vfs_ops.c

Go to the documentation of this file.

/*
 * Copyright (c) 2008 Jakub Jermar
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * - Redistributions of source code must retain the above copyright
 *   notice, this list of conditions and the following disclaimer.
 * - Redistributions in binary form must reproduce the above copyright
 *   notice, this list of conditions and the following disclaimer in the
 *   documentation and/or other materials provided with the distribution.
 * - The name of the author may not be used to endorse or promote products
 *   derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include "vfs.h"
#include <macros.h>
#include <stdint.h>
#include <async.h>
#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <str.h>
#include <bool.h>
#include <fibril_synch.h>
#include <adt/list.h>
#include <unistd.h>
#include <ctype.h>
#include <fcntl.h>
#include <assert.h>
#include <vfs/canonify.h>

/* Forward declarations of static functions. */
static int vfs_truncate_internal(fs_handle_t, devmap_handle_t, fs_index_t,
    aoff64_t);

FIBRIL_RWLOCK_INITIALIZE(namespace_rwlock);

vfs_pair_t rootfs = {
        .fs_handle = 0,
        .devmap_handle = 0
};

static void vfs_mount_internal(ipc_callid_t rid, devmap_handle_t devmap_handle,
    fs_handle_t fs_handle, char *mp, char *opts)
{
        vfs_lookup_res_t mp_res;
        vfs_lookup_res_t mr_res;
        vfs_node_t *mp_node = NULL;
        vfs_node_t *mr_node;
        fs_index_t rindex;
        size_t rsize;
        unsigned rlnkcnt;
        sysarg_t rc;
        int phone;
        aid_t msg;
        ipc_call_t answer;
        
        /* Resolve the path to the mountpoint. */
        fibril_rwlock_write_lock(&namespace_rwlock);
        if (rootfs.fs_handle) {
                /* We already have the root FS. */
                if (str_cmp(mp, "/") == 0) {
                        /* Trying to mount root FS over root FS */
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        async_answer_0(rid, EBUSY);
                        return;
                }
                
                rc = vfs_lookup_internal(mp, L_MP, &mp_res, NULL);
                if (rc != EOK) {
                        /* The lookup failed for some reason. */
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        async_answer_0(rid, rc);
                        return;
                }
                
                mp_node = vfs_node_get(&mp_res);
                if (!mp_node) {
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        async_answer_0(rid, ENOMEM);
                        return;
                }
                
                /*
                 * Now we hold a reference to mp_node.
                 * It will be dropped upon the corresponding VFS_IN_UNMOUNT.
                 * This prevents the mount point from being deleted.
                 */
        } else {
                /* We still don't have the root file system mounted. */
                if (str_cmp(mp, "/") == 0) {
                        /*
                         * For this simple, but important case,
                         * we are almost done.
                         */
                        
                        /* Tell the mountee that it is being mounted. */
                        phone = vfs_grab_phone(fs_handle);
                        msg = async_send_1(phone, VFS_OUT_MOUNTED,
                            (sysarg_t) devmap_handle, &answer);
                        /* send the mount options */
                        rc = async_data_write_start(phone, (void *)opts,
                            str_size(opts));
                        if (rc != EOK) {
                                async_wait_for(msg, NULL);
                                vfs_release_phone(fs_handle, phone);
                                fibril_rwlock_write_unlock(&namespace_rwlock);
                                async_answer_0(rid, rc);
                                return;
                        }
                        async_wait_for(msg, &rc);
                        vfs_release_phone(fs_handle, phone);
                        
                        if (rc != EOK) {
                                fibril_rwlock_write_unlock(&namespace_rwlock);
                                async_answer_0(rid, rc);
                                return;
                        }

                        rindex = (fs_index_t) IPC_GET_ARG1(answer);
                        rsize = (size_t) IPC_GET_ARG2(answer);
                        rlnkcnt = (unsigned) IPC_GET_ARG3(answer);
                        
                        mr_res.triplet.fs_handle = fs_handle;
                        mr_res.triplet.devmap_handle = devmap_handle;
                        mr_res.triplet.index = rindex;
                        mr_res.size = rsize;
                        mr_res.lnkcnt = rlnkcnt;
                        mr_res.type = VFS_NODE_DIRECTORY;
                        
                        rootfs.fs_handle = fs_handle;
                        rootfs.devmap_handle = devmap_handle;
                        
                        /* Add reference to the mounted root. */
                        mr_node = vfs_node_get(&mr_res); 
                        assert(mr_node);
                        
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        async_answer_0(rid, rc);
                        return;
                } else {
                        /*
                         * We can't resolve this without the root filesystem
                         * being mounted first.
                         */
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        async_answer_0(rid, ENOENT);
                        return;
                }
        }
        
        /*
         * At this point, we have all necessary pieces: file system and device
         * handles, and we know the mount point VFS node.
         */
        
        int mountee_phone = vfs_grab_phone(fs_handle);
        assert(mountee_phone >= 0);

        phone = vfs_grab_phone(mp_res.triplet.fs_handle);
        msg = async_send_4(phone, VFS_OUT_MOUNT,
            (sysarg_t) mp_res.triplet.devmap_handle,
            (sysarg_t) mp_res.triplet.index,
            (sysarg_t) fs_handle,
            (sysarg_t) devmap_handle, &answer);
        
