fs/xfs/xfs_extfree_item.c

   1 /*
   2  * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_log_format.h"
  21 #include "xfs_trans_resv.h"
  22 #include "xfs_sb.h"
  23 #include "xfs_ag.h"
  24 #include "xfs_mount.h"
  25 #include "xfs_trans.h"
  26 #include "xfs_trans_priv.h"
  27 #include "xfs_buf_item.h"
  28 #include "xfs_extfree_item.h"
  29 #include "xfs_log.h"
  30
  31
  32 kmem_zone_t     *xfs_efi_zone;
  33 kmem_zone_t     *xfs_efd_zone;
  34
  35 static inline struct xfs_efi_log_item *EFI_ITEM(struct xfs_log_item *lip)
  36 {
  37         return container_of(lip, struct xfs_efi_log_item, efi_item);
  38 }
  39
  40 void
  41 xfs_efi_item_free(
  42         struct xfs_efi_log_item *efip)
  43 {
  44         if (efip->efi_format.efi_nextents > XFS_EFI_MAX_FAST_EXTENTS)
  45                 kmem_free(efip);
  46         else
  47                 kmem_zone_free(xfs_efi_zone, efip);
  48 }
  49
  50 /*
  51  * Freeing the efi requires that we remove it from the AIL if it has already
  52  * been placed there. However, the EFI may not yet have been placed in the AIL
  53  * when called by xfs_efi_release() from EFD processing due to the ordering of
  54  * committed vs unpin operations in bulk insert operations. Hence the reference
  55  * count to ensure only the last caller frees the EFI.
  56  */
  57 STATIC void
  58 __xfs_efi_release(
  59         struct xfs_efi_log_item *efip)
  60 {
  61         struct xfs_ail          *ailp = efip->efi_item.li_ailp;
  62
  63         if (atomic_dec_and_test(&efip->efi_refcount)) {
  64                 spin_lock(&ailp->xa_lock);
  65                 /* xfs_trans_ail_delete() drops the AIL lock. */
  66                 xfs_trans_ail_delete(ailp, &efip->efi_item,
  67                                      SHUTDOWN_LOG_IO_ERROR);
  68                 xfs_efi_item_free(efip);
  69         }
  70 }
  71
  72 /*
  73  * This returns the number of iovecs needed to log the given efi item.
  74  * We only need 1 iovec for an efi item.  It just logs the efi_log_format
  75  * structure.
  76  */
  77 static inline int
  78 xfs_efi_item_sizeof(
  79         struct xfs_efi_log_item *efip)
  80 {
  81         return sizeof(struct xfs_efi_log_format) +
  82                (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
  83 }
  84
  85 STATIC void
  86 xfs_efi_item_size(
  87         struct xfs_log_item     *lip,
  88         int                     *nvecs,
  89         int                     *nbytes)
  90 {
  91         *nvecs += 1;
  92         *nbytes += xfs_efi_item_sizeof(EFI_ITEM(lip));
  93 }
  94
  95 /*
  96  * This is called to fill in the vector of log iovecs for the
  97  * given efi log item. We use only 1 iovec, and we point that
  98  * at the efi_log_format structure embedded in the efi item.
  99  * It is at this point that we assert that all of the extent
 100  * slots in the efi item have been filled.
 101  */
 102 STATIC void
 103 xfs_efi_item_format(
 104         struct xfs_log_item     *lip,
 105         struct xfs_log_vec      *lv)
 106 {
 107         struct xfs_efi_log_item *efip = EFI_ITEM(lip);
 108         struct xfs_log_iovec    *vecp = NULL;
 109
 110         ASSERT(atomic_read(&efip->efi_next_extent) ==
 111                                 efip->efi_format.efi_nextents);
 112
 113         efip->efi_format.efi_type = XFS_LI_EFI;
 114         efip->efi_format.efi_size = 1;
 115
 116         xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFI_FORMAT,
 117                         &efip->efi_format,
 118                         xfs_efi_item_sizeof(efip));
 119 }
 120
 121
 122 /*
 123  * Pinning has no meaning for an efi item, so just return.
 124  */
 125 STATIC void
 126 xfs_efi_item_pin(
 127         struct xfs_log_item     *lip)
 128 {
 129 }
 130
 131 /*
 132  * While EFIs cannot really be pinned, the unpin operation is the last place at
 133  * which the EFI is manipulated during a transaction.  If we are being asked to
 134  * remove the EFI it's because the transaction has been cancelled and by
 135  * definition that means the EFI cannot be in the AIL so remove it from the
 136  * transaction and free it.  Otherwise coordinate with xfs_efi_release()
 137  * to determine who gets to free the EFI.
