module/spl/spl-generic.c

   1 /*
   2  *  This file is part of the SPL: Solaris Porting Layer.
   3  *
   4  *  Copyright (c) 2008 Lawrence Livermore National Security, LLC.
   5  *  Produced at Lawrence Livermore National Laboratory
   6  *  Written by:
   7  *          Brian Behlendorf <behlendorf1@llnl.gov>,
   8  *          Herb Wartens <wartens2@llnl.gov>,
   9  *          Jim Garlick <garlick@llnl.gov>
  10  *  UCRL-CODE-235197
  11  *
  12  *  This is free software; you can redistribute it and/or modify it
  13  *  under the terms of the GNU General Public License as published by
  14  *  the Free Software Foundation; either version 2 of the License, or
  15  *  (at your option) any later version.
  16  *
  17  *  This is distributed in the hope that it will be useful, but WITHOUT
  18  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  19  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  20  *  for more details.
  21  *
  22  *  You should have received a copy of the GNU General Public License along
  23  *  with this program; if not, write to the Free Software Foundation, Inc.,
  24  *  51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  25  */
  26
  27 #include <sys/sysmacros.h>
  28 #include <sys/systeminfo.h>
  29 #include <sys/vmsystm.h>
  30 #include <sys/vnode.h>
  31 #include <sys/kmem.h>
  32 #include <sys/mutex.h>
  33 #include <sys/rwlock.h>
  34 #include <sys/taskq.h>
  35 #include <sys/debug.h>
  36 #include <sys/proc.h>
  37 #include <sys/kstat.h>
  38 #include <sys/utsname.h>
  39 #include <sys/file.h>
  40 #include <linux/kmod.h>
  41 #include "spl_config.h"
  42
  43 #ifdef DEBUG_SUBSYSTEM
  44 #undef DEBUG_SUBSYSTEM
  45 #endif
  46
  47 #define DEBUG_SUBSYSTEM S_GENERIC
  48
  49 char spl_version[16] = "SPL v" SPL_META_VERSION;
  50
  51 long spl_hostid = 0;
  52 EXPORT_SYMBOL(spl_hostid);
  53
  54 char hw_serial[HW_HOSTID_LEN] = "<none>";
  55 EXPORT_SYMBOL(hw_serial);
  56
  57 int p0 = 0;
  58 EXPORT_SYMBOL(p0);
  59
  60 #ifndef HAVE_KALLSYMS_LOOKUP_NAME
  61 kallsyms_lookup_name_t spl_kallsyms_lookup_name_fn = SYMBOL_POISON;
  62 #endif
  63
  64 int
  65 highbit(unsigned long i)
  66 {
  67         register int h = 1;
  68         ENTRY;
  69
  70         if (i == 0)
  71                 RETURN(0);
  72 #if BITS_PER_LONG == 64
  73         if (i & 0xffffffff00000000ul) {
  74                 h += 32; i >>= 32;
  75         }
  76 #endif
  77         if (i & 0xffff0000) {
  78                 h += 16; i >>= 16;
  79         }
  80         if (i & 0xff00) {
  81                 h += 8; i >>= 8;
  82         }
  83         if (i & 0xf0) {
  84                 h += 4; i >>= 4;
  85         }
  86         if (i & 0xc) {
  87                 h += 2; i >>= 2;
  88         }
  89         if (i & 0x2) {
  90                 h += 1;
  91         }
  92         RETURN(h);
  93 }
  94 EXPORT_SYMBOL(highbit);
  95
  96 /*
  97  * Implementation of 64 bit division for 32-bit machines.
  98  */
  99 #if BITS_PER_LONG == 32
 100 uint64_t __udivdi3(uint64_t dividend, uint64_t divisor)
 101 {
 102 #if defined(HAVE_DIV64_64) /* 2.6.22 - 2.6.25 API */
 103         return div64_64(dividend, divisor);
 104 #elif defined(HAVE_DIV64_U64) /* 2.6.26 - 2.6.x API */
 105         return div64_u64(dividend, divisor);
 106 #else
 107         /* Implementation from 2.6.30 kernel */
 108         uint32_t high, d;
 109
 110         high = divisor >> 32;
 111         if (high) {
 112                 unsigned int shift = fls(high);
 113
 114                 d = divisor >> shift;
 115                 dividend >>= shift;
 116         } else
 117                 d = divisor;
 118
 119         return do_div(dividend, d);
 120 #endif /* HAVE_DIV64_64, HAVE_DIV64_U64 */
 121 }
 122 EXPORT_SYMBOL(__udivdi3);
 123
 124 /*
 125  * Implementation of 64 bit modulo for 32-bit machines.
