aboutsummaryrefslogtreecommitdiffstats
path: root/fs/xfs
diff options
context:
space:
mode:
Diffstat (limited to 'fs/xfs')
-rw-r--r--fs/xfs/Makefile1
-rw-r--r--fs/xfs/xfs_acl.c1
-rw-r--r--fs/xfs/xfs_aops.c1
-rw-r--r--fs/xfs/xfs_attr.c1
-rw-r--r--fs/xfs/xfs_attr.h8
-rw-r--r--fs/xfs/xfs_bmap.c1
-rw-r--r--fs/xfs/xfs_bmap_util.c861
-rw-r--r--fs/xfs/xfs_bmap_util.h11
-rw-r--r--fs/xfs/xfs_dfrag.c1
-rw-r--r--fs/xfs/xfs_dir2.c1
-rw-r--r--fs/xfs/xfs_dir2.h6
-rw-r--r--fs/xfs/xfs_dir2_priv.h6
-rw-r--r--fs/xfs/xfs_export.c1
-rw-r--r--fs/xfs/xfs_file.c3
-rw-r--r--fs/xfs/xfs_icache.c2
-rw-r--r--fs/xfs/xfs_inode.c1004
-rw-r--r--fs/xfs/xfs_inode.h23
-rw-r--r--fs/xfs/xfs_inode_buf.h1
-rw-r--r--fs/xfs/xfs_inode_fork.c1
-rw-r--r--fs/xfs/xfs_ioctl.c36
-rw-r--r--fs/xfs/xfs_ioctl.h6
-rw-r--r--fs/xfs/xfs_ioctl32.c1
-rw-r--r--fs/xfs/xfs_iops.c2
-rw-r--r--fs/xfs/xfs_iops.h13
-rw-r--r--fs/xfs/xfs_rename.c1
-rw-r--r--fs/xfs/xfs_super.c1
-rw-r--r--fs/xfs/xfs_vnodeops.c1872
-rw-r--r--fs/xfs/xfs_vnodeops.h55
-rw-r--r--fs/xfs/xfs_xattr.c1
29 files changed, 1931 insertions, 1991 deletions
diff --git a/fs/xfs/Makefile b/fs/xfs/Makefile
index dcd21b95cd98..8691dba546cd 100644
--- a/fs/xfs/Makefile
+++ b/fs/xfs/Makefile
@@ -55,7 +55,6 @@ xfs-y += xfs_aops.o \
55 xfs_symlink.o \ 55 xfs_symlink.o \
56 xfs_trans.o \ 56 xfs_trans.o \
57 xfs_utils.o \ 57 xfs_utils.o \
58 xfs_vnodeops.o \
59 xfs_xattr.o \ 58 xfs_xattr.o \
60 kmem.o \ 59 kmem.o \
61 uuid.o 60 uuid.o
diff --git a/fs/xfs/xfs_acl.c b/fs/xfs/xfs_acl.c
index 150c670e15fa..0c4368aac449 100644
--- a/fs/xfs/xfs_acl.c
+++ b/fs/xfs/xfs_acl.c
@@ -22,7 +22,6 @@
22#include "xfs_attr.h" 22#include "xfs_attr.h"
23#include "xfs_bmap_btree.h" 23#include "xfs_bmap_btree.h"
24#include "xfs_inode.h" 24#include "xfs_inode.h"
25#include "xfs_vnodeops.h"
26#include "xfs_sb.h" 25#include "xfs_sb.h"
27#include "xfs_mount.h" 26#include "xfs_mount.h"
28#include "xfs_trace.h" 27#include "xfs_trace.h"
diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
index 709347d6f468..a82c83707324 100644
--- a/fs/xfs/xfs_aops.c
+++ b/fs/xfs/xfs_aops.c
@@ -28,7 +28,6 @@
28#include "xfs_alloc.h" 28#include "xfs_alloc.h"
29#include "xfs_error.h" 29#include "xfs_error.h"
30#include "xfs_iomap.h" 30#include "xfs_iomap.h"
31#include "xfs_vnodeops.h"
32#include "xfs_trace.h" 31#include "xfs_trace.h"
33#include "xfs_bmap.h" 32#include "xfs_bmap.h"
34#include "xfs_bmap_util.h" 33#include "xfs_bmap_util.h"
diff --git a/fs/xfs/xfs_attr.c b/fs/xfs/xfs_attr.c
index 91acc38f664b..046cf3d7f52f 100644
--- a/fs/xfs/xfs_attr.c
+++ b/fs/xfs/xfs_attr.c
@@ -40,7 +40,6 @@
40#include "xfs_error.h" 40#include "xfs_error.h"
41#include "xfs_quota.h" 41#include "xfs_quota.h"
42#include "xfs_trans_space.h" 42#include "xfs_trans_space.h"
43#include "xfs_vnodeops.h"
44#include "xfs_trace.h" 43#include "xfs_trace.h"
45 44
46/* 45/*
diff --git a/fs/xfs/xfs_attr.h b/fs/xfs/xfs_attr.h
index cb604b53e2f6..dd4824589470 100644
--- a/fs/xfs/xfs_attr.h
+++ b/fs/xfs/xfs_attr.h
@@ -142,5 +142,13 @@ typedef struct xfs_attr_list_context {
142int xfs_attr_inactive(struct xfs_inode *dp); 142int xfs_attr_inactive(struct xfs_inode *dp);
143int xfs_attr_list_int(struct xfs_attr_list_context *); 143int xfs_attr_list_int(struct xfs_attr_list_context *);
144int xfs_inode_hasattr(struct xfs_inode *ip); 144int xfs_inode_hasattr(struct xfs_inode *ip);
145int xfs_attr_get(struct xfs_inode *ip, const unsigned char *name,
146 unsigned char *value, int *valuelenp, int flags);
147int xfs_attr_set(struct xfs_inode *dp, const unsigned char *name,
148 unsigned char *value, int valuelen, int flags);
149int xfs_attr_remove(struct xfs_inode *dp, const unsigned char *name, int flags);
150int xfs_attr_list(struct xfs_inode *dp, char *buffer, int bufsize,
151 int flags, struct attrlist_cursor_kern *cursor);
152
145 153
146#endif /* __XFS_ATTR_H__ */ 154#endif /* __XFS_ATTR_H__ */
diff --git a/fs/xfs/xfs_bmap.c b/fs/xfs/xfs_bmap.c
index a9331a6b36fb..51e07e7a0092 100644
--- a/fs/xfs/xfs_bmap.c
+++ b/fs/xfs/xfs_bmap.c
@@ -48,7 +48,6 @@
48#include "xfs_trans_space.h" 48#include "xfs_trans_space.h"
49#include "xfs_buf_item.h" 49#include "xfs_buf_item.h"
50#include "xfs_filestream.h" 50#include "xfs_filestream.h"
51#include "xfs_vnodeops.h"
52#include "xfs_trace.h" 51#include "xfs_trace.h"
53#include "xfs_symlink.h" 52#include "xfs_symlink.h"
54 53
diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
index 8e48ddf1c11c..f557022bd0e7 100644
--- a/fs/xfs/xfs_bmap_util.c
+++ b/fs/xfs/xfs_bmap_util.c
@@ -1,5 +1,6 @@
1/* 1/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * Copyright (c) 2012 Red Hat, Inc.
3 * All Rights Reserved. 4 * All Rights Reserved.
4 * 5 *
5 * This program is free software; you can redistribute it and/or 6 * This program is free software; you can redistribute it and/or
@@ -41,6 +42,7 @@
41#include "xfs_quota.h" 42#include "xfs_quota.h"
42#include "xfs_trans_space.h" 43#include "xfs_trans_space.h"
43#include "xfs_trace.h" 44#include "xfs_trace.h"
45#include "xfs_icache.h"
44 46
45/* Kernel only BMAP related definitions and functions */ 47/* Kernel only BMAP related definitions and functions */
46 48
@@ -832,3 +834,862 @@ next_block:
832 834
833 return error; 835 return error;
834} 836}
837
838/*
839 * Test whether it is appropriate to check an inode for and free post EOF
840 * blocks. The 'force' parameter determines whether we should also consider
841 * regular files that are marked preallocated or append-only.
842 */
843bool
844xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
845{
846 /* prealloc/delalloc exists only on regular files */
847 if (!S_ISREG(ip->i_d.di_mode))
848 return false;
849
850 /*
851 * Zero sized files with no cached pages and delalloc blocks will not
852 * have speculative prealloc/delalloc blocks to remove.
853 */
854 if (VFS_I(ip)->i_size == 0 &&
855 VN_CACHED(VFS_I(ip)) == 0 &&
856 ip->i_delayed_blks == 0)
857 return false;
858
859 /* If we haven't read in the extent list, then don't do it now. */
860 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
861 return false;
862
863 /*
864 * Do not free real preallocated or append-only files unless the file
865 * has delalloc blocks and we are forced to remove them.
866 */
867 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
868 if (!force || ip->i_delayed_blks == 0)
869 return false;
870
871 return true;
872}
873
874/*
875 * This is called by xfs_inactive to free any blocks beyond eof
876 * when the link count isn't zero and by xfs_dm_punch_hole() when
877 * punching a hole to EOF.
878 */
879int
880xfs_free_eofblocks(
881 xfs_mount_t *mp,
882 xfs_inode_t *ip,
883 bool need_iolock)
884{
885 xfs_trans_t *tp;
886 int error;
887 xfs_fileoff_t end_fsb;
888 xfs_fileoff_t last_fsb;
889 xfs_filblks_t map_len;
890 int nimaps;
891 xfs_bmbt_irec_t imap;
892
893 /*
894 * Figure out if there are any blocks beyond the end
895 * of the file. If not, then there is nothing to do.
896 */
897 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
898 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
899 if (last_fsb <= end_fsb)
900 return 0;
901 map_len = last_fsb - end_fsb;
902
903 nimaps = 1;
904 xfs_ilock(ip, XFS_ILOCK_SHARED);
905 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
906 xfs_iunlock(ip, XFS_ILOCK_SHARED);
907
908 if (!error && (nimaps != 0) &&
909 (imap.br_startblock != HOLESTARTBLOCK ||
910 ip->i_delayed_blks)) {
911 /*
912 * Attach the dquots to the inode up front.
913 */
914 error = xfs_qm_dqattach(ip, 0);
915 if (error)
916 return error;
917
918 /*
919 * There are blocks after the end of file.
920 * Free them up now by truncating the file to
921 * its current size.
922 */
923 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
924
925 if (need_iolock) {
926 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
927 xfs_trans_cancel(tp, 0);
928 return EAGAIN;
929 }
930 }
931
932 error = xfs_trans_reserve(tp, 0,
933 XFS_ITRUNCATE_LOG_RES(mp),
934 0, XFS_TRANS_PERM_LOG_RES,
935 XFS_ITRUNCATE_LOG_COUNT);
936 if (error) {
937 ASSERT(XFS_FORCED_SHUTDOWN(mp));
938 xfs_trans_cancel(tp, 0);
939 if (need_iolock)
940 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
941 return error;
942 }
943
944 xfs_ilock(ip, XFS_ILOCK_EXCL);
945 xfs_trans_ijoin(tp, ip, 0);
946
947 /*
948 * Do not update the on-disk file size. If we update the
949 * on-disk file size and then the system crashes before the
950 * contents of the file are flushed to disk then the files
951 * may be full of holes (ie NULL files bug).
952 */
953 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
954 XFS_ISIZE(ip));
955 if (error) {
956 /*
957 * If we get an error at this point we simply don't
958 * bother truncating the file.
959 */
960 xfs_trans_cancel(tp,
961 (XFS_TRANS_RELEASE_LOG_RES |
962 XFS_TRANS_ABORT));
963 } else {
964 error = xfs_trans_commit(tp,
965 XFS_TRANS_RELEASE_LOG_RES);
966 if (!error)
967 xfs_inode_clear_eofblocks_tag(ip);
968 }
969
970 xfs_iunlock(ip, XFS_ILOCK_EXCL);
971 if (need_iolock)
972 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
973 }
974 return error;
975}
976
977/*
978 * xfs_alloc_file_space()
979 * This routine allocates disk space for the given file.
980 *
981 * If alloc_type == 0, this request is for an ALLOCSP type
982 * request which will change the file size. In this case, no
983 * DMAPI event will be generated by the call. A TRUNCATE event
984 * will be generated later by xfs_setattr.
985 *
986 * If alloc_type != 0, this request is for a RESVSP type
987 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
988 * lower block boundary byte address is less than the file's
989 * length.
990 *
991 * RETURNS:
992 * 0 on success
993 * errno on error
994 *
995 */
996STATIC int
997xfs_alloc_file_space(
998 xfs_inode_t *ip,
999 xfs_off_t offset,
1000 xfs_off_t len,
1001 int alloc_type,
1002 int attr_flags)
1003{
1004 xfs_mount_t *mp = ip->i_mount;
1005 xfs_off_t count;
1006 xfs_filblks_t allocated_fsb;
1007 xfs_filblks_t allocatesize_fsb;
1008 xfs_extlen_t extsz, temp;
1009 xfs_fileoff_t startoffset_fsb;
1010 xfs_fsblock_t firstfsb;
1011 int nimaps;
1012 int quota_flag;
1013 int rt;
1014 xfs_trans_t *tp;
1015 xfs_bmbt_irec_t imaps[1], *imapp;
1016 xfs_bmap_free_t free_list;
1017 uint qblocks, resblks, resrtextents;
1018 int committed;
1019 int error;
1020
1021 trace_xfs_alloc_file_space(ip);
1022
1023 if (XFS_FORCED_SHUTDOWN(mp))
1024 return XFS_ERROR(EIO);
1025
1026 error = xfs_qm_dqattach(ip, 0);
1027 if (error)
1028 return error;
1029
1030 if (len <= 0)
1031 return XFS_ERROR(EINVAL);
1032
1033 rt = XFS_IS_REALTIME_INODE(ip);
1034 extsz = xfs_get_extsz_hint(ip);
1035
1036 count = len;
1037 imapp = &imaps[0];
1038 nimaps = 1;
1039 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1040 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1041
1042 /*
1043 * Allocate file space until done or until there is an error
1044 */
1045 while (allocatesize_fsb && !error) {
1046 xfs_fileoff_t s, e;
1047
1048 /*
1049 * Determine space reservations for data/realtime.
1050 */
1051 if (unlikely(extsz)) {
1052 s = startoffset_fsb;
1053 do_div(s, extsz);
1054 s *= extsz;
1055 e = startoffset_fsb + allocatesize_fsb;
1056 if ((temp = do_mod(startoffset_fsb, extsz)))
1057 e += temp;
1058 if ((temp = do_mod(e, extsz)))
1059 e += extsz - temp;
1060 } else {
1061 s = 0;
1062 e = allocatesize_fsb;
1063 }
1064
1065 /*
1066 * The transaction reservation is limited to a 32-bit block
1067 * count, hence we need to limit the number of blocks we are
1068 * trying to reserve to avoid an overflow. We can't allocate
1069 * more than @nimaps extents, and an extent is limited on disk
1070 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1071 */
1072 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1073 if (unlikely(rt)) {
1074 resrtextents = qblocks = resblks;
1075 resrtextents /= mp->m_sb.sb_rextsize;
1076 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1077 quota_flag = XFS_QMOPT_RES_RTBLKS;
1078 } else {
1079 resrtextents = 0;
1080 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1081 quota_flag = XFS_QMOPT_RES_REGBLKS;
1082 }
1083
1084 /*
1085 * Allocate and setup the transaction.
1086 */
1087 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1088 error = xfs_trans_reserve(tp, resblks,
1089 XFS_WRITE_LOG_RES(mp), resrtextents,
1090 XFS_TRANS_PERM_LOG_RES,
1091 XFS_WRITE_LOG_COUNT);
1092 /*
1093 * Check for running out of space
1094 */
1095 if (error) {
1096 /*
1097 * Free the transaction structure.
1098 */
1099 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1100 xfs_trans_cancel(tp, 0);
1101 break;
1102 }
1103 xfs_ilock(ip, XFS_ILOCK_EXCL);
1104 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1105 0, quota_flag);
1106 if (error)
1107 goto error1;
1108
1109 xfs_trans_ijoin(tp, ip, 0);
1110
1111 xfs_bmap_init(&free_list, &firstfsb);
1112 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1113 allocatesize_fsb, alloc_type, &firstfsb,
1114 0, imapp, &nimaps, &free_list);
1115 if (error) {
1116 goto error0;
1117 }
1118
1119 /*
1120 * Complete the transaction
1121 */
1122 error = xfs_bmap_finish(&tp, &free_list, &committed);
1123 if (error) {
1124 goto error0;
1125 }
1126
1127 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1128 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1129 if (error) {
1130 break;
1131 }
1132
1133 allocated_fsb = imapp->br_blockcount;
1134
1135 if (nimaps == 0) {
1136 error = XFS_ERROR(ENOSPC);
1137 break;
1138 }
1139
1140 startoffset_fsb += allocated_fsb;
1141 allocatesize_fsb -= allocated_fsb;
1142 }
1143
1144 return error;
1145
1146error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1147 xfs_bmap_cancel(&free_list);
1148 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1149
1150error1: /* Just cancel transaction */
1151 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1152 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1153 return error;
1154}
1155
1156/*
1157 * Zero file bytes between startoff and endoff inclusive.
1158 * The iolock is held exclusive and no blocks are buffered.
1159 *
1160 * This function is used by xfs_free_file_space() to zero
1161 * partial blocks when the range to free is not block aligned.
1162 * When unreserving space with boundaries that are not block
1163 * aligned we round up the start and round down the end
1164 * boundaries and then use this function to zero the parts of
1165 * the blocks that got dropped during the rounding.
1166 */
1167STATIC int
1168xfs_zero_remaining_bytes(
1169 xfs_inode_t *ip,
1170 xfs_off_t startoff,
1171 xfs_off_t endoff)
1172{
1173 xfs_bmbt_irec_t imap;
1174 xfs_fileoff_t offset_fsb;
1175 xfs_off_t lastoffset;
1176 xfs_off_t offset;
1177 xfs_buf_t *bp;
1178 xfs_mount_t *mp = ip->i_mount;
1179 int nimap;
1180 int error = 0;
1181
1182 /*
1183 * Avoid doing I/O beyond eof - it's not necessary
1184 * since nothing can read beyond eof. The space will
1185 * be zeroed when the file is extended anyway.
