source: mainline/uspace/srv/vfs/vfs_ops.c@ 4cc2ddd

/*
 * 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.
 */

/** @addtogroup fs
 * @{
 */ 

/**
 * @file        vfs_ops.c
 * @brief       Operations that VFS offers to its clients.
 */

#include <ipc/ipc.h>
#include <ipc/services.h>
#include <async.h>
#include <fibril.h>
#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <bool.h>
#include <futex.h>
#include <rwlock.h>
#include <libadt/list.h>
#include <unistd.h>
#include <ctype.h>
#include <as.h>
#include <assert.h>
#include <atomic.h>
#include "vfs.h"

#define min(a, b)       ((a) < (b) ? (a) : (b))

/**
 * This rwlock prevents the race between a triplet-to-VFS-node resolution and a
 * concurrent VFS operation which modifies the file system namespace.
 */
RWLOCK_INITIALIZE(namespace_rwlock);

atomic_t plb_futex = FUTEX_INITIALIZER;
link_t plb_head;        /**< PLB entry ring buffer. */
uint8_t *plb = NULL;

/** Perform a path lookup.
 *
 * @param path          Path to be resolved; it needn't be an ASCIIZ string.
 * @param len           Number of path characters pointed by path.
 * @param result        Empty structure where the lookup result will be stored.
 * @param altroot       If non-empty, will be used instead of rootfs as the root
 *                      of the whole VFS tree.
 *
 * @return              EOK on success or an error code from errno.h.
 */
int vfs_lookup_internal(char *path, size_t len, vfs_lookup_res_t *result,
    vfs_pair_t *altroot)
{
        vfs_pair_t *root;

        if (!len)
                return EINVAL;

        if (altroot)
                root = altroot;
        else
                root = (vfs_pair_t *) &rootfs;

        if (!root->fs_handle)
                return ENOENT;
        
        futex_down(&plb_futex);

        plb_entry_t entry;
        link_initialize(&entry.plb_link);
        entry.len = len;

        off_t first;    /* the first free index */
        off_t last;     /* the last free index */

        if (list_empty(&plb_head)) {
                first = 0;
                last = PLB_SIZE - 1;
        } else {
                plb_entry_t *oldest = list_get_instance(plb_head.next,
                    plb_entry_t, plb_link);
                plb_entry_t *newest = list_get_instance(plb_head.prev,
                    plb_entry_t, plb_link);

                first = (newest->index + newest->len) % PLB_SIZE;
                last = (oldest->index - 1) % PLB_SIZE;
        }

        if (first <= last) {
                if ((last - first) + 1 < len) {
                        /*
                         * The buffer cannot absorb the path.
                         */
                        futex_up(&plb_futex);
                        return ELIMIT;
                }
        } else {
                if (PLB_SIZE - ((first - last) + 1) < len) {
                        /*
                         * The buffer cannot absorb the path.
                         */
                        futex_up(&plb_futex);
                        return ELIMIT;
                }
        }

        /*
         * We know the first free index in PLB and we also know that there is
         * enough space in the buffer to hold our path.
         */

        entry.index = first;
        entry.len = len;

        /*
         * Claim PLB space by inserting the entry into the PLB entry ring
         * buffer.
         */
        list_append(&entry.plb_link, &plb_head);
        
        futex_up(&plb_futex);

        /*
         * Copy the path into PLB.
         */
        size_t cnt1 = min(len, (PLB_SIZE - first) + 1);
        size_t cnt2 = len - cnt1;
        
        memcpy(&plb[first], path, cnt1);
        memcpy(plb, &path[cnt1], cnt2);

        ipc_call_t answer;
        int phone = vfs_grab_phone(root->fs_handle);
        aid_t req = async_send_3(phone, VFS_LOOKUP, (ipcarg_t) first,
            (ipcarg_t) (first + len - 1) % PLB_SIZE,
            (ipcarg_t) root->dev_handle, &answer);
        vfs_release_phone(phone);

        ipcarg_t rc;
        async_wait_for(req, &rc);

