source: mainline/uspace/srv/fs/fat/fat_ops.c@ 6ebaff9

/*
 * 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        fat_ops.c
 * @brief       Implementation of VFS operations for the FAT file system server.
 */

#include "fat.h"
#include "../../vfs/vfs.h"
#include <libfs.h>
#include <ipc/ipc.h>
#include <ipc/services.h>
#include <ipc/devmap.h>
#include <async.h>
#include <errno.h>
#include <string.h>
#include <byteorder.h>
#include <libadt/hash_table.h>
#include <libadt/list.h>
#include <assert.h>
#include <futex.h>
#include <sys/mman.h>
#include <align.h>

#define BS_BLOCK                0
#define BS_SIZE                 512

/** Futex protecting the list of cached free FAT nodes. */
static futex_t ffn_futex = FUTEX_INITIALIZER;

/** List of cached free FAT nodes. */
static LIST_INITIALIZE(ffn_head);

#define FAT_NAME_LEN            8
#define FAT_EXT_LEN             3

#define FAT_PAD                 ' ' 

#define FAT_DENTRY_UNUSED       0x00
#define FAT_DENTRY_E5_ESC       0x05
#define FAT_DENTRY_DOT          0x2e
#define FAT_DENTRY_ERASED       0xe5

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

static void dentry_name_canonify(fat_dentry_t *d, char *buf)
{
        int i;

        for (i = 0; i < FAT_NAME_LEN; i++) {
                if (d->name[i] == FAT_PAD)
                        break;
                if (d->name[i] == FAT_DENTRY_E5_ESC)
                        *buf++ = 0xe5;
                else
                        *buf++ = d->name[i];
        }
        if (d->ext[0] != FAT_PAD)
                *buf++ = '.';
        for (i = 0; i < FAT_EXT_LEN; i++) {
                if (d->ext[i] == FAT_PAD) {
                        *buf = '\0';
                        return;
                }
                if (d->ext[i] == FAT_DENTRY_E5_ESC)
                        *buf++ = 0xe5;
                else
                        *buf++ = d->ext[i];
        }
        *buf = '\0';
}

static int dev_phone = -1;              /* FIXME */
static void *dev_buffer = NULL;         /* FIXME */

/* TODO move somewhere else */
typedef struct {
        void *data;
        size_t size;
        bool dirty;
} block_t;

static block_t *block_get(dev_handle_t dev_handle, off_t offset, size_t bs)
{
        /* FIXME */
        block_t *b;
        off_t bufpos = 0;
        size_t buflen = 0;
        off_t pos = offset * bs;

        assert(dev_phone != -1);
        assert(dev_buffer);

        b = malloc(sizeof(block_t));
        if (!b)
                return NULL;
        
        b->data = malloc(bs);
        if (!b->data) {
                free(b);
                return NULL;
        }
        b->size = bs;

        if (!libfs_blockread(dev_phone, dev_buffer, &bufpos, &buflen, &pos,
            b->data, bs, bs)) {
                free(b->data);
                free(b);
                return NULL;
        }

        return b;
}

static void block_put(block_t *block)
{
        /* FIXME */
        free(block->data);
        free(block);
}

#define FAT1            0

#define FAT_BS(b)               ((fat_bs_t *)((b)->data))

#define FAT_CLST_RES0   0x0000
#define FAT_CLST_RES1   0x0001
#define FAT_CLST_FIRST  0x0002
#define FAT_CLST_BAD    0xfff7
#define FAT_CLST_LAST1  0xfff8
#define FAT_CLST_LAST8  0xffff

/* internally used to mark root directory's parent */
#define FAT_CLST_ROOTPAR        FAT_CLST_RES0
/* internally used to mark root directory */
#define FAT_CLST_ROOT           FAT_CLST_RES1