 138  */
 139 STATIC void
 140 xfs_efi_item_unpin(
 141         struct xfs_log_item     *lip,
 142         int                     remove)
 143 {
 144         struct xfs_efi_log_item *efip = EFI_ITEM(lip);
 145
 146         if (remove) {
 147                 ASSERT(!(lip->li_flags & XFS_LI_IN_AIL));
 148                 if (lip->li_desc)
 149                         xfs_trans_del_item(lip);
 150                 xfs_efi_item_free(efip);
 151                 return;
 152         }
 153         __xfs_efi_release(efip);
 154 }
 155
 156 /*
 157  * Efi items have no locking or pushing.  However, since EFIs are pulled from
 158  * the AIL when their corresponding EFDs are committed to disk, their situation
 159  * is very similar to being pinned.  Return XFS_ITEM_PINNED so that the caller
 160  * will eventually flush the log.  This should help in getting the EFI out of
 161  * the AIL.
 162  */
 163 STATIC uint
 164 xfs_efi_item_push(
 165         struct xfs_log_item     *lip,
 166         struct list_head        *buffer_list)
 167 {
 168         return XFS_ITEM_PINNED;
 169 }
 170
 171 STATIC void
 172 xfs_efi_item_unlock(
 173         struct xfs_log_item     *lip)
 174 {
 175         if (lip->li_flags & XFS_LI_ABORTED)
 176                 xfs_efi_item_free(EFI_ITEM(lip));
 177 }
 178
 179 /*
 180  * The EFI is logged only once and cannot be moved in the log, so simply return
 181  * the lsn at which it's been logged.
 182  */
 183 STATIC xfs_lsn_t
 184 xfs_efi_item_committed(
 185         struct xfs_log_item     *lip,
 186         xfs_lsn_t               lsn)
 187 {
 188         return lsn;
 189 }
 190
 191 /*
 192  * The EFI dependency tracking op doesn't do squat.  It can't because
 193  * it doesn't know where the free extent is coming from.  The dependency
 194  * tracking has to be handled by the "enclosing" metadata object.  For
 195  * example, for inodes, the inode is locked throughout the extent freeing
 196  * so the dependency should be recorded there.
 197  */
 198 STATIC void
 199 xfs_efi_item_committing(
 200         struct xfs_log_item     *lip,
 201         xfs_lsn_t               lsn)
 202 {
 203 }
 204
 205 /*
 206  * This is the ops vector shared by all efi log items.
 207  */
 208 static const struct xfs_item_ops xfs_efi_item_ops = {
 209         .iop_size       = xfs_efi_item_size,
 210         .iop_format     = xfs_efi_item_format,
 211         .iop_pin        = xfs_efi_item_pin,
 212         .iop_unpin      = xfs_efi_item_unpin,
 213         .iop_unlock     = xfs_efi_item_unlock,
 214         .iop_committed  = xfs_efi_item_committed,
 215         .iop_push       = xfs_efi_item_push,
 216         .iop_committing = xfs_efi_item_committing
 217 };
 218
 219
 220 /*
 221  * Allocate and initialize an efi item with the given number of extents.
 222  */
 223 struct xfs_efi_log_item *
 224 xfs_efi_init(
 225         struct xfs_mount        *mp,
 226         uint                    nextents)
 227
 228 {
 229         struct xfs_efi_log_item *efip;
 230         uint                    size;
 231
 232         ASSERT(nextents > 0);
 233         if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
 234                 size = (uint)(sizeof(xfs_efi_log_item_t) +
 235                         ((nextents - 1) * sizeof(xfs_extent_t)));
 236                 efip = kmem_zalloc(size, KM_SLEEP);
 237         } else {
 238                 efip = kmem_zone_zalloc(xfs_efi_zone, KM_SLEEP);
 239         }
 240
 241         xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
 242         efip->efi_format.efi_nextents = nextents;
 243         efip->efi_format.efi_id = (__psint_t)(void*)efip;
 244         atomic_set(&efip->efi_next_extent, 0);
 245         atomic_set(&efip->efi_refcount, 2);
 246
 247         return efip;
 248 }
 249
 250 /*
 251  * Copy an EFI format buffer from the given buf, and into the destination
 252  * EFI format structure.
 253  * The given buffer can be in 32 bit or 64 bit form (which has different padding),
 254  * one of which will be the native format for this kernel.
 255  * It will handle the conversion of formats if necessary.
