Reduce memory footprint of fsck_msdosfs.

This utility was initially written for FAT12/16, which were inherently
small. When FAT32 support was added, the old data structure and
algorithms remain used with minimal changes.

With growing size of FAT32 media, the current data structure that
requires 4 32-bit variables per each FAT32 table entry would consume up
to 4 GiB of RAM, which can be too big for systems with limited RAM
available.

Address this by taking a different approach of validating the FAT.

The FAT is essentially a set of linked lists of chains that was
referenced by directory entries, and the checker needs to make sure that
the linked chains of clusters do not have cross-linked chains, and every
chain were referenced by one and only one directory entry.  Instead of
keeping track of the chain's 'head' cluster number, the size of the
chain, the used status of the chain and the "next" pointer which is
content of the FAT table, we create accessors for the FAT table data
for the "next" pointer, and keep only one bit to indicate if the
current cluster is a 'head' node of a cluster chain, in a bitmap.

We further overhaul the FAT checker to find out the possible head nodes
by excluding ones that are not (in other words, nodes that have some
other nodes claiming them as the next node) instead of marking the head
nodes for each node on the chain.  This approach greatly reduced the
complexiety of computation from O(N^2) worst case, to an O(N) scan for
worst case.  The file (cluster chain) length is not useful for the FAT
checker, so don't bother to calculate them in the FAT checker and
instead leave the task to the directory structure check, at which point
we would have non-crossed cluster chains, and we are guaranteed that
each cluster will be visited for at most one time.

When checking the directory structures, we use the head node indicator
to as the visited (used) flag: every cluster chain can only be
referenced by one directory entry, so we clear them when calculating
the length of the chain, and we can immediately tell if there are
anomalies in the directory entry.

As a result, the required RAM size is now 1 bit per each entry of
the FAT table, plus memory needed to hold the FAT table in memory,
instead of 16 bytes (=128 bits) per each entry.  For FAT12 and FAT16,
we will load the whole FAT table into memory as they are smaller than
128KiB, and for FAT32, we first attempt to mmap() it into memory, and
when that fails, we would fall back to a simple LRU cache of 4 MiB of
RAM.

sbin/fsck_msdosfs/boot.c:

 - Added additional sanity checks for valid FAT32/FAT16/FAT12 cluster
   number.
 - FAT32: check if root directory starts with a valid cluster number,
   moved from dir.c.  There is no point to proceed if the filesystem
   is already damaged beyond repair.

sbin/fsck_msdosfs/check.c:

 - Combine phase 1 and phase 2, now that the readfat() is able to
   detect cross chains.

sbin/fsck_msdosfs/dir.c:

 - Refactor code to use FAT accessor instead of accessing the internal
   representation of FAT table.
 - Make use of the cluster chain head bitmap.
 - Clarify and simplify directory entry check, remove unnecessary
   checks that are would be done at a later time (for example, whether
   the directory's second cluster is a valid one, which is examined
   more throughly in a later checkchain() and does not prevent us
   from proceeding further).

sbin/fsck_msdosfs/dosfs.h:

 - Remove internal representation of FAT table, which is replaced by
   the head bitmap that is opaque to other code.
 - Added a special CLUST_DEAD cluster type to indicate errors.

sbin/fsck_msdosfs/ext.h:

 - Added a flag that overrides mmap(2) setting.  The corresponding
   command line option, -M is intentionally undocumented as we do not
   expect users to need it.
 - Added accessors for FAT table and convert existing interface to use
   it.

sbin/fsck_msdosfs/fat.c:

 - Added head bitmap to represent whether a cluster is a head cluster.
 - Converted FAT internal representation to accessors.
 - Implemented a LRU cache for FAT32 when mmap(2) should not or can not
   be used.
 - _readfat: Attempt a mmap(2) and fall back to regular read for
   non-FAT32 file systems; use the LRU cache for FAT32 and prepopulate
   the cache with the first 4MiB of the entries.
 - readfat: Added support of head bitmap and use the population scan to
   detect bogus chains.
 - clusterdiff: removed, FATs are copied from the checked copy via
   writefat()/copyfat().
 - checkchain: calculates the length of a cluster chain and make sure
   that it ends with a valid EOF marker.
 - clearchain: follow and clear a chain and maintain the free cluster
   count.
 - checklost: convert to use head bitmap. At the end of all other scans,
   the remaining 'head' nodes are leaders of lost cluster chains.

sbin/fsck_msdosfs/fat.c:

 - Added a new -M option which is intentionally undocumented, to disable
   the use of mmap().

