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                                      SHUTDOWN_LOG_IO_ERROR);
  69                 xfs_efi_item_free(efip);
  70         }
  71 }
  72
  73 /*
  74  * This returns the number of iovecs needed to log the given efi item.
  75  * We only need 1 iovec for an efi item.  It just logs the efi_log_format
  76  * structure.
  77  */
  78 STATIC uint
  79 xfs_efi_item_size(
  80         struct xfs_log_item     *lip)
  81 {
  82         return 1;
  83 }
  84
  85 /*
  86  * This is called to fill in the vector of log iovecs for the
  87  * given efi log item. We use only 1 iovec, and we point that
  88  * at the efi_log_format structure embedded in the efi item.
  89  * It is at this point that we assert that all of the extent
  90  * slots in the efi item have been filled.
  91  */
  92 STATIC void
  93 xfs_efi_item_format(
  94         struct xfs_log_item     *lip,
  95         struct xfs_log_iovec    *log_vector)
  96 {
  97         struct xfs_efi_log_item *efip = EFI_ITEM(lip);
  98         uint                    size;
  99
 100         ASSERT(atomic_read(&efip->efi_next_extent) ==
 101                                 efip->efi_format.efi_nextents);
 102
 103         efip->efi_format.efi_type = XFS_LI_EFI;
 104
 105         size = sizeof(xfs_efi_log_format_t);
 106         size += (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
 107         efip->efi_format.efi_size = 1;
 108
 109         log_vector->i_addr = &efip->efi_format;
 110         log_vector->i_len = size;
 111         log_vector->i_type = XLOG_REG_TYPE_EFI_FORMAT;
 112         ASSERT(size >= sizeof(xfs_efi_log_format_t));
 113 }
 114
 115
 116 /*
 117  * Pinning has no meaning for an efi item, so just return.
 118  */
 119 STATIC void
 120 xfs_efi_item_pin(
 121         struct xfs_log_item     *lip)
 122 {
 123 }
 124
 125 /*
 126  * While EFIs cannot really be pinned, the unpin operation is the last place at
 127  * which the EFI is manipulated during a transaction.  If we are being asked to
 128  * remove the EFI it's because the transaction has been cancelled and by
 129  * definition that means the EFI cannot be in the AIL so remove it from the
 130  * transaction and free it.  Otherwise coordinate with xfs_efi_release() (via
 131  * XFS_EFI_COMMITTED) to determine who gets to free the EFI.
 132  */
 133 STATIC void
 134 xfs_efi_item_unpin(
 135         struct xfs_log_item     *lip,
 136         int                     remove)
 137 {
 138         struct xfs_efi_log_item *efip = EFI_ITEM(lip);
 139
 140         if (remove) {
 141                 ASSERT(!(lip->li_flags & XFS_LI_IN_AIL));
 142                 if (lip->li_desc)
 143                         xfs_trans_del_item(lip);
 144                 xfs_efi_item_free(efip);
 145                 return;
 146         }
 147         __xfs_efi_release(efip);
 148 }
 149
 150 /*
 151  * Efi items have no locking or pushing.  However, since EFIs are pulled from
 152  * the AIL when their corresponding EFDs are committed to disk, their situation
 153  * is very similar to being pinned.  Return XFS_ITEM_PINNED so that the caller
 154  * will eventually flush the log.  This should help in getting the EFI out of
 155  * the AIL.
 156  */
 157 STATIC uint
 158 xfs_efi_item_push(
 159         struct xfs_log_item     *lip,
 160         struct list_head        *buffer_list)
 161 {
 162         return XFS_ITEM_PINNED;
 163 }
 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  * 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
 246         return efip;
 247 }
 248
 249 /*
 250  * Copy an EFI format buffer from the given buf, and into the destination
 251  * EFI format structure.
 252  * The given buffer can be in 32 bit or 64 bit form (which has different padding),
 253  * one of which will be the native format for this kernel.
 254  * It will handle the conversion of formats if necessary.