        futex_down(&plb_futex);
        list_remove(&entry.plb_link);
        /*
         * Erasing the path from PLB will come handy for debugging purposes.
         */
        memset(&plb[first], 0, cnt1);
        memset(plb, 0, cnt2);
        futex_up(&plb_futex);

        if (rc == EOK) {
                result->triplet.fs_handle = (int) IPC_GET_ARG1(answer);
                result->triplet.dev_handle = (int) IPC_GET_ARG2(answer);
                result->triplet.index = (int) IPC_GET_ARG3(answer);
                result->size = (size_t) IPC_GET_ARG4(answer);
        }

        return rc;
}

atomic_t rootfs_futex = FUTEX_INITIALIZER;
vfs_triplet_t rootfs = {
        .fs_handle = 0,
        .dev_handle = 0,
        .index = 0,
};

static int lookup_root(int fs_handle, int dev_handle, vfs_lookup_res_t *result)
{
        vfs_pair_t altroot = {
                .fs_handle = fs_handle,
                .dev_handle = dev_handle,
        };

        return vfs_lookup_internal("/", strlen("/"), result, &altroot);
}

void vfs_mount(ipc_callid_t rid, ipc_call_t *request)
{
        int dev_handle;
        vfs_node_t *mp_node = NULL;

        /*
         * 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.
         */
        dev_handle = IPC_GET_ARG1(*request);

        /*
         * For now, don't make use of ARG2 and ARG3, but they can be used to
         * carry mount options in the future.
         */

        ipc_callid_t callid;
        size_t size;

        /*
         * Now, we expect the client to send us data with the name of the file
         * system.
         */
        if (!ipc_data_write_receive(&callid, &size)) {
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }

        /*
         * Don't receive more than is necessary for storing a full file system
         * name.
         */
        if (size < 1 || size > FS_NAME_MAXLEN) {
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }

        /*
         * Deliver the file system name.
         */
        char fs_name[FS_NAME_MAXLEN + 1];
        (void) ipc_data_write_finalize(callid, fs_name, size);
        fs_name[size] = '\0';
        
        /*
         * 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.
         */
        int fs_handle = fs_name_to_handle(fs_name, true);
        if (!fs_handle) {
                ipc_answer_0(rid, ENOENT);
                return;
        }

        /*
         * Now, we want the client to send us the mount point.
         */
        if (!ipc_data_write_receive(&callid, &size)) {
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }

        /*
         * Check whether size is reasonable wrt. the mount point.
         */
        if (size < 1 || size > MAX_PATH_LEN) {
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }
        /*
         * Allocate buffer for the mount point data being received.
         */
        uint8_t *buf;
        buf = malloc(size);
        if (!buf) {
                ipc_answer_0(callid, ENOMEM);
                ipc_answer_0(rid, ENOMEM);
                return;
        }

        /*
         * Deliver the mount point.
         */
        (void) ipc_data_write_finalize(callid, buf, size);

        /*
         * Lookup the root node of the filesystem being mounted.
         * In this case, we don't need to take the namespace_futex as the root
         * node cannot be removed. However, we do take a reference to it so
         * that we can track how many times it has been mounted.
         */
        int rc;
        vfs_lookup_res_t mr_res;
        rc = lookup_root(fs_handle, dev_handle, &mr_res);
        if (rc != EOK) {
                free(buf);
                ipc_answer_0(rid, rc);
                return;
        }
        vfs_node_t *mr_node = vfs_node_get(&mr_res);
        if (!mr_node) {
                free(buf);
                ipc_answer_0(rid, ENOMEM);
                return;
        }

