linux/fs/xfs/xfs_trans.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3 * Copyright (C) 2010 Red Hat, Inc.
   4 * All Rights Reserved.
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License as
   8 * published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it would be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write the Free Software Foundation,
  17 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18 */
  19#include "xfs.h"
  20#include "xfs_fs.h"
  21#include "xfs_types.h"
  22#include "xfs_log.h"
  23#include "xfs_trans.h"
  24#include "xfs_sb.h"
  25#include "xfs_ag.h"
  26#include "xfs_mount.h"
  27#include "xfs_error.h"
  28#include "xfs_da_btree.h"
  29#include "xfs_bmap_btree.h"
  30#include "xfs_alloc_btree.h"
  31#include "xfs_ialloc_btree.h"
  32#include "xfs_dinode.h"
  33#include "xfs_inode.h"
  34#include "xfs_btree.h"
  35#include "xfs_ialloc.h"
  36#include "xfs_alloc.h"
  37#include "xfs_extent_busy.h"
  38#include "xfs_bmap.h"
  39#include "xfs_quota.h"
  40#include "xfs_trans_priv.h"
  41#include "xfs_trans_space.h"
  42#include "xfs_inode_item.h"
  43#include "xfs_trace.h"
  44
  45kmem_zone_t     *xfs_trans_zone;
  46kmem_zone_t     *xfs_log_item_desc_zone;
  47
  48
  49/*
  50 * Various log reservation values.
  51 *
  52 * These are based on the size of the file system block because that is what
  53 * most transactions manipulate.  Each adds in an additional 128 bytes per
  54 * item logged to try to account for the overhead of the transaction mechanism.
  55 *
  56 * Note:  Most of the reservations underestimate the number of allocation
  57 * groups into which they could free extents in the xfs_bmap_finish() call.
  58 * This is because the number in the worst case is quite high and quite
  59 * unusual.  In order to fix this we need to change xfs_bmap_finish() to free
  60 * extents in only a single AG at a time.  This will require changes to the
  61 * EFI code as well, however, so that the EFI for the extents not freed is
  62 * logged again in each transaction.  See SGI PV #261917.
  63 *
  64 * Reservation functions here avoid a huge stack in xfs_trans_init due to
  65 * register overflow from temporaries in the calculations.
  66 */
  67
  68
  69/*
  70 * In a write transaction we can allocate a maximum of 2
  71 * extents.  This gives:
  72 *    the inode getting the new extents: inode size
  73 *    the inode's bmap btree: max depth * block size
  74 *    the agfs of the ags from which the extents are allocated: 2 * sector
  75 *    the superblock free block counter: sector size
  76 *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
  77 * And the bmap_finish transaction can free bmap blocks in a join:
  78 *    the agfs of the ags containing the blocks: 2 * sector size
  79 *    the agfls of the ags containing the blocks: 2 * sector size
  80 *    the super block free block counter: sector size
  81 *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
  82 */
  83STATIC uint
  84xfs_calc_write_reservation(
  85        struct xfs_mount        *mp)
  86{
  87        return XFS_DQUOT_LOGRES(mp) +
  88                MAX((mp->m_sb.sb_inodesize +
  89                     XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK)) +
  90                     2 * mp->m_sb.sb_sectsize +
  91                     mp->m_sb.sb_sectsize +
  92                     XFS_ALLOCFREE_LOG_RES(mp, 2) +
  93                     128 * (4 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) +
  94                            XFS_ALLOCFREE_LOG_COUNT(mp, 2))),
  95                    (2 * mp->m_sb.sb_sectsize +
  96                     2 * mp->m_sb.sb_sectsize +
  97                     mp->m_sb.sb_sectsize +
  98                     XFS_ALLOCFREE_LOG_RES(mp, 2) +
  99                     128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2))));
 100}
 101
 102/*
 103 * In truncating a file we free up to two extents at once.  We can modify:
 104 *    the inode being truncated: inode size
 105 *    the inode's bmap btree: (max depth + 1) * block size
 106 * And the bmap_finish transaction can free the blocks and bmap blocks:
 107 *    the agf for each of the ags: 4 * sector size
 108 *    the agfl for each of the ags: 4 * sector size
 109 *    the super block to reflect the freed blocks: sector size
 110 *    worst case split in allocation btrees per extent assuming 4 extents:
 111 *              4 exts * 2 trees * (2 * max depth - 1) * block size
 112 *    the inode btree: max depth * blocksize
 113 *    the allocation btrees: 2 trees * (max depth - 1) * block size
 114 */
 115STATIC uint
 116xfs_calc_itruncate_reservation(
 117        struct xfs_mount        *mp)
 118{
 119        return XFS_DQUOT_LOGRES(mp) +
 120                MAX((mp->m_sb.sb_inodesize +
 121                     XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) + 1) +
 122                     128 * (2 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))),
 123                    (4 * mp->m_sb.sb_sectsize +
 124                     4 * mp->m_sb.sb_sectsize +
 125                     mp->m_sb.sb_sectsize +
 126                     XFS_ALLOCFREE_LOG_RES(mp, 4) +
 127                     128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4)) +
 128                     128 * 5 +
 129                     XFS_ALLOCFREE_LOG_RES(mp, 1) +
 130                     128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
 131                            XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
 132}
 133
 134/*
 135 * In renaming a files we can modify:
 136 *    the four inodes involved: 4 * inode size
 137 *    the two directory btrees: 2 * (max depth + v2) * dir block size
 138 *    the two directory bmap btrees: 2 * max depth * block size
 139 * And the bmap_finish transaction can free dir and bmap blocks (two sets
 140 *      of bmap blocks) giving:
 141 *    the agf for the ags in which the blocks live: 3 * sector size
 142 *    the agfl for the ags in which the blocks live: 3 * sector size
 143 *    the superblock for the free block count: sector size
 144 *    the allocation btrees: 3 exts * 2 trees * (2 * max depth - 1) * block size
 145 */
 146STATIC uint
 147xfs_calc_rename_reservation(
 148        struct xfs_mount        *mp)
 149{
 150        return XFS_DQUOT_LOGRES(mp) +
 151                MAX((4 * mp->m_sb.sb_inodesize +
 152                     2 * XFS_DIROP_LOG_RES(mp) +
 153                     128 * (4 + 2 * XFS_DIROP_LOG_COUNT(mp))),
 154                    (3 * mp->m_sb.sb_sectsize +
 155                     3 * mp->m_sb.sb_sectsize +
 156                     mp->m_sb.sb_sectsize +
 157                     XFS_ALLOCFREE_LOG_RES(mp, 3) +
 158                     128 * (7 + XFS_ALLOCFREE_LOG_COUNT(mp, 3))));
 159}
 160
 161/*
 162 * For creating a link to an inode:
 163 *    the parent directory inode: inode size
 164 *    the linked inode: inode size
 165 *    the directory btree could split: (max depth + v2) * dir block size
 166 *    the directory bmap btree could join or split: (max depth + v2) * blocksize
 167 * And the bmap_finish transaction can free some bmap blocks giving:
