src/volser/vol_split.c

   1 /*
   2  * Copyright (c) 2007, Hartmut Reuter,
   3  * RZG, Max-Planck-Institut f. Plasmaphysik.
   4  * All Rights Reserved.
   5  *
   6  */
   7
   8 #include <afsconfig.h>
   9 #include <afs/param.h>
  10
  11 #include <roken.h>
  12
  13 #ifdef HAVE_SYS_FILE_H
  14 #include <sys/file.h>
  15 #endif
  16
  17 #if defined(AFS_NAMEI_ENV) && !defined(AFS_NT40_ENV)
  18 #include <afs/dir.h>
  19 #include <rx/xdr.h>
  20 #include <rx/rx_queue.h>
  21 #include <afs/afsint.h>
  22 #include <afs/nfs.h>
  23 #include <lwp.h>
  24 #include <lock.h>
  25 #include <afs/afssyscalls.h>
  26 #include <afs/ihandle.h>
  27 #include <afs/vnode.h>
  28 #include <afs/volume.h>
  29 #include <afs/partition.h>
  30 #include <afs/viceinode.h>
  31
  32 #include "vol.h"
  33 #include "volint.h"
  34 #include "volser.h"
  35 #include "physio.h"
  36 #include "volser_internal.h"
  37 #ifdef AFS_RXOSD_SUPPORT
  38 #include "rxosd.h"
  39 #include "vol_osd.h"
  40 #include "../vol/vol_osd_prototypes.h"
  41 #endif
  42
  43 #define NEEDED  1
  44 #define PARENT  2
  45 #define CHANGEPARENT 4
  46
  47 #define NAMEI_VNODEMASK    0x03ffffff
  48 #define NAMEI_TAGMASK      0x7
  49 #define NAMEI_TAGSHIFT     26
  50 #define NAMEI_UNIQMASK     0xffffffff
  51 #define NAMEI_UNIQSHIFT    32
  52
  53 struct VnodeExtract {
  54     afs_uint32 vN;
  55     afs_uint32 parent;
  56     afs_uint32 flag;
  57 };
  58
  59 struct Msg {
  60     struct rx_call * call;
  61     int verbose;
  62     char line[1024];
  63 };
  64
  65 static afs_int32
  66 ExtractVnodes(struct Msg *m, Volume *vol, afs_int32 class,
  67               struct VnodeExtract **list,
  68               afs_uint32 *length, afs_uint32 where,
  69               struct VnodeDiskObject *vd,
  70               afs_uint32 *parent, struct VnodeDiskObject *parentvd)
  71 {
  72     afs_int32 code = 0;
  73     char buf[SIZEOF_LARGEDISKVNODE];
  74     struct VnodeDiskObject *vnode = (struct VnodeDiskObject *)&buf;
  75     FdHandle_t *fdP = 0;
  76     StreamHandle_t *stream = 0;
  77     struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
  78     struct VnodeExtract *e;
  79     afs_sfsize_t size;
  80     afs_foff_t offset;
  81
  82     *length = 0;
  83     if (parent)
  84         *parent = 0;
  85
  86     fdP = IH_OPEN(vol->vnodeIndex[class].handle);
  87     if (!fdP) {
  88         sprintf(m->line, "Couldn't open %s Index of volume %" AFS_VOLID_FMT "\n",
  89                 class ? "small":"large", afs_printable_VolumeId_lu(V_id(vol)));
  90         rx_Write(m->call, m->line, strlen(m->line));
  91         code = EIO;
  92         goto Bad_Extract;
  93     }
  94     size = FDH_SIZE(fdP);
  95     *list = calloc(size / vcp->diskSize, sizeof(struct VnodeExtract));
  96     if (!(*list)) {
  97         code = ENOMEM;
  98         goto Bad_Extract;
  99     }
 100     stream = FDH_FDOPEN(fdP, "r");
 101     if (!stream) {
 102         sprintf(m->line, "Couldn't stream open %s Index of volume %" AFS_VOLID_FMT "\n",
 103                 class ? "small":"large", afs_printable_VolumeId_lu(V_id(vol)));
 104         rx_Write(m->call, m->line, strlen(m->line));
 105         code = EIO;
 106         goto Bad_Extract;
 107     }
 108     code = STREAM_ASEEK(stream, vcp->diskSize);
 109     if (code)
 110         goto Bad_Extract;
 111
 112     offset = vcp->diskSize;
 113     e = *list;
 114     while (!STREAM_EOF(stream)) {
 115         afs_int32 vN = (offset >> (vcp->logSize -1)) - 1 + class;
 116         if (STREAM_READ(vnode, vcp->diskSize, 1, stream) == 1) {
 117             if (vnode->type != vNull) {
 118                 e->vN = vN;
 119                 e->parent = vnode->parent;
 120                 if (vN == where && class == vLarge) {
 121                     memcpy(vd, vnode, vcp->diskSize);
 122                     *parent = vnode->parent;
 123                 }
 124                 e++;
 125             }
 126             offset += vcp->diskSize;
 127         }
 128     }
 129     *length = (e - *list);
 130     if (class == vLarge) {
 131         if (*parent) {
 132             offset = (*parent + 1 - class) << (vcp->logSize -1);
 133             code = STREAM_ASEEK(stream, offset);
 134             if (STREAM_READ(vnode, vcp->diskSize, 1, stream) == 1)
 135                 memcpy(parentvd, vnode, vcp->diskSize);
 136             else
