]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
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 | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
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. | |
1da177e4 | 13 | * |
7b718769 NS |
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 | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
a844f451 | 26 | #include "xfs_ag.h" |
1da177e4 LT |
27 | #include "xfs_mount.h" |
28 | #include "xfs_trans_priv.h" | |
1da177e4 | 29 | #include "xfs_bmap_btree.h" |
1da177e4 | 30 | #include "xfs_dinode.h" |
1da177e4 | 31 | #include "xfs_inode.h" |
a844f451 | 32 | #include "xfs_inode_item.h" |
db7a19f2 | 33 | #include "xfs_error.h" |
0b1b213f | 34 | #include "xfs_trace.h" |
1da177e4 LT |
35 | |
36 | ||
37 | kmem_zone_t *xfs_ili_zone; /* inode log item zone */ | |
38 | ||
7bfa31d8 CH |
39 | static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip) |
40 | { | |
41 | return container_of(lip, struct xfs_inode_log_item, ili_item); | |
42 | } | |
43 | ||
44 | ||
1da177e4 LT |
45 | /* |
46 | * This returns the number of iovecs needed to log the given inode item. | |
47 | * | |
48 | * We need one iovec for the inode log format structure, one for the | |
49 | * inode core, and possibly one for the inode data/extents/b-tree root | |
50 | * and one for the inode attribute data/extents/b-tree root. | |
51 | */ | |
52 | STATIC uint | |
53 | xfs_inode_item_size( | |
7bfa31d8 | 54 | struct xfs_log_item *lip) |
1da177e4 | 55 | { |
7bfa31d8 CH |
56 | struct xfs_inode_log_item *iip = INODE_ITEM(lip); |
57 | struct xfs_inode *ip = iip->ili_inode; | |
58 | uint nvecs = 2; | |
1da177e4 | 59 | |
1da177e4 LT |
60 | switch (ip->i_d.di_format) { |
61 | case XFS_DINODE_FMT_EXTENTS: | |
f5d8d5c4 | 62 | if ((iip->ili_fields & XFS_ILOG_DEXT) && |
339a5f5d CH |
63 | ip->i_d.di_nextents > 0 && |
64 | ip->i_df.if_bytes > 0) | |
1da177e4 | 65 | nvecs++; |
1da177e4 LT |
66 | break; |
67 | ||
68 | case XFS_DINODE_FMT_BTREE: | |
f5d8d5c4 | 69 | if ((iip->ili_fields & XFS_ILOG_DBROOT) && |
339a5f5d | 70 | ip->i_df.if_broot_bytes > 0) |
1da177e4 | 71 | nvecs++; |
1da177e4 LT |
72 | break; |
73 | ||
74 | case XFS_DINODE_FMT_LOCAL: | |
f5d8d5c4 | 75 | if ((iip->ili_fields & XFS_ILOG_DDATA) && |
339a5f5d | 76 | ip->i_df.if_bytes > 0) |
1da177e4 | 77 | nvecs++; |
1da177e4 LT |
78 | break; |
79 | ||
80 | case XFS_DINODE_FMT_DEV: | |
1da177e4 | 81 | case XFS_DINODE_FMT_UUID: |
1da177e4 LT |
82 | break; |
83 | ||
84 | default: | |
85 | ASSERT(0); | |
86 | break; | |
87 | } | |
88 | ||
339a5f5d | 89 | if (!XFS_IFORK_Q(ip)) |
1da177e4 | 90 | return nvecs; |
339a5f5d | 91 | |
1da177e4 LT |
92 | |
93 | /* | |
94 | * Log any necessary attribute data. | |
95 | */ | |
96 | switch (ip->i_d.di_aformat) { | |
97 | case XFS_DINODE_FMT_EXTENTS: | |
f5d8d5c4 | 98 | if ((iip->ili_fields & XFS_ILOG_AEXT) && |
339a5f5d CH |
99 | ip->i_d.di_anextents > 0 && |
100 | ip->i_afp->if_bytes > 0) | |
1da177e4 | 101 | nvecs++; |
1da177e4 LT |
102 | break; |
103 | ||
104 | case XFS_DINODE_FMT_BTREE: | |
f5d8d5c4 | 105 | if ((iip->ili_fields & XFS_ILOG_ABROOT) && |
339a5f5d | 106 | ip->i_afp->if_broot_bytes > 0) |
1da177e4 | 107 | nvecs++; |
1da177e4 LT |
108 | break; |
109 | ||
110 | case XFS_DINODE_FMT_LOCAL: | |
f5d8d5c4 | 111 | if ((iip->ili_fields & XFS_ILOG_ADATA) && |
339a5f5d | 112 | ip->i_afp->if_bytes > 0) |
1da177e4 | 113 | nvecs++; |
1da177e4 LT |
114 | break; |
115 | ||
116 | default: | |
117 | ASSERT(0); | |
118 | break; | |
119 | } | |
120 | ||
121 | return nvecs; | |
122 | } | |
123 | ||
e828776a DC |
124 | /* |
125 | * xfs_inode_item_format_extents - convert in-core extents to on-disk form | |
126 | * | |
127 | * For either the data or attr fork in extent format, we need to endian convert | |
128 | * the in-core extent as we place them into the on-disk inode. In this case, we | |
129 | * need to do this conversion before we write the extents into the log. Because | |
130 | * we don't have the disk inode to write into here, we allocate a buffer and | |
131 | * format the extents into it via xfs_iextents_copy(). We free the buffer in | |
132 | * the unlock routine after the copy for the log has been made. | |
133 | * | |
134 | * In the case of the data fork, the in-core and on-disk fork sizes can be | |
