Use PC-relative relocations for USDT probe sites on i386 and amd64.

When recording probe site addresses in the output DOF file, dtrace -G
needs to emit relocations for the .SUNW_dof section in order to obtain
the addresses of functions containing probe sites. DTrace expects the
addresses to be relative to the base address of the final ELF file,
and the amd64 USDT implementation was relying on some unspecified and
incorrect behaviour in the base system GNU ld to achieve this.

This change reimplements the probe site relocation handling to allow
USDT to be used with lld and newer GNU binutils. Specifically, it
makes use of R_X86_64_PC64/R_386_PC32 relocations to obtain the
probe site address relative to the DOF file address, and adds and uses a
new DOF relocation type which computes the final probe site address using
these relative offsets.

Reported by and discussed with:	Rafael Espíndola
MFC after:	1 month
Differential Revision:	https://reviews.freebsd.org/D9374
This commit is contained in:
Mark Johnston 2017-02-05 02:39:12 +00:00
parent 69d2418faa
commit e801af6fba
Notes: svn2git 2020-12-20 02:59:44 +00:00
svn path=/head/; revision=313262
6 changed files with 73 additions and 117 deletions

View File

@ -462,18 +462,8 @@ dof_add_probe(dt_idhash_t *dhp, dt_ident_t *idp, void *data)
dt_buf_write(dtp, &ddo->ddo_enoffs, pip->pi_enoffs,
pip->pi_nenoffs * sizeof (uint32_t), sizeof (uint32_t));
/*
* If pi_rname isn't set, the relocation will be against the
* function name. If it is, the relocation will be against
* pi_rname. This will be used if the function is scoped
* locally so an alternate symbol is added for the purpose
* of this relocation.
*/
if (pip->pi_rname == NULL)
dofr.dofr_name = dofpr.dofpr_func;
else
dofr.dofr_name = dof_add_string(ddo, pip->pi_rname);
dofr.dofr_type = DOF_RELO_SETX;
dofr.dofr_name = dof_add_string(ddo, pip->pi_rname);
dofr.dofr_type = DOF_RELO_DOFREL;
dofr.dofr_offset = dt_buf_len(&ddo->ddo_probes);
dofr.dofr_data = 0;

