OS-4469 journald needs fallocate(2) syscall

6 files changed, 236 insertions, 5 deletions

diff --git a/usr/src/lib/brand/lx/testing/ltp_skiplist b/usr/src/lib/brand/lx/testing/ltp_skiplist
index fdd2eb6206..7568a28682 100644
--- a/usr/src/lib/brand/lx/testing/ltp_skiplist
+++ b/usr/src/lib/brand/lx/testing/ltp_skiplist
@@ -127,9 +127,7 @@ add_key02
 clone02
 clone08
 crash02
-fallocate01
-fallocate01
-fallocate01
+fallocate04		# needs SEEK_HOLE
 fanotify01
 fanotify02
 fanotify03
diff --git a/usr/src/uts/common/brand/lx/os/lx_syscall.c b/usr/src/uts/common/brand/lx/os/lx_syscall.c
index 6b8f60efda..993af09f12 100644
--- a/usr/src/uts/common/brand/lx/os/lx_syscall.c
+++ b/usr/src/uts/common/brand/lx/os/lx_syscall.c
@@ -851,7 +851,7 @@ lx_sysent_t lx_sysent32[] = {
 	{"signalfd",	NULL,			0,		3}, /* 321 */
 	{"timerfd_create", NULL,		0,		2}, /* 322 */
 	{"eventfd",	NULL,			0,		1}, /* 323 */
-	{"fallocate",	NULL,			NOSYS_NULL,	0}, /* 324 */
+	{"fallocate",	lx_fallocate,		0,		4}, /* 324 */
 	{"timerfd_settime", NULL,		0,		4}, /* 325 */
 	{"timerfd_gettime", NULL,		0,		2}, /* 326 */
 	{"signalfd4",	NULL,			0,		4}, /* 327 */
@@ -1179,7 +1179,7 @@ lx_sysent_t lx_sysent64[] = {
 	{"signalfd",	NULL,			0,		3}, /* 282 */
 	{"timerfd_create", NULL,		0,		2}, /* 283 */
 	{"eventfd",	NULL,			0,		1}, /* 284 */
-	{"fallocate",	NULL,			NOSYS_NULL,	0}, /* 285 */
+	{"fallocate",	lx_fallocate,		0,		4}, /* 285 */
 	{"timerfd_settime", NULL,		0,		4}, /* 286 */
 	{"timerfd_gettime", NULL,		0,		2}, /* 287 */
 	{"accept4",	NULL,			0,		4}, /* 288 */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_syscalls.h b/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
index 989991e6c0..41587591fd 100644
--- a/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
+++ b/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
@@ -43,6 +43,7 @@ extern long lx_clock_getres();
 extern long lx_clock_gettime();
 extern long lx_clock_settime();
 extern long lx_connect();
+extern long lx_fallocate();
 extern long lx_fchmod();
 extern long lx_fchmodat();
 extern long lx_fchown();
diff --git a/usr/src/uts/common/brand/lx/sys/lx_types.h b/usr/src/uts/common/brand/lx/sys/lx_types.h
index 922c412020..e91484c16b 100644
--- a/usr/src/uts/common/brand/lx/sys/lx_types.h
+++ b/usr/src/uts/common/brand/lx/sys/lx_types.h
@@ -40,6 +40,10 @@ extern "C" {
 #define	INT_MAX		2147483647	/* max value of an "int" */
 #define	UINT_MAX	4294967295U	/* max value of an "unsigned int" */
 
+#ifndef LLONG_MAX
+#define	LLONG_MAX	9223372036854775807LL
+#endif
+
 #if defined(_LP64)
 #define	LONG_MAX	9223372036854775807L
 #define	ULONG_MAX	18446744073709551615UL
diff --git a/usr/src/uts/common/brand/lx/syscall/lx_fallocate.c b/usr/src/uts/common/brand/lx/syscall/lx_fallocate.c
new file mode 100644
index 0000000000..2c214bfb87
--- /dev/null
+++ b/usr/src/uts/common/brand/lx/syscall/lx_fallocate.c
@@ -0,0 +1,227 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2015 Joyent, Inc.