#define fat_block_get(np, off) \
    _fat_block_get((np)->idx->dev_handle, (np)->firstc, (off))

static block_t *
_fat_block_get(dev_handle_t dev_handle, fat_cluster_t firstc, off_t offset)
{
        block_t *bb;
        block_t *b;
        unsigned bps;
        unsigned spc;
        unsigned rscnt;         /* block address of the first FAT */
        unsigned fatcnt;
        unsigned rde;
        unsigned rds;           /* root directory size */
        unsigned sf;
        unsigned ssa;           /* size of the system area */
        unsigned clusters;
        fat_cluster_t clst = firstc;
        unsigned i;

        bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
        bps = uint16_t_le2host(FAT_BS(bb)->bps);
        spc = FAT_BS(bb)->spc;
        rscnt = uint16_t_le2host(FAT_BS(bb)->rscnt);
        fatcnt = FAT_BS(bb)->fatcnt;
        rde = uint16_t_le2host(FAT_BS(bb)->root_ent_max);
        sf = uint16_t_le2host(FAT_BS(bb)->sec_per_fat);
        block_put(bb);

        rds = (sizeof(fat_dentry_t) * rde) / bps;
        rds += ((sizeof(fat_dentry_t) * rde) % bps != 0);
        ssa = rscnt + fatcnt * sf + rds;

        if (firstc == FAT_CLST_ROOT) {
                /* root directory special case */
                assert(offset < rds);
                b = block_get(dev_handle, rscnt + fatcnt * sf + offset, bps);
                return b;
        }

        clusters = offset / spc; 
        for (i = 0; i < clusters; i++) {
                unsigned fsec;  /* sector offset relative to FAT1 */
                unsigned fidx;  /* FAT1 entry index */

                assert(clst >= FAT_CLST_FIRST && clst < FAT_CLST_BAD);
                fsec = (clst * sizeof(fat_cluster_t)) / bps;
                fidx = clst % (bps / sizeof(fat_cluster_t));
                /* read FAT1 */
                b = block_get(dev_handle, rscnt + fsec, bps);
                clst = uint16_t_le2host(((fat_cluster_t *)b->data)[fidx]);
                assert(clst != FAT_CLST_BAD);
                assert(clst < FAT_CLST_LAST1);
                block_put(b);
        }

        b = block_get(dev_handle, ssa + (clst - FAT_CLST_FIRST) * spc +
            offset % spc, bps);

        return b;
}

/** Return number of blocks allocated to a file.
 *
 * @param dev_handle    Device handle of the device with the file.
 * @param firstc        First cluster of the file.
 *
 * @return              Number of blocks allocated to the file.
 */
static uint16_t 
_fat_blcks_get(dev_handle_t dev_handle, fat_cluster_t firstc)
{
        block_t *bb;
        block_t *b;
        unsigned bps;
        unsigned spc;
        unsigned rscnt;         /* block address of the first FAT */
        unsigned clusters = 0;
        fat_cluster_t clst = firstc;

        bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
        bps = uint16_t_le2host(FAT_BS(bb)->bps);
        spc = FAT_BS(bb)->spc;
        rscnt = uint16_t_le2host(FAT_BS(bb)->rscnt);
        block_put(bb);

        if (firstc == FAT_CLST_RES0) {
                /* No space allocated to the file. */
                return 0;
        }

        while (clst < FAT_CLST_LAST1) {
                unsigned fsec;  /* sector offset relative to FAT1 */
                unsigned fidx;  /* FAT1 entry index */

                assert(clst >= FAT_CLST_FIRST);
                fsec = (clst * sizeof(fat_cluster_t)) / bps;
                fidx = clst % (bps / sizeof(fat_cluster_t));
                /* read FAT1 */
                b = block_get(dev_handle, rscnt + fsec, bps);
                clst = uint16_t_le2host(((fat_cluster_t *)b->data)[fidx]);
                assert(clst != FAT_CLST_BAD);
                block_put(b);
                clusters++;
        }

        return clusters * spc;
}

static void fat_node_initialize(fat_node_t *node)
{
        futex_initialize(&node->lock, 1);
        node->idx = NULL;
        node->type = 0;
        link_initialize(&node->ffn_link);
        node->size = 0;
        node->lnkcnt = 0;
        node->refcnt = 0;
        node->dirty = false;
}

static uint16_t fat_bps_get(dev_handle_t dev_handle)
{
        block_t *bb;
        uint16_t bps;
        
        bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
        assert(bb != NULL);
        bps = uint16_t_le2host(FAT_BS(bb)->bps);
        block_put(bb);

        return bps;
}

typedef enum {
        FAT_DENTRY_SKIP,
        FAT_DENTRY_LAST,
        FAT_DENTRY_VALID
} fat_dentry_clsf_t;

