ext4: avoid hangs in ext4_da_should_update_i_disksize()
[linux-2.6.git] / fs / read_write.c
index df3468a..179f1c3 100644 (file)
 #include <linux/fcntl.h>
 #include <linux/file.h>
 #include <linux/uio.h>
-#include <linux/smp_lock.h>
 #include <linux/fsnotify.h>
 #include <linux/security.h>
 #include <linux/module.h>
 #include <linux/syscalls.h>
 #include <linux/pagemap.h>
+#include <linux/splice.h>
+#include "read_write.h"
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
 
-struct file_operations generic_ro_fops = {
+const struct file_operations generic_ro_fops = {
        .llseek         = generic_file_llseek,
-       .read           = generic_file_read,
+       .read           = do_sync_read,
+       .aio_read       = generic_file_aio_read,
        .mmap           = generic_file_readonly_mmap,
-       .sendfile       = generic_file_sendfile,
+       .splice_read    = generic_file_splice_read,
 };
 
 EXPORT_SYMBOL(generic_ro_fops);
 
-loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
+static inline int unsigned_offsets(struct file *file)
+{
+       return file->f_mode & FMODE_UNSIGNED_OFFSET;
+}
+
+/**
+ * generic_file_llseek_unlocked - lockless generic llseek implementation
+ * @file:      file structure to seek on
+ * @offset:    file offset to seek to
+ * @origin:    type of seek
+ *
+ * Updates the file offset to the value specified by @offset and @origin.
+ * Locking must be provided by the caller.
+ */
+loff_t
+generic_file_llseek_unlocked(struct file *file, loff_t offset, int origin)
 {
-       long long retval;
        struct inode *inode = file->f_mapping->host;
 
-       down(&inode->i_sem);
        switch (origin) {
-               case 2:
-                       offset += inode->i_size;
-                       break;
-               case 1:
-                       offset += file->f_pos;
+       case SEEK_END:
+               offset += inode->i_size;
+               break;
+       case SEEK_CUR:
+               /*
+                * Here we special-case the lseek(fd, 0, SEEK_CUR)
+                * position-querying operation.  Avoid rewriting the "same"
+                * f_pos value back to the file because a concurrent read(),
+                * write() or lseek() might have altered it
+                */
+               if (offset == 0)
+                       return file->f_pos;
+               offset += file->f_pos;
+               break;
+       case SEEK_DATA:
+               /*
+                * In the generic case the entire file is data, so as long as
+                * offset isn't at the end of the file then the offset is data.
+                */
+               if (offset >= inode->i_size)
+                       return -ENXIO;
+               break;
+       case SEEK_HOLE:
+               /*
+                * There is a virtual hole at the end of the file, so as long as
+                * offset isn't i_size or larger, return i_size.
+                */
+               if (offset >= inode->i_size)
+                       return -ENXIO;
+               offset = inode->i_size;
+               break;
        }
-       retval = -EINVAL;
-       if (offset>=0 && offset<=inode->i_sb->s_maxbytes) {
-               if (offset != file->f_pos) {
-                       file->f_pos = offset;
-                       file->f_version = 0;
-               }
-               retval = offset;
+
+       if (offset < 0 && !unsigned_offsets(file))
+               return -EINVAL;
+       if (offset > inode->i_sb->s_maxbytes)
+               return -EINVAL;
+
+       /* Special lock needed here? */
+       if (offset != file->f_pos) {
+               file->f_pos = offset;
+               file->f_version = 0;
        }
-       up(&inode->i_sem);
-       return retval;
+
+       return offset;
 }
+EXPORT_SYMBOL(generic_file_llseek_unlocked);
+
+/**
+ * generic_file_llseek - generic llseek implementation for regular files
+ * @file:      file structure to seek on
+ * @offset:    file offset to seek to
+ * @origin:    type of seek
+ *
+ * This is a generic implemenation of ->llseek useable for all normal local
+ * filesystems.  It just updates the file offset to the value specified by
+ * @offset and @origin under i_mutex.
+ */
+loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
+{
+       loff_t rval;
+
+       mutex_lock(&file->f_dentry->d_inode->i_mutex);
+       rval = generic_file_llseek_unlocked(file, offset, origin);
+       mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 
+       return rval;
+}
 EXPORT_SYMBOL(generic_file_llseek);
 
