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_types.h"
  21 #include "xfs_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_buf_item.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_trans_priv.h"
  29 #include "xfs_extfree_item.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
  55  * test_and_clear_bit(XFS_EFI_COMMITTED) to ensure only the last caller frees
  56  * the EFI.
  57  */
  58 STATIC void
  59 __xfs_efi_release(
  60         struct xfs_efi_log_item *efip)
  61 {
  62         struct xfs_ail          *ailp = efip->efi_item.li_ailp;
  63
  64         if (!test_and_clear_bit(XFS_EFI_COMMITTED, &efip->efi_flags)) {
  65                 spin_lock(&ailp->xa_lock);
  66                 /* xfs_trans_ail_delete() drops the AIL lock. */
  67                 xfs_trans_ail_delete(ailp, &efip->efi_item);
  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 uint
  78 xfs_efi_item_size(
  79         struct xfs_log_item     *lip)
  80 {
  81         return 1;
  82 }
  83
  84 /*
  85  * This is called to fill in the vector of log iovecs for the
  86  * given efi log item. We use only 1 iovec, and we point that
  87  * at the efi_log_format structure embedded in the efi item.
  88  * It is at this point that we assert that all of the extent
  89  * slots in the efi item have been filled.
  90  */
  91 STATIC void
  92 xfs_efi_item_format(
  93         struct xfs_log_item     *lip,
  94         struct xfs_log_iovec    *log_vector)
  95 {
  96         struct xfs_efi_log_item *efip = EFI_ITEM(lip);
  97         uint                    size;
  98
  99         ASSERT(atomic_read(&efip->efi_next_extent) ==
 100                                 efip->efi_format.efi_nextents);
 101
 102         efip->efi_format.efi_type = XFS_LI_EFI;
 103
 104         size = sizeof(xfs_efi_log_format_t);
 105         size += (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
 106         efip->efi_format.efi_size = 1;
 107
 108         log_vector->i_addr = &efip->efi_format;
 109         log_vector->i_len = size;
 110         log_vector->i_type = XLOG_REG_TYPE_EFI_FORMAT;
 111         ASSERT(size >= sizeof(xfs_efi_log_format_t));
 112 }
 113
 114
 115 /*
 116  * Pinning has no meaning for an efi item, so just return.
 117  */
 118 STATIC void
 119 xfs_efi_item_pin(
 120         struct xfs_log_item     *lip)
 121 {
 122 }
 123
 124 /*
 125  * While EFIs cannot really be pinned, the unpin operation is the last place at
 126  * which the EFI is manipulated during a transaction.  If we are being asked to
 127  * remove the EFI it's because the transaction has been cancelled and by
 128  * definition that means the EFI cannot be in the AIL so remove it from the
 129  * transaction and free it.  Otherwise coordinate with xfs_efi_release() (via
 130  * XFS_EFI_COMMITTED) to determine who gets to free the EFI.
 131  */
 132 STATIC void
 133 xfs_efi_item_unpin(
 134         struct xfs_log_item     *lip,
 135         int                     remove)
 136 {
 137         struct xfs_efi_log_item *efip = EFI_ITEM(lip);
 138
 139         if (remove) {
 140                 ASSERT(!(lip->li_flags & XFS_LI_IN_AIL));
 141                 xfs_trans_del_item(lip);
 142                 xfs_efi_item_free(efip);
 143                 return;
 144         }
 145         __xfs_efi_release(efip);
 146 }
 147
 148 /*
 149  * Efi items have no locking or pushing.  However, since EFIs are
 150  * pulled from the AIL when their corresponding EFDs are committed
 151  * to disk, their situation is very similar to being pinned.  Return
 152  * XFS_ITEM_PINNED so that the caller will eventually flush the log.
 153  * This should help in getting the EFI out of the AIL.
 154  */
 155 STATIC uint
 156 xfs_efi_item_trylock(
 157         struct xfs_log_item     *lip)
 158 {
 159         return XFS_ITEM_PINNED;
 160 }
 161
 162 /*
 163  * Efi items have no locking, so just return.