View File

@ -237,7 +237,7 @@ printf("%s:%s(%d): DOODAD\n",__FUNCTION__,__FILE__,__LINE__);
rel->r_offset = s->dofs_offset +
dofr[j].dofr_offset;
rel->r_info = ELF32_R_INFO(count + dep->de_global,
R_386_32);
R_386_PC32);
#elif defined(__mips__)
/* XXX */
printf("%s:%s(%d): DOODAD\n",__FUNCTION__,__FILE__,__LINE__);
@ -253,15 +253,6 @@ printf("%s:%s(%d): DOODAD\n",__FUNCTION__,__FILE__,__LINE__);
#elif defined(__riscv__)
/* XXX */
printf("%s:%s(%d): DOODAD\n",__FUNCTION__,__FILE__,__LINE__);
#elif defined(__sparc)
/*
* Add 4 bytes to hit the low half of this 64-bit
* big-endian address.
*/
rel->r_offset = s->dofs_offset +
dofr[j].dofr_offset + 4;
rel->r_info = ELF32_R_INFO(count + dep->de_global,
R_SPARC_32);
#else
#error unknown ISA
#endif
@ -270,7 +261,7 @@ printf("%s:%s(%d): DOODAD\n",__FUNCTION__,__FILE__,__LINE__);
sym->st_value = 0;
sym->st_size = 0;
sym->st_info = ELF32_ST_INFO(STB_GLOBAL, STT_FUNC);
sym->st_other = 0;
sym->st_other = ELF32_ST_VISIBILITY(STV_HIDDEN);
sym->st_shndx = SHN_UNDEF;
rel++;
@ -287,11 +278,7 @@ printf("%s:%s(%d): DOODAD\n",__FUNCTION__,__FILE__,__LINE__);
sym->st_value = 0;
sym->st_size = dof->dofh_filesz;
sym->st_info = ELF32_ST_INFO(STB_GLOBAL, STT_OBJECT);
#ifdef illumos
sym->st_other = 0;
#else
sym->st_other = ELF32_ST_VISIBILITY(STV_HIDDEN);
#endif
sym->st_shndx = ESHDR_DOF;
sym++;
@ -448,18 +435,8 @@ prepare_elf64(dtrace_hdl_t *dtp, const dof_hdr_t *dof, dof_elf64_t *dep)
#elif defined(__i386) || defined(__amd64)
rel->r_offset = s->dofs_offset +
dofr[j].dofr_offset;
#ifdef illumos
rel->r_info = ELF64_R_INFO(count + dep->de_global,
R_AMD64_64);
#else
rel->r_info = ELF64_R_INFO(count + dep->de_global,
R_X86_64_RELATIVE);
#endif
#elif defined(__sparc)
rel->r_offset = s->dofs_offset +
dofr[j].dofr_offset;
rel->r_info = ELF64_R_INFO(count + dep->de_global,
R_SPARC_64);
R_X86_64_PC64);
#else
#error unknown ISA
#endif
@ -468,7 +445,7 @@ prepare_elf64(dtrace_hdl_t *dtp, const dof_hdr_t *dof, dof_elf64_t *dep)
sym->st_value = 0;
sym->st_size = 0;
sym->st_info = GELF_ST_INFO(STB_GLOBAL, STT_FUNC);
sym->st_other = 0;
sym->st_other = ELF64_ST_VISIBILITY(STV_HIDDEN);
sym->st_shndx = SHN_UNDEF;
rel++;
@ -485,11 +462,7 @@ prepare_elf64(dtrace_hdl_t *dtp, const dof_hdr_t *dof, dof_elf64_t *dep)
sym->st_value = 0;
sym->st_size = dof->dofh_filesz;
sym->st_info = GELF_ST_INFO(STB_GLOBAL, STT_OBJECT);
#ifdef illumos
sym->st_other = 0;
#else
sym->st_other = ELF64_ST_VISIBILITY(STV_HIDDEN);
#endif
sym->st_shndx = ESHDR_DOF;
sym++;
@ -797,16 +770,15 @@ dump_elf64(dtrace_hdl_t *dtp, const dof_hdr_t *dof, int fd)
}
static int
dt_symtab_lookup(Elf_Data *data_sym, int nsym, uintptr_t addr, uint_t shn,
GElf_Sym *sym, int uses_funcdesc, Elf *elf)
dt_symtab_lookup(Elf_Data *data_sym, int start, int end, uintptr_t addr,
uint_t shn, GElf_Sym *sym, int uses_funcdesc, Elf *elf)
{
int i, ret = -1;
Elf64_Addr symval;
Elf_Scn *opd_scn;
Elf_Data *opd_desc;
GElf_Sym s;
int i;
for (i = 0; i < nsym && gelf_getsym(data_sym, i, sym) != NULL; i++) {
for (i = start; i < end && gelf_getsym(data_sym, i, sym) != NULL; i++) {
if (GELF_ST_TYPE(sym->st_info) == STT_FUNC) {
symval = sym->st_value;
if (uses_funcdesc) {
@ -816,20 +788,12 @@ dt_symtab_lookup(Elf_Data *data_sym, int nsym, uintptr_t addr, uint_t shn,
*(uint64_t*)((char *)opd_desc->d_buf + symval);
}
if ((uses_funcdesc || shn == sym->st_shndx) &&
symval <= addr &&
addr < symval + sym->st_size) {
if (GELF_ST_BIND(sym->st_info) == STB_GLOBAL)
return (0);
ret = 0;
s = *sym;
}
symval <= addr && addr < symval + sym->st_size)
return (0);
}
}
if (ret == 0)
*sym = s;
return (ret);
return (-1);
}
#if defined(__aarch64__)
@ -1237,7 +1201,7 @@ process_obj(dtrace_hdl_t *dtp, const char *obj, int *eprobesp)
dt_provider_t *pvp;
dt_probe_t *prp;
uint32_t off, eclass, emachine1, emachine2;
size_t symsize, nsym, isym, istr, len;
size_t symsize, osym, nsym, isym, istr, len;
key_t objkey;
dt_link_pair_t *pair, *bufs = NULL;
dt_strtab_t *strtab;
@ -1374,12 +1338,13 @@ process_obj(dtrace_hdl_t *dtp, const char *obj, int *eprobesp)
* target (text) section to replace the call instruction with
* one or more nops.
*
* If the function containing the probe is locally scoped
