summaryrefslogtreecommitdiff
path: root/usr/src/uts/common/syscall
diff options
context:
space:
mode:
Diffstat (limited to 'usr/src/uts/common/syscall')
-rw-r--r--usr/src/uts/common/syscall/brandsys.c8
-rw-r--r--usr/src/uts/common/syscall/chdir.c29
-rw-r--r--usr/src/uts/common/syscall/fcntl.c3
-rw-r--r--usr/src/uts/common/syscall/memcntl.c8
-rw-r--r--usr/src/uts/common/syscall/open.c8
-rw-r--r--usr/src/uts/common/syscall/poll.c358
-rw-r--r--usr/src/uts/common/syscall/rusagesys.c1
-rw-r--r--usr/src/uts/common/syscall/rw.c375
-rw-r--r--usr/src/uts/common/syscall/sendfile.c19
-rw-r--r--usr/src/uts/common/syscall/stat.c2
-rw-r--r--usr/src/uts/common/syscall/sysconfig.c51
-rw-r--r--usr/src/uts/common/syscall/uadmin.c10
-rw-r--r--usr/src/uts/common/syscall/umount.c11
13 files changed, 526 insertions, 357 deletions
diff --git a/usr/src/uts/common/syscall/brandsys.c b/usr/src/uts/common/syscall/brandsys.c
index 9b4bd38baa..245ef9f14f 100644
--- a/usr/src/uts/common/syscall/brandsys.c
+++ b/usr/src/uts/common/syscall/brandsys.c
@@ -23,7 +23,9 @@
* Use is subject to license terms.
*/
-#pragma ident "%Z%%M% %I% %E% SMI"
+/*
+ * Copyright 2016 Joyent, Inc.
+ */
#include <sys/brand.h>
#include <sys/systm.h>
@@ -35,7 +37,7 @@
*/
int64_t
brandsys(int cmd, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3,
- uintptr_t arg4, uintptr_t arg5, uintptr_t arg6)
+ uintptr_t arg4)
{
struct proc *p = curthread->t_procp;
int64_t rval = 0;
@@ -49,7 +51,7 @@ brandsys(int cmd, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3,
return (set_errno(ENOSYS));
if ((err = ZBROP(p->p_zone)->b_brandsys(cmd, &rval, arg1, arg2, arg3,
- arg4, arg5, arg6)) != 0)
+ arg4)) != 0)
return (set_errno(err));
return (rval);
diff --git a/usr/src/uts/common/syscall/chdir.c b/usr/src/uts/common/syscall/chdir.c
index 84c924f570..deb5532b50 100644
--- a/usr/src/uts/common/syscall/chdir.c
+++ b/usr/src/uts/common/syscall/chdir.c
@@ -21,6 +21,7 @@
/*
* Copyright 2010 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
+ * Copyright 2016 Joyent, Inc.
*/
/* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
@@ -62,7 +63,7 @@
/*
* Change current working directory (".").
*/
-static int chdirec(vnode_t *, int ischroot, int do_traverse);
+static int chdirec(vnode_t *, boolean_t ischroot, boolean_t do_traverse);
int
chdir(char *fname)
@@ -78,7 +79,7 @@ lookup:
return (set_errno(error));
}
- error = chdirec(vp, 0, 1);
+ error = chdirec(vp, B_FALSE, B_TRUE);
if (error) {
if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
goto lookup;
@@ -102,7 +103,7 @@ fchdir(int fd)
vp = fp->f_vnode;
VN_HOLD(vp);
releasef(fd);
- error = chdirec(vp, 0, 0);
+ error = chdirec(vp, B_FALSE, B_FALSE);
if (error)
return (set_errno(error));
return (0);
@@ -125,7 +126,7 @@ lookup:
return (set_errno(error));
}
- error = chdirec(vp, 1, 1);
+ error = chdirec(vp, B_TRUE, B_TRUE);
if (error) {
if ((error == ESTALE) && fs_need_estale_retry(estale_retry++))
goto lookup;
@@ -152,18 +153,18 @@ fchroot(int fd)
vp = fp->f_vnode;
VN_HOLD(vp);
releasef(fd);
- error = chdirec(vp, 1, 0);
+ error = chdirec(vp, B_TRUE, B_FALSE);
if (error)
return (set_errno(error));
return (0);
}
static int
-chdirec(vnode_t *vp, int ischroot, int do_traverse)
+chdirec_common(proc_t *pp, vnode_t *vp, boolean_t ischroot,
+ boolean_t do_traverse)
{
int error;
vnode_t *oldvp;
- proc_t *pp = curproc;
vnode_t **vpp;
refstr_t *cwd;
int newcwd = 1;
@@ -194,7 +195,7 @@ chdirec(vnode_t *vp, int ischroot, int do_traverse)
if (ischroot) {
struct vattr tattr;
struct vattr rattr;
- vnode_t *zonevp = curproc->p_zone->zone_rootvp;
+ vnode_t *zonevp = pp->p_zone->zone_rootvp;
tattr.va_mask = AT_FSID|AT_NODEID;
if (error = VOP_GETATTR(vp, &tattr, 0, CRED(), NULL))
@@ -243,3 +244,15 @@ bad:
VN_RELE(vp);
return (error);
}
+
+int
+chdir_proc(proc_t *pp, vnode_t *vp, boolean_t ischroot, boolean_t do_traverse)
+{
+ return (chdirec_common(pp, vp, ischroot, do_traverse));
+}
+
+static int
+chdirec(vnode_t *vp, boolean_t ischroot, boolean_t do_traverse)
+{
+ return (chdirec_common(curproc, vp, ischroot, do_traverse));
+}
diff --git a/usr/src/uts/common/syscall/fcntl.c b/usr/src/uts/common/syscall/fcntl.c
index 7b787a4acb..b029d92f1b 100644
--- a/usr/src/uts/common/syscall/fcntl.c
+++ b/usr/src/uts/common/syscall/fcntl.c
@@ -54,7 +54,8 @@
#include <sys/cmn_err.h>
-static int flock_check(vnode_t *, flock64_t *, offset_t, offset_t);
+/* This is global so that it can be used by brand emulation. */
+int flock_check(vnode_t *, flock64_t *, offset_t, offset_t);
static int flock_get_start(vnode_t *, flock64_t *, offset_t, u_offset_t *);
static void fd_too_big(proc_t *);
diff --git a/usr/src/uts/common/syscall/memcntl.c b/usr/src/uts/common/syscall/memcntl.c
index 1ee4b6a395..721f884a7e 100644
--- a/usr/src/uts/common/syscall/memcntl.c
+++ b/usr/src/uts/common/syscall/memcntl.c
@@ -115,13 +115,17 @@ memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg, int attr, int mask)
* MS_SYNC used to be defined to be zero but is now non-zero.
* For binary compatibility we still accept zero
* (the absence of MS_ASYNC) to mean the same thing.
+ * Binary compatibility is not an issue for MS_INVALCURPROC.
