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_location = "";
 181         fp->fp_flags |= FAIL_POINT_DYNAMIC_NAME;
 182         fp->fp_sleep_fn = NULL;
 183         fp->fp_sleep_arg = NULL;
 184 }
 185
 186 /**
 187  * Free the resources held by a fail_point.
 188  *
 189  * @ingroup failpoint
 190  */
 191 void
 192 fail_point_destroy(struct fail_point *fp)
 193 {
 194         struct fail_point_entry *ent;
 195
 196         if (fp->fp_flags & FAIL_POINT_DYNAMIC_NAME && fp->fp_name != NULL) {
 197                 fp_free((void *)(intptr_t)fp->fp_name);
 198                 fp->fp_name = NULL;
 199         }
 200         fp->fp_flags = 0;
 201
 202         while (!TAILQ_EMPTY(&fp->fp_entries)) {
 203                 ent = TAILQ_FIRST(&fp->fp_entries);
 204                 TAILQ_REMOVE(&fp->fp_entries, ent, fe_entries);
 205                 fp_free(ent);
 206         }
 207 }
 208
 209 /**
 210  * This does the real work of evaluating a fail point. If the fail point tells
 211  * us to return a value, this function returns 1 and fills in 'return_value'
 212  * (return_value is allowed to be null). If the fail point tells us to panic,
 213  * we never return. Otherwise we just return 0 after doing some work, which
 214  * means "keep going".
 215  */
 216 enum fail_point_return_code
 217 fail_point_eval_nontrivial(struct fail_point *fp, int *return_value)
 218 {
 219         enum fail_point_return_code ret = FAIL_POINT_RC_CONTINUE;
 220         struct fail_point_entry *ent, *next;
 221         int msecs;
 222
 223         FP_LOCK();
 224
 225         ent = TAILQ_FIRST(&fp->fp_entries);
 226         while (ent) {
 227                 int cont = 0; /* don't continue by default */
 228                 next = TAILQ_NEXT(ent, fe_entries);
 229
 230                 if (ent->fe_prob < PROB_MAX &&
 231                     ent->fe_prob < random() % PROB_MAX) {
 232                         cont = 1;
 233                         goto loop_end;
 234                 }
 235
 236                 switch (ent->fe_type) {
 237                 case FAIL_POINT_PANIC:
 238                         panic("fail point %s panicking", fp->fp_name);
 239                         /* NOTREACHED */
 240
 241                 case FAIL_POINT_RETURN:
 242                         if (return_value)
 243                                 *return_value = ent->fe_arg;
 244                         ret = FAIL_POINT_RC_RETURN;
 245                         break;
 246
 247                 case FAIL_POINT_BREAK:
 248                         printf("fail point %s breaking to debugger\n", fp->fp_name);
 249                         breakpoint();
 250                         break;
 251
 252                 case FAIL_POINT_PRINT:
 253                         printf("fail point %s executing\n", fp->fp_name);
 254                         cont = ent->fe_arg;
 255                         break;
 256
 257                 case FAIL_POINT_SLEEP:
 258                         /*
 259                          * Free the entry now if necessary, since
 260                          * we're about to drop the mutex and sleep.
 261                          */
 262                         msecs = ent->fe_arg;
 263                         if (ent->fe_count > 0 && --ent->fe_count == 0) {
 264                                 free_entry(&fp->fp_entries, ent);
 265                                 ent = NULL;
 266                         }
 267
 268                         if (msecs)
 269                                 fail_point_sleep(fp, ent, msecs, &ret);
 270                         break;
 271
 272                 default:
 273                         break;
 274                 }
 275
 276                 if (ent && ent->fe_count > 0 && --ent->fe_count == 0)
 277                         free_entry(&fp->fp_entries, ent);
 278
 279 loop_end:
 280                 if (cont)
 281                         ent = next;
 282                 else
 283                         break;
 284         }
 285
 286         /* Get rid of "off"s at the end. */
 287         while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
 288                ent->fe_type == FAIL_POINT_OFF)
 289                 free_entry(&fp->fp_entries, ent);
 290
 291         FP_UNLOCK();
 292
 293         return ret;
 294 }
 295
 296 /**
 297  * Translate internal fail_point structure into human-readable text.
