common/os/vm_subr.c

	vm_subr.c revision 7c478bd95313f5f23a4c958a745db2134aa03244
/*
 * 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 2004 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

/*  Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
/*    All Rights Reserved   */

/*
 * University Copyright- Copyright (c) 1982, 1986, 1988
 * The Regents of the University of California
 * All Rights Reserved
 *
 * University Acknowledgment- Portions of this document are derived from
 * software developed by the University of California, Berkeley, and its
 * contributors.
 */

#pragma ident   "%Z%%M% %I% %E% SMI"

#include <sys/types.h>
#include <sys/t_lock.h>
#include <sys/param.h>
#include <sys/errno.h>
#include <sys/debug.h>
#include <sys/cmn_err.h>
#include <sys/kmem.h>
#include <sys/sysmacros.h>
#include <sys/inline.h>
#include <sys/buf.h>
#include <sys/uio.h>
#include <sys/user.h>
#include <sys/proc.h>
#include <sys/systm.h>
#include <sys/vmsystm.h>
#include <sys/cpuvar.h>
#include <sys/mman.h>
#include <sys/cred.h>
#include <sys/vnode.h>
#include <sys/file.h>
#include <sys/vm.h>

#include <sys/swap.h>
#include <sys/vtrace.h>
#include <sys/tnf_probe.h>
#include <sys/fs/snode.h>
#include <sys/copyops.h>
#include <sys/conf.h>
#include <sys/sdt.h>

#include <vm/anon.h>
#include <vm/hat.h>
#include <vm/as.h>
#include <vm/seg.h>
#include <vm/page.h>
#include <vm/seg_vn.h>
#include <vm/seg_kmem.h>

extern int maxphys;

void
minphys(struct buf *bp)
{
    if (bp->b_bcount > maxphys)
        bp->b_bcount = maxphys;
}

/*
 * use kmem_cache_create for physio buffers. This has shown
 * a better cache distribution compared to buffers on the
 * stack. It also avoids semaphore construction/deconstruction
 * per request
 */

static struct kmem_cache *physio_buf_cache;

/* ARGSUSED */
static int
physio_buf_constructor(void *buf, void *cdrarg, int kmflags)
{
    bioinit((struct buf *)buf);
    return (0);
}

/* ARGSUSED */
static void
physio_buf_destructor(void *buf, void *cdrarg)
{
    biofini((struct buf *)buf);
}

void
physio_bufs_init(void)
{
    physio_buf_cache = kmem_cache_create("physio_buf_cache",
        sizeof (struct buf), 0,
        physio_buf_constructor, physio_buf_destructor,
        NULL, NULL, NULL, 0);
}


/*
 * initiate raw I/O request
 *
 * allocate buf header if necessary
 * adjust max size of each I/O request
 * lock down user pages and verify access protections
 * call driver's strategy routine to submit request
 * wait for I/O completion
 * unlock user pages and free allocated buf header
 */

int
default_physio(int (*strat)(struct buf *), struct buf *bp, dev_t dev,
    int rw, void (*mincnt)(struct buf *), struct uio *uio)
{
    struct iovec *iov;
    struct proc *procp;
    struct as *asp;
    ssize_t c;
    char *a;
    int error = 0;
    page_t **pplist;
    int allocbuf = 0;

    TRACE_1(TR_FAC_PHYSIO, TR_PHYSIO_START, "physio_start: bp %p", bp);

    /* Kernel probe */
    TNF_PROBE_4(physio_start, "io rawio", /* CSTYLED */,
        tnf_device, device,     dev,
        tnf_offset, offset,     uio->uio_loffset,
        tnf_size,   size,       uio->uio_resid,
        tnf_bioflags,   rw,     rw);

    if (rw == B_READ) {
        CPU_STATS_ADD_K(sys, phread, 1);
    } else {
        CPU_STATS_ADD_K(sys, phwrite, 1);
    }

