filters/multi-conn/multi-conn.c

   1 /* nbdkit
   2  * Copyright Red Hat
   3  *
   4  * Redistribution and use in source and binary forms, with or without
   5  * modification, are permitted provided that the following conditions are
   6  * met:
   7  *
   8  * * Redistributions of source code must retain the above copyright
   9  * notice, this list of conditions and the following disclaimer.
  10  *
  11  * * Redistributions in binary form must reproduce the above copyright
  12  * notice, this list of conditions and the following disclaimer in the
  13  * documentation and/or other materials provided with the distribution.
  14  *
  15  * * Neither the name of Red Hat nor the names of its contributors may be
  16  * used to endorse or promote products derived from this software without
  17  * specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY RED HAT AND CONTRIBUTORS ''AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
  21  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
  22  * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL RED HAT OR
  23  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  24  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  25  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
  26  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
  27  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  28  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  29  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 #include <config.h>
  34
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <stdint.h>
  38 #include <string.h>
  39 #include <stdbool.h>
  40 #include <assert.h>
  41 #include <pthread.h>
  42
  43 #include <nbdkit-filter.h>
  44
  45 #include "cleanup.h"
  46 #include "vector.h"
  47
  48 /* Track results of .config */
  49 static enum MultiConnMode {
  50   AUTO,
  51   EMULATE,
  52   PLUGIN,
  53   UNSAFE,
  54   DISABLE,
  55 } mode;
  56
  57 static enum TrackDirtyMode {
  58   CONN,
  59   FAST,
  60   OFF,
  61 } track;
  62
  63 static bool byname = false;
  64
  65 enum dirty {
  66   WRITE = 1, /* A write may have populated a cache */
  67   READ = 2, /* A read may have populated a cache */
  68 };
  69
  70 /* Coordination between connections. */
  71 static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;
  72
  73 /* The list of handles to active connections. */
  74 struct handle {
  75   nbdkit_next *next;
  76   enum MultiConnMode mode; /* Runtime resolution of mode==AUTO */
  77   enum dirty dirty; /* What aspects of this connection are dirty */
  78   char *name; /* Used when byname==true to assign group */
  79   struct group *group; /* All connections grouped with this one */
  80 };
  81 DEFINE_VECTOR_TYPE (conns_vector, struct handle *);
  82 struct group {
  83   conns_vector conns;
  84   char *name;
  85   bool dirty; /* True if any connection in group is dirty */
  86 };
  87 DEFINE_VECTOR_TYPE (group_vector, struct group *);
  88 static group_vector groups = empty_vector;
  89
  90 /* Accept 'multi-conn-mode=mode', 'multi-conn-track-dirty=level', and
  91  * 'multi-conn-exportname=bool'.
  92  */
  93 static int
  94 multi_conn_config (nbdkit_next_config *next, nbdkit_backend *nxdata,
  95                    const char *key, const char *value)
  96 {
  97   if (strcmp (key, "multi-conn-mode") == 0) {
  98     if (strcmp (value, "auto") == 0)
  99       mode = AUTO;
 100     else if (strcmp (value, "emulate") == 0)
 101       mode = EMULATE;
 102     else if (strcmp (value, "plugin") == 0)
 103       mode = PLUGIN;
 104     else if (strcmp (value, "disable") == 0)
 105       mode = DISABLE;
 106     else if (strcmp (value, "unsafe") == 0)
 107       mode = UNSAFE;
 108     else {
 109       nbdkit_error ("unknown multi-conn mode '%s'", value);
 110       return -1;
 111     }
 112     return 0;
 113   }
 114   else if (strcmp (key, "multi-conn-track-dirty") == 0) {
 115     if (strcmp (value, "connection") == 0 ||
 116         strcmp (value, "conn") == 0)
 117       track = CONN;
 118     else if (strcmp (value, "fast") == 0)
 119       track = FAST;
 120     else if (strcmp (value, "off") == 0)
 121       track = OFF;
 122     else {
 123       nbdkit_error ("unknown multi-conn track-dirty setting '%s'", value);
 124       return -1;
 125     }
 126     return 0;
 127   }
 128   else if (strcmp (key, "multi-conn-exportname") == 0 ||
 129            strcmp (key, "multi-conn-export-name") == 0) {
 130     int r;
 131
 132     r = nbdkit_parse_bool (value);