Reviewed by:	kevlo
MFC after:	1 month
Relnotes:	yes
Differential Revision:	https://reviews.freebsd.org/D22965

1 files changed, 201 insertions, 161 deletions

diff --git a/sbin/fsck_msdosfs/dir.c b/sbin/fsck_msdosfs/dir.c
index 7e459c75d4ff..6bdc3b4d0f94 100644
--- a/sbin/fsck_msdosfs/dir.c
+++ b/sbin/fsck_msdosfs/dir.c
@@ -1,6 +1,7 @@
 /*-
  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
  *
+ * Copyright (c) 2019 Google LLC
  * Copyright (C) 1995, 1996, 1997 Wolfgang Solfrank
  * Copyright (c) 1995 Martin Husemann
  * Some structure declaration borrowed from Paul Popelka
@@ -95,14 +96,11 @@ static struct dirTodoNode *newDirTodo(void);
 static void freeDirTodo(struct dirTodoNode *);
 static char *fullpath(struct dosDirEntry *);
 static u_char calcShortSum(u_char *);
-static int delete(int, struct bootblock *, struct fatEntry *, cl_t, int,
-    cl_t, int, int);
-static int removede(int, struct bootblock *, struct fatEntry *, u_char *,
-    u_char *, cl_t, cl_t, cl_t, char *, int);
-static int checksize(struct bootblock *, struct fatEntry *, u_char *,
-    struct dosDirEntry *);
-static int readDosDirSection(int, struct bootblock *, struct fatEntry *,
-    struct dosDirEntry *);
+static int delete(struct fat_descriptor *, cl_t, int, cl_t, int, int);
+static int removede(struct fat_descriptor *, u_char *, u_char *,
+    cl_t, cl_t, cl_t, char *, int);
+static int checksize(struct fat_descriptor *, u_char *, struct dosDirEntry *);
+static int readDosDirSection(struct fat_descriptor *, struct dosDirEntry *);
 
 /*
  * Manage free dosDirEntry structures.
@@ -116,7 +114,7 @@ newDosDirEntry(void)
 
 	if (!(de = freede)) {
 		if (!(de = malloc(sizeof *de)))
-			return 0;
+			return (NULL);
 	} else
 		freede = de->next;
 	return de;
@@ -193,7 +191,7 @@ fullpath(struct dosDirEntry *dir)
 /*
  * Calculate a checksum over an 8.3 alias name
  */
-static u_char
+static inline u_char
 calcShortSum(u_char *p)
 {
 	u_char sum = 0;
@@ -221,21 +219,24 @@ static struct dosDirEntry *lostDir;
  * Init internal state for a new directory scan.
  */
 int
-resetDosDirSection(struct bootblock *boot, struct fatEntry *fat)
+resetDosDirSection(struct fat_descriptor *fat)
 {
-	int b1, b2;
+	int rootdir_size, cluster_size;
 	int ret = FSOK;
 	size_t len;
+	struct bootblock *boot;
+
+	boot = fat_get_boot(fat);
 
-	b1 = boot->bpbRootDirEnts * 32;
-	b2 = boot->bpbSecPerClust * boot->bpbBytesPerSec;
+	rootdir_size = boot->bpbRootDirEnts * 32;
+	cluster_size = boot->bpbSecPerClust * boot->bpbBytesPerSec;
 
-	if ((buffer = malloc(len = MAX(b1, b2))) == NULL) {
+	if ((buffer = malloc(len = MAX(rootdir_size, cluster_size))) == NULL) {
 		perr("No space for directory buffer (%zu)", len);
 		return FSFATAL;
 	}
 
