common/os/watchpoint.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License, Version 1.0 only
 * (the "License").  You may not use this file except in compliance
 * with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#include <sys/types.h>
#include <sys/t_lock.h>
#include <sys/param.h>
#include <sys/cred.h>
#include <sys/debug.h>
#include <sys/inline.h>
#include <sys/kmem.h>
#include <sys/proc.h>
#include <sys/regset.h>
#include <sys/sysmacros.h>
#include <sys/systm.h>
#include <sys/prsystm.h>
#include <sys/buf.h>
#include <sys/signal.h>
#include <sys/user.h>
#include <sys/cpuvar.h>

#include <sys/fault.h>
#include <sys/syscall.h>
#include <sys/procfs.h>
#include <sys/cmn_err.h>
#include <sys/stack.h>
#include <sys/watchpoint.h>
#include <sys/copyops.h>
#include <sys/schedctl.h>

#include <sys/mman.h>
#include <vm/as.h>
#include <vm/seg.h>

/*
 * Copy ops vector for watchpoints.
 */
static int  watch_copyin(const void *, void *, size_t);
static int  watch_xcopyin(const void *, void *, size_t);
static int  watch_copyout(const void *, void *, size_t);
static int  watch_xcopyout(const void *, void *, size_t);
static int  watch_copyinstr(const char *, char *, size_t, size_t *);
static int  watch_copyoutstr(const char *, char *, size_t, size_t *);
static int  watch_fuword8(const void *, uint8_t *);
static int  watch_fuword16(const void *, uint16_t *);
static int  watch_fuword32(const void *, uint32_t *);
static int  watch_suword8(void *, uint8_t);
static int  watch_suword16(void *, uint16_t);
static int  watch_suword32(void *, uint32_t);
static int  watch_physio(int (*)(struct buf *), struct buf *,
    dev_t, int, void (*)(struct buf *), struct uio *);
#ifdef _LP64
static int  watch_fuword64(const void *, uint64_t *);
static int  watch_suword64(void *, uint64_t);
#endif

struct copyops watch_copyops = {
    watch_copyin,
    watch_xcopyin,
    watch_copyout,
    watch_xcopyout,
    watch_copyinstr,
    watch_copyoutstr,
    watch_fuword8,
    watch_fuword16,
    watch_fuword32,
#ifdef _LP64
    watch_fuword64,
#else
    NULL,
#endif
    watch_suword8,
    watch_suword16,
    watch_suword32,
#ifdef _LP64
    watch_suword64,
#else
    NULL,
#endif
    watch_physio
};

/*
 * Map the 'rw' argument to a protection flag.
 */
static int
rw_to_prot(enum seg_rw rw)
{
    switch (rw) {
    case S_EXEC:
        return (PROT_EXEC);
    case S_READ:
        return (PROT_READ);
    case S_WRITE:
        return (PROT_WRITE);
    default:
        return (PROT_NONE); /* can't happen */
    }
}

/*
 * Map the 'rw' argument to an index into an array of exec/write/read things.
 * The index follows the precedence order:  exec .. write .. read
 */
static int
rw_to_index(enum seg_rw rw)
{
    switch (rw) {
    default:    /* default case "can't happen" */
    case S_EXEC:
        return (0);
    case S_WRITE:
        return (1);
    case S_READ:
        return (2);
    }
}

/*
 * Map an index back to a seg_rw.
 */
static enum seg_rw S_rw[4] = {
    S_EXEC,
    S_WRITE,
    S_READ,
    S_READ,
};

#define X   0
#define W   1
#define R   2
#define sum(a)  (a[X] + a[W] + a[R])

/*
 * Common code for pr_mappage() and pr_unmappage().
 */
static int
pr_do_mappage(caddr_t addr, size_t size, int mapin, enum seg_rw rw, int kernel)
{
    proc_t *p = curproc;
    struct as *as = p->p_as;
    char *eaddr = addr + size;
    int prot_rw = rw_to_prot(rw);
    int xrw = rw_to_index(rw);
    int rv = 0;
    struct watched_page *pwp;
    struct watched_page tpw;
    avl_index_t where;
    uint_t prot;

    ASSERT(as != &kas);

startover:
    ASSERT(rv == 0);
    if (avl_numnodes(&as->a_wpage) == 0)
        return (0);

    /*
     * as->a_wpage can only be changed while the process is totally stopped.
     * Don't grab p_lock here.  Holding p_lock while grabbing the address
     * space lock leads to deadlocks with the clock thread.
     *
     * p_maplock prevents simultaneous execution of this function.  Under
     * normal circumstances, holdwatch() will stop all other threads, so the
     * lock isn't really needed.  But there may be multiple threads within
     * stop() when SWATCHOK is set, so we need to handle multiple threads
     * at once.  See holdwatch() for the details of this dance.
     */

    mutex_enter(&p->p_maplock);

    tpw.wp_vaddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK);
    if ((pwp = avl_find(&as->a_wpage, &tpw, &where)) == NULL)
        pwp = avl_nearest(&as->a_wpage, where, AVL_AFTER);

    for (; pwp != NULL && pwp->wp_vaddr < eaddr;
        pwp = AVL_NEXT(&as->a_wpage, pwp)) {