 168 *    the agf for the ag in which the blocks live: sector size
 169 *    the agfl for the ag in which the blocks live: sector size
 170 *    the superblock for the free block count: sector size
 171 *    the allocation btrees: 2 trees * (2 * max depth - 1) * block size
 172 */
 173STATIC uint
 174xfs_calc_link_reservation(
 175        struct xfs_mount        *mp)
 176{
 177        return XFS_DQUOT_LOGRES(mp) +
 178                MAX((mp->m_sb.sb_inodesize +
 179                     mp->m_sb.sb_inodesize +
 180                     XFS_DIROP_LOG_RES(mp) +
 181                     128 * (2 + XFS_DIROP_LOG_COUNT(mp))),
 182                    (mp->m_sb.sb_sectsize +
 183                     mp->m_sb.sb_sectsize +
 184                     mp->m_sb.sb_sectsize +
 185                     XFS_ALLOCFREE_LOG_RES(mp, 1) +
 186                     128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
 187}
 188
 189/*
 190 * For removing a directory entry we can modify:
 191 *    the parent directory inode: inode size
 192 *    the removed inode: inode size
 193 *    the directory btree could join: (max depth + v2) * dir block size
 194 *    the directory bmap btree could join or split: (max depth + v2) * blocksize
 195 * And the bmap_finish transaction can free the dir and bmap blocks giving:
 196 *    the agf for the ag in which the blocks live: 2 * sector size
 197 *    the agfl for the ag in which the blocks live: 2 * sector size
 198 *    the superblock for the free block count: sector size
 199 *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
 200 */
 201STATIC uint
 202xfs_calc_remove_reservation(
 203        struct xfs_mount        *mp)
 204{
 205        return XFS_DQUOT_LOGRES(mp) +
 206                MAX((mp->m_sb.sb_inodesize +
 207                     mp->m_sb.sb_inodesize +
 208                     XFS_DIROP_LOG_RES(mp) +
 209                     128 * (2 + XFS_DIROP_LOG_COUNT(mp))),
 210                    (2 * mp->m_sb.sb_sectsize +
 211                     2 * mp->m_sb.sb_sectsize +
 212                     mp->m_sb.sb_sectsize +
 213                     XFS_ALLOCFREE_LOG_RES(mp, 2) +
 214                     128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2))));
 215}
 216
 217/*
 218 * For symlink we can modify:
 219 *    the parent directory inode: inode size
 220 *    the new inode: inode size
 221 *    the inode btree entry: 1 block
 222 *    the directory btree: (max depth + v2) * dir block size
 223 *    the directory inode's bmap btree: (max depth + v2) * block size
 224 *    the blocks for the symlink: 1 kB
 225 * Or in the first xact we allocate some inodes giving:
 226 *    the agi and agf of the ag getting the new inodes: 2 * sectorsize
 227 *    the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize
 228 *    the inode btree: max depth * blocksize
 229 *    the allocation btrees: 2 trees * (2 * max depth - 1) * block size
 230 */
 231STATIC uint
 232xfs_calc_symlink_reservation(
 233        struct xfs_mount        *mp)
 234{
 235        return XFS_DQUOT_LOGRES(mp) +
 236                MAX((mp->m_sb.sb_inodesize +
 237                     mp->m_sb.sb_inodesize +
 238                     XFS_FSB_TO_B(mp, 1) +
 239                     XFS_DIROP_LOG_RES(mp) +
 240                     1024 +
 241                     128 * (4 + XFS_DIROP_LOG_COUNT(mp))),
 242                    (2 * mp->m_sb.sb_sectsize +
 243                     XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) +
 244                     XFS_FSB_TO_B(mp, mp->m_in_maxlevels) +
 245                     XFS_ALLOCFREE_LOG_RES(mp, 1) +
 246                     128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
 247                            XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
 248}
 249
 250/*
 251 * For create we can modify:
 252 *    the parent directory inode: inode size
 253 *    the new inode: inode size
 254 *    the inode btree entry: block size
 255 *    the superblock for the nlink flag: sector size
 256 *    the directory btree: (max depth + v2) * dir block size
 257 *    the directory inode's bmap btree: (max depth + v2) * block size
 258 * Or in the first xact we allocate some inodes giving:
 259 *    the agi and agf of the ag getting the new inodes: 2 * sectorsize
 260 *    the superblock for the nlink flag: sector size
 261 *    the inode blocks allocated: XFS_IALLOC_BLOCKS * blocksize
 262 *    the inode btree: max depth * blocksize
 263 *    the allocation btrees: 2 trees * (max depth - 1) * block size
 264 */
 265STATIC uint
 266xfs_calc_create_reservation(
 267        struct xfs_mount        *mp)
 268{
 269        return XFS_DQUOT_LOGRES(mp) +
 270                MAX((mp->m_sb.sb_inodesize +
 271                     mp->m_sb.sb_inodesize +
 272                     mp->m_sb.sb_sectsize +
 273                     XFS_FSB_TO_B(mp, 1) +
 274                     XFS_DIROP_LOG_RES(mp) +
 275                     128 * (3 + XFS_DIROP_LOG_COUNT(mp))),
 276                    (3 * mp->m_sb.sb_sectsize +
 277                     XFS_FSB_TO_B(mp, XFS_IALLOC_BLOCKS(mp)) +
 278                     XFS_FSB_TO_B(mp, mp->m_in_maxlevels) +
 279                     XFS_ALLOCFREE_LOG_RES(mp, 1) +
 280                     128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
 281                            XFS_ALLOCFREE_LOG_COUNT(mp, 1))));
 282}
 283
 284/*
 285 * Making a new directory is the same as creating a new file.
 286 */
 287STATIC uint
 288xfs_calc_mkdir_reservation(
 289        struct xfs_mount        *mp)
 290{
 291        return xfs_calc_create_reservation(mp);
 292}
 293
 294/*
 295 * In freeing an inode we can modify:
 296 *    the inode being freed: inode size
 297 *    the super block free inode counter: sector size
 298 *    the agi hash list and counters: sector size
 299 *    the inode btree entry: block size
 300 *    the on disk inode before ours in the agi hash list: inode cluster size
 301 *    the inode btree: max depth * blocksize
 302 *    the allocation btrees: 2 trees * (max depth - 1) * block size
 303 */
 304STATIC uint
 305xfs_calc_ifree_reservation(
 306        struct xfs_mount        *mp)
 307{
 308        return XFS_DQUOT_LOGRES(mp) +
 309                mp->m_sb.sb_inodesize +
 310                mp->m_sb.sb_sectsize +
 311                mp->m_sb.sb_sectsize +
 312                XFS_FSB_TO_B(mp, 1) +
 313                MAX((__uint16_t)XFS_FSB_TO_B(mp, 1),
 314                    XFS_INODE_CLUSTER_SIZE(mp)) +
 315                128 * 5 +
 316                XFS_ALLOCFREE_LOG_RES(mp, 1) +
 317                128 * (2 + XFS_IALLOC_BLOCKS(mp) + mp->m_in_maxlevels +
 318                       XFS_ALLOCFREE_LOG_COUNT(mp, 1));
 319}
 320
 321/*
 322 * When only changing the inode we log the inode and possibly the superblock
 323 * We also add a bit of slop for the transaction stuff.
 324 */
 325STATIC uint
 326xfs_calc_ichange_reservation(
 327        struct xfs_mount        *mp)
 328{
 329        return XFS_DQUOT_LOGRES(mp) +
 330                mp->m_sb.sb_inodesize +
 331                mp->m_sb.sb_sectsize +
 332                512;
 333
 334}
 335
 336/*
 337 * Growing the data section of the filesystem.
