_PROT_NONE);
                        } else {
                                /*
                                 * The page is not valid and can be made
                                 * part of the read.
                                 */
                                vm_page_protect(m, VM_PROT_NONE);
                        }
                        vm_page_wakeup(m);
                }
                if (bogus) {
                        pmap_qenter_noinval(trunc_page((vm_offset_t)bp->b_data),
                                            bp->b_xio.xio_pages,
                                            bp->b_xio.xio_npages);
                        bkvareset(bp);
                }
        }

        /*
         * This is the easiest place to put the process accounting for the I/O
         * for now.
         */
        if (lp != NULL) {
                if (bp->b_cmd == BUF_CMD_READ)
                        lp->lwp_ru.ru_inblock++;
                else
                        lp->lwp_ru.ru_oublock++;
        }
}

/*
 * Tell the VM system that the pages associated with this buffer
 * are clean.  This is used for delayed writes where the data is
 * going to go to disk eventually without additional VM intevention.
 *
 * NOTE: While we only really need to clean through to b_bcount, we
 *       just go ahead and clean through to b_bufsize.
 */
static void
vfs_clean_pages(struct buf *bp)
{
        vm_page_t m;
        int i;

        if ((bp->b_flags & B_VMIO) == 0)
                return;

        KASSERT(bp->b_loffset != NOOFFSET,
                ("vfs_clean_pages: no buffer offset"));

        for (i = 0; i < bp->b_xio.xio_npages; i++) {
                m = bp->b_xio.xio_pages[i];
                vfs_clean_one_page(bp, i, m);
        }
}

/*
 * vfs_clean_one_page:
 *
 *      Set the valid bits and clear the dirty bits in a page within a
 *      buffer.  The range is restricted to the buffer's size and the
 *      buffer's logical offset might index into the first page.
 *
 *      The caller has busied or soft-busied the page and it is not mapped,
 *      test and incorporate the dirty bits into b_dirtyoff/end before
 *      clearing them.  Note that we need to clear the pmap modified bits
 *      after determining the the page was dirty, vm_page_set_validclean()
 *      does not do it for us.
 *
 *      This routine is typically called after a read completes (dirty should
 *      be zero in that case as we are not called on bogus-replace pages),
 *      or before a write is initiated.
 */
static void
vfs_clean_one_page(struct buf *bp, int pageno, vm_page_t m)
{
        int bcount;
        int xoff;
        int soff;
        int eoff;

        /*
         * Calculate offset range within the page but relative to buffer's
         * loffset.  loffset might be offset into the first page.
         */
        xoff = (int)bp->b_loffset & PAGE_MASK;  /* loffset offset into pg 0 */
        bcount = bp->b_bcount + xoff;           /* offset adjusted */

        if (pageno == 0) {
                soff = xoff;
                eoff = PAGE_SIZE;
        } else {
                soff = (pageno << PAGE_SHIFT);
                eoff = soff + PAGE_SIZE;
        }
        if (eoff > bcount)
                eoff = bcount;
        if (soff >= eoff)
                return;

        /*
         * Test dirty bits and adjust b_dirtyoff/end.
         *
         * If dirty pages are incorporated into the bp any prior
         * B_NEEDCOMMIT state (NFS) must be cleared because the
         * caller has not taken into account the new dirty data.
         *
         * If the page was memory mapped the dirty bits might go beyond the
         * end of the buffer, but we can't really make the assumption that
         * a file EOF straddles the buffer (even though this is the case for
         * NFS if B_NEEDCOMMIT is also set).  So for the purposes of clearing
         * B_NEEDCOMMIT we only test the dirty bits covered by the buffer.
         * This also saves some console spam.
         *
         * When clearing B_NEEDCOMMIT we must also clear B_CLUSTEROK,
         * NFS can handle huge commits but not huge writes.
         */
        vm_page_test_dirty(m);
        if (m->dirty) {
                if ((bp->b_flags & B_NEEDCOMMIT) &&
                    (m->dirty & vm_page_bits(soff & PAGE_MASK, eoff - soff))) {
                        if (debug_commit)
                                kprintf("Warning: vfs_clean_one_page: bp %p "
                                    "loff=%jx,%d flgs=%08x clr B_NEEDCOMMIT"
                                    " cmd %d vd %02x/%02x x/s/e %d %d %d "
                                    "doff/end %d %d\n",
                                    bp, (uintmax_t)bp->b_loffset, bp->b_bcount,
                                    bp->b_flags, bp->b_cmd,
                                    m->valid, m->dirty, xoff, soff, eoff,
                                    bp->b_dirtyoff, bp->b_dirtyend);
                        bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
                        if (debug_commit)
                                print_backtrace(-1);
                }
                /*
                 * Only clear the pmap modified bits if ALL the dirty bits
                 * are set, otherwise the system might mis-clear portions
                 * of a page.
                 */
                if (m->dirty == VM_PAGE_BITS_ALL &&
                    (bp->b_flags & B_NEEDCOMMIT) == 0) {
                        pmap_clear_modify(m);
                }
                if (bp->b_dirtyoff > soff - xoff)
                        bp->b_dirtyoff = soff - xoff;
                if (bp->b_dirtyend < eoff - xoff)
                        bp->b_dirtyend = eoff - xoff;
        }

