blob: edc0aef4e51ebf8bf18642baefbc32bebc3f8dec [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_inode_item.h"
39#include "xfs_alloc.h"
40#include "xfs_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_attr.h"
43#include "xfs_attr_leaf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_dir2_data.h"
45#include "xfs_dir2_leaf.h"
46#include "xfs_dir2_block.h"
47#include "xfs_dir2_node.h"
48#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50/*
51 * xfs_da_btree.c
52 *
53 * Routines to implement directories as Btrees of hashed names.
54 */
55
56/*========================================================================
57 * Function prototypes for the kernel.
58 *========================================================================*/
59
60/*
61 * Routines used for growing the Btree.
62 */
63STATIC int xfs_da_root_split(xfs_da_state_t *state,
64 xfs_da_state_blk_t *existing_root,
65 xfs_da_state_blk_t *new_child);
66STATIC int xfs_da_node_split(xfs_da_state_t *state,
67 xfs_da_state_blk_t *existing_blk,
68 xfs_da_state_blk_t *split_blk,
69 xfs_da_state_blk_t *blk_to_add,
70 int treelevel,
71 int *result);
72STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
73 xfs_da_state_blk_t *node_blk_1,
74 xfs_da_state_blk_t *node_blk_2);
75STATIC void xfs_da_node_add(xfs_da_state_t *state,
76 xfs_da_state_blk_t *old_node_blk,
77 xfs_da_state_blk_t *new_node_blk);
78
79/*
80 * Routines used for shrinking the Btree.
81 */
82STATIC int xfs_da_root_join(xfs_da_state_t *state,
83 xfs_da_state_blk_t *root_blk);
84STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
85STATIC void xfs_da_node_remove(xfs_da_state_t *state,
86 xfs_da_state_blk_t *drop_blk);
87STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
88 xfs_da_state_blk_t *src_node_blk,
89 xfs_da_state_blk_t *dst_node_blk);
90
91/*
92 * Utility routines.
93 */
94STATIC uint xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
95STATIC int xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
96STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra);
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100097STATIC int xfs_da_blk_unlink(xfs_da_state_t *state,
98 xfs_da_state_blk_t *drop_blk,
99 xfs_da_state_blk_t *save_blk);
100STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102/*========================================================================
103 * Routines used for growing the Btree.
104 *========================================================================*/
105
106/*
107 * Create the initial contents of an intermediate node.
108 */
109int
110xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
111 xfs_dabuf_t **bpp, int whichfork)
112{
113 xfs_da_intnode_t *node;
114 xfs_dabuf_t *bp;
115 int error;
116 xfs_trans_t *tp;
117
118 tp = args->trans;
119 error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
120 if (error)
121 return(error);
122 ASSERT(bp != NULL);
123 node = bp->data;
124 node->hdr.info.forw = 0;
125 node->hdr.info.back = 0;
Nathan Scott89da0542006-03-17 17:28:40 +1100126 node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 node->hdr.info.pad = 0;
128 node->hdr.count = 0;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100129 node->hdr.level = cpu_to_be16(level);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
131 xfs_da_log_buf(tp, bp,
132 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
133
134 *bpp = bp;
135 return(0);
136}
137
138/*
139 * Split a leaf node, rebalance, then possibly split
140 * intermediate nodes, rebalance, etc.
141 */
142int /* error */
143xfs_da_split(xfs_da_state_t *state)
144{
145 xfs_da_state_blk_t *oldblk, *newblk, *addblk;
146 xfs_da_intnode_t *node;
147 xfs_dabuf_t *bp;
148 int max, action, error, i;
149
150 /*
151 * Walk back up the tree splitting/inserting/adjusting as necessary.
152 * If we need to insert and there isn't room, split the node, then
153 * decide which fragment to insert the new block from below into.
154 * Note that we may split the root this way, but we need more fixup.
155 */
156 max = state->path.active - 1;
157 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
158 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000159 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160
161 addblk = &state->path.blk[max]; /* initial dummy value */
162 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
163 oldblk = &state->path.blk[i];
164 newblk = &state->altpath.blk[i];
165
166 /*
167 * If a leaf node then
168 * Allocate a new leaf node, then rebalance across them.
169 * else if an intermediate node then
170 * We split on the last layer, must we split the node?
171 */
172 switch (oldblk->magic) {
173 case XFS_ATTR_LEAF_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 error = xfs_attr_leaf_split(state, oldblk, newblk);
175 if ((error != 0) && (error != ENOSPC)) {
176 return(error); /* GROT: attr is inconsistent */
177 }
178 if (!error) {
179 addblk = newblk;
180 break;
181 }
182 /*
183 * Entry wouldn't fit, split the leaf again.
184 */
185 state->extravalid = 1;
186 if (state->inleaf) {
187 state->extraafter = 0; /* before newblk */
188 error = xfs_attr_leaf_split(state, oldblk,
189 &state->extrablk);
190 } else {
191 state->extraafter = 1; /* after newblk */
192 error = xfs_attr_leaf_split(state, newblk,
193 &state->extrablk);
194 }
195 if (error)
196 return(error); /* GROT: attr inconsistent */
197 addblk = newblk;
198 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 error = xfs_dir2_leafn_split(state, oldblk, newblk);
201 if (error)
202 return error;
203 addblk = newblk;
204 break;
205 case XFS_DA_NODE_MAGIC:
206 error = xfs_da_node_split(state, oldblk, newblk, addblk,
207 max - i, &action);
208 xfs_da_buf_done(addblk->bp);
209 addblk->bp = NULL;
210 if (error)
211 return(error); /* GROT: dir is inconsistent */
212 /*
213 * Record the newly split block for the next time thru?
214 */
215 if (action)
216 addblk = newblk;
217 else
218 addblk = NULL;
219 break;
220 }
221
222 /*
223 * Update the btree to show the new hashval for this child.
224 */
225 xfs_da_fixhashpath(state, &state->path);
226 /*
227 * If we won't need this block again, it's getting dropped
228 * from the active path by the loop control, so we need
229 * to mark it done now.
230 */
231 if (i > 0 || !addblk)
232 xfs_da_buf_done(oldblk->bp);
233 }
234 if (!addblk)
235 return(0);
236
237 /*
238 * Split the root node.
239 */
240 ASSERT(state->path.active == 0);
241 oldblk = &state->path.blk[0];
242 error = xfs_da_root_split(state, oldblk, addblk);
243 if (error) {
244 xfs_da_buf_done(oldblk->bp);
245 xfs_da_buf_done(addblk->bp);
246 addblk->bp = NULL;
247 return(error); /* GROT: dir is inconsistent */
248 }
249
250 /*
251 * Update pointers to the node which used to be block 0 and
252 * just got bumped because of the addition of a new root node.
253 * There might be three blocks involved if a double split occurred,
254 * and the original block 0 could be at any position in the list.
255 */
256
257 node = oldblk->bp->data;
258 if (node->hdr.info.forw) {
Nathan Scott89da0542006-03-17 17:28:40 +1100259 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 bp = addblk->bp;
261 } else {
262 ASSERT(state->extravalid);
263 bp = state->extrablk.bp;
264 }
265 node = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100266 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 xfs_da_log_buf(state->args->trans, bp,
268 XFS_DA_LOGRANGE(node, &node->hdr.info,
269 sizeof(node->hdr.info)));
270 }
271 node = oldblk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100272 if (node->hdr.info.back) {
273 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 bp = addblk->bp;
275 } else {
276 ASSERT(state->extravalid);
277 bp = state->extrablk.bp;
278 }
279 node = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100280 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 xfs_da_log_buf(state->args->trans, bp,
282 XFS_DA_LOGRANGE(node, &node->hdr.info,
283 sizeof(node->hdr.info)));
284 }
285 xfs_da_buf_done(oldblk->bp);
286 xfs_da_buf_done(addblk->bp);
287 addblk->bp = NULL;
288 return(0);
289}
290
291/*
292 * Split the root. We have to create a new root and point to the two
293 * parts (the split old root) that we just created. Copy block zero to
294 * the EOF, extending the inode in process.
295 */
296STATIC int /* error */
297xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
298 xfs_da_state_blk_t *blk2)
299{
300 xfs_da_intnode_t *node, *oldroot;
301 xfs_da_args_t *args;
302 xfs_dablk_t blkno;
303 xfs_dabuf_t *bp;
304 int error, size;
305 xfs_inode_t *dp;
306 xfs_trans_t *tp;
307 xfs_mount_t *mp;
308 xfs_dir2_leaf_t *leaf;
309
310 /*
311 * Copy the existing (incorrect) block from the root node position
312 * to a free space somewhere.
313 */
314 args = state->args;
315 ASSERT(args != NULL);
316 error = xfs_da_grow_inode(args, &blkno);
317 if (error)
318 return(error);
319 dp = args->dp;
320 tp = args->trans;
321 mp = state->mp;
322 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
323 if (error)
324 return(error);
325 ASSERT(bp != NULL);
326 node = bp->data;
327 oldroot = blk1->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100328 if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC) {
Nathan Scottfac80cc2006-03-17 17:29:56 +1100329 size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 (char *)oldroot);
331 } else {
Nathan Scott89da0542006-03-17 17:28:40 +1100332 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 leaf = (xfs_dir2_leaf_t *)oldroot;
Nathan Scotta818e5d2006-03-17 17:28:07 +1100334 size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 (char *)leaf);
336 }
337 memcpy(node, oldroot, size);
338 xfs_da_log_buf(tp, bp, 0, size - 1);
339 xfs_da_buf_done(blk1->bp);
340 blk1->bp = bp;
341 blk1->blkno = blkno;
342
343 /*
344 * Set up the new root node.
