| 1 |
/* Main program and global state for tmpfs. |
| 2 |
Copyright (C) 2000,01,02 Free Software Foundation, Inc. |
| 3 |
|
| 4 |
This file is part of the GNU Hurd. |
| 5 |
|
| 6 |
The GNU Hurd is free software; you can redistribute it and/or modify |
| 7 |
it under the terms of the GNU General Public License as published by |
| 8 |
the Free Software Foundation; either version 2, or (at your option) |
| 9 |
any later version. |
| 10 |
|
| 11 |
The GNU Hurd is distributed in the hope that it will be useful, |
| 12 |
but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 |
GNU General Public License for more details. |
| 15 |
|
| 16 |
You should have received a copy of the GNU General Public License |
| 17 |
along with the GNU Hurd; see the file COPYING. If not, write to |
| 18 |
the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */ |
| 19 |
|
| 20 |
#include <argp.h> |
| 21 |
#include <argz.h> |
| 22 |
#include <string.h> |
| 23 |
#include <inttypes.h> |
| 24 |
#include <error.h> |
| 25 |
#include <pthread.h> |
| 26 |
|
| 27 |
#include "tmpfs.h" |
| 28 |
#include <limits.h> |
| 29 |
#include <version.h> |
| 30 |
#include <fcntl.h> |
| 31 |
#include <hurd.h> |
| 32 |
#include <hurd/paths.h> |
| 33 |
|
| 34 |
char *diskfs_server_name = "tmpfs"; |
| 35 |
char *diskfs_server_version = HURD_VERSION; |
| 36 |
char *diskfs_extra_version = "GNU Hurd"; |
| 37 |
char *diskfs_disk_name = "none"; |
| 38 |
|
| 39 |
/* We ain't got to show you no stinkin' sync'ing. */ |
| 40 |
int diskfs_default_sync_interval = 0; |
| 41 |
|
| 42 |
/* We must supply some claimed limits, though we don't impose any new ones. */ |
| 43 |
int diskfs_link_max = INT_MAX; |
| 44 |
int diskfs_name_max = 255; /* dirent d_namlen limit */ |
| 45 |
int diskfs_maxsymlinks = 8; |
| 46 |
|
| 47 |
/* Yeah, baby, we do it all! */ |
| 48 |
int diskfs_shortcut_symlink = 1; |
| 49 |
int diskfs_shortcut_chrdev = 1; |
| 50 |
int diskfs_shortcut_blkdev = 1; |
| 51 |
int diskfs_shortcut_fifo = 1; |
| 52 |
int diskfs_shortcut_ifsock = 1; |
| 53 |
|
| 54 |
struct node *diskfs_root_node; |
| 55 |
mach_port_t default_pager; |
| 56 |
|
| 57 |
off_t tmpfs_page_limit, tmpfs_space_used; |
| 58 |
mode_t tmpfs_root_mode = -1; |
| 59 |
|
| 60 |
error_t |
| 61 |
diskfs_set_statfs (struct statfs *st) |
| 62 |
{ |
| 63 |
fsblkcnt_t pages; |
| 64 |
|
| 65 |
st->f_type = FSTYPE_MEMFS; |
| 66 |
st->f_fsid = getpid (); |
| 67 |
|
| 68 |
st->f_bsize = vm_page_size; |
| 69 |
st->f_blocks = tmpfs_page_limit; |
| 70 |
|
| 71 |
st->f_files = __atomic_load_n (&num_files, __ATOMIC_RELAXED); |
| 72 |
pages = round_page (get_used ()) / vm_page_size; |
| 73 |
|
| 74 |
st->f_bfree = pages < tmpfs_page_limit ? tmpfs_page_limit - pages : 0; |
| 75 |
st->f_bavail = st->f_bfree; |
| 76 |
st->f_ffree = st->f_bavail / sizeof (struct disknode); /* Well, sort of. */ |
| 77 |
|
| 78 |
return 0; |
| 79 |
} |
| 80 |
|
| 81 |
|
| 82 |
error_t |
| 83 |
diskfs_set_hypermetadata (int wait, int clean) |
| 84 |
{ |
| 85 |
/* All the state always just lives in core, so we have nothing to do. */ |
| 86 |
return 0; |
| 87 |
} |
| 88 |
|
| 89 |