        /*
         * Set related valid bits, clear related dirty bits.
         * Does not mess with the pmap modified bit.
         *
         * WARNING!  We cannot just clear all of m->dirty here as the
         *           buffer cache buffers may use a DEV_BSIZE'd aligned
         *           block size, or have an odd size (e.g. NFS at file EOF).
         *           The putpages code can clear m->dirty to 0.
         *
         *           If a VOP_WRITE generates a buffer cache buffer which
         *           covers the same space as mapped writable pages the
         *           buffer flush might not be able to clear all the dirty
         *           bits and still require a putpages from the VM system
         *           to finish it off.
         *
         * WARNING!  vm_page_set_validclean() currently assumes vm_token
         *           is held.  The page might not be busied (bdwrite() case).
         *           XXX remove this comment once we've validated that this
         *           is no longer an issue.
         */
        vm_page_set_validclean(m, soff & PAGE_MASK, eoff - soff);
}

#if 0
/*
 * Similar to vfs_clean_one_page() but sets the bits to valid and dirty.
 * The page data is assumed to be valid (there is no zeroing here).
 */
static void
vfs_dirty_one_page(struct buf *bp, int pageno, vm_page_t m)
{
        int bcount;
        int xoff;
        int soff;
        int eoff;

        /*
         * Calculate offset range within the page but relative to buffer's
         * loffset.  loffset might be offset into the first page.
         */
        xoff = (int)bp->b_loffset & PAGE_MASK;  /* loffset offset into pg 0 */
        bcount = bp->b_bcount + xoff;           /* offset adjusted */

        if (pageno == 0) {
                soff = xoff;
                eoff = PAGE_SIZE;
        } else {
                soff = (pageno << PAGE_SHIFT);
                eoff = soff + PAGE_SIZE;
        }
        if (eoff > bcount)
                eoff = bcount;
        if (soff >= eoff)
                return;
        vm_page_set_validdirty(m, soff & PAGE_MASK, eoff - soff);
}
#endif

/*
 * vfs_bio_clrbuf:
 *
 *      Clear a buffer.  This routine essentially fakes an I/O, so we need
 *      to clear B_ERROR and B_INVAL.
 *
 *      Note that while we only theoretically need to clear through b_bcount,
 *      we go ahead and clear through b_bufsize.
 */
void
vfs_bio_clrbuf(struct buf *bp)
{
        int i, mask = 0;
        caddr_t sa, ea;
        KKASSERT(bp->b_flags & B_VMIO);

        bp->b_flags &= ~(B_INVAL | B_EINTR | B_ERROR);
        bkvasync(bp);