*/
iarg = (uintptr_t)arg;
if ((iarg & ~MS_INVALIDATE) == 0)
iarg |= MS_SYNC;
- if (((iarg & ~(MS_SYNC|MS_ASYNC|MS_INVALIDATE)) != 0) ||
- ((iarg & (MS_SYNC|MS_ASYNC)) == (MS_SYNC|MS_ASYNC))) {
+ if (((iarg &
+ ~(MS_SYNC|MS_ASYNC|MS_INVALIDATE|MS_INVALCURPROC)) != 0) ||
+ ((iarg & (MS_SYNC|MS_ASYNC)) == (MS_SYNC|MS_ASYNC)) ||
+ ((iarg & (MS_INVALIDATE|MS_INVALCURPROC)) ==
+ (MS_INVALIDATE|MS_INVALCURPROC))) {
error = set_errno(EINVAL);
} else {
error = as_ctl(as, addr, len, cmd, attr, iarg, NULL, 0);
diff --git a/usr/src/uts/common/syscall/open.c b/usr/src/uts/common/syscall/open.c
index edb04c824b..874e31869c 100644
--- a/usr/src/uts/common/syscall/open.c
+++ b/usr/src/uts/common/syscall/open.c
@@ -74,12 +74,12 @@ copen(int startfd, char *fname, int filemode, int createmode)
if (filemode & (FSEARCH|FEXEC)) {
/*
- * Must be one or the other and neither FREAD nor FWRITE
+ * Must be one or the other.
* Must not be any of FAPPEND FCREAT FTRUNC FXATTR FXATTRDIROPEN
- * XXX: Should these just be silently ignored?
+ * XXX: Should these just be silently ignored like we
+ * silently ignore FREAD|FWRITE?
*/
- if ((filemode & (FREAD|FWRITE)) ||
- (filemode & (FSEARCH|FEXEC)) == (FSEARCH|FEXEC) ||
+ if ((filemode & (FSEARCH|FEXEC)) == (FSEARCH|FEXEC) ||
(filemode & (FAPPEND|FCREAT|FTRUNC|FXATTR|FXATTRDIROPEN)))
return (set_errno(EINVAL));
}
diff --git a/usr/src/uts/common/syscall/poll.c b/usr/src/uts/common/syscall/poll.c
index ae34556f14..2214bacaf8 100644
--- a/usr/src/uts/common/syscall/poll.c
+++ b/usr/src/uts/common/syscall/poll.c
@@ -29,7 +29,7 @@
/*
* Copyright (c) 2012, 2016 by Delphix. All rights reserved.
- * Copyright 2015, Joyent, Inc.
+ * Copyright (c) 2017, Joyent, Inc.
*/
/*
@@ -317,20 +317,57 @@ polllock(pollhead_t *php, kmutex_t *lp)
return (0);
}
-static int
-poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
+int
+poll_copyin(pollstate_t *ps, pollfd_t *fds, nfds_t nfds)
+{
+ pollfd_t *pollfdp;
+ nfds_t old_nfds;
+
+ /*
+ * NOTE: for performance, buffers are saved across poll() calls.
+ * The theory is that if a process polls heavily, it tends to poll
+ * on the same set of descriptors. Therefore, we only reallocate
+ * buffers when nfds changes. There is no hysteresis control,
+ * because there is no data to suggest that this is necessary;
+ * the penalty of reallocating is not *that* great in any event.
+ */
+ old_nfds = ps->ps_nfds;
+ if (nfds != old_nfds) {
+ kmem_free(ps->ps_pollfd, old_nfds * sizeof (pollfd_t));
+ pollfdp = kmem_alloc(nfds * sizeof (pollfd_t), KM_SLEEP);
+ ps->ps_pollfd = pollfdp;
+ ps->ps_nfds = nfds;
+ }
+
+ pollfdp = ps->ps_pollfd;
+ if (copyin(fds, pollfdp, nfds * sizeof (pollfd_t))) {
+ return (EFAULT);
+ }
+
+ if (fds == NULL) {
+ /*
+ * If the process has page 0 mapped, then the copyin() above
+ * will succeed even if fds is NULL. However, our cached
+ * poll lists are keyed by the address of the passed-in fds
+ * structure, and we use the value NULL to indicate an unused
+ * poll cache list entry. As such, we elect not to support
+ * NULL as a valid (user) memory address and fail the poll()
+ * call.
+ */
+ return (EFAULT);
+ }
+ return (0);
+}
+
+int
+poll_common(pollstate_t *ps, pollfd_t *fds, nfds_t nfds, timespec_t *tsp,
+ int *fdcnt)
{
kthread_t *t = curthread;
- klwp_t *lwp = ttolwp(t);
- proc_t *p = ttoproc(t);
- int fdcnt = 0;
- int i;
hrtime_t deadline; /* hrtime value when we want to return */
pollfd_t *pollfdp;
- pollstate_t *ps;
pollcache_t *pcp;
int error = 0;
- nfds_t old_nfds;
int cacheindex = 0; /* which cache set is used */
/*
@@ -340,33 +377,34 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
deadline = -1;
} else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) {
deadline = 0;
+ } else if (tsp->tv_sec >= HRTIME_MAX/NANOSEC) {
+ /* Use an indefinite timeout if tv_sec would cause overflow */
+ deadline = -1;
} else {
+ /*
+ * The above check, when combined with the protections offered
+ * by itimerspecfix (ensuring that neither field is negative
+ * and that tv_nsec represents less than a whole second), will
+ * prevent overflow during the conversion from timespec_t to
+ * uhrtime_t.
+ */
+ uhrtime_t utime = tsp->tv_sec * NANOSEC;
+ utime += tsp->tv_nsec;
+
/* They must wait at least a tick. */
- deadline = ((hrtime_t)tsp->tv_sec * NANOSEC) + tsp->tv_nsec;
- deadline = MAX(deadline, nsec_per_tick);
- deadline += gethrtime();
- }
+ utime = MAX(utime, nsec_per_tick);
- /*
- * Reset our signal mask, if requested.
- */
- if (ksetp != NULL) {
- mutex_enter(&p->p_lock);
- schedctl_finish_sigblock(t);
- lwp->lwp_sigoldmask = t->t_hold;
- t->t_hold = *ksetp;
- t->t_flag |= T_TOMASK;
/*
- * Call cv_reltimedwait_sig() just to check for signals.
- * We will return immediately with either 0 or -1.
+ * Since utime has an upper bound of HRTIME_MAX, adding the
+ * gethrtime() result cannot incur an overflow as the unsigned
+ * type has an adequate bound.
*/
- if (!cv_reltimedwait_sig(&t->t_delay_cv, &p->p_lock, 0,
- TR_CLOCK_TICK)) {
- mutex_exit(&p->p_lock);
- error = EINTR;
- goto pollout;
+ utime += (uhrtime_t)gethrtime();
+ if (utime > HRTIME_MAX) {
+ deadline = -1;
+ } else {
+ deadline = (hrtime_t)utime;
}
- mutex_exit(&p->p_lock);
}
/*
@@ -374,6 +412,7 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
* If yes then bypass all the other stuff and make it sleep.
*/
if (nfds == 0) {
+ *fdcnt = 0;
/*
* Sleep until we have passed the requested future
* time or until interrupted by a signal.
@@ -385,66 +424,14 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
&t->t_delay_lock, deadline)) > 0)
continue;
mutex_exit(&t->t_delay_lock);
- error = (error == 0) ? EINTR : 0;
+ return ((error == 0) ? EINTR : 0);
}
- goto pollout;
- }
-
- if (nfds > p->p_fno_ctl) {
- mutex_enter(&p->p_lock);
- (void) rctl_action(rctlproc_legacy[RLIMIT_NOFILE],
- p->p_rctls, p, RCA_SAFE);
- mutex_exit(&p->p_lock);
- error = EINVAL;
- goto pollout;
- }
-
- /*
- * Need to allocate memory for pollstate before anything because
- * the mutex and cv are created in this space
- */
- ps = pollstate_create();
-
- if (ps->ps_pcache == NULL)
- ps->ps_pcache = pcache_alloc();
- pcp = ps->ps_pcache;
-
- /*
- * NOTE: for performance, buffers are saved across poll() calls.
