10b61f8a4SDave Chinner // SPDX-License-Identifier: GPL-2.0 268988114SDave Chinner /* 368988114SDave Chinner * Copyright (c) 2000-2006 Silicon Graphics, Inc. 4c24b5dfaSDave Chinner * Copyright (c) 2012 Red Hat, Inc. 568988114SDave Chinner * All Rights Reserved. 668988114SDave Chinner */ 768988114SDave Chinner #include "xfs.h" 868988114SDave Chinner #include "xfs_fs.h" 970a9883cSDave Chinner #include "xfs_shared.h" 10239880efSDave Chinner #include "xfs_format.h" 11239880efSDave Chinner #include "xfs_log_format.h" 12239880efSDave Chinner #include "xfs_trans_resv.h" 1368988114SDave Chinner #include "xfs_bit.h" 1468988114SDave Chinner #include "xfs_mount.h" 153ab78df2SDarrick J. Wong #include "xfs_defer.h" 1668988114SDave Chinner #include "xfs_inode.h" 1768988114SDave Chinner #include "xfs_btree.h" 18239880efSDave Chinner #include "xfs_trans.h" 1968988114SDave Chinner #include "xfs_alloc.h" 2068988114SDave Chinner #include "xfs_bmap.h" 2168988114SDave Chinner #include "xfs_bmap_util.h" 22a4fbe6abSDave Chinner #include "xfs_bmap_btree.h" 2368988114SDave Chinner #include "xfs_rtalloc.h" 2468988114SDave Chinner #include "xfs_error.h" 2568988114SDave Chinner #include "xfs_quota.h" 2668988114SDave Chinner #include "xfs_trans_space.h" 2768988114SDave Chinner #include "xfs_trace.h" 28c24b5dfaSDave Chinner #include "xfs_icache.h" 29f86f4037SDarrick J. Wong #include "xfs_iomap.h" 30f86f4037SDarrick J. Wong #include "xfs_reflink.h" 3168988114SDave Chinner 3268988114SDave Chinner /* Kernel only BMAP related definitions and functions */ 3368988114SDave Chinner 3468988114SDave Chinner /* 3568988114SDave Chinner * Convert the given file system block to a disk block. We have to treat it 3668988114SDave Chinner * differently based on whether the file is a real time file or not, because the 3768988114SDave Chinner * bmap code does. 3868988114SDave Chinner */ 3968988114SDave Chinner xfs_daddr_t 4068988114SDave Chinner xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb) 4168988114SDave Chinner { 42ecfc28a4SChristoph Hellwig if (XFS_IS_REALTIME_INODE(ip)) 43ecfc28a4SChristoph Hellwig return XFS_FSB_TO_BB(ip->i_mount, fsb); 44ecfc28a4SChristoph Hellwig return XFS_FSB_TO_DADDR(ip->i_mount, fsb); 4568988114SDave Chinner } 4668988114SDave Chinner 4768988114SDave Chinner /* 483fbbbea3SDave Chinner * Routine to zero an extent on disk allocated to the specific inode. 493fbbbea3SDave Chinner * 503fbbbea3SDave Chinner * The VFS functions take a linearised filesystem block offset, so we have to 513fbbbea3SDave Chinner * convert the sparse xfs fsb to the right format first. 523fbbbea3SDave Chinner * VFS types are real funky, too. 533fbbbea3SDave Chinner */ 543fbbbea3SDave Chinner int 553fbbbea3SDave Chinner xfs_zero_extent( 563fbbbea3SDave Chinner struct xfs_inode *ip, 573fbbbea3SDave Chinner xfs_fsblock_t start_fsb, 583fbbbea3SDave Chinner xfs_off_t count_fsb) 593fbbbea3SDave Chinner { 603fbbbea3SDave Chinner struct xfs_mount *mp = ip->i_mount; 6130fa529eSChristoph Hellwig struct xfs_buftarg *target = xfs_inode_buftarg(ip); 623fbbbea3SDave Chinner xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb); 633fbbbea3SDave Chinner sector_t block = XFS_BB_TO_FSBT(mp, sector); 643fbbbea3SDave Chinner 6530fa529eSChristoph Hellwig return blkdev_issue_zeroout(target->bt_bdev, 663dc29161SMatthew Wilcox block << (mp->m_super->s_blocksize_bits - 9), 673dc29161SMatthew Wilcox count_fsb << (mp->m_super->s_blocksize_bits - 9), 68ee472d83SChristoph Hellwig GFP_NOFS, 0); 693fbbbea3SDave Chinner } 703fbbbea3SDave Chinner 71bb9c2e54SDave Chinner #ifdef CONFIG_XFS_RT 7268988114SDave Chinner int 7368988114SDave Chinner xfs_bmap_rtalloc( 749d5e8492SDarrick J. Wong struct xfs_bmalloca *ap) 7568988114SDave Chinner { 769d5e8492SDarrick J. Wong struct xfs_mount *mp = ap->ip->i_mount; 779d5e8492SDarrick J. Wong xfs_fileoff_t orig_offset = ap->offset; 789d5e8492SDarrick J. Wong xfs_rtblock_t rtb; 7968988114SDave Chinner xfs_extlen_t prod = 0; /* product factor for allocators */ 800703a8e1SDave Chinner xfs_extlen_t mod = 0; /* product factor for allocators */ 8168988114SDave Chinner xfs_extlen_t ralen = 0; /* realtime allocation length */ 8268988114SDave Chinner xfs_extlen_t align; /* minimum allocation alignment */ 839d5e8492SDarrick J. Wong xfs_extlen_t orig_length = ap->length; 849d5e8492SDarrick J. Wong xfs_extlen_t minlen = mp->m_sb.sb_rextsize; 859d5e8492SDarrick J. Wong xfs_extlen_t raminlen; 869d5e8492SDarrick J. Wong bool rtlocked = false; 87676a659bSDarrick J. Wong bool ignore_locality = false; 889d5e8492SDarrick J. Wong int error; 8968988114SDave Chinner 9068988114SDave Chinner align = xfs_get_extsz_hint(ap->ip); 919d5e8492SDarrick J. Wong retry: 9268988114SDave Chinner prod = align / mp->m_sb.sb_rextsize; 9368988114SDave Chinner error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, 9468988114SDave Chinner align, 1, ap->eof, 0, 9568988114SDave Chinner ap->conv, &ap->offset, &ap->length); 9668988114SDave Chinner if (error) 9768988114SDave Chinner return error; 9868988114SDave Chinner ASSERT(ap->length); 9968988114SDave Chinner ASSERT(ap->length % mp->m_sb.sb_rextsize == 0); 10068988114SDave Chinner 10168988114SDave Chinner /* 1029d5e8492SDarrick J. Wong * If we shifted the file offset downward to satisfy an extent size 1039d5e8492SDarrick J. Wong * hint, increase minlen by that amount so that the allocator won't 1049d5e8492SDarrick J. Wong * give us an allocation that's too short to cover at least one of the 1059d5e8492SDarrick J. Wong * blocks that the caller asked for. 1069d5e8492SDarrick J. Wong */ 1079d5e8492SDarrick J. Wong if (ap->offset != orig_offset) 1089d5e8492SDarrick J. Wong minlen += orig_offset - ap->offset; 1099d5e8492SDarrick J. Wong 1109d5e8492SDarrick J. Wong /* 11168988114SDave Chinner * If the offset & length are not perfectly aligned 11268988114SDave Chinner * then kill prod, it will just get us in trouble. 11368988114SDave Chinner */ 1140703a8e1SDave Chinner div_u64_rem(ap->offset, align, &mod); 1150703a8e1SDave Chinner if (mod || ap->length % align) 11668988114SDave Chinner prod = 1; 11768988114SDave Chinner /* 11868988114SDave Chinner * Set ralen to be the actual requested length in rtextents. 11968988114SDave Chinner */ 12068988114SDave Chinner ralen = ap->length / mp->m_sb.sb_rextsize; 12168988114SDave Chinner /* 12295f0b95eSChandan Babu R * If the old value was close enough to XFS_BMBT_MAX_EXTLEN that 12368988114SDave Chinner * we rounded up to it, cut it back so it's valid again. 12468988114SDave Chinner * Note that if it's a really large request (bigger than 12595f0b95eSChandan Babu R * XFS_BMBT_MAX_EXTLEN), we don't hear about that number, and can't 12668988114SDave Chinner * adjust the starting point to match it. 12768988114SDave Chinner */ 12895f0b95eSChandan Babu R if (ralen * mp->m_sb.sb_rextsize >= XFS_MAX_BMBT_EXTLEN) 12995f0b95eSChandan Babu R ralen = XFS_MAX_BMBT_EXTLEN / mp->m_sb.sb_rextsize; 13068988114SDave Chinner 13168988114SDave Chinner /* 1324b680afbSDave Chinner * Lock out modifications to both the RT bitmap and summary inodes 13368988114SDave Chinner */ 1349d5e8492SDarrick J. Wong if (!rtlocked) { 135f4a0660dSDarrick J. Wong xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); 13668988114SDave Chinner xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL); 137f4a0660dSDarrick J. Wong xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); 1384b680afbSDave Chinner xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL); 1399d5e8492SDarrick J. Wong rtlocked = true; 1409d5e8492SDarrick J. Wong } 14168988114SDave Chinner 14268988114SDave Chinner /* 14368988114SDave Chinner * If it's an allocation to an empty file at offset 0, 14468988114SDave Chinner * pick an extent that will space things out in the rt area. 14568988114SDave Chinner */ 14668988114SDave Chinner if (ap->eof && ap->offset == 0) { 1473f649ab7SKees Cook xfs_rtblock_t rtx; /* realtime extent no */ 14868988114SDave Chinner 14968988114SDave Chinner error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx); 15068988114SDave Chinner if (error) 15168988114SDave Chinner return error; 15268988114SDave Chinner ap->blkno = rtx * mp->m_sb.sb_rextsize; 15368988114SDave Chinner } else { 15468988114SDave Chinner ap->blkno = 0; 15568988114SDave Chinner } 15668988114SDave Chinner 15768988114SDave Chinner xfs_bmap_adjacent(ap); 15868988114SDave Chinner 15968988114SDave Chinner /* 16068988114SDave Chinner * Realtime allocation, done through xfs_rtallocate_extent. 16168988114SDave Chinner */ 162676a659bSDarrick J. Wong if (ignore_locality) 163676a659bSDarrick J. Wong ap->blkno = 0; 164676a659bSDarrick J. Wong else 16568988114SDave Chinner do_div(ap->blkno, mp->m_sb.sb_rextsize); 16668988114SDave Chinner rtb = ap->blkno; 16768988114SDave Chinner ap->length = ralen; 1689d5e8492SDarrick J. Wong raminlen = max_t(xfs_extlen_t, 1, minlen / mp->m_sb.sb_rextsize); 1699d5e8492SDarrick J. Wong error = xfs_rtallocate_extent(ap->tp, ap->blkno, raminlen, ap->length, 170089ec2f8SChristoph Hellwig &ralen, ap->wasdel, prod, &rtb); 171089ec2f8SChristoph Hellwig if (error) 17268988114SDave Chinner return error; 173089ec2f8SChristoph Hellwig 1749d5e8492SDarrick J. Wong if (rtb != NULLRTBLOCK) { 1759d5e8492SDarrick J. Wong ap->blkno = rtb * mp->m_sb.sb_rextsize; 1769d5e8492SDarrick J. Wong ap->length = ralen * mp->m_sb.sb_rextsize; 1779d5e8492SDarrick J. Wong ap->ip->i_nblocks += ap->length; 17868988114SDave Chinner xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); 17968988114SDave Chinner if (ap->wasdel) 1809d5e8492SDarrick J. Wong ap->ip->i_delayed_blks -= ap->length; 18168988114SDave Chinner /* 18268988114SDave Chinner * Adjust the disk quota also. This was reserved 18368988114SDave Chinner * earlier. 18468988114SDave Chinner */ 18568988114SDave Chinner xfs_trans_mod_dquot_byino(ap->tp, ap->ip, 18668988114SDave Chinner ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT : 1879d5e8492SDarrick J. Wong XFS_TRANS_DQ_RTBCOUNT, ap->length); 1889d5e8492SDarrick J. Wong return 0; 18968988114SDave Chinner } 1909d5e8492SDarrick J. Wong 1919d5e8492SDarrick J. Wong if (align > mp->m_sb.sb_rextsize) { 1929d5e8492SDarrick J. Wong /* 1939d5e8492SDarrick J. Wong * We previously enlarged the request length to try to satisfy 1949d5e8492SDarrick J. Wong * an extent size hint. The allocator didn't return anything, 1959d5e8492SDarrick J. Wong * so reset the parameters to the original values and try again 1969d5e8492SDarrick J. Wong * without alignment criteria. 1979d5e8492SDarrick J. Wong */ 1989d5e8492SDarrick J. Wong ap->offset = orig_offset; 1999d5e8492SDarrick J. Wong ap->length = orig_length; 2009d5e8492SDarrick J. Wong minlen = align = mp->m_sb.sb_rextsize; 2019d5e8492SDarrick J. Wong goto retry; 2029d5e8492SDarrick J. Wong } 2039d5e8492SDarrick J. Wong 204676a659bSDarrick J. Wong if (!ignore_locality && ap->blkno != 0) { 205676a659bSDarrick J. Wong /* 206676a659bSDarrick J. Wong * If we can't allocate near a specific rt extent, try again 207676a659bSDarrick J. Wong * without locality criteria. 208676a659bSDarrick J. Wong */ 209676a659bSDarrick J. Wong ignore_locality = true; 210676a659bSDarrick J. Wong goto retry; 211676a659bSDarrick J. Wong } 212676a659bSDarrick J. Wong 2139d5e8492SDarrick J. Wong ap->blkno = NULLFSBLOCK; 2149d5e8492SDarrick J. Wong ap->length = 0; 21568988114SDave Chinner return 0; 21668988114SDave Chinner } 217bb9c2e54SDave Chinner #endif /* CONFIG_XFS_RT */ 21868988114SDave Chinner 21968988114SDave Chinner /* 22068988114SDave Chinner * Extent tree block counting routines. 22168988114SDave Chinner */ 22268988114SDave Chinner 22368988114SDave Chinner /* 224d29cb3e4SDarrick J. Wong * Count leaf blocks given a range of extent records. Delayed allocation 225d29cb3e4SDarrick J. Wong * extents are not counted towards the totals. 