-loff_t remote_llseek(struct file *file, loff_t offset, int origin)
+/**
+ * noop_llseek - No Operation Performed llseek implementation
+ * @file:      file structure to seek on
+ * @offset:    file offset to seek to
+ * @origin:    type of seek
+ *
+ * This is an implementation of ->llseek useable for the rare special case when
+ * userspace expects the seek to succeed but the (device) file is actually not
+ * able to perform the seek. In this case you use noop_llseek() instead of
+ * falling back to the default implementation of ->llseek.
+ */
+loff_t noop_llseek(struct file *file, loff_t offset, int origin)
 {
-       long long retval;
-
-       lock_kernel();
-       switch (origin) {
-               case 2:
-                       offset += i_size_read(file->f_dentry->d_inode);
-                       break;
-               case 1:
-                       offset += file->f_pos;
-       }
-       retval = -EINVAL;
-       if (offset>=0 && offset<=file->f_dentry->d_inode->i_sb->s_maxbytes) {
-               if (offset != file->f_pos) {
-                       file->f_pos = offset;
-                       file->f_version = 0;
-               }
-               retval = offset;
-       }
-       unlock_kernel();
-       return retval;
+       return file->f_pos;
 }
-EXPORT_SYMBOL(remote_llseek);
+EXPORT_SYMBOL(noop_llseek);
 
 loff_t no_llseek(struct file *file, loff_t offset, int origin)
 {
@@ -88,25 +145,55 @@ EXPORT_SYMBOL(no_llseek);
 
 loff_t default_llseek(struct file *file, loff_t offset, int origin)
 {
-       long long retval;
+       struct inode *inode = file->f_path.dentry->d_inode;
+       loff_t retval;
 
-       lock_kernel();
+       mutex_lock(&inode->i_mutex);
        switch (origin) {
-               case 2:
-                       offset += i_size_read(file->f_dentry->d_inode);
+               case SEEK_END:
+                       offset += i_size_read(inode);
                        break;
-               case 1:
+               case SEEK_CUR:
+                       if (offset == 0) {
+                               retval = file->f_pos;
+                               goto out;
+                       }
                        offset += file->f_pos;
+                       break;
+               case SEEK_DATA:
+                       /*
+                        * In the generic case the entire file is data, so as
+                        * long as offset isn't at the end of the file then the
+                        * offset is data.
+                        */
+                       if (offset >= inode->i_size) {
+                               retval = -ENXIO;
+                               goto out;
+                       }
+                       break;
+               case SEEK_HOLE:
+                       /*
+                        * There is a virtual hole at the end of the file, so
+                        * as long as offset isn't i_size or larger, return
+                        * i_size.
+                        */
+                       if (offset >= inode->i_size) {
+                               retval = -ENXIO;
+                               goto out;
+                       }
+                       offset = inode->i_size;
+                       break;
        }
        retval = -EINVAL;
-       if (offset >= 0) {
+       if (offset >= 0 || unsigned_offsets(file)) {
                if (offset != file->f_pos) {
                        file->f_pos = offset;
                        file->f_version = 0;
                }
                retval = offset;
        }
-       unlock_kernel();
+out:
+       mutex_unlock(&inode->i_mutex);
        return retval;
 }
 EXPORT_SYMBOL(default_llseek);
@@ -117,7 +204,6 @@ loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
 
        fn = no_llseek;
        if (file->f_mode & FMODE_LSEEK) {
-               fn = default_llseek;
                if (file->f_op && file->f_op->llseek)
                        fn = file->f_op->llseek;
        }
@@ -125,7 +211,7 @@ loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
 }
 EXPORT_SYMBOL(vfs_llseek);
 
-asmlinkage off_t sys_lseek(unsigned int fd, off_t offset, unsigned int origin)
+SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
 {
        off_t retval;
        struct file * file;
@@ -137,7 +223,7 @@ asmlinkage off_t sys_lseek(unsigned int fd, off_t offset, unsigned int origin)
                goto bad;
 
        retval = -EINVAL;
-       if (origin <= 2) {
+       if (origin <= SEEK_MAX) {
                loff_t res = vfs_llseek(file, offset, origin);
                retval = res;
                if (res != (loff_t)retval)
@@ -149,9 +235,9 @@ bad:
 }
 