- * The theory is that if a process polls heavily, it tends to poll
- * on the same set of descriptors. Therefore, we only reallocate
- * buffers when nfds changes. There is no hysteresis control,
- * because there is no data to suggest that this is necessary;
- * the penalty of reallocating is not *that* great in any event.
- */
- old_nfds = ps->ps_nfds;
- if (nfds != old_nfds) {
-
- kmem_free(ps->ps_pollfd, old_nfds * sizeof (pollfd_t));
- pollfdp = kmem_alloc(nfds * sizeof (pollfd_t), KM_SLEEP);
- ps->ps_pollfd = pollfdp;
- ps->ps_nfds = nfds;
+ return (0);
}
+ VERIFY(ps != NULL);
pollfdp = ps->ps_pollfd;
- if (copyin(fds, pollfdp, nfds * sizeof (pollfd_t))) {
- error = EFAULT;
- goto pollout;
- }
-
- if (fds == NULL) {
- /*
- * If the process has page 0 mapped, then the copyin() above
- * will succeed even if fds is NULL. However, our cached
- * poll lists are keyed by the address of the passed-in fds
- * structure, and we use the value NULL to indicate an unused
- * poll cache list entry. As such, we elect not to support
- * NULL as a valid (user) memory address and fail the poll()
- * call.
- */
- error = EINVAL;
- goto pollout;
- }
+ VERIFY(pollfdp != NULL);
/*
* If this thread polls for the first time, allocate ALL poll
@@ -460,10 +447,10 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
/*
* poll and cache this poll fd list in ps_pcacheset[0].
*/
- error = pcacheset_cache_list(ps, fds, &fdcnt, cacheindex);
- if (fdcnt || error) {
+ error = pcacheset_cache_list(ps, fds, fdcnt, cacheindex);
+ if (error || *fdcnt) {
mutex_exit(&ps->ps_lock);
- goto pollout;
+ return (error);
}
} else {
pollcacheset_t *pcset = ps->ps_pcacheset;
@@ -488,11 +475,11 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
* the callee will guarantee the consistency
* of cached poll list and cache content.
*/
- error = pcacheset_resolve(ps, nfds, &fdcnt,
+ error = pcacheset_resolve(ps, nfds, fdcnt,
cacheindex);
if (error) {
mutex_exit(&ps->ps_lock);
- goto pollout;
+ return (error);
}
break;
}
@@ -509,11 +496,11 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
* found an unused entry. Use it to cache
* this poll list.
*/
- error = pcacheset_cache_list(ps, fds, &fdcnt,
+ error = pcacheset_cache_list(ps, fds, fdcnt,
cacheindex);
- if (fdcnt || error) {
+ if (error || *fdcnt) {
mutex_exit(&ps->ps_lock);
- goto pollout;
+ return (error);
}
break;
}
@@ -527,10 +514,10 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
cacheindex = pcacheset_replace(ps);
ASSERT(cacheindex < ps->ps_nsets);
pcset[cacheindex].pcs_usradr = (uintptr_t)fds;
- error = pcacheset_resolve(ps, nfds, &fdcnt, cacheindex);
+ error = pcacheset_resolve(ps, nfds, fdcnt, cacheindex);
if (error) {
mutex_exit(&ps->ps_lock);
- goto pollout;
+ return (error);
}
}
}
@@ -548,8 +535,8 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
mutex_enter(&pcp->pc_lock);
for (;;) {
pcp->pc_flag = 0;
- error = pcache_poll(pollfdp, ps, nfds, &fdcnt, cacheindex);
- if (fdcnt || error) {
+ error = pcache_poll(pollfdp, ps, nfds, fdcnt, cacheindex);
+ if (error || *fdcnt) {
mutex_exit(&pcp->pc_lock);
mutex_exit(&ps->ps_lock);
break;
@@ -595,13 +582,116 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
mutex_enter(&pcp->pc_lock);
}
+ return (error);
+}
+
+/*
+ * This is the system call trap that poll(),
+ * select() and pselect() are built upon.
+ * It is a private interface between libc and the kernel.
+ */
+int
+pollsys(pollfd_t *fds, nfds_t nfds, timespec_t *timeoutp, sigset_t *setp)
+{
+ kthread_t *t = curthread;
+ klwp_t *lwp = ttolwp(t);
+ proc_t *p = ttoproc(t);
+ timespec_t ts;
+ timespec_t *tsp;
+ k_sigset_t kset;
+ pollstate_t *ps = NULL;
+ pollfd_t *pollfdp = NULL;
+ int error = 0, fdcnt = 0;
+
+ /*
+ * Copy in timeout
+ */
+ if (timeoutp == NULL) {
+ tsp = NULL;
+ } else {
+ if (get_udatamodel() == DATAMODEL_NATIVE) {
+ if (copyin(timeoutp, &ts, sizeof (ts)))
+ return (set_errno(EFAULT));
+ } else {
+ timespec32_t ts32;
+
+ if (copyin(timeoutp, &ts32, sizeof (ts32)))
+ return (set_errno(EFAULT));
+ TIMESPEC32_TO_TIMESPEC(&ts, &ts32)
+ }
+
+ if (itimerspecfix(&ts))
+ return (set_errno(EINVAL));
+ tsp = &ts;
+ }
+
+ /*
+ * Copy in and reset signal mask, if requested.
+ */
+ if (setp != NULL) {
+ sigset_t set;
+
+ if (copyin(setp, &set, sizeof (set)))
+ return (set_errno(EFAULT));
+ sigutok(&set, &kset);
+
+ mutex_enter(&p->p_lock);
+ schedctl_finish_sigblock(t);
+ lwp->lwp_sigoldmask = t->t_hold;
+ t->t_hold = kset;
+ t->t_flag |= T_TOMASK;
+ /*
+ * Call cv_reltimedwait_sig() just to check for signals.
+ * We will return immediately with either 0 or -1.
+ */
+ if (!cv_reltimedwait_sig(&t->t_delay_cv, &p->p_lock, 0,
+ TR_CLOCK_TICK)) {
+ mutex_exit(&p->p_lock);
+ error = EINTR;
+ goto pollout;
+ }
+ mutex_exit(&p->p_lock);
+ }
+
+ /*
+ * Initialize pollstate and copy in pollfd data if present.
+ * If nfds == 0, we will skip all of the copying and check steps and
+ * proceed directly into poll_common to process the supplied timeout.
+ */
+ if (nfds != 0) {
+ if (nfds > p->p_fno_ctl) {
+ mutex_enter(&p->p_lock);
+ (void) rctl_action(rctlproc_legacy[RLIMIT_NOFILE],
+ p->p_rctls, p, RCA_SAFE);
+ mutex_exit(&p->p_lock);
+ error = EINVAL;
+ goto pollout;
+ }
+
+ /*
+ * Need to allocate memory for pollstate before anything
+ * because the mutex and cv are created in this space
+ */
+ ps = pollstate_create();
+ if (ps->ps_pcache == NULL)
+ ps->ps_pcache = pcache_alloc();
+
+ if ((error = poll_copyin(ps, fds, nfds)) != 0)
+ goto pollout;
+ pollfdp = ps->ps_pollfd;
+ }
+
+ /*
+ * Perform the actual poll.