22668988114SDave Chinner */ 227e17a5c6fSChristoph Hellwig xfs_extnum_t 22868988114SDave Chinner xfs_bmap_count_leaves( 229d29cb3e4SDarrick J. Wong struct xfs_ifork *ifp, 230e7f5d5caSDarrick J. Wong xfs_filblks_t *count) 23168988114SDave Chinner { 232b2b1712aSChristoph Hellwig struct xfs_iext_cursor icur; 233e17a5c6fSChristoph Hellwig struct xfs_bmbt_irec got; 234b2b1712aSChristoph Hellwig xfs_extnum_t numrecs = 0; 23568988114SDave Chinner 236b2b1712aSChristoph Hellwig for_each_xfs_iext(ifp, &icur, &got) { 237e17a5c6fSChristoph Hellwig if (!isnullstartblock(got.br_startblock)) { 238e17a5c6fSChristoph Hellwig *count += got.br_blockcount; 239e17a5c6fSChristoph Hellwig numrecs++; 24068988114SDave Chinner } 24168988114SDave Chinner } 242b2b1712aSChristoph Hellwig 243e17a5c6fSChristoph Hellwig return numrecs; 244d29cb3e4SDarrick J. Wong } 24568988114SDave Chinner 24668988114SDave Chinner /* 247d29cb3e4SDarrick J. Wong * Count fsblocks of the given fork. Delayed allocation extents are 248d29cb3e4SDarrick J. Wong * not counted towards the totals. 24968988114SDave Chinner */ 250e7f5d5caSDarrick J. Wong int 25168988114SDave Chinner xfs_bmap_count_blocks( 252e7f5d5caSDarrick J. Wong struct xfs_trans *tp, 253e7f5d5caSDarrick J. Wong struct xfs_inode *ip, 254e7f5d5caSDarrick J. Wong int whichfork, 255e7f5d5caSDarrick J. Wong xfs_extnum_t *nextents, 256e7f5d5caSDarrick J. Wong xfs_filblks_t *count) 25768988114SDave Chinner { 258fec40e22SDarrick J. Wong struct xfs_mount *mp = ip->i_mount; 259732436efSDarrick J. Wong struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork); 260fec40e22SDarrick J. Wong struct xfs_btree_cur *cur; 261fec40e22SDarrick J. Wong xfs_extlen_t btblocks = 0; 262e7f5d5caSDarrick J. Wong int error; 26368988114SDave Chinner 264e7f5d5caSDarrick J. Wong *nextents = 0; 265e7f5d5caSDarrick J. Wong *count = 0; 266fec40e22SDarrick J. Wong 267e7f5d5caSDarrick J. Wong if (!ifp) 26868988114SDave Chinner return 0; 269e7f5d5caSDarrick J. Wong 270f7e67b20SChristoph Hellwig switch (ifp->if_format) { 271e7f5d5caSDarrick J. Wong case XFS_DINODE_FMT_BTREE: 272e7f5d5caSDarrick J. Wong error = xfs_iread_extents(tp, ip, whichfork); 273e7f5d5caSDarrick J. Wong if (error) 274e7f5d5caSDarrick J. Wong return error; 27568988114SDave Chinner 276fec40e22SDarrick J. Wong cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); 277fec40e22SDarrick J. Wong error = xfs_btree_count_blocks(cur, &btblocks); 278fec40e22SDarrick J. Wong xfs_btree_del_cursor(cur, error); 279fec40e22SDarrick J. Wong if (error) 280fec40e22SDarrick J. Wong return error; 28168988114SDave Chinner 282fec40e22SDarrick J. Wong /* 283fec40e22SDarrick J. Wong * xfs_btree_count_blocks includes the root block contained in 284fec40e22SDarrick J. Wong * the inode fork in @btblocks, so subtract one because we're 285fec40e22SDarrick J. Wong * only interested in allocated disk blocks. 286fec40e22SDarrick J. Wong */ 287fec40e22SDarrick J. Wong *count += btblocks - 1; 288fec40e22SDarrick J. Wong 28953004ee7SGustavo A. R. Silva fallthrough; 290fec40e22SDarrick J. Wong case XFS_DINODE_FMT_EXTENTS: 291fec40e22SDarrick J. Wong *nextents = xfs_bmap_count_leaves(ifp, count); 292fec40e22SDarrick J. Wong break; 293e7f5d5caSDarrick J. Wong } 29468988114SDave Chinner 29568988114SDave Chinner return 0; 29668988114SDave Chinner } 29768988114SDave Chinner 298abbf9e8aSChristoph Hellwig static int 299abbf9e8aSChristoph Hellwig xfs_getbmap_report_one( 300f86f4037SDarrick J. Wong struct xfs_inode *ip, 301abbf9e8aSChristoph Hellwig struct getbmapx *bmv, 302232b5194SChristoph Hellwig struct kgetbmap *out, 303abbf9e8aSChristoph Hellwig int64_t bmv_end, 304abbf9e8aSChristoph Hellwig struct xfs_bmbt_irec *got) 305f86f4037SDarrick J. Wong { 306232b5194SChristoph Hellwig struct kgetbmap *p = out + bmv->bmv_entries; 307d392bc81SChristoph Hellwig bool shared = false; 308f86f4037SDarrick J. Wong int error; 309f86f4037SDarrick J. Wong 310d392bc81SChristoph Hellwig error = xfs_reflink_trim_around_shared(ip, got, &shared); 311f86f4037SDarrick J. Wong if (error) 312f86f4037SDarrick J. Wong return error; 313f86f4037SDarrick J. Wong 314abbf9e8aSChristoph Hellwig if (isnullstartblock(got->br_startblock) || 315abbf9e8aSChristoph Hellwig got->br_startblock == DELAYSTARTBLOCK) { 316f86f4037SDarrick J. Wong /* 3178ee81ed5SYe Bin * Take the flush completion as being a point-in-time snapshot 3188ee81ed5SYe Bin * where there are no delalloc extents, and if any new ones 3198ee81ed5SYe Bin * have been created racily, just skip them as being 'after' 3208ee81ed5SYe Bin * the flush and so don't get reported. 321f86f4037SDarrick J. Wong */ 3228ee81ed5SYe Bin if (!(bmv->bmv_iflags & BMV_IF_DELALLOC)) 3238ee81ed5SYe Bin return 0; 324abbf9e8aSChristoph Hellwig 325abbf9e8aSChristoph Hellwig p->bmv_oflags |= BMV_OF_DELALLOC; 326abbf9e8aSChristoph Hellwig p->bmv_block = -2; 327f86f4037SDarrick J. Wong } else { 328abbf9e8aSChristoph Hellwig p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock); 329f86f4037SDarrick J. Wong } 330f86f4037SDarrick J. Wong 331abbf9e8aSChristoph Hellwig if (got->br_state == XFS_EXT_UNWRITTEN && 332abbf9e8aSChristoph Hellwig (bmv->bmv_iflags & BMV_IF_PREALLOC)) 333abbf9e8aSChristoph Hellwig p->bmv_oflags |= BMV_OF_PREALLOC; 334abbf9e8aSChristoph Hellwig 335abbf9e8aSChristoph Hellwig if (shared) 336abbf9e8aSChristoph Hellwig p->bmv_oflags |= BMV_OF_SHARED; 337abbf9e8aSChristoph Hellwig 338abbf9e8aSChristoph Hellwig p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff); 339abbf9e8aSChristoph Hellwig p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount); 340abbf9e8aSChristoph Hellwig 341abbf9e8aSChristoph Hellwig bmv->bmv_offset = p->bmv_offset + p->bmv_length; 342abbf9e8aSChristoph Hellwig bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset); 343abbf9e8aSChristoph Hellwig bmv->bmv_entries++; 344f86f4037SDarrick J. Wong return 0; 345f86f4037SDarrick J. Wong } 346f86f4037SDarrick J. Wong 347abbf9e8aSChristoph Hellwig static void 348abbf9e8aSChristoph Hellwig xfs_getbmap_report_hole( 349abbf9e8aSChristoph Hellwig struct xfs_inode *ip, 350abbf9e8aSChristoph Hellwig struct getbmapx *bmv, 351232b5194SChristoph Hellwig struct kgetbmap *out, 352abbf9e8aSChristoph Hellwig int64_t bmv_end, 353abbf9e8aSChristoph Hellwig xfs_fileoff_t bno, 354abbf9e8aSChristoph Hellwig xfs_fileoff_t end) 355abbf9e8aSChristoph Hellwig { 356232b5194SChristoph Hellwig struct kgetbmap *p = out + bmv->bmv_entries; 357abbf9e8aSChristoph Hellwig 358abbf9e8aSChristoph Hellwig if (bmv->bmv_iflags & BMV_IF_NO_HOLES) 359abbf9e8aSChristoph Hellwig return; 360abbf9e8aSChristoph Hellwig 361abbf9e8aSChristoph Hellwig p->bmv_block = -1; 362abbf9e8aSChristoph Hellwig p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno); 363abbf9e8aSChristoph Hellwig p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno); 364abbf9e8aSChristoph Hellwig 365abbf9e8aSChristoph Hellwig bmv->bmv_offset = p->bmv_offset + p->bmv_length; 366abbf9e8aSChristoph Hellwig bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset); 367abbf9e8aSChristoph Hellwig bmv->bmv_entries++; 368abbf9e8aSChristoph Hellwig } 369abbf9e8aSChristoph Hellwig 370abbf9e8aSChristoph Hellwig static inline bool 371abbf9e8aSChristoph Hellwig xfs_getbmap_full( 372abbf9e8aSChristoph Hellwig struct getbmapx *bmv) 373abbf9e8aSChristoph Hellwig { 374abbf9e8aSChristoph Hellwig return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1; 375abbf9e8aSChristoph Hellwig } 376abbf9e8aSChristoph Hellwig 377abbf9e8aSChristoph Hellwig static bool 378abbf9e8aSChristoph Hellwig xfs_getbmap_next_rec( 379abbf9e8aSChristoph Hellwig struct xfs_bmbt_irec *rec, 380abbf9e8aSChristoph Hellwig xfs_fileoff_t total_end) 381abbf9e8aSChristoph Hellwig { 382abbf9e8aSChristoph Hellwig xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount; 383abbf9e8aSChristoph Hellwig 384abbf9e8aSChristoph Hellwig if (end == total_end) 385abbf9e8aSChristoph Hellwig return false; 386abbf9e8aSChristoph Hellwig 387abbf9e8aSChristoph Hellwig rec->br_startoff += rec->br_blockcount; 388abbf9e8aSChristoph Hellwig if (!isnullstartblock(rec->br_startblock) && 389abbf9e8aSChristoph Hellwig rec->br_startblock != DELAYSTARTBLOCK) 390abbf9e8aSChristoph Hellwig rec->br_startblock += rec->br_blockcount; 391abbf9e8aSChristoph Hellwig rec->br_blockcount = total_end - end; 392abbf9e8aSChristoph Hellwig return true; 393abbf9e8aSChristoph Hellwig } 394abbf9e8aSChristoph Hellwig 39568988114SDave Chinner /* 39668988114SDave Chinner * Get inode's extents as described in bmv, and format for output. 39768988114SDave Chinner * Calls formatter to fill the user's buffer until all extents 39868988114SDave Chinner * are mapped, until the passed-in bmv->bmv_count slots have 39968988114SDave Chinner * been filled, or until the formatter short-circuits the loop, 40068988114SDave Chinner * if it is tracking filled-in extents on its own. 40168988114SDave Chinner */ 40268988114SDave Chinner int /* error code */ 40368988114SDave Chinner xfs_getbmap( 404232b5194SChristoph Hellwig struct xfs_inode *ip, 40568988114SDave Chinner struct getbmapx *bmv, /* user bmap structure */ 406232b5194SChristoph Hellwig struct kgetbmap *out) 40768988114SDave Chinner { 408abbf9e8aSChristoph Hellwig struct xfs_mount *mp = ip->i_mount; 409abbf9e8aSChristoph Hellwig int iflags = bmv->bmv_iflags; 410232b5194SChristoph Hellwig int whichfork, lock, error = 0; 411abbf9e8aSChristoph Hellwig int64_t bmv_end, max_len; 412abbf9e8aSChristoph Hellwig xfs_fileoff_t bno, first_bno; 413abbf9e8aSChristoph Hellwig struct xfs_ifork *ifp; 414abbf9e8aSChristoph Hellwig struct xfs_bmbt_irec got, rec; 415abbf9e8aSChristoph Hellwig xfs_filblks_t len; 416b2b1712aSChristoph Hellwig struct xfs_iext_cursor icur; 41768988114SDave Chinner 418232b5194SChristoph Hellwig if (bmv->bmv_iflags & ~BMV_IF_VALID) 419232b5194SChristoph Hellwig return -EINVAL; 420f86f4037SDarrick J. Wong #ifndef DEBUG 421f86f4037SDarrick J. Wong /* Only allow CoW fork queries if we're debugging. */ 422f86f4037SDarrick J. Wong if (iflags & BMV_IF_COWFORK) 423f86f4037SDarrick J. Wong return -EINVAL; 424f86f4037SDarrick J. Wong #endif 425f86f4037SDarrick J. Wong if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK)) 426f86f4037SDarrick J. Wong return -EINVAL; 427f86f4037SDarrick J. Wong 428abbf9e8aSChristoph Hellwig if (bmv->bmv_length < -1) 429abbf9e8aSChristoph Hellwig return -EINVAL; 430abbf9e8aSChristoph Hellwig bmv->bmv_entries = 0; 431abbf9e8aSChristoph Hellwig if (bmv->bmv_length == 0) 432abbf9e8aSChristoph Hellwig return 0; 433abbf9e8aSChristoph Hellwig 434f86f4037SDarrick J. Wong if (iflags & BMV_IF_ATTRFORK) 435f86f4037SDarrick J. Wong whichfork = XFS_ATTR_FORK; 436f86f4037SDarrick J. Wong else if (iflags & BMV_IF_COWFORK) 437f86f4037SDarrick J. Wong whichfork = XFS_COW_FORK; 438f86f4037SDarrick J. Wong else 439f86f4037SDarrick J. Wong whichfork = XFS_DATA_FORK; 44068988114SDave Chinner 44168988114SDave Chinner xfs_ilock(ip, XFS_IOLOCK_SHARED); 442f86f4037SDarrick J. Wong switch (whichfork) { 443abbf9e8aSChristoph Hellwig case XFS_ATTR_FORK: 444001c179cSChenXiaoSong lock = xfs_ilock_attr_map_shared(ip); 445932b42c6SDarrick J. Wong if (!xfs_inode_has_attr_fork(ip)) 446001c179cSChenXiaoSong goto out_unlock_ilock; 447abbf9e8aSChristoph Hellwig 448abbf9e8aSChristoph Hellwig max_len = 1LL << 32; 449abbf9e8aSChristoph Hellwig break; 450abbf9e8aSChristoph Hellwig case XFS_COW_FORK: 451001c179cSChenXiaoSong lock = XFS_ILOCK_SHARED; 452001c179cSChenXiaoSong xfs_ilock(ip, lock); 453001c179cSChenXiaoSong 454abbf9e8aSChristoph Hellwig /* No CoW fork? Just return */ 455001c179cSChenXiaoSong if (!xfs_ifork_ptr(ip, whichfork)) 456001c179cSChenXiaoSong goto out_unlock_ilock; 457abbf9e8aSChristoph Hellwig 458abbf9e8aSChristoph Hellwig if (xfs_get_cowextsz_hint(ip)) 459abbf9e8aSChristoph Hellwig max_len = mp->m_super->s_maxbytes; 460abbf9e8aSChristoph Hellwig else 461abbf9e8aSChristoph Hellwig max_len = XFS_ISIZE(ip); 462abbf9e8aSChristoph Hellwig break; 463f86f4037SDarrick J. Wong case XFS_DATA_FORK: 464efa70be1SChristoph Hellwig if (!(iflags & BMV_IF_DELALLOC) && 46513d2c10bSChristoph Hellwig (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_disk_size)) { 4662451337dSDave Chinner error = filemap_write_and_wait(VFS_I(ip)->i_mapping); 46768988114SDave Chinner if (error) 46868988114SDave Chinner goto out_unlock_iolock; 469efa70be1SChristoph Hellwig 47068988114SDave Chinner /* 471efa70be1SChristoph Hellwig * Even after flushing the inode, there can still be 472efa70be1SChristoph Hellwig * delalloc blocks on the inode beyond EOF due to 473efa70be1SChristoph Hellwig * speculative preallocation. These are not removed 474efa70be1SChristoph Hellwig * until the release function is called or the inode 475efa70be1SChristoph Hellwig * is inactivated. Hence we cannot assert here that 476efa70be1SChristoph Hellwig * ip->i_delayed_blks == 0. 