static ephemeral_zsd_t *
get_ephemeral_zsd(zone_t *zone)
{
ephemeral_zsd_t *eph_zsd;
eph_zsd = zone_getspecific(ephemeral_zone_key, zone);
if (eph_zsd != NULL) {
return (eph_zsd);
}
mutex_enter(&ephemeral_zone_mutex);
eph_zsd = zone_getspecific(ephemeral_zone_key, zone);
if (eph_zsd == NULL) {
eph_zsd = kmem_zalloc(sizeof (ephemeral_zsd_t), KM_SLEEP);
eph_zsd->min_uid = MAXUID;
eph_zsd->last_uid = IDMAP_WK__MAX_UID;
eph_zsd->min_gid = MAXUID;
eph_zsd->last_gid = IDMAP_WK__MAX_GID;
mutex_init(&eph_zsd->eph_lock, NULL, MUTEX_DEFAULT, NULL);
/*
* nobody is used to map SID containing CRs.
*/
eph_zsd->eph_nobody = crdup(zone->zone_kcred);
(void) crsetugid(eph_zsd->eph_nobody, UID_NOBODY, GID_NOBODY);
CR_FLAGS(eph_zsd->eph_nobody) = 0;
eph_zsd->eph_nobody->cr_zone = zone;
(void) zone_setspecific(ephemeral_zone_key, zone, eph_zsd);
}
mutex_exit(&ephemeral_zone_mutex);
return (eph_zsd);
}
/*
* Get the client credential. Used for Renew and recovery.
*/
struct ucred *
newnfs_getcred(void)
{
struct ucred *cred;
struct thread *td = curthread;
cred = crdup(td->td_ucred);
newnfs_setroot(cred);
return (cred);
}
/*
* If the reply status is NFSERR_EACCES, it may be because we are
* root (no root net access). Check the real uid, if it isn't root
* make that the uid instead and retry the call.
* Private interface for NFS.
*/
cred_t *
crnetadjust(cred_t *cr)
{
if (cr->cr_uid == 0 && cr->cr_ruid != 0) {
cr = crdup(cr);
cr->cr_uid = cr->cr_ruid;
return (cr);
}
return (NULL);
}
/*
* XXX This is wrong
*/
static struct ucred *
nnpfs_crcopy(struct ucred *cr)
{
struct ucred *ncr;
if (crshared(cr)) {
ncr = crdup(cr);
crfree(cr);
return ncr;
}
return cr;
}
int
linux_setgroups16(struct proc *p, struct linux_setgroups16_args *args)
{
struct ucred *newcred, *oldcred;
l_gid16_t linux_gidset[NGROUPS];
gid_t *bsd_gidset;
int ngrp, error;
#ifdef DEBUG
if (ldebug(setgroups16))
printf(ARGS(setgroups16, "%d, *"), args->gidsetsize);
#endif
ngrp = args->gidsetsize;
oldcred = p->p_ucred;
/*
* cr_groups[0] holds egid. Setting the whole set from
* the supplied set will cause egid to be changed too.
* Keep cr_groups[0] unchanged to prevent that.
*/
if ((error = suser_xxx(oldcred, NULL, PRISON_ROOT)) != 0)
return (error);
if (ngrp >= NGROUPS)
return (EINVAL);
newcred = crdup(oldcred);
if (ngrp > 0) {
error = copyin((caddr_t)args->gidset, linux_gidset,
ngrp * sizeof(l_gid16_t));
if (error)
return (error);
newcred->cr_ngroups = ngrp + 1;
bsd_gidset = newcred->cr_groups;
ngrp--;
while (ngrp >= 0) {
bsd_gidset[ngrp + 1] = linux_gidset[ngrp];
ngrp--;
}
}
else
newcred->cr_ngroups = 1;
setsugid(p);
p->p_ucred = newcred;
crfree(oldcred);
return (0);
}
void
osi_Init(void)
{
static int once = 0;
if (once++ > 0) /* just in case */
return;
osi_InitGlock();
/* Initialize a lock for the kernel hcrypto bits. */
#ifndef UKERNEL
init_hckernel_mutex();
#endif
if (!afs_osicred_initialized) {
#if defined(AFS_DARWIN80_ENV)
afs_osi_ctxtp_initialized = 0;
afs_osi_ctxtp = NULL; /* initialized in afs_Daemon since it has
a proc reference that cannot be changed */
#endif
#if defined(AFS_XBSD_ENV)
/* Can't just invent one, must use crget() because of mutex */
afs_osi_credp =
crdup(osi_curcred());
#elif defined(AFS_SUN5_ENV)
afs_osi_credp = kcred;
#else
memset(&afs_osi_cred, 0, sizeof(afs_ucred_t));
#if defined(AFS_LINUX26_ENV)
afs_set_cr_group_info(&afs_osi_cred, groups_alloc(0));
#endif
#if defined(AFS_DARWIN80_ENV)
afs_osi_cred.cr_ref = 1; /* kauth_cred_get_ref needs 1 existing ref */
#else
# if !(defined(AFS_LINUX26_ENV) && defined(STRUCT_TASK_STRUCT_HAS_CRED))
crhold(&afs_osi_cred); /* don't let it evaporate */
# endif
#endif
afs_osi_credp = &afs_osi_cred;
#endif
afs_osicred_initialized = 1;
}
#ifdef AFS_SGI64_ENV
osi_flid.fl_pid = osi_flid.fl_sysid = 0;
#endif
init_et_to_sys_error();
}
void
smb_credinit(struct smb_cred *scred, cred_t *cr)
{
/* cr arg is optional */
if (cr == NULL)
cr = ddi_get_cred();
if (is_system_labeled()) {
cr = crdup(cr);
(void) setpflags(NET_MAC_AWARE, 1, cr);
} else {
crhold(cr);
}
scred->scr_cred = cr;
}
CLIENT *
clnt_reconnect_create(
