Subversion Repositories HelenOS

Rev

Rev 2690 | Go to most recent revision | Blame | Compare with Previous | Last modification | View Log | Download | RSS feed

  1. /*
  2.  * Copyright (c) 2008 Jakub Jermar
  3.  * All rights reserved.
  4.  *
  5.  * Redistribution and use in source and binary forms, with or without
  6.  * modification, are permitted provided that the following conditions
  7.  * are met:
  8.  *
  9.  * - Redistributions of source code must retain the above copyright
  10.  *   notice, this list of conditions and the following disclaimer.
  11.  * - Redistributions in binary form must reproduce the above copyright
  12.  *   notice, this list of conditions and the following disclaimer in the
  13.  *   documentation and/or other materials provided with the distribution.
  14.  * - The name of the author may not be used to endorse or promote products
  15.  *   derived from this software without specific prior written permission.
  16.  *
  17.  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  18.  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  19.  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  20.  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  21.  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  22.  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  23.  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  24.  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  25.  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  26.  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  27.  */
  28.  
  29. /** @addtogroup fs
  30.  * @{
  31.  */
  32.  
  33. /**
  34.  * @file    vfs_ops.c
  35.  * @brief   Operations that VFS offers to its clients.
  36.  */
  37.  
  38. #include <ipc/ipc.h>
  39. #include <ipc/services.h>
  40. #include <async.h>
  41. #include <fibril.h>
  42. #include <errno.h>
  43. #include <stdio.h>
  44. #include <stdlib.h>
  45. #include <string.h>
  46. #include <bool.h>
  47. #include <futex.h>
  48. #include <rwlock.h>
  49. #include <libadt/list.h>
  50. #include <unistd.h>
  51. #include <ctype.h>
  52. #include <as.h>
  53. #include <assert.h>
  54. #include <atomic.h>
  55. #include "vfs.h"
  56.  
  57. #define min(a, b)   ((a) < (b) ? (a) : (b))
  58.  
  59. /**
  60.  * This rwlock prevents the race between a triplet-to-VFS-node resolution and a
  61.  * concurrent VFS operation which modifies the file system namespace.
  62.  */
  63. RWLOCK_INITIALIZE(namespace_rwlock);
  64.  
  65. atomic_t plb_futex = FUTEX_INITIALIZER;
  66. link_t plb_head;    /**< PLB entry ring buffer. */
  67. uint8_t *plb = NULL;
  68.  
  69. /** Perform a path lookup.
  70.  *
  71.  * @param path      Path to be resolved; it needn't be an ASCIIZ string.
  72.  * @param len       Number of path characters pointed by path.
  73.  * @param result    Empty node structure where the result will be stored.
  74.  * @param size      Storage where the size of the node will be stored. Can
  75.  *          be NULL.
  76.  * @param altroot   If non-empty, will be used instead of rootfs as the root
  77.  *          of the whole VFS tree.
  78.  *
  79.  * @return      EOK on success or an error code from errno.h.
  80.  */
  81. int vfs_lookup_internal(char *path, size_t len, vfs_triplet_t *result,
  82.     size_t *size, vfs_pair_t *altroot)
  83. {
  84.     vfs_pair_t *root;
  85.  
  86.     if (!len)
  87.         return EINVAL;
  88.  
  89.     if (altroot)
  90.         root = altroot;
  91.     else
  92.         root = (vfs_pair_t *) &rootfs;
  93.  
  94.     if (!root->fs_handle)
  95.         return ENOENT;
  96.    
  97.     futex_down(&plb_futex);
  98.  
  99.     plb_entry_t entry;
  100.     link_initialize(&entry.plb_link);
  101.     entry.len = len;
  102.  
  103.     off_t first;    /* the first free index */
  104.     off_t last; /* the last free index */
  105.  
  106.     if (list_empty(&plb_head)) {
  107.         first = 0;
  108.         last = PLB_SIZE - 1;
  109.     } else {
  110.         plb_entry_t *oldest = list_get_instance(plb_head.next,
  111.             plb_entry_t, plb_link);
  112.         plb_entry_t *newest = list_get_instance(plb_head.prev,
  113.             plb_entry_t, plb_link);
  114.  
  115.         first = (newest->index + newest->len) % PLB_SIZE;
  116.         last = (oldest->index - 1) % PLB_SIZE;
  117.     }
  118.  
  119.     if (first <= last) {
  120.         if ((last - first) + 1 < len) {
  121.             /*
  122.              * The buffer cannot absorb the path.
  123.              */
  124.             futex_up(&plb_futex);
  125.             return ELIMIT;
  126.         }
  127.     } else {
  128.         if (PLB_SIZE - ((first - last) + 1) < len) {
  129.             /*
  130.              * The buffer cannot absorb the path.
  131.              */
  132.             futex_up(&plb_futex);
  133.             return ELIMIT;
  134.         }
  135.     }
  136.  
  137.     /*
  138.      * We know the first free index in PLB and we also know that there is
  139.      * enough space in the buffer to hold our path.
  140.      */
  141.  
  142.     entry.index = first;
  143.     entry.len = len;
  144.  
