diff options
author | Bjoern A. Zeeb <bz@FreeBSD.org> | 2009-07-23 21:12:21 +0000 |
---|---|---|
committer | Bjoern A. Zeeb <bz@FreeBSD.org> | 2009-07-23 21:12:21 +0000 |
commit | 7cf8b4b9335f3632051e32a9819e87c663ee70d2 (patch) | |
tree | b8a5e23e08212e3f59199c99ea9ba792ed035e7b | |
parent | d0728d71742e6b2304a3b370f4f4bec9e7d1dc5e (diff) |
Notes
-rw-r--r-- | lib/libkvm/Makefile | 2 | ||||
-rw-r--r-- | lib/libkvm/kvm.c | 172 | ||||
-rw-r--r-- | lib/libkvm/kvm_private.h | 10 | ||||
-rw-r--r-- | lib/libkvm/kvm_vnet.c | 237 |
4 files changed, 412 insertions, 9 deletions
diff --git a/lib/libkvm/Makefile b/lib/libkvm/Makefile index 52a22caf8f47c..47dc613ab7ed1 100644 --- a/lib/libkvm/Makefile +++ b/lib/libkvm/Makefile @@ -10,7 +10,7 @@ CFLAGS+=-DSUN4V .endif SRCS= kvm.c kvm_${MACHINE_ARCH}.c kvm_cptime.c kvm_file.c kvm_getloadavg.c \ - kvm_getswapinfo.c kvm_pcpu.c kvm_proc.c + kvm_getswapinfo.c kvm_pcpu.c kvm_proc.c kvm_vnet.c .if ${MACHINE_ARCH} == "amd64" || ${MACHINE_ARCH} == "i386" || ${MACHINE_ARCH} == "arm" SRCS+= kvm_minidump_${MACHINE_ARCH}.c .endif diff --git a/lib/libkvm/kvm.c b/lib/libkvm/kvm.c index 8ff8a2b9f9864..50f752aa659db 100644 --- a/lib/libkvm/kvm.c +++ b/lib/libkvm/kvm.c @@ -41,6 +41,9 @@ static char sccsid[] = "@(#)kvm.c 8.2 (Berkeley) 2/13/94"; #endif /* LIBC_SCCS and not lint */ #include <sys/param.h> + +#define _WANT_VNET + #include <sys/user.h> #include <sys/proc.h> #include <sys/ioctl.h> @@ -48,6 +51,8 @@ static char sccsid[] = "@(#)kvm.c 8.2 (Berkeley) 2/13/94"; #include <sys/sysctl.h> #include <sys/linker.h> +#include <net/vnet.h> + #include <vm/vm.h> #include <vm/vm_param.h> @@ -62,6 +67,7 @@ static char sccsid[] = "@(#)kvm.c 8.2 (Berkeley) 2/13/94"; #include <stdio.h> #include <stdlib.h> #include <string.h> +#include <strings.h> #include <unistd.h> #include "kvm_private.h" @@ -299,34 +305,154 @@ kvm_close(kd) return (0); } +/* + * Walk the list of unresolved symbols, generate a new list and prefix the + * symbol names, try again, and merge back what we could resolve. + */ +static int +kvm_fdnlist_prefix(kvm_t *kd, struct nlist *nl, int missing, const char *prefix, + uintptr_t (*validate_fn)(kvm_t *, uintptr_t)) +{ + struct nlist *n, *np, *p; + char *cp, *ce; + size_t len; + int unresolved; + + /* + * Calculate the space we need to malloc for nlist and names. + * We are going to store the name twice for later lookups: once + * with the prefix and once the unmodified name delmited by \0. + */ + len = 0; + unresolved = 0; + for (p = nl; p->n_name && p->n_name[0]; ++p) { + if (p->n_type != N_UNDF) + continue; + len += sizeof(struct nlist) + strlen(prefix) + + 2 * (strlen(p->n_name) + 1); + unresolved++; + } + if (unresolved == 0) + return (unresolved); + /* Add space for the terminating nlist entry. */ + len += sizeof(struct nlist); + unresolved++; + + /* Alloc one chunk for (nlist, [names]) and setup pointers. */ + n = np = malloc(len); + bzero(n, len); + if (n == NULL) + return (missing); + cp = ce = (char *)np; + cp += unresolved * sizeof(struct nlist); + ce += len; + + /* Generate shortened nlist with special prefix. */ + unresolved = 0; + for (p = nl; p->n_name && p->n_name[0]; ++p) { + if (p->n_type != N_UNDF) + continue; + bcopy(p, np, sizeof(struct nlist)); + /* Save the new\0orig. name so we can later match it again. */ + len = snprintf(cp, ce - cp, "%s%s%c%s", prefix, + (prefix[0] != '\0' && p->n_name[0] == '_') ? + (p->n_name + 