    TRACE_1(TR_FAC_PHYSIO, TR_PHYSIO_GETBUF_START,
        "getbuf_start: bp %p", bp);

    if (bp == NULL) {
        bp = kmem_cache_alloc(physio_buf_cache, KM_SLEEP);
        bp->b_iodone = NULL;
        bp->b_resid = 0;
        allocbuf = 1;
    }
    TRACE_1(TR_FAC_PHYSIO, TR_PHYSIO_GETBUF_END, "getbuf_end: bp %p", bp);

    if (uio->uio_segflg == UIO_USERSPACE) {
        procp = ttoproc(curthread);
        asp = procp->p_as;
    } else {
        procp = NULL;
        asp = &kas;
    }
    ASSERT(SEMA_HELD(&bp->b_sem));

    /*
     * We need to prepare this buffer for the io:::start probe, including
     * NULL'ing out the file, clearing the offset, and filling in the
     * b_dip field.
     */
    bp->b_file = NULL;
    bp->b_offset = -1;

    if (dev != NODEV) {
        (void) devopsp[getmajor(dev)]->devo_getinfo(NULL,
            DDI_INFO_DEVT2DEVINFO, (void *)dev, (void **)&bp->b_dip);
    } else {
        bp->b_dip = NULL;
    }

    while (uio->uio_iovcnt > 0) {
        iov = uio->uio_iov;

        bp->b_error = 0;
        bp->b_proc = procp;

        while (iov->iov_len > 0) {
            if (uio->uio_resid == 0)
                break;
            if (uio->uio_loffset < 0) {
                error = EINVAL;
                break;
            }
#ifdef  _ILP32
            /*
             * For 32-bit kernels, check against SPEC_MAXOFFSET_T
             * which represents the maximum size that can be
             * supported by the IO subsystem.
             * XXX this code assumes a D_64BIT driver.
             */
            if (uio->uio_loffset > SPEC_MAXOFFSET_T) {
                error = EINVAL;
                break;
            }
#endif  /* _ILP32 */
            bp->b_flags = B_BUSY | B_PHYS | rw;
            bp->b_edev = dev;
            bp->b_lblkno = btodt(uio->uio_loffset);

            /*
             * Don't count on b_addr remaining untouched by the
             * code below (it may be reset because someone does
             * a bp_mapin on the buffer) -- reset from the iov
             * each time through, updating the iov's base address
             * instead.
             */
            a = bp->b_un.b_addr = iov->iov_base;
            bp->b_bcount = MIN(iov->iov_len, uio->uio_resid);
            (*mincnt)(bp);
            c = bp->b_bcount;

            TRACE_1(TR_FAC_PHYSIO, TR_PHYSIO_LOCK_START,
                "as_pagelock_start: bp %p", bp);

            error = as_pagelock(asp, &pplist, a,
                c, rw == B_READ? S_WRITE : S_READ);

            TRACE_0(TR_FAC_PHYSIO, TR_PHYSIO_LOCK_END,
                "as_pagelock_end:");

            if (error != 0) {
                bp->b_flags |= B_ERROR;
                bp->b_error = error;
                bp->b_flags &=
                    ~(B_BUSY|B_WANTED|B_PHYS);
                break;
            }
            bp->b_shadow = pplist;
            if (pplist != NULL) {
                bp->b_flags |= B_SHADOW;
            }

            DTRACE_IO1(start, struct buf *, bp);
            bp->b_flags |= B_STARTED;

            (void) (*strat)(bp);
            error = biowait(bp);

            /*
             * unlock the pages
             */
            TRACE_1(TR_FAC_PHYSIO, TR_PHYSIO_UNLOCK_START,
                "as_pageunlock_start: bp %p", bp);

            as_pageunlock(asp, pplist, a, c,
                rw == B_READ? S_WRITE : S_READ);

            TRACE_0(TR_FAC_PHYSIO, TR_PHYSIO_UNLOCK_END,
                "as_pageunlock_end:");

            c -= bp->b_resid;
            iov->iov_base += c;
            iov->iov_len -= c;
            uio->uio_resid -= c;
            uio->uio_loffset += c;
            /* bp->b_resid - temp kludge for tape drives */
            if (bp->b_resid || error)
                break;
        }
        bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_SHADOW);
        /* bp->b_resid - temp kludge for tape drives */
        if (bp->b_resid || error)
            break;
        uio->uio_iov++;
        uio->uio_iovcnt--;
    }

    if (allocbuf) {
        kmem_cache_free(physio_buf_cache, bp);
    }

    /* Kernel probe */
    TNF_PROBE_1(physio_end, "io rawio", /* CSTYLED */,
        tnf_device, device,     dev);