 #ifdef __ARCH_WANT_SYS_LLSEEK
-asmlinkage long sys_llseek(unsigned int fd, unsigned long offset_high,
-                          unsigned long offset_low, loff_t __user * result,
-                          unsigned int origin)
+SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
+               unsigned long, offset_low, loff_t __user *, result,
+               unsigned int, origin)
 {
        int retval;
        struct file * file;
@@ -164,7 +250,7 @@ asmlinkage long sys_llseek(unsigned int fd, unsigned long offset_high,
                goto bad;
 
        retval = -EINVAL;
-       if (origin > 2)
+       if (origin > SEEK_MAX)
                goto out_putf;
 
        offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
@@ -183,36 +269,44 @@ bad:
 }
 #endif
 
+
 /*
  * rw_verify_area doesn't like huge counts. We limit
  * them to something that fits in "int" so that others
  * won't have to do range checks all the time.
  */
-#define MAX_RW_COUNT (INT_MAX & PAGE_CACHE_MASK)
-
 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
 {
        struct inode *inode;
        loff_t pos;
+       int retval = -EINVAL;
 
+       inode = file->f_path.dentry->d_inode;
        if (unlikely((ssize_t) count < 0))
-               goto Einval;
+               return retval;
        pos = *ppos;
-       if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))
-               goto Einval;
+       if (unlikely(pos < 0)) {
+               if (!unsigned_offsets(file))
+                       return retval;
+               if (count >= -pos) /* both values are in 0..LLONG_MAX */
+                       return -EOVERFLOW;
+       } else if (unlikely((loff_t) (pos + count) < 0)) {
+               if (!unsigned_offsets(file))
+                       return retval;
+       }
 
-       inode = file->f_dentry->d_inode;
-       if (inode->i_flock && MANDATORY_LOCK(inode)) {
-               int retval = locks_mandatory_area(
+       if (unlikely(inode->i_flock && mandatory_lock(inode))) {
+               retval = locks_mandatory_area(
                        read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
                        inode, file, pos, count);
                if (retval < 0)
                        return retval;
        }
+       retval = security_file_permission(file,
+                               read_write == READ ? MAY_READ : MAY_WRITE);
+       if (retval)
+               return retval;
        return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
-
-Einval:
-       return -EINVAL;
 }
 
 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
@@ -227,14 +321,21 @@ static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
 
 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 {
+       struct iovec iov = { .iov_base = buf, .iov_len = len };
        struct kiocb kiocb;
        ssize_t ret;
 
        init_sync_kiocb(&kiocb, filp);
        kiocb.ki_pos = *ppos;
-       while (-EIOCBRETRY ==
-               (ret = filp->f_op->aio_read(&kiocb, buf, len, kiocb.ki_pos)))
+       kiocb.ki_left = len;
+       kiocb.ki_nbytes = len;
+
+       for (;;) {
+               ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
+               if (ret != -EIOCBRETRY)
+                       break;
                wait_on_retry_sync_kiocb(&kiocb);
+       }
 
        if (-EIOCBQUEUED == ret)
                ret = wait_on_sync_kiocb(&kiocb);
@@ -258,18 +359,15 @@ ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
        ret = rw_verify_area(READ, file, pos, count);
        if (ret >= 0) {
                count = ret;
-               ret = security_file_permission (file, MAY_READ);
-               if (!ret) {
-                       if (file->f_op->read)
-                               ret = file->f_op->read(file, buf, count, pos);
-                       else
-                               ret = do_sync_read(file, buf, count, pos);
-                       if (ret > 0) {
-                               fsnotify_access(file->f_dentry);
-                               current->rchar += ret;
-                       }
-                       current->syscr++;
+               if (file->f_op->read)
+                       ret = file->f_op->read(file, buf, count, pos);
+               else
+                       ret = do_sync_read(file, buf, count, pos);
+               if (ret > 0) {
+                       fsnotify_access(file);
+                       add_rchar(current, ret);
                }
+               inc_syscr(current);
        }
 
        return ret;
@@ -279,14 +377,21 @@ EXPORT_SYMBOL(vfs_read);
 