-	if ((delbuf = malloc(len = b2)) == NULL) {
+	if ((delbuf = malloc(len = cluster_size)) == NULL) {
 		free(buffer);
 		perr("No space for directory delbuf (%zu)", len);
 		return FSFATAL;
@@ -250,18 +251,10 @@ resetDosDirSection(struct bootblock *boot, struct fatEntry *fat)
 
 	memset(rootDir, 0, sizeof *rootDir);
 	if (boot->flags & FAT32) {
-		if (boot->bpbRootClust < CLUST_FIRST ||
-		    boot->bpbRootClust >= boot->NumClusters) {
-			pfatal("Root directory starts with cluster out of range(%u)",
-			       boot->bpbRootClust);
-			return FSFATAL;
-		}
-		if (fat[boot->bpbRootClust].head != boot->bpbRootClust) {
+		if (!fat_is_cl_head(fat, boot->bpbRootClust)) {
 			pfatal("Root directory doesn't start a cluster chain");
 			return FSFATAL;
 		}
-
-		fat[boot->bpbRootClust].flags |= FAT_USED;
 		rootDir->head = boot->bpbRootClust;
 	}
 
@@ -302,16 +295,21 @@ finishDosDirSection(void)
  * Delete directory entries between startcl, startoff and endcl, endoff.
  */
 static int
-delete(int f, struct bootblock *boot, struct fatEntry *fat, cl_t startcl,
+delete(struct fat_descriptor *fat, cl_t startcl,
     int startoff, cl_t endcl, int endoff, int notlast)
 {
 	u_char *s, *e;
 	off_t off;
-	int clsz = boot->bpbSecPerClust * boot->bpbBytesPerSec;
+	int clsz, fd;
+	struct bootblock *boot;
+
+	boot = fat_get_boot(fat);
+	fd = fat_get_fd(fat);
+	clsz = boot->bpbSecPerClust * boot->bpbBytesPerSec;
 
 	s = delbuf + startoff;
 	e = delbuf + clsz;
-	while (startcl >= CLUST_FIRST && startcl < boot->NumClusters) {
+	while (fat_is_valid_cl(fat, startcl)) {
 		if (startcl == endcl) {
 			if (notlast)
 				break;
@@ -320,11 +318,11 @@ delete(int f, struct bootblock *boot, struct fatEntry *fat, cl_t startcl,
 		off = (startcl - CLUST_FIRST) * boot->bpbSecPerClust + boot->FirstCluster;
 
 		off *= boot->bpbBytesPerSec;
-		if (lseek(f, off, SEEK_SET) != off) {
+		if (lseek(fd, off, SEEK_SET) != off) {
 			perr("Unable to lseek to %" PRId64, off);
 			return FSFATAL;
 		}
-		if (read(f, delbuf, clsz) != clsz) {
+		if (read(fd, delbuf, clsz) != clsz) {
 			perr("Unable to read directory");
 			return FSFATAL;
 		}
@@ -332,25 +330,26 @@ delete(int f, struct bootblock *boot, struct fatEntry *fat, cl_t startcl,
 			*s = SLOT_DELETED;
 			s += 32;
 		}
-		if (lseek(f, off, SEEK_SET) != off) {
+		if (lseek(fd, off, SEEK_SET) != off) {
 			perr("Unable to lseek to %" PRId64, off);
 			return FSFATAL;
 		}
-		if (write(f, delbuf, clsz) != clsz) {
+		if (write(fd, delbuf, clsz) != clsz) {
 			perr("Unable to write directory");
 			return FSFATAL;
 		}
 		if (startcl == endcl)
 			break;
-		startcl = fat[startcl].next;
+		startcl = fat_get_cl_next(fat, startcl);
 		s = delbuf;
 	}
 	return FSOK;
 }
 