+ */
+ error = poll_common(ps, fds, nfds, tsp, &fdcnt);
+
pollout:
/*
- * If we changed the signal mask but we received
- * no signal then restore the signal mask.
- * Otherwise psig() will deal with the signal mask.
+ * If we changed the signal mask but we received no signal then restore
+ * the signal mask. Otherwise psig() will deal with the signal mask.
*/
- if (ksetp != NULL) {
+ if (setp != NULL) {
mutex_enter(&p->p_lock);
if (lwp->lwp_cursig == 0) {
t->t_hold = lwp->lwp_sigoldmask;
@@ -612,12 +702,10 @@ pollout:
if (error)
return (set_errno(error));
-
/*
* Copy out the events and return the fdcnt to the user.
*/
- if (nfds != 0 &&
- copyout(pollfdp, fds, nfds * sizeof (pollfd_t)))
+ if (nfds != 0 && copyout(pollfdp, fds, nfds * sizeof (pollfd_t)))
return (set_errno(EFAULT));
#ifdef DEBUG
@@ -625,7 +713,7 @@ pollout:
* Another sanity check:
*/
if (fdcnt) {
- int reventcnt = 0;
+ int i, reventcnt = 0;
for (i = 0; i < nfds; i++) {
if (pollfdp[i].fd < 0) {
@@ -638,6 +726,8 @@ pollout:
}
ASSERT(fdcnt == reventcnt);
} else {
+ int i;
+
for (i = 0; i < nfds; i++) {
ASSERT(pollfdp[i].revents == 0);
}
@@ -648,52 +738,6 @@ pollout:
}
/*
- * This is the system call trap that poll(),
- * select() and pselect() are built upon.
- * It is a private interface between libc and the kernel.
- */
-int
-pollsys(pollfd_t *fds, nfds_t nfds, timespec_t *timeoutp, sigset_t *setp)
-{
- timespec_t ts;
- timespec_t *tsp;
- sigset_t set;
- k_sigset_t kset;
- k_sigset_t *ksetp;
- model_t datamodel = get_udatamodel();
-
- if (timeoutp == NULL)
- tsp = NULL;
- else {
- if (datamodel == DATAMODEL_NATIVE) {
- if (copyin(timeoutp, &ts, sizeof (ts)))
- return (set_errno(EFAULT));
- } else {
- timespec32_t ts32;
-
- if (copyin(timeoutp, &ts32, sizeof (ts32)))
- return (set_errno(EFAULT));
- TIMESPEC32_TO_TIMESPEC(&ts, &ts32)
- }
-
- if (itimerspecfix(&ts))
- return (set_errno(EINVAL));
- tsp = &ts;
- }
-
- if (setp == NULL)
- ksetp = NULL;
- else {
- if (copyin(setp, &set, sizeof (set)))
- return (set_errno(EFAULT));
- sigutok(&set, &kset);
- ksetp = &kset;
- }
-
- return (poll_common(fds, nfds, tsp, ksetp));
-}
-
-/*
* Clean up any state left around by poll(2). Called when a thread exits.
*/
void
@@ -1277,8 +1321,8 @@ pcache_insert(pollstate_t *ps, file_t *fp, pollfd_t *pollfdp, int *fdcntp,
* be OK too.
*/
ASSERT(curthread->t_pollcache == NULL);
- error = VOP_POLL(fp->f_vnode, pollfdp->events, 0, &pollfdp->revents,
- &memphp, NULL);
+ error = VOP_POLL(fp->f_vnode, pollfdp->events | ps->ps_implicit_ev, 0,
+ &pollfdp->revents, &memphp, NULL);
if (error) {
return (error);
}
@@ -1992,7 +2036,8 @@ retry:
* flag.
*/
ASSERT(curthread->t_pollcache == NULL);
- error = VOP_POLL(fp->f_vnode, pollfdp[entry].events, 0,
+ error = VOP_POLL(fp->f_vnode,
+ pollfdp[entry].events | ps->ps_implicit_ev, 0,
&pollfdp[entry].revents, &php, NULL);
/*
* releasef after completely done with this cached
@@ -2291,6 +2336,7 @@ pollstate_create()
} else {
ASSERT(ps->ps_depth == 0);
ASSERT(ps->ps_flags == 0);
+ ASSERT(ps->ps_implicit_ev == 0);
ASSERT(ps->ps_pc_stack[0] == 0);
}
return (ps);
@@ -3025,7 +3071,7 @@ plist_chkdupfd(file_t *fp, polldat_t *pdp, pollstate_t *psp, pollfd_t *pollfdp,
php = NULL;
ASSERT(curthread->t_pollcache == NULL);
error = VOP_POLL(fp->f_vnode,
- pollfdp[i].events, 0,
+ pollfdp[i].events | psp->ps_implicit_ev, 0,
&pollfdp[i].revents, &php, NULL);
if (error) {
return (error);
diff --git a/usr/src/uts/common/syscall/rusagesys.c b/usr/src/uts/common/syscall/rusagesys.c
index 3e0e63f4c0..09f3266ab4 100644
--- a/usr/src/uts/common/syscall/rusagesys.c
+++ b/usr/src/uts/common/syscall/rusagesys.c
@@ -21,6 +21,7 @@
/*
* Copyright 2008 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
+ * Copyright 2017 Joyent, Inc. All rights reserved.
*/
/*
diff --git a/usr/src/uts/common/syscall/rw.c b/usr/src/uts/common/syscall/rw.c
index a28894b2c9..23f03e841d 100644
--- a/usr/src/uts/common/syscall/rw.c
+++ b/usr/src/uts/common/syscall/rw.c
@@ -22,7 +22,7 @@
/*
* Copyright 2008 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
- * Copyright (c) 2015, Joyent, Inc. All rights reserved.
+ * Copyright 2017, Joyent, Inc.
*/
/* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
@@ -50,6 +50,7 @@
#include <sys/debug.h>
#include <sys/rctl.h>
#include <sys/nbmlock.h>
+#include <sys/limits.h>
#define COPYOUT_MAX_CACHE (1<<17) /* 128K */
@@ -607,19 +608,12 @@ out:
return (bcount);
}
-/*
- * XXX -- The SVID refers to IOV_MAX, but doesn't define it. Grrrr....
- * XXX -- However, SVVS expects readv() and writev() to fail if
- * XXX -- iovcnt > 16 (yes, it's hard-coded in the SVVS source),
- * XXX -- so I guess that's the "interface".
- */
-#define DEF_IOV_MAX 16
-
ssize_t
readv(int fdes, struct iovec *iovp, int iovcnt)
{
struct uio auio;
- struct iovec aiov[DEF_IOV_MAX];
+ struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+ int aiovlen = 0;
file_t *fp;
register vnode_t *vp;
struct cpu *cp;
@@ -630,9 +624,14 @@ readv(int fdes, struct iovec *iovp, int iovcnt)
u_offset_t fileoff;
int in_crit = 0;
- if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX)
+ if (iovcnt <= 0 || iovcnt > IOV_MAX)
return (set_errno(EINVAL));
+ if (iovcnt > IOV_MAX_STACK) {
+ aiovlen = iovcnt * sizeof (iovec_t);
+ aiov = kmem_alloc(aiovlen, KM_SLEEP);
+ }
+
#ifdef _SYSCALL32_IMPL
/*
* 32-bit callers need to have their iovec expanded,
@@ -640,36 +639,63 @@ readv(int fdes, struct iovec *iovp, int iovcnt)
* of data in a single call.