 298  */
 299 static void
 300 fail_point_get(struct fail_point *fp, struct sbuf *sb)
 301 {
 302         struct fail_point_entry *ent;
 303
 304         FP_LOCK();
 305
 306         TAILQ_FOREACH(ent, &fp->fp_entries, fe_entries) {
 307                 if (ent->fe_prob < PROB_MAX) {
 308                         int decimal = ent->fe_prob % (PROB_MAX / 100);
 309                         int units = ent->fe_prob / (PROB_MAX / 100);
 310                         sbuf_printf(sb, "%d", units);
 311                         if (decimal) {
 312                                 int digits = PROB_DIGITS - 2;
 313                                 while (!(decimal % 10)) {
 314                                         digits--;
 315                                         decimal /= 10;
 316                                 }
 317                                 sbuf_printf(sb, ".%0*d", digits, decimal);
 318                         }
 319                         sbuf_printf(sb, "%%");
 320                 }
 321                 if (ent->fe_count > 0)
 322                         sbuf_printf(sb, "%d*", ent->fe_count);
 323                 sbuf_printf(sb, "%s", fail_type_strings[ent->fe_type]);
 324                 if (ent->fe_arg)
 325                         sbuf_printf(sb, "(%d)", ent->fe_arg);
 326                 if (TAILQ_NEXT(ent, fe_entries))
 327                         sbuf_printf(sb, "->");
 328         }
 329         if (TAILQ_EMPTY(&fp->fp_entries))
 330                 sbuf_printf(sb, "off");
 331
 332         FP_UNLOCK();
 333 }
 334
 335 /**
 336  * Set an internal fail_point structure from a human-readable failpoint string
 337  * in a lock-safe manner.
 338  */
 339 static int
 340 fail_point_set(struct fail_point *fp, char *buf)
 341 {
 342         int error = 0;
 343         struct fail_point_entry *ent, *ent_next;
 344         struct fail_point_entries new_entries;
 345
 346         /* Parse new entries. */
 347         TAILQ_INIT(&new_entries);
 348         if (!parse_fail_point(&new_entries, buf)) {
 349                 clear_entries(&new_entries);
 350                 error = EINVAL;
 351                 goto end;
 352         }
 353
 354         FP_LOCK();
 355
 356         /* Move new entries in. */
 357         TAILQ_SWAP(&fp->fp_entries, &new_entries, fail_point_entry, fe_entries);
 358         clear_entries(&new_entries);
 359
 360         /* Get rid of useless zero probability entries. */
 361         TAILQ_FOREACH_SAFE(ent, &fp->fp_entries, fe_entries, ent_next) {
 362                 if (ent->fe_prob == 0)
 363                         free_entry(&fp->fp_entries, ent);
 364         }
 365
 366         /* Get rid of "off"s at the end. */
 367         while ((ent = TAILQ_LAST(&fp->fp_entries, fail_point_entries)) &&
 368                 ent->fe_type == FAIL_POINT_OFF)
 369                 free_entry(&fp->fp_entries, ent);
 370
 371         FP_UNLOCK();
 372
 373  end:
 374 #ifdef IWARNING
 375         if (error)
 376                 IWARNING("Failed to set %s %s to %s",
 377                     fp->fp_name, fp->fp_location, buf);
 378         else
 379                 INOTICE("Set %s %s to %s",
 380                     fp->fp_name, fp->fp_location, buf);
 381 #endif /* IWARNING */
 382
 383         return error;
 384 }
 385
 386 #define MAX_FAIL_POINT_BUF      1023
 387
 388 /**
 389  * Handle kernel failpoint set/get.
 390  */
 391 int
 392 fail_point_sysctl(SYSCTL_HANDLER_ARGS)
 393 {
 394         struct fail_point *fp = arg1;
 395         char *buf = NULL;
 396         struct sbuf sb;
 397         int error;
 398
 399         /* Retrieving */
 400         sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND);
 401         fail_point_get(fp, &sb);
 402         sbuf_trim(&sb);
 403         sbuf_finish(&sb);
 404         error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
 405         sbuf_delete(&sb);
 406
 407         /* Setting */
 408         if (!error && req->newptr) {
 409                 if (req->newlen > MAX_FAIL_POINT_BUF) {
 410                         error = EINVAL;
 411                         goto out;
 412                 }
 413
 414                 buf = fp_malloc(req->newlen + 1, M_WAITOK);
 415
 416                 error = SYSCTL_IN(req, buf, req->newlen);
 417                 if (error)
 418                         goto out;
 419                 buf[req->newlen] = '\0';
 420
 421                 error = fail_point_set(fp, buf);
 422         }
 423
 424 out:
 425         if (buf)
 426                 fp_free(buf);
 427         return error;
 428 }
 429
 430 /**
 431  * Internal helper function to translate a human-readable failpoint string
 432  * into a internally-parsable fail_point structure.
 433  */
 434 static char *
 435 parse_fail_point(struct fail_point_entries *ents, char *p)
 436 {
 437         /*  <fail_point> ::
 438          *      <term> ( "->" <term> )*
 439          */
 440         if (!(p = parse_term(ents, p)))
 441                 return 0;
 442         while (*p)
 443                 if (p[0] != '-' || p[1] != '>' || !(p = parse_term(ents, p+2)))
 444                         return 0;
 445         return p;
 446 }
 447
 448 /**
 449  * Internal helper function to parse an individual term from a failpoint.