 164  */
 165 STATIC void
 166 xfs_efi_item_unlock(
 167         struct xfs_log_item     *lip)
 168 {
 169         if (lip->li_flags & XFS_LI_ABORTED)
 170                 xfs_efi_item_free(EFI_ITEM(lip));
 171 }
 172
 173 /*
 174  * The EFI is logged only once and cannot be moved in the log, so simply return
 175  * the lsn at which it's been logged.  For bulk transaction committed
 176  * processing, the EFI may be processed but not yet unpinned prior to the EFD
 177  * being processed. Set the XFS_EFI_COMMITTED flag so this case can be detected
 178  * when processing the EFD.
 179  */
 180 STATIC xfs_lsn_t
 181 xfs_efi_item_committed(
 182         struct xfs_log_item     *lip,
 183         xfs_lsn_t               lsn)
 184 {
 185         struct xfs_efi_log_item *efip = EFI_ITEM(lip);
 186
 187         set_bit(XFS_EFI_COMMITTED, &efip->efi_flags);
 188         return lsn;
 189 }
 190
 191 /*
 192  * There isn't much you can do to push on an efi item.  It is simply
 193  * stuck waiting for all of its corresponding efd items to be
 194  * committed to disk.
 195  */
 196 STATIC void
 197 xfs_efi_item_push(
 198         struct xfs_log_item     *lip)
 199 {
 200 }
 201
 202 /*
 203  * The EFI dependency tracking op doesn't do squat.  It can't because
 204  * it doesn't know where the free extent is coming from.  The dependency
 205  * tracking has to be handled by the "enclosing" metadata object.  For
 206  * example, for inodes, the inode is locked throughout the extent freeing
 207  * so the dependency should be recorded there.
 208  */
 209 STATIC void
 210 xfs_efi_item_committing(
 211         struct xfs_log_item     *lip,
 212         xfs_lsn_t               lsn)
 213 {
 214 }
 215
 216 /*
 217  * This is the ops vector shared by all efi log items.
 218  */
 219 static struct xfs_item_ops xfs_efi_item_ops = {
 220         .iop_size       = xfs_efi_item_size,
 221         .iop_format     = xfs_efi_item_format,
 222         .iop_pin        = xfs_efi_item_pin,
 223         .iop_unpin      = xfs_efi_item_unpin,
 224         .iop_trylock    = xfs_efi_item_trylock,
 225         .iop_unlock     = xfs_efi_item_unlock,
 226         .iop_committed  = xfs_efi_item_committed,
 227         .iop_push       = xfs_efi_item_push,
 228         .iop_committing = xfs_efi_item_committing
 229 };
 230
 231
 232 /*
 233  * Allocate and initialize an efi item with the given number of extents.
 234  */
 235 struct xfs_efi_log_item *
 236 xfs_efi_init(
 237         struct xfs_mount        *mp,
 238         uint                    nextents)
 239
 240 {
 241         struct xfs_efi_log_item *efip;
 242         uint                    size;
 243
 244         ASSERT(nextents > 0);
 245         if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
 246                 size = (uint)(sizeof(xfs_efi_log_item_t) +
 247                         ((nextents - 1) * sizeof(xfs_extent_t)));
 248                 efip = kmem_zalloc(size, KM_SLEEP);
 249         } else {
 250                 efip = kmem_zone_zalloc(xfs_efi_zone, KM_SLEEP);
 251         }
 252
 253         xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
 254         efip->efi_format.efi_nextents = nextents;
 255         efip->efi_format.efi_id = (__psint_t)(void*)efip;
 256         atomic_set(&efip->efi_next_extent, 0);
 257
 258         return efip;
 259 }
 260
 261 /*
 262  * Copy an EFI format buffer from the given buf, and into the destination
 263  * EFI format structure.
 264  * The given buffer can be in 32 bit or 64 bit form (which has different padding),
 265  * one of which will be the native format for this kernel.
