Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
| 3 | * All Rights Reserved. |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License as |
| 7 | * published by the Free Software Foundation. |
| 8 | * |
| 9 | * This program is distributed in the hope that it would be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
| 13 | * |
| 14 | * You should have received a copy of the GNU General Public License |
| 15 | * along with this program; if not, write the Free Software Foundation, |
| 16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
| 17 | */ |
| 18 | #include <linux/log2.h> |
| 19 | |
| 20 | #include "xfs.h" |
| 21 | #include "xfs_fs.h" |
| 22 | #include "xfs_format.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 23 | #include "xfs_log_format.h" |
| 24 | #include "xfs_trans_resv.h" |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 25 | #include "xfs_mount.h" |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 26 | #include "xfs_inode.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 27 | #include "xfs_trans.h" |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 28 | #include "xfs_inode_item.h" |
Dave Chinner | a4fbe6a | 2013-10-23 10:51:50 +1100 | [diff] [blame] | 29 | #include "xfs_bmap_btree.h" |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 30 | #include "xfs_bmap.h" |
| 31 | #include "xfs_error.h" |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 32 | #include "xfs_trace.h" |
Dave Chinner | a4fbe6a | 2013-10-23 10:51:50 +1100 | [diff] [blame] | 33 | #include "xfs_attr_sf.h" |
Darrick J. Wong | 244efea | 2016-02-08 15:00:01 +1100 | [diff] [blame] | 34 | #include "xfs_da_format.h" |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 35 | |
| 36 | kmem_zone_t *xfs_ifork_zone; |
| 37 | |
| 38 | STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); |
| 39 | STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); |
| 40 | STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); |
| 41 | |
| 42 | #ifdef DEBUG |
| 43 | /* |
| 44 | * Make sure that the extents in the given memory buffer |
| 45 | * are valid. |
| 46 | */ |
| 47 | void |
| 48 | xfs_validate_extents( |
| 49 | xfs_ifork_t *ifp, |
| 50 | int nrecs, |
| 51 | xfs_exntfmt_t fmt) |
| 52 | { |
| 53 | xfs_bmbt_irec_t irec; |
| 54 | xfs_bmbt_rec_host_t rec; |
| 55 | int i; |
| 56 | |
| 57 | for (i = 0; i < nrecs; i++) { |
| 58 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
| 59 | rec.l0 = get_unaligned(&ep->l0); |
| 60 | rec.l1 = get_unaligned(&ep->l1); |
| 61 | xfs_bmbt_get_all(&rec, &irec); |
| 62 | if (fmt == XFS_EXTFMT_NOSTATE) |
| 63 | ASSERT(irec.br_state == XFS_EXT_NORM); |
| 64 | } |
| 65 | } |
| 66 | #else /* DEBUG */ |
| 67 | #define xfs_validate_extents(ifp, nrecs, fmt) |
| 68 | #endif /* DEBUG */ |
| 69 | |
| 70 | |
| 71 | /* |
| 72 | * Move inode type and inode format specific information from the |
| 73 | * on-disk inode to the in-core inode. For fifos, devs, and sockets |
| 74 | * this means set if_rdev to the proper value. For files, directories, |
| 75 | * and symlinks this means to bring in the in-line data or extent |
| 76 | * pointers. For a file in B-tree format, only the root is immediately |
| 77 | * brought in-core. The rest will be in-lined in if_extents when it |
| 78 | * is first referenced (see xfs_iread_extents()). |
| 79 | */ |
| 80 | int |
| 81 | xfs_iformat_fork( |
| 82 | xfs_inode_t *ip, |
| 83 | xfs_dinode_t *dip) |
| 84 | { |
| 85 | xfs_attr_shortform_t *atp; |
| 86 | int size; |
| 87 | int error = 0; |
| 88 | xfs_fsize_t di_size; |
| 89 | |
| 90 | if (unlikely(be32_to_cpu(dip->di_nextents) + |
| 91 | be16_to_cpu(dip->di_anextents) > |
| 92 | be64_to_cpu(dip->di_nblocks))) { |
| 93 | xfs_warn(ip->i_mount, |
| 94 | "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", |
| 95 | (unsigned long long)ip->i_ino, |
| 96 | (int)(be32_to_cpu(dip->di_nextents) + |
| 97 | be16_to_cpu(dip->di_anextents)), |
| 98 | (unsigned long long) |
| 99 | be64_to_cpu(dip->di_nblocks)); |
| 100 | XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, |
| 101 | ip->i_mount, dip); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 102 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 103 | } |
| 104 | |
| 105 | if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) { |
| 106 | xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.", |
| 107 | (unsigned long long)ip->i_ino, |
| 108 | dip->di_forkoff); |
| 109 | XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, |
| 110 | ip->i_mount, dip); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 111 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 112 | } |
| 113 | |
| 114 | if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && |
| 115 | !ip->i_mount->m_rtdev_targp)) { |
| 116 | xfs_warn(ip->i_mount, |
| 117 | "corrupt dinode %Lu, has realtime flag set.", |
| 118 | ip->i_ino); |
| 119 | XFS_CORRUPTION_ERROR("xfs_iformat(realtime)", |
| 120 | XFS_ERRLEVEL_LOW, ip->i_mount, dip); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 121 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 122 | } |
| 123 | |
Darrick J. Wong | 11715a2 | 2016-10-03 09:11:31 -0700 | [diff] [blame] | 124 | if (unlikely(xfs_is_reflink_inode(ip) && |
| 125 | (VFS_I(ip)->i_mode & S_IFMT) != S_IFREG)) { |
| 126 | xfs_warn(ip->i_mount, |
| 127 | "corrupt dinode %llu, wrong file type for reflink.", |
| 128 | ip->i_ino); |
| 129 | XFS_CORRUPTION_ERROR("xfs_iformat(reflink)", |
| 130 | XFS_ERRLEVEL_LOW, ip->i_mount, dip); |
| 131 | return -EFSCORRUPTED; |
| 132 | } |
| 133 | |
| 134 | if (unlikely(xfs_is_reflink_inode(ip) && |
| 135 | (ip->i_d.di_flags & XFS_DIFLAG_REALTIME))) { |
| 136 | xfs_warn(ip->i_mount, |
| 137 | "corrupt dinode %llu, has reflink+realtime flag set.", |
| 138 | ip->i_ino); |
| 139 | XFS_CORRUPTION_ERROR("xfs_iformat(reflink)", |
| 140 | XFS_ERRLEVEL_LOW, ip->i_mount, dip); |
| 141 | return -EFSCORRUPTED; |
| 142 | } |
| 143 | |
Dave Chinner | c19b3b05 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 144 | switch (VFS_I(ip)->i_mode & S_IFMT) { |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 145 | case S_IFIFO: |
| 146 | case S_IFCHR: |
| 147 | case S_IFBLK: |
| 148 | case S_IFSOCK: |
| 149 | if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) { |
| 150 | XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, |
| 151 | ip->i_mount, dip); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 152 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 153 | } |
| 154 | ip->i_d.di_size = 0; |
| 155 | ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip); |
| 156 | break; |
| 157 | |
| 158 | case S_IFREG: |
| 159 | case S_IFLNK: |
| 160 | case S_IFDIR: |
| 161 | switch (dip->di_format) { |
| 162 | case XFS_DINODE_FMT_LOCAL: |
| 163 | /* |
| 164 | * no local regular files yet |
| 165 | */ |
| 166 | if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) { |
| 167 | xfs_warn(ip->i_mount, |
| 168 | "corrupt inode %Lu (local format for regular file).", |
| 169 | (unsigned long long) ip->i_ino); |
| 170 | XFS_CORRUPTION_ERROR("xfs_iformat(4)", |
| 171 | XFS_ERRLEVEL_LOW, |
| 172 | ip->i_mount, dip); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 173 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 174 | } |
| 175 | |
| 176 | di_size = be64_to_cpu(dip->di_size); |
Dan Carpenter | 0d0ab12 | 2013-08-15 08:53:38 +0300 | [diff] [blame] | 177 | if (unlikely(di_size < 0 || |
| 178 | di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 179 | xfs_warn(ip->i_mount, |
| 180 | "corrupt inode %Lu (bad size %Ld for local inode).", |
| 181 | (unsigned long long) ip->i_ino, |
| 182 | (long long) di_size); |
| 183 | XFS_CORRUPTION_ERROR("xfs_iformat(5)", |
| 184 | XFS_ERRLEVEL_LOW, |
| 185 | ip->i_mount, dip); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 186 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 187 | } |
| 188 | |
| 189 | size = (int)di_size; |
| 190 | error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); |
| 191 | break; |
| 192 | case XFS_DINODE_FMT_EXTENTS: |
| 193 | error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); |
| 194 | break; |
| 195 | case XFS_DINODE_FMT_BTREE: |
| 196 | error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); |
| 197 | break; |
| 198 | default: |
| 199 | XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, |
| 200 | ip->i_mount); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 201 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 202 | } |
| 203 | break; |
| 204 | |
| 205 | default: |
| 206 | XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 207 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 208 | } |
Darrick J. Wong | 3993bae | 2016-10-03 09:11:32 -0700 | [diff] [blame] | 209 | if (error) |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 210 | return error; |
Darrick J. Wong | 3993bae | 2016-10-03 09:11:32 -0700 | [diff] [blame] | 211 | |
| 212 | if (xfs_is_reflink_inode(ip)) { |
| 213 | ASSERT(ip->i_cowfp == NULL); |
| 214 | xfs_ifork_init_cow(ip); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 215 | } |
Darrick J. Wong | 3993bae | 2016-10-03 09:11:32 -0700 | [diff] [blame] | 216 | |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 217 | if (!XFS_DFORK_Q(dip)) |
| 218 | return 0; |
| 219 | |
| 220 | ASSERT(ip->i_afp == NULL); |
| 221 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS); |
| 222 | |
| 223 | switch (dip->di_aformat) { |
| 224 | case XFS_DINODE_FMT_LOCAL: |
| 225 | atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); |
| 226 | size = be16_to_cpu(atp->hdr.totsize); |
| 227 | |
| 228 | if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) { |
| 229 | xfs_warn(ip->i_mount, |
| 230 | "corrupt inode %Lu (bad attr fork size %Ld).", |
| 231 | (unsigned long long) ip->i_ino, |
| 232 | (long long) size); |
| 233 | XFS_CORRUPTION_ERROR("xfs_iformat(8)", |
| 234 | XFS_ERRLEVEL_LOW, |
| 235 | ip->i_mount, dip); |
Darrick J. Wong | 11715a2 | 2016-10-03 09:11:31 -0700 | [diff] [blame] | 236 | error = -EFSCORRUPTED; |
| 237 | break; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 238 | } |
| 239 | |
| 240 | error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); |
| 241 | break; |
| 242 | case XFS_DINODE_FMT_EXTENTS: |
| 243 | error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); |
| 244 | break; |
| 245 | case XFS_DINODE_FMT_BTREE: |
| 246 | error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); |
| 247 | break; |
| 248 | default: |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 249 | error = -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 250 | break; |
| 251 | } |
| 252 | if (error) { |