 static int
-removede(int f, struct bootblock *boot, struct fatEntry *fat, u_char *start,
-    u_char *end, cl_t startcl, cl_t endcl, cl_t curcl, char *path, int type)
+removede(struct fat_descriptor *fat, u_char *start,
+    u_char *end, cl_t startcl, cl_t endcl, cl_t curcl,
+    char *path, int type)
 {
 	switch (type) {
 	case 0:
@@ -366,14 +365,14 @@ removede(int f, struct bootblock *boot, struct fatEntry *fat, u_char *start,
 	}
 	if (ask(0, "Remove")) {
 		if (startcl != curcl) {
-			if (delete(f, boot, fat,
+			if (delete(fat,
 				   startcl, start - buffer,
 				   endcl, end - buffer,
 				   endcl == curcl) == FSFATAL)
 				return FSFATAL;
 			start = buffer;
 		}
-		/* startcl is < CLUST_FIRST for !fat32 root */
+		/* startcl is < CLUST_FIRST for !FAT32 root */
 		if ((endcl == curcl) || (startcl < CLUST_FIRST))
 			for (; start < end; start += 32)
 				*start = SLOT_DELETED;
@@ -386,23 +385,37 @@ removede(int f, struct bootblock *boot, struct fatEntry *fat, u_char *start,
  * Check an in-memory file entry
  */
 static int
-checksize(struct bootblock *boot, struct fatEntry *fat, u_char *p,
-    struct dosDirEntry *dir)
+checksize(struct fat_descriptor *fat, u_char *p, struct dosDirEntry *dir)
 {
+	int ret = FSOK;
+	size_t physicalSize;
+	struct bootblock *boot;
+
+	boot = fat_get_boot(fat);
+
 	/*
 	 * Check size on ordinary files
 	 */
-	u_int32_t physicalSize;
-
-	if (dir->head == CLUST_FREE)
+	if (dir->head == CLUST_FREE) {
 		physicalSize = 0;
-	else {
-		if (dir->head < CLUST_FIRST || dir->head >= boot->NumClusters)
+	} else {
+		if (!fat_is_valid_cl(fat, dir->head))
 			return FSERROR;
-		physicalSize = fat[dir->head].length * boot->ClusterSize;
+		ret = checkchain(fat, dir->head, &physicalSize);
+		/*
+		 * Upon return, physicalSize would hold the chain length
+		 * that checkchain() was able to validate, but if the user
+		 * refused the proposed repair, it would be unsafe to
+		 * proceed with directory entry fix, so bail out in that
+		 * case.
+		 */
+		if (ret == FSERROR) {
+			return (FSERROR);
+		}
+		physicalSize *= boot->ClusterSize;
 	}
 	if (physicalSize < dir->size) {
-		pwarn("size of %s is %u, should at most be %u\n",
+		pwarn("size of %s is %u, should at most be %zu\n",
 		      fullpath(dir), dir->size, physicalSize);
 		if (ask(1, "Truncate")) {
 			dir->size = physicalSize;
@@ -422,11 +435,10 @@ checksize(struct bootblock *boot, struct fatEntry *fat, u_char *p,
 
 			for (cl = dir->head, len = sz = 0;
 			    (sz += boot->ClusterSize) < dir->size; len++)
-				cl = fat[cl].next;
-			clearchain(boot, fat, fat[cl].next);
-			fat[cl].next = CLUST_EOF;
-			fat[dir->head].length = len;
-			return FSFATMOD;
+				cl = fat_get_cl_next(fat, cl);
+			clearchain(fat, fat_get_cl_next(fat, cl));
+			ret = fat_set_cl_next(fat, cl, CLUST_EOF);
+			return (FSFATMOD | ret);
 		} else
 			return FSERROR;
 	}
@@ -442,15 +454,20 @@ static const u_char dotdot_name[11] = "..         ";
  * when we traverse into it.
  */
 static int
-check_subdirectory(int f, struct bootblock *boot, struct dosDirEntry *dir)
+check_subdirectory(struct fat_descriptor *fat, struct dosDirEntry *dir)
 {
 	u_char *buf, *cp;
 	off_t off;
 	cl_t cl;
 	int retval = FSOK;
+	int fd;
+	struct bootblock *boot;
+
+	boot = fat_get_boot(fat);
+	fd = fat_get_fd(fat);
 
 	cl = dir->head;
-	if (dir->parent && (cl < CLUST_FIRST || cl >= boot->NumClusters)) {
+	if (dir->parent && !fat_is_valid_cl(fat, cl)) {
 		return FSERROR;
 	}
 