 256  */
 257 int
 258 xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
 259 {
 260         xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
 261         uint i;
 262         uint len = sizeof(xfs_efi_log_format_t) +
 263                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
 264         uint len32 = sizeof(xfs_efi_log_format_32_t) +
 265                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
 266         uint len64 = sizeof(xfs_efi_log_format_64_t) +
 267                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
 268
 269         if (buf->i_len == len) {
 270                 memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
 271                 return 0;
 272         } else if (buf->i_len == len32) {
 273                 xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
 274
 275                 dst_efi_fmt->efi_type     = src_efi_fmt_32->efi_type;
 276                 dst_efi_fmt->efi_size     = src_efi_fmt_32->efi_size;
 277                 dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
 278                 dst_efi_fmt->efi_id       = src_efi_fmt_32->efi_id;
 279                 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
 280                         dst_efi_fmt->efi_extents[i].ext_start =
 281                                 src_efi_fmt_32->efi_extents[i].ext_start;
 282                         dst_efi_fmt->efi_extents[i].ext_len =
 283                                 src_efi_fmt_32->efi_extents[i].ext_len;
 284                 }
 285                 return 0;
 286         } else if (buf->i_len == len64) {
 287                 xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
 288
 289                 dst_efi_fmt->efi_type     = src_efi_fmt_64->efi_type;
 290                 dst_efi_fmt->efi_size     = src_efi_fmt_64->efi_size;
 291                 dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
 292                 dst_efi_fmt->efi_id       = src_efi_fmt_64->efi_id;
 293                 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
 294                         dst_efi_fmt->efi_extents[i].ext_start =
 295                                 src_efi_fmt_64->efi_extents[i].ext_start;
 296                         dst_efi_fmt->efi_extents[i].ext_len =
 297                                 src_efi_fmt_64->efi_extents[i].ext_len;
 298                 }
 299                 return 0;
 300         }
 301         return -EFSCORRUPTED;
 302 }
 303
 304 /*
 305  * This is called by the efd item code below to release references to the given
 306  * efi item.  Each efd calls this with the number of extents that it has
 307  * logged, and when the sum of these reaches the total number of extents logged
 308  * by this efi item we can free the efi item.
 309  */
 310 void
 311 xfs_efi_release(xfs_efi_log_item_t      *efip,
 312                 uint                    nextents)
 313 {
 314         ASSERT(atomic_read(&efip->efi_next_extent) >= nextents);
 315         if (atomic_sub_and_test(nextents, &efip->efi_next_extent)) {
 316                 /* recovery needs us to drop the EFI reference, too */
 317                 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags))
 318                         __xfs_efi_release(efip);
 319
 320                 __xfs_efi_release(efip);
 321                 /* efip may now have been freed, do not reference it again. */
 322         }
 323 }
 324
 325 static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
 326 {
 327         return container_of(lip, struct xfs_efd_log_item, efd_item);
 328 }
 329
 330 STATIC void
 331 xfs_efd_item_free(struct xfs_efd_log_item *efdp)
 332 {
 333         if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
 334                 kmem_free(efdp);
 335         else
 336                 kmem_zone_free(xfs_efd_zone, efdp);
 337 }
 338
 339 /*
 340  * This returns the number of iovecs needed to log the given efd item.
 341  * We only need 1 iovec for an efd item.  It just logs the efd_log_format
 342  * structure.
 343  */
 344 static inline int
 345 xfs_efd_item_sizeof(
 346         struct xfs_efd_log_item *efdp)
 347 {
 348         return sizeof(xfs_efd_log_format_t) +
 349                (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
 350 }
 351
 352 STATIC void
 353 xfs_efd_item_size(
 354         struct xfs_log_item     *lip,
 355         int                     *nvecs,
 356         int                     *nbytes)
 357 {
 358         *nvecs += 1;
 359         *nbytes += xfs_efd_item_sizeof(EFD_ITEM(lip));
 360 }
 361
 362 /*
 363  * This is called to fill in the vector of log iovecs for the
 364  * given efd log item. We use only 1 iovec, and we point that
 365  * at the efd_log_format structure embedded in the efd item.
 366  * It is at this point that we assert that all of the extent
 367  * slots in the efd item have been filled.
 368  */
 369 STATIC void
 370 xfs_efd_item_format(
 371         struct xfs_log_item     *lip,
 372         struct xfs_log_vec      *lv)
 373 {
 374         struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
 375         struct xfs_log_iovec    *vecp = NULL;
 376
 377         ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
 378
 379         efdp->efd_format.efd_type = XFS_LI_EFD;
 380         efdp->efd_format.efd_size = 1;
 381
 382         xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFD_FORMAT,
 383                         &efdp->efd_format,
 384                         xfs_efd_item_sizeof(efdp));
 385 }
 386
 387 /*
 388  * Pinning has no meaning for an efd item, so just return.