        /* send connection */
        rc = async_req_1_0(phone, IPC_M_CONNECTION_CLONE, mountee_phone);
        if (rc != EOK) {
                async_wait_for(msg, NULL);
                vfs_release_phone(fs_handle, mountee_phone);
                vfs_release_phone(mp_res.triplet.fs_handle, phone);
                /* Mount failed, drop reference to mp_node. */
                if (mp_node)
                        vfs_node_put(mp_node);
                async_answer_0(rid, rc);
                fibril_rwlock_write_unlock(&namespace_rwlock);
                return;
        }

        vfs_release_phone(fs_handle, mountee_phone);
        
        /* send the mount options */
        rc = async_data_write_start(phone, (void *)opts, str_size(opts));
        if (rc != EOK) {
                async_wait_for(msg, NULL);
                vfs_release_phone(mp_res.triplet.fs_handle, phone);
                /* Mount failed, drop reference to mp_node. */
                if (mp_node)
                        vfs_node_put(mp_node);
                fibril_rwlock_write_unlock(&namespace_rwlock);
                async_answer_0(rid, rc);
                return;
        }
        async_wait_for(msg, &rc);
        vfs_release_phone(mp_res.triplet.fs_handle, phone);
        
        if (rc == EOK) {
                rindex = (fs_index_t) IPC_GET_ARG1(answer);
                rsize = (size_t) IPC_GET_ARG2(answer);
                rlnkcnt = (unsigned) IPC_GET_ARG3(answer);
        
                mr_res.triplet.fs_handle = fs_handle;
                mr_res.triplet.devmap_handle = devmap_handle;
                mr_res.triplet.index = rindex;
                mr_res.size = rsize;
                mr_res.lnkcnt = rlnkcnt;
                mr_res.type = VFS_NODE_DIRECTORY;
        
                /* Add reference to the mounted root. */
                mr_node = vfs_node_get(&mr_res); 
                assert(mr_node);
        } else {
                /* Mount failed, drop reference to mp_node. */
                if (mp_node)
                        vfs_node_put(mp_node);
        }

        async_answer_0(rid, rc);
        fibril_rwlock_write_unlock(&namespace_rwlock);
}

void vfs_mount(ipc_callid_t rid, ipc_call_t *request)
{
        devmap_handle_t devmap_handle;

        /*
         * We expect the library to do the device-name to device-handle
         * translation for us, thus the device handle will arrive as ARG1
         * in the request.
         */
        devmap_handle = (devmap_handle_t) IPC_GET_ARG1(*request);
        
        /*
         * Mount flags are passed as ARG2.
         */
        unsigned int flags = (unsigned int) IPC_GET_ARG2(*request);
        
        /*
         * For now, don't make use of ARG3, but it can be used to
         * carry mount options in the future.
         */
        
        /* We want the client to send us the mount point. */
        char *mp;
        int rc = async_data_write_accept((void **) &mp, true, 0, MAX_PATH_LEN,
            0, NULL);
        if (rc != EOK) {
                async_answer_0(rid, rc);
                return;
        }
        
        /* Now we expect to receive the mount options. */
        char *opts;
        rc = async_data_write_accept((void **) &opts, true, 0, MAX_MNTOPTS_LEN,
            0, NULL);
        if (rc != EOK) {
                free(mp);
                async_answer_0(rid, rc);
                return;
        }
        
        /*
         * Now, we expect the client to send us data with the name of the file
         * system.
         */
        char *fs_name;
        rc = async_data_write_accept((void **) &fs_name, true, 0,
            FS_NAME_MAXLEN, 0, NULL);
        if (rc != EOK) {
                free(mp);
                free(opts);
                async_answer_0(rid, rc);
                return;
        }
        
        /*
         * Wait for IPC_M_PING so that we can return an error if we don't know
         * fs_name.
         */
        ipc_call_t data;
        ipc_callid_t callid = async_get_call(&data);
        if (IPC_GET_IMETHOD(data) != IPC_M_PING) {
                async_answer_0(callid, ENOTSUP);
                async_answer_0(rid, ENOTSUP);
                free(mp);
                free(opts);
                free(fs_name);
                return;
        }

        /*
         * Check if we know a file system with the same name as is in fs_name.
         * This will also give us its file system handle.
         */
        fibril_mutex_lock(&fs_head_lock);
        fs_handle_t fs_handle;
recheck:
        fs_handle = fs_name_to_handle(fs_name, false);
        if (!fs_handle) {
                if (flags & IPC_FLAG_BLOCKING) {
                        fibril_condvar_wait(&fs_head_cv, &fs_head_lock);
                        goto recheck;
                }
                
                fibril_mutex_unlock(&fs_head_lock);
                async_answer_0(callid, ENOENT);
                async_answer_0(rid, ENOENT);
                free(mp);
                free(fs_name);
                free(opts);
                return;
        }
        fibril_mutex_unlock(&fs_head_lock);
        
        /* Acknowledge that we know fs_name. */
        async_answer_0(callid, EOK);
        
        /* Do the mount */
        vfs_mount_internal(rid, devmap_handle, fs_handle, mp, opts);
        free(mp);
        free(fs_name);
        free(opts);
}

void vfs_unmount(ipc_callid_t rid, ipc_call_t *request)
{
        int rc;
        char *mp;
        vfs_lookup_res_t mp_res;
        vfs_lookup_res_t mr_res;
        vfs_node_t *mr_node;
        int phone;