 255  */
 256 int
 257 xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
 258 {
 259         xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
 260         uint i;
 261         uint len = sizeof(xfs_efi_log_format_t) +
 262                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
 263         uint len32 = sizeof(xfs_efi_log_format_32_t) +
 264                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
 265         uint len64 = sizeof(xfs_efi_log_format_64_t) +
 266                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
 267
 268         if (buf->i_len == len) {
 269                 memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
 270                 return 0;
 271         } else if (buf->i_len == len32) {
 272                 xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
 273
 274                 dst_efi_fmt->efi_type     = src_efi_fmt_32->efi_type;
 275                 dst_efi_fmt->efi_size     = src_efi_fmt_32->efi_size;
 276                 dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
 277                 dst_efi_fmt->efi_id       = src_efi_fmt_32->efi_id;
 278                 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
 279                         dst_efi_fmt->efi_extents[i].ext_start =
 280                                 src_efi_fmt_32->efi_extents[i].ext_start;
 281                         dst_efi_fmt->efi_extents[i].ext_len =
 282                                 src_efi_fmt_32->efi_extents[i].ext_len;
 283                 }
 284                 return 0;
 285         } else if (buf->i_len == len64) {
 286                 xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
 287
 288                 dst_efi_fmt->efi_type     = src_efi_fmt_64->efi_type;
 289                 dst_efi_fmt->efi_size     = src_efi_fmt_64->efi_size;
 290                 dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
 291                 dst_efi_fmt->efi_id       = src_efi_fmt_64->efi_id;
 292                 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
 293                         dst_efi_fmt->efi_extents[i].ext_start =
 294                                 src_efi_fmt_64->efi_extents[i].ext_start;
 295                         dst_efi_fmt->efi_extents[i].ext_len =
 296                                 src_efi_fmt_64->efi_extents[i].ext_len;
 297                 }
 298                 return 0;
 299         }
 300         return EFSCORRUPTED;
 301 }
 302
 303 /*
 304  * This is called by the efd item code below to release references to the given
 305  * efi item.  Each efd calls this with the number of extents that it has
 306  * logged, and when the sum of these reaches the total number of extents logged
 307  * by this efi item we can free the efi item.
 308  */
 309 void
 310 xfs_efi_release(xfs_efi_log_item_t      *efip,
 311                 uint                    nextents)
 312 {
 313         ASSERT(atomic_read(&efip->efi_next_extent) >= nextents);
 314         if (atomic_sub_and_test(nextents, &efip->efi_next_extent))
 315                 __xfs_efi_release(efip);
 316 }
 317
 318 static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
 319 {
 320         return container_of(lip, struct xfs_efd_log_item, efd_item);
 321 }
 322
 323 STATIC void
 324 xfs_efd_item_free(struct xfs_efd_log_item *efdp)
 325 {
 326         if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
 327                 kmem_free(efdp);
 328         else
 329                 kmem_zone_free(xfs_efd_zone, efdp);
 330 }
 331
 332 /*
 333  * This returns the number of iovecs needed to log the given efd item.
 334  * We only need 1 iovec for an efd item.  It just logs the efd_log_format
 335  * structure.
 336  */
 337 STATIC uint
 338 xfs_efd_item_size(
 339         struct xfs_log_item     *lip)
 340 {
 341         return 1;
 342 }
 343
 344 /*
 345  * This is called to fill in the vector of log iovecs for the
 346  * given efd log item. We use only 1 iovec, and we point that
 347  * at the efd_log_format structure embedded in the efd item.
 348  * It is at this point that we assert that all of the extent
 349  * slots in the efd item have been filled.
 350  */
 351 STATIC void
 352 xfs_efd_item_format(
 353         struct xfs_log_item     *lip,
 354         struct xfs_log_iovec    *log_vector)
 355 {
 356         struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
 357         uint                    size;
 358
 359         ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
 360
 361         efdp->efd_format.efd_type = XFS_LI_EFD;
 362
 363         size = sizeof(xfs_efd_log_format_t);
 364         size += (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
 365         efdp->efd_format.efd_size = 1;
 366
 367         log_vector->i_addr = &efdp->efd_format;
 368         log_vector->i_len = size;
 369         log_vector->i_type = XLOG_REG_TYPE_EFD_FORMAT;
 370         ASSERT(size >= sizeof(xfs_efd_log_format_t));
 371 }
 372
 373 /*
 374  * Pinning has no meaning for an efd item, so just return.
