src/vol/clone.c

   1 /*
   2  * Copyright 2000, International Business Machines Corporation and others.
   3  * All Rights Reserved.
   4  *
   5  * This software has been released under the terms of the IBM Public
   6  * License.  For details, see the LICENSE file in the top-level source
   7  * directory or online at http://www.openafs.org/dl/license10.html
   8  */
   9
  10 /*
  11         System:         VICE-TWO
  12         Module:         clone.c
  13
  14  */
  15
  16 /* Clone a volume.  Assumes the new volume is already created */
  17
  18 #include <afsconfig.h>
  19 #include <afs/param.h>
  20
  21 #include <roken.h>
  22
  23 #ifdef AFS_NT40_ENV
  24 #include <windows.h>
  25 #include <winbase.h>
  26 #endif
  27
  28 #include <rx/xdr.h>
  29 #include <afs/afsint.h>
  30 #include <afs/afssyscalls.h>
  31 #include <rx/rx_queue.h>
  32
  33 #include "nfs.h"
  34 #include "lwp.h"
  35 #include "lock.h"
  36 #include "ihandle.h"
  37 #include "vnode.h"
  38 #include "volume.h"
  39 #include "partition.h"
  40 #include "viceinode.h"
  41 #include "vol_prototypes.h"
  42 #include "common.h"
  43
  44 int (*vol_PollProc) (void) = 0; /* someone must init this */
  45
  46 #define ERROR_EXIT(code) do { \
  47     error = code; \
  48     goto error_exit; \
  49 } while (0)
  50
  51 /* parameters for idec call - this could just be an IHandle_t, but leaving
  52  * open the possibility of decrementing the special files as well.
  53  */
  54 struct clone_rock {
  55     IHandle_t *h;
  56     VolId vol;
  57 };
  58
  59 #define CLONE_MAXITEMS  100
  60 struct clone_items {
  61     struct clone_items *next;
  62     afs_int32 nitems;
  63     Inode data[CLONE_MAXITEMS];
  64 };
  65
  66 struct clone_head {
  67     struct clone_items *first;
  68     struct clone_items *last;
  69 };
  70
  71 void CloneVolume(Error *, Volume *, Volume *, Volume *);
  72
  73 static int
  74 ci_AddItem(struct clone_head *ah, Inode aino)
  75 {
  76     struct clone_items *ti;
  77
  78     /* if no last elt (first call) or last item full, get a new one */
  79     if ((!ah->last) || ah->last->nitems >= CLONE_MAXITEMS) {
  80         ti = malloc(sizeof(struct clone_items));
  81         if (!ti) {
  82             Log("ci_AddItem: malloc failed\n");
  83             osi_Panic("ci_AddItem: malloc failed\n");
  84         }
  85         ti->nitems = 0;
  86         ti->next = (struct clone_items *)0;
  87         if (ah->last) {
  88             ah->last->next = ti;
  89             ah->last = ti;
  90         } else {
  91             /* first dude in the list */
  92             ah->first = ah->last = ti;
  93         }
  94     } else
  95         ti = ah->last;
  96
  97     /* now ti points to the end of the list, to a clone_item with room
  98      * for at least one more element.  Add it.
