sys/kern/kern_fail.c

   1 /*-
   2  * Copyright (c) 2009 Isilon Inc http://www.isilon.com/
   3  *
   4  * Redistribution and use in source and binary forms, with or without
   5  * modification, are permitted provided that the following conditions
   6  * are met:
   7  * 1. Redistributions of source code must retain the above copyright
   8  *    notice, this list of conditions and the following disclaimer.
   9  * 2. Redistributions in binary form must reproduce the above copyright
  10  *    notice, this list of conditions and the following disclaimer in the
  11  *    documentation and/or other materials provided with the distribution.
  12  *
  13  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  14  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  15  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  16  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  17  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  18  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  19  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  20  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  21  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  22  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  23  * SUCH DAMAGE.
  24  */
  25 /**
  26  * @file
  27  *
  28  * fail(9) Facility.
  29  *
  30  * @ingroup failpoint_private
  31  */
  32 /**
  33  * @defgroup failpoint fail(9) Facility
  34  *
  35  * Failpoints allow for injecting fake errors into running code on the fly,
  36  * without modifying code or recompiling with flags.  Failpoints are always
  37  * present, and are very efficient when disabled.  Failpoints are described
  38  * in man fail(9).
  39  */
  40 /**
  41  * @defgroup failpoint_private Private fail(9) Implementation functions
  42  *
  43  * Private implementations for the actual failpoint code.
  44  *
  45  * @ingroup failpoint
  46  */
  47 /**
  48  * @addtogroup failpoint_private
  49  * @{
  50  */
  51
  52 #include <sys/cdefs.h>
  53 __FBSDID("$FreeBSD$");
  54
  55 #include <sys/errno.h>
  56 #include <sys/fail.h>
  57 #include <sys/kernel.h>
  58 #include <sys/libkern.h>
  59 #include <sys/lock.h>
  60 #include <sys/malloc.h>
  61 #include <sys/mutex.h>
  62 #include <sys/sbuf.h>
  63
  64 #include <machine/stdarg.h>
  65
  66 #ifdef ILOG_DEFINE_FOR_FILE
  67 ILOG_DEFINE_FOR_FILE(L_ISI_FAIL_POINT, L_ILOG, fail_point);
  68 #endif
  69
  70 MALLOC_DEFINE(M_FAIL_POINT, "Fail Points", "fail points system");
  71 #define fp_free(ptr) free(ptr, M_FAIL_POINT)
  72 #define fp_malloc(size, flags) malloc((size), M_FAIL_POINT, (flags))
  73
  74 static struct mtx g_fp_mtx;
  75 MTX_SYSINIT(g_fp_mtx, &g_fp_mtx, "fail point mtx", MTX_DEF);
  76 #define FP_LOCK()       mtx_lock(&g_fp_mtx)
  77 #define FP_UNLOCK()     mtx_unlock(&g_fp_mtx)
  78
  79 static inline void
  80 fail_point_sleep(struct fail_point *fp, struct fail_point_entry *ent,
  81     int msecs, enum fail_point_return_code *pret)
  82 {
  83         /* convert from millisecs to ticks, rounding up */
  84         int timo = ((msecs * hz) + 999) / 1000;
  85
  86         if (timo) {
  87                 if (fp->fp_sleep_fn == NULL) {
  88                         msleep(fp, &g_fp_mtx, PWAIT, "failpt", timo);
  89                 } else {
  90                         timeout(fp->fp_sleep_fn, fp->fp_sleep_arg, timo);
  91                         *pret = FAIL_POINT_RC_QUEUED;
  92                 }
  93         }
  94 }
  95
  96
  97 /**
  98  * Defines stating the equivalent of probablilty one (100%)
  99  */
 100 enum {
 101         PROB_MAX = 1000000,     /* probability between zero and this number */
 102         PROB_DIGITS = 6,        /* number of zero's in above number */
 103 };
 104
 105 static const char *fail_type_strings[] = {
 106         "off",
 107         "panic",
 108         "return",
 109         "break",
 110         "print",
 111         "sleep",
 112 };
 113
 114 static char *parse_fail_point(struct fail_point_entries *, char *);
 115 static char *parse_term(struct fail_point_entries *, char *);
 116 static char *parse_number(int *out_units, int *out_decimal, char *);
 117 static char *parse_type(struct fail_point_entry *, char *);
 118 static void free_entry(struct fail_point_entries *, struct fail_point_entry *);
 119 static void clear_entries(struct fail_point_entries *);
 120
 121 /**
 122  * Initialize a fail_point.  The name is formed in a printf-like fashion
 123  * from "fmt" and subsequent arguments.  This function is generally used
 124  * for custom failpoints located at odd places in the sysctl tree, and is
 125  * not explicitly needed for standard in-line-declared failpoints.
