blob: 1c00bedb3175c1f5e58d006b0cdf0c8089bc0df5 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110029#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_dir2_sf.h"
32#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_inode_item.h"
35#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_attr.h"
38#include "xfs_attr_leaf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_dir2_data.h"
40#include "xfs_dir2_leaf.h"
41#include "xfs_dir2_block.h"
42#include "xfs_dir2_node.h"
43#include "xfs_error.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000044#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46/*
47 * xfs_da_btree.c
48 *
49 * Routines to implement directories as Btrees of hashed names.
50 */
51
52/*========================================================================
53 * Function prototypes for the kernel.
54 *========================================================================*/
55
56/*
57 * Routines used for growing the Btree.
58 */
59STATIC int xfs_da_root_split(xfs_da_state_t *state,
60 xfs_da_state_blk_t *existing_root,
61 xfs_da_state_blk_t *new_child);
62STATIC int xfs_da_node_split(xfs_da_state_t *state,
63 xfs_da_state_blk_t *existing_blk,
64 xfs_da_state_blk_t *split_blk,
65 xfs_da_state_blk_t *blk_to_add,
66 int treelevel,
67 int *result);
68STATIC void xfs_da_node_rebalance(xfs_da_state_t *state,
69 xfs_da_state_blk_t *node_blk_1,
70 xfs_da_state_blk_t *node_blk_2);
71STATIC void xfs_da_node_add(xfs_da_state_t *state,
72 xfs_da_state_blk_t *old_node_blk,
73 xfs_da_state_blk_t *new_node_blk);
74
75/*
76 * Routines used for shrinking the Btree.
77 */
78STATIC int xfs_da_root_join(xfs_da_state_t *state,
79 xfs_da_state_blk_t *root_blk);
80STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval);
81STATIC void xfs_da_node_remove(xfs_da_state_t *state,
82 xfs_da_state_blk_t *drop_blk);
83STATIC void xfs_da_node_unbalance(xfs_da_state_t *state,
84 xfs_da_state_blk_t *src_node_blk,
85 xfs_da_state_blk_t *dst_node_blk);
86
87/*
88 * Utility routines.
89 */
90STATIC uint xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count);
91STATIC int xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp);
92STATIC xfs_dabuf_t *xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra);
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100093STATIC int xfs_da_blk_unlink(xfs_da_state_t *state,
94 xfs_da_state_blk_t *drop_blk,
95 xfs_da_state_blk_t *save_blk);
96STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
98/*========================================================================
99 * Routines used for growing the Btree.
100 *========================================================================*/
101
102/*
103 * Create the initial contents of an intermediate node.
104 */
105int
106xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level,
107 xfs_dabuf_t **bpp, int whichfork)
108{
109 xfs_da_intnode_t *node;
110 xfs_dabuf_t *bp;
111 int error;
112 xfs_trans_t *tp;
113
114 tp = args->trans;
115 error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork);
116 if (error)
117 return(error);
118 ASSERT(bp != NULL);
119 node = bp->data;
120 node->hdr.info.forw = 0;
121 node->hdr.info.back = 0;
Nathan Scott89da0542006-03-17 17:28:40 +1100122 node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 node->hdr.info.pad = 0;
124 node->hdr.count = 0;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100125 node->hdr.level = cpu_to_be16(level);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127 xfs_da_log_buf(tp, bp,
128 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
129
130 *bpp = bp;
131 return(0);
132}
133
134/*
135 * Split a leaf node, rebalance, then possibly split
136 * intermediate nodes, rebalance, etc.
137 */
138int /* error */
139xfs_da_split(xfs_da_state_t *state)
140{
141 xfs_da_state_blk_t *oldblk, *newblk, *addblk;
142 xfs_da_intnode_t *node;
143 xfs_dabuf_t *bp;
144 int max, action, error, i;
145
146 /*
147 * Walk back up the tree splitting/inserting/adjusting as necessary.
148 * If we need to insert and there isn't room, split the node, then
149 * decide which fragment to insert the new block from below into.
150 * Note that we may split the root this way, but we need more fixup.
151 */
152 max = state->path.active - 1;
153 ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH));
154 ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000155 state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 addblk = &state->path.blk[max]; /* initial dummy value */
158 for (i = max; (i >= 0) && addblk; state->path.active--, i--) {
159 oldblk = &state->path.blk[i];
160 newblk = &state->altpath.blk[i];
161
162 /*
163 * If a leaf node then
164 * Allocate a new leaf node, then rebalance across them.
165 * else if an intermediate node then
166 * We split on the last layer, must we split the node?
167 */
168 switch (oldblk->magic) {
169 case XFS_ATTR_LEAF_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 error = xfs_attr_leaf_split(state, oldblk, newblk);
171 if ((error != 0) && (error != ENOSPC)) {
172 return(error); /* GROT: attr is inconsistent */
173 }
174 if (!error) {
175 addblk = newblk;
176 break;
177 }
178 /*
179 * Entry wouldn't fit, split the leaf again.
180 */
181 state->extravalid = 1;
182 if (state->inleaf) {
183 state->extraafter = 0; /* before newblk */
184 error = xfs_attr_leaf_split(state, oldblk,
185 &state->extrablk);
186 } else {
187 state->extraafter = 1; /* after newblk */
188 error = xfs_attr_leaf_split(state, newblk,
189 &state->extrablk);
190 }
191 if (error)
192 return(error); /* GROT: attr inconsistent */
193 addblk = newblk;
194 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 error = xfs_dir2_leafn_split(state, oldblk, newblk);
197 if (error)
198 return error;
199 addblk = newblk;
200 break;
201 case XFS_DA_NODE_MAGIC:
202 error = xfs_da_node_split(state, oldblk, newblk, addblk,
203 max - i, &action);
204 xfs_da_buf_done(addblk->bp);
205 addblk->bp = NULL;
206 if (error)
207 return(error); /* GROT: dir is inconsistent */
208 /*
209 * Record the newly split block for the next time thru?
210 */
211 if (action)
212 addblk = newblk;
213 else
214 addblk = NULL;
215 break;
216 }
217
218 /*
219 * Update the btree to show the new hashval for this child.
220 */
221 xfs_da_fixhashpath(state, &state->path);
222 /*
223 * If we won't need this block again, it's getting dropped
224 * from the active path by the loop control, so we need
225 * to mark it done now.
226 */
227 if (i > 0 || !addblk)
228 xfs_da_buf_done(oldblk->bp);
229 }
230 if (!addblk)
231 return(0);
232
233 /*
234 * Split the root node.
235 */
236 ASSERT(state->path.active == 0);
237 oldblk = &state->path.blk[0];
238 error = xfs_da_root_split(state, oldblk, addblk);
239 if (error) {
240 xfs_da_buf_done(oldblk->bp);
241 xfs_da_buf_done(addblk->bp);
242 addblk->bp = NULL;
243 return(error); /* GROT: dir is inconsistent */
244 }
245
246 /*
247 * Update pointers to the node which used to be block 0 and
248 * just got bumped because of the addition of a new root node.
249 * There might be three blocks involved if a double split occurred,
250 * and the original block 0 could be at any position in the list.
251 */
252
253 node = oldblk->bp->data;
254 if (node->hdr.info.forw) {
Nathan Scott89da0542006-03-17 17:28:40 +1100255 if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 bp = addblk->bp;
257 } else {
258 ASSERT(state->extravalid);
259 bp = state->extrablk.bp;
260 }
261 node = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100262 node->hdr.info.back = cpu_to_be32(oldblk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 xfs_da_log_buf(state->args->trans, bp,
264 XFS_DA_LOGRANGE(node, &node->hdr.info,
265 sizeof(node->hdr.info)));
266 }
267 node = oldblk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100268 if (node->hdr.info.back) {
269 if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 bp = addblk->bp;
271 } else {
272 ASSERT(state->extravalid);
273 bp = state->extrablk.bp;
274 }
275 node = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100276 node->hdr.info.forw = cpu_to_be32(oldblk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 xfs_da_log_buf(state->args->trans, bp,
278 XFS_DA_LOGRANGE(node, &node->hdr.info,
279 sizeof(node->hdr.info)));
280 }
281 xfs_da_buf_done(oldblk->bp);
282 xfs_da_buf_done(addblk->bp);
283 addblk->bp = NULL;
284 return(0);
285}
286
287/*
288 * Split the root. We have to create a new root and point to the two
289 * parts (the split old root) that we just created. Copy block zero to
290 * the EOF, extending the inode in process.
291 */
292STATIC int /* error */
293xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
294 xfs_da_state_blk_t *blk2)
295{
296 xfs_da_intnode_t *node, *oldroot;
297 xfs_da_args_t *args;
298 xfs_dablk_t blkno;
299 xfs_dabuf_t *bp;
300 int error, size;
301 xfs_inode_t *dp;
302 xfs_trans_t *tp;
303 xfs_mount_t *mp;
304 xfs_dir2_leaf_t *leaf;
305
306 /*
307 * Copy the existing (incorrect) block from the root node position
308 * to a free space somewhere.
309 */
310 args = state->args;
311 ASSERT(args != NULL);
312 error = xfs_da_grow_inode(args, &blkno);
313 if (error)
314 return(error);
315 dp = args->dp;
316 tp = args->trans;
317 mp = state->mp;
318 error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork);
319 if (error)
320 return(error);
321 ASSERT(bp != NULL);
322 node = bp->data;
323 oldroot = blk1->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100324 if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC) {
Nathan Scottfac80cc2006-03-17 17:29:56 +1100325 size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] -
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 (char *)oldroot);
327 } else {
Nathan Scott89da0542006-03-17 17:28:40 +1100328 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 leaf = (xfs_dir2_leaf_t *)oldroot;
Nathan Scotta818e5d2006-03-17 17:28:07 +1100330 size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] -
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 (char *)leaf);
332 }
333 memcpy(node, oldroot, size);
334 xfs_da_log_buf(tp, bp, 0, size - 1);
335 xfs_da_buf_done(blk1->bp);
336 blk1->bp = bp;
337 blk1->blkno = blkno;
338
339 /*
340 * Set up the new root node.
341 */
342 error = xfs_da_node_create(args,
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000343 (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0,
Nathan Scottfac80cc2006-03-17 17:29:56 +1100344 be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (error)
346 return(error);
347 node = bp->data;
Nathan Scott403432d2006-03-17 17:29:46 +1100348 node->btree[0].hashval = cpu_to_be32(blk1->hashval);
349 node->btree[0].before = cpu_to_be32(blk1->blkno);
350 node->btree[1].hashval = cpu_to_be32(blk2->hashval);
351 node->btree[1].before = cpu_to_be32(blk2->blkno);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100352 node->hdr.count = cpu_to_be16(2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354#ifdef DEBUG
Nathan Scott89da0542006-03-17 17:28:40 +1100355 if (be16_to_cpu(oldroot->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 ASSERT(blk1->blkno >= mp->m_dirleafblk &&
357 blk1->blkno < mp->m_dirfreeblk);
358 ASSERT(blk2->blkno >= mp->m_dirleafblk &&
359 blk2->blkno < mp->m_dirfreeblk);
360 }
361#endif
362
363 /* Header is already logged by xfs_da_node_create */
364 xfs_da_log_buf(tp, bp,
365 XFS_DA_LOGRANGE(node, node->btree,
366 sizeof(xfs_da_node_entry_t) * 2));
367 xfs_da_buf_done(bp);
368
369 return(0);
370}
371
372/*
373 * Split the node, rebalance, then add the new entry.
