posix-extensions-cvs.patch

diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/arch/i386/kernel/syscall_table.S vanilla-new/arch/i386/kernel/syscall_table.S
--- vanilla/arch/i386/kernel/syscall_table.S	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/arch/i386/kernel/syscall_table.S	2006-04-07 10:28:39.000000000 -0500
@@ -310,3 +310,5 @@
 	.long sys_pselect6
 	.long sys_ppoll
 	.long sys_unshare		/* 310 */
+	.long sys_openg
+	.long sys_openfh
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/arch/ia64/kernel/entry.S vanilla-new/arch/ia64/kernel/entry.S
--- vanilla/arch/ia64/kernel/entry.S	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/arch/ia64/kernel/entry.S	2006-04-07 10:28:39.000000000 -0500
@@ -1619,5 +1619,7 @@
 	data8 sys_ni_syscall			// reserved for pselect
 	data8 sys_ni_syscall			// 1295 reserved for ppoll
 	data8 sys_unshare
+	data8 sys_openg
+	data8 sys_openfh
 
 	.org sys_call_table + 8*NR_syscalls	// guard against failures to increase NR_syscalls
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/arch/powerpc/kernel/systbl.S vanilla-new/arch/powerpc/kernel/systbl.S
--- vanilla/arch/powerpc/kernel/systbl.S	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/arch/powerpc/kernel/systbl.S	2006-04-07 10:28:39.000000000 -0500
@@ -322,3 +322,5 @@
 COMPAT_SYS(pselect6)
 COMPAT_SYS(ppoll)
 SYSCALL(unshare)
+SYSCALL(openg)
+SYSCALL(openfh)
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/arch/x86_64/ia32/ia32entry.S vanilla-new/arch/x86_64/ia32/ia32entry.S
--- vanilla/arch/x86_64/ia32/ia32entry.S	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/arch/x86_64/ia32/ia32entry.S	2006-04-07 10:28:39.000000000 -0500
@@ -688,6 +688,8 @@
 	.quad sys_ni_syscall		/* pselect6 for now */
 	.quad sys_ni_syscall		/* ppoll for now */
 	.quad sys_unshare		/* 310 */
+	.quad compat_sys_openg
+	.quad compat_sys_openfh
 ia32_syscall_end:		
 	.rept IA32_NR_syscalls-(ia32_syscall_end-ia32_sys_call_table)/8
 		.quad ni_syscall
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/fs/compat.c vanilla-new/fs/compat.c
--- vanilla/fs/compat.c	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/fs/compat.c	2006-04-07 13:39:13.000000000 -0500
@@ -1332,6 +1332,27 @@
 }
 
 /*
+ * Exactly like fs/open.c: sys_openg(), except that it doesn't set the 
+ * O_LARGEFILE flag.
+ */
+asmlinkage long
+compat_sys_openg(const char __user *pathname, void __user *uhandle, size_t __user *uhandle_len,
+		int flags, int mode)
+{
+	return do_sys_openg(pathname, uhandle, uhandle_len, flags, mode);
+}
+
+/*
+ * Exactly like fs/open.c: sys_openg(), except that it doesn't set the 
+ * O_LARGEFILE flag.
+ */
+asmlinkage long
+compat_sys_openfh(const char __user *uhandle, int uhandle_len)
+{
+	return do_sys_openfh(uhandle, uhandle_len);
+}
+
+/*
  * compat_count() counts the number of arguments/envelopes. It is basically
  * a copy of count() from fs/exec.c, except that it works with 32 bit argv
  * and envp pointers.
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/fs/open.c vanilla-new/fs/open.c
--- vanilla/fs/open.c	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/fs/open.c	2006-04-26 22:06:15.000000000 -0500
@@ -27,9 +27,37 @@
 #include <linux/pagemap.h>
 #include <linux/syscalls.h>
 #include <linux/rcupdate.h>
+#include <linux/scatterlist.h>
+#include <linux/crypto.h>
+#include <linux/types.h>
 
 #include <asm/unistd.h>
 
+void get_filesystem(struct file_system_type *fs);
+void put_filesystem(struct file_system_type *fs);
+struct file_system_type *get_fs_type(const char *name);
+
+int vfs_statfs_lite(struct super_block *sb, struct kstatfs *buf, int statfs_mask)
+{
+	int retval = -ENODEV;
+
+	if (sb) {
+		retval = -ENOSYS;
+		if (sb->s_op->statfs_lite) {
+			memset(buf, 0, sizeof(*buf));
+			retval = security_sb_statfs(sb);
+			if (retval)
+				return retval;
+			retval = sb->s_op->statfs_lite(sb, buf, statfs_mask);
+			if (retval == 0 && buf->f_frsize == 0 && (statfs_mask & STATFS_M_FRSIZE) != 0)
+				buf->f_frsize = buf->f_bsize;
+		}
+	}
+	return retval;
+}
+
+EXPORT_SYMBOL(vfs_statfs_lite);
+
 int vfs_statfs(struct super_block *sb, struct kstatfs *buf)
 {
 	int retval = -ENODEV;
@@ -1100,6 +1128,638 @@
 }
 EXPORT_SYMBOL_GPL(sys_openat);
 
