fs/swapfs/swap_vnops.c

	swap_vnops.c revision c6939658adb0a356a77bc28f7df252ceb4a8f6cc
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (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 2006 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

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

#include <sys/types.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/buf.h>
#include <sys/cred.h>
#include <sys/errno.h>
#include <sys/vnode.h>
#include <sys/cmn_err.h>
#include <sys/swap.h>
#include <sys/mman.h>
#include <sys/vmsystm.h>
#include <sys/vtrace.h>
#include <sys/debug.h>
#include <sys/sysmacros.h>
#include <sys/vm.h>

#include <sys/fs/swapnode.h>

#include <vm/seg.h>
#include <vm/page.h>
#include <vm/pvn.h>
#include <fs/fs_subr.h>

#include <vm/seg_kp.h>

/*
 * Define the routines within this file.
 */
static int  swap_getpage(struct vnode *vp, offset_t off, size_t len,
    uint_t *protp, struct page **plarr, size_t plsz,
    struct seg *seg, caddr_t addr, enum seg_rw rw, struct cred *cr);
static int  swap_putpage(struct vnode *vp, offset_t off, size_t len,
    int flags, struct cred *cr);
static void swap_inactive(struct vnode *vp, struct cred *cr);
static void swap_dispose(vnode_t *vp, page_t *pp, int fl, int dn,
    cred_t *cr);

static int  swap_getapage(struct vnode *vp, u_offset_t off, size_t len,
    uint_t *protp, page_t **plarr, size_t plsz,
    struct seg *seg, caddr_t addr, enum seg_rw rw, struct cred *cr);

int swap_getconpage(struct vnode *vp, u_offset_t off, size_t len,
    uint_t *protp, page_t **plarr, size_t plsz, page_t *conpp,
    uint_t *pszc, spgcnt_t *nreloc, struct seg *seg, caddr_t addr,
    enum seg_rw rw, struct cred *cr);

static int  swap_putapage(struct vnode *vp, page_t *pp, u_offset_t *off,
    size_t *lenp, int flags, struct cred *cr);

const fs_operation_def_t swap_vnodeops_template[] = {
    VOPNAME_INACTIVE, (fs_generic_func_p) swap_inactive,
    VOPNAME_GETPAGE, swap_getpage,
    VOPNAME_PUTPAGE, swap_putpage,
    VOPNAME_DISPOSE, (fs_generic_func_p) swap_dispose,
    VOPNAME_SETFL, fs_error,
    VOPNAME_POLL, fs_error,
    VOPNAME_PATHCONF, fs_error,
    VOPNAME_GETSECATTR, fs_error,
    VOPNAME_SHRLOCK, fs_error,
    NULL, NULL
};

vnodeops_t *swap_vnodeops;

/* ARGSUSED */
static void
swap_inactive(
    struct vnode *vp,
    struct cred *cr)
{
    SWAPFS_PRINT(SWAP_VOPS, "swap_inactive: vp %x\n", vp, 0, 0, 0, 0);
}

/*
 * Return all the pages from [off..off+len] in given file
 */
static int
swap_getpage(
    struct vnode *vp,
    offset_t off,
    size_t len,
    uint_t *protp,
    page_t *pl[],
    size_t plsz,
    struct seg *seg,
    caddr_t addr,
    enum seg_rw rw,
    struct cred *cr)
{
    int err;

    SWAPFS_PRINT(SWAP_VOPS, "swap_getpage: vp %p, off %llx, len %lx\n",
        (void *)vp, off, len, 0, 0);