struct netconfig *nconf, /* network type */
struct sockaddr *svcaddr, /* servers address */
rpcprog_t program, /* program number */
rpcvers_t version, /* version number */
size_t sendsz, /* buffer recv size */
size_t recvsz) /* buffer send size */
{
CLIENT *cl = NULL; /* client handle */
struct rc_data *rc = NULL; /* private data */
if (svcaddr == NULL) {
rpc_createerr.cf_stat = RPC_UNKNOWNADDR;
return (NULL);
}
cl = mem_alloc(sizeof (CLIENT));
rc = mem_alloc(sizeof (*rc));
mtx_init(&rc->rc_lock, "rc->rc_lock", NULL, MTX_DEF);
(void) memcpy(&rc->rc_addr, svcaddr, (size_t)svcaddr->sa_len);
rc->rc_nconf = nconf;
rc->rc_prog = program;
rc->rc_vers = version;
rc->rc_sendsz = sendsz;
rc->rc_recvsz = recvsz;
rc->rc_timeout.tv_sec = -1;
rc->rc_timeout.tv_usec = -1;
rc->rc_retry.tv_sec = 3;
rc->rc_retry.tv_usec = 0;
rc->rc_retries = INT_MAX;
rc->rc_privport = FALSE;
rc->rc_waitchan = "rpcrecv";
rc->rc_intr = 0;
rc->rc_connecting = FALSE;
rc->rc_closed = FALSE;
rc->rc_ucred = crdup(curthread->td_ucred);
rc->rc_client = NULL;
cl->cl_refs = 1;
cl->cl_ops = &clnt_reconnect_ops;
cl->cl_private = (caddr_t)(void *)rc;
cl->cl_auth = authnone_create();
cl->cl_tp = NULL;
cl->cl_netid = NULL;
return (cl);
}
/*
* Duplicate the current processes' credentials. Since we are called only
* as the result of a SET ioctl and only root can do that, any future access
* to this "disk" is essentially as root. Note that credentials may change
* if some other uid can write directly to the mapped file (NFS).
*/
int
vndsetcred(struct vnd_softc *sc, struct ucred *cred)
{
void *buf;
size_t size;
int error;
sc->sc_cred = crdup(cred);
buf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
size = MIN(DEV_BSIZE, sc->sc_size * sc->sc_secsize);
/* XXX: Horrible kludge to establish credentials for NFS */
error = vn_rdwr(UIO_READ, sc->sc_vp, buf, size, 0, UIO_SYSSPACE, 0,
sc->sc_cred, NULL, curproc);
free(buf, M_TEMP, DEV_BSIZE);
return (error);
}
/*
* In-kernel implementation of execve(). All arguments are assumed to be
* userspace pointers from the passed thread.
*/
static int
do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
{
struct proc *p = td->td_proc;
struct nameidata nd;
struct ucred *oldcred;
struct uidinfo *euip = NULL;
register_t *stack_base;
int error, i;
struct image_params image_params, *imgp;
struct vattr attr;
int (*img_first)(struct image_params *);
struct pargs *oldargs = NULL, *newargs = NULL;
struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
#ifdef KTRACE
struct vnode *tracevp = NULL;
struct ucred *tracecred = NULL;
#endif
struct vnode *oldtextvp = NULL, *newtextvp;
int credential_changing;
int textset;
#ifdef MAC
struct label *interpvplabel = NULL;
int will_transition;
#endif
#ifdef HWPMC_HOOKS
struct pmckern_procexec pe;
#endif
static const char fexecv_proc_title[] = "(fexecv)";
imgp = &image_params;
/*
* Lock the process and set the P_INEXEC flag to indicate that
* it should be left alone until we're done here. This is
* necessary to avoid race conditions - e.g. in ptrace() -
* that might allow a local user to illicitly obtain elevated
* privileges.
*/
PROC_LOCK(p);
KASSERT((p->p_flag & P_INEXEC) == 0,
("%s(): process already has P_INEXEC flag", __func__));
p->p_flag |= P_INEXEC;
PROC_UNLOCK(p);
/*
* Initialize part of the common data
*/
bzero(imgp, sizeof(*imgp));
imgp->proc = p;
imgp->attr = &attr;
imgp->args = args;
oldcred = p->p_ucred;
#ifdef MAC
error = mac_execve_enter(imgp, mac_p);
if (error)
goto exec_fail;
#endif
/*
* Translate the file name. namei() returns a vnode pointer
* in ni_vp among other things.
*
* XXXAUDIT: It would be desirable to also audit the name of the
* interpreter if this is an interpreted binary.
*/
if (args->fname != NULL) {
NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
| AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
}
SDT_PROBE1(proc, , , exec, args->fname);
interpret:
if (args->fname != NULL) {
#ifdef CAPABILITY_MODE
/*
* While capability mode can't reach this point via direct
* path arguments to execve(), we also don't allow
* interpreters to be used in capability mode (for now).