1186 */
1187 if (startoff >= XFS_ISIZE(ip))
1188 return 0;
1189
1190 if (endoff > XFS_ISIZE(ip))
1191 endoff = XFS_ISIZE(ip);
1192
1193 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1194 mp->m_rtdev_targp : mp->m_ddev_targp,
1195 BTOBB(mp->m_sb.sb_blocksize), 0);
1196 if (!bp)
1197 return XFS_ERROR(ENOMEM);
1198
1199 xfs_buf_unlock(bp);
1200
1201 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1202 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1203 nimap = 1;
1204 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1205 if (error || nimap < 1)
1206 break;
1207 ASSERT(imap.br_blockcount >= 1);
1208 ASSERT(imap.br_startoff == offset_fsb);
1209 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1210 if (lastoffset > endoff)
1211 lastoffset = endoff;
1212 if (imap.br_startblock == HOLESTARTBLOCK)
1213 continue;
1214 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1215 if (imap.br_state == XFS_EXT_UNWRITTEN)
1216 continue;
1217 XFS_BUF_UNDONE(bp);
1218 XFS_BUF_UNWRITE(bp);
1219 XFS_BUF_READ(bp);
1220 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1221 xfsbdstrat(mp, bp);
1222 error = xfs_buf_iowait(bp);
1223 if (error) {
1224 xfs_buf_ioerror_alert(bp,
1225 "xfs_zero_remaining_bytes(read)");
1226 break;
1227 }
1228 memset(bp->b_addr +
1229 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1230 0, lastoffset - offset + 1);
1231 XFS_BUF_UNDONE(bp);
1232 XFS_BUF_UNREAD(bp);
1233 XFS_BUF_WRITE(bp);
1234 xfsbdstrat(mp, bp);
1235 error = xfs_buf_iowait(bp);
1236 if (error) {
1237 xfs_buf_ioerror_alert(bp,
1238 "xfs_zero_remaining_bytes(write)");
1239 break;
1240 }
1241 }
1242 xfs_buf_free(bp);
1243 return error;
1244}
1245
1246/*
1247 * xfs_free_file_space()
1248 * This routine frees disk space for the given file.
1249 *
1250 * This routine is only called by xfs_change_file_space
1251 * for an UNRESVSP type call.
1252 *
1253 * RETURNS:
1254 * 0 on success
1255 * errno on error
1256 *
1257 */
1258STATIC int
1259xfs_free_file_space(
1260 xfs_inode_t *ip,
1261 xfs_off_t offset,
1262 xfs_off_t len,
1263 int attr_flags)
1264{
1265 int committed;
1266 int done;
1267 xfs_fileoff_t endoffset_fsb;
1268 int error;
1269 xfs_fsblock_t firstfsb;
1270 xfs_bmap_free_t free_list;
1271 xfs_bmbt_irec_t imap;
1272 xfs_off_t ioffset;
1273 xfs_extlen_t mod=0;
1274 xfs_mount_t *mp;
1275 int nimap;
1276 uint resblks;
1277 xfs_off_t rounding;
1278 int rt;
1279 xfs_fileoff_t startoffset_fsb;
1280 xfs_trans_t *tp;
1281 int need_iolock = 1;
1282
1283 mp = ip->i_mount;
1284
1285 trace_xfs_free_file_space(ip);
1286
1287 error = xfs_qm_dqattach(ip, 0);
1288 if (error)
1289 return error;
1290
1291 error = 0;
1292 if (len <= 0) /* if nothing being freed */
1293 return error;
1294 rt = XFS_IS_REALTIME_INODE(ip);
1295 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1296 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1297
1298 if (attr_flags & XFS_ATTR_NOLOCK)
1299 need_iolock = 0;
1300 if (need_iolock) {
1301 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1302 /* wait for the completion of any pending DIOs */
1303 inode_dio_wait(VFS_I(ip));
1304 }
1305
1306 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1307 ioffset = offset & ~(rounding - 1);
1308 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1309 ioffset, -1);
1310 if (error)
1311 goto out_unlock_iolock;
1312 truncate_pagecache_range(VFS_I(ip), ioffset, -1);
1313
1314 /*
1315 * Need to zero the stuff we're not freeing, on disk.
1316 * If it's a realtime file & can't use unwritten extents then we
1317 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1318 * will take care of it for us.
1319 */
1320 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1321 nimap = 1;
1322 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1323 &imap, &nimap, 0);
1324 if (error)
1325 goto out_unlock_iolock;
1326 ASSERT(nimap == 0 || nimap == 1);
1327 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1328 xfs_daddr_t block;
1329
1330 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1331 block = imap.br_startblock;
1332 mod = do_div(block, mp->m_sb.sb_rextsize);
1333 if (mod)
1334 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1335 }
1336 nimap = 1;
1337 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1338 &imap, &nimap, 0);
1339 if (error)
1340 goto out_unlock_iolock;
1341 ASSERT(nimap == 0 || nimap == 1);
1342 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1343 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1344 mod++;
1345 if (mod && (mod != mp->m_sb.sb_rextsize))
1346 endoffset_fsb -= mod;
1347 }
1348 }
1349 if ((done = (endoffset_fsb <= startoffset_fsb)))
1350 /*
1351 * One contiguous piece to clear
1352 */
1353 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1354 else {
1355 /*
1356 * Some full blocks, possibly two pieces to clear
1357 */
1358 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1359 error = xfs_zero_remaining_bytes(ip, offset,
1360 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1361 if (!error &&
1362 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1363 error = xfs_zero_remaining_bytes(ip,
1364 XFS_FSB_TO_B(mp, endoffset_fsb),
1365 offset + len - 1);
1366 }
1367
1368 /*
1369 * free file space until done or until there is an error
1370 */
1371 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1372 while (!error && !done) {
1373
1374 /*
1375 * allocate and setup the transaction. Allow this
1376 * transaction to dip into the reserve blocks to ensure
1377 * the freeing of the space succeeds at ENOSPC.
1378 */
1379 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1380 tp->t_flags |= XFS_TRANS_RESERVE;
1381 error = xfs_trans_reserve(tp,
1382 resblks,
1383 XFS_WRITE_LOG_RES(mp),
1384 0,
1385 XFS_TRANS_PERM_LOG_RES,
1386 XFS_WRITE_LOG_COUNT);
1387
1388 /*
1389 * check for running out of space
1390 */
1391 if (error) {
1392 /*
1393 * Free the transaction structure.
1394 */
1395 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1396 xfs_trans_cancel(tp, 0);
1397 break;
1398 }
1399 xfs_ilock(ip, XFS_ILOCK_EXCL);
1400 error = xfs_trans_reserve_quota(tp, mp,
1401 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1402 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1403 if (error)
1404 goto error1;
1405
1406 xfs_trans_ijoin(tp, ip, 0);
1407
1408 /*
1409 * issue the bunmapi() call to free the blocks
1410 */
1411 xfs_bmap_init(&free_list, &firstfsb);
1412 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1413 endoffset_fsb - startoffset_fsb,
1414 0, 2, &firstfsb, &free_list, &done);
1415 if (error) {
1416 goto error0;
1417 }
1418
1419 /*
1420 * complete the transaction
1421 */
1422 error = xfs_bmap_finish(&tp, &free_list, &committed);
1423 if (error) {
1424 goto error0;
1425 }
1426
1427 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1428 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1429 }
1430
1431 out_unlock_iolock:
1432 if (need_iolock)
1433 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1434 return error;
1435
1436 error0:
1437 xfs_bmap_cancel(&free_list);
1438 error1:
1439 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1440 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
1441 XFS_ILOCK_EXCL);
1442 return error;
1443}
1444
1445
1446STATIC int
1447xfs_zero_file_space(
1448 struct xfs_inode *ip,
1449 xfs_off_t offset,
1450 xfs_off_t len,
1451 int attr_flags)
1452{
1453 struct xfs_mount *mp = ip->i_mount;
1454 uint granularity;
1455 xfs_off_t start_boundary;
1456 xfs_off_t end_boundary;
1457 int error;
1458
1459 granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1460
1461 /*
1462 * Round the range of extents we are going to convert inwards. If the
1463 * offset is aligned, then it doesn't get changed so we zero from the
1464 * start of the block offset points to.
1465 */
1466 start_boundary = round_up(offset, granularity);
1467 end_boundary = round_down(offset + len, granularity);
1468
1469 ASSERT(start_boundary >= offset);
1470 ASSERT(end_boundary <= offset + len);
1471
1472 if (!(attr_flags & XFS_ATTR_NOLOCK))
1473 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1474
1475 if (start_boundary < end_boundary - 1) {
1476 /* punch out the page cache over the conversion range */
1477 truncate_pagecache_range(VFS_I(ip), start_boundary,
1478 end_boundary - 1);
1479 /* convert the blocks */
1480 error = xfs_alloc_file_space(ip, start_boundary,
1481 end_boundary - start_boundary - 1,
1482 XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT,
1483 attr_flags);
1484 if (error)
1485 goto out_unlock;
1486
1487 /* We've handled the interior of the range, now for the edges */
1488 if (start_boundary != offset)
1489 error = xfs_iozero(ip, offset, start_boundary - offset);
1490 if (error)
1491 goto out_unlock;
1492
1493 if (end_boundary != offset + len)
1494 error = xfs_iozero(ip, end_boundary,
1495 offset + len - end_boundary);
1496
1497 } else {
1498 /*
1499 * It's either a sub-granularity range or the range spanned lies
1500 * partially across two adjacent blocks.
1501 */
1502 error = xfs_iozero(ip, offset, len);
1503 }
1504
1505out_unlock:
1506 if (!(attr_flags & XFS_ATTR_NOLOCK))
1507 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1508 return error;
1509
1510}
1511
1512/*
1513 * xfs_change_file_space()
1514 * This routine allocates or frees disk space for the given file.
1515 * The user specified parameters are checked for alignment and size
1516 * limitations.
1517 *
1518 * RETURNS:
1519 * 0 on success
1520 * errno on error
1521 *
1522 */
1523int
1524xfs_change_file_space(
1525 xfs_inode_t *ip,
1526 int cmd,
1527 xfs_flock64_t *bf,
1528 xfs_off_t offset,
1529 int attr_flags)
1530{
1531 xfs_mount_t *mp = ip->i_mount;
1532 int clrprealloc;
1533 int error;
1534 xfs_fsize_t fsize;
1535 int setprealloc;
1536 xfs_off_t startoffset;
1537 xfs_trans_t *tp;
1538 struct iattr iattr;
1539
1540 if (!S_ISREG(ip->i_d.di_mode))
1541 return XFS_ERROR(EINVAL);
1542
1543 switch (bf->l_whence) {
1544 case 0: /*SEEK_SET*/
1545 break;
1546 case 1: /*SEEK_CUR*/
1547 bf->l_start += offset;
1548 break;
1549 case 2: /*SEEK_END*/
1550 bf->l_start += XFS_ISIZE(ip);
1551 break;
1552 default:
1553 return XFS_ERROR(EINVAL);
1554 }
1555
1556 /*
1557 * length of <= 0 for resv/unresv/zero is invalid. length for
1558 * alloc/free is ignored completely and we have no idea what userspace
1559 * might have set it to, so set it to zero to allow range
1560 * checks to pass.
1561 */
1562 switch (cmd) {
1563 case XFS_IOC_ZERO_RANGE:
1564 case XFS_IOC_RESVSP:
1565 case XFS_IOC_RESVSP64:
1566 case XFS_IOC_UNRESVSP:
1567 case XFS_IOC_UNRESVSP64:
1568 if (bf->l_len <= 0)
1569 return XFS_ERROR(EINVAL);
1570 break;
1571 default:
1572 bf->l_len = 0;
1573 break;
1574 }
1575
1576 if (bf->l_start < 0 ||
1577 bf->l_start > mp->m_super->s_maxbytes ||
1578 bf->l_start + bf->l_len < 0 ||
1579 bf->l_start + bf->l_len >= mp->m_super->s_maxbytes)
1580 return XFS_ERROR(EINVAL);
1581
1582 bf->l_whence = 0;
1583
1584 startoffset = bf->l_start;
1585 fsize = XFS_ISIZE(ip);
1586
1587 setprealloc = clrprealloc = 0;
1588 switch (cmd) {
1589 case XFS_IOC_ZERO_RANGE:
1590 error = xfs_zero_file_space(ip, startoffset, bf->l_len,
1591 attr_flags);
1592 if (error)
1593 return error;
1594 setprealloc = 1;
1595 break;
1596
1597 case XFS_IOC_RESVSP:
1598 case XFS_IOC_RESVSP64:
1599 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
1600 XFS_BMAPI_PREALLOC, attr_flags);
1601 if (error)
1602 return error;
1603 setprealloc = 1;
1604 break;
1605
1606 case XFS_IOC_UNRESVSP:
1607 case XFS_IOC_UNRESVSP64:
1608 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
1609 attr_flags)))
1610 return error;
1611 break;
1612
1613 case XFS_IOC_ALLOCSP:
1614 case XFS_IOC_ALLOCSP64:
1615 case XFS_IOC_FREESP:
1616 case XFS_IOC_FREESP64:
1617 /*
1618 * These operations actually do IO when extending the file, but
1619 * the allocation is done seperately to the zeroing that is
1620 * done. This set of operations need to be serialised against
1621 * other IO operations, such as truncate and buffered IO. We
1622 * need to take the IOLOCK here to serialise the allocation and
1623 * zeroing IO to prevent other IOLOCK holders (e.g. getbmap,
1624 * truncate, direct IO) from racing against the transient
1625 * allocated but not written state we can have here.
1626 */
1627 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1628 if (startoffset > fsize) {
1629 error = xfs_alloc_file_space(ip, fsize,
1630 startoffset - fsize, 0,
1631 attr_flags | XFS_ATTR_NOLOCK);
1632 if (error) {
1633 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1634 break;
1635 }
1636 }
1637
1638 iattr.ia_valid = ATTR_SIZE;
1639 iattr.ia_size = startoffset;
1640
1641 error = xfs_setattr_size(ip, &iattr,
1642 attr_flags | XFS_ATTR_NOLOCK);
1643 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1644
1645 if (error)
1646 return error;
1647
1648 clrprealloc = 1;
1649 break;
1650
1651 default:
1652 ASSERT(0);
1653 return XFS_ERROR(EINVAL);
1654 }
1655
1656 /*
1657 * update the inode timestamp, mode, and prealloc flag bits
1658 */
1659 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
1660
1661 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
1662 0, 0, 0))) {
1663 /* ASSERT(0); */
1664 xfs_trans_cancel(tp, 0);
1665 return error;
1666 }
1667
1668 xfs_ilock(ip, XFS_ILOCK_EXCL);
1669 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1670
1671 if ((attr_flags & XFS_ATTR_DMI) == 0) {
1672 ip->i_d.di_mode &= ~S_ISUID;
1673
1674 /*
1675 * Note that we don't have to worry about mandatory
1676 * file locking being disabled here because we only
1677 * clear the S_ISGID bit if the Group execute bit is
1678 * on, but if it was on then mandatory locking wouldn't
1679 * have been enabled.
1680 */
1681 if (ip->i_d.di_mode & S_IXGRP)
1682 ip->i_d.di_mode &= ~S_ISGID;
1683
1684 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1685 }
1686 if (setprealloc)
1687 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
1688 else if (clrprealloc)
1689 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
1690
1691 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1692 if (attr_flags & XFS_ATTR_SYNC)
1693 xfs_trans_set_sync(tp);
1694 return xfs_trans_commit(tp, 0);
1695}
diff --git a/fs/xfs/xfs_bmap_util.h b/fs/xfs/xfs_bmap_util.h
index 004fe4b28d49..a6d207769dee 100644
--- a/fs/xfs/xfs_bmap_util.h
+++ b/fs/xfs/xfs_bmap_util.h
@@ -21,6 +21,7 @@
21/* Kernel only BMAP related definitions and functions */ 21/* Kernel only BMAP related definitions and functions */
22 22
23struct xfs_bmbt_irec; 23struct xfs_bmbt_irec;
24struct xfs_bmap_free_item;
24struct xfs_ifork; 25struct xfs_ifork;
25struct xfs_inode; 26struct xfs_inode;
26struct xfs_mount; 27struct xfs_mount;
@@ -91,6 +92,16 @@ int xfs_bmap_last_extent(struct xfs_trans *tp, struct xfs_inode *ip,
91 int whichfork, struct xfs_bmbt_irec *rec, 92 int whichfork, struct xfs_bmbt_irec *rec,
92 int *is_empty); 93 int *is_empty);
93 94
95/* preallocation and hole punch interface */
96int xfs_change_file_space(struct xfs_inode *ip, int cmd,
97 xfs_flock64_t *bf, xfs_off_t offset,
98 int attr_flags);
99
100/* EOF block manipulation functions */
101bool xfs_can_free_eofblocks(struct xfs_inode *ip, bool force);
102int xfs_free_eofblocks(struct xfs_mount *mp, struct xfs_inode *ip,
103 bool need_iolock);
104
94xfs_daddr_t xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb); 105xfs_daddr_t xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb);
95 106
96#endif /* __XFS_BMAP_UTIL_H__ */ 107#endif /* __XFS_BMAP_UTIL_H__ */
diff --git a/fs/xfs/xfs_dfrag.c b/fs/xfs/xfs_dfrag.c
index 815dace5e57a..b83d45f8cb12 100644
--- a/fs/xfs/xfs_dfrag.c
+++ b/fs/xfs/xfs_dfrag.c
@@ -35,7 +35,6 @@
35#include "xfs_itable.h" 35#include "xfs_itable.h"
36#include "xfs_dfrag.h" 36#include "xfs_dfrag.h"
37#include "xfs_error.h" 37#include "xfs_error.h"
38#include "xfs_vnodeops.h"
39#include "xfs_trace.h" 38#include "xfs_trace.h"
40 39
41 40
diff --git a/fs/xfs/xfs_dir2.c b/fs/xfs/xfs_dir2.c
index c3263a56b861..841933c9e80f 100644
--- a/fs/xfs/xfs_dir2.c
+++ b/fs/xfs/xfs_dir2.c
@@ -35,7 +35,6 @@
35#include "xfs_dir2.h" 35#include "xfs_dir2.h"
36#include "xfs_dir2_priv.h" 36#include "xfs_dir2_priv.h"
37#include "xfs_error.h" 37#include "xfs_error.h"
38#include "xfs_vnodeops.h"
39#include "xfs_trace.h" 38#include "xfs_trace.h"
40 39
41struct xfs_name xfs_name_dotdot = { (unsigned char *)"..", 2}; 40struct xfs_name xfs_name_dotdot = { (unsigned char *)"..", 2};
diff --git a/fs/xfs/xfs_dir2.h b/fs/xfs/xfs_dir2.h
index 7ef6b0fb951a..7fe2b8f0a9e3 100644
--- a/fs/xfs/xfs_dir2.h
+++ b/fs/xfs/xfs_dir2.h
@@ -63,12 +63,6 @@ extern int xfs_dir_canenter(struct xfs_trans *tp, struct xfs_inode *dp,
63extern int xfs_dir2_sf_to_block(struct xfs_da_args *args); 63extern int xfs_dir2_sf_to_block(struct xfs_da_args *args);
64 64
65/* 65/*
66 * Direct call on directory open, before entering the readdir code.