  145.     /*
  146.      * Claim PLB space by inserting the entry into the PLB entry ring
  147.      * buffer.
  148.      */
  149.     list_append(&entry.plb_link, &plb_head);
  150.    
  151.     futex_up(&plb_futex);
  152.  
  153.     /*
  154.      * Copy the path into PLB.
  155.      */
  156.     size_t cnt1 = min(len, (PLB_SIZE - first) + 1);
  157.     size_t cnt2 = len - cnt1;
  158.    
  159.     memcpy(&plb[first], path, cnt1);
  160.     memcpy(plb, &path[cnt1], cnt2);
  161.  
  162.     ipc_call_t answer;
  163.     int phone = vfs_grab_phone(root->fs_handle);
  164.     aid_t req = async_send_3(phone, VFS_LOOKUP, (ipcarg_t) first,
  165.         (ipcarg_t) (first + len - 1) % PLB_SIZE,
  166.         (ipcarg_t) root->dev_handle, &answer);
  167.     vfs_release_phone(phone);
  168.  
  169.     ipcarg_t rc;
  170.     async_wait_for(req, &rc);
  171.  
  172.     futex_down(&plb_futex);
  173.     list_remove(&entry.plb_link);
  174.     /*
  175.      * Erasing the path from PLB will come handy for debugging purposes.
  176.      */
  177.     memset(&plb[first], 0, cnt1);
  178.     memset(plb, 0, cnt2);
  179.     futex_up(&plb_futex);
  180.  
  181.     if (rc == EOK) {
  182.         result->fs_handle = (int) IPC_GET_ARG1(answer);
  183.         result->dev_handle = (int) IPC_GET_ARG2(answer);
  184.         result->index = (int) IPC_GET_ARG3(answer);
  185.         if (size)
  186.             *size = (size_t) IPC_GET_ARG4(answer);
  187.     }
  188.  
  189.     return rc;
  190. }
  191.  
  192. atomic_t rootfs_futex = FUTEX_INITIALIZER;
  193. vfs_triplet_t rootfs = {
  194.     .fs_handle = 0,
  195.     .dev_handle = 0,
  196.     .index = 0,
  197. };
  198.  
  199. static int lookup_root(int fs_handle, int dev_handle, vfs_triplet_t *root,
  200.     size_t *size)
  201. {
  202.     vfs_pair_t altroot = {
  203.         .fs_handle = fs_handle,
  204.         .dev_handle = dev_handle,
  205.     };
  206.  
  207.     return vfs_lookup_internal("/", strlen("/"), root, size, &altroot);
  208. }
  209.  
  210. void vfs_mount(ipc_callid_t rid, ipc_call_t *request)
  211. {
  212.     int dev_handle;
  213.     vfs_node_t *mp_node = NULL;
  214.  
  215.     /*
  216.      * We expect the library to do the device-name to device-handle
  217.      * translation for us, thus the device handle will arrive as ARG1
  218.      * in the request.
  219.      */
  220.     dev_handle = IPC_GET_ARG1(*request);
  221.  
  222.     /*
  223.      * For now, don't make use of ARG2 and ARG3, but they can be used to
  224.      * carry mount options in the future.
  225.      */
  226.  
  227.     ipc_callid_t callid;
  228.     size_t size;
  229.  
  230.     /*
  231.      * Now, we expect the client to send us data with the name of the file
  232.      * system.
  233.      */
  234.     if (!ipc_data_write_receive(&callid, &size)) {
  235.         ipc_answer_0(callid, EINVAL);
  236.         ipc_answer_0(rid, EINVAL);
  237.         return;
  238.     }
  239.  
  240.     /*
  241.      * Don't receive more than is necessary for storing a full file system
  242.      * name.
  243.      */
  244.     if (size < 1 || size > FS_NAME_MAXLEN) {
  245.         ipc_answer_0(callid, EINVAL);
  246.         ipc_answer_0(rid, EINVAL);
  247.         return;
  248.     }
  249.  
  250.     /*
  251.      * Deliver the file system name.
  252.      */
  253.     char fs_name[FS_NAME_MAXLEN + 1];
  254.     (void) ipc_data_write_finalize(callid, fs_name, size);
  255.     fs_name[size] = '\0';
  256.    
  257.     /*
  258.      * Check if we know a file system with the same name as is in fs_name.
  259.      * This will also give us its file system handle.
  260.      */
  261.     int fs_handle = fs_name_to_handle(fs_name, true);
  262.     if (!fs_handle) {
  263.         ipc_answer_0(rid, ENOENT);
  264.         return;
  265.     }
  266.  
  267.     /*
  268.      * Now, we want the client to send us the mount point.
  269.      */
  270.     if (!ipc_data_write_receive(&callid, &size)) {
  271.         ipc_answer_0(callid, EINVAL);
  272.         ipc_answer_0(rid, EINVAL);
  273.         return;
  274.     }
  275.  
  276.     /*
  277.      * Check whether size is reasonable wrt. the mount point.