        /*
         * Finally, we need to resolve the path to the mountpoint. 
         */
        vfs_lookup_res_t mp_res;
        futex_down(&rootfs_futex);
        if (rootfs.fs_handle) {
                /*
                 * We already have the root FS.
                 */
                rwlock_write_lock(&namespace_rwlock);
                rc = vfs_lookup_internal(buf, size, &mp_res, NULL);
                if (rc != EOK) {
                        /*
                         * The lookup failed for some reason.
                         */
                        rwlock_write_unlock(&namespace_rwlock);
                        futex_up(&rootfs_futex);
                        vfs_node_put(mr_node);  /* failed -> drop reference */
                        free(buf);
                        ipc_answer_0(rid, rc);
                        return;
                }
                mp_node = vfs_node_get(&mp_res);
                if (!mp_node) {
                        rwlock_write_unlock(&namespace_rwlock);
                        futex_up(&rootfs_futex);
                        vfs_node_put(mr_node);  /* failed -> drop reference */
                        free(buf);
                        ipc_answer_0(rid, ENOMEM);
                        return;
                }
                /*
                 * Now we hold a reference to mp_node.
                 * It will be dropped upon the corresponding VFS_UNMOUNT.
                 * This prevents the mount point from being deleted.
                 */
                rwlock_write_unlock(&namespace_rwlock);
        } else {
                /*
                 * We still don't have the root file system mounted.
                 */
                if ((size == 1) && (buf[0] == '/')) {
                        /*
                         * For this simple, but important case, we are done.
                         */
                        rootfs = mr_res.triplet;
                        futex_up(&rootfs_futex);
                        free(buf);
                        ipc_answer_0(rid, EOK);
                        return;
                } else {
                        /*
                         * We can't resolve this without the root filesystem
                         * being mounted first.
                         */
                        futex_up(&rootfs_futex);
                        free(buf);
                        vfs_node_put(mr_node);  /* failed -> drop reference */
                        ipc_answer_0(rid, ENOENT);
                        return;
                }
        }
        futex_up(&rootfs_futex);
        
        free(buf);      /* The buffer is not needed anymore. */
        
        /*
         * At this point, we have all necessary pieces: file system and device
         * handles, and we know the mount point VFS node and also the root node
         * of the file system being mounted.
         */

        int phone = vfs_grab_phone(mp_res.triplet.fs_handle);
        /* Later we can use ARG3 to pass mode/flags. */
        aid_t req1 = async_send_3(phone, VFS_MOUNT,
            (ipcarg_t) mp_res.triplet.dev_handle,
            (ipcarg_t) mp_res.triplet.index, 0, NULL);
        /* The second call uses the same method. */
        aid_t req2 = async_send_3(phone, VFS_MOUNT,
            (ipcarg_t) mr_res.triplet.fs_handle,
            (ipcarg_t) mr_res.triplet.dev_handle,
            (ipcarg_t) mr_res.triplet.index, NULL);
        vfs_release_phone(phone);

        ipcarg_t rc1;
        ipcarg_t rc2;
        async_wait_for(req1, &rc1);
        async_wait_for(req2, &rc2);

        if ((rc1 != EOK) || (rc2 != EOK)) {
                /* Mount failed, drop references to mr_node and mp_node. */
                vfs_node_put(mr_node);
                if (mp_node)
                        vfs_node_put(mp_node);
        }
        
        if (rc2 == EOK)
                ipc_answer_0(rid, rc1);
        else if (rc1 == EOK)
                ipc_answer_0(rid, rc2);
        else
                ipc_answer_0(rid, rc1);
}

void vfs_open(ipc_callid_t rid, ipc_call_t *request)
{
        if (!vfs_files_init()) {
                ipc_answer_0(rid, ENOMEM);
                return;
        }

        /*
         * The POSIX interface is open(path, flags, mode).
         * We can receive flags and mode along with the VFS_OPEN call; the path
         * will need to arrive in another call.
         */
        int flags = IPC_GET_ARG1(*request);
        int mode = IPC_GET_ARG2(*request);
        size_t len;

        ipc_callid_t callid;

        if (!ipc_data_write_receive(&callid, &len)) {
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }

        /*
         * Now we are on the verge of accepting the path.
         *
         * There is one optimization we could do in the future: copy the path
         * directly into the PLB using some kind of a callback.
         */
        char *path = malloc(len);
        
        if (!path) {
                ipc_answer_0(callid, ENOMEM);
                ipc_answer_0(rid, ENOMEM);
                return;
        }

        int rc;
        if ((rc = ipc_data_write_finalize(callid, path, len))) {
                ipc_answer_0(rid, rc);
                free(path);
                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.
         */
        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, len, &lr, NULL);
        if (rc) {
                rwlock_read_unlock(&namespace_rwlock);
                ipc_answer_0(rid, rc);
                free(path);
                return;
        }