 133     if (r == -1)
 134       return -1;
 135     byname = r;
 136     return 0;
 137   }
 138   return next (nxdata, key, value);
 139 }
 140
 141 #define multi_conn_config_help \
 142   "multi-conn-mode=<MODE>          'auto' (default), 'emulate', 'plugin',\n" \
 143   "                                'disable', or 'unsafe'.\n" \
 144   "multi-conn-track-dirty=<LEVEL>  'conn' (default), 'fast', or 'off'.\n" \
 145   "multi-conn-exportname=<BOOL>    true to limit emulation by export name.\n"
 146
 147 static int
 148 multi_conn_get_ready (int thread_model)
 149 {
 150   if (thread_model == NBDKIT_THREAD_MODEL_SERIALIZE_CONNECTIONS &&
 151       mode == AUTO)
 152     mode = DISABLE;
 153   return 0;
 154 }
 155
 156 static void
 157 multi_conn_unload (void)
 158 {
 159   assert (groups.len == 0);
 160   group_vector_reset (&groups);
 161 }
 162
 163 static void *
 164 multi_conn_open (nbdkit_next_open *next, nbdkit_context *nxdata,
 165                  int readonly, const char *exportname, int is_tls)
 166 {
 167   struct handle *h;
 168
 169   if (next (nxdata, readonly, exportname) == -1)
 170     return NULL;
 171
 172   /* Allocate here, but populate and insert into list in .prepare */
 173   h = calloc (1, sizeof *h);
 174   if (h == NULL) {
 175     nbdkit_error ("calloc: %m");
 176     return NULL;
 177   }
 178   if (byname) {
 179     h->name = strdup (exportname);
 180     if (h->name == NULL) {
 181       nbdkit_error ("strdup: %m");
 182       free (h);
 183       return NULL;
 184     }
 185   }
 186   return h;
 187 }
 188
 189 static int
 190 multi_conn_prepare (nbdkit_next *next, void *handle, int readonly)
 191 {
 192   struct handle *h = handle;
 193   struct group *g;
 194   bool new_group = false;
 195   int r;
 196   size_t i;
 197
 198   h->next = next;
 199   if (mode == AUTO) { /* See also .get_ready turning AUTO into DISABLE */
 200     r = next->can_multi_conn (next);
 201     if (r == -1)
 202       return -1;
 203     if (r == 0)
 204       h->mode = EMULATE;
 205     else
 206       h->mode = PLUGIN;
 207   }
 208   else
 209     h->mode = mode;
 210   if (h->mode == EMULATE && next->can_flush (next) != 1) {
 211     nbdkit_error ("emulating multi-conn requires working flush");
 212     return -1;
 213   }
 214
 215   ACQUIRE_LOCK_FOR_CURRENT_SCOPE (&lock);
 216   if (byname) {
 217     g = NULL;
 218     for (i = 0; i < groups.len; i++)
 219       if (strcmp (groups.ptr[i]->name, h->name) == 0) {
 220         g = groups.ptr[i];
 221         break;
 222       }
 223   }
 224   else
 225     g = groups.len ? groups.ptr[0] : NULL;
 226
 227   if (!g) {
 228     g = calloc (1, sizeof *g);
 229     if (g == NULL) {
 230       nbdkit_error ("calloc: %m");
 231       return -1;
 232     }
 233     if (group_vector_append (&groups, g) == -1) {
 234       free (g);
 235       return -1;
 236     }
 237     g->name = h->name;
 238     h->name = NULL;
 239     new_group = true;
 240   }
 241   if (conns_vector_append (&g->conns, h) == -1) {
 242     if (new_group) {
 243       group_vector_remove (&groups, groups.len - 1);
 244       free (g->name);
 245       free (g);
 246     }
 247     return -1;
 248   }
 249   h->group = g;
 250   return 0;
 251 }
 252
 253 static int
 254 multi_conn_finalize (nbdkit_next *next, void *handle)
 255 {
 256   struct handle *h = handle;
 257   size_t i;
 258
 259   ACQUIRE_LOCK_FOR_CURRENT_SCOPE (&lock);
 260   assert (h->next == next);
 261   assert (h->group);
 262
 263   /* XXX should we add a config param to flush if the client forgot? */
 264   for (i = 0; i < h->group->conns.len; i++) {
 265     if (h->group->conns.ptr[i] == h) {
 266       conns_vector_remove (&h->group->conns, i);
 267       break;
 268     }
 269   }
 270   if (h->group->conns.len == 0) {
 271     for (i = 0; i < groups.len; i++)
 272       if (groups.ptr[i] == h->group) {
 273         group_vector_remove (&groups, i);
 274         free (h->group->name);
 275         free (h->group);
 276         break;
 277       }
 278   }
 279   h->group = NULL;
 280   return 0;
 281 }
 282
 283 static void
 284 multi_conn_close (void *handle)
 285 {
 286   struct handle *h = handle;
 287
 288   assert (h->group == NULL);
 289   free (h->name);
 290   free (h);
 291 }
 292
 293 static int
 294 multi_conn_can_fua (nbdkit_next *next, void *handle)
 295 {
 296   /* If the backend has native FUA support but is not multi-conn
 297    * consistent, and we have to flush on every connection, then we are
 298    * better off advertising emulated fua rather than native.