1) : p->n_name, '\0', p->n_name); + if (len >= ce - cp) + continue; + np->n_name = cp; + cp += len + 1; + np++; + unresolved++; + } + + /* Do lookup on the reduced list. */ + np = n; + unresolved = __fdnlist(kd->nlfd, np); + + /* Check if we could resolve further symbols and update the list. */ + if (unresolved >= 0 && unresolved < missing) { + /* Find the first freshly resolved entry. */ + for (; np->n_name && np->n_name[0]; np++) + if (np->n_type != N_UNDF) + break; + /* + * The lists are both in the same order, + * so we can walk them in parallel. + */ + for (p = nl; np->n_name && np->n_name[0] && + p->n_name && p->n_name[0]; ++p) { + if (p->n_type != N_UNDF) + continue; + /* Skip expanded name and compare to orig. one. */ + cp = np->n_name + strlen(np->n_name) + 1; + if (strcmp(cp, p->n_name)) + continue; + /* Update nlist with new, translated results. */ + p->n_type = np->n_type; + p->n_other = np->n_other; + p->n_desc = np->n_desc; + if (validate_fn) + p->n_value = (*validate_fn)(kd, np->n_value); + else + p->n_value = np->n_value; + missing--; + /* Find next freshly resolved entry. */ + for (np++; np->n_name && np->n_name[0]; np++) + if (np->n_type != N_UNDF) + break; + } + } + /* We could assert missing = unresolved here. */ + + free(n); + return (unresolved); +} + int -kvm_nlist(kd, nl) - kvm_t *kd; - struct nlist *nl; +_kvm_nlist(kvm_t *kd, struct nlist *nl, int initialize) { struct nlist *p; int nvalid; struct kld_sym_lookup lookup; int error; - + char *prefix = "", symname[1024]; /* XXX-BZ symbol name length limit? */ /* * If we can't use the kld symbol lookup, revert to the * slow library call. */ - if (!ISALIVE(kd)) - return (__fdnlist(kd->nlfd, nl)); + if (!ISALIVE(kd)) { + error = __fdnlist(kd->nlfd, nl); + if (error <= 0) /* Hard error or success. */ + return (error); + + if (_kvm_vnet_initialized(kd, initialize)) + error = kvm_fdnlist_prefix(kd, nl, error, + VNET_SYMPREFIX, _kvm_vnet_validaddr); + + return (error); + } /* * We can use the kld lookup syscall. Go through each nlist entry * and look it up with a kldsym(2) syscall. */ nvalid = 0; +again: for (p = nl; p->n_name && p->n_name[0]; ++p) { + if (p->n_type != N_UNDF) + continue; + lookup.version = sizeof(lookup); - lookup.symname = p->n_name; lookup.symvalue = 0; lookup.symsize = 0; + error = snprintf(symname, sizeof(symname), "%s%s", prefix, + (prefix[0] != '\0' && p->n_name[0] == '_') ? + (p->n_name + 1) : p->n_name); + if (error >= sizeof(symname)) + continue; + + lookup.symname = symname; if (lookup.symname[0] == '_') lookup.symname++; @@ -334,11 +460,28 @@ kvm_nlist(kd, nl) p->n_type = N_TEXT; p->n_other = 0; p->n_desc = 0; - p->n_value = lookup.symvalue; + if (_kvm_vnet_initialized(kd, initialize) && + !strcmp(prefix, VNET_SYMPREFIX)) + p->n_value = + _kvm_vnet_validaddr(kd, lookup.symvalue); + else + p->n_value = lookup.symvalue; ++nvalid; /* lookup.symsize */ } } + + /* + * Check the number of entries that weren't found. If they exist, + * try again with a prefix for virtualized symbol names. + */ + error = ((p - nl) - nvalid); + if (error && _kvm_vnet_initialized(kd, initialize) && + strcmp(prefix, VNET_SYMPREFIX)) { + prefix = VNET_SYMPREFIX; + goto again; + } + /* * Return the number of entries that weren't found. If they exist, * also fill internal error buffer. @@ -349,6 +492,19 @@ kvm_nlist(kd, nl) return (error); } +int +kvm_nlist(kd, nl) + kvm_t *kd; + struct nlist *nl; +{ + + /* + * If