345 */
346 error = xfs_da_node_create(args,
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000347 (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
Nathan Scottfac80cc2006-03-17 17:29:56 +1100348 be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 if (error)
350 return(error);
351 node = bp->data;
Nathan Scott403432d2006-03-17 17:29:46 +1100352 node->btree[0].hashval = cpu_to_be32(blk1->hashval);
353 node->btree[0].before = cpu_to_be32(blk1->blkno);
354 node->btree[1].hashval = cpu_to_be32(blk2->hashval);
355 node->btree[1].before = cpu_to_be32(blk2->blkno);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100356 node->hdr.count = cpu_to_be16(2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358#ifdef DEBUG
Nathan Scott89da0542006-03-17 17:28:40 +1100359 if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
361 blk1->blkno < mp->m_dirfreeblk);
362 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
363 blk2->blkno < mp->m_dirfreeblk);
364 }
365#endif
366
367 /* Header is already logged by xfs_da_node_create */
368 xfs_da_log_buf(tp, bp,
369 XFS_DA_LOGRANGE(node, node->btree,
370 sizeof(xfs_da_node_entry_t) * 2));
371 xfs_da_buf_done(bp);
372
373 return(0);
374}
375
376/*
377 * Split the node, rebalance, then add the new entry.
378 */
379STATIC int /* error */
380xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
381 xfs_da_state_blk_t *newblk,
382 xfs_da_state_blk_t *addblk,
383 int treelevel, int *result)
384{
385 xfs_da_intnode_t *node;
386 xfs_dablk_t blkno;
387 int newcount, error;
388 int useextra;
389
390 node = oldblk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100391 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
393 /*
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000394 * With V2 dirs the extra block is data or freespace.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000396 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 newcount = 1 + useextra;
398 /*
399 * Do we have to split the node?
400 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100401 if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 /*
403 * Allocate a new node, add to the doubly linked chain of
404 * nodes, then move some of our excess entries into it.
405 */
406 error = xfs_da_grow_inode(state->args, &blkno);
407 if (error)
408 return(error); /* GROT: dir is inconsistent */
409
410 error = xfs_da_node_create(state->args, blkno, treelevel,
411 &newblk->bp, state->args->whichfork);
412 if (error)
413 return(error); /* GROT: dir is inconsistent */
414 newblk->blkno = blkno;
415 newblk->magic = XFS_DA_NODE_MAGIC;
416 xfs_da_node_rebalance(state, oldblk, newblk);
417 error = xfs_da_blk_link(state, oldblk, newblk);
418 if (error)
419 return(error);
420 *result = 1;
421 } else {
422 *result = 0;
423 }
424
425 /*
426 * Insert the new entry(s) into the correct block
427 * (updating last hashval in the process).
428 *
429 * xfs_da_node_add() inserts BEFORE the given index,
430 * and as a result of using node_lookup_int() we always
431 * point to a valid entry (not after one), but a split
432 * operation always results in a new block whose hashvals
433 * FOLLOW the current block.
434 *
435 * If we had double-split op below us, then add the extra block too.
436 */
437 node = oldblk->bp->data;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100438 if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 oldblk->index++;
440 xfs_da_node_add(state, oldblk, addblk);
441 if (useextra) {
442 if (state->extraafter)
443 oldblk->index++;
444 xfs_da_node_add(state, oldblk, &state->extrablk);
445 state->extravalid = 0;
446 }
447 } else {
448 newblk->index++;
449 xfs_da_node_add(state, newblk, addblk);
450 if (useextra) {
451 if (state->extraafter)
452 newblk->index++;
453 xfs_da_node_add(state, newblk, &state->extrablk);
454 state->extravalid = 0;
455 }
456 }
457
458 return(0);
459}
460
461/*
462 * Balance the btree elements between two intermediate nodes,
463 * usually one full and one empty.
464 *
465 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
466 */
467STATIC void
468xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
469 xfs_da_state_blk_t *blk2)
470{
471 xfs_da_intnode_t *node1, *node2, *tmpnode;
472 xfs_da_node_entry_t *btree_s, *btree_d;
473 int count, tmp;
474 xfs_trans_t *tp;
475
476 node1 = blk1->bp->data;
477 node2 = blk2->bp->data;
478 /*
479 * Figure out how many entries need to move, and in which direction.
480 * Swap the nodes around if that makes it simpler.
481 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100482 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
Nathan Scott403432d2006-03-17 17:29:46 +1100483 ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
Nathan Scottfac80cc2006-03-17 17:29:56 +1100484 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
485 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 tmpnode = node1;
487 node1 = node2;
488 node2 = tmpnode;
489 }
Nathan Scott89da0542006-03-17 17:28:40 +1100490 ASSERT(be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC);
491 ASSERT(be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100492 count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 if (count == 0)
494 return;
495 tp = state->args->trans;
496 /*
497 * Two cases: high-to-low and low-to-high.
498 */
499 if (count > 0) {
500 /*
501 * Move elements in node2 up to make a hole.
502 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100503 if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 tmp *= (uint)sizeof(xfs_da_node_entry_t);
505 btree_s = &node2->btree[0];
506 btree_d = &node2->btree[count];
507 memmove(btree_d, btree_s, tmp);
508 }
509
510 /*
511 * Move the req'd B-tree elements from high in node1 to
512 * low in node2.
513 */
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800514 be16_add_cpu(&node2->hdr.count, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100516 btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 btree_d = &node2->btree[0];
518 memcpy(btree_d, btree_s, tmp);
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800519 be16_add_cpu(&node1->hdr.count, -count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 } else {
521 /*
522 * Move the req'd B-tree elements from low in node2 to
523 * high in node1.
524 */
525 count = -count;
526 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
527 btree_s = &node2->btree[0];
Nathan Scottfac80cc2006-03-17 17:29:56 +1100528 btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 memcpy(btree_d, btree_s, tmp);
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800530 be16_add_cpu(&node1->hdr.count, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 xfs_da_log_buf(tp, blk1->bp,
532 XFS_DA_LOGRANGE(node1, btree_d, tmp));
533
534 /*
535 * Move elements in node2 down to fill the hole.
536 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100537 tmp = be16_to_cpu(node2->hdr.count) - count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 tmp *= (uint)sizeof(xfs_da_node_entry_t);
539 btree_s = &node2->btree[count];
540 btree_d = &node2->btree[0];
541 memmove(btree_d, btree_s, tmp);
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800542 be16_add_cpu(&node2->hdr.count, -count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 }
544
545 /*
546 * Log header of node 1 and all current bits of node 2.
547 */
548 xfs_da_log_buf(tp, blk1->bp,
549 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
550 xfs_da_log_buf(tp, blk2->bp,
551 XFS_DA_LOGRANGE(node2, &node2->hdr,
552 sizeof(node2->hdr) +
Nathan Scottfac80cc2006-03-17 17:29:56 +1100553 sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 /*
556 * Record the last hashval from each block for upward propagation.
557 * (note: don't use the swapped node pointers)
558 */
559 node1 = blk1->bp->data;
560 node2 = blk2->bp->data;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100561 blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
562 blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564 /*
565 * Adjust the expected index for insertion.
566 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100567 if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
568 blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
569 blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 }
571}
572
573/*
574 * Add a new entry to an intermediate node.
575 */
576STATIC void
577xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
578 xfs_da_state_blk_t *newblk)
579{
580 xfs_da_intnode_t *node;
581 xfs_da_node_entry_t *btree;
582 int tmp;
583 xfs_mount_t *mp;
584
585 node = oldblk->bp->data;
586 mp = state->mp;
Nathan Scott89da0542006-03-17 17:28:40 +1100587 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100588 ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 ASSERT(newblk->blkno != 0);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000590 if (state->args->whichfork == XFS_DATA_FORK)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 ASSERT(newblk->blkno >= mp->m_dirleafblk &&
592 newblk->blkno < mp->m_dirfreeblk);
593
594 /*
595 * We may need to make some room before we insert the new node.
596 */
597 tmp = 0;
598 btree = &node->btree[ oldblk->index ];
Nathan Scottfac80cc2006-03-17 17:29:56 +1100599 if (oldblk->index < be16_to_cpu(node->hdr.count)) {
600 tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 memmove(btree + 1, btree, tmp);
602 }
Nathan Scott403432d2006-03-17 17:29:46 +1100603 btree->hashval = cpu_to_be32(newblk->hashval);
604 btree->before = cpu_to_be32(newblk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 xfs_da_log_buf(state->args->trans, oldblk->bp,
606 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800607 be16_add_cpu(&node->hdr.count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 xfs_da_log_buf(state->args->trans, oldblk->bp,
609 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
610
611 /*
612 * Copy the last hash value from the oldblk to propagate upwards.
613 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100614 oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615}
616
617/*========================================================================
618 * Routines used for shrinking the Btree.
619 *========================================================================*/
620
621/*
622 * Deallocate an empty leaf node, remove it from its parent,
623 * possibly deallocating that block, etc...
624 */
625int
626xfs_da_join(xfs_da_state_t *state)
627{
628 xfs_da_state_blk_t *drop_blk, *save_blk;
629 int action, error;
630
631 action = 0;
632 drop_blk = &state->path.blk[ state->path.active-1 ];
633 save_blk = &state->altpath.blk[ state->path.active-1 ];
634 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
635 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000636 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
638 /*
639 * Walk back up the tree joining/deallocating as necessary.
640 * When we stop dropping blocks, break out.
641 */
642 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
643 state->path.active--) {
644 /*
645 * See if we can combine the block with a neighbor.
646 * (action == 0) => no options, just leave
647 * (action == 1) => coalesce, then unlink
648 * (action == 2) => block empty, unlink it
649 */
650 switch (drop_blk->magic) {
651 case XFS_ATTR_LEAF_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 error = xfs_attr_leaf_toosmall(state, &action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 if (error)
654 return(error);
655 if (action == 0)
656 return(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 error = xfs_dir2_leafn_toosmall(state, &action);
661 if (error)
662 return error;
663 if (action == 0)
664 return 0;
665 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
666 break;
667 case XFS_DA_NODE_MAGIC:
668 /*
669 * Remove the offending node, fixup hashvals,
670 * check for a toosmall neighbor.
