| 1 | /*
 | 
|---|
| 2 |  * Copyright (c) James Peach 2005-2006
 | 
|---|
| 3 |  *
 | 
|---|
| 4 |  * This program is free software; you can redistribute it and/or modify
 | 
|---|
| 5 |  * it under the terms of the GNU General Public License as published by
 | 
|---|
| 6 |  * the Free Software Foundation; either version 3 of the License, or
 | 
|---|
| 7 |  * (at your option) any later version.
 | 
|---|
| 8 |  *
 | 
|---|
| 9 |  * This program is distributed in the hope that it will be useful,
 | 
|---|
| 10 |  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 | 
|---|
| 11 |  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 | 
|---|
| 12 |  * GNU General Public License for more details.
 | 
|---|
| 13 |  *
 | 
|---|
| 14 |  * You should have received a copy of the GNU General Public License
 | 
|---|
| 15 |  * along with this program; if not, see <http://www.gnu.org/licenses/>.
 | 
|---|
| 16 |  */
 | 
|---|
| 17 | 
 | 
|---|
| 18 | #include "includes.h"
 | 
|---|
| 19 | 
 | 
|---|
| 20 | /* Cache priming module.
 | 
|---|
| 21 |  *
 | 
|---|
| 22 |  * The purpose of this module is to do RAID stripe width reads to prime the
 | 
|---|
| 23 |  * buffer cache to do zero-copy I/O for subsequent sendfile calls. The idea is
 | 
|---|
| 24 |  * to do a single large read at the start of the file to make sure that most or
 | 
|---|
| 25 |  * all of the file is pulled into the buffer cache. Subsequent I/Os have
 | 
|---|
| 26 |  * reduced latency.
 | 
|---|
| 27 |  *
 | 
|---|
| 28 |  * Tunables.
 | 
|---|
| 29 |  *
 | 
|---|
| 30 |  *      cacheprime:rsize    Amount of readahead in bytes. This should be a
 | 
|---|
| 31 |  *                          multiple of the RAID stripe width.
 | 
|---|
| 32 |  *      cacheprime:debug    Debug level at which to emit messages.
 | 
|---|
| 33 |  */
 | 
|---|
| 34 | 
 | 
|---|
| 35 | #define READAHEAD_MIN       (128 * 1024)        /* min is 128 KiB */
 | 
|---|
| 36 | #define READAHEAD_MAX       (100 * 1024 * 1024) /* max is 100 MiB */
 | 
|---|
| 37 | 
 | 
|---|
| 38 | #define MODULE "cacheprime"
 | 
|---|
| 39 | 
 | 
|---|
| 40 | static int module_debug;
 | 
|---|
| 41 | static ssize_t g_readsz = 0;
 | 
|---|
| 42 | static void * g_readbuf = NULL;
 | 
|---|
| 43 | 
 | 
|---|
| 44 | /* Prime the kernel buffer cache with data from the specified file. We use
 | 
|---|
| 45 |  * per-fsp data to make sure we only ever do this once. If pread is being
 | 
|---|
| 46 |  * emulated by seek/read/seek, when this will suck quite a lot.