        if ((bp->b_xio.xio_npages == 1) && (bp->b_bufsize < PAGE_SIZE) &&
            (bp->b_loffset & PAGE_MASK) == 0) {
                mask = (1 << (bp->b_bufsize / DEV_BSIZE)) - 1;
                if ((bp->b_xio.xio_pages[0]->valid & mask) == mask) {
                        bp->b_resid = 0;
                        return;
                }
                if ((bp->b_xio.xio_pages[0]->valid & mask) == 0) {
                        bzero(bp->b_data, bp->b_bufsize);
                        bp->b_xio.xio_pages[0]->valid |= mask;
                        bp->b_resid = 0;
                        return;
                }
        }
        sa = bp->b_data;
        for(i = 0; i < bp->b_xio.xio_npages; i++, sa=ea) {
                int j = ((vm_offset_t)sa & PAGE_MASK) / DEV_BSIZE;
                ea = (caddr_t)trunc_page((vm_offset_t)sa + PAGE_SIZE);
                ea = (caddr_t)(vm_offset_t)ulmin(
                            (u_long)(vm_offset_t)ea,
                            (u_long)(vm_offset_t)bp->b_data + bp->b_bufsize);
                mask = ((1 << ((ea - sa) / DEV_BSIZE)) - 1) << j;
                if ((bp->b_xio.xio_pages[i]->valid & mask) == mask)
                        continue;
                if ((bp->b_xio.xio_pages[i]->valid & mask) == 0) {
                        bzero(sa, ea - sa);
                } else {
                        for (; sa < ea; sa += DEV_BSIZE, j++) {
                                if ((bp->b_xio.xio_pages[i]->valid &
                                    (1<<j)) == 0) {
                                        bzero(sa, DEV_BSIZE);
                                }
                        }
                }
                bp->b_xio.xio_pages[i]->valid |= mask;
        }
        bp->b_resid = 0;
}

/*
 * Allocate a page for a buffer cache buffer.
 *
 * If NULL is returned the caller is expected to retry (typically check if
 * the page already exists on retry before trying to allocate one).
 *
 * NOTE! Low-memory handling is dealt with in b[q]relse(), not here.  This
 *       function will use the system reserve with the hope that the page
 *       allocations can be returned to PQ_CACHE/PQ_FREE when the caller
 *       is done with the buffer.
 *
 * NOTE! However, TMPFS is a special case because flushing a dirty buffer
 *       to TMPFS doesn't clean the page.  For TMPFS, only the pagedaemon
 *       is capable of retiring pages (to swap).  For TMPFS we don't dig
 *       into the system reserve because doing so could stall out pretty
 *       much every process running on the system.
 */
static
vm_page_t
bio_page_alloc(struct buf *bp, vm_object_t obj, vm_pindex_t pg, int deficit)
{
        int vmflags = VM_ALLOC_NORMAL | VM_ALLOC_NULL_OK;
        vm_page_t p;

        ASSERT_LWKT_TOKEN_HELD(vm_object_token(obj));

        /*
         * Try a normal allocation first.
         */
        p = vm_page_alloc(obj, pg, vmflags);
        if (p)
                return(p);
        if (vm_page_lookup(obj, pg))
                return(NULL);
        vm_pageout_deficit += deficit;

        /*
         * Try again, digging into the system reserve.
         *
         * Trying to recover pages from the buffer cache here can deadlock
         * against other threads trying to busy underlying pages so we
         * depend on the code in brelse() and bqrelse() to free/cache the
         * underlying buffer cache pages when memory is low.
         */
        if (curthread->td_flags & TDF_SYSTHREAD)
                vmflags |= VM_ALLOC_SYSTEM | VM_ALLOC_INTERRUPT;
        else if (bp->b_vp && bp->b_vp->v_tag == VT_TMPFS)
                vmflags |= 0;
        else
                vmflags |= VM_ALLOC_SYSTEM;

        /*recoverbufpages();*/
        p = vm_page_alloc(obj, pg, vmflags);
        if (p)
                return(p);
        if (vm_page_lookup(obj, pg))
                return(NULL);

        /*
         * Wait for memory to free up and try again
         */
        if (vm_page_count_severe())
                ++lowmempgallocs;
        vm_wait(hz / 20 + 1);

        p = vm_page_alloc(obj, pg, vmflags);
        if (p)
                return(p);
        if (vm_page_lookup(obj, pg))
                return(NULL);

        /*
         * Ok, now we are really in trouble.
         */
        if (bootverbose) {
                static struct krate biokrate = { .freq = 1 };
                krateprintf(&biokrate,
                            "Warning: bio_page_alloc: memory exhausted "
                            "during buffer cache page allocation from %s\n",
                            curthread->td_comm);
        }
        if (curthread->td_flags & TDF_SYSTHREAD)
                vm_wait(hz / 20 + 1);
        else
                vm_wait(hz / 2 + 1);
        return (NULL);
}

