src/backend/port/sysv_shmem.c

   1 /*-------------------------------------------------------------------------
   2  *
   3  * sysv_shmem.c
   4  *        Implement shared memory using SysV facilities
   5  *
   6  * These routines represent a fairly thin layer on top of SysV shared
   7  * memory functionality.
   8  *
   9  * Portions Copyright (c) 1996-2009, PostgreSQL Global Development Group
  10  * Portions Copyright (c) 1994, Regents of the University of California
  11  *
  12  * IDENTIFICATION
  13  *        $PostgreSQL$
  14  *
  15  *-------------------------------------------------------------------------
  16  */
  17 #include "postgres.h"
  18
  19 #include <signal.h>
  20 #include <unistd.h>
  21 #include <sys/file.h>
  22 #include <sys/stat.h>
  23 #ifdef HAVE_SYS_IPC_H
  24 #include <sys/ipc.h>
  25 #endif
  26 #ifdef HAVE_SYS_SHM_H
  27 #include <sys/shm.h>
  28 #endif
  29 #ifdef HAVE_KERNEL_OS_H
  30 #include <kernel/OS.h>
  31 #endif
  32
  33 #include "miscadmin.h"
  34 #include "storage/ipc.h"
  35 #include "storage/pg_shmem.h"
  36
  37
  38 typedef key_t IpcMemoryKey;             /* shared memory key passed to shmget(2) */
  39 typedef int IpcMemoryId;                /* shared memory ID returned by shmget(2) */
  40
  41 #define IPCProtection   (0600)  /* access/modify by user only */
  42
  43 #ifdef SHM_SHARE_MMU                    /* use intimate shared memory on Solaris */
  44 #define PG_SHMAT_FLAGS                  SHM_SHARE_MMU
  45 #else
  46 #define PG_SHMAT_FLAGS                  0
  47 #endif
  48
  49
  50 unsigned long UsedShmemSegID = 0;
  51 void       *UsedShmemSegAddr = NULL;
  52
  53 static void *InternalIpcMemoryCreate(IpcMemoryKey memKey, Size size);
  54 static void IpcMemoryDetach(int status, Datum shmaddr);
  55 static void IpcMemoryDelete(int status, Datum shmId);
  56 static PGShmemHeader *PGSharedMemoryAttach(IpcMemoryKey key,
  57                                          IpcMemoryId *shmid);
  58
  59
  60 /*
  61  *      InternalIpcMemoryCreate(memKey, size)
  62  *
  63  * Attempt to create a new shared memory segment with the specified key.
  64  * Will fail (return NULL) if such a segment already exists.  If successful,
  65  * attach the segment to the current process and return its attached address.
  66  * On success, callbacks are registered with on_shmem_exit to detach and
  67  * delete the segment when on_shmem_exit is called.
  68  *
  69  * If we fail with a failure code other than collision-with-existing-segment,
  70  * print out an error and abort.  Other types of errors are not recoverable.
  71  */
  72 static void *
  73 InternalIpcMemoryCreate(IpcMemoryKey memKey, Size size)
  74 {
  75         IpcMemoryId shmid;
  76         void       *memAddress;
  77
  78         shmid = shmget(memKey, size, IPC_CREAT | IPC_EXCL | IPCProtection);
  79
  80         if (shmid < 0)
  81         {
  82                 /*
  83                  * Fail quietly if error indicates a collision with existing segment.
  84                  * One would expect EEXIST, given that we said IPC_EXCL, but perhaps
  85                  * we could get a permission violation instead?  Also, EIDRM might
  86                  * occur if an old seg is slated for destruction but not gone yet.
  87                  */
  88                 if (errno == EEXIST || errno == EACCES
  89 #ifdef EIDRM
  90                         || errno == EIDRM
  91 #endif
  92                         )
  93                         return NULL;
  94
  95                 /*
  96                  * Else complain and abort
  97                  */
  98                 ereport(FATAL,
  99                                 (errmsg("could not create shared memory segment: %m"),
 100                   errdetail("Failed system call was shmget(key=%lu, size=%lu, 0%o).",
 101                                         (unsigned long) memKey, (unsigned long) size,
 102                                         IPC_CREAT | IPC_EXCL | IPCProtection),
 103                                  (errno == EINVAL) ?