static fat_dentry_clsf_t fat_classify_dentry(fat_dentry_t *d)
{
        if (d->attr & FAT_ATTR_VOLLABEL) {
                /* volume label entry */
                return FAT_DENTRY_SKIP;
        }
        if (d->name[0] == FAT_DENTRY_ERASED) {
                /* not-currently-used entry */
                return FAT_DENTRY_SKIP;
        }
        if (d->name[0] == FAT_DENTRY_UNUSED) {
                /* never used entry */
                return FAT_DENTRY_LAST;
        }
        if (d->name[0] == FAT_DENTRY_DOT) {
                /*
                 * Most likely '.' or '..'.
                 * It cannot occur in a regular file name.
                 */
                return FAT_DENTRY_SKIP;
        }
        return FAT_DENTRY_VALID;
}

static void fat_node_sync(fat_node_t *node)
{
        /* TODO */
}

/** Internal version of fat_node_get().
 *
 * @param idxp          Locked index structure.
 */
static void *fat_node_get_core(fat_idx_t *idxp)
{
        block_t *b;
        fat_dentry_t *d;
        fat_node_t *nodep = NULL;
        unsigned bps;
        unsigned dps;

        if (idxp->nodep) {
                /*
                 * We are lucky.
                 * The node is already instantiated in memory.
                 */
                futex_down(&idxp->nodep->lock);
                if (!idxp->nodep->refcnt++)
                        list_remove(&idxp->nodep->ffn_link);
                futex_up(&idxp->nodep->lock);
                return idxp->nodep;
        }

        /*
         * We must instantiate the node from the file system.
         */
        
        assert(idxp->pfc);

        futex_down(&ffn_futex);
        if (!list_empty(&ffn_head)) {
                /* Try to use a cached free node structure. */
                fat_idx_t *idxp_tmp;
                nodep = list_get_instance(ffn_head.next, fat_node_t, ffn_link);
                if (futex_trydown(&nodep->lock) == ESYNCH_WOULD_BLOCK)
                        goto skip_cache;
                idxp_tmp = nodep->idx;
                if (futex_trydown(&idxp_tmp->lock) == ESYNCH_WOULD_BLOCK) {
                        futex_up(&nodep->lock);
                        goto skip_cache;
                }
                list_remove(&nodep->ffn_link);
                futex_up(&ffn_futex);
                if (nodep->dirty)
                        fat_node_sync(nodep);
                idxp_tmp->nodep = NULL;
                futex_up(&nodep->lock);
                futex_up(&idxp_tmp->lock);
        } else {
skip_cache:
                /* Try to allocate a new node structure. */
                futex_up(&ffn_futex);
                nodep = (fat_node_t *)malloc(sizeof(fat_node_t));
                if (!nodep)
                        return NULL;
        }
        fat_node_initialize(nodep);

        bps = fat_bps_get(idxp->dev_handle);
        dps = bps / sizeof(fat_dentry_t);

        /* Read the block that contains the dentry of interest. */
        b = _fat_block_get(idxp->dev_handle, idxp->pfc,
            (idxp->pdi * sizeof(fat_dentry_t)) / bps);
        assert(b);

        d = ((fat_dentry_t *)b->data) + (idxp->pdi % dps);
        if (d->attr & FAT_ATTR_SUBDIR) {
                /* 
                 * The only directory which does not have this bit set is the
                 * root directory itself. The root directory node is handled
                 * and initialized elsewhere.
                 */
                nodep->type = FAT_DIRECTORY;
                /*
                 * Unfortunately, the 'size' field of the FAT dentry is not
                 * defined for the directory entry type. We must determine the
                 * size of the directory by walking the FAT.
                 */
                nodep->size = bps * _fat_blcks_get(idxp->dev_handle,
                    uint16_t_le2host(d->firstc));
        } else {
                nodep->type = FAT_FILE;
                nodep->size = uint32_t_le2host(d->size);
        }
        nodep->firstc = uint16_t_le2host(d->firstc);
        nodep->lnkcnt = 1;
        nodep->refcnt = 1;

        block_put(b);

