int
umask(int mask)
{
mode_t t;
t = PTOU(curproc)->u_cmask;
PTOU(curproc)->u_cmask = (mode_t)(mask & PERMMASK);
return ((int)t);
}
/*
* Kernel setup code, called from startup().
*/
void
kern_setup1(void)
{
proc_t *pp;
pp = &p0;
proc_sched = pp;
/*
* Initialize process 0 data structures
*/
pp->p_stat = SRUN;
pp->p_flag = SSYS;
pp->p_pidp = &pid0;
pp->p_pgidp = &pid0;
pp->p_sessp = &session0;
pp->p_tlist = &t0;
pid0.pid_pglink = pp;
pid0.pid_pgtail = pp;
/*
* XXX - we asssume that the u-area is zeroed out except for
* ttolwp(curthread)->lwp_regs.
*/
PTOU(curproc)->u_cmask = (mode_t)CMASK;
thread_init(); /* init thread_free list */
pid_init(); /* initialize pid (proc) table */
contract_init(); /* initialize contracts */
init_pages_pp_maximum();
}
/*
* Lookup the user file name from a given vp, using a specific credential.
*/
int
lookuppnatcred(
struct pathname *pnp, /* pathname to lookup */
struct pathname *rpnp, /* if non-NULL, return resolved path */
int followlink, /* (don't) follow sym links */
vnode_t **dirvpp, /* ptr for parent vnode */
vnode_t **compvpp, /* ptr for entry vnode */
vnode_t *startvp, /* start search from this vp */
cred_t *cr) /* user credential */
{
vnode_t *vp; /* current directory vp */
vnode_t *rootvp;
proc_t *p = curproc;
if (pnp->pn_pathlen == 0)
return (ENOENT);
mutex_enter(&p->p_lock); /* for u_rdir and u_cdir */
if ((rootvp = PTOU(p)->u_rdir) == NULL)
rootvp = rootdir;
else if (rootvp != rootdir) /* no need to VN_HOLD rootdir */
VN_HOLD(rootvp);
if (pnp->pn_path[0] == '/') {
vp = rootvp;
} else {
vp = (startvp == NULL) ? PTOU(p)->u_cdir : startvp;
}
VN_HOLD(vp);
mutex_exit(&p->p_lock);
/*
* Skip over leading slashes
*/
if (pnp->pn_path[0] == '/') {
do {
pnp->pn_path++;
pnp->pn_pathlen--;
} while (pnp->pn_path[0] == '/');
}
return (lookuppnvp(pnp, rpnp, followlink, dirvpp,
compvpp, rootvp, vp, cr));
}
/*
* Make a directory.
*/
int
mkdir(char *dname, int dmode)
{
vnode_t *vp;
struct vattr vattr;
int error;
vattr.va_type = VDIR;
vattr.va_mode = dmode & PERMMASK;
vattr.va_mask = AT_TYPE|AT_MODE;
error = vn_create(dname, UIO_USERSPACE, &vattr, EXCL, 0, &vp, CRMKDIR,
0, PTOU(curproc)->u_cmask);
if (error)
return (set_errno(error));
VN_RELE(vp);
return (0);
}
/*
* Native processes are started with the native ld.so.1 as the command. This
* brand op is invoked by s10_npreload to fix up the command and arguments
* so that apps like pgrep or ps see the expected command strings.
*/
int
s10_native(void *cmd, void *args)
{
struct user *up = PTOU(curproc);
char cmd_buf[MAXCOMLEN + 1];
char arg_buf[PSARGSZ];
if (copyin(cmd, &cmd_buf, sizeof (cmd_buf)) != 0)
return (EFAULT);
if (copyin(args, &arg_buf, sizeof (arg_buf)) != 0)
return (EFAULT);
/*
* Make sure that the process' interpreter is the native dynamic linker.
* Convention dictates that native processes executing within solaris10-
* branded zones are interpreted by the native dynamic linker (the
* process and its arguments are specified as arguments to the dynamic
* linker). If this convention is violated (i.e.,
* brandsys(B_S10_NATIVE, ...) is invoked by a process that shouldn't be
* native), then do nothing and silently indicate success.
*/
if (strcmp(up->u_comm, S10_LINKER_NAME) != 0)
return (0);
/*
* The sizeof has an extra value for the trailing '\0' so this covers
* the appended " " in the following strcmps.
*/
if (strncmp(up->u_psargs, BRAND_NATIVE_LINKER64 " ",
sizeof (BRAND_NATIVE_LINKER64)) != 0 &&
strncmp(up->u_psargs, BRAND_NATIVE_LINKER32 " ",
sizeof (BRAND_NATIVE_LINKER32)) != 0)
return (0);
mutex_enter(&curproc->p_lock);
(void) strlcpy(up->u_comm, cmd_buf, sizeof (up->u_comm));
(void) strlcpy(up->u_psargs, arg_buf, sizeof (up->u_psargs));
mutex_exit(&curproc->p_lock);
return (0);
}
/*
* Perform pre-system-call processing, including stopping for tracing,
* auditing, microstate-accounting, etc.
*
* This routine is called only if the t_pre_sys flag is set. Any condition
* requiring pre-syscall handling must set the t_pre_sys flag. If the
* condition is persistent, this routine will repost t_pre_sys.
*/
int
pre_syscall(int arg0)
{
unsigned int code;
kthread_t *t = curthread;
proc_t *p = ttoproc(t);
klwp_t *lwp = ttolwp(t);
struct regs *rp = lwptoregs(lwp);
int repost;
t->t_pre_sys = repost = 0; /* clear pre-syscall processing flag */
ASSERT(t->t_schedflag & TS_DONT_SWAP);
syscall_mstate(LMS_USER, LMS_SYSTEM);
/*
* The syscall arguments in the out registers should be pointed to
* by lwp_ap. If the args need to be copied so that the outs can
* be changed without losing the ability to get the args for /proc,
* they can be saved by save_syscall_args(), and lwp_ap will be
* restored by post_syscall().
*/
ASSERT(lwp->lwp_ap == (long *)&rp->r_o0);
/*
* Make sure the thread is holding the latest credentials for the
* process. The credentials in the process right now apply to this
* thread for the entire system call.
*/
if (t->t_cred != p->p_cred) {
cred_t *oldcred = t->t_cred;
/*
* DTrace accesses t_cred in probe context. t_cred must
* always be either NULL, or point to a valid, allocated cred
* structure.
*/
t->t_cred = crgetcred();
crfree(oldcred);
}
/*
* Undo special arrangements to single-step the lwp
* so that a debugger will see valid register contents.
* Also so that the pc is valid for syncfpu().
* Also so that a syscall like exec() can be stepped.
*/
if (lwp->lwp_pcb.pcb_step != STEP_NONE) {
(void) prundostep();
repost = 1;
}
/*
* Check for indirect system call in case we stop for tracing.
* Don't allow multiple indirection.
*/
code = t->t_sysnum;
if (code == 0 && arg0 != 0) { /* indirect syscall */
code = arg0;
t->t_sysnum = arg0;
}
/*
* From the proc(4) manual page:
* When entry to a system call is being traced, the traced process
* stops after having begun the call to the system but before the
* system call arguments have been fetched from the process.
* If proc changes the args we must refetch them after starting.
*/
if (PTOU(p)->u_systrap) {
if (prismember(&PTOU(p)->u_entrymask, code)) {
/*
* Recheck stop condition, now that lock is held.
*/
mutex_enter(&p->p_lock);
if (PTOU(p)->u_systrap &&
prismember(&PTOU(p)->u_entrymask, code)) {
stop(PR_SYSENTRY, code);
/*
* Must refetch args since they were
* possibly modified by /proc. Indicate
* that the valid copy is in the
* registers.
*/
lwp->lwp_argsaved = 0;
lwp->lwp_ap = (long *)&rp->r_o0;
}
mutex_exit(&p->p_lock);
}
repost = 1;
}
if (lwp->lwp_sysabort) {
/*
* lwp_sysabort may have been set via /proc while the process
* was stopped on PR_SYSENTRY. If so, abort the system call.
* Override any error from the copyin() of the arguments.
*/
lwp->lwp_sysabort = 0;
(void) set_errno(EINTR); /* sets post-sys processing */
t->t_pre_sys = 1; /* repost anyway */
return (1); /* don't do system call, return EINTR */
}
#ifdef C2_AUDIT
if (audit_active) { /* begin auditing for this syscall */
int error;
if (error = audit_start(T_SYSCALL, code, 0, lwp)) {
t->t_pre_sys = 1; /* repost anyway */
lwp->lwp_error = 0; /* for old drivers */
return (error);
}
repost = 1;
}
#endif /* C2_AUDIT */
#ifndef NPROBE
/* Kernel probe */
if (tnf_tracing_active) {
TNF_PROBE_1(syscall_start, "syscall thread", /* CSTYLED */,
tnf_sysnum, sysnum, t->t_sysnum);
t->t_post_sys = 1; /* make sure post_syscall runs */
repost = 1;
}
#endif /* NPROBE */
#ifdef SYSCALLTRACE
if (syscalltrace) {
int i;
long *ap;
char *cp;
char *sysname;
struct sysent *callp;
if (code >= NSYSCALL)
callp = &nosys_ent; /* nosys has no args */
else
callp = LWP_GETSYSENT(lwp) + code;
(void) save_syscall_args();
mutex_enter(&systrace_lock);
printf("%d: ", p->p_pid);
if (code >= NSYSCALL)
printf("0x%x", code);
else {
sysname = mod_getsysname(code);
printf("%s[0x%x]", sysname == NULL ? "NULL" :
sysname, code);
}
cp = "(";
for (i = 0, ap = lwp->lwp_ap; i < callp->sy_narg; i++, ap++) {
printf("%s%lx", cp, *ap);
cp = ", ";
}
if (i)
printf(")");
printf(" %s id=0x%p\n", PTOU(p)->u_comm, curthread);
mutex_exit(&systrace_lock);
}
#endif /* SYSCALLTRACE */
/*
* If there was a continuing reason for pre-syscall processing,
* set the t_pre_sys flag for the next system call.
*/
if (repost)
t->t_pre_sys = 1;
lwp->lwp_error = 0; /* for old drivers */
lwp->lwp_badpriv = PRIV_NONE; /* for privilege tracing */
return (0);
}
static int
chdirec(vnode_t *vp, int ischroot, int do_traverse)
{
int error;
vnode_t *oldvp;
proc_t *pp = curproc;
vnode_t **vpp;
refstr_t *cwd;
int newcwd = 1;
if (vp->v_type != VDIR) {
error = ENOTDIR;
goto bad;
}
if (error = VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL))
goto bad;
/*
* The VOP_ACCESS() may have covered 'vp' with a new filesystem,
* if 'vp' is an autoFS vnode. Traverse the mountpoint so
* that we don't end up with a covered current directory.
*/
if (vn_mountedvfs(vp) != NULL && do_traverse) {
if (error = traverse(&vp))
goto bad;
}
/*
* Special chroot semantics: chroot is allowed if privileged
* or if the target is really a loopback mount of the root (or
* root of the zone) as determined by comparing dev and inode
* numbers
*/
if (ischroot) {
struct vattr tattr;
struct vattr rattr;
vnode_t *zonevp = curproc->p_zone->zone_rootvp;
tattr.va_mask = AT_FSID|AT_NODEID;
if (error = VOP_GETATTR(vp, &tattr, 0, CRED(), NULL))
goto bad;
rattr.va_mask = AT_FSID|AT_NODEID;
if (error = VOP_GETATTR(zonevp, &rattr, 0, CRED(), NULL))
goto bad;
if ((tattr.va_fsid != rattr.va_fsid ||
tattr.va_nodeid != rattr.va_nodeid) &&
(error = secpolicy_chroot(CRED())) != 0)
goto bad;
vpp = &PTOU(pp)->u_rdir;
} else {
vpp = &PTOU(pp)->u_cdir;
}
if (audit_active) /* update abs cwd/root path see c2audit.c */
audit_chdirec(vp, vpp);
mutex_enter(&pp->p_lock);
/*
* This bit of logic prevents us from overwriting u_cwd if we are
* changing to the same directory. We set the cwd to NULL so that we
* don't try to do the lookup on the next call to getcwd().
*/
if (!ischroot && *vpp != NULL && vp != NULL && VN_CMP(*vpp, vp))
newcwd = 0;
oldvp = *vpp;
*vpp = vp;
if ((cwd = PTOU(pp)->u_cwd) != NULL && newcwd)
PTOU(pp)->u_cwd = NULL;
mutex_exit(&pp->p_lock);
if (cwd && newcwd)
refstr_rele(cwd);
if (oldvp)
VN_RELE(oldvp);
return (0);
bad:
VN_RELE(vp);
return (error);
}
/*
* Perform pre-system-call processing, including stopping for tracing,
* auditing, etc.
*
* This routine is called only if the t_pre_sys flag is set. Any condition
* requiring pre-syscall handling must set the t_pre_sys flag. If the
* condition is persistent, this routine will repost t_pre_sys.
*/
int
pre_syscall()
{
kthread_t *t = curthread;
unsigned code = t->t_sysnum;
klwp_t *lwp = ttolwp(t);
proc_t *p = ttoproc(t);
int repost;
t->t_pre_sys = repost = 0; /* clear pre-syscall processing flag */
ASSERT(t->t_schedflag & TS_DONT_SWAP);
#if defined(DEBUG)
/*
* On the i386 kernel, lwp_ap points at the piece of the thread
* stack that we copy the users arguments into.
*
* On the amd64 kernel, the syscall arguments in the rdi..r9
* registers should be pointed at by lwp_ap. If the args need to
* be copied so that those registers can be changed without losing
* the ability to get the args for /proc, they can be saved by
* save_syscall_args(), and lwp_ap will be restored by post_syscall().
*/
if (lwp_getdatamodel(lwp) == DATAMODEL_NATIVE) {
#if defined(_LP64)
ASSERT(lwp->lwp_ap == (long *)&lwptoregs(lwp)->r_rdi);
} else {
#endif
ASSERT((caddr_t)lwp->lwp_ap > t->t_stkbase &&
(caddr_t)lwp->lwp_ap < t->t_stk);
}
#endif /* DEBUG */
/*
* Make sure the thread is holding the latest credentials for the
* process. The credentials in the process right now apply to this
* thread for the entire system call.
*/
if (t->t_cred != p->p_cred) {
cred_t *oldcred = t->t_cred;
/*
* DTrace accesses t_cred in probe context. t_cred must
* always be either NULL, or point to a valid, allocated cred
* structure.
*/
t->t_cred = crgetcred();
crfree(oldcred);
}
/*
* From the proc(4) manual page:
* When entry to a system call is being traced, the traced process
* stops after having begun the call to the system but before the
* system call arguments have been fetched from the process.
*/
if (PTOU(p)->u_systrap) {
if (prismember(&PTOU(p)->u_entrymask, code)) {
mutex_enter(&p->p_lock);
/*
* Recheck stop condition, now that lock is held.
*/
if (PTOU(p)->u_systrap &&
prismember(&PTOU(p)->u_entrymask, code)) {
stop(PR_SYSENTRY, code);
/*
* /proc may have modified syscall args,
* either in regs for amd64 or on ustack
* for ia32. Either way, arrange to
* copy them again, both for the syscall
* handler and for other consumers in
* post_syscall (like audit). Here, we
* only do amd64, and just set lwp_ap
* back to the kernel-entry stack copy;
* the syscall ml code redoes
* move-from-regs to set up for the
* syscall handler after we return. For
* ia32, save_syscall_args() below makes
* an lwp_ap-accessible copy.
