freebsd-dev/sys/dev/ksyms/ksyms.c
Hans Petter Selasky 07da61a6cc Streamline use of cdevpriv and correct some corner cases.
1) It is not useful to call "devfs_clear_cdevpriv()" from
"d_close" callbacks, hence for example read, write, ioctl and
so on might be sleeping at the time of "d_close" being called
and then then freed private data can still be accessed.
Examples: dtrace, linux_compat, ksyms (all fixed by this patch)

2) In sys/dev/drm* there are some cases in which memory will
be freed twice, if open fails, first by code in the open
routine, secondly by the cdevpriv destructor. Move registration
of the cdevpriv to the end of the drm open routines.

3) devfs_clear_cdevpriv() is not called if the "d_open" callback
registered cdevpriv data and the "d_open" callback function
returned an error. Fix this.

Discussed with:	phk
MFC after:	2 weeks
2012-08-15 16:19:39 +00:00

617 lines
15 KiB
C

/*-
* Copyright (c) 2008-2009, Stacey Son <sson@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.
*
* $FreeBSD$
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/conf.h>
#include <sys/elf.h>
#include <sys/ksyms.h>
#include <sys/linker.h>
#include <sys/malloc.h>
#include <sys/mman.h>
#include <sys/module.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/queue.h>
#include <sys/resourcevar.h>
#include <sys/stat.h>
#include <sys/uio.h>
#include <machine/elf.h>
#include <vm/pmap.h>
#include <vm/vm.h>
#include <vm/vm_extern.h>
#include <vm/vm_map.h>
#include "linker_if.h"
#define SHDR_NULL 0
#define SHDR_SYMTAB 1
#define SHDR_STRTAB 2
#define SHDR_SHSTRTAB 3
#define SHDR_NUM 4
#define STR_SYMTAB ".symtab"
#define STR_STRTAB ".strtab"
#define STR_SHSTRTAB ".shstrtab"
#define KSYMS_DNAME "ksyms"
static d_open_t ksyms_open;
static d_read_t ksyms_read;
static d_close_t ksyms_close;
static d_ioctl_t ksyms_ioctl;
static d_mmap_t ksyms_mmap;
static struct cdevsw ksyms_cdevsw = {
.d_version = D_VERSION,
.d_flags = D_TRACKCLOSE,
.d_open = ksyms_open,
.d_close = ksyms_close,
.d_read = ksyms_read,
.d_ioctl = ksyms_ioctl,
.d_mmap = ksyms_mmap,
.d_name = KSYMS_DNAME
};
struct ksyms_softc {
LIST_ENTRY(ksyms_softc) sc_list;
vm_offset_t sc_uaddr;
size_t sc_usize;
pmap_t sc_pmap;
struct proc *sc_proc;
};
static struct mtx ksyms_mtx;
static struct cdev *ksyms_dev;
static LIST_HEAD(, ksyms_softc) ksyms_list =
LIST_HEAD_INITIALIZER(ksyms_list);
static const char ksyms_shstrtab[] =
"\0" STR_SYMTAB "\0" STR_STRTAB "\0" STR_SHSTRTAB "\0";
struct ksyms_hdr {
Elf_Ehdr kh_ehdr;
Elf_Phdr kh_txtphdr;
Elf_Phdr kh_datphdr;
Elf_Shdr kh_shdr[SHDR_NUM];
char kh_shstrtab[sizeof(ksyms_shstrtab)];
};
struct tsizes {
size_t ts_symsz;
size_t ts_strsz;
};
struct toffsets {
vm_offset_t to_symoff;
vm_offset_t to_stroff;
unsigned to_stridx;
size_t to_resid;
};
static MALLOC_DEFINE(M_KSYMS, "KSYMS", "Kernel Symbol Table");
/*
* Get the symbol and string table sizes for a kernel module. Add it to the
* running total.