        /* Link the idx structure with the node structure. */
        nodep->idx = idxp;
        idxp->nodep = nodep;

        return nodep;
}

/** Instantiate a FAT in-core node. */
static void *fat_node_get(dev_handle_t dev_handle, fs_index_t index)
{
        void *node;
        fat_idx_t *idxp;

        idxp = fat_idx_get_by_index(dev_handle, index);
        if (!idxp)
                return NULL;
        /* idxp->lock held */
        node = fat_node_get_core(idxp);
        futex_up(&idxp->lock);
        return node;
}

static void fat_node_put(void *node)
{
        fat_node_t *nodep = (fat_node_t *)node;

        futex_down(&nodep->lock);
        if (!--nodep->refcnt) {
                futex_down(&ffn_futex);
                list_append(&nodep->ffn_link, &ffn_head);
                futex_up(&ffn_futex);
        }
        futex_up(&nodep->lock);
}

static void *fat_create(int flags)
{
        return NULL;    /* not supported at the moment */
}

static int fat_destroy(void *node)
{
        return ENOTSUP; /* not supported at the moment */
}

static bool fat_link(void *prnt, void *chld, const char *name)
{
        return false;   /* not supported at the moment */
}

static int fat_unlink(void *prnt, void *chld)
{
        return ENOTSUP; /* not supported at the moment */
}

static void *fat_match(void *prnt, const char *component)
{
        fat_node_t *parentp = (fat_node_t *)prnt;
        char name[FAT_NAME_LEN + 1 + FAT_EXT_LEN + 1];
        unsigned i, j;
        unsigned bps;           /* bytes per sector */
        unsigned dps;           /* dentries per sector */
        unsigned blocks;
        fat_dentry_t *d;
        block_t *b;

        futex_down(&parentp->idx->lock);
        bps = fat_bps_get(parentp->idx->dev_handle);
        dps = bps / sizeof(fat_dentry_t);
        blocks = parentp->size / bps + (parentp->size % bps != 0);
        for (i = 0; i < blocks; i++) {
                unsigned dentries;
                
                b = fat_block_get(parentp, i);
                dentries = (i == blocks - 1) ?
                    parentp->size % sizeof(fat_dentry_t) :
                    dps;
                for (j = 0; j < dentries; j++) { 
                        d = ((fat_dentry_t *)b->data) + j;
                        switch (fat_classify_dentry(d)) {
                        case FAT_DENTRY_SKIP:
                                continue;
                        case FAT_DENTRY_LAST:
                                block_put(b);
                                futex_up(&parentp->idx->lock);
                                return NULL;
                        default:
                        case FAT_DENTRY_VALID:
                                dentry_name_canonify(d, name);
                                break;
                        }
                        if (stricmp(name, component) == 0) {
                                /* hit */
                                void *node;
                                /*
                                 * Assume tree hierarchy for locking.  We
                                 * already have the parent and now we are going
                                 * to lock the child.  Never lock in the oposite
                                 * order.
                                 */
                                fat_idx_t *idx = fat_idx_get_by_pos(
                                    parentp->idx->dev_handle, parentp->firstc,
                                    i * dps + j);
                                futex_up(&parentp->idx->lock);
                                if (!idx) {
                                        /*
                                         * Can happen if memory is low or if we
                                         * run out of 32-bit indices.
                                         */
                                        block_put(b);
                                        return NULL;
                                }
                                node = fat_node_get_core(idx);
                                futex_up(&idx->lock);
                                block_put(b);
                                return node;
                        }
                }
                block_put(b);
        }
        futex_up(&parentp->idx->lock);
        return NULL;
}

static fs_index_t fat_index_get(void *node)
{
        fat_node_t *fnodep = (fat_node_t *)node;
        if (!fnodep)
                return 0;
        return fnodep->idx->index;
}

static size_t fat_size_get(void *node)
{
        return ((fat_node_t *)node)->size;
}

static unsigned fat_lnkcnt_get(void *node)
{
        return ((fat_node_t *)node)->lnkcnt;
}

static bool fat_has_children(void *node)
{
        fat_node_t *nodep = (fat_node_t *)node;
        unsigned bps;
        unsigned dps;
        unsigned blocks;
        block_t *b;
        unsigned i, j;

        if (nodep->type != FAT_DIRECTORY)
                return false;