    TRACE_3(TR_FAC_SWAPFS, TR_SWAPFS_GETPAGE,
        "swapfs getpage:vp %p off %llx len %ld",
        (void *)vp, off, len);

    if (len <= PAGESIZE) {
        err = swap_getapage(vp, (u_offset_t)off, len, protp, pl, plsz,
            seg, addr, rw, cr);
    } else {
        err = pvn_getpages(swap_getapage, vp, (u_offset_t)off, len,
            protp, pl, plsz, seg, addr, rw, cr);
    }

    return (err);
}

/*
 * Called from pvn_getpages or swap_getpage to get a particular page.
 */
/*ARGSUSED*/
static int
swap_getapage(
    struct vnode *vp,
    u_offset_t off,
    size_t len,
    uint_t *protp,
    page_t *pl[],
    size_t plsz,
    struct seg *seg,
    caddr_t addr,
    enum seg_rw rw,
    struct cred *cr)
{
    struct page *pp, *rpp;
    int flags;
    int err = 0;
    struct vnode *pvp = NULL;
    u_offset_t poff;
    int flag_noreloc;
    se_t lock;
    extern int kcage_on;
    int upgrade = 0;

    SWAPFS_PRINT(SWAP_VOPS, "swap_getapage: vp %p, off %llx, len %lx\n",
        vp, off, len, 0, 0);

    /*
     * Until there is a call-back mechanism to cause SEGKP
     * pages to be unlocked, make them non-relocatable.
     */
    if (SEG_IS_SEGKP(seg))
        flag_noreloc = PG_NORELOC;
    else
        flag_noreloc = 0;

    if (protp != NULL)
        *protp = PROT_ALL;

    lock = (rw == S_CREATE ? SE_EXCL : SE_SHARED);

again:
    if (pp = page_lookup(vp, off, lock)) {
        /*
         * In very rare instances, a segkp page may have been
         * relocated outside of the kernel by the kernel cage
         * due to the window between page_unlock() and
         * VOP_PUTPAGE() in segkp_unlock().  Due to the
         * rareness of these occurances, the solution is to
         * relocate the page to a P_NORELOC page.
         */
        if (flag_noreloc != 0) {
            if (!PP_ISNORELOC(pp) && kcage_on) {
                if (lock != SE_EXCL) {
                    upgrade = 1;
                    if (!page_tryupgrade(pp)) {
                        page_unlock(pp);
                        lock = SE_EXCL;
                        goto again;
                    }
                }

                if (page_relocate_cage(&pp, &rpp) != 0)
                    panic("swap_getapage: "
                        "page_relocate_cage failed");

                pp = rpp;
            }
        }

        if (pl) {
            if (upgrade)
                page_downgrade(pp);

            pl[0] = pp;
            pl[1] = NULL;
        } else {
            page_unlock(pp);
        }
    } else {
        pp = page_create_va(vp, off, PAGESIZE,
            PG_WAIT | PG_EXCL | flag_noreloc,
            seg, addr);
        /*
         * Someone raced in and created the page after we did the
         * lookup but before we did the create, so go back and
         * try to look it up again.
         */
        if (pp == NULL)
            goto again;
        if (rw != S_CREATE) {
            err = swap_getphysname(vp, off, &pvp, &poff);
            if (pvp) {
                struct anon *ap;
                kmutex_t *ahm;

                flags = (pl == NULL ? B_ASYNC|B_READ : B_READ);
                err = VOP_PAGEIO(pvp, pp, poff,
                    PAGESIZE, flags, cr);

                if (!err) {
                    ahm = &anonhash_lock[AH_LOCK(vp, off)];
                    mutex_enter(ahm);

                    ap = swap_anon(vp, off);
                    if (ap == NULL)
                        panic("swap_getapage: null anon");

                    if (ap->an_pvp == pvp &&
                        ap->an_poff == poff) {
                        swap_phys_free(pvp, poff,
                            PAGESIZE);
                        ap->an_pvp = NULL;
                        ap->an_poff = NULL;
                        hat_setmod(pp);
                    }

                    mutex_exit(ahm);
                }
            } else {
                if (!err)
                    pagezero(pp, 0, PAGESIZE);