 299    */
 300   struct handle *h = handle;
 301   int fua = next->can_fua (next);
 302
 303   assert (h->mode != AUTO);
 304   if (fua == NBDKIT_FUA_NATIVE && h->mode == EMULATE)
 305     return NBDKIT_FUA_EMULATE;
 306   return fua;
 307 }
 308
 309 static int
 310 multi_conn_can_multi_conn (nbdkit_next *next, void *handle)
 311 {
 312   struct handle *h = handle;
 313
 314   switch (h->mode) {
 315   case EMULATE:
 316     return 1;
 317   case PLUGIN:
 318     return next->can_multi_conn (next);
 319   case DISABLE:
 320     return 0;
 321   case UNSAFE:
 322     return 1;
 323   case AUTO: /* Not possible, see .prepare */
 324   default:
 325     abort ();
 326   }
 327 }
 328
 329 static void
 330 mark_dirty (struct handle *h, bool is_write)
 331 {
 332   /* No need to grab lock here: the NBD spec is clear that a client
 333    * must wait for the response to a flush before sending the next
 334    * command that expects to see the result of that flush, so any race
 335    * in accessing dirty can be traced back to the client improperly
 336    * sending a flush in parallel with other live commands.
 337    */
 338   switch (track) {
 339   case CONN:
 340     h->dirty |= is_write ? WRITE : READ;
 341     /* fallthrough */
 342   case FAST:
 343     if (is_write)
 344       h->group->dirty = true;
 345     break;
 346   case OFF:
 347     break;
 348   default:
 349     abort ();
 350   }
 351 }
 352
 353 static int
 354 multi_conn_flush (nbdkit_next *next,
 355                   void *handle, uint32_t flags, int *err);
 356
 357 static int
 358 multi_conn_pread (nbdkit_next *next,
 359                   void *handle, void *buf, uint32_t count, uint64_t offs,
 360                   uint32_t flags, int *err)
 361 {
 362   struct handle *h = handle;
 363
 364   mark_dirty (h, false);
 365   return next->pread (next, buf, count, offs, flags, err);
 366 }
 367
 368 static int
 369 multi_conn_pwrite (nbdkit_next *next,
 370                    void *handle, const void *buf, uint32_t count,
 371                    uint64_t offs, uint32_t flags, int *err)
 372 {
 373   struct handle *h = handle;
 374   bool need_flush = false;
 375
 376   if (flags & NBDKIT_FLAG_FUA) {
 377     if (h->mode == EMULATE) {
 378       mark_dirty (h, true);
 379       need_flush = true;
 380       flags &= ~NBDKIT_FLAG_FUA;
 381     }
 382   }
 383   else
 384     mark_dirty (h, true);
 385
 386   if (next->pwrite (next, buf, count, offs, flags, err) == -1)
 387     return -1;
 388   if (need_flush)
 389     return multi_conn_flush (next, h, 0, err);
 390   return 0;
 391 }
 392
 393 static int
 394 multi_conn_zero (nbdkit_next *next,
 395                  void *handle, uint32_t count, uint64_t offs, uint32_t flags,
 396                  int *err)
 397 {
 398   struct handle *h = handle;
 399   bool need_flush = false;
 400
 401   if (flags & NBDKIT_FLAG_FUA) {
 402     if (h->mode == EMULATE) {
 403       mark_dirty (h, true);
 404       need_flush = true;
 405       flags &= ~NBDKIT_FLAG_FUA;
 406     }
 407   }
 408   else
 409     mark_dirty (h, true);
 410
 411   if (next->zero (next, count, offs, flags, err) == -1)
 412     return -1;
 413   if (need_flush)
 414     return multi_conn_flush (next, h, 0, err);
 415   return 0;
 416 }
 417
 418 static int
 419 multi_conn_trim (nbdkit_next *next,
 420                  void *handle, uint32_t count, uint64_t offs, uint32_t flags,