135 | * different due to delayed allocation extents. We only log on-disk extents | |
136 | * here, so always use the physical fork size to determine the size of the | |
137 | * buffer we need to allocate. | |
138 | */ | |
139 | STATIC void | |
140 | xfs_inode_item_format_extents( | |
141 | struct xfs_inode *ip, | |
142 | struct xfs_log_iovec *vecp, | |
143 | int whichfork, | |
144 | int type) | |
145 | { | |
146 | xfs_bmbt_rec_t *ext_buffer; | |
147 | ||
148 | ext_buffer = kmem_alloc(XFS_IFORK_SIZE(ip, whichfork), KM_SLEEP); | |
149 | if (whichfork == XFS_DATA_FORK) | |
150 | ip->i_itemp->ili_extents_buf = ext_buffer; | |
151 | else | |
152 | ip->i_itemp->ili_aextents_buf = ext_buffer; | |
153 | ||
154 | vecp->i_addr = ext_buffer; | |
155 | vecp->i_len = xfs_iextents_copy(ip, ext_buffer, whichfork); | |
156 | vecp->i_type = type; | |
157 | } | |
158 | ||
1da177e4 LT |
159 | /* |
160 | * This is called to fill in the vector of log iovecs for the | |
161 | * given inode log item. It fills the first item with an inode | |
162 | * log format structure, the second with the on-disk inode structure, | |
163 | * and a possible third and/or fourth with the inode data/extents/b-tree | |
164 | * root and inode attributes data/extents/b-tree root. | |
165 | */ | |
166 | STATIC void | |
167 | xfs_inode_item_format( | |
7bfa31d8 CH |
168 | struct xfs_log_item *lip, |
169 | struct xfs_log_iovec *vecp) | |
1da177e4 | 170 | { |
7bfa31d8 CH |
171 | struct xfs_inode_log_item *iip = INODE_ITEM(lip); |
172 | struct xfs_inode *ip = iip->ili_inode; | |
1da177e4 | 173 | uint nvecs; |
1da177e4 | 174 | size_t data_bytes; |
1da177e4 LT |
175 | xfs_mount_t *mp; |
176 | ||
4e0d5f92 | 177 | vecp->i_addr = &iip->ili_format; |
1da177e4 | 178 | vecp->i_len = sizeof(xfs_inode_log_format_t); |
4139b3b3 | 179 | vecp->i_type = XLOG_REG_TYPE_IFORMAT; |
1da177e4 LT |
180 | vecp++; |
181 | nvecs = 1; | |
182 | ||
4e0d5f92 | 183 | vecp->i_addr = &ip->i_d; |
81591fe2 | 184 | vecp->i_len = sizeof(struct xfs_icdinode); |
4139b3b3 | 185 | vecp->i_type = XLOG_REG_TYPE_ICORE; |
1da177e4 LT |
186 | vecp++; |
187 | nvecs++; | |
1da177e4 LT |
188 | |
189 | /* | |
190 | * If this is really an old format inode, then we need to | |
191 | * log it as such. This means that we have to copy the link | |
192 | * count from the new field to the old. We don't have to worry | |
193 | * about the new fields, because nothing trusts them as long as | |
194 | * the old inode version number is there. If the superblock already | |
195 | * has a new version number, then we don't bother converting back. | |
196 | */ | |
197 | mp = ip->i_mount; | |
51ce16d5 CH |
198 | ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb)); |
199 | if (ip->i_d.di_version == 1) { | |
62118709 | 200 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
1da177e4 LT |
201 | /* |
202 | * Convert it back. | |
203 | */ | |
204 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
205 | ip->i_d.di_onlink = ip->i_d.di_nlink; | |
206 | } else { | |
207 | /* | |
208 | * The superblock version has already been bumped, | |
209 | * so just make the conversion to the new inode | |
210 | * format permanent. | |
211 | */ | |
51ce16d5 | 212 | ip->i_d.di_version = 2; |
1da177e4 LT |
213 | ip->i_d.di_onlink = 0; |
214 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
215 | } | |
216 | } | |
217 | ||
218 | switch (ip->i_d.di_format) { | |
219 | case XFS_DINODE_FMT_EXTENTS: | |
f5d8d5c4 | 220 | iip->ili_fields &= |
339a5f5d CH |
221 | ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | |
222 | XFS_ILOG_DEV | XFS_ILOG_UUID); | |
223 | ||
f5d8d5c4 | 224 | if ((iip->ili_fields & XFS_ILOG_DEXT) && |
339a5f5d CH |
225 | ip->i_d.di_nextents > 0 && |
226 | ip->i_df.if_bytes > 0) { | |
1da177e4 | 227 | ASSERT(ip->i_df.if_u1.if_extents != NULL); |
339a5f5d | 228 | ASSERT(ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) > 0); |
1da177e4 | 229 | ASSERT(iip->ili_extents_buf == NULL); |
339a5f5d | 230 | |
f016bad6 | 231 | #ifdef XFS_NATIVE_HOST |
696123fc DC |
232 | if (ip->i_d.di_nextents == ip->i_df.if_bytes / |
233 | (uint)sizeof(xfs_bmbt_rec_t)) { | |
1da177e4 LT |
234 | /* |
235 | * There are no delayed allocation | |
236 | * extents, so just point to the | |
237 | * real extents array. | |