* Catch indirect lookups and return a permissions error.
*/
if (IN_CAPABILITY_MODE(td)) {
error = ECAPMODE;
goto exec_fail;
}
#endif
error = namei(&nd);
if (error)
goto exec_fail;
newtextvp = nd.ni_vp;
imgp->vp = newtextvp;
} else {
AUDIT_ARG_FD(args->fd);
/*
* Descriptors opened only with O_EXEC or O_RDONLY are allowed.
*/
error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
if (error)
goto exec_fail;
vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
AUDIT_ARG_VNODE1(newtextvp);
imgp->vp = newtextvp;
}
/*
* Check file permissions (also 'opens' file)
*/
error = exec_check_permissions(imgp);
if (error)
goto exec_fail_dealloc;
imgp->object = imgp->vp->v_object;
if (imgp->object != NULL)
vm_object_reference(imgp->object);
/*
* Set VV_TEXT now so no one can write to the executable while we're
* activating it.
*
* Remember if this was set before and unset it in case this is not
* actually an executable image.
*/
textset = VOP_IS_TEXT(imgp->vp);
VOP_SET_TEXT(imgp->vp);
error = exec_map_first_page(imgp);
if (error)
goto exec_fail_dealloc;
imgp->proc->p_osrel = 0;
imgp->proc->p_fctl0 = 0;
/*
* Implement image setuid/setgid.
*
* Determine new credentials before attempting image activators
* so that it can be used by process_exec handlers to determine
* credential/setid changes.
*
* Don't honor setuid/setgid if the filesystem prohibits it or if
* the process is being traced.
*
* We disable setuid/setgid/etc in capability mode on the basis
* that most setugid applications are not written with that
* environment in mind, and will therefore almost certainly operate
* incorrectly. In principle there's no reason that setugid
* applications might not be useful in capability mode, so we may want
* to reconsider this conservative design choice in the future.
*
* XXXMAC: For the time being, use NOSUID to also prohibit
* transitions on the file system.
*/
credential_changing = 0;
credential_changing |= (attr.va_mode & S_ISUID) &&
oldcred->cr_uid != attr.va_uid;
credential_changing |= (attr.va_mode & S_ISGID) &&
oldcred->cr_gid != attr.va_gid;
#ifdef MAC
will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
interpvplabel, imgp);
credential_changing |= will_transition;
#endif
/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
if (credential_changing)
imgp->proc->p_pdeathsig = 0;
if (credential_changing &&
#ifdef CAPABILITY_MODE
((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
#endif
(imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
(p->p_flag & P_TRACED) == 0) {
imgp->credential_setid = true;
VOP_UNLOCK(imgp->vp, 0);
imgp->newcred = crdup(oldcred);
if (attr.va_mode & S_ISUID) {
euip = uifind(attr.va_uid);
change_euid(imgp->newcred, euip);
}
vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
if (attr.va_mode & S_ISGID)
change_egid(imgp->newcred, attr.va_gid);
/*
* Implement correct POSIX saved-id behavior.
*
* XXXMAC: Note that the current logic will save the
* uid and gid if a MAC domain transition occurs, even
* though maybe it shouldn't.
*/
change_svuid(imgp->newcred, imgp->newcred->cr_uid);
change_svgid(imgp->newcred, imgp->newcred->cr_gid);
} else {
/*
* Implement correct POSIX saved-id behavior.
*
* XXX: It's not clear that the existing behavior is
* POSIX-compliant. A number of sources indicate that the
* saved uid/gid should only be updated if the new ruid is
* not equal to the old ruid, or the new euid is not equal
* to the old euid and the new euid is not equal to the old
* ruid. The FreeBSD code always updates the saved uid/gid.
* Also, this code uses the new (replaced) euid and egid as
* the source, which may or may not be the right ones to use.
*/
if (oldcred->cr_svuid != oldcred->cr_uid ||
oldcred->cr_svgid != oldcred->cr_gid) {
VOP_UNLOCK(imgp->vp, 0);
imgp->newcred = crdup(oldcred);
vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
change_svuid(imgp->newcred, imgp->newcred->cr_uid);
change_svgid(imgp->newcred, imgp->newcred->cr_gid);
}
}
/* The new credentials are installed into the process later. */
/*
* Do the best to calculate the full path to the image file.
*/
if (args->fname != NULL && args->fname[0] == '/')
imgp->execpath = args->fname;
else {
VOP_UNLOCK(imgp->vp, 0);
if (vn_fullpath(td, imgp->vp, &imgp->execpath,
&imgp->freepath) != 0)
imgp->execpath = args->fname;
vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
}
/*
* If the current process has a special image activator it
* wants to try first, call it. For example, emulating shell
* scripts differently.
*/
error = -1;
if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
error = img_first(imgp);
/*
* Loop through the list of image activators, calling each one.
* An activator returns -1 if there is no match, 0 on success,
* and an error otherwise.
*/
for (i = 0; error == -1 && execsw[i]; ++i) {
if (execsw[i]->ex_imgact == NULL ||
execsw[i]->ex_imgact == img_first) {
continue;
}
error = (*execsw[i]->ex_imgact)(imgp);
}
if (error) {
if (error == -1) {
if (textset == 0)
VOP_UNSET_TEXT(imgp->vp);
error = ENOEXEC;
}
goto exec_fail_dealloc;
}
/*
* Special interpreter operation, cleanup and loop up to try to
* activate the interpreter.