671 */
672 xfs_da_node_remove(state, drop_blk);
673 xfs_da_fixhashpath(state, &state->path);
674 error = xfs_da_node_toosmall(state, &action);
675 if (error)
676 return(error);
677 if (action == 0)
678 return 0;
679 xfs_da_node_unbalance(state, drop_blk, save_blk);
680 break;
681 }
682 xfs_da_fixhashpath(state, &state->altpath);
683 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
684 xfs_da_state_kill_altpath(state);
685 if (error)
686 return(error);
687 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
688 drop_blk->bp);
689 drop_blk->bp = NULL;
690 if (error)
691 return(error);
692 }
693 /*
694 * We joined all the way to the top. If it turns out that
695 * we only have one entry in the root, make the child block
696 * the new root.
697 */
698 xfs_da_node_remove(state, drop_blk);
699 xfs_da_fixhashpath(state, &state->path);
700 error = xfs_da_root_join(state, &state->path.blk[0]);
701 return(error);
702}
703
704/*
705 * We have only one entry in the root. Copy the only remaining child of
706 * the old root to block 0 as the new root node.
707 */
708STATIC int
709xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
710{
711 xfs_da_intnode_t *oldroot;
712 /* REFERENCED */
713 xfs_da_blkinfo_t *blkinfo;
714 xfs_da_args_t *args;
715 xfs_dablk_t child;
716 xfs_dabuf_t *bp;
717 int error;
718
719 args = state->args;
720 ASSERT(args != NULL);
721 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
722 oldroot = root_blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100723 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 ASSERT(!oldroot->hdr.info.forw);
725 ASSERT(!oldroot->hdr.info.back);
726
727 /*
728 * If the root has more than one child, then don't do anything.
729 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100730 if (be16_to_cpu(oldroot->hdr.count) > 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 return(0);
732
733 /*
734 * Read in the (only) child block, then copy those bytes into
735 * the root block's buffer and free the original child block.
736 */
Nathan Scott403432d2006-03-17 17:29:46 +1100737 child = be32_to_cpu(oldroot->btree[0].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 ASSERT(child != 0);
739 error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
740 args->whichfork);
741 if (error)
742 return(error);
743 ASSERT(bp != NULL);
744 blkinfo = bp->data;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100745 if (be16_to_cpu(oldroot->hdr.level) == 1) {
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000746 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DIR2_LEAFN_MAGIC ||
Nathan Scott89da0542006-03-17 17:28:40 +1100747 be16_to_cpu(blkinfo->magic) == XFS_ATTR_LEAF_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 } else {
Nathan Scott89da0542006-03-17 17:28:40 +1100749 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
751 ASSERT(!blkinfo->forw);
752 ASSERT(!blkinfo->back);
753 memcpy(root_blk->bp->data, bp->data, state->blocksize);
754 xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
755 error = xfs_da_shrink_inode(args, child, bp);
756 return(error);
757}
758
759/*
760 * Check a node block and its neighbors to see if the block should be
761 * collapsed into one or the other neighbor. Always keep the block
762 * with the smaller block number.
763 * If the current block is over 50% full, don't try to join it, return 0.
764 * If the block is empty, fill in the state structure and return 2.
765 * If it can be collapsed, fill in the state structure and return 1.
766 * If nothing can be done, return 0.
767 */
768STATIC int
769xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
770{
771 xfs_da_intnode_t *node;
772 xfs_da_state_blk_t *blk;
773 xfs_da_blkinfo_t *info;
774 int count, forward, error, retval, i;
775 xfs_dablk_t blkno;
776 xfs_dabuf_t *bp;
777
778 /*
779 * Check for the degenerate case of the block being over 50% full.
780 * If so, it's not worth even looking to see if we might be able
781 * to coalesce with a sibling.
782 */
783 blk = &state->path.blk[ state->path.active-1 ];
784 info = blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100785 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 node = (xfs_da_intnode_t *)info;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100787 count = be16_to_cpu(node->hdr.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 if (count > (state->node_ents >> 1)) {
789 *action = 0; /* blk over 50%, don't try to join */
790 return(0); /* blk over 50%, don't try to join */
791 }
792
793 /*
794 * Check for the degenerate case of the block being empty.
795 * If the block is empty, we'll simply delete it, no need to
Nathan Scottc41564b2006-03-29 08:55:14 +1000796 * coalesce it with a sibling block. We choose (arbitrarily)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 * to merge with the forward block unless it is NULL.
798 */
799 if (count == 0) {
800 /*
801 * Make altpath point to the block we want to keep and
802 * path point to the block we want to drop (this one).
803 */
Nathan Scott89da0542006-03-17 17:28:40 +1100804 forward = (info->forw != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 memcpy(&state->altpath, &state->path, sizeof(state->path));
806 error = xfs_da_path_shift(state, &state->altpath, forward,
807 0, &retval);
808 if (error)
809 return(error);
810 if (retval) {
811 *action = 0;
812 } else {
813 *action = 2;
814 }
815 return(0);
816 }
817
818 /*
819 * Examine each sibling block to see if we can coalesce with
820 * at least 25% free space to spare. We need to figure out
821 * whether to merge with the forward or the backward block.
822 * We prefer coalescing with the lower numbered sibling so as
823 * to shrink a directory over time.
824 */
825 /* start with smaller blk num */
Nathan Scott89da0542006-03-17 17:28:40 +1100826 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 for (i = 0; i < 2; forward = !forward, i++) {
828 if (forward)
Nathan Scott89da0542006-03-17 17:28:40 +1100829 blkno = be32_to_cpu(info->forw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 else
Nathan Scott89da0542006-03-17 17:28:40 +1100831 blkno = be32_to_cpu(info->back);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 if (blkno == 0)
833 continue;
834 error = xfs_da_read_buf(state->args->trans, state->args->dp,
835 blkno, -1, &bp, state->args->whichfork);
836 if (error)
837 return(error);
838 ASSERT(bp != NULL);
839
840 node = (xfs_da_intnode_t *)info;
841 count = state->node_ents;
842 count -= state->node_ents >> 2;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100843 count -= be16_to_cpu(node->hdr.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 node = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100845 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100846 count -= be16_to_cpu(node->hdr.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 xfs_da_brelse(state->args->trans, bp);
848 if (count >= 0)
849 break; /* fits with at least 25% to spare */
850 }
851 if (i >= 2) {
852 *action = 0;
853 return(0);
854 }
855
856 /*
857 * Make altpath point to the block we want to keep (the lower
858 * numbered block) and path point to the block we want to drop.
859 */
860 memcpy(&state->altpath, &state->path, sizeof(state->path));
861 if (blkno < blk->blkno) {
862 error = xfs_da_path_shift(state, &state->altpath, forward,
863 0, &retval);
864 if (error) {
865 return(error);
866 }
867 if (retval) {
868 *action = 0;
869 return(0);
870 }
871 } else {
872 error = xfs_da_path_shift(state, &state->path, forward,
873 0, &retval);
874 if (error) {
875 return(error);
876 }
877 if (retval) {
878 *action = 0;
879 return(0);
880 }
881 }
882 *action = 1;
883 return(0);
884}
885
886/*
887 * Walk back up the tree adjusting hash values as necessary,
888 * when we stop making changes, return.
889 */
890void
891xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
892{
893 xfs_da_state_blk_t *blk;
894 xfs_da_intnode_t *node;
895 xfs_da_node_entry_t *btree;
896 xfs_dahash_t lasthash=0;
897 int level, count;
898
899 level = path->active-1;
900 blk = &path->blk[ level ];
901 switch (blk->magic) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 case XFS_ATTR_LEAF_MAGIC:
903 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
904 if (count == 0)
905 return;
906 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
909 if (count == 0)
910 return;
911 break;
912 case XFS_DA_NODE_MAGIC:
913 lasthash = xfs_da_node_lasthash(blk->bp, &count);
914 if (count == 0)
915 return;
916 break;
917 }
918 for (blk--, level--; level >= 0; blk--, level--) {
919 node = blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100920 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 btree = &node->btree[ blk->index ];
Nathan Scott403432d2006-03-17 17:29:46 +1100922 if (be32_to_cpu(btree->hashval) == lasthash)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 break;
924 blk->hashval = lasthash;
Nathan Scott403432d2006-03-17 17:29:46 +1100925 btree->hashval = cpu_to_be32(lasthash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 xfs_da_log_buf(state->args->trans, blk->bp,
927 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
928
Nathan Scottfac80cc2006-03-17 17:29:56 +1100929 lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 }
931}
932
933/*
934 * Remove an entry from an intermediate node.
935 */
936STATIC void
937xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
938{
939 xfs_da_intnode_t *node;
940 xfs_da_node_entry_t *btree;
941 int tmp;
942
943 node = drop_blk->bp->data;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100944 ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 ASSERT(drop_blk->index >= 0);
946
947 /*
948 * Copy over the offending entry, or just zero it out.
949 */
950 btree = &node->btree[drop_blk->index];
Nathan Scottfac80cc2006-03-17 17:29:56 +1100951 if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
952 tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 tmp *= (uint)sizeof(xfs_da_node_entry_t);
954 memmove(btree, btree + 1, tmp);
955 xfs_da_log_buf(state->args->trans, drop_blk->bp,
956 XFS_DA_LOGRANGE(node, btree, tmp));
Nathan Scottfac80cc2006-03-17 17:29:56 +1100957 btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 }
959 memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
960 xfs_da_log_buf(state->args->trans, drop_blk->bp,
961 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800962 be16_add_cpu(&node->hdr.count, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 xfs_da_log_buf(state->args->trans, drop_blk->bp,
964 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
965
966 /*
967 * Copy the last hash value from the block to propagate upwards.
968 */
969 btree--;
Nathan Scott403432d2006-03-17 17:29:46 +1100970 drop_blk->hashval = be32_to_cpu(btree->hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
972
973/*
974 * Unbalance the btree elements between two intermediate nodes,
975 * move all Btree elements from one node into another.