* (static), we create an alias used by the relocation in the
* generated object. The alias, a new symbol, will be global
* (so that the relocation from the generated object can be
* resolved), and hidden (so that it is converted to a local
* symbol at link time). Such aliases have this form:
* To avoid runtime overhead, the relocations added to the
* generated object should be resolved at static link time. We
* therefore create aliases for the functions that contain
* probes. An alias is global (so that the relocation from the
* generated object can be resolved), and hidden (so that its
* address is known at static link time). Such aliases have this
* form:
*
* $dtrace<key>.<function>
*
@ -1417,16 +1382,13 @@ process_obj(dtrace_hdl_t *dtp, const char *obj, int *eprobesp)
if (strncmp(s, dt_prefix, sizeof (dt_prefix) - 1) != 0)
continue;
if (dt_symtab_lookup(data_sym, isym, rela.r_offset,
shdr_rel.sh_info, &fsym,
(emachine1 == EM_PPC64), elf) != 0) {
if (dt_symtab_lookup(data_sym, 0, isym, rela.r_offset,
shdr_rel.sh_info, &fsym, (emachine1 == EM_PPC64),
elf) != 0) {
dt_strtab_destroy(strtab);
goto err;
}
if (GELF_ST_BIND(fsym.st_info) != STB_LOCAL)
continue;
if (fsym.st_name > data_str->d_size) {
dt_strtab_destroy(strtab);
goto err;
@ -1462,12 +1424,12 @@ process_obj(dtrace_hdl_t *dtp, const char *obj, int *eprobesp)
}
/*
* If needed, allocate the additional space for the symbol
* table and string table copying the old data into the new
* buffers, and marking the buffers as dirty. We inject those
* newly allocated buffers into the libelf data structures, but
* are still responsible for freeing them once we're done with
* the elf handle.
* If any probes were found, allocate the additional space for
* the symbol table and string table, copying the old data into
* the new buffers, and marking the buffers as dirty. We inject
* those newly allocated buffers into the libelf data
* structures, but are still responsible for freeing them once
* we're done with the elf handle.
*/
if (nsym > 0) {
/*
@ -1516,9 +1478,11 @@ process_obj(dtrace_hdl_t *dtp, const char *obj, int *eprobesp)
shdr_sym.sh_size += nsym * symsize;
(void) gelf_update_shdr(scn_sym, &shdr_sym);
osym = isym;
nsym += isym;
} else {
dt_strtab_destroy(strtab);
continue;
}
/*
@ -1577,8 +1541,11 @@ process_obj(dtrace_hdl_t *dtp, const char *obj, int *eprobesp)
bcopy(s, pname, p - s);
pname[p - s] = '\0';
if (dt_symtab_lookup(data_sym, isym, rela.r_offset,
shdr_rel.sh_info, &fsym,
if (dt_symtab_lookup(data_sym, osym, isym,
rela.r_offset, shdr_rel.sh_info, &fsym,
(emachine1 == EM_PPC64), elf) != 0 &&
dt_symtab_lookup(data_sym, 0, osym,
rela.r_offset, shdr_rel.sh_info, &fsym,
(emachine1 == EM_PPC64), elf) != 0)
goto err;
@ -1588,37 +1555,30 @@ process_obj(dtrace_hdl_t *dtp, const char *obj, int *eprobesp)
assert(GELF_ST_TYPE(fsym.st_info) == STT_FUNC);
/*
* If a NULL relocation name is passed to
* dt_probe_define(), the function name is used for the
* relocation. The relocation needs to use a mangled
* name if the symbol is locally scoped; the function
* name may need to change if we've found the global
* alias for the locally scoped symbol (we prefer
* global symbols to locals in dt_symtab_lookup()).
* If this is our first time encountering this symbol,
* emit an alias.
*/
s = (char *)data_str->d_buf + fsym.st_name;
r = NULL;
if (GELF_ST_BIND(fsym.st_info) == STB_LOCAL) {
if (strncmp(s, dt_symprefix,
sizeof (dt_symprefix) - 1) != 0) {
u_int bind = GELF_ST_BIND(fsym.st_info);
dsym = fsym;
dsym.st_name = istr;
dsym.st_info = GELF_ST_INFO(STB_GLOBAL,
STT_FUNC);
dsym.st_other =
ELF64_ST_VISIBILITY(STV_ELIMINATE);
dsym.st_info = GELF_ST_INFO(bind == STB_LOCAL ?
STB_GLOBAL : bind, STT_FUNC);
dsym.st_other = GELF_ST_VISIBILITY(STV_HIDDEN);
(void) gelf_update_sym(data_sym, isym, &dsym);
r = (char *)data_str->d_buf + istr;
istr += 1 + sprintf(r, dt_symfmt,
dt_symprefix, objkey, s);
r = (char *) data_str->d_buf + istr;
istr += 1 + sprintf(r, dt_symfmt, dt_symprefix, objkey,
s);
isym++;
assert(isym <= nsym);
} else if (strncmp(s, dt_symprefix,
strlen(dt_symprefix)) == 0) {
} else {
r = s;
if ((s = strchr(s, '.')) == NULL)
goto err;
s = strchr(s, '.');
assert(s != NULL);
s++;
}