 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 {
+       struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
        struct kiocb kiocb;
        ssize_t ret;
 
        init_sync_kiocb(&kiocb, filp);
        kiocb.ki_pos = *ppos;
-       while (-EIOCBRETRY ==
-              (ret = filp->f_op->aio_write(&kiocb, buf, len, kiocb.ki_pos)))
+       kiocb.ki_left = len;
+       kiocb.ki_nbytes = len;
+
+       for (;;) {
+               ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
+               if (ret != -EIOCBRETRY)
+                       break;
                wait_on_retry_sync_kiocb(&kiocb);
+       }
 
        if (-EIOCBQUEUED == ret)
                ret = wait_on_sync_kiocb(&kiocb);
@@ -310,18 +415,15 @@ ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_
        ret = rw_verify_area(WRITE, file, pos, count);
        if (ret >= 0) {
                count = ret;
-               ret = security_file_permission (file, MAY_WRITE);
-               if (!ret) {
-                       if (file->f_op->write)
-                               ret = file->f_op->write(file, buf, count, pos);
-                       else
-                               ret = do_sync_write(file, buf, count, pos);
-                       if (ret > 0) {
-                               fsnotify_modify(file->f_dentry);
-                               current->wchar += ret;
-                       }
-                       current->syscw++;
+               if (file->f_op->write)
+                       ret = file->f_op->write(file, buf, count, pos);
+               else
+                       ret = do_sync_write(file, buf, count, pos);
+               if (ret > 0) {
+                       fsnotify_modify(file);
+                       add_wchar(current, ret);
                }
+               inc_syscw(current);
        }
 
        return ret;
@@ -339,7 +441,7 @@ static inline void file_pos_write(struct file *file, loff_t pos)
        file->f_pos = pos;
 }
 
-asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count)
+SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
 {
        struct file *file;
        ssize_t ret = -EBADF;
@@ -355,9 +457,9 @@ asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count)
 
        return ret;
 }
-EXPORT_SYMBOL_GPL(sys_read);
 
-asmlinkage ssize_t sys_write(unsigned int fd, const char __user * buf, size_t count)
+SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
+               size_t, count)
 {
        struct file *file;
        ssize_t ret = -EBADF;
@@ -374,8 +476,8 @@ asmlinkage ssize_t sys_write(unsigned int fd, const char __user * buf, size_t co
        return ret;
 }
 
-asmlinkage ssize_t sys_pread64(unsigned int fd, char __user *buf,
-                            size_t count, loff_t pos)
+SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
+                       size_t count, loff_t pos)
 {
        struct file *file;
        ssize_t ret = -EBADF;
@@ -394,9 +496,17 @@ asmlinkage ssize_t sys_pread64(unsigned int fd, char __user *buf,
 
        return ret;
 }
+#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
+asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
+{
+       return SYSC_pread64((unsigned int) fd, (char __user *) buf,
+                           (size_t) count, pos);
+}
+SYSCALL_ALIAS(sys_pread64, SyS_pread64);
+#endif
 
-asmlinkage ssize_t sys_pwrite64(unsigned int fd, const char __user *buf,
-                             size_t count, loff_t pos)
+SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
+                        size_t count, loff_t pos)
 {
        struct file *file;
        ssize_t ret = -EBADF;
@@ -415,6 +525,14 @@ asmlinkage ssize_t sys_pwrite64(unsigned int fd, const char __user *buf,
 
        return ret;
 }
+#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
+asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
+{
+       return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
+                            (size_t) count, pos);
+}
+SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
+#endif
 
 /*
  * Reduce an iovec's length in-place.  Return the resulting number of segments
@@ -435,137 +553,192 @@ unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
        }
        return seg;
 }
-
 EXPORT_SYMBOL(iov_shorten);
 
+ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
+               unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
+{
+       struct kiocb kiocb;
+       ssize_t ret;
+
+       init_sync_kiocb(&kiocb, filp);
+       kiocb.ki_pos = *ppos;
+       kiocb.ki_left = len;
+       kiocb.ki_nbytes = len;
+
+       for (;;) {
+               ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
+               if (ret != -EIOCBRETRY)
+                       break;
+               wait_on_retry_sync_kiocb(&kiocb);
+       }
+
+       if (ret == -EIOCBQUEUED)
+               ret = wait_on_sync_kiocb(&kiocb);
+       *ppos = kiocb.ki_pos;
+       return ret;
+}
+
+/* Do it by hand, with file-ops */
+ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
+               unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
+{
+       struct iovec *vector = iov;
+       ssize_t ret = 0;
+
+       while (nr_segs > 0) {
+               void __user *base;
+               size_t len;
+               ssize_t nr;
+
+               base = vector->iov_base;
+               len = vector->iov_len;
+               vector++;
+               nr_segs--;
+
+               nr = fn(filp, base, len, ppos);
+
+               if (nr < 0) {
+                       if (!ret)
+                               ret = nr;
+                       break;
+               }
+               ret += nr;
+               if (nr != len)
+                       break;
+       }
+
+       return ret;
+}
+
 /* A write operation does a read from user space and vice versa */
 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
 