        /*
         * Path is no longer needed.
         */
        free(path);

        vfs_node_t *node = vfs_node_get(&lr);
        rwlock_read_unlock(&namespace_rwlock);

        /*
         * Get ourselves a file descriptor and the corresponding vfs_file_t
         * structure.
         */
        int fd = vfs_fd_alloc();
        if (fd < 0) {
                vfs_node_put(node);
                ipc_answer_0(rid, fd);
                return;
        }
        vfs_file_t *file = vfs_file_get(fd);
        file->node = node;

        /*
         * 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_CLOSE.
         */
        vfs_node_addref(node);
        vfs_node_put(node);

        /*
         * Success! Return the new file descriptor to the client.
         */
        ipc_answer_1(rid, EOK, fd);
}

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

        /*
         * 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) {
                ipc_answer_0(rid, ENOENT);
                return;
        }

        /*
         * Now we need to receive a call with client's
         * IPC_M_DATA_READ/IPC_M_DATA_WRITE request.
         */
        ipc_callid_t callid;
        int res;
        if (read)
                res = ipc_data_read_receive(&callid, NULL);
        else 
                res = ipc_data_write_receive(&callid, NULL);
        if (!res) {
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }

        /*
         * Lock the open file structure so that no other thread can manipulate
         * the same open file at a time.
         */
        futex_down(&file->lock);

        /*
         * Lock the file's node so that no other client can read/write to it at
         * the same time.
         */
        if (read)
                rwlock_read_lock(&file->node->contents_rwlock);
        else
                rwlock_write_lock(&file->node->contents_rwlock);

        int fs_phone = vfs_grab_phone(file->node->fs_handle);   
        
        /*
         * Make a VFS_READ/VFS_WRITE request at the destination FS server.
         */
        aid_t msg;
        ipc_call_t answer;
        msg = async_send_3(fs_phone, IPC_GET_METHOD(*request),
            file->node->dev_handle, file->node->index, file->pos, &answer);
        
        /*
         * 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.
         */
        ipc_forward_fast(callid, fs_phone, 0, 0, 0, IPC_FF_ROUTE_FROM_ME);

        vfs_release_phone(fs_phone);

        /*
         * Wait for reply from the FS server.
         */
        ipcarg_t rc;
        async_wait_for(msg, &rc);
        size_t bytes = IPC_GET_ARG1(answer);

        /*
         * Unlock the VFS node.
         */
        if (read)
                rwlock_read_unlock(&file->node->contents_rwlock);
        else {
                /* Update the cached version of node's size. */
                file->node->size = IPC_GET_ARG2(answer); 
                rwlock_write_unlock(&file->node->contents_rwlock);
        }

        /*
         * Update the position pointer and unlock the open file.
         */
        file->pos += bytes;
        futex_up(&file->lock);

        /*
         * FS server's reply is the final result of the whole operation we
         * return to the client.
         */
        ipc_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);
        off_t off = (off_t) IPC_GET_ARG2(*request);
        int whence = (int) IPC_GET_ARG3(*request);


        /*
         * Lookup the file structure corresponding to the file descriptor.
         */
        vfs_file_t *file = vfs_file_get(fd);
        if (!file) {
                ipc_answer_0(rid, ENOENT);
                return;
        }