    TRACE_1(TR_FAC_PHYSIO, TR_PHYSIO_END, "physio_end: bp %p", bp);

    return (error);
}

/*
 * Returns 0 on success, or an error on failure.
 *
 * This function is no longer a part of the DDI/DKI.
 * However, for compatibility, its interface should not
 * be changed and it should not be removed from the kernel.
 */
int
useracc(void *addr, size_t count, int access)
{
    uint_t prot;

    prot = PROT_USER | ((access == B_READ) ? PROT_READ : PROT_WRITE);
    return (as_checkprot(ttoproc(curthread)->p_as, addr, count, prot));
}

#define MAX_MAPIN_PAGES 8

/*
 * This function temporarily "borrows" user pages for kernel use. If
 * "cow" is on, it also sets up copy-on-write protection (only feasible
 * on MAP_PRIVATE segment) on the user mappings, to protect the borrowed
 * pages from any changes by the user. The caller is responsible for
 * unlocking and tearing down cow settings when it's done with the pages.
 * For an example, see kcfree().
 *
 * Pages behind [uaddr..uaddr+*lenp] under address space "as" are locked
 * (shared), and mapped into kernel address range [kaddr..kaddr+*lenp] if
 * kaddr != -1. On entering this function, cached_ppp contains a list
 * of pages that are mapped into [kaddr..kaddr+*lenp] already (from a
 * previous call). Thus if same pages remain behind [uaddr..uaddr+*lenp],
 * the kernel map won't need to be reloaded again.
 *
 * For cow == 1, if the pages are anonymous pages, it also bumps the anon
 * reference count, and change the user-mapping to read-only. This
 * scheme should work on all types of segment drivers. But to be safe,
 * we check against segvn here.
 *
 * Since this function is used to emulate copyin() semantic, it checks
 * to make sure the user-mappings allow "user-read".
 *
 * On exit "lenp" contains the number of bytes successfully locked and
 * mapped in. For the unsuccessful ones, the caller can fall back to
 * copyin().
 *
 * Error return:
 * ENOTSUP - operation like this is not supported either on this segment
 * type, or on this platform type.
 */
int
cow_mapin(struct as *as, caddr_t uaddr, caddr_t kaddr, struct page **cached_ppp,
    struct anon **app, size_t *lenp, int cow)
{
    struct      hat *hat;
    struct seg  *seg;
    caddr_t     base;
    page_t      *pp, *ppp[MAX_MAPIN_PAGES];
    long        i;
    int     flags;
    size_t      size, total = *lenp;
    char        first = 1;
    faultcode_t res;