*/
if (imgp->interpreted) {
exec_unmap_first_page(imgp);
/*
* VV_TEXT needs to be unset for scripts. There is a short
* period before we determine that something is a script where
* VV_TEXT will be set. The vnode lock is held over this
* entire period so nothing should illegitimately be blocked.
*/
VOP_UNSET_TEXT(imgp->vp);
/* free name buffer and old vnode */
if (args->fname != NULL)
NDFREE(&nd, NDF_ONLY_PNBUF);
#ifdef MAC
mac_execve_interpreter_enter(newtextvp, &interpvplabel);
#endif
if (imgp->opened) {
VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
imgp->opened = 0;
}
vput(newtextvp);
vm_object_deallocate(imgp->object);
imgp->object = NULL;
imgp->credential_setid = false;
if (imgp->newcred != NULL) {
crfree(imgp->newcred);
imgp->newcred = NULL;
}
imgp->execpath = NULL;
free(imgp->freepath, M_TEMP);
imgp->freepath = NULL;
/* set new name to that of the interpreter */
NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
UIO_SYSSPACE, imgp->interpreter_name, td);
args->fname = imgp->interpreter_name;
goto interpret;
}
/*
* NB: We unlock the vnode here because it is believed that none
* of the sv_copyout_strings/sv_fixup operations require the vnode.
*/
VOP_UNLOCK(imgp->vp, 0);
if (disallow_high_osrel &&
P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
error = ENOEXEC;
uprintf("Osrel %d for image %s too high\n", p->p_osrel,
imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
goto exec_fail_dealloc;
}
/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
if (SV_PROC_FLAG(p, SV_CAPSICUM))
sys_cap_enter(td, NULL);
/*
* Copy out strings (args and env) and initialize stack base.
*/
stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
/*
* Stack setup.
*/
error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
if (error != 0) {
vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
goto exec_fail_dealloc;
}
if (args->fdp != NULL) {
/* Install a brand new file descriptor table. */
fdinstall_remapped(td, args->fdp);
args->fdp = NULL;
} else {
/*
* Keep on using the existing file descriptor table. For
* security and other reasons, the file descriptor table
* cannot be shared after an exec.
*/
fdunshare(td);
/* close files on exec */
fdcloseexec(td);
}
/*
* Malloc things before we need locks.
*/
i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
/* Cache arguments if they fit inside our allowance */
if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
newargs = pargs_alloc(i);
bcopy(imgp->args->begin_argv, newargs->ar_args, i);
}
/*
* For security and other reasons, signal handlers cannot
* be shared after an exec. The new process gets a copy of the old
* handlers. In execsigs(), the new process will have its signals
* reset.
*/
if (sigacts_shared(p->p_sigacts)) {
oldsigacts = p->p_sigacts;
newsigacts = sigacts_alloc();
sigacts_copy(newsigacts, oldsigacts);
}
vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
PROC_LOCK(p);
if (oldsigacts)
p->p_sigacts = newsigacts;
/* Stop profiling */
stopprofclock(p);
/* reset caught signals */
execsigs(p);
/* name this process - nameiexec(p, ndp) */
bzero(p->p_comm, sizeof(p->p_comm));
if (args->fname)
bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
#ifdef KTR
sched_clear_tdname(td);
#endif
/*
* mark as execed, wakeup the process that vforked (if any) and tell
* it that it now has its own resources back
*/
p->p_flag |= P_EXEC;
if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
p->p_flag2 &= ~P2_NOTRACE;
if (p->p_flag & P_PPWAIT) {
p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
cv_broadcast(&p->p_pwait);
/* STOPs are no longer ignored, arrange for AST */
signotify(td);
}
/*
* Implement image setuid/setgid installation.
*/
if (imgp->credential_setid) {
/*
* Turn off syscall tracing for set-id programs, except for
* root. Record any set-id flags first to make sure that
* we do not regain any tracing during a possible block.
*/
setsugid(p);
#ifdef KTRACE
if (p->p_tracecred != NULL &&
priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
ktrprocexec(p, &tracecred, &tracevp);
#endif
/*
* Close any file descriptors 0..2 that reference procfs,
* then make sure file descriptors 0..2 are in use.
*
* Both fdsetugidsafety() and fdcheckstd() may call functions
* taking sleepable locks, so temporarily drop our locks.
*/
PROC_UNLOCK(p);
VOP_UNLOCK(imgp->vp, 0);
fdsetugidsafety(td);
error = fdcheckstd(td);
vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
if (error != 0)
goto exec_fail_dealloc;
PROC_LOCK(p);
#ifdef MAC
if (will_transition) {
mac_vnode_execve_transition(oldcred, imgp->newcred,
imgp->vp, interpvplabel, imgp);
}
#endif
} else {
if (oldcred->cr_uid == oldcred->cr_ruid &&
oldcred->cr_gid == oldcred->cr_rgid)
p->p_flag &= ~P_SUGID;
}
/*
* Set the new credentials.
*/
if (imgp->newcred != NULL) {
proc_set_cred(p, imgp->newcred);
crfree(oldcred);
oldcred = NULL;
}
/*
* Store the vp for use in procfs. This vnode was referenced by namei
* or fgetvp_exec.