 137                 code = EIO;
 138         } else {
 139             sprintf(m->line, "SplitVolume: extract didn't see directory %u\n", where);
 140             rx_Write(m->call, m->line, strlen(m->line));
 141             code = ENOENT;
 142         }
 143     }
 144     if (m->verbose) {
 145         sprintf(m->line, "Volume %" AFS_VOLID_FMT " has %u %s vnodes in volume %" AFS_VOLID_FMT "\n",
 146                 afs_printable_VolumeId_lu(V_parentId(vol)), *length,
 147                 class? "small":"large", afs_printable_VolumeId_lu(V_id(vol)));
 148         rx_Write(m->call, m->line, strlen(m->line));
 149     }
 150
 151 Bad_Extract:
 152     if (stream)
 153         STREAM_CLOSE(stream);
 154     if (fdP)
 155         FDH_CLOSE(fdP);
 156     if (code) {
 157         free(*list);
 158         *list = 0;
 159     }
 160     return code;
 161 }
 162
 163 static afs_int32
 164 FindVnodes(struct Msg *m, afs_uint32 where,
 165            struct VnodeExtract *list, afs_int32 length,
 166            struct VnodeExtract *dlist, afs_int32 dlength,
 167            afs_uint32 *needed, afs_int32 class)
 168 {
 169     afs_int32 i, j, found = 0;
 170     afs_int32 parent = 0;
 171
 172     *needed = 0;
 173     for (i=0; i<length; i++) {
 174         if (list[i].vN == where) {        /* dir to be replaced by mount point */
 175             list[i].flag |= NEEDED;
 176             parent = list[i].parent;
 177             found = 1;
 178             (*needed)++;
 179         }
 180         if (list[i].parent == where) {          /* all 1st generation children */
 181             list[i].flag |= (NEEDED + CHANGEPARENT);
 182             (*needed)++;
 183         }
 184     }
 185     if (list[0].vN & 1) {               /* only for directories */
 186         if (!found) {
 187             sprintf(m->line,
 188                 "SplitVolume: directory %u where to start new volume not found\n",
 189                  where);
 190             rx_Write(m->call, m->line, strlen(m->line));
 191             return ENOENT;
 192         }
 193         found = 0;
 194         for (i=0; i<length; i++) {
 195             if (list[i].vN == parent) { /* dir where to create mount point */
 196                 list[i].flag |= PARENT;
 197                 found = 1;
 198                 break;
 199             }
 200         }
 201         if (!found) {
 202             sprintf(m->line, "SplitVolume: parent directory %u not found\n",
 203                         parent);
 204             rx_Write(m->call, m->line, strlen(m->line));
 205             return ENOENT;
 206         }
 207     }
 208     found = 1;
 209     while (found) {
 210         found = 0;
 211         for (i=0; i<dlength; i++) {
 212             if (!(dlist[i].flag & NEEDED)) /* dirs to remain in old volume */
 213                 continue;
 214             for (j=0; j<length; j++) {
 215                 if (list[j].parent == dlist[i].vN && !(list[j].flag & NEEDED)) {
 216                     list[j].flag |= NEEDED;
 217                     (*needed)++;
 218                     found = 1;
 219                 }
 220             }
 221         }
 222     }
 223     if (m->verbose) {
 224         sprintf(m->line, "%u %s vnodes will go into the new volume\n",
 225                         *needed, class ? "small" : "large");
 226         rx_Write(m->call, m->line, strlen(m->line));
 227     }
 228     return 0;
 229 }
 230
 231 static afs_int32
 232 copyDir(struct Msg *m, IHandle_t *inh, IHandle_t *outh)
 233 {
 234     FdHandle_t *infdP, *outfdP;
 235     char *tbuf;
 236     afs_sfsize_t size;
 237     afs_foff_t offset;
 238
 239     infdP = IH_OPEN(inh);
 240     if (!infdP) {
 241         sprintf(m->line, "Couldn't open input directory %" AFS_VOLID_FMT ".%u.%u\n",
 242                 afs_printable_VolumeId_lu(inh->ih_vid),
 243                 (afs_uint32)(inh->ih_ino & NAMEI_VNODEMASK),
 244                 (afs_uint32)(inh->ih_ino >> NAMEI_UNIQSHIFT));
 245         rx_Write(m->call, m->line, strlen(m->line));
 246         return EIO;
 247     }
 248     outfdP = IH_OPEN(outh);
 249     /*
 250      * In case that a file with the same (NAMEI) name existed before and is still
 251      * open outfdP may point to the wrong (unlinked) file. To make sure we write
 252      * into the correct file it's safer to 1st FDH_REALLYCLOSE it and then to
 253      * re-open it.