374 */
375STATIC int /* error */
376xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
377 xfs_da_state_blk_t *newblk,
378 xfs_da_state_blk_t *addblk,
379 int treelevel, int *result)
380{
381 xfs_da_intnode_t *node;
382 xfs_dablk_t blkno;
383 int newcount, error;
384 int useextra;
385
386 node = oldblk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100387 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389 /*
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000390 * With V2 dirs the extra block is data or freespace.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000392 useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 newcount = 1 + useextra;
394 /*
395 * Do we have to split the node?
396 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100397 if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 /*
399 * Allocate a new node, add to the doubly linked chain of
400 * nodes, then move some of our excess entries into it.
401 */
402 error = xfs_da_grow_inode(state->args, &blkno);
403 if (error)
404 return(error); /* GROT: dir is inconsistent */
405
406 error = xfs_da_node_create(state->args, blkno, treelevel,
407 &newblk->bp, state->args->whichfork);
408 if (error)
409 return(error); /* GROT: dir is inconsistent */
410 newblk->blkno = blkno;
411 newblk->magic = XFS_DA_NODE_MAGIC;
412 xfs_da_node_rebalance(state, oldblk, newblk);
413 error = xfs_da_blk_link(state, oldblk, newblk);
414 if (error)
415 return(error);
416 *result = 1;
417 } else {
418 *result = 0;
419 }
420
421 /*
422 * Insert the new entry(s) into the correct block
423 * (updating last hashval in the process).
424 *
425 * xfs_da_node_add() inserts BEFORE the given index,
426 * and as a result of using node_lookup_int() we always
427 * point to a valid entry (not after one), but a split
428 * operation always results in a new block whose hashvals
429 * FOLLOW the current block.
430 *
431 * If we had double-split op below us, then add the extra block too.
432 */
433 node = oldblk->bp->data;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100434 if (oldblk->index <= be16_to_cpu(node->hdr.count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 oldblk->index++;
436 xfs_da_node_add(state, oldblk, addblk);
437 if (useextra) {
438 if (state->extraafter)
439 oldblk->index++;
440 xfs_da_node_add(state, oldblk, &state->extrablk);
441 state->extravalid = 0;
442 }
443 } else {
444 newblk->index++;
445 xfs_da_node_add(state, newblk, addblk);
446 if (useextra) {
447 if (state->extraafter)
448 newblk->index++;
449 xfs_da_node_add(state, newblk, &state->extrablk);
450 state->extravalid = 0;
451 }
452 }
453
454 return(0);
455}
456
457/*
458 * Balance the btree elements between two intermediate nodes,
459 * usually one full and one empty.
460 *
461 * NOTE: if blk2 is empty, then it will get the upper half of blk1.
462 */
463STATIC void
464xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
465 xfs_da_state_blk_t *blk2)
466{
467 xfs_da_intnode_t *node1, *node2, *tmpnode;
468 xfs_da_node_entry_t *btree_s, *btree_d;
469 int count, tmp;
470 xfs_trans_t *tp;
471
472 node1 = blk1->bp->data;
473 node2 = blk2->bp->data;
474 /*
475 * Figure out how many entries need to move, and in which direction.
476 * Swap the nodes around if that makes it simpler.
477 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100478 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
Nathan Scott403432d2006-03-17 17:29:46 +1100479 ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) ||
Nathan Scottfac80cc2006-03-17 17:29:56 +1100480 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
481 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 tmpnode = node1;
483 node1 = node2;
484 node2 = tmpnode;
485 }
Nathan Scott89da0542006-03-17 17:28:40 +1100486 ASSERT(be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC);
487 ASSERT(be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100488 count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if (count == 0)
490 return;
491 tp = state->args->trans;
492 /*
493 * Two cases: high-to-low and low-to-high.
494 */
495 if (count > 0) {
496 /*
497 * Move elements in node2 up to make a hole.
498 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100499 if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 tmp *= (uint)sizeof(xfs_da_node_entry_t);
501 btree_s = &node2->btree[0];
502 btree_d = &node2->btree[count];
503 memmove(btree_d, btree_s, tmp);
504 }
505
506 /*
507 * Move the req'd B-tree elements from high in node1 to
508 * low in node2.
509 */
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800510 be16_add_cpu(&node2->hdr.count, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100512 btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 btree_d = &node2->btree[0];
514 memcpy(btree_d, btree_s, tmp);
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800515 be16_add_cpu(&node1->hdr.count, -count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 } else {
517 /*
518 * Move the req'd B-tree elements from low in node2 to
519 * high in node1.
520 */
521 count = -count;
522 tmp = count * (uint)sizeof(xfs_da_node_entry_t);
523 btree_s = &node2->btree[0];
Nathan Scottfac80cc2006-03-17 17:29:56 +1100524 btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 memcpy(btree_d, btree_s, tmp);
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800526 be16_add_cpu(&node1->hdr.count, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 xfs_da_log_buf(tp, blk1->bp,
528 XFS_DA_LOGRANGE(node1, btree_d, tmp));
529
530 /*
531 * Move elements in node2 down to fill the hole.
532 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100533 tmp = be16_to_cpu(node2->hdr.count) - count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 tmp *= (uint)sizeof(xfs_da_node_entry_t);
535 btree_s = &node2->btree[count];
536 btree_d = &node2->btree[0];
537 memmove(btree_d, btree_s, tmp);
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800538 be16_add_cpu(&node2->hdr.count, -count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
540
541 /*
542 * Log header of node 1 and all current bits of node 2.
543 */
544 xfs_da_log_buf(tp, blk1->bp,
545 XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr)));
546 xfs_da_log_buf(tp, blk2->bp,
547 XFS_DA_LOGRANGE(node2, &node2->hdr,
548 sizeof(node2->hdr) +
Nathan Scottfac80cc2006-03-17 17:29:56 +1100549 sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
551 /*
552 * Record the last hashval from each block for upward propagation.
553 * (note: don't use the swapped node pointers)
554 */
555 node1 = blk1->bp->data;
556 node2 = blk2->bp->data;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100557 blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval);
558 blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
560 /*
561 * Adjust the expected index for insertion.
562 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100563 if (blk1->index >= be16_to_cpu(node1->hdr.count)) {
564 blk2->index = blk1->index - be16_to_cpu(node1->hdr.count);
565 blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 }
567}
568
569/*
570 * Add a new entry to an intermediate node.
571 */
572STATIC void
573xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
574 xfs_da_state_blk_t *newblk)
575{
576 xfs_da_intnode_t *node;
577 xfs_da_node_entry_t *btree;
578 int tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 node = oldblk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100581 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100582 ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 ASSERT(newblk->blkno != 0);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000584 if (state->args->whichfork == XFS_DATA_FORK)
Christoph Hellwig73523a22010-07-20 17:54:45 +1000585 ASSERT(newblk->blkno >= state->mp->m_dirleafblk &&
586 newblk->blkno < state->mp->m_dirfreeblk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 /*
589 * We may need to make some room before we insert the new node.
590 */
591 tmp = 0;
592 btree = &node->btree[ oldblk->index ];
Nathan Scottfac80cc2006-03-17 17:29:56 +1100593 if (oldblk->index < be16_to_cpu(node->hdr.count)) {
594 tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 memmove(btree + 1, btree, tmp);
596 }
Nathan Scott403432d2006-03-17 17:29:46 +1100597 btree->hashval = cpu_to_be32(newblk->hashval);
598 btree->before = cpu_to_be32(newblk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 xfs_da_log_buf(state->args->trans, oldblk->bp,
600 XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree)));
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800601 be16_add_cpu(&node->hdr.count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 xfs_da_log_buf(state->args->trans, oldblk->bp,
603 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
604
605 /*
606 * Copy the last hash value from the oldblk to propagate upwards.
607 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100608 oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
611/*========================================================================
612 * Routines used for shrinking the Btree.
613 *========================================================================*/
614
615/*
616 * Deallocate an empty leaf node, remove it from its parent,
617 * possibly deallocating that block, etc...
618 */
619int
620xfs_da_join(xfs_da_state_t *state)
621{
622 xfs_da_state_blk_t *drop_blk, *save_blk;
623 int action, error;
624
625 action = 0;
626 drop_blk = &state->path.blk[ state->path.active-1 ];
627 save_blk = &state->altpath.blk[ state->path.active-1 ];
628 ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC);
629 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000630 drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
632 /*
633 * Walk back up the tree joining/deallocating as necessary.
634 * When we stop dropping blocks, break out.
635 */
636 for ( ; state->path.active >= 2; drop_blk--, save_blk--,
637 state->path.active--) {
638 /*
639 * See if we can combine the block with a neighbor.
640 * (action == 0) => no options, just leave
641 * (action == 1) => coalesce, then unlink
642 * (action == 2) => block empty, unlink it
643 */
644 switch (drop_blk->magic) {
645 case XFS_ATTR_LEAF_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 error = xfs_attr_leaf_toosmall(state, &action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 if (error)
648 return(error);
649 if (action == 0)
650 return(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 xfs_attr_leaf_unbalance(state, drop_blk, save_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 error = xfs_dir2_leafn_toosmall(state, &action);
655 if (error)
656 return error;
657 if (action == 0)
658 return 0;
659 xfs_dir2_leafn_unbalance(state, drop_blk, save_blk);
660 break;
661 case XFS_DA_NODE_MAGIC:
662 /*
663 * Remove the offending node, fixup hashvals,
664 * check for a toosmall neighbor.
665 */
666 xfs_da_node_remove(state, drop_blk);
667 xfs_da_fixhashpath(state, &state->path);
668 error = xfs_da_node_toosmall(state, &action);
669 if (error)
670 return(error);
671 if (action == 0)
672 return 0;
673 xfs_da_node_unbalance(state, drop_blk, save_blk);
674 break;
675 }
676 xfs_da_fixhashpath(state, &state->altpath);
677 error = xfs_da_blk_unlink(state, drop_blk, save_blk);
678 xfs_da_state_kill_altpath(state);
679 if (error)
680 return(error);
681 error = xfs_da_shrink_inode(state->args, drop_blk->blkno,
682 drop_blk->bp);
683 drop_blk->bp = NULL;
684 if (error)
685 return(error);
686 }
687 /*
688 * We joined all the way to the top. If it turns out that
689 * we only have one entry in the root, make the child block
690 * the new root.
691 */
692 xfs_da_node_remove(state, drop_blk);
693 xfs_da_fixhashpath(state, &state->path);
694 error = xfs_da_root_join(state, &state->path.blk[0]);
695 return(error);
696}
697
698/*
699 * We have only one entry in the root. Copy the only remaining child of
700 * the old root to block 0 as the new root node.