        /*
         * If the requested protection has not been
         * removed, we need not remap this page.
         */
        prot = pwp->wp_prot;
        if (kernel || (prot & PROT_USER))
            if (prot & prot_rw)
                continue;
        /*
         * If the requested access does not exist in the page's
         * original protections, we need not remap this page.
         * If the page does not exist yet, we can't test it.
         */
        if ((prot = pwp->wp_oprot) != 0) {
            if (!(kernel || (prot & PROT_USER)))
                continue;
            if (!(prot & prot_rw))
                continue;
        }

        if (mapin) {
            /*
             * Before mapping the page in, ensure that
             * all other lwps are held in the kernel.
             */
            if (p->p_mapcnt == 0) {
                mutex_exit(&p->p_maplock);
                if (holdwatch() != 0) {
                    /*
                     * We stopped in holdwatch().
                     * Start all over again because the
                     * watched page list may have changed.
                     */
                    goto startover;
                }
                mutex_enter(&p->p_maplock);
            }
            p->p_mapcnt++;
        }

        addr = pwp->wp_vaddr;
        rv++;

        prot = pwp->wp_prot;
        if (mapin) {
            if (kernel)
                pwp->wp_kmap[xrw]++;
            else
                pwp->wp_umap[xrw]++;
            pwp->wp_flags |= WP_NOWATCH;
            if (pwp->wp_kmap[X] + pwp->wp_umap[X])
                /* cannot have exec-only protection */
                prot |= PROT_READ|PROT_EXEC;
            if (pwp->wp_kmap[R] + pwp->wp_umap[R])
                prot |= PROT_READ;
            if (pwp->wp_kmap[W] + pwp->wp_umap[W])
                /* cannot have write-only protection */
                prot |= PROT_READ|PROT_WRITE;
#if 0   /* damned broken mmu feature! */
            if (sum(pwp->wp_umap) == 0)
                prot &= ~PROT_USER;
#endif
        } else {
            ASSERT(pwp->wp_flags & WP_NOWATCH);
            if (kernel) {
                ASSERT(pwp->wp_kmap[xrw] != 0);
                --pwp->wp_kmap[xrw];
            } else {
                ASSERT(pwp->wp_umap[xrw] != 0);
                --pwp->wp_umap[xrw];
            }
            if (sum(pwp->wp_kmap) + sum(pwp->wp_umap) == 0)
                pwp->wp_flags &= ~WP_NOWATCH;
            else {
                if (pwp->wp_kmap[X] + pwp->wp_umap[X])
                    /* cannot have exec-only protection */
                    prot |= PROT_READ|PROT_EXEC;
                if (pwp->wp_kmap[R] + pwp->wp_umap[R])
                    prot |= PROT_READ;
                if (pwp->wp_kmap[W] + pwp->wp_umap[W])
                    /* cannot have write-only protection */
                    prot |= PROT_READ|PROT_WRITE;
#if 0   /* damned broken mmu feature! */
                if (sum(pwp->wp_umap) == 0)
                    prot &= ~PROT_USER;
#endif
            }
        }


        if (pwp->wp_oprot != 0) {   /* if page exists */
            struct seg *seg;
            uint_t oprot;
            int err, retrycnt = 0;

            AS_LOCK_ENTER(as, RW_WRITER);
        retry:
            seg = as_segat(as, addr);
            ASSERT(seg != NULL);
            SEGOP_GETPROT(seg, addr, 0, &oprot);
            if (prot != oprot) {
                err = SEGOP_SETPROT(seg, addr, PAGESIZE, prot);
                if (err == IE_RETRY) {
                    ASSERT(retrycnt == 0);
                    retrycnt++;
                    goto retry;
                }
            }
            AS_LOCK_EXIT(as);
        }

        /*
         * When all pages are mapped back to their normal state,
         * continue the other lwps.
         */
        if (!mapin) {
            ASSERT(p->p_mapcnt > 0);
            p->p_mapcnt--;
            if (p->p_mapcnt == 0) {
                mutex_exit(&p->p_maplock);
                mutex_enter(&p->p_lock);
                continuelwps(p);
                mutex_exit(&p->p_lock);
                mutex_enter(&p->p_maplock);
            }
        }
    }

    mutex_exit(&p->p_maplock);

    return (rv);
}

/*
 * Restore the original page protections on an address range.
 * If 'kernel' is non-zero, just do it for the kernel.
 * pr_mappage() returns non-zero if it actually changed anything.
 *
 * pr_mappage() and pr_unmappage() must be executed in matched pairs,
 * but pairs may be nested within other pairs.  The reference counts
 * sort it all out.  See pr_do_mappage(), above.
 */
static int
pr_mappage(const caddr_t addr, size_t size, enum seg_rw rw, int kernel)
{
    return (pr_do_mappage(addr, size, 1, rw, kernel));
}

/*
 * Set the modified page protections on a watched page.
 * Inverse of pr_mappage().
 * Needs to be called only if pr_mappage() returned non-zero.
 */
static void
pr_unmappage(const caddr_t addr, size_t size, enum seg_rw rw, int kernel)
{
    (void) pr_do_mappage(addr, size, 0, rw, kernel);
}

/*
 * Function called by an lwp after it resumes from stop().
 */
void
setallwatch(void)
{
    proc_t *p = curproc;
    struct as *as = curproc->p_as;
    struct watched_page *pwp, *next;
    struct seg *seg;
    caddr_t vaddr;
    uint_t prot;
    int err, retrycnt;

    if (p->p_wprot == NULL)
        return;

    ASSERT(MUTEX_NOT_HELD(&curproc->p_lock));

    AS_LOCK_ENTER(as, RW_WRITER);

    pwp = p->p_wprot;
    while (pwp != NULL) {

        vaddr = pwp->wp_vaddr;
        retrycnt = 0;
    retry:
        ASSERT(pwp->wp_flags & WP_SETPROT);
        if ((seg = as_segat(as, vaddr)) != NULL &&
            !(pwp->wp_flags & WP_NOWATCH)) {
            prot = pwp->wp_prot;
            err = SEGOP_SETPROT(seg, vaddr, PAGESIZE, prot);
            if (err == IE_RETRY) {
                ASSERT(retrycnt == 0);
                retrycnt++;
                goto retry;
            }
        }

        next = pwp->wp_list;

        if (pwp->wp_read + pwp->wp_write + pwp->wp_exec == 0) {
            /*
             * No watched areas remain in this page.
             * Free the watched_page structure.
             */
            avl_remove(&as->a_wpage, pwp);
            kmem_free(pwp, sizeof (struct watched_page));
        } else {
            pwp->wp_flags &= ~WP_SETPROT;
        }

        pwp = next;
    }
    p->p_wprot = NULL;

    AS_LOCK_EXIT(as);
}


int
pr_is_watchpage_as(caddr_t addr, enum seg_rw rw, struct as *as)
{
    register struct watched_page *pwp;
    struct watched_page tpw;
    uint_t prot;
    int rv = 0;

    switch (rw) {
    case S_READ:
    case S_WRITE:
    case S_EXEC:
        break;
    default:
        return (0);
    }