| 253 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); |
| 254 | ip->i_afp = NULL; |
Darrick J. Wong | 3993bae | 2016-10-03 09:11:32 -0700 | [diff] [blame] | 255 | if (ip->i_cowfp) |
| 256 | kmem_zone_free(xfs_ifork_zone, ip->i_cowfp); |
| 257 | ip->i_cowfp = NULL; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 258 | xfs_idestroy_fork(ip, XFS_DATA_FORK); |
| 259 | } |
| 260 | return error; |
| 261 | } |
| 262 | |
Christoph Hellwig | 143f4ae | 2016-04-06 07:41:43 +1000 | [diff] [blame] | 263 | void |
| 264 | xfs_init_local_fork( |
| 265 | struct xfs_inode *ip, |
| 266 | int whichfork, |
| 267 | const void *data, |
| 268 | int size) |
| 269 | { |
| 270 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
Christoph Hellwig | 30ee052 | 2016-04-06 07:53:29 +1000 | [diff] [blame] | 271 | int mem_size = size, real_size = 0; |
| 272 | bool zero_terminate; |
| 273 | |
| 274 | /* |
| 275 | * If we are using the local fork to store a symlink body we need to |
| 276 | * zero-terminate it so that we can pass it back to the VFS directly. |
| 277 | * Overallocate the in-memory fork by one for that and add a zero |
| 278 | * to terminate it below. |
| 279 | */ |
| 280 | zero_terminate = S_ISLNK(VFS_I(ip)->i_mode); |
| 281 | if (zero_terminate) |
| 282 | mem_size++; |
Christoph Hellwig | 143f4ae | 2016-04-06 07:41:43 +1000 | [diff] [blame] | 283 | |
| 284 | if (size == 0) |
| 285 | ifp->if_u1.if_data = NULL; |
Christoph Hellwig | 30ee052 | 2016-04-06 07:53:29 +1000 | [diff] [blame] | 286 | else if (mem_size <= sizeof(ifp->if_u2.if_inline_data)) |
Christoph Hellwig | 143f4ae | 2016-04-06 07:41:43 +1000 | [diff] [blame] | 287 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; |
| 288 | else { |
Christoph Hellwig | 30ee052 | 2016-04-06 07:53:29 +1000 | [diff] [blame] | 289 | real_size = roundup(mem_size, 4); |
Christoph Hellwig | 143f4ae | 2016-04-06 07:41:43 +1000 | [diff] [blame] | 290 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS); |
| 291 | } |
| 292 | |
Christoph Hellwig | 30ee052 | 2016-04-06 07:53:29 +1000 | [diff] [blame] | 293 | if (size) { |
Christoph Hellwig | 143f4ae | 2016-04-06 07:41:43 +1000 | [diff] [blame] | 294 | memcpy(ifp->if_u1.if_data, data, size); |
Christoph Hellwig | 30ee052 | 2016-04-06 07:53:29 +1000 | [diff] [blame] | 295 | if (zero_terminate) |
| 296 | ifp->if_u1.if_data[size] = '\0'; |
| 297 | } |
Christoph Hellwig | 143f4ae | 2016-04-06 07:41:43 +1000 | [diff] [blame] | 298 | |
| 299 | ifp->if_bytes = size; |
| 300 | ifp->if_real_bytes = real_size; |
| 301 | ifp->if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); |
| 302 | ifp->if_flags |= XFS_IFINLINE; |
| 303 | } |
| 304 | |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 305 | /* |
| 306 | * The file is in-lined in the on-disk inode. |
| 307 | * If it fits into if_inline_data, then copy |
| 308 | * it there, otherwise allocate a buffer for it |
| 309 | * and copy the data there. Either way, set |
| 310 | * if_data to point at the data. |
| 311 | * If we allocate a buffer for the data, make |
| 312 | * sure that its size is a multiple of 4 and |
| 313 | * record the real size in i_real_bytes. |
| 314 | */ |
| 315 | STATIC int |
| 316 | xfs_iformat_local( |
| 317 | xfs_inode_t *ip, |
| 318 | xfs_dinode_t *dip, |
| 319 | int whichfork, |
| 320 | int size) |
| 321 | { |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 322 | |
| 323 | /* |
| 324 | * If the size is unreasonable, then something |
| 325 | * is wrong and we just bail out rather than crash in |
| 326 | * kmem_alloc() or memcpy() below. |
| 327 | */ |
| 328 | if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { |
| 329 | xfs_warn(ip->i_mount, |
| 330 | "corrupt inode %Lu (bad size %d for local fork, size = %d).", |
| 331 | (unsigned long long) ip->i_ino, size, |
| 332 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); |
| 333 | XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, |
| 334 | ip->i_mount, dip); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 335 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 336 | } |
Christoph Hellwig | 143f4ae | 2016-04-06 07:41:43 +1000 | [diff] [blame] | 337 | |
| 338 | xfs_init_local_fork(ip, whichfork, XFS_DFORK_PTR(dip, whichfork), size); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 339 | return 0; |
| 340 | } |
| 341 | |
| 342 | /* |
| 343 | * The file consists of a set of extents all |
| 344 | * of which fit into the on-disk inode. |
| 345 | * If there are few enough extents to fit into |
| 346 | * the if_inline_ext, then copy them there. |
| 347 | * Otherwise allocate a buffer for them and copy |
| 348 | * them into it. Either way, set if_extents |
| 349 | * to point at the extents. |
| 350 | */ |
| 351 | STATIC int |
| 352 | xfs_iformat_extents( |
| 353 | xfs_inode_t *ip, |
| 354 | xfs_dinode_t *dip, |
| 355 | int whichfork) |
| 356 | { |
| 357 | xfs_bmbt_rec_t *dp; |
| 358 | xfs_ifork_t *ifp; |
| 359 | int nex; |
| 360 | int size; |
| 361 | int i; |
| 362 | |
| 363 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 364 | nex = XFS_DFORK_NEXTENTS(dip, whichfork); |
| 365 | size = nex * (uint)sizeof(xfs_bmbt_rec_t); |
| 366 | |
| 367 | /* |
| 368 | * If the number of extents is unreasonable, then something |
| 369 | * is wrong and we just bail out rather than crash in |
| 370 | * kmem_alloc() or memcpy() below. |
| 371 | */ |
| 372 | if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { |
| 373 | xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).", |
| 374 | (unsigned long long) ip->i_ino, nex); |
| 375 | XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, |
| 376 | ip->i_mount, dip); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 377 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 378 | } |
| 379 | |
| 380 | ifp->if_real_bytes = 0; |
| 381 | if (nex == 0) |
| 382 | ifp->if_u1.if_extents = NULL; |
| 383 | else if (nex <= XFS_INLINE_EXTS) |
| 384 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; |
| 385 | else |
| 386 | xfs_iext_add(ifp, 0, nex); |
| 387 | |
| 388 | ifp->if_bytes = size; |
| 389 | if (size) { |
| 390 | dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); |
| 391 | xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip)); |
| 392 | for (i = 0; i < nex; i++, dp++) { |
| 393 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
| 394 | ep->l0 = get_unaligned_be64(&dp->l0); |
| 395 | ep->l1 = get_unaligned_be64(&dp->l1); |
| 396 | } |
| 397 | XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork); |
| 398 | if (whichfork != XFS_DATA_FORK || |
| 399 | XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) |
| 400 | if (unlikely(xfs_check_nostate_extents( |
| 401 | ifp, 0, nex))) { |
| 402 | XFS_ERROR_REPORT("xfs_iformat_extents(2)", |
| 403 | XFS_ERRLEVEL_LOW, |
| 404 | ip->i_mount); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 405 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 406 | } |
| 407 | } |
| 408 | ifp->if_flags |= XFS_IFEXTENTS; |
| 409 | return 0; |
| 410 | } |
| 411 | |
| 412 | /* |
| 413 | * The file has too many extents to fit into |
| 414 | * the inode, so they are in B-tree format. |
| 415 | * Allocate a buffer for the root of the B-tree |
| 416 | * and copy the root into it. The i_extents |
| 417 | * field will remain NULL until all of the |
| 418 | * extents are read in (when they are needed). |
| 419 | */ |
| 420 | STATIC int |
| 421 | xfs_iformat_btree( |
| 422 | xfs_inode_t *ip, |
| 423 | xfs_dinode_t *dip, |
| 424 | int whichfork) |
| 425 | { |
| 426 | struct xfs_mount *mp = ip->i_mount; |
| 427 | xfs_bmdr_block_t *dfp; |
| 428 | xfs_ifork_t *ifp; |
| 429 | /* REFERENCED */ |
| 430 | int nrecs; |
| 431 | int size; |
| 432 | |
| 433 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 434 | dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); |
| 435 | size = XFS_BMAP_BROOT_SPACE(mp, dfp); |
| 436 | nrecs = be16_to_cpu(dfp->bb_numrecs); |
| 437 | |
| 438 | /* |
| 439 | * blow out if -- fork has less extents than can fit in |
| 440 | * fork (fork shouldn't be a btree format), root btree |
| 441 | * block has more records than can fit into the fork, |
| 442 | * or the number of extents is greater than the number of |
| 443 | * blocks. |
| 444 | */ |
| 445 | if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= |
| 446 | XFS_IFORK_MAXEXT(ip, whichfork) || |
| 447 | XFS_BMDR_SPACE_CALC(nrecs) > |
| 448 | XFS_DFORK_SIZE(dip, mp, whichfork) || |
| 449 | XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { |
| 450 | xfs_warn(mp, "corrupt inode %Lu (btree).", |
| 451 | (unsigned long long) ip->i_ino); |
| 452 | XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW, |
| 453 | mp, dip); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 454 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 455 | } |
| 456 | |
| 457 | ifp->if_broot_bytes = size; |
| 458 | ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS); |
| 459 | ASSERT(ifp->if_broot != NULL); |
| 460 | /* |
| 461 | * Copy and convert from the on-disk structure |
| 462 | * to the in-memory structure. |
| 463 | */ |
| 464 | xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), |
| 465 | ifp->if_broot, size); |
| 466 | ifp->if_flags &= ~XFS_IFEXTENTS; |
| 467 | ifp->if_flags |= XFS_IFBROOT; |
| 468 | |
| 469 | return 0; |
| 470 | } |
| 471 | |
| 472 | /* |
| 473 | * Read in extents from a btree-format inode. |
| 474 | * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. |
| 475 | */ |
| 476 | int |
| 477 | xfs_iread_extents( |
| 478 | xfs_trans_t *tp, |
| 479 | xfs_inode_t *ip, |
| 480 | int whichfork) |
| 481 | { |
| 482 | int error; |
| 483 | xfs_ifork_t *ifp; |
| 484 | xfs_extnum_t nextents; |
| 485 | |
Christoph Hellwig | eef334e | 2013-12-06 12:30:17 -0800 | [diff] [blame] | 486 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
| 487 | |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 488 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { |
| 489 | XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, |
| 490 | ip->i_mount); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 491 | return -EFSCORRUPTED; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 492 | } |
| 493 | nextents = XFS_IFORK_NEXTENTS(ip, whichfork); |
| 494 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 495 | |
| 496 | /* |
| 497 | * We know that the size is valid (it's checked in iformat_btree) |
| 498 | */ |
| 499 | ifp->if_bytes = ifp->if_real_bytes = 0; |
| 500 | ifp->if_flags |= XFS_IFEXTENTS; |
| 501 | xfs_iext_add(ifp, 0, nextents); |
| 502 | error = xfs_bmap_read_extents(tp, ip, whichfork); |
| 503 | if (error) { |
| 504 | xfs_iext_destroy(ifp); |
| 505 | ifp->if_flags &= ~XFS_IFEXTENTS; |
| 506 | return error; |
