/*
* Read a symbolic link
*/
static int
pfs_readlink(struct vop_readlink_args *va)
{
struct vnode *vn = va->a_vp;
struct pfs_vdata *pvd = vn->v_data;
struct pfs_node *pn = pvd->pvd_pn;
struct uio *uio = va->a_uio;
struct proc *proc = NULL;
struct thread *td = curthread;
char buf[PATH_MAX];
struct sbuf sb;
int error, locked;
PFS_TRACE(("%s", pn->pn_name));
pfs_assert_not_owned(pn);
if (vn->v_type != VLNK)
PFS_RETURN (EINVAL);
KASSERT_PN_IS_LINK(pn);
if (pn->pn_fill == NULL)
PFS_RETURN (EIO);
if (pvd->pvd_pid != NO_PID) {
if ((proc = pfind(pvd->pvd_pid)) == NULL)
PFS_RETURN (EIO);
if (proc->p_flag & P_WEXIT) {
PROC_UNLOCK(proc);
PFS_RETURN (EIO);
}
_PHOLD(proc);
PROC_UNLOCK(proc);
}
vhold(vn);
locked = VOP_ISLOCKED(vn, td);
VOP_UNLOCK(vn, 0, td);
/* sbuf_new() can't fail with a static buffer */
sbuf_new(&sb, buf, sizeof buf, 0);
error = pn_fill(td, proc, pn, &sb, NULL);
if (proc != NULL)
PRELE(proc);
vn_lock(vn, locked | LK_RETRY, td);
vdrop(vn);
if (error) {
sbuf_delete(&sb);
PFS_RETURN (error);
}
sbuf_finish(&sb);
error = uiomove_frombuf(sbuf_data(&sb), sbuf_len(&sb), uio);
sbuf_delete(&sb);
PFS_RETURN (error);
}
/*
* Write to a file
*/
static int
pfs_write(struct vop_write_args *va)
{
struct vnode *vn = va->a_vp;
struct pfs_vdata *pvd = vn->v_data;
struct pfs_node *pn = pvd->pvd_pn;
struct uio *uio = va->a_uio;
struct proc *proc;
struct sbuf sb;
int error;
PFS_TRACE(("%s", pn->pn_name));
pfs_assert_not_owned(pn);
if (vn->v_type != VREG)
PFS_RETURN (EINVAL);
KASSERT_PN_IS_FILE(pn);
if (!(pn->pn_flags & PFS_WR))
PFS_RETURN (EBADF);
if (pn->pn_fill == NULL)
PFS_RETURN (EIO);
/*
* This is necessary because either process' privileges may
* have changed since the open() call.
*/
if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
PFS_RETURN (EIO);
if (proc != NULL) {
_PHOLD(proc);
PROC_UNLOCK(proc);
}
if (pn->pn_flags & PFS_RAWWR) {
error = pn_fill(curthread, proc, pn, NULL, uio);
if (proc != NULL)
PRELE(proc);
PFS_RETURN (error);
}
sbuf_uionew(&sb, uio, &error);
if (error) {
if (proc != NULL)
PRELE(proc);
PFS_RETURN (error);
}
error = pn_fill(curthread, proc, pn, &sb, uio);
sbuf_delete(&sb);
if (proc != NULL)
PRELE(proc);
PFS_RETURN (error);
}
int
fasttrap_pid_probe(struct reg *rp)
{
proc_t *p = curproc;
uintptr_t pc = rp->r_rip - 1;
uintptr_t new_pc = 0;
fasttrap_bucket_t *bucket;
#if defined(sun)
kmutex_t *pid_mtx;
#endif
fasttrap_tracepoint_t *tp, tp_local;
pid_t pid;
dtrace_icookie_t cookie;
uint_t is_enabled = 0;
/*
* It's possible that a user (in a veritable orgy of bad planning)
* could redirect this thread's flow of control before it reached the
* return probe fasttrap. In this case we need to kill the process
* since it's in a unrecoverable state.
*/
if (curthread->t_dtrace_step) {
ASSERT(curthread->t_dtrace_on);
fasttrap_sigtrap(p, curthread, pc);
return (0);
}
/*
* Clear all user tracing flags.
*/
curthread->t_dtrace_ft = 0;
curthread->t_dtrace_pc = 0;
curthread->t_dtrace_npc = 0;
curthread->t_dtrace_scrpc = 0;
curthread->t_dtrace_astpc = 0;
#ifdef __amd64
curthread->t_dtrace_regv = 0;
#endif
#if defined(sun)
/*
* Treat a child created by a call to vfork(2) as if it were its
* parent. We know that there's only one thread of control in such a
* process: this one.
*/
while (p->p_flag & SVFORK) {
p = p->p_parent;
}
#endif
PROC_LOCK(p);
_PHOLD(p);
pid = p->p_pid;
#if defined(sun)
pid_mtx = &cpu_core[CPU->cpu_id].cpuc_pid_lock;
mutex_enter(pid_mtx);
#endif
bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)];
/*
* Lookup the tracepoint that the process just hit.
*/
for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) {
if (pid == tp->ftt_pid && pc == tp->ftt_pc &&
tp->ftt_proc->ftpc_acount != 0)
break;
}
/*
* If we couldn't find a matching tracepoint, either a tracepoint has
* been inserted without using the pid<pid> ioctl interface (see
* fasttrap_ioctl), or somehow we have mislaid this tracepoint.
*/
if (tp == NULL) {
#if defined(sun)
mutex_exit(pid_mtx);
#endif
_PRELE(p);
PROC_UNLOCK(p);
return (-1);
}
/*
* Set the program counter to the address of the traced instruction
* so that it looks right in ustack() output.
*/
rp->r_rip = pc;
if (tp->ftt_ids != NULL) {
fasttrap_id_t *id;
#ifdef __amd64
if (p->p_model == DATAMODEL_LP64) {
for (id = tp->ftt_ids; id != NULL; id = id->fti_next) {
fasttrap_probe_t *probe = id->fti_probe;
if (id->fti_ptype == DTFTP_ENTRY) {
/*
* We note that this was an entry
* probe to help ustack() find the
* first caller.
*/
cookie = dtrace_interrupt_disable();
DTRACE_CPUFLAG_SET(CPU_DTRACE_ENTRY);
dtrace_probe(probe->ftp_id, rp->r_rdi,
rp->r_rsi, rp->r_rdx, rp->r_rcx,
rp->r_r8);
DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_ENTRY);
dtrace_interrupt_enable(cookie);
} else if (id->fti_ptype == DTFTP_IS_ENABLED) {
/*
* Note that in this case, we don't
* call dtrace_probe() since it's only
* an artificial probe meant to change
* the flow of control so that it
* encounters the true probe.
*/
is_enabled = 1;
} else if (probe->ftp_argmap == NULL) {
dtrace_probe(probe->ftp_id, rp->r_rdi,
rp->r_rsi, rp->r_rdx, rp->r_rcx,
rp->r_r8);
} else {
uintptr_t t[5];
fasttrap_usdt_args64(probe, rp,
sizeof (t) / sizeof (t[0]), t);
dtrace_probe(probe->ftp_id, t[0], t[1],
t[2], t[3], t[4]);
}
}
} else {
#else /* __amd64 */
uintptr_t s0, s1, s2, s3, s4, s5;
uint32_t *stack = (uint32_t *)rp->r_esp;
/*
* In 32-bit mode, all arguments are passed on the
* stack. If this is a function entry probe, we need
* to skip the first entry on the stack as it
* represents the return address rather than a
* parameter to the function.
*/
s0 = fasttrap_fuword32_noerr(&stack[0]);
s1 = fasttrap_fuword32_noerr(&stack[1]);
s2 = fasttrap_fuword32_noerr(&stack[2]);
s3 = fasttrap_fuword32_noerr(&stack[3]);
s4 = fasttrap_fuword32_noerr(&stack[4]);
s5 = fasttrap_fuword32_noerr(&stack[5]);
for (id = tp->ftt_ids; id != NULL; id = id->fti_next) {
fasttrap_probe_t *probe = id->fti_probe;
if (id->fti_ptype == DTFTP_ENTRY) {
/*
* We note that this was an entry
* probe to help ustack() find the
* first caller.
*/
cookie = dtrace_interrupt_disable();
DTRACE_CPUFLAG_SET(CPU_DTRACE_ENTRY);
dtrace_probe(probe->ftp_id, s1, s2,
s3, s4, s5);
DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_ENTRY);
dtrace_interrupt_enable(cookie);
} else if (id->fti_ptype == DTFTP_IS_ENABLED) {
/*
* Note that in this case, we don't
* call dtrace_probe() since it's only
* an artificial probe meant to change
* the flow of control so that it
* encounters the true probe.