67 */
68extern int xfs_dir3_data_readahead(struct xfs_trans *tp, struct xfs_inode *dp,
69 xfs_dablk_t bno, xfs_daddr_t mapped_bno);
70
71/*
72 * Interface routines used by userspace utilities 66 * Interface routines used by userspace utilities
73 */ 67 */
74extern xfs_ino_t xfs_dir2_sf_get_parent_ino(struct xfs_dir2_sf_hdr *sfp); 68extern xfs_ino_t xfs_dir2_sf_get_parent_ino(struct xfs_dir2_sf_hdr *sfp);
diff --git a/fs/xfs/xfs_dir2_priv.h b/fs/xfs/xfs_dir2_priv.h
index 807eb658107c..6d2a99c224b7 100644
--- a/fs/xfs/xfs_dir2_priv.h
+++ b/fs/xfs/xfs_dir2_priv.h
@@ -45,6 +45,8 @@ extern int xfs_dir2_leaf_to_block(struct xfs_da_args *args,
45extern int __xfs_dir3_data_check(struct xfs_inode *dp, struct xfs_buf *bp); 45extern int __xfs_dir3_data_check(struct xfs_inode *dp, struct xfs_buf *bp);
46extern int xfs_dir3_data_read(struct xfs_trans *tp, struct xfs_inode *dp, 46extern int xfs_dir3_data_read(struct xfs_trans *tp, struct xfs_inode *dp,
47 xfs_dablk_t bno, xfs_daddr_t mapped_bno, struct xfs_buf **bpp); 47 xfs_dablk_t bno, xfs_daddr_t mapped_bno, struct xfs_buf **bpp);
48extern int xfs_dir3_data_readahead(struct xfs_trans *tp, struct xfs_inode *dp,
49 xfs_dablk_t bno, xfs_daddr_t mapped_bno);
48 50
49extern struct xfs_dir2_data_free * 51extern struct xfs_dir2_data_free *
50xfs_dir2_data_freeinsert(struct xfs_dir2_data_hdr *hdr, 52xfs_dir2_data_freeinsert(struct xfs_dir2_data_hdr *hdr,
@@ -124,4 +126,8 @@ extern int xfs_dir2_sf_lookup(struct xfs_da_args *args);
124extern int xfs_dir2_sf_removename(struct xfs_da_args *args); 126extern int xfs_dir2_sf_removename(struct xfs_da_args *args);
125extern int xfs_dir2_sf_replace(struct xfs_da_args *args); 127extern int xfs_dir2_sf_replace(struct xfs_da_args *args);
126 128
129/* xfs_dir2_readdir.c */
130extern int xfs_readdir(struct xfs_inode *dp, struct dir_context *ctx,
131 size_t bufsize);
132
127#endif /* __XFS_DIR2_PRIV_H__ */ 133#endif /* __XFS_DIR2_PRIV_H__ */
diff --git a/fs/xfs/xfs_export.c b/fs/xfs/xfs_export.c
index 29c880fbdb1b..066df425c14f 100644
--- a/fs/xfs/xfs_export.c
+++ b/fs/xfs/xfs_export.c
@@ -26,7 +26,6 @@
26#include "xfs_dir2_format.h" 26#include "xfs_dir2_format.h"
27#include "xfs_dir2.h" 27#include "xfs_dir2.h"
28#include "xfs_export.h" 28#include "xfs_export.h"
29#include "xfs_vnodeops.h"
30#include "xfs_bmap_btree.h" 29#include "xfs_bmap_btree.h"
31#include "xfs_inode.h" 30#include "xfs_inode.h"
32#include "xfs_inode_item.h" 31#include "xfs_inode_item.h"
diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index 1ba3c7998b9a..4c749ab543d0 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -28,11 +28,12 @@
28#include "xfs_inode.h" 28#include "xfs_inode.h"
29#include "xfs_inode_item.h" 29#include "xfs_inode_item.h"
30#include "xfs_bmap.h" 30#include "xfs_bmap.h"
31#include "xfs_bmap_util.h"
31#include "xfs_error.h" 32#include "xfs_error.h"
32#include "xfs_vnodeops.h"
33#include "xfs_da_btree.h" 33#include "xfs_da_btree.h"
34#include "xfs_dir2_format.h" 34#include "xfs_dir2_format.h"
35#include "xfs_dir2.h" 35#include "xfs_dir2.h"
36#include "xfs_dir2_priv.h"
36#include "xfs_ioctl.h" 37#include "xfs_ioctl.h"
37#include "xfs_trace.h" 38#include "xfs_trace.h"
38 39
diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index d9a41671a4ad..2abc9450ce20 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -32,12 +32,12 @@
32#include "xfs_dinode.h" 32#include "xfs_dinode.h"
33#include "xfs_error.h" 33#include "xfs_error.h"
34#include "xfs_filestream.h" 34#include "xfs_filestream.h"
35#include "xfs_vnodeops.h"
36#include "xfs_inode_item.h" 35#include "xfs_inode_item.h"
37#include "xfs_quota.h" 36#include "xfs_quota.h"
38#include "xfs_trace.h" 37#include "xfs_trace.h"
39#include "xfs_fsops.h" 38#include "xfs_fsops.h"
40#include "xfs_icache.h" 39#include "xfs_icache.h"
40#include "xfs_bmap_util.h"
41 41
42#include <linux/kthread.h> 42#include <linux/kthread.h>
43#include <linux/freezer.h> 43#include <linux/freezer.h>
diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
index 1029f5670207..73a7a1d84243 100644
--- a/fs/xfs/xfs_inode.c
+++ b/fs/xfs/xfs_inode.c
@@ -23,14 +23,19 @@
23#include "xfs_log.h" 23#include "xfs_log.h"
24#include "xfs_inum.h" 24#include "xfs_inum.h"
25#include "xfs_trans.h" 25#include "xfs_trans.h"
26#include "xfs_trans_space.h"
26#include "xfs_trans_priv.h" 27#include "xfs_trans_priv.h"
27#include "xfs_sb.h" 28#include "xfs_sb.h"
28#include "xfs_ag.h" 29#include "xfs_ag.h"
29#include "xfs_mount.h" 30#include "xfs_mount.h"
31#include "xfs_da_btree.h"
32#include "xfs_dir2_format.h"
33#include "xfs_dir2.h"
30#include "xfs_bmap_btree.h" 34#include "xfs_bmap_btree.h"
31#include "xfs_alloc_btree.h" 35#include "xfs_alloc_btree.h"
32#include "xfs_ialloc_btree.h" 36#include "xfs_ialloc_btree.h"
33#include "xfs_attr_sf.h" 37#include "xfs_attr_sf.h"
38#include "xfs_attr.h"
34#include "xfs_dinode.h" 39#include "xfs_dinode.h"
35#include "xfs_inode.h" 40#include "xfs_inode.h"
36#include "xfs_buf_item.h" 41#include "xfs_buf_item.h"
@@ -44,10 +49,10 @@
44#include "xfs_utils.h" 49#include "xfs_utils.h"
45#include "xfs_quota.h" 50#include "xfs_quota.h"
46#include "xfs_filestream.h" 51#include "xfs_filestream.h"
47#include "xfs_vnodeops.h"
48#include "xfs_cksum.h" 52#include "xfs_cksum.h"
49#include "xfs_trace.h" 53#include "xfs_trace.h"
50#include "xfs_icache.h" 54#include "xfs_icache.h"
55#include "xfs_symlink.h"
51 56
52kmem_zone_t *xfs_inode_zone; 57kmem_zone_t *xfs_inode_zone;
53 58
@@ -307,6 +312,188 @@ xfs_isilocked(
307} 312}
308#endif 313#endif
309 314
315#ifdef DEBUG
316int xfs_locked_n;
317int xfs_small_retries;
318int xfs_middle_retries;
319int xfs_lots_retries;
320int xfs_lock_delays;
321#endif
322
323/*
324 * Bump the subclass so xfs_lock_inodes() acquires each lock with
325 * a different value
326 */
327static inline int
328xfs_lock_inumorder(int lock_mode, int subclass)
329{
330 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
331 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
332 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
333 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
334
335 return lock_mode;
336}
337
338/*
339 * The following routine will lock n inodes in exclusive mode.
340 * We assume the caller calls us with the inodes in i_ino order.
341 *
342 * We need to detect deadlock where an inode that we lock
343 * is in the AIL and we start waiting for another inode that is locked
344 * by a thread in a long running transaction (such as truncate). This can
345 * result in deadlock since the long running trans might need to wait
346 * for the inode we just locked in order to push the tail and free space
347 * in the log.
348 */
349void
350xfs_lock_inodes(
351 xfs_inode_t **ips,
352 int inodes,
353 uint lock_mode)
354{
355 int attempts = 0, i, j, try_lock;
356 xfs_log_item_t *lp;
357
358 ASSERT(ips && (inodes >= 2)); /* we need at least two */
359
360 try_lock = 0;
361 i = 0;
362
363again:
364 for (; i < inodes; i++) {
365 ASSERT(ips[i]);
366
367 if (i && (ips[i] == ips[i-1])) /* Already locked */
368 continue;
369
370 /*
371 * If try_lock is not set yet, make sure all locked inodes
372 * are not in the AIL.
373 * If any are, set try_lock to be used later.
374 */
375
376 if (!try_lock) {
377 for (j = (i - 1); j >= 0 && !try_lock; j--) {
378 lp = (xfs_log_item_t *)ips[j]->i_itemp;
379 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
380 try_lock++;
381 }
382 }
383 }
384
385 /*
386 * If any of the previous locks we have locked is in the AIL,
387 * we must TRY to get the second and subsequent locks. If
388 * we can't get any, we must release all we have
389 * and try again.
390 */
391
392 if (try_lock) {
393 /* try_lock must be 0 if i is 0. */
394 /*
395 * try_lock means we have an inode locked
396 * that is in the AIL.
397 */
398 ASSERT(i != 0);
399 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
400 attempts++;
401
402 /*
403 * Unlock all previous guys and try again.
404 * xfs_iunlock will try to push the tail
405 * if the inode is in the AIL.
406 */
407
408 for(j = i - 1; j >= 0; j--) {
409
410 /*
411 * Check to see if we've already
412 * unlocked this one.
413 * Not the first one going back,
414 * and the inode ptr is the same.
415 */
416 if ((j != (i - 1)) && ips[j] ==
417 ips[j+1])
418 continue;
419
420 xfs_iunlock(ips[j], lock_mode);
421 }
422
423 if ((attempts % 5) == 0) {
424 delay(1); /* Don't just spin the CPU */
425#ifdef DEBUG
426 xfs_lock_delays++;
427#endif
428 }
429 i = 0;
430 try_lock = 0;
431 goto again;
432 }
433 } else {
434 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
435 }
436 }
437
438#ifdef DEBUG
439 if (attempts) {
440 if (attempts < 5) xfs_small_retries++;
441 else if (attempts < 100) xfs_middle_retries++;
442 else xfs_lots_retries++;
443 } else {
444 xfs_locked_n++;
445 }
446#endif
447}
448
449/*
450 * xfs_lock_two_inodes() can only be used to lock one type of lock
451 * at a time - the iolock or the ilock, but not both at once. If
452 * we lock both at once, lockdep will report false positives saying
453 * we have violated locking orders.
454 */
455void
456xfs_lock_two_inodes(
457 xfs_inode_t *ip0,
458 xfs_inode_t *ip1,
459 uint lock_mode)
460{
461 xfs_inode_t *temp;
462 int attempts = 0;
463 xfs_log_item_t *lp;
464
465 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
466 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
467 ASSERT(ip0->i_ino != ip1->i_ino);
468
469 if (ip0->i_ino > ip1->i_ino) {
470 temp = ip0;
471 ip0 = ip1;
472 ip1 = temp;
473 }
474
475 again:
476 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
477
478 /*
479 * If the first lock we have locked is in the AIL, we must TRY to get
480 * the second lock. If we can't get it, we must release the first one
481 * and try again.
482 */
483 lp = (xfs_log_item_t *)ip0->i_itemp;
484 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
485 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
486 xfs_iunlock(ip0, lock_mode);
487 if ((++attempts % 5) == 0)
488 delay(1); /* Don't just spin the CPU */
489 goto again;
490 }
491 } else {
492 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
493 }
494}
495
496
310void 497void
311__xfs_iflock( 498__xfs_iflock(
312 struct xfs_inode *ip) 499 struct xfs_inode *ip)
@@ -382,6 +569,49 @@ xfs_dic2xflags(
382} 569}
383 570
384/* 571/*
572 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
573 * is allowed, otherwise it has to be an exact match. If a CI match is found,
574 * ci_name->name will point to a the actual name (caller must free) or
575 * will be set to NULL if an exact match is found.
576 */
577int
578xfs_lookup(
579 xfs_inode_t *dp,
580 struct xfs_name *name,
581 xfs_inode_t **ipp,
582 struct xfs_name *ci_name)
583{
584 xfs_ino_t inum;
585 int error;
586 uint lock_mode;
587
588 trace_xfs_lookup(dp, name);
589
590 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
591 return XFS_ERROR(EIO);
592
593 lock_mode = xfs_ilock_map_shared(dp);
594 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
595 xfs_iunlock_map_shared(dp, lock_mode);
596
597 if (error)
598 goto out;
599
600 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
601 if (error)
602 goto out_free_name;
603
604 return 0;
605
606out_free_name:
607 if (ci_name)
608 kmem_free(ci_name->name);
609out:
610 *ipp = NULL;
611 return error;
612}
613
614/*
385 * Allocate an inode on disk and return a copy of its in-core version. 615 * Allocate an inode on disk and return a copy of its in-core version.
386 * The in-core inode is locked exclusively. Set mode, nlink, and rdev 616 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
387 * appropriately within the inode. The uid and gid for the inode are 617 * appropriately within the inode. The uid and gid for the inode are
@@ -632,6 +862,308 @@ xfs_ialloc(
632 return 0; 862 return 0;
633} 863}
634 864
865int
866xfs_create(
867 xfs_inode_t *dp,
868 struct xfs_name *name,
869 umode_t mode,
870 xfs_dev_t rdev,
871 xfs_inode_t **ipp)
872{
873 int is_dir = S_ISDIR(mode);
874 struct xfs_mount *mp = dp->i_mount;
875 struct xfs_inode *ip = NULL;
876 struct xfs_trans *tp = NULL;
877 int error;
878 xfs_bmap_free_t free_list;
879 xfs_fsblock_t first_block;
880 bool unlock_dp_on_error = false;
881 uint cancel_flags;
882 int committed;
883 prid_t prid;
884 struct xfs_dquot *udqp = NULL;
885 struct xfs_dquot *gdqp = NULL;
886 struct xfs_dquot *pdqp = NULL;
887 uint resblks;
888 uint log_res;
889 uint log_count;
890
891 trace_xfs_create(dp, name);
892
893 if (XFS_FORCED_SHUTDOWN(mp))
894 return XFS_ERROR(EIO);
895
896 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
897 prid = xfs_get_projid(dp);
898 else
899 prid = XFS_PROJID_DEFAULT;
900
901 /*
902 * Make sure that we have allocated dquot(s) on disk.
903 */
904 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
905 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
906 &udqp, &gdqp, &pdqp);
907 if (error)
908 return error;
909
910 if (is_dir) {
911 rdev = 0;
912 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
913 log_res = XFS_MKDIR_LOG_RES(mp);
914 log_count = XFS_MKDIR_LOG_COUNT;
915 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
916 } else {
917 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
918 log_res = XFS_CREATE_LOG_RES(mp);
919 log_count = XFS_CREATE_LOG_COUNT;
920 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
921 }
922
923 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
924
925 /*
926 * Initially assume that the file does not exist and
927 * reserve the resources for that case. If that is not
928 * the case we'll drop the one we have and get a more
929 * appropriate transaction later.
930 */
931 error = xfs_trans_reserve(tp, resblks, log_res, 0,
932 XFS_TRANS_PERM_LOG_RES, log_count);
933 if (error == ENOSPC) {
934 /* flush outstanding delalloc blocks and retry */
935 xfs_flush_inodes(mp);
936 error = xfs_trans_reserve(tp, resblks, log_res, 0,
937 XFS_TRANS_PERM_LOG_RES, log_count);
938 }
939 if (error == ENOSPC) {
940 /* No space at all so try a "no-allocation" reservation */
941 resblks = 0;
942 error = xfs_trans_reserve(tp, 0, log_res, 0,
943 XFS_TRANS_PERM_LOG_RES, log_count);
944 }
945 if (error) {
946 cancel_flags = 0;
947 goto out_trans_cancel;
948 }
949
950 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
951 unlock_dp_on_error = true;
952
953 xfs_bmap_init(&free_list, &first_block);
954
955 /*
956 * Reserve disk quota and the inode.