701 */
702STATIC int
703xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk)
704{
705 xfs_da_intnode_t *oldroot;
706 /* REFERENCED */
707 xfs_da_blkinfo_t *blkinfo;
708 xfs_da_args_t *args;
709 xfs_dablk_t child;
710 xfs_dabuf_t *bp;
711 int error;
712
713 args = state->args;
714 ASSERT(args != NULL);
715 ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC);
716 oldroot = root_blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100717 ASSERT(be16_to_cpu(oldroot->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 ASSERT(!oldroot->hdr.info.forw);
719 ASSERT(!oldroot->hdr.info.back);
720
721 /*
722 * If the root has more than one child, then don't do anything.
723 */
Nathan Scottfac80cc2006-03-17 17:29:56 +1100724 if (be16_to_cpu(oldroot->hdr.count) > 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 return(0);
726
727 /*
728 * Read in the (only) child block, then copy those bytes into
729 * the root block's buffer and free the original child block.
730 */
Nathan Scott403432d2006-03-17 17:29:46 +1100731 child = be32_to_cpu(oldroot->btree[0].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 ASSERT(child != 0);
733 error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp,
734 args->whichfork);
735 if (error)
736 return(error);
737 ASSERT(bp != NULL);
738 blkinfo = bp->data;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100739 if (be16_to_cpu(oldroot->hdr.level) == 1) {
Nathan Scottf6c2d1f2006-06-20 13:04:51 +1000740 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DIR2_LEAFN_MAGIC ||
Nathan Scott89da0542006-03-17 17:28:40 +1100741 be16_to_cpu(blkinfo->magic) == XFS_ATTR_LEAF_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 } else {
Nathan Scott89da0542006-03-17 17:28:40 +1100743 ASSERT(be16_to_cpu(blkinfo->magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
745 ASSERT(!blkinfo->forw);
746 ASSERT(!blkinfo->back);
747 memcpy(root_blk->bp->data, bp->data, state->blocksize);
748 xfs_da_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1);
749 error = xfs_da_shrink_inode(args, child, bp);
750 return(error);
751}
752
753/*
754 * Check a node block and its neighbors to see if the block should be
755 * collapsed into one or the other neighbor. Always keep the block
756 * with the smaller block number.
757 * If the current block is over 50% full, don't try to join it, return 0.
758 * If the block is empty, fill in the state structure and return 2.
759 * If it can be collapsed, fill in the state structure and return 1.
760 * If nothing can be done, return 0.
761 */
762STATIC int
763xfs_da_node_toosmall(xfs_da_state_t *state, int *action)
764{
765 xfs_da_intnode_t *node;
766 xfs_da_state_blk_t *blk;
767 xfs_da_blkinfo_t *info;
768 int count, forward, error, retval, i;
769 xfs_dablk_t blkno;
770 xfs_dabuf_t *bp;
771
772 /*
773 * Check for the degenerate case of the block being over 50% full.
774 * If so, it's not worth even looking to see if we might be able
775 * to coalesce with a sibling.
776 */
777 blk = &state->path.blk[ state->path.active-1 ];
778 info = blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100779 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 node = (xfs_da_intnode_t *)info;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100781 count = be16_to_cpu(node->hdr.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 if (count > (state->node_ents >> 1)) {
783 *action = 0; /* blk over 50%, don't try to join */
784 return(0); /* blk over 50%, don't try to join */
785 }
786
787 /*
788 * Check for the degenerate case of the block being empty.
789 * If the block is empty, we'll simply delete it, no need to
Nathan Scottc41564b2006-03-29 08:55:14 +1000790 * coalesce it with a sibling block. We choose (arbitrarily)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 * to merge with the forward block unless it is NULL.
792 */
793 if (count == 0) {
794 /*
795 * Make altpath point to the block we want to keep and
796 * path point to the block we want to drop (this one).
797 */
Nathan Scott89da0542006-03-17 17:28:40 +1100798 forward = (info->forw != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 memcpy(&state->altpath, &state->path, sizeof(state->path));
800 error = xfs_da_path_shift(state, &state->altpath, forward,
801 0, &retval);
802 if (error)
803 return(error);
804 if (retval) {
805 *action = 0;
806 } else {
807 *action = 2;
808 }
809 return(0);
810 }
811
812 /*
813 * Examine each sibling block to see if we can coalesce with
814 * at least 25% free space to spare. We need to figure out
815 * whether to merge with the forward or the backward block.
816 * We prefer coalescing with the lower numbered sibling so as
817 * to shrink a directory over time.
818 */
819 /* start with smaller blk num */
Nathan Scott89da0542006-03-17 17:28:40 +1100820 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 for (i = 0; i < 2; forward = !forward, i++) {
822 if (forward)
Nathan Scott89da0542006-03-17 17:28:40 +1100823 blkno = be32_to_cpu(info->forw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 else
Nathan Scott89da0542006-03-17 17:28:40 +1100825 blkno = be32_to_cpu(info->back);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 if (blkno == 0)
827 continue;
828 error = xfs_da_read_buf(state->args->trans, state->args->dp,
829 blkno, -1, &bp, state->args->whichfork);
830 if (error)
831 return(error);
832 ASSERT(bp != NULL);
833
834 node = (xfs_da_intnode_t *)info;
835 count = state->node_ents;
836 count -= state->node_ents >> 2;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100837 count -= be16_to_cpu(node->hdr.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 node = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100839 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Nathan Scottfac80cc2006-03-17 17:29:56 +1100840 count -= be16_to_cpu(node->hdr.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 xfs_da_brelse(state->args->trans, bp);
842 if (count >= 0)
843 break; /* fits with at least 25% to spare */
844 }
845 if (i >= 2) {
846 *action = 0;
847 return(0);
848 }
849
850 /*
851 * Make altpath point to the block we want to keep (the lower
852 * numbered block) and path point to the block we want to drop.
853 */
854 memcpy(&state->altpath, &state->path, sizeof(state->path));
855 if (blkno < blk->blkno) {
856 error = xfs_da_path_shift(state, &state->altpath, forward,
857 0, &retval);
858 if (error) {
859 return(error);
860 }
861 if (retval) {
862 *action = 0;
863 return(0);
864 }
865 } else {
866 error = xfs_da_path_shift(state, &state->path, forward,
867 0, &retval);
868 if (error) {
869 return(error);
870 }
871 if (retval) {
872 *action = 0;
873 return(0);
874 }
875 }
876 *action = 1;
877 return(0);
878}
879
880/*
881 * Walk back up the tree adjusting hash values as necessary,
882 * when we stop making changes, return.
883 */
884void
885xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path)
886{
887 xfs_da_state_blk_t *blk;
888 xfs_da_intnode_t *node;
889 xfs_da_node_entry_t *btree;
890 xfs_dahash_t lasthash=0;
891 int level, count;
892
893 level = path->active-1;
894 blk = &path->blk[ level ];
895 switch (blk->magic) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 case XFS_ATTR_LEAF_MAGIC:
897 lasthash = xfs_attr_leaf_lasthash(blk->bp, &count);
898 if (count == 0)
899 return;
900 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count);
903 if (count == 0)
904 return;
905 break;
906 case XFS_DA_NODE_MAGIC:
907 lasthash = xfs_da_node_lasthash(blk->bp, &count);
908 if (count == 0)
909 return;
910 break;
911 }
912 for (blk--, level--; level >= 0; blk--, level--) {
913 node = blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100914 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 btree = &node->btree[ blk->index ];
Nathan Scott403432d2006-03-17 17:29:46 +1100916 if (be32_to_cpu(btree->hashval) == lasthash)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 break;
918 blk->hashval = lasthash;
Nathan Scott403432d2006-03-17 17:29:46 +1100919 btree->hashval = cpu_to_be32(lasthash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 xfs_da_log_buf(state->args->trans, blk->bp,
921 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
922
Nathan Scottfac80cc2006-03-17 17:29:56 +1100923 lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 }
925}
926
927/*
928 * Remove an entry from an intermediate node.
929 */
930STATIC void
931xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk)
932{
933 xfs_da_intnode_t *node;
934 xfs_da_node_entry_t *btree;
935 int tmp;
936
937 node = drop_blk->bp->data;
Nathan Scottfac80cc2006-03-17 17:29:56 +1100938 ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 ASSERT(drop_blk->index >= 0);
940
941 /*
942 * Copy over the offending entry, or just zero it out.
943 */
944 btree = &node->btree[drop_blk->index];
Nathan Scottfac80cc2006-03-17 17:29:56 +1100945 if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) {
946 tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 tmp *= (uint)sizeof(xfs_da_node_entry_t);
948 memmove(btree, btree + 1, tmp);
949 xfs_da_log_buf(state->args->trans, drop_blk->bp,
950 XFS_DA_LOGRANGE(node, btree, tmp));
Nathan Scottfac80cc2006-03-17 17:29:56 +1100951 btree = &node->btree[be16_to_cpu(node->hdr.count)-1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 }
953 memset((char *)btree, 0, sizeof(xfs_da_node_entry_t));
954 xfs_da_log_buf(state->args->trans, drop_blk->bp,
955 XFS_DA_LOGRANGE(node, btree, sizeof(*btree)));
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800956 be16_add_cpu(&node->hdr.count, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 xfs_da_log_buf(state->args->trans, drop_blk->bp,
958 XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr)));
959
960 /*
961 * Copy the last hash value from the block to propagate upwards.
962 */
963 btree--;
Nathan Scott403432d2006-03-17 17:29:46 +1100964 drop_blk->hashval = be32_to_cpu(btree->hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965}
966
967/*
968 * Unbalance the btree elements between two intermediate nodes,
969 * move all Btree elements from one node into another.
970 */
971STATIC void
972xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
973 xfs_da_state_blk_t *save_blk)
974{
975 xfs_da_intnode_t *drop_node, *save_node;
976 xfs_da_node_entry_t *btree;
977 int tmp;
978 xfs_trans_t *tp;
979
980 drop_node = drop_blk->bp->data;
981 save_node = save_blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +1100982 ASSERT(be16_to_cpu(drop_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
983 ASSERT(be16_to_cpu(save_node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 tp = state->args->trans;
985
986 /*
987 * If the dying block has lower hashvals, then move all the
988 * elements in the remaining block up to make a hole.
989 */
Nathan Scott403432d2006-03-17 17:29:46 +1100990 if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) ||
Nathan Scottfac80cc2006-03-17 17:29:56 +1100991 (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) <
992 be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 {
Nathan Scottfac80cc2006-03-17 17:29:56 +1100994 btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)];
995 tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 memmove(btree, &save_node->btree[0], tmp);
997 btree = &save_node->btree[0];
998 xfs_da_log_buf(tp, save_blk->bp,
999 XFS_DA_LOGRANGE(save_node, btree,
Nathan Scottfac80cc2006-03-17 17:29:56 +11001000 (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 sizeof(xfs_da_node_entry_t)));
1002 } else {
Nathan Scottfac80cc2006-03-17 17:29:56 +11001003 btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 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(drop_node->hdr.count) *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 sizeof(xfs_da_node_entry_t)));
1008 }
1009
1010 /*
1011 * Move all the B-tree elements from drop_blk to save_blk.
