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_format.h"
  21 #include "xfs_log_format.h"
  22 #include "xfs_trans_resv.h"
  23 #include "xfs_mount.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_trans_priv.h"
  26 #include "xfs_buf_item.h"
  27 #include "xfs_extfree_item.h"
  28 #include "xfs_log.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                 /*
  65                  * We don't know whether the EFI made it to the AIL. Remove it
  66                  * if so. Note that xfs_trans_ail_delete() drops the AIL lock.
  67                  */
  68                 if (efip->efi_item.li_flags & XFS_LI_IN_AIL)
  69                         xfs_trans_ail_delete(ailp, &efip->efi_item,
  70                                              SHUTDOWN_LOG_IO_ERROR);
  71                 else
  72                         spin_unlock(&ailp->xa_lock);
  73                 xfs_efi_item_free(efip);
  74         }
  75 }
  76
  77 /*
  78  * This returns the number of iovecs needed to log the given efi item.
  79  * We only need 1 iovec for an efi item.  It just logs the efi_log_format
  80  * structure.
  81  */
  82 static inline int
  83 xfs_efi_item_sizeof(
  84         struct xfs_efi_log_item *efip)
  85 {
  86         return sizeof(struct xfs_efi_log_format) +
  87                (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
  88 }
  89
  90 STATIC void
  91 xfs_efi_item_size(
  92         struct xfs_log_item     *lip,
  93         int                     *nvecs,
  94         int                     *nbytes)
  95 {
  96         *nvecs += 1;
  97         *nbytes += xfs_efi_item_sizeof(EFI_ITEM(lip));
  98 }
  99
 100 /*
 101  * This is called to fill in the vector of log iovecs for the
 102  * given efi log item. We use only 1 iovec, and we point that
 103  * at the efi_log_format structure embedded in the efi item.
 104  * It is at this point that we assert that all of the extent
 105  * slots in the efi item have been filled.
 106  */
 107 STATIC void
 108 xfs_efi_item_format(
 109         struct xfs_log_item     *lip,
 110         struct xfs_log_vec      *lv)
 111 {
 112         struct xfs_efi_log_item *efip = EFI_ITEM(lip);
 113         struct xfs_log_iovec    *vecp = NULL;
 114
 115         ASSERT(atomic_read(&efip->efi_next_extent) ==
 116                                 efip->efi_format.efi_nextents);
 117
 118         efip->efi_format.efi_type = XFS_LI_EFI;
 119         efip->efi_format.efi_size = 1;
 120
 121         xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFI_FORMAT,
 122                         &efip->efi_format,
 123                         xfs_efi_item_sizeof(efip));
 124 }
 125
 126
 127 /*
 128  * Pinning has no meaning for an efi item, so just return.
 129  */
 130 STATIC void
 131 xfs_efi_item_pin(
 132         struct xfs_log_item     *lip)
 133 {
 134 }
 135
 136 /*
 137  * The unpin operation is the last place an EFI is manipulated in the log. It is
 138  * either inserted in the AIL or aborted in the event of a log I/O error. In
 139  * either case, the EFI transaction has been successfully committed to make it
 140  * this far. Therefore, we expect whoever committed the EFI to either construct
 141  * and commit the EFD or drop the EFD's reference in the event of error. Simply
 142  * drop the log's EFI reference now that the log is done with it.
 143  */
 144 STATIC void
 145 xfs_efi_item_unpin(
 146         struct xfs_log_item     *lip,
 147         int                     remove)
 148 {
 149         struct xfs_efi_log_item *efip = EFI_ITEM(lip);
 150         xfs_efi_release(efip);
 151 }
 152
 153 /*
 154  * Efi items have no locking or pushing.  However, since EFIs are pulled from
 155  * the AIL when their corresponding EFDs are committed to disk, their situation
 156  * is very similar to being pinned.  Return XFS_ITEM_PINNED so that the caller
 157  * will eventually flush the log.  This should help in getting the EFI out of
 158  * the AIL.