*/
if (get_udatamodel() == DATAMODEL_ILP32) {
- struct iovec32 aiov32[DEF_IOV_MAX];
+ struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+ int aiov32len;
ssize32_t count32;
- if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32)))
+ aiov32len = iovcnt * sizeof (iovec32_t);
+ if (aiovlen != 0)
+ aiov32 = kmem_alloc(aiov32len, KM_SLEEP);
+
+ if (copyin(iovp, aiov32, aiov32len)) {
+ if (aiovlen != 0) {
+ kmem_free(aiov32, aiov32len);
+ kmem_free(aiov, aiovlen);
+ }
return (set_errno(EFAULT));
+ }
count32 = 0;
for (i = 0; i < iovcnt; i++) {
ssize32_t iovlen32 = aiov32[i].iov_len;
count32 += iovlen32;
- if (iovlen32 < 0 || count32 < 0)
+ if (iovlen32 < 0 || count32 < 0) {
+ if (aiovlen != 0) {
+ kmem_free(aiov32, aiov32len);
+ kmem_free(aiov, aiovlen);
+ }
return (set_errno(EINVAL));
+ }
aiov[i].iov_len = iovlen32;
aiov[i].iov_base =
(caddr_t)(uintptr_t)aiov32[i].iov_base;
}
+
+ if (aiovlen != 0)
+ kmem_free(aiov32, aiov32len);
} else
#endif
- if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec)))
+ if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EFAULT));
+ }
count = 0;
for (i = 0; i < iovcnt; i++) {
ssize_t iovlen = aiov[i].iov_len;
count += iovlen;
- if (iovlen < 0 || count < 0)
+ if (iovlen < 0 || count < 0) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EINVAL));
+ }
}
- if ((fp = getf(fdes)) == NULL)
+ if ((fp = getf(fdes)) == NULL) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EBADF));
+ }
if (((fflag = fp->f_flag) & FREAD) == 0) {
error = EBADF;
goto out;
@@ -768,6 +794,8 @@ out:
if (in_crit)
nbl_end_crit(vp);
releasef(fdes);
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
if (error)
return (set_errno(error));
return (count);
@@ -777,7 +805,8 @@ ssize_t
writev(int fdes, struct iovec *iovp, int iovcnt)
{
struct uio auio;
- struct iovec aiov[DEF_IOV_MAX];
+ struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+ int aiovlen = 0;
file_t *fp;
register vnode_t *vp;
struct cpu *cp;
@@ -788,9 +817,14 @@ writev(int fdes, struct iovec *iovp, int iovcnt)
u_offset_t fileoff;
int in_crit = 0;
- if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX)
+ if (iovcnt <= 0 || iovcnt > IOV_MAX)
return (set_errno(EINVAL));
+ if (iovcnt > IOV_MAX_STACK) {
+ aiovlen = iovcnt * sizeof (iovec_t);
+ aiov = kmem_alloc(aiovlen, KM_SLEEP);
+ }
+
#ifdef _SYSCALL32_IMPL
/*
* 32-bit callers need to have their iovec expanded,
@@ -798,36 +832,62 @@ writev(int fdes, struct iovec *iovp, int iovcnt)
* of data in a single call.
*/
if (get_udatamodel() == DATAMODEL_ILP32) {
- struct iovec32 aiov32[DEF_IOV_MAX];
+ struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+ int aiov32len;
ssize32_t count32;
- if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32)))
+ aiov32len = iovcnt * sizeof (iovec32_t);
+ if (aiovlen != 0)
+ aiov32 = kmem_alloc(aiov32len, KM_SLEEP);
+
+ if (copyin(iovp, aiov32, aiov32len)) {
+ if (aiovlen != 0) {
+ kmem_free(aiov32, aiov32len);
+ kmem_free(aiov, aiovlen);
+ }
return (set_errno(EFAULT));
+ }
count32 = 0;
for (i = 0; i < iovcnt; i++) {
ssize32_t iovlen = aiov32[i].iov_len;
count32 += iovlen;
- if (iovlen < 0 || count32 < 0)
+ if (iovlen < 0 || count32 < 0) {
+ if (aiovlen != 0) {
+ kmem_free(aiov32, aiov32len);
+ kmem_free(aiov, aiovlen);
+ }
return (set_errno(EINVAL));
+ }
aiov[i].iov_len = iovlen;
aiov[i].iov_base =
(caddr_t)(uintptr_t)aiov32[i].iov_base;
}
+ if (aiovlen != 0)
+ kmem_free(aiov32, aiov32len);
} else
#endif
- if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec)))
+ if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EFAULT));
+ }
count = 0;
for (i = 0; i < iovcnt; i++) {
ssize_t iovlen = aiov[i].iov_len;
count += iovlen;
- if (iovlen < 0 || count < 0)
+ if (iovlen < 0 || count < 0) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EINVAL));
+ }
}
- if ((fp = getf(fdes)) == NULL)
+ if ((fp = getf(fdes)) == NULL) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EBADF));
+ }
if (((fflag = fp->f_flag) & FWRITE) == 0) {
error = EBADF;
goto out;
@@ -917,6 +977,8 @@ out:
if (in_crit)
nbl_end_crit(vp);
releasef(fdes);
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
if (error)
return (set_errno(error));
return (count);
@@ -927,7 +989,8 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
off_t extended_offset)
{
struct uio auio;
- struct iovec aiov[DEF_IOV_MAX];
+ struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+ int aiovlen = 0;
file_t *fp;
register vnode_t *vp;
struct cpu *cp;
@@ -936,25 +999,35 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
int error = 0;
int i;
+ /*
+ * In a 64-bit kernel, this interface supports native 64-bit
+ * applications as well as 32-bit applications using both standard and
+ * large-file access. For 32-bit large-file aware applications, the
+ * offset is passed as two parameters which are joined into the actual
+ * offset used. The 64-bit libc always passes 0 for the extended_offset.
+ * Note that off_t is a signed value, but the preadv/pwritev API treats
+ * the offset as a position in the file for the operation, so passing
+ * a negative value will likely fail the maximum offset checks below
+ * because we convert it to an unsigned value which will be larger than
+ * the maximum valid offset.
+ */
#if defined(_SYSCALL32_IMPL) || defined(_ILP32)
u_offset_t fileoff = ((u_offset_t)extended_offset << 32) |
(u_offset_t)offset;
#else /* _SYSCALL32_IMPL || _ILP32 */
u_offset_t fileoff = (u_offset_t)(ulong_t)offset;
#endif /* _SYSCALL32_IMPR || _ILP32 */
-#ifdef _SYSCALL32_IMPL
- const u_offset_t maxoff = get_udatamodel() == DATAMODEL_ILP32 &&
- extended_offset == 0?
- MAXOFF32_T : MAXOFFSET_T;
-#else /* _SYSCALL32_IMPL */
- const u_offset_t maxoff = MAXOFF32_T;
-#endif /* _SYSCALL32_IMPL */
int in_crit = 0;
- if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX)
+ if (iovcnt <= 0 || iovcnt > IOV_MAX)
return (set_errno(EINVAL));
+ if (iovcnt > IOV_MAX_STACK) {
+ aiovlen = iovcnt * sizeof (iovec_t);
+ aiov = kmem_alloc(aiovlen, KM_SLEEP);
+ }
+
#ifdef _SYSCALL32_IMPL
/*
* 32-bit callers need to have their iovec expanded,
@@ -962,61 +1035,104 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
* of data in a single call.