called via the public interface, permit intialization of + * further virtualized modules on demand. + */ + return (_kvm_nlist(kd, nl, 1)); +} + ssize_t kvm_read(kd, kva, buf, len) kvm_t *kd; diff --git a/lib/libkvm/kvm_private.h b/lib/libkvm/kvm_private.h index 88ff7166b59b3..cc073db5fe424 100644 --- a/lib/libkvm/kvm_private.h +++ b/lib/libkvm/kvm_private.h @@ -62,6 +62,12 @@ struct __kvm { */ struct vmstate *vmst; int rawdump; /* raw dump format */ + + int vnet_initialized; /* vnet fields set up */ + uintptr_t vnet_start; /* start of kernel's vnet region */ + uintptr_t vnet_stop; /* stop of kernel's vnet region */ + uintptr_t vnet_current; /* vnet we're working with */ + uintptr_t vnet_base; /* vnet base of current vnet */ }; /* @@ -74,10 +80,14 @@ void _kvm_freevtop(kvm_t *); int _kvm_initvtop(kvm_t *); int _kvm_kvatop(kvm_t *, u_long, off_t *); void *_kvm_malloc(kvm_t *kd, size_t); +int _kvm_nlist(kvm_t *, struct nlist *, int); void *_kvm_realloc(kvm_t *kd, void *, size_t); void _kvm_syserr (kvm_t *kd, const char *program, const char *fmt, ...) __printflike(3, 4); int _kvm_uvatop(kvm_t *, const struct proc *, u_long, u_long *); +int _kvm_vnet_selectpid(kvm_t *, pid_t); +int _kvm_vnet_initialized(kvm_t *, int); +uintptr_t _kvm_vnet_validaddr(kvm_t *, uintptr_t); #if defined(__amd64__) || defined(__i386__) || defined(__arm__) void _kvm_minidump_freevtop(kvm_t *); diff --git a/lib/libkvm/kvm_vnet.c b/lib/libkvm/kvm_vnet.c new file mode 100644 index 0000000000000..13f7cf4c664fe --- /dev/null +++ b/lib/libkvm/kvm_vnet.c @@ -0,0 +1,237 @@ +/*- + * Copyright (c) 2009 Robert N. M. Watson + * Copyright (c) 2009 Bjoern A. Zeeb <bz@FreeBSD.org> + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +#include <sys/param.h> + +#define _WANT_PRISON +#define _WANT_UCRED +#define _WANT_VNET + +#include <sys/_lock.h> +#include <sys/_mutex.h> +#include <sys/_task.h> +#include <sys/jail.h> +#include <sys/proc.h> +#include <sys/types.h> +#include <sys/vimage.h> + +#include <net/vnet.h> + +#include <nlist.h> +#include <kvm.h> +#include <limits.h> +#include <stdlib.h> +#include <unistd.h> + +#include "kvm_private.h" + +/* + * Set up libkvm to handle virtual network stack symbols by selecting a + * starting pid. + */ +int +_kvm_vnet_selectpid(kvm_t *kd, pid_t pid) +{ + struct proc proc; + struct thread td; + struct ucred cred; + struct prison prison; + struct vnet vnet; + struct nlist nl[] = { + /* + * Note: kvm_nlist strips the first '_' so add an extra one + * here to __{start,stop}_set_vnet. + */ +#define NLIST_START_VNET 0 + { .n_name = "___start_" VNET_SETNAME }, +#define NLIST_STOP_VNET 1 + { .n_name = "___stop_" VNET_SETNAME }, +#define NLIST_VNET_HEAD 2 + { .n_name = "vnet_head" }, +#define NLIST_ALLPROC 3 + { .n_name = "allproc" }, +#define NLIST_DUMPTID 4 + { .n_name = "dumptid" }, +#define NLIST_PROC0 5 + { .n_name = "proc0" }, + { .n_name = NULL }, + }; + uintptr_t procp, tdp, credp; + lwpid_t dumptid; + + /* + * Locate and cache locations of important symbols + * using the internal version of _kvm_nlist, turning + * off initialization to avoid recursion in case of + * unresolveable symbols. + */ + if (_kvm_nlist(kd, nl, 0) != 0) { + /* + * XXX-BZ: ___start_/___stop_VNET_SETNAME may fail. + * For now do not report an error here as we are called + * internally and in `void context' until we merge the + * functionality to optionally activate this into programs. + * By that time we