+static unsigned int diff(struct timeval *end, struct timeval *begin)
+{
+    if (end->tv_usec < begin->tv_usec) {
+        end->tv_usec += 1000000; end->tv_sec--;
+    }
+    end->tv_sec  -= begin->tv_sec;
+    end->tv_usec -= begin->tv_usec;
+    return ((end->tv_sec * 1000000) + (end->tv_usec));
+}
+
+/*
+ * Compute a simple crc checksum
+ */
+static inline __u32 simple_crc_csum(struct file_handle *fhandle)
+{
+	struct file_handle_generic *fhg;
+	__le32 *u;
+	__u8 *c;
+	__u32 i, j;
+	
+	fhg = &fhandle->fh_generic;
+	/* upto and not including the fhg_crc_csum field */
+	for (i = 0, u = (__le32 *) fhg; u < (__le32 *)(&fhg->fhg_crc_csum); ++u)
+		i += le32_to_cpup(u);
+	/* and over the entire fh_private buffer */
+	c = (__u8 *) fhandle->fh_private;
+	for (j = 0; j < fhandle->fh_private_length; ++j) 
+		i += *(c + j);
+	return i;
+}
+
+/*
+ * Compute and return a crc32c checksum using the crypto subsystem
+ * API
+ */
+static __u32 compute_crc32_csum(struct file_handle *fhandle)
+{
+#define MAX_SG_LIST 8
+	__u32 csum;
+	__u32 i;
+	struct crypto_tfm *tfm;
+	struct scatterlist sg[MAX_SG_LIST];
+	struct timeval begin, end;
+
+	do_gettimeofday(&begin);
+	if (crypto_alg_available("crc32c", 0) == 0) {
+		csum = simple_crc_csum(fhandle);
+		do_gettimeofday(&end);
+		printk(KERN_DEBUG "compute_crc32_csum[simple]: took %d usecs\n", diff(&end, &begin));
+		goto out;
+	}
+
+	tfm = crypto_alloc_tfm("crc32c", 0);
+	if (tfm == NULL) {
+		csum = simple_crc_csum(fhandle);
+		goto out;
+	}
+
+	crypto_digest_init(tfm);
+
+	i = 0;
+	sg_set_buf(&sg[i++], &fhandle->fh_generic, 
+			offsetof(struct file_handle_generic, fhg_crc_csum));
+	sg_set_buf(&sg[i++], fhandle->fh_private, fhandle->fh_private_length);
+
+	crypto_digest_update(tfm, sg, i);
+	crypto_digest_final(tfm, (__u8*)&csum);
+	crypto_free_tfm(tfm);
+	do_gettimeofday(&end);
+	printk(KERN_DEBUG "compute_crc32_csum[tfm]: took %d usecs\n", diff(&end, &begin));
+out:
+	return csum;
+#undef MAX_SG_LIST
+}
+
+/* 
+ * Store a crc32 check sum into the handle buffer
+ * NOTE: We expect fhandle->fh_private and fhandle->fh_private_length
+ * to be initialized prior to this call!
+ */
+static inline void store_crc32_csum(struct file_handle *fhandle)
+{
+	/* Compute the CRC checksum over the handle buffer */
+	__u32 csum = compute_crc32_csum(fhandle);
+	/* Set the CRC checksum into the handle buffer */
+	set_fh_field(&fhandle->fh_generic, crc_csum, csum);
+	return;
+}
+
+/* 
+ * Verifies if a file handle's structure's check sum
+ * matches with what user-space indicates.
+ * This should catch most of the simple buffer handle corruption
+ * cases (but not the maliciously crafted ones!) for which
+ * we have the keyed SHA1 (HMAC-SHA1) algorithm verification.
+ * NOTE: We expect fhandle->fh_private and fhandle->fh_private_length
+ * to be initialized prior to this call!
+ * Returns 1 in case verification was successful and 0
+ * otherwise.
+ */
+static inline int verify_crc32_csum(struct file_handle *fhandle)
+{
+	__u32 csum, crc_csum;
+
+	/* Compute the CRC checksum over the handle buffer */
+	csum = compute_crc32_csum(fhandle);
+	/* Retrieve the CRC checksum from the handle buffer */
+	get_fh_field(&fhandle->fh_generic, crc_csum, crc_csum);
+	return (crc_csum == csum);
+}
+
+/* 
+ * Compute a hmac-sha1 crypto check sum using a specified
+ * key and store it in the buffer pointed to by result.
+ * NOTE: We assume that fhandle->fh_private and
+ * fhandle->fh_private_length are initialized prior to this call.
+ * Returns 0 in case SHA1/HMAC-SHA1 was not built-in/compiled
+ * for the kernel
+ * Else returns length of the digest generated.
+ */
+static unsigned int compute_hmac_sha1_csum(struct file_handle *fhandle, 
+		char *key, int keylen, char *result)
+{
+#define MAX_SG_LIST 8
+	__u32 i;
+	struct crypto_tfm *tfm;
+	struct scatterlist sg[MAX_SG_LIST];
+	unsigned int ret;
+	struct timeval begin, end;
+ 
+	do_gettimeofday(&begin);
+
+	/* SHA1 is not available */
+	if (crypto_alg_available("sha1", 0) == 0)
+		return 0;
+
+	tfm = crypto_alloc_tfm("sha1", 0);
+	if (tfm == NULL)
+		return 0;
+
+	i = 0;
+	sg_set_buf(&sg[i++], &fhandle->fh_generic, 
+			offsetof(struct file_handle_generic, fhg_hmac_sha1));
+	sg_set_buf(&sg[i++], fhandle->fh_private, fhandle->fh_private_length);
+
+	crypto_hmac(tfm, key, &keylen, sg, i, result);
+	ret = crypto_tfm_alg_digestsize(tfm);
+	crypto_free_tfm(tfm);
+	do_gettimeofday(&end);
+	printk(KERN_DEBUG "compute_hmac_sha1_csum: took %d usecs\n", diff(&end, &begin));
+	return ret;
+#undef MAX_SG_LIST
+}
+
+/*
+ * Store a hmac-sha1 crypto check sum into the handle
+ * buffer.
+ */
+static inline void store_hmac_csum(struct file_handle *fhandle,
+		char *key, int keylen)
+{