  99      */
 100     ti->data[ti->nitems++] = aino;
 101     return 0;
 102 }
 103
 104 /* initialize a clone header */
 105 int
 106 ci_InitHead(struct clone_head *ah)
 107 {
 108     memset(ah, 0, sizeof(*ah));
 109     return 0;
 110 }
 111
 112 /* apply a function to all dudes in the set */
 113 int
 114 ci_Apply(struct clone_head *ah, int (*aproc) (Inode,  void *), void *arock)
 115 {
 116     struct clone_items *ti;
 117     int i;
 118
 119     for (ti = ah->first; ti; ti = ti->next) {
 120         for (i = 0; i < ti->nitems; i++) {
 121             (*aproc) (ti->data[i], arock);
 122         }
 123     }
 124     return 0;
 125 }
 126
 127 /* free all dudes in the list */
 128 int
 129 ci_Destroy(struct clone_head *ah)
 130 {
 131     struct clone_items *ti, *ni;
 132
 133     for (ti = ah->first; ti; ti = ni) {
 134         ni = ti->next;          /* guard against freeing */
 135         free(ti);
 136     }
 137     return 0;
 138 }
 139
 140 static int
 141 IDecProc(Inode adata, void *arock)
 142 {
 143     struct clone_rock *aparm = (struct clone_rock *)arock;
 144     IH_DEC(aparm->h, adata, aparm->vol);
 145     DOPOLL;
 146     return 0;
 147 }
 148
 149 afs_int32
 150 DoCloneIndex(Volume * rwvp, Volume * clvp, VnodeClass class, int reclone)
 151 {
 152     afs_int32 code, error = 0;
 153     FdHandle_t *rwFd = 0, *clFdIn = 0, *clFdOut = 0;
 154     StreamHandle_t *rwfile = 0, *clfilein = 0, *clfileout = 0;
 155     IHandle_t *rwH = 0, *clHin = 0, *clHout = 0;
 156     char buf[SIZEOF_LARGEDISKVNODE], dbuf[SIZEOF_LARGEDISKVNODE];
 157     struct VnodeDiskObject *rwvnode = (struct VnodeDiskObject *)buf;
 158     struct VnodeDiskObject *clvnode = (struct VnodeDiskObject *)dbuf;
 159     Inode rwinode = 0;
 160     Inode clinode;
 161     struct clone_head decHead;
 162     struct clone_rock decRock;
 163     afs_foff_t offset = 0;
 164     afs_int32 dircloned, inodeinced;
 165     afs_int32 filecount = 0, diskused = 0;
 166     afs_ino_str_t stmp;
 167
 168     struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
 169     /*
 170      * The fileserver's -readonly switch should make this false, but we
 171      * have no useful way to know in the volserver.
 172      * This doesn't make client data mutable.
 173      */
 174     int ReadWriteOriginal = 1;
 175
 176     /* Correct number of files in volume: this assumes indexes are always
 177        cloned starting with vLarge */
 178     if (ReadWriteOriginal && class != vLarge) {
 179         filecount = V_filecount(rwvp);
 180         diskused = V_diskused(rwvp);
 181     }
 182
 183     /* Initialize list of inodes to nuke - must do this before any calls
 184      * to ERROR_EXIT, as the error handler requires an initialised list
 185      */
 186     ci_InitHead(&decHead);
 187     decRock.h = V_linkHandle(rwvp);
 188     decRock.vol = V_parentId(rwvp);
 189
 190     /* Open the RW volume's index file and seek to beginning */
 191     IH_COPY(rwH, rwvp->vnodeIndex[class].handle);
 192     rwFd = IH_OPEN(rwH);
 193     if (!rwFd)
 194         ERROR_EXIT(EIO);
 195     rwfile = FDH_FDOPEN(rwFd, ReadWriteOriginal ? "r+" : "r");
 196     if (!rwfile)
 197         ERROR_EXIT(EIO);
 198     STREAM_ASEEK(rwfile, vcp->diskSize);        /* Will fail if no vnodes */
 199
 200     /* Open the clone volume's index file and seek to beginning */
 201     IH_COPY(clHout, clvp->vnodeIndex[class].handle);
 202     clFdOut = IH_OPEN(clHout);
 203     if (!clFdOut)
 204         ERROR_EXIT(EIO);
 205     clfileout = FDH_FDOPEN(clFdOut, "a");
 206     if (!clfileout)
 207         ERROR_EXIT(EIO);
 208     code = STREAM_ASEEK(clfileout, vcp->diskSize);
 209     if (code)
 210         ERROR_EXIT(EIO);
 211
 212     /* If recloning, open the new volume's index; this time for
 213      * reading. We never read anything that we're simultaneously
 214      * writing, so this all works.
 215      */
 216     if (reclone) {
 217         IH_COPY(clHin, clvp->vnodeIndex[class].handle);
 218         clFdIn = IH_OPEN(clHin);
 219         if (!clFdIn)
 220             ERROR_EXIT(EIO);
 221         clfilein = FDH_FDOPEN(clFdIn, "r");
 222         if (!clfilein)
 223             ERROR_EXIT(EIO);
 224         STREAM_ASEEK(clfilein, vcp->diskSize);  /* Will fail if no vnodes */
 225     }
 226
 227     /* Read each vnode in the old volume's index file */
 228     for (offset = vcp->diskSize;
 229          STREAM_READ(rwvnode, vcp->diskSize, 1, rwfile) == 1;
 230          offset += vcp->diskSize) {
 231         dircloned = inodeinced = 0;
 232
 233         /* If we are recloning the volume, read the corresponding vnode
 234          * from the clone and determine its inode number.