 | 
|---|
| 47 |  */
 | 
|---|
| 48 | static bool prime_cache(
 | 
|---|
| 49 |             struct vfs_handle_struct *  handle,
 | 
|---|
| 50 |                         files_struct *                  fsp,
 | 
|---|
| 51 |                         SMB_OFF_T                           offset,
 | 
|---|
| 52 |                         size_t                              count)
 | 
|---|
| 53 | {
 | 
|---|
| 54 |         SMB_OFF_T * last;
 | 
|---|
| 55 |         ssize_t nread;
 | 
|---|
| 56 | 
 | 
|---|
| 57 |         last = (SMB_OFF_T *)VFS_ADD_FSP_EXTENSION(handle, fsp, SMB_OFF_T, NULL);
 | 
|---|
| 58 |         if (!last) {
 | 
|---|
| 59 |                 return False;
 | 
|---|
| 60 |         }
 | 
|---|
| 61 | 
 | 
|---|
| 62 |         if (*last == -1) {
 | 
|---|
| 63 |             /* Readahead disabled. */
 | 
|---|
| 64 |             return False;
 | 
|---|
| 65 |         }
 | 
|---|
| 66 | 
 | 
|---|
| 67 |         if ((*last + g_readsz) > (offset + count)) {
 | 
|---|
| 68 |             /* Skip readahead ... we've already been here. */
 | 
|---|
| 69 |             return False;
 | 
|---|
| 70 |         }
 | 
|---|
| 71 | 
 | 
|---|
| 72 |         DEBUG(module_debug,
 | 
|---|
| 73 |             ("%s: doing readahead of %lld bytes at %lld for %s\n",
 | 
|---|
| 74 |             MODULE, (long long)g_readsz, (long long)*last,
 | 
|---|
| 75 |             fsp_str_dbg(fsp)));
 | 
|---|
| 76 | 
 | 
|---|
| 77 |         nread = sys_pread(fsp->fh->fd, g_readbuf, g_readsz, *last);
 | 
|---|
| 78 |         if (nread < 0) {
 | 
|---|
| 79 |             *last = -1;
 | 
|---|
| 80 |             return False;
 | 
|---|
| 81 |         }
 | 
|---|
| 82 | 
 | 
|---|
| 83 |         *last += nread;
 | 
|---|
| 84 |         return True;
 | 
|---|
| 85 | }
 | 
|---|
| 86 | 
 | 
|---|
| 87 | static int cprime_connect(
 | 
|---|
| 88 |                 struct vfs_handle_struct *  handle,
 | 
|---|
| 89 |                 const char *                service,
 | 
|---|
| 90 |                 const char *                user)
 | 
|---|
| 91 | {
 | 
|---|
| 92 |         int ret;
 | 
|---|
| 93 | 
 | 
|---|
| 94 |         module_debug = lp_parm_int(SNUM(handle->conn), MODULE, "debug", 100);
 | 
|---|
| 95 |         if (g_readbuf) {
 | 
|---|
| 96 |                 /* Only allocate g_readbuf once. If the config changes and
 | 
|---|
| 97 |                  * another client multiplexes onto this smbd, we don't want
 | 
|---|
| 98 |                  * to risk memory corruption.
 | 
|---|
| 99 |                  */
 | 
|---|
| 100 |                 return SMB_VFS_NEXT_CONNECT(handle, service, user);
 | 
|---|
| 101 |         }
 | 
|---|
| 102 | 
 | 
|---|
| 103 |         ret = SMB_VFS_NEXT_CONNECT(handle, service, user);
 | 
|---|
| 104 |         if (ret < 0) {
 | 
|---|
| 105 |                 return ret;
 | 
|---|
| 106 |         }
 | 
|---|
| 107 | 
 | 
|---|
| 108 |         g_readsz = conv_str_size(lp_parm_const_string(SNUM(handle->conn),
 | 
|---|
| 109 |                                         MODULE, "rsize", NULL));
 | 
|---|
| 110 | 
 | 
|---|
| 111 |         if (g_readsz < READAHEAD_MIN) {
 | 
|---|
| 112 |                 DEBUG(module_debug, ("%s: %ld bytes of readahead "
 | 
|---|
| 113 |                             "requested, using minimum of %u\n",
 | 
|---|
| 114 |                             MODULE, (long)g_readsz, READAHEAD_MIN));
 | 
|---|
| 115 |                 g_readsz = READAHEAD_MIN;
 | 
|---|
| 116 |         } else if (g_readsz > READAHEAD_MAX) {
 | 
|---|
| 117 |                 DEBUG(module_debug, ("%s: %ld bytes of readahead "
 | 
|---|
| 118 |                             "requested, using maximum of %u\n",
 | 
|---|
| 119 |                             MODULE, (long)g_readsz, READAHEAD_MAX));
 | 
|---|