 126  */
 127 uint64_t __umoddi3(uint64_t dividend, uint64_t divisor)
 128 {
 129         return dividend - divisor * (dividend / divisor);
 130 }
 131 EXPORT_SYMBOL(__umoddi3);
 132 #endif /* BITS_PER_LONG */
 133
 134 /* NOTE: The strtoxx behavior is solely based on my reading of the Solaris
 135  * ddi_strtol(9F) man page.  I have not verified the behavior of these
 136  * functions against their Solaris counterparts.  It is possible that I
 137  * may have misinterpreted the man page or the man page is incorrect.
 138  */
 139 int ddi_strtoul(const char *, char **, int, unsigned long *);
 140 int ddi_strtol(const char *, char **, int, long *);
 141 int ddi_strtoull(const char *, char **, int, unsigned long long *);
 142 int ddi_strtoll(const char *, char **, int, long long *);
 143
 144 #define define_ddi_strtoux(type, valtype)                               \
 145 int ddi_strtou##type(const char *str, char **endptr,                    \
 146                      int base, valtype *result)                         \
 147 {                                                                       \
 148         valtype last_value, value = 0;                                  \
 149         char *ptr = (char *)str;                                        \
 150         int flag = 1, digit;                                            \
 151                                                                         \
 152         if (strlen(ptr) == 0)                                           \
 153                 return EINVAL;                                          \
 154                                                                         \
 155         /* Auto-detect base based on prefix */                          \
 156         if (!base) {                                                    \
 157                 if (str[0] == '0') {                                    \
 158                         if (tolower(str[1])=='x' && isxdigit(str[2])) { \
 159                                 base = 16; /* hex */                    \
 160                                 ptr += 2;                               \
 161                         } else if (str[1] >= '0' && str[1] < 8) {       \
 162                                 base = 8; /* octal */                   \
 163                                 ptr += 1;                               \
 164                         } else {                                        \
 165                                 return EINVAL;                          \
 166                         }                                               \
 167                 } else {                                                \
 168                         base = 10; /* decimal */                        \
 169                 }                                                       \
 170         }                                                               \
 171                                                                         \
 172         while (1) {                                                     \
 173                 if (isdigit(*ptr))                                      \
 174                         digit = *ptr - '0';                             \
 175                 else if (isalpha(*ptr))                                 \
 176                         digit = tolower(*ptr) - 'a' + 10;               \
 177                 else                                                    \
 178                         break;                                          \
 179                                                                         \
 180                 if (digit >= base)                                      \
 181                         break;                                          \
 182                                                                         \
 183                 last_value = value;                                     \
 184                 value = value * base + digit;                           \
 185                 if (last_value > value) /* Overflow */                  \
 186                         return ERANGE;                                  \
 187                                                                         \
 188                 flag = 1;                                               \
 189                 ptr++;                                                  \
 190         }                                                               \
 191                                                                         \
 192         if (flag)                                                       \
 193                 *result = value;                                        \
 194                                                                         \
 195         if (endptr)                                                     \
 196                 *endptr = (char *)(flag ? ptr : str);                   \
 197                                                                         \
 198         return 0;                                                       \
 199 }                                                                       \
 200
 201 #define define_ddi_strtox(type, valtype)                                \