 338 *      superblock
 339 *      agi and agf
 340 *      allocation btrees
 341 */
 342STATIC uint
 343xfs_calc_growdata_reservation(
 344        struct xfs_mount        *mp)
 345{
 346        return mp->m_sb.sb_sectsize * 3 +
 347                XFS_ALLOCFREE_LOG_RES(mp, 1) +
 348                128 * (3 + XFS_ALLOCFREE_LOG_COUNT(mp, 1));
 349}
 350
 351/*
 352 * Growing the rt section of the filesystem.
 353 * In the first set of transactions (ALLOC) we allocate space to the
 354 * bitmap or summary files.
 355 *      superblock: sector size
 356 *      agf of the ag from which the extent is allocated: sector size
 357 *      bmap btree for bitmap/summary inode: max depth * blocksize
 358 *      bitmap/summary inode: inode size
 359 *      allocation btrees for 1 block alloc: 2 * (2 * maxdepth - 1) * blocksize
 360 */
 361STATIC uint
 362xfs_calc_growrtalloc_reservation(
 363        struct xfs_mount        *mp)
 364{
 365        return 2 * mp->m_sb.sb_sectsize +
 366                XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK)) +
 367                mp->m_sb.sb_inodesize +
 368                XFS_ALLOCFREE_LOG_RES(mp, 1) +
 369                128 * (3 + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) +
 370                       XFS_ALLOCFREE_LOG_COUNT(mp, 1));
 371}
 372
 373/*
 374 * Growing the rt section of the filesystem.
 375 * In the second set of transactions (ZERO) we zero the new metadata blocks.
 376 *      one bitmap/summary block: blocksize
 377 */
 378STATIC uint
 379xfs_calc_growrtzero_reservation(
 380        struct xfs_mount        *mp)
 381{
 382        return mp->m_sb.sb_blocksize + 128;
 383}
 384
 385/*
 386 * Growing the rt section of the filesystem.
 387 * In the third set of transactions (FREE) we update metadata without
 388 * allocating any new blocks.
 389 *      superblock: sector size
 390 *      bitmap inode: inode size
 391 *      summary inode: inode size
 392 *      one bitmap block: blocksize
 393 *      summary blocks: new summary size
 394 */
 395STATIC uint
 396xfs_calc_growrtfree_reservation(
 397        struct xfs_mount        *mp)
 398{
 399        return mp->m_sb.sb_sectsize +
 400                2 * mp->m_sb.sb_inodesize +
 401                mp->m_sb.sb_blocksize +
 402                mp->m_rsumsize +
 403                128 * 5;
 404}
 405
 406/*
 407 * Logging the inode modification timestamp on a synchronous write.
 408 *      inode
 409 */
 410STATIC uint
 411xfs_calc_swrite_reservation(
 412        struct xfs_mount        *mp)
 413{
 414        return mp->m_sb.sb_inodesize + 128;
 415}
 416
 417/*
 418 * Logging the inode mode bits when writing a setuid/setgid file
 419 *      inode
 420 */
 421STATIC uint
 422xfs_calc_writeid_reservation(xfs_mount_t *mp)
 423{
 424        return mp->m_sb.sb_inodesize + 128;
 425}
 426
 427/*
 428 * Converting the inode from non-attributed to attributed.
 429 *      the inode being converted: inode size
 430 *      agf block and superblock (for block allocation)
 431 *      the new block (directory sized)
 432 *      bmap blocks for the new directory block
 433 *      allocation btrees
 434 */
 435STATIC uint
 436xfs_calc_addafork_reservation(
 437        struct xfs_mount        *mp)
 438{
 439        return XFS_DQUOT_LOGRES(mp) +
 440                mp->m_sb.sb_inodesize +
 441                mp->m_sb.sb_sectsize * 2 +
 442                mp->m_dirblksize +
 443                XFS_FSB_TO_B(mp, XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1) +
 444                XFS_ALLOCFREE_LOG_RES(mp, 1) +
 445                128 * (4 + XFS_DAENTER_BMAP1B(mp, XFS_DATA_FORK) + 1 +
 446                       XFS_ALLOCFREE_LOG_COUNT(mp, 1));
 447}
 448
 449/*
 450 * Removing the attribute fork of a file
 451 *    the inode being truncated: inode size
 452 *    the inode's bmap btree: max depth * block size
 453 * And the bmap_finish transaction can free the blocks and bmap blocks:
 454 *    the agf for each of the ags: 4 * sector size
 455 *    the agfl for each of the ags: 4 * sector size
 456 *    the super block to reflect the freed blocks: sector size
 457 *    worst case split in allocation btrees per extent assuming 4 extents:
 458 *              4 exts * 2 trees * (2 * max depth - 1) * block size
 459 */
 460STATIC uint
 461xfs_calc_attrinval_reservation(
 462        struct xfs_mount        *mp)
 463{
 464        return MAX((mp->m_sb.sb_inodesize +
 465                    XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
 466                    128 * (1 + XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK))),
 467                   (4 * mp->m_sb.sb_sectsize +
 468                    4 * mp->m_sb.sb_sectsize +
 469                    mp->m_sb.sb_sectsize +
 470                    XFS_ALLOCFREE_LOG_RES(mp, 4) +
 471                    128 * (9 + XFS_ALLOCFREE_LOG_COUNT(mp, 4))));
 472}
 473
 474/*
 475 * Setting an attribute.
 476 *      the inode getting the attribute
 477 *      the superblock for allocations
 478 *      the agfs extents are allocated from
 479 *      the attribute btree * max depth
 480 *      the inode allocation btree
 481 * Since attribute transaction space is dependent on the size of the attribute,
 482 * the calculation is done partially at mount time and partially at runtime.
 483 */
 484STATIC uint
 485xfs_calc_attrset_reservation(
 486        struct xfs_mount        *mp)
 487{
 488        return XFS_DQUOT_LOGRES(mp) +
 489                mp->m_sb.sb_inodesize +
 490                mp->m_sb.sb_sectsize +
 491                XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) +
 492                128 * (2 + XFS_DA_NODE_MAXDEPTH);
 493}
 494
 495/*
 496 * Removing an attribute.
 497 *    the inode: inode size
 498 *    the attribute btree could join: max depth * block size
 499 *    the inode bmap btree could join or split: max depth * block size
 500 * And the bmap_finish transaction can free the attr blocks freed giving:
 501 *    the agf for the ag in which the blocks live: 2 * sector size
 502 *    the agfl for the ag in which the blocks live: 2 * sector size
 503 *    the superblock for the free block count: sector size
 504 *    the allocation btrees: 2 exts * 2 trees * (2 * max depth - 1) * block size
 505 */
 506STATIC uint
 507xfs_calc_attrrm_reservation(
 508        struct xfs_mount        *mp)
 509{
 510        return XFS_DQUOT_LOGRES(mp) +
 511                MAX((mp->m_sb.sb_inodesize +
 512                     XFS_FSB_TO_B(mp, XFS_DA_NODE_MAXDEPTH) +
 513                     XFS_FSB_TO_B(mp, XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)) +
 514                     128 * (1 + XFS_DA_NODE_MAXDEPTH +
 515                            XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK))),
 516                    (2 * mp->m_sb.sb_sectsize +
 517                     2 * mp->m_sb.sb_sectsize +
 518                     mp->m_sb.sb_sectsize +
 519                     XFS_ALLOCFREE_LOG_RES(mp, 2) +
 520                     128 * (5 + XFS_ALLOCFREE_LOG_COUNT(mp, 2))));
 521}
 522
 523/*
 524 * Clearing a bad agino number in an agi hash bucket.