 159  */
 160 STATIC uint
 161 xfs_efi_item_push(
 162         struct xfs_log_item     *lip,
 163         struct list_head        *buffer_list)
 164 {
 165         return XFS_ITEM_PINNED;
 166 }
 167
 168 /*
 169  * The EFI has been either committed or aborted if the transaction has been
 170  * cancelled. If the transaction was cancelled, an EFD isn't going to be
 171  * constructed and thus we free the EFI here directly.
 172  */
 173 STATIC void
 174 xfs_efi_item_unlock(
 175         struct xfs_log_item     *lip)
 176 {
 177         if (lip->li_flags & XFS_LI_ABORTED)
 178                 xfs_efi_item_free(EFI_ITEM(lip));
 179 }
 180
 181 /*
 182  * The EFI is logged only once and cannot be moved in the log, so simply return
 183  * the lsn at which it's been logged.
 184  */
 185 STATIC xfs_lsn_t
 186 xfs_efi_item_committed(
 187         struct xfs_log_item     *lip,
 188         xfs_lsn_t               lsn)
 189 {
 190         return lsn;
 191 }
 192
 193 /*
 194  * The EFI dependency tracking op doesn't do squat.  It can't because
 195  * it doesn't know where the free extent is coming from.  The dependency
 196  * tracking has to be handled by the "enclosing" metadata object.  For
 197  * example, for inodes, the inode is locked throughout the extent freeing
 198  * so the dependency should be recorded there.
 199  */
 200 STATIC void
 201 xfs_efi_item_committing(
 202         struct xfs_log_item     *lip,
 203         xfs_lsn_t               lsn)
 204 {
 205 }
 206
 207 /*
 208  * This is the ops vector shared by all efi log items.
 209  */
 210 static const struct xfs_item_ops xfs_efi_item_ops = {
 211         .iop_size       = xfs_efi_item_size,
 212         .iop_format     = xfs_efi_item_format,
 213         .iop_pin        = xfs_efi_item_pin,
 214         .iop_unpin      = xfs_efi_item_unpin,
 215         .iop_unlock     = xfs_efi_item_unlock,
 216         .iop_committed  = xfs_efi_item_committed,
 217         .iop_push       = xfs_efi_item_push,
 218         .iop_committing = xfs_efi_item_committing
 219 };
 220
 221
 222 /*
 223  * Allocate and initialize an efi item with the given number of extents.
 224  */
 225 struct xfs_efi_log_item *
 226 xfs_efi_init(
 227         struct xfs_mount        *mp,
 228         uint                    nextents)
 229
 230 {
 231         struct xfs_efi_log_item *efip;
 232         uint                    size;
 233
 234         ASSERT(nextents > 0);
 235         if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
 236                 size = (uint)(sizeof(xfs_efi_log_item_t) +
 237                         ((nextents - 1) * sizeof(xfs_extent_t)));
 238                 efip = kmem_zalloc(size, KM_SLEEP);
 239         } else {
 240                 efip = kmem_zone_zalloc(xfs_efi_zone, KM_SLEEP);
 241         }
 242
 243         xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
 244         efip->efi_format.efi_nextents = nextents;
 245         efip->efi_format.efi_id = (uintptr_t)(void *)efip;
 246         atomic_set(&efip->efi_next_extent, 0);
 247         atomic_set(&efip->efi_refcount, 2);
 248
 249         return efip;
 250 }
 251
 252 /*
 253  * Copy an EFI format buffer from the given buf, and into the destination
 254  * EFI format structure.
 255  * The given buffer can be in 32 bit or 64 bit form (which has different padding),
 256  * one of which will be the native format for this kernel.
 257  * It will handle the conversion of formats if necessary.
