sys/vfs/hammer/hammer_pfs.c

   1 /*
   2  * Copyright (c) 2008 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vfs/hammer/hammer_pfs.c,v 1.1.2.3 2008/07/19 18:46:20 dillon Exp $
  35  */
  36 /*
  37  * HAMMER PFS ioctls - Manage pseudo-fs configurations
  38  */
  39
  40 #include "hammer.h"
  41
  42 static int hammer_pfs_autodetect(struct hammer_ioc_pseudofs_rw *pfs,
  43                                 hammer_inode_t ip);
  44 static int hammer_pfs_rollback(hammer_transaction_t trans,
  45                                 hammer_pseudofs_inmem_t pfsm,
  46                                 hammer_tid_t trunc_tid);
  47 static int hammer_pfs_delete_at_cursor(hammer_cursor_t cursor,
  48                                 hammer_tid_t trunc_tid);
  49
  50 /*
  51  * Get mirroring/pseudo-fs information
  52  *
  53  * NOTE: The ip used for ioctl is not necessarily related to the PFS
  54  */
  55 int
  56 hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
  57                         struct hammer_ioc_pseudofs_rw *pfs)
  58 {
  59         hammer_pseudofs_inmem_t pfsm;
  60         u_int32_t localization;
  61         int error;
  62
  63         if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
  64                 return(error);
  65         localization = (u_int32_t)pfs->pfs_id << 16;
  66         pfs->bytes = sizeof(struct hammer_pseudofs_data);
  67         pfs->version = HAMMER_IOC_PSEUDOFS_VERSION;
  68
  69         pfsm = hammer_load_pseudofs(trans, localization, &error);
  70         if (error) {
  71                 hammer_rel_pseudofs(trans->hmp, pfsm);
  72                 return(error);
  73         }
  74
  75         /*
  76          * If the PFS is a master the sync tid is set by normal operation
  77          * rather then the mirroring code, and will always track the
  78          * real HAMMER filesystem.
  79          */
  80         if ((pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0)
  81                 pfsm->pfsd.sync_end_tid = trans->rootvol->ondisk->vol0_next_tid;
  82
  83         /*
  84          * Copy out to userland.
  85          */
  86         error = 0;
  87         if (pfs->ondisk && error == 0)
  88                 error = copyout(&pfsm->pfsd, pfs->ondisk, sizeof(pfsm->pfsd));
  89         hammer_rel_pseudofs(trans->hmp, pfsm);
  90         return(error);
  91 }
  92
  93 /*
  94  * Set mirroring/pseudo-fs information
  95  *
  96  * NOTE: The ip used for ioctl is not necessarily related to the PFS
  97  */
  98 int
  99 hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
 100                         struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs)
 101 {
 102         hammer_pseudofs_inmem_t pfsm;
 103         u_int32_t localization;
 104         int error;
 105
 106         if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
 107                 return(error);
 108         localization = (u_int32_t)pfs->pfs_id << 16;
 109         if (pfs->version != HAMMER_IOC_PSEUDOFS_VERSION)
 110                 error = EINVAL;
 111         localization = (u_int32_t)pfs->pfs_id << 16;
 112
 113         if (error == 0 && pfs->ondisk) {
 114                 /*
 115                  * Load the PFS so we can modify our in-core copy.  Ignore
 116                  * ENOENT errors.
 117                  */
 118                 pfsm = hammer_load_pseudofs(trans, localization, &error);
 119                 error = copyin(pfs->ondisk, &pfsm->pfsd, sizeof(pfsm->pfsd));
 120
 121                 /*
 122                  * Save it back, create a root inode if we are in master
 123                  * mode and no root exists.
 124                  */
 125                 if (error == 0)
 126                         error = hammer_mkroot_pseudofs(trans, cred, pfsm);
 127                 if (error == 0)
 128                         error = hammer_save_pseudofs(trans, pfsm);
 129                 hammer_rel_pseudofs(trans->hmp, pfsm);
 130         }
 131         return(error);
 132 }
 133
 134 /*
 135  * Upgrade a slave to a master
 136  *
 137  * This is fairly easy to do, but we must physically undo any partial syncs
 138  * for transaction ids > sync_end_tid.  Effective, we must do a partial
 139  * rollback.
 140  *
 141  * NOTE: The ip used for ioctl is not necessarily related to the PFS
 142  */
 143 int
 144 hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
 145                         struct hammer_ioc_pseudofs_rw *pfs)
 146 {
 147         hammer_pseudofs_inmem_t pfsm;
 148         u_int32_t localization;
 149         int error;
 150
 151         if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
 152                 return(error);
 153         localization = (u_int32_t)pfs->pfs_id << 16;
 154         if ((error = hammer_unload_pseudofs(trans, localization)) != 0)
 155                 return(error);
 156
 157         /*
 158          * A master id must be set when upgrading
 159          */
 160         pfsm = hammer_load_pseudofs(trans, localization, &error);
 161         if (error == 0) {
 162                 if ((pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) != 0) {
 163                         error = hammer_pfs_rollback(trans, pfsm,
 164                                             pfsm->pfsd.sync_end_tid + 1);
 165                         if (error == 0) {
 166                                 pfsm->pfsd.mirror_flags &= ~HAMMER_PFSD_SLAVE;
 167                                 error = hammer_save_pseudofs(trans, pfsm);
 168                         }
 169                 }
 170         }
 171         hammer_rel_pseudofs(trans->hmp, pfsm);
 172         if (error == EINTR) {
 173                 pfs->head.flags |= HAMMER_IOC_HEAD_INTR;
 174                 error = 0;
 175         }
 176         return (error);
 177 }
 178
 179 /*
 180  * Downgrade a master to a slave
 181  *
 182  * This is really easy to do, just set the SLAVE flag.