1012 */
Nathan Scottfac80cc2006-03-17 17:29:56 +11001013 tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 memcpy(btree, &drop_node->btree[0], tmp);
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001015 be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
1017 xfs_da_log_buf(tp, save_blk->bp,
1018 XFS_DA_LOGRANGE(save_node, &save_node->hdr,
1019 sizeof(save_node->hdr)));
1020
1021 /*
1022 * Save the last hashval in the remaining block for upward propagation.
1023 */
Nathan Scottfac80cc2006-03-17 17:29:56 +11001024 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 -07001025}
1026
1027/*========================================================================
1028 * Routines used for finding things in the Btree.
1029 *========================================================================*/
1030
1031/*
1032 * Walk down the Btree looking for a particular filename, filling
1033 * in the state structure as we go.
1034 *
1035 * We will set the state structure to point to each of the elements
1036 * in each of the nodes where either the hashval is or should be.
1037 *
1038 * We support duplicate hashval's so for each entry in the current
1039 * node that could contain the desired hashval, descend. This is a
1040 * pruned depth-first tree search.
1041 */
1042int /* error */
1043xfs_da_node_lookup_int(xfs_da_state_t *state, int *result)
1044{
1045 xfs_da_state_blk_t *blk;
1046 xfs_da_blkinfo_t *curr;
1047 xfs_da_intnode_t *node;
1048 xfs_da_node_entry_t *btree;
1049 xfs_dablk_t blkno;
1050 int probe, span, max, error, retval;
Nathan Scottd432c802006-09-28 11:03:44 +10001051 xfs_dahash_t hashval, btreehashval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 xfs_da_args_t *args;
1053
1054 args = state->args;
1055
1056 /*
1057 * Descend thru the B-tree searching each level for the right
1058 * node to use, until the right hashval is found.
1059 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001060 blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 for (blk = &state->path.blk[0], state->path.active = 1;
1062 state->path.active <= XFS_DA_NODE_MAXDEPTH;
1063 blk++, state->path.active++) {
1064 /*
1065 * Read the next node down in the tree.
1066 */
1067 blk->blkno = blkno;
1068 error = xfs_da_read_buf(args->trans, args->dp, blkno,
1069 -1, &blk->bp, args->whichfork);
1070 if (error) {
1071 blk->blkno = 0;
1072 state->path.active--;
1073 return(error);
1074 }
1075 curr = blk->bp->data;
Nathan Scottd432c802006-09-28 11:03:44 +10001076 blk->magic = be16_to_cpu(curr->magic);
1077 ASSERT(blk->magic == XFS_DA_NODE_MAGIC ||
1078 blk->magic == XFS_DIR2_LEAFN_MAGIC ||
1079 blk->magic == XFS_ATTR_LEAF_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 /*
1082 * Search an intermediate node for a match.
1083 */
Nathan Scott89da0542006-03-17 17:28:40 +11001084 if (blk->magic == XFS_DA_NODE_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 node = blk->bp->data;
Nathan Scottd432c802006-09-28 11:03:44 +10001086 max = be16_to_cpu(node->hdr.count);
Lachlan McIlroy1c91ad32007-02-10 18:35:27 +11001087 blk->hashval = be32_to_cpu(node->btree[max-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
1089 /*
1090 * Binary search. (note: small blocks will skip loop)
1091 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 probe = span = max / 2;
1093 hashval = args->hashval;
1094 for (btree = &node->btree[probe]; span > 4;
1095 btree = &node->btree[probe]) {
1096 span /= 2;
Nathan Scottd432c802006-09-28 11:03:44 +10001097 btreehashval = be32_to_cpu(btree->hashval);
1098 if (btreehashval < hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 probe += span;
Nathan Scottd432c802006-09-28 11:03:44 +10001100 else if (btreehashval > hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 probe -= span;
1102 else
1103 break;
1104 }
1105 ASSERT((probe >= 0) && (probe < max));
Nathan Scott403432d2006-03-17 17:29:46 +11001106 ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
1108 /*
1109 * Since we may have duplicate hashval's, find the first
1110 * matching hashval in the node.
1111 */
Nathan Scott403432d2006-03-17 17:29:46 +11001112 while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 btree--;
1114 probe--;
1115 }
Nathan Scott403432d2006-03-17 17:29:46 +11001116 while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 btree++;
1118 probe++;
1119 }
1120
1121 /*
1122 * Pick the right block to descend on.
1123 */
1124 if (probe == max) {
1125 blk->index = max-1;
Nathan Scott403432d2006-03-17 17:29:46 +11001126 blkno = be32_to_cpu(node->btree[max-1].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 } else {
1128 blk->index = probe;
Nathan Scott403432d2006-03-17 17:29:46 +11001129 blkno = be32_to_cpu(btree->before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 }
Nathan Scottd432c802006-09-28 11:03:44 +10001131 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL);
1133 break;
Nathan Scottd432c802006-09-28 11:03:44 +10001134 } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL);
1136 break;
1137 }
1138 }
1139
1140 /*
1141 * A leaf block that ends in the hashval that we are interested in
1142 * (final hashval == search hashval) means that the next block may
1143 * contain more entries with the same hashval, shift upward to the
1144 * next leaf and keep searching.
1145 */
1146 for (;;) {
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001147 if (blk->magic == XFS_DIR2_LEAFN_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 retval = xfs_dir2_leafn_lookup_int(blk->bp, args,
1149 &blk->index, state);
Nathan Scottd432c802006-09-28 11:03:44 +10001150 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 retval = xfs_attr_leaf_lookup_int(blk->bp, args);
1152 blk->index = args->index;
1153 args->blkno = blk->blkno;
Nathan Scottd432c802006-09-28 11:03:44 +10001154 } else {
1155 ASSERT(0);
1156 return XFS_ERROR(EFSCORRUPTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 if (((retval == ENOENT) || (retval == ENOATTR)) &&
1159 (blk->hashval == args->hashval)) {
1160 error = xfs_da_path_shift(state, &state->path, 1, 1,
1161 &retval);
1162 if (error)
1163 return(error);
1164 if (retval == 0) {
1165 continue;
Nathan Scottd432c802006-09-28 11:03:44 +10001166 } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 /* path_shift() gives ENOENT */
1168 retval = XFS_ERROR(ENOATTR);
1169 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 }
1171 break;
1172 }
1173 *result = retval;
1174 return(0);
1175}
1176
1177/*========================================================================
1178 * Utility routines.
1179 *========================================================================*/
1180
1181/*
1182 * Link a new block into a doubly linked list of blocks (of whatever type).
1183 */
1184int /* error */
1185xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk,
1186 xfs_da_state_blk_t *new_blk)
1187{
1188 xfs_da_blkinfo_t *old_info, *new_info, *tmp_info;
1189 xfs_da_args_t *args;
1190 int before=0, error;
1191 xfs_dabuf_t *bp;
1192
1193 /*
1194 * Set up environment.
1195 */
1196 args = state->args;
1197 ASSERT(args != NULL);
1198 old_info = old_blk->bp->data;
1199 new_info = new_blk->bp->data;
1200 ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001201 old_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 old_blk->magic == XFS_ATTR_LEAF_MAGIC);
Nathan Scott89da0542006-03-17 17:28:40 +11001203 ASSERT(old_blk->magic == be16_to_cpu(old_info->magic));
1204 ASSERT(new_blk->magic == be16_to_cpu(new_info->magic));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 ASSERT(old_blk->magic == new_blk->magic);
1206
1207 switch (old_blk->magic) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 case XFS_ATTR_LEAF_MAGIC:
1209 before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp);
1210 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp);
1213 break;
1214 case XFS_DA_NODE_MAGIC:
1215 before = xfs_da_node_order(old_blk->bp, new_blk->bp);
1216 break;
1217 }
1218
1219 /*
1220 * Link blocks in appropriate order.
1221 */
1222 if (before) {
1223 /*
1224 * Link new block in before existing block.
1225 */
Nathan Scott89da0542006-03-17 17:28:40 +11001226 new_info->forw = cpu_to_be32(old_blk->blkno);
1227 new_info->back = old_info->back;
1228 if (old_info->back) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 error = xfs_da_read_buf(args->trans, args->dp,
Nathan Scott89da0542006-03-17 17:28:40 +11001230 be32_to_cpu(old_info->back),
1231 -1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 if (error)
1233 return(error);
1234 ASSERT(bp != NULL);
1235 tmp_info = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001236 ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic));
1237 ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno);
1238 tmp_info->forw = cpu_to_be32(new_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1240 xfs_da_buf_done(bp);
1241 }
Nathan Scott89da0542006-03-17 17:28:40 +11001242 old_info->back = cpu_to_be32(new_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 } else {
1244 /*
1245 * Link new block in after existing block.
1246 */
Nathan Scott89da0542006-03-17 17:28:40 +11001247 new_info->forw = old_info->forw;
1248 new_info->back = cpu_to_be32(old_blk->blkno);
1249 if (old_info->forw) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 error = xfs_da_read_buf(args->trans, args->dp,
Nathan Scott89da0542006-03-17 17:28:40 +11001251 be32_to_cpu(old_info->forw),
1252 -1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 if (error)
1254 return(error);
1255 ASSERT(bp != NULL);
1256 tmp_info = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001257 ASSERT(tmp_info->magic == old_info->magic);
1258 ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno);
1259 tmp_info->back = cpu_to_be32(new_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 xfs_da_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1);
1261 xfs_da_buf_done(bp);
1262 }
Nathan Scott89da0542006-03-17 17:28:40 +11001263 old_info->forw = cpu_to_be32(new_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 }
1265
1266 xfs_da_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1);
1267 xfs_da_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1);
1268 return(0);
1269}
1270
1271/*
1272 * Compare two intermediate nodes for "order".
1273 */
1274STATIC int
1275xfs_da_node_order(xfs_dabuf_t *node1_bp, xfs_dabuf_t *node2_bp)
1276{
1277 xfs_da_intnode_t *node1, *node2;
1278
1279 node1 = node1_bp->data;
1280 node2 = node2_bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001281 ASSERT((be16_to_cpu(node1->hdr.info.magic) == XFS_DA_NODE_MAGIC) &&
1282 (be16_to_cpu(node2->hdr.info.magic) == XFS_DA_NODE_MAGIC));
Nathan Scottfac80cc2006-03-17 17:29:56 +11001283 if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) &&
Nathan Scott403432d2006-03-17 17:29:46 +11001284 ((be32_to_cpu(node2->btree[0].hashval) <
1285 be32_to_cpu(node1->btree[0].hashval)) ||
Nathan Scottfac80cc2006-03-17 17:29:56 +11001286 (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) <
1287 be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 return(1);
1289 }
1290 return(0);
1291}
1292
1293/*
1294 * Pick up the last hashvalue from an intermediate node.
1295 */
1296STATIC uint
1297xfs_da_node_lasthash(xfs_dabuf_t *bp, int *count)
1298{
1299 xfs_da_intnode_t *node;
1300
1301 node = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001302 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 if (count)
Nathan Scottfac80cc2006-03-17 17:29:56 +11001304 *count = be16_to_cpu(node->hdr.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if (!node->hdr.count)
1306 return(0);
Nathan Scottfac80cc2006-03-17 17:29:56 +11001307 return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308}
1309
1310/*
1311 * Unlink a block from a doubly linked list of blocks.