 126  *
 127  * @ingroup failpoint
 128  */
 129 void
 130 fail_point_init(struct fail_point *fp, const char *fmt, ...)
 131 {
 132         va_list ap;
 133         char *name;
 134         int n;
 135
 136         TAILQ_INIT(&fp->fp_entries);
 137         fp->fp_flags = 0;
 138
 139         /* Figure out the size of the name. */
 140         va_start(ap, fmt);
 141         n = vsnprintf(NULL, 0, fmt, ap);
 142         va_end(ap);
 143
 144         /* Allocate the name and fill it in. */
 145         name = fp_malloc(n + 1, M_WAITOK);
 146         if (name != NULL) {
 147                 va_start(ap, fmt);
 148                 vsnprintf(name, n + 1, fmt, ap);
 149                 va_end(ap);
 150         }
 151         fp->fp_name = name;
 152         fp->fp_flags |= FAIL_POINT_DYNAMIC_NAME;
 153         fp->fp_sleep_fn = NULL;
 154         fp->fp_sleep_arg = NULL;
 155 }
 156
 157 /**
 158  * Free the resources held by a fail_point.
 159  *
 160  * @ingroup failpoint
 161  */
 162 void
 163 fail_point_destroy(struct fail_point *fp)
 164 {
 165         struct fail_point_entry *ent;
 166
 167         if (fp->fp_flags & FAIL_POINT_DYNAMIC_NAME && fp->fp_name != NULL) {
 168                 fp_free((void *)(intptr_t)fp->fp_name);
 169                 fp->fp_name = NULL;
 170         }
 171         fp->fp_flags = 0;
 172
 173         while (!TAILQ_EMPTY(&fp->fp_entries)) {
 174                 ent = TAILQ_FIRST(&fp->fp_entries);
 175                 TAILQ_REMOVE(&fp->fp_entries, ent, fe_entries);
 176                 fp_free(ent);
 177         }
 178 }
 179
 180 /**
 181  * This does the real work of evaluating a fail point. If the fail point tells
 182  * us to return a value, this function returns 1 and fills in 'return_value'
 183  * (return_value is allowed to be null). If the fail point tells us to panic,
 184  * we never return. Otherwise we just return 0 after doing some work, which
 185  * means "keep going".
 186  */
 187 enum fail_point_return_code
 188 fail_point_eval_nontrivial(struct fail_point *fp, int *return_value)
 189 {
 190         enum fail_point_return_code ret = FAIL_POINT_RC_CONTINUE;
 191         struct fail_point_entry *ent, *next;
 192         int msecs;
 193
 194         FP_LOCK();
 195
 196         ent = TAILQ_FIRST(&fp->fp_entries);
 197         while (ent) {
 198                 int cont = 0; /* don't continue by default */
 199                 next = TAILQ_NEXT(ent, fe_entries);
 200
 201                 if (ent->fe_prob < PROB_MAX &&
 202                     ent->fe_prob < random() % PROB_MAX) {
 203                         cont = 1;
 204                         goto loop_end;
 205                 }
 206
 207                 switch (ent->fe_type) {
 208                 case FAIL_POINT_PANIC:
 209                         panic("fail point %s panicking", fp->fp_name);
 210                         /* NOTREACHED */
 211
 212                 case FAIL_POINT_RETURN:
 213                         if (return_value)
 214                                 *return_value = ent->fe_arg;
 215                         ret = FAIL_POINT_RC_RETURN;
 216                         break;
 217
 218                 case FAIL_POINT_BREAK:
 219                         printf("fail point %s breaking to debugger\n", fp->fp_name);
 220                         breakpoint();
 221                         break;
 222
 223                 case FAIL_POINT_PRINT:
 224                         printf("fail point %s executing\n", fp->fp_name);
 225                         cont = ent->fe_arg;
 226                         break;
 227
 228                 case FAIL_POINT_SLEEP:
 229                         /*
 230                          * Free the entry now if necessary, since
 231                          * we're about to drop the mutex and sleep.