47768988114SDave Chinner */ 47868988114SDave Chinner } 47968988114SDave Chinner 480abbf9e8aSChristoph Hellwig if (xfs_get_extsz_hint(ip) || 481db07349dSChristoph Hellwig (ip->i_diflags & 482abbf9e8aSChristoph Hellwig (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))) 483abbf9e8aSChristoph Hellwig max_len = mp->m_super->s_maxbytes; 484abbf9e8aSChristoph Hellwig else 485abbf9e8aSChristoph Hellwig max_len = XFS_ISIZE(ip); 486abbf9e8aSChristoph Hellwig 487309ecac8SChristoph Hellwig lock = xfs_ilock_data_map_shared(ip); 488f86f4037SDarrick J. Wong break; 489efa70be1SChristoph Hellwig } 49068988114SDave Chinner 491001c179cSChenXiaoSong ifp = xfs_ifork_ptr(ip, whichfork); 492001c179cSChenXiaoSong 493f7e67b20SChristoph Hellwig switch (ifp->if_format) { 494abbf9e8aSChristoph Hellwig case XFS_DINODE_FMT_EXTENTS: 495abbf9e8aSChristoph Hellwig case XFS_DINODE_FMT_BTREE: 496abbf9e8aSChristoph Hellwig break; 497abbf9e8aSChristoph Hellwig case XFS_DINODE_FMT_LOCAL: 498abbf9e8aSChristoph Hellwig /* Local format inode forks report no extents. */ 49968988114SDave Chinner goto out_unlock_ilock; 500abbf9e8aSChristoph Hellwig default: 501abbf9e8aSChristoph Hellwig error = -EINVAL; 502abbf9e8aSChristoph Hellwig goto out_unlock_ilock; 50368988114SDave Chinner } 50468988114SDave Chinner 505abbf9e8aSChristoph Hellwig if (bmv->bmv_length == -1) { 506abbf9e8aSChristoph Hellwig max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len)); 507abbf9e8aSChristoph Hellwig bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset); 508abbf9e8aSChristoph Hellwig } 509abbf9e8aSChristoph Hellwig 510abbf9e8aSChristoph Hellwig bmv_end = bmv->bmv_offset + bmv->bmv_length; 511abbf9e8aSChristoph Hellwig 512abbf9e8aSChristoph Hellwig first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset); 513abbf9e8aSChristoph Hellwig len = XFS_BB_TO_FSB(mp, bmv->bmv_length); 514abbf9e8aSChristoph Hellwig 515abbf9e8aSChristoph Hellwig error = xfs_iread_extents(NULL, ip, whichfork); 516abbf9e8aSChristoph Hellwig if (error) 517abbf9e8aSChristoph Hellwig goto out_unlock_ilock; 518abbf9e8aSChristoph Hellwig 519b2b1712aSChristoph Hellwig if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) { 520abbf9e8aSChristoph Hellwig /* 521abbf9e8aSChristoph Hellwig * Report a whole-file hole if the delalloc flag is set to 522abbf9e8aSChristoph Hellwig * stay compatible with the old implementation. 523abbf9e8aSChristoph Hellwig */ 524abbf9e8aSChristoph Hellwig if (iflags & BMV_IF_DELALLOC) 525abbf9e8aSChristoph Hellwig xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno, 526abbf9e8aSChristoph Hellwig XFS_B_TO_FSB(mp, XFS_ISIZE(ip))); 527abbf9e8aSChristoph Hellwig goto out_unlock_ilock; 528abbf9e8aSChristoph Hellwig } 529abbf9e8aSChristoph Hellwig 530abbf9e8aSChristoph Hellwig while (!xfs_getbmap_full(bmv)) { 531abbf9e8aSChristoph Hellwig xfs_trim_extent(&got, first_bno, len); 532abbf9e8aSChristoph Hellwig 533abbf9e8aSChristoph Hellwig /* 534abbf9e8aSChristoph Hellwig * Report an entry for a hole if this extent doesn't directly 535abbf9e8aSChristoph Hellwig * follow the previous one. 536abbf9e8aSChristoph Hellwig */ 537abbf9e8aSChristoph Hellwig if (got.br_startoff > bno) { 538abbf9e8aSChristoph Hellwig xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno, 539abbf9e8aSChristoph Hellwig got.br_startoff); 540abbf9e8aSChristoph Hellwig if (xfs_getbmap_full(bmv)) 541abbf9e8aSChristoph Hellwig break; 542abbf9e8aSChristoph Hellwig } 543abbf9e8aSChristoph Hellwig 544abbf9e8aSChristoph Hellwig /* 545abbf9e8aSChristoph Hellwig * In order to report shared extents accurately, we report each 546abbf9e8aSChristoph Hellwig * distinct shared / unshared part of a single bmbt record with 547abbf9e8aSChristoph Hellwig * an individual getbmapx record. 548abbf9e8aSChristoph Hellwig */ 549abbf9e8aSChristoph Hellwig bno = got.br_startoff + got.br_blockcount; 550abbf9e8aSChristoph Hellwig rec = got; 55168988114SDave Chinner do { 552abbf9e8aSChristoph Hellwig error = xfs_getbmap_report_one(ip, bmv, out, bmv_end, 553abbf9e8aSChristoph Hellwig &rec); 554abbf9e8aSChristoph Hellwig if (error || xfs_getbmap_full(bmv)) 555abbf9e8aSChristoph Hellwig goto out_unlock_ilock; 556abbf9e8aSChristoph Hellwig } while (xfs_getbmap_next_rec(&rec, bno)); 55768988114SDave Chinner 558b2b1712aSChristoph Hellwig if (!xfs_iext_next_extent(ifp, &icur, &got)) { 559abbf9e8aSChristoph Hellwig xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); 56068988114SDave Chinner 561*1bba82feSDarrick J. Wong if (bmv->bmv_entries > 0) 562*1bba82feSDarrick J. Wong out[bmv->bmv_entries - 1].bmv_oflags |= 563*1bba82feSDarrick J. Wong BMV_OF_LAST; 56468988114SDave Chinner 565abbf9e8aSChristoph Hellwig if (whichfork != XFS_ATTR_FORK && bno < end && 566abbf9e8aSChristoph Hellwig !xfs_getbmap_full(bmv)) { 567abbf9e8aSChristoph Hellwig xfs_getbmap_report_hole(ip, bmv, out, bmv_end, 568abbf9e8aSChristoph Hellwig bno, end); 569abbf9e8aSChristoph Hellwig } 570abbf9e8aSChristoph Hellwig break; 57168988114SDave Chinner } 57268988114SDave Chinner 573abbf9e8aSChristoph Hellwig if (bno >= first_bno + len) 574abbf9e8aSChristoph Hellwig break; 57568988114SDave Chinner } 57668988114SDave Chinner 57768988114SDave Chinner out_unlock_ilock: 57801f4f327SChristoph Hellwig xfs_iunlock(ip, lock); 57968988114SDave Chinner out_unlock_iolock: 58068988114SDave Chinner xfs_iunlock(ip, XFS_IOLOCK_SHARED); 58168988114SDave Chinner return error; 58268988114SDave Chinner } 58368988114SDave Chinner 58468988114SDave Chinner /* 585e2ac8363SChristoph Hellwig * Dead simple method of punching delalyed allocation blocks from a range in 586e2ac8363SChristoph Hellwig * the inode. This will always punch out both the start and end blocks, even 587e2ac8363SChristoph Hellwig * if the ranges only partially overlap them, so it is up to the caller to 588e2ac8363SChristoph Hellwig * ensure that partial blocks are not passed in. 58968988114SDave Chinner */ 59068988114SDave Chinner int 59168988114SDave Chinner xfs_bmap_punch_delalloc_range( 59268988114SDave Chinner struct xfs_inode *ip, 5937348b322SDave Chinner xfs_off_t start_byte, 5947348b322SDave Chinner xfs_off_t end_byte) 59568988114SDave Chinner { 5967348b322SDave Chinner struct xfs_mount *mp = ip->i_mount; 597e2ac8363SChristoph Hellwig struct xfs_ifork *ifp = &ip->i_df; 5987348b322SDave Chinner xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, start_byte); 5997348b322SDave Chinner xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, end_byte); 600e2ac8363SChristoph Hellwig struct xfs_bmbt_irec got, del; 601e2ac8363SChristoph Hellwig struct xfs_iext_cursor icur; 60268988114SDave Chinner int error = 0; 60368988114SDave Chinner 604b2197a36SChristoph Hellwig ASSERT(!xfs_need_iread_extents(ifp)); 60568988114SDave Chinner 6060065b541SChristoph Hellwig xfs_ilock(ip, XFS_ILOCK_EXCL); 607e2ac8363SChristoph Hellwig if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got)) 608d4380177SChristoph Hellwig goto out_unlock; 609e2ac8363SChristoph Hellwig 610e2ac8363SChristoph Hellwig while (got.br_startoff + got.br_blockcount > start_fsb) { 611e2ac8363SChristoph Hellwig del = got; 6127348b322SDave Chinner xfs_trim_extent(&del, start_fsb, end_fsb - start_fsb); 613e2ac8363SChristoph Hellwig 614e2ac8363SChristoph Hellwig /* 615e2ac8363SChristoph Hellwig * A delete can push the cursor forward. Step back to the 616e2ac8363SChristoph Hellwig * previous extent on non-delalloc or extents outside the 617e2ac8363SChristoph Hellwig * target range. 618e2ac8363SChristoph Hellwig */ 619e2ac8363SChristoph Hellwig if (!del.br_blockcount || 620e2ac8363SChristoph Hellwig !isnullstartblock(del.br_startblock)) { 621e2ac8363SChristoph Hellwig if (!xfs_iext_prev_extent(ifp, &icur, &got)) 622e2ac8363SChristoph Hellwig break; 623e2ac8363SChristoph Hellwig continue; 624e2ac8363SChristoph Hellwig } 625e2ac8363SChristoph Hellwig 626e2ac8363SChristoph Hellwig error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur, 627e2ac8363SChristoph Hellwig &got, &del); 628e2ac8363SChristoph Hellwig if (error || !xfs_iext_get_extent(ifp, &icur, &got)) 629e2ac8363SChristoph Hellwig break; 630e2ac8363SChristoph Hellwig } 63168988114SDave Chinner 632d4380177SChristoph Hellwig out_unlock: 633d4380177SChristoph Hellwig xfs_iunlock(ip, XFS_ILOCK_EXCL); 63468988114SDave Chinner return error; 63568988114SDave Chinner } 636c24b5dfaSDave Chinner 637c24b5dfaSDave Chinner /* 638c24b5dfaSDave Chinner * Test whether it is appropriate to check an inode for and free post EOF 639c24b5dfaSDave Chinner * blocks. The 'force' parameter determines whether we should also consider 640c24b5dfaSDave Chinner * regular files that are marked preallocated or append-only. 641c24b5dfaSDave Chinner */ 642c24b5dfaSDave Chinner bool 6437d88329eSDarrick J. Wong xfs_can_free_eofblocks( 6447d88329eSDarrick J. Wong struct xfs_inode *ip, 6457d88329eSDarrick J. Wong bool force) 646c24b5dfaSDave Chinner { 6477d88329eSDarrick J. Wong struct xfs_bmbt_irec imap; 6487d88329eSDarrick J. Wong struct xfs_mount *mp = ip->i_mount; 6497d88329eSDarrick J. Wong xfs_fileoff_t end_fsb; 6507d88329eSDarrick J. Wong xfs_fileoff_t last_fsb; 6517d88329eSDarrick J. Wong int nimaps = 1; 6527d88329eSDarrick J. Wong int error; 6537d88329eSDarrick J. Wong 6547d88329eSDarrick J. Wong /* 6557d88329eSDarrick J. Wong * Caller must either hold the exclusive io lock; or be inactivating 6567d88329eSDarrick J. Wong * the inode, which guarantees there are no other users of the inode. 6577d88329eSDarrick J. Wong */ 6587d88329eSDarrick J. Wong ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL) || 6597d88329eSDarrick J. Wong (VFS_I(ip)->i_state & I_FREEING)); 6607d88329eSDarrick J. Wong 661c24b5dfaSDave Chinner /* prealloc/delalloc exists only on regular files */ 662c19b3b05SDave Chinner if (!S_ISREG(VFS_I(ip)->i_mode)) 663c24b5dfaSDave Chinner return false; 664c24b5dfaSDave Chinner 665c24b5dfaSDave Chinner /* 666c24b5dfaSDave Chinner * Zero sized files with no cached pages and delalloc blocks will not 667c24b5dfaSDave Chinner * have speculative prealloc/delalloc blocks to remove. 668c24b5dfaSDave Chinner */ 669c24b5dfaSDave Chinner if (VFS_I(ip)->i_size == 0 && 6702667c6f9SDave Chinner VFS_I(ip)->i_mapping->nrpages == 0 && 671c24b5dfaSDave Chinner ip->i_delayed_blks == 0) 672c24b5dfaSDave Chinner return false; 673c24b5dfaSDave Chinner 674c24b5dfaSDave Chinner /* If we haven't read in the extent list, then don't do it now. */ 675b2197a36SChristoph Hellwig if (xfs_need_iread_extents(&ip->i_df)) 676c24b5dfaSDave Chinner return false; 677c24b5dfaSDave Chinner 678c24b5dfaSDave Chinner /* 679c24b5dfaSDave Chinner * Do not free real preallocated or append-only files unless the file 680c24b5dfaSDave Chinner * has delalloc blocks and we are forced to remove them. 681c24b5dfaSDave Chinner */ 682db07349dSChristoph Hellwig if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) 683c24b5dfaSDave Chinner if (!force || ip->i_delayed_blks == 0) 684c24b5dfaSDave Chinner return false; 685c24b5dfaSDave Chinner 6867d88329eSDarrick J. Wong /* 6877d88329eSDarrick J. Wong * Do not try to free post-EOF blocks if EOF is beyond the end of the 6887d88329eSDarrick J. Wong * range supported by the page cache, because the truncation will loop 6897d88329eSDarrick J. Wong * forever. 