  278.      */
  279.     if (size < 1 || size > MAX_PATH_LEN) {
  280.         ipc_answer_0(callid, EINVAL);
  281.         ipc_answer_0(rid, EINVAL);
  282.         return;
  283.     }
  284.     /*
  285.      * Allocate buffer for the mount point data being received.
  286.      */
  287.     uint8_t *buf;
  288.     buf = malloc(size);
  289.     if (!buf) {
  290.         ipc_answer_0(callid, ENOMEM);
  291.         ipc_answer_0(rid, ENOMEM);
  292.         return;
  293.     }
  294.  
  295.     /*
  296.      * Deliver the mount point.
  297.      */
  298.     (void) ipc_data_write_finalize(callid, buf, size);
  299.  
  300.     /*
  301.      * Lookup the root node of the filesystem being mounted.
  302.      * In this case, we don't need to take the namespace_futex as the root
  303.      * node cannot be removed. However, we do take a reference to it so
  304.      * that we can track how many times it has been mounted.
  305.      */
  306.     int rc;
  307.     vfs_triplet_t mounted_root;
  308.     size_t mrsz;
  309.     rc = lookup_root(fs_handle, dev_handle, &mounted_root, &mrsz);
  310.     if (rc != EOK) {
  311.         free(buf);
  312.         ipc_answer_0(rid, rc);
  313.         return;
  314.     }
  315.     vfs_node_t *mr_node = vfs_node_get(&mounted_root, mrsz);
  316.     if (!mr_node) {
  317.         free(buf);
  318.         ipc_answer_0(rid, ENOMEM);
  319.         return;
  320.     }
  321.  
  322.     /*
  323.      * Finally, we need to resolve the path to the mountpoint.
  324.      */
  325.     vfs_triplet_t mp;
  326.     size_t mpsz;
  327.     futex_down(&rootfs_futex);
  328.     if (rootfs.fs_handle) {
  329.         /*
  330.          * We already have the root FS.
  331.          */
  332.         rwlock_write_lock(&namespace_rwlock);
  333.         rc = vfs_lookup_internal(buf, size, &mp, &mpsz, NULL);
  334.         if (rc != EOK) {
  335.             /*
  336.              * The lookup failed for some reason.
  337.              */
  338.             rwlock_write_unlock(&namespace_rwlock);
  339.             futex_up(&rootfs_futex);
  340.             vfs_node_put(mr_node);  /* failed -> drop reference */
  341.             free(buf);
  342.             ipc_answer_0(rid, rc);
  343.             return;
  344.         }
  345.         mp_node = vfs_node_get(&mp, mpsz);
  346.         if (!mp_node) {
  347.             rwlock_write_unlock(&namespace_rwlock);
  348.             futex_up(&rootfs_futex);
  349.             vfs_node_put(mr_node);  /* failed -> drop reference */
  350.             free(buf);
  351.             ipc_answer_0(rid, ENOMEM);
  352.             return;
  353.         }
  354.         /*
  355.          * Now we hold a reference to mp_node.
  356.          * It will be dropped upon the corresponding VFS_UNMOUNT.
  357.          * This prevents the mount point from being deleted.
  358.          */
  359.         rwlock_write_unlock(&namespace_rwlock);
  360.     } else {
  361.         /*
  362.          * We still don't have the root file system mounted.
  363.          */
  364.         if ((size == 1) && (buf[0] == '/')) {
  365.             /*
  366.              * For this simple, but important case, we are done.
  367.              */
  368.             rootfs = mounted_root;
  369.             futex_up(&rootfs_futex);
  370.             free(buf);
  371.             ipc_answer_0(rid, EOK);
  372.             return;
  373.         } else {
  374.             /*
  375.              * We can't resolve this without the root filesystem
  376.              * being mounted first.
  377.              */
  378.             futex_up(&rootfs_futex);
  379.             free(buf);
  380.             vfs_node_put(mr_node);  /* failed -> drop reference */
  381.             ipc_answer_0(rid, ENOENT);
  382.             return;
  383.         }
  384.     }
  385.     futex_up(&rootfs_futex);
  386.    
  387.     free(buf);  /* The buffer is not needed anymore. */
  388.    
  389.     /*
  390.      * At this point, we have all necessary pieces: file system and device
  391.      * handles, and we know the mount point VFS node and also the root node
  392.      * of the file system being mounted.
  393.      */
  394.  
  395.     int phone = vfs_grab_phone(mp.fs_handle);
  396.     /* Later we can use ARG3 to pass mode/flags. */
  397.     aid_t req1 = async_send_3(phone, VFS_MOUNT, (ipcarg_t) mp.dev_handle,
  398.         (ipcarg_t) mp.index, 0, NULL);
  399.     /* The second call uses the same method. */
  400.     aid_t req2 = async_send_3(phone, VFS_MOUNT,
  401.         (ipcarg_t) mounted_root.fs_handle,
  402.         (ipcarg_t) mounted_root.dev_handle, (ipcarg_t) mounted_root.index,
  403.         NULL);
  404.     vfs_release_phone(phone);
  405.  