void |
| 90 |
diskfs_sync_everything (int wait) |
| 91 |
{ |
| 92 |
} |
| 93 |
|
| 94 |
error_t |
| 95 |
diskfs_reload_global_state (void) |
| 96 |
{ |
| 97 |
return 0; |
| 98 |
} |
| 99 |
|
| 100 |
int diskfs_synchronous = 0; |
| 101 |
|
| 102 |
#define OPT_SIZE 600 /* --size */ |
| 103 |
|
| 104 |
static const struct argp_option options[] = |
| 105 |
{ |
| 106 |
{"mode", 'm', "MODE", 0, "Permissions (octal) for root directory"}, |
| 107 |
{"size", OPT_SIZE, "MAX-BYTES", 0, "Maximum size"}, |
| 108 |
{NULL,} |
| 109 |
}; |
| 110 |
|
| 111 |
struct option_values |
| 112 |
{ |
| 113 |
off_t size; |
| 114 |
mode_t mode; |
| 115 |
}; |
| 116 |
|
| 117 |
/* Parse the size string ARG, and set *NEWSIZE with the resulting size. */ |
| 118 |
static error_t |
| 119 |
parse_opt_size (const char *arg, struct argp_state *state, off_t *newsize) |
| 120 |
{ |
| 121 |
char *end = NULL; |
| 122 |
intmax_t size = strtoimax (arg, &end, 0); |
| 123 |
if (end == NULL || end == arg) |
| 124 |
{ |
| 125 |
argp_error (state, "argument must be a number"); |
| 126 |
return EINVAL; |
| 127 |
} |
| 128 |
if (size < 0) |
| 129 |
{ |
| 130 |
argp_error (state, "negative size not meaningful"); |
| 131 |
return EINVAL; |
| 132 |
} |
| 133 |
switch (*end) |
| 134 |
{ |
| 135 |
case 'g': |
| 136 |
case 'G': |
| 137 |
size <<= 10; |
| 138 |
case 'm': |
| 139 |
case 'M': |
| 140 |
size <<= 10; |
| 141 |
case 'k': |
| 142 |
case 'K': |
| 143 |
size <<= 10; |
| 144 |
break; |
| 145 |
case '%': |
| 146 |
{ |
| 147 |
/* Set as a percentage of the machine's physical memory. */ |
| 148 |
struct vm_statistics vmstats; |
| 149 |
error_t err = vm_statistics (mach_task_self (), &vmstats); |
| 150 |
if (err) |
| 151 |
{ |
| 152 |
argp_error (state, "cannot find total physical memory: %s", |
| 153 |
strerror (err)); |
| 154 |
return err; |
| 155 |
} |
| 156 |
size = round_page ((((vmstats.free_count |
| 157 |
+ vmstats.active_count |
| 158 |
+ vmstats.inactive_count |
| 159 |
+ vmstats.wire_count) |
| 160 |
* vm_page_size) |
| 161 |
* size + 99) / 100); |
| 162 |
break; |
| 163 |
} |
| 164 |
} |
| 165 |
size = (off_t) size; |
| 166 |
if (size < 0) |
| 167 |
{ |
| 168 |
argp_error (state, "size too large"); |
| 169 |
return EINVAL; |
| 170 |
} |
| 171 |
|
| 172 |
*newsize = size; |
| 173 |
|
| 174 |
return 0; |
| 175 |
} |
| 176 |
|
| 177 |
/* Parse a command line option. */ |
| 178 |
static error_t |
| 179 |
parse_opt (int key, char *arg, struct argp_state *state) |
| 180 |
{ |
| 181 |
/* We save our parsed values in this structure, hung off STATE->hook. |
| 182 |
Only after parsing all options successfully will we use these values. */ |
| 183 |
struct option_values *values = state->hook; |
| 184 |
|
| 185 |
switch (key) |
| 186 |
{ |
| 187 |
case ARGP_KEY_INIT: |
| 188 |
state->child_inputs[0] = state->input; |
| 189 |
values = malloc (sizeof *values); |
| 190 |
if (values == 0) |
| 191 |
return ENOMEM; |
| 192 |
state->hook = values; |
| 193 |
values->size = -1; |
| 194 |
values->mode = -1; |
| 195 |
break; |
| 196 |
case ARGP_KEY_FINI: |
| 197 |
free (values); |
| 198 |
state->hook = 0; |