        off_t newpos;
        futex_down(&file->lock);
        if (whence == SEEK_SET) {
                file->pos = off;
                futex_up(&file->lock);
                ipc_answer_1(rid, EOK, off);
                return;
        }
        if (whence == SEEK_CUR) {
                if (file->pos + off < file->pos) {
                        futex_up(&file->lock);
                        ipc_answer_0(rid, EOVERFLOW);
                        return;
                }
                file->pos += off;
                newpos = file->pos;
                futex_up(&file->lock);
                ipc_answer_1(rid, EOK, newpos);
                return;
        }
        if (whence == SEEK_END) {
                rwlock_read_lock(&file->node->contents_rwlock);
                size_t size = file->node->size;
                rwlock_read_unlock(&file->node->contents_rwlock);
                if (size + off < size) {
                        futex_up(&file->lock);
                        ipc_answer_0(rid, EOVERFLOW);
                        return;
                }
                newpos = size + off;
                futex_up(&file->lock);
                ipc_answer_1(rid, EOK, newpos);
                return;
        }
        futex_up(&file->lock);
        ipc_answer_0(rid, EINVAL);
}

void vfs_truncate(ipc_callid_t rid, ipc_call_t *request)
{
        int fd = IPC_GET_ARG1(*request);
        size_t size = IPC_GET_ARG2(*request);
        ipcarg_t rc;

        vfs_file_t *file = vfs_file_get(fd);
        if (!file) {
                ipc_answer_0(rid, ENOENT);
                return;
        }
        futex_down(&file->lock);

        rwlock_write_lock(&file->node->contents_rwlock);
        int fs_phone = vfs_grab_phone(file->node->fs_handle);
        rc = async_req_3_0(fs_phone, VFS_TRUNCATE, (ipcarg_t)file->node->dev_handle,
            (ipcarg_t)file->node->index, (ipcarg_t)size);
        vfs_release_phone(fs_phone);
        if (rc == EOK)
                file->node->size = size;
        rwlock_write_unlock(&file->node->contents_rwlock);

        futex_up(&file->lock);

        return rc;      
}

atomic_t fs_head_futex = FUTEX_INITIALIZER;
link_t fs_head;

atomic_t fs_handle_next = {
        .count = 1
};

/** Verify the VFS info structure.
 *
 * @param info          Info structure to be verified.
 *
 * @return              Non-zero if the info structure is sane, zero otherwise.
 */
static bool vfs_info_sane(vfs_info_t *info)
{
        int i;

        /*
         * Check if the name is non-empty and is composed solely of ASCII
         * characters [a-z]+[a-z0-9_-]*.
         */
        if (!islower(info->name[0])) {
                dprintf("The name doesn't start with a lowercase character.\n");
                return false;
        }
        for (i = 1; i < FS_NAME_MAXLEN; i++) {
                if (!(islower(info->name[i]) || isdigit(info->name[i])) &&
                    (info->name[i] != '-') && (info->name[i] != '_')) {
                        if (info->name[i] == '\0') {
                                break;
                        } else {
                                dprintf("The name contains illegal "
                                    "characters.\n");
                                return false;
                        }
                }
        }
        /*
         * This check is not redundant. It ensures that the name is
         * NULL-terminated, even if FS_NAME_MAXLEN characters are used.
         */
        if (info->name[i] != '\0') {
                dprintf("The name is not properly NULL-terminated.\n"); 
                return false;
        }
        

        /*
         * Check if the FS implements mandatory VFS operations.
         */
        if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_LOOKUP)] != VFS_OP_DEFINED) {
                dprintf("Operation VFS_LOOKUP not defined by the client.\n");
                return false;
        }
        if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_OPEN)] != VFS_OP_DEFINED) {
                dprintf("Operation VFS_OPEN not defined by the client.\n");
                return false;
        }
        if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_CLOSE)] != VFS_OP_DEFINED) {
                dprintf("Operation VFS_CLOSE not defined by the client.\n");
                return false;
        }
        if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_READ)] != VFS_OP_DEFINED) {
                dprintf("Operation VFS_READ not defined by the client.\n");
                return false;
        }
        
        /*
         * Check if each operation is either not defined, defined or default.
         */
        for (i = VFS_FIRST; i < VFS_LAST_CLNT; i++) {
                if ((info->ops[IPC_METHOD_TO_VFS_OP(i)] != VFS_OP_NULL) && 
                    (info->ops[IPC_METHOD_TO_VFS_OP(i)] != VFS_OP_DEFAULT) && 
                    (info->ops[IPC_METHOD_TO_VFS_OP(i)] != VFS_OP_DEFINED)) {
                        dprintf("Operation info not understood.\n");
                        return false;
                }
        }
        return true;
}