  406.     ipcarg_t rc1;
  407.     ipcarg_t rc2;
  408.     async_wait_for(req1, &rc1);
  409.     async_wait_for(req2, &rc2);
  410.  
  411.     if ((rc1 != EOK) || (rc2 != EOK)) {
  412.         /* Mount failed, drop references to mr_node and mp_node. */
  413.         vfs_node_put(mr_node);
  414.         if (mp_node)
  415.             vfs_node_put(mp_node);
  416.     }
  417.    
  418.     if (rc2 == EOK)
  419.         ipc_answer_0(rid, rc1);
  420.     else if (rc1 == EOK)
  421.         ipc_answer_0(rid, rc2);
  422.     else
  423.         ipc_answer_0(rid, rc1);
  424. }
  425.  
  426. void vfs_open(ipc_callid_t rid, ipc_call_t *request)
  427. {
  428.     if (!vfs_files_init()) {
  429.         ipc_answer_0(rid, ENOMEM);
  430.         return;
  431.     }
  432.  
  433.     /*
  434.      * The POSIX interface is open(path, flags, mode).
  435.      * We can receive flags and mode along with the VFS_OPEN call; the path
  436.      * will need to arrive in another call.
  437.      */
  438.     int flags = IPC_GET_ARG1(*request);
  439.     int mode = IPC_GET_ARG2(*request);
  440.     size_t len;
  441.  
  442.     ipc_callid_t callid;
  443.  
  444.     if (!ipc_data_write_receive(&callid, &len)) {
  445.         ipc_answer_0(callid, EINVAL);
  446.         ipc_answer_0(rid, EINVAL);
  447.         return;
  448.     }
  449.  
  450.     /*
  451.      * Now we are on the verge of accepting the path.
  452.      *
  453.      * There is one optimization we could do in the future: copy the path
  454.      * directly into the PLB using some kind of a callback.
  455.      */
  456.     char *path = malloc(len);
  457.    
  458.     if (!path) {
  459.         ipc_answer_0(callid, ENOMEM);
  460.         ipc_answer_0(rid, ENOMEM);
  461.         return;
  462.     }
  463.  
  464.     int rc;
  465.     if ((rc = ipc_data_write_finalize(callid, path, len))) {
  466.         ipc_answer_0(rid, rc);
  467.         free(path);
  468.         return;
  469.     }
  470.    
  471.     /*
  472.      * Avoid the race condition in which the file can be deleted before we
  473.      * find/create-and-lock the VFS node corresponding to the looked-up
  474.      * triplet.
  475.      */
  476.     rwlock_read_lock(&namespace_rwlock);
  477.  
  478.     /*
  479.      * The path is now populated and we can call vfs_lookup_internal().
  480.      */
  481.     vfs_triplet_t triplet;
  482.     size_t size;
  483.     rc = vfs_lookup_internal(path, len, &triplet, &size, NULL);
  484.     if (rc) {
  485.         rwlock_read_unlock(&namespace_rwlock);
  486.         ipc_answer_0(rid, rc);
  487.         free(path);
  488.         return;
  489.     }
  490.  
  491.     /*
  492.      * Path is no longer needed.
  493.      */
  494.     free(path);
  495.  
  496.     vfs_node_t *node = vfs_node_get(&triplet, size);
  497.     rwlock_read_unlock(&namespace_rwlock);
  498.  
  499.     /*
  500.      * Get ourselves a file descriptor and the corresponding vfs_file_t
  501.      * structure.
  502.      */
  503.     int fd = vfs_fd_alloc();
  504.     if (fd < 0) {
  505.         vfs_node_put(node);
  506.         ipc_answer_0(rid, fd);
  507.         return;
  508.     }
  509.     vfs_file_t *file = vfs_file_get(fd);
  510.     file->node = node;
  511.  
  512.     /*
  513.      * The following increase in reference count is for the fact that the
  514.      * file is being opened and that a file structure is pointing to it.
  515.      * It is necessary so that the file will not disappear when
  516.      * vfs_node_put() is called. The reference will be dropped by the
  517.      * respective VFS_CLOSE.
  518.      */
  519.     vfs_node_addref(node);
  520.     vfs_node_put(node);
  521.  
  522.     /*
  523.      * Success! Return the new file descriptor to the client.
  524.      */
  525.     ipc_answer_1(rid, EOK, fd);
  526. }
  527.  
  528. static void vfs_rdwr(ipc_callid_t rid, ipc_call_t *request, bool read)
  529. {
  530.  
  531.     /*
  532.      * The following code strongly depends on the fact that the files data
  533.      * structure can be only accessed by a single fibril and all file
  534.      * operations are serialized (i.e. the reads and writes cannot
  535.      * interleave and a file cannot be closed while it is being read).
  536.      *
  537.      * Additional synchronization needs to be added once the table of
  538.      * open files supports parallel access!
  539.      */
  540.  
  541.     int fd = IPC_GET_ARG1(*request);
  542.  
  543.     /*
  544.      * Lookup the file structure corresponding to the file descriptor.
  545.      */
  546.     vfs_file_t *file = vfs_file_get(fd);
  547.     if (!file) {
  548.         ipc_answer_0(rid, ENOENT);
  549.         return;
  550.     }
  551.  