 525 */
 526STATIC uint
 527xfs_calc_clear_agi_bucket_reservation(
 528        struct xfs_mount        *mp)
 529{
 530        return mp->m_sb.sb_sectsize + 128;
 531}
 532
 533/*
 534 * Initialize the precomputed transaction reservation values
 535 * in the mount structure.
 536 */
 537void
 538xfs_trans_init(
 539        struct xfs_mount        *mp)
 540{
 541        struct xfs_trans_reservations *resp = &mp->m_reservations;
 542
 543        resp->tr_write = xfs_calc_write_reservation(mp);
 544        resp->tr_itruncate = xfs_calc_itruncate_reservation(mp);
 545        resp->tr_rename = xfs_calc_rename_reservation(mp);
 546        resp->tr_link = xfs_calc_link_reservation(mp);
 547        resp->tr_remove = xfs_calc_remove_reservation(mp);
 548        resp->tr_symlink = xfs_calc_symlink_reservation(mp);
 549        resp->tr_create = xfs_calc_create_reservation(mp);
 550        resp->tr_mkdir = xfs_calc_mkdir_reservation(mp);
 551        resp->tr_ifree = xfs_calc_ifree_reservation(mp);
 552        resp->tr_ichange = xfs_calc_ichange_reservation(mp);
 553        resp->tr_growdata = xfs_calc_growdata_reservation(mp);
 554        resp->tr_swrite = xfs_calc_swrite_reservation(mp);
 555        resp->tr_writeid = xfs_calc_writeid_reservation(mp);
 556        resp->tr_addafork = xfs_calc_addafork_reservation(mp);
 557        resp->tr_attrinval = xfs_calc_attrinval_reservation(mp);
 558        resp->tr_attrset = xfs_calc_attrset_reservation(mp);
 559        resp->tr_attrrm = xfs_calc_attrrm_reservation(mp);
 560        resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp);
 561        resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp);
 562        resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp);
 563        resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp);
 564}
 565
 566/*
 567 * This routine is called to allocate a transaction structure.
 568 * The type parameter indicates the type of the transaction.  These
 569 * are enumerated in xfs_trans.h.
 570 *
 571 * Dynamically allocate the transaction structure from the transaction
 572 * zone, initialize it, and return it to the caller.
 573 */
 574xfs_trans_t *
 575xfs_trans_alloc(
 576        xfs_mount_t     *mp,
 577        uint            type)
 578{
 579        xfs_trans_t     *tp;
 580
 581        sb_start_intwrite(mp->m_super);
 582        tp = _xfs_trans_alloc(mp, type, KM_SLEEP);
 583        tp->t_flags |= XFS_TRANS_FREEZE_PROT;
 584        return tp;
 585}
 586
 587xfs_trans_t *
 588_xfs_trans_alloc(
 589        xfs_mount_t     *mp,
 590        uint            type,
 591        xfs_km_flags_t  memflags)
 592{
 593        xfs_trans_t     *tp;
 594
 595        WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
 596        atomic_inc(&mp->m_active_trans);
 597
 598        tp = kmem_zone_zalloc(xfs_trans_zone, memflags);
 599        tp->t_magic = XFS_TRANS_MAGIC;
 600        tp->t_type = type;
 601        tp->t_mountp = mp;
 602        INIT_LIST_HEAD(&tp->t_items);
 603        INIT_LIST_HEAD(&tp->t_busy);
 604        return tp;
 605}
 606
 607/*
 608 * Free the transaction structure.  If there is more clean up
 609 * to do when the structure is freed, add it here.
 610 */
 611STATIC void
 612xfs_trans_free(
 613        struct xfs_trans        *tp)
 614{
 615        xfs_extent_busy_sort(&tp->t_busy);
 616        xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
 617
 618        atomic_dec(&tp->t_mountp->m_active_trans);
 619        if (tp->t_flags & XFS_TRANS_FREEZE_PROT)
 620                sb_end_intwrite(tp->t_mountp->m_super);
 621        xfs_trans_free_dqinfo(tp);
 622        kmem_zone_free(xfs_trans_zone, tp);
 623}
 624
 625/*
 626 * This is called to create a new transaction which will share the
 627 * permanent log reservation of the given transaction.  The remaining
 628 * unused block and rt extent reservations are also inherited.  This
 629 * implies that the original transaction is no longer allowed to allocate
 630 * blocks.  Locks and log items, however, are no inherited.  They must
 631 * be added to the new transaction explicitly.
 632 */
 633xfs_trans_t *
 634xfs_trans_dup(
 635        xfs_trans_t     *tp)
 636{
 637        xfs_trans_t     *ntp;
 638
 639        ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
 640
 641        /*
 642         * Initialize the new transaction structure.
 643         */
 644        ntp->t_magic = XFS_TRANS_MAGIC;
 645        ntp->t_type = tp->t_type;
 646        ntp->t_mountp = tp->t_mountp;
 647        INIT_LIST_HEAD(&ntp->t_items);
 648        INIT_LIST_HEAD(&ntp->t_busy);
 649
 650        ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 651        ASSERT(tp->t_ticket != NULL);
 652
 653        ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
 654                       (tp->t_flags & XFS_TRANS_RESERVE) |
 655                       (tp->t_flags & XFS_TRANS_FREEZE_PROT);
 656        /* We gave our writer reference to the new transaction */
 657        tp->t_flags &= ~XFS_TRANS_FREEZE_PROT;
 658        ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
 659        ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
 660        tp->t_blk_res = tp->t_blk_res_used;
 661        ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
 662        tp->t_rtx_res = tp->t_rtx_res_used;
 663        ntp->t_pflags = tp->t_pflags;
 664
 665        xfs_trans_dup_dqinfo(tp, ntp);
 666
 667        atomic_inc(&tp->t_mountp->m_active_trans);
 668        return ntp;
 669}
 670
 671/*
 672 * This is called to reserve free disk blocks and log space for the
 673 * given transaction.  This must be done before allocating any resources
 674 * within the transaction.
 675 *
 676 * This will return ENOSPC if there are not enough blocks available.
 677 * It will sleep waiting for available log space.
 678 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
 679 * is used by long running transactions.  If any one of the reservations
 680 * fails then they will all be backed out.
 681 *
 682 * This does not do quota reservations. That typically is done by the
 683 * caller afterwards.
 684 */
 685int
 686xfs_trans_reserve(
 687        xfs_trans_t     *tp,
 688        uint            blocks,
 689        uint            logspace,
 690        uint            rtextents,
 691        uint            flags,
 692        uint            logcount)
 693{
 694        int             error = 0;
 695        int             rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
 696
 697        /* Mark this thread as being in a transaction */
 698        current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
 699
 700        /*
 701         * Attempt to reserve the needed disk blocks by decrementing
 702         * the number needed from the number available.  This will
 703         * fail if the count would go below zero.