 202 int ddi_strto##type(const char *str, char **endptr,                     \
 203                        int base, valtype *result)                       \
 204 {                                                                       \
 205         int rc;                                                         \
 206                                                                         \
 207         if (*str == '-') {                                              \
 208                 rc = ddi_strtou##type(str + 1, endptr, base, result);   \
 209                 if (!rc) {                                              \
 210                         if (*endptr == str + 1)                         \
 211                                 *endptr = (char *)str;                  \
 212                         else                                            \
 213                                 *result = -*result;                     \
 214                 }                                                       \
 215         } else {                                                        \
 216                 rc = ddi_strtou##type(str, endptr, base, result);       \
 217         }                                                               \
 218                                                                         \
 219         return rc;                                                      \
 220 }
 221
 222 define_ddi_strtoux(l, unsigned long)
 223 define_ddi_strtox(l, long)
 224 define_ddi_strtoux(ll, unsigned long long)
 225 define_ddi_strtox(ll, long long)
 226
 227 EXPORT_SYMBOL(ddi_strtoul);
 228 EXPORT_SYMBOL(ddi_strtol);
 229 EXPORT_SYMBOL(ddi_strtoll);
 230 EXPORT_SYMBOL(ddi_strtoull);
 231
 232 int
 233 ddi_copyin(const void *from, void *to, size_t len, int flags)
 234 {
 235         /* Fake ioctl() issued by kernel, 'from' is a kernel address */
 236         if (flags & FKIOCTL) {
 237                 memcpy(to, from, len);
 238                 return 0;
 239         }
 240
 241         return copyin(from, to, len);
 242 }
 243 EXPORT_SYMBOL(ddi_copyin);
 244
 245 int
 246 ddi_copyout(const void *from, void *to, size_t len, int flags)
 247 {
 248         /* Fake ioctl() issued by kernel, 'from' is a kernel address */
 249         if (flags & FKIOCTL) {
 250                 memcpy(to, from, len);
 251                 return 0;
 252         }
 253
 254         return copyout(from, to, len);
 255 }
 256 EXPORT_SYMBOL(ddi_copyout);
 257
 258 #ifndef HAVE_PUT_TASK_STRUCT
 259 /*
 260  * This is only a stub function which should never be used.  The SPL should
 261  * never be putting away the last reference on a task structure so this will
 262  * not be called.  However, we still need to define it so the module does not
 263  * have undefined symbol at load time.  That all said if this impossible
 264  * thing does somehow happen SBUG() immediately so we know about it.
 265  */
 266 void
 267 __put_task_struct(struct task_struct *t)
 268 {
 269         SBUG();
 270 }
 271 EXPORT_SYMBOL(__put_task_struct);
 272 #endif /* HAVE_PUT_TASK_STRUCT */
 273
 274 struct new_utsname *__utsname(void)
 275 {
 276 #ifdef HAVE_INIT_UTSNAME
 277         return init_utsname();
 278 #else
 279         return &system_utsname;
 280 #endif
 281 }
 282 EXPORT_SYMBOL(__utsname);
 283
 284 static int
 285 set_hostid(void)
 286 {
 287         char sh_path[] = "/bin/sh";
 288         char *argv[] = { sh_path,
 289                          "-c",
 290                          "/usr/bin/hostid >/proc/sys/kernel/spl/hostid",
 291                          NULL };
 292         char *envp[] = { "HOME=/",
 293                          "TERM=linux",
 294                          "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 295                          NULL };
 296         int rc;
 297
 298         /* Doing address resolution in the kernel is tricky and just
 299          * not a good idea in general.  So to set the proper 'hw_serial'
 300          * use the usermodehelper support to ask '/bin/sh' to run
 301          * '/usr/bin/hostid' and redirect the result to /proc/sys/spl/hostid
 302          * for us to use.  It's a horrific solution but it will do for now.
 303          */
 304         rc = call_usermodehelper(sh_path, argv, envp, 1);
 305         if (rc)
 306                 printk("SPL: Failed user helper '%s %s %s', rc = %d\n",
 307                        argv[0], argv[1], argv[2], rc);
 308
 309         return rc;
 310 }
 311
 312 uint32_t
 313 zone_get_hostid(void *zone)
 314 {
 315         unsigned long hostid;
 316
 317         /* Only the global zone is supported */
 318         ASSERT(zone == NULL);
 319
 320         if (ddi_strtoul(hw_serial, NULL, HW_HOSTID_LEN-1, &hostid) != 0)
 321                 return HW_INVALID_HOSTID;
 322
 323         return (uint32_t)hostid;
 324 }
 325 EXPORT_SYMBOL(zone_get_hostid);
 326
 327 #ifndef HAVE_KALLSYMS_LOOKUP_NAME
 328 /*
 329  * Because kallsyms_lookup_name() is no longer exported in the
 330  * mainline kernel we are forced to resort to somewhat drastic
 331  * measures.  This function replaces the functionality by performing
 332  * an upcall to user space where /proc/kallsyms is consulted for
 333  * the requested address.