*/
#if defined(_LP64)
if (lwp_getdatamodel(lwp) == DATAMODEL_NATIVE) {
lwp->lwp_argsaved = 0;
lwp->lwp_ap =
(long *)&lwptoregs(lwp)->r_rdi;
}
#endif
}
mutex_exit(&p->p_lock);
}
repost = 1;
}
/*
* ia32 kernel, or ia32 proc on amd64 kernel: keep args in
* lwp_arg for post-syscall processing, regardless of whether
* they might have been changed in /proc above.
*/
#if defined(_LP64)
if (lwp_getdatamodel(lwp) != DATAMODEL_NATIVE)
#endif
(void) save_syscall_args();
if (lwp->lwp_sysabort) {
/*
* lwp_sysabort may have been set via /proc while the process
* was stopped on PR_SYSENTRY. If so, abort the system call.
* Override any error from the copyin() of the arguments.
*/
lwp->lwp_sysabort = 0;
(void) set_errno(EINTR); /* forces post_sys */
t->t_pre_sys = 1; /* repost anyway */
return (1); /* don't do system call, return EINTR */
}
#ifdef C2_AUDIT
if (audit_active) { /* begin auditing for this syscall */
int error;
if (error = audit_start(T_SYSCALL, code, 0, lwp)) {
t->t_pre_sys = 1; /* repost anyway */
(void) set_errno(error);
return (1);
}
repost = 1;
}
#endif /* C2_AUDIT */
#ifndef NPROBE
/* Kernel probe */
if (tnf_tracing_active) {
TNF_PROBE_1(syscall_start, "syscall thread", /* CSTYLED */,
tnf_sysnum, sysnum, t->t_sysnum);
t->t_post_sys = 1; /* make sure post_syscall runs */
repost = 1;
}
#endif /* NPROBE */
#ifdef SYSCALLTRACE
if (syscalltrace) {
int i;
long *ap;
char *cp;
char *sysname;
struct sysent *callp;
if (code >= NSYSCALL)
callp = &nosys_ent; /* nosys has no args */
else
callp = LWP_GETSYSENT(lwp) + code;
(void) save_syscall_args();
mutex_enter(&systrace_lock);
printf("%d: ", p->p_pid);
if (code >= NSYSCALL)
printf("0x%x", code);
else {
sysname = mod_getsysname(code);
printf("%s[0x%x/0x%p]", sysname == NULL ? "NULL" :
sysname, code, callp->sy_callc);
}
cp = "(";
for (i = 0, ap = lwp->lwp_ap; i < callp->sy_narg; i++, ap++) {
printf("%s%lx", cp, *ap);
cp = ", ";
}
if (i)
printf(")");
printf(" %s id=0x%p\n", PTOU(p)->u_comm, curthread);
mutex_exit(&systrace_lock);
}
#endif /* SYSCALLTRACE */
/*
* If there was a continuing reason for pre-syscall processing,
* set the t_pre_sys flag for the next system call.
*/
if (repost)
t->t_pre_sys = 1;
lwp->lwp_error = 0; /* for old drivers */
lwp->lwp_badpriv = PRIV_NONE;
return (0);
}
static int
copen(int startfd, char *fname, int filemode, int createmode)
{
struct pathname pn;
vnode_t *vp, *sdvp;
file_t *fp, *startfp;
enum vtype type;
int error;
int fd, dupfd;
vnode_t *startvp;
proc_t *p = curproc;
if (startfd == AT_FDCWD) {
/*
* Regular open()
*/
startvp = NULL;
} else {
/*
* We're here via openat()
*/
char startchar;
if (copyin(fname, &startchar, sizeof (char)))
return (set_errno(EFAULT));
/*
* if startchar is / then startfd is ignored
*/
if (startchar == '/')
startvp = NULL;
else {
if ((startfp = getf(startfd)) == NULL)
return (set_errno(EBADF));
startvp = startfp->f_vnode;
VN_HOLD(startvp);
releasef(startfd);
}
}
if (filemode & FXATTR) {
/*
* Make sure we have a valid request.
* We must either have a real fd or AT_FDCWD
*/
if (startfd != AT_FDCWD && startvp == NULL) {
error = EINVAL;
goto out;
}
if (error = pn_get(fname, UIO_USERSPACE, &pn)) {
goto out;
}
if (startfd == AT_FDCWD) {
mutex_enter(&p->p_lock);
startvp = PTOU(p)->u_cdir;
VN_HOLD(startvp);
mutex_exit(&p->p_lock);
}
/*
* Verify permission to put attributes on file
*/
if ((VOP_ACCESS(startvp, VREAD, 0, CRED()) != 0) &&
(VOP_ACCESS(startvp, VWRITE, 0, CRED()) != 0) &&
(VOP_ACCESS(startvp, VEXEC, 0, CRED()) != 0)) {
error = EACCES;
pn_free(&pn);
goto out;
}
if ((startvp->v_vfsp->vfs_flag & VFS_XATTR) != 0) {
error = VOP_LOOKUP(startvp, "", &sdvp, &pn,
LOOKUP_XATTR|CREATE_XATTR_DIR, rootvp, CRED());
} else {
error = EINVAL;
}
pn_free(&pn);
if (error != 0)
goto out;
VN_RELE(startvp);
startvp = sdvp;
}
if ((filemode & (FREAD|FWRITE)) != 0) {
if ((filemode & (FNONBLOCK|FNDELAY)) == (FNONBLOCK|FNDELAY))
filemode &= ~FNDELAY;
error = falloc((vnode_t *)NULL, filemode, &fp, &fd);
if (error == 0) {
#ifdef C2_AUDIT
if (audit_active)
audit_setfsat_path(1);
#endif /* C2_AUDIT */
/*
* Last arg is a don't-care term if
* !(filemode & FCREAT).
*/
error = vn_openat(fname, UIO_USERSPACE, filemode,
(int)(createmode & MODEMASK), &vp, CRCREAT,
u.u_cmask, startvp);
if (startvp != NULL)
VN_RELE(startvp);
if (error == 0) {
#ifdef C2_AUDIT
if (audit_active)
audit_copen(fd, fp, vp);
#endif /* C2_AUDIT */
if ((vp->v_flag & VDUP) == 0) {
fp->f_vnode = vp;
mutex_exit(&fp->f_tlock);
/*
* We must now fill in the slot
* falloc reserved.
*/
setf(fd, fp);
return (fd);
} else {
/*
* Special handling for /dev/fd.
* Give up the file pointer
* and dup the indicated file descriptor
* (in v_rdev). This is ugly, but I've
* seen worse.
*/
unfalloc(fp);
dupfd = getminor(vp->v_rdev);
type = vp->v_type;
mutex_enter(&vp->v_lock);
vp->v_flag &= ~VDUP;
mutex_exit(&vp->v_lock);
VN_RELE(vp);
if (type != VCHR)
return (set_errno(EINVAL));
if ((fp = getf(dupfd)) == NULL) {
setf(fd, NULL);
return (set_errno(EBADF));
}
mutex_enter(&fp->f_tlock);
fp->f_count++;
mutex_exit(&fp->f_tlock);
setf(fd, fp);
releasef(dupfd);
}
return (fd);
} else {
setf(fd, NULL);
unfalloc(fp);
return (set_errno(error));
}
}
} else {
error = EINVAL;
}
out:
if (startvp != NULL)
VN_RELE(startvp);
return (set_errno(error));
}
int
sendsig(int sig, k_siginfo_t *sip, void (*hdlr)())
{
volatile int minstacksz;
int newstack;
label_t ljb;
volatile caddr_t sp;
caddr_t fp;
volatile struct regs *rp;
volatile greg_t upc;
volatile proc_t *p = ttoproc(curthread);
struct as *as = p->p_as;
klwp_t *lwp = ttolwp(curthread);
ucontext_t *volatile tuc = NULL;
ucontext_t *uc;
siginfo_t *sip_addr;
volatile int watched;
/*
* This routine is utterly dependent upon STACK_ALIGN being
* 16 and STACK_ENTRY_ALIGN being 8. Let's just acknowledge
* that and require it.
*/
#if STACK_ALIGN != 16 || STACK_ENTRY_ALIGN != 8
#error "sendsig() amd64 did not find the expected stack alignments"
#endif
rp = lwptoregs(lwp);
upc = rp->r_pc;
/*
* Since we're setting up to run the signal handler we have to
* arrange that the stack at entry to the handler is (only)
* STACK_ENTRY_ALIGN (i.e. 8) byte aligned so that when the handler
* executes its push of %rbp, the stack realigns to STACK_ALIGN
* (i.e. 16) correctly.
*
* The new sp will point to the sigframe and the ucontext_t. The
* above means that sp (and thus sigframe) will be 8-byte aligned,
* but not 16-byte aligned. ucontext_t, however, contains %xmm regs
* which must be 16-byte aligned. Because of this, for correct
* alignment, sigframe must be a multiple of 8-bytes in length, but
* not 16-bytes. This will place ucontext_t at a nice 16-byte boundary.
*/
/* LINTED: logical expression always true: op "||" */
ASSERT((sizeof (struct sigframe) % 16) == 8);
minstacksz = sizeof (struct sigframe) + SA(sizeof (*uc));
if (sip != NULL)
minstacksz += SA(sizeof (siginfo_t));
ASSERT((minstacksz & (STACK_ENTRY_ALIGN - 1ul)) == 0);
/*
* Figure out whether we will be handling this signal on
* an alternate stack specified by the user. Then allocate
* and validate the stack requirements for the signal handler
* context. on_fault will catch any faults.
*/
newstack = sigismember(&PTOU(curproc)->u_sigonstack, sig) &&
!(lwp->lwp_sigaltstack.ss_flags & (SS_ONSTACK|SS_DISABLE));
if (newstack) {
fp = (caddr_t)(SA((uintptr_t)lwp->lwp_sigaltstack.ss_sp) +
SA(lwp->lwp_sigaltstack.ss_size) - STACK_ALIGN);
} else {
/*
* Drop below the 128-byte reserved region of the stack frame
* we're interrupting.
*/
fp = (caddr_t)rp->r_sp - STACK_RESERVE;
}
/*
* Force proper stack pointer alignment, even in the face of a
* misaligned stack pointer from user-level before the signal.
*/
fp = (caddr_t)((uintptr_t)fp & ~(STACK_ENTRY_ALIGN - 1ul));
/*
* Most of the time during normal execution, the stack pointer
* is aligned on a STACK_ALIGN (i.e. 16 byte) boundary. However,
* (for example) just after a call instruction (which pushes
* the return address), the callers stack misaligns until the
* 'push %rbp' happens in the callee prolog. So while we should
* expect the stack pointer to be always at least STACK_ENTRY_ALIGN
* aligned, we should -not- expect it to always be STACK_ALIGN aligned.
* We now adjust to ensure that the new sp is aligned to
* STACK_ENTRY_ALIGN but not to STACK_ALIGN.
*/
sp = fp - minstacksz;
if (((uintptr_t)sp & (STACK_ALIGN - 1ul)) == 0) {
sp -= STACK_ENTRY_ALIGN;
minstacksz = fp - sp;
}
/*
* Now, make sure the resulting signal frame address is sane
*/
if (sp >= as->a_userlimit || fp >= as->a_userlimit) {
#ifdef DEBUG
printf("sendsig: bad signal stack cmd=%s, pid=%d, sig=%d\n",
PTOU(p)->u_comm, p->p_pid, sig);
printf("sigsp = 0x%p, action = 0x%p, upc = 0x%lx\n",
(void *)sp, (void *)hdlr, (uintptr_t)upc);
printf("sp above USERLIMIT\n");
#endif
return (0);
}
watched = watch_disable_addr((caddr_t)sp, minstacksz, S_WRITE);
if (on_fault(&ljb))
goto badstack;
if (sip != NULL) {
zoneid_t zoneid;
fp -= SA(sizeof (siginfo_t));
uzero(fp, sizeof (siginfo_t));
if (SI_FROMUSER(sip) &&
(zoneid = p->p_zone->zone_id) != GLOBAL_ZONEID &&
zoneid != sip->si_zoneid) {
k_siginfo_t sani_sip = *sip;
sani_sip.si_pid = p->p_zone->zone_zsched->p_pid;
sani_sip.si_uid = 0;
sani_sip.si_ctid = -1;
sani_sip.si_zoneid = zoneid;
copyout_noerr(&sani_sip, fp, sizeof (sani_sip));
} else
copyout_noerr(sip, fp, sizeof (*sip));
sip_addr = (siginfo_t *)fp;
if (sig == SIGPROF &&
curthread->t_rprof != NULL &&
curthread->t_rprof->rp_anystate) {
/*
* We stand on our head to deal with
* the real time profiling signal.
* Fill in the stuff that doesn't fit
* in a normal k_siginfo structure.
*/
int i = sip->si_nsysarg;
while (--i >= 0)
sulword_noerr(
(ulong_t *)&(sip_addr->si_sysarg[i]),
(ulong_t)lwp->lwp_arg[i]);
copyout_noerr(curthread->t_rprof->rp_state,
sip_addr->si_mstate,
sizeof (curthread->t_rprof->rp_state));
}
} else
sip_addr = NULL;
/*
* save the current context on the user stack directly after the
* sigframe. Since sigframe is 8-byte-but-not-16-byte aligned,
* and since sizeof (struct sigframe) is 24, this guarantees
* 16-byte alignment for ucontext_t and its %xmm registers.
*/
uc = (ucontext_t *)(sp + sizeof (struct sigframe));
tuc = kmem_alloc(sizeof (*tuc), KM_SLEEP);
no_fault();
savecontext(tuc, &lwp->lwp_sigoldmask);
if (on_fault(&ljb))
goto badstack;
copyout_noerr(tuc, uc, sizeof (*tuc));
kmem_free(tuc, sizeof (*tuc));
tuc = NULL;
lwp->lwp_oldcontext = (uintptr_t)uc;
if (newstack) {
lwp->lwp_sigaltstack.ss_flags |= SS_ONSTACK;
if (lwp->lwp_ustack)
copyout_noerr(&lwp->lwp_sigaltstack,
(stack_t *)lwp->lwp_ustack, sizeof (stack_t));
}
/*
* Set up signal handler return and stack linkage
*/
{
struct sigframe frame;
/*
* ensure we never return "normally"
*/
frame.retaddr = (caddr_t)(uintptr_t)-1L;
frame.signo = sig;
frame.sip = sip_addr;
copyout_noerr(&frame, sp, sizeof (frame));
}
no_fault();
if (watched)
watch_enable_addr((caddr_t)sp, minstacksz, S_WRITE);
/*
* Set up user registers for execution of signal handler.
*/
rp->r_sp = (greg_t)sp;
rp->r_pc = (greg_t)hdlr;
rp->r_ps = PSL_USER | (rp->r_ps & PS_IOPL);
rp->r_rdi = sig;
rp->r_rsi = (uintptr_t)sip_addr;
rp->r_rdx = (uintptr_t)uc;
if ((rp->r_cs & 0xffff) != UCS_SEL ||
(rp->r_ss & 0xffff) != UDS_SEL) {
/*
* Try our best to deliver the signal.
*/
rp->r_cs = UCS_SEL;
rp->r_ss = UDS_SEL;
}
/*
* Don't set lwp_eosys here. sendsig() is called via psig() after
* lwp_eosys is handled, so setting it here would affect the next
* system call.