| 120 |                 g_readsz = READAHEAD_MAX;
 | 
|---|
| 121 |         }
 | 
|---|
| 122 | 
 | 
|---|
| 123 |         if ((g_readbuf = SMB_MALLOC(g_readsz)) == NULL) {
 | 
|---|
| 124 |                 /* Turn off readahead if we can't get a buffer. */
 | 
|---|
| 125 |                 g_readsz = 0;
 | 
|---|
| 126 |         }
 | 
|---|
| 127 | 
 | 
|---|
| 128 |         return 0;
 | 
|---|
| 129 | }
 | 
|---|
| 130 | 
 | 
|---|
| 131 | static ssize_t cprime_sendfile(
 | 
|---|
| 132 |                 struct vfs_handle_struct *  handle,
 | 
|---|
| 133 |                 int                         tofd,
 | 
|---|
| 134 |                 files_struct *              fromfsp,
 | 
|---|
| 135 |                 const DATA_BLOB *           header,
 | 
|---|
| 136 |                 SMB_OFF_T                   offset,
 | 
|---|
| 137 |                 size_t                      count)
 | 
|---|
| 138 | {
 | 
|---|
| 139 |         if (g_readbuf && offset == 0) {
 | 
|---|
| 140 |                 prime_cache(handle, fromfsp, offset, count);
 | 
|---|
| 141 |         }
 | 
|---|
| 142 | 
 | 
|---|
| 143 |         return SMB_VFS_NEXT_SENDFILE(handle, tofd, fromfsp,
 | 
|---|
| 144 |                                      header, offset, count);
 | 
|---|
| 145 | }
 | 
|---|
| 146 | 
 | 
|---|
| 147 | static ssize_t cprime_read(
 | 
|---|
| 148 |                 vfs_handle_struct * handle,
 | 
|---|
| 149 |                 files_struct *      fsp,
 | 
|---|
| 150 |                 void *              data,
 | 
|---|
| 151 |                 size_t              count)
 | 
|---|
| 152 | {
 | 
|---|
| 153 |         SMB_OFF_T offset;
 | 
|---|
| 154 | 
 | 
|---|
| 155 |         offset = SMB_VFS_LSEEK(fsp, 0, SEEK_CUR);
 | 
|---|
| 156 |         if (offset >= 0 && g_readbuf)  {
 | 
|---|
| 157 |                 prime_cache(handle, fsp, offset, count);
 | 
|---|
| 158 |                 SMB_VFS_LSEEK(fsp, offset, SEEK_SET);
 | 
|---|
| 159 |         }
 | 
|---|
| 160 | 
 | 
|---|
| 161 |         return SMB_VFS_NEXT_READ(handle, fsp, data, count);
 | 
|---|
| 162 | }
 | 
|---|
| 163 | 
 | 
|---|
| 164 | static ssize_t cprime_pread(
 | 
|---|
| 165 |                 vfs_handle_struct * handle,
 | 
|---|
| 166 |                 files_struct *      fsp,
 | 
|---|
| 167 |                 void *              data,
 | 
|---|
| 168 |                         size_t              count,
 | 
|---|
| 169 |                 SMB_OFF_T           offset)
 | 
|---|
| 170 | {
 | 
|---|
| 171 |         if (g_readbuf) {
 | 
|---|
| 172 |                 prime_cache(handle, fsp, offset, count);
 | 
|---|
| 173 |         }
 | 
|---|
| 174 | 
 | 
|---|
| 175 |         return SMB_VFS_NEXT_PREAD(handle, fsp, data, count, offset);
 | 
|---|
| 176 | }
 | 
|---|
| 177 | 
 | 
|---|
| 178 | static struct vfs_fn_pointers vfs_cacheprime_fns = {
 | 
|---|
| 179 |         .sendfile = cprime_sendfile,
 | 
|---|
| 180 |         .pread = cprime_pread,
 | 
|---|
| 181 |         .vfs_read = cprime_read,
 | 
|---|
| 182 |         .connect_fn = cprime_connect,
 | 
|---|
| 183 | };
 | 
|---|
| 184 | 
 | 
|---|
| 185 | /* -------------------------------------------------------------------------
 | 
|---|
| 186 |  * Samba module initialisation entry point.
 | 
|---|
| 187 |  * -------------------------------------------------------------------------
 | 
|---|
| 188 |  */
 | 
|---|
| 189 | 
 | 
|---|
| 190 | NTSTATUS vfs_cacheprime_init(void);
 | 
|---|
| 191 | NTSTATUS vfs_cacheprime_init(void)
 | 
|---|
| 192 | {
 | 
|---|
| 193 |         return smb_register_vfs(SMB_VFS_INTERFACE_VERSION, MODULE,
 | 
|---|
| 194 |                                 &vfs_cacheprime_fns);
 | 
|---|
| 195 | }
 | 
|---|
| 196 | 
 | 
|---|
| 197 | /* vim: set sw=4 ts=4 tw=79 et: */
 | 
|---|