@@ -474,8 +491,8 @@ check_subdirectory(int f, struct bootblock *boot, struct dosDirEntry *dir)
 	}
 
 	off *= boot->bpbBytesPerSec;
-	if (lseek(f, off, SEEK_SET) != off ||
-	    read(f, buf, boot->bpbBytesPerSec) != (ssize_t)boot->bpbBytesPerSec) {
+	if (lseek(fd, off, SEEK_SET) != off ||
+	    read(fd, buf, boot->bpbBytesPerSec) != (ssize_t)boot->bpbBytesPerSec) {
 		perr("Unable to read directory");
 		free(buf);
 		return FSFATAL;
@@ -509,22 +526,27 @@ check_subdirectory(int f, struct bootblock *boot, struct dosDirEntry *dir)
  *   - push directories onto the todo-stack
  */
 static int
-readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
-    struct dosDirEntry *dir)
+readDosDirSection(struct fat_descriptor *fat, struct dosDirEntry *dir)
 {
+	struct bootblock *boot;
 	struct dosDirEntry dirent, *d;
 	u_char *p, *vallfn, *invlfn, *empty;
 	off_t off;
-	int i, j, k, last;
+	int fd, i, j, k, iosize, entries;
+	bool is_legacyroot;
 	cl_t cl, valcl = ~0, invcl = ~0, empcl = ~0;
 	char *t;
 	u_int lidx = 0;
 	int shortSum;
 	int mod = FSOK;
+	size_t dirclusters;
 #define	THISMOD	0x8000			/* Only used within this routine */
 
+	boot = fat_get_boot(fat);
+	fd = fat_get_fd(fat);
+
 	cl = dir->head;
-	if (dir->parent && (cl < CLUST_FIRST || cl >= boot->NumClusters)) {
+	if (dir->parent && (!fat_is_valid_cl(fat, cl))) {
 		/*
 		 * Already handled somewhere else.
 		 */
@@ -532,24 +554,50 @@ readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
 	}
 	shortSum = -1;
 	vallfn = invlfn = empty = NULL;
+
+	/*
+	 * If we are checking the legacy root (for FAT12/FAT16),
+	 * we will operate on the whole directory; otherwise, we
+	 * will operate on one cluster at a time, and also take
+	 * this opportunity to examine the chain.
+	 *
+	 * Derive how many entries we are going to encounter from
+	 * the I/O size.
+	 */
+	is_legacyroot = (dir->parent == NULL && !(boot->flags & FAT32));
+	if (is_legacyroot) {
+		iosize = boot->bpbRootDirEnts * 32;
+		entries = boot->bpbRootDirEnts;
+	} else {
+		iosize = boot->bpbSecPerClust * boot->bpbBytesPerSec;
+		entries = iosize / 32;
+		mod |= checkchain(fat, dir->head, &dirclusters);
+	}
+
 	do {
-		if (!(boot->flags & FAT32) && !dir->parent) {
-			last = boot->bpbRootDirEnts * 32;
+		if (is_legacyroot) {
+			/*
+			 * Special case for FAT12/FAT16 root -- read
+			 * in the whole root directory.
+			 */
 			off = boot->bpbResSectors + boot->bpbFATs *
 			    boot->FATsecs;
 		} else {
-			last = boot->bpbSecPerClust * boot->bpbBytesPerSec;
+			/*
+			 * Otherwise, read in a cluster of the
+			 * directory.
+			 */
 			off = (cl - CLUST_FIRST) * boot->bpbSecPerClust + boot->FirstCluster;
 		}
 
 		off *= boot->bpbBytesPerSec;
-		if (lseek(f, off, SEEK_SET) != off
-		    || read(f, buffer, last) != last) {
+		if (lseek(fd, off, SEEK_SET) != off ||
+		    read(fd, buffer, iosize) != iosize) {
 			perr("Unable to read directory");
 			return FSFATAL;
 		}
-		last /= 32;
-		for (p = buffer, i = 0; i < last; i++, p += 32) {
+
+		for (p = buffer, i = 0; i < entries; i++, p += 32) {
 			if (dir->fsckflags & DIREMPWARN) {
 				*p = SLOT_EMPTY;
 				continue;
@@ -572,7 +620,7 @@ readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
 						u_char *q;
 