 104                                  errhint("This error usually means that PostgreSQL's request for a shared memory "
 105                   "segment exceeded your kernel's SHMMAX parameter.  You can either "
 106                                                  "reduce the request size or reconfigure the kernel with larger SHMMAX.  "
 107                                   "To reduce the request size (currently %lu bytes), reduce "
 108                            "PostgreSQL's shared_buffers parameter (currently %d) and/or "
 109                                                  "its max_connections parameter (currently %d).\n"
 110                                                  "If the request size is already small, it's possible that it is less than "
 111                                                  "your kernel's SHMMIN parameter, in which case raising the request size or "
 112                                                  "reconfiguring SHMMIN is called for.\n"
 113                 "The PostgreSQL documentation contains more information about shared "
 114                                                  "memory configuration.",
 115                                                  (unsigned long) size, NBuffers, MaxBackends) : 0,
 116                                  (errno == ENOMEM) ?
 117                                  errhint("This error usually means that PostgreSQL's request for a shared "
 118                                    "memory segment exceeded available memory or swap space. "
 119                                   "To reduce the request size (currently %lu bytes), reduce "
 120                            "PostgreSQL's shared_buffers parameter (currently %d) and/or "
 121                                                  "its max_connections parameter (currently %d).\n"
 122                 "The PostgreSQL documentation contains more information about shared "
 123                                                  "memory configuration.",
 124                                                  (unsigned long) size, NBuffers, MaxBackends) : 0,
 125                                  (errno == ENOSPC) ?
 126                                  errhint("This error does *not* mean that you have run out of disk space. "
 127                                                  "It occurs either if all available shared memory IDs have been taken, "
 128                                                  "in which case you need to raise the SHMMNI parameter in your kernel, "
 129                   "or because the system's overall limit for shared memory has been "
 130                                  "reached.  If you cannot increase the shared memory limit, "
 131                   "reduce PostgreSQL's shared memory request (currently %lu bytes), "
 132                         "by reducing its shared_buffers parameter (currently %d) and/or "
 133                                                  "its max_connections parameter (currently %d).\n"
 134                 "The PostgreSQL documentation contains more information about shared "
 135                                                  "memory configuration.",
 136                                                  (unsigned long) size, NBuffers, MaxBackends) : 0));
 137         }
 138
 139         /* Register on-exit routine to delete the new segment */
 140         on_shmem_exit(IpcMemoryDelete, Int32GetDatum(shmid));
 141
 142         /* OK, should be able to attach to the segment */
 143         memAddress = shmat(shmid, NULL, PG_SHMAT_FLAGS);
 144
 145         if (memAddress == (void *) -1)
 146                 elog(FATAL, "shmat(id=%d) failed: %m", shmid);
 147
 148         /* Register on-exit routine to detach new segment before deleting */
 149         on_shmem_exit(IpcMemoryDetach, PointerGetDatum(memAddress));
 150
 151         /* Record key and ID in lockfile for data directory. */
 152         RecordSharedMemoryInLockFile((unsigned long) memKey,
 153                                                                  (unsigned long) shmid);
 154
 155         return memAddress;
 156 }
 157
 158 /****************************************************************************/
 159 /*      IpcMemoryDetach(status, shmaddr)        removes a shared memory segment         */
 160 /*                                                                              from process' address spaceq            */
 161 /*      (called as an on_shmem_exit callback, hence funny argument list)                */
 162 /****************************************************************************/
 163 static void
 164 IpcMemoryDetach(int status, Datum shmaddr)
 165 {
 166         if (shmdt(DatumGetPointer(shmaddr)) < 0)
 167                 elog(LOG, "shmdt(%p) failed: %m", DatumGetPointer(shmaddr));
 168 }
 169
 170 /****************************************************************************/
 171 /*      IpcMemoryDelete(status, shmId)          deletes a shared memory segment         */
 172 /*      (called as an on_shmem_exit callback, hence funny argument list)                */
 173 /****************************************************************************/
 174 static void
 175 IpcMemoryDelete(int status, Datum shmId)
 176 {
 177         if (shmctl(DatumGetInt32(shmId), IPC_RMID, NULL) < 0)
 178                 elog(LOG, "shmctl(%d, %d, 0) failed: %m",
 179                          DatumGetInt32(shmId), IPC_RMID);
 180 }
 181
 182 /*
 183  * PGSharedMemoryIsInUse
 184  *
 185  * Is a previously-existing shmem segment still existing and in use?