*/
return (1);
badstack:
no_fault();
if (watched)
watch_enable_addr((caddr_t)sp, minstacksz, S_WRITE);
if (tuc)
kmem_free(tuc, sizeof (*tuc));
#ifdef DEBUG
printf("sendsig: bad signal stack cmd=%s, pid=%d, sig=%d\n",
PTOU(p)->u_comm, p->p_pid, sig);
printf("on fault, sigsp = 0x%p, action = 0x%p, upc = 0x%lx\n",
(void *)sp, (void *)hdlr, (uintptr_t)upc);
#endif
return (0);
}
/*
* The additional flag, LOOKUP_CHECKREAD, is used to enforce artificial
* constraints in order to be standards compliant. For example, if we have
* the cached path of '/foo/bar', and '/foo' has permissions 100 (execute
* only), then we can legitimately look up the path to the current working
* directory without needing read permission. Existing standards tests,
* however, assume that we are determining the path by repeatedly looking up
* "..". We need to keep this behavior in order to maintain backwards
* compatibility.
*/
static int
vnodetopath_common(vnode_t *vrootp, vnode_t *vp, char *buf, size_t buflen,
cred_t *cr, int flags)
{
pathname_t pn, rpn;
int ret, len;
vnode_t *compvp, *pvp, *realvp;
proc_t *p = curproc;
char path[MAXNAMELEN];
int doclose = 0;
/*
* If vrootp is NULL, get the root for curproc. Callers with any other
* requirements should pass in a different vrootp.
*/
if (vrootp == NULL) {
mutex_enter(&p->p_lock);
if ((vrootp = PTOU(p)->u_rdir) == NULL)
vrootp = rootdir;
VN_HOLD(vrootp);
mutex_exit(&p->p_lock);
} else {
VN_HOLD(vrootp);
}
/*
* This is to get around an annoying artifact of the /proc filesystem,
* which is the behavior of {cwd/root}. Trying to resolve this path
* will result in /proc/pid/cwd instead of whatever the real working
* directory is. We can't rely on VOP_REALVP(), since that will break
* lofs. The only difference between procfs and lofs is that opening
* the file will return the underling vnode in the case of procfs.
*/
if (vp->v_type == VDIR && VOP_REALVP(vp, &realvp, NULL) == 0 &&
realvp != vp) {
VN_HOLD(vp);
if (VOP_OPEN(&vp, FREAD, cr, NULL) == 0)
doclose = 1;
else
VN_RELE(vp);
}
pn_alloc(&pn);
/*
* Check to see if we have a cached path in the vnode.
*/
mutex_enter(&vp->v_lock);
if (vp->v_path != NULL) {
(void) pn_set(&pn, vp->v_path);
mutex_exit(&vp->v_lock);
pn_alloc(&rpn);
/* We should only cache absolute paths */
ASSERT(pn.pn_buf[0] == '/');
/*
* If we are in a zone or a chroot environment, then we have to
* take additional steps, since the path to the root might not
* be readable with the current credentials, even though the
* process can legitmately access the file. In this case, we
* do the following:
*
* lookuppnvp() with all privileges to get the resolved path.
* call localpath() to get the local portion of the path, and
* continue as normal.
*
* If the the conversion to a local path fails, then we continue
* as normal. This is a heuristic to make process object file
* paths available from within a zone. Because lofs doesn't
* support page operations, the vnode stored in the seg_t is
* actually the underlying real vnode, not the lofs node itself.
* Most of the time, the lofs path is the same as the underlying
* vnode (for example, /usr/lib/libc.so.1).
*/
if (vrootp != rootdir) {
char *local = NULL;
VN_HOLD(rootdir);
if (lookuppnvp(&pn, &rpn, FOLLOW,
NULL, &compvp, rootdir, rootdir, kcred) == 0) {
local = localpath(rpn.pn_path, vrootp,
kcred);
VN_RELE(compvp);
}
/*
* The original pn was changed through lookuppnvp().
* Set it to local for next validation attempt.
*/
if (local) {
(void) pn_set(&pn, local);
} else {
goto notcached;
}
}
/*
* We should have a local path at this point, so start the
* search from the root of the current process.
*/
VN_HOLD(vrootp);
if (vrootp != rootdir)
VN_HOLD(vrootp);
ret = lookuppnvp(&pn, &rpn, FOLLOW | flags, NULL,
&compvp, vrootp, vrootp, cr);
if (ret == 0) {
/*
* Check to see if the returned vnode is the same as
* the one we expect. If not, give up.
*/
if (!vn_compare(vp, compvp) &&
!vnode_match(vp, compvp, cr)) {
VN_RELE(compvp);
goto notcached;
}
VN_RELE(compvp);
/*
* Return the result.
*/
if (buflen <= rpn.pn_pathlen)
goto notcached;
bcopy(rpn.pn_path, buf, rpn.pn_pathlen + 1);
pn_free(&pn);
pn_free(&rpn);
VN_RELE(vrootp);
if (doclose) {
(void) VOP_CLOSE(vp, FREAD, 1, 0, cr, NULL);
VN_RELE(vp);
}
return (0);
}
notcached:
pn_free(&rpn);
} else {
mutex_exit(&vp->v_lock);
}
pn_free(&pn);
if (vp->v_type != VDIR) {
/*
* If we don't have a directory, try to find it in the dnlc via
* reverse lookup. Once this is found, we can use the regular
* directory search to find the full path.
*/
if ((pvp = dnlc_reverse_lookup(vp, path, MAXNAMELEN)) != NULL) {
/*
* Check if we have read privilege so, that
* we can lookup the path in the directory
*/
ret = 0;
if ((flags & LOOKUP_CHECKREAD)) {
ret = VOP_ACCESS(pvp, VREAD, 0, cr, NULL);
}
if (ret == 0) {
ret = dirtopath(vrootp, pvp, buf, buflen,
flags, cr);
}
if (ret == 0) {
len = strlen(buf);
if (len + strlen(path) + 1 >= buflen) {
ret = ENAMETOOLONG;
} else {
if (buf[len - 1] != '/')
buf[len++] = '/';
bcopy(path, buf + len,
strlen(path) + 1);
}
}
VN_RELE(pvp);
} else
ret = ENOENT;
} else
ret = dirtopath(vrootp, vp, buf, buflen, flags, cr);
VN_RELE(vrootp);
if (doclose) {
(void) VOP_CLOSE(vp, FREAD, 1, 0, cr, NULL);
VN_RELE(vp);
}
return (ret);
}
void
main(void)
{
proc_t *p = ttoproc(curthread); /* &p0 */
int (**initptr)();
extern void sched();
extern void fsflush();
extern int (*init_tbl[])();
extern int (*mp_init_tbl[])();
extern id_t syscid, defaultcid;
extern int swaploaded;
extern int netboot;
extern ib_boot_prop_t *iscsiboot_prop;
extern void vm_init(void);
extern void cbe_init_pre(void);
extern void cbe_init(void);
extern void clock_tick_init_pre(void);
extern void clock_tick_init_post(void);
extern void clock_init(void);
extern void physio_bufs_init(void);
extern void pm_cfb_setup_intr(void);
extern int pm_adjust_timestamps(dev_info_t *, void *);
extern void start_other_cpus(int);
extern void sysevent_evc_thrinit();
extern kmutex_t ualock;
#if defined(__x86)
extern void fastboot_post_startup(void);
extern void progressbar_start(void);
#endif
/*
* In the horrible world of x86 in-lines, you can't get symbolic
* structure offsets a la genassym. This assertion is here so
* that the next poor slob who innocently changes the offset of
* cpu_thread doesn't waste as much time as I just did finding
* out that it's hard-coded in i86/ml/i86.il. Similarly for
* curcpup. You're welcome.
*/
ASSERT(CPU == CPU->cpu_self);
ASSERT(curthread == CPU->cpu_thread);
ASSERT_STACK_ALIGNED();
/*
* We take the ualock until we have completed the startup
* to prevent kadmin() from disrupting this work. In particular,
* we don't want kadmin() to bring the system down while we are
* trying to start it up.
*/
mutex_enter(&ualock);
/*
* Setup root lgroup and leaf lgroup for CPU 0
*/
lgrp_init(LGRP_INIT_STAGE2);
/*
* Once 'startup()' completes, the thread_reaper() daemon would be
* created(in thread_init()). After that, it is safe to create threads
* that could exit. These exited threads will get reaped.
*/
startup();
segkmem_gc();
callb_init();
cbe_init_pre(); /* x86 must initialize gethrtimef before timer_init */
ddi_periodic_init();
cbe_init();
callout_init(); /* callout table MUST be init'd after cyclics */
clock_tick_init_pre();
clock_init();
#if defined(__x86)
/*
* The progressbar thread uses cv_reltimedwait() and hence needs to be
* started after the callout mechanism has been initialized.
*/
progressbar_start();
#endif
/*
* On some platforms, clkinitf() changes the timing source that
* gethrtime_unscaled() uses to generate timestamps. cbe_init() calls
* clkinitf(), so re-initialize the microstate counters after the
* timesource has been chosen.
*/
init_mstate(&t0, LMS_SYSTEM);
init_cpu_mstate(CPU, CMS_SYSTEM);
/*
* May need to probe to determine latencies from CPU 0 after
* gethrtime() comes alive in cbe_init() and before enabling interrupts
* and copy and release any temporary memory allocated with BOP_ALLOC()
* before release_bootstrap() frees boot memory
*/
lgrp_init(LGRP_INIT_STAGE3);
/*
* Call all system initialization functions.
*/
for (initptr = &init_tbl[0]; *initptr; initptr++)
(**initptr)();
/*
* Load iSCSI boot properties
*/
ld_ib_prop();
/*
* initialize vm related stuff.
*/
vm_init();
/*
* initialize buffer pool for raw I/O requests
*/
physio_bufs_init();
ttolwp(curthread)->lwp_error = 0; /* XXX kludge for SCSI driver */
/*
* Drop the interrupt level and allow interrupts. At this point
* the DDI guarantees that interrupts are enabled.
*/
(void) spl0();
interrupts_unleashed = 1;
/*
* Create kmem cache for proc structures
*/
process_cache = kmem_cache_create("process_cache", sizeof (proc_t),
0, NULL, NULL, NULL, NULL, NULL, 0);
vfs_mountroot(); /* Mount the root file system */
errorq_init(); /* after vfs_mountroot() so DDI root is ready */
cpu_kstat_init(CPU); /* after vfs_mountroot() so TOD is valid */
ddi_walk_devs(ddi_root_node(), pm_adjust_timestamps, NULL);
/* after vfs_mountroot() so hrestime is valid */
post_startup();
swaploaded = 1;
/*
* Initialize Solaris Audit Subsystem
*/
audit_init();
/*
* Plumb the protocol modules and drivers only if we are not
* networked booted, in this case we already did it in rootconf().
*/
if (netboot == 0 && iscsiboot_prop == NULL)
(void) strplumb();
gethrestime(&PTOU(curproc)->u_start);
curthread->t_start = PTOU(curproc)->u_start.tv_sec;
p->p_mstart = gethrtime();
/*
* Perform setup functions that can only be done after root
* and swap have been set up.
*/
consconfig();
#ifndef __sparc
release_bootstrap();
#endif
/*
* attach drivers with ddi-forceattach prop
* It must be done early enough to load hotplug drivers (e.g.
* pcmcia nexus) so that devices enumerated via hotplug is
* available before I/O subsystem is fully initialized.
*/
i_ddi_forceattach_drivers();
/*
* Set the scan rate and other parameters of the paging subsystem.
*/
setupclock(0);
/*
* Initialize process 0's lwp directory and lwpid hash table.
*/
p->p_lwpdir = p->p_lwpfree = p0_lwpdir;
p->p_lwpdir->ld_next = p->p_lwpdir + 1;
p->p_lwpdir_sz = 2;
p->p_tidhash = p0_tidhash;
p->p_tidhash_sz = 2;
p0_lep.le_thread = curthread;
p0_lep.le_lwpid = curthread->t_tid;
p0_lep.le_start = curthread->t_start;
lwp_hash_in(p, &p0_lep, p0_tidhash, 2, 0);
/*
* Initialize extended accounting.
*/
exacct_init();
/*
* Initialize threads of sysevent event channels
*/
sysevent_evc_thrinit();
/*
* This must be done after post_startup() but before
* start_other_cpus()
*/
lgrp_init(LGRP_INIT_STAGE4);
/*
* Perform MP initialization, if any.
*/
start_other_cpus(0);
#ifdef __sparc
/*
* Release bootstrap here since PROM interfaces are
* used to start other CPUs above.
*/
release_bootstrap();
#endif
/*
* Finish lgrp initialization after all CPUS are brought online.
*/
lgrp_init(LGRP_INIT_STAGE5);
/*
* After mp_init(), number of cpus are known (this is
* true for the time being, when there are actually
* hot pluggable cpus then this scheme would not do).
* Any per cpu initialization is done here.
*/
kmem_mp_init();
vmem_update(NULL);
clock_tick_init_post();
for (initptr = &mp_init_tbl[0]; *initptr; initptr++)
(**initptr)();
/*
* These must be called after start_other_cpus
*/
pm_cfb_setup_intr();
#if defined(__x86)
fastboot_post_startup();
#endif
/*
* Make init process; enter scheduling loop with system process.
*
* Note that we manually assign the pids for these processes, for
* historical reasons. If more pre-assigned pids are needed,
* FAMOUS_PIDS will have to be updated.
*/
/* create init process */
if (newproc(start_init, NULL, defaultcid, 59, NULL,
FAMOUS_PID_INIT))
panic("main: unable to fork init.");
/* create pageout daemon */
if (newproc(pageout, NULL, syscid, maxclsyspri - 1, NULL,
FAMOUS_PID_PAGEOUT))
panic("main: unable to fork pageout()");
/* create fsflush daemon */
if (newproc(fsflush, NULL, syscid, minclsyspri, NULL,
FAMOUS_PID_FSFLUSH))
panic("main: unable to fork fsflush()");
/* create cluster process if we're a member of one */
if (cluster_bootflags & CLUSTER_BOOTED) {
if (newproc(cluster_wrapper, NULL, syscid, minclsyspri,
NULL, 0)) {
panic("main: unable to fork cluster()");
}
}
/*
* Create system threads (threads are associated with p0)
*/
/* create module uninstall daemon */
/* BugID 1132273. If swapping over NFS need a bigger stack */
(void) thread_create(NULL, 0, (void (*)())mod_uninstall_daemon,
NULL, 0, &p0, TS_RUN, minclsyspri);
(void) thread_create(NULL, 0, seg_pasync_thread,
NULL, 0, &p0, TS_RUN, minclsyspri);
pid_setmin();
/* system is now ready */
mutex_exit(&ualock);
bcopy("sched", PTOU(curproc)->u_psargs, 6);
bcopy("sched", PTOU(curproc)->u_comm, 5);
sched();
/* NOTREACHED */
}
static int
copen(int startfd, char *fname, int filemode, int createmode)
{
struct pathname pn;
vnode_t *vp, *sdvp;
file_t *fp, *startfp;
enum vtype type;
int error;
int fd, dupfd;
vnode_t *startvp;
proc_t *p = curproc;
uio_seg_t seg = UIO_USERSPACE;
char *open_filename = fname;
uint32_t auditing = AU_AUDITING();
char startchar;
if (filemode & (FSEARCH|FEXEC)) {
/*
* Must be one or the other and neither FREAD nor FWRITE
* Must not be any of FAPPEND FCREAT FTRUNC FXATTR FXATTRDIROPEN
* XXX: Should these just be silently ignored?