 704         */
 705        if (blocks > 0) {
 706                error = xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
 707                                          -((int64_t)blocks), rsvd);
 708                if (error != 0) {
 709                        current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 710                        return (XFS_ERROR(ENOSPC));
 711                }
 712                tp->t_blk_res += blocks;
 713        }
 714
 715        /*
 716         * Reserve the log space needed for this transaction.
 717         */
 718        if (logspace > 0) {
 719                bool    permanent = false;
 720
 721                ASSERT(tp->t_log_res == 0 || tp->t_log_res == logspace);
 722                ASSERT(tp->t_log_count == 0 || tp->t_log_count == logcount);
 723
 724                if (flags & XFS_TRANS_PERM_LOG_RES) {
 725                        tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
 726                        permanent = true;
 727                } else {
 728                        ASSERT(tp->t_ticket == NULL);
 729                        ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
 730                }
 731
 732                if (tp->t_ticket != NULL) {
 733                        ASSERT(flags & XFS_TRANS_PERM_LOG_RES);
 734                        error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
 735                } else {
 736                        error = xfs_log_reserve(tp->t_mountp, logspace,
 737                                                logcount, &tp->t_ticket,
 738                                                XFS_TRANSACTION, permanent,
 739                                                tp->t_type);
 740                }
 741
 742                if (error)
 743                        goto undo_blocks;
 744
 745                tp->t_log_res = logspace;
 746                tp->t_log_count = logcount;
 747        }
 748
 749        /*
 750         * Attempt to reserve the needed realtime extents by decrementing
 751         * the number needed from the number available.  This will
 752         * fail if the count would go below zero.
 753         */
 754        if (rtextents > 0) {
 755                error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
 756                                          -((int64_t)rtextents), rsvd);
 757                if (error) {
 758                        error = XFS_ERROR(ENOSPC);
 759                        goto undo_log;
 760                }
 761                tp->t_rtx_res += rtextents;
 762        }
 763
 764        return 0;
 765
 766        /*
 767         * Error cases jump to one of these labels to undo any
 768         * reservations which have already been performed.
 769         */
 770undo_log:
 771        if (logspace > 0) {
 772                int             log_flags;
 773
 774                if (flags & XFS_TRANS_PERM_LOG_RES) {
 775                        log_flags = XFS_LOG_REL_PERM_RESERV;
 776                } else {
 777                        log_flags = 0;
 778                }
 779                xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
 780                tp->t_ticket = NULL;
 781                tp->t_log_res = 0;
 782                tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
 783        }
 784
 785undo_blocks:
 786        if (blocks > 0) {
 787                xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS,
 788                                         (int64_t)blocks, rsvd);
 789                tp->t_blk_res = 0;
 790        }
 791
 792        current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 793
 794        return error;
 795}
 796
 797/*
 798 * Record the indicated change to the given field for application
 799 * to the file system's superblock when the transaction commits.
 800 * For now, just store the change in the transaction structure.
 801 *
 802 * Mark the transaction structure to indicate that the superblock
 803 * needs to be updated before committing.
 804 *
 805 * Because we may not be keeping track of allocated/free inodes and
 806 * used filesystem blocks in the superblock, we do not mark the
 807 * superblock dirty in this transaction if we modify these fields.
 808 * We still need to update the transaction deltas so that they get
 809 * applied to the incore superblock, but we don't want them to
 810 * cause the superblock to get locked and logged if these are the
 811 * only fields in the superblock that the transaction modifies.
 812 */
 813void
 814xfs_trans_mod_sb(
 815        xfs_trans_t     *tp,
 816        uint            field,
 817        int64_t         delta)
 818{
 819        uint32_t        flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
 820        xfs_mount_t     *mp = tp->t_mountp;
 821
 822        switch (field) {
 823        case XFS_TRANS_SB_ICOUNT:
 824                tp->t_icount_delta += delta;
 825                if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 826                        flags &= ~XFS_TRANS_SB_DIRTY;
 827                break;
 828        case XFS_TRANS_SB_IFREE:
 829                tp->t_ifree_delta += delta;
 830                if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 831                        flags &= ~XFS_TRANS_SB_DIRTY;
 832                break;
 833        case XFS_TRANS_SB_FDBLOCKS:
 834                /*
 835                 * Track the number of blocks allocated in the
 836                 * transaction.  Make sure it does not exceed the
 837                 * number reserved.
 838                 */
 839                if (delta < 0) {
 840                        tp->t_blk_res_used += (uint)-delta;
 841                        ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
 842                }
 843                tp->t_fdblocks_delta += delta;
 844                if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 845                        flags &= ~XFS_TRANS_SB_DIRTY;
 846                break;
 847        case XFS_TRANS_SB_RES_FDBLOCKS:
 848                /*
 849                 * The allocation has already been applied to the
 850                 * in-core superblock's counter.  This should only
 851                 * be applied to the on-disk superblock.
 852                 */
 853                ASSERT(delta < 0);
 854                tp->t_res_fdblocks_delta += delta;
 855                if (xfs_sb_version_haslazysbcount(&mp->m_sb))
 856                        flags &= ~XFS_TRANS_SB_DIRTY;
 857                break;
 858        case XFS_TRANS_SB_FREXTENTS:
 859                /*
 860                 * Track the number of blocks allocated in the
 861                 * transaction.  Make sure it does not exceed the
 862                 * number reserved.
 863                 */
 864                if (delta < 0) {
 865                        tp->t_rtx_res_used += (uint)-delta;
 866                        ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
 867                }
 868                tp->t_frextents_delta += delta;
 869                break;
 870        case XFS_TRANS_SB_RES_FREXTENTS:
 871                /*
 872                 * The allocation has already been applied to the
 873                 * in-core superblock's counter.  This should only
 874                 * be applied to the on-disk superblock.
 875                 */
 876                ASSERT(delta < 0);
 877                tp->t_res_frextents_delta += delta;
 878                break;
 879        case XFS_TRANS_SB_DBLOCKS:
 880                ASSERT(delta > 0);
 881                tp->t_dblocks_delta += delta;
 882                break;
 883        case XFS_TRANS_SB_AGCOUNT:
 884                ASSERT(delta > 0);
 885                tp->t_agcount_delta += delta;
 886                break;
 887        case XFS_TRANS_SB_IMAXPCT:
 888                tp->t_imaxpct_delta += delta;
 889                break;
 890        case XFS_TRANS_SB_REXTSIZE:
 891                tp->t_rextsize_delta += delta;
 892                break;
 893        case XFS_TRANS_SB_RBMBLOCKS:
 894                tp->t_rbmblocks_delta += delta;
 895                break;
 896        case XFS_TRANS_SB_RBLOCKS:
 897                tp->t_rblocks_delta += delta;
 898                break;
 899        case XFS_TRANS_SB_REXTENTS:
 900                tp->t_rextents_delta += delta;
 901                break;
 902        case XFS_TRANS_SB_REXTSLOG:
 903                tp->t_rextslog_delta += delta;
 904                break;
 905        default:
 906                ASSERT(0);
 907                return;
 908        }
 909
 910        tp->t_flags |= flags;
 911}
 912
 913/*
 914 * xfs_trans_apply_sb_deltas() is called from the commit code
 915 * to bring the superblock buffer into the current transaction
 916 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
 917 *
 918 * For now we just look at each field allowed to change and change
 919 * it if necessary.
