libkvm/common/kvm.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 2004 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

/*
 * Copyright (c) 2013, Joyent, Inc.  All rights reserved.
 */

#include <kvm.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>
#include <unistd.h>
#include <limits.h>
#include <fcntl.h>
#include <strings.h>
#include <errno.h>
#include <sys/mem.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <sys/dumphdr.h>
#include <sys/sysmacros.h>

struct _kvmd {
    struct dumphdr  kvm_dump;
    char        *kvm_debug;
    int     kvm_openflag;
    int     kvm_corefd;
    int     kvm_kmemfd;
    int     kvm_memfd;
    size_t      kvm_coremapsize;
    char        *kvm_core;
    dump_map_t  *kvm_map;
    pfn_t       *kvm_pfn;
    struct as   *kvm_kas;
    proc_t      *kvm_practive;
    pid_t       kvm_pid;
    char        kvm_namelist[MAXNAMELEN + 1];
    boolean_t   kvm_namelist_core;
    proc_t      kvm_proc;
};

#define PREAD   (ssize_t (*)(int, void *, size_t, offset_t))pread64
#define PWRITE  (ssize_t (*)(int, void *, size_t, offset_t))pwrite64

static int kvm_nlist_core(kvm_t *kd, struct nlist nl[], const char *err);

static kvm_t *
fail(kvm_t *kd, const char *err, const char *message, ...)
{
    va_list args;

    va_start(args, message);
    if (err || (kd && kd->kvm_debug)) {
        (void) fprintf(stderr, "%s: ", err ? err : "KVM_DEBUG");
        (void) vfprintf(stderr, message, args);
        (void) fprintf(stderr, "\n");
    }
    va_end(args);
    if (kd != NULL)
        (void) kvm_close(kd);
    return (NULL);
}

/*ARGSUSED*/
kvm_t *
kvm_open(const char *namelist, const char *corefile, const char *swapfile,
    int flag, const char *err)
{
    kvm_t *kd;
    struct stat64 memstat, kmemstat, allkmemstat, corestat;
    struct nlist nl[3] = { { "kas" }, { "practive" }, { "" } };

    if ((kd = calloc(1, sizeof (kvm_t))) == NULL)
        return (fail(NULL, err, "cannot allocate space for kvm_t"));

    kd->kvm_corefd = kd->kvm_kmemfd = kd->kvm_memfd = -1;
    kd->kvm_debug = getenv("KVM_DEBUG");

    if ((kd->kvm_openflag = flag) != O_RDONLY && flag != O_RDWR)
        return (fail(kd, err, "illegal flag 0x%x to kvm_open()", flag));

    if (corefile == NULL)
        corefile = "/dev/kmem";

    if (stat64(corefile, &corestat) == -1)
        return (fail(kd, err, "cannot stat %s", corefile));

    if (S_ISCHR(corestat.st_mode)) {
        if (stat64("/dev/mem", &memstat) == -1)
            return (fail(kd, err, "cannot stat /dev/mem"));

        if (stat64("/dev/kmem", &kmemstat) == -1)
            return (fail(kd, err, "cannot stat /dev/kmem"));

        if (stat64("/dev/allkmem", &allkmemstat) == -1)
            return (fail(kd, err, "cannot stat /dev/allkmem"));
        if (corestat.st_rdev == memstat.st_rdev ||
            corestat.st_rdev == kmemstat.st_rdev ||
            corestat.st_rdev == allkmemstat.st_rdev) {
            char *kmem = (corestat.st_rdev == allkmemstat.st_rdev ?
                "/dev/allkmem" : "/dev/kmem");