 186  *
 187  * The point of this exercise is to detect the case where a prior postmaster
 188  * crashed, but it left child backends that are still running.  Therefore
 189  * we only care about shmem segments that are associated with the intended
 190  * DataDir.  This is an important consideration since accidental matches of
 191  * shmem segment IDs are reasonably common.
 192  */
 193 bool
 194 PGSharedMemoryIsInUse(unsigned long id1, unsigned long id2)
 195 {
 196         IpcMemoryId shmId = (IpcMemoryId) id2;
 197         struct shmid_ds shmStat;
 198         struct stat statbuf;
 199         PGShmemHeader *hdr;
 200
 201         /*
 202          * We detect whether a shared memory segment is in use by seeing whether
 203          * it (a) exists and (b) has any processes attached to it.
 204          */
 205         if (shmctl(shmId, IPC_STAT, &shmStat) < 0)
 206         {
 207                 /*
 208                  * EINVAL actually has multiple possible causes documented in the
 209                  * shmctl man page, but we assume it must mean the segment no longer
 210                  * exists.
 211                  */
 212                 if (errno == EINVAL)
 213                         return false;
 214
 215                 /*
 216                  * EACCES implies that the segment belongs to some other userid, which
 217                  * means it is not a Postgres shmem segment (or at least, not one that
 218                  * is relevant to our data directory).
 219                  */
 220                 if (errno == EACCES)
 221                         return false;
 222
 223                 /*
 224                  * Some Linux kernel versions (in fact, all of them as of July 2007)
 225                  * sometimes return EIDRM when EINVAL is correct.  The Linux kernel
 226                  * actually does not have any internal state that would justify
 227                  * returning EIDRM, so we can get away with assuming that EIDRM is
 228                  * equivalent to EINVAL on that platform.
 229                  */
 230 #ifdef HAVE_LINUX_EIDRM_BUG
 231                 if (errno == EIDRM)
 232                         return false;
 233 #endif
 234
 235                 /*
 236                  * Otherwise, we had better assume that the segment is in use. The
 237                  * only likely case is EIDRM, which implies that the segment has been
 238                  * IPC_RMID'd but there are still processes attached to it.
 239                  */
 240                 return true;
 241         }
 242
 243         /* If it has no attached processes, it's not in use */
 244         if (shmStat.shm_nattch == 0)
 245                 return false;
 246
 247         /*
 248          * Try to attach to the segment and see if it matches our data directory.
 249          * This avoids shmid-conflict problems on machines that are running
 250          * several postmasters under the same userid.
 251          */
 252         if (stat(DataDir, &statbuf) < 0)
 253                 return true;                    /* if can't stat, be conservative */
 254
 255         hdr = (PGShmemHeader *) shmat(shmId, NULL, PG_SHMAT_FLAGS);
 256
 257         if (hdr == (PGShmemHeader *) -1)
 258                 return true;                    /* if can't attach, be conservative */
 259
 260         if (hdr->magic != PGShmemMagic ||
 261                 hdr->device != statbuf.st_dev ||
 262                 hdr->inode != statbuf.st_ino)
 263         {
 264                 /*
 265                  * It's either not a Postgres segment, or not one for my data
 266                  * directory.  In either case it poses no threat.
 267                  */
 268                 shmdt((void *) hdr);
 269                 return false;
 270         }
 271
 272         /* Trouble --- looks a lot like there's still live backends */
 273         shmdt((void *) hdr);
 274
 275         return true;
 276 }
 277
 278
 279 /*
 280  * PGSharedMemoryCreate
 281  *
 282  * Create a shared memory segment of the given size and initialize its
 283  * standard header.  Also, register an on_shmem_exit callback to release
 284  * the storage.