        /*
         * Receive the mount point path.
         */
        rc = async_data_write_accept((void **) &mp, true, 0, MAX_PATH_LEN,
            0, NULL);
        if (rc != EOK)
                async_answer_0(rid, rc);

        /*
         * Taking the namespace lock will do two things for us. First, it will
         * prevent races with other lookup operations. Second, it will stop new
         * references to already existing VFS nodes and creation of new VFS
         * nodes. This is because new references are added as a result of some
         * lookup operation or at least of some operation which is protected by
         * the namespace lock.
         */
        fibril_rwlock_write_lock(&namespace_rwlock);
        
        /*
         * Lookup the mounted root and instantiate it.
         */
        rc = vfs_lookup_internal(mp, L_ROOT, &mr_res, NULL);
        if (rc != EOK) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                free(mp);
                async_answer_0(rid, rc);
                return;
        }
        mr_node = vfs_node_get(&mr_res);
        if (!mr_node) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                free(mp);
                async_answer_0(rid, ENOMEM);
                return;
        }

        /*
         * Count the total number of references for the mounted file system. We
         * are expecting at least two. One which we got above and one which we
         * got when the file system was mounted. If we find more, it means that
         * the file system cannot be gracefully unmounted at the moment because
         * someone is working with it.
         */
        if (vfs_nodes_refcount_sum_get(mr_node->fs_handle,
            mr_node->devmap_handle) != 2) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                vfs_node_put(mr_node);
                free(mp);
                async_answer_0(rid, EBUSY);
                return;
        }

        if (str_cmp(mp, "/") == 0) {

                /*
                 * Unmounting the root file system.
                 *
                 * In this case, there is no mount point node and we send
                 * VFS_OUT_UNMOUNTED directly to the mounted file system.
                 */

                free(mp);
                phone = vfs_grab_phone(mr_node->fs_handle);
                rc = async_req_1_0(phone, VFS_OUT_UNMOUNTED,
                    mr_node->devmap_handle);
                vfs_release_phone(mr_node->fs_handle, phone);
                if (rc != EOK) {
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        vfs_node_put(mr_node);
                        async_answer_0(rid, rc);
                        return;
                }
                rootfs.fs_handle = 0;
                rootfs.devmap_handle = 0;
        } else {

                /*
                 * Unmounting a non-root file system.
                 *
                 * We have a regular mount point node representing the parent
                 * file system, so we delegate the operation to it.
                 */

                rc = vfs_lookup_internal(mp, L_MP, &mp_res, NULL);
                free(mp);
                if (rc != EOK) {
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        vfs_node_put(mr_node);
                        async_answer_0(rid, rc);
                        return;
                }
                vfs_node_t *mp_node = vfs_node_get(&mp_res);
                if (!mp_node) {
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        vfs_node_put(mr_node);
                        async_answer_0(rid, ENOMEM);
                        return;
                }

                phone = vfs_grab_phone(mp_node->fs_handle);
                rc = async_req_2_0(phone, VFS_OUT_UNMOUNT,
                    mp_node->devmap_handle, mp_node->index);
                vfs_release_phone(mp_node->fs_handle, phone);
                if (rc != EOK) {
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        vfs_node_put(mp_node);
                        vfs_node_put(mr_node);
                        async_answer_0(rid, rc);
                        return;
                }

                /* Drop the reference we got above. */
                vfs_node_put(mp_node);
                /* Drop the reference from when the file system was mounted. */
                vfs_node_put(mp_node);
        }


        /*
         * All went well, the mounted file system was successfully unmounted.
         * The only thing left is to forget the unmounted root VFS node.
         */
        vfs_node_forget(mr_node);

        fibril_rwlock_write_unlock(&namespace_rwlock);
        async_answer_0(rid, EOK);
}

void vfs_open(ipc_callid_t rid, ipc_call_t *request)
{
        /*
         * The POSIX interface is open(path, oflag, mode).
         * We can receive oflags and mode along with the VFS_IN_OPEN call;
         * the path will need to arrive in another call.
         *
         * We also receive one private, non-POSIX set of flags called lflag
         * used to pass information to vfs_lookup_internal().
         */
        int lflag = IPC_GET_ARG1(*request);
        int oflag = IPC_GET_ARG2(*request);
        int mode = IPC_GET_ARG3(*request);

        /* Ignore mode for now. */
        (void) mode;
        
        /*
         * Make sure that we are called with exactly one of L_FILE and
         * L_DIRECTORY. Make sure that the user does not pass L_OPEN,
         * L_ROOT or L_MP.
         */
        if (((lflag & (L_FILE | L_DIRECTORY)) == 0) ||
            ((lflag & (L_FILE | L_DIRECTORY)) == (L_FILE | L_DIRECTORY)) ||
            (lflag & (L_OPEN | L_ROOT | L_MP))) {
                async_answer_0(rid, EINVAL);
                return;
        }
        
        if (oflag & O_CREAT)
                lflag |= L_CREATE;
        if (oflag & O_EXCL)
                lflag |= L_EXCLUSIVE;
        
        char *path;
        int rc = async_data_write_accept((void **) &path, true, 0, 0, 0, NULL);
        if (rc != EOK) {
                async_answer_0(rid, rc);
                return;
        }
        