*/
oldtextvp = p->p_textvp;
p->p_textvp = newtextvp;
#ifdef KDTRACE_HOOKS
/*
* Tell the DTrace fasttrap provider about the exec if it
* has declared an interest.
*/
if (dtrace_fasttrap_exec)
dtrace_fasttrap_exec(p);
#endif
/*
* Notify others that we exec'd, and clear the P_INEXEC flag
* as we're now a bona fide freshly-execed process.
*/
KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
p->p_flag &= ~P_INEXEC;
/* clear "fork but no exec" flag, as we _are_ execing */
p->p_acflag &= ~AFORK;
/*
* Free any previous argument cache and replace it with
* the new argument cache, if any.
*/
oldargs = p->p_args;
p->p_args = newargs;
newargs = NULL;
PROC_UNLOCK(p);
#ifdef HWPMC_HOOKS
/*
* Check if system-wide sampling is in effect or if the
* current process is using PMCs. If so, do exec() time
* processing. This processing needs to happen AFTER the
* P_INEXEC flag is cleared.
*/
if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
VOP_UNLOCK(imgp->vp, 0);
pe.pm_credentialschanged = credential_changing;
pe.pm_entryaddr = imgp->entry_addr;
PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
}
#endif
/* Set values passed into the program in registers. */
(*p->p_sysent->sv_setregs)(td, imgp, (u_long)(uintptr_t)stack_base);
vfs_mark_atime(imgp->vp, td->td_ucred);
SDT_PROBE1(proc, , , exec__success, args->fname);
exec_fail_dealloc:
if (imgp->firstpage != NULL)
exec_unmap_first_page(imgp);
if (imgp->vp != NULL) {
if (args->fname)
NDFREE(&nd, NDF_ONLY_PNBUF);
if (imgp->opened)
VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
if (error != 0)
vput(imgp->vp);
else
VOP_UNLOCK(imgp->vp, 0);
}
if (imgp->object != NULL)
vm_object_deallocate(imgp->object);
free(imgp->freepath, M_TEMP);
if (error == 0) {
if (p->p_ptevents & PTRACE_EXEC) {
PROC_LOCK(p);
if (p->p_ptevents & PTRACE_EXEC)
td->td_dbgflags |= TDB_EXEC;
PROC_UNLOCK(p);
}
/*
* Stop the process here if its stop event mask has
* the S_EXEC bit set.
*/
STOPEVENT(p, S_EXEC, 0);
} else {
exec_fail:
/* we're done here, clear P_INEXEC */
PROC_LOCK(p);
p->p_flag &= ~P_INEXEC;
PROC_UNLOCK(p);
SDT_PROBE1(proc, , , exec__failure, error);
}
if (imgp->newcred != NULL && oldcred != NULL)
crfree(imgp->newcred);
#ifdef MAC
mac_execve_exit(imgp);
mac_execve_interpreter_exit(interpvplabel);
#endif
exec_free_args(args);
/*
* Handle deferred decrement of ref counts.
*/
if (oldtextvp != NULL)
vrele(oldtextvp);
#ifdef KTRACE
if (tracevp != NULL)
vrele(tracevp);
if (tracecred != NULL)
crfree(tracecred);
#endif
pargs_drop(oldargs);
pargs_drop(newargs);
if (oldsigacts != NULL)
sigacts_free(oldsigacts);
if (euip != NULL)
uifree(euip);
if (error && imgp->vmspace_destroyed) {
/* sorry, no more process anymore. exit gracefully */
exit1(td, 0, SIGABRT);
/* NOT REACHED */
}
#ifdef KTRACE
if (error == 0)
ktrprocctor(p);
#endif
/*
* We don't want cpu_set_syscall_retval() to overwrite any of
* the register values put in place by exec_setregs().
* Implementations of cpu_set_syscall_retval() will leave
* registers unmodified when returning EJUSTRETURN.
*/
return (error == 0 ? EJUSTRETURN : error);
}
/*
* task_join()
*
* Overview
* task_join() contains the actions that must be executed when the first
* member (curproc) of a newly created task joins it. It may never fail.
*
* The caller must make sure holdlwps() is called so that all other lwps are
* stopped prior to calling this function.
*
* NB: It returns with curproc->p_lock held.
*
* Return values
* Pointer to the old task.
*
* Caller's context
* cpu_lock must be held entering the function. It will acquire pidlock,
* p_crlock and p_lock during execution.
*/
task_t *
task_join(task_t *tk, uint_t flags)
{
proc_t *p = ttoproc(curthread);
task_t *prev_tk;
void *projbuf, *zonebuf;
zone_t *zone = tk->tk_zone;
projid_t projid = tk->tk_proj->kpj_id;
cred_t *oldcr;
/*
* We can't know for sure if holdlwps() was called, but we can check to
* ensure we're single-threaded.
*/
ASSERT(curthread == p->p_agenttp || p->p_lwprcnt == 1);
/*
* Changing the credential is always hard because we cannot
* allocate memory when holding locks but we don't know whether
* we need to change it. We first get a reference to the current
* cred if we need to change it. Then we create a credential
* with an updated project id. Finally we install it, first
* releasing the reference we had on the p_cred at the time we
* acquired the lock the first time and later we release the
* reference to p_cred at the time we acquired the lock the
* second time.