6907d88329eSDarrick J. Wong */ 6917d88329eSDarrick J. Wong end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); 6928944c6fbSDarrick J. Wong if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) 6938944c6fbSDarrick J. Wong end_fsb = roundup_64(end_fsb, mp->m_sb.sb_rextsize); 6947d88329eSDarrick J. Wong last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 6957d88329eSDarrick J. Wong if (last_fsb <= end_fsb) 6967d88329eSDarrick J. Wong return false; 6977d88329eSDarrick J. Wong 6987d88329eSDarrick J. Wong /* 6997d88329eSDarrick J. Wong * Look up the mapping for the first block past EOF. If we can't find 7007d88329eSDarrick J. Wong * it, there's nothing to free. 7017d88329eSDarrick J. Wong */ 7027d88329eSDarrick J. Wong xfs_ilock(ip, XFS_ILOCK_SHARED); 7037d88329eSDarrick J. Wong error = xfs_bmapi_read(ip, end_fsb, last_fsb - end_fsb, &imap, &nimaps, 7047d88329eSDarrick J. Wong 0); 7057d88329eSDarrick J. Wong xfs_iunlock(ip, XFS_ILOCK_SHARED); 7067d88329eSDarrick J. Wong if (error || nimaps == 0) 7077d88329eSDarrick J. Wong return false; 7087d88329eSDarrick J. Wong 7097d88329eSDarrick J. Wong /* 7107d88329eSDarrick J. Wong * If there's a real mapping there or there are delayed allocation 7117d88329eSDarrick J. Wong * reservations, then we have post-EOF blocks to try to free. 7127d88329eSDarrick J. Wong */ 7137d88329eSDarrick J. Wong return imap.br_startblock != HOLESTARTBLOCK || ip->i_delayed_blks; 714c24b5dfaSDave Chinner } 715c24b5dfaSDave Chinner 716c24b5dfaSDave Chinner /* 7173b4683c2SBrian Foster * This is called to free any blocks beyond eof. The caller must hold 7183b4683c2SBrian Foster * IOLOCK_EXCL unless we are in the inode reclaim path and have the only 7193b4683c2SBrian Foster * reference to the inode. 720c24b5dfaSDave Chinner */ 721c24b5dfaSDave Chinner int 722c24b5dfaSDave Chinner xfs_free_eofblocks( 723a36b9261SBrian Foster struct xfs_inode *ip) 724c24b5dfaSDave Chinner { 725a36b9261SBrian Foster struct xfs_trans *tp; 726a36b9261SBrian Foster struct xfs_mount *mp = ip->i_mount; 7277d88329eSDarrick J. Wong int error; 728a36b9261SBrian Foster 7297d88329eSDarrick J. Wong /* Attach the dquots to the inode up front. */ 730c14cfccaSDarrick J. Wong error = xfs_qm_dqattach(ip); 731c24b5dfaSDave Chinner if (error) 732c24b5dfaSDave Chinner return error; 733c24b5dfaSDave Chinner 7347d88329eSDarrick J. Wong /* Wait on dio to ensure i_size has settled. */ 735e4229d6bSBrian Foster inode_dio_wait(VFS_I(ip)); 736e4229d6bSBrian Foster 7377d88329eSDarrick J. Wong error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 738c24b5dfaSDave Chinner if (error) { 73975c8c50fSDave Chinner ASSERT(xfs_is_shutdown(mp)); 740c24b5dfaSDave Chinner return error; 741c24b5dfaSDave Chinner } 742c24b5dfaSDave Chinner 743c24b5dfaSDave Chinner xfs_ilock(ip, XFS_ILOCK_EXCL); 744c24b5dfaSDave Chinner xfs_trans_ijoin(tp, ip, 0); 745c24b5dfaSDave Chinner 746c24b5dfaSDave Chinner /* 7477d88329eSDarrick J. Wong * Do not update the on-disk file size. If we update the on-disk file 7487d88329eSDarrick J. Wong * size and then the system crashes before the contents of the file are 7497d88329eSDarrick J. Wong * flushed to disk then the files may be full of holes (ie NULL files 7507d88329eSDarrick J. Wong * bug). 751c24b5dfaSDave Chinner */ 7524e529339SBrian Foster error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK, 7534e529339SBrian Foster XFS_ISIZE(ip), XFS_BMAPI_NODISCARD); 7547d88329eSDarrick J. Wong if (error) 7557d88329eSDarrick J. Wong goto err_cancel; 7567d88329eSDarrick J. Wong 7577d88329eSDarrick J. Wong error = xfs_trans_commit(tp); 7587d88329eSDarrick J. Wong if (error) 7597d88329eSDarrick J. Wong goto out_unlock; 7607d88329eSDarrick J. Wong 7617d88329eSDarrick J. Wong xfs_inode_clear_eofblocks_tag(ip); 7627d88329eSDarrick J. Wong goto out_unlock; 7637d88329eSDarrick J. Wong 7647d88329eSDarrick J. Wong err_cancel: 765c24b5dfaSDave Chinner /* 766c24b5dfaSDave Chinner * If we get an error at this point we simply don't 767c24b5dfaSDave Chinner * bother truncating the file. 768c24b5dfaSDave Chinner */ 7694906e215SChristoph Hellwig xfs_trans_cancel(tp); 7707d88329eSDarrick J. Wong out_unlock: 771c24b5dfaSDave Chinner xfs_iunlock(ip, XFS_ILOCK_EXCL); 772c24b5dfaSDave Chinner return error; 773c24b5dfaSDave Chinner } 774c24b5dfaSDave Chinner 77583aee9e4SChristoph Hellwig int 776c24b5dfaSDave Chinner xfs_alloc_file_space( 77783aee9e4SChristoph Hellwig struct xfs_inode *ip, 778c24b5dfaSDave Chinner xfs_off_t offset, 7794d1b97f9SDarrick J. Wong xfs_off_t len) 780c24b5dfaSDave Chinner { 781c24b5dfaSDave Chinner xfs_mount_t *mp = ip->i_mount; 782c24b5dfaSDave Chinner xfs_off_t count; 783c24b5dfaSDave Chinner xfs_filblks_t allocated_fsb; 784c24b5dfaSDave Chinner xfs_filblks_t allocatesize_fsb; 785c24b5dfaSDave Chinner xfs_extlen_t extsz, temp; 786c24b5dfaSDave Chinner xfs_fileoff_t startoffset_fsb; 787e093c4beSMax Reitz xfs_fileoff_t endoffset_fsb; 788c24b5dfaSDave Chinner int nimaps; 789c24b5dfaSDave Chinner int rt; 790c24b5dfaSDave Chinner xfs_trans_t *tp; 791c24b5dfaSDave Chinner xfs_bmbt_irec_t imaps[1], *imapp; 792c24b5dfaSDave Chinner int error; 793c24b5dfaSDave Chinner 794c24b5dfaSDave Chinner trace_xfs_alloc_file_space(ip); 795c24b5dfaSDave Chinner 79675c8c50fSDave Chinner if (xfs_is_shutdown(mp)) 7972451337dSDave Chinner return -EIO; 798c24b5dfaSDave Chinner 799c14cfccaSDarrick J. Wong error = xfs_qm_dqattach(ip); 800c24b5dfaSDave Chinner if (error) 801c24b5dfaSDave Chinner return error; 802c24b5dfaSDave Chinner 803c24b5dfaSDave Chinner if (len <= 0) 8042451337dSDave Chinner return -EINVAL; 805c24b5dfaSDave Chinner 806c24b5dfaSDave Chinner rt = XFS_IS_REALTIME_INODE(ip); 807c24b5dfaSDave Chinner extsz = xfs_get_extsz_hint(ip); 808c24b5dfaSDave Chinner 809c24b5dfaSDave Chinner count = len; 810c24b5dfaSDave Chinner imapp = &imaps[0]; 811c24b5dfaSDave Chinner nimaps = 1; 812c24b5dfaSDave Chinner startoffset_fsb = XFS_B_TO_FSBT(mp, offset); 813e093c4beSMax Reitz endoffset_fsb = XFS_B_TO_FSB(mp, offset + count); 814e093c4beSMax Reitz allocatesize_fsb = endoffset_fsb - startoffset_fsb; 815c24b5dfaSDave Chinner 816c24b5dfaSDave Chinner /* 817c24b5dfaSDave Chinner * Allocate file space until done or until there is an error 818c24b5dfaSDave Chinner */ 819c24b5dfaSDave Chinner while (allocatesize_fsb && !error) { 820c24b5dfaSDave Chinner xfs_fileoff_t s, e; 8213de4eb10SDarrick J. Wong unsigned int dblocks, rblocks, resblks; 822c24b5dfaSDave Chinner 823c24b5dfaSDave Chinner /* 824c24b5dfaSDave Chinner * Determine space reservations for data/realtime. 825c24b5dfaSDave Chinner */ 826c24b5dfaSDave Chinner if (unlikely(extsz)) { 827c24b5dfaSDave Chinner s = startoffset_fsb; 828c24b5dfaSDave Chinner do_div(s, extsz); 829c24b5dfaSDave Chinner s *= extsz; 830c24b5dfaSDave Chinner e = startoffset_fsb + allocatesize_fsb; 8310703a8e1SDave Chinner div_u64_rem(startoffset_fsb, extsz, &temp); 8320703a8e1SDave Chinner if (temp) 833c24b5dfaSDave Chinner e += temp; 8340703a8e1SDave Chinner div_u64_rem(e, extsz, &temp); 8350703a8e1SDave Chinner if (temp) 836c24b5dfaSDave Chinner e += extsz - temp; 837c24b5dfaSDave Chinner } else { 838c24b5dfaSDave Chinner s = 0; 839c24b5dfaSDave Chinner e = allocatesize_fsb; 840c24b5dfaSDave Chinner } 841c24b5dfaSDave Chinner 842c24b5dfaSDave Chinner /* 843c24b5dfaSDave Chinner * The transaction reservation is limited to a 32-bit block 844c24b5dfaSDave Chinner * count, hence we need to limit the number of blocks we are 845c24b5dfaSDave Chinner * trying to reserve to avoid an overflow. We can't allocate 846c24b5dfaSDave Chinner * more than @nimaps extents, and an extent is limited on disk 84795f0b95eSChandan Babu R * to XFS_BMBT_MAX_EXTLEN (21 bits), so use that to enforce the 84895f0b95eSChandan Babu R * limit. 849c24b5dfaSDave Chinner */ 85095f0b95eSChandan Babu R resblks = min_t(xfs_fileoff_t, (e - s), 85195f0b95eSChandan Babu R (XFS_MAX_BMBT_EXTLEN * nimaps)); 852c24b5dfaSDave Chinner if (unlikely(rt)) { 85302b7ee4eSDarrick J. Wong dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 85402b7ee4eSDarrick J. Wong rblocks = resblks; 855c24b5dfaSDave Chinner } else { 85602b7ee4eSDarrick J. Wong dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); 85702b7ee4eSDarrick J. Wong rblocks = 0; 858c24b5dfaSDave Chinner } 859c24b5dfaSDave Chinner 8603de4eb10SDarrick J. Wong error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, 8613de4eb10SDarrick J. Wong dblocks, rblocks, false, &tp); 862c24b5dfaSDave Chinner if (error) 8633de4eb10SDarrick J. Wong break; 864c24b5dfaSDave Chinner 865727e1acdSChandan Babu R error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, 866727e1acdSChandan Babu R XFS_IEXT_ADD_NOSPLIT_CNT); 8674f86bb4bSChandan Babu R if (error == -EFBIG) 8684f86bb4bSChandan Babu R error = xfs_iext_count_upgrade(tp, ip, 8694f86bb4bSChandan Babu R XFS_IEXT_ADD_NOSPLIT_CNT); 870727e1acdSChandan Babu R if (error) 87135b11010SDarrick J. Wong goto error; 872727e1acdSChandan Babu R 873c24b5dfaSDave Chinner error = xfs_bmapi_write(tp, ip, startoffset_fsb, 8744d1b97f9SDarrick J. Wong allocatesize_fsb, XFS_BMAPI_PREALLOC, 0, imapp, 875da781e64SBrian Foster &nimaps); 876f6106efaSEric Sandeen if (error) 87735b11010SDarrick J. Wong goto error; 878c24b5dfaSDave Chinner 8790b02c8c0SDave Chinner ip->i_diflags |= XFS_DIFLAG_PREALLOC; 8800b02c8c0SDave Chinner xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 8810b02c8c0SDave Chinner 88270393313SChristoph Hellwig error = xfs_trans_commit(tp); 883c24b5dfaSDave Chinner xfs_iunlock(ip, XFS_ILOCK_EXCL); 884f6106efaSEric Sandeen if (error) 885c24b5dfaSDave Chinner break; 886c24b5dfaSDave Chinner 887c24b5dfaSDave Chinner allocated_fsb = imapp->br_blockcount; 888c24b5dfaSDave Chinner 889c24b5dfaSDave Chinner if (nimaps == 0) { 8902451337dSDave Chinner error = -ENOSPC; 891c24b5dfaSDave Chinner break; 892c24b5dfaSDave Chinner } 893c24b5dfaSDave Chinner 894c24b5dfaSDave Chinner startoffset_fsb += allocated_fsb; 895c24b5dfaSDave Chinner allocatesize_fsb -= allocated_fsb; 896c24b5dfaSDave Chinner } 897c24b5dfaSDave Chinner 898c24b5dfaSDave Chinner return error; 899c24b5dfaSDave Chinner 90035b11010SDarrick J. Wong error: 9014906e215SChristoph Hellwig xfs_trans_cancel(tp); 902c24b5dfaSDave Chinner xfs_iunlock(ip, XFS_ILOCK_EXCL); 903c24b5dfaSDave Chinner return error; 904c24b5dfaSDave Chinner } 905c24b5dfaSDave Chinner 906bdb0d04fSChristoph Hellwig static int 907bdb0d04fSChristoph Hellwig xfs_unmap_extent( 90883aee9e4SChristoph Hellwig struct xfs_inode *ip, 909bdb0d04fSChristoph Hellwig xfs_fileoff_t startoffset_fsb, 910bdb0d04fSChristoph Hellwig xfs_filblks_t len_fsb, 911bdb0d04fSChristoph Hellwig int *done) 912c24b5dfaSDave Chinner { 913bdb0d04fSChristoph Hellwig struct xfs_mount *mp = ip->i_mount; 914bdb0d04fSChristoph Hellwig struct xfs_trans *tp; 915bdb0d04fSChristoph Hellwig uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 916bdb0d04fSChristoph Hellwig int error; 917c24b5dfaSDave Chinner 9183de4eb10SDarrick J. Wong error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0, 9193a1af6c3SDarrick J. Wong false, &tp); 920bdb0d04fSChristoph Hellwig if (error) 9213a1af6c3SDarrick J. Wong return error; 922c24b5dfaSDave Chinner 92385ef08b5SChandan Babu R error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, 92485ef08b5SChandan Babu R XFS_IEXT_PUNCH_HOLE_CNT); 9254f86bb4bSChandan Babu R if (error == -EFBIG) 9264f86bb4bSChandan Babu R error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT); 92785ef08b5SChandan Babu R if (error) 92885ef08b5SChandan Babu R goto out_trans_cancel; 92985ef08b5SChandan Babu R 9302af52842SBrian Foster error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done); 931bdb0d04fSChristoph Hellwig if (error) 932c8eac49eSBrian Foster goto out_trans_cancel; 933bdb0d04fSChristoph Hellwig 934bdb0d04fSChristoph Hellwig error = xfs_trans_commit(tp); 935bdb0d04fSChristoph Hellwig out_unlock: 936bdb0d04fSChristoph Hellwig xfs_iunlock(ip, XFS_ILOCK_EXCL); 937bdb0d04fSChristoph Hellwig return error; 938bdb0d04fSChristoph Hellwig 939bdb0d04fSChristoph Hellwig out_trans_cancel: 940bdb0d04fSChristoph Hellwig xfs_trans_cancel(tp); 941bdb0d04fSChristoph Hellwig goto out_unlock; 942bdb0d04fSChristoph Hellwig } 943bdb0d04fSChristoph Hellwig 944249bd908SDave Chinner /* Caller must first wait for the completion of any pending DIOs if required. */ 9452c307174SDave Chinner int 946bdb0d04fSChristoph Hellwig xfs_flush_unmap_range( 947bdb0d04fSChristoph Hellwig struct xfs_inode *ip, 948bdb0d04fSChristoph Hellwig xfs_off_t offset, 949bdb0d04fSChristoph Hellwig xfs_off_t len) 950bdb0d04fSChristoph Hellwig { 951bdb0d04fSChristoph Hellwig struct xfs_mount *mp = ip->i_mount; 952bdb0d04fSChristoph Hellwig struct inode *inode = VFS_I(ip); 953bdb0d04fSChristoph Hellwig xfs_off_t rounding, start, end; 954bdb0d04fSChristoph Hellwig int error; 955bdb0d04fSChristoph Hellwig 95620bd8e63SDarrick J. Wong rounding = max_t(xfs_off_t, mp->m_sb.sb_blocksize, PAGE_SIZE); 