                /*
                 * If it's a fault ahead, release page_io_lock
                 * and SE_EXCL we grabbed in page_create_va
                 *
                 * If we are here, we haven't called VOP_PAGEIO
                 * and thus calling pvn_read_done(pp, B_READ)
                 * below may mislead that we tried i/o. Besides,
                 * in case of async, pvn_read_done() should
                 * not be called by *getpage()
                 */
                if (pl == NULL) {
                    /*
                     * swap_getphysname can return error
                     * only when we are getting called from
                     * swapslot_free which passes non-NULL
                     * pl to VOP_GETPAGE.
                     */
                    ASSERT(err == 0);
                    page_io_unlock(pp);
                    page_unlock(pp);
                }
            }
        }

        ASSERT(pp != NULL);

        if (err && pl)
            pvn_read_done(pp, B_ERROR);

        if (!err && pl)
            pvn_plist_init(pp, pl, plsz, off, PAGESIZE, rw);
    }
    TRACE_3(TR_FAC_SWAPFS, TR_SWAPFS_GETAPAGE,
        "swapfs getapage:pp %p vp %p off %llx", pp, vp, off);
    return (err);
}

/*
 * Called from large page anon routines only! This is an ugly hack where
 * the anon layer directly calls into swapfs with a preallocated large page.
 * Another method would have been to change to VOP and add an extra arg for
 * the preallocated large page. This all could be cleaned up later when we
 * solve the anonymous naming problem and no longer need to loop across of
 * the VOP in PAGESIZE increments to fill in or initialize a large page as
 * is done today. I think the latter is better since it avoid a change to
 * the VOP interface that could later be avoided.
 */
int
swap_getconpage(
    struct vnode *vp,
    u_offset_t off,
    size_t len,
    uint_t *protp,
    page_t *pl[],
    size_t plsz,
    page_t  *conpp,
    uint_t  *pszc,
    spgcnt_t *nreloc,
    struct seg *seg,
    caddr_t addr,
    enum seg_rw rw,
    struct cred *cr)
{
    struct page *pp;
    int         err = 0;
    struct vnode    *pvp = NULL;
    u_offset_t  poff;

    ASSERT(len == PAGESIZE);
    ASSERT(pl != NULL);
    ASSERT(plsz == PAGESIZE);
    ASSERT(protp == NULL);
    ASSERT(nreloc != NULL);
    ASSERT(!SEG_IS_SEGKP(seg)); /* XXX for now not supported */
    SWAPFS_PRINT(SWAP_VOPS, "swap_getconpage: vp %p, off %llx, len %lx\n",
        vp, off, len, 0, 0);

    /*
     * If we are not using a preallocated page then we know one already
     * exists. So just let the old code handle it.
     */
    if (conpp == NULL) {
        err = swap_getapage(vp, (u_offset_t)off, len, protp, pl, plsz,
            seg, addr, rw, cr);
        return (err);
    }
    ASSERT(conpp->p_szc != 0);
    ASSERT(PAGE_EXCL(conpp));


    ASSERT(conpp->p_next == conpp);
    ASSERT(conpp->p_prev == conpp);
    ASSERT(!PP_ISAGED(conpp));
    ASSERT(!PP_ISFREE(conpp));

    *nreloc = 0;
    pp = page_lookup_create(vp, off, SE_SHARED, conpp, nreloc, 0);

    /*
     * If existing page is found we may need to relocate.
     */
    if (pp != conpp) {
        ASSERT(rw != S_CREATE);
        ASSERT(pszc != NULL);
        ASSERT(PAGE_SHARED(pp));
        if (pp->p_szc < conpp->p_szc) {
            *pszc = pp->p_szc;
            page_unlock(pp);
            err = -1;
        } else if (pp->p_szc > conpp->p_szc &&
            seg->s_szc > conpp->p_szc) {
            *pszc = MIN(pp->p_szc, seg->s_szc);
            page_unlock(pp);
            err = -2;
        } else {
            pl[0] = pp;
            pl[1] = NULL;
            if (page_pptonum(pp) &
                (page_get_pagecnt(conpp->p_szc) - 1))
                cmn_err(CE_PANIC, "swap_getconpage: no root");
        }
        return (err);
    }