-static ssize_t do_readv_writev(int type, struct file *file,
-                              const struct iovec __user * uvector,
-                              unsigned long nr_segs, loff_t *pos)
+ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
+                             unsigned long nr_segs, unsigned long fast_segs,
+                             struct iovec *fast_pointer,
+                             struct iovec **ret_pointer)
 {
-       typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
-       typedef ssize_t (*iov_fn_t)(struct file *, const struct iovec *, unsigned long, loff_t *);
-
-       size_t tot_len;
-       struct iovec iovstack[UIO_FASTIOV];
-       struct iovec *iov=iovstack, *vector;
+       unsigned long seg;
        ssize_t ret;
-       int seg;
-       io_fn_t fn;
-       iov_fn_t fnv;
+       struct iovec *iov = fast_pointer;
 
        /*
         * SuS says "The readv() function *may* fail if the iovcnt argument
         * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
         * traditionally returned zero for zero segments, so...
         */
-       ret = 0;
-       if (nr_segs == 0)
+       if (nr_segs == 0) {
+               ret = 0;
                goto out;
+       }
 
        /*
         * First get the "struct iovec" from user memory and
         * verify all the pointers
         */
-       ret = -EINVAL;
-       if ((nr_segs > UIO_MAXIOV) || (nr_segs <= 0))
+       if (nr_segs > UIO_MAXIOV) {
+               ret = -EINVAL;
                goto out;
-       if (!file->f_op)
-               goto out;
-       if (nr_segs > UIO_FASTIOV) {
-               ret = -ENOMEM;
+       }
+       if (nr_segs > fast_segs) {
                iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
-               if (!iov)
+               if (iov == NULL) {
+                       ret = -ENOMEM;
                        goto out;
+               }
        }
-       ret = -EFAULT;
-       if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector)))
+       if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
+               ret = -EFAULT;
                goto out;
+       }
 
        /*
-        * Single unix specification:
-        * We should -EINVAL if an element length is not >= 0 and fitting an
-        * ssize_t.  The total length is fitting an ssize_t
+        * According to the Single Unix Specification we should return EINVAL
+        * if an element length is < 0 when cast to ssize_t or if the
+        * total length would overflow the ssize_t return value of the
+        * system call.
         *
-        * Be careful here because iov_len is a size_t not an ssize_t
+        * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
+        * overflow case.
         */
-       tot_len = 0;
-       ret = -EINVAL;
+       ret = 0;
        for (seg = 0; seg < nr_segs; seg++) {
                void __user *buf = iov[seg].iov_base;
                ssize_t len = (ssize_t)iov[seg].iov_len;
 
-               if (len < 0)    /* size_t not fitting an ssize_t .. */
+               /* see if we we're about to use an invalid len or if
+                * it's about to overflow ssize_t */
+               if (len < 0) {
+                       ret = -EINVAL;
                        goto out;
-               if (unlikely(!access_ok(vrfy_dir(type), buf, len)))
-                       goto Efault;
-               tot_len += len;
-               if ((ssize_t)tot_len < 0) /* maths overflow on the ssize_t */
+               }
+               if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
+                       ret = -EFAULT;
                        goto out;
+               }
+               if (len > MAX_RW_COUNT - ret) {
+                       len = MAX_RW_COUNT - ret;
+                       iov[seg].iov_len = len;
+               }
+               ret += len;
        }
-       if (tot_len == 0) {
-               ret = 0;
+out:
+       *ret_pointer = iov;
+       return ret;
+}
+
+static ssize_t do_readv_writev(int type, struct file *file,
+                              const struct iovec __user * uvector,
+                              unsigned long nr_segs, loff_t *pos)
+{
+       size_t tot_len;
+       struct iovec iovstack[UIO_FASTIOV];
+       struct iovec *iov = iovstack;
+       ssize_t ret;
+       io_fn_t fn;
+       iov_fn_t fnv;
+
+       if (!file->f_op) {
+               ret = -EINVAL;
                goto out;
        }
 