    /*
     * as->a_wpage can only be modified while the process is totally
     * stopped.  We need, and should use, no locks here.
     */
    if (as != &kas && avl_numnodes(&as->a_wpage) != 0) {
        tpw.wp_vaddr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK);
        pwp = avl_find(&as->a_wpage, &tpw, NULL);
        if (pwp != NULL) {
            ASSERT(addr >= pwp->wp_vaddr &&
                addr < pwp->wp_vaddr + PAGESIZE);
            if (pwp->wp_oprot != 0) {
                prot = pwp->wp_prot;
                switch (rw) {
                case S_READ:
                    rv = ((prot & (PROT_USER|PROT_READ))
                        != (PROT_USER|PROT_READ));
                    break;
                case S_WRITE:
                    rv = ((prot & (PROT_USER|PROT_WRITE))
                        != (PROT_USER|PROT_WRITE));
                    break;
                case S_EXEC:
                    rv = ((prot & (PROT_USER|PROT_EXEC))
                        != (PROT_USER|PROT_EXEC));
                    break;
                default:
                    /* can't happen! */
                    break;
                }
            }
        }
    }

    return (rv);
}


/*
 * trap() calls here to determine if a fault is in a watched page.
 * We return nonzero if this is true and the load/store would fail.
 */
int
pr_is_watchpage(caddr_t addr, enum seg_rw rw)
{
    struct as *as = curproc->p_as;

    if ((as == &kas) || avl_numnodes(&as->a_wpage) == 0)
        return (0);

    return (pr_is_watchpage_as(addr, rw, as));
}


/*
 * trap() calls here to determine if a fault is a watchpoint.
 */
int
pr_is_watchpoint(caddr_t *paddr, int *pta, size_t size, size_t *plen,
    enum seg_rw rw)
{
    proc_t *p = curproc;
    caddr_t addr = *paddr;
    caddr_t eaddr = addr + size;
    register struct watched_area *pwa;
    struct watched_area twa;
    int rv = 0;
    int ta = 0;
    size_t len = 0;

    switch (rw) {
    case S_READ:
    case S_WRITE:
    case S_EXEC:
        break;
    default:
        *pta = 0;
        return (0);
    }

    /*
     * p->p_warea is protected by p->p_lock.
     */
    mutex_enter(&p->p_lock);

    /* BEGIN CSTYLED */
    /*
     * This loop is somewhat complicated because the fault region can span
     * multiple watched areas.  For example:
     *
     *            addr              eaddr
     *      +-----------------+
     *      | fault region    |
     *  +-------+--------+----+---+------------+
     *      | prot not right |    | prot correct   |
     *      +----------------+    +----------------+
     *    wa_vaddr        wa_eaddr
     *                wa_vaddr      wa_eaddr
     *
     * We start at the area greater than or equal to the starting address.
     * As long as some portion of the fault region overlaps the current
     * area, we continue checking permissions until we find an appropriate
     * match.
     */
    /* END CSTYLED */
    twa.wa_vaddr = addr;
    twa.wa_eaddr = eaddr;

    for (pwa = pr_find_watched_area(p, &twa, NULL);
        pwa != NULL && eaddr > pwa->wa_vaddr && addr < pwa->wa_eaddr;
        pwa = AVL_NEXT(&p->p_warea, pwa)) {

        switch (rw) {
        case S_READ:
            if (pwa->wa_flags & WA_READ)
                rv = TRAP_RWATCH;
            break;
        case S_WRITE:
            if (pwa->wa_flags & WA_WRITE)
                rv = TRAP_WWATCH;
            break;
        case S_EXEC:
            if (pwa->wa_flags & WA_EXEC)
                rv = TRAP_XWATCH;
            break;
        default:
            /* can't happen */
            break;
        }

        /*
         * If protections didn't match, check the next watched
         * area
         */
        if (rv != 0) {
            if (addr < pwa->wa_vaddr)
                addr = pwa->wa_vaddr;
            len = pwa->wa_eaddr - addr;
            if (pwa->wa_flags & WA_TRAPAFTER)
                ta = 1;
            break;
        }
    }

    mutex_exit(&p->p_lock);

    *paddr = addr;
    *pta = ta;
    if (plen != NULL)
        *plen = len;
    return (rv);
}

/*
 * Set up to perform a single-step at user level for the
 * case of a trapafter watchpoint.  Called from trap().
 */
void
do_watch_step(caddr_t vaddr, size_t sz, enum seg_rw rw,
    int watchcode, greg_t pc)
{
    register klwp_t *lwp = ttolwp(curthread);
    struct lwp_watch *pw = &lwp->lwp_watch[rw_to_index(rw)];

    /*
     * Check to see if we are already performing this special
     * watchpoint single-step.  We must not do pr_mappage() twice.
     */

    /* special check for two read traps on the same instruction */
    if (rw == S_READ && pw->wpaddr != NULL &&
        !(pw->wpaddr <= vaddr && vaddr < pw->wpaddr + pw->wpsize)) {
        ASSERT(lwp->lwp_watchtrap != 0);
        pw++;   /* use the extra S_READ struct */
    }

    if (pw->wpaddr != NULL) {
        ASSERT(lwp->lwp_watchtrap != 0);
        ASSERT(pw->wpaddr <= vaddr && vaddr < pw->wpaddr + pw->wpsize);
        if (pw->wpcode == 0) {
            pw->wpcode = watchcode;
            pw->wppc = pc;
        }
    } else {
        int mapped = pr_mappage(vaddr, sz, rw, 0);
        prstep(lwp, 1);
        lwp->lwp_watchtrap = 1;
        pw->wpaddr = vaddr;
        pw->wpsize = sz;
        pw->wpcode = watchcode;
        pw->wpmapped = mapped;
        pw->wppc = pc;
    }
}

