Since r219452 the alignment of __dmadat has changed, revealing that fsread()

bogusly casts its contents around causing alignment faults on sparc64 and
most likely also on at least powerpc. Fix this by copying the contents
bytewise instead as partly already done here. Solving this the right way
costs some space, i.e. 148 bytes with GCC and 16 bytes with clang on x86
there are still some bytes left there though, and an acceptable hack which
tricks the compiler into only using a 2-byte alignment instead of the native
one when accessing the contents turned out to even take up more space that.

1 files changed, 35 insertions, 31 deletions

diff --git a/sys/boot/common/ufsread.c b/sys/boot/common/ufsread.c
index 2724d7c..b767732 100644
--- a/sys/boot/common/ufsread.c
+++ b/sys/boot/common/ufsread.c
@@ -49,6 +49,7 @@ __FBSDID("$FreeBSD$");
 #include <ufs/ufs/dinode.h>
 #include <ufs/ufs/dir.h>
 #include <ufs/ffs/fs.h>
+
 #ifdef UFS_SMALL_CGBASE
 /* XXX: Revert to old (broken for over 1.5Tb filesystems) version of cgbase
    (see sys/ufs/ffs/fs.h rev 1.39) so that small boot loaders (e.g. boot2) can
@@ -90,7 +91,7 @@ static ssize_t fsread(ino_t, void *, size_t);
 static uint8_t ls, dsk_meta;
 static uint32_t fs_off;
 
-static __inline int
+static __inline uint8_t
 fsfind(const char *name, ino_t * ino)
 {
 	char buf[DEV_BSIZE];
@@ -158,7 +159,7 @@ static int sblock_try[] = SBLOCKSEARCH;
 #elif defined(UFS1_ONLY)
 #define DIP(field) dp1.field
 #else
-#define DIP(field) fs->fs_magic == FS_UFS1_MAGIC ? dp1.field : dp2.field
+#define DIP(field) fs.fs_magic == FS_UFS1_MAGIC ? dp1.field : dp2.field
 #endif
 
 static ssize_t
@@ -173,7 +174,7 @@ fsread(ino_t inode, void *buf, size_t nbyte)
 	static ino_t inomap;
 	char *blkbuf;
 	void *indbuf;
-	struct fs *fs;
+	struct fs fs;
 	char *s;
 	size_t n, nb, size, off, vboff;
 	ufs_lbn_t lbn;
@@ -181,30 +182,29 @@ fsread(ino_t inode, void *buf, size_t nbyte)
 	static ufs2_daddr_t blkmap, indmap;
 	u_int u;
 