 389  */
 390 STATIC void
 391 xfs_efd_item_pin(
 392         struct xfs_log_item     *lip)
 393 {
 394 }
 395
 396 /*
 397  * Since pinning has no meaning for an efd item, unpinning does
 398  * not either.
 399  */
 400 STATIC void
 401 xfs_efd_item_unpin(
 402         struct xfs_log_item     *lip,
 403         int                     remove)
 404 {
 405 }
 406
 407 /*
 408  * There isn't much you can do to push on an efd item.  It is simply stuck
 409  * waiting for the log to be flushed to disk.
 410  */
 411 STATIC uint
 412 xfs_efd_item_push(
 413         struct xfs_log_item     *lip,
 414         struct list_head        *buffer_list)
 415 {
 416         return XFS_ITEM_PINNED;
 417 }
 418
 419 STATIC void
 420 xfs_efd_item_unlock(
 421         struct xfs_log_item     *lip)
 422 {
 423         if (lip->li_flags & XFS_LI_ABORTED)
 424                 xfs_efd_item_free(EFD_ITEM(lip));
 425 }
 426
 427 /*
 428  * When the efd item is committed to disk, all we need to do
 429  * is delete our reference to our partner efi item and then
 430  * free ourselves.  Since we're freeing ourselves we must
 431  * return -1 to keep the transaction code from further referencing
 432  * this item.
 433  */
 434 STATIC xfs_lsn_t
 435 xfs_efd_item_committed(
 436         struct xfs_log_item     *lip,
 437         xfs_lsn_t               lsn)
 438 {
 439         struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
 440
 441         /*
 442          * If we got a log I/O error, it's always the case that the LR with the
 443          * EFI got unpinned and freed before the EFD got aborted.
 444          */
 445         if (!(lip->li_flags & XFS_LI_ABORTED))
 446                 xfs_efi_release(efdp->efd_efip, efdp->efd_format.efd_nextents);
 447
 448         xfs_efd_item_free(efdp);
 449         return (xfs_lsn_t)-1;
 450 }
 451
 452 /*
 453  * The EFD dependency tracking op doesn't do squat.  It can't because
 454  * it doesn't know where the free extent is coming from.  The dependency
 455  * tracking has to be handled by the "enclosing" metadata object.  For
 456  * example, for inodes, the inode is locked throughout the extent freeing
 457  * so the dependency should be recorded there.
 458  */
 459 STATIC void
 460 xfs_efd_item_committing(
 461         struct xfs_log_item     *lip,
 462         xfs_lsn_t               lsn)
 463 {
 464 }
 465
 466 /*
 467  * This is the ops vector shared by all efd log items.
 468  */
 469 static const struct xfs_item_ops xfs_efd_item_ops = {
 470         .iop_size       = xfs_efd_item_size,
 471         .iop_format     = xfs_efd_item_format,
 472         .iop_pin        = xfs_efd_item_pin,
 473         .iop_unpin      = xfs_efd_item_unpin,
 474         .iop_unlock     = xfs_efd_item_unlock,
 475         .iop_committed  = xfs_efd_item_committed,
 476         .iop_push       = xfs_efd_item_push,
 477         .iop_committing = xfs_efd_item_committing
 478 };
 479
 480 /*
 481  * Allocate and initialize an efd item with the given number of extents.
 482  */
 483 struct xfs_efd_log_item *
 484 xfs_efd_init(
 485         struct xfs_mount        *mp,
 486         struct xfs_efi_log_item *efip,
 487         uint                    nextents)
 488
 489 {
 490         struct xfs_efd_log_item *efdp;
 491         uint                    size;
 492
 493         ASSERT(nextents > 0);
 494         if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
 495                 size = (uint)(sizeof(xfs_efd_log_item_t) +
 496                         ((nextents - 1) * sizeof(xfs_extent_t)));
 497                 efdp = kmem_zalloc(size, KM_SLEEP);
 498         } else {
 499                 efdp = kmem_zone_zalloc(xfs_efd_zone, KM_SLEEP);
 500         }
 501
 502         xfs_log_item_init(mp, &efdp->efd_item, XFS_LI_EFD, &xfs_efd_item_ops);
 503         efdp->efd_efip = efip;
 504         efdp->efd_format.efd_nextents = nextents;
 505         efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
 506
 507         return efdp;
 508 }