976 */
977STATIC void
978xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
979 xfs_da_state_blk_t *save_blk)
980{
981 xfs_da_intnode_t *drop_node, *save_node;
982 xfs_da_node_entry_t *btree;
983 int tmp;
984 xfs_trans_t *tp;
985
986 drop_node = drop_blk->bp->data;
987 save_node = save_blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100988 ASSERT(be16_to_cpu(drop_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
989 ASSERT(be16_to_cpu(save_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 tp = state->args->trans;
991
992 /*
993 * If the dying block has lower hashvals, then move all the
994 * elements in the remaining block up to make a hole.
995 */
Nathan Scott403432d2006-03-17 17:29:46 +1100996 if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
Nathan Scottfac80cc2006-03-17 17:29:56 +1100997 (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
998 be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 {
Nathan Scottfac80cc2006-03-17 17:29:56 +11001000 btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
1001 tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 memmove(btree, &save_node->btree[0], tmp);
1003 btree = &save_node->btree[0];
1004 xfs_da_log_buf(tp, save_blk->bp,
1005 XFS_DA_LOGRANGE(save_node, btree,
Nathan Scottfac80cc2006-03-17 17:29:56 +11001006 (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 sizeof(xfs_da_node_entry_t)));
1008 } else {
Nathan Scottfac80cc2006-03-17 17:29:56 +11001009 btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 xfs_da_log_buf(tp, save_blk->bp,
1011 XFS_DA_LOGRANGE(save_node, btree,
Nathan Scottfac80cc2006-03-17 17:29:56 +11001012 be16_to_cpu(drop_node->hdr.count) *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 sizeof(xfs_da_node_entry_t)));
1014 }
1015
1016 /*
1017 * Move all the B-tree elements from drop_blk to save_blk.
1018 */
Nathan Scottfac80cc2006-03-17 17:29:56 +11001019 tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 memcpy(btree, &drop_node->btree[0], tmp);
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001021 be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
1023 xfs_da_log_buf(tp, save_blk->bp,
1024 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1025 sizeof(save_node->hdr)));
1026
1027 /*
1028 * Save the last hashval in the remaining block for upward propagation.
1029 */
Nathan Scottfac80cc2006-03-17 17:29:56 +11001030 save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031}
1032
1033/*========================================================================
1034 * Routines used for finding things in the Btree.
1035 *========================================================================*/
1036
1037/*
1038 * Walk down the Btree looking for a particular filename, filling
1039 * in the state structure as we go.
1040 *
1041 * We will set the state structure to point to each of the elements
1042 * in each of the nodes where either the hashval is or should be.
1043 *
1044 * We support duplicate hashval's so for each entry in the current
1045 * node that could contain the desired hashval, descend. This is a
1046 * pruned depth-first tree search.
1047 */
1048int /* error */
1049xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1050{
1051 xfs_da_state_blk_t *blk;
1052 xfs_da_blkinfo_t *curr;
1053 xfs_da_intnode_t *node;
1054 xfs_da_node_entry_t *btree;
1055 xfs_dablk_t blkno;
1056 int probe, span, max, error, retval;
Nathan Scottd432c802006-09-28 11:03:44 +10001057 xfs_dahash_t hashval, btreehashval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 xfs_da_args_t *args;
1059
1060 args = state->args;
1061
1062 /*
1063 * Descend thru the B-tree searching each level for the right
1064 * node to use, until the right hashval is found.
1065 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001066 blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 for (blk = &state->path.blk[0], state->path.active = 1;
1068 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1069 blk++, state->path.active++) {
1070 /*
1071 * Read the next node down in the tree.
1072 */
1073 blk->blkno = blkno;
1074 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1075 -1, &blk->bp, args->whichfork);
1076 if (error) {
1077 blk->blkno = 0;
1078 state->path.active--;
1079 return(error);
1080 }
1081 curr = blk->bp->data;
Nathan Scottd432c802006-09-28 11:03:44 +10001082 blk->magic = be16_to_cpu(curr->magic);
1083 ASSERT(blk->magic == XFS_DA_NODE_MAGIC ||
1084 blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1085 blk->magic == XFS_ATTR_LEAF_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087 /*
1088 * Search an intermediate node for a match.
1089 */
Nathan Scott89da0542006-03-17 17:28:40 +11001090 if (blk->magic == XFS_DA_NODE_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 node = blk->bp->data;
Nathan Scottd432c802006-09-28 11:03:44 +10001092 max = be16_to_cpu(node->hdr.count);
Lachlan McIlroy1c91ad32007-02-10 18:35:27 +11001093 blk->hashval = be32_to_cpu(node->btree[max-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
1095 /*
1096 * Binary search. (note: small blocks will skip loop)
1097 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 probe = span = max / 2;
1099 hashval = args->hashval;
1100 for (btree = &node->btree[probe]; span > 4;
1101 btree = &node->btree[probe]) {
1102 span /= 2;
Nathan Scottd432c802006-09-28 11:03:44 +10001103 btreehashval = be32_to_cpu(btree->hashval);
1104 if (btreehashval < hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 probe += span;
Nathan Scottd432c802006-09-28 11:03:44 +10001106 else if (btreehashval > hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 probe -= span;
1108 else
1109 break;
1110 }
1111 ASSERT((probe >= 0) && (probe < max));
Nathan Scott403432d2006-03-17 17:29:46 +11001112 ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
1114 /*
1115 * Since we may have duplicate hashval's, find the first
1116 * matching hashval in the node.
1117 */
Nathan Scott403432d2006-03-17 17:29:46 +11001118 while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 btree--;
1120 probe--;
1121 }
Nathan Scott403432d2006-03-17 17:29:46 +11001122 while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 btree++;
1124 probe++;
1125 }
1126
1127 /*
1128 * Pick the right block to descend on.
1129 */
1130 if (probe == max) {
1131 blk->index = max-1;
Nathan Scott403432d2006-03-17 17:29:46 +11001132 blkno = be32_to_cpu(node->btree[max-1].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 } else {
1134 blk->index = probe;
Nathan Scott403432d2006-03-17 17:29:46 +11001135 blkno = be32_to_cpu(btree->before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 }
Nathan Scottd432c802006-09-28 11:03:44 +10001137 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1139 break;
Nathan Scottd432c802006-09-28 11:03:44 +10001140 } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1142 break;
1143 }
1144 }
1145
1146 /*
1147 * A leaf block that ends in the hashval that we are interested in
1148 * (final hashval == search hashval) means that the next block may
1149 * contain more entries with the same hashval, shift upward to the
1150 * next leaf and keep searching.
1151 */
1152 for (;;) {
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001153 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1155 &blk->index, state);
Nathan Scottd432c802006-09-28 11:03:44 +10001156 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1158 blk->index = args->index;
1159 args->blkno = blk->blkno;
Nathan Scottd432c802006-09-28 11:03:44 +10001160 } else {
1161 ASSERT(0);
1162 return XFS_ERROR(EFSCORRUPTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1165 (blk->hashval == args->hashval)) {
1166 error = xfs_da_path_shift(state, &state->path, 1, 1,
1167 &retval);
1168 if (error)
1169 return(error);
1170 if (retval == 0) {
1171 continue;
Nathan Scottd432c802006-09-28 11:03:44 +10001172 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 /* path_shift() gives ENOENT */
1174 retval = XFS_ERROR(ENOATTR);
1175 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 }
1177 break;
1178 }
1179 *result = retval;
1180 return(0);
1181}
1182
1183/*========================================================================
1184 * Utility routines.
1185 *========================================================================*/
1186
1187/*
1188 * Link a new block into a doubly linked list of blocks (of whatever type).
1189 */
1190int /* error */
1191xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1192 xfs_da_state_blk_t *new_blk)
1193{
1194 xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1195 xfs_da_args_t *args;
1196 int before=0, error;
1197 xfs_dabuf_t *bp;
1198
1199 /*
1200 * Set up environment.
1201 */
1202 args = state->args;
1203 ASSERT(args != NULL);
1204 old_info = old_blk->bp->data;
1205 new_info = new_blk->bp->data;
1206 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001207 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
Nathan Scott89da0542006-03-17 17:28:40 +11001209 ASSERT(old_blk->magic == be16_to_cpu(old_info->magic));
1210 ASSERT(new_blk->magic == be16_to_cpu(new_info->magic));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 ASSERT(old_blk->magic == new_blk->magic);
1212
1213 switch (old_blk->magic) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 case XFS_ATTR_LEAF_MAGIC:
1215 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1216 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1219 break;
1220 case XFS_DA_NODE_MAGIC:
1221 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1222 break;
1223 }
1224
1225 /*
1226 * Link blocks in appropriate order.
1227 */
1228 if (before) {
1229 /*
1230 * Link new block in before existing block.
1231 */
Nathan Scott89da0542006-03-17 17:28:40 +11001232 new_info->forw = cpu_to_be32(old_blk->blkno);
1233 new_info->back = old_info->back;
1234 if (old_info->back) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 error = xfs_da_read_buf(args->trans, args->dp,
Nathan Scott89da0542006-03-17 17:28:40 +11001236 be32_to_cpu(old_info->back),
1237 -1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 if (error)
1239 return(error);
1240 ASSERT(bp != NULL);
1241 tmp_info = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001242 ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic));
1243 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1244 tmp_info->forw = cpu_to_be32(new_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1246 xfs_da_buf_done(bp);
1247 }
Nathan Scott89da0542006-03-17 17:28:40 +11001248 old_info->back = cpu_to_be32(new_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 } else {
1250 /*
1251 * Link new block in after existing block.