        /*
         * Avoid the race condition in which the file can be deleted before we
         * find/create-and-lock the VFS node corresponding to the looked-up
         * triplet.
         */
        if (lflag & L_CREATE)
                fibril_rwlock_write_lock(&namespace_rwlock);
        else
                fibril_rwlock_read_lock(&namespace_rwlock);
        
        /* The path is now populated and we can call vfs_lookup_internal(). */
        vfs_lookup_res_t lr;
        rc = vfs_lookup_internal(path, lflag | L_OPEN, &lr, NULL);
        if (rc != EOK) {
                if (lflag & L_CREATE)
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                else
                        fibril_rwlock_read_unlock(&namespace_rwlock);
                async_answer_0(rid, rc);
                free(path);
                return;
        }
        
        /* Path is no longer needed. */
        free(path);
        
        vfs_node_t *node = vfs_node_get(&lr);
        if (lflag & L_CREATE)
                fibril_rwlock_write_unlock(&namespace_rwlock);
        else
                fibril_rwlock_read_unlock(&namespace_rwlock);
        
        /* Truncate the file if requested and if necessary. */
        if (oflag & O_TRUNC) {
                fibril_rwlock_write_lock(&node->contents_rwlock);
                if (node->size) {
                        rc = vfs_truncate_internal(node->fs_handle,
                            node->devmap_handle, node->index, 0);
                        if (rc) {
                                fibril_rwlock_write_unlock(&node->contents_rwlock);
                                vfs_node_put(node);
                                async_answer_0(rid, rc);
                                return;
                        }
                        node->size = 0;
                }
                fibril_rwlock_write_unlock(&node->contents_rwlock);
        }
        
        /*
         * Get ourselves a file descriptor and the corresponding vfs_file_t
         * structure.
         */
        int fd = vfs_fd_alloc((oflag & O_DESC) != 0);
        if (fd < 0) {
                vfs_node_put(node);
                async_answer_0(rid, fd);
                return;
        }
        vfs_file_t *file = vfs_file_get(fd);
        assert(file);
        file->node = node;
        if (oflag & O_APPEND)
                file->append = true;
        
        /*
         * The following increase in reference count is for the fact that the
         * file is being opened and that a file structure is pointing to it.
         * It is necessary so that the file will not disappear when
         * vfs_node_put() is called. The reference will be dropped by the
         * respective VFS_IN_CLOSE.
         */
        vfs_node_addref(node);
        vfs_node_put(node);
        vfs_file_put(file);
        
        /* Success! Return the new file descriptor to the client. */
        async_answer_1(rid, EOK, fd);
}

void vfs_open_node(ipc_callid_t rid, ipc_call_t *request)
{
        // FIXME: check for sanity of the supplied fs, dev and index
        
        /*
         * The interface is open_node(fs, dev, index, oflag).
         */
        vfs_lookup_res_t lr;
        
        lr.triplet.fs_handle = IPC_GET_ARG1(*request);
        lr.triplet.devmap_handle = IPC_GET_ARG2(*request);
        lr.triplet.index = IPC_GET_ARG3(*request);
        int oflag = IPC_GET_ARG4(*request);
        
        fibril_rwlock_read_lock(&namespace_rwlock);
        
        int rc = vfs_open_node_internal(&lr);
        if (rc != EOK) {
                fibril_rwlock_read_unlock(&namespace_rwlock);
                async_answer_0(rid, rc);
                return;
        }
        
        vfs_node_t *node = vfs_node_get(&lr);
        fibril_rwlock_read_unlock(&namespace_rwlock);
        
        /* Truncate the file if requested and if necessary. */
        if (oflag & O_TRUNC) {
                fibril_rwlock_write_lock(&node->contents_rwlock);
                if (node->size) {
                        rc = vfs_truncate_internal(node->fs_handle,
                            node->devmap_handle, node->index, 0);
                        if (rc) {
                                fibril_rwlock_write_unlock(&node->contents_rwlock);
                                vfs_node_put(node);
                                async_answer_0(rid, rc);
                                return;
                        }
                        node->size = 0;
                }
                fibril_rwlock_write_unlock(&node->contents_rwlock);
        }
        
        /*
         * Get ourselves a file descriptor and the corresponding vfs_file_t
         * structure.
         */
        int fd = vfs_fd_alloc((oflag & O_DESC) != 0);
        if (fd < 0) {
                vfs_node_put(node);
                async_answer_0(rid, fd);
                return;
        }
        vfs_file_t *file = vfs_file_get(fd);
        file->node = node;
        if (oflag & O_APPEND)
                file->append = true;
        
        /*
         * The following increase in reference count is for the fact that the
         * file is being opened and that a file structure is pointing to it.
         * It is necessary so that the file will not disappear when
         * vfs_node_put() is called. The reference will be dropped by the
         * respective VFS_IN_CLOSE.
         */
        vfs_node_addref(node);
        vfs_node_put(node);
        vfs_file_put(file);
        
        /* Success! Return the new file descriptor to the client. */
        async_answer_1(rid, EOK, fd);
}

void vfs_sync(ipc_callid_t rid, ipc_call_t *request)
{
        int fd = IPC_GET_ARG1(*request);
        