*/
if ((filemode & (FREAD|FWRITE)) ||
(filemode & (FSEARCH|FEXEC)) == (FSEARCH|FEXEC) ||
(filemode & (FAPPEND|FCREAT|FTRUNC|FXATTR|FXATTRDIROPEN)))
return (set_errno(EINVAL));
}
if (startfd == AT_FDCWD) {
/*
* Regular open()
*/
startvp = NULL;
} else {
/*
* We're here via openat()
*/
if (copyin(fname, &startchar, sizeof (char)))
return (set_errno(EFAULT));
/*
* if startchar is / then startfd is ignored
*/
if (startchar == '/')
startvp = NULL;
else {
if ((startfp = getf(startfd)) == NULL)
return (set_errno(EBADF));
startvp = startfp->f_vnode;
VN_HOLD(startvp);
releasef(startfd);
}
}
/*
* Handle __openattrdirat() requests
*/
if (filemode & FXATTRDIROPEN) {
if (auditing && startvp != NULL)
audit_setfsat_path(1);
if (error = lookupnameat(fname, seg, FOLLOW,
NULLVPP, &vp, startvp))
return (set_errno(error));
if (startvp != NULL)
VN_RELE(startvp);
startvp = vp;
}
/*
* Do we need to go into extended attribute space?
*/
if (filemode & FXATTR) {
if (startfd == AT_FDCWD) {
if (copyin(fname, &startchar, sizeof (char)))
return (set_errno(EFAULT));
/*
* If startchar == '/' then no extended attributes
* are looked up.
*/
if (startchar == '/') {
startvp = NULL;
} else {
mutex_enter(&p->p_lock);
startvp = PTOU(p)->u_cdir;
VN_HOLD(startvp);
mutex_exit(&p->p_lock);
}
}
/*
* Make sure we have a valid extended attribute request.
* We must either have a real fd or AT_FDCWD and a relative
* pathname.
*/
if (startvp == NULL) {
goto noxattr;
}
}
if (filemode & (FXATTR|FXATTRDIROPEN)) {
vattr_t vattr;
if (error = pn_get(fname, UIO_USERSPACE, &pn)) {
goto out;
}
/*
* In order to access hidden attribute directory the
* user must be able to stat() the file
*/
vattr.va_mask = AT_ALL;
if (error = VOP_GETATTR(startvp, &vattr, 0, CRED(), NULL)) {
pn_free(&pn);
goto out;
}
if ((startvp->v_vfsp->vfs_flag & VFS_XATTR) != 0 ||
vfs_has_feature(startvp->v_vfsp, VFSFT_SYSATTR_VIEWS)) {
error = VOP_LOOKUP(startvp, "", &sdvp, &pn,
(filemode & FXATTRDIROPEN) ? LOOKUP_XATTR :
LOOKUP_XATTR|CREATE_XATTR_DIR, rootvp, CRED(),
NULL, NULL, NULL);
} else {
error = EINVAL;
}
/*
* For __openattrdirat() use "." as filename to open
* as part of vn_openat()
*/
if (error == 0 && (filemode & FXATTRDIROPEN)) {
open_filename = ".";
seg = UIO_SYSSPACE;
}
pn_free(&pn);
if (error != 0)
goto out;
VN_RELE(startvp);
startvp = sdvp;
}
noxattr:
if ((filemode & (FREAD|FWRITE|FSEARCH|FEXEC|FXATTRDIROPEN)) != 0) {
if ((filemode & (FNONBLOCK|FNDELAY)) == (FNONBLOCK|FNDELAY))
filemode &= ~FNDELAY;
error = falloc((vnode_t *)NULL, filemode, &fp, &fd);
if (error == 0) {
if (auditing && startvp != NULL)
audit_setfsat_path(1);
/*
* Last arg is a don't-care term if
* !(filemode & FCREAT).
*/
error = vn_openat(open_filename, seg, filemode,
(int)(createmode & MODEMASK),
&vp, CRCREAT, PTOU(curproc)->u_cmask,
startvp, fd);
if (startvp != NULL)
VN_RELE(startvp);
if (error == 0) {
if ((vp->v_flag & VDUP) == 0) {
fp->f_vnode = vp;
mutex_exit(&fp->f_tlock);
/*
* We must now fill in the slot
* falloc reserved.
*/
setf(fd, fp);
return (fd);
} else {
/*
* Special handling for /dev/fd.
* Give up the file pointer
* and dup the indicated file descriptor
* (in v_rdev). This is ugly, but I've
* seen worse.
*/
unfalloc(fp);
dupfd = getminor(vp->v_rdev);
type = vp->v_type;
mutex_enter(&vp->v_lock);
vp->v_flag &= ~VDUP;
mutex_exit(&vp->v_lock);
VN_RELE(vp);
if (type != VCHR)
return (set_errno(EINVAL));
if ((fp = getf(dupfd)) == NULL) {
setf(fd, NULL);
return (set_errno(EBADF));
}
mutex_enter(&fp->f_tlock);
fp->f_count++;
mutex_exit(&fp->f_tlock);
setf(fd, fp);
releasef(dupfd);
}
return (fd);
} else {
setf(fd, NULL);
unfalloc(fp);
return (set_errno(error));
}
}
} else {
error = EINVAL;
}
out:
if (startvp != NULL)
VN_RELE(startvp);
return (set_errno(error));
}
/*
* setppriv (priv_op_t, priv_ptype_t, priv_set_t)
*/
static int
setppriv(priv_op_t op, priv_ptype_t type, priv_set_t *in_pset)
{
priv_set_t pset, *target;
cred_t *cr, *pcr;
proc_t *p;
boolean_t donocd = B_FALSE;
if (!PRIV_VALIDSET(type) || !PRIV_VALIDOP(op))
return (set_errno(EINVAL));
if (copyin(in_pset, &pset, sizeof (priv_set_t)))
return (set_errno(EFAULT));
p = ttoproc(curthread);
cr = cralloc();
mutex_enter(&p->p_crlock);
retry:
pcr = p->p_cred;
if (AU_AUDITING())
audit_setppriv(op, type, &pset, pcr);
/*
* Filter out unallowed request (bad op and bad type)
*/
switch (op) {
case PRIV_ON:
case PRIV_SET:
/*
* Turning on privileges; the limit set cannot grow,
* other sets can but only as long as they remain subsets
* of P. Only immediately after exec holds that P <= L.
*/
if (type == PRIV_LIMIT &&
!priv_issubset(&pset, &CR_LPRIV(pcr))) {
mutex_exit(&p->p_crlock);
crfree(cr);
return (set_errno(EPERM));
}
if (!priv_issubset(&pset, &CR_OPPRIV(pcr)) &&
!priv_issubset(&pset, priv_getset(pcr, type))) {
mutex_exit(&p->p_crlock);
/* Policy override should not grow beyond L either */
if (type != PRIV_INHERITABLE ||
!priv_issubset(&pset, &CR_LPRIV(pcr)) ||
secpolicy_require_privs(CRED(), &pset) != 0) {
crfree(cr);
return (set_errno(EPERM));
}
mutex_enter(&p->p_crlock);
if (pcr != p->p_cred)
goto retry;
donocd = B_TRUE;
}
break;
case PRIV_OFF:
/* PRIV_OFF is always allowed */
break;
}
/*
* OK! everything is cool.
* Do cred COW.
*/
crcopy_to(pcr, cr);
/*
* If we change the effective, permitted or limit set, we attain
* "privilege awareness".
*/
if (type != PRIV_INHERITABLE)
priv_set_PA(cr);
target = &(CR_PRIVS(cr)->crprivs[type]);
switch (op) {
case PRIV_ON:
priv_union(&pset, target);
break;
case PRIV_OFF:
priv_inverse(&pset);
priv_intersect(target, &pset);
/*
* Fall-thru to set target and change other process
* privilege sets.
*/
/*FALLTHRU*/
case PRIV_SET:
*target = pset;
/*
* Take privileges no longer permitted out
* of other effective sets as well.
* Limit set is enforced at exec() time.
*/
if (type == PRIV_PERMITTED)
priv_intersect(&pset, &CR_EPRIV(cr));
break;
}
/*
* When we give up privileges not in the inheritable set,
* set SNOCD if not already set; first we compute the
* privileges removed from P using Diff = (~P') & P
* and then we check whether the removed privileges are
* a subset of I. If we retain uid 0, all privileges
* are required anyway so don't set SNOCD.
*/
if (type == PRIV_PERMITTED && (p->p_flag & SNOCD) == 0 &&
cr->cr_uid != 0 && cr->cr_ruid != 0 && cr->cr_suid != 0) {
priv_set_t diff = CR_OPPRIV(cr);
priv_inverse(&diff);
priv_intersect(&CR_OPPRIV(pcr), &diff);
donocd = !priv_issubset(&diff, &CR_IPRIV(cr));
}
p->p_cred = cr;
mutex_exit(&p->p_crlock);
if (donocd) {
mutex_enter(&p->p_lock);
p->p_flag |= SNOCD;
mutex_exit(&p->p_lock);
}
/*
* The basic_test privilege should not be removed from E;
* if that has happened, then some programmer typically set the E/P to
* empty. That is not portable.
*/
if ((type == PRIV_EFFECTIVE || type == PRIV_PERMITTED) &&
priv_basic_test >= 0 && !PRIV_ISASSERT(target, priv_basic_test)) {
proc_t *p = curproc;
pid_t pid = p->p_pid;
char *fn = PTOU(p)->u_comm;
cmn_err(CE_WARN, "%s[%d]: setppriv: basic_test privilege "
"removed from E/P", fn, pid);
}
crset(p, cr); /* broadcast to process threads */
return (0);
}
int
dogetcwd(char *buf, size_t buflen)
{
int ret;
vnode_t *vp;
vnode_t *compvp;
refstr_t *cwd, *oldcwd;
const char *value;
pathname_t rpnp, pnp;
proc_t *p = curproc;
/*
* Check to see if there is a cached version of the cwd. If so, lookup
* the cached value and make sure it is the same vnode.
*/
mutex_enter(&p->p_lock);
if ((cwd = PTOU(p)->u_cwd) != NULL)
refstr_hold(cwd);
vp = PTOU(p)->u_cdir;
VN_HOLD(vp);
mutex_exit(&p->p_lock);
/*
* Make sure we have permission to access the current directory.
*/
if ((ret = VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) != 0) {
if (cwd != NULL)
refstr_rele(cwd);
VN_RELE(vp);
return (ret);
}
if (cwd) {
value = refstr_value(cwd);
if ((ret = pn_get((char *)value, UIO_SYSSPACE, &pnp)) != 0) {
refstr_rele(cwd);
VN_RELE(vp);
return (ret);
}
pn_alloc(&rpnp);
if (lookuppn(&pnp, &rpnp, NO_FOLLOW, NULL, &compvp) == 0) {
if (VN_CMP(vp, compvp) &&
strcmp(value, rpnp.pn_path) == 0) {
VN_RELE(compvp);
VN_RELE(vp);
pn_free(&pnp);
pn_free(&rpnp);
if (strlen(value) + 1 > buflen) {
refstr_rele(cwd);
return (ENAMETOOLONG);
}
bcopy(value, buf, strlen(value) + 1);
refstr_rele(cwd);
return (0);
}
VN_RELE(compvp);
}
pn_free(&rpnp);
pn_free(&pnp);
refstr_rele(cwd);
}
ret = vnodetopath_common(NULL, vp, buf, buflen, CRED(),
LOOKUP_CHECKREAD);
VN_RELE(vp);
/*
* Store the new cwd and replace the existing cached copy.
*/
if (ret == 0)
cwd = refstr_alloc(buf);
else
cwd = NULL;
mutex_enter(&p->p_lock);
oldcwd = PTOU(p)->u_cwd;
PTOU(p)->u_cwd = cwd;
mutex_exit(&p->p_lock);
if (oldcwd)
refstr_rele(oldcwd);
return (ret);
}
int
kadmin(int cmd, int fcn, void *mdep, cred_t *credp)
{
int error = 0;
char *buf;
size_t buflen = 0;
boolean_t invoke_cb = B_FALSE;
/*
* We might be called directly by the kernel's fault-handling code, so
* we can't assert that the caller is in the global zone.
*/
/*
* Make sure that cmd is one of the valid <sys/uadmin.h> command codes
* and that we have appropriate privileges for this action.
*/
switch (cmd) {
case A_FTRACE:
case A_SHUTDOWN:
case A_REBOOT:
case A_REMOUNT:
case A_FREEZE:
case A_DUMP:
case A_SDTTEST:
case A_CONFIG:
if (secpolicy_sys_config(credp, B_FALSE) != 0)
return (EPERM);
break;
default:
return (EINVAL);
}
/*
* Serialize these operations on ualock. If it is held, the
* system should shutdown, reboot, or remount shortly, unless there is
* an error. We need a cv rather than just a mutex because proper
* functioning of A_REBOOT relies on being able to interrupt blocked
* userland callers.
*
* We only clear ua_shutdown_thread after A_REMOUNT or A_CONFIG.
* Other commands should never return.
*/
if (cmd == A_SHUTDOWN || cmd == A_REBOOT || cmd == A_REMOUNT ||
cmd == A_CONFIG) {
mutex_enter(&ualock);
while (ua_shutdown_thread != NULL) {
if (cv_wait_sig(&uacond, &ualock) == 0) {
/*
* If we were interrupted, leave, and handle
* the signal (or exit, depending on what
* happened)
*/
mutex_exit(&ualock);
return (EINTR);
}
}
ua_shutdown_thread = curthread;
mutex_exit(&ualock);
}
switch (cmd) {
case A_SHUTDOWN:
{
proc_t *p = ttoproc(curthread);
/*
* Release (almost) all of our own resources if we are called
* from a user context, however if we are calling kadmin() from
* a kernel context then we do not release these resources.
*/
if (p != &p0) {
proc_is_exiting(p);
if ((error = exitlwps(0)) != 0) {
/*
* Another thread in this process also called
* exitlwps().
*/
mutex_enter(&ualock);
ua_shutdown_thread = NULL;
cv_signal(&uacond);
mutex_exit(&ualock);
return (error);
}
mutex_enter(&p->p_lock);
p->p_flag |= SNOWAIT;
sigfillset(&p->p_ignore);
curthread->t_lwp->lwp_cursig = 0;
curthread->t_lwp->lwp_extsig = 0;
if (p->p_exec) {
vnode_t *exec_vp = p->p_exec;
p->p_exec = NULLVP;
mutex_exit(&p->p_lock);
VN_RELE(exec_vp);
} else {
mutex_exit(&p->p_lock);
}
pollcleanup();
closeall(P_FINFO(curproc));
relvm();
} else {
/*
* Reset t_cred if not set because much of the
* filesystem code depends on CRED() being valid.
*/
if (curthread->t_cred == NULL)
curthread->t_cred = kcred;
}
/* indicate shutdown in progress */
sys_shutdown = 1;
/*
* Communcate that init shouldn't be restarted.
*/
zone_shutdown_global();
killall(ALL_ZONES);
/*
* If we are calling kadmin() from a kernel context then we
* do not release these resources.
*/
if (ttoproc(curthread) != &p0) {
VN_RELE(PTOU(curproc)->u_cdir);
if (PTOU(curproc)->u_rdir)
VN_RELE(PTOU(curproc)->u_rdir);
if (PTOU(curproc)->u_cwd)
refstr_rele(PTOU(curproc)->u_cwd);
PTOU(curproc)->u_cdir = rootdir;
PTOU(curproc)->u_rdir = NULL;
PTOU(curproc)->u_cwd = NULL;
}
/*
* Allow the reboot/halt/poweroff code a chance to do
* anything it needs to whilst we still have filesystems
* mounted, like loading any modules necessary for later
* performing the actual poweroff.