 285  *
 286  * Dead Postgres segments are recycled if found, but we do not fail upon
 287  * collision with non-Postgres shmem segments.  The idea here is to detect and
 288  * re-use keys that may have been assigned by a crashed postmaster or backend.
 289  *
 290  * makePrivate means to always create a new segment, rather than attach to
 291  * or recycle any existing segment.
 292  *
 293  * The port number is passed for possible use as a key (for SysV, we use
 294  * it to generate the starting shmem key).      In a standalone backend,
 295  * zero will be passed.
 296  */
 297 PGShmemHeader *
 298 PGSharedMemoryCreate(Size size, bool makePrivate, int port)
 299 {
 300         IpcMemoryKey NextShmemSegID;
 301         void       *memAddress;
 302         PGShmemHeader *hdr;
 303         IpcMemoryId shmid;
 304         struct stat statbuf;
 305
 306         /* Room for a header? */
 307         Assert(size > MAXALIGN(sizeof(PGShmemHeader)));
 308
 309         /* Make sure PGSharedMemoryAttach doesn't fail without need */
 310         UsedShmemSegAddr = NULL;
 311
 312         /* Loop till we find a free IPC key */
 313         NextShmemSegID = port * 1000;
 314
 315         for (NextShmemSegID++;; NextShmemSegID++)
 316         {
 317                 /* Try to create new segment */
 318                 memAddress = InternalIpcMemoryCreate(NextShmemSegID, size);
 319                 if (memAddress)
 320                         break;                          /* successful create and attach */
 321
 322                 /* Check shared memory and possibly remove and recreate */
 323
 324                 if (makePrivate)                /* a standalone backend shouldn't do this */
 325                         continue;
 326
 327                 if ((memAddress = PGSharedMemoryAttach(NextShmemSegID, &shmid)) == NULL)
 328                         continue;                       /* can't attach, not one of mine */
 329
 330                 /*
 331                  * If I am not the creator and it belongs to an extant process,
 332                  * continue.
 333                  */
 334                 hdr = (PGShmemHeader *) memAddress;
 335                 if (hdr->creatorPID != getpid())
 336                 {
 337                         if (kill(hdr->creatorPID, 0) == 0 || errno != ESRCH)
 338                         {
 339                                 shmdt(memAddress);
 340                                 continue;               /* segment belongs to a live process */
 341                         }
 342                 }
 343
 344                 /*
 345                  * The segment appears to be from a dead Postgres process, or from a
 346                  * previous cycle of life in this same process.  Zap it, if possible.
 347                  * This probably shouldn't fail, but if it does, assume the segment
 348                  * belongs to someone else after all, and continue quietly.
 349                  */
 350                 shmdt(memAddress);
 351                 if (shmctl(shmid, IPC_RMID, NULL) < 0)
 352                         continue;
 353
 354                 /*
 355                  * Now try again to create the segment.
 356                  */
 357                 memAddress = InternalIpcMemoryCreate(NextShmemSegID, size);
 358                 if (memAddress)
 359                         break;                          /* successful create and attach */
 360
 361                 /*
 362                  * Can only get here if some other process managed to create the same
 363                  * shmem key before we did.  Let him have that one, loop around to try
 364                  * next key.
 365                  */
 366         }
 367
 368         /*
 369          * OK, we created a new segment.  Mark it as created by this process. The
 370          * order of assignments here is critical so that another Postgres process
 371          * can't see the header as valid but belonging to an invalid PID!
 372          */
 373         hdr = (PGShmemHeader *) memAddress;
 374         hdr->creatorPID = getpid();
 375         hdr->magic = PGShmemMagic;
 376
 377         /* Fill in the data directory ID info, too */
 378         if (stat(DataDir, &statbuf) < 0)
 379                 ereport(FATAL,
 380                                 (errcode_for_file_access(),
 381                                  errmsg("could not stat data directory \"%s\": %m",
 382                                                 DataDir)));
 383         hdr->device = statbuf.st_dev;
 384         hdr->inode = statbuf.st_ino;
 385
 386         /*
 387          * Initialize space allocation status for segment.