*/
static int
ksyms_size_permod(linker_file_t lf, void *arg)
{
struct tsizes *ts;
const Elf_Sym *symtab;
caddr_t strtab;
long syms;
ts = arg;
syms = LINKER_SYMTAB_GET(lf, &symtab);
ts->ts_symsz += syms * sizeof(Elf_Sym);
ts->ts_strsz += LINKER_STRTAB_GET(lf, &strtab);
return (0);
}
/*
* For kernel module get the symbol and string table sizes, returning the
* totals in *ts.
*/
static void
ksyms_size_calc(struct tsizes *ts)
{
ts->ts_symsz = 0;
ts->ts_strsz = 0;
(void) linker_file_foreach(ksyms_size_permod, ts);
}
#define KSYMS_EMIT(src, des, sz) do { \
copyout(src, (void *)des, sz); \
des += sz; \
} while (0)
#define SYMBLKSZ 256 * sizeof (Elf_Sym)
/*
* For a kernel module, add the symbol and string tables into the
* snapshot buffer. Fix up the offsets in the tables.
*/
static int
ksyms_add(linker_file_t lf, void *arg)
{
struct toffsets *to;
const Elf_Sym *symtab;
Elf_Sym *symp;
caddr_t strtab;
long symsz;
size_t strsz, numsyms;
linker_symval_t symval;
char *buf;
int i, nsyms, len;
to = arg;
MOD_SLOCK;
numsyms = LINKER_SYMTAB_GET(lf, &symtab);
strsz = LINKER_STRTAB_GET(lf, &strtab);
symsz = numsyms * sizeof(Elf_Sym);
buf = malloc(SYMBLKSZ, M_KSYMS, M_WAITOK);
while (symsz > 0) {
len = min(SYMBLKSZ, symsz);
bcopy(symtab, buf, len);
/*
* Fix up symbol table for kernel modules:
* string offsets need adjusted
* symbol values made absolute
*/
symp = (Elf_Sym *) buf;
nsyms = len / sizeof (Elf_Sym);
for (i = 0; i < nsyms; i++) {
symp[i].st_name += to->to_stridx;
if (lf->id > 1 && LINKER_SYMBOL_VALUES(lf,
(c_linker_sym_t) &symtab[i], &symval) == 0) {
symp[i].st_value = (uintptr_t) symval.value;
}
}
if (len > to->to_resid) {
MOD_SUNLOCK;
free(buf, M_KSYMS);
return (ENXIO);
} else
to->to_resid -= len;
KSYMS_EMIT(buf, to->to_symoff, len);
symtab += nsyms;
symsz -= len;
}
free(buf, M_KSYMS);
MOD_SUNLOCK;
if (strsz > to->to_resid)
return (ENXIO);
else
to->to_resid -= strsz;
KSYMS_EMIT(strtab, to->to_stroff, strsz);
to->to_stridx += strsz;
return (0);
}
/*
* Create a single ELF symbol table for the kernel and kernel modules loaded
* at this time. Write this snapshot out in the process address space. Return
* 0 on success, otherwise error.
*/
static int
ksyms_snapshot(struct tsizes *ts, vm_offset_t uaddr, size_t resid)
{
struct ksyms_hdr *hdr;
struct toffsets to;
int error = 0;
/* Be kernel stack friendly */
hdr = malloc(sizeof (*hdr), M_KSYMS, M_WAITOK|M_ZERO);
/*
* Create the ELF header.