*/
if ((mdep != NULL) && (*(char *)mdep == '/')) {
buf = i_convert_boot_device_name(mdep, NULL, &buflen);
mdpreboot(cmd, fcn, buf);
} else
mdpreboot(cmd, fcn, mdep);
/*
* Allow fsflush to finish running and then prevent it
* from ever running again so that vfs_unmountall() and
* vfs_syncall() can acquire the vfs locks they need.
*/
sema_p(&fsflush_sema);
(void) callb_execute_class(CB_CL_UADMIN_PRE_VFS, NULL);
vfs_unmountall();
(void) VFS_MOUNTROOT(rootvfs, ROOT_UNMOUNT);
vfs_syncall();
dump_ereports();
dump_messages();
invoke_cb = B_TRUE;
/* FALLTHROUGH */
}
case A_REBOOT:
if ((mdep != NULL) && (*(char *)mdep == '/')) {
buf = i_convert_boot_device_name(mdep, NULL, &buflen);
mdboot(cmd, fcn, buf, invoke_cb);
} else
mdboot(cmd, fcn, mdep, invoke_cb);
/* no return expected */
break;
case A_CONFIG:
switch (fcn) {
case AD_UPDATE_BOOT_CONFIG:
#ifndef __sparc
{
extern void fastboot_update_config(const char *);
fastboot_update_config(mdep);
}
#endif
break;
}
/* Let other threads enter the shutdown path now */
mutex_enter(&ualock);
ua_shutdown_thread = NULL;
cv_signal(&uacond);
mutex_exit(&ualock);
break;
case A_REMOUNT:
(void) VFS_MOUNTROOT(rootvfs, ROOT_REMOUNT);
/* Let other threads enter the shutdown path now */
mutex_enter(&ualock);
ua_shutdown_thread = NULL;
cv_signal(&uacond);
mutex_exit(&ualock);
break;
case A_FREEZE:
{
/*
* This is the entrypoint for all suspend/resume actions.
*/
extern int cpr(int, void *);
if (modload("misc", "cpr") == -1)
return (ENOTSUP);
/* Let the CPR module decide what to do with mdep */
error = cpr(fcn, mdep);
break;
}
case A_FTRACE:
{
switch (fcn) {
case AD_FTRACE_START:
(void) FTRACE_START();
break;
case AD_FTRACE_STOP:
(void) FTRACE_STOP();
break;
default:
error = EINVAL;
}
break;
}
case A_DUMP:
{
if (fcn == AD_NOSYNC) {
in_sync = 1;
break;
}
panic_bootfcn = fcn;
panic_forced = 1;
if ((mdep != NULL) && (*(char *)mdep == '/')) {
panic_bootstr = i_convert_boot_device_name(mdep,
NULL, &buflen);
} else
panic_bootstr = mdep;
#ifndef __sparc
extern void fastboot_update_and_load(int, char *);
fastboot_update_and_load(fcn, mdep);
#endif
panic("forced crash dump initiated at user request");
/*NOTREACHED*/
}
case A_SDTTEST:
{
DTRACE_PROBE7(test, int, 1, int, 2, int, 3, int, 4, int, 5,
int, 6, int, 7);
break;
}
default:
error = EINVAL;
}
return (error);
}
/*
* Remove zombie children from the process table.
*/
void
freeproc(proc_t *p)
{
proc_t *q;
task_t *tk;
ASSERT(p->p_stat == SZOMB);
ASSERT(p->p_tlist == NULL);
ASSERT(MUTEX_HELD(&pidlock));
sigdelq(p, NULL, 0);
if (p->p_killsqp) {
siginfofree(p->p_killsqp);
p->p_killsqp = NULL;
}
prfree(p); /* inform /proc */
/*
* Don't free the init processes.
* Other dying processes will access it.
*/
if (p == proc_init)
return;
/*
* We wait until now to free the cred structure because a
* zombie process's credentials may be examined by /proc.
* No cred locking needed because there are no threads at this point.
*/
upcount_dec(crgetruid(p->p_cred), crgetzoneid(p->p_cred));
crfree(p->p_cred);
if (p->p_corefile != NULL) {
corectl_path_rele(p->p_corefile);
p->p_corefile = NULL;
}
if (p->p_content != NULL) {
corectl_content_rele(p->p_content);
p->p_content = NULL;
}
if (p->p_nextofkin && !((p->p_nextofkin->p_flag & SNOWAIT) ||
(PTOU(p->p_nextofkin)->u_signal[SIGCLD - 1] == SIG_IGN))) {
/*
* This should still do the right thing since p_utime/stime
* get set to the correct value on process exit, so it
* should get properly updated
*/
p->p_nextofkin->p_cutime += p->p_utime;
p->p_nextofkin->p_cstime += p->p_stime;
p->p_nextofkin->p_cacct[LMS_USER] += p->p_acct[LMS_USER];
p->p_nextofkin->p_cacct[LMS_SYSTEM] += p->p_acct[LMS_SYSTEM];
p->p_nextofkin->p_cacct[LMS_TRAP] += p->p_acct[LMS_TRAP];
p->p_nextofkin->p_cacct[LMS_TFAULT] += p->p_acct[LMS_TFAULT];
p->p_nextofkin->p_cacct[LMS_DFAULT] += p->p_acct[LMS_DFAULT];
p->p_nextofkin->p_cacct[LMS_KFAULT] += p->p_acct[LMS_KFAULT];
p->p_nextofkin->p_cacct[LMS_USER_LOCK]
+= p->p_acct[LMS_USER_LOCK];
p->p_nextofkin->p_cacct[LMS_SLEEP] += p->p_acct[LMS_SLEEP];
p->p_nextofkin->p_cacct[LMS_WAIT_CPU]
+= p->p_acct[LMS_WAIT_CPU];
p->p_nextofkin->p_cacct[LMS_STOPPED] += p->p_acct[LMS_STOPPED];
p->p_nextofkin->p_cru.minflt += p->p_ru.minflt;
p->p_nextofkin->p_cru.majflt += p->p_ru.majflt;
p->p_nextofkin->p_cru.nswap += p->p_ru.nswap;
p->p_nextofkin->p_cru.inblock += p->p_ru.inblock;
p->p_nextofkin->p_cru.oublock += p->p_ru.oublock;
p->p_nextofkin->p_cru.msgsnd += p->p_ru.msgsnd;
p->p_nextofkin->p_cru.msgrcv += p->p_ru.msgrcv;
p->p_nextofkin->p_cru.nsignals += p->p_ru.nsignals;
p->p_nextofkin->p_cru.nvcsw += p->p_ru.nvcsw;
p->p_nextofkin->p_cru.nivcsw += p->p_ru.nivcsw;
p->p_nextofkin->p_cru.sysc += p->p_ru.sysc;
p->p_nextofkin->p_cru.ioch += p->p_ru.ioch;
}
q = p->p_nextofkin;
if (q && q->p_orphan == p)
q->p_orphan = p->p_nextorph;
else if (q) {
for (q = q->p_orphan; q; q = q->p_nextorph)
if (q->p_nextorph == p)
break;
ASSERT(q && q->p_nextorph == p);
q->p_nextorph = p->p_nextorph;
}
/*
* The process table slot is being freed, so it is now safe to give up
* task and project membership.
*/
mutex_enter(&p->p_lock);
tk = p->p_task;
task_detach(p);
mutex_exit(&p->p_lock);
proc_detach(p);
pid_exit(p, tk); /* frees pid and proc structure */
task_rele(tk);
}
int
sigaction(int sig, struct sigaction *actp, struct sigaction *oactp)
{
struct sigaction act;
struct sigaction oact;
k_sigset_t set;
proc_t *p;
user_t *ua;
int sigcld_look = 0;
if (sig <= 0 || sig >= NSIG ||
(actp != NULL && sigismember(&cantmask, sig)))
return (set_errno(EINVAL));
/*
* act and oact might be the same address, so copyin act first.
*/
if (actp) {
#if defined(__sparc)
void (*handler)();
#endif
if (copyin(actp, &act, sizeof (act)))
return (set_errno(EFAULT));
#if defined(__sparc)
/*
* Check alignment of handler
*/
handler = act.sa_handler;
if (handler != SIG_IGN && handler != SIG_DFL &&
((uintptr_t)handler & 0x3) != 0)
return (set_errno(EINVAL));
#endif
}
p = curproc;
ua = PTOU(p);
mutex_enter(&p->p_lock);
if (oactp) {
int flags;
void (*disp)();
disp = ua->u_signal[sig - 1];
flags = 0;
if (disp != SIG_DFL && disp != SIG_IGN) {
set = ua->u_sigmask[sig-1];
if (sigismember(&p->p_siginfo, sig))
flags |= SA_SIGINFO;
if (sigismember(&ua->u_sigrestart, sig))
flags |= SA_RESTART;
if (sigismember(&ua->u_sigonstack, sig))
flags |= SA_ONSTACK;
if (sigismember(&ua->u_sigresethand, sig))
flags |= SA_RESETHAND;
if (sigismember(&ua->u_signodefer, sig))
flags |= SA_NODEFER;
} else
sigemptyset(&set);
if (sig == SIGCLD) {
if (p->p_flag & SNOWAIT)
flags |= SA_NOCLDWAIT;
if (!(p->p_flag & SJCTL))
flags |= SA_NOCLDSTOP;
}
oact.sa_handler = disp;
oact.sa_flags = flags;
sigktou(&set, &oact.sa_mask);
}
if (actp) {
if (sig == SIGCLD)
sigcld_look = 1;
sigutok(&act.sa_mask, &set);
setsigact(sig, act.sa_handler, &set, act.sa_flags);
}
mutex_exit(&p->p_lock);
if (sigcld_look)
sigcld_repost();
if (oactp &&
copyout(&oact, oactp, sizeof (oact)))
return (set_errno(EFAULT));
return (0);
}
/*
* Called by proc_exit() when a zone's init exits, presumably because
* it failed. As long as the given zone is still in the "running"
* state, we will re-exec() init, but first we need to reset things
* which are usually inherited across exec() but will break init's
* assumption that it is being exec()'d from a virgin process. Most
* importantly this includes closing all file descriptors (exec only
* closes those marked close-on-exec) and resetting signals (exec only
* resets handled signals, and we need to clear any signals which
* killed init). Anything else that exec(2) says would be inherited,
* but would affect the execution of init, needs to be reset.
*/
static int
restart_init(int what, int why)
{
kthread_t *t = curthread;
klwp_t *lwp = ttolwp(t);
proc_t *p = ttoproc(t);
user_t *up = PTOU(p);
vnode_t *oldcd, *oldrd;
int i, err;
char reason_buf[64];
/*
* Let zone admin (and global zone admin if this is for a non-global
* zone) know that init has failed and will be restarted.
*/
zcmn_err(p->p_zone->zone_id, CE_WARN,
"init(1M) %s: restarting automatically",
exit_reason(reason_buf, sizeof (reason_buf), what, why));
if (!INGLOBALZONE(p)) {
cmn_err(CE_WARN, "init(1M) for zone %s (pid %d) %s: "
"restarting automatically",
p->p_zone->zone_name, p->p_pid, reason_buf);
}
/*
* Remove any fpollinfo_t's for this (last) thread from our file
* descriptors so closeall() can ASSERT() that they're all gone.
* Then close all open file descriptors in the process.
*/
pollcleanup();
closeall(P_FINFO(p));
/*
* Grab p_lock and begin clearing miscellaneous global process
* state that needs to be reset before we exec the new init(1M).
*/
mutex_enter(&p->p_lock);
prbarrier(p);
p->p_flag &= ~(SKILLED | SEXTKILLED | SEXITING | SDOCORE);
up->u_cmask = CMASK;
sigemptyset(&t->t_hold);
sigemptyset(&t->t_sig);
sigemptyset(&t->t_extsig);
sigemptyset(&p->p_sig);
sigemptyset(&p->p_extsig);
sigdelq(p, t, 0);
sigdelq(p, NULL, 0);
if (p->p_killsqp) {
siginfofree(p->p_killsqp);
p->p_killsqp = NULL;
}
/*
* Reset any signals that are ignored back to the default disposition.
* Other u_signal members will be cleared when exec calls sigdefault().
*/
for (i = 1; i < NSIG; i++) {
if (up->u_signal[i - 1] == SIG_IGN) {
up->u_signal[i - 1] = SIG_DFL;
sigemptyset(&up->u_sigmask[i - 1]);
}
}
/*
* Clear the current signal, any signal info associated with it, and
* any signal information from contracts and/or contract templates.
*/
lwp->lwp_cursig = 0;
lwp->lwp_extsig = 0;
if (lwp->lwp_curinfo != NULL) {
siginfofree(lwp->lwp_curinfo);
lwp->lwp_curinfo = NULL;
}
lwp_ctmpl_clear(lwp);
/*
* Reset both the process root directory and the current working
* directory to the root of the zone just as we do during boot.
*/
VN_HOLD(p->p_zone->zone_rootvp);
oldrd = up->u_rdir;
up->u_rdir = p->p_zone->zone_rootvp;
VN_HOLD(p->p_zone->zone_rootvp);
oldcd = up->u_cdir;
up->u_cdir = p->p_zone->zone_rootvp;
if (up->u_cwd != NULL) {
refstr_rele(up->u_cwd);
up->u_cwd = NULL;
}
mutex_exit(&p->p_lock);
if (oldrd != NULL)
VN_RELE(oldrd);
if (oldcd != NULL)
VN_RELE(oldcd);
/* Free the controlling tty. (freectty() always assumes curproc.) */
ASSERT(p == curproc);
(void) freectty(B_TRUE);
/*
* Now exec() the new init(1M) on top of the current process. If we
* succeed, the caller will treat this like a successful system call.
* If we fail, we issue messages and the caller will proceed with exit.
*/
err = exec_init(p->p_zone->zone_initname, NULL);
if (err == 0)
return (0);
zcmn_err(p->p_zone->zone_id, CE_WARN,
"failed to restart init(1M) (err=%d): system reboot required", err);
if (!INGLOBALZONE(p)) {
cmn_err(CE_WARN, "failed to restart init(1M) for zone %s "
"(pid %d, err=%d): zoneadm(1M) boot required",
p->p_zone->zone_name, p->p_pid, err);
}
return (-1);
}
int
sendsig(int sig, k_siginfo_t *sip, void (*hdlr)())
{
volatile int minstacksz;
int newstack;
label_t ljb;
volatile caddr_t sp;
caddr_t fp;
struct regs *rp;
volatile greg_t upc;
volatile proc_t *p = ttoproc(curthread);
klwp_t *lwp = ttolwp(curthread);
ucontext_t *volatile tuc = NULL;
ucontext_t *uc;
siginfo_t *sip_addr;
volatile int watched;
rp = lwptoregs(lwp);
upc = rp->r_pc;
minstacksz = SA(sizeof (struct sigframe)) + SA(sizeof (*uc));
if (sip != NULL)
minstacksz += SA(sizeof (siginfo_t));
ASSERT((minstacksz & (STACK_ALIGN - 1ul)) == 0);
/*
* Figure out whether we will be handling this signal on
* an alternate stack specified by the user. Then allocate
* and validate the stack requirements for the signal handler
* context. on_fault will catch any faults.