/*
 * The buffer's mapping has changed.  Adjust the buffer's memory
 * synchronization.  The caller is the exclusive holder of the buffer
 * and has set or cleared B_KVABIO according to preference.
 *
 * WARNING! If the caller is using B_KVABIO mode, this function will
 *          not map the data to the current cpu.  The caller must also
 *          call bkvasync(bp).
 */
void
bkvareset(struct buf *bp)
{
        if (bp->b_flags & B_KVABIO) {
                CPUMASK_ASSZERO(bp->b_cpumask);
        } else {
                CPUMASK_ORMASK(bp->b_cpumask, smp_active_mask);
                smp_invltlb();
                cpu_invltlb();
        }
}

/*
 * The buffer will be used by the caller on the caller's cpu, synchronize
 * its data to the current cpu.
 *
 * If B_KVABIO is not set, the buffer is already fully synchronized.
 */
void
bkvasync(struct buf *bp)
{
        int cpuid = mycpu->gd_cpuid;
        char *bdata;

        if ((bp->b_flags & B_KVABIO) &&
            CPUMASK_TESTBIT(bp->b_cpumask, cpuid) == 0) {
                bdata = bp->b_data;
                while (bdata < bp->b_data + bp->b_bufsize) {
                        cpu_invlpg(bdata);
                        bdata += PAGE_SIZE -
                                 ((intptr_t)bdata & PAGE_MASK);
                }
                ATOMIC_CPUMASK_ORBIT(bp->b_cpumask, cpuid);
        }
}

/*
 * The buffer will be used by a subsystem that does not understand
 * the KVABIO API.  Make sure its data is synchronized to all cpus.
 *
 * If B_KVABIO is not set, the buffer is already fully synchronized.
 *
 * NOTE! This is the only safe way to clear B_KVABIO on a buffer.
 */
void
bkvasync_all(struct buf *bp)
{
        if ((bp->b_flags & B_KVABIO) &&
            CPUMASK_CMPMASKNEQ(bp->b_cpumask, smp_active_mask)) {
                smp_invltlb();
                cpu_invltlb();
                ATOMIC_CPUMASK_ORMASK(bp->b_cpumask, smp_active_mask);
        }
        bp->b_flags &= ~B_KVABIO;
}

/*
 * Scan all buffers in the system and issue the callback.
 */
int
scan_all_buffers(int (*callback)(struct buf *, void *), void *info)
{
        int count = 0;
        int error;
        long n;

        for (n = 0; n < nbuf; ++n) {
                if ((error = callback(&buf[n], info)) < 0) {
                        count = error;
                        break;
                }
                count += error;
        }
        return (count);
}

/*
 * nestiobuf_iodone: biodone callback for nested buffers and propagate
 * completion to the master buffer.
 */
static void
nestiobuf_iodone(struct bio *bio)
{
        struct bio *mbio;
        struct buf *mbp, *bp;
        struct devstat *stats;
        int error;
        int donebytes;

        bp = bio->bio_buf;
        mbio = bio->bio_caller_info1.ptr;
        stats = bio->bio_caller_info2.ptr;
        mbp = mbio->bio_buf;

        KKASSERT(bp->b_bcount <= bp->b_bufsize);
        KKASSERT(mbp != bp);

        error = bp->b_error;
        if (bp->b_error == 0 &&
            (bp->b_bcount < bp->b_bufsize || bp->b_resid > 0)) {
                /*
                 * Not all got transfered, raise an error. We have no way to
                 * propagate these conditions to mbp.
                 */
                error = EIO;
        }

        donebytes = bp->b_bufsize;

        relpbuf(bp, NULL);

        nestiobuf_done(mbio, donebytes, error, stats);
}

void
nestiobuf_done(struct bio *mbio, int donebytes, int error, struct devstat *stats)
{
        struct buf *mbp;

        mbp = mbio->bio_buf;    

        KKASSERT((int)(intptr_t)mbio->bio_driver_info > 0);

        /*
         * If an error occured, propagate it to the master buffer.
         *
         * Several biodone()s may wind up running concurrently so
         * use an atomic op to adjust b_flags.
         */
        if (error) {
                mbp->b_error = error;
                atomic_set_int(&mbp->b_flags, B_ERROR);
        }

        /*
         * Decrement the operations in progress counter and terminate the
         * I/O if this was the last bit.
         */
        if (atomic_fetchadd_int((int *)&mbio->bio_driver_info, -1) == 1) {
                mbp->b_resid = 0;
                if (stats)
                        devstat_end_transaction_buf(stats, mbp);
                biodone(mbio);
        }
}