 258  */
 259 int
 260 xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
 261 {
 262         xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
 263         uint i;
 264         uint len = sizeof(xfs_efi_log_format_t) +
 265                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
 266         uint len32 = sizeof(xfs_efi_log_format_32_t) +
 267                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
 268         uint len64 = sizeof(xfs_efi_log_format_64_t) +
 269                 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
 270
 271         if (buf->i_len == len) {
 272                 memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
 273                 return 0;
 274         } else if (buf->i_len == len32) {
 275                 xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
 276
 277                 dst_efi_fmt->efi_type     = src_efi_fmt_32->efi_type;
 278                 dst_efi_fmt->efi_size     = src_efi_fmt_32->efi_size;
 279                 dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
 280                 dst_efi_fmt->efi_id       = src_efi_fmt_32->efi_id;
 281                 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
 282                         dst_efi_fmt->efi_extents[i].ext_start =
 283                                 src_efi_fmt_32->efi_extents[i].ext_start;
 284                         dst_efi_fmt->efi_extents[i].ext_len =
 285                                 src_efi_fmt_32->efi_extents[i].ext_len;
 286                 }
 287                 return 0;
 288         } else if (buf->i_len == len64) {
 289                 xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
 290
 291                 dst_efi_fmt->efi_type     = src_efi_fmt_64->efi_type;
 292                 dst_efi_fmt->efi_size     = src_efi_fmt_64->efi_size;
 293                 dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
 294                 dst_efi_fmt->efi_id       = src_efi_fmt_64->efi_id;
 295                 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
 296                         dst_efi_fmt->efi_extents[i].ext_start =
 297                                 src_efi_fmt_64->efi_extents[i].ext_start;
 298                         dst_efi_fmt->efi_extents[i].ext_len =
 299                                 src_efi_fmt_64->efi_extents[i].ext_len;
 300                 }
 301                 return 0;
 302         }
 303         return -EFSCORRUPTED;
 304 }
 305
 306 /*
 307  * This is called by the efd item code below to release references to the given
 308  * efi item.  Each efd calls this with the number of extents that it has
 309  * logged, and when the sum of these reaches the total number of extents logged
 310  * by this efi item we can free the efi item.
 311  */
 312 void
 313 xfs_efi_release(
 314         struct xfs_efi_log_item *efip)
 315 {
 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 static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
 325 {
 326         return container_of(lip, struct xfs_efd_log_item, efd_item);
 327 }
 328
 329 STATIC void
 330 xfs_efd_item_free(struct xfs_efd_log_item *efdp)
 331 {
 332         if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
 333                 kmem_free(efdp);
 334         else
 335                 kmem_zone_free(xfs_efd_zone, efdp);
 336 }
 337
 338 /*
 339  * This returns the number of iovecs needed to log the given efd item.
 340  * We only need 1 iovec for an efd item.  It just logs the efd_log_format
 341  * structure.
 342  */
 343 static inline int
 344 xfs_efd_item_sizeof(
 345         struct xfs_efd_log_item *efdp)
 346 {
 347         return sizeof(xfs_efd_log_format_t) +
 348                (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
 349 }
 350
 351 STATIC void
 352 xfs_efd_item_size(
 353         struct xfs_log_item     *lip,
 354         int                     *nvecs,
 355         int                     *nbytes)
 356 {
 357         *nvecs += 1;
 358         *nbytes += xfs_efd_item_sizeof(EFD_ITEM(lip));
 359 }
 360
 361 /*
 362  * This is called to fill in the vector of log iovecs for the
 363  * given efd log item. We use only 1 iovec, and we point that
 364  * at the efd_log_format structure embedded in the efd item.
 365  * It is at this point that we assert that all of the extent
 366  * slots in the efd item have been filled.
 367  */
 368 STATIC void
 369 xfs_efd_item_format(
 370         struct xfs_log_item     *lip,
 371         struct xfs_log_vec      *lv)
 372 {
 373         struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
 374         struct xfs_log_iovec    *vecp = NULL;
 375
 376         ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
 377
 378         efdp->efd_format.efd_type = XFS_LI_EFD;
 379         efdp->efd_format.efd_size = 1;
 380
 381         xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFD_FORMAT,
 382                         &efdp->efd_format,
 383                         xfs_efd_item_sizeof(efdp));
 384 }
 385
 386 /*
 387  * Pinning has no meaning for an efd item, so just return.
 388  */
 389 STATIC void
 390 xfs_efd_item_pin(
 391         struct xfs_log_item     *lip)
 392 {
 393 }
 394
 395 /*
 396  * Since pinning has no meaning for an efd item, unpinning does
 397  * not either.
 398  */
 399 STATIC void
 400 xfs_efd_item_unpin(
 401         struct xfs_log_item     *lip,
 402         int                     remove)
 403 {
 404 }
 405
 406 /*
 407  * There isn't much you can do to push on an efd item.  It is simply stuck
 408  * waiting for the log to be flushed to disk.
