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 /**
  80  * Failpoint types.
  81  * Don't change these without changing fail_type_strings in fail.c.
  82  * @ingroup failpoint_private
  83  */
  84 enum fail_point_t {
  85         FAIL_POINT_OFF,         /**< don't fail */
  86         FAIL_POINT_PANIC,       /**< panic */
  87         FAIL_POINT_RETURN,      /**< return an errorcode */
  88         FAIL_POINT_BREAK,       /**< break into the debugger */
  89         FAIL_POINT_PRINT,       /**< print a message */
  90         FAIL_POINT_SLEEP,       /**< sleep for some msecs */
  91         FAIL_POINT_INVALID,     /**< placeholder */
  92 };
  93
  94 static const char *fail_type_strings[] = {
  95         "off",
  96         "panic",
  97         "return",
  98         "break",
  99         "print",
 100         "sleep",
 101 };
 102
 103 /**
 104  * Internal structure tracking a single term of a complete failpoint.
 105  * @ingroup failpoint_private
 106  */
 107 struct fail_point_entry {
 108         enum fail_point_t fe_type;      /**< type of entry */
 109         int             fe_arg;         /**< argument to type (e.g. return value) */
 110         int             fe_prob;        /**< likelihood of firing in millionths */
 111         int             fe_count;       /**< number of times to fire, 0 means always */
 112
 113         TAILQ_ENTRY(fail_point_entry) fe_entries; /**< next entry in fail point */
 114 };
 115
 116 static inline void
 117 fail_point_sleep(struct fail_point *fp, struct fail_point_entry *ent,
 118     int msecs, enum fail_point_return_code *pret)
 119 {
 120         /* convert from millisecs to ticks, rounding up */
 121         int timo = ((msecs * hz) + 999) / 1000;
 122
 123         if (timo) {
 124                 if (fp->fp_sleep_fn == NULL) {
 125                         msleep(fp, &g_fp_mtx, PWAIT, "failpt", timo);
 126                 } else {
 127                         timeout(fp->fp_sleep_fn, fp->fp_sleep_arg, timo);
 128                         *pret = FAIL_POINT_RC_QUEUED;
 129                 }
 130         }
 131 }
 132
 133
 134 /**
 135  * Defines stating the equivalent of probablilty one (100%)
 136  */
 137 enum {
 138         PROB_MAX = 1000000,     /* probability between zero and this number */
 139         PROB_DIGITS = 6,        /* number of zero's in above number */
 140 };
 141
 142 static char *parse_fail_point(struct fail_point_entries *, char *);
 143 static char *parse_term(struct fail_point_entries *, char *);
 144 static char *parse_number(int *out_units, int *out_decimal, char *);
 145 static char *parse_type(struct fail_point_entry *, char *);
 146 static void free_entry(struct fail_point_entries *, struct fail_point_entry *);
 147 static void clear_entries(struct fail_point_entries *);
 148
 149 /**
 150  * Initialize a fail_point.  The name is formed in a printf-like fashion
 151  * from "fmt" and subsequent arguments.  This function is generally used
 152  * for custom failpoints located at odd places in the sysctl tree, and is
 153  * not explicitly needed for standard in-line-declared failpoints.
 154  *
 155  * @ingroup failpoint
 156  */
 157 void
 158 fail_point_init(struct fail_point *fp, const char *fmt, ...)
 159 {
 160         va_list ap;
 161         char *name;
 162         int n;
 163
 164         TAILQ_INIT(&fp->fp_entries);
 165         fp->fp_flags = 0;
 166
 167         /* Figure out the size of the name. */
 168         va_start(ap, fmt);
 169         n = vsnprintf(NULL, 0, fmt, ap);
 170         va_end(ap);
 171
 172         /* Allocate the name and fill it in. */
 173         name = fp_malloc(n + 1, M_WAITOK);
 174         if (name != NULL) {
 175                 va_start(ap, fmt);
 176                 vsnprintf(name, n + 1, fmt, ap);
 177                 va_end(ap);
 178         }
 179         fp->fp_name = name;
 180         fp->fp_flags |= FAIL_POINT_DYNAMIC_NAME;
 181         fp->fp_sleep_fn = NULL;
 182         fp->fp_sleep_arg = NULL;
 183 }
 184
 185 /**
 186  * Free the resources held by a fail_point.