*/
is_enabled = 1;
} else if (probe->ftp_argmap == NULL) {
dtrace_probe(probe->ftp_id, s0, s1,
s2, s3, s4);
} else {
uint32_t t[5];
fasttrap_usdt_args32(probe, rp,
sizeof (t) / sizeof (t[0]), t);
dtrace_probe(probe->ftp_id, t[0], t[1],
t[2], t[3], t[4]);
}
}
#endif /* __amd64 */
#ifdef __amd64
}
#endif
}
/*
* We're about to do a bunch of work so we cache a local copy of
* the tracepoint to emulate the instruction, and then find the
* tracepoint again later if we need to light up any return probes.
*/
tp_local = *tp;
PROC_UNLOCK(p);
#if defined(sun)
mutex_exit(pid_mtx);
#endif
tp = &tp_local;
/*
* Set the program counter to appear as though the traced instruction
* had completely executed. This ensures that fasttrap_getreg() will
* report the expected value for REG_RIP.
*/
rp->r_rip = pc + tp->ftt_size;
/*
* If there's an is-enabled probe connected to this tracepoint it
* means that there was a 'xorl %eax, %eax' or 'xorq %rax, %rax'
* instruction that was placed there by DTrace when the binary was
* linked. As this probe is, in fact, enabled, we need to stuff 1
* into %eax or %rax. Accordingly, we can bypass all the instruction
* emulation logic since we know the inevitable result. It's possible
* that a user could construct a scenario where the 'is-enabled'
* probe was on some other instruction, but that would be a rather
* exotic way to shoot oneself in the foot.
*/
if (is_enabled) {
rp->r_rax = 1;
new_pc = rp->r_rip;
goto done;
}
/*
* We emulate certain types of instructions to ensure correctness
* (in the case of position dependent instructions) or optimize
* common cases. The rest we have the thread execute back in user-
* land.
*/
switch (tp->ftt_type) {
case FASTTRAP_T_RET:
case FASTTRAP_T_RET16:
{
uintptr_t dst = 0;
uintptr_t addr = 0;
int ret = 0;
/*
* We have to emulate _every_ facet of the behavior of a ret
* instruction including what happens if the load from %esp
* fails; in that case, we send a SIGSEGV.
*/
#ifdef __amd64
if (p->p_model == DATAMODEL_NATIVE) {
ret = dst = fasttrap_fulword((void *)rp->r_rsp);
addr = rp->r_rsp + sizeof (uintptr_t);
} else {
#endif
#ifdef __i386__
uint32_t dst32;
ret = dst32 = fasttrap_fuword32((void *)rp->r_esp);
dst = dst32;
addr = rp->r_esp + sizeof (uint32_t);
#endif
#ifdef __amd64
}
#endif
if (ret == -1) {
fasttrap_sigsegv(p, curthread, rp->r_rsp);
new_pc = pc;
break;
}
if (tp->ftt_type == FASTTRAP_T_RET16)
addr += tp->ftt_dest;
rp->r_rsp = addr;
new_pc = dst;
break;
}
case FASTTRAP_T_JCC:
{
uint_t taken = 0;
switch (tp->ftt_code) {
case FASTTRAP_JO:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_OF) != 0;
break;
case FASTTRAP_JNO:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_OF) == 0;
break;
case FASTTRAP_JB:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_CF) != 0;
break;
case FASTTRAP_JAE:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_CF) == 0;
break;
case FASTTRAP_JE:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_ZF) != 0;
break;
case FASTTRAP_JNE:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_ZF) == 0;
break;
case FASTTRAP_JBE:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_CF) != 0 ||
(rp->r_rflags & FASTTRAP_EFLAGS_ZF) != 0;
break;
case FASTTRAP_JA:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_CF) == 0 &&
(rp->r_rflags & FASTTRAP_EFLAGS_ZF) == 0;
break;
case FASTTRAP_JS:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_SF) != 0;
break;
case FASTTRAP_JNS:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_SF) == 0;
break;
case FASTTRAP_JP:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_PF) != 0;
break;
case FASTTRAP_JNP:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_PF) == 0;
break;
case FASTTRAP_JL:
taken = ((rp->r_rflags & FASTTRAP_EFLAGS_SF) == 0) !=
((rp->r_rflags & FASTTRAP_EFLAGS_OF) == 0);
break;
case FASTTRAP_JGE:
taken = ((rp->r_rflags & FASTTRAP_EFLAGS_SF) == 0) ==
((rp->r_rflags & FASTTRAP_EFLAGS_OF) == 0);
break;
case FASTTRAP_JLE:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_ZF) != 0 ||
((rp->r_rflags & FASTTRAP_EFLAGS_SF) == 0) !=
((rp->r_rflags & FASTTRAP_EFLAGS_OF) == 0);
break;
case FASTTRAP_JG:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_ZF) == 0 &&
((rp->r_rflags & FASTTRAP_EFLAGS_SF) == 0) ==
((rp->r_rflags & FASTTRAP_EFLAGS_OF) == 0);
break;
}
if (taken)
new_pc = tp->ftt_dest;
else
new_pc = pc + tp->ftt_size;
break;
}
case FASTTRAP_T_LOOP:
{
uint_t taken = 0;
#ifdef __amd64
greg_t cx = rp->r_rcx--;
#else
greg_t cx = rp->r_ecx--;
#endif
switch (tp->ftt_code) {
case FASTTRAP_LOOPNZ:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_ZF) == 0 &&
cx != 0;
break;
case FASTTRAP_LOOPZ:
taken = (rp->r_rflags & FASTTRAP_EFLAGS_ZF) != 0 &&
cx != 0;
break;
case FASTTRAP_LOOP:
taken = (cx != 0);
break;
}
if (taken)
new_pc = tp->ftt_dest;
else
new_pc = pc + tp->ftt_size;
break;
}
case FASTTRAP_T_JCXZ:
{
#ifdef __amd64
greg_t cx = rp->r_rcx;
#else
greg_t cx = rp->r_ecx;
#endif
if (cx == 0)
new_pc = tp->ftt_dest;
else
new_pc = pc + tp->ftt_size;
break;
}
case FASTTRAP_T_PUSHL_EBP:
{
int ret = 0;
uintptr_t addr = 0;
#ifdef __amd64
if (p->p_model == DATAMODEL_NATIVE) {
addr = rp->r_rsp - sizeof (uintptr_t);
ret = fasttrap_sulword((void *)addr, &rp->r_rsp);
} else {
#endif
#ifdef __i386__
addr = rp->r_rsp - sizeof (uint32_t);
ret = fasttrap_suword32((void *)addr, &rp->r_rsp);
#endif
#ifdef __amd64
}
#endif
if (ret == -1) {
fasttrap_sigsegv(p, curthread, addr);
new_pc = pc;
break;
}
rp->r_rsp = addr;
new_pc = pc + tp->ftt_size;
break;
}
case FASTTRAP_T_NOP:
new_pc = pc + tp->ftt_size;
break;
case FASTTRAP_T_JMP:
case FASTTRAP_T_CALL:
if (tp->ftt_code == 0) {
new_pc = tp->ftt_dest;
} else {
#ifdef __amd64
uintptr_t value;
#endif
uintptr_t addr = tp->ftt_dest;
if (tp->ftt_base != FASTTRAP_NOREG)
addr += fasttrap_getreg(rp, tp->ftt_base);
if (tp->ftt_index != FASTTRAP_NOREG)
addr += fasttrap_getreg(rp, tp->ftt_index) <<
tp->ftt_scale;
if (tp->ftt_code == 1) {
/*
* If there's a segment prefix for this
* instruction, we'll need to check permissions
* and bounds on the given selector, and adjust
* the address accordingly.