*/
mutex_enter(&p->p_crlock);
if (crgetprojid(p->p_cred) == projid)
oldcr = NULL;
else
crhold(oldcr = p->p_cred);
mutex_exit(&p->p_crlock);
if (oldcr != NULL) {
cred_t *newcr = crdup(oldcr);
crsetprojid(newcr, projid);
crfree(oldcr);
mutex_enter(&p->p_crlock);
oldcr = p->p_cred;
p->p_cred = newcr;
mutex_exit(&p->p_crlock);
crfree(oldcr);
}
/*
* Make sure that the number of processor sets is constant
* across this operation.
*/
ASSERT(MUTEX_HELD(&cpu_lock));
projbuf = fss_allocbuf(FSS_NPSET_BUF, FSS_ALLOC_PROJ);
zonebuf = fss_allocbuf(FSS_NPSET_BUF, FSS_ALLOC_ZONE);
mutex_enter(&pidlock);
mutex_enter(&p->p_lock);
prev_tk = p->p_task;
task_change(tk, p);
/*
* Now move threads one by one to their new project.
*/
changeproj(p, tk->tk_proj, zone, projbuf, zonebuf);
if (flags & TASK_FINAL)
p->p_task->tk_flags |= TASK_FINAL;
mutex_exit(&pidlock);
fss_freebuf(zonebuf, FSS_ALLOC_ZONE);
fss_freebuf(projbuf, FSS_ALLOC_PROJ);
return (prev_tk);
}
/*
* Guts of pr_spriv:
*
* Set the privileges of a process.
*
* In order to set the privileges, the setting process will need to
* have those privileges in its effective set in order to prevent
* specially privileged processes to easily gain additional privileges.
* Pre-existing privileges can be retained. To change any privileges,
* PRIV_PROC_OWNER needs to be asserted.
*
* In formula:
*
* S' <= S || S' <= S + Ea
*
* the new set must either be subset of the old set or a subset of
* the oldset merged with the effective set of the acting process; or just:
*
* S' <= S + Ea
*
* It's not legal to grow the limit set this way.
*
*/
int
priv_pr_spriv(proc_t *p, prpriv_t *prpriv, const cred_t *cr)
{
cred_t *oldcred;
cred_t *newcred;
int i;
int err = EPERM;
cred_priv_t *cp, *ocp;
priv_set_t eset;
ASSERT(MUTEX_HELD(&p->p_lock));
/*
* Set must have proper dimension; infosize must be absent
* or properly sized.
*/
if (prpriv->pr_nsets != PRIV_NSET ||
prpriv->pr_setsize != PRIV_SETSIZE ||
(prpriv->pr_infosize & (sizeof (uint32_t) - 1)) != 0 ||
prpriv->pr_infosize > priv_info->priv_infosize ||
prpriv->pr_infosize < 0)
return (EINVAL);
mutex_exit(&p->p_lock);
if (priv_proc_cred_perm(cr, p, &oldcred, VWRITE) != 0) {
mutex_enter(&p->p_lock);
return (EPERM);
}
newcred = crdup(oldcred);
/* Copy the privilege sets from prpriv to newcred */
bcopy(prpriv->pr_sets, CR_PRIVSETS(newcred), PRIV_SETBYTES);
cp = &newcred->cr_priv;
ocp = &oldcred->cr_priv;
eset = CR_OEPRIV(cr);
priv_intersect(&CR_LPRIV(oldcred), &eset);
/*
* Verify the constraints laid out:
* for the limit set, we require that the new set is a subset
* of the old limit set.
* for all other sets, we require that the new set is either a
* subset of the old set or a subset of the intersection of
* the old limit set and the effective set of the acting process.
*/
for (i = 0; i < PRIV_NSET; i++)
if (!priv_issubset(&cp->crprivs[i], &ocp->crprivs[i]) &&
(i == PRIV_LIMIT || !priv_issubset(&cp->crprivs[i], &eset)))
break;
crfree(oldcred);
if (i < PRIV_NSET || !priv_valid(newcred))
goto err;
/* Load the settable privilege information */
if (prpriv->pr_infosize > 0) {
char *x = (char *)prpriv + PRIV_PRPRIV_INFO_OFFSET(prpriv);
char *lastx = x + prpriv->pr_infosize;
while (x < lastx) {
priv_info_t *pi = (priv_info_t *)x;
priv_info_uint_t *pii;
switch (pi->priv_info_type) {
case PRIV_INFO_FLAGS:
pii = (priv_info_uint_t *)x;
if (pii->info.priv_info_size != sizeof (*pii)) {
err = EINVAL;
goto err;
}
CR_FLAGS(newcred) &= ~PRIV_USER;
CR_FLAGS(newcred) |= (pii->val & PRIV_USER);
break;
default:
err = EINVAL;
goto err;
}
/* Guarantee alignment and forward progress */
if ((pi->priv_info_size & (sizeof (uint32_t) - 1)) ||
pi->priv_info_size < sizeof (*pi) ||
lastx - x > pi->priv_info_size) {
err = EINVAL;
goto err;
}
x += pi->priv_info_size;
}
}
/*
* We'll try to copy the privilege aware flag; but since the
* privileges sets are all individually set, they are set
* as if we're privilege aware. If PRIV_AWARE wasn't set
* or was explicitely unset, we need to set the flag and then
* try to get rid of it.