+	compute_hmac_sha1_csum(fhandle, key, keylen,
+			fhandle->fh_generic.fhg_hmac_sha1);
+	return;
+}
+
+/*
+ * Verifies if a file handle buffer's computed hmac-sha1 crypto
+ * check sum matches with whatever the user-space buffer claims
+ * Returns 1 in case verification was succesful and 0
+ * otherwise
+ */
+static int verify_hmac_csum(struct file_handle *fhandle, char *key, int keylen)
+{
+	__u8 result[24];
+	unsigned int result_len;
+
+	result_len = compute_hmac_sha1_csum(fhandle, key, keylen, result);
+	/* Crypto is not supported. deny verification */
+	if (result_len == 0) 
+		return 0;
+
+	/* Check if user has tampered with the buffer */
+	if (memcmp(fhandle->fh_generic.fhg_hmac_sha1, result, result_len) == 0)
+		return 1;
+	return 0;
+}
+
+/*
+ * Copy the constructed file handle to a user-specified buffer.
+ * NOTE: We assume that fhandle->fh_private and
+ * fhandle->fh_private_length have been initialized
+ * prior to this routine
+ */
+static int copy_handle(const void __user *uaddr, struct file_handle *fhandle)
+{
+	char __user *ptr = (char __user *) uaddr;
+	struct timeval begin, end;
+ 
+	do_gettimeofday(&begin);
+
+	/* Copy the VFS generic part of the handle */
+	if (copy_to_user(ptr, &fhandle->fh_generic, 
+				sizeof(struct file_handle_generic)))
+		return -EFAULT;
+
+	/* Advance the user-buffer pointer */
+	ptr += sizeof(struct file_handle_generic);
+	/* Copy the FS specific part of the handle */
+	if (copy_to_user(ptr, fhandle->fh_private, fhandle->fh_private_length))
+		return -EFAULT;
+	do_gettimeofday(&end);
+	printk(KERN_DEBUG "copy_handle: took %d usecs\n", diff(&end, &begin));
+	return 0;
+}
+
+/*
+ * Copy a user-specified handle buffer to a temporary
+ * kernel buffer.
+ */
+static void* move_handle_to_kernel(const void __user *uaddr, 
+		size_t ulen, void *kaddr)
+{
+	if (uaddr != NULL) {
+		if (copy_from_user(kaddr, uaddr, ulen))
+			return ERR_PTR(-EFAULT);
+	}
+	return kaddr;
+}
+
+/*
+ * Copy a specified size user-space handle buffer to a temporary
+ * kernel buffer and return a pointer to the kernel buffer.
+ * Encodes error in return pointer in case 
+ * operation failed for some reason.
+ */
+static void *gethandle(const void __user *uhandle, size_t uhandle_len)
+{
+	void *tmp = NULL, *result = ERR_PTR(-EINVAL);
+	struct timeval begin, end;
+ 
+	do_gettimeofday(&begin);
+
+	do {
+		if (uhandle_len <= 0 || uhandle_len > MAX_HANDLE_LEN) 
+			break;
+		result = ERR_PTR(-ENOMEM);
+		tmp = kmalloc(uhandle_len, GFP_KERNEL);
+		if (tmp == NULL) 
+			break;
+		result = move_handle_to_kernel(uhandle, uhandle_len, tmp);
+	} while (0);
+	do_gettimeofday(&end);
+	printk(KERN_DEBUG "gethandle: took %d usecs\n", diff(&end, &begin));
+	return result;
+}
+/*
+ * Frees memory allocated to a temporary kernel space buffer
+ * used for staging the handle copy from user-space.
+ */
+static void puthandle(void *handle)
+{
+	if (handle)
+		kfree(handle);
+	return;
+}
+
+/*
+ * Wrapper routine to obtain any FS specific keys that can be used
+ * for the HMAC-SHA1 calculation to make the handle returned
+ * by openg() tamper-proof or atleast tamper-detectable.
+ *	We dont use any keys in case FS does not provide one!
+ */
+static void get_fs_key(struct super_block *sb, char **ppkey, int *keylen)
+{
+	struct timeval begin, end;
+
+	*ppkey = NULL;
+	*keylen = 0;
+	do_gettimeofday(&begin);
+	/* underlying fs wishes to provide a key */
+	if (sb && sb->s_op && sb->s_op->get_fs_key)
+		sb->s_op->get_fs_key(sb, ppkey, keylen);
+	do_gettimeofday(&end);
+	printk(KERN_DEBUG "get_fs_key: took %d usecs\n", diff(&end, &begin));
+	return;
+}
+
+/*
+ * Free the memory allocated to fhandle->fh_private.
+ * This is done by either calling the provided destructor
+ * function if any (supplied by fill_handle callback of FS).
+ * or by calling kfree(). We expect FS to either use kmalloc()
+ * or supply a dtor function so that fh_private can be properly
+ * deallocated.
+ */
+static void drop_file_handle_private(struct file_handle *fhandle)
+{
+	if (!fhandle || !fhandle->fh_private)
+		return;
+	if (fhandle->fh_private_dtor) 
+		fhandle->fh_private_dtor(fhandle->fh_private);
+	else 
+		kfree(fhandle->fh_private);
+	fhandle->fh_private = NULL;
+	return;
+}
+
+/*
+ * Associate and establish a file handle to a struct file mapping.
+ * NOTE: Linux VFS is inherently based off a dentry cache model
+ * and since we dont have any pathnames/filenames we need to create a anon.
+ * dentry to represent files opened using this system call.
+ */
+static struct file *do_fh_open(const void *handle, size_t handle_len)
+{
+	struct file *filp = NULL;
+	struct super_block *sb;
+	struct inode *inode = NULL;
+	struct file_handle fhandle;
+	size_t min_len = sizeof(struct file_handle_generic);