 232                          */
 233                         msecs = ent->fe_arg;
 234                         if (ent->fe_count > 0 && --ent->fe_count == 0) {
 235                                 free_entry(&fp->fp_entries, ent);
 236                                 ent = NULL;
 237                         }
 238
 239                         if (msecs)
 240                                 fail_point_sleep(fp, ent, msecs, &ret);
 241                         break;
 242
 243                 default:
 244                         break;
 245                 }
 246
 247                 if (ent && ent->fe_count > 0 && --ent->fe_count == 0)
 248                         free_entry(&fp->fp_entries, ent);
 249
 250 loop_end:
 251                 if (cont)
 252                         ent = next;
 253                 else
 254                         break;
 255         }
 256
 257         /* Get rid of "off"s at the end. */
 258         while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
 259                ent->fe_type == FAIL_POINT_OFF)
 260                 free_entry(&fp->fp_entries, ent);
 261
 262         FP_UNLOCK();
 263
 264         return ret;
 265 }
 266
 267 /**
 268  * Translate internal fail_point structure into human-readable text.
 269  */
 270 static void
 271 fail_point_get(struct fail_point *fp, struct sbuf *sb)
 272 {
 273         struct fail_point_entry *ent;
 274
 275         FP_LOCK();
 276
 277         TAILQ_FOREACH(ent, &fp->fp_entries, fe_entries) {
 278                 if (ent->fe_prob < PROB_MAX) {
 279                         int decimal = ent->fe_prob % (PROB_MAX / 100);
 280                         int units = ent->fe_prob / (PROB_MAX / 100);
 281                         sbuf_printf(sb, "%d", units);
 282                         if (decimal) {
 283                                 int digits = PROB_DIGITS - 2;
 284                                 while (!(decimal % 10)) {
 285                                         digits--;
 286                                         decimal /= 10;
 287                                 }
 288                                 sbuf_printf(sb, ".%0*d", digits, decimal);
 289                         }
 290                         sbuf_printf(sb, "%%");
 291                 }
 292                 if (ent->fe_count > 0)
 293                         sbuf_printf(sb, "%d*", ent->fe_count);
 294                 sbuf_printf(sb, "%s", fail_type_strings[ent->fe_type]);
 295                 if (ent->fe_arg)
 296                         sbuf_printf(sb, "(%d)", ent->fe_arg);
 297                 if (TAILQ_NEXT(ent, fe_entries))
 298                         sbuf_printf(sb, "->");
 299         }
 300         if (TAILQ_EMPTY(&fp->fp_entries))
 301                 sbuf_printf(sb, "off");
 302
 303         FP_UNLOCK();
 304 }
 305
 306 /**
 307  * Set an internal fail_point structure from a human-readable failpoint string
 308  * in a lock-safe manner.