/*
 * Undo the effects of do_watch_step().
 * Called from trap() after the single-step is finished.
 * Also called from issig_forreal() and stop() with a NULL
 * argument to avoid having these things set more than once.
 */
int
undo_watch_step(k_siginfo_t *sip)
{
    register klwp_t *lwp = ttolwp(curthread);
    int fault = 0;

    if (lwp->lwp_watchtrap) {
        struct lwp_watch *pw = lwp->lwp_watch;
        int i;

        for (i = 0; i < 4; i++, pw++) {
            if (pw->wpaddr == NULL)
                continue;
            if (pw->wpmapped)
                pr_unmappage(pw->wpaddr, pw->wpsize, S_rw[i],
                    0);
            if (pw->wpcode != 0) {
                if (sip != NULL) {
                    sip->si_signo = SIGTRAP;
                    sip->si_code = pw->wpcode;
                    sip->si_addr = pw->wpaddr;
                    sip->si_trapafter = 1;
                    sip->si_pc = (caddr_t)pw->wppc;
                }
                fault = FLTWATCH;
                pw->wpcode = 0;
            }
            pw->wpaddr = NULL;
            pw->wpsize = 0;
            pw->wpmapped = 0;
        }
        lwp->lwp_watchtrap = 0;
    }

    return (fault);
}

/*
 * Handle a watchpoint that occurs while doing copyin()
 * or copyout() in a system call.
 * Return non-zero if the fault or signal is cleared
 * by a debugger while the lwp is stopped.
 */
static int
sys_watchpoint(caddr_t addr, int watchcode, int ta)
{
    extern greg_t getuserpc(void);  /* XXX header file */
    k_sigset_t smask;
    register proc_t *p = ttoproc(curthread);
    register klwp_t *lwp = ttolwp(curthread);
    register sigqueue_t *sqp;
    int rval;

    /* assert no locks are held */
    /* ASSERT(curthread->t_nlocks == 0); */

    sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP);
    sqp->sq_info.si_signo = SIGTRAP;
    sqp->sq_info.si_code = watchcode;
    sqp->sq_info.si_addr = addr;
    sqp->sq_info.si_trapafter = ta;
    sqp->sq_info.si_pc = (caddr_t)getuserpc();

    mutex_enter(&p->p_lock);

    /* this will be tested and cleared by the caller */
    lwp->lwp_sysabort = 0;

    if (prismember(&p->p_fltmask, FLTWATCH)) {
        lwp->lwp_curflt = (uchar_t)FLTWATCH;
        lwp->lwp_siginfo = sqp->sq_info;
        stop(PR_FAULTED, FLTWATCH);
        if (lwp->lwp_curflt == 0) {
            mutex_exit(&p->p_lock);
            kmem_free(sqp, sizeof (sigqueue_t));
            return (1);
        }
        lwp->lwp_curflt = 0;
    }

    /*
     * post the SIGTRAP signal.
     * Block all other signals so we only stop showing SIGTRAP.
     */
    if (signal_is_blocked(curthread, SIGTRAP) ||
        sigismember(&p->p_ignore, SIGTRAP)) {
        /* SIGTRAP is blocked or ignored, forget the rest. */
        mutex_exit(&p->p_lock);
        kmem_free(sqp, sizeof (sigqueue_t));
        return (0);
    }
    sigdelq(p, curthread, SIGTRAP);
    sigaddqa(p, curthread, sqp);
    schedctl_finish_sigblock(curthread);
    smask = curthread->t_hold;
    sigfillset(&curthread->t_hold);
    sigdiffset(&curthread->t_hold, &cantmask);
    sigdelset(&curthread->t_hold, SIGTRAP);
    mutex_exit(&p->p_lock);

    rval = ((ISSIG_FAST(curthread, lwp, p, FORREAL))? 0 : 1);

    /* restore the original signal mask */
    mutex_enter(&p->p_lock);
    curthread->t_hold = smask;
    mutex_exit(&p->p_lock);

    return (rval);
}

/*
 * Wrappers for the copyin()/copyout() functions to deal
 * with watchpoints that fire while in system calls.
 */

static int
watch_xcopyin(const void *uaddr, void *kaddr, size_t count)
{
    klwp_t *lwp = ttolwp(curthread);
    caddr_t watch_uaddr = (caddr_t)uaddr;
    caddr_t watch_kaddr = (caddr_t)kaddr;
    int error = 0;
    label_t ljb;
    size_t part;
    int mapped;

    while (count && error == 0) {
        int watchcode;
        caddr_t vaddr;
        size_t len;
        int ta;

        if ((part = PAGESIZE -
            (((uintptr_t)uaddr) & PAGEOFFSET)) > count)
            part = count;

        if (!pr_is_watchpage(watch_uaddr, S_READ))
            watchcode = 0;
        else {
            vaddr = watch_uaddr;
            watchcode = pr_is_watchpoint(&vaddr, &ta,
                part, &len, S_READ);
            if (watchcode && ta == 0)
                part = vaddr - watch_uaddr;
        }

        /*
         * Copy the initial part, up to a watched address, if any.
         */
        if (part != 0) {
            mapped = pr_mappage(watch_uaddr, part, S_READ, 1);
            if (on_fault(&ljb))
                error = EFAULT;
            else
                copyin_noerr(watch_uaddr, watch_kaddr, part);
            no_fault();
            if (mapped)
                pr_unmappage(watch_uaddr, part, S_READ, 1);
            watch_uaddr += part;
            watch_kaddr += part;
            count -= part;
        }
        /*
         * If trapafter was specified, then copy through the
         * watched area before taking the watchpoint trap.
         */
        while (count && watchcode && ta && len > part && error == 0) {
            len -= part;
            if ((part = PAGESIZE) > count)
                part = count;
            if (part > len)
                part = len;
            mapped = pr_mappage(watch_uaddr, part, S_READ, 1);
            if (on_fault(&ljb))
                error = EFAULT;
            else
                copyin_noerr(watch_uaddr, watch_kaddr, part);
            no_fault();
            if (mapped)
                pr_unmappage(watch_uaddr, part, S_READ, 1);
            watch_uaddr += part;
            watch_kaddr += part;
            count -= part;
        }

