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