 920 */
 921STATIC void
 922xfs_trans_apply_sb_deltas(
 923        xfs_trans_t     *tp)
 924{
 925        xfs_dsb_t       *sbp;
 926        xfs_buf_t       *bp;
 927        int             whole = 0;
 928
 929        bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
 930        sbp = XFS_BUF_TO_SBP(bp);
 931
 932        /*
 933         * Check that superblock mods match the mods made to AGF counters.
 934         */
 935        ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
 936               (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
 937                tp->t_ag_btree_delta));
 938
 939        /*
 940         * Only update the superblock counters if we are logging them
 941         */
 942        if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
 943                if (tp->t_icount_delta)
 944                        be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
 945                if (tp->t_ifree_delta)
 946                        be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
 947                if (tp->t_fdblocks_delta)
 948                        be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
 949                if (tp->t_res_fdblocks_delta)
 950                        be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
 951        }
 952
 953        if (tp->t_frextents_delta)
 954                be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
 955        if (tp->t_res_frextents_delta)
 956                be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
 957
 958        if (tp->t_dblocks_delta) {
 959                be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
 960                whole = 1;
 961        }
 962        if (tp->t_agcount_delta) {
 963                be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
 964                whole = 1;
 965        }
 966        if (tp->t_imaxpct_delta) {
 967                sbp->sb_imax_pct += tp->t_imaxpct_delta;
 968                whole = 1;
 969        }
 970        if (tp->t_rextsize_delta) {
 971                be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
 972                whole = 1;
 973        }
 974        if (tp->t_rbmblocks_delta) {
 975                be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
 976                whole = 1;
 977        }
 978        if (tp->t_rblocks_delta) {
 979                be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
 980                whole = 1;
 981        }
 982        if (tp->t_rextents_delta) {
 983                be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
 984                whole = 1;
 985        }
 986        if (tp->t_rextslog_delta) {
 987                sbp->sb_rextslog += tp->t_rextslog_delta;
 988                whole = 1;
 989        }
 990
 991        if (whole)
 992                /*
 993                 * Log the whole thing, the fields are noncontiguous.
 994                 */
 995                xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
 996        else
 997                /*
 998                 * Since all the modifiable fields are contiguous, we
 999                 * can get away with this.
1000                 */
1001                xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
1002                                  offsetof(xfs_dsb_t, sb_frextents) +
1003                                  sizeof(sbp->sb_frextents) - 1);
1004}
1005
1006/*
1007 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
1008 * and apply superblock counter changes to the in-core superblock.  The
1009 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
1010 * applied to the in-core superblock.  The idea is that that has already been
1011 * done.
1012 *
1013 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
1014 * However, we have to ensure that we only modify each superblock field only
1015 * once because the application of the delta values may not be atomic. That can
1016 * lead to ENOSPC races occurring if we have two separate modifcations of the
1017 * free space counter to put back the entire reservation and then take away
1018 * what we used.
1019 *
1020 * If we are not logging superblock counters, then the inode allocated/free and
1021 * used block counts are not updated in the on disk superblock. In this case,
1022 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
1023 * still need to update the incore superblock with the changes.
1024 */
1025void
1026xfs_trans_unreserve_and_mod_sb(
1027        xfs_trans_t     *tp)
1028{
1029        xfs_mod_sb_t    msb[9]; /* If you add cases, add entries */
1030        xfs_mod_sb_t    *msbp;
1031        xfs_mount_t     *mp = tp->t_mountp;
1032        /* REFERENCED */
1033        int             error;
1034        int             rsvd;
1035        int64_t         blkdelta = 0;
1036        int64_t         rtxdelta = 0;
1037        int64_t         idelta = 0;
1038        int64_t         ifreedelta = 0;
1039
1040        msbp = msb;
1041        rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
1042
1043        /* calculate deltas */
1044        if (tp->t_blk_res > 0)
1045                blkdelta = tp->t_blk_res;
1046        if ((tp->t_fdblocks_delta != 0) &&
1047            (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1048             (tp->t_flags & XFS_TRANS_SB_DIRTY)))
1049                blkdelta += tp->t_fdblocks_delta;
1050
1051        if (tp->t_rtx_res > 0)
1052                rtxdelta = tp->t_rtx_res;
1053        if ((tp->t_frextents_delta != 0) &&
1054            (tp->t_flags & XFS_TRANS_SB_DIRTY))
1055                rtxdelta += tp->t_frextents_delta;
1056
1057        if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
1058             (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
1059                idelta = tp->t_icount_delta;
1060                ifreedelta = tp->t_ifree_delta;
1061        }
1062
1063        /* apply the per-cpu counters */
1064        if (blkdelta) {
1065                error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
1066                                                 blkdelta, rsvd);
1067                if (error)
1068                        goto out;
1069        }
1070
1071        if (idelta) {
1072                error = xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT,
1073                                                 idelta, rsvd);
1074                if (error)
1075                        goto out_undo_fdblocks;
1076        }
1077
1078        if (ifreedelta) {
1079                error = xfs_icsb_modify_counters(mp, XFS_SBS_IFREE,
1080                                                 ifreedelta, rsvd);
1081                if (error)
1082                        goto out_undo_icount;
1083        }
1084
1085        /* apply remaining deltas */
1086        if (rtxdelta != 0) {
1087                msbp->msb_field = XFS_SBS_FREXTENTS;
1088                msbp->msb_delta = rtxdelta;
1089                msbp++;
1090        }
1091
1092        if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
1093                if (tp->t_dblocks_delta != 0) {
1094                        msbp->msb_field = XFS_SBS_DBLOCKS;
1095                        msbp->msb_delta = tp->t_dblocks_delta;
1096                        msbp++;
1097                }
1098                if (tp->t_agcount_delta != 0) {
1099                        msbp->msb_field = XFS_SBS_AGCOUNT;
1100                        msbp->msb_delta = tp->t_agcount_delta;
1101                        msbp++;
1102                }
1103                if (tp->t_imaxpct_delta != 0) {
1104                        msbp->msb_field = XFS_SBS_IMAX_PCT;
1105                        msbp->msb_delta = tp->t_imaxpct_delta;
1106                        msbp++;
1107                }
1108                if (tp->t_rextsize_delta != 0) {
1109                        msbp->msb_field = XFS_SBS_REXTSIZE;
1110                        msbp->msb_delta = tp->t_rextsize_delta;
1111                        msbp++;
1112                }
1113                if (tp->t_rbmblocks_delta != 0) {
1114                        msbp->msb_field = XFS_SBS_RBMBLOCKS;
1115                        msbp->msb_delta = tp->t_rbmblocks_delta;
1116                        msbp++;
1117                }
1118                if (tp->t_rblocks_delta != 0) {
1119                        msbp->msb_field = XFS_SBS_RBLOCKS;
1120                        msbp->msb_delta = tp->t_rblocks_delta;
1121                        msbp++;
1122                }
1123                if (tp->t_rextents_delta != 0) {
1124                        msbp->msb_field = XFS_SBS_REXTENTS;
1125                        msbp->msb_delta = tp->t_rextents_delta;
1126                        msbp++;
1127                }
1128                if (tp->t_rextslog_delta != 0) {
1129                        msbp->msb_field = XFS_SBS_REXTSLOG;
1130                        msbp->msb_delta = tp->t_rextslog_delta;
1131                        msbp++;
1132                }
1133        }
1134
1135        /*
1136         * If we need to change anything, do it.