1312 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001313STATIC int /* error */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1315 xfs_da_state_blk_t *save_blk)
1316{
1317 xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info;
1318 xfs_da_args_t *args;
1319 xfs_dabuf_t *bp;
1320 int error;
1321
1322 /*
1323 * Set up environment.
1324 */
1325 args = state->args;
1326 ASSERT(args != NULL);
1327 save_info = save_blk->bp->data;
1328 drop_info = drop_blk->bp->data;
1329 ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001330 save_blk->magic == XFS_DIR2_LEAFN_MAGIC ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 save_blk->magic == XFS_ATTR_LEAF_MAGIC);
Nathan Scott89da0542006-03-17 17:28:40 +11001332 ASSERT(save_blk->magic == be16_to_cpu(save_info->magic));
1333 ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 ASSERT(save_blk->magic == drop_blk->magic);
Nathan Scott89da0542006-03-17 17:28:40 +11001335 ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) ||
1336 (be32_to_cpu(save_info->back) == drop_blk->blkno));
1337 ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) ||
1338 (be32_to_cpu(drop_info->back) == save_blk->blkno));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
1340 /*
1341 * Unlink the leaf block from the doubly linked chain of leaves.
1342 */
Nathan Scott89da0542006-03-17 17:28:40 +11001343 if (be32_to_cpu(save_info->back) == drop_blk->blkno) {
1344 save_info->back = drop_info->back;
1345 if (drop_info->back) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 error = xfs_da_read_buf(args->trans, args->dp,
Nathan Scott89da0542006-03-17 17:28:40 +11001347 be32_to_cpu(drop_info->back),
1348 -1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 if (error)
1350 return(error);
1351 ASSERT(bp != NULL);
1352 tmp_info = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001353 ASSERT(tmp_info->magic == save_info->magic);
1354 ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno);
1355 tmp_info->forw = cpu_to_be32(save_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 xfs_da_log_buf(args->trans, bp, 0,
1357 sizeof(*tmp_info) - 1);
1358 xfs_da_buf_done(bp);
1359 }
1360 } else {
Nathan Scott89da0542006-03-17 17:28:40 +11001361 save_info->forw = drop_info->forw;
1362 if (drop_info->forw) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 error = xfs_da_read_buf(args->trans, args->dp,
Nathan Scott89da0542006-03-17 17:28:40 +11001364 be32_to_cpu(drop_info->forw),
1365 -1, &bp, args->whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if (error)
1367 return(error);
1368 ASSERT(bp != NULL);
1369 tmp_info = bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001370 ASSERT(tmp_info->magic == save_info->magic);
1371 ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno);
1372 tmp_info->back = cpu_to_be32(save_blk->blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 xfs_da_log_buf(args->trans, bp, 0,
1374 sizeof(*tmp_info) - 1);
1375 xfs_da_buf_done(bp);
1376 }
1377 }
1378
1379 xfs_da_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1);
1380 return(0);
1381}
1382
1383/*
1384 * Move a path "forward" or "!forward" one block at the current level.
1385 *
1386 * This routine will adjust a "path" to point to the next block
1387 * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the
1388 * Btree, including updating pointers to the intermediate nodes between
1389 * the new bottom and the root.
1390 */
1391int /* error */
1392xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path,
1393 int forward, int release, int *result)
1394{
1395 xfs_da_state_blk_t *blk;
1396 xfs_da_blkinfo_t *info;
1397 xfs_da_intnode_t *node;
1398 xfs_da_args_t *args;
1399 xfs_dablk_t blkno=0;
1400 int level, error;
1401
1402 /*
1403 * Roll up the Btree looking for the first block where our
1404 * current index is not at the edge of the block. Note that
1405 * we skip the bottom layer because we want the sibling block.
1406 */
1407 args = state->args;
1408 ASSERT(args != NULL);
1409 ASSERT(path != NULL);
1410 ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1411 level = (path->active-1) - 1; /* skip bottom layer in path */
1412 for (blk = &path->blk[level]; level >= 0; blk--, level--) {
1413 ASSERT(blk->bp != NULL);
1414 node = blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001415 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
Nathan Scottfac80cc2006-03-17 17:29:56 +11001416 if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 blk->index++;
Nathan Scott403432d2006-03-17 17:29:46 +11001418 blkno = be32_to_cpu(node->btree[blk->index].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 break;
1420 } else if (!forward && (blk->index > 0)) {
1421 blk->index--;
Nathan Scott403432d2006-03-17 17:29:46 +11001422 blkno = be32_to_cpu(node->btree[blk->index].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 break;
1424 }
1425 }
1426 if (level < 0) {
1427 *result = XFS_ERROR(ENOENT); /* we're out of our tree */
Barry Naujok6a178102008-05-21 16:42:05 +10001428 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 return(0);
1430 }
1431
1432 /*
1433 * Roll down the edge of the subtree until we reach the
1434 * same depth we were at originally.
1435 */
1436 for (blk++, level++; level < path->active; blk++, level++) {
1437 /*
1438 * Release the old block.
1439 * (if it's dirty, trans won't actually let go)
1440 */
1441 if (release)
1442 xfs_da_brelse(args->trans, blk->bp);
1443
1444 /*
1445 * Read the next child block.
1446 */
1447 blk->blkno = blkno;
1448 error = xfs_da_read_buf(args->trans, args->dp, blkno, -1,
1449 &blk->bp, args->whichfork);
1450 if (error)
1451 return(error);
1452 ASSERT(blk->bp != NULL);
1453 info = blk->bp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11001454 ASSERT(be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001455 be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC ||
Nathan Scott89da0542006-03-17 17:28:40 +11001456 be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1457 blk->magic = be16_to_cpu(info->magic);
1458 if (blk->magic == XFS_DA_NODE_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 node = (xfs_da_intnode_t *)info;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001460 blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 if (forward)
1462 blk->index = 0;
1463 else
Nathan Scottfac80cc2006-03-17 17:29:56 +11001464 blk->index = be16_to_cpu(node->hdr.count)-1;
Nathan Scott403432d2006-03-17 17:29:46 +11001465 blkno = be32_to_cpu(node->btree[blk->index].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 } else {
1467 ASSERT(level == path->active-1);
1468 blk->index = 0;
1469 switch(blk->magic) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 case XFS_ATTR_LEAF_MAGIC:
1471 blk->hashval = xfs_attr_leaf_lasthash(blk->bp,
1472 NULL);
1473 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 case XFS_DIR2_LEAFN_MAGIC:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 blk->hashval = xfs_dir2_leafn_lasthash(blk->bp,
1476 NULL);
1477 break;
1478 default:
1479 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC ||
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001480 blk->magic == XFS_DIR2_LEAFN_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 break;
1482 }
1483 }
1484 }
1485 *result = 0;
1486 return(0);
1487}
1488
1489
1490/*========================================================================
1491 * Utility routines.
1492 *========================================================================*/
1493
1494/*
1495 * Implement a simple hash on a character string.
1496 * Rotate the hash value by 7 bits, then XOR each character in.
1497 * This is implemented with some source-level loop unrolling.
1498 */
1499xfs_dahash_t
Christoph Hellwiga5687782009-02-09 08:37:39 +01001500xfs_da_hashname(const __uint8_t *name, int namelen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501{
1502 xfs_dahash_t hash;
1503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 /*
1505 * Do four characters at a time as long as we can.
1506 */
1507 for (hash = 0; namelen >= 4; namelen -= 4, name += 4)
1508 hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^
1509 (name[3] << 0) ^ rol32(hash, 7 * 4);
1510
1511 /*
1512 * Now do the rest of the characters.
1513 */
1514 switch (namelen) {
1515 case 3:
1516 return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^
1517 rol32(hash, 7 * 3);
1518 case 2:
1519 return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2);
1520 case 1:
1521 return (name[0] << 0) ^ rol32(hash, 7 * 1);
Nathan Scott2b3b6d02005-11-02 15:12:28 +11001522 default: /* case 0: */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 return hash;
1524 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525}
1526
Barry Naujok5163f952008-05-21 16:41:01 +10001527enum xfs_dacmp
1528xfs_da_compname(
1529 struct xfs_da_args *args,
Dave Chinner2bc75422010-01-20 10:47:17 +11001530 const unsigned char *name,
1531 int len)
Barry Naujok5163f952008-05-21 16:41:01 +10001532{
1533 return (args->namelen == len && memcmp(args->name, name, len) == 0) ?
1534 XFS_CMP_EXACT : XFS_CMP_DIFFERENT;
1535}
1536
1537static xfs_dahash_t
1538xfs_default_hashname(
1539 struct xfs_name *name)
1540{
1541 return xfs_da_hashname(name->name, name->len);
1542}
1543
1544const struct xfs_nameops xfs_default_nameops = {
1545 .hashname = xfs_default_hashname,
1546 .compname = xfs_da_compname
1547};
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549/*
1550 * Add a block to the btree ahead of the file.
1551 * Return the new block number to the caller.
1552 */
1553int
1554xfs_da_grow_inode(xfs_da_args_t *args, xfs_dablk_t *new_blkno)
1555{
1556 xfs_fileoff_t bno, b;
1557 xfs_bmbt_irec_t map;
1558 xfs_bmbt_irec_t *mapp;
1559 xfs_inode_t *dp;
1560 int nmap, error, w, count, c, got, i, mapi;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 xfs_trans_t *tp;
1562 xfs_mount_t *mp;
David Chinnera7444052008-10-30 17:38:12 +11001563 xfs_drfsbno_t nblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
1565 dp = args->dp;
1566 mp = dp->i_mount;
1567 w = args->whichfork;
1568 tp = args->trans;
David Chinnera7444052008-10-30 17:38:12 +11001569 nblks = dp->i_d.di_nblocks;
1570
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 /*
1572 * For new directories adjust the file offset and block count.
1573 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001574 if (w == XFS_DATA_FORK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 bno = mp->m_dirleafblk;
1576 count = mp->m_dirblkfsbs;
1577 } else {
1578 bno = 0;
1579 count = 1;
1580 }
1581 /*
1582 * Find a spot in the file space to put the new block.
1583 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001584 if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, w)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 return error;
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001586 if (w == XFS_DATA_FORK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 ASSERT(bno >= mp->m_dirleafblk && bno < mp->m_dirfreeblk);
1588 /*
1589 * Try mapping it in one filesystem block.
1590 */
1591 nmap = 1;
1592 ASSERT(args->firstblock != NULL);
1593 if ((error = xfs_bmapi(tp, dp, bno, count,
Eric Sandeen9d87c312009-01-14 23:22:07 -06001594 xfs_bmapi_aflag(w)|XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 XFS_BMAPI_CONTIG,
1596 args->firstblock, args->total, &map, &nmap,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001597 args->flist))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 return error;
1599 }
1600 ASSERT(nmap <= 1);
1601 if (nmap == 1) {
1602 mapp = &map;
1603 mapi = 1;
1604 }
1605 /*
1606 * If we didn't get it and the block might work if fragmented,
1607 * try without the CONTIG flag. Loop until we get it all.