        /* Lookup the file structure corresponding to the file descriptor. */
        vfs_file_t *file = vfs_file_get(fd);
        if (!file) {
                async_answer_0(rid, ENOENT);
                return;
        }
        
        /*
         * Lock the open file structure so that no other thread can manipulate
         * the same open file at a time.
         */
        fibril_mutex_lock(&file->lock);
        int fs_phone = vfs_grab_phone(file->node->fs_handle);
        
        /* Make a VFS_OUT_SYMC request at the destination FS server. */
        aid_t msg;
        ipc_call_t answer;
        msg = async_send_2(fs_phone, VFS_OUT_SYNC, file->node->devmap_handle,
            file->node->index, &answer);

        /* Wait for reply from the FS server. */
        sysarg_t rc;
        async_wait_for(msg, &rc);
        
        vfs_release_phone(file->node->fs_handle, fs_phone);
        fibril_mutex_unlock(&file->lock);

        vfs_file_put(file);
        async_answer_0(rid, rc);
}

void vfs_close(ipc_callid_t rid, ipc_call_t *request)
{
        int fd = IPC_GET_ARG1(*request);
        int ret;
        
        ret = vfs_fd_free(fd);
        async_answer_0(rid, ret);
}

static void vfs_rdwr(ipc_callid_t rid, ipc_call_t *request, bool read)
{
        vfs_info_t *vi;

        /*
         * The following code strongly depends on the fact that the files data
         * structure can be only accessed by a single fibril and all file
         * operations are serialized (i.e. the reads and writes cannot
         * interleave and a file cannot be closed while it is being read).
         *
         * Additional synchronization needs to be added once the table of
         * open files supports parallel access!
         */

        int fd = IPC_GET_ARG1(*request);
        
        /* Lookup the file structure corresponding to the file descriptor. */
        vfs_file_t *file = vfs_file_get(fd);
        if (!file) {
                async_answer_0(rid, ENOENT);
                return;
        }
        
        /*
         * Lock the open file structure so that no other thread can manipulate
         * the same open file at a time.
         */
        fibril_mutex_lock(&file->lock);

        vi = fs_handle_to_info(file->node->fs_handle);
        assert(vi);

        /*
         * Lock the file's node so that no other client can read/write to it at
         * the same time unless the FS supports concurrent reads/writes and its
         * write implementation does not modify the file size.
         */
        if (read || (vi->concurrent_read_write && vi->write_retains_size))
                fibril_rwlock_read_lock(&file->node->contents_rwlock);
        else
                fibril_rwlock_write_lock(&file->node->contents_rwlock);

        if (file->node->type == VFS_NODE_DIRECTORY) {
                /*
                 * Make sure that no one is modifying the namespace
                 * while we are in readdir().
                 */
                assert(read);
                fibril_rwlock_read_lock(&namespace_rwlock);
        }
        
        int fs_phone = vfs_grab_phone(file->node->fs_handle);
        
        /*
         * Make a VFS_READ/VFS_WRITE request at the destination FS server
         * and forward the IPC_M_DATA_READ/IPC_M_DATA_WRITE request to the
         * destination FS server. The call will be routed as if sent by
         * ourselves. Note that call arguments are immutable in this case so we
         * don't have to bother.
         */
        sysarg_t rc;
        ipc_call_t answer;
        if (read) {
                rc = async_data_read_forward_3_1(fs_phone, VFS_OUT_READ,
                    file->node->devmap_handle, file->node->index, file->pos,
                    &answer);
        } else {
                if (file->append)
                        file->pos = file->node->size;
                
                rc = async_data_write_forward_3_1(fs_phone, VFS_OUT_WRITE,
                    file->node->devmap_handle, file->node->index, file->pos,
                    &answer);
        }
        
        vfs_release_phone(file->node->fs_handle, fs_phone);
        
        size_t bytes = IPC_GET_ARG1(answer);
        
        if (file->node->type == VFS_NODE_DIRECTORY)
                fibril_rwlock_read_unlock(&namespace_rwlock);
        
        /* Unlock the VFS node. */
        if (read || (vi->concurrent_read_write && vi->write_retains_size))
                fibril_rwlock_read_unlock(&file->node->contents_rwlock);
        else {
                /* Update the cached version of node's size. */
                if (rc == EOK)
                        file->node->size = IPC_GET_ARG2(answer); 
                fibril_rwlock_write_unlock(&file->node->contents_rwlock);
        }
        
        /* Update the position pointer and unlock the open file. */
        if (rc == EOK)
                file->pos += bytes;
        fibril_mutex_unlock(&file->lock);
        vfs_file_put(file);     

        /*
         * FS server's reply is the final result of the whole operation we
         * return to the client.
         */
        async_answer_1(rid, rc, bytes);
}

void vfs_read(ipc_callid_t rid, ipc_call_t *request)
{
        vfs_rdwr(rid, request, true);
}

void vfs_write(ipc_callid_t rid, ipc_call_t *request)
{
        vfs_rdwr(rid, request, false);
}

void vfs_seek(ipc_callid_t rid, ipc_call_t *request)
{
        int fd = (int) IPC_GET_ARG1(*request);
        off64_t off = (off64_t) MERGE_LOUP32(IPC_GET_ARG2(*request),
            IPC_GET_ARG3(*request));
        int whence = (int) IPC_GET_ARG4(*request);
        