238 | */ | |
4e0d5f92 | 239 | vecp->i_addr = ip->i_df.if_u1.if_extents; |
1da177e4 | 240 | vecp->i_len = ip->i_df.if_bytes; |
4139b3b3 | 241 | vecp->i_type = XLOG_REG_TYPE_IEXT; |
1da177e4 LT |
242 | } else |
243 | #endif | |
244 | { | |
e828776a DC |
245 | xfs_inode_item_format_extents(ip, vecp, |
246 | XFS_DATA_FORK, XLOG_REG_TYPE_IEXT); | |
1da177e4 LT |
247 | } |
248 | ASSERT(vecp->i_len <= ip->i_df.if_bytes); | |
249 | iip->ili_format.ilf_dsize = vecp->i_len; | |
250 | vecp++; | |
251 | nvecs++; | |
339a5f5d | 252 | } else { |
f5d8d5c4 | 253 | iip->ili_fields &= ~XFS_ILOG_DEXT; |
1da177e4 LT |
254 | } |
255 | break; | |
256 | ||
257 | case XFS_DINODE_FMT_BTREE: | |
f5d8d5c4 | 258 | iip->ili_fields &= |
339a5f5d CH |
259 | ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | |
260 | XFS_ILOG_DEV | XFS_ILOG_UUID); | |
261 | ||
f5d8d5c4 | 262 | if ((iip->ili_fields & XFS_ILOG_DBROOT) && |
339a5f5d | 263 | ip->i_df.if_broot_bytes > 0) { |
1da177e4 | 264 | ASSERT(ip->i_df.if_broot != NULL); |
4e0d5f92 | 265 | vecp->i_addr = ip->i_df.if_broot; |
1da177e4 | 266 | vecp->i_len = ip->i_df.if_broot_bytes; |
4139b3b3 | 267 | vecp->i_type = XLOG_REG_TYPE_IBROOT; |
1da177e4 LT |
268 | vecp++; |
269 | nvecs++; | |
270 | iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes; | |
339a5f5d | 271 | } else { |
f5d8d5c4 | 272 | ASSERT(!(iip->ili_fields & |
339a5f5d CH |
273 | XFS_ILOG_DBROOT)); |
274 | #ifdef XFS_TRANS_DEBUG | |
275 | if (iip->ili_root_size > 0) { | |
276 | ASSERT(iip->ili_root_size == | |
277 | ip->i_df.if_broot_bytes); | |
278 | ASSERT(memcmp(iip->ili_orig_root, | |
279 | ip->i_df.if_broot, | |
280 | iip->ili_root_size) == 0); | |
281 | } else { | |
282 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
283 | } | |
284 | #endif | |
f5d8d5c4 | 285 | iip->ili_fields &= ~XFS_ILOG_DBROOT; |
1da177e4 LT |
286 | } |
287 | break; | |
288 | ||
289 | case XFS_DINODE_FMT_LOCAL: | |
f5d8d5c4 | 290 | iip->ili_fields &= |
339a5f5d CH |
291 | ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | |
292 | XFS_ILOG_DEV | XFS_ILOG_UUID); | |
f5d8d5c4 | 293 | if ((iip->ili_fields & XFS_ILOG_DDATA) && |
339a5f5d | 294 | ip->i_df.if_bytes > 0) { |
1da177e4 LT |
295 | ASSERT(ip->i_df.if_u1.if_data != NULL); |
296 | ASSERT(ip->i_d.di_size > 0); | |
297 | ||
4e0d5f92 | 298 | vecp->i_addr = ip->i_df.if_u1.if_data; |
1da177e4 LT |
299 | /* |
300 | * Round i_bytes up to a word boundary. | |
301 | * The underlying memory is guaranteed to | |
302 | * to be there by xfs_idata_realloc(). | |
303 | */ | |
304 | data_bytes = roundup(ip->i_df.if_bytes, 4); | |
305 | ASSERT((ip->i_df.if_real_bytes == 0) || | |
306 | (ip->i_df.if_real_bytes == data_bytes)); | |
307 | vecp->i_len = (int)data_bytes; | |
4139b3b3 | 308 | vecp->i_type = XLOG_REG_TYPE_ILOCAL; |
1da177e4 LT |
309 | vecp++; |
310 | nvecs++; | |
311 | iip->ili_format.ilf_dsize = (unsigned)data_bytes; | |
339a5f5d | 312 | } else { |
f5d8d5c4 | 313 | iip->ili_fields &= ~XFS_ILOG_DDATA; |
1da177e4 LT |
314 | } |
315 | break; | |
316 | ||
317 | case XFS_DINODE_FMT_DEV: | |
f5d8d5c4 | 318 | iip->ili_fields &= |
339a5f5d CH |
319 | ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | |
320 | XFS_ILOG_DEXT | XFS_ILOG_UUID); | |
f5d8d5c4 | 321 | if (iip->ili_fields & XFS_ILOG_DEV) { |
1da177e4 LT |
322 | iip->ili_format.ilf_u.ilfu_rdev = |
323 | ip->i_df.if_u2.if_rdev; | |
324 | } | |
325 | break; | |
326 | ||
327 | case XFS_DINODE_FMT_UUID: | |
f5d8d5c4 | 328 | iip->ili_fields &= |
339a5f5d CH |
329 | ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | |
330 | XFS_ILOG_DEXT | XFS_ILOG_DEV); | |
f5d8d5c4 | 331 | if (iip->ili_fields & XFS_ILOG_UUID) { |
1da177e4 LT |
332 | iip->ili_format.ilf_u.ilfu_uuid = |
333 | ip->i_df.if_u2.if_uuid; | |
334 | } | |
335 | break; | |
336 | ||
337 | default: | |
338 | ASSERT(0); | |
339 | break; | |
340 | } | |
341 | ||
342 | /* | |
339a5f5d | 343 | * If there are no attributes associated with the file, then we're done. |
1da177e4 LT |
344 | */ |
345 | if (!XFS_IFORK_Q(ip)) { | |
f5d8d5c4 | 346 | iip->ili_fields &= |
339a5f5d | 347 | ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT); |
f5d8d5c4 | 348 | goto out; |
1da177e4 LT |
349 | } |
350 | ||
351 | switch (ip->i_d.di_aformat) { | |
352 | case XFS_DINODE_FMT_EXTENTS: | |