1252 */
Nathan Scott89da0542006-03-17 17:28:40 +11001253 new_info->forw = old_info->forw;
1254 new_info->back = cpu_to_be32(old_blk->blkno);
1255 if (old_info->forw) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 error = xfs_da_read_buf(args->trans, args->dp,
Nathan Scott89da0542006-03-17 17:28:40 +11001257 be32_to_cpu(old_info->forw),
1258 -1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 if (error)
1260 return(error);
1261 ASSERT(bp != NULL);
1262 tmp_info = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001263 ASSERT(tmp_info->magic == old_info->magic);
1264 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1265 tmp_info->back = cpu_to_be32(new_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1267 xfs_da_buf_done(bp);
1268 }
Nathan Scott89da0542006-03-17 17:28:40 +11001269 old_info->forw = cpu_to_be32(new_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 }
1271
1272 xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1273 xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1274 return(0);
1275}
1276
1277/*
1278 * Compare two intermediate nodes for "order".
1279 */
1280STATIC int
1281xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1282{
1283 xfs_da_intnode_t *node1, *node2;
1284
1285 node1 = node1_bp->data;
1286 node2 = node2_bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001287 ASSERT((be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC) &&
1288 (be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC));
Nathan Scottfac80cc2006-03-17 17:29:56 +11001289 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
Nathan Scott403432d2006-03-17 17:29:46 +11001290 ((be32_to_cpu(node2->btree[0].hashval) <
1291 be32_to_cpu(node1->btree[0].hashval)) ||
Nathan Scottfac80cc2006-03-17 17:29:56 +11001292 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
1293 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 return(1);
1295 }
1296 return(0);
1297}
1298
1299/*
1300 * Pick up the last hashvalue from an intermediate node.
1301 */
1302STATIC uint
1303xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1304{
1305 xfs_da_intnode_t *node;
1306
1307 node = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001308 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 if (count)
Nathan Scottfac80cc2006-03-17 17:29:56 +11001310 *count = be16_to_cpu(node->hdr.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 if (!node->hdr.count)
1312 return(0);
Nathan Scottfac80cc2006-03-17 17:29:56 +11001313 return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
1315
1316/*
1317 * Unlink a block from a doubly linked list of blocks.
1318 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001319STATIC int /* error */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1321 xfs_da_state_blk_t *save_blk)
1322{
1323 xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1324 xfs_da_args_t *args;
1325 xfs_dabuf_t *bp;
1326 int error;
1327
1328 /*
1329 * Set up environment.
1330 */
1331 args = state->args;
1332 ASSERT(args != NULL);
1333 save_info = save_blk->bp->data;
1334 drop_info = drop_blk->bp->data;
1335 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001336 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
Nathan Scott89da0542006-03-17 17:28:40 +11001338 ASSERT(save_blk->magic == be16_to_cpu(save_info->magic));
1339 ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 ASSERT(save_blk->magic == drop_blk->magic);
Nathan Scott89da0542006-03-17 17:28:40 +11001341 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1342 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1343 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1344 (be32_to_cpu(drop_info->back) == save_blk->blkno));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
1346 /*
1347 * Unlink the leaf block from the doubly linked chain of leaves.
1348 */
Nathan Scott89da0542006-03-17 17:28:40 +11001349 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1350 save_info->back = drop_info->back;
1351 if (drop_info->back) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 error = xfs_da_read_buf(args->trans, args->dp,
Nathan Scott89da0542006-03-17 17:28:40 +11001353 be32_to_cpu(drop_info->back),
1354 -1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 if (error)
1356 return(error);
1357 ASSERT(bp != NULL);
1358 tmp_info = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001359 ASSERT(tmp_info->magic == save_info->magic);
1360 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1361 tmp_info->forw = cpu_to_be32(save_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 xfs_da_log_buf(args->trans, bp, 0,
1363 sizeof(*tmp_info) - 1);
1364 xfs_da_buf_done(bp);
1365 }
1366 } else {
Nathan Scott89da0542006-03-17 17:28:40 +11001367 save_info->forw = drop_info->forw;
1368 if (drop_info->forw) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 error = xfs_da_read_buf(args->trans, args->dp,
Nathan Scott89da0542006-03-17 17:28:40 +11001370 be32_to_cpu(drop_info->forw),
1371 -1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (error)
1373 return(error);
1374 ASSERT(bp != NULL);
1375 tmp_info = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001376 ASSERT(tmp_info->magic == save_info->magic);
1377 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1378 tmp_info->back = cpu_to_be32(save_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 xfs_da_log_buf(args->trans, bp, 0,
1380 sizeof(*tmp_info) - 1);
1381 xfs_da_buf_done(bp);
1382 }
1383 }
1384
1385 xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1386 return(0);
1387}
1388
1389/*
1390 * Move a path "forward" or "!forward" one block at the current level.
1391 *
1392 * This routine will adjust a "path" to point to the next block
1393 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1394 * Btree, including updating pointers to the intermediate nodes between
1395 * the new bottom and the root.
1396 */
1397int /* error */
1398xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1399 int forward, int release, int *result)
1400{
1401 xfs_da_state_blk_t *blk;
1402 xfs_da_blkinfo_t *info;
1403 xfs_da_intnode_t *node;
1404 xfs_da_args_t *args;
1405 xfs_dablk_t blkno=0;
1406 int level, error;
1407
1408 /*
1409 * Roll up the Btree looking for the first block where our
1410 * current index is not at the edge of the block. Note that
1411 * we skip the bottom layer because we want the sibling block.
1412 */
1413 args = state->args;
1414 ASSERT(args != NULL);
1415 ASSERT(path != NULL);
1416 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1417 level = (path->active-1) - 1; /* skip bottom layer in path */
1418 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1419 ASSERT(blk->bp != NULL);
1420 node = blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001421 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Nathan Scottfac80cc2006-03-17 17:29:56 +11001422 if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 blk->index++;
Nathan Scott403432d2006-03-17 17:29:46 +11001424 blkno = be32_to_cpu(node->btree[blk->index].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 break;
1426 } else if (!forward && (blk->index > 0)) {
1427 blk->index--;
Nathan Scott403432d2006-03-17 17:29:46 +11001428 blkno = be32_to_cpu(node->btree[blk->index].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 break;
1430 }
1431 }
1432 if (level < 0) {
1433 *result = XFS_ERROR(ENOENT); /* we're out of our tree */
Barry Naujok6a178102008-05-21 16:42:05 +10001434 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 return(0);
1436 }
1437
1438 /*
1439 * Roll down the edge of the subtree until we reach the
1440 * same depth we were at originally.
1441 */
1442 for (blk++, level++; level < path->active; blk++, level++) {
1443 /*
1444 * Release the old block.
1445 * (if it's dirty, trans won't actually let go)
1446 */
1447 if (release)
1448 xfs_da_brelse(args->trans, blk->bp);
1449
1450 /*
1451 * Read the next child block.
1452 */
1453 blk->blkno = blkno;
1454 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1455 &blk->bp, args->whichfork);
1456 if (error)
1457 return(error);
1458 ASSERT(blk->bp != NULL);
1459 info = blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001460 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001461 be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC ||
Nathan Scott89da0542006-03-17 17:28:40 +11001462 be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1463 blk->magic = be16_to_cpu(info->magic);
1464 if (blk->magic == XFS_DA_NODE_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 node = (xfs_da_intnode_t *)info;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001466 blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 if (forward)
1468 blk->index = 0;
1469 else
Nathan Scottfac80cc2006-03-17 17:29:56 +11001470 blk->index = be16_to_cpu(node->hdr.count)-1;
Nathan Scott403432d2006-03-17 17:29:46 +11001471 blkno = be32_to_cpu(node->btree[blk->index].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 } else {
1473 ASSERT(level == path->active-1);
1474 blk->index = 0;
1475 switch(blk->magic) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 case XFS_ATTR_LEAF_MAGIC:
1477 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1478 NULL);
1479 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1482 NULL);
1483 break;
1484 default:
1485 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001486 blk->magic == XFS_DIR2_LEAFN_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 break;
1488 }
1489 }
1490 }
1491 *result = 0;
1492 return(0);
1493}
1494
1495
1496/*========================================================================
1497 * Utility routines.
1498 *========================================================================*/
1499
1500/*
1501 * Implement a simple hash on a character string.
1502 * Rotate the hash value by 7 bits, then XOR each character in.
1503 * This is implemented with some source-level loop unrolling.
1504 */
1505xfs_dahash_t
Nathan Scott80cce772005-11-02 11:43:04 +11001506xfs_da_hashname(const uchar_t *name, int namelen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507{
1508 xfs_dahash_t hash;
1509
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 /*
1511 * Do four characters at a time as long as we can.
1512 */
1513 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1514 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1515 (name[3] << 0) ^ rol32(hash, 7 * 4);
1516
1517 /*
1518 * Now do the rest of the characters.
1519 */
1520 switch (namelen) {
1521 case 3:
1522 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1523 rol32(hash, 7 * 3);
1524 case 2:
1525 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1526 case 1:
1527 return (name[0] << 0) ^ rol32(hash, 7 * 1);
Nathan Scott2b3b6d02005-11-02 15:12:28 +11001528 default: /* case 0: */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 return hash;
1530 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531}
1532
Barry Naujok5163f952008-05-21 16:41:01 +10001533enum xfs_dacmp
1534xfs_da_compname(
1535 struct xfs_da_args *args,
1536 const char *name,
1537 int len)
1538{
1539 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1540 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1541}
1542
1543static xfs_dahash_t
1544xfs_default_hashname(
1545 struct xfs_name *name)
1546{
1547 return xfs_da_hashname(name->name, name->len);
1548}
1549
1550const struct xfs_nameops xfs_default_nameops = {
1551 .hashname = xfs_default_hashname,
1552 .compname = xfs_da_compname
1553};
1554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555/*
1556 * Add a block to the btree ahead of the file.
1557 * Return the new block number to the caller.
1558 */
1559int
1560xfs_da_grow_inode(xfs_da_args_t *args, xfs_dablk_t *new_blkno)
1561{
1562 xfs_fileoff_t bno, b;
1563 xfs_bmbt_irec_t map;
1564 xfs_bmbt_irec_t *mapp;
1565 xfs_inode_t *dp;
1566 int nmap, error, w, count, c, got, i, mapi;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 xfs_trans_t *tp;
1568 xfs_mount_t *mp;
1569
1570 dp = args->dp;
1571 mp = dp->i_mount;
1572 w = args->whichfork;
1573 tp = args->trans;
1574 /*
1575 * For new directories adjust the file offset and block count.