+       ret = rw_copy_check_uvector(type, uvector, nr_segs,
+                       ARRAY_SIZE(iovstack), iovstack, &iov);
+       if (ret <= 0)
+               goto out;
+
+       tot_len = ret;
        ret = rw_verify_area(type, file, pos, tot_len);
        if (ret < 0)
                goto out;
-       ret = security_file_permission(file, type == READ ? MAY_READ : MAY_WRITE);
-       if (ret)
-               goto out;
 
        fnv = NULL;
        if (type == READ) {
                fn = file->f_op->read;
-               fnv = file->f_op->readv;
+               fnv = file->f_op->aio_read;
        } else {
                fn = (io_fn_t)file->f_op->write;
-               fnv = file->f_op->writev;
-       }
-       if (fnv) {
-               ret = fnv(file, iov, nr_segs, pos);
-               goto out;
+               fnv = file->f_op->aio_write;
        }
 
-       /* Do it by hand, with file-ops */
-       ret = 0;
-       vector = iov;
-       while (nr_segs > 0) {
-               void __user * base;
-               size_t len;
-               ssize_t nr;
-
-               base = vector->iov_base;
-               len = vector->iov_len;
-               vector++;
-               nr_segs--;
-
-               nr = fn(file, base, len, pos);
+       if (fnv)
+               ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
+                                               pos, fnv);
+       else
+               ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
 
-               if (nr < 0) {
-                       if (!ret) ret = nr;
-                       break;
-               }
-               ret += nr;
-               if (nr != len)
-                       break;
-       }
 out:
        if (iov != iovstack)
                kfree(iov);
        if ((ret + (type == READ)) > 0) {
                if (type == READ)
-                       fsnotify_access(file->f_dentry);
+                       fsnotify_access(file);
                else
-                       fsnotify_modify(file->f_dentry);
+                       fsnotify_modify(file);
        }
        return ret;
-Efault:
-       ret = -EFAULT;
-       goto out;
 }
 
 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
@@ -573,7 +746,7 @@ ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
 {
        if (!(file->f_mode & FMODE_READ))
                return -EBADF;
-       if (!file->f_op || (!file->f_op->readv && !file->f_op->read))
+       if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
                return -EINVAL;
 
        return do_readv_writev(READ, file, vec, vlen, pos);
@@ -586,7 +759,7 @@ ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 {
        if (!(file->f_mode & FMODE_WRITE))
                return -EBADF;
-       if (!file->f_op || (!file->f_op->writev && !file->f_op->write))
+       if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
                return -EINVAL;
 
        return do_readv_writev(WRITE, file, vec, vlen, pos);
@@ -594,8 +767,8 @@ ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
 
 EXPORT_SYMBOL(vfs_writev);
 
-asmlinkage ssize_t
-sys_readv(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
+SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
+               unsigned long, vlen)
 {
        struct file *file;
        ssize_t ret = -EBADF;
@@ -610,13 +783,13 @@ sys_readv(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
        }
 
        if (ret > 0)
-               current->rchar += ret;
-       current->syscr++;
+               add_rchar(current, ret);
+       inc_syscr(current);
        return ret;
 }
 
-asmlinkage ssize_t
-sys_writev(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
+SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
+               unsigned long, vlen)
 {
        struct file *file;
        ssize_t ret = -EBADF;
@@ -631,8 +804,64 @@ sys_writev(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
        }
 
        if (ret > 0)
-               current->wchar += ret;
-       current->syscw++;
+               add_wchar(current, ret);
+       inc_syscw(current);
+       return ret;
+}
+
+static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
+{
+#define HALF_LONG_BITS (BITS_PER_LONG / 2)
+       return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
+}
+
+SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
+               unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
+{
+       loff_t pos = pos_from_hilo(pos_h, pos_l);
+       struct file *file;
+       ssize_t ret = -EBADF;
+       int fput_needed;
+
+       if (pos < 0)
+               return -EINVAL;
+
+       file = fget_light(fd, &fput_needed);
+       if (file) {
+               ret = -ESPIPE;
+               if (file->f_mode & FMODE_PREAD)
+                       ret = vfs_readv(file, vec, vlen, &pos);
+               fput_light(file, fput_needed);
+       }
+
+       if (ret > 0)
+               add_rchar(current, ret);
+       inc_syscr(current);
+       return ret;
+}
+
+SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
+               unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
+{
+       loff_t pos = pos_from_hilo(pos_h, pos_l);
+       struct file *file;
+       ssize_t ret = -EBADF;
+       int fput_needed;
+
+       if (pos < 0)
+               return -EINVAL;
+
+       file = fget_light(fd, &fput_needed);
+       if (file) {
+               ret = -ESPIPE;
+               if (file->f_mode & FMODE_PWRITE)
+                       ret = vfs_writev(file, vec, vlen, &pos);
+               fput_light(file, fput_needed);
+       }
+
+       if (ret > 0)
+               add_wchar(current, ret);
+       inc_syscw(current);
        return ret;
 }
 