 187  *
 188  * @ingroup failpoint
 189  */
 190 void
 191 fail_point_destroy(struct fail_point *fp)
 192 {
 193         struct fail_point_entry *ent;
 194
 195         if (fp->fp_flags & FAIL_POINT_DYNAMIC_NAME && fp->fp_name != NULL) {
 196                 fp_free((void *)(intptr_t)fp->fp_name);
 197                 fp->fp_name = NULL;
 198         }
 199         fp->fp_flags = 0;
 200
 201         while (!TAILQ_EMPTY(&fp->fp_entries)) {
 202                 ent = TAILQ_FIRST(&fp->fp_entries);
 203                 TAILQ_REMOVE(&fp->fp_entries, ent, fe_entries);
 204                 fp_free(ent);
 205         }
 206 }
 207
 208 /**
 209  * This does the real work of evaluating a fail point. If the fail point tells
 210  * us to return a value, this function returns 1 and fills in 'return_value'
 211  * (return_value is allowed to be null). If the fail point tells us to panic,
 212  * we never return. Otherwise we just return 0 after doing some work, which
 213  * means "keep going".
 214  */
 215 enum fail_point_return_code
 216 fail_point_eval_nontrivial(struct fail_point *fp, int *return_value)
 217 {
 218         enum fail_point_return_code ret = FAIL_POINT_RC_CONTINUE;
 219         struct fail_point_entry *ent, *next;
 220         int msecs;
 221
 222         FP_LOCK();
 223
 224         ent = TAILQ_FIRST(&fp->fp_entries);
 225         while (ent) {
 226                 int cont = 0; /* don't continue by default */
 227                 next = TAILQ_NEXT(ent, fe_entries);
 228
 229                 if (ent->fe_prob < PROB_MAX &&
 230                     ent->fe_prob < random() % PROB_MAX) {
 231                         cont = 1;
 232                         goto loop_end;
 233                 }
 234
 235                 switch (ent->fe_type) {
 236                 case FAIL_POINT_PANIC:
 237                         panic("fail point %s panicking", fp->fp_name);
 238                         /* NOTREACHED */
 239
 240                 case FAIL_POINT_RETURN:
 241                         if (return_value)
 242                                 *return_value = ent->fe_arg;
 243                         ret = FAIL_POINT_RC_RETURN;
 244                         break;
 245
 246                 case FAIL_POINT_BREAK:
 247                         printf("fail point %s breaking to debugger\n", fp->fp_name);
 248                         breakpoint();
 249                         break;
 250
 251                 case FAIL_POINT_PRINT:
 252                         printf("fail point %s executing\n", fp->fp_name);
 253                         cont = ent->fe_arg;
 254                         break;
 255
 256                 case FAIL_POINT_SLEEP:
 257                         /*
 258                          * Free the entry now if necessary, since
 259                          * we're about to drop the mutex and sleep.
 260                          */
 261                         msecs = ent->fe_arg;
 262                         if (ent->fe_count > 0 && --ent->fe_count == 0) {
 263                                 free_entry(&fp->fp_entries, ent);
 264                                 ent = NULL;
 265                         }
 266
 267                         if (msecs)
 268                                 fail_point_sleep(fp, ent, msecs, &ret);
 269                         break;
 270
 271                 default:
 272                         break;
 273                 }
 274
 275                 if (ent && ent->fe_count > 0 && --ent->fe_count == 0)
 276                         free_entry(&fp->fp_entries, ent);
 277
 278 loop_end:
 279                 if (cont)
 280                         ent = next;
 281                 else
 282                         break;
 283         }
 284
 285         /* Get rid of "off"s at the end. */
 286         while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
 287                ent->fe_type == FAIL_POINT_OFF)
 288                 free_entry(&fp->fp_entries, ent);
 289
 290         FP_UNLOCK();
 291
 292         return ret;
 293 }
 294
 295 /**
 296  * Translate internal fail_point structure into human-readable text.
 297  */
 298 static void
 299 fail_point_get(struct fail_point *fp, struct sbuf *sb)
 300 {
 301         struct fail_point_entry *ent;
 302
 303         FP_LOCK();
 304
 305         TAILQ_FOREACH(ent, &fp->fp_entries, fe_entries) {
 306                 if (ent->fe_prob < PROB_MAX) {
 307                         int decimal = ent->fe_prob % (PROB_MAX / 100);
 308                         int units = ent->fe_prob / (PROB_MAX / 100);
 309                         sbuf_printf(sb, "%d", units);
 310                         if (decimal) {
 311                                 int digits = PROB_DIGITS - 2;
 312                                 while (!(decimal % 10)) {
 313                                         digits--;
 314                                         decimal /= 10;
 315                                 }
 316                                 sbuf_printf(sb, ".%0*d", digits, decimal);
 317                         }
 318                         sbuf_printf(sb, "%%");
 319                 }
 320                 if (ent->fe_count > 0)
 321                         sbuf_printf(sb, "%d*", ent->fe_count);
 322                 sbuf_printf(sb, "%s", fail_type_strings[ent->fe_type]);
 323                 if (ent->fe_arg)
 324                         sbuf_printf(sb, "(%d)", ent->fe_arg);
 325                 if (TAILQ_NEXT(ent, fe_entries))
 326                         sbuf_printf(sb, "->");
 327         }
 328         if (TAILQ_EMPTY(&fp->fp_entries))
 329                 sbuf_printf(sb, "off");
 330
 331         FP_UNLOCK();
 332 }
 333
 334 /**
 335  * Set an internal fail_point structure from a human-readable failpoint string
 336  * in a lock-safe manner.