 309  */
 310 static int
 311 fail_point_set(struct fail_point *fp, char *buf)
 312 {
 313         int error = 0;
 314         struct fail_point_entry *ent, *ent_next;
 315         struct fail_point_entries new_entries;
 316
 317         /* Parse new entries. */
 318         TAILQ_INIT(&new_entries);
 319         if (!parse_fail_point(&new_entries, buf)) {
 320                 clear_entries(&new_entries);
 321                 error = EINVAL;
 322                 goto end;
 323         }
 324
 325         FP_LOCK();
 326
 327         /* Move new entries in. */
 328         TAILQ_SWAP(&fp->fp_entries, &new_entries, fail_point_entry, fe_entries);
 329         clear_entries(&new_entries);
 330
 331         /* Get rid of useless zero probability entries. */
 332         TAILQ_FOREACH_SAFE(ent, &fp->fp_entries, fe_entries, ent_next) {
 333                 if (ent->fe_prob == 0)
 334                         free_entry(&fp->fp_entries, ent);
 335         }
 336
 337         /* Get rid of "off"s at the end. */
 338         while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
 339                 ent->fe_type == FAIL_POINT_OFF)
 340                 free_entry(&fp->fp_entries, ent);
 341
 342         FP_UNLOCK();
 343
 344  end:
 345 #ifdef IWARNING
 346         if (error)
 347                 IWARNING("Failed to set %s (%s) to %s",
 348                     fp->fp_name, fp->fp_location, buf);
 349         else
 350                 INOTICE("Set %s (%s) to %s",
 351                     fp->fp_name, fp->fp_location, buf);
 352 #endif /* IWARNING */
 353
 354         return error;
 355 }
 356
 357 #define MAX_FAIL_POINT_BUF      1023
 358
 359 /**
 360  * Handle kernel failpoint set/get.
 361  */
 362 int
 363 fail_point_sysctl(SYSCTL_HANDLER_ARGS)
 364 {
 365         struct fail_point *fp = arg1;
 366         char *buf = NULL;
 367         struct sbuf sb;
 368         int error;
 369
 370         /* Retrieving */
 371         sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND);
 372         fail_point_get(fp, &sb);
 373         sbuf_trim(&sb);
 374         sbuf_finish(&sb);
 375         error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
 376         sbuf_delete(&sb);
 377
 378         /* Setting */
 379         if (!error && req->newptr) {
 380                 if (req->newlen > MAX_FAIL_POINT_BUF) {
 381                         error = EINVAL;
 382                         goto out;
 383                 }
 384
 385                 buf = fp_malloc(req->newlen + 1, M_WAITOK);
 386
 387                 error = SYSCTL_IN(req, buf, req->newlen);
 388                 if (error)
 389                         goto out;
 390                 buf[req->newlen] = '\0';
 391
 392                 error = fail_point_set(fp, buf);
 393         }
 394
 395 out:
 396         if (buf)
 397                 fp_free(buf);
 398         return error;
 399 }
 400
 401 /**
 402  * Internal helper function to translate a human-readable failpoint string
 403  * into a internally-parsable fail_point structure.
 404  */
 405 static char *
 406 parse_fail_point(struct fail_point_entries *ents, char *p)
 407 {
 408         /*  <fail_point> ::
 409          *      <term> ( "->" <term> )*
 410          */
 411         if (!(p = parse_term(ents, p)))
 412                 return 0;
 413         while (*p)
 414                 if (p[0] != '-' || p[1] != '>' || !(p = parse_term(ents, p+2)))
 415                         return 0;
 416         return p;
 417 }
 418
 419 /**
 420  * Internal helper function to parse an individual term from a failpoint.
 421  */
 422 static char *
 423 parse_term(struct fail_point_entries *ents, char *p)
 424 {
 425         struct fail_point_entry *ent;
 426
 427         ent = fp_malloc(sizeof *ent, M_WAITOK | M_ZERO);
 428         ent->fe_prob = PROB_MAX;
 429         TAILQ_INSERT_TAIL(ents, ent, fe_entries);
 430
 431         /*
 432          * <term> ::
 433          *     ( (<float> "%") | (<integer> "*" ) )*
 434          *     <type>
 435          *     [ "(" <integer> ")" ]
 436          */
 437
 438         /* ( (<float> "%") | (<integer> "*" ) )* */
 439         while (('0' <= *p && *p <= '9') || *p == '.') {
 440                 int units, decimal;
 441
 442                 if (!(p = parse_number(&units, &decimal, p)))
 443                         return 0;
 444
 445                 if (*p == '%') {
 446                         if (units > 100) /* prevent overflow early */
 447                                 units = 100;
 448                         ent->fe_prob = units * (PROB_MAX / 100) + decimal;
 449                         if (ent->fe_prob > PROB_MAX)
 450                                 ent->fe_prob = PROB_MAX;
 451
 452                 } else if (*p == '*') {
 453                         if (!units || decimal)
 454                                 return 0;
 455                         ent->fe_count = units;
 456
 457                 } else {
 458                         return 0;
 459                 }
 460
 461                 p++;
 462         }
 463
 464         /* <type> */
 465         if (!(p = parse_type(ent, p)))
 466                 return 0;
 467         if (*p == '\0')
 468                 return p;
 469
 470         /* [ "(" <integer> ")" ] */
 471         if (*p != '(')
 472                 return p;
 473         p++;
 474         if (('0' <= *p && *p <= '9') || *p == '-')
 475                 ent->fe_arg = strtol(p, &p, 0);
 476         else
 477                 return 0;
 478         if (*p++ != ')')
 479                 return 0;
 480
 481         return p;
 482 }
 483
 484 /**
 485  * Internal helper function to parse a numeric for a failpoint term.