-
 	blkbuf = dmadat->blkbuf;
 	indbuf = dmadat->indbuf;
-	fs = (struct fs *)dmadat->sbbuf;
 	if (!dsk_meta) {
 		inomap = 0;
 		for (n = 0; sblock_try[n] != -1; n++) {
-			if (dskread(fs, sblock_try[n] / DEV_BSIZE,
+			if (dskread(dmadat->sbbuf, sblock_try[n] / DEV_BSIZE,
 			    SBLOCKSIZE / DEV_BSIZE))
 				return -1;
+			memcpy(&fs, dmadat->sbbuf, sizeof(struct fs));
 			if ((
 #if defined(UFS1_ONLY)
-			     fs->fs_magic == FS_UFS1_MAGIC
+			    fs.fs_magic == FS_UFS1_MAGIC
 #elif defined(UFS2_ONLY)
-			    (fs->fs_magic == FS_UFS2_MAGIC &&
-			    fs->fs_sblockloc == sblock_try[n])
+			    (fs.fs_magic == FS_UFS2_MAGIC &&
+			    fs.fs_sblockloc == sblock_try[n])
 #else
-			     fs->fs_magic == FS_UFS1_MAGIC ||
-			    (fs->fs_magic == FS_UFS2_MAGIC &&
-			    fs->fs_sblockloc == sblock_try[n])
+			    fs.fs_magic == FS_UFS1_MAGIC ||
+			    (fs.fs_magic == FS_UFS2_MAGIC &&
+			    fs.fs_sblockloc == sblock_try[n])
 #endif
 			    ) &&
-			    fs->fs_bsize <= MAXBSIZE &&
-			    fs->fs_bsize >= sizeof(struct fs))
+			    fs.fs_bsize <= MAXBSIZE &&
+			    fs.fs_bsize >= sizeof(struct fs))
 				break;
 		}
 		if (sblock_try[n] == -1) {
@@ -212,12 +212,13 @@ fsread(ino_t inode, void *buf, size_t nbyte)
 			return -1;
 		}
 		dsk_meta++;
-	}
+	} else
+		memcpy(&fs, dmadat->sbbuf, sizeof(struct fs));
 	if (!inode)
 		return 0;
 	if (inomap != inode) {
-		n = IPERVBLK(fs);
-		if (dskread(blkbuf, INO_TO_VBA(fs, n, inode), DBPERVBLK))
+		n = IPERVBLK(&fs);
+		if (dskread(blkbuf, INO_TO_VBA(&fs, n, inode), DBPERVBLK))
 			return -1;
 		n = INO_TO_VBO(n, inode);
 #if defined(UFS1_ONLY)
@@ -227,13 +228,12 @@ fsread(ino_t inode, void *buf, size_t nbyte)
 		memcpy(&dp2, (struct ufs2_dinode *)blkbuf + n,
 		    sizeof(struct ufs2_dinode));
 #else
-		if (fs->fs_magic == FS_UFS1_MAGIC)
+		if (fs.fs_magic == FS_UFS1_MAGIC)
 			memcpy(&dp1, (struct ufs1_dinode *)blkbuf + n,
 			    sizeof(struct ufs1_dinode));
 		else
 			memcpy(&dp2, (struct ufs2_dinode *)blkbuf + n,
 			    sizeof(struct ufs2_dinode));
-
 #endif
 		inomap = inode;
 		fs_off = 0;
@@ -246,15 +246,15 @@ fsread(ino_t inode, void *buf, size_t nbyte)
 		nbyte = n;
 	nb = nbyte;
 	while (nb) {
-		lbn = lblkno(fs, fs_off);
-		off = blkoff(fs, fs_off);
+		lbn = lblkno(&fs, fs_off);
+		off = blkoff(&fs, fs_off);
 		if (lbn < NDADDR) {
 			addr = DIP(di_db[lbn]);
-		} else if (lbn < NDADDR + NINDIR(fs)) {
-			n = INDIRPERVBLK(fs);
+		} else if (lbn < NDADDR + NINDIR(&fs)) {
+			n = INDIRPERVBLK(&fs);
 			addr = DIP(di_ib[0]);
 			u = (u_int)(lbn - NDADDR) / n * DBPERVBLK;
-			vbaddr = fsbtodb(fs, addr) + u;
+			vbaddr = fsbtodb(&fs, addr) + u;
 			if (indmap != vbaddr) {
 				if (dskread(indbuf, vbaddr, DBPERVBLK))
 					return -1;
@@ -262,21 +262,25 @@ fsread(ino_t inode, void *buf, size_t nbyte)
 			}
 			n = (lbn - NDADDR) & (n - 1);
 #if defined(UFS1_ONLY)
-			addr = ((ufs1_daddr_t *)indbuf)[n];
+			memcpy(&addr, (ufs1_daddr_t *)indbuf + n,
+			    sizeof(ufs1_daddr_t));
 #elif defined(UFS2_ONLY)
-			addr = ((ufs2_daddr_t *)indbuf)[n];
+			memcpy(&addr, (ufs2_daddr_t *)indbuf + n,
+			    sizeof(ufs2_daddr_t));
 #else
-			if (fs->fs_magic == FS_UFS1_MAGIC)
-				addr = ((ufs1_daddr_t *)indbuf)[n];
+			if (fs.fs_magic == FS_UFS1_MAGIC)
+				memcpy(&addr, (ufs1_daddr_t *)indbuf + n,
+				    sizeof(ufs1_daddr_t));
 			else
-				addr = ((ufs2_daddr_t *)indbuf)[n];
+				memcpy(&addr, (ufs2_daddr_t *)indbuf + n,
+				    sizeof(ufs2_daddr_t));
 #endif
 		} else {
 			return -1;
 		}
-		vbaddr = fsbtodb(fs, addr) + (off >> VBLKSHIFT) * DBPERVBLK;
+		vbaddr = fsbtodb(&fs, addr) + (off >> VBLKSHIFT) * DBPERVBLK;
 		vboff = off & VBLKMASK;
-		n = sblksize(fs, size, lbn) - (off & ~VBLKMASK);
+		n = sblksize(&fs, size, lbn) - (off & ~VBLKMASK);
 		if (n > VBLKSIZE)
 			n = VBLKSIZE;
 		if (blkmap != vbaddr) {