| 507 | } |
| 508 | xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip)); |
| 509 | return 0; |
| 510 | } |
| 511 | /* |
| 512 | * Reallocate the space for if_broot based on the number of records |
| 513 | * being added or deleted as indicated in rec_diff. Move the records |
| 514 | * and pointers in if_broot to fit the new size. When shrinking this |
| 515 | * will eliminate holes between the records and pointers created by |
| 516 | * the caller. When growing this will create holes to be filled in |
| 517 | * by the caller. |
| 518 | * |
| 519 | * The caller must not request to add more records than would fit in |
| 520 | * the on-disk inode root. If the if_broot is currently NULL, then |
Zhi Yong Wu | f6c2734 | 2013-08-07 10:11:04 +0000 | [diff] [blame] | 521 | * if we are adding records, one will be allocated. The caller must also |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 522 | * not request that the number of records go below zero, although |
| 523 | * it can go to zero. |
| 524 | * |
| 525 | * ip -- the inode whose if_broot area is changing |
| 526 | * ext_diff -- the change in the number of records, positive or negative, |
| 527 | * requested for the if_broot array. |
| 528 | */ |
| 529 | void |
| 530 | xfs_iroot_realloc( |
| 531 | xfs_inode_t *ip, |
| 532 | int rec_diff, |
| 533 | int whichfork) |
| 534 | { |
| 535 | struct xfs_mount *mp = ip->i_mount; |
| 536 | int cur_max; |
| 537 | xfs_ifork_t *ifp; |
| 538 | struct xfs_btree_block *new_broot; |
| 539 | int new_max; |
| 540 | size_t new_size; |
| 541 | char *np; |
| 542 | char *op; |
| 543 | |
| 544 | /* |
| 545 | * Handle the degenerate case quietly. |
| 546 | */ |
| 547 | if (rec_diff == 0) { |
| 548 | return; |
| 549 | } |
| 550 | |
| 551 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 552 | if (rec_diff > 0) { |
| 553 | /* |
| 554 | * If there wasn't any memory allocated before, just |
| 555 | * allocate it now and get out. |
| 556 | */ |
| 557 | if (ifp->if_broot_bytes == 0) { |
| 558 | new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff); |
| 559 | ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); |
| 560 | ifp->if_broot_bytes = (int)new_size; |
| 561 | return; |
| 562 | } |
| 563 | |
| 564 | /* |
| 565 | * If there is already an existing if_broot, then we need |
| 566 | * to realloc() it and shift the pointers to their new |
| 567 | * location. The records don't change location because |
| 568 | * they are kept butted up against the btree block header. |
| 569 | */ |
| 570 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
| 571 | new_max = cur_max + rec_diff; |
| 572 | new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max); |
| 573 | ifp->if_broot = kmem_realloc(ifp->if_broot, new_size, |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 574 | KM_SLEEP | KM_NOFS); |
| 575 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
| 576 | ifp->if_broot_bytes); |
| 577 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
| 578 | (int)new_size); |
| 579 | ifp->if_broot_bytes = (int)new_size; |
| 580 | ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <= |
| 581 | XFS_IFORK_SIZE(ip, whichfork)); |
Christoph Hellwig | d5cf09b | 2014-07-30 09:12:05 +1000 | [diff] [blame] | 582 | memmove(np, op, cur_max * (uint)sizeof(xfs_fsblock_t)); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 583 | return; |
| 584 | } |
| 585 | |
| 586 | /* |
| 587 | * rec_diff is less than 0. In this case, we are shrinking the |
| 588 | * if_broot buffer. It must already exist. If we go to zero |
| 589 | * records, just get rid of the root and clear the status bit. |
| 590 | */ |
| 591 | ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); |
| 592 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
| 593 | new_max = cur_max + rec_diff; |
| 594 | ASSERT(new_max >= 0); |
| 595 | if (new_max > 0) |
| 596 | new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max); |
| 597 | else |
| 598 | new_size = 0; |
| 599 | if (new_size > 0) { |
| 600 | new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); |
| 601 | /* |
| 602 | * First copy over the btree block header. |
| 603 | */ |
| 604 | memcpy(new_broot, ifp->if_broot, |
| 605 | XFS_BMBT_BLOCK_LEN(ip->i_mount)); |
| 606 | } else { |
| 607 | new_broot = NULL; |
| 608 | ifp->if_flags &= ~XFS_IFBROOT; |
| 609 | } |
| 610 | |
| 611 | /* |
| 612 | * Only copy the records and pointers if there are any. |
| 613 | */ |
| 614 | if (new_max > 0) { |
| 615 | /* |
| 616 | * First copy the records. |
| 617 | */ |
| 618 | op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1); |
| 619 | np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1); |
| 620 | memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); |
| 621 | |
| 622 | /* |
| 623 | * Then copy the pointers. |
| 624 | */ |
| 625 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
| 626 | ifp->if_broot_bytes); |
| 627 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1, |
| 628 | (int)new_size); |
Christoph Hellwig | d5cf09b | 2014-07-30 09:12:05 +1000 | [diff] [blame] | 629 | memcpy(np, op, new_max * (uint)sizeof(xfs_fsblock_t)); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 630 | } |
| 631 | kmem_free(ifp->if_broot); |
| 632 | ifp->if_broot = new_broot; |
| 633 | ifp->if_broot_bytes = (int)new_size; |
| 634 | if (ifp->if_broot) |
| 635 | ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <= |
| 636 | XFS_IFORK_SIZE(ip, whichfork)); |
| 637 | return; |
| 638 | } |
| 639 | |
| 640 | |
| 641 | /* |
| 642 | * This is called when the amount of space needed for if_data |
| 643 | * is increased or decreased. The change in size is indicated by |
| 644 | * the number of bytes that need to be added or deleted in the |
| 645 | * byte_diff parameter. |
| 646 | * |
| 647 | * If the amount of space needed has decreased below the size of the |
| 648 | * inline buffer, then switch to using the inline buffer. Otherwise, |
| 649 | * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer |
| 650 | * to what is needed. |
| 651 | * |
| 652 | * ip -- the inode whose if_data area is changing |
| 653 | * byte_diff -- the change in the number of bytes, positive or negative, |
| 654 | * requested for the if_data array. |
| 655 | */ |
| 656 | void |
| 657 | xfs_idata_realloc( |
| 658 | xfs_inode_t *ip, |
| 659 | int byte_diff, |
| 660 | int whichfork) |
| 661 | { |
| 662 | xfs_ifork_t *ifp; |
| 663 | int new_size; |
| 664 | int real_size; |
| 665 | |
| 666 | if (byte_diff == 0) { |
| 667 | return; |
| 668 | } |
| 669 | |
| 670 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 671 | new_size = (int)ifp->if_bytes + byte_diff; |
| 672 | ASSERT(new_size >= 0); |
| 673 | |
| 674 | if (new_size == 0) { |
| 675 | if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { |
| 676 | kmem_free(ifp->if_u1.if_data); |
| 677 | } |
| 678 | ifp->if_u1.if_data = NULL; |
| 679 | real_size = 0; |
| 680 | } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { |
| 681 | /* |
| 682 | * If the valid extents/data can fit in if_inline_ext/data, |
| 683 | * copy them from the malloc'd vector and free it. |
| 684 | */ |
| 685 | if (ifp->if_u1.if_data == NULL) { |
| 686 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; |
| 687 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { |
| 688 | ASSERT(ifp->if_real_bytes != 0); |
| 689 | memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, |
| 690 | new_size); |
| 691 | kmem_free(ifp->if_u1.if_data); |
| 692 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; |
| 693 | } |
| 694 | real_size = 0; |
| 695 | } else { |
| 696 | /* |
| 697 | * Stuck with malloc/realloc. |
| 698 | * For inline data, the underlying buffer must be |
| 699 | * a multiple of 4 bytes in size so that it can be |
| 700 | * logged and stay on word boundaries. We enforce |
| 701 | * that here. |
| 702 | */ |
| 703 | real_size = roundup(new_size, 4); |
| 704 | if (ifp->if_u1.if_data == NULL) { |
| 705 | ASSERT(ifp->if_real_bytes == 0); |
| 706 | ifp->if_u1.if_data = kmem_alloc(real_size, |
| 707 | KM_SLEEP | KM_NOFS); |
| 708 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { |
| 709 | /* |
| 710 | * Only do the realloc if the underlying size |
| 711 | * is really changing. |
| 712 | */ |
| 713 | if (ifp->if_real_bytes != real_size) { |
| 714 | ifp->if_u1.if_data = |
| 715 | kmem_realloc(ifp->if_u1.if_data, |
| 716 | real_size, |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 717 | KM_SLEEP | KM_NOFS); |
| 718 | } |
| 719 | } else { |
| 720 | ASSERT(ifp->if_real_bytes == 0); |
| 721 | ifp->if_u1.if_data = kmem_alloc(real_size, |
| 722 | KM_SLEEP | KM_NOFS); |
| 723 | memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, |
| 724 | ifp->if_bytes); |
| 725 | } |
| 726 | } |
| 727 | ifp->if_real_bytes = real_size; |
| 728 | ifp->if_bytes = new_size; |
| 729 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); |
| 730 | } |
| 731 | |
| 732 | void |
| 733 | xfs_idestroy_fork( |
| 734 | xfs_inode_t *ip, |
| 735 | int whichfork) |
| 736 | { |
| 737 | xfs_ifork_t *ifp; |
| 738 | |
| 739 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 740 | if (ifp->if_broot != NULL) { |
| 741 | kmem_free(ifp->if_broot); |
| 742 | ifp->if_broot = NULL; |
| 743 | } |
| 744 | |
| 745 | /* |
| 746 | * If the format is local, then we can't have an extents |
| 747 | * array so just look for an inline data array. If we're |
| 748 | * not local then we may or may not have an extents list, |
| 749 | * so check and free it up if we do. |
| 750 | */ |
| 751 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { |
| 752 | if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && |
| 753 | (ifp->if_u1.if_data != NULL)) { |
| 754 | ASSERT(ifp->if_real_bytes != 0); |
| 755 | kmem_free(ifp->if_u1.if_data); |
| 756 | ifp->if_u1.if_data = NULL; |
| 757 | ifp->if_real_bytes = 0; |
| 758 | } |
| 759 | } else if ((ifp->if_flags & XFS_IFEXTENTS) && |
| 760 | ((ifp->if_flags & XFS_IFEXTIREC) || |
| 761 | ((ifp->if_u1.if_extents != NULL) && |
| 762 | (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { |
| 763 | ASSERT(ifp->if_real_bytes != 0); |
| 764 | xfs_iext_destroy(ifp); |
| 765 | } |
| 766 | ASSERT(ifp->if_u1.if_extents == NULL || |
| 767 | ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); |
| 768 | ASSERT(ifp->if_real_bytes == 0); |
| 769 | if (whichfork == XFS_ATTR_FORK) { |
| 770 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); |
| 771 | ip->i_afp = NULL; |
Darrick J. Wong | 3993bae | 2016-10-03 09:11:32 -0700 | [diff] [blame] | 772 | } else if (whichfork == XFS_COW_FORK) { |
| 773 | kmem_zone_free(xfs_ifork_zone, ip->i_cowfp); |
| 774 | ip->i_cowfp = NULL; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 775 | } |