1137         */
1138        if (msbp > msb) {
1139                error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
1140                        (uint)(msbp - msb), rsvd);
1141                if (error)
1142                        goto out_undo_ifreecount;
1143        }
1144
1145        return;
1146
1147out_undo_ifreecount:
1148        if (ifreedelta)
1149                xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, -ifreedelta, rsvd);
1150out_undo_icount:
1151        if (idelta)
1152                xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, -idelta, rsvd);
1153out_undo_fdblocks:
1154        if (blkdelta)
1155                xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, -blkdelta, rsvd);
1156out:
1157        ASSERT(error == 0);
1158        return;
1159}
1160
1161/*
1162 * Add the given log item to the transaction's list of log items.
1163 *
1164 * The log item will now point to its new descriptor with its li_desc field.
1165 */
1166void
1167xfs_trans_add_item(
1168        struct xfs_trans        *tp,
1169        struct xfs_log_item     *lip)
1170{
1171        struct xfs_log_item_desc *lidp;
1172
1173        ASSERT(lip->li_mountp == tp->t_mountp);
1174        ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
1175
1176        lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
1177
1178        lidp->lid_item = lip;
1179        lidp->lid_flags = 0;
1180        list_add_tail(&lidp->lid_trans, &tp->t_items);
1181
1182        lip->li_desc = lidp;
1183}
1184
1185STATIC void
1186xfs_trans_free_item_desc(
1187        struct xfs_log_item_desc *lidp)
1188{
1189        list_del_init(&lidp->lid_trans);
1190        kmem_zone_free(xfs_log_item_desc_zone, lidp);
1191}
1192
1193/*
1194 * Unlink and free the given descriptor.
1195 */
1196void
1197xfs_trans_del_item(
1198        struct xfs_log_item     *lip)
1199{
1200        xfs_trans_free_item_desc(lip->li_desc);
1201        lip->li_desc = NULL;
1202}
1203
1204/*
1205 * Unlock all of the items of a transaction and free all the descriptors
1206 * of that transaction.
1207 */
1208void
1209xfs_trans_free_items(
1210        struct xfs_trans        *tp,
1211        xfs_lsn_t               commit_lsn,
1212        int                     flags)
1213{
1214        struct xfs_log_item_desc *lidp, *next;
1215
1216        list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
1217                struct xfs_log_item     *lip = lidp->lid_item;
1218
1219                lip->li_desc = NULL;
1220
1221                if (commit_lsn != NULLCOMMITLSN)
1222                        IOP_COMMITTING(lip, commit_lsn);
1223                if (flags & XFS_TRANS_ABORT)
1224                        lip->li_flags |= XFS_LI_ABORTED;
1225                IOP_UNLOCK(lip);
1226
1227                xfs_trans_free_item_desc(lidp);
1228        }
1229}
1230
1231static inline void
1232xfs_log_item_batch_insert(
1233        struct xfs_ail          *ailp,
1234        struct xfs_ail_cursor   *cur,
1235        struct xfs_log_item     **log_items,
1236        int                     nr_items,
1237        xfs_lsn_t               commit_lsn)
1238{
1239        int     i;
1240
1241        spin_lock(&ailp->xa_lock);
1242        /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
1243        xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
1244
1245        for (i = 0; i < nr_items; i++)
1246                IOP_UNPIN(log_items[i], 0);
1247}
1248
1249/*
1250 * Bulk operation version of xfs_trans_committed that takes a log vector of
1251 * items to insert into the AIL. This uses bulk AIL insertion techniques to
1252 * minimise lock traffic.
1253 *
1254 * If we are called with the aborted flag set, it is because a log write during
1255 * a CIL checkpoint commit has failed. In this case, all the items in the
1256 * checkpoint have already gone through IOP_COMMITED and IOP_UNLOCK, which
1257 * means that checkpoint commit abort handling is treated exactly the same
1258 * as an iclog write error even though we haven't started any IO yet. Hence in
1259 * this case all we need to do is IOP_COMMITTED processing, followed by an
1260 * IOP_UNPIN(aborted) call.
1261 *
1262 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
1263 * at the end of the AIL, the insert cursor avoids the need to walk
1264 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
1265 * call. This saves a lot of needless list walking and is a net win, even
1266 * though it slightly increases that amount of AIL lock traffic to set it up
1267 * and tear it down.
1268 */
1269void
1270xfs_trans_committed_bulk(
1271        struct xfs_ail          *ailp,
1272        struct xfs_log_vec      *log_vector,
1273        xfs_lsn_t               commit_lsn,
1274        int                     aborted)
1275{
1276#define LOG_ITEM_BATCH_SIZE     32
1277        struct xfs_log_item     *log_items[LOG_ITEM_BATCH_SIZE];
1278        struct xfs_log_vec      *lv;
1279        struct xfs_ail_cursor   cur;
1280        int                     i = 0;
1281
1282        spin_lock(&ailp->xa_lock);
1283        xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
1284        spin_unlock(&ailp->xa_lock);
1285
1286        /* unpin all the log items */
1287        for (lv = log_vector; lv; lv = lv->lv_next ) {
1288                struct xfs_log_item     *lip = lv->lv_item;
1289                xfs_lsn_t               item_lsn;
1290
1291                if (aborted)
1292                        lip->li_flags |= XFS_LI_ABORTED;
1293                item_lsn = IOP_COMMITTED(lip, commit_lsn);
1294
1295                /* item_lsn of -1 means the item needs no further processing */
1296                if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
1297                        continue;
1298
1299                /*
1300                 * if we are aborting the operation, no point in inserting the
1301                 * object into the AIL as we are in a shutdown situation.
1302                 */
1303                if (aborted) {
1304                        ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount));
1305                        IOP_UNPIN(lip, 1);
1306                        continue;
1307                }
1308
1309                if (item_lsn != commit_lsn) {
1310
1311                        /*
1312                         * Not a bulk update option due to unusual item_lsn.
1313                         * Push into AIL immediately, rechecking the lsn once
1314                         * we have the ail lock. Then unpin the item. This does
1315                         * not affect the AIL cursor the bulk insert path is
1316                         * using.
1317                         */
1318                        spin_lock(&ailp->xa_lock);
1319                        if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
1320                                xfs_trans_ail_update(ailp, lip, item_lsn);
1321                        else
1322                                spin_unlock(&ailp->xa_lock);
1323                        IOP_UNPIN(lip, 0);
1324                        continue;
1325                }
1326
1327                /* Item is a candidate for bulk AIL insert.  */
1328                log_items[i++] = lv->lv_item;
1329                if (i >= LOG_ITEM_BATCH_SIZE) {
1330                        xfs_log_item_batch_insert(ailp, &cur, log_items,
1331                                        LOG_ITEM_BATCH_SIZE, commit_lsn);
1332                        i = 0;
1333                }
1334        }
1335
1336        /* make sure we insert the remainder! */
1337        if (i)
1338                xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
1339
1340        spin_lock(&ailp->xa_lock);
1341        xfs_trans_ail_cursor_done(ailp, &cur);
1342        spin_unlock(&ailp->xa_lock);
1343}
1344
1345/*
1346 * Commit the given transaction to the log.