957 */
958 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
959 pdqp, resblks, 1, 0);
960 if (error)
961 goto out_trans_cancel;
962
963 error = xfs_dir_canenter(tp, dp, name, resblks);
964 if (error)
965 goto out_trans_cancel;
966
967 /*
968 * A newly created regular or special file just has one directory
969 * entry pointing to them, but a directory also the "." entry
970 * pointing to itself.
971 */
972 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
973 prid, resblks > 0, &ip, &committed);
974 if (error) {
975 if (error == ENOSPC)
976 goto out_trans_cancel;
977 goto out_trans_abort;
978 }
979
980 /*
981 * Now we join the directory inode to the transaction. We do not do it
982 * earlier because xfs_dir_ialloc might commit the previous transaction
983 * (and release all the locks). An error from here on will result in
984 * the transaction cancel unlocking dp so don't do it explicitly in the
985 * error path.
986 */
987 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
988 unlock_dp_on_error = false;
989
990 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
991 &first_block, &free_list, resblks ?
992 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
993 if (error) {
994 ASSERT(error != ENOSPC);
995 goto out_trans_abort;
996 }
997 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
998 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
999
1000 if (is_dir) {
1001 error = xfs_dir_init(tp, ip, dp);
1002 if (error)
1003 goto out_bmap_cancel;
1004
1005 error = xfs_bumplink(tp, dp);
1006 if (error)
1007 goto out_bmap_cancel;
1008 }
1009
1010 /*
1011 * If this is a synchronous mount, make sure that the
1012 * create transaction goes to disk before returning to
1013 * the user.
1014 */
1015 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1016 xfs_trans_set_sync(tp);
1017
1018 /*
1019 * Attach the dquot(s) to the inodes and modify them incore.
1020 * These ids of the inode couldn't have changed since the new
1021 * inode has been locked ever since it was created.
1022 */
1023 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1024
1025 error = xfs_bmap_finish(&tp, &free_list, &committed);
1026 if (error)
1027 goto out_bmap_cancel;
1028
1029 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1030 if (error)
1031 goto out_release_inode;
1032
1033 xfs_qm_dqrele(udqp);
1034 xfs_qm_dqrele(gdqp);
1035 xfs_qm_dqrele(pdqp);
1036
1037 *ipp = ip;
1038 return 0;
1039
1040 out_bmap_cancel:
1041 xfs_bmap_cancel(&free_list);
1042 out_trans_abort:
1043 cancel_flags |= XFS_TRANS_ABORT;
1044 out_trans_cancel:
1045 xfs_trans_cancel(tp, cancel_flags);
1046 out_release_inode:
1047 /*
1048 * Wait until after the current transaction is aborted to
1049 * release the inode. This prevents recursive transactions
1050 * and deadlocks from xfs_inactive.
1051 */
1052 if (ip)
1053 IRELE(ip);
1054
1055 xfs_qm_dqrele(udqp);
1056 xfs_qm_dqrele(gdqp);
1057 xfs_qm_dqrele(pdqp);
1058
1059 if (unlock_dp_on_error)
1060 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1061 return error;
1062}
1063
1064int
1065xfs_link(
1066 xfs_inode_t *tdp,
1067 xfs_inode_t *sip,
1068 struct xfs_name *target_name)
1069{
1070 xfs_mount_t *mp = tdp->i_mount;
1071 xfs_trans_t *tp;
1072 int error;
1073 xfs_bmap_free_t free_list;
1074 xfs_fsblock_t first_block;
1075 int cancel_flags;
1076 int committed;
1077 int resblks;
1078
1079 trace_xfs_link(tdp, target_name);
1080
1081 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1082
1083 if (XFS_FORCED_SHUTDOWN(mp))
1084 return XFS_ERROR(EIO);
1085
1086 error = xfs_qm_dqattach(sip, 0);
1087 if (error)
1088 goto std_return;
1089
1090 error = xfs_qm_dqattach(tdp, 0);
1091 if (error)
1092 goto std_return;
1093
1094 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1095 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1096 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1097 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1098 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1099 if (error == ENOSPC) {
1100 resblks = 0;
1101 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1102 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1103 }
1104 if (error) {
1105 cancel_flags = 0;
1106 goto error_return;
1107 }
1108
1109 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1110
1111 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1112 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1113
1114 /*
1115 * If we are using project inheritance, we only allow hard link
1116 * creation in our tree when the project IDs are the same; else
1117 * the tree quota mechanism could be circumvented.
1118 */
1119 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1120 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1121 error = XFS_ERROR(EXDEV);
1122 goto error_return;
1123 }
1124
1125 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1126 if (error)
1127 goto error_return;
1128
1129 xfs_bmap_init(&free_list, &first_block);
1130
1131 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1132 &first_block, &free_list, resblks);
1133 if (error)
1134 goto abort_return;
1135 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1136 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1137
1138 error = xfs_bumplink(tp, sip);
1139 if (error)
1140 goto abort_return;
1141
1142 /*
1143 * If this is a synchronous mount, make sure that the
1144 * link transaction goes to disk before returning to
1145 * the user.
1146 */
1147 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1148 xfs_trans_set_sync(tp);
1149 }
1150
1151 error = xfs_bmap_finish (&tp, &free_list, &committed);
1152 if (error) {
1153 xfs_bmap_cancel(&free_list);
1154 goto abort_return;
1155 }
1156
1157 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1158
1159 abort_return:
1160 cancel_flags |= XFS_TRANS_ABORT;
1161 error_return:
1162 xfs_trans_cancel(tp, cancel_flags);
1163 std_return:
1164 return error;
1165}
1166
635/* 1167/*
636 * Free up the underlying blocks past new_size. The new size must be smaller 1168 * Free up the underlying blocks past new_size. The new size must be smaller
637 * than the current size. This routine can be used both for the attribute and 1169 * than the current size. This routine can be used both for the attribute and
@@ -771,6 +1303,276 @@ out_bmap_cancel:
771 goto out; 1303 goto out;
772} 1304}
773 1305
1306int
1307xfs_release(
1308 xfs_inode_t *ip)
1309{
1310 xfs_mount_t *mp = ip->i_mount;
1311 int error;
1312
1313 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1314 return 0;
1315
1316 /* If this is a read-only mount, don't do this (would generate I/O) */
1317 if (mp->m_flags & XFS_MOUNT_RDONLY)
1318 return 0;
1319
1320 if (!XFS_FORCED_SHUTDOWN(mp)) {
1321 int truncated;
1322
1323 /*
1324 * If we are using filestreams, and we have an unlinked
1325 * file that we are processing the last close on, then nothing
1326 * will be able to reopen and write to this file. Purge this
1327 * inode from the filestreams cache so that it doesn't delay
1328 * teardown of the inode.
1329 */
1330 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1331 xfs_filestream_deassociate(ip);
1332
1333 /*
1334 * If we previously truncated this file and removed old data
1335 * in the process, we want to initiate "early" writeout on
1336 * the last close. This is an attempt to combat the notorious
1337 * NULL files problem which is particularly noticeable from a
1338 * truncate down, buffered (re-)write (delalloc), followed by
1339 * a crash. What we are effectively doing here is
1340 * significantly reducing the time window where we'd otherwise
1341 * be exposed to that problem.
1342 */
1343 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1344 if (truncated) {
1345 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1346 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
1347 error = -filemap_flush(VFS_I(ip)->i_mapping);
1348 if (error)
1349 return error;
1350 }
1351 }
1352 }
1353
1354 if (ip->i_d.di_nlink == 0)
1355 return 0;
1356
1357 if (xfs_can_free_eofblocks(ip, false)) {
1358
1359 /*
1360 * If we can't get the iolock just skip truncating the blocks
1361 * past EOF because we could deadlock with the mmap_sem
1362 * otherwise. We'll get another chance to drop them once the
1363 * last reference to the inode is dropped, so we'll never leak
1364 * blocks permanently.
1365 *
1366 * Further, check if the inode is being opened, written and
1367 * closed frequently and we have delayed allocation blocks
1368 * outstanding (e.g. streaming writes from the NFS server),
1369 * truncating the blocks past EOF will cause fragmentation to
1370 * occur.
1371 *
1372 * In this case don't do the truncation, either, but we have to
1373 * be careful how we detect this case. Blocks beyond EOF show
1374 * up as i_delayed_blks even when the inode is clean, so we
1375 * need to truncate them away first before checking for a dirty
1376 * release. Hence on the first dirty close we will still remove
1377 * the speculative allocation, but after that we will leave it
1378 * in place.
1379 */
1380 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1381 return 0;
1382
1383 error = xfs_free_eofblocks(mp, ip, true);
1384 if (error && error != EAGAIN)
1385 return error;
1386
1387 /* delalloc blocks after truncation means it really is dirty */
1388 if (ip->i_delayed_blks)
1389 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1390 }
1391 return 0;
1392}
1393
1394/*
1395 * xfs_inactive
1396 *
1397 * This is called when the vnode reference count for the vnode
1398 * goes to zero. If the file has been unlinked, then it must
1399 * now be truncated. Also, we clear all of the read-ahead state
1400 * kept for the inode here since the file is now closed.
1401 */
1402int
1403xfs_inactive(
1404 xfs_inode_t *ip)
1405{
1406 xfs_bmap_free_t free_list;
1407 xfs_fsblock_t first_block;
1408 int committed;
1409 xfs_trans_t *tp;
1410 xfs_mount_t *mp;
1411 int error;
1412 int truncate = 0;
1413
1414 /*
1415 * If the inode is already free, then there can be nothing
1416 * to clean up here.
1417 */
1418 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
1419 ASSERT(ip->i_df.if_real_bytes == 0);
1420 ASSERT(ip->i_df.if_broot_bytes == 0);
1421 return VN_INACTIVE_CACHE;
1422 }
1423
1424 mp = ip->i_mount;
1425
1426 error = 0;
1427
1428 /* If this is a read-only mount, don't do this (would generate I/O) */
1429 if (mp->m_flags & XFS_MOUNT_RDONLY)
1430 goto out;
1431
1432 if (ip->i_d.di_nlink != 0) {
1433 /*
1434 * force is true because we are evicting an inode from the
1435 * cache. Post-eof blocks must be freed, lest we end up with
1436 * broken free space accounting.
1437 */
1438 if (xfs_can_free_eofblocks(ip, true)) {
1439 error = xfs_free_eofblocks(mp, ip, false);
1440 if (error)
1441 return VN_INACTIVE_CACHE;
1442 }
1443 goto out;
1444 }
1445
1446 if (S_ISREG(ip->i_d.di_mode) &&
1447 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1448 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1449 truncate = 1;
1450
1451 error = xfs_qm_dqattach(ip, 0);
1452 if (error)
1453 return VN_INACTIVE_CACHE;
1454
1455 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1456 error = xfs_trans_reserve(tp, 0,
1457 (truncate || S_ISLNK(ip->i_d.di_mode)) ?
1458 XFS_ITRUNCATE_LOG_RES(mp) :
1459 XFS_IFREE_LOG_RES(mp),
1460 0,
1461 XFS_TRANS_PERM_LOG_RES,
1462 XFS_ITRUNCATE_LOG_COUNT);
1463 if (error) {
1464 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1465 xfs_trans_cancel(tp, 0);
1466 return VN_INACTIVE_CACHE;
1467 }
1468
1469 xfs_ilock(ip, XFS_ILOCK_EXCL);
1470 xfs_trans_ijoin(tp, ip, 0);
1471
1472 if (S_ISLNK(ip->i_d.di_mode)) {
1473 error = xfs_inactive_symlink(ip, &tp);
1474 if (error)
1475 goto out_cancel;
1476 } else if (truncate) {
1477 ip->i_d.di_size = 0;
1478 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1479
1480 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1481 if (error)
1482 goto out_cancel;
1483
1484 ASSERT(ip->i_d.di_nextents == 0);
1485 }
1486
1487 /*
1488 * If there are attributes associated with the file then blow them away
1489 * now. The code calls a routine that recursively deconstructs the
1490 * attribute fork. We need to just commit the current transaction
1491 * because we can't use it for xfs_attr_inactive().
1492 */
1493 if (ip->i_d.di_anextents > 0) {
1494 ASSERT(ip->i_d.di_forkoff != 0);
1495
1496 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1497 if (error)
1498 goto out_unlock;
1499
1500 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1501
1502 error = xfs_attr_inactive(ip);
1503 if (error)
1504 goto out;
1505
1506 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1507 error = xfs_trans_reserve(tp, 0,
1508 XFS_IFREE_LOG_RES(mp),
1509 0, XFS_TRANS_PERM_LOG_RES,
1510 XFS_INACTIVE_LOG_COUNT);
1511 if (error) {
1512 xfs_trans_cancel(tp, 0);
1513 goto out;
1514 }
1515
1516 xfs_ilock(ip, XFS_ILOCK_EXCL);
1517 xfs_trans_ijoin(tp, ip, 0);
1518 }
1519
1520 if (ip->i_afp)
1521 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1522
1523 ASSERT(ip->i_d.di_anextents == 0);
1524
1525 /*
1526 * Free the inode.
1527 */
1528 xfs_bmap_init(&free_list, &first_block);
1529 error = xfs_ifree(tp, ip, &free_list);
1530 if (error) {
1531 /*
1532 * If we fail to free the inode, shut down. The cancel
1533 * might do that, we need to make sure. Otherwise the
1534 * inode might be lost for a long time or forever.
1535 */
1536 if (!XFS_FORCED_SHUTDOWN(mp)) {
1537 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1538 __func__, error);
1539 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1540 }
1541 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1542 } else {
1543 /*
1544 * Credit the quota account(s). The inode is gone.
1545 */
1546 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1547
1548 /*
1549 * Just ignore errors at this point. There is nothing we can
1550 * do except to try to keep going. Make sure it's not a silent
1551 * error.
1552 */
1553 error = xfs_bmap_finish(&tp, &free_list, &committed);
1554 if (error)
1555 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1556 __func__, error);
1557 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1558 if (error)
1559 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1560 __func__, error);
1561 }
1562
1563 /*
1564 * Release the dquots held by inode, if any.
1565 */
1566 xfs_qm_dqdetach(ip);
1567out_unlock:
1568 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1569out:
1570 return VN_INACTIVE_CACHE;
1571out_cancel:
1572 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1573 goto out_unlock;
1574}
1575
774/* 1576/*
775 * This is called when the inode's link count goes to 0. 1577 * This is called when the inode's link count goes to 0.
776 * We place the on-disk inode on a list in the AGI. It 1578 * We place the on-disk inode on a list in the AGI. It
@@ -1302,6 +2104,170 @@ xfs_iunpin_wait(
1302 __xfs_iunpin_wait(ip); 2104 __xfs_iunpin_wait(ip);
1303} 2105}
1304 2106
2107int
2108xfs_remove(
2109 xfs_inode_t *dp,
2110 struct xfs_name *name,
2111 xfs_inode_t *ip)
2112{
2113 xfs_mount_t *mp = dp->i_mount;
2114 xfs_trans_t *tp = NULL;
2115 int is_dir = S_ISDIR(ip->i_d.di_mode);
2116 int error = 0;
2117 xfs_bmap_free_t free_list;
2118 xfs_fsblock_t first_block;
2119 int cancel_flags;
2120 int committed;
2121 int link_zero;
2122 uint resblks;
2123 uint log_count;
2124
2125 trace_xfs_remove(dp, name);
2126
2127 if (XFS_FORCED_SHUTDOWN(mp))
2128 return XFS_ERROR(EIO);
2129
2130 error = xfs_qm_dqattach(dp, 0);
2131 if (error)
2132 goto std_return;
2133
2134 error = xfs_qm_dqattach(ip, 0);
2135 if (error)
2136 goto std_return;
2137
2138 if (is_dir) {
2139 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2140 log_count = XFS_DEFAULT_LOG_COUNT;
2141 } else {
2142 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2143 log_count = XFS_REMOVE_LOG_COUNT;
2144 }
2145 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2146
2147 /*
2148 * We try to get the real space reservation first,
2149 * allowing for directory btree deletion(s) implying
2150 * possible bmap insert(s). If we can't get the space
2151 * reservation then we use 0 instead, and avoid the bmap
2152 * btree insert(s) in the directory code by, if the bmap
2153 * insert tries to happen, instead trimming the LAST
2154 * block from the directory.
2155 */
2156 resblks = XFS_REMOVE_SPACE_RES(mp);
2157 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2158 XFS_TRANS_PERM_LOG_RES, log_count);
2159 if (error == ENOSPC) {
2160 resblks = 0;
2161 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2162 XFS_TRANS_PERM_LOG_RES, log_count);
2163 }
2164 if (error) {
2165 ASSERT(error != ENOSPC);
2166 cancel_flags = 0;
2167 goto out_trans_cancel;
2168 }
2169
2170 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2171
2172 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2173 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2174
2175 /*
2176 * If we're removing a directory perform some additional validation.
2177 */
2178 if (is_dir) {
2179 ASSERT(ip->i_d.di_nlink >= 2);
2180 if (ip->i_d.di_nlink != 2) {
2181 error = XFS_ERROR(ENOTEMPTY);
2182 goto out_trans_cancel;
2183 }
2184 if (!xfs_dir_isempty(ip)) {
2185 error = XFS_ERROR(ENOTEMPTY);
2186 goto out_trans_cancel;
2187 }
2188 }
2189
2190 xfs_bmap_init(&free_list, &first_block);
2191 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2192 &first_block, &free_list, resblks);
2193 if (error) {
2194 ASSERT(error != ENOENT);
2195 goto out_bmap_cancel;
2196 }
2197 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2198
2199 if (is_dir) {
2200 /*
2201 * Drop the link from ip's "..".
2202 */
2203 error = xfs_droplink(tp, dp);
2204 if (error)
2205 goto out_bmap_cancel;
2206
2207 /*
2208 * Drop the "." link from ip to self.
2209 */
2210 error = xfs_droplink(tp, ip);
2211 if (error)
2212 goto out_bmap_cancel;
2213 } else {
2214 /*
2215 * When removing a non-directory we need to log the parent
2216 * inode here. For a directory this is done implicitly
2217 * by the xfs_droplink call for the ".." entry.
2218 */
2219 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2220 }
2221
2222 /*
2223 * Drop the link from dp to ip.