/** VFS_REGISTER protocol function.
 *
 * @param rid           Hash of the call with the request.
 * @param request       Call structure with the request.
 */
void vfs_register(ipc_callid_t rid, ipc_call_t *request)
{
        ipc_callid_t callid;
        ipc_call_t call;
        int rc;
        size_t size;

        dprintf("Processing VFS_REGISTER request received from %p.\n",
            request->in_phone_hash);

        /*
         * The first call has to be IPC_M_DATA_SEND in which we receive the
         * VFS info structure from the client FS.
         */
        if (!ipc_data_write_receive(&callid, &size)) {
                /*
                 * The client doesn't obey the same protocol as we do.
                 */
                dprintf("Receiving of VFS info failed.\n");
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }
        
        dprintf("VFS info received, size = %d\n", size);
        
        /*
         * We know the size of the VFS info structure. See if the client
         * understands this easy concept too.
         */
        if (size != sizeof(vfs_info_t)) {
                /*
                 * The client is sending us something, which cannot be
                 * the info structure.
                 */
                dprintf("Received VFS info has bad size.\n");
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }

        /*
         * Allocate and initialize a buffer for the fs_info structure.
         */
        fs_info_t *fs_info;
        fs_info = (fs_info_t *) malloc(sizeof(fs_info_t));
        if (!fs_info) {
                dprintf("Could not allocate memory for FS info.\n");
                ipc_answer_0(callid, ENOMEM);
                ipc_answer_0(rid, ENOMEM);
                return;
        }
        link_initialize(&fs_info->fs_link);
        futex_initialize(&fs_info->phone_futex, 1);
                
        rc = ipc_data_write_finalize(callid, &fs_info->vfs_info, size);
        if (rc != EOK) {
                dprintf("Failed to deliver the VFS info into our AS, rc=%d.\n",
                    rc);
                free(fs_info);
                ipc_answer_0(callid, rc);
                ipc_answer_0(rid, rc);
                return;
        }

        dprintf("VFS info delivered.\n");
                
        if (!vfs_info_sane(&fs_info->vfs_info)) {
                free(fs_info);
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }
                
        futex_down(&fs_head_futex);

        /*
         * Check for duplicit registrations.
         */
        if (fs_name_to_handle(fs_info->vfs_info.name, false)) {
                /*
                 * We already register a fs like this.
                 */
                dprintf("FS is already registered.\n");
                futex_up(&fs_head_futex);
                free(fs_info);
                ipc_answer_0(callid, EEXISTS);
                ipc_answer_0(rid, EEXISTS);
                return;
        }

        /*
         * Add fs_info to the list of registered FS's.
         */
        dprintf("Inserting FS into the list of registered file systems.\n");
        list_append(&fs_info->fs_link, &fs_head);

        /*
         * Now we want the client to send us the IPC_M_CONNECT_TO_ME call so
         * that a callback connection is created and we have a phone through
         * which to forward VFS requests to it.
         */
        callid = async_get_call(&call);
        if (IPC_GET_METHOD(call) != IPC_M_CONNECT_TO_ME) {
                dprintf("Unexpected call, method = %d\n", IPC_GET_METHOD(call));
                list_remove(&fs_info->fs_link);
                futex_up(&fs_head_futex);
                free(fs_info);
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }
        fs_info->phone = IPC_GET_ARG5(call);
        ipc_answer_0(callid, EOK);

        dprintf("Callback connection to FS created.\n");

        /*
         * The client will want us to send him the address space area with PLB.
         */

        if (!ipc_share_in_receive(&callid, &size)) {
                dprintf("Unexpected call, method = %d\n", IPC_GET_METHOD(call));
                list_remove(&fs_info->fs_link);
                futex_up(&fs_head_futex);
                ipc_hangup(fs_info->phone);
                free(fs_info);
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }
        