/*
 * Initialize a nestiobuf for use.  Set an initial count of 1 to prevent
 * the mbio from being biodone()'d while we are still adding sub-bios to
 * it.
 */
void
nestiobuf_init(struct bio *bio)
{
        bio->bio_driver_info = (void *)1;
}

/*
 * The BIOs added to the nestedio have already been started, remove the
 * count that placeheld our mbio and biodone() it if the count would
 * transition to 0.
 */
void
nestiobuf_start(struct bio *mbio)
{
        struct buf *mbp = mbio->bio_buf;

        /*
         * Decrement the operations in progress counter and terminate the
         * I/O if this was the last bit.
         */
        if (atomic_fetchadd_int((int *)&mbio->bio_driver_info, -1) == 1) {
                if (mbp->b_flags & B_ERROR)
                        mbp->b_resid = mbp->b_bcount;
                else
                        mbp->b_resid = 0;
                biodone(mbio);
        }
}

/*
 * Set an intermediate error prior to calling nestiobuf_start()
 */
void
nestiobuf_error(struct bio *mbio, int error)
{
        struct buf *mbp = mbio->bio_buf;

        if (error) {
                mbp->b_error = error;
                atomic_set_int(&mbp->b_flags, B_ERROR);
        }
}

/*
 * nestiobuf_add: setup a "nested" buffer.
 *
 * => 'mbp' is a "master" buffer which is being divided into sub pieces.
 * => 'bp' should be a buffer allocated by getiobuf.
 * => 'offset' is a byte offset in the master buffer.
 * => 'size' is a size in bytes of this nested buffer.
 */
void
nestiobuf_add(struct bio *mbio, struct buf *bp, int offset, size_t size, struct devstat *stats)
{
        struct buf *mbp = mbio->bio_buf;
        struct vnode *vp = mbp->b_vp;

        KKASSERT(mbp->b_bcount >= offset + size);

        atomic_add_int((int *)&mbio->bio_driver_info, 1);

        /* kernel needs to own the lock for it to be released in biodone */
        BUF_KERNPROC(bp);
        bp->b_vp = vp;
        bp->b_cmd = mbp->b_cmd;
        bp->b_bio1.bio_done = nestiobuf_iodone;
        bp->b_data = (char *)mbp->b_data + offset;
        bp->b_resid = bp->b_bcount = size;
        bp->b_bufsize = bp->b_bcount;

        bp->b_bio1.bio_track = NULL;
        bp->b_bio1.bio_caller_info1.ptr = mbio;
        bp->b_bio1.bio_caller_info2.ptr = stats;
}

#ifdef DDB

DB_SHOW_COMMAND(buffer, db_show_buffer)
{
        /* get args */
        struct buf *bp = (struct buf *)addr;

        if (!have_addr) {
                db_printf("usage: show buffer <addr>\n");
                return;
        }

        db_printf("b_flags = 0x%b\n", (u_int)bp->b_flags, PRINT_BUF_FLAGS);
        db_printf("b_cmd = %d\n", bp->b_cmd);
        db_printf("b_error = %d, b_bufsize = %d, b_bcount = %d, "
                  "b_resid = %d\n, b_data = %p, "
                  "bio_offset(disk) = %lld, bio_offset(phys) = %lld\n",
                  bp->b_error, bp->b_bufsize, bp->b_bcount, bp->b_resid,
                  bp->b_data,
                  (long long)bp->b_bio2.bio_offset,
                  (long long)(bp->b_bio2.bio_next ?
                                bp->b_bio2.bio_next->bio_offset : (off_t)-1));
        if (bp->b_xio.xio_npages) {
                int i;
                db_printf("b_xio.xio_npages = %d, pages(OBJ, IDX, PA): ",
                        bp->b_xio.xio_npages);
                for (i = 0; i < bp->b_xio.xio_npages; i++) {
                        vm_page_t m;
                        m = bp->b_xio.xio_pages[i];
                        db_printf("(%p, 0x%lx, 0x%lx)", (void *)m->object,
                            (u_long)m->pindex, (u_long)VM_PAGE_TO_PHYS(m));
                        if ((i + 1) < bp->b_xio.xio_npages)
                                db_printf(",");
                }
                db_printf("\n");
        }
}
#endif /* DDB */