1608 */
1609 else if (nmap == 0 && count > 1) {
1610 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
1611 for (b = bno, mapi = 0; b < bno + count; ) {
1612 nmap = MIN(XFS_BMAP_MAX_NMAP, count);
1613 c = (int)(bno + count - b);
1614 if ((error = xfs_bmapi(tp, dp, b, c,
Eric Sandeen9d87c312009-01-14 23:22:07 -06001615 xfs_bmapi_aflag(w)|XFS_BMAPI_WRITE|
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 XFS_BMAPI_METADATA,
1617 args->firstblock, args->total,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001618 &mapp[mapi], &nmap, args->flist))) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001619 kmem_free(mapp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 return error;
1621 }
1622 if (nmap < 1)
1623 break;
1624 mapi += nmap;
1625 b = mapp[mapi - 1].br_startoff +
1626 mapp[mapi - 1].br_blockcount;
1627 }
1628 } else {
1629 mapi = 0;
1630 mapp = NULL;
1631 }
1632 /*
1633 * Count the blocks we got, make sure it matches the total.
1634 */
1635 for (i = 0, got = 0; i < mapi; i++)
1636 got += mapp[i].br_blockcount;
1637 if (got != count || mapp[0].br_startoff != bno ||
1638 mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
1639 bno + count) {
1640 if (mapp != &map)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001641 kmem_free(mapp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 return XFS_ERROR(ENOSPC);
1643 }
1644 if (mapp != &map)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001645 kmem_free(mapp);
David Chinnera7444052008-10-30 17:38:12 +11001646 /* account for newly allocated blocks in reserved blocks total */
1647 args->total -= dp->i_d.di_nblocks - nblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 *new_blkno = (xfs_dablk_t)bno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 return 0;
1650}
1651
1652/*
1653 * Ick. We need to always be able to remove a btree block, even
1654 * if there's no space reservation because the filesystem is full.
1655 * This is called if xfs_bunmapi on a btree block fails due to ENOSPC.
1656 * It swaps the target block with the last block in the file. The
1657 * last block in the file can always be removed since it can't cause
1658 * a bmap btree split to do that.
1659 */
1660STATIC int
1661xfs_da_swap_lastblock(xfs_da_args_t *args, xfs_dablk_t *dead_blknop,
1662 xfs_dabuf_t **dead_bufp)
1663{
1664 xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno;
1665 xfs_dabuf_t *dead_buf, *last_buf, *sib_buf, *par_buf;
1666 xfs_fileoff_t lastoff;
1667 xfs_inode_t *ip;
1668 xfs_trans_t *tp;
1669 xfs_mount_t *mp;
1670 int error, w, entno, level, dead_level;
1671 xfs_da_blkinfo_t *dead_info, *sib_info;
1672 xfs_da_intnode_t *par_node, *dead_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 xfs_dir2_leaf_t *dead_leaf2;
1674 xfs_dahash_t dead_hash;
1675
1676 dead_buf = *dead_bufp;
1677 dead_blkno = *dead_blknop;
1678 tp = args->trans;
1679 ip = args->dp;
1680 w = args->whichfork;
1681 ASSERT(w == XFS_DATA_FORK);
1682 mp = ip->i_mount;
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001683 lastoff = mp->m_dirfreeblk;
1684 error = xfs_bmap_last_before(tp, ip, &lastoff, w);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 if (error)
1686 return error;
1687 if (unlikely(lastoff == 0)) {
1688 XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW,
1689 mp);
1690 return XFS_ERROR(EFSCORRUPTED);
1691 }
1692 /*
1693 * Read the last block in the btree space.
1694 */
1695 last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs;
1696 if ((error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w)))
1697 return error;
1698 /*
1699 * Copy the last block into the dead buffer and log it.
1700 */
1701 memcpy(dead_buf->data, last_buf->data, mp->m_dirblksize);
1702 xfs_da_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1);
1703 dead_info = dead_buf->data;
1704 /*
1705 * Get values from the moved block.
1706 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001707 if (be16_to_cpu(dead_info->magic) == XFS_DIR2_LEAFN_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 dead_leaf2 = (xfs_dir2_leaf_t *)dead_info;
1709 dead_level = 0;
Nathan Scott3c1f9c12006-03-17 17:28:18 +11001710 dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 } else {
Nathan Scott89da0542006-03-17 17:28:40 +11001712 ASSERT(be16_to_cpu(dead_info->magic) == XFS_DA_NODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 dead_node = (xfs_da_intnode_t *)dead_info;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001714 dead_level = be16_to_cpu(dead_node->hdr.level);
1715 dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 }
1717 sib_buf = par_buf = NULL;
1718 /*
1719 * If the moved block has a left sibling, fix up the pointers.
1720 */
Nathan Scott89da0542006-03-17 17:28:40 +11001721 if ((sib_blkno = be32_to_cpu(dead_info->back))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1723 goto done;
1724 sib_info = sib_buf->data;
1725 if (unlikely(
Nathan Scott89da0542006-03-17 17:28:40 +11001726 be32_to_cpu(sib_info->forw) != last_blkno ||
1727 sib_info->magic != dead_info->magic)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)",
1729 XFS_ERRLEVEL_LOW, mp);
1730 error = XFS_ERROR(EFSCORRUPTED);
1731 goto done;
1732 }
Nathan Scott89da0542006-03-17 17:28:40 +11001733 sib_info->forw = cpu_to_be32(dead_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 xfs_da_log_buf(tp, sib_buf,
1735 XFS_DA_LOGRANGE(sib_info, &sib_info->forw,
1736 sizeof(sib_info->forw)));
1737 xfs_da_buf_done(sib_buf);
1738 sib_buf = NULL;
1739 }
1740 /*
1741 * If the moved block has a right sibling, fix up the pointers.
1742 */
Nathan Scott89da0542006-03-17 17:28:40 +11001743 if ((sib_blkno = be32_to_cpu(dead_info->forw))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 if ((error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w)))
1745 goto done;
1746 sib_info = sib_buf->data;
1747 if (unlikely(
Nathan Scott89da0542006-03-17 17:28:40 +11001748 be32_to_cpu(sib_info->back) != last_blkno ||
1749 sib_info->magic != dead_info->magic)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)",
1751 XFS_ERRLEVEL_LOW, mp);
1752 error = XFS_ERROR(EFSCORRUPTED);
1753 goto done;
1754 }
Nathan Scott89da0542006-03-17 17:28:40 +11001755 sib_info->back = cpu_to_be32(dead_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 xfs_da_log_buf(tp, sib_buf,
1757 XFS_DA_LOGRANGE(sib_info, &sib_info->back,
1758 sizeof(sib_info->back)));
1759 xfs_da_buf_done(sib_buf);
1760 sib_buf = NULL;
1761 }
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001762 par_blkno = mp->m_dirleafblk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 level = -1;
1764 /*
1765 * Walk down the tree looking for the parent of the moved block.
1766 */
1767 for (;;) {
1768 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1769 goto done;
1770 par_node = par_buf->data;
1771 if (unlikely(
Nathan Scott89da0542006-03-17 17:28:40 +11001772 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC ||
Nathan Scottfac80cc2006-03-17 17:29:56 +11001773 (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)",
1775 XFS_ERRLEVEL_LOW, mp);
1776 error = XFS_ERROR(EFSCORRUPTED);
1777 goto done;
1778 }
Nathan Scottfac80cc2006-03-17 17:29:56 +11001779 level = be16_to_cpu(par_node->hdr.level);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 for (entno = 0;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001781 entno < be16_to_cpu(par_node->hdr.count) &&
Nathan Scott403432d2006-03-17 17:29:46 +11001782 be32_to_cpu(par_node->btree[entno].hashval) < dead_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 entno++)
1784 continue;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001785 if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)",
1787 XFS_ERRLEVEL_LOW, mp);
1788 error = XFS_ERROR(EFSCORRUPTED);
1789 goto done;
1790 }
Nathan Scott403432d2006-03-17 17:29:46 +11001791 par_blkno = be32_to_cpu(par_node->btree[entno].before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 if (level == dead_level + 1)
1793 break;
1794 xfs_da_brelse(tp, par_buf);
1795 par_buf = NULL;
1796 }
1797 /*
1798 * We're in the right parent block.
1799 * Look for the right entry.
1800 */
1801 for (;;) {
1802 for (;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001803 entno < be16_to_cpu(par_node->hdr.count) &&
Nathan Scott403432d2006-03-17 17:29:46 +11001804 be32_to_cpu(par_node->btree[entno].before) != last_blkno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 entno++)
1806 continue;
Nathan Scottfac80cc2006-03-17 17:29:56 +11001807 if (entno < be16_to_cpu(par_node->hdr.count))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 break;
Nathan Scott89da0542006-03-17 17:28:40 +11001809 par_blkno = be32_to_cpu(par_node->hdr.info.forw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 xfs_da_brelse(tp, par_buf);
1811 par_buf = NULL;
1812 if (unlikely(par_blkno == 0)) {
1813 XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)",
1814 XFS_ERRLEVEL_LOW, mp);
1815 error = XFS_ERROR(EFSCORRUPTED);
1816 goto done;
1817 }
1818 if ((error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w)))
1819 goto done;
1820 par_node = par_buf->data;
1821 if (unlikely(
Nathan Scottfac80cc2006-03-17 17:29:56 +11001822 be16_to_cpu(par_node->hdr.level) != level ||
Nathan Scott89da0542006-03-17 17:28:40 +11001823 be16_to_cpu(par_node->hdr.info.magic) != XFS_DA_NODE_MAGIC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)",
1825 XFS_ERRLEVEL_LOW, mp);
1826 error = XFS_ERROR(EFSCORRUPTED);
1827 goto done;
1828 }
1829 entno = 0;
1830 }
1831 /*
1832 * Update the parent entry pointing to the moved block.
1833 */
Nathan Scott403432d2006-03-17 17:29:46 +11001834 par_node->btree[entno].before = cpu_to_be32(dead_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 xfs_da_log_buf(tp, par_buf,
1836 XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before,
1837 sizeof(par_node->btree[entno].before)));
1838 xfs_da_buf_done(par_buf);
1839 xfs_da_buf_done(dead_buf);
1840 *dead_blknop = last_blkno;
1841 *dead_bufp = last_buf;
1842 return 0;
1843done:
1844 if (par_buf)
1845 xfs_da_brelse(tp, par_buf);
1846 if (sib_buf)
1847 xfs_da_brelse(tp, sib_buf);
1848 xfs_da_brelse(tp, last_buf);
1849 return error;
1850}
1851
1852/*
1853 * Remove a btree block from a directory or attribute.
1854 */
1855int
1856xfs_da_shrink_inode(xfs_da_args_t *args, xfs_dablk_t dead_blkno,
1857 xfs_dabuf_t *dead_buf)
1858{
1859 xfs_inode_t *dp;
1860 int done, error, w, count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 xfs_trans_t *tp;
1862 xfs_mount_t *mp;
1863
1864 dp = args->dp;
1865 w = args->whichfork;
1866 tp = args->trans;
1867 mp = dp->i_mount;
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001868 if (w == XFS_DATA_FORK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 count = mp->m_dirblkfsbs;
1870 else
1871 count = 1;
1872 for (;;) {
1873 /*
1874 * Remove extents. If we get ENOSPC for a dir we have to move
1875 * the last block to the place we want to kill.