+ */
+
+#include <sys/systm.h>
+#include <sys/zone.h>
+#include <sys/types.h>
+#include <sys/filio.h>
+#include <sys/fcntl.h>
+#include <sys/stat.h>
+#include <sys/nbmlock.h>
+#include <sys/lx_impl.h>
+#include <sys/lx_brand.h>
+#include <sys/sdt.h>
+
+extern int flock_check(vnode_t *, flock64_t *, offset_t, offset_t);
+
+#define	LX_FALLOC_FL_KEEP_SIZE		0x01
+#define	LX_FALLOC_FL_PUNCH_HOLE		0x02
+#define	LX_FALLOC_FL_NO_HIDE_STALE	0x04
+#define	LX_FALLOC_FL_COLLAPSE_RANGE	0x08
+#define	LX_FALLOC_FL_ZERO_RANGE		0x10
+
+#define	LX_FALLOC_VALID	(LX_FALLOC_FL_KEEP_SIZE | LX_FALLOC_FL_PUNCH_HOLE | \
+	LX_FALLOC_FL_NO_HIDE_STALE | LX_FALLOC_FL_COLLAPSE_RANGE | \
+	LX_FALLOC_FL_ZERO_RANGE)
+
+long
+lx_fallocate(int fd, int mode, off_t offset, off_t len)
+{
+	int error = 0;
+	file_t *fp;
+	vnode_t *vp;
+	int64_t tot;
+	struct flock64 bf;
+	vattr_t vattr;
+	u_offset_t f_offset;
+	boolean_t in_crit = B_FALSE;
+
+	/*
+	 * Error checking is in a specific order to make LTP happy.
+	 */
+
+	tot = offset + len;
+	if (tot > (LLONG_MAX / (int64_t)1024))
+		return (set_errno(EFBIG));
+
+	if (mode & LX_FALLOC_FL_COLLAPSE_RANGE)
+		return (set_errno(EOPNOTSUPP));
+
+	if ((fp = getf(fd)) == NULL)
+		return (set_errno(EBADF));
+
+	if ((fp->f_flag & FWRITE) == 0) {
+		error = EBADF;
+		goto done;
+	}
+
+	vp = fp->f_vnode;
+	if (vp->v_type != VREG) {
+		error = EINVAL;
+		goto done;
+	}
+
+	if (offset < 0 || len <= 0) {
+		error = EINVAL;
+		goto done;
+	}
+
+	if (tot < 0LL) {
+		error = EFBIG;
+		goto done;
+	}
+
+	if ((mode & ~LX_FALLOC_VALID) != 0) {
+		error = EINVAL;
+		goto done;
+	}
+
+	/*
+	 * If this is the only flag then we don't actually do any work.
+	 */
+	if (mode == LX_FALLOC_FL_KEEP_SIZE)
+		goto done;
+
+	bzero(&bf, sizeof (bf));
+
+	vattr.va_mask = AT_SIZE;
+	if ((error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) != 0)
+		goto done;
+
+	if (mode == 0) {
+		/* Nothing to do if not extending the file */
+		if (vattr.va_size >= tot)
+			goto done;
+
+		/* Extend the file. */
+		bf.l_start = (off64_t)tot;
+		bf.l_len = (off64_t)0;
+
+	} else if (mode & LX_FALLOC_FL_PUNCH_HOLE) {
+		/*
+		 * Deallocate space in the file.