 						dir->fsckflags &= ~DIREMPTY;
-						if (delete(f, boot, fat,
+						if (delete(fat,
 							   empcl, empty - buffer,
 							   cl, p - buffer, 1) == FSFATAL)
 							return FSFATAL;
@@ -701,7 +749,7 @@ readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
 
 			if (dirent.flags & ATTR_VOLUME) {
 				if (vallfn || invlfn) {
-					mod |= removede(f, boot, fat,
+					mod |= removede(fat,
 							invlfn ? invlfn : vallfn, p,
 							invlfn ? invcl : valcl, -1, 0,
 							fullpath(dir), 2);
@@ -741,7 +789,7 @@ readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
 			dirent.next = dir->child;
 
 			if (invlfn) {
-				mod |= k = removede(f, boot, fat,
+				mod |= k = removede(fat,
 						    invlfn, vallfn ? vallfn : p,
 						    invcl, vallfn ? valcl : cl, cl,
 						    fullpath(&dirent), 0);
@@ -757,74 +805,61 @@ readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
 			vallfn = NULL; /* not used any longer */
 			invlfn = NULL;
 
-			if (dirent.size == 0 && !(dirent.flags & ATTR_DIRECTORY)) {
-				if (dirent.head != 0) {
-					pwarn("%s has clusters, but size 0\n",
-					      fullpath(&dirent));
-					if (ask(1, "Drop allocated clusters")) {
-						p[26] = p[27] = 0;
-						if (boot->ClustMask == CLUST32_MASK)
-							p[20] = p[21] = 0;
-						clearchain(boot, fat, dirent.head);
-						dirent.head = 0;
-						mod |= THISMOD|FSDIRMOD|FSFATMOD;
-					} else
-						mod |= FSERROR;
-				}
-			} else if (dirent.head == 0
-				   && !strcmp(dirent.name, "..")
-				   && dir->parent			/* XXX */
-				   && !dir->parent->parent) {
-				/*
-				 *  Do nothing, the parent is the root
-				 */
-			} else if (dirent.head < CLUST_FIRST
-				   || dirent.head >= boot->NumClusters
-				   || fat[dirent.head].next == CLUST_FREE
-				   || (fat[dirent.head].next >= CLUST_RSRVD
-				       && fat[dirent.head].next < CLUST_EOFS)
-				   || fat[dirent.head].head != dirent.head) {
-				if (dirent.head == 0)
-					pwarn("%s has no clusters\n",
-					      fullpath(&dirent));
-				else if (dirent.head < CLUST_FIRST
-					 || dirent.head >= boot->NumClusters)
-					pwarn("%s starts with cluster out of range(%u)\n",
-					      fullpath(&dirent),
-					      dirent.head);
-				else if (fat[dirent.head].next == CLUST_FREE)
-					pwarn("%s starts with free cluster\n",
-					      fullpath(&dirent));
-				else if (fat[dirent.head].next >= CLUST_RSRVD)
-					pwarn("%s starts with cluster marked %s\n",
-					      fullpath(&dirent),
-					      rsrvdcltype(fat[dirent.head].next));
-				else
-					pwarn("%s doesn't start a cluster chain\n",
-					      fullpath(&dirent));
-				if (dirent.flags & ATTR_DIRECTORY) {
-					if (ask(0, "Remove")) {
-						*p = SLOT_DELETED;
-						mod |= THISMOD|FSDIRMOD;
-					} else
-						mod |= FSERROR;
-					continue;
-				} else {
-					if (ask(1, "Truncate")) {
-						p[28] = p[29] = p[30] = p[31] = 0;
-						p[26] = p[27] = 0;
-						if (boot->ClustMask == CLUST32_MASK)
-							p[20] = p[21] = 0;
-						dirent.size = 0;
-						mod |= THISMOD|FSDIRMOD;
-					} else
-						mod |= FSERROR;
+			/*
+			 * Check if the directory entry is sane.
+			 *
+			 * '.' and '..' are skipped, their sanity is
+			 * checked somewhere else.
+			 *
+			 * For everything else, check if we have a new,
+			 * valid cluster chain (beginning of a file or
+			 * directory that was never previously claimed
+			 * by another file) when it's a non-empty file