957bdb0d04fSChristoph Hellwig start = round_down(offset, rounding); 958bdb0d04fSChristoph Hellwig end = round_up(offset + len, rounding) - 1; 959bdb0d04fSChristoph Hellwig 960bdb0d04fSChristoph Hellwig error = filemap_write_and_wait_range(inode->i_mapping, start, end); 961c24b5dfaSDave Chinner if (error) 962c24b5dfaSDave Chinner return error; 963bdb0d04fSChristoph Hellwig truncate_pagecache_range(inode, start, end); 964bdb0d04fSChristoph Hellwig return 0; 965c24b5dfaSDave Chinner } 966c24b5dfaSDave Chinner 967c24b5dfaSDave Chinner int 968c24b5dfaSDave Chinner xfs_free_file_space( 969c24b5dfaSDave Chinner struct xfs_inode *ip, 970c24b5dfaSDave Chinner xfs_off_t offset, 971c24b5dfaSDave Chinner xfs_off_t len) 972c24b5dfaSDave Chinner { 973bdb0d04fSChristoph Hellwig struct xfs_mount *mp = ip->i_mount; 974c24b5dfaSDave Chinner xfs_fileoff_t startoffset_fsb; 975bdb0d04fSChristoph Hellwig xfs_fileoff_t endoffset_fsb; 9763c2bdc91SChristoph Hellwig int done = 0, error; 977c24b5dfaSDave Chinner 978c24b5dfaSDave Chinner trace_xfs_free_file_space(ip); 979c24b5dfaSDave Chinner 980c14cfccaSDarrick J. Wong error = xfs_qm_dqattach(ip); 981c24b5dfaSDave Chinner if (error) 982c24b5dfaSDave Chinner return error; 983c24b5dfaSDave Chinner 984c24b5dfaSDave Chinner if (len <= 0) /* if nothing being freed */ 985bdb0d04fSChristoph Hellwig return 0; 986bdb0d04fSChristoph Hellwig 987c24b5dfaSDave Chinner startoffset_fsb = XFS_B_TO_FSB(mp, offset); 988c24b5dfaSDave Chinner endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); 989c24b5dfaSDave Chinner 990fe341eb1SDarrick J. Wong /* We can only free complete realtime extents. */ 99125219dbfSDarrick J. Wong if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) { 99225219dbfSDarrick J. Wong startoffset_fsb = roundup_64(startoffset_fsb, 99325219dbfSDarrick J. Wong mp->m_sb.sb_rextsize); 99425219dbfSDarrick J. Wong endoffset_fsb = rounddown_64(endoffset_fsb, 99525219dbfSDarrick J. Wong mp->m_sb.sb_rextsize); 996fe341eb1SDarrick J. Wong } 997fe341eb1SDarrick J. Wong 998bdb0d04fSChristoph Hellwig /* 999daa79baeSChristoph Hellwig * Need to zero the stuff we're not freeing, on disk. 1000bdb0d04fSChristoph Hellwig */ 10013c2bdc91SChristoph Hellwig if (endoffset_fsb > startoffset_fsb) { 10023c2bdc91SChristoph Hellwig while (!done) { 1003bdb0d04fSChristoph Hellwig error = xfs_unmap_extent(ip, startoffset_fsb, 1004bdb0d04fSChristoph Hellwig endoffset_fsb - startoffset_fsb, &done); 10053c2bdc91SChristoph Hellwig if (error) 10063c2bdc91SChristoph Hellwig return error; 10073c2bdc91SChristoph Hellwig } 1008c24b5dfaSDave Chinner } 1009c24b5dfaSDave Chinner 10103c2bdc91SChristoph Hellwig /* 10113c2bdc91SChristoph Hellwig * Now that we've unmap all full blocks we'll have to zero out any 1012f1ba5fafSShiyang Ruan * partial block at the beginning and/or end. xfs_zero_range is smart 1013f5c54717SChristoph Hellwig * enough to skip any holes, including those we just created, but we 1014f5c54717SChristoph Hellwig * must take care not to zero beyond EOF and enlarge i_size. 10153c2bdc91SChristoph Hellwig */ 10163dd09d5aSCalvin Owens if (offset >= XFS_ISIZE(ip)) 10173dd09d5aSCalvin Owens return 0; 10183dd09d5aSCalvin Owens if (offset + len > XFS_ISIZE(ip)) 10193dd09d5aSCalvin Owens len = XFS_ISIZE(ip) - offset; 1020f1ba5fafSShiyang Ruan error = xfs_zero_range(ip, offset, len, NULL); 1021e53c4b59SDarrick J. Wong if (error) 1022e53c4b59SDarrick J. Wong return error; 1023e53c4b59SDarrick J. Wong 1024e53c4b59SDarrick J. Wong /* 1025e53c4b59SDarrick J. Wong * If we zeroed right up to EOF and EOF straddles a page boundary we 1026e53c4b59SDarrick J. Wong * must make sure that the post-EOF area is also zeroed because the 1027f1ba5fafSShiyang Ruan * page could be mmap'd and xfs_zero_range doesn't do that for us. 1028e53c4b59SDarrick J. Wong * Writeback of the eof page will do this, albeit clumsily. 1029e53c4b59SDarrick J. Wong */ 1030a579121fSDarrick J. Wong if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) { 1031e53c4b59SDarrick J. Wong error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 1032a579121fSDarrick J. Wong round_down(offset + len, PAGE_SIZE), LLONG_MAX); 1033e53c4b59SDarrick J. Wong } 1034e53c4b59SDarrick J. Wong 1035e53c4b59SDarrick J. Wong return error; 1036c24b5dfaSDave Chinner } 1037c24b5dfaSDave Chinner 103872c1a739Skbuild test robot static int 10394ed36c6bSChristoph Hellwig xfs_prepare_shift( 1040e1d8fb88SNamjae Jeon struct xfs_inode *ip, 10414ed36c6bSChristoph Hellwig loff_t offset) 1042e1d8fb88SNamjae Jeon { 1043d0c22041SBrian Foster struct xfs_mount *mp = ip->i_mount; 1044e1d8fb88SNamjae Jeon int error; 1045f71721d0SBrian Foster 1046f71721d0SBrian Foster /* 1047f71721d0SBrian Foster * Trim eofblocks to avoid shifting uninitialized post-eof preallocation 1048f71721d0SBrian Foster * into the accessible region of the file. 1049f71721d0SBrian Foster */ 105041b9d726SBrian Foster if (xfs_can_free_eofblocks(ip, true)) { 1051a36b9261SBrian Foster error = xfs_free_eofblocks(ip); 105241b9d726SBrian Foster if (error) 105341b9d726SBrian Foster return error; 105441b9d726SBrian Foster } 10551669a8caSDave Chinner 1056f71721d0SBrian Foster /* 1057d0c22041SBrian Foster * Shift operations must stabilize the start block offset boundary along 1058d0c22041SBrian Foster * with the full range of the operation. If we don't, a COW writeback 1059d0c22041SBrian Foster * completion could race with an insert, front merge with the start 1060d0c22041SBrian Foster * extent (after split) during the shift and corrupt the file. Start 1061d0c22041SBrian Foster * with the block just prior to the start to stabilize the boundary. 1062d0c22041SBrian Foster */ 106320bd8e63SDarrick J. Wong offset = round_down(offset, mp->m_sb.sb_blocksize); 1064d0c22041SBrian Foster if (offset) 106520bd8e63SDarrick J. Wong offset -= mp->m_sb.sb_blocksize; 1066d0c22041SBrian Foster 1067d0c22041SBrian Foster /* 1068f71721d0SBrian Foster * Writeback and invalidate cache for the remainder of the file as we're 1069a904b1caSNamjae Jeon * about to shift down every extent from offset to EOF. 1070f71721d0SBrian Foster */ 10717f9f71beSDave Chinner error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip)); 10721749d1eaSBrian Foster if (error) 10731749d1eaSBrian Foster return error; 1074e1d8fb88SNamjae Jeon 1075a904b1caSNamjae Jeon /* 10763af423b0SDarrick J. Wong * Clean out anything hanging around in the cow fork now that 10773af423b0SDarrick J. Wong * we've flushed all the dirty data out to disk to avoid having 10783af423b0SDarrick J. Wong * CoW extents at the wrong offsets. 10793af423b0SDarrick J. Wong */ 108051d62690SChristoph Hellwig if (xfs_inode_has_cow_data(ip)) { 10813af423b0SDarrick J. Wong error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF, 10823af423b0SDarrick J. Wong true); 10833af423b0SDarrick J. Wong if (error) 10843af423b0SDarrick J. Wong return error; 10853af423b0SDarrick J. Wong } 10863af423b0SDarrick J. Wong 10874ed36c6bSChristoph Hellwig return 0; 1088e1d8fb88SNamjae Jeon } 1089e1d8fb88SNamjae Jeon 1090e1d8fb88SNamjae Jeon /* 1091a904b1caSNamjae Jeon * xfs_collapse_file_space() 1092a904b1caSNamjae Jeon * This routine frees disk space and shift extent for the given file. 1093a904b1caSNamjae Jeon * The first thing we do is to free data blocks in the specified range 1094a904b1caSNamjae Jeon * by calling xfs_free_file_space(). It would also sync dirty data 1095a904b1caSNamjae Jeon * and invalidate page cache over the region on which collapse range 1096a904b1caSNamjae Jeon * is working. And Shift extent records to the left to cover a hole. 1097a904b1caSNamjae Jeon * RETURNS: 1098a904b1caSNamjae Jeon * 0 on success 1099a904b1caSNamjae Jeon * errno on error 1100a904b1caSNamjae Jeon * 1101a904b1caSNamjae Jeon */ 1102a904b1caSNamjae Jeon int 1103a904b1caSNamjae Jeon xfs_collapse_file_space( 1104a904b1caSNamjae Jeon struct xfs_inode *ip, 1105a904b1caSNamjae Jeon xfs_off_t offset, 1106a904b1caSNamjae Jeon xfs_off_t len) 1107a904b1caSNamjae Jeon { 11084ed36c6bSChristoph Hellwig struct xfs_mount *mp = ip->i_mount; 11094ed36c6bSChristoph Hellwig struct xfs_trans *tp; 1110a904b1caSNamjae Jeon int error; 11114ed36c6bSChristoph Hellwig xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len); 11124ed36c6bSChristoph Hellwig xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len); 1113ecfea3f0SChristoph Hellwig bool done = false; 1114a904b1caSNamjae Jeon 1115a904b1caSNamjae Jeon ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 11169ad1a23aSChristoph Hellwig ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 11179ad1a23aSChristoph Hellwig 1118a904b1caSNamjae Jeon trace_xfs_collapse_file_space(ip); 1119a904b1caSNamjae Jeon 1120a904b1caSNamjae Jeon error = xfs_free_file_space(ip, offset, len); 1121a904b1caSNamjae Jeon if (error) 1122a904b1caSNamjae Jeon return error; 1123a904b1caSNamjae Jeon 11244ed36c6bSChristoph Hellwig error = xfs_prepare_shift(ip, offset); 11254ed36c6bSChristoph Hellwig if (error) 11264ed36c6bSChristoph Hellwig return error; 11274ed36c6bSChristoph Hellwig 1128211683b2SBrian Foster error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp); 11294ed36c6bSChristoph Hellwig if (error) 1130211683b2SBrian Foster return error; 11314ed36c6bSChristoph Hellwig 11324ed36c6bSChristoph Hellwig xfs_ilock(ip, XFS_ILOCK_EXCL); 1133211683b2SBrian Foster xfs_trans_ijoin(tp, ip, 0); 11344ed36c6bSChristoph Hellwig 1135211683b2SBrian Foster while (!done) { 1136ecfea3f0SChristoph Hellwig error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb, 1137333f950cSBrian Foster &done); 11384ed36c6bSChristoph Hellwig if (error) 1139c8eac49eSBrian Foster goto out_trans_cancel; 1140211683b2SBrian Foster if (done) 1141211683b2SBrian Foster break; 11424ed36c6bSChristoph Hellwig 1143211683b2SBrian Foster /* finish any deferred frees and roll the transaction */ 1144211683b2SBrian Foster error = xfs_defer_finish(&tp); 1145211683b2SBrian Foster if (error) 1146211683b2SBrian Foster goto out_trans_cancel; 11474ed36c6bSChristoph Hellwig } 11484ed36c6bSChristoph Hellwig 1149211683b2SBrian Foster error = xfs_trans_commit(tp); 1150211683b2SBrian Foster xfs_iunlock(ip, XFS_ILOCK_EXCL); 11514ed36c6bSChristoph Hellwig return error; 11524ed36c6bSChristoph Hellwig 11534ed36c6bSChristoph Hellwig out_trans_cancel: 11544ed36c6bSChristoph Hellwig xfs_trans_cancel(tp); 1155211683b2SBrian Foster xfs_iunlock(ip, XFS_ILOCK_EXCL); 11564ed36c6bSChristoph Hellwig return error; 1157a904b1caSNamjae Jeon } 1158a904b1caSNamjae Jeon 1159a904b1caSNamjae Jeon /* 1160a904b1caSNamjae Jeon * xfs_insert_file_space() 1161a904b1caSNamjae Jeon * This routine create hole space by shifting extents for the given file. 1162a904b1caSNamjae Jeon * The first thing we do is to sync dirty data and invalidate page cache 1163a904b1caSNamjae Jeon * over the region on which insert range is working. And split an extent 1164a904b1caSNamjae Jeon * to two extents at given offset by calling xfs_bmap_split_extent. 1165a904b1caSNamjae Jeon * And shift all extent records which are laying between [offset, 1166a904b1caSNamjae Jeon * last allocated extent] to the right to reserve hole range. 