*/
newstack = sigismember(&PTOU(curproc)->u_sigonstack, sig) &&
!(lwp->lwp_sigaltstack.ss_flags & (SS_ONSTACK|SS_DISABLE));
if (newstack) {
fp = (caddr_t)(SA((uintptr_t)lwp->lwp_sigaltstack.ss_sp) +
SA(lwp->lwp_sigaltstack.ss_size) - STACK_ALIGN);
} else if ((rp->r_ss & 0xffff) != UDS_SEL) {
user_desc_t *ldt;
/*
* If the stack segment selector is -not- pointing at
* the UDS_SEL descriptor and we have an LDT entry for
* it instead, add the base address to find the effective va.
*/
if ((ldt = p->p_ldt) != NULL)
fp = (caddr_t)rp->r_sp +
USEGD_GETBASE(&ldt[SELTOIDX(rp->r_ss)]);
else
fp = (caddr_t)rp->r_sp;
} else
fp = (caddr_t)rp->r_sp;
/*
* Force proper stack pointer alignment, even in the face of a
* misaligned stack pointer from user-level before the signal.
* Don't use the SA() macro because that rounds up, not down.
*/
fp = (caddr_t)((uintptr_t)fp & ~(STACK_ALIGN - 1ul));
sp = fp - minstacksz;
/*
* Make sure lwp hasn't trashed its stack.
*/
if (sp >= (caddr_t)USERLIMIT || fp >= (caddr_t)USERLIMIT) {
#ifdef DEBUG
printf("sendsig: bad signal stack cmd=%s, pid=%d, sig=%d\n",
PTOU(p)->u_comm, p->p_pid, sig);
printf("sigsp = 0x%p, action = 0x%p, upc = 0x%lx\n",
(void *)sp, (void *)hdlr, (uintptr_t)upc);
printf("sp above USERLIMIT\n");
#endif
return (0);
}
watched = watch_disable_addr((caddr_t)sp, minstacksz, S_WRITE);
if (on_fault(&ljb))
goto badstack;
if (sip != NULL) {
zoneid_t zoneid;
fp -= SA(sizeof (siginfo_t));
uzero(fp, sizeof (siginfo_t));
if (SI_FROMUSER(sip) &&
(zoneid = p->p_zone->zone_id) != GLOBAL_ZONEID &&
zoneid != sip->si_zoneid) {
k_siginfo_t sani_sip = *sip;
sani_sip.si_pid = p->p_zone->zone_zsched->p_pid;
sani_sip.si_uid = 0;
sani_sip.si_ctid = -1;
sani_sip.si_zoneid = zoneid;
copyout_noerr(&sani_sip, fp, sizeof (sani_sip));
} else
copyout_noerr(sip, fp, sizeof (*sip));
sip_addr = (siginfo_t *)fp;
if (sig == SIGPROF &&
curthread->t_rprof != NULL &&
curthread->t_rprof->rp_anystate) {
/*
* We stand on our head to deal with
* the real time profiling signal.
* Fill in the stuff that doesn't fit
* in a normal k_siginfo structure.
*/
int i = sip->si_nsysarg;
while (--i >= 0)
suword32_noerr(&(sip_addr->si_sysarg[i]),
(uint32_t)lwp->lwp_arg[i]);
copyout_noerr(curthread->t_rprof->rp_state,
sip_addr->si_mstate,
sizeof (curthread->t_rprof->rp_state));
}
} else
sip_addr = NULL;
/* save the current context on the user stack */
fp -= SA(sizeof (*tuc));
uc = (ucontext_t *)fp;
tuc = kmem_alloc(sizeof (*tuc), KM_SLEEP);
savecontext(tuc, &lwp->lwp_sigoldmask);
copyout_noerr(tuc, uc, sizeof (*tuc));
kmem_free(tuc, sizeof (*tuc));
tuc = NULL;
lwp->lwp_oldcontext = (uintptr_t)uc;
if (newstack) {
lwp->lwp_sigaltstack.ss_flags |= SS_ONSTACK;
if (lwp->lwp_ustack)
copyout_noerr(&lwp->lwp_sigaltstack,
(stack_t *)lwp->lwp_ustack, sizeof (stack_t));
}
/*
* Set up signal handler arguments
*/
{
struct sigframe frame;
frame.sip = sip_addr;
frame.ucp = uc;
frame.signo = sig;
frame.retaddr = (void (*)())0xffffffff; /* never return! */
copyout_noerr(&frame, sp, sizeof (frame));
}
no_fault();
if (watched)
watch_enable_addr((caddr_t)sp, minstacksz, S_WRITE);
rp->r_sp = (greg_t)sp;
rp->r_pc = (greg_t)hdlr;
rp->r_ps = PSL_USER | (rp->r_ps & PS_IOPL);
if ((rp->r_cs & 0xffff) != UCS_SEL ||
(rp->r_ss & 0xffff) != UDS_SEL) {
rp->r_cs = UCS_SEL;
rp->r_ss = UDS_SEL;
}
/*
* Don't set lwp_eosys here. sendsig() is called via psig() after
* lwp_eosys is handled, so setting it here would affect the next
* system call.
*/
return (1);
badstack:
no_fault();
if (watched)
watch_enable_addr((caddr_t)sp, minstacksz, S_WRITE);
if (tuc)
kmem_free(tuc, sizeof (*tuc));
#ifdef DEBUG
printf("sendsig: bad signal stack cmd=%s, pid=%d, sig=%d\n",
PTOU(p)->u_comm, p->p_pid, sig);
printf("on fault, sigsp = 0x%p, action = 0x%p, upc = 0x%lx\n",
(void *)sp, (void *)hdlr, (uintptr_t)upc);
#endif
return (0);
}
/*
* Return value:
* 1 - exitlwps() failed, call (or continue) lwp_exit()
* 0 - restarting init. Return through system call path
*/
int
proc_exit(int why, int what)
{
kthread_t *t = curthread;
klwp_t *lwp = ttolwp(t);
proc_t *p = ttoproc(t);
zone_t *z = p->p_zone;
timeout_id_t tmp_id;
int rv;
proc_t *q;
task_t *tk;
vnode_t *exec_vp, *execdir_vp, *cdir, *rdir;
sigqueue_t *sqp;
lwpdir_t *lwpdir;
uint_t lwpdir_sz;
tidhash_t *tidhash;
uint_t tidhash_sz;
ret_tidhash_t *ret_tidhash;
refstr_t *cwd;
hrtime_t hrutime, hrstime;
int evaporate;
/*
* Stop and discard the process's lwps except for the current one,
* unless some other lwp beat us to it. If exitlwps() fails then
* return and the calling lwp will call (or continue in) lwp_exit().
*/
proc_is_exiting(p);
if (exitlwps(0) != 0)
return (1);
mutex_enter(&p->p_lock);
if (p->p_ttime > 0) {
/*
* Account any remaining ticks charged to this process
* on its way out.
*/
(void) task_cpu_time_incr(p->p_task, p->p_ttime);
p->p_ttime = 0;
}
mutex_exit(&p->p_lock);
DTRACE_PROC(lwp__exit);
DTRACE_PROC1(exit, int, why);
/*
* Will perform any brand specific proc exit processing, since this
* is always the last lwp, will also perform lwp_exit and free brand
* data
*/
if (PROC_IS_BRANDED(p)) {
lwp_detach_brand_hdlrs(lwp);
brand_clearbrand(p, B_FALSE);
}
/*
* Don't let init exit unless zone_start_init() failed its exec, or
* we are shutting down the zone or the machine.
*
* Since we are single threaded, we don't need to lock the
* following accesses to zone_proc_initpid.
*/
if (p->p_pid == z->zone_proc_initpid) {
if (z->zone_boot_err == 0 &&
zone_status_get(z) < ZONE_IS_SHUTTING_DOWN &&
zone_status_get(global_zone) < ZONE_IS_SHUTTING_DOWN &&
z->zone_restart_init == B_TRUE &&
restart_init(what, why) == 0)
return (0);
/*
* Since we didn't or couldn't restart init, we clear
* the zone's init state and proceed with exit
* processing.
*/
z->zone_proc_initpid = -1;
}
lwp_pcb_exit();
/*
* Allocate a sigqueue now, before we grab locks.
* It will be given to sigcld(), below.
* Special case: If we will be making the process disappear
* without a trace because it is either:
* * an exiting SSYS process, or
* * a posix_spawn() vfork child who requests it,
* we don't bother to allocate a useless sigqueue.
*/
evaporate = (p->p_flag & SSYS) || ((p->p_flag & SVFORK) &&
why == CLD_EXITED && what == _EVAPORATE);
if (!evaporate)
sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP);
/*
* revoke any doors created by the process.
*/
if (p->p_door_list)
door_exit();
/*
* Release schedctl data structures.
*/
if (p->p_pagep)
schedctl_proc_cleanup();
/*
* make sure all pending kaio has completed.
*/
if (p->p_aio)
aio_cleanup_exit();
/*
* discard the lwpchan cache.
*/
if (p->p_lcp != NULL)
lwpchan_destroy_cache(0);
/*
* Clean up any DTrace helper actions or probes for the process.
*/
if (p->p_dtrace_helpers != NULL) {
ASSERT(dtrace_helpers_cleanup != NULL);
(*dtrace_helpers_cleanup)();
}
/* untimeout the realtime timers */
if (p->p_itimer != NULL)
timer_exit();
if ((tmp_id = p->p_alarmid) != 0) {
p->p_alarmid = 0;
(void) untimeout(tmp_id);
}
/*
* Remove any fpollinfo_t's for this (last) thread from our file
* descriptors so closeall() can ASSERT() that they're all gone.
*/
pollcleanup();
if (p->p_rprof_cyclic != CYCLIC_NONE) {
mutex_enter(&cpu_lock);
cyclic_remove(p->p_rprof_cyclic);
mutex_exit(&cpu_lock);
}
mutex_enter(&p->p_lock);
/*
* Clean up any DTrace probes associated with this process.
*/
if (p->p_dtrace_probes) {
ASSERT(dtrace_fasttrap_exit_ptr != NULL);
dtrace_fasttrap_exit_ptr(p);
}
while ((tmp_id = p->p_itimerid) != 0) {
p->p_itimerid = 0;
mutex_exit(&p->p_lock);
(void) untimeout(tmp_id);
mutex_enter(&p->p_lock);
}
lwp_cleanup();
/*
* We are about to exit; prevent our resource associations from
* being changed.
*/
pool_barrier_enter();
/*
* Block the process against /proc now that we have really
* acquired p->p_lock (to manipulate p_tlist at least).
*/
prbarrier(p);
sigfillset(&p->p_ignore);
sigemptyset(&p->p_siginfo);
sigemptyset(&p->p_sig);
sigemptyset(&p->p_extsig);
sigemptyset(&t->t_sig);
sigemptyset(&t->t_extsig);
sigemptyset(&p->p_sigmask);
sigdelq(p, t, 0);
lwp->lwp_cursig = 0;
lwp->lwp_extsig = 0;
p->p_flag &= ~(SKILLED | SEXTKILLED);
if (lwp->lwp_curinfo) {
siginfofree(lwp->lwp_curinfo);
lwp->lwp_curinfo = NULL;
}
t->t_proc_flag |= TP_LWPEXIT;
ASSERT(p->p_lwpcnt == 1 && p->p_zombcnt == 0);
prlwpexit(t); /* notify /proc */
lwp_hash_out(p, t->t_tid);
prexit(p);
p->p_lwpcnt = 0;
p->p_tlist = NULL;
sigqfree(p);
term_mstate(t);
p->p_mterm = gethrtime();
exec_vp = p->p_exec;
execdir_vp = p->p_execdir;
p->p_exec = NULLVP;
p->p_execdir = NULLVP;
mutex_exit(&p->p_lock);
pr_free_watched_pages(p);
closeall(P_FINFO(p));
/* Free the controlling tty. (freectty() always assumes curproc.) */
ASSERT(p == curproc);
(void) freectty(B_TRUE);
#if defined(__sparc)
if (p->p_utraps != NULL)
utrap_free(p);
#endif
if (p->p_semacct) /* IPC semaphore exit */
semexit(p);
rv = wstat(why, what);
acct(rv & 0xff);
exacct_commit_proc(p, rv);
/*
* Release any resources associated with C2 auditing
*/
if (AU_AUDITING()) {
/*
* audit exit system call
*/
audit_exit(why, what);
}
/*
* Free address space.
*/
relvm();
if (exec_vp) {
/*
* Close this executable which has been opened when the process
* was created by getproc().
*/
(void) VOP_CLOSE(exec_vp, FREAD, 1, (offset_t)0, CRED(), NULL);
VN_RELE(exec_vp);
}
if (execdir_vp)
VN_RELE(execdir_vp);
/*
* Release held contracts.
*/
contract_exit(p);
/*
* Depart our encapsulating process contract.
*/
if ((p->p_flag & SSYS) == 0) {
ASSERT(p->p_ct_process);
contract_process_exit(p->p_ct_process, p, rv);
}
/*
* Remove pool association, and block if requested by pool_do_bind.
*/
mutex_enter(&p->p_lock);
ASSERT(p->p_pool->pool_ref > 0);
atomic_add_32(&p->p_pool->pool_ref, -1);
p->p_pool = pool_default;
/*
* Now that our address space has been freed and all other threads
* in this process have exited, set the PEXITED pool flag. This
* tells the pools subsystems to ignore this process if it was
* requested to rebind this process to a new pool.
*/
p->p_poolflag |= PEXITED;
pool_barrier_exit();
mutex_exit(&p->p_lock);
mutex_enter(&pidlock);
/*
* Delete this process from the newstate list of its parent. We
* will put it in the right place in the sigcld in the end.
*/
delete_ns(p->p_parent, p);
/*
* Reassign the orphans to the next of kin.
* Don't rearrange init's orphanage.
*/
if ((q = p->p_orphan) != NULL && p != proc_init) {
proc_t *nokp = p->p_nextofkin;
for (;;) {
q->p_nextofkin = nokp;
if (q->p_nextorph == NULL)
break;
q = q->p_nextorph;
}
q->p_nextorph = nokp->p_orphan;
nokp->p_orphan = p->p_orphan;
p->p_orphan = NULL;
}
/*
* Reassign the children to init.
* Don't try to assign init's children to init.
*/
if ((q = p->p_child) != NULL && p != proc_init) {
struct proc *np;
struct proc *initp = proc_init;
boolean_t setzonetop = B_FALSE;
if (!INGLOBALZONE(curproc))
setzonetop = B_TRUE;
pgdetach(p);
do {
np = q->p_sibling;
/*
* Delete it from its current parent new state
* list and add it to init new state list
*/
delete_ns(q->p_parent, q);
q->p_ppid = 1;
q->p_pidflag &= ~(CLDNOSIGCHLD | CLDWAITPID);
if (setzonetop) {
mutex_enter(&q->p_lock);
q->p_flag |= SZONETOP;
mutex_exit(&q->p_lock);
}
q->p_parent = initp;
/*
* Since q will be the first child,
* it will not have a previous sibling.
*/
q->p_psibling = NULL;
if (initp->p_child) {
initp->p_child->p_psibling = q;
}
q->p_sibling = initp->p_child;
initp->p_child = q;
if (q->p_proc_flag & P_PR_PTRACE) {
mutex_enter(&q->p_lock);
sigtoproc(q, NULL, SIGKILL);
mutex_exit(&q->p_lock);
}
/*
* sigcld() will add the child to parents
* newstate list.