f5d8d5c4 | 353 | iip->ili_fields &= |
339a5f5d CH |
354 | ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT); |
355 | ||
f5d8d5c4 | 356 | if ((iip->ili_fields & XFS_ILOG_AEXT) && |
339a5f5d CH |
357 | ip->i_d.di_anextents > 0 && |
358 | ip->i_afp->if_bytes > 0) { | |
359 | ASSERT(ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) == | |
360 | ip->i_d.di_anextents); | |
73523a2e | 361 | ASSERT(ip->i_afp->if_u1.if_extents != NULL); |
f016bad6 | 362 | #ifdef XFS_NATIVE_HOST |
1da177e4 LT |
363 | /* |
364 | * There are not delayed allocation extents | |
365 | * for attributes, so just point at the array. | |
366 | */ | |
4e0d5f92 | 367 | vecp->i_addr = ip->i_afp->if_u1.if_extents; |
1da177e4 | 368 | vecp->i_len = ip->i_afp->if_bytes; |
e828776a | 369 | vecp->i_type = XLOG_REG_TYPE_IATTR_EXT; |
1da177e4 LT |
370 | #else |
371 | ASSERT(iip->ili_aextents_buf == NULL); | |
e828776a DC |
372 | xfs_inode_item_format_extents(ip, vecp, |
373 | XFS_ATTR_FORK, XLOG_REG_TYPE_IATTR_EXT); | |
1da177e4 LT |
374 | #endif |
375 | iip->ili_format.ilf_asize = vecp->i_len; | |
376 | vecp++; | |
377 | nvecs++; | |
339a5f5d | 378 | } else { |
f5d8d5c4 | 379 | iip->ili_fields &= ~XFS_ILOG_AEXT; |
1da177e4 LT |
380 | } |
381 | break; | |
382 | ||
383 | case XFS_DINODE_FMT_BTREE: | |
f5d8d5c4 | 384 | iip->ili_fields &= |
339a5f5d CH |
385 | ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT); |
386 | ||
f5d8d5c4 | 387 | if ((iip->ili_fields & XFS_ILOG_ABROOT) && |
339a5f5d | 388 | ip->i_afp->if_broot_bytes > 0) { |
1da177e4 | 389 | ASSERT(ip->i_afp->if_broot != NULL); |
339a5f5d | 390 | |
4e0d5f92 | 391 | vecp->i_addr = ip->i_afp->if_broot; |
1da177e4 | 392 | vecp->i_len = ip->i_afp->if_broot_bytes; |
4139b3b3 | 393 | vecp->i_type = XLOG_REG_TYPE_IATTR_BROOT; |
1da177e4 LT |
394 | vecp++; |
395 | nvecs++; | |
396 | iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes; | |
339a5f5d | 397 | } else { |
f5d8d5c4 | 398 | iip->ili_fields &= ~XFS_ILOG_ABROOT; |
1da177e4 LT |
399 | } |
400 | break; | |
401 | ||
402 | case XFS_DINODE_FMT_LOCAL: | |
f5d8d5c4 | 403 | iip->ili_fields &= |
339a5f5d CH |
404 | ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT); |
405 | ||
f5d8d5c4 | 406 | if ((iip->ili_fields & XFS_ILOG_ADATA) && |
339a5f5d | 407 | ip->i_afp->if_bytes > 0) { |
1da177e4 LT |
408 | ASSERT(ip->i_afp->if_u1.if_data != NULL); |
409 | ||
4e0d5f92 | 410 | vecp->i_addr = ip->i_afp->if_u1.if_data; |
1da177e4 LT |
411 | /* |
412 | * Round i_bytes up to a word boundary. | |
413 | * The underlying memory is guaranteed to | |
414 | * to be there by xfs_idata_realloc(). | |
415 | */ | |
416 | data_bytes = roundup(ip->i_afp->if_bytes, 4); | |
417 | ASSERT((ip->i_afp->if_real_bytes == 0) || | |
418 | (ip->i_afp->if_real_bytes == data_bytes)); | |
419 | vecp->i_len = (int)data_bytes; | |
4139b3b3 | 420 | vecp->i_type = XLOG_REG_TYPE_IATTR_LOCAL; |
1da177e4 LT |
421 | vecp++; |
422 | nvecs++; | |
423 | iip->ili_format.ilf_asize = (unsigned)data_bytes; | |
339a5f5d | 424 | } else { |
f5d8d5c4 | 425 | iip->ili_fields &= ~XFS_ILOG_ADATA; |
1da177e4 LT |
426 | } |
427 | break; | |
428 | ||
429 | default: | |
430 | ASSERT(0); | |
431 | break; | |
432 | } | |
433 | ||
f5d8d5c4 CH |
434 | out: |
435 | /* | |
436 | * Now update the log format that goes out to disk from the in-core | |
437 | * values. We always write the inode core to make the arithmetic | |
438 | * games in recovery easier, which isn't a big deal as just about any | |
439 | * transaction would dirty it anyway. | |
440 | */ | |
8f639dde CH |
441 | iip->ili_format.ilf_fields = XFS_ILOG_CORE | |
442 | (iip->ili_fields & ~XFS_ILOG_TIMESTAMP); | |
1da177e4 LT |
443 | iip->ili_format.ilf_size = nvecs; |
444 | } | |
445 | ||
446 | ||
447 | /* | |
448 | * This is called to pin the inode associated with the inode log | |
a14a5ab5 | 449 | * item in memory so it cannot be written out. |
1da177e4 LT |
450 | */ |
451 | STATIC void | |
452 | xfs_inode_item_pin( | |
7bfa31d8 | 453 | struct xfs_log_item *lip) |
1da177e4 | 454 | { |
7bfa31d8 | 455 | struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode; |
a14a5ab5 | 456 | |
7bfa31d8 CH |
457 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
458 | ||
459 | trace_xfs_inode_pin(ip, _RET_IP_); | |
460 | atomic_inc(&ip->i_pincount); | |