        futex_down(&nodep->idx->lock);
        bps = fat_bps_get(nodep->idx->dev_handle);
        dps = bps / sizeof(fat_dentry_t);

        blocks = nodep->size / bps + (nodep->size % bps != 0);

        for (i = 0; i < blocks; i++) {
                unsigned dentries;
                fat_dentry_t *d;
        
                b = fat_block_get(nodep, i);
                dentries = (i == blocks - 1) ?
                    nodep->size % sizeof(fat_dentry_t) :
                    dps;
                for (j = 0; j < dentries; j++) {
                        d = ((fat_dentry_t *)b->data) + j;
                        switch (fat_classify_dentry(d)) {
                        case FAT_DENTRY_SKIP:
                                continue;
                        case FAT_DENTRY_LAST:
                                block_put(b);
                                futex_up(&nodep->idx->lock);
                                return false;
                        default:
                        case FAT_DENTRY_VALID:
                                block_put(b);
                                futex_up(&nodep->idx->lock);
                                return true;
                        }
                        block_put(b);
                        futex_up(&nodep->idx->lock);
                        return true;
                }
                block_put(b);
        }

        futex_up(&nodep->idx->lock);
        return false;
}

static void *fat_root_get(dev_handle_t dev_handle)
{
        return fat_node_get(dev_handle, 0);
}

static char fat_plb_get_char(unsigned pos)
{
        return fat_reg.plb_ro[pos % PLB_SIZE];
}

static bool fat_is_directory(void *node)
{
        return ((fat_node_t *)node)->type == FAT_DIRECTORY;
}

static bool fat_is_file(void *node)
{
        return ((fat_node_t *)node)->type == FAT_FILE;
}

/** libfs operations */
libfs_ops_t fat_libfs_ops = {
        .match = fat_match,
        .node_get = fat_node_get,
        .node_put = fat_node_put,
        .create = fat_create,
        .destroy = fat_destroy,
        .link = fat_link,
        .unlink = fat_unlink,
        .index_get = fat_index_get,
        .size_get = fat_size_get,
        .lnkcnt_get = fat_lnkcnt_get,
        .has_children = fat_has_children,
        .root_get = fat_root_get,
        .plb_get_char = fat_plb_get_char,
        .is_directory = fat_is_directory,
        .is_file = fat_is_file
};

void fat_mounted(ipc_callid_t rid, ipc_call_t *request)
{
        dev_handle_t dev_handle = (dev_handle_t) IPC_GET_ARG1(*request);
        block_t *bb;
        uint16_t bps;
        uint16_t rde;
        int rc;

        /*
         * For now, we don't bother to remember dev_handle, dev_phone or
         * dev_buffer in some data structure. We use global variables because we
         * know there will be at most one mount on this file system.
         * Of course, this is a huge TODO item.
         */
        dev_buffer = mmap(NULL, BS_SIZE, PROTO_READ | PROTO_WRITE,
            MAP_ANONYMOUS | MAP_PRIVATE, 0, 0);
        
        if (!dev_buffer) {
                ipc_answer_0(rid, ENOMEM);
                return;
        }

        dev_phone = ipc_connect_me_to(PHONE_NS, SERVICE_DEVMAP,
            DEVMAP_CONNECT_TO_DEVICE, dev_handle);

        if (dev_phone < 0) {
                munmap(dev_buffer, BS_SIZE);
                ipc_answer_0(rid, dev_phone);
                return;
        }

        rc = ipc_share_out_start(dev_phone, dev_buffer,
            AS_AREA_READ | AS_AREA_WRITE);
        if (rc != EOK) {
                munmap(dev_buffer, BS_SIZE);
                ipc_answer_0(rid, rc);
                return;
        }

        /* Read the number of root directory entries. */
        bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
        bps = uint16_t_le2host(FAT_BS(bb)->bps);
        rde = uint16_t_le2host(FAT_BS(bb)->root_ent_max);
        block_put(bb);

        if (bps != BS_SIZE) {
                munmap(dev_buffer, BS_SIZE);
                ipc_answer_0(rid, ENOTSUP);
                return;
        }

        rc = fat_idx_init_by_dev_handle(dev_handle);
        if (rc != EOK) {
                munmap(dev_buffer, BS_SIZE);
                ipc_answer_0(rid, rc);
                return;
        }