| 199 |
break; |
| 200 |
|
| 201 |
case 'm': /* --mode=OCTAL */ |
| 202 |
{ |
| 203 |
char *end = NULL; |
| 204 |
mode_t mode = strtoul (arg, &end, 8); |
| 205 |
if (end == NULL || end == arg) |
| 206 |
{ |
| 207 |
argp_error (state, "argument must be an octal number"); |
| 208 |
return EINVAL; |
| 209 |
} |
| 210 |
if (mode & S_IFMT) |
| 211 |
{ |
| 212 |
argp_error (state, "invalid bits in mode"); |
| 213 |
return EINVAL; |
| 214 |
} |
| 215 |
values->mode = mode; |
| 216 |
} |
| 217 |
break; |
| 218 |
|
| 219 |
case OPT_SIZE: /* --size=MAX-BYTES */ |
| 220 |
{ |
| 221 |
error_t err = parse_opt_size (arg, state, &values->size); |
| 222 |
if (err) |
| 223 |
return err; |
| 224 |
} |
| 225 |
break; |
| 226 |
|
| 227 |
case ARGP_KEY_NO_ARGS: |
| 228 |
if (values->size < 0) |
| 229 |
{ |
| 230 |
argp_error (state, "must supply maximum size"); |
| 231 |
return EINVAL; |
| 232 |
} |
| 233 |
break; |
| 234 |
|
| 235 |
case ARGP_KEY_ARGS: |
| 236 |
if (state->argv[state->next + 1] != 0) |
| 237 |
{ |
| 238 |
argp_error (state, "too many arguments"); |
| 239 |
return EINVAL; |
| 240 |
} |
| 241 |
else if (values->size >= 0) |
| 242 |
{ |
| 243 |
if (strcmp (state->argv[state->next], "tmpfs") != 0) |
| 244 |
{ |
| 245 |
argp_error (state, "size specified with --size and argument is not \"tmpfs\""); |
| 246 |
return EINVAL; |
| 247 |
} |
| 248 |
} |
| 249 |
else |
| 250 |
{ |
| 251 |
error_t err = parse_opt_size (state->argv[state->next], state, |
| 252 |
&values->size); |
| 253 |
if (err) |
| 254 |
return err; |
| 255 |
} |
| 256 |
break; |
| 257 |
|
| 258 |
case ARGP_KEY_SUCCESS: |
| 259 |
/* All options parse successfully, so implement ours if possible. */ |
| 260 |
tmpfs_page_limit = values->size / vm_page_size; |
| 261 |
tmpfs_root_mode = values->mode; |
| 262 |
break; |
| 263 |
|
| 264 |
default: |
| 265 |
return ARGP_ERR_UNKNOWN; |
| 266 |
} |
| 267 |
return 0; |
| 268 |
} |
| 269 |
|
| 270 |
/* Override the standard diskfs routine so we can add our own output. */ |
| 271 |
error_t |
| 272 |
diskfs_append_args (char **argz, size_t *argz_len) |
| 273 |
{ |
| 274 |
error_t err; |
| 275 |
|
| 276 |
/* Get the standard things. */ |
| 277 |
err = diskfs_append_std_options (argz, argz_len); |
| 278 |
|
| 279 |
if (!err) |
| 280 |
{ |
| 281 |
off_t lim = tmpfs_page_limit * vm_page_size; |
| 282 |
char buf[100], sfx; |
| 283 |
#define S(n, c) if ((lim & ((1 << n) - 1)) == 0) sfx = c, lim >>= n |
| 284 |
S (30, 'G'); else S (20, 'M'); else S (10, 'K'); else sfx = '\0'; |
| 285 |
#undef S |
| 286 |
snprintf (buf, sizeof buf, "%" PRIi64 "%c", lim, sfx); |
| 287 |
err = argz_add (argz, argz_len, buf); |
| 288 |
} |
| 289 |
|
| 290 |
return err; |
| 291 |
} |
| 292 |
|
| 293 |
/* Handling of operations for the ports in diskfs_port_bucket, calling |
| 294 |
* demuxer for each incoming message */ |
| 295 |
static void * |
| 296 |
diskfs_thread_function (void *demuxer) |
| 297 |
{ |
| 298 |
static int thread_timeout = 1000 * 60 * 2; /* two minutes */ |
| 299 |
error_t err; |
| 300 |
|
| 301 |
pthread_setname_np (pthread_self (), "diskfs"); |
| 302 |
|
| 303 |
