]> git.proxmox.com Git - mirror_spl-debian.git/blob - module/spl/spl-generic.c
b2bd1b837a9f1c6cd5a035aa0244b002446d64e6
[mirror_spl-debian.git] / module / spl / spl-generic.c
1 /*****************************************************************************\
2 * Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC.
3 * Copyright (C) 2007 The Regents of the University of California.
4 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
5 * Written by Brian Behlendorf <behlendorf1@llnl.gov>.
6 * UCRL-CODE-235197
7 *
8 * This file is part of the SPL, Solaris Porting Layer.
9 * For details, see <http://zfsonlinux.org/>.
10 *
11 * The SPL is free software; you can redistribute it and/or modify it
12 * under the terms of the GNU General Public License as published by the
13 * Free Software Foundation; either version 2 of the License, or (at your
14 * option) any later version.
15 *
16 * The SPL is distributed in the hope that it will be useful, but WITHOUT
17 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
18 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
19 * for more details.
20 *
21 * You should have received a copy of the GNU General Public License along
22 * with the SPL. If not, see <http://www.gnu.org/licenses/>.
23 *****************************************************************************
24 * Solaris Porting Layer (SPL) Generic Implementation.
25 \*****************************************************************************/
26
27 #include <sys/sysmacros.h>
28 #include <sys/systeminfo.h>
29 #include <sys/vmsystm.h>
30 #include <sys/kobj.h>
31 #include <sys/kmem.h>
32 #include <sys/kmem_cache.h>
33 #include <sys/vmem.h>
34 #include <sys/mutex.h>
35 #include <sys/rwlock.h>
36 #include <sys/taskq.h>
37 #include <sys/tsd.h>
38 #include <sys/zmod.h>
39 #include <sys/debug.h>
40 #include <sys/proc.h>
41 #include <sys/kstat.h>
42 #include <sys/file.h>
43 #include <linux/ctype.h>
44 #include <linux/kmod.h>
45 #include <linux/math64_compat.h>
46 #include <linux/proc_compat.h>
47
48 char spl_version[32] = "SPL v" SPL_META_VERSION "-" SPL_META_RELEASE;
49 EXPORT_SYMBOL(spl_version);
50
51 unsigned long spl_hostid = 0;
52 EXPORT_SYMBOL(spl_hostid);
53 module_param(spl_hostid, ulong, 0644);
54 MODULE_PARM_DESC(spl_hostid, "The system hostid.");
55
56 proc_t p0;
57 EXPORT_SYMBOL(p0);
58
59 #if BITS_PER_LONG == 32
60 /*
61 * Support 64/64 => 64 division on a 32-bit platform. While the kernel
62 * provides a div64_u64() function for this we do not use it because the
63 * implementation is flawed. There are cases which return incorrect
64 * results as late as linux-2.6.35. Until this is fixed upstream the
65 * spl must provide its own implementation.
66 *
67 * This implementation is a slightly modified version of the algorithm
68 * proposed by the book 'Hacker's Delight'. The original source can be
69 * found here and is available for use without restriction.
70 *
71 * http://www.hackersdelight.org/HDcode/newCode/divDouble.c
72 */
73
74 /*
75 * Calculate number of leading of zeros for a 64-bit value.
76 */
77 static int
78 nlz64(uint64_t x) {
79 register int n = 0;
80
81 if (x == 0)
82 return 64;
83
84 if (x <= 0x00000000FFFFFFFFULL) {n = n + 32; x = x << 32;}
85 if (x <= 0x0000FFFFFFFFFFFFULL) {n = n + 16; x = x << 16;}
86 if (x <= 0x00FFFFFFFFFFFFFFULL) {n = n + 8; x = x << 8;}
87 if (x <= 0x0FFFFFFFFFFFFFFFULL) {n = n + 4; x = x << 4;}
88 if (x <= 0x3FFFFFFFFFFFFFFFULL) {n = n + 2; x = x << 2;}
89 if (x <= 0x7FFFFFFFFFFFFFFFULL) {n = n + 1;}
90
91 return n;
92 }
93
94 /*
95 * Newer kernels have a div_u64() function but we define our own
96 * to simplify portibility between kernel versions.