 266  * It will handle the conversion of formats if necessary.
 267  */
 268 int
 269 xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
 270 {
 271         xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
 272         uint i;
 273         uint len = sizeof(xfs_efi_log_format_t) +
 274                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
 275         uint len32 = sizeof(xfs_efi_log_format_32_t) +
 276                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
 277         uint len64 = sizeof(xfs_efi_log_format_64_t) +
 278                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
 279
 280         if (buf->i_len == len) {
 281                 memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
 282                 return 0;
 283         } else if (buf->i_len == len32) {
 284                 xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
 285
 286                 dst_efi_fmt->efi_type     = src_efi_fmt_32->efi_type;
 287                 dst_efi_fmt->efi_size     = src_efi_fmt_32->efi_size;
 288                 dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
 289                 dst_efi_fmt->efi_id       = src_efi_fmt_32->efi_id;
 290                 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
 291                         dst_efi_fmt->efi_extents[i].ext_start =
 292                                 src_efi_fmt_32->efi_extents[i].ext_start;
 293                         dst_efi_fmt->efi_extents[i].ext_len =
 294                                 src_efi_fmt_32->efi_extents[i].ext_len;
 295                 }
 296                 return 0;
 297         } else if (buf->i_len == len64) {
 298                 xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
 299
 300                 dst_efi_fmt->efi_type     = src_efi_fmt_64->efi_type;
 301                 dst_efi_fmt->efi_size     = src_efi_fmt_64->efi_size;
 302                 dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
 303                 dst_efi_fmt->efi_id       = src_efi_fmt_64->efi_id;
 304                 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
 305                         dst_efi_fmt->efi_extents[i].ext_start =
 306                                 src_efi_fmt_64->efi_extents[i].ext_start;
 307                         dst_efi_fmt->efi_extents[i].ext_len =
 308                                 src_efi_fmt_64->efi_extents[i].ext_len;
 309                 }
 310                 return 0;
 311         }
 312         return EFSCORRUPTED;
 313 }
 314
 315 /*
 316  * This is called by the efd item code below to release references to the given
 317  * efi item.  Each efd calls this with the number of extents that it has
 318  * logged, and when the sum of these reaches the total number of extents logged
 319  * by this efi item we can free the efi item.
 320  */
 321 void
 322 xfs_efi_release(xfs_efi_log_item_t      *efip,
 323                 uint                    nextents)
 324 {
 325         ASSERT(atomic_read(&efip->efi_next_extent) >= nextents);
 326         if (atomic_sub_and_test(nextents, &efip->efi_next_extent))
 327                 __xfs_efi_release(efip);
 328 }
 329
 330 static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
 331 {
 332         return container_of(lip, struct xfs_efd_log_item, efd_item);
 333 }
 334
 335 STATIC void
 336 xfs_efd_item_free(struct xfs_efd_log_item *efdp)
 337 {
 338         if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
 339                 kmem_free(efdp);
 340         else
 341                 kmem_zone_free(xfs_efd_zone, efdp);
 342 }
 343
 344 /*
 345  * This returns the number of iovecs needed to log the given efd item.
 346  * We only need 1 iovec for an efd item.  It just logs the efd_log_format
 347  * structure.
 348  */
 349 STATIC uint
 350 xfs_efd_item_size(
 351         struct xfs_log_item     *lip)
 352 {
 353         return 1;
 354 }
 355
 356 /*
 357  * This is called to fill in the vector of log iovecs for the
 358  * given efd log item. We use only 1 iovec, and we point that
 359  * at the efd_log_format structure embedded in the efd item.
 360  * It is at this point that we assert that all of the extent
 361  * slots in the efd item have been filled.