error:
        /* if we hit a watched address, do the watchpoint logic */
        if (watchcode &&
            (!sys_watchpoint(vaddr, watchcode, ta) ||
            lwp->lwp_sysabort)) {
            lwp->lwp_sysabort = 0;
            error = EFAULT;
            break;
        }
    }

    return (error);
}

static int
watch_copyin(const void *kaddr, void *uaddr, size_t count)
{
    return (watch_xcopyin(kaddr, uaddr, count) ? -1 : 0);
}


static int
watch_xcopyout(const void *kaddr, void *uaddr, size_t count)
{
    klwp_t *lwp = ttolwp(curthread);
    caddr_t watch_uaddr = (caddr_t)uaddr;
    caddr_t watch_kaddr = (caddr_t)kaddr;
    int error = 0;
    label_t ljb;

    while (count && error == 0) {
        int watchcode;
        caddr_t vaddr;
        size_t part;
        size_t len;
        int ta;
        int mapped;

        if ((part = PAGESIZE -
            (((uintptr_t)uaddr) & PAGEOFFSET)) > count)
            part = count;

        if (!pr_is_watchpage(watch_uaddr, S_WRITE))
            watchcode = 0;
        else {
            vaddr = watch_uaddr;
            watchcode = pr_is_watchpoint(&vaddr, &ta,
                part, &len, S_WRITE);
            if (watchcode) {
                if (ta == 0)
                    part = vaddr - watch_uaddr;
                else {
                    len += vaddr - watch_uaddr;
                    if (part > len)
                        part = len;
                }
            }
        }

        /*
         * Copy the initial part, up to a watched address, if any.
         */
        if (part != 0) {
            mapped = pr_mappage(watch_uaddr, part, S_WRITE, 1);
            if (on_fault(&ljb))
                error = EFAULT;
            else
                copyout_noerr(watch_kaddr, watch_uaddr, part);
            no_fault();
            if (mapped)
                pr_unmappage(watch_uaddr, part, S_WRITE, 1);
            watch_uaddr += part;
            watch_kaddr += part;
            count -= part;
        }

        /*
         * If trapafter was specified, then copy through the
         * watched area before taking the watchpoint trap.
         */
        while (count && watchcode && ta && len > part && error == 0) {
            len -= part;
            if ((part = PAGESIZE) > count)
                part = count;
            if (part > len)
                part = len;
            mapped = pr_mappage(watch_uaddr, part, S_WRITE, 1);
            if (on_fault(&ljb))
                error = EFAULT;
            else
                copyout_noerr(watch_kaddr, watch_uaddr, part);
            no_fault();
            if (mapped)
                pr_unmappage(watch_uaddr, part, S_WRITE, 1);
            watch_uaddr += part;
            watch_kaddr += part;
            count -= part;
        }

        /* if we hit a watched address, do the watchpoint logic */
        if (watchcode &&
            (!sys_watchpoint(vaddr, watchcode, ta) ||
            lwp->lwp_sysabort)) {
            lwp->lwp_sysabort = 0;
            error = EFAULT;
            break;
        }
    }

    return (error);
}

static int
watch_copyout(const void *kaddr, void *uaddr, size_t count)
{
    return (watch_xcopyout(kaddr, uaddr, count) ? -1 : 0);
}

static int
watch_copyinstr(
    const char *uaddr,
    char *kaddr,
    size_t maxlength,
    size_t *lencopied)
{
    klwp_t *lwp = ttolwp(curthread);
    size_t resid;
    int error = 0;
    label_t ljb;

    if ((resid = maxlength) == 0)
        return (ENAMETOOLONG);

    while (resid && error == 0) {
        int watchcode;
        caddr_t vaddr;
        size_t part;
        size_t len;
        size_t size;
        int ta;
        int mapped;

        if ((part = PAGESIZE -
            (((uintptr_t)uaddr) & PAGEOFFSET)) > resid)
            part = resid;

        if (!pr_is_watchpage((caddr_t)uaddr, S_READ))
            watchcode = 0;
        else {
            vaddr = (caddr_t)uaddr;
            watchcode = pr_is_watchpoint(&vaddr, &ta,
                part, &len, S_READ);
            if (watchcode) {
                if (ta == 0)
                    part = vaddr - uaddr;
                else {
                    len += vaddr - uaddr;
                    if (part > len)
                        part = len;
                }
            }
        }

        /*
         * Copy the initial part, up to a watched address, if any.
         */
        if (part != 0) {
            mapped = pr_mappage((caddr_t)uaddr, part, S_READ, 1);
            if (on_fault(&ljb))
                error = EFAULT;
            else
                error = copyinstr_noerr(uaddr, kaddr, part,
                    &size);
            no_fault();
            if (mapped)
                pr_unmappage((caddr_t)uaddr, part, S_READ, 1);
            uaddr += size;
            kaddr += size;
            resid -= size;
            if (error == ENAMETOOLONG && resid > 0)
                error = 0;
            if (error != 0 || (watchcode &&
                (uaddr < vaddr || kaddr[-1] == '\0')))
                break;  /* didn't reach the watched area */
        }

        /*
         * If trapafter was specified, then copy through the
         * watched area before taking the watchpoint trap.
         */
        while (resid && watchcode && ta && len > part && error == 0 &&
            size == part && kaddr[-1] != '\0') {
            len -= part;
            if ((part = PAGESIZE) > resid)
                part = resid;
            if (part > len)
                part = len;
            mapped = pr_mappage((caddr_t)uaddr, part, S_READ, 1);
            if (on_fault(&ljb))
                error = EFAULT;
            else
                error = copyinstr_noerr(uaddr, kaddr, part,
                    &size);
            no_fault();
            if (mapped)
                pr_unmappage((caddr_t)uaddr, part, S_READ, 1);
            uaddr += size;
            kaddr += size;
            resid -= size;
            if (error == ENAMETOOLONG && resid > 0)
                error = 0;
        }