 254      */
 255     if (outfdP)
 256         FDH_REALLYCLOSE(outfdP);
 257     outfdP = IH_OPEN(outh);
 258     if (!outfdP) {
 259         sprintf(m->line, "Couldn't open output directory %" AFS_VOLID_FMT ".%u.%u\n",
 260                 afs_printable_VolumeId_lu(outh->ih_vid),
 261                 (afs_uint32)(outh->ih_ino & NAMEI_VNODEMASK),
 262                 (afs_uint32)(outh->ih_ino >> NAMEI_UNIQSHIFT));
 263         rx_Write(m->call, m->line, strlen(m->line));
 264         FDH_REALLYCLOSE(infdP);
 265         return EIO;
 266     }
 267     tbuf = malloc(2048);
 268     offset = 0;
 269     size = FDH_SIZE(infdP);
 270     while (size) {
 271         size_t tlen;
 272         tlen = size > 2048 ? 2048 : size;
 273         if (FDH_PREAD(infdP, tbuf, tlen, offset) != tlen) {
 274             sprintf(m->line, "Couldn't read directory %" AFS_VOLID_FMT ".%u.%u\n",
 275                     afs_printable_VolumeId_lu(infdP->fd_ih->ih_vid),
 276                     (afs_uint32)(infdP->fd_ih->ih_ino & NAMEI_VNODEMASK),
 277                     (afs_uint32)(infdP->fd_ih->ih_ino >> NAMEI_UNIQSHIFT));
 278             rx_Write(m->call, m->line, strlen(m->line));
 279             FDH_REALLYCLOSE(infdP);
 280             FDH_REALLYCLOSE(outfdP);
 281             free(tbuf);
 282             return EIO;
 283         }
 284         if (FDH_PWRITE(outfdP, tbuf, tlen, offset) != tlen) {
 285             sprintf(m->line, "Couldn't write directory %" AFS_VOLID_FMT ".%u.%u\n",
 286                     afs_printable_VolumeId_lu(outfdP->fd_ih->ih_vid),
 287                     (afs_uint32)(outfdP->fd_ih->ih_ino & NAMEI_VNODEMASK),
 288                     (afs_uint32)(outfdP->fd_ih->ih_ino >> NAMEI_UNIQSHIFT));
 289             rx_Write(m->call, m->line, strlen(m->line));
 290             FDH_REALLYCLOSE(infdP);
 291             FDH_REALLYCLOSE(outfdP);
 292             free(tbuf);
 293             return EIO;
 294         }
 295         size -= tlen;
 296         offset += tlen;
 297     }
 298     free(tbuf);
 299     FDH_CLOSE(outfdP);
 300     FDH_REALLYCLOSE(infdP);
 301     return 0;
 302 }
 303
 304 afs_int32 copyVnodes(struct Msg *m, Volume *vol, Volume *newvol,
 305                         afs_int32 class,
 306                         struct VnodeExtract *list, afs_int32 length,
 307                         afs_int32 where, afs_uint64 *blocks,
 308                         struct VnodeDiskObject *parVnode)
 309 {
 310     afs_int32 i, code = 0;
 311     char buf[SIZEOF_LARGEDISKVNODE];
 312     struct VnodeDiskObject *vnode = (struct VnodeDiskObject *)&buf;
 313     FdHandle_t *fdP = 0;
 314     FdHandle_t *newfdP = 0;
 315     struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
 316     struct VnodeExtract *e;
 317     afs_uint64 offset;
 318     Inode ino, newino;
 319
 320     fdP = IH_OPEN(vol->vnodeIndex[class].handle);
 321     if (!fdP) {
 322         Log("Couldn't open %s Index of volume %" AFS_VOLID_FMT "\n",
 323             class ? "small":"large", afs_printable_VolumeId_lu(V_id(vol)));
 324         code = EIO;
 325         goto Bad_Copy;
 326     }
 327     newfdP = IH_OPEN(newvol->vnodeIndex[class].handle);
 328     if (!newfdP) {
 329         Log("Couldn't open %s Index of volume %" AFS_VOLID_FMT "\n",
 330             class ? "small":"large", afs_printable_VolumeId_lu(V_id(newvol)));
 331         code = EIO;
 332         goto Bad_Copy;
 333     }
 334
 335     for (i=0; i<length; i++) {
 336         e = &list[i];
 337         if (e->flag) {
 338             afs_uint64 size;
 339             offset = (e->vN + 1 - class) << (vcp->logSize -1);
 340             if (FDH_PREAD(fdP, vnode, vcp->diskSize, offset) != vcp->diskSize) {
 341                 Log("Couldn't read in %s Index of volume %" AFS_VOLID_FMT " at offset %llu\n",
 342                     class ? "small":"large",
 343                     afs_printable_VolumeId_lu(V_id(vol)), offset);
 344                 code = EIO;
 345                 goto Bad_Copy;
 346             }
 347             if (e->flag & PARENT) {
 348                 /*
 349                  *   do a preventive copy on write for later update
 350                  */
 351                 IHandle_t *newh = 0;
 352                 IHandle_t *h = 0;
 353 #if defined(NEARINODE_HINT) && !defined(AFS_NAMEI_ENV)
 354                 Inode nearInode;
 355                 V_pref(vol,nearInode)
 356 #endif
 357
 358                 newino = IH_CREATE(V_linkHandle(vol), V_device(vol),