 235          */
 236         if (reclone && !STREAM_EOF(clfilein)
 237             && (STREAM_READ(clvnode, vcp->diskSize, 1, clfilein) == 1)) {
 238             clinode = VNDISK_GET_INO(clvnode);
 239         } else {
 240             clinode = 0;
 241         }
 242
 243         if (rwvnode->type != vNull) {
 244             afs_fsize_t ll;
 245
 246             if (rwvnode->vnodeMagic != vcp->magic)
 247                 ERROR_EXIT(-1);
 248             rwinode = VNDISK_GET_INO(rwvnode);
 249             filecount++;
 250             VNDISK_GET_LEN(ll, rwvnode);
 251             diskused += nBlocks(ll);
 252
 253             /* Increment the inode if not already */
 254             if (clinode && (clinode == rwinode)) {
 255                 clinode = 0;    /* already cloned - don't delete later */
 256             } else if (rwinode) {
 257                 if (IH_INC(V_linkHandle(rwvp), rwinode, V_parentId(rwvp)) ==
 258                     -1) {
 259                     Log("IH_INC failed: %"AFS_PTR_FMT", %s, %" AFS_VOLID_FMT " errno %d\n",
 260                         V_linkHandle(rwvp), PrintInode(stmp, rwinode),
 261                         afs_printable_VolumeId_lu(V_parentId(rwvp)), errno);
 262                     VForceOffline(rwvp);
 263                     ERROR_EXIT(EIO);
 264                 }
 265                 inodeinced = 1;
 266             }
 267
 268             /* If a directory, mark vnode in old volume as cloned */
 269             if ((rwvnode->type == vDirectory) && ReadWriteOriginal) {
 270 #ifdef DVINC
 271                 /*
 272                  * It is my firmly held belief that immediately after
 273                  * copy-on-write, the two directories can be identical.
 274                  * If the new copy is changed (presumably, that is the very
 275                  * next thing that will happen) then the dataVersion will
 276                  * get bumped.
 277                  */
 278                 /* NOTE:  the dataVersion++ is incredibly important!!!.
 279                  * This will cause the inode created by the file server
 280                  * on copy-on-write to be stamped with a dataVersion bigger
 281                  * than the current one.  The salvager will then do the
 282                  * right thing */
 283                 rwvnode->dataVersion++;
 284 #endif /* DVINC */
 285                 rwvnode->cloned = 1;
 286                 code = STREAM_ASEEK(rwfile, offset);
 287                 if (code == -1)
 288                     goto clonefailed;
 289                 code = STREAM_WRITE(rwvnode, vcp->diskSize, 1, rwfile);
 290                 if (code != 1)
 291                     goto clonefailed;
 292                 dircloned = 1;
 293                 code = STREAM_ASEEK(rwfile, offset + vcp->diskSize);
 294                 if (code == -1)
 295                     goto clonefailed;
 296 #ifdef DVINC
 297                 rwvnode->dataVersion--; /* Really needs to be set to the value in the inode,
 298                                          * for the read-only volume */
 299 #endif /* DVINC */
 300             }
 301         }
 302
 303         /* Overwrite the vnode entry in the clone volume */
 304         rwvnode->cloned = 0;
 305         code = STREAM_WRITE(rwvnode, vcp->diskSize, 1, clfileout);
 306         if (code != 1) {
 307           clonefailed:
 308             /* Couldn't clone, go back and decrement the inode's link count */
 309             if (inodeinced) {
 310                 if (IH_DEC(V_linkHandle(rwvp), rwinode, V_parentId(rwvp)) ==
 311                     -1) {
 312                     Log("IH_DEC failed: %"AFS_PTR_FMT", %s, %" AFS_VOLID_FMT " errno %d\n",
 313                         V_linkHandle(rwvp), PrintInode(stmp, rwinode),
 314                         afs_printable_VolumeId_lu(V_parentId(rwvp)), errno);
 315                     VForceOffline(rwvp);
 316                     ERROR_EXIT(EIO);
 317                 }
 318             }
 319             /* And if the directory was marked clone, unmark it */
 320             if (dircloned) {
 321                 rwvnode->cloned = 0;
 322                 if (STREAM_ASEEK(rwfile, offset) != -1)
 323                     (void)STREAM_WRITE(rwvnode, vcp->diskSize, 1, rwfile);
 324             }
 325             ERROR_EXIT(EIO);
 326         }
 327
 328         /* Removal of the old cloned inode */
 329         if (clinode) {
 330             ci_AddItem(&decHead, clinode);      /* just queue it */
 331         }
 332
 333         DOPOLL;
 334     }
 335     if (STREAM_ERROR(clfileout))
 336         ERROR_EXIT(EIO);
 337
 338     /* Clean out any junk at end of clone file */
 339     if (reclone) {
 340         STREAM_ASEEK(clfilein, offset);
 341         while (STREAM_READ(clvnode, vcp->diskSize, 1, clfilein) == 1) {
 342             if (clvnode->type != vNull && VNDISK_GET_INO(clvnode) != 0) {
 343                 ci_AddItem(&decHead, VNDISK_GET_INO(clvnode));
 344             }
 345             DOPOLL;
 346         }
 347     }
 348
 349     /* come here to finish up.  If code is non-zero, we've already run into problems,
 350      * and shouldn't do the idecs.