 337  */
 338 static int
 339 fail_point_set(struct fail_point *fp, char *buf)
 340 {
 341         int error = 0;
 342         struct fail_point_entry *ent, *ent_next;
 343         struct fail_point_entries new_entries;
 344
 345         /* Parse new entries. */
 346         TAILQ_INIT(&new_entries);
 347         if (!parse_fail_point(&new_entries, buf)) {
 348                 clear_entries(&new_entries);
 349                 error = EINVAL;
 350                 goto end;
 351         }
 352
 353         FP_LOCK();
 354
 355         /* Move new entries in. */
 356         TAILQ_SWAP(&fp->fp_entries, &new_entries, fail_point_entry, fe_entries);
 357         clear_entries(&new_entries);
 358
 359         /* Get rid of useless zero probability entries. */
 360         TAILQ_FOREACH_SAFE(ent, &fp->fp_entries, fe_entries, ent_next) {
 361                 if (ent->fe_prob == 0)
 362                         free_entry(&fp->fp_entries, ent);
 363         }
 364
 365         /* Get rid of "off"s at the end. */
 366         while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
 367                 ent->fe_type == FAIL_POINT_OFF)
 368                 free_entry(&fp->fp_entries, ent);
 369
 370         FP_UNLOCK();
 371
 372  end:
 373 #ifdef IWARNING
 374         if (error)
 375                 IWARNING("Failed to set %s (%s) to %s",
 376                     fp->fp_name, fp->fp_location, buf);
 377         else
 378                 INOTICE("Set %s (%s) to %s",
 379                     fp->fp_name, fp->fp_location, buf);
 380 #endif /* IWARNING */
 381
 382         return error;
 383 }
 384
 385 #define MAX_FAIL_POINT_BUF      1023
 386
 387 /**
 388  * Handle kernel failpoint set/get.
 389  */
 390 int
 391 fail_point_sysctl(SYSCTL_HANDLER_ARGS)
 392 {
 393         struct fail_point *fp = arg1;
 394         char *buf = NULL;
 395         struct sbuf sb;
 396         int error;
 397
 398         /* Retrieving */
 399         sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND);
 400         fail_point_get(fp, &sb);
 401         sbuf_trim(&sb);
 402         sbuf_finish(&sb);
 403         error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
 404         sbuf_delete(&sb);
 405
 406         /* Setting */
 407         if (!error && req->newptr) {
 408                 if (req->newlen > MAX_FAIL_POINT_BUF) {
 409                         error = EINVAL;
 410                         goto out;
 411                 }
 412
 413                 buf = fp_malloc(req->newlen + 1, M_WAITOK);
 414
 415                 error = SYSCTL_IN(req, buf, req->newlen);
 416                 if (error)
 417                         goto out;
 418                 buf[req->newlen] = '\0';
 419
 420                 error = fail_point_set(fp, buf);
 421         }
 422
 423 out:
 424         if (buf)
 425                 fp_free(buf);
 426         return error;
 427 }
 428
 429 /**
 430  * Internal helper function to translate a human-readable failpoint string
 431  * into a internally-parsable fail_point structure.
 432  */
 433 static char *
 434 parse_fail_point(struct fail_point_entries *ents, char *p)
 435 {
 436         /*  <fail_point> ::
 437          *      <term> ( "->" <term> )*
 438          */
 439         if (!(p = parse_term(ents, p)))
 440                 return 0;
 441         while (*p)
 442                 if (p[0] != '-' || p[1] != '>' || !(p = parse_term(ents, p+2)))
 443                         return 0;
 444         return p;
 445 }
 446
 447 /**
 448  * Internal helper function to parse an individual term from a failpoint.