*/
if (tp->ftt_segment != FASTTRAP_SEG_NONE &&
fasttrap_do_seg(tp, rp, &addr) != 0) {
fasttrap_sigsegv(p, curthread, addr);
new_pc = pc;
break;
}
#ifdef __amd64
if (p->p_model == DATAMODEL_NATIVE) {
if ((value = fasttrap_fulword((void *)addr))
== -1) {
fasttrap_sigsegv(p, curthread,
addr);
new_pc = pc;
break;
}
new_pc = value;
} else {
#endif
#ifdef __i386__
uint32_t value32;
addr = (uintptr_t)(uint32_t)addr;
if ((value32 = fasttrap_fuword32((void *)addr))
== -1) {
fasttrap_sigsegv(p, curthread,
addr);
new_pc = pc;
break;
}
new_pc = value32;
#endif
}
#ifdef __amd64
} else {
new_pc = addr;
}
#endif
}
/*
* If this is a call instruction, we need to push the return
* address onto the stack. If this fails, we send the process
* a SIGSEGV and reset the pc to emulate what would happen if
* this instruction weren't traced.
*/
if (tp->ftt_type == FASTTRAP_T_CALL) {
int ret = 0;
uintptr_t addr = 0, pcps;
#ifdef __amd64
if (p->p_model == DATAMODEL_NATIVE) {
addr = rp->r_rsp - sizeof (uintptr_t);
pcps = pc + tp->ftt_size;
ret = fasttrap_sulword((void *)addr, &pcps);
} else {
#endif
#ifdef __i386__
addr = rp->r_rsp - sizeof (uint32_t);
pcps = (uint32_t)(pc + tp->ftt_size);
ret = fasttrap_suword32((void *)addr, &pcps);
#endif
#ifdef __amd64
}
#endif
if (ret == -1) {
fasttrap_sigsegv(p, curthread, addr);
new_pc = pc;
break;
}
rp->r_rsp = addr;
}
break;
case FASTTRAP_T_COMMON:
{
uintptr_t addr;
#if defined(__amd64)
uint8_t scratch[2 * FASTTRAP_MAX_INSTR_SIZE + 22];
#else
uint8_t scratch[2 * FASTTRAP_MAX_INSTR_SIZE + 7];
#endif
uint_t i = 0;
#if defined(sun)
klwp_t *lwp = ttolwp(curthread);
#endif
/*
* Compute the address of the ulwp_t and step over the
* ul_self pointer. The method used to store the user-land
* thread pointer is very different on 32- and 64-bit
* kernels.
*/
#if defined(sun)
#if defined(__amd64)
if (p->p_model == DATAMODEL_LP64) {
addr = lwp->lwp_pcb.pcb_fsbase;
addr += sizeof (void *);
} else {
addr = lwp->lwp_pcb.pcb_gsbase;
addr += sizeof (caddr32_t);
}
#else
addr = USD_GETBASE(&lwp->lwp_pcb.pcb_gsdesc);
addr += sizeof (void *);
#endif
#endif /* sun */
#ifdef __i386__
addr = USD_GETBASE(&curthread->td_pcb->pcb_gsd);
#else
addr = curthread->td_pcb->pcb_gsbase;
#endif
addr += sizeof (void *);
/*
* Generic Instruction Tracing
* ---------------------------
*
* This is the layout of the scratch space in the user-land
* thread structure for our generated instructions.
*
* 32-bit mode bytes
* ------------------------ -----
* a: <original instruction> <= 15
* jmp <pc + tp->ftt_size> 5
* b: <original instruction> <= 15
* int T_DTRACE_RET 2
* -----
* <= 37
*
* 64-bit mode bytes
* ------------------------ -----
* a: <original instruction> <= 15
* jmp 0(%rip) 6
* <pc + tp->ftt_size> 8
* b: <original instruction> <= 15
* int T_DTRACE_RET 2
* -----
* <= 46
*
* The %pc is set to a, and curthread->t_dtrace_astpc is set
* to b. If we encounter a signal on the way out of the
* kernel, trap() will set %pc to curthread->t_dtrace_astpc
* so that we execute the original instruction and re-enter
* the kernel rather than redirecting to the next instruction.
*
* If there are return probes (so we know that we're going to
* need to reenter the kernel after executing the original
* instruction), the scratch space will just contain the
* original instruction followed by an interrupt -- the same
* data as at b.
*
* %rip-relative Addressing
* ------------------------
*
* There's a further complication in 64-bit mode due to %rip-
* relative addressing. While this is clearly a beneficial
* architectural decision for position independent code, it's
* hard not to see it as a personal attack against the pid
* provider since before there was a relatively small set of
* instructions to emulate; with %rip-relative addressing,
* almost every instruction can potentially depend on the
* address at which it's executed. Rather than emulating
* the broad spectrum of instructions that can now be
* position dependent, we emulate jumps and others as in
* 32-bit mode, and take a different tack for instructions
* using %rip-relative addressing.
*
* For every instruction that uses the ModRM byte, the
* in-kernel disassembler reports its location. We use the
* ModRM byte to identify that an instruction uses
* %rip-relative addressing and to see what other registers
* the instruction uses. To emulate those instructions,
* we modify the instruction to be %rax-relative rather than
* %rip-relative (or %rcx-relative if the instruction uses
* %rax; or %r8- or %r9-relative if the REX.B is present so
* we don't have to rewrite the REX prefix). We then load
* the value that %rip would have been into the scratch
* register and generate an instruction to reset the scratch
* register back to its original value. The instruction
* sequence looks like this:
*
* 64-mode %rip-relative bytes
* ------------------------ -----
* a: <modified instruction> <= 15
* movq $<value>, %<scratch> 6
* jmp 0(%rip) 6
* <pc + tp->ftt_size> 8
* b: <modified instruction> <= 15
* int T_DTRACE_RET 2
* -----
* 52
*
* We set curthread->t_dtrace_regv so that upon receiving
* a signal we can reset the value of the scratch register.
*/
ASSERT(tp->ftt_size < FASTTRAP_MAX_INSTR_SIZE);
curthread->t_dtrace_scrpc = addr;
bcopy(tp->ftt_instr, &scratch[i], tp->ftt_size);
i += tp->ftt_size;
#ifdef __amd64
if (tp->ftt_ripmode != 0) {
greg_t *reg = NULL;
ASSERT(p->p_model == DATAMODEL_LP64);
ASSERT(tp->ftt_ripmode &
(FASTTRAP_RIP_1 | FASTTRAP_RIP_2));
/*
* If this was a %rip-relative instruction, we change
* it to be either a %rax- or %rcx-relative
* instruction (depending on whether those registers
* are used as another operand; or %r8- or %r9-
* relative depending on the value of REX.B). We then
* set that register and generate a movq instruction
* to reset the value.
*/
if (tp->ftt_ripmode & FASTTRAP_RIP_X)
scratch[i++] = FASTTRAP_REX(1, 0, 0, 1);
else
scratch[i++] = FASTTRAP_REX(1, 0, 0, 0);
if (tp->ftt_ripmode & FASTTRAP_RIP_1)
scratch[i++] = FASTTRAP_MOV_EAX;
else
scratch[i++] = FASTTRAP_MOV_ECX;
switch (tp->ftt_ripmode) {
case FASTTRAP_RIP_1:
reg = &rp->r_rax;
curthread->t_dtrace_reg = REG_RAX;
break;
case FASTTRAP_RIP_2:
reg = &rp->r_rcx;
curthread->t_dtrace_reg = REG_RCX;
break;
case FASTTRAP_RIP_1 | FASTTRAP_RIP_X:
reg = &rp->r_r8;
curthread->t_dtrace_reg = REG_R8;
break;
case FASTTRAP_RIP_2 | FASTTRAP_RIP_X:
reg = &rp->r_r9;
curthread->t_dtrace_reg = REG_R9;
break;
}
/* LINTED - alignment */
*(uint64_t *)&scratch[i] = *reg;
curthread->t_dtrace_regv = *reg;
*reg = pc + tp->ftt_size;
i += sizeof (uint64_t);
}
#endif
/*
* Generate the branch instruction to what would have
* normally been the subsequent instruction. In 32-bit mode,
* this is just a relative branch; in 64-bit mode this is a
* %rip-relative branch that loads the 64-bit pc value
* immediately after the jmp instruction.
*/
#ifdef __amd64
if (p->p_model == DATAMODEL_LP64) {
scratch[i++] = FASTTRAP_GROUP5_OP;
scratch[i++] = FASTTRAP_MODRM(0, 4, 5);
/* LINTED - alignment */
*(uint32_t *)&scratch[i] = 0;
i += sizeof (uint32_t);
/* LINTED - alignment */
*(uint64_t *)&scratch[i] = pc + tp->ftt_size;
i += sizeof (uint64_t);
} else {
#endif
#ifdef __i386__
/*
* Set up the jmp to the next instruction; note that
* the size of the traced instruction cancels out.