*/
hdr->kh_ehdr.e_ident[EI_PAD] = 0;
hdr->kh_ehdr.e_ident[EI_MAG0] = ELFMAG0;
hdr->kh_ehdr.e_ident[EI_MAG1] = ELFMAG1;
hdr->kh_ehdr.e_ident[EI_MAG2] = ELFMAG2;
hdr->kh_ehdr.e_ident[EI_MAG3] = ELFMAG3;
hdr->kh_ehdr.e_ident[EI_DATA] = ELF_DATA;
hdr->kh_ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
hdr->kh_ehdr.e_ident[EI_CLASS] = ELF_CLASS;
hdr->kh_ehdr.e_ident[EI_VERSION] = EV_CURRENT;
hdr->kh_ehdr.e_ident[EI_ABIVERSION] = 0;
hdr->kh_ehdr.e_type = ET_EXEC;
hdr->kh_ehdr.e_machine = ELF_ARCH;
hdr->kh_ehdr.e_version = EV_CURRENT;
hdr->kh_ehdr.e_entry = 0;
hdr->kh_ehdr.e_phoff = offsetof(struct ksyms_hdr, kh_txtphdr);
hdr->kh_ehdr.e_shoff = offsetof(struct ksyms_hdr, kh_shdr);
hdr->kh_ehdr.e_flags = 0;
hdr->kh_ehdr.e_ehsize = sizeof(Elf_Ehdr);
hdr->kh_ehdr.e_phentsize = sizeof(Elf_Phdr);
hdr->kh_ehdr.e_phnum = 2; /* Text and Data */
hdr->kh_ehdr.e_shentsize = sizeof(Elf_Shdr);
hdr->kh_ehdr.e_shnum = SHDR_NUM;
hdr->kh_ehdr.e_shstrndx = SHDR_SHSTRTAB;
/*
* Add both the text and data Program headers.
*/
hdr->kh_txtphdr.p_type = PT_LOAD;
/* XXX - is there a way to put the actual .text addr/size here? */
hdr->kh_txtphdr.p_vaddr = 0;
hdr->kh_txtphdr.p_memsz = 0;
hdr->kh_txtphdr.p_flags = PF_R | PF_X;
hdr->kh_datphdr.p_type = PT_LOAD;
/* XXX - is there a way to put the actual .data addr/size here? */
hdr->kh_datphdr.p_vaddr = 0;
hdr->kh_datphdr.p_memsz = 0;
hdr->kh_datphdr.p_flags = PF_R | PF_W | PF_X;
/*
* Add the Section headers: null, symtab, strtab, shstrtab,
*/
/* First section header - null */
/* Second section header - symtab */
hdr->kh_shdr[SHDR_SYMTAB].sh_name = 1; /* String offset (skip null) */
hdr->kh_shdr[SHDR_SYMTAB].sh_type = SHT_SYMTAB;
hdr->kh_shdr[SHDR_SYMTAB].sh_flags = 0;
hdr->kh_shdr[SHDR_SYMTAB].sh_addr = 0;
hdr->kh_shdr[SHDR_SYMTAB].sh_offset = sizeof(*hdr);
hdr->kh_shdr[SHDR_SYMTAB].sh_size = ts->ts_symsz;
hdr->kh_shdr[SHDR_SYMTAB].sh_link = SHDR_STRTAB;
hdr->kh_shdr[SHDR_SYMTAB].sh_info = ts->ts_symsz / sizeof(Elf_Sym);
hdr->kh_shdr[SHDR_SYMTAB].sh_addralign = sizeof(long);
hdr->kh_shdr[SHDR_SYMTAB].sh_entsize = sizeof(Elf_Sym);
/* Third section header - strtab */
hdr->kh_shdr[SHDR_STRTAB].sh_name = 1 + sizeof(STR_SYMTAB);
hdr->kh_shdr[SHDR_STRTAB].sh_type = SHT_STRTAB;
hdr->kh_shdr[SHDR_STRTAB].sh_flags = 0;
hdr->kh_shdr[SHDR_STRTAB].sh_addr = 0;
hdr->kh_shdr[SHDR_STRTAB].sh_offset =
hdr->kh_shdr[SHDR_SYMTAB].sh_offset + ts->ts_symsz;
hdr->kh_shdr[SHDR_STRTAB].sh_size = ts->ts_strsz;
hdr->kh_shdr[SHDR_STRTAB].sh_link = 0;
hdr->kh_shdr[SHDR_STRTAB].sh_info = 0;
hdr->kh_shdr[SHDR_STRTAB].sh_addralign = sizeof(char);
hdr->kh_shdr[SHDR_STRTAB].sh_entsize = 0;