2224 */
2225 error = xfs_droplink(tp, ip);
2226 if (error)
2227 goto out_bmap_cancel;
2228
2229 /*
2230 * Determine if this is the last link while
2231 * we are in the transaction.
2232 */
2233 link_zero = (ip->i_d.di_nlink == 0);
2234
2235 /*
2236 * If this is a synchronous mount, make sure that the
2237 * remove transaction goes to disk before returning to
2238 * the user.
2239 */
2240 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2241 xfs_trans_set_sync(tp);
2242
2243 error = xfs_bmap_finish(&tp, &free_list, &committed);
2244 if (error)
2245 goto out_bmap_cancel;
2246
2247 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2248 if (error)
2249 goto std_return;
2250
2251 /*
2252 * If we are using filestreams, kill the stream association.
2253 * If the file is still open it may get a new one but that
2254 * will get killed on last close in xfs_close() so we don't
2255 * have to worry about that.
2256 */
2257 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2258 xfs_filestream_deassociate(ip);
2259
2260 return 0;
2261
2262 out_bmap_cancel:
2263 xfs_bmap_cancel(&free_list);
2264 cancel_flags |= XFS_TRANS_ABORT;
2265 out_trans_cancel:
2266 xfs_trans_cancel(tp, cancel_flags);
2267 std_return:
2268 return error;
2269}
2270
1305STATIC int 2271STATIC int
1306xfs_iflush_cluster( 2272xfs_iflush_cluster(
1307 xfs_inode_t *ip, 2273 xfs_inode_t *ip,
@@ -1746,39 +2712,3 @@ xfs_iflush_int(
1746corrupt_out: 2712corrupt_out:
1747 return XFS_ERROR(EFSCORRUPTED); 2713 return XFS_ERROR(EFSCORRUPTED);
1748} 2714}
1749
1750/*
1751 * Test whether it is appropriate to check an inode for and free post EOF
1752 * blocks. The 'force' parameter determines whether we should also consider
1753 * regular files that are marked preallocated or append-only.
1754 */
1755bool
1756xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
1757{
1758 /* prealloc/delalloc exists only on regular files */
1759 if (!S_ISREG(ip->i_d.di_mode))
1760 return false;
1761
1762 /*
1763 * Zero sized files with no cached pages and delalloc blocks will not
1764 * have speculative prealloc/delalloc blocks to remove.
1765 */
1766 if (VFS_I(ip)->i_size == 0 &&
1767 VN_CACHED(VFS_I(ip)) == 0 &&
1768 ip->i_delayed_blks == 0)
1769 return false;
1770
1771 /* If we haven't read in the extent list, then don't do it now. */
1772 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
1773 return false;
1774
1775 /*
1776 * Do not free real preallocated or append-only files unless the file
1777 * has delalloc blocks and we are forced to remove them.
1778 */
1779 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
1780 if (!force || ip->i_delayed_blks == 0)
1781 return false;
1782
1783 return true;
1784}
diff --git a/fs/xfs/xfs_inode.h b/fs/xfs/xfs_inode.h
index 8f775ed722a9..2a6347038cd8 100644
--- a/fs/xfs/xfs_inode.h
+++ b/fs/xfs/xfs_inode.h
@@ -315,9 +315,21 @@ static inline int xfs_isiflocked(struct xfs_inode *ip)
315 ((pip)->i_d.di_mode & S_ISGID)) 315 ((pip)->i_d.di_mode & S_ISGID))
316 316
317 317
318/* 318int xfs_release(struct xfs_inode *ip);
319 * xfs_inode.c prototypes. 319int xfs_inactive(struct xfs_inode *ip);
320 */ 320int xfs_lookup(struct xfs_inode *dp, struct xfs_name *name,
321 struct xfs_inode **ipp, struct xfs_name *ci_name);
322int xfs_create(struct xfs_inode *dp, struct xfs_name *name,
323 umode_t mode, xfs_dev_t rdev, struct xfs_inode **ipp);
324int xfs_remove(struct xfs_inode *dp, struct xfs_name *name,
325 struct xfs_inode *ip);
326int xfs_link(struct xfs_inode *tdp, struct xfs_inode *sip,
327 struct xfs_name *target_name);
328int xfs_rename(struct xfs_inode *src_dp, struct xfs_name *src_name,
329 struct xfs_inode *src_ip, struct xfs_inode *target_dp,
330 struct xfs_name *target_name,
331 struct xfs_inode *target_ip);
332
321void xfs_ilock(xfs_inode_t *, uint); 333void xfs_ilock(xfs_inode_t *, uint);
322int xfs_ilock_nowait(xfs_inode_t *, uint); 334int xfs_ilock_nowait(xfs_inode_t *, uint);
323void xfs_iunlock(xfs_inode_t *, uint); 335void xfs_iunlock(xfs_inode_t *, uint);
@@ -348,6 +360,11 @@ void xfs_lock_two_inodes(xfs_inode_t *, xfs_inode_t *, uint);
348 360
349xfs_extlen_t xfs_get_extsz_hint(struct xfs_inode *ip); 361xfs_extlen_t xfs_get_extsz_hint(struct xfs_inode *ip);
350 362
363/* from xfs_file.c */
364int xfs_zero_eof(struct xfs_inode *, xfs_off_t, xfs_fsize_t);
365int xfs_iozero(struct xfs_inode *, loff_t, size_t);
366
367
351#define IHOLD(ip) \ 368#define IHOLD(ip) \
352do { \ 369do { \
353 ASSERT(atomic_read(&VFS_I(ip)->i_count) > 0) ; \ 370 ASSERT(atomic_read(&VFS_I(ip)->i_count) > 0) ; \
diff --git a/fs/xfs/xfs_inode_buf.h b/fs/xfs/xfs_inode_buf.h
index b5f1e22bf44e..aae9fc465fe0 100644
--- a/fs/xfs/xfs_inode_buf.h
+++ b/fs/xfs/xfs_inode_buf.h
@@ -40,7 +40,6 @@ int xfs_iread(struct xfs_mount *, struct xfs_trans *,
40void xfs_dinode_calc_crc(struct xfs_mount *, struct xfs_dinode *); 40void xfs_dinode_calc_crc(struct xfs_mount *, struct xfs_dinode *);
41void xfs_dinode_to_disk(struct xfs_dinode *, 41void xfs_dinode_to_disk(struct xfs_dinode *,
42 struct xfs_icdinode *); 42 struct xfs_icdinode *);
43bool xfs_can_free_eofblocks(struct xfs_inode *, bool);
44 43
45#if defined(DEBUG) 44#if defined(DEBUG)
46void xfs_inobp_check(struct xfs_mount *, struct xfs_buf *); 45void xfs_inobp_check(struct xfs_mount *, struct xfs_buf *);
diff --git a/fs/xfs/xfs_inode_fork.c b/fs/xfs/xfs_inode_fork.c
index f06570bb36bd..3f9ff99d6007 100644
--- a/fs/xfs/xfs_inode_fork.c
+++ b/fs/xfs/xfs_inode_fork.c
@@ -43,7 +43,6 @@
43#include "xfs_utils.h" 43#include "xfs_utils.h"
44#include "xfs_quota.h" 44#include "xfs_quota.h"
45#include "xfs_filestream.h" 45#include "xfs_filestream.h"
46#include "xfs_vnodeops.h"
47#include "xfs_cksum.h" 46#include "xfs_cksum.h"
48#include "xfs_trace.h" 47#include "xfs_trace.h"
49#include "xfs_icache.h" 48#include "xfs_icache.h"
diff --git a/fs/xfs/xfs_ioctl.c b/fs/xfs/xfs_ioctl.c
index 7c275380d40a..ce636bf0d550 100644
--- a/fs/xfs/xfs_ioctl.c
+++ b/fs/xfs/xfs_ioctl.c
@@ -38,13 +38,13 @@
38#include "xfs_utils.h" 38#include "xfs_utils.h"
39#include "xfs_dfrag.h" 39#include "xfs_dfrag.h"
40#include "xfs_fsops.h" 40#include "xfs_fsops.h"
41#include "xfs_vnodeops.h"
42#include "xfs_discard.h" 41#include "xfs_discard.h"
43#include "xfs_quota.h" 42#include "xfs_quota.h"
44#include "xfs_inode_item.h" 43#include "xfs_inode_item.h"
45#include "xfs_export.h" 44#include "xfs_export.h"
46#include "xfs_trace.h" 45#include "xfs_trace.h"
47#include "xfs_icache.h" 46#include "xfs_icache.h"
47#include "xfs_symlink.h"
48 48
49#include <linux/capability.h> 49#include <linux/capability.h>
50#include <linux/dcache.h> 50#include <linux/dcache.h>
@@ -352,6 +352,40 @@ xfs_readlink_by_handle(
352 return error; 352 return error;
353} 353}
354 354
355int
356xfs_set_dmattrs(
357 xfs_inode_t *ip,
358 u_int evmask,
359 u_int16_t state)
360{
361 xfs_mount_t *mp = ip->i_mount;
362 xfs_trans_t *tp;
363 int error;
364
365 if (!capable(CAP_SYS_ADMIN))
366 return XFS_ERROR(EPERM);
367
368 if (XFS_FORCED_SHUTDOWN(mp))
369 return XFS_ERROR(EIO);
370
371 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
372 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
373 if (error) {
374 xfs_trans_cancel(tp, 0);
375 return error;
376 }
377 xfs_ilock(ip, XFS_ILOCK_EXCL);
378 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
379
380 ip->i_d.di_dmevmask = evmask;
381 ip->i_d.di_dmstate = state;
382
383 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
384 error = xfs_trans_commit(tp, 0);
385
386 return error;
387}
388
355STATIC int 389STATIC int
356xfs_fssetdm_by_handle( 390xfs_fssetdm_by_handle(
357 struct file *parfilp, 391 struct file *parfilp,
diff --git a/fs/xfs/xfs_ioctl.h b/fs/xfs/xfs_ioctl.h
index d56173b34a2a..1233dee4fef0 100644
--- a/fs/xfs/xfs_ioctl.h
+++ b/fs/xfs/xfs_ioctl.h
@@ -82,4 +82,10 @@ xfs_file_compat_ioctl(
82 unsigned int cmd, 82 unsigned int cmd,
83 unsigned long arg); 83 unsigned long arg);
84 84
85extern int
86xfs_set_dmattrs(
87 struct xfs_inode *ip,
88 u_int evmask,
89 u_int16_t state);
90
85#endif 91#endif
diff --git a/fs/xfs/xfs_ioctl32.c b/fs/xfs/xfs_ioctl32.c
index c0c66259cc91..55a3072e7f56 100644
--- a/fs/xfs/xfs_ioctl32.c
+++ b/fs/xfs/xfs_ioctl32.c
@@ -34,7 +34,6 @@
34#include "xfs_itable.h" 34#include "xfs_itable.h"
35#include "xfs_error.h" 35#include "xfs_error.h"
36#include "xfs_dfrag.h" 36#include "xfs_dfrag.h"
37#include "xfs_vnodeops.h"
38#include "xfs_fsops.h" 37#include "xfs_fsops.h"
39#include "xfs_alloc.h" 38#include "xfs_alloc.h"
40#include "xfs_rtalloc.h" 39#include "xfs_rtalloc.h"
diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c
index 17b70f0ccf8a..e0cefd417bf8 100644
--- a/fs/xfs/xfs_iops.c
+++ b/fs/xfs/xfs_iops.c
@@ -37,10 +37,10 @@
37#include "xfs_attr.h" 37#include "xfs_attr.h"
38#include "xfs_buf_item.h" 38#include "xfs_buf_item.h"
39#include "xfs_utils.h" 39#include "xfs_utils.h"
40#include "xfs_vnodeops.h"
41#include "xfs_inode_item.h" 40#include "xfs_inode_item.h"
42#include "xfs_trace.h" 41#include "xfs_trace.h"
43#include "xfs_icache.h" 42#include "xfs_icache.h"
43#include "xfs_symlink.h"
44 44
45#include <linux/capability.h> 45#include <linux/capability.h>
46#include <linux/xattr.h> 46#include <linux/xattr.h>
diff --git a/fs/xfs/xfs_iops.h b/fs/xfs/xfs_iops.h
index ef41c92ce66e..d81fb41205ec 100644
--- a/fs/xfs/xfs_iops.h
+++ b/fs/xfs/xfs_iops.h
@@ -27,4 +27,17 @@ extern ssize_t xfs_vn_listxattr(struct dentry *, char *data, size_t size);
27 27
28extern void xfs_setup_inode(struct xfs_inode *); 28extern void xfs_setup_inode(struct xfs_inode *);
29 29
30/*
31 * Internal setattr interfaces.
32 */
33#define XFS_ATTR_DMI 0x01 /* invocation from a DMI function */
34#define XFS_ATTR_NONBLOCK 0x02 /* return EAGAIN if op would block */
35#define XFS_ATTR_NOLOCK 0x04 /* Don't grab any conflicting locks */
36#define XFS_ATTR_NOACL 0x08 /* Don't call xfs_acl_chmod */
37#define XFS_ATTR_SYNC 0x10 /* synchronous operation required */
38
39extern int xfs_setattr_nonsize(struct xfs_inode *ip, struct iattr *vap,
40 int flags);
41extern int xfs_setattr_size(struct xfs_inode *ip, struct iattr *vap, int flags);
42
30#endif /* __XFS_IOPS_H__ */ 43#endif /* __XFS_IOPS_H__ */
diff --git a/fs/xfs/xfs_rename.c b/fs/xfs/xfs_rename.c
index 68cb7251cb99..c15c6c581d56 100644
--- a/fs/xfs/xfs_rename.c
+++ b/fs/xfs/xfs_rename.c
@@ -36,7 +36,6 @@
36#include "xfs_quota.h" 36#include "xfs_quota.h"
37#include "xfs_utils.h" 37#include "xfs_utils.h"
38#include "xfs_trans_space.h" 38#include "xfs_trans_space.h"
39#include "xfs_vnodeops.h"
40#include "xfs_trace.h" 39#include "xfs_trace.h"
41 40
42 41
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index aefc0044237d..978a2845f9af 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -41,7 +41,6 @@
41#include "xfs_attr.h" 41#include "xfs_attr.h"
42#include "xfs_buf_item.h" 42#include "xfs_buf_item.h"
43#include "xfs_utils.h" 43#include "xfs_utils.h"
44#include "xfs_vnodeops.h"
45#include "xfs_log_priv.h" 44#include "xfs_log_priv.h"
46#include "xfs_trans_priv.h" 45#include "xfs_trans_priv.h"
47#include "xfs_filestream.h" 46#include "xfs_filestream.h"
diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
deleted file mode 100644
index dc81a8d86ba5..000000000000
--- a/fs/xfs/xfs_vnodeops.c
+++ /dev/null
@@ -1,1872 +0,0 @@
1/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * Copyright (c) 2012 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19
20#include "xfs.h"
21#include "xfs_fs.h"
22#include "xfs_format.h"
23#include "xfs_bit.h"
24#include "xfs_log.h"
25#include "xfs_trans.h"
26#include "xfs_sb.h"
27#include "xfs_ag.h"
28#include "xfs_mount.h"
29#include "xfs_da_btree.h"
30#include "xfs_dir2_format.h"
31#include "xfs_dir2.h"
32#include "xfs_bmap_btree.h"
33#include "xfs_ialloc_btree.h"
34#include "xfs_dinode.h"
35#include "xfs_inode.h"
36#include "xfs_inode_item.h"
37#include "xfs_itable.h"
38#include "xfs_ialloc.h"
39#include "xfs_alloc.h"
40#include "xfs_bmap.h"
41#include "xfs_bmap_util.h"
42#include "xfs_acl.h"
43#include "xfs_attr.h"
44#include "xfs_error.h"
45#include "xfs_quota.h"
46#include "xfs_utils.h"
47#include "xfs_rtalloc.h"
48#include "xfs_trans_space.h"
49#include "xfs_log_priv.h"
50#include "xfs_filestream.h"
51#include "xfs_vnodeops.h"
52#include "xfs_trace.h"
53#include "xfs_icache.h"
54#include "xfs_symlink.h"
55
56
57/*
58 * This is called by xfs_inactive to free any blocks beyond eof
59 * when the link count isn't zero and by xfs_dm_punch_hole() when
60 * punching a hole to EOF.
61 */
62int
63xfs_free_eofblocks(
64 xfs_mount_t *mp,
65 xfs_inode_t *ip,
66 bool need_iolock)
67{
68 xfs_trans_t *tp;
69 int error;
70 xfs_fileoff_t end_fsb;
71 xfs_fileoff_t last_fsb;
72 xfs_filblks_t map_len;
73 int nimaps;
74 xfs_bmbt_irec_t imap;
75
76 /*
77 * Figure out if there are any blocks beyond the end
78 * of the file. If not, then there is nothing to do.
79 */
80 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
81 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
82 if (last_fsb <= end_fsb)
83 return 0;
84 map_len = last_fsb - end_fsb;
85
86 nimaps = 1;
87 xfs_ilock(ip, XFS_ILOCK_SHARED);
88 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
89 xfs_iunlock(ip, XFS_ILOCK_SHARED);
90
91 if (!error && (nimaps != 0) &&
92 (imap.br_startblock != HOLESTARTBLOCK ||
93 ip->i_delayed_blks)) {
94 /*
95 * Attach the dquots to the inode up front.
96 */
97 error = xfs_qm_dqattach(ip, 0);
98 if (error)
99 return error;
100
101 /*
102 * There are blocks after the end of file.
103 * Free them up now by truncating the file to
104 * its current size.
105 */
106 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
107
108 if (need_iolock) {
109 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
110 xfs_trans_cancel(tp, 0);
111 return EAGAIN;
112 }
113 }
114
115 error = xfs_trans_reserve(tp, 0,
116 XFS_ITRUNCATE_LOG_RES(mp),
117 0, XFS_TRANS_PERM_LOG_RES,
118 XFS_ITRUNCATE_LOG_COUNT);
119 if (error) {
120 ASSERT(XFS_FORCED_SHUTDOWN(mp));
121 xfs_trans_cancel(tp, 0);
122 if (need_iolock)
123 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
124 return error;
125 }
126
127 xfs_ilock(ip, XFS_ILOCK_EXCL);
128 xfs_trans_ijoin(tp, ip, 0);
129
130 /*
131 * Do not update the on-disk file size. If we update the
132 * on-disk file size and then the system crashes before the
133 * contents of the file are flushed to disk then the files
134 * may be full of holes (ie NULL files bug).