            if ((kd->kvm_kmemfd = open64(kmem, flag)) == -1)
                return (fail(kd, err, "cannot open %s", kmem));
            if ((kd->kvm_memfd = open64("/dev/mem", flag)) == -1)
                return (fail(kd, err, "cannot open /dev/mem"));
        }
    } else {
        if ((kd->kvm_corefd = open64(corefile, flag)) == -1)
            return (fail(kd, err, "cannot open %s", corefile));
        if (pread64(kd->kvm_corefd, &kd->kvm_dump,
            sizeof (kd->kvm_dump), 0) != sizeof (kd->kvm_dump))
            return (fail(kd, err, "cannot read dump header"));
        if (kd->kvm_dump.dump_magic != DUMP_MAGIC)
            return (fail(kd, err, "%s is not a kernel core file "
                "(bad magic number %x)", corefile,
                kd->kvm_dump.dump_magic));
        if (kd->kvm_dump.dump_version != DUMP_VERSION)
            return (fail(kd, err,
                "libkvm version (%u) != corefile version (%u)",
                DUMP_VERSION, kd->kvm_dump.dump_version));
        if (kd->kvm_dump.dump_wordsize != DUMP_WORDSIZE)
            return (fail(kd, err, "%s is a %d-bit core file - "
                "cannot examine with %d-bit libkvm", corefile,
                kd->kvm_dump.dump_wordsize, DUMP_WORDSIZE));
        /*
         * We try to mmap(2) the entire corefile for performance
         * (so we can use bcopy(3C) rather than pread(2)).  Failing
         * that, we insist on at least mmap(2)ing the dump map.
         */
        kd->kvm_coremapsize = (size_t)corestat.st_size;
        if (corestat.st_size > LONG_MAX ||
            (kd->kvm_core = mmap64(0, kd->kvm_coremapsize,
            PROT_READ, MAP_SHARED, kd->kvm_corefd, 0)) == MAP_FAILED) {
            kd->kvm_coremapsize = kd->kvm_dump.dump_data;
            if ((kd->kvm_core = mmap64(0, kd->kvm_coremapsize,
                PROT_READ, MAP_SHARED, kd->kvm_corefd, 0)) ==
                MAP_FAILED)
                return (fail(kd, err, "cannot mmap corefile"));
        }
        kd->kvm_map = (void *)(kd->kvm_core + kd->kvm_dump.dump_map);
        kd->kvm_pfn = (void *)(kd->kvm_core + kd->kvm_dump.dump_pfn);
    }

    if (namelist == NULL)
        namelist = "/dev/ksyms";

    (void) strncpy(kd->kvm_namelist, namelist, MAXNAMELEN);

    if (kvm_nlist(kd, nl) == -1) {
        if (kd->kvm_corefd == -1) {
            return (fail(kd, err, "%s is not a %d-bit "
                "kernel namelist", namelist, DUMP_WORDSIZE));
        }

        if (kvm_nlist_core(kd, nl, err) == -1)
            return (NULL);      /* fail() already called */
    }

    kd->kvm_kas = (struct as *)nl[0].n_value;
    kd->kvm_practive = (proc_t *)nl[1].n_value;

    (void) kvm_setproc(kd);
    return (kd);
}

int
kvm_close(kvm_t *kd)
{
    if (kd->kvm_core != NULL && kd->kvm_core != MAP_FAILED)
        (void) munmap(kd->kvm_core, kd->kvm_coremapsize);
    if (kd->kvm_corefd != -1)
        (void) close(kd->kvm_corefd);
    if (kd->kvm_kmemfd != -1)
        (void) close(kd->kvm_kmemfd);
    if (kd->kvm_memfd != -1)
        (void) close(kd->kvm_memfd);
    if (kd->kvm_namelist_core)
        (void) unlink(kd->kvm_namelist);
    free(kd);
    return (0);
}

const char *
kvm_namelist(kvm_t *kd)
{
    return (kd->kvm_namelist);
}

int
kvm_nlist(kvm_t *kd, struct nlist nl[])
{
    return (nlist(kd->kvm_namelist, nl));
}

/*
 * If we don't have a name list, try to dig it out of the kernel crash dump.
 * (The symbols have been present in the dump, uncompressed, for nearly a
 * decade as of this writing -- and it is frankly surprising that the archaic
 * notion of a disjoint symbol table managed to survive that change.)
 */
static int
kvm_nlist_core(kvm_t *kd, struct nlist nl[], const char *err)
{
    dumphdr_t *dump = &kd->kvm_dump;
    char *msg = "couldn't extract symbols from dump";
    char *template = "/tmp/.libkvm.kvm_nlist_core.pid%d.XXXXXX";
    int fd, rval;

    if (dump->dump_ksyms_size != dump->dump_ksyms_csize) {
        (void) fail(kd, err, "%s: kernel symbols are compressed", msg);
        return (-1);
    }

    if (dump->dump_ksyms + dump->dump_ksyms_size > kd->kvm_coremapsize) {
        (void) fail(kd, err, "%s: kernel symbols not mapped", msg);
        return (-1);
    }