97 */
98 static inline uint64_t
99 __div_u64(uint64_t u, uint32_t v)
100 {
101 (void) do_div(u, v);
102 return u;
103 }
104
105 /*
106 * Implementation of 64-bit unsigned division for 32-bit machines.
107 *
108 * First the procedure takes care of the case in which the divisor is a
109 * 32-bit quantity. There are two subcases: (1) If the left half of the
110 * dividend is less than the divisor, one execution of do_div() is all that
111 * is required (overflow is not possible). (2) Otherwise it does two
112 * divisions, using the grade school method.
113 */
114 uint64_t
115 __udivdi3(uint64_t u, uint64_t v)
116 {
117 uint64_t u0, u1, v1, q0, q1, k;
118 int n;
119
120 if (v >> 32 == 0) { // If v < 2**32:
121 if (u >> 32 < v) { // If u/v cannot overflow,
122 return __div_u64(u, v); // just do one division.
123 } else { // If u/v would overflow:
124 u1 = u >> 32; // Break u into two halves.
125 u0 = u & 0xFFFFFFFF;
126 q1 = __div_u64(u1, v); // First quotient digit.
127 k = u1 - q1 * v; // First remainder, < v.
128 u0 += (k << 32);
129 q0 = __div_u64(u0, v); // Seconds quotient digit.
130 return (q1 << 32) + q0;
131 }
132 } else { // If v >= 2**32:
133 n = nlz64(v); // 0 <= n <= 31.
134 v1 = (v << n) >> 32; // Normalize divisor, MSB is 1.
135 u1 = u >> 1; // To ensure no overflow.
136 q1 = __div_u64(u1, v1); // Get quotient from
137 q0 = (q1 << n) >> 31; // Undo normalization and
138 // division of u by 2.
139 if (q0 != 0) // Make q0 correct or
140 q0 = q0 - 1; // too small by 1.
141 if ((u - q0 * v) >= v)
142 q0 = q0 + 1; // Now q0 is correct.
143
144 return q0;
145 }
146 }
147 EXPORT_SYMBOL(__udivdi3);
148
149 /*
150 * Implementation of 64-bit signed division for 32-bit machines.
151 */
152 int64_t
153 __divdi3(int64_t u, int64_t v)
154 {
155 int64_t q, t;
156 q = __udivdi3(abs64(u), abs64(v));
157 t = (u ^ v) >> 63; // If u, v have different
158 return (q ^ t) - t; // signs, negate q.
159 }
160 EXPORT_SYMBOL(__divdi3);
161
162 /*
163 * Implementation of 64-bit unsigned modulo for 32-bit machines.
164 */
165 uint64_t
166 __umoddi3(uint64_t dividend, uint64_t divisor)
167 {
168 return (dividend - (divisor * __udivdi3(dividend, divisor)));
169 }
170 EXPORT_SYMBOL(__umoddi3);
171
172 #if defined(__arm) || defined(__arm__)
173 /*
174 * Implementation of 64-bit (un)signed division for 32-bit arm machines.