 421                  int *err)
 422 {
 423   struct handle *h = handle;
 424   bool need_flush = false;
 425
 426   if (flags & NBDKIT_FLAG_FUA) {
 427     if (h->mode == EMULATE) {
 428       mark_dirty (h, true);
 429       need_flush = true;
 430       flags &= ~NBDKIT_FLAG_FUA;
 431     }
 432   }
 433   else
 434     mark_dirty (h, true);
 435
 436   if (next->trim (next, count, offs, flags, err) == -1)
 437     return -1;
 438   if (need_flush)
 439     return multi_conn_flush (next, h, 0, err);
 440   return 0;
 441 }
 442
 443 static int
 444 multi_conn_cache (nbdkit_next *next,
 445                   void *handle, uint32_t count, uint64_t offs, uint32_t flags,
 446                   int *err)
 447 {
 448   struct handle *h = handle;
 449
 450   mark_dirty (h, false);
 451   return next->cache (next, count, offs, flags, err);
 452 }
 453
 454 static int
 455 multi_conn_flush (nbdkit_next *next,
 456                   void *handle, uint32_t flags, int *err)
 457 {
 458   struct handle *h = handle, *h2;
 459   size_t i;
 460
 461   assert (h->group);
 462   if (h->mode == EMULATE) {
 463     ACQUIRE_LOCK_FOR_CURRENT_SCOPE (&lock);
 464     for (i = 0; i < h->group->conns.len; i++) {
 465       h2 = h->group->conns.ptr[i];
 466       if (track == OFF || (h->group->dirty &&
 467                            (track == FAST || h2->dirty & READ)) ||
 468           h2->dirty & WRITE) {
 469         if (h2->next->flush (h2->next, flags, err) == -1)
 470           return -1;
 471         h2->dirty = 0;
 472       }
 473     }
 474     h->group->dirty = 0;
 475   }
 476   else {
 477     /* !EMULATE: Check if the image is clean, allowing us to skip a flush. */
 478     if (track != OFF && !h->group->dirty)
 479       return 0;
 480     /* Perform the flush, then update dirty tracking. */
 481     if (next->flush (next, flags, err) == -1)
 482       return -1;
 483     switch (track) {
 484     case CONN:
 485       if (next->can_multi_conn (next) == 1) {
 486         ACQUIRE_LOCK_FOR_CURRENT_SCOPE (&lock);
 487         for (i = 0; i < h->group->conns.len; i++)
 488           h->group->conns.ptr[i]->dirty = 0;
 489         h->group->dirty = 0;
 490       }
 491       else
 492         h->dirty = 0;
 493       break;
 494     case FAST:
 495       h->group->dirty = false;
 496       break;
 497     case OFF:
 498       break;
 499     default:
 500       abort ();
 501     }
 502   }
 503   return 0;
 504 }
 505
 506 static struct nbdkit_filter filter = {
 507   .name              = "multi-conn",
 508   .longname          = "nbdkit multi-conn filter",
 509   .config            = multi_conn_config,
 510   .config_help       = multi_conn_config_help,
 511   .get_ready         = multi_conn_get_ready,
 512   .unload            = multi_conn_unload,
 513   .open              = multi_conn_open,
 514   .prepare           = multi_conn_prepare,
 515   .finalize          = multi_conn_finalize,
 516   .close             = multi_conn_close,
 517   .can_fua           = multi_conn_can_fua,
 518   .can_multi_conn    = multi_conn_can_multi_conn,
 519   .pread             = multi_conn_pread,
 520   .pwrite            = multi_conn_pwrite,
 521   .trim              = multi_conn_trim,
 522   .zero              = multi_conn_zero,
 523   .cache             = multi_conn_cache,
 524   .flush             = multi_conn_flush,
 525 };
 526
 527 NBDKIT_REGISTER_FILTER (filter)