1576 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001577 if (w == XFS_DATA_FORK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 bno = mp->m_dirleafblk;
1579 count = mp->m_dirblkfsbs;
1580 } else {
1581 bno = 0;
1582 count = 1;
1583 }
1584 /*
1585 * Find a spot in the file space to put the new block.
1586 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001587 if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, w)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 return error;
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001589 if (w == XFS_DATA_FORK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 ASSERT(bno >= mp->m_dirleafblk && bno < mp->m_dirfreeblk);
1591 /*
1592 * Try mapping it in one filesystem block.
1593 */
1594 nmap = 1;
1595 ASSERT(args->firstblock != NULL);
1596 if ((error = xfs_bmapi(tp, dp, bno, count,
1597 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|
1598 XFS_BMAPI_CONTIG,
1599 args->firstblock, args->total, &map, &nmap,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001600 args->flist, NULL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 return error;
1602 }
1603 ASSERT(nmap <= 1);
1604 if (nmap == 1) {
1605 mapp = &map;
1606 mapi = 1;
1607 }
1608 /*
1609 * If we didn't get it and the block might work if fragmented,
1610 * try without the CONTIG flag. Loop until we get it all.
1611 */
1612 else if (nmap == 0 && count > 1) {
1613 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1614 for (b = bno, mapi = 0; b < bno + count; ) {
1615 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1616 c = (int)(bno + count - b);
1617 if ((error = xfs_bmapi(tp, dp, b, c,
1618 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_WRITE|
1619 XFS_BMAPI_METADATA,
1620 args->firstblock, args->total,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001621 &mapp[mapi], &nmap, args->flist,
1622 NULL))) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001623 kmem_free(mapp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 return error;
1625 }
1626 if (nmap < 1)
1627 break;
1628 mapi += nmap;
1629 b = mapp[mapi - 1].br_startoff +
1630 mapp[mapi - 1].br_blockcount;
1631 }
1632 } else {
1633 mapi = 0;
1634 mapp = NULL;
1635 }
1636 /*
1637 * Count the blocks we got, make sure it matches the total.
1638 */
1639 for (i = 0, got = 0; i < mapi; i++)
1640 got += mapp[i].br_blockcount;
1641 if (got != count || mapp[0].br_startoff != bno ||
1642 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1643 bno + count) {
1644 if (mapp != &map)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001645 kmem_free(mapp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 return XFS_ERROR(ENOSPC);
1647 }
1648 if (mapp != &map)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001649 kmem_free(mapp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 *new_blkno = (xfs_dablk_t)bno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 return 0;
1652}
1653
1654/*
1655 * Ick. We need to always be able to remove a btree block, even
1656 * if there's no space reservation because the filesystem is full.
1657 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1658 * It swaps the target block with the last block in the file. The
1659 * last block in the file can always be removed since it can't cause
1660 * a bmap btree split to do that.
1661 */
1662STATIC int
1663xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1664 xfs_dabuf_t **dead_bufp)
1665{
1666 xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1667 xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1668 xfs_fileoff_t lastoff;
1669 xfs_inode_t *ip;
1670 xfs_trans_t *tp;
1671 xfs_mount_t *mp;
1672 int error, w, entno, level, dead_level;
1673 xfs_da_blkinfo_t *dead_info, *sib_info;
1674 xfs_da_intnode_t *par_node, *dead_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 xfs_dir2_leaf_t *dead_leaf2;
1676 xfs_dahash_t dead_hash;
1677
1678 dead_buf = *dead_bufp;
1679 dead_blkno = *dead_blknop;
1680 tp = args->trans;
1681 ip = args->dp;
1682 w = args->whichfork;
1683 ASSERT(w == XFS_DATA_FORK);
1684 mp = ip->i_mount;
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001685 lastoff = mp->m_dirfreeblk;
1686 error = xfs_bmap_last_before(tp, ip, &lastoff, w);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 if (error)
1688 return error;
1689 if (unlikely(lastoff == 0)) {
1690 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1691 mp);
1692 return XFS_ERROR(EFSCORRUPTED);
1693 }
1694 /*
1695 * Read the last block in the btree space.
1696 */
1697 last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1698 if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1699 return error;
1700 /*
1701 * Copy the last block into the dead buffer and log it.
1702 */
1703 memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1704 xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1705 dead_info = dead_buf->data;
1706 /*
1707 * Get values from the moved block.
1708 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001709 if (be16_to_cpu(dead_info->magic) == XFS_DIR2_LEAFN_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1711 dead_level = 0;
Nathan Scott3c1f9c12006-03-17 17:28:18 +11001712 dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 } else {
Nathan Scott89da0542006-03-17 17:28:40 +11001714 ASSERT(be16_to_cpu(dead_info->magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 dead_node = (xfs_da_intnode_t *)dead_info;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001716 dead_level = be16_to_cpu(dead_node->hdr.level);
1717 dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
1719 sib_buf = par_buf = NULL;
1720 /*
1721 * If the moved block has a left sibling, fix up the pointers.
1722 */
Nathan Scott89da0542006-03-17 17:28:40 +11001723 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1725 goto done;
1726 sib_info = sib_buf->data;
1727 if (unlikely(
Nathan Scott89da0542006-03-17 17:28:40 +11001728 be32_to_cpu(sib_info->forw) != last_blkno ||
1729 sib_info->magic != dead_info->magic)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1731 XFS_ERRLEVEL_LOW, mp);
1732 error = XFS_ERROR(EFSCORRUPTED);
1733 goto done;
1734 }
Nathan Scott89da0542006-03-17 17:28:40 +11001735 sib_info->forw = cpu_to_be32(dead_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 xfs_da_log_buf(tp, sib_buf,
1737 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1738 sizeof(sib_info->forw)));
1739 xfs_da_buf_done(sib_buf);
1740 sib_buf = NULL;
1741 }
1742 /*
1743 * If the moved block has a right sibling, fix up the pointers.
1744 */
Nathan Scott89da0542006-03-17 17:28:40 +11001745 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1747 goto done;
1748 sib_info = sib_buf->data;
1749 if (unlikely(
Nathan Scott89da0542006-03-17 17:28:40 +11001750 be32_to_cpu(sib_info->back) != last_blkno ||
1751 sib_info->magic != dead_info->magic)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1753 XFS_ERRLEVEL_LOW, mp);
1754 error = XFS_ERROR(EFSCORRUPTED);
1755 goto done;
1756 }
Nathan Scott89da0542006-03-17 17:28:40 +11001757 sib_info->back = cpu_to_be32(dead_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 xfs_da_log_buf(tp, sib_buf,
1759 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1760 sizeof(sib_info->back)));
1761 xfs_da_buf_done(sib_buf);
1762 sib_buf = NULL;
1763 }
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001764 par_blkno = mp->m_dirleafblk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 level = -1;
1766 /*
1767 * Walk down the tree looking for the parent of the moved block.
1768 */
1769 for (;;) {
1770 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1771 goto done;
1772 par_node = par_buf->data;
1773 if (unlikely(
Nathan Scott89da0542006-03-17 17:28:40 +11001774 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC ||
Nathan Scottfac80cc2006-03-17 17:29:56 +11001775 (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1777 XFS_ERRLEVEL_LOW, mp);
1778 error = XFS_ERROR(EFSCORRUPTED);
1779 goto done;
1780 }
Nathan Scottfac80cc2006-03-17 17:29:56 +11001781 level = be16_to_cpu(par_node->hdr.level);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 for (entno = 0;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001783 entno < be16_to_cpu(par_node->hdr.count) &&
Nathan Scott403432d2006-03-17 17:29:46 +11001784 be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 entno++)
1786 continue;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001787 if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1789 XFS_ERRLEVEL_LOW, mp);
1790 error = XFS_ERROR(EFSCORRUPTED);
1791 goto done;
1792 }
Nathan Scott403432d2006-03-17 17:29:46 +11001793 par_blkno = be32_to_cpu(par_node->btree[entno].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 if (level == dead_level + 1)
1795 break;
1796 xfs_da_brelse(tp, par_buf);
1797 par_buf = NULL;
1798 }
1799 /*
1800 * We're in the right parent block.
1801 * Look for the right entry.
1802 */
1803 for (;;) {
1804 for (;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001805 entno < be16_to_cpu(par_node->hdr.count) &&
Nathan Scott403432d2006-03-17 17:29:46 +11001806 be32_to_cpu(par_node->btree[entno].before) != last_blkno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 entno++)
1808 continue;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001809 if (entno < be16_to_cpu(par_node->hdr.count))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 break;
Nathan Scott89da0542006-03-17 17:28:40 +11001811 par_blkno = be32_to_cpu(par_node->hdr.info.forw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 xfs_da_brelse(tp, par_buf);
1813 par_buf = NULL;
1814 if (unlikely(par_blkno == 0)) {
1815 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1816 XFS_ERRLEVEL_LOW, mp);
1817 error = XFS_ERROR(EFSCORRUPTED);
1818 goto done;
1819 }
1820 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1821 goto done;
1822 par_node = par_buf->data;
1823 if (unlikely(
Nathan Scottfac80cc2006-03-17 17:29:56 +11001824 be16_to_cpu(par_node->hdr.level) != level ||
Nathan Scott89da0542006-03-17 17:28:40 +11001825 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1827 XFS_ERRLEVEL_LOW, mp);
1828 error = XFS_ERROR(EFSCORRUPTED);
1829 goto done;
1830 }
1831 entno = 0;
1832 }
1833 /*
1834 * Update the parent entry pointing to the moved block.