175 *
176 * Run-time ABI for the ARM Architecture (page 20). A pair of (unsigned)
177 * long longs is returned in {{r0, r1}, {r2,r3}}, the quotient in {r0, r1},
178 * and the remainder in {r2, r3}. The return type is specifically left
179 * set to 'void' to ensure the compiler does not overwrite these registers
180 * during the return. All results are in registers as per ABI
181 */
182 void
183 __aeabi_uldivmod(uint64_t u, uint64_t v)
184 {
185 uint64_t res;
186 uint64_t mod;
187
188 res = __udivdi3(u, v);
189 mod = __umoddi3(u, v);
190 {
191 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
192 register uint32_t r1 asm("r1") = (res >> 32);
193 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
194 register uint32_t r3 asm("r3") = (mod >> 32);
195
196 asm volatile(""
197 : "+r"(r0), "+r"(r1), "+r"(r2),"+r"(r3) /* output */
198 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */
199
200 return; /* r0; */
201 }
202 }
203 EXPORT_SYMBOL(__aeabi_uldivmod);
204
205 void
206 __aeabi_ldivmod(int64_t u, int64_t v)
207 {
208 int64_t res;
209 uint64_t mod;
210
211 res = __divdi3(u, v);
212 mod = __umoddi3(u, v);
213 {
214 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
215 register uint32_t r1 asm("r1") = (res >> 32);
216 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
217 register uint32_t r3 asm("r3") = (mod >> 32);
218
219 asm volatile(""
220 : "+r"(r0), "+r"(r1), "+r"(r2),"+r"(r3) /* output */
221 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */
222
223 return; /* r0; */
224 }
225 }
226 EXPORT_SYMBOL(__aeabi_ldivmod);
227 #endif /* __arm || __arm__ */
228 #endif /* BITS_PER_LONG */
229
230 /* NOTE: The strtoxx behavior is solely based on my reading of the Solaris
231 * ddi_strtol(9F) man page. I have not verified the behavior of these
232 * functions against their Solaris counterparts. It is possible that I
233 * may have misinterpreted the man page or the man page is incorrect.
234 */
235 int ddi_strtoul(const char *, char **, int, unsigned long *);
236 int ddi_strtol(const char *, char **, int, long *);
237 int ddi_strtoull(const char *, char **, int, unsigned long long *);
238 int ddi_strtoll(const char *, char **, int, long long *);
239
240 #define define_ddi_strtoux(type, valtype) \
241 int ddi_strtou##type(const char *str, char **endptr, \
242 int base, valtype *result) \
243 { \
244 valtype last_value, value = 0; \
245 char *ptr = (char *)str; \
246 int flag = 1, digit; \
247 \
248 if (strlen(ptr) == 0) \
249 return EINVAL; \
250 \
251 /* Auto-detect base based on prefix */ \
252 if (!base) { \
253 if (str[0] == '0') { \
254 if (tolower(str[1])=='x' && isxdigit(str[2])) { \
255 base = 16; /* hex */ \
256 ptr += 2; \
257 } else if (str[1] >= '0' && str[1] < 8) { \
258 base = 8; /* octal */ \
259 ptr += 1; \
260 } else { \
261 return EINVAL; \
262 } \
263 } else { \
264 base = 10; /* decimal */ \
265 } \
266 } \
267 \
268 while (1) { \
269 if (isdigit(*ptr)) \
270 digit = *ptr - '0'; \
271 else if (isalpha(*ptr)) \
272 digit = tolower(*ptr) - 'a' + 10; \
273 else \
274 break; \
275 \
276 if (digit >= base) \
277 break; \
278 \
279 last_value = value; \
280 value = value * base + digit; \
281 if (last_value > value) /* Overflow */ \
282 return ERANGE; \
283 \
284 flag = 1; \
285 ptr++; \
286 } \
287 \
288 if (flag) \
289 *result = value; \
290 \
291 if (endptr) \
292 *endptr = (char *)(flag ? ptr : str); \
293 \
294 return 0; \
295 } \
296
297 #define define_ddi_strtox(type, valtype) \
298 int ddi_strto##type(const char *str, char **endptr, \
299 int base, valtype *result) \
300 { \
301 int rc; \
302 \
303 if (*str == '-') { \
304 rc = ddi_strtou##type(str + 1, endptr, base, result); \
305 if (!rc) { \
306 if (*endptr == str + 1) \
307 *endptr = (char *)str; \
308 else \
309 *result = -*result; \
310 } \
311 } else { \
312 rc = ddi_strtou##type(str, endptr, base, result); \
313 } \
314 \
315 return rc; \
316 }
317
318 define_ddi_strtoux(l, unsigned long)
319 define_ddi_strtox(l, long)
320 define_ddi_strtoux(ll, unsigned long long)
321 define_ddi_strtox(ll, long long)
322
323 EXPORT_SYMBOL(ddi_strtoul);
324 EXPORT_SYMBOL(ddi_strtol);
325 EXPORT_SYMBOL(ddi_strtoll);
326 EXPORT_SYMBOL(ddi_strtoull);
327
328 int
329 ddi_copyin(const void *from, void *to, size_t len, int flags)
330 {
331 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
332 if (flags & FKIOCTL) {
333 memcpy(to, from, len);
334 return 0;
335 }
336
337 return copyin(from, to, len);
338 }
339 EXPORT_SYMBOL(ddi_copyin);
340
341 int
342 ddi_copyout(const void *from, void *to, size_t len, int flags)
343 {
344 /* Fake ioctl() issued by kernel, 'from' is a kernel address */
345 if (flags & FKIOCTL) {
346 memcpy(to, from, len);
347 return 0;
348 }
349
350 return copyout(from, to, len);
351 }
352 EXPORT_SYMBOL(ddi_copyout);
353
354 /*
355 * Read the unique system identifier from the /etc/hostid file.