*/
if (get_udatamodel() == DATAMODEL_ILP32) {
- struct iovec32 aiov32[DEF_IOV_MAX];
+ struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+ int aiov32len;
ssize32_t count32;
- if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32)))
+ aiov32len = iovcnt * sizeof (iovec32_t);
+ if (aiovlen != 0)
+ aiov32 = kmem_alloc(aiov32len, KM_SLEEP);
+
+ if (copyin(iovp, aiov32, aiov32len)) {
+ if (aiovlen != 0) {
+ kmem_free(aiov32, aiov32len);
+ kmem_free(aiov, aiovlen);
+ }
return (set_errno(EFAULT));
+ }
count32 = 0;
for (i = 0; i < iovcnt; i++) {
ssize32_t iovlen32 = aiov32[i].iov_len;
count32 += iovlen32;
- if (iovlen32 < 0 || count32 < 0)
+ if (iovlen32 < 0 || count32 < 0) {
+ if (aiovlen != 0) {
+ kmem_free(aiov32, aiov32len);
+ kmem_free(aiov, aiovlen);
+ }
return (set_errno(EINVAL));
+ }
aiov[i].iov_len = iovlen32;
aiov[i].iov_base =
(caddr_t)(uintptr_t)aiov32[i].iov_base;
}
+ if (aiovlen != 0)
+ kmem_free(aiov32, aiov32len);
} else
#endif /* _SYSCALL32_IMPL */
- if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec)))
+ if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EFAULT));
+ }
count = 0;
for (i = 0; i < iovcnt; i++) {
ssize_t iovlen = aiov[i].iov_len;
count += iovlen;
- if (iovlen < 0 || count < 0)
+ if (iovlen < 0 || count < 0) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EINVAL));
+ }
}
- if ((bcount = (ssize_t)count) < 0)
+ if ((bcount = count) < 0) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EINVAL));
- if ((fp = getf(fdes)) == NULL)
+ }
+ if ((fp = getf(fdes)) == NULL) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EBADF));
+ }
if (((fflag = fp->f_flag) & FREAD) == 0) {
error = EBADF;
goto out;
}
vp = fp->f_vnode;
rwflag = 0;
- if (vp->v_type == VREG) {
+ /*
+ * Behaviour is same as read(2). Please see comments in read(2).
+ */
+ if (vp->v_type == VREG) {
if (bcount == 0)
goto out;
- /*
- * return EINVAL for offsets that cannot be
- * represented in an off_t.
- */
- if (fileoff > maxoff) {
- error = EINVAL;
+ /* Handle offset past maximum offset allowed for file. */
+ if (fileoff >= OFFSET_MAX(fp)) {
+ struct vattr va;
+ va.va_mask = AT_SIZE;
+
+ error = VOP_GETATTR(vp, &va, 0, fp->f_cred, NULL);
+ if (error == 0) {
+ if (fileoff >= va.va_size) {
+ count = 0;
+ } else {
+ error = EOVERFLOW;
+ }
+ }
goto out;
}
- if (fileoff + bcount > maxoff)
- bcount = (ssize_t)((u_offset_t)maxoff - fileoff);
+ ASSERT(bcount == count);
+
+ /* Note: modified count used in nbl_conflict() call below. */
+ if ((fileoff + count) > OFFSET_MAX(fp))
+ count = (ssize_t)(OFFSET_MAX(fp) - fileoff);
+
} else if (vp->v_type == VFIFO) {
error = ESPIPE;
goto out;
@@ -1033,8 +1149,7 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
error = nbl_svmand(vp, fp->f_cred, &svmand);
if (error != 0)
goto out;
- if (nbl_conflict(vp, NBL_WRITE, fileoff, count, svmand,
- NULL)) {
+ if (nbl_conflict(vp, NBL_WRITE, fileoff, count, svmand, NULL)) {
error = EACCES;
goto out;
}
@@ -1042,33 +1157,6 @@ preadv(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
(void) VOP_RWLOCK(vp, rwflag, NULL);
- /*
- * Behaviour is same as read(2). Please see comments in
- * read(2).
- */
-
- if ((vp->v_type == VREG) && (fileoff >= OFFSET_MAX(fp))) {
- struct vattr va;
- va.va_mask = AT_SIZE;
- if ((error =
- VOP_GETATTR(vp, &va, 0, fp->f_cred, NULL))) {
- VOP_RWUNLOCK(vp, rwflag, NULL);
- goto out;
- }
- if (fileoff >= va.va_size) {
- VOP_RWUNLOCK(vp, rwflag, NULL);
- count = 0;
- goto out;
- } else {
- VOP_RWUNLOCK(vp, rwflag, NULL);
- error = EOVERFLOW;
- goto out;
- }
- }
- if ((vp->v_type == VREG) &&
- (fileoff + count > OFFSET_MAX(fp))) {
- count = (ssize_t)(OFFSET_MAX(fp) - fileoff);
- }
auio.uio_loffset = fileoff;
auio.uio_iov = aiov;
auio.uio_iovcnt = iovcnt;
@@ -1099,6 +1187,8 @@ out:
if (in_crit)
nbl_end_crit(vp);
releasef(fdes);
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
if (error)
return (set_errno(error));
return (count);
@@ -1109,7 +1199,8 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
off_t extended_offset)
{
struct uio auio;
- struct iovec aiov[DEF_IOV_MAX];
+ struct iovec buf[IOV_MAX_STACK], *aiov = buf;
+ int aiovlen = 0;
file_t *fp;
register vnode_t *vp;
struct cpu *cp;
@@ -1118,25 +1209,26 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
int error = 0;
int i;
+ /*
+ * See the comment in preadv for how the offset is handled.
+ */
#if defined(_SYSCALL32_IMPL) || defined(_ILP32)
u_offset_t fileoff = ((u_offset_t)extended_offset << 32) |
(u_offset_t)offset;
#else /* _SYSCALL32_IMPL || _ILP32 */
u_offset_t fileoff = (u_offset_t)(ulong_t)offset;
#endif /* _SYSCALL32_IMPR || _ILP32 */
-#ifdef _SYSCALL32_IMPL
- const u_offset_t maxoff = get_udatamodel() == DATAMODEL_ILP32 &&
- extended_offset == 0?
- MAXOFF32_T : MAXOFFSET_T;
-#else /* _SYSCALL32_IMPL */
- const u_offset_t maxoff = MAXOFF32_T;
-#endif /* _SYSCALL32_IMPL */
int in_crit = 0;
- if (iovcnt <= 0 || iovcnt > DEF_IOV_MAX)
+ if (iovcnt <= 0 || iovcnt > IOV_MAX)
return (set_errno(EINVAL));
+ if (iovcnt > IOV_MAX_STACK) {
+ aiovlen = iovcnt * sizeof (iovec_t);
+ aiov = kmem_alloc(aiovlen, KM_SLEEP);
+ }
+
#ifdef _SYSCALL32_IMPL
/*
* 32-bit callers need to have their iovec expanded,
@@ -1144,58 +1236,92 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
* of data in a single call.