do |
| 304 |
{ |
| 305 |
ports_manage_port_operations_multithread (diskfs_port_bucket, |
| 306 |
(ports_demuxer_type) demuxer, |
| 307 |
thread_timeout, |
| 308 |
0, |
| 309 |
0); |
| 310 |
err = diskfs_shutdown (0); |
| 311 |
} |
| 312 |
while (err); |
| 313 |
|
| 314 |
exit (0); |
| 315 |
/* NOTREACHED */ |
| 316 |
return NULL; |
| 317 |
} |
| 318 |
|
| 319 |
|
| 320 |
/* Add our startup arguments to the standard diskfs set. */ |
| 321 |
static const struct argp_child startup_children[] = |
| 322 |
{{&diskfs_startup_argp}, {0}}; |
| 323 |
static struct argp startup_argp = {options, parse_opt, "MAX-BYTES", "\ |
| 324 |
\v\ |
| 325 |
MAX-BYTES may be followed by k or K for kilobytes,\n\ |
| 326 |
m or M for megabytes, g or G for gigabytes.", |
| 327 |
startup_children}; |
| 328 |
|
| 329 |
/* Similarly at runtime. */ |
| 330 |
static const struct argp_child runtime_children[] = |
| 331 |
{{&diskfs_std_runtime_argp}, {0}}; |
| 332 |
static struct argp runtime_argp = {options, parse_opt, 0, 0, runtime_children}; |
| 333 |
|
| 334 |
struct argp *diskfs_runtime_argp = (struct argp *)&runtime_argp; |
| 335 |
|
| 336 |
|
| 337 |
|
| 338 |
int |
| 339 |
main (int argc, char **argv) |
| 340 |
{ |
| 341 |
error_t err; |
| 342 |
mach_port_t bootstrap, realnode, host_priv; |
| 343 |
pthread_t pthread_id; |
| 344 |
struct stat st; |
| 345 |
|
| 346 |
err = argp_parse (&startup_argp, argc, argv, ARGP_IN_ORDER, NULL, NULL); |
| 347 |
assert_perror_backtrace (err); |
| 348 |
|
| 349 |
task_get_bootstrap_port (mach_task_self (), &bootstrap); |
| 350 |
if (bootstrap == MACH_PORT_NULL) |
| 351 |
error (2, 0, "Must be started as a translator"); |
| 352 |
|
| 353 |
/* Get our port to the default pager. Without that, |
| 354 |
we have no place to put file contents. */ |
| 355 |
err = get_privileged_ports (&host_priv, NULL); |
| 356 |
if (err == EPERM) |
| 357 |
{ |
| 358 |
default_pager = file_name_lookup (_SERVERS_DEFPAGER, O_EXEC, 0); |
| 359 |
if (default_pager == MACH_PORT_NULL) |
| 360 |
error (0, errno, _SERVERS_DEFPAGER); |
| 361 |
} |
| 362 |
else if (err) |
| 363 |
error (0, err, "Cannot get host privileged port"); |
| 364 |
else |
| 365 |
{ |
| 366 |
err = vm_set_default_memory_manager (host_priv, &default_pager); |
| 367 |
if (err) |
| 368 |
error (0, err, "Cannot get default pager port"); |
| 369 |
else |
| 370 |
if (default_pager == MACH_PORT_NULL) |
| 371 |
{ |
| 372 |
error (0, 0, "No default pager (memory manager) is running"); |
| 373 |
/* Try to auto-start it. */ |
| 374 |
err = system ("/hurd/mach-defpager"); |
| 375 |
if (err) |
| 376 |
error (0, err, "Could not start it"); |
| 377 |
else |
| 378 |
{ |
| 379 |
fprintf (stderr, "Started it\n"); |
| 380 |
err = vm_set_default_memory_manager (host_priv, &default_pager); |
| 381 |
if (err) |
| 382 |
error(0, err, "Cannot get default pager port"); |
| 383 |
} |
| 384 |
} |
| 385 |
mach_port_deallocate (mach_task_self (), host_priv); |
| 386 |
} |
| 387 |
if (default_pager == MACH_PORT_NULL) |
| 388 |
error (0, 0, "files cannot have contents with no default pager port"); |
| 389 |
|
| 390 |