356 *
357 * The behavior of /usr/bin/hostid on Linux systems with the
358 * regular eglibc and coreutils is:
359 *
360 * 1. Generate the value if the /etc/hostid file does not exist
361 * or if the /etc/hostid file is less than four bytes in size.
362 *
363 * 2. If the /etc/hostid file is at least 4 bytes, then return
364 * the first four bytes [0..3] in native endian order.
365 *
366 * 3. Always ignore bytes [4..] if they exist in the file.
367 *
368 * Only the first four bytes are significant, even on systems that
369 * have a 64-bit word size.
370 *
371 * See:
372 *
373 * eglibc: sysdeps/unix/sysv/linux/gethostid.c
374 * coreutils: src/hostid.c
375 *
376 * Notes:
377 *
378 * The /etc/hostid file on Solaris is a text file that often reads:
379 *
380 * # DO NOT EDIT
381 * "0123456789"
382 *
383 * Directly copying this file to Linux results in a constant
384 * hostid of 4f442023 because the default comment constitutes
385 * the first four bytes of the file.
386 *
387 */
388
389 char *spl_hostid_path = HW_HOSTID_PATH;
390 module_param(spl_hostid_path, charp, 0444);
391 MODULE_PARM_DESC(spl_hostid_path, "The system hostid file (/etc/hostid)");
392
393 static int
394 hostid_read(void)
395 {
396 int result;
397 uint64_t size;
398 struct _buf *file;
399 uint32_t hostid = 0;
400
401 file = kobj_open_file(spl_hostid_path);
402
403 if (file == (struct _buf *)-1)
404 return -1;
405
406 result = kobj_get_filesize(file, &size);
407
408 if (result != 0) {
409 printk(KERN_WARNING
410 "SPL: kobj_get_filesize returned %i on %s\n",
411 result, spl_hostid_path);
412 kobj_close_file(file);
413 return -2;
414 }
415
416 if (size < sizeof(HW_HOSTID_MASK)) {
417 printk(KERN_WARNING
418 "SPL: Ignoring the %s file because it is %llu bytes; "
419 "expecting %lu bytes instead.\n", spl_hostid_path,
420 size, (unsigned long)sizeof(HW_HOSTID_MASK));
421 kobj_close_file(file);
422 return -3;
423 }
424
425 /* Read directly into the variable like eglibc does. */
426 /* Short reads are okay; native behavior is preserved. */
427 result = kobj_read_file(file, (char *)&hostid, sizeof(hostid), 0);
428
429 if (result < 0) {
430 printk(KERN_WARNING
431 "SPL: kobj_read_file returned %i on %s\n",
432 result, spl_hostid_path);
433 kobj_close_file(file);
434 return -4;
435 }
436
437 /* Mask down to 32 bits like coreutils does. */
438 spl_hostid = hostid & HW_HOSTID_MASK;
439 kobj_close_file(file);
440 return 0;
441 }
442
443 uint32_t
444 zone_get_hostid(void *zone)
445 {
446 static int first = 1;
447
448 /* Only the global zone is supported */
449 ASSERT(zone == NULL);
450
451 if (first) {
452 first = 0;
453
454 spl_hostid &= HW_HOSTID_MASK;
455 /*
456 * Get the hostid if it was not passed as a module parameter.