*/
if (q->p_stat == SZOMB)
sigcld(q, NULL);
} while ((q = np) != NULL);
p->p_child = NULL;
ASSERT(p->p_child_ns == NULL);
}
TRACE_1(TR_FAC_PROC, TR_PROC_EXIT, "proc_exit: %p", p);
mutex_enter(&p->p_lock);
CL_EXIT(curthread); /* tell the scheduler that curthread is exiting */
/*
* Have our task accummulate our resource usage data before they
* become contaminated by p_cacct etc., and before we renounce
* membership of the task.
*
* We do this regardless of whether or not task accounting is active.
* This is to avoid having nonsense data reported for this task if
* task accounting is subsequently enabled. The overhead is minimal;
* by this point, this process has accounted for the usage of all its
* LWPs. We nonetheless do the work here, and under the protection of
* pidlock, so that the movement of the process's usage to the task
* happens at the same time as the removal of the process from the
* task, from the point of view of exacct_snapshot_task_usage().
*/
exacct_update_task_mstate(p);
hrutime = mstate_aggr_state(p, LMS_USER);
hrstime = mstate_aggr_state(p, LMS_SYSTEM);
p->p_utime = (clock_t)NSEC_TO_TICK(hrutime) + p->p_cutime;
p->p_stime = (clock_t)NSEC_TO_TICK(hrstime) + p->p_cstime;
p->p_acct[LMS_USER] += p->p_cacct[LMS_USER];
p->p_acct[LMS_SYSTEM] += p->p_cacct[LMS_SYSTEM];
p->p_acct[LMS_TRAP] += p->p_cacct[LMS_TRAP];
p->p_acct[LMS_TFAULT] += p->p_cacct[LMS_TFAULT];
p->p_acct[LMS_DFAULT] += p->p_cacct[LMS_DFAULT];
p->p_acct[LMS_KFAULT] += p->p_cacct[LMS_KFAULT];
p->p_acct[LMS_USER_LOCK] += p->p_cacct[LMS_USER_LOCK];
p->p_acct[LMS_SLEEP] += p->p_cacct[LMS_SLEEP];
p->p_acct[LMS_WAIT_CPU] += p->p_cacct[LMS_WAIT_CPU];
p->p_acct[LMS_STOPPED] += p->p_cacct[LMS_STOPPED];
p->p_ru.minflt += p->p_cru.minflt;
p->p_ru.majflt += p->p_cru.majflt;
p->p_ru.nswap += p->p_cru.nswap;
p->p_ru.inblock += p->p_cru.inblock;
p->p_ru.oublock += p->p_cru.oublock;
p->p_ru.msgsnd += p->p_cru.msgsnd;
p->p_ru.msgrcv += p->p_cru.msgrcv;
p->p_ru.nsignals += p->p_cru.nsignals;
p->p_ru.nvcsw += p->p_cru.nvcsw;
p->p_ru.nivcsw += p->p_cru.nivcsw;
p->p_ru.sysc += p->p_cru.sysc;
p->p_ru.ioch += p->p_cru.ioch;
p->p_stat = SZOMB;
p->p_proc_flag &= ~P_PR_PTRACE;
p->p_wdata = what;
p->p_wcode = (char)why;
cdir = PTOU(p)->u_cdir;
rdir = PTOU(p)->u_rdir;
cwd = PTOU(p)->u_cwd;
ASSERT(cdir != NULL || p->p_parent == &p0);
/*
* Release resource controls, as they are no longer enforceable.
*/
rctl_set_free(p->p_rctls);
/*
* Decrement tk_nlwps counter for our task.max-lwps resource control.
* An extended accounting record, if that facility is active, is
* scheduled to be written. We cannot give up task and project
* membership at this point because that would allow zombies to escape
* from the max-processes resource controls. Zombies stay in their
* current task and project until the process table slot is released
* in freeproc().
*/
tk = p->p_task;
mutex_enter(&p->p_zone->zone_nlwps_lock);
tk->tk_nlwps--;
tk->tk_proj->kpj_nlwps--;
p->p_zone->zone_nlwps--;
mutex_exit(&p->p_zone->zone_nlwps_lock);
/*
* Clear the lwp directory and the lwpid hash table
* now that /proc can't bother us any more.
* We free the memory below, after dropping p->p_lock.
*/
lwpdir = p->p_lwpdir;
lwpdir_sz = p->p_lwpdir_sz;
tidhash = p->p_tidhash;
tidhash_sz = p->p_tidhash_sz;
ret_tidhash = p->p_ret_tidhash;
p->p_lwpdir = NULL;
p->p_lwpfree = NULL;
p->p_lwpdir_sz = 0;
p->p_tidhash = NULL;
p->p_tidhash_sz = 0;
p->p_ret_tidhash = NULL;
/*
* If the process has context ops installed, call the exit routine
* on behalf of this last remaining thread. Normally exitpctx() is
* called during thread_exit() or lwp_exit(), but because this is the
* last thread in the process, we must call it here. By the time
* thread_exit() is called (below), the association with the relevant
* process has been lost.
*
* We also free the context here.
*/
if (p->p_pctx) {
kpreempt_disable();
exitpctx(p);
kpreempt_enable();
freepctx(p, 0);
}
/*
* curthread's proc pointer is changed to point to the 'sched'
* process for the corresponding zone, except in the case when
* the exiting process is in fact a zsched instance, in which
* case the proc pointer is set to p0. We do so, so that the
* process still points at the right zone when we call the VN_RELE()
* below.
*
* This is because curthread's original proc pointer can be freed as
* soon as the child sends a SIGCLD to its parent. We use zsched so
* that for user processes, even in the final moments of death, the
* process is still associated with its zone.
*/
if (p != t->t_procp->p_zone->zone_zsched)
t->t_procp = t->t_procp->p_zone->zone_zsched;
else
t->t_procp = &p0;
mutex_exit(&p->p_lock);
if (!evaporate) {
p->p_pidflag &= ~CLDPEND;
sigcld(p, sqp);
} else {
/*
* Do what sigcld() would do if the disposition
* of the SIGCHLD signal were set to be ignored.
*/
cv_broadcast(&p->p_srwchan_cv);
freeproc(p);
}
mutex_exit(&pidlock);
/*
* We don't release u_cdir and u_rdir until SZOMB is set.
* This protects us against dofusers().
*/
if (cdir)
VN_RELE(cdir);
if (rdir)
VN_RELE(rdir);
if (cwd)
refstr_rele(cwd);
/*
* task_rele() may ultimately cause the zone to go away (or
* may cause the last user process in a zone to go away, which
* signals zsched to go away). So prior to this call, we must
* no longer point at zsched.
*/
t->t_procp = &p0;
kmem_free(lwpdir, lwpdir_sz * sizeof (lwpdir_t));
kmem_free(tidhash, tidhash_sz * sizeof (tidhash_t));
while (ret_tidhash != NULL) {
ret_tidhash_t *next = ret_tidhash->rth_next;
kmem_free(ret_tidhash->rth_tidhash,
ret_tidhash->rth_tidhash_sz * sizeof (tidhash_t));
kmem_free(ret_tidhash, sizeof (*ret_tidhash));
ret_tidhash = next;
}
thread_exit();
/* NOTREACHED */
}
/*
* Allocate a new lxproc node
*
* This also allocates the vnode associated with it
*/
lxpr_node_t *
lxpr_getnode(vnode_t *dp, lxpr_nodetype_t type, proc_t *p, int fd)
{
lxpr_node_t *lxpnp;
vnode_t *vp;
user_t *up;
timestruc_t now;
/*
* Allocate a new node. It is deallocated in vop_innactive
*/
lxpnp = kmem_cache_alloc(lxpr_node_cache, KM_SLEEP);
/*
* Set defaults (may be overridden below)
*/
gethrestime(&now);
lxpnp->lxpr_type = type;
lxpnp->lxpr_realvp = NULL;
lxpnp->lxpr_parent = dp;
VN_HOLD(dp);
if (p != NULL) {
lxpnp->lxpr_pid = ((p->p_pid ==
curproc->p_zone->zone_proc_initpid) ? 1 : p->p_pid);
lxpnp->lxpr_time = PTOU(p)->u_start;
lxpnp->lxpr_uid = crgetruid(p->p_cred);
lxpnp->lxpr_gid = crgetrgid(p->p_cred);
lxpnp->lxpr_ino = lxpr_inode(type, p->p_pid, fd);
} else {
/* Pretend files without a proc belong to sched */
lxpnp->lxpr_pid = 0;
lxpnp->lxpr_time = now;
lxpnp->lxpr_uid = lxpnp->lxpr_gid = 0;
lxpnp->lxpr_ino = lxpr_inode(type, 0, 0);
}
/* initialize the vnode data */
vp = lxpnp->lxpr_vnode;
vn_reinit(vp);
vp->v_flag = VNOCACHE|VNOMAP|VNOSWAP|VNOMOUNT;
vp->v_vfsp = dp->v_vfsp;
/*
* Do node specific stuff
*/
switch (type) {
case LXPR_PROCDIR:
vp->v_flag |= VROOT;
vp->v_type = VDIR;
lxpnp->lxpr_mode = 0555; /* read-search by everyone */
break;
case LXPR_PID_CURDIR:
ASSERT(p != NULL);
/*
* Zombie check. p_stat is officially protected by pidlock,
* but we can't grab pidlock here because we already hold
* p_lock. Luckily if we look at the process exit code
* we see that p_stat only transisions from SRUN to SZOMB
* while p_lock is held. Aside from this, the only other
* p_stat transition that we need to be aware about is
* SIDL to SRUN, but that's not a problem since lxpr_lock()
* ignores nodes in the SIDL state so we'll never get a node
* that isn't already in the SRUN state.
*/
if (p->p_stat == SZOMB) {
lxpnp->lxpr_realvp = NULL;
} else {
up = PTOU(p);
lxpnp->lxpr_realvp = up->u_cdir;
ASSERT(lxpnp->lxpr_realvp != NULL);
VN_HOLD(lxpnp->lxpr_realvp);
}
vp->v_type = VLNK;
lxpnp->lxpr_mode = 0777; /* anyone does anything ! */
break;
case LXPR_PID_ROOTDIR:
ASSERT(p != NULL);
/* Zombie check. see locking comment above */
if (p->p_stat == SZOMB) {
lxpnp->lxpr_realvp = NULL;
} else {
up = PTOU(p);
lxpnp->lxpr_realvp =
up->u_rdir != NULL ? up->u_rdir : rootdir;
ASSERT(lxpnp->lxpr_realvp != NULL);
VN_HOLD(lxpnp->lxpr_realvp);
}
vp->v_type = VLNK;
lxpnp->lxpr_mode = 0777; /* anyone does anything ! */
break;
case LXPR_PID_EXE:
ASSERT(p != NULL);
lxpnp->lxpr_realvp = p->p_exec;
if (lxpnp->lxpr_realvp != NULL) {
VN_HOLD(lxpnp->lxpr_realvp);
}
vp->v_type = VLNK;
lxpnp->lxpr_mode = 0777;
break;
case LXPR_SELF:
vp->v_type = VLNK;
lxpnp->lxpr_mode = 0777; /* anyone does anything ! */
break;
case LXPR_PID_FD_FD:
ASSERT(p != NULL);
/* lxpr_realvp is set after we return */
vp->v_type = VLNK;
lxpnp->lxpr_mode = 0700; /* read-write-exe owner only */
break;
case LXPR_PID_FDDIR:
ASSERT(p != NULL);
vp->v_type = VDIR;
lxpnp->lxpr_mode = 0500; /* read-search by owner only */
break;
case LXPR_PIDDIR:
ASSERT(p != NULL);
vp->v_type = VDIR;
lxpnp->lxpr_mode = 0511;
break;
case LXPR_SYSDIR:
case LXPR_SYS_FSDIR:
case LXPR_SYS_FS_INOTIFYDIR:
case LXPR_SYS_KERNELDIR:
case LXPR_NETDIR:
vp->v_type = VDIR;
lxpnp->lxpr_mode = 0555; /* read-search by all */
break;
case LXPR_PID_ENV:
case LXPR_PID_MEM:
ASSERT(p != NULL);
/*FALLTHRU*/
case LXPR_KCORE:
vp->v_type = VREG;
lxpnp->lxpr_mode = 0400; /* read-only by owner only */
break;
default:
vp->v_type = VREG;
lxpnp->lxpr_mode = 0444; /* read-only by all */
break;
}
return (lxpnp);
}
void
exacct_calculate_proc_usage(proc_t *p, proc_usage_t *pu, ulong_t *mask,
int flag, int wstat)
{
timestruc_t ts, ts_run;
ASSERT(MUTEX_HELD(&p->p_lock));
/*
* Convert CPU and execution times to sec/nsec format.
*/
if (BT_TEST(mask, AC_PROC_CPU)) {
hrt2ts(mstate_aggr_state(p, LMS_USER), &ts);
pu->pu_utimesec = (uint64_t)(ulong_t)ts.tv_sec;
pu->pu_utimensec = (uint64_t)(ulong_t)ts.tv_nsec;
hrt2ts(mstate_aggr_state(p, LMS_SYSTEM), &ts);
pu->pu_stimesec = (uint64_t)(ulong_t)ts.tv_sec;
pu->pu_stimensec = (uint64_t)(ulong_t)ts.tv_nsec;
}
if (BT_TEST(mask, AC_PROC_TIME)) {
gethrestime(&ts);
pu->pu_finishsec = (uint64_t)(ulong_t)ts.tv_sec;
pu->pu_finishnsec = (uint64_t)(ulong_t)ts.tv_nsec;
hrt2ts(gethrtime() - p->p_mstart, &ts_run);
ts.tv_sec -= ts_run.tv_sec;
ts.tv_nsec -= ts_run.tv_nsec;
if (ts.tv_nsec < 0) {
ts.tv_sec--;
if ((ts.tv_nsec = ts.tv_nsec + NANOSEC) >= NANOSEC) {
ts.tv_sec++;
ts.tv_nsec -= NANOSEC;
}
}
pu->pu_startsec = (uint64_t)(ulong_t)ts.tv_sec;
pu->pu_startnsec = (uint64_t)(ulong_t)ts.tv_nsec;
}
pu->pu_pid = p->p_pidp->pid_id;
pu->pu_acflag = p->p_user.u_acflag;
pu->pu_projid = p->p_task->tk_proj->kpj_id;
pu->pu_taskid = p->p_task->tk_tkid;
pu->pu_major = getmajor(p->p_sessp->s_dev);
pu->pu_minor = getminor(p->p_sessp->s_dev);
pu->pu_ancpid = p->p_ancpid;
pu->pu_wstat = wstat;
/*
* Compute average RSS in K. The denominator is the number of
* samples: the number of clock ticks plus the initial value.
*/
pu->pu_mem_rss_avg = (PTOU(p)->u_mem / (p->p_stime + p->p_utime + 1)) *
(PAGESIZE / 1024);
pu->pu_mem_rss_max = PTOU(p)->u_mem_max * (PAGESIZE / 1024);
mutex_enter(&p->p_crlock);
pu->pu_ruid = crgetruid(p->p_cred);
pu->pu_rgid = crgetrgid(p->p_cred);
mutex_exit(&p->p_crlock);
bcopy(p->p_user.u_comm, pu->pu_command, strlen(p->p_user.u_comm) + 1);
bcopy(p->p_zone->zone_name, pu->pu_zonename,
strlen(p->p_zone->zone_name) + 1);
bcopy(p->p_zone->zone_nodename, pu->pu_nodename,
strlen(p->p_zone->zone_nodename) + 1);
/*
* Calculate microstate accounting data for a process that is still
* running. Presently, we explicitly collect all of the LWP usage into
* the proc usage structure here.