 359                                 VPartitionPath(V_partition(vol)),
 360                                 nearInode, V_parentId(vol),
 361                                 e->vN, vnode->uniquifier,
 362                                 vnode->dataVersion);
 363                 IH_INIT(newh, V_device(vol), V_parentId(vol), newino);
 364                 ino = VNDISK_GET_INO(vnode);
 365                 IH_INIT(h, V_device(vol), V_parentId(vol), ino);
 366                 code = copyDir(m, h, newh);
 367                 if (code)
 368                     goto Bad_Copy;
 369                 /* Now update the vnode and write it back to disk */
 370                 VNDISK_SET_INO(vnode, newino);
 371                 vnode->cloned = 0;
 372                 if (FDH_PWRITE(fdP, vnode, vcp->diskSize, offset) != vcp->diskSize) {
 373                     Log("Couldn't write in %s Index of volume %" AFS_VOLID_FMT " at offset %llu\n",
 374                         class ? "small":"large", afs_printable_VolumeId_lu(V_id(vol)), offset);
 375                     code = EIO;
 376                     goto Bad_Copy;
 377                 }
 378                 if (parVnode != NULL)
 379                     memcpy(parVnode, vnode, sizeof(struct VnodeDiskObject));
 380             }
 381             if (e->flag & NEEDED && e->vN != where) {
 382                 VNDISK_GET_LEN(size, vnode);
 383                 *blocks += (size + 0x3ff) >> 10;
 384                 ino = VNDISK_GET_INO(vnode);
 385                 if (ino) {
 386                     IHandle_t *h, *newh;
 387                     Inode AFS_UNUSED nearInode;
 388 #if defined(NEARINODE_HINT) && !defined(AFS_NAMEI_ENV)
 389                     V_pref(vol,nearInode)
 390 #endif
 391                     IH_INIT(h, vol->device, V_parentId(vol), ino);
 392                     if (e->parent == where)
 393                         vnode->parent = 1;
 394                     newino = IH_CREATE(V_linkHandle(newvol), V_device(newvol),
 395                                 VPartitionPath(V_partition(newvol)),
 396                                 nearInode, V_parentId(newvol),
 397                                 e->vN, vnode->uniquifier,
 398                                 vnode->dataVersion);
 399                     if (!VALID_INO(newino)) {
 400                         Log("IH_CREATE failed for %" AFS_VOLID_FMT ".%u.%u\n",
 401                             afs_printable_VolumeId_lu(V_id(newvol)), e->vN, vnode->uniquifier);
 402                         code = EIO;
 403                         goto Bad_Copy;
 404                     }
 405                     nearInode = newino;
 406                     IH_INIT(newh, newvol->device, V_parentId(newvol), newino);
 407                     code = namei_replace_file_by_hardlink(newh, h);
 408                     VNDISK_SET_INO(vnode, newino);
 409 #ifdef AFS_RXOSD_SUPPORT
 410                 } else {
 411                     code = osd_split_objects(vol, newvol, vnode, e->vN);
 412 #endif /*  AFS_RXOSD_SUPPORT */
 413                 }
 414                 if (code)
 415                     goto Bad_Copy;
 416                 if (e->flag & CHANGEPARENT)
 417                     vnode->parent = 1; /* in new root-directory */
 418                 vnode->cloned = 0;
 419                 if (FDH_PWRITE(newfdP, vnode, vcp->diskSize, offset) != vcp->diskSize) {
 420                     Log("Couldn't write in %s Index of volume %" AFS_VOLID_FMT " to offset %llu\n",
 421                         class ? "small":"large", afs_printable_VolumeId_lu(V_id(newvol)), offset);
 422                     code = EIO;
 423                     goto Bad_Copy;
 424                 }
 425             }
 426         }
 427     }
 428     /*
 429      *  Now copy the root directory from old to new volume
 430      */
 431     if (class == vLarge) {
 432         IHandle_t *h, *newh;
 433         char buf2[SIZEOF_LARGEDISKVNODE];
 434         struct VnodeDiskObject *vnode2 = (struct VnodeDiskObject *)&buf2;
 435         afs_uint64 newoffset, size;
 436
 437         newoffset = vcp->diskSize;
 438         if (FDH_PREAD(newfdP, vnode2, vcp->diskSize, newoffset) != vcp->diskSize) {
 439             Log("splitvolume: couldn't read in large Index of new volume %" AFS_VOLID_FMT " at offset %u\n",
 440                 afs_printable_VolumeId_lu(V_id(newvol)), vcp->diskSize);
 441             code = EIO;
 442             goto Bad_Copy;