        /* Initialize the root node. */
        fat_node_t *rootp = (fat_node_t *)malloc(sizeof(fat_node_t));
        if (!rootp) {
                munmap(dev_buffer, BS_SIZE);
                fat_idx_fini_by_dev_handle(dev_handle);
                ipc_answer_0(rid, ENOMEM);
                return;
        }
        fat_node_initialize(rootp);

        fat_idx_t *ridxp = fat_idx_get_by_pos(dev_handle, FAT_CLST_ROOTPAR, 0);
        if (!ridxp) {
                munmap(dev_buffer, BS_SIZE);
                free(rootp);
                fat_idx_fini_by_dev_handle(dev_handle);
                ipc_answer_0(rid, ENOMEM);
                return;
        }
        assert(ridxp->index == 0);
        /* ridxp->lock held */

        rootp->type = FAT_DIRECTORY;
        rootp->firstc = FAT_CLST_ROOT;
        rootp->refcnt = 1;
        rootp->lnkcnt = 0;      /* FS root is not linked */
        rootp->size = rde * sizeof(fat_dentry_t);
        rootp->idx = ridxp;
        ridxp->nodep = rootp;
        
        futex_up(&ridxp->lock);

        ipc_answer_3(rid, EOK, ridxp->index, rootp->size, rootp->lnkcnt);
}

void fat_mount(ipc_callid_t rid, ipc_call_t *request)
{
        ipc_answer_0(rid, ENOTSUP);
}

void fat_lookup(ipc_callid_t rid, ipc_call_t *request)
{
        libfs_lookup(&fat_libfs_ops, fat_reg.fs_handle, rid, request);
}

void fat_read(ipc_callid_t rid, ipc_call_t *request)
{
        dev_handle_t dev_handle = (dev_handle_t)IPC_GET_ARG1(*request);
        fs_index_t index = (fs_index_t)IPC_GET_ARG2(*request);
        off_t pos = (off_t)IPC_GET_ARG3(*request);
        fat_node_t *nodep = (fat_node_t *)fat_node_get(dev_handle, index);
        uint16_t bps = fat_bps_get(dev_handle);
        size_t bytes;
        block_t *b;

        if (!nodep) {
                ipc_answer_0(rid, ENOENT);
                return;
        }

        ipc_callid_t callid;
        size_t len;
        if (!ipc_data_read_receive(&callid, &len)) {
                fat_node_put(nodep);
                ipc_answer_0(callid, EINVAL);
                ipc_answer_0(rid, EINVAL);
                return;
        }

        if (nodep->type == FAT_FILE) {
                /*
                 * Our strategy for regular file reads is to read one block at
                 * most and make use of the possibility to return less data than
                 * requested. This keeps the code very simple.
                 */
                bytes = min(len, bps - pos % bps);
                b = fat_block_get(nodep, pos / bps);
                (void) ipc_data_read_finalize(callid, b->data + pos % bps,
                    bytes);
                block_put(b);
        } else {
                unsigned bnum;
                off_t spos = pos;
                char name[FAT_NAME_LEN + 1 + FAT_EXT_LEN + 1];
                fat_dentry_t *d;

                assert(nodep->type == FAT_DIRECTORY);
                assert(nodep->size % bps == 0);
                assert(bps % sizeof(fat_dentry_t) == 0);

                /*
                 * Our strategy for readdir() is to use the position pointer as
                 * an index into the array of all dentries. On entry, it points
                 * to the first unread dentry. If we skip any dentries, we bump
                 * the position pointer accordingly.
                 */
                bnum = (pos * sizeof(fat_dentry_t)) / bps;
                while (bnum < nodep->size / bps) {
                        off_t o;