 450  */
 451 static char *
 452 parse_term(struct fail_point_entries *ents, char *p)
 453 {
 454         struct fail_point_entry *ent;
 455
 456         ent = fp_malloc(sizeof *ent, M_WAITOK | M_ZERO);
 457         ent->fe_prob = PROB_MAX;
 458         TAILQ_INSERT_TAIL(ents, ent, fe_entries);
 459
 460         /*
 461          * <term> ::
 462          *     ( (<float> "%") | (<integer> "*" ) )*
 463          *     <type>
 464          *     [ "(" <integer> ")" ]
 465          */
 466
 467         /* ( (<float> "%") | (<integer> "*" ) )* */
 468         while (('0' <= *p && *p <= '9') || *p == '.') {
 469                 int units, decimal;
 470
 471                 if (!(p = parse_number(&units, &decimal, p)))
 472                         return 0;
 473
 474                 if (*p == '%') {
 475                         if (units > 100) /* prevent overflow early */
 476                                 units = 100;
 477                         ent->fe_prob = units * (PROB_MAX / 100) + decimal;
 478                         if (ent->fe_prob > PROB_MAX)
 479                                 ent->fe_prob = PROB_MAX;
 480
 481                 } else if (*p == '*') {
 482                         if (!units || decimal)
 483                                 return 0;
 484                         ent->fe_count = units;
 485
 486                 } else {
 487                         return 0;
 488                 }
 489
 490                 p++;
 491         }
 492
 493         /* <type> */
 494         if (!(p = parse_type(ent, p)))
 495                 return 0;
 496         if (*p == '\0')
 497                 return p;
 498
 499         /* [ "(" <integer> ")" ] */
 500         if (*p != '(')
 501                 return p;
 502         p++;
 503         if (('0' <= *p && *p <= '9') || *p == '-')
 504                 ent->fe_arg = strtol(p, &p, 0);
 505         else
 506                 return 0;
 507         if (*p++ != ')')
 508                 return 0;
 509
 510         return p;
 511 }
 512
 513 /**
 514  * Internal helper function to parse a numeric for a failpoint term.
 515  */
 516 static char *
 517 parse_number(int *out_units, int *out_decimal, char *p)
 518 {
 519         char *old_p;
 520
 521         /*
 522          *  <number> ::
 523          *      <integer> [ "." <integer> ] |
 524          *      "." <integer>
 525          */
 526
 527         /* whole part */
 528         old_p = p;
 529         *out_units = strtol(p, &p, 10);
 530         if (p == old_p && *p != '.')
 531                 return 0;
 532
 533         /* fractional part */
 534         *out_decimal = 0;
 535         if (*p == '.') {
 536                 int digits = 0;
 537                 p++;
 538                 while ('0' <= *p && *p <= '9') {
 539                         int digit = *p - '0';
 540                         if (digits < PROB_DIGITS - 2)
 541                                 *out_decimal = *out_decimal * 10 + digit;
 542                         else if (digits == PROB_DIGITS - 2 && digit >= 5)
 543                                 (*out_decimal)++;
 544                         digits++;
 545                         p++;
 546                 }
 547                 if (!digits) /* need at least one digit after '.' */
 548                         return 0;
 549                 while (digits++ < PROB_DIGITS - 2) /* add implicit zeros */
 550                         *out_decimal *= 10;
 551         }
 552
 553         return p; /* success */
 554 }
 555
 556 /**
 557  * Internal helper function to parse an individual type for a failpoint term.
 558  */
 559 static char *
 560 parse_type(struct fail_point_entry *ent, char *beg)
 561 {
 562         enum fail_point_t type;
 563         char *end = beg;
 564         while ('a' <= *end && *end <= 'z')
 565                 end++;
 566         if (beg == end)
 567                 return 0;
 568         for (type = FAIL_POINT_OFF; type != FAIL_POINT_INVALID; type++) {
 569                 const char *p = fail_type_strings[type];
 570                 const char *q = beg;
 571                 while (q < end && *p++ == *q++);
 572                 if (q == end && *p == '\0') {
 573                         ent->fe_type = type;
 574                         return end;
 575                 }
 576         }
 577         return 0;
 578 }
 579
 580 /**
 581  * Internal helper function to free an individual failpoint term.
 582  */
 583 static void
 584 free_entry(struct fail_point_entries *ents, struct fail_point_entry *ent)
 585 {
 586         TAILQ_REMOVE(ents, ent, fe_entries);
 587         fp_free(ent);
 588 }
 589
 590 /**
 591  * Internal helper function to clear out all failpoint terms for a single
 592  * failpoint.
 593  */
 594 static void
 595 clear_entries(struct fail_point_entries *ents)
 596 {
 597         struct fail_point_entry *ent, *ent_next;
 598         TAILQ_FOREACH_SAFE(ent, ents, fe_entries, ent_next)
 599                 fp_free(ent);
 600         TAILQ_INIT(ents);
 601 }
 602
 603 /* The fail point sysctl tree. */
 604 SYSCTL_NODE(_debug, OID_AUTO, fail_point, CTLFLAG_RW, 0, "fail points");