        /* if we hit a watched address, do the watchpoint logic */
        if (watchcode &&
            (!sys_watchpoint(vaddr, watchcode, ta) ||
            lwp->lwp_sysabort)) {
            lwp->lwp_sysabort = 0;
            error = EFAULT;
            break;
        }

        if (error == 0 && part != 0 &&
            (size < part || kaddr[-1] == '\0'))
            break;
    }

    if (error != EFAULT && lencopied)
        *lencopied = maxlength - resid;
    return (error);
}

static int
watch_copyoutstr(
    const char *kaddr,
    char *uaddr,
    size_t maxlength,
    size_t *lencopied)
{
    klwp_t *lwp = ttolwp(curthread);
    size_t resid;
    int error = 0;
    label_t ljb;

    if ((resid = maxlength) == 0)
        return (ENAMETOOLONG);

    while (resid && error == 0) {
        int watchcode;
        caddr_t vaddr;
        size_t part;
        size_t len;
        size_t size;
        int ta;
        int mapped;

        if ((part = PAGESIZE -
            (((uintptr_t)uaddr) & PAGEOFFSET)) > resid)
            part = resid;

        if (!pr_is_watchpage(uaddr, S_WRITE)) {
            watchcode = 0;
        } else {
            vaddr = uaddr;
            watchcode = pr_is_watchpoint(&vaddr, &ta,
                part, &len, S_WRITE);
            if (watchcode && ta == 0)
                part = vaddr - uaddr;
        }

        /*
         * Copy the initial part, up to a watched address, if any.
         */
        if (part != 0) {
            mapped = pr_mappage(uaddr, part, S_WRITE, 1);
            if (on_fault(&ljb))
                error = EFAULT;
            else
                error = copyoutstr_noerr(kaddr, uaddr, part,
                    &size);
            no_fault();
            if (mapped)
                pr_unmappage(uaddr, part, S_WRITE, 1);
            uaddr += size;
            kaddr += size;
            resid -= size;
            if (error == ENAMETOOLONG && resid > 0)
                error = 0;
            if (error != 0 || (watchcode &&
                (uaddr < vaddr || kaddr[-1] == '\0')))
                break;  /* didn't reach the watched area */
        }

        /*
         * If trapafter was specified, then copy through the
         * watched area before taking the watchpoint trap.
         */
        while (resid && watchcode && ta && len > part && error == 0 &&
            size == part && kaddr[-1] != '\0') {
            len -= part;
            if ((part = PAGESIZE) > resid)
                part = resid;
            if (part > len)
                part = len;
            mapped = pr_mappage(uaddr, part, S_WRITE, 1);
            if (on_fault(&ljb))
                error = EFAULT;
            else
                error = copyoutstr_noerr(kaddr, uaddr, part,
                    &size);
            no_fault();
            if (mapped)
                pr_unmappage(uaddr, part, S_WRITE, 1);
            uaddr += size;
            kaddr += size;
            resid -= size;
            if (error == ENAMETOOLONG && resid > 0)
                error = 0;
        }

        /* if we hit a watched address, do the watchpoint logic */
        if (watchcode &&
            (!sys_watchpoint(vaddr, watchcode, ta) ||
            lwp->lwp_sysabort)) {
            lwp->lwp_sysabort = 0;
            error = EFAULT;
            break;
        }

        if (error == 0 && part != 0 &&
            (size < part || kaddr[-1] == '\0'))
            break;
    }

    if (error != EFAULT && lencopied)
        *lencopied = maxlength - resid;
    return (error);
}

typedef int (*fuword_func)(const void *, void *);

/*
 * Generic form of watch_fuword8(), watch_fuword16(), etc.
 */
static int
watch_fuword(const void *addr, void *dst, fuword_func func, size_t size)
{
    klwp_t *lwp = ttolwp(curthread);
    int watchcode;
    caddr_t vaddr;
    int mapped;
    int rv = 0;
    int ta;
    label_t ljb;

    for (;;) {

        vaddr = (caddr_t)addr;
        watchcode = pr_is_watchpoint(&vaddr, &ta, size, NULL, S_READ);
        if (watchcode == 0 || ta != 0) {
            mapped = pr_mappage((caddr_t)addr, size, S_READ, 1);
            if (on_fault(&ljb))
                rv = -1;
            else
                (*func)(addr, dst);
            no_fault();
            if (mapped)
                pr_unmappage((caddr_t)addr, size, S_READ, 1);
        }
        if (watchcode &&
            (!sys_watchpoint(vaddr, watchcode, ta) ||
            lwp->lwp_sysabort)) {
            lwp->lwp_sysabort = 0;
            rv = -1;
            break;
        }
        if (watchcode == 0 || ta != 0)
            break;
    }

    return (rv);
}

static int
watch_fuword8(const void *addr, uint8_t *dst)
{
    return (watch_fuword(addr, dst, (fuword_func)fuword8_noerr,
        sizeof (*dst)));
}

static int
watch_fuword16(const void *addr, uint16_t *dst)
{
    return (watch_fuword(addr, dst, (fuword_func)fuword16_noerr,
        sizeof (*dst)));
}

static int
watch_fuword32(const void *addr, uint32_t *dst)
{
    return (watch_fuword(addr, dst, (fuword_func)fuword32_noerr,
        sizeof (*dst)));
}

#ifdef _LP64
static int
watch_fuword64(const void *addr, uint64_t *dst)
{
    return (watch_fuword(addr, dst, (fuword_func)fuword64_noerr,
        sizeof (*dst)));
}
#endif


static int
watch_suword8(void *addr, uint8_t value)
{
    klwp_t *lwp = ttolwp(curthread);
    int watchcode;
    caddr_t vaddr;
    int mapped;
    int rv = 0;
    int ta;
    label_t ljb;

    for (;;) {

        vaddr = (caddr_t)addr;
        watchcode = pr_is_watchpoint(&vaddr, &ta, sizeof (value), NULL,
            S_WRITE);
        if (watchcode == 0 || ta != 0) {
            mapped = pr_mappage((caddr_t)addr, sizeof (value),
                S_WRITE, 1);
            if (on_fault(&ljb))
                rv = -1;
            else
                suword8_noerr(addr, value);
            no_fault();
            if (mapped)
                pr_unmappage((caddr_t)addr, sizeof (value),
                    S_WRITE, 1);
        }
        if (watchcode &&
            (!sys_watchpoint(vaddr, watchcode, ta) ||
            lwp->lwp_sysabort)) {
            lwp->lwp_sysabort = 0;
            rv = -1;
            break;
        }
        if (watchcode == 0 || ta != 0)
            break;
    }