/* Fourth section - shstrtab */
hdr->kh_shdr[SHDR_SHSTRTAB].sh_name = 1 + sizeof(STR_SYMTAB) +
sizeof(STR_STRTAB);
hdr->kh_shdr[SHDR_SHSTRTAB].sh_type = SHT_STRTAB;
hdr->kh_shdr[SHDR_SHSTRTAB].sh_flags = 0;
hdr->kh_shdr[SHDR_SHSTRTAB].sh_addr = 0;
hdr->kh_shdr[SHDR_SHSTRTAB].sh_offset =
offsetof(struct ksyms_hdr, kh_shstrtab);
hdr->kh_shdr[SHDR_SHSTRTAB].sh_size = sizeof(ksyms_shstrtab);
hdr->kh_shdr[SHDR_SHSTRTAB].sh_link = 0;
hdr->kh_shdr[SHDR_SHSTRTAB].sh_info = 0;
hdr->kh_shdr[SHDR_SHSTRTAB].sh_addralign = 0 /* sizeof(char) */;
hdr->kh_shdr[SHDR_SHSTRTAB].sh_entsize = 0;
/* Copy shstrtab into the header */
bcopy(ksyms_shstrtab, hdr->kh_shstrtab, sizeof(ksyms_shstrtab));
to.to_symoff = uaddr + hdr->kh_shdr[SHDR_SYMTAB].sh_offset;
to.to_stroff = uaddr + hdr->kh_shdr[SHDR_STRTAB].sh_offset;
to.to_stridx = 0;
if (sizeof(struct ksyms_hdr) > resid) {
free(hdr, M_KSYMS);
return (ENXIO);
}
to.to_resid = resid - sizeof(struct ksyms_hdr);
/* Emit Header */
copyout(hdr, (void *)uaddr, sizeof(struct ksyms_hdr));
free(hdr, M_KSYMS);
/* Add symbol and string tables for each kernelmodule */
error = linker_file_foreach(ksyms_add, &to);
if (to.to_resid != 0)
return (ENXIO);
return (error);
}
static void
ksyms_cdevpriv_dtr(void *data)
{
struct ksyms_softc *sc;
sc = (struct ksyms_softc *)data;
mtx_lock(&ksyms_mtx);
LIST_REMOVE(sc, sc_list);
mtx_unlock(&ksyms_mtx);
free(sc, M_KSYMS);
}
/* ARGSUSED */
static int
ksyms_open(struct cdev *dev, int flags, int fmt __unused, struct thread *td)
{
struct tsizes ts;
size_t total_elf_sz;
int error, try;
struct ksyms_softc *sc;
/*
* Limit one open() per process. The process must close()
* before open()'ing again.
*/
mtx_lock(&ksyms_mtx);
LIST_FOREACH(sc, &ksyms_list, sc_list) {
if (sc->sc_proc == td->td_proc) {
mtx_unlock(&ksyms_mtx);
return (EBUSY);
}
}
sc = (struct ksyms_softc *) malloc(sizeof (*sc), M_KSYMS,
M_NOWAIT|M_ZERO);
if (sc == NULL) {
mtx_unlock(&ksyms_mtx);
return (ENOMEM);
}
sc->sc_proc = td->td_proc;
sc->sc_pmap = &td->td_proc->p_vmspace->vm_pmap;
LIST_INSERT_HEAD(&ksyms_list, sc, sc_list);
mtx_unlock(&ksyms_mtx);
error = devfs_set_cdevpriv(sc, ksyms_cdevpriv_dtr);
if (error)
goto failed;
/*
* MOD_SLOCK doesn't work here (because of a lock reversal with
* KLD_SLOCK). Therefore, simply try upto 3 times to get a "clean"
* snapshot of the kernel symbol table. This should work fine in the
* rare case of a kernel module being loaded/unloaded at the same
* time.
*/
for(try = 0; try < 3; try++) {
/*
* Map a buffer in the calling process memory space and
* create a snapshot of the kernel symbol table in it.