+			 * or a directory. The sanity of the cluster
+			 * chain is checked at a later time when we
+			 * traverse into the directory, or examine the
+			 * file's directory entry.
+			 *
+			 * The only possible fix is to delete the entry
+			 * if it's a directory; for file, we have to
+			 * truncate the size to 0.
+			 */
+			if (!(dirent.flags & ATTR_DIRECTORY) ||
+			    (strcmp(dirent.name, ".") != 0 &&
+			    strcmp(dirent.name, "..") != 0)) {
+				if ((dirent.size != 0 || (dirent.flags & ATTR_DIRECTORY)) &&
+				    ((!fat_is_valid_cl(fat, dirent.head) ||
+				    !fat_is_cl_head(fat, dirent.head)))) {
+					if (!fat_is_valid_cl(fat, dirent.head)) {
+						pwarn("%s starts with cluster out of range(%u)\n",
+						    fullpath(&dirent),
+						    dirent.head);
+					} else {
+						pwarn("%s doesn't start a new cluster chain\n",
+						    fullpath(&dirent));
+					}
+
+					if (dirent.flags & ATTR_DIRECTORY) {
+						if (ask(0, "Remove")) {
+							*p = SLOT_DELETED;
+							mod |= THISMOD|FSDIRMOD;
+						} else
+							mod |= FSERROR;
+						continue;
+					} else {
+						if (ask(1, "Truncate")) {
+							p[28] = p[29] = p[30] = p[31] = 0;
+							p[26] = p[27] = 0;
+							if (boot->ClustMask == CLUST32_MASK)
+								p[20] = p[21] = 0;
+							dirent.size = 0;
+							dirent.head = 0;
+							mod |= THISMOD|FSDIRMOD;
+						} else
+							mod |= FSERROR;
+					}
 				}
 			}
-
-			if (dirent.head >= CLUST_FIRST && dirent.head < boot->NumClusters)
-				fat[dirent.head].flags |= FAT_USED;
-
 			if (dirent.flags & ATTR_DIRECTORY) {
 				/*
 				 * gather more info for directories
@@ -861,8 +896,7 @@ readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
 							mod |= FSERROR;
 					}
 					continue;
-				}
-				if (strcmp(dirent.name, "..") == 0) {
+				} else if (strcmp(dirent.name, "..") == 0) {
 					if (dir->parent) {		/* XXX */
 						if (!dir->parent->parent) {
 							if (dirent.head) {
@@ -908,7 +942,7 @@ readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
 						} else
 							mod |= FSERROR;
 						continue;
-					} else if ((check_subdirectory(f, boot,
+					} else if ((check_subdirectory(fat,
 					    &dirent) & FSERROR) == FSERROR) {
 						/*
 						 * A subdirectory should have
@@ -944,39 +978,43 @@ readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
 				n->dir = d;
 				pendingDirectories = n;
 			} else {
-				mod |= k = checksize(boot, fat, p, &dirent);
+				mod |= k = checksize(fat, p, &dirent);
 				if (k & FSDIRMOD)
 					mod |= THISMOD;
 			}
 			boot->NumFiles++;
 		}
 
-		if (!(boot->flags & FAT32) && !dir->parent)
+		if (is_legacyroot) {
+			/*
+			 * Don't bother to write back right now because
+			 * we may continue to make modification to the
+			 * non-FAT32 root directory below.
+			 */
 			break;
-
-		if (mod & THISMOD) {
-			last *= 32;
-			if (lseek(f, off, SEEK_SET) != off
-			    || write(f, buffer, last) != last) {
+		} else if (mod & THISMOD) {
+			if (lseek(fd, off, SEEK_SET) != off
+			    || write(fd, buffer, iosize) != iosize) {
 				perr("Unable to write directory");
 				return FSFATAL;
 			}
 			mod &= ~THISMOD;
 		}
-	} while ((cl = fat[cl].next) >= CLUST_FIRST && cl < boot->NumClusters);
+	} while (fat_is_valid_cl(fat, (cl = fat_get_cl_next(fat, cl))));
 	if (invlfn || vallfn)
-		mod |= removede(f, boot, fat,
+		mod |= removede(fat,
 				invlfn ? invlfn : vallfn, p,
 				invlfn ? invcl : valcl, -1, 0,
 				fullpath(dir), 1);
 