| 776 | } |
| 777 | |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 778 | /* Count number of incore extents based on if_bytes */ |
| 779 | xfs_extnum_t |
| 780 | xfs_iext_count(struct xfs_ifork *ifp) |
| 781 | { |
| 782 | return ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); |
| 783 | } |
| 784 | |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 785 | /* |
Christoph Hellwig | da77650 | 2013-12-13 11:34:04 +1100 | [diff] [blame] | 786 | * Convert in-core extents to on-disk form |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 787 | * |
Christoph Hellwig | da77650 | 2013-12-13 11:34:04 +1100 | [diff] [blame] | 788 | * For either the data or attr fork in extent format, we need to endian convert |
| 789 | * the in-core extent as we place them into the on-disk inode. |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 790 | * |
Christoph Hellwig | da77650 | 2013-12-13 11:34:04 +1100 | [diff] [blame] | 791 | * In the case of the data fork, the in-core and on-disk fork sizes can be |
| 792 | * different due to delayed allocation extents. We only copy on-disk extents |
| 793 | * here, so callers must always use the physical fork size to determine the |
| 794 | * size of the buffer passed to this routine. We will return the size actually |
| 795 | * used. |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 796 | */ |
| 797 | int |
| 798 | xfs_iextents_copy( |
| 799 | xfs_inode_t *ip, |
| 800 | xfs_bmbt_rec_t *dp, |
| 801 | int whichfork) |
| 802 | { |
| 803 | int copied; |
| 804 | int i; |
| 805 | xfs_ifork_t *ifp; |
| 806 | int nrecs; |
| 807 | xfs_fsblock_t start_block; |
| 808 | |
| 809 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 810 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
| 811 | ASSERT(ifp->if_bytes > 0); |
| 812 | |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 813 | nrecs = xfs_iext_count(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 814 | XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork); |
| 815 | ASSERT(nrecs > 0); |
| 816 | |
| 817 | /* |
| 818 | * There are some delayed allocation extents in the |
| 819 | * inode, so copy the extents one at a time and skip |
| 820 | * the delayed ones. There must be at least one |
| 821 | * non-delayed extent. |
| 822 | */ |
| 823 | copied = 0; |
| 824 | for (i = 0; i < nrecs; i++) { |
| 825 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
| 826 | start_block = xfs_bmbt_get_startblock(ep); |
| 827 | if (isnullstartblock(start_block)) { |
| 828 | /* |
| 829 | * It's a delayed allocation extent, so skip it. |
| 830 | */ |
| 831 | continue; |
| 832 | } |
| 833 | |
| 834 | /* Translate to on disk format */ |
Dave Chinner | c5c249b | 2013-08-12 20:49:43 +1000 | [diff] [blame] | 835 | put_unaligned_be64(ep->l0, &dp->l0); |
| 836 | put_unaligned_be64(ep->l1, &dp->l1); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 837 | dp++; |
| 838 | copied++; |
| 839 | } |
| 840 | ASSERT(copied != 0); |
| 841 | xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip)); |
| 842 | |
| 843 | return (copied * (uint)sizeof(xfs_bmbt_rec_t)); |
| 844 | } |
| 845 | |
| 846 | /* |
| 847 | * Each of the following cases stores data into the same region |
| 848 | * of the on-disk inode, so only one of them can be valid at |
| 849 | * any given time. While it is possible to have conflicting formats |
| 850 | * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is |
| 851 | * in EXTENTS format, this can only happen when the fork has |
| 852 | * changed formats after being modified but before being flushed. |
| 853 | * In these cases, the format always takes precedence, because the |
| 854 | * format indicates the current state of the fork. |
| 855 | */ |
| 856 | void |
| 857 | xfs_iflush_fork( |
| 858 | xfs_inode_t *ip, |
| 859 | xfs_dinode_t *dip, |
| 860 | xfs_inode_log_item_t *iip, |
Eric Sandeen | fd9fdba | 2014-04-14 19:04:46 +1000 | [diff] [blame] | 861 | int whichfork) |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 862 | { |
| 863 | char *cp; |
| 864 | xfs_ifork_t *ifp; |
| 865 | xfs_mount_t *mp; |
| 866 | static const short brootflag[2] = |
| 867 | { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; |
| 868 | static const short dataflag[2] = |
| 869 | { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; |
| 870 | static const short extflag[2] = |
| 871 | { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; |
| 872 | |
| 873 | if (!iip) |
| 874 | return; |
| 875 | ifp = XFS_IFORK_PTR(ip, whichfork); |
| 876 | /* |
| 877 | * This can happen if we gave up in iformat in an error path, |
| 878 | * for the attribute fork. |
| 879 | */ |
| 880 | if (!ifp) { |
| 881 | ASSERT(whichfork == XFS_ATTR_FORK); |
| 882 | return; |
| 883 | } |
| 884 | cp = XFS_DFORK_PTR(dip, whichfork); |
| 885 | mp = ip->i_mount; |
| 886 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { |
| 887 | case XFS_DINODE_FMT_LOCAL: |
| 888 | if ((iip->ili_fields & dataflag[whichfork]) && |
| 889 | (ifp->if_bytes > 0)) { |
| 890 | ASSERT(ifp->if_u1.if_data != NULL); |
| 891 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); |
| 892 | memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); |
| 893 | } |
| 894 | break; |
| 895 | |
| 896 | case XFS_DINODE_FMT_EXTENTS: |
| 897 | ASSERT((ifp->if_flags & XFS_IFEXTENTS) || |
| 898 | !(iip->ili_fields & extflag[whichfork])); |
| 899 | if ((iip->ili_fields & extflag[whichfork]) && |
| 900 | (ifp->if_bytes > 0)) { |
| 901 | ASSERT(xfs_iext_get_ext(ifp, 0)); |
| 902 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); |
| 903 | (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, |
| 904 | whichfork); |
| 905 | } |
| 906 | break; |
| 907 | |
| 908 | case XFS_DINODE_FMT_BTREE: |
| 909 | if ((iip->ili_fields & brootflag[whichfork]) && |
| 910 | (ifp->if_broot_bytes > 0)) { |
| 911 | ASSERT(ifp->if_broot != NULL); |
| 912 | ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <= |
| 913 | XFS_IFORK_SIZE(ip, whichfork)); |
| 914 | xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes, |
| 915 | (xfs_bmdr_block_t *)cp, |
| 916 | XFS_DFORK_SIZE(dip, mp, whichfork)); |
| 917 | } |
| 918 | break; |
| 919 | |
| 920 | case XFS_DINODE_FMT_DEV: |
| 921 | if (iip->ili_fields & XFS_ILOG_DEV) { |
| 922 | ASSERT(whichfork == XFS_DATA_FORK); |
| 923 | xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev); |
| 924 | } |
| 925 | break; |
| 926 | |
| 927 | case XFS_DINODE_FMT_UUID: |
| 928 | if (iip->ili_fields & XFS_ILOG_UUID) { |
| 929 | ASSERT(whichfork == XFS_DATA_FORK); |
| 930 | memcpy(XFS_DFORK_DPTR(dip), |
| 931 | &ip->i_df.if_u2.if_uuid, |
| 932 | sizeof(uuid_t)); |
| 933 | } |
| 934 | break; |
| 935 | |
| 936 | default: |
| 937 | ASSERT(0); |
| 938 | break; |
| 939 | } |
| 940 | } |
| 941 | |
| 942 | /* |
| 943 | * Return a pointer to the extent record at file index idx. |
| 944 | */ |
| 945 | xfs_bmbt_rec_host_t * |
| 946 | xfs_iext_get_ext( |
| 947 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 948 | xfs_extnum_t idx) /* index of target extent */ |
| 949 | { |
| 950 | ASSERT(idx >= 0); |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 951 | ASSERT(idx < xfs_iext_count(ifp)); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 952 | |
| 953 | if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { |
| 954 | return ifp->if_u1.if_ext_irec->er_extbuf; |
| 955 | } else if (ifp->if_flags & XFS_IFEXTIREC) { |
| 956 | xfs_ext_irec_t *erp; /* irec pointer */ |
| 957 | int erp_idx = 0; /* irec index */ |
| 958 | xfs_extnum_t page_idx = idx; /* ext index in target list */ |
| 959 | |
| 960 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); |
| 961 | return &erp->er_extbuf[page_idx]; |
| 962 | } else if (ifp->if_bytes) { |
| 963 | return &ifp->if_u1.if_extents[idx]; |
| 964 | } else { |
| 965 | return NULL; |
| 966 | } |
| 967 | } |
| 968 | |
Darrick J. Wong | 3993bae | 2016-10-03 09:11:32 -0700 | [diff] [blame] | 969 | /* Convert bmap state flags to an inode fork. */ |
| 970 | struct xfs_ifork * |
| 971 | xfs_iext_state_to_fork( |
| 972 | struct xfs_inode *ip, |
| 973 | int state) |
| 974 | { |
| 975 | if (state & BMAP_COWFORK) |
| 976 | return ip->i_cowfp; |
| 977 | else if (state & BMAP_ATTRFORK) |
| 978 | return ip->i_afp; |
| 979 | return &ip->i_df; |
| 980 | } |
| 981 | |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 982 | /* |
| 983 | * Insert new item(s) into the extent records for incore inode |
| 984 | * fork 'ifp'. 'count' new items are inserted at index 'idx'. |
| 985 | */ |
| 986 | void |
| 987 | xfs_iext_insert( |
| 988 | xfs_inode_t *ip, /* incore inode pointer */ |
| 989 | xfs_extnum_t idx, /* starting index of new items */ |
| 990 | xfs_extnum_t count, /* number of inserted items */ |
| 991 | xfs_bmbt_irec_t *new, /* items to insert */ |
| 992 | int state) /* type of extent conversion */ |
| 993 | { |
Darrick J. Wong | 3993bae | 2016-10-03 09:11:32 -0700 | [diff] [blame] | 994 | xfs_ifork_t *ifp = xfs_iext_state_to_fork(ip, state); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 995 | xfs_extnum_t i; /* extent record index */ |
| 996 | |
| 997 | trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_); |
| 998 | |
| 999 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
| 1000 | xfs_iext_add(ifp, idx, count); |
| 1001 | for (i = idx; i < idx + count; i++, new++) |
| 1002 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new); |
| 1003 | } |
| 1004 | |
| 1005 | /* |
| 1006 | * This is called when the amount of space required for incore file |
| 1007 | * extents needs to be increased. The ext_diff parameter stores the |
| 1008 | * number of new extents being added and the idx parameter contains |
| 1009 | * the extent index where the new extents will be added. If the new |
| 1010 | * extents are being appended, then we just need to (re)allocate and |
| 1011 | * initialize the space. Otherwise, if the new extents are being |
| 1012 | * inserted into the middle of the existing entries, a bit more work |
| 1013 | * is required to make room for the new extents to be inserted. The |
| 1014 | * caller is responsible for filling in the new extent entries upon |
| 1015 | * return. |
| 1016 | */ |
| 1017 | void |
| 1018 | xfs_iext_add( |
| 1019 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1020 | xfs_extnum_t idx, /* index to begin adding exts */ |
| 1021 | int ext_diff) /* number of extents to add */ |
| 1022 | { |
| 1023 | int byte_diff; /* new bytes being added */ |
| 1024 | int new_size; /* size of extents after adding */ |
| 1025 | xfs_extnum_t nextents; /* number of extents in file */ |