1167a904b1caSNamjae Jeon * RETURNS: 1168a904b1caSNamjae Jeon * 0 on success 1169a904b1caSNamjae Jeon * errno on error 1170a904b1caSNamjae Jeon */ 1171a904b1caSNamjae Jeon int 1172a904b1caSNamjae Jeon xfs_insert_file_space( 1173a904b1caSNamjae Jeon struct xfs_inode *ip, 1174a904b1caSNamjae Jeon loff_t offset, 1175a904b1caSNamjae Jeon loff_t len) 1176a904b1caSNamjae Jeon { 11774ed36c6bSChristoph Hellwig struct xfs_mount *mp = ip->i_mount; 11784ed36c6bSChristoph Hellwig struct xfs_trans *tp; 11794ed36c6bSChristoph Hellwig int error; 11804ed36c6bSChristoph Hellwig xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset); 11814ed36c6bSChristoph Hellwig xfs_fileoff_t next_fsb = NULLFSBLOCK; 11824ed36c6bSChristoph Hellwig xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len); 1183ecfea3f0SChristoph Hellwig bool done = false; 11844ed36c6bSChristoph Hellwig 1185a904b1caSNamjae Jeon ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 11869ad1a23aSChristoph Hellwig ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 11879ad1a23aSChristoph Hellwig 1188a904b1caSNamjae Jeon trace_xfs_insert_file_space(ip); 1189a904b1caSNamjae Jeon 1190f62cb48eSDarrick J. Wong error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb); 1191f62cb48eSDarrick J. Wong if (error) 1192f62cb48eSDarrick J. Wong return error; 1193f62cb48eSDarrick J. Wong 11944ed36c6bSChristoph Hellwig error = xfs_prepare_shift(ip, offset); 11954ed36c6bSChristoph Hellwig if (error) 11964ed36c6bSChristoph Hellwig return error; 11974ed36c6bSChristoph Hellwig 1198b73df17eSBrian Foster error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 1199b73df17eSBrian Foster XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); 1200b73df17eSBrian Foster if (error) 1201b73df17eSBrian Foster return error; 1202b73df17eSBrian Foster 1203b73df17eSBrian Foster xfs_ilock(ip, XFS_ILOCK_EXCL); 1204dd87f87dSBrian Foster xfs_trans_ijoin(tp, ip, 0); 1205b73df17eSBrian Foster 120685ef08b5SChandan Babu R error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, 120785ef08b5SChandan Babu R XFS_IEXT_PUNCH_HOLE_CNT); 12084f86bb4bSChandan Babu R if (error == -EFBIG) 12094f86bb4bSChandan Babu R error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT); 121085ef08b5SChandan Babu R if (error) 121185ef08b5SChandan Babu R goto out_trans_cancel; 121285ef08b5SChandan Babu R 1213dd87f87dSBrian Foster /* 1214dd87f87dSBrian Foster * The extent shifting code works on extent granularity. So, if stop_fsb 1215dd87f87dSBrian Foster * is not the starting block of extent, we need to split the extent at 1216dd87f87dSBrian Foster * stop_fsb. 1217dd87f87dSBrian Foster */ 1218b73df17eSBrian Foster error = xfs_bmap_split_extent(tp, ip, stop_fsb); 1219b73df17eSBrian Foster if (error) 1220b73df17eSBrian Foster goto out_trans_cancel; 1221b73df17eSBrian Foster 1222dd87f87dSBrian Foster do { 12239c516e0eSBrian Foster error = xfs_defer_finish(&tp); 12244ed36c6bSChristoph Hellwig if (error) 1225dd87f87dSBrian Foster goto out_trans_cancel; 12264ed36c6bSChristoph Hellwig 1227ecfea3f0SChristoph Hellwig error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb, 1228333f950cSBrian Foster &done, stop_fsb); 12294ed36c6bSChristoph Hellwig if (error) 1230c8eac49eSBrian Foster goto out_trans_cancel; 1231dd87f87dSBrian Foster } while (!done); 12324ed36c6bSChristoph Hellwig 12334ed36c6bSChristoph Hellwig error = xfs_trans_commit(tp); 1234dd87f87dSBrian Foster xfs_iunlock(ip, XFS_ILOCK_EXCL); 12354ed36c6bSChristoph Hellwig return error; 12364ed36c6bSChristoph Hellwig 1237c8eac49eSBrian Foster out_trans_cancel: 12384ed36c6bSChristoph Hellwig xfs_trans_cancel(tp); 1239dd87f87dSBrian Foster xfs_iunlock(ip, XFS_ILOCK_EXCL); 12404ed36c6bSChristoph Hellwig return error; 1241a904b1caSNamjae Jeon } 1242a904b1caSNamjae Jeon 1243a904b1caSNamjae Jeon /* 1244a133d952SDave Chinner * We need to check that the format of the data fork in the temporary inode is 1245a133d952SDave Chinner * valid for the target inode before doing the swap. This is not a problem with 1246a133d952SDave Chinner * attr1 because of the fixed fork offset, but attr2 has a dynamically sized 1247a133d952SDave Chinner * data fork depending on the space the attribute fork is taking so we can get 1248a133d952SDave Chinner * invalid formats on the target inode. 1249a133d952SDave Chinner * 1250a133d952SDave Chinner * E.g. target has space for 7 extents in extent format, temp inode only has 1251a133d952SDave Chinner * space for 6. If we defragment down to 7 extents, then the tmp format is a 1252a133d952SDave Chinner * btree, but when swapped it needs to be in extent format. Hence we can't just 1253a133d952SDave Chinner * blindly swap data forks on attr2 filesystems. 1254a133d952SDave Chinner * 1255a133d952SDave Chinner * Note that we check the swap in both directions so that we don't end up with 1256a133d952SDave Chinner * a corrupt temporary inode, either. 1257a133d952SDave Chinner * 1258a133d952SDave Chinner * Note that fixing the way xfs_fsr sets up the attribute fork in the source 1259a133d952SDave Chinner * inode will prevent this situation from occurring, so all we do here is 1260a133d952SDave Chinner * reject and log the attempt. basically we are putting the responsibility on 1261a133d952SDave Chinner * userspace to get this right. 1262a133d952SDave Chinner */ 1263a133d952SDave Chinner static int 1264a133d952SDave Chinner xfs_swap_extents_check_format( 1265e06259aaSDarrick J. Wong struct xfs_inode *ip, /* target inode */ 1266e06259aaSDarrick J. Wong struct xfs_inode *tip) /* tmp inode */ 1267a133d952SDave Chinner { 1268f7e67b20SChristoph Hellwig struct xfs_ifork *ifp = &ip->i_df; 1269f7e67b20SChristoph Hellwig struct xfs_ifork *tifp = &tip->i_df; 1270a133d952SDave Chinner 1271765d3c39SDarrick J. Wong /* User/group/project quota ids must match if quotas are enforced. */ 1272765d3c39SDarrick J. Wong if (XFS_IS_QUOTA_ON(ip->i_mount) && 1273765d3c39SDarrick J. Wong (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) || 1274765d3c39SDarrick J. Wong !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) || 1275ceaf603cSChristoph Hellwig ip->i_projid != tip->i_projid)) 1276765d3c39SDarrick J. Wong return -EINVAL; 1277765d3c39SDarrick J. Wong 1278a133d952SDave Chinner /* Should never get a local format */ 1279f7e67b20SChristoph Hellwig if (ifp->if_format == XFS_DINODE_FMT_LOCAL || 1280f7e67b20SChristoph Hellwig tifp->if_format == XFS_DINODE_FMT_LOCAL) 12812451337dSDave Chinner return -EINVAL; 1282a133d952SDave Chinner 1283a133d952SDave Chinner /* 1284a133d952SDave Chinner * if the target inode has less extents that then temporary inode then 1285a133d952SDave Chinner * why did userspace call us? 1286a133d952SDave Chinner */ 1287f7e67b20SChristoph Hellwig if (ifp->if_nextents < tifp->if_nextents) 12882451337dSDave Chinner return -EINVAL; 1289a133d952SDave Chinner 1290a133d952SDave Chinner /* 12911f08af52SDarrick J. Wong * If we have to use the (expensive) rmap swap method, we can 12921f08af52SDarrick J. Wong * handle any number of extents and any format. 12931f08af52SDarrick J. Wong */ 129438c26bfdSDave Chinner if (xfs_has_rmapbt(ip->i_mount)) 12951f08af52SDarrick J. Wong return 0; 12961f08af52SDarrick J. Wong 12971f08af52SDarrick J. Wong /* 1298a133d952SDave Chinner * if the target inode is in extent form and the temp inode is in btree 1299a133d952SDave Chinner * form then we will end up with the target inode in the wrong format 1300a133d952SDave Chinner * as we already know there are less extents in the temp inode. 1301a133d952SDave Chinner */ 1302f7e67b20SChristoph Hellwig if (ifp->if_format == XFS_DINODE_FMT_EXTENTS && 1303f7e67b20SChristoph Hellwig tifp->if_format == XFS_DINODE_FMT_BTREE) 13042451337dSDave Chinner return -EINVAL; 1305a133d952SDave Chinner 1306a133d952SDave Chinner /* Check temp in extent form to max in target */ 1307f7e67b20SChristoph Hellwig if (tifp->if_format == XFS_DINODE_FMT_EXTENTS && 1308f7e67b20SChristoph Hellwig tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) 13092451337dSDave Chinner return -EINVAL; 1310a133d952SDave Chinner 1311a133d952SDave Chinner /* Check target in extent form to max in temp */ 1312f7e67b20SChristoph Hellwig if (ifp->if_format == XFS_DINODE_FMT_EXTENTS && 1313f7e67b20SChristoph Hellwig ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) 13142451337dSDave Chinner return -EINVAL; 1315a133d952SDave Chinner 1316a133d952SDave Chinner /* 1317a133d952SDave Chinner * If we are in a btree format, check that the temp root block will fit 1318a133d952SDave Chinner * in the target and that it has enough extents to be in btree format 1319a133d952SDave Chinner * in the target. 1320a133d952SDave Chinner * 1321a133d952SDave Chinner * Note that we have to be careful to allow btree->extent conversions 1322a133d952SDave Chinner * (a common defrag case) which will occur when the temp inode is in 1323a133d952SDave Chinner * extent format... 1324a133d952SDave Chinner */ 1325f7e67b20SChristoph Hellwig if (tifp->if_format == XFS_DINODE_FMT_BTREE) { 1326932b42c6SDarrick J. Wong if (xfs_inode_has_attr_fork(ip) && 1327c01147d9SDarrick J. Wong XFS_BMAP_BMDR_SPACE(tifp->if_broot) > xfs_inode_fork_boff(ip)) 13282451337dSDave Chinner return -EINVAL; 1329f7e67b20SChristoph Hellwig if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) 13302451337dSDave Chinner return -EINVAL; 1331a133d952SDave Chinner } 1332a133d952SDave Chinner 1333a133d952SDave Chinner /* Reciprocal target->temp btree format checks */ 1334f7e67b20SChristoph Hellwig if (ifp->if_format == XFS_DINODE_FMT_BTREE) { 1335932b42c6SDarrick J. Wong if (xfs_inode_has_attr_fork(tip) && 1336c01147d9SDarrick J. Wong XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > xfs_inode_fork_boff(tip)) 13372451337dSDave Chinner return -EINVAL; 1338f7e67b20SChristoph Hellwig if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) 13392451337dSDave Chinner return -EINVAL; 1340a133d952SDave Chinner } 1341a133d952SDave Chinner 1342a133d952SDave Chinner return 0; 1343a133d952SDave Chinner } 1344a133d952SDave Chinner 13457abbb8f9SDave Chinner static int 13464ef897a2SDave Chinner xfs_swap_extent_flush( 13474ef897a2SDave Chinner struct xfs_inode *ip) 13484ef897a2SDave Chinner { 13494ef897a2SDave Chinner int error; 13504ef897a2SDave Chinner 13514ef897a2SDave Chinner error = filemap_write_and_wait(VFS_I(ip)->i_mapping); 13524ef897a2SDave Chinner if (error) 13534ef897a2SDave Chinner return error; 13544ef897a2SDave Chinner truncate_pagecache_range(VFS_I(ip), 0, -1); 13554ef897a2SDave Chinner 13564ef897a2SDave Chinner /* Verify O_DIRECT for ftmp */ 13574ef897a2SDave Chinner if (VFS_I(ip)->i_mapping->nrpages) 13584ef897a2SDave Chinner return -EINVAL; 13594ef897a2SDave Chinner return 0; 13604ef897a2SDave Chinner } 13614ef897a2SDave Chinner 13621f08af52SDarrick J. Wong /* 13631f08af52SDarrick J. Wong * Move extents from one file to another, when rmap is enabled. 13641f08af52SDarrick J. Wong */ 13651f08af52SDarrick J. Wong STATIC int 13661f08af52SDarrick J. Wong xfs_swap_extent_rmap( 13671f08af52SDarrick J. Wong struct xfs_trans **tpp, 13681f08af52SDarrick J. Wong struct xfs_inode *ip, 13691f08af52SDarrick J. Wong struct xfs_inode *tip) 13701f08af52SDarrick J. Wong { 13717a7943c7SBrian Foster struct xfs_trans *tp = *tpp; 13721f08af52SDarrick J. Wong struct xfs_bmbt_irec irec; 13731f08af52SDarrick J. Wong struct xfs_bmbt_irec uirec; 13741f08af52SDarrick J. Wong struct xfs_bmbt_irec tirec; 13751f08af52SDarrick J. Wong xfs_fileoff_t offset_fsb; 13761f08af52SDarrick J. Wong xfs_fileoff_t end_fsb; 13771f08af52SDarrick J. Wong xfs_filblks_t count_fsb; 13781f08af52SDarrick J. Wong int error; 13791f08af52SDarrick J. Wong xfs_filblks_t ilen; 13801f08af52SDarrick J. Wong xfs_filblks_t rlen; 13811f08af52SDarrick J. Wong int nimaps; 1382c8ce540dSDarrick J. Wong uint64_t tip_flags2; 13831f08af52SDarrick J. Wong 13841f08af52SDarrick J. Wong /* 13851f08af52SDarrick J. Wong * If the source file has shared blocks, we must flag the donor 13861f08af52SDarrick J. Wong * file as having shared blocks so that we get the shared-block 13871f08af52SDarrick J. Wong * rmap functions when we go to fix up the rmaps. The flags 13881f08af52SDarrick J. Wong * will be switch for reals later. 