1da177e4 LT |
461 | } |
462 | ||
463 | ||
464 | /* | |
465 | * This is called to unpin the inode associated with the inode log | |
466 | * item which was previously pinned with a call to xfs_inode_item_pin(). | |
a14a5ab5 CH |
467 | * |
468 | * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0. | |
1da177e4 | 469 | */ |
1da177e4 LT |
470 | STATIC void |
471 | xfs_inode_item_unpin( | |
7bfa31d8 | 472 | struct xfs_log_item *lip, |
9412e318 | 473 | int remove) |
1da177e4 | 474 | { |
7bfa31d8 | 475 | struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode; |
a14a5ab5 | 476 | |
4aaf15d1 | 477 | trace_xfs_inode_unpin(ip, _RET_IP_); |
a14a5ab5 CH |
478 | ASSERT(atomic_read(&ip->i_pincount) > 0); |
479 | if (atomic_dec_and_test(&ip->i_pincount)) | |
f392e631 | 480 | wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT); |
1da177e4 LT |
481 | } |
482 | ||
1da177e4 | 483 | STATIC uint |
43ff2122 CH |
484 | xfs_inode_item_push( |
485 | struct xfs_log_item *lip, | |
486 | struct list_head *buffer_list) | |
1da177e4 | 487 | { |
7bfa31d8 CH |
488 | struct xfs_inode_log_item *iip = INODE_ITEM(lip); |
489 | struct xfs_inode *ip = iip->ili_inode; | |
43ff2122 CH |
490 | struct xfs_buf *bp = NULL; |
491 | uint rval = XFS_ITEM_SUCCESS; | |
492 | int error; | |
1da177e4 | 493 | |
7bfa31d8 | 494 | if (xfs_ipincount(ip) > 0) |
1da177e4 | 495 | return XFS_ITEM_PINNED; |
1da177e4 | 496 | |
7bfa31d8 | 497 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) |
1da177e4 | 498 | return XFS_ITEM_LOCKED; |
1da177e4 | 499 | |
4c46819a CH |
500 | /* |
501 | * Re-check the pincount now that we stabilized the value by | |
502 | * taking the ilock. | |
503 | */ | |
504 | if (xfs_ipincount(ip) > 0) { | |
43ff2122 CH |
505 | rval = XFS_ITEM_PINNED; |
506 | goto out_unlock; | |
4c46819a CH |
507 | } |
508 | ||
43ff2122 CH |
509 | /* |
510 | * Someone else is already flushing the inode. Nothing we can do | |
511 | * here but wait for the flush to finish and remove the item from | |
512 | * the AIL. | |
513 | */ | |
1da177e4 | 514 | if (!xfs_iflock_nowait(ip)) { |
43ff2122 CH |
515 | rval = XFS_ITEM_FLUSHING; |
516 | goto out_unlock; | |
1da177e4 LT |
517 | } |
518 | ||
43ff2122 CH |
519 | /* |
520 | * Stale inode items should force out the iclog. | |
521 | */ | |
1da177e4 LT |
522 | if (ip->i_flags & XFS_ISTALE) { |
523 | xfs_ifunlock(ip); | |
5b03ff1b | 524 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
525 | return XFS_ITEM_PINNED; |
526 | } | |
527 | ||
43ff2122 CH |
528 | ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount)); |
529 | ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
530 | ||
531 | spin_unlock(&lip->li_ailp->xa_lock); | |
532 | ||
533 | error = xfs_iflush(ip, &bp); | |
534 | if (!error) { | |
535 | if (!xfs_buf_delwri_queue(bp, buffer_list)) | |
536 | rval = XFS_ITEM_FLUSHING; | |
537 | xfs_buf_relse(bp); | |
1da177e4 | 538 | } |
43ff2122 CH |
539 | |
540 | spin_lock(&lip->li_ailp->xa_lock); | |
541 | out_unlock: | |
542 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
543 | return rval; | |
1da177e4 LT |
544 | } |
545 | ||
546 | /* | |
547 | * Unlock the inode associated with the inode log item. | |
548 | * Clear the fields of the inode and inode log item that | |
549 | * are specific to the current transaction. If the | |
550 | * hold flags is set, do not unlock the inode. | |
551 | */ | |
552 | STATIC void | |
553 | xfs_inode_item_unlock( | |
7bfa31d8 | 554 | struct xfs_log_item *lip) |
1da177e4 | 555 | { |
7bfa31d8 CH |
556 | struct xfs_inode_log_item *iip = INODE_ITEM(lip); |
557 | struct xfs_inode *ip = iip->ili_inode; | |
898621d5 | 558 | unsigned short lock_flags; |
1da177e4 | 559 | |
f3ca8738 CH |
560 | ASSERT(ip->i_itemp != NULL); |
561 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
1da177e4 LT |
562 | |
563 | /* | |
564 | * If the inode needed a separate buffer with which to log | |
565 | * its extents, then free it now. | |
566 | */ | |
567 | if (iip->ili_extents_buf != NULL) { | |
568 | ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS); | |
569 | ASSERT(ip->i_d.di_nextents > 0); | |
f5d8d5c4 | 570 | ASSERT(iip->ili_fields & XFS_ILOG_DEXT); |
1da177e4 | 571 | ASSERT(ip->i_df.if_bytes > 0); |
f0e2d93c | 572 | kmem_free(iip->ili_extents_buf); |
1da177e4 LT |