| 1026 | |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 1027 | nextents = xfs_iext_count(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1028 | ASSERT((idx >= 0) && (idx <= nextents)); |
| 1029 | byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); |
| 1030 | new_size = ifp->if_bytes + byte_diff; |
| 1031 | /* |
| 1032 | * If the new number of extents (nextents + ext_diff) |
| 1033 | * fits inside the inode, then continue to use the inline |
| 1034 | * extent buffer. |
| 1035 | */ |
| 1036 | if (nextents + ext_diff <= XFS_INLINE_EXTS) { |
| 1037 | if (idx < nextents) { |
| 1038 | memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], |
| 1039 | &ifp->if_u2.if_inline_ext[idx], |
| 1040 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); |
| 1041 | memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); |
| 1042 | } |
| 1043 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; |
| 1044 | ifp->if_real_bytes = 0; |
| 1045 | } |
| 1046 | /* |
| 1047 | * Otherwise use a linear (direct) extent list. |
| 1048 | * If the extents are currently inside the inode, |
| 1049 | * xfs_iext_realloc_direct will switch us from |
| 1050 | * inline to direct extent allocation mode. |
| 1051 | */ |
| 1052 | else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { |
| 1053 | xfs_iext_realloc_direct(ifp, new_size); |
| 1054 | if (idx < nextents) { |
| 1055 | memmove(&ifp->if_u1.if_extents[idx + ext_diff], |
| 1056 | &ifp->if_u1.if_extents[idx], |
| 1057 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); |
| 1058 | memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); |
| 1059 | } |
| 1060 | } |
| 1061 | /* Indirection array */ |
| 1062 | else { |
| 1063 | xfs_ext_irec_t *erp; |
| 1064 | int erp_idx = 0; |
| 1065 | int page_idx = idx; |
| 1066 | |
| 1067 | ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); |
| 1068 | if (ifp->if_flags & XFS_IFEXTIREC) { |
| 1069 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); |
| 1070 | } else { |
| 1071 | xfs_iext_irec_init(ifp); |
| 1072 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1073 | erp = ifp->if_u1.if_ext_irec; |
| 1074 | } |
| 1075 | /* Extents fit in target extent page */ |
| 1076 | if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { |
| 1077 | if (page_idx < erp->er_extcount) { |
| 1078 | memmove(&erp->er_extbuf[page_idx + ext_diff], |
| 1079 | &erp->er_extbuf[page_idx], |
| 1080 | (erp->er_extcount - page_idx) * |
| 1081 | sizeof(xfs_bmbt_rec_t)); |
| 1082 | memset(&erp->er_extbuf[page_idx], 0, byte_diff); |
| 1083 | } |
| 1084 | erp->er_extcount += ext_diff; |
| 1085 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); |
| 1086 | } |
| 1087 | /* Insert a new extent page */ |
| 1088 | else if (erp) { |
| 1089 | xfs_iext_add_indirect_multi(ifp, |
| 1090 | erp_idx, page_idx, ext_diff); |
| 1091 | } |
| 1092 | /* |
| 1093 | * If extent(s) are being appended to the last page in |
| 1094 | * the indirection array and the new extent(s) don't fit |
| 1095 | * in the page, then erp is NULL and erp_idx is set to |
| 1096 | * the next index needed in the indirection array. |
| 1097 | */ |
| 1098 | else { |
Jie Liu | bb86d21 | 2013-10-25 14:52:44 +0800 | [diff] [blame] | 1099 | uint count = ext_diff; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1100 | |
| 1101 | while (count) { |
| 1102 | erp = xfs_iext_irec_new(ifp, erp_idx); |
Jie Liu | bb86d21 | 2013-10-25 14:52:44 +0800 | [diff] [blame] | 1103 | erp->er_extcount = min(count, XFS_LINEAR_EXTS); |
| 1104 | count -= erp->er_extcount; |
| 1105 | if (count) |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1106 | erp_idx++; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1107 | } |
| 1108 | } |
| 1109 | } |
| 1110 | ifp->if_bytes = new_size; |
| 1111 | } |
| 1112 | |
| 1113 | /* |
| 1114 | * This is called when incore extents are being added to the indirection |
| 1115 | * array and the new extents do not fit in the target extent list. The |
| 1116 | * erp_idx parameter contains the irec index for the target extent list |
| 1117 | * in the indirection array, and the idx parameter contains the extent |
| 1118 | * index within the list. The number of extents being added is stored |
| 1119 | * in the count parameter. |
| 1120 | * |
| 1121 | * |-------| |-------| |
| 1122 | * | | | | idx - number of extents before idx |
| 1123 | * | idx | | count | |
| 1124 | * | | | | count - number of extents being inserted at idx |
| 1125 | * |-------| |-------| |
| 1126 | * | count | | nex2 | nex2 - number of extents after idx + count |
| 1127 | * |-------| |-------| |
| 1128 | */ |
| 1129 | void |
| 1130 | xfs_iext_add_indirect_multi( |
| 1131 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1132 | int erp_idx, /* target extent irec index */ |
| 1133 | xfs_extnum_t idx, /* index within target list */ |
| 1134 | int count) /* new extents being added */ |
| 1135 | { |
| 1136 | int byte_diff; /* new bytes being added */ |
| 1137 | xfs_ext_irec_t *erp; /* pointer to irec entry */ |
| 1138 | xfs_extnum_t ext_diff; /* number of extents to add */ |
| 1139 | xfs_extnum_t ext_cnt; /* new extents still needed */ |
| 1140 | xfs_extnum_t nex2; /* extents after idx + count */ |
| 1141 | xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ |
| 1142 | int nlists; /* number of irec's (lists) */ |
| 1143 | |
| 1144 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1145 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; |
| 1146 | nex2 = erp->er_extcount - idx; |
| 1147 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1148 | |
| 1149 | /* |
| 1150 | * Save second part of target extent list |
| 1151 | * (all extents past */ |
| 1152 | if (nex2) { |
| 1153 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); |
| 1154 | nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS); |
| 1155 | memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); |
| 1156 | erp->er_extcount -= nex2; |
| 1157 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); |
| 1158 | memset(&erp->er_extbuf[idx], 0, byte_diff); |
| 1159 | } |
| 1160 | |
| 1161 | /* |
| 1162 | * Add the new extents to the end of the target |
| 1163 | * list, then allocate new irec record(s) and |
| 1164 | * extent buffer(s) as needed to store the rest |
| 1165 | * of the new extents. |
| 1166 | */ |
| 1167 | ext_cnt = count; |
| 1168 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); |
| 1169 | if (ext_diff) { |
| 1170 | erp->er_extcount += ext_diff; |
| 1171 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); |
| 1172 | ext_cnt -= ext_diff; |
| 1173 | } |
| 1174 | while (ext_cnt) { |
| 1175 | erp_idx++; |
| 1176 | erp = xfs_iext_irec_new(ifp, erp_idx); |
| 1177 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); |
| 1178 | erp->er_extcount = ext_diff; |
| 1179 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); |
| 1180 | ext_cnt -= ext_diff; |
| 1181 | } |
| 1182 | |
| 1183 | /* Add nex2 extents back to indirection array */ |
| 1184 | if (nex2) { |
| 1185 | xfs_extnum_t ext_avail; |
| 1186 | int i; |
| 1187 | |
| 1188 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); |
| 1189 | ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; |
| 1190 | i = 0; |
| 1191 | /* |
| 1192 | * If nex2 extents fit in the current page, append |
| 1193 | * nex2_ep after the new extents. |
| 1194 | */ |
| 1195 | if (nex2 <= ext_avail) { |
| 1196 | i = erp->er_extcount; |
| 1197 | } |
| 1198 | /* |
| 1199 | * Otherwise, check if space is available in the |
| 1200 | * next page. |
| 1201 | */ |
| 1202 | else if ((erp_idx < nlists - 1) && |
| 1203 | (nex2 <= (ext_avail = XFS_LINEAR_EXTS - |
| 1204 | ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { |
| 1205 | erp_idx++; |
| 1206 | erp++; |
| 1207 | /* Create a hole for nex2 extents */ |
| 1208 | memmove(&erp->er_extbuf[nex2], erp->er_extbuf, |
| 1209 | erp->er_extcount * sizeof(xfs_bmbt_rec_t)); |
| 1210 | } |
| 1211 | /* |
| 1212 | * Final choice, create a new extent page for |
| 1213 | * nex2 extents. |
| 1214 | */ |
| 1215 | else { |
| 1216 | erp_idx++; |
| 1217 | erp = xfs_iext_irec_new(ifp, erp_idx); |
| 1218 | } |
| 1219 | memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); |
| 1220 | kmem_free(nex2_ep); |
| 1221 | erp->er_extcount += nex2; |
| 1222 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); |
| 1223 | } |
| 1224 | } |
| 1225 | |
| 1226 | /* |
| 1227 | * This is called when the amount of space required for incore file |
| 1228 | * extents needs to be decreased. The ext_diff parameter stores the |
| 1229 | * number of extents to be removed and the idx parameter contains |
| 1230 | * the extent index where the extents will be removed from. |
| 1231 | * |
| 1232 | * If the amount of space needed has decreased below the linear |
| 1233 | * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous |
| 1234 | * extent array. Otherwise, use kmem_realloc() to adjust the |
| 1235 | * size to what is needed. |
| 1236 | */ |
| 1237 | void |
| 1238 | xfs_iext_remove( |
| 1239 | xfs_inode_t *ip, /* incore inode pointer */ |
| 1240 | xfs_extnum_t idx, /* index to begin removing exts */ |
| 1241 | int ext_diff, /* number of extents to remove */ |
| 1242 | int state) /* type of extent conversion */ |
| 1243 | { |
Darrick J. Wong | 3993bae | 2016-10-03 09:11:32 -0700 | [diff] [blame] | 1244 | xfs_ifork_t *ifp = xfs_iext_state_to_fork(ip, state); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1245 | xfs_extnum_t nextents; /* number of extents in file */ |
| 1246 | int new_size; /* size of extents after removal */ |
| 1247 | |
| 1248 | trace_xfs_iext_remove(ip, idx, state, _RET_IP_); |
| 1249 | |
| 1250 | ASSERT(ext_diff > 0); |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 1251 | nextents = xfs_iext_count(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1252 | new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); |
| 1253 | |
| 1254 | if (new_size == 0) { |
| 1255 | xfs_iext_destroy(ifp); |
| 1256 | } else if (ifp->if_flags & XFS_IFEXTIREC) { |
| 1257 | xfs_iext_remove_indirect(ifp, idx, ext_diff); |
| 1258 | } else if (ifp->if_real_bytes) { |
| 1259 | xfs_iext_remove_direct(ifp, idx, ext_diff); |
| 1260 | } else { |
| 1261 | xfs_iext_remove_inline(ifp, idx, ext_diff); |
| 1262 | } |
| 1263 | ifp->if_bytes = new_size; |
| 1264 | } |
| 1265 | |
| 1266 | /* |
| 1267 | * This removes ext_diff extents from the inline buffer, beginning |
| 1268 | * at extent index idx. |
| 1269 | */ |
| 1270 | void |
| 1271 | xfs_iext_remove_inline( |
| 1272 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1273 | xfs_extnum_t idx, /* index to begin removing exts */ |
| 1274 | int ext_diff) /* number of extents to remove */ |
| 1275 | { |
| 1276 | int nextents; /* number of extents in file */ |