        /* Lookup the file structure corresponding to the file descriptor. */
        vfs_file_t *file = vfs_file_get(fd);
        if (!file) {
                async_answer_0(rid, ENOENT);
                return;
        }
        
        fibril_mutex_lock(&file->lock);
        
        off64_t newoff;
        switch (whence) {
        case SEEK_SET:
                if (off >= 0) {
                        file->pos = (aoff64_t) off;
                        fibril_mutex_unlock(&file->lock);
                        vfs_file_put(file);
                        async_answer_1(rid, EOK, off);
                        return;
                }
                break;
        case SEEK_CUR:
                if ((off >= 0) && (file->pos + off < file->pos)) {
                        fibril_mutex_unlock(&file->lock);
                        vfs_file_put(file);
                        async_answer_0(rid, EOVERFLOW);
                        return;
                }
                
                if ((off < 0) && (file->pos < (aoff64_t) -off)) {
                        fibril_mutex_unlock(&file->lock);
                        vfs_file_put(file);
                        async_answer_0(rid, EOVERFLOW);
                        return;
                }
                
                file->pos += off;
                newoff = (file->pos > OFF64_MAX) ?  OFF64_MAX : file->pos;
                
                fibril_mutex_unlock(&file->lock);
                vfs_file_put(file);
                async_answer_2(rid, EOK, LOWER32(newoff),
                    UPPER32(newoff));
                return;
        case SEEK_END:
                fibril_rwlock_read_lock(&file->node->contents_rwlock);
                aoff64_t size = file->node->size;
                
                if ((off >= 0) && (size + off < size)) {
                        fibril_rwlock_read_unlock(&file->node->contents_rwlock);
                        fibril_mutex_unlock(&file->lock);
                        vfs_file_put(file);
                        async_answer_0(rid, EOVERFLOW);
                        return;
                }
                
                if ((off < 0) && (size < (aoff64_t) -off)) {
                        fibril_rwlock_read_unlock(&file->node->contents_rwlock);
                        fibril_mutex_unlock(&file->lock);
                        vfs_file_put(file);
                        async_answer_0(rid, EOVERFLOW);
                        return;
                }
                
                file->pos = size + off;
                newoff = (file->pos > OFF64_MAX) ?  OFF64_MAX : file->pos;
                
                fibril_rwlock_read_unlock(&file->node->contents_rwlock);
                fibril_mutex_unlock(&file->lock);
                vfs_file_put(file);
                async_answer_2(rid, EOK, LOWER32(newoff), UPPER32(newoff));
                return;
        }
        
        fibril_mutex_unlock(&file->lock);
        vfs_file_put(file);
        async_answer_0(rid, EINVAL);
}

int vfs_truncate_internal(fs_handle_t fs_handle, devmap_handle_t devmap_handle,
    fs_index_t index, aoff64_t size)
{
        sysarg_t rc;
        int fs_phone;
        
        fs_phone = vfs_grab_phone(fs_handle);
        rc = async_req_4_0(fs_phone, VFS_OUT_TRUNCATE, (sysarg_t) devmap_handle,
            (sysarg_t) index, LOWER32(size), UPPER32(size));
        vfs_release_phone(fs_handle, fs_phone);
        return (int)rc;
}

void vfs_truncate(ipc_callid_t rid, ipc_call_t *request)
{
        int fd = IPC_GET_ARG1(*request);
        aoff64_t size = (aoff64_t) MERGE_LOUP32(IPC_GET_ARG2(*request),
            IPC_GET_ARG3(*request));
        int rc;

        vfs_file_t *file = vfs_file_get(fd);
        if (!file) {
                async_answer_0(rid, ENOENT);
                return;
        }
        fibril_mutex_lock(&file->lock);

        fibril_rwlock_write_lock(&file->node->contents_rwlock);
        rc = vfs_truncate_internal(file->node->fs_handle,
            file->node->devmap_handle, file->node->index, size);
        if (rc == EOK)
                file->node->size = size;
        fibril_rwlock_write_unlock(&file->node->contents_rwlock);

        fibril_mutex_unlock(&file->lock);
        vfs_file_put(file);
        async_answer_0(rid, (sysarg_t)rc);
}

void vfs_fstat(ipc_callid_t rid, ipc_call_t *request)
{
        int fd = IPC_GET_ARG1(*request);
        sysarg_t rc;

        vfs_file_t *file = vfs_file_get(fd);
        if (!file) {
                async_answer_0(rid, ENOENT);
                return;
        }

        ipc_callid_t callid;
        if (!async_data_read_receive(&callid, NULL)) {
                vfs_file_put(file);
                async_answer_0(callid, EINVAL);
                async_answer_0(rid, EINVAL);
                return;
        }

        fibril_mutex_lock(&file->lock);

        int fs_phone = vfs_grab_phone(file->node->fs_handle);
        
        aid_t msg;
        msg = async_send_3(fs_phone, VFS_OUT_STAT, file->node->devmap_handle,
            file->node->index, true, NULL);
        async_forward_fast(callid, fs_phone, 0, 0, 0, IPC_FF_ROUTE_FROM_ME);
        async_wait_for(msg, &rc);
        vfs_release_phone(file->node->fs_handle, fs_phone);