 409  */
 410 STATIC uint
 411 xfs_efd_item_push(
 412         struct xfs_log_item     *lip,
 413         struct list_head        *buffer_list)
 414 {
 415         return XFS_ITEM_PINNED;
 416 }
 417
 418 /*
 419  * The EFD is either committed or aborted if the transaction is cancelled. If
 420  * the transaction is cancelled, drop our reference to the EFI and free the EFD.
 421  */
 422 STATIC void
 423 xfs_efd_item_unlock(
 424         struct xfs_log_item     *lip)
 425 {
 426         struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
 427
 428         if (lip->li_flags & XFS_LI_ABORTED) {
 429                 xfs_efi_release(efdp->efd_efip);
 430                 xfs_efd_item_free(efdp);
 431         }
 432 }
 433
 434 /*
 435  * When the efd item is committed to disk, all we need to do is delete our
 436  * reference to our partner efi item and then free ourselves. Since we're
 437  * freeing ourselves we must return -1 to keep the transaction code from further
 438  * referencing this item.
 439  */
 440 STATIC xfs_lsn_t
 441 xfs_efd_item_committed(
 442         struct xfs_log_item     *lip,
 443         xfs_lsn_t               lsn)
 444 {
 445         struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
 446
 447         /*
 448          * Drop the EFI reference regardless of whether the EFD has been
 449          * aborted. Once the EFD transaction is constructed, it is the sole
 450          * responsibility of the EFD to release the EFI (even if the EFI is
 451          * aborted due to log I/O error).
 452          */
 453         xfs_efi_release(efdp->efd_efip);
 454         xfs_efd_item_free(efdp);
 455
 456         return (xfs_lsn_t)-1;
 457 }
 458
 459 /*
 460  * The EFD dependency tracking op doesn't do squat.  It can't because
 461  * it doesn't know where the free extent is coming from.  The dependency
 462  * tracking has to be handled by the "enclosing" metadata object.  For
 463  * example, for inodes, the inode is locked throughout the extent freeing
 464  * so the dependency should be recorded there.
 465  */
 466 STATIC void
 467 xfs_efd_item_committing(
 468         struct xfs_log_item     *lip,
 469         xfs_lsn_t               lsn)
 470 {
 471 }
 472
 473 /*
 474  * This is the ops vector shared by all efd log items.
 475  */
 476 static const struct xfs_item_ops xfs_efd_item_ops = {
 477         .iop_size       = xfs_efd_item_size,
 478         .iop_format     = xfs_efd_item_format,
 479         .iop_pin        = xfs_efd_item_pin,
 480         .iop_unpin      = xfs_efd_item_unpin,
 481         .iop_unlock     = xfs_efd_item_unlock,
 482         .iop_committed  = xfs_efd_item_committed,
 483         .iop_push       = xfs_efd_item_push,
 484         .iop_committing = xfs_efd_item_committing
 485 };
 486
 487 /*
 488  * Allocate and initialize an efd item with the given number of extents.
 489  */
 490 struct xfs_efd_log_item *
 491 xfs_efd_init(
 492         struct xfs_mount        *mp,
 493         struct xfs_efi_log_item *efip,
 494         uint                    nextents)
 495
 496 {
 497         struct xfs_efd_log_item *efdp;
 498         uint                    size;
 499
 500         ASSERT(nextents > 0);
 501         if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
 502                 size = (uint)(sizeof(xfs_efd_log_item_t) +
 503                         ((nextents - 1) * sizeof(xfs_extent_t)));
 504                 efdp = kmem_zalloc(size, KM_SLEEP);
 505         } else {
 506                 efdp = kmem_zone_zalloc(xfs_efd_zone, KM_SLEEP);
 507         }
 508
 509         xfs_log_item_init(mp, &efdp->efd_item, XFS_LI_EFD, &xfs_efd_item_ops);
 510         efdp->efd_efip = efip;
 511         efdp->efd_format.efd_nextents = nextents;
 512         efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
 513
 514         return efdp;
 515 }