 443         }
 444         offset = (where + 1 - class) << (vcp->logSize -1);
 445         if (FDH_PREAD(fdP, vnode, vcp->diskSize, offset) != vcp->diskSize) {
 446             Log("Couldn't read in large Index of old volume %" AFS_VOLID_FMT " at offset %llu\n",
 447                 afs_printable_VolumeId_lu(V_id(vol)), offset);
 448             code = EIO;
 449             goto Bad_Copy;
 450         }
 451         VNDISK_GET_LEN(size, vnode);
 452         *blocks += (size + 0x3ff) >> 10;
 453         ino = VNDISK_GET_INO(vnode);
 454         IH_INIT(h, vol->device, V_parentId(vol), ino);
 455         newino = VNDISK_GET_INO(vnode2);
 456         IH_INIT(newh, newvol->device, V_parentId(newvol), newino);
 457         code = copyDir(m, h, newh);
 458         if (code) {
 459             Log("splitvolume: copyDir failed for new root from "
 460                 "%" AFS_VOLID_FMT ".%u.%u to %" AFS_VOLID_FMT ".1.1\n",
 461                 afs_printable_VolumeId_lu(V_id(vol)), where, vnode->uniquifier,
 462                 afs_printable_VolumeId_lu(V_id(newvol)));
 463             code = EIO;
 464             goto Bad_Copy;
 465         }
 466         VNDISK_SET_INO(vnode, newino);
 467         vnode->uniquifier = 1;
 468         vnode->cloned = 0;
 469         vnode->parent = vnode2->parent;
 470         vnode->serverModifyTime = vnode2->serverModifyTime;
 471         if (FDH_PWRITE(newfdP, vnode, vcp->diskSize, newoffset) != vcp->diskSize) {
 472             Log("splitvolume: couldn't write in large Index of %" AFS_VOLID_FMT " at offset %u\n",
 473                 afs_printable_VolumeId_lu(V_id(newvol)), vcp->diskSize);
 474             code = EIO;
 475         }
 476     }
 477 Bad_Copy:
 478     if (fdP)
 479         FDH_CLOSE(fdP);
 480     if (newfdP)
 481         FDH_CLOSE(newfdP);
 482     return code;
 483 }
 484
 485 static afs_int32
 486 findName(Volume *vol, struct VnodeDiskObject *vd, afs_uint32 vN,
 487          afs_uint32 un, char *name,afs_int32 length)
 488 {
 489     afs_int32 code;
 490     Inode ino;
 491     DirHandle dir;
 492
 493     ino = VNDISK_GET_INO(vd);
 494     SetSalvageDirHandle(&dir, V_id(vol), V_device(vol), ino);
 495
 496     code = afs_dir_InverseLookup(&dir, vN, un, name, length);
 497     FidZap(&dir);
 498     return code;
 499 }
 500
 501 static afs_int32
 502 createMountpoint(Volume *vol, Volume *newvol, struct VnodeDiskObject *parent,
 503                 afs_uint32 vN,  struct VnodeDiskObject *vd, char *name)
 504 {
 505     afs_int32 code;
 506     Inode ino, newino;
 507     DirHandle dir;
 508     IHandle_t *h, *hp;
 509     struct VnodeDiskObject vnode;
 510     FdHandle_t *fdP, *fdP2;
 511     afs_uint64 size;
 512     afs_foff_t offset;
 513     afs_int32 class = vSmall;
 514     struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
 515 #if defined(NEARINODE_HINT) && !defined(AFS_NAMEI_ENV)
 516     Inode nearInode = 0;
 517 #endif
 518     AFSFid fid;
 519     struct timeval now;
 520     afs_uint32 newvN;
 521     char symlink[VNAMESIZE + 1];
 522     ssize_t rc;
 523
 524     FT_GetTimeOfDay(&now, 0);
 525     fdP = IH_OPEN(vol->vnodeIndex[vSmall].handle);
 526     if (!fdP) {
 527         Log("split volume: error opening small vnode index of %" AFS_VOLID_FMT "\n", afs_printable_VolumeId_lu(V_id(vol)));
 528         return EIO;
 529     }
 530     offset = vcp->diskSize;
 531     while (1) {
 532         rc = FDH_PREAD(fdP, &vnode, vcp->diskSize, offset);
 533         if (rc != vcp->diskSize) {
 534             if (rc < 0) {
 535                 Log("split volume: error reading small vnode index of %" AFS_VOLID_FMT "\n", afs_printable_VolumeId_lu(V_id(vol)));
 536                 return EIO;
 537              }
 538              if (rc == 0)
 539                  break;
 540              if (rc < vcp->diskSize)
 541                  break;
 542         }
 543         if (vnode.type == vNull)
 544             break;
 545         offset += vcp->diskSize;
 546     }
 547     memset(&vnode, 0, sizeof(vnode));
 548     vnode.type = vSymlink;
 549     V_nextVnodeUnique(vol)++;
 550     vnode.uniquifier = V_nextVnodeUnique(vol);
 551     vnode.author = vd->author;
 552     vnode.owner = vd->owner;
 553     vnode.group = vd->group;
 554     vnode.modeBits = 0644;
 555     vnode.unixModifyTime = now.tv_sec;
 556     vnode.serverModifyTime = now.tv_sec;
 557     vnode.dataVersion = 1;
 558     vnode.linkCount = 1;
 559     vnode.parent = vN;