    /*
     * Beause this temporary file may be left as a turd if the caller
     * does not properly call kvm_close(), we make sure that it clearly
     * indicates its origins.
     */
    (void) snprintf(kd->kvm_namelist, MAXNAMELEN, template, getpid());

    if ((fd = mkstemp(kd->kvm_namelist)) == -1) {
        (void) fail(kd, err, "%s: couldn't create temporary "
            "symbols file: %s", msg, strerror(errno));
        return (-1);
    }

    kd->kvm_namelist_core = B_TRUE;

    do {
        rval = write(fd, (caddr_t)((uintptr_t)kd->kvm_core +
            (uintptr_t)dump->dump_ksyms), dump->dump_ksyms_size);
    } while (rval < dump->dump_ksyms_size && errno == EINTR);

    if (rval < dump->dump_ksyms_size) {
        (void) fail(kd, err, "%s: couldn't write to temporary "
            "symbols file: %s", msg, strerror(errno));
        (void) close(fd);
        return (-1);
    }

    (void) close(fd);

    if (kvm_nlist(kd, nl) == -1) {
        (void) fail(kd, err, "%s: symbols not valid", msg);
        return (-1);
    }

    return (0);
}

static offset_t
kvm_lookup(kvm_t *kd, struct as *as, uint64_t addr)
{
    uintptr_t pageoff = addr & (kd->kvm_dump.dump_pagesize - 1);
    uint64_t page = addr - pageoff;
    offset_t off = 0;

    if (kd->kvm_debug)
        fprintf(stderr, "kvm_lookup(%p, %llx):", (void *)as, addr);

    if (as == NULL) {       /* physical addressing mode */
        long first = 0;
        long last = kd->kvm_dump.dump_npages - 1;
        pfn_t target = (pfn_t)(page >> kd->kvm_dump.dump_pageshift);
        while (last >= first) {
            long middle = (first + last) / 2;
            pfn_t pfn = kd->kvm_pfn[middle];
            if (kd->kvm_debug)
                fprintf(stderr, " %ld ->", middle);
            if (pfn == target) {
                off = kd->kvm_dump.dump_data + pageoff +
                    ((uint64_t)middle <<
                    kd->kvm_dump.dump_pageshift);
                break;
            }
            if (pfn < target)
                first = middle + 1;
            else
                last = middle - 1;
        }
    } else {
        long hash = DUMP_HASH(&kd->kvm_dump, as, page);
        off = kd->kvm_map[hash].dm_first;
        while (off != 0) {
            dump_map_t *dmp = (void *)(kd->kvm_core + off);
            if (kd->kvm_debug)
                fprintf(stderr, " %llx ->", off);
            if (dmp < kd->kvm_map ||
                dmp > kd->kvm_map + kd->kvm_dump.dump_hashmask ||
                (off & (sizeof (offset_t) - 1)) != 0 ||
                DUMP_HASH(&kd->kvm_dump, dmp->dm_as, dmp->dm_va) !=
                hash) {
                if (kd->kvm_debug)
                    fprintf(stderr, " dump map corrupt\n");
                return (0);
            }
            if (dmp->dm_va == page && dmp->dm_as == as) {
                off = dmp->dm_data + pageoff;
                break;
            }
            off = dmp->dm_next;
        }
    }
    if (kd->kvm_debug)
        fprintf(stderr, "%s found: %llx\n", off ? "" : " not", off);
    return (off);
}

static ssize_t
kvm_rw(kvm_t *kd, uint64_t addr, void *buf, size_t size,
    struct as *as, ssize_t (*prw)(int, void *, size_t, offset_t))
{
    offset_t off;
    size_t resid = size;

    /*
     * read/write of zero bytes always succeeds
     */
    if (size == 0)
        return (0);

    if (kd->kvm_core == NULL) {
        char procbuf[100];
        int procfd;
        ssize_t rval;

        if (as == kd->kvm_kas)
            return (prw(kd->kvm_kmemfd, buf, size, addr));
        if (as == NULL)
            return (prw(kd->kvm_memfd, buf, size, addr));

        (void) sprintf(procbuf, "/proc/%ld/as", kd->kvm_pid);
        if ((procfd = open64(procbuf, kd->kvm_openflag)) == -1)
            return (-1);
        rval = prw(procfd, buf, size, addr);
        (void) close(procfd);
        return (rval);
    }

    while (resid != 0) {
        uintptr_t pageoff = addr & (kd->kvm_dump.dump_pagesize - 1);
        ssize_t len = MIN(resid, kd->kvm_dump.dump_pagesize - pageoff);

        if ((off = kvm_lookup(kd, as, addr)) == 0)
            break;

        if (prw == PREAD && off < kd->kvm_coremapsize)
            bcopy(kd->kvm_core + off, buf, len);
        else if ((len = prw(kd->kvm_corefd, buf, len, off)) <= 0)
            break;
        resid -= len;
        addr += len;
        buf = (char *)buf + len;
    }
    return (resid < size ? size - resid : -1);
}