-	/* The root directory of non fat32 filesystems is in a special
-	 * area and may have been modified above without being written out.
+	/*
+	 * The root directory of non-FAT32 filesystems is in a special
+	 * area and may have been modified above removede() without
+	 * being written out.
 	 */
-	if ((mod & FSDIRMOD) && !(boot->flags & FAT32) && !dir->parent) {
-		last *= 32;
-		if (lseek(f, off, SEEK_SET) != off
-		    || write(f, buffer, last) != last) {
+	if ((mod & FSDIRMOD) && is_legacyroot) {
+		if (lseek(fd, off, SEEK_SET) != off
+		    || write(fd, buffer, iosize) != iosize) {
 			perr("Unable to write directory");
 			return FSFATAL;
 		}
@@ -986,11 +1024,11 @@ readDosDirSection(int f, struct bootblock *boot, struct fatEntry *fat,
 }
 
 int
-handleDirTree(int dosfs, struct bootblock *boot, struct fatEntry *fat)
+handleDirTree(struct fat_descriptor *fat)
 {
 	int mod;
 
-	mod = readDosDirSection(dosfs, boot, fat, rootDir);
+	mod = readDosDirSection(fat, rootDir);
 	if (mod & FSFATAL)
 		return FSFATAL;
 
@@ -1011,7 +1049,7 @@ handleDirTree(int dosfs, struct bootblock *boot, struct fatEntry *fat)
 		/*
 		 * handle subdirectory
 		 */
-		mod |= readDosDirSection(dosfs, boot, fat, dir);
+		mod |= readDosDirSection(fat, dir);
 		if (mod & FSFATAL)
 			return FSFATAL;
 	}
@@ -1027,12 +1065,15 @@ static cl_t lfcl;
 static off_t lfoff;
 
 int
-reconnect(int dosfs, struct bootblock *boot, struct fatEntry *fat, cl_t head)
+reconnect(struct fat_descriptor *fat, cl_t head, size_t length)
 {
+	struct bootblock *boot = fat_get_boot(fat);
 	struct dosDirEntry d;
-	int len;
+	int len, dosfs;
 	u_char *p;
 
+	dosfs = fat_get_fd(fat);
+
 	if (!ask(1, "Reconnect"))
 		return FSERROR;
 
@@ -1063,7 +1104,7 @@ reconnect(int dosfs, struct bootblock *boot, struct fatEntry *fat, cl_t head)
 					break;
 		if (p && p < lfbuf + boot->ClusterSize)
 			break;
-		lfcl = p ? fat[lfcl].next : lostDir->head;
+		lfcl = p ? fat_get_cl_next(fat, lfcl) : lostDir->head;
 		if (lfcl < CLUST_FIRST || lfcl >= boot->NumClusters) {
 			/* Extend LOSTDIR?				XXX */
 			pwarn("No space in %s\n", LOSTDIR);
@@ -1088,7 +1129,7 @@ reconnect(int dosfs, struct bootblock *boot, struct fatEntry *fat, cl_t head)
 	len = snprintf(d.name, sizeof(d.name), "%u", head);
 	d.flags = 0;
 	d.head = head;
-	d.size = fat[head].length * boot->ClusterSize;
+	d.size = length * boot->ClusterSize;
 
 	memcpy(p, d.name, len);
 	memset(p + len, ' ', 11 - len);
@@ -1103,7 +1144,6 @@ reconnect(int dosfs, struct bootblock *boot, struct fatEntry *fat, cl_t head)
 	p[29] = (u_char)(d.size >> 8);
 	p[30] = (u_char)(d.size >> 16);
 	p[31] = (u_char)(d.size >> 24);
-	fat[head].flags |= FAT_USED;
 	if (lseek(dosfs, lfoff, SEEK_SET) != lfoff
 	    || (size_t)write(dosfs, lfbuf, boot->ClusterSize) != boot->ClusterSize) {
 		perr("could not write LOST.DIR");