 560
 561     newvN = (offset >> (VnodeClassInfo[class].logSize - 1)) - 1 + class;
 562 #if defined(NEARINODE_HINT) && !defined(AFS_NAMEI_ENV)
 563     V_pref(vol,nearInode)
 564 #endif
 565     newino = IH_CREATE(V_linkHandle(vol), V_device(vol),
 566                 VPartitionPath(V_partition(vol)), nearInode,
 567                 V_parentId(vol), newvN, vnode.uniquifier, 1);
 568
 569     IH_INIT(h, V_device(vol), V_parentId(vol), newino);
 570     fdP2 = IH_OPEN(h);
 571     if (!fdP2) {
 572         Log("split volume: couldn't open inode for mountpoint %" AFS_VOLID_FMT ".%u.%u\n",
 573             afs_printable_VolumeId_lu(V_id(vol)), newvN, vnode.uniquifier);
 574         return EIO;
 575     }
 576     sprintf(symlink, "#%s", V_name(newvol));
 577     size = strlen(symlink) + 1;
 578     if (FDH_PWRITE(fdP2, symlink, size, 0) != size) {
 579         Log("split volume: couldn't write mountpoint %" AFS_VOLID_FMT ".%u.%u\n",
 580             afs_printable_VolumeId_lu(V_id(vol)), newvN, vnode.uniquifier);
 581         return EIO;
 582     }
 583     FDH_REALLYCLOSE(fdP2);
 584     IH_RELEASE(h);
 585     VNDISK_SET_INO(&vnode, newino);
 586     VNDISK_SET_LEN(&vnode, size);
 587 #ifndef AFS_RXOSD_SUPPORT
 588     vnode.vnodeMagic = SMALLVNODEMAGIC;
 589 #endif
 590     if (FDH_PWRITE(fdP, &vnode, vcp->diskSize, offset) != vcp->diskSize) {
 591         Log("split volume: couldn't write vnode for mountpoint %" AFS_VOLID_FMT ".%u.%u\n",
 592             afs_printable_VolumeId_lu(V_id(vol)), newvN, vnode.uniquifier);
 593         return EIO;
 594     }
 595     FDH_REALLYCLOSE(fdP);
 596
 597     fid.Volume = V_id(vol);
 598     fid.Vnode = newvN;
 599     fid.Unique = vnode.uniquifier;
 600
 601     /*
 602      * Now  update the parent directory.
 603      */
 604
 605     ino = VNDISK_GET_INO(parent);
 606     SetSalvageDirHandle(&dir, V_id(vol), V_device(vol), ino);
 607
 608     code = afs_dir_Delete(&dir, name);
 609     if (code) {
 610         Log("splitvolume: couldn't delete directory entry for %s in %" AFS_VOLID_FMT ".%u.%u, code = %d\n",
 611             name, afs_printable_VolumeId_lu(V_id(vol)), vN, parent->uniquifier, code);
 612         return code;
 613     }
 614     code = afs_dir_Create(&dir, name, &fid);
 615     FidZap(&dir);
 616
 617     /* Make sure the directory file doesn't remain open */
 618     IH_INIT(hp, V_device(vol), V_parentId(vol), ino);
 619     fdP = IH_OPEN(hp);
 620     if (fdP)
 621         FDH_REALLYCLOSE(fdP);
 622     IH_RELEASE(hp);
 623
 624     class = vLarge;
 625     vcp = &VnodeClassInfo[class];
 626     fdP = IH_OPEN(vol->vnodeIndex[class].handle);
 627     offset = (vN + 1 - class) << (vcp->logSize -1);
 628     parent->dataVersion++;
 629     if (FDH_PWRITE(fdP, parent, vcp->diskSize, offset) != vcp->diskSize) {
 630         Log("split volume: couldn't write vnode for parent directory %" AFS_VOLID_FMT ".%u.%u\n",
 631             afs_printable_VolumeId_lu(V_id(vol)), vN, parent->uniquifier);
 632         return EIO;
 633     }
 634     FDH_REALLYCLOSE(fdP);
 635     return code;
 636 }
 637
 638 static afs_int32
 639 deleteVnodes(Volume *vol, afs_int32 class,
 640              struct VnodeExtract *list, afs_int32 length,
 641              afs_uint64 *blocks)
 642 {
 643     afs_int32 i, code = 0;
 644     char buf[SIZEOF_LARGEDISKVNODE];
 645     struct VnodeDiskObject *vnode = (struct VnodeDiskObject *)&buf;
 646     FdHandle_t *fdP = 0;
 647     struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
 648     struct VnodeExtract *e;
 649     afs_sfsize_t size;
 650     afs_uint64 offset;
 651     Inode ino;
 652
 653     fdP = IH_OPEN(vol->vnodeIndex[class].handle);
 654     if (!fdP) {
 655         Log("Couldn't open %s Index of volume %" AFS_VOLID_FMT "\n",
 656             class ? "small":"large", afs_printable_VolumeId_lu(V_id(vol)));
 657         code = EIO;
 658         goto Bad_Delete;
 659     }
 660
 661     for (i=0; i<length; i++) {
 662         e = &list[i];
 663         if (e->flag & NEEDED) {
 664             offset = (e->vN + 1 - class) << (vcp->logSize -1);
 665             if (FDH_PREAD(fdP, vnode, vcp->diskSize, offset) != vcp->diskSize) {
 666                 Log("Couldn't read in %s Index of volume %" AFS_VOLID_FMT " at offset %llu\n",
 667                     class ? "small":"large", afs_printable_VolumeId_lu(V_id(vol)), offset);
 668                 code = EIO;
 669                 goto Bad_Delete;
 670             }
