| 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 2 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, write to the Free Software
|
|---|
| 16 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
|
|---|
| 17 | */
|
|---|
| 18 |
|
|---|
| 19 | #include "includes.h"
|
|---|
| 20 |
|
|---|
| 21 | /* Cache priming module.
|
|---|
| 22 | *
|
|---|
| 23 | * The purpose of this module is to do RAID stripe width reads to prime the
|
|---|
| 24 | * buffer cache to do zero-copy I/O for subsequent sendfile calls. The idea is
|
|---|
| 25 | * to do a single large read at the start of the file to make sure that most or
|
|---|
| 26 | * all of the file is pulled into the buffer cache. Subsequent I/Os have
|
|---|
| 27 | * reduced latency.
|
|---|
| 28 | *
|
|---|
| 29 | * Tunables.
|
|---|
| 30 | *
|
|---|
| 31 | * cacheprime:rsize Amount of readahead in bytes. This should be a
|
|---|
| 32 | * multiple of the RAID stripe width.
|
|---|
| 33 | * cacheprime:debug Debug level at which to emit messages.
|
|---|
| 34 | */
|
|---|
| 35 |
|
|---|
| 36 | #define READAHEAD_MIN (128 * 1024) /* min is 128 KiB */
|
|---|
| 37 | #define READAHEAD_MAX (100 * 1024 * 1024) /* max is 100 MiB */
|
|---|
| 38 |
|
|---|
| 39 | #define MODULE "cacheprime"
|
|---|
| 40 |
|
|---|
| 41 | static int module_debug;
|
|---|
| 42 | static ssize_t g_readsz = 0;
|
|---|
| 43 | static void * g_readbuf = NULL;
|
|---|
| 44 |
|
|---|
| 45 | /* Prime the kernel buffer cache with data from the specified file. We use
|
|---|
| 46 | * per-fsp data to make sure we only ever do this once. If pread is being
|
|---|
| 47 | * emulated by seek/read/seek, when this will suck quite a lot.
|
|---|
| 48 | */
|
|---|
| 49 | static BOOL prime_cache(
|
|---|
| 50 | struct vfs_handle_struct * handle,
|
|---|
| 51 | files_struct * fsp,
|
|---|
| 52 | int fd,
|
|---|
| 53 | SMB_OFF_T offset,
|
|---|
| 54 | size_t count)
|
|---|
| 55 | {
|
|---|
| 56 | SMB_OFF_T * last;
|
|---|
| 57 | ssize_t nread;
|
|---|
| 58 |
|
|---|
| 59 | last = VFS_ADD_FSP_EXTENSION(handle, fsp, SMB_OFF_T);
|
|---|
| 60 | if (!last) {
|
|---|
| 61 | return False;
|
|---|
| 62 | }
|
|---|
| 63 |
|
|---|
| 64 | if (*last == -1) {
|
|---|
| 65 | /* Readahead disabled. */
|
|---|
| 66 | return False;
|
|---|
| 67 | }
|
|---|
| 68 |
|
|---|
| 69 | if ((*last + g_readsz) > (offset + count)) {
|
|---|
| 70 | /* Skip readahead ... we've already been here. */
|
|---|
| 71 | return False;
|
|---|
| 72 | }
|
|---|
| 73 |
|
|---|
| 74 | DEBUG(module_debug,
|
|---|
| 75 | ("%s: doing readahead of %lld bytes at %lld for %s\n",
|
|---|
| 76 | MODULE, (long long)g_readsz, (long long)*last,
|
|---|
| 77 | fsp->fsp_name));
|
|---|
| 78 |
|
|---|
| 79 | nread = sys_pread(fd, g_readbuf, g_readsz, *last);
|
|---|
| 80 | if (nread < 0) {
|
|---|
| 81 | *last = -1;
|
|---|
| 82 | return False;
|
|---|
| 83 | }
|
|---|
| 84 |
|
|---|
| 85 | *last += nread;
|
|---|
| 86 | return True;
|
|---|
| 87 | }
|
|---|
| 88 |
|
|---|
| 89 | static int cprime_connect(
|
|---|
| 90 | struct vfs_handle_struct * handle,
|
|---|
| 91 | const char * service,
|
|---|
| 92 | const char * user)
|
|---|
| 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 | g_readsz = conv_str_size(lp_parm_const_string(SNUM(handle->conn),
|
|---|
| 104 | MODULE, "rsize", NULL));
|
|---|
| 105 |
|
|---|
| 106 | if (g_readsz < READAHEAD_MIN) {
|
|---|
| 107 | DEBUG(module_debug, ("%s: %ld bytes of readahead "
|
|---|
| 108 | "requested, using minimum of %u\n",
|
|---|
| 109 | MODULE, (long)g_readsz, READAHEAD_MIN));
|
|---|
| 110 | g_readsz = READAHEAD_MIN;