573 | iip->ili_extents_buf = NULL; |
574 | } | |
575 | if (iip->ili_aextents_buf != NULL) { | |
576 | ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS); | |
577 | ASSERT(ip->i_d.di_anextents > 0); | |
f5d8d5c4 | 578 | ASSERT(iip->ili_fields & XFS_ILOG_AEXT); |
1da177e4 | 579 | ASSERT(ip->i_afp->if_bytes > 0); |
f0e2d93c | 580 | kmem_free(iip->ili_aextents_buf); |
1da177e4 LT |
581 | iip->ili_aextents_buf = NULL; |
582 | } | |
583 | ||
898621d5 CH |
584 | lock_flags = iip->ili_lock_flags; |
585 | iip->ili_lock_flags = 0; | |
ddc3415a | 586 | if (lock_flags) |
f3ca8738 | 587 | xfs_iunlock(ip, lock_flags); |
1da177e4 LT |
588 | } |
589 | ||
590 | /* | |
de25c181 DC |
591 | * This is called to find out where the oldest active copy of the inode log |
592 | * item in the on disk log resides now that the last log write of it completed | |
593 | * at the given lsn. Since we always re-log all dirty data in an inode, the | |
594 | * latest copy in the on disk log is the only one that matters. Therefore, | |
595 | * simply return the given lsn. | |
596 | * | |
597 | * If the inode has been marked stale because the cluster is being freed, we | |
598 | * don't want to (re-)insert this inode into the AIL. There is a race condition | |
599 | * where the cluster buffer may be unpinned before the inode is inserted into | |
600 | * the AIL during transaction committed processing. If the buffer is unpinned | |
601 | * before the inode item has been committed and inserted, then it is possible | |
1316d4da | 602 | * for the buffer to be written and IO completes before the inode is inserted |
de25c181 DC |
603 | * into the AIL. In that case, we'd be inserting a clean, stale inode into the |
604 | * AIL which will never get removed. It will, however, get reclaimed which | |
605 | * triggers an assert in xfs_inode_free() complaining about freein an inode | |
606 | * still in the AIL. | |
607 | * | |
1316d4da DC |
608 | * To avoid this, just unpin the inode directly and return a LSN of -1 so the |
609 | * transaction committed code knows that it does not need to do any further | |
610 | * processing on the item. | |
1da177e4 | 611 | */ |
1da177e4 LT |
612 | STATIC xfs_lsn_t |
613 | xfs_inode_item_committed( | |
7bfa31d8 | 614 | struct xfs_log_item *lip, |
1da177e4 LT |
615 | xfs_lsn_t lsn) |
616 | { | |
de25c181 DC |
617 | struct xfs_inode_log_item *iip = INODE_ITEM(lip); |
618 | struct xfs_inode *ip = iip->ili_inode; | |
619 | ||
1316d4da DC |
620 | if (xfs_iflags_test(ip, XFS_ISTALE)) { |
621 | xfs_inode_item_unpin(lip, 0); | |
622 | return -1; | |
623 | } | |
7bfa31d8 | 624 | return lsn; |
1da177e4 LT |
625 | } |
626 | ||
1da177e4 LT |
627 | /* |
628 | * XXX rcc - this one really has to do something. Probably needs | |
629 | * to stamp in a new field in the incore inode. | |
630 | */ | |
1da177e4 LT |
631 | STATIC void |
632 | xfs_inode_item_committing( | |
7bfa31d8 | 633 | struct xfs_log_item *lip, |
1da177e4 LT |
634 | xfs_lsn_t lsn) |
635 | { | |
7bfa31d8 | 636 | INODE_ITEM(lip)->ili_last_lsn = lsn; |
1da177e4 LT |
637 | } |
638 | ||
639 | /* | |
640 | * This is the ops vector shared by all buf log items. | |
641 | */ | |
272e42b2 | 642 | static const struct xfs_item_ops xfs_inode_item_ops = { |
7bfa31d8 CH |
643 | .iop_size = xfs_inode_item_size, |
644 | .iop_format = xfs_inode_item_format, | |
645 | .iop_pin = xfs_inode_item_pin, | |
646 | .iop_unpin = xfs_inode_item_unpin, | |
7bfa31d8 CH |
647 | .iop_unlock = xfs_inode_item_unlock, |
648 | .iop_committed = xfs_inode_item_committed, | |
649 | .iop_push = xfs_inode_item_push, | |
7bfa31d8 | 650 | .iop_committing = xfs_inode_item_committing |
1da177e4 LT |
651 | }; |
652 | ||
653 | ||
654 | /* | |
655 | * Initialize the inode log item for a newly allocated (in-core) inode. | |
656 | */ | |
657 | void | |
658 | xfs_inode_item_init( | |
7bfa31d8 CH |
659 | struct xfs_inode *ip, |
660 | struct xfs_mount *mp) | |
1da177e4 | 661 | { |
7bfa31d8 | 662 | struct xfs_inode_log_item *iip; |
1da177e4 LT |
663 | |
664 | ASSERT(ip->i_itemp == NULL); | |
665 | iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP); | |
666 | ||
1da177e4 | 667 | iip->ili_inode = ip; |
43f5efc5 DC |
668 | xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE, |
669 | &xfs_inode_item_ops); | |