    ASSERT(PAGE_EXCL(pp));

    if (*nreloc != 0) {
        ASSERT(rw != S_CREATE);
        pl[0] = pp;
        pl[1] = NULL;
        return (0);
    }

    *nreloc = 1;

    /*
     * If necessary do the page io.
     */
    if (rw != S_CREATE) {
        /*
         * Since we are only called now on behalf of an
         * address space operation it's impossible for
         * us to fail unlike swap_getapge() which
         * also gets called from swapslot_free().
         */
        if (swap_getphysname(vp, off, &pvp, &poff)) {
            cmn_err(CE_PANIC,
                "swap_getconpage: swap_getphysname failed!");
        }

        if (pvp) {
            err = VOP_PAGEIO(pvp, pp, poff, PAGESIZE, B_READ, cr);
        } else {
            pagezero(pp, 0, PAGESIZE);
        }
    }

    /*
     * Normally we would let pvn_read_done() destroy
     * the page on IO error. But since this is a preallocated
     * page we'll let the anon layer handle it.
     */
    page_io_unlock(pp);
    if (err != 0)
        page_hashout(pp, NULL);
    ASSERT(pp->p_next == pp);
    ASSERT(pp->p_prev == pp);

    TRACE_3(TR_FAC_SWAPFS, TR_SWAPFS_GETAPAGE,
        "swapfs getconpage:pp %p vp %p off %llx", pp, vp, off);

    pl[0] = pp;
    pl[1] = NULL;
    return (err);
}

/* Async putpage klustering stuff */
int sw_pending_size;
extern int klustsize;
extern struct async_reqs *sw_getreq();
extern void sw_putreq(struct async_reqs *);
extern void sw_putbackreq(struct async_reqs *);
extern struct async_reqs *sw_getfree();
extern void sw_putfree(struct async_reqs *);

static size_t swap_putpagecnt, swap_pagespushed;
static size_t swap_otherfail, swap_otherpages;
static size_t swap_klustfail, swap_klustpages;
static size_t swap_getiofail, swap_getiopages;

/*
 * Flags are composed of {B_INVAL, B_DIRTY B_FREE, B_DONTNEED}.
 * If len == 0, do from off to EOF.
 */
static int swap_nopage = 0; /* Don't do swap_putpage's if set */

/* ARGSUSED */
static int
swap_putpage(
    struct vnode *vp,
    offset_t off,
    size_t len,
    int flags,
    struct cred *cr)
{
    page_t *pp;
    u_offset_t io_off;
    size_t io_len = 0;
    int err = 0;
    struct async_reqs *arg;

    if (swap_nopage)
        return (0);

    ASSERT(vp->v_count != 0);

    SWAPFS_PRINT(SWAP_VOPS,
        "swap_putpage: vp %p, off %llx len %lx, flags %x\n",
        (void *)vp, off, len, flags, 0);
    TRACE_3(TR_FAC_SWAPFS, TR_SWAPFS_PUTPAGE,
        "swapfs putpage:vp %p off %llx len %ld", (void *)vp, off, len);

    if (vp->v_flag & VNOMAP)
        return (ENOSYS);

    if (!vn_has_cached_data(vp))
        return (0);

    if (len == 0) {
        if (curproc == proc_pageout)
            cmn_err(CE_PANIC, "swapfs: pageout can't block");

        /* Search the entire vp list for pages >= off. */
        err = pvn_vplist_dirty(vp, (u_offset_t)off, swap_putapage,
            flags, cr);
    } else {
        u_offset_t eoff;