*/
/* Compute the size of buffer needed. */
ksyms_size_calc(&ts);
total_elf_sz = sizeof(struct ksyms_hdr) + ts.ts_symsz +
ts.ts_strsz;
error = copyout_map(td, &(sc->sc_uaddr),
(vm_size_t) total_elf_sz);
if (error)
break;
sc->sc_usize = total_elf_sz;
error = ksyms_snapshot(&ts, sc->sc_uaddr, total_elf_sz);
if (!error) {
/* Successful Snapshot */
return (0);
}
/* Snapshot failed, unmap the memory and try again */
(void) copyout_unmap(td, sc->sc_uaddr, sc->sc_usize);
}
failed:
ksyms_cdevpriv_dtr(sc);
return (error);
}
/* ARGSUSED */
static int
ksyms_read(struct cdev *dev, struct uio *uio, int flags __unused)
{
int error;
size_t len, sz;
struct ksyms_softc *sc;
off_t off;
char *buf;
vm_size_t ubase;
error = devfs_get_cdevpriv((void **)&sc);
if (error)
return (error);
off = uio->uio_offset;
len = uio->uio_resid;
if (off < 0 || off > sc->sc_usize)
return (EFAULT);
if (len > (sc->sc_usize - off))
len = sc->sc_usize - off;
if (len == 0)
return (0);
/*
* Since the snapshot buffer is in the user space we have to copy it
* in to the kernel and then back out. The extra copy saves valuable
* kernel memory.
*/
buf = malloc(PAGE_SIZE, M_KSYMS, M_WAITOK);
ubase = sc->sc_uaddr + off;
while (len) {
sz = min(PAGE_SIZE, len);
if (copyin((void *)ubase, buf, sz))
error = EFAULT;
else
error = uiomove(buf, sz, uio);
if (error)
break;
len -= sz;
ubase += sz;
}
free(buf, M_KSYMS);
return (error);
}
/* ARGSUSED */
static int
ksyms_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int32_t flag __unused,
struct thread *td __unused)
{
int error = 0;
struct ksyms_softc *sc;
error = devfs_get_cdevpriv((void **)&sc);
if (error)
return (error);
switch (cmd) {
case KIOCGSIZE:
/*
* Return the size (in bytes) of the symbol table
* snapshot.
*/
*(size_t *)data = sc->sc_usize;
break;
case KIOCGADDR:
/*
* Return the address of the symbol table snapshot.
* XXX - compat32 version of this?
*/
*(void **)data = (void *)sc->sc_uaddr;
break;
default:
error = ENOTTY;
break;
}
return (error);
}
/* ARGUSED */
static int
ksyms_mmap(struct cdev *dev, vm_ooffset_t offset, vm_paddr_t *paddr,
int prot __unused, vm_memattr_t *memattr __unused)
{
struct ksyms_softc *sc;
int error;
error = devfs_get_cdevpriv((void **)&sc);
if (error)
return (error);
/*
* XXX mmap() will actually map the symbol table into the process
* address space again.
*/
if (offset > round_page(sc->sc_usize) ||
(*paddr = pmap_extract(sc->sc_pmap,
(vm_offset_t)sc->sc_uaddr + offset)) == 0)
return (-1);
return (0);
}
/* ARGUSED */
static int
ksyms_close(struct cdev *dev, int flags __unused, int fmt __unused,
struct thread *td)
{
int error = 0;
struct ksyms_softc *sc;
error = devfs_get_cdevpriv((void **)&sc);
if (error)
return (error);
/* Unmap the buffer from the process address space. */
error = copyout_unmap(td, sc->sc_uaddr, sc->sc_usize);
return (error);
}
/* ARGSUSED */
static int
ksyms_modevent(module_t mod __unused, int type, void *data __unused)
{
int error = 0;
switch (type) {
case MOD_LOAD:
mtx_init(&ksyms_mtx, "KSyms mtx", NULL, MTX_DEF);
ksyms_dev = make_dev(&ksyms_cdevsw, 0, UID_ROOT, GID_WHEEL,
0444, KSYMS_DNAME);
break;
case MOD_UNLOAD:
if (!LIST_EMPTY(&ksyms_list))
return (EBUSY);
destroy_dev(ksyms_dev);
mtx_destroy(&ksyms_mtx);
break;
case MOD_SHUTDOWN:
break;
default:
error = EOPNOTSUPP;
break;
}
return (error);
}
DEV_MODULE(ksyms, ksyms_modevent, NULL);
MODULE_VERSION(ksyms, 1);