 375  */
 376 STATIC void
 377 xfs_efd_item_pin(
 378         struct xfs_log_item     *lip)
 379 {
 380 }
 381
 382 /*
 383  * Since pinning has no meaning for an efd item, unpinning does
 384  * not either.
 385  */
 386 STATIC void
 387 xfs_efd_item_unpin(
 388         struct xfs_log_item     *lip,
 389         int                     remove)
 390 {
 391 }
 392
 393 /*
 394  * There isn't much you can do to push on an efd item.  It is simply stuck
 395  * waiting for the log to be flushed to disk.
 396  */
 397 STATIC uint
 398 xfs_efd_item_push(
 399         struct xfs_log_item     *lip,
 400         struct list_head        *buffer_list)
 401 {
 402         return XFS_ITEM_PINNED;
 403 }
 404
 405 STATIC void
 406 xfs_efd_item_unlock(
 407         struct xfs_log_item     *lip)
 408 {
 409         if (lip->li_flags & XFS_LI_ABORTED)
 410                 xfs_efd_item_free(EFD_ITEM(lip));
 411 }
 412
 413 /*
 414  * When the efd item is committed to disk, all we need to do
 415  * is delete our reference to our partner efi item and then
 416  * free ourselves.  Since we're freeing ourselves we must
 417  * return -1 to keep the transaction code from further referencing
 418  * this item.
 419  */
 420 STATIC xfs_lsn_t
 421 xfs_efd_item_committed(
 422         struct xfs_log_item     *lip,
 423         xfs_lsn_t               lsn)
 424 {
 425         struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
 426
 427         /*
 428          * If we got a log I/O error, it's always the case that the LR with the
 429          * EFI got unpinned and freed before the EFD got aborted.
 430          */
 431         if (!(lip->li_flags & XFS_LI_ABORTED))
 432                 xfs_efi_release(efdp->efd_efip, efdp->efd_format.efd_nextents);
 433
 434         xfs_efd_item_free(efdp);
 435         return (xfs_lsn_t)-1;
 436 }
 437
 438 /*
 439  * The EFD dependency tracking op doesn't do squat.  It can't because
 440  * it doesn't know where the free extent is coming from.  The dependency
 441  * tracking has to be handled by the "enclosing" metadata object.  For
 442  * example, for inodes, the inode is locked throughout the extent freeing
 443  * so the dependency should be recorded there.
 444  */
 445 STATIC void
 446 xfs_efd_item_committing(
 447         struct xfs_log_item     *lip,
 448         xfs_lsn_t               lsn)
 449 {
 450 }
 451
 452 /*
 453  * This is the ops vector shared by all efd log items.
 454  */
 455 static const struct xfs_item_ops xfs_efd_item_ops = {
 456         .iop_size       = xfs_efd_item_size,
 457         .iop_format     = xfs_efd_item_format,
 458         .iop_pin        = xfs_efd_item_pin,
 459         .iop_unpin      = xfs_efd_item_unpin,
 460         .iop_unlock     = xfs_efd_item_unlock,
 461         .iop_committed  = xfs_efd_item_committed,
 462         .iop_push       = xfs_efd_item_push,
 463         .iop_committing = xfs_efd_item_committing
 464 };
 465
 466 /*
 467  * Allocate and initialize an efd item with the given number of extents.
 468  */
 469 struct xfs_efd_log_item *
 470 xfs_efd_init(
 471         struct xfs_mount        *mp,
 472         struct xfs_efi_log_item *efip,
 473         uint                    nextents)
 474
 475 {
 476         struct xfs_efd_log_item *efdp;
 477         uint                    size;
 478
 479         ASSERT(nextents > 0);
 480         if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
 481                 size = (uint)(sizeof(xfs_efd_log_item_t) +
 482                         ((nextents - 1) * sizeof(xfs_extent_t)));
 483                 efdp = kmem_zalloc(size, KM_SLEEP);
 484         } else {
 485                 efdp = kmem_zone_zalloc(xfs_efd_zone, KM_SLEEP);
 486         }
 487
 488         xfs_log_item_init(mp, &efdp->efd_item, XFS_LI_EFD, &xfs_efd_item_ops);
 489         efdp->efd_efip = efip;
 490         efdp->efd_format.efd_nextents = nextents;
 491         efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
 492
 493         return efdp;
 494 }