1876 */
1877 if ((error = xfs_bunmapi(tp, dp, dead_blkno, count,
Eric Sandeen9d87c312009-01-14 23:22:07 -06001878 xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001879 0, args->firstblock, args->flist,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 &done)) == ENOSPC) {
1881 if (w != XFS_DATA_FORK)
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001882 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 if ((error = xfs_da_swap_lastblock(args, &dead_blkno,
1884 &dead_buf)))
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001885 break;
1886 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 }
1889 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 xfs_da_binval(tp, dead_buf);
1891 return error;
1892}
1893
1894/*
1895 * See if the mapping(s) for this btree block are valid, i.e.
1896 * don't contain holes, are logically contiguous, and cover the whole range.
1897 */
1898STATIC int
1899xfs_da_map_covers_blocks(
1900 int nmap,
1901 xfs_bmbt_irec_t *mapp,
1902 xfs_dablk_t bno,
1903 int count)
1904{
1905 int i;
1906 xfs_fileoff_t off;
1907
1908 for (i = 0, off = bno; i < nmap; i++) {
1909 if (mapp[i].br_startblock == HOLESTARTBLOCK ||
1910 mapp[i].br_startblock == DELAYSTARTBLOCK) {
1911 return 0;
1912 }
1913 if (off != mapp[i].br_startoff) {
1914 return 0;
1915 }
1916 off += mapp[i].br_blockcount;
1917 }
1918 return off == bno + count;
1919}
1920
1921/*
1922 * Make a dabuf.
1923 * Used for get_buf, read_buf, read_bufr, and reada_buf.
1924 */
1925STATIC int
1926xfs_da_do_buf(
1927 xfs_trans_t *trans,
1928 xfs_inode_t *dp,
1929 xfs_dablk_t bno,
1930 xfs_daddr_t *mappedbnop,
1931 xfs_dabuf_t **bpp,
1932 int whichfork,
1933 int caller,
1934 inst_t *ra)
1935{
1936 xfs_buf_t *bp = NULL;
1937 xfs_buf_t **bplist;
1938 int error=0;
1939 int i;
1940 xfs_bmbt_irec_t map;
1941 xfs_bmbt_irec_t *mapp;
1942 xfs_daddr_t mappedbno;
1943 xfs_mount_t *mp;
1944 int nbplist=0;
1945 int nfsb;
1946 int nmap;
1947 xfs_dabuf_t *rbp;
1948
1949 mp = dp->i_mount;
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001950 nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 mappedbno = *mappedbnop;
1952 /*
1953 * Caller doesn't have a mapping. -2 means don't complain
1954 * if we land in a hole.
1955 */
1956 if (mappedbno == -1 || mappedbno == -2) {
1957 /*
1958 * Optimize the one-block case.
1959 */
1960 if (nfsb == 1) {
1961 xfs_fsblock_t fsb;
1962
1963 if ((error =
1964 xfs_bmapi_single(trans, dp, whichfork, &fsb,
1965 (xfs_fileoff_t)bno))) {
1966 return error;
1967 }
1968 mapp = &map;
1969 if (fsb == NULLFSBLOCK) {
1970 nmap = 0;
1971 } else {
1972 map.br_startblock = fsb;
1973 map.br_startoff = (xfs_fileoff_t)bno;
1974 map.br_blockcount = 1;
1975 nmap = 1;
1976 }
1977 } else {
1978 mapp = kmem_alloc(sizeof(*mapp) * nfsb, KM_SLEEP);
1979 nmap = nfsb;
1980 if ((error = xfs_bmapi(trans, dp, (xfs_fileoff_t)bno,
1981 nfsb,
1982 XFS_BMAPI_METADATA |
Eric Sandeen9d87c312009-01-14 23:22:07 -06001983 xfs_bmapi_aflag(whichfork),
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001984 NULL, 0, mapp, &nmap, NULL)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 goto exit0;
1986 }
1987 } else {
1988 map.br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno);
1989 map.br_startoff = (xfs_fileoff_t)bno;
1990 map.br_blockcount = nfsb;
1991 mapp = &map;
1992 nmap = 1;
1993 }
1994 if (!xfs_da_map_covers_blocks(nmap, mapp, bno, nfsb)) {
1995 error = mappedbno == -2 ? 0 : XFS_ERROR(EFSCORRUPTED);
1996 if (unlikely(error == EFSCORRUPTED)) {
1997 if (xfs_error_level >= XFS_ERRLEVEL_LOW) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 cmn_err(CE_ALERT, "xfs_da_do_buf: bno %lld\n",
1999 (long long)bno);
2000 cmn_err(CE_ALERT, "dir: inode %lld\n",
2001 (long long)dp->i_ino);
2002 for (i = 0; i < nmap; i++) {
2003 cmn_err(CE_ALERT,
2004 "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d\n",
2005 i,
2006 (long long)mapp[i].br_startoff,
2007 (long long)mapp[i].br_startblock,
2008 (long long)mapp[i].br_blockcount,
2009 mapp[i].br_state);
2010 }
2011 }
2012 XFS_ERROR_REPORT("xfs_da_do_buf(1)",
2013 XFS_ERRLEVEL_LOW, mp);
2014 }
2015 goto exit0;
2016 }
2017 if (caller != 3 && nmap > 1) {
2018 bplist = kmem_alloc(sizeof(*bplist) * nmap, KM_SLEEP);
2019 nbplist = 0;
2020 } else
2021 bplist = NULL;
2022 /*
2023 * Turn the mapping(s) into buffer(s).
2024 */
2025 for (i = 0; i < nmap; i++) {
2026 int nmapped;
2027
2028 mappedbno = XFS_FSB_TO_DADDR(mp, mapp[i].br_startblock);
2029 if (i == 0)
2030 *mappedbnop = mappedbno;
2031 nmapped = (int)XFS_FSB_TO_BB(mp, mapp[i].br_blockcount);
2032 switch (caller) {
2033 case 0:
2034 bp = xfs_trans_get_buf(trans, mp->m_ddev_targp,
2035 mappedbno, nmapped, 0);
2036 error = bp ? XFS_BUF_GETERROR(bp) : XFS_ERROR(EIO);
2037 break;
2038 case 1:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 case 2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 bp = NULL;
2041 error = xfs_trans_read_buf(mp, trans, mp->m_ddev_targp,
2042 mappedbno, nmapped, 0, &bp);
2043 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 case 3:
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00002045 xfs_buf_readahead(mp->m_ddev_targp, mappedbno, nmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 error = 0;
2047 bp = NULL;
2048 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 }
2050 if (error) {
2051 if (bp)
2052 xfs_trans_brelse(trans, bp);
2053 goto exit1;
2054 }
2055 if (!bp)
2056 continue;
2057 if (caller == 1) {
2058 if (whichfork == XFS_ATTR_FORK) {
2059 XFS_BUF_SET_VTYPE_REF(bp, B_FS_ATTR_BTREE,
2060 XFS_ATTR_BTREE_REF);
2061 } else {
2062 XFS_BUF_SET_VTYPE_REF(bp, B_FS_DIR_BTREE,
2063 XFS_DIR_BTREE_REF);
2064 }
2065 }
2066 if (bplist) {
2067 bplist[nbplist++] = bp;
2068 }
2069 }
2070 /*
2071 * Build a dabuf structure.
2072 */
2073 if (bplist) {
2074 rbp = xfs_da_buf_make(nbplist, bplist, ra);
2075 } else if (bp)
2076 rbp = xfs_da_buf_make(1, &bp, ra);
2077 else
2078 rbp = NULL;
2079 /*
2080 * For read_buf, check the magic number.
2081 */
2082 if (caller == 1) {
2083 xfs_dir2_data_t *data;
2084 xfs_dir2_free_t *free;
2085 xfs_da_blkinfo_t *info;
2086 uint magic, magic1;
2087
2088 info = rbp->data;
2089 data = rbp->data;
2090 free = rbp->data;
Nathan Scott89da0542006-03-17 17:28:40 +11002091 magic = be16_to_cpu(info->magic);
Nathan Scott70e73f52006-03-17 17:26:52 +11002092 magic1 = be32_to_cpu(data->hdr.magic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 if (unlikely(
2094 XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 (magic != XFS_ATTR_LEAF_MAGIC) &&
2096 (magic != XFS_DIR2_LEAF1_MAGIC) &&
2097 (magic != XFS_DIR2_LEAFN_MAGIC) &&
2098 (magic1 != XFS_DIR2_BLOCK_MAGIC) &&
2099 (magic1 != XFS_DIR2_DATA_MAGIC) &&
Nathan Scott0ba962e2006-03-17 17:27:07 +11002100 (be32_to_cpu(free->hdr.magic) != XFS_DIR2_FREE_MAGIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 mp, XFS_ERRTAG_DA_READ_BUF,
2102 XFS_RANDOM_DA_READ_BUF))) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002103 trace_xfs_da_btree_corrupt(rbp->bps[0], _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)",
2105 XFS_ERRLEVEL_LOW, mp, info);
2106 error = XFS_ERROR(EFSCORRUPTED);
2107 xfs_da_brelse(trans, rbp);
2108 nbplist = 0;
2109 goto exit1;
2110 }
2111 }
2112 if (bplist) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002113 kmem_free(bplist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 }
2115 if (mapp != &map) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002116 kmem_free(mapp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 }
2118 if (bpp)
2119 *bpp = rbp;
2120 return 0;
2121exit1:
2122 if (bplist) {
2123 for (i = 0; i < nbplist; i++)
2124 xfs_trans_brelse(trans, bplist[i]);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002125 kmem_free(bplist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 }
2127exit0:
2128 if (mapp != &map)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002129 kmem_free(mapp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 if (bpp)
2131 *bpp = NULL;
2132 return error;
2133}
2134
2135/*
2136 * Get a buffer for the dir/attr block.
2137 */
2138int
2139xfs_da_get_buf(
2140 xfs_trans_t *trans,
2141 xfs_inode_t *dp,
2142 xfs_dablk_t bno,
2143 xfs_daddr_t mappedbno,
2144 xfs_dabuf_t **bpp,
2145 int whichfork)
2146{
2147 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 0,
2148 (inst_t *)__return_address);
2149}
2150
2151/*
2152 * Get a buffer for the dir/attr block, fill in the contents.
2153 */
2154int
2155xfs_da_read_buf(
2156 xfs_trans_t *trans,
2157 xfs_inode_t *dp,
2158 xfs_dablk_t bno,
2159 xfs_daddr_t mappedbno,
2160 xfs_dabuf_t **bpp,
2161 int whichfork)
2162{
2163 return xfs_da_do_buf(trans, dp, bno, &mappedbno, bpp, whichfork, 1,
2164 (inst_t *)__return_address);
2165}
2166
2167/*
2168 * Readahead the dir/attr block.