+	char *key = NULL;
+	int   keylen = 0;
+	struct timeval begin, end;
+	
+	/* Do some sanity checks on the handle lengths and handle buffer */
+	if (handle_len <= min_len)
+		return ERR_PTR(-EINVAL);
+
+	memcpy(&fhandle.fh_generic, handle, min_len);
+	/* set the fields of the file handle prior to verification */
+	fhandle.fh_private = (char *) handle + min_len;
+	fhandle.fh_private_length = (handle_len - min_len);
+	fhandle.fh_private_dtor = NULL;
+	/* verify crc32 checksum on the handle buffer */
+	if (!verify_crc32_csum(&fhandle))
+		return ERR_PTR(-EACCES);
+
+	/*
+	 * Verification of HMAC SHA1 csum on the handle buffer cannot be 
+	 * done at this point  since we don't know which file
+	 * system/superblock it belongs to.
+	 */
+
+	/* Locate and get an active reference to a 
+	 * super-block that matches a given handle 
+	 */
+	do_gettimeofday(&begin);
+	sb = get_sb_handle(&fhandle);
+	/* could not find any valid FS that owns this handle */
+	if (!sb)
+		return ERR_PTR(-EINVAL);
+
+	do_gettimeofday(&end);
+	printk(KERN_DEBUG "get_sb_handle: took %d usecs\n", diff(&end, &begin));
+	/* FS has not implemented a routine to query inode based on handle */
+	if (!sb->s_op || !sb->s_op->find_inode_handle) {
+		filp = ERR_PTR(-EOPNOTSUPP);
+		goto out;
+	}
+	/* Check if underlying FS wishes to use a key */
+	get_fs_key(sb, &key, &keylen);
+	/* and verify the authenticity of the handle buffer */
+	if (verify_hmac_csum(&fhandle, key, keylen) == 0) {
+		/* Permission denied */
+		filp = ERR_PTR(-EACCES);
+		goto out;
+	}
+	do_gettimeofday(&begin);
+	/* find an inode or allocate/lookup an inode based on the handle */
+	inode = sb->s_op->find_inode_handle(sb, &fhandle);
+	if (!IS_ERR(inode)) {
+		struct dentry *anon_dentry;
+		int flags = 0;
+
+		do_gettimeofday(&end);
+		printk(KERN_DEBUG "s_op->find_inode_handle: took %d usecs\n", diff(&end, &begin));
+
+		get_fh_field(&fhandle.fh_generic, flags, flags);
+		/* 
+		 * Associate a anonymous dentry to this inode in case there is none.
+		 * d_prune_aliases() will take care of freeeing all these dentries
+		 * which is called when the last reference to iput() calls
+		 * put_inode() which in turn calls the above function.
+		 */
+		do_gettimeofday(&begin);
+		anon_dentry = d_alloc_anon(inode);
+		if (anon_dentry == NULL) {
+			filp = ERR_PTR(-ENOMEM);
+			goto drop_inode;
+		}
+		/* and setup file and inode mapping */
+		filp = get_empty_filp();
+		if (filp == NULL) {
+			filp = ERR_PTR(-ENFILE);
+			goto drop_dentry;
+		}
+		filp = __dentry_open(anon_dentry, NULL, flags, filp, NULL);
+		if (IS_ERR(filp)) {
+			/* in case of error __dentry_open drops the anon_dentry */
+			goto drop_inode;
+	drop_dentry:
+			dput(anon_dentry);
+	drop_inode:
+			iput(inode);
+		}
+		else {
+			do_gettimeofday(&end);
+			printk(KERN_DEBUG "d_alloc_anon + misc.: took %d usecs\n", diff(&end, &begin));
+		}
+	}
+	else {
+		/* in case of error */
+		filp = ERR_PTR(PTR_ERR(inode));
+	}
+out:
+	/* drop our active reference to sb */
+	drop_super(sb);
+	return filp;
+}
+
+long do_sys_openfh(const void __user *uhandle, size_t handle_len)
+{
+	void *tmp = gethandle(uhandle, handle_len);
+	int fd = PTR_ERR(tmp);
+
+	if (!IS_ERR(tmp)) {
+		struct timeval begin, end;
+
+		do_gettimeofday(&begin);
+		fd = get_unused_fd();
+		do_gettimeofday(&end);
+		printk(KERN_DEBUG "get_unused_fd: took %d usecs\n", diff(&end, &begin));
+		if (fd >= 0) {
+			struct file *f;
+
+			do_gettimeofday(&begin);
+			f = do_fh_open(tmp, handle_len);
+			if (IS_ERR(f)) {
+				put_unused_fd(fd);
+				fd = PTR_ERR(f);
+			} else {
+				do_gettimeofday(&end);
+				printk(KERN_DEBUG "do_fh_open: took %d usecs\n", diff(&end, &begin));
+				do_gettimeofday(&begin);
+				/* this file descriptor needs to be closed on exec */
+				set_close_on_exec(fd, 1);
+				fd_install(fd, f);
+				do_gettimeofday(&end);
+				printk(KERN_DEBUG "set_close, fd_install: took %d usecs\n", diff(&end, &begin));
+			}
+		}
+		puthandle(tmp);
+	}
+	return fd;
+}
+
+asmlinkage long sys_openfh(const void __user *uhandle, size_t handle_len)
+{
+	return do_sys_openfh(uhandle, handle_len);
+}
+
+EXPORT_SYMBOL_GPL(sys_openfh);
+
+long do_sys_openg(const char __user *pathname, void __user *uhandle,
+		size_t __user *uhandle_len, int flags, int mode)
+{
+	char *tmp = getname(pathname);
+	long err = PTR_ERR(tmp);
+
+	if (!IS_ERR(tmp)) 
+	{
+		struct file *f = NULL;
+		size_t handle_len = 0;
+		size_t min_len = sizeof(struct file_handle_generic);
+		struct file_handle fhandle;
+		struct timeval begin, end;
+
+		memset(&fhandle, 0, sizeof(fhandle));
+		if (get_user(handle_len, uhandle_len)) {
+			err = -EFAULT;
+			goto drop_name;
+		}
+		/* discard bogus values */