can properly fail and let the callers + * handle the error. + */ + /* _kvm_err(kd, kd->program, "%s: no namelist", __func__); */ + return (-1); + } + + /* + * Auto-detect if this is a crashdump by reading dumptid. + */ + dumptid = 0; + if (nl[NLIST_DUMPTID].n_value) { + if (kvm_read(kd, nl[NLIST_DUMPTID].n_value, &dumptid, + sizeof(dumptid)) != sizeof(dumptid)) { + _kvm_err(kd, kd->program, "%s: dumptid", __func__); + return (-1); + } + } + + /* + * First, find the process for this pid. If we are workig on a dump, + * either locate the thread dumptid is refering to or proc0. + * Based on either, take the address of the ucred. + */ + credp = 0; + + procp = nl[NLIST_ALLPROC].n_value; +#define VMCORE_VNET_OF_PROC0 +#ifdef VMCORE_VNET_OF_PROC0 + if (dumptid > 0) { + procp = nl[NLIST_PROC0].n_value; + pid = 0; + } +#endif + while (procp != 0) { + if (kvm_read(kd, procp, &proc, sizeof(proc)) != sizeof(proc)) { + _kvm_err(kd, kd->program, "%s: proc", __func__); + return (-1); + } +#ifndef VMCORE_VNET_OF_PROC0 + if (dumptid > 0) { + tdp = (uintptr_t)TAILQ_FIRST(&proc.p_threads); + while (tdp != 0) { + if (kvm_read(kd, tdp, &td, sizeof(td)) != + sizeof(td)) { + _kvm_err(kd, kd->program, "%s: thread", + __func__); + return (-1); + } + if (td.td_tid == dumptid) { + credp = (uintptr_t)td.td_ucred; + break; + } + tdp = (uintptr_t)TAILQ_NEXT(&td, td_plist); + } + } else +#endif + if (proc.p_pid == pid) + credp = (uintptr_t)proc.p_ucred; + if (credp != 0) + break; + procp = (uintptr_t)LIST_NEXT(&proc, p_list); + } + if (credp == 0) { + _kvm_err(kd, kd->program, "%s: pid/tid not found", __func__); + return (-1); + } + if (kvm_read(kd, (uintptr_t)credp, &cred, sizeof(cred)) != + sizeof(cred)) { + _kvm_err(kd, kd->program, "%s: cred", __func__); + return (-1); + } + if (cred.cr_prison == NULL) { + _kvm_err(kd, kd->program, "%s: no jail", __func__); + return (-1); + } + if (kvm_read(kd, (uintptr_t)cred.cr_prison, &prison, sizeof(prison)) != + sizeof(prison)) { + _kvm_err(kd, kd->program, "%s: prison", __func__); + return (-1); + } + if (prison.pr_vnet == NULL) { + _kvm_err(kd, kd->program, "%s: no vnet", __func__); + return (-1); + } + if (kvm_read(kd, (uintptr_t)prison.pr_vnet, &vnet, sizeof(vnet)) != + sizeof(vnet)) { + _kvm_err(kd, kd->program, "%s: vnet", __func__); + return (-1); + } + if (vnet.vnet_magic_n != VNET_MAGIC_N) { + _kvm_err(kd, kd->program, "%s: invalid vnet magic#", __func__); + return (-1); + } + kd->vnet_initialized = 1; + kd->vnet_start = nl[NLIST_START_VNET].n_value; + kd->vnet_stop = nl[NLIST_STOP_VNET].n_value; + kd->vnet_current = (uintptr_t)prison.pr_vnet; + kd->vnet_base = (uintptr_t)vnet.vnet_data_mem - kd->vnet_start; + return (0); +} + +/* + * Check whether the vnet module has been initialized sucessfully + * or not, intialize it if permitted. + */ +int +_kvm_vnet_initialized(kvm_t *kd, int intialize) +{ + + if (kd->vnet_initialized || !intialize) + return (kd->vnet_initialized); + + (void) _kvm_vnet_selectpid(kd, getpid()); + + return (kd->vnet_initialized); +} + +/* + * Check whether the value is within the vnet symbol range and + * only if so adjust the offset relative to the current base. + */ +uintptr_t +_kvm_vnet_validaddr(kvm_t *kd, uintptr_t value) +{ + + if (value == 0) + return (value); + + if (!kd->vnet_initialized) + return (value); + + if (value < kd->vnet_start || value >= kd->vnet_stop) + return (value); + + return (kd->vnet_base + value); +} |