135 */
136 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
137 XFS_ISIZE(ip));
138 if (error) {
139 /*
140 * If we get an error at this point we simply don't
141 * bother truncating the file.
142 */
143 xfs_trans_cancel(tp,
144 (XFS_TRANS_RELEASE_LOG_RES |
145 XFS_TRANS_ABORT));
146 } else {
147 error = xfs_trans_commit(tp,
148 XFS_TRANS_RELEASE_LOG_RES);
149 if (!error)
150 xfs_inode_clear_eofblocks_tag(ip);
151 }
152
153 xfs_iunlock(ip, XFS_ILOCK_EXCL);
154 if (need_iolock)
155 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
156 }
157 return error;
158}
159
160int
161xfs_release(
162 xfs_inode_t *ip)
163{
164 xfs_mount_t *mp = ip->i_mount;
165 int error;
166
167 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
168 return 0;
169
170 /* If this is a read-only mount, don't do this (would generate I/O) */
171 if (mp->m_flags & XFS_MOUNT_RDONLY)
172 return 0;
173
174 if (!XFS_FORCED_SHUTDOWN(mp)) {
175 int truncated;
176
177 /*
178 * If we are using filestreams, and we have an unlinked
179 * file that we are processing the last close on, then nothing
180 * will be able to reopen and write to this file. Purge this
181 * inode from the filestreams cache so that it doesn't delay
182 * teardown of the inode.
183 */
184 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
185 xfs_filestream_deassociate(ip);
186
187 /*
188 * If we previously truncated this file and removed old data
189 * in the process, we want to initiate "early" writeout on
190 * the last close. This is an attempt to combat the notorious
191 * NULL files problem which is particularly noticeable from a
192 * truncate down, buffered (re-)write (delalloc), followed by
193 * a crash. What we are effectively doing here is
194 * significantly reducing the time window where we'd otherwise
195 * be exposed to that problem.
196 */
197 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
198 if (truncated) {
199 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
200 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
201 error = -filemap_flush(VFS_I(ip)->i_mapping);
202 if (error)
203 return error;
204 }
205 }
206 }
207
208 if (ip->i_d.di_nlink == 0)
209 return 0;
210
211 if (xfs_can_free_eofblocks(ip, false)) {
212
213 /*
214 * If we can't get the iolock just skip truncating the blocks
215 * past EOF because we could deadlock with the mmap_sem
216 * otherwise. We'll get another chance to drop them once the
217 * last reference to the inode is dropped, so we'll never leak
218 * blocks permanently.
219 *
220 * Further, check if the inode is being opened, written and
221 * closed frequently and we have delayed allocation blocks
222 * outstanding (e.g. streaming writes from the NFS server),
223 * truncating the blocks past EOF will cause fragmentation to
224 * occur.
225 *
226 * In this case don't do the truncation, either, but we have to
227 * be careful how we detect this case. Blocks beyond EOF show
228 * up as i_delayed_blks even when the inode is clean, so we
229 * need to truncate them away first before checking for a dirty
230 * release. Hence on the first dirty close we will still remove
231 * the speculative allocation, but after that we will leave it
232 * in place.
233 */
234 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
235 return 0;
236
237 error = xfs_free_eofblocks(mp, ip, true);
238 if (error && error != EAGAIN)
239 return error;
240
241 /* delalloc blocks after truncation means it really is dirty */
242 if (ip->i_delayed_blks)
243 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
244 }
245 return 0;
246}
247
248/*
249 * xfs_inactive
250 *
251 * This is called when the vnode reference count for the vnode
252 * goes to zero. If the file has been unlinked, then it must
253 * now be truncated. Also, we clear all of the read-ahead state
254 * kept for the inode here since the file is now closed.
255 */
256int
257xfs_inactive(
258 xfs_inode_t *ip)
259{
260 xfs_bmap_free_t free_list;
261 xfs_fsblock_t first_block;
262 int committed;
263 xfs_trans_t *tp;
264 xfs_mount_t *mp;
265 int error;
266 int truncate = 0;
267
268 /*
269 * If the inode is already free, then there can be nothing
270 * to clean up here.
271 */
272 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
273 ASSERT(ip->i_df.if_real_bytes == 0);
274 ASSERT(ip->i_df.if_broot_bytes == 0);
275 return VN_INACTIVE_CACHE;
276 }
277
278 mp = ip->i_mount;
279
280 error = 0;
281
282 /* If this is a read-only mount, don't do this (would generate I/O) */
283 if (mp->m_flags & XFS_MOUNT_RDONLY)
284 goto out;
285
286 if (ip->i_d.di_nlink != 0) {
287 /*
288 * force is true because we are evicting an inode from the
289 * cache. Post-eof blocks must be freed, lest we end up with
290 * broken free space accounting.
291 */
292 if (xfs_can_free_eofblocks(ip, true)) {
293 error = xfs_free_eofblocks(mp, ip, false);
294 if (error)
295 return VN_INACTIVE_CACHE;
296 }
297 goto out;
298 }
299
300 if (S_ISREG(ip->i_d.di_mode) &&
301 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
302 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
303 truncate = 1;
304
305 error = xfs_qm_dqattach(ip, 0);
306 if (error)
307 return VN_INACTIVE_CACHE;
308
309 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
310 error = xfs_trans_reserve(tp, 0,
311 (truncate || S_ISLNK(ip->i_d.di_mode)) ?
312 XFS_ITRUNCATE_LOG_RES(mp) :
313 XFS_IFREE_LOG_RES(mp),
314 0,
315 XFS_TRANS_PERM_LOG_RES,
316 XFS_ITRUNCATE_LOG_COUNT);
317 if (error) {
318 ASSERT(XFS_FORCED_SHUTDOWN(mp));
319 xfs_trans_cancel(tp, 0);
320 return VN_INACTIVE_CACHE;
321 }
322
323 xfs_ilock(ip, XFS_ILOCK_EXCL);
324 xfs_trans_ijoin(tp, ip, 0);
325
326 if (S_ISLNK(ip->i_d.di_mode)) {
327 error = xfs_inactive_symlink(ip, &tp);
328 if (error)
329 goto out_cancel;
330 } else if (truncate) {
331 ip->i_d.di_size = 0;
332 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
333
334 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
335 if (error)
336 goto out_cancel;
337
338 ASSERT(ip->i_d.di_nextents == 0);
339 }
340
341 /*
342 * If there are attributes associated with the file then blow them away
343 * now. The code calls a routine that recursively deconstructs the
344 * attribute fork. We need to just commit the current transaction
345 * because we can't use it for xfs_attr_inactive().
346 */
347 if (ip->i_d.di_anextents > 0) {
348 ASSERT(ip->i_d.di_forkoff != 0);
349
350 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
351 if (error)
352 goto out_unlock;
353
354 xfs_iunlock(ip, XFS_ILOCK_EXCL);
355
356 error = xfs_attr_inactive(ip);
357 if (error)
358 goto out;
359
360 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
361 error = xfs_trans_reserve(tp, 0,
362 XFS_IFREE_LOG_RES(mp),
363 0, XFS_TRANS_PERM_LOG_RES,
364 XFS_INACTIVE_LOG_COUNT);
365 if (error) {
366 xfs_trans_cancel(tp, 0);
367 goto out;
368 }
369
370 xfs_ilock(ip, XFS_ILOCK_EXCL);
371 xfs_trans_ijoin(tp, ip, 0);
372 }
373
374 if (ip->i_afp)
375 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
376
377 ASSERT(ip->i_d.di_anextents == 0);
378
379 /*
380 * Free the inode.
381 */
382 xfs_bmap_init(&free_list, &first_block);
383 error = xfs_ifree(tp, ip, &free_list);
384 if (error) {
385 /*
386 * If we fail to free the inode, shut down. The cancel
387 * might do that, we need to make sure. Otherwise the
388 * inode might be lost for a long time or forever.
389 */
390 if (!XFS_FORCED_SHUTDOWN(mp)) {
391 xfs_notice(mp, "%s: xfs_ifree returned error %d",
392 __func__, error);
393 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
394 }
395 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
396 } else {
397 /*
398 * Credit the quota account(s). The inode is gone.
399 */
400 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
401
402 /*
403 * Just ignore errors at this point. There is nothing we can
404 * do except to try to keep going. Make sure it's not a silent
405 * error.
406 */
407 error = xfs_bmap_finish(&tp, &free_list, &committed);
408 if (error)
409 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
410 __func__, error);
411 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
412 if (error)
413 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
414 __func__, error);
415 }
416
417 /*
418 * Release the dquots held by inode, if any.
419 */
420 xfs_qm_dqdetach(ip);
421out_unlock:
422 xfs_iunlock(ip, XFS_ILOCK_EXCL);
423out:
424 return VN_INACTIVE_CACHE;
425out_cancel:
426 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
427 goto out_unlock;
428}
429
430/*
431 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
432 * is allowed, otherwise it has to be an exact match. If a CI match is found,
433 * ci_name->name will point to a the actual name (caller must free) or
434 * will be set to NULL if an exact match is found.
435 */
436int
437xfs_lookup(
438 xfs_inode_t *dp,
439 struct xfs_name *name,
440 xfs_inode_t **ipp,
441 struct xfs_name *ci_name)
442{
443 xfs_ino_t inum;
444 int error;
445 uint lock_mode;
446
447 trace_xfs_lookup(dp, name);
448
449 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
450 return XFS_ERROR(EIO);
451
452 lock_mode = xfs_ilock_map_shared(dp);
453 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
454 xfs_iunlock_map_shared(dp, lock_mode);
455
456 if (error)
457 goto out;
458
459 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
460 if (error)
461 goto out_free_name;
462
463 return 0;
464
465out_free_name:
466 if (ci_name)
467 kmem_free(ci_name->name);
468out:
469 *ipp = NULL;
470 return error;
471}
472
473int
474xfs_create(
475 xfs_inode_t *dp,
476 struct xfs_name *name,
477 umode_t mode,
478 xfs_dev_t rdev,
479 xfs_inode_t **ipp)
480{
481 int is_dir = S_ISDIR(mode);
482 struct xfs_mount *mp = dp->i_mount;
483 struct xfs_inode *ip = NULL;
484 struct xfs_trans *tp = NULL;
485 int error;
486 xfs_bmap_free_t free_list;
487 xfs_fsblock_t first_block;
488 bool unlock_dp_on_error = false;
489 uint cancel_flags;
490 int committed;
491 prid_t prid;
492 struct xfs_dquot *udqp = NULL;
493 struct xfs_dquot *gdqp = NULL;
494 struct xfs_dquot *pdqp = NULL;
495 uint resblks;
496 uint log_res;
497 uint log_count;
498
499 trace_xfs_create(dp, name);
500
501 if (XFS_FORCED_SHUTDOWN(mp))
502 return XFS_ERROR(EIO);
503
504 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
505 prid = xfs_get_projid(dp);
506 else
507 prid = XFS_PROJID_DEFAULT;
508
509 /*
510 * Make sure that we have allocated dquot(s) on disk.
511 */
512 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
513 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
514 &udqp, &gdqp, &pdqp);
515 if (error)
516 return error;
517
518 if (is_dir) {
519 rdev = 0;
520 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
521 log_res = XFS_MKDIR_LOG_RES(mp);
522 log_count = XFS_MKDIR_LOG_COUNT;
523 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
524 } else {
525 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
526 log_res = XFS_CREATE_LOG_RES(mp);
527 log_count = XFS_CREATE_LOG_COUNT;
528 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
529 }
530
531 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
532
533 /*
534 * Initially assume that the file does not exist and
535 * reserve the resources for that case. If that is not
536 * the case we'll drop the one we have and get a more
537 * appropriate transaction later.
538 */
539 error = xfs_trans_reserve(tp, resblks, log_res, 0,
540 XFS_TRANS_PERM_LOG_RES, log_count);
541 if (error == ENOSPC) {
542 /* flush outstanding delalloc blocks and retry */
543 xfs_flush_inodes(mp);
544 error = xfs_trans_reserve(tp, resblks, log_res, 0,
545 XFS_TRANS_PERM_LOG_RES, log_count);
546 }
547 if (error == ENOSPC) {
548 /* No space at all so try a "no-allocation" reservation */
549 resblks = 0;
550 error = xfs_trans_reserve(tp, 0, log_res, 0,
551 XFS_TRANS_PERM_LOG_RES, log_count);
552 }
553 if (error) {
554 cancel_flags = 0;
555 goto out_trans_cancel;
556 }
557
558 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
559 unlock_dp_on_error = true;
560
561 xfs_bmap_init(&free_list, &first_block);
562
563 /*
564 * Reserve disk quota and the inode.
565 */
566 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
567 pdqp, resblks, 1, 0);
568 if (error)
569 goto out_trans_cancel;
570
571 error = xfs_dir_canenter(tp, dp, name, resblks);
572 if (error)
573 goto out_trans_cancel;
574
575 /*
576 * A newly created regular or special file just has one directory
577 * entry pointing to them, but a directory also the "." entry
578 * pointing to itself.
579 */
580 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
581 prid, resblks > 0, &ip, &committed);
582 if (error) {
583 if (error == ENOSPC)
584 goto out_trans_cancel;
585 goto out_trans_abort;
586 }
587
588 /*
589 * Now we join the directory inode to the transaction. We do not do it
590 * earlier because xfs_dir_ialloc might commit the previous transaction
591 * (and release all the locks). An error from here on will result in
592 * the transaction cancel unlocking dp so don't do it explicitly in the
593 * error path.
594 */
595 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
596 unlock_dp_on_error = false;
597
598 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
599 &first_block, &free_list, resblks ?
600 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
601 if (error) {
602 ASSERT(error != ENOSPC);
603 goto out_trans_abort;
604 }
605 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
606 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
607
608 if (is_dir) {
609 error = xfs_dir_init(tp, ip, dp);
610 if (error)
611 goto out_bmap_cancel;
612
613 error = xfs_bumplink(tp, dp);
614 if (error)
615 goto out_bmap_cancel;
616 }
617
618 /*
619 * If this is a synchronous mount, make sure that the
620 * create transaction goes to disk before returning to
621 * the user.
622 */
623 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
624 xfs_trans_set_sync(tp);
625
626 /*
627 * Attach the dquot(s) to the inodes and modify them incore.
628 * These ids of the inode couldn't have changed since the new
629 * inode has been locked ever since it was created.
630 */
631 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
632
633 error = xfs_bmap_finish(&tp, &free_list, &committed);
634 if (error)
635 goto out_bmap_cancel;
636
637 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
638 if (error)
639 goto out_release_inode;
640
641 xfs_qm_dqrele(udqp);
642 xfs_qm_dqrele(gdqp);
643 xfs_qm_dqrele(pdqp);
644
645 *ipp = ip;
646 return 0;
647
648 out_bmap_cancel:
649 xfs_bmap_cancel(&free_list);
650 out_trans_abort:
651 cancel_flags |= XFS_TRANS_ABORT;
652 out_trans_cancel:
653 xfs_trans_cancel(tp, cancel_flags);
654 out_release_inode:
655 /*
656 * Wait until after the current transaction is aborted to
657 * release the inode. This prevents recursive transactions
658 * and deadlocks from xfs_inactive.
659 */
660 if (ip)
661 IRELE(ip);
662
663 xfs_qm_dqrele(udqp);
664 xfs_qm_dqrele(gdqp);
665 xfs_qm_dqrele(pdqp);
666
667 if (unlock_dp_on_error)
668 xfs_iunlock(dp, XFS_ILOCK_EXCL);
669 return error;
670}
671
672#ifdef DEBUG
673int xfs_locked_n;
674int xfs_small_retries;
675int xfs_middle_retries;
676int xfs_lots_retries;
677int xfs_lock_delays;
678#endif
679
680/*
681 * Bump the subclass so xfs_lock_inodes() acquires each lock with
682 * a different value
683 */
684static inline int
685xfs_lock_inumorder(int lock_mode, int subclass)
686{
687 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
688 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
689 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
690 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
691
692 return lock_mode;
693}
694
695/*
696 * The following routine will lock n inodes in exclusive mode.
697 * We assume the caller calls us with the inodes in i_ino order.
698 *
699 * We need to detect deadlock where an inode that we lock
700 * is in the AIL and we start waiting for another inode that is locked
701 * by a thread in a long running transaction (such as truncate). This can
702 * result in deadlock since the long running trans might need to wait
703 * for the inode we just locked in order to push the tail and free space
704 * in the log.
705 */
706void
707xfs_lock_inodes(
708 xfs_inode_t **ips,
709 int inodes,
710 uint lock_mode)
711{
712 int attempts = 0, i, j, try_lock;
713 xfs_log_item_t *lp;
714
715 ASSERT(ips && (inodes >= 2)); /* we need at least two */
716
717 try_lock = 0;
718 i = 0;
719
720again:
721 for (; i < inodes; i++) {
722 ASSERT(ips[i]);
723
724 if (i && (ips[i] == ips[i-1])) /* Already locked */
725 continue;
726
727 /*
728 * If try_lock is not set yet, make sure all locked inodes
729 * are not in the AIL.
730 * If any are, set try_lock to be used later.
731 */
732
733 if (!try_lock) {
734 for (j = (i - 1); j >= 0 && !try_lock; j--) {
735 lp = (xfs_log_item_t *)ips[j]->i_itemp;
736 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
737 try_lock++;
738 }
739 }
740 }
741
742 /*
743 * If any of the previous locks we have locked is in the AIL,
744 * we must TRY to get the second and subsequent locks. If
745 * we can't get any, we must release all we have
746 * and try again.
747 */
748
749 if (try_lock) {
750 /* try_lock must be 0 if i is 0. */
751 /*
752 * try_lock means we have an inode locked
753 * that is in the AIL.
754 */
755 ASSERT(i != 0);
756 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
757 attempts++;
758
759 /*
760 * Unlock all previous guys and try again.
761 * xfs_iunlock will try to push the tail
762 * if the inode is in the AIL.