457 * Try reading the /etc/hostid file directly.
458 */
459 if (spl_hostid == 0 && hostid_read())
460 spl_hostid = 0;
461
462
463 printk(KERN_NOTICE "SPL: using hostid 0x%08x\n",
464 (unsigned int) spl_hostid);
465 }
466
467 return spl_hostid;
468 }
469 EXPORT_SYMBOL(zone_get_hostid);
470
471 static int
472 spl_kvmem_init(void)
473 {
474 int rc = 0;
475
476 rc = spl_kmem_init();
477 if (rc)
478 goto out1;
479
480 rc = spl_vmem_init();
481 if (rc)
482 goto out2;
483
484 rc = spl_kmem_cache_init();
485 if (rc)
486 goto out3;
487
488 return (rc);
489 out3:
490 spl_vmem_fini();
491 out2:
492 spl_kmem_fini();
493 out1:
494 return (rc);
495 }
496
497 static void
498 spl_kvmem_fini(void)
499 {
500 spl_kmem_cache_fini();
501 spl_vmem_fini();
502 spl_kmem_fini();
503 }
504
505 static int __init
506 spl_init(void)
507 {
508 int rc = 0;
509
510 bzero(&p0, sizeof (proc_t));
511
512 if ((rc = spl_kvmem_init()))
513 goto out1;
514
515 if ((rc = spl_mutex_init()))
516 goto out2;
517
518 if ((rc = spl_rw_init()))
519 goto out3;
520
521 if ((rc = spl_taskq_init()))
522 goto out4;
523
524 if ((rc = spl_vn_init()))
525 goto out5;
526
527 if ((rc = spl_proc_init()))
528 goto out6;
529
530 if ((rc = spl_kstat_init()))
531 goto out7;
532
533 if ((rc = spl_tsd_init()))
534 goto out8;
535
536 if ((rc = spl_zlib_init()))
537 goto out9;
538
539 printk(KERN_NOTICE "SPL: Loaded module v%s-%s%s\n", SPL_META_VERSION,
540 SPL_META_RELEASE, SPL_DEBUG_STR);
541 return (rc);
542
543 out9:
544 spl_tsd_fini();
545 out8:
546 spl_kstat_fini();
547 out7:
548 spl_proc_fini();
549 out6:
550 spl_vn_fini();
551 out5:
552 spl_taskq_fini();
553 out4:
554 spl_rw_fini();
555 out3:
556 spl_mutex_fini();
557 out2:
558 spl_kvmem_fini();
559 out1:
560 printk(KERN_NOTICE "SPL: Failed to Load Solaris Porting Layer "
561 "v%s-%s%s, rc = %d\n", SPL_META_VERSION, SPL_META_RELEASE,
562 SPL_DEBUG_STR, rc);
563
564 return (rc);
565 }
566
567 static void __exit
568 spl_fini(void)
569 {
570 printk(KERN_NOTICE "SPL: Unloaded module v%s-%s%s\n",
571 SPL_META_VERSION, SPL_META_RELEASE, SPL_DEBUG_STR);
572 spl_zlib_fini();
573 spl_tsd_fini();
574 spl_kstat_fini();
575 spl_proc_fini();
576 spl_vn_fini();
577 spl_taskq_fini();
578 spl_rw_fini();
579 spl_mutex_fini();
580 spl_kvmem_fini();
581 }
582
583 module_init(spl_init);
584 module_exit(spl_fini);
585
586 MODULE_DESCRIPTION("Solaris Porting Layer");
587 MODULE_AUTHOR(SPL_META_AUTHOR);
588 MODULE_LICENSE(SPL_META_LICENSE);
589 MODULE_VERSION(SPL_META_VERSION "-" SPL_META_RELEASE);