*/
scratch[i++] = FASTTRAP_JMP32;
/* LINTED - alignment */
*(uint32_t *)&scratch[i] = pc - addr - 5;
i += sizeof (uint32_t);
#endif
#ifdef __amd64
}
#endif
curthread->t_dtrace_astpc = addr + i;
bcopy(tp->ftt_instr, &scratch[i], tp->ftt_size);
i += tp->ftt_size;
scratch[i++] = FASTTRAP_INT;
scratch[i++] = T_DTRACE_RET;
ASSERT(i <= sizeof (scratch));
#if defined(sun)
if (fasttrap_copyout(scratch, (char *)addr, i)) {
#else
if (uwrite(curproc, scratch, i, addr)) {
#endif
fasttrap_sigtrap(p, curthread, pc);
new_pc = pc;
break;
}
if (tp->ftt_retids != NULL) {
curthread->t_dtrace_step = 1;
curthread->t_dtrace_ret = 1;
new_pc = curthread->t_dtrace_astpc;
} else {
new_pc = curthread->t_dtrace_scrpc;
}
curthread->t_dtrace_pc = pc;
curthread->t_dtrace_npc = pc + tp->ftt_size;
curthread->t_dtrace_on = 1;
break;
}
default:
panic("fasttrap: mishandled an instruction");
}
done:
/*
* If there were no return probes when we first found the tracepoint,
* we should feel no obligation to honor any return probes that were
* subsequently enabled -- they'll just have to wait until the next
* time around.
*/
if (tp->ftt_retids != NULL) {
/*
* We need to wait until the results of the instruction are
* apparent before invoking any return probes. If this
* instruction was emulated we can just call
* fasttrap_return_common(); if it needs to be executed, we
* need to wait until the user thread returns to the kernel.
*/
if (tp->ftt_type != FASTTRAP_T_COMMON) {
/*
* Set the program counter to the address of the traced
* instruction so that it looks right in ustack()
* output. We had previously set it to the end of the
* instruction to simplify %rip-relative addressing.
*/
rp->r_rip = pc;
fasttrap_return_common(rp, pc, pid, new_pc);
} else {
ASSERT(curthread->t_dtrace_ret != 0);
ASSERT(curthread->t_dtrace_pc == pc);
ASSERT(curthread->t_dtrace_scrpc != 0);
ASSERT(new_pc == curthread->t_dtrace_astpc);
}
}
rp->r_rip = new_pc;
PROC_LOCK(p);
proc_write_regs(curthread, rp);
_PRELE(p);
PROC_UNLOCK(p);
return (0);
}
int
fasttrap_return_probe(struct reg *rp)
{
proc_t *p = curproc;
uintptr_t pc = curthread->t_dtrace_pc;
uintptr_t npc = curthread->t_dtrace_npc;
curthread->t_dtrace_pc = 0;
curthread->t_dtrace_npc = 0;
curthread->t_dtrace_scrpc = 0;
curthread->t_dtrace_astpc = 0;
#if defined(sun)
/*
* Treat a child created by a call to vfork(2) as if it were its
* parent. We know that there's only one thread of control in such a
* process: this one.
*/
while (p->p_flag & SVFORK) {
p = p->p_parent;
}
#endif
/*
* We set rp->r_rip to the address of the traced instruction so
* that it appears to dtrace_probe() that we're on the original
* instruction, and so that the user can't easily detect our
* complex web of lies. dtrace_return_probe() (our caller)
* will correctly set %pc after we return.
*/
rp->r_rip = pc;
fasttrap_return_common(rp, pc, p->p_pid, npc);
return (0);
}
int
linux_ptrace(struct thread *td, struct linux_ptrace_args *uap)
{
union {
struct linux_pt_reg reg;
struct linux_pt_fpreg fpreg;
struct linux_pt_fpxreg fpxreg;
} r;
union {
struct reg bsd_reg;
struct fpreg bsd_fpreg;
struct dbreg bsd_dbreg;
} u;
void *addr;
pid_t pid;
int error, req;
error = 0;
/* by default, just copy data intact */
req = uap->req;
pid = (pid_t)uap->pid;
addr = (void *)uap->addr;
switch (req) {
case PTRACE_TRACEME:
case PTRACE_POKETEXT:
case PTRACE_POKEDATA:
case PTRACE_KILL:
error = kern_ptrace(td, req, pid, addr, uap->data);
break;
case PTRACE_PEEKTEXT:
case PTRACE_PEEKDATA: {
/* need to preserve return value */
int rval = td->td_retval[0];
error = kern_ptrace(td, req, pid, addr, 0);
if (error == 0)
error = copyout(td->td_retval, (void *)uap->data,
sizeof(l_int));
td->td_retval[0] = rval;
break;
}
case PTRACE_DETACH:
error = kern_ptrace(td, PT_DETACH, pid, (void *)1,
map_signum(uap->data));
break;
case PTRACE_SINGLESTEP:
case PTRACE_CONT:
error = kern_ptrace(td, req, pid, (void *)1,
map_signum(uap->data));
break;
case PTRACE_ATTACH:
error = kern_ptrace(td, PT_ATTACH, pid, addr, uap->data);
break;
case PTRACE_GETREGS:
/* Linux is using data where FreeBSD is using addr */
error = kern_ptrace(td, PT_GETREGS, pid, &u.bsd_reg, 0);
if (error == 0) {
map_regs_to_linux(&u.bsd_reg, &r.reg);
error = copyout(&r.reg, (void *)uap->data,
sizeof(r.reg));
}
break;
case PTRACE_SETREGS:
/* Linux is using data where FreeBSD is using addr */
error = copyin((void *)uap->data, &r.reg, sizeof(r.reg));
if (error == 0) {
map_regs_from_linux(&u.bsd_reg, &r.reg);
error = kern_ptrace(td, PT_SETREGS, pid, &u.bsd_reg, 0);
}
break;
case PTRACE_GETFPREGS:
/* Linux is using data where FreeBSD is using addr */
error = kern_ptrace(td, PT_GETFPREGS, pid, &u.bsd_fpreg, 0);
if (error == 0) {
map_fpregs_to_linux(&u.bsd_fpreg, &r.fpreg);
error = copyout(&r.fpreg, (void *)uap->data,
sizeof(r.fpreg));
}
break;
case PTRACE_SETFPREGS:
/* Linux is using data where FreeBSD is using addr */
error = copyin((void *)uap->data, &r.fpreg, sizeof(r.fpreg));
if (error == 0) {
map_fpregs_from_linux(&u.bsd_fpreg, &r.fpreg);
error = kern_ptrace(td, PT_SETFPREGS, pid,
&u.bsd_fpreg, 0);
}
break;
case PTRACE_SETFPXREGS:
#ifdef CPU_ENABLE_SSE
error = copyin((void *)uap->data, &r.fpxreg, sizeof(r.fpxreg));
if (error)
break;
#endif
/* FALL THROUGH */
case PTRACE_GETFPXREGS: {
#ifdef CPU_ENABLE_SSE
struct proc *p;
struct thread *td2;
if (sizeof(struct linux_pt_fpxreg) != sizeof(struct savexmm)) {
static int once = 0;
if (!once) {
printf("linux: savexmm != linux_pt_fpxreg\n");
once = 1;
}
error = EIO;
break;
}
if ((p = pfind(uap->pid)) == NULL) {
error = ESRCH;
break;
}
/* Exiting processes can't be debugged. */
if ((p->p_flag & P_WEXIT) != 0) {
error = ESRCH;
goto fail;
}
if ((error = p_candebug(td, p)) != 0)
goto fail;
/* System processes can't be debugged. */
if ((p->p_flag & P_SYSTEM) != 0) {
error = EINVAL;
goto fail;
}
/* not being traced... */
if ((p->p_flag & P_TRACED) == 0) {
error = EPERM;
goto fail;
}
/* not being traced by YOU */
if (p->p_pptr != td->td_proc) {
error = EBUSY;
goto fail;
}
/* not currently stopped */
if (!P_SHOULDSTOP(p) || (p->p_flag & P_WAITED) == 0) {
error = EBUSY;
goto fail;
}
if (req == PTRACE_GETFPXREGS) {
_PHOLD(p); /* may block */
td2 = FIRST_THREAD_IN_PROC(p);
error = linux_proc_read_fpxregs(td2, &r.fpxreg);
_PRELE(p);
PROC_UNLOCK(p);
if (error == 0)
error = copyout(&r.fpxreg, (void *)uap->data,
sizeof(r.fpxreg));
} else {
/* clear dangerous bits exactly as Linux does*/
r.fpxreg.mxcsr &= 0xffbf;
_PHOLD(p); /* may block */
td2 = FIRST_THREAD_IN_PROC(p);
error = linux_proc_write_fpxregs(td2, &r.fpxreg);
_PRELE(p);
PROC_UNLOCK(p);
}
break;
fail:
PROC_UNLOCK(p);
#else
error = EIO;
#endif
break;
}
case PTRACE_PEEKUSR:
case PTRACE_POKEUSR: {
error = EIO;
/* check addr for alignment */
if (uap->addr < 0 || uap->addr & (sizeof(l_int) - 1))
break;
/*
* Allow linux programs to access register values in
* user struct. We simulate this through PT_GET/SETREGS
* as necessary.