/* Initialize the diskfs library. Must come before any other diskfs call. */ |
| 391 |
err = diskfs_init_diskfs (); |
| 392 |
if (err) |
| 393 |
error (4, err, "init"); |
| 394 |
|
| 395 |
err = diskfs_alloc_node (0, S_IFDIR, &diskfs_root_node); |
| 396 |
if (err) |
| 397 |
error (4, err, "cannot create root directory"); |
| 398 |
|
| 399 |
/* Like diskfs_spawn_first_thread. But do it manually, without timeout */ |
| 400 |
err = pthread_create (&pthread_id, NULL, diskfs_thread_function, |
| 401 |
diskfs_demuxer); |
| 402 |
if (!err) |
| 403 |
pthread_detach (pthread_id); |
| 404 |
else |
| 405 |
{ |
| 406 |
errno = err; |
| 407 |
perror ("pthread_create"); |
| 408 |
} |
| 409 |
|
| 410 |
/* Now that we are all set up to handle requests, and diskfs_root_node is |
| 411 |
set properly, it is safe to export our fsys control port to the |
| 412 |
outside world. */ |
| 413 |
realnode = diskfs_startup_diskfs (bootstrap, 0); |
| 414 |
diskfs_root_node->dn_stat.st_mode = S_IFDIR; |
| 415 |
|
| 416 |
/* Propagate permissions, owner, etc. from underlying node to |
| 417 |
the root directory of the new (empty) filesystem. */ |
| 418 |
err = io_stat (realnode, &st); |
| 419 |
if (err) |
| 420 |
{ |
| 421 |
error (0, err, "cannot stat underlying node"); |
| 422 |
if (tmpfs_root_mode == -1) |
| 423 |
diskfs_root_node->dn_stat.st_mode |= 0777 | S_ISVTX; |
| 424 |
else |
| 425 |
diskfs_root_node->dn_stat.st_mode |= tmpfs_root_mode; |
| 426 |
diskfs_root_node->dn_set_ctime = 1; |
| 427 |
diskfs_root_node->dn_set_mtime = 1; |
| 428 |
diskfs_root_node->dn_set_atime = 1; |
| 429 |
} |
| 430 |
else |
| 431 |
{ |
| 432 |
if (tmpfs_root_mode == -1) |
| 433 |
{ |
| 434 |
diskfs_root_node->dn_stat.st_mode |= st.st_mode &~ S_IFMT; |
| 435 |
if (S_ISREG (st.st_mode) && (st.st_mode & 0111) == 0) |
| 436 |
/* There are no execute bits set, as by default on a plain file. |
| 437 |
For the virtual directory, set execute bits where read bits are |
| 438 |
set on the underlying plain file. */ |
| 439 |
diskfs_root_node->dn_stat.st_mode |= (st.st_mode & 0444) >> 2; |
| 440 |
} |
| 441 |
else |
| 442 |
diskfs_root_node->dn_stat.st_mode |= tmpfs_root_mode; |
| 443 |
diskfs_root_node->dn_stat.st_uid = st.st_uid; |
| 444 |
diskfs_root_node->dn_stat.st_author = st.st_author; |
| 445 |
diskfs_root_node->dn_stat.st_gid = st.st_gid; |
| 446 |
diskfs_root_node->dn_stat.st_atim = st.st_atim; |
| 447 |
diskfs_root_node->dn_stat.st_mtim = st.st_mtim; |
| 448 |
diskfs_root_node->dn_stat.st_ctim = st.st_ctim; |
| 449 |
diskfs_root_node->dn_stat.st_flags = st.st_flags; |
| 450 |
} |
| 451 |
diskfs_root_node->dn_stat.st_mode &= ~S_ITRANS; |
| 452 |
diskfs_root_node->dn_stat.st_mode |= S_IROOT; |
| 453 |
diskfs_root_node->dn_stat.st_nlink = 2; |
| 454 |
|
| 455 |
/* We must keep the REALNODE send right to remain the active |
| 456 |
translator for the underlying node. */ |
| 457 |
|
| 458 |
pthread_mutex_unlock (&diskfs_root_node->lock); |
| 459 |
|
| 460 |
/* and so we die, leaving others to do the real work. */ |
| 461 |
pthread_exit (NULL); |
| 462 |
/* NOTREACHED */ |
| 463 |
return 0; |
| 464 |
} |