1347 *
1348 * XFS disk error handling mechanism is not based on a typical
1349 * transaction abort mechanism. Logically after the filesystem
1350 * gets marked 'SHUTDOWN', we can't let any new transactions
1351 * be durable - ie. committed to disk - because some metadata might
1352 * be inconsistent. In such cases, this returns an error, and the
1353 * caller may assume that all locked objects joined to the transaction
1354 * have already been unlocked as if the commit had succeeded.
1355 * Do not reference the transaction structure after this call.
1356 */
1357int
1358xfs_trans_commit(
1359        struct xfs_trans        *tp,
1360        uint                    flags)
1361{
1362        struct xfs_mount        *mp = tp->t_mountp;
1363        xfs_lsn_t               commit_lsn = -1;
1364        int                     error = 0;
1365        int                     log_flags = 0;
1366        int                     sync = tp->t_flags & XFS_TRANS_SYNC;
1367
1368        /*
1369         * Determine whether this commit is releasing a permanent
1370         * log reservation or not.
1371         */
1372        if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1373                ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1374                log_flags = XFS_LOG_REL_PERM_RESERV;
1375        }
1376
1377        /*
1378         * If there is nothing to be logged by the transaction,
1379         * then unlock all of the items associated with the
1380         * transaction and free the transaction structure.
1381         * Also make sure to return any reserved blocks to
1382         * the free pool.
1383         */
1384        if (!(tp->t_flags & XFS_TRANS_DIRTY))
1385                goto out_unreserve;
1386
1387        if (XFS_FORCED_SHUTDOWN(mp)) {
1388                error = XFS_ERROR(EIO);
1389                goto out_unreserve;
1390        }
1391
1392        ASSERT(tp->t_ticket != NULL);
1393
1394        /*
1395         * If we need to update the superblock, then do it now.
1396         */
1397        if (tp->t_flags & XFS_TRANS_SB_DIRTY)
1398                xfs_trans_apply_sb_deltas(tp);
1399        xfs_trans_apply_dquot_deltas(tp);
1400
1401        error = xfs_log_commit_cil(mp, tp, &commit_lsn, flags);
1402        if (error == ENOMEM) {
1403                xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
1404                error = XFS_ERROR(EIO);
1405                goto out_unreserve;
1406        }
1407
1408        current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1409        xfs_trans_free(tp);
1410
1411        /*
1412         * If the transaction needs to be synchronous, then force the
1413         * log out now and wait for it.
1414         */
1415        if (sync) {
1416                if (!error) {
1417                        error = _xfs_log_force_lsn(mp, commit_lsn,
1418                                      XFS_LOG_SYNC, NULL);
1419                }
1420                XFS_STATS_INC(xs_trans_sync);
1421        } else {
1422                XFS_STATS_INC(xs_trans_async);
1423        }
1424
1425        return error;
1426
1427out_unreserve:
1428        xfs_trans_unreserve_and_mod_sb(tp);
1429
1430        /*
1431         * It is indeed possible for the transaction to be not dirty but
1432         * the dqinfo portion to be.  All that means is that we have some
1433         * (non-persistent) quota reservations that need to be unreserved.
1434         */
1435        xfs_trans_unreserve_and_mod_dquots(tp);
1436        if (tp->t_ticket) {
1437                commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1438                if (commit_lsn == -1 && !error)
1439                        error = XFS_ERROR(EIO);
1440        }
1441        current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1442        xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0);
1443        xfs_trans_free(tp);
1444
1445        XFS_STATS_INC(xs_trans_empty);
1446        return error;
1447}
1448
1449/*
1450 * Unlock all of the transaction's items and free the transaction.
1451 * The transaction must not have modified any of its items, because
1452 * there is no way to restore them to their previous state.
1453 *
1454 * If the transaction has made a log reservation, make sure to release
1455 * it as well.
1456 */
1457void
1458xfs_trans_cancel(
1459        xfs_trans_t             *tp,
1460        int                     flags)
1461{
1462        int                     log_flags;
1463        xfs_mount_t             *mp = tp->t_mountp;
1464
1465        /*
1466         * See if the caller is being too lazy to figure out if
1467         * the transaction really needs an abort.
1468         */
1469        if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1470                flags &= ~XFS_TRANS_ABORT;
1471        /*
1472         * See if the caller is relying on us to shut down the
1473         * filesystem.  This happens in paths where we detect
1474         * corruption and decide to give up.
1475         */
1476        if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
1477                XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1478                xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1479        }
1480#ifdef DEBUG
1481        if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) {
1482                struct xfs_log_item_desc *lidp;
1483
1484                list_for_each_entry(lidp, &tp->t_items, lid_trans)
1485                        ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
1486        }
1487#endif
1488        xfs_trans_unreserve_and_mod_sb(tp);
1489        xfs_trans_unreserve_and_mod_dquots(tp);
1490
1491        if (tp->t_ticket) {
1492                if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1493                        ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1494                        log_flags = XFS_LOG_REL_PERM_RESERV;
1495                } else {
1496                        log_flags = 0;
1497                }
1498                xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1499        }
1500
1501        /* mark this thread as no longer being in a transaction */
1502        current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1503
1504        xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
1505        xfs_trans_free(tp);
1506}
1507
1508/*
1509 * Roll from one trans in the sequence of PERMANENT transactions to
1510 * the next: permanent transactions are only flushed out when
1511 * committed with XFS_TRANS_RELEASE_LOG_RES, but we still want as soon
1512 * as possible to let chunks of it go to the log. So we commit the
1513 * chunk we've been working on and get a new transaction to continue.
1514 */
1515int
1516xfs_trans_roll(
1517        struct xfs_trans        **tpp,
1518        struct xfs_inode        *dp)
1519{
1520        struct xfs_trans        *trans;
1521        unsigned int            logres, count;
1522        int                     error;
1523
1524        /*
1525         * Ensure that the inode is always logged.
1526         */
1527        trans = *tpp;
1528        xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1529
1530        /*
1531         * Copy the critical parameters from one trans to the next.
1532         */
1533        logres = trans->t_log_res;
1534        count = trans->t_log_count;
1535        *tpp = xfs_trans_dup(trans);
1536
1537        /*
1538         * Commit the current transaction.
1539         * If this commit failed, then it'd just unlock those items that
1540         * are not marked ihold. That also means that a filesystem shutdown
1541         * is in progress. The caller takes the responsibility to cancel
1542         * the duplicate transaction that gets returned.
1543         */
1544        error = xfs_trans_commit(trans, 0);
1545        if (error)
1546                return (error);
1547
1548        trans = *tpp;
1549
1550        /*
1551         * transaction commit worked ok so we can drop the extra ticket
1552         * reference that we gained in xfs_trans_dup()
1553         */
1554        xfs_log_ticket_put(trans->t_ticket);
1555
1556
1557        /*
1558         * Reserve space in the log for th next transaction.
1559         * This also pushes items in the "AIL", the list of logged items,
1560         * out to disk if they are taking up space at the tail of the log
1561         * that we want to use.  This requires that either nothing be locked
1562         * across this call, or that anything that is locked be logged in
1563         * the prior and the next transactions.
1564         */
1565        error = xfs_trans_reserve(trans, 0, logres, 0,
1566                                  XFS_TRANS_PERM_LOG_RES, count);
1567        /*
1568         *  Ensure that the inode is in the new transaction and locked.
1569         */
1570        if (error)
1571                return error;
1572
1573        xfs_trans_ijoin(trans, dp, 0);
1574        return 0;
1575}
1576
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.