    return (rv);
}

static int
watch_suword16(void *addr, uint16_t value)
{
    klwp_t *lwp = ttolwp(curthread);
    int watchcode;
    caddr_t vaddr;
    int mapped;
    int rv = 0;
    int ta;
    label_t ljb;

    for (;;) {

        vaddr = (caddr_t)addr;
        watchcode = pr_is_watchpoint(&vaddr, &ta, sizeof (value), NULL,
            S_WRITE);
        if (watchcode == 0 || ta != 0) {
            mapped = pr_mappage((caddr_t)addr, sizeof (value),
                S_WRITE, 1);
            if (on_fault(&ljb))
                rv = -1;
            else
                suword16_noerr(addr, value);
            no_fault();
            if (mapped)
                pr_unmappage((caddr_t)addr, sizeof (value),
                    S_WRITE, 1);
        }
        if (watchcode &&
            (!sys_watchpoint(vaddr, watchcode, ta) ||
            lwp->lwp_sysabort)) {
            lwp->lwp_sysabort = 0;
            rv = -1;
            break;
        }
        if (watchcode == 0 || ta != 0)
            break;
    }

    return (rv);
}

static int
watch_suword32(void *addr, uint32_t value)
{
    klwp_t *lwp = ttolwp(curthread);
    int watchcode;
    caddr_t vaddr;
    int mapped;
    int rv = 0;
    int ta;
    label_t ljb;

    for (;;) {

        vaddr = (caddr_t)addr;
        watchcode = pr_is_watchpoint(&vaddr, &ta, sizeof (value), NULL,
            S_WRITE);
        if (watchcode == 0 || ta != 0) {
            mapped = pr_mappage((caddr_t)addr, sizeof (value),
                S_WRITE, 1);
            if (on_fault(&ljb))
                rv = -1;
            else
                suword32_noerr(addr, value);
            no_fault();
            if (mapped)
                pr_unmappage((caddr_t)addr, sizeof (value),
                    S_WRITE, 1);
        }
        if (watchcode &&
            (!sys_watchpoint(vaddr, watchcode, ta) ||
            lwp->lwp_sysabort)) {
            lwp->lwp_sysabort = 0;
            rv = -1;
            break;
        }
        if (watchcode == 0 || ta != 0)
            break;
    }

    return (rv);
}

#ifdef _LP64
static int
watch_suword64(void *addr, uint64_t value)
{
    klwp_t *lwp = ttolwp(curthread);
    int watchcode;
    caddr_t vaddr;
    int mapped;
    int rv = 0;
    int ta;
    label_t ljb;

    for (;;) {

        vaddr = (caddr_t)addr;
        watchcode = pr_is_watchpoint(&vaddr, &ta, sizeof (value), NULL,
            S_WRITE);
        if (watchcode == 0 || ta != 0) {
            mapped = pr_mappage((caddr_t)addr, sizeof (value),
                S_WRITE, 1);
            if (on_fault(&ljb))
                rv = -1;
            else
                suword64_noerr(addr, value);
            no_fault();
            if (mapped)
                pr_unmappage((caddr_t)addr, sizeof (value),
                    S_WRITE, 1);
        }
        if (watchcode &&
            (!sys_watchpoint(vaddr, watchcode, ta) ||
            lwp->lwp_sysabort)) {
            lwp->lwp_sysabort = 0;
            rv = -1;
            break;
        }
        if (watchcode == 0 || ta != 0)
            break;
    }

    return (rv);
}
#endif /* _LP64 */

/*
 * Check for watched addresses in the given address space.
 * Return 1 if this is true, otherwise 0.
 */
static int
pr_is_watched(caddr_t base, size_t len, int rw)
{
    caddr_t saddr = (caddr_t)((uintptr_t)base & (uintptr_t)PAGEMASK);
    caddr_t eaddr = base + len;
    caddr_t paddr;

    for (paddr = saddr; paddr < eaddr; paddr += PAGESIZE) {
        if (pr_is_watchpage(paddr, rw))
            return (1);
    }

    return (0);
}

/*
 * Wrapper for the physio() function.
 * Splits one uio operation with multiple iovecs into uio operations with
 * only one iovecs to do the watchpoint handling separately for each iovecs.
 */
static int
watch_physio(int (*strat)(struct buf *), struct buf *bp, dev_t dev,
    int rw, void (*mincnt)(struct buf *), struct uio *uio)
{
    struct uio auio;
    struct iovec *iov;
    caddr_t  base;
    size_t len;
    int seg_rw;
    int error = 0;

    if (uio->uio_segflg == UIO_SYSSPACE)
        return (default_physio(strat, bp, dev, rw, mincnt, uio));

    seg_rw = (rw == B_READ) ? S_WRITE : S_READ;

    while (uio->uio_iovcnt > 0) {
        if (uio->uio_resid == 0) {
            /*
             * Make sure to return the uio structure with the
             * same values as default_physio() does.
             */
            uio->uio_iov++;
            uio->uio_iovcnt--;
            continue;
        }

        iov = uio->uio_iov;
        len = MIN(iov->iov_len, uio->uio_resid);

        auio.uio_iovcnt = 1;
        auio.uio_iov = iov;
        auio.uio_resid = len;
        auio.uio_loffset = uio->uio_loffset;
        auio.uio_llimit = uio->uio_llimit;
        auio.uio_fmode = uio->uio_fmode;
        auio.uio_extflg = uio->uio_extflg;
        auio.uio_segflg = uio->uio_segflg;

        base = iov->iov_base;

        if (!pr_is_watched(base, len, seg_rw)) {
            /*
             * The given memory references don't cover a
             * watched page.
             */
            error = default_physio(strat, bp, dev, rw, mincnt,
                &auio);

            /* Update uio with values from auio. */
            len -= auio.uio_resid;
            uio->uio_resid -= len;
            uio->uio_loffset += len;