2169 */
2170xfs_daddr_t
2171xfs_da_reada_buf(
2172 xfs_trans_t *trans,
2173 xfs_inode_t *dp,
2174 xfs_dablk_t bno,
2175 int whichfork)
2176{
2177 xfs_daddr_t rval;
2178
2179 rval = -1;
2180 if (xfs_da_do_buf(trans, dp, bno, &rval, NULL, whichfork, 3,
2181 (inst_t *)__return_address))
2182 return -1;
2183 else
2184 return rval;
2185}
2186
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */
2188kmem_zone_t *xfs_dabuf_zone; /* dabuf zone */
2189
2190/*
2191 * Allocate a dir-state structure.
2192 * We don't put them on the stack since they're large.
2193 */
2194xfs_da_state_t *
2195xfs_da_state_alloc(void)
2196{
Christoph Hellwigf41d7fb2009-07-18 18:14:55 -04002197 return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198}
2199
2200/*
2201 * Kill the altpath contents of a da-state structure.
2202 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002203STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204xfs_da_state_kill_altpath(xfs_da_state_t *state)
2205{
2206 int i;
2207
2208 for (i = 0; i < state->altpath.active; i++) {
2209 if (state->altpath.blk[i].bp) {
2210 if (state->altpath.blk[i].bp != state->path.blk[i].bp)
2211 xfs_da_buf_done(state->altpath.blk[i].bp);
2212 state->altpath.blk[i].bp = NULL;
2213 }
2214 }
2215 state->altpath.active = 0;
2216}
2217
2218/*
2219 * Free a da-state structure.
2220 */
2221void
2222xfs_da_state_free(xfs_da_state_t *state)
2223{
2224 int i;
2225
2226 xfs_da_state_kill_altpath(state);
2227 for (i = 0; i < state->path.active; i++) {
2228 if (state->path.blk[i].bp)
2229 xfs_da_buf_done(state->path.blk[i].bp);
2230 }
2231 if (state->extravalid && state->extrablk.bp)
2232 xfs_da_buf_done(state->extrablk.bp);
2233#ifdef DEBUG
2234 memset((char *)state, 0, sizeof(*state));
2235#endif /* DEBUG */
2236 kmem_zone_free(xfs_da_state_zone, state);
2237}
2238
2239#ifdef XFS_DABUF_DEBUG
2240xfs_dabuf_t *xfs_dabuf_global_list;
Christoph Hellwig9f8868f2008-07-18 17:11:46 +10002241static DEFINE_SPINLOCK(xfs_dabuf_global_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242#endif
2243
2244/*
2245 * Create a dabuf.
2246 */
2247/* ARGSUSED */
2248STATIC xfs_dabuf_t *
2249xfs_da_buf_make(int nbuf, xfs_buf_t **bps, inst_t *ra)
2250{
2251 xfs_buf_t *bp;
2252 xfs_dabuf_t *dabuf;
2253 int i;
2254 int off;
2255
2256 if (nbuf == 1)
Christoph Hellwig73195ed2009-07-18 18:14:56 -04002257 dabuf = kmem_zone_alloc(xfs_dabuf_zone, KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 else
Christoph Hellwig73195ed2009-07-18 18:14:56 -04002259 dabuf = kmem_alloc(XFS_DA_BUF_SIZE(nbuf), KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 dabuf->dirty = 0;
2261#ifdef XFS_DABUF_DEBUG
2262 dabuf->ra = ra;
2263 dabuf->target = XFS_BUF_TARGET(bps[0]);
2264 dabuf->blkno = XFS_BUF_ADDR(bps[0]);
2265#endif
2266 if (nbuf == 1) {
2267 dabuf->nbuf = 1;
2268 bp = bps[0];
2269 dabuf->bbcount = (short)BTOBB(XFS_BUF_COUNT(bp));
2270 dabuf->data = XFS_BUF_PTR(bp);
2271 dabuf->bps[0] = bp;
2272 } else {
2273 dabuf->nbuf = nbuf;
2274 for (i = 0, dabuf->bbcount = 0; i < nbuf; i++) {
2275 dabuf->bps[i] = bp = bps[i];
2276 dabuf->bbcount += BTOBB(XFS_BUF_COUNT(bp));
2277 }
2278 dabuf->data = kmem_alloc(BBTOB(dabuf->bbcount), KM_SLEEP);
2279 for (i = off = 0; i < nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2280 bp = bps[i];
2281 memcpy((char *)dabuf->data + off, XFS_BUF_PTR(bp),
2282 XFS_BUF_COUNT(bp));
2283 }
2284 }
2285#ifdef XFS_DABUF_DEBUG
2286 {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 xfs_dabuf_t *p;
2288
Eric Sandeen703e1f02007-10-11 17:41:21 +10002289 spin_lock(&xfs_dabuf_global_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 for (p = xfs_dabuf_global_list; p; p = p->next) {
2291 ASSERT(p->blkno != dabuf->blkno ||
2292 p->target != dabuf->target);
2293 }
2294 dabuf->prev = NULL;
2295 if (xfs_dabuf_global_list)
2296 xfs_dabuf_global_list->prev = dabuf;
2297 dabuf->next = xfs_dabuf_global_list;
2298 xfs_dabuf_global_list = dabuf;
Eric Sandeen703e1f02007-10-11 17:41:21 +10002299 spin_unlock(&xfs_dabuf_global_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 }
2301#endif
2302 return dabuf;
2303}
2304
2305/*
2306 * Un-dirty a dabuf.
2307 */
2308STATIC void
2309xfs_da_buf_clean(xfs_dabuf_t *dabuf)
2310{
2311 xfs_buf_t *bp;
2312 int i;
2313 int off;
2314
2315 if (dabuf->dirty) {
2316 ASSERT(dabuf->nbuf > 1);
2317 dabuf->dirty = 0;
2318 for (i = off = 0; i < dabuf->nbuf;
2319 i++, off += XFS_BUF_COUNT(bp)) {
2320 bp = dabuf->bps[i];
2321 memcpy(XFS_BUF_PTR(bp), (char *)dabuf->data + off,
2322 XFS_BUF_COUNT(bp));
2323 }
2324 }
2325}
2326
2327/*
2328 * Release a dabuf.
2329 */
2330void
2331xfs_da_buf_done(xfs_dabuf_t *dabuf)
2332{
2333 ASSERT(dabuf);
2334 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2335 if (dabuf->dirty)
2336 xfs_da_buf_clean(dabuf);
2337 if (dabuf->nbuf > 1)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002338 kmem_free(dabuf->data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339#ifdef XFS_DABUF_DEBUG
2340 {
Eric Sandeen703e1f02007-10-11 17:41:21 +10002341 spin_lock(&xfs_dabuf_global_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 if (dabuf->prev)
2343 dabuf->prev->next = dabuf->next;
2344 else
2345 xfs_dabuf_global_list = dabuf->next;
2346 if (dabuf->next)
2347 dabuf->next->prev = dabuf->prev;
Eric Sandeen703e1f02007-10-11 17:41:21 +10002348 spin_unlock(&xfs_dabuf_global_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 }
2350 memset(dabuf, 0, XFS_DA_BUF_SIZE(dabuf->nbuf));
2351#endif
2352 if (dabuf->nbuf == 1)
2353 kmem_zone_free(xfs_dabuf_zone, dabuf);
2354 else
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002355 kmem_free(dabuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356}
2357
2358/*
2359 * Log transaction from a dabuf.
2360 */
2361void
2362xfs_da_log_buf(xfs_trans_t *tp, xfs_dabuf_t *dabuf, uint first, uint last)
2363{
2364 xfs_buf_t *bp;
2365 uint f;
2366 int i;
2367 uint l;
2368 int off;
2369
2370 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2371 if (dabuf->nbuf == 1) {
2372 ASSERT(dabuf->data == (void *)XFS_BUF_PTR(dabuf->bps[0]));
2373 xfs_trans_log_buf(tp, dabuf->bps[0], first, last);
2374 return;
2375 }
2376 dabuf->dirty = 1;
2377 ASSERT(first <= last);
2378 for (i = off = 0; i < dabuf->nbuf; i++, off += XFS_BUF_COUNT(bp)) {
2379 bp = dabuf->bps[i];
2380 f = off;
2381 l = f + XFS_BUF_COUNT(bp) - 1;
2382 if (f < first)
2383 f = first;
2384 if (l > last)
2385 l = last;
2386 if (f <= l)
2387 xfs_trans_log_buf(tp, bp, f - off, l - off);
2388 /*
2389 * B_DONE is set by xfs_trans_log buf.
2390 * If we don't set it on a new buffer (get not read)
2391 * then if we don't put anything in the buffer it won't
2392 * be set, and at commit it it released into the cache,
2393 * and then a read will fail.
2394 */
2395 else if (!(XFS_BUF_ISDONE(bp)))
2396 XFS_BUF_DONE(bp);
2397 }
2398 ASSERT(last < off);
2399}
2400
2401/*
2402 * Release dabuf from a transaction.
2403 * Have to free up the dabuf before the buffers are released,
2404 * since the synchronization on the dabuf is really the lock on the buffer.
2405 */
2406void
2407xfs_da_brelse(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2408{
2409 xfs_buf_t *bp;
2410 xfs_buf_t **bplist;
2411 int i;
2412 int nbuf;
2413
2414 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2415 if ((nbuf = dabuf->nbuf) == 1) {
2416 bplist = &bp;
2417 bp = dabuf->bps[0];
2418 } else {
2419 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2420 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2421 }
2422 xfs_da_buf_done(dabuf);
2423 for (i = 0; i < nbuf; i++)
2424 xfs_trans_brelse(tp, bplist[i]);
2425 if (bplist != &bp)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002426 kmem_free(bplist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427}
2428
2429/*
2430 * Invalidate dabuf from a transaction.
2431 */
2432void
2433xfs_da_binval(xfs_trans_t *tp, xfs_dabuf_t *dabuf)
2434{
2435 xfs_buf_t *bp;
2436 xfs_buf_t **bplist;
2437 int i;
2438 int nbuf;
2439
2440 ASSERT(dabuf->nbuf && dabuf->data && dabuf->bbcount && dabuf->bps[0]);
2441 if ((nbuf = dabuf->nbuf) == 1) {
2442 bplist = &bp;
2443 bp = dabuf->bps[0];
2444 } else {
2445 bplist = kmem_alloc(nbuf * sizeof(*bplist), KM_SLEEP);
2446 memcpy(bplist, dabuf->bps, nbuf * sizeof(*bplist));
2447 }
2448 xfs_da_buf_done(dabuf);
2449 for (i = 0; i < nbuf; i++)
2450 xfs_trans_binval(tp, bplist[i]);
2451 if (bplist != &bp)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002452 kmem_free(bplist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453}
2454
2455/*
2456 * Get the first daddr from a dabuf.
2457 */
2458xfs_daddr_t
2459xfs_da_blkno(xfs_dabuf_t *dabuf)
2460{
2461 ASSERT(dabuf->nbuf);
2462 ASSERT(dabuf->data);
2463 return XFS_BUF_ADDR(dabuf->bps[0]);
2464}