| 1277 | |
| 1278 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
| 1279 | ASSERT(idx < XFS_INLINE_EXTS); |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 1280 | nextents = xfs_iext_count(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1281 | ASSERT(((nextents - ext_diff) > 0) && |
| 1282 | (nextents - ext_diff) < XFS_INLINE_EXTS); |
| 1283 | |
| 1284 | if (idx + ext_diff < nextents) { |
| 1285 | memmove(&ifp->if_u2.if_inline_ext[idx], |
| 1286 | &ifp->if_u2.if_inline_ext[idx + ext_diff], |
| 1287 | (nextents - (idx + ext_diff)) * |
| 1288 | sizeof(xfs_bmbt_rec_t)); |
| 1289 | memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], |
| 1290 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); |
| 1291 | } else { |
| 1292 | memset(&ifp->if_u2.if_inline_ext[idx], 0, |
| 1293 | ext_diff * sizeof(xfs_bmbt_rec_t)); |
| 1294 | } |
| 1295 | } |
| 1296 | |
| 1297 | /* |
| 1298 | * This removes ext_diff extents from a linear (direct) extent list, |
| 1299 | * beginning at extent index idx. If the extents are being removed |
| 1300 | * from the end of the list (ie. truncate) then we just need to re- |
| 1301 | * allocate the list to remove the extra space. Otherwise, if the |
| 1302 | * extents are being removed from the middle of the existing extent |
| 1303 | * entries, then we first need to move the extent records beginning |
| 1304 | * at idx + ext_diff up in the list to overwrite the records being |
| 1305 | * removed, then remove the extra space via kmem_realloc. |
| 1306 | */ |
| 1307 | void |
| 1308 | xfs_iext_remove_direct( |
| 1309 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1310 | xfs_extnum_t idx, /* index to begin removing exts */ |
| 1311 | int ext_diff) /* number of extents to remove */ |
| 1312 | { |
| 1313 | xfs_extnum_t nextents; /* number of extents in file */ |
| 1314 | int new_size; /* size of extents after removal */ |
| 1315 | |
| 1316 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
| 1317 | new_size = ifp->if_bytes - |
| 1318 | (ext_diff * sizeof(xfs_bmbt_rec_t)); |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 1319 | nextents = xfs_iext_count(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1320 | |
| 1321 | if (new_size == 0) { |
| 1322 | xfs_iext_destroy(ifp); |
| 1323 | return; |
| 1324 | } |
| 1325 | /* Move extents up in the list (if needed) */ |
| 1326 | if (idx + ext_diff < nextents) { |
| 1327 | memmove(&ifp->if_u1.if_extents[idx], |
| 1328 | &ifp->if_u1.if_extents[idx + ext_diff], |
| 1329 | (nextents - (idx + ext_diff)) * |
| 1330 | sizeof(xfs_bmbt_rec_t)); |
| 1331 | } |
| 1332 | memset(&ifp->if_u1.if_extents[nextents - ext_diff], |
| 1333 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); |
| 1334 | /* |
| 1335 | * Reallocate the direct extent list. If the extents |
| 1336 | * will fit inside the inode then xfs_iext_realloc_direct |
| 1337 | * will switch from direct to inline extent allocation |
| 1338 | * mode for us. |
| 1339 | */ |
| 1340 | xfs_iext_realloc_direct(ifp, new_size); |
| 1341 | ifp->if_bytes = new_size; |
| 1342 | } |
| 1343 | |
| 1344 | /* |
| 1345 | * This is called when incore extents are being removed from the |
| 1346 | * indirection array and the extents being removed span multiple extent |
| 1347 | * buffers. The idx parameter contains the file extent index where we |
| 1348 | * want to begin removing extents, and the count parameter contains |
| 1349 | * how many extents need to be removed. |
| 1350 | * |
| 1351 | * |-------| |-------| |
| 1352 | * | nex1 | | | nex1 - number of extents before idx |
| 1353 | * |-------| | count | |
| 1354 | * | | | | count - number of extents being removed at idx |
| 1355 | * | count | |-------| |
| 1356 | * | | | nex2 | nex2 - number of extents after idx + count |
| 1357 | * |-------| |-------| |
| 1358 | */ |
| 1359 | void |
| 1360 | xfs_iext_remove_indirect( |
| 1361 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1362 | xfs_extnum_t idx, /* index to begin removing extents */ |
| 1363 | int count) /* number of extents to remove */ |
| 1364 | { |
| 1365 | xfs_ext_irec_t *erp; /* indirection array pointer */ |
| 1366 | int erp_idx = 0; /* indirection array index */ |
| 1367 | xfs_extnum_t ext_cnt; /* extents left to remove */ |
| 1368 | xfs_extnum_t ext_diff; /* extents to remove in current list */ |
| 1369 | xfs_extnum_t nex1; /* number of extents before idx */ |
| 1370 | xfs_extnum_t nex2; /* extents after idx + count */ |
| 1371 | int page_idx = idx; /* index in target extent list */ |
| 1372 | |
| 1373 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1374 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); |
| 1375 | ASSERT(erp != NULL); |
| 1376 | nex1 = page_idx; |
| 1377 | ext_cnt = count; |
| 1378 | while (ext_cnt) { |
| 1379 | nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); |
| 1380 | ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); |
| 1381 | /* |
| 1382 | * Check for deletion of entire list; |
| 1383 | * xfs_iext_irec_remove() updates extent offsets. |
| 1384 | */ |
| 1385 | if (ext_diff == erp->er_extcount) { |
| 1386 | xfs_iext_irec_remove(ifp, erp_idx); |
| 1387 | ext_cnt -= ext_diff; |
| 1388 | nex1 = 0; |
| 1389 | if (ext_cnt) { |
| 1390 | ASSERT(erp_idx < ifp->if_real_bytes / |
| 1391 | XFS_IEXT_BUFSZ); |
| 1392 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; |
| 1393 | nex1 = 0; |
| 1394 | continue; |
| 1395 | } else { |
| 1396 | break; |
| 1397 | } |
| 1398 | } |
| 1399 | /* Move extents up (if needed) */ |
| 1400 | if (nex2) { |
| 1401 | memmove(&erp->er_extbuf[nex1], |
| 1402 | &erp->er_extbuf[nex1 + ext_diff], |
| 1403 | nex2 * sizeof(xfs_bmbt_rec_t)); |
| 1404 | } |
| 1405 | /* Zero out rest of page */ |
| 1406 | memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - |
| 1407 | ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); |
| 1408 | /* Update remaining counters */ |
| 1409 | erp->er_extcount -= ext_diff; |
| 1410 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); |
| 1411 | ext_cnt -= ext_diff; |
| 1412 | nex1 = 0; |
| 1413 | erp_idx++; |
| 1414 | erp++; |
| 1415 | } |
| 1416 | ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); |
| 1417 | xfs_iext_irec_compact(ifp); |
| 1418 | } |
| 1419 | |
| 1420 | /* |
| 1421 | * Create, destroy, or resize a linear (direct) block of extents. |
| 1422 | */ |
| 1423 | void |
| 1424 | xfs_iext_realloc_direct( |
| 1425 | xfs_ifork_t *ifp, /* inode fork pointer */ |
Jie Liu | 17ec81c | 2013-09-22 16:25:15 +0800 | [diff] [blame] | 1426 | int new_size) /* new size of extents after adding */ |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1427 | { |
| 1428 | int rnew_size; /* real new size of extents */ |
| 1429 | |
| 1430 | rnew_size = new_size; |
| 1431 | |
| 1432 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || |
| 1433 | ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && |
| 1434 | (new_size != ifp->if_real_bytes))); |
| 1435 | |
| 1436 | /* Free extent records */ |
| 1437 | if (new_size == 0) { |
| 1438 | xfs_iext_destroy(ifp); |
| 1439 | } |
| 1440 | /* Resize direct extent list and zero any new bytes */ |
| 1441 | else if (ifp->if_real_bytes) { |
| 1442 | /* Check if extents will fit inside the inode */ |
| 1443 | if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { |
| 1444 | xfs_iext_direct_to_inline(ifp, new_size / |
| 1445 | (uint)sizeof(xfs_bmbt_rec_t)); |
| 1446 | ifp->if_bytes = new_size; |
| 1447 | return; |
| 1448 | } |
| 1449 | if (!is_power_of_2(new_size)){ |
| 1450 | rnew_size = roundup_pow_of_two(new_size); |
| 1451 | } |
| 1452 | if (rnew_size != ifp->if_real_bytes) { |
| 1453 | ifp->if_u1.if_extents = |
| 1454 | kmem_realloc(ifp->if_u1.if_extents, |
Christoph Hellwig | 664b60f | 2016-04-06 09:47:01 +1000 | [diff] [blame] | 1455 | rnew_size, KM_NOFS); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1456 | } |
| 1457 | if (rnew_size > ifp->if_real_bytes) { |
| 1458 | memset(&ifp->if_u1.if_extents[ifp->if_bytes / |
| 1459 | (uint)sizeof(xfs_bmbt_rec_t)], 0, |
| 1460 | rnew_size - ifp->if_real_bytes); |
| 1461 | } |
| 1462 | } |
Jie Liu | 17ec81c | 2013-09-22 16:25:15 +0800 | [diff] [blame] | 1463 | /* Switch from the inline extent buffer to a direct extent list */ |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1464 | else { |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1465 | if (!is_power_of_2(new_size)) { |
| 1466 | rnew_size = roundup_pow_of_two(new_size); |
| 1467 | } |
| 1468 | xfs_iext_inline_to_direct(ifp, rnew_size); |
| 1469 | } |
| 1470 | ifp->if_real_bytes = rnew_size; |
| 1471 | ifp->if_bytes = new_size; |
| 1472 | } |
| 1473 | |
| 1474 | /* |
| 1475 | * Switch from linear (direct) extent records to inline buffer. |
| 1476 | */ |
| 1477 | void |
| 1478 | xfs_iext_direct_to_inline( |
| 1479 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1480 | xfs_extnum_t nextents) /* number of extents in file */ |
| 1481 | { |
| 1482 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
| 1483 | ASSERT(nextents <= XFS_INLINE_EXTS); |
| 1484 | /* |
| 1485 | * The inline buffer was zeroed when we switched |
| 1486 | * from inline to direct extent allocation mode, |
| 1487 | * so we don't need to clear it here. |
| 1488 | */ |
| 1489 | memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, |
| 1490 | nextents * sizeof(xfs_bmbt_rec_t)); |
| 1491 | kmem_free(ifp->if_u1.if_extents); |
| 1492 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; |
| 1493 | ifp->if_real_bytes = 0; |
| 1494 | } |
| 1495 | |
| 1496 | /* |
| 1497 | * Switch from inline buffer to linear (direct) extent records. |
| 1498 | * new_size should already be rounded up to the next power of 2 |
| 1499 | * by the caller (when appropriate), so use new_size as it is. |
| 1500 | * However, since new_size may be rounded up, we can't update |
| 1501 | * if_bytes here. It is the caller's responsibility to update |
| 1502 | * if_bytes upon return. |
| 1503 | */ |
| 1504 | void |
| 1505 | xfs_iext_inline_to_direct( |
| 1506 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1507 | int new_size) /* number of extents in file */ |
| 1508 | { |
| 1509 | ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS); |
| 1510 | memset(ifp->if_u1.if_extents, 0, new_size); |
| 1511 | if (ifp->if_bytes) { |
| 1512 | memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, |
| 1513 | ifp->if_bytes); |
| 1514 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * |
| 1515 | sizeof(xfs_bmbt_rec_t)); |
| 1516 | } |
| 1517 | ifp->if_real_bytes = new_size; |
| 1518 | } |
| 1519 | |
| 1520 | /* |
| 1521 | * Resize an extent indirection array to new_size bytes. |
| 1522 | */ |
| 1523 | STATIC void |
| 1524 | xfs_iext_realloc_indirect( |