 449  */
 450 static char *
 451 parse_term(struct fail_point_entries *ents, char *p)
 452 {
 453         struct fail_point_entry *ent;
 454
 455         ent = fp_malloc(sizeof *ent, M_WAITOK | M_ZERO);
 456         ent->fe_prob = PROB_MAX;
 457         TAILQ_INSERT_TAIL(ents, ent, fe_entries);
 458
 459         /*
 460          * <term> ::
 461          *     ( (<float> "%") | (<integer> "*" ) )*
 462          *     <type>
 463          *     [ "(" <integer> ")" ]
 464          */
 465
 466         /* ( (<float> "%") | (<integer> "*" ) )* */
 467         while (('0' <= *p && *p <= '9') || *p == '.') {
 468                 int units, decimal;
 469
 470                 if (!(p = parse_number(&units, &decimal, p)))
 471                         return 0;
 472
 473                 if (*p == '%') {
 474                         if (units > 100) /* prevent overflow early */
 475                                 units = 100;
 476                         ent->fe_prob = units * (PROB_MAX / 100) + decimal;
 477                         if (ent->fe_prob > PROB_MAX)
 478                                 ent->fe_prob = PROB_MAX;
 479
 480                 } else if (*p == '*') {
 481                         if (!units || decimal)
 482                                 return 0;
 483                         ent->fe_count = units;
 484
 485                 } else {
 486                         return 0;
 487                 }
 488
 489                 p++;
 490         }
 491
 492         /* <type> */
 493         if (!(p = parse_type(ent, p)))
 494                 return 0;
 495         if (*p == '\0')
 496                 return p;
 497
 498         /* [ "(" <integer> ")" ] */
 499         if (*p != '(')
 500                 return p;
 501         p++;
 502         if (('0' <= *p && *p <= '9') || *p == '-')
 503                 ent->fe_arg = strtol(p, &p, 0);
 504         else
 505                 return 0;
 506         if (*p++ != ')')
 507                 return 0;
 508
 509         return p;
 510 }
 511
 512 /**
 513  * Internal helper function to parse a numeric for a failpoint term.
 514  */
 515 static char *
 516 parse_number(int *out_units, int *out_decimal, char *p)
 517 {
 518         char *old_p;
 519
 520         /*
 521          *  <number> ::
 522          *      <integer> [ "." <integer> ] |
 523          *      "." <integer>
 524          */
 525
 526         /* whole part */
 527         old_p = p;
 528         *out_units = strtol(p, &p, 10);
 529         if (p == old_p && *p != '.')
 530                 return 0;
 531
 532         /* fractional part */
 533         *out_decimal = 0;
 534         if (*p == '.') {
 535                 int digits = 0;
 536                 p++;
 537                 while ('0' <= *p && *p <= '9') {
 538                         int digit = *p - '0';
 539                         if (digits < PROB_DIGITS - 2)
 540                                 *out_decimal = *out_decimal * 10 + digit;
 541                         else if (digits == PROB_DIGITS - 2 && digit >= 5)
 542                                 (*out_decimal)++;
 543                         digits++;
 544                         p++;
 545                 }
 546                 if (!digits) /* need at least one digit after '.' */
 547                         return 0;
 548                 while (digits++ < PROB_DIGITS - 2) /* add implicit zeros */
 549                         *out_decimal *= 10;
 550         }
 551
 552         return p; /* success */
 553 }
 554
 555 /**
 556  * Internal helper function to parse an individual type for a failpoint term.
 557  */
 558 static char *
 559 parse_type(struct fail_point_entry *ent, char *beg)
 560 {
 561         enum fail_point_t type;
 562         char *end = beg;
 563         while ('a' <= *end && *end <= 'z')
 564                 end++;
 565         if (beg == end)
 566                 return 0;
 567         for (type = FAIL_POINT_OFF; type != FAIL_POINT_INVALID; type++) {
 568                 const char *p = fail_type_strings[type];
 569                 const char *q = beg;
 570                 while (q < end && *p++ == *q++);
 571                 if (q == end && *p == '\0') {
 572                         ent->fe_type = type;
 573                         return end;
 574                 }
 575         }
 576         return 0;
 577 }
 578
 579 /**
 580  * Internal helper function to free an individual failpoint term.
 581  */
 582 static void
 583 free_entry(struct fail_point_entries *ents, struct fail_point_entry *ent)
 584 {
 585         TAILQ_REMOVE(ents, ent, fe_entries);
 586         fp_free(ent);
 587 }
 588
 589 /**
 590  * Internal helper function to clear out all failpoint terms for a single
 591  * failpoint.
 592  */
 593 static void
 594 clear_entries(struct fail_point_entries *ents)
 595 {
 596         struct fail_point_entry *ent, *ent_next;
 597         TAILQ_FOREACH_SAFE(ent, ents, fe_entries, ent_next)
 598                 fp_free(ent);
 599         TAILQ_INIT(ents);
 600 }
 601
 602 /* The fail point sysctl tree. */
 603 SYSCTL_NODE(_debug, OID_AUTO, fail_point, CTLFLAG_RW, 0, "fail points");