1da177e4 LT |
670 | iip->ili_format.ilf_type = XFS_LI_INODE; |
671 | iip->ili_format.ilf_ino = ip->i_ino; | |
92bfc6e7 CH |
672 | iip->ili_format.ilf_blkno = ip->i_imap.im_blkno; |
673 | iip->ili_format.ilf_len = ip->i_imap.im_len; | |
674 | iip->ili_format.ilf_boffset = ip->i_imap.im_boffset; | |
1da177e4 LT |
675 | } |
676 | ||
677 | /* | |
678 | * Free the inode log item and any memory hanging off of it. | |
679 | */ | |
680 | void | |
681 | xfs_inode_item_destroy( | |
682 | xfs_inode_t *ip) | |
683 | { | |
684 | #ifdef XFS_TRANS_DEBUG | |
685 | if (ip->i_itemp->ili_root_size != 0) { | |
f0e2d93c | 686 | kmem_free(ip->i_itemp->ili_orig_root); |
1da177e4 LT |
687 | } |
688 | #endif | |
689 | kmem_zone_free(xfs_ili_zone, ip->i_itemp); | |
690 | } | |
691 | ||
692 | ||
693 | /* | |
694 | * This is the inode flushing I/O completion routine. It is called | |
695 | * from interrupt level when the buffer containing the inode is | |
696 | * flushed to disk. It is responsible for removing the inode item | |
697 | * from the AIL if it has not been re-logged, and unlocking the inode's | |
698 | * flush lock. | |
30136832 DC |
699 | * |
700 | * To reduce AIL lock traffic as much as possible, we scan the buffer log item | |
701 | * list for other inodes that will run this function. We remove them from the | |
702 | * buffer list so we can process all the inode IO completions in one AIL lock | |
703 | * traversal. | |
1da177e4 | 704 | */ |
1da177e4 LT |
705 | void |
706 | xfs_iflush_done( | |
ca30b2a7 CH |
707 | struct xfs_buf *bp, |
708 | struct xfs_log_item *lip) | |
1da177e4 | 709 | { |
30136832 DC |
710 | struct xfs_inode_log_item *iip; |
711 | struct xfs_log_item *blip; | |
712 | struct xfs_log_item *next; | |
713 | struct xfs_log_item *prev; | |
ca30b2a7 | 714 | struct xfs_ail *ailp = lip->li_ailp; |
30136832 DC |
715 | int need_ail = 0; |
716 | ||
717 | /* | |
718 | * Scan the buffer IO completions for other inodes being completed and | |
719 | * attach them to the current inode log item. | |
720 | */ | |
adadbeef | 721 | blip = bp->b_fspriv; |
30136832 DC |
722 | prev = NULL; |
723 | while (blip != NULL) { | |
724 | if (lip->li_cb != xfs_iflush_done) { | |
725 | prev = blip; | |
726 | blip = blip->li_bio_list; | |
727 | continue; | |
728 | } | |
729 | ||
730 | /* remove from list */ | |
731 | next = blip->li_bio_list; | |
732 | if (!prev) { | |
adadbeef | 733 | bp->b_fspriv = next; |
30136832 DC |
734 | } else { |
735 | prev->li_bio_list = next; | |
736 | } | |
737 | ||
738 | /* add to current list */ | |
739 | blip->li_bio_list = lip->li_bio_list; | |
740 | lip->li_bio_list = blip; | |
741 | ||
742 | /* | |
743 | * while we have the item, do the unlocked check for needing | |
744 | * the AIL lock. | |
745 | */ | |
746 | iip = INODE_ITEM(blip); | |
747 | if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn) | |
748 | need_ail++; | |
749 | ||
750 | blip = next; | |
751 | } | |
752 | ||
753 | /* make sure we capture the state of the initial inode. */ | |
754 | iip = INODE_ITEM(lip); | |
755 | if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn) | |
756 | need_ail++; | |
1da177e4 LT |
757 | |
758 | /* | |
759 | * We only want to pull the item from the AIL if it is | |
760 | * actually there and its location in the log has not | |
761 | * changed since we started the flush. Thus, we only bother | |
762 | * if the ili_logged flag is set and the inode's lsn has not | |
763 | * changed. First we check the lsn outside | |
764 | * the lock since it's cheaper, and then we recheck while | |
765 | * holding the lock before removing the inode from the AIL. | |
766 | */ | |
30136832 DC |
767 | if (need_ail) { |
768 | struct xfs_log_item *log_items[need_ail]; | |
769 | int i = 0; | |
783a2f65 | 770 | spin_lock(&ailp->xa_lock); |
30136832 DC |
771 | for (blip = lip; blip; blip = blip->li_bio_list) { |
772 | iip = INODE_ITEM(blip); | |
773 | if (iip->ili_logged && | |
774 | blip->li_lsn == iip->ili_flush_lsn) { | |
775 | log_items[i++] = blip; | |
776 | } | |
777 | ASSERT(i <= need_ail); | |
1da177e4 | 778 | } |
30136832 DC |
779 | /* xfs_trans_ail_delete_bulk() drops the AIL lock. */ |
780 | xfs_trans_ail_delete_bulk(ailp, log_items, i); | |
1da177e4 LT |
781 | } |
782 | ||
1da177e4 LT |
783 | |
784 | /* | |
30136832 DC |