 183  *
 184  * We also leave sync_end_tid intact... the field is not used in master
 185  * mode (vol0_next_tid overrides it), but if someone switches to master
 186  * mode accidently and then back to slave mode we don't want it to change.
 187  * Eventually it will be used as the cross-synchronization TID in
 188  * multi-master mode, and we don't want to mess with it for that feature
 189  * either.
 190  *
 191  * NOTE: The ip used for ioctl is not necessarily related to the PFS
 192  */
 193 int
 194 hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
 195                         struct hammer_ioc_pseudofs_rw *pfs)
 196 {
 197         hammer_pseudofs_inmem_t pfsm;
 198         u_int32_t localization;
 199         int error;
 200
 201         if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
 202                 return(error);
 203         localization = (u_int32_t)pfs->pfs_id << 16;
 204         if ((error = hammer_unload_pseudofs(trans, localization)) != 0)
 205                 return(error);
 206
 207         pfsm = hammer_load_pseudofs(trans, localization, &error);
 208         if (error == 0) {
 209                 if ((pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0) {
 210                         pfsm->pfsd.mirror_flags |= HAMMER_PFSD_SLAVE;
 211                         error = hammer_save_pseudofs(trans, pfsm);
 212                 }
 213         }
 214         hammer_rel_pseudofs(trans->hmp, pfsm);
 215         return (error);
 216 }
 217
 218 /*
 219  * Destroy a PFS
 220  *
 221  * We can destroy a PFS by scanning and deleting all of its records in the
 222  * B-Tree.  The hammer utility will delete the softlink in the primary
 223  * filesystem.
 224  *
 225  * NOTE: The ip used for ioctl is not necessarily related to the PFS
 226  */
 227 int
 228 hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
 229                         struct hammer_ioc_pseudofs_rw *pfs)
 230 {
 231         hammer_pseudofs_inmem_t pfsm;
 232         u_int32_t localization;
 233         int error;
 234
 235         if ((error = hammer_pfs_autodetect(pfs, ip)) != 0)
 236                 return(error);
 237         localization = (u_int32_t)pfs->pfs_id << 16;
 238
 239         if ((error = hammer_unload_pseudofs(trans, localization)) != 0)
 240                 return(error);
 241
 242         pfsm = hammer_load_pseudofs(trans, localization, &error);
 243         if (error == 0) {
 244                 error = hammer_pfs_rollback(trans, pfsm, 0);
 245                 if (error == 0) {
 246                         pfsm->pfsd.mirror_flags |= HAMMER_PFSD_DELETED;
 247                         error = hammer_save_pseudofs(trans, pfsm);
 248                 }
 249         }
 250         hammer_rel_pseudofs(trans->hmp, pfsm);
 251         if (error == EINTR) {
 252                 pfs->head.flags |= HAMMER_IOC_HEAD_INTR;
 253                 error = 0;
 254         }
 255         return(error);
 256 }
 257
 258 /*
 259  * Auto-detect the pseudofs and do basic bounds checking.
 260  */
 261 static
 262 int
 263 hammer_pfs_autodetect(struct hammer_ioc_pseudofs_rw *pfs, hammer_inode_t ip)
 264 {
 265         int error = 0;
 266
 267         if (pfs->pfs_id == -1)
 268                 pfs->pfs_id = (int)(ip->obj_localization >> 16);
 269         if (pfs->pfs_id < 0 || pfs->pfs_id >= HAMMER_MAX_PFS)
 270                 error = EINVAL;
 271         if (pfs->bytes < sizeof(struct hammer_pseudofs_data))
 272                 error = EINVAL;
 273         return(error);
 274 }
 275
 276 /*
 277  * Rollback the specified PFS to (trunc_tid - 1), removing everything
 278  * greater or equal to trunc_tid.  The PFS must not have been in no-mirror
 279  * mode or the MIRROR_FILTERED scan will not work properly.
 280  *
 281  * This is typically used to remove any partial syncs when upgrading a
 282  * slave to a master.  It can theoretically also be used to rollback
 283  * any PFS, including PFS#0, BUT ONLY TO POINTS THAT HAVE NOT YET BEEN
 284  * PRUNED, and to points that are older only if they are on a retained
 285  * (pruning softlink) boundary.
 286  *
 287  * Rollbacks destroy information.  If you don't mind inode numbers changing
 288  * a better way would be to cpdup a snapshot back onto the master.