763 */
764
765 for(j = i - 1; j >= 0; j--) {
766
767 /*
768 * Check to see if we've already
769 * unlocked this one.
770 * Not the first one going back,
771 * and the inode ptr is the same.
772 */
773 if ((j != (i - 1)) && ips[j] ==
774 ips[j+1])
775 continue;
776
777 xfs_iunlock(ips[j], lock_mode);
778 }
779
780 if ((attempts % 5) == 0) {
781 delay(1); /* Don't just spin the CPU */
782#ifdef DEBUG
783 xfs_lock_delays++;
784#endif
785 }
786 i = 0;
787 try_lock = 0;
788 goto again;
789 }
790 } else {
791 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
792 }
793 }
794
795#ifdef DEBUG
796 if (attempts) {
797 if (attempts < 5) xfs_small_retries++;
798 else if (attempts < 100) xfs_middle_retries++;
799 else xfs_lots_retries++;
800 } else {
801 xfs_locked_n++;
802 }
803#endif
804}
805
806/*
807 * xfs_lock_two_inodes() can only be used to lock one type of lock
808 * at a time - the iolock or the ilock, but not both at once. If
809 * we lock both at once, lockdep will report false positives saying
810 * we have violated locking orders.
811 */
812void
813xfs_lock_two_inodes(
814 xfs_inode_t *ip0,
815 xfs_inode_t *ip1,
816 uint lock_mode)
817{
818 xfs_inode_t *temp;
819 int attempts = 0;
820 xfs_log_item_t *lp;
821
822 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
823 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
824 ASSERT(ip0->i_ino != ip1->i_ino);
825
826 if (ip0->i_ino > ip1->i_ino) {
827 temp = ip0;
828 ip0 = ip1;
829 ip1 = temp;
830 }
831
832 again:
833 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
834
835 /*
836 * If the first lock we have locked is in the AIL, we must TRY to get
837 * the second lock. If we can't get it, we must release the first one
838 * and try again.
839 */
840 lp = (xfs_log_item_t *)ip0->i_itemp;
841 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
842 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
843 xfs_iunlock(ip0, lock_mode);
844 if ((++attempts % 5) == 0)
845 delay(1); /* Don't just spin the CPU */
846 goto again;
847 }
848 } else {
849 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
850 }
851}
852
853int
854xfs_remove(
855 xfs_inode_t *dp,
856 struct xfs_name *name,
857 xfs_inode_t *ip)
858{
859 xfs_mount_t *mp = dp->i_mount;
860 xfs_trans_t *tp = NULL;
861 int is_dir = S_ISDIR(ip->i_d.di_mode);
862 int error = 0;
863 xfs_bmap_free_t free_list;
864 xfs_fsblock_t first_block;
865 int cancel_flags;
866 int committed;
867 int link_zero;
868 uint resblks;
869 uint log_count;
870
871 trace_xfs_remove(dp, name);
872
873 if (XFS_FORCED_SHUTDOWN(mp))
874 return XFS_ERROR(EIO);
875
876 error = xfs_qm_dqattach(dp, 0);
877 if (error)
878 goto std_return;
879
880 error = xfs_qm_dqattach(ip, 0);
881 if (error)
882 goto std_return;
883
884 if (is_dir) {
885 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
886 log_count = XFS_DEFAULT_LOG_COUNT;
887 } else {
888 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
889 log_count = XFS_REMOVE_LOG_COUNT;
890 }
891 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
892
893 /*
894 * We try to get the real space reservation first,
895 * allowing for directory btree deletion(s) implying
896 * possible bmap insert(s). If we can't get the space
897 * reservation then we use 0 instead, and avoid the bmap
898 * btree insert(s) in the directory code by, if the bmap
899 * insert tries to happen, instead trimming the LAST
900 * block from the directory.
901 */
902 resblks = XFS_REMOVE_SPACE_RES(mp);
903 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
904 XFS_TRANS_PERM_LOG_RES, log_count);
905 if (error == ENOSPC) {
906 resblks = 0;
907 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
908 XFS_TRANS_PERM_LOG_RES, log_count);
909 }
910 if (error) {
911 ASSERT(error != ENOSPC);
912 cancel_flags = 0;
913 goto out_trans_cancel;
914 }
915
916 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
917
918 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
919 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
920
921 /*
922 * If we're removing a directory perform some additional validation.
923 */
924 if (is_dir) {
925 ASSERT(ip->i_d.di_nlink >= 2);
926 if (ip->i_d.di_nlink != 2) {
927 error = XFS_ERROR(ENOTEMPTY);
928 goto out_trans_cancel;
929 }
930 if (!xfs_dir_isempty(ip)) {
931 error = XFS_ERROR(ENOTEMPTY);
932 goto out_trans_cancel;
933 }
934 }
935
936 xfs_bmap_init(&free_list, &first_block);
937 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
938 &first_block, &free_list, resblks);
939 if (error) {
940 ASSERT(error != ENOENT);
941 goto out_bmap_cancel;
942 }
943 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
944
945 if (is_dir) {
946 /*
947 * Drop the link from ip's "..".
948 */
949 error = xfs_droplink(tp, dp);
950 if (error)
951 goto out_bmap_cancel;
952
953 /*
954 * Drop the "." link from ip to self.
955 */
956 error = xfs_droplink(tp, ip);
957 if (error)
958 goto out_bmap_cancel;
959 } else {
960 /*
961 * When removing a non-directory we need to log the parent
962 * inode here. For a directory this is done implicitly
963 * by the xfs_droplink call for the ".." entry.
964 */
965 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
966 }
967
968 /*
969 * Drop the link from dp to ip.
970 */
971 error = xfs_droplink(tp, ip);
972 if (error)
973 goto out_bmap_cancel;
974
975 /*
976 * Determine if this is the last link while
977 * we are in the transaction.
978 */
979 link_zero = (ip->i_d.di_nlink == 0);
980
981 /*
982 * If this is a synchronous mount, make sure that the
983 * remove transaction goes to disk before returning to
984 * the user.
985 */
986 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
987 xfs_trans_set_sync(tp);
988
989 error = xfs_bmap_finish(&tp, &free_list, &committed);
990 if (error)
991 goto out_bmap_cancel;
992
993 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
994 if (error)
995 goto std_return;
996
997 /*
998 * If we are using filestreams, kill the stream association.
999 * If the file is still open it may get a new one but that
1000 * will get killed on last close in xfs_close() so we don't
1001 * have to worry about that.
1002 */
1003 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
1004 xfs_filestream_deassociate(ip);
1005
1006 return 0;
1007
1008 out_bmap_cancel:
1009 xfs_bmap_cancel(&free_list);
1010 cancel_flags |= XFS_TRANS_ABORT;
1011 out_trans_cancel:
1012 xfs_trans_cancel(tp, cancel_flags);
1013 std_return:
1014 return error;
1015}
1016
1017int
1018xfs_link(
1019 xfs_inode_t *tdp,
1020 xfs_inode_t *sip,
1021 struct xfs_name *target_name)
1022{
1023 xfs_mount_t *mp = tdp->i_mount;
1024 xfs_trans_t *tp;
1025 int error;
1026 xfs_bmap_free_t free_list;
1027 xfs_fsblock_t first_block;
1028 int cancel_flags;
1029 int committed;
1030 int resblks;
1031
1032 trace_xfs_link(tdp, target_name);
1033
1034 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1035
1036 if (XFS_FORCED_SHUTDOWN(mp))
1037 return XFS_ERROR(EIO);
1038
1039 error = xfs_qm_dqattach(sip, 0);
1040 if (error)
1041 goto std_return;
1042
1043 error = xfs_qm_dqattach(tdp, 0);
1044 if (error)
1045 goto std_return;
1046
1047 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1048 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1049 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1050 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1051 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1052 if (error == ENOSPC) {
1053 resblks = 0;
1054 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1055 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1056 }
1057 if (error) {
1058 cancel_flags = 0;
1059 goto error_return;
1060 }
1061
1062 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1063
1064 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1065 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1066
1067 /*
1068 * If we are using project inheritance, we only allow hard link
1069 * creation in our tree when the project IDs are the same; else
1070 * the tree quota mechanism could be circumvented.
1071 */
1072 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1073 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1074 error = XFS_ERROR(EXDEV);
1075 goto error_return;
1076 }
1077
1078 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1079 if (error)
1080 goto error_return;
1081
1082 xfs_bmap_init(&free_list, &first_block);
1083
1084 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1085 &first_block, &free_list, resblks);
1086 if (error)
1087 goto abort_return;
1088 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1089 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1090
1091 error = xfs_bumplink(tp, sip);
1092 if (error)
1093 goto abort_return;
1094
1095 /*
1096 * If this is a synchronous mount, make sure that the
1097 * link transaction goes to disk before returning to
1098 * the user.
1099 */
1100 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1101 xfs_trans_set_sync(tp);
1102 }
1103
1104 error = xfs_bmap_finish (&tp, &free_list, &committed);
1105 if (error) {
1106 xfs_bmap_cancel(&free_list);
1107 goto abort_return;
1108 }
1109
1110 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1111
1112 abort_return:
1113 cancel_flags |= XFS_TRANS_ABORT;
1114 error_return:
1115 xfs_trans_cancel(tp, cancel_flags);
1116 std_return:
1117 return error;
1118}
1119
1120int
1121xfs_set_dmattrs(
1122 xfs_inode_t *ip,
1123 u_int evmask,
1124 u_int16_t state)
1125{
1126 xfs_mount_t *mp = ip->i_mount;
1127 xfs_trans_t *tp;
1128 int error;
1129
1130 if (!capable(CAP_SYS_ADMIN))
1131 return XFS_ERROR(EPERM);
1132
1133 if (XFS_FORCED_SHUTDOWN(mp))
1134 return XFS_ERROR(EIO);
1135
1136 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
1137 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
1138 if (error) {
1139 xfs_trans_cancel(tp, 0);
1140 return error;
1141 }
1142 xfs_ilock(ip, XFS_ILOCK_EXCL);
1143 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1144
1145 ip->i_d.di_dmevmask = evmask;
1146 ip->i_d.di_dmstate = state;
1147
1148 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1149 error = xfs_trans_commit(tp, 0);
1150
1151 return error;
1152}
1153
1154/*
1155 * xfs_alloc_file_space()
1156 * This routine allocates disk space for the given file.
1157 *
1158 * If alloc_type == 0, this request is for an ALLOCSP type
1159 * request which will change the file size. In this case, no
1160 * DMAPI event will be generated by the call. A TRUNCATE event
1161 * will be generated later by xfs_setattr.
1162 *
1163 * If alloc_type != 0, this request is for a RESVSP type
1164 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
1165 * lower block boundary byte address is less than the file's
1166 * length.
1167 *
1168 * RETURNS:
1169 * 0 on success
1170 * errno on error
1171 *
1172 */
1173STATIC int
1174xfs_alloc_file_space(
1175 xfs_inode_t *ip,
1176 xfs_off_t offset,
1177 xfs_off_t len,
1178 int alloc_type,
1179 int attr_flags)
1180{
1181 xfs_mount_t *mp = ip->i_mount;
1182 xfs_off_t count;
1183 xfs_filblks_t allocated_fsb;
1184 xfs_filblks_t allocatesize_fsb;
1185 xfs_extlen_t extsz, temp;
1186 xfs_fileoff_t startoffset_fsb;
1187 xfs_fsblock_t firstfsb;
1188 int nimaps;
1189 int quota_flag;
1190 int rt;
1191 xfs_trans_t *tp;
1192 xfs_bmbt_irec_t imaps[1], *imapp;
1193 xfs_bmap_free_t free_list;
1194 uint qblocks, resblks, resrtextents;
1195 int committed;
1196 int error;
1197
1198 trace_xfs_alloc_file_space(ip);
1199
1200 if (XFS_FORCED_SHUTDOWN(mp))
1201 return XFS_ERROR(EIO);
1202
1203 error = xfs_qm_dqattach(ip, 0);
1204 if (error)
1205 return error;
1206
1207 if (len <= 0)
1208 return XFS_ERROR(EINVAL);
1209
1210 rt = XFS_IS_REALTIME_INODE(ip);
1211 extsz = xfs_get_extsz_hint(ip);
1212
1213 count = len;
1214 imapp = &imaps[0];
1215 nimaps = 1;
1216 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1217 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1218
1219 /*
1220 * Allocate file space until done or until there is an error
1221 */
1222 while (allocatesize_fsb && !error) {
1223 xfs_fileoff_t s, e;
1224
1225 /*
1226 * Determine space reservations for data/realtime.
1227 */
1228 if (unlikely(extsz)) {
1229 s = startoffset_fsb;
1230 do_div(s, extsz);
1231 s *= extsz;
1232 e = startoffset_fsb + allocatesize_fsb;
1233 if ((temp = do_mod(startoffset_fsb, extsz)))
1234 e += temp;
1235 if ((temp = do_mod(e, extsz)))
1236 e += extsz - temp;
1237 } else {
1238 s = 0;
1239 e = allocatesize_fsb;
1240 }
1241
1242 /*
1243 * The transaction reservation is limited to a 32-bit block
1244 * count, hence we need to limit the number of blocks we are
1245 * trying to reserve to avoid an overflow. We can't allocate
1246 * more than @nimaps extents, and an extent is limited on disk
1247 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1248 */
1249 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1250 if (unlikely(rt)) {
1251 resrtextents = qblocks = resblks;
1252 resrtextents /= mp->m_sb.sb_rextsize;
1253 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1254 quota_flag = XFS_QMOPT_RES_RTBLKS;
1255 } else {
1256 resrtextents = 0;
1257 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1258 quota_flag = XFS_QMOPT_RES_REGBLKS;
1259 }
1260
1261 /*
1262 * Allocate and setup the transaction.
1263 */
1264 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1265 error = xfs_trans_reserve(tp, resblks,
1266 XFS_WRITE_LOG_RES(mp), resrtextents,
1267 XFS_TRANS_PERM_LOG_RES,
1268 XFS_WRITE_LOG_COUNT);
1269 /*
1270 * Check for running out of space
1271 */
1272 if (error) {
1273 /*
1274 * Free the transaction structure.
1275 */
1276 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1277 xfs_trans_cancel(tp, 0);
1278 break;
1279 }
1280 xfs_ilock(ip, XFS_ILOCK_EXCL);
1281 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1282 0, quota_flag);
1283 if (error)
1284 goto error1;
1285
1286 xfs_trans_ijoin(tp, ip, 0);
1287
1288 xfs_bmap_init(&free_list, &firstfsb);
1289 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1290 allocatesize_fsb, alloc_type, &firstfsb,
1291 0, imapp, &nimaps, &free_list);
1292 if (error) {
1293 goto error0;
1294 }
1295
1296 /*
1297 * Complete the transaction
1298 */
1299 error = xfs_bmap_finish(&tp, &free_list, &committed);
1300 if (error) {
1301 goto error0;
1302 }
1303
1304 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1305 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1306 if (error) {
1307 break;
1308 }
1309
1310 allocated_fsb = imapp->br_blockcount;
1311
1312 if (nimaps == 0) {
1313 error = XFS_ERROR(ENOSPC);
1314 break;
1315 }
1316
1317 startoffset_fsb += allocated_fsb;
1318 allocatesize_fsb -= allocated_fsb;
1319 }
1320
1321 return error;
1322
1323error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1324 xfs_bmap_cancel(&free_list);
1325 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1326
1327error1: /* Just cancel transaction */
1328 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1329 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1330 return error;
1331}
1332
1333/*
1334 * Zero file bytes between startoff and endoff inclusive.
1335 * The iolock is held exclusive and no blocks are buffered.
1336 *
1337 * This function is used by xfs_free_file_space() to zero
1338 * partial blocks when the range to free is not block aligned.
1339 * When unreserving space with boundaries that are not block
1340 * aligned we round up the start and round down the end
1341 * boundaries and then use this function to zero the parts of
1342 * the blocks that got dropped during the rounding.
1343 */
1344STATIC int
1345xfs_zero_remaining_bytes(
1346 xfs_inode_t *ip,
1347 xfs_off_t startoff,
1348 xfs_off_t endoff)
1349{
1350 xfs_bmbt_irec_t imap;
1351 xfs_fileoff_t offset_fsb;
1352 xfs_off_t lastoffset;
1353 xfs_off_t offset;
1354 xfs_buf_t *bp;
1355 xfs_mount_t *mp = ip->i_mount;
1356 int nimap;
1357 int error = 0;
1358
1359 /*
1360 * Avoid doing I/O beyond eof - it's not necessary
1361 * since nothing can read beyond eof. The space will
1362 * be zeroed when the file is extended anyway.
1363 */
1364 if (startoff >= XFS_ISIZE(ip))
1365 return 0;
1366
1367 if (endoff > XFS_ISIZE(ip))
1368 endoff = XFS_ISIZE(ip);
1369
1370 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1371 mp->m_rtdev_targp : mp->m_ddev_targp,
1372 BTOBB(mp->m_sb.sb_blocksize), 0);
1373 if (!bp)
1374 return XFS_ERROR(ENOMEM);
1375
1376 xfs_buf_unlock(bp);
1377
1378 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1379 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1380 nimap = 1;
1381 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1382 if (error || nimap < 1)
1383 break;
1384 ASSERT(imap.br_blockcount >= 1);
1385 ASSERT(imap.br_startoff == offset_fsb);
1386 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1387 if (lastoffset > endoff)
1388 lastoffset = endoff;
1389 if (imap.br_startblock == HOLESTARTBLOCK)
1390 continue;
1391 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1392 if (imap.br_state == XFS_EXT_UNWRITTEN)
1393 continue;
1394 XFS_BUF_UNDONE(bp);
1395 XFS_BUF_UNWRITE(bp);
1396 XFS_BUF_READ(bp);
1397 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1398 xfsbdstrat(mp, bp);
1399 error = xfs_buf_iowait(bp);
1400 if (error) {
1401 xfs_buf_ioerror_alert(bp,
1402 "xfs_zero_remaining_bytes(read)");
1403 break;
1404 }
1405 memset(bp->b_addr +
1406 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1407 0, lastoffset - offset + 1);
1408 XFS_BUF_UNDONE(bp);
1409 XFS_BUF_UNREAD(bp);
1410 XFS_BUF_WRITE(bp);
1411 xfsbdstrat(mp, bp);
1412 error = xfs_buf_iowait(bp);
1413 if (error) {
1414 xfs_buf_ioerror_alert(bp,
1415 "xfs_zero_remaining_bytes(write)");
1416 break;
1417 }
1418 }
1419 xfs_buf_free(bp);
1420 return error;
1421}
1422
1423/*
1424 * xfs_free_file_space()
1425 * This routine frees disk space for the given file.