+		if (handle_len < 0)
+		{
+			err = -EINVAL;
+			goto drop_name;
+		}
+		/* or definitely insufficient buffer length */
+		else if (handle_len > 0 && 
+				handle_len < min_len)
+		{
+			err = -ERANGE;
+			goto drop_name;
+		}
+		err = 0;
+		do_gettimeofday(&begin);
+		/* Try to open the file now */
+		f = do_filp_open(AT_FDCWD, tmp, flags, mode);
+		if (IS_ERR(f)) {
+			/* File does not exist */
+			err = PTR_ERR(f);
+			goto drop_name;
+		}
+		else {
+			/* Is this even possible? */
+			if (!f->f_dentry || !f->f_dentry->d_inode)
+			{
+				err = -EINVAL;
+				goto drop_filp;
+			}
+			/* Diallow anything but regular files */
+			if (!S_ISREG(f->f_dentry->d_inode->i_mode)) {
+				err = -EACCES;
+				goto drop_filp;
+			}
+			else {
+				/* Does FS define callback for openg? */
+				if (!f->f_dentry->d_inode->i_op->fill_handle) {
+					/* not supported */
+					err = -EOPNOTSUPP;
+					goto drop_filp;
+				}
+				else {
+					do_gettimeofday(&end);
+					printk(KERN_DEBUG "do_filp_open: took %d usecs\n", diff(&end, &begin));
+					fhandle.fh_private = NULL;
+					fhandle.fh_private_length = 
+						handle_len ? (handle_len - min_len)	: 0;
+					fhandle.fh_private_dtor = NULL;
+					/* 
+					 * if handle_len == 0, 
+					 * FS is expected to give us the handle length it needs. 
+					 */
+					do_gettimeofday(&begin);
+					err = f->f_dentry->d_inode->i_op->fill_handle(
+								f->f_dentry->d_inode, &fhandle);
+					if (err)
+						goto drop_filp;
+
+					/* make sure that FS does not set it to an invalid length */
+					if (fhandle.fh_private_length <= 0) {
+						err = -EINVAL;
+						goto drop_private;
+					}
+					
+					/* try to copy the handle if requested */
+					if (handle_len > 0) {
+						__u32 magic, fsid;
+						struct kstatfs st;
+						char *key = NULL;
+						int  keylen = 0, statfs_mask = STATFS_M_TYPE | STATFS_M_FSID;
+						/*
+						 * Check if fhandle.fh_private was not set.
+						 */
+						if (fhandle.fh_private == NULL) {
+							err = -EINVAL;
+							goto drop_filp;
+						}
+						do_gettimeofday(&end);
+						printk(KERN_DEBUG "fill_handle: took %d usecs\n", diff(&end, &begin));
+						/*
+						 * Check if the underlying FS wishes to supply a key for 
+						 * the message authentication code.
+						 * If it does not or in case of errors, we use a NULL key
+						 * for the MAC
+						 */
+						get_fs_key(f->f_dentry->d_inode->i_sb, &key, &keylen);
+						/* 
+						 * Currently, we issue a statfs on the superblock to figure
+						 * out the magic number and fsid for the file system.
+						 * But this might translate to many network messages for many
+						 * distributed/parallel file systems. Therefore, we need a 
+						 * sb->statfs_lite() callback mechanism to get only those
+						 * fields that don't require a network message.
+						 * If FS does not support the callback, fallback to using statfs.
+						 */
+						do_gettimeofday(&begin);
+						if ((err = vfs_statfs_lite(f->f_dentry->d_inode->i_sb, &st, statfs_mask)) < 0) {
+							//err = vfs_statfs(f->f_dentry->d_inode->i_sb, &st);
+							//if (err) 
+								goto drop_private;
+						}
+						do_gettimeofday(&end);
+						printk(KERN_DEBUG "statfs_lite: took %d usecs\n", diff(&end, &begin));
+						magic = st.f_type;
+						memcpy(&fsid, &st.f_fsid, sizeof(__u32));
+						/* set magic number, fsid, flags */
+						set_fh_field(&fhandle.fh_generic, magic, magic);
+						set_fh_field(&fhandle.fh_generic, fsid,  fsid);
+						set_fh_field(&fhandle.fh_generic, flags, flags);
+
+						do_gettimeofday(&begin);
+						/* Compute and store crc check sum */
+						store_crc32_csum(&fhandle);
+						/* Compute and store HMAC-SHA1 crypto check sum */
+						store_hmac_csum(&fhandle, key, keylen);
+						/* copy opaque handle to user buffer */
+						err = copy_handle(uhandle, &fhandle);
+						if (err)
+							goto drop_private;
+						do_gettimeofday(&end);
+						printk(KERN_DEBUG "(crc, hmac, copy_handle): took %d usecs\n", diff(&end, &begin));
+					}
+					/* update the handle length reported to user-space */
+					handle_len = fhandle.fh_private_length + min_len;
+					/* and copy the new length to user space */
+					err = put_user(handle_len, uhandle_len);
+			drop_private:
+					drop_file_handle_private(&fhandle);
+				}
+			}
+drop_filp:
+			filp_close(f, NULL);
+		}
+drop_name:
+		putname(tmp);
+	}
+	return err;
+}
+
+asmlinkage long sys_openg(const char __user *pathname, void __user *uhandle, 
+		size_t __user *uhandle_len, int flags, int mode)
+{
+	if (force_o_largefile())
+		flags |= O_LARGEFILE;
+	return do_sys_openg(pathname, uhandle, uhandle_len, flags, mode);
+}
+
+EXPORT_SYMBOL_GPL(sys_openg);
+
 #ifndef __alpha__
 