13891f08af52SDarrick J. Wong */ 13903e09ab8fSChristoph Hellwig tip_flags2 = tip->i_diflags2; 13913e09ab8fSChristoph Hellwig if (ip->i_diflags2 & XFS_DIFLAG2_REFLINK) 13923e09ab8fSChristoph Hellwig tip->i_diflags2 |= XFS_DIFLAG2_REFLINK; 13931f08af52SDarrick J. Wong 13941f08af52SDarrick J. Wong offset_fsb = 0; 13951f08af52SDarrick J. Wong end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip))); 13961f08af52SDarrick J. Wong count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb); 13971f08af52SDarrick J. Wong 13981f08af52SDarrick J. Wong while (count_fsb) { 13991f08af52SDarrick J. Wong /* Read extent from the donor file */ 14001f08af52SDarrick J. Wong nimaps = 1; 14011f08af52SDarrick J. Wong error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec, 14021f08af52SDarrick J. Wong &nimaps, 0); 14031f08af52SDarrick J. Wong if (error) 14041f08af52SDarrick J. Wong goto out; 14051f08af52SDarrick J. Wong ASSERT(nimaps == 1); 14061f08af52SDarrick J. Wong ASSERT(tirec.br_startblock != DELAYSTARTBLOCK); 14071f08af52SDarrick J. Wong 14081f08af52SDarrick J. Wong trace_xfs_swap_extent_rmap_remap(tip, &tirec); 14091f08af52SDarrick J. Wong ilen = tirec.br_blockcount; 14101f08af52SDarrick J. Wong 14111f08af52SDarrick J. Wong /* Unmap the old blocks in the source file. */ 14121f08af52SDarrick J. Wong while (tirec.br_blockcount) { 1413692b6cddSDave Chinner ASSERT(tp->t_highest_agno == NULLAGNUMBER); 14141f08af52SDarrick J. Wong trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec); 14151f08af52SDarrick J. Wong 14161f08af52SDarrick J. Wong /* Read extent from the source file */ 14171f08af52SDarrick J. Wong nimaps = 1; 14181f08af52SDarrick J. Wong error = xfs_bmapi_read(ip, tirec.br_startoff, 14191f08af52SDarrick J. Wong tirec.br_blockcount, &irec, 14201f08af52SDarrick J. Wong &nimaps, 0); 14211f08af52SDarrick J. Wong if (error) 1422d5a2e289SBrian Foster goto out; 14231f08af52SDarrick J. Wong ASSERT(nimaps == 1); 14241f08af52SDarrick J. Wong ASSERT(tirec.br_startoff == irec.br_startoff); 14251f08af52SDarrick J. Wong trace_xfs_swap_extent_rmap_remap_piece(ip, &irec); 14261f08af52SDarrick J. Wong 14271f08af52SDarrick J. Wong /* Trim the extent. */ 14281f08af52SDarrick J. Wong uirec = tirec; 14291f08af52SDarrick J. Wong uirec.br_blockcount = rlen = min_t(xfs_filblks_t, 14301f08af52SDarrick J. Wong tirec.br_blockcount, 14311f08af52SDarrick J. Wong irec.br_blockcount); 14321f08af52SDarrick J. Wong trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec); 14331f08af52SDarrick J. Wong 1434bcc561f2SChandan Babu R if (xfs_bmap_is_real_extent(&uirec)) { 1435bcc561f2SChandan Babu R error = xfs_iext_count_may_overflow(ip, 1436bcc561f2SChandan Babu R XFS_DATA_FORK, 1437bcc561f2SChandan Babu R XFS_IEXT_SWAP_RMAP_CNT); 14384f86bb4bSChandan Babu R if (error == -EFBIG) 14394f86bb4bSChandan Babu R error = xfs_iext_count_upgrade(tp, ip, 14404f86bb4bSChandan Babu R XFS_IEXT_SWAP_RMAP_CNT); 1441bcc561f2SChandan Babu R if (error) 1442bcc561f2SChandan Babu R goto out; 1443bcc561f2SChandan Babu R } 1444bcc561f2SChandan Babu R 1445bcc561f2SChandan Babu R if (xfs_bmap_is_real_extent(&irec)) { 1446bcc561f2SChandan Babu R error = xfs_iext_count_may_overflow(tip, 1447bcc561f2SChandan Babu R XFS_DATA_FORK, 1448bcc561f2SChandan Babu R XFS_IEXT_SWAP_RMAP_CNT); 14494f86bb4bSChandan Babu R if (error == -EFBIG) 14504f86bb4bSChandan Babu R error = xfs_iext_count_upgrade(tp, ip, 14514f86bb4bSChandan Babu R XFS_IEXT_SWAP_RMAP_CNT); 1452bcc561f2SChandan Babu R if (error) 1453bcc561f2SChandan Babu R goto out; 1454bcc561f2SChandan Babu R } 1455bcc561f2SChandan Babu R 14561f08af52SDarrick J. Wong /* Remove the mapping from the donor file. */ 14573e08f42aSDarrick J. Wong xfs_bmap_unmap_extent(tp, tip, &uirec); 14581f08af52SDarrick J. Wong 14591f08af52SDarrick J. Wong /* Remove the mapping from the source file. */ 14603e08f42aSDarrick J. Wong xfs_bmap_unmap_extent(tp, ip, &irec); 14611f08af52SDarrick J. Wong 14621f08af52SDarrick J. Wong /* Map the donor file's blocks into the source file. */ 14633e08f42aSDarrick J. Wong xfs_bmap_map_extent(tp, ip, &uirec); 14641f08af52SDarrick J. Wong 14651f08af52SDarrick J. Wong /* Map the source file's blocks into the donor file. */ 14663e08f42aSDarrick J. Wong xfs_bmap_map_extent(tp, tip, &irec); 14671f08af52SDarrick J. Wong 14689e28a242SBrian Foster error = xfs_defer_finish(tpp); 14697a7943c7SBrian Foster tp = *tpp; 14701f08af52SDarrick J. Wong if (error) 14719b1f4e98SBrian Foster goto out; 14721f08af52SDarrick J. Wong 14731f08af52SDarrick J. Wong tirec.br_startoff += rlen; 14741f08af52SDarrick J. Wong if (tirec.br_startblock != HOLESTARTBLOCK && 14751f08af52SDarrick J. Wong tirec.br_startblock != DELAYSTARTBLOCK) 14761f08af52SDarrick J. Wong tirec.br_startblock += rlen; 14771f08af52SDarrick J. Wong tirec.br_blockcount -= rlen; 14781f08af52SDarrick J. Wong } 14791f08af52SDarrick J. Wong 14801f08af52SDarrick J. Wong /* Roll on... */ 14811f08af52SDarrick J. Wong count_fsb -= ilen; 14821f08af52SDarrick J. Wong offset_fsb += ilen; 14831f08af52SDarrick J. Wong } 14841f08af52SDarrick J. Wong 14853e09ab8fSChristoph Hellwig tip->i_diflags2 = tip_flags2; 14861f08af52SDarrick J. Wong return 0; 14871f08af52SDarrick J. Wong 14881f08af52SDarrick J. Wong out: 14891f08af52SDarrick J. Wong trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_); 14903e09ab8fSChristoph Hellwig tip->i_diflags2 = tip_flags2; 14911f08af52SDarrick J. Wong return error; 14921f08af52SDarrick J. Wong } 14931f08af52SDarrick J. Wong 149439aff5fdSDarrick J. Wong /* Swap the extents of two files by swapping data forks. */ 149539aff5fdSDarrick J. Wong STATIC int 149639aff5fdSDarrick J. Wong xfs_swap_extent_forks( 149739aff5fdSDarrick J. Wong struct xfs_trans *tp, 149839aff5fdSDarrick J. Wong struct xfs_inode *ip, 149939aff5fdSDarrick J. Wong struct xfs_inode *tip, 150039aff5fdSDarrick J. Wong int *src_log_flags, 150139aff5fdSDarrick J. Wong int *target_log_flags) 150239aff5fdSDarrick J. Wong { 1503e7f5d5caSDarrick J. Wong xfs_filblks_t aforkblks = 0; 1504e7f5d5caSDarrick J. Wong xfs_filblks_t taforkblks = 0; 1505e7f5d5caSDarrick J. Wong xfs_extnum_t junk; 1506c8ce540dSDarrick J. Wong uint64_t tmp; 150739aff5fdSDarrick J. Wong int error; 150839aff5fdSDarrick J. Wong 150939aff5fdSDarrick J. Wong /* 151039aff5fdSDarrick J. Wong * Count the number of extended attribute blocks 151139aff5fdSDarrick J. Wong */ 1512932b42c6SDarrick J. Wong if (xfs_inode_has_attr_fork(ip) && ip->i_af.if_nextents > 0 && 15132ed5b09bSDarrick J. Wong ip->i_af.if_format != XFS_DINODE_FMT_LOCAL) { 1514e7f5d5caSDarrick J. Wong error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk, 151539aff5fdSDarrick J. Wong &aforkblks); 151639aff5fdSDarrick J. Wong if (error) 151739aff5fdSDarrick J. Wong return error; 151839aff5fdSDarrick J. Wong } 1519932b42c6SDarrick J. Wong if (xfs_inode_has_attr_fork(tip) && tip->i_af.if_nextents > 0 && 15202ed5b09bSDarrick J. Wong tip->i_af.if_format != XFS_DINODE_FMT_LOCAL) { 1521e7f5d5caSDarrick J. Wong error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk, 152239aff5fdSDarrick J. Wong &taforkblks); 152339aff5fdSDarrick J. Wong if (error) 152439aff5fdSDarrick J. Wong return error; 152539aff5fdSDarrick J. Wong } 152639aff5fdSDarrick J. Wong 152739aff5fdSDarrick J. Wong /* 15286fb10d6dSBrian Foster * Btree format (v3) inodes have the inode number stamped in the bmbt 15296fb10d6dSBrian Foster * block headers. We can't start changing the bmbt blocks until the 15306fb10d6dSBrian Foster * inode owner change is logged so recovery does the right thing in the 15316fb10d6dSBrian Foster * event of a crash. Set the owner change log flags now and leave the 15326fb10d6dSBrian Foster * bmbt scan as the last step. 153339aff5fdSDarrick J. Wong */ 153438c26bfdSDave Chinner if (xfs_has_v3inodes(ip->i_mount)) { 1535f7e67b20SChristoph Hellwig if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE) 153639aff5fdSDarrick J. Wong (*target_log_flags) |= XFS_ILOG_DOWNER; 1537f7e67b20SChristoph Hellwig if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE) 153839aff5fdSDarrick J. Wong (*src_log_flags) |= XFS_ILOG_DOWNER; 15396471e9c5SChristoph Hellwig } 154039aff5fdSDarrick J. Wong 154139aff5fdSDarrick J. Wong /* 154239aff5fdSDarrick J. Wong * Swap the data forks of the inodes 154339aff5fdSDarrick J. Wong */ 1544897992b7SGustavo A. R. Silva swap(ip->i_df, tip->i_df); 154539aff5fdSDarrick J. Wong 154639aff5fdSDarrick J. Wong /* 154739aff5fdSDarrick J. Wong * Fix the on-disk inode values 154839aff5fdSDarrick J. Wong */ 15496e73a545SChristoph Hellwig tmp = (uint64_t)ip->i_nblocks; 15506e73a545SChristoph Hellwig ip->i_nblocks = tip->i_nblocks - taforkblks + aforkblks; 15516e73a545SChristoph Hellwig tip->i_nblocks = tmp + taforkblks - aforkblks; 155239aff5fdSDarrick J. Wong 155339aff5fdSDarrick J. Wong /* 155439aff5fdSDarrick J. Wong * The extents in the source inode could still contain speculative 155539aff5fdSDarrick J. Wong * preallocation beyond EOF (e.g. the file is open but not modified 155639aff5fdSDarrick J. Wong * while defrag is in progress). In that case, we need to copy over the 155739aff5fdSDarrick J. Wong * number of delalloc blocks the data fork in the source inode is 155839aff5fdSDarrick J. Wong * tracking beyond EOF so that when the fork is truncated away when the 155939aff5fdSDarrick J. Wong * temporary inode is unlinked we don't underrun the i_delayed_blks 156039aff5fdSDarrick J. Wong * counter on that inode. 156139aff5fdSDarrick J. Wong */ 156239aff5fdSDarrick J. Wong ASSERT(tip->i_delayed_blks == 0); 156339aff5fdSDarrick J. Wong tip->i_delayed_blks = ip->i_delayed_blks; 156439aff5fdSDarrick J. Wong ip->i_delayed_blks = 0; 156539aff5fdSDarrick J. Wong 1566f7e67b20SChristoph Hellwig switch (ip->i_df.if_format) { 156739aff5fdSDarrick J. Wong case XFS_DINODE_FMT_EXTENTS: 156839aff5fdSDarrick J. Wong (*src_log_flags) |= XFS_ILOG_DEXT; 156939aff5fdSDarrick J. Wong break; 157039aff5fdSDarrick J. Wong case XFS_DINODE_FMT_BTREE: 157138c26bfdSDave Chinner ASSERT(!xfs_has_v3inodes(ip->i_mount) || 157239aff5fdSDarrick J. Wong (*src_log_flags & XFS_ILOG_DOWNER)); 157339aff5fdSDarrick J. Wong (*src_log_flags) |= XFS_ILOG_DBROOT; 157439aff5fdSDarrick J. Wong break; 157539aff5fdSDarrick J. Wong } 157639aff5fdSDarrick J. Wong 1577f7e67b20SChristoph Hellwig switch (tip->i_df.if_format) { 157839aff5fdSDarrick J. Wong case XFS_DINODE_FMT_EXTENTS: 157939aff5fdSDarrick J. Wong (*target_log_flags) |= XFS_ILOG_DEXT; 158039aff5fdSDarrick J. Wong break; 158139aff5fdSDarrick J. Wong case XFS_DINODE_FMT_BTREE: 158239aff5fdSDarrick J. Wong (*target_log_flags) |= XFS_ILOG_DBROOT; 158338c26bfdSDave Chinner ASSERT(!xfs_has_v3inodes(ip->i_mount) || 158439aff5fdSDarrick J. Wong (*target_log_flags & XFS_ILOG_DOWNER)); 158539aff5fdSDarrick J. Wong break; 158639aff5fdSDarrick J. Wong } 158739aff5fdSDarrick J. Wong 158839aff5fdSDarrick J. Wong return 0; 158939aff5fdSDarrick J. Wong } 159039aff5fdSDarrick J. Wong 15912dd3d709SBrian Foster /* 15922dd3d709SBrian Foster * Fix up the owners of the bmbt blocks to refer to the current inode. The 15932dd3d709SBrian Foster * change owner scan attempts to order all modified buffers in the current 15942dd3d709SBrian Foster * transaction. In the event of ordered buffer failure, the offending buffer is 15952dd3d709SBrian Foster * physically logged as a fallback and the scan returns -EAGAIN. We must roll 15962dd3d709SBrian Foster * the transaction in this case to replenish the fallback log reservation and 15972dd3d709SBrian Foster * restart the scan. This process repeats until the scan completes. 15982dd3d709SBrian Foster */ 15992dd3d709SBrian Foster static int 16002dd3d709SBrian Foster xfs_swap_change_owner( 16012dd3d709SBrian Foster struct xfs_trans **tpp, 16022dd3d709SBrian Foster struct xfs_inode *ip, 16032dd3d709SBrian Foster struct xfs_inode *tmpip) 16042dd3d709SBrian Foster { 16052dd3d709SBrian Foster int error; 16062dd3d709SBrian Foster struct xfs_trans *tp = *tpp; 16072dd3d709SBrian Foster 16082dd3d709SBrian Foster do { 16092dd3d709SBrian Foster error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino, 16102dd3d709SBrian Foster NULL); 16112dd3d709SBrian Foster /* success or fatal error */ 16122dd3d709SBrian Foster if (error != -EAGAIN) 16132dd3d709SBrian Foster break; 16142dd3d709SBrian Foster 16152dd3d709SBrian Foster error = xfs_trans_roll(tpp); 16162dd3d709SBrian Foster if (error) 16172dd3d709SBrian Foster break; 16182dd3d709SBrian Foster tp = *tpp; 16192dd3d709SBrian Foster 16202dd3d709SBrian Foster /* 16212dd3d709SBrian Foster * Redirty both inodes so they can relog and keep the log tail 16222dd3d709SBrian Foster * moving forward. 