 334  */
 335 #define GET_KALLSYMS_ADDR_CMD                                           \
 336         "awk '{ if ( $3 == \"kallsyms_lookup_name\") { print $1 } }' "  \
 337         "/proc/kallsyms >/proc/sys/kernel/spl/kallsyms_lookup_name"
 338
 339 static int
 340 set_kallsyms_lookup_name(void)
 341 {
 342         char sh_path[] = "/bin/sh";
 343         char *argv[] = { sh_path,
 344                          "-c",
 345                          GET_KALLSYMS_ADDR_CMD,
 346                          NULL };
 347         char *envp[] = { "HOME=/",
 348                          "TERM=linux",
 349                          "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 350                          NULL };
 351         int rc;
 352
 353         rc = call_usermodehelper(sh_path, argv, envp, 1);
 354         if (rc)
 355                 printk("SPL: Failed user helper '%s %s %s', rc = %d\n",
 356                        argv[0], argv[1], argv[2], rc);
 357
 358         return rc;
 359 }
 360 #endif
 361
 362 static int
 363 __init spl_init(void)
 364 {
 365         int rc = 0;
 366
 367         if ((rc = debug_init()))
 368                 return rc;
 369
 370         if ((rc = spl_kmem_init()))
 371                 GOTO(out1, rc);
 372
 373         if ((rc = spl_mutex_init()))
 374                 GOTO(out2, rc);
 375
 376         if ((rc = spl_rw_init()))
 377                 GOTO(out3, rc);
 378
 379         if ((rc = spl_taskq_init()))
 380                 GOTO(out4, rc);
 381
 382         if ((rc = vn_init()))
 383                 GOTO(out5, rc);
 384
 385         if ((rc = proc_init()))
 386                 GOTO(out6, rc);
 387
 388         if ((rc = kstat_init()))
 389                 GOTO(out7, rc);
 390
 391         if ((rc = set_hostid()))
 392                 GOTO(out8, rc = -EADDRNOTAVAIL);
 393
 394 #ifndef HAVE_KALLSYMS_LOOKUP_NAME
 395         if ((rc = set_kallsyms_lookup_name()))
 396                 GOTO(out8, rc = -EADDRNOTAVAIL);
 397 #endif /* HAVE_KALLSYMS_LOOKUP_NAME */
 398
 399         if ((rc = spl_kmem_init_kallsyms_lookup()))
 400                 GOTO(out8, rc);
 401
 402         printk("SPL: Loaded Solaris Porting Layer v%s\n", SPL_META_VERSION);
 403         RETURN(rc);
 404 out8:
 405         kstat_fini();
 406 out7:
 407         proc_fini();
 408 out6:
 409         vn_fini();
 410 out5:
 411         spl_taskq_fini();
 412 out4:
 413         spl_rw_fini();
 414 out3:
 415         spl_mutex_fini();
 416 out2:
 417         spl_kmem_fini();
 418 out1:
 419         debug_fini();
 420
 421         printk("SPL: Failed to Load Solaris Porting Layer v%s, "
 422                "rc = %d\n", SPL_META_VERSION, rc);
 423         return rc;
 424 }
 425
 426 static void
 427 spl_fini(void)
 428 {
 429         ENTRY;
 430
 431         printk("SPL: Unloaded Solaris Porting Layer v%s\n", SPL_META_VERSION);
 432         kstat_fini();
 433         proc_fini();
 434         vn_fini();
 435         spl_taskq_fini();
 436         spl_rw_fini();
 437         spl_mutex_fini();
 438         spl_kmem_fini();
 439         debug_fini();
 440 }
 441
 442 /* Called when a dependent module is loaded */
 443 void
 444 spl_setup(void)
 445 {
 446         /*
 447          * At module load time the pwd is set to '/' on a Solaris system.
 448          * On a Linux system will be set to whatever directory the caller
 449          * was in when executing insmod/modprobe.
 450          */
 451         vn_set_pwd("/");
 452 }
 453 EXPORT_SYMBOL(spl_setup);
 454
 455 /* Called when a dependent module is unloaded */
 456 void
 457 spl_cleanup(void)
 458 {
 459 }
 460 EXPORT_SYMBOL(spl_cleanup);
 461
 462 module_init(spl_init);
 463 module_exit(spl_fini);
 464
 465 MODULE_AUTHOR("Lawrence Livermore National Labs");
 466 MODULE_DESCRIPTION("Solaris Porting Layer");
 467 MODULE_LICENSE("GPL");