ssize_t
kvm_read(kvm_t *kd, uintptr_t addr, void *buf, size_t size)
{
    return (kvm_rw(kd, addr, buf, size, kd->kvm_kas, PREAD));
}

ssize_t
kvm_kread(kvm_t *kd, uintptr_t addr, void *buf, size_t size)
{
    return (kvm_rw(kd, addr, buf, size, kd->kvm_kas, PREAD));
}

ssize_t
kvm_uread(kvm_t *kd, uintptr_t addr, void *buf, size_t size)
{
    return (kvm_rw(kd, addr, buf, size, kd->kvm_proc.p_as, PREAD));
}

ssize_t
kvm_aread(kvm_t *kd, uintptr_t addr, void *buf, size_t size, struct as *as)
{
    return (kvm_rw(kd, addr, buf, size, as, PREAD));
}

ssize_t
kvm_pread(kvm_t *kd, uint64_t addr, void *buf, size_t size)
{
    return (kvm_rw(kd, addr, buf, size, NULL, PREAD));
}

ssize_t
kvm_write(kvm_t *kd, uintptr_t addr, const void *buf, size_t size)
{
    return (kvm_rw(kd, addr, (void *)buf, size, kd->kvm_kas, PWRITE));
}

ssize_t
kvm_kwrite(kvm_t *kd, uintptr_t addr, const void *buf, size_t size)
{
    return (kvm_rw(kd, addr, (void *)buf, size, kd->kvm_kas, PWRITE));
}

ssize_t
kvm_uwrite(kvm_t *kd, uintptr_t addr, const void *buf, size_t size)
{
    return (kvm_rw(kd, addr, (void *)buf, size, kd->kvm_proc.p_as, PWRITE));
}

ssize_t
kvm_awrite(kvm_t *kd, uintptr_t addr, const void *buf, size_t size,
    struct as *as)
{
    return (kvm_rw(kd, addr, (void *)buf, size, as, PWRITE));
}

ssize_t
kvm_pwrite(kvm_t *kd, uint64_t addr, const void *buf, size_t size)
{
    return (kvm_rw(kd, addr, (void *)buf, size, NULL, PWRITE));
}

uint64_t
kvm_physaddr(kvm_t *kd, struct as *as, uintptr_t addr)
{
    mem_vtop_t mem_vtop;
    offset_t off;

    if (kd->kvm_core == NULL) {
        mem_vtop.m_as = as;
        mem_vtop.m_va = (void *)addr;
        if (ioctl(kd->kvm_kmemfd, MEM_VTOP, &mem_vtop) == 0)
            return ((uint64_t)mem_vtop.m_pfn * getpagesize() +
                (addr & (getpagesize() - 1)));
    } else {
        if ((off = kvm_lookup(kd, as, addr)) != 0) {
            long pfn_index =
                (u_offset_t)(off - kd->kvm_dump.dump_data) >>
                kd->kvm_dump.dump_pageshift;
            return (((uint64_t)kd->kvm_pfn[pfn_index] <<
                kd->kvm_dump.dump_pageshift) +
                (addr & (kd->kvm_dump.dump_pagesize - 1)));
        }
    }
    return (-1ULL);
}

struct proc *
kvm_getproc(kvm_t *kd, pid_t pid)
{
    (void) kvm_setproc(kd);
    while (kvm_nextproc(kd) != NULL)
        if (kd->kvm_pid == pid)
            return (&kd->kvm_proc);
    return (NULL);
}

struct proc *
kvm_nextproc(kvm_t *kd)
{
    if (kd->kvm_proc.p_next == NULL ||
        kvm_kread(kd, (uintptr_t)kd->kvm_proc.p_next,
        &kd->kvm_proc, sizeof (proc_t)) != sizeof (proc_t) ||
        kvm_kread(kd, (uintptr_t)&kd->kvm_proc.p_pidp->pid_id,
        &kd->kvm_pid, sizeof (pid_t)) != sizeof (pid_t))
        return (NULL);

    return (&kd->kvm_proc);
}

int
kvm_setproc(kvm_t *kd)
{
    (void) kvm_kread(kd, (uintptr_t)kd->kvm_practive,
        &kd->kvm_proc.p_next, sizeof (proc_t *));
    kd->kvm_pid = -1;
    return (0);
}

/*ARGSUSED*/
struct user *
kvm_getu(kvm_t *kd, struct proc *p)
{
    return (&p->p_user);
}