*/
if (uap->addr < sizeof(struct linux_pt_reg)) {
error = kern_ptrace(td, PT_GETREGS, pid, &u.bsd_reg, 0);
if (error != 0)
break;
map_regs_to_linux(&u.bsd_reg, &r.reg);
if (req == PTRACE_PEEKUSR) {
error = copyout((char *)&r.reg + uap->addr,
(void *)uap->data, sizeof(l_int));
break;
}
*(l_int *)((char *)&r.reg + uap->addr) =
(l_int)uap->data;
map_regs_from_linux(&u.bsd_reg, &r.reg);
error = kern_ptrace(td, PT_SETREGS, pid, &u.bsd_reg, 0);
}
/*
* Simulate debug registers access
*/
if (uap->addr >= LINUX_DBREG_OFFSET &&
uap->addr <= LINUX_DBREG_OFFSET + LINUX_DBREG_SIZE) {
error = kern_ptrace(td, PT_GETDBREGS, pid, &u.bsd_dbreg,
0);
if (error != 0)
break;
uap->addr -= LINUX_DBREG_OFFSET;
if (req == PTRACE_PEEKUSR) {
error = copyout((char *)&u.bsd_dbreg +
uap->addr, (void *)uap->data,
sizeof(l_int));
break;
}
*(l_int *)((char *)&u.bsd_dbreg + uap->addr) =
uap->data;
error = kern_ptrace(td, PT_SETDBREGS, pid,
&u.bsd_dbreg, 0);
}
break;
}
case PTRACE_SYSCALL:
/* fall through */
default:
printf("linux: ptrace(%u, ...) not implemented\n",
(unsigned int)uap->req);
error = EINVAL;
break;
}
return (error);
}
/*
* Read from a file
*/
static int
pfs_read(struct vop_read_args *va)
{
struct vnode *vn = va->a_vp;
struct pfs_vdata *pvd = vn->v_data;
struct pfs_node *pn = pvd->pvd_pn;
struct uio *uio = va->a_uio;
struct proc *proc;
struct sbuf *sb = NULL;
int error, locked;
unsigned int buflen, offset, resid;
PFS_TRACE(("%s", pn->pn_name));
pfs_assert_not_owned(pn);
if (vn->v_type != VREG)
PFS_RETURN (EINVAL);
KASSERT_PN_IS_FILE(pn);
if (!(pn->pn_flags & PFS_RD))
PFS_RETURN (EBADF);
if (pn->pn_fill == NULL)
PFS_RETURN (EIO);
/*
* This is necessary because either process' privileges may
* have changed since the open() call.
*/
if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
PFS_RETURN (EIO);
if (proc != NULL) {
_PHOLD(proc);
PROC_UNLOCK(proc);
}
vhold(vn);
locked = VOP_ISLOCKED(vn, curthread);
VOP_UNLOCK(vn, 0, curthread);
if (pn->pn_flags & PFS_RAWRD) {
PFS_TRACE(("%lu resid", (unsigned long)uio->uio_resid));
error = pn_fill(curthread, proc, pn, NULL, uio);
PFS_TRACE(("%lu resid", (unsigned long)uio->uio_resid));
goto ret;
}
/* beaucoup sanity checks so we don't ask for bogus allocation */
if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
(offset = uio->uio_offset) != uio->uio_offset ||
(resid = uio->uio_resid) != uio->uio_resid ||
(buflen = offset + resid + 1) < offset || buflen > INT_MAX) {
if (proc != NULL)
PRELE(proc);
error = EINVAL;
goto ret;
}
if (buflen > MAXPHYS + 1) {
error = EIO;
goto ret;
}
sb = sbuf_new(sb, NULL, buflen, 0);
if (sb == NULL) {
error = EIO;
goto ret;
}
error = pn_fill(curthread, proc, pn, sb, uio);
if (error) {
sbuf_delete(sb);
goto ret;
}
sbuf_finish(sb);
error = uiomove_frombuf(sbuf_data(sb), sbuf_len(sb), uio);
sbuf_delete(sb);
ret:
vn_lock(vn, locked | LK_RETRY, curthread);
vdrop(vn);
if (proc != NULL)
PRELE(proc);
PFS_RETURN (error);
}
static void
do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
struct vmspace *vm2, struct file *fp_procdesc)
{
struct proc *p1, *pptr;
int trypid;
struct filedesc *fd;
struct filedesc_to_leader *fdtol;
struct sigacts *newsigacts;
sx_assert(&proctree_lock, SX_SLOCKED);
sx_assert(&allproc_lock, SX_XLOCKED);
p1 = td->td_proc;
trypid = fork_findpid(fr->fr_flags);
sx_sunlock(&proctree_lock);
p2->p_state = PRS_NEW; /* protect against others */
p2->p_pid = trypid;
AUDIT_ARG_PID(p2->p_pid);
LIST_INSERT_HEAD(&allproc, p2, p_list);
allproc_gen++;
LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
tidhash_add(td2);
PROC_LOCK(p2);
PROC_LOCK(p1);
sx_xunlock(&allproc_lock);
bcopy(&p1->p_startcopy, &p2->p_startcopy,
__rangeof(struct proc, p_startcopy, p_endcopy));
pargs_hold(p2->p_args);
PROC_UNLOCK(p1);
bzero(&p2->p_startzero,
__rangeof(struct proc, p_startzero, p_endzero));
/* Tell the prison that we exist. */
prison_proc_hold(p2->p_ucred->cr_prison);
PROC_UNLOCK(p2);
/*
* Malloc things while we don't hold any locks.
*/
if (fr->fr_flags & RFSIGSHARE)
newsigacts = NULL;
else
newsigacts = sigacts_alloc();
/*
* Copy filedesc.
*/
if (fr->fr_flags & RFCFDG) {
fd = fdinit(p1->p_fd, false);
fdtol = NULL;
} else if (fr->fr_flags & RFFDG) {
fd = fdcopy(p1->p_fd);
fdtol = NULL;
} else {
fd = fdshare(p1->p_fd);
if (p1->p_fdtol == NULL)
p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
p1->p_leader);
if ((fr->fr_flags & RFTHREAD) != 0) {
/*
* Shared file descriptor table, and shared
* process leaders.
*/
fdtol = p1->p_fdtol;
FILEDESC_XLOCK(p1->p_fd);
fdtol->fdl_refcount++;
FILEDESC_XUNLOCK(p1->p_fd);
} else {
/*
* Shared file descriptor table, and different
* process leaders.
*/
fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
p1->p_fd, p2);
}
}
/*
* Make a proc table entry for the new process.
* Start by zeroing the section of proc that is zero-initialized,
* then copy the section that is copied directly from the parent.
*/
PROC_LOCK(p2);
PROC_LOCK(p1);
bzero(&td2->td_startzero,
__rangeof(struct thread, td_startzero, td_endzero));
bcopy(&td->td_startcopy, &td2->td_startcopy,
__rangeof(struct thread, td_startcopy, td_endcopy));
bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
td2->td_sigstk = td->td_sigstk;
td2->td_flags = TDF_INMEM;
td2->td_lend_user_pri = PRI_MAX;
#ifdef VIMAGE
td2->td_vnet = NULL;
td2->td_vnet_lpush = NULL;
#endif
/*
* Allow the scheduler to initialize the child.
*/
thread_lock(td);
sched_fork(td, td2);
thread_unlock(td);
/*
* Duplicate sub-structures as needed.
* Increase reference counts on shared objects.
*/
p2->p_flag = P_INMEM;
p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP);
p2->p_swtick = ticks;
if (p1->p_flag & P_PROFIL)
startprofclock(p2);
/*
* Whilst the proc lock is held, copy the VM domain data out
* using the VM domain method.