*/
if (flag & EW_PARTIAL)
exacct_calculate_proc_mstate(p, pu);
if (flag & EW_FINAL)
exacct_copy_proc_mstate(p, pu);
}
/*
* Post-syscall processing. Perform abnormal system call completion
* actions such as /proc tracing, profiling, signals, preemption, etc.
*
* This routine is called only if t_post_sys, t_sig_check, or t_astflag is set.
* Any condition requiring pre-syscall handling must set one of these.
* If the condition is persistent, this routine will repost t_post_sys.
*/
void
post_syscall(long rval1, long rval2)
{
kthread_t *t = curthread;
klwp_t *lwp = ttolwp(t);
proc_t *p = ttoproc(t);
struct regs *rp = lwptoregs(lwp);
uint_t error;
uint_t code = t->t_sysnum;
int repost = 0;
int proc_stop = 0; /* non-zero if stopping */
int sigprof = 0; /* non-zero if sending SIGPROF */
t->t_post_sys = 0;
error = lwp->lwp_errno;
/*
* Code can be zero if this is a new LWP returning after a forkall(),
* other than the one which matches the one in the parent which called
* forkall(). In these LWPs, skip most of post-syscall activity.
*/
if (code == 0)
goto sig_check;
/*
* If the trace flag is set, mark the lwp to take a single-step trap
* on return to user level (below). The x86 lcall interface and
* sysenter has already done this, and turned off the flag, but
* amd64 syscall interface has not.
*/
if (rp->r_ps & PS_T) {
lwp->lwp_pcb.pcb_flags |= DEBUG_PENDING;
rp->r_ps &= ~PS_T;
aston(curthread);
}
#ifdef C2_AUDIT
if (audit_active) { /* put out audit record for this syscall */
rval_t rval;
/* XX64 -- truncation of 64-bit return values? */
rval.r_val1 = (int)rval1;
rval.r_val2 = (int)rval2;
audit_finish(T_SYSCALL, code, error, &rval);
repost = 1;
}
#endif /* C2_AUDIT */
if (curthread->t_pdmsg != NULL) {
char *m = curthread->t_pdmsg;
uprintf("%s", m);
kmem_free(m, strlen(m) + 1);
curthread->t_pdmsg = NULL;
}
/*
* If we're going to stop for /proc tracing, set the flag and
* save the arguments so that the return values don't smash them.
*/
if (PTOU(p)->u_systrap) {
if (prismember(&PTOU(p)->u_exitmask, code)) {
if (lwp_getdatamodel(lwp) == DATAMODEL_LP64)
(void) save_syscall_args();
proc_stop = 1;
}
repost = 1;
}
/*
* Similarly check to see if SIGPROF might be sent.
*/
if (curthread->t_rprof != NULL &&
curthread->t_rprof->rp_anystate != 0) {
if (lwp_getdatamodel(lwp) == DATAMODEL_LP64)
(void) save_syscall_args();
sigprof = 1;
}
if (lwp->lwp_eosys == NORMALRETURN) {
if (error == 0) {
#ifdef SYSCALLTRACE
if (syscalltrace) {
mutex_enter(&systrace_lock);
printf(
"%d: r_val1=0x%lx, r_val2=0x%lx, id 0x%p\n",
p->p_pid, rval1, rval2, curthread);
mutex_exit(&systrace_lock);
}
#endif /* SYSCALLTRACE */
rp->r_ps &= ~PS_C;
rp->r_r0 = rval1;
rp->r_r1 = rval2;
} else {
int sig;
#ifdef SYSCALLTRACE
if (syscalltrace) {
mutex_enter(&systrace_lock);
printf("%d: error=%d, id 0x%p\n",
p->p_pid, error, curthread);
mutex_exit(&systrace_lock);
}
#endif /* SYSCALLTRACE */
if (error == EINTR && t->t_activefd.a_stale)
error = EBADF;
if (error == EINTR &&
(sig = lwp->lwp_cursig) != 0 &&
sigismember(&PTOU(p)->u_sigrestart, sig) &&
PTOU(p)->u_signal[sig - 1] != SIG_DFL &&
PTOU(p)->u_signal[sig - 1] != SIG_IGN)
error = ERESTART;
rp->r_r0 = error;
rp->r_ps |= PS_C;
}
}
/*
* From the proc(4) manual page:
* When exit from a system call is being traced, the traced process
* stops on completion of the system call just prior to checking for
* signals and returning to user level. At this point all return
* values have been stored into the traced process's saved registers.
*/
if (proc_stop) {
mutex_enter(&p->p_lock);
if (PTOU(p)->u_systrap &&
prismember(&PTOU(p)->u_exitmask, code))
stop(PR_SYSEXIT, code);
mutex_exit(&p->p_lock);
}
/*
* If we are the parent returning from a successful
* vfork, wait for the child to exec or exit.
* This code must be here and not in the bowels of the system
* so that /proc can intercept exit from vfork in a timely way.
*/
if (code == SYS_vfork && rp->r_r1 == 0 && error == 0)
vfwait((pid_t)rval1);
/*
* If profiling is active, bill the current PC in user-land
* and keep reposting until profiling is disabled.
*/
if (p->p_prof.pr_scale) {
if (lwp->lwp_oweupc)
profil_tick(rp->r_pc);
repost = 1;
}
sig_check:
/*
* Reset flag for next time.
* We must do this after stopping on PR_SYSEXIT
* because /proc uses the information in lwp_eosys.
*/
lwp->lwp_eosys = NORMALRETURN;
clear_stale_fd();
t->t_flag &= ~T_FORKALL;
if (t->t_astflag | t->t_sig_check) {
/*
* Turn off the AST flag before checking all the conditions that
* may have caused an AST. This flag is on whenever a signal or
* unusual condition should be handled after the next trap or
* syscall.
*/
astoff(t);
/*
* If a single-step trap occurred on a syscall (see trap())
* recognize it now. Do this before checking for signals
* because deferred_singlestep_trap() may generate a SIGTRAP to
* the LWP or may otherwise mark the LWP to call issig(FORREAL).
*/
if (lwp->lwp_pcb.pcb_flags & DEBUG_PENDING)
deferred_singlestep_trap((caddr_t)rp->r_pc);
t->t_sig_check = 0;
/*
* The following check is legal for the following reasons:
* 1) The thread we are checking, is ourselves, so there is
* no way the proc can go away.
* 2) The only time we need to be protected by the
* lock is if the binding is changed.
*
* Note we will still take the lock and check the binding
* if the condition was true without the lock held. This
* prevents lock contention among threads owned by the
* same proc.
*/
if (curthread->t_proc_flag & TP_CHANGEBIND) {
mutex_enter(&p->p_lock);
if (curthread->t_proc_flag & TP_CHANGEBIND) {
timer_lwpbind();
curthread->t_proc_flag &= ~TP_CHANGEBIND;
}
mutex_exit(&p->p_lock);
}
/*
* for kaio requests on the special kaio poll queue,
* copyout their results to user memory.
*/
if (p->p_aio)
aio_cleanup(0);
/*
* If this LWP was asked to hold, call holdlwp(), which will
* stop. holdlwps() sets this up and calls pokelwps() which
* sets the AST flag.
*
* Also check TP_EXITLWP, since this is used by fresh new LWPs
* through lwp_rtt(). That flag is set if the lwp_create(2)
* syscall failed after creating the LWP.
*/
if (ISHOLD(p) || (t->t_proc_flag & TP_EXITLWP))
holdlwp();
/*
* All code that sets signals and makes ISSIG_PENDING
* evaluate true must set t_sig_check afterwards.
*/
if (ISSIG_PENDING(t, lwp, p)) {
if (issig(FORREAL))
psig();
t->t_sig_check = 1; /* recheck next time */
}
if (sigprof) {
realsigprof(code, error);
t->t_sig_check = 1; /* recheck next time */
}
/*
* If a performance counter overflow interrupt was
* delivered *during* the syscall, then re-enable the
* AST so that we take a trip through trap() to cause
* the SIGEMT to be delivered.
*/
if (lwp->lwp_pcb.pcb_flags & CPC_OVERFLOW)
aston(t);
/*
* /proc can't enable/disable the trace bit itself
* because that could race with the call gate used by
* system calls via "lcall". If that happened, an
* invalid EFLAGS would result. prstep()/prnostep()
* therefore schedule an AST for the purpose.
*/
if (lwp->lwp_pcb.pcb_flags & REQUEST_STEP) {
lwp->lwp_pcb.pcb_flags &= ~REQUEST_STEP;
rp->r_ps |= PS_T;
}
if (lwp->lwp_pcb.pcb_flags & REQUEST_NOSTEP) {
lwp->lwp_pcb.pcb_flags &= ~REQUEST_NOSTEP;
rp->r_ps &= ~PS_T;
}
}
lwp->lwp_errno = 0; /* clear error for next time */
#ifndef NPROBE
/* Kernel probe */
if (tnf_tracing_active) {
TNF_PROBE_3(syscall_end, "syscall thread", /* CSTYLED */,
tnf_long, rval1, rval1,
tnf_long, rval2, rval2,
tnf_long, errno, (long)error);
repost = 1;
}
#endif /* NPROBE */
/*
* Set state to LWP_USER here so preempt won't give us a kernel
* priority if it occurs after this point. Call CL_TRAPRET() to
* restore the user-level priority.
*
* It is important that no locks (other than spinlocks) be entered
* after this point before returning to user mode (unless lwp_state
* is set back to LWP_SYS).
*
* XXX Sampled times past this point are charged to the user.
*/
lwp->lwp_state = LWP_USER;
if (t->t_trapret) {
t->t_trapret = 0;
thread_lock(t);
CL_TRAPRET(t);
thread_unlock(t);
}
if (CPU->cpu_runrun)
preempt();
lwp->lwp_errno = 0; /* clear error for next time */
/*
* The thread lock must be held in order to clear sysnum and reset
* lwp_ap atomically with respect to other threads in the system that
* may be looking at the args via lwp_ap from get_syscall_args().
*/
thread_lock(t);
t->t_sysnum = 0; /* no longer in a system call */
if (lwp_getdatamodel(lwp) == DATAMODEL_NATIVE) {
#if defined(_LP64)
/*
* In case the args were copied to the lwp, reset the
* pointer so the next syscall will have the right
* lwp_ap pointer.
*/
lwp->lwp_ap = (long *)&rp->r_rdi;
} else {
#endif
lwp->lwp_ap = NULL; /* reset on every syscall entry */
}
thread_unlock(t);
lwp->lwp_argsaved = 0;
/*
* If there was a continuing reason for post-syscall processing,
* set the t_post_sys flag for the next system call.
*/
if (repost)
t->t_post_sys = 1;
/*
* If there is a ustack registered for this lwp, and the stack rlimit
* has been altered, read in the ustack. If the saved stack rlimit
* matches the bounds of the ustack, update the ustack to reflect
* the new rlimit. If the new stack rlimit is RLIM_INFINITY, disable
* stack checking by setting the size to 0.
*/
if (lwp->lwp_ustack != 0 && lwp->lwp_old_stk_ctl != 0) {
rlim64_t new_size;
caddr_t top;
stack_t stk;
struct rlimit64 rl;
mutex_enter(&p->p_lock);
new_size = p->p_stk_ctl;
top = p->p_usrstack;
(void) rctl_rlimit_get(rctlproc_legacy[RLIMIT_STACK], p, &rl);
mutex_exit(&p->p_lock);
if (rl.rlim_cur == RLIM64_INFINITY)
new_size = 0;
if (copyin((stack_t *)lwp->lwp_ustack, &stk,
sizeof (stack_t)) == 0 &&
(stk.ss_size == lwp->lwp_old_stk_ctl ||
stk.ss_size == 0) &&
stk.ss_sp == top - stk.ss_size) {
stk.ss_sp = (void *)((uintptr_t)stk.ss_sp +
stk.ss_size - (uintptr_t)new_size);
stk.ss_size = new_size;
(void) copyout(&stk, (stack_t *)lwp->lwp_ustack,
sizeof (stack_t));
}
lwp->lwp_old_stk_ctl = 0;
}
}
/*
* This routine assumes that the stack grows downward.
* Returns 0 on success, errno on failure.
*/
int
grow_internal(caddr_t sp, uint_t growszc)
{
struct proc *p = curproc;
size_t newsize;
size_t oldsize;
int error;
size_t pgsz;
uint_t szc;
struct segvn_crargs crargs = SEGVN_ZFOD_ARGS(PROT_ZFOD, PROT_ALL);
ASSERT(sp < p->p_usrstack);
sp = (caddr_t)P2ALIGN((uintptr_t)sp, PAGESIZE);
/*
* grow to growszc alignment but use current p->p_stkpageszc for
* the segvn_crargs szc passed to segvn_create. For memcntl to
* increase the szc, this allows the new extension segment to be
* concatenated successfully with the existing stack segment.
*/
if ((szc = growszc) != 0) {
pgsz = page_get_pagesize(szc);
ASSERT(pgsz > PAGESIZE);
newsize = p->p_usrstack - (caddr_t)P2ALIGN((uintptr_t)sp, pgsz);
if (newsize > (size_t)p->p_stk_ctl) {
szc = 0;
pgsz = PAGESIZE;
newsize = p->p_usrstack - sp;
}
} else {
pgsz = PAGESIZE;
newsize = p->p_usrstack - sp;
}
if (newsize > (size_t)p->p_stk_ctl) {
(void) rctl_action(rctlproc_legacy[RLIMIT_STACK], p->p_rctls, p,
RCA_UNSAFE_ALL);
return (ENOMEM);
}
oldsize = p->p_stksize;
ASSERT(P2PHASE(oldsize, PAGESIZE) == 0);
if (newsize <= oldsize) { /* prevent the stack from shrinking */
return (0);
}
if (!(p->p_stkprot & PROT_EXEC)) {
crargs.prot &= ~PROT_EXEC;
}
/*
* extend stack with the proposed new growszc, which is different
* than p_stkpageszc only on a memcntl to increase the stack pagesize.
* AS_MAP_NO_LPOOB means use 0, and don't reapply OOB policies via
* map_pgszcvec(). Use AS_MAP_STACK to get intermediate page sizes
* if not aligned to szc's pgsz.
*/
if (szc > 0) {
caddr_t oldsp = p->p_usrstack - oldsize;
caddr_t austk = (caddr_t)P2ALIGN((uintptr_t)p->p_usrstack,
pgsz);
if (IS_P2ALIGNED(p->p_usrstack, pgsz) || oldsp < austk) {
crargs.szc = p->p_stkpageszc ? p->p_stkpageszc :
AS_MAP_NO_LPOOB;
} else if (oldsp == austk) {
crargs.szc = szc;
} else {
crargs.szc = AS_MAP_STACK;
}
} else {
crargs.szc = AS_MAP_NO_LPOOB;
}
crargs.lgrp_mem_policy_flags = LGRP_MP_FLAG_EXTEND_DOWN;
if ((error = as_map(p->p_as, p->p_usrstack - newsize, newsize - oldsize,
segvn_create, &crargs)) != 0) {
if (error == EAGAIN) {
cmn_err(CE_WARN, "Sorry, no swap space to grow stack "
"for pid %d (%s)", p->p_pid, PTOU(p)->u_comm);
}
return (error);
}
p->p_stksize = newsize;
return (0);
}