 289  */
 290 static
 291 int
 292 hammer_pfs_rollback(hammer_transaction_t trans,
 293                     hammer_pseudofs_inmem_t pfsm,
 294                     hammer_tid_t trunc_tid)
 295 {
 296         struct hammer_cmirror cmirror;
 297         struct hammer_cursor cursor;
 298         struct hammer_base_elm key_cur;
 299         int error;
 300         int seq;
 301
 302         bzero(&cmirror, sizeof(cmirror));
 303         bzero(&key_cur, sizeof(key_cur));
 304         key_cur.localization = HAMMER_MIN_LOCALIZATION + pfsm->localization;
 305         key_cur.obj_id = HAMMER_MIN_OBJID;
 306         key_cur.key = HAMMER_MIN_KEY;
 307         key_cur.create_tid = 1;
 308         key_cur.rec_type = HAMMER_MIN_RECTYPE;
 309
 310         seq = trans->hmp->flusher.act;
 311
 312 retry:
 313         error = hammer_init_cursor(trans, &cursor, NULL, NULL);
 314         if (error) {
 315                 hammer_done_cursor(&cursor);
 316                 goto failed;
 317         }
 318         cursor.key_beg = key_cur;
 319         cursor.key_end.localization = HAMMER_MAX_LOCALIZATION +
 320                                       pfsm->localization;
 321         cursor.key_end.obj_id = HAMMER_MAX_OBJID;
 322         cursor.key_end.key = HAMMER_MAX_KEY;
 323         cursor.key_end.create_tid = HAMMER_MAX_TID;
 324         cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
 325
 326         cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
 327         cursor.flags |= HAMMER_CURSOR_BACKEND;
 328
 329         /*
 330          * Do an optimized scan of only records created or modified
 331          * >= trunc_tid, so we can fix up those records.  We must
 332          * still check the TIDs but this greatly reduces the size of
 333          * the scan.
 334          */
 335         cursor.flags |= HAMMER_CURSOR_MIRROR_FILTERED;
 336         cursor.cmirror = &cmirror;
 337         cmirror.mirror_tid = trunc_tid;
 338
 339         error = hammer_btree_first(&cursor);
 340         while (error == 0) {
 341                 /*
 342                  * Abort the rollback.
 343                  */
 344                 if (error == 0) {
 345                         error = hammer_signal_check(trans->hmp);
 346                         if (error)
 347                                 break;
 348                 }
 349
 350                 /*
 351                  * We only care about leafs.  Internal nodes can be returned
 352                  * in mirror-filtered mode (they are used to generate SKIP
 353                  * mrecords), but we don't need them for this code.
 354                  */
 355                 cursor.flags |= HAMMER_CURSOR_ATEDISK;
 356                 if (cursor.node->ondisk->type == HAMMER_BTREE_TYPE_LEAF) {
 357                         key_cur = cursor.node->ondisk->elms[cursor.index].base;
 358                         error = hammer_pfs_delete_at_cursor(&cursor, trunc_tid);
 359                 }
 360
 361                 while (hammer_flusher_meta_halflimit(trans->hmp) ||
 362                        hammer_flusher_undo_exhausted(trans, 2)) {
 363                         hammer_unlock_cursor(&cursor, 0);
 364                         hammer_flusher_wait(trans->hmp, seq);
 365                         hammer_lock_cursor(&cursor, 0);
 366                         seq = hammer_flusher_async_one(trans->hmp);
 367                 }
 368
 369                 if (error == 0)
 370                         error = hammer_btree_iterate(&cursor);
 371         }
 372         if (error == ENOENT)
 373                 error = 0;
 374         hammer_done_cursor(&cursor);
 375         if (error == EDEADLK)
 376                 goto retry;
 377 failed:
 378         return(error);
 379 }
 380
 381 /*
 382  * Helper function - perform rollback on a B-Tree element given trunc_tid.
 383  *
 384  * If create_tid >= trunc_tid the record is physically destroyed.
 385  * If delete_tid >= trunc_tid it will be set to 0, undeleting the record.
 386  */
 387 static
 388 int
 389 hammer_pfs_delete_at_cursor(hammer_cursor_t cursor, hammer_tid_t trunc_tid)
 390 {
 391         hammer_btree_leaf_elm_t elm;
 392         hammer_transaction_t trans;
 393         int error;
 394
 395         elm = &cursor->node->ondisk->elms[cursor->index].leaf;
 396         if (elm->base.create_tid < trunc_tid &&
 397             elm->base.delete_tid < trunc_tid) {
 398                 return(0);
 399         }
 400         trans = cursor->trans;
 401
 402         if (elm->base.create_tid >= trunc_tid) {
 403                 error = hammer_delete_at_cursor(
 404                                 cursor, HAMMER_DELETE_DESTROY,
 405                                 cursor->trans->tid, cursor->trans->time32,
 406                                 1, NULL);
 407         } else if (elm->base.delete_tid >= trunc_tid) {
 408                 error = hammer_delete_at_cursor(
 409                                 cursor, HAMMER_DELETE_ADJUST,
 410                                 0, 0,
 411                                 1, NULL);
 412         } else {
 413                 error = 0;
 414         }
 415         return(error);
 416 }
 417