*/
vm_domain_policy_init(&p2->p_vm_dom_policy);
vm_domain_policy_localcopy(&p2->p_vm_dom_policy,
&p1->p_vm_dom_policy);
if (fr->fr_flags & RFSIGSHARE) {
p2->p_sigacts = sigacts_hold(p1->p_sigacts);
} else {
sigacts_copy(newsigacts, p1->p_sigacts);
p2->p_sigacts = newsigacts;
}
if (fr->fr_flags & RFTSIGZMB)
p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
else if (fr->fr_flags & RFLINUXTHPN)
p2->p_sigparent = SIGUSR1;
else
p2->p_sigparent = SIGCHLD;
p2->p_textvp = p1->p_textvp;
p2->p_fd = fd;
p2->p_fdtol = fdtol;
if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
p2->p_flag |= P_PROTECTED;
p2->p_flag2 |= P2_INHERIT_PROTECTED;
}
/*
* p_limit is copy-on-write. Bump its refcount.
*/
lim_fork(p1, p2);
thread_cow_get_proc(td2, p2);
pstats_fork(p1->p_stats, p2->p_stats);
PROC_UNLOCK(p1);
PROC_UNLOCK(p2);
/* Bump references to the text vnode (for procfs). */
if (p2->p_textvp)
vrefact(p2->p_textvp);
/*
* Set up linkage for kernel based threading.
*/
if ((fr->fr_flags & RFTHREAD) != 0) {
mtx_lock(&ppeers_lock);
p2->p_peers = p1->p_peers;
p1->p_peers = p2;
p2->p_leader = p1->p_leader;
mtx_unlock(&ppeers_lock);
PROC_LOCK(p1->p_leader);
if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
PROC_UNLOCK(p1->p_leader);
/*
* The task leader is exiting, so process p1 is
* going to be killed shortly. Since p1 obviously
* isn't dead yet, we know that the leader is either
* sending SIGKILL's to all the processes in this
* task or is sleeping waiting for all the peers to
* exit. We let p1 complete the fork, but we need
* to go ahead and kill the new process p2 since
* the task leader may not get a chance to send
* SIGKILL to it. We leave it on the list so that
* the task leader will wait for this new process
* to commit suicide.
*/
PROC_LOCK(p2);
kern_psignal(p2, SIGKILL);
PROC_UNLOCK(p2);
} else
PROC_UNLOCK(p1->p_leader);
} else {
p2->p_peers = NULL;
p2->p_leader = p2;
}
sx_xlock(&proctree_lock);
PGRP_LOCK(p1->p_pgrp);
PROC_LOCK(p2);
PROC_LOCK(p1);
/*
* Preserve some more flags in subprocess. P_PROFIL has already
* been preserved.
*/
p2->p_flag |= p1->p_flag & P_SUGID;
td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING;
SESS_LOCK(p1->p_session);
if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
p2->p_flag |= P_CONTROLT;
SESS_UNLOCK(p1->p_session);
if (fr->fr_flags & RFPPWAIT)
p2->p_flag |= P_PPWAIT;
p2->p_pgrp = p1->p_pgrp;
LIST_INSERT_AFTER(p1, p2, p_pglist);
PGRP_UNLOCK(p1->p_pgrp);
LIST_INIT(&p2->p_children);
LIST_INIT(&p2->p_orphans);
callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
/*
* If PF_FORK is set, the child process inherits the
* procfs ioctl flags from its parent.
*/
if (p1->p_pfsflags & PF_FORK) {
p2->p_stops = p1->p_stops;
p2->p_pfsflags = p1->p_pfsflags;
}
/*
* This begins the section where we must prevent the parent
* from being swapped.
*/
_PHOLD(p1);
PROC_UNLOCK(p1);
/*
* Attach the new process to its parent.
*
* If RFNOWAIT is set, the newly created process becomes a child
* of init. This effectively disassociates the child from the
* parent.
*/
if ((fr->fr_flags & RFNOWAIT) != 0) {
pptr = p1->p_reaper;
p2->p_reaper = pptr;
} else {
p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
p1 : p1->p_reaper;
pptr = p1;
}
p2->p_pptr = pptr;
LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
LIST_INIT(&p2->p_reaplist);
LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
if (p2->p_reaper == p1)
p2->p_reapsubtree = p2->p_pid;
sx_xunlock(&proctree_lock);
/* Inform accounting that we have forked. */
p2->p_acflag = AFORK;
PROC_UNLOCK(p2);
#ifdef KTRACE
ktrprocfork(p1, p2);
#endif
/*
* Finish creating the child process. It will return via a different
* execution path later. (ie: directly into user mode)
*/
vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
if (fr->fr_flags == (RFFDG | RFPROC)) {
VM_CNT_INC(v_forks);
VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
p2->p_vmspace->vm_ssize);
} else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
VM_CNT_INC(v_vforks);
VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
p2->p_vmspace->vm_ssize);
} else if (p1 == &proc0) {
VM_CNT_INC(v_kthreads);
VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
p2->p_vmspace->vm_ssize);
} else {
VM_CNT_INC(v_rforks);
VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
p2->p_vmspace->vm_ssize);
}
/*
* Associate the process descriptor with the process before anything
* can happen that might cause that process to need the descriptor.
* However, don't do this until after fork(2) can no longer fail.
*/
if (fr->fr_flags & RFPROCDESC)
procdesc_new(p2, fr->fr_pd_flags);
/*
* Both processes are set up, now check if any loadable modules want
* to adjust anything.
*/
EVENTHANDLER_INVOKE(process_fork, p1, p2, fr->fr_flags);
/*
* Set the child start time and mark the process as being complete.
*/
PROC_LOCK(p2);
PROC_LOCK(p1);
microuptime(&p2->p_stats->p_start);
PROC_SLOCK(p2);
p2->p_state = PRS_NORMAL;
PROC_SUNLOCK(p2);
#ifdef KDTRACE_HOOKS
/*
* Tell the DTrace fasttrap provider about the new process so that any
* tracepoints inherited from the parent can be removed. We have to do
* this only after p_state is PRS_NORMAL since the fasttrap module will
* use pfind() later on.
*/
if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
dtrace_fasttrap_fork(p1, p2);
#endif
/*
* Hold the process so that it cannot exit after we make it runnable,
* but before we wait for the debugger.
*/
_PHOLD(p2);
if (p1->p_ptevents & PTRACE_FORK) {
/*
* Arrange for debugger to receive the fork event.
*
* We can report PL_FLAG_FORKED regardless of
* P_FOLLOWFORK settings, but it does not make a sense
* for runaway child.
*/
td->td_dbgflags |= TDB_FORK;
td->td_dbg_forked = p2->p_pid;
td2->td_dbgflags |= TDB_STOPATFORK;
}
if (fr->fr_flags & RFPPWAIT) {
td->td_pflags |= TDP_RFPPWAIT;
td->td_rfppwait_p = p2;
td->td_dbgflags |= TDB_VFORK;
}
PROC_UNLOCK(p2);
/*
* Now can be swapped.
*/
_PRELE(p1);
PROC_UNLOCK(p1);
/*
* Tell any interested parties about the new process.
*/
knote_fork(p1->p_klist, p2->p_pid);
SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
if (fr->fr_flags & RFPROCDESC) {
procdesc_finit(p2->p_procdesc, fp_procdesc);
fdrop(fp_procdesc, td);
}
if ((fr->fr_flags & RFSTOPPED) == 0) {
/*
* If RFSTOPPED not requested, make child runnable and
* add to run queue.
*/
thread_lock(td2);
TD_SET_CAN_RUN(td2);
sched_add(td2, SRQ_BORING);
thread_unlock(td2);
if (fr->fr_pidp != NULL)
*fr->fr_pidp = p2->p_pid;
} else {
*fr->fr_procp = p2;
}
PROC_LOCK(p2);
/*
* Wait until debugger is attached to child.