 486  */
 487 static char *
 488 parse_number(int *out_units, int *out_decimal, char *p)
 489 {
 490         char *old_p;
 491
 492         /*
 493          *  <number> ::
 494          *      <integer> [ "." <integer> ] |
 495          *      "." <integer>
 496          */
 497
 498         /* whole part */
 499         old_p = p;
 500         *out_units = strtol(p, &p, 10);
 501         if (p == old_p && *p != '.')
 502                 return 0;
 503
 504         /* fractional part */
 505         *out_decimal = 0;
 506         if (*p == '.') {
 507                 int digits = 0;
 508                 p++;
 509                 while ('0' <= *p && *p <= '9') {
 510                         int digit = *p - '0';
 511                         if (digits < PROB_DIGITS - 2)
 512                                 *out_decimal = *out_decimal * 10 + digit;
 513                         else if (digits == PROB_DIGITS - 2 && digit >= 5)
 514                                 (*out_decimal)++;
 515                         digits++;
 516                         p++;
 517                 }
 518                 if (!digits) /* need at least one digit after '.' */
 519                         return 0;
 520                 while (digits++ < PROB_DIGITS - 2) /* add implicit zeros */
 521                         *out_decimal *= 10;
 522         }
 523
 524         return p; /* success */
 525 }
 526
 527 /**
 528  * Internal helper function to parse an individual type for a failpoint term.
 529  */
 530 static char *
 531 parse_type(struct fail_point_entry *ent, char *beg)
 532 {
 533         enum fail_point_t type;
 534         char *end = beg;
 535         while ('a' <= *end && *end <= 'z')
 536                 end++;
 537         if (beg == end)
 538                 return 0;
 539         for (type = FAIL_POINT_OFF; type != FAIL_POINT_INVALID; type++) {
 540                 const char *p = fail_type_strings[type];
 541                 const char *q = beg;
 542                 while (q < end && *p++ == *q++);
 543                 if (q == end && *p == '\0') {
 544                         ent->fe_type = type;
 545                         return end;
 546                 }
 547         }
 548         return 0;
 549 }
 550
 551 /**
 552  * Internal helper function to free an individual failpoint term.
 553  */
 554 static void
 555 free_entry(struct fail_point_entries *ents, struct fail_point_entry *ent)
 556 {
 557         TAILQ_REMOVE(ents, ent, fe_entries);
 558         fp_free(ent);
 559 }
 560
 561 /**
 562  * Internal helper function to clear out all failpoint terms for a single
 563  * failpoint.
 564  */
 565 static void
 566 clear_entries(struct fail_point_entries *ents)
 567 {
 568         struct fail_point_entry *ent, *ent_next;
 569         TAILQ_FOREACH_SAFE(ent, ents, fe_entries, ent_next)
 570                 fp_free(ent);
 571         TAILQ_INIT(ents);
 572 }
 573
 574 /* The fail point sysctl tree. */
 575 SYSCTL_NODE(_debug, OID_AUTO, fail_point, CTLFLAG_RW, 0, "fail points");