View File

@ -545,9 +545,7 @@ dt_probe_define(dt_provider_t *pvp, dt_probe_t *prp,
for (pip = prp->pr_inst; pip != NULL; pip = pip->pi_next) {
if (strcmp(pip->pi_fname, fname) == 0 &&
((rname == NULL && pip->pi_rname == NULL) ||
(rname != NULL && pip->pi_rname != NULL &&
strcmp(pip->pi_rname, rname) == 0)))
strcmp(pip->pi_rname, rname) == 0)
break;
}
@ -565,7 +563,7 @@ dt_probe_define(dt_provider_t *pvp, dt_probe_t *prp,
if ((pip->pi_fname = strdup(fname)) == NULL)
goto nomem;
if (rname != NULL && (pip->pi_rname = strdup(rname)) == NULL)
if ((pip->pi_rname = strdup(rname)) == NULL)
goto nomem;
pip->pi_noffs = 0;
@ -605,7 +603,7 @@ dt_probe_define(dt_provider_t *pvp, dt_probe_t *prp,
dt_dprintf("defined probe %s %s:%s %s() +0x%x (%s)\n",
isenabled ? "(is-enabled)" : "",
pvp->pv_desc.dtvd_name, prp->pr_ident->di_name, fname, offset,
rname != NULL ? rname : fname);
rname);
assert(*noffs < *maxoffs);
(*offs)[(*noffs)++] = offset;

View File

@ -13917,12 +13917,13 @@ dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
/*
* Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the
* specified DOF. At present, this amounts to simply adding 'ubase' to the
* site of any user SETX relocations to account for load object base address.
* In the future, if we need other relocations, this function can be extended.
* specified DOF. SETX relocations are computed using 'ubase', the base load
* address of the object containing the DOF, and DOFREL relocations are relative
* to the relocation offset within the DOF.
*/
static int
dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase)
dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase,
uint64_t udaddr)
{
uintptr_t daddr = (uintptr_t)dof;
dof_relohdr_t *dofr =
@ -13960,6 +13961,7 @@ dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase)
case DOF_RELO_NONE:
break;
case DOF_RELO_SETX:
case DOF_RELO_DOFREL:
if (r->dofr_offset >= ts->dofs_size || r->dofr_offset +
sizeof (uint64_t) > ts->dofs_size) {
dtrace_dof_error(dof, "bad relocation offset");
@ -13971,7 +13973,11 @@ dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase)
return (-1);
}
*(uint64_t *)taddr += ubase;
if (r->dofr_type == DOF_RELO_SETX)
*(uint64_t *)taddr += ubase;
else
*(uint64_t *)taddr +=
udaddr + ts->dofs_offset + r->dofr_offset;
break;
default:
dtrace_dof_error(dof, "invalid relocation type");
@ -13992,7 +13998,7 @@ dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase)
*/
static int
dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr,
dtrace_enabling_t **enabp, uint64_t ubase, int noprobes)
dtrace_enabling_t **enabp, uint64_t ubase, uint64_t udaddr, int noprobes)
{
uint64_t len = dof->dofh_loadsz, seclen;
uintptr_t daddr = (uintptr_t)dof;
@ -14154,7 +14160,7 @@ dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr,
switch (sec->dofs_type) {
case DOF_SECT_URELHDR:
if (dtrace_dof_relocate(dof, sec, ubase) != 0)
if (dtrace_dof_relocate(dof, sec, ubase, udaddr) != 0)
return (-1);
break;
}
@ -15519,7 +15525,7 @@ dtrace_anon_property(void)
}
rv = dtrace_dof_slurp(dof, &state->dts_vstate, CRED(),
&dtrace_anon.dta_enabling, 0, B_TRUE);
&dtrace_anon.dta_enabling, 0, 0, B_TRUE);
if (rv == 0)
rv = dtrace_dof_options(dof, state);
@ -16290,7 +16296,7 @@ dtrace_helper_slurp(dof_hdr_t *dof, dof_helper_t *dhp, struct proc *p)
vstate = &help->dthps_vstate;
if ((rv = dtrace_dof_slurp(dof, vstate, NULL, &enab, dhp->dofhp_addr,
B_FALSE)) != 0) {
dhp->dofhp_dof, B_FALSE)) != 0) {
dtrace_dof_destroy(dof);
return (rv);
}

View File

@ -784,6 +784,7 @@ typedef struct dof_relodesc {
#define DOF_RELO_NONE 0 /* empty relocation entry */
#define DOF_RELO_SETX 1 /* relocate setx value */
#define DOF_RELO_DOFREL 2 /* relocate DOF-relative value */
typedef struct dof_optdesc {
uint32_t dofo_option; /* option identifier */

View File

@ -429,7 +429,8 @@ dtrace_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
return (EBUSY);
}
if (dtrace_dof_slurp(dof, vstate, td->td_ucred, &enab, 0, B_TRUE) != 0) {
if (dtrace_dof_slurp(dof, vstate, td->td_ucred, &enab, 0, 0,
B_TRUE) != 0) {
mutex_exit(&dtrace_lock);
mutex_exit(&cpu_lock);
dtrace_dof_destroy(dof);