*/
while (td2->td_proc == p2 && (td2->td_dbgflags & TDB_STOPATFORK) != 0)
cv_wait(&p2->p_dbgwait, &p2->p_mtx);
_PRELE(p2);
racct_proc_fork_done(p2);
PROC_UNLOCK(p2);
}
/*
* Process ioctls
*/
int
procfs_ioctl(PFS_IOCTL_ARGS)
{
struct procfs_status *ps;
#ifdef COMPAT_FREEBSD32
struct procfs_status32 *ps32;
#endif
int error, flags, sig;
#ifdef COMPAT_FREEBSD6
int ival;
#endif
KASSERT(p != NULL,
("%s() called without a process", __func__));
PROC_LOCK_ASSERT(p, MA_OWNED);
error = 0;
switch (cmd) {
#if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
case _IOC(IOC_IN, 'p', 1, 0):
#endif
#ifdef COMPAT_FREEBSD6
case _IO('p', 1):
ival = IOCPARM_IVAL(data);
data = &ival;
#endif
case PIOCBIS:
p->p_stops |= *(unsigned int *)data;
break;
#if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
case _IOC(IOC_IN, 'p', 2, 0):
#endif
#ifdef COMPAT_FREEBSD6
case _IO('p', 2):
ival = IOCPARM_IVAL(data);
data = &ival;
#endif
case PIOCBIC:
p->p_stops &= ~*(unsigned int *)data;
break;
#if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
case _IOC(IOC_IN, 'p', 3, 0):
#endif
#ifdef COMPAT_FREEBSD6
case _IO('p', 3):
ival = IOCPARM_IVAL(data);
data = &ival;
#endif
case PIOCSFL:
flags = *(unsigned int *)data;
if (flags & PF_ISUGID) {
/*
* XXXRW: Is this specific check required here, as
* p_candebug() should implement it, or other checks
* are missing.
*/
error = priv_check(td, PRIV_DEBUG_SUGID);
if (error)
break;
}
p->p_pfsflags = flags;
break;
case PIOCGFL:
*(unsigned int *)data = p->p_pfsflags;
break;
case PIOCWAIT:
while (p->p_step == 0 && (p->p_flag & P_WEXIT) == 0) {
/* sleep until p stops */
_PHOLD(p);
error = msleep(&p->p_stype, &p->p_mtx,
PWAIT|PCATCH, "pioctl", 0);
_PRELE(p);
if (error != 0)
break;
}
/* fall through to PIOCSTATUS */
case PIOCSTATUS:
ps = (struct procfs_status *)data;
ps->state = (p->p_step == 0);
ps->flags = 0; /* nope */
ps->events = p->p_stops;
ps->why = p->p_step ? p->p_stype : 0;
ps->val = p->p_step ? p->p_xstat : 0;
break;
#ifdef COMPAT_FREEBSD32
case PIOCWAIT32:
while (p->p_step == 0 && (p->p_flag & P_WEXIT) == 0) {
/* sleep until p stops */
_PHOLD(p);
error = msleep(&p->p_stype, &p->p_mtx,
PWAIT|PCATCH, "pioctl", 0);
_PRELE(p);
if (error != 0)
break;
}
/* fall through to PIOCSTATUS32 */
case PIOCSTATUS32:
ps32 = (struct procfs_status32 *)data;
ps32->state = (p->p_step == 0);
ps32->flags = 0; /* nope */
ps32->events = p->p_stops;
ps32->why = p->p_step ? p->p_stype : 0;
ps32->val = p->p_step ? p->p_xstat : 0;
break;
#endif
#if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
case _IOC(IOC_IN, 'p', 5, 0):
#endif
#ifdef COMPAT_FREEBSD6
case _IO('p', 5):
ival = IOCPARM_IVAL(data);
data = &ival;
#endif
case PIOCCONT:
if (p->p_step == 0)
break;
sig = *(unsigned int *)data;
if (sig != 0 && !_SIG_VALID(sig)) {
error = EINVAL;
break;
}
#if 0
p->p_step = 0;
if (P_SHOULDSTOP(p)) {
p->p_xstat = sig;
p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG);
PROC_SLOCK(p);
thread_unsuspend(p);
PROC_SUNLOCK(p);
} else if (sig)
kern_psignal(p, sig);
#else
if (sig)
kern_psignal(p, sig);
p->p_step = 0;
wakeup(&p->p_step);
#endif
break;
default:
error = (ENOTTY);
}
return (error);
}
void
undefinedinstruction(struct trapframe *frame)
{
struct thread *td;
u_int fault_pc;
int fault_instruction;
int fault_code;
int coprocessor;
struct undefined_handler *uh;
int error;
#ifdef VERBOSE_ARM32
int s;
#endif
ksiginfo_t ksi;
/* Enable interrupts if they were enabled before the exception. */
if (__predict_true(frame->tf_spsr & PSR_I) == 0)
enable_interrupts(PSR_I);
if (__predict_true(frame->tf_spsr & PSR_F) == 0)
enable_interrupts(PSR_F);
VM_CNT_INC(v_trap);
fault_pc = frame->tf_pc;
/*
* Get the current thread/proc structure or thread0/proc0 if there is
* none.
*/
td = curthread == NULL ? &thread0 : curthread;
coprocessor = 0;
if ((frame->tf_spsr & PSR_T) == 0) {
/*
* Make sure the program counter is correctly aligned so we
* don't take an alignment fault trying to read the opcode.
*/
if (__predict_false((fault_pc & 3) != 0)) {
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = SIGILL;
ksi.ksi_code = ILL_ILLADR;
ksi.ksi_addr = (u_int32_t *)(intptr_t) fault_pc;
trapsignal(td, &ksi);
userret(td, frame);
return;
}
/*
* Should use fuword() here .. but in the interests of
* squeezing every bit of speed we will just use ReadWord().
* We know the instruction can be read as was just executed
* so this will never fail unless the kernel is screwed up
* in which case it does not really matter does it ?
*/
fault_instruction = *(u_int32_t *)fault_pc;
/* Check for coprocessor instruction */
/*
* According to the datasheets you only need to look at bit
* 27 of the instruction to tell the difference between and
* undefined instruction and a coprocessor instruction
* following an undefined instruction trap.
*/
if (ARM_COPROC_INSN(fault_instruction))
coprocessor = ARM_COPROC(fault_instruction);
else { /* check for special instructions */
if (ARM_VFP_INSN(fault_instruction))
coprocessor = COPROC_VFP; /* vfp / simd */
}
} else {
#if __ARM_ARCH >= 7
fault_instruction = *(uint16_t *)fault_pc;
if (THUMB_32BIT_INSN(fault_instruction)) {
fault_instruction <<= 16;
fault_instruction |= *(uint16_t *)(fault_pc + 2);
/*
* Is it a Coprocessor, Advanced SIMD, or
* Floating-point instruction.
*/
if (THUMB_COPROC_INSN(fault_instruction)) {
if (THUMB_COPROC_UNDEFINED(fault_instruction)) {
/* undefined insn */
} else if (THUMB_VFP_INSN(fault_instruction))
coprocessor = COPROC_VFP;
else
coprocessor =
THUMB_COPROC(fault_instruction);
}
}
#else
/*
* No support for Thumb-2 on this cpu
*/
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = SIGILL;
ksi.ksi_code = ILL_ILLADR;
ksi.ksi_addr = (u_int32_t *)(intptr_t) fault_pc;
trapsignal(td, &ksi);
userret(td, frame);
return;
#endif
}
if ((frame->tf_spsr & PSR_MODE) == PSR_USR32_MODE) {
/*
* Modify the fault_code to reflect the USR/SVC state at
* time of fault.
*/
fault_code = FAULT_USER;
td->td_frame = frame;
} else
fault_code = 0;
/* OK this is were we do something about the instruction. */
LIST_FOREACH(uh, &undefined_handlers[coprocessor], uh_link)
if (uh->uh_handler(fault_pc, fault_instruction, frame,
fault_code) == 0)
break;
if (fault_code & FAULT_USER) {
/* TODO: No support for ptrace from Thumb-2 */
if ((frame->tf_spsr & PSR_T) == 0 &&
fault_instruction == PTRACE_BREAKPOINT) {
PROC_LOCK(td->td_proc);
_PHOLD(td->td_proc);
error = ptrace_clear_single_step(td);
_PRELE(td->td_proc);
PROC_UNLOCK(td->td_proc);
if (error != 0) {
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = SIGILL;
ksi.ksi_code = ILL_ILLOPC;
ksi.ksi_addr = (u_int32_t *)(intptr_t) fault_pc;
trapsignal(td, &ksi);
}
return;
}
}
if (uh == NULL && (fault_code & FAULT_USER)) {
/* Fault has not been handled */
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = SIGILL;
ksi.ksi_code = ILL_ILLOPC;
ksi.ksi_addr = (u_int32_t *)(intptr_t) fault_pc;
trapsignal(td, &ksi);
}
if ((fault_code & FAULT_USER) == 0) {
if (fault_instruction == KERNEL_BREAKPOINT) {
#ifdef KDB
kdb_trap(T_BREAKPOINT, 0, frame);
#else
printf("No debugger in kernel.\n");
#endif
return;
}
else
panic("Undefined instruction in kernel.\n");
}
userret(td, frame);
}
int
kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
{
struct iovec iov;
struct uio uio;
struct proc *curp, *p, *pp;
struct thread *td2;
struct ptrace_io_desc *piod;
int error, write, tmp;
int proctree_locked = 0;
curp = td->td_proc;
/* Lock proctree before locking the process. */
switch (req) {
case PT_TRACE_ME:
case PT_ATTACH:
case PT_STEP:
case PT_CONTINUE:
case PT_DETACH:
sx_xlock(&proctree_lock);
proctree_locked = 1;
break;
default:
break;
}
write = 0;
if (req == PT_TRACE_ME) {
p = td->td_proc;
PROC_LOCK(p);
} else {
if ((p = pfind(pid)) == NULL) {
if (proctree_locked)
sx_xunlock(&proctree_lock);
return (ESRCH);
}
}
if ((error = p_cansee(td, p)) != 0)
goto fail;
if ((error = p_candebug(td, p)) != 0)
goto fail;
/*
* System processes can't be debugged.