            /*
             * Return if an error occurred or not all data
             * was copied.
             */
            if (auio.uio_resid || error)
                break;
            uio->uio_iov++;
            uio->uio_iovcnt--;
        } else {
            int mapped, watchcode, ta;
            caddr_t vaddr = base;
            klwp_t *lwp = ttolwp(curthread);

            watchcode = pr_is_watchpoint(&vaddr, &ta, len,
                NULL, seg_rw);

            if (watchcode == 0 || ta != 0) {
                /*
                 * Do the io if the given memory references
                 * don't cover a watched area (watchcode=0)
                 * or if WA_TRAPAFTER was specified.
                 */
                mapped = pr_mappage(base, len, seg_rw, 1);
                error = default_physio(strat, bp, dev, rw,
                    mincnt, &auio);
                if (mapped)
                    pr_unmappage(base, len, seg_rw, 1);

                len -= auio.uio_resid;
                uio->uio_resid -= len;
                uio->uio_loffset += len;
            }

            /*
             * If we hit a watched address, do the watchpoint logic.
             */
            if (watchcode &&
                (!sys_watchpoint(vaddr, watchcode, ta) ||
                lwp->lwp_sysabort)) {
                lwp->lwp_sysabort = 0;
                return (EFAULT);
            }

            /*
             * Check for errors from default_physio().
             */
            if (watchcode == 0 || ta != 0) {
                if (auio.uio_resid || error)
                    break;
                uio->uio_iov++;
                uio->uio_iovcnt--;
            }
        }
    }

    return (error);
}

int
wa_compare(const void *a, const void *b)
{
    const watched_area_t *pa = a;
    const watched_area_t *pb = b;

    if (pa->wa_vaddr < pb->wa_vaddr)
        return (-1);
    else if (pa->wa_vaddr > pb->wa_vaddr)
        return (1);
    else
        return (0);
}

int
wp_compare(const void *a, const void *b)
{
    const watched_page_t *pa = a;
    const watched_page_t *pb = b;

    if (pa->wp_vaddr < pb->wp_vaddr)
        return (-1);
    else if (pa->wp_vaddr > pb->wp_vaddr)
        return (1);
    else
        return (0);
}

/*
 * Given an address range, finds the first watched area which overlaps some or
 * all of the range.
 */
watched_area_t *
pr_find_watched_area(proc_t *p, watched_area_t *pwa, avl_index_t *where)
{
    caddr_t vaddr = pwa->wa_vaddr;
    caddr_t eaddr = pwa->wa_eaddr;
    watched_area_t *wap;
    avl_index_t real_where;

    /* First, check if there is an exact match.  */
    wap = avl_find(&p->p_warea, pwa, &real_where);


    /* Check to see if we overlap with the previous area.  */
    if (wap == NULL) {
        wap = avl_nearest(&p->p_warea, real_where, AVL_BEFORE);
        if (wap != NULL &&
            (vaddr >= wap->wa_eaddr || eaddr <= wap->wa_vaddr))
            wap = NULL;
    }

    /* Try the next area.  */
    if (wap == NULL) {
        wap = avl_nearest(&p->p_warea, real_where, AVL_AFTER);
        if (wap != NULL &&
            (vaddr >= wap->wa_eaddr || eaddr <= wap->wa_vaddr))
            wap = NULL;
    }

    if (where)
        *where = real_where;

    return (wap);
}

void
watch_enable(kthread_id_t t)
{
    t->t_proc_flag |= TP_WATCHPT;
    install_copyops(t, &watch_copyops);
}

void
watch_disable(kthread_id_t t)
{
    t->t_proc_flag &= ~TP_WATCHPT;
    remove_copyops(t);
}

int
copyin_nowatch(const void *uaddr, void *kaddr, size_t len)
{
    int watched, ret;

    watched = watch_disable_addr(uaddr, len, S_READ);
    ret = copyin(uaddr, kaddr, len);
    if (watched)
        watch_enable_addr(uaddr, len, S_READ);

    return (ret);
}

int
copyout_nowatch(const void *kaddr, void *uaddr, size_t len)
{
    int watched, ret;

    watched = watch_disable_addr(uaddr, len, S_WRITE);
    ret = copyout(kaddr, uaddr, len);
    if (watched)
        watch_enable_addr(uaddr, len, S_WRITE);

    return (ret);
}

#ifdef _LP64
int
fuword64_nowatch(const void *addr, uint64_t *value)
{
    int watched, ret;

    watched = watch_disable_addr(addr, sizeof (*value), S_READ);
    ret = fuword64(addr, value);
    if (watched)
        watch_enable_addr(addr, sizeof (*value), S_READ);

    return (ret);
}
#endif

int
fuword32_nowatch(const void *addr, uint32_t *value)
{
    int watched, ret;

    watched = watch_disable_addr(addr, sizeof (*value), S_READ);
    ret = fuword32(addr, value);
    if (watched)
        watch_enable_addr(addr, sizeof (*value), S_READ);

    return (ret);
}

#ifdef _LP64
int
suword64_nowatch(void *addr, uint64_t value)
{
    int watched, ret;

    watched = watch_disable_addr(addr, sizeof (value), S_WRITE);
    ret = suword64(addr, value);
    if (watched)
        watch_enable_addr(addr, sizeof (value), S_WRITE);

    return (ret);
}
#endif

int
suword32_nowatch(void *addr, uint32_t value)
{
    int watched, ret;

    watched = watch_disable_addr(addr, sizeof (value), S_WRITE);
    ret = suword32(addr, value);
    if (watched)
        watch_enable_addr(addr, sizeof (value), S_WRITE);

    return (ret);
}

int
watch_disable_addr(const void *addr, size_t len, enum seg_rw rw)
{
    if (pr_watch_active(curproc))
        return (pr_mappage((caddr_t)addr, len, rw, 1));
    return (0);
}

void
watch_enable_addr(const void *addr, size_t len, enum seg_rw rw)
{
    if (pr_watch_active(curproc))
        pr_unmappage((caddr_t)addr, len, rw, 1);
}