16232dd3d709SBrian Foster */ 16242dd3d709SBrian Foster xfs_trans_ijoin(tp, ip, 0); 16252dd3d709SBrian Foster xfs_trans_ijoin(tp, tmpip, 0); 16262dd3d709SBrian Foster xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 16272dd3d709SBrian Foster xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE); 16282dd3d709SBrian Foster } while (true); 16292dd3d709SBrian Foster 16302dd3d709SBrian Foster return error; 16312dd3d709SBrian Foster } 16322dd3d709SBrian Foster 16334ef897a2SDave Chinner int 1634a133d952SDave Chinner xfs_swap_extents( 1635e06259aaSDarrick J. Wong struct xfs_inode *ip, /* target inode */ 1636e06259aaSDarrick J. Wong struct xfs_inode *tip, /* tmp inode */ 1637e06259aaSDarrick J. Wong struct xfs_swapext *sxp) 1638a133d952SDave Chinner { 1639e06259aaSDarrick J. Wong struct xfs_mount *mp = ip->i_mount; 1640e06259aaSDarrick J. Wong struct xfs_trans *tp; 1641e06259aaSDarrick J. Wong struct xfs_bstat *sbp = &sxp->sx_stat; 1642a133d952SDave Chinner int src_log_flags, target_log_flags; 1643a133d952SDave Chinner int error = 0; 1644c8ce540dSDarrick J. Wong uint64_t f; 16452dd3d709SBrian Foster int resblks = 0; 1646f74681baSBrian Foster unsigned int flags = 0; 1647a133d952SDave Chinner 1648a133d952SDave Chinner /* 1649723cac48SDave Chinner * Lock the inodes against other IO, page faults and truncate to 1650723cac48SDave Chinner * begin with. Then we can ensure the inodes are flushed and have no 1651723cac48SDave Chinner * page cache safely. Once we have done this we can take the ilocks and 1652723cac48SDave Chinner * do the rest of the checks. 1653a133d952SDave Chinner */ 165465523218SChristoph Hellwig lock_two_nondirectories(VFS_I(ip), VFS_I(tip)); 1655d2c292d8SJan Kara filemap_invalidate_lock_two(VFS_I(ip)->i_mapping, 1656d2c292d8SJan Kara VFS_I(tip)->i_mapping); 1657a133d952SDave Chinner 1658a133d952SDave Chinner /* Verify that both files have the same format */ 1659c19b3b05SDave Chinner if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) { 16602451337dSDave Chinner error = -EINVAL; 1661a133d952SDave Chinner goto out_unlock; 1662a133d952SDave Chinner } 1663a133d952SDave Chinner 1664a133d952SDave Chinner /* Verify both files are either real-time or non-realtime */ 1665a133d952SDave Chinner if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) { 16662451337dSDave Chinner error = -EINVAL; 1667a133d952SDave Chinner goto out_unlock; 1668a133d952SDave Chinner } 1669a133d952SDave Chinner 16702713fefaSDarrick J. Wong error = xfs_qm_dqattach(ip); 16712713fefaSDarrick J. Wong if (error) 16722713fefaSDarrick J. Wong goto out_unlock; 16732713fefaSDarrick J. Wong 16742713fefaSDarrick J. Wong error = xfs_qm_dqattach(tip); 16752713fefaSDarrick J. Wong if (error) 16762713fefaSDarrick J. Wong goto out_unlock; 16772713fefaSDarrick J. Wong 16784ef897a2SDave Chinner error = xfs_swap_extent_flush(ip); 1679a133d952SDave Chinner if (error) 1680a133d952SDave Chinner goto out_unlock; 16814ef897a2SDave Chinner error = xfs_swap_extent_flush(tip); 16824ef897a2SDave Chinner if (error) 16834ef897a2SDave Chinner goto out_unlock; 1684a133d952SDave Chinner 168596987eeaSChristoph Hellwig if (xfs_inode_has_cow_data(tip)) { 168696987eeaSChristoph Hellwig error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true); 168796987eeaSChristoph Hellwig if (error) 16888bc3b5e4SDarrick J. Wong goto out_unlock; 168996987eeaSChristoph Hellwig } 169096987eeaSChristoph Hellwig 16911f08af52SDarrick J. Wong /* 16921f08af52SDarrick J. Wong * Extent "swapping" with rmap requires a permanent reservation and 16931f08af52SDarrick J. Wong * a block reservation because it's really just a remap operation 16941f08af52SDarrick J. Wong * performed with log redo items! 16951f08af52SDarrick J. Wong */ 169638c26bfdSDave Chinner if (xfs_has_rmapbt(mp)) { 1697b3fed434SBrian Foster int w = XFS_DATA_FORK; 1698daf83964SChristoph Hellwig uint32_t ipnext = ip->i_df.if_nextents; 1699daf83964SChristoph Hellwig uint32_t tipnext = tip->i_df.if_nextents; 1700b3fed434SBrian Foster 17011f08af52SDarrick J. Wong /* 1702b3fed434SBrian Foster * Conceptually this shouldn't affect the shape of either bmbt, 1703b3fed434SBrian Foster * but since we atomically move extents one by one, we reserve 1704b3fed434SBrian Foster * enough space to rebuild both trees. 17051f08af52SDarrick J. Wong */ 1706b3fed434SBrian Foster resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w); 1707b3fed434SBrian Foster resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w); 1708b3fed434SBrian Foster 1709b3fed434SBrian Foster /* 1710f74681baSBrian Foster * If either inode straddles a bmapbt block allocation boundary, 1711f74681baSBrian Foster * the rmapbt algorithm triggers repeated allocs and frees as 1712f74681baSBrian Foster * extents are remapped. This can exhaust the block reservation 1713f74681baSBrian Foster * prematurely and cause shutdown. Return freed blocks to the 1714f74681baSBrian Foster * transaction reservation to counter this behavior. 1715b3fed434SBrian Foster */ 1716f74681baSBrian Foster flags |= XFS_TRANS_RES_FDBLKS; 17172dd3d709SBrian Foster } 1718f74681baSBrian Foster error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags, 1719f74681baSBrian Foster &tp); 1720253f4911SChristoph Hellwig if (error) 1721a133d952SDave Chinner goto out_unlock; 1722723cac48SDave Chinner 1723723cac48SDave Chinner /* 1724723cac48SDave Chinner * Lock and join the inodes to the tansaction so that transaction commit 1725723cac48SDave Chinner * or cancel will unlock the inodes from this point onwards. 1726723cac48SDave Chinner */ 17277c2d238aSDarrick J. Wong xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL); 172839aff5fdSDarrick J. Wong xfs_trans_ijoin(tp, ip, 0); 172939aff5fdSDarrick J. Wong xfs_trans_ijoin(tp, tip, 0); 1730723cac48SDave Chinner 1731a133d952SDave Chinner 1732a133d952SDave Chinner /* Verify all data are being swapped */ 1733a133d952SDave Chinner if (sxp->sx_offset != 0 || 173413d2c10bSChristoph Hellwig sxp->sx_length != ip->i_disk_size || 173513d2c10bSChristoph Hellwig sxp->sx_length != tip->i_disk_size) { 17362451337dSDave Chinner error = -EFAULT; 17374ef897a2SDave Chinner goto out_trans_cancel; 1738a133d952SDave Chinner } 1739a133d952SDave Chinner 1740a133d952SDave Chinner trace_xfs_swap_extent_before(ip, 0); 1741a133d952SDave Chinner trace_xfs_swap_extent_before(tip, 1); 1742a133d952SDave Chinner 1743a133d952SDave Chinner /* check inode formats now that data is flushed */ 1744a133d952SDave Chinner error = xfs_swap_extents_check_format(ip, tip); 1745a133d952SDave Chinner if (error) { 1746a133d952SDave Chinner xfs_notice(mp, 1747a133d952SDave Chinner "%s: inode 0x%llx format is incompatible for exchanging.", 1748a133d952SDave Chinner __func__, ip->i_ino); 17494ef897a2SDave Chinner goto out_trans_cancel; 1750a133d952SDave Chinner } 1751a133d952SDave Chinner 1752a133d952SDave Chinner /* 1753a133d952SDave Chinner * Compare the current change & modify times with that 1754a133d952SDave Chinner * passed in. If they differ, we abort this swap. 1755a133d952SDave Chinner * This is the mechanism used to ensure the calling 1756a133d952SDave Chinner * process that the file was not changed out from 1757a133d952SDave Chinner * under it. 1758a133d952SDave Chinner */ 1759a133d952SDave Chinner if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) || 1760a133d952SDave Chinner (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) || 1761a133d952SDave Chinner (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) || 1762a133d952SDave Chinner (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) { 17632451337dSDave Chinner error = -EBUSY; 176481217683SDave Chinner goto out_trans_cancel; 1765a133d952SDave Chinner } 1766a133d952SDave Chinner 176721b5c978SDave Chinner /* 176821b5c978SDave Chinner * Note the trickiness in setting the log flags - we set the owner log 176921b5c978SDave Chinner * flag on the opposite inode (i.e. the inode we are setting the new 177021b5c978SDave Chinner * owner to be) because once we swap the forks and log that, log 177121b5c978SDave Chinner * recovery is going to see the fork as owned by the swapped inode, 177221b5c978SDave Chinner * not the pre-swapped inodes. 177321b5c978SDave Chinner */ 177421b5c978SDave Chinner src_log_flags = XFS_ILOG_CORE; 177521b5c978SDave Chinner target_log_flags = XFS_ILOG_CORE; 177639aff5fdSDarrick J. Wong 177738c26bfdSDave Chinner if (xfs_has_rmapbt(mp)) 17781f08af52SDarrick J. Wong error = xfs_swap_extent_rmap(&tp, ip, tip); 17791f08af52SDarrick J. Wong else 178039aff5fdSDarrick J. Wong error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags, 178139aff5fdSDarrick J. Wong &target_log_flags); 178221b5c978SDave Chinner if (error) 178321b5c978SDave Chinner goto out_trans_cancel; 1784a133d952SDave Chinner 1785f0bc4d13SDarrick J. Wong /* Do we have to swap reflink flags? */ 17863e09ab8fSChristoph Hellwig if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^ 17873e09ab8fSChristoph Hellwig (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) { 17883e09ab8fSChristoph Hellwig f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK; 17893e09ab8fSChristoph Hellwig ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK; 17903e09ab8fSChristoph Hellwig ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK; 17913e09ab8fSChristoph Hellwig tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK; 17923e09ab8fSChristoph Hellwig tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK; 179352bfcdd7SDarrick J. Wong } 179452bfcdd7SDarrick J. Wong 179552bfcdd7SDarrick J. Wong /* Swap the cow forks. */ 179638c26bfdSDave Chinner if (xfs_has_reflink(mp)) { 1797f7e67b20SChristoph Hellwig ASSERT(!ip->i_cowfp || 1798f7e67b20SChristoph Hellwig ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS); 1799f7e67b20SChristoph Hellwig ASSERT(!tip->i_cowfp || 1800f7e67b20SChristoph Hellwig tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS); 180152bfcdd7SDarrick J. Wong 1802897992b7SGustavo A. R. Silva swap(ip->i_cowfp, tip->i_cowfp); 180352bfcdd7SDarrick J. Wong 18045bcffe30SChristoph Hellwig if (ip->i_cowfp && ip->i_cowfp->if_bytes) 180583104d44SDarrick J. Wong xfs_inode_set_cowblocks_tag(ip); 180652bfcdd7SDarrick J. Wong else 180752bfcdd7SDarrick J. Wong xfs_inode_clear_cowblocks_tag(ip); 18085bcffe30SChristoph Hellwig if (tip->i_cowfp && tip->i_cowfp->if_bytes) 180983104d44SDarrick J. Wong xfs_inode_set_cowblocks_tag(tip); 181052bfcdd7SDarrick J. Wong else 181152bfcdd7SDarrick J. Wong xfs_inode_clear_cowblocks_tag(tip); 1812f0bc4d13SDarrick J. Wong } 1813f0bc4d13SDarrick J. Wong 1814a133d952SDave Chinner xfs_trans_log_inode(tp, ip, src_log_flags); 1815a133d952SDave Chinner xfs_trans_log_inode(tp, tip, target_log_flags); 1816a133d952SDave Chinner 1817a133d952SDave Chinner /* 18186fb10d6dSBrian Foster * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems 18196fb10d6dSBrian Foster * have inode number owner values in the bmbt blocks that still refer to 18206fb10d6dSBrian Foster * the old inode. Scan each bmbt to fix up the owner values with the 18216fb10d6dSBrian Foster * inode number of the current inode. 18226fb10d6dSBrian Foster */ 18236fb10d6dSBrian Foster if (src_log_flags & XFS_ILOG_DOWNER) { 18242dd3d709SBrian Foster error = xfs_swap_change_owner(&tp, ip, tip); 18256fb10d6dSBrian Foster if (error) 18266fb10d6dSBrian Foster goto out_trans_cancel; 18276fb10d6dSBrian Foster } 18286fb10d6dSBrian Foster if (target_log_flags & XFS_ILOG_DOWNER) { 18292dd3d709SBrian Foster error = xfs_swap_change_owner(&tp, tip, ip); 18306fb10d6dSBrian Foster if (error) 18316fb10d6dSBrian Foster goto out_trans_cancel; 18326fb10d6dSBrian Foster } 18336fb10d6dSBrian Foster 18346fb10d6dSBrian Foster /* 1835a133d952SDave Chinner * If this is a synchronous mount, make sure that the 1836a133d952SDave Chinner * transaction goes to disk before returning to the user. 1837a133d952SDave Chinner */ 18380560f31aSDave Chinner if (xfs_has_wsync(mp)) 1839a133d952SDave Chinner xfs_trans_set_sync(tp); 1840a133d952SDave Chinner 184170393313SChristoph Hellwig error = xfs_trans_commit(tp); 1842a133d952SDave Chinner 1843a133d952SDave Chinner trace_xfs_swap_extent_after(ip, 0); 1844a133d952SDave Chinner trace_xfs_swap_extent_after(tip, 1); 184539aff5fdSDarrick J. Wong 1846d2c292d8SJan Kara out_unlock_ilock: 1847d2c292d8SJan Kara xfs_iunlock(ip, XFS_ILOCK_EXCL); 1848d2c292d8SJan Kara xfs_iunlock(tip, XFS_ILOCK_EXCL); 184965523218SChristoph Hellwig out_unlock: 1850d2c292d8SJan Kara filemap_invalidate_unlock_two(VFS_I(ip)->i_mapping, 1851d2c292d8SJan Kara VFS_I(tip)->i_mapping); 185265523218SChristoph Hellwig unlock_two_nondirectories(VFS_I(ip), VFS_I(tip)); 1853a133d952SDave Chinner return error; 1854a133d952SDave Chinner 185539aff5fdSDarrick J. Wong out_trans_cancel: 185639aff5fdSDarrick J. Wong xfs_trans_cancel(tp); 1857d2c292d8SJan Kara goto out_unlock_ilock; 1858a133d952SDave Chinner } 1859