                        b = fat_block_get(nodep, bnum);
                        for (o = pos % (bps / sizeof(fat_dentry_t));
                            o < bps / sizeof(fat_dentry_t);
                            o++, pos++) {
                                d = ((fat_dentry_t *)b->data) + o;
                                switch (fat_classify_dentry(d)) {
                                case FAT_DENTRY_SKIP:
                                        continue;
                                case FAT_DENTRY_LAST:
                                        block_put(b);
                                        goto miss;
                                default:
                                case FAT_DENTRY_VALID:
                                        dentry_name_canonify(d, name);
                                        block_put(b);
                                        goto hit;
                                }
                        }
                        block_put(b);
                        bnum++;
                }
miss:
                fat_node_put(nodep);
                ipc_answer_0(callid, ENOENT);
                ipc_answer_1(rid, ENOENT, 0);
                return;
hit:
                (void) ipc_data_read_finalize(callid, name, strlen(name) + 1);
                bytes = (pos - spos) + 1;
        }

        fat_node_put(nodep);
        ipc_answer_1(rid, EOK, (ipcarg_t)bytes);
}

/** Fill the gap between EOF and a new file position.
 *
 * @param nodep         FAT node with the gap.
 * @param mcl           First cluster in an independent cluster chain that will
 *                      be later appended to the end of the node's own cluster
 *                      chain. If pos is still in the last allocated cluster,
 *                      this argument is ignored.
 * @param pos           Position in the last node block.
 */
static void
fat_fill_gap(fat_node_t *nodep, fat_cluster_t mcl, off_t pos)
{
        uint16_t bps;
        unsigned spc;
        block_t *bb, *b;
        off_t o, boundary;

        bb = block_get(nodep->idx->dev_handle, BS_BLOCK, BS_SIZE);
        bps = uint16_t_le2host(FAT_BS(bb)->bps);
        spc = FAT_BS(bb)->spc;
        block_put(bb);
        
        boundary = ROUND_UP(nodep->size, bps * spc);

        /* zero out already allocated space */
        for (o = nodep->size - 1; o < pos && o < boundary;
            o = ALIGN_DOWN(o + bps, bps)) {
                b = fat_block_get(nodep, o / bps);
                memset(b->data + o % bps, 0, bps - o % bps);
                b->dirty = true;                /* need to sync node */
                block_put(b);
        }
        
        if (o >= pos)
                return;
        
        /* zero out the initial part of the new cluster chain */
        for (o = boundary; o < pos; o += bps) {
                b = _fat_block_get(nodep->idx->dev_handle, mcl,
                    (o - boundary) / bps);
                memset(b->data, 0, min(bps, pos - o));
                b->dirty = true;                /* need to sync node */
                block_put(b);
        }
}

static void
fat_mark_cluster(dev_handle_t dev_handle, unsigned fatno, fat_cluster_t clst,
    fat_cluster_t value)
{
        /* TODO */
}

static void
fat_alloc_shadow_clusters(dev_handle_t dev_handle, fat_cluster_t *lifo,
    unsigned nclsts)
{
        /* TODO */
}

static int
fat_alloc_clusters(dev_handle_t dev_handle, unsigned nclsts, fat_cluster_t *mcl,
    fat_cluster_t *lcl)
{
        uint16_t bps;
        uint16_t rscnt;
        uint16_t sf;
        block_t *bb, *blk;
        fat_cluster_t *lifo;    /* stack for storing free cluster numbers */ 
        unsigned found = 0;     /* top of the free cluster number stack */
        unsigned b, c, cl; 

        lifo = (fat_cluster_t *) malloc(nclsts * sizeof(fat_cluster_t));
        if (lifo)
                return ENOMEM;
        
        bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
        bps = uint16_t_le2host(FAT_BS(bb)->bps);
        rscnt = uint16_t_le2host(FAT_BS(bb)->rscnt);
        sf = uint16_t_le2host(FAT_BS(bb)->sec_per_fat);
        block_put(bb);
        
        /*
         * Search FAT1 for unused clusters.
         */
        for (b = 0, cl = 0; b < sf; blk++) {
                blk = block_get(dev_handle, rscnt + b, bps);
                for (c = 0; c < bps / sizeof(fat_cluster_t); c++, cl++) {
                        fat_cluster_t *clst = (fat_cluster_t *)blk->data + c;
                        if (*clst == FAT_CLST_RES0) {
                                /*
                                 * The cluster is free. Put it into our stack
                                 * of found clusters and mark it as non-free.
                                 */
                                lifo[found] = cl;
                                if (found == 0)
                                        *clst = FAT_CLST_LAST1;
                                else
                                        *clst = lifo[found - 1];
                                blk->dirty = true;      /* need to sync block */
                                if (++found == nclsts) {
                                        /* we are almost done */
                                        block_put(blk);
                                        /* update the shadow copies of FAT */
                                        fat_alloc_shadow_clusters(dev_handle,
                                            lifo, nclsts);
                                        *mcl = lifo[found - 1];
                                        *lcl = lifo[0];
                                        free(lifo);
                                        return EOK;
                                }
                        }
                }
                block_put(blk);
        }