 /*
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/fs/super.c vanilla-new/fs/super.c
--- vanilla/fs/super.c	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/fs/super.c	2006-04-26 22:06:36.000000000 -0500
@@ -799,6 +799,58 @@
 
 EXPORT_SYMBOL(get_sb_single);
 
+struct super_block *get_sb_handle(const struct file_handle *fhandle)
+{
+	/* walk through the list of superblocks */
+	struct super_block *sb;
+
+	spin_lock(&sb_lock);
+	list_for_each_entry (sb, &super_blocks, s_list) {
+		struct kstatfs st;
+
+		sb->s_count++;
+		spin_unlock(&sb_lock);
+		down_read(&sb->s_umount);
+		if (sb->s_root)
+		{
+			int err, statfs_mask = STATFS_M_TYPE | STATFS_M_FSID;
+			/* 
+			 * Currently, we issue a statfs on the superblock to figure
+			 * out the type and fsid for the file system.
+			 * But this might translate to network messages for many
+			 * distributed/parallel file systems. Therefore, we need a 
+			 * statfs_lite() callback mechanism to get only those
+			 * fields that don't require a network message. We fall back to a
+			 * regular statfs in case the file system does not support a statfs_lite()
+			 * callback.
+			 */
+			if ((err = vfs_statfs_lite(sb, &st, statfs_mask)) < 0) {
+				//err = vfs_statfs(sb, &st);
+			}
+			
+			if (err == 0) {
+				__u32 fsid, magic; 
+
+				get_fh_field(&fhandle->fh_generic, fsid, fsid);
+				get_fh_field(&fhandle->fh_generic, magic, magic);
+				/* check if magic numbers and fsid matches */
+				if (st.f_type == magic
+						&& memcmp(&st.f_fsid, &fsid, sizeof(__u32)) == 0)
+				{
+					return sb;
+				}
+			}
+		}
+		drop_super(sb);
+		spin_lock(&sb_lock);
+	}
+	spin_unlock(&sb_lock);
+	/* This function will not create a new superblock! */
+	return NULL;
+}
+
+EXPORT_SYMBOL_GPL(get_sb_handle);
+
 struct vfsmount *
 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
 {
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/include/asm-i386/unistd.h vanilla-new/include/asm-i386/unistd.h
--- vanilla/include/asm-i386/unistd.h	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/include/asm-i386/unistd.h	2006-04-07 10:28:39.000000000 -0500
@@ -316,8 +316,10 @@
 #define __NR_pselect6		308
 #define __NR_ppoll		309
 #define __NR_unshare		310
+#define __NR_openg      311
+#define __NR_openfh     312
 
