Print this page
5880 Increase IOV_MAX to at least 1024
Portions contributed by: Jerry Jelinek <jerry.jelinek@joyent.com>
@@ -20,11 +20,11 @@
*/
/*
* Copyright 2008 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
- * Copyright (c) 2015, Joyent, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc. All rights reserved.
*/
/* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
/* All Rights Reserved */
@@ -48,10 +48,11 @@
#include <sys/cpuvar.h>
#include <sys/uio.h>
#include <sys/debug.h>
#include <sys/rctl.h>
#include <sys/nbmlock.h>
+#include <sys/limits.h>
#define COPYOUT_MAX_CACHE (1<<17) /* 128K */
size_t copyout_max_cached = COPYOUT_MAX_CACHE; /* global so it's patchable */
@@ -605,23 +606,16 @@
if (error)
return (set_errno(error));
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;
int fflag, ioflag, rwflag;
ssize_t count, bcount;
@@ -628,50 +622,82 @@
int error = 0;
int i;
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,
* while ensuring that they can't move more than 2Gbytes
* 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;
}
vp = fp->f_vnode;
@@ -766,20 +792,23 @@
error = 0;
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);
}
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;
int fflag, ioflag, rwflag;
ssize_t count, bcount;
@@ -786,50 +815,81 @@
int error = 0;
int i;
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,
* while ensuring that they can't move more than 2Gbytes
* 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;
}
vp = fp->f_vnode;
@@ -915,10 +975,12 @@
error = 0;
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);
}
@@ -925,11 +987,12 @@
ssize_t
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;
int fflag, ioflag, rwflag;
ssize_t count, bcount;
@@ -950,53 +1013,87 @@
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,
* while ensuring that they can't move more than 2Gbytes
* 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 = (ssize_t)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;
@@ -1097,10 +1194,12 @@
error = 0;
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);
}
@@ -1107,11 +1206,12 @@
ssize_t
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;
int fflag, ioflag, rwflag;
ssize_t count, bcount;
@@ -1132,53 +1232,87 @@
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,
* while ensuring that they can't move more than 2Gbytes
* 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 = (ssize_t)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;
@@ -1306,10 +1440,12 @@
error = 0;
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);
}