| 1525 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1526 | int new_size) /* new indirection array size */ |
| 1527 | { |
| 1528 | int nlists; /* number of irec's (ex lists) */ |
| 1529 | int size; /* current indirection array size */ |
| 1530 | |
| 1531 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1532 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1533 | size = nlists * sizeof(xfs_ext_irec_t); |
| 1534 | ASSERT(ifp->if_real_bytes); |
| 1535 | ASSERT((new_size >= 0) && (new_size != size)); |
| 1536 | if (new_size == 0) { |
| 1537 | xfs_iext_destroy(ifp); |
| 1538 | } else { |
Christoph Hellwig | 664b60f | 2016-04-06 09:47:01 +1000 | [diff] [blame] | 1539 | ifp->if_u1.if_ext_irec = |
| 1540 | kmem_realloc(ifp->if_u1.if_ext_irec, new_size, KM_NOFS); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1541 | } |
| 1542 | } |
| 1543 | |
| 1544 | /* |
| 1545 | * Switch from indirection array to linear (direct) extent allocations. |
| 1546 | */ |
| 1547 | STATIC void |
| 1548 | xfs_iext_indirect_to_direct( |
| 1549 | xfs_ifork_t *ifp) /* inode fork pointer */ |
| 1550 | { |
| 1551 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ |
| 1552 | xfs_extnum_t nextents; /* number of extents in file */ |
| 1553 | int size; /* size of file extents */ |
| 1554 | |
| 1555 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 1556 | nextents = xfs_iext_count(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1557 | ASSERT(nextents <= XFS_LINEAR_EXTS); |
| 1558 | size = nextents * sizeof(xfs_bmbt_rec_t); |
| 1559 | |
| 1560 | xfs_iext_irec_compact_pages(ifp); |
| 1561 | ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); |
| 1562 | |
| 1563 | ep = ifp->if_u1.if_ext_irec->er_extbuf; |
| 1564 | kmem_free(ifp->if_u1.if_ext_irec); |
| 1565 | ifp->if_flags &= ~XFS_IFEXTIREC; |
| 1566 | ifp->if_u1.if_extents = ep; |
| 1567 | ifp->if_bytes = size; |
| 1568 | if (nextents < XFS_LINEAR_EXTS) { |
| 1569 | xfs_iext_realloc_direct(ifp, size); |
| 1570 | } |
| 1571 | } |
| 1572 | |
| 1573 | /* |
Alex Lyakas | 32b43ab | 2016-05-18 14:01:52 +1000 | [diff] [blame] | 1574 | * Remove all records from the indirection array. |
| 1575 | */ |
| 1576 | STATIC void |
| 1577 | xfs_iext_irec_remove_all( |
| 1578 | struct xfs_ifork *ifp) |
| 1579 | { |
| 1580 | int nlists; |
| 1581 | int i; |
| 1582 | |
| 1583 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1584 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1585 | for (i = 0; i < nlists; i++) |
| 1586 | kmem_free(ifp->if_u1.if_ext_irec[i].er_extbuf); |
| 1587 | kmem_free(ifp->if_u1.if_ext_irec); |
| 1588 | ifp->if_flags &= ~XFS_IFEXTIREC; |
| 1589 | } |
| 1590 | |
| 1591 | /* |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1592 | * Free incore file extents. |
| 1593 | */ |
| 1594 | void |
| 1595 | xfs_iext_destroy( |
| 1596 | xfs_ifork_t *ifp) /* inode fork pointer */ |
| 1597 | { |
| 1598 | if (ifp->if_flags & XFS_IFEXTIREC) { |
Alex Lyakas | 32b43ab | 2016-05-18 14:01:52 +1000 | [diff] [blame] | 1599 | xfs_iext_irec_remove_all(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1600 | } else if (ifp->if_real_bytes) { |
| 1601 | kmem_free(ifp->if_u1.if_extents); |
| 1602 | } else if (ifp->if_bytes) { |
| 1603 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * |
| 1604 | sizeof(xfs_bmbt_rec_t)); |
| 1605 | } |
| 1606 | ifp->if_u1.if_extents = NULL; |
| 1607 | ifp->if_real_bytes = 0; |
| 1608 | ifp->if_bytes = 0; |
| 1609 | } |
| 1610 | |
| 1611 | /* |
| 1612 | * Return a pointer to the extent record for file system block bno. |
| 1613 | */ |
| 1614 | xfs_bmbt_rec_host_t * /* pointer to found extent record */ |
| 1615 | xfs_iext_bno_to_ext( |
| 1616 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1617 | xfs_fileoff_t bno, /* block number to search for */ |
| 1618 | xfs_extnum_t *idxp) /* index of target extent */ |
| 1619 | { |
| 1620 | xfs_bmbt_rec_host_t *base; /* pointer to first extent */ |
| 1621 | xfs_filblks_t blockcount = 0; /* number of blocks in extent */ |
| 1622 | xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */ |
| 1623 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ |
| 1624 | int high; /* upper boundary in search */ |
| 1625 | xfs_extnum_t idx = 0; /* index of target extent */ |
| 1626 | int low; /* lower boundary in search */ |
| 1627 | xfs_extnum_t nextents; /* number of file extents */ |
| 1628 | xfs_fileoff_t startoff = 0; /* start offset of extent */ |
| 1629 | |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 1630 | nextents = xfs_iext_count(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1631 | if (nextents == 0) { |
| 1632 | *idxp = 0; |
| 1633 | return NULL; |
| 1634 | } |
| 1635 | low = 0; |
| 1636 | if (ifp->if_flags & XFS_IFEXTIREC) { |
| 1637 | /* Find target extent list */ |
| 1638 | int erp_idx = 0; |
| 1639 | erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); |
| 1640 | base = erp->er_extbuf; |
| 1641 | high = erp->er_extcount - 1; |
| 1642 | } else { |
| 1643 | base = ifp->if_u1.if_extents; |
| 1644 | high = nextents - 1; |
| 1645 | } |
| 1646 | /* Binary search extent records */ |
| 1647 | while (low <= high) { |
| 1648 | idx = (low + high) >> 1; |
| 1649 | ep = base + idx; |
| 1650 | startoff = xfs_bmbt_get_startoff(ep); |
| 1651 | blockcount = xfs_bmbt_get_blockcount(ep); |
| 1652 | if (bno < startoff) { |
| 1653 | high = idx - 1; |
| 1654 | } else if (bno >= startoff + blockcount) { |
| 1655 | low = idx + 1; |
| 1656 | } else { |
| 1657 | /* Convert back to file-based extent index */ |
| 1658 | if (ifp->if_flags & XFS_IFEXTIREC) { |
| 1659 | idx += erp->er_extoff; |
| 1660 | } |
| 1661 | *idxp = idx; |
| 1662 | return ep; |
| 1663 | } |
| 1664 | } |
| 1665 | /* Convert back to file-based extent index */ |
| 1666 | if (ifp->if_flags & XFS_IFEXTIREC) { |
| 1667 | idx += erp->er_extoff; |
| 1668 | } |
| 1669 | if (bno >= startoff + blockcount) { |
| 1670 | if (++idx == nextents) { |
| 1671 | ep = NULL; |
| 1672 | } else { |
| 1673 | ep = xfs_iext_get_ext(ifp, idx); |
| 1674 | } |
| 1675 | } |
| 1676 | *idxp = idx; |
| 1677 | return ep; |
| 1678 | } |
| 1679 | |
| 1680 | /* |
| 1681 | * Return a pointer to the indirection array entry containing the |
| 1682 | * extent record for filesystem block bno. Store the index of the |
| 1683 | * target irec in *erp_idxp. |
| 1684 | */ |
| 1685 | xfs_ext_irec_t * /* pointer to found extent record */ |
| 1686 | xfs_iext_bno_to_irec( |
| 1687 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1688 | xfs_fileoff_t bno, /* block number to search for */ |
| 1689 | int *erp_idxp) /* irec index of target ext list */ |
| 1690 | { |
| 1691 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ |
| 1692 | xfs_ext_irec_t *erp_next; /* next indirection array entry */ |
| 1693 | int erp_idx; /* indirection array index */ |
| 1694 | int nlists; /* number of extent irec's (lists) */ |
| 1695 | int high; /* binary search upper limit */ |
| 1696 | int low; /* binary search lower limit */ |
| 1697 | |
| 1698 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1699 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1700 | erp_idx = 0; |
| 1701 | low = 0; |
| 1702 | high = nlists - 1; |
| 1703 | while (low <= high) { |
| 1704 | erp_idx = (low + high) >> 1; |
| 1705 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; |
| 1706 | erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; |
| 1707 | if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { |
| 1708 | high = erp_idx - 1; |
| 1709 | } else if (erp_next && bno >= |
| 1710 | xfs_bmbt_get_startoff(erp_next->er_extbuf)) { |
| 1711 | low = erp_idx + 1; |
| 1712 | } else { |
| 1713 | break; |
| 1714 | } |
| 1715 | } |
| 1716 | *erp_idxp = erp_idx; |
| 1717 | return erp; |
| 1718 | } |
| 1719 | |
| 1720 | /* |
| 1721 | * Return a pointer to the indirection array entry containing the |
| 1722 | * extent record at file extent index *idxp. Store the index of the |
| 1723 | * target irec in *erp_idxp and store the page index of the target |
| 1724 | * extent record in *idxp. |
| 1725 | */ |
| 1726 | xfs_ext_irec_t * |
| 1727 | xfs_iext_idx_to_irec( |
| 1728 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1729 | xfs_extnum_t *idxp, /* extent index (file -> page) */ |
| 1730 | int *erp_idxp, /* pointer to target irec */ |
| 1731 | int realloc) /* new bytes were just added */ |
| 1732 | { |
| 1733 | xfs_ext_irec_t *prev; /* pointer to previous irec */ |
| 1734 | xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ |
| 1735 | int erp_idx; /* indirection array index */ |
| 1736 | int nlists; /* number of irec's (ex lists) */ |
| 1737 | int high; /* binary search upper limit */ |
| 1738 | int low; /* binary search lower limit */ |
| 1739 | xfs_extnum_t page_idx = *idxp; /* extent index in target list */ |
| 1740 | |
| 1741 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1742 | ASSERT(page_idx >= 0); |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 1743 | ASSERT(page_idx <= xfs_iext_count(ifp)); |
| 1744 | ASSERT(page_idx < xfs_iext_count(ifp) || realloc); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1745 | |
| 1746 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1747 | erp_idx = 0; |
| 1748 | low = 0; |
| 1749 | high = nlists - 1; |
| 1750 | |
| 1751 | /* Binary search extent irec's */ |
| 1752 | while (low <= high) { |
| 1753 | erp_idx = (low + high) >> 1; |
| 1754 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; |
| 1755 | prev = erp_idx > 0 ? erp - 1 : NULL; |
| 1756 | if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && |
| 1757 | realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { |
| 1758 | high = erp_idx - 1; |
| 1759 | } else if (page_idx > erp->er_extoff + erp->er_extcount || |
| 1760 | (page_idx == erp->er_extoff + erp->er_extcount && |
| 1761 | !realloc)) { |
| 1762 | low = erp_idx + 1; |
| 1763 | } else if (page_idx == erp->er_extoff + erp->er_extcount && |
| 1764 | erp->er_extcount == XFS_LINEAR_EXTS) { |
| 1765 | ASSERT(realloc); |
| 1766 | page_idx = 0; |
| 1767 | erp_idx++; |
| 1768 | erp = erp_idx < nlists ? erp + 1 : NULL; |
| 1769 | break; |
| 1770 | } else { |
| 1771 | page_idx -= erp->er_extoff; |
| 1772 | break; |
| 1773 | } |
| 1774 | } |
| 1775 | *idxp = page_idx; |
| 1776 | *erp_idxp = erp_idx; |
Eric Sandeen | d99831f | 2014-06-22 15:03:54 +1000 | [diff] [blame] | 1777 | return erp; |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1778 | } |
| 1779 | |
| 1780 | /* |
| 1781 | * Allocate and initialize an indirection array once the space needed |
| 1782 | * for incore extents increases above XFS_IEXT_BUFSZ. |
| 1783 | */ |
| 1784 | void |
| 1785 | xfs_iext_irec_init( |
| 1786 | xfs_ifork_t *ifp) /* inode fork pointer */ |