        fibril_mutex_unlock(&file->lock);
        vfs_file_put(file);
        async_answer_0(rid, rc);
}

void vfs_stat(ipc_callid_t rid, ipc_call_t *request)
{
        char *path;
        int rc = async_data_write_accept((void **) &path, true, 0, 0, 0, NULL);
        if (rc != EOK) {
                async_answer_0(rid, rc);
                return;
        }
        
        ipc_callid_t callid;
        if (!async_data_read_receive(&callid, NULL)) {
                free(path);
                async_answer_0(callid, EINVAL);
                async_answer_0(rid, EINVAL);
                return;
        }

        vfs_lookup_res_t lr;
        fibril_rwlock_read_lock(&namespace_rwlock);
        rc = vfs_lookup_internal(path, L_NONE, &lr, NULL);
        free(path);
        if (rc != EOK) {
                fibril_rwlock_read_unlock(&namespace_rwlock);
                async_answer_0(callid, rc);
                async_answer_0(rid, rc);
                return;
        }
        vfs_node_t *node = vfs_node_get(&lr);
        if (!node) {
                fibril_rwlock_read_unlock(&namespace_rwlock);
                async_answer_0(callid, ENOMEM);
                async_answer_0(rid, ENOMEM);
                return;
        }

        fibril_rwlock_read_unlock(&namespace_rwlock);

        int fs_phone = vfs_grab_phone(node->fs_handle);
        aid_t msg;
        msg = async_send_3(fs_phone, VFS_OUT_STAT, node->devmap_handle,
            node->index, false, NULL);
        async_forward_fast(callid, fs_phone, 0, 0, 0, IPC_FF_ROUTE_FROM_ME);
        
        sysarg_t rv;
        async_wait_for(msg, &rv);
        vfs_release_phone(node->fs_handle, fs_phone);

        async_answer_0(rid, rv);

        vfs_node_put(node);
}

void vfs_mkdir(ipc_callid_t rid, ipc_call_t *request)
{
        int mode = IPC_GET_ARG1(*request);
        
        char *path;
        int rc = async_data_write_accept((void **) &path, true, 0, 0, 0, NULL);
        if (rc != EOK) {
                async_answer_0(rid, rc);
                return;
        }
        
        /* Ignore mode for now. */
        (void) mode;
        
        fibril_rwlock_write_lock(&namespace_rwlock);
        int lflag = L_DIRECTORY | L_CREATE | L_EXCLUSIVE;
        rc = vfs_lookup_internal(path, lflag, NULL, NULL);
        fibril_rwlock_write_unlock(&namespace_rwlock);
        free(path);
        async_answer_0(rid, rc);
}

void vfs_unlink(ipc_callid_t rid, ipc_call_t *request)
{
        int lflag = IPC_GET_ARG1(*request);
        
        char *path;
        int rc = async_data_write_accept((void **) &path, true, 0, 0, 0, NULL);
        if (rc != EOK) {
                async_answer_0(rid, rc);
                return;
        }
        
        fibril_rwlock_write_lock(&namespace_rwlock);
        lflag &= L_DIRECTORY;   /* sanitize lflag */
        vfs_lookup_res_t lr;
        rc = vfs_lookup_internal(path, lflag | L_UNLINK, &lr, NULL);
        free(path);
        if (rc != EOK) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                async_answer_0(rid, rc);
                return;
        }

        /*
         * The name has already been unlinked by vfs_lookup_internal().
         * We have to get and put the VFS node to ensure that it is
         * VFS_OUT_DESTROY'ed after the last reference to it is dropped.
         */
        vfs_node_t *node = vfs_node_get(&lr);
        fibril_mutex_lock(&nodes_mutex);
        node->lnkcnt--;
        fibril_mutex_unlock(&nodes_mutex);
        fibril_rwlock_write_unlock(&namespace_rwlock);
        vfs_node_put(node);
        async_answer_0(rid, EOK);
}

void vfs_rename(ipc_callid_t rid, ipc_call_t *request)
{
        /* Retrieve the old path. */
        char *old;
        int rc = async_data_write_accept((void **) &old, true, 0, 0, 0, NULL);
        if (rc != EOK) {
                async_answer_0(rid, rc);
                return;
        }
        
        /* Retrieve the new path. */
        char *new;
        rc = async_data_write_accept((void **) &new, true, 0, 0, 0, NULL);
        if (rc != EOK) {
                free(old);
                async_answer_0(rid, rc);
                return;
        }
        
        size_t olen;
        size_t nlen;
        char *oldc = canonify(old, &olen);
        char *newc = canonify(new, &nlen);
        
        if ((!oldc) || (!newc)) {
                async_answer_0(rid, EINVAL);
                free(old);
                free(new);
                return;
        }
        
        oldc[olen] = '\0';
        newc[nlen] = '\0';
        
        if ((!str_lcmp(newc, oldc, str_length(oldc))) &&
            ((newc[str_length(oldc)] == '/') ||
            (str_length(oldc) == 1) ||
            (str_length(oldc) == str_length(newc)))) {
                /*
                 * oldc is a prefix of newc and either
                 * - newc continues with a / where oldc ends, or
                 * - oldc was / itself, or
                 * - oldc and newc are equal.
                 */
                async_answer_0(rid, EINVAL);
                free(old);
                free(new);
                return;
        }
        
        vfs_lookup_res_t old_lr;
        vfs_lookup_res_t new_lr;
        vfs_lookup_res_t new_par_lr;
        fibril_rwlock_write_lock(&namespace_rwlock);
        