*/
if (get_udatamodel() == DATAMODEL_ILP32) {
- struct iovec32 aiov32[DEF_IOV_MAX];
+ struct iovec32 buf32[IOV_MAX_STACK], *aiov32 = buf32;
+ int aiov32len;
ssize32_t count32;
- if (copyin(iovp, aiov32, iovcnt * sizeof (struct iovec32)))
+ aiov32len = iovcnt * sizeof (iovec32_t);
+ if (aiovlen != 0)
+ aiov32 = kmem_alloc(aiov32len, KM_SLEEP);
+
+ if (copyin(iovp, aiov32, aiov32len)) {
+ if (aiovlen != 0) {
+ kmem_free(aiov32, aiov32len);
+ kmem_free(aiov, aiovlen);
+ }
return (set_errno(EFAULT));
+ }
count32 = 0;
for (i = 0; i < iovcnt; i++) {
ssize32_t iovlen32 = aiov32[i].iov_len;
count32 += iovlen32;
- if (iovlen32 < 0 || count32 < 0)
+ if (iovlen32 < 0 || count32 < 0) {
+ if (aiovlen != 0) {
+ kmem_free(aiov32, aiov32len);
+ kmem_free(aiov, aiovlen);
+ }
return (set_errno(EINVAL));
+ }
aiov[i].iov_len = iovlen32;
aiov[i].iov_base =
(caddr_t)(uintptr_t)aiov32[i].iov_base;
}
+ if (aiovlen != 0)
+ kmem_free(aiov32, aiov32len);
} else
#endif /* _SYSCALL32_IMPL */
- if (copyin(iovp, aiov, iovcnt * sizeof (struct iovec)))
+ if (copyin(iovp, aiov, iovcnt * sizeof (iovec_t))) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EFAULT));
+ }
count = 0;
for (i = 0; i < iovcnt; i++) {
ssize_t iovlen = aiov[i].iov_len;
count += iovlen;
- if (iovlen < 0 || count < 0)
+ if (iovlen < 0 || count < 0) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EINVAL));
+ }
}
- if ((bcount = (ssize_t)count) < 0)
+ if ((bcount = count) < 0) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EINVAL));
- if ((fp = getf(fdes)) == NULL)
+ }
+ if ((fp = getf(fdes)) == NULL) {
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
return (set_errno(EBADF));
+ }
if (((fflag = fp->f_flag) & FWRITE) == 0) {
error = EBADF;
goto out;
}
vp = fp->f_vnode;
rwflag = 1;
- if (vp->v_type == VREG) {
+ /*
+ * The kernel's write(2) code checks the rctl & OFFSET_MAX and returns
+ * EFBIG when fileoff exceeds either limit. We do the same.
+ */
+ if (vp->v_type == VREG) {
if (bcount == 0)
goto out;
/*
- * return EINVAL for offsets that cannot be
- * represented in an off_t.
+ * Don't allow pwritev to cause file size to exceed the proper
+ * offset limit.
*/
- if (fileoff > maxoff) {
- error = EINVAL;
+ if (fileoff >= OFFSET_MAX(fp)) {
+ error = EFBIG;
goto out;
}
+
/*
* Take appropriate action if we are trying
* to write above the resource limit.
@@ -1218,17 +1344,13 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
error = EFBIG;
goto out;
}
- /*
- * Don't allow pwritev to cause file sizes to exceed
- * maxoff.
- */
- if (fileoff == maxoff) {
- error = EFBIG;
- goto out;
- }
- if (fileoff + bcount > maxoff)
- bcount = (ssize_t)((u_offset_t)maxoff - fileoff);
+ ASSERT(bcount == count);
+
+ /* Note: modified count used in nbl_conflict() call below. */
+ if ((fileoff + count) > OFFSET_MAX(fp))
+ count = (ssize_t)(OFFSET_MAX(fp) - fileoff);
+
} else if (vp->v_type == VFIFO) {
error = ESPIPE;
goto out;
@@ -1245,8 +1367,7 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
error = nbl_svmand(vp, fp->f_cred, &svmand);
if (error != 0)
goto out;
- if (nbl_conflict(vp, NBL_WRITE, fileoff, count, svmand,
- NULL)) {
+ if (nbl_conflict(vp, NBL_WRITE, fileoff, count, svmand, NULL)) {
error = EACCES;
goto out;
}
@@ -1254,34 +1375,6 @@ pwritev(int fdes, struct iovec *iovp, int iovcnt, off_t offset,
(void) VOP_RWLOCK(vp, rwflag, NULL);
-
- /*
- * Behaviour is same as write(2). Please see comments for
- * write(2).
- */
-
- if (vp->v_type == VREG) {
- if (fileoff >= curproc->p_fsz_ctl) {
- VOP_RWUNLOCK(vp, rwflag, NULL);
- mutex_enter(&curproc->p_lock);
- /* see above rctl_action comment */
- (void) rctl_action(
- rctlproc_legacy[RLIMIT_FSIZE],
- curproc->p_rctls,
- curproc, RCA_UNSAFE_SIGINFO);
- mutex_exit(&curproc->p_lock);
- error = EFBIG;
- goto out;
- }
- if (fileoff >= OFFSET_MAX(fp)) {
- VOP_RWUNLOCK(vp, rwflag, NULL);
- error = EFBIG;
- goto out;
- }
- if (fileoff + count > OFFSET_MAX(fp))
- count = (ssize_t)(OFFSET_MAX(fp) - fileoff);
- }
-
auio.uio_loffset = fileoff;
auio.uio_iov = aiov;
auio.uio_iovcnt = iovcnt;
@@ -1308,6 +1401,8 @@ out:
if (in_crit)
nbl_end_crit(vp);
releasef(fdes);
+ if (aiovlen != 0)
+ kmem_free(aiov, aiovlen);
if (error)
return (set_errno(error));
return (count);
diff --git a/usr/src/uts/common/syscall/sendfile.c b/usr/src/uts/common/syscall/sendfile.c
index 0cfafbf13f..16c6fdd27e 100644
--- a/usr/src/uts/common/syscall/sendfile.c
+++ b/usr/src/uts/common/syscall/sendfile.c
@@ -82,7 +82,7 @@ extern sotpi_info_t *sotpi_sototpi(struct sonode *);
* 64 bit kernel or 32 bit kernel. For 32 bit apps, we can't transfer
* more than 2GB of data.
*/
-int
+static int
sendvec_chunk64(file_t *fp, u_offset_t *fileoff, struct ksendfilevec64 *sfv,
int copy_cnt, ssize32_t *count)
{
@@ -343,7 +343,7 @@ sendvec_chunk64(file_t *fp, u_offset_t *fileoff, struct ksendfilevec64 *sfv,
return (0);
}
-ssize32_t
+static ssize32_t
sendvec64(file_t *fp, const struct ksendfilevec64 *vec, int sfvcnt,
size32_t *xferred, int fildes)
{
@@ -390,7 +390,7 @@ sendvec64(file_t *fp, const struct ksendfilevec64 *vec, int sfvcnt,
}
#endif
-int
+static int
sendvec_small_chunk(file_t *fp, u_offset_t *fileoff, struct sendfilevec *sfv,
int copy_cnt, ssize_t total_size, int maxblk, ssize_t *count)
{
@@ -680,7 +680,7 @@ sendvec_small_chunk(file_t *fp, u_offset_t *fileoff, struct sendfilevec *sfv,
}
-int
+static int
sendvec_chunk(file_t *fp, u_offset_t *fileoff, struct sendfilevec *sfv,
int copy_cnt, ssize_t *count)
{
@@ -1174,6 +1174,17 @@ sendfilev(int opcode, int fildes, const struct sendfilevec *vec, int sfvcnt,
} else {
maxblk = (int)vp->v_stream->sd_maxblk;
}
+
+ /*
+ * We need to make sure that the socket that we're sending on
+ * supports sendfile behavior. sockfs doesn't know that the APIs
+ * we want to use are coming from sendfile, so we can't rely on
+ * it to check for us.