| 1787 | { |
| 1788 | xfs_ext_irec_t *erp; /* indirection array pointer */ |
| 1789 | xfs_extnum_t nextents; /* number of extents in file */ |
| 1790 | |
| 1791 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 1792 | nextents = xfs_iext_count(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1793 | ASSERT(nextents <= XFS_LINEAR_EXTS); |
| 1794 | |
| 1795 | erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS); |
| 1796 | |
| 1797 | if (nextents == 0) { |
| 1798 | ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); |
| 1799 | } else if (!ifp->if_real_bytes) { |
| 1800 | xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); |
| 1801 | } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { |
| 1802 | xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); |
| 1803 | } |
| 1804 | erp->er_extbuf = ifp->if_u1.if_extents; |
| 1805 | erp->er_extcount = nextents; |
| 1806 | erp->er_extoff = 0; |
| 1807 | |
| 1808 | ifp->if_flags |= XFS_IFEXTIREC; |
| 1809 | ifp->if_real_bytes = XFS_IEXT_BUFSZ; |
| 1810 | ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); |
| 1811 | ifp->if_u1.if_ext_irec = erp; |
| 1812 | |
| 1813 | return; |
| 1814 | } |
| 1815 | |
| 1816 | /* |
| 1817 | * Allocate and initialize a new entry in the indirection array. |
| 1818 | */ |
| 1819 | xfs_ext_irec_t * |
| 1820 | xfs_iext_irec_new( |
| 1821 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1822 | int erp_idx) /* index for new irec */ |
| 1823 | { |
| 1824 | xfs_ext_irec_t *erp; /* indirection array pointer */ |
| 1825 | int i; /* loop counter */ |
| 1826 | int nlists; /* number of irec's (ex lists) */ |
| 1827 | |
| 1828 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1829 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1830 | |
| 1831 | /* Resize indirection array */ |
| 1832 | xfs_iext_realloc_indirect(ifp, ++nlists * |
| 1833 | sizeof(xfs_ext_irec_t)); |
| 1834 | /* |
| 1835 | * Move records down in the array so the |
| 1836 | * new page can use erp_idx. |
| 1837 | */ |
| 1838 | erp = ifp->if_u1.if_ext_irec; |
| 1839 | for (i = nlists - 1; i > erp_idx; i--) { |
| 1840 | memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); |
| 1841 | } |
| 1842 | ASSERT(i == erp_idx); |
| 1843 | |
| 1844 | /* Initialize new extent record */ |
| 1845 | erp = ifp->if_u1.if_ext_irec; |
| 1846 | erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); |
| 1847 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; |
| 1848 | memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); |
| 1849 | erp[erp_idx].er_extcount = 0; |
| 1850 | erp[erp_idx].er_extoff = erp_idx > 0 ? |
| 1851 | erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; |
| 1852 | return (&erp[erp_idx]); |
| 1853 | } |
| 1854 | |
| 1855 | /* |
| 1856 | * Remove a record from the indirection array. |
| 1857 | */ |
| 1858 | void |
| 1859 | xfs_iext_irec_remove( |
| 1860 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1861 | int erp_idx) /* irec index to remove */ |
| 1862 | { |
| 1863 | xfs_ext_irec_t *erp; /* indirection array pointer */ |
| 1864 | int i; /* loop counter */ |
| 1865 | int nlists; /* number of irec's (ex lists) */ |
| 1866 | |
| 1867 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1868 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1869 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; |
| 1870 | if (erp->er_extbuf) { |
| 1871 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, |
| 1872 | -erp->er_extcount); |
| 1873 | kmem_free(erp->er_extbuf); |
| 1874 | } |
| 1875 | /* Compact extent records */ |
| 1876 | erp = ifp->if_u1.if_ext_irec; |
| 1877 | for (i = erp_idx; i < nlists - 1; i++) { |
| 1878 | memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); |
| 1879 | } |
| 1880 | /* |
| 1881 | * Manually free the last extent record from the indirection |
| 1882 | * array. A call to xfs_iext_realloc_indirect() with a size |
| 1883 | * of zero would result in a call to xfs_iext_destroy() which |
| 1884 | * would in turn call this function again, creating a nasty |
| 1885 | * infinite loop. |
| 1886 | */ |
| 1887 | if (--nlists) { |
| 1888 | xfs_iext_realloc_indirect(ifp, |
| 1889 | nlists * sizeof(xfs_ext_irec_t)); |
| 1890 | } else { |
| 1891 | kmem_free(ifp->if_u1.if_ext_irec); |
| 1892 | } |
| 1893 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; |
| 1894 | } |
| 1895 | |
| 1896 | /* |
| 1897 | * This is called to clean up large amounts of unused memory allocated |
| 1898 | * by the indirection array. Before compacting anything though, verify |
| 1899 | * that the indirection array is still needed and switch back to the |
| 1900 | * linear extent list (or even the inline buffer) if possible. The |
| 1901 | * compaction policy is as follows: |
| 1902 | * |
| 1903 | * Full Compaction: Extents fit into a single page (or inline buffer) |
| 1904 | * Partial Compaction: Extents occupy less than 50% of allocated space |
| 1905 | * No Compaction: Extents occupy at least 50% of allocated space |
| 1906 | */ |
| 1907 | void |
| 1908 | xfs_iext_irec_compact( |
| 1909 | xfs_ifork_t *ifp) /* inode fork pointer */ |
| 1910 | { |
| 1911 | xfs_extnum_t nextents; /* number of extents in file */ |
| 1912 | int nlists; /* number of irec's (ex lists) */ |
| 1913 | |
| 1914 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1915 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
Eric Sandeen | 5d82930 | 2016-11-08 12:59:42 +1100 | [diff] [blame] | 1916 | nextents = xfs_iext_count(ifp); |
Dave Chinner | 5c4d97d | 2013-08-12 20:49:33 +1000 | [diff] [blame] | 1917 | |
| 1918 | if (nextents == 0) { |
| 1919 | xfs_iext_destroy(ifp); |
| 1920 | } else if (nextents <= XFS_INLINE_EXTS) { |
| 1921 | xfs_iext_indirect_to_direct(ifp); |
| 1922 | xfs_iext_direct_to_inline(ifp, nextents); |
| 1923 | } else if (nextents <= XFS_LINEAR_EXTS) { |
| 1924 | xfs_iext_indirect_to_direct(ifp); |
| 1925 | } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { |
| 1926 | xfs_iext_irec_compact_pages(ifp); |
| 1927 | } |
| 1928 | } |
| 1929 | |
| 1930 | /* |
| 1931 | * Combine extents from neighboring extent pages. |
| 1932 | */ |
| 1933 | void |
| 1934 | xfs_iext_irec_compact_pages( |
| 1935 | xfs_ifork_t *ifp) /* inode fork pointer */ |
| 1936 | { |
| 1937 | xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ |
| 1938 | int erp_idx = 0; /* indirection array index */ |
| 1939 | int nlists; /* number of irec's (ex lists) */ |
| 1940 | |
| 1941 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1942 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1943 | while (erp_idx < nlists - 1) { |
| 1944 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; |
| 1945 | erp_next = erp + 1; |
| 1946 | if (erp_next->er_extcount <= |
| 1947 | (XFS_LINEAR_EXTS - erp->er_extcount)) { |
| 1948 | memcpy(&erp->er_extbuf[erp->er_extcount], |
| 1949 | erp_next->er_extbuf, erp_next->er_extcount * |
| 1950 | sizeof(xfs_bmbt_rec_t)); |
| 1951 | erp->er_extcount += erp_next->er_extcount; |
| 1952 | /* |
| 1953 | * Free page before removing extent record |
| 1954 | * so er_extoffs don't get modified in |
| 1955 | * xfs_iext_irec_remove. |
| 1956 | */ |
| 1957 | kmem_free(erp_next->er_extbuf); |
| 1958 | erp_next->er_extbuf = NULL; |
| 1959 | xfs_iext_irec_remove(ifp, erp_idx + 1); |
| 1960 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1961 | } else { |
| 1962 | erp_idx++; |
| 1963 | } |
| 1964 | } |
| 1965 | } |
| 1966 | |
| 1967 | /* |
| 1968 | * This is called to update the er_extoff field in the indirection |
| 1969 | * array when extents have been added or removed from one of the |
| 1970 | * extent lists. erp_idx contains the irec index to begin updating |
| 1971 | * at and ext_diff contains the number of extents that were added |
| 1972 | * or removed. |
| 1973 | */ |
| 1974 | void |
| 1975 | xfs_iext_irec_update_extoffs( |
| 1976 | xfs_ifork_t *ifp, /* inode fork pointer */ |
| 1977 | int erp_idx, /* irec index to update */ |
| 1978 | int ext_diff) /* number of new extents */ |
| 1979 | { |
| 1980 | int i; /* loop counter */ |
| 1981 | int nlists; /* number of irec's (ex lists */ |
| 1982 | |
| 1983 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); |
| 1984 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
| 1985 | for (i = erp_idx; i < nlists; i++) { |
| 1986 | ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; |
| 1987 | } |
| 1988 | } |
Darrick J. Wong | 3993bae | 2016-10-03 09:11:32 -0700 | [diff] [blame] | 1989 | |
| 1990 | /* |
| 1991 | * Initialize an inode's copy-on-write fork. |
| 1992 | */ |
| 1993 | void |
| 1994 | xfs_ifork_init_cow( |
| 1995 | struct xfs_inode *ip) |
| 1996 | { |
| 1997 | if (ip->i_cowfp) |
| 1998 | return; |
| 1999 | |
| 2000 | ip->i_cowfp = kmem_zone_zalloc(xfs_ifork_zone, |
| 2001 | KM_SLEEP | KM_NOFS); |
| 2002 | ip->i_cowfp->if_flags = XFS_IFEXTENTS; |
| 2003 | ip->i_cformat = XFS_DINODE_FMT_EXTENTS; |
| 2004 | ip->i_cnextents = 0; |
| 2005 | } |
Christoph Hellwig | 93533c7 | 2016-11-24 11:39:32 +1100 | [diff] [blame] | 2006 | |
| 2007 | /* |
| 2008 | * Lookup the extent covering bno. |
| 2009 | * |
| 2010 | * If there is an extent covering bno return the extent index, and store the |
| 2011 | * expanded extent structure in *gotp, and the extent index in *idx. |
| 2012 | * If there is no extent covering bno, but there is an extent after it (e.g. |
| 2013 | * it lies in a hole) return that extent in *gotp and its index in *idx |
| 2014 | * instead. |
| 2015 | * If bno is beyond the last extent return false, and return the index after |
| 2016 | * the last valid index in *idxp. |
| 2017 | */ |
| 2018 | bool |
| 2019 | xfs_iext_lookup_extent( |
| 2020 | struct xfs_inode *ip, |
| 2021 | struct xfs_ifork *ifp, |
| 2022 | xfs_fileoff_t bno, |
| 2023 | xfs_extnum_t *idxp, |
| 2024 | struct xfs_bmbt_irec *gotp) |
| 2025 | { |
| 2026 | struct xfs_bmbt_rec_host *ep; |
| 2027 | |
| 2028 | XFS_STATS_INC(ip->i_mount, xs_look_exlist); |
| 2029 | |
| 2030 | ep = xfs_iext_bno_to_ext(ifp, bno, idxp); |
| 2031 | if (!ep) |
| 2032 | return false; |
| 2033 | xfs_bmbt_get_all(ep, gotp); |
| 2034 | return true; |
| 2035 | } |
| 2036 | |
| 2037 | /* |
| 2038 | * Return true if there is an extent at index idx, and return the expanded |
| 2039 | * extent structure at idx in that case. Else return false. |
| 2040 | */ |
| 2041 | bool |
| 2042 | xfs_iext_get_extent( |
| 2043 | struct xfs_ifork *ifp, |
| 2044 | xfs_extnum_t idx, |
| 2045 | struct xfs_bmbt_irec *gotp) |
| 2046 | { |
| 2047 | if (idx < 0 || idx >= xfs_iext_count(ifp)) |
| 2048 | return false; |
| 2049 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), gotp); |
| 2050 | return true; |
| 2051 | } |