*/
if ((CR_FLAGS(newcred) & PRIV_AWARE) == 0) {
CR_FLAGS(newcred) |= PRIV_AWARE;
priv_adjust_PA(newcred);
}
mutex_enter(&p->p_crlock);
oldcred = p->p_cred;
p->p_cred = newcred;
mutex_exit(&p->p_crlock);
crfree(oldcred);
mutex_enter(&p->p_lock);
return (0);
err:
crfree(newcred);
mutex_enter(&p->p_lock);
return (err);
}
void
osi_Init(void)
{
static int once = 0;
if (once++ > 0) /* just in case */
return;
#if defined(AFS_HPUX_ENV)
osi_InitGlock();
#else /* AFS_HPUX_ENV */
#if defined(AFS_GLOBAL_SUNLOCK)
#if defined(AFS_SGI62_ENV)
mutex_init(&afs_global_lock, MUTEX_DEFAULT, "afs_global_lock");
#elif defined(AFS_OSF_ENV)
usimple_lock_init(&afs_global_lock);
afs_global_owner = (thread_t) 0;
#elif defined(AFS_FBSD50_ENV)
#if defined(AFS_FBSD80_ENV) && defined(WITNESS)
/* "lock_initalized" (sic) can panic, checks a flag bit
* is unset _before_ init */
memset(&afs_global_mtx, 0, sizeof(struct mtx));
#endif
mtx_init(&afs_global_mtx, "AFS global lock", NULL, MTX_DEF);
#elif defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
#if !defined(AFS_DARWIN80_ENV)
lockinit(&afs_global_lock, PLOCK, "afs global lock", 0, 0);
#endif
afs_global_owner = 0;
#elif defined(AFS_AIX41_ENV)
lock_alloc((void *)&afs_global_lock, LOCK_ALLOC_PIN, 1, 1);
simple_lock_init((void *)&afs_global_lock);
#elif !defined(AFS_LINUX22_ENV)
/* Linux initialization in osi directory. Should move the others. */
mutex_init(&afs_global_lock, "afs_global_lock", MUTEX_DEFAULT, NULL);
#endif
#endif /* AFS_GLOBAL_SUNLOCK */
#endif /* AFS_HPUX_ENV */
if (!afs_osicred_initialized) {
#if defined(AFS_DARWIN80_ENV)
afs_osi_ctxtp_initialized = 0;
afs_osi_ctxtp = NULL; /* initialized in afs_Daemon since it has
a proc reference that cannot be changed */
#endif
#if defined(AFS_XBSD_ENV)
/* Can't just invent one, must use crget() because of mutex */
afs_osi_credp = crdup(osi_curcred());
#else
memset(&afs_osi_cred, 0, sizeof(struct AFS_UCRED));
#if defined(AFS_LINUX26_ENV)
afs_osi_cred.cr_group_info = groups_alloc(0);
#endif
#if defined(AFS_DARWIN80_ENV)
afs_osi_cred.cr_ref = 1; /* kauth_cred_get_ref needs 1 existing ref */
#else
crhold(&afs_osi_cred); /* don't let it evaporate */
#endif
afs_osi_credp = &afs_osi_cred;
#endif
afs_osicred_initialized = 1;
}
#ifdef AFS_SGI64_ENV
osi_flid.fl_pid = osi_flid.fl_sysid = 0;
#endif
init_et_to_sys_error();
}
int
domount(kthread_t *td, vnode_t *vp, const char *fstype, char *fspath,
char *fspec, int fsflags)
{
struct mount *mp;
struct vfsconf *vfsp;
struct ucred *newcr, *oldcr;
int error;
/*
* Be ultra-paranoid about making sure the type and fspath
* variables will fit in our mp buffers, including the
* terminating NUL.
*/
if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
return (ENAMETOOLONG);
vfsp = vfs_byname_kld(fstype, td, &error);
if (vfsp == NULL)
return (ENODEV);
if (vp->v_type != VDIR)
return (ENOTDIR);
simple_lock(&vp->v_interlock);
if ((vp->v_iflag & VI_MOUNT) != 0 ||
vp->v_mountedhere != NULL) {
simple_unlock(&vp->v_interlock);
return (EBUSY);
}
vp->v_iflag |= VI_MOUNT;
simple_unlock(&vp->v_interlock);
/*
* Allocate and initialize the filesystem.
*/
vn_lock(vp, LK_SHARED | LK_RETRY);
mp = vfs_mount_alloc(vp, vfsp, fspath, td);
VOP_UNLOCK(vp);
mp->mnt_optnew = NULL;
vfs_setmntopt(mp, "from", fspec, 0);
mp->mnt_optnew = mp->mnt_opt;
mp->mnt_opt = NULL;
/*
* Set the mount level flags.
* crdup() can sleep, so do it before acquiring a mutex.
*/
newcr = crdup(kcred);
MNT_ILOCK(mp);
if (fsflags & MNT_RDONLY)
mp->mnt_flag |= MNT_RDONLY;
mp->mnt_flag &=~ MNT_UPDATEMASK;
mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS);
/*
* Unprivileged user can trigger mounting a snapshot, but we don't want
* him to unmount it, so we switch to privileged credentials.
*/
oldcr = mp->mnt_cred;
mp->mnt_cred = newcr;
mp->mnt_stat.f_owner = mp->mnt_cred->cr_uid;
MNT_IUNLOCK(mp);
crfree(oldcr);
/*
* Mount the filesystem.