+ */
+ if ((so->so_mode & SM_SENDFILESUPP) == 0) {
+ error = EOPNOTSUPP;
+ goto err;
+ }
break;
case VREG:
break;
diff --git a/usr/src/uts/common/syscall/stat.c b/usr/src/uts/common/syscall/stat.c
index 4085104cc7..93f26121bc 100644
--- a/usr/src/uts/common/syscall/stat.c
+++ b/usr/src/uts/common/syscall/stat.c
@@ -61,7 +61,7 @@
* to VOP_GETATTR
*/
-static int
+int
cstatat_getvp(int fd, char *name, int follow, vnode_t **vp, cred_t **cred)
{
vnode_t *startvp;
diff --git a/usr/src/uts/common/syscall/sysconfig.c b/usr/src/uts/common/syscall/sysconfig.c
index 03f2fabe13..e09f4e85a2 100644
--- a/usr/src/uts/common/syscall/sysconfig.c
+++ b/usr/src/uts/common/syscall/sysconfig.c
@@ -22,6 +22,7 @@
/*
* Copyright 2008 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
+ * Copyright 2017 Joyent, Inc.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -111,6 +112,9 @@ sysconfig(int which)
case _CONFIG_NPROC_MAX:
return (max_ncpus);
+ case _CONFIG_NPROC_NCPU:
+ return (NCPU); /* Private sysconfig for direct NCPU access */
+
case _CONFIG_STACK_PROT:
return (curproc->p_stkprot & ~PROT_USER);
@@ -167,44 +171,29 @@ sysconfig(int which)
/*
* If the non-global zone has a phys. memory cap, use that.
* We always report the system-wide value for the global zone,
- * even though rcapd can be used on the global zone too.
+ * even though memory capping can be used on the global zone
+ * too.
*/
- if (!INGLOBALZONE(curproc) &&
- curproc->p_zone->zone_phys_mcap != 0)
- return (MIN(btop(curproc->p_zone->zone_phys_mcap),
- physinstalled));
+ if (!INGLOBALZONE(curproc)) {
+ pgcnt_t cap, free;
+
+ zone_get_physmem_data(curzone->zone_id, &cap, &free);
+ return (MIN(cap, physinstalled));
+ }
return (physinstalled);
case _CONFIG_AVPHYS_PAGES:
/*
- * If the non-global zone has a phys. memory cap, use
- * the phys. memory cap - zone's current rss. We always
- * report the system-wide value for the global zone, even
- * though rcapd can be used on the global zone too.
+ * If the non-global zone has a phys. memory cap, use its
+ * free value. We always report the system-wide value for the
+ * global zone, even though memory capping can be used on the
+ * global zone too.
*/
- if (!INGLOBALZONE(curproc) &&
- curproc->p_zone->zone_phys_mcap != 0) {
- pgcnt_t cap, rss, free;
- vmusage_t in_use;
- size_t cnt = 1;
-
- cap = btop(curproc->p_zone->zone_phys_mcap);
- if (cap > physinstalled)
- return (freemem);
-
- if (vm_getusage(VMUSAGE_ZONE, 1, &in_use, &cnt,
- FKIOCTL) != 0)
- in_use.vmu_rss_all = 0;
- rss = btop(in_use.vmu_rss_all);
- /*
- * Because rcapd implements a soft cap, it is possible
- * for rss to be temporarily over the cap.
- */
- if (cap > rss)
- free = cap - rss;
- else
- free = 0;
+ if (!INGLOBALZONE(curproc)) {
+ pgcnt_t cap, free;
+
+ zone_get_physmem_data(curzone->zone_id, &cap, &free);
return (MIN(free, freemem));
}
diff --git a/usr/src/uts/common/syscall/uadmin.c b/usr/src/uts/common/syscall/uadmin.c
index 858305504d..dfe7f22d44 100644
--- a/usr/src/uts/common/syscall/uadmin.c
+++ b/usr/src/uts/common/syscall/uadmin.c
@@ -22,7 +22,7 @@
/*
* Copyright 2009 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
- * Copyright 2013 Joyent, Inc. All rights reserved.
+ * Copyright 2016 Joyent, Inc.
*/
#include <sys/param.h>
@@ -78,7 +78,7 @@ volatile int fastreboot_dryrun = 0;
* system with many zones.
*/
void
-killall(zoneid_t zoneid)
+killall(zoneid_t zoneid, boolean_t force)
{
proc_t *p;
@@ -108,7 +108,7 @@ killall(zoneid_t zoneid)
p->p_stat != SIDL &&
p->p_stat != SZOMB) {
mutex_enter(&p->p_lock);
- if (sigismember(&p->p_sig, SIGKILL)) {
+ if (!force && sigismember(&p->p_sig, SIGKILL)) {
mutex_exit(&p->p_lock);
p = p->p_next;
} else {
@@ -245,12 +245,13 @@ kadmin(int cmd, int fcn, void *mdep, cred_t *credp)
*/
zone_shutdown_global();
- killall(ALL_ZONES);
+ killall(ALL_ZONES, B_FALSE);
/*
* If we are calling kadmin() from a kernel context then we
* do not release these resources.
*/
if (ttoproc(curthread) != &p0) {
+ mutex_enter(&curproc->p_lock);
VN_RELE(PTOU(curproc)->u_cdir);
if (PTOU(curproc)->u_rdir)
VN_RELE(PTOU(curproc)->u_rdir);
@@ -260,6 +261,7 @@ kadmin(int cmd, int fcn, void *mdep, cred_t *credp)
PTOU(curproc)->u_cdir = rootdir;
PTOU(curproc)->u_rdir = NULL;
PTOU(curproc)->u_cwd = NULL;
+ mutex_exit(&curproc->p_lock);
}
/*
diff --git a/usr/src/uts/common/syscall/umount.c b/usr/src/uts/common/syscall/umount.c
index a2deedb163..b25f89b6d5 100644
--- a/usr/src/uts/common/syscall/umount.c
+++ b/usr/src/uts/common/syscall/umount.c
@@ -22,6 +22,7 @@
/*
* Copyright 2010 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
+ * Copyright 2016 Joyent, Inc.
*/
/* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
@@ -125,6 +126,7 @@ umount2(char *pathp, int flag)
struct pathname pn;
struct vfs *vfsp;
int error;
+ boolean_t altroot;
/*
* Some flags are disallowed through the system call interface.
@@ -154,9 +156,12 @@ umount2(char *pathp, int flag)
* isn't in an environment with an alternate root (to the zone's root)
* directory, i.e. chroot(2).
*/
- if (secpolicy_fs_unmount(CRED(), NULL) != 0 ||
- (PTOU(curproc)->u_rdir != NULL &&
- PTOU(curproc)->u_rdir != curproc->p_zone->zone_rootvp) ||
+ mutex_enter(&curproc->p_lock);
+ altroot = (PTOU(curproc)->u_rdir != NULL &&
+ PTOU(curproc)->u_rdir != curproc->p_zone->zone_rootvp);
+ mutex_exit(&curproc->p_lock);
+
+ if (secpolicy_fs_unmount(CRED(), NULL) != 0 || altroot ||
(vfsp = vfs_mntpoint2vfsp(pn.pn_path)) == NULL) {
vnode_t *fsrootvp;
generated by cgit v1.2.3 (git 2.39.1) at 2025-06-05 20:13:04 +0000