1426 *
1427 * This routine is only called by xfs_change_file_space
1428 * for an UNRESVSP type call.
1429 *
1430 * RETURNS:
1431 * 0 on success
1432 * errno on error
1433 *
1434 */
1435STATIC int
1436xfs_free_file_space(
1437 xfs_inode_t *ip,
1438 xfs_off_t offset,
1439 xfs_off_t len,
1440 int attr_flags)
1441{
1442 int committed;
1443 int done;
1444 xfs_fileoff_t endoffset_fsb;
1445 int error;
1446 xfs_fsblock_t firstfsb;
1447 xfs_bmap_free_t free_list;
1448 xfs_bmbt_irec_t imap;
1449 xfs_off_t ioffset;
1450 xfs_extlen_t mod=0;
1451 xfs_mount_t *mp;
1452 int nimap;
1453 uint resblks;
1454 xfs_off_t rounding;
1455 int rt;
1456 xfs_fileoff_t startoffset_fsb;
1457 xfs_trans_t *tp;
1458 int need_iolock = 1;
1459
1460 mp = ip->i_mount;
1461
1462 trace_xfs_free_file_space(ip);
1463
1464 error = xfs_qm_dqattach(ip, 0);
1465 if (error)
1466 return error;
1467
1468 error = 0;
1469 if (len <= 0) /* if nothing being freed */
1470 return error;
1471 rt = XFS_IS_REALTIME_INODE(ip);
1472 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1473 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1474
1475 if (attr_flags & XFS_ATTR_NOLOCK)
1476 need_iolock = 0;
1477 if (need_iolock) {
1478 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1479 /* wait for the completion of any pending DIOs */
1480 inode_dio_wait(VFS_I(ip));
1481 }
1482
1483 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1484 ioffset = offset & ~(rounding - 1);
1485 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1486 ioffset, -1);
1487 if (error)
1488 goto out_unlock_iolock;
1489 truncate_pagecache_range(VFS_I(ip), ioffset, -1);
1490
1491 /*
1492 * Need to zero the stuff we're not freeing, on disk.
1493 * If it's a realtime file & can't use unwritten extents then we
1494 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1495 * will take care of it for us.
1496 */
1497 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1498 nimap = 1;
1499 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1500 &imap, &nimap, 0);
1501 if (error)
1502 goto out_unlock_iolock;
1503 ASSERT(nimap == 0 || nimap == 1);
1504 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1505 xfs_daddr_t block;
1506
1507 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1508 block = imap.br_startblock;
1509 mod = do_div(block, mp->m_sb.sb_rextsize);
1510 if (mod)
1511 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1512 }
1513 nimap = 1;
1514 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1515 &imap, &nimap, 0);
1516 if (error)
1517 goto out_unlock_iolock;
1518 ASSERT(nimap == 0 || nimap == 1);
1519 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1520 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1521 mod++;
1522 if (mod && (mod != mp->m_sb.sb_rextsize))
1523 endoffset_fsb -= mod;
1524 }
1525 }
1526 if ((done = (endoffset_fsb <= startoffset_fsb)))
1527 /*
1528 * One contiguous piece to clear
1529 */
1530 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1531 else {
1532 /*
1533 * Some full blocks, possibly two pieces to clear
1534 */
1535 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1536 error = xfs_zero_remaining_bytes(ip, offset,
1537 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1538 if (!error &&
1539 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1540 error = xfs_zero_remaining_bytes(ip,
1541 XFS_FSB_TO_B(mp, endoffset_fsb),
1542 offset + len - 1);
1543 }
1544
1545 /*
1546 * free file space until done or until there is an error
1547 */
1548 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1549 while (!error && !done) {
1550
1551 /*
1552 * allocate and setup the transaction. Allow this
1553 * transaction to dip into the reserve blocks to ensure
1554 * the freeing of the space succeeds at ENOSPC.
1555 */
1556 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1557 tp->t_flags |= XFS_TRANS_RESERVE;
1558 error = xfs_trans_reserve(tp,
1559 resblks,
1560 XFS_WRITE_LOG_RES(mp),
1561 0,
1562 XFS_TRANS_PERM_LOG_RES,
1563 XFS_WRITE_LOG_COUNT);
1564
1565 /*
1566 * check for running out of space
1567 */
1568 if (error) {
1569 /*
1570 * Free the transaction structure.
1571 */
1572 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1573 xfs_trans_cancel(tp, 0);
1574 break;
1575 }
1576 xfs_ilock(ip, XFS_ILOCK_EXCL);
1577 error = xfs_trans_reserve_quota(tp, mp,
1578 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1579 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1580 if (error)
1581 goto error1;
1582
1583 xfs_trans_ijoin(tp, ip, 0);
1584
1585 /*
1586 * issue the bunmapi() call to free the blocks
1587 */
1588 xfs_bmap_init(&free_list, &firstfsb);
1589 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1590 endoffset_fsb - startoffset_fsb,
1591 0, 2, &firstfsb, &free_list, &done);
1592 if (error) {
1593 goto error0;
1594 }
1595
1596 /*
1597 * complete the transaction
1598 */
1599 error = xfs_bmap_finish(&tp, &free_list, &committed);
1600 if (error) {
1601 goto error0;
1602 }
1603
1604 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1605 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1606 }
1607
1608 out_unlock_iolock:
1609 if (need_iolock)
1610 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1611 return error;
1612
1613 error0:
1614 xfs_bmap_cancel(&free_list);
1615 error1:
1616 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1617 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
1618 XFS_ILOCK_EXCL);
1619 return error;
1620}
1621
1622
1623STATIC int
1624xfs_zero_file_space(
1625 struct xfs_inode *ip,
1626 xfs_off_t offset,
1627 xfs_off_t len,
1628 int attr_flags)
1629{
1630 struct xfs_mount *mp = ip->i_mount;
1631 uint granularity;
1632 xfs_off_t start_boundary;
1633 xfs_off_t end_boundary;
1634 int error;
1635
1636 granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1637
1638 /*
1639 * Round the range of extents we are going to convert inwards. If the
1640 * offset is aligned, then it doesn't get changed so we zero from the
1641 * start of the block offset points to.
1642 */
1643 start_boundary = round_up(offset, granularity);
1644 end_boundary = round_down(offset + len, granularity);
1645
1646 ASSERT(start_boundary >= offset);
1647 ASSERT(end_boundary <= offset + len);
1648
1649 if (!(attr_flags & XFS_ATTR_NOLOCK))
1650 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1651
1652 if (start_boundary < end_boundary - 1) {
1653 /* punch out the page cache over the conversion range */
1654 truncate_pagecache_range(VFS_I(ip), start_boundary,
1655 end_boundary - 1);
1656 /* convert the blocks */
1657 error = xfs_alloc_file_space(ip, start_boundary,
1658 end_boundary - start_boundary - 1,
1659 XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT,
1660 attr_flags);
1661 if (error)
1662 goto out_unlock;
1663
1664 /* We've handled the interior of the range, now for the edges */
1665 if (start_boundary != offset)
1666 error = xfs_iozero(ip, offset, start_boundary - offset);
1667 if (error)
1668 goto out_unlock;
1669
1670 if (end_boundary != offset + len)
1671 error = xfs_iozero(ip, end_boundary,
1672 offset + len - end_boundary);
1673
1674 } else {
1675 /*
1676 * It's either a sub-granularity range or the range spanned lies
1677 * partially across two adjacent blocks.
1678 */
1679 error = xfs_iozero(ip, offset, len);
1680 }
1681
1682out_unlock:
1683 if (!(attr_flags & XFS_ATTR_NOLOCK))
1684 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1685 return error;
1686
1687}
1688
1689/*
1690 * xfs_change_file_space()
1691 * This routine allocates or frees disk space for the given file.
1692 * The user specified parameters are checked for alignment and size
1693 * limitations.
1694 *
1695 * RETURNS:
1696 * 0 on success
1697 * errno on error
1698 *
1699 */
1700int
1701xfs_change_file_space(
1702 xfs_inode_t *ip,
1703 int cmd,
1704 xfs_flock64_t *bf,
1705 xfs_off_t offset,
1706 int attr_flags)
1707{
1708 xfs_mount_t *mp = ip->i_mount;
1709 int clrprealloc;
1710 int error;
1711 xfs_fsize_t fsize;
1712 int setprealloc;
1713 xfs_off_t startoffset;
1714 xfs_trans_t *tp;
1715 struct iattr iattr;
1716
1717 if (!S_ISREG(ip->i_d.di_mode))
1718 return XFS_ERROR(EINVAL);
1719
1720 switch (bf->l_whence) {
1721 case 0: /*SEEK_SET*/
1722 break;
1723 case 1: /*SEEK_CUR*/
1724 bf->l_start += offset;
1725 break;
1726 case 2: /*SEEK_END*/
1727 bf->l_start += XFS_ISIZE(ip);
1728 break;
1729 default:
1730 return XFS_ERROR(EINVAL);
1731 }
1732
1733 /*
1734 * length of <= 0 for resv/unresv/zero is invalid. length for
1735 * alloc/free is ignored completely and we have no idea what userspace
1736 * might have set it to, so set it to zero to allow range
1737 * checks to pass.
1738 */
1739 switch (cmd) {
1740 case XFS_IOC_ZERO_RANGE:
1741 case XFS_IOC_RESVSP:
1742 case XFS_IOC_RESVSP64:
1743 case XFS_IOC_UNRESVSP:
1744 case XFS_IOC_UNRESVSP64:
1745 if (bf->l_len <= 0)
1746 return XFS_ERROR(EINVAL);
1747 break;
1748 default:
1749 bf->l_len = 0;
1750 break;
1751 }
1752
1753 if (bf->l_start < 0 ||
1754 bf->l_start > mp->m_super->s_maxbytes ||
1755 bf->l_start + bf->l_len < 0 ||
1756 bf->l_start + bf->l_len >= mp->m_super->s_maxbytes)
1757 return XFS_ERROR(EINVAL);
1758
1759 bf->l_whence = 0;
1760
1761 startoffset = bf->l_start;
1762 fsize = XFS_ISIZE(ip);
1763
1764 setprealloc = clrprealloc = 0;
1765 switch (cmd) {
1766 case XFS_IOC_ZERO_RANGE:
1767 error = xfs_zero_file_space(ip, startoffset, bf->l_len,
1768 attr_flags);
1769 if (error)
1770 return error;
1771 setprealloc = 1;
1772 break;
1773
1774 case XFS_IOC_RESVSP:
1775 case XFS_IOC_RESVSP64:
1776 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
1777 XFS_BMAPI_PREALLOC, attr_flags);
1778 if (error)
1779 return error;
1780 setprealloc = 1;
1781 break;
1782
1783 case XFS_IOC_UNRESVSP:
1784 case XFS_IOC_UNRESVSP64:
1785 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
1786 attr_flags)))
1787 return error;
1788 break;
1789
1790 case XFS_IOC_ALLOCSP:
1791 case XFS_IOC_ALLOCSP64:
1792 case XFS_IOC_FREESP:
1793 case XFS_IOC_FREESP64:
1794 /*
1795 * These operations actually do IO when extending the file, but
1796 * the allocation is done seperately to the zeroing that is
1797 * done. This set of operations need to be serialised against
1798 * other IO operations, such as truncate and buffered IO. We
1799 * need to take the IOLOCK here to serialise the allocation and
1800 * zeroing IO to prevent other IOLOCK holders (e.g. getbmap,
1801 * truncate, direct IO) from racing against the transient
1802 * allocated but not written state we can have here.
1803 */
1804 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1805 if (startoffset > fsize) {
1806 error = xfs_alloc_file_space(ip, fsize,
1807 startoffset - fsize, 0,
1808 attr_flags | XFS_ATTR_NOLOCK);
1809 if (error) {
1810 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1811 break;
1812 }
1813 }
1814
1815 iattr.ia_valid = ATTR_SIZE;
1816 iattr.ia_size = startoffset;
1817
1818 error = xfs_setattr_size(ip, &iattr,
1819 attr_flags | XFS_ATTR_NOLOCK);
1820 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1821
1822 if (error)
1823 return error;
1824
1825 clrprealloc = 1;
1826 break;
1827
1828 default:
1829 ASSERT(0);
1830 return XFS_ERROR(EINVAL);
1831 }
1832
1833 /*
1834 * update the inode timestamp, mode, and prealloc flag bits
1835 */
1836 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
1837
1838 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
1839 0, 0, 0))) {
1840 /* ASSERT(0); */
1841 xfs_trans_cancel(tp, 0);
1842 return error;
1843 }
1844
1845 xfs_ilock(ip, XFS_ILOCK_EXCL);
1846 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1847
1848 if ((attr_flags & XFS_ATTR_DMI) == 0) {
1849 ip->i_d.di_mode &= ~S_ISUID;
1850
1851 /*
1852 * Note that we don't have to worry about mandatory
1853 * file locking being disabled here because we only
1854 * clear the S_ISGID bit if the Group execute bit is
1855 * on, but if it was on then mandatory locking wouldn't
1856 * have been enabled.
1857 */
1858 if (ip->i_d.di_mode & S_IXGRP)
1859 ip->i_d.di_mode &= ~S_ISGID;
1860
1861 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1862 }
1863 if (setprealloc)
1864 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
1865 else if (clrprealloc)
1866 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
1867
1868 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1869 if (attr_flags & XFS_ATTR_SYNC)
1870 xfs_trans_set_sync(tp);
1871 return xfs_trans_commit(tp, 0);
1872}
diff --git a/fs/xfs/xfs_vnodeops.h b/fs/xfs/xfs_vnodeops.h
deleted file mode 100644
index 38c67c34d73f..000000000000
--- a/fs/xfs/xfs_vnodeops.h
+++ /dev/null
@@ -1,55 +0,0 @@
1#ifndef _XFS_VNODEOPS_H
2#define _XFS_VNODEOPS_H 1
3
4struct attrlist_cursor_kern;
5struct file;
6struct iattr;
7struct inode;
8struct iovec;
9struct kiocb;
10struct pipe_inode_info;
11struct uio;
12struct xfs_inode;
13
14
15int xfs_setattr_nonsize(struct xfs_inode *ip, struct iattr *vap, int flags);
16int xfs_setattr_size(struct xfs_inode *ip, struct iattr *vap, int flags);
17#define XFS_ATTR_DMI 0x01 /* invocation from a DMI function */
18#define XFS_ATTR_NONBLOCK 0x02 /* return EAGAIN if operation would block */
19#define XFS_ATTR_NOLOCK 0x04 /* Don't grab any conflicting locks */
20#define XFS_ATTR_NOACL 0x08 /* Don't call xfs_acl_chmod */
21#define XFS_ATTR_SYNC 0x10 /* synchronous operation required */
22
23int xfs_readlink(struct xfs_inode *ip, char *link);
24int xfs_release(struct xfs_inode *ip);
25int xfs_inactive(struct xfs_inode *ip);
26int xfs_lookup(struct xfs_inode *dp, struct xfs_name *name,
27 struct xfs_inode **ipp, struct xfs_name *ci_name);
28int xfs_create(struct xfs_inode *dp, struct xfs_name *name, umode_t mode,
29 xfs_dev_t rdev, struct xfs_inode **ipp);
30int xfs_remove(struct xfs_inode *dp, struct xfs_name *name,
31 struct xfs_inode *ip);
32int xfs_link(struct xfs_inode *tdp, struct xfs_inode *sip,
33 struct xfs_name *target_name);
34int xfs_readdir(struct xfs_inode *dp, struct dir_context *ctx, size_t bufsize);
35int xfs_symlink(struct xfs_inode *dp, struct xfs_name *link_name,
36 const char *target_path, umode_t mode, struct xfs_inode **ipp);
37int xfs_set_dmattrs(struct xfs_inode *ip, u_int evmask, u_int16_t state);
38int xfs_change_file_space(struct xfs_inode *ip, int cmd,
39 xfs_flock64_t *bf, xfs_off_t offset, int attr_flags);
40int xfs_rename(struct xfs_inode *src_dp, struct xfs_name *src_name,
41 struct xfs_inode *src_ip, struct xfs_inode *target_dp,
42 struct xfs_name *target_name, struct xfs_inode *target_ip);
43int xfs_attr_get(struct xfs_inode *ip, const unsigned char *name,
44 unsigned char *value, int *valuelenp, int flags);
45int xfs_attr_set(struct xfs_inode *dp, const unsigned char *name,
46 unsigned char *value, int valuelen, int flags);
47int xfs_attr_remove(struct xfs_inode *dp, const unsigned char *name, int flags);
48int xfs_attr_list(struct xfs_inode *dp, char *buffer, int bufsize,
49 int flags, struct attrlist_cursor_kern *cursor);
50
51int xfs_iozero(struct xfs_inode *, loff_t, size_t);
52int xfs_zero_eof(struct xfs_inode *, xfs_off_t, xfs_fsize_t);
53int xfs_free_eofblocks(struct xfs_mount *, struct xfs_inode *, bool);
54
55#endif /* _XFS_VNODEOPS_H */
diff --git a/fs/xfs/xfs_xattr.c b/fs/xfs/xfs_xattr.c
index b7311aaa5f0e..e01f35ea76ba 100644
--- a/fs/xfs/xfs_xattr.c
+++ b/fs/xfs/xfs_xattr.c
@@ -24,7 +24,6 @@
24#include "xfs_attr.h" 24#include "xfs_attr.h"
25#include "xfs_attr_leaf.h" 25#include "xfs_attr_leaf.h"
26#include "xfs_acl.h" 26#include "xfs_acl.h"
27#include "xfs_vnodeops.h"
28 27
29#include <linux/posix_acl_xattr.h> 28#include <linux/posix_acl_xattr.h>
30#include <linux/xattr.h> 29#include <linux/xattr.h>