 351      */
 352   error_exit:
 353     if (rwfile)
 354         STREAM_CLOSE(rwfile);
 355     if (clfilein)
 356         STREAM_CLOSE(clfilein);
 357     if (clfileout)
 358         STREAM_CLOSE(clfileout);
 359
 360     if (rwFd)
 361         FDH_CLOSE(rwFd);
 362     if (clFdIn)
 363         FDH_CLOSE(clFdIn);
 364     if (clFdOut)
 365         FDH_CLOSE(clFdOut);
 366
 367     if (rwH)
 368         IH_RELEASE(rwH);
 369     if (clHout)
 370         IH_RELEASE(clHout);
 371     if (clHin)
 372         IH_RELEASE(clHin);
 373
 374     /* Next, we sync the disk. We have to reopen in case we're truncating,
 375      * since we were using stdio above, and don't know when the buffers
 376      * would otherwise be flushed.  There's no stdio fftruncate call.
 377      */
 378     rwFd = IH_OPEN(clvp->vnodeIndex[class].handle);
 379     if (rwFd == NULL) {
 380         if (!error)
 381             error = EIO;
 382     } else {
 383         if (reclone) {
 384             /* If doing a reclone, we're keeping the clone. We need to
 385              * truncate the file to offset bytes.
 386              */
 387             if (reclone && !error) {
 388                 error = FDH_TRUNC(rwFd, offset);
 389             }
 390         }
 391         (void)FDH_SYNC(rwFd);
 392         FDH_CLOSE(rwFd);
 393     }
 394
 395     /* Now finally do the idec's.  At this point, all potential
 396      * references have been cleaned up and sent to the disk
 397      * (see above fclose and fsync). No matter what happens, we
 398      * no longer need to keep these references around.
 399      */
 400     code = ci_Apply(&decHead, IDecProc, (char *)&decRock);
 401     if (!error)
 402         error = code;
 403     ci_Destroy(&decHead);
 404
 405     if (ReadWriteOriginal && filecount > 0)
 406         V_filecount(rwvp) = filecount;
 407     if (ReadWriteOriginal && diskused > 0)
 408         V_diskused(rwvp) = diskused;
 409     return error;
 410 }
 411
 412 void
 413 CloneVolume(Error * rerror, Volume * original, Volume * new, Volume * old)
 414 {
 415     afs_int32 code, error = 0;
 416     afs_int32 reclone;
 417     afs_int32 filecount = V_filecount(original), diskused = V_diskused(original);
 418
 419     *rerror = 0;
 420     reclone = ((new == old) ? 1 : 0);
 421
 422     code = DoCloneIndex(original, new, vLarge, reclone);
 423     if (code)
 424         ERROR_EXIT(code);
 425     code = DoCloneIndex(original, new, vSmall, reclone);
 426     if (code)
 427         ERROR_EXIT(code);
 428     if (filecount != V_filecount(original) || diskused != V_diskused(original))
 429         Log("Clone %" AFS_VOLID_FMT ": filecount %d -> %d diskused %d -> %d\n",
 430             afs_printable_VolumeId_lu(V_id(original)), filecount,
 431             V_filecount(original), diskused, V_diskused(original));
 432
 433     code = CopyVolumeHeader(&V_disk(original), &V_disk(new));
 434     if (code)
 435         ERROR_EXIT(code);
 436
 437   error_exit:
 438     *rerror = error;
 439 }