Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[pandora-kernel.git] / fs / read_write.c
index d4cb318..1f8dc37 100644 (file)
 #include <linux/module.h>
 #include <linux/syscalls.h>
 #include <linux/pagemap.h>
+#include "read_write.h"
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
 
 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,
 };
@@ -62,13 +64,13 @@ loff_t remote_llseek(struct file *file, loff_t offset, int origin)
        lock_kernel();
        switch (origin) {
                case 2:
-                       offset += i_size_read(file->f_dentry->d_inode);
+                       offset += i_size_read(file->f_path.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>=0 && offset<=file->f_path.dentry->d_inode->i_sb->s_maxbytes) {
                if (offset != file->f_pos) {
                        file->f_pos = offset;
                        file->f_version = 0;
@@ -93,7 +95,7 @@ loff_t default_llseek(struct file *file, loff_t offset, int origin)
        lock_kernel();
        switch (origin) {
                case 2:
-                       offset += i_size_read(file->f_dentry->d_inode);
+                       offset += i_size_read(file->f_path.dentry->d_inode);
                        break;
                case 1:
                        offset += file->f_pos;
@@ -195,13 +197,13 @@ int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count
        struct inode *inode;
        loff_t pos;
 
+       inode = file->f_path.dentry->d_inode;
        if (unlikely((ssize_t) count < 0))
                goto Einval;
        pos = *ppos;
        if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))
                goto Einval;
 
-       inode = file->f_dentry->d_inode;
        if (unlikely(inode->i_flock && MANDATORY_LOCK(inode))) {
                int retval = locks_mandatory_area(
                        read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
@@ -227,14 +229,20 @@ 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;
+
+       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);
@@ -265,10 +273,10 @@ ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
                        else
                                ret = do_sync_read(file, buf, count, pos);
                        if (ret > 0) {
-                               fsnotify_access(file->f_dentry);
-                               current->rchar += ret;
+                               fsnotify_access(file->f_path.dentry);
+                               add_rchar(current, ret);
                        }
-                       current->syscr++;
+                       inc_syscr(current);
                }
        }
 
@@ -279,14 +287,20 @@ 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;
+
+       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);
@@ -317,10 +331,10 @@ ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_
                        else
                                ret = do_sync_write(file, buf, count, pos);
                        if (ret > 0) {
-                               fsnotify_modify(file->f_dentry);
-                               current->wchar += ret;
+                               fsnotify_modify(file->f_path.dentry);
+                               add_wchar(current, ret);
                        }
-                       current->syscw++;
+                       inc_syscw(current);
                }
        }
 
@@ -436,80 +450,155 @@ unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
        return seg;
 }
 
-EXPORT_UNUSED_SYMBOL(iov_shorten);  /*  June 2006  */
+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)
 
+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)
+  {
+       unsigned long seg;
+       ssize_t ret;
+       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...
+        */
+       if (nr_segs == 0) {
+               ret = 0;
+               goto out;
+       }
+
+       /*
+        * First get the "struct iovec" from user memory and
+        * verify all the pointers
+        */
+       if (nr_segs > UIO_MAXIOV) {
+               ret = -EINVAL;
+               goto out;
+       }
+       if (nr_segs > fast_segs) {
+               iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
+               if (iov == NULL) {
+                       ret = -ENOMEM;
+                       goto out;
+               }
+       }
+       if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
+               ret = -EFAULT;
+               goto out;
+       }
+
+       /*
+        * 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.
+        */
+       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;
+
+               /* see if we we're about to use an invalid len or if
+                * it's about to overflow ssize_t */
+               if (len < 0 || (ret + len < ret)) {
+                       ret = -EINVAL;
+                       goto out;
+               }
+               if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
+                       ret = -EFAULT;
+                       goto out;
+               }
+
+               ret += len;
+       }
+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)
 {
-       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;
+       struct iovec *iov = iovstack;
        ssize_t ret;
-       int seg;
        io_fn_t fn;
        iov_fn_t fnv;
 
-       /*
-        * 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 (!file->f_op) {
+               ret = -EINVAL;
                goto out;
+       }
 
-       /*
-        * First get the "struct iovec" from user memory and
-        * verify all the pointers
-        */
-       ret = -EINVAL;
-       if (nr_segs > UIO_MAXIOV)
-               goto out;
-       if (!file->f_op)
-               goto out;
-       if (nr_segs > UIO_FASTIOV) {
-               ret = -ENOMEM;
-               iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
-               if (!iov)
-                       goto out;
-       }
-       ret = -EFAULT;
-       if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector)))
+       ret = rw_copy_check_uvector(type, uvector, nr_segs,
+                       ARRAY_SIZE(iovstack), iovstack, &iov);
+       if (ret <= 0)
                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
-        *
-        * Be careful here because iov_len is a size_t not an ssize_t
-        */
-       tot_len = 0;
-       ret = -EINVAL;
-       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 .. */
-                       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 */
-                       goto out;
-       }
-       if (tot_len == 0) {
-               ret = 0;
-               goto out;
-       }
-
+       tot_len = ret;
        ret = rw_verify_area(type, file, pos, tot_len);
        if (ret < 0)
                goto out;
@@ -520,52 +609,28 @@ static ssize_t do_readv_writev(int type, struct file *file,
        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;
+               fnv = file->f_op->aio_write;
        }
-       if (fnv) {
-               ret = fnv(file, iov, nr_segs, pos);
-               goto out;
-       }
-
-       /* 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->f_path.dentry);
                else
-                       fsnotify_modify(file->f_dentry);
+                       fsnotify_modify(file->f_path.dentry);
        }
        return ret;
-Efault:
-       ret = -EFAULT;
-       goto out;
 }
 
 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
@@ -573,7 +638,7 @@ 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 +651,7 @@ 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);
@@ -610,8 +675,8 @@ 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;
 }
 
@@ -631,8 +696,8 @@ 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;
 }
 
@@ -655,7 +720,7 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
        if (!(in_file->f_mode & FMODE_READ))
                goto fput_in;
        retval = -EINVAL;
-       in_inode = in_file->f_dentry->d_inode;
+       in_inode = in_file->f_path.dentry->d_inode;
        if (!in_inode)
                goto fput_in;
        if (!in_file->f_op || !in_file->f_op->sendfile)
@@ -687,7 +752,7 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
        retval = -EINVAL;
        if (!out_file->f_op || !out_file->f_op->sendpage)
                goto fput_out;
-       out_inode = out_file->f_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;
@@ -714,12 +779,12 @@ static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
        retval = in_file->f_op->sendfile(in_file, ppos, count, file_send_actor, out_file);
 
        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;