        /*
         * We could not find enough clusters. Now we need to free the clusters
         * we have allocated so far.
         */
        while (found--)
                fat_mark_cluster(dev_handle, FAT1, lifo[found], FAT_CLST_RES0);
        
        free(lifo);
        return ENOSPC;
}

static void
fat_append_clusters(fat_node_t *nodep, fat_cluster_t mcl)
{
}

void fat_write(ipc_callid_t rid, ipc_call_t *request)
{
        dev_handle_t dev_handle = (dev_handle_t)IPC_GET_ARG1(*request);
        fs_index_t index = (fs_index_t)IPC_GET_ARG2(*request);
        off_t pos = (off_t)IPC_GET_ARG3(*request);
        fat_node_t *nodep = (fat_node_t *)fat_node_get(dev_handle, index);
        size_t bytes;
        block_t *b, *bb;
        uint16_t bps;
        unsigned spc;
        off_t boundary;
        
        if (!nodep) {
                ipc_answer_0(rid, ENOENT);
                return;
        }
        
        /* XXX remove me when you are ready */
        {
                ipc_answer_0(rid, ENOTSUP);
                fat_node_put(nodep);
                return;
        }

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

        /*
         * In all scenarios, we will attempt to write out only one block worth
         * of data at maximum. There might be some more efficient approaches,
         * but this one greatly simplifies fat_write(). Note that we can afford
         * to do this because the client must be ready to handle the return
         * value signalizing a smaller number of bytes written. 
         */ 
        bytes = min(len, bps - pos % bps);

        bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
        bps = uint16_t_le2host(FAT_BS(bb)->bps);
        spc = FAT_BS(bb)->spc;
        block_put(bb);
        
        boundary = ROUND_UP(nodep->size, bps * spc);
        if (pos < boundary) {
                /*
                 * This is the easier case - we are either overwriting already
                 * existing contents or writing behind the EOF, but still within
                 * the limits of the last cluster. The node size may grow to the
                 * next block size boundary.
                 */
                fat_fill_gap(nodep, FAT_CLST_RES0, pos);
                b = fat_block_get(nodep, pos / bps);
                (void) ipc_data_write_finalize(callid, b->data + pos % bps,
                    bytes);
                b->dirty = true;                /* need to sync block */
                block_put(b);
                if (pos + bytes > nodep->size) {
                        nodep->size = pos + bytes;
                        nodep->dirty = true;    /* need to sync node */
                }
                fat_node_put(nodep);
                ipc_answer_1(rid, EOK, bytes);  
                return;
        } else {
                /*
                 * This is the more difficult case. We must allocate new
                 * clusters for the node and zero them out.
                 */
                int status;
                unsigned nclsts;
                fat_cluster_t mcl, lcl;
        
                nclsts = (ROUND_UP(pos + bytes, bps * spc) - boundary) /
                    bps * spc;
                /* create an independent chain of nclsts clusters in all FATs */
                status = fat_alloc_clusters(dev_handle, nclsts, &mcl, &lcl);
                if (status != EOK) {
                        /* could not allocate a chain of nclsts clusters */
                        fat_node_put(nodep);
                        ipc_answer_0(callid, status);
                        ipc_answer_0(rid, status);
                        return;
                }
                /* zero fill any gaps */
                fat_fill_gap(nodep, mcl, pos);
                b = _fat_block_get(dev_handle, lcl, (pos / bps) % spc);
                (void) ipc_data_write_finalize(callid, b->data + pos % bps,
                    bytes);
                b->dirty = true;                /* need to sync block */
                block_put(b);
                /*
                 * Append the cluster chain starting in mcl to the end of the
                 * node's cluster chain.
                 */
                fat_append_clusters(nodep, mcl);
                nodep->size = pos + bytes;
                nodep->dirty = true;            /* need to sync node */
                fat_node_put(nodep);
                ipc_answer_1(rid, EOK, bytes);
                return;
        }
}

/**
 * @}
 */