* XXX The final recipients of VFS_MOUNT just overwrite the ndp they
* get. No freeing of cn_pnbuf.
*/
error = VFS_MOUNT(mp, td);
if (!error) {
if (mp->mnt_opt != NULL)
vfs_freeopts(mp->mnt_opt);
mp->mnt_opt = mp->mnt_optnew;
(void)VFS_STATFS(mp, &mp->mnt_stat, td);
}
/*
* Prevent external consumers of mount options from reading
* mnt_optnew.
*/
mp->mnt_optnew = NULL;
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
/*
* Put the new filesystem on the mount list after root.
*/
#ifdef FREEBSD_NAMECACHE
cache_purge(vp);
#endif
if (!error) {
vnode_t *mvp;
simple_lock(&vp->v_interlock);
vp->v_iflag &= ~VI_MOUNT;
simple_unlock(&vp->v_interlock);
vp->v_mountedhere = mp;
mountlist_append(mp);
vfs_event_signal(NULL, VQ_MOUNT, 0);
if (VFS_ROOT(mp, LK_EXCLUSIVE, &mvp, td))
panic("mount: lost mount");
mountcheckdirs(vp, mvp);
vput(mvp);
VOP_UNLOCK(vp);
if ((mp->mnt_flag & MNT_RDONLY) == 0)
vfs_syncer_add_to_worklist(mp);
vfs_unbusy(mp, td);
vfs_mountedfrom(mp, fspec);
} else {
simple_lock(&vp->v_interlock);
vp->v_iflag &= ~VI_MOUNT;
simple_unlock(&vp->v_interlock);
VOP_UNLOCK(vp);
vfs_unbusy(mp, td);
vfs_mount_destroy(mp);
}
return (error);
}
int
mount_snapshot(kthread_t *td, vnode_t **vpp, const char *fstype, char *fspath,
char *fspec, int fsflags)
{
struct mount *mp;
struct vfsconf *vfsp;
struct ucred *cr;
vnode_t *vp;
int error;
/*
* Be ultra-paranoid about making sure the type and fspath
* variables will fit in our mp buffers, including the
* terminating NUL.
*/
if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
return (ENAMETOOLONG);
vfsp = vfs_byname_kld(fstype, td, &error);
if (vfsp == NULL)
return (ENODEV);
vp = *vpp;
if (vp->v_type != VDIR)
return (ENOTDIR);
/*
* We need vnode lock to protect v_mountedhere and vnode interlock
* to protect v_iflag.
*/
vn_lock(vp, LK_SHARED | LK_RETRY);
VI_LOCK(vp);
if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
VI_UNLOCK(vp);
VOP_UNLOCK(vp, 0);
return (EBUSY);
}
vp->v_iflag |= VI_MOUNT;
VI_UNLOCK(vp);
VOP_UNLOCK(vp, 0);
/*
* Allocate and initialize the filesystem.
*/
mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
mp->mnt_optnew = NULL;
vfs_setmntopt(mp, "from", fspec, 0);
mp->mnt_optnew = mp->mnt_opt;
mp->mnt_opt = NULL;
/*
* Set the mount level flags.
*/
mp->mnt_flag &= ~MNT_UPDATEMASK;
mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS);
/*
* Snapshots are always read-only.
*/
mp->mnt_flag |= MNT_RDONLY;
/*
* We don't want snapshots to be visible in regular
* mount(8) and df(1) output.
*/
mp->mnt_flag |= MNT_IGNORE;
/*
* Unprivileged user can trigger mounting a snapshot, but we don't want
* him to unmount it, so we switch to privileged of original mount.
*/
crfree(mp->mnt_cred);
mp->mnt_cred = crdup(vp->v_mount->mnt_cred);
mp->mnt_stat.f_owner = mp->mnt_cred->cr_uid;
/*
* XXX: This is evil, but we can't mount a snapshot as a regular user.
* XXX: Is is safe when snapshot is mounted from within a jail?
*/
cr = td->td_ucred;
td->td_ucred = kcred;
error = VFS_MOUNT(mp);
td->td_ucred = cr;
if (error == 0) {
if (mp->mnt_opt != NULL)
vfs_freeopts(mp->mnt_opt);
mp->mnt_opt = mp->mnt_optnew;
(void)VFS_STATFS(mp, &mp->mnt_stat);
}
/*
* Prevent external consumers of mount options from reading
* mnt_optnew.
*/
mp->mnt_optnew = NULL;
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
#ifdef FREEBSD_NAMECACHE
cache_purge(vp);
#endif
VI_LOCK(vp);
vp->v_iflag &= ~VI_MOUNT;
VI_UNLOCK(vp);
if (error == 0) {
vnode_t *mvp;
vp->v_mountedhere = mp;
/*
* Put the new filesystem on the mount list.
*/
mtx_lock(&mountlist_mtx);
TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
mtx_unlock(&mountlist_mtx);
vfs_event_signal(NULL, VQ_MOUNT, 0);
if (VFS_ROOT(mp, LK_EXCLUSIVE, &mvp))
panic("mount: lost mount");
vput(vp);
vfs_unbusy(mp);
*vpp = mvp;
} else {
vput(vp);
vfs_unbusy(mp);
vfs_mount_destroy(mp);
*vpp = NULL;
}
return (error);
}