  552.     /*
  553.      * Now we need to receive a call with client's
  554.      * IPC_M_DATA_READ/IPC_M_DATA_WRITE request.
  555.      */
  556.     ipc_callid_t callid;
  557.     int res;
  558.     if (read)
  559.         res = ipc_data_read_receive(&callid, NULL);
  560.     else
  561.         res = ipc_data_write_receive(&callid, NULL);
  562.     if (!res) {
  563.         ipc_answer_0(callid, EINVAL);
  564.         ipc_answer_0(rid, EINVAL);
  565.         return;
  566.     }
  567.  
  568.     /*
  569.      * Lock the open file structure so that no other thread can manipulate
  570.      * the same open file at a time.
  571.      */
  572.     futex_down(&file->lock);
  573.  
  574.     /*
  575.      * Lock the file's node so that no other client can read/write to it at
  576.      * the same time.
  577.      */
  578.     if (read)
  579.         rwlock_read_lock(&file->node->contents_rwlock);
  580.     else
  581.         rwlock_write_lock(&file->node->contents_rwlock);
  582.  
  583.     int fs_phone = vfs_grab_phone(file->node->fs_handle);  
  584.    
  585.     /*
  586.      * Make a VFS_READ/VFS_WRITE request at the destination FS server.
  587.      */
  588.     aid_t msg;
  589.     ipc_call_t answer;
  590.     msg = async_send_3(fs_phone, IPC_GET_METHOD(*request),
  591.         file->node->dev_handle, file->node->index, file->pos, &answer);
  592.    
  593.     /*
  594.      * Forward the IPC_M_DATA_READ/IPC_M_DATA_WRITE request to the
  595.      * destination FS server. The call will be routed as if sent by
  596.      * ourselves. Note that call arguments are immutable in this case so we
  597.      * don't have to bother.
  598.      */
  599.     ipc_forward_fast(callid, fs_phone, 0, 0, 0, IPC_FF_ROUTE_FROM_ME);
  600.  
  601.     vfs_release_phone(fs_phone);
  602.  
  603.     /*
  604.      * Wait for reply from the FS server.
  605.      */
  606.     ipcarg_t rc;
  607.     async_wait_for(msg, &rc);
  608.     size_t bytes = IPC_GET_ARG1(answer);
  609.  
  610.     /*
  611.      * Unlock the VFS node.
  612.      */
  613.     if (read)
  614.         rwlock_read_unlock(&file->node->contents_rwlock);
  615.     else {
  616.         /* Update the cached version of node's size. */
  617.         file->node->size = IPC_GET_ARG2(answer);
  618.         rwlock_write_unlock(&file->node->contents_rwlock);
  619.     }
  620.  
  621.     /*
  622.      * Update the position pointer and unlock the open file.
  623.      */
  624.     file->pos += bytes;
  625.     futex_up(&file->lock);
  626.  
  627.     /*
  628.      * FS server's reply is the final result of the whole operation we
  629.      * return to the client.
  630.      */
  631.     ipc_answer_1(rid, rc, bytes);
  632. }
  633.  
  634. void vfs_read(ipc_callid_t rid, ipc_call_t *request)
  635. {
  636.     vfs_rdwr(rid, request, true);
  637. }
  638.  
  639. void vfs_write(ipc_callid_t rid, ipc_call_t *request)
  640. {
  641.     vfs_rdwr(rid, request, false);
  642. }
  643.  
  644. void vfs_seek(ipc_callid_t rid, ipc_call_t *request)
  645. {
  646.     int fd = (int) IPC_GET_ARG1(*request);
  647.     off_t off = (off_t) IPC_GET_ARG2(*request);
  648.     int whence = (int) IPC_GET_ARG3(*request);
  649.  
  650.  
  651.     /*
  652.      * Lookup the file structure corresponding to the file descriptor.
  653.      */
  654.     vfs_file_t *file = vfs_file_get(fd);
  655.     if (!file) {
  656.         ipc_answer_0(rid, ENOENT);
  657.         return;
  658.     }
  659.  
  660.     off_t newpos;
  661.     futex_down(&file->lock);
  662.     if (whence == SEEK_SET) {
  663.         file->pos = off;
  664.         futex_up(&file->lock);
  665.         ipc_answer_1(rid, EOK, off);
  666.         return;
  667.     }
  668.     if (whence == SEEK_CUR) {
  669.         if (file->pos + off < file->pos) {
  670.             futex_up(&file->lock);
  671.             ipc_answer_0(rid, EOVERFLOW);
  672.             return;
  673.         }
  674.         file->pos += off;
  675.         newpos = file->pos;
  676.         futex_up(&file->lock);
  677.         ipc_answer_1(rid, EOK, newpos);
  678.         return;
  679.     }
  680.     if (whence == SEEK_END) {
  681.         rwlock_read_lock(&file->node->contents_rwlock);
  682.         size_t size = file->node->size;
  683.         rwlock_read_unlock(&file->node->contents_rwlock);
  684.         if (size + off < size) {
  685.             futex_up(&file->lock);
  686.             ipc_answer_0(rid, EOVERFLOW);
  687.             return;
  688.         }
  689.         newpos = size + off;
  690.         futex_up(&file->lock);
  691.         ipc_answer_1(rid, EOK, newpos);
  692.         return;
  693.     }
  694.     futex_up(&file->lock);
  695.     ipc_answer_0(rid, EINVAL);
  696. }
  697.  