*/
if ((p->p_flag & P_SYSTEM) != 0) {
error = EINVAL;
goto fail;
}
/*
* Permissions check
*/
switch (req) {
case PT_TRACE_ME:
/* Always legal. */
break;
case PT_ATTACH:
/* Self */
if (p->p_pid == td->td_proc->p_pid) {
error = EINVAL;
goto fail;
}
/* Already traced */
if (p->p_flag & P_TRACED) {
error = EBUSY;
goto fail;
}
/* Can't trace an ancestor if you're being traced. */
if (curp->p_flag & P_TRACED) {
for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
if (pp == p) {
error = EINVAL;
goto fail;
}
}
}
/* OK */
break;
case PT_READ_I:
case PT_READ_D:
case PT_WRITE_I:
case PT_WRITE_D:
case PT_IO:
case PT_CONTINUE:
case PT_KILL:
case PT_STEP:
case PT_DETACH:
case PT_GETREGS:
case PT_SETREGS:
case PT_GETFPREGS:
case PT_SETFPREGS:
case PT_GETDBREGS:
case PT_SETDBREGS:
/* not being traced... */
if ((p->p_flag & P_TRACED) == 0) {
error = EPERM;
goto fail;
}
/* not being traced by YOU */
if (p->p_pptr != td->td_proc) {
error = EBUSY;
goto fail;
}
/* not currently stopped */
if (!P_SHOULDSTOP(p) || (p->p_flag & P_WAITED) == 0) {
error = EBUSY;
goto fail;
}
/* OK */
break;
default:
error = EINVAL;
goto fail;
}
td2 = FIRST_THREAD_IN_PROC(p);
#ifdef FIX_SSTEP
/*
* Single step fixup ala procfs
*/
FIX_SSTEP(td2); /* XXXKSE */
#endif
/*
* Actually do the requests
*/
td->td_retval[0] = 0;
switch (req) {
case PT_TRACE_ME:
/* set my trace flag and "owner" so it can read/write me */
p->p_flag |= P_TRACED;
p->p_oppid = p->p_pptr->p_pid;
PROC_UNLOCK(p);
sx_xunlock(&proctree_lock);
return (0);
case PT_ATTACH:
/* security check done above */
p->p_flag |= P_TRACED;
p->p_oppid = p->p_pptr->p_pid;
if (p->p_pptr != td->td_proc)
proc_reparent(p, td->td_proc);
data = SIGSTOP;
goto sendsig; /* in PT_CONTINUE below */
case PT_STEP:
case PT_CONTINUE:
case PT_DETACH:
/* XXX data is used even in the PT_STEP case. */
if (req != PT_STEP && (unsigned)data > _SIG_MAXSIG) {
error = EINVAL;
goto fail;
}
_PHOLD(p);
if (req == PT_STEP) {
error = ptrace_single_step(td2);
if (error) {
_PRELE(p);
goto fail;
}
}
if (addr != (void *)1) {
error = ptrace_set_pc(td2, (u_long)(uintfptr_t)addr);
if (error) {
_PRELE(p);
goto fail;
}
}
_PRELE(p);
if (req == PT_DETACH) {
/* reset process parent */
if (p->p_oppid != p->p_pptr->p_pid) {
struct proc *pp;
PROC_UNLOCK(p);
pp = pfind(p->p_oppid);
if (pp == NULL)
pp = initproc;
else
PROC_UNLOCK(pp);
PROC_LOCK(p);
proc_reparent(p, pp);
}
p->p_flag &= ~(P_TRACED | P_WAITED);
p->p_oppid = 0;
/* should we send SIGCHLD? */
}
sendsig:
if (proctree_locked)
sx_xunlock(&proctree_lock);
/* deliver or queue signal */
if (P_SHOULDSTOP(p)) {
p->p_xstat = data;
mtx_lock_spin(&sched_lock);
p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG);
thread_unsuspend(p);
setrunnable(td2); /* XXXKSE */
/* Need foreach kse in proc, ... make_kse_queued(). */
mtx_unlock_spin(&sched_lock);
} else if (data)
psignal(p, data);
PROC_UNLOCK(p);
return (0);
case PT_WRITE_I:
case PT_WRITE_D:
write = 1;
/* FALLTHROUGH */
case PT_READ_I:
case PT_READ_D:
PROC_UNLOCK(p);
tmp = 0;
/* write = 0 set above */
iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
iov.iov_len = sizeof(int);
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = (off_t)(uintptr_t)addr;
uio.uio_resid = sizeof(int);
uio.uio_segflg = UIO_SYSSPACE; /* i.e.: the uap */
uio.uio_rw = write ? UIO_WRITE : UIO_READ;
uio.uio_td = td;
error = proc_rwmem(p, &uio);
if (uio.uio_resid != 0) {
/*
* XXX proc_rwmem() doesn't currently return ENOSPC,
* so I think write() can bogusly return 0.
* XXX what happens for short writes? We don't want
* to write partial data.
* XXX proc_rwmem() returns EPERM for other invalid
* addresses. Convert this to EINVAL. Does this
* clobber returns of EPERM for other reasons?
*/
if (error == 0 || error == ENOSPC || error == EPERM)
error = EINVAL; /* EOF */
}
if (!write)
td->td_retval[0] = tmp;
return (error);
case PT_IO:
PROC_UNLOCK(p);
piod = addr;
iov.iov_base = piod->piod_addr;
iov.iov_len = piod->piod_len;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
uio.uio_resid = piod->piod_len;
uio.uio_segflg = UIO_USERSPACE;
uio.uio_td = td;
switch (piod->piod_op) {
case PIOD_READ_D:
case PIOD_READ_I:
uio.uio_rw = UIO_READ;
break;
case PIOD_WRITE_D:
case PIOD_WRITE_I:
uio.uio_rw = UIO_WRITE;
break;
default:
return (EINVAL);
}
error = proc_rwmem(p, &uio);
piod->piod_len -= uio.uio_resid;
return (error);
case PT_KILL:
data = SIGKILL;
goto sendsig; /* in PT_CONTINUE above */
case PT_SETREGS:
_PHOLD(p);
error = proc_write_regs(td2, addr);
_PRELE(p);
PROC_UNLOCK(p);
return (error);
case PT_GETREGS:
_PHOLD(p);
error = proc_read_regs(td2, addr);
_PRELE(p);
PROC_UNLOCK(p);
return (error);
case PT_SETFPREGS:
_PHOLD(p);
error = proc_write_fpregs(td2, addr);
_PRELE(p);
PROC_UNLOCK(p);
return (error);
case PT_GETFPREGS:
_PHOLD(p);
error = proc_read_fpregs(td2, addr);
_PRELE(p);
PROC_UNLOCK(p);
return (error);
case PT_SETDBREGS:
_PHOLD(p);
error = proc_write_dbregs(td2, addr);
_PRELE(p);
PROC_UNLOCK(p);
return (error);
case PT_GETDBREGS:
_PHOLD(p);
error = proc_read_dbregs(td2, addr);
_PRELE(p);
PROC_UNLOCK(p);
return (error);
default:
KASSERT(0, ("unreachable code\n"));
break;
}
KASSERT(0, ("unreachable code\n"));
return (0);
fail:
PROC_UNLOCK(p);
if (proctree_locked)
sx_xunlock(&proctree_lock);
return (error);
}