785 | * clean up and unlock the flush lock now we are done. We can clear the |
786 | * ili_last_fields bits now that we know that the data corresponding to | |
787 | * them is safely on disk. | |
1da177e4 | 788 | */ |
30136832 DC |
789 | for (blip = lip; blip; blip = next) { |
790 | next = blip->li_bio_list; | |
791 | blip->li_bio_list = NULL; | |
792 | ||
793 | iip = INODE_ITEM(blip); | |
794 | iip->ili_logged = 0; | |
795 | iip->ili_last_fields = 0; | |
796 | xfs_ifunlock(iip->ili_inode); | |
797 | } | |
1da177e4 LT |
798 | } |
799 | ||
800 | /* | |
801 | * This is the inode flushing abort routine. It is called | |
802 | * from xfs_iflush when the filesystem is shutting down to clean | |
803 | * up the inode state. | |
804 | * It is responsible for removing the inode item | |
805 | * from the AIL if it has not been re-logged, and unlocking the inode's | |
806 | * flush lock. | |
807 | */ | |
808 | void | |
809 | xfs_iflush_abort( | |
810 | xfs_inode_t *ip) | |
811 | { | |
783a2f65 | 812 | xfs_inode_log_item_t *iip = ip->i_itemp; |
1da177e4 | 813 | |
1da177e4 | 814 | if (iip) { |
783a2f65 | 815 | struct xfs_ail *ailp = iip->ili_item.li_ailp; |
1da177e4 | 816 | if (iip->ili_item.li_flags & XFS_LI_IN_AIL) { |
783a2f65 | 817 | spin_lock(&ailp->xa_lock); |
1da177e4 | 818 | if (iip->ili_item.li_flags & XFS_LI_IN_AIL) { |
783a2f65 DC |
819 | /* xfs_trans_ail_delete() drops the AIL lock. */ |
820 | xfs_trans_ail_delete(ailp, (xfs_log_item_t *)iip); | |
1da177e4 | 821 | } else |
783a2f65 | 822 | spin_unlock(&ailp->xa_lock); |
1da177e4 LT |
823 | } |
824 | iip->ili_logged = 0; | |
825 | /* | |
826 | * Clear the ili_last_fields bits now that we know that the | |
827 | * data corresponding to them is safely on disk. | |
828 | */ | |
829 | iip->ili_last_fields = 0; | |
830 | /* | |
831 | * Clear the inode logging fields so no more flushes are | |
832 | * attempted. | |
833 | */ | |
f5d8d5c4 | 834 | iip->ili_fields = 0; |
1da177e4 LT |
835 | } |
836 | /* | |
837 | * Release the inode's flush lock since we're done with it. | |
838 | */ | |
839 | xfs_ifunlock(ip); | |
840 | } | |
841 | ||
842 | void | |
843 | xfs_istale_done( | |
ca30b2a7 CH |
844 | struct xfs_buf *bp, |
845 | struct xfs_log_item *lip) | |
1da177e4 | 846 | { |
ca30b2a7 | 847 | xfs_iflush_abort(INODE_ITEM(lip)->ili_inode); |
1da177e4 | 848 | } |
6d192a9b TS |
849 | |
850 | /* | |
851 | * convert an xfs_inode_log_format struct from either 32 or 64 bit versions | |
852 | * (which can have different field alignments) to the native version | |
853 | */ | |
854 | int | |
855 | xfs_inode_item_format_convert( | |
856 | xfs_log_iovec_t *buf, | |
857 | xfs_inode_log_format_t *in_f) | |
858 | { | |
859 | if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) { | |
4e0d5f92 | 860 | xfs_inode_log_format_32_t *in_f32 = buf->i_addr; |
6d192a9b | 861 | |
6d192a9b TS |
862 | in_f->ilf_type = in_f32->ilf_type; |
863 | in_f->ilf_size = in_f32->ilf_size; | |
864 | in_f->ilf_fields = in_f32->ilf_fields; | |
865 | in_f->ilf_asize = in_f32->ilf_asize; | |
866 | in_f->ilf_dsize = in_f32->ilf_dsize; | |
867 | in_f->ilf_ino = in_f32->ilf_ino; | |
868 | /* copy biggest field of ilf_u */ | |
869 | memcpy(in_f->ilf_u.ilfu_uuid.__u_bits, | |
870 | in_f32->ilf_u.ilfu_uuid.__u_bits, | |
871 | sizeof(uuid_t)); | |
872 | in_f->ilf_blkno = in_f32->ilf_blkno; | |
873 | in_f->ilf_len = in_f32->ilf_len; | |
874 | in_f->ilf_boffset = in_f32->ilf_boffset; | |
875 | return 0; | |
876 | } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){ | |
4e0d5f92 | 877 | xfs_inode_log_format_64_t *in_f64 = buf->i_addr; |
6d192a9b | 878 | |
6d192a9b TS |
879 | in_f->ilf_type = in_f64->ilf_type; |
880 | in_f->ilf_size = in_f64->ilf_size; | |
881 | in_f->ilf_fields = in_f64->ilf_fields; | |
882 | in_f->ilf_asize = in_f64->ilf_asize; | |
883 | in_f->ilf_dsize = in_f64->ilf_dsize; | |
884 | in_f->ilf_ino = in_f64->ilf_ino; | |
885 | /* copy biggest field of ilf_u */ | |
886 | memcpy(in_f->ilf_u.ilfu_uuid.__u_bits, | |
887 | in_f64->ilf_u.ilfu_uuid.__u_bits, | |
888 | sizeof(uuid_t)); | |
889 | in_f->ilf_blkno = in_f64->ilf_blkno; | |
890 | in_f->ilf_len = in_f64->ilf_len; | |
891 | in_f->ilf_boffset = in_f64->ilf_boffset; | |
892 | return 0; | |
893 | } | |
894 | return EFSCORRUPTED; | |
895 | } |