@@ -643,7 +872,7 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
        struct inode * in_inode, * out_inode;
        loff_t pos;
        ssize_t retval;
-       int fput_needed_in, fput_needed_out;
+       int fput_needed_in, fput_needed_out, fl;
 
        /*
         * Get input file, and verify that it is ok..
@@ -654,12 +883,6 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
                goto out;
        if (!(in_file->f_mode & FMODE_READ))
                goto fput_in;
-       retval = -EINVAL;
-       in_inode = in_file->f_dentry->d_inode;
-       if (!in_inode)
-               goto fput_in;
-       if (!in_file->f_op || !in_file->f_op->sendfile)
-               goto fput_in;
        retval = -ESPIPE;
        if (!ppos)
                ppos = &in_file->f_pos;
@@ -671,10 +894,6 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
                goto fput_in;
        count = retval;
 
-       retval = security_file_permission (in_file, MAY_READ);
-       if (retval)
-               goto fput_in;
-
        /*
         * Get output file, and verify that it is ok..
         */
@@ -685,25 +904,17 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
        if (!(out_file->f_mode & FMODE_WRITE))
                goto fput_out;
        retval = -EINVAL;
-       if (!out_file->f_op || !out_file->f_op->sendpage)
-               goto fput_out;
-       out_inode = out_file->f_dentry->d_inode;
+       in_inode = in_file->f_path.dentry->d_inode;
+       out_inode = out_file->f_path.dentry->d_inode;
        retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
        if (retval < 0)
                goto fput_out;
        count = retval;
 
-       retval = security_file_permission (out_file, MAY_WRITE);
-       if (retval)
-               goto fput_out;
-
        if (!max)
                max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
 
        pos = *ppos;
-       retval = -EINVAL;
-       if (unlikely(pos < 0))
-               goto fput_out;
        if (unlikely(pos + count > max)) {
                retval = -EOVERFLOW;
                if (pos >= max)
@@ -711,15 +922,26 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
                count = max - pos;
        }
 
-       retval = in_file->f_op->sendfile(in_file, ppos, count, file_send_actor, out_file);
+       fl = 0;
+#if 0
+       /*
+        * We need to debate whether we can enable this or not. The
+        * man page documents EAGAIN return for the output at least,
+        * and the application is arguably buggy if it doesn't expect
+        * EAGAIN on a non-blocking file descriptor.
+        */
+       if (in_file->f_flags & O_NONBLOCK)
+               fl = SPLICE_F_NONBLOCK;
+#endif
+       retval = do_splice_direct(in_file, ppos, out_file, count, fl);
 
        if (retval > 0) {
-               current->rchar += retval;
-               current->wchar += retval;
+               add_rchar(current, retval);
+               add_wchar(current, retval);
        }
-       current->syscr++;
-       current->syscw++;
 
+       inc_syscr(current);
+       inc_syscw(current);
        if (*ppos > max)
                retval = -EOVERFLOW;
 
@@ -731,7 +953,7 @@ out:
        return retval;
 }
 
-asmlinkage ssize_t sys_sendfile(int out_fd, int in_fd, off_t __user *offset, size_t count)
+SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
 {
        loff_t pos;
        off_t off;
@@ -750,7 +972,7 @@ asmlinkage ssize_t sys_sendfile(int out_fd, int in_fd, off_t __user *offset, siz
        return do_sendfile(out_fd, in_fd, NULL, count, 0);
 }
 
-asmlinkage ssize_t sys_sendfile64(int out_fd, int in_fd, loff_t __user *offset, size_t count)
+SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
 {
        loff_t pos;
        ssize_t ret;