 388          */
 389         hdr->totalsize = size;
 390         hdr->freeoffset = MAXALIGN(sizeof(PGShmemHeader));
 391
 392         /* Save info for possible future use */
 393         UsedShmemSegAddr = memAddress;
 394         UsedShmemSegID = (unsigned long) NextShmemSegID;
 395
 396         return hdr;
 397 }
 398
 399 #ifdef EXEC_BACKEND
 400
 401 /*
 402  * PGSharedMemoryReAttach
 403  *
 404  * Re-attach to an already existing shared memory segment.      In the non
 405  * EXEC_BACKEND case this is not used, because postmaster children inherit
 406  * the shared memory segment attachment via fork().
 407  *
 408  * UsedShmemSegID and UsedShmemSegAddr are implicit parameters to this
 409  * routine.  The caller must have already restored them to the postmaster's
 410  * values.
 411  */
 412 void
 413 PGSharedMemoryReAttach(void)
 414 {
 415         IpcMemoryId shmid;
 416         void       *hdr;
 417         void       *origUsedShmemSegAddr = UsedShmemSegAddr;
 418
 419         Assert(UsedShmemSegAddr != NULL);
 420         Assert(IsUnderPostmaster);
 421
 422 #ifdef __CYGWIN__
 423         /* cygipc (currently) appears to not detach on exec. */
 424         PGSharedMemoryDetach();
 425         UsedShmemSegAddr = origUsedShmemSegAddr;
 426 #endif
 427
 428         elog(DEBUG3, "attaching to %p", UsedShmemSegAddr);
 429         hdr = (void *) PGSharedMemoryAttach((IpcMemoryKey) UsedShmemSegID, &shmid);
 430         if (hdr == NULL)
 431                 elog(FATAL, "could not reattach to shared memory (key=%d, addr=%p): %m",
 432                          (int) UsedShmemSegID, UsedShmemSegAddr);
 433         if (hdr != origUsedShmemSegAddr)
 434                 elog(FATAL, "reattaching to shared memory returned unexpected address (got %p, expected %p)",
 435                          hdr, origUsedShmemSegAddr);
 436
 437         UsedShmemSegAddr = hdr;         /* probably redundant */
 438 }
 439 #endif   /* EXEC_BACKEND */
 440
 441 /*
 442  * PGSharedMemoryDetach
 443  *
 444  * Detach from the shared memory segment, if still attached.  This is not
 445  * intended for use by the process that originally created the segment
 446  * (it will have an on_shmem_exit callback registered to do that).      Rather,
 447  * this is for subprocesses that have inherited an attachment and want to
 448  * get rid of it.
 449  */
 450 void
 451 PGSharedMemoryDetach(void)
 452 {
 453         if (UsedShmemSegAddr != NULL)
 454         {
 455                 if ((shmdt(UsedShmemSegAddr) < 0)
 456 #if defined(EXEC_BACKEND) && defined(__CYGWIN__)
 457                 /* Work-around for cygipc exec bug */
 458                         && shmdt(NULL) < 0
 459 #endif
 460                         )
 461                         elog(LOG, "shmdt(%p) failed: %m", UsedShmemSegAddr);
 462                 UsedShmemSegAddr = NULL;
 463         }
 464 }
 465
 466
 467 /*
 468  * Attach to shared memory and make sure it has a Postgres header
 469  *
 470  * Returns attach address if OK, else NULL
 471  */
 472 static PGShmemHeader *
 473 PGSharedMemoryAttach(IpcMemoryKey key, IpcMemoryId *shmid)
 474 {
 475         PGShmemHeader *hdr;
 476
 477         if ((*shmid = shmget(key, sizeof(PGShmemHeader), 0)) < 0)
 478                 return NULL;
 479
 480         hdr = (PGShmemHeader *) shmat(*shmid, UsedShmemSegAddr, PG_SHMAT_FLAGS);
 481
 482         if (hdr == (PGShmemHeader *) -1)
 483                 return NULL;                    /* failed: must be some other app's */
 484
 485         if (hdr->magic != PGShmemMagic)
 486         {
 487                 shmdt((void *) hdr);
 488                 return NULL;                    /* segment belongs to a non-Postgres app */
 489         }
 490
 491         return hdr;
 492 }