        /*
         * Loop over all offsets in the range [off...off + len]
         * looking for pages to deal with.
         */
        eoff = off + len;
        for (io_off = (u_offset_t)off; io_off < eoff;
            io_off += io_len) {
            /*
             * If we run out of the async req slot, put the page
             * now instead of queuing.
             */
            if (flags == (B_ASYNC | B_FREE) &&
                sw_pending_size < klustsize &&
                (arg = sw_getfree())) {
                /*
                 * If we are clustering, we should allow
                 * pageout to feed us more pages because # of
                 * pushes is limited by # of I/Os, and one
                 * cluster is considered to be one I/O.
                 */
                if (pushes)
                    pushes--;

                arg->a_vp = vp;
                arg->a_off = io_off;
                arg->a_len = PAGESIZE;
                arg->a_flags = B_ASYNC | B_FREE;
                arg->a_cred = kcred;
                sw_putreq(arg);
                io_len = PAGESIZE;
                continue;
            }
            /*
             * If we are not invalidating pages, use the
             * routine page_lookup_nowait() to prevent
             * reclaiming them from the free list.
             */
            if ((flags & B_INVAL) ||
                (flags & (B_ASYNC | B_FREE)) == B_FREE)
                pp = page_lookup(vp, io_off, SE_EXCL);
            else
                pp = page_lookup_nowait(vp, io_off,
                    (flags & B_FREE) ? SE_EXCL : SE_SHARED);

            if (pp == NULL || pvn_getdirty(pp, flags) == 0)
                io_len = PAGESIZE;
            else {
                err = swap_putapage(vp, pp, &io_off, &io_len,
                    flags, cr);
                if (err != 0)
                    break;
            }
        }
    }
    /* If invalidating, verify all pages on vnode list are gone. */
    if (err == 0 && off == 0 && len == 0 &&
        (flags & B_INVAL) && vn_has_cached_data(vp)) {
        cmn_err(CE_WARN,
            "swap_putpage: B_INVAL, pages not gone");
    }
    return (err);
}

/*
 * Write out a single page.
 * For swapfs this means choose a physical swap slot and write the page
 * out using VOP_PAGEIO.
 * In the (B_ASYNC | B_FREE) case we try to find a bunch of other dirty
 * swapfs pages, a bunch of contiguous swap slots and then write them
 * all out in one clustered i/o.
 */
/*ARGSUSED*/
static int
swap_putapage(
    struct vnode *vp,
    page_t *pp,
    u_offset_t *offp,
    size_t *lenp,
    int flags,
    struct cred *cr)
{
    int err;
    struct vnode *pvp;
    u_offset_t poff, off;
    u_offset_t doff;
    size_t dlen;
    size_t klsz = 0;
    u_offset_t klstart = 0;
    struct vnode *klvp = NULL;
    page_t *pplist;
    se_t se;
    struct async_reqs *arg;
    size_t swap_klustsize;

    /*
     * Clear force flag so that p_lckcnt pages are not invalidated.
     */
    flags &= ~B_FORCE;

    /*
     * This check is added for callers who access swap_putpage with len = 0.
     * swap_putpage calls swap_putapage page-by-page via pvn_vplist_dirty.
     * And it's necessary to do the same queuing if users have the same
     * B_ASYNC|B_FREE flags on.
     */
    if (flags == (B_ASYNC | B_FREE) &&
        sw_pending_size < klustsize && (arg = sw_getfree())) {

        hat_setmod(pp);
        page_io_unlock(pp);
        page_unlock(pp);

        arg->a_vp = vp;
        arg->a_off = pp->p_offset;
        arg->a_len = PAGESIZE;
        arg->a_flags = B_ASYNC | B_FREE;
        arg->a_cred = kcred;
        sw_putreq(arg);

        return (0);
    }

    SWAPFS_PRINT(SWAP_PUTP,
        "swap_putapage: pp %p, vp %p, off %llx, flags %x\n",
        pp, vp, pp->p_offset, flags, 0);