1835 */
Nathan Scott403432d2006-03-17 17:29:46 +11001836 par_node->btree[entno].before = cpu_to_be32(dead_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 xfs_da_log_buf(tp, par_buf,
1838 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1839 sizeof(par_node->btree[entno].before)));
1840 xfs_da_buf_done(par_buf);
1841 xfs_da_buf_done(dead_buf);
1842 *dead_blknop = last_blkno;
1843 *dead_bufp = last_buf;
1844 return 0;
1845done:
1846 if (par_buf)
1847 xfs_da_brelse(tp, par_buf);
1848 if (sib_buf)
1849 xfs_da_brelse(tp, sib_buf);
1850 xfs_da_brelse(tp, last_buf);
1851 return error;
1852}
1853
1854/*
1855 * Remove a btree block from a directory or attribute.
1856 */
1857int
1858xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1859 xfs_dabuf_t *dead_buf)
1860{
1861 xfs_inode_t *dp;
1862 int done, error, w, count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 xfs_trans_t *tp;
1864 xfs_mount_t *mp;
1865
1866 dp = args->dp;
1867 w = args->whichfork;
1868 tp = args->trans;
1869 mp = dp->i_mount;
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001870 if (w == XFS_DATA_FORK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 count = mp->m_dirblkfsbs;
1872 else
1873 count = 1;
1874 for (;;) {
1875 /*
1876 * Remove extents. If we get ENOSPC for a dir we have to move
1877 * the last block to the place we want to kill.
1878 */
1879 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
1880 XFS_BMAPI_AFLAG(w)|XFS_BMAPI_METADATA,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001881 0, args->firstblock, args->flist, NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 &done)) == ENOSPC) {
1883 if (w != XFS_DATA_FORK)
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001884 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1886 &dead_buf)))
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001887 break;
1888 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 }
1891 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 xfs_da_binval(tp, dead_buf);
1893 return error;
1894}
1895
1896/*
1897 * See if the mapping(s) for this btree block are valid, i.e.
1898 * don't contain holes, are logically contiguous, and cover the whole range.
1899 */
1900STATIC int
1901xfs_da_map_covers_blocks(
1902 int nmap,
1903 xfs_bmbt_irec_t *mapp,
1904 xfs_dablk_t bno,
1905 int count)
1906{
1907 int i;
1908 xfs_fileoff_t off;
1909
1910 for (i = 0, off = bno; i < nmap; i++) {
1911 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
1912 mapp[i].br_startblock == DELAYSTARTBLOCK) {
1913 return 0;
1914 }
1915 if (off != mapp[i].br_startoff) {
1916 return 0;
1917 }
1918 off += mapp[i].br_blockcount;
1919 }
1920 return off == bno + count;
1921}
1922
1923/*
1924 * Make a dabuf.
1925 * Used for get_buf, read_buf, read_bufr, and reada_buf.
1926 */
1927STATIC int
1928xfs_da_do_buf(
1929 xfs_trans_t *trans,
1930 xfs_inode_t *dp,
1931 xfs_dablk_t bno,
1932 xfs_daddr_t *mappedbnop,
1933 xfs_dabuf_t **bpp,
1934 int whichfork,
1935 int caller,
1936 inst_t *ra)
1937{
1938 xfs_buf_t *bp = NULL;
1939 xfs_buf_t **bplist;
1940 int error=0;
1941 int i;
1942 xfs_bmbt_irec_t map;
1943 xfs_bmbt_irec_t *mapp;
1944 xfs_daddr_t mappedbno;
1945 xfs_mount_t *mp;
1946 int nbplist=0;
1947 int nfsb;
1948 int nmap;
1949 xfs_dabuf_t *rbp;
1950
1951 mp = dp->i_mount;
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001952 nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 mappedbno = *mappedbnop;
1954 /*
1955 * Caller doesn't have a mapping. -2 means don't complain
1956 * if we land in a hole.
1957 */
1958 if (mappedbno == -1 || mappedbno == -2) {
1959 /*
1960 * Optimize the one-block case.
1961 */
1962 if (nfsb == 1) {
1963 xfs_fsblock_t fsb;
1964
1965 if ((error =
1966 xfs_bmapi_single(trans, dp, whichfork, &fsb,
1967 (xfs_fileoff_t)bno))) {
1968 return error;
1969 }
1970 mapp = &map;
1971 if (fsb == NULLFSBLOCK) {
1972 nmap = 0;
1973 } else {
1974 map.br_startblock = fsb;
1975 map.br_startoff = (xfs_fileoff_t)bno;
1976 map.br_blockcount = 1;
1977 nmap = 1;
1978 }
1979 } else {
1980 mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
1981 nmap = nfsb;
1982 if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno,
1983 nfsb,
1984 XFS_BMAPI_METADATA |
1985 XFS_BMAPI_AFLAG(whichfork),
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001986 NULL, 0, mapp, &nmap, NULL, NULL)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 goto exit0;
1988 }
1989 } else {
1990 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
1991 map.br_startoff = (xfs_fileoff_t)bno;
1992 map.br_blockcount = nfsb;
1993 mapp = &map;
1994 nmap = 1;
1995 }
1996 if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
1997 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
1998 if (unlikely(error == EFSCORRUPTED)) {
1999 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 cmn_err(CE_ALERT, "xfs_da_do_buf: bno %lld\n",
2001 (long long)bno);
2002 cmn_err(CE_ALERT, "dir: inode %lld\n",
2003 (long long)dp->i_ino);
2004 for (i = 0; i < nmap; i++) {
2005 cmn_err(CE_ALERT,
2006 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d\n",
2007 i,
2008 (long long)mapp[i].br_startoff,
2009 (long long)mapp[i].br_startblock,
2010 (long long)mapp[i].br_blockcount,
2011 mapp[i].br_state);
2012 }
2013 }
2014 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2015 XFS_ERRLEVEL_LOW, mp);
2016 }
2017 goto exit0;
2018 }
2019 if (caller != 3 && nmap > 1) {
2020 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
2021 nbplist = 0;
2022 } else
2023 bplist = NULL;
2024 /*
2025 * Turn the mapping(s) into buffer(s).
2026 */
2027 for (i = 0; i < nmap; i++) {
2028 int nmapped;
2029
2030 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2031 if (i == 0)
2032 *mappedbnop = mappedbno;
2033 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2034 switch (caller) {
2035 case 0:
2036 bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2037 mappedbno, nmapped, 0);
2038 error = bp ? XFS_BUF_GETERROR(bp) : XFS_ERROR(EIO);
2039 break;
2040 case 1:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 case 2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 bp = NULL;
2043 error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2044 mappedbno, nmapped, 0, &bp);
2045 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 case 3:
2047 xfs_baread(mp->m_ddev_targp, mappedbno, nmapped);
2048 error = 0;
2049 bp = NULL;
2050 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 }
2052 if (error) {
2053 if (bp)
2054 xfs_trans_brelse(trans, bp);
2055 goto exit1;
2056 }
2057 if (!bp)
2058 continue;
2059 if (caller == 1) {
2060 if (whichfork == XFS_ATTR_FORK) {
2061 XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE,
2062 XFS_ATTR_BTREE_REF);
2063 } else {
2064 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE,
2065 XFS_DIR_BTREE_REF);
2066 }
2067 }
2068 if (bplist) {
2069 bplist[nbplist++] = bp;
2070 }
2071 }
2072 /*
2073 * Build a dabuf structure.
2074 */
2075 if (bplist) {
2076 rbp = xfs_da_buf_make(nbplist, bplist, ra);
2077 } else if (bp)
2078 rbp = xfs_da_buf_make(1, &bp, ra);
2079 else
2080 rbp = NULL;
2081 /*
2082 * For read_buf, check the magic number.
2083 */
2084 if (caller == 1) {
2085 xfs_dir2_data_t *data;
2086 xfs_dir2_free_t *free;
2087 xfs_da_blkinfo_t *info;
2088 uint magic, magic1;
2089
2090 info = rbp->data;
2091 data = rbp->data;
2092 free = rbp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11002093 magic = be16_to_cpu(info->magic);
Nathan Scott70e73f52006-03-17 17:26:52 +11002094 magic1 = be32_to_cpu(data->hdr.magic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 if (unlikely(
2096 XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 (magic != XFS_ATTR_LEAF_MAGIC) &&
2098 (magic != XFS_DIR2_LEAF1_MAGIC) &&
2099 (magic != XFS_DIR2_LEAFN_MAGIC) &&
2100 (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2101 (magic1 != XFS_DIR2_DATA_MAGIC) &&
Nathan Scott0ba962e2006-03-17 17:27:07 +11002102 (be32_to_cpu(free->hdr.magic) != XFS_DIR2_FREE_MAGIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 mp, XFS_ERRTAG_DA_READ_BUF,
2104 XFS_RANDOM_DA_READ_BUF))) {
2105 xfs_buftrace("DA READ ERROR", rbp->bps[0]);
2106 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2107 XFS_ERRLEVEL_LOW, mp, info);
2108 error = XFS_ERROR(EFSCORRUPTED);
2109 xfs_da_brelse(trans, rbp);
2110 nbplist = 0;
2111 goto exit1;
2112 }
2113 }
2114 if (bplist) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002115 kmem_free(bplist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 }
2117 if (mapp != &map) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002118 kmem_free(mapp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 }
2120 if (bpp)
2121 *bpp = rbp;
2122 return 0;
2123exit1:
2124 if (bplist) {
2125 for (i = 0; i < nbplist; i++)
2126 xfs_trans_brelse(trans, bplist[i]);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002127 kmem_free(bplist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 }
2129exit0:
2130 if (mapp != &map)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002131 kmem_free(mapp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 if (bpp)
2133 *bpp = NULL;
2134 return error;
2135}
2136
2137/*
2138 * Get a buffer for the dir/attr block.
2139 */
2140int
2141xfs_da_get_buf(
2142 xfs_trans_t *trans,
2143 xfs_inode_t *dp,
2144 xfs_dablk_t bno,
2145 xfs_daddr_t mappedbno,
2146 xfs_dabuf_t **bpp,
2147 int whichfork)
2148{
2149 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0,
2150 (inst_t *)__return_address);
2151}
2152
2153/*
2154 * Get a buffer for the dir/attr block, fill in the contents.