  698. atomic_t fs_head_futex = FUTEX_INITIALIZER;
  699. link_t fs_head;
  700.  
  701. atomic_t fs_handle_next = {
  702.     .count = 1
  703. };
  704.  
  705. /** Verify the VFS info structure.
  706.  *
  707.  * @param info      Info structure to be verified.
  708.  *
  709.  * @return      Non-zero if the info structure is sane, zero otherwise.
  710.  */
  711. static bool vfs_info_sane(vfs_info_t *info)
  712. {
  713.     int i;
  714.  
  715.     /*
  716.      * Check if the name is non-empty and is composed solely of ASCII
  717.      * characters [a-z]+[a-z0-9_-]*.
  718.      */
  719.     if (!islower(info->name[0])) {
  720.         dprintf("The name doesn't start with a lowercase character.\n");
  721.         return false;
  722.     }
  723.     for (i = 1; i < FS_NAME_MAXLEN; i++) {
  724.         if (!(islower(info->name[i]) || isdigit(info->name[i])) &&
  725.             (info->name[i] != '-') && (info->name[i] != '_')) {
  726.             if (info->name[i] == '\0') {
  727.                 break;
  728.             } else {
  729.                 dprintf("The name contains illegal "
  730.                     "characters.\n");
  731.                 return false;
  732.             }
  733.         }
  734.     }
  735.     /*
  736.      * This check is not redundant. It ensures that the name is
  737.      * NULL-terminated, even if FS_NAME_MAXLEN characters are used.
  738.      */
  739.     if (info->name[i] != '\0') {
  740.         dprintf("The name is not properly NULL-terminated.\n");
  741.         return false;
  742.     }
  743.    
  744.  
  745.     /*
  746.      * Check if the FS implements mandatory VFS operations.
  747.      */
  748.     if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_REGISTER)] != VFS_OP_DEFINED) {
  749.         dprintf("Operation VFS_REGISTER not defined by the client.\n");
  750.         return false;
  751.     }
  752.     if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_MOUNT)] != VFS_OP_DEFINED) {
  753.         dprintf("Operation VFS_MOUNT not defined by the client.\n");
  754.         return false;
  755.     }
  756.     if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_UNMOUNT)] != VFS_OP_DEFINED) {
  757.         dprintf("Operation VFS_UNMOUNT not defined by the client.\n");
  758.         return false;
  759.     }
  760.     if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_LOOKUP)] != VFS_OP_DEFINED) {
  761.         dprintf("Operation VFS_LOOKUP not defined by the client.\n");
  762.         return false;
  763.     }
  764.     if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_OPEN)] != VFS_OP_DEFINED) {
  765.         dprintf("Operation VFS_OPEN not defined by the client.\n");
  766.         return false;
  767.     }
  768.     if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_CLOSE)] != VFS_OP_DEFINED) {
  769.         dprintf("Operation VFS_CLOSE not defined by the client.\n");
  770.         return false;
  771.     }
  772.     if (info->ops[IPC_METHOD_TO_VFS_OP(VFS_READ)] != VFS_OP_DEFINED) {
  773.         dprintf("Operation VFS_READ not defined by the client.\n");
  774.         return false;
  775.     }
  776.    
  777.     /*
  778.      * Check if each operation is either not defined, defined or default.
  779.      */
  780.     for (i = VFS_FIRST; i < VFS_LAST; i++) {
  781.         if ((info->ops[IPC_METHOD_TO_VFS_OP(i)] != VFS_OP_NULL) &&
  782.             (info->ops[IPC_METHOD_TO_VFS_OP(i)] != VFS_OP_DEFAULT) &&
  783.             (info->ops[IPC_METHOD_TO_VFS_OP(i)] != VFS_OP_DEFINED)) {
  784.             dprintf("Operation info not understood.\n");
  785.             return false;
  786.         }
  787.     }
  788.     return true;
  789. }
  790.  
  791. /** VFS_REGISTER protocol function.
  792.  *
  793.  * @param rid       Hash of the call with the request.
  794.  * @param request   Call structure with the request.
  795.  */
  796. void vfs_register(ipc_callid_t rid, ipc_call_t *request)
  797. {
  798.     ipc_callid_t callid;
  799.     ipc_call_t call;
  800.     int rc;
  801.     size_t size;
  802.  
  803.     dprintf("Processing VFS_REGISTER request received from %p.\n",
  804.         request->in_phone_hash);
  805.  
  806.     /*
  807.      * The first call has to be IPC_M_DATA_SEND in which we receive the
  808.      * VFS info structure from the client FS.
  809.      */
  810.     if (!ipc_data_write_receive(&callid, &size)) {
  811.         /*
  812.          * The client doesn't obey the same protocol as we do.