|
|---|
| 111 | } else if (g_readsz > READAHEAD_MAX) {
|
|---|
| 112 | DEBUG(module_debug, ("%s: %ld bytes of readahead "
|
|---|
| 113 | "requested, using maximum of %u\n",
|
|---|
| 114 | MODULE, (long)g_readsz, READAHEAD_MAX));
|
|---|
| 115 | g_readsz = READAHEAD_MAX;
|
|---|
| 116 | }
|
|---|
| 117 |
|
|---|
| 118 | if ((g_readbuf = SMB_MALLOC(g_readsz)) == NULL) {
|
|---|
| 119 | /* Turn off readahead if we can't get a buffer. */
|
|---|
| 120 | g_readsz = 0;
|
|---|
| 121 | }
|
|---|
| 122 |
|
|---|
| 123 | return SMB_VFS_NEXT_CONNECT(handle, service, user);
|
|---|
| 124 | }
|
|---|
| 125 |
|
|---|
| 126 | static ssize_t cprime_sendfile(
|
|---|
| 127 | struct vfs_handle_struct * handle,
|
|---|
| 128 | int tofd,
|
|---|
| 129 | files_struct * fsp,
|
|---|
| 130 | int fromfd,
|
|---|
| 131 | const DATA_BLOB * header,
|
|---|
| 132 | SMB_OFF_T offset,
|
|---|
| 133 | size_t count)
|
|---|
| 134 | {
|
|---|
| 135 | if (g_readbuf && offset == 0) {
|
|---|
| 136 | prime_cache(handle, fsp, fromfd, offset, count);
|
|---|
| 137 | }
|
|---|
| 138 |
|
|---|
| 139 | return SMB_VFS_NEXT_SENDFILE(handle, tofd, fsp, fromfd,
|
|---|
| 140 | header, offset, count);
|
|---|
| 141 | }
|
|---|
| 142 |
|
|---|
| 143 | static ssize_t cprime_read(
|
|---|
| 144 | vfs_handle_struct * handle,
|
|---|
| 145 | files_struct * fsp,
|
|---|
| 146 | int fd,
|
|---|
| 147 | void * data,
|
|---|
| 148 | size_t count)
|
|---|
| 149 | {
|
|---|
| 150 | SMB_OFF_T offset;
|
|---|
| 151 |
|
|---|
| 152 | offset = SMB_VFS_LSEEK(fsp, fd, 0, SEEK_CUR);
|
|---|
| 153 | if (offset >= 0 && g_readbuf) {
|
|---|
| 154 | prime_cache(handle, fsp, fd, offset, count);
|
|---|
| 155 | SMB_VFS_LSEEK(fsp, fd, offset, SEEK_SET);
|
|---|
| 156 | }
|
|---|
| 157 |
|
|---|
| 158 | return SMB_VFS_NEXT_READ(handle, fsp, fd, data, count);
|
|---|
| 159 | }
|
|---|
| 160 |
|
|---|
| 161 | static ssize_t cprime_pread(
|
|---|
| 162 | vfs_handle_struct * handle,
|
|---|
| 163 | files_struct * fsp,
|
|---|
| 164 | int fd,
|
|---|
| 165 | void * data,
|
|---|
| 166 | size_t count,
|
|---|
| 167 | SMB_OFF_T offset)
|
|---|
| 168 | {
|
|---|
| 169 | if (g_readbuf) {
|
|---|
| 170 | prime_cache(handle, fsp, fd, offset, count);
|
|---|
| 171 | }
|
|---|
| 172 |
|
|---|
| 173 | return SMB_VFS_NEXT_PREAD(handle, fsp, fd, data, count, offset);
|
|---|
| 174 | }
|
|---|
| 175 |
|
|---|
| 176 | static vfs_op_tuple cprime_ops [] =
|
|---|
| 177 | {
|
|---|
| 178 | {SMB_VFS_OP(cprime_sendfile),
|
|---|
| 179 | SMB_VFS_OP_SENDFILE, SMB_VFS_LAYER_TRANSPARENT},
|
|---|
| 180 | {SMB_VFS_OP(cprime_pread),
|
|---|
| 181 | SMB_VFS_OP_PREAD, SMB_VFS_LAYER_TRANSPARENT},
|
|---|
| 182 | {SMB_VFS_OP(cprime_read),
|
|---|
| 183 | SMB_VFS_OP_READ, SMB_VFS_LAYER_TRANSPARENT},
|
|---|
| 184 | {SMB_VFS_OP(cprime_connect),
|
|---|
| 185 | SMB_VFS_OP_CONNECT, SMB_VFS_LAYER_TRANSPARENT},
|
|---|
| 186 |
|
|---|
| 187 | {SMB_VFS_OP(NULL), SMB_VFS_OP_NOOP, SMB_VFS_LAYER_NOOP}
|
|---|
| 188 | };
|
|---|
| 189 |
|
|---|
| 190 | /* -------------------------------------------------------------------------
|
|---|
| 191 | * Samba module initialisation entry point.
|
|---|
| 192 | * -------------------------------------------------------------------------
|
|---|
| 193 | */
|
|---|
| 194 |
|
|---|
| 195 | NTSTATUS vfs_cacheprime_init(void);
|
|---|
| 196 | NTSTATUS vfs_cacheprime_init(void)
|
|---|
| 197 | {
|
|---|
| 198 | return smb_register_vfs(SMB_VFS_INTERFACE_VERSION, MODULE, cprime_ops);
|
|---|
| 199 | }
|
|---|
| 200 |
|
|---|
| 201 | /* vim: set sw=4 ts=4 tw=79 et: */
|
|---|