    ASSERT(PAGE_LOCKED(pp));

    off = pp->p_offset;

    doff = off;
    dlen = PAGESIZE;

    if (err = swap_newphysname(vp, off, &doff, &dlen, &pvp, &poff)) {
        err = (flags == (B_ASYNC | B_FREE) ? ENOMEM : 0);
        hat_setmod(pp);
        page_io_unlock(pp);
        page_unlock(pp);
        goto out;
    }

    klvp = pvp;
    klstart = poff;
    pplist = pp;
    /*
     * If this is ASYNC | FREE and we've accumulated a bunch of such
     * pending requests, kluster.
     */
    if (flags == (B_ASYNC | B_FREE))
        swap_klustsize = klustsize;
    else
        swap_klustsize = PAGESIZE;
    se = (flags & B_FREE ? SE_EXCL : SE_SHARED);
    klsz = PAGESIZE;
    while (klsz < swap_klustsize) {
        if ((arg = sw_getreq()) == NULL) {
            swap_getiofail++;
            swap_getiopages += btop(klsz);
            break;
        }
        ASSERT(vn_matchops(arg->a_vp, swap_vnodeops));
        vp = arg->a_vp;
        off = arg->a_off;

        if ((pp = page_lookup_nowait(vp, off, se)) == NULL) {
            swap_otherfail++;
            swap_otherpages += btop(klsz);
            sw_putfree(arg);
            break;
        }
        if (pvn_getdirty(pp, flags | B_DELWRI) == 0) {
            sw_putfree(arg);
            continue;
        }
        /* Get new physical backing store for the page */
        doff = off;
        dlen = PAGESIZE;
        if (err = swap_newphysname(vp, off, &doff, &dlen,
                        &pvp, &poff)) {
            swap_otherfail++;
            swap_otherpages += btop(klsz);
            hat_setmod(pp);
            page_io_unlock(pp);
            page_unlock(pp);
            sw_putbackreq(arg);
            break;
        }
        /* Try to cluster new physical name with previous ones */
        if (klvp == pvp && poff == klstart + klsz) {
            klsz += PAGESIZE;
            page_add(&pplist, pp);
            pplist = pplist->p_next;
            sw_putfree(arg);
        } else if (klvp == pvp && poff == klstart - PAGESIZE) {
            klsz += PAGESIZE;
            klstart -= PAGESIZE;
            page_add(&pplist, pp);
            sw_putfree(arg);
        } else {
            swap_klustfail++;
            swap_klustpages += btop(klsz);
            hat_setmod(pp);
            page_io_unlock(pp);
            page_unlock(pp);
            sw_putbackreq(arg);
            break;
        }
    }

    err = VOP_PAGEIO(klvp, pplist, klstart, klsz,
            B_WRITE | flags, cr);

    if ((flags & B_ASYNC) == 0)
        pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);

    /* Statistics */
    if (!err) {
        swap_putpagecnt++;
        swap_pagespushed += btop(klsz);
    }
out:
    TRACE_4(TR_FAC_SWAPFS, TR_SWAPFS_PUTAPAGE,
        "swapfs putapage:vp %p klvp %p, klstart %lx, klsz %lx",
        vp, klvp, klstart, klsz);
    if (err && err != ENOMEM)
        cmn_err(CE_WARN, "swapfs_putapage: err %d\n", err);
    if (lenp)
        *lenp = PAGESIZE;
    return (err);
}

static void
swap_dispose(vnode_t *vp, page_t *pp, int fl, int dn, cred_t *cr)
{
    int err;
    u_offset_t off = pp->p_offset;
    vnode_t *pvp;
    u_offset_t poff;

    ASSERT(PAGE_EXCL(pp));

    /*
     * The caller will free/invalidate large page in one shot instead of
     * one small page at a time.
     */
    if (pp->p_szc != 0) {
        page_unlock(pp);
        return;
    }

    err = swap_getphysname(vp, off, &pvp, &poff);
    if (!err && pvp != NULL)
        VOP_DISPOSE(pvp, pp, fl, dn, cr);
    else
        fs_dispose(vp, pp, fl, dn, cr);
}