  813.          */
  814.         dprintf("Receiving of VFS info failed.\n");
  815.         ipc_answer_0(callid, EINVAL);
  816.         ipc_answer_0(rid, EINVAL);
  817.         return;
  818.     }
  819.    
  820.     dprintf("VFS info received, size = %d\n", size);
  821.    
  822.     /*
  823.      * We know the size of the VFS info structure. See if the client
  824.      * understands this easy concept too.
  825.      */
  826.     if (size != sizeof(vfs_info_t)) {
  827.         /*
  828.          * The client is sending us something, which cannot be
  829.          * the info structure.
  830.          */
  831.         dprintf("Received VFS info has bad size.\n");
  832.         ipc_answer_0(callid, EINVAL);
  833.         ipc_answer_0(rid, EINVAL);
  834.         return;
  835.     }
  836.  
  837.     /*
  838.      * Allocate and initialize a buffer for the fs_info structure.
  839.      */
  840.     fs_info_t *fs_info;
  841.     fs_info = (fs_info_t *) malloc(sizeof(fs_info_t));
  842.     if (!fs_info) {
  843.         dprintf("Could not allocate memory for FS info.\n");
  844.         ipc_answer_0(callid, ENOMEM);
  845.         ipc_answer_0(rid, ENOMEM);
  846.         return;
  847.     }
  848.     link_initialize(&fs_info->fs_link);
  849.     futex_initialize(&fs_info->phone_futex, 1);
  850.        
  851.     rc = ipc_data_write_finalize(callid, &fs_info->vfs_info, size);
  852.     if (rc != EOK) {
  853.         dprintf("Failed to deliver the VFS info into our AS, rc=%d.\n",
  854.             rc);
  855.         free(fs_info);
  856.         ipc_answer_0(callid, rc);
  857.         ipc_answer_0(rid, rc);
  858.         return;
  859.     }
  860.  
  861.     dprintf("VFS info delivered.\n");
  862.        
  863.     if (!vfs_info_sane(&fs_info->vfs_info)) {
  864.         free(fs_info);
  865.         ipc_answer_0(callid, EINVAL);
  866.         ipc_answer_0(rid, EINVAL);
  867.         return;
  868.     }
  869.        
  870.     futex_down(&fs_head_futex);
  871.  
  872.     /*
  873.      * Check for duplicit registrations.
  874.      */
  875.     if (fs_name_to_handle(fs_info->vfs_info.name, false)) {
  876.         /*
  877.          * We already register a fs like this.
  878.          */
  879.         dprintf("FS is already registered.\n");
  880.         futex_up(&fs_head_futex);
  881.         free(fs_info);
  882.         ipc_answer_0(callid, EEXISTS);
  883.         ipc_answer_0(rid, EEXISTS);
  884.         return;
  885.     }
  886.  
  887.     /*
  888.      * Add fs_info to the list of registered FS's.
  889.      */
  890.     dprintf("Inserting FS into the list of registered file systems.\n");
  891.     list_append(&fs_info->fs_link, &fs_head);
  892.  
  893.     /*
  894.      * Now we want the client to send us the IPC_M_CONNECT_TO_ME call so
  895.      * that a callback connection is created and we have a phone through
  896.      * which to forward VFS requests to it.
  897.      */
  898.     callid = async_get_call(&call);
  899.     if (IPC_GET_METHOD(call) != IPC_M_CONNECT_TO_ME) {
  900.         dprintf("Unexpected call, method = %d\n", IPC_GET_METHOD(call));
  901.         list_remove(&fs_info->fs_link);
  902.         futex_up(&fs_head_futex);
  903.         free(fs_info);
  904.         ipc_answer_0(callid, EINVAL);
  905.         ipc_answer_0(rid, EINVAL);
  906.         return;
  907.     }
  908.     fs_info->phone = IPC_GET_ARG5(call);
  909.     ipc_answer_0(callid, EOK);
  910.  
  911.     dprintf("Callback connection to FS created.\n");
  912.  
  913.     /*
  914.      * The client will want us to send him the address space area with PLB.
  915.      */
  916.  
  917.     if (!ipc_share_in_receive(&callid, &size)) {
  918.         dprintf("Unexpected call, method = %d\n", IPC_GET_METHOD(call));
  919.         list_remove(&fs_info->fs_link);
  920.         futex_up(&fs_head_futex);
  921.         ipc_hangup(fs_info->phone);
  922.         free(fs_info);
  923.         ipc_answer_0(callid, EINVAL);
  924.         ipc_answer_0(rid, EINVAL);
  925.         return;
  926.     }
  927.    
  928.     /*
  929.      * We can only send the client address space area PLB_SIZE bytes long.
  930.      */
  931.     if (size != PLB_SIZE) {
  932.         dprintf("Client suggests wrong size of PFB, size = %d\n", size);
  933.         list_remove(&fs_info->fs_link);
  934.         futex_up(&fs_head_futex);
  935.         ipc_hangup(fs_info->phone);
  936.         free(fs_info);
  937.         ipc_answer_0(callid, EINVAL);
  938.         ipc_answer_0(rid, EINVAL);
  939.         return;
  940.     }
  941.  