-#define NR_syscalls 311
+#define NR_syscalls 313
 
 /*
  * user-visible error numbers are in the range -1 - -128: see
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/include/asm-ia64/unistd.h vanilla-new/include/asm-ia64/unistd.h
--- vanilla/include/asm-ia64/unistd.h	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/include/asm-ia64/unistd.h	2006-04-07 10:28:39.000000000 -0500
@@ -285,12 +285,14 @@
 #define __NR_faccessat			1293
 /* 1294, 1295 reserved for pselect/ppoll */
 #define __NR_unshare			1296
+#define __NR_openg         1297
+#define __NR_openfh        1298
 
 #ifdef __KERNEL__
 
 #include <linux/config.h>
 
-#define NR_syscalls			273 /* length of syscall table */
+#define NR_syscalls			275 /* length of syscall table */
 
 #define __ARCH_WANT_SYS_RT_SIGACTION
 
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/include/asm-powerpc/unistd.h vanilla-new/include/asm-powerpc/unistd.h
--- vanilla/include/asm-powerpc/unistd.h	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/include/asm-powerpc/unistd.h	2006-04-07 10:28:39.000000000 -0500
@@ -301,8 +301,10 @@
 #define __NR_pselect6		280
 #define __NR_ppoll		281
 #define __NR_unshare		282
+#define __NR_openg		283
+#define __NR_openfh		284
 
-#define __NR_syscalls		283
+#define __NR_syscalls		285
 
 #ifdef __KERNEL__
 #define __NR__exit __NR_exit
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/include/asm-x86_64/unistd.h vanilla-new/include/asm-x86_64/unistd.h
--- vanilla/include/asm-x86_64/unistd.h	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/include/asm-x86_64/unistd.h	2006-04-07 10:28:39.000000000 -0500
@@ -605,8 +605,12 @@
 __SYSCALL(__NR_ppoll,	sys_ni_syscall)		/* for now */
 #define __NR_unshare		272
 __SYSCALL(__NR_unshare,	sys_unshare)
+#define __NR_openg      273
+__SYSCALL(__NR_openg,   sys_openg)
+#define __NR_openfh     274
+__SYSCALL(__NR_openfh,  sys_openfh)
 
-#define __NR_syscall_max __NR_unshare
+#define __NR_syscall_max __NR_openfh
 
 #ifndef __NO_STUBS
 
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/include/linux/fs.h vanilla-new/include/linux/fs.h
--- vanilla/include/linux/fs.h	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/include/linux/fs.h	2006-04-17 13:35:37.000000000 -0500
@@ -991,6 +991,27 @@
 #define HAVE_COMPAT_IOCTL 1
 #define HAVE_UNLOCKED_IOCTL 1
 
+struct file_handle_generic {
+	/* Filled by VFS */
+	__le32   fhg_magic; /* magic number */
+	__le32   fhg_fsid; /* file system identifier */
+	__le32   fhg_flags; 	/* flags associated with the file object */
+	__le32   fhg_crc_csum; /* crc32c check sum of the blob */
+	__u8     fhg_hmac_sha1[24]; /* hmac-sha1 message authentication code */
+};
+
+struct file_handle {
+	/* generic part filled by VFS */
+	struct file_handle_generic fh_generic;
+	/* FS specific part */
+	void                      *fh_private;
+	size_t							fh_private_length;
+	void                      (*fh_private_dtor)(void *);
+};
+
+#define set_fh_field(X, Y, Z) (X)->fhg_##Y = cpu_to_le32(Z)
+#define get_fh_field(X, Y, Z) Z = le32_to_cpu((X)->fhg_##Y)
+
 /*
  * NOTE:
  * read, write, poll, fsync, readv, writev, unlocked_ioctl and compat_ioctl
@@ -1049,6 +1070,8 @@
 	ssize_t (*listxattr) (struct dentry *, char *, size_t);
 	int (*removexattr) (struct dentry *, const char *);
 	void (*truncate_range)(struct inode *, loff_t, loff_t);
+	/* method needed for filling up handle */
+	int (*fill_handle)(struct inode *, struct file_handle *);
 };
 
 struct seq_file;
@@ -1065,12 +1088,12 @@
  * without the big kernel lock held in all filesystems.
  */
 struct super_operations {
-   	struct inode *(*alloc_inode)(struct super_block *sb);
+   struct inode *(*alloc_inode)(struct super_block *sb);
 	void (*destroy_inode)(struct inode *);
 
 	void (*read_inode) (struct inode *);
   
-   	void (*dirty_inode) (struct inode *);
+  	void (*dirty_inode) (struct inode *);
 	int (*write_inode) (struct inode *, int);
 	void (*put_inode) (struct inode *);
 	void (*drop_inode) (struct inode *);
@@ -1081,6 +1104,7 @@
 	void (*write_super_lockfs) (struct super_block *);
 	void (*unlockfs) (struct super_block *);
 	int (*statfs) (struct super_block *, struct kstatfs *);
+	int (*statfs_lite) (struct super_block *, struct kstatfs *, int statfs_mask);
 	int (*remount_fs) (struct super_block *, int *, char *);
 	void (*clear_inode) (struct inode *);
 	void (*umount_begin) (struct super_block *);
@@ -1089,6 +1113,8 @@
 