 671             VNDISK_GET_LEN(size, vnode);
 672             *blocks += (size + 0x3ff) >> 10;
 673             ino = VNDISK_GET_INO(vnode);
 674             if (ino) {
 675                 IHandle_t *h;
 676                 IH_INIT(h, vol->device, V_parentId(vol), ino);
 677                     IH_DEC(h, ino, V_parentId(vol));
 678 #ifdef AFS_RXOSD_SUPPORT
 679             } else {
 680                 code = osdRemove(vol, vnode, e->vN);
 681 #endif /*  AFS_RXOSD_SUPPORT */
 682             }
 683             memset(vnode, 0, vcp->diskSize);
 684             vnode->type = vNull;
 685             if (FDH_PWRITE(fdP, vnode, vcp->diskSize, offset) != vcp->diskSize) {
 686                    Log("Couldn't write in %s Index of volume %" AFS_VOLID_FMT " to offset %llu\n",
 687                        class ? "small":"large", afs_printable_VolumeId_lu(V_id(vol)), offset);
 688             }
 689         }
 690     }
 691 Bad_Delete:
 692     if (fdP)
 693         FDH_CLOSE(fdP);
 694     return code;
 695 }
 696
 697 afs_int32
 698 split_volume(struct rx_call *call, Volume *vol, Volume *newvol,
 699              afs_uint32 where, afs_int32 verbose)
 700 {
 701     Error code = 0;
 702     struct VnodeExtract *dirList = 0;
 703     struct VnodeExtract *fileList = 0;
 704     afs_uint64 blocks = 0;
 705     afs_uint32 filesNeeded, dirsNeeded;
 706     afs_uint32 dl, fl;
 707     char buf[SIZEOF_LARGEDISKVNODE];
 708     char buf2[SIZEOF_LARGEDISKVNODE];
 709     struct VnodeDiskObject *rootVnode = (struct VnodeDiskObject *)&buf;
 710     struct VnodeDiskObject *parVnode = (struct VnodeDiskObject *)&buf2;
 711     char name[256];
 712     afs_uint32 parent;
 713     struct Msg *m;
 714
 715     m = calloc(1, sizeof(struct Msg));
 716     m->call = call;
 717     m->verbose = verbose;
 718
 719     /*
 720      *  First step: planning
 721      *
 722      *  Find out which directories will belong to the new volume
 723      *
 724      */
 725     if (verbose) {
 726         sprintf(m->line,
 727                 "1st step: extract vnode essence from large vnode file\n");
 728         rx_Write(m->call, m->line, strlen(m->line));
 729     }
 730
 731     code = ExtractVnodes(m, vol, vLarge, &dirList, &dl, where, rootVnode,
 732                         &parent, parVnode);
 733     if (code) {
 734         sprintf(m->line,
 735                 "ExtractVnodes failed for %" AFS_VOLID_FMT " for directories with code %d\n",
 736                 afs_printable_VolumeId_lu(V_id(vol)), code);
 737         rx_Write(m->call, m->line, strlen(m->line));
 738         return code;
 739     }
 740
 741     if (verbose) {
 742         sprintf(m->line, "2nd step: look for name of vnode %u in directory %" AFS_VOLID_FMT ".%u.%u\n",
 743                 where, afs_printable_VolumeId_lu(V_id(vol)), parent, parVnode->uniquifier);
 744         rx_Write(m->call, m->line, strlen(m->line));
 745     }
 746     code = findName(vol, parVnode, where, rootVnode->uniquifier,
 747                     name,  sizeof(name));
 748     if (code) {
 749         sprintf(m->line,
 750                 "splitvolume: could'nt find name of %u in directory %" AFS_VOLID_FMT ".%u.%u.\n",
 751                 where, afs_printable_VolumeId_lu(V_id(vol)), parent, parVnode->uniquifier);
 752         rx_Write(m->call, m->line, strlen(m->line));
 753         return code;
 754     }
 755     if (verbose) {
 756         sprintf(m->line, "name of %u is %s\n", where, name);
 757         rx_Write(m->call, m->line, strlen(m->line));
 758     }
 759
 760     if (verbose) {
 761         sprintf(m->line, "3rd step: find all directory vnodes belonging to the subtree under %u \"%s\"\n",
 762                         where, name);
 763         rx_Write(m->call, m->line, strlen(m->line));
 764     }
 765     code = FindVnodes(m, where, dirList, dl, dirList, dl, &dirsNeeded, 1);
 766     if (code) {
 767         sprintf(m->line,
 768                 "FindVnodes for directories failed with code %d\n", code);
 769         rx_Write(m->call, m->line, strlen(m->line));
 770         return code;
 771     }
 772
 773     if (verbose) {
 774         sprintf(m->line, "4th step extract vnode essence from small vnode file\n");
 775         rx_Write(m->call, m->line, strlen(m->line));
 776     }
 777     code = ExtractVnodes(m, vol, vSmall, &fileList, &fl, where, 0, 0, 0);
 778     if (code) {
 779         sprintf(m->line,
 780                 "ExtractVnodes failed for %" AFS_VOLID_FMT " for files with code %d\n",