  942.     /*
  943.      * Commit to read-only sharing the PLB with the client.
  944.      */
  945.     (void) ipc_share_in_finalize(callid, plb,
  946.         AS_AREA_READ | AS_AREA_CACHEABLE);
  947.  
  948.     dprintf("Sharing PLB.\n");
  949.  
  950.     /*
  951.      * That was it. The FS has been registered.
  952.      * In reply to the VFS_REGISTER request, we assign the client file
  953.      * system a global file system handle.
  954.      */
  955.     fs_info->fs_handle = (int) atomic_postinc(&fs_handle_next);
  956.     ipc_answer_1(rid, EOK, (ipcarg_t) fs_info->fs_handle);
  957.    
  958.     futex_up(&fs_head_futex);
  959.    
  960.     dprintf("\"%.*s\" filesystem successfully registered, handle=%d.\n",
  961.         FS_NAME_MAXLEN, fs_info->vfs_info.name, fs_info->fs_handle);
  962. }
  963.  
  964. /** For a given file system handle, implement policy for allocating a phone.
  965.  *
  966.  * @param handle    File system handle.
  967.  *
  968.  * @return      Phone over which a multi-call request can be safely
  969.  *          sent. Return 0 if no phone was found.
  970.  */
  971. int vfs_grab_phone(int handle)
  972. {
  973.     /*
  974.      * For now, we don't try to be very clever and very fast.
  975.      * We simply lookup the phone in the fs_head list. We currently don't
  976.      * open any additional phones (even though that itself would be pretty
  977.      * straightforward; housekeeping multiple open phones to a FS task would
  978.      * be more demanding). Instead, we simply take the respective
  979.      * phone_futex and keep it until vfs_release_phone().
  980.      */
  981.     futex_down(&fs_head_futex);
  982.     link_t *cur;
  983.     fs_info_t *fs;
  984.     for (cur = fs_head.next; cur != &fs_head; cur = cur->next) {
  985.         fs = list_get_instance(cur, fs_info_t, fs_link);
  986.         if (fs->fs_handle == handle) {
  987.             futex_up(&fs_head_futex);
  988.             /*
  989.              * For now, take the futex unconditionally.
  990.              * Oh yeah, serialization rocks.
  991.              * It will be up'ed in vfs_release_phone().
  992.              */
  993.             futex_down(&fs->phone_futex);
  994.             /*
  995.              * Avoid deadlock with other fibrils in the same thread
  996.              * by disabling fibril preemption.
  997.              */
  998.             fibril_inc_sercount();
  999.             return fs->phone;
  1000.         }
  1001.     }
  1002.     futex_up(&fs_head_futex);
  1003.     return 0;
  1004. }
  1005.  
  1006. /** Tell VFS that the phone is in use for any request.
  1007.  *
  1008.  * @param phone     Phone to FS task.
  1009.  */
  1010. void vfs_release_phone(int phone)
  1011. {
  1012.     bool found = false;
  1013.  
  1014.     /*
  1015.      * Undo the fibril_inc_sercount() done in vfs_grab_phone().
  1016.      */
  1017.     fibril_dec_sercount();
  1018.    
  1019.     futex_down(&fs_head_futex);
  1020.     link_t *cur;
  1021.     for (cur = fs_head.next; cur != &fs_head; cur = cur->next) {
  1022.         fs_info_t *fs = list_get_instance(cur, fs_info_t, fs_link);
  1023.         if (fs->phone == phone) {
  1024.             found = true;
  1025.             futex_up(&fs_head_futex);
  1026.             futex_up(&fs->phone_futex);
  1027.             return;
  1028.         }
  1029.     }
  1030.     futex_up(&fs_head_futex);
  1031.  
  1032.     /*
  1033.      * Not good to get here.
  1034.      */
  1035.     assert(found == true);
  1036. }
  1037.  
  1038. /** Convert file system name to its handle.
  1039.  *
  1040.  * @param name      File system name.
  1041.  * @param lock      If true, the function will down and up the
  1042.  *          fs_head_futex.
  1043.  *
  1044.  * @return      File system handle or zero if file system not found.
  1045.  */
  1046. int fs_name_to_handle(char *name, bool lock)
  1047. {
  1048.     int handle = 0;
  1049.    
  1050.     if (lock)
  1051.         futex_down(&fs_head_futex);
  1052.     link_t *cur;
  1053.     for (cur = fs_head.next; cur != &fs_head; cur = cur->next) {
  1054.         fs_info_t *fs = list_get_instance(cur, fs_info_t, fs_link);
  1055.         if (strncmp(fs->vfs_info.name, name,
  1056.             sizeof(fs->vfs_info.name)) == 0) {
  1057.             handle = fs->fs_handle;
  1058.             break;
  1059.         }
  1060.     }
  1061.     if (lock)
  1062.         futex_up(&fs_head_futex);
  1063.     return handle;
  1064. }
  1065.  
  1066. /**
  1067.  * @}
  1068.  */
  1069.