 	ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
 	ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
+	struct inode *(*find_inode_handle)(struct super_block *, const struct file_handle *);
+	void    (*get_fs_key)(struct super_block *, char **ppkey, int *keylen);
 };
 
 /* Inode state bits.  Protected by inode_lock. */
@@ -1257,6 +1283,7 @@
 struct super_block *get_sb_nodev(struct file_system_type *fs_type,
 	int flags, void *data,
 	int (*fill_super)(struct super_block *, void *, int));
+struct super_block *get_sb_handle(const struct file_handle *handle);
 void generic_shutdown_super(struct super_block *sb);
 void kill_block_super(struct super_block *sb);
 void kill_anon_super(struct super_block *sb);
@@ -1292,6 +1319,7 @@
 				  struct vfsmount *);
 
 extern int vfs_statfs(struct super_block *, struct kstatfs *);
+extern int vfs_statfs_lite(struct super_block *, struct kstatfs *, int statfs_mask);
 
 /* /sys/fs */
 extern struct subsystem fs_subsys;
@@ -1345,6 +1373,10 @@
 		       struct file *filp);
 extern long do_sys_open(int fdf, const char __user *filename, int flags,
 			int mode);
+
+extern long do_sys_openfh(const void __user *uhandle, size_t handle_len);
+extern long do_sys_openg(const char __user *pathname, void __user *uhandle, 
+			size_t __user *uhandle_len, int flags, int mode);
 extern struct file *filp_open(const char *, int, int);
 extern struct file * dentry_open(struct dentry *, struct vfsmount *, int);
 extern int filp_close(struct file *, fl_owner_t id);
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/include/linux/limits.h vanilla-new/include/linux/limits.h
--- vanilla/include/linux/limits.h	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/include/linux/limits.h	2006-04-07 17:25:07.000000000 -0500
@@ -16,6 +16,7 @@
 #define XATTR_NAME_MAX   255	/* # chars in an extended attribute name */
 #define XATTR_SIZE_MAX 65536	/* size of an extended attribute value (64k) */
 #define XATTR_LIST_MAX 65536	/* size of extended attribute namelist (64k) */
+#define MAX_HANDLE_LEN   128  /* maximum size of a handle used by the openfh system call */
 
 #define RTSIG_MAX	  32
 
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/include/linux/statfs.h vanilla-new/include/linux/statfs.h
--- vanilla/include/linux/statfs.h	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/include/linux/statfs.h	2006-04-17 11:38:07.000000000 -0500
@@ -5,6 +5,18 @@
 
 #include <asm/statfs.h>
 
+/* Masks used by statfs_lite callback */
+#define STATFS_M_TYPE  		(1 << 0)
+#define STATFS_M_BSIZE  	(1 << 1)
+#define STATFS_M_BLOCKS		(1 << 2)
+#define STATFS_M_BFREE  	(1 << 3)
+#define STATFS_M_BAVAIL 	(1 << 4)
+#define STATFS_M_FILES  	(1 << 5)
+#define STATFS_M_FFREE  	(1 << 6)
+#define STATFS_M_FSID   	(1 << 7)
+#define STATFS_M_NAMELEN	(1 << 8)
+#define STATFS_M_FRSIZE    (1 << 9)
+
 struct kstatfs {
 	long f_type;
 	long f_bsize;
diff -Naur --exclude-from=/home/vilayann/redhat/BUILD/kernel-2.6.16/exclude vanilla/include/linux/syscalls.h vanilla-new/include/linux/syscalls.h
--- vanilla/include/linux/syscalls.h	2006-03-19 23:53:29.000000000 -0600
+++ vanilla-new/include/linux/syscalls.h	2006-04-07 13:56:07.000000000 -0500
@@ -327,6 +327,10 @@
 asmlinkage long sys_creat(const char __user *pathname, int mode);
 asmlinkage long sys_open(const char __user *filename,
 				int flags, int mode);
+asmlinkage long sys_openfh(const void __user *uhandle, 
+				size_t handle_len);
+asmlinkage long sys_openg(const char __user *pathname, void __user *uhandle, 
+				size_t *uhandle_len, int flags, int mode);
 asmlinkage long sys_close(unsigned int fd);
 asmlinkage long sys_access(const char __user *filename, int mode);
 asmlinkage long sys_vhangup(void);
@@ -566,7 +570,10 @@
 asmlinkage long compat_sys_newfstatat(unsigned int dfd, char __user * filename,
 				      struct compat_stat __user *statbuf,
 				      int flag);
+asmlinkage long compat_sys_open(const char __user *filename, int flags, int mode);
 asmlinkage long compat_sys_openat(unsigned int dfd, const char __user *filename,
 				   int flags, int mode);
-
+asmlinkage long compat_sys_openg(const char __user *pathname, void __user *uhandle,
+					size_t __user *uhandle_len, int flags, int mode);
+asmlinkage long compat_sys_openfh(const void __user *uhandle, size_t uhandle_len);
 #endif