        /*
         * We can only send the client address space area PLB_SIZE bytes long.
         */
        if (size != PLB_SIZE) {
                dprintf("Client suggests wrong size of PFB, size = %d\n", size);
                list_remove(&fs_info->fs_link);
                futex_up(&fs_head_futex);
                ipc_hangup(fs_info->phone);
                free(fs_info);
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }

        /*
         * Commit to read-only sharing the PLB with the client.
         */
        (void) ipc_share_in_finalize(callid, plb,
            AS_AREA_READ | AS_AREA_CACHEABLE);

        dprintf("Sharing PLB.\n");

        /*
         * That was it. The FS has been registered.
         * In reply to the VFS_REGISTER request, we assign the client file
         * system a global file system handle.
         */
        fs_info->fs_handle = (int) atomic_postinc(&fs_handle_next);
        ipc_answer_1(rid, EOK, (ipcarg_t) fs_info->fs_handle);
        
        futex_up(&fs_head_futex);
        
        dprintf("\"%.*s\" filesystem successfully registered, handle=%d.\n",
            FS_NAME_MAXLEN, fs_info->vfs_info.name, fs_info->fs_handle);
}

/** For a given file system handle, implement policy for allocating a phone.
 *
 * @param handle        File system handle.
 *
 * @return              Phone over which a multi-call request can be safely
 *                      sent. Return 0 if no phone was found.
 */
int vfs_grab_phone(int handle)
{
        /*
         * For now, we don't try to be very clever and very fast.
         * We simply lookup the phone in the fs_head list. We currently don't
         * open any additional phones (even though that itself would be pretty
         * straightforward; housekeeping multiple open phones to a FS task would
         * be more demanding). Instead, we simply take the respective
         * phone_futex and keep it until vfs_release_phone().
         */
        futex_down(&fs_head_futex);
        link_t *cur;
        fs_info_t *fs;
        for (cur = fs_head.next; cur != &fs_head; cur = cur->next) {
                fs = list_get_instance(cur, fs_info_t, fs_link);
                if (fs->fs_handle == handle) {
                        futex_up(&fs_head_futex);
                        /*
                         * For now, take the futex unconditionally.
                         * Oh yeah, serialization rocks.
                         * It will be up'ed in vfs_release_phone().
                         */
                        futex_down(&fs->phone_futex);
                        /*
                         * Avoid deadlock with other fibrils in the same thread
                         * by disabling fibril preemption.
                         */
                        fibril_inc_sercount();
                        return fs->phone; 
                }
        }
        futex_up(&fs_head_futex);
        return 0;
}

/** Tell VFS that the phone is in use for any request.
 *
 * @param phone         Phone to FS task.
 */
void vfs_release_phone(int phone)
{
        bool found = false;

        /*
         * Undo the fibril_inc_sercount() done in vfs_grab_phone().
         */
        fibril_dec_sercount();
        
        futex_down(&fs_head_futex);
        link_t *cur;
        for (cur = fs_head.next; cur != &fs_head; cur = cur->next) {
                fs_info_t *fs = list_get_instance(cur, fs_info_t, fs_link);
                if (fs->phone == phone) {
                        found = true;
                        futex_up(&fs_head_futex);
                        futex_up(&fs->phone_futex);
                        return;
                }
        }
        futex_up(&fs_head_futex);

        /*
         * Not good to get here.
         */
        assert(found == true);
}

/** Convert file system name to its handle.
 *
 * @param name          File system name.
 * @param lock          If true, the function will down and up the
 *                      fs_head_futex.
 *
 * @return              File system handle or zero if file system not found.
 */
int fs_name_to_handle(char *name, bool lock)
{
        int handle = 0;
        
        if (lock)
                futex_down(&fs_head_futex);
        link_t *cur;
        for (cur = fs_head.next; cur != &fs_head; cur = cur->next) {
                fs_info_t *fs = list_get_instance(cur, fs_info_t, fs_link);
                if (strncmp(fs->vfs_info.name, name,
                    sizeof(fs->vfs_info.name)) == 0) { 
                        handle = fs->fs_handle;
                        break;
                }
        }
        if (lock)
                futex_up(&fs_head_futex);
        return handle;
}

/**
 * @}
 */