 781                 afs_printable_VolumeId_lu(V_id(vol)), code);
 782         rx_Write(m->call, m->line, strlen(m->line));
 783         return code;
 784     }
 785     if (verbose) {
 786         sprintf(m->line, "5th step: find all small vnodes belonging to the subtree under %u \"%s\"\n",
 787                         where, name);
 788         rx_Write(m->call, m->line, strlen(m->line));
 789     }
 790     FindVnodes(m, where, fileList, fl, dirList, dl, &filesNeeded, 0);
 791
 792     /*
 793      *  Third step: create hard links for all files needed
 794      *
 795      */
 796
 797     V_destroyMe(newvol) = DESTROY_ME;
 798     V_inService(newvol) = 0;
 799     if (verbose) {
 800         sprintf(m->line, "6th step: create hard links in the AFSIDat tree between files of the old and new volume\n");
 801         rx_Write(m->call, m->line, strlen(m->line));
 802     }
 803     code = copyVnodes(m, vol, newvol, 1, fileList, fl, where, &blocks, 0);
 804     if (code) {
 805         sprintf(m->line, "copyVnodes for files failed with code %d\n", code);
 806         rx_Write(m->call, m->line, strlen(m->line));
 807         return code;
 808     }
 809
 810     /*
 811      *  Forth step: create hard links for all directories and copy
 812      *  split directory to new root directory
 813      */
 814
 815     if (verbose) {
 816         sprintf(m->line, "7th step: create hard links in the AFSIDat tree between directories of the old and new volume and make dir %u to new volume's root directory.\n",
 817                 where);
 818         rx_Write(m->call, m->line, strlen(m->line));
 819     }
 820     code = copyVnodes(m, vol, newvol, 0, dirList, dl, where, &blocks, parVnode);
 821     if (code) {
 822         sprintf(m->line, "copyVnodes for directories failed with code %d\n", code);
 823         rx_Write(m->call, m->line, strlen(m->line));
 824         return code;
 825     }
 826
 827     /*
 828      *  Finalize new volume
 829      *
 830      */
 831     if (verbose) {
 832         sprintf(m->line, "8th step: write new volume's metadata to disk\n");
 833         rx_Write(m->call, m->line, strlen(m->line));
 834     }
 835
 836     V_diskused(newvol) = blocks;
 837 #ifdef AFS_RXOSD_SUPPORT
 838     V_osdFlag(newvol) = V_osdFlag(vol);
 839 #endif
 840     V_filecount(newvol) = filesNeeded + dirsNeeded;
 841     V_destroyMe(newvol) = 0;
 842     V_maxquota(newvol) = V_maxquota(vol);
 843     V_uniquifier(newvol) = V_uniquifier(vol);
 844     V_inService(newvol) = 1;
 845     VUpdateVolume(&code, newvol);
 846
 847     /*
 848      *  Sixth step: change directory entry in old volume:
 849      *  rename old tree and create mount point for new volume.
 850      */
 851     if (verbose) {
 852         sprintf(m->line, "9th step: create mountpoint \"%s\" for new volume in old volume's directory %u.\n", name, parent);
 853         rx_Write(m->call, m->line, strlen(m->line));
 854     }
 855
 856     code = createMountpoint(vol, newvol, parVnode, parent, rootVnode, name);
 857     if (code) {
 858         sprintf(m->line, "createMountpoint failed with code %d\n", code);
 859         rx_Write(m->call, m->line, strlen(m->line));
 860         return code;
 861     }
 862     /*
 863      * Now both volumes should be ready and consistent, but the old volume
 864      * contains still the vnodes and data we transferred into the new one.
 865      * Delete orphaned vnodes and data.
 866      */
 867
 868     blocks = 0;
 869     if (verbose) {
 870         sprintf(m->line, "10th step: delete large vnodes belonging to subtree in the old volume.\n");
 871         rx_Write(m->call, m->line, strlen(m->line));
 872     }
 873     deleteVnodes(vol, vLarge, dirList, dl, &blocks);
 874     if (verbose) {
 875         sprintf(m->line, "11th step: delete small vnodes belonging to subtree in the old volume.\n");
 876         rx_Write(m->call, m->line, strlen(m->line));
 877     }
 878     deleteVnodes(vol, vSmall, fileList, fl, &blocks);
 879     V_diskused(vol) -= blocks;
 880     V_filecount(vol) -= (filesNeeded + dirsNeeded + 1);
 881     VUpdateVolume(&code, vol);
 882
 883     sprintf(m->line, "Finished!\n");
 884     rx_Write(m->call, m->line, strlen(m->line));
 885     m->line[0] = 0;
 886     m->line[1] = 0;
 887     m->line[2] = 0;
 888     m->line[3] = 0;
 889     rx_Write(m->call, m->line, 4);
 890     free(m);
 891     return code;
 892 }
 893 #endif