 362  */
 363 STATIC void
 364 xfs_efd_item_format(
 365         struct xfs_log_item     *lip,
 366         struct xfs_log_iovec    *log_vector)
 367 {
 368         struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
 369         uint                    size;
 370
 371         ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
 372
 373         efdp->efd_format.efd_type = XFS_LI_EFD;
 374
 375         size = sizeof(xfs_efd_log_format_t);
 376         size += (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
 377         efdp->efd_format.efd_size = 1;
 378
 379         log_vector->i_addr = &efdp->efd_format;
 380         log_vector->i_len = size;
 381         log_vector->i_type = XLOG_REG_TYPE_EFD_FORMAT;
 382         ASSERT(size >= sizeof(xfs_efd_log_format_t));
 383 }
 384
 385 /*
 386  * Pinning has no meaning for an efd item, so just return.
 387  */
 388 STATIC void
 389 xfs_efd_item_pin(
 390         struct xfs_log_item     *lip)
 391 {
 392 }
 393
 394 /*
 395  * Since pinning has no meaning for an efd item, unpinning does
 396  * not either.
 397  */
 398 STATIC void
 399 xfs_efd_item_unpin(
 400         struct xfs_log_item     *lip,
 401         int                     remove)
 402 {
 403 }
 404
 405 /*
 406  * Efd items have no locking, so just return success.
 407  */
 408 STATIC uint
 409 xfs_efd_item_trylock(
 410         struct xfs_log_item     *lip)
 411 {
 412         return XFS_ITEM_LOCKED;
 413 }
 414
 415 /*
 416  * Efd items have no locking or pushing, so return failure
 417  * so that the caller doesn't bother with us.
 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  * There isn't much you can do to push on an efd item.  It is simply
 454  * stuck waiting for the log to be flushed to disk.
 455  */
 456 STATIC void
 457 xfs_efd_item_push(
 458         struct xfs_log_item     *lip)
 459 {
 460 }
 461
 462 /*
 463  * The EFD dependency tracking op doesn't do squat.  It can't because
 464  * it doesn't know where the free extent is coming from.  The dependency
 465  * tracking has to be handled by the "enclosing" metadata object.  For
 466  * example, for inodes, the inode is locked throughout the extent freeing
 467  * so the dependency should be recorded there.
 468  */
 469 STATIC void
 470 xfs_efd_item_committing(
 471         struct xfs_log_item     *lip,
 472         xfs_lsn_t               lsn)
 473 {
 474 }
 475
 476 /*
 477  * This is the ops vector shared by all efd log items.
 478  */
 479 static struct xfs_item_ops xfs_efd_item_ops = {
 480         .iop_size       = xfs_efd_item_size,
 481         .iop_format     = xfs_efd_item_format,
 482         .iop_pin        = xfs_efd_item_pin,
 483         .iop_unpin      = xfs_efd_item_unpin,
 484         .iop_trylock    = xfs_efd_item_trylock,
 485         .iop_unlock     = xfs_efd_item_unlock,
 486         .iop_committed  = xfs_efd_item_committed,
 487         .iop_push       = xfs_efd_item_push,
 488         .iop_committing = xfs_efd_item_committing
 489 };
 490
 491 /*
 492  * Allocate and initialize an efd item with the given number of extents.
 493  */
 494 struct xfs_efd_log_item *
 495 xfs_efd_init(
 496         struct xfs_mount        *mp,
 497         struct xfs_efi_log_item *efip,
 498         uint                    nextents)
 499
 500 {
 501         struct xfs_efd_log_item *efdp;
 502         uint                    size;
 503
 504         ASSERT(nextents > 0);
 505         if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
 506                 size = (uint)(sizeof(xfs_efd_log_item_t) +
 507                         ((nextents - 1) * sizeof(xfs_extent_t)));
 508                 efdp = kmem_zalloc(size, KM_SLEEP);
 509         } else {
 510                 efdp = kmem_zone_zalloc(xfs_efd_zone, KM_SLEEP);
 511         }
 512
 513         xfs_log_item_init(mp, &efdp->efd_item, XFS_LI_EFD, &xfs_efd_item_ops);
 514         efdp->efd_efip = efip;
 515         efdp->efd_format.efd_nextents = nextents;
 516         efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
 517
 518         return efdp;
 519 }