2155 */
2156int
2157xfs_da_read_buf(
2158 xfs_trans_t *trans,
2159 xfs_inode_t *dp,
2160 xfs_dablk_t bno,
2161 xfs_daddr_t mappedbno,
2162 xfs_dabuf_t **bpp,
2163 int whichfork)
2164{
2165 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1,
2166 (inst_t *)__return_address);
2167}
2168
2169/*
2170 * Readahead the dir/attr block.
2171 */
2172xfs_daddr_t
2173xfs_da_reada_buf(
2174 xfs_trans_t *trans,
2175 xfs_inode_t *dp,
2176 xfs_dablk_t bno,
2177 int whichfork)
2178{
2179 xfs_daddr_t rval;
2180
2181 rval = -1;
2182 if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3,
2183 (inst_t *)__return_address))
2184 return -1;
2185 else
2186 return rval;
2187}
2188
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2190kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */
2191
2192/*
2193 * Allocate a dir-state structure.
2194 * We don't put them on the stack since they're large.
2195 */
2196xfs_da_state_t *
2197xfs_da_state_alloc(void)
2198{
2199 return kmem_zone_zalloc(xfs_da_state_zone, KM_SLEEP);
2200}
2201
2202/*
2203 * Kill the altpath contents of a da-state structure.
2204 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002205STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206xfs_da_state_kill_altpath(xfs_da_state_t *state)
2207{
2208 int i;
2209
2210 for (i = 0; i < state->altpath.active; i++) {
2211 if (state->altpath.blk[i].bp) {
2212 if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2213 xfs_da_buf_done(state->altpath.blk[i].bp);
2214 state->altpath.blk[i].bp = NULL;
2215 }
2216 }
2217 state->altpath.active = 0;
2218}
2219
2220/*
2221 * Free a da-state structure.
2222 */
2223void
2224xfs_da_state_free(xfs_da_state_t *state)
2225{
2226 int i;
2227
2228 xfs_da_state_kill_altpath(state);
2229 for (i = 0; i < state->path.active; i++) {
2230 if (state->path.blk[i].bp)
2231 xfs_da_buf_done(state->path.blk[i].bp);
2232 }
2233 if (state->extravalid && state->extrablk.bp)
2234 xfs_da_buf_done(state->extrablk.bp);
2235#ifdef DEBUG
2236 memset((char *)state, 0, sizeof(*state));
2237#endif /* DEBUG */
2238 kmem_zone_free(xfs_da_state_zone, state);
2239}
2240
2241#ifdef XFS_DABUF_DEBUG
2242xfs_dabuf_t *xfs_dabuf_global_list;
Eric Sandeen703e1f02007-10-11 17:41:21 +10002243spinlock_t xfs_dabuf_global_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244#endif
2245
2246/*
2247 * Create a dabuf.
2248 */
2249/* ARGSUSED */
2250STATIC xfs_dabuf_t *
2251xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra)
2252{
2253 xfs_buf_t *bp;
2254 xfs_dabuf_t *dabuf;
2255 int i;
2256 int off;
2257
2258 if (nbuf == 1)
2259 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_SLEEP);
2260 else
2261 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_SLEEP);
2262 dabuf->dirty = 0;
2263#ifdef XFS_DABUF_DEBUG
2264 dabuf->ra = ra;
2265 dabuf->target = XFS_BUF_TARGET(bps[0]);
2266 dabuf->blkno = XFS_BUF_ADDR(bps[0]);
2267#endif
2268 if (nbuf == 1) {
2269 dabuf->nbuf = 1;
2270 bp = bps[0];
2271 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2272 dabuf->data = XFS_BUF_PTR(bp);
2273 dabuf->bps[0] = bp;
2274 } else {
2275 dabuf->nbuf = nbuf;
2276 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2277 dabuf->bps[i] = bp = bps[i];
2278 dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2279 }
2280 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2281 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2282 bp = bps[i];
2283 memcpy((char *)dabuf->data + off, XFS_BUF_PTR(bp),
2284 XFS_BUF_COUNT(bp));
2285 }
2286 }
2287#ifdef XFS_DABUF_DEBUG
2288 {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 xfs_dabuf_t *p;
2290
Eric Sandeen703e1f02007-10-11 17:41:21 +10002291 spin_lock(&xfs_dabuf_global_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 for (p = xfs_dabuf_global_list; p; p = p->next) {
2293 ASSERT(p->blkno != dabuf->blkno ||
2294 p->target != dabuf->target);
2295 }
2296 dabuf->prev = NULL;
2297 if (xfs_dabuf_global_list)
2298 xfs_dabuf_global_list->prev = dabuf;
2299 dabuf->next = xfs_dabuf_global_list;
2300 xfs_dabuf_global_list = dabuf;
Eric Sandeen703e1f02007-10-11 17:41:21 +10002301 spin_unlock(&xfs_dabuf_global_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 }
2303#endif
2304 return dabuf;
2305}
2306
2307/*
2308 * Un-dirty a dabuf.
2309 */
2310STATIC void
2311xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2312{
2313 xfs_buf_t *bp;
2314 int i;
2315 int off;
2316
2317 if (dabuf->dirty) {
2318 ASSERT(dabuf->nbuf > 1);
2319 dabuf->dirty = 0;
2320 for (i = off = 0; i < dabuf->nbuf;
2321 i++, off += XFS_BUF_COUNT(bp)) {
2322 bp = dabuf->bps[i];
2323 memcpy(XFS_BUF_PTR(bp), (char *)dabuf->data + off,
2324 XFS_BUF_COUNT(bp));
2325 }
2326 }
2327}
2328
2329/*
2330 * Release a dabuf.
2331 */
2332void
2333xfs_da_buf_done(xfs_dabuf_t *dabuf)
2334{
2335 ASSERT(dabuf);
2336 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2337 if (dabuf->dirty)
2338 xfs_da_buf_clean(dabuf);
2339 if (dabuf->nbuf > 1)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002340 kmem_free(dabuf->data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341#ifdef XFS_DABUF_DEBUG
2342 {
Eric Sandeen703e1f02007-10-11 17:41:21 +10002343 spin_lock(&xfs_dabuf_global_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 if (dabuf->prev)
2345 dabuf->prev->next = dabuf->next;
2346 else
2347 xfs_dabuf_global_list = dabuf->next;
2348 if (dabuf->next)
2349 dabuf->next->prev = dabuf->prev;
Eric Sandeen703e1f02007-10-11 17:41:21 +10002350 spin_unlock(&xfs_dabuf_global_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 }
2352 memset(dabuf, 0, XFS_DA_BUF_SIZE(dabuf->nbuf));
2353#endif
2354 if (dabuf->nbuf == 1)
2355 kmem_zone_free(xfs_dabuf_zone, dabuf);
2356 else
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002357 kmem_free(dabuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358}
2359
2360/*
2361 * Log transaction from a dabuf.
2362 */
2363void
2364xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2365{
2366 xfs_buf_t *bp;
2367 uint f;
2368 int i;
2369 uint l;
2370 int off;
2371
2372 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2373 if (dabuf->nbuf == 1) {
2374 ASSERT(dabuf->data == (void *)XFS_BUF_PTR(dabuf->bps[0]));
2375 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2376 return;
2377 }
2378 dabuf->dirty = 1;
2379 ASSERT(first <= last);
2380 for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2381 bp = dabuf->bps[i];
2382 f = off;
2383 l = f + XFS_BUF_COUNT(bp) - 1;
2384 if (f < first)
2385 f = first;
2386 if (l > last)
2387 l = last;
2388 if (f <= l)
2389 xfs_trans_log_buf(tp, bp, f - off, l - off);
2390 /*
2391 * B_DONE is set by xfs_trans_log buf.
2392 * If we don't set it on a new buffer (get not read)
2393 * then if we don't put anything in the buffer it won't
2394 * be set, and at commit it it released into the cache,
2395 * and then a read will fail.
2396 */
2397 else if (!(XFS_BUF_ISDONE(bp)))
2398 XFS_BUF_DONE(bp);
2399 }
2400 ASSERT(last < off);
2401}
2402
2403/*
2404 * Release dabuf from a transaction.
2405 * Have to free up the dabuf before the buffers are released,
2406 * since the synchronization on the dabuf is really the lock on the buffer.
2407 */
2408void
2409xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2410{
2411 xfs_buf_t *bp;
2412 xfs_buf_t **bplist;
2413 int i;
2414 int nbuf;
2415
2416 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2417 if ((nbuf = dabuf->nbuf) == 1) {
2418 bplist = &bp;
2419 bp = dabuf->bps[0];
2420 } else {
2421 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2422 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2423 }
2424 xfs_da_buf_done(dabuf);
2425 for (i = 0; i < nbuf; i++)
2426 xfs_trans_brelse(tp, bplist[i]);
2427 if (bplist != &bp)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002428 kmem_free(bplist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429}
2430
2431/*
2432 * Invalidate dabuf from a transaction.
2433 */
2434void
2435xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2436{
2437 xfs_buf_t *bp;
2438 xfs_buf_t **bplist;
2439 int i;
2440 int nbuf;
2441
2442 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2443 if ((nbuf = dabuf->nbuf) == 1) {
2444 bplist = &bp;
2445 bp = dabuf->bps[0];
2446 } else {
2447 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2448 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2449 }
2450 xfs_da_buf_done(dabuf);
2451 for (i = 0; i < nbuf; i++)
2452 xfs_trans_binval(tp, bplist[i]);
2453 if (bplist != &bp)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002454 kmem_free(bplist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455}
2456
2457/*
2458 * Get the first daddr from a dabuf.
2459 */
2460xfs_daddr_t
2461xfs_da_blkno(xfs_dabuf_t *dabuf)
2462{
2463 ASSERT(dabuf->nbuf);
2464 ASSERT(dabuf->data);
2465 return XFS_BUF_ADDR(dabuf->bps[0]);
2466}