    *lenp = 0;
    if (cow) {
        AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
        seg = as_findseg(as, uaddr, 0);
        if ((seg == NULL) || ((base = seg->s_base) > uaddr) ||
            (uaddr + total) > base + seg->s_size) {
            AS_LOCK_EXIT(as, &as->a_lock);
            return (EINVAL);
        }
        /*
         * The COW scheme should work for all segment types.
         * But to be safe, we check against segvn.
         */
        if (seg->s_ops != &segvn_ops) {
            AS_LOCK_EXIT(as, &as->a_lock);
            return (ENOTSUP);
        } else if ((SEGOP_GETTYPE(seg, uaddr) & MAP_PRIVATE) == 0) {
            AS_LOCK_EXIT(as, &as->a_lock);
            return (ENOTSUP);
        }
    }
    hat = as->a_hat;
    size = total;
tryagain:
    /*
     * If (cow), hat_softlock will also change the usr protection to RO.
     * This is the first step toward setting up cow. Before we
     * bump up an_refcnt, we can't allow any cow-fault on this
     * address. Otherwise segvn_fault will change the protection back
     * to RW upon seeing an_refcnt == 1.
     * The solution is to hold the writer lock on "as".
     */
    res = hat_softlock(hat, uaddr, &size, &ppp[0], cow ? HAT_COW : 0);
    size = total - size;
    *lenp += size;
    size = size >> PAGESHIFT;
    i = 0;
    while (i < size) {
        pp = ppp[i];
        if (cow) {
            kmutex_t *ahm;
            /*
             * Another solution is to hold SE_EXCL on pp, and
             * disable PROT_WRITE. This also works for MAP_SHARED
             * segment. The disadvantage is that it locks the
             * page from being used by anybody else.
             */
            ahm = &anonhash_lock[
                AH_LOCK(pp->p_vnode, pp->p_offset)];
            mutex_enter(ahm);
            *app = swap_anon(pp->p_vnode, pp->p_offset);
            /*
             * Since we are holding the as lock, this avoids a
             * potential race with anon_decref. (segvn_unmap and
             * segvn_free needs the as writer lock to do anon_free.)
             */
            if (*app != NULL) {
#if 0
                if ((*app)->an_refcnt == 0)
                /*
                 * Consider the following senario (unlikey
                 * though):
                 * 1. an_refcnt == 2
                 * 2. we solftlock the page.
                 * 3. cow ocurrs on this addr. So a new ap,
                 * page and mapping is established on addr.
                 * 4. an_refcnt drops to 1 (segvn_faultpage
                 * -> anon_decref(oldap))
                 * 5. the last ref to ap also drops (from
                 * another as). It ends up blocked inside
                 * anon_decref trying to get page's excl lock.
                 * 6. Later kcfree unlocks the page, call
                 * anon_decref -> oops, ap is gone already.
                 *
                 * Holding as writer lock solves all problems.
                 */
                    *app = NULL;
                else
#endif
                    (*app)->an_refcnt++;
            }
            mutex_exit(ahm);
        } else {
            *app = NULL;
        }
        if (kaddr != (caddr_t)-1) {
            if (pp != *cached_ppp) {
                if (*cached_ppp == NULL)
                    flags = HAT_LOAD_LOCK | HAT_NOSYNC |
                        HAT_LOAD_NOCONSIST;
                else
                    flags = HAT_LOAD_REMAP |
                        HAT_LOAD_NOCONSIST;
                /*
                 * In order to cache the kernel mapping after
                 * the user page is unlocked, we call
                 * hat_devload instead of hat_memload so
                 * that the kernel mapping we set up here is
                 * "invisible" to the rest of the world. This
                 * is not very pretty. But as long as the
                 * caller bears the responsibility of keeping
                 * cache consistency, we should be ok -
                 * HAT_NOCONSIST will get us a uncached
                 * mapping on VAC. hat_softlock will flush
                 * a VAC_WRITEBACK cache. Therefore the kaddr
                 * doesn't have to be of the same vcolor as
                 * uaddr.
                 * The alternative is - change hat_devload
                 * to get a cached mapping. Allocate a kaddr
                 * with the same vcolor as uaddr. Then
                 * hat_softlock won't need to flush the VAC.
                 */
                hat_devload(kas.a_hat, kaddr, PAGESIZE,
                    page_pptonum(pp), PROT_READ, flags);
                *cached_ppp = pp;
            }
            kaddr += PAGESIZE;
        }
        cached_ppp++;
        app++;
        ++i;
    }
    if (cow) {
        AS_LOCK_EXIT(as, &as->a_lock);
    }
    if (first && res == FC_NOMAP) {
        /*
         * If the address is not mapped yet, we call as_fault to
         * fault the pages in. We could've fallen back to copy and
         * let it fault in the pages. But for a mapped file, we
         * normally reference each page only once. For zero-copy to
         * be of any use, we'd better fall in the page now and try
         * again.
         */
        first = 0;
        size = size << PAGESHIFT;
        uaddr += size;
        total -= size;
        size = total;
        res = as_fault(as->a_hat, as, uaddr, size, F_INVAL, S_READ);
        if (cow)
            AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
        goto tryagain;
    }
    switch (res) {
    case FC_NOSUPPORT:
        return (ENOTSUP);
    case FC_PROT:   /* Pretend we don't know about it. This will be */
            /* caught by the caller when uiomove fails. */
    case FC_NOMAP:
    case FC_OBJERR:
    default:
        return (0);
    }
}