        /* Lookup the node belonging to the old file name. */
        rc = vfs_lookup_internal(oldc, L_NONE, &old_lr, NULL);
        if (rc != EOK) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                async_answer_0(rid, rc);
                free(old);
                free(new);
                return;
        }
        
        vfs_node_t *old_node = vfs_node_get(&old_lr);
        if (!old_node) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                async_answer_0(rid, ENOMEM);
                free(old);
                free(new);
                return;
        }
        
        /* Determine the path to the parent of the node with the new name. */
        char *parentc = str_dup(newc);
        if (!parentc) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                vfs_node_put(old_node);
                async_answer_0(rid, rc);
                free(old);
                free(new);
                return;
        }
        
        char *lastsl = str_rchr(parentc + 1, '/');
        if (lastsl)
                *lastsl = '\0';
        else
                parentc[1] = '\0';
        
        /* Lookup parent of the new file name. */
        rc = vfs_lookup_internal(parentc, L_NONE, &new_par_lr, NULL);
        free(parentc);  /* not needed anymore */
        if (rc != EOK) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                vfs_node_put(old_node);
                async_answer_0(rid, rc);
                free(old);
                free(new);
                return;
        }
        
        /* Check whether linking to the same file system instance. */
        if ((old_node->fs_handle != new_par_lr.triplet.fs_handle) ||
            (old_node->devmap_handle != new_par_lr.triplet.devmap_handle)) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                vfs_node_put(old_node);
                async_answer_0(rid, EXDEV);     /* different file systems */
                free(old);
                free(new);
                return;
        }
        
        /* Destroy the old link for the new name. */
        vfs_node_t *new_node = NULL;
        rc = vfs_lookup_internal(newc, L_UNLINK, &new_lr, NULL);
        
        switch (rc) {
        case ENOENT:
                /* simply not in our way */
                break;
        case EOK:
                new_node = vfs_node_get(&new_lr);
                if (!new_node) {
                        fibril_rwlock_write_unlock(&namespace_rwlock);
                        vfs_node_put(old_node);
                        async_answer_0(rid, ENOMEM);
                        free(old);
                        free(new);
                        return;
                }
                fibril_mutex_lock(&nodes_mutex);
                new_node->lnkcnt--;
                fibril_mutex_unlock(&nodes_mutex);
                break;
        default:
                fibril_rwlock_write_unlock(&namespace_rwlock);
                vfs_node_put(old_node);
                async_answer_0(rid, ENOTEMPTY);
                free(old);
                free(new);
                return;
        }
        
        /* Create the new link for the new name. */
        rc = vfs_lookup_internal(newc, L_LINK, NULL, NULL, old_node->index);
        if (rc != EOK) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                vfs_node_put(old_node);
                if (new_node)
                        vfs_node_put(new_node);
                async_answer_0(rid, rc);
                free(old);
                free(new);
                return;
        }
        
        fibril_mutex_lock(&nodes_mutex);
        old_node->lnkcnt++;
        fibril_mutex_unlock(&nodes_mutex);
        
        /* Destroy the link for the old name. */
        rc = vfs_lookup_internal(oldc, L_UNLINK, NULL, NULL);
        if (rc != EOK) {
                fibril_rwlock_write_unlock(&namespace_rwlock);
                vfs_node_put(old_node);
                if (new_node)
                        vfs_node_put(new_node);
                async_answer_0(rid, rc);
                free(old);
                free(new);
                return;
        }
        
        fibril_mutex_lock(&nodes_mutex);
        old_node->lnkcnt--;
        fibril_mutex_unlock(&nodes_mutex);
        fibril_rwlock_write_unlock(&namespace_rwlock);
        vfs_node_put(old_node);
        
        if (new_node)
                vfs_node_put(new_node);
        
        free(old);
        free(new);
        async_answer_0(rid, EOK);
}

void vfs_dup(ipc_callid_t rid, ipc_call_t *request)
{
        int oldfd = IPC_GET_ARG1(*request);
        int newfd = IPC_GET_ARG2(*request);
        
        /* If the file descriptors are the same, do nothing. */
        if (oldfd == newfd) {
                async_answer_1(rid, EOK, newfd);
                return;
        }
        
        /* Lookup the file structure corresponding to oldfd. */
        vfs_file_t *oldfile = vfs_file_get(oldfd);
        if (!oldfile) {
                async_answer_0(rid, EBADF);
                return;
        }
        
        /*
         * Lock the open file structure so that no other thread can manipulate
         * the same open file at a time.
         */
        fibril_mutex_lock(&oldfile->lock);
        
        /* Make sure newfd is closed. */
        (void) vfs_fd_free(newfd);
        
        /* Assign the old file to newfd. */
        int ret = vfs_fd_assign(oldfile, newfd);
        fibril_mutex_unlock(&oldfile->lock);
        vfs_file_put(oldfile);
        
        if (ret != EOK)
                async_answer_0(rid, ret);
        else
                async_answer_1(rid, EOK, newfd);
}

---