+		 */
+		if ((mode & LX_FALLOC_FL_KEEP_SIZE) == 0) {
+			/* this flag is required with punch hole */
+			error = EINVAL;
+			goto done;
+		}
+
+		if (mode &
+		    ~(LX_FALLOC_FL_PUNCH_HOLE | LX_FALLOC_FL_KEEP_SIZE)) {
+			error = EINVAL;
+			goto done;
+		}
+
+		/* Make sure we don't extend since keep_size is set. */
+		if (vattr.va_size < tot) {
+			if (offset > vattr.va_size)
+				goto done;
+			len = (off_t)vattr.va_size - offset;
+		}
+
+		bf.l_start = (off64_t)offset;
+		bf.l_len = (off64_t)len;
+
+	} else if (mode & LX_FALLOC_FL_ZERO_RANGE) {
+		/*
+		 * Zero out the space in the file.
+		 */
+		if (mode &
+		    ~(LX_FALLOC_FL_ZERO_RANGE | LX_FALLOC_FL_KEEP_SIZE)) {
+			error = EINVAL;
+			goto done;
+		}
+
+		/* Make sure we don't extend when keep_size is set. */
+		if (mode & LX_FALLOC_FL_KEEP_SIZE && vattr.va_size < tot) {
+			if (offset > vattr.va_size)
+				goto done;
+			len = vattr.va_size - offset;
+		}
+
+		bf.l_start = (off64_t)offset;
+		bf.l_len = (off64_t)len;
+	} else {
+		/* We should have already handled all flags */
+		VERIFY(0);
+	}
+
+	/*
+	 * Check for locks in the range.
+	 */
+	f_offset = fp->f_offset;
+	error = flock_check(vp, &bf, f_offset, MAXOFF_T);
+	if (error != 0)
+		goto done;
+
+	/*
+	 * Check for conflicting non-blocking mandatory locks.
+	 * We need to get the size again under nbl_start_crit.
+	 */
+	if (nbl_need_check(vp)) {
+		u_offset_t	begin;
+		ssize_t		length;
+
+		nbl_start_crit(vp, RW_READER);
+		in_crit = B_TRUE;
+		vattr.va_mask = AT_SIZE;
+		if ((error = VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)) != 0)
+			goto done;
+
+		/*
+		 * Make sure we don't extend when keep_size is set.
+		 */
+		if (mode & LX_FALLOC_FL_KEEP_SIZE && vattr.va_size < tot) {
+			ASSERT(mode & (LX_FALLOC_FL_PUNCH_HOLE |
+			    LX_FALLOC_FL_ZERO_RANGE));
+
+			/*
+			 * If the size grew we can short-circuit the rest of
+			 * the work, otherwise adjust bf for the vop_space
+			 * call.
+			 */
+			if (offset >= vattr.va_size)
+				goto done;
+			len = vattr.va_size - offset;
+			bf.l_len = (off64_t)len;
+		}
+
+		if (offset > vattr.va_size) {
+			begin = vattr.va_size;
+			length = offset - vattr.va_size;
+		} else {
+			begin = offset;
+			length = vattr.va_size - offset;
+		}
+
+		if (nbl_conflict(vp, NBL_WRITE, begin, length, 0, NULL)) {
+			error = EACCES;
+			goto done;
+		}
+	}
+
+	error = VOP_SPACE(vp, F_FREESP, &bf, 0, f_offset, fp->f_cred, NULL);
+
+done:
+	if (in_crit)
+		nbl_end_crit(vp);
+
+	releasef(fd);
+	if (error != 0)
+		return (set_errno(error));
+
+	return (0);
+}
diff --git a/usr/src/uts/intel/Makefile.files b/usr/src/uts/intel/Makefile.files
index 4e69d03511..7e30c5843f 100644
--- a/usr/src/uts/intel/Makefile.files
+++ b/usr/src/uts/intel/Makefile.files
@@ -294,6 +294,7 @@ LX_BRAND_OBJS  =		\
 	lx_clone.o		\
 	lx_cpu.o		\
 	lx_errno.o		\
